1/*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
35 */
36
37/*
38 * External virtual filesystem routines
39 */
40
41#include <sys/cdefs.h>
42__FBSDID("$FreeBSD: head/sys/kern/vfs_subr.c 250551 2013-05-12 04:05:01Z jeff $");
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
125/*
126 * Conversion tables for conversion from vnode types to inode formats
127 * and back.
128 */
129enum vtype iftovt_tab[16] = {
130 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
131 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
132};
133int vttoif_tab[10] = {
134 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
135 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
136};
137
138/*
139 * List of vnodes that are ready for recycling.
140 */
141static TAILQ_HEAD(freelst, vnode) vnode_free_list;
142
143/*
144 * Free vnode target. Free vnodes may simply be files which have been stat'd
145 * but not read. This is somewhat common, and a small cache of such files
146 * should be kept to avoid recreation costs.
147 */
148static u_long wantfreevnodes;
149SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
150/* Number of vnodes in the free list. */
151static u_long freevnodes;
152SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
153 "Number of vnodes in the free list");
154
155static int vlru_allow_cache_src;
156SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
157 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
158
159/*
160 * Various variables used for debugging the new implementation of
161 * reassignbuf().
162 * XXX these are probably of (very) limited utility now.
163 */
164static int reassignbufcalls;
165SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
166 "Number of calls to reassignbuf");
167
168/*
169 * Cache for the mount type id assigned to NFS. This is used for
170 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
171 */
172int nfs_mount_type = -1;
173
174/* To keep more than one thread at a time from running vfs_getnewfsid */
175static struct mtx mntid_mtx;
176
177/*
178 * Lock for any access to the following:
179 * vnode_free_list
180 * numvnodes
181 * freevnodes
182 */
183static struct mtx vnode_free_list_mtx;
184
185/* Publicly exported FS */
186struct nfs_public nfs_pub;
187
188static uma_zone_t buf_trie_zone;
189
190/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
191static uma_zone_t vnode_zone;
192static uma_zone_t vnodepoll_zone;
193
194/*
195 * The workitem queue.
196 *
197 * It is useful to delay writes of file data and filesystem metadata
198 * for tens of seconds so that quickly created and deleted files need
199 * not waste disk bandwidth being created and removed. To realize this,
200 * we append vnodes to a "workitem" queue. When running with a soft
201 * updates implementation, most pending metadata dependencies should
202 * not wait for more than a few seconds. Thus, mounted on block devices
203 * are delayed only about a half the time that file data is delayed.
204 * Similarly, directory updates are more critical, so are only delayed
205 * about a third the time that file data is delayed. Thus, there are
206 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207 * one each second (driven off the filesystem syncer process). The
208 * syncer_delayno variable indicates the next queue that is to be processed.
209 * Items that need to be processed soon are placed in this queue:
210 *
211 * syncer_workitem_pending[syncer_delayno]
212 *
213 * A delay of fifteen seconds is done by placing the request fifteen
214 * entries later in the queue:
215 *
216 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
217 *
218 */
219static int syncer_delayno;
220static long syncer_mask;
221LIST_HEAD(synclist, bufobj);
222static struct synclist *syncer_workitem_pending;
223/*
224 * The sync_mtx protects:
225 * bo->bo_synclist
226 * sync_vnode_count
227 * syncer_delayno
228 * syncer_state
229 * syncer_workitem_pending
230 * syncer_worklist_len
231 * rushjob
232 */
233static struct mtx sync_mtx;
234static struct cv sync_wakeup;
235
236#define SYNCER_MAXDELAY 32
237static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
238static int syncdelay = 30; /* max time to delay syncing data */
239static int filedelay = 30; /* time to delay syncing files */
240SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
241 "Time to delay syncing files (in seconds)");
242static int dirdelay = 29; /* time to delay syncing directories */
243SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
244 "Time to delay syncing directories (in seconds)");
245static int metadelay = 28; /* time to delay syncing metadata */
246SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
247 "Time to delay syncing metadata (in seconds)");
248static int rushjob; /* number of slots to run ASAP */
249static int stat_rush_requests; /* number of times I/O speeded up */
250SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
251 "Number of times I/O speeded up (rush requests)");
252
253/*
254 * When shutting down the syncer, run it at four times normal speed.
255 */
256#define SYNCER_SHUTDOWN_SPEEDUP 4
257static int sync_vnode_count;
258static int syncer_worklist_len;
259static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
260 syncer_state;
261
262/*
263 * Number of vnodes we want to exist at any one time. This is mostly used
264 * to size hash tables in vnode-related code. It is normally not used in
265 * getnewvnode(), as wantfreevnodes is normally nonzero.)
266 *
267 * XXX desiredvnodes is historical cruft and should not exist.
268 */
269int desiredvnodes;
270SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
271 &desiredvnodes, 0, "Maximum number of vnodes");
272SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
273 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
274static int vnlru_nowhere;
275SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
276 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
277
278/*
279 * Macros to control when a vnode is freed and recycled. All require
280 * the vnode interlock.
281 */
282#define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283#define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
284#define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
285
286/* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
287static int vnsz2log;
288
289/*
290 * Support for the bufobj clean & dirty pctrie.
291 */
292static void *
293buf_trie_alloc(struct pctrie *ptree)
294{
295
296 return uma_zalloc(buf_trie_zone, M_NOWAIT);
297}
298
299static void
300buf_trie_free(struct pctrie *ptree, void *node)
301{
302
303 uma_zfree(buf_trie_zone, node);
304}
305PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
306
307/*
308 * Initialize the vnode management data structures.
309 *
310 * Reevaluate the following cap on the number of vnodes after the physical
311 * memory size exceeds 512GB. In the limit, as the physical memory size
312 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
313 */
314#ifndef MAXVNODES_MAX
315#define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
316#endif
317static void
318vntblinit(void *dummy __unused)
319{
320 u_int i;
321 int physvnodes, virtvnodes;
322
323 /*
324 * Desiredvnodes is a function of the physical memory size and the
325 * kernel's heap size. Generally speaking, it scales with the
326 * physical memory size. The ratio of desiredvnodes to physical pages
327 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
328 * marginal ratio of desiredvnodes to physical pages is one to
329 * sixteen. However, desiredvnodes is limited by the kernel's heap
330 * size. The memory required by desiredvnodes vnodes and vm objects
331 * may not exceed one seventh of the kernel's heap size.
332 */
333 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
334 cnt.v_page_count) / 16;
335 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
336 sizeof(struct vnode)));
337 desiredvnodes = min(physvnodes, virtvnodes);
338 if (desiredvnodes > MAXVNODES_MAX) {
339 if (bootverbose)
340 printf("Reducing kern.maxvnodes %d -> %d\n",
341 desiredvnodes, MAXVNODES_MAX);
342 desiredvnodes = MAXVNODES_MAX;
343 }
344 wantfreevnodes = desiredvnodes / 4;
345 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
346 TAILQ_INIT(&vnode_free_list);
347 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
348 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
349 NULL, NULL, UMA_ALIGN_PTR, 0);
350 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
351 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
352 /*
353 * Preallocate enough nodes to support one-per buf so that
354 * we can not fail an insert. reassignbuf() callers can not
355 * tolerate the insertion failure.
356 */
357 buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
358 NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
359 UMA_ZONE_NOFREE | UMA_ZONE_VM);
360 uma_prealloc(buf_trie_zone, nbuf);
361 /*
362 * Initialize the filesystem syncer.
363 */
364 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
365 &syncer_mask);
366 syncer_maxdelay = syncer_mask + 1;
367 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
368 cv_init(&sync_wakeup, "syncer");
369 for (i = 1; i <= sizeof(struct vnode); i <<= 1)
370 vnsz2log++;
371 vnsz2log--;
372}
373SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
374
375
376/*
377 * Mark a mount point as busy. Used to synchronize access and to delay
378 * unmounting. Eventually, mountlist_mtx is not released on failure.
379 *
380 * vfs_busy() is a custom lock, it can block the caller.
381 * vfs_busy() only sleeps if the unmount is active on the mount point.
382 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
383 * vnode belonging to mp.
384 *
385 * Lookup uses vfs_busy() to traverse mount points.
386 * root fs var fs
387 * / vnode lock A / vnode lock (/var) D
388 * /var vnode lock B /log vnode lock(/var/log) E
389 * vfs_busy lock C vfs_busy lock F
390 *
391 * Within each file system, the lock order is C->A->B and F->D->E.
392 *
393 * When traversing across mounts, the system follows that lock order:
394 *
395 * C->A->B
396 * |
397 * +->F->D->E
398 *
399 * The lookup() process for namei("/var") illustrates the process:
400 * VOP_LOOKUP() obtains B while A is held
401 * vfs_busy() obtains a shared lock on F while A and B are held
402 * vput() releases lock on B
403 * vput() releases lock on A
404 * VFS_ROOT() obtains lock on D while shared lock on F is held
405 * vfs_unbusy() releases shared lock on F
406 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
407 * Attempt to lock A (instead of vp_crossmp) while D is held would
408 * violate the global order, causing deadlocks.
409 *
410 * dounmount() locks B while F is drained.
411 */
412int
413vfs_busy(struct mount *mp, int flags)
414{
415
416 MPASS((flags & ~MBF_MASK) == 0);
417 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
418
419 MNT_ILOCK(mp);
420 MNT_REF(mp);
421 /*
422 * If mount point is currenly being unmounted, sleep until the
423 * mount point fate is decided. If thread doing the unmounting fails,
424 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
425 * that this mount point has survived the unmount attempt and vfs_busy
426 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
427 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
428 * about to be really destroyed. vfs_busy needs to release its
429 * reference on the mount point in this case and return with ENOENT,
430 * telling the caller that mount mount it tried to busy is no longer
431 * valid.
432 */
433 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
434 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
435 MNT_REL(mp);
436 MNT_IUNLOCK(mp);
437 CTR1(KTR_VFS, "%s: failed busying before sleeping",
438 __func__);
439 return (ENOENT);
440 }
441 if (flags & MBF_MNTLSTLOCK)
442 mtx_unlock(&mountlist_mtx);
443 mp->mnt_kern_flag |= MNTK_MWAIT;
444 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
445 if (flags & MBF_MNTLSTLOCK)
446 mtx_lock(&mountlist_mtx);
447 MNT_ILOCK(mp);
448 }
449 if (flags & MBF_MNTLSTLOCK)
450 mtx_unlock(&mountlist_mtx);
451 mp->mnt_lockref++;
452 MNT_IUNLOCK(mp);
453 return (0);
454}
455
456/*
457 * Free a busy filesystem.
458 */
459void
460vfs_unbusy(struct mount *mp)
461{
462
463 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
464 MNT_ILOCK(mp);
465 MNT_REL(mp);
466 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
467 mp->mnt_lockref--;
468 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
469 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
470 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
471 mp->mnt_kern_flag &= ~MNTK_DRAINING;
472 wakeup(&mp->mnt_lockref);
473 }
474 MNT_IUNLOCK(mp);
475}
476
477/*
478 * Lookup a mount point by filesystem identifier.
479 */
480struct mount *
481vfs_getvfs(fsid_t *fsid)
482{
483 struct mount *mp;
484
485 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
486 mtx_lock(&mountlist_mtx);
487 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
488 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
489 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
490 vfs_ref(mp);
491 mtx_unlock(&mountlist_mtx);
492 return (mp);
493 }
494 }
495 mtx_unlock(&mountlist_mtx);
496 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
497 return ((struct mount *) 0);
498}
499
500/*
501 * Lookup a mount point by filesystem identifier, busying it before
502 * returning.
503 */
504struct mount *
505vfs_busyfs(fsid_t *fsid)
506{
507 struct mount *mp;
508 int error;
509
510 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
511 mtx_lock(&mountlist_mtx);
512 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
513 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
514 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
515 error = vfs_busy(mp, MBF_MNTLSTLOCK);
516 if (error) {
517 mtx_unlock(&mountlist_mtx);
518 return (NULL);
519 }
520 return (mp);
521 }
522 }
523 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
524 mtx_unlock(&mountlist_mtx);
525 return ((struct mount *) 0);
526}
527
528/*
529 * Check if a user can access privileged mount options.
530 */
531int
532vfs_suser(struct mount *mp, struct thread *td)
533{
534 int error;
535
536 /*
537 * If the thread is jailed, but this is not a jail-friendly file
538 * system, deny immediately.
539 */
540 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
541 return (EPERM);
542
543 /*
544 * If the file system was mounted outside the jail of the calling
545 * thread, deny immediately.
546 */
547 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
548 return (EPERM);
549
550 /*
551 * If file system supports delegated administration, we don't check
552 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
553 * by the file system itself.
554 * If this is not the user that did original mount, we check for
555 * the PRIV_VFS_MOUNT_OWNER privilege.
556 */
557 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
558 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
559 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
560 return (error);
561 }
562 return (0);
563}
564
565/*
566 * Get a new unique fsid. Try to make its val[0] unique, since this value
567 * will be used to create fake device numbers for stat(). Also try (but
568 * not so hard) make its val[0] unique mod 2^16, since some emulators only
569 * support 16-bit device numbers. We end up with unique val[0]'s for the
570 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
571 *
572 * Keep in mind that several mounts may be running in parallel. Starting
573 * the search one past where the previous search terminated is both a
574 * micro-optimization and a defense against returning the same fsid to
575 * different mounts.
576 */
577void
578vfs_getnewfsid(struct mount *mp)
579{
580 static uint16_t mntid_base;
581 struct mount *nmp;
582 fsid_t tfsid;
583 int mtype;
584
585 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
586 mtx_lock(&mntid_mtx);
587 mtype = mp->mnt_vfc->vfc_typenum;
588 tfsid.val[1] = mtype;
589 mtype = (mtype & 0xFF) << 24;
590 for (;;) {
591 tfsid.val[0] = makedev(255,
592 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
593 mntid_base++;
594 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
595 break;
596 vfs_rel(nmp);
597 }
598 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
599 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
600 mtx_unlock(&mntid_mtx);
601}
602
603/*
604 * Knob to control the precision of file timestamps:
605 *
606 * 0 = seconds only; nanoseconds zeroed.
607 * 1 = seconds and nanoseconds, accurate within 1/HZ.
608 * 2 = seconds and nanoseconds, truncated to microseconds.
609 * >=3 = seconds and nanoseconds, maximum precision.
610 */
611enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
612
613static int timestamp_precision = TSP_SEC;
614SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
615 &timestamp_precision, 0, "File timestamp precision (0: seconds, "
616 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
617 "3+: sec + ns (max. precision))");
618
619/*
620 * Get a current timestamp.
621 */
622void
623vfs_timestamp(struct timespec *tsp)
624{
625 struct timeval tv;
626
627 switch (timestamp_precision) {
628 case TSP_SEC:
629 tsp->tv_sec = time_second;
630 tsp->tv_nsec = 0;
631 break;
632 case TSP_HZ:
633 getnanotime(tsp);
634 break;
635 case TSP_USEC:
636 microtime(&tv);
637 TIMEVAL_TO_TIMESPEC(&tv, tsp);
638 break;
639 case TSP_NSEC:
640 default:
641 nanotime(tsp);
642 break;
643 }
644}
645
646/*
647 * Set vnode attributes to VNOVAL
648 */
649void
650vattr_null(struct vattr *vap)
651{
652
653 vap->va_type = VNON;
654 vap->va_size = VNOVAL;
655 vap->va_bytes = VNOVAL;
656 vap->va_mode = VNOVAL;
657 vap->va_nlink = VNOVAL;
658 vap->va_uid = VNOVAL;
659 vap->va_gid = VNOVAL;
660 vap->va_fsid = VNOVAL;
661 vap->va_fileid = VNOVAL;
662 vap->va_blocksize = VNOVAL;
663 vap->va_rdev = VNOVAL;
664 vap->va_atime.tv_sec = VNOVAL;
665 vap->va_atime.tv_nsec = VNOVAL;
666 vap->va_mtime.tv_sec = VNOVAL;
667 vap->va_mtime.tv_nsec = VNOVAL;
668 vap->va_ctime.tv_sec = VNOVAL;
669 vap->va_ctime.tv_nsec = VNOVAL;
670 vap->va_birthtime.tv_sec = VNOVAL;
671 vap->va_birthtime.tv_nsec = VNOVAL;
672 vap->va_flags = VNOVAL;
673 vap->va_gen = VNOVAL;
674 vap->va_vaflags = 0;
675}
676
677/*
678 * This routine is called when we have too many vnodes. It attempts
679 * to free <count> vnodes and will potentially free vnodes that still
680 * have VM backing store (VM backing store is typically the cause
681 * of a vnode blowout so we want to do this). Therefore, this operation
682 * is not considered cheap.
683 *
684 * A number of conditions may prevent a vnode from being reclaimed.
685 * the buffer cache may have references on the vnode, a directory
686 * vnode may still have references due to the namei cache representing
687 * underlying files, or the vnode may be in active use. It is not
688 * desireable to reuse such vnodes. These conditions may cause the
689 * number of vnodes to reach some minimum value regardless of what
690 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
691 */
692static int
693vlrureclaim(struct mount *mp)
694{
695 struct vnode *vp;
696 int done;
697 int trigger;
698 int usevnodes;
699 int count;
700
701 /*
702 * Calculate the trigger point, don't allow user
703 * screwups to blow us up. This prevents us from
704 * recycling vnodes with lots of resident pages. We
705 * aren't trying to free memory, we are trying to
706 * free vnodes.
707 */
708 usevnodes = desiredvnodes;
709 if (usevnodes <= 0)
710 usevnodes = 1;
711 trigger = cnt.v_page_count * 2 / usevnodes;
712 done = 0;
713 vn_start_write(NULL, &mp, V_WAIT);
714 MNT_ILOCK(mp);
715 count = mp->mnt_nvnodelistsize / 10 + 1;
716 while (count != 0) {
717 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
718 while (vp != NULL && vp->v_type == VMARKER)
719 vp = TAILQ_NEXT(vp, v_nmntvnodes);
720 if (vp == NULL)
721 break;
722 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
723 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
724 --count;
725 if (!VI_TRYLOCK(vp))
726 goto next_iter;
727 /*
728 * If it's been deconstructed already, it's still
729 * referenced, or it exceeds the trigger, skip it.
730 */
731 if (vp->v_usecount ||
732 (!vlru_allow_cache_src &&
733 !LIST_EMPTY(&(vp)->v_cache_src)) ||
734 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
735 vp->v_object->resident_page_count > trigger)) {
736 VI_UNLOCK(vp);
737 goto next_iter;
738 }
739 MNT_IUNLOCK(mp);
740 vholdl(vp);
741 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
742 vdrop(vp);
743 goto next_iter_mntunlocked;
744 }
745 VI_LOCK(vp);
746 /*
747 * v_usecount may have been bumped after VOP_LOCK() dropped
748 * the vnode interlock and before it was locked again.
749 *
750 * It is not necessary to recheck VI_DOOMED because it can
751 * only be set by another thread that holds both the vnode
752 * lock and vnode interlock. If another thread has the
753 * vnode lock before we get to VOP_LOCK() and obtains the
754 * vnode interlock after VOP_LOCK() drops the vnode
755 * interlock, the other thread will be unable to drop the
756 * vnode lock before our VOP_LOCK() call fails.
757 */
758 if (vp->v_usecount ||
759 (!vlru_allow_cache_src &&
760 !LIST_EMPTY(&(vp)->v_cache_src)) ||
761 (vp->v_object != NULL &&
762 vp->v_object->resident_page_count > trigger)) {
763 VOP_UNLOCK(vp, LK_INTERLOCK);
764 vdrop(vp);
765 goto next_iter_mntunlocked;
766 }
767 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
768 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
769 vgonel(vp);
770 VOP_UNLOCK(vp, 0);
771 vdropl(vp);
772 done++;
773next_iter_mntunlocked:
774 if (!should_yield())
775 goto relock_mnt;
776 goto yield;
777next_iter:
778 if (!should_yield())
779 continue;
780 MNT_IUNLOCK(mp);
781yield:
782 kern_yield(PRI_USER);
783relock_mnt:
784 MNT_ILOCK(mp);
785 }
786 MNT_IUNLOCK(mp);
787 vn_finished_write(mp);
788 return done;
789}
790
791/*
792 * Attempt to keep the free list at wantfreevnodes length.
793 */
794static void
795vnlru_free(int count)
796{
797 struct vnode *vp;
798
799 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
800 for (; count > 0; count--) {
801 vp = TAILQ_FIRST(&vnode_free_list);
802 /*
803 * The list can be modified while the free_list_mtx
804 * has been dropped and vp could be NULL here.
805 */
806 if (!vp)
807 break;
808 VNASSERT(vp->v_op != NULL, vp,
809 ("vnlru_free: vnode already reclaimed."));
810 KASSERT((vp->v_iflag & VI_FREE) != 0,
811 ("Removing vnode not on freelist"));
812 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
813 ("Mangling active vnode"));
814 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
815 /*
816 * Don't recycle if we can't get the interlock.
817 */
818 if (!VI_TRYLOCK(vp)) {
819 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
820 continue;
821 }
822 VNASSERT(VCANRECYCLE(vp), vp,
823 ("vp inconsistent on freelist"));
824 freevnodes--;
825 vp->v_iflag &= ~VI_FREE;
826 vholdl(vp);
827 mtx_unlock(&vnode_free_list_mtx);
828 VI_UNLOCK(vp);
829 vtryrecycle(vp);
830 /*
831 * If the recycled succeeded this vdrop will actually free
832 * the vnode. If not it will simply place it back on
833 * the free list.
834 */
835 vdrop(vp);
836 mtx_lock(&vnode_free_list_mtx);
837 }
838}
839/*
840 * Attempt to recycle vnodes in a context that is always safe to block.
841 * Calling vlrurecycle() from the bowels of filesystem code has some
842 * interesting deadlock problems.
843 */
844static struct proc *vnlruproc;
845static int vnlruproc_sig;
846
847static void
848vnlru_proc(void)
849{
850 struct mount *mp, *nmp;
851 int done;
852 struct proc *p = vnlruproc;
853
854 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
855 SHUTDOWN_PRI_FIRST);
856
857 for (;;) {
858 kproc_suspend_check(p);
859 mtx_lock(&vnode_free_list_mtx);
860 if (freevnodes > wantfreevnodes)
861 vnlru_free(freevnodes - wantfreevnodes);
862 if (numvnodes <= desiredvnodes * 9 / 10) {
863 vnlruproc_sig = 0;
864 wakeup(&vnlruproc_sig);
865 msleep(vnlruproc, &vnode_free_list_mtx,
866 PVFS|PDROP, "vlruwt", hz);
867 continue;
868 }
869 mtx_unlock(&vnode_free_list_mtx);
870 done = 0;
871 mtx_lock(&mountlist_mtx);
872 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
873 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
874 nmp = TAILQ_NEXT(mp, mnt_list);
875 continue;
876 }
877 done += vlrureclaim(mp);
878 mtx_lock(&mountlist_mtx);
879 nmp = TAILQ_NEXT(mp, mnt_list);
880 vfs_unbusy(mp);
881 }
882 mtx_unlock(&mountlist_mtx);
883 if (done == 0) {
884#if 0
885 /* These messages are temporary debugging aids */
886 if (vnlru_nowhere < 5)
887 printf("vnlru process getting nowhere..\n");
888 else if (vnlru_nowhere == 5)
889 printf("vnlru process messages stopped.\n");
890#endif
891 vnlru_nowhere++;
892 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
893 } else
894 kern_yield(PRI_USER);
895 }
896}
897
898static struct kproc_desc vnlru_kp = {
899 "vnlru",
900 vnlru_proc,
901 &vnlruproc
902};
903SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
904 &vnlru_kp);
905
906/*
907 * Routines having to do with the management of the vnode table.
908 */
909
910/*
911 * Try to recycle a freed vnode. We abort if anyone picks up a reference
912 * before we actually vgone(). This function must be called with the vnode
913 * held to prevent the vnode from being returned to the free list midway
914 * through vgone().
915 */
916static int
917vtryrecycle(struct vnode *vp)
918{
919 struct mount *vnmp;
920
921 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
922 VNASSERT(vp->v_holdcnt, vp,
923 ("vtryrecycle: Recycling vp %p without a reference.", vp));
924 /*
925 * This vnode may found and locked via some other list, if so we
926 * can't recycle it yet.
927 */
928 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
929 CTR2(KTR_VFS,
930 "%s: impossible to recycle, vp %p lock is already held",
931 __func__, vp);
932 return (EWOULDBLOCK);
933 }
934 /*
935 * Don't recycle if its filesystem is being suspended.
936 */
937 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
938 VOP_UNLOCK(vp, 0);
939 CTR2(KTR_VFS,
940 "%s: impossible to recycle, cannot start the write for %p",
941 __func__, vp);
942 return (EBUSY);
943 }
944 /*
945 * If we got this far, we need to acquire the interlock and see if
946 * anyone picked up this vnode from another list. If not, we will
947 * mark it with DOOMED via vgonel() so that anyone who does find it
948 * will skip over it.
949 */
950 VI_LOCK(vp);
951 if (vp->v_usecount) {
952 VOP_UNLOCK(vp, LK_INTERLOCK);
953 vn_finished_write(vnmp);
954 CTR2(KTR_VFS,
955 "%s: impossible to recycle, %p is already referenced",
956 __func__, vp);
957 return (EBUSY);
958 }
959 if ((vp->v_iflag & VI_DOOMED) == 0)
960 vgonel(vp);
961 VOP_UNLOCK(vp, LK_INTERLOCK);
962 vn_finished_write(vnmp);
963 return (0);
964}
965
966/*
967 * Wait for available vnodes.
968 */
969static int
970getnewvnode_wait(int suspended)
971{
972
973 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
974 if (numvnodes > desiredvnodes) {
975 if (suspended) {
976 /*
977 * File system is beeing suspended, we cannot risk a
978 * deadlock here, so allocate new vnode anyway.
979 */
980 if (freevnodes > wantfreevnodes)
981 vnlru_free(freevnodes - wantfreevnodes);
982 return (0);
983 }
984 if (vnlruproc_sig == 0) {
985 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
986 wakeup(vnlruproc);
987 }
988 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
989 "vlruwk", hz);
990 }
991 return (numvnodes > desiredvnodes ? ENFILE : 0);
992}
993
994void
995getnewvnode_reserve(u_int count)
996{
997 struct thread *td;
998
999 td = curthread;
1000 mtx_lock(&vnode_free_list_mtx);
1001 while (count > 0) {
1002 if (getnewvnode_wait(0) == 0) {
1003 count--;
1004 td->td_vp_reserv++;
1005 numvnodes++;
1006 }
1007 }
1008 mtx_unlock(&vnode_free_list_mtx);
1009}
1010
1011void
1012getnewvnode_drop_reserve(void)
1013{
1014 struct thread *td;
1015
1016 td = curthread;
1017 mtx_lock(&vnode_free_list_mtx);
1018 KASSERT(numvnodes >= td->td_vp_reserv, ("reserve too large"));
1019 numvnodes -= td->td_vp_reserv;
1020 mtx_unlock(&vnode_free_list_mtx);
1021 td->td_vp_reserv = 0;
1022}
1023
1024/*
1025 * Return the next vnode from the free list.
1026 */
1027int
1028getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1029 struct vnode **vpp)
1030{
1031 struct vnode *vp;
1032 struct bufobj *bo;
1033 struct thread *td;
1034 int error;
1035
1036 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1037 vp = NULL;
1038 td = curthread;
1039 if (td->td_vp_reserv > 0) {
1040 td->td_vp_reserv -= 1;
1041 goto alloc;
1042 }
1043 mtx_lock(&vnode_free_list_mtx);
1044 /*
1045 * Lend our context to reclaim vnodes if they've exceeded the max.
1046 */
1047 if (freevnodes > wantfreevnodes)
1048 vnlru_free(1);
1049 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1050 MNTK_SUSPEND));
1051#if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1052 if (error != 0) {
1053 mtx_unlock(&vnode_free_list_mtx);
1054 return (error);
1055 }
1056#endif
1057 numvnodes++;
1058 mtx_unlock(&vnode_free_list_mtx);
1059alloc:
1060 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1061 /*
1062 * Setup locks.
1063 */
1064 vp->v_vnlock = &vp->v_lock;
1065 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1066 /*
1067 * By default, don't allow shared locks unless filesystems
1068 * opt-in.
1069 */
1070 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
1071 /*
1072 * Initialize bufobj.
1073 */
1074 bo = &vp->v_bufobj;
1075 bo->__bo_vnode = vp;
1076 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1077 bo->bo_ops = &buf_ops_bio;
1078 bo->bo_private = vp;
1079 TAILQ_INIT(&bo->bo_clean.bv_hd);
1080 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1081 /*
1082 * Initialize namecache.
1083 */
1084 LIST_INIT(&vp->v_cache_src);
1085 TAILQ_INIT(&vp->v_cache_dst);
1086 /*
1087 * Finalize various vnode identity bits.
1088 */
1089 vp->v_type = VNON;
1090 vp->v_tag = tag;
1091 vp->v_op = vops;
1092 v_incr_usecount(vp);
1093 vp->v_data = NULL;
1094#ifdef MAC
1095 mac_vnode_init(vp);
1096 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1097 mac_vnode_associate_singlelabel(mp, vp);
1098 else if (mp == NULL && vops != &dead_vnodeops)
1099 printf("NULL mp in getnewvnode()\n");
1100#endif
1101 if (mp != NULL) {
1102 bo->bo_bsize = mp->mnt_stat.f_iosize;
1103 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1104 vp->v_vflag |= VV_NOKNOTE;
1105 }
1106 rangelock_init(&vp->v_rl);
1107
1108 /*
1109 * For the filesystems which do not use vfs_hash_insert(),
1110 * still initialize v_hash to have vfs_hash_index() useful.
1111 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1112 * its own hashing.
1113 */
1114 vp->v_hash = (uintptr_t)vp >> vnsz2log;
1115
1116 *vpp = vp;
1117 return (0);
1118}
1119
1120/*
1121 * Delete from old mount point vnode list, if on one.
1122 */
1123static void
1124delmntque(struct vnode *vp)
1125{
1126 struct mount *mp;
1127 int active;
1128
1129 mp = vp->v_mount;
1130 if (mp == NULL)
1131 return;
1132 MNT_ILOCK(mp);
1133 VI_LOCK(vp);
1134 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1135 ("Active vnode list size %d > Vnode list size %d",
1136 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1137 active = vp->v_iflag & VI_ACTIVE;
1138 vp->v_iflag &= ~VI_ACTIVE;
1139 if (active) {
1140 mtx_lock(&vnode_free_list_mtx);
1141 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1142 mp->mnt_activevnodelistsize--;
1143 mtx_unlock(&vnode_free_list_mtx);
1144 }
1145 vp->v_mount = NULL;
1146 VI_UNLOCK(vp);
1147 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1148 ("bad mount point vnode list size"));
1149 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1150 mp->mnt_nvnodelistsize--;
1151 MNT_REL(mp);
1152 MNT_IUNLOCK(mp);
1153}
1154
1155static void
1156insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1157{
1158
1159 vp->v_data = NULL;
1160 vp->v_op = &dead_vnodeops;
1161 vgone(vp);
1162 vput(vp);
1163}
1164
1165/*
1166 * Insert into list of vnodes for the new mount point, if available.
1167 */
1168int
1169insmntque1(struct vnode *vp, struct mount *mp,
1170 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1171{
1172
1173 KASSERT(vp->v_mount == NULL,
1174 ("insmntque: vnode already on per mount vnode list"));
1175 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1176 ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1177
1178 /*
1179 * We acquire the vnode interlock early to ensure that the
1180 * vnode cannot be recycled by another process releasing a
1181 * holdcnt on it before we get it on both the vnode list
1182 * and the active vnode list. The mount mutex protects only
1183 * manipulation of the vnode list and the vnode freelist
1184 * mutex protects only manipulation of the active vnode list.
1185 * Hence the need to hold the vnode interlock throughout.
1186 */
1187 MNT_ILOCK(mp);
1188 VI_LOCK(vp);
1189 if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1190 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1191 mp->mnt_nvnodelistsize == 0)) &&
1192 (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1193 VI_UNLOCK(vp);
1194 MNT_IUNLOCK(mp);
1195 if (dtr != NULL)
1196 dtr(vp, dtr_arg);
1197 return (EBUSY);
1198 }
1199 vp->v_mount = mp;
1200 MNT_REF(mp);
1201 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1202 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1203 ("neg mount point vnode list size"));
1204 mp->mnt_nvnodelistsize++;
1205 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1206 ("Activating already active vnode"));
1207 vp->v_iflag |= VI_ACTIVE;
1208 mtx_lock(&vnode_free_list_mtx);
1209 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1210 mp->mnt_activevnodelistsize++;
1211 mtx_unlock(&vnode_free_list_mtx);
1212 VI_UNLOCK(vp);
1213 MNT_IUNLOCK(mp);
1214 return (0);
1215}
1216
1217int
1218insmntque(struct vnode *vp, struct mount *mp)
1219{
1220
1221 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1222}
1223
1224/*
1225 * Flush out and invalidate all buffers associated with a bufobj
1226 * Called with the underlying object locked.
1227 */
1228int
1229bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1230{
1231 int error;
1232
1233 BO_LOCK(bo);
1234 if (flags & V_SAVE) {
1235 error = bufobj_wwait(bo, slpflag, slptimeo);
1236 if (error) {
1237 BO_UNLOCK(bo);
1238 return (error);
1239 }
1240 if (bo->bo_dirty.bv_cnt > 0) {
1241 BO_UNLOCK(bo);
1242 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1243 return (error);
1244 /*
1245 * XXX We could save a lock/unlock if this was only
1246 * enabled under INVARIANTS
1247 */
1248 BO_LOCK(bo);
1249 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1250 panic("vinvalbuf: dirty bufs");
1251 }
1252 }
1253 /*
1254 * If you alter this loop please notice that interlock is dropped and
1255 * reacquired in flushbuflist. Special care is needed to ensure that
1256 * no race conditions occur from this.
1257 */
1258 do {
1259 error = flushbuflist(&bo->bo_clean,
1260 flags, bo, slpflag, slptimeo);
1261 if (error == 0 && !(flags & V_CLEANONLY))
1262 error = flushbuflist(&bo->bo_dirty,
1263 flags, bo, slpflag, slptimeo);
1264 if (error != 0 && error != EAGAIN) {
1265 BO_UNLOCK(bo);
1266 return (error);
1267 }
1268 } while (error != 0);
1269
1270 /*
1271 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1272 * have write I/O in-progress but if there is a VM object then the
1273 * VM object can also have read-I/O in-progress.
1274 */
1275 do {
1276 bufobj_wwait(bo, 0, 0);
1277 BO_UNLOCK(bo);
1278 if (bo->bo_object != NULL) {
1279 VM_OBJECT_WLOCK(bo->bo_object);
1280 vm_object_pip_wait(bo->bo_object, "bovlbx");
1281 VM_OBJECT_WUNLOCK(bo->bo_object);
1282 }
1283 BO_LOCK(bo);
1284 } while (bo->bo_numoutput > 0);
1285 BO_UNLOCK(bo);
1286
1287 /*
1288 * Destroy the copy in the VM cache, too.
1289 */
1290 if (bo->bo_object != NULL &&
1291 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1292 VM_OBJECT_WLOCK(bo->bo_object);
1293 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1294 OBJPR_CLEANONLY : 0);
1295 VM_OBJECT_WUNLOCK(bo->bo_object);
1296 }
1297
1298#ifdef INVARIANTS
1299 BO_LOCK(bo);
1300 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1301 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1302 panic("vinvalbuf: flush failed");
1303 BO_UNLOCK(bo);
1304#endif
1305 return (0);
1306}
1307
1308/*
1309 * Flush out and invalidate all buffers associated with a vnode.
1310 * Called with the underlying object locked.
1311 */
1312int
1313vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1314{
1315
1316 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1317 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1318 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1319}
1320
1321/*
1322 * Flush out buffers on the specified list.
1323 *
1324 */
1325static int
1326flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1327 int slptimeo)
1328{
1329 struct buf *bp, *nbp;
1330 int retval, error;
1331 daddr_t lblkno;
1332 b_xflags_t xflags;
1333
1334 ASSERT_BO_LOCKED(bo);
1335
1336 retval = 0;
1337 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1338 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1339 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1340 continue;
1341 }
1342 lblkno = 0;
1343 xflags = 0;
1344 if (nbp != NULL) {
1345 lblkno = nbp->b_lblkno;
1346 xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1347 }
1348 retval = EAGAIN;
1349 error = BUF_TIMELOCK(bp,
1350 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1351 "flushbuf", slpflag, slptimeo);
1352 if (error) {
1353 BO_LOCK(bo);
1354 return (error != ENOLCK ? error : EAGAIN);
1355 }
1356 KASSERT(bp->b_bufobj == bo,
1357 ("bp %p wrong b_bufobj %p should be %p",
1358 bp, bp->b_bufobj, bo));
1359 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1360 BUF_UNLOCK(bp);
1361 BO_LOCK(bo);
1362 return (EAGAIN);
1363 }
1364 /*
1365 * XXX Since there are no node locks for NFS, I
1366 * believe there is a slight chance that a delayed
1367 * write will occur while sleeping just above, so
1368 * check for it.
1369 */
1370 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1371 (flags & V_SAVE)) {
1372 BO_LOCK(bo);
1373 bremfree(bp);
1374 BO_UNLOCK(bo);
1375 bp->b_flags |= B_ASYNC;
1376 bwrite(bp);
1377 BO_LOCK(bo);
1378 return (EAGAIN); /* XXX: why not loop ? */
1379 }
1380 BO_LOCK(bo);
1381 bremfree(bp);
1382 BO_UNLOCK(bo);
1383 bp->b_flags |= (B_INVAL | B_RELBUF);
1384 bp->b_flags &= ~B_ASYNC;
1385 brelse(bp);
1386 BO_LOCK(bo);
1387 if (nbp != NULL &&
1388 (nbp->b_bufobj != bo ||
1389 nbp->b_lblkno != lblkno ||
1390 (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1391 break; /* nbp invalid */
1392 }
1393 return (retval);
1394}
1395
1396/*
1397 * Truncate a file's buffer and pages to a specified length. This
1398 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1399 * sync activity.
1400 */
1401int
1402vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1403{
1404 struct buf *bp, *nbp;
1405 int anyfreed;
1406 int trunclbn;
1407 struct bufobj *bo;
1408
1409 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1410 vp, cred, blksize, (uintmax_t)length);
1411
1412 /*
1413 * Round up to the *next* lbn.
1414 */
1415 trunclbn = (length + blksize - 1) / blksize;
1416
1417 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1418restart:
1419 bo = &vp->v_bufobj;
1420 BO_LOCK(bo);
1421 anyfreed = 1;
1422 for (;anyfreed;) {
1423 anyfreed = 0;
1424 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1425 if (bp->b_lblkno < trunclbn)
1426 continue;
1427 if (BUF_LOCK(bp,
1428 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1429 BO_MTX(bo)) == ENOLCK)
1430 goto restart;
1431
1432 BO_LOCK(bo);
1433 bremfree(bp);
1434 BO_UNLOCK(bo);
1435 bp->b_flags |= (B_INVAL | B_RELBUF);
1436 bp->b_flags &= ~B_ASYNC;
1437 brelse(bp);
1438 anyfreed = 1;
1439
1440 BO_LOCK(bo);
1441 if (nbp != NULL &&
1442 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1443 (nbp->b_vp != vp) ||
1444 (nbp->b_flags & B_DELWRI))) {
1445 BO_UNLOCK(bo);
1446 goto restart;
1447 }
1448 }
1449
1450 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1451 if (bp->b_lblkno < trunclbn)
1452 continue;
1453 if (BUF_LOCK(bp,
1454 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1455 BO_MTX(bo)) == ENOLCK)
1456 goto restart;
1457 BO_LOCK(bo);
1458 bremfree(bp);
1459 BO_UNLOCK(bo);
1460 bp->b_flags |= (B_INVAL | B_RELBUF);
1461 bp->b_flags &= ~B_ASYNC;
1462 brelse(bp);
1463 anyfreed = 1;
1464
1465 BO_LOCK(bo);
1466 if (nbp != NULL &&
1467 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1468 (nbp->b_vp != vp) ||
1469 (nbp->b_flags & B_DELWRI) == 0)) {
1470 BO_UNLOCK(bo);
1471 goto restart;
1472 }
1473 }
1474 }
1475
1476 if (length > 0) {
1477restartsync:
1478 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1479 if (bp->b_lblkno > 0)
1480 continue;
1481 /*
1482 * Since we hold the vnode lock this should only
1483 * fail if we're racing with the buf daemon.
1484 */
1485 if (BUF_LOCK(bp,
1486 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1487 BO_MTX(bo)) == ENOLCK) {
1488 goto restart;
1489 }
1490 VNASSERT((bp->b_flags & B_DELWRI), vp,
1491 ("buf(%p) on dirty queue without DELWRI", bp));
1492
1493 BO_LOCK(bo);
1494 bremfree(bp);
1495 BO_UNLOCK(bo);
1496 bawrite(bp);
1497 BO_LOCK(bo);
1498 goto restartsync;
1499 }
1500 }
1501
1502 bufobj_wwait(bo, 0, 0);
1503 BO_UNLOCK(bo);
1504 vnode_pager_setsize(vp, length);
1505
1506 return (0);
1507}
1508
1509static void
1510buf_vlist_remove(struct buf *bp)
1511{
1512 struct bufv *bv;
1513
1514 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1515 ASSERT_BO_LOCKED(bp->b_bufobj);
1516 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1517 (BX_VNDIRTY|BX_VNCLEAN),
1518 ("buf_vlist_remove: Buf %p is on two lists", bp));
1519 if (bp->b_xflags & BX_VNDIRTY)
1520 bv = &bp->b_bufobj->bo_dirty;
1521 else
1522 bv = &bp->b_bufobj->bo_clean;
1523 BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1524 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1525 bv->bv_cnt--;
1526 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1527}
1528
1529/*
1530 * Add the buffer to the sorted clean or dirty block list.
1531 *
1532 * NOTE: xflags is passed as a constant, optimizing this inline function!
1533 */
1534static void
1535buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1536{
1537 struct bufv *bv;
1538 struct buf *n;
1539 int error;
1540
1541 ASSERT_BO_LOCKED(bo);
1542 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1543 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1544 bp->b_xflags |= xflags;
1545 if (xflags & BX_VNDIRTY)
1546 bv = &bo->bo_dirty;
1547 else
1548 bv = &bo->bo_clean;
1549
1550 /*
1551 * Keep the list ordered. Optimize empty list insertion. Assume
1552 * we tend to grow at the tail so lookup_le should usually be cheaper
1553 * than _ge.
1554 */
1555 if (bv->bv_cnt == 0 ||
1556 bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1557 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1558 else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1559 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1560 else
1561 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1562 error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1563 if (error)
1564 panic("buf_vlist_add: Preallocated nodes insufficient.");
1565 bv->bv_cnt++;
1566}
1567
1568/*
1569 * Lookup a buffer using the splay tree. Note that we specifically avoid
1570 * shadow buffers used in background bitmap writes.
1571 *
1572 * This code isn't quite efficient as it could be because we are maintaining
1573 * two sorted lists and do not know which list the block resides in.
1574 *
1575 * During a "make buildworld" the desired buffer is found at one of
1576 * the roots more than 60% of the time. Thus, checking both roots
1577 * before performing either splay eliminates unnecessary splays on the
1578 * first tree splayed.
1579 */
1580struct buf *
1581gbincore(struct bufobj *bo, daddr_t lblkno)
1582{
1583 struct buf *bp;
1584
1585 ASSERT_BO_LOCKED(bo);
1586 bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1587 if (bp != NULL)
1588 return (bp);
1589 return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1590}
1591
1592/*
1593 * Associate a buffer with a vnode.
1594 */
1595void
1596bgetvp(struct vnode *vp, struct buf *bp)
1597{
1598 struct bufobj *bo;
1599
1600 bo = &vp->v_bufobj;
1601 ASSERT_BO_LOCKED(bo);
1602 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1603
1604 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1605 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1606 ("bgetvp: bp already attached! %p", bp));
1607
1608 vhold(vp);
1609 bp->b_vp = vp;
1610 bp->b_bufobj = bo;
1611 /*
1612 * Insert onto list for new vnode.
1613 */
1614 buf_vlist_add(bp, bo, BX_VNCLEAN);
1615}
1616
1617/*
1618 * Disassociate a buffer from a vnode.
1619 */
1620void
1621brelvp(struct buf *bp)
1622{
1623 struct bufobj *bo;
1624 struct vnode *vp;
1625
1626 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1627 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1628
1629 /*
1630 * Delete from old vnode list, if on one.
1631 */
1632 vp = bp->b_vp; /* XXX */
1633 bo = bp->b_bufobj;
1634 BO_LOCK(bo);
1635 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1636 buf_vlist_remove(bp);
1637 else
1638 panic("brelvp: Buffer %p not on queue.", bp);
1639 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1640 bo->bo_flag &= ~BO_ONWORKLST;
1641 mtx_lock(&sync_mtx);
1642 LIST_REMOVE(bo, bo_synclist);
1643 syncer_worklist_len--;
1644 mtx_unlock(&sync_mtx);
1645 }
1646 bp->b_vp = NULL;
1647 bp->b_bufobj = NULL;
1648 BO_UNLOCK(bo);
1649 vdrop(vp);
1650}
1651
1652/*
1653 * Add an item to the syncer work queue.
1654 */
1655static void
1656vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1657{
1658 int slot;
1659
1660 ASSERT_BO_LOCKED(bo);
1661
1662 mtx_lock(&sync_mtx);
1663 if (bo->bo_flag & BO_ONWORKLST)
1664 LIST_REMOVE(bo, bo_synclist);
1665 else {
1666 bo->bo_flag |= BO_ONWORKLST;
1667 syncer_worklist_len++;
1668 }
1669
1670 if (delay > syncer_maxdelay - 2)
1671 delay = syncer_maxdelay - 2;
1672 slot = (syncer_delayno + delay) & syncer_mask;
1673
1674 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1675 mtx_unlock(&sync_mtx);
1676}
1677
1678static int
1679sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1680{
1681 int error, len;
1682
1683 mtx_lock(&sync_mtx);
1684 len = syncer_worklist_len - sync_vnode_count;
1685 mtx_unlock(&sync_mtx);
1686 error = SYSCTL_OUT(req, &len, sizeof(len));
1687 return (error);
1688}
1689
1690SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1691 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1692
1693static struct proc *updateproc;
1694static void sched_sync(void);
1695static struct kproc_desc up_kp = {
1696 "syncer",
1697 sched_sync,
1698 &updateproc
1699};
1700SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1701
1702static int
1703sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1704{
1705 struct vnode *vp;
1706 struct mount *mp;
1707
1708 *bo = LIST_FIRST(slp);
1709 if (*bo == NULL)
1710 return (0);
1711 vp = (*bo)->__bo_vnode; /* XXX */
1712 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1713 return (1);
1714 /*
1715 * We use vhold in case the vnode does not
1716 * successfully sync. vhold prevents the vnode from
1717 * going away when we unlock the sync_mtx so that
1718 * we can acquire the vnode interlock.
1719 */
1720 vholdl(vp);
1721 mtx_unlock(&sync_mtx);
1722 VI_UNLOCK(vp);
1723 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1724 vdrop(vp);
1725 mtx_lock(&sync_mtx);
1726 return (*bo == LIST_FIRST(slp));
1727 }
1728 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1729 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1730 VOP_UNLOCK(vp, 0);
1731 vn_finished_write(mp);
1732 BO_LOCK(*bo);
1733 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1734 /*
1735 * Put us back on the worklist. The worklist
1736 * routine will remove us from our current
1737 * position and then add us back in at a later
1738 * position.
1739 */
1740 vn_syncer_add_to_worklist(*bo, syncdelay);
1741 }
1742 BO_UNLOCK(*bo);
1743 vdrop(vp);
1744 mtx_lock(&sync_mtx);
1745 return (0);
1746}
1747
1748/*
1749 * System filesystem synchronizer daemon.
1750 */
1751static void
1752sched_sync(void)
1753{
1754 struct synclist *next, *slp;
1755 struct bufobj *bo;
1756 long starttime;
1757 struct thread *td = curthread;
1758 int last_work_seen;
1759 int net_worklist_len;
1760 int syncer_final_iter;
1761 int first_printf;
1762 int error;
1763
1764 last_work_seen = 0;
1765 syncer_final_iter = 0;
1766 first_printf = 1;
1767 syncer_state = SYNCER_RUNNING;
1768 starttime = time_uptime;
1769 td->td_pflags |= TDP_NORUNNINGBUF;
1770
1771 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1772 SHUTDOWN_PRI_LAST);
1773
1774 mtx_lock(&sync_mtx);
1775 for (;;) {
1776 if (syncer_state == SYNCER_FINAL_DELAY &&
1777 syncer_final_iter == 0) {
1778 mtx_unlock(&sync_mtx);
1779 kproc_suspend_check(td->td_proc);
1780 mtx_lock(&sync_mtx);
1781 }
1782 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1783 if (syncer_state != SYNCER_RUNNING &&
1784 starttime != time_uptime) {
1785 if (first_printf) {
1786 printf("\nSyncing disks, vnodes remaining...");
1787 first_printf = 0;
1788 }
1789 printf("%d ", net_worklist_len);
1790 }
1791 starttime = time_uptime;
1792
1793 /*
1794 * Push files whose dirty time has expired. Be careful
1795 * of interrupt race on slp queue.
1796 *
1797 * Skip over empty worklist slots when shutting down.
1798 */
1799 do {
1800 slp = &syncer_workitem_pending[syncer_delayno];
1801 syncer_delayno += 1;
1802 if (syncer_delayno == syncer_maxdelay)
1803 syncer_delayno = 0;
1804 next = &syncer_workitem_pending[syncer_delayno];
1805 /*
1806 * If the worklist has wrapped since the
1807 * it was emptied of all but syncer vnodes,
1808 * switch to the FINAL_DELAY state and run
1809 * for one more second.
1810 */
1811 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1812 net_worklist_len == 0 &&
1813 last_work_seen == syncer_delayno) {
1814 syncer_state = SYNCER_FINAL_DELAY;
1815 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1816 }
1817 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1818 syncer_worklist_len > 0);
1819
1820 /*
1821 * Keep track of the last time there was anything
1822 * on the worklist other than syncer vnodes.
1823 * Return to the SHUTTING_DOWN state if any
1824 * new work appears.
1825 */
1826 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1827 last_work_seen = syncer_delayno;
1828 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1829 syncer_state = SYNCER_SHUTTING_DOWN;
1830 while (!LIST_EMPTY(slp)) {
1831 error = sync_vnode(slp, &bo, td);
1832 if (error == 1) {
1833 LIST_REMOVE(bo, bo_synclist);
1834 LIST_INSERT_HEAD(next, bo, bo_synclist);
1835 continue;
1836 }
1837
1838 if (first_printf == 0)
1839 wdog_kern_pat(WD_LASTVAL);
1840
1841 }
1842 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1843 syncer_final_iter--;
1844 /*
1845 * The variable rushjob allows the kernel to speed up the
1846 * processing of the filesystem syncer process. A rushjob
1847 * value of N tells the filesystem syncer to process the next
1848 * N seconds worth of work on its queue ASAP. Currently rushjob
1849 * is used by the soft update code to speed up the filesystem
1850 * syncer process when the incore state is getting so far
1851 * ahead of the disk that the kernel memory pool is being
1852 * threatened with exhaustion.
1853 */
1854 if (rushjob > 0) {
1855 rushjob -= 1;
1856 continue;
1857 }
1858 /*
1859 * Just sleep for a short period of time between
1860 * iterations when shutting down to allow some I/O
1861 * to happen.
1862 *
1863 * If it has taken us less than a second to process the
1864 * current work, then wait. Otherwise start right over
1865 * again. We can still lose time if any single round
1866 * takes more than two seconds, but it does not really
1867 * matter as we are just trying to generally pace the
1868 * filesystem activity.
1869 */
1870 if (syncer_state != SYNCER_RUNNING ||
1871 time_uptime == starttime) {
1872 thread_lock(td);
1873 sched_prio(td, PPAUSE);
1874 thread_unlock(td);
1875 }
1876 if (syncer_state != SYNCER_RUNNING)
1877 cv_timedwait(&sync_wakeup, &sync_mtx,
1878 hz / SYNCER_SHUTDOWN_SPEEDUP);
1879 else if (time_uptime == starttime)
1880 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1881 }
1882}
1883
1884/*
1885 * Request the syncer daemon to speed up its work.
1886 * We never push it to speed up more than half of its
1887 * normal turn time, otherwise it could take over the cpu.
1888 */
1889int
1890speedup_syncer(void)
1891{
1892 int ret = 0;
1893
1894 mtx_lock(&sync_mtx);
1895 if (rushjob < syncdelay / 2) {
1896 rushjob += 1;
1897 stat_rush_requests += 1;
1898 ret = 1;
1899 }
1900 mtx_unlock(&sync_mtx);
1901 cv_broadcast(&sync_wakeup);
1902 return (ret);
1903}
1904
1905/*
1906 * Tell the syncer to speed up its work and run though its work
1907 * list several times, then tell it to shut down.
1908 */
1909static void
1910syncer_shutdown(void *arg, int howto)
1911{
1912
1913 if (howto & RB_NOSYNC)
1914 return;
1915 mtx_lock(&sync_mtx);
1916 syncer_state = SYNCER_SHUTTING_DOWN;
1917 rushjob = 0;
1918 mtx_unlock(&sync_mtx);
1919 cv_broadcast(&sync_wakeup);
1920 kproc_shutdown(arg, howto);
1921}
1922
1923/*
1924 * Reassign a buffer from one vnode to another.
1925 * Used to assign file specific control information
1926 * (indirect blocks) to the vnode to which they belong.
1927 */
1928void
1929reassignbuf(struct buf *bp)
1930{
1931 struct vnode *vp;
1932 struct bufobj *bo;
1933 int delay;
1934#ifdef INVARIANTS
1935 struct bufv *bv;
1936#endif
1937
1938 vp = bp->b_vp;
1939 bo = bp->b_bufobj;
1940 ++reassignbufcalls;
1941
1942 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1943 bp, bp->b_vp, bp->b_flags);
1944 /*
1945 * B_PAGING flagged buffers cannot be reassigned because their vp
1946 * is not fully linked in.
1947 */
1948 if (bp->b_flags & B_PAGING)
1949 panic("cannot reassign paging buffer");
1950
1951 /*
1952 * Delete from old vnode list, if on one.
1953 */
1954 BO_LOCK(bo);
1955 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1956 buf_vlist_remove(bp);
1957 else
1958 panic("reassignbuf: Buffer %p not on queue.", bp);
1959 /*
1960 * If dirty, put on list of dirty buffers; otherwise insert onto list
1961 * of clean buffers.
1962 */
1963 if (bp->b_flags & B_DELWRI) {
1964 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1965 switch (vp->v_type) {
1966 case VDIR:
1967 delay = dirdelay;
1968 break;
1969 case VCHR:
1970 delay = metadelay;
1971 break;
1972 default:
1973 delay = filedelay;
1974 }
1975 vn_syncer_add_to_worklist(bo, delay);
1976 }
1977 buf_vlist_add(bp, bo, BX_VNDIRTY);
1978 } else {
1979 buf_vlist_add(bp, bo, BX_VNCLEAN);
1980
1981 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1982 mtx_lock(&sync_mtx);
1983 LIST_REMOVE(bo, bo_synclist);
1984 syncer_worklist_len--;
1985 mtx_unlock(&sync_mtx);
1986 bo->bo_flag &= ~BO_ONWORKLST;
1987 }
1988 }
1989#ifdef INVARIANTS
1990 bv = &bo->bo_clean;
1991 bp = TAILQ_FIRST(&bv->bv_hd);
1992 KASSERT(bp == NULL || bp->b_bufobj == bo,
1993 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1994 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1995 KASSERT(bp == NULL || bp->b_bufobj == bo,
1996 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1997 bv = &bo->bo_dirty;
1998 bp = TAILQ_FIRST(&bv->bv_hd);
1999 KASSERT(bp == NULL || bp->b_bufobj == bo,
2000 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2001 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2002 KASSERT(bp == NULL || bp->b_bufobj == bo,
2003 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2004#endif
2005 BO_UNLOCK(bo);
2006}
2007
2008/*
2009 * Increment the use and hold counts on the vnode, taking care to reference
2010 * the driver's usecount if this is a chardev. The vholdl() will remove
2011 * the vnode from the free list if it is presently free. Requires the
2012 * vnode interlock and returns with it held.
2013 */
2014static void
2015v_incr_usecount(struct vnode *vp)
2016{
2017
2018 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2019 vp->v_usecount++;
2020 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2021 dev_lock();
2022 vp->v_rdev->si_usecount++;
2023 dev_unlock();
2024 }
2025 vholdl(vp);
2026}
2027
2028/*
2029 * Turn a holdcnt into a use+holdcnt such that only one call to
2030 * v_decr_usecount is needed.
2031 */
2032static void
2033v_upgrade_usecount(struct vnode *vp)
2034{
2035
2036 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2037 vp->v_usecount++;
2038 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2039 dev_lock();
2040 vp->v_rdev->si_usecount++;
2041 dev_unlock();
2042 }
2043}
2044
2045/*
2046 * Decrement the vnode use and hold count along with the driver's usecount
2047 * if this is a chardev. The vdropl() below releases the vnode interlock
2048 * as it may free the vnode.
2049 */
2050static void
2051v_decr_usecount(struct vnode *vp)
2052{
2053
2054 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2055 VNASSERT(vp->v_usecount > 0, vp,
2056 ("v_decr_usecount: negative usecount"));
2057 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2058 vp->v_usecount--;
2059 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2060 dev_lock();
2061 vp->v_rdev->si_usecount--;
2062 dev_unlock();
2063 }
2064 vdropl(vp);
2065}
2066
2067/*
2068 * Decrement only the use count and driver use count. This is intended to
2069 * be paired with a follow on vdropl() to release the remaining hold count.
2070 * In this way we may vgone() a vnode with a 0 usecount without risk of
2071 * having it end up on a free list because the hold count is kept above 0.
2072 */
2073static void
2074v_decr_useonly(struct vnode *vp)
2075{
2076
2077 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2078 VNASSERT(vp->v_usecount > 0, vp,
2079 ("v_decr_useonly: negative usecount"));
2080 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2081 vp->v_usecount--;
2082 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2083 dev_lock();
2084 vp->v_rdev->si_usecount--;
2085 dev_unlock();
2086 }
2087}
2088
2089/*
2090 * Grab a particular vnode from the free list, increment its
2091 * reference count and lock it. VI_DOOMED is set if the vnode
2092 * is being destroyed. Only callers who specify LK_RETRY will
2093 * see doomed vnodes. If inactive processing was delayed in
2094 * vput try to do it here.
2095 */
2096int
2097vget(struct vnode *vp, int flags, struct thread *td)
2098{
2099 int error;
2100
2101 error = 0;
2102 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2103 ("vget: invalid lock operation"));
2104 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2105
2106 if ((flags & LK_INTERLOCK) == 0)
2107 VI_LOCK(vp);
2108 vholdl(vp);
2109 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2110 vdrop(vp);
2111 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2112 vp);
2113 return (error);
2114 }
2115 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2116 panic("vget: vn_lock failed to return ENOENT\n");
2117 VI_LOCK(vp);
2118 /* Upgrade our holdcnt to a usecount. */
2119 v_upgrade_usecount(vp);
2120 /*
2121 * We don't guarantee that any particular close will
2122 * trigger inactive processing so just make a best effort
2123 * here at preventing a reference to a removed file. If
2124 * we don't succeed no harm is done.
2125 */
2126 if (vp->v_iflag & VI_OWEINACT) {
2127 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2128 (flags & LK_NOWAIT) == 0)
2129 vinactive(vp, td);
2130 vp->v_iflag &= ~VI_OWEINACT;
2131 }
2132 VI_UNLOCK(vp);
2133 return (0);
2134}
2135
2136/*
2137 * Increase the reference count of a vnode.
2138 */
2139void
2140vref(struct vnode *vp)
2141{
2142
2143 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2144 VI_LOCK(vp);
2145 v_incr_usecount(vp);
2146 VI_UNLOCK(vp);
2147}
2148
2149/*
2150 * Return reference count of a vnode.
2151 *
2152 * The results of this call are only guaranteed when some mechanism other
2153 * than the VI lock is used to stop other processes from gaining references
2154 * to the vnode. This may be the case if the caller holds the only reference.
2155 * This is also useful when stale data is acceptable as race conditions may
2156 * be accounted for by some other means.
2157 */
2158int
2159vrefcnt(struct vnode *vp)
2160{
2161 int usecnt;
2162
2163 VI_LOCK(vp);
2164 usecnt = vp->v_usecount;
2165 VI_UNLOCK(vp);
2166
2167 return (usecnt);
2168}
2169
2170#define VPUTX_VRELE 1
2171#define VPUTX_VPUT 2
2172#define VPUTX_VUNREF 3
2173
2174static void
2175vputx(struct vnode *vp, int func)
2176{
2177 int error;
2178
2179 KASSERT(vp != NULL, ("vputx: null vp"));
2180 if (func == VPUTX_VUNREF)
2181 ASSERT_VOP_LOCKED(vp, "vunref");
2182 else if (func == VPUTX_VPUT)
2183 ASSERT_VOP_LOCKED(vp, "vput");
2184 else
2185 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2186 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2187 VI_LOCK(vp);
2188
2189 /* Skip this v_writecount check if we're going to panic below. */
2190 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2191 ("vputx: missed vn_close"));
2192 error = 0;
2193
2194 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2195 vp->v_usecount == 1)) {
2196 if (func == VPUTX_VPUT)
2197 VOP_UNLOCK(vp, 0);
2198 v_decr_usecount(vp);
2199 return;
2200 }
2201
2202 if (vp->v_usecount != 1) {
2203 vprint("vputx: negative ref count", vp);
2204 panic("vputx: negative ref cnt");
2205 }
2206 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2207 /*
2208 * We want to hold the vnode until the inactive finishes to
2209 * prevent vgone() races. We drop the use count here and the
2210 * hold count below when we're done.
2211 */
2212 v_decr_useonly(vp);
2213 /*
2214 * We must call VOP_INACTIVE with the node locked. Mark
2215 * as VI_DOINGINACT to avoid recursion.
2216 */
2217 vp->v_iflag |= VI_OWEINACT;
2218 switch (func) {
2219 case VPUTX_VRELE:
2220 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2221 VI_LOCK(vp);
2222 break;
2223 case VPUTX_VPUT:
2224 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2225 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2226 LK_NOWAIT);
2227 VI_LOCK(vp);
2228 }
2229 break;
2230 case VPUTX_VUNREF:
2231 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2232 error = EBUSY;
2233 break;
2234 }
2235 if (vp->v_usecount > 0)
2236 vp->v_iflag &= ~VI_OWEINACT;
2237 if (error == 0) {
2238 if (vp->v_iflag & VI_OWEINACT)
2239 vinactive(vp, curthread);
2240 if (func != VPUTX_VUNREF)
2241 VOP_UNLOCK(vp, 0);
2242 }
2243 vdropl(vp);
2244}
2245
2246/*
2247 * Vnode put/release.
2248 * If count drops to zero, call inactive routine and return to freelist.
2249 */
2250void
2251vrele(struct vnode *vp)
2252{
2253
2254 vputx(vp, VPUTX_VRELE);
2255}
2256
2257/*
2258 * Release an already locked vnode. This give the same effects as
2259 * unlock+vrele(), but takes less time and avoids releasing and
2260 * re-aquiring the lock (as vrele() acquires the lock internally.)
2261 */
2262void
2263vput(struct vnode *vp)
2264{
2265
2266 vputx(vp, VPUTX_VPUT);
2267}
2268
2269/*
2270 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2271 */
2272void
2273vunref(struct vnode *vp)
2274{
2275
2276 vputx(vp, VPUTX_VUNREF);
2277}
2278
2279/*
2280 * Somebody doesn't want the vnode recycled.
2281 */
2282void
2283vhold(struct vnode *vp)
2284{
2285
2286 VI_LOCK(vp);
2287 vholdl(vp);
2288 VI_UNLOCK(vp);
2289}
2290
2291/*
2292 * Increase the hold count and activate if this is the first reference.
2293 */
2294void
2295vholdl(struct vnode *vp)
2296{
2297 struct mount *mp;
2298
2299 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2300 vp->v_holdcnt++;
2301 if (!VSHOULDBUSY(vp))
2302 return;
2303 ASSERT_VI_LOCKED(vp, "vholdl");
2304 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2305 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2306 /*
2307 * Remove a vnode from the free list, mark it as in use,
2308 * and put it on the active list.
2309 */
2310 mtx_lock(&vnode_free_list_mtx);
2311 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2312 freevnodes--;
2313 vp->v_iflag &= ~(VI_FREE|VI_AGE);
2314 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2315 ("Activating already active vnode"));
2316 vp->v_iflag |= VI_ACTIVE;
2317 mp = vp->v_mount;
2318 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2319 mp->mnt_activevnodelistsize++;
2320 mtx_unlock(&vnode_free_list_mtx);
2321}
2322
2323/*
2324 * Note that there is one less who cares about this vnode.
2325 * vdrop() is the opposite of vhold().
2326 */
2327void
2328vdrop(struct vnode *vp)
2329{
2330
2331 VI_LOCK(vp);
2332 vdropl(vp);
2333}
2334
2335/*
2336 * Drop the hold count of the vnode. If this is the last reference to
2337 * the vnode we place it on the free list unless it has been vgone'd
2338 * (marked VI_DOOMED) in which case we will free it.
2339 */
2340void
2341vdropl(struct vnode *vp)
2342{
2343 struct bufobj *bo;
2344 struct mount *mp;
2345 int active;
2346
2347 ASSERT_VI_LOCKED(vp, "vdropl");
2348 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2349 if (vp->v_holdcnt <= 0)
2350 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2351 vp->v_holdcnt--;
2352 if (vp->v_holdcnt > 0) {
2353 VI_UNLOCK(vp);
2354 return;
2355 }
2356 if ((vp->v_iflag & VI_DOOMED) == 0) {
2357 /*
2358 * Mark a vnode as free: remove it from its active list
2359 * and put it up for recycling on the freelist.
2360 */
2361 VNASSERT(vp->v_op != NULL, vp,
2362 ("vdropl: vnode already reclaimed."));
2363 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2364 ("vnode already free"));
2365 VNASSERT(VSHOULDFREE(vp), vp,
2366 ("vdropl: freeing when we shouldn't"));
2367 active = vp->v_iflag & VI_ACTIVE;
2368 vp->v_iflag &= ~VI_ACTIVE;
2369 mp = vp->v_mount;
2370 mtx_lock(&vnode_free_list_mtx);
2371 if (active) {
2372 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2373 v_actfreelist);
2374 mp->mnt_activevnodelistsize--;
2375 }
2376 if (vp->v_iflag & VI_AGE) {
2377 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2378 } else {
2379 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2380 }
2381 freevnodes++;
2382 vp->v_iflag &= ~VI_AGE;
2383 vp->v_iflag |= VI_FREE;
2384 mtx_unlock(&vnode_free_list_mtx);
2385 VI_UNLOCK(vp);
2386 return;
2387 }
2388 /*
2389 * The vnode has been marked for destruction, so free it.
2390 */
2391 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2392 mtx_lock(&vnode_free_list_mtx);
2393 numvnodes--;
2394 mtx_unlock(&vnode_free_list_mtx);
2395 bo = &vp->v_bufobj;
2396 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2397 ("cleaned vnode still on the free list."));
2398 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2399 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2400 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2401 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2402 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2403 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2404 VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2405 ("clean blk trie not empty"));
2406 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2407 VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2408 ("dirty blk trie not empty"));
2409 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2410 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2411 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2412 VI_UNLOCK(vp);
2413#ifdef MAC
2414 mac_vnode_destroy(vp);
2415#endif
2416 if (vp->v_pollinfo != NULL)
2417 destroy_vpollinfo(vp->v_pollinfo);
2418#ifdef INVARIANTS
2419 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2420 vp->v_op = NULL;
2421#endif
2422 rangelock_destroy(&vp->v_rl);
2423 lockdestroy(vp->v_vnlock);
2424 mtx_destroy(&vp->v_interlock);
2425 mtx_destroy(BO_MTX(bo));
2426 uma_zfree(vnode_zone, vp);
2427}
2428
2429/*
2430 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2431 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2432 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2433 * failed lock upgrade.
2434 */
2435void
2436vinactive(struct vnode *vp, struct thread *td)
2437{
2438 struct vm_object *obj;
2439
2440 ASSERT_VOP_ELOCKED(vp, "vinactive");
2441 ASSERT_VI_LOCKED(vp, "vinactive");
2442 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2443 ("vinactive: recursed on VI_DOINGINACT"));
2444 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2445 vp->v_iflag |= VI_DOINGINACT;
2446 vp->v_iflag &= ~VI_OWEINACT;
2447 VI_UNLOCK(vp);
2448 /*
2449 * Before moving off the active list, we must be sure that any
2450 * modified pages are on the vnode's dirty list since these will
2451 * no longer be checked once the vnode is on the inactive list.
2452 * Because the vnode vm object keeps a hold reference on the vnode
2453 * if there is at least one resident non-cached page, the vnode
2454 * cannot leave the active list without the page cleanup done.
2455 */
2456 obj = vp->v_object;
2457 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2458 VM_OBJECT_WLOCK(obj);
2459 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2460 VM_OBJECT_WUNLOCK(obj);
2461 }
2462 VOP_INACTIVE(vp, td);
2463 VI_LOCK(vp);
2464 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2465 ("vinactive: lost VI_DOINGINACT"));
2466 vp->v_iflag &= ~VI_DOINGINACT;
2467}
2468
2469/*
2470 * Remove any vnodes in the vnode table belonging to mount point mp.
2471 *
2472 * If FORCECLOSE is not specified, there should not be any active ones,
2473 * return error if any are found (nb: this is a user error, not a
2474 * system error). If FORCECLOSE is specified, detach any active vnodes
2475 * that are found.
2476 *
2477 * If WRITECLOSE is set, only flush out regular file vnodes open for
2478 * writing.
2479 *
2480 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2481 *
2482 * `rootrefs' specifies the base reference count for the root vnode
2483 * of this filesystem. The root vnode is considered busy if its
2484 * v_usecount exceeds this value. On a successful return, vflush(, td)
2485 * will call vrele() on the root vnode exactly rootrefs times.
2486 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2487 * be zero.
2488 */
2489#ifdef DIAGNOSTIC
2490static int busyprt = 0; /* print out busy vnodes */
2491SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2492#endif
2493
2494int
2495vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2496{
2497 struct vnode *vp, *mvp, *rootvp = NULL;
2498 struct vattr vattr;
2499 int busy = 0, error;
2500
2501 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2502 rootrefs, flags);
2503 if (rootrefs > 0) {
2504 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2505 ("vflush: bad args"));
2506 /*
2507 * Get the filesystem root vnode. We can vput() it
2508 * immediately, since with rootrefs > 0, it won't go away.
2509 */
2510 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2511 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2512 __func__, error);
2513 return (error);
2514 }
2515 vput(rootvp);
2516 }
2517loop:
2518 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2519 vholdl(vp);
2520 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2521 if (error) {
2522 vdrop(vp);
2523 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2524 goto loop;
2525 }
2526 /*
2527 * Skip over a vnodes marked VV_SYSTEM.
2528 */
2529 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2530 VOP_UNLOCK(vp, 0);
2531 vdrop(vp);
2532 continue;
2533 }
2534 /*
2535 * If WRITECLOSE is set, flush out unlinked but still open
2536 * files (even if open only for reading) and regular file
2537 * vnodes open for writing.
2538 */
2539 if (flags & WRITECLOSE) {
2540 if (vp->v_object != NULL) {
2541 VM_OBJECT_WLOCK(vp->v_object);
2542 vm_object_page_clean(vp->v_object, 0, 0, 0);
2543 VM_OBJECT_WUNLOCK(vp->v_object);
2544 }
2545 error = VOP_FSYNC(vp, MNT_WAIT, td);
2546 if (error != 0) {
2547 VOP_UNLOCK(vp, 0);
2548 vdrop(vp);
2549 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2550 return (error);
2551 }
2552 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2553 VI_LOCK(vp);
2554
2555 if ((vp->v_type == VNON ||
2556 (error == 0 && vattr.va_nlink > 0)) &&
2557 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2558 VOP_UNLOCK(vp, 0);
2559 vdropl(vp);
2560 continue;
2561 }
2562 } else
2563 VI_LOCK(vp);
2564 /*
2565 * With v_usecount == 0, all we need to do is clear out the
2566 * vnode data structures and we are done.
2567 *
2568 * If FORCECLOSE is set, forcibly close the vnode.
2569 */
2570 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2571 VNASSERT(vp->v_usecount == 0 ||
2572 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2573 ("device VNODE %p is FORCECLOSED", vp));
2574 vgonel(vp);
2575 } else {
2576 busy++;
2577#ifdef DIAGNOSTIC
2578 if (busyprt)
2579 vprint("vflush: busy vnode", vp);
2580#endif
2581 }
2582 VOP_UNLOCK(vp, 0);
2583 vdropl(vp);
2584 }
2585 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2586 /*
2587 * If just the root vnode is busy, and if its refcount
2588 * is equal to `rootrefs', then go ahead and kill it.
2589 */
2590 VI_LOCK(rootvp);
2591 KASSERT(busy > 0, ("vflush: not busy"));
2592 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2593 ("vflush: usecount %d < rootrefs %d",
2594 rootvp->v_usecount, rootrefs));
2595 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2596 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2597 vgone(rootvp);
2598 VOP_UNLOCK(rootvp, 0);
2599 busy = 0;
2600 } else
2601 VI_UNLOCK(rootvp);
2602 }
2603 if (busy) {
2604 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2605 busy);
2606 return (EBUSY);
2607 }
2608 for (; rootrefs > 0; rootrefs--)
2609 vrele(rootvp);
2610 return (0);
2611}
2612
2613/*
2614 * Recycle an unused vnode to the front of the free list.
2615 */
2616int
2617vrecycle(struct vnode *vp)
2618{
2619 int recycled;
2620
2621 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2622 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2623 recycled = 0;
2624 VI_LOCK(vp);
2625 if (vp->v_usecount == 0) {
2626 recycled = 1;
2627 vgonel(vp);
2628 }
2629 VI_UNLOCK(vp);
2630 return (recycled);
2631}
2632
2633/*
2634 * Eliminate all activity associated with a vnode
2635 * in preparation for reuse.
2636 */
2637void
2638vgone(struct vnode *vp)
2639{
2640 VI_LOCK(vp);
2641 vgonel(vp);
2642 VI_UNLOCK(vp);
2643}
2644
2645static void
2646notify_lowervp_vfs_dummy(struct mount *mp __unused,
2647 struct vnode *lowervp __unused)
2648{
2649}
2650
2651/*
2652 * Notify upper mounts about reclaimed or unlinked vnode.
2653 */
2654void
2655vfs_notify_upper(struct vnode *vp, int event)
2656{
2657 static struct vfsops vgonel_vfsops = {
2658 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2659 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2660 };
2661 struct mount *mp, *ump, *mmp;
2662
2663 mp = vp->v_mount;
2664 if (mp == NULL)
2665 return;
2666
2667 MNT_ILOCK(mp);
2668 if (TAILQ_EMPTY(&mp->mnt_uppers))
2669 goto unlock;
2670 MNT_IUNLOCK(mp);
2671 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2672 mmp->mnt_op = &vgonel_vfsops;
2673 mmp->mnt_kern_flag |= MNTK_MARKER;
2674 MNT_ILOCK(mp);
2675 mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2676 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2677 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2678 ump = TAILQ_NEXT(ump, mnt_upper_link);
2679 continue;
2680 }
2681 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2682 MNT_IUNLOCK(mp);
2683 switch (event) {
2684 case VFS_NOTIFY_UPPER_RECLAIM:
2685 VFS_RECLAIM_LOWERVP(ump, vp);
2686 break;
2687 case VFS_NOTIFY_UPPER_UNLINK:
2688 VFS_UNLINK_LOWERVP(ump, vp);
2689 break;
2690 default:
2691 KASSERT(0, ("invalid event %d", event));
2692 break;
2693 }
2694 MNT_ILOCK(mp);
2695 ump = TAILQ_NEXT(mmp, mnt_upper_link);
2696 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2697 }
2698 free(mmp, M_TEMP);
2699 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2700 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2701 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2702 wakeup(&mp->mnt_uppers);
2703 }
2704unlock:
2705 MNT_IUNLOCK(mp);
2706}
2707
2708/*
2709 * vgone, with the vp interlock held.
2710 */
2711void
2712vgonel(struct vnode *vp)
2713{
2714 struct thread *td;
2715 int oweinact;
2716 int active;
2717 struct mount *mp;
2718
2719 ASSERT_VOP_ELOCKED(vp, "vgonel");
2720 ASSERT_VI_LOCKED(vp, "vgonel");
2721 VNASSERT(vp->v_holdcnt, vp,
2722 ("vgonel: vp %p has no reference.", vp));
2723 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2724 td = curthread;
2725
2726 /*
2727 * Don't vgonel if we're already doomed.
2728 */
2729 if (vp->v_iflag & VI_DOOMED)
2730 return;
2731 vp->v_iflag |= VI_DOOMED;
2732
2733 /*
2734 * Check to see if the vnode is in use. If so, we have to call
2735 * VOP_CLOSE() and VOP_INACTIVE().
2736 */
2737 active = vp->v_usecount;
2738 oweinact = (vp->v_iflag & VI_OWEINACT);
2739 VI_UNLOCK(vp);
2740 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2741
2742 /*
2743 * Clean out any buffers associated with the vnode.
2744 * If the flush fails, just toss the buffers.
2745 */
2746 mp = NULL;
2747 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2748 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2749 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2750 vinvalbuf(vp, 0, 0, 0);
2751
2752 /*
2753 * If purging an active vnode, it must be closed and
2754 * deactivated before being reclaimed.
2755 */
2756 if (active)
2757 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2758 if (oweinact || active) {
2759 VI_LOCK(vp);
2760 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2761 vinactive(vp, td);
2762 VI_UNLOCK(vp);
2763 }
2764 if (vp->v_type == VSOCK)
2765 vfs_unp_reclaim(vp);
2766 /*
2767 * Reclaim the vnode.
2768 */
2769 if (VOP_RECLAIM(vp, td))
2770 panic("vgone: cannot reclaim");
2771 if (mp != NULL)
2772 vn_finished_secondary_write(mp);
2773 VNASSERT(vp->v_object == NULL, vp,
2774 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2775 /*
2776 * Clear the advisory locks and wake up waiting threads.
2777 */
2778 (void)VOP_ADVLOCKPURGE(vp);
2779 /*
2780 * Delete from old mount point vnode list.
2781 */
2782 delmntque(vp);
2783 cache_purge(vp);
2784 /*
2785 * Done with purge, reset to the standard lock and invalidate
2786 * the vnode.
2787 */
2788 VI_LOCK(vp);
2789 vp->v_vnlock = &vp->v_lock;
2790 vp->v_op = &dead_vnodeops;
2791 vp->v_tag = "none";
2792 vp->v_type = VBAD;
2793}
2794
2795/*
2796 * Calculate the total number of references to a special device.
2797 */
2798int
2799vcount(struct vnode *vp)
2800{
2801 int count;
2802
2803 dev_lock();
2804 count = vp->v_rdev->si_usecount;
2805 dev_unlock();
2806 return (count);
2807}
2808
2809/*
2810 * Same as above, but using the struct cdev *as argument
2811 */
2812int
2813count_dev(struct cdev *dev)
2814{
2815 int count;
2816
2817 dev_lock();
2818 count = dev->si_usecount;
2819 dev_unlock();
2820 return(count);
2821}
2822
2823/*
2824 * Print out a description of a vnode.
2825 */
2826static char *typename[] =
2827{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2828 "VMARKER"};
2829
2830void
2831vn_printf(struct vnode *vp, const char *fmt, ...)
2832{
2833 va_list ap;
2834 char buf[256], buf2[16];
2835 u_long flags;
2836
2837 va_start(ap, fmt);
2838 vprintf(fmt, ap);
2839 va_end(ap);
2840 printf("%p: ", (void *)vp);
2841 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2842 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2843 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2844 buf[0] = '\0';
2845 buf[1] = '\0';
2846 if (vp->v_vflag & VV_ROOT)
2847 strlcat(buf, "|VV_ROOT", sizeof(buf));
2848 if (vp->v_vflag & VV_ISTTY)
2849 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2850 if (vp->v_vflag & VV_NOSYNC)
2851 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2852 if (vp->v_vflag & VV_ETERNALDEV)
2853 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
2854 if (vp->v_vflag & VV_CACHEDLABEL)
2855 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2856 if (vp->v_vflag & VV_TEXT)
2857 strlcat(buf, "|VV_TEXT", sizeof(buf));
2858 if (vp->v_vflag & VV_COPYONWRITE)
2859 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2860 if (vp->v_vflag & VV_SYSTEM)
2861 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2862 if (vp->v_vflag & VV_PROCDEP)
2863 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2864 if (vp->v_vflag & VV_NOKNOTE)
2865 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2866 if (vp->v_vflag & VV_DELETED)
2867 strlcat(buf, "|VV_DELETED", sizeof(buf));
2868 if (vp->v_vflag & VV_MD)
2869 strlcat(buf, "|VV_MD", sizeof(buf));
2870 if (vp->v_vflag & VV_FORCEINSMQ)
2871 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
2872 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
2873 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2874 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
2875 if (flags != 0) {
2876 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2877 strlcat(buf, buf2, sizeof(buf));
2878 }
2879 if (vp->v_iflag & VI_MOUNT)
2880 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2881 if (vp->v_iflag & VI_AGE)
2882 strlcat(buf, "|VI_AGE", sizeof(buf));
2883 if (vp->v_iflag & VI_DOOMED)
2884 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2885 if (vp->v_iflag & VI_FREE)
2886 strlcat(buf, "|VI_FREE", sizeof(buf));
2887 if (vp->v_iflag & VI_ACTIVE)
2888 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
2889 if (vp->v_iflag & VI_DOINGINACT)
2890 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2891 if (vp->v_iflag & VI_OWEINACT)
2892 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2893 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2894 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
2895 if (flags != 0) {
2896 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2897 strlcat(buf, buf2, sizeof(buf));
2898 }
2899 printf(" flags (%s)\n", buf + 1);
2900 if (mtx_owned(VI_MTX(vp)))
2901 printf(" VI_LOCKed");
2902 if (vp->v_object != NULL)
2903 printf(" v_object %p ref %d pages %d\n",
2904 vp->v_object, vp->v_object->ref_count,
2905 vp->v_object->resident_page_count);
2906 printf(" ");
2907 lockmgr_printinfo(vp->v_vnlock);
2908 if (vp->v_data != NULL)
2909 VOP_PRINT(vp);
2910}
2911
2912#ifdef DDB
2913/*
2914 * List all of the locked vnodes in the system.
2915 * Called when debugging the kernel.
2916 */
2917DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2918{
2919 struct mount *mp, *nmp;
2920 struct vnode *vp;
2921
2922 /*
2923 * Note: because this is DDB, we can't obey the locking semantics
2924 * for these structures, which means we could catch an inconsistent
2925 * state and dereference a nasty pointer. Not much to be done
2926 * about that.
2927 */
2928 db_printf("Locked vnodes\n");
2929 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2930 nmp = TAILQ_NEXT(mp, mnt_list);
2931 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2932 if (vp->v_type != VMARKER &&
2933 VOP_ISLOCKED(vp))
2934 vprint("", vp);
2935 }
2936 nmp = TAILQ_NEXT(mp, mnt_list);
2937 }
2938}
2939
2940/*
2941 * Show details about the given vnode.
2942 */
2943DB_SHOW_COMMAND(vnode, db_show_vnode)
2944{
2945 struct vnode *vp;
2946
2947 if (!have_addr)
2948 return;
2949 vp = (struct vnode *)addr;
2950 vn_printf(vp, "vnode ");
2951}
2952
2953/*
2954 * Show details about the given mount point.
2955 */
2956DB_SHOW_COMMAND(mount, db_show_mount)
2957{
2958 struct mount *mp;
2959 struct vfsopt *opt;
2960 struct statfs *sp;
2961 struct vnode *vp;
2962 char buf[512];
2963 uint64_t mflags;
2964 u_int flags;
2965
2966 if (!have_addr) {
2967 /* No address given, print short info about all mount points. */
2968 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2969 db_printf("%p %s on %s (%s)\n", mp,
2970 mp->mnt_stat.f_mntfromname,
2971 mp->mnt_stat.f_mntonname,
2972 mp->mnt_stat.f_fstypename);
2973 if (db_pager_quit)
2974 break;
2975 }
2976 db_printf("\nMore info: show mount <addr>\n");
2977 return;
2978 }
2979
2980 mp = (struct mount *)addr;
2981 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2982 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2983
2984 buf[0] = '\0';
2985 mflags = mp->mnt_flag;
2986#define MNT_FLAG(flag) do { \
2987 if (mflags & (flag)) { \
2988 if (buf[0] != '\0') \
2989 strlcat(buf, ", ", sizeof(buf)); \
2990 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2991 mflags &= ~(flag); \
2992 } \
2993} while (0)
2994 MNT_FLAG(MNT_RDONLY);
2995 MNT_FLAG(MNT_SYNCHRONOUS);
2996 MNT_FLAG(MNT_NOEXEC);
2997 MNT_FLAG(MNT_NOSUID);
2998 MNT_FLAG(MNT_NFS4ACLS);
2999 MNT_FLAG(MNT_UNION);
3000 MNT_FLAG(MNT_ASYNC);
3001 MNT_FLAG(MNT_SUIDDIR);
3002 MNT_FLAG(MNT_SOFTDEP);
3003 MNT_FLAG(MNT_NOSYMFOLLOW);
3004 MNT_FLAG(MNT_GJOURNAL);
3005 MNT_FLAG(MNT_MULTILABEL);
3006 MNT_FLAG(MNT_ACLS);
3007 MNT_FLAG(MNT_NOATIME);
3008 MNT_FLAG(MNT_NOCLUSTERR);
3009 MNT_FLAG(MNT_NOCLUSTERW);
3010 MNT_FLAG(MNT_SUJ);
3011 MNT_FLAG(MNT_EXRDONLY);
3012 MNT_FLAG(MNT_EXPORTED);
3013 MNT_FLAG(MNT_DEFEXPORTED);
3014 MNT_FLAG(MNT_EXPORTANON);
3015 MNT_FLAG(MNT_EXKERB);
3016 MNT_FLAG(MNT_EXPUBLIC);
3017 MNT_FLAG(MNT_LOCAL);
3018 MNT_FLAG(MNT_QUOTA);
3019 MNT_FLAG(MNT_ROOTFS);
3020 MNT_FLAG(MNT_USER);
3021 MNT_FLAG(MNT_IGNORE);
3022 MNT_FLAG(MNT_UPDATE);
3023 MNT_FLAG(MNT_DELEXPORT);
3024 MNT_FLAG(MNT_RELOAD);
3025 MNT_FLAG(MNT_FORCE);
3026 MNT_FLAG(MNT_SNAPSHOT);
3027 MNT_FLAG(MNT_BYFSID);
3028#undef MNT_FLAG
3029 if (mflags != 0) {
3030 if (buf[0] != '\0')
3031 strlcat(buf, ", ", sizeof(buf));
3032 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3033 "0x%016jx", mflags);
3034 }
3035 db_printf(" mnt_flag = %s\n", buf);
3036
3037 buf[0] = '\0';
3038 flags = mp->mnt_kern_flag;
3039#define MNT_KERN_FLAG(flag) do { \
3040 if (flags & (flag)) { \
3041 if (buf[0] != '\0') \
3042 strlcat(buf, ", ", sizeof(buf)); \
3043 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3044 flags &= ~(flag); \
3045 } \
3046} while (0)
3047 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3048 MNT_KERN_FLAG(MNTK_ASYNC);
3049 MNT_KERN_FLAG(MNTK_SOFTDEP);
3050 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3051 MNT_KERN_FLAG(MNTK_DRAINING);
3052 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3053 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3054 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3055 MNT_KERN_FLAG(MNTK_NO_IOPF);
3056 MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3057 MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3058 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3059 MNT_KERN_FLAG(MNTK_MARKER);
3060 MNT_KERN_FLAG(MNTK_NOASYNC);
3061 MNT_KERN_FLAG(MNTK_UNMOUNT);
3062 MNT_KERN_FLAG(MNTK_MWAIT);
3063 MNT_KERN_FLAG(MNTK_SUSPEND);
3064 MNT_KERN_FLAG(MNTK_SUSPEND2);
3065 MNT_KERN_FLAG(MNTK_SUSPENDED);
3066 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3067 MNT_KERN_FLAG(MNTK_NOKNOTE);
3068#undef MNT_KERN_FLAG
3069 if (flags != 0) {
3070 if (buf[0] != '\0')
3071 strlcat(buf, ", ", sizeof(buf));
3072 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3073 "0x%08x", flags);
3074 }
3075 db_printf(" mnt_kern_flag = %s\n", buf);
3076
3077 db_printf(" mnt_opt = ");
3078 opt = TAILQ_FIRST(mp->mnt_opt);
3079 if (opt != NULL) {
3080 db_printf("%s", opt->name);
3081 opt = TAILQ_NEXT(opt, link);
3082 while (opt != NULL) {
3083 db_printf(", %s", opt->name);
3084 opt = TAILQ_NEXT(opt, link);
3085 }
3086 }
3087 db_printf("\n");
3088
3089 sp = &mp->mnt_stat;
3090 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3091 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3092 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3093 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3094 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3095 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3096 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3097 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3098 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3099 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3100 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3101 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3102
3103 db_printf(" mnt_cred = { uid=%u ruid=%u",
3104 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3105 if (jailed(mp->mnt_cred))
3106 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3107 db_printf(" }\n");
3108 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3109 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3110 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3111 db_printf(" mnt_activevnodelistsize = %d\n",
3112 mp->mnt_activevnodelistsize);
3113 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3114 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3115 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3116 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3117 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3118 db_printf(" mnt_secondary_accwrites = %d\n",
3119 mp->mnt_secondary_accwrites);
3120 db_printf(" mnt_gjprovider = %s\n",
3121 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3122
3123 db_printf("\n\nList of active vnodes\n");
3124 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3125 if (vp->v_type != VMARKER) {
3126 vn_printf(vp, "vnode ");
3127 if (db_pager_quit)
3128 break;
3129 }
3130 }
3131 db_printf("\n\nList of inactive vnodes\n");
3132 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3133 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3134 vn_printf(vp, "vnode ");
3135 if (db_pager_quit)
3136 break;
3137 }
3138 }
3139}
3140#endif /* DDB */
3141
3142/*
3143 * Fill in a struct xvfsconf based on a struct vfsconf.
3144 */
3145static int
3146vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3147{
3148 struct xvfsconf xvfsp;
3149
3150 bzero(&xvfsp, sizeof(xvfsp));
3151 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3152 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3153 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3154 xvfsp.vfc_flags = vfsp->vfc_flags;
3155 /*
3156 * These are unused in userland, we keep them
3157 * to not break binary compatibility.
3158 */
3159 xvfsp.vfc_vfsops = NULL;
3160 xvfsp.vfc_next = NULL;
3161 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3162}
3163
3164#ifdef COMPAT_FREEBSD32
3165struct xvfsconf32 {
3166 uint32_t vfc_vfsops;
3167 char vfc_name[MFSNAMELEN];
3168 int32_t vfc_typenum;
3169 int32_t vfc_refcount;
3170 int32_t vfc_flags;
3171 uint32_t vfc_next;
3172};
3173
3174static int
3175vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3176{
3177 struct xvfsconf32 xvfsp;
3178
3179 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3180 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3181 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3182 xvfsp.vfc_flags = vfsp->vfc_flags;
3183 xvfsp.vfc_vfsops = 0;
3184 xvfsp.vfc_next = 0;
3185 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3186}
3187#endif
3188
3189/*
3190 * Top level filesystem related information gathering.
3191 */
3192static int
3193sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3194{
3195 struct vfsconf *vfsp;
3196 int error;
3197
3198 error = 0;
3199 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3200#ifdef COMPAT_FREEBSD32
3201 if (req->flags & SCTL_MASK32)
3202 error = vfsconf2x32(req, vfsp);
3203 else
3204#endif
3205 error = vfsconf2x(req, vfsp);
3206 if (error)
3207 break;
3208 }
3209 return (error);
3210}
3211
3212SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3213 NULL, 0, sysctl_vfs_conflist,
3214 "S,xvfsconf", "List of all configured filesystems");
3215
3216#ifndef BURN_BRIDGES
3217static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3218
3219static int
3220vfs_sysctl(SYSCTL_HANDLER_ARGS)
3221{
3222 int *name = (int *)arg1 - 1; /* XXX */
3223 u_int namelen = arg2 + 1; /* XXX */
3224 struct vfsconf *vfsp;
3225
3226 log(LOG_WARNING, "userland calling deprecated sysctl, "
3227 "please rebuild world\n");
3228
3229#if 1 || defined(COMPAT_PRELITE2)
3230 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3231 if (namelen == 1)
3232 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3233#endif
3234
3235 switch (name[1]) {
3236 case VFS_MAXTYPENUM:
3237 if (namelen != 2)
3238 return (ENOTDIR);
3239 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3240 case VFS_CONF:
3241 if (namelen != 3)
3242 return (ENOTDIR); /* overloaded */
3243 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3244 if (vfsp->vfc_typenum == name[2])
3245 break;
3246 if (vfsp == NULL)
3247 return (EOPNOTSUPP);
3248#ifdef COMPAT_FREEBSD32
3249 if (req->flags & SCTL_MASK32)
3250 return (vfsconf2x32(req, vfsp));
3251 else
3252#endif
3253 return (vfsconf2x(req, vfsp));
3254 }
3255 return (EOPNOTSUPP);
3256}
3257
3258static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3259 vfs_sysctl, "Generic filesystem");
3260
3261#if 1 || defined(COMPAT_PRELITE2)
3262
3263static int
3264sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3265{
3266 int error;
3267 struct vfsconf *vfsp;
3268 struct ovfsconf ovfs;
3269
3270 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3271 bzero(&ovfs, sizeof(ovfs));
3272 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3273 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3274 ovfs.vfc_index = vfsp->vfc_typenum;
3275 ovfs.vfc_refcount = vfsp->vfc_refcount;
3276 ovfs.vfc_flags = vfsp->vfc_flags;
3277 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3278 if (error)
3279 return error;
3280 }
3281 return 0;
3282}
3283
3284#endif /* 1 || COMPAT_PRELITE2 */
3285#endif /* !BURN_BRIDGES */
3286
3287#define KINFO_VNODESLOP 10
3288#ifdef notyet
3289/*
3290 * Dump vnode list (via sysctl).
3291 */
3292/* ARGSUSED */
3293static int
3294sysctl_vnode(SYSCTL_HANDLER_ARGS)
3295{
3296 struct xvnode *xvn;
3297 struct mount *mp;
3298 struct vnode *vp;
3299 int error, len, n;
3300
3301 /*
3302 * Stale numvnodes access is not fatal here.
3303 */
3304 req->lock = 0;
3305 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3306 if (!req->oldptr)
3307 /* Make an estimate */
3308 return (SYSCTL_OUT(req, 0, len));
3309
3310 error = sysctl_wire_old_buffer(req, 0);
3311 if (error != 0)
3312 return (error);
3313 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3314 n = 0;
3315 mtx_lock(&mountlist_mtx);
3316 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3317 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3318 continue;
3319 MNT_ILOCK(mp);
3320 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3321 if (n == len)
3322 break;
3323 vref(vp);
3324 xvn[n].xv_size = sizeof *xvn;
3325 xvn[n].xv_vnode = vp;
3326 xvn[n].xv_id = 0; /* XXX compat */
3327#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3328 XV_COPY(usecount);
3329 XV_COPY(writecount);
3330 XV_COPY(holdcnt);
3331 XV_COPY(mount);
3332 XV_COPY(numoutput);
3333 XV_COPY(type);
3334#undef XV_COPY
3335 xvn[n].xv_flag = vp->v_vflag;
3336
3337 switch (vp->v_type) {
3338 case VREG:
3339 case VDIR:
3340 case VLNK:
3341 break;
3342 case VBLK:
3343 case VCHR:
3344 if (vp->v_rdev == NULL) {
3345 vrele(vp);
3346 continue;
3347 }
3348 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3349 break;
3350 case VSOCK:
3351 xvn[n].xv_socket = vp->v_socket;
3352 break;
3353 case VFIFO:
3354 xvn[n].xv_fifo = vp->v_fifoinfo;
3355 break;
3356 case VNON:
3357 case VBAD:
3358 default:
3359 /* shouldn't happen? */
3360 vrele(vp);
3361 continue;
3362 }
3363 vrele(vp);
3364 ++n;
3365 }
3366 MNT_IUNLOCK(mp);
3367 mtx_lock(&mountlist_mtx);
3368 vfs_unbusy(mp);
3369 if (n == len)
3370 break;
3371 }
3372 mtx_unlock(&mountlist_mtx);
3373
3374 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3375 free(xvn, M_TEMP);
3376 return (error);
3377}
3378
3379SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3380 0, 0, sysctl_vnode, "S,xvnode", "");
3381#endif
3382
3383/*
3384 * Unmount all filesystems. The list is traversed in reverse order
3385 * of mounting to avoid dependencies.
3386 */
3387void
3388vfs_unmountall(void)
3389{
3390 struct mount *mp;
3391 struct thread *td;
3392 int error;
3393
3394 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3395 td = curthread;
3396
3397 /*
3398 * Since this only runs when rebooting, it is not interlocked.
3399 */
3400 while(!TAILQ_EMPTY(&mountlist)) {
3401 mp = TAILQ_LAST(&mountlist, mntlist);
3402 error = dounmount(mp, MNT_FORCE, td);
3403 if (error) {
3404 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3405 /*
3406 * XXX: Due to the way in which we mount the root
3407 * file system off of devfs, devfs will generate a
3408 * "busy" warning when we try to unmount it before
3409 * the root. Don't print a warning as a result in
3410 * order to avoid false positive errors that may
3411 * cause needless upset.
3412 */
3413 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3414 printf("unmount of %s failed (",
3415 mp->mnt_stat.f_mntonname);
3416 if (error == EBUSY)
3417 printf("BUSY)\n");
3418 else
3419 printf("%d)\n", error);
3420 }
3421 } else {
3422 /* The unmount has removed mp from the mountlist */
3423 }
3424 }
3425}
3426
3427/*
3428 * perform msync on all vnodes under a mount point
3429 * the mount point must be locked.
3430 */
3431void
3432vfs_msync(struct mount *mp, int flags)
3433{
3434 struct vnode *vp, *mvp;
3435 struct vm_object *obj;
3436
3437 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3438 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3439 obj = vp->v_object;
3440 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3441 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3442 if (!vget(vp,
3443 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3444 curthread)) {
3445 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3446 vput(vp);
3447 continue;
3448 }
3449
3450 obj = vp->v_object;
3451 if (obj != NULL) {
3452 VM_OBJECT_WLOCK(obj);
3453 vm_object_page_clean(obj, 0, 0,
3454 flags == MNT_WAIT ?
3455 OBJPC_SYNC : OBJPC_NOSYNC);
3456 VM_OBJECT_WUNLOCK(obj);
3457 }
3458 vput(vp);
3459 }
3460 } else
3461 VI_UNLOCK(vp);
3462 }
3463}
3464
3465static void
3466destroy_vpollinfo(struct vpollinfo *vi)
3467{
3468 seldrain(&vi->vpi_selinfo);
3469 knlist_destroy(&vi->vpi_selinfo.si_note);
3470 mtx_destroy(&vi->vpi_lock);
3471 uma_zfree(vnodepoll_zone, vi);
3472}
3473
3474/*
3475 * Initalize per-vnode helper structure to hold poll-related state.
3476 */
3477void
3478v_addpollinfo(struct vnode *vp)
3479{
3480 struct vpollinfo *vi;
3481
3482 if (vp->v_pollinfo != NULL)
3483 return;
3484 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3485 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3486 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3487 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3488 VI_LOCK(vp);
3489 if (vp->v_pollinfo != NULL) {
3490 VI_UNLOCK(vp);
3491 destroy_vpollinfo(vi);
3492 return;
3493 }
3494 vp->v_pollinfo = vi;
3495 VI_UNLOCK(vp);
3496}
3497
3498/*
3499 * Record a process's interest in events which might happen to
3500 * a vnode. Because poll uses the historic select-style interface
3501 * internally, this routine serves as both the ``check for any
3502 * pending events'' and the ``record my interest in future events''
3503 * functions. (These are done together, while the lock is held,
3504 * to avoid race conditions.)
3505 */
3506int
3507vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3508{
3509
3510 v_addpollinfo(vp);
3511 mtx_lock(&vp->v_pollinfo->vpi_lock);
3512 if (vp->v_pollinfo->vpi_revents & events) {
3513 /*
3514 * This leaves events we are not interested
3515 * in available for the other process which
3516 * which presumably had requested them
3517 * (otherwise they would never have been
3518 * recorded).
3519 */
3520 events &= vp->v_pollinfo->vpi_revents;
3521 vp->v_pollinfo->vpi_revents &= ~events;
3522
3523 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3524 return (events);
3525 }
3526 vp->v_pollinfo->vpi_events |= events;
3527 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3528 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3529 return (0);
3530}
3531
3532/*
3533 * Routine to create and manage a filesystem syncer vnode.
3534 */
3535#define sync_close ((int (*)(struct vop_close_args *))nullop)
3536static int sync_fsync(struct vop_fsync_args *);
3537static int sync_inactive(struct vop_inactive_args *);
3538static int sync_reclaim(struct vop_reclaim_args *);
3539
3540static struct vop_vector sync_vnodeops = {
3541 .vop_bypass = VOP_EOPNOTSUPP,
3542 .vop_close = sync_close, /* close */
3543 .vop_fsync = sync_fsync, /* fsync */
3544 .vop_inactive = sync_inactive, /* inactive */
3545 .vop_reclaim = sync_reclaim, /* reclaim */
3546 .vop_lock1 = vop_stdlock, /* lock */
3547 .vop_unlock = vop_stdunlock, /* unlock */
3548 .vop_islocked = vop_stdislocked, /* islocked */
3549};
3550
3551/*
3552 * Create a new filesystem syncer vnode for the specified mount point.
3553 */
3554void
3555vfs_allocate_syncvnode(struct mount *mp)
3556{
3557 struct vnode *vp;
3558 struct bufobj *bo;
3559 static long start, incr, next;
3560 int error;
3561
3562 /* Allocate a new vnode */
3563 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3564 if (error != 0)
3565 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3566 vp->v_type = VNON;
3567 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3568 vp->v_vflag |= VV_FORCEINSMQ;
3569 error = insmntque(vp, mp);
3570 if (error != 0)
3571 panic("vfs_allocate_syncvnode: insmntque() failed");
3572 vp->v_vflag &= ~VV_FORCEINSMQ;
3573 VOP_UNLOCK(vp, 0);
3574 /*
3575 * Place the vnode onto the syncer worklist. We attempt to
3576 * scatter them about on the list so that they will go off
3577 * at evenly distributed times even if all the filesystems
3578 * are mounted at once.
3579 */
3580 next += incr;
3581 if (next == 0 || next > syncer_maxdelay) {
3582 start /= 2;
3583 incr /= 2;
3584 if (start == 0) {
3585 start = syncer_maxdelay / 2;
3586 incr = syncer_maxdelay;
3587 }
3588 next = start;
3589 }
3590 bo = &vp->v_bufobj;
3591 BO_LOCK(bo);
3592 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3593 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3594 mtx_lock(&sync_mtx);
3595 sync_vnode_count++;
3596 if (mp->mnt_syncer == NULL) {
3597 mp->mnt_syncer = vp;
3598 vp = NULL;
3599 }
3600 mtx_unlock(&sync_mtx);
3601 BO_UNLOCK(bo);
3602 if (vp != NULL) {
3603 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3604 vgone(vp);
3605 vput(vp);
3606 }
3607}
3608
3609void
3610vfs_deallocate_syncvnode(struct mount *mp)
3611{
3612 struct vnode *vp;
3613
3614 mtx_lock(&sync_mtx);
3615 vp = mp->mnt_syncer;
3616 if (vp != NULL)
3617 mp->mnt_syncer = NULL;
3618 mtx_unlock(&sync_mtx);
3619 if (vp != NULL)
3620 vrele(vp);
3621}
3622
3623/*
3624 * Do a lazy sync of the filesystem.
3625 */
3626static int
3627sync_fsync(struct vop_fsync_args *ap)
3628{
3629 struct vnode *syncvp = ap->a_vp;
3630 struct mount *mp = syncvp->v_mount;
3631 int error, save;
3632 struct bufobj *bo;
3633
3634 /*
3635 * We only need to do something if this is a lazy evaluation.
3636 */
3637 if (ap->a_waitfor != MNT_LAZY)
3638 return (0);
3639
3640 /*
3641 * Move ourselves to the back of the sync list.
3642 */
3643 bo = &syncvp->v_bufobj;
3644 BO_LOCK(bo);
3645 vn_syncer_add_to_worklist(bo, syncdelay);
3646 BO_UNLOCK(bo);
3647
3648 /*
3649 * Walk the list of vnodes pushing all that are dirty and
3650 * not already on the sync list.
3651 */
3652 mtx_lock(&mountlist_mtx);
3653 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3654 mtx_unlock(&mountlist_mtx);
3655 return (0);
3656 }
3657 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3658 vfs_unbusy(mp);
3659 return (0);
3660 }
3661 save = curthread_pflags_set(TDP_SYNCIO);
3662 vfs_msync(mp, MNT_NOWAIT);
3663 error = VFS_SYNC(mp, MNT_LAZY);
3664 curthread_pflags_restore(save);
3665 vn_finished_write(mp);
3666 vfs_unbusy(mp);
3667 return (error);
3668}
3669
3670/*
3671 * The syncer vnode is no referenced.
3672 */
3673static int
3674sync_inactive(struct vop_inactive_args *ap)
3675{
3676
3677 vgone(ap->a_vp);
3678 return (0);
3679}
3680
3681/*
3682 * The syncer vnode is no longer needed and is being decommissioned.
3683 *
3684 * Modifications to the worklist must be protected by sync_mtx.
3685 */
3686static int
3687sync_reclaim(struct vop_reclaim_args *ap)
3688{
3689 struct vnode *vp = ap->a_vp;
3690 struct bufobj *bo;
3691
3692 bo = &vp->v_bufobj;
3693 BO_LOCK(bo);
3694 mtx_lock(&sync_mtx);
3695 if (vp->v_mount->mnt_syncer == vp)
3696 vp->v_mount->mnt_syncer = NULL;
3697 if (bo->bo_flag & BO_ONWORKLST) {
3698 LIST_REMOVE(bo, bo_synclist);
3699 syncer_worklist_len--;
3700 sync_vnode_count--;
3701 bo->bo_flag &= ~BO_ONWORKLST;
3702 }
3703 mtx_unlock(&sync_mtx);
3704 BO_UNLOCK(bo);
3705
3706 return (0);
3707}
3708
3709/*
3710 * Check if vnode represents a disk device
3711 */
3712int
3713vn_isdisk(struct vnode *vp, int *errp)
3714{
3715 int error;
3716
3717 error = 0;
3718 dev_lock();
3719 if (vp->v_type != VCHR)
3720 error = ENOTBLK;
3721 else if (vp->v_rdev == NULL)
3722 error = ENXIO;
3723 else if (vp->v_rdev->si_devsw == NULL)
3724 error = ENXIO;
3725 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3726 error = ENOTBLK;
3727 dev_unlock();
3728 if (errp != NULL)
3729 *errp = error;
3730 return (error == 0);
3731}
3732
3733/*
3734 * Common filesystem object access control check routine. Accepts a
3735 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3736 * and optional call-by-reference privused argument allowing vaccess()
3737 * to indicate to the caller whether privilege was used to satisfy the
3738 * request (obsoleted). Returns 0 on success, or an errno on failure.
3739 */
3740int
3741vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3742 accmode_t accmode, struct ucred *cred, int *privused)
3743{
3744 accmode_t dac_granted;
3745 accmode_t priv_granted;
3746
3747 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3748 ("invalid bit in accmode"));
3749 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3750 ("VAPPEND without VWRITE"));
3751
3752 /*
3753 * Look for a normal, non-privileged way to access the file/directory
3754 * as requested. If it exists, go with that.
3755 */
3756
3757 if (privused != NULL)
3758 *privused = 0;
3759
3760 dac_granted = 0;
3761
3762 /* Check the owner. */
3763 if (cred->cr_uid == file_uid) {
3764 dac_granted |= VADMIN;
3765 if (file_mode & S_IXUSR)
3766 dac_granted |= VEXEC;
3767 if (file_mode & S_IRUSR)
3768 dac_granted |= VREAD;
3769 if (file_mode & S_IWUSR)
3770 dac_granted |= (VWRITE | VAPPEND);
3771
3772 if ((accmode & dac_granted) == accmode)
3773 return (0);
3774
3775 goto privcheck;
3776 }
3777
3778 /* Otherwise, check the groups (first match) */
3779 if (groupmember(file_gid, cred)) {
3780 if (file_mode & S_IXGRP)
3781 dac_granted |= VEXEC;
3782 if (file_mode & S_IRGRP)
3783 dac_granted |= VREAD;
3784 if (file_mode & S_IWGRP)
3785 dac_granted |= (VWRITE | VAPPEND);
3786
3787 if ((accmode & dac_granted) == accmode)
3788 return (0);
3789
3790 goto privcheck;
3791 }
3792
3793 /* Otherwise, check everyone else. */
3794 if (file_mode & S_IXOTH)
3795 dac_granted |= VEXEC;
3796 if (file_mode & S_IROTH)
3797 dac_granted |= VREAD;
3798 if (file_mode & S_IWOTH)
3799 dac_granted |= (VWRITE | VAPPEND);
3800 if ((accmode & dac_granted) == accmode)
3801 return (0);
3802
3803privcheck:
3804 /*
3805 * Build a privilege mask to determine if the set of privileges
3806 * satisfies the requirements when combined with the granted mask
3807 * from above. For each privilege, if the privilege is required,
3808 * bitwise or the request type onto the priv_granted mask.
3809 */
3810 priv_granted = 0;
3811
3812 if (type == VDIR) {
3813 /*
3814 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3815 * requests, instead of PRIV_VFS_EXEC.
3816 */
3817 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3818 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3819 priv_granted |= VEXEC;
3820 } else {
3821 /*
3822 * Ensure that at least one execute bit is on. Otherwise,
3823 * a privileged user will always succeed, and we don't want
3824 * this to happen unless the file really is executable.
3825 */
3826 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3827 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3828 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3829 priv_granted |= VEXEC;
3830 }
3831
3832 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3833 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3834 priv_granted |= VREAD;
3835
3836 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3837 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3838 priv_granted |= (VWRITE | VAPPEND);
3839
3840 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3841 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3842 priv_granted |= VADMIN;
3843
3844 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3845 /* XXX audit: privilege used */
3846 if (privused != NULL)
3847 *privused = 1;
3848 return (0);
3849 }
3850
3851 return ((accmode & VADMIN) ? EPERM : EACCES);
3852}
3853
3854/*
3855 * Credential check based on process requesting service, and per-attribute
3856 * permissions.
3857 */
3858int
3859extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3860 struct thread *td, accmode_t accmode)
3861{
3862
3863 /*
3864 * Kernel-invoked always succeeds.
3865 */
3866 if (cred == NOCRED)
3867 return (0);
3868
3869 /*
3870 * Do not allow privileged processes in jail to directly manipulate
3871 * system attributes.
3872 */
3873 switch (attrnamespace) {
3874 case EXTATTR_NAMESPACE_SYSTEM:
3875 /* Potentially should be: return (EPERM); */
3876 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3877 case EXTATTR_NAMESPACE_USER:
3878 return (VOP_ACCESS(vp, accmode, cred, td));
3879 default:
3880 return (EPERM);
3881 }
3882}
3883
3884#ifdef DEBUG_VFS_LOCKS
3885/*
3886 * This only exists to supress warnings from unlocked specfs accesses. It is
3887 * no longer ok to have an unlocked VFS.
3888 */
3889#define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3890 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3891
3892int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3893SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3894 "Drop into debugger on lock violation");
3895
3896int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3897SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3898 0, "Check for interlock across VOPs");
3899
3900int vfs_badlock_print = 1; /* Print lock violations. */
3901SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3902 0, "Print lock violations");
3903
3904#ifdef KDB
3905int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3906SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3907 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3908#endif
3909
3910static void
3911vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3912{
3913
3914#ifdef KDB
3915 if (vfs_badlock_backtrace)
3916 kdb_backtrace();
3917#endif
3918 if (vfs_badlock_print)
3919 printf("%s: %p %s\n", str, (void *)vp, msg);
3920 if (vfs_badlock_ddb)
3921 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3922}
3923
3924void
3925assert_vi_locked(struct vnode *vp, const char *str)
3926{
3927
3928 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3929 vfs_badlock("interlock is not locked but should be", str, vp);
3930}
3931
3932void
3933assert_vi_unlocked(struct vnode *vp, const char *str)
3934{
3935
3936 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3937 vfs_badlock("interlock is locked but should not be", str, vp);
3938}
3939
3940void
3941assert_vop_locked(struct vnode *vp, const char *str)
3942{
3943 int locked;
3944
3945 if (!IGNORE_LOCK(vp)) {
3946 locked = VOP_ISLOCKED(vp);
3947 if (locked == 0 || locked == LK_EXCLOTHER)
3948 vfs_badlock("is not locked but should be", str, vp);
3949 }
3950}
3951
3952void
3953assert_vop_unlocked(struct vnode *vp, const char *str)
3954{
3955
3956 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3957 vfs_badlock("is locked but should not be", str, vp);
3958}
3959
3960void
3961assert_vop_elocked(struct vnode *vp, const char *str)
3962{
3963
3964 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3965 vfs_badlock("is not exclusive locked but should be", str, vp);
3966}
3967
3968#if 0
3969void
3970assert_vop_elocked_other(struct vnode *vp, const char *str)
3971{
3972
3973 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3974 vfs_badlock("is not exclusive locked by another thread",
3975 str, vp);
3976}
3977
3978void
3979assert_vop_slocked(struct vnode *vp, const char *str)
3980{
3981
3982 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3983 vfs_badlock("is not locked shared but should be", str, vp);
3984}
3985#endif /* 0 */
3986#endif /* DEBUG_VFS_LOCKS */
3987
3988void
3989vop_rename_fail(struct vop_rename_args *ap)
3990{
3991
3992 if (ap->a_tvp != NULL)
3993 vput(ap->a_tvp);
3994 if (ap->a_tdvp == ap->a_tvp)
3995 vrele(ap->a_tdvp);
3996 else
3997 vput(ap->a_tdvp);
3998 vrele(ap->a_fdvp);
3999 vrele(ap->a_fvp);
4000}
4001
4002void
4003vop_rename_pre(void *ap)
4004{
4005 struct vop_rename_args *a = ap;
4006
4007#ifdef DEBUG_VFS_LOCKS
4008 if (a->a_tvp)
4009 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4010 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4011 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4012 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4013
4014 /* Check the source (from). */
4015 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4016 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4017 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4018 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4019 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4020
4021 /* Check the target. */
4022 if (a->a_tvp)
4023 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4024 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4025#endif
4026 if (a->a_tdvp != a->a_fdvp)
4027 vhold(a->a_fdvp);
4028 if (a->a_tvp != a->a_fvp)
4029 vhold(a->a_fvp);
4030 vhold(a->a_tdvp);
4031 if (a->a_tvp)
4032 vhold(a->a_tvp);
4033}
4034
4035void
4036vop_strategy_pre(void *ap)
4037{
4038#ifdef DEBUG_VFS_LOCKS
4039 struct vop_strategy_args *a;
4040 struct buf *bp;
4041
4042 a = ap;
4043 bp = a->a_bp;
4044
4045 /*
4046 * Cluster ops lock their component buffers but not the IO container.
4047 */
4048 if ((bp->b_flags & B_CLUSTER) != 0)
4049 return;
4050
4051 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4052 if (vfs_badlock_print)
4053 printf(
4054 "VOP_STRATEGY: bp is not locked but should be\n");
4055 if (vfs_badlock_ddb)
4056 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4057 }
4058#endif
4059}
4060
4061void
4062vop_lock_pre(void *ap)
4063{
4064#ifdef DEBUG_VFS_LOCKS
4065 struct vop_lock1_args *a = ap;
4066
4067 if ((a->a_flags & LK_INTERLOCK) == 0)
4068 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4069 else
4070 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4071#endif
4072}
4073
4074void
4075vop_lock_post(void *ap, int rc)
4076{
4077#ifdef DEBUG_VFS_LOCKS
4078 struct vop_lock1_args *a = ap;
4079
4080 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4081 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4082 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4083#endif
4084}
4085
4086void
4087vop_unlock_pre(void *ap)
4088{
4089#ifdef DEBUG_VFS_LOCKS
4090 struct vop_unlock_args *a = ap;
4091
4092 if (a->a_flags & LK_INTERLOCK)
4093 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4094 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4095#endif
4096}
4097
4098void
4099vop_unlock_post(void *ap, int rc)
4100{
4101#ifdef DEBUG_VFS_LOCKS
4102 struct vop_unlock_args *a = ap;
4103
4104 if (a->a_flags & LK_INTERLOCK)
4105 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4106#endif
4107}
4108
4109void
4110vop_create_post(void *ap, int rc)
4111{
4112 struct vop_create_args *a = ap;
4113
4114 if (!rc)
4115 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4116}
4117
4118void
4119vop_deleteextattr_post(void *ap, int rc)
4120{
4121 struct vop_deleteextattr_args *a = ap;
4122
4123 if (!rc)
4124 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4125}
4126
4127void
4128vop_link_post(void *ap, int rc)
4129{
4130 struct vop_link_args *a = ap;
4131
4132 if (!rc) {
4133 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4134 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4135 }
4136}
4137
4138void
4139vop_mkdir_post(void *ap, int rc)
4140{
4141 struct vop_mkdir_args *a = ap;
4142
4143 if (!rc)
4144 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4145}
4146
4147void
4148vop_mknod_post(void *ap, int rc)
4149{
4150 struct vop_mknod_args *a = ap;
4151
4152 if (!rc)
4153 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4154}
4155
4156void
4157vop_remove_post(void *ap, int rc)
4158{
4159 struct vop_remove_args *a = ap;
4160
4161 if (!rc) {
4162 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4163 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4164 }
4165}
4166
4167void
4168vop_rename_post(void *ap, int rc)
4169{
4170 struct vop_rename_args *a = ap;
4171
4172 if (!rc) {
4173 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4174 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4175 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4176 if (a->a_tvp)
4177 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4178 }
4179 if (a->a_tdvp != a->a_fdvp)
4180 vdrop(a->a_fdvp);
4181 if (a->a_tvp != a->a_fvp)
4182 vdrop(a->a_fvp);
4183 vdrop(a->a_tdvp);
4184 if (a->a_tvp)
4185 vdrop(a->a_tvp);
4186}
4187
4188void
4189vop_rmdir_post(void *ap, int rc)
4190{
4191 struct vop_rmdir_args *a = ap;
4192
4193 if (!rc) {
4194 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4195 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4196 }
4197}
4198
4199void
4200vop_setattr_post(void *ap, int rc)
4201{
4202 struct vop_setattr_args *a = ap;
4203
4204 if (!rc)
4205 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4206}
4207
4208void
4209vop_setextattr_post(void *ap, int rc)
4210{
4211 struct vop_setextattr_args *a = ap;
4212
4213 if (!rc)
4214 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4215}
4216
4217void
4218vop_symlink_post(void *ap, int rc)
4219{
4220 struct vop_symlink_args *a = ap;
4221
4222 if (!rc)
4223 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4224}
4225
4226static struct knlist fs_knlist;
4227
4228static void
4229vfs_event_init(void *arg)
4230{
4231 knlist_init_mtx(&fs_knlist, NULL);
4232}
4233/* XXX - correct order? */
4234SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4235
4236void
4237vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4238{
4239
4240 KNOTE_UNLOCKED(&fs_knlist, event);
4241}
4242
4243static int filt_fsattach(struct knote *kn);
4244static void filt_fsdetach(struct knote *kn);
4245static int filt_fsevent(struct knote *kn, long hint);
4246
4247struct filterops fs_filtops = {
4248 .f_isfd = 0,
4249 .f_attach = filt_fsattach,
4250 .f_detach = filt_fsdetach,
4251 .f_event = filt_fsevent
4252};
4253
4254static int
4255filt_fsattach(struct knote *kn)
4256{
4257
4258 kn->kn_flags |= EV_CLEAR;
4259 knlist_add(&fs_knlist, kn, 0);
4260 return (0);
4261}
4262
4263static void
4264filt_fsdetach(struct knote *kn)
4265{
4266
4267 knlist_remove(&fs_knlist, kn, 0);
4268}
4269
4270static int
4271filt_fsevent(struct knote *kn, long hint)
4272{
4273
4274 kn->kn_fflags |= hint;
4275 return (kn->kn_fflags != 0);
4276}
4277
4278static int
4279sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4280{
4281 struct vfsidctl vc;
4282 int error;
4283 struct mount *mp;
4284
4285 error = SYSCTL_IN(req, &vc, sizeof(vc));
4286 if (error)
4287 return (error);
4288 if (vc.vc_vers != VFS_CTL_VERS1)
4289 return (EINVAL);
4290 mp = vfs_getvfs(&vc.vc_fsid);
4291 if (mp == NULL)
4292 return (ENOENT);
4293 /* ensure that a specific sysctl goes to the right filesystem. */
4294 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4295 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4296 vfs_rel(mp);
4297 return (EINVAL);
4298 }
4299 VCTLTOREQ(&vc, req);
4300 error = VFS_SYSCTL(mp, vc.vc_op, req);
4301 vfs_rel(mp);
4302 return (error);
4303}
4304
4305SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4306 NULL, 0, sysctl_vfs_ctl, "",
4307 "Sysctl by fsid");
4308
4309/*
4310 * Function to initialize a va_filerev field sensibly.
4311 * XXX: Wouldn't a random number make a lot more sense ??
4312 */
4313u_quad_t
4314init_va_filerev(void)
4315{
4316 struct bintime bt;
4317
4318 getbinuptime(&bt);
4319 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4320}
4321
4322static int filt_vfsread(struct knote *kn, long hint);
4323static int filt_vfswrite(struct knote *kn, long hint);
4324static int filt_vfsvnode(struct knote *kn, long hint);
4325static void filt_vfsdetach(struct knote *kn);
4326static struct filterops vfsread_filtops = {
4327 .f_isfd = 1,
4328 .f_detach = filt_vfsdetach,
4329 .f_event = filt_vfsread
4330};
4331static struct filterops vfswrite_filtops = {
4332 .f_isfd = 1,
4333 .f_detach = filt_vfsdetach,
4334 .f_event = filt_vfswrite
4335};
4336static struct filterops vfsvnode_filtops = {
4337 .f_isfd = 1,
4338 .f_detach = filt_vfsdetach,
4339 .f_event = filt_vfsvnode
4340};
4341
4342static void
4343vfs_knllock(void *arg)
4344{
4345 struct vnode *vp = arg;
4346
4347 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4348}
4349
4350static void
4351vfs_knlunlock(void *arg)
4352{
4353 struct vnode *vp = arg;
4354
4355 VOP_UNLOCK(vp, 0);
4356}
4357
4358static void
4359vfs_knl_assert_locked(void *arg)
4360{
4361#ifdef DEBUG_VFS_LOCKS
4362 struct vnode *vp = arg;
4363
4364 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4365#endif
4366}
4367
4368static void
4369vfs_knl_assert_unlocked(void *arg)
4370{
4371#ifdef DEBUG_VFS_LOCKS
4372 struct vnode *vp = arg;
4373
4374 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4375#endif
4376}
4377
4378int
4379vfs_kqfilter(struct vop_kqfilter_args *ap)
4380{
4381 struct vnode *vp = ap->a_vp;
4382 struct knote *kn = ap->a_kn;
4383 struct knlist *knl;
4384
4385 switch (kn->kn_filter) {
4386 case EVFILT_READ:
4387 kn->kn_fop = &vfsread_filtops;
4388 break;
4389 case EVFILT_WRITE:
4390 kn->kn_fop = &vfswrite_filtops;
4391 break;
4392 case EVFILT_VNODE:
4393 kn->kn_fop = &vfsvnode_filtops;
4394 break;
4395 default:
4396 return (EINVAL);
4397 }
4398
4399 kn->kn_hook = (caddr_t)vp;
4400
4401 v_addpollinfo(vp);
4402 if (vp->v_pollinfo == NULL)
4403 return (ENOMEM);
4404 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4405 knlist_add(knl, kn, 0);
4406
4407 return (0);
4408}
4409
4410/*
4411 * Detach knote from vnode
4412 */
4413static void
4414filt_vfsdetach(struct knote *kn)
4415{
4416 struct vnode *vp = (struct vnode *)kn->kn_hook;
4417
4418 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4419 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4420}
4421
4422/*ARGSUSED*/
4423static int
4424filt_vfsread(struct knote *kn, long hint)
4425{
4426 struct vnode *vp = (struct vnode *)kn->kn_hook;
4427 struct vattr va;
4428 int res;
4429
4430 /*
4431 * filesystem is gone, so set the EOF flag and schedule
4432 * the knote for deletion.
4433 */
4434 if (hint == NOTE_REVOKE) {
4435 VI_LOCK(vp);
4436 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4437 VI_UNLOCK(vp);
4438 return (1);
4439 }
4440
4441 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4442 return (0);
4443
4444 VI_LOCK(vp);
4445 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4446 res = (kn->kn_data != 0);
4447 VI_UNLOCK(vp);
4448 return (res);
4449}
4450
4451/*ARGSUSED*/
4452static int
4453filt_vfswrite(struct knote *kn, long hint)
4454{
4455 struct vnode *vp = (struct vnode *)kn->kn_hook;
4456
4457 VI_LOCK(vp);
4458
4459 /*
4460 * filesystem is gone, so set the EOF flag and schedule
4461 * the knote for deletion.
4462 */
4463 if (hint == NOTE_REVOKE)
4464 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4465
4466 kn->kn_data = 0;
4467 VI_UNLOCK(vp);
4468 return (1);
4469}
4470
4471static int
4472filt_vfsvnode(struct knote *kn, long hint)
4473{
4474 struct vnode *vp = (struct vnode *)kn->kn_hook;
4475 int res;
4476
4477 VI_LOCK(vp);
4478 if (kn->kn_sfflags & hint)
4479 kn->kn_fflags |= hint;
4480 if (hint == NOTE_REVOKE) {
4481 kn->kn_flags |= EV_EOF;
4482 VI_UNLOCK(vp);
4483 return (1);
4484 }
4485 res = (kn->kn_fflags != 0);
4486 VI_UNLOCK(vp);
4487 return (res);
4488}
4489
4490int
4491vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4492{
4493 int error;
4494
4495 if (dp->d_reclen > ap->a_uio->uio_resid)
4496 return (ENAMETOOLONG);
4497 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4498 if (error) {
4499 if (ap->a_ncookies != NULL) {
4500 if (ap->a_cookies != NULL)
4501 free(ap->a_cookies, M_TEMP);
4502 ap->a_cookies = NULL;
4503 *ap->a_ncookies = 0;
4504 }
4505 return (error);
4506 }
4507 if (ap->a_ncookies == NULL)
4508 return (0);
4509
4510 KASSERT(ap->a_cookies,
4511 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4512
4513 *ap->a_cookies = realloc(*ap->a_cookies,
4514 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4515 (*ap->a_cookies)[*ap->a_ncookies] = off;
4516 return (0);
4517}
4518
4519/*
4520 * Mark for update the access time of the file if the filesystem
4521 * supports VOP_MARKATIME. This functionality is used by execve and
4522 * mmap, so we want to avoid the I/O implied by directly setting
4523 * va_atime for the sake of efficiency.
4524 */
4525void
4526vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4527{
4528 struct mount *mp;
4529
4530 mp = vp->v_mount;
4531 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4532 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4533 (void)VOP_MARKATIME(vp);
4534}
4535
4536/*
4537 * The purpose of this routine is to remove granularity from accmode_t,
4538 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4539 * VADMIN and VAPPEND.
4540 *
4541 * If it returns 0, the caller is supposed to continue with the usual
4542 * access checks using 'accmode' as modified by this routine. If it
4543 * returns nonzero value, the caller is supposed to return that value
4544 * as errno.
4545 *
4546 * Note that after this routine runs, accmode may be zero.
4547 */
4548int
4549vfs_unixify_accmode(accmode_t *accmode)
4550{
4551 /*
4552 * There is no way to specify explicit "deny" rule using
4553 * file mode or POSIX.1e ACLs.
4554 */
4555 if (*accmode & VEXPLICIT_DENY) {
4556 *accmode = 0;
4557 return (0);
4558 }
4559
4560 /*
4561 * None of these can be translated into usual access bits.
4562 * Also, the common case for NFSv4 ACLs is to not contain
4563 * either of these bits. Caller should check for VWRITE
4564 * on the containing directory instead.
4565 */
4566 if (*accmode & (VDELETE_CHILD | VDELETE))
4567 return (EPERM);
4568
4569 if (*accmode & VADMIN_PERMS) {
4570 *accmode &= ~VADMIN_PERMS;
4571 *accmode |= VADMIN;
4572 }
4573
4574 /*
4575 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4576 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4577 */
4578 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4579
4580 return (0);
4581}
4582
4583/*
4584 * These are helper functions for filesystems to traverse all
4585 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4586 *
4587 * This interface replaces MNT_VNODE_FOREACH.
4588 */
4589
4590MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4591
4592struct vnode *
4593__mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4594{
4595 struct vnode *vp;
4596
4597 if (should_yield())
4598 kern_yield(PRI_USER);
4599 MNT_ILOCK(mp);
4600 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4601 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4602 while (vp != NULL && (vp->v_type == VMARKER ||
4603 (vp->v_iflag & VI_DOOMED) != 0))
4604 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4605
4606 /* Check if we are done */
4607 if (vp == NULL) {
4608 __mnt_vnode_markerfree_all(mvp, mp);
4609 /* MNT_IUNLOCK(mp); -- done in above function */
4610 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4611 return (NULL);
4612 }
4613 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4614 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4615 VI_LOCK(vp);
4616 MNT_IUNLOCK(mp);
4617 return (vp);
4618}
4619
4620struct vnode *
4621__mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4622{
4623 struct vnode *vp;
4624
4625 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4626 MNT_ILOCK(mp);
4627 MNT_REF(mp);
4628 (*mvp)->v_type = VMARKER;
4629
4630 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4631 while (vp != NULL && (vp->v_type == VMARKER ||
4632 (vp->v_iflag & VI_DOOMED) != 0))
4633 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4634
4635 /* Check if we are done */
4636 if (vp == NULL) {
4637 MNT_REL(mp);
4638 MNT_IUNLOCK(mp);
4639 free(*mvp, M_VNODE_MARKER);
4640 *mvp = NULL;
4641 return (NULL);
4642 }
4643 (*mvp)->v_mount = mp;
4644 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4645 VI_LOCK(vp);
4646 MNT_IUNLOCK(mp);
4647 return (vp);
4648}
4649
4650
4651void
4652__mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4653{
4654
4655 if (*mvp == NULL) {
4656 MNT_IUNLOCK(mp);
4657 return;
4658 }
4659
4660 mtx_assert(MNT_MTX(mp), MA_OWNED);
4661
4662 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4663 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4664 MNT_REL(mp);
4665 MNT_IUNLOCK(mp);
4666 free(*mvp, M_VNODE_MARKER);
4667 *mvp = NULL;
4668}
4669
4670/*
4671 * These are helper functions for filesystems to traverse their
4672 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4673 */
4674static void
4675mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4676{
4677
4678 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4679
4680 MNT_ILOCK(mp);
4681 MNT_REL(mp);
4682 MNT_IUNLOCK(mp);
4683 free(*mvp, M_VNODE_MARKER);
4684 *mvp = NULL;
4685}
4686
4687static struct vnode *
4688mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4689{
4690 struct vnode *vp, *nvp;
4691
4692 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4693 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4694restart:
4695 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4696 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4697 while (vp != NULL) {
4698 if (vp->v_type == VMARKER) {
4699 vp = TAILQ_NEXT(vp, v_actfreelist);
4700 continue;
4701 }
4702 if (!VI_TRYLOCK(vp)) {
4703 if (mp_ncpus == 1 || should_yield()) {
4704 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4705 mtx_unlock(&vnode_free_list_mtx);
4706 kern_yield(PRI_USER);
4707 mtx_lock(&vnode_free_list_mtx);
4708 goto restart;
4709 }
4710 continue;
4711 }
4712 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4713 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4714 ("alien vnode on the active list %p %p", vp, mp));
4715 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4716 break;
4717 nvp = TAILQ_NEXT(vp, v_actfreelist);
4718 VI_UNLOCK(vp);
4719 vp = nvp;
4720 }
4721
4722 /* Check if we are done */
4723 if (vp == NULL) {
4724 mtx_unlock(&vnode_free_list_mtx);
4725 mnt_vnode_markerfree_active(mvp, mp);
4726 return (NULL);
4727 }
4728 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4729 mtx_unlock(&vnode_free_list_mtx);
4730 ASSERT_VI_LOCKED(vp, "active iter");
4731 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4732 return (vp);
4733}
4734
4735struct vnode *
4736__mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4737{
4738
4739 if (should_yield())
4740 kern_yield(PRI_USER);
4741 mtx_lock(&vnode_free_list_mtx);
4742 return (mnt_vnode_next_active(mvp, mp));
4743}
4744
4745struct vnode *
4746__mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4747{
4748 struct vnode *vp;
4749
4750 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4751 MNT_ILOCK(mp);
4752 MNT_REF(mp);
4753 MNT_IUNLOCK(mp);
4754 (*mvp)->v_type = VMARKER;
4755 (*mvp)->v_mount = mp;
4756
4757 mtx_lock(&vnode_free_list_mtx);
4758 vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4759 if (vp == NULL) {
4760 mtx_unlock(&vnode_free_list_mtx);
4761 mnt_vnode_markerfree_active(mvp, mp);
4762 return (NULL);
4763 }
4764 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4765 return (mnt_vnode_next_active(mvp, mp));
4766}
4767
4768void
4769__mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4770{
4771
4772 if (*mvp == NULL)
4773 return;
4774
4775 mtx_lock(&vnode_free_list_mtx);
4776 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4777 mtx_unlock(&vnode_free_list_mtx);
4778 mnt_vnode_markerfree_active(mvp, mp);
4779}