vfs_subr.c revision 99220
1159967Sobrien/* 2159952Sobrien * Copyright (c) 1989, 1993 3159952Sobrien * The Regents of the University of California. All rights reserved. 4159952Sobrien * (c) UNIX System Laboratories, Inc. 5159952Sobrien * All or some portions of this file are derived from material licensed 6159952Sobrien * to the University of California by American Telephone and Telegraph 7159952Sobrien * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8159952Sobrien * the permission of UNIX System Laboratories, Inc. 9159952Sobrien * 10159952Sobrien * Redistribution and use in source and binary forms, with or without 11159952Sobrien * modification, are permitted provided that the following conditions 12159952Sobrien * are met: 13159952Sobrien * 1. Redistributions of source code must retain the above copyright 14159952Sobrien * notice, this list of conditions and the following disclaimer. 15159952Sobrien * 2. Redistributions in binary form must reproduce the above copyright 16159952Sobrien * notice, this list of conditions and the following disclaimer in the 17159967Sobrien * documentation and/or other materials provided with the distribution. 18159967Sobrien * 3. All advertising materials mentioning features or use of this software 19159952Sobrien * must display the following acknowledgement: 20159952Sobrien * This product includes software developed by the University of 21170589Syongari * California, Berkeley and its contributors. 22170589Syongari * 4. Neither the name of the University nor the names of its contributors 23163504Sobrien * may be used to endorse or promote products derived from this software 24159952Sobrien * without specific prior written permission. 25170589Syongari * 26170589Syongari * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27170589Syongari * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28170589Syongari * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29170589Syongari * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30170589Syongari * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31164650Sobrien * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32164650Sobrien * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33164650Sobrien * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34164650Sobrien * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35164650Sobrien * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36164650Sobrien * SUCH DAMAGE. 37159952Sobrien * 38170589Syongari * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 39170589Syongari * $FreeBSD: head/sys/kern/vfs_subr.c 99220 2002-07-01 17:59:40Z iedowse $ 40170589Syongari */ 41170589Syongari 42159952Sobrien/* 43170589Syongari * External virtual filesystem routines 44170595Syongari */ 45170604Syongari#include "opt_ddb.h" 46170589Syongari 47163504Sobrien#include <sys/param.h> 48163504Sobrien#include <sys/systm.h> 49163504Sobrien#include <sys/bio.h> 50163504Sobrien#include <sys/buf.h> 51170589Syongari#include <sys/conf.h> 52170589Syongari#include <sys/eventhandler.h> 53170589Syongari#include <sys/fcntl.h> 54170589Syongari#include <sys/kernel.h> 55163504Sobrien#include <sys/kthread.h> 56163504Sobrien#include <sys/malloc.h> 57163504Sobrien#include <sys/mount.h> 58163504Sobrien#include <sys/namei.h> 59163504Sobrien#include <sys/stat.h> 60163504Sobrien#include <sys/sysctl.h> 61163504Sobrien#include <sys/syslog.h> 62163504Sobrien#include <sys/vmmeter.h> 63163504Sobrien#include <sys/vnode.h> 64163504Sobrien 65163504Sobrien#include <vm/vm.h> 66163504Sobrien#include <vm/vm_object.h> 67163504Sobrien#include <vm/vm_extern.h> 68163504Sobrien#include <vm/pmap.h> 69163504Sobrien#include <vm/vm_map.h> 70163504Sobrien#include <vm/vm_page.h> 71163504Sobrien#include <vm/uma.h> 72163504Sobrien 73163504Sobrienstatic MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 74163504Sobrien 75163504Sobrienstatic void addalias(struct vnode *vp, dev_t nvp_rdev); 76163504Sobrienstatic void insmntque(struct vnode *vp, struct mount *mp); 77163504Sobrienstatic void vclean(struct vnode *vp, int flags, struct thread *td); 78163504Sobrienstatic void vlruvp(struct vnode *vp); 79163504Sobrien 80163504Sobrien/* 81163504Sobrien * Number of vnodes in existence. Increased whenever getnewvnode() 82163504Sobrien * allocates a new vnode, never decreased. 83170589Syongari */ 84163504Sobrienstatic unsigned long numvnodes; 85163504Sobrien 86163504SobrienSYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 87163504Sobrien 88163504Sobrien/* 89163504Sobrien * Conversion tables for conversion from vnode types to inode formats 90183561Syongari * and back. 91183561Syongari */ 92183561Syongarienum vtype iftovt_tab[16] = { 93163504Sobrien VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 94163504Sobrien VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 95163504Sobrien}; 96163504Sobrienint vttoif_tab[9] = { 97163504Sobrien 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 98183561Syongari S_IFSOCK, S_IFIFO, S_IFMT, 99183561Syongari}; 100183561Syongari 101183561Syongari/* 102183561Syongari * List of vnodes that are ready for recycling. 103183561Syongari */ 104183561Syongaristatic TAILQ_HEAD(freelst, vnode) vnode_free_list; 105183561Syongari 106183561Syongari/* 107183561Syongari * Minimum number of free vnodes. If there are fewer than this free vnodes, 108183561Syongari * getnewvnode() will return a newly allocated vnode. 109183561Syongari */ 110183561Syongaristatic u_long wantfreevnodes = 25; 111183561SyongariSYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 112183561Syongari/* Number of vnodes in the free list. */ 113183561Syongaristatic u_long freevnodes; 114183561SyongariSYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 115183561Syongari 116183561Syongari/* 117183561Syongari * Various variables used for debugging the new implementation of 118183561Syongari * reassignbuf(). 119183561Syongari * XXX these are probably of (very) limited utility now. 120183561Syongari */ 121183561Syongaristatic int reassignbufcalls; 122183561SyongariSYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 123183561Syongaristatic int reassignbufloops; 124183561SyongariSYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, ""); 125163504Sobrienstatic int reassignbufsortgood; 126170589SyongariSYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, ""); 127170589Syongaristatic int reassignbufsortbad; 128170589SyongariSYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, ""); 129170589Syongari/* Set to 0 for old insertion-sort based reassignbuf, 1 for modern method. */ 130159952Sobrienstatic int reassignbufmethod = 1; 131170589SyongariSYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, ""); 132170589Syongaristatic int nameileafonly; 133170589SyongariSYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, ""); 134170589Syongari 135163504Sobrien#ifdef ENABLE_VFS_IOOPT 136163504Sobrien/* See NOTES for a description of this setting. */ 137163504Sobrienint vfs_ioopt; 138163504SobrienSYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 139159952Sobrien#endif 140163504Sobrien 141159952Sobrien/* List of mounted filesystems. */ 142170589Syongaristruct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 143170589Syongari 144170589Syongari/* For any iteration/modification of mountlist */ 145163504Sobrienstruct mtx mountlist_mtx; 146163504Sobrien 147163504Sobrien/* For any iteration/modification of mnt_vnodelist */ 148163504Sobrienstruct mtx mntvnode_mtx; 149163504Sobrien 150163504Sobrien/* 151159952Sobrien * Cache for the mount type id assigned to NFS. This is used for 152163504Sobrien * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 153163504Sobrien */ 154163504Sobrienint nfs_mount_type = -1; 155163504Sobrien 156163504Sobrien/* To keep more than one thread at a time from running vfs_getnewfsid */ 157163504Sobrienstatic struct mtx mntid_mtx; 158163504Sobrien 159163504Sobrien/* For any iteration/modification of vnode_free_list */ 160163504Sobrienstatic struct mtx vnode_free_list_mtx; 161159952Sobrien 162163504Sobrien/* 163159952Sobrien * For any iteration/modification of dev->si_hlist (linked through 164159952Sobrien * v_specnext) 165159952Sobrien */ 166159952Sobrienstatic struct mtx spechash_mtx; 167163504Sobrien 168163504Sobrien/* Publicly exported FS */ 169163504Sobrienstruct nfs_public nfs_pub; 170163504Sobrien 171163504Sobrien/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 172163504Sobrienstatic uma_zone_t vnode_zone; 173163504Sobrienstatic uma_zone_t vnodepoll_zone; 174163504Sobrien 175163504Sobrien/* Set to 1 to print out reclaim of active vnodes */ 176170589Syongariint prtactive; 177170589Syongari 178170589Syongari/* 179170589Syongari * The workitem queue. 180170589Syongari * 181164656Sobrien * It is useful to delay writes of file data and filesystem metadata 182159952Sobrien * for tens of seconds so that quickly created and deleted files need 183159952Sobrien * not waste disk bandwidth being created and removed. To realize this, 184163504Sobrien * we append vnodes to a "workitem" queue. When running with a soft 185159952Sobrien * updates implementation, most pending metadata dependencies should 186163504Sobrien * not wait for more than a few seconds. Thus, mounted on block devices 187159952Sobrien * are delayed only about a half the time that file data is delayed. 188163504Sobrien * Similarly, directory updates are more critical, so are only delayed 189163504Sobrien * about a third the time that file data is delayed. Thus, there are 190159952Sobrien * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 191170589Syongari * one each second (driven off the filesystem syncer process). The 192170589Syongari * syncer_delayno variable indicates the next queue that is to be processed. 193215132Syongari * Items that need to be processed soon are placed in this queue: 194170589Syongari * 195163504Sobrien * syncer_workitem_pending[syncer_delayno] 196159952Sobrien * 197163504Sobrien * A delay of fifteen seconds is done by placing the request fifteen 198163504Sobrien * entries later in the queue: 199159952Sobrien * 200163504Sobrien * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 201163504Sobrien * 202163504Sobrien */ 203159952Sobrienstatic int syncer_delayno; 204163504Sobrienstatic long syncer_mask; 205170589SyongariLIST_HEAD(synclist, vnode); 206163504Sobrienstatic struct synclist *syncer_workitem_pending; 207159952Sobrien 208170589Syongari#define SYNCER_MAXDELAY 32 209170589Syongaristatic int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 210170589Syongaristatic int syncdelay = 30; /* max time to delay syncing data */ 211163504Sobrienstatic int filedelay = 30; /* time to delay syncing files */ 212163504SobrienSYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 213163504Sobrienstatic int dirdelay = 29; /* time to delay syncing directories */ 214163504SobrienSYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 215159952Sobrienstatic int metadelay = 28; /* time to delay syncing metadata */ 216183561SyongariSYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 217183561Syongaristatic int rushjob; /* number of slots to run ASAP */ 218183561Syongaristatic int stat_rush_requests; /* number of times I/O speeded up */ 219183561SyongariSYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 220170589Syongari 221170589Syongari/* 222170589Syongari * Number of vnodes we want to exist at any one time. This is mostly used 223170589Syongari * to size hash tables in vnode-related code. It is normally not used in 224170589Syongari * getnewvnode(), as wantfreevnodes is normally nonzero.) 225170589Syongari * 226170589Syongari * XXX desiredvnodes is historical cruft and should not exist. 227170589Syongari */ 228170589Syongariint desiredvnodes; 229170589SyongariSYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 230170589Syongari &desiredvnodes, 0, "Maximum number of vnodes"); 231170589Syongaristatic int minvnodes; 232170589SyongariSYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 233170589Syongari &minvnodes, 0, "Minimum number of vnodes"); 234170589Syongaristatic int vnlru_nowhere; 235159952SobrienSYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, 236159952Sobrien "Number of times the vnlru process ran without success"); 237159952Sobrien 238159952Sobrien/* Hook for calling soft updates */ 239159952Sobrienint (*softdep_process_worklist_hook)(struct mount *); 240163504Sobrien 241163504Sobrien#ifdef DEBUG_VFS_LOCKS 242163504Sobrien/* Print lock violations */ 243163504Sobrienint vfs_badlock_print = 1; 244163504Sobrien/* Panic on violation */ 245163504Sobrienint vfs_badlock_panic = 1; 246163504Sobrien#endif 247163504Sobrien 248159952Sobrienvoid 249159952Sobrienv_addpollinfo(struct vnode *vp) 250163504Sobrien{ 251159952Sobrien vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK); 252159952Sobrien mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 253159952Sobrien} 254159952Sobrien 255159952Sobrien/* 256159952Sobrien * Initialize the vnode management data structures. 257159952Sobrien */ 258163504Sobrienstatic void 259163504Sobrienvntblinit(void *dummy __unused) 260159952Sobrien{ 261159952Sobrien 262163504Sobrien desiredvnodes = maxproc + cnt.v_page_count / 4; 263163504Sobrien minvnodes = desiredvnodes / 4; 264163504Sobrien mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF); 265163504Sobrien mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF); 266163504Sobrien mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 267163504Sobrien mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF); 268163504Sobrien TAILQ_INIT(&vnode_free_list); 269163504Sobrien mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 270159952Sobrien vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 271159952Sobrien NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 272163504Sobrien vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 273159952Sobrien NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 274159952Sobrien /* 275170589Syongari * Initialize the filesystem syncer. 276170589Syongari */ 277159952Sobrien syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 278170589Syongari &syncer_mask); 279170589Syongari syncer_maxdelay = syncer_mask + 1; 280170589Syongari} 281163504SobrienSYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL) 282163504Sobrien 283170589Syongari 284170589Syongari/* 285163504Sobrien * Mark a mount point as busy. Used to synchronize access and to delay 286170589Syongari * unmounting. Interlock is not released on failure. 287170589Syongari */ 288163504Sobrienint 289159952Sobrienvfs_busy(mp, flags, interlkp, td) 290163504Sobrien struct mount *mp; 291170589Syongari int flags; 292159952Sobrien struct mtx *interlkp; 293163504Sobrien struct thread *td; 294170589Syongari{ 295159967Sobrien int lkflags; 296170589Syongari 297170589Syongari if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 298159967Sobrien if (flags & LK_NOWAIT) 299159967Sobrien return (ENOENT); 300159967Sobrien mp->mnt_kern_flag |= MNTK_MWAIT; 301159967Sobrien /* 302159967Sobrien * Since all busy locks are shared except the exclusive 303159967Sobrien * lock granted when unmounting, the only place that a 304159967Sobrien * wakeup needs to be done is at the release of the 305159967Sobrien * exclusive lock at the end of dounmount. 306159967Sobrien */ 307159967Sobrien msleep(mp, interlkp, PVFS, "vfs_busy", 0); 308159967Sobrien return (ENOENT); 309159967Sobrien } 310159967Sobrien lkflags = LK_SHARED | LK_NOPAUSE; 311159967Sobrien if (interlkp) 312159967Sobrien lkflags |= LK_INTERLOCK; 313159967Sobrien if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td)) 314159967Sobrien panic("vfs_busy: unexpected lock failure"); 315159967Sobrien return (0); 316159967Sobrien} 317162212Sobrien 318162212Sobrien/* 319162212Sobrien * Free a busy filesystem. 320162212Sobrien */ 321162212Sobrienvoid 322162212Sobrienvfs_unbusy(mp, td) 323162212Sobrien struct mount *mp; 324162212Sobrien struct thread *td; 325170589Syongari{ 326170589Syongari 327170589Syongari lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 328170589Syongari} 329178055Syongari 330178055Syongari/* 331178055Syongari * Lookup a filesystem type, and if found allocate and initialize 332178055Syongari * a mount structure for it. 333183509Syongari * 334183509Syongari * Devname is usually updated by mount(8) after booting. 335183509Syongari */ 336183509Syongariint 337183509Syongarivfs_rootmountalloc(fstypename, devname, mpp) 338183509Syongari char *fstypename; 339183509Syongari char *devname; 340183509Syongari struct mount **mpp; 341159967Sobrien{ 342159967Sobrien struct thread *td = curthread; /* XXX */ 343159967Sobrien struct vfsconf *vfsp; 344159967Sobrien struct mount *mp; 345159967Sobrien 346159967Sobrien if (fstypename == NULL) 347159967Sobrien return (ENODEV); 348159967Sobrien for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 349159967Sobrien if (!strcmp(vfsp->vfc_name, fstypename)) 350163504Sobrien break; 351163504Sobrien if (vfsp == NULL) 352163504Sobrien return (ENODEV); 353163504Sobrien mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 354163504Sobrien lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); 355163504Sobrien (void)vfs_busy(mp, LK_NOWAIT, 0, td); 356163504Sobrien TAILQ_INIT(&mp->mnt_nvnodelist); 357163504Sobrien TAILQ_INIT(&mp->mnt_reservedvnlist); 358163504Sobrien mp->mnt_vfc = vfsp; 359163504Sobrien mp->mnt_op = vfsp->vfc_vfsops; 360163504Sobrien mp->mnt_flag = MNT_RDONLY; 361 mp->mnt_vnodecovered = NULLVP; 362 vfsp->vfc_refcount++; 363 mp->mnt_iosize_max = DFLTPHYS; 364 mp->mnt_stat.f_type = vfsp->vfc_typenum; 365 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 366 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 367 mp->mnt_stat.f_mntonname[0] = '/'; 368 mp->mnt_stat.f_mntonname[1] = 0; 369 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 370 *mpp = mp; 371 return (0); 372} 373 374/* 375 * Find an appropriate filesystem to use for the root. If a filesystem 376 * has not been preselected, walk through the list of known filesystems 377 * trying those that have mountroot routines, and try them until one 378 * works or we have tried them all. 379 */ 380#ifdef notdef /* XXX JH */ 381int 382lite2_vfs_mountroot() 383{ 384 struct vfsconf *vfsp; 385 extern int (*lite2_mountroot)(void); 386 int error; 387 388 if (lite2_mountroot != NULL) 389 return ((*lite2_mountroot)()); 390 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 391 if (vfsp->vfc_mountroot == NULL) 392 continue; 393 if ((error = (*vfsp->vfc_mountroot)()) == 0) 394 return (0); 395 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 396 } 397 return (ENODEV); 398} 399#endif 400 401/* 402 * Lookup a mount point by filesystem identifier. 403 */ 404struct mount * 405vfs_getvfs(fsid) 406 fsid_t *fsid; 407{ 408 register struct mount *mp; 409 410 mtx_lock(&mountlist_mtx); 411 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 412 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 413 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 414 mtx_unlock(&mountlist_mtx); 415 return (mp); 416 } 417 } 418 mtx_unlock(&mountlist_mtx); 419 return ((struct mount *) 0); 420} 421 422/* 423 * Get a new unique fsid. Try to make its val[0] unique, since this value 424 * will be used to create fake device numbers for stat(). Also try (but 425 * not so hard) make its val[0] unique mod 2^16, since some emulators only 426 * support 16-bit device numbers. We end up with unique val[0]'s for the 427 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 428 * 429 * Keep in mind that several mounts may be running in parallel. Starting 430 * the search one past where the previous search terminated is both a 431 * micro-optimization and a defense against returning the same fsid to 432 * different mounts. 433 */ 434void 435vfs_getnewfsid(mp) 436 struct mount *mp; 437{ 438 static u_int16_t mntid_base; 439 fsid_t tfsid; 440 int mtype; 441 442 mtx_lock(&mntid_mtx); 443 mtype = mp->mnt_vfc->vfc_typenum; 444 tfsid.val[1] = mtype; 445 mtype = (mtype & 0xFF) << 24; 446 for (;;) { 447 tfsid.val[0] = makeudev(255, 448 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 449 mntid_base++; 450 if (vfs_getvfs(&tfsid) == NULL) 451 break; 452 } 453 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 454 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 455 mtx_unlock(&mntid_mtx); 456} 457 458/* 459 * Knob to control the precision of file timestamps: 460 * 461 * 0 = seconds only; nanoseconds zeroed. 462 * 1 = seconds and nanoseconds, accurate within 1/HZ. 463 * 2 = seconds and nanoseconds, truncated to microseconds. 464 * >=3 = seconds and nanoseconds, maximum precision. 465 */ 466enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 467 468static int timestamp_precision = TSP_SEC; 469SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 470 ×tamp_precision, 0, ""); 471 472/* 473 * Get a current timestamp. 474 */ 475void 476vfs_timestamp(tsp) 477 struct timespec *tsp; 478{ 479 struct timeval tv; 480 481 switch (timestamp_precision) { 482 case TSP_SEC: 483 tsp->tv_sec = time_second; 484 tsp->tv_nsec = 0; 485 break; 486 case TSP_HZ: 487 getnanotime(tsp); 488 break; 489 case TSP_USEC: 490 microtime(&tv); 491 TIMEVAL_TO_TIMESPEC(&tv, tsp); 492 break; 493 case TSP_NSEC: 494 default: 495 nanotime(tsp); 496 break; 497 } 498} 499 500/* 501 * Build a linked list of mount options from a struct uio. 502 */ 503int 504vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 505{ 506 struct vfsoptlist *opts; 507 struct vfsopt *opt; 508 unsigned int i, iovcnt; 509 int error, namelen, optlen; 510 511 iovcnt = auio->uio_iovcnt; 512 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 513 TAILQ_INIT(opts); 514 for (i = 0; i < iovcnt; i += 2) { 515 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 516 namelen = auio->uio_iov[i].iov_len; 517 optlen = auio->uio_iov[i + 1].iov_len; 518 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 519 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 520 opt->len = optlen; 521 if (auio->uio_segflg == UIO_SYSSPACE) { 522 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 523 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 524 optlen); 525 } else { 526 error = copyin(auio->uio_iov[i].iov_base, opt->name, 527 namelen); 528 if (!error) 529 error = copyin(auio->uio_iov[i + 1].iov_base, 530 opt->value, optlen); 531 if (error) 532 goto bad; 533 } 534 TAILQ_INSERT_TAIL(opts, opt, link); 535 } 536 *options = opts; 537 return (0); 538bad: 539 vfs_freeopts(opts); 540 return (error); 541} 542 543/* 544 * Get a mount option by its name. 545 * 546 * Return 0 if the option was found, ENOENT otherwise. 547 * If len is non-NULL it will be filled with the length 548 * of the option. If buf is non-NULL, it will be filled 549 * with the address of the option. 550 */ 551int 552vfs_getopt(opts, name, buf, len) 553 struct vfsoptlist *opts; 554 const char *name; 555 void **buf; 556 int *len; 557{ 558 struct vfsopt *opt; 559 560 TAILQ_FOREACH(opt, opts, link) { 561 if (strcmp(name, opt->name) == 0) { 562 if (len != NULL) 563 *len = opt->len; 564 if (buf != NULL) 565 *buf = opt->value; 566 return (0); 567 } 568 } 569 return (ENOENT); 570} 571 572/* 573 * Find and copy a mount option. 574 * 575 * The size of the buffer has to be specified 576 * in len, if it is not the same length as the 577 * mount option, EINVAL is returned. 578 * Returns ENOENT if the option is not found. 579 */ 580int 581vfs_copyopt(opts, name, dest, len) 582 struct vfsoptlist *opts; 583 const char *name; 584 void *dest; 585 int len; 586{ 587 struct vfsopt *opt; 588 589 TAILQ_FOREACH(opt, opts, link) { 590 if (strcmp(name, opt->name) == 0) { 591 if (len != opt->len) 592 return (EINVAL); 593 bcopy(opt->value, dest, opt->len); 594 return (0); 595 } 596 } 597 return (ENOENT); 598} 599 600/* 601 * Set vnode attributes to VNOVAL 602 */ 603void 604vattr_null(vap) 605 register struct vattr *vap; 606{ 607 608 vap->va_type = VNON; 609 vap->va_size = VNOVAL; 610 vap->va_bytes = VNOVAL; 611 vap->va_mode = VNOVAL; 612 vap->va_nlink = VNOVAL; 613 vap->va_uid = VNOVAL; 614 vap->va_gid = VNOVAL; 615 vap->va_fsid = VNOVAL; 616 vap->va_fileid = VNOVAL; 617 vap->va_blocksize = VNOVAL; 618 vap->va_rdev = VNOVAL; 619 vap->va_atime.tv_sec = VNOVAL; 620 vap->va_atime.tv_nsec = VNOVAL; 621 vap->va_mtime.tv_sec = VNOVAL; 622 vap->va_mtime.tv_nsec = VNOVAL; 623 vap->va_ctime.tv_sec = VNOVAL; 624 vap->va_ctime.tv_nsec = VNOVAL; 625 vap->va_flags = VNOVAL; 626 vap->va_gen = VNOVAL; 627 vap->va_vaflags = 0; 628} 629 630/* 631 * This routine is called when we have too many vnodes. It attempts 632 * to free <count> vnodes and will potentially free vnodes that still 633 * have VM backing store (VM backing store is typically the cause 634 * of a vnode blowout so we want to do this). Therefore, this operation 635 * is not considered cheap. 636 * 637 * A number of conditions may prevent a vnode from being reclaimed. 638 * the buffer cache may have references on the vnode, a directory 639 * vnode may still have references due to the namei cache representing 640 * underlying files, or the vnode may be in active use. It is not 641 * desireable to reuse such vnodes. These conditions may cause the 642 * number of vnodes to reach some minimum value regardless of what 643 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 644 */ 645static int 646vlrureclaim(struct mount *mp, int count) 647{ 648 struct vnode *vp; 649 int done; 650 int trigger; 651 int usevnodes; 652 653 /* 654 * Calculate the trigger point, don't allow user 655 * screwups to blow us up. This prevents us from 656 * recycling vnodes with lots of resident pages. We 657 * aren't trying to free memory, we are trying to 658 * free vnodes. 659 */ 660 usevnodes = desiredvnodes; 661 if (usevnodes <= 0) 662 usevnodes = 1; 663 trigger = cnt.v_page_count * 2 / usevnodes; 664 665 done = 0; 666 mtx_lock(&mntvnode_mtx); 667 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) { 668 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 669 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 670 671 if (vp->v_type != VNON && 672 vp->v_type != VBAD && 673 VMIGHTFREE(vp) && /* critical path opt */ 674 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) && 675 mtx_trylock(&vp->v_interlock) 676 ) { 677 mtx_unlock(&mntvnode_mtx); 678 if (VMIGHTFREE(vp)) { 679 vgonel(vp, curthread); 680 done++; 681 } else { 682 mtx_unlock(&vp->v_interlock); 683 } 684 mtx_lock(&mntvnode_mtx); 685 } 686 --count; 687 } 688 mtx_unlock(&mntvnode_mtx); 689 return done; 690} 691 692/* 693 * Attempt to recycle vnodes in a context that is always safe to block. 694 * Calling vlrurecycle() from the bowels of filesystem code has some 695 * interesting deadlock problems. 696 */ 697static struct proc *vnlruproc; 698static int vnlruproc_sig; 699 700static void 701vnlru_proc(void) 702{ 703 struct mount *mp, *nmp; 704 int s; 705 int done; 706 struct proc *p = vnlruproc; 707 struct thread *td = FIRST_THREAD_IN_PROC(p); /* XXXKSE */ 708 709 mtx_lock(&Giant); 710 711 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 712 SHUTDOWN_PRI_FIRST); 713 714 s = splbio(); 715 for (;;) { 716 kthread_suspend_check(p); 717 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) { 718 vnlruproc_sig = 0; 719 tsleep(vnlruproc, PVFS, "vlruwt", 0); 720 continue; 721 } 722 done = 0; 723 mtx_lock(&mountlist_mtx); 724 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 725 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) { 726 nmp = TAILQ_NEXT(mp, mnt_list); 727 continue; 728 } 729 done += vlrureclaim(mp, 10); 730 mtx_lock(&mountlist_mtx); 731 nmp = TAILQ_NEXT(mp, mnt_list); 732 vfs_unbusy(mp, td); 733 } 734 mtx_unlock(&mountlist_mtx); 735 if (done == 0) { 736#if 0 737 /* These messages are temporary debugging aids */ 738 if (vnlru_nowhere < 5) 739 printf("vnlru process getting nowhere..\n"); 740 else if (vnlru_nowhere == 5) 741 printf("vnlru process messages stopped.\n"); 742#endif 743 vnlru_nowhere++; 744 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 745 } 746 } 747 splx(s); 748} 749 750static struct kproc_desc vnlru_kp = { 751 "vnlru", 752 vnlru_proc, 753 &vnlruproc 754}; 755SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp) 756 757 758/* 759 * Routines having to do with the management of the vnode table. 760 */ 761 762/* 763 * Return the next vnode from the free list. 764 */ 765int 766getnewvnode(tag, mp, vops, vpp) 767 enum vtagtype tag; 768 struct mount *mp; 769 vop_t **vops; 770 struct vnode **vpp; 771{ 772 int s; 773 struct thread *td = curthread; /* XXX */ 774 struct vnode *vp = NULL; 775 struct mount *vnmp; 776 vm_object_t object; 777 778 s = splbio(); 779 /* 780 * Try to reuse vnodes if we hit the max. This situation only 781 * occurs in certain large-memory (2G+) situations. We cannot 782 * attempt to directly reclaim vnodes due to nasty recursion 783 * problems. 784 */ 785 if (vnlruproc_sig == 0 && numvnodes - freevnodes > desiredvnodes) { 786 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 787 wakeup(vnlruproc); 788 } 789 790 /* 791 * Attempt to reuse a vnode already on the free list, allocating 792 * a new vnode if we can't find one or if we have not reached a 793 * good minimum for good LRU performance. 794 */ 795 796 mtx_lock(&vnode_free_list_mtx); 797 798 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) { 799 int count; 800 801 for (count = 0; count < freevnodes; count++) { 802 vp = TAILQ_FIRST(&vnode_free_list); 803 if (vp == NULL || vp->v_usecount) 804 panic("getnewvnode: free vnode isn't"); 805 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 806 807 /* Don't recycle if we can't get the interlock */ 808 if (!mtx_trylock(&vp->v_interlock)) { 809 vp = NULL; 810 continue; 811 } 812 813 /* We should be able to immediately acquire this */ 814 if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0) 815 continue; 816 /* 817 * Don't recycle if we still have cached pages. 818 */ 819 if (VOP_GETVOBJECT(vp, &object) == 0 && 820 (object->resident_page_count || 821 object->ref_count)) { 822 TAILQ_INSERT_TAIL(&vnode_free_list, vp, 823 v_freelist); 824 VOP_UNLOCK(vp, 0, td); 825 vp = NULL; 826 continue; 827 } 828 if (LIST_FIRST(&vp->v_cache_src)) { 829 /* 830 * note: nameileafonly sysctl is temporary, 831 * for debugging only, and will eventually be 832 * removed. 833 */ 834 if (nameileafonly > 0) { 835 /* 836 * Do not reuse namei-cached directory 837 * vnodes that have cached 838 * subdirectories. 839 */ 840 if (cache_leaf_test(vp) < 0) { 841 VOP_UNLOCK(vp, 0, td); 842 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 843 vp = NULL; 844 continue; 845 } 846 } else if (nameileafonly < 0 || 847 vmiodirenable == 0) { 848 /* 849 * Do not reuse namei-cached directory 850 * vnodes if nameileafonly is -1 or 851 * if VMIO backing for directories is 852 * turned off (otherwise we reuse them 853 * too quickly). 854 */ 855 VOP_UNLOCK(vp, 0, td); 856 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 857 vp = NULL; 858 continue; 859 } 860 } 861 /* 862 * Skip over it if its filesystem is being suspended. 863 */ 864 if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0) 865 break; 866 VOP_UNLOCK(vp, 0, td); 867 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 868 vp = NULL; 869 } 870 } 871 if (vp) { 872 vp->v_flag |= VDOOMED; 873 vp->v_flag &= ~VFREE; 874 freevnodes--; 875 mtx_unlock(&vnode_free_list_mtx); 876 cache_purge(vp); 877 if (vp->v_type != VBAD) { 878 VOP_UNLOCK(vp, 0, td); 879 vgone(vp); 880 } else { 881 VOP_UNLOCK(vp, 0, td); 882 } 883 vn_finished_write(vnmp); 884 885#ifdef INVARIANTS 886 { 887 int s; 888 889 if (vp->v_data) 890 panic("cleaned vnode isn't"); 891 s = splbio(); 892 if (vp->v_numoutput) 893 panic("Clean vnode has pending I/O's"); 894 splx(s); 895 if (vp->v_writecount != 0) 896 panic("Non-zero write count"); 897 } 898#endif 899 if (vp->v_pollinfo) { 900 mtx_destroy(&vp->v_pollinfo->vpi_lock); 901 uma_zfree(vnodepoll_zone, vp->v_pollinfo); 902 } 903 vp->v_pollinfo = NULL; 904 vp->v_flag = 0; 905 vp->v_lastw = 0; 906 vp->v_lasta = 0; 907 vp->v_cstart = 0; 908 vp->v_clen = 0; 909 vp->v_socket = 0; 910 } else { 911 mtx_unlock(&vnode_free_list_mtx); 912 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK); 913 bzero((char *) vp, sizeof *vp); 914 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 915 vp->v_dd = vp; 916 cache_purge(vp); 917 LIST_INIT(&vp->v_cache_src); 918 TAILQ_INIT(&vp->v_cache_dst); 919 numvnodes++; 920 } 921 922 TAILQ_INIT(&vp->v_cleanblkhd); 923 TAILQ_INIT(&vp->v_dirtyblkhd); 924 vp->v_type = VNON; 925 vp->v_tag = tag; 926 vp->v_op = vops; 927 lockinit(&vp->v_lock, PVFS, "vnlock", VLKTIMEOUT, LK_NOPAUSE); 928 insmntque(vp, mp); 929 *vpp = vp; 930 vp->v_usecount = 1; 931 vp->v_data = 0; 932 933 splx(s); 934 935#if 0 936 vnodeallocs++; 937 if (vnodeallocs % vnoderecycleperiod == 0 && 938 freevnodes < vnoderecycleminfreevn && 939 vnoderecyclemintotalvn < numvnodes) { 940 /* Recycle vnodes. */ 941 cache_purgeleafdirs(vnoderecyclenumber); 942 } 943#endif 944 945 return (0); 946} 947 948/* 949 * Move a vnode from one mount queue to another. 950 */ 951static void 952insmntque(vp, mp) 953 register struct vnode *vp; 954 register struct mount *mp; 955{ 956 957 mtx_lock(&mntvnode_mtx); 958 /* 959 * Delete from old mount point vnode list, if on one. 960 */ 961 if (vp->v_mount != NULL) 962 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 963 /* 964 * Insert into list of vnodes for the new mount point, if available. 965 */ 966 if ((vp->v_mount = mp) == NULL) { 967 mtx_unlock(&mntvnode_mtx); 968 return; 969 } 970 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 971 mtx_unlock(&mntvnode_mtx); 972} 973 974/* 975 * Update outstanding I/O count and do wakeup if requested. 976 */ 977void 978vwakeup(bp) 979 register struct buf *bp; 980{ 981 register struct vnode *vp; 982 983 bp->b_flags &= ~B_WRITEINPROG; 984 if ((vp = bp->b_vp)) { 985 vp->v_numoutput--; 986 if (vp->v_numoutput < 0) 987 panic("vwakeup: neg numoutput"); 988 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 989 vp->v_flag &= ~VBWAIT; 990 wakeup(&vp->v_numoutput); 991 } 992 } 993} 994 995/* 996 * Flush out and invalidate all buffers associated with a vnode. 997 * Called with the underlying object locked. 998 */ 999int 1000vinvalbuf(vp, flags, cred, td, slpflag, slptimeo) 1001 register struct vnode *vp; 1002 int flags; 1003 struct ucred *cred; 1004 struct thread *td; 1005 int slpflag, slptimeo; 1006{ 1007 register struct buf *bp; 1008 struct buf *nbp, *blist; 1009 int s, error; 1010 vm_object_t object; 1011 1012 GIANT_REQUIRED; 1013 1014 if (flags & V_SAVE) { 1015 s = splbio(); 1016 while (vp->v_numoutput) { 1017 vp->v_flag |= VBWAIT; 1018 error = tsleep(&vp->v_numoutput, 1019 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo); 1020 if (error) { 1021 splx(s); 1022 return (error); 1023 } 1024 } 1025 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 1026 splx(s); 1027 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0) 1028 return (error); 1029 s = splbio(); 1030 if (vp->v_numoutput > 0 || 1031 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 1032 panic("vinvalbuf: dirty bufs"); 1033 } 1034 splx(s); 1035 } 1036 s = splbio(); 1037 for (;;) { 1038 blist = TAILQ_FIRST(&vp->v_cleanblkhd); 1039 if (!blist) 1040 blist = TAILQ_FIRST(&vp->v_dirtyblkhd); 1041 if (!blist) 1042 break; 1043 1044 for (bp = blist; bp; bp = nbp) { 1045 nbp = TAILQ_NEXT(bp, b_vnbufs); 1046 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1047 error = BUF_TIMELOCK(bp, 1048 LK_EXCLUSIVE | LK_SLEEPFAIL, 1049 "vinvalbuf", slpflag, slptimeo); 1050 if (error == ENOLCK) 1051 break; 1052 splx(s); 1053 return (error); 1054 } 1055 /* 1056 * XXX Since there are no node locks for NFS, I 1057 * believe there is a slight chance that a delayed 1058 * write will occur while sleeping just above, so 1059 * check for it. Note that vfs_bio_awrite expects 1060 * buffers to reside on a queue, while BUF_WRITE and 1061 * brelse do not. 1062 */ 1063 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1064 (flags & V_SAVE)) { 1065 1066 if (bp->b_vp == vp) { 1067 if (bp->b_flags & B_CLUSTEROK) { 1068 BUF_UNLOCK(bp); 1069 vfs_bio_awrite(bp); 1070 } else { 1071 bremfree(bp); 1072 bp->b_flags |= B_ASYNC; 1073 BUF_WRITE(bp); 1074 } 1075 } else { 1076 bremfree(bp); 1077 (void) BUF_WRITE(bp); 1078 } 1079 break; 1080 } 1081 bremfree(bp); 1082 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 1083 bp->b_flags &= ~B_ASYNC; 1084 brelse(bp); 1085 } 1086 } 1087 1088 /* 1089 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1090 * have write I/O in-progress but if there is a VM object then the 1091 * VM object can also have read-I/O in-progress. 1092 */ 1093 do { 1094 while (vp->v_numoutput > 0) { 1095 vp->v_flag |= VBWAIT; 1096 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 1097 } 1098 if (VOP_GETVOBJECT(vp, &object) == 0) { 1099 while (object->paging_in_progress) 1100 vm_object_pip_sleep(object, "vnvlbx"); 1101 } 1102 } while (vp->v_numoutput > 0); 1103 1104 splx(s); 1105 1106 /* 1107 * Destroy the copy in the VM cache, too. 1108 */ 1109 mtx_lock(&vp->v_interlock); 1110 if (VOP_GETVOBJECT(vp, &object) == 0) { 1111 vm_object_page_remove(object, 0, 0, 1112 (flags & V_SAVE) ? TRUE : FALSE); 1113 } 1114 mtx_unlock(&vp->v_interlock); 1115 1116 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd)) 1117 panic("vinvalbuf: flush failed"); 1118 return (0); 1119} 1120 1121/* 1122 * Truncate a file's buffer and pages to a specified length. This 1123 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1124 * sync activity. 1125 */ 1126int 1127vtruncbuf(vp, cred, td, length, blksize) 1128 register struct vnode *vp; 1129 struct ucred *cred; 1130 struct thread *td; 1131 off_t length; 1132 int blksize; 1133{ 1134 register struct buf *bp; 1135 struct buf *nbp; 1136 int s, anyfreed; 1137 int trunclbn; 1138 1139 /* 1140 * Round up to the *next* lbn. 1141 */ 1142 trunclbn = (length + blksize - 1) / blksize; 1143 1144 s = splbio(); 1145restart: 1146 anyfreed = 1; 1147 for (;anyfreed;) { 1148 anyfreed = 0; 1149 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 1150 nbp = TAILQ_NEXT(bp, b_vnbufs); 1151 if (bp->b_lblkno >= trunclbn) { 1152 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1153 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 1154 goto restart; 1155 } else { 1156 bremfree(bp); 1157 bp->b_flags |= (B_INVAL | B_RELBUF); 1158 bp->b_flags &= ~B_ASYNC; 1159 brelse(bp); 1160 anyfreed = 1; 1161 } 1162 if (nbp && 1163 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1164 (nbp->b_vp != vp) || 1165 (nbp->b_flags & B_DELWRI))) { 1166 goto restart; 1167 } 1168 } 1169 } 1170 1171 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1172 nbp = TAILQ_NEXT(bp, b_vnbufs); 1173 if (bp->b_lblkno >= trunclbn) { 1174 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1175 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 1176 goto restart; 1177 } else { 1178 bremfree(bp); 1179 bp->b_flags |= (B_INVAL | B_RELBUF); 1180 bp->b_flags &= ~B_ASYNC; 1181 brelse(bp); 1182 anyfreed = 1; 1183 } 1184 if (nbp && 1185 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1186 (nbp->b_vp != vp) || 1187 (nbp->b_flags & B_DELWRI) == 0)) { 1188 goto restart; 1189 } 1190 } 1191 } 1192 } 1193 1194 if (length > 0) { 1195restartsync: 1196 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1197 nbp = TAILQ_NEXT(bp, b_vnbufs); 1198 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) { 1199 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1200 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 1201 goto restart; 1202 } else { 1203 bremfree(bp); 1204 if (bp->b_vp == vp) { 1205 bp->b_flags |= B_ASYNC; 1206 } else { 1207 bp->b_flags &= ~B_ASYNC; 1208 } 1209 BUF_WRITE(bp); 1210 } 1211 goto restartsync; 1212 } 1213 1214 } 1215 } 1216 1217 while (vp->v_numoutput > 0) { 1218 vp->v_flag |= VBWAIT; 1219 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0); 1220 } 1221 1222 splx(s); 1223 1224 vnode_pager_setsize(vp, length); 1225 1226 return (0); 1227} 1228 1229/* 1230 * Associate a buffer with a vnode. 1231 */ 1232void 1233bgetvp(vp, bp) 1234 register struct vnode *vp; 1235 register struct buf *bp; 1236{ 1237 int s; 1238 1239 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 1240 1241 vhold(vp); 1242 bp->b_vp = vp; 1243 bp->b_dev = vn_todev(vp); 1244 /* 1245 * Insert onto list for new vnode. 1246 */ 1247 s = splbio(); 1248 bp->b_xflags |= BX_VNCLEAN; 1249 bp->b_xflags &= ~BX_VNDIRTY; 1250 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 1251 splx(s); 1252} 1253 1254/* 1255 * Disassociate a buffer from a vnode. 1256 */ 1257void 1258brelvp(bp) 1259 register struct buf *bp; 1260{ 1261 struct vnode *vp; 1262 struct buflists *listheadp; 1263 int s; 1264 1265 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1266 1267 /* 1268 * Delete from old vnode list, if on one. 1269 */ 1270 vp = bp->b_vp; 1271 s = splbio(); 1272 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1273 if (bp->b_xflags & BX_VNDIRTY) 1274 listheadp = &vp->v_dirtyblkhd; 1275 else 1276 listheadp = &vp->v_cleanblkhd; 1277 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1278 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1279 } 1280 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 1281 vp->v_flag &= ~VONWORKLST; 1282 LIST_REMOVE(vp, v_synclist); 1283 } 1284 splx(s); 1285 bp->b_vp = (struct vnode *) 0; 1286 vdrop(vp); 1287} 1288 1289/* 1290 * Add an item to the syncer work queue. 1291 */ 1292static void 1293vn_syncer_add_to_worklist(struct vnode *vp, int delay) 1294{ 1295 int s, slot; 1296 1297 s = splbio(); 1298 1299 if (vp->v_flag & VONWORKLST) { 1300 LIST_REMOVE(vp, v_synclist); 1301 } 1302 1303 if (delay > syncer_maxdelay - 2) 1304 delay = syncer_maxdelay - 2; 1305 slot = (syncer_delayno + delay) & syncer_mask; 1306 1307 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist); 1308 vp->v_flag |= VONWORKLST; 1309 splx(s); 1310} 1311 1312struct proc *updateproc; 1313static void sched_sync(void); 1314static struct kproc_desc up_kp = { 1315 "syncer", 1316 sched_sync, 1317 &updateproc 1318}; 1319SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 1320 1321/* 1322 * System filesystem synchronizer daemon. 1323 */ 1324void 1325sched_sync(void) 1326{ 1327 struct synclist *slp; 1328 struct vnode *vp; 1329 struct mount *mp; 1330 long starttime; 1331 int s; 1332 struct thread *td = FIRST_THREAD_IN_PROC(updateproc); /* XXXKSE */ 1333 1334 mtx_lock(&Giant); 1335 1336 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc, 1337 SHUTDOWN_PRI_LAST); 1338 1339 for (;;) { 1340 kthread_suspend_check(td->td_proc); 1341 1342 starttime = time_second; 1343 1344 /* 1345 * Push files whose dirty time has expired. Be careful 1346 * of interrupt race on slp queue. 1347 */ 1348 s = splbio(); 1349 slp = &syncer_workitem_pending[syncer_delayno]; 1350 syncer_delayno += 1; 1351 if (syncer_delayno == syncer_maxdelay) 1352 syncer_delayno = 0; 1353 splx(s); 1354 1355 while ((vp = LIST_FIRST(slp)) != NULL) { 1356 if (VOP_ISLOCKED(vp, NULL) == 0 && 1357 vn_start_write(vp, &mp, V_NOWAIT) == 0) { 1358 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1359 (void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td); 1360 VOP_UNLOCK(vp, 0, td); 1361 vn_finished_write(mp); 1362 } 1363 s = splbio(); 1364 if (LIST_FIRST(slp) == vp) { 1365 /* 1366 * Note: v_tag VT_VFS vps can remain on the 1367 * worklist too with no dirty blocks, but 1368 * since sync_fsync() moves it to a different 1369 * slot we are safe. 1370 */ 1371 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) && 1372 !vn_isdisk(vp, NULL)) 1373 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag); 1374 /* 1375 * Put us back on the worklist. The worklist 1376 * routine will remove us from our current 1377 * position and then add us back in at a later 1378 * position. 1379 */ 1380 vn_syncer_add_to_worklist(vp, syncdelay); 1381 } 1382 splx(s); 1383 } 1384 1385 /* 1386 * Do soft update processing. 1387 */ 1388 if (softdep_process_worklist_hook != NULL) 1389 (*softdep_process_worklist_hook)(NULL); 1390 1391 /* 1392 * The variable rushjob allows the kernel to speed up the 1393 * processing of the filesystem syncer process. A rushjob 1394 * value of N tells the filesystem syncer to process the next 1395 * N seconds worth of work on its queue ASAP. Currently rushjob 1396 * is used by the soft update code to speed up the filesystem 1397 * syncer process when the incore state is getting so far 1398 * ahead of the disk that the kernel memory pool is being 1399 * threatened with exhaustion. 1400 */ 1401 if (rushjob > 0) { 1402 rushjob -= 1; 1403 continue; 1404 } 1405 /* 1406 * If it has taken us less than a second to process the 1407 * current work, then wait. Otherwise start right over 1408 * again. We can still lose time if any single round 1409 * takes more than two seconds, but it does not really 1410 * matter as we are just trying to generally pace the 1411 * filesystem activity. 1412 */ 1413 if (time_second == starttime) 1414 tsleep(&lbolt, PPAUSE, "syncer", 0); 1415 } 1416} 1417 1418/* 1419 * Request the syncer daemon to speed up its work. 1420 * We never push it to speed up more than half of its 1421 * normal turn time, otherwise it could take over the cpu. 1422 * XXXKSE only one update? 1423 */ 1424int 1425speedup_syncer() 1426{ 1427 1428 mtx_lock_spin(&sched_lock); 1429 if (FIRST_THREAD_IN_PROC(updateproc)->td_wchan == &lbolt) /* XXXKSE */ 1430 setrunnable(FIRST_THREAD_IN_PROC(updateproc)); 1431 mtx_unlock_spin(&sched_lock); 1432 if (rushjob < syncdelay / 2) { 1433 rushjob += 1; 1434 stat_rush_requests += 1; 1435 return (1); 1436 } 1437 return(0); 1438} 1439 1440/* 1441 * Associate a p-buffer with a vnode. 1442 * 1443 * Also sets B_PAGING flag to indicate that vnode is not fully associated 1444 * with the buffer. i.e. the bp has not been linked into the vnode or 1445 * ref-counted. 1446 */ 1447void 1448pbgetvp(vp, bp) 1449 register struct vnode *vp; 1450 register struct buf *bp; 1451{ 1452 1453 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 1454 1455 bp->b_vp = vp; 1456 bp->b_flags |= B_PAGING; 1457 bp->b_dev = vn_todev(vp); 1458} 1459 1460/* 1461 * Disassociate a p-buffer from a vnode. 1462 */ 1463void 1464pbrelvp(bp) 1465 register struct buf *bp; 1466{ 1467 1468 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 1469 1470 /* XXX REMOVE ME */ 1471 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) { 1472 panic( 1473 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 1474 bp, 1475 (int)bp->b_flags 1476 ); 1477 } 1478 bp->b_vp = (struct vnode *) 0; 1479 bp->b_flags &= ~B_PAGING; 1480} 1481 1482/* 1483 * Reassign a buffer from one vnode to another. 1484 * Used to assign file specific control information 1485 * (indirect blocks) to the vnode to which they belong. 1486 */ 1487void 1488reassignbuf(bp, newvp) 1489 register struct buf *bp; 1490 register struct vnode *newvp; 1491{ 1492 struct buflists *listheadp; 1493 int delay; 1494 int s; 1495 1496 if (newvp == NULL) { 1497 printf("reassignbuf: NULL"); 1498 return; 1499 } 1500 ++reassignbufcalls; 1501 1502 /* 1503 * B_PAGING flagged buffers cannot be reassigned because their vp 1504 * is not fully linked in. 1505 */ 1506 if (bp->b_flags & B_PAGING) 1507 panic("cannot reassign paging buffer"); 1508 1509 s = splbio(); 1510 /* 1511 * Delete from old vnode list, if on one. 1512 */ 1513 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1514 if (bp->b_xflags & BX_VNDIRTY) 1515 listheadp = &bp->b_vp->v_dirtyblkhd; 1516 else 1517 listheadp = &bp->b_vp->v_cleanblkhd; 1518 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1519 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1520 if (bp->b_vp != newvp) { 1521 vdrop(bp->b_vp); 1522 bp->b_vp = NULL; /* for clarification */ 1523 } 1524 } 1525 /* 1526 * If dirty, put on list of dirty buffers; otherwise insert onto list 1527 * of clean buffers. 1528 */ 1529 if (bp->b_flags & B_DELWRI) { 1530 struct buf *tbp; 1531 1532 listheadp = &newvp->v_dirtyblkhd; 1533 if ((newvp->v_flag & VONWORKLST) == 0) { 1534 switch (newvp->v_type) { 1535 case VDIR: 1536 delay = dirdelay; 1537 break; 1538 case VCHR: 1539 if (newvp->v_rdev->si_mountpoint != NULL) { 1540 delay = metadelay; 1541 break; 1542 } 1543 /* fall through */ 1544 default: 1545 delay = filedelay; 1546 } 1547 vn_syncer_add_to_worklist(newvp, delay); 1548 } 1549 bp->b_xflags |= BX_VNDIRTY; 1550 tbp = TAILQ_FIRST(listheadp); 1551 if (tbp == NULL || 1552 bp->b_lblkno == 0 || 1553 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) || 1554 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) { 1555 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs); 1556 ++reassignbufsortgood; 1557 } else if (bp->b_lblkno < 0) { 1558 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs); 1559 ++reassignbufsortgood; 1560 } else if (reassignbufmethod == 1) { 1561 /* 1562 * New sorting algorithm, only handle sequential case, 1563 * otherwise append to end (but before metadata) 1564 */ 1565 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL && 1566 (tbp->b_xflags & BX_VNDIRTY)) { 1567 /* 1568 * Found the best place to insert the buffer 1569 */ 1570 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1571 ++reassignbufsortgood; 1572 } else { 1573 /* 1574 * Missed, append to end, but before meta-data. 1575 * We know that the head buffer in the list is 1576 * not meta-data due to prior conditionals. 1577 * 1578 * Indirect effects: NFS second stage write 1579 * tends to wind up here, giving maximum 1580 * distance between the unstable write and the 1581 * commit rpc. 1582 */ 1583 tbp = TAILQ_LAST(listheadp, buflists); 1584 while (tbp && tbp->b_lblkno < 0) 1585 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs); 1586 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1587 ++reassignbufsortbad; 1588 } 1589 } else { 1590 /* 1591 * Old sorting algorithm, scan queue and insert 1592 */ 1593 struct buf *ttbp; 1594 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) && 1595 (ttbp->b_lblkno < bp->b_lblkno)) { 1596 ++reassignbufloops; 1597 tbp = ttbp; 1598 } 1599 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1600 } 1601 } else { 1602 bp->b_xflags |= BX_VNCLEAN; 1603 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs); 1604 if ((newvp->v_flag & VONWORKLST) && 1605 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 1606 newvp->v_flag &= ~VONWORKLST; 1607 LIST_REMOVE(newvp, v_synclist); 1608 } 1609 } 1610 if (bp->b_vp != newvp) { 1611 bp->b_vp = newvp; 1612 vhold(bp->b_vp); 1613 } 1614 splx(s); 1615} 1616 1617/* 1618 * Create a vnode for a device. 1619 * Used for mounting the root filesystem. 1620 */ 1621int 1622bdevvp(dev, vpp) 1623 dev_t dev; 1624 struct vnode **vpp; 1625{ 1626 register struct vnode *vp; 1627 struct vnode *nvp; 1628 int error; 1629 1630 if (dev == NODEV) { 1631 *vpp = NULLVP; 1632 return (ENXIO); 1633 } 1634 if (vfinddev(dev, VCHR, vpp)) 1635 return (0); 1636 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 1637 if (error) { 1638 *vpp = NULLVP; 1639 return (error); 1640 } 1641 vp = nvp; 1642 vp->v_type = VCHR; 1643 addalias(vp, dev); 1644 *vpp = vp; 1645 return (0); 1646} 1647 1648/* 1649 * Add vnode to the alias list hung off the dev_t. 1650 * 1651 * The reason for this gunk is that multiple vnodes can reference 1652 * the same physical device, so checking vp->v_usecount to see 1653 * how many users there are is inadequate; the v_usecount for 1654 * the vnodes need to be accumulated. vcount() does that. 1655 */ 1656struct vnode * 1657addaliasu(nvp, nvp_rdev) 1658 struct vnode *nvp; 1659 udev_t nvp_rdev; 1660{ 1661 struct vnode *ovp; 1662 vop_t **ops; 1663 dev_t dev; 1664 1665 if (nvp->v_type == VBLK) 1666 return (nvp); 1667 if (nvp->v_type != VCHR) 1668 panic("addaliasu on non-special vnode"); 1669 dev = udev2dev(nvp_rdev, 0); 1670 /* 1671 * Check to see if we have a bdevvp vnode with no associated 1672 * filesystem. If so, we want to associate the filesystem of 1673 * the new newly instigated vnode with the bdevvp vnode and 1674 * discard the newly created vnode rather than leaving the 1675 * bdevvp vnode lying around with no associated filesystem. 1676 */ 1677 if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) { 1678 addalias(nvp, dev); 1679 return (nvp); 1680 } 1681 /* 1682 * Discard unneeded vnode, but save its node specific data. 1683 * Note that if there is a lock, it is carried over in the 1684 * node specific data to the replacement vnode. 1685 */ 1686 vref(ovp); 1687 ovp->v_data = nvp->v_data; 1688 ovp->v_tag = nvp->v_tag; 1689 nvp->v_data = NULL; 1690 lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg, 1691 nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK); 1692 if (nvp->v_vnlock) 1693 ovp->v_vnlock = &ovp->v_lock; 1694 ops = ovp->v_op; 1695 ovp->v_op = nvp->v_op; 1696 if (VOP_ISLOCKED(nvp, curthread)) { 1697 VOP_UNLOCK(nvp, 0, curthread); 1698 vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread); 1699 } 1700 nvp->v_op = ops; 1701 insmntque(ovp, nvp->v_mount); 1702 vrele(nvp); 1703 vgone(nvp); 1704 return (ovp); 1705} 1706 1707/* This is a local helper function that do the same as addaliasu, but for a 1708 * dev_t instead of an udev_t. */ 1709static void 1710addalias(nvp, dev) 1711 struct vnode *nvp; 1712 dev_t dev; 1713{ 1714 1715 KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode")); 1716 nvp->v_rdev = dev; 1717 mtx_lock(&spechash_mtx); 1718 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext); 1719 mtx_unlock(&spechash_mtx); 1720} 1721 1722/* 1723 * Grab a particular vnode from the free list, increment its 1724 * reference count and lock it. The vnode lock bit is set if the 1725 * vnode is being eliminated in vgone. The process is awakened 1726 * when the transition is completed, and an error returned to 1727 * indicate that the vnode is no longer usable (possibly having 1728 * been changed to a new filesystem type). 1729 */ 1730int 1731vget(vp, flags, td) 1732 register struct vnode *vp; 1733 int flags; 1734 struct thread *td; 1735{ 1736 int error; 1737 1738 /* 1739 * If the vnode is in the process of being cleaned out for 1740 * another use, we wait for the cleaning to finish and then 1741 * return failure. Cleaning is determined by checking that 1742 * the VXLOCK flag is set. 1743 */ 1744 if ((flags & LK_INTERLOCK) == 0) 1745 mtx_lock(&vp->v_interlock); 1746 if (vp->v_flag & VXLOCK) { 1747 if (vp->v_vxproc == curthread) { 1748#if 0 1749 /* this can now occur in normal operation */ 1750 log(LOG_INFO, "VXLOCK interlock avoided\n"); 1751#endif 1752 } else { 1753 vp->v_flag |= VXWANT; 1754 msleep(vp, &vp->v_interlock, PINOD | PDROP, "vget", 0); 1755 return (ENOENT); 1756 } 1757 } 1758 1759 vp->v_usecount++; 1760 1761 if (VSHOULDBUSY(vp)) 1762 vbusy(vp); 1763 if (flags & LK_TYPE_MASK) { 1764 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) { 1765 /* 1766 * must expand vrele here because we do not want 1767 * to call VOP_INACTIVE if the reference count 1768 * drops back to zero since it was never really 1769 * active. We must remove it from the free list 1770 * before sleeping so that multiple processes do 1771 * not try to recycle it. 1772 */ 1773 mtx_lock(&vp->v_interlock); 1774 vp->v_usecount--; 1775 if (VSHOULDFREE(vp)) 1776 vfree(vp); 1777 else 1778 vlruvp(vp); 1779 mtx_unlock(&vp->v_interlock); 1780 } 1781 return (error); 1782 } 1783 mtx_unlock(&vp->v_interlock); 1784 return (0); 1785} 1786 1787/* 1788 * Increase the reference count of a vnode. 1789 */ 1790void 1791vref(struct vnode *vp) 1792{ 1793 mtx_lock(&vp->v_interlock); 1794 vp->v_usecount++; 1795 mtx_unlock(&vp->v_interlock); 1796} 1797 1798/* 1799 * Vnode put/release. 1800 * If count drops to zero, call inactive routine and return to freelist. 1801 */ 1802void 1803vrele(vp) 1804 struct vnode *vp; 1805{ 1806 struct thread *td = curthread; /* XXX */ 1807 1808 KASSERT(vp != NULL, ("vrele: null vp")); 1809 1810 mtx_lock(&vp->v_interlock); 1811 1812 /* Skip this v_writecount check if we're going to panic below. */ 1813 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, 1814 ("vrele: missed vn_close")); 1815 1816 if (vp->v_usecount > 1) { 1817 1818 vp->v_usecount--; 1819 mtx_unlock(&vp->v_interlock); 1820 1821 return; 1822 } 1823 1824 if (vp->v_usecount == 1) { 1825 vp->v_usecount--; 1826 /* 1827 * We must call VOP_INACTIVE with the node locked. 1828 * If we are doing a vput, the node is already locked, 1829 * but, in the case of vrele, we must explicitly lock 1830 * the vnode before calling VOP_INACTIVE. 1831 */ 1832 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) 1833 VOP_INACTIVE(vp, td); 1834 if (VSHOULDFREE(vp)) 1835 vfree(vp); 1836 else 1837 vlruvp(vp); 1838 1839 } else { 1840#ifdef DIAGNOSTIC 1841 vprint("vrele: negative ref count", vp); 1842 mtx_unlock(&vp->v_interlock); 1843#endif 1844 panic("vrele: negative ref cnt"); 1845 } 1846} 1847 1848/* 1849 * Release an already locked vnode. This give the same effects as 1850 * unlock+vrele(), but takes less time and avoids releasing and 1851 * re-aquiring the lock (as vrele() aquires the lock internally.) 1852 */ 1853void 1854vput(vp) 1855 struct vnode *vp; 1856{ 1857 struct thread *td = curthread; /* XXX */ 1858 1859 GIANT_REQUIRED; 1860 1861 KASSERT(vp != NULL, ("vput: null vp")); 1862 mtx_lock(&vp->v_interlock); 1863 /* Skip this v_writecount check if we're going to panic below. */ 1864 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, 1865 ("vput: missed vn_close")); 1866 1867 if (vp->v_usecount > 1) { 1868 vp->v_usecount--; 1869 VOP_UNLOCK(vp, LK_INTERLOCK, td); 1870 return; 1871 } 1872 1873 if (vp->v_usecount == 1) { 1874 vp->v_usecount--; 1875 /* 1876 * We must call VOP_INACTIVE with the node locked. 1877 * If we are doing a vput, the node is already locked, 1878 * so we just need to release the vnode mutex. 1879 */ 1880 mtx_unlock(&vp->v_interlock); 1881 VOP_INACTIVE(vp, td); 1882 if (VSHOULDFREE(vp)) 1883 vfree(vp); 1884 else 1885 vlruvp(vp); 1886 1887 } else { 1888#ifdef DIAGNOSTIC 1889 vprint("vput: negative ref count", vp); 1890#endif 1891 panic("vput: negative ref cnt"); 1892 } 1893} 1894 1895/* 1896 * Somebody doesn't want the vnode recycled. 1897 */ 1898void 1899vhold(vp) 1900 register struct vnode *vp; 1901{ 1902 int s; 1903 1904 s = splbio(); 1905 vp->v_holdcnt++; 1906 if (VSHOULDBUSY(vp)) 1907 vbusy(vp); 1908 splx(s); 1909} 1910 1911/* 1912 * Note that there is one less who cares about this vnode. vdrop() is the 1913 * opposite of vhold(). 1914 */ 1915void 1916vdrop(vp) 1917 register struct vnode *vp; 1918{ 1919 int s; 1920 1921 s = splbio(); 1922 if (vp->v_holdcnt <= 0) 1923 panic("vdrop: holdcnt"); 1924 vp->v_holdcnt--; 1925 if (VSHOULDFREE(vp)) 1926 vfree(vp); 1927 else 1928 vlruvp(vp); 1929 splx(s); 1930} 1931 1932/* 1933 * Remove any vnodes in the vnode table belonging to mount point mp. 1934 * 1935 * If FORCECLOSE is not specified, there should not be any active ones, 1936 * return error if any are found (nb: this is a user error, not a 1937 * system error). If FORCECLOSE is specified, detach any active vnodes 1938 * that are found. 1939 * 1940 * If WRITECLOSE is set, only flush out regular file vnodes open for 1941 * writing. 1942 * 1943 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 1944 * 1945 * `rootrefs' specifies the base reference count for the root vnode 1946 * of this filesystem. The root vnode is considered busy if its 1947 * v_usecount exceeds this value. On a successful return, vflush() 1948 * will call vrele() on the root vnode exactly rootrefs times. 1949 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 1950 * be zero. 1951 */ 1952#ifdef DIAGNOSTIC 1953static int busyprt = 0; /* print out busy vnodes */ 1954SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1955#endif 1956 1957int 1958vflush(mp, rootrefs, flags) 1959 struct mount *mp; 1960 int rootrefs; 1961 int flags; 1962{ 1963 struct thread *td = curthread; /* XXX */ 1964 struct vnode *vp, *nvp, *rootvp = NULL; 1965 struct vattr vattr; 1966 int busy = 0, error; 1967 1968 if (rootrefs > 0) { 1969 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 1970 ("vflush: bad args")); 1971 /* 1972 * Get the filesystem root vnode. We can vput() it 1973 * immediately, since with rootrefs > 0, it won't go away. 1974 */ 1975 if ((error = VFS_ROOT(mp, &rootvp)) != 0) 1976 return (error); 1977 vput(rootvp); 1978 } 1979 mtx_lock(&mntvnode_mtx); 1980loop: 1981 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) { 1982 /* 1983 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1984 * Start over if it has (it won't be on the list anymore). 1985 */ 1986 if (vp->v_mount != mp) 1987 goto loop; 1988 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1989 1990 mtx_unlock(&mntvnode_mtx); 1991 mtx_lock(&vp->v_interlock); 1992 /* 1993 * Skip over a vnodes marked VSYSTEM. 1994 */ 1995 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1996 mtx_unlock(&vp->v_interlock); 1997 mtx_lock(&mntvnode_mtx); 1998 continue; 1999 } 2000 /* 2001 * If WRITECLOSE is set, flush out unlinked but still open 2002 * files (even if open only for reading) and regular file 2003 * vnodes open for writing. 2004 */ 2005 if ((flags & WRITECLOSE) && 2006 (vp->v_type == VNON || 2007 (VOP_GETATTR(vp, &vattr, td->td_ucred, td) == 0 && 2008 vattr.va_nlink > 0)) && 2009 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2010 mtx_unlock(&vp->v_interlock); 2011 mtx_lock(&mntvnode_mtx); 2012 continue; 2013 } 2014 2015 /* 2016 * With v_usecount == 0, all we need to do is clear out the 2017 * vnode data structures and we are done. 2018 */ 2019 if (vp->v_usecount == 0) { 2020 vgonel(vp, td); 2021 mtx_lock(&mntvnode_mtx); 2022 continue; 2023 } 2024 2025 /* 2026 * If FORCECLOSE is set, forcibly close the vnode. For block 2027 * or character devices, revert to an anonymous device. For 2028 * all other files, just kill them. 2029 */ 2030 if (flags & FORCECLOSE) { 2031 if (vp->v_type != VCHR) { 2032 vgonel(vp, td); 2033 } else { 2034 vclean(vp, 0, td); 2035 vp->v_op = spec_vnodeop_p; 2036 insmntque(vp, (struct mount *) 0); 2037 } 2038 mtx_lock(&mntvnode_mtx); 2039 continue; 2040 } 2041#ifdef DIAGNOSTIC 2042 if (busyprt) 2043 vprint("vflush: busy vnode", vp); 2044#endif 2045 mtx_unlock(&vp->v_interlock); 2046 mtx_lock(&mntvnode_mtx); 2047 busy++; 2048 } 2049 mtx_unlock(&mntvnode_mtx); 2050 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2051 /* 2052 * If just the root vnode is busy, and if its refcount 2053 * is equal to `rootrefs', then go ahead and kill it. 2054 */ 2055 mtx_lock(&rootvp->v_interlock); 2056 KASSERT(busy > 0, ("vflush: not busy")); 2057 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs")); 2058 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2059 vgonel(rootvp, td); 2060 busy = 0; 2061 } else 2062 mtx_unlock(&rootvp->v_interlock); 2063 } 2064 if (busy) 2065 return (EBUSY); 2066 for (; rootrefs > 0; rootrefs--) 2067 vrele(rootvp); 2068 return (0); 2069} 2070 2071/* 2072 * This moves a now (likely recyclable) vnode to the end of the 2073 * mountlist. XXX However, it is temporarily disabled until we 2074 * can clean up ffs_sync() and friends, which have loop restart 2075 * conditions which this code causes to operate O(N^2). 2076 */ 2077static void 2078vlruvp(struct vnode *vp) 2079{ 2080#if 0 2081 struct mount *mp; 2082 2083 if ((mp = vp->v_mount) != NULL) { 2084 mtx_lock(&mntvnode_mtx); 2085 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 2086 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 2087 mtx_unlock(&mntvnode_mtx); 2088 } 2089#endif 2090} 2091 2092/* 2093 * Disassociate the underlying filesystem from a vnode. 2094 */ 2095static void 2096vclean(vp, flags, td) 2097 struct vnode *vp; 2098 int flags; 2099 struct thread *td; 2100{ 2101 int active; 2102 2103 /* 2104 * Check to see if the vnode is in use. If so we have to reference it 2105 * before we clean it out so that its count cannot fall to zero and 2106 * generate a race against ourselves to recycle it. 2107 */ 2108 if ((active = vp->v_usecount)) 2109 vp->v_usecount++; 2110 2111 /* 2112 * Prevent the vnode from being recycled or brought into use while we 2113 * clean it out. 2114 */ 2115 if (vp->v_flag & VXLOCK) 2116 panic("vclean: deadlock"); 2117 vp->v_flag |= VXLOCK; 2118 vp->v_vxproc = curthread; 2119 /* 2120 * Even if the count is zero, the VOP_INACTIVE routine may still 2121 * have the object locked while it cleans it out. The VOP_LOCK 2122 * ensures that the VOP_INACTIVE routine is done with its work. 2123 * For active vnodes, it ensures that no other activity can 2124 * occur while the underlying object is being cleaned out. 2125 */ 2126 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td); 2127 2128 /* 2129 * Clean out any buffers associated with the vnode. 2130 * If the flush fails, just toss the buffers. 2131 */ 2132 if (flags & DOCLOSE) { 2133 if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL) 2134 (void) vn_write_suspend_wait(vp, NULL, V_WAIT); 2135 if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0) 2136 vinvalbuf(vp, 0, NOCRED, td, 0, 0); 2137 } 2138 2139 VOP_DESTROYVOBJECT(vp); 2140 2141 /* 2142 * If purging an active vnode, it must be closed and 2143 * deactivated before being reclaimed. Note that the 2144 * VOP_INACTIVE will unlock the vnode. 2145 */ 2146 if (active) { 2147 if (flags & DOCLOSE) 2148 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2149 VOP_INACTIVE(vp, td); 2150 } else { 2151 /* 2152 * Any other processes trying to obtain this lock must first 2153 * wait for VXLOCK to clear, then call the new lock operation. 2154 */ 2155 VOP_UNLOCK(vp, 0, td); 2156 } 2157 /* 2158 * Reclaim the vnode. 2159 */ 2160 if (VOP_RECLAIM(vp, td)) 2161 panic("vclean: cannot reclaim"); 2162 2163 if (active) { 2164 /* 2165 * Inline copy of vrele() since VOP_INACTIVE 2166 * has already been called. 2167 */ 2168 mtx_lock(&vp->v_interlock); 2169 if (--vp->v_usecount <= 0) { 2170#ifdef DIAGNOSTIC 2171 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 2172 vprint("vclean: bad ref count", vp); 2173 panic("vclean: ref cnt"); 2174 } 2175#endif 2176 vfree(vp); 2177 } 2178 mtx_unlock(&vp->v_interlock); 2179 } 2180 2181 cache_purge(vp); 2182 vp->v_vnlock = NULL; 2183 lockdestroy(&vp->v_lock); 2184 2185 if (VSHOULDFREE(vp)) 2186 vfree(vp); 2187 2188 /* 2189 * Done with purge, notify sleepers of the grim news. 2190 */ 2191 vp->v_op = dead_vnodeop_p; 2192 if (vp->v_pollinfo != NULL) 2193 vn_pollgone(vp); 2194 vp->v_tag = VT_NON; 2195 vp->v_flag &= ~VXLOCK; 2196 vp->v_vxproc = NULL; 2197 if (vp->v_flag & VXWANT) { 2198 vp->v_flag &= ~VXWANT; 2199 wakeup(vp); 2200 } 2201} 2202 2203/* 2204 * Eliminate all activity associated with the requested vnode 2205 * and with all vnodes aliased to the requested vnode. 2206 */ 2207int 2208vop_revoke(ap) 2209 struct vop_revoke_args /* { 2210 struct vnode *a_vp; 2211 int a_flags; 2212 } */ *ap; 2213{ 2214 struct vnode *vp, *vq; 2215 dev_t dev; 2216 2217 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 2218 2219 vp = ap->a_vp; 2220 /* 2221 * If a vgone (or vclean) is already in progress, 2222 * wait until it is done and return. 2223 */ 2224 if (vp->v_flag & VXLOCK) { 2225 vp->v_flag |= VXWANT; 2226 msleep(vp, &vp->v_interlock, PINOD | PDROP, 2227 "vop_revokeall", 0); 2228 return (0); 2229 } 2230 dev = vp->v_rdev; 2231 for (;;) { 2232 mtx_lock(&spechash_mtx); 2233 vq = SLIST_FIRST(&dev->si_hlist); 2234 mtx_unlock(&spechash_mtx); 2235 if (!vq) 2236 break; 2237 vgone(vq); 2238 } 2239 return (0); 2240} 2241 2242/* 2243 * Recycle an unused vnode to the front of the free list. 2244 * Release the passed interlock if the vnode will be recycled. 2245 */ 2246int 2247vrecycle(vp, inter_lkp, td) 2248 struct vnode *vp; 2249 struct mtx *inter_lkp; 2250 struct thread *td; 2251{ 2252 2253 mtx_lock(&vp->v_interlock); 2254 if (vp->v_usecount == 0) { 2255 if (inter_lkp) { 2256 mtx_unlock(inter_lkp); 2257 } 2258 vgonel(vp, td); 2259 return (1); 2260 } 2261 mtx_unlock(&vp->v_interlock); 2262 return (0); 2263} 2264 2265/* 2266 * Eliminate all activity associated with a vnode 2267 * in preparation for reuse. 2268 */ 2269void 2270vgone(vp) 2271 register struct vnode *vp; 2272{ 2273 struct thread *td = curthread; /* XXX */ 2274 2275 mtx_lock(&vp->v_interlock); 2276 vgonel(vp, td); 2277} 2278 2279/* 2280 * vgone, with the vp interlock held. 2281 */ 2282void 2283vgonel(vp, td) 2284 struct vnode *vp; 2285 struct thread *td; 2286{ 2287 int s; 2288 2289 /* 2290 * If a vgone (or vclean) is already in progress, 2291 * wait until it is done and return. 2292 */ 2293 if (vp->v_flag & VXLOCK) { 2294 vp->v_flag |= VXWANT; 2295 msleep(vp, &vp->v_interlock, PINOD | PDROP, "vgone", 0); 2296 return; 2297 } 2298 2299 /* 2300 * Clean out the filesystem specific data. 2301 */ 2302 vclean(vp, DOCLOSE, td); 2303 mtx_lock(&vp->v_interlock); 2304 2305 /* 2306 * Delete from old mount point vnode list, if on one. 2307 */ 2308 if (vp->v_mount != NULL) 2309 insmntque(vp, (struct mount *)0); 2310 /* 2311 * If special device, remove it from special device alias list 2312 * if it is on one. 2313 */ 2314 if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) { 2315 mtx_lock(&spechash_mtx); 2316 SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext); 2317 freedev(vp->v_rdev); 2318 mtx_unlock(&spechash_mtx); 2319 vp->v_rdev = NULL; 2320 } 2321 2322 /* 2323 * If it is on the freelist and not already at the head, 2324 * move it to the head of the list. The test of the 2325 * VDOOMED flag and the reference count of zero is because 2326 * it will be removed from the free list by getnewvnode, 2327 * but will not have its reference count incremented until 2328 * after calling vgone. If the reference count were 2329 * incremented first, vgone would (incorrectly) try to 2330 * close the previous instance of the underlying object. 2331 */ 2332 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 2333 s = splbio(); 2334 mtx_lock(&vnode_free_list_mtx); 2335 if (vp->v_flag & VFREE) 2336 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2337 else 2338 freevnodes++; 2339 vp->v_flag |= VFREE; 2340 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2341 mtx_unlock(&vnode_free_list_mtx); 2342 splx(s); 2343 } 2344 2345 vp->v_type = VBAD; 2346 mtx_unlock(&vp->v_interlock); 2347} 2348 2349/* 2350 * Lookup a vnode by device number. 2351 */ 2352int 2353vfinddev(dev, type, vpp) 2354 dev_t dev; 2355 enum vtype type; 2356 struct vnode **vpp; 2357{ 2358 struct vnode *vp; 2359 2360 mtx_lock(&spechash_mtx); 2361 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 2362 if (type == vp->v_type) { 2363 *vpp = vp; 2364 mtx_unlock(&spechash_mtx); 2365 return (1); 2366 } 2367 } 2368 mtx_unlock(&spechash_mtx); 2369 return (0); 2370} 2371 2372/* 2373 * Calculate the total number of references to a special device. 2374 */ 2375int 2376vcount(vp) 2377 struct vnode *vp; 2378{ 2379 struct vnode *vq; 2380 int count; 2381 2382 count = 0; 2383 mtx_lock(&spechash_mtx); 2384 SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext) 2385 count += vq->v_usecount; 2386 mtx_unlock(&spechash_mtx); 2387 return (count); 2388} 2389 2390/* 2391 * Same as above, but using the dev_t as argument 2392 */ 2393int 2394count_dev(dev) 2395 dev_t dev; 2396{ 2397 struct vnode *vp; 2398 2399 vp = SLIST_FIRST(&dev->si_hlist); 2400 if (vp == NULL) 2401 return (0); 2402 return(vcount(vp)); 2403} 2404 2405/* 2406 * Print out a description of a vnode. 2407 */ 2408static char *typename[] = 2409{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 2410 2411void 2412vprint(label, vp) 2413 char *label; 2414 struct vnode *vp; 2415{ 2416 char buf[96]; 2417 2418 if (label != NULL) 2419 printf("%s: %p: ", label, (void *)vp); 2420 else 2421 printf("%p: ", (void *)vp); 2422 printf("type %s, usecount %d, writecount %d, refcount %d,", 2423 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 2424 vp->v_holdcnt); 2425 buf[0] = '\0'; 2426 if (vp->v_flag & VROOT) 2427 strcat(buf, "|VROOT"); 2428 if (vp->v_flag & VTEXT) 2429 strcat(buf, "|VTEXT"); 2430 if (vp->v_flag & VSYSTEM) 2431 strcat(buf, "|VSYSTEM"); 2432 if (vp->v_flag & VXLOCK) 2433 strcat(buf, "|VXLOCK"); 2434 if (vp->v_flag & VXWANT) 2435 strcat(buf, "|VXWANT"); 2436 if (vp->v_flag & VBWAIT) 2437 strcat(buf, "|VBWAIT"); 2438 if (vp->v_flag & VDOOMED) 2439 strcat(buf, "|VDOOMED"); 2440 if (vp->v_flag & VFREE) 2441 strcat(buf, "|VFREE"); 2442 if (vp->v_flag & VOBJBUF) 2443 strcat(buf, "|VOBJBUF"); 2444 if (buf[0] != '\0') 2445 printf(" flags (%s)", &buf[1]); 2446 if (vp->v_data == NULL) { 2447 printf("\n"); 2448 } else { 2449 printf("\n\t"); 2450 VOP_PRINT(vp); 2451 } 2452} 2453 2454#ifdef DDB 2455#include <ddb/ddb.h> 2456/* 2457 * List all of the locked vnodes in the system. 2458 * Called when debugging the kernel. 2459 */ 2460DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2461{ 2462 struct thread *td = curthread; /* XXX */ 2463 struct mount *mp, *nmp; 2464 struct vnode *vp; 2465 2466 printf("Locked vnodes\n"); 2467 mtx_lock(&mountlist_mtx); 2468 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2469 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) { 2470 nmp = TAILQ_NEXT(mp, mnt_list); 2471 continue; 2472 } 2473 mtx_lock(&mntvnode_mtx); 2474 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2475 if (VOP_ISLOCKED(vp, NULL)) 2476 vprint((char *)0, vp); 2477 } 2478 mtx_unlock(&mntvnode_mtx); 2479 mtx_lock(&mountlist_mtx); 2480 nmp = TAILQ_NEXT(mp, mnt_list); 2481 vfs_unbusy(mp, td); 2482 } 2483 mtx_unlock(&mountlist_mtx); 2484} 2485#endif 2486 2487/* 2488 * Top level filesystem related information gathering. 2489 */ 2490static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 2491 2492static int 2493vfs_sysctl(SYSCTL_HANDLER_ARGS) 2494{ 2495 int *name = (int *)arg1 - 1; /* XXX */ 2496 u_int namelen = arg2 + 1; /* XXX */ 2497 struct vfsconf *vfsp; 2498 2499#if 1 || defined(COMPAT_PRELITE2) 2500 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2501 if (namelen == 1) 2502 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2503#endif 2504 2505 /* XXX the below code does not compile; vfs_sysctl does not exist. */ 2506#ifdef notyet 2507 /* all sysctl names at this level are at least name and field */ 2508 if (namelen < 2) 2509 return (ENOTDIR); /* overloaded */ 2510 if (name[0] != VFS_GENERIC) { 2511 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2512 if (vfsp->vfc_typenum == name[0]) 2513 break; 2514 if (vfsp == NULL) 2515 return (EOPNOTSUPP); 2516 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 2517 oldp, oldlenp, newp, newlen, td)); 2518 } 2519#endif 2520 switch (name[1]) { 2521 case VFS_MAXTYPENUM: 2522 if (namelen != 2) 2523 return (ENOTDIR); 2524 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2525 case VFS_CONF: 2526 if (namelen != 3) 2527 return (ENOTDIR); /* overloaded */ 2528 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2529 if (vfsp->vfc_typenum == name[2]) 2530 break; 2531 if (vfsp == NULL) 2532 return (EOPNOTSUPP); 2533 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 2534 } 2535 return (EOPNOTSUPP); 2536} 2537 2538SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 2539 "Generic filesystem"); 2540 2541#if 1 || defined(COMPAT_PRELITE2) 2542 2543static int 2544sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2545{ 2546 int error; 2547 struct vfsconf *vfsp; 2548 struct ovfsconf ovfs; 2549 2550 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 2551 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2552 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2553 ovfs.vfc_index = vfsp->vfc_typenum; 2554 ovfs.vfc_refcount = vfsp->vfc_refcount; 2555 ovfs.vfc_flags = vfsp->vfc_flags; 2556 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2557 if (error) 2558 return error; 2559 } 2560 return 0; 2561} 2562 2563#endif /* 1 || COMPAT_PRELITE2 */ 2564 2565#if COMPILING_LINT 2566#define KINFO_VNODESLOP 10 2567/* 2568 * Dump vnode list (via sysctl). 2569 * Copyout address of vnode followed by vnode. 2570 */ 2571/* ARGSUSED */ 2572static int 2573sysctl_vnode(SYSCTL_HANDLER_ARGS) 2574{ 2575 struct thread *td = curthread; /* XXX */ 2576 struct mount *mp, *nmp; 2577 struct vnode *nvp, *vp; 2578 int error; 2579 2580#define VPTRSZ sizeof (struct vnode *) 2581#define VNODESZ sizeof (struct vnode) 2582 2583 req->lock = 0; 2584 if (!req->oldptr) /* Make an estimate */ 2585 return (SYSCTL_OUT(req, 0, 2586 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 2587 2588 mtx_lock(&mountlist_mtx); 2589 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2590 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) { 2591 nmp = TAILQ_NEXT(mp, mnt_list); 2592 continue; 2593 } 2594 mtx_lock(&mntvnode_mtx); 2595again: 2596 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2597 vp != NULL; 2598 vp = nvp) { 2599 /* 2600 * Check that the vp is still associated with 2601 * this filesystem. RACE: could have been 2602 * recycled onto the same filesystem. 2603 */ 2604 if (vp->v_mount != mp) 2605 goto again; 2606 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2607 mtx_unlock(&mntvnode_mtx); 2608 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 2609 (error = SYSCTL_OUT(req, vp, VNODESZ))) 2610 return (error); 2611 mtx_lock(&mntvnode_mtx); 2612 } 2613 mtx_unlock(&mntvnode_mtx); 2614 mtx_lock(&mountlist_mtx); 2615 nmp = TAILQ_NEXT(mp, mnt_list); 2616 vfs_unbusy(mp, td); 2617 } 2618 mtx_unlock(&mountlist_mtx); 2619 2620 return (0); 2621} 2622 2623/* 2624 * XXX 2625 * Exporting the vnode list on large systems causes them to crash. 2626 * Exporting the vnode list on medium systems causes sysctl to coredump. 2627 */ 2628SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 2629 0, 0, sysctl_vnode, "S,vnode", ""); 2630#endif 2631 2632/* 2633 * Check to see if a filesystem is mounted on a block device. 2634 */ 2635int 2636vfs_mountedon(vp) 2637 struct vnode *vp; 2638{ 2639 2640 if (vp->v_rdev->si_mountpoint != NULL) 2641 return (EBUSY); 2642 return (0); 2643} 2644 2645/* 2646 * Unmount all filesystems. The list is traversed in reverse order 2647 * of mounting to avoid dependencies. 2648 */ 2649void 2650vfs_unmountall() 2651{ 2652 struct mount *mp; 2653 struct thread *td; 2654 int error; 2655 2656 if (curthread != NULL) 2657 td = curthread; 2658 else 2659 td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */ 2660 /* 2661 * Since this only runs when rebooting, it is not interlocked. 2662 */ 2663 while(!TAILQ_EMPTY(&mountlist)) { 2664 mp = TAILQ_LAST(&mountlist, mntlist); 2665 error = dounmount(mp, MNT_FORCE, td); 2666 if (error) { 2667 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2668 printf("unmount of %s failed (", 2669 mp->mnt_stat.f_mntonname); 2670 if (error == EBUSY) 2671 printf("BUSY)\n"); 2672 else 2673 printf("%d)\n", error); 2674 } else { 2675 /* The unmount has removed mp from the mountlist */ 2676 } 2677 } 2678} 2679 2680/* 2681 * perform msync on all vnodes under a mount point 2682 * the mount point must be locked. 2683 */ 2684void 2685vfs_msync(struct mount *mp, int flags) 2686{ 2687 struct vnode *vp, *nvp; 2688 struct vm_object *obj; 2689 int tries; 2690 2691 GIANT_REQUIRED; 2692 2693 tries = 5; 2694 mtx_lock(&mntvnode_mtx); 2695loop: 2696 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) { 2697 if (vp->v_mount != mp) { 2698 if (--tries > 0) 2699 goto loop; 2700 break; 2701 } 2702 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2703 2704 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */ 2705 continue; 2706 2707 if (vp->v_flag & VNOSYNC) /* unlinked, skip it */ 2708 continue; 2709 2710 if ((vp->v_flag & VOBJDIRTY) && 2711 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 2712 mtx_unlock(&mntvnode_mtx); 2713 if (!vget(vp, 2714 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curthread)) { 2715 if (VOP_GETVOBJECT(vp, &obj) == 0) { 2716 vm_object_page_clean(obj, 0, 0, 2717 flags == MNT_WAIT ? 2718 OBJPC_SYNC : OBJPC_NOSYNC); 2719 } 2720 vput(vp); 2721 } 2722 mtx_lock(&mntvnode_mtx); 2723 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) { 2724 if (--tries > 0) 2725 goto loop; 2726 break; 2727 } 2728 } 2729 } 2730 mtx_unlock(&mntvnode_mtx); 2731} 2732 2733/* 2734 * Create the VM object needed for VMIO and mmap support. This 2735 * is done for all VREG files in the system. Some filesystems might 2736 * afford the additional metadata buffering capability of the 2737 * VMIO code by making the device node be VMIO mode also. 2738 * 2739 * vp must be locked when vfs_object_create is called. 2740 */ 2741int 2742vfs_object_create(vp, td, cred) 2743 struct vnode *vp; 2744 struct thread *td; 2745 struct ucred *cred; 2746{ 2747 GIANT_REQUIRED; 2748 return (VOP_CREATEVOBJECT(vp, cred, td)); 2749} 2750 2751/* 2752 * Mark a vnode as free, putting it up for recycling. 2753 */ 2754void 2755vfree(vp) 2756 struct vnode *vp; 2757{ 2758 int s; 2759 2760 s = splbio(); 2761 mtx_lock(&vnode_free_list_mtx); 2762 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free")); 2763 if (vp->v_flag & VAGE) { 2764 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2765 } else { 2766 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2767 } 2768 freevnodes++; 2769 mtx_unlock(&vnode_free_list_mtx); 2770 vp->v_flag &= ~VAGE; 2771 vp->v_flag |= VFREE; 2772 splx(s); 2773} 2774 2775/* 2776 * Opposite of vfree() - mark a vnode as in use. 2777 */ 2778void 2779vbusy(vp) 2780 struct vnode *vp; 2781{ 2782 int s; 2783 2784 s = splbio(); 2785 mtx_lock(&vnode_free_list_mtx); 2786 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free")); 2787 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2788 freevnodes--; 2789 mtx_unlock(&vnode_free_list_mtx); 2790 vp->v_flag &= ~(VFREE|VAGE); 2791 splx(s); 2792} 2793 2794/* 2795 * Record a process's interest in events which might happen to 2796 * a vnode. Because poll uses the historic select-style interface 2797 * internally, this routine serves as both the ``check for any 2798 * pending events'' and the ``record my interest in future events'' 2799 * functions. (These are done together, while the lock is held, 2800 * to avoid race conditions.) 2801 */ 2802int 2803vn_pollrecord(vp, td, events) 2804 struct vnode *vp; 2805 struct thread *td; 2806 short events; 2807{ 2808 2809 if (vp->v_pollinfo == NULL) 2810 v_addpollinfo(vp); 2811 mtx_lock(&vp->v_pollinfo->vpi_lock); 2812 if (vp->v_pollinfo->vpi_revents & events) { 2813 /* 2814 * This leaves events we are not interested 2815 * in available for the other process which 2816 * which presumably had requested them 2817 * (otherwise they would never have been 2818 * recorded). 2819 */ 2820 events &= vp->v_pollinfo->vpi_revents; 2821 vp->v_pollinfo->vpi_revents &= ~events; 2822 2823 mtx_unlock(&vp->v_pollinfo->vpi_lock); 2824 return events; 2825 } 2826 vp->v_pollinfo->vpi_events |= events; 2827 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 2828 mtx_unlock(&vp->v_pollinfo->vpi_lock); 2829 return 0; 2830} 2831 2832/* 2833 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2834 * it is possible for us to miss an event due to race conditions, but 2835 * that condition is expected to be rare, so for the moment it is the 2836 * preferred interface. 2837 */ 2838void 2839vn_pollevent(vp, events) 2840 struct vnode *vp; 2841 short events; 2842{ 2843 2844 if (vp->v_pollinfo == NULL) 2845 v_addpollinfo(vp); 2846 mtx_lock(&vp->v_pollinfo->vpi_lock); 2847 if (vp->v_pollinfo->vpi_events & events) { 2848 /* 2849 * We clear vpi_events so that we don't 2850 * call selwakeup() twice if two events are 2851 * posted before the polling process(es) is 2852 * awakened. This also ensures that we take at 2853 * most one selwakeup() if the polling process 2854 * is no longer interested. However, it does 2855 * mean that only one event can be noticed at 2856 * a time. (Perhaps we should only clear those 2857 * event bits which we note?) XXX 2858 */ 2859 vp->v_pollinfo->vpi_events = 0; /* &= ~events ??? */ 2860 vp->v_pollinfo->vpi_revents |= events; 2861 selwakeup(&vp->v_pollinfo->vpi_selinfo); 2862 } 2863 mtx_unlock(&vp->v_pollinfo->vpi_lock); 2864} 2865 2866/* 2867 * Wake up anyone polling on vp because it is being revoked. 2868 * This depends on dead_poll() returning POLLHUP for correct 2869 * behavior. 2870 */ 2871void 2872vn_pollgone(vp) 2873 struct vnode *vp; 2874{ 2875 2876 mtx_lock(&vp->v_pollinfo->vpi_lock); 2877 VN_KNOTE(vp, NOTE_REVOKE); 2878 if (vp->v_pollinfo->vpi_events) { 2879 vp->v_pollinfo->vpi_events = 0; 2880 selwakeup(&vp->v_pollinfo->vpi_selinfo); 2881 } 2882 mtx_unlock(&vp->v_pollinfo->vpi_lock); 2883} 2884 2885 2886 2887/* 2888 * Routine to create and manage a filesystem syncer vnode. 2889 */ 2890#define sync_close ((int (*)(struct vop_close_args *))nullop) 2891static int sync_fsync(struct vop_fsync_args *); 2892static int sync_inactive(struct vop_inactive_args *); 2893static int sync_reclaim(struct vop_reclaim_args *); 2894#define sync_lock ((int (*)(struct vop_lock_args *))vop_nolock) 2895#define sync_unlock ((int (*)(struct vop_unlock_args *))vop_nounlock) 2896static int sync_print(struct vop_print_args *); 2897#define sync_islocked ((int(*)(struct vop_islocked_args *))vop_noislocked) 2898 2899static vop_t **sync_vnodeop_p; 2900static struct vnodeopv_entry_desc sync_vnodeop_entries[] = { 2901 { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 2902 { &vop_close_desc, (vop_t *) sync_close }, /* close */ 2903 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */ 2904 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */ 2905 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */ 2906 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */ 2907 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */ 2908 { &vop_print_desc, (vop_t *) sync_print }, /* print */ 2909 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */ 2910 { NULL, NULL } 2911}; 2912static struct vnodeopv_desc sync_vnodeop_opv_desc = 2913 { &sync_vnodeop_p, sync_vnodeop_entries }; 2914 2915VNODEOP_SET(sync_vnodeop_opv_desc); 2916 2917/* 2918 * Create a new filesystem syncer vnode for the specified mount point. 2919 */ 2920int 2921vfs_allocate_syncvnode(mp) 2922 struct mount *mp; 2923{ 2924 struct vnode *vp; 2925 static long start, incr, next; 2926 int error; 2927 2928 /* Allocate a new vnode */ 2929 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) { 2930 mp->mnt_syncer = NULL; 2931 return (error); 2932 } 2933 vp->v_type = VNON; 2934 /* 2935 * Place the vnode onto the syncer worklist. We attempt to 2936 * scatter them about on the list so that they will go off 2937 * at evenly distributed times even if all the filesystems 2938 * are mounted at once. 2939 */ 2940 next += incr; 2941 if (next == 0 || next > syncer_maxdelay) { 2942 start /= 2; 2943 incr /= 2; 2944 if (start == 0) { 2945 start = syncer_maxdelay / 2; 2946 incr = syncer_maxdelay; 2947 } 2948 next = start; 2949 } 2950 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0); 2951 mp->mnt_syncer = vp; 2952 return (0); 2953} 2954 2955/* 2956 * Do a lazy sync of the filesystem. 2957 */ 2958static int 2959sync_fsync(ap) 2960 struct vop_fsync_args /* { 2961 struct vnode *a_vp; 2962 struct ucred *a_cred; 2963 int a_waitfor; 2964 struct thread *a_td; 2965 } */ *ap; 2966{ 2967 struct vnode *syncvp = ap->a_vp; 2968 struct mount *mp = syncvp->v_mount; 2969 struct thread *td = ap->a_td; 2970 int asyncflag; 2971 2972 /* 2973 * We only need to do something if this is a lazy evaluation. 2974 */ 2975 if (ap->a_waitfor != MNT_LAZY) 2976 return (0); 2977 2978 /* 2979 * Move ourselves to the back of the sync list. 2980 */ 2981 vn_syncer_add_to_worklist(syncvp, syncdelay); 2982 2983 /* 2984 * Walk the list of vnodes pushing all that are dirty and 2985 * not already on the sync list. 2986 */ 2987 mtx_lock(&mountlist_mtx); 2988 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) { 2989 mtx_unlock(&mountlist_mtx); 2990 return (0); 2991 } 2992 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 2993 vfs_unbusy(mp, td); 2994 return (0); 2995 } 2996 asyncflag = mp->mnt_flag & MNT_ASYNC; 2997 mp->mnt_flag &= ~MNT_ASYNC; 2998 vfs_msync(mp, MNT_NOWAIT); 2999 VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td); 3000 if (asyncflag) 3001 mp->mnt_flag |= MNT_ASYNC; 3002 vn_finished_write(mp); 3003 vfs_unbusy(mp, td); 3004 return (0); 3005} 3006 3007/* 3008 * The syncer vnode is no referenced. 3009 */ 3010static int 3011sync_inactive(ap) 3012 struct vop_inactive_args /* { 3013 struct vnode *a_vp; 3014 struct thread *a_td; 3015 } */ *ap; 3016{ 3017 3018 vgone(ap->a_vp); 3019 return (0); 3020} 3021 3022/* 3023 * The syncer vnode is no longer needed and is being decommissioned. 3024 * 3025 * Modifications to the worklist must be protected at splbio(). 3026 */ 3027static int 3028sync_reclaim(ap) 3029 struct vop_reclaim_args /* { 3030 struct vnode *a_vp; 3031 } */ *ap; 3032{ 3033 struct vnode *vp = ap->a_vp; 3034 int s; 3035 3036 s = splbio(); 3037 vp->v_mount->mnt_syncer = NULL; 3038 if (vp->v_flag & VONWORKLST) { 3039 LIST_REMOVE(vp, v_synclist); 3040 vp->v_flag &= ~VONWORKLST; 3041 } 3042 splx(s); 3043 3044 return (0); 3045} 3046 3047/* 3048 * Print out a syncer vnode. 3049 */ 3050static int 3051sync_print(ap) 3052 struct vop_print_args /* { 3053 struct vnode *a_vp; 3054 } */ *ap; 3055{ 3056 struct vnode *vp = ap->a_vp; 3057 3058 printf("syncer vnode"); 3059 if (vp->v_vnlock != NULL) 3060 lockmgr_printinfo(vp->v_vnlock); 3061 printf("\n"); 3062 return (0); 3063} 3064 3065/* 3066 * extract the dev_t from a VCHR 3067 */ 3068dev_t 3069vn_todev(vp) 3070 struct vnode *vp; 3071{ 3072 if (vp->v_type != VCHR) 3073 return (NODEV); 3074 return (vp->v_rdev); 3075} 3076 3077/* 3078 * Check if vnode represents a disk device 3079 */ 3080int 3081vn_isdisk(vp, errp) 3082 struct vnode *vp; 3083 int *errp; 3084{ 3085 struct cdevsw *cdevsw; 3086 3087 if (vp->v_type != VCHR) { 3088 if (errp != NULL) 3089 *errp = ENOTBLK; 3090 return (0); 3091 } 3092 if (vp->v_rdev == NULL) { 3093 if (errp != NULL) 3094 *errp = ENXIO; 3095 return (0); 3096 } 3097 cdevsw = devsw(vp->v_rdev); 3098 if (cdevsw == NULL) { 3099 if (errp != NULL) 3100 *errp = ENXIO; 3101 return (0); 3102 } 3103 if (!(cdevsw->d_flags & D_DISK)) { 3104 if (errp != NULL) 3105 *errp = ENOTBLK; 3106 return (0); 3107 } 3108 if (errp != NULL) 3109 *errp = 0; 3110 return (1); 3111} 3112 3113/* 3114 * Free data allocated by namei(); see namei(9) for details. 3115 */ 3116void 3117NDFREE(ndp, flags) 3118 struct nameidata *ndp; 3119 const uint flags; 3120{ 3121 if (!(flags & NDF_NO_FREE_PNBUF) && 3122 (ndp->ni_cnd.cn_flags & HASBUF)) { 3123 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 3124 ndp->ni_cnd.cn_flags &= ~HASBUF; 3125 } 3126 if (!(flags & NDF_NO_DVP_UNLOCK) && 3127 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 3128 ndp->ni_dvp != ndp->ni_vp) 3129 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread); 3130 if (!(flags & NDF_NO_DVP_RELE) && 3131 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 3132 vrele(ndp->ni_dvp); 3133 ndp->ni_dvp = NULL; 3134 } 3135 if (!(flags & NDF_NO_VP_UNLOCK) && 3136 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 3137 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread); 3138 if (!(flags & NDF_NO_VP_RELE) && 3139 ndp->ni_vp) { 3140 vrele(ndp->ni_vp); 3141 ndp->ni_vp = NULL; 3142 } 3143 if (!(flags & NDF_NO_STARTDIR_RELE) && 3144 (ndp->ni_cnd.cn_flags & SAVESTART)) { 3145 vrele(ndp->ni_startdir); 3146 ndp->ni_startdir = NULL; 3147 } 3148} 3149 3150/* 3151 * Common filesystem object access control check routine. Accepts a 3152 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3153 * and optional call-by-reference privused argument allowing vaccess() 3154 * to indicate to the caller whether privilege was used to satisfy the 3155 * request. Returns 0 on success, or an errno on failure. 3156 */ 3157int 3158vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused) 3159 enum vtype type; 3160 mode_t file_mode; 3161 uid_t file_uid; 3162 gid_t file_gid; 3163 mode_t acc_mode; 3164 struct ucred *cred; 3165 int *privused; 3166{ 3167 mode_t dac_granted; 3168#ifdef CAPABILITIES 3169 mode_t cap_granted; 3170#endif 3171 3172 /* 3173 * Look for a normal, non-privileged way to access the file/directory 3174 * as requested. If it exists, go with that. 3175 */ 3176 3177 if (privused != NULL) 3178 *privused = 0; 3179 3180 dac_granted = 0; 3181 3182 /* Check the owner. */ 3183 if (cred->cr_uid == file_uid) { 3184 dac_granted |= VADMIN; 3185 if (file_mode & S_IXUSR) 3186 dac_granted |= VEXEC; 3187 if (file_mode & S_IRUSR) 3188 dac_granted |= VREAD; 3189 if (file_mode & S_IWUSR) 3190 dac_granted |= VWRITE; 3191 3192 if ((acc_mode & dac_granted) == acc_mode) 3193 return (0); 3194 3195 goto privcheck; 3196 } 3197 3198 /* Otherwise, check the groups (first match) */ 3199 if (groupmember(file_gid, cred)) { 3200 if (file_mode & S_IXGRP) 3201 dac_granted |= VEXEC; 3202 if (file_mode & S_IRGRP) 3203 dac_granted |= VREAD; 3204 if (file_mode & S_IWGRP) 3205 dac_granted |= VWRITE; 3206 3207 if ((acc_mode & dac_granted) == acc_mode) 3208 return (0); 3209 3210 goto privcheck; 3211 } 3212 3213 /* Otherwise, check everyone else. */ 3214 if (file_mode & S_IXOTH) 3215 dac_granted |= VEXEC; 3216 if (file_mode & S_IROTH) 3217 dac_granted |= VREAD; 3218 if (file_mode & S_IWOTH) 3219 dac_granted |= VWRITE; 3220 if ((acc_mode & dac_granted) == acc_mode) 3221 return (0); 3222 3223privcheck: 3224 if (!suser_cred(cred, PRISON_ROOT)) { 3225 /* XXX audit: privilege used */ 3226 if (privused != NULL) 3227 *privused = 1; 3228 return (0); 3229 } 3230 3231#ifdef CAPABILITIES 3232 /* 3233 * Build a capability mask to determine if the set of capabilities 3234 * satisfies the requirements when combined with the granted mask 3235 * from above. 3236 * For each capability, if the capability is required, bitwise 3237 * or the request type onto the cap_granted mask. 3238 */ 3239 cap_granted = 0; 3240 3241 if (type == VDIR) { 3242 /* 3243 * For directories, use CAP_DAC_READ_SEARCH to satisfy 3244 * VEXEC requests, instead of CAP_DAC_EXECUTE. 3245 */ 3246 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3247 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT)) 3248 cap_granted |= VEXEC; 3249 } else { 3250 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3251 !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT)) 3252 cap_granted |= VEXEC; 3253 } 3254 3255 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) && 3256 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT)) 3257 cap_granted |= VREAD; 3258 3259 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3260 !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT)) 3261 cap_granted |= VWRITE; 3262 3263 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3264 !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT)) 3265 cap_granted |= VADMIN; 3266 3267 if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) { 3268 /* XXX audit: privilege used */ 3269 if (privused != NULL) 3270 *privused = 1; 3271 return (0); 3272 } 3273#endif 3274 3275 return ((acc_mode & VADMIN) ? EPERM : EACCES); 3276} 3277