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/7.3/sys/kern/vfs_subr.c 202814 2010-01-22 17:02:07Z jhb $");
43
44 #include "opt_ddb.h"
45 #include "opt_mac.h"
46
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/dirent.h>
53 #include <sys/event.h>
54 #include <sys/eventhandler.h>
55 #include <sys/extattr.h>
56 #include <sys/file.h>
57 #include <sys/fcntl.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/lockf.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/namei.h>
66 #include <sys/priv.h>
67 #include <sys/reboot.h>
68 #include <sys/sleepqueue.h>
69 #include <sys/stat.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/vmmeter.h>
73 #include <sys/vnode.h>
74
75 #include <machine/stdarg.h>
76
77 #include <security/mac/mac_framework.h>
78
79 #include <vm/vm.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_kern.h>
86 #include <vm/uma.h>
87
88 #ifdef DDB
89 #include <ddb/ddb.h>
90 #endif
91
92 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
93
94 static void delmntque(struct vnode *vp);
95 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
96 int slpflag, int slptimeo);
97 static void syncer_shutdown(void *arg, int howto);
98 static int vtryrecycle(struct vnode *vp);
99 static void vbusy(struct vnode *vp);
100 static void vinactive(struct vnode *, struct thread *);
101 static void v_incr_usecount(struct vnode *);
102 static void v_decr_usecount(struct vnode *);
103 static void v_decr_useonly(struct vnode *);
104 static void v_upgrade_usecount(struct vnode *);
105 static void vfree(struct vnode *);
106 static void vnlru_free(int);
107 static void vgonel(struct vnode *);
108 static void vfs_knllock(void *arg);
109 static void vfs_knlunlock(void *arg);
110 static void destroy_vpollinfo(struct vpollinfo *vi);
111
112 /*
113 * Enable Giant pushdown based on whether or not the vm is mpsafe in this
114 * build. Without mpsafevm the buffer cache can not run Giant free.
115 */
116 int mpsafe_vfs = 1;
117 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
118 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
119 "MPSAFE VFS");
120
121 /*
122 * Number of vnodes in existence. Increased whenever getnewvnode()
123 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
124 * vnode.
125 */
126 static unsigned long numvnodes;
127
128 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
129
130 /*
131 * Conversion tables for conversion from vnode types to inode formats
132 * and back.
133 */
134 enum vtype iftovt_tab[16] = {
135 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
137 };
138 int vttoif_tab[10] = {
139 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
141 };
142
143 /*
144 * List of vnodes that are ready for recycling.
145 */
146 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
147
148 /*
149 * Free vnode target. Free vnodes may simply be files which have been stat'd
150 * but not read. This is somewhat common, and a small cache of such files
151 * should be kept to avoid recreation costs.
152 */
153 static u_long wantfreevnodes;
154 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
155 /* Number of vnodes in the free list. */
156 static u_long freevnodes;
157 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
158
159 /*
160 * Various variables used for debugging the new implementation of
161 * reassignbuf().
162 * XXX these are probably of (very) limited utility now.
163 */
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
166
167 /*
168 * Cache for the mount type id assigned to NFS. This is used for
169 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
170 */
171 int nfs_mount_type = -1;
172
173 /* To keep more than one thread at a time from running vfs_getnewfsid */
174 static struct mtx mntid_mtx;
175
176 /*
177 * Lock for any access to the following:
178 * vnode_free_list
179 * numvnodes
180 * freevnodes
181 */
182 static struct mtx vnode_free_list_mtx;
183
184 /* Publicly exported FS */
185 struct nfs_public nfs_pub;
186
187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
188 static uma_zone_t vnode_zone;
189 static uma_zone_t vnodepoll_zone;
190
191 /* Set to 1 to print out reclaim of active vnodes */
192 int prtactive;
193
194 /*
195 * The workitem queue.
196 *
197 * It is useful to delay writes of file data and filesystem metadata
198 * for tens of seconds so that quickly created and deleted files need
199 * not waste disk bandwidth being created and removed. To realize this,
200 * we append vnodes to a "workitem" queue. When running with a soft
201 * updates implementation, most pending metadata dependencies should
202 * not wait for more than a few seconds. Thus, mounted on block devices
203 * are delayed only about a half the time that file data is delayed.
204 * Similarly, directory updates are more critical, so are only delayed
205 * about a third the time that file data is delayed. Thus, there are
206 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207 * one each second (driven off the filesystem syncer process). The
208 * syncer_delayno variable indicates the next queue that is to be processed.
209 * Items that need to be processed soon are placed in this queue:
210 *
211 * syncer_workitem_pending[syncer_delayno]
212 *
213 * A delay of fifteen seconds is done by placing the request fifteen
214 * entries later in the queue:
215 *
216 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
217 *
218 */
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending;
223 /*
224 * The sync_mtx protects:
225 * bo->bo_synclist
226 * sync_vnode_count
227 * syncer_delayno
228 * syncer_state
229 * syncer_workitem_pending
230 * syncer_worklist_len
231 * rushjob
232 */
233 static struct mtx sync_mtx;
234
235 #define SYNCER_MAXDELAY 32
236 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
237 static int syncdelay = 30; /* max time to delay syncing data */
238 static int filedelay = 30; /* time to delay syncing files */
239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
240 static int dirdelay = 29; /* time to delay syncing directories */
241 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
242 static int metadelay = 28; /* time to delay syncing metadata */
243 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
244 static int rushjob; /* number of slots to run ASAP */
245 static int stat_rush_requests; /* number of times I/O speeded up */
246 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
247
248 /*
249 * When shutting down the syncer, run it at four times normal speed.
250 */
251 #define SYNCER_SHUTDOWN_SPEEDUP 4
252 static int sync_vnode_count;
253 static int syncer_worklist_len;
254 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
255 syncer_state;
256
257 /*
258 * Number of vnodes we want to exist at any one time. This is mostly used
259 * to size hash tables in vnode-related code. It is normally not used in
260 * getnewvnode(), as wantfreevnodes is normally nonzero.)
261 *
262 * XXX desiredvnodes is historical cruft and should not exist.
263 */
264 int desiredvnodes;
265 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
266 &desiredvnodes, 0, "Maximum number of vnodes");
267 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
268 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
269 static int vnlru_nowhere;
270 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
271 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
272
273 /*
274 * Macros to control when a vnode is freed and recycled. All require
275 * the vnode interlock.
276 */
277 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
278 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
280
281
282 /*
283 * Initialize the vnode management data structures.
284 */
285 #ifndef MAXVNODES_MAX
286 #define MAXVNODES_MAX 100000
287 #endif
288 static void
289 vntblinit(void *dummy __unused)
290 {
291
292 /*
293 * Desiredvnodes is a function of the physical memory size and
294 * the kernel's heap size. Specifically, desiredvnodes scales
295 * in proportion to the physical memory size until two fifths
296 * of the kernel's heap size is consumed by vnodes and vm
297 * objects.
298 */
299 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
300 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
301 if (desiredvnodes > MAXVNODES_MAX) {
302 if (bootverbose)
303 printf("Reducing kern.maxvnodes %d -> %d\n",
304 desiredvnodes, MAXVNODES_MAX);
305 desiredvnodes = MAXVNODES_MAX;
306 }
307 wantfreevnodes = desiredvnodes / 4;
308 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
309 TAILQ_INIT(&vnode_free_list);
310 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
311 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
312 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
313 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
314 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
315 /*
316 * Initialize the filesystem syncer.
317 */
318 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
319 &syncer_mask);
320 syncer_maxdelay = syncer_mask + 1;
321 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
322 }
323 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
324
325
326 /*
327 * Mark a mount point as busy. Used to synchronize access and to delay
328 * unmounting. Interlock is not released on failure.
329 */
330 int
331 vfs_busy(struct mount *mp, int flags, struct mtx *interlkp,
332 struct thread *td)
333 {
334 int lkflags;
335
336 MNT_ILOCK(mp);
337 MNT_REF(mp);
338 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
339 if (flags & LK_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
340 MNT_REL(mp);
341 MNT_IUNLOCK(mp);
342 return (ENOENT);
343 }
344 if (interlkp)
345 mtx_unlock(interlkp);
346 mp->mnt_kern_flag |= MNTK_MWAIT;
347 /*
348 * Since all busy locks are shared except the exclusive
349 * lock granted when unmounting, the only place that a
350 * wakeup needs to be done is at the release of the
351 * exclusive lock at the end of dounmount.
352 */
353 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
354 if (interlkp)
355 mtx_lock(interlkp);
356 }
357 if (interlkp)
358 mtx_unlock(interlkp);
359 lkflags = LK_SHARED | LK_INTERLOCK | LK_NOWAIT;
360 if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp), td))
361 panic("vfs_busy: unexpected lock failure");
362 return (0);
363 }
364
365 /*
366 * Free a busy filesystem.
367 */
368 void
369 vfs_unbusy(struct mount *mp, struct thread *td)
370 {
371
372 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
373 vfs_rel(mp);
374 }
375
376 /*
377 * Lookup a mount point by filesystem identifier.
378 */
379 struct mount *
380 vfs_getvfs(fsid_t *fsid)
381 {
382 struct mount *mp;
383
384 mtx_lock(&mountlist_mtx);
385 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
386 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
387 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
388 vfs_ref(mp);
389 mtx_unlock(&mountlist_mtx);
390 return (mp);
391 }
392 }
393 mtx_unlock(&mountlist_mtx);
394 return ((struct mount *) 0);
395 }
396
397 /*
398 * Check if a user can access privileged mount options.
399 */
400 int
401 vfs_suser(struct mount *mp, struct thread *td)
402 {
403 int error;
404
405 /*
406 * If the thread is jailed, but this is not a jail-friendly file
407 * system, deny immediately.
408 */
409 if (jailed(td->td_ucred) && !(mp->mnt_vfc->vfc_flags & VFCF_JAIL))
410 return (EPERM);
411
412 /*
413 * If the file system was mounted outside a jail and a jailed thread
414 * tries to access it, deny immediately.
415 */
416 if (!jailed(mp->mnt_cred) && jailed(td->td_ucred))
417 return (EPERM);
418
419 /*
420 * If the file system was mounted inside different jail that the jail of
421 * the calling thread, deny immediately.
422 */
423 if (jailed(mp->mnt_cred) && jailed(td->td_ucred) &&
424 mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) {
425 return (EPERM);
426 }
427
428 if ((mp->mnt_flag & MNT_USER) == 0 ||
429 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
430 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
431 return (error);
432 }
433 return (0);
434 }
435
436 /*
437 * Get a new unique fsid. Try to make its val[0] unique, since this value
438 * will be used to create fake device numbers for stat(). Also try (but
439 * not so hard) make its val[0] unique mod 2^16, since some emulators only
440 * support 16-bit device numbers. We end up with unique val[0]'s for the
441 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
442 *
443 * Keep in mind that several mounts may be running in parallel. Starting
444 * the search one past where the previous search terminated is both a
445 * micro-optimization and a defense against returning the same fsid to
446 * different mounts.
447 */
448 void
449 vfs_getnewfsid(struct mount *mp)
450 {
451 static u_int16_t mntid_base;
452 struct mount *nmp;
453 fsid_t tfsid;
454 int mtype;
455
456 mtx_lock(&mntid_mtx);
457 mtype = mp->mnt_vfc->vfc_typenum;
458 tfsid.val[1] = mtype;
459 mtype = (mtype & 0xFF) << 24;
460 for (;;) {
461 tfsid.val[0] = makedev(255,
462 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
463 mntid_base++;
464 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
465 break;
466 vfs_rel(nmp);
467 }
468 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
469 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
470 mtx_unlock(&mntid_mtx);
471 }
472
473 /*
474 * Knob to control the precision of file timestamps:
475 *
476 * 0 = seconds only; nanoseconds zeroed.
477 * 1 = seconds and nanoseconds, accurate within 1/HZ.
478 * 2 = seconds and nanoseconds, truncated to microseconds.
479 * >=3 = seconds and nanoseconds, maximum precision.
480 */
481 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
482
483 static int timestamp_precision = TSP_SEC;
484 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
485 ×tamp_precision, 0, "");
486
487 /*
488 * Get a current timestamp.
489 */
490 void
491 vfs_timestamp(struct timespec *tsp)
492 {
493 struct timeval tv;
494
495 switch (timestamp_precision) {
496 case TSP_SEC:
497 tsp->tv_sec = time_second;
498 tsp->tv_nsec = 0;
499 break;
500 case TSP_HZ:
501 getnanotime(tsp);
502 break;
503 case TSP_USEC:
504 microtime(&tv);
505 TIMEVAL_TO_TIMESPEC(&tv, tsp);
506 break;
507 case TSP_NSEC:
508 default:
509 nanotime(tsp);
510 break;
511 }
512 }
513
514 /*
515 * Set vnode attributes to VNOVAL
516 */
517 void
518 vattr_null(struct vattr *vap)
519 {
520
521 vap->va_type = VNON;
522 vap->va_size = VNOVAL;
523 vap->va_bytes = VNOVAL;
524 vap->va_mode = VNOVAL;
525 vap->va_nlink = VNOVAL;
526 vap->va_uid = VNOVAL;
527 vap->va_gid = VNOVAL;
528 vap->va_fsid = VNOVAL;
529 vap->va_fileid = VNOVAL;
530 vap->va_blocksize = VNOVAL;
531 vap->va_rdev = VNOVAL;
532 vap->va_atime.tv_sec = VNOVAL;
533 vap->va_atime.tv_nsec = VNOVAL;
534 vap->va_mtime.tv_sec = VNOVAL;
535 vap->va_mtime.tv_nsec = VNOVAL;
536 vap->va_ctime.tv_sec = VNOVAL;
537 vap->va_ctime.tv_nsec = VNOVAL;
538 vap->va_birthtime.tv_sec = VNOVAL;
539 vap->va_birthtime.tv_nsec = VNOVAL;
540 vap->va_flags = VNOVAL;
541 vap->va_gen = VNOVAL;
542 vap->va_vaflags = 0;
543 }
544
545 /*
546 * This routine is called when we have too many vnodes. It attempts
547 * to free <count> vnodes and will potentially free vnodes that still
548 * have VM backing store (VM backing store is typically the cause
549 * of a vnode blowout so we want to do this). Therefore, this operation
550 * is not considered cheap.
551 *
552 * A number of conditions may prevent a vnode from being reclaimed.
553 * the buffer cache may have references on the vnode, a directory
554 * vnode may still have references due to the namei cache representing
555 * underlying files, or the vnode may be in active use. It is not
556 * desireable to reuse such vnodes. These conditions may cause the
557 * number of vnodes to reach some minimum value regardless of what
558 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
559 */
560 static int
561 vlrureclaim(struct mount *mp)
562 {
563 struct thread *td;
564 struct vnode *vp;
565 int done;
566 int trigger;
567 int usevnodes;
568 int count;
569
570 /*
571 * Calculate the trigger point, don't allow user
572 * screwups to blow us up. This prevents us from
573 * recycling vnodes with lots of resident pages. We
574 * aren't trying to free memory, we are trying to
575 * free vnodes.
576 */
577 usevnodes = desiredvnodes;
578 if (usevnodes <= 0)
579 usevnodes = 1;
580 trigger = cnt.v_page_count * 2 / usevnodes;
581 done = 0;
582 td = curthread;
583 vn_start_write(NULL, &mp, V_WAIT);
584 MNT_ILOCK(mp);
585 count = mp->mnt_nvnodelistsize / 10 + 1;
586 while (count != 0) {
587 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
588 while (vp != NULL && vp->v_type == VMARKER)
589 vp = TAILQ_NEXT(vp, v_nmntvnodes);
590 if (vp == NULL)
591 break;
592 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
593 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
594 --count;
595 if (!VI_TRYLOCK(vp))
596 goto next_iter;
597 /*
598 * If it's been deconstructed already, it's still
599 * referenced, or it exceeds the trigger, skip it.
600 */
601 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
602 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
603 vp->v_object->resident_page_count > trigger)) {
604 VI_UNLOCK(vp);
605 goto next_iter;
606 }
607 MNT_IUNLOCK(mp);
608 vholdl(vp);
609 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT, td)) {
610 vdrop(vp);
611 goto next_iter_mntunlocked;
612 }
613 VI_LOCK(vp);
614 /*
615 * v_usecount may have been bumped after VOP_LOCK() dropped
616 * the vnode interlock and before it was locked again.
617 *
618 * It is not necessary to recheck VI_DOOMED because it can
619 * only be set by another thread that holds both the vnode
620 * lock and vnode interlock. If another thread has the
621 * vnode lock before we get to VOP_LOCK() and obtains the
622 * vnode interlock after VOP_LOCK() drops the vnode
623 * interlock, the other thread will be unable to drop the
624 * vnode lock before our VOP_LOCK() call fails.
625 */
626 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
627 (vp->v_object != NULL &&
628 vp->v_object->resident_page_count > trigger)) {
629 VOP_UNLOCK(vp, LK_INTERLOCK, td);
630 goto next_iter_mntunlocked;
631 }
632 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
633 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
634 vgonel(vp);
635 VOP_UNLOCK(vp, 0, td);
636 vdropl(vp);
637 done++;
638 next_iter_mntunlocked:
639 if ((count % 256) != 0)
640 goto relock_mnt;
641 goto yield;
642 next_iter:
643 if ((count % 256) != 0)
644 continue;
645 MNT_IUNLOCK(mp);
646 yield:
647 uio_yield();
648 relock_mnt:
649 MNT_ILOCK(mp);
650 }
651 MNT_IUNLOCK(mp);
652 vn_finished_write(mp);
653 return done;
654 }
655
656 /*
657 * Attempt to keep the free list at wantfreevnodes length.
658 */
659 static void
660 vnlru_free(int count)
661 {
662 struct vnode *vp;
663 int vfslocked;
664
665 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
666 for (; count > 0; count--) {
667 vp = TAILQ_FIRST(&vnode_free_list);
668 /*
669 * The list can be modified while the free_list_mtx
670 * has been dropped and vp could be NULL here.
671 */
672 if (!vp)
673 break;
674 VNASSERT(vp->v_op != NULL, vp,
675 ("vnlru_free: vnode already reclaimed."));
676 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
677 /*
678 * Don't recycle if we can't get the interlock.
679 */
680 if (!VI_TRYLOCK(vp)) {
681 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
682 continue;
683 }
684 VNASSERT(VCANRECYCLE(vp), vp,
685 ("vp inconsistent on freelist"));
686 freevnodes--;
687 vp->v_iflag &= ~VI_FREE;
688 vholdl(vp);
689 mtx_unlock(&vnode_free_list_mtx);
690 VI_UNLOCK(vp);
691 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
692 vtryrecycle(vp);
693 VFS_UNLOCK_GIANT(vfslocked);
694 /*
695 * If the recycled succeeded this vdrop will actually free
696 * the vnode. If not it will simply place it back on
697 * the free list.
698 */
699 vdrop(vp);
700 mtx_lock(&vnode_free_list_mtx);
701 }
702 }
703 /*
704 * Attempt to recycle vnodes in a context that is always safe to block.
705 * Calling vlrurecycle() from the bowels of filesystem code has some
706 * interesting deadlock problems.
707 */
708 static struct proc *vnlruproc;
709 static int vnlruproc_sig;
710
711 static void
712 vnlru_proc(void)
713 {
714 struct mount *mp, *nmp;
715 int done, vfslocked;
716 struct proc *p = vnlruproc;
717 struct thread *td = curthread;
718
719 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
720 SHUTDOWN_PRI_FIRST);
721
722 for (;;) {
723 kthread_suspend_check(p);
724 mtx_lock(&vnode_free_list_mtx);
725 if (freevnodes > wantfreevnodes)
726 vnlru_free(freevnodes - wantfreevnodes);
727 if (numvnodes <= desiredvnodes * 9 / 10) {
728 vnlruproc_sig = 0;
729 wakeup(&vnlruproc_sig);
730 msleep(vnlruproc, &vnode_free_list_mtx,
731 PVFS|PDROP, "vlruwt", hz);
732 continue;
733 }
734 mtx_unlock(&vnode_free_list_mtx);
735 done = 0;
736 mtx_lock(&mountlist_mtx);
737 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
738 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
739 nmp = TAILQ_NEXT(mp, mnt_list);
740 continue;
741 }
742 vfslocked = VFS_LOCK_GIANT(mp);
743 done += vlrureclaim(mp);
744 VFS_UNLOCK_GIANT(vfslocked);
745 mtx_lock(&mountlist_mtx);
746 nmp = TAILQ_NEXT(mp, mnt_list);
747 vfs_unbusy(mp, td);
748 }
749 mtx_unlock(&mountlist_mtx);
750 if (done == 0) {
751 EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
752 #if 0
753 /* These messages are temporary debugging aids */
754 if (vnlru_nowhere < 5)
755 printf("vnlru process getting nowhere..\n");
756 else if (vnlru_nowhere == 5)
757 printf("vnlru process messages stopped.\n");
758 #endif
759 vnlru_nowhere++;
760 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
761 } else
762 uio_yield();
763 }
764 }
765
766 static struct kproc_desc vnlru_kp = {
767 "vnlru",
768 vnlru_proc,
769 &vnlruproc
770 };
771 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
772 &vnlru_kp);
773
774 /*
775 * Routines having to do with the management of the vnode table.
776 */
777
778 void
779 vdestroy(struct vnode *vp)
780 {
781 struct bufobj *bo;
782
783 CTR1(KTR_VFS, "vdestroy vp %p", vp);
784 mtx_lock(&vnode_free_list_mtx);
785 numvnodes--;
786 mtx_unlock(&vnode_free_list_mtx);
787 bo = &vp->v_bufobj;
788 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
789 ("cleaned vnode still on the free list."));
790 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
791 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
792 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
793 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
794 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
795 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
796 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
797 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
798 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
799 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
800 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
801 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
802 VI_UNLOCK(vp);
803 #ifdef MAC
804 mac_destroy_vnode(vp);
805 #endif
806 if (vp->v_pollinfo != NULL)
807 destroy_vpollinfo(vp->v_pollinfo);
808 #ifdef INVARIANTS
809 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
810 vp->v_op = NULL;
811 #endif
812 lockdestroy(vp->v_vnlock);
813 mtx_destroy(&vp->v_interlock);
814 uma_zfree(vnode_zone, vp);
815 }
816
817 /*
818 * Try to recycle a freed vnode. We abort if anyone picks up a reference
819 * before we actually vgone(). This function must be called with the vnode
820 * held to prevent the vnode from being returned to the free list midway
821 * through vgone().
822 */
823 static int
824 vtryrecycle(struct vnode *vp)
825 {
826 struct thread *td = curthread;
827 struct mount *vnmp;
828
829 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
830 VNASSERT(vp->v_holdcnt, vp,
831 ("vtryrecycle: Recycling vp %p without a reference.", vp));
832 /*
833 * This vnode may found and locked via some other list, if so we
834 * can't recycle it yet.
835 */
836 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
837 return (EWOULDBLOCK);
838 /*
839 * Don't recycle if its filesystem is being suspended.
840 */
841 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
842 VOP_UNLOCK(vp, 0, td);
843 return (EBUSY);
844 }
845 /*
846 * If we got this far, we need to acquire the interlock and see if
847 * anyone picked up this vnode from another list. If not, we will
848 * mark it with DOOMED via vgonel() so that anyone who does find it
849 * will skip over it.
850 */
851 VI_LOCK(vp);
852 if (vp->v_usecount) {
853 VOP_UNLOCK(vp, LK_INTERLOCK, td);
854 vn_finished_write(vnmp);
855 return (EBUSY);
856 }
857 if ((vp->v_iflag & VI_DOOMED) == 0)
858 vgonel(vp);
859 VOP_UNLOCK(vp, LK_INTERLOCK, td);
860 vn_finished_write(vnmp);
861 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
862 return (0);
863 }
864
865 /*
866 * Return the next vnode from the free list.
867 */
868 int
869 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
870 struct vnode **vpp)
871 {
872 struct vnode *vp = NULL;
873 struct bufobj *bo;
874
875 mtx_lock(&vnode_free_list_mtx);
876 /*
877 * Lend our context to reclaim vnodes if they've exceeded the max.
878 */
879 if (freevnodes > wantfreevnodes)
880 vnlru_free(1);
881 /*
882 * Wait for available vnodes.
883 */
884 if (numvnodes > desiredvnodes) {
885 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
886 /*
887 * File system is beeing suspended, we cannot risk a
888 * deadlock here, so allocate new vnode anyway.
889 */
890 if (freevnodes > wantfreevnodes)
891 vnlru_free(freevnodes - wantfreevnodes);
892 goto alloc;
893 }
894 if (vnlruproc_sig == 0) {
895 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
896 wakeup(vnlruproc);
897 }
898 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
899 "vlruwk", hz);
900 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
901 if (numvnodes > desiredvnodes) {
902 mtx_unlock(&vnode_free_list_mtx);
903 return (ENFILE);
904 }
905 #endif
906 }
907 alloc:
908 numvnodes++;
909 mtx_unlock(&vnode_free_list_mtx);
910 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
911 /*
912 * Setup locks.
913 */
914 vp->v_vnlock = &vp->v_lock;
915 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
916 /*
917 * By default, don't allow shared locks unless filesystems
918 * opt-in.
919 */
920 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
921 /*
922 * Initialize bufobj.
923 */
924 bo = &vp->v_bufobj;
925 bo->__bo_vnode = vp;
926 bo->bo_mtx = &vp->v_interlock;
927 bo->bo_ops = &buf_ops_bio;
928 bo->bo_private = vp;
929 TAILQ_INIT(&bo->bo_clean.bv_hd);
930 TAILQ_INIT(&bo->bo_dirty.bv_hd);
931 /*
932 * Initialize namecache.
933 */
934 LIST_INIT(&vp->v_cache_src);
935 TAILQ_INIT(&vp->v_cache_dst);
936 /*
937 * Finalize various vnode identity bits.
938 */
939 vp->v_type = VNON;
940 vp->v_tag = tag;
941 vp->v_op = vops;
942 v_incr_usecount(vp);
943 vp->v_data = 0;
944 #ifdef MAC
945 mac_init_vnode(vp);
946 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
947 mac_associate_vnode_singlelabel(mp, vp);
948 else if (mp == NULL && vops != &dead_vnodeops)
949 printf("NULL mp in getnewvnode()\n");
950 #endif
951 if (mp != NULL) {
952 bo->bo_bsize = mp->mnt_stat.f_iosize;
953 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
954 vp->v_vflag |= VV_NOKNOTE;
955 }
956
957 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
958 *vpp = vp;
959 return (0);
960 }
961
962 /*
963 * Delete from old mount point vnode list, if on one.
964 */
965 static void
966 delmntque(struct vnode *vp)
967 {
968 struct mount *mp;
969
970 mp = vp->v_mount;
971 if (mp == NULL)
972 return;
973 MNT_ILOCK(mp);
974 vp->v_mount = NULL;
975 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
976 ("bad mount point vnode list size"));
977 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
978 mp->mnt_nvnodelistsize--;
979 MNT_REL(mp);
980 MNT_IUNLOCK(mp);
981 }
982
983 static void
984 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
985 {
986 struct thread *td;
987
988 td = curthread; /* XXX ? */
989 vp->v_data = NULL;
990 vp->v_op = &dead_vnodeops;
991 /* XXX non mp-safe fs may still call insmntque with vnode
992 unlocked */
993 if (!VOP_ISLOCKED(vp, td))
994 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
995 vgone(vp);
996 vput(vp);
997 }
998
999 /*
1000 * Insert into list of vnodes for the new mount point, if available.
1001 */
1002 int
1003 insmntque1(struct vnode *vp, struct mount *mp,
1004 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1005 {
1006 int locked;
1007
1008 KASSERT(vp->v_mount == NULL,
1009 ("insmntque: vnode already on per mount vnode list"));
1010 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1011 #if 0
1012 #ifdef DEBUG_VFS_LOCKS
1013 if (!VFS_NEEDSGIANT(mp))
1014 ASSERT_VOP_ELOCKED(vp,
1015 "insmntque: mp-safe fs and non-locked vp");
1016 #endif
1017 #endif
1018 MNT_ILOCK(mp);
1019 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1020 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1021 mp->mnt_nvnodelistsize == 0)) {
1022 locked = VOP_ISLOCKED(vp, curthread);
1023 if (!locked || (locked == LK_EXCLUSIVE &&
1024 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1025 MNT_IUNLOCK(mp);
1026 if (dtr != NULL)
1027 dtr(vp, dtr_arg);
1028 return (EBUSY);
1029 }
1030 }
1031 vp->v_mount = mp;
1032 MNT_REF(mp);
1033 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1034 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1035 ("neg mount point vnode list size"));
1036 mp->mnt_nvnodelistsize++;
1037 MNT_IUNLOCK(mp);
1038 return (0);
1039 }
1040
1041 int
1042 insmntque(struct vnode *vp, struct mount *mp)
1043 {
1044
1045 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1046 }
1047
1048 /*
1049 * Flush out and invalidate all buffers associated with a bufobj
1050 * Called with the underlying object locked.
1051 */
1052 int
1053 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag,
1054 int slptimeo)
1055 {
1056 int error;
1057
1058 BO_LOCK(bo);
1059 if (flags & V_SAVE) {
1060 error = bufobj_wwait(bo, slpflag, slptimeo);
1061 if (error) {
1062 BO_UNLOCK(bo);
1063 return (error);
1064 }
1065 if (bo->bo_dirty.bv_cnt > 0) {
1066 BO_UNLOCK(bo);
1067 if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1068 return (error);
1069 /*
1070 * XXX We could save a lock/unlock if this was only
1071 * enabled under INVARIANTS
1072 */
1073 BO_LOCK(bo);
1074 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1075 panic("vinvalbuf: dirty bufs");
1076 }
1077 }
1078 /*
1079 * If you alter this loop please notice that interlock is dropped and
1080 * reacquired in flushbuflist. Special care is needed to ensure that
1081 * no race conditions occur from this.
1082 */
1083 do {
1084 error = flushbuflist(&bo->bo_clean,
1085 flags, bo, slpflag, slptimeo);
1086 if (error == 0)
1087 error = flushbuflist(&bo->bo_dirty,
1088 flags, bo, slpflag, slptimeo);
1089 if (error != 0 && error != EAGAIN) {
1090 BO_UNLOCK(bo);
1091 return (error);
1092 }
1093 } while (error != 0);
1094
1095 /*
1096 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1097 * have write I/O in-progress but if there is a VM object then the
1098 * VM object can also have read-I/O in-progress.
1099 */
1100 do {
1101 bufobj_wwait(bo, 0, 0);
1102 BO_UNLOCK(bo);
1103 if (bo->bo_object != NULL) {
1104 VM_OBJECT_LOCK(bo->bo_object);
1105 vm_object_pip_wait(bo->bo_object, "bovlbx");
1106 VM_OBJECT_UNLOCK(bo->bo_object);
1107 }
1108 BO_LOCK(bo);
1109 } while (bo->bo_numoutput > 0);
1110 BO_UNLOCK(bo);
1111
1112 /*
1113 * Destroy the copy in the VM cache, too.
1114 */
1115 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1116 VM_OBJECT_LOCK(bo->bo_object);
1117 vm_object_page_remove(bo->bo_object, 0, 0,
1118 (flags & V_SAVE) ? TRUE : FALSE);
1119 VM_OBJECT_UNLOCK(bo->bo_object);
1120 }
1121
1122 #ifdef INVARIANTS
1123 BO_LOCK(bo);
1124 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1125 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1126 panic("vinvalbuf: flush failed");
1127 BO_UNLOCK(bo);
1128 #endif
1129 return (0);
1130 }
1131
1132 /*
1133 * Flush out and invalidate all buffers associated with a vnode.
1134 * Called with the underlying object locked.
1135 */
1136 int
1137 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag,
1138 int slptimeo)
1139 {
1140
1141 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1142 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1143 return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1144 }
1145
1146 /*
1147 * Flush out buffers on the specified list.
1148 *
1149 */
1150 static int
1151 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1152 int slptimeo)
1153 {
1154 struct buf *bp, *nbp;
1155 int retval, error;
1156 daddr_t lblkno;
1157 b_xflags_t xflags;
1158
1159 ASSERT_BO_LOCKED(bo);
1160
1161 retval = 0;
1162 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1163 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1164 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1165 continue;
1166 }
1167 lblkno = 0;
1168 xflags = 0;
1169 if (nbp != NULL) {
1170 lblkno = nbp->b_lblkno;
1171 xflags = nbp->b_xflags &
1172 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1173 }
1174 retval = EAGAIN;
1175 error = BUF_TIMELOCK(bp,
1176 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1177 "flushbuf", slpflag, slptimeo);
1178 if (error) {
1179 BO_LOCK(bo);
1180 return (error != ENOLCK ? error : EAGAIN);
1181 }
1182 KASSERT(bp->b_bufobj == bo,
1183 ("bp %p wrong b_bufobj %p should be %p",
1184 bp, bp->b_bufobj, bo));
1185 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1186 BUF_UNLOCK(bp);
1187 BO_LOCK(bo);
1188 return (EAGAIN);
1189 }
1190 /*
1191 * XXX Since there are no node locks for NFS, I
1192 * believe there is a slight chance that a delayed
1193 * write will occur while sleeping just above, so
1194 * check for it.
1195 */
1196 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1197 (flags & V_SAVE)) {
1198 bremfree(bp);
1199 bp->b_flags |= B_ASYNC;
1200 bwrite(bp);
1201 BO_LOCK(bo);
1202 return (EAGAIN); /* XXX: why not loop ? */
1203 }
1204 bremfree(bp);
1205 bp->b_flags |= (B_INVAL | B_RELBUF);
1206 bp->b_flags &= ~B_ASYNC;
1207 brelse(bp);
1208 BO_LOCK(bo);
1209 if (nbp != NULL &&
1210 (nbp->b_bufobj != bo ||
1211 nbp->b_lblkno != lblkno ||
1212 (nbp->b_xflags &
1213 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1214 break; /* nbp invalid */
1215 }
1216 return (retval);
1217 }
1218
1219 /*
1220 * Truncate a file's buffer and pages to a specified length. This
1221 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1222 * sync activity.
1223 */
1224 int
1225 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1226 off_t length, int blksize)
1227 {
1228 struct buf *bp, *nbp;
1229 int anyfreed;
1230 int trunclbn;
1231 struct bufobj *bo;
1232
1233 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1234 /*
1235 * Round up to the *next* lbn.
1236 */
1237 trunclbn = (length + blksize - 1) / blksize;
1238
1239 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1240 restart:
1241 VI_LOCK(vp);
1242 bo = &vp->v_bufobj;
1243 anyfreed = 1;
1244 for (;anyfreed;) {
1245 anyfreed = 0;
1246 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1247 if (bp->b_lblkno < trunclbn)
1248 continue;
1249 if (BUF_LOCK(bp,
1250 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1251 VI_MTX(vp)) == ENOLCK)
1252 goto restart;
1253
1254 bremfree(bp);
1255 bp->b_flags |= (B_INVAL | B_RELBUF);
1256 bp->b_flags &= ~B_ASYNC;
1257 brelse(bp);
1258 anyfreed = 1;
1259
1260 if (nbp != NULL &&
1261 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1262 (nbp->b_vp != vp) ||
1263 (nbp->b_flags & B_DELWRI))) {
1264 goto restart;
1265 }
1266 VI_LOCK(vp);
1267 }
1268
1269 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1270 if (bp->b_lblkno < trunclbn)
1271 continue;
1272 if (BUF_LOCK(bp,
1273 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1274 VI_MTX(vp)) == ENOLCK)
1275 goto restart;
1276 bremfree(bp);
1277 bp->b_flags |= (B_INVAL | B_RELBUF);
1278 bp->b_flags &= ~B_ASYNC;
1279 brelse(bp);
1280 anyfreed = 1;
1281 if (nbp != NULL &&
1282 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1283 (nbp->b_vp != vp) ||
1284 (nbp->b_flags & B_DELWRI) == 0)) {
1285 goto restart;
1286 }
1287 VI_LOCK(vp);
1288 }
1289 }
1290
1291 if (length > 0) {
1292 restartsync:
1293 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1294 if (bp->b_lblkno > 0)
1295 continue;
1296 /*
1297 * Since we hold the vnode lock this should only
1298 * fail if we're racing with the buf daemon.
1299 */
1300 if (BUF_LOCK(bp,
1301 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1302 VI_MTX(vp)) == ENOLCK) {
1303 goto restart;
1304 }
1305 VNASSERT((bp->b_flags & B_DELWRI), vp,
1306 ("buf(%p) on dirty queue without DELWRI", bp));
1307
1308 bremfree(bp);
1309 bawrite(bp);
1310 VI_LOCK(vp);
1311 goto restartsync;
1312 }
1313 }
1314
1315 bufobj_wwait(bo, 0, 0);
1316 VI_UNLOCK(vp);
1317 vnode_pager_setsize(vp, length);
1318
1319 return (0);
1320 }
1321
1322 /*
1323 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1324 * a vnode.
1325 *
1326 * NOTE: We have to deal with the special case of a background bitmap
1327 * buffer, a situation where two buffers will have the same logical
1328 * block offset. We want (1) only the foreground buffer to be accessed
1329 * in a lookup and (2) must differentiate between the foreground and
1330 * background buffer in the splay tree algorithm because the splay
1331 * tree cannot normally handle multiple entities with the same 'index'.
1332 * We accomplish this by adding differentiating flags to the splay tree's
1333 * numerical domain.
1334 */
1335 static
1336 struct buf *
1337 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1338 {
1339 struct buf dummy;
1340 struct buf *lefttreemax, *righttreemin, *y;
1341
1342 if (root == NULL)
1343 return (NULL);
1344 lefttreemax = righttreemin = &dummy;
1345 for (;;) {
1346 if (lblkno < root->b_lblkno ||
1347 (lblkno == root->b_lblkno &&
1348 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1349 if ((y = root->b_left) == NULL)
1350 break;
1351 if (lblkno < y->b_lblkno) {
1352 /* Rotate right. */
1353 root->b_left = y->b_right;
1354 y->b_right = root;
1355 root = y;
1356 if ((y = root->b_left) == NULL)
1357 break;
1358 }
1359 /* Link into the new root's right tree. */
1360 righttreemin->b_left = root;
1361 righttreemin = root;
1362 } else if (lblkno > root->b_lblkno ||
1363 (lblkno == root->b_lblkno &&
1364 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1365 if ((y = root->b_right) == NULL)
1366 break;
1367 if (lblkno > y->b_lblkno) {
1368 /* Rotate left. */
1369 root->b_right = y->b_left;
1370 y->b_left = root;
1371 root = y;
1372 if ((y = root->b_right) == NULL)
1373 break;
1374 }
1375 /* Link into the new root's left tree. */
1376 lefttreemax->b_right = root;
1377 lefttreemax = root;
1378 } else {
1379 break;
1380 }
1381 root = y;
1382 }
1383 /* Assemble the new root. */
1384 lefttreemax->b_right = root->b_left;
1385 righttreemin->b_left = root->b_right;
1386 root->b_left = dummy.b_right;
1387 root->b_right = dummy.b_left;
1388 return (root);
1389 }
1390
1391 static void
1392 buf_vlist_remove(struct buf *bp)
1393 {
1394 struct buf *root;
1395 struct bufv *bv;
1396
1397 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1398 ASSERT_BO_LOCKED(bp->b_bufobj);
1399 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1400 (BX_VNDIRTY|BX_VNCLEAN),
1401 ("buf_vlist_remove: Buf %p is on two lists", bp));
1402 if (bp->b_xflags & BX_VNDIRTY)
1403 bv = &bp->b_bufobj->bo_dirty;
1404 else
1405 bv = &bp->b_bufobj->bo_clean;
1406 if (bp != bv->bv_root) {
1407 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1408 KASSERT(root == bp, ("splay lookup failed in remove"));
1409 }
1410 if (bp->b_left == NULL) {
1411 root = bp->b_right;
1412 } else {
1413 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1414 root->b_right = bp->b_right;
1415 }
1416 bv->bv_root = root;
1417 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1418 bv->bv_cnt--;
1419 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1420 }
1421
1422 /*
1423 * Add the buffer to the sorted clean or dirty block list using a
1424 * splay tree algorithm.
1425 *
1426 * NOTE: xflags is passed as a constant, optimizing this inline function!
1427 */
1428 static void
1429 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1430 {
1431 struct buf *root;
1432 struct bufv *bv;
1433
1434 ASSERT_BO_LOCKED(bo);
1435 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1436 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1437 bp->b_xflags |= xflags;
1438 if (xflags & BX_VNDIRTY)
1439 bv = &bo->bo_dirty;
1440 else
1441 bv = &bo->bo_clean;
1442
1443 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1444 if (root == NULL) {
1445 bp->b_left = NULL;
1446 bp->b_right = NULL;
1447 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1448 } else if (bp->b_lblkno < root->b_lblkno ||
1449 (bp->b_lblkno == root->b_lblkno &&
1450 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1451 bp->b_left = root->b_left;
1452 bp->b_right = root;
1453 root->b_left = NULL;
1454 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1455 } else {
1456 bp->b_right = root->b_right;
1457 bp->b_left = root;
1458 root->b_right = NULL;
1459 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1460 }
1461 bv->bv_cnt++;
1462 bv->bv_root = bp;
1463 }
1464
1465 /*
1466 * Lookup a buffer using the splay tree. Note that we specifically avoid
1467 * shadow buffers used in background bitmap writes.
1468 *
1469 * This code isn't quite efficient as it could be because we are maintaining
1470 * two sorted lists and do not know which list the block resides in.
1471 *
1472 * During a "make buildworld" the desired buffer is found at one of
1473 * the roots more than 60% of the time. Thus, checking both roots
1474 * before performing either splay eliminates unnecessary splays on the
1475 * first tree splayed.
1476 */
1477 struct buf *
1478 gbincore(struct bufobj *bo, daddr_t lblkno)
1479 {
1480 struct buf *bp;
1481
1482 ASSERT_BO_LOCKED(bo);
1483 if ((bp = bo->bo_clean.bv_root) != NULL &&
1484 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1485 return (bp);
1486 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1487 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1488 return (bp);
1489 if ((bp = bo->bo_clean.bv_root) != NULL) {
1490 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1491 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1492 return (bp);
1493 }
1494 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1495 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1496 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1497 return (bp);
1498 }
1499 return (NULL);
1500 }
1501
1502 /*
1503 * Associate a buffer with a vnode.
1504 */
1505 void
1506 bgetvp(struct vnode *vp, struct buf *bp)
1507 {
1508
1509 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1510
1511 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1512 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1513 ("bgetvp: bp already attached! %p", bp));
1514
1515 ASSERT_VI_LOCKED(vp, "bgetvp");
1516 vholdl(vp);
1517 if (VFS_NEEDSGIANT(vp->v_mount) ||
1518 vp->v_bufobj.bo_flag & BO_NEEDSGIANT)
1519 bp->b_flags |= B_NEEDSGIANT;
1520 bp->b_vp = vp;
1521 bp->b_bufobj = &vp->v_bufobj;
1522 /*
1523 * Insert onto list for new vnode.
1524 */
1525 buf_vlist_add(bp, &vp->v_bufobj, BX_VNCLEAN);
1526 }
1527
1528 /*
1529 * Disassociate a buffer from a vnode.
1530 */
1531 void
1532 brelvp(struct buf *bp)
1533 {
1534 struct bufobj *bo;
1535 struct vnode *vp;
1536
1537 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1538 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1539
1540 /*
1541 * Delete from old vnode list, if on one.
1542 */
1543 vp = bp->b_vp; /* XXX */
1544 bo = bp->b_bufobj;
1545 BO_LOCK(bo);
1546 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1547 buf_vlist_remove(bp);
1548 else
1549 panic("brelvp: Buffer %p not on queue.", bp);
1550 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1551 bo->bo_flag &= ~BO_ONWORKLST;
1552 mtx_lock(&sync_mtx);
1553 LIST_REMOVE(bo, bo_synclist);
1554 syncer_worklist_len--;
1555 mtx_unlock(&sync_mtx);
1556 }
1557 bp->b_flags &= ~B_NEEDSGIANT;
1558 bp->b_vp = NULL;
1559 bp->b_bufobj = NULL;
1560 vdropl(vp);
1561 }
1562
1563 /*
1564 * Add an item to the syncer work queue.
1565 */
1566 static void
1567 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1568 {
1569 int slot;
1570
1571 ASSERT_BO_LOCKED(bo);
1572
1573 mtx_lock(&sync_mtx);
1574 if (bo->bo_flag & BO_ONWORKLST)
1575 LIST_REMOVE(bo, bo_synclist);
1576 else {
1577 bo->bo_flag |= BO_ONWORKLST;
1578 syncer_worklist_len++;
1579 }
1580
1581 if (delay > syncer_maxdelay - 2)
1582 delay = syncer_maxdelay - 2;
1583 slot = (syncer_delayno + delay) & syncer_mask;
1584
1585 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1586 mtx_unlock(&sync_mtx);
1587 }
1588
1589 static int
1590 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1591 {
1592 int error, len;
1593
1594 mtx_lock(&sync_mtx);
1595 len = syncer_worklist_len - sync_vnode_count;
1596 mtx_unlock(&sync_mtx);
1597 error = SYSCTL_OUT(req, &len, sizeof(len));
1598 return (error);
1599 }
1600
1601 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1602 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1603
1604 static struct proc *updateproc;
1605 static void sched_sync(void);
1606 static struct kproc_desc up_kp = {
1607 "syncer",
1608 sched_sync,
1609 &updateproc
1610 };
1611 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1612
1613 static int
1614 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1615 {
1616 struct vnode *vp;
1617 struct mount *mp;
1618 int vfslocked;
1619
1620 vfslocked = 0;
1621 restart:
1622 *bo = LIST_FIRST(slp);
1623 if (*bo == NULL) {
1624 VFS_UNLOCK_GIANT(vfslocked);
1625 return (0);
1626 }
1627 vp = (*bo)->__bo_vnode; /* XXX */
1628 if (VFS_NEEDSGIANT(vp->v_mount)) {
1629 if (!vfslocked) {
1630 vfslocked = 1;
1631 if (mtx_trylock(&Giant) == 0) {
1632 mtx_unlock(&sync_mtx);
1633 mtx_lock(&Giant);
1634 mtx_lock(&sync_mtx);
1635 goto restart;
1636 }
1637 }
1638 } else {
1639 VFS_UNLOCK_GIANT(vfslocked);
1640 vfslocked = 0;
1641 }
1642 if (VOP_ISLOCKED(vp, NULL) != 0) {
1643 VFS_UNLOCK_GIANT(vfslocked);
1644 return (1);
1645 }
1646 if (VI_TRYLOCK(vp) == 0) {
1647 VFS_UNLOCK_GIANT(vfslocked);
1648 return (1);
1649 }
1650 /*
1651 * We use vhold in case the vnode does not
1652 * successfully sync. vhold prevents the vnode from
1653 * going away when we unlock the sync_mtx so that
1654 * we can acquire the vnode interlock.
1655 */
1656 vholdl(vp);
1657 mtx_unlock(&sync_mtx);
1658 VI_UNLOCK(vp);
1659 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1660 vdrop(vp);
1661 VFS_UNLOCK_GIANT(vfslocked);
1662 mtx_lock(&sync_mtx);
1663 return (*bo == LIST_FIRST(slp));
1664 }
1665 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1666 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1667 VOP_UNLOCK(vp, 0, td);
1668 vn_finished_write(mp);
1669 VI_LOCK(vp);
1670 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1671 /*
1672 * Put us back on the worklist. The worklist
1673 * routine will remove us from our current
1674 * position and then add us back in at a later
1675 * position.
1676 */
1677 vn_syncer_add_to_worklist(*bo, syncdelay);
1678 }
1679 vdropl(vp);
1680 VFS_UNLOCK_GIANT(vfslocked);
1681 mtx_lock(&sync_mtx);
1682 return (0);
1683 }
1684
1685 /*
1686 * System filesystem synchronizer daemon.
1687 */
1688 static void
1689 sched_sync(void)
1690 {
1691 struct synclist *next;
1692 struct synclist *slp;
1693 struct bufobj *bo;
1694 long starttime;
1695 struct thread *td = curthread;
1696 static int dummychan;
1697 int last_work_seen;
1698 int net_worklist_len;
1699 int syncer_final_iter;
1700 int first_printf;
1701 int error;
1702
1703 last_work_seen = 0;
1704 syncer_final_iter = 0;
1705 first_printf = 1;
1706 syncer_state = SYNCER_RUNNING;
1707 starttime = time_uptime;
1708 td->td_pflags |= TDP_NORUNNINGBUF;
1709
1710 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1711 SHUTDOWN_PRI_LAST);
1712
1713 mtx_lock(&sync_mtx);
1714 for (;;) {
1715 if (syncer_state == SYNCER_FINAL_DELAY &&
1716 syncer_final_iter == 0) {
1717 mtx_unlock(&sync_mtx);
1718 kthread_suspend_check(td->td_proc);
1719 mtx_lock(&sync_mtx);
1720 }
1721 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1722 if (syncer_state != SYNCER_RUNNING &&
1723 starttime != time_uptime) {
1724 if (first_printf) {
1725 printf("\nSyncing disks, vnodes remaining...");
1726 first_printf = 0;
1727 }
1728 printf("%d ", net_worklist_len);
1729 }
1730 starttime = time_uptime;
1731
1732 /*
1733 * Push files whose dirty time has expired. Be careful
1734 * of interrupt race on slp queue.
1735 *
1736 * Skip over empty worklist slots when shutting down.
1737 */
1738 do {
1739 slp = &syncer_workitem_pending[syncer_delayno];
1740 syncer_delayno += 1;
1741 if (syncer_delayno == syncer_maxdelay)
1742 syncer_delayno = 0;
1743 next = &syncer_workitem_pending[syncer_delayno];
1744 /*
1745 * If the worklist has wrapped since the
1746 * it was emptied of all but syncer vnodes,
1747 * switch to the FINAL_DELAY state and run
1748 * for one more second.
1749 */
1750 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1751 net_worklist_len == 0 &&
1752 last_work_seen == syncer_delayno) {
1753 syncer_state = SYNCER_FINAL_DELAY;
1754 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1755 }
1756 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1757 syncer_worklist_len > 0);
1758
1759 /*
1760 * Keep track of the last time there was anything
1761 * on the worklist other than syncer vnodes.
1762 * Return to the SHUTTING_DOWN state if any
1763 * new work appears.
1764 */
1765 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1766 last_work_seen = syncer_delayno;
1767 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1768 syncer_state = SYNCER_SHUTTING_DOWN;
1769 while (!LIST_EMPTY(slp)) {
1770 error = sync_vnode(slp, &bo, td);
1771 if (error == 1) {
1772 LIST_REMOVE(bo, bo_synclist);
1773 LIST_INSERT_HEAD(next, bo, bo_synclist);
1774 continue;
1775 }
1776 }
1777 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1778 syncer_final_iter--;
1779 /*
1780 * The variable rushjob allows the kernel to speed up the
1781 * processing of the filesystem syncer process. A rushjob
1782 * value of N tells the filesystem syncer to process the next
1783 * N seconds worth of work on its queue ASAP. Currently rushjob
1784 * is used by the soft update code to speed up the filesystem
1785 * syncer process when the incore state is getting so far
1786 * ahead of the disk that the kernel memory pool is being
1787 * threatened with exhaustion.
1788 */
1789 if (rushjob > 0) {
1790 rushjob -= 1;
1791 continue;
1792 }
1793 /*
1794 * Just sleep for a short period of time between
1795 * iterations when shutting down to allow some I/O
1796 * to happen.
1797 *
1798 * If it has taken us less than a second to process the
1799 * current work, then wait. Otherwise start right over
1800 * again. We can still lose time if any single round
1801 * takes more than two seconds, but it does not really
1802 * matter as we are just trying to generally pace the
1803 * filesystem activity.
1804 */
1805 if (syncer_state != SYNCER_RUNNING)
1806 msleep(&dummychan, &sync_mtx, PPAUSE, "syncfnl",
1807 hz / SYNCER_SHUTDOWN_SPEEDUP);
1808 else if (time_uptime == starttime)
1809 msleep(&lbolt, &sync_mtx, PPAUSE, "syncer", 0);
1810 }
1811 }
1812
1813 /*
1814 * Request the syncer daemon to speed up its work.
1815 * We never push it to speed up more than half of its
1816 * normal turn time, otherwise it could take over the cpu.
1817 */
1818 int
1819 speedup_syncer(void)
1820 {
1821 struct thread *td;
1822 int ret = 0;
1823
1824 td = FIRST_THREAD_IN_PROC(updateproc);
1825 mtx_lock(&sync_mtx);
1826 if (rushjob < syncdelay / 2) {
1827 rushjob += 1;
1828 stat_rush_requests += 1;
1829 ret = 1;
1830 }
1831 mtx_unlock(&sync_mtx);
1832 sleepq_remove(td, &lbolt);
1833 return (ret);
1834 }
1835
1836 /*
1837 * Tell the syncer to speed up its work and run though its work
1838 * list several times, then tell it to shut down.
1839 */
1840 static void
1841 syncer_shutdown(void *arg, int howto)
1842 {
1843 struct thread *td;
1844
1845 if (howto & RB_NOSYNC)
1846 return;
1847 td = FIRST_THREAD_IN_PROC(updateproc);
1848 mtx_lock(&sync_mtx);
1849 syncer_state = SYNCER_SHUTTING_DOWN;
1850 rushjob = 0;
1851 mtx_unlock(&sync_mtx);
1852 sleepq_remove(td, &lbolt);
1853 kproc_shutdown(arg, howto);
1854 }
1855
1856 /*
1857 * Reassign a buffer from one vnode to another.
1858 * Used to assign file specific control information
1859 * (indirect blocks) to the vnode to which they belong.
1860 */
1861 void
1862 reassignbuf(struct buf *bp)
1863 {
1864 struct vnode *vp;
1865 struct bufobj *bo;
1866 int delay;
1867 #ifdef INVARIANTS
1868 struct bufv *bv;
1869 #endif
1870
1871 vp = bp->b_vp;
1872 bo = bp->b_bufobj;
1873 ++reassignbufcalls;
1874
1875 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1876 bp, bp->b_vp, bp->b_flags);
1877 /*
1878 * B_PAGING flagged buffers cannot be reassigned because their vp
1879 * is not fully linked in.
1880 */
1881 if (bp->b_flags & B_PAGING)
1882 panic("cannot reassign paging buffer");
1883
1884 /*
1885 * Delete from old vnode list, if on one.
1886 */
1887 VI_LOCK(vp);
1888 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1889 buf_vlist_remove(bp);
1890 else
1891 panic("reassignbuf: Buffer %p not on queue.", bp);
1892 /*
1893 * If dirty, put on list of dirty buffers; otherwise insert onto list
1894 * of clean buffers.
1895 */
1896 if (bp->b_flags & B_DELWRI) {
1897 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1898 switch (vp->v_type) {
1899 case VDIR:
1900 delay = dirdelay;
1901 break;
1902 case VCHR:
1903 delay = metadelay;
1904 break;
1905 default:
1906 delay = filedelay;
1907 }
1908 vn_syncer_add_to_worklist(bo, delay);
1909 }
1910 buf_vlist_add(bp, bo, BX_VNDIRTY);
1911 } else {
1912 buf_vlist_add(bp, bo, BX_VNCLEAN);
1913
1914 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1915 mtx_lock(&sync_mtx);
1916 LIST_REMOVE(bo, bo_synclist);
1917 syncer_worklist_len--;
1918 mtx_unlock(&sync_mtx);
1919 bo->bo_flag &= ~BO_ONWORKLST;
1920 }
1921 }
1922 #ifdef INVARIANTS
1923 bv = &bo->bo_clean;
1924 bp = TAILQ_FIRST(&bv->bv_hd);
1925 KASSERT(bp == NULL || bp->b_bufobj == bo,
1926 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1927 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1928 KASSERT(bp == NULL || bp->b_bufobj == bo,
1929 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1930 bv = &bo->bo_dirty;
1931 bp = TAILQ_FIRST(&bv->bv_hd);
1932 KASSERT(bp == NULL || bp->b_bufobj == bo,
1933 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1934 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1935 KASSERT(bp == NULL || bp->b_bufobj == bo,
1936 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1937 #endif
1938 VI_UNLOCK(vp);
1939 }
1940
1941 /*
1942 * Increment the use and hold counts on the vnode, taking care to reference
1943 * the driver's usecount if this is a chardev. The vholdl() will remove
1944 * the vnode from the free list if it is presently free. Requires the
1945 * vnode interlock and returns with it held.
1946 */
1947 static void
1948 v_incr_usecount(struct vnode *vp)
1949 {
1950
1951 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1952 vp, vp->v_holdcnt, vp->v_usecount);
1953 vp->v_usecount++;
1954 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1955 dev_lock();
1956 vp->v_rdev->si_usecount++;
1957 dev_unlock();
1958 }
1959 vholdl(vp);
1960 }
1961
1962 /*
1963 * Turn a holdcnt into a use+holdcnt such that only one call to
1964 * v_decr_usecount is needed.
1965 */
1966 static void
1967 v_upgrade_usecount(struct vnode *vp)
1968 {
1969
1970 CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
1971 vp, vp->v_holdcnt, vp->v_usecount);
1972 vp->v_usecount++;
1973 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1974 dev_lock();
1975 vp->v_rdev->si_usecount++;
1976 dev_unlock();
1977 }
1978 }
1979
1980 /*
1981 * Decrement the vnode use and hold count along with the driver's usecount
1982 * if this is a chardev. The vdropl() below releases the vnode interlock
1983 * as it may free the vnode.
1984 */
1985 static void
1986 v_decr_usecount(struct vnode *vp)
1987 {
1988
1989 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
1990 vp, vp->v_holdcnt, vp->v_usecount);
1991 ASSERT_VI_LOCKED(vp, __FUNCTION__);
1992 VNASSERT(vp->v_usecount > 0, vp,
1993 ("v_decr_usecount: negative usecount"));
1994 vp->v_usecount--;
1995 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1996 dev_lock();
1997 vp->v_rdev->si_usecount--;
1998 dev_unlock();
1999 }
2000 vdropl(vp);
2001 }
2002
2003 /*
2004 * Decrement only the use count and driver use count. This is intended to
2005 * be paired with a follow on vdropl() to release the remaining hold count.
2006 * In this way we may vgone() a vnode with a 0 usecount without risk of
2007 * having it end up on a free list because the hold count is kept above 0.
2008 */
2009 static void
2010 v_decr_useonly(struct vnode *vp)
2011 {
2012
2013 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
2014 vp, vp->v_holdcnt, vp->v_usecount);
2015 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2016 VNASSERT(vp->v_usecount > 0, vp,
2017 ("v_decr_useonly: negative usecount"));
2018 vp->v_usecount--;
2019 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2020 dev_lock();
2021 vp->v_rdev->si_usecount--;
2022 dev_unlock();
2023 }
2024 }
2025
2026 /*
2027 * Grab a particular vnode from the free list, increment its
2028 * reference count and lock it. The vnode lock bit is set if the
2029 * vnode is being eliminated in vgone. The process is awakened
2030 * when the transition is completed, and an error returned to
2031 * indicate that the vnode is no longer usable (possibly having
2032 * been changed to a new filesystem type).
2033 */
2034 int
2035 vget(struct vnode *vp, int flags, struct thread *td)
2036 {
2037 int oweinact;
2038 int oldflags;
2039 int error;
2040
2041 error = 0;
2042 oldflags = flags;
2043 oweinact = 0;
2044 VFS_ASSERT_GIANT(vp->v_mount);
2045 if ((flags & LK_INTERLOCK) == 0)
2046 VI_LOCK(vp);
2047 /*
2048 * If the inactive call was deferred because vput() was called
2049 * with a shared lock, we have to do it here before another thread
2050 * gets a reference to data that should be dead.
2051 */
2052 if (vp->v_iflag & VI_OWEINACT) {
2053 if (flags & LK_NOWAIT) {
2054 VI_UNLOCK(vp);
2055 return (EBUSY);
2056 }
2057 flags &= ~LK_TYPE_MASK;
2058 flags |= LK_EXCLUSIVE;
2059 oweinact = 1;
2060 }
2061 vholdl(vp);
2062 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2063 vdrop(vp);
2064 return (error);
2065 }
2066 VI_LOCK(vp);
2067 /* Upgrade our holdcnt to a usecount. */
2068 v_upgrade_usecount(vp);
2069 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2070 panic("vget: vn_lock failed to return ENOENT\n");
2071 if (oweinact) {
2072 if (vp->v_iflag & VI_OWEINACT)
2073 vinactive(vp, td);
2074 VI_UNLOCK(vp);
2075 if ((oldflags & LK_TYPE_MASK) == 0)
2076 VOP_UNLOCK(vp, 0, td);
2077 } else
2078 VI_UNLOCK(vp);
2079 return (0);
2080 }
2081
2082 /*
2083 * Increase the reference count of a vnode.
2084 */
2085 void
2086 vref(struct vnode *vp)
2087 {
2088
2089 VI_LOCK(vp);
2090 v_incr_usecount(vp);
2091 VI_UNLOCK(vp);
2092 }
2093
2094 /*
2095 * Return reference count of a vnode.
2096 *
2097 * The results of this call are only guaranteed when some mechanism other
2098 * than the VI lock is used to stop other processes from gaining references
2099 * to the vnode. This may be the case if the caller holds the only reference.
2100 * This is also useful when stale data is acceptable as race conditions may
2101 * be accounted for by some other means.
2102 */
2103 int
2104 vrefcnt(struct vnode *vp)
2105 {
2106 int usecnt;
2107
2108 VI_LOCK(vp);
2109 usecnt = vp->v_usecount;
2110 VI_UNLOCK(vp);
2111
2112 return (usecnt);
2113 }
2114
2115
2116 /*
2117 * Vnode put/release.
2118 * If count drops to zero, call inactive routine and return to freelist.
2119 */
2120 void
2121 vrele(struct vnode *vp)
2122 {
2123 struct thread *td = curthread; /* XXX */
2124
2125 KASSERT(vp != NULL, ("vrele: null vp"));
2126 VFS_ASSERT_GIANT(vp->v_mount);
2127
2128 VI_LOCK(vp);
2129
2130 /* Skip this v_writecount check if we're going to panic below. */
2131 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2132 ("vrele: missed vn_close"));
2133
2134 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2135 vp->v_usecount == 1)) {
2136 v_decr_usecount(vp);
2137 return;
2138 }
2139 if (vp->v_usecount != 1) {
2140 #ifdef DIAGNOSTIC
2141 vprint("vrele: negative ref count", vp);
2142 #endif
2143 VI_UNLOCK(vp);
2144 panic("vrele: negative ref cnt");
2145 }
2146 /*
2147 * We want to hold the vnode until the inactive finishes to
2148 * prevent vgone() races. We drop the use count here and the
2149 * hold count below when we're done.
2150 */
2151 v_decr_useonly(vp);
2152 /*
2153 * We must call VOP_INACTIVE with the node locked. Mark
2154 * as VI_DOINGINACT to avoid recursion.
2155 */
2156 vp->v_iflag |= VI_OWEINACT;
2157 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2158 VI_LOCK(vp);
2159 if (vp->v_usecount > 0)
2160 vp->v_iflag &= ~VI_OWEINACT;
2161 if (vp->v_iflag & VI_OWEINACT)
2162 vinactive(vp, td);
2163 VOP_UNLOCK(vp, 0, td);
2164 } else {
2165 VI_LOCK(vp);
2166 if (vp->v_usecount > 0)
2167 vp->v_iflag &= ~VI_OWEINACT;
2168 }
2169 vdropl(vp);
2170 }
2171
2172 /*
2173 * Release an already locked vnode. This give the same effects as
2174 * unlock+vrele(), but takes less time and avoids releasing and
2175 * re-aquiring the lock (as vrele() acquires the lock internally.)
2176 */
2177 void
2178 vput(struct vnode *vp)
2179 {
2180 struct thread *td = curthread; /* XXX */
2181 int error;
2182
2183 KASSERT(vp != NULL, ("vput: null vp"));
2184 ASSERT_VOP_LOCKED(vp, "vput");
2185 VFS_ASSERT_GIANT(vp->v_mount);
2186 VI_LOCK(vp);
2187 /* Skip this v_writecount check if we're going to panic below. */
2188 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2189 ("vput: missed vn_close"));
2190 error = 0;
2191
2192 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2193 vp->v_usecount == 1)) {
2194 VOP_UNLOCK(vp, 0, td);
2195 v_decr_usecount(vp);
2196 return;
2197 }
2198
2199 if (vp->v_usecount != 1) {
2200 #ifdef DIAGNOSTIC
2201 vprint("vput: negative ref count", vp);
2202 #endif
2203 panic("vput: negative ref cnt");
2204 }
2205 /*
2206 * We want to hold the vnode until the inactive finishes to
2207 * prevent vgone() races. We drop the use count here and the
2208 * hold count below when we're done.
2209 */
2210 v_decr_useonly(vp);
2211 vp->v_iflag |= VI_OWEINACT;
2212 if (VOP_ISLOCKED(vp, NULL) != LK_EXCLUSIVE) {
2213 error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT, td);
2214 VI_LOCK(vp);
2215 if (error) {
2216 if (vp->v_usecount > 0)
2217 vp->v_iflag &= ~VI_OWEINACT;
2218 goto done;
2219 }
2220 }
2221 if (vp->v_usecount > 0)
2222 vp->v_iflag &= ~VI_OWEINACT;
2223 if (vp->v_iflag & VI_OWEINACT)
2224 vinactive(vp, td);
2225 VOP_UNLOCK(vp, 0, td);
2226 done:
2227 vdropl(vp);
2228 }
2229
2230 /*
2231 * Somebody doesn't want the vnode recycled.
2232 */
2233 void
2234 vhold(struct vnode *vp)
2235 {
2236
2237 VI_LOCK(vp);
2238 vholdl(vp);
2239 VI_UNLOCK(vp);
2240 }
2241
2242 void
2243 vholdl(struct vnode *vp)
2244 {
2245
2246 vp->v_holdcnt++;
2247 if (VSHOULDBUSY(vp))
2248 vbusy(vp);
2249 }
2250
2251 /*
2252 * Note that there is one less who cares about this vnode. vdrop() is the
2253 * opposite of vhold().
2254 */
2255 void
2256 vdrop(struct vnode *vp)
2257 {
2258
2259 VI_LOCK(vp);
2260 vdropl(vp);
2261 }
2262
2263 /*
2264 * Drop the hold count of the vnode. If this is the last reference to
2265 * the vnode we will free it if it has been vgone'd otherwise it is
2266 * placed on the free list.
2267 */
2268 void
2269 vdropl(struct vnode *vp)
2270 {
2271
2272 ASSERT_VI_LOCKED(vp, "vdropl");
2273 if (vp->v_holdcnt <= 0)
2274 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2275 vp->v_holdcnt--;
2276 if (vp->v_holdcnt == 0) {
2277 if (vp->v_iflag & VI_DOOMED) {
2278 vdestroy(vp);
2279 return;
2280 } else
2281 vfree(vp);
2282 }
2283 VI_UNLOCK(vp);
2284 }
2285
2286 /*
2287 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2288 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2289 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2290 * failed lock upgrade.
2291 */
2292 static void
2293 vinactive(struct vnode *vp, struct thread *td)
2294 {
2295
2296 ASSERT_VOP_LOCKED(vp, "vinactive");
2297 ASSERT_VI_LOCKED(vp, "vinactive");
2298 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2299 ("vinactive: recursed on VI_DOINGINACT"));
2300 vp->v_iflag |= VI_DOINGINACT;
2301 vp->v_iflag &= ~VI_OWEINACT;
2302 VI_UNLOCK(vp);
2303 VOP_INACTIVE(vp, td);
2304 VI_LOCK(vp);
2305 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2306 ("vinactive: lost VI_DOINGINACT"));
2307 vp->v_iflag &= ~VI_DOINGINACT;
2308 }
2309
2310 /*
2311 * Remove any vnodes in the vnode table belonging to mount point mp.
2312 *
2313 * If FORCECLOSE is not specified, there should not be any active ones,
2314 * return error if any are found (nb: this is a user error, not a
2315 * system error). If FORCECLOSE is specified, detach any active vnodes
2316 * that are found.
2317 *
2318 * If WRITECLOSE is set, only flush out regular file vnodes open for
2319 * writing.
2320 *
2321 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2322 *
2323 * `rootrefs' specifies the base reference count for the root vnode
2324 * of this filesystem. The root vnode is considered busy if its
2325 * v_usecount exceeds this value. On a successful return, vflush(, td)
2326 * will call vrele() on the root vnode exactly rootrefs times.
2327 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2328 * be zero.
2329 */
2330 #ifdef DIAGNOSTIC
2331 static int busyprt = 0; /* print out busy vnodes */
2332 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2333 #endif
2334
2335 int
2336 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2337 {
2338 struct vnode *vp, *mvp, *rootvp = NULL;
2339 struct vattr vattr;
2340 int busy = 0, error;
2341
2342 CTR1(KTR_VFS, "vflush: mp %p", mp);
2343 if (rootrefs > 0) {
2344 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2345 ("vflush: bad args"));
2346 /*
2347 * Get the filesystem root vnode. We can vput() it
2348 * immediately, since with rootrefs > 0, it won't go away.
2349 */
2350 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2351 return (error);
2352 vput(rootvp);
2353
2354 }
2355 MNT_ILOCK(mp);
2356 loop:
2357 MNT_VNODE_FOREACH(vp, mp, mvp) {
2358
2359 VI_LOCK(vp);
2360 vholdl(vp);
2361 MNT_IUNLOCK(mp);
2362 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2363 if (error) {
2364 vdrop(vp);
2365 MNT_ILOCK(mp);
2366 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2367 goto loop;
2368 }
2369 /*
2370 * Skip over a vnodes marked VV_SYSTEM.
2371 */
2372 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2373 VOP_UNLOCK(vp, 0, td);
2374 vdrop(vp);
2375 MNT_ILOCK(mp);
2376 continue;
2377 }
2378 /*
2379 * If WRITECLOSE is set, flush out unlinked but still open
2380 * files (even if open only for reading) and regular file
2381 * vnodes open for writing.
2382 */
2383 if (flags & WRITECLOSE) {
2384 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2385 VI_LOCK(vp);
2386
2387 if ((vp->v_type == VNON ||
2388 (error == 0 && vattr.va_nlink > 0)) &&
2389 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2390 VOP_UNLOCK(vp, 0, td);
2391 vdropl(vp);
2392 MNT_ILOCK(mp);
2393 continue;
2394 }
2395 } else
2396 VI_LOCK(vp);
2397 /*
2398 * With v_usecount == 0, all we need to do is clear out the
2399 * vnode data structures and we are done.
2400 *
2401 * If FORCECLOSE is set, forcibly close the vnode.
2402 */
2403 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2404 VNASSERT(vp->v_usecount == 0 ||
2405 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2406 ("device VNODE %p is FORCECLOSED", vp));
2407 vgonel(vp);
2408 } else {
2409 busy++;
2410 #ifdef DIAGNOSTIC
2411 if (busyprt)
2412 vprint("vflush: busy vnode", vp);
2413 #endif
2414 }
2415 VOP_UNLOCK(vp, 0, td);
2416 vdropl(vp);
2417 MNT_ILOCK(mp);
2418 }
2419 MNT_IUNLOCK(mp);
2420 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2421 /*
2422 * If just the root vnode is busy, and if its refcount
2423 * is equal to `rootrefs', then go ahead and kill it.
2424 */
2425 VI_LOCK(rootvp);
2426 KASSERT(busy > 0, ("vflush: not busy"));
2427 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2428 ("vflush: usecount %d < rootrefs %d",
2429 rootvp->v_usecount, rootrefs));
2430 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2431 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK, td);
2432 vgone(rootvp);
2433 VOP_UNLOCK(rootvp, 0, td);
2434 busy = 0;
2435 } else
2436 VI_UNLOCK(rootvp);
2437 }
2438 if (busy)
2439 return (EBUSY);
2440 for (; rootrefs > 0; rootrefs--)
2441 vrele(rootvp);
2442 return (0);
2443 }
2444
2445 /*
2446 * Recycle an unused vnode to the front of the free list.
2447 */
2448 int
2449 vrecycle(struct vnode *vp, struct thread *td)
2450 {
2451 int recycled;
2452
2453 ASSERT_VOP_LOCKED(vp, "vrecycle");
2454 recycled = 0;
2455 VI_LOCK(vp);
2456 if (vp->v_usecount == 0) {
2457 recycled = 1;
2458 vgonel(vp);
2459 }
2460 VI_UNLOCK(vp);
2461 return (recycled);
2462 }
2463
2464 /*
2465 * Eliminate all activity associated with a vnode
2466 * in preparation for reuse.
2467 */
2468 void
2469 vgone(struct vnode *vp)
2470 {
2471 VI_LOCK(vp);
2472 vgonel(vp);
2473 VI_UNLOCK(vp);
2474 }
2475
2476 /*
2477 * vgone, with the vp interlock held.
2478 */
2479 void
2480 vgonel(struct vnode *vp)
2481 {
2482 struct thread *td;
2483 int oweinact;
2484 int active;
2485 struct mount *mp;
2486
2487 CTR1(KTR_VFS, "vgonel: vp %p", vp);
2488 ASSERT_VOP_LOCKED(vp, "vgonel");
2489 ASSERT_VI_LOCKED(vp, "vgonel");
2490 VNASSERT(vp->v_holdcnt, vp,
2491 ("vgonel: vp %p has no reference.", vp));
2492 td = curthread;
2493
2494 /*
2495 * Don't vgonel if we're already doomed.
2496 */
2497 if (vp->v_iflag & VI_DOOMED)
2498 return;
2499 vp->v_iflag |= VI_DOOMED;
2500 /*
2501 * Check to see if the vnode is in use. If so, we have to call
2502 * VOP_CLOSE() and VOP_INACTIVE().
2503 */
2504 active = vp->v_usecount;
2505 oweinact = (vp->v_iflag & VI_OWEINACT);
2506 VI_UNLOCK(vp);
2507 /*
2508 * Clean out any buffers associated with the vnode.
2509 * If the flush fails, just toss the buffers.
2510 */
2511 mp = NULL;
2512 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2513 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2514 if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2515 vinvalbuf(vp, 0, td, 0, 0);
2516
2517 /*
2518 * If purging an active vnode, it must be closed and
2519 * deactivated before being reclaimed.
2520 */
2521 if (active)
2522 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2523 if (oweinact || active) {
2524 VI_LOCK(vp);
2525 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2526 vinactive(vp, td);
2527 VI_UNLOCK(vp);
2528 }
2529 /*
2530 * Reclaim the vnode.
2531 */
2532 if (VOP_RECLAIM(vp, td))
2533 panic("vgone: cannot reclaim");
2534 if (mp != NULL)
2535 vn_finished_secondary_write(mp);
2536 VNASSERT(vp->v_object == NULL, vp,
2537 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2538 /*
2539 * Clear the advisory locks and wake up waiting threads.
2540 */
2541 lf_purgelocks(vp, &(vp->v_lockf));
2542 /*
2543 * Delete from old mount point vnode list.
2544 */
2545 delmntque(vp);
2546 cache_purge(vp);
2547 /*
2548 * Done with purge, reset to the standard lock and invalidate
2549 * the vnode.
2550 */
2551 VI_LOCK(vp);
2552 vp->v_vnlock = &vp->v_lock;
2553 vp->v_op = &dead_vnodeops;
2554 vp->v_tag = "none";
2555 vp->v_type = VBAD;
2556 }
2557
2558 /*
2559 * Calculate the total number of references to a special device.
2560 */
2561 int
2562 vcount(struct vnode *vp)
2563 {
2564 int count;
2565
2566 dev_lock();
2567 count = vp->v_rdev->si_usecount;
2568 dev_unlock();
2569 return (count);
2570 }
2571
2572 /*
2573 * Same as above, but using the struct cdev *as argument
2574 */
2575 int
2576 count_dev(struct cdev *dev)
2577 {
2578 int count;
2579
2580 dev_lock();
2581 count = dev->si_usecount;
2582 dev_unlock();
2583 return(count);
2584 }
2585
2586 /*
2587 * Print out a description of a vnode.
2588 */
2589 static char *typename[] =
2590 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2591 "VMARKER"};
2592
2593 void
2594 vn_printf(struct vnode *vp, const char *fmt, ...)
2595 {
2596 va_list ap;
2597 char buf[256], buf2[16];
2598 u_long flags;
2599
2600 va_start(ap, fmt);
2601 vprintf(fmt, ap);
2602 va_end(ap);
2603 printf("%p: ", (void *)vp);
2604 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2605 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2606 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2607 buf[0] = '\0';
2608 buf[1] = '\0';
2609 if (vp->v_vflag & VV_ROOT)
2610 strlcat(buf, "|VV_ROOT", sizeof(buf));
2611 if (vp->v_vflag & VV_ISTTY)
2612 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2613 if (vp->v_vflag & VV_NOSYNC)
2614 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2615 if (vp->v_vflag & VV_CACHEDLABEL)
2616 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2617 if (vp->v_vflag & VV_TEXT)
2618 strlcat(buf, "|VV_TEXT", sizeof(buf));
2619 if (vp->v_vflag & VV_COPYONWRITE)
2620 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2621 if (vp->v_vflag & VV_SYSTEM)
2622 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2623 if (vp->v_vflag & VV_PROCDEP)
2624 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2625 if (vp->v_vflag & VV_NOKNOTE)
2626 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2627 if (vp->v_vflag & VV_DELETED)
2628 strlcat(buf, "|VV_DELETED", sizeof(buf));
2629 if (vp->v_vflag & VV_MD)
2630 strlcat(buf, "|VV_MD", sizeof(buf));
2631 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2632 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2633 VV_NOKNOTE | VV_DELETED | VV_MD);
2634 if (flags != 0) {
2635 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2636 strlcat(buf, buf2, sizeof(buf));
2637 }
2638 if (vp->v_iflag & VI_MOUNT)
2639 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2640 if (vp->v_iflag & VI_AGE)
2641 strlcat(buf, "|VI_AGE", sizeof(buf));
2642 if (vp->v_iflag & VI_DOOMED)
2643 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2644 if (vp->v_iflag & VI_FREE)
2645 strlcat(buf, "|VI_FREE", sizeof(buf));
2646 if (vp->v_iflag & VI_OBJDIRTY)
2647 strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2648 if (vp->v_iflag & VI_DOINGINACT)
2649 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2650 if (vp->v_iflag & VI_OWEINACT)
2651 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2652 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2653 VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2654 if (flags != 0) {
2655 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2656 strlcat(buf, buf2, sizeof(buf));
2657 }
2658 printf(" flags (%s)\n", buf + 1);
2659 if (mtx_owned(VI_MTX(vp)))
2660 printf(" VI_LOCKed");
2661 if (vp->v_object != NULL)
2662 printf(" v_object %p ref %d pages %d\n",
2663 vp->v_object, vp->v_object->ref_count,
2664 vp->v_object->resident_page_count);
2665 printf(" ");
2666 lockmgr_printinfo(vp->v_vnlock);
2667 printf("\n");
2668 if (vp->v_data != NULL)
2669 VOP_PRINT(vp);
2670 }
2671
2672 #ifdef DDB
2673 /*
2674 * List all of the locked vnodes in the system.
2675 * Called when debugging the kernel.
2676 */
2677 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2678 {
2679 struct mount *mp, *nmp;
2680 struct vnode *vp;
2681
2682 /*
2683 * Note: because this is DDB, we can't obey the locking semantics
2684 * for these structures, which means we could catch an inconsistent
2685 * state and dereference a nasty pointer. Not much to be done
2686 * about that.
2687 */
2688 db_printf("Locked vnodes\n");
2689 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2690 nmp = TAILQ_NEXT(mp, mnt_list);
2691 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2692 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp, NULL))
2693 vprint("", vp);
2694 }
2695 nmp = TAILQ_NEXT(mp, mnt_list);
2696 }
2697 }
2698
2699 /*
2700 * Show details about the given vnode.
2701 */
2702 DB_SHOW_COMMAND(vnode, db_show_vnode)
2703 {
2704 struct vnode *vp;
2705
2706 if (!have_addr)
2707 return;
2708 vp = (struct vnode *)addr;
2709 vn_printf(vp, "vnode ");
2710 }
2711
2712 /*
2713 * Show details about the given mount point.
2714 */
2715 DB_SHOW_COMMAND(mount, db_show_mount)
2716 {
2717 struct mount *mp;
2718 struct statfs *sp;
2719 struct vnode *vp;
2720 char buf[512];
2721 u_int flags;
2722
2723 if (!have_addr) {
2724 /* No address given, print short info about all mount points. */
2725 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2726 db_printf("%p %s on %s (%s)\n", mp,
2727 mp->mnt_stat.f_mntfromname,
2728 mp->mnt_stat.f_mntonname,
2729 mp->mnt_stat.f_fstypename);
2730 if (db_pager_quit)
2731 break;
2732 }
2733 db_printf("\nMore info: show mount <addr>\n");
2734 return;
2735 }
2736
2737 mp = (struct mount *)addr;
2738 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2739 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2740
2741 buf[0] = '\0';
2742 flags = mp->mnt_flag;
2743 #define MNT_FLAG(flag) do { \
2744 if (flags & (flag)) { \
2745 if (buf[0] != '\0') \
2746 strlcat(buf, ", ", sizeof(buf)); \
2747 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2748 flags &= ~(flag); \
2749 } \
2750 } while (0)
2751 MNT_FLAG(MNT_RDONLY);
2752 MNT_FLAG(MNT_SYNCHRONOUS);
2753 MNT_FLAG(MNT_NOEXEC);
2754 MNT_FLAG(MNT_NOSUID);
2755 MNT_FLAG(MNT_UNION);
2756 MNT_FLAG(MNT_ASYNC);
2757 MNT_FLAG(MNT_SUIDDIR);
2758 MNT_FLAG(MNT_SOFTDEP);
2759 MNT_FLAG(MNT_NOSYMFOLLOW);
2760 MNT_FLAG(MNT_GJOURNAL);
2761 MNT_FLAG(MNT_MULTILABEL);
2762 MNT_FLAG(MNT_ACLS);
2763 MNT_FLAG(MNT_NOATIME);
2764 MNT_FLAG(MNT_NOCLUSTERR);
2765 MNT_FLAG(MNT_NOCLUSTERW);
2766 MNT_FLAG(MNT_EXRDONLY);
2767 MNT_FLAG(MNT_EXPORTED);
2768 MNT_FLAG(MNT_DEFEXPORTED);
2769 MNT_FLAG(MNT_EXPORTANON);
2770 MNT_FLAG(MNT_EXKERB);
2771 MNT_FLAG(MNT_EXPUBLIC);
2772 MNT_FLAG(MNT_LOCAL);
2773 MNT_FLAG(MNT_QUOTA);
2774 MNT_FLAG(MNT_ROOTFS);
2775 MNT_FLAG(MNT_USER);
2776 MNT_FLAG(MNT_IGNORE);
2777 MNT_FLAG(MNT_UPDATE);
2778 MNT_FLAG(MNT_DELEXPORT);
2779 MNT_FLAG(MNT_RELOAD);
2780 MNT_FLAG(MNT_FORCE);
2781 MNT_FLAG(MNT_SNAPSHOT);
2782 MNT_FLAG(MNT_BYFSID);
2783 #undef MNT_FLAG
2784 if (flags != 0) {
2785 if (buf[0] != '\0')
2786 strlcat(buf, ", ", sizeof(buf));
2787 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2788 "0x%08x", flags);
2789 }
2790 db_printf(" mnt_flag = %s\n", buf);
2791
2792 buf[0] = '\0';
2793 flags = mp->mnt_kern_flag;
2794 #define MNT_KERN_FLAG(flag) do { \
2795 if (flags & (flag)) { \
2796 if (buf[0] != '\0') \
2797 strlcat(buf, ", ", sizeof(buf)); \
2798 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2799 flags &= ~(flag); \
2800 } \
2801 } while (0)
2802 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2803 MNT_KERN_FLAG(MNTK_ASYNC);
2804 MNT_KERN_FLAG(MNTK_SOFTDEP);
2805 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2806 MNT_KERN_FLAG(MNTK_UNMOUNT);
2807 MNT_KERN_FLAG(MNTK_MWAIT);
2808 MNT_KERN_FLAG(MNTK_SUSPEND);
2809 MNT_KERN_FLAG(MNTK_SUSPEND2);
2810 MNT_KERN_FLAG(MNTK_SUSPENDED);
2811 MNT_KERN_FLAG(MNTK_MPSAFE);
2812 MNT_KERN_FLAG(MNTK_NOKNOTE);
2813 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2814 #undef MNT_KERN_FLAG
2815 if (flags != 0) {
2816 if (buf[0] != '\0')
2817 strlcat(buf, ", ", sizeof(buf));
2818 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2819 "0x%08x", flags);
2820 }
2821 db_printf(" mnt_kern_flag = %s\n", buf);
2822
2823 sp = &mp->mnt_stat;
2824 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2825 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2826 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2827 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2828 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2829 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2830 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2831 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2832 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2833 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2834 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2835 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2836
2837 db_printf(" mnt_cred = { uid=%u ruid=%u",
2838 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2839 if (mp->mnt_cred->cr_prison != NULL)
2840 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2841 db_printf(" }\n");
2842 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2843 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2844 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2845 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2846 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
2847 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2848 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2849 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2850 db_printf(" mnt_markercnt = %d\n", mp->mnt_markercnt);
2851 db_printf(" mnt_holdcnt = %d\n", mp->mnt_holdcnt);
2852 db_printf(" mnt_holdcntwaiters = %d\n", mp->mnt_holdcntwaiters);
2853 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2854 db_printf(" mnt_secondary_accwrites = %d\n",
2855 mp->mnt_secondary_accwrites);
2856 db_printf(" mnt_gjprovider = %s\n",
2857 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2858 db_printf("\n");
2859
2860 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2861 if (vp->v_type != VMARKER) {
2862 vn_printf(vp, "vnode ");
2863 if (db_pager_quit)
2864 break;
2865 }
2866 }
2867 }
2868 #endif /* DDB */
2869
2870 /*
2871 * Fill in a struct xvfsconf based on a struct vfsconf.
2872 */
2873 static void
2874 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2875 {
2876
2877 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2878 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2879 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2880 xvfsp->vfc_flags = vfsp->vfc_flags;
2881 /*
2882 * These are unused in userland, we keep them
2883 * to not break binary compatibility.
2884 */
2885 xvfsp->vfc_vfsops = NULL;
2886 xvfsp->vfc_next = NULL;
2887 }
2888
2889 /*
2890 * Top level filesystem related information gathering.
2891 */
2892 static int
2893 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2894 {
2895 struct vfsconf *vfsp;
2896 struct xvfsconf xvfsp;
2897 int error;
2898
2899 error = 0;
2900 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2901 bzero(&xvfsp, sizeof(xvfsp));
2902 vfsconf2x(vfsp, &xvfsp);
2903 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2904 if (error)
2905 break;
2906 }
2907 return (error);
2908 }
2909
2910 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2911 "S,xvfsconf", "List of all configured filesystems");
2912
2913 #ifndef BURN_BRIDGES
2914 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2915
2916 static int
2917 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2918 {
2919 int *name = (int *)arg1 - 1; /* XXX */
2920 u_int namelen = arg2 + 1; /* XXX */
2921 struct vfsconf *vfsp;
2922 struct xvfsconf xvfsp;
2923
2924 printf("WARNING: userland calling deprecated sysctl, "
2925 "please rebuild world\n");
2926
2927 #if 1 || defined(COMPAT_PRELITE2)
2928 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2929 if (namelen == 1)
2930 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2931 #endif
2932
2933 switch (name[1]) {
2934 case VFS_MAXTYPENUM:
2935 if (namelen != 2)
2936 return (ENOTDIR);
2937 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2938 case VFS_CONF:
2939 if (namelen != 3)
2940 return (ENOTDIR); /* overloaded */
2941 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2942 if (vfsp->vfc_typenum == name[2])
2943 break;
2944 if (vfsp == NULL)
2945 return (EOPNOTSUPP);
2946 bzero(&xvfsp, sizeof(xvfsp));
2947 vfsconf2x(vfsp, &xvfsp);
2948 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2949 }
2950 return (EOPNOTSUPP);
2951 }
2952
2953 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2954 vfs_sysctl, "Generic filesystem");
2955
2956 #if 1 || defined(COMPAT_PRELITE2)
2957
2958 static int
2959 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2960 {
2961 int error;
2962 struct vfsconf *vfsp;
2963 struct ovfsconf ovfs;
2964
2965 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2966 bzero(&ovfs, sizeof(ovfs));
2967 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2968 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2969 ovfs.vfc_index = vfsp->vfc_typenum;
2970 ovfs.vfc_refcount = vfsp->vfc_refcount;
2971 ovfs.vfc_flags = vfsp->vfc_flags;
2972 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2973 if (error)
2974 return error;
2975 }
2976 return 0;
2977 }
2978
2979 #endif /* 1 || COMPAT_PRELITE2 */
2980 #endif /* !BURN_BRIDGES */
2981
2982 #define KINFO_VNODESLOP 10
2983 #ifdef notyet
2984 /*
2985 * Dump vnode list (via sysctl).
2986 */
2987 /* ARGSUSED */
2988 static int
2989 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2990 {
2991 struct xvnode *xvn;
2992 struct thread *td = req->td;
2993 struct mount *mp;
2994 struct vnode *vp;
2995 int error, len, n;
2996
2997 /*
2998 * Stale numvnodes access is not fatal here.
2999 */
3000 req->lock = 0;
3001 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3002 if (!req->oldptr)
3003 /* Make an estimate */
3004 return (SYSCTL_OUT(req, 0, len));
3005
3006 error = sysctl_wire_old_buffer(req, 0);
3007 if (error != 0)
3008 return (error);
3009 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3010 n = 0;
3011 mtx_lock(&mountlist_mtx);
3012 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3013 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3014 continue;
3015 MNT_ILOCK(mp);
3016 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3017 if (n == len)
3018 break;
3019 vref(vp);
3020 xvn[n].xv_size = sizeof *xvn;
3021 xvn[n].xv_vnode = vp;
3022 xvn[n].xv_id = 0; /* XXX compat */
3023 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3024 XV_COPY(usecount);
3025 XV_COPY(writecount);
3026 XV_COPY(holdcnt);
3027 XV_COPY(mount);
3028 XV_COPY(numoutput);
3029 XV_COPY(type);
3030 #undef XV_COPY
3031 xvn[n].xv_flag = vp->v_vflag;
3032
3033 switch (vp->v_type) {
3034 case VREG:
3035 case VDIR:
3036 case VLNK:
3037 break;
3038 case VBLK:
3039 case VCHR:
3040 if (vp->v_rdev == NULL) {
3041 vrele(vp);
3042 continue;
3043 }
3044 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3045 break;
3046 case VSOCK:
3047 xvn[n].xv_socket = vp->v_socket;
3048 break;
3049 case VFIFO:
3050 xvn[n].xv_fifo = vp->v_fifoinfo;
3051 break;
3052 case VNON:
3053 case VBAD:
3054 default:
3055 /* shouldn't happen? */
3056 vrele(vp);
3057 continue;
3058 }
3059 vrele(vp);
3060 ++n;
3061 }
3062 MNT_IUNLOCK(mp);
3063 mtx_lock(&mountlist_mtx);
3064 vfs_unbusy(mp, td);
3065 if (n == len)
3066 break;
3067 }
3068 mtx_unlock(&mountlist_mtx);
3069
3070 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3071 free(xvn, M_TEMP);
3072 return (error);
3073 }
3074
3075 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3076 0, 0, sysctl_vnode, "S,xvnode", "");
3077 #endif
3078
3079 /*
3080 * Unmount all filesystems. The list is traversed in reverse order
3081 * of mounting to avoid dependencies.
3082 */
3083 void
3084 vfs_unmountall(void)
3085 {
3086 struct mount *mp;
3087 struct thread *td;
3088 int error;
3089
3090 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3091 td = curthread;
3092 /*
3093 * Since this only runs when rebooting, it is not interlocked.
3094 */
3095 while(!TAILQ_EMPTY(&mountlist)) {
3096 mp = TAILQ_LAST(&mountlist, mntlist);
3097 error = dounmount(mp, MNT_FORCE, td);
3098 if (error) {
3099 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3100 /*
3101 * XXX: Due to the way in which we mount the root
3102 * file system off of devfs, devfs will generate a
3103 * "busy" warning when we try to unmount it before
3104 * the root. Don't print a warning as a result in
3105 * order to avoid false positive errors that may
3106 * cause needless upset.
3107 */
3108 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3109 printf("unmount of %s failed (",
3110 mp->mnt_stat.f_mntonname);
3111 if (error == EBUSY)
3112 printf("BUSY)\n");
3113 else
3114 printf("%d)\n", error);
3115 }
3116 } else {
3117 /* The unmount has removed mp from the mountlist */
3118 }
3119 }
3120 }
3121
3122 /*
3123 * perform msync on all vnodes under a mount point
3124 * the mount point must be locked.
3125 */
3126 void
3127 vfs_msync(struct mount *mp, int flags)
3128 {
3129 struct vnode *vp, *mvp;
3130 struct vm_object *obj;
3131
3132 MNT_ILOCK(mp);
3133 MNT_VNODE_FOREACH(vp, mp, mvp) {
3134 VI_LOCK(vp);
3135 if ((vp->v_iflag & VI_OBJDIRTY) &&
3136 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3137 MNT_IUNLOCK(mp);
3138 if (!vget(vp,
3139 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3140 curthread)) {
3141 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3142 vput(vp);
3143 MNT_ILOCK(mp);
3144 continue;
3145 }
3146
3147 obj = vp->v_object;
3148 if (obj != NULL) {
3149 VM_OBJECT_LOCK(obj);
3150 vm_object_page_clean(obj, 0, 0,
3151 flags == MNT_WAIT ?
3152 OBJPC_SYNC : OBJPC_NOSYNC);
3153 VM_OBJECT_UNLOCK(obj);
3154 }
3155 vput(vp);
3156 }
3157 MNT_ILOCK(mp);
3158 } else
3159 VI_UNLOCK(vp);
3160 }
3161 MNT_IUNLOCK(mp);
3162 }
3163
3164 /*
3165 * Mark a vnode as free, putting it up for recycling.
3166 */
3167 static void
3168 vfree(struct vnode *vp)
3169 {
3170
3171 CTR1(KTR_VFS, "vfree vp %p", vp);
3172 ASSERT_VI_LOCKED(vp, "vfree");
3173 mtx_lock(&vnode_free_list_mtx);
3174 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3175 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3176 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3177 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3178 ("vfree: Freeing doomed vnode"));
3179 if (vp->v_iflag & VI_AGE) {
3180 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3181 } else {
3182 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3183 }
3184 freevnodes++;
3185 vp->v_iflag &= ~VI_AGE;
3186 vp->v_iflag |= VI_FREE;
3187 mtx_unlock(&vnode_free_list_mtx);
3188 }
3189
3190 /*
3191 * Opposite of vfree() - mark a vnode as in use.
3192 */
3193 static void
3194 vbusy(struct vnode *vp)
3195 {
3196 CTR1(KTR_VFS, "vbusy vp %p", vp);
3197 ASSERT_VI_LOCKED(vp, "vbusy");
3198 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3199 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3200
3201 mtx_lock(&vnode_free_list_mtx);
3202 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3203 freevnodes--;
3204 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3205 mtx_unlock(&vnode_free_list_mtx);
3206 }
3207
3208 static void
3209 destroy_vpollinfo(struct vpollinfo *vi)
3210 {
3211 knlist_destroy(&vi->vpi_selinfo.si_note);
3212 mtx_destroy(&vi->vpi_lock);
3213 uma_zfree(vnodepoll_zone, vi);
3214 }
3215
3216 /*
3217 * Initalize per-vnode helper structure to hold poll-related state.
3218 */
3219 void
3220 v_addpollinfo(struct vnode *vp)
3221 {
3222 struct vpollinfo *vi;
3223
3224 if (vp->v_pollinfo != NULL)
3225 return;
3226 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3227 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3228 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3229 vfs_knlunlock, NULL);
3230 VI_LOCK(vp);
3231 if (vp->v_pollinfo != NULL) {
3232 VI_UNLOCK(vp);
3233 destroy_vpollinfo(vi);
3234 return;
3235 }
3236 vp->v_pollinfo = vi;
3237 VI_UNLOCK(vp);
3238 }
3239
3240 /*
3241 * Record a process's interest in events which might happen to
3242 * a vnode. Because poll uses the historic select-style interface
3243 * internally, this routine serves as both the ``check for any
3244 * pending events'' and the ``record my interest in future events''
3245 * functions. (These are done together, while the lock is held,
3246 * to avoid race conditions.)
3247 */
3248 int
3249 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3250 {
3251
3252 v_addpollinfo(vp);
3253 mtx_lock(&vp->v_pollinfo->vpi_lock);
3254 if (vp->v_pollinfo->vpi_revents & events) {
3255 /*
3256 * This leaves events we are not interested
3257 * in available for the other process which
3258 * which presumably had requested them
3259 * (otherwise they would never have been
3260 * recorded).
3261 */
3262 events &= vp->v_pollinfo->vpi_revents;
3263 vp->v_pollinfo->vpi_revents &= ~events;
3264
3265 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3266 return events;
3267 }
3268 vp->v_pollinfo->vpi_events |= events;
3269 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3270 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3271 return 0;
3272 }
3273
3274 /*
3275 * Routine to create and manage a filesystem syncer vnode.
3276 */
3277 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3278 static int sync_fsync(struct vop_fsync_args *);
3279 static int sync_inactive(struct vop_inactive_args *);
3280 static int sync_reclaim(struct vop_reclaim_args *);
3281
3282 static struct vop_vector sync_vnodeops = {
3283 .vop_bypass = VOP_EOPNOTSUPP,
3284 .vop_close = sync_close, /* close */
3285 .vop_fsync = sync_fsync, /* fsync */
3286 .vop_inactive = sync_inactive, /* inactive */
3287 .vop_reclaim = sync_reclaim, /* reclaim */
3288 .vop_lock1 = vop_stdlock, /* lock */
3289 .vop_unlock = vop_stdunlock, /* unlock */
3290 .vop_islocked = vop_stdislocked, /* islocked */
3291 };
3292
3293 /*
3294 * Create a new filesystem syncer vnode for the specified mount point.
3295 */
3296 int
3297 vfs_allocate_syncvnode(struct mount *mp)
3298 {
3299 struct vnode *vp;
3300 static long start, incr, next;
3301 int error;
3302
3303 /* Allocate a new vnode */
3304 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3305 mp->mnt_syncer = NULL;
3306 return (error);
3307 }
3308 vp->v_type = VNON;
3309 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
3310 vp->v_vflag |= VV_FORCEINSMQ;
3311 error = insmntque(vp, mp);
3312 if (error != 0)
3313 panic("vfs_allocate_syncvnode: insmntque failed");
3314 vp->v_vflag &= ~VV_FORCEINSMQ;
3315 VOP_UNLOCK(vp, 0, curthread);
3316 /*
3317 * Place the vnode onto the syncer worklist. We attempt to
3318 * scatter them about on the list so that they will go off
3319 * at evenly distributed times even if all the filesystems
3320 * are mounted at once.
3321 */
3322 next += incr;
3323 if (next == 0 || next > syncer_maxdelay) {
3324 start /= 2;
3325 incr /= 2;
3326 if (start == 0) {
3327 start = syncer_maxdelay / 2;
3328 incr = syncer_maxdelay;
3329 }
3330 next = start;
3331 }
3332 VI_LOCK(vp);
3333 vn_syncer_add_to_worklist(&vp->v_bufobj,
3334 syncdelay > 0 ? next % syncdelay : 0);
3335 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3336 mtx_lock(&sync_mtx);
3337 sync_vnode_count++;
3338 mtx_unlock(&sync_mtx);
3339 VI_UNLOCK(vp);
3340 mp->mnt_syncer = vp;
3341 return (0);
3342 }
3343
3344 /*
3345 * Do a lazy sync of the filesystem.
3346 */
3347 static int
3348 sync_fsync(struct vop_fsync_args *ap)
3349 {
3350 struct vnode *syncvp = ap->a_vp;
3351 struct mount *mp = syncvp->v_mount;
3352 struct thread *td = ap->a_td;
3353 int error;
3354 struct bufobj *bo;
3355
3356 /*
3357 * We only need to do something if this is a lazy evaluation.
3358 */
3359 if (ap->a_waitfor != MNT_LAZY)
3360 return (0);
3361
3362 /*
3363 * Move ourselves to the back of the sync list.
3364 */
3365 bo = &syncvp->v_bufobj;
3366 BO_LOCK(bo);
3367 vn_syncer_add_to_worklist(bo, syncdelay);
3368 BO_UNLOCK(bo);
3369
3370 /*
3371 * Walk the list of vnodes pushing all that are dirty and
3372 * not already on the sync list.
3373 */
3374 mtx_lock(&mountlist_mtx);
3375 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3376 mtx_unlock(&mountlist_mtx);
3377 return (0);
3378 }
3379 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3380 vfs_unbusy(mp, td);
3381 return (0);
3382 }
3383 MNT_ILOCK(mp);
3384 mp->mnt_noasync++;
3385 mp->mnt_kern_flag &= ~MNTK_ASYNC;
3386 MNT_IUNLOCK(mp);
3387 vfs_msync(mp, MNT_NOWAIT);
3388 error = VFS_SYNC(mp, MNT_LAZY, td);
3389 MNT_ILOCK(mp);
3390 mp->mnt_noasync--;
3391 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3392 mp->mnt_kern_flag |= MNTK_ASYNC;
3393 MNT_IUNLOCK(mp);
3394 vn_finished_write(mp);
3395 vfs_unbusy(mp, td);
3396 return (error);
3397 }
3398
3399 /*
3400 * The syncer vnode is no referenced.
3401 */
3402 static int
3403 sync_inactive(struct vop_inactive_args *ap)
3404 {
3405
3406 vgone(ap->a_vp);
3407 return (0);
3408 }
3409
3410 /*
3411 * The syncer vnode is no longer needed and is being decommissioned.
3412 *
3413 * Modifications to the worklist must be protected by sync_mtx.
3414 */
3415 static int
3416 sync_reclaim(struct vop_reclaim_args *ap)
3417 {
3418 struct vnode *vp = ap->a_vp;
3419 struct bufobj *bo;
3420
3421 VI_LOCK(vp);
3422 bo = &vp->v_bufobj;
3423 vp->v_mount->mnt_syncer = NULL;
3424 if (bo->bo_flag & BO_ONWORKLST) {
3425 mtx_lock(&sync_mtx);
3426 LIST_REMOVE(bo, bo_synclist);
3427 syncer_worklist_len--;
3428 sync_vnode_count--;
3429 mtx_unlock(&sync_mtx);
3430 bo->bo_flag &= ~BO_ONWORKLST;
3431 }
3432 VI_UNLOCK(vp);
3433
3434 return (0);
3435 }
3436
3437 /*
3438 * Check if vnode represents a disk device
3439 */
3440 int
3441 vn_isdisk(struct vnode *vp, int *errp)
3442 {
3443 int error;
3444
3445 error = 0;
3446 dev_lock();
3447 if (vp->v_type != VCHR)
3448 error = ENOTBLK;
3449 else if (vp->v_rdev == NULL)
3450 error = ENXIO;
3451 else if (vp->v_rdev->si_devsw == NULL)
3452 error = ENXIO;
3453 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3454 error = ENOTBLK;
3455 dev_unlock();
3456 if (errp != NULL)
3457 *errp = error;
3458 return (error == 0);
3459 }
3460
3461 /*
3462 * Common filesystem object access control check routine. Accepts a
3463 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3464 * and optional call-by-reference privused argument allowing vaccess()
3465 * to indicate to the caller whether privilege was used to satisfy the
3466 * request (obsoleted). Returns 0 on success, or an errno on failure.
3467 *
3468 * The ifdef'd CAPABILITIES version is here for reference, but is not
3469 * actually used.
3470 */
3471 int
3472 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3473 mode_t acc_mode, struct ucred *cred, int *privused)
3474 {
3475 mode_t dac_granted;
3476 mode_t priv_granted;
3477
3478 /*
3479 * Look for a normal, non-privileged way to access the file/directory
3480 * as requested. If it exists, go with that.
3481 */
3482
3483 if (privused != NULL)
3484 *privused = 0;
3485
3486 dac_granted = 0;
3487
3488 /* Check the owner. */
3489 if (cred->cr_uid == file_uid) {
3490 dac_granted |= VADMIN;
3491 if (file_mode & S_IXUSR)
3492 dac_granted |= VEXEC;
3493 if (file_mode & S_IRUSR)
3494 dac_granted |= VREAD;
3495 if (file_mode & S_IWUSR)
3496 dac_granted |= (VWRITE | VAPPEND);
3497
3498 if ((acc_mode & dac_granted) == acc_mode)
3499 return (0);
3500
3501 goto privcheck;
3502 }
3503
3504 /* Otherwise, check the groups (first match) */
3505 if (groupmember(file_gid, cred)) {
3506 if (file_mode & S_IXGRP)
3507 dac_granted |= VEXEC;
3508 if (file_mode & S_IRGRP)
3509 dac_granted |= VREAD;
3510 if (file_mode & S_IWGRP)
3511 dac_granted |= (VWRITE | VAPPEND);
3512
3513 if ((acc_mode & dac_granted) == acc_mode)
3514 return (0);
3515
3516 goto privcheck;
3517 }
3518
3519 /* Otherwise, check everyone else. */
3520 if (file_mode & S_IXOTH)
3521 dac_granted |= VEXEC;
3522 if (file_mode & S_IROTH)
3523 dac_granted |= VREAD;
3524 if (file_mode & S_IWOTH)
3525 dac_granted |= (VWRITE | VAPPEND);
3526 if ((acc_mode & dac_granted) == acc_mode)
3527 return (0);
3528
3529 privcheck:
3530 /*
3531 * Build a privilege mask to determine if the set of privileges
3532 * satisfies the requirements when combined with the granted mask
3533 * from above. For each privilege, if the privilege is required,
3534 * bitwise or the request type onto the priv_granted mask.
3535 */
3536 priv_granted = 0;
3537
3538 if (type == VDIR) {
3539 /*
3540 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3541 * requests, instead of PRIV_VFS_EXEC.
3542 */
3543 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3544 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3545 priv_granted |= VEXEC;
3546 } else {
3547 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3548 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3549 priv_granted |= VEXEC;
3550 }
3551
3552 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3553 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3554 priv_granted |= VREAD;
3555
3556 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3557 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3558 priv_granted |= (VWRITE | VAPPEND);
3559
3560 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3561 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3562 priv_granted |= VADMIN;
3563
3564 if ((acc_mode & (priv_granted | dac_granted)) == acc_mode) {
3565 /* XXX audit: privilege used */
3566 if (privused != NULL)
3567 *privused = 1;
3568 return (0);
3569 }
3570
3571 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3572 }
3573
3574 /*
3575 * Credential check based on process requesting service, and per-attribute
3576 * permissions.
3577 */
3578 int
3579 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3580 struct thread *td, int access)
3581 {
3582
3583 /*
3584 * Kernel-invoked always succeeds.
3585 */
3586 if (cred == NOCRED)
3587 return (0);
3588
3589 /*
3590 * Do not allow privileged processes in jail to directly manipulate
3591 * system attributes.
3592 */
3593 switch (attrnamespace) {
3594 case EXTATTR_NAMESPACE_SYSTEM:
3595 /* Potentially should be: return (EPERM); */
3596 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3597 case EXTATTR_NAMESPACE_USER:
3598 return (VOP_ACCESS(vp, access, cred, td));
3599 default:
3600 return (EPERM);
3601 }
3602 }
3603
3604 #ifdef DEBUG_VFS_LOCKS
3605 /*
3606 * This only exists to supress warnings from unlocked specfs accesses. It is
3607 * no longer ok to have an unlocked VFS.
3608 */
3609 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3610
3611 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3612 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3613
3614 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3615 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3616
3617 int vfs_badlock_print = 1; /* Print lock violations. */
3618 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3619
3620 #ifdef KDB
3621 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3622 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3623 #endif
3624
3625 static void
3626 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3627 {
3628
3629 #ifdef KDB
3630 if (vfs_badlock_backtrace)
3631 kdb_backtrace();
3632 #endif
3633 if (vfs_badlock_print)
3634 printf("%s: %p %s\n", str, (void *)vp, msg);
3635 if (vfs_badlock_ddb)
3636 kdb_enter_why(KDB_WHY_VFSLOCK, "lock violation");
3637 }
3638
3639 void
3640 assert_vi_locked(struct vnode *vp, const char *str)
3641 {
3642
3643 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3644 vfs_badlock("interlock is not locked but should be", str, vp);
3645 }
3646
3647 void
3648 assert_vi_unlocked(struct vnode *vp, const char *str)
3649 {
3650
3651 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3652 vfs_badlock("interlock is locked but should not be", str, vp);
3653 }
3654
3655 void
3656 assert_vop_locked(struct vnode *vp, const char *str)
3657 {
3658
3659 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
3660 vfs_badlock("is not locked but should be", str, vp);
3661 }
3662
3663 void
3664 assert_vop_unlocked(struct vnode *vp, const char *str)
3665 {
3666
3667 if (vp && !IGNORE_LOCK(vp) &&
3668 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
3669 vfs_badlock("is locked but should not be", str, vp);
3670 }
3671
3672 void
3673 assert_vop_elocked(struct vnode *vp, const char *str)
3674 {
3675
3676 if (vp && !IGNORE_LOCK(vp) &&
3677 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
3678 vfs_badlock("is not exclusive locked but should be", str, vp);
3679 }
3680
3681 #if 0
3682 void
3683 assert_vop_elocked_other(struct vnode *vp, const char *str)
3684 {
3685
3686 if (vp && !IGNORE_LOCK(vp) &&
3687 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
3688 vfs_badlock("is not exclusive locked by another thread",
3689 str, vp);
3690 }
3691
3692 void
3693 assert_vop_slocked(struct vnode *vp, const char *str)
3694 {
3695
3696 if (vp && !IGNORE_LOCK(vp) &&
3697 VOP_ISLOCKED(vp, curthread) != LK_SHARED)
3698 vfs_badlock("is not locked shared but should be", str, vp);
3699 }
3700 #endif /* 0 */
3701 #endif /* DEBUG_VFS_LOCKS */
3702
3703 void
3704 vop_rename_pre(void *ap)
3705 {
3706 struct vop_rename_args *a = ap;
3707
3708 #ifdef DEBUG_VFS_LOCKS
3709 if (a->a_tvp)
3710 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3711 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3712 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3713 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3714
3715 /* Check the source (from). */
3716 if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3717 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3718 if (a->a_tvp != a->a_fvp)
3719 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3720
3721 /* Check the target. */
3722 if (a->a_tvp)
3723 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3724 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3725 #endif
3726 if (a->a_tdvp != a->a_fdvp)
3727 vhold(a->a_fdvp);
3728 if (a->a_tvp != a->a_fvp)
3729 vhold(a->a_fvp);
3730 vhold(a->a_tdvp);
3731 if (a->a_tvp)
3732 vhold(a->a_tvp);
3733 }
3734
3735 void
3736 vop_strategy_pre(void *ap)
3737 {
3738 #ifdef DEBUG_VFS_LOCKS
3739 struct vop_strategy_args *a;
3740 struct buf *bp;
3741
3742 a = ap;
3743 bp = a->a_bp;
3744
3745 /*
3746 * Cluster ops lock their component buffers but not the IO container.
3747 */
3748 if ((bp->b_flags & B_CLUSTER) != 0)
3749 return;
3750
3751 if (BUF_REFCNT(bp) < 1) {
3752 if (vfs_badlock_print)
3753 printf(
3754 "VOP_STRATEGY: bp is not locked but should be\n");
3755 if (vfs_badlock_ddb)
3756 kdb_enter_why(KDB_WHY_VFSLOCK, "lock violation");
3757 }
3758 #endif
3759 }
3760
3761 void
3762 vop_lookup_pre(void *ap)
3763 {
3764 #ifdef DEBUG_VFS_LOCKS
3765 struct vop_lookup_args *a;
3766 struct vnode *dvp;
3767
3768 a = ap;
3769 dvp = a->a_dvp;
3770 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3771 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3772 #endif
3773 }
3774
3775 void
3776 vop_lookup_post(void *ap, int rc)
3777 {
3778 #ifdef DEBUG_VFS_LOCKS
3779 struct vop_lookup_args *a;
3780 struct vnode *dvp;
3781 struct vnode *vp;
3782
3783 a = ap;
3784 dvp = a->a_dvp;
3785 vp = *(a->a_vpp);
3786
3787 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3788 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3789
3790 if (!rc)
3791 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3792 #endif
3793 }
3794
3795 void
3796 vop_lock_pre(void *ap)
3797 {
3798 #ifdef DEBUG_VFS_LOCKS
3799 struct vop_lock1_args *a = ap;
3800
3801 if ((a->a_flags & LK_INTERLOCK) == 0)
3802 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3803 else
3804 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3805 #endif
3806 }
3807
3808 void
3809 vop_lock_post(void *ap, int rc)
3810 {
3811 #ifdef DEBUG_VFS_LOCKS
3812 struct vop_lock1_args *a = ap;
3813
3814 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3815 if (rc == 0)
3816 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3817 #endif
3818 }
3819
3820 void
3821 vop_unlock_pre(void *ap)
3822 {
3823 #ifdef DEBUG_VFS_LOCKS
3824 struct vop_unlock_args *a = ap;
3825
3826 if (a->a_flags & LK_INTERLOCK)
3827 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3828 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3829 #endif
3830 }
3831
3832 void
3833 vop_unlock_post(void *ap, int rc)
3834 {
3835 #ifdef DEBUG_VFS_LOCKS
3836 struct vop_unlock_args *a = ap;
3837
3838 if (a->a_flags & LK_INTERLOCK)
3839 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3840 #endif
3841 }
3842
3843 void
3844 vop_create_post(void *ap, int rc)
3845 {
3846 struct vop_create_args *a = ap;
3847
3848 if (!rc)
3849 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3850 }
3851
3852 void
3853 vop_link_post(void *ap, int rc)
3854 {
3855 struct vop_link_args *a = ap;
3856
3857 if (!rc) {
3858 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3859 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3860 }
3861 }
3862
3863 void
3864 vop_mkdir_post(void *ap, int rc)
3865 {
3866 struct vop_mkdir_args *a = ap;
3867
3868 if (!rc)
3869 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3870 }
3871
3872 void
3873 vop_mknod_post(void *ap, int rc)
3874 {
3875 struct vop_mknod_args *a = ap;
3876
3877 if (!rc)
3878 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3879 }
3880
3881 void
3882 vop_remove_post(void *ap, int rc)
3883 {
3884 struct vop_remove_args *a = ap;
3885
3886 if (!rc) {
3887 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3888 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3889 }
3890 }
3891
3892 void
3893 vop_rename_post(void *ap, int rc)
3894 {
3895 struct vop_rename_args *a = ap;
3896
3897 if (!rc) {
3898 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3899 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3900 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3901 if (a->a_tvp)
3902 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3903 }
3904 if (a->a_tdvp != a->a_fdvp)
3905 vdrop(a->a_fdvp);
3906 if (a->a_tvp != a->a_fvp)
3907 vdrop(a->a_fvp);
3908 vdrop(a->a_tdvp);
3909 if (a->a_tvp)
3910 vdrop(a->a_tvp);
3911 }
3912
3913 void
3914 vop_rmdir_post(void *ap, int rc)
3915 {
3916 struct vop_rmdir_args *a = ap;
3917
3918 if (!rc) {
3919 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3920 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3921 }
3922 }
3923
3924 void
3925 vop_setattr_post(void *ap, int rc)
3926 {
3927 struct vop_setattr_args *a = ap;
3928
3929 if (!rc)
3930 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3931 }
3932
3933 void
3934 vop_symlink_post(void *ap, int rc)
3935 {
3936 struct vop_symlink_args *a = ap;
3937
3938 if (!rc)
3939 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3940 }
3941
3942 static struct knlist fs_knlist;
3943
3944 static void
3945 vfs_event_init(void *arg)
3946 {
3947 knlist_init_mtx(&fs_knlist, NULL);
3948 }
3949 /* XXX - correct order? */
3950 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3951
3952 void
3953 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3954 {
3955
3956 KNOTE_UNLOCKED(&fs_knlist, event);
3957 }
3958
3959 static int filt_fsattach(struct knote *kn);
3960 static void filt_fsdetach(struct knote *kn);
3961 static int filt_fsevent(struct knote *kn, long hint);
3962
3963 struct filterops fs_filtops =
3964 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3965
3966 static int
3967 filt_fsattach(struct knote *kn)
3968 {
3969
3970 kn->kn_flags |= EV_CLEAR;
3971 knlist_add(&fs_knlist, kn, 0);
3972 return (0);
3973 }
3974
3975 static void
3976 filt_fsdetach(struct knote *kn)
3977 {
3978
3979 knlist_remove(&fs_knlist, kn, 0);
3980 }
3981
3982 static int
3983 filt_fsevent(struct knote *kn, long hint)
3984 {
3985
3986 kn->kn_fflags |= hint;
3987 return (kn->kn_fflags != 0);
3988 }
3989
3990 static int
3991 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
3992 {
3993 struct vfsidctl vc;
3994 int error;
3995 struct mount *mp;
3996
3997 error = SYSCTL_IN(req, &vc, sizeof(vc));
3998 if (error)
3999 return (error);
4000 if (vc.vc_vers != VFS_CTL_VERS1)
4001 return (EINVAL);
4002 mp = vfs_getvfs(&vc.vc_fsid);
4003 if (mp == NULL)
4004 return (ENOENT);
4005 /* ensure that a specific sysctl goes to the right filesystem. */
4006 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4007 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4008 vfs_rel(mp);
4009 return (EINVAL);
4010 }
4011 VCTLTOREQ(&vc, req);
4012 error = VFS_SYSCTL(mp, vc.vc_op, req);
4013 vfs_rel(mp);
4014 return (error);
4015 }
4016
4017 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4018 "Sysctl by fsid");
4019
4020 /*
4021 * Function to initialize a va_filerev field sensibly.
4022 * XXX: Wouldn't a random number make a lot more sense ??
4023 */
4024 u_quad_t
4025 init_va_filerev(void)
4026 {
4027 struct bintime bt;
4028
4029 getbinuptime(&bt);
4030 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4031 }
4032
4033 static int filt_vfsread(struct knote *kn, long hint);
4034 static int filt_vfswrite(struct knote *kn, long hint);
4035 static int filt_vfsvnode(struct knote *kn, long hint);
4036 static void filt_vfsdetach(struct knote *kn);
4037 static struct filterops vfsread_filtops =
4038 { 1, NULL, filt_vfsdetach, filt_vfsread };
4039 static struct filterops vfswrite_filtops =
4040 { 1, NULL, filt_vfsdetach, filt_vfswrite };
4041 static struct filterops vfsvnode_filtops =
4042 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
4043
4044 static void
4045 vfs_knllock(void *arg)
4046 {
4047 struct vnode *vp = arg;
4048
4049 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
4050 }
4051
4052 static void
4053 vfs_knlunlock(void *arg)
4054 {
4055 struct vnode *vp = arg;
4056
4057 VOP_UNLOCK(vp, 0, curthread);
4058 }
4059
4060 int
4061 vfs_kqfilter(struct vop_kqfilter_args *ap)
4062 {
4063 struct vnode *vp = ap->a_vp;
4064 struct knote *kn = ap->a_kn;
4065 struct knlist *knl;
4066
4067 switch (kn->kn_filter) {
4068 case EVFILT_READ:
4069 kn->kn_fop = &vfsread_filtops;
4070 break;
4071 case EVFILT_WRITE:
4072 kn->kn_fop = &vfswrite_filtops;
4073 break;
4074 case EVFILT_VNODE:
4075 kn->kn_fop = &vfsvnode_filtops;
4076 break;
4077 default:
4078 return (EINVAL);
4079 }
4080
4081 kn->kn_hook = (caddr_t)vp;
4082
4083 v_addpollinfo(vp);
4084 if (vp->v_pollinfo == NULL)
4085 return (ENOMEM);
4086 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4087 knlist_add(knl, kn, 0);
4088
4089 return (0);
4090 }
4091
4092 /*
4093 * Detach knote from vnode
4094 */
4095 static void
4096 filt_vfsdetach(struct knote *kn)
4097 {
4098 struct vnode *vp = (struct vnode *)kn->kn_hook;
4099
4100 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4101 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4102 }
4103
4104 /*ARGSUSED*/
4105 static int
4106 filt_vfsread(struct knote *kn, long hint)
4107 {
4108 struct vnode *vp = (struct vnode *)kn->kn_hook;
4109 struct vattr va;
4110 int res;
4111
4112 /*
4113 * filesystem is gone, so set the EOF flag and schedule
4114 * the knote for deletion.
4115 */
4116 if (hint == NOTE_REVOKE) {
4117 VI_LOCK(vp);
4118 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4119 VI_UNLOCK(vp);
4120 return (1);
4121 }
4122
4123 if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
4124 return (0);
4125
4126 VI_LOCK(vp);
4127 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4128 res = (kn->kn_data != 0);
4129 VI_UNLOCK(vp);
4130 return (res);
4131 }
4132
4133 /*ARGSUSED*/
4134 static int
4135 filt_vfswrite(struct knote *kn, long hint)
4136 {
4137 struct vnode *vp = (struct vnode *)kn->kn_hook;
4138
4139 VI_LOCK(vp);
4140
4141 /*
4142 * filesystem is gone, so set the EOF flag and schedule
4143 * the knote for deletion.
4144 */
4145 if (hint == NOTE_REVOKE)
4146 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4147
4148 kn->kn_data = 0;
4149 VI_UNLOCK(vp);
4150 return (1);
4151 }
4152
4153 static int
4154 filt_vfsvnode(struct knote *kn, long hint)
4155 {
4156 struct vnode *vp = (struct vnode *)kn->kn_hook;
4157 int res;
4158
4159 VI_LOCK(vp);
4160 if (kn->kn_sfflags & hint)
4161 kn->kn_fflags |= hint;
4162 if (hint == NOTE_REVOKE) {
4163 kn->kn_flags |= EV_EOF;
4164 VI_UNLOCK(vp);
4165 return (1);
4166 }
4167 res = (kn->kn_fflags != 0);
4168 VI_UNLOCK(vp);
4169 return (res);
4170 }
4171
4172 int
4173 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4174 {
4175 int error;
4176
4177 if (dp->d_reclen > ap->a_uio->uio_resid)
4178 return (ENAMETOOLONG);
4179 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4180 if (error) {
4181 if (ap->a_ncookies != NULL) {
4182 if (ap->a_cookies != NULL)
4183 free(ap->a_cookies, M_TEMP);
4184 ap->a_cookies = NULL;
4185 *ap->a_ncookies = 0;
4186 }
4187 return (error);
4188 }
4189 if (ap->a_ncookies == NULL)
4190 return (0);
4191
4192 KASSERT(ap->a_cookies,
4193 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4194
4195 *ap->a_cookies = realloc(*ap->a_cookies,
4196 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4197 (*ap->a_cookies)[*ap->a_ncookies] = off;
4198 return (0);
4199 }
4200
4201 /*
4202 * Mark for update the access time of the file if the filesystem
4203 * supports VA_MARK_ATIME. This functionality is used by execve
4204 * and mmap, so we want to avoid the synchronous I/O implied by
4205 * directly setting va_atime for the sake of efficiency.
4206 */
4207 void
4208 vfs_mark_atime(struct vnode *vp, struct thread *td)
4209 {
4210 struct vattr atimeattr;
4211
4212 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
4213 VATTR_NULL(&atimeattr);
4214 atimeattr.va_vaflags |= VA_MARK_ATIME;
4215 (void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td);
4216 }
4217 }
Cache object: 62823cb129e44238fed487424a672fa9
|