FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_subr.c
1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
35 */
36
37 /*
38 * External virtual filesystem routines
39 */
40
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD: releng/10.4/sys/kern/vfs_subr.c 316543 2017-04-05 21:08:26Z brooks $");
43
44 #include "opt_compat.h"
45 #include "opt_ddb.h"
46 #include "opt_watchdog.h"
47
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/condvar.h>
53 #include <sys/conf.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
58 #include <sys/file.h>
59 #include <sys/fcntl.h>
60 #include <sys/jail.h>
61 #include <sys/kdb.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
68 #include <sys/pctrie.h>
69 #include <sys/priv.h>
70 #include <sys/reboot.h>
71 #include <sys/rwlock.h>
72 #include <sys/sched.h>
73 #include <sys/sleepqueue.h>
74 #include <sys/smp.h>
75 #include <sys/stat.h>
76 #include <sys/sysctl.h>
77 #include <sys/syslog.h>
78 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/watchdog.h>
81
82 #include <machine/stdarg.h>
83
84 #include <security/mac/mac_framework.h>
85
86 #include <vm/vm.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_extern.h>
89 #include <vm/pmap.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_kern.h>
93 #include <vm/uma.h>
94
95 #ifdef DDB
96 #include <ddb/ddb.h>
97 #endif
98
99 static void delmntque(struct vnode *vp);
100 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 int slpflag, int slptimeo);
102 static void syncer_shutdown(void *arg, int howto);
103 static int vtryrecycle(struct vnode *vp);
104 static void v_incr_usecount(struct vnode *);
105 static void v_decr_usecount(struct vnode *);
106 static void v_decr_useonly(struct vnode *);
107 static void v_upgrade_usecount(struct vnode *);
108 static void vnlru_free(int);
109 static void vgonel(struct vnode *);
110 static void vfs_knllock(void *arg);
111 static void vfs_knlunlock(void *arg);
112 static void vfs_knl_assert_locked(void *arg);
113 static void vfs_knl_assert_unlocked(void *arg);
114 static void destroy_vpollinfo(struct vpollinfo *vi);
115
116 /*
117 * Number of vnodes in existence. Increased whenever getnewvnode()
118 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
119 */
120 static unsigned long numvnodes;
121
122 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123 "Number of vnodes in existence");
124
125 static u_long vnodes_created;
126 SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
127 0, "Number of vnodes created by getnewvnode");
128
129 /*
130 * Conversion tables for conversion from vnode types to inode formats
131 * and back.
132 */
133 enum vtype iftovt_tab[16] = {
134 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
135 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
136 };
137 int vttoif_tab[10] = {
138 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
139 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
140 };
141
142 /*
143 * List of vnodes that are ready for recycling.
144 */
145 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
146
147 /*
148 * Free vnode target. Free vnodes may simply be files which have been stat'd
149 * but not read. This is somewhat common, and a small cache of such files
150 * should be kept to avoid recreation costs.
151 */
152 static u_long wantfreevnodes;
153 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
154 /* Number of vnodes in the free list. */
155 static u_long freevnodes;
156 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
157 "Number of vnodes in the free list");
158
159 static int vlru_allow_cache_src;
160 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
161 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
162
163 static u_long recycles_count;
164 SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
165 "Number of vnodes recycled to avoid exceding kern.maxvnodes");
166
167 /*
168 * Various variables used for debugging the new implementation of
169 * reassignbuf().
170 * XXX these are probably of (very) limited utility now.
171 */
172 static int reassignbufcalls;
173 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
174 "Number of calls to reassignbuf");
175
176 static u_long free_owe_inact;
177 SYSCTL_ULONG(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, 0,
178 "Number of times free vnodes kept on active list due to VFS "
179 "owing inactivation");
180
181 /*
182 * Cache for the mount type id assigned to NFS. This is used for
183 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
184 */
185 int nfs_mount_type = -1;
186
187 /* To keep more than one thread at a time from running vfs_getnewfsid */
188 static struct mtx mntid_mtx;
189
190 /*
191 * Lock for any access to the following:
192 * vnode_free_list
193 * numvnodes
194 * freevnodes
195 */
196 static struct mtx vnode_free_list_mtx;
197
198 /* Publicly exported FS */
199 struct nfs_public nfs_pub;
200
201 static uma_zone_t buf_trie_zone;
202
203 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
204 static uma_zone_t vnode_zone;
205 static uma_zone_t vnodepoll_zone;
206
207 /*
208 * The workitem queue.
209 *
210 * It is useful to delay writes of file data and filesystem metadata
211 * for tens of seconds so that quickly created and deleted files need
212 * not waste disk bandwidth being created and removed. To realize this,
213 * we append vnodes to a "workitem" queue. When running with a soft
214 * updates implementation, most pending metadata dependencies should
215 * not wait for more than a few seconds. Thus, mounted on block devices
216 * are delayed only about a half the time that file data is delayed.
217 * Similarly, directory updates are more critical, so are only delayed
218 * about a third the time that file data is delayed. Thus, there are
219 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
220 * one each second (driven off the filesystem syncer process). The
221 * syncer_delayno variable indicates the next queue that is to be processed.
222 * Items that need to be processed soon are placed in this queue:
223 *
224 * syncer_workitem_pending[syncer_delayno]
225 *
226 * A delay of fifteen seconds is done by placing the request fifteen
227 * entries later in the queue:
228 *
229 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
230 *
231 */
232 static int syncer_delayno;
233 static long syncer_mask;
234 LIST_HEAD(synclist, bufobj);
235 static struct synclist *syncer_workitem_pending;
236 /*
237 * The sync_mtx protects:
238 * bo->bo_synclist
239 * sync_vnode_count
240 * syncer_delayno
241 * syncer_state
242 * syncer_workitem_pending
243 * syncer_worklist_len
244 * rushjob
245 */
246 static struct mtx sync_mtx;
247 static struct cv sync_wakeup;
248
249 #define SYNCER_MAXDELAY 32
250 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
251 static int syncdelay = 30; /* max time to delay syncing data */
252 static int filedelay = 30; /* time to delay syncing files */
253 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
254 "Time to delay syncing files (in seconds)");
255 static int dirdelay = 29; /* time to delay syncing directories */
256 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
257 "Time to delay syncing directories (in seconds)");
258 static int metadelay = 28; /* time to delay syncing metadata */
259 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
260 "Time to delay syncing metadata (in seconds)");
261 static int rushjob; /* number of slots to run ASAP */
262 static int stat_rush_requests; /* number of times I/O speeded up */
263 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
264 "Number of times I/O speeded up (rush requests)");
265
266 /*
267 * When shutting down the syncer, run it at four times normal speed.
268 */
269 #define SYNCER_SHUTDOWN_SPEEDUP 4
270 static int sync_vnode_count;
271 static int syncer_worklist_len;
272 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
273 syncer_state;
274
275 /*
276 * Number of vnodes we want to exist at any one time. This is mostly used
277 * to size hash tables in vnode-related code. It is normally not used in
278 * getnewvnode(), as wantfreevnodes is normally nonzero.)
279 *
280 * XXX desiredvnodes is historical cruft and should not exist.
281 */
282 int desiredvnodes;
283
284 static int
285 sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS)
286 {
287 int error, old_desiredvnodes;
288
289 old_desiredvnodes = desiredvnodes;
290 if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0)
291 return (error);
292 if (old_desiredvnodes != desiredvnodes) {
293 vfs_hash_changesize(desiredvnodes);
294 cache_changesize(desiredvnodes);
295 }
296 return (0);
297 }
298
299 SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes,
300 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0,
301 sysctl_update_desiredvnodes, "I", "Maximum number of vnodes");
302 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
303 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
304 static int vnlru_nowhere;
305 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
306 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
307
308 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
309 static int vnsz2log;
310
311 /*
312 * Support for the bufobj clean & dirty pctrie.
313 */
314 static void *
315 buf_trie_alloc(struct pctrie *ptree)
316 {
317
318 return uma_zalloc(buf_trie_zone, M_NOWAIT);
319 }
320
321 static void
322 buf_trie_free(struct pctrie *ptree, void *node)
323 {
324
325 uma_zfree(buf_trie_zone, node);
326 }
327 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
328
329 /*
330 * Initialize the vnode management data structures.
331 *
332 * Reevaluate the following cap on the number of vnodes after the physical
333 * memory size exceeds 512GB. In the limit, as the physical memory size
334 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
335 */
336 #ifndef MAXVNODES_MAX
337 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
338 #endif
339
340 /*
341 * Initialize a vnode as it first enters the zone.
342 */
343 static int
344 vnode_init(void *mem, int size, int flags)
345 {
346 struct vnode *vp;
347 struct bufobj *bo;
348
349 vp = mem;
350 bzero(vp, size);
351 /*
352 * Setup locks.
353 */
354 vp->v_vnlock = &vp->v_lock;
355 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
356 /*
357 * By default, don't allow shared locks unless filesystems opt-in.
358 */
359 lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT,
360 LK_NOSHARE | LK_IS_VNODE);
361 /*
362 * Initialize bufobj.
363 */
364 bo = &vp->v_bufobj;
365 bo->__bo_vnode = vp;
366 rw_init(BO_LOCKPTR(bo), "bufobj interlock");
367 bo->bo_private = vp;
368 TAILQ_INIT(&bo->bo_clean.bv_hd);
369 TAILQ_INIT(&bo->bo_dirty.bv_hd);
370 /*
371 * Initialize namecache.
372 */
373 LIST_INIT(&vp->v_cache_src);
374 TAILQ_INIT(&vp->v_cache_dst);
375 /*
376 * Initialize rangelocks.
377 */
378 rangelock_init(&vp->v_rl);
379 return (0);
380 }
381
382 /*
383 * Free a vnode when it is cleared from the zone.
384 */
385 static void
386 vnode_fini(void *mem, int size)
387 {
388 struct vnode *vp;
389 struct bufobj *bo;
390
391 vp = mem;
392 rangelock_destroy(&vp->v_rl);
393 lockdestroy(vp->v_vnlock);
394 mtx_destroy(&vp->v_interlock);
395 bo = &vp->v_bufobj;
396 rw_destroy(BO_LOCKPTR(bo));
397 }
398
399 static void
400 vntblinit(void *dummy __unused)
401 {
402 u_int i;
403 int physvnodes, virtvnodes;
404
405 /*
406 * Desiredvnodes is a function of the physical memory size and the
407 * kernel's heap size. Generally speaking, it scales with the
408 * physical memory size. The ratio of desiredvnodes to physical pages
409 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
410 * marginal ratio of desiredvnodes to physical pages is one to
411 * sixteen. However, desiredvnodes is limited by the kernel's heap
412 * size. The memory required by desiredvnodes vnodes and vm objects
413 * may not exceed one seventh of the kernel's heap size.
414 */
415 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
416 cnt.v_page_count) / 16;
417 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
418 sizeof(struct vnode)));
419 desiredvnodes = min(physvnodes, virtvnodes);
420 if (desiredvnodes > MAXVNODES_MAX) {
421 if (bootverbose)
422 printf("Reducing kern.maxvnodes %d -> %d\n",
423 desiredvnodes, MAXVNODES_MAX);
424 desiredvnodes = MAXVNODES_MAX;
425 }
426 wantfreevnodes = desiredvnodes / 4;
427 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
428 TAILQ_INIT(&vnode_free_list);
429 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
430 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
431 vnode_init, vnode_fini, UMA_ALIGN_PTR, 0);
432 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
433 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
434 /*
435 * Preallocate enough nodes to support one-per buf so that
436 * we can not fail an insert. reassignbuf() callers can not
437 * tolerate the insertion failure.
438 */
439 buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
440 NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
441 UMA_ZONE_NOFREE | UMA_ZONE_VM);
442 uma_prealloc(buf_trie_zone, nbuf);
443 /*
444 * Initialize the filesystem syncer.
445 */
446 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
447 &syncer_mask);
448 syncer_maxdelay = syncer_mask + 1;
449 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
450 cv_init(&sync_wakeup, "syncer");
451 for (i = 1; i <= sizeof(struct vnode); i <<= 1)
452 vnsz2log++;
453 vnsz2log--;
454 }
455 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
456
457
458 /*
459 * Mark a mount point as busy. Used to synchronize access and to delay
460 * unmounting. Eventually, mountlist_mtx is not released on failure.
461 *
462 * vfs_busy() is a custom lock, it can block the caller.
463 * vfs_busy() only sleeps if the unmount is active on the mount point.
464 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
465 * vnode belonging to mp.
466 *
467 * Lookup uses vfs_busy() to traverse mount points.
468 * root fs var fs
469 * / vnode lock A / vnode lock (/var) D
470 * /var vnode lock B /log vnode lock(/var/log) E
471 * vfs_busy lock C vfs_busy lock F
472 *
473 * Within each file system, the lock order is C->A->B and F->D->E.
474 *
475 * When traversing across mounts, the system follows that lock order:
476 *
477 * C->A->B
478 * |
479 * +->F->D->E
480 *
481 * The lookup() process for namei("/var") illustrates the process:
482 * VOP_LOOKUP() obtains B while A is held
483 * vfs_busy() obtains a shared lock on F while A and B are held
484 * vput() releases lock on B
485 * vput() releases lock on A
486 * VFS_ROOT() obtains lock on D while shared lock on F is held
487 * vfs_unbusy() releases shared lock on F
488 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
489 * Attempt to lock A (instead of vp_crossmp) while D is held would
490 * violate the global order, causing deadlocks.
491 *
492 * dounmount() locks B while F is drained.
493 */
494 int
495 vfs_busy(struct mount *mp, int flags)
496 {
497
498 MPASS((flags & ~MBF_MASK) == 0);
499 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
500
501 MNT_ILOCK(mp);
502 MNT_REF(mp);
503 /*
504 * If mount point is currently being unmounted, sleep until the
505 * mount point fate is decided. If thread doing the unmounting fails,
506 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
507 * that this mount point has survived the unmount attempt and vfs_busy
508 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
509 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
510 * about to be really destroyed. vfs_busy needs to release its
511 * reference on the mount point in this case and return with ENOENT,
512 * telling the caller that mount mount it tried to busy is no longer
513 * valid.
514 */
515 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
516 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
517 MNT_REL(mp);
518 MNT_IUNLOCK(mp);
519 CTR1(KTR_VFS, "%s: failed busying before sleeping",
520 __func__);
521 return (ENOENT);
522 }
523 if (flags & MBF_MNTLSTLOCK)
524 mtx_unlock(&mountlist_mtx);
525 mp->mnt_kern_flag |= MNTK_MWAIT;
526 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
527 if (flags & MBF_MNTLSTLOCK)
528 mtx_lock(&mountlist_mtx);
529 MNT_ILOCK(mp);
530 }
531 if (flags & MBF_MNTLSTLOCK)
532 mtx_unlock(&mountlist_mtx);
533 mp->mnt_lockref++;
534 MNT_IUNLOCK(mp);
535 return (0);
536 }
537
538 /*
539 * Free a busy filesystem.
540 */
541 void
542 vfs_unbusy(struct mount *mp)
543 {
544
545 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
546 MNT_ILOCK(mp);
547 MNT_REL(mp);
548 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
549 mp->mnt_lockref--;
550 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
551 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
552 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
553 mp->mnt_kern_flag &= ~MNTK_DRAINING;
554 wakeup(&mp->mnt_lockref);
555 }
556 MNT_IUNLOCK(mp);
557 }
558
559 /*
560 * Lookup a mount point by filesystem identifier.
561 */
562 struct mount *
563 vfs_getvfs(fsid_t *fsid)
564 {
565 struct mount *mp;
566
567 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
568 mtx_lock(&mountlist_mtx);
569 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
570 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
571 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
572 vfs_ref(mp);
573 mtx_unlock(&mountlist_mtx);
574 return (mp);
575 }
576 }
577 mtx_unlock(&mountlist_mtx);
578 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
579 return ((struct mount *) 0);
580 }
581
582 /*
583 * Lookup a mount point by filesystem identifier, busying it before
584 * returning.
585 *
586 * To avoid congestion on mountlist_mtx, implement simple direct-mapped
587 * cache for popular filesystem identifiers. The cache is lockess, using
588 * the fact that struct mount's are never freed. In worst case we may
589 * get pointer to unmounted or even different filesystem, so we have to
590 * check what we got, and go slow way if so.
591 */
592 struct mount *
593 vfs_busyfs(fsid_t *fsid)
594 {
595 #define FSID_CACHE_SIZE 256
596 typedef struct mount * volatile vmp_t;
597 static vmp_t cache[FSID_CACHE_SIZE];
598 struct mount *mp;
599 int error;
600 uint32_t hash;
601
602 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
603 hash = fsid->val[0] ^ fsid->val[1];
604 hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
605 mp = cache[hash];
606 if (mp == NULL ||
607 mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
608 mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
609 goto slow;
610 if (vfs_busy(mp, 0) != 0) {
611 cache[hash] = NULL;
612 goto slow;
613 }
614 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
615 mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
616 return (mp);
617 else
618 vfs_unbusy(mp);
619
620 slow:
621 mtx_lock(&mountlist_mtx);
622 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
623 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
624 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
625 error = vfs_busy(mp, MBF_MNTLSTLOCK);
626 if (error) {
627 cache[hash] = NULL;
628 mtx_unlock(&mountlist_mtx);
629 return (NULL);
630 }
631 cache[hash] = mp;
632 return (mp);
633 }
634 }
635 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
636 mtx_unlock(&mountlist_mtx);
637 return ((struct mount *) 0);
638 }
639
640 /*
641 * Check if a user can access privileged mount options.
642 */
643 int
644 vfs_suser(struct mount *mp, struct thread *td)
645 {
646 int error;
647
648 /*
649 * If the thread is jailed, but this is not a jail-friendly file
650 * system, deny immediately.
651 */
652 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
653 return (EPERM);
654
655 /*
656 * If the file system was mounted outside the jail of the calling
657 * thread, deny immediately.
658 */
659 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
660 return (EPERM);
661
662 /*
663 * If file system supports delegated administration, we don't check
664 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
665 * by the file system itself.
666 * If this is not the user that did original mount, we check for
667 * the PRIV_VFS_MOUNT_OWNER privilege.
668 */
669 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
670 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
671 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
672 return (error);
673 }
674 return (0);
675 }
676
677 /*
678 * Get a new unique fsid. Try to make its val[0] unique, since this value
679 * will be used to create fake device numbers for stat(). Also try (but
680 * not so hard) make its val[0] unique mod 2^16, since some emulators only
681 * support 16-bit device numbers. We end up with unique val[0]'s for the
682 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
683 *
684 * Keep in mind that several mounts may be running in parallel. Starting
685 * the search one past where the previous search terminated is both a
686 * micro-optimization and a defense against returning the same fsid to
687 * different mounts.
688 */
689 void
690 vfs_getnewfsid(struct mount *mp)
691 {
692 static uint16_t mntid_base;
693 struct mount *nmp;
694 fsid_t tfsid;
695 int mtype;
696
697 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
698 mtx_lock(&mntid_mtx);
699 mtype = mp->mnt_vfc->vfc_typenum;
700 tfsid.val[1] = mtype;
701 mtype = (mtype & 0xFF) << 24;
702 for (;;) {
703 tfsid.val[0] = makedev(255,
704 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
705 mntid_base++;
706 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
707 break;
708 vfs_rel(nmp);
709 }
710 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
711 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
712 mtx_unlock(&mntid_mtx);
713 }
714
715 /*
716 * Knob to control the precision of file timestamps:
717 *
718 * 0 = seconds only; nanoseconds zeroed.
719 * 1 = seconds and nanoseconds, accurate within 1/HZ.
720 * 2 = seconds and nanoseconds, truncated to microseconds.
721 * >=3 = seconds and nanoseconds, maximum precision.
722 */
723 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
724
725 static int timestamp_precision = TSP_USEC;
726 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
727 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
728 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
729 "3+: sec + ns (max. precision))");
730
731 /*
732 * Get a current timestamp.
733 */
734 void
735 vfs_timestamp(struct timespec *tsp)
736 {
737 struct timeval tv;
738
739 switch (timestamp_precision) {
740 case TSP_SEC:
741 tsp->tv_sec = time_second;
742 tsp->tv_nsec = 0;
743 break;
744 case TSP_HZ:
745 getnanotime(tsp);
746 break;
747 case TSP_USEC:
748 microtime(&tv);
749 TIMEVAL_TO_TIMESPEC(&tv, tsp);
750 break;
751 case TSP_NSEC:
752 default:
753 nanotime(tsp);
754 break;
755 }
756 }
757
758 /*
759 * Set vnode attributes to VNOVAL
760 */
761 void
762 vattr_null(struct vattr *vap)
763 {
764
765 vap->va_type = VNON;
766 vap->va_size = VNOVAL;
767 vap->va_bytes = VNOVAL;
768 vap->va_mode = VNOVAL;
769 vap->va_nlink = VNOVAL;
770 vap->va_uid = VNOVAL;
771 vap->va_gid = VNOVAL;
772 vap->va_fsid = VNOVAL;
773 vap->va_fileid = VNOVAL;
774 vap->va_blocksize = VNOVAL;
775 vap->va_rdev = VNOVAL;
776 vap->va_atime.tv_sec = VNOVAL;
777 vap->va_atime.tv_nsec = VNOVAL;
778 vap->va_mtime.tv_sec = VNOVAL;
779 vap->va_mtime.tv_nsec = VNOVAL;
780 vap->va_ctime.tv_sec = VNOVAL;
781 vap->va_ctime.tv_nsec = VNOVAL;
782 vap->va_birthtime.tv_sec = VNOVAL;
783 vap->va_birthtime.tv_nsec = VNOVAL;
784 vap->va_flags = VNOVAL;
785 vap->va_gen = VNOVAL;
786 vap->va_vaflags = 0;
787 }
788
789 /*
790 * This routine is called when we have too many vnodes. It attempts
791 * to free <count> vnodes and will potentially free vnodes that still
792 * have VM backing store (VM backing store is typically the cause
793 * of a vnode blowout so we want to do this). Therefore, this operation
794 * is not considered cheap.
795 *
796 * A number of conditions may prevent a vnode from being reclaimed.
797 * the buffer cache may have references on the vnode, a directory
798 * vnode may still have references due to the namei cache representing
799 * underlying files, or the vnode may be in active use. It is not
800 * desirable to reuse such vnodes. These conditions may cause the
801 * number of vnodes to reach some minimum value regardless of what
802 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
803 */
804 static int
805 vlrureclaim(struct mount *mp)
806 {
807 struct vnode *vp;
808 int done;
809 int trigger;
810 int usevnodes;
811 int count;
812
813 /*
814 * Calculate the trigger point, don't allow user
815 * screwups to blow us up. This prevents us from
816 * recycling vnodes with lots of resident pages. We
817 * aren't trying to free memory, we are trying to
818 * free vnodes.
819 */
820 usevnodes = desiredvnodes;
821 if (usevnodes <= 0)
822 usevnodes = 1;
823 trigger = cnt.v_page_count * 2 / usevnodes;
824 done = 0;
825 vn_start_write(NULL, &mp, V_WAIT);
826 MNT_ILOCK(mp);
827 count = mp->mnt_nvnodelistsize / 10 + 1;
828 while (count != 0) {
829 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
830 while (vp != NULL && vp->v_type == VMARKER)
831 vp = TAILQ_NEXT(vp, v_nmntvnodes);
832 if (vp == NULL)
833 break;
834 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
835 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
836 --count;
837 if (!VI_TRYLOCK(vp))
838 goto next_iter;
839 /*
840 * If it's been deconstructed already, it's still
841 * referenced, or it exceeds the trigger, skip it.
842 */
843 if (vp->v_usecount ||
844 (!vlru_allow_cache_src &&
845 !LIST_EMPTY(&(vp)->v_cache_src)) ||
846 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
847 vp->v_object->resident_page_count > trigger)) {
848 VI_UNLOCK(vp);
849 goto next_iter;
850 }
851 MNT_IUNLOCK(mp);
852 vholdl(vp);
853 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
854 vdrop(vp);
855 goto next_iter_mntunlocked;
856 }
857 VI_LOCK(vp);
858 /*
859 * v_usecount may have been bumped after VOP_LOCK() dropped
860 * the vnode interlock and before it was locked again.
861 *
862 * It is not necessary to recheck VI_DOOMED because it can
863 * only be set by another thread that holds both the vnode
864 * lock and vnode interlock. If another thread has the
865 * vnode lock before we get to VOP_LOCK() and obtains the
866 * vnode interlock after VOP_LOCK() drops the vnode
867 * interlock, the other thread will be unable to drop the
868 * vnode lock before our VOP_LOCK() call fails.
869 */
870 if (vp->v_usecount ||
871 (!vlru_allow_cache_src &&
872 !LIST_EMPTY(&(vp)->v_cache_src)) ||
873 (vp->v_object != NULL &&
874 vp->v_object->resident_page_count > trigger)) {
875 VOP_UNLOCK(vp, LK_INTERLOCK);
876 vdrop(vp);
877 goto next_iter_mntunlocked;
878 }
879 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
880 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
881 atomic_add_long(&recycles_count, 1);
882 vgonel(vp);
883 VOP_UNLOCK(vp, 0);
884 vdropl(vp);
885 done++;
886 next_iter_mntunlocked:
887 if (!should_yield())
888 goto relock_mnt;
889 goto yield;
890 next_iter:
891 if (!should_yield())
892 continue;
893 MNT_IUNLOCK(mp);
894 yield:
895 kern_yield(PRI_USER);
896 relock_mnt:
897 MNT_ILOCK(mp);
898 }
899 MNT_IUNLOCK(mp);
900 vn_finished_write(mp);
901 return done;
902 }
903
904 /*
905 * Attempt to keep the free list at wantfreevnodes length.
906 */
907 static void
908 vnlru_free(int count)
909 {
910 struct vnode *vp;
911
912 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
913 for (; count > 0; count--) {
914 vp = TAILQ_FIRST(&vnode_free_list);
915 /*
916 * The list can be modified while the free_list_mtx
917 * has been dropped and vp could be NULL here.
918 */
919 if (!vp)
920 break;
921 VNASSERT(vp->v_op != NULL, vp,
922 ("vnlru_free: vnode already reclaimed."));
923 KASSERT((vp->v_iflag & VI_FREE) != 0,
924 ("Removing vnode not on freelist"));
925 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
926 ("Mangling active vnode"));
927 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
928 /*
929 * Don't recycle if we can't get the interlock.
930 */
931 if (!VI_TRYLOCK(vp)) {
932 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
933 continue;
934 }
935 VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
936 vp, ("vp inconsistent on freelist"));
937
938 /*
939 * The clear of VI_FREE prevents activation of the
940 * vnode. There is no sense in putting the vnode on
941 * the mount point active list, only to remove it
942 * later during recycling. Inline the relevant part
943 * of vholdl(), to avoid triggering assertions or
944 * activating.
945 */
946 freevnodes--;
947 vp->v_iflag &= ~VI_FREE;
948 vp->v_holdcnt++;
949
950 mtx_unlock(&vnode_free_list_mtx);
951 VI_UNLOCK(vp);
952 vtryrecycle(vp);
953 /*
954 * If the recycled succeeded this vdrop will actually free
955 * the vnode. If not it will simply place it back on
956 * the free list.
957 */
958 vdrop(vp);
959 mtx_lock(&vnode_free_list_mtx);
960 }
961 }
962 /*
963 * Attempt to recycle vnodes in a context that is always safe to block.
964 * Calling vlrurecycle() from the bowels of filesystem code has some
965 * interesting deadlock problems.
966 */
967 static struct proc *vnlruproc;
968 static int vnlruproc_sig;
969
970 static void
971 vnlru_proc(void)
972 {
973 struct mount *mp, *nmp;
974 int done;
975 struct proc *p = vnlruproc;
976
977 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
978 SHUTDOWN_PRI_FIRST);
979
980 for (;;) {
981 kproc_suspend_check(p);
982 mtx_lock(&vnode_free_list_mtx);
983 if (freevnodes > wantfreevnodes)
984 vnlru_free(freevnodes - wantfreevnodes);
985 if (numvnodes <= desiredvnodes * 9 / 10) {
986 vnlruproc_sig = 0;
987 wakeup(&vnlruproc_sig);
988 msleep(vnlruproc, &vnode_free_list_mtx,
989 PVFS|PDROP, "vlruwt", hz);
990 continue;
991 }
992 mtx_unlock(&vnode_free_list_mtx);
993 done = 0;
994 mtx_lock(&mountlist_mtx);
995 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
996 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
997 nmp = TAILQ_NEXT(mp, mnt_list);
998 continue;
999 }
1000 done += vlrureclaim(mp);
1001 mtx_lock(&mountlist_mtx);
1002 nmp = TAILQ_NEXT(mp, mnt_list);
1003 vfs_unbusy(mp);
1004 }
1005 mtx_unlock(&mountlist_mtx);
1006 if (done == 0) {
1007 #if 0
1008 /* These messages are temporary debugging aids */
1009 if (vnlru_nowhere < 5)
1010 printf("vnlru process getting nowhere..\n");
1011 else if (vnlru_nowhere == 5)
1012 printf("vnlru process messages stopped.\n");
1013 #endif
1014 vnlru_nowhere++;
1015 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
1016 } else
1017 kern_yield(PRI_USER);
1018 }
1019 }
1020
1021 static struct kproc_desc vnlru_kp = {
1022 "vnlru",
1023 vnlru_proc,
1024 &vnlruproc
1025 };
1026 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
1027 &vnlru_kp);
1028
1029 /*
1030 * Routines having to do with the management of the vnode table.
1031 */
1032
1033 /*
1034 * Try to recycle a freed vnode. We abort if anyone picks up a reference
1035 * before we actually vgone(). This function must be called with the vnode
1036 * held to prevent the vnode from being returned to the free list midway
1037 * through vgone().
1038 */
1039 static int
1040 vtryrecycle(struct vnode *vp)
1041 {
1042 struct mount *vnmp;
1043
1044 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
1045 VNASSERT(vp->v_holdcnt, vp,
1046 ("vtryrecycle: Recycling vp %p without a reference.", vp));
1047 /*
1048 * This vnode may found and locked via some other list, if so we
1049 * can't recycle it yet.
1050 */
1051 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1052 CTR2(KTR_VFS,
1053 "%s: impossible to recycle, vp %p lock is already held",
1054 __func__, vp);
1055 return (EWOULDBLOCK);
1056 }
1057 /*
1058 * Don't recycle if its filesystem is being suspended.
1059 */
1060 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
1061 VOP_UNLOCK(vp, 0);
1062 CTR2(KTR_VFS,
1063 "%s: impossible to recycle, cannot start the write for %p",
1064 __func__, vp);
1065 return (EBUSY);
1066 }
1067 /*
1068 * If we got this far, we need to acquire the interlock and see if
1069 * anyone picked up this vnode from another list. If not, we will
1070 * mark it with DOOMED via vgonel() so that anyone who does find it
1071 * will skip over it.
1072 */
1073 VI_LOCK(vp);
1074 if (vp->v_usecount) {
1075 VOP_UNLOCK(vp, LK_INTERLOCK);
1076 vn_finished_write(vnmp);
1077 CTR2(KTR_VFS,
1078 "%s: impossible to recycle, %p is already referenced",
1079 __func__, vp);
1080 return (EBUSY);
1081 }
1082 if ((vp->v_iflag & VI_DOOMED) == 0) {
1083 atomic_add_long(&recycles_count, 1);
1084 vgonel(vp);
1085 }
1086 VOP_UNLOCK(vp, LK_INTERLOCK);
1087 vn_finished_write(vnmp);
1088 return (0);
1089 }
1090
1091 /*
1092 * Wait for available vnodes.
1093 */
1094 static int
1095 getnewvnode_wait(int suspended)
1096 {
1097
1098 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
1099 if (numvnodes > desiredvnodes) {
1100 if (suspended) {
1101 /*
1102 * File system is beeing suspended, we cannot risk a
1103 * deadlock here, so allocate new vnode anyway.
1104 */
1105 if (freevnodes > wantfreevnodes)
1106 vnlru_free(freevnodes - wantfreevnodes);
1107 return (0);
1108 }
1109 if (vnlruproc_sig == 0) {
1110 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
1111 wakeup(vnlruproc);
1112 }
1113 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1114 "vlruwk", hz);
1115 }
1116 return (numvnodes > desiredvnodes ? ENFILE : 0);
1117 }
1118
1119 void
1120 getnewvnode_reserve(u_int count)
1121 {
1122 struct thread *td;
1123
1124 td = curthread;
1125 /* First try to be quick and racy. */
1126 if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1127 td->td_vp_reserv += count;
1128 return;
1129 } else
1130 atomic_subtract_long(&numvnodes, count);
1131
1132 mtx_lock(&vnode_free_list_mtx);
1133 while (count > 0) {
1134 if (getnewvnode_wait(0) == 0) {
1135 count--;
1136 td->td_vp_reserv++;
1137 atomic_add_long(&numvnodes, 1);
1138 }
1139 }
1140 mtx_unlock(&vnode_free_list_mtx);
1141 }
1142
1143 void
1144 getnewvnode_drop_reserve(void)
1145 {
1146 struct thread *td;
1147
1148 td = curthread;
1149 atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1150 td->td_vp_reserv = 0;
1151 }
1152
1153 /*
1154 * Return the next vnode from the free list.
1155 */
1156 int
1157 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1158 struct vnode **vpp)
1159 {
1160 struct vnode *vp;
1161 struct thread *td;
1162 struct lock_object *lo;
1163 int error;
1164
1165 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1166 vp = NULL;
1167 td = curthread;
1168 if (td->td_vp_reserv > 0) {
1169 td->td_vp_reserv -= 1;
1170 goto alloc;
1171 }
1172 mtx_lock(&vnode_free_list_mtx);
1173 /*
1174 * Lend our context to reclaim vnodes if they've exceeded the max.
1175 */
1176 if (freevnodes > wantfreevnodes)
1177 vnlru_free(1);
1178 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1179 MNTK_SUSPEND));
1180 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1181 if (error != 0) {
1182 mtx_unlock(&vnode_free_list_mtx);
1183 return (error);
1184 }
1185 #endif
1186 atomic_add_long(&numvnodes, 1);
1187 mtx_unlock(&vnode_free_list_mtx);
1188 alloc:
1189 atomic_add_long(&vnodes_created, 1);
1190 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK);
1191 /*
1192 * Locks are given the generic name "vnode" when created.
1193 * Follow the historic practice of using the filesystem
1194 * name when they allocated, e.g., "zfs", "ufs", "nfs, etc.
1195 *
1196 * Locks live in a witness group keyed on their name. Thus,
1197 * when a lock is renamed, it must also move from the witness
1198 * group of its old name to the witness group of its new name.
1199 *
1200 * The change only needs to be made when the vnode moves
1201 * from one filesystem type to another. We ensure that each
1202 * filesystem use a single static name pointer for its tag so
1203 * that we can compare pointers rather than doing a strcmp().
1204 */
1205 lo = &vp->v_vnlock->lock_object;
1206 if (lo->lo_name != tag) {
1207 lo->lo_name = tag;
1208 WITNESS_DESTROY(lo);
1209 WITNESS_INIT(lo, tag);
1210 }
1211 /*
1212 * By default, don't allow shared locks unless filesystems opt-in.
1213 */
1214 vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE;
1215 /*
1216 * Finalize various vnode identity bits.
1217 */
1218 KASSERT(vp->v_object == NULL, ("stale v_object %p", vp));
1219 KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp));
1220 KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp));
1221 vp->v_type = VNON;
1222 vp->v_tag = tag;
1223 vp->v_op = vops;
1224 v_incr_usecount(vp);
1225 vp->v_bufobj.bo_ops = &buf_ops_bio;
1226 #ifdef MAC
1227 mac_vnode_init(vp);
1228 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1229 mac_vnode_associate_singlelabel(mp, vp);
1230 else if (mp == NULL && vops != &dead_vnodeops)
1231 printf("NULL mp in getnewvnode()\n");
1232 #endif
1233 if (mp != NULL) {
1234 vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize;
1235 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1236 vp->v_vflag |= VV_NOKNOTE;
1237 }
1238
1239 /*
1240 * For the filesystems which do not use vfs_hash_insert(),
1241 * still initialize v_hash to have vfs_hash_index() useful.
1242 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1243 * its own hashing.
1244 */
1245 vp->v_hash = (uintptr_t)vp >> vnsz2log;
1246
1247 *vpp = vp;
1248 return (0);
1249 }
1250
1251 /*
1252 * Delete from old mount point vnode list, if on one.
1253 */
1254 static void
1255 delmntque(struct vnode *vp)
1256 {
1257 struct mount *mp;
1258 int active;
1259
1260 mp = vp->v_mount;
1261 if (mp == NULL)
1262 return;
1263 MNT_ILOCK(mp);
1264 VI_LOCK(vp);
1265 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1266 ("Active vnode list size %d > Vnode list size %d",
1267 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1268 active = vp->v_iflag & VI_ACTIVE;
1269 vp->v_iflag &= ~VI_ACTIVE;
1270 if (active) {
1271 mtx_lock(&vnode_free_list_mtx);
1272 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1273 mp->mnt_activevnodelistsize--;
1274 mtx_unlock(&vnode_free_list_mtx);
1275 }
1276 vp->v_mount = NULL;
1277 VI_UNLOCK(vp);
1278 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1279 ("bad mount point vnode list size"));
1280 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1281 mp->mnt_nvnodelistsize--;
1282 MNT_REL(mp);
1283 MNT_IUNLOCK(mp);
1284 }
1285
1286 static void
1287 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1288 {
1289
1290 vp->v_data = NULL;
1291 vp->v_op = &dead_vnodeops;
1292 vgone(vp);
1293 vput(vp);
1294 }
1295
1296 /*
1297 * Insert into list of vnodes for the new mount point, if available.
1298 */
1299 int
1300 insmntque1(struct vnode *vp, struct mount *mp,
1301 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1302 {
1303
1304 KASSERT(vp->v_mount == NULL,
1305 ("insmntque: vnode already on per mount vnode list"));
1306 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1307 ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1308
1309 /*
1310 * We acquire the vnode interlock early to ensure that the
1311 * vnode cannot be recycled by another process releasing a
1312 * holdcnt on it before we get it on both the vnode list
1313 * and the active vnode list. The mount mutex protects only
1314 * manipulation of the vnode list and the vnode freelist
1315 * mutex protects only manipulation of the active vnode list.
1316 * Hence the need to hold the vnode interlock throughout.
1317 */
1318 MNT_ILOCK(mp);
1319 VI_LOCK(vp);
1320 if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1321 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1322 mp->mnt_nvnodelistsize == 0)) &&
1323 (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1324 VI_UNLOCK(vp);
1325 MNT_IUNLOCK(mp);
1326 if (dtr != NULL)
1327 dtr(vp, dtr_arg);
1328 return (EBUSY);
1329 }
1330 vp->v_mount = mp;
1331 MNT_REF(mp);
1332 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1333 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1334 ("neg mount point vnode list size"));
1335 mp->mnt_nvnodelistsize++;
1336 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1337 ("Activating already active vnode"));
1338 vp->v_iflag |= VI_ACTIVE;
1339 mtx_lock(&vnode_free_list_mtx);
1340 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1341 mp->mnt_activevnodelistsize++;
1342 mtx_unlock(&vnode_free_list_mtx);
1343 VI_UNLOCK(vp);
1344 MNT_IUNLOCK(mp);
1345 return (0);
1346 }
1347
1348 int
1349 insmntque(struct vnode *vp, struct mount *mp)
1350 {
1351
1352 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1353 }
1354
1355 /*
1356 * Flush out and invalidate all buffers associated with a bufobj
1357 * Called with the underlying object locked.
1358 */
1359 int
1360 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1361 {
1362 int error;
1363
1364 BO_LOCK(bo);
1365 if (flags & V_SAVE) {
1366 error = bufobj_wwait(bo, slpflag, slptimeo);
1367 if (error) {
1368 BO_UNLOCK(bo);
1369 return (error);
1370 }
1371 if (bo->bo_dirty.bv_cnt > 0) {
1372 BO_UNLOCK(bo);
1373 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1374 return (error);
1375 /*
1376 * XXX We could save a lock/unlock if this was only
1377 * enabled under INVARIANTS
1378 */
1379 BO_LOCK(bo);
1380 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1381 panic("vinvalbuf: dirty bufs");
1382 }
1383 }
1384 /*
1385 * If you alter this loop please notice that interlock is dropped and
1386 * reacquired in flushbuflist. Special care is needed to ensure that
1387 * no race conditions occur from this.
1388 */
1389 do {
1390 error = flushbuflist(&bo->bo_clean,
1391 flags, bo, slpflag, slptimeo);
1392 if (error == 0 && !(flags & V_CLEANONLY))
1393 error = flushbuflist(&bo->bo_dirty,
1394 flags, bo, slpflag, slptimeo);
1395 if (error != 0 && error != EAGAIN) {
1396 BO_UNLOCK(bo);
1397 return (error);
1398 }
1399 } while (error != 0);
1400
1401 /*
1402 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1403 * have write I/O in-progress but if there is a VM object then the
1404 * VM object can also have read-I/O in-progress.
1405 */
1406 do {
1407 bufobj_wwait(bo, 0, 0);
1408 BO_UNLOCK(bo);
1409 if (bo->bo_object != NULL) {
1410 VM_OBJECT_WLOCK(bo->bo_object);
1411 vm_object_pip_wait(bo->bo_object, "bovlbx");
1412 VM_OBJECT_WUNLOCK(bo->bo_object);
1413 }
1414 BO_LOCK(bo);
1415 } while (bo->bo_numoutput > 0);
1416 BO_UNLOCK(bo);
1417
1418 /*
1419 * Destroy the copy in the VM cache, too.
1420 */
1421 if (bo->bo_object != NULL &&
1422 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1423 VM_OBJECT_WLOCK(bo->bo_object);
1424 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1425 OBJPR_CLEANONLY : 0);
1426 VM_OBJECT_WUNLOCK(bo->bo_object);
1427 }
1428
1429 #ifdef INVARIANTS
1430 BO_LOCK(bo);
1431 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1432 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1433 panic("vinvalbuf: flush failed");
1434 BO_UNLOCK(bo);
1435 #endif
1436 return (0);
1437 }
1438
1439 /*
1440 * Flush out and invalidate all buffers associated with a vnode.
1441 * Called with the underlying object locked.
1442 */
1443 int
1444 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1445 {
1446
1447 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1448 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1449 if (vp->v_object != NULL && vp->v_object->handle != vp)
1450 return (0);
1451 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1452 }
1453
1454 /*
1455 * Flush out buffers on the specified list.
1456 *
1457 */
1458 static int
1459 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1460 int slptimeo)
1461 {
1462 struct buf *bp, *nbp;
1463 int retval, error;
1464 daddr_t lblkno;
1465 b_xflags_t xflags;
1466
1467 ASSERT_BO_WLOCKED(bo);
1468
1469 retval = 0;
1470 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1471 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1472 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1473 continue;
1474 }
1475 lblkno = 0;
1476 xflags = 0;
1477 if (nbp != NULL) {
1478 lblkno = nbp->b_lblkno;
1479 xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1480 }
1481 retval = EAGAIN;
1482 error = BUF_TIMELOCK(bp,
1483 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1484 "flushbuf", slpflag, slptimeo);
1485 if (error) {
1486 BO_LOCK(bo);
1487 return (error != ENOLCK ? error : EAGAIN);
1488 }
1489 KASSERT(bp->b_bufobj == bo,
1490 ("bp %p wrong b_bufobj %p should be %p",
1491 bp, bp->b_bufobj, bo));
1492 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1493 BUF_UNLOCK(bp);
1494 BO_LOCK(bo);
1495 return (EAGAIN);
1496 }
1497 /*
1498 * XXX Since there are no node locks for NFS, I
1499 * believe there is a slight chance that a delayed
1500 * write will occur while sleeping just above, so
1501 * check for it.
1502 */
1503 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1504 (flags & V_SAVE)) {
1505 bremfree(bp);
1506 bp->b_flags |= B_ASYNC;
1507 bwrite(bp);
1508 BO_LOCK(bo);
1509 return (EAGAIN); /* XXX: why not loop ? */
1510 }
1511 bremfree(bp);
1512 bp->b_flags |= (B_INVAL | B_RELBUF);
1513 bp->b_flags &= ~B_ASYNC;
1514 brelse(bp);
1515 BO_LOCK(bo);
1516 if (nbp != NULL &&
1517 (nbp->b_bufobj != bo ||
1518 nbp->b_lblkno != lblkno ||
1519 (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1520 break; /* nbp invalid */
1521 }
1522 return (retval);
1523 }
1524
1525 /*
1526 * Truncate a file's buffer and pages to a specified length. This
1527 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1528 * sync activity.
1529 */
1530 int
1531 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1532 {
1533 struct buf *bp, *nbp;
1534 int anyfreed;
1535 int trunclbn;
1536 struct bufobj *bo;
1537
1538 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1539 vp, cred, blksize, (uintmax_t)length);
1540
1541 /*
1542 * Round up to the *next* lbn.
1543 */
1544 trunclbn = (length + blksize - 1) / blksize;
1545
1546 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1547 restart:
1548 bo = &vp->v_bufobj;
1549 BO_LOCK(bo);
1550 anyfreed = 1;
1551 for (;anyfreed;) {
1552 anyfreed = 0;
1553 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1554 if (bp->b_lblkno < trunclbn)
1555 continue;
1556 if (BUF_LOCK(bp,
1557 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1558 BO_LOCKPTR(bo)) == ENOLCK)
1559 goto restart;
1560
1561 bremfree(bp);
1562 bp->b_flags |= (B_INVAL | B_RELBUF);
1563 bp->b_flags &= ~B_ASYNC;
1564 brelse(bp);
1565 anyfreed = 1;
1566
1567 BO_LOCK(bo);
1568 if (nbp != NULL &&
1569 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1570 (nbp->b_vp != vp) ||
1571 (nbp->b_flags & B_DELWRI))) {
1572 BO_UNLOCK(bo);
1573 goto restart;
1574 }
1575 }
1576
1577 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1578 if (bp->b_lblkno < trunclbn)
1579 continue;
1580 if (BUF_LOCK(bp,
1581 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1582 BO_LOCKPTR(bo)) == ENOLCK)
1583 goto restart;
1584 bremfree(bp);
1585 bp->b_flags |= (B_INVAL | B_RELBUF);
1586 bp->b_flags &= ~B_ASYNC;
1587 brelse(bp);
1588 anyfreed = 1;
1589
1590 BO_LOCK(bo);
1591 if (nbp != NULL &&
1592 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1593 (nbp->b_vp != vp) ||
1594 (nbp->b_flags & B_DELWRI) == 0)) {
1595 BO_UNLOCK(bo);
1596 goto restart;
1597 }
1598 }
1599 }
1600
1601 if (length > 0) {
1602 restartsync:
1603 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1604 if (bp->b_lblkno > 0)
1605 continue;
1606 /*
1607 * Since we hold the vnode lock this should only
1608 * fail if we're racing with the buf daemon.
1609 */
1610 if (BUF_LOCK(bp,
1611 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1612 BO_LOCKPTR(bo)) == ENOLCK) {
1613 goto restart;
1614 }
1615 VNASSERT((bp->b_flags & B_DELWRI), vp,
1616 ("buf(%p) on dirty queue without DELWRI", bp));
1617
1618 bremfree(bp);
1619 bawrite(bp);
1620 BO_LOCK(bo);
1621 goto restartsync;
1622 }
1623 }
1624
1625 bufobj_wwait(bo, 0, 0);
1626 BO_UNLOCK(bo);
1627 vnode_pager_setsize(vp, length);
1628
1629 return (0);
1630 }
1631
1632 static void
1633 buf_vlist_remove(struct buf *bp)
1634 {
1635 struct bufv *bv;
1636
1637 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1638 ASSERT_BO_WLOCKED(bp->b_bufobj);
1639 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1640 (BX_VNDIRTY|BX_VNCLEAN),
1641 ("buf_vlist_remove: Buf %p is on two lists", bp));
1642 if (bp->b_xflags & BX_VNDIRTY)
1643 bv = &bp->b_bufobj->bo_dirty;
1644 else
1645 bv = &bp->b_bufobj->bo_clean;
1646 BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1647 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1648 bv->bv_cnt--;
1649 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1650 }
1651
1652 /*
1653 * Add the buffer to the sorted clean or dirty block list.
1654 *
1655 * NOTE: xflags is passed as a constant, optimizing this inline function!
1656 */
1657 static void
1658 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1659 {
1660 struct bufv *bv;
1661 struct buf *n;
1662 int error;
1663
1664 ASSERT_BO_WLOCKED(bo);
1665 KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0,
1666 ("dead bo %p", bo));
1667 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1668 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1669 bp->b_xflags |= xflags;
1670 if (xflags & BX_VNDIRTY)
1671 bv = &bo->bo_dirty;
1672 else
1673 bv = &bo->bo_clean;
1674
1675 /*
1676 * Keep the list ordered. Optimize empty list insertion. Assume
1677 * we tend to grow at the tail so lookup_le should usually be cheaper
1678 * than _ge.
1679 */
1680 if (bv->bv_cnt == 0 ||
1681 bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1682 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1683 else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1684 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1685 else
1686 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1687 error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1688 if (error)
1689 panic("buf_vlist_add: Preallocated nodes insufficient.");
1690 bv->bv_cnt++;
1691 }
1692
1693 /*
1694 * Lookup a buffer using the splay tree. Note that we specifically avoid
1695 * shadow buffers used in background bitmap writes.
1696 *
1697 * This code isn't quite efficient as it could be because we are maintaining
1698 * two sorted lists and do not know which list the block resides in.
1699 *
1700 * During a "make buildworld" the desired buffer is found at one of
1701 * the roots more than 60% of the time. Thus, checking both roots
1702 * before performing either splay eliminates unnecessary splays on the
1703 * first tree splayed.
1704 */
1705 struct buf *
1706 gbincore(struct bufobj *bo, daddr_t lblkno)
1707 {
1708 struct buf *bp;
1709
1710 ASSERT_BO_LOCKED(bo);
1711 bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1712 if (bp != NULL)
1713 return (bp);
1714 return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1715 }
1716
1717 /*
1718 * Associate a buffer with a vnode.
1719 */
1720 void
1721 bgetvp(struct vnode *vp, struct buf *bp)
1722 {
1723 struct bufobj *bo;
1724
1725 bo = &vp->v_bufobj;
1726 ASSERT_BO_WLOCKED(bo);
1727 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1728
1729 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1730 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1731 ("bgetvp: bp already attached! %p", bp));
1732
1733 vhold(vp);
1734 bp->b_vp = vp;
1735 bp->b_bufobj = bo;
1736 /*
1737 * Insert onto list for new vnode.
1738 */
1739 buf_vlist_add(bp, bo, BX_VNCLEAN);
1740 }
1741
1742 /*
1743 * Disassociate a buffer from a vnode.
1744 */
1745 void
1746 brelvp(struct buf *bp)
1747 {
1748 struct bufobj *bo;
1749 struct vnode *vp;
1750
1751 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1752 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1753
1754 /*
1755 * Delete from old vnode list, if on one.
1756 */
1757 vp = bp->b_vp; /* XXX */
1758 bo = bp->b_bufobj;
1759 BO_LOCK(bo);
1760 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1761 buf_vlist_remove(bp);
1762 else
1763 panic("brelvp: Buffer %p not on queue.", bp);
1764 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1765 bo->bo_flag &= ~BO_ONWORKLST;
1766 mtx_lock(&sync_mtx);
1767 LIST_REMOVE(bo, bo_synclist);
1768 syncer_worklist_len--;
1769 mtx_unlock(&sync_mtx);
1770 }
1771 bp->b_vp = NULL;
1772 bp->b_bufobj = NULL;
1773 BO_UNLOCK(bo);
1774 vdrop(vp);
1775 }
1776
1777 /*
1778 * Add an item to the syncer work queue.
1779 */
1780 static void
1781 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1782 {
1783 int slot;
1784
1785 ASSERT_BO_WLOCKED(bo);
1786
1787 mtx_lock(&sync_mtx);
1788 if (bo->bo_flag & BO_ONWORKLST)
1789 LIST_REMOVE(bo, bo_synclist);
1790 else {
1791 bo->bo_flag |= BO_ONWORKLST;
1792 syncer_worklist_len++;
1793 }
1794
1795 if (delay > syncer_maxdelay - 2)
1796 delay = syncer_maxdelay - 2;
1797 slot = (syncer_delayno + delay) & syncer_mask;
1798
1799 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1800 mtx_unlock(&sync_mtx);
1801 }
1802
1803 static int
1804 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1805 {
1806 int error, len;
1807
1808 mtx_lock(&sync_mtx);
1809 len = syncer_worklist_len - sync_vnode_count;
1810 mtx_unlock(&sync_mtx);
1811 error = SYSCTL_OUT(req, &len, sizeof(len));
1812 return (error);
1813 }
1814
1815 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1816 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1817
1818 static struct proc *updateproc;
1819 static void sched_sync(void);
1820 static struct kproc_desc up_kp = {
1821 "syncer",
1822 sched_sync,
1823 &updateproc
1824 };
1825 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1826
1827 static int
1828 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1829 {
1830 struct vnode *vp;
1831 struct mount *mp;
1832
1833 *bo = LIST_FIRST(slp);
1834 if (*bo == NULL)
1835 return (0);
1836 vp = (*bo)->__bo_vnode; /* XXX */
1837 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1838 return (1);
1839 /*
1840 * We use vhold in case the vnode does not
1841 * successfully sync. vhold prevents the vnode from
1842 * going away when we unlock the sync_mtx so that
1843 * we can acquire the vnode interlock.
1844 */
1845 vholdl(vp);
1846 mtx_unlock(&sync_mtx);
1847 VI_UNLOCK(vp);
1848 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1849 vdrop(vp);
1850 mtx_lock(&sync_mtx);
1851 return (*bo == LIST_FIRST(slp));
1852 }
1853 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1854 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1855 VOP_UNLOCK(vp, 0);
1856 vn_finished_write(mp);
1857 BO_LOCK(*bo);
1858 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1859 /*
1860 * Put us back on the worklist. The worklist
1861 * routine will remove us from our current
1862 * position and then add us back in at a later
1863 * position.
1864 */
1865 vn_syncer_add_to_worklist(*bo, syncdelay);
1866 }
1867 BO_UNLOCK(*bo);
1868 vdrop(vp);
1869 mtx_lock(&sync_mtx);
1870 return (0);
1871 }
1872
1873 static int first_printf = 1;
1874
1875 /*
1876 * System filesystem synchronizer daemon.
1877 */
1878 static void
1879 sched_sync(void)
1880 {
1881 struct synclist *next, *slp;
1882 struct bufobj *bo;
1883 long starttime;
1884 struct thread *td = curthread;
1885 int last_work_seen;
1886 int net_worklist_len;
1887 int syncer_final_iter;
1888 int error;
1889
1890 last_work_seen = 0;
1891 syncer_final_iter = 0;
1892 syncer_state = SYNCER_RUNNING;
1893 starttime = time_uptime;
1894 td->td_pflags |= TDP_NORUNNINGBUF;
1895
1896 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1897 SHUTDOWN_PRI_LAST);
1898
1899 mtx_lock(&sync_mtx);
1900 for (;;) {
1901 if (syncer_state == SYNCER_FINAL_DELAY &&
1902 syncer_final_iter == 0) {
1903 mtx_unlock(&sync_mtx);
1904 kproc_suspend_check(td->td_proc);
1905 mtx_lock(&sync_mtx);
1906 }
1907 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1908 if (syncer_state != SYNCER_RUNNING &&
1909 starttime != time_uptime) {
1910 if (first_printf) {
1911 printf("\nSyncing disks, vnodes remaining...");
1912 first_printf = 0;
1913 }
1914 printf("%d ", net_worklist_len);
1915 }
1916 starttime = time_uptime;
1917
1918 /*
1919 * Push files whose dirty time has expired. Be careful
1920 * of interrupt race on slp queue.
1921 *
1922 * Skip over empty worklist slots when shutting down.
1923 */
1924 do {
1925 slp = &syncer_workitem_pending[syncer_delayno];
1926 syncer_delayno += 1;
1927 if (syncer_delayno == syncer_maxdelay)
1928 syncer_delayno = 0;
1929 next = &syncer_workitem_pending[syncer_delayno];
1930 /*
1931 * If the worklist has wrapped since the
1932 * it was emptied of all but syncer vnodes,
1933 * switch to the FINAL_DELAY state and run
1934 * for one more second.
1935 */
1936 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1937 net_worklist_len == 0 &&
1938 last_work_seen == syncer_delayno) {
1939 syncer_state = SYNCER_FINAL_DELAY;
1940 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1941 }
1942 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1943 syncer_worklist_len > 0);
1944
1945 /*
1946 * Keep track of the last time there was anything
1947 * on the worklist other than syncer vnodes.
1948 * Return to the SHUTTING_DOWN state if any
1949 * new work appears.
1950 */
1951 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1952 last_work_seen = syncer_delayno;
1953 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1954 syncer_state = SYNCER_SHUTTING_DOWN;
1955 while (!LIST_EMPTY(slp)) {
1956 error = sync_vnode(slp, &bo, td);
1957 if (error == 1) {
1958 LIST_REMOVE(bo, bo_synclist);
1959 LIST_INSERT_HEAD(next, bo, bo_synclist);
1960 continue;
1961 }
1962
1963 if (first_printf == 0)
1964 wdog_kern_pat(WD_LASTVAL);
1965
1966 }
1967 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1968 syncer_final_iter--;
1969 /*
1970 * The variable rushjob allows the kernel to speed up the
1971 * processing of the filesystem syncer process. A rushjob
1972 * value of N tells the filesystem syncer to process the next
1973 * N seconds worth of work on its queue ASAP. Currently rushjob
1974 * is used by the soft update code to speed up the filesystem
1975 * syncer process when the incore state is getting so far
1976 * ahead of the disk that the kernel memory pool is being
1977 * threatened with exhaustion.
1978 */
1979 if (rushjob > 0) {
1980 rushjob -= 1;
1981 continue;
1982 }
1983 /*
1984 * Just sleep for a short period of time between
1985 * iterations when shutting down to allow some I/O
1986 * to happen.
1987 *
1988 * If it has taken us less than a second to process the
1989 * current work, then wait. Otherwise start right over
1990 * again. We can still lose time if any single round
1991 * takes more than two seconds, but it does not really
1992 * matter as we are just trying to generally pace the
1993 * filesystem activity.
1994 */
1995 if (syncer_state != SYNCER_RUNNING ||
1996 time_uptime == starttime) {
1997 thread_lock(td);
1998 sched_prio(td, PPAUSE);
1999 thread_unlock(td);
2000 }
2001 if (syncer_state != SYNCER_RUNNING)
2002 cv_timedwait(&sync_wakeup, &sync_mtx,
2003 hz / SYNCER_SHUTDOWN_SPEEDUP);
2004 else if (time_uptime == starttime)
2005 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
2006 }
2007 }
2008
2009 /*
2010 * Request the syncer daemon to speed up its work.
2011 * We never push it to speed up more than half of its
2012 * normal turn time, otherwise it could take over the cpu.
2013 */
2014 int
2015 speedup_syncer(void)
2016 {
2017 int ret = 0;
2018
2019 mtx_lock(&sync_mtx);
2020 if (rushjob < syncdelay / 2) {
2021 rushjob += 1;
2022 stat_rush_requests += 1;
2023 ret = 1;
2024 }
2025 mtx_unlock(&sync_mtx);
2026 cv_broadcast(&sync_wakeup);
2027 return (ret);
2028 }
2029
2030 /*
2031 * Tell the syncer to speed up its work and run though its work
2032 * list several times, then tell it to shut down.
2033 */
2034 static void
2035 syncer_shutdown(void *arg, int howto)
2036 {
2037
2038 if (howto & RB_NOSYNC)
2039 return;
2040 mtx_lock(&sync_mtx);
2041 syncer_state = SYNCER_SHUTTING_DOWN;
2042 rushjob = 0;
2043 mtx_unlock(&sync_mtx);
2044 cv_broadcast(&sync_wakeup);
2045 kproc_shutdown(arg, howto);
2046 }
2047
2048 void
2049 syncer_suspend(void)
2050 {
2051
2052 syncer_shutdown(updateproc, 0);
2053 }
2054
2055 void
2056 syncer_resume(void)
2057 {
2058
2059 mtx_lock(&sync_mtx);
2060 first_printf = 1;
2061 syncer_state = SYNCER_RUNNING;
2062 mtx_unlock(&sync_mtx);
2063 cv_broadcast(&sync_wakeup);
2064 kproc_resume(updateproc);
2065 }
2066
2067 /*
2068 * Reassign a buffer from one vnode to another.
2069 * Used to assign file specific control information
2070 * (indirect blocks) to the vnode to which they belong.
2071 */
2072 void
2073 reassignbuf(struct buf *bp)
2074 {
2075 struct vnode *vp;
2076 struct bufobj *bo;
2077 int delay;
2078 #ifdef INVARIANTS
2079 struct bufv *bv;
2080 #endif
2081
2082 vp = bp->b_vp;
2083 bo = bp->b_bufobj;
2084 ++reassignbufcalls;
2085
2086 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2087 bp, bp->b_vp, bp->b_flags);
2088 /*
2089 * B_PAGING flagged buffers cannot be reassigned because their vp
2090 * is not fully linked in.
2091 */
2092 if (bp->b_flags & B_PAGING)
2093 panic("cannot reassign paging buffer");
2094
2095 /*
2096 * Delete from old vnode list, if on one.
2097 */
2098 BO_LOCK(bo);
2099 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2100 buf_vlist_remove(bp);
2101 else
2102 panic("reassignbuf: Buffer %p not on queue.", bp);
2103 /*
2104 * If dirty, put on list of dirty buffers; otherwise insert onto list
2105 * of clean buffers.
2106 */
2107 if (bp->b_flags & B_DELWRI) {
2108 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2109 switch (vp->v_type) {
2110 case VDIR:
2111 delay = dirdelay;
2112 break;
2113 case VCHR:
2114 delay = metadelay;
2115 break;
2116 default:
2117 delay = filedelay;
2118 }
2119 vn_syncer_add_to_worklist(bo, delay);
2120 }
2121 buf_vlist_add(bp, bo, BX_VNDIRTY);
2122 } else {
2123 buf_vlist_add(bp, bo, BX_VNCLEAN);
2124
2125 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2126 mtx_lock(&sync_mtx);
2127 LIST_REMOVE(bo, bo_synclist);
2128 syncer_worklist_len--;
2129 mtx_unlock(&sync_mtx);
2130 bo->bo_flag &= ~BO_ONWORKLST;
2131 }
2132 }
2133 #ifdef INVARIANTS
2134 bv = &bo->bo_clean;
2135 bp = TAILQ_FIRST(&bv->bv_hd);
2136 KASSERT(bp == NULL || bp->b_bufobj == bo,
2137 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2138 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2139 KASSERT(bp == NULL || bp->b_bufobj == bo,
2140 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2141 bv = &bo->bo_dirty;
2142 bp = TAILQ_FIRST(&bv->bv_hd);
2143 KASSERT(bp == NULL || bp->b_bufobj == bo,
2144 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2145 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2146 KASSERT(bp == NULL || bp->b_bufobj == bo,
2147 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2148 #endif
2149 BO_UNLOCK(bo);
2150 }
2151
2152 /*
2153 * Increment the use and hold counts on the vnode, taking care to reference
2154 * the driver's usecount if this is a chardev. The vholdl() will remove
2155 * the vnode from the free list if it is presently free. Requires the
2156 * vnode interlock and returns with it held.
2157 */
2158 static void
2159 v_incr_usecount(struct vnode *vp)
2160 {
2161
2162 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2163 vholdl(vp);
2164 vp->v_usecount++;
2165 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2166 dev_lock();
2167 vp->v_rdev->si_usecount++;
2168 dev_unlock();
2169 }
2170 }
2171
2172 /*
2173 * Turn a holdcnt into a use+holdcnt such that only one call to
2174 * v_decr_usecount is needed.
2175 */
2176 static void
2177 v_upgrade_usecount(struct vnode *vp)
2178 {
2179
2180 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2181 vp->v_usecount++;
2182 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2183 dev_lock();
2184 vp->v_rdev->si_usecount++;
2185 dev_unlock();
2186 }
2187 }
2188
2189 /*
2190 * Decrement the vnode use and hold count along with the driver's usecount
2191 * if this is a chardev. The vdropl() below releases the vnode interlock
2192 * as it may free the vnode.
2193 */
2194 static void
2195 v_decr_usecount(struct vnode *vp)
2196 {
2197
2198 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2199 VNASSERT(vp->v_usecount > 0, vp,
2200 ("v_decr_usecount: negative usecount"));
2201 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2202 vp->v_usecount--;
2203 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2204 dev_lock();
2205 vp->v_rdev->si_usecount--;
2206 dev_unlock();
2207 }
2208 vdropl(vp);
2209 }
2210
2211 /*
2212 * Decrement only the use count and driver use count. This is intended to
2213 * be paired with a follow on vdropl() to release the remaining hold count.
2214 * In this way we may vgone() a vnode with a 0 usecount without risk of
2215 * having it end up on a free list because the hold count is kept above 0.
2216 */
2217 static void
2218 v_decr_useonly(struct vnode *vp)
2219 {
2220
2221 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2222 VNASSERT(vp->v_usecount > 0, vp,
2223 ("v_decr_useonly: negative usecount"));
2224 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2225 vp->v_usecount--;
2226 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2227 dev_lock();
2228 vp->v_rdev->si_usecount--;
2229 dev_unlock();
2230 }
2231 }
2232
2233 /*
2234 * Grab a particular vnode from the free list, increment its
2235 * reference count and lock it. VI_DOOMED is set if the vnode
2236 * is being destroyed. Only callers who specify LK_RETRY will
2237 * see doomed vnodes. If inactive processing was delayed in
2238 * vput try to do it here.
2239 */
2240 int
2241 vget(struct vnode *vp, int flags, struct thread *td)
2242 {
2243 int error;
2244
2245 error = 0;
2246 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2247 ("vget: invalid lock operation"));
2248 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2249
2250 if ((flags & LK_INTERLOCK) == 0)
2251 VI_LOCK(vp);
2252 vholdl(vp);
2253 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2254 vdrop(vp);
2255 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2256 vp);
2257 return (error);
2258 }
2259 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2260 panic("vget: vn_lock failed to return ENOENT\n");
2261 VI_LOCK(vp);
2262 /* Upgrade our holdcnt to a usecount. */
2263 v_upgrade_usecount(vp);
2264 /*
2265 * We don't guarantee that any particular close will
2266 * trigger inactive processing so just make a best effort
2267 * here at preventing a reference to a removed file. If
2268 * we don't succeed no harm is done.
2269 */
2270 if (vp->v_iflag & VI_OWEINACT) {
2271 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2272 (flags & LK_NOWAIT) == 0)
2273 vinactive(vp, td);
2274 vp->v_iflag &= ~VI_OWEINACT;
2275 }
2276 VI_UNLOCK(vp);
2277 return (0);
2278 }
2279
2280 /*
2281 * Increase the reference count of a vnode.
2282 */
2283 void
2284 vref(struct vnode *vp)
2285 {
2286
2287 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2288 VI_LOCK(vp);
2289 v_incr_usecount(vp);
2290 VI_UNLOCK(vp);
2291 }
2292
2293 /*
2294 * Return reference count of a vnode.
2295 *
2296 * The results of this call are only guaranteed when some mechanism other
2297 * than the VI lock is used to stop other processes from gaining references
2298 * to the vnode. This may be the case if the caller holds the only reference.
2299 * This is also useful when stale data is acceptable as race conditions may
2300 * be accounted for by some other means.
2301 */
2302 int
2303 vrefcnt(struct vnode *vp)
2304 {
2305 int usecnt;
2306
2307 VI_LOCK(vp);
2308 usecnt = vp->v_usecount;
2309 VI_UNLOCK(vp);
2310
2311 return (usecnt);
2312 }
2313
2314 #define VPUTX_VRELE 1
2315 #define VPUTX_VPUT 2
2316 #define VPUTX_VUNREF 3
2317
2318 static void
2319 vputx(struct vnode *vp, int func)
2320 {
2321 int error;
2322
2323 KASSERT(vp != NULL, ("vputx: null vp"));
2324 if (func == VPUTX_VUNREF)
2325 ASSERT_VOP_LOCKED(vp, "vunref");
2326 else if (func == VPUTX_VPUT)
2327 ASSERT_VOP_LOCKED(vp, "vput");
2328 else
2329 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2330 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2331 VI_LOCK(vp);
2332
2333 /* Skip this v_writecount check if we're going to panic below. */
2334 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2335 ("vputx: missed vn_close"));
2336 error = 0;
2337
2338 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2339 vp->v_usecount == 1)) {
2340 if (func == VPUTX_VPUT)
2341 VOP_UNLOCK(vp, 0);
2342 v_decr_usecount(vp);
2343 return;
2344 }
2345
2346 if (vp->v_usecount != 1) {
2347 vprint("vputx: negative ref count", vp);
2348 panic("vputx: negative ref cnt");
2349 }
2350 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2351 /*
2352 * We want to hold the vnode until the inactive finishes to
2353 * prevent vgone() races. We drop the use count here and the
2354 * hold count below when we're done.
2355 */
2356 v_decr_useonly(vp);
2357 /*
2358 * We must call VOP_INACTIVE with the node locked. Mark
2359 * as VI_DOINGINACT to avoid recursion.
2360 */
2361 vp->v_iflag |= VI_OWEINACT;
2362 switch (func) {
2363 case VPUTX_VRELE:
2364 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2365 VI_LOCK(vp);
2366 break;
2367 case VPUTX_VPUT:
2368 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2369 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2370 LK_NOWAIT);
2371 VI_LOCK(vp);
2372 }
2373 break;
2374 case VPUTX_VUNREF:
2375 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2376 error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2377 VI_LOCK(vp);
2378 }
2379 break;
2380 }
2381 if (vp->v_usecount > 0)
2382 vp->v_iflag &= ~VI_OWEINACT;
2383 if (error == 0) {
2384 if (vp->v_iflag & VI_OWEINACT)
2385 vinactive(vp, curthread);
2386 if (func != VPUTX_VUNREF)
2387 VOP_UNLOCK(vp, 0);
2388 }
2389 vdropl(vp);
2390 }
2391
2392 /*
2393 * Vnode put/release.
2394 * If count drops to zero, call inactive routine and return to freelist.
2395 */
2396 void
2397 vrele(struct vnode *vp)
2398 {
2399
2400 vputx(vp, VPUTX_VRELE);
2401 }
2402
2403 /*
2404 * Release an already locked vnode. This give the same effects as
2405 * unlock+vrele(), but takes less time and avoids releasing and
2406 * re-aquiring the lock (as vrele() acquires the lock internally.)
2407 */
2408 void
2409 vput(struct vnode *vp)
2410 {
2411
2412 vputx(vp, VPUTX_VPUT);
2413 }
2414
2415 /*
2416 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2417 */
2418 void
2419 vunref(struct vnode *vp)
2420 {
2421
2422 vputx(vp, VPUTX_VUNREF);
2423 }
2424
2425 /*
2426 * Somebody doesn't want the vnode recycled.
2427 */
2428 void
2429 vhold(struct vnode *vp)
2430 {
2431
2432 VI_LOCK(vp);
2433 vholdl(vp);
2434 VI_UNLOCK(vp);
2435 }
2436
2437 /*
2438 * Increase the hold count and activate if this is the first reference.
2439 */
2440 void
2441 vholdl(struct vnode *vp)
2442 {
2443 struct mount *mp;
2444
2445 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2446 #ifdef INVARIANTS
2447 /* getnewvnode() calls v_incr_usecount() without holding interlock. */
2448 if (vp->v_type != VNON || vp->v_data != NULL)
2449 ASSERT_VI_LOCKED(vp, "vholdl");
2450 #endif
2451 vp->v_holdcnt++;
2452 if ((vp->v_iflag & VI_FREE) == 0)
2453 return;
2454 VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count"));
2455 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2456 /*
2457 * Remove a vnode from the free list, mark it as in use,
2458 * and put it on the active list.
2459 */
2460 mtx_lock(&vnode_free_list_mtx);
2461 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2462 freevnodes--;
2463 vp->v_iflag &= ~VI_FREE;
2464 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2465 ("Activating already active vnode"));
2466 vp->v_iflag |= VI_ACTIVE;
2467 mp = vp->v_mount;
2468 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2469 mp->mnt_activevnodelistsize++;
2470 mtx_unlock(&vnode_free_list_mtx);
2471 }
2472
2473 /*
2474 * Note that there is one less who cares about this vnode.
2475 * vdrop() is the opposite of vhold().
2476 */
2477 void
2478 vdrop(struct vnode *vp)
2479 {
2480
2481 VI_LOCK(vp);
2482 vdropl(vp);
2483 }
2484
2485 /*
2486 * Drop the hold count of the vnode. If this is the last reference to
2487 * the vnode we place it on the free list unless it has been vgone'd
2488 * (marked VI_DOOMED) in which case we will free it.
2489 *
2490 * Because the vnode vm object keeps a hold reference on the vnode if
2491 * there is at least one resident non-cached page, the vnode cannot
2492 * leave the active list without the page cleanup done.
2493 */
2494 void
2495 vdropl(struct vnode *vp)
2496 {
2497 struct bufobj *bo;
2498 struct mount *mp;
2499 int active;
2500
2501 ASSERT_VI_LOCKED(vp, "vdropl");
2502 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2503 if (vp->v_holdcnt <= 0)
2504 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2505 vp->v_holdcnt--;
2506 if (vp->v_holdcnt > 0) {
2507 VI_UNLOCK(vp);
2508 return;
2509 }
2510 if ((vp->v_iflag & VI_DOOMED) == 0) {
2511 /*
2512 * Mark a vnode as free: remove it from its active list
2513 * and put it up for recycling on the freelist.
2514 */
2515 VNASSERT(vp->v_op != NULL, vp,
2516 ("vdropl: vnode already reclaimed."));
2517 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2518 ("vnode already free"));
2519 VNASSERT(vp->v_holdcnt == 0, vp,
2520 ("vdropl: freeing when we shouldn't"));
2521 active = vp->v_iflag & VI_ACTIVE;
2522 if ((vp->v_iflag & VI_OWEINACT) == 0) {
2523 vp->v_iflag &= ~VI_ACTIVE;
2524 mp = vp->v_mount;
2525 mtx_lock(&vnode_free_list_mtx);
2526 if (active) {
2527 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2528 v_actfreelist);
2529 mp->mnt_activevnodelistsize--;
2530 }
2531 TAILQ_INSERT_TAIL(&vnode_free_list, vp,
2532 v_actfreelist);
2533 freevnodes++;
2534 vp->v_iflag |= VI_FREE;
2535 mtx_unlock(&vnode_free_list_mtx);
2536 } else {
2537 atomic_add_long(&free_owe_inact, 1);
2538 }
2539 VI_UNLOCK(vp);
2540 return;
2541 }
2542 /*
2543 * The vnode has been marked for destruction, so free it.
2544 *
2545 * The vnode will be returned to the zone where it will
2546 * normally remain until it is needed for another vnode. We
2547 * need to cleanup (or verify that the cleanup has already
2548 * been done) any residual data left from its current use
2549 * so as not to contaminate the freshly allocated vnode.
2550 */
2551 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2552 atomic_subtract_long(&numvnodes, 1);
2553 bo = &vp->v_bufobj;
2554 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2555 ("cleaned vnode still on the free list."));
2556 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2557 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2558 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2559 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2560 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2561 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2562 VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2563 ("clean blk trie not empty"));
2564 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2565 VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2566 ("dirty blk trie not empty"));
2567 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2568 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2569 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2570 VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp,
2571 ("Dangling rangelock waiters"));
2572 VI_UNLOCK(vp);
2573 #ifdef MAC
2574 mac_vnode_destroy(vp);
2575 #endif
2576 if (vp->v_pollinfo != NULL) {
2577 destroy_vpollinfo(vp->v_pollinfo);
2578 vp->v_pollinfo = NULL;
2579 }
2580 #ifdef INVARIANTS
2581 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2582 vp->v_op = NULL;
2583 #endif
2584 bzero(&vp->v_un, sizeof(vp->v_un));
2585 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
2586 vp->v_iflag = 0;
2587 vp->v_vflag = 0;
2588 bo->bo_flag = 0;
2589 uma_zfree(vnode_zone, vp);
2590 }
2591
2592 /*
2593 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2594 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2595 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2596 * failed lock upgrade.
2597 */
2598 void
2599 vinactive(struct vnode *vp, struct thread *td)
2600 {
2601 struct vm_object *obj;
2602
2603 ASSERT_VOP_ELOCKED(vp, "vinactive");
2604 ASSERT_VI_LOCKED(vp, "vinactive");
2605 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2606 ("vinactive: recursed on VI_DOINGINACT"));
2607 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2608 vp->v_iflag |= VI_DOINGINACT;
2609 vp->v_iflag &= ~VI_OWEINACT;
2610 VI_UNLOCK(vp);
2611 /*
2612 * Before moving off the active list, we must be sure that any
2613 * modified pages are converted into the vnode's dirty
2614 * buffers, since these will no longer be checked once the
2615 * vnode is on the inactive list.
2616 *
2617 * The write-out of the dirty pages is asynchronous. At the
2618 * point that VOP_INACTIVE() is called, there could still be
2619 * pending I/O and dirty pages in the object.
2620 */
2621 obj = vp->v_object;
2622 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2623 VM_OBJECT_WLOCK(obj);
2624 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2625 VM_OBJECT_WUNLOCK(obj);
2626 }
2627 VOP_INACTIVE(vp, td);
2628 VI_LOCK(vp);
2629 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2630 ("vinactive: lost VI_DOINGINACT"));
2631 vp->v_iflag &= ~VI_DOINGINACT;
2632 }
2633
2634 /*
2635 * Remove any vnodes in the vnode table belonging to mount point mp.
2636 *
2637 * If FORCECLOSE is not specified, there should not be any active ones,
2638 * return error if any are found (nb: this is a user error, not a
2639 * system error). If FORCECLOSE is specified, detach any active vnodes
2640 * that are found.
2641 *
2642 * If WRITECLOSE is set, only flush out regular file vnodes open for
2643 * writing.
2644 *
2645 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2646 *
2647 * `rootrefs' specifies the base reference count for the root vnode
2648 * of this filesystem. The root vnode is considered busy if its
2649 * v_usecount exceeds this value. On a successful return, vflush(, td)
2650 * will call vrele() on the root vnode exactly rootrefs times.
2651 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2652 * be zero.
2653 */
2654 #ifdef DIAGNOSTIC
2655 static int busyprt = 0; /* print out busy vnodes */
2656 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2657 #endif
2658
2659 int
2660 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2661 {
2662 struct vnode *vp, *mvp, *rootvp = NULL;
2663 struct vattr vattr;
2664 int busy = 0, error;
2665
2666 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2667 rootrefs, flags);
2668 if (rootrefs > 0) {
2669 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2670 ("vflush: bad args"));
2671 /*
2672 * Get the filesystem root vnode. We can vput() it
2673 * immediately, since with rootrefs > 0, it won't go away.
2674 */
2675 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2676 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2677 __func__, error);
2678 return (error);
2679 }
2680 vput(rootvp);
2681 }
2682 loop:
2683 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2684 vholdl(vp);
2685 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2686 if (error) {
2687 vdrop(vp);
2688 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2689 goto loop;
2690 }
2691 /*
2692 * Skip over a vnodes marked VV_SYSTEM.
2693 */
2694 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2695 VOP_UNLOCK(vp, 0);
2696 vdrop(vp);
2697 continue;
2698 }
2699 /*
2700 * If WRITECLOSE is set, flush out unlinked but still open
2701 * files (even if open only for reading) and regular file
2702 * vnodes open for writing.
2703 */
2704 if (flags & WRITECLOSE) {
2705 if (vp->v_object != NULL) {
2706 VM_OBJECT_WLOCK(vp->v_object);
2707 vm_object_page_clean(vp->v_object, 0, 0, 0);
2708 VM_OBJECT_WUNLOCK(vp->v_object);
2709 }
2710 error = VOP_FSYNC(vp, MNT_WAIT, td);
2711 if (error != 0) {
2712 VOP_UNLOCK(vp, 0);
2713 vdrop(vp);
2714 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2715 return (error);
2716 }
2717 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2718 VI_LOCK(vp);
2719
2720 if ((vp->v_type == VNON ||
2721 (error == 0 && vattr.va_nlink > 0)) &&
2722 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2723 VOP_UNLOCK(vp, 0);
2724 vdropl(vp);
2725 continue;
2726 }
2727 } else
2728 VI_LOCK(vp);
2729 /*
2730 * With v_usecount == 0, all we need to do is clear out the
2731 * vnode data structures and we are done.
2732 *
2733 * If FORCECLOSE is set, forcibly close the vnode.
2734 */
2735 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2736 vgonel(vp);
2737 } else {
2738 busy++;
2739 #ifdef DIAGNOSTIC
2740 if (busyprt)
2741 vprint("vflush: busy vnode", vp);
2742 #endif
2743 }
2744 VOP_UNLOCK(vp, 0);
2745 vdropl(vp);
2746 }
2747 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2748 /*
2749 * If just the root vnode is busy, and if its refcount
2750 * is equal to `rootrefs', then go ahead and kill it.
2751 */
2752 VI_LOCK(rootvp);
2753 KASSERT(busy > 0, ("vflush: not busy"));
2754 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2755 ("vflush: usecount %d < rootrefs %d",
2756 rootvp->v_usecount, rootrefs));
2757 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2758 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2759 vgone(rootvp);
2760 VOP_UNLOCK(rootvp, 0);
2761 busy = 0;
2762 } else
2763 VI_UNLOCK(rootvp);
2764 }
2765 if (busy) {
2766 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2767 busy);
2768 return (EBUSY);
2769 }
2770 for (; rootrefs > 0; rootrefs--)
2771 vrele(rootvp);
2772 return (0);
2773 }
2774
2775 /*
2776 * Recycle an unused vnode to the front of the free list.
2777 */
2778 int
2779 vrecycle(struct vnode *vp)
2780 {
2781 int recycled;
2782
2783 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2784 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2785 recycled = 0;
2786 VI_LOCK(vp);
2787 if (vp->v_usecount == 0) {
2788 recycled = 1;
2789 vgonel(vp);
2790 }
2791 VI_UNLOCK(vp);
2792 return (recycled);
2793 }
2794
2795 /*
2796 * Eliminate all activity associated with a vnode
2797 * in preparation for reuse.
2798 */
2799 void
2800 vgone(struct vnode *vp)
2801 {
2802 VI_LOCK(vp);
2803 vgonel(vp);
2804 VI_UNLOCK(vp);
2805 }
2806
2807 static void
2808 notify_lowervp_vfs_dummy(struct mount *mp __unused,
2809 struct vnode *lowervp __unused)
2810 {
2811 }
2812
2813 /*
2814 * Notify upper mounts about reclaimed or unlinked vnode.
2815 */
2816 void
2817 vfs_notify_upper(struct vnode *vp, int event)
2818 {
2819 static struct vfsops vgonel_vfsops = {
2820 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2821 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2822 };
2823 struct mount *mp, *ump, *mmp;
2824
2825 mp = vp->v_mount;
2826 if (mp == NULL)
2827 return;
2828
2829 MNT_ILOCK(mp);
2830 if (TAILQ_EMPTY(&mp->mnt_uppers))
2831 goto unlock;
2832 MNT_IUNLOCK(mp);
2833 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2834 mmp->mnt_op = &vgonel_vfsops;
2835 mmp->mnt_kern_flag |= MNTK_MARKER;
2836 MNT_ILOCK(mp);
2837 mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2838 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2839 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2840 ump = TAILQ_NEXT(ump, mnt_upper_link);
2841 continue;
2842 }
2843 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2844 MNT_IUNLOCK(mp);
2845 switch (event) {
2846 case VFS_NOTIFY_UPPER_RECLAIM:
2847 VFS_RECLAIM_LOWERVP(ump, vp);
2848 break;
2849 case VFS_NOTIFY_UPPER_UNLINK:
2850 VFS_UNLINK_LOWERVP(ump, vp);
2851 break;
2852 default:
2853 KASSERT(0, ("invalid event %d", event));
2854 break;
2855 }
2856 MNT_ILOCK(mp);
2857 ump = TAILQ_NEXT(mmp, mnt_upper_link);
2858 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2859 }
2860 free(mmp, M_TEMP);
2861 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2862 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2863 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2864 wakeup(&mp->mnt_uppers);
2865 }
2866 unlock:
2867 MNT_IUNLOCK(mp);
2868 }
2869
2870 /*
2871 * vgone, with the vp interlock held.
2872 */
2873 static void
2874 vgonel(struct vnode *vp)
2875 {
2876 struct thread *td;
2877 int oweinact;
2878 int active;
2879 struct mount *mp;
2880
2881 ASSERT_VOP_ELOCKED(vp, "vgonel");
2882 ASSERT_VI_LOCKED(vp, "vgonel");
2883 VNASSERT(vp->v_holdcnt, vp,
2884 ("vgonel: vp %p has no reference.", vp));
2885 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2886 td = curthread;
2887
2888 /*
2889 * Don't vgonel if we're already doomed.
2890 */
2891 if (vp->v_iflag & VI_DOOMED)
2892 return;
2893 vp->v_iflag |= VI_DOOMED;
2894
2895 /*
2896 * Check to see if the vnode is in use. If so, we have to call
2897 * VOP_CLOSE() and VOP_INACTIVE().
2898 */
2899 active = vp->v_usecount;
2900 oweinact = (vp->v_iflag & VI_OWEINACT);
2901 VI_UNLOCK(vp);
2902 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2903
2904 /*
2905 * If purging an active vnode, it must be closed and
2906 * deactivated before being reclaimed.
2907 */
2908 if (active)
2909 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2910 if (oweinact || active) {
2911 VI_LOCK(vp);
2912 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2913 vinactive(vp, td);
2914 VI_UNLOCK(vp);
2915 }
2916 if (vp->v_type == VSOCK)
2917 vfs_unp_reclaim(vp);
2918
2919 /*
2920 * Clean out any buffers associated with the vnode.
2921 * If the flush fails, just toss the buffers.
2922 */
2923 mp = NULL;
2924 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2925 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2926 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
2927 while (vinvalbuf(vp, 0, 0, 0) != 0)
2928 ;
2929 }
2930
2931 BO_LOCK(&vp->v_bufobj);
2932 KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
2933 vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
2934 TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
2935 vp->v_bufobj.bo_clean.bv_cnt == 0,
2936 ("vp %p bufobj not invalidated", vp));
2937
2938 /*
2939 * For VMIO bufobj, BO_DEAD is set in vm_object_terminate()
2940 * after the object's page queue is flushed.
2941 */
2942 if (vp->v_bufobj.bo_object == NULL)
2943 vp->v_bufobj.bo_flag |= BO_DEAD;
2944 BO_UNLOCK(&vp->v_bufobj);
2945
2946 /*
2947 * Reclaim the vnode.
2948 */
2949 if (VOP_RECLAIM(vp, td))
2950 panic("vgone: cannot reclaim");
2951 if (mp != NULL)
2952 vn_finished_secondary_write(mp);
2953 VNASSERT(vp->v_object == NULL, vp,
2954 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2955 /*
2956 * Clear the advisory locks and wake up waiting threads.
2957 */
2958 (void)VOP_ADVLOCKPURGE(vp);
2959 vp->v_lockf = NULL;
2960 /*
2961 * Delete from old mount point vnode list.
2962 */
2963 delmntque(vp);
2964 cache_purge(vp);
2965 /*
2966 * Done with purge, reset to the standard lock and invalidate
2967 * the vnode.
2968 */
2969 VI_LOCK(vp);
2970 vp->v_vnlock = &vp->v_lock;
2971 vp->v_op = &dead_vnodeops;
2972 vp->v_tag = "none";
2973 vp->v_type = VBAD;
2974 }
2975
2976 /*
2977 * Calculate the total number of references to a special device.
2978 */
2979 int
2980 vcount(struct vnode *vp)
2981 {
2982 int count;
2983
2984 dev_lock();
2985 count = vp->v_rdev->si_usecount;
2986 dev_unlock();
2987 return (count);
2988 }
2989
2990 /*
2991 * Same as above, but using the struct cdev *as argument
2992 */
2993 int
2994 count_dev(struct cdev *dev)
2995 {
2996 int count;
2997
2998 dev_lock();
2999 count = dev->si_usecount;
3000 dev_unlock();
3001 return(count);
3002 }
3003
3004 /*
3005 * Print out a description of a vnode.
3006 */
3007 static char *typename[] =
3008 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
3009 "VMARKER"};
3010
3011 void
3012 vn_printf(struct vnode *vp, const char *fmt, ...)
3013 {
3014 va_list ap;
3015 char buf[256], buf2[16];
3016 u_long flags;
3017
3018 va_start(ap, fmt);
3019 vprintf(fmt, ap);
3020 va_end(ap);
3021 printf("%p: ", (void *)vp);
3022 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
3023 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
3024 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
3025 buf[0] = '\0';
3026 buf[1] = '\0';
3027 if (vp->v_vflag & VV_ROOT)
3028 strlcat(buf, "|VV_ROOT", sizeof(buf));
3029 if (vp->v_vflag & VV_ISTTY)
3030 strlcat(buf, "|VV_ISTTY", sizeof(buf));
3031 if (vp->v_vflag & VV_NOSYNC)
3032 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
3033 if (vp->v_vflag & VV_ETERNALDEV)
3034 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
3035 if (vp->v_vflag & VV_CACHEDLABEL)
3036 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
3037 if (vp->v_vflag & VV_TEXT)
3038 strlcat(buf, "|VV_TEXT", sizeof(buf));
3039 if (vp->v_vflag & VV_COPYONWRITE)
3040 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
3041 if (vp->v_vflag & VV_SYSTEM)
3042 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
3043 if (vp->v_vflag & VV_PROCDEP)
3044 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
3045 if (vp->v_vflag & VV_NOKNOTE)
3046 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
3047 if (vp->v_vflag & VV_DELETED)
3048 strlcat(buf, "|VV_DELETED", sizeof(buf));
3049 if (vp->v_vflag & VV_MD)
3050 strlcat(buf, "|VV_MD", sizeof(buf));
3051 if (vp->v_vflag & VV_FORCEINSMQ)
3052 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
3053 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
3054 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
3055 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
3056 if (flags != 0) {
3057 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
3058 strlcat(buf, buf2, sizeof(buf));
3059 }
3060 if (vp->v_iflag & VI_MOUNT)
3061 strlcat(buf, "|VI_MOUNT", sizeof(buf));
3062 if (vp->v_iflag & VI_DOOMED)
3063 strlcat(buf, "|VI_DOOMED", sizeof(buf));
3064 if (vp->v_iflag & VI_FREE)
3065 strlcat(buf, "|VI_FREE", sizeof(buf));
3066 if (vp->v_iflag & VI_ACTIVE)
3067 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
3068 if (vp->v_iflag & VI_DOINGINACT)
3069 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
3070 if (vp->v_iflag & VI_OWEINACT)
3071 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
3072 flags = vp->v_iflag & ~(VI_MOUNT | VI_DOOMED | VI_FREE |
3073 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
3074 if (flags != 0) {
3075 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
3076 strlcat(buf, buf2, sizeof(buf));
3077 }
3078 printf(" flags (%s)\n", buf + 1);
3079 if (mtx_owned(VI_MTX(vp)))
3080 printf(" VI_LOCKed");
3081 if (vp->v_object != NULL)
3082 printf(" v_object %p ref %d pages %d "
3083 "cleanbuf %d dirtybuf %d\n",
3084 vp->v_object, vp->v_object->ref_count,
3085 vp->v_object->resident_page_count,
3086 vp->v_bufobj.bo_clean.bv_cnt,
3087 vp->v_bufobj.bo_dirty.bv_cnt);
3088 printf(" ");
3089 lockmgr_printinfo(vp->v_vnlock);
3090 if (vp->v_data != NULL)
3091 VOP_PRINT(vp);
3092 }
3093
3094 #ifdef DDB
3095 /*
3096 * List all of the locked vnodes in the system.
3097 * Called when debugging the kernel.
3098 */
3099 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
3100 {
3101 struct mount *mp;
3102 struct vnode *vp;
3103
3104 /*
3105 * Note: because this is DDB, we can't obey the locking semantics
3106 * for these structures, which means we could catch an inconsistent
3107 * state and dereference a nasty pointer. Not much to be done
3108 * about that.
3109 */
3110 db_printf("Locked vnodes\n");
3111 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3112 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3113 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
3114 vprint("", vp);
3115 }
3116 }
3117 }
3118
3119 /*
3120 * Show details about the given vnode.
3121 */
3122 DB_SHOW_COMMAND(vnode, db_show_vnode)
3123 {
3124 struct vnode *vp;
3125
3126 if (!have_addr)
3127 return;
3128 vp = (struct vnode *)addr;
3129 vn_printf(vp, "vnode ");
3130 }
3131
3132 /*
3133 * Show details about the given mount point.
3134 */
3135 DB_SHOW_COMMAND(mount, db_show_mount)
3136 {
3137 struct mount *mp;
3138 struct vfsopt *opt;
3139 struct statfs *sp;
3140 struct vnode *vp;
3141 char buf[512];
3142 uint64_t mflags;
3143 u_int flags;
3144
3145 if (!have_addr) {
3146 /* No address given, print short info about all mount points. */
3147 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3148 db_printf("%p %s on %s (%s)\n", mp,
3149 mp->mnt_stat.f_mntfromname,
3150 mp->mnt_stat.f_mntonname,
3151 mp->mnt_stat.f_fstypename);
3152 if (db_pager_quit)
3153 break;
3154 }
3155 db_printf("\nMore info: show mount <addr>\n");
3156 return;
3157 }
3158
3159 mp = (struct mount *)addr;
3160 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3161 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3162
3163 buf[0] = '\0';
3164 mflags = mp->mnt_flag;
3165 #define MNT_FLAG(flag) do { \
3166 if (mflags & (flag)) { \
3167 if (buf[0] != '\0') \
3168 strlcat(buf, ", ", sizeof(buf)); \
3169 strlcat(buf, (#flag) + 4, sizeof(buf)); \
3170 mflags &= ~(flag); \
3171 } \
3172 } while (0)
3173 MNT_FLAG(MNT_RDONLY);
3174 MNT_FLAG(MNT_SYNCHRONOUS);
3175 MNT_FLAG(MNT_NOEXEC);
3176 MNT_FLAG(MNT_NOSUID);
3177 MNT_FLAG(MNT_NFS4ACLS);
3178 MNT_FLAG(MNT_UNION);
3179 MNT_FLAG(MNT_ASYNC);
3180 MNT_FLAG(MNT_SUIDDIR);
3181 MNT_FLAG(MNT_SOFTDEP);
3182 MNT_FLAG(MNT_NOSYMFOLLOW);
3183 MNT_FLAG(MNT_GJOURNAL);
3184 MNT_FLAG(MNT_MULTILABEL);
3185 MNT_FLAG(MNT_ACLS);
3186 MNT_FLAG(MNT_NOATIME);
3187 MNT_FLAG(MNT_NOCLUSTERR);
3188 MNT_FLAG(MNT_NOCLUSTERW);
3189 MNT_FLAG(MNT_SUJ);
3190 MNT_FLAG(MNT_EXRDONLY);
3191 MNT_FLAG(MNT_EXPORTED);
3192 MNT_FLAG(MNT_DEFEXPORTED);
3193 MNT_FLAG(MNT_EXPORTANON);
3194 MNT_FLAG(MNT_EXKERB);
3195 MNT_FLAG(MNT_EXPUBLIC);
3196 MNT_FLAG(MNT_LOCAL);
3197 MNT_FLAG(MNT_QUOTA);
3198 MNT_FLAG(MNT_ROOTFS);
3199 MNT_FLAG(MNT_USER);
3200 MNT_FLAG(MNT_IGNORE);
3201 MNT_FLAG(MNT_UPDATE);
3202 MNT_FLAG(MNT_DELEXPORT);
3203 MNT_FLAG(MNT_RELOAD);
3204 MNT_FLAG(MNT_FORCE);
3205 MNT_FLAG(MNT_SNAPSHOT);
3206 MNT_FLAG(MNT_BYFSID);
3207 #undef MNT_FLAG
3208 if (mflags != 0) {
3209 if (buf[0] != '\0')
3210 strlcat(buf, ", ", sizeof(buf));
3211 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3212 "0x%016jx", mflags);
3213 }
3214 db_printf(" mnt_flag = %s\n", buf);
3215
3216 buf[0] = '\0';
3217 flags = mp->mnt_kern_flag;
3218 #define MNT_KERN_FLAG(flag) do { \
3219 if (flags & (flag)) { \
3220 if (buf[0] != '\0') \
3221 strlcat(buf, ", ", sizeof(buf)); \
3222 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3223 flags &= ~(flag); \
3224 } \
3225 } while (0)
3226 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3227 MNT_KERN_FLAG(MNTK_ASYNC);
3228 MNT_KERN_FLAG(MNTK_SOFTDEP);
3229 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3230 MNT_KERN_FLAG(MNTK_DRAINING);
3231 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3232 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3233 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3234 MNT_KERN_FLAG(MNTK_NO_IOPF);
3235 MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3236 MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3237 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3238 MNT_KERN_FLAG(MNTK_MARKER);
3239 MNT_KERN_FLAG(MNTK_USES_BCACHE);
3240 MNT_KERN_FLAG(MNTK_NOASYNC);
3241 MNT_KERN_FLAG(MNTK_UNMOUNT);
3242 MNT_KERN_FLAG(MNTK_MWAIT);
3243 MNT_KERN_FLAG(MNTK_SUSPEND);
3244 MNT_KERN_FLAG(MNTK_SUSPEND2);
3245 MNT_KERN_FLAG(MNTK_SUSPENDED);
3246 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3247 MNT_KERN_FLAG(MNTK_NOKNOTE);
3248 #undef MNT_KERN_FLAG
3249 if (flags != 0) {
3250 if (buf[0] != '\0')
3251 strlcat(buf, ", ", sizeof(buf));
3252 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3253 "0x%08x", flags);
3254 }
3255 db_printf(" mnt_kern_flag = %s\n", buf);
3256
3257 db_printf(" mnt_opt = ");
3258 opt = TAILQ_FIRST(mp->mnt_opt);
3259 if (opt != NULL) {
3260 db_printf("%s", opt->name);
3261 opt = TAILQ_NEXT(opt, link);
3262 while (opt != NULL) {
3263 db_printf(", %s", opt->name);
3264 opt = TAILQ_NEXT(opt, link);
3265 }
3266 }
3267 db_printf("\n");
3268
3269 sp = &mp->mnt_stat;
3270 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3271 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3272 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3273 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3274 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3275 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3276 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3277 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3278 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3279 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3280 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3281 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3282
3283 db_printf(" mnt_cred = { uid=%u ruid=%u",
3284 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3285 if (jailed(mp->mnt_cred))
3286 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3287 db_printf(" }\n");
3288 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3289 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3290 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3291 db_printf(" mnt_activevnodelistsize = %d\n",
3292 mp->mnt_activevnodelistsize);
3293 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3294 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3295 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3296 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3297 db_printf(" mnt_lockref = %d\n", mp->mnt_lockref);
3298 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3299 db_printf(" mnt_secondary_accwrites = %d\n",
3300 mp->mnt_secondary_accwrites);
3301 db_printf(" mnt_gjprovider = %s\n",
3302 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3303
3304 db_printf("\n\nList of active vnodes\n");
3305 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3306 if (vp->v_type != VMARKER) {
3307 vn_printf(vp, "vnode ");
3308 if (db_pager_quit)
3309 break;
3310 }
3311 }
3312 db_printf("\n\nList of inactive vnodes\n");
3313 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3314 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3315 vn_printf(vp, "vnode ");
3316 if (db_pager_quit)
3317 break;
3318 }
3319 }
3320 }
3321 #endif /* DDB */
3322
3323 /*
3324 * Fill in a struct xvfsconf based on a struct vfsconf.
3325 */
3326 static int
3327 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3328 {
3329 struct xvfsconf xvfsp;
3330
3331 bzero(&xvfsp, sizeof(xvfsp));
3332 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3333 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3334 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3335 xvfsp.vfc_flags = vfsp->vfc_flags;
3336 /*
3337 * These are unused in userland, we keep them
3338 * to not break binary compatibility.
3339 */
3340 xvfsp.vfc_vfsops = NULL;
3341 xvfsp.vfc_next = NULL;
3342 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3343 }
3344
3345 #ifdef COMPAT_FREEBSD32
3346 struct xvfsconf32 {
3347 uint32_t vfc_vfsops;
3348 char vfc_name[MFSNAMELEN];
3349 int32_t vfc_typenum;
3350 int32_t vfc_refcount;
3351 int32_t vfc_flags;
3352 uint32_t vfc_next;
3353 };
3354
3355 static int
3356 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3357 {
3358 struct xvfsconf32 xvfsp;
3359
3360 bzero(&xvfsp, sizeof(xvfsp));
3361 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3362 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3363 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3364 xvfsp.vfc_flags = vfsp->vfc_flags;
3365 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3366 }
3367 #endif
3368
3369 /*
3370 * Top level filesystem related information gathering.
3371 */
3372 static int
3373 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3374 {
3375 struct vfsconf *vfsp;
3376 int error;
3377
3378 error = 0;
3379 vfsconf_slock();
3380 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3381 #ifdef COMPAT_FREEBSD32
3382 if (req->flags & SCTL_MASK32)
3383 error = vfsconf2x32(req, vfsp);
3384 else
3385 #endif
3386 error = vfsconf2x(req, vfsp);
3387 if (error)
3388 break;
3389 }
3390 vfsconf_sunlock();
3391 return (error);
3392 }
3393
3394 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
3395 CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
3396 "S,xvfsconf", "List of all configured filesystems");
3397
3398 #ifndef BURN_BRIDGES
3399 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3400
3401 static int
3402 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3403 {
3404 int *name = (int *)arg1 - 1; /* XXX */
3405 u_int namelen = arg2 + 1; /* XXX */
3406 struct vfsconf *vfsp;
3407
3408 log(LOG_WARNING, "userland calling deprecated sysctl, "
3409 "please rebuild world\n");
3410
3411 #if 1 || defined(COMPAT_PRELITE2)
3412 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3413 if (namelen == 1)
3414 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3415 #endif
3416
3417 switch (name[1]) {
3418 case VFS_MAXTYPENUM:
3419 if (namelen != 2)
3420 return (ENOTDIR);
3421 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3422 case VFS_CONF:
3423 if (namelen != 3)
3424 return (ENOTDIR); /* overloaded */
3425 vfsconf_slock();
3426 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3427 if (vfsp->vfc_typenum == name[2])
3428 break;
3429 }
3430 vfsconf_sunlock();
3431 if (vfsp == NULL)
3432 return (EOPNOTSUPP);
3433 #ifdef COMPAT_FREEBSD32
3434 if (req->flags & SCTL_MASK32)
3435 return (vfsconf2x32(req, vfsp));
3436 else
3437 #endif
3438 return (vfsconf2x(req, vfsp));
3439 }
3440 return (EOPNOTSUPP);
3441 }
3442
3443 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
3444 CTLFLAG_MPSAFE, vfs_sysctl,
3445 "Generic filesystem");
3446
3447 #if 1 || defined(COMPAT_PRELITE2)
3448
3449 static int
3450 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3451 {
3452 int error;
3453 struct vfsconf *vfsp;
3454 struct ovfsconf ovfs;
3455
3456 vfsconf_slock();
3457 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3458 bzero(&ovfs, sizeof(ovfs));
3459 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3460 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3461 ovfs.vfc_index = vfsp->vfc_typenum;
3462 ovfs.vfc_refcount = vfsp->vfc_refcount;
3463 ovfs.vfc_flags = vfsp->vfc_flags;
3464 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3465 if (error != 0) {
3466 vfsconf_sunlock();
3467 return (error);
3468 }
3469 }
3470 vfsconf_sunlock();
3471 return (0);
3472 }
3473
3474 #endif /* 1 || COMPAT_PRELITE2 */
3475 #endif /* !BURN_BRIDGES */
3476
3477 #define KINFO_VNODESLOP 10
3478 #ifdef notyet
3479 /*
3480 * Dump vnode list (via sysctl).
3481 */
3482 /* ARGSUSED */
3483 static int
3484 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3485 {
3486 struct xvnode *xvn;
3487 struct mount *mp;
3488 struct vnode *vp;
3489 int error, len, n;
3490
3491 /*
3492 * Stale numvnodes access is not fatal here.
3493 */
3494 req->lock = 0;
3495 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3496 if (!req->oldptr)
3497 /* Make an estimate */
3498 return (SYSCTL_OUT(req, 0, len));
3499
3500 error = sysctl_wire_old_buffer(req, 0);
3501 if (error != 0)
3502 return (error);
3503 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3504 n = 0;
3505 mtx_lock(&mountlist_mtx);
3506 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3507 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3508 continue;
3509 MNT_ILOCK(mp);
3510 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3511 if (n == len)
3512 break;
3513 vref(vp);
3514 xvn[n].xv_size = sizeof *xvn;
3515 xvn[n].xv_vnode = vp;
3516 xvn[n].xv_id = 0; /* XXX compat */
3517 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3518 XV_COPY(usecount);
3519 XV_COPY(writecount);
3520 XV_COPY(holdcnt);
3521 XV_COPY(mount);
3522 XV_COPY(numoutput);
3523 XV_COPY(type);
3524 #undef XV_COPY
3525 xvn[n].xv_flag = vp->v_vflag;
3526
3527 switch (vp->v_type) {
3528 case VREG:
3529 case VDIR:
3530 case VLNK:
3531 break;
3532 case VBLK:
3533 case VCHR:
3534 if (vp->v_rdev == NULL) {
3535 vrele(vp);
3536 continue;
3537 }
3538 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3539 break;
3540 case VSOCK:
3541 xvn[n].xv_socket = vp->v_socket;
3542 break;
3543 case VFIFO:
3544 xvn[n].xv_fifo = vp->v_fifoinfo;
3545 break;
3546 case VNON:
3547 case VBAD:
3548 default:
3549 /* shouldn't happen? */
3550 vrele(vp);
3551 continue;
3552 }
3553 vrele(vp);
3554 ++n;
3555 }
3556 MNT_IUNLOCK(mp);
3557 mtx_lock(&mountlist_mtx);
3558 vfs_unbusy(mp);
3559 if (n == len)
3560 break;
3561 }
3562 mtx_unlock(&mountlist_mtx);
3563
3564 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3565 free(xvn, M_TEMP);
3566 return (error);
3567 }
3568
3569 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
3570 CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
3571 "");
3572 #endif
3573
3574 static void
3575 unmount_or_warn(struct mount *mp)
3576 {
3577 int error;
3578
3579 error = dounmount(mp, MNT_FORCE, curthread);
3580 if (error != 0 && strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3581 printf("unmount of %s failed (", mp->mnt_stat.f_mntonname);
3582 if (error == EBUSY)
3583 printf("BUSY)\n");
3584 else
3585 printf("%d)\n", error);
3586 }
3587 }
3588
3589 /*
3590 * Unmount all filesystems. The list is traversed in reverse order
3591 * of mounting to avoid dependencies.
3592 */
3593 void
3594 vfs_unmountall(void)
3595 {
3596 struct mount *mp, *tmp;
3597
3598 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3599
3600 /*
3601 * Since this only runs when rebooting, it is not interlocked.
3602 */
3603 TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) {
3604 vfs_ref(mp);
3605
3606 /*
3607 * Forcibly unmounting "/dev" before "/" would prevent clean
3608 * unmount of the latter.
3609 */
3610 if (mp == rootdevmp)
3611 continue;
3612
3613 unmount_or_warn(mp);
3614 }
3615
3616 if (rootdevmp != NULL)
3617 unmount_or_warn(rootdevmp);
3618 }
3619
3620 /*
3621 * perform msync on all vnodes under a mount point
3622 * the mount point must be locked.
3623 */
3624 void
3625 vfs_msync(struct mount *mp, int flags)
3626 {
3627 struct vnode *vp, *mvp;
3628 struct vm_object *obj;
3629
3630 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3631 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3632 obj = vp->v_object;
3633 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3634 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3635 if (!vget(vp,
3636 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3637 curthread)) {
3638 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3639 vput(vp);
3640 continue;
3641 }
3642
3643 obj = vp->v_object;
3644 if (obj != NULL) {
3645 VM_OBJECT_WLOCK(obj);
3646 vm_object_page_clean(obj, 0, 0,
3647 flags == MNT_WAIT ?
3648 OBJPC_SYNC : OBJPC_NOSYNC);
3649 VM_OBJECT_WUNLOCK(obj);
3650 }
3651 vput(vp);
3652 }
3653 } else
3654 VI_UNLOCK(vp);
3655 }
3656 }
3657
3658 static void
3659 destroy_vpollinfo_free(struct vpollinfo *vi)
3660 {
3661
3662 knlist_destroy(&vi->vpi_selinfo.si_note);
3663 mtx_destroy(&vi->vpi_lock);
3664 uma_zfree(vnodepoll_zone, vi);
3665 }
3666
3667 static void
3668 destroy_vpollinfo(struct vpollinfo *vi)
3669 {
3670
3671 knlist_clear(&vi->vpi_selinfo.si_note, 1);
3672 seldrain(&vi->vpi_selinfo);
3673 destroy_vpollinfo_free(vi);
3674 }
3675
3676 /*
3677 * Initialize per-vnode helper structure to hold poll-related state.
3678 */
3679 void
3680 v_addpollinfo(struct vnode *vp)
3681 {
3682 struct vpollinfo *vi;
3683
3684 if (vp->v_pollinfo != NULL)
3685 return;
3686 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3687 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3688 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3689 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3690 VI_LOCK(vp);
3691 if (vp->v_pollinfo != NULL) {
3692 VI_UNLOCK(vp);
3693 destroy_vpollinfo_free(vi);
3694 return;
3695 }
3696 vp->v_pollinfo = vi;
3697 VI_UNLOCK(vp);
3698 }
3699
3700 /*
3701 * Record a process's interest in events which might happen to
3702 * a vnode. Because poll uses the historic select-style interface
3703 * internally, this routine serves as both the ``check for any
3704 * pending events'' and the ``record my interest in future events''
3705 * functions. (These are done together, while the lock is held,
3706 * to avoid race conditions.)
3707 */
3708 int
3709 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3710 {
3711
3712 v_addpollinfo(vp);
3713 mtx_lock(&vp->v_pollinfo->vpi_lock);
3714 if (vp->v_pollinfo->vpi_revents & events) {
3715 /*
3716 * This leaves events we are not interested
3717 * in available for the other process which
3718 * which presumably had requested them
3719 * (otherwise they would never have been
3720 * recorded).
3721 */
3722 events &= vp->v_pollinfo->vpi_revents;
3723 vp->v_pollinfo->vpi_revents &= ~events;
3724
3725 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3726 return (events);
3727 }
3728 vp->v_pollinfo->vpi_events |= events;
3729 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3730 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3731 return (0);
3732 }
3733
3734 /*
3735 * Routine to create and manage a filesystem syncer vnode.
3736 */
3737 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3738 static int sync_fsync(struct vop_fsync_args *);
3739 static int sync_inactive(struct vop_inactive_args *);
3740 static int sync_reclaim(struct vop_reclaim_args *);
3741
3742 static struct vop_vector sync_vnodeops = {
3743 .vop_bypass = VOP_EOPNOTSUPP,
3744 .vop_close = sync_close, /* close */
3745 .vop_fsync = sync_fsync, /* fsync */
3746 .vop_inactive = sync_inactive, /* inactive */
3747 .vop_reclaim = sync_reclaim, /* reclaim */
3748 .vop_lock1 = vop_stdlock, /* lock */
3749 .vop_unlock = vop_stdunlock, /* unlock */
3750 .vop_islocked = vop_stdislocked, /* islocked */
3751 };
3752
3753 /*
3754 * Create a new filesystem syncer vnode for the specified mount point.
3755 */
3756 void
3757 vfs_allocate_syncvnode(struct mount *mp)
3758 {
3759 struct vnode *vp;
3760 struct bufobj *bo;
3761 static long start, incr, next;
3762 int error;
3763
3764 /* Allocate a new vnode */
3765 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3766 if (error != 0)
3767 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3768 vp->v_type = VNON;
3769 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3770 vp->v_vflag |= VV_FORCEINSMQ;
3771 error = insmntque(vp, mp);
3772 if (error != 0)
3773 panic("vfs_allocate_syncvnode: insmntque() failed");
3774 vp->v_vflag &= ~VV_FORCEINSMQ;
3775 VOP_UNLOCK(vp, 0);
3776 /*
3777 * Place the vnode onto the syncer worklist. We attempt to
3778 * scatter them about on the list so that they will go off
3779 * at evenly distributed times even if all the filesystems
3780 * are mounted at once.
3781 */
3782 next += incr;
3783 if (next == 0 || next > syncer_maxdelay) {
3784 start /= 2;
3785 incr /= 2;
3786 if (start == 0) {
3787 start = syncer_maxdelay / 2;
3788 incr = syncer_maxdelay;
3789 }
3790 next = start;
3791 }
3792 bo = &vp->v_bufobj;
3793 BO_LOCK(bo);
3794 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3795 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3796 mtx_lock(&sync_mtx);
3797 sync_vnode_count++;
3798 if (mp->mnt_syncer == NULL) {
3799 mp->mnt_syncer = vp;
3800 vp = NULL;
3801 }
3802 mtx_unlock(&sync_mtx);
3803 BO_UNLOCK(bo);
3804 if (vp != NULL) {
3805 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3806 vgone(vp);
3807 vput(vp);
3808 }
3809 }
3810
3811 void
3812 vfs_deallocate_syncvnode(struct mount *mp)
3813 {
3814 struct vnode *vp;
3815
3816 mtx_lock(&sync_mtx);
3817 vp = mp->mnt_syncer;
3818 if (vp != NULL)
3819 mp->mnt_syncer = NULL;
3820 mtx_unlock(&sync_mtx);
3821 if (vp != NULL)
3822 vrele(vp);
3823 }
3824
3825 /*
3826 * Do a lazy sync of the filesystem.
3827 */
3828 static int
3829 sync_fsync(struct vop_fsync_args *ap)
3830 {
3831 struct vnode *syncvp = ap->a_vp;
3832 struct mount *mp = syncvp->v_mount;
3833 int error, save;
3834 struct bufobj *bo;
3835
3836 /*
3837 * We only need to do something if this is a lazy evaluation.
3838 */
3839 if (ap->a_waitfor != MNT_LAZY)
3840 return (0);
3841
3842 /*
3843 * Move ourselves to the back of the sync list.
3844 */
3845 bo = &syncvp->v_bufobj;
3846 BO_LOCK(bo);
3847 vn_syncer_add_to_worklist(bo, syncdelay);
3848 BO_UNLOCK(bo);
3849
3850 /*
3851 * Walk the list of vnodes pushing all that are dirty and
3852 * not already on the sync list.
3853 */
3854 if (vfs_busy(mp, MBF_NOWAIT) != 0)
3855 return (0);
3856 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3857 vfs_unbusy(mp);
3858 return (0);
3859 }
3860 save = curthread_pflags_set(TDP_SYNCIO);
3861 vfs_msync(mp, MNT_NOWAIT);
3862 error = VFS_SYNC(mp, MNT_LAZY);
3863 curthread_pflags_restore(save);
3864 vn_finished_write(mp);
3865 vfs_unbusy(mp);
3866 return (error);
3867 }
3868
3869 /*
3870 * The syncer vnode is no referenced.
3871 */
3872 static int
3873 sync_inactive(struct vop_inactive_args *ap)
3874 {
3875
3876 vgone(ap->a_vp);
3877 return (0);
3878 }
3879
3880 /*
3881 * The syncer vnode is no longer needed and is being decommissioned.
3882 *
3883 * Modifications to the worklist must be protected by sync_mtx.
3884 */
3885 static int
3886 sync_reclaim(struct vop_reclaim_args *ap)
3887 {
3888 struct vnode *vp = ap->a_vp;
3889 struct bufobj *bo;
3890
3891 bo = &vp->v_bufobj;
3892 BO_LOCK(bo);
3893 mtx_lock(&sync_mtx);
3894 if (vp->v_mount->mnt_syncer == vp)
3895 vp->v_mount->mnt_syncer = NULL;
3896 if (bo->bo_flag & BO_ONWORKLST) {
3897 LIST_REMOVE(bo, bo_synclist);
3898 syncer_worklist_len--;
3899 sync_vnode_count--;
3900 bo->bo_flag &= ~BO_ONWORKLST;
3901 }
3902 mtx_unlock(&sync_mtx);
3903 BO_UNLOCK(bo);
3904
3905 return (0);
3906 }
3907
3908 /*
3909 * Check if vnode represents a disk device
3910 */
3911 int
3912 vn_isdisk(struct vnode *vp, int *errp)
3913 {
3914 int error;
3915
3916 if (vp->v_type != VCHR) {
3917 error = ENOTBLK;
3918 goto out;
3919 }
3920 error = 0;
3921 dev_lock();
3922 if (vp->v_rdev == NULL)
3923 error = ENXIO;
3924 else if (vp->v_rdev->si_devsw == NULL)
3925 error = ENXIO;
3926 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3927 error = ENOTBLK;
3928 dev_unlock();
3929 out:
3930 if (errp != NULL)
3931 *errp = error;
3932 return (error == 0);
3933 }
3934
3935 /*
3936 * Common filesystem object access control check routine. Accepts a
3937 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3938 * and optional call-by-reference privused argument allowing vaccess()
3939 * to indicate to the caller whether privilege was used to satisfy the
3940 * request (obsoleted). Returns 0 on success, or an errno on failure.
3941 */
3942 int
3943 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3944 accmode_t accmode, struct ucred *cred, int *privused)
3945 {
3946 accmode_t dac_granted;
3947 accmode_t priv_granted;
3948
3949 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3950 ("invalid bit in accmode"));
3951 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3952 ("VAPPEND without VWRITE"));
3953
3954 /*
3955 * Look for a normal, non-privileged way to access the file/directory
3956 * as requested. If it exists, go with that.
3957 */
3958
3959 if (privused != NULL)
3960 *privused = 0;
3961
3962 dac_granted = 0;
3963
3964 /* Check the owner. */
3965 if (cred->cr_uid == file_uid) {
3966 dac_granted |= VADMIN;
3967 if (file_mode & S_IXUSR)
3968 dac_granted |= VEXEC;
3969 if (file_mode & S_IRUSR)
3970 dac_granted |= VREAD;
3971 if (file_mode & S_IWUSR)
3972 dac_granted |= (VWRITE | VAPPEND);
3973
3974 if ((accmode & dac_granted) == accmode)
3975 return (0);
3976
3977 goto privcheck;
3978 }
3979
3980 /* Otherwise, check the groups (first match) */
3981 if (groupmember(file_gid, cred)) {
3982 if (file_mode & S_IXGRP)
3983 dac_granted |= VEXEC;
3984 if (file_mode & S_IRGRP)
3985 dac_granted |= VREAD;
3986 if (file_mode & S_IWGRP)
3987 dac_granted |= (VWRITE | VAPPEND);
3988
3989 if ((accmode & dac_granted) == accmode)
3990 return (0);
3991
3992 goto privcheck;
3993 }
3994
3995 /* Otherwise, check everyone else. */
3996 if (file_mode & S_IXOTH)
3997 dac_granted |= VEXEC;
3998 if (file_mode & S_IROTH)
3999 dac_granted |= VREAD;
4000 if (file_mode & S_IWOTH)
4001 dac_granted |= (VWRITE | VAPPEND);
4002 if ((accmode & dac_granted) == accmode)
4003 return (0);
4004
4005 privcheck:
4006 /*
4007 * Build a privilege mask to determine if the set of privileges
4008 * satisfies the requirements when combined with the granted mask
4009 * from above. For each privilege, if the privilege is required,
4010 * bitwise or the request type onto the priv_granted mask.
4011 */
4012 priv_granted = 0;
4013
4014 if (type == VDIR) {
4015 /*
4016 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
4017 * requests, instead of PRIV_VFS_EXEC.
4018 */
4019 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4020 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
4021 priv_granted |= VEXEC;
4022 } else {
4023 /*
4024 * Ensure that at least one execute bit is on. Otherwise,
4025 * a privileged user will always succeed, and we don't want
4026 * this to happen unless the file really is executable.
4027 */
4028 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4029 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
4030 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
4031 priv_granted |= VEXEC;
4032 }
4033
4034 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
4035 !priv_check_cred(cred, PRIV_VFS_READ, 0))
4036 priv_granted |= VREAD;
4037
4038 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
4039 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
4040 priv_granted |= (VWRITE | VAPPEND);
4041
4042 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
4043 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
4044 priv_granted |= VADMIN;
4045
4046 if ((accmode & (priv_granted | dac_granted)) == accmode) {
4047 /* XXX audit: privilege used */
4048 if (privused != NULL)
4049 *privused = 1;
4050 return (0);
4051 }
4052
4053 return ((accmode & VADMIN) ? EPERM : EACCES);
4054 }
4055
4056 /*
4057 * Credential check based on process requesting service, and per-attribute
4058 * permissions.
4059 */
4060 int
4061 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
4062 struct thread *td, accmode_t accmode)
4063 {
4064
4065 /*
4066 * Kernel-invoked always succeeds.
4067 */
4068 if (cred == NOCRED)
4069 return (0);
4070
4071 /*
4072 * Do not allow privileged processes in jail to directly manipulate
4073 * system attributes.
4074 */
4075 switch (attrnamespace) {
4076 case EXTATTR_NAMESPACE_SYSTEM:
4077 /* Potentially should be: return (EPERM); */
4078 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
4079 case EXTATTR_NAMESPACE_USER:
4080 return (VOP_ACCESS(vp, accmode, cred, td));
4081 default:
4082 return (EPERM);
4083 }
4084 }
4085
4086 #ifdef DEBUG_VFS_LOCKS
4087 /*
4088 * This only exists to suppress warnings from unlocked specfs accesses. It is
4089 * no longer ok to have an unlocked VFS.
4090 */
4091 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
4092 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
4093
4094 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
4095 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
4096 "Drop into debugger on lock violation");
4097
4098 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
4099 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
4100 0, "Check for interlock across VOPs");
4101
4102 int vfs_badlock_print = 1; /* Print lock violations. */
4103 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
4104 0, "Print lock violations");
4105
4106 #ifdef KDB
4107 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
4108 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
4109 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
4110 #endif
4111
4112 static void
4113 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
4114 {
4115
4116 #ifdef KDB
4117 if (vfs_badlock_backtrace)
4118 kdb_backtrace();
4119 #endif
4120 if (vfs_badlock_print)
4121 printf("%s: %p %s\n", str, (void *)vp, msg);
4122 if (vfs_badlock_ddb)
4123 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4124 }
4125
4126 void
4127 assert_vi_locked(struct vnode *vp, const char *str)
4128 {
4129
4130 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4131 vfs_badlock("interlock is not locked but should be", str, vp);
4132 }
4133
4134 void
4135 assert_vi_unlocked(struct vnode *vp, const char *str)
4136 {
4137
4138 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4139 vfs_badlock("interlock is locked but should not be", str, vp);
4140 }
4141
4142 void
4143 assert_vop_locked(struct vnode *vp, const char *str)
4144 {
4145 int locked;
4146
4147 if (!IGNORE_LOCK(vp)) {
4148 locked = VOP_ISLOCKED(vp);
4149 if (locked == 0 || locked == LK_EXCLOTHER)
4150 vfs_badlock("is not locked but should be", str, vp);
4151 }
4152 }
4153
4154 void
4155 assert_vop_unlocked(struct vnode *vp, const char *str)
4156 {
4157
4158 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4159 vfs_badlock("is locked but should not be", str, vp);
4160 }
4161
4162 void
4163 assert_vop_elocked(struct vnode *vp, const char *str)
4164 {
4165
4166 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4167 vfs_badlock("is not exclusive locked but should be", str, vp);
4168 }
4169
4170 #if 0
4171 void
4172 assert_vop_elocked_other(struct vnode *vp, const char *str)
4173 {
4174
4175 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4176 vfs_badlock("is not exclusive locked by another thread",
4177 str, vp);
4178 }
4179
4180 void
4181 assert_vop_slocked(struct vnode *vp, const char *str)
4182 {
4183
4184 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4185 vfs_badlock("is not locked shared but should be", str, vp);
4186 }
4187 #endif /* 0 */
4188 #endif /* DEBUG_VFS_LOCKS */
4189
4190 void
4191 vop_rename_fail(struct vop_rename_args *ap)
4192 {
4193
4194 if (ap->a_tvp != NULL)
4195 vput(ap->a_tvp);
4196 if (ap->a_tdvp == ap->a_tvp)
4197 vrele(ap->a_tdvp);
4198 else
4199 vput(ap->a_tdvp);
4200 vrele(ap->a_fdvp);
4201 vrele(ap->a_fvp);
4202 }
4203
4204 void
4205 vop_rename_pre(void *ap)
4206 {
4207 struct vop_rename_args *a = ap;
4208
4209 #ifdef DEBUG_VFS_LOCKS
4210 if (a->a_tvp)
4211 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4212 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4213 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4214 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4215
4216 /* Check the source (from). */
4217 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4218 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4219 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4220 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4221 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4222
4223 /* Check the target. */
4224 if (a->a_tvp)
4225 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4226 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4227 #endif
4228 if (a->a_tdvp != a->a_fdvp)
4229 vhold(a->a_fdvp);
4230 if (a->a_tvp != a->a_fvp)
4231 vhold(a->a_fvp);
4232 vhold(a->a_tdvp);
4233 if (a->a_tvp)
4234 vhold(a->a_tvp);
4235 }
4236
4237 void
4238 vop_strategy_pre(void *ap)
4239 {
4240 #ifdef DEBUG_VFS_LOCKS
4241 struct vop_strategy_args *a;
4242 struct buf *bp;
4243
4244 a = ap;
4245 bp = a->a_bp;
4246
4247 /*
4248 * Cluster ops lock their component buffers but not the IO container.
4249 */
4250 if ((bp->b_flags & B_CLUSTER) != 0)
4251 return;
4252
4253 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4254 if (vfs_badlock_print)
4255 printf(
4256 "VOP_STRATEGY: bp is not locked but should be\n");
4257 if (vfs_badlock_ddb)
4258 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4259 }
4260 #endif
4261 }
4262
4263 void
4264 vop_lock_pre(void *ap)
4265 {
4266 #ifdef DEBUG_VFS_LOCKS
4267 struct vop_lock1_args *a = ap;
4268
4269 if ((a->a_flags & LK_INTERLOCK) == 0)
4270 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4271 else
4272 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4273 #endif
4274 }
4275
4276 void
4277 vop_lock_post(void *ap, int rc)
4278 {
4279 #ifdef DEBUG_VFS_LOCKS
4280 struct vop_lock1_args *a = ap;
4281
4282 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4283 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4284 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4285 #endif
4286 }
4287
4288 void
4289 vop_unlock_pre(void *ap)
4290 {
4291 #ifdef DEBUG_VFS_LOCKS
4292 struct vop_unlock_args *a = ap;
4293
4294 if (a->a_flags & LK_INTERLOCK)
4295 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4296 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4297 #endif
4298 }
4299
4300 void
4301 vop_unlock_post(void *ap, int rc)
4302 {
4303 #ifdef DEBUG_VFS_LOCKS
4304 struct vop_unlock_args *a = ap;
4305
4306 if (a->a_flags & LK_INTERLOCK)
4307 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4308 #endif
4309 }
4310
4311 void
4312 vop_create_post(void *ap, int rc)
4313 {
4314 struct vop_create_args *a = ap;
4315
4316 if (!rc)
4317 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4318 }
4319
4320 void
4321 vop_deleteextattr_post(void *ap, int rc)
4322 {
4323 struct vop_deleteextattr_args *a = ap;
4324
4325 if (!rc)
4326 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4327 }
4328
4329 void
4330 vop_link_post(void *ap, int rc)
4331 {
4332 struct vop_link_args *a = ap;
4333
4334 if (!rc) {
4335 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4336 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4337 }
4338 }
4339
4340 void
4341 vop_mkdir_post(void *ap, int rc)
4342 {
4343 struct vop_mkdir_args *a = ap;
4344
4345 if (!rc)
4346 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4347 }
4348
4349 void
4350 vop_mknod_post(void *ap, int rc)
4351 {
4352 struct vop_mknod_args *a = ap;
4353
4354 if (!rc)
4355 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4356 }
4357
4358 void
4359 vop_reclaim_post(void *ap, int rc)
4360 {
4361 struct vop_reclaim_args *a = ap;
4362
4363 if (!rc)
4364 VFS_KNOTE_LOCKED(a->a_vp, NOTE_REVOKE);
4365 }
4366
4367 void
4368 vop_remove_post(void *ap, int rc)
4369 {
4370 struct vop_remove_args *a = ap;
4371
4372 if (!rc) {
4373 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4374 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4375 }
4376 }
4377
4378 void
4379 vop_rename_post(void *ap, int rc)
4380 {
4381 struct vop_rename_args *a = ap;
4382 long hint;
4383
4384 if (!rc) {
4385 hint = NOTE_WRITE;
4386 if (a->a_fdvp == a->a_tdvp) {
4387 if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR)
4388 hint |= NOTE_LINK;
4389 VFS_KNOTE_UNLOCKED(a->a_fdvp, hint);
4390 VFS_KNOTE_UNLOCKED(a->a_tdvp, hint);
4391 } else {
4392 hint |= NOTE_EXTEND;
4393 if (a->a_fvp->v_type == VDIR)
4394 hint |= NOTE_LINK;
4395 VFS_KNOTE_UNLOCKED(a->a_fdvp, hint);
4396
4397 if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL &&
4398 a->a_tvp->v_type == VDIR)
4399 hint &= ~NOTE_LINK;
4400 VFS_KNOTE_UNLOCKED(a->a_tdvp, hint);
4401 }
4402
4403 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4404 if (a->a_tvp)
4405 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4406 }
4407 if (a->a_tdvp != a->a_fdvp)
4408 vdrop(a->a_fdvp);
4409 if (a->a_tvp != a->a_fvp)
4410 vdrop(a->a_fvp);
4411 vdrop(a->a_tdvp);
4412 if (a->a_tvp)
4413 vdrop(a->a_tvp);
4414 }
4415
4416 void
4417 vop_rmdir_post(void *ap, int rc)
4418 {
4419 struct vop_rmdir_args *a = ap;
4420
4421 if (!rc) {
4422 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4423 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4424 }
4425 }
4426
4427 void
4428 vop_setattr_post(void *ap, int rc)
4429 {
4430 struct vop_setattr_args *a = ap;
4431
4432 if (!rc)
4433 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4434 }
4435
4436 void
4437 vop_setextattr_post(void *ap, int rc)
4438 {
4439 struct vop_setextattr_args *a = ap;
4440
4441 if (!rc)
4442 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4443 }
4444
4445 void
4446 vop_symlink_post(void *ap, int rc)
4447 {
4448 struct vop_symlink_args *a = ap;
4449
4450 if (!rc)
4451 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4452 }
4453
4454 void
4455 vop_open_post(void *ap, int rc)
4456 {
4457 struct vop_open_args *a = ap;
4458
4459 if (!rc)
4460 VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN);
4461 }
4462
4463 void
4464 vop_close_post(void *ap, int rc)
4465 {
4466 struct vop_close_args *a = ap;
4467
4468 if (!rc && (a->a_cred != NOCRED || /* filter out revokes */
4469 (a->a_vp->v_iflag & VI_DOOMED) == 0)) {
4470 VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ?
4471 NOTE_CLOSE_WRITE : NOTE_CLOSE);
4472 }
4473 }
4474
4475 void
4476 vop_read_post(void *ap, int rc)
4477 {
4478 struct vop_read_args *a = ap;
4479
4480 if (!rc)
4481 VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ);
4482 }
4483
4484 void
4485 vop_readdir_post(void *ap, int rc)
4486 {
4487 struct vop_readdir_args *a = ap;
4488
4489 if (!rc)
4490 VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ);
4491 }
4492
4493 static struct knlist fs_knlist;
4494
4495 static void
4496 vfs_event_init(void *arg)
4497 {
4498 knlist_init_mtx(&fs_knlist, NULL);
4499 }
4500 /* XXX - correct order? */
4501 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4502
4503 void
4504 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4505 {
4506
4507 KNOTE_UNLOCKED(&fs_knlist, event);
4508 }
4509
4510 static int filt_fsattach(struct knote *kn);
4511 static void filt_fsdetach(struct knote *kn);
4512 static int filt_fsevent(struct knote *kn, long hint);
4513
4514 struct filterops fs_filtops = {
4515 .f_isfd = 0,
4516 .f_attach = filt_fsattach,
4517 .f_detach = filt_fsdetach,
4518 .f_event = filt_fsevent
4519 };
4520
4521 static int
4522 filt_fsattach(struct knote *kn)
4523 {
4524
4525 kn->kn_flags |= EV_CLEAR;
4526 knlist_add(&fs_knlist, kn, 0);
4527 return (0);
4528 }
4529
4530 static void
4531 filt_fsdetach(struct knote *kn)
4532 {
4533
4534 knlist_remove(&fs_knlist, kn, 0);
4535 }
4536
4537 static int
4538 filt_fsevent(struct knote *kn, long hint)
4539 {
4540
4541 kn->kn_fflags |= hint;
4542 return (kn->kn_fflags != 0);
4543 }
4544
4545 static int
4546 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4547 {
4548 struct vfsidctl vc;
4549 int error;
4550 struct mount *mp;
4551
4552 error = SYSCTL_IN(req, &vc, sizeof(vc));
4553 if (error)
4554 return (error);
4555 if (vc.vc_vers != VFS_CTL_VERS1)
4556 return (EINVAL);
4557 mp = vfs_getvfs(&vc.vc_fsid);
4558 if (mp == NULL)
4559 return (ENOENT);
4560 /* ensure that a specific sysctl goes to the right filesystem. */
4561 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4562 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4563 vfs_rel(mp);
4564 return (EINVAL);
4565 }
4566 VCTLTOREQ(&vc, req);
4567 error = VFS_SYSCTL(mp, vc.vc_op, req);
4568 vfs_rel(mp);
4569 return (error);
4570 }
4571
4572 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4573 NULL, 0, sysctl_vfs_ctl, "",
4574 "Sysctl by fsid");
4575
4576 /*
4577 * Function to initialize a va_filerev field sensibly.
4578 * XXX: Wouldn't a random number make a lot more sense ??
4579 */
4580 u_quad_t
4581 init_va_filerev(void)
4582 {
4583 struct bintime bt;
4584
4585 getbinuptime(&bt);
4586 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4587 }
4588
4589 static int filt_vfsread(struct knote *kn, long hint);
4590 static int filt_vfswrite(struct knote *kn, long hint);
4591 static int filt_vfsvnode(struct knote *kn, long hint);
4592 static void filt_vfsdetach(struct knote *kn);
4593 static struct filterops vfsread_filtops = {
4594 .f_isfd = 1,
4595 .f_detach = filt_vfsdetach,
4596 .f_event = filt_vfsread
4597 };
4598 static struct filterops vfswrite_filtops = {
4599 .f_isfd = 1,
4600 .f_detach = filt_vfsdetach,
4601 .f_event = filt_vfswrite
4602 };
4603 static struct filterops vfsvnode_filtops = {
4604 .f_isfd = 1,
4605 .f_detach = filt_vfsdetach,
4606 .f_event = filt_vfsvnode
4607 };
4608
4609 static void
4610 vfs_knllock(void *arg)
4611 {
4612 struct vnode *vp = arg;
4613
4614 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4615 }
4616
4617 static void
4618 vfs_knlunlock(void *arg)
4619 {
4620 struct vnode *vp = arg;
4621
4622 VOP_UNLOCK(vp, 0);
4623 }
4624
4625 static void
4626 vfs_knl_assert_locked(void *arg)
4627 {
4628 #ifdef DEBUG_VFS_LOCKS
4629 struct vnode *vp = arg;
4630
4631 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4632 #endif
4633 }
4634
4635 static void
4636 vfs_knl_assert_unlocked(void *arg)
4637 {
4638 #ifdef DEBUG_VFS_LOCKS
4639 struct vnode *vp = arg;
4640
4641 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4642 #endif
4643 }
4644
4645 int
4646 vfs_kqfilter(struct vop_kqfilter_args *ap)
4647 {
4648 struct vnode *vp = ap->a_vp;
4649 struct knote *kn = ap->a_kn;
4650 struct knlist *knl;
4651
4652 switch (kn->kn_filter) {
4653 case EVFILT_READ:
4654 kn->kn_fop = &vfsread_filtops;
4655 break;
4656 case EVFILT_WRITE:
4657 kn->kn_fop = &vfswrite_filtops;
4658 break;
4659 case EVFILT_VNODE:
4660 kn->kn_fop = &vfsvnode_filtops;
4661 break;
4662 default:
4663 return (EINVAL);
4664 }
4665
4666 kn->kn_hook = (caddr_t)vp;
4667
4668 v_addpollinfo(vp);
4669 if (vp->v_pollinfo == NULL)
4670 return (ENOMEM);
4671 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4672 vhold(vp);
4673 knlist_add(knl, kn, 0);
4674
4675 return (0);
4676 }
4677
4678 /*
4679 * Detach knote from vnode
4680 */
4681 static void
4682 filt_vfsdetach(struct knote *kn)
4683 {
4684 struct vnode *vp = (struct vnode *)kn->kn_hook;
4685
4686 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4687 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4688 vdrop(vp);
4689 }
4690
4691 /*ARGSUSED*/
4692 static int
4693 filt_vfsread(struct knote *kn, long hint)
4694 {
4695 struct vnode *vp = (struct vnode *)kn->kn_hook;
4696 struct vattr va;
4697 int res;
4698
4699 /*
4700 * filesystem is gone, so set the EOF flag and schedule
4701 * the knote for deletion.
4702 */
4703 if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) {
4704 VI_LOCK(vp);
4705 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4706 VI_UNLOCK(vp);
4707 return (1);
4708 }
4709
4710 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4711 return (0);
4712
4713 VI_LOCK(vp);
4714 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4715 res = (kn->kn_data != 0);
4716 VI_UNLOCK(vp);
4717 return (res);
4718 }
4719
4720 /*ARGSUSED*/
4721 static int
4722 filt_vfswrite(struct knote *kn, long hint)
4723 {
4724 struct vnode *vp = (struct vnode *)kn->kn_hook;
4725
4726 VI_LOCK(vp);
4727
4728 /*
4729 * filesystem is gone, so set the EOF flag and schedule
4730 * the knote for deletion.
4731 */
4732 if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD))
4733 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4734
4735 kn->kn_data = 0;
4736 VI_UNLOCK(vp);
4737 return (1);
4738 }
4739
4740 static int
4741 filt_vfsvnode(struct knote *kn, long hint)
4742 {
4743 struct vnode *vp = (struct vnode *)kn->kn_hook;
4744 int res;
4745
4746 VI_LOCK(vp);
4747 if (kn->kn_sfflags & hint)
4748 kn->kn_fflags |= hint;
4749 if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) {
4750 kn->kn_flags |= EV_EOF;
4751 VI_UNLOCK(vp);
4752 return (1);
4753 }
4754 res = (kn->kn_fflags != 0);
4755 VI_UNLOCK(vp);
4756 return (res);
4757 }
4758
4759 int
4760 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4761 {
4762 int error;
4763
4764 if (dp->d_reclen > ap->a_uio->uio_resid)
4765 return (ENAMETOOLONG);
4766 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4767 if (error) {
4768 if (ap->a_ncookies != NULL) {
4769 if (ap->a_cookies != NULL)
4770 free(ap->a_cookies, M_TEMP);
4771 ap->a_cookies = NULL;
4772 *ap->a_ncookies = 0;
4773 }
4774 return (error);
4775 }
4776 if (ap->a_ncookies == NULL)
4777 return (0);
4778
4779 KASSERT(ap->a_cookies,
4780 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4781
4782 *ap->a_cookies = realloc(*ap->a_cookies,
4783 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4784 (*ap->a_cookies)[*ap->a_ncookies] = off;
4785 *ap->a_ncookies += 1;
4786 return (0);
4787 }
4788
4789 /*
4790 * Mark for update the access time of the file if the filesystem
4791 * supports VOP_MARKATIME. This functionality is used by execve and
4792 * mmap, so we want to avoid the I/O implied by directly setting
4793 * va_atime for the sake of efficiency.
4794 */
4795 void
4796 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4797 {
4798 struct mount *mp;
4799
4800 mp = vp->v_mount;
4801 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4802 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4803 (void)VOP_MARKATIME(vp);
4804 }
4805
4806 /*
4807 * The purpose of this routine is to remove granularity from accmode_t,
4808 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4809 * VADMIN and VAPPEND.
4810 *
4811 * If it returns 0, the caller is supposed to continue with the usual
4812 * access checks using 'accmode' as modified by this routine. If it
4813 * returns nonzero value, the caller is supposed to return that value
4814 * as errno.
4815 *
4816 * Note that after this routine runs, accmode may be zero.
4817 */
4818 int
4819 vfs_unixify_accmode(accmode_t *accmode)
4820 {
4821 /*
4822 * There is no way to specify explicit "deny" rule using
4823 * file mode or POSIX.1e ACLs.
4824 */
4825 if (*accmode & VEXPLICIT_DENY) {
4826 *accmode = 0;
4827 return (0);
4828 }
4829
4830 /*
4831 * None of these can be translated into usual access bits.
4832 * Also, the common case for NFSv4 ACLs is to not contain
4833 * either of these bits. Caller should check for VWRITE
4834 * on the containing directory instead.
4835 */
4836 if (*accmode & (VDELETE_CHILD | VDELETE))
4837 return (EPERM);
4838
4839 if (*accmode & VADMIN_PERMS) {
4840 *accmode &= ~VADMIN_PERMS;
4841 *accmode |= VADMIN;
4842 }
4843
4844 /*
4845 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4846 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4847 */
4848 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4849
4850 return (0);
4851 }
4852
4853 /*
4854 * These are helper functions for filesystems to traverse all
4855 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4856 *
4857 * This interface replaces MNT_VNODE_FOREACH.
4858 */
4859
4860 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4861
4862 struct vnode *
4863 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4864 {
4865 struct vnode *vp;
4866
4867 if (should_yield())
4868 kern_yield(PRI_USER);
4869 MNT_ILOCK(mp);
4870 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4871 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4872 while (vp != NULL && (vp->v_type == VMARKER ||
4873 (vp->v_iflag & VI_DOOMED) != 0))
4874 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4875
4876 /* Check if we are done */
4877 if (vp == NULL) {
4878 __mnt_vnode_markerfree_all(mvp, mp);
4879 /* MNT_IUNLOCK(mp); -- done in above function */
4880 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4881 return (NULL);
4882 }
4883 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4884 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4885 VI_LOCK(vp);
4886 MNT_IUNLOCK(mp);
4887 return (vp);
4888 }
4889
4890 struct vnode *
4891 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4892 {
4893 struct vnode *vp;
4894
4895 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4896 MNT_ILOCK(mp);
4897 MNT_REF(mp);
4898 (*mvp)->v_type = VMARKER;
4899
4900 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4901 while (vp != NULL && (vp->v_type == VMARKER ||
4902 (vp->v_iflag & VI_DOOMED) != 0))
4903 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4904
4905 /* Check if we are done */
4906 if (vp == NULL) {
4907 MNT_REL(mp);
4908 MNT_IUNLOCK(mp);
4909 free(*mvp, M_VNODE_MARKER);
4910 *mvp = NULL;
4911 return (NULL);
4912 }
4913 (*mvp)->v_mount = mp;
4914 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4915 VI_LOCK(vp);
4916 MNT_IUNLOCK(mp);
4917 return (vp);
4918 }
4919
4920
4921 void
4922 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4923 {
4924
4925 if (*mvp == NULL) {
4926 MNT_IUNLOCK(mp);
4927 return;
4928 }
4929
4930 mtx_assert(MNT_MTX(mp), MA_OWNED);
4931
4932 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4933 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4934 MNT_REL(mp);
4935 MNT_IUNLOCK(mp);
4936 free(*mvp, M_VNODE_MARKER);
4937 *mvp = NULL;
4938 }
4939
4940 /*
4941 * These are helper functions for filesystems to traverse their
4942 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4943 */
4944 static void
4945 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4946 {
4947
4948 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4949
4950 MNT_ILOCK(mp);
4951 MNT_REL(mp);
4952 MNT_IUNLOCK(mp);
4953 free(*mvp, M_VNODE_MARKER);
4954 *mvp = NULL;
4955 }
4956
4957 static struct vnode *
4958 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4959 {
4960 struct vnode *vp, *nvp;
4961
4962 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4963 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4964 restart:
4965 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4966 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4967 while (vp != NULL) {
4968 if (vp->v_type == VMARKER) {
4969 vp = TAILQ_NEXT(vp, v_actfreelist);
4970 continue;
4971 }
4972 if (!VI_TRYLOCK(vp)) {
4973 if (mp_ncpus == 1 || should_yield()) {
4974 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4975 mtx_unlock(&vnode_free_list_mtx);
4976 pause("vnacti", 1);
4977 mtx_lock(&vnode_free_list_mtx);
4978 goto restart;
4979 }
4980 continue;
4981 }
4982 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4983 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4984 ("alien vnode on the active list %p %p", vp, mp));
4985 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4986 break;
4987 nvp = TAILQ_NEXT(vp, v_actfreelist);
4988 VI_UNLOCK(vp);
4989 vp = nvp;
4990 }
4991
4992 /* Check if we are done */
4993 if (vp == NULL) {
4994 mtx_unlock(&vnode_free_list_mtx);
4995 mnt_vnode_markerfree_active(mvp, mp);
4996 return (NULL);
4997 }
4998 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4999 mtx_unlock(&vnode_free_list_mtx);
5000 ASSERT_VI_LOCKED(vp, "active iter");
5001 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
5002 return (vp);
5003 }
5004
5005 struct vnode *
5006 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
5007 {
5008
5009 if (should_yield())
5010 kern_yield(PRI_USER);
5011 mtx_lock(&vnode_free_list_mtx);
5012 return (mnt_vnode_next_active(mvp, mp));
5013 }
5014
5015 struct vnode *
5016 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
5017 {
5018 struct vnode *vp;
5019
5020 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
5021 MNT_ILOCK(mp);
5022 MNT_REF(mp);
5023 MNT_IUNLOCK(mp);
5024 (*mvp)->v_type = VMARKER;
5025 (*mvp)->v_mount = mp;
5026
5027 mtx_lock(&vnode_free_list_mtx);
5028 vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
5029 if (vp == NULL) {
5030 mtx_unlock(&vnode_free_list_mtx);
5031 mnt_vnode_markerfree_active(mvp, mp);
5032 return (NULL);
5033 }
5034 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
5035 return (mnt_vnode_next_active(mvp, mp));
5036 }
5037
5038 void
5039 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
5040 {
5041
5042 if (*mvp == NULL)
5043 return;
5044
5045 mtx_lock(&vnode_free_list_mtx);
5046 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
5047 mtx_unlock(&vnode_free_list_mtx);
5048 mnt_vnode_markerfree_active(mvp, mp);
5049 }
Cache object: 3bd99849e49e0d992a314b028619db40
|