43
44#include "opt_compat.h"
45#include "opt_ddb.h"
46#include "opt_watchdog.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/bio.h>
51#include <sys/buf.h>
52#include <sys/condvar.h>
53#include <sys/conf.h>
54#include <sys/dirent.h>
55#include <sys/event.h>
56#include <sys/eventhandler.h>
57#include <sys/extattr.h>
58#include <sys/file.h>
59#include <sys/fcntl.h>
60#include <sys/jail.h>
61#include <sys/kdb.h>
62#include <sys/kernel.h>
63#include <sys/kthread.h>
64#include <sys/lockf.h>
65#include <sys/malloc.h>
66#include <sys/mount.h>
67#include <sys/namei.h>
68#include <sys/pctrie.h>
69#include <sys/priv.h>
70#include <sys/reboot.h>
71#include <sys/rwlock.h>
72#include <sys/sched.h>
73#include <sys/sleepqueue.h>
74#include <sys/smp.h>
75#include <sys/stat.h>
76#include <sys/sysctl.h>
77#include <sys/syslog.h>
78#include <sys/vmmeter.h>
79#include <sys/vnode.h>
80#include <sys/watchdog.h>
81
82#include <machine/stdarg.h>
83
84#include <security/mac/mac_framework.h>
85
86#include <vm/vm.h>
87#include <vm/vm_object.h>
88#include <vm/vm_extern.h>
89#include <vm/pmap.h>
90#include <vm/vm_map.h>
91#include <vm/vm_page.h>
92#include <vm/vm_kern.h>
93#include <vm/uma.h>
94
95#ifdef DDB
96#include <ddb/ddb.h>
97#endif
98
99static void delmntque(struct vnode *vp);
100static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 int slpflag, int slptimeo);
102static void syncer_shutdown(void *arg, int howto);
103static int vtryrecycle(struct vnode *vp);
104static void v_incr_usecount(struct vnode *);
105static void v_decr_usecount(struct vnode *);
106static void v_decr_useonly(struct vnode *);
107static void v_upgrade_usecount(struct vnode *);
108static void vnlru_free(int);
109static void vgonel(struct vnode *);
110static void vfs_knllock(void *arg);
111static void vfs_knlunlock(void *arg);
112static void vfs_knl_assert_locked(void *arg);
113static void vfs_knl_assert_unlocked(void *arg);
114static void destroy_vpollinfo(struct vpollinfo *vi);
115
116/*
117 * Number of vnodes in existence. Increased whenever getnewvnode()
118 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
119 */
120static unsigned long numvnodes;
121
122SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123 "Number of vnodes in existence");
124
125static u_long vnodes_created;
126SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
127 0, "Number of vnodes created by getnewvnode");
128
129/*
130 * Conversion tables for conversion from vnode types to inode formats
131 * and back.
132 */
133enum vtype iftovt_tab[16] = {
134 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
135 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
136};
137int vttoif_tab[10] = {
138 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
139 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
140};
141
142/*
143 * List of vnodes that are ready for recycling.
144 */
145static TAILQ_HEAD(freelst, vnode) vnode_free_list;
146
147/*
148 * Free vnode target. Free vnodes may simply be files which have been stat'd
149 * but not read. This is somewhat common, and a small cache of such files
150 * should be kept to avoid recreation costs.
151 */
152static u_long wantfreevnodes;
153SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
154/* Number of vnodes in the free list. */
155static u_long freevnodes;
156SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
157 "Number of vnodes in the free list");
158
159static int vlru_allow_cache_src;
160SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
161 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
162
163static u_long recycles_count;
164SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
165 "Number of vnodes recycled to avoid exceding kern.maxvnodes");
166
167/*
168 * Various variables used for debugging the new implementation of
169 * reassignbuf().
170 * XXX these are probably of (very) limited utility now.
171 */
172static int reassignbufcalls;
173SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
174 "Number of calls to reassignbuf");
175
176static u_long free_owe_inact;
177SYSCTL_ULONG(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, 0,
178 "Number of times free vnodes kept on active list due to VFS "
179 "owing inactivation");
180
181/*
182 * Cache for the mount type id assigned to NFS. This is used for
183 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
184 */
185int nfs_mount_type = -1;
186
187/* To keep more than one thread at a time from running vfs_getnewfsid */
188static struct mtx mntid_mtx;
189
190/*
191 * Lock for any access to the following:
192 * vnode_free_list
193 * numvnodes
194 * freevnodes
195 */
196static struct mtx vnode_free_list_mtx;
197
198/* Publicly exported FS */
199struct nfs_public nfs_pub;
200
201static uma_zone_t buf_trie_zone;
202
203/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
204static uma_zone_t vnode_zone;
205static uma_zone_t vnodepoll_zone;
206
207/*
208 * The workitem queue.
209 *
210 * It is useful to delay writes of file data and filesystem metadata
211 * for tens of seconds so that quickly created and deleted files need
212 * not waste disk bandwidth being created and removed. To realize this,
213 * we append vnodes to a "workitem" queue. When running with a soft
214 * updates implementation, most pending metadata dependencies should
215 * not wait for more than a few seconds. Thus, mounted on block devices
216 * are delayed only about a half the time that file data is delayed.
217 * Similarly, directory updates are more critical, so are only delayed
218 * about a third the time that file data is delayed. Thus, there are
219 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
220 * one each second (driven off the filesystem syncer process). The
221 * syncer_delayno variable indicates the next queue that is to be processed.
222 * Items that need to be processed soon are placed in this queue:
223 *
224 * syncer_workitem_pending[syncer_delayno]
225 *
226 * A delay of fifteen seconds is done by placing the request fifteen
227 * entries later in the queue:
228 *
229 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
230 *
231 */
232static int syncer_delayno;
233static long syncer_mask;
234LIST_HEAD(synclist, bufobj);
235static struct synclist *syncer_workitem_pending;
236/*
237 * The sync_mtx protects:
238 * bo->bo_synclist
239 * sync_vnode_count
240 * syncer_delayno
241 * syncer_state
242 * syncer_workitem_pending
243 * syncer_worklist_len
244 * rushjob
245 */
246static struct mtx sync_mtx;
247static struct cv sync_wakeup;
248
249#define SYNCER_MAXDELAY 32
250static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
251static int syncdelay = 30; /* max time to delay syncing data */
252static int filedelay = 30; /* time to delay syncing files */
253SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
254 "Time to delay syncing files (in seconds)");
255static int dirdelay = 29; /* time to delay syncing directories */
256SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
257 "Time to delay syncing directories (in seconds)");
258static int metadelay = 28; /* time to delay syncing metadata */
259SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
260 "Time to delay syncing metadata (in seconds)");
261static int rushjob; /* number of slots to run ASAP */
262static int stat_rush_requests; /* number of times I/O speeded up */
263SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
264 "Number of times I/O speeded up (rush requests)");
265
266/*
267 * When shutting down the syncer, run it at four times normal speed.
268 */
269#define SYNCER_SHUTDOWN_SPEEDUP 4
270static int sync_vnode_count;
271static int syncer_worklist_len;
272static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
273 syncer_state;
274
275/*
276 * Number of vnodes we want to exist at any one time. This is mostly used
277 * to size hash tables in vnode-related code. It is normally not used in
278 * getnewvnode(), as wantfreevnodes is normally nonzero.)
279 *
280 * XXX desiredvnodes is historical cruft and should not exist.
281 */
282int desiredvnodes;
283
284static int
285sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS)
286{
287 int error, old_desiredvnodes;
288
289 old_desiredvnodes = desiredvnodes;
290 if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0)
291 return (error);
292 if (old_desiredvnodes != desiredvnodes) {
293 vfs_hash_changesize(desiredvnodes);
294 cache_changesize(desiredvnodes);
295 }
296 return (0);
297}
298
299SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes,
300 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0,
301 sysctl_update_desiredvnodes, "I", "Maximum number of vnodes");
302SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
303 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
304static int vnlru_nowhere;
305SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
306 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
307
308/* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
309static int vnsz2log;
310
311/*
312 * Support for the bufobj clean & dirty pctrie.
313 */
314static void *
315buf_trie_alloc(struct pctrie *ptree)
316{
317
318 return uma_zalloc(buf_trie_zone, M_NOWAIT);
319}
320
321static void
322buf_trie_free(struct pctrie *ptree, void *node)
323{
324
325 uma_zfree(buf_trie_zone, node);
326}
327PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
328
329/*
330 * Initialize the vnode management data structures.
331 *
332 * Reevaluate the following cap on the number of vnodes after the physical
333 * memory size exceeds 512GB. In the limit, as the physical memory size
334 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
335 */
336#ifndef MAXVNODES_MAX
337#define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
338#endif
339
340/*
341 * Initialize a vnode as it first enters the zone.
342 */
343static int
344vnode_init(void *mem, int size, int flags)
345{
346 struct vnode *vp;
347 struct bufobj *bo;
348
349 vp = mem;
350 bzero(vp, size);
351 /*
352 * Setup locks.
353 */
354 vp->v_vnlock = &vp->v_lock;
355 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
356 /*
357 * By default, don't allow shared locks unless filesystems opt-in.
358 */
359 lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT,
360 LK_NOSHARE | LK_IS_VNODE);
361 /*
362 * Initialize bufobj.
363 */
364 bo = &vp->v_bufobj;
365 bo->__bo_vnode = vp;
366 rw_init(BO_LOCKPTR(bo), "bufobj interlock");
367 bo->bo_private = vp;
368 TAILQ_INIT(&bo->bo_clean.bv_hd);
369 TAILQ_INIT(&bo->bo_dirty.bv_hd);
370 /*
371 * Initialize namecache.
372 */
373 LIST_INIT(&vp->v_cache_src);
374 TAILQ_INIT(&vp->v_cache_dst);
375 /*
376 * Initialize rangelocks.
377 */
378 rangelock_init(&vp->v_rl);
379 return (0);
380}
381
382/*
383 * Free a vnode when it is cleared from the zone.
384 */
385static void
386vnode_fini(void *mem, int size)
387{
388 struct vnode *vp;
389 struct bufobj *bo;
390
391 vp = mem;
392 rangelock_destroy(&vp->v_rl);
393 lockdestroy(vp->v_vnlock);
394 mtx_destroy(&vp->v_interlock);
395 bo = &vp->v_bufobj;
396 rw_destroy(BO_LOCKPTR(bo));
397}
398
399static void
400vntblinit(void *dummy __unused)
401{
402 u_int i;
403 int physvnodes, virtvnodes;
404
405 /*
406 * Desiredvnodes is a function of the physical memory size and the
407 * kernel's heap size. Generally speaking, it scales with the
408 * physical memory size. The ratio of desiredvnodes to physical pages
409 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
410 * marginal ratio of desiredvnodes to physical pages is one to
411 * sixteen. However, desiredvnodes is limited by the kernel's heap
412 * size. The memory required by desiredvnodes vnodes and vm objects
413 * may not exceed one seventh of the kernel's heap size.
414 */
415 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
416 cnt.v_page_count) / 16;
417 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
418 sizeof(struct vnode)));
419 desiredvnodes = min(physvnodes, virtvnodes);
420 if (desiredvnodes > MAXVNODES_MAX) {
421 if (bootverbose)
422 printf("Reducing kern.maxvnodes %d -> %d\n",
423 desiredvnodes, MAXVNODES_MAX);
424 desiredvnodes = MAXVNODES_MAX;
425 }
426 wantfreevnodes = desiredvnodes / 4;
427 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
428 TAILQ_INIT(&vnode_free_list);
429 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
430 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
431 vnode_init, vnode_fini, UMA_ALIGN_PTR, 0);
432 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
433 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
434 /*
435 * Preallocate enough nodes to support one-per buf so that
436 * we can not fail an insert. reassignbuf() callers can not
437 * tolerate the insertion failure.
438 */
439 buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
440 NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
441 UMA_ZONE_NOFREE | UMA_ZONE_VM);
442 uma_prealloc(buf_trie_zone, nbuf);
443 /*
444 * Initialize the filesystem syncer.
445 */
446 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
447 &syncer_mask);
448 syncer_maxdelay = syncer_mask + 1;
449 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
450 cv_init(&sync_wakeup, "syncer");
451 for (i = 1; i <= sizeof(struct vnode); i <<= 1)
452 vnsz2log++;
453 vnsz2log--;
454}
455SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
456
457
458/*
459 * Mark a mount point as busy. Used to synchronize access and to delay
460 * unmounting. Eventually, mountlist_mtx is not released on failure.
461 *
462 * vfs_busy() is a custom lock, it can block the caller.
463 * vfs_busy() only sleeps if the unmount is active on the mount point.
464 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
465 * vnode belonging to mp.
466 *
467 * Lookup uses vfs_busy() to traverse mount points.
468 * root fs var fs
469 * / vnode lock A / vnode lock (/var) D
470 * /var vnode lock B /log vnode lock(/var/log) E
471 * vfs_busy lock C vfs_busy lock F
472 *
473 * Within each file system, the lock order is C->A->B and F->D->E.
474 *
475 * When traversing across mounts, the system follows that lock order:
476 *
477 * C->A->B
478 * |
479 * +->F->D->E
480 *
481 * The lookup() process for namei("/var") illustrates the process:
482 * VOP_LOOKUP() obtains B while A is held
483 * vfs_busy() obtains a shared lock on F while A and B are held
484 * vput() releases lock on B
485 * vput() releases lock on A
486 * VFS_ROOT() obtains lock on D while shared lock on F is held
487 * vfs_unbusy() releases shared lock on F
488 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
489 * Attempt to lock A (instead of vp_crossmp) while D is held would
490 * violate the global order, causing deadlocks.
491 *
492 * dounmount() locks B while F is drained.
493 */
494int
495vfs_busy(struct mount *mp, int flags)
496{
497
498 MPASS((flags & ~MBF_MASK) == 0);
499 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
500
501 MNT_ILOCK(mp);
502 MNT_REF(mp);
503 /*
504 * If mount point is currenly being unmounted, sleep until the
505 * mount point fate is decided. If thread doing the unmounting fails,
506 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
507 * that this mount point has survived the unmount attempt and vfs_busy
508 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
509 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
510 * about to be really destroyed. vfs_busy needs to release its
511 * reference on the mount point in this case and return with ENOENT,
512 * telling the caller that mount mount it tried to busy is no longer
513 * valid.
514 */
515 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
516 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
517 MNT_REL(mp);
518 MNT_IUNLOCK(mp);
519 CTR1(KTR_VFS, "%s: failed busying before sleeping",
520 __func__);
521 return (ENOENT);
522 }
523 if (flags & MBF_MNTLSTLOCK)
524 mtx_unlock(&mountlist_mtx);
525 mp->mnt_kern_flag |= MNTK_MWAIT;
526 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
527 if (flags & MBF_MNTLSTLOCK)
528 mtx_lock(&mountlist_mtx);
529 MNT_ILOCK(mp);
530 }
531 if (flags & MBF_MNTLSTLOCK)
532 mtx_unlock(&mountlist_mtx);
533 mp->mnt_lockref++;
534 MNT_IUNLOCK(mp);
535 return (0);
536}
537
538/*
539 * Free a busy filesystem.
540 */
541void
542vfs_unbusy(struct mount *mp)
543{
544
545 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
546 MNT_ILOCK(mp);
547 MNT_REL(mp);
548 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
549 mp->mnt_lockref--;
550 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
551 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
552 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
553 mp->mnt_kern_flag &= ~MNTK_DRAINING;
554 wakeup(&mp->mnt_lockref);
555 }
556 MNT_IUNLOCK(mp);
557}
558
559/*
560 * Lookup a mount point by filesystem identifier.
561 */
562struct mount *
563vfs_getvfs(fsid_t *fsid)
564{
565 struct mount *mp;
566
567 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
568 mtx_lock(&mountlist_mtx);
569 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
570 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
571 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
572 vfs_ref(mp);
573 mtx_unlock(&mountlist_mtx);
574 return (mp);
575 }
576 }
577 mtx_unlock(&mountlist_mtx);
578 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
579 return ((struct mount *) 0);
580}
581
582/*
583 * Lookup a mount point by filesystem identifier, busying it before
584 * returning.
585 *
586 * To avoid congestion on mountlist_mtx, implement simple direct-mapped
587 * cache for popular filesystem identifiers. The cache is lockess, using
588 * the fact that struct mount's are never freed. In worst case we may
589 * get pointer to unmounted or even different filesystem, so we have to
590 * check what we got, and go slow way if so.
591 */
592struct mount *
593vfs_busyfs(fsid_t *fsid)
594{
595#define FSID_CACHE_SIZE 256
596 typedef struct mount * volatile vmp_t;
597 static vmp_t cache[FSID_CACHE_SIZE];
598 struct mount *mp;
599 int error;
600 uint32_t hash;
601
602 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
603 hash = fsid->val[0] ^ fsid->val[1];
604 hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
605 mp = cache[hash];
606 if (mp == NULL ||
607 mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
608 mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
609 goto slow;
610 if (vfs_busy(mp, 0) != 0) {
611 cache[hash] = NULL;
612 goto slow;
613 }
614 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
615 mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
616 return (mp);
617 else
618 vfs_unbusy(mp);
619
620slow:
621 mtx_lock(&mountlist_mtx);
622 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
623 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
624 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
625 error = vfs_busy(mp, MBF_MNTLSTLOCK);
626 if (error) {
627 cache[hash] = NULL;
628 mtx_unlock(&mountlist_mtx);
629 return (NULL);
630 }
631 cache[hash] = mp;
632 return (mp);
633 }
634 }
635 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
636 mtx_unlock(&mountlist_mtx);
637 return ((struct mount *) 0);
638}
639
640/*
641 * Check if a user can access privileged mount options.
642 */
643int
644vfs_suser(struct mount *mp, struct thread *td)
645{
646 int error;
647
648 /*
649 * If the thread is jailed, but this is not a jail-friendly file
650 * system, deny immediately.
651 */
652 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
653 return (EPERM);
654
655 /*
656 * If the file system was mounted outside the jail of the calling
657 * thread, deny immediately.
658 */
659 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
660 return (EPERM);
661
662 /*
663 * If file system supports delegated administration, we don't check
664 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
665 * by the file system itself.
666 * If this is not the user that did original mount, we check for
667 * the PRIV_VFS_MOUNT_OWNER privilege.
668 */
669 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
670 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
671 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
672 return (error);
673 }
674 return (0);
675}
676
677/*
678 * Get a new unique fsid. Try to make its val[0] unique, since this value
679 * will be used to create fake device numbers for stat(). Also try (but
680 * not so hard) make its val[0] unique mod 2^16, since some emulators only
681 * support 16-bit device numbers. We end up with unique val[0]'s for the
682 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
683 *
684 * Keep in mind that several mounts may be running in parallel. Starting
685 * the search one past where the previous search terminated is both a
686 * micro-optimization and a defense against returning the same fsid to
687 * different mounts.
688 */
689void
690vfs_getnewfsid(struct mount *mp)
691{
692 static uint16_t mntid_base;
693 struct mount *nmp;
694 fsid_t tfsid;
695 int mtype;
696
697 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
698 mtx_lock(&mntid_mtx);
699 mtype = mp->mnt_vfc->vfc_typenum;
700 tfsid.val[1] = mtype;
701 mtype = (mtype & 0xFF) << 24;
702 for (;;) {
703 tfsid.val[0] = makedev(255,
704 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
705 mntid_base++;
706 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
707 break;
708 vfs_rel(nmp);
709 }
710 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
711 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
712 mtx_unlock(&mntid_mtx);
713}
714
715/*
716 * Knob to control the precision of file timestamps:
717 *
718 * 0 = seconds only; nanoseconds zeroed.
719 * 1 = seconds and nanoseconds, accurate within 1/HZ.
720 * 2 = seconds and nanoseconds, truncated to microseconds.
721 * >=3 = seconds and nanoseconds, maximum precision.
722 */
723enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
724
725static int timestamp_precision = TSP_USEC;
726SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
727 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
728 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
729 "3+: sec + ns (max. precision))");
730
731/*
732 * Get a current timestamp.
733 */
734void
735vfs_timestamp(struct timespec *tsp)
736{
737 struct timeval tv;
738
739 switch (timestamp_precision) {
740 case TSP_SEC:
741 tsp->tv_sec = time_second;
742 tsp->tv_nsec = 0;
743 break;
744 case TSP_HZ:
745 getnanotime(tsp);
746 break;
747 case TSP_USEC:
748 microtime(&tv);
749 TIMEVAL_TO_TIMESPEC(&tv, tsp);
750 break;
751 case TSP_NSEC:
752 default:
753 nanotime(tsp);
754 break;
755 }
756}
757
758/*
759 * Set vnode attributes to VNOVAL
760 */
761void
762vattr_null(struct vattr *vap)
763{
764
765 vap->va_type = VNON;
766 vap->va_size = VNOVAL;
767 vap->va_bytes = VNOVAL;
768 vap->va_mode = VNOVAL;
769 vap->va_nlink = VNOVAL;
770 vap->va_uid = VNOVAL;
771 vap->va_gid = VNOVAL;
772 vap->va_fsid = VNOVAL;
773 vap->va_fileid = VNOVAL;
774 vap->va_blocksize = VNOVAL;
775 vap->va_rdev = VNOVAL;
776 vap->va_atime.tv_sec = VNOVAL;
777 vap->va_atime.tv_nsec = VNOVAL;
778 vap->va_mtime.tv_sec = VNOVAL;
779 vap->va_mtime.tv_nsec = VNOVAL;
780 vap->va_ctime.tv_sec = VNOVAL;
781 vap->va_ctime.tv_nsec = VNOVAL;
782 vap->va_birthtime.tv_sec = VNOVAL;
783 vap->va_birthtime.tv_nsec = VNOVAL;
784 vap->va_flags = VNOVAL;
785 vap->va_gen = VNOVAL;
786 vap->va_vaflags = 0;
787}
788
789/*
790 * This routine is called when we have too many vnodes. It attempts
791 * to free <count> vnodes and will potentially free vnodes that still
792 * have VM backing store (VM backing store is typically the cause
793 * of a vnode blowout so we want to do this). Therefore, this operation
794 * is not considered cheap.
795 *
796 * A number of conditions may prevent a vnode from being reclaimed.
797 * the buffer cache may have references on the vnode, a directory
798 * vnode may still have references due to the namei cache representing
799 * underlying files, or the vnode may be in active use. It is not
800 * desireable to reuse such vnodes. These conditions may cause the
801 * number of vnodes to reach some minimum value regardless of what
802 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
803 */
804static int
805vlrureclaim(struct mount *mp)
806{
807 struct vnode *vp;
808 int done;
809 int trigger;
810 int usevnodes;
811 int count;
812
813 /*
814 * Calculate the trigger point, don't allow user
815 * screwups to blow us up. This prevents us from
816 * recycling vnodes with lots of resident pages. We
817 * aren't trying to free memory, we are trying to
818 * free vnodes.
819 */
820 usevnodes = desiredvnodes;
821 if (usevnodes <= 0)
822 usevnodes = 1;
823 trigger = cnt.v_page_count * 2 / usevnodes;
824 done = 0;
825 vn_start_write(NULL, &mp, V_WAIT);
826 MNT_ILOCK(mp);
827 count = mp->mnt_nvnodelistsize / 10 + 1;
828 while (count != 0) {
829 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
830 while (vp != NULL && vp->v_type == VMARKER)
831 vp = TAILQ_NEXT(vp, v_nmntvnodes);
832 if (vp == NULL)
833 break;
834 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
835 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
836 --count;
837 if (!VI_TRYLOCK(vp))
838 goto next_iter;
839 /*
840 * If it's been deconstructed already, it's still
841 * referenced, or it exceeds the trigger, skip it.
842 */
843 if (vp->v_usecount ||
844 (!vlru_allow_cache_src &&
845 !LIST_EMPTY(&(vp)->v_cache_src)) ||
846 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
847 vp->v_object->resident_page_count > trigger)) {
848 VI_UNLOCK(vp);
849 goto next_iter;
850 }
851 MNT_IUNLOCK(mp);
852 vholdl(vp);
853 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
854 vdrop(vp);
855 goto next_iter_mntunlocked;
856 }
857 VI_LOCK(vp);
858 /*
859 * v_usecount may have been bumped after VOP_LOCK() dropped
860 * the vnode interlock and before it was locked again.
861 *
862 * It is not necessary to recheck VI_DOOMED because it can
863 * only be set by another thread that holds both the vnode
864 * lock and vnode interlock. If another thread has the
865 * vnode lock before we get to VOP_LOCK() and obtains the
866 * vnode interlock after VOP_LOCK() drops the vnode
867 * interlock, the other thread will be unable to drop the
868 * vnode lock before our VOP_LOCK() call fails.
869 */
870 if (vp->v_usecount ||
871 (!vlru_allow_cache_src &&
872 !LIST_EMPTY(&(vp)->v_cache_src)) ||
873 (vp->v_object != NULL &&
874 vp->v_object->resident_page_count > trigger)) {
875 VOP_UNLOCK(vp, LK_INTERLOCK);
876 vdrop(vp);
877 goto next_iter_mntunlocked;
878 }
879 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
880 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
881 atomic_add_long(&recycles_count, 1);
882 vgonel(vp);
883 VOP_UNLOCK(vp, 0);
884 vdropl(vp);
885 done++;
886next_iter_mntunlocked:
887 if (!should_yield())
888 goto relock_mnt;
889 goto yield;
890next_iter:
891 if (!should_yield())
892 continue;
893 MNT_IUNLOCK(mp);
894yield:
895 kern_yield(PRI_USER);
896relock_mnt:
897 MNT_ILOCK(mp);
898 }
899 MNT_IUNLOCK(mp);
900 vn_finished_write(mp);
901 return done;
902}
903
904/*
905 * Attempt to keep the free list at wantfreevnodes length.
906 */
907static void
908vnlru_free(int count)
909{
910 struct vnode *vp;
911
912 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
913 for (; count > 0; count--) {
914 vp = TAILQ_FIRST(&vnode_free_list);
915 /*
916 * The list can be modified while the free_list_mtx
917 * has been dropped and vp could be NULL here.
918 */
919 if (!vp)
920 break;
921 VNASSERT(vp->v_op != NULL, vp,
922 ("vnlru_free: vnode already reclaimed."));
923 KASSERT((vp->v_iflag & VI_FREE) != 0,
924 ("Removing vnode not on freelist"));
925 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
926 ("Mangling active vnode"));
927 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
928 /*
929 * Don't recycle if we can't get the interlock.
930 */
931 if (!VI_TRYLOCK(vp)) {
932 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
933 continue;
934 }
935 VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
936 vp, ("vp inconsistent on freelist"));
937
938 /*
939 * The clear of VI_FREE prevents activation of the
940 * vnode. There is no sense in putting the vnode on
941 * the mount point active list, only to remove it
942 * later during recycling. Inline the relevant part
943 * of vholdl(), to avoid triggering assertions or
944 * activating.
945 */
946 freevnodes--;
947 vp->v_iflag &= ~VI_FREE;
948 vp->v_holdcnt++;
949
950 mtx_unlock(&vnode_free_list_mtx);
951 VI_UNLOCK(vp);
952 vtryrecycle(vp);
953 /*
954 * If the recycled succeeded this vdrop will actually free
955 * the vnode. If not it will simply place it back on
956 * the free list.
957 */
958 vdrop(vp);
959 mtx_lock(&vnode_free_list_mtx);
960 }
961}
962/*
963 * Attempt to recycle vnodes in a context that is always safe to block.
964 * Calling vlrurecycle() from the bowels of filesystem code has some
965 * interesting deadlock problems.
966 */
967static struct proc *vnlruproc;
968static int vnlruproc_sig;
969
970static void
971vnlru_proc(void)
972{
973 struct mount *mp, *nmp;
974 int done;
975 struct proc *p = vnlruproc;
976
977 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
978 SHUTDOWN_PRI_FIRST);
979
980 for (;;) {
981 kproc_suspend_check(p);
982 mtx_lock(&vnode_free_list_mtx);
983 if (freevnodes > wantfreevnodes)
984 vnlru_free(freevnodes - wantfreevnodes);
985 if (numvnodes <= desiredvnodes * 9 / 10) {
986 vnlruproc_sig = 0;
987 wakeup(&vnlruproc_sig);
988 msleep(vnlruproc, &vnode_free_list_mtx,
989 PVFS|PDROP, "vlruwt", hz);
990 continue;
991 }
992 mtx_unlock(&vnode_free_list_mtx);
993 done = 0;
994 mtx_lock(&mountlist_mtx);
995 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
996 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
997 nmp = TAILQ_NEXT(mp, mnt_list);
998 continue;
999 }
1000 done += vlrureclaim(mp);
1001 mtx_lock(&mountlist_mtx);
1002 nmp = TAILQ_NEXT(mp, mnt_list);
1003 vfs_unbusy(mp);
1004 }
1005 mtx_unlock(&mountlist_mtx);
1006 if (done == 0) {
1007#if 0
1008 /* These messages are temporary debugging aids */
1009 if (vnlru_nowhere < 5)
1010 printf("vnlru process getting nowhere..\n");
1011 else if (vnlru_nowhere == 5)
1012 printf("vnlru process messages stopped.\n");
1013#endif
1014 vnlru_nowhere++;
1015 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
1016 } else
1017 kern_yield(PRI_USER);
1018 }
1019}
1020
1021static struct kproc_desc vnlru_kp = {
1022 "vnlru",
1023 vnlru_proc,
1024 &vnlruproc
1025};
1026SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
1027 &vnlru_kp);
1028
1029/*
1030 * Routines having to do with the management of the vnode table.
1031 */
1032
1033/*
1034 * Try to recycle a freed vnode. We abort if anyone picks up a reference
1035 * before we actually vgone(). This function must be called with the vnode
1036 * held to prevent the vnode from being returned to the free list midway
1037 * through vgone().
1038 */
1039static int
1040vtryrecycle(struct vnode *vp)
1041{
1042 struct mount *vnmp;
1043
1044 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
1045 VNASSERT(vp->v_holdcnt, vp,
1046 ("vtryrecycle: Recycling vp %p without a reference.", vp));
1047 /*
1048 * This vnode may found and locked via some other list, if so we
1049 * can't recycle it yet.
1050 */
1051 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1052 CTR2(KTR_VFS,
1053 "%s: impossible to recycle, vp %p lock is already held",
1054 __func__, vp);
1055 return (EWOULDBLOCK);
1056 }
1057 /*
1058 * Don't recycle if its filesystem is being suspended.
1059 */
1060 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
1061 VOP_UNLOCK(vp, 0);
1062 CTR2(KTR_VFS,
1063 "%s: impossible to recycle, cannot start the write for %p",
1064 __func__, vp);
1065 return (EBUSY);
1066 }
1067 /*
1068 * If we got this far, we need to acquire the interlock and see if
1069 * anyone picked up this vnode from another list. If not, we will
1070 * mark it with DOOMED via vgonel() so that anyone who does find it
1071 * will skip over it.
1072 */
1073 VI_LOCK(vp);
1074 if (vp->v_usecount) {
1075 VOP_UNLOCK(vp, LK_INTERLOCK);
1076 vn_finished_write(vnmp);
1077 CTR2(KTR_VFS,
1078 "%s: impossible to recycle, %p is already referenced",
1079 __func__, vp);
1080 return (EBUSY);
1081 }
1082 if ((vp->v_iflag & VI_DOOMED) == 0) {
1083 atomic_add_long(&recycles_count, 1);
1084 vgonel(vp);
1085 }
1086 VOP_UNLOCK(vp, LK_INTERLOCK);
1087 vn_finished_write(vnmp);
1088 return (0);
1089}
1090
1091/*
1092 * Wait for available vnodes.
1093 */
1094static int
1095getnewvnode_wait(int suspended)
1096{
1097
1098 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
1099 if (numvnodes > desiredvnodes) {
1100 if (suspended) {
1101 /*
1102 * File system is beeing suspended, we cannot risk a
1103 * deadlock here, so allocate new vnode anyway.
1104 */
1105 if (freevnodes > wantfreevnodes)
1106 vnlru_free(freevnodes - wantfreevnodes);
1107 return (0);
1108 }
1109 if (vnlruproc_sig == 0) {
1110 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
1111 wakeup(vnlruproc);
1112 }
1113 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1114 "vlruwk", hz);
1115 }
1116 return (numvnodes > desiredvnodes ? ENFILE : 0);
1117}
1118
1119void
1120getnewvnode_reserve(u_int count)
1121{
1122 struct thread *td;
1123
1124 td = curthread;
1125 /* First try to be quick and racy. */
1126 if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1127 td->td_vp_reserv += count;
1128 return;
1129 } else
1130 atomic_subtract_long(&numvnodes, count);
1131
1132 mtx_lock(&vnode_free_list_mtx);
1133 while (count > 0) {
1134 if (getnewvnode_wait(0) == 0) {
1135 count--;
1136 td->td_vp_reserv++;
1137 atomic_add_long(&numvnodes, 1);
1138 }
1139 }
1140 mtx_unlock(&vnode_free_list_mtx);
1141}
1142
1143void
1144getnewvnode_drop_reserve(void)
1145{
1146 struct thread *td;
1147
1148 td = curthread;
1149 atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1150 td->td_vp_reserv = 0;
1151}
1152
1153/*
1154 * Return the next vnode from the free list.
1155 */
1156int
1157getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1158 struct vnode **vpp)
1159{
1160 struct vnode *vp;
1161 struct thread *td;
1162 struct lock_object *lo;
1163 int error;
1164
1165 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1166 vp = NULL;
1167 td = curthread;
1168 if (td->td_vp_reserv > 0) {
1169 td->td_vp_reserv -= 1;
1170 goto alloc;
1171 }
1172 mtx_lock(&vnode_free_list_mtx);
1173 /*
1174 * Lend our context to reclaim vnodes if they've exceeded the max.
1175 */
1176 if (freevnodes > wantfreevnodes)
1177 vnlru_free(1);
1178 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1179 MNTK_SUSPEND));
1180#if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1181 if (error != 0) {
1182 mtx_unlock(&vnode_free_list_mtx);
1183 return (error);
1184 }
1185#endif
1186 atomic_add_long(&numvnodes, 1);
1187 mtx_unlock(&vnode_free_list_mtx);
1188alloc:
1189 atomic_add_long(&vnodes_created, 1);
1190 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK);
1191 /*
1192 * Locks are given the generic name "vnode" when created.
1193 * Follow the historic practice of using the filesystem
1194 * name when they allocated, e.g., "zfs", "ufs", "nfs, etc.
1195 *
1196 * Locks live in a witness group keyed on their name. Thus,
1197 * when a lock is renamed, it must also move from the witness
1198 * group of its old name to the witness group of its new name.
1199 *
1200 * The change only needs to be made when the vnode moves
1201 * from one filesystem type to another. We ensure that each
1202 * filesystem use a single static name pointer for its tag so
1203 * that we can compare pointers rather than doing a strcmp().
1204 */
1205 lo = &vp->v_vnlock->lock_object;
1206 if (lo->lo_name != tag) {
1207 lo->lo_name = tag;
1208 WITNESS_DESTROY(lo);
1209 WITNESS_INIT(lo, tag);
1210 }
1211 /*
1212 * By default, don't allow shared locks unless filesystems opt-in.
1213 */
1214 vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE;
1215 /*
1216 * Finalize various vnode identity bits.
1217 */
1218 KASSERT(vp->v_object == NULL, ("stale v_object %p", vp));
1219 KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp));
1220 KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp));
1221 vp->v_type = VNON;
1222 vp->v_tag = tag;
1223 vp->v_op = vops;
1224 v_incr_usecount(vp);
1225 vp->v_bufobj.bo_ops = &buf_ops_bio;
1226#ifdef MAC
1227 mac_vnode_init(vp);
1228 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1229 mac_vnode_associate_singlelabel(mp, vp);
1230 else if (mp == NULL && vops != &dead_vnodeops)
1231 printf("NULL mp in getnewvnode()\n");
1232#endif
1233 if (mp != NULL) {
1234 vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize;
1235 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1236 vp->v_vflag |= VV_NOKNOTE;
1237 }
1238
1239 /*
1240 * For the filesystems which do not use vfs_hash_insert(),
1241 * still initialize v_hash to have vfs_hash_index() useful.
1242 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1243 * its own hashing.
1244 */
1245 vp->v_hash = (uintptr_t)vp >> vnsz2log;
1246
1247 *vpp = vp;
1248 return (0);
1249}
1250
1251/*
1252 * Delete from old mount point vnode list, if on one.
1253 */
1254static void
1255delmntque(struct vnode *vp)
1256{
1257 struct mount *mp;
1258 int active;
1259
1260 mp = vp->v_mount;
1261 if (mp == NULL)
1262 return;
1263 MNT_ILOCK(mp);
1264 VI_LOCK(vp);
1265 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1266 ("Active vnode list size %d > Vnode list size %d",
1267 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1268 active = vp->v_iflag & VI_ACTIVE;
1269 vp->v_iflag &= ~VI_ACTIVE;
1270 if (active) {
1271 mtx_lock(&vnode_free_list_mtx);
1272 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1273 mp->mnt_activevnodelistsize--;
1274 mtx_unlock(&vnode_free_list_mtx);
1275 }
1276 vp->v_mount = NULL;
1277 VI_UNLOCK(vp);
1278 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1279 ("bad mount point vnode list size"));
1280 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1281 mp->mnt_nvnodelistsize--;
1282 MNT_REL(mp);
1283 MNT_IUNLOCK(mp);
1284}
1285
1286static void
1287insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1288{
1289
1290 vp->v_data = NULL;
1291 vp->v_op = &dead_vnodeops;
1292 vgone(vp);
1293 vput(vp);
1294}
1295
1296/*
1297 * Insert into list of vnodes for the new mount point, if available.
1298 */
1299int
1300insmntque1(struct vnode *vp, struct mount *mp,
1301 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1302{
1303
1304 KASSERT(vp->v_mount == NULL,
1305 ("insmntque: vnode already on per mount vnode list"));
1306 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1307 ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1308
1309 /*
1310 * We acquire the vnode interlock early to ensure that the
1311 * vnode cannot be recycled by another process releasing a
1312 * holdcnt on it before we get it on both the vnode list
1313 * and the active vnode list. The mount mutex protects only
1314 * manipulation of the vnode list and the vnode freelist
1315 * mutex protects only manipulation of the active vnode list.
1316 * Hence the need to hold the vnode interlock throughout.
1317 */
1318 MNT_ILOCK(mp);
1319 VI_LOCK(vp);
1320 if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1321 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1322 mp->mnt_nvnodelistsize == 0)) &&
1323 (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1324 VI_UNLOCK(vp);
1325 MNT_IUNLOCK(mp);
1326 if (dtr != NULL)
1327 dtr(vp, dtr_arg);
1328 return (EBUSY);
1329 }
1330 vp->v_mount = mp;
1331 MNT_REF(mp);
1332 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1333 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1334 ("neg mount point vnode list size"));
1335 mp->mnt_nvnodelistsize++;
1336 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1337 ("Activating already active vnode"));
1338 vp->v_iflag |= VI_ACTIVE;
1339 mtx_lock(&vnode_free_list_mtx);
1340 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1341 mp->mnt_activevnodelistsize++;
1342 mtx_unlock(&vnode_free_list_mtx);
1343 VI_UNLOCK(vp);
1344 MNT_IUNLOCK(mp);
1345 return (0);
1346}
1347
1348int
1349insmntque(struct vnode *vp, struct mount *mp)
1350{
1351
1352 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1353}
1354
1355/*
1356 * Flush out and invalidate all buffers associated with a bufobj
1357 * Called with the underlying object locked.
1358 */
1359int
1360bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1361{
1362 int error;
1363
1364 BO_LOCK(bo);
1365 if (flags & V_SAVE) {
1366 error = bufobj_wwait(bo, slpflag, slptimeo);
1367 if (error) {
1368 BO_UNLOCK(bo);
1369 return (error);
1370 }
1371 if (bo->bo_dirty.bv_cnt > 0) {
1372 BO_UNLOCK(bo);
1373 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1374 return (error);
1375 /*
1376 * XXX We could save a lock/unlock if this was only
1377 * enabled under INVARIANTS
1378 */
1379 BO_LOCK(bo);
1380 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1381 panic("vinvalbuf: dirty bufs");
1382 }
1383 }
1384 /*
1385 * If you alter this loop please notice that interlock is dropped and
1386 * reacquired in flushbuflist. Special care is needed to ensure that
1387 * no race conditions occur from this.
1388 */
1389 do {
1390 error = flushbuflist(&bo->bo_clean,
1391 flags, bo, slpflag, slptimeo);
1392 if (error == 0 && !(flags & V_CLEANONLY))
1393 error = flushbuflist(&bo->bo_dirty,
1394 flags, bo, slpflag, slptimeo);
1395 if (error != 0 && error != EAGAIN) {
1396 BO_UNLOCK(bo);
1397 return (error);
1398 }
1399 } while (error != 0);
1400
1401 /*
1402 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1403 * have write I/O in-progress but if there is a VM object then the
1404 * VM object can also have read-I/O in-progress.
1405 */
1406 do {
1407 bufobj_wwait(bo, 0, 0);
1408 BO_UNLOCK(bo);
1409 if (bo->bo_object != NULL) {
1410 VM_OBJECT_WLOCK(bo->bo_object);
1411 vm_object_pip_wait(bo->bo_object, "bovlbx");
1412 VM_OBJECT_WUNLOCK(bo->bo_object);
1413 }
1414 BO_LOCK(bo);
1415 } while (bo->bo_numoutput > 0);
1416 BO_UNLOCK(bo);
1417
1418 /*
1419 * Destroy the copy in the VM cache, too.
1420 */
1421 if (bo->bo_object != NULL &&
1422 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1423 VM_OBJECT_WLOCK(bo->bo_object);
1424 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1425 OBJPR_CLEANONLY : 0);
1426 VM_OBJECT_WUNLOCK(bo->bo_object);
1427 }
1428
1429#ifdef INVARIANTS
1430 BO_LOCK(bo);
1431 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1432 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1433 panic("vinvalbuf: flush failed");
1434 BO_UNLOCK(bo);
1435#endif
1436 return (0);
1437}
1438
1439/*
1440 * Flush out and invalidate all buffers associated with a vnode.
1441 * Called with the underlying object locked.
1442 */
1443int
1444vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1445{
1446
1447 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1448 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1449 if (vp->v_object != NULL && vp->v_object->handle != vp)
1450 return (0);
1451 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1452}
1453
1454/*
1455 * Flush out buffers on the specified list.
1456 *
1457 */
1458static int
1459flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1460 int slptimeo)
1461{
1462 struct buf *bp, *nbp;
1463 int retval, error;
1464 daddr_t lblkno;
1465 b_xflags_t xflags;
1466
1467 ASSERT_BO_WLOCKED(bo);
1468
1469 retval = 0;
1470 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1471 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1472 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1473 continue;
1474 }
1475 lblkno = 0;
1476 xflags = 0;
1477 if (nbp != NULL) {
1478 lblkno = nbp->b_lblkno;
1479 xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1480 }
1481 retval = EAGAIN;
1482 error = BUF_TIMELOCK(bp,
1483 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1484 "flushbuf", slpflag, slptimeo);
1485 if (error) {
1486 BO_LOCK(bo);
1487 return (error != ENOLCK ? error : EAGAIN);
1488 }
1489 KASSERT(bp->b_bufobj == bo,
1490 ("bp %p wrong b_bufobj %p should be %p",
1491 bp, bp->b_bufobj, bo));
1492 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1493 BUF_UNLOCK(bp);
1494 BO_LOCK(bo);
1495 return (EAGAIN);
1496 }
1497 /*
1498 * XXX Since there are no node locks for NFS, I
1499 * believe there is a slight chance that a delayed
1500 * write will occur while sleeping just above, so
1501 * check for it.
1502 */
1503 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1504 (flags & V_SAVE)) {
1505 bremfree(bp);
1506 bp->b_flags |= B_ASYNC;
1507 bwrite(bp);
1508 BO_LOCK(bo);
1509 return (EAGAIN); /* XXX: why not loop ? */
1510 }
1511 bremfree(bp);
1512 bp->b_flags |= (B_INVAL | B_RELBUF);
1513 bp->b_flags &= ~B_ASYNC;
1514 brelse(bp);
1515 BO_LOCK(bo);
1516 if (nbp != NULL &&
1517 (nbp->b_bufobj != bo ||
1518 nbp->b_lblkno != lblkno ||
1519 (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1520 break; /* nbp invalid */
1521 }
1522 return (retval);
1523}
1524
1525/*
1526 * Truncate a file's buffer and pages to a specified length. This
1527 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1528 * sync activity.
1529 */
1530int
1531vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1532{
1533 struct buf *bp, *nbp;
1534 int anyfreed;
1535 int trunclbn;
1536 struct bufobj *bo;
1537
1538 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1539 vp, cred, blksize, (uintmax_t)length);
1540
1541 /*
1542 * Round up to the *next* lbn.
1543 */
1544 trunclbn = (length + blksize - 1) / blksize;
1545
1546 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1547restart:
1548 bo = &vp->v_bufobj;
1549 BO_LOCK(bo);
1550 anyfreed = 1;
1551 for (;anyfreed;) {
1552 anyfreed = 0;
1553 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1554 if (bp->b_lblkno < trunclbn)
1555 continue;
1556 if (BUF_LOCK(bp,
1557 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1558 BO_LOCKPTR(bo)) == ENOLCK)
1559 goto restart;
1560
1561 bremfree(bp);
1562 bp->b_flags |= (B_INVAL | B_RELBUF);
1563 bp->b_flags &= ~B_ASYNC;
1564 brelse(bp);
1565 anyfreed = 1;
1566
1567 BO_LOCK(bo);
1568 if (nbp != NULL &&
1569 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1570 (nbp->b_vp != vp) ||
1571 (nbp->b_flags & B_DELWRI))) {
1572 BO_UNLOCK(bo);
1573 goto restart;
1574 }
1575 }
1576
1577 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1578 if (bp->b_lblkno < trunclbn)
1579 continue;
1580 if (BUF_LOCK(bp,
1581 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1582 BO_LOCKPTR(bo)) == ENOLCK)
1583 goto restart;
1584 bremfree(bp);
1585 bp->b_flags |= (B_INVAL | B_RELBUF);
1586 bp->b_flags &= ~B_ASYNC;
1587 brelse(bp);
1588 anyfreed = 1;
1589
1590 BO_LOCK(bo);
1591 if (nbp != NULL &&
1592 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1593 (nbp->b_vp != vp) ||
1594 (nbp->b_flags & B_DELWRI) == 0)) {
1595 BO_UNLOCK(bo);
1596 goto restart;
1597 }
1598 }
1599 }
1600
1601 if (length > 0) {
1602restartsync:
1603 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1604 if (bp->b_lblkno > 0)
1605 continue;
1606 /*
1607 * Since we hold the vnode lock this should only
1608 * fail if we're racing with the buf daemon.
1609 */
1610 if (BUF_LOCK(bp,
1611 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1612 BO_LOCKPTR(bo)) == ENOLCK) {
1613 goto restart;
1614 }
1615 VNASSERT((bp->b_flags & B_DELWRI), vp,
1616 ("buf(%p) on dirty queue without DELWRI", bp));
1617
1618 bremfree(bp);
1619 bawrite(bp);
1620 BO_LOCK(bo);
1621 goto restartsync;
1622 }
1623 }
1624
1625 bufobj_wwait(bo, 0, 0);
1626 BO_UNLOCK(bo);
1627 vnode_pager_setsize(vp, length);
1628
1629 return (0);
1630}
1631
1632static void
1633buf_vlist_remove(struct buf *bp)
1634{
1635 struct bufv *bv;
1636
1637 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1638 ASSERT_BO_WLOCKED(bp->b_bufobj);
1639 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1640 (BX_VNDIRTY|BX_VNCLEAN),
1641 ("buf_vlist_remove: Buf %p is on two lists", bp));
1642 if (bp->b_xflags & BX_VNDIRTY)
1643 bv = &bp->b_bufobj->bo_dirty;
1644 else
1645 bv = &bp->b_bufobj->bo_clean;
1646 BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1647 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1648 bv->bv_cnt--;
1649 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1650}
1651
1652/*
1653 * Add the buffer to the sorted clean or dirty block list.
1654 *
1655 * NOTE: xflags is passed as a constant, optimizing this inline function!
1656 */
1657static void
1658buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1659{
1660 struct bufv *bv;
1661 struct buf *n;
1662 int error;
1663
1664 ASSERT_BO_WLOCKED(bo);
1665 KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0,
1666 ("dead bo %p", bo));
1667 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1668 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1669 bp->b_xflags |= xflags;
1670 if (xflags & BX_VNDIRTY)
1671 bv = &bo->bo_dirty;
1672 else
1673 bv = &bo->bo_clean;
1674
1675 /*
1676 * Keep the list ordered. Optimize empty list insertion. Assume
1677 * we tend to grow at the tail so lookup_le should usually be cheaper
1678 * than _ge.
1679 */
1680 if (bv->bv_cnt == 0 ||
1681 bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1682 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1683 else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1684 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1685 else
1686 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1687 error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1688 if (error)
1689 panic("buf_vlist_add: Preallocated nodes insufficient.");
1690 bv->bv_cnt++;
1691}
1692
1693/*
1694 * Lookup a buffer using the splay tree. Note that we specifically avoid
1695 * shadow buffers used in background bitmap writes.
1696 *
1697 * This code isn't quite efficient as it could be because we are maintaining
1698 * two sorted lists and do not know which list the block resides in.
1699 *
1700 * During a "make buildworld" the desired buffer is found at one of
1701 * the roots more than 60% of the time. Thus, checking both roots
1702 * before performing either splay eliminates unnecessary splays on the
1703 * first tree splayed.
1704 */
1705struct buf *
1706gbincore(struct bufobj *bo, daddr_t lblkno)
1707{
1708 struct buf *bp;
1709
1710 ASSERT_BO_LOCKED(bo);
1711 bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1712 if (bp != NULL)
1713 return (bp);
1714 return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1715}
1716
1717/*
1718 * Associate a buffer with a vnode.
1719 */
1720void
1721bgetvp(struct vnode *vp, struct buf *bp)
1722{
1723 struct bufobj *bo;
1724
1725 bo = &vp->v_bufobj;
1726 ASSERT_BO_WLOCKED(bo);
1727 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1728
1729 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1730 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1731 ("bgetvp: bp already attached! %p", bp));
1732
1733 vhold(vp);
1734 bp->b_vp = vp;
1735 bp->b_bufobj = bo;
1736 /*
1737 * Insert onto list for new vnode.
1738 */
1739 buf_vlist_add(bp, bo, BX_VNCLEAN);
1740}
1741
1742/*
1743 * Disassociate a buffer from a vnode.
1744 */
1745void
1746brelvp(struct buf *bp)
1747{
1748 struct bufobj *bo;
1749 struct vnode *vp;
1750
1751 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1752 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1753
1754 /*
1755 * Delete from old vnode list, if on one.
1756 */
1757 vp = bp->b_vp; /* XXX */
1758 bo = bp->b_bufobj;
1759 BO_LOCK(bo);
1760 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1761 buf_vlist_remove(bp);
1762 else
1763 panic("brelvp: Buffer %p not on queue.", bp);
1764 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1765 bo->bo_flag &= ~BO_ONWORKLST;
1766 mtx_lock(&sync_mtx);
1767 LIST_REMOVE(bo, bo_synclist);
1768 syncer_worklist_len--;
1769 mtx_unlock(&sync_mtx);
1770 }
1771 bp->b_vp = NULL;
1772 bp->b_bufobj = NULL;
1773 BO_UNLOCK(bo);
1774 vdrop(vp);
1775}
1776
1777/*
1778 * Add an item to the syncer work queue.
1779 */
1780static void
1781vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1782{
1783 int slot;
1784
1785 ASSERT_BO_WLOCKED(bo);
1786
1787 mtx_lock(&sync_mtx);
1788 if (bo->bo_flag & BO_ONWORKLST)
1789 LIST_REMOVE(bo, bo_synclist);
1790 else {
1791 bo->bo_flag |= BO_ONWORKLST;
1792 syncer_worklist_len++;
1793 }
1794
1795 if (delay > syncer_maxdelay - 2)
1796 delay = syncer_maxdelay - 2;
1797 slot = (syncer_delayno + delay) & syncer_mask;
1798
1799 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1800 mtx_unlock(&sync_mtx);
1801}
1802
1803static int
1804sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1805{
1806 int error, len;
1807
1808 mtx_lock(&sync_mtx);
1809 len = syncer_worklist_len - sync_vnode_count;
1810 mtx_unlock(&sync_mtx);
1811 error = SYSCTL_OUT(req, &len, sizeof(len));
1812 return (error);
1813}
1814
1815SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1816 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1817
1818static struct proc *updateproc;
1819static void sched_sync(void);
1820static struct kproc_desc up_kp = {
1821 "syncer",
1822 sched_sync,
1823 &updateproc
1824};
1825SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1826
1827static int
1828sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1829{
1830 struct vnode *vp;
1831 struct mount *mp;
1832
1833 *bo = LIST_FIRST(slp);
1834 if (*bo == NULL)
1835 return (0);
1836 vp = (*bo)->__bo_vnode; /* XXX */
1837 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1838 return (1);
1839 /*
1840 * We use vhold in case the vnode does not
1841 * successfully sync. vhold prevents the vnode from
1842 * going away when we unlock the sync_mtx so that
1843 * we can acquire the vnode interlock.
1844 */
1845 vholdl(vp);
1846 mtx_unlock(&sync_mtx);
1847 VI_UNLOCK(vp);
1848 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1849 vdrop(vp);
1850 mtx_lock(&sync_mtx);
1851 return (*bo == LIST_FIRST(slp));
1852 }
1853 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1854 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1855 VOP_UNLOCK(vp, 0);
1856 vn_finished_write(mp);
1857 BO_LOCK(*bo);
1858 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1859 /*
1860 * Put us back on the worklist. The worklist
1861 * routine will remove us from our current
1862 * position and then add us back in at a later
1863 * position.
1864 */
1865 vn_syncer_add_to_worklist(*bo, syncdelay);
1866 }
1867 BO_UNLOCK(*bo);
1868 vdrop(vp);
1869 mtx_lock(&sync_mtx);
1870 return (0);
1871}
1872
1873static int first_printf = 1;
1874
1875/*
1876 * System filesystem synchronizer daemon.
1877 */
1878static void
1879sched_sync(void)
1880{
1881 struct synclist *next, *slp;
1882 struct bufobj *bo;
1883 long starttime;
1884 struct thread *td = curthread;
1885 int last_work_seen;
1886 int net_worklist_len;
1887 int syncer_final_iter;
1888 int error;
1889
1890 last_work_seen = 0;
1891 syncer_final_iter = 0;
1892 syncer_state = SYNCER_RUNNING;
1893 starttime = time_uptime;
1894 td->td_pflags |= TDP_NORUNNINGBUF;
1895
1896 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1897 SHUTDOWN_PRI_LAST);
1898
1899 mtx_lock(&sync_mtx);
1900 for (;;) {
1901 if (syncer_state == SYNCER_FINAL_DELAY &&
1902 syncer_final_iter == 0) {
1903 mtx_unlock(&sync_mtx);
1904 kproc_suspend_check(td->td_proc);
1905 mtx_lock(&sync_mtx);
1906 }
1907 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1908 if (syncer_state != SYNCER_RUNNING &&
1909 starttime != time_uptime) {
1910 if (first_printf) {
1911 printf("\nSyncing disks, vnodes remaining...");
1912 first_printf = 0;
1913 }
1914 printf("%d ", net_worklist_len);
1915 }
1916 starttime = time_uptime;
1917
1918 /*
1919 * Push files whose dirty time has expired. Be careful
1920 * of interrupt race on slp queue.
1921 *
1922 * Skip over empty worklist slots when shutting down.
1923 */
1924 do {
1925 slp = &syncer_workitem_pending[syncer_delayno];
1926 syncer_delayno += 1;
1927 if (syncer_delayno == syncer_maxdelay)
1928 syncer_delayno = 0;
1929 next = &syncer_workitem_pending[syncer_delayno];
1930 /*
1931 * If the worklist has wrapped since the
1932 * it was emptied of all but syncer vnodes,
1933 * switch to the FINAL_DELAY state and run
1934 * for one more second.
1935 */
1936 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1937 net_worklist_len == 0 &&
1938 last_work_seen == syncer_delayno) {
1939 syncer_state = SYNCER_FINAL_DELAY;
1940 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1941 }
1942 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1943 syncer_worklist_len > 0);
1944
1945 /*
1946 * Keep track of the last time there was anything
1947 * on the worklist other than syncer vnodes.
1948 * Return to the SHUTTING_DOWN state if any
1949 * new work appears.
1950 */
1951 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1952 last_work_seen = syncer_delayno;
1953 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1954 syncer_state = SYNCER_SHUTTING_DOWN;
1955 while (!LIST_EMPTY(slp)) {
1956 error = sync_vnode(slp, &bo, td);
1957 if (error == 1) {
1958 LIST_REMOVE(bo, bo_synclist);
1959 LIST_INSERT_HEAD(next, bo, bo_synclist);
1960 continue;
1961 }
1962
1963 if (first_printf == 0)
1964 wdog_kern_pat(WD_LASTVAL);
1965
1966 }
1967 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1968 syncer_final_iter--;
1969 /*
1970 * The variable rushjob allows the kernel to speed up the
1971 * processing of the filesystem syncer process. A rushjob
1972 * value of N tells the filesystem syncer to process the next
1973 * N seconds worth of work on its queue ASAP. Currently rushjob
1974 * is used by the soft update code to speed up the filesystem
1975 * syncer process when the incore state is getting so far
1976 * ahead of the disk that the kernel memory pool is being
1977 * threatened with exhaustion.
1978 */
1979 if (rushjob > 0) {
1980 rushjob -= 1;
1981 continue;
1982 }
1983 /*
1984 * Just sleep for a short period of time between
1985 * iterations when shutting down to allow some I/O
1986 * to happen.
1987 *
1988 * If it has taken us less than a second to process the
1989 * current work, then wait. Otherwise start right over
1990 * again. We can still lose time if any single round
1991 * takes more than two seconds, but it does not really
1992 * matter as we are just trying to generally pace the
1993 * filesystem activity.
1994 */
1995 if (syncer_state != SYNCER_RUNNING ||
1996 time_uptime == starttime) {
1997 thread_lock(td);
1998 sched_prio(td, PPAUSE);
1999 thread_unlock(td);
2000 }
2001 if (syncer_state != SYNCER_RUNNING)
2002 cv_timedwait(&sync_wakeup, &sync_mtx,
2003 hz / SYNCER_SHUTDOWN_SPEEDUP);
2004 else if (time_uptime == starttime)
2005 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
2006 }
2007}
2008
2009/*
2010 * Request the syncer daemon to speed up its work.
2011 * We never push it to speed up more than half of its
2012 * normal turn time, otherwise it could take over the cpu.
2013 */
2014int
2015speedup_syncer(void)
2016{
2017 int ret = 0;
2018
2019 mtx_lock(&sync_mtx);
2020 if (rushjob < syncdelay / 2) {
2021 rushjob += 1;
2022 stat_rush_requests += 1;
2023 ret = 1;
2024 }
2025 mtx_unlock(&sync_mtx);
2026 cv_broadcast(&sync_wakeup);
2027 return (ret);
2028}
2029
2030/*
2031 * Tell the syncer to speed up its work and run though its work
2032 * list several times, then tell it to shut down.
2033 */
2034static void
2035syncer_shutdown(void *arg, int howto)
2036{
2037
2038 if (howto & RB_NOSYNC)
2039 return;
2040 mtx_lock(&sync_mtx);
2041 syncer_state = SYNCER_SHUTTING_DOWN;
2042 rushjob = 0;
2043 mtx_unlock(&sync_mtx);
2044 cv_broadcast(&sync_wakeup);
2045 kproc_shutdown(arg, howto);
2046}
2047
2048void
2049syncer_suspend(void)
2050{
2051
2052 syncer_shutdown(updateproc, 0);
2053}
2054
2055void
2056syncer_resume(void)
2057{
2058
2059 mtx_lock(&sync_mtx);
2060 first_printf = 1;
2061 syncer_state = SYNCER_RUNNING;
2062 mtx_unlock(&sync_mtx);
2063 cv_broadcast(&sync_wakeup);
2064 kproc_resume(updateproc);
2065}
2066
2067/*
2068 * Reassign a buffer from one vnode to another.
2069 * Used to assign file specific control information
2070 * (indirect blocks) to the vnode to which they belong.
2071 */
2072void
2073reassignbuf(struct buf *bp)
2074{
2075 struct vnode *vp;
2076 struct bufobj *bo;
2077 int delay;
2078#ifdef INVARIANTS
2079 struct bufv *bv;
2080#endif
2081
2082 vp = bp->b_vp;
2083 bo = bp->b_bufobj;
2084 ++reassignbufcalls;
2085
2086 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2087 bp, bp->b_vp, bp->b_flags);
2088 /*
2089 * B_PAGING flagged buffers cannot be reassigned because their vp
2090 * is not fully linked in.
2091 */
2092 if (bp->b_flags & B_PAGING)
2093 panic("cannot reassign paging buffer");
2094
2095 /*
2096 * Delete from old vnode list, if on one.
2097 */
2098 BO_LOCK(bo);
2099 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2100 buf_vlist_remove(bp);
2101 else
2102 panic("reassignbuf: Buffer %p not on queue.", bp);
2103 /*
2104 * If dirty, put on list of dirty buffers; otherwise insert onto list
2105 * of clean buffers.
2106 */
2107 if (bp->b_flags & B_DELWRI) {
2108 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2109 switch (vp->v_type) {
2110 case VDIR:
2111 delay = dirdelay;
2112 break;
2113 case VCHR:
2114 delay = metadelay;
2115 break;
2116 default:
2117 delay = filedelay;
2118 }
2119 vn_syncer_add_to_worklist(bo, delay);
2120 }
2121 buf_vlist_add(bp, bo, BX_VNDIRTY);
2122 } else {
2123 buf_vlist_add(bp, bo, BX_VNCLEAN);
2124
2125 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2126 mtx_lock(&sync_mtx);
2127 LIST_REMOVE(bo, bo_synclist);
2128 syncer_worklist_len--;
2129 mtx_unlock(&sync_mtx);
2130 bo->bo_flag &= ~BO_ONWORKLST;
2131 }
2132 }
2133#ifdef INVARIANTS
2134 bv = &bo->bo_clean;
2135 bp = TAILQ_FIRST(&bv->bv_hd);
2136 KASSERT(bp == NULL || bp->b_bufobj == bo,
2137 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2138 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2139 KASSERT(bp == NULL || bp->b_bufobj == bo,
2140 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2141 bv = &bo->bo_dirty;
2142 bp = TAILQ_FIRST(&bv->bv_hd);
2143 KASSERT(bp == NULL || bp->b_bufobj == bo,
2144 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2145 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2146 KASSERT(bp == NULL || bp->b_bufobj == bo,
2147 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2148#endif
2149 BO_UNLOCK(bo);
2150}
2151
2152/*
2153 * Increment the use and hold counts on the vnode, taking care to reference
2154 * the driver's usecount if this is a chardev. The vholdl() will remove
2155 * the vnode from the free list if it is presently free. Requires the
2156 * vnode interlock and returns with it held.
2157 */
2158static void
2159v_incr_usecount(struct vnode *vp)
2160{
2161
2162 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2163 vholdl(vp);
2164 vp->v_usecount++;
2165 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2166 dev_lock();
2167 vp->v_rdev->si_usecount++;
2168 dev_unlock();
2169 }
2170}
2171
2172/*
2173 * Turn a holdcnt into a use+holdcnt such that only one call to
2174 * v_decr_usecount is needed.
2175 */
2176static void
2177v_upgrade_usecount(struct vnode *vp)
2178{
2179
2180 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2181 vp->v_usecount++;
2182 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2183 dev_lock();
2184 vp->v_rdev->si_usecount++;
2185 dev_unlock();
2186 }
2187}
2188
2189/*
2190 * Decrement the vnode use and hold count along with the driver's usecount
2191 * if this is a chardev. The vdropl() below releases the vnode interlock
2192 * as it may free the vnode.
2193 */
2194static void
2195v_decr_usecount(struct vnode *vp)
2196{
2197
2198 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2199 VNASSERT(vp->v_usecount > 0, vp,
2200 ("v_decr_usecount: negative usecount"));
2201 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2202 vp->v_usecount--;
2203 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2204 dev_lock();
2205 vp->v_rdev->si_usecount--;
2206 dev_unlock();
2207 }
2208 vdropl(vp);
2209}
2210
2211/*
2212 * Decrement only the use count and driver use count. This is intended to
2213 * be paired with a follow on vdropl() to release the remaining hold count.
2214 * In this way we may vgone() a vnode with a 0 usecount without risk of
2215 * having it end up on a free list because the hold count is kept above 0.
2216 */
2217static void
2218v_decr_useonly(struct vnode *vp)
2219{
2220
2221 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2222 VNASSERT(vp->v_usecount > 0, vp,
2223 ("v_decr_useonly: negative usecount"));
2224 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2225 vp->v_usecount--;
2226 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2227 dev_lock();
2228 vp->v_rdev->si_usecount--;
2229 dev_unlock();
2230 }
2231}
2232
2233/*
2234 * Grab a particular vnode from the free list, increment its
2235 * reference count and lock it. VI_DOOMED is set if the vnode
2236 * is being destroyed. Only callers who specify LK_RETRY will
2237 * see doomed vnodes. If inactive processing was delayed in
2238 * vput try to do it here.
2239 */
2240int
2241vget(struct vnode *vp, int flags, struct thread *td)
2242{
2243 int error;
2244
2245 error = 0;
2246 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2247 ("vget: invalid lock operation"));
2248 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2249
2250 if ((flags & LK_INTERLOCK) == 0)
2251 VI_LOCK(vp);
2252 vholdl(vp);
2253 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2254 vdrop(vp);
2255 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2256 vp);
2257 return (error);
2258 }
2259 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2260 panic("vget: vn_lock failed to return ENOENT\n");
2261 VI_LOCK(vp);
2262 /* Upgrade our holdcnt to a usecount. */
2263 v_upgrade_usecount(vp);
2264 /*
2265 * We don't guarantee that any particular close will
2266 * trigger inactive processing so just make a best effort
2267 * here at preventing a reference to a removed file. If
2268 * we don't succeed no harm is done.
2269 */
2270 if (vp->v_iflag & VI_OWEINACT) {
2271 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2272 (flags & LK_NOWAIT) == 0)
2273 vinactive(vp, td);
2274 vp->v_iflag &= ~VI_OWEINACT;
2275 }
2276 VI_UNLOCK(vp);
2277 return (0);
2278}
2279
2280/*
2281 * Increase the reference count of a vnode.
2282 */
2283void
2284vref(struct vnode *vp)
2285{
2286
2287 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2288 VI_LOCK(vp);
2289 v_incr_usecount(vp);
2290 VI_UNLOCK(vp);
2291}
2292
2293/*
2294 * Return reference count of a vnode.
2295 *
2296 * The results of this call are only guaranteed when some mechanism other
2297 * than the VI lock is used to stop other processes from gaining references
2298 * to the vnode. This may be the case if the caller holds the only reference.
2299 * This is also useful when stale data is acceptable as race conditions may
2300 * be accounted for by some other means.
2301 */
2302int
2303vrefcnt(struct vnode *vp)
2304{
2305 int usecnt;
2306
2307 VI_LOCK(vp);
2308 usecnt = vp->v_usecount;
2309 VI_UNLOCK(vp);
2310
2311 return (usecnt);
2312}
2313
2314#define VPUTX_VRELE 1
2315#define VPUTX_VPUT 2
2316#define VPUTX_VUNREF 3
2317
2318static void
2319vputx(struct vnode *vp, int func)
2320{
2321 int error;
2322
2323 KASSERT(vp != NULL, ("vputx: null vp"));
2324 if (func == VPUTX_VUNREF)
2325 ASSERT_VOP_LOCKED(vp, "vunref");
2326 else if (func == VPUTX_VPUT)
2327 ASSERT_VOP_LOCKED(vp, "vput");
2328 else
2329 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2330 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2331 VI_LOCK(vp);
2332
2333 /* Skip this v_writecount check if we're going to panic below. */
2334 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2335 ("vputx: missed vn_close"));
2336 error = 0;
2337
2338 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2339 vp->v_usecount == 1)) {
2340 if (func == VPUTX_VPUT)
2341 VOP_UNLOCK(vp, 0);
2342 v_decr_usecount(vp);
2343 return;
2344 }
2345
2346 if (vp->v_usecount != 1) {
2347 vprint("vputx: negative ref count", vp);
2348 panic("vputx: negative ref cnt");
2349 }
2350 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2351 /*
2352 * We want to hold the vnode until the inactive finishes to
2353 * prevent vgone() races. We drop the use count here and the
2354 * hold count below when we're done.
2355 */
2356 v_decr_useonly(vp);
2357 /*
2358 * We must call VOP_INACTIVE with the node locked. Mark
2359 * as VI_DOINGINACT to avoid recursion.
2360 */
2361 vp->v_iflag |= VI_OWEINACT;
2362 switch (func) {
2363 case VPUTX_VRELE:
2364 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2365 VI_LOCK(vp);
2366 break;
2367 case VPUTX_VPUT:
2368 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2369 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2370 LK_NOWAIT);
2371 VI_LOCK(vp);
2372 }
2373 break;
2374 case VPUTX_VUNREF:
2375 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2376 error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2377 VI_LOCK(vp);
2378 }
2379 break;
2380 }
2381 if (vp->v_usecount > 0)
2382 vp->v_iflag &= ~VI_OWEINACT;
2383 if (error == 0) {
2384 if (vp->v_iflag & VI_OWEINACT)
2385 vinactive(vp, curthread);
2386 if (func != VPUTX_VUNREF)
2387 VOP_UNLOCK(vp, 0);
2388 }
2389 vdropl(vp);
2390}
2391
2392/*
2393 * Vnode put/release.
2394 * If count drops to zero, call inactive routine and return to freelist.
2395 */
2396void
2397vrele(struct vnode *vp)
2398{
2399
2400 vputx(vp, VPUTX_VRELE);
2401}
2402
2403/*
2404 * Release an already locked vnode. This give the same effects as
2405 * unlock+vrele(), but takes less time and avoids releasing and
2406 * re-aquiring the lock (as vrele() acquires the lock internally.)
2407 */
2408void
2409vput(struct vnode *vp)
2410{
2411
2412 vputx(vp, VPUTX_VPUT);
2413}
2414
2415/*
2416 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2417 */
2418void
2419vunref(struct vnode *vp)
2420{
2421
2422 vputx(vp, VPUTX_VUNREF);
2423}
2424
2425/*
2426 * Somebody doesn't want the vnode recycled.
2427 */
2428void
2429vhold(struct vnode *vp)
2430{
2431
2432 VI_LOCK(vp);
2433 vholdl(vp);
2434 VI_UNLOCK(vp);
2435}
2436
2437/*
2438 * Increase the hold count and activate if this is the first reference.
2439 */
2440void
2441vholdl(struct vnode *vp)
2442{
2443 struct mount *mp;
2444
2445 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2446#ifdef INVARIANTS
2447 /* getnewvnode() calls v_incr_usecount() without holding interlock. */
2448 if (vp->v_type != VNON || vp->v_data != NULL)
2449 ASSERT_VI_LOCKED(vp, "vholdl");
2450#endif
2451 vp->v_holdcnt++;
2452 if ((vp->v_iflag & VI_FREE) == 0)
2453 return;
2454 VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count"));
2455 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2456 /*
2457 * Remove a vnode from the free list, mark it as in use,
2458 * and put it on the active list.
2459 */
2460 mtx_lock(&vnode_free_list_mtx);
2461 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2462 freevnodes--;
2463 vp->v_iflag &= ~VI_FREE;
2464 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2465 ("Activating already active vnode"));
2466 vp->v_iflag |= VI_ACTIVE;
2467 mp = vp->v_mount;
2468 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2469 mp->mnt_activevnodelistsize++;
2470 mtx_unlock(&vnode_free_list_mtx);
2471}
2472
2473/*
2474 * Note that there is one less who cares about this vnode.
2475 * vdrop() is the opposite of vhold().
2476 */
2477void
2478vdrop(struct vnode *vp)
2479{
2480
2481 VI_LOCK(vp);
2482 vdropl(vp);
2483}
2484
2485/*
2486 * Drop the hold count of the vnode. If this is the last reference to
2487 * the vnode we place it on the free list unless it has been vgone'd
2488 * (marked VI_DOOMED) in which case we will free it.
2489 *
2490 * Because the vnode vm object keeps a hold reference on the vnode if
2491 * there is at least one resident non-cached page, the vnode cannot
2492 * leave the active list without the page cleanup done.
2493 */
2494void
2495vdropl(struct vnode *vp)
2496{
2497 struct bufobj *bo;
2498 struct mount *mp;
2499 int active;
2500
2501 ASSERT_VI_LOCKED(vp, "vdropl");
2502 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2503 if (vp->v_holdcnt <= 0)
2504 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2505 vp->v_holdcnt--;
2506 if (vp->v_holdcnt > 0) {
2507 VI_UNLOCK(vp);
2508 return;
2509 }
2510 if ((vp->v_iflag & VI_DOOMED) == 0) {
2511 /*
2512 * Mark a vnode as free: remove it from its active list
2513 * and put it up for recycling on the freelist.
2514 */
2515 VNASSERT(vp->v_op != NULL, vp,
2516 ("vdropl: vnode already reclaimed."));
2517 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2518 ("vnode already free"));
2519 VNASSERT(vp->v_holdcnt == 0, vp,
2520 ("vdropl: freeing when we shouldn't"));
2521 active = vp->v_iflag & VI_ACTIVE;
2522 if ((vp->v_iflag & VI_OWEINACT) == 0) {
2523 vp->v_iflag &= ~VI_ACTIVE;
2524 mp = vp->v_mount;
2525 mtx_lock(&vnode_free_list_mtx);
2526 if (active) {
2527 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2528 v_actfreelist);
2529 mp->mnt_activevnodelistsize--;
2530 }
2531 TAILQ_INSERT_TAIL(&vnode_free_list, vp,
2532 v_actfreelist);
2533 freevnodes++;
2534 vp->v_iflag |= VI_FREE;
2535 mtx_unlock(&vnode_free_list_mtx);
2536 } else {
2537 atomic_add_long(&free_owe_inact, 1);
2538 }
2539 VI_UNLOCK(vp);
2540 return;
2541 }
2542 /*
2543 * The vnode has been marked for destruction, so free it.
2544 *
2545 * The vnode will be returned to the zone where it will
2546 * normally remain until it is needed for another vnode. We
2547 * need to cleanup (or verify that the cleanup has already
2548 * been done) any residual data left from its current use
2549 * so as not to contaminate the freshly allocated vnode.
2550 */
2551 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2552 atomic_subtract_long(&numvnodes, 1);
2553 bo = &vp->v_bufobj;
2554 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2555 ("cleaned vnode still on the free list."));
2556 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2557 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2558 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2559 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2560 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2561 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2562 VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2563 ("clean blk trie not empty"));
2564 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2565 VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2566 ("dirty blk trie not empty"));
2567 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2568 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2569 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2570 VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp,
2571 ("Dangling rangelock waiters"));
2572 VI_UNLOCK(vp);
2573#ifdef MAC
2574 mac_vnode_destroy(vp);
2575#endif
2576 if (vp->v_pollinfo != NULL) {
2577 destroy_vpollinfo(vp->v_pollinfo);
2578 vp->v_pollinfo = NULL;
2579 }
2580#ifdef INVARIANTS
2581 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2582 vp->v_op = NULL;
2583#endif
2584 bzero(&vp->v_un, sizeof(vp->v_un));
2585 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
2586 vp->v_iflag = 0;
2587 vp->v_vflag = 0;
2588 bo->bo_flag = 0;
2589 uma_zfree(vnode_zone, vp);
2590}
2591
2592/*
2593 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2594 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2595 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2596 * failed lock upgrade.
2597 */
2598void
2599vinactive(struct vnode *vp, struct thread *td)
2600{
2601 struct vm_object *obj;
2602
2603 ASSERT_VOP_ELOCKED(vp, "vinactive");
2604 ASSERT_VI_LOCKED(vp, "vinactive");
2605 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2606 ("vinactive: recursed on VI_DOINGINACT"));
2607 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2608 vp->v_iflag |= VI_DOINGINACT;
2609 vp->v_iflag &= ~VI_OWEINACT;
2610 VI_UNLOCK(vp);
2611 /*
2612 * Before moving off the active list, we must be sure that any
2613 * modified pages are converted into the vnode's dirty
2614 * buffers, since these will no longer be checked once the
2615 * vnode is on the inactive list.
2616 *
2617 * The write-out of the dirty pages is asynchronous. At the
2618 * point that VOP_INACTIVE() is called, there could still be
2619 * pending I/O and dirty pages in the object.
2620 */
2621 obj = vp->v_object;
2622 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2623 VM_OBJECT_WLOCK(obj);
2624 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2625 VM_OBJECT_WUNLOCK(obj);
2626 }
2627 VOP_INACTIVE(vp, td);
2628 VI_LOCK(vp);
2629 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2630 ("vinactive: lost VI_DOINGINACT"));
2631 vp->v_iflag &= ~VI_DOINGINACT;
2632}
2633
2634/*
2635 * Remove any vnodes in the vnode table belonging to mount point mp.
2636 *
2637 * If FORCECLOSE is not specified, there should not be any active ones,
2638 * return error if any are found (nb: this is a user error, not a
2639 * system error). If FORCECLOSE is specified, detach any active vnodes
2640 * that are found.
2641 *
2642 * If WRITECLOSE is set, only flush out regular file vnodes open for
2643 * writing.
2644 *
2645 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2646 *
2647 * `rootrefs' specifies the base reference count for the root vnode
2648 * of this filesystem. The root vnode is considered busy if its
2649 * v_usecount exceeds this value. On a successful return, vflush(, td)
2650 * will call vrele() on the root vnode exactly rootrefs times.
2651 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2652 * be zero.
2653 */
2654#ifdef DIAGNOSTIC
2655static int busyprt = 0; /* print out busy vnodes */
2656SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2657#endif
2658
2659int
2660vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2661{
2662 struct vnode *vp, *mvp, *rootvp = NULL;
2663 struct vattr vattr;
2664 int busy = 0, error;
2665
2666 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2667 rootrefs, flags);
2668 if (rootrefs > 0) {
2669 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2670 ("vflush: bad args"));
2671 /*
2672 * Get the filesystem root vnode. We can vput() it
2673 * immediately, since with rootrefs > 0, it won't go away.
2674 */
2675 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2676 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2677 __func__, error);
2678 return (error);
2679 }
2680 vput(rootvp);
2681 }
2682loop:
2683 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2684 vholdl(vp);
2685 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2686 if (error) {
2687 vdrop(vp);
2688 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2689 goto loop;
2690 }
2691 /*
2692 * Skip over a vnodes marked VV_SYSTEM.
2693 */
2694 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2695 VOP_UNLOCK(vp, 0);
2696 vdrop(vp);
2697 continue;
2698 }
2699 /*
2700 * If WRITECLOSE is set, flush out unlinked but still open
2701 * files (even if open only for reading) and regular file
2702 * vnodes open for writing.
2703 */
2704 if (flags & WRITECLOSE) {
2705 if (vp->v_object != NULL) {
2706 VM_OBJECT_WLOCK(vp->v_object);
2707 vm_object_page_clean(vp->v_object, 0, 0, 0);
2708 VM_OBJECT_WUNLOCK(vp->v_object);
2709 }
2710 error = VOP_FSYNC(vp, MNT_WAIT, td);
2711 if (error != 0) {
2712 VOP_UNLOCK(vp, 0);
2713 vdrop(vp);
2714 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2715 return (error);
2716 }
2717 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2718 VI_LOCK(vp);
2719
2720 if ((vp->v_type == VNON ||
2721 (error == 0 && vattr.va_nlink > 0)) &&
2722 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2723 VOP_UNLOCK(vp, 0);
2724 vdropl(vp);
2725 continue;
2726 }
2727 } else
2728 VI_LOCK(vp);
2729 /*
2730 * With v_usecount == 0, all we need to do is clear out the
2731 * vnode data structures and we are done.
2732 *
2733 * If FORCECLOSE is set, forcibly close the vnode.
2734 */
2735 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2736 vgonel(vp);
2737 } else {
2738 busy++;
2739#ifdef DIAGNOSTIC
2740 if (busyprt)
2741 vprint("vflush: busy vnode", vp);
2742#endif
2743 }
2744 VOP_UNLOCK(vp, 0);
2745 vdropl(vp);
2746 }
2747 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2748 /*
2749 * If just the root vnode is busy, and if its refcount
2750 * is equal to `rootrefs', then go ahead and kill it.
2751 */
2752 VI_LOCK(rootvp);
2753 KASSERT(busy > 0, ("vflush: not busy"));
2754 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2755 ("vflush: usecount %d < rootrefs %d",
2756 rootvp->v_usecount, rootrefs));
2757 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2758 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2759 vgone(rootvp);
2760 VOP_UNLOCK(rootvp, 0);
2761 busy = 0;
2762 } else
2763 VI_UNLOCK(rootvp);
2764 }
2765 if (busy) {
2766 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2767 busy);
2768 return (EBUSY);
2769 }
2770 for (; rootrefs > 0; rootrefs--)
2771 vrele(rootvp);
2772 return (0);
2773}
2774
2775/*
2776 * Recycle an unused vnode to the front of the free list.
2777 */
2778int
2779vrecycle(struct vnode *vp)
2780{
2781 int recycled;
2782
2783 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2784 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2785 recycled = 0;
2786 VI_LOCK(vp);
2787 if (vp->v_usecount == 0) {
2788 recycled = 1;
2789 vgonel(vp);
2790 }
2791 VI_UNLOCK(vp);
2792 return (recycled);
2793}
2794
2795/*
2796 * Eliminate all activity associated with a vnode
2797 * in preparation for reuse.
2798 */
2799void
2800vgone(struct vnode *vp)
2801{
2802 VI_LOCK(vp);
2803 vgonel(vp);
2804 VI_UNLOCK(vp);
2805}
2806
2807static void
2808notify_lowervp_vfs_dummy(struct mount *mp __unused,
2809 struct vnode *lowervp __unused)
2810{
2811}
2812
2813/*
2814 * Notify upper mounts about reclaimed or unlinked vnode.
2815 */
2816void
2817vfs_notify_upper(struct vnode *vp, int event)
2818{
2819 static struct vfsops vgonel_vfsops = {
2820 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2821 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2822 };
2823 struct mount *mp, *ump, *mmp;
2824
2825 mp = vp->v_mount;
2826 if (mp == NULL)
2827 return;
2828
2829 MNT_ILOCK(mp);
2830 if (TAILQ_EMPTY(&mp->mnt_uppers))
2831 goto unlock;
2832 MNT_IUNLOCK(mp);
2833 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2834 mmp->mnt_op = &vgonel_vfsops;
2835 mmp->mnt_kern_flag |= MNTK_MARKER;
2836 MNT_ILOCK(mp);
2837 mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2838 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2839 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2840 ump = TAILQ_NEXT(ump, mnt_upper_link);
2841 continue;
2842 }
2843 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2844 MNT_IUNLOCK(mp);
2845 switch (event) {
2846 case VFS_NOTIFY_UPPER_RECLAIM:
2847 VFS_RECLAIM_LOWERVP(ump, vp);
2848 break;
2849 case VFS_NOTIFY_UPPER_UNLINK:
2850 VFS_UNLINK_LOWERVP(ump, vp);
2851 break;
2852 default:
2853 KASSERT(0, ("invalid event %d", event));
2854 break;
2855 }
2856 MNT_ILOCK(mp);
2857 ump = TAILQ_NEXT(mmp, mnt_upper_link);
2858 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2859 }
2860 free(mmp, M_TEMP);
2861 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2862 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2863 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2864 wakeup(&mp->mnt_uppers);
2865 }
2866unlock:
2867 MNT_IUNLOCK(mp);
2868}
2869
2870/*
2871 * vgone, with the vp interlock held.
2872 */
2873static void
2874vgonel(struct vnode *vp)
2875{
2876 struct thread *td;
2877 int oweinact;
2878 int active;
2879 struct mount *mp;
2880
2881 ASSERT_VOP_ELOCKED(vp, "vgonel");
2882 ASSERT_VI_LOCKED(vp, "vgonel");
2883 VNASSERT(vp->v_holdcnt, vp,
2884 ("vgonel: vp %p has no reference.", vp));
2885 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2886 td = curthread;
2887
2888 /*
2889 * Don't vgonel if we're already doomed.
2890 */
2891 if (vp->v_iflag & VI_DOOMED)
2892 return;
2893 vp->v_iflag |= VI_DOOMED;
2894
2895 /*
2896 * Check to see if the vnode is in use. If so, we have to call
2897 * VOP_CLOSE() and VOP_INACTIVE().
2898 */
2899 active = vp->v_usecount;
2900 oweinact = (vp->v_iflag & VI_OWEINACT);
2901 VI_UNLOCK(vp);
2902 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2903
2904 /*
2905 * If purging an active vnode, it must be closed and
2906 * deactivated before being reclaimed.
2907 */
2908 if (active)
2909 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2910 if (oweinact || active) {
2911 VI_LOCK(vp);
2912 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2913 vinactive(vp, td);
2914 VI_UNLOCK(vp);
2915 }
2916 if (vp->v_type == VSOCK)
2917 vfs_unp_reclaim(vp);
2918
2919 /*
2920 * Clean out any buffers associated with the vnode.
2921 * If the flush fails, just toss the buffers.
2922 */
2923 mp = NULL;
2924 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2925 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2926 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
2927 while (vinvalbuf(vp, 0, 0, 0) != 0)
2928 ;
2929 }
2930
2931 BO_LOCK(&vp->v_bufobj);
2932 KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
2933 vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
2934 TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
2935 vp->v_bufobj.bo_clean.bv_cnt == 0,
2936 ("vp %p bufobj not invalidated", vp));
| 43
44#include "opt_compat.h"
45#include "opt_ddb.h"
46#include "opt_watchdog.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/bio.h>
51#include <sys/buf.h>
52#include <sys/condvar.h>
53#include <sys/conf.h>
54#include <sys/dirent.h>
55#include <sys/event.h>
56#include <sys/eventhandler.h>
57#include <sys/extattr.h>
58#include <sys/file.h>
59#include <sys/fcntl.h>
60#include <sys/jail.h>
61#include <sys/kdb.h>
62#include <sys/kernel.h>
63#include <sys/kthread.h>
64#include <sys/lockf.h>
65#include <sys/malloc.h>
66#include <sys/mount.h>
67#include <sys/namei.h>
68#include <sys/pctrie.h>
69#include <sys/priv.h>
70#include <sys/reboot.h>
71#include <sys/rwlock.h>
72#include <sys/sched.h>
73#include <sys/sleepqueue.h>
74#include <sys/smp.h>
75#include <sys/stat.h>
76#include <sys/sysctl.h>
77#include <sys/syslog.h>
78#include <sys/vmmeter.h>
79#include <sys/vnode.h>
80#include <sys/watchdog.h>
81
82#include <machine/stdarg.h>
83
84#include <security/mac/mac_framework.h>
85
86#include <vm/vm.h>
87#include <vm/vm_object.h>
88#include <vm/vm_extern.h>
89#include <vm/pmap.h>
90#include <vm/vm_map.h>
91#include <vm/vm_page.h>
92#include <vm/vm_kern.h>
93#include <vm/uma.h>
94
95#ifdef DDB
96#include <ddb/ddb.h>
97#endif
98
99static void delmntque(struct vnode *vp);
100static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 int slpflag, int slptimeo);
102static void syncer_shutdown(void *arg, int howto);
103static int vtryrecycle(struct vnode *vp);
104static void v_incr_usecount(struct vnode *);
105static void v_decr_usecount(struct vnode *);
106static void v_decr_useonly(struct vnode *);
107static void v_upgrade_usecount(struct vnode *);
108static void vnlru_free(int);
109static void vgonel(struct vnode *);
110static void vfs_knllock(void *arg);
111static void vfs_knlunlock(void *arg);
112static void vfs_knl_assert_locked(void *arg);
113static void vfs_knl_assert_unlocked(void *arg);
114static void destroy_vpollinfo(struct vpollinfo *vi);
115
116/*
117 * Number of vnodes in existence. Increased whenever getnewvnode()
118 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
119 */
120static unsigned long numvnodes;
121
122SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123 "Number of vnodes in existence");
124
125static u_long vnodes_created;
126SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
127 0, "Number of vnodes created by getnewvnode");
128
129/*
130 * Conversion tables for conversion from vnode types to inode formats
131 * and back.
132 */
133enum vtype iftovt_tab[16] = {
134 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
135 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
136};
137int vttoif_tab[10] = {
138 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
139 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
140};
141
142/*
143 * List of vnodes that are ready for recycling.
144 */
145static TAILQ_HEAD(freelst, vnode) vnode_free_list;
146
147/*
148 * Free vnode target. Free vnodes may simply be files which have been stat'd
149 * but not read. This is somewhat common, and a small cache of such files
150 * should be kept to avoid recreation costs.
151 */
152static u_long wantfreevnodes;
153SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
154/* Number of vnodes in the free list. */
155static u_long freevnodes;
156SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
157 "Number of vnodes in the free list");
158
159static int vlru_allow_cache_src;
160SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
161 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
162
163static u_long recycles_count;
164SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
165 "Number of vnodes recycled to avoid exceding kern.maxvnodes");
166
167/*
168 * Various variables used for debugging the new implementation of
169 * reassignbuf().
170 * XXX these are probably of (very) limited utility now.
171 */
172static int reassignbufcalls;
173SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
174 "Number of calls to reassignbuf");
175
176static u_long free_owe_inact;
177SYSCTL_ULONG(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, 0,
178 "Number of times free vnodes kept on active list due to VFS "
179 "owing inactivation");
180
181/*
182 * Cache for the mount type id assigned to NFS. This is used for
183 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
184 */
185int nfs_mount_type = -1;
186
187/* To keep more than one thread at a time from running vfs_getnewfsid */
188static struct mtx mntid_mtx;
189
190/*
191 * Lock for any access to the following:
192 * vnode_free_list
193 * numvnodes
194 * freevnodes
195 */
196static struct mtx vnode_free_list_mtx;
197
198/* Publicly exported FS */
199struct nfs_public nfs_pub;
200
201static uma_zone_t buf_trie_zone;
202
203/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
204static uma_zone_t vnode_zone;
205static uma_zone_t vnodepoll_zone;
206
207/*
208 * The workitem queue.
209 *
210 * It is useful to delay writes of file data and filesystem metadata
211 * for tens of seconds so that quickly created and deleted files need
212 * not waste disk bandwidth being created and removed. To realize this,
213 * we append vnodes to a "workitem" queue. When running with a soft
214 * updates implementation, most pending metadata dependencies should
215 * not wait for more than a few seconds. Thus, mounted on block devices
216 * are delayed only about a half the time that file data is delayed.
217 * Similarly, directory updates are more critical, so are only delayed
218 * about a third the time that file data is delayed. Thus, there are
219 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
220 * one each second (driven off the filesystem syncer process). The
221 * syncer_delayno variable indicates the next queue that is to be processed.
222 * Items that need to be processed soon are placed in this queue:
223 *
224 * syncer_workitem_pending[syncer_delayno]
225 *
226 * A delay of fifteen seconds is done by placing the request fifteen
227 * entries later in the queue:
228 *
229 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
230 *
231 */
232static int syncer_delayno;
233static long syncer_mask;
234LIST_HEAD(synclist, bufobj);
235static struct synclist *syncer_workitem_pending;
236/*
237 * The sync_mtx protects:
238 * bo->bo_synclist
239 * sync_vnode_count
240 * syncer_delayno
241 * syncer_state
242 * syncer_workitem_pending
243 * syncer_worklist_len
244 * rushjob
245 */
246static struct mtx sync_mtx;
247static struct cv sync_wakeup;
248
249#define SYNCER_MAXDELAY 32
250static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
251static int syncdelay = 30; /* max time to delay syncing data */
252static int filedelay = 30; /* time to delay syncing files */
253SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
254 "Time to delay syncing files (in seconds)");
255static int dirdelay = 29; /* time to delay syncing directories */
256SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
257 "Time to delay syncing directories (in seconds)");
258static int metadelay = 28; /* time to delay syncing metadata */
259SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
260 "Time to delay syncing metadata (in seconds)");
261static int rushjob; /* number of slots to run ASAP */
262static int stat_rush_requests; /* number of times I/O speeded up */
263SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
264 "Number of times I/O speeded up (rush requests)");
265
266/*
267 * When shutting down the syncer, run it at four times normal speed.
268 */
269#define SYNCER_SHUTDOWN_SPEEDUP 4
270static int sync_vnode_count;
271static int syncer_worklist_len;
272static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
273 syncer_state;
274
275/*
276 * Number of vnodes we want to exist at any one time. This is mostly used
277 * to size hash tables in vnode-related code. It is normally not used in
278 * getnewvnode(), as wantfreevnodes is normally nonzero.)
279 *
280 * XXX desiredvnodes is historical cruft and should not exist.
281 */
282int desiredvnodes;
283
284static int
285sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS)
286{
287 int error, old_desiredvnodes;
288
289 old_desiredvnodes = desiredvnodes;
290 if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0)
291 return (error);
292 if (old_desiredvnodes != desiredvnodes) {
293 vfs_hash_changesize(desiredvnodes);
294 cache_changesize(desiredvnodes);
295 }
296 return (0);
297}
298
299SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes,
300 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0,
301 sysctl_update_desiredvnodes, "I", "Maximum number of vnodes");
302SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
303 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
304static int vnlru_nowhere;
305SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
306 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
307
308/* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
309static int vnsz2log;
310
311/*
312 * Support for the bufobj clean & dirty pctrie.
313 */
314static void *
315buf_trie_alloc(struct pctrie *ptree)
316{
317
318 return uma_zalloc(buf_trie_zone, M_NOWAIT);
319}
320
321static void
322buf_trie_free(struct pctrie *ptree, void *node)
323{
324
325 uma_zfree(buf_trie_zone, node);
326}
327PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
328
329/*
330 * Initialize the vnode management data structures.
331 *
332 * Reevaluate the following cap on the number of vnodes after the physical
333 * memory size exceeds 512GB. In the limit, as the physical memory size
334 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
335 */
336#ifndef MAXVNODES_MAX
337#define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
338#endif
339
340/*
341 * Initialize a vnode as it first enters the zone.
342 */
343static int
344vnode_init(void *mem, int size, int flags)
345{
346 struct vnode *vp;
347 struct bufobj *bo;
348
349 vp = mem;
350 bzero(vp, size);
351 /*
352 * Setup locks.
353 */
354 vp->v_vnlock = &vp->v_lock;
355 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
356 /*
357 * By default, don't allow shared locks unless filesystems opt-in.
358 */
359 lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT,
360 LK_NOSHARE | LK_IS_VNODE);
361 /*
362 * Initialize bufobj.
363 */
364 bo = &vp->v_bufobj;
365 bo->__bo_vnode = vp;
366 rw_init(BO_LOCKPTR(bo), "bufobj interlock");
367 bo->bo_private = vp;
368 TAILQ_INIT(&bo->bo_clean.bv_hd);
369 TAILQ_INIT(&bo->bo_dirty.bv_hd);
370 /*
371 * Initialize namecache.
372 */
373 LIST_INIT(&vp->v_cache_src);
374 TAILQ_INIT(&vp->v_cache_dst);
375 /*
376 * Initialize rangelocks.
377 */
378 rangelock_init(&vp->v_rl);
379 return (0);
380}
381
382/*
383 * Free a vnode when it is cleared from the zone.
384 */
385static void
386vnode_fini(void *mem, int size)
387{
388 struct vnode *vp;
389 struct bufobj *bo;
390
391 vp = mem;
392 rangelock_destroy(&vp->v_rl);
393 lockdestroy(vp->v_vnlock);
394 mtx_destroy(&vp->v_interlock);
395 bo = &vp->v_bufobj;
396 rw_destroy(BO_LOCKPTR(bo));
397}
398
399static void
400vntblinit(void *dummy __unused)
401{
402 u_int i;
403 int physvnodes, virtvnodes;
404
405 /*
406 * Desiredvnodes is a function of the physical memory size and the
407 * kernel's heap size. Generally speaking, it scales with the
408 * physical memory size. The ratio of desiredvnodes to physical pages
409 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
410 * marginal ratio of desiredvnodes to physical pages is one to
411 * sixteen. However, desiredvnodes is limited by the kernel's heap
412 * size. The memory required by desiredvnodes vnodes and vm objects
413 * may not exceed one seventh of the kernel's heap size.
414 */
415 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
416 cnt.v_page_count) / 16;
417 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
418 sizeof(struct vnode)));
419 desiredvnodes = min(physvnodes, virtvnodes);
420 if (desiredvnodes > MAXVNODES_MAX) {
421 if (bootverbose)
422 printf("Reducing kern.maxvnodes %d -> %d\n",
423 desiredvnodes, MAXVNODES_MAX);
424 desiredvnodes = MAXVNODES_MAX;
425 }
426 wantfreevnodes = desiredvnodes / 4;
427 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
428 TAILQ_INIT(&vnode_free_list);
429 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
430 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
431 vnode_init, vnode_fini, UMA_ALIGN_PTR, 0);
432 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
433 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
434 /*
435 * Preallocate enough nodes to support one-per buf so that
436 * we can not fail an insert. reassignbuf() callers can not
437 * tolerate the insertion failure.
438 */
439 buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
440 NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
441 UMA_ZONE_NOFREE | UMA_ZONE_VM);
442 uma_prealloc(buf_trie_zone, nbuf);
443 /*
444 * Initialize the filesystem syncer.
445 */
446 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
447 &syncer_mask);
448 syncer_maxdelay = syncer_mask + 1;
449 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
450 cv_init(&sync_wakeup, "syncer");
451 for (i = 1; i <= sizeof(struct vnode); i <<= 1)
452 vnsz2log++;
453 vnsz2log--;
454}
455SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
456
457
458/*
459 * Mark a mount point as busy. Used to synchronize access and to delay
460 * unmounting. Eventually, mountlist_mtx is not released on failure.
461 *
462 * vfs_busy() is a custom lock, it can block the caller.
463 * vfs_busy() only sleeps if the unmount is active on the mount point.
464 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
465 * vnode belonging to mp.
466 *
467 * Lookup uses vfs_busy() to traverse mount points.
468 * root fs var fs
469 * / vnode lock A / vnode lock (/var) D
470 * /var vnode lock B /log vnode lock(/var/log) E
471 * vfs_busy lock C vfs_busy lock F
472 *
473 * Within each file system, the lock order is C->A->B and F->D->E.
474 *
475 * When traversing across mounts, the system follows that lock order:
476 *
477 * C->A->B
478 * |
479 * +->F->D->E
480 *
481 * The lookup() process for namei("/var") illustrates the process:
482 * VOP_LOOKUP() obtains B while A is held
483 * vfs_busy() obtains a shared lock on F while A and B are held
484 * vput() releases lock on B
485 * vput() releases lock on A
486 * VFS_ROOT() obtains lock on D while shared lock on F is held
487 * vfs_unbusy() releases shared lock on F
488 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
489 * Attempt to lock A (instead of vp_crossmp) while D is held would
490 * violate the global order, causing deadlocks.
491 *
492 * dounmount() locks B while F is drained.
493 */
494int
495vfs_busy(struct mount *mp, int flags)
496{
497
498 MPASS((flags & ~MBF_MASK) == 0);
499 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
500
501 MNT_ILOCK(mp);
502 MNT_REF(mp);
503 /*
504 * If mount point is currenly being unmounted, sleep until the
505 * mount point fate is decided. If thread doing the unmounting fails,
506 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
507 * that this mount point has survived the unmount attempt and vfs_busy
508 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
509 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
510 * about to be really destroyed. vfs_busy needs to release its
511 * reference on the mount point in this case and return with ENOENT,
512 * telling the caller that mount mount it tried to busy is no longer
513 * valid.
514 */
515 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
516 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
517 MNT_REL(mp);
518 MNT_IUNLOCK(mp);
519 CTR1(KTR_VFS, "%s: failed busying before sleeping",
520 __func__);
521 return (ENOENT);
522 }
523 if (flags & MBF_MNTLSTLOCK)
524 mtx_unlock(&mountlist_mtx);
525 mp->mnt_kern_flag |= MNTK_MWAIT;
526 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
527 if (flags & MBF_MNTLSTLOCK)
528 mtx_lock(&mountlist_mtx);
529 MNT_ILOCK(mp);
530 }
531 if (flags & MBF_MNTLSTLOCK)
532 mtx_unlock(&mountlist_mtx);
533 mp->mnt_lockref++;
534 MNT_IUNLOCK(mp);
535 return (0);
536}
537
538/*
539 * Free a busy filesystem.
540 */
541void
542vfs_unbusy(struct mount *mp)
543{
544
545 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
546 MNT_ILOCK(mp);
547 MNT_REL(mp);
548 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
549 mp->mnt_lockref--;
550 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
551 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
552 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
553 mp->mnt_kern_flag &= ~MNTK_DRAINING;
554 wakeup(&mp->mnt_lockref);
555 }
556 MNT_IUNLOCK(mp);
557}
558
559/*
560 * Lookup a mount point by filesystem identifier.
561 */
562struct mount *
563vfs_getvfs(fsid_t *fsid)
564{
565 struct mount *mp;
566
567 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
568 mtx_lock(&mountlist_mtx);
569 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
570 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
571 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
572 vfs_ref(mp);
573 mtx_unlock(&mountlist_mtx);
574 return (mp);
575 }
576 }
577 mtx_unlock(&mountlist_mtx);
578 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
579 return ((struct mount *) 0);
580}
581
582/*
583 * Lookup a mount point by filesystem identifier, busying it before
584 * returning.
585 *
586 * To avoid congestion on mountlist_mtx, implement simple direct-mapped
587 * cache for popular filesystem identifiers. The cache is lockess, using
588 * the fact that struct mount's are never freed. In worst case we may
589 * get pointer to unmounted or even different filesystem, so we have to
590 * check what we got, and go slow way if so.
591 */
592struct mount *
593vfs_busyfs(fsid_t *fsid)
594{
595#define FSID_CACHE_SIZE 256
596 typedef struct mount * volatile vmp_t;
597 static vmp_t cache[FSID_CACHE_SIZE];
598 struct mount *mp;
599 int error;
600 uint32_t hash;
601
602 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
603 hash = fsid->val[0] ^ fsid->val[1];
604 hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
605 mp = cache[hash];
606 if (mp == NULL ||
607 mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
608 mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
609 goto slow;
610 if (vfs_busy(mp, 0) != 0) {
611 cache[hash] = NULL;
612 goto slow;
613 }
614 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
615 mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
616 return (mp);
617 else
618 vfs_unbusy(mp);
619
620slow:
621 mtx_lock(&mountlist_mtx);
622 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
623 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
624 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
625 error = vfs_busy(mp, MBF_MNTLSTLOCK);
626 if (error) {
627 cache[hash] = NULL;
628 mtx_unlock(&mountlist_mtx);
629 return (NULL);
630 }
631 cache[hash] = mp;
632 return (mp);
633 }
634 }
635 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
636 mtx_unlock(&mountlist_mtx);
637 return ((struct mount *) 0);
638}
639
640/*
641 * Check if a user can access privileged mount options.
642 */
643int
644vfs_suser(struct mount *mp, struct thread *td)
645{
646 int error;
647
648 /*
649 * If the thread is jailed, but this is not a jail-friendly file
650 * system, deny immediately.
651 */
652 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
653 return (EPERM);
654
655 /*
656 * If the file system was mounted outside the jail of the calling
657 * thread, deny immediately.
658 */
659 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
660 return (EPERM);
661
662 /*
663 * If file system supports delegated administration, we don't check
664 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
665 * by the file system itself.
666 * If this is not the user that did original mount, we check for
667 * the PRIV_VFS_MOUNT_OWNER privilege.
668 */
669 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
670 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
671 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
672 return (error);
673 }
674 return (0);
675}
676
677/*
678 * Get a new unique fsid. Try to make its val[0] unique, since this value
679 * will be used to create fake device numbers for stat(). Also try (but
680 * not so hard) make its val[0] unique mod 2^16, since some emulators only
681 * support 16-bit device numbers. We end up with unique val[0]'s for the
682 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
683 *
684 * Keep in mind that several mounts may be running in parallel. Starting
685 * the search one past where the previous search terminated is both a
686 * micro-optimization and a defense against returning the same fsid to
687 * different mounts.
688 */
689void
690vfs_getnewfsid(struct mount *mp)
691{
692 static uint16_t mntid_base;
693 struct mount *nmp;
694 fsid_t tfsid;
695 int mtype;
696
697 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
698 mtx_lock(&mntid_mtx);
699 mtype = mp->mnt_vfc->vfc_typenum;
700 tfsid.val[1] = mtype;
701 mtype = (mtype & 0xFF) << 24;
702 for (;;) {
703 tfsid.val[0] = makedev(255,
704 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
705 mntid_base++;
706 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
707 break;
708 vfs_rel(nmp);
709 }
710 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
711 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
712 mtx_unlock(&mntid_mtx);
713}
714
715/*
716 * Knob to control the precision of file timestamps:
717 *
718 * 0 = seconds only; nanoseconds zeroed.
719 * 1 = seconds and nanoseconds, accurate within 1/HZ.
720 * 2 = seconds and nanoseconds, truncated to microseconds.
721 * >=3 = seconds and nanoseconds, maximum precision.
722 */
723enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
724
725static int timestamp_precision = TSP_USEC;
726SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
727 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
728 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
729 "3+: sec + ns (max. precision))");
730
731/*
732 * Get a current timestamp.
733 */
734void
735vfs_timestamp(struct timespec *tsp)
736{
737 struct timeval tv;
738
739 switch (timestamp_precision) {
740 case TSP_SEC:
741 tsp->tv_sec = time_second;
742 tsp->tv_nsec = 0;
743 break;
744 case TSP_HZ:
745 getnanotime(tsp);
746 break;
747 case TSP_USEC:
748 microtime(&tv);
749 TIMEVAL_TO_TIMESPEC(&tv, tsp);
750 break;
751 case TSP_NSEC:
752 default:
753 nanotime(tsp);
754 break;
755 }
756}
757
758/*
759 * Set vnode attributes to VNOVAL
760 */
761void
762vattr_null(struct vattr *vap)
763{
764
765 vap->va_type = VNON;
766 vap->va_size = VNOVAL;
767 vap->va_bytes = VNOVAL;
768 vap->va_mode = VNOVAL;
769 vap->va_nlink = VNOVAL;
770 vap->va_uid = VNOVAL;
771 vap->va_gid = VNOVAL;
772 vap->va_fsid = VNOVAL;
773 vap->va_fileid = VNOVAL;
774 vap->va_blocksize = VNOVAL;
775 vap->va_rdev = VNOVAL;
776 vap->va_atime.tv_sec = VNOVAL;
777 vap->va_atime.tv_nsec = VNOVAL;
778 vap->va_mtime.tv_sec = VNOVAL;
779 vap->va_mtime.tv_nsec = VNOVAL;
780 vap->va_ctime.tv_sec = VNOVAL;
781 vap->va_ctime.tv_nsec = VNOVAL;
782 vap->va_birthtime.tv_sec = VNOVAL;
783 vap->va_birthtime.tv_nsec = VNOVAL;
784 vap->va_flags = VNOVAL;
785 vap->va_gen = VNOVAL;
786 vap->va_vaflags = 0;
787}
788
789/*
790 * This routine is called when we have too many vnodes. It attempts
791 * to free <count> vnodes and will potentially free vnodes that still
792 * have VM backing store (VM backing store is typically the cause
793 * of a vnode blowout so we want to do this). Therefore, this operation
794 * is not considered cheap.
795 *
796 * A number of conditions may prevent a vnode from being reclaimed.
797 * the buffer cache may have references on the vnode, a directory
798 * vnode may still have references due to the namei cache representing
799 * underlying files, or the vnode may be in active use. It is not
800 * desireable to reuse such vnodes. These conditions may cause the
801 * number of vnodes to reach some minimum value regardless of what
802 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
803 */
804static int
805vlrureclaim(struct mount *mp)
806{
807 struct vnode *vp;
808 int done;
809 int trigger;
810 int usevnodes;
811 int count;
812
813 /*
814 * Calculate the trigger point, don't allow user
815 * screwups to blow us up. This prevents us from
816 * recycling vnodes with lots of resident pages. We
817 * aren't trying to free memory, we are trying to
818 * free vnodes.
819 */
820 usevnodes = desiredvnodes;
821 if (usevnodes <= 0)
822 usevnodes = 1;
823 trigger = cnt.v_page_count * 2 / usevnodes;
824 done = 0;
825 vn_start_write(NULL, &mp, V_WAIT);
826 MNT_ILOCK(mp);
827 count = mp->mnt_nvnodelistsize / 10 + 1;
828 while (count != 0) {
829 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
830 while (vp != NULL && vp->v_type == VMARKER)
831 vp = TAILQ_NEXT(vp, v_nmntvnodes);
832 if (vp == NULL)
833 break;
834 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
835 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
836 --count;
837 if (!VI_TRYLOCK(vp))
838 goto next_iter;
839 /*
840 * If it's been deconstructed already, it's still
841 * referenced, or it exceeds the trigger, skip it.
842 */
843 if (vp->v_usecount ||
844 (!vlru_allow_cache_src &&
845 !LIST_EMPTY(&(vp)->v_cache_src)) ||
846 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
847 vp->v_object->resident_page_count > trigger)) {
848 VI_UNLOCK(vp);
849 goto next_iter;
850 }
851 MNT_IUNLOCK(mp);
852 vholdl(vp);
853 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
854 vdrop(vp);
855 goto next_iter_mntunlocked;
856 }
857 VI_LOCK(vp);
858 /*
859 * v_usecount may have been bumped after VOP_LOCK() dropped
860 * the vnode interlock and before it was locked again.
861 *
862 * It is not necessary to recheck VI_DOOMED because it can
863 * only be set by another thread that holds both the vnode
864 * lock and vnode interlock. If another thread has the
865 * vnode lock before we get to VOP_LOCK() and obtains the
866 * vnode interlock after VOP_LOCK() drops the vnode
867 * interlock, the other thread will be unable to drop the
868 * vnode lock before our VOP_LOCK() call fails.
869 */
870 if (vp->v_usecount ||
871 (!vlru_allow_cache_src &&
872 !LIST_EMPTY(&(vp)->v_cache_src)) ||
873 (vp->v_object != NULL &&
874 vp->v_object->resident_page_count > trigger)) {
875 VOP_UNLOCK(vp, LK_INTERLOCK);
876 vdrop(vp);
877 goto next_iter_mntunlocked;
878 }
879 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
880 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
881 atomic_add_long(&recycles_count, 1);
882 vgonel(vp);
883 VOP_UNLOCK(vp, 0);
884 vdropl(vp);
885 done++;
886next_iter_mntunlocked:
887 if (!should_yield())
888 goto relock_mnt;
889 goto yield;
890next_iter:
891 if (!should_yield())
892 continue;
893 MNT_IUNLOCK(mp);
894yield:
895 kern_yield(PRI_USER);
896relock_mnt:
897 MNT_ILOCK(mp);
898 }
899 MNT_IUNLOCK(mp);
900 vn_finished_write(mp);
901 return done;
902}
903
904/*
905 * Attempt to keep the free list at wantfreevnodes length.
906 */
907static void
908vnlru_free(int count)
909{
910 struct vnode *vp;
911
912 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
913 for (; count > 0; count--) {
914 vp = TAILQ_FIRST(&vnode_free_list);
915 /*
916 * The list can be modified while the free_list_mtx
917 * has been dropped and vp could be NULL here.
918 */
919 if (!vp)
920 break;
921 VNASSERT(vp->v_op != NULL, vp,
922 ("vnlru_free: vnode already reclaimed."));
923 KASSERT((vp->v_iflag & VI_FREE) != 0,
924 ("Removing vnode not on freelist"));
925 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
926 ("Mangling active vnode"));
927 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
928 /*
929 * Don't recycle if we can't get the interlock.
930 */
931 if (!VI_TRYLOCK(vp)) {
932 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
933 continue;
934 }
935 VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
936 vp, ("vp inconsistent on freelist"));
937
938 /*
939 * The clear of VI_FREE prevents activation of the
940 * vnode. There is no sense in putting the vnode on
941 * the mount point active list, only to remove it
942 * later during recycling. Inline the relevant part
943 * of vholdl(), to avoid triggering assertions or
944 * activating.
945 */
946 freevnodes--;
947 vp->v_iflag &= ~VI_FREE;
948 vp->v_holdcnt++;
949
950 mtx_unlock(&vnode_free_list_mtx);
951 VI_UNLOCK(vp);
952 vtryrecycle(vp);
953 /*
954 * If the recycled succeeded this vdrop will actually free
955 * the vnode. If not it will simply place it back on
956 * the free list.
957 */
958 vdrop(vp);
959 mtx_lock(&vnode_free_list_mtx);
960 }
961}
962/*
963 * Attempt to recycle vnodes in a context that is always safe to block.
964 * Calling vlrurecycle() from the bowels of filesystem code has some
965 * interesting deadlock problems.
966 */
967static struct proc *vnlruproc;
968static int vnlruproc_sig;
969
970static void
971vnlru_proc(void)
972{
973 struct mount *mp, *nmp;
974 int done;
975 struct proc *p = vnlruproc;
976
977 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
978 SHUTDOWN_PRI_FIRST);
979
980 for (;;) {
981 kproc_suspend_check(p);
982 mtx_lock(&vnode_free_list_mtx);
983 if (freevnodes > wantfreevnodes)
984 vnlru_free(freevnodes - wantfreevnodes);
985 if (numvnodes <= desiredvnodes * 9 / 10) {
986 vnlruproc_sig = 0;
987 wakeup(&vnlruproc_sig);
988 msleep(vnlruproc, &vnode_free_list_mtx,
989 PVFS|PDROP, "vlruwt", hz);
990 continue;
991 }
992 mtx_unlock(&vnode_free_list_mtx);
993 done = 0;
994 mtx_lock(&mountlist_mtx);
995 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
996 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
997 nmp = TAILQ_NEXT(mp, mnt_list);
998 continue;
999 }
1000 done += vlrureclaim(mp);
1001 mtx_lock(&mountlist_mtx);
1002 nmp = TAILQ_NEXT(mp, mnt_list);
1003 vfs_unbusy(mp);
1004 }
1005 mtx_unlock(&mountlist_mtx);
1006 if (done == 0) {
1007#if 0
1008 /* These messages are temporary debugging aids */
1009 if (vnlru_nowhere < 5)
1010 printf("vnlru process getting nowhere..\n");
1011 else if (vnlru_nowhere == 5)
1012 printf("vnlru process messages stopped.\n");
1013#endif
1014 vnlru_nowhere++;
1015 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
1016 } else
1017 kern_yield(PRI_USER);
1018 }
1019}
1020
1021static struct kproc_desc vnlru_kp = {
1022 "vnlru",
1023 vnlru_proc,
1024 &vnlruproc
1025};
1026SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
1027 &vnlru_kp);
1028
1029/*
1030 * Routines having to do with the management of the vnode table.
1031 */
1032
1033/*
1034 * Try to recycle a freed vnode. We abort if anyone picks up a reference
1035 * before we actually vgone(). This function must be called with the vnode
1036 * held to prevent the vnode from being returned to the free list midway
1037 * through vgone().
1038 */
1039static int
1040vtryrecycle(struct vnode *vp)
1041{
1042 struct mount *vnmp;
1043
1044 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
1045 VNASSERT(vp->v_holdcnt, vp,
1046 ("vtryrecycle: Recycling vp %p without a reference.", vp));
1047 /*
1048 * This vnode may found and locked via some other list, if so we
1049 * can't recycle it yet.
1050 */
1051 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1052 CTR2(KTR_VFS,
1053 "%s: impossible to recycle, vp %p lock is already held",
1054 __func__, vp);
1055 return (EWOULDBLOCK);
1056 }
1057 /*
1058 * Don't recycle if its filesystem is being suspended.
1059 */
1060 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
1061 VOP_UNLOCK(vp, 0);
1062 CTR2(KTR_VFS,
1063 "%s: impossible to recycle, cannot start the write for %p",
1064 __func__, vp);
1065 return (EBUSY);
1066 }
1067 /*
1068 * If we got this far, we need to acquire the interlock and see if
1069 * anyone picked up this vnode from another list. If not, we will
1070 * mark it with DOOMED via vgonel() so that anyone who does find it
1071 * will skip over it.
1072 */
1073 VI_LOCK(vp);
1074 if (vp->v_usecount) {
1075 VOP_UNLOCK(vp, LK_INTERLOCK);
1076 vn_finished_write(vnmp);
1077 CTR2(KTR_VFS,
1078 "%s: impossible to recycle, %p is already referenced",
1079 __func__, vp);
1080 return (EBUSY);
1081 }
1082 if ((vp->v_iflag & VI_DOOMED) == 0) {
1083 atomic_add_long(&recycles_count, 1);
1084 vgonel(vp);
1085 }
1086 VOP_UNLOCK(vp, LK_INTERLOCK);
1087 vn_finished_write(vnmp);
1088 return (0);
1089}
1090
1091/*
1092 * Wait for available vnodes.
1093 */
1094static int
1095getnewvnode_wait(int suspended)
1096{
1097
1098 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
1099 if (numvnodes > desiredvnodes) {
1100 if (suspended) {
1101 /*
1102 * File system is beeing suspended, we cannot risk a
1103 * deadlock here, so allocate new vnode anyway.
1104 */
1105 if (freevnodes > wantfreevnodes)
1106 vnlru_free(freevnodes - wantfreevnodes);
1107 return (0);
1108 }
1109 if (vnlruproc_sig == 0) {
1110 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
1111 wakeup(vnlruproc);
1112 }
1113 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1114 "vlruwk", hz);
1115 }
1116 return (numvnodes > desiredvnodes ? ENFILE : 0);
1117}
1118
1119void
1120getnewvnode_reserve(u_int count)
1121{
1122 struct thread *td;
1123
1124 td = curthread;
1125 /* First try to be quick and racy. */
1126 if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1127 td->td_vp_reserv += count;
1128 return;
1129 } else
1130 atomic_subtract_long(&numvnodes, count);
1131
1132 mtx_lock(&vnode_free_list_mtx);
1133 while (count > 0) {
1134 if (getnewvnode_wait(0) == 0) {
1135 count--;
1136 td->td_vp_reserv++;
1137 atomic_add_long(&numvnodes, 1);
1138 }
1139 }
1140 mtx_unlock(&vnode_free_list_mtx);
1141}
1142
1143void
1144getnewvnode_drop_reserve(void)
1145{
1146 struct thread *td;
1147
1148 td = curthread;
1149 atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1150 td->td_vp_reserv = 0;
1151}
1152
1153/*
1154 * Return the next vnode from the free list.
1155 */
1156int
1157getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1158 struct vnode **vpp)
1159{
1160 struct vnode *vp;
1161 struct thread *td;
1162 struct lock_object *lo;
1163 int error;
1164
1165 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1166 vp = NULL;
1167 td = curthread;
1168 if (td->td_vp_reserv > 0) {
1169 td->td_vp_reserv -= 1;
1170 goto alloc;
1171 }
1172 mtx_lock(&vnode_free_list_mtx);
1173 /*
1174 * Lend our context to reclaim vnodes if they've exceeded the max.
1175 */
1176 if (freevnodes > wantfreevnodes)
1177 vnlru_free(1);
1178 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1179 MNTK_SUSPEND));
1180#if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1181 if (error != 0) {
1182 mtx_unlock(&vnode_free_list_mtx);
1183 return (error);
1184 }
1185#endif
1186 atomic_add_long(&numvnodes, 1);
1187 mtx_unlock(&vnode_free_list_mtx);
1188alloc:
1189 atomic_add_long(&vnodes_created, 1);
1190 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK);
1191 /*
1192 * Locks are given the generic name "vnode" when created.
1193 * Follow the historic practice of using the filesystem
1194 * name when they allocated, e.g., "zfs", "ufs", "nfs, etc.
1195 *
1196 * Locks live in a witness group keyed on their name. Thus,
1197 * when a lock is renamed, it must also move from the witness
1198 * group of its old name to the witness group of its new name.
1199 *
1200 * The change only needs to be made when the vnode moves
1201 * from one filesystem type to another. We ensure that each
1202 * filesystem use a single static name pointer for its tag so
1203 * that we can compare pointers rather than doing a strcmp().
1204 */
1205 lo = &vp->v_vnlock->lock_object;
1206 if (lo->lo_name != tag) {
1207 lo->lo_name = tag;
1208 WITNESS_DESTROY(lo);
1209 WITNESS_INIT(lo, tag);
1210 }
1211 /*
1212 * By default, don't allow shared locks unless filesystems opt-in.
1213 */
1214 vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE;
1215 /*
1216 * Finalize various vnode identity bits.
1217 */
1218 KASSERT(vp->v_object == NULL, ("stale v_object %p", vp));
1219 KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp));
1220 KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp));
1221 vp->v_type = VNON;
1222 vp->v_tag = tag;
1223 vp->v_op = vops;
1224 v_incr_usecount(vp);
1225 vp->v_bufobj.bo_ops = &buf_ops_bio;
1226#ifdef MAC
1227 mac_vnode_init(vp);
1228 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1229 mac_vnode_associate_singlelabel(mp, vp);
1230 else if (mp == NULL && vops != &dead_vnodeops)
1231 printf("NULL mp in getnewvnode()\n");
1232#endif
1233 if (mp != NULL) {
1234 vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize;
1235 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1236 vp->v_vflag |= VV_NOKNOTE;
1237 }
1238
1239 /*
1240 * For the filesystems which do not use vfs_hash_insert(),
1241 * still initialize v_hash to have vfs_hash_index() useful.
1242 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1243 * its own hashing.
1244 */
1245 vp->v_hash = (uintptr_t)vp >> vnsz2log;
1246
1247 *vpp = vp;
1248 return (0);
1249}
1250
1251/*
1252 * Delete from old mount point vnode list, if on one.
1253 */
1254static void
1255delmntque(struct vnode *vp)
1256{
1257 struct mount *mp;
1258 int active;
1259
1260 mp = vp->v_mount;
1261 if (mp == NULL)
1262 return;
1263 MNT_ILOCK(mp);
1264 VI_LOCK(vp);
1265 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1266 ("Active vnode list size %d > Vnode list size %d",
1267 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1268 active = vp->v_iflag & VI_ACTIVE;
1269 vp->v_iflag &= ~VI_ACTIVE;
1270 if (active) {
1271 mtx_lock(&vnode_free_list_mtx);
1272 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1273 mp->mnt_activevnodelistsize--;
1274 mtx_unlock(&vnode_free_list_mtx);
1275 }
1276 vp->v_mount = NULL;
1277 VI_UNLOCK(vp);
1278 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1279 ("bad mount point vnode list size"));
1280 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1281 mp->mnt_nvnodelistsize--;
1282 MNT_REL(mp);
1283 MNT_IUNLOCK(mp);
1284}
1285
1286static void
1287insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1288{
1289
1290 vp->v_data = NULL;
1291 vp->v_op = &dead_vnodeops;
1292 vgone(vp);
1293 vput(vp);
1294}
1295
1296/*
1297 * Insert into list of vnodes for the new mount point, if available.
1298 */
1299int
1300insmntque1(struct vnode *vp, struct mount *mp,
1301 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1302{
1303
1304 KASSERT(vp->v_mount == NULL,
1305 ("insmntque: vnode already on per mount vnode list"));
1306 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1307 ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1308
1309 /*
1310 * We acquire the vnode interlock early to ensure that the
1311 * vnode cannot be recycled by another process releasing a
1312 * holdcnt on it before we get it on both the vnode list
1313 * and the active vnode list. The mount mutex protects only
1314 * manipulation of the vnode list and the vnode freelist
1315 * mutex protects only manipulation of the active vnode list.
1316 * Hence the need to hold the vnode interlock throughout.
1317 */
1318 MNT_ILOCK(mp);
1319 VI_LOCK(vp);
1320 if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1321 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1322 mp->mnt_nvnodelistsize == 0)) &&
1323 (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1324 VI_UNLOCK(vp);
1325 MNT_IUNLOCK(mp);
1326 if (dtr != NULL)
1327 dtr(vp, dtr_arg);
1328 return (EBUSY);
1329 }
1330 vp->v_mount = mp;
1331 MNT_REF(mp);
1332 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1333 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1334 ("neg mount point vnode list size"));
1335 mp->mnt_nvnodelistsize++;
1336 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1337 ("Activating already active vnode"));
1338 vp->v_iflag |= VI_ACTIVE;
1339 mtx_lock(&vnode_free_list_mtx);
1340 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1341 mp->mnt_activevnodelistsize++;
1342 mtx_unlock(&vnode_free_list_mtx);
1343 VI_UNLOCK(vp);
1344 MNT_IUNLOCK(mp);
1345 return (0);
1346}
1347
1348int
1349insmntque(struct vnode *vp, struct mount *mp)
1350{
1351
1352 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1353}
1354
1355/*
1356 * Flush out and invalidate all buffers associated with a bufobj
1357 * Called with the underlying object locked.
1358 */
1359int
1360bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1361{
1362 int error;
1363
1364 BO_LOCK(bo);
1365 if (flags & V_SAVE) {
1366 error = bufobj_wwait(bo, slpflag, slptimeo);
1367 if (error) {
1368 BO_UNLOCK(bo);
1369 return (error);
1370 }
1371 if (bo->bo_dirty.bv_cnt > 0) {
1372 BO_UNLOCK(bo);
1373 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1374 return (error);
1375 /*
1376 * XXX We could save a lock/unlock if this was only
1377 * enabled under INVARIANTS
1378 */
1379 BO_LOCK(bo);
1380 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1381 panic("vinvalbuf: dirty bufs");
1382 }
1383 }
1384 /*
1385 * If you alter this loop please notice that interlock is dropped and
1386 * reacquired in flushbuflist. Special care is needed to ensure that
1387 * no race conditions occur from this.
1388 */
1389 do {
1390 error = flushbuflist(&bo->bo_clean,
1391 flags, bo, slpflag, slptimeo);
1392 if (error == 0 && !(flags & V_CLEANONLY))
1393 error = flushbuflist(&bo->bo_dirty,
1394 flags, bo, slpflag, slptimeo);
1395 if (error != 0 && error != EAGAIN) {
1396 BO_UNLOCK(bo);
1397 return (error);
1398 }
1399 } while (error != 0);
1400
1401 /*
1402 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1403 * have write I/O in-progress but if there is a VM object then the
1404 * VM object can also have read-I/O in-progress.
1405 */
1406 do {
1407 bufobj_wwait(bo, 0, 0);
1408 BO_UNLOCK(bo);
1409 if (bo->bo_object != NULL) {
1410 VM_OBJECT_WLOCK(bo->bo_object);
1411 vm_object_pip_wait(bo->bo_object, "bovlbx");
1412 VM_OBJECT_WUNLOCK(bo->bo_object);
1413 }
1414 BO_LOCK(bo);
1415 } while (bo->bo_numoutput > 0);
1416 BO_UNLOCK(bo);
1417
1418 /*
1419 * Destroy the copy in the VM cache, too.
1420 */
1421 if (bo->bo_object != NULL &&
1422 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1423 VM_OBJECT_WLOCK(bo->bo_object);
1424 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1425 OBJPR_CLEANONLY : 0);
1426 VM_OBJECT_WUNLOCK(bo->bo_object);
1427 }
1428
1429#ifdef INVARIANTS
1430 BO_LOCK(bo);
1431 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1432 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1433 panic("vinvalbuf: flush failed");
1434 BO_UNLOCK(bo);
1435#endif
1436 return (0);
1437}
1438
1439/*
1440 * Flush out and invalidate all buffers associated with a vnode.
1441 * Called with the underlying object locked.
1442 */
1443int
1444vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1445{
1446
1447 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1448 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1449 if (vp->v_object != NULL && vp->v_object->handle != vp)
1450 return (0);
1451 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1452}
1453
1454/*
1455 * Flush out buffers on the specified list.
1456 *
1457 */
1458static int
1459flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1460 int slptimeo)
1461{
1462 struct buf *bp, *nbp;
1463 int retval, error;
1464 daddr_t lblkno;
1465 b_xflags_t xflags;
1466
1467 ASSERT_BO_WLOCKED(bo);
1468
1469 retval = 0;
1470 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1471 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1472 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1473 continue;
1474 }
1475 lblkno = 0;
1476 xflags = 0;
1477 if (nbp != NULL) {
1478 lblkno = nbp->b_lblkno;
1479 xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1480 }
1481 retval = EAGAIN;
1482 error = BUF_TIMELOCK(bp,
1483 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1484 "flushbuf", slpflag, slptimeo);
1485 if (error) {
1486 BO_LOCK(bo);
1487 return (error != ENOLCK ? error : EAGAIN);
1488 }
1489 KASSERT(bp->b_bufobj == bo,
1490 ("bp %p wrong b_bufobj %p should be %p",
1491 bp, bp->b_bufobj, bo));
1492 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1493 BUF_UNLOCK(bp);
1494 BO_LOCK(bo);
1495 return (EAGAIN);
1496 }
1497 /*
1498 * XXX Since there are no node locks for NFS, I
1499 * believe there is a slight chance that a delayed
1500 * write will occur while sleeping just above, so
1501 * check for it.
1502 */
1503 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1504 (flags & V_SAVE)) {
1505 bremfree(bp);
1506 bp->b_flags |= B_ASYNC;
1507 bwrite(bp);
1508 BO_LOCK(bo);
1509 return (EAGAIN); /* XXX: why not loop ? */
1510 }
1511 bremfree(bp);
1512 bp->b_flags |= (B_INVAL | B_RELBUF);
1513 bp->b_flags &= ~B_ASYNC;
1514 brelse(bp);
1515 BO_LOCK(bo);
1516 if (nbp != NULL &&
1517 (nbp->b_bufobj != bo ||
1518 nbp->b_lblkno != lblkno ||
1519 (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1520 break; /* nbp invalid */
1521 }
1522 return (retval);
1523}
1524
1525/*
1526 * Truncate a file's buffer and pages to a specified length. This
1527 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1528 * sync activity.
1529 */
1530int
1531vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1532{
1533 struct buf *bp, *nbp;
1534 int anyfreed;
1535 int trunclbn;
1536 struct bufobj *bo;
1537
1538 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1539 vp, cred, blksize, (uintmax_t)length);
1540
1541 /*
1542 * Round up to the *next* lbn.
1543 */
1544 trunclbn = (length + blksize - 1) / blksize;
1545
1546 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1547restart:
1548 bo = &vp->v_bufobj;
1549 BO_LOCK(bo);
1550 anyfreed = 1;
1551 for (;anyfreed;) {
1552 anyfreed = 0;
1553 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1554 if (bp->b_lblkno < trunclbn)
1555 continue;
1556 if (BUF_LOCK(bp,
1557 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1558 BO_LOCKPTR(bo)) == ENOLCK)
1559 goto restart;
1560
1561 bremfree(bp);
1562 bp->b_flags |= (B_INVAL | B_RELBUF);
1563 bp->b_flags &= ~B_ASYNC;
1564 brelse(bp);
1565 anyfreed = 1;
1566
1567 BO_LOCK(bo);
1568 if (nbp != NULL &&
1569 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1570 (nbp->b_vp != vp) ||
1571 (nbp->b_flags & B_DELWRI))) {
1572 BO_UNLOCK(bo);
1573 goto restart;
1574 }
1575 }
1576
1577 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1578 if (bp->b_lblkno < trunclbn)
1579 continue;
1580 if (BUF_LOCK(bp,
1581 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1582 BO_LOCKPTR(bo)) == ENOLCK)
1583 goto restart;
1584 bremfree(bp);
1585 bp->b_flags |= (B_INVAL | B_RELBUF);
1586 bp->b_flags &= ~B_ASYNC;
1587 brelse(bp);
1588 anyfreed = 1;
1589
1590 BO_LOCK(bo);
1591 if (nbp != NULL &&
1592 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1593 (nbp->b_vp != vp) ||
1594 (nbp->b_flags & B_DELWRI) == 0)) {
1595 BO_UNLOCK(bo);
1596 goto restart;
1597 }
1598 }
1599 }
1600
1601 if (length > 0) {
1602restartsync:
1603 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1604 if (bp->b_lblkno > 0)
1605 continue;
1606 /*
1607 * Since we hold the vnode lock this should only
1608 * fail if we're racing with the buf daemon.
1609 */
1610 if (BUF_LOCK(bp,
1611 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1612 BO_LOCKPTR(bo)) == ENOLCK) {
1613 goto restart;
1614 }
1615 VNASSERT((bp->b_flags & B_DELWRI), vp,
1616 ("buf(%p) on dirty queue without DELWRI", bp));
1617
1618 bremfree(bp);
1619 bawrite(bp);
1620 BO_LOCK(bo);
1621 goto restartsync;
1622 }
1623 }
1624
1625 bufobj_wwait(bo, 0, 0);
1626 BO_UNLOCK(bo);
1627 vnode_pager_setsize(vp, length);
1628
1629 return (0);
1630}
1631
1632static void
1633buf_vlist_remove(struct buf *bp)
1634{
1635 struct bufv *bv;
1636
1637 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1638 ASSERT_BO_WLOCKED(bp->b_bufobj);
1639 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1640 (BX_VNDIRTY|BX_VNCLEAN),
1641 ("buf_vlist_remove: Buf %p is on two lists", bp));
1642 if (bp->b_xflags & BX_VNDIRTY)
1643 bv = &bp->b_bufobj->bo_dirty;
1644 else
1645 bv = &bp->b_bufobj->bo_clean;
1646 BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1647 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1648 bv->bv_cnt--;
1649 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1650}
1651
1652/*
1653 * Add the buffer to the sorted clean or dirty block list.
1654 *
1655 * NOTE: xflags is passed as a constant, optimizing this inline function!
1656 */
1657static void
1658buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1659{
1660 struct bufv *bv;
1661 struct buf *n;
1662 int error;
1663
1664 ASSERT_BO_WLOCKED(bo);
1665 KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0,
1666 ("dead bo %p", bo));
1667 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1668 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1669 bp->b_xflags |= xflags;
1670 if (xflags & BX_VNDIRTY)
1671 bv = &bo->bo_dirty;
1672 else
1673 bv = &bo->bo_clean;
1674
1675 /*
1676 * Keep the list ordered. Optimize empty list insertion. Assume
1677 * we tend to grow at the tail so lookup_le should usually be cheaper
1678 * than _ge.
1679 */
1680 if (bv->bv_cnt == 0 ||
1681 bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1682 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1683 else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1684 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1685 else
1686 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1687 error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1688 if (error)
1689 panic("buf_vlist_add: Preallocated nodes insufficient.");
1690 bv->bv_cnt++;
1691}
1692
1693/*
1694 * Lookup a buffer using the splay tree. Note that we specifically avoid
1695 * shadow buffers used in background bitmap writes.
1696 *
1697 * This code isn't quite efficient as it could be because we are maintaining
1698 * two sorted lists and do not know which list the block resides in.
1699 *
1700 * During a "make buildworld" the desired buffer is found at one of
1701 * the roots more than 60% of the time. Thus, checking both roots
1702 * before performing either splay eliminates unnecessary splays on the
1703 * first tree splayed.
1704 */
1705struct buf *
1706gbincore(struct bufobj *bo, daddr_t lblkno)
1707{
1708 struct buf *bp;
1709
1710 ASSERT_BO_LOCKED(bo);
1711 bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1712 if (bp != NULL)
1713 return (bp);
1714 return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1715}
1716
1717/*
1718 * Associate a buffer with a vnode.
1719 */
1720void
1721bgetvp(struct vnode *vp, struct buf *bp)
1722{
1723 struct bufobj *bo;
1724
1725 bo = &vp->v_bufobj;
1726 ASSERT_BO_WLOCKED(bo);
1727 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1728
1729 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1730 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1731 ("bgetvp: bp already attached! %p", bp));
1732
1733 vhold(vp);
1734 bp->b_vp = vp;
1735 bp->b_bufobj = bo;
1736 /*
1737 * Insert onto list for new vnode.
1738 */
1739 buf_vlist_add(bp, bo, BX_VNCLEAN);
1740}
1741
1742/*
1743 * Disassociate a buffer from a vnode.
1744 */
1745void
1746brelvp(struct buf *bp)
1747{
1748 struct bufobj *bo;
1749 struct vnode *vp;
1750
1751 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1752 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1753
1754 /*
1755 * Delete from old vnode list, if on one.
1756 */
1757 vp = bp->b_vp; /* XXX */
1758 bo = bp->b_bufobj;
1759 BO_LOCK(bo);
1760 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1761 buf_vlist_remove(bp);
1762 else
1763 panic("brelvp: Buffer %p not on queue.", bp);
1764 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1765 bo->bo_flag &= ~BO_ONWORKLST;
1766 mtx_lock(&sync_mtx);
1767 LIST_REMOVE(bo, bo_synclist);
1768 syncer_worklist_len--;
1769 mtx_unlock(&sync_mtx);
1770 }
1771 bp->b_vp = NULL;
1772 bp->b_bufobj = NULL;
1773 BO_UNLOCK(bo);
1774 vdrop(vp);
1775}
1776
1777/*
1778 * Add an item to the syncer work queue.
1779 */
1780static void
1781vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1782{
1783 int slot;
1784
1785 ASSERT_BO_WLOCKED(bo);
1786
1787 mtx_lock(&sync_mtx);
1788 if (bo->bo_flag & BO_ONWORKLST)
1789 LIST_REMOVE(bo, bo_synclist);
1790 else {
1791 bo->bo_flag |= BO_ONWORKLST;
1792 syncer_worklist_len++;
1793 }
1794
1795 if (delay > syncer_maxdelay - 2)
1796 delay = syncer_maxdelay - 2;
1797 slot = (syncer_delayno + delay) & syncer_mask;
1798
1799 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1800 mtx_unlock(&sync_mtx);
1801}
1802
1803static int
1804sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1805{
1806 int error, len;
1807
1808 mtx_lock(&sync_mtx);
1809 len = syncer_worklist_len - sync_vnode_count;
1810 mtx_unlock(&sync_mtx);
1811 error = SYSCTL_OUT(req, &len, sizeof(len));
1812 return (error);
1813}
1814
1815SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1816 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1817
1818static struct proc *updateproc;
1819static void sched_sync(void);
1820static struct kproc_desc up_kp = {
1821 "syncer",
1822 sched_sync,
1823 &updateproc
1824};
1825SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1826
1827static int
1828sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1829{
1830 struct vnode *vp;
1831 struct mount *mp;
1832
1833 *bo = LIST_FIRST(slp);
1834 if (*bo == NULL)
1835 return (0);
1836 vp = (*bo)->__bo_vnode; /* XXX */
1837 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1838 return (1);
1839 /*
1840 * We use vhold in case the vnode does not
1841 * successfully sync. vhold prevents the vnode from
1842 * going away when we unlock the sync_mtx so that
1843 * we can acquire the vnode interlock.
1844 */
1845 vholdl(vp);
1846 mtx_unlock(&sync_mtx);
1847 VI_UNLOCK(vp);
1848 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1849 vdrop(vp);
1850 mtx_lock(&sync_mtx);
1851 return (*bo == LIST_FIRST(slp));
1852 }
1853 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1854 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1855 VOP_UNLOCK(vp, 0);
1856 vn_finished_write(mp);
1857 BO_LOCK(*bo);
1858 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1859 /*
1860 * Put us back on the worklist. The worklist
1861 * routine will remove us from our current
1862 * position and then add us back in at a later
1863 * position.
1864 */
1865 vn_syncer_add_to_worklist(*bo, syncdelay);
1866 }
1867 BO_UNLOCK(*bo);
1868 vdrop(vp);
1869 mtx_lock(&sync_mtx);
1870 return (0);
1871}
1872
1873static int first_printf = 1;
1874
1875/*
1876 * System filesystem synchronizer daemon.
1877 */
1878static void
1879sched_sync(void)
1880{
1881 struct synclist *next, *slp;
1882 struct bufobj *bo;
1883 long starttime;
1884 struct thread *td = curthread;
1885 int last_work_seen;
1886 int net_worklist_len;
1887 int syncer_final_iter;
1888 int error;
1889
1890 last_work_seen = 0;
1891 syncer_final_iter = 0;
1892 syncer_state = SYNCER_RUNNING;
1893 starttime = time_uptime;
1894 td->td_pflags |= TDP_NORUNNINGBUF;
1895
1896 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1897 SHUTDOWN_PRI_LAST);
1898
1899 mtx_lock(&sync_mtx);
1900 for (;;) {
1901 if (syncer_state == SYNCER_FINAL_DELAY &&
1902 syncer_final_iter == 0) {
1903 mtx_unlock(&sync_mtx);
1904 kproc_suspend_check(td->td_proc);
1905 mtx_lock(&sync_mtx);
1906 }
1907 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1908 if (syncer_state != SYNCER_RUNNING &&
1909 starttime != time_uptime) {
1910 if (first_printf) {
1911 printf("\nSyncing disks, vnodes remaining...");
1912 first_printf = 0;
1913 }
1914 printf("%d ", net_worklist_len);
1915 }
1916 starttime = time_uptime;
1917
1918 /*
1919 * Push files whose dirty time has expired. Be careful
1920 * of interrupt race on slp queue.
1921 *
1922 * Skip over empty worklist slots when shutting down.
1923 */
1924 do {
1925 slp = &syncer_workitem_pending[syncer_delayno];
1926 syncer_delayno += 1;
1927 if (syncer_delayno == syncer_maxdelay)
1928 syncer_delayno = 0;
1929 next = &syncer_workitem_pending[syncer_delayno];
1930 /*
1931 * If the worklist has wrapped since the
1932 * it was emptied of all but syncer vnodes,
1933 * switch to the FINAL_DELAY state and run
1934 * for one more second.
1935 */
1936 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1937 net_worklist_len == 0 &&
1938 last_work_seen == syncer_delayno) {
1939 syncer_state = SYNCER_FINAL_DELAY;
1940 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1941 }
1942 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1943 syncer_worklist_len > 0);
1944
1945 /*
1946 * Keep track of the last time there was anything
1947 * on the worklist other than syncer vnodes.
1948 * Return to the SHUTTING_DOWN state if any
1949 * new work appears.
1950 */
1951 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1952 last_work_seen = syncer_delayno;
1953 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1954 syncer_state = SYNCER_SHUTTING_DOWN;
1955 while (!LIST_EMPTY(slp)) {
1956 error = sync_vnode(slp, &bo, td);
1957 if (error == 1) {
1958 LIST_REMOVE(bo, bo_synclist);
1959 LIST_INSERT_HEAD(next, bo, bo_synclist);
1960 continue;
1961 }
1962
1963 if (first_printf == 0)
1964 wdog_kern_pat(WD_LASTVAL);
1965
1966 }
1967 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1968 syncer_final_iter--;
1969 /*
1970 * The variable rushjob allows the kernel to speed up the
1971 * processing of the filesystem syncer process. A rushjob
1972 * value of N tells the filesystem syncer to process the next
1973 * N seconds worth of work on its queue ASAP. Currently rushjob
1974 * is used by the soft update code to speed up the filesystem
1975 * syncer process when the incore state is getting so far
1976 * ahead of the disk that the kernel memory pool is being
1977 * threatened with exhaustion.
1978 */
1979 if (rushjob > 0) {
1980 rushjob -= 1;
1981 continue;
1982 }
1983 /*
1984 * Just sleep for a short period of time between
1985 * iterations when shutting down to allow some I/O
1986 * to happen.
1987 *
1988 * If it has taken us less than a second to process the
1989 * current work, then wait. Otherwise start right over
1990 * again. We can still lose time if any single round
1991 * takes more than two seconds, but it does not really
1992 * matter as we are just trying to generally pace the
1993 * filesystem activity.
1994 */
1995 if (syncer_state != SYNCER_RUNNING ||
1996 time_uptime == starttime) {
1997 thread_lock(td);
1998 sched_prio(td, PPAUSE);
1999 thread_unlock(td);
2000 }
2001 if (syncer_state != SYNCER_RUNNING)
2002 cv_timedwait(&sync_wakeup, &sync_mtx,
2003 hz / SYNCER_SHUTDOWN_SPEEDUP);
2004 else if (time_uptime == starttime)
2005 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
2006 }
2007}
2008
2009/*
2010 * Request the syncer daemon to speed up its work.
2011 * We never push it to speed up more than half of its
2012 * normal turn time, otherwise it could take over the cpu.
2013 */
2014int
2015speedup_syncer(void)
2016{
2017 int ret = 0;
2018
2019 mtx_lock(&sync_mtx);
2020 if (rushjob < syncdelay / 2) {
2021 rushjob += 1;
2022 stat_rush_requests += 1;
2023 ret = 1;
2024 }
2025 mtx_unlock(&sync_mtx);
2026 cv_broadcast(&sync_wakeup);
2027 return (ret);
2028}
2029
2030/*
2031 * Tell the syncer to speed up its work and run though its work
2032 * list several times, then tell it to shut down.
2033 */
2034static void
2035syncer_shutdown(void *arg, int howto)
2036{
2037
2038 if (howto & RB_NOSYNC)
2039 return;
2040 mtx_lock(&sync_mtx);
2041 syncer_state = SYNCER_SHUTTING_DOWN;
2042 rushjob = 0;
2043 mtx_unlock(&sync_mtx);
2044 cv_broadcast(&sync_wakeup);
2045 kproc_shutdown(arg, howto);
2046}
2047
2048void
2049syncer_suspend(void)
2050{
2051
2052 syncer_shutdown(updateproc, 0);
2053}
2054
2055void
2056syncer_resume(void)
2057{
2058
2059 mtx_lock(&sync_mtx);
2060 first_printf = 1;
2061 syncer_state = SYNCER_RUNNING;
2062 mtx_unlock(&sync_mtx);
2063 cv_broadcast(&sync_wakeup);
2064 kproc_resume(updateproc);
2065}
2066
2067/*
2068 * Reassign a buffer from one vnode to another.
2069 * Used to assign file specific control information
2070 * (indirect blocks) to the vnode to which they belong.
2071 */
2072void
2073reassignbuf(struct buf *bp)
2074{
2075 struct vnode *vp;
2076 struct bufobj *bo;
2077 int delay;
2078#ifdef INVARIANTS
2079 struct bufv *bv;
2080#endif
2081
2082 vp = bp->b_vp;
2083 bo = bp->b_bufobj;
2084 ++reassignbufcalls;
2085
2086 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2087 bp, bp->b_vp, bp->b_flags);
2088 /*
2089 * B_PAGING flagged buffers cannot be reassigned because their vp
2090 * is not fully linked in.
2091 */
2092 if (bp->b_flags & B_PAGING)
2093 panic("cannot reassign paging buffer");
2094
2095 /*
2096 * Delete from old vnode list, if on one.
2097 */
2098 BO_LOCK(bo);
2099 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2100 buf_vlist_remove(bp);
2101 else
2102 panic("reassignbuf: Buffer %p not on queue.", bp);
2103 /*
2104 * If dirty, put on list of dirty buffers; otherwise insert onto list
2105 * of clean buffers.
2106 */
2107 if (bp->b_flags & B_DELWRI) {
2108 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2109 switch (vp->v_type) {
2110 case VDIR:
2111 delay = dirdelay;
2112 break;
2113 case VCHR:
2114 delay = metadelay;
2115 break;
2116 default:
2117 delay = filedelay;
2118 }
2119 vn_syncer_add_to_worklist(bo, delay);
2120 }
2121 buf_vlist_add(bp, bo, BX_VNDIRTY);
2122 } else {
2123 buf_vlist_add(bp, bo, BX_VNCLEAN);
2124
2125 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2126 mtx_lock(&sync_mtx);
2127 LIST_REMOVE(bo, bo_synclist);
2128 syncer_worklist_len--;
2129 mtx_unlock(&sync_mtx);
2130 bo->bo_flag &= ~BO_ONWORKLST;
2131 }
2132 }
2133#ifdef INVARIANTS
2134 bv = &bo->bo_clean;
2135 bp = TAILQ_FIRST(&bv->bv_hd);
2136 KASSERT(bp == NULL || bp->b_bufobj == bo,
2137 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2138 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2139 KASSERT(bp == NULL || bp->b_bufobj == bo,
2140 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2141 bv = &bo->bo_dirty;
2142 bp = TAILQ_FIRST(&bv->bv_hd);
2143 KASSERT(bp == NULL || bp->b_bufobj == bo,
2144 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2145 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2146 KASSERT(bp == NULL || bp->b_bufobj == bo,
2147 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2148#endif
2149 BO_UNLOCK(bo);
2150}
2151
2152/*
2153 * Increment the use and hold counts on the vnode, taking care to reference
2154 * the driver's usecount if this is a chardev. The vholdl() will remove
2155 * the vnode from the free list if it is presently free. Requires the
2156 * vnode interlock and returns with it held.
2157 */
2158static void
2159v_incr_usecount(struct vnode *vp)
2160{
2161
2162 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2163 vholdl(vp);
2164 vp->v_usecount++;
2165 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2166 dev_lock();
2167 vp->v_rdev->si_usecount++;
2168 dev_unlock();
2169 }
2170}
2171
2172/*
2173 * Turn a holdcnt into a use+holdcnt such that only one call to
2174 * v_decr_usecount is needed.
2175 */
2176static void
2177v_upgrade_usecount(struct vnode *vp)
2178{
2179
2180 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2181 vp->v_usecount++;
2182 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2183 dev_lock();
2184 vp->v_rdev->si_usecount++;
2185 dev_unlock();
2186 }
2187}
2188
2189/*
2190 * Decrement the vnode use and hold count along with the driver's usecount
2191 * if this is a chardev. The vdropl() below releases the vnode interlock
2192 * as it may free the vnode.
2193 */
2194static void
2195v_decr_usecount(struct vnode *vp)
2196{
2197
2198 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2199 VNASSERT(vp->v_usecount > 0, vp,
2200 ("v_decr_usecount: negative usecount"));
2201 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2202 vp->v_usecount--;
2203 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2204 dev_lock();
2205 vp->v_rdev->si_usecount--;
2206 dev_unlock();
2207 }
2208 vdropl(vp);
2209}
2210
2211/*
2212 * Decrement only the use count and driver use count. This is intended to
2213 * be paired with a follow on vdropl() to release the remaining hold count.
2214 * In this way we may vgone() a vnode with a 0 usecount without risk of
2215 * having it end up on a free list because the hold count is kept above 0.
2216 */
2217static void
2218v_decr_useonly(struct vnode *vp)
2219{
2220
2221 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2222 VNASSERT(vp->v_usecount > 0, vp,
2223 ("v_decr_useonly: negative usecount"));
2224 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2225 vp->v_usecount--;
2226 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2227 dev_lock();
2228 vp->v_rdev->si_usecount--;
2229 dev_unlock();
2230 }
2231}
2232
2233/*
2234 * Grab a particular vnode from the free list, increment its
2235 * reference count and lock it. VI_DOOMED is set if the vnode
2236 * is being destroyed. Only callers who specify LK_RETRY will
2237 * see doomed vnodes. If inactive processing was delayed in
2238 * vput try to do it here.
2239 */
2240int
2241vget(struct vnode *vp, int flags, struct thread *td)
2242{
2243 int error;
2244
2245 error = 0;
2246 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2247 ("vget: invalid lock operation"));
2248 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2249
2250 if ((flags & LK_INTERLOCK) == 0)
2251 VI_LOCK(vp);
2252 vholdl(vp);
2253 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2254 vdrop(vp);
2255 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2256 vp);
2257 return (error);
2258 }
2259 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2260 panic("vget: vn_lock failed to return ENOENT\n");
2261 VI_LOCK(vp);
2262 /* Upgrade our holdcnt to a usecount. */
2263 v_upgrade_usecount(vp);
2264 /*
2265 * We don't guarantee that any particular close will
2266 * trigger inactive processing so just make a best effort
2267 * here at preventing a reference to a removed file. If
2268 * we don't succeed no harm is done.
2269 */
2270 if (vp->v_iflag & VI_OWEINACT) {
2271 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2272 (flags & LK_NOWAIT) == 0)
2273 vinactive(vp, td);
2274 vp->v_iflag &= ~VI_OWEINACT;
2275 }
2276 VI_UNLOCK(vp);
2277 return (0);
2278}
2279
2280/*
2281 * Increase the reference count of a vnode.
2282 */
2283void
2284vref(struct vnode *vp)
2285{
2286
2287 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2288 VI_LOCK(vp);
2289 v_incr_usecount(vp);
2290 VI_UNLOCK(vp);
2291}
2292
2293/*
2294 * Return reference count of a vnode.
2295 *
2296 * The results of this call are only guaranteed when some mechanism other
2297 * than the VI lock is used to stop other processes from gaining references
2298 * to the vnode. This may be the case if the caller holds the only reference.
2299 * This is also useful when stale data is acceptable as race conditions may
2300 * be accounted for by some other means.
2301 */
2302int
2303vrefcnt(struct vnode *vp)
2304{
2305 int usecnt;
2306
2307 VI_LOCK(vp);
2308 usecnt = vp->v_usecount;
2309 VI_UNLOCK(vp);
2310
2311 return (usecnt);
2312}
2313
2314#define VPUTX_VRELE 1
2315#define VPUTX_VPUT 2
2316#define VPUTX_VUNREF 3
2317
2318static void
2319vputx(struct vnode *vp, int func)
2320{
2321 int error;
2322
2323 KASSERT(vp != NULL, ("vputx: null vp"));
2324 if (func == VPUTX_VUNREF)
2325 ASSERT_VOP_LOCKED(vp, "vunref");
2326 else if (func == VPUTX_VPUT)
2327 ASSERT_VOP_LOCKED(vp, "vput");
2328 else
2329 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2330 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2331 VI_LOCK(vp);
2332
2333 /* Skip this v_writecount check if we're going to panic below. */
2334 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2335 ("vputx: missed vn_close"));
2336 error = 0;
2337
2338 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2339 vp->v_usecount == 1)) {
2340 if (func == VPUTX_VPUT)
2341 VOP_UNLOCK(vp, 0);
2342 v_decr_usecount(vp);
2343 return;
2344 }
2345
2346 if (vp->v_usecount != 1) {
2347 vprint("vputx: negative ref count", vp);
2348 panic("vputx: negative ref cnt");
2349 }
2350 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2351 /*
2352 * We want to hold the vnode until the inactive finishes to
2353 * prevent vgone() races. We drop the use count here and the
2354 * hold count below when we're done.
2355 */
2356 v_decr_useonly(vp);
2357 /*
2358 * We must call VOP_INACTIVE with the node locked. Mark
2359 * as VI_DOINGINACT to avoid recursion.
2360 */
2361 vp->v_iflag |= VI_OWEINACT;
2362 switch (func) {
2363 case VPUTX_VRELE:
2364 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2365 VI_LOCK(vp);
2366 break;
2367 case VPUTX_VPUT:
2368 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2369 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2370 LK_NOWAIT);
2371 VI_LOCK(vp);
2372 }
2373 break;
2374 case VPUTX_VUNREF:
2375 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2376 error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2377 VI_LOCK(vp);
2378 }
2379 break;
2380 }
2381 if (vp->v_usecount > 0)
2382 vp->v_iflag &= ~VI_OWEINACT;
2383 if (error == 0) {
2384 if (vp->v_iflag & VI_OWEINACT)
2385 vinactive(vp, curthread);
2386 if (func != VPUTX_VUNREF)
2387 VOP_UNLOCK(vp, 0);
2388 }
2389 vdropl(vp);
2390}
2391
2392/*
2393 * Vnode put/release.
2394 * If count drops to zero, call inactive routine and return to freelist.
2395 */
2396void
2397vrele(struct vnode *vp)
2398{
2399
2400 vputx(vp, VPUTX_VRELE);
2401}
2402
2403/*
2404 * Release an already locked vnode. This give the same effects as
2405 * unlock+vrele(), but takes less time and avoids releasing and
2406 * re-aquiring the lock (as vrele() acquires the lock internally.)
2407 */
2408void
2409vput(struct vnode *vp)
2410{
2411
2412 vputx(vp, VPUTX_VPUT);
2413}
2414
2415/*
2416 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2417 */
2418void
2419vunref(struct vnode *vp)
2420{
2421
2422 vputx(vp, VPUTX_VUNREF);
2423}
2424
2425/*
2426 * Somebody doesn't want the vnode recycled.
2427 */
2428void
2429vhold(struct vnode *vp)
2430{
2431
2432 VI_LOCK(vp);
2433 vholdl(vp);
2434 VI_UNLOCK(vp);
2435}
2436
2437/*
2438 * Increase the hold count and activate if this is the first reference.
2439 */
2440void
2441vholdl(struct vnode *vp)
2442{
2443 struct mount *mp;
2444
2445 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2446#ifdef INVARIANTS
2447 /* getnewvnode() calls v_incr_usecount() without holding interlock. */
2448 if (vp->v_type != VNON || vp->v_data != NULL)
2449 ASSERT_VI_LOCKED(vp, "vholdl");
2450#endif
2451 vp->v_holdcnt++;
2452 if ((vp->v_iflag & VI_FREE) == 0)
2453 return;
2454 VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count"));
2455 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2456 /*
2457 * Remove a vnode from the free list, mark it as in use,
2458 * and put it on the active list.
2459 */
2460 mtx_lock(&vnode_free_list_mtx);
2461 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2462 freevnodes--;
2463 vp->v_iflag &= ~VI_FREE;
2464 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2465 ("Activating already active vnode"));
2466 vp->v_iflag |= VI_ACTIVE;
2467 mp = vp->v_mount;
2468 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2469 mp->mnt_activevnodelistsize++;
2470 mtx_unlock(&vnode_free_list_mtx);
2471}
2472
2473/*
2474 * Note that there is one less who cares about this vnode.
2475 * vdrop() is the opposite of vhold().
2476 */
2477void
2478vdrop(struct vnode *vp)
2479{
2480
2481 VI_LOCK(vp);
2482 vdropl(vp);
2483}
2484
2485/*
2486 * Drop the hold count of the vnode. If this is the last reference to
2487 * the vnode we place it on the free list unless it has been vgone'd
2488 * (marked VI_DOOMED) in which case we will free it.
2489 *
2490 * Because the vnode vm object keeps a hold reference on the vnode if
2491 * there is at least one resident non-cached page, the vnode cannot
2492 * leave the active list without the page cleanup done.
2493 */
2494void
2495vdropl(struct vnode *vp)
2496{
2497 struct bufobj *bo;
2498 struct mount *mp;
2499 int active;
2500
2501 ASSERT_VI_LOCKED(vp, "vdropl");
2502 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2503 if (vp->v_holdcnt <= 0)
2504 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2505 vp->v_holdcnt--;
2506 if (vp->v_holdcnt > 0) {
2507 VI_UNLOCK(vp);
2508 return;
2509 }
2510 if ((vp->v_iflag & VI_DOOMED) == 0) {
2511 /*
2512 * Mark a vnode as free: remove it from its active list
2513 * and put it up for recycling on the freelist.
2514 */
2515 VNASSERT(vp->v_op != NULL, vp,
2516 ("vdropl: vnode already reclaimed."));
2517 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2518 ("vnode already free"));
2519 VNASSERT(vp->v_holdcnt == 0, vp,
2520 ("vdropl: freeing when we shouldn't"));
2521 active = vp->v_iflag & VI_ACTIVE;
2522 if ((vp->v_iflag & VI_OWEINACT) == 0) {
2523 vp->v_iflag &= ~VI_ACTIVE;
2524 mp = vp->v_mount;
2525 mtx_lock(&vnode_free_list_mtx);
2526 if (active) {
2527 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2528 v_actfreelist);
2529 mp->mnt_activevnodelistsize--;
2530 }
2531 TAILQ_INSERT_TAIL(&vnode_free_list, vp,
2532 v_actfreelist);
2533 freevnodes++;
2534 vp->v_iflag |= VI_FREE;
2535 mtx_unlock(&vnode_free_list_mtx);
2536 } else {
2537 atomic_add_long(&free_owe_inact, 1);
2538 }
2539 VI_UNLOCK(vp);
2540 return;
2541 }
2542 /*
2543 * The vnode has been marked for destruction, so free it.
2544 *
2545 * The vnode will be returned to the zone where it will
2546 * normally remain until it is needed for another vnode. We
2547 * need to cleanup (or verify that the cleanup has already
2548 * been done) any residual data left from its current use
2549 * so as not to contaminate the freshly allocated vnode.
2550 */
2551 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2552 atomic_subtract_long(&numvnodes, 1);
2553 bo = &vp->v_bufobj;
2554 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2555 ("cleaned vnode still on the free list."));
2556 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2557 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2558 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2559 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2560 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2561 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2562 VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2563 ("clean blk trie not empty"));
2564 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2565 VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2566 ("dirty blk trie not empty"));
2567 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2568 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2569 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2570 VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp,
2571 ("Dangling rangelock waiters"));
2572 VI_UNLOCK(vp);
2573#ifdef MAC
2574 mac_vnode_destroy(vp);
2575#endif
2576 if (vp->v_pollinfo != NULL) {
2577 destroy_vpollinfo(vp->v_pollinfo);
2578 vp->v_pollinfo = NULL;
2579 }
2580#ifdef INVARIANTS
2581 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2582 vp->v_op = NULL;
2583#endif
2584 bzero(&vp->v_un, sizeof(vp->v_un));
2585 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
2586 vp->v_iflag = 0;
2587 vp->v_vflag = 0;
2588 bo->bo_flag = 0;
2589 uma_zfree(vnode_zone, vp);
2590}
2591
2592/*
2593 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2594 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2595 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2596 * failed lock upgrade.
2597 */
2598void
2599vinactive(struct vnode *vp, struct thread *td)
2600{
2601 struct vm_object *obj;
2602
2603 ASSERT_VOP_ELOCKED(vp, "vinactive");
2604 ASSERT_VI_LOCKED(vp, "vinactive");
2605 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2606 ("vinactive: recursed on VI_DOINGINACT"));
2607 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2608 vp->v_iflag |= VI_DOINGINACT;
2609 vp->v_iflag &= ~VI_OWEINACT;
2610 VI_UNLOCK(vp);
2611 /*
2612 * Before moving off the active list, we must be sure that any
2613 * modified pages are converted into the vnode's dirty
2614 * buffers, since these will no longer be checked once the
2615 * vnode is on the inactive list.
2616 *
2617 * The write-out of the dirty pages is asynchronous. At the
2618 * point that VOP_INACTIVE() is called, there could still be
2619 * pending I/O and dirty pages in the object.
2620 */
2621 obj = vp->v_object;
2622 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2623 VM_OBJECT_WLOCK(obj);
2624 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2625 VM_OBJECT_WUNLOCK(obj);
2626 }
2627 VOP_INACTIVE(vp, td);
2628 VI_LOCK(vp);
2629 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2630 ("vinactive: lost VI_DOINGINACT"));
2631 vp->v_iflag &= ~VI_DOINGINACT;
2632}
2633
2634/*
2635 * Remove any vnodes in the vnode table belonging to mount point mp.
2636 *
2637 * If FORCECLOSE is not specified, there should not be any active ones,
2638 * return error if any are found (nb: this is a user error, not a
2639 * system error). If FORCECLOSE is specified, detach any active vnodes
2640 * that are found.
2641 *
2642 * If WRITECLOSE is set, only flush out regular file vnodes open for
2643 * writing.
2644 *
2645 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2646 *
2647 * `rootrefs' specifies the base reference count for the root vnode
2648 * of this filesystem. The root vnode is considered busy if its
2649 * v_usecount exceeds this value. On a successful return, vflush(, td)
2650 * will call vrele() on the root vnode exactly rootrefs times.
2651 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2652 * be zero.
2653 */
2654#ifdef DIAGNOSTIC
2655static int busyprt = 0; /* print out busy vnodes */
2656SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2657#endif
2658
2659int
2660vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2661{
2662 struct vnode *vp, *mvp, *rootvp = NULL;
2663 struct vattr vattr;
2664 int busy = 0, error;
2665
2666 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2667 rootrefs, flags);
2668 if (rootrefs > 0) {
2669 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2670 ("vflush: bad args"));
2671 /*
2672 * Get the filesystem root vnode. We can vput() it
2673 * immediately, since with rootrefs > 0, it won't go away.
2674 */
2675 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2676 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2677 __func__, error);
2678 return (error);
2679 }
2680 vput(rootvp);
2681 }
2682loop:
2683 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2684 vholdl(vp);
2685 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2686 if (error) {
2687 vdrop(vp);
2688 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2689 goto loop;
2690 }
2691 /*
2692 * Skip over a vnodes marked VV_SYSTEM.
2693 */
2694 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2695 VOP_UNLOCK(vp, 0);
2696 vdrop(vp);
2697 continue;
2698 }
2699 /*
2700 * If WRITECLOSE is set, flush out unlinked but still open
2701 * files (even if open only for reading) and regular file
2702 * vnodes open for writing.
2703 */
2704 if (flags & WRITECLOSE) {
2705 if (vp->v_object != NULL) {
2706 VM_OBJECT_WLOCK(vp->v_object);
2707 vm_object_page_clean(vp->v_object, 0, 0, 0);
2708 VM_OBJECT_WUNLOCK(vp->v_object);
2709 }
2710 error = VOP_FSYNC(vp, MNT_WAIT, td);
2711 if (error != 0) {
2712 VOP_UNLOCK(vp, 0);
2713 vdrop(vp);
2714 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2715 return (error);
2716 }
2717 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2718 VI_LOCK(vp);
2719
2720 if ((vp->v_type == VNON ||
2721 (error == 0 && vattr.va_nlink > 0)) &&
2722 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2723 VOP_UNLOCK(vp, 0);
2724 vdropl(vp);
2725 continue;
2726 }
2727 } else
2728 VI_LOCK(vp);
2729 /*
2730 * With v_usecount == 0, all we need to do is clear out the
2731 * vnode data structures and we are done.
2732 *
2733 * If FORCECLOSE is set, forcibly close the vnode.
2734 */
2735 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2736 vgonel(vp);
2737 } else {
2738 busy++;
2739#ifdef DIAGNOSTIC
2740 if (busyprt)
2741 vprint("vflush: busy vnode", vp);
2742#endif
2743 }
2744 VOP_UNLOCK(vp, 0);
2745 vdropl(vp);
2746 }
2747 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2748 /*
2749 * If just the root vnode is busy, and if its refcount
2750 * is equal to `rootrefs', then go ahead and kill it.
2751 */
2752 VI_LOCK(rootvp);
2753 KASSERT(busy > 0, ("vflush: not busy"));
2754 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2755 ("vflush: usecount %d < rootrefs %d",
2756 rootvp->v_usecount, rootrefs));
2757 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2758 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2759 vgone(rootvp);
2760 VOP_UNLOCK(rootvp, 0);
2761 busy = 0;
2762 } else
2763 VI_UNLOCK(rootvp);
2764 }
2765 if (busy) {
2766 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2767 busy);
2768 return (EBUSY);
2769 }
2770 for (; rootrefs > 0; rootrefs--)
2771 vrele(rootvp);
2772 return (0);
2773}
2774
2775/*
2776 * Recycle an unused vnode to the front of the free list.
2777 */
2778int
2779vrecycle(struct vnode *vp)
2780{
2781 int recycled;
2782
2783 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2784 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2785 recycled = 0;
2786 VI_LOCK(vp);
2787 if (vp->v_usecount == 0) {
2788 recycled = 1;
2789 vgonel(vp);
2790 }
2791 VI_UNLOCK(vp);
2792 return (recycled);
2793}
2794
2795/*
2796 * Eliminate all activity associated with a vnode
2797 * in preparation for reuse.
2798 */
2799void
2800vgone(struct vnode *vp)
2801{
2802 VI_LOCK(vp);
2803 vgonel(vp);
2804 VI_UNLOCK(vp);
2805}
2806
2807static void
2808notify_lowervp_vfs_dummy(struct mount *mp __unused,
2809 struct vnode *lowervp __unused)
2810{
2811}
2812
2813/*
2814 * Notify upper mounts about reclaimed or unlinked vnode.
2815 */
2816void
2817vfs_notify_upper(struct vnode *vp, int event)
2818{
2819 static struct vfsops vgonel_vfsops = {
2820 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2821 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2822 };
2823 struct mount *mp, *ump, *mmp;
2824
2825 mp = vp->v_mount;
2826 if (mp == NULL)
2827 return;
2828
2829 MNT_ILOCK(mp);
2830 if (TAILQ_EMPTY(&mp->mnt_uppers))
2831 goto unlock;
2832 MNT_IUNLOCK(mp);
2833 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2834 mmp->mnt_op = &vgonel_vfsops;
2835 mmp->mnt_kern_flag |= MNTK_MARKER;
2836 MNT_ILOCK(mp);
2837 mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2838 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2839 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2840 ump = TAILQ_NEXT(ump, mnt_upper_link);
2841 continue;
2842 }
2843 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2844 MNT_IUNLOCK(mp);
2845 switch (event) {
2846 case VFS_NOTIFY_UPPER_RECLAIM:
2847 VFS_RECLAIM_LOWERVP(ump, vp);
2848 break;
2849 case VFS_NOTIFY_UPPER_UNLINK:
2850 VFS_UNLINK_LOWERVP(ump, vp);
2851 break;
2852 default:
2853 KASSERT(0, ("invalid event %d", event));
2854 break;
2855 }
2856 MNT_ILOCK(mp);
2857 ump = TAILQ_NEXT(mmp, mnt_upper_link);
2858 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2859 }
2860 free(mmp, M_TEMP);
2861 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2862 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2863 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2864 wakeup(&mp->mnt_uppers);
2865 }
2866unlock:
2867 MNT_IUNLOCK(mp);
2868}
2869
2870/*
2871 * vgone, with the vp interlock held.
2872 */
2873static void
2874vgonel(struct vnode *vp)
2875{
2876 struct thread *td;
2877 int oweinact;
2878 int active;
2879 struct mount *mp;
2880
2881 ASSERT_VOP_ELOCKED(vp, "vgonel");
2882 ASSERT_VI_LOCKED(vp, "vgonel");
2883 VNASSERT(vp->v_holdcnt, vp,
2884 ("vgonel: vp %p has no reference.", vp));
2885 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2886 td = curthread;
2887
2888 /*
2889 * Don't vgonel if we're already doomed.
2890 */
2891 if (vp->v_iflag & VI_DOOMED)
2892 return;
2893 vp->v_iflag |= VI_DOOMED;
2894
2895 /*
2896 * Check to see if the vnode is in use. If so, we have to call
2897 * VOP_CLOSE() and VOP_INACTIVE().
2898 */
2899 active = vp->v_usecount;
2900 oweinact = (vp->v_iflag & VI_OWEINACT);
2901 VI_UNLOCK(vp);
2902 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2903
2904 /*
2905 * If purging an active vnode, it must be closed and
2906 * deactivated before being reclaimed.
2907 */
2908 if (active)
2909 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2910 if (oweinact || active) {
2911 VI_LOCK(vp);
2912 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2913 vinactive(vp, td);
2914 VI_UNLOCK(vp);
2915 }
2916 if (vp->v_type == VSOCK)
2917 vfs_unp_reclaim(vp);
2918
2919 /*
2920 * Clean out any buffers associated with the vnode.
2921 * If the flush fails, just toss the buffers.
2922 */
2923 mp = NULL;
2924 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2925 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2926 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
2927 while (vinvalbuf(vp, 0, 0, 0) != 0)
2928 ;
2929 }
2930
2931 BO_LOCK(&vp->v_bufobj);
2932 KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
2933 vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
2934 TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
2935 vp->v_bufobj.bo_clean.bv_cnt == 0,
2936 ("vp %p bufobj not invalidated", vp));
|
2938 BO_UNLOCK(&vp->v_bufobj);
2939
2940 /*
2941 * Reclaim the vnode.
2942 */
2943 if (VOP_RECLAIM(vp, td))
2944 panic("vgone: cannot reclaim");
2945 if (mp != NULL)
2946 vn_finished_secondary_write(mp);
2947 VNASSERT(vp->v_object == NULL, vp,
2948 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2949 /*
2950 * Clear the advisory locks and wake up waiting threads.
2951 */
2952 (void)VOP_ADVLOCKPURGE(vp);
2953 vp->v_lockf = NULL;
2954 /*
2955 * Delete from old mount point vnode list.
2956 */
2957 delmntque(vp);
2958 cache_purge(vp);
2959 /*
2960 * Done with purge, reset to the standard lock and invalidate
2961 * the vnode.
2962 */
2963 VI_LOCK(vp);
2964 vp->v_vnlock = &vp->v_lock;
2965 vp->v_op = &dead_vnodeops;
2966 vp->v_tag = "none";
2967 vp->v_type = VBAD;
2968}
2969
2970/*
2971 * Calculate the total number of references to a special device.
2972 */
2973int
2974vcount(struct vnode *vp)
2975{
2976 int count;
2977
2978 dev_lock();
2979 count = vp->v_rdev->si_usecount;
2980 dev_unlock();
2981 return (count);
2982}
2983
2984/*
2985 * Same as above, but using the struct cdev *as argument
2986 */
2987int
2988count_dev(struct cdev *dev)
2989{
2990 int count;
2991
2992 dev_lock();
2993 count = dev->si_usecount;
2994 dev_unlock();
2995 return(count);
2996}
2997
2998/*
2999 * Print out a description of a vnode.
3000 */
3001static char *typename[] =
3002{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
3003 "VMARKER"};
3004
3005void
3006vn_printf(struct vnode *vp, const char *fmt, ...)
3007{
3008 va_list ap;
3009 char buf[256], buf2[16];
3010 u_long flags;
3011
3012 va_start(ap, fmt);
3013 vprintf(fmt, ap);
3014 va_end(ap);
3015 printf("%p: ", (void *)vp);
3016 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
3017 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
3018 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
3019 buf[0] = '\0';
3020 buf[1] = '\0';
3021 if (vp->v_vflag & VV_ROOT)
3022 strlcat(buf, "|VV_ROOT", sizeof(buf));
3023 if (vp->v_vflag & VV_ISTTY)
3024 strlcat(buf, "|VV_ISTTY", sizeof(buf));
3025 if (vp->v_vflag & VV_NOSYNC)
3026 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
3027 if (vp->v_vflag & VV_ETERNALDEV)
3028 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
3029 if (vp->v_vflag & VV_CACHEDLABEL)
3030 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
3031 if (vp->v_vflag & VV_TEXT)
3032 strlcat(buf, "|VV_TEXT", sizeof(buf));
3033 if (vp->v_vflag & VV_COPYONWRITE)
3034 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
3035 if (vp->v_vflag & VV_SYSTEM)
3036 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
3037 if (vp->v_vflag & VV_PROCDEP)
3038 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
3039 if (vp->v_vflag & VV_NOKNOTE)
3040 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
3041 if (vp->v_vflag & VV_DELETED)
3042 strlcat(buf, "|VV_DELETED", sizeof(buf));
3043 if (vp->v_vflag & VV_MD)
3044 strlcat(buf, "|VV_MD", sizeof(buf));
3045 if (vp->v_vflag & VV_FORCEINSMQ)
3046 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
3047 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
3048 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
3049 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
3050 if (flags != 0) {
3051 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
3052 strlcat(buf, buf2, sizeof(buf));
3053 }
3054 if (vp->v_iflag & VI_MOUNT)
3055 strlcat(buf, "|VI_MOUNT", sizeof(buf));
3056 if (vp->v_iflag & VI_DOOMED)
3057 strlcat(buf, "|VI_DOOMED", sizeof(buf));
3058 if (vp->v_iflag & VI_FREE)
3059 strlcat(buf, "|VI_FREE", sizeof(buf));
3060 if (vp->v_iflag & VI_ACTIVE)
3061 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
3062 if (vp->v_iflag & VI_DOINGINACT)
3063 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
3064 if (vp->v_iflag & VI_OWEINACT)
3065 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
3066 flags = vp->v_iflag & ~(VI_MOUNT | VI_DOOMED | VI_FREE |
3067 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
3068 if (flags != 0) {
3069 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
3070 strlcat(buf, buf2, sizeof(buf));
3071 }
3072 printf(" flags (%s)\n", buf + 1);
3073 if (mtx_owned(VI_MTX(vp)))
3074 printf(" VI_LOCKed");
3075 if (vp->v_object != NULL)
3076 printf(" v_object %p ref %d pages %d "
3077 "cleanbuf %d dirtybuf %d\n",
3078 vp->v_object, vp->v_object->ref_count,
3079 vp->v_object->resident_page_count,
3080 vp->v_bufobj.bo_clean.bv_cnt,
3081 vp->v_bufobj.bo_dirty.bv_cnt);
3082 printf(" ");
3083 lockmgr_printinfo(vp->v_vnlock);
3084 if (vp->v_data != NULL)
3085 VOP_PRINT(vp);
3086}
3087
3088#ifdef DDB
3089/*
3090 * List all of the locked vnodes in the system.
3091 * Called when debugging the kernel.
3092 */
3093DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
3094{
3095 struct mount *mp;
3096 struct vnode *vp;
3097
3098 /*
3099 * Note: because this is DDB, we can't obey the locking semantics
3100 * for these structures, which means we could catch an inconsistent
3101 * state and dereference a nasty pointer. Not much to be done
3102 * about that.
3103 */
3104 db_printf("Locked vnodes\n");
3105 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3106 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3107 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
3108 vprint("", vp);
3109 }
3110 }
3111}
3112
3113/*
3114 * Show details about the given vnode.
3115 */
3116DB_SHOW_COMMAND(vnode, db_show_vnode)
3117{
3118 struct vnode *vp;
3119
3120 if (!have_addr)
3121 return;
3122 vp = (struct vnode *)addr;
3123 vn_printf(vp, "vnode ");
3124}
3125
3126/*
3127 * Show details about the given mount point.
3128 */
3129DB_SHOW_COMMAND(mount, db_show_mount)
3130{
3131 struct mount *mp;
3132 struct vfsopt *opt;
3133 struct statfs *sp;
3134 struct vnode *vp;
3135 char buf[512];
3136 uint64_t mflags;
3137 u_int flags;
3138
3139 if (!have_addr) {
3140 /* No address given, print short info about all mount points. */
3141 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3142 db_printf("%p %s on %s (%s)\n", mp,
3143 mp->mnt_stat.f_mntfromname,
3144 mp->mnt_stat.f_mntonname,
3145 mp->mnt_stat.f_fstypename);
3146 if (db_pager_quit)
3147 break;
3148 }
3149 db_printf("\nMore info: show mount <addr>\n");
3150 return;
3151 }
3152
3153 mp = (struct mount *)addr;
3154 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3155 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3156
3157 buf[0] = '\0';
3158 mflags = mp->mnt_flag;
3159#define MNT_FLAG(flag) do { \
3160 if (mflags & (flag)) { \
3161 if (buf[0] != '\0') \
3162 strlcat(buf, ", ", sizeof(buf)); \
3163 strlcat(buf, (#flag) + 4, sizeof(buf)); \
3164 mflags &= ~(flag); \
3165 } \
3166} while (0)
3167 MNT_FLAG(MNT_RDONLY);
3168 MNT_FLAG(MNT_SYNCHRONOUS);
3169 MNT_FLAG(MNT_NOEXEC);
3170 MNT_FLAG(MNT_NOSUID);
3171 MNT_FLAG(MNT_NFS4ACLS);
3172 MNT_FLAG(MNT_UNION);
3173 MNT_FLAG(MNT_ASYNC);
3174 MNT_FLAG(MNT_SUIDDIR);
3175 MNT_FLAG(MNT_SOFTDEP);
3176 MNT_FLAG(MNT_NOSYMFOLLOW);
3177 MNT_FLAG(MNT_GJOURNAL);
3178 MNT_FLAG(MNT_MULTILABEL);
3179 MNT_FLAG(MNT_ACLS);
3180 MNT_FLAG(MNT_NOATIME);
3181 MNT_FLAG(MNT_NOCLUSTERR);
3182 MNT_FLAG(MNT_NOCLUSTERW);
3183 MNT_FLAG(MNT_SUJ);
3184 MNT_FLAG(MNT_EXRDONLY);
3185 MNT_FLAG(MNT_EXPORTED);
3186 MNT_FLAG(MNT_DEFEXPORTED);
3187 MNT_FLAG(MNT_EXPORTANON);
3188 MNT_FLAG(MNT_EXKERB);
3189 MNT_FLAG(MNT_EXPUBLIC);
3190 MNT_FLAG(MNT_LOCAL);
3191 MNT_FLAG(MNT_QUOTA);
3192 MNT_FLAG(MNT_ROOTFS);
3193 MNT_FLAG(MNT_USER);
3194 MNT_FLAG(MNT_IGNORE);
3195 MNT_FLAG(MNT_UPDATE);
3196 MNT_FLAG(MNT_DELEXPORT);
3197 MNT_FLAG(MNT_RELOAD);
3198 MNT_FLAG(MNT_FORCE);
3199 MNT_FLAG(MNT_SNAPSHOT);
3200 MNT_FLAG(MNT_BYFSID);
3201#undef MNT_FLAG
3202 if (mflags != 0) {
3203 if (buf[0] != '\0')
3204 strlcat(buf, ", ", sizeof(buf));
3205 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3206 "0x%016jx", mflags);
3207 }
3208 db_printf(" mnt_flag = %s\n", buf);
3209
3210 buf[0] = '\0';
3211 flags = mp->mnt_kern_flag;
3212#define MNT_KERN_FLAG(flag) do { \
3213 if (flags & (flag)) { \
3214 if (buf[0] != '\0') \
3215 strlcat(buf, ", ", sizeof(buf)); \
3216 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3217 flags &= ~(flag); \
3218 } \
3219} while (0)
3220 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3221 MNT_KERN_FLAG(MNTK_ASYNC);
3222 MNT_KERN_FLAG(MNTK_SOFTDEP);
3223 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3224 MNT_KERN_FLAG(MNTK_DRAINING);
3225 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3226 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3227 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3228 MNT_KERN_FLAG(MNTK_NO_IOPF);
3229 MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3230 MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3231 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3232 MNT_KERN_FLAG(MNTK_MARKER);
3233 MNT_KERN_FLAG(MNTK_USES_BCACHE);
3234 MNT_KERN_FLAG(MNTK_NOASYNC);
3235 MNT_KERN_FLAG(MNTK_UNMOUNT);
3236 MNT_KERN_FLAG(MNTK_MWAIT);
3237 MNT_KERN_FLAG(MNTK_SUSPEND);
3238 MNT_KERN_FLAG(MNTK_SUSPEND2);
3239 MNT_KERN_FLAG(MNTK_SUSPENDED);
3240 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3241 MNT_KERN_FLAG(MNTK_NOKNOTE);
3242#undef MNT_KERN_FLAG
3243 if (flags != 0) {
3244 if (buf[0] != '\0')
3245 strlcat(buf, ", ", sizeof(buf));
3246 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3247 "0x%08x", flags);
3248 }
3249 db_printf(" mnt_kern_flag = %s\n", buf);
3250
3251 db_printf(" mnt_opt = ");
3252 opt = TAILQ_FIRST(mp->mnt_opt);
3253 if (opt != NULL) {
3254 db_printf("%s", opt->name);
3255 opt = TAILQ_NEXT(opt, link);
3256 while (opt != NULL) {
3257 db_printf(", %s", opt->name);
3258 opt = TAILQ_NEXT(opt, link);
3259 }
3260 }
3261 db_printf("\n");
3262
3263 sp = &mp->mnt_stat;
3264 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3265 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3266 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3267 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3268 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3269 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3270 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3271 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3272 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3273 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3274 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3275 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3276
3277 db_printf(" mnt_cred = { uid=%u ruid=%u",
3278 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3279 if (jailed(mp->mnt_cred))
3280 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3281 db_printf(" }\n");
3282 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3283 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3284 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3285 db_printf(" mnt_activevnodelistsize = %d\n",
3286 mp->mnt_activevnodelistsize);
3287 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3288 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3289 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3290 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3291 db_printf(" mnt_lockref = %d\n", mp->mnt_lockref);
3292 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3293 db_printf(" mnt_secondary_accwrites = %d\n",
3294 mp->mnt_secondary_accwrites);
3295 db_printf(" mnt_gjprovider = %s\n",
3296 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3297
3298 db_printf("\n\nList of active vnodes\n");
3299 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3300 if (vp->v_type != VMARKER) {
3301 vn_printf(vp, "vnode ");
3302 if (db_pager_quit)
3303 break;
3304 }
3305 }
3306 db_printf("\n\nList of inactive vnodes\n");
3307 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3308 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3309 vn_printf(vp, "vnode ");
3310 if (db_pager_quit)
3311 break;
3312 }
3313 }
3314}
3315#endif /* DDB */
3316
3317/*
3318 * Fill in a struct xvfsconf based on a struct vfsconf.
3319 */
3320static int
3321vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3322{
3323 struct xvfsconf xvfsp;
3324
3325 bzero(&xvfsp, sizeof(xvfsp));
3326 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3327 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3328 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3329 xvfsp.vfc_flags = vfsp->vfc_flags;
3330 /*
3331 * These are unused in userland, we keep them
3332 * to not break binary compatibility.
3333 */
3334 xvfsp.vfc_vfsops = NULL;
3335 xvfsp.vfc_next = NULL;
3336 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3337}
3338
3339#ifdef COMPAT_FREEBSD32
3340struct xvfsconf32 {
3341 uint32_t vfc_vfsops;
3342 char vfc_name[MFSNAMELEN];
3343 int32_t vfc_typenum;
3344 int32_t vfc_refcount;
3345 int32_t vfc_flags;
3346 uint32_t vfc_next;
3347};
3348
3349static int
3350vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3351{
3352 struct xvfsconf32 xvfsp;
3353
3354 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3355 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3356 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3357 xvfsp.vfc_flags = vfsp->vfc_flags;
3358 xvfsp.vfc_vfsops = 0;
3359 xvfsp.vfc_next = 0;
3360 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3361}
3362#endif
3363
3364/*
3365 * Top level filesystem related information gathering.
3366 */
3367static int
3368sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3369{
3370 struct vfsconf *vfsp;
3371 int error;
3372
3373 error = 0;
3374 vfsconf_slock();
3375 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3376#ifdef COMPAT_FREEBSD32
3377 if (req->flags & SCTL_MASK32)
3378 error = vfsconf2x32(req, vfsp);
3379 else
3380#endif
3381 error = vfsconf2x(req, vfsp);
3382 if (error)
3383 break;
3384 }
3385 vfsconf_sunlock();
3386 return (error);
3387}
3388
3389SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
3390 CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
3391 "S,xvfsconf", "List of all configured filesystems");
3392
3393#ifndef BURN_BRIDGES
3394static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3395
3396static int
3397vfs_sysctl(SYSCTL_HANDLER_ARGS)
3398{
3399 int *name = (int *)arg1 - 1; /* XXX */
3400 u_int namelen = arg2 + 1; /* XXX */
3401 struct vfsconf *vfsp;
3402
3403 log(LOG_WARNING, "userland calling deprecated sysctl, "
3404 "please rebuild world\n");
3405
3406#if 1 || defined(COMPAT_PRELITE2)
3407 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3408 if (namelen == 1)
3409 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3410#endif
3411
3412 switch (name[1]) {
3413 case VFS_MAXTYPENUM:
3414 if (namelen != 2)
3415 return (ENOTDIR);
3416 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3417 case VFS_CONF:
3418 if (namelen != 3)
3419 return (ENOTDIR); /* overloaded */
3420 vfsconf_slock();
3421 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3422 if (vfsp->vfc_typenum == name[2])
3423 break;
3424 }
3425 vfsconf_sunlock();
3426 if (vfsp == NULL)
3427 return (EOPNOTSUPP);
3428#ifdef COMPAT_FREEBSD32
3429 if (req->flags & SCTL_MASK32)
3430 return (vfsconf2x32(req, vfsp));
3431 else
3432#endif
3433 return (vfsconf2x(req, vfsp));
3434 }
3435 return (EOPNOTSUPP);
3436}
3437
3438static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
3439 CTLFLAG_MPSAFE, vfs_sysctl,
3440 "Generic filesystem");
3441
3442#if 1 || defined(COMPAT_PRELITE2)
3443
3444static int
3445sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3446{
3447 int error;
3448 struct vfsconf *vfsp;
3449 struct ovfsconf ovfs;
3450
3451 vfsconf_slock();
3452 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3453 bzero(&ovfs, sizeof(ovfs));
3454 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3455 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3456 ovfs.vfc_index = vfsp->vfc_typenum;
3457 ovfs.vfc_refcount = vfsp->vfc_refcount;
3458 ovfs.vfc_flags = vfsp->vfc_flags;
3459 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3460 if (error != 0) {
3461 vfsconf_sunlock();
3462 return (error);
3463 }
3464 }
3465 vfsconf_sunlock();
3466 return (0);
3467}
3468
3469#endif /* 1 || COMPAT_PRELITE2 */
3470#endif /* !BURN_BRIDGES */
3471
3472#define KINFO_VNODESLOP 10
3473#ifdef notyet
3474/*
3475 * Dump vnode list (via sysctl).
3476 */
3477/* ARGSUSED */
3478static int
3479sysctl_vnode(SYSCTL_HANDLER_ARGS)
3480{
3481 struct xvnode *xvn;
3482 struct mount *mp;
3483 struct vnode *vp;
3484 int error, len, n;
3485
3486 /*
3487 * Stale numvnodes access is not fatal here.
3488 */
3489 req->lock = 0;
3490 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3491 if (!req->oldptr)
3492 /* Make an estimate */
3493 return (SYSCTL_OUT(req, 0, len));
3494
3495 error = sysctl_wire_old_buffer(req, 0);
3496 if (error != 0)
3497 return (error);
3498 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3499 n = 0;
3500 mtx_lock(&mountlist_mtx);
3501 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3502 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3503 continue;
3504 MNT_ILOCK(mp);
3505 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3506 if (n == len)
3507 break;
3508 vref(vp);
3509 xvn[n].xv_size = sizeof *xvn;
3510 xvn[n].xv_vnode = vp;
3511 xvn[n].xv_id = 0; /* XXX compat */
3512#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3513 XV_COPY(usecount);
3514 XV_COPY(writecount);
3515 XV_COPY(holdcnt);
3516 XV_COPY(mount);
3517 XV_COPY(numoutput);
3518 XV_COPY(type);
3519#undef XV_COPY
3520 xvn[n].xv_flag = vp->v_vflag;
3521
3522 switch (vp->v_type) {
3523 case VREG:
3524 case VDIR:
3525 case VLNK:
3526 break;
3527 case VBLK:
3528 case VCHR:
3529 if (vp->v_rdev == NULL) {
3530 vrele(vp);
3531 continue;
3532 }
3533 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3534 break;
3535 case VSOCK:
3536 xvn[n].xv_socket = vp->v_socket;
3537 break;
3538 case VFIFO:
3539 xvn[n].xv_fifo = vp->v_fifoinfo;
3540 break;
3541 case VNON:
3542 case VBAD:
3543 default:
3544 /* shouldn't happen? */
3545 vrele(vp);
3546 continue;
3547 }
3548 vrele(vp);
3549 ++n;
3550 }
3551 MNT_IUNLOCK(mp);
3552 mtx_lock(&mountlist_mtx);
3553 vfs_unbusy(mp);
3554 if (n == len)
3555 break;
3556 }
3557 mtx_unlock(&mountlist_mtx);
3558
3559 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3560 free(xvn, M_TEMP);
3561 return (error);
3562}
3563
3564SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
3565 CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
3566 "");
3567#endif
3568
3569static void
3570unmount_or_warn(struct mount *mp)
3571{
3572 int error;
3573
3574 error = dounmount(mp, MNT_FORCE, curthread);
3575 if (error != 0 && strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3576 printf("unmount of %s failed (", mp->mnt_stat.f_mntonname);
3577 if (error == EBUSY)
3578 printf("BUSY)\n");
3579 else
3580 printf("%d)\n", error);
3581 }
3582}
3583
3584/*
3585 * Unmount all filesystems. The list is traversed in reverse order
3586 * of mounting to avoid dependencies.
3587 */
3588void
3589vfs_unmountall(void)
3590{
3591 struct mount *mp, *tmp;
3592
3593 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3594
3595 /*
3596 * Since this only runs when rebooting, it is not interlocked.
3597 */
3598 TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) {
3599 vfs_ref(mp);
3600
3601 /*
3602 * Forcibly unmounting "/dev" before "/" would prevent clean
3603 * unmount of the latter.
3604 */
3605 if (mp == rootdevmp)
3606 continue;
3607
3608 unmount_or_warn(mp);
3609 }
3610
3611 if (rootdevmp != NULL)
3612 unmount_or_warn(rootdevmp);
3613}
3614
3615/*
3616 * perform msync on all vnodes under a mount point
3617 * the mount point must be locked.
3618 */
3619void
3620vfs_msync(struct mount *mp, int flags)
3621{
3622 struct vnode *vp, *mvp;
3623 struct vm_object *obj;
3624
3625 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3626 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3627 obj = vp->v_object;
3628 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3629 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3630 if (!vget(vp,
3631 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3632 curthread)) {
3633 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3634 vput(vp);
3635 continue;
3636 }
3637
3638 obj = vp->v_object;
3639 if (obj != NULL) {
3640 VM_OBJECT_WLOCK(obj);
3641 vm_object_page_clean(obj, 0, 0,
3642 flags == MNT_WAIT ?
3643 OBJPC_SYNC : OBJPC_NOSYNC);
3644 VM_OBJECT_WUNLOCK(obj);
3645 }
3646 vput(vp);
3647 }
3648 } else
3649 VI_UNLOCK(vp);
3650 }
3651}
3652
3653static void
3654destroy_vpollinfo_free(struct vpollinfo *vi)
3655{
3656
3657 knlist_destroy(&vi->vpi_selinfo.si_note);
3658 mtx_destroy(&vi->vpi_lock);
3659 uma_zfree(vnodepoll_zone, vi);
3660}
3661
3662static void
3663destroy_vpollinfo(struct vpollinfo *vi)
3664{
3665
3666 knlist_clear(&vi->vpi_selinfo.si_note, 1);
3667 seldrain(&vi->vpi_selinfo);
3668 destroy_vpollinfo_free(vi);
3669}
3670
3671/*
3672 * Initalize per-vnode helper structure to hold poll-related state.
3673 */
3674void
3675v_addpollinfo(struct vnode *vp)
3676{
3677 struct vpollinfo *vi;
3678
3679 if (vp->v_pollinfo != NULL)
3680 return;
3681 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3682 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3683 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3684 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3685 VI_LOCK(vp);
3686 if (vp->v_pollinfo != NULL) {
3687 VI_UNLOCK(vp);
3688 destroy_vpollinfo_free(vi);
3689 return;
3690 }
3691 vp->v_pollinfo = vi;
3692 VI_UNLOCK(vp);
3693}
3694
3695/*
3696 * Record a process's interest in events which might happen to
3697 * a vnode. Because poll uses the historic select-style interface
3698 * internally, this routine serves as both the ``check for any
3699 * pending events'' and the ``record my interest in future events''
3700 * functions. (These are done together, while the lock is held,
3701 * to avoid race conditions.)
3702 */
3703int
3704vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3705{
3706
3707 v_addpollinfo(vp);
3708 mtx_lock(&vp->v_pollinfo->vpi_lock);
3709 if (vp->v_pollinfo->vpi_revents & events) {
3710 /*
3711 * This leaves events we are not interested
3712 * in available for the other process which
3713 * which presumably had requested them
3714 * (otherwise they would never have been
3715 * recorded).
3716 */
3717 events &= vp->v_pollinfo->vpi_revents;
3718 vp->v_pollinfo->vpi_revents &= ~events;
3719
3720 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3721 return (events);
3722 }
3723 vp->v_pollinfo->vpi_events |= events;
3724 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3725 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3726 return (0);
3727}
3728
3729/*
3730 * Routine to create and manage a filesystem syncer vnode.
3731 */
3732#define sync_close ((int (*)(struct vop_close_args *))nullop)
3733static int sync_fsync(struct vop_fsync_args *);
3734static int sync_inactive(struct vop_inactive_args *);
3735static int sync_reclaim(struct vop_reclaim_args *);
3736
3737static struct vop_vector sync_vnodeops = {
3738 .vop_bypass = VOP_EOPNOTSUPP,
3739 .vop_close = sync_close, /* close */
3740 .vop_fsync = sync_fsync, /* fsync */
3741 .vop_inactive = sync_inactive, /* inactive */
3742 .vop_reclaim = sync_reclaim, /* reclaim */
3743 .vop_lock1 = vop_stdlock, /* lock */
3744 .vop_unlock = vop_stdunlock, /* unlock */
3745 .vop_islocked = vop_stdislocked, /* islocked */
3746};
3747
3748/*
3749 * Create a new filesystem syncer vnode for the specified mount point.
3750 */
3751void
3752vfs_allocate_syncvnode(struct mount *mp)
3753{
3754 struct vnode *vp;
3755 struct bufobj *bo;
3756 static long start, incr, next;
3757 int error;
3758
3759 /* Allocate a new vnode */
3760 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3761 if (error != 0)
3762 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3763 vp->v_type = VNON;
3764 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3765 vp->v_vflag |= VV_FORCEINSMQ;
3766 error = insmntque(vp, mp);
3767 if (error != 0)
3768 panic("vfs_allocate_syncvnode: insmntque() failed");
3769 vp->v_vflag &= ~VV_FORCEINSMQ;
3770 VOP_UNLOCK(vp, 0);
3771 /*
3772 * Place the vnode onto the syncer worklist. We attempt to
3773 * scatter them about on the list so that they will go off
3774 * at evenly distributed times even if all the filesystems
3775 * are mounted at once.
3776 */
3777 next += incr;
3778 if (next == 0 || next > syncer_maxdelay) {
3779 start /= 2;
3780 incr /= 2;
3781 if (start == 0) {
3782 start = syncer_maxdelay / 2;
3783 incr = syncer_maxdelay;
3784 }
3785 next = start;
3786 }
3787 bo = &vp->v_bufobj;
3788 BO_LOCK(bo);
3789 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3790 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3791 mtx_lock(&sync_mtx);
3792 sync_vnode_count++;
3793 if (mp->mnt_syncer == NULL) {
3794 mp->mnt_syncer = vp;
3795 vp = NULL;
3796 }
3797 mtx_unlock(&sync_mtx);
3798 BO_UNLOCK(bo);
3799 if (vp != NULL) {
3800 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3801 vgone(vp);
3802 vput(vp);
3803 }
3804}
3805
3806void
3807vfs_deallocate_syncvnode(struct mount *mp)
3808{
3809 struct vnode *vp;
3810
3811 mtx_lock(&sync_mtx);
3812 vp = mp->mnt_syncer;
3813 if (vp != NULL)
3814 mp->mnt_syncer = NULL;
3815 mtx_unlock(&sync_mtx);
3816 if (vp != NULL)
3817 vrele(vp);
3818}
3819
3820/*
3821 * Do a lazy sync of the filesystem.
3822 */
3823static int
3824sync_fsync(struct vop_fsync_args *ap)
3825{
3826 struct vnode *syncvp = ap->a_vp;
3827 struct mount *mp = syncvp->v_mount;
3828 int error, save;
3829 struct bufobj *bo;
3830
3831 /*
3832 * We only need to do something if this is a lazy evaluation.
3833 */
3834 if (ap->a_waitfor != MNT_LAZY)
3835 return (0);
3836
3837 /*
3838 * Move ourselves to the back of the sync list.
3839 */
3840 bo = &syncvp->v_bufobj;
3841 BO_LOCK(bo);
3842 vn_syncer_add_to_worklist(bo, syncdelay);
3843 BO_UNLOCK(bo);
3844
3845 /*
3846 * Walk the list of vnodes pushing all that are dirty and
3847 * not already on the sync list.
3848 */
3849 if (vfs_busy(mp, MBF_NOWAIT) != 0)
3850 return (0);
3851 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3852 vfs_unbusy(mp);
3853 return (0);
3854 }
3855 save = curthread_pflags_set(TDP_SYNCIO);
3856 vfs_msync(mp, MNT_NOWAIT);
3857 error = VFS_SYNC(mp, MNT_LAZY);
3858 curthread_pflags_restore(save);
3859 vn_finished_write(mp);
3860 vfs_unbusy(mp);
3861 return (error);
3862}
3863
3864/*
3865 * The syncer vnode is no referenced.
3866 */
3867static int
3868sync_inactive(struct vop_inactive_args *ap)
3869{
3870
3871 vgone(ap->a_vp);
3872 return (0);
3873}
3874
3875/*
3876 * The syncer vnode is no longer needed and is being decommissioned.
3877 *
3878 * Modifications to the worklist must be protected by sync_mtx.
3879 */
3880static int
3881sync_reclaim(struct vop_reclaim_args *ap)
3882{
3883 struct vnode *vp = ap->a_vp;
3884 struct bufobj *bo;
3885
3886 bo = &vp->v_bufobj;
3887 BO_LOCK(bo);
3888 mtx_lock(&sync_mtx);
3889 if (vp->v_mount->mnt_syncer == vp)
3890 vp->v_mount->mnt_syncer = NULL;
3891 if (bo->bo_flag & BO_ONWORKLST) {
3892 LIST_REMOVE(bo, bo_synclist);
3893 syncer_worklist_len--;
3894 sync_vnode_count--;
3895 bo->bo_flag &= ~BO_ONWORKLST;
3896 }
3897 mtx_unlock(&sync_mtx);
3898 BO_UNLOCK(bo);
3899
3900 return (0);
3901}
3902
3903/*
3904 * Check if vnode represents a disk device
3905 */
3906int
3907vn_isdisk(struct vnode *vp, int *errp)
3908{
3909 int error;
3910
3911 if (vp->v_type != VCHR) {
3912 error = ENOTBLK;
3913 goto out;
3914 }
3915 error = 0;
3916 dev_lock();
3917 if (vp->v_rdev == NULL)
3918 error = ENXIO;
3919 else if (vp->v_rdev->si_devsw == NULL)
3920 error = ENXIO;
3921 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3922 error = ENOTBLK;
3923 dev_unlock();
3924out:
3925 if (errp != NULL)
3926 *errp = error;
3927 return (error == 0);
3928}
3929
3930/*
3931 * Common filesystem object access control check routine. Accepts a
3932 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3933 * and optional call-by-reference privused argument allowing vaccess()
3934 * to indicate to the caller whether privilege was used to satisfy the
3935 * request (obsoleted). Returns 0 on success, or an errno on failure.
3936 */
3937int
3938vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3939 accmode_t accmode, struct ucred *cred, int *privused)
3940{
3941 accmode_t dac_granted;
3942 accmode_t priv_granted;
3943
3944 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3945 ("invalid bit in accmode"));
3946 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3947 ("VAPPEND without VWRITE"));
3948
3949 /*
3950 * Look for a normal, non-privileged way to access the file/directory
3951 * as requested. If it exists, go with that.
3952 */
3953
3954 if (privused != NULL)
3955 *privused = 0;
3956
3957 dac_granted = 0;
3958
3959 /* Check the owner. */
3960 if (cred->cr_uid == file_uid) {
3961 dac_granted |= VADMIN;
3962 if (file_mode & S_IXUSR)
3963 dac_granted |= VEXEC;
3964 if (file_mode & S_IRUSR)
3965 dac_granted |= VREAD;
3966 if (file_mode & S_IWUSR)
3967 dac_granted |= (VWRITE | VAPPEND);
3968
3969 if ((accmode & dac_granted) == accmode)
3970 return (0);
3971
3972 goto privcheck;
3973 }
3974
3975 /* Otherwise, check the groups (first match) */
3976 if (groupmember(file_gid, cred)) {
3977 if (file_mode & S_IXGRP)
3978 dac_granted |= VEXEC;
3979 if (file_mode & S_IRGRP)
3980 dac_granted |= VREAD;
3981 if (file_mode & S_IWGRP)
3982 dac_granted |= (VWRITE | VAPPEND);
3983
3984 if ((accmode & dac_granted) == accmode)
3985 return (0);
3986
3987 goto privcheck;
3988 }
3989
3990 /* Otherwise, check everyone else. */
3991 if (file_mode & S_IXOTH)
3992 dac_granted |= VEXEC;
3993 if (file_mode & S_IROTH)
3994 dac_granted |= VREAD;
3995 if (file_mode & S_IWOTH)
3996 dac_granted |= (VWRITE | VAPPEND);
3997 if ((accmode & dac_granted) == accmode)
3998 return (0);
3999
4000privcheck:
4001 /*
4002 * Build a privilege mask to determine if the set of privileges
4003 * satisfies the requirements when combined with the granted mask
4004 * from above. For each privilege, if the privilege is required,
4005 * bitwise or the request type onto the priv_granted mask.
4006 */
4007 priv_granted = 0;
4008
4009 if (type == VDIR) {
4010 /*
4011 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
4012 * requests, instead of PRIV_VFS_EXEC.
4013 */
4014 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4015 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
4016 priv_granted |= VEXEC;
4017 } else {
4018 /*
4019 * Ensure that at least one execute bit is on. Otherwise,
4020 * a privileged user will always succeed, and we don't want
4021 * this to happen unless the file really is executable.
4022 */
4023 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4024 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
4025 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
4026 priv_granted |= VEXEC;
4027 }
4028
4029 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
4030 !priv_check_cred(cred, PRIV_VFS_READ, 0))
4031 priv_granted |= VREAD;
4032
4033 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
4034 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
4035 priv_granted |= (VWRITE | VAPPEND);
4036
4037 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
4038 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
4039 priv_granted |= VADMIN;
4040
4041 if ((accmode & (priv_granted | dac_granted)) == accmode) {
4042 /* XXX audit: privilege used */
4043 if (privused != NULL)
4044 *privused = 1;
4045 return (0);
4046 }
4047
4048 return ((accmode & VADMIN) ? EPERM : EACCES);
4049}
4050
4051/*
4052 * Credential check based on process requesting service, and per-attribute
4053 * permissions.
4054 */
4055int
4056extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
4057 struct thread *td, accmode_t accmode)
4058{
4059
4060 /*
4061 * Kernel-invoked always succeeds.
4062 */
4063 if (cred == NOCRED)
4064 return (0);
4065
4066 /*
4067 * Do not allow privileged processes in jail to directly manipulate
4068 * system attributes.
4069 */
4070 switch (attrnamespace) {
4071 case EXTATTR_NAMESPACE_SYSTEM:
4072 /* Potentially should be: return (EPERM); */
4073 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
4074 case EXTATTR_NAMESPACE_USER:
4075 return (VOP_ACCESS(vp, accmode, cred, td));
4076 default:
4077 return (EPERM);
4078 }
4079}
4080
4081#ifdef DEBUG_VFS_LOCKS
4082/*
4083 * This only exists to supress warnings from unlocked specfs accesses. It is
4084 * no longer ok to have an unlocked VFS.
4085 */
4086#define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
4087 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
4088
4089int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
4090SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
4091 "Drop into debugger on lock violation");
4092
4093int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
4094SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
4095 0, "Check for interlock across VOPs");
4096
4097int vfs_badlock_print = 1; /* Print lock violations. */
4098SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
4099 0, "Print lock violations");
4100
4101#ifdef KDB
4102int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
4103SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
4104 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
4105#endif
4106
4107static void
4108vfs_badlock(const char *msg, const char *str, struct vnode *vp)
4109{
4110
4111#ifdef KDB
4112 if (vfs_badlock_backtrace)
4113 kdb_backtrace();
4114#endif
4115 if (vfs_badlock_print)
4116 printf("%s: %p %s\n", str, (void *)vp, msg);
4117 if (vfs_badlock_ddb)
4118 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4119}
4120
4121void
4122assert_vi_locked(struct vnode *vp, const char *str)
4123{
4124
4125 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4126 vfs_badlock("interlock is not locked but should be", str, vp);
4127}
4128
4129void
4130assert_vi_unlocked(struct vnode *vp, const char *str)
4131{
4132
4133 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4134 vfs_badlock("interlock is locked but should not be", str, vp);
4135}
4136
4137void
4138assert_vop_locked(struct vnode *vp, const char *str)
4139{
4140 int locked;
4141
4142 if (!IGNORE_LOCK(vp)) {
4143 locked = VOP_ISLOCKED(vp);
4144 if (locked == 0 || locked == LK_EXCLOTHER)
4145 vfs_badlock("is not locked but should be", str, vp);
4146 }
4147}
4148
4149void
4150assert_vop_unlocked(struct vnode *vp, const char *str)
4151{
4152
4153 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4154 vfs_badlock("is locked but should not be", str, vp);
4155}
4156
4157void
4158assert_vop_elocked(struct vnode *vp, const char *str)
4159{
4160
4161 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4162 vfs_badlock("is not exclusive locked but should be", str, vp);
4163}
4164
4165#if 0
4166void
4167assert_vop_elocked_other(struct vnode *vp, const char *str)
4168{
4169
4170 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4171 vfs_badlock("is not exclusive locked by another thread",
4172 str, vp);
4173}
4174
4175void
4176assert_vop_slocked(struct vnode *vp, const char *str)
4177{
4178
4179 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4180 vfs_badlock("is not locked shared but should be", str, vp);
4181}
4182#endif /* 0 */
4183#endif /* DEBUG_VFS_LOCKS */
4184
4185void
4186vop_rename_fail(struct vop_rename_args *ap)
4187{
4188
4189 if (ap->a_tvp != NULL)
4190 vput(ap->a_tvp);
4191 if (ap->a_tdvp == ap->a_tvp)
4192 vrele(ap->a_tdvp);
4193 else
4194 vput(ap->a_tdvp);
4195 vrele(ap->a_fdvp);
4196 vrele(ap->a_fvp);
4197}
4198
4199void
4200vop_rename_pre(void *ap)
4201{
4202 struct vop_rename_args *a = ap;
4203
4204#ifdef DEBUG_VFS_LOCKS
4205 if (a->a_tvp)
4206 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4207 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4208 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4209 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4210
4211 /* Check the source (from). */
4212 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4213 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4214 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4215 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4216 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4217
4218 /* Check the target. */
4219 if (a->a_tvp)
4220 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4221 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4222#endif
4223 if (a->a_tdvp != a->a_fdvp)
4224 vhold(a->a_fdvp);
4225 if (a->a_tvp != a->a_fvp)
4226 vhold(a->a_fvp);
4227 vhold(a->a_tdvp);
4228 if (a->a_tvp)
4229 vhold(a->a_tvp);
4230}
4231
4232void
4233vop_strategy_pre(void *ap)
4234{
4235#ifdef DEBUG_VFS_LOCKS
4236 struct vop_strategy_args *a;
4237 struct buf *bp;
4238
4239 a = ap;
4240 bp = a->a_bp;
4241
4242 /*
4243 * Cluster ops lock their component buffers but not the IO container.
4244 */
4245 if ((bp->b_flags & B_CLUSTER) != 0)
4246 return;
4247
4248 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4249 if (vfs_badlock_print)
4250 printf(
4251 "VOP_STRATEGY: bp is not locked but should be\n");
4252 if (vfs_badlock_ddb)
4253 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4254 }
4255#endif
4256}
4257
4258void
4259vop_lock_pre(void *ap)
4260{
4261#ifdef DEBUG_VFS_LOCKS
4262 struct vop_lock1_args *a = ap;
4263
4264 if ((a->a_flags & LK_INTERLOCK) == 0)
4265 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4266 else
4267 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4268#endif
4269}
4270
4271void
4272vop_lock_post(void *ap, int rc)
4273{
4274#ifdef DEBUG_VFS_LOCKS
4275 struct vop_lock1_args *a = ap;
4276
4277 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4278 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4279 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4280#endif
4281}
4282
4283void
4284vop_unlock_pre(void *ap)
4285{
4286#ifdef DEBUG_VFS_LOCKS
4287 struct vop_unlock_args *a = ap;
4288
4289 if (a->a_flags & LK_INTERLOCK)
4290 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4291 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4292#endif
4293}
4294
4295void
4296vop_unlock_post(void *ap, int rc)
4297{
4298#ifdef DEBUG_VFS_LOCKS
4299 struct vop_unlock_args *a = ap;
4300
4301 if (a->a_flags & LK_INTERLOCK)
4302 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4303#endif
4304}
4305
4306void
4307vop_create_post(void *ap, int rc)
4308{
4309 struct vop_create_args *a = ap;
4310
4311 if (!rc)
4312 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4313}
4314
4315void
4316vop_deleteextattr_post(void *ap, int rc)
4317{
4318 struct vop_deleteextattr_args *a = ap;
4319
4320 if (!rc)
4321 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4322}
4323
4324void
4325vop_link_post(void *ap, int rc)
4326{
4327 struct vop_link_args *a = ap;
4328
4329 if (!rc) {
4330 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4331 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4332 }
4333}
4334
4335void
4336vop_mkdir_post(void *ap, int rc)
4337{
4338 struct vop_mkdir_args *a = ap;
4339
4340 if (!rc)
4341 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4342}
4343
4344void
4345vop_mknod_post(void *ap, int rc)
4346{
4347 struct vop_mknod_args *a = ap;
4348
4349 if (!rc)
4350 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4351}
4352
4353void
4354vop_remove_post(void *ap, int rc)
4355{
4356 struct vop_remove_args *a = ap;
4357
4358 if (!rc) {
4359 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4360 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4361 }
4362}
4363
4364void
4365vop_rename_post(void *ap, int rc)
4366{
4367 struct vop_rename_args *a = ap;
4368
4369 if (!rc) {
4370 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4371 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4372 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4373 if (a->a_tvp)
4374 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4375 }
4376 if (a->a_tdvp != a->a_fdvp)
4377 vdrop(a->a_fdvp);
4378 if (a->a_tvp != a->a_fvp)
4379 vdrop(a->a_fvp);
4380 vdrop(a->a_tdvp);
4381 if (a->a_tvp)
4382 vdrop(a->a_tvp);
4383}
4384
4385void
4386vop_rmdir_post(void *ap, int rc)
4387{
4388 struct vop_rmdir_args *a = ap;
4389
4390 if (!rc) {
4391 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4392 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4393 }
4394}
4395
4396void
4397vop_setattr_post(void *ap, int rc)
4398{
4399 struct vop_setattr_args *a = ap;
4400
4401 if (!rc)
4402 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4403}
4404
4405void
4406vop_setextattr_post(void *ap, int rc)
4407{
4408 struct vop_setextattr_args *a = ap;
4409
4410 if (!rc)
4411 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4412}
4413
4414void
4415vop_symlink_post(void *ap, int rc)
4416{
4417 struct vop_symlink_args *a = ap;
4418
4419 if (!rc)
4420 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4421}
4422
4423static struct knlist fs_knlist;
4424
4425static void
4426vfs_event_init(void *arg)
4427{
4428 knlist_init_mtx(&fs_knlist, NULL);
4429}
4430/* XXX - correct order? */
4431SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4432
4433void
4434vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4435{
4436
4437 KNOTE_UNLOCKED(&fs_knlist, event);
4438}
4439
4440static int filt_fsattach(struct knote *kn);
4441static void filt_fsdetach(struct knote *kn);
4442static int filt_fsevent(struct knote *kn, long hint);
4443
4444struct filterops fs_filtops = {
4445 .f_isfd = 0,
4446 .f_attach = filt_fsattach,
4447 .f_detach = filt_fsdetach,
4448 .f_event = filt_fsevent
4449};
4450
4451static int
4452filt_fsattach(struct knote *kn)
4453{
4454
4455 kn->kn_flags |= EV_CLEAR;
4456 knlist_add(&fs_knlist, kn, 0);
4457 return (0);
4458}
4459
4460static void
4461filt_fsdetach(struct knote *kn)
4462{
4463
4464 knlist_remove(&fs_knlist, kn, 0);
4465}
4466
4467static int
4468filt_fsevent(struct knote *kn, long hint)
4469{
4470
4471 kn->kn_fflags |= hint;
4472 return (kn->kn_fflags != 0);
4473}
4474
4475static int
4476sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4477{
4478 struct vfsidctl vc;
4479 int error;
4480 struct mount *mp;
4481
4482 error = SYSCTL_IN(req, &vc, sizeof(vc));
4483 if (error)
4484 return (error);
4485 if (vc.vc_vers != VFS_CTL_VERS1)
4486 return (EINVAL);
4487 mp = vfs_getvfs(&vc.vc_fsid);
4488 if (mp == NULL)
4489 return (ENOENT);
4490 /* ensure that a specific sysctl goes to the right filesystem. */
4491 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4492 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4493 vfs_rel(mp);
4494 return (EINVAL);
4495 }
4496 VCTLTOREQ(&vc, req);
4497 error = VFS_SYSCTL(mp, vc.vc_op, req);
4498 vfs_rel(mp);
4499 return (error);
4500}
4501
4502SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4503 NULL, 0, sysctl_vfs_ctl, "",
4504 "Sysctl by fsid");
4505
4506/*
4507 * Function to initialize a va_filerev field sensibly.
4508 * XXX: Wouldn't a random number make a lot more sense ??
4509 */
4510u_quad_t
4511init_va_filerev(void)
4512{
4513 struct bintime bt;
4514
4515 getbinuptime(&bt);
4516 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4517}
4518
4519static int filt_vfsread(struct knote *kn, long hint);
4520static int filt_vfswrite(struct knote *kn, long hint);
4521static int filt_vfsvnode(struct knote *kn, long hint);
4522static void filt_vfsdetach(struct knote *kn);
4523static struct filterops vfsread_filtops = {
4524 .f_isfd = 1,
4525 .f_detach = filt_vfsdetach,
4526 .f_event = filt_vfsread
4527};
4528static struct filterops vfswrite_filtops = {
4529 .f_isfd = 1,
4530 .f_detach = filt_vfsdetach,
4531 .f_event = filt_vfswrite
4532};
4533static struct filterops vfsvnode_filtops = {
4534 .f_isfd = 1,
4535 .f_detach = filt_vfsdetach,
4536 .f_event = filt_vfsvnode
4537};
4538
4539static void
4540vfs_knllock(void *arg)
4541{
4542 struct vnode *vp = arg;
4543
4544 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4545}
4546
4547static void
4548vfs_knlunlock(void *arg)
4549{
4550 struct vnode *vp = arg;
4551
4552 VOP_UNLOCK(vp, 0);
4553}
4554
4555static void
4556vfs_knl_assert_locked(void *arg)
4557{
4558#ifdef DEBUG_VFS_LOCKS
4559 struct vnode *vp = arg;
4560
4561 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4562#endif
4563}
4564
4565static void
4566vfs_knl_assert_unlocked(void *arg)
4567{
4568#ifdef DEBUG_VFS_LOCKS
4569 struct vnode *vp = arg;
4570
4571 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4572#endif
4573}
4574
4575int
4576vfs_kqfilter(struct vop_kqfilter_args *ap)
4577{
4578 struct vnode *vp = ap->a_vp;
4579 struct knote *kn = ap->a_kn;
4580 struct knlist *knl;
4581
4582 switch (kn->kn_filter) {
4583 case EVFILT_READ:
4584 kn->kn_fop = &vfsread_filtops;
4585 break;
4586 case EVFILT_WRITE:
4587 kn->kn_fop = &vfswrite_filtops;
4588 break;
4589 case EVFILT_VNODE:
4590 kn->kn_fop = &vfsvnode_filtops;
4591 break;
4592 default:
4593 return (EINVAL);
4594 }
4595
4596 kn->kn_hook = (caddr_t)vp;
4597
4598 v_addpollinfo(vp);
4599 if (vp->v_pollinfo == NULL)
4600 return (ENOMEM);
4601 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4602 vhold(vp);
4603 knlist_add(knl, kn, 0);
4604
4605 return (0);
4606}
4607
4608/*
4609 * Detach knote from vnode
4610 */
4611static void
4612filt_vfsdetach(struct knote *kn)
4613{
4614 struct vnode *vp = (struct vnode *)kn->kn_hook;
4615
4616 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4617 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4618 vdrop(vp);
4619}
4620
4621/*ARGSUSED*/
4622static int
4623filt_vfsread(struct knote *kn, long hint)
4624{
4625 struct vnode *vp = (struct vnode *)kn->kn_hook;
4626 struct vattr va;
4627 int res;
4628
4629 /*
4630 * filesystem is gone, so set the EOF flag and schedule
4631 * the knote for deletion.
4632 */
4633 if (hint == NOTE_REVOKE) {
4634 VI_LOCK(vp);
4635 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4636 VI_UNLOCK(vp);
4637 return (1);
4638 }
4639
4640 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4641 return (0);
4642
4643 VI_LOCK(vp);
4644 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4645 res = (kn->kn_data != 0);
4646 VI_UNLOCK(vp);
4647 return (res);
4648}
4649
4650/*ARGSUSED*/
4651static int
4652filt_vfswrite(struct knote *kn, long hint)
4653{
4654 struct vnode *vp = (struct vnode *)kn->kn_hook;
4655
4656 VI_LOCK(vp);
4657
4658 /*
4659 * filesystem is gone, so set the EOF flag and schedule
4660 * the knote for deletion.
4661 */
4662 if (hint == NOTE_REVOKE)
4663 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4664
4665 kn->kn_data = 0;
4666 VI_UNLOCK(vp);
4667 return (1);
4668}
4669
4670static int
4671filt_vfsvnode(struct knote *kn, long hint)
4672{
4673 struct vnode *vp = (struct vnode *)kn->kn_hook;
4674 int res;
4675
4676 VI_LOCK(vp);
4677 if (kn->kn_sfflags & hint)
4678 kn->kn_fflags |= hint;
4679 if (hint == NOTE_REVOKE) {
4680 kn->kn_flags |= EV_EOF;
4681 VI_UNLOCK(vp);
4682 return (1);
4683 }
4684 res = (kn->kn_fflags != 0);
4685 VI_UNLOCK(vp);
4686 return (res);
4687}
4688
4689int
4690vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4691{
4692 int error;
4693
4694 if (dp->d_reclen > ap->a_uio->uio_resid)
4695 return (ENAMETOOLONG);
4696 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4697 if (error) {
4698 if (ap->a_ncookies != NULL) {
4699 if (ap->a_cookies != NULL)
4700 free(ap->a_cookies, M_TEMP);
4701 ap->a_cookies = NULL;
4702 *ap->a_ncookies = 0;
4703 }
4704 return (error);
4705 }
4706 if (ap->a_ncookies == NULL)
4707 return (0);
4708
4709 KASSERT(ap->a_cookies,
4710 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4711
4712 *ap->a_cookies = realloc(*ap->a_cookies,
4713 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4714 (*ap->a_cookies)[*ap->a_ncookies] = off;
4715 return (0);
4716}
4717
4718/*
4719 * Mark for update the access time of the file if the filesystem
4720 * supports VOP_MARKATIME. This functionality is used by execve and
4721 * mmap, so we want to avoid the I/O implied by directly setting
4722 * va_atime for the sake of efficiency.
4723 */
4724void
4725vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4726{
4727 struct mount *mp;
4728
4729 mp = vp->v_mount;
4730 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4731 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4732 (void)VOP_MARKATIME(vp);
4733}
4734
4735/*
4736 * The purpose of this routine is to remove granularity from accmode_t,
4737 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4738 * VADMIN and VAPPEND.
4739 *
4740 * If it returns 0, the caller is supposed to continue with the usual
4741 * access checks using 'accmode' as modified by this routine. If it
4742 * returns nonzero value, the caller is supposed to return that value
4743 * as errno.
4744 *
4745 * Note that after this routine runs, accmode may be zero.
4746 */
4747int
4748vfs_unixify_accmode(accmode_t *accmode)
4749{
4750 /*
4751 * There is no way to specify explicit "deny" rule using
4752 * file mode or POSIX.1e ACLs.
4753 */
4754 if (*accmode & VEXPLICIT_DENY) {
4755 *accmode = 0;
4756 return (0);
4757 }
4758
4759 /*
4760 * None of these can be translated into usual access bits.
4761 * Also, the common case for NFSv4 ACLs is to not contain
4762 * either of these bits. Caller should check for VWRITE
4763 * on the containing directory instead.
4764 */
4765 if (*accmode & (VDELETE_CHILD | VDELETE))
4766 return (EPERM);
4767
4768 if (*accmode & VADMIN_PERMS) {
4769 *accmode &= ~VADMIN_PERMS;
4770 *accmode |= VADMIN;
4771 }
4772
4773 /*
4774 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4775 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4776 */
4777 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4778
4779 return (0);
4780}
4781
4782/*
4783 * These are helper functions for filesystems to traverse all
4784 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4785 *
4786 * This interface replaces MNT_VNODE_FOREACH.
4787 */
4788
4789MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4790
4791struct vnode *
4792__mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4793{
4794 struct vnode *vp;
4795
4796 if (should_yield())
4797 kern_yield(PRI_USER);
4798 MNT_ILOCK(mp);
4799 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4800 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4801 while (vp != NULL && (vp->v_type == VMARKER ||
4802 (vp->v_iflag & VI_DOOMED) != 0))
4803 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4804
4805 /* Check if we are done */
4806 if (vp == NULL) {
4807 __mnt_vnode_markerfree_all(mvp, mp);
4808 /* MNT_IUNLOCK(mp); -- done in above function */
4809 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4810 return (NULL);
4811 }
4812 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4813 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4814 VI_LOCK(vp);
4815 MNT_IUNLOCK(mp);
4816 return (vp);
4817}
4818
4819struct vnode *
4820__mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4821{
4822 struct vnode *vp;
4823
4824 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4825 MNT_ILOCK(mp);
4826 MNT_REF(mp);
4827 (*mvp)->v_type = VMARKER;
4828
4829 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4830 while (vp != NULL && (vp->v_type == VMARKER ||
4831 (vp->v_iflag & VI_DOOMED) != 0))
4832 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4833
4834 /* Check if we are done */
4835 if (vp == NULL) {
4836 MNT_REL(mp);
4837 MNT_IUNLOCK(mp);
4838 free(*mvp, M_VNODE_MARKER);
4839 *mvp = NULL;
4840 return (NULL);
4841 }
4842 (*mvp)->v_mount = mp;
4843 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4844 VI_LOCK(vp);
4845 MNT_IUNLOCK(mp);
4846 return (vp);
4847}
4848
4849
4850void
4851__mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4852{
4853
4854 if (*mvp == NULL) {
4855 MNT_IUNLOCK(mp);
4856 return;
4857 }
4858
4859 mtx_assert(MNT_MTX(mp), MA_OWNED);
4860
4861 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4862 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4863 MNT_REL(mp);
4864 MNT_IUNLOCK(mp);
4865 free(*mvp, M_VNODE_MARKER);
4866 *mvp = NULL;
4867}
4868
4869/*
4870 * These are helper functions for filesystems to traverse their
4871 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4872 */
4873static void
4874mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4875{
4876
4877 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4878
4879 MNT_ILOCK(mp);
4880 MNT_REL(mp);
4881 MNT_IUNLOCK(mp);
4882 free(*mvp, M_VNODE_MARKER);
4883 *mvp = NULL;
4884}
4885
4886static struct vnode *
4887mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4888{
4889 struct vnode *vp, *nvp;
4890
4891 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4892 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4893restart:
4894 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4895 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4896 while (vp != NULL) {
4897 if (vp->v_type == VMARKER) {
4898 vp = TAILQ_NEXT(vp, v_actfreelist);
4899 continue;
4900 }
4901 if (!VI_TRYLOCK(vp)) {
4902 if (mp_ncpus == 1 || should_yield()) {
4903 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4904 mtx_unlock(&vnode_free_list_mtx);
4905 pause("vnacti", 1);
4906 mtx_lock(&vnode_free_list_mtx);
4907 goto restart;
4908 }
4909 continue;
4910 }
4911 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4912 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4913 ("alien vnode on the active list %p %p", vp, mp));
4914 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4915 break;
4916 nvp = TAILQ_NEXT(vp, v_actfreelist);
4917 VI_UNLOCK(vp);
4918 vp = nvp;
4919 }
4920
4921 /* Check if we are done */
4922 if (vp == NULL) {
4923 mtx_unlock(&vnode_free_list_mtx);
4924 mnt_vnode_markerfree_active(mvp, mp);
4925 return (NULL);
4926 }
4927 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4928 mtx_unlock(&vnode_free_list_mtx);
4929 ASSERT_VI_LOCKED(vp, "active iter");
4930 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4931 return (vp);
4932}
4933
4934struct vnode *
4935__mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4936{
4937
4938 if (should_yield())
4939 kern_yield(PRI_USER);
4940 mtx_lock(&vnode_free_list_mtx);
4941 return (mnt_vnode_next_active(mvp, mp));
4942}
4943
4944struct vnode *
4945__mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4946{
4947 struct vnode *vp;
4948
4949 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4950 MNT_ILOCK(mp);
4951 MNT_REF(mp);
4952 MNT_IUNLOCK(mp);
4953 (*mvp)->v_type = VMARKER;
4954 (*mvp)->v_mount = mp;
4955
4956 mtx_lock(&vnode_free_list_mtx);
4957 vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4958 if (vp == NULL) {
4959 mtx_unlock(&vnode_free_list_mtx);
4960 mnt_vnode_markerfree_active(mvp, mp);
4961 return (NULL);
4962 }
4963 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4964 return (mnt_vnode_next_active(mvp, mp));
4965}
4966
4967void
4968__mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4969{
4970
4971 if (*mvp == NULL)
4972 return;
4973
4974 mtx_lock(&vnode_free_list_mtx);
4975 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4976 mtx_unlock(&vnode_free_list_mtx);
4977 mnt_vnode_markerfree_active(mvp, mp);
4978}
| 2944 BO_UNLOCK(&vp->v_bufobj);
2945
2946 /*
2947 * Reclaim the vnode.
2948 */
2949 if (VOP_RECLAIM(vp, td))
2950 panic("vgone: cannot reclaim");
2951 if (mp != NULL)
2952 vn_finished_secondary_write(mp);
2953 VNASSERT(vp->v_object == NULL, vp,
2954 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2955 /*
2956 * Clear the advisory locks and wake up waiting threads.
2957 */
2958 (void)VOP_ADVLOCKPURGE(vp);
2959 vp->v_lockf = NULL;
2960 /*
2961 * Delete from old mount point vnode list.
2962 */
2963 delmntque(vp);
2964 cache_purge(vp);
2965 /*
2966 * Done with purge, reset to the standard lock and invalidate
2967 * the vnode.
2968 */
2969 VI_LOCK(vp);
2970 vp->v_vnlock = &vp->v_lock;
2971 vp->v_op = &dead_vnodeops;
2972 vp->v_tag = "none";
2973 vp->v_type = VBAD;
2974}
2975
2976/*
2977 * Calculate the total number of references to a special device.
2978 */
2979int
2980vcount(struct vnode *vp)
2981{
2982 int count;
2983
2984 dev_lock();
2985 count = vp->v_rdev->si_usecount;
2986 dev_unlock();
2987 return (count);
2988}
2989
2990/*
2991 * Same as above, but using the struct cdev *as argument
2992 */
2993int
2994count_dev(struct cdev *dev)
2995{
2996 int count;
2997
2998 dev_lock();
2999 count = dev->si_usecount;
3000 dev_unlock();
3001 return(count);
3002}
3003
3004/*
3005 * Print out a description of a vnode.
3006 */
3007static char *typename[] =
3008{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
3009 "VMARKER"};
3010
3011void
3012vn_printf(struct vnode *vp, const char *fmt, ...)
3013{
3014 va_list ap;
3015 char buf[256], buf2[16];
3016 u_long flags;
3017
3018 va_start(ap, fmt);
3019 vprintf(fmt, ap);
3020 va_end(ap);
3021 printf("%p: ", (void *)vp);
3022 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
3023 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
3024 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
3025 buf[0] = '\0';
3026 buf[1] = '\0';
3027 if (vp->v_vflag & VV_ROOT)
3028 strlcat(buf, "|VV_ROOT", sizeof(buf));
3029 if (vp->v_vflag & VV_ISTTY)
3030 strlcat(buf, "|VV_ISTTY", sizeof(buf));
3031 if (vp->v_vflag & VV_NOSYNC)
3032 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
3033 if (vp->v_vflag & VV_ETERNALDEV)
3034 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
3035 if (vp->v_vflag & VV_CACHEDLABEL)
3036 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
3037 if (vp->v_vflag & VV_TEXT)
3038 strlcat(buf, "|VV_TEXT", sizeof(buf));
3039 if (vp->v_vflag & VV_COPYONWRITE)
3040 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
3041 if (vp->v_vflag & VV_SYSTEM)
3042 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
3043 if (vp->v_vflag & VV_PROCDEP)
3044 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
3045 if (vp->v_vflag & VV_NOKNOTE)
3046 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
3047 if (vp->v_vflag & VV_DELETED)
3048 strlcat(buf, "|VV_DELETED", sizeof(buf));
3049 if (vp->v_vflag & VV_MD)
3050 strlcat(buf, "|VV_MD", sizeof(buf));
3051 if (vp->v_vflag & VV_FORCEINSMQ)
3052 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
3053 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
3054 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
3055 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
3056 if (flags != 0) {
3057 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
3058 strlcat(buf, buf2, sizeof(buf));
3059 }
3060 if (vp->v_iflag & VI_MOUNT)
3061 strlcat(buf, "|VI_MOUNT", sizeof(buf));
3062 if (vp->v_iflag & VI_DOOMED)
3063 strlcat(buf, "|VI_DOOMED", sizeof(buf));
3064 if (vp->v_iflag & VI_FREE)
3065 strlcat(buf, "|VI_FREE", sizeof(buf));
3066 if (vp->v_iflag & VI_ACTIVE)
3067 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
3068 if (vp->v_iflag & VI_DOINGINACT)
3069 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
3070 if (vp->v_iflag & VI_OWEINACT)
3071 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
3072 flags = vp->v_iflag & ~(VI_MOUNT | VI_DOOMED | VI_FREE |
3073 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
3074 if (flags != 0) {
3075 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
3076 strlcat(buf, buf2, sizeof(buf));
3077 }
3078 printf(" flags (%s)\n", buf + 1);
3079 if (mtx_owned(VI_MTX(vp)))
3080 printf(" VI_LOCKed");
3081 if (vp->v_object != NULL)
3082 printf(" v_object %p ref %d pages %d "
3083 "cleanbuf %d dirtybuf %d\n",
3084 vp->v_object, vp->v_object->ref_count,
3085 vp->v_object->resident_page_count,
3086 vp->v_bufobj.bo_clean.bv_cnt,
3087 vp->v_bufobj.bo_dirty.bv_cnt);
3088 printf(" ");
3089 lockmgr_printinfo(vp->v_vnlock);
3090 if (vp->v_data != NULL)
3091 VOP_PRINT(vp);
3092}
3093
3094#ifdef DDB
3095/*
3096 * List all of the locked vnodes in the system.
3097 * Called when debugging the kernel.
3098 */
3099DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
3100{
3101 struct mount *mp;
3102 struct vnode *vp;
3103
3104 /*
3105 * Note: because this is DDB, we can't obey the locking semantics
3106 * for these structures, which means we could catch an inconsistent
3107 * state and dereference a nasty pointer. Not much to be done
3108 * about that.
3109 */
3110 db_printf("Locked vnodes\n");
3111 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3112 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3113 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
3114 vprint("", vp);
3115 }
3116 }
3117}
3118
3119/*
3120 * Show details about the given vnode.
3121 */
3122DB_SHOW_COMMAND(vnode, db_show_vnode)
3123{
3124 struct vnode *vp;
3125
3126 if (!have_addr)
3127 return;
3128 vp = (struct vnode *)addr;
3129 vn_printf(vp, "vnode ");
3130}
3131
3132/*
3133 * Show details about the given mount point.
3134 */
3135DB_SHOW_COMMAND(mount, db_show_mount)
3136{
3137 struct mount *mp;
3138 struct vfsopt *opt;
3139 struct statfs *sp;
3140 struct vnode *vp;
3141 char buf[512];
3142 uint64_t mflags;
3143 u_int flags;
3144
3145 if (!have_addr) {
3146 /* No address given, print short info about all mount points. */
3147 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3148 db_printf("%p %s on %s (%s)\n", mp,
3149 mp->mnt_stat.f_mntfromname,
3150 mp->mnt_stat.f_mntonname,
3151 mp->mnt_stat.f_fstypename);
3152 if (db_pager_quit)
3153 break;
3154 }
3155 db_printf("\nMore info: show mount <addr>\n");
3156 return;
3157 }
3158
3159 mp = (struct mount *)addr;
3160 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3161 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3162
3163 buf[0] = '\0';
3164 mflags = mp->mnt_flag;
3165#define MNT_FLAG(flag) do { \
3166 if (mflags & (flag)) { \
3167 if (buf[0] != '\0') \
3168 strlcat(buf, ", ", sizeof(buf)); \
3169 strlcat(buf, (#flag) + 4, sizeof(buf)); \
3170 mflags &= ~(flag); \
3171 } \
3172} while (0)
3173 MNT_FLAG(MNT_RDONLY);
3174 MNT_FLAG(MNT_SYNCHRONOUS);
3175 MNT_FLAG(MNT_NOEXEC);
3176 MNT_FLAG(MNT_NOSUID);
3177 MNT_FLAG(MNT_NFS4ACLS);
3178 MNT_FLAG(MNT_UNION);
3179 MNT_FLAG(MNT_ASYNC);
3180 MNT_FLAG(MNT_SUIDDIR);
3181 MNT_FLAG(MNT_SOFTDEP);
3182 MNT_FLAG(MNT_NOSYMFOLLOW);
3183 MNT_FLAG(MNT_GJOURNAL);
3184 MNT_FLAG(MNT_MULTILABEL);
3185 MNT_FLAG(MNT_ACLS);
3186 MNT_FLAG(MNT_NOATIME);
3187 MNT_FLAG(MNT_NOCLUSTERR);
3188 MNT_FLAG(MNT_NOCLUSTERW);
3189 MNT_FLAG(MNT_SUJ);
3190 MNT_FLAG(MNT_EXRDONLY);
3191 MNT_FLAG(MNT_EXPORTED);
3192 MNT_FLAG(MNT_DEFEXPORTED);
3193 MNT_FLAG(MNT_EXPORTANON);
3194 MNT_FLAG(MNT_EXKERB);
3195 MNT_FLAG(MNT_EXPUBLIC);
3196 MNT_FLAG(MNT_LOCAL);
3197 MNT_FLAG(MNT_QUOTA);
3198 MNT_FLAG(MNT_ROOTFS);
3199 MNT_FLAG(MNT_USER);
3200 MNT_FLAG(MNT_IGNORE);
3201 MNT_FLAG(MNT_UPDATE);
3202 MNT_FLAG(MNT_DELEXPORT);
3203 MNT_FLAG(MNT_RELOAD);
3204 MNT_FLAG(MNT_FORCE);
3205 MNT_FLAG(MNT_SNAPSHOT);
3206 MNT_FLAG(MNT_BYFSID);
3207#undef MNT_FLAG
3208 if (mflags != 0) {
3209 if (buf[0] != '\0')
3210 strlcat(buf, ", ", sizeof(buf));
3211 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3212 "0x%016jx", mflags);
3213 }
3214 db_printf(" mnt_flag = %s\n", buf);
3215
3216 buf[0] = '\0';
3217 flags = mp->mnt_kern_flag;
3218#define MNT_KERN_FLAG(flag) do { \
3219 if (flags & (flag)) { \
3220 if (buf[0] != '\0') \
3221 strlcat(buf, ", ", sizeof(buf)); \
3222 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3223 flags &= ~(flag); \
3224 } \
3225} while (0)
3226 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3227 MNT_KERN_FLAG(MNTK_ASYNC);
3228 MNT_KERN_FLAG(MNTK_SOFTDEP);
3229 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3230 MNT_KERN_FLAG(MNTK_DRAINING);
3231 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3232 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3233 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3234 MNT_KERN_FLAG(MNTK_NO_IOPF);
3235 MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3236 MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3237 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3238 MNT_KERN_FLAG(MNTK_MARKER);
3239 MNT_KERN_FLAG(MNTK_USES_BCACHE);
3240 MNT_KERN_FLAG(MNTK_NOASYNC);
3241 MNT_KERN_FLAG(MNTK_UNMOUNT);
3242 MNT_KERN_FLAG(MNTK_MWAIT);
3243 MNT_KERN_FLAG(MNTK_SUSPEND);
3244 MNT_KERN_FLAG(MNTK_SUSPEND2);
3245 MNT_KERN_FLAG(MNTK_SUSPENDED);
3246 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3247 MNT_KERN_FLAG(MNTK_NOKNOTE);
3248#undef MNT_KERN_FLAG
3249 if (flags != 0) {
3250 if (buf[0] != '\0')
3251 strlcat(buf, ", ", sizeof(buf));
3252 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3253 "0x%08x", flags);
3254 }
3255 db_printf(" mnt_kern_flag = %s\n", buf);
3256
3257 db_printf(" mnt_opt = ");
3258 opt = TAILQ_FIRST(mp->mnt_opt);
3259 if (opt != NULL) {
3260 db_printf("%s", opt->name);
3261 opt = TAILQ_NEXT(opt, link);
3262 while (opt != NULL) {
3263 db_printf(", %s", opt->name);
3264 opt = TAILQ_NEXT(opt, link);
3265 }
3266 }
3267 db_printf("\n");
3268
3269 sp = &mp->mnt_stat;
3270 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3271 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3272 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3273 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3274 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3275 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3276 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3277 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3278 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3279 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3280 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3281 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3282
3283 db_printf(" mnt_cred = { uid=%u ruid=%u",
3284 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3285 if (jailed(mp->mnt_cred))
3286 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3287 db_printf(" }\n");
3288 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3289 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3290 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3291 db_printf(" mnt_activevnodelistsize = %d\n",
3292 mp->mnt_activevnodelistsize);
3293 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3294 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3295 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3296 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3297 db_printf(" mnt_lockref = %d\n", mp->mnt_lockref);
3298 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3299 db_printf(" mnt_secondary_accwrites = %d\n",
3300 mp->mnt_secondary_accwrites);
3301 db_printf(" mnt_gjprovider = %s\n",
3302 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3303
3304 db_printf("\n\nList of active vnodes\n");
3305 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3306 if (vp->v_type != VMARKER) {
3307 vn_printf(vp, "vnode ");
3308 if (db_pager_quit)
3309 break;
3310 }
3311 }
3312 db_printf("\n\nList of inactive vnodes\n");
3313 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3314 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3315 vn_printf(vp, "vnode ");
3316 if (db_pager_quit)
3317 break;
3318 }
3319 }
3320}
3321#endif /* DDB */
3322
3323/*
3324 * Fill in a struct xvfsconf based on a struct vfsconf.
3325 */
3326static int
3327vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3328{
3329 struct xvfsconf xvfsp;
3330
3331 bzero(&xvfsp, sizeof(xvfsp));
3332 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3333 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3334 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3335 xvfsp.vfc_flags = vfsp->vfc_flags;
3336 /*
3337 * These are unused in userland, we keep them
3338 * to not break binary compatibility.
3339 */
3340 xvfsp.vfc_vfsops = NULL;
3341 xvfsp.vfc_next = NULL;
3342 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3343}
3344
3345#ifdef COMPAT_FREEBSD32
3346struct xvfsconf32 {
3347 uint32_t vfc_vfsops;
3348 char vfc_name[MFSNAMELEN];
3349 int32_t vfc_typenum;
3350 int32_t vfc_refcount;
3351 int32_t vfc_flags;
3352 uint32_t vfc_next;
3353};
3354
3355static int
3356vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3357{
3358 struct xvfsconf32 xvfsp;
3359
3360 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3361 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3362 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3363 xvfsp.vfc_flags = vfsp->vfc_flags;
3364 xvfsp.vfc_vfsops = 0;
3365 xvfsp.vfc_next = 0;
3366 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3367}
3368#endif
3369
3370/*
3371 * Top level filesystem related information gathering.
3372 */
3373static int
3374sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3375{
3376 struct vfsconf *vfsp;
3377 int error;
3378
3379 error = 0;
3380 vfsconf_slock();
3381 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3382#ifdef COMPAT_FREEBSD32
3383 if (req->flags & SCTL_MASK32)
3384 error = vfsconf2x32(req, vfsp);
3385 else
3386#endif
3387 error = vfsconf2x(req, vfsp);
3388 if (error)
3389 break;
3390 }
3391 vfsconf_sunlock();
3392 return (error);
3393}
3394
3395SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
3396 CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
3397 "S,xvfsconf", "List of all configured filesystems");
3398
3399#ifndef BURN_BRIDGES
3400static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3401
3402static int
3403vfs_sysctl(SYSCTL_HANDLER_ARGS)
3404{
3405 int *name = (int *)arg1 - 1; /* XXX */
3406 u_int namelen = arg2 + 1; /* XXX */
3407 struct vfsconf *vfsp;
3408
3409 log(LOG_WARNING, "userland calling deprecated sysctl, "
3410 "please rebuild world\n");
3411
3412#if 1 || defined(COMPAT_PRELITE2)
3413 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3414 if (namelen == 1)
3415 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3416#endif
3417
3418 switch (name[1]) {
3419 case VFS_MAXTYPENUM:
3420 if (namelen != 2)
3421 return (ENOTDIR);
3422 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3423 case VFS_CONF:
3424 if (namelen != 3)
3425 return (ENOTDIR); /* overloaded */
3426 vfsconf_slock();
3427 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3428 if (vfsp->vfc_typenum == name[2])
3429 break;
3430 }
3431 vfsconf_sunlock();
3432 if (vfsp == NULL)
3433 return (EOPNOTSUPP);
3434#ifdef COMPAT_FREEBSD32
3435 if (req->flags & SCTL_MASK32)
3436 return (vfsconf2x32(req, vfsp));
3437 else
3438#endif
3439 return (vfsconf2x(req, vfsp));
3440 }
3441 return (EOPNOTSUPP);
3442}
3443
3444static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
3445 CTLFLAG_MPSAFE, vfs_sysctl,
3446 "Generic filesystem");
3447
3448#if 1 || defined(COMPAT_PRELITE2)
3449
3450static int
3451sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3452{
3453 int error;
3454 struct vfsconf *vfsp;
3455 struct ovfsconf ovfs;
3456
3457 vfsconf_slock();
3458 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3459 bzero(&ovfs, sizeof(ovfs));
3460 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3461 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3462 ovfs.vfc_index = vfsp->vfc_typenum;
3463 ovfs.vfc_refcount = vfsp->vfc_refcount;
3464 ovfs.vfc_flags = vfsp->vfc_flags;
3465 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3466 if (error != 0) {
3467 vfsconf_sunlock();
3468 return (error);
3469 }
3470 }
3471 vfsconf_sunlock();
3472 return (0);
3473}
3474
3475#endif /* 1 || COMPAT_PRELITE2 */
3476#endif /* !BURN_BRIDGES */
3477
3478#define KINFO_VNODESLOP 10
3479#ifdef notyet
3480/*
3481 * Dump vnode list (via sysctl).
3482 */
3483/* ARGSUSED */
3484static int
3485sysctl_vnode(SYSCTL_HANDLER_ARGS)
3486{
3487 struct xvnode *xvn;
3488 struct mount *mp;
3489 struct vnode *vp;
3490 int error, len, n;
3491
3492 /*
3493 * Stale numvnodes access is not fatal here.
3494 */
3495 req->lock = 0;
3496 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3497 if (!req->oldptr)
3498 /* Make an estimate */
3499 return (SYSCTL_OUT(req, 0, len));
3500
3501 error = sysctl_wire_old_buffer(req, 0);
3502 if (error != 0)
3503 return (error);
3504 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3505 n = 0;
3506 mtx_lock(&mountlist_mtx);
3507 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3508 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3509 continue;
3510 MNT_ILOCK(mp);
3511 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3512 if (n == len)
3513 break;
3514 vref(vp);
3515 xvn[n].xv_size = sizeof *xvn;
3516 xvn[n].xv_vnode = vp;
3517 xvn[n].xv_id = 0; /* XXX compat */
3518#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3519 XV_COPY(usecount);
3520 XV_COPY(writecount);
3521 XV_COPY(holdcnt);
3522 XV_COPY(mount);
3523 XV_COPY(numoutput);
3524 XV_COPY(type);
3525#undef XV_COPY
3526 xvn[n].xv_flag = vp->v_vflag;
3527
3528 switch (vp->v_type) {
3529 case VREG:
3530 case VDIR:
3531 case VLNK:
3532 break;
3533 case VBLK:
3534 case VCHR:
3535 if (vp->v_rdev == NULL) {
3536 vrele(vp);
3537 continue;
3538 }
3539 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3540 break;
3541 case VSOCK:
3542 xvn[n].xv_socket = vp->v_socket;
3543 break;
3544 case VFIFO:
3545 xvn[n].xv_fifo = vp->v_fifoinfo;
3546 break;
3547 case VNON:
3548 case VBAD:
3549 default:
3550 /* shouldn't happen? */
3551 vrele(vp);
3552 continue;
3553 }
3554 vrele(vp);
3555 ++n;
3556 }
3557 MNT_IUNLOCK(mp);
3558 mtx_lock(&mountlist_mtx);
3559 vfs_unbusy(mp);
3560 if (n == len)
3561 break;
3562 }
3563 mtx_unlock(&mountlist_mtx);
3564
3565 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3566 free(xvn, M_TEMP);
3567 return (error);
3568}
3569
3570SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
3571 CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
3572 "");
3573#endif
3574
3575static void
3576unmount_or_warn(struct mount *mp)
3577{
3578 int error;
3579
3580 error = dounmount(mp, MNT_FORCE, curthread);
3581 if (error != 0 && strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3582 printf("unmount of %s failed (", mp->mnt_stat.f_mntonname);
3583 if (error == EBUSY)
3584 printf("BUSY)\n");
3585 else
3586 printf("%d)\n", error);
3587 }
3588}
3589
3590/*
3591 * Unmount all filesystems. The list is traversed in reverse order
3592 * of mounting to avoid dependencies.
3593 */
3594void
3595vfs_unmountall(void)
3596{
3597 struct mount *mp, *tmp;
3598
3599 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3600
3601 /*
3602 * Since this only runs when rebooting, it is not interlocked.
3603 */
3604 TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) {
3605 vfs_ref(mp);
3606
3607 /*
3608 * Forcibly unmounting "/dev" before "/" would prevent clean
3609 * unmount of the latter.
3610 */
3611 if (mp == rootdevmp)
3612 continue;
3613
3614 unmount_or_warn(mp);
3615 }
3616
3617 if (rootdevmp != NULL)
3618 unmount_or_warn(rootdevmp);
3619}
3620
3621/*
3622 * perform msync on all vnodes under a mount point
3623 * the mount point must be locked.
3624 */
3625void
3626vfs_msync(struct mount *mp, int flags)
3627{
3628 struct vnode *vp, *mvp;
3629 struct vm_object *obj;
3630
3631 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3632 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3633 obj = vp->v_object;
3634 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3635 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3636 if (!vget(vp,
3637 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3638 curthread)) {
3639 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3640 vput(vp);
3641 continue;
3642 }
3643
3644 obj = vp->v_object;
3645 if (obj != NULL) {
3646 VM_OBJECT_WLOCK(obj);
3647 vm_object_page_clean(obj, 0, 0,
3648 flags == MNT_WAIT ?
3649 OBJPC_SYNC : OBJPC_NOSYNC);
3650 VM_OBJECT_WUNLOCK(obj);
3651 }
3652 vput(vp);
3653 }
3654 } else
3655 VI_UNLOCK(vp);
3656 }
3657}
3658
3659static void
3660destroy_vpollinfo_free(struct vpollinfo *vi)
3661{
3662
3663 knlist_destroy(&vi->vpi_selinfo.si_note);
3664 mtx_destroy(&vi->vpi_lock);
3665 uma_zfree(vnodepoll_zone, vi);
3666}
3667
3668static void
3669destroy_vpollinfo(struct vpollinfo *vi)
3670{
3671
3672 knlist_clear(&vi->vpi_selinfo.si_note, 1);
3673 seldrain(&vi->vpi_selinfo);
3674 destroy_vpollinfo_free(vi);
3675}
3676
3677/*
3678 * Initalize per-vnode helper structure to hold poll-related state.
3679 */
3680void
3681v_addpollinfo(struct vnode *vp)
3682{
3683 struct vpollinfo *vi;
3684
3685 if (vp->v_pollinfo != NULL)
3686 return;
3687 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3688 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3689 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3690 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3691 VI_LOCK(vp);
3692 if (vp->v_pollinfo != NULL) {
3693 VI_UNLOCK(vp);
3694 destroy_vpollinfo_free(vi);
3695 return;
3696 }
3697 vp->v_pollinfo = vi;
3698 VI_UNLOCK(vp);
3699}
3700
3701/*
3702 * Record a process's interest in events which might happen to
3703 * a vnode. Because poll uses the historic select-style interface
3704 * internally, this routine serves as both the ``check for any
3705 * pending events'' and the ``record my interest in future events''
3706 * functions. (These are done together, while the lock is held,
3707 * to avoid race conditions.)
3708 */
3709int
3710vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3711{
3712
3713 v_addpollinfo(vp);
3714 mtx_lock(&vp->v_pollinfo->vpi_lock);
3715 if (vp->v_pollinfo->vpi_revents & events) {
3716 /*
3717 * This leaves events we are not interested
3718 * in available for the other process which
3719 * which presumably had requested them
3720 * (otherwise they would never have been
3721 * recorded).
3722 */
3723 events &= vp->v_pollinfo->vpi_revents;
3724 vp->v_pollinfo->vpi_revents &= ~events;
3725
3726 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3727 return (events);
3728 }
3729 vp->v_pollinfo->vpi_events |= events;
3730 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3731 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3732 return (0);
3733}
3734
3735/*
3736 * Routine to create and manage a filesystem syncer vnode.
3737 */
3738#define sync_close ((int (*)(struct vop_close_args *))nullop)
3739static int sync_fsync(struct vop_fsync_args *);
3740static int sync_inactive(struct vop_inactive_args *);
3741static int sync_reclaim(struct vop_reclaim_args *);
3742
3743static struct vop_vector sync_vnodeops = {
3744 .vop_bypass = VOP_EOPNOTSUPP,
3745 .vop_close = sync_close, /* close */
3746 .vop_fsync = sync_fsync, /* fsync */
3747 .vop_inactive = sync_inactive, /* inactive */
3748 .vop_reclaim = sync_reclaim, /* reclaim */
3749 .vop_lock1 = vop_stdlock, /* lock */
3750 .vop_unlock = vop_stdunlock, /* unlock */
3751 .vop_islocked = vop_stdislocked, /* islocked */
3752};
3753
3754/*
3755 * Create a new filesystem syncer vnode for the specified mount point.
3756 */
3757void
3758vfs_allocate_syncvnode(struct mount *mp)
3759{
3760 struct vnode *vp;
3761 struct bufobj *bo;
3762 static long start, incr, next;
3763 int error;
3764
3765 /* Allocate a new vnode */
3766 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3767 if (error != 0)
3768 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3769 vp->v_type = VNON;
3770 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3771 vp->v_vflag |= VV_FORCEINSMQ;
3772 error = insmntque(vp, mp);
3773 if (error != 0)
3774 panic("vfs_allocate_syncvnode: insmntque() failed");
3775 vp->v_vflag &= ~VV_FORCEINSMQ;
3776 VOP_UNLOCK(vp, 0);
3777 /*
3778 * Place the vnode onto the syncer worklist. We attempt to
3779 * scatter them about on the list so that they will go off
3780 * at evenly distributed times even if all the filesystems
3781 * are mounted at once.
3782 */
3783 next += incr;
3784 if (next == 0 || next > syncer_maxdelay) {
3785 start /= 2;
3786 incr /= 2;
3787 if (start == 0) {
3788 start = syncer_maxdelay / 2;
3789 incr = syncer_maxdelay;
3790 }
3791 next = start;
3792 }
3793 bo = &vp->v_bufobj;
3794 BO_LOCK(bo);
3795 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3796 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3797 mtx_lock(&sync_mtx);
3798 sync_vnode_count++;
3799 if (mp->mnt_syncer == NULL) {
3800 mp->mnt_syncer = vp;
3801 vp = NULL;
3802 }
3803 mtx_unlock(&sync_mtx);
3804 BO_UNLOCK(bo);
3805 if (vp != NULL) {
3806 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3807 vgone(vp);
3808 vput(vp);
3809 }
3810}
3811
3812void
3813vfs_deallocate_syncvnode(struct mount *mp)
3814{
3815 struct vnode *vp;
3816
3817 mtx_lock(&sync_mtx);
3818 vp = mp->mnt_syncer;
3819 if (vp != NULL)
3820 mp->mnt_syncer = NULL;
3821 mtx_unlock(&sync_mtx);
3822 if (vp != NULL)
3823 vrele(vp);
3824}
3825
3826/*
3827 * Do a lazy sync of the filesystem.
3828 */
3829static int
3830sync_fsync(struct vop_fsync_args *ap)
3831{
3832 struct vnode *syncvp = ap->a_vp;
3833 struct mount *mp = syncvp->v_mount;
3834 int error, save;
3835 struct bufobj *bo;
3836
3837 /*
3838 * We only need to do something if this is a lazy evaluation.
3839 */
3840 if (ap->a_waitfor != MNT_LAZY)
3841 return (0);
3842
3843 /*
3844 * Move ourselves to the back of the sync list.
3845 */
3846 bo = &syncvp->v_bufobj;
3847 BO_LOCK(bo);
3848 vn_syncer_add_to_worklist(bo, syncdelay);
3849 BO_UNLOCK(bo);
3850
3851 /*
3852 * Walk the list of vnodes pushing all that are dirty and
3853 * not already on the sync list.
3854 */
3855 if (vfs_busy(mp, MBF_NOWAIT) != 0)
3856 return (0);
3857 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3858 vfs_unbusy(mp);
3859 return (0);
3860 }
3861 save = curthread_pflags_set(TDP_SYNCIO);
3862 vfs_msync(mp, MNT_NOWAIT);
3863 error = VFS_SYNC(mp, MNT_LAZY);
3864 curthread_pflags_restore(save);
3865 vn_finished_write(mp);
3866 vfs_unbusy(mp);
3867 return (error);
3868}
3869
3870/*
3871 * The syncer vnode is no referenced.
3872 */
3873static int
3874sync_inactive(struct vop_inactive_args *ap)
3875{
3876
3877 vgone(ap->a_vp);
3878 return (0);
3879}
3880
3881/*
3882 * The syncer vnode is no longer needed and is being decommissioned.
3883 *
3884 * Modifications to the worklist must be protected by sync_mtx.
3885 */
3886static int
3887sync_reclaim(struct vop_reclaim_args *ap)
3888{
3889 struct vnode *vp = ap->a_vp;
3890 struct bufobj *bo;
3891
3892 bo = &vp->v_bufobj;
3893 BO_LOCK(bo);
3894 mtx_lock(&sync_mtx);
3895 if (vp->v_mount->mnt_syncer == vp)
3896 vp->v_mount->mnt_syncer = NULL;
3897 if (bo->bo_flag & BO_ONWORKLST) {
3898 LIST_REMOVE(bo, bo_synclist);
3899 syncer_worklist_len--;
3900 sync_vnode_count--;
3901 bo->bo_flag &= ~BO_ONWORKLST;
3902 }
3903 mtx_unlock(&sync_mtx);
3904 BO_UNLOCK(bo);
3905
3906 return (0);
3907}
3908
3909/*
3910 * Check if vnode represents a disk device
3911 */
3912int
3913vn_isdisk(struct vnode *vp, int *errp)
3914{
3915 int error;
3916
3917 if (vp->v_type != VCHR) {
3918 error = ENOTBLK;
3919 goto out;
3920 }
3921 error = 0;
3922 dev_lock();
3923 if (vp->v_rdev == NULL)
3924 error = ENXIO;
3925 else if (vp->v_rdev->si_devsw == NULL)
3926 error = ENXIO;
3927 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3928 error = ENOTBLK;
3929 dev_unlock();
3930out:
3931 if (errp != NULL)
3932 *errp = error;
3933 return (error == 0);
3934}
3935
3936/*
3937 * Common filesystem object access control check routine. Accepts a
3938 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3939 * and optional call-by-reference privused argument allowing vaccess()
3940 * to indicate to the caller whether privilege was used to satisfy the
3941 * request (obsoleted). Returns 0 on success, or an errno on failure.
3942 */
3943int
3944vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3945 accmode_t accmode, struct ucred *cred, int *privused)
3946{
3947 accmode_t dac_granted;
3948 accmode_t priv_granted;
3949
3950 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3951 ("invalid bit in accmode"));
3952 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3953 ("VAPPEND without VWRITE"));
3954
3955 /*
3956 * Look for a normal, non-privileged way to access the file/directory
3957 * as requested. If it exists, go with that.
3958 */
3959
3960 if (privused != NULL)
3961 *privused = 0;
3962
3963 dac_granted = 0;
3964
3965 /* Check the owner. */
3966 if (cred->cr_uid == file_uid) {
3967 dac_granted |= VADMIN;
3968 if (file_mode & S_IXUSR)
3969 dac_granted |= VEXEC;
3970 if (file_mode & S_IRUSR)
3971 dac_granted |= VREAD;
3972 if (file_mode & S_IWUSR)
3973 dac_granted |= (VWRITE | VAPPEND);
3974
3975 if ((accmode & dac_granted) == accmode)
3976 return (0);
3977
3978 goto privcheck;
3979 }
3980
3981 /* Otherwise, check the groups (first match) */
3982 if (groupmember(file_gid, cred)) {
3983 if (file_mode & S_IXGRP)
3984 dac_granted |= VEXEC;
3985 if (file_mode & S_IRGRP)
3986 dac_granted |= VREAD;
3987 if (file_mode & S_IWGRP)
3988 dac_granted |= (VWRITE | VAPPEND);
3989
3990 if ((accmode & dac_granted) == accmode)
3991 return (0);
3992
3993 goto privcheck;
3994 }
3995
3996 /* Otherwise, check everyone else. */
3997 if (file_mode & S_IXOTH)
3998 dac_granted |= VEXEC;
3999 if (file_mode & S_IROTH)
4000 dac_granted |= VREAD;
4001 if (file_mode & S_IWOTH)
4002 dac_granted |= (VWRITE | VAPPEND);
4003 if ((accmode & dac_granted) == accmode)
4004 return (0);
4005
4006privcheck:
4007 /*
4008 * Build a privilege mask to determine if the set of privileges
4009 * satisfies the requirements when combined with the granted mask
4010 * from above. For each privilege, if the privilege is required,
4011 * bitwise or the request type onto the priv_granted mask.
4012 */
4013 priv_granted = 0;
4014
4015 if (type == VDIR) {
4016 /*
4017 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
4018 * requests, instead of PRIV_VFS_EXEC.
4019 */
4020 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4021 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
4022 priv_granted |= VEXEC;
4023 } else {
4024 /*
4025 * Ensure that at least one execute bit is on. Otherwise,
4026 * a privileged user will always succeed, and we don't want
4027 * this to happen unless the file really is executable.
4028 */
4029 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4030 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
4031 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
4032 priv_granted |= VEXEC;
4033 }
4034
4035 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
4036 !priv_check_cred(cred, PRIV_VFS_READ, 0))
4037 priv_granted |= VREAD;
4038
4039 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
4040 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
4041 priv_granted |= (VWRITE | VAPPEND);
4042
4043 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
4044 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
4045 priv_granted |= VADMIN;
4046
4047 if ((accmode & (priv_granted | dac_granted)) == accmode) {
4048 /* XXX audit: privilege used */
4049 if (privused != NULL)
4050 *privused = 1;
4051 return (0);
4052 }
4053
4054 return ((accmode & VADMIN) ? EPERM : EACCES);
4055}
4056
4057/*
4058 * Credential check based on process requesting service, and per-attribute
4059 * permissions.
4060 */
4061int
4062extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
4063 struct thread *td, accmode_t accmode)
4064{
4065
4066 /*
4067 * Kernel-invoked always succeeds.
4068 */
4069 if (cred == NOCRED)
4070 return (0);
4071
4072 /*
4073 * Do not allow privileged processes in jail to directly manipulate
4074 * system attributes.
4075 */
4076 switch (attrnamespace) {
4077 case EXTATTR_NAMESPACE_SYSTEM:
4078 /* Potentially should be: return (EPERM); */
4079 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
4080 case EXTATTR_NAMESPACE_USER:
4081 return (VOP_ACCESS(vp, accmode, cred, td));
4082 default:
4083 return (EPERM);
4084 }
4085}
4086
4087#ifdef DEBUG_VFS_LOCKS
4088/*
4089 * This only exists to supress warnings from unlocked specfs accesses. It is
4090 * no longer ok to have an unlocked VFS.
4091 */
4092#define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
4093 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
4094
4095int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
4096SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
4097 "Drop into debugger on lock violation");
4098
4099int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
4100SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
4101 0, "Check for interlock across VOPs");
4102
4103int vfs_badlock_print = 1; /* Print lock violations. */
4104SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
4105 0, "Print lock violations");
4106
4107#ifdef KDB
4108int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
4109SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
4110 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
4111#endif
4112
4113static void
4114vfs_badlock(const char *msg, const char *str, struct vnode *vp)
4115{
4116
4117#ifdef KDB
4118 if (vfs_badlock_backtrace)
4119 kdb_backtrace();
4120#endif
4121 if (vfs_badlock_print)
4122 printf("%s: %p %s\n", str, (void *)vp, msg);
4123 if (vfs_badlock_ddb)
4124 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4125}
4126
4127void
4128assert_vi_locked(struct vnode *vp, const char *str)
4129{
4130
4131 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4132 vfs_badlock("interlock is not locked but should be", str, vp);
4133}
4134
4135void
4136assert_vi_unlocked(struct vnode *vp, const char *str)
4137{
4138
4139 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4140 vfs_badlock("interlock is locked but should not be", str, vp);
4141}
4142
4143void
4144assert_vop_locked(struct vnode *vp, const char *str)
4145{
4146 int locked;
4147
4148 if (!IGNORE_LOCK(vp)) {
4149 locked = VOP_ISLOCKED(vp);
4150 if (locked == 0 || locked == LK_EXCLOTHER)
4151 vfs_badlock("is not locked but should be", str, vp);
4152 }
4153}
4154
4155void
4156assert_vop_unlocked(struct vnode *vp, const char *str)
4157{
4158
4159 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4160 vfs_badlock("is locked but should not be", str, vp);
4161}
4162
4163void
4164assert_vop_elocked(struct vnode *vp, const char *str)
4165{
4166
4167 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4168 vfs_badlock("is not exclusive locked but should be", str, vp);
4169}
4170
4171#if 0
4172void
4173assert_vop_elocked_other(struct vnode *vp, const char *str)
4174{
4175
4176 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4177 vfs_badlock("is not exclusive locked by another thread",
4178 str, vp);
4179}
4180
4181void
4182assert_vop_slocked(struct vnode *vp, const char *str)
4183{
4184
4185 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4186 vfs_badlock("is not locked shared but should be", str, vp);
4187}
4188#endif /* 0 */
4189#endif /* DEBUG_VFS_LOCKS */
4190
4191void
4192vop_rename_fail(struct vop_rename_args *ap)
4193{
4194
4195 if (ap->a_tvp != NULL)
4196 vput(ap->a_tvp);
4197 if (ap->a_tdvp == ap->a_tvp)
4198 vrele(ap->a_tdvp);
4199 else
4200 vput(ap->a_tdvp);
4201 vrele(ap->a_fdvp);
4202 vrele(ap->a_fvp);
4203}
4204
4205void
4206vop_rename_pre(void *ap)
4207{
4208 struct vop_rename_args *a = ap;
4209
4210#ifdef DEBUG_VFS_LOCKS
4211 if (a->a_tvp)
4212 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4213 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4214 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4215 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4216
4217 /* Check the source (from). */
4218 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4219 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4220 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4221 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4222 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4223
4224 /* Check the target. */
4225 if (a->a_tvp)
4226 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4227 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4228#endif
4229 if (a->a_tdvp != a->a_fdvp)
4230 vhold(a->a_fdvp);
4231 if (a->a_tvp != a->a_fvp)
4232 vhold(a->a_fvp);
4233 vhold(a->a_tdvp);
4234 if (a->a_tvp)
4235 vhold(a->a_tvp);
4236}
4237
4238void
4239vop_strategy_pre(void *ap)
4240{
4241#ifdef DEBUG_VFS_LOCKS
4242 struct vop_strategy_args *a;
4243 struct buf *bp;
4244
4245 a = ap;
4246 bp = a->a_bp;
4247
4248 /*
4249 * Cluster ops lock their component buffers but not the IO container.
4250 */
4251 if ((bp->b_flags & B_CLUSTER) != 0)
4252 return;
4253
4254 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4255 if (vfs_badlock_print)
4256 printf(
4257 "VOP_STRATEGY: bp is not locked but should be\n");
4258 if (vfs_badlock_ddb)
4259 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4260 }
4261#endif
4262}
4263
4264void
4265vop_lock_pre(void *ap)
4266{
4267#ifdef DEBUG_VFS_LOCKS
4268 struct vop_lock1_args *a = ap;
4269
4270 if ((a->a_flags & LK_INTERLOCK) == 0)
4271 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4272 else
4273 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4274#endif
4275}
4276
4277void
4278vop_lock_post(void *ap, int rc)
4279{
4280#ifdef DEBUG_VFS_LOCKS
4281 struct vop_lock1_args *a = ap;
4282
4283 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4284 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4285 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4286#endif
4287}
4288
4289void
4290vop_unlock_pre(void *ap)
4291{
4292#ifdef DEBUG_VFS_LOCKS
4293 struct vop_unlock_args *a = ap;
4294
4295 if (a->a_flags & LK_INTERLOCK)
4296 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4297 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4298#endif
4299}
4300
4301void
4302vop_unlock_post(void *ap, int rc)
4303{
4304#ifdef DEBUG_VFS_LOCKS
4305 struct vop_unlock_args *a = ap;
4306
4307 if (a->a_flags & LK_INTERLOCK)
4308 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4309#endif
4310}
4311
4312void
4313vop_create_post(void *ap, int rc)
4314{
4315 struct vop_create_args *a = ap;
4316
4317 if (!rc)
4318 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4319}
4320
4321void
4322vop_deleteextattr_post(void *ap, int rc)
4323{
4324 struct vop_deleteextattr_args *a = ap;
4325
4326 if (!rc)
4327 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4328}
4329
4330void
4331vop_link_post(void *ap, int rc)
4332{
4333 struct vop_link_args *a = ap;
4334
4335 if (!rc) {
4336 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4337 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4338 }
4339}
4340
4341void
4342vop_mkdir_post(void *ap, int rc)
4343{
4344 struct vop_mkdir_args *a = ap;
4345
4346 if (!rc)
4347 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4348}
4349
4350void
4351vop_mknod_post(void *ap, int rc)
4352{
4353 struct vop_mknod_args *a = ap;
4354
4355 if (!rc)
4356 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4357}
4358
4359void
4360vop_remove_post(void *ap, int rc)
4361{
4362 struct vop_remove_args *a = ap;
4363
4364 if (!rc) {
4365 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4366 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4367 }
4368}
4369
4370void
4371vop_rename_post(void *ap, int rc)
4372{
4373 struct vop_rename_args *a = ap;
4374
4375 if (!rc) {
4376 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4377 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4378 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4379 if (a->a_tvp)
4380 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4381 }
4382 if (a->a_tdvp != a->a_fdvp)
4383 vdrop(a->a_fdvp);
4384 if (a->a_tvp != a->a_fvp)
4385 vdrop(a->a_fvp);
4386 vdrop(a->a_tdvp);
4387 if (a->a_tvp)
4388 vdrop(a->a_tvp);
4389}
4390
4391void
4392vop_rmdir_post(void *ap, int rc)
4393{
4394 struct vop_rmdir_args *a = ap;
4395
4396 if (!rc) {
4397 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4398 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4399 }
4400}
4401
4402void
4403vop_setattr_post(void *ap, int rc)
4404{
4405 struct vop_setattr_args *a = ap;
4406
4407 if (!rc)
4408 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4409}
4410
4411void
4412vop_setextattr_post(void *ap, int rc)
4413{
4414 struct vop_setextattr_args *a = ap;
4415
4416 if (!rc)
4417 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4418}
4419
4420void
4421vop_symlink_post(void *ap, int rc)
4422{
4423 struct vop_symlink_args *a = ap;
4424
4425 if (!rc)
4426 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4427}
4428
4429static struct knlist fs_knlist;
4430
4431static void
4432vfs_event_init(void *arg)
4433{
4434 knlist_init_mtx(&fs_knlist, NULL);
4435}
4436/* XXX - correct order? */
4437SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4438
4439void
4440vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4441{
4442
4443 KNOTE_UNLOCKED(&fs_knlist, event);
4444}
4445
4446static int filt_fsattach(struct knote *kn);
4447static void filt_fsdetach(struct knote *kn);
4448static int filt_fsevent(struct knote *kn, long hint);
4449
4450struct filterops fs_filtops = {
4451 .f_isfd = 0,
4452 .f_attach = filt_fsattach,
4453 .f_detach = filt_fsdetach,
4454 .f_event = filt_fsevent
4455};
4456
4457static int
4458filt_fsattach(struct knote *kn)
4459{
4460
4461 kn->kn_flags |= EV_CLEAR;
4462 knlist_add(&fs_knlist, kn, 0);
4463 return (0);
4464}
4465
4466static void
4467filt_fsdetach(struct knote *kn)
4468{
4469
4470 knlist_remove(&fs_knlist, kn, 0);
4471}
4472
4473static int
4474filt_fsevent(struct knote *kn, long hint)
4475{
4476
4477 kn->kn_fflags |= hint;
4478 return (kn->kn_fflags != 0);
4479}
4480
4481static int
4482sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4483{
4484 struct vfsidctl vc;
4485 int error;
4486 struct mount *mp;
4487
4488 error = SYSCTL_IN(req, &vc, sizeof(vc));
4489 if (error)
4490 return (error);
4491 if (vc.vc_vers != VFS_CTL_VERS1)
4492 return (EINVAL);
4493 mp = vfs_getvfs(&vc.vc_fsid);
4494 if (mp == NULL)
4495 return (ENOENT);
4496 /* ensure that a specific sysctl goes to the right filesystem. */
4497 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4498 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4499 vfs_rel(mp);
4500 return (EINVAL);
4501 }
4502 VCTLTOREQ(&vc, req);
4503 error = VFS_SYSCTL(mp, vc.vc_op, req);
4504 vfs_rel(mp);
4505 return (error);
4506}
4507
4508SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4509 NULL, 0, sysctl_vfs_ctl, "",
4510 "Sysctl by fsid");
4511
4512/*
4513 * Function to initialize a va_filerev field sensibly.
4514 * XXX: Wouldn't a random number make a lot more sense ??
4515 */
4516u_quad_t
4517init_va_filerev(void)
4518{
4519 struct bintime bt;
4520
4521 getbinuptime(&bt);
4522 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4523}
4524
4525static int filt_vfsread(struct knote *kn, long hint);
4526static int filt_vfswrite(struct knote *kn, long hint);
4527static int filt_vfsvnode(struct knote *kn, long hint);
4528static void filt_vfsdetach(struct knote *kn);
4529static struct filterops vfsread_filtops = {
4530 .f_isfd = 1,
4531 .f_detach = filt_vfsdetach,
4532 .f_event = filt_vfsread
4533};
4534static struct filterops vfswrite_filtops = {
4535 .f_isfd = 1,
4536 .f_detach = filt_vfsdetach,
4537 .f_event = filt_vfswrite
4538};
4539static struct filterops vfsvnode_filtops = {
4540 .f_isfd = 1,
4541 .f_detach = filt_vfsdetach,
4542 .f_event = filt_vfsvnode
4543};
4544
4545static void
4546vfs_knllock(void *arg)
4547{
4548 struct vnode *vp = arg;
4549
4550 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4551}
4552
4553static void
4554vfs_knlunlock(void *arg)
4555{
4556 struct vnode *vp = arg;
4557
4558 VOP_UNLOCK(vp, 0);
4559}
4560
4561static void
4562vfs_knl_assert_locked(void *arg)
4563{
4564#ifdef DEBUG_VFS_LOCKS
4565 struct vnode *vp = arg;
4566
4567 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4568#endif
4569}
4570
4571static void
4572vfs_knl_assert_unlocked(void *arg)
4573{
4574#ifdef DEBUG_VFS_LOCKS
4575 struct vnode *vp = arg;
4576
4577 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4578#endif
4579}
4580
4581int
4582vfs_kqfilter(struct vop_kqfilter_args *ap)
4583{
4584 struct vnode *vp = ap->a_vp;
4585 struct knote *kn = ap->a_kn;
4586 struct knlist *knl;
4587
4588 switch (kn->kn_filter) {
4589 case EVFILT_READ:
4590 kn->kn_fop = &vfsread_filtops;
4591 break;
4592 case EVFILT_WRITE:
4593 kn->kn_fop = &vfswrite_filtops;
4594 break;
4595 case EVFILT_VNODE:
4596 kn->kn_fop = &vfsvnode_filtops;
4597 break;
4598 default:
4599 return (EINVAL);
4600 }
4601
4602 kn->kn_hook = (caddr_t)vp;
4603
4604 v_addpollinfo(vp);
4605 if (vp->v_pollinfo == NULL)
4606 return (ENOMEM);
4607 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4608 vhold(vp);
4609 knlist_add(knl, kn, 0);
4610
4611 return (0);
4612}
4613
4614/*
4615 * Detach knote from vnode
4616 */
4617static void
4618filt_vfsdetach(struct knote *kn)
4619{
4620 struct vnode *vp = (struct vnode *)kn->kn_hook;
4621
4622 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4623 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4624 vdrop(vp);
4625}
4626
4627/*ARGSUSED*/
4628static int
4629filt_vfsread(struct knote *kn, long hint)
4630{
4631 struct vnode *vp = (struct vnode *)kn->kn_hook;
4632 struct vattr va;
4633 int res;
4634
4635 /*
4636 * filesystem is gone, so set the EOF flag and schedule
4637 * the knote for deletion.
4638 */
4639 if (hint == NOTE_REVOKE) {
4640 VI_LOCK(vp);
4641 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4642 VI_UNLOCK(vp);
4643 return (1);
4644 }
4645
4646 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4647 return (0);
4648
4649 VI_LOCK(vp);
4650 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4651 res = (kn->kn_data != 0);
4652 VI_UNLOCK(vp);
4653 return (res);
4654}
4655
4656/*ARGSUSED*/
4657static int
4658filt_vfswrite(struct knote *kn, long hint)
4659{
4660 struct vnode *vp = (struct vnode *)kn->kn_hook;
4661
4662 VI_LOCK(vp);
4663
4664 /*
4665 * filesystem is gone, so set the EOF flag and schedule
4666 * the knote for deletion.
4667 */
4668 if (hint == NOTE_REVOKE)
4669 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4670
4671 kn->kn_data = 0;
4672 VI_UNLOCK(vp);
4673 return (1);
4674}
4675
4676static int
4677filt_vfsvnode(struct knote *kn, long hint)
4678{
4679 struct vnode *vp = (struct vnode *)kn->kn_hook;
4680 int res;
4681
4682 VI_LOCK(vp);
4683 if (kn->kn_sfflags & hint)
4684 kn->kn_fflags |= hint;
4685 if (hint == NOTE_REVOKE) {
4686 kn->kn_flags |= EV_EOF;
4687 VI_UNLOCK(vp);
4688 return (1);
4689 }
4690 res = (kn->kn_fflags != 0);
4691 VI_UNLOCK(vp);
4692 return (res);
4693}
4694
4695int
4696vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4697{
4698 int error;
4699
4700 if (dp->d_reclen > ap->a_uio->uio_resid)
4701 return (ENAMETOOLONG);
4702 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4703 if (error) {
4704 if (ap->a_ncookies != NULL) {
4705 if (ap->a_cookies != NULL)
4706 free(ap->a_cookies, M_TEMP);
4707 ap->a_cookies = NULL;
4708 *ap->a_ncookies = 0;
4709 }
4710 return (error);
4711 }
4712 if (ap->a_ncookies == NULL)
4713 return (0);
4714
4715 KASSERT(ap->a_cookies,
4716 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4717
4718 *ap->a_cookies = realloc(*ap->a_cookies,
4719 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4720 (*ap->a_cookies)[*ap->a_ncookies] = off;
4721 return (0);
4722}
4723
4724/*
4725 * Mark for update the access time of the file if the filesystem
4726 * supports VOP_MARKATIME. This functionality is used by execve and
4727 * mmap, so we want to avoid the I/O implied by directly setting
4728 * va_atime for the sake of efficiency.
4729 */
4730void
4731vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4732{
4733 struct mount *mp;
4734
4735 mp = vp->v_mount;
4736 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4737 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4738 (void)VOP_MARKATIME(vp);
4739}
4740
4741/*
4742 * The purpose of this routine is to remove granularity from accmode_t,
4743 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4744 * VADMIN and VAPPEND.
4745 *
4746 * If it returns 0, the caller is supposed to continue with the usual
4747 * access checks using 'accmode' as modified by this routine. If it
4748 * returns nonzero value, the caller is supposed to return that value
4749 * as errno.
4750 *
4751 * Note that after this routine runs, accmode may be zero.
4752 */
4753int
4754vfs_unixify_accmode(accmode_t *accmode)
4755{
4756 /*
4757 * There is no way to specify explicit "deny" rule using
4758 * file mode or POSIX.1e ACLs.
4759 */
4760 if (*accmode & VEXPLICIT_DENY) {
4761 *accmode = 0;
4762 return (0);
4763 }
4764
4765 /*
4766 * None of these can be translated into usual access bits.
4767 * Also, the common case for NFSv4 ACLs is to not contain
4768 * either of these bits. Caller should check for VWRITE
4769 * on the containing directory instead.
4770 */
4771 if (*accmode & (VDELETE_CHILD | VDELETE))
4772 return (EPERM);
4773
4774 if (*accmode & VADMIN_PERMS) {
4775 *accmode &= ~VADMIN_PERMS;
4776 *accmode |= VADMIN;
4777 }
4778
4779 /*
4780 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4781 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4782 */
4783 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4784
4785 return (0);
4786}
4787
4788/*
4789 * These are helper functions for filesystems to traverse all
4790 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4791 *
4792 * This interface replaces MNT_VNODE_FOREACH.
4793 */
4794
4795MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4796
4797struct vnode *
4798__mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4799{
4800 struct vnode *vp;
4801
4802 if (should_yield())
4803 kern_yield(PRI_USER);
4804 MNT_ILOCK(mp);
4805 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4806 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4807 while (vp != NULL && (vp->v_type == VMARKER ||
4808 (vp->v_iflag & VI_DOOMED) != 0))
4809 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4810
4811 /* Check if we are done */
4812 if (vp == NULL) {
4813 __mnt_vnode_markerfree_all(mvp, mp);
4814 /* MNT_IUNLOCK(mp); -- done in above function */
4815 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4816 return (NULL);
4817 }
4818 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4819 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4820 VI_LOCK(vp);
4821 MNT_IUNLOCK(mp);
4822 return (vp);
4823}
4824
4825struct vnode *
4826__mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4827{
4828 struct vnode *vp;
4829
4830 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4831 MNT_ILOCK(mp);
4832 MNT_REF(mp);
4833 (*mvp)->v_type = VMARKER;
4834
4835 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4836 while (vp != NULL && (vp->v_type == VMARKER ||
4837 (vp->v_iflag & VI_DOOMED) != 0))
4838 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4839
4840 /* Check if we are done */
4841 if (vp == NULL) {
4842 MNT_REL(mp);
4843 MNT_IUNLOCK(mp);
4844 free(*mvp, M_VNODE_MARKER);
4845 *mvp = NULL;
4846 return (NULL);
4847 }
4848 (*mvp)->v_mount = mp;
4849 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4850 VI_LOCK(vp);
4851 MNT_IUNLOCK(mp);
4852 return (vp);
4853}
4854
4855
4856void
4857__mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4858{
4859
4860 if (*mvp == NULL) {
4861 MNT_IUNLOCK(mp);
4862 return;
4863 }
4864
4865 mtx_assert(MNT_MTX(mp), MA_OWNED);
4866
4867 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4868 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4869 MNT_REL(mp);
4870 MNT_IUNLOCK(mp);
4871 free(*mvp, M_VNODE_MARKER);
4872 *mvp = NULL;
4873}
4874
4875/*
4876 * These are helper functions for filesystems to traverse their
4877 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4878 */
4879static void
4880mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4881{
4882
4883 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4884
4885 MNT_ILOCK(mp);
4886 MNT_REL(mp);
4887 MNT_IUNLOCK(mp);
4888 free(*mvp, M_VNODE_MARKER);
4889 *mvp = NULL;
4890}
4891
4892static struct vnode *
4893mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4894{
4895 struct vnode *vp, *nvp;
4896
4897 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4898 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4899restart:
4900 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4901 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4902 while (vp != NULL) {
4903 if (vp->v_type == VMARKER) {
4904 vp = TAILQ_NEXT(vp, v_actfreelist);
4905 continue;
4906 }
4907 if (!VI_TRYLOCK(vp)) {
4908 if (mp_ncpus == 1 || should_yield()) {
4909 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4910 mtx_unlock(&vnode_free_list_mtx);
4911 pause("vnacti", 1);
4912 mtx_lock(&vnode_free_list_mtx);
4913 goto restart;
4914 }
4915 continue;
4916 }
4917 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4918 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4919 ("alien vnode on the active list %p %p", vp, mp));
4920 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4921 break;
4922 nvp = TAILQ_NEXT(vp, v_actfreelist);
4923 VI_UNLOCK(vp);
4924 vp = nvp;
4925 }
4926
4927 /* Check if we are done */
4928 if (vp == NULL) {
4929 mtx_unlock(&vnode_free_list_mtx);
4930 mnt_vnode_markerfree_active(mvp, mp);
4931 return (NULL);
4932 }
4933 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4934 mtx_unlock(&vnode_free_list_mtx);
4935 ASSERT_VI_LOCKED(vp, "active iter");
4936 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4937 return (vp);
4938}
4939
4940struct vnode *
4941__mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4942{
4943
4944 if (should_yield())
4945 kern_yield(PRI_USER);
4946 mtx_lock(&vnode_free_list_mtx);
4947 return (mnt_vnode_next_active(mvp, mp));
4948}
4949
4950struct vnode *
4951__mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4952{
4953 struct vnode *vp;
4954
4955 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4956 MNT_ILOCK(mp);
4957 MNT_REF(mp);
4958 MNT_IUNLOCK(mp);
4959 (*mvp)->v_type = VMARKER;
4960 (*mvp)->v_mount = mp;
4961
4962 mtx_lock(&vnode_free_list_mtx);
4963 vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4964 if (vp == NULL) {
4965 mtx_unlock(&vnode_free_list_mtx);
4966 mnt_vnode_markerfree_active(mvp, mp);
4967 return (NULL);
4968 }
4969 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4970 return (mnt_vnode_next_active(mvp, mp));
4971}
4972
4973void
4974__mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4975{
4976
4977 if (*mvp == NULL)
4978 return;
4979
4980 mtx_lock(&vnode_free_list_mtx);
4981 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4982 mtx_unlock(&vnode_free_list_mtx);
4983 mnt_vnode_markerfree_active(mvp, mp);
4984}
|