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("$FreeBSD$");
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 void delmntque(struct vnode *vp);
93 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
94 int slpflag, int slptimeo);
95 static void syncer_shutdown(void *arg, int howto);
96 static int vtryrecycle(struct vnode *vp);
97 static void vbusy(struct vnode *vp);
98 static void vinactive(struct vnode *, struct thread *);
99 static void v_incr_usecount(struct vnode *);
100 static void v_decr_usecount(struct vnode *);
101 static void v_decr_useonly(struct vnode *);
102 static void v_upgrade_usecount(struct vnode *);
103 static void vfree(struct vnode *);
104 static void vnlru_free(int);
105 static void vgonel(struct vnode *);
106 static void vfs_knllock(void *arg);
107 static void vfs_knlunlock(void *arg);
108 static void destroy_vpollinfo(struct vpollinfo *vi);
109
110 /*
111 * Enable Giant pushdown based on whether or not the vm is mpsafe in this
112 * build. Without mpsafevm the buffer cache can not run Giant free.
113 */
114 int mpsafe_vfs = 1;
115 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
116 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
117 "MPSAFE VFS");
118
119 /*
120 * Number of vnodes in existence. Increased whenever getnewvnode()
121 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
122 * vnode.
123 */
124 static unsigned long numvnodes;
125
126 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
127
128 /*
129 * Conversion tables for conversion from vnode types to inode formats
130 * and back.
131 */
132 enum vtype iftovt_tab[16] = {
133 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
134 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
135 };
136 int vttoif_tab[10] = {
137 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
138 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
139 };
140
141 /*
142 * List of vnodes that are ready for recycling.
143 */
144 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
145
146 /*
147 * Free vnode target. Free vnodes may simply be files which have been stat'd
148 * but not read. This is somewhat common, and a small cache of such files
149 * should be kept to avoid recreation costs.
150 */
151 static u_long wantfreevnodes;
152 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
153 /* Number of vnodes in the free list. */
154 static u_long freevnodes;
155 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
156
157 /*
158 * Various variables used for debugging the new implementation of
159 * reassignbuf().
160 * XXX these are probably of (very) limited utility now.
161 */
162 static int reassignbufcalls;
163 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
164
165 /*
166 * Cache for the mount type id assigned to NFS. This is used for
167 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
168 */
169 int nfs_mount_type = -1;
170
171 /* To keep more than one thread at a time from running vfs_getnewfsid */
172 static struct mtx mntid_mtx;
173
174 /*
175 * Lock for any access to the following:
176 * vnode_free_list
177 * numvnodes
178 * freevnodes
179 */
180 static struct mtx vnode_free_list_mtx;
181
182 /* Publicly exported FS */
183 struct nfs_public nfs_pub;
184
185 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
186 static uma_zone_t vnode_zone;
187 static uma_zone_t vnodepoll_zone;
188
189 /* Set to 1 to print out reclaim of active vnodes */
190 int prtactive;
191
192 /*
193 * The workitem queue.
194 *
195 * It is useful to delay writes of file data and filesystem metadata
196 * for tens of seconds so that quickly created and deleted files need
197 * not waste disk bandwidth being created and removed. To realize this,
198 * we append vnodes to a "workitem" queue. When running with a soft
199 * updates implementation, most pending metadata dependencies should
200 * not wait for more than a few seconds. Thus, mounted on block devices
201 * are delayed only about a half the time that file data is delayed.
202 * Similarly, directory updates are more critical, so are only delayed
203 * about a third the time that file data is delayed. Thus, there are
204 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
205 * one each second (driven off the filesystem syncer process). The
206 * syncer_delayno variable indicates the next queue that is to be processed.
207 * Items that need to be processed soon are placed in this queue:
208 *
209 * syncer_workitem_pending[syncer_delayno]
210 *
211 * A delay of fifteen seconds is done by placing the request fifteen
212 * entries later in the queue:
213 *
214 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
215 *
216 */
217 static int syncer_delayno;
218 static long syncer_mask;
219 LIST_HEAD(synclist, bufobj);
220 static struct synclist *syncer_workitem_pending;
221 /*
222 * The sync_mtx protects:
223 * bo->bo_synclist
224 * sync_vnode_count
225 * syncer_delayno
226 * syncer_state
227 * syncer_workitem_pending
228 * syncer_worklist_len
229 * rushjob
230 */
231 static struct mtx sync_mtx;
232
233 #define SYNCER_MAXDELAY 32
234 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
235 static int syncdelay = 30; /* max time to delay syncing data */
236 static int filedelay = 30; /* time to delay syncing files */
237 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
238 static int dirdelay = 29; /* time to delay syncing directories */
239 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
240 static int metadelay = 28; /* time to delay syncing metadata */
241 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
242 static int rushjob; /* number of slots to run ASAP */
243 static int stat_rush_requests; /* number of times I/O speeded up */
244 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
245
246 /*
247 * When shutting down the syncer, run it at four times normal speed.
248 */
249 #define SYNCER_SHUTDOWN_SPEEDUP 4
250 static int sync_vnode_count;
251 static int syncer_worklist_len;
252 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
253 syncer_state;
254
255 /*
256 * Number of vnodes we want to exist at any one time. This is mostly used
257 * to size hash tables in vnode-related code. It is normally not used in
258 * getnewvnode(), as wantfreevnodes is normally nonzero.)
259 *
260 * XXX desiredvnodes is historical cruft and should not exist.
261 */
262 int desiredvnodes;
263 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
264 &desiredvnodes, 0, "Maximum number of vnodes");
265 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
266 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
267 static int vnlru_nowhere;
268 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
269 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
270
271 /*
272 * Macros to control when a vnode is freed and recycled. All require
273 * the vnode interlock.
274 */
275 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
276 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
277 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
278
279
280 /*
281 * Initialize the vnode management data structures.
282 *
283 * Reevaluate the following cap on the number of vnodes after the physical
284 * memory size exceeds 512GB. In the limit, as the physical memory size
285 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
286 */
287 #ifndef MAXVNODES_MAX
288 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
289 #endif
290 static void
291 vntblinit(void *dummy __unused)
292 {
293 int physvnodes, virtvnodes;
294
295 /*
296 * Desiredvnodes is a function of the physical memory size and the
297 * kernel's heap size. Generally speaking, it scales with the
298 * physical memory size. The ratio of desiredvnodes to physical pages
299 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
300 * marginal ratio of desiredvnodes to physical pages is one to
301 * sixteen. However, desiredvnodes is limited by the kernel's heap
302 * size. The memory required by desiredvnodes vnodes and vm objects
303 * may not exceed one seventh of the kernel's heap size.
304 */
305 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
306 cnt.v_page_count) / 16;
307 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
308 sizeof(struct vnode)));
309 desiredvnodes = min(physvnodes, virtvnodes);
310 if (desiredvnodes > MAXVNODES_MAX) {
311 if (bootverbose)
312 printf("Reducing kern.maxvnodes %d -> %d\n",
313 desiredvnodes, MAXVNODES_MAX);
314 desiredvnodes = MAXVNODES_MAX;
315 }
316 wantfreevnodes = desiredvnodes / 4;
317 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
318 TAILQ_INIT(&vnode_free_list);
319 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
320 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
321 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
322 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
323 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
324 /*
325 * Initialize the filesystem syncer.
326 */
327 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
328 &syncer_mask);
329 syncer_maxdelay = syncer_mask + 1;
330 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
331 }
332 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
333
334
335 /*
336 * Mark a mount point as busy. Used to synchronize access and to delay
337 * unmounting. Interlock is not released on failure.
338 *
339 * vfs_busy() only sleeps if the unmount is active on the mount point.
340 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
341 * vnode belonging to mp.
342 *
343 * Lookup uses vfs_busy() to traverse mount points.
344 * root fs var fs
345 * / vnode lock A / vnode lock (/var) D
346 * /var vnode lock B /log vnode lock(/var/log) E
347 * vfs_busy lock C vfs_busy lock F
348 *
349 * Within each file system, the lock order is C->A->B and F->D->E.
350 *
351 * When traversing across mounts, the system follows that lock order:
352 *
353 * C->A->B
354 * |
355 * +->F->D->E
356 *
357 * The lookup() process for namei("/var") illustrates the process:
358 * VOP_LOOKUP() obtains B while A is held
359 * vfs_busy() obtains a shared lock on F while A and B are held
360 * vput() releases lock on B
361 * vput() releases lock on A
362 * VFS_ROOT() obtains lock on D while shared lock on F is held
363 * vfs_unbusy() releases shared lock on F
364 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
365 * Attempt to lock A (instead of vp_crossmp) while D is held would
366 * violate the global order, causing deadlocks.
367 *
368 * dounmount() locks B while F is drained.
369 */
370 int
371 vfs_busy(struct mount *mp, int flags, struct mtx *interlkp,
372 struct thread *td)
373 {
374 int lkflags;
375
376 MNT_ILOCK(mp);
377 MNT_REF(mp);
378 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
379 if (flags & LK_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
380 MNT_REL(mp);
381 MNT_IUNLOCK(mp);
382 return (ENOENT);
383 }
384 if (interlkp)
385 mtx_unlock(interlkp);
386 mp->mnt_kern_flag |= MNTK_MWAIT;
387 /*
388 * Since all busy locks are shared except the exclusive
389 * lock granted when unmounting, the only place that a
390 * wakeup needs to be done is at the release of the
391 * exclusive lock at the end of dounmount.
392 */
393 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
394 if (interlkp)
395 mtx_lock(interlkp);
396 }
397 if (interlkp)
398 mtx_unlock(interlkp);
399 lkflags = LK_SHARED | LK_INTERLOCK | LK_NOWAIT;
400 if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp), td))
401 panic("vfs_busy: unexpected lock failure");
402 return (0);
403 }
404
405 /*
406 * Free a busy filesystem.
407 */
408 void
409 vfs_unbusy(struct mount *mp, struct thread *td)
410 {
411
412 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
413 vfs_rel(mp);
414 }
415
416 /*
417 * Lookup a mount point by filesystem identifier.
418 */
419 struct mount *
420 vfs_getvfs(fsid_t *fsid)
421 {
422 struct mount *mp;
423
424 mtx_lock(&mountlist_mtx);
425 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
426 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
427 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
428 vfs_ref(mp);
429 mtx_unlock(&mountlist_mtx);
430 return (mp);
431 }
432 }
433 mtx_unlock(&mountlist_mtx);
434 return ((struct mount *) 0);
435 }
436
437 /*
438 * Check if a user can access privileged mount options.
439 */
440 int
441 vfs_suser(struct mount *mp, struct thread *td)
442 {
443 int error;
444
445 /*
446 * If the thread is jailed, but this is not a jail-friendly file
447 * system, deny immediately.
448 */
449 if (jailed(td->td_ucred) && !(mp->mnt_vfc->vfc_flags & VFCF_JAIL))
450 return (EPERM);
451
452 /*
453 * If the file system was mounted outside a jail and a jailed thread
454 * tries to access it, deny immediately.
455 */
456 if (!jailed(mp->mnt_cred) && jailed(td->td_ucred))
457 return (EPERM);
458
459 /*
460 * If the file system was mounted inside different jail that the jail of
461 * the calling thread, deny immediately.
462 */
463 if (jailed(mp->mnt_cred) && jailed(td->td_ucred) &&
464 mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) {
465 return (EPERM);
466 }
467
468 if ((mp->mnt_flag & MNT_USER) == 0 ||
469 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
470 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
471 return (error);
472 }
473 return (0);
474 }
475
476 /*
477 * Get a new unique fsid. Try to make its val[0] unique, since this value
478 * will be used to create fake device numbers for stat(). Also try (but
479 * not so hard) make its val[0] unique mod 2^16, since some emulators only
480 * support 16-bit device numbers. We end up with unique val[0]'s for the
481 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
482 *
483 * Keep in mind that several mounts may be running in parallel. Starting
484 * the search one past where the previous search terminated is both a
485 * micro-optimization and a defense against returning the same fsid to
486 * different mounts.
487 */
488 void
489 vfs_getnewfsid(struct mount *mp)
490 {
491 static u_int16_t mntid_base;
492 struct mount *nmp;
493 fsid_t tfsid;
494 int mtype;
495
496 mtx_lock(&mntid_mtx);
497 mtype = mp->mnt_vfc->vfc_typenum;
498 tfsid.val[1] = mtype;
499 mtype = (mtype & 0xFF) << 24;
500 for (;;) {
501 tfsid.val[0] = makedev(255,
502 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
503 mntid_base++;
504 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
505 break;
506 vfs_rel(nmp);
507 }
508 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
509 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
510 mtx_unlock(&mntid_mtx);
511 }
512
513 /*
514 * Knob to control the precision of file timestamps:
515 *
516 * 0 = seconds only; nanoseconds zeroed.
517 * 1 = seconds and nanoseconds, accurate within 1/HZ.
518 * 2 = seconds and nanoseconds, truncated to microseconds.
519 * >=3 = seconds and nanoseconds, maximum precision.
520 */
521 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
522
523 static int timestamp_precision = TSP_SEC;
524 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
525 ×tamp_precision, 0, "");
526
527 /*
528 * Get a current timestamp.
529 */
530 void
531 vfs_timestamp(struct timespec *tsp)
532 {
533 struct timeval tv;
534
535 switch (timestamp_precision) {
536 case TSP_SEC:
537 tsp->tv_sec = time_second;
538 tsp->tv_nsec = 0;
539 break;
540 case TSP_HZ:
541 getnanotime(tsp);
542 break;
543 case TSP_USEC:
544 microtime(&tv);
545 TIMEVAL_TO_TIMESPEC(&tv, tsp);
546 break;
547 case TSP_NSEC:
548 default:
549 nanotime(tsp);
550 break;
551 }
552 }
553
554 /*
555 * Set vnode attributes to VNOVAL
556 */
557 void
558 vattr_null(struct vattr *vap)
559 {
560
561 vap->va_type = VNON;
562 vap->va_size = VNOVAL;
563 vap->va_bytes = VNOVAL;
564 vap->va_mode = VNOVAL;
565 vap->va_nlink = VNOVAL;
566 vap->va_uid = VNOVAL;
567 vap->va_gid = VNOVAL;
568 vap->va_fsid = VNOVAL;
569 vap->va_fileid = VNOVAL;
570 vap->va_blocksize = VNOVAL;
571 vap->va_rdev = VNOVAL;
572 vap->va_atime.tv_sec = VNOVAL;
573 vap->va_atime.tv_nsec = VNOVAL;
574 vap->va_mtime.tv_sec = VNOVAL;
575 vap->va_mtime.tv_nsec = VNOVAL;
576 vap->va_ctime.tv_sec = VNOVAL;
577 vap->va_ctime.tv_nsec = VNOVAL;
578 vap->va_birthtime.tv_sec = VNOVAL;
579 vap->va_birthtime.tv_nsec = VNOVAL;
580 vap->va_flags = VNOVAL;
581 vap->va_gen = VNOVAL;
582 vap->va_vaflags = 0;
583 }
584
585 /*
586 * This routine is called when we have too many vnodes. It attempts
587 * to free <count> vnodes and will potentially free vnodes that still
588 * have VM backing store (VM backing store is typically the cause
589 * of a vnode blowout so we want to do this). Therefore, this operation
590 * is not considered cheap.
591 *
592 * A number of conditions may prevent a vnode from being reclaimed.
593 * the buffer cache may have references on the vnode, a directory
594 * vnode may still have references due to the namei cache representing
595 * underlying files, or the vnode may be in active use. It is not
596 * desireable to reuse such vnodes. These conditions may cause the
597 * number of vnodes to reach some minimum value regardless of what
598 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
599 */
600 static int
601 vlrureclaim(struct mount *mp)
602 {
603 struct thread *td;
604 struct vnode *vp;
605 int done;
606 int trigger;
607 int usevnodes;
608 int count;
609
610 /*
611 * Calculate the trigger point, don't allow user
612 * screwups to blow us up. This prevents us from
613 * recycling vnodes with lots of resident pages. We
614 * aren't trying to free memory, we are trying to
615 * free vnodes.
616 */
617 usevnodes = desiredvnodes;
618 if (usevnodes <= 0)
619 usevnodes = 1;
620 trigger = cnt.v_page_count * 2 / usevnodes;
621 done = 0;
622 td = curthread;
623 vn_start_write(NULL, &mp, V_WAIT);
624 MNT_ILOCK(mp);
625 count = mp->mnt_nvnodelistsize / 10 + 1;
626 while (count != 0) {
627 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
628 while (vp != NULL && vp->v_type == VMARKER)
629 vp = TAILQ_NEXT(vp, v_nmntvnodes);
630 if (vp == NULL)
631 break;
632 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
633 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
634 --count;
635 if (!VI_TRYLOCK(vp))
636 goto next_iter;
637 /*
638 * If it's been deconstructed already, it's still
639 * referenced, or it exceeds the trigger, skip it.
640 */
641 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
642 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
643 vp->v_object->resident_page_count > trigger)) {
644 VI_UNLOCK(vp);
645 goto next_iter;
646 }
647 MNT_IUNLOCK(mp);
648 vholdl(vp);
649 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT, td)) {
650 vdrop(vp);
651 goto next_iter_mntunlocked;
652 }
653 VI_LOCK(vp);
654 /*
655 * v_usecount may have been bumped after VOP_LOCK() dropped
656 * the vnode interlock and before it was locked again.
657 *
658 * It is not necessary to recheck VI_DOOMED because it can
659 * only be set by another thread that holds both the vnode
660 * lock and vnode interlock. If another thread has the
661 * vnode lock before we get to VOP_LOCK() and obtains the
662 * vnode interlock after VOP_LOCK() drops the vnode
663 * interlock, the other thread will be unable to drop the
664 * vnode lock before our VOP_LOCK() call fails.
665 */
666 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
667 (vp->v_object != NULL &&
668 vp->v_object->resident_page_count > trigger)) {
669 VOP_UNLOCK(vp, LK_INTERLOCK, td);
670 goto next_iter_mntunlocked;
671 }
672 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
673 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
674 vgonel(vp);
675 VOP_UNLOCK(vp, 0, td);
676 vdropl(vp);
677 done++;
678 next_iter_mntunlocked:
679 if ((count % 256) != 0)
680 goto relock_mnt;
681 goto yield;
682 next_iter:
683 if ((count % 256) != 0)
684 continue;
685 MNT_IUNLOCK(mp);
686 yield:
687 uio_yield();
688 relock_mnt:
689 MNT_ILOCK(mp);
690 }
691 MNT_IUNLOCK(mp);
692 vn_finished_write(mp);
693 return done;
694 }
695
696 /*
697 * Attempt to keep the free list at wantfreevnodes length.
698 */
699 static void
700 vnlru_free(int count)
701 {
702 struct vnode *vp;
703 int vfslocked;
704
705 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
706 for (; count > 0; count--) {
707 vp = TAILQ_FIRST(&vnode_free_list);
708 /*
709 * The list can be modified while the free_list_mtx
710 * has been dropped and vp could be NULL here.
711 */
712 if (!vp)
713 break;
714 VNASSERT(vp->v_op != NULL, vp,
715 ("vnlru_free: vnode already reclaimed."));
716 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
717 /*
718 * Don't recycle if we can't get the interlock.
719 */
720 if (!VI_TRYLOCK(vp)) {
721 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
722 continue;
723 }
724 VNASSERT(VCANRECYCLE(vp), vp,
725 ("vp inconsistent on freelist"));
726 freevnodes--;
727 vp->v_iflag &= ~VI_FREE;
728 vholdl(vp);
729 mtx_unlock(&vnode_free_list_mtx);
730 VI_UNLOCK(vp);
731 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
732 vtryrecycle(vp);
733 VFS_UNLOCK_GIANT(vfslocked);
734 /*
735 * If the recycled succeeded this vdrop will actually free
736 * the vnode. If not it will simply place it back on
737 * the free list.
738 */
739 vdrop(vp);
740 mtx_lock(&vnode_free_list_mtx);
741 }
742 }
743 /*
744 * Attempt to recycle vnodes in a context that is always safe to block.
745 * Calling vlrurecycle() from the bowels of filesystem code has some
746 * interesting deadlock problems.
747 */
748 static struct proc *vnlruproc;
749 static int vnlruproc_sig;
750
751 static void
752 vnlru_proc(void)
753 {
754 struct mount *mp, *nmp;
755 int done, vfslocked;
756 struct proc *p = vnlruproc;
757 struct thread *td = curthread;
758
759 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
760 SHUTDOWN_PRI_FIRST);
761
762 for (;;) {
763 kthread_suspend_check(p);
764 mtx_lock(&vnode_free_list_mtx);
765 if (freevnodes > wantfreevnodes)
766 vnlru_free(freevnodes - wantfreevnodes);
767 if (numvnodes <= desiredvnodes * 9 / 10) {
768 vnlruproc_sig = 0;
769 wakeup(&vnlruproc_sig);
770 msleep(vnlruproc, &vnode_free_list_mtx,
771 PVFS|PDROP, "vlruwt", hz);
772 continue;
773 }
774 mtx_unlock(&vnode_free_list_mtx);
775 done = 0;
776 mtx_lock(&mountlist_mtx);
777 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
778 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
779 nmp = TAILQ_NEXT(mp, mnt_list);
780 continue;
781 }
782 vfslocked = VFS_LOCK_GIANT(mp);
783 done += vlrureclaim(mp);
784 VFS_UNLOCK_GIANT(vfslocked);
785 mtx_lock(&mountlist_mtx);
786 nmp = TAILQ_NEXT(mp, mnt_list);
787 vfs_unbusy(mp, td);
788 }
789 mtx_unlock(&mountlist_mtx);
790 if (done == 0) {
791 EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
792 #if 0
793 /* These messages are temporary debugging aids */
794 if (vnlru_nowhere < 5)
795 printf("vnlru process getting nowhere..\n");
796 else if (vnlru_nowhere == 5)
797 printf("vnlru process messages stopped.\n");
798 #endif
799 vnlru_nowhere++;
800 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
801 } else
802 uio_yield();
803 }
804 }
805
806 static struct kproc_desc vnlru_kp = {
807 "vnlru",
808 vnlru_proc,
809 &vnlruproc
810 };
811 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
812 &vnlru_kp);
813
814 /*
815 * Routines having to do with the management of the vnode table.
816 */
817
818 void
819 vdestroy(struct vnode *vp)
820 {
821 struct bufobj *bo;
822
823 CTR1(KTR_VFS, "vdestroy vp %p", vp);
824 mtx_lock(&vnode_free_list_mtx);
825 numvnodes--;
826 mtx_unlock(&vnode_free_list_mtx);
827 bo = &vp->v_bufobj;
828 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
829 ("cleaned vnode still on the free list."));
830 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
831 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
832 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
833 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
834 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
835 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
836 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
837 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
838 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
839 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
840 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
841 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
842 VI_UNLOCK(vp);
843 #ifdef MAC
844 mac_destroy_vnode(vp);
845 #endif
846 if (vp->v_pollinfo != NULL)
847 destroy_vpollinfo(vp->v_pollinfo);
848 #ifdef INVARIANTS
849 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
850 vp->v_op = NULL;
851 #endif
852 lockdestroy(vp->v_vnlock);
853 mtx_destroy(&vp->v_interlock);
854 uma_zfree(vnode_zone, vp);
855 }
856
857 /*
858 * Try to recycle a freed vnode. We abort if anyone picks up a reference
859 * before we actually vgone(). This function must be called with the vnode
860 * held to prevent the vnode from being returned to the free list midway
861 * through vgone().
862 */
863 static int
864 vtryrecycle(struct vnode *vp)
865 {
866 struct thread *td = curthread;
867 struct mount *vnmp;
868
869 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
870 VNASSERT(vp->v_holdcnt, vp,
871 ("vtryrecycle: Recycling vp %p without a reference.", vp));
872 /*
873 * This vnode may found and locked via some other list, if so we
874 * can't recycle it yet.
875 */
876 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
877 return (EWOULDBLOCK);
878 /*
879 * Don't recycle if its filesystem is being suspended.
880 */
881 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
882 VOP_UNLOCK(vp, 0, td);
883 return (EBUSY);
884 }
885 /*
886 * If we got this far, we need to acquire the interlock and see if
887 * anyone picked up this vnode from another list. If not, we will
888 * mark it with DOOMED via vgonel() so that anyone who does find it
889 * will skip over it.
890 */
891 VI_LOCK(vp);
892 if (vp->v_usecount) {
893 VOP_UNLOCK(vp, LK_INTERLOCK, td);
894 vn_finished_write(vnmp);
895 return (EBUSY);
896 }
897 if ((vp->v_iflag & VI_DOOMED) == 0)
898 vgonel(vp);
899 VOP_UNLOCK(vp, LK_INTERLOCK, td);
900 vn_finished_write(vnmp);
901 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
902 return (0);
903 }
904
905 /*
906 * Return the next vnode from the free list.
907 */
908 int
909 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
910 struct vnode **vpp)
911 {
912 struct vnode *vp = NULL;
913 struct bufobj *bo;
914
915 mtx_lock(&vnode_free_list_mtx);
916 /*
917 * Lend our context to reclaim vnodes if they've exceeded the max.
918 */
919 if (freevnodes > wantfreevnodes)
920 vnlru_free(1);
921 /*
922 * Wait for available vnodes.
923 */
924 if (numvnodes > desiredvnodes) {
925 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
926 /*
927 * File system is beeing suspended, we cannot risk a
928 * deadlock here, so allocate new vnode anyway.
929 */
930 if (freevnodes > wantfreevnodes)
931 vnlru_free(freevnodes - wantfreevnodes);
932 goto alloc;
933 }
934 if (vnlruproc_sig == 0) {
935 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
936 wakeup(vnlruproc);
937 }
938 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
939 "vlruwk", hz);
940 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
941 if (numvnodes > desiredvnodes) {
942 mtx_unlock(&vnode_free_list_mtx);
943 return (ENFILE);
944 }
945 #endif
946 }
947 alloc:
948 numvnodes++;
949 mtx_unlock(&vnode_free_list_mtx);
950 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
951 /*
952 * Setup locks.
953 */
954 vp->v_vnlock = &vp->v_lock;
955 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
956 /*
957 * By default, don't allow shared locks unless filesystems
958 * opt-in.
959 */
960 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
961 /*
962 * Initialize bufobj.
963 */
964 bo = &vp->v_bufobj;
965 bo->__bo_vnode = vp;
966 bo->bo_mtx = &vp->v_interlock;
967 bo->bo_ops = &buf_ops_bio;
968 bo->bo_private = vp;
969 TAILQ_INIT(&bo->bo_clean.bv_hd);
970 TAILQ_INIT(&bo->bo_dirty.bv_hd);
971 /*
972 * Initialize namecache.
973 */
974 LIST_INIT(&vp->v_cache_src);
975 TAILQ_INIT(&vp->v_cache_dst);
976 /*
977 * Finalize various vnode identity bits.
978 */
979 vp->v_type = VNON;
980 vp->v_tag = tag;
981 vp->v_op = vops;
982 v_incr_usecount(vp);
983 vp->v_data = 0;
984 #ifdef MAC
985 mac_init_vnode(vp);
986 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
987 mac_associate_vnode_singlelabel(mp, vp);
988 else if (mp == NULL && vops != &dead_vnodeops)
989 printf("NULL mp in getnewvnode()\n");
990 #endif
991 if (mp != NULL) {
992 bo->bo_bsize = mp->mnt_stat.f_iosize;
993 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
994 vp->v_vflag |= VV_NOKNOTE;
995 }
996
997 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
998 *vpp = vp;
999 return (0);
1000 }
1001
1002 /*
1003 * Delete from old mount point vnode list, if on one.
1004 */
1005 static void
1006 delmntque(struct vnode *vp)
1007 {
1008 struct mount *mp;
1009
1010 mp = vp->v_mount;
1011 if (mp == NULL)
1012 return;
1013 MNT_ILOCK(mp);
1014 vp->v_mount = NULL;
1015 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1016 ("bad mount point vnode list size"));
1017 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1018 mp->mnt_nvnodelistsize--;
1019 MNT_REL(mp);
1020 MNT_IUNLOCK(mp);
1021 }
1022
1023 static void
1024 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1025 {
1026 struct thread *td;
1027
1028 td = curthread; /* XXX ? */
1029 vp->v_data = NULL;
1030 vp->v_op = &dead_vnodeops;
1031 /* XXX non mp-safe fs may still call insmntque with vnode
1032 unlocked */
1033 if (!VOP_ISLOCKED(vp, td))
1034 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1035 vgone(vp);
1036 vput(vp);
1037 }
1038
1039 /*
1040 * Insert into list of vnodes for the new mount point, if available.
1041 */
1042 int
1043 insmntque1(struct vnode *vp, struct mount *mp,
1044 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1045 {
1046 int locked;
1047
1048 KASSERT(vp->v_mount == NULL,
1049 ("insmntque: vnode already on per mount vnode list"));
1050 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1051 #if 0
1052 #ifdef DEBUG_VFS_LOCKS
1053 if (!VFS_NEEDSGIANT(mp))
1054 ASSERT_VOP_ELOCKED(vp,
1055 "insmntque: mp-safe fs and non-locked vp");
1056 #endif
1057 #endif
1058 MNT_ILOCK(mp);
1059 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1060 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1061 mp->mnt_nvnodelistsize == 0)) {
1062 locked = VOP_ISLOCKED(vp, curthread);
1063 if (!locked || (locked == LK_EXCLUSIVE &&
1064 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1065 MNT_IUNLOCK(mp);
1066 if (dtr != NULL)
1067 dtr(vp, dtr_arg);
1068 return (EBUSY);
1069 }
1070 }
1071 vp->v_mount = mp;
1072 MNT_REF(mp);
1073 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1074 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1075 ("neg mount point vnode list size"));
1076 mp->mnt_nvnodelistsize++;
1077 MNT_IUNLOCK(mp);
1078 return (0);
1079 }
1080
1081 int
1082 insmntque(struct vnode *vp, struct mount *mp)
1083 {
1084
1085 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1086 }
1087
1088 /*
1089 * Flush out and invalidate all buffers associated with a bufobj
1090 * Called with the underlying object locked.
1091 */
1092 int
1093 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag,
1094 int slptimeo)
1095 {
1096 int error;
1097
1098 BO_LOCK(bo);
1099 if (flags & V_SAVE) {
1100 error = bufobj_wwait(bo, slpflag, slptimeo);
1101 if (error) {
1102 BO_UNLOCK(bo);
1103 return (error);
1104 }
1105 if (bo->bo_dirty.bv_cnt > 0) {
1106 BO_UNLOCK(bo);
1107 if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1108 return (error);
1109 /*
1110 * XXX We could save a lock/unlock if this was only
1111 * enabled under INVARIANTS
1112 */
1113 BO_LOCK(bo);
1114 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1115 panic("vinvalbuf: dirty bufs");
1116 }
1117 }
1118 /*
1119 * If you alter this loop please notice that interlock is dropped and
1120 * reacquired in flushbuflist. Special care is needed to ensure that
1121 * no race conditions occur from this.
1122 */
1123 do {
1124 error = flushbuflist(&bo->bo_clean,
1125 flags, bo, slpflag, slptimeo);
1126 if (error == 0)
1127 error = flushbuflist(&bo->bo_dirty,
1128 flags, bo, slpflag, slptimeo);
1129 if (error != 0 && error != EAGAIN) {
1130 BO_UNLOCK(bo);
1131 return (error);
1132 }
1133 } while (error != 0);
1134
1135 /*
1136 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1137 * have write I/O in-progress but if there is a VM object then the
1138 * VM object can also have read-I/O in-progress.
1139 */
1140 do {
1141 bufobj_wwait(bo, 0, 0);
1142 BO_UNLOCK(bo);
1143 if (bo->bo_object != NULL) {
1144 VM_OBJECT_LOCK(bo->bo_object);
1145 vm_object_pip_wait(bo->bo_object, "bovlbx");
1146 VM_OBJECT_UNLOCK(bo->bo_object);
1147 }
1148 BO_LOCK(bo);
1149 } while (bo->bo_numoutput > 0);
1150 BO_UNLOCK(bo);
1151
1152 /*
1153 * Destroy the copy in the VM cache, too.
1154 */
1155 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1156 VM_OBJECT_LOCK(bo->bo_object);
1157 vm_object_page_remove(bo->bo_object, 0, 0,
1158 (flags & V_SAVE) ? TRUE : FALSE);
1159 VM_OBJECT_UNLOCK(bo->bo_object);
1160 }
1161
1162 #ifdef INVARIANTS
1163 BO_LOCK(bo);
1164 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1165 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1166 panic("vinvalbuf: flush failed");
1167 BO_UNLOCK(bo);
1168 #endif
1169 return (0);
1170 }
1171
1172 /*
1173 * Flush out and invalidate all buffers associated with a vnode.
1174 * Called with the underlying object locked.
1175 */
1176 int
1177 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag,
1178 int slptimeo)
1179 {
1180
1181 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1182 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1183 return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1184 }
1185
1186 /*
1187 * Flush out buffers on the specified list.
1188 *
1189 */
1190 static int
1191 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1192 int slptimeo)
1193 {
1194 struct buf *bp, *nbp;
1195 int retval, error;
1196 daddr_t lblkno;
1197 b_xflags_t xflags;
1198
1199 ASSERT_BO_LOCKED(bo);
1200
1201 retval = 0;
1202 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1203 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1204 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1205 continue;
1206 }
1207 lblkno = 0;
1208 xflags = 0;
1209 if (nbp != NULL) {
1210 lblkno = nbp->b_lblkno;
1211 xflags = nbp->b_xflags &
1212 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1213 }
1214 retval = EAGAIN;
1215 error = BUF_TIMELOCK(bp,
1216 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1217 "flushbuf", slpflag, slptimeo);
1218 if (error) {
1219 BO_LOCK(bo);
1220 return (error != ENOLCK ? error : EAGAIN);
1221 }
1222 KASSERT(bp->b_bufobj == bo,
1223 ("bp %p wrong b_bufobj %p should be %p",
1224 bp, bp->b_bufobj, bo));
1225 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1226 BUF_UNLOCK(bp);
1227 BO_LOCK(bo);
1228 return (EAGAIN);
1229 }
1230 /*
1231 * XXX Since there are no node locks for NFS, I
1232 * believe there is a slight chance that a delayed
1233 * write will occur while sleeping just above, so
1234 * check for it.
1235 */
1236 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1237 (flags & V_SAVE)) {
1238 bremfree(bp);
1239 bp->b_flags |= B_ASYNC;
1240 bwrite(bp);
1241 BO_LOCK(bo);
1242 return (EAGAIN); /* XXX: why not loop ? */
1243 }
1244 bremfree(bp);
1245 bp->b_flags |= (B_INVAL | B_RELBUF);
1246 bp->b_flags &= ~B_ASYNC;
1247 brelse(bp);
1248 BO_LOCK(bo);
1249 if (nbp != NULL &&
1250 (nbp->b_bufobj != bo ||
1251 nbp->b_lblkno != lblkno ||
1252 (nbp->b_xflags &
1253 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1254 break; /* nbp invalid */
1255 }
1256 return (retval);
1257 }
1258
1259 /*
1260 * Truncate a file's buffer and pages to a specified length. This
1261 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1262 * sync activity.
1263 */
1264 int
1265 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1266 off_t length, int blksize)
1267 {
1268 struct buf *bp, *nbp;
1269 int anyfreed;
1270 int trunclbn;
1271 struct bufobj *bo;
1272
1273 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1274 /*
1275 * Round up to the *next* lbn.
1276 */
1277 trunclbn = (length + blksize - 1) / blksize;
1278
1279 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1280 restart:
1281 VI_LOCK(vp);
1282 bo = &vp->v_bufobj;
1283 anyfreed = 1;
1284 for (;anyfreed;) {
1285 anyfreed = 0;
1286 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1287 if (bp->b_lblkno < trunclbn)
1288 continue;
1289 if (BUF_LOCK(bp,
1290 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1291 VI_MTX(vp)) == ENOLCK)
1292 goto restart;
1293
1294 bremfree(bp);
1295 bp->b_flags |= (B_INVAL | B_RELBUF);
1296 bp->b_flags &= ~B_ASYNC;
1297 brelse(bp);
1298 anyfreed = 1;
1299
1300 if (nbp != NULL &&
1301 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1302 (nbp->b_vp != vp) ||
1303 (nbp->b_flags & B_DELWRI))) {
1304 goto restart;
1305 }
1306 VI_LOCK(vp);
1307 }
1308
1309 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1310 if (bp->b_lblkno < trunclbn)
1311 continue;
1312 if (BUF_LOCK(bp,
1313 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1314 VI_MTX(vp)) == ENOLCK)
1315 goto restart;
1316 bremfree(bp);
1317 bp->b_flags |= (B_INVAL | B_RELBUF);
1318 bp->b_flags &= ~B_ASYNC;
1319 brelse(bp);
1320 anyfreed = 1;
1321 if (nbp != NULL &&
1322 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1323 (nbp->b_vp != vp) ||
1324 (nbp->b_flags & B_DELWRI) == 0)) {
1325 goto restart;
1326 }
1327 VI_LOCK(vp);
1328 }
1329 }
1330
1331 if (length > 0) {
1332 restartsync:
1333 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1334 if (bp->b_lblkno > 0)
1335 continue;
1336 /*
1337 * Since we hold the vnode lock this should only
1338 * fail if we're racing with the buf daemon.
1339 */
1340 if (BUF_LOCK(bp,
1341 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1342 VI_MTX(vp)) == ENOLCK) {
1343 goto restart;
1344 }
1345 VNASSERT((bp->b_flags & B_DELWRI), vp,
1346 ("buf(%p) on dirty queue without DELWRI", bp));
1347
1348 bremfree(bp);
1349 bawrite(bp);
1350 VI_LOCK(vp);
1351 goto restartsync;
1352 }
1353 }
1354
1355 bufobj_wwait(bo, 0, 0);
1356 VI_UNLOCK(vp);
1357 vnode_pager_setsize(vp, length);
1358
1359 return (0);
1360 }
1361
1362 /*
1363 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1364 * a vnode.
1365 *
1366 * NOTE: We have to deal with the special case of a background bitmap
1367 * buffer, a situation where two buffers will have the same logical
1368 * block offset. We want (1) only the foreground buffer to be accessed
1369 * in a lookup and (2) must differentiate between the foreground and
1370 * background buffer in the splay tree algorithm because the splay
1371 * tree cannot normally handle multiple entities with the same 'index'.
1372 * We accomplish this by adding differentiating flags to the splay tree's
1373 * numerical domain.
1374 */
1375 static
1376 struct buf *
1377 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1378 {
1379 struct buf dummy;
1380 struct buf *lefttreemax, *righttreemin, *y;
1381
1382 if (root == NULL)
1383 return (NULL);
1384 lefttreemax = righttreemin = &dummy;
1385 for (;;) {
1386 if (lblkno < root->b_lblkno ||
1387 (lblkno == root->b_lblkno &&
1388 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1389 if ((y = root->b_left) == NULL)
1390 break;
1391 if (lblkno < y->b_lblkno) {
1392 /* Rotate right. */
1393 root->b_left = y->b_right;
1394 y->b_right = root;
1395 root = y;
1396 if ((y = root->b_left) == NULL)
1397 break;
1398 }
1399 /* Link into the new root's right tree. */
1400 righttreemin->b_left = root;
1401 righttreemin = root;
1402 } else if (lblkno > root->b_lblkno ||
1403 (lblkno == root->b_lblkno &&
1404 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1405 if ((y = root->b_right) == NULL)
1406 break;
1407 if (lblkno > y->b_lblkno) {
1408 /* Rotate left. */
1409 root->b_right = y->b_left;
1410 y->b_left = root;
1411 root = y;
1412 if ((y = root->b_right) == NULL)
1413 break;
1414 }
1415 /* Link into the new root's left tree. */
1416 lefttreemax->b_right = root;
1417 lefttreemax = root;
1418 } else {
1419 break;
1420 }
1421 root = y;
1422 }
1423 /* Assemble the new root. */
1424 lefttreemax->b_right = root->b_left;
1425 righttreemin->b_left = root->b_right;
1426 root->b_left = dummy.b_right;
1427 root->b_right = dummy.b_left;
1428 return (root);
1429 }
1430
1431 static void
1432 buf_vlist_remove(struct buf *bp)
1433 {
1434 struct buf *root;
1435 struct bufv *bv;
1436
1437 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1438 ASSERT_BO_LOCKED(bp->b_bufobj);
1439 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1440 (BX_VNDIRTY|BX_VNCLEAN),
1441 ("buf_vlist_remove: Buf %p is on two lists", bp));
1442 if (bp->b_xflags & BX_VNDIRTY)
1443 bv = &bp->b_bufobj->bo_dirty;
1444 else
1445 bv = &bp->b_bufobj->bo_clean;
1446 if (bp != bv->bv_root) {
1447 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1448 KASSERT(root == bp, ("splay lookup failed in remove"));
1449 }
1450 if (bp->b_left == NULL) {
1451 root = bp->b_right;
1452 } else {
1453 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1454 root->b_right = bp->b_right;
1455 }
1456 bv->bv_root = root;
1457 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1458 bv->bv_cnt--;
1459 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1460 }
1461
1462 /*
1463 * Add the buffer to the sorted clean or dirty block list using a
1464 * splay tree algorithm.
1465 *
1466 * NOTE: xflags is passed as a constant, optimizing this inline function!
1467 */
1468 static void
1469 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1470 {
1471 struct buf *root;
1472 struct bufv *bv;
1473
1474 ASSERT_BO_LOCKED(bo);
1475 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1476 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1477 bp->b_xflags |= xflags;
1478 if (xflags & BX_VNDIRTY)
1479 bv = &bo->bo_dirty;
1480 else
1481 bv = &bo->bo_clean;
1482
1483 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1484 if (root == NULL) {
1485 bp->b_left = NULL;
1486 bp->b_right = NULL;
1487 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1488 } else if (bp->b_lblkno < root->b_lblkno ||
1489 (bp->b_lblkno == root->b_lblkno &&
1490 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1491 bp->b_left = root->b_left;
1492 bp->b_right = root;
1493 root->b_left = NULL;
1494 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1495 } else {
1496 bp->b_right = root->b_right;
1497 bp->b_left = root;
1498 root->b_right = NULL;
1499 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1500 }
1501 bv->bv_cnt++;
1502 bv->bv_root = bp;
1503 }
1504
1505 /*
1506 * Lookup a buffer using the splay tree. Note that we specifically avoid
1507 * shadow buffers used in background bitmap writes.
1508 *
1509 * This code isn't quite efficient as it could be because we are maintaining
1510 * two sorted lists and do not know which list the block resides in.
1511 *
1512 * During a "make buildworld" the desired buffer is found at one of
1513 * the roots more than 60% of the time. Thus, checking both roots
1514 * before performing either splay eliminates unnecessary splays on the
1515 * first tree splayed.
1516 */
1517 struct buf *
1518 gbincore(struct bufobj *bo, daddr_t lblkno)
1519 {
1520 struct buf *bp;
1521
1522 ASSERT_BO_LOCKED(bo);
1523 if ((bp = bo->bo_clean.bv_root) != NULL &&
1524 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1525 return (bp);
1526 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1527 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1528 return (bp);
1529 if ((bp = bo->bo_clean.bv_root) != NULL) {
1530 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1531 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1532 return (bp);
1533 }
1534 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1535 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1536 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1537 return (bp);
1538 }
1539 return (NULL);
1540 }
1541
1542 /*
1543 * Associate a buffer with a vnode.
1544 */
1545 void
1546 bgetvp(struct vnode *vp, struct buf *bp)
1547 {
1548
1549 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1550
1551 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1552 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1553 ("bgetvp: bp already attached! %p", bp));
1554
1555 ASSERT_VI_LOCKED(vp, "bgetvp");
1556 vholdl(vp);
1557 if (VFS_NEEDSGIANT(vp->v_mount) ||
1558 vp->v_bufobj.bo_flag & BO_NEEDSGIANT)
1559 bp->b_flags |= B_NEEDSGIANT;
1560 bp->b_vp = vp;
1561 bp->b_bufobj = &vp->v_bufobj;
1562 /*
1563 * Insert onto list for new vnode.
1564 */
1565 buf_vlist_add(bp, &vp->v_bufobj, BX_VNCLEAN);
1566 }
1567
1568 /*
1569 * Disassociate a buffer from a vnode.
1570 */
1571 void
1572 brelvp(struct buf *bp)
1573 {
1574 struct bufobj *bo;
1575 struct vnode *vp;
1576
1577 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1578 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1579
1580 /*
1581 * Delete from old vnode list, if on one.
1582 */
1583 vp = bp->b_vp; /* XXX */
1584 bo = bp->b_bufobj;
1585 BO_LOCK(bo);
1586 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1587 buf_vlist_remove(bp);
1588 else
1589 panic("brelvp: Buffer %p not on queue.", bp);
1590 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1591 bo->bo_flag &= ~BO_ONWORKLST;
1592 mtx_lock(&sync_mtx);
1593 LIST_REMOVE(bo, bo_synclist);
1594 syncer_worklist_len--;
1595 mtx_unlock(&sync_mtx);
1596 }
1597 bp->b_flags &= ~B_NEEDSGIANT;
1598 bp->b_vp = NULL;
1599 bp->b_bufobj = NULL;
1600 vdropl(vp);
1601 }
1602
1603 /*
1604 * Add an item to the syncer work queue.
1605 */
1606 static void
1607 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1608 {
1609 int slot;
1610
1611 ASSERT_BO_LOCKED(bo);
1612
1613 mtx_lock(&sync_mtx);
1614 if (bo->bo_flag & BO_ONWORKLST)
1615 LIST_REMOVE(bo, bo_synclist);
1616 else {
1617 bo->bo_flag |= BO_ONWORKLST;
1618 syncer_worklist_len++;
1619 }
1620
1621 if (delay > syncer_maxdelay - 2)
1622 delay = syncer_maxdelay - 2;
1623 slot = (syncer_delayno + delay) & syncer_mask;
1624
1625 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1626 mtx_unlock(&sync_mtx);
1627 }
1628
1629 static int
1630 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1631 {
1632 int error, len;
1633
1634 mtx_lock(&sync_mtx);
1635 len = syncer_worklist_len - sync_vnode_count;
1636 mtx_unlock(&sync_mtx);
1637 error = SYSCTL_OUT(req, &len, sizeof(len));
1638 return (error);
1639 }
1640
1641 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1642 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1643
1644 static struct proc *updateproc;
1645 static void sched_sync(void);
1646 static struct kproc_desc up_kp = {
1647 "syncer",
1648 sched_sync,
1649 &updateproc
1650 };
1651 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1652
1653 static int
1654 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1655 {
1656 struct vnode *vp;
1657 struct mount *mp;
1658 int vfslocked;
1659
1660 vfslocked = 0;
1661 restart:
1662 *bo = LIST_FIRST(slp);
1663 if (*bo == NULL) {
1664 VFS_UNLOCK_GIANT(vfslocked);
1665 return (0);
1666 }
1667 vp = (*bo)->__bo_vnode; /* XXX */
1668 if (VFS_NEEDSGIANT(vp->v_mount)) {
1669 if (!vfslocked) {
1670 vfslocked = 1;
1671 if (mtx_trylock(&Giant) == 0) {
1672 mtx_unlock(&sync_mtx);
1673 mtx_lock(&Giant);
1674 mtx_lock(&sync_mtx);
1675 goto restart;
1676 }
1677 }
1678 } else {
1679 VFS_UNLOCK_GIANT(vfslocked);
1680 vfslocked = 0;
1681 }
1682 if (VOP_ISLOCKED(vp, NULL) != 0) {
1683 VFS_UNLOCK_GIANT(vfslocked);
1684 return (1);
1685 }
1686 if (VI_TRYLOCK(vp) == 0) {
1687 VFS_UNLOCK_GIANT(vfslocked);
1688 return (1);
1689 }
1690 /*
1691 * We use vhold in case the vnode does not
1692 * successfully sync. vhold prevents the vnode from
1693 * going away when we unlock the sync_mtx so that
1694 * we can acquire the vnode interlock.
1695 */
1696 vholdl(vp);
1697 mtx_unlock(&sync_mtx);
1698 VI_UNLOCK(vp);
1699 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1700 vdrop(vp);
1701 VFS_UNLOCK_GIANT(vfslocked);
1702 mtx_lock(&sync_mtx);
1703 return (*bo == LIST_FIRST(slp));
1704 }
1705 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1706 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1707 VOP_UNLOCK(vp, 0, td);
1708 vn_finished_write(mp);
1709 VI_LOCK(vp);
1710 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1711 /*
1712 * Put us back on the worklist. The worklist
1713 * routine will remove us from our current
1714 * position and then add us back in at a later
1715 * position.
1716 */
1717 vn_syncer_add_to_worklist(*bo, syncdelay);
1718 }
1719 vdropl(vp);
1720 VFS_UNLOCK_GIANT(vfslocked);
1721 mtx_lock(&sync_mtx);
1722 return (0);
1723 }
1724
1725 /*
1726 * System filesystem synchronizer daemon.
1727 */
1728 static void
1729 sched_sync(void)
1730 {
1731 struct synclist *next;
1732 struct synclist *slp;
1733 struct bufobj *bo;
1734 long starttime;
1735 struct thread *td = curthread;
1736 static int dummychan;
1737 int last_work_seen;
1738 int net_worklist_len;
1739 int syncer_final_iter;
1740 int first_printf;
1741 int error;
1742
1743 last_work_seen = 0;
1744 syncer_final_iter = 0;
1745 first_printf = 1;
1746 syncer_state = SYNCER_RUNNING;
1747 starttime = time_uptime;
1748 td->td_pflags |= TDP_NORUNNINGBUF;
1749
1750 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1751 SHUTDOWN_PRI_LAST);
1752
1753 mtx_lock(&sync_mtx);
1754 for (;;) {
1755 if (syncer_state == SYNCER_FINAL_DELAY &&
1756 syncer_final_iter == 0) {
1757 mtx_unlock(&sync_mtx);
1758 kthread_suspend_check(td->td_proc);
1759 mtx_lock(&sync_mtx);
1760 }
1761 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1762 if (syncer_state != SYNCER_RUNNING &&
1763 starttime != time_uptime) {
1764 if (first_printf) {
1765 printf("\nSyncing disks, vnodes remaining...");
1766 first_printf = 0;
1767 }
1768 printf("%d ", net_worklist_len);
1769 }
1770 starttime = time_uptime;
1771
1772 /*
1773 * Push files whose dirty time has expired. Be careful
1774 * of interrupt race on slp queue.
1775 *
1776 * Skip over empty worklist slots when shutting down.
1777 */
1778 do {
1779 slp = &syncer_workitem_pending[syncer_delayno];
1780 syncer_delayno += 1;
1781 if (syncer_delayno == syncer_maxdelay)
1782 syncer_delayno = 0;
1783 next = &syncer_workitem_pending[syncer_delayno];
1784 /*
1785 * If the worklist has wrapped since the
1786 * it was emptied of all but syncer vnodes,
1787 * switch to the FINAL_DELAY state and run
1788 * for one more second.
1789 */
1790 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1791 net_worklist_len == 0 &&
1792 last_work_seen == syncer_delayno) {
1793 syncer_state = SYNCER_FINAL_DELAY;
1794 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1795 }
1796 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1797 syncer_worklist_len > 0);
1798
1799 /*
1800 * Keep track of the last time there was anything
1801 * on the worklist other than syncer vnodes.
1802 * Return to the SHUTTING_DOWN state if any
1803 * new work appears.
1804 */
1805 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1806 last_work_seen = syncer_delayno;
1807 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1808 syncer_state = SYNCER_SHUTTING_DOWN;
1809 while (!LIST_EMPTY(slp)) {
1810 error = sync_vnode(slp, &bo, td);
1811 if (error == 1) {
1812 LIST_REMOVE(bo, bo_synclist);
1813 LIST_INSERT_HEAD(next, bo, bo_synclist);
1814 continue;
1815 }
1816 }
1817 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1818 syncer_final_iter--;
1819 /*
1820 * The variable rushjob allows the kernel to speed up the
1821 * processing of the filesystem syncer process. A rushjob
1822 * value of N tells the filesystem syncer to process the next
1823 * N seconds worth of work on its queue ASAP. Currently rushjob
1824 * is used by the soft update code to speed up the filesystem
1825 * syncer process when the incore state is getting so far
1826 * ahead of the disk that the kernel memory pool is being
1827 * threatened with exhaustion.
1828 */
1829 if (rushjob > 0) {
1830 rushjob -= 1;
1831 continue;
1832 }
1833 /*
1834 * Just sleep for a short period of time between
1835 * iterations when shutting down to allow some I/O
1836 * to happen.
1837 *
1838 * If it has taken us less than a second to process the
1839 * current work, then wait. Otherwise start right over
1840 * again. We can still lose time if any single round
1841 * takes more than two seconds, but it does not really
1842 * matter as we are just trying to generally pace the
1843 * filesystem activity.
1844 */
1845 if (syncer_state != SYNCER_RUNNING)
1846 msleep(&dummychan, &sync_mtx, PPAUSE, "syncfnl",
1847 hz / SYNCER_SHUTDOWN_SPEEDUP);
1848 else if (time_uptime == starttime)
1849 msleep(&lbolt, &sync_mtx, PPAUSE, "syncer", 0);
1850 }
1851 }
1852
1853 /*
1854 * Request the syncer daemon to speed up its work.
1855 * We never push it to speed up more than half of its
1856 * normal turn time, otherwise it could take over the cpu.
1857 */
1858 int
1859 speedup_syncer(void)
1860 {
1861 struct thread *td;
1862 int ret = 0;
1863
1864 td = FIRST_THREAD_IN_PROC(updateproc);
1865 mtx_lock(&sync_mtx);
1866 if (rushjob < syncdelay / 2) {
1867 rushjob += 1;
1868 stat_rush_requests += 1;
1869 ret = 1;
1870 }
1871 mtx_unlock(&sync_mtx);
1872 sleepq_remove(td, &lbolt);
1873 return (ret);
1874 }
1875
1876 /*
1877 * Tell the syncer to speed up its work and run though its work
1878 * list several times, then tell it to shut down.
1879 */
1880 static void
1881 syncer_shutdown(void *arg, int howto)
1882 {
1883 struct thread *td;
1884
1885 if (howto & RB_NOSYNC)
1886 return;
1887 td = FIRST_THREAD_IN_PROC(updateproc);
1888 mtx_lock(&sync_mtx);
1889 syncer_state = SYNCER_SHUTTING_DOWN;
1890 rushjob = 0;
1891 mtx_unlock(&sync_mtx);
1892 sleepq_remove(td, &lbolt);
1893 kproc_shutdown(arg, howto);
1894 }
1895
1896 /*
1897 * Reassign a buffer from one vnode to another.
1898 * Used to assign file specific control information
1899 * (indirect blocks) to the vnode to which they belong.
1900 */
1901 void
1902 reassignbuf(struct buf *bp)
1903 {
1904 struct vnode *vp;
1905 struct bufobj *bo;
1906 int delay;
1907 #ifdef INVARIANTS
1908 struct bufv *bv;
1909 #endif
1910
1911 vp = bp->b_vp;
1912 bo = bp->b_bufobj;
1913 ++reassignbufcalls;
1914
1915 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1916 bp, bp->b_vp, bp->b_flags);
1917 /*
1918 * B_PAGING flagged buffers cannot be reassigned because their vp
1919 * is not fully linked in.
1920 */
1921 if (bp->b_flags & B_PAGING)
1922 panic("cannot reassign paging buffer");
1923
1924 /*
1925 * Delete from old vnode list, if on one.
1926 */
1927 VI_LOCK(vp);
1928 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1929 buf_vlist_remove(bp);
1930 else
1931 panic("reassignbuf: Buffer %p not on queue.", bp);
1932 /*
1933 * If dirty, put on list of dirty buffers; otherwise insert onto list
1934 * of clean buffers.
1935 */
1936 if (bp->b_flags & B_DELWRI) {
1937 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1938 switch (vp->v_type) {
1939 case VDIR:
1940 delay = dirdelay;
1941 break;
1942 case VCHR:
1943 delay = metadelay;
1944 break;
1945 default:
1946 delay = filedelay;
1947 }
1948 vn_syncer_add_to_worklist(bo, delay);
1949 }
1950 buf_vlist_add(bp, bo, BX_VNDIRTY);
1951 } else {
1952 buf_vlist_add(bp, bo, BX_VNCLEAN);
1953
1954 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1955 mtx_lock(&sync_mtx);
1956 LIST_REMOVE(bo, bo_synclist);
1957 syncer_worklist_len--;
1958 mtx_unlock(&sync_mtx);
1959 bo->bo_flag &= ~BO_ONWORKLST;
1960 }
1961 }
1962 #ifdef INVARIANTS
1963 bv = &bo->bo_clean;
1964 bp = TAILQ_FIRST(&bv->bv_hd);
1965 KASSERT(bp == NULL || bp->b_bufobj == bo,
1966 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1967 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1968 KASSERT(bp == NULL || bp->b_bufobj == bo,
1969 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1970 bv = &bo->bo_dirty;
1971 bp = TAILQ_FIRST(&bv->bv_hd);
1972 KASSERT(bp == NULL || bp->b_bufobj == bo,
1973 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1974 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1975 KASSERT(bp == NULL || bp->b_bufobj == bo,
1976 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1977 #endif
1978 VI_UNLOCK(vp);
1979 }
1980
1981 /*
1982 * Increment the use and hold counts on the vnode, taking care to reference
1983 * the driver's usecount if this is a chardev. The vholdl() will remove
1984 * the vnode from the free list if it is presently free. Requires the
1985 * vnode interlock and returns with it held.
1986 */
1987 static void
1988 v_incr_usecount(struct vnode *vp)
1989 {
1990
1991 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1992 vp, vp->v_holdcnt, vp->v_usecount);
1993 vp->v_usecount++;
1994 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1995 dev_lock();
1996 vp->v_rdev->si_usecount++;
1997 dev_unlock();
1998 }
1999 vholdl(vp);
2000 }
2001
2002 /*
2003 * Turn a holdcnt into a use+holdcnt such that only one call to
2004 * v_decr_usecount is needed.
2005 */
2006 static void
2007 v_upgrade_usecount(struct vnode *vp)
2008 {
2009
2010 CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
2011 vp, vp->v_holdcnt, vp->v_usecount);
2012 vp->v_usecount++;
2013 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2014 dev_lock();
2015 vp->v_rdev->si_usecount++;
2016 dev_unlock();
2017 }
2018 }
2019
2020 /*
2021 * Decrement the vnode use and hold count along with the driver's usecount
2022 * if this is a chardev. The vdropl() below releases the vnode interlock
2023 * as it may free the vnode.
2024 */
2025 static void
2026 v_decr_usecount(struct vnode *vp)
2027 {
2028
2029 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
2030 vp, vp->v_holdcnt, vp->v_usecount);
2031 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2032 VNASSERT(vp->v_usecount > 0, vp,
2033 ("v_decr_usecount: negative usecount"));
2034 vp->v_usecount--;
2035 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2036 dev_lock();
2037 vp->v_rdev->si_usecount--;
2038 dev_unlock();
2039 }
2040 vdropl(vp);
2041 }
2042
2043 /*
2044 * Decrement only the use count and driver use count. This is intended to
2045 * be paired with a follow on vdropl() to release the remaining hold count.
2046 * In this way we may vgone() a vnode with a 0 usecount without risk of
2047 * having it end up on a free list because the hold count is kept above 0.
2048 */
2049 static void
2050 v_decr_useonly(struct vnode *vp)
2051 {
2052
2053 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
2054 vp, vp->v_holdcnt, vp->v_usecount);
2055 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2056 VNASSERT(vp->v_usecount > 0, vp,
2057 ("v_decr_useonly: negative usecount"));
2058 vp->v_usecount--;
2059 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2060 dev_lock();
2061 vp->v_rdev->si_usecount--;
2062 dev_unlock();
2063 }
2064 }
2065
2066 /*
2067 * Grab a particular vnode from the free list, increment its
2068 * reference count and lock it. The vnode lock bit is set if the
2069 * vnode is being eliminated in vgone. The process is awakened
2070 * when the transition is completed, and an error returned to
2071 * indicate that the vnode is no longer usable (possibly having
2072 * been changed to a new filesystem type).
2073 */
2074 int
2075 vget(struct vnode *vp, int flags, struct thread *td)
2076 {
2077 int oweinact;
2078 int oldflags;
2079 int error;
2080
2081 error = 0;
2082 oldflags = flags;
2083 oweinact = 0;
2084 VFS_ASSERT_GIANT(vp->v_mount);
2085 if ((flags & LK_INTERLOCK) == 0)
2086 VI_LOCK(vp);
2087 /*
2088 * If the inactive call was deferred because vput() was called
2089 * with a shared lock, we have to do it here before another thread
2090 * gets a reference to data that should be dead.
2091 */
2092 if (vp->v_iflag & VI_OWEINACT) {
2093 if (flags & LK_NOWAIT) {
2094 VI_UNLOCK(vp);
2095 return (EBUSY);
2096 }
2097 flags &= ~LK_TYPE_MASK;
2098 flags |= LK_EXCLUSIVE;
2099 oweinact = 1;
2100 }
2101 vholdl(vp);
2102 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2103 vdrop(vp);
2104 return (error);
2105 }
2106 VI_LOCK(vp);
2107 /*
2108 * Deal with a timing window when the interlock is not held
2109 * and VI_DOOMED can be set, since we only have a holdcnt,
2110 * not a usecount.
2111 */
2112 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) {
2113 KASSERT((flags & LK_TYPE_MASK) == 0, ("Unexpected flags %x", flags));
2114 vdropl(vp);
2115 return (ENOENT);
2116 }
2117 /* Upgrade our holdcnt to a usecount. */
2118 v_upgrade_usecount(vp);
2119 if (oweinact) {
2120 if (vp->v_iflag & VI_OWEINACT)
2121 vinactive(vp, td);
2122 VI_UNLOCK(vp);
2123 if ((oldflags & LK_TYPE_MASK) == 0)
2124 VOP_UNLOCK(vp, 0, td);
2125 } else
2126 VI_UNLOCK(vp);
2127 return (0);
2128 }
2129
2130 /*
2131 * Increase the reference count of a vnode.
2132 */
2133 void
2134 vref(struct vnode *vp)
2135 {
2136
2137 VI_LOCK(vp);
2138 v_incr_usecount(vp);
2139 VI_UNLOCK(vp);
2140 }
2141
2142 /*
2143 * Return reference count of a vnode.
2144 *
2145 * The results of this call are only guaranteed when some mechanism other
2146 * than the VI lock is used to stop other processes from gaining references
2147 * to the vnode. This may be the case if the caller holds the only reference.
2148 * This is also useful when stale data is acceptable as race conditions may
2149 * be accounted for by some other means.
2150 */
2151 int
2152 vrefcnt(struct vnode *vp)
2153 {
2154 int usecnt;
2155
2156 VI_LOCK(vp);
2157 usecnt = vp->v_usecount;
2158 VI_UNLOCK(vp);
2159
2160 return (usecnt);
2161 }
2162
2163
2164 /*
2165 * Vnode put/release.
2166 * If count drops to zero, call inactive routine and return to freelist.
2167 */
2168 void
2169 vrele(struct vnode *vp)
2170 {
2171 struct thread *td = curthread; /* XXX */
2172
2173 KASSERT(vp != NULL, ("vrele: null vp"));
2174 VFS_ASSERT_GIANT(vp->v_mount);
2175
2176 VI_LOCK(vp);
2177
2178 /* Skip this v_writecount check if we're going to panic below. */
2179 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2180 ("vrele: missed vn_close"));
2181
2182 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2183 vp->v_usecount == 1)) {
2184 v_decr_usecount(vp);
2185 return;
2186 }
2187 if (vp->v_usecount != 1) {
2188 #ifdef DIAGNOSTIC
2189 vprint("vrele: negative ref count", vp);
2190 #endif
2191 VI_UNLOCK(vp);
2192 panic("vrele: negative ref cnt");
2193 }
2194 /*
2195 * We want to hold the vnode until the inactive finishes to
2196 * prevent vgone() races. We drop the use count here and the
2197 * hold count below when we're done.
2198 */
2199 v_decr_useonly(vp);
2200 /*
2201 * We must call VOP_INACTIVE with the node locked. Mark
2202 * as VI_DOINGINACT to avoid recursion.
2203 */
2204 vp->v_iflag |= VI_OWEINACT;
2205 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2206 VI_LOCK(vp);
2207 if (vp->v_usecount > 0)
2208 vp->v_iflag &= ~VI_OWEINACT;
2209 if (vp->v_iflag & VI_OWEINACT)
2210 vinactive(vp, td);
2211 VOP_UNLOCK(vp, 0, td);
2212 } else {
2213 VI_LOCK(vp);
2214 if (vp->v_usecount > 0)
2215 vp->v_iflag &= ~VI_OWEINACT;
2216 }
2217 vdropl(vp);
2218 }
2219
2220 /*
2221 * Release an already locked vnode. This give the same effects as
2222 * unlock+vrele(), but takes less time and avoids releasing and
2223 * re-aquiring the lock (as vrele() acquires the lock internally.)
2224 */
2225 void
2226 vput(struct vnode *vp)
2227 {
2228 struct thread *td = curthread; /* XXX */
2229 int error;
2230
2231 KASSERT(vp != NULL, ("vput: null vp"));
2232 ASSERT_VOP_LOCKED(vp, "vput");
2233 VFS_ASSERT_GIANT(vp->v_mount);
2234 VI_LOCK(vp);
2235 /* Skip this v_writecount check if we're going to panic below. */
2236 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2237 ("vput: missed vn_close"));
2238 error = 0;
2239
2240 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2241 vp->v_usecount == 1)) {
2242 VOP_UNLOCK(vp, 0, td);
2243 v_decr_usecount(vp);
2244 return;
2245 }
2246
2247 if (vp->v_usecount != 1) {
2248 #ifdef DIAGNOSTIC
2249 vprint("vput: negative ref count", vp);
2250 #endif
2251 panic("vput: negative ref cnt");
2252 }
2253 /*
2254 * We want to hold the vnode until the inactive finishes to
2255 * prevent vgone() races. We drop the use count here and the
2256 * hold count below when we're done.
2257 */
2258 v_decr_useonly(vp);
2259 vp->v_iflag |= VI_OWEINACT;
2260 if (VOP_ISLOCKED(vp, NULL) != LK_EXCLUSIVE) {
2261 error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT, td);
2262 VI_LOCK(vp);
2263 if (error) {
2264 if (vp->v_usecount > 0)
2265 vp->v_iflag &= ~VI_OWEINACT;
2266 goto done;
2267 }
2268 }
2269 if (vp->v_usecount > 0)
2270 vp->v_iflag &= ~VI_OWEINACT;
2271 if (vp->v_iflag & VI_OWEINACT)
2272 vinactive(vp, td);
2273 VOP_UNLOCK(vp, 0, td);
2274 done:
2275 vdropl(vp);
2276 }
2277
2278 /*
2279 * Somebody doesn't want the vnode recycled.
2280 */
2281 void
2282 vhold(struct vnode *vp)
2283 {
2284
2285 VI_LOCK(vp);
2286 vholdl(vp);
2287 VI_UNLOCK(vp);
2288 }
2289
2290 void
2291 vholdl(struct vnode *vp)
2292 {
2293
2294 vp->v_holdcnt++;
2295 if (VSHOULDBUSY(vp))
2296 vbusy(vp);
2297 }
2298
2299 /*
2300 * Note that there is one less who cares about this vnode. vdrop() is the
2301 * opposite of vhold().
2302 */
2303 void
2304 vdrop(struct vnode *vp)
2305 {
2306
2307 VI_LOCK(vp);
2308 vdropl(vp);
2309 }
2310
2311 /*
2312 * Drop the hold count of the vnode. If this is the last reference to
2313 * the vnode we will free it if it has been vgone'd otherwise it is
2314 * placed on the free list.
2315 */
2316 void
2317 vdropl(struct vnode *vp)
2318 {
2319
2320 ASSERT_VI_LOCKED(vp, "vdropl");
2321 if (vp->v_holdcnt <= 0)
2322 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2323 vp->v_holdcnt--;
2324 if (vp->v_holdcnt == 0) {
2325 if (vp->v_iflag & VI_DOOMED) {
2326 vdestroy(vp);
2327 return;
2328 } else
2329 vfree(vp);
2330 }
2331 VI_UNLOCK(vp);
2332 }
2333
2334 /*
2335 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2336 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2337 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2338 * failed lock upgrade.
2339 */
2340 static void
2341 vinactive(struct vnode *vp, struct thread *td)
2342 {
2343
2344 ASSERT_VOP_LOCKED(vp, "vinactive");
2345 ASSERT_VI_LOCKED(vp, "vinactive");
2346 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2347 ("vinactive: recursed on VI_DOINGINACT"));
2348 vp->v_iflag |= VI_DOINGINACT;
2349 vp->v_iflag &= ~VI_OWEINACT;
2350 VI_UNLOCK(vp);
2351 VOP_INACTIVE(vp, td);
2352 VI_LOCK(vp);
2353 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2354 ("vinactive: lost VI_DOINGINACT"));
2355 vp->v_iflag &= ~VI_DOINGINACT;
2356 }
2357
2358 /*
2359 * Remove any vnodes in the vnode table belonging to mount point mp.
2360 *
2361 * If FORCECLOSE is not specified, there should not be any active ones,
2362 * return error if any are found (nb: this is a user error, not a
2363 * system error). If FORCECLOSE is specified, detach any active vnodes
2364 * that are found.
2365 *
2366 * If WRITECLOSE is set, only flush out regular file vnodes open for
2367 * writing.
2368 *
2369 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2370 *
2371 * `rootrefs' specifies the base reference count for the root vnode
2372 * of this filesystem. The root vnode is considered busy if its
2373 * v_usecount exceeds this value. On a successful return, vflush(, td)
2374 * will call vrele() on the root vnode exactly rootrefs times.
2375 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2376 * be zero.
2377 */
2378 #ifdef DIAGNOSTIC
2379 static int busyprt = 0; /* print out busy vnodes */
2380 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2381 #endif
2382
2383 int
2384 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2385 {
2386 struct vnode *vp, *mvp, *rootvp = NULL;
2387 struct vattr vattr;
2388 int busy = 0, error;
2389
2390 CTR1(KTR_VFS, "vflush: mp %p", mp);
2391 if (rootrefs > 0) {
2392 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2393 ("vflush: bad args"));
2394 /*
2395 * Get the filesystem root vnode. We can vput() it
2396 * immediately, since with rootrefs > 0, it won't go away.
2397 */
2398 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2399 return (error);
2400 vput(rootvp);
2401
2402 }
2403 MNT_ILOCK(mp);
2404 loop:
2405 MNT_VNODE_FOREACH(vp, mp, mvp) {
2406
2407 VI_LOCK(vp);
2408 vholdl(vp);
2409 MNT_IUNLOCK(mp);
2410 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2411 if (error) {
2412 vdrop(vp);
2413 MNT_ILOCK(mp);
2414 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2415 goto loop;
2416 }
2417 /*
2418 * Skip over a vnodes marked VV_SYSTEM.
2419 */
2420 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2421 VOP_UNLOCK(vp, 0, td);
2422 vdrop(vp);
2423 MNT_ILOCK(mp);
2424 continue;
2425 }
2426 /*
2427 * If WRITECLOSE is set, flush out unlinked but still open
2428 * files (even if open only for reading) and regular file
2429 * vnodes open for writing.
2430 */
2431 if (flags & WRITECLOSE) {
2432 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2433 VI_LOCK(vp);
2434
2435 if ((vp->v_type == VNON ||
2436 (error == 0 && vattr.va_nlink > 0)) &&
2437 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2438 VOP_UNLOCK(vp, 0, td);
2439 vdropl(vp);
2440 MNT_ILOCK(mp);
2441 continue;
2442 }
2443 } else
2444 VI_LOCK(vp);
2445 /*
2446 * With v_usecount == 0, all we need to do is clear out the
2447 * vnode data structures and we are done.
2448 *
2449 * If FORCECLOSE is set, forcibly close the vnode.
2450 */
2451 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2452 VNASSERT(vp->v_usecount == 0 ||
2453 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2454 ("device VNODE %p is FORCECLOSED", vp));
2455 vgonel(vp);
2456 } else {
2457 busy++;
2458 #ifdef DIAGNOSTIC
2459 if (busyprt)
2460 vprint("vflush: busy vnode", vp);
2461 #endif
2462 }
2463 VOP_UNLOCK(vp, 0, td);
2464 vdropl(vp);
2465 MNT_ILOCK(mp);
2466 }
2467 MNT_IUNLOCK(mp);
2468 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2469 /*
2470 * If just the root vnode is busy, and if its refcount
2471 * is equal to `rootrefs', then go ahead and kill it.
2472 */
2473 VI_LOCK(rootvp);
2474 KASSERT(busy > 0, ("vflush: not busy"));
2475 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2476 ("vflush: usecount %d < rootrefs %d",
2477 rootvp->v_usecount, rootrefs));
2478 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2479 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK, td);
2480 vgone(rootvp);
2481 VOP_UNLOCK(rootvp, 0, td);
2482 busy = 0;
2483 } else
2484 VI_UNLOCK(rootvp);
2485 }
2486 if (busy)
2487 return (EBUSY);
2488 for (; rootrefs > 0; rootrefs--)
2489 vrele(rootvp);
2490 return (0);
2491 }
2492
2493 /*
2494 * Recycle an unused vnode to the front of the free list.
2495 */
2496 int
2497 vrecycle(struct vnode *vp, struct thread *td)
2498 {
2499 int recycled;
2500
2501 ASSERT_VOP_LOCKED(vp, "vrecycle");
2502 recycled = 0;
2503 VI_LOCK(vp);
2504 if (vp->v_usecount == 0) {
2505 recycled = 1;
2506 vgonel(vp);
2507 }
2508 VI_UNLOCK(vp);
2509 return (recycled);
2510 }
2511
2512 /*
2513 * Eliminate all activity associated with a vnode
2514 * in preparation for reuse.
2515 */
2516 void
2517 vgone(struct vnode *vp)
2518 {
2519 VI_LOCK(vp);
2520 vgonel(vp);
2521 VI_UNLOCK(vp);
2522 }
2523
2524 /*
2525 * vgone, with the vp interlock held.
2526 */
2527 void
2528 vgonel(struct vnode *vp)
2529 {
2530 struct thread *td;
2531 int oweinact;
2532 int active;
2533 struct mount *mp;
2534
2535 CTR1(KTR_VFS, "vgonel: vp %p", vp);
2536 ASSERT_VOP_LOCKED(vp, "vgonel");
2537 ASSERT_VI_LOCKED(vp, "vgonel");
2538 VNASSERT(vp->v_holdcnt, vp,
2539 ("vgonel: vp %p has no reference.", vp));
2540 td = curthread;
2541
2542 /*
2543 * Don't vgonel if we're already doomed.
2544 */
2545 if (vp->v_iflag & VI_DOOMED)
2546 return;
2547 vp->v_iflag |= VI_DOOMED;
2548 /*
2549 * Check to see if the vnode is in use. If so, we have to call
2550 * VOP_CLOSE() and VOP_INACTIVE().
2551 */
2552 active = vp->v_usecount;
2553 oweinact = (vp->v_iflag & VI_OWEINACT);
2554 VI_UNLOCK(vp);
2555 /*
2556 * Clean out any buffers associated with the vnode.
2557 * If the flush fails, just toss the buffers.
2558 */
2559 mp = NULL;
2560 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2561 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2562 if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2563 vinvalbuf(vp, 0, td, 0, 0);
2564
2565 /*
2566 * If purging an active vnode, it must be closed and
2567 * deactivated before being reclaimed.
2568 */
2569 if (active)
2570 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2571 if (oweinact || active) {
2572 VI_LOCK(vp);
2573 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2574 vinactive(vp, td);
2575 VI_UNLOCK(vp);
2576 }
2577 /*
2578 * Reclaim the vnode.
2579 */
2580 if (VOP_RECLAIM(vp, td))
2581 panic("vgone: cannot reclaim");
2582 if (mp != NULL)
2583 vn_finished_secondary_write(mp);
2584 VNASSERT(vp->v_object == NULL, vp,
2585 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2586 /*
2587 * Clear the advisory locks and wake up waiting threads.
2588 */
2589 lf_purgelocks(vp, &(vp->v_lockf));
2590 /*
2591 * Delete from old mount point vnode list.
2592 */
2593 delmntque(vp);
2594 cache_purge(vp);
2595 /*
2596 * Done with purge, reset to the standard lock and invalidate
2597 * the vnode.
2598 */
2599 VI_LOCK(vp);
2600 vp->v_vnlock = &vp->v_lock;
2601 vp->v_op = &dead_vnodeops;
2602 vp->v_tag = "none";
2603 vp->v_type = VBAD;
2604 }
2605
2606 /*
2607 * Calculate the total number of references to a special device.
2608 */
2609 int
2610 vcount(struct vnode *vp)
2611 {
2612 int count;
2613
2614 dev_lock();
2615 count = vp->v_rdev->si_usecount;
2616 dev_unlock();
2617 return (count);
2618 }
2619
2620 /*
2621 * Same as above, but using the struct cdev *as argument
2622 */
2623 int
2624 count_dev(struct cdev *dev)
2625 {
2626 int count;
2627
2628 dev_lock();
2629 count = dev->si_usecount;
2630 dev_unlock();
2631 return(count);
2632 }
2633
2634 /*
2635 * Print out a description of a vnode.
2636 */
2637 static char *typename[] =
2638 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2639 "VMARKER"};
2640
2641 void
2642 vn_printf(struct vnode *vp, const char *fmt, ...)
2643 {
2644 va_list ap;
2645 char buf[256], buf2[16];
2646 u_long flags;
2647
2648 va_start(ap, fmt);
2649 vprintf(fmt, ap);
2650 va_end(ap);
2651 printf("%p: ", (void *)vp);
2652 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2653 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2654 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2655 buf[0] = '\0';
2656 buf[1] = '\0';
2657 if (vp->v_vflag & VV_ROOT)
2658 strlcat(buf, "|VV_ROOT", sizeof(buf));
2659 if (vp->v_vflag & VV_ISTTY)
2660 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2661 if (vp->v_vflag & VV_NOSYNC)
2662 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2663 if (vp->v_vflag & VV_CACHEDLABEL)
2664 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2665 if (vp->v_vflag & VV_TEXT)
2666 strlcat(buf, "|VV_TEXT", sizeof(buf));
2667 if (vp->v_vflag & VV_COPYONWRITE)
2668 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2669 if (vp->v_vflag & VV_SYSTEM)
2670 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2671 if (vp->v_vflag & VV_PROCDEP)
2672 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2673 if (vp->v_vflag & VV_NOKNOTE)
2674 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2675 if (vp->v_vflag & VV_DELETED)
2676 strlcat(buf, "|VV_DELETED", sizeof(buf));
2677 if (vp->v_vflag & VV_MD)
2678 strlcat(buf, "|VV_MD", sizeof(buf));
2679 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2680 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2681 VV_NOKNOTE | VV_DELETED | VV_MD);
2682 if (flags != 0) {
2683 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2684 strlcat(buf, buf2, sizeof(buf));
2685 }
2686 if (vp->v_iflag & VI_MOUNT)
2687 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2688 if (vp->v_iflag & VI_AGE)
2689 strlcat(buf, "|VI_AGE", sizeof(buf));
2690 if (vp->v_iflag & VI_DOOMED)
2691 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2692 if (vp->v_iflag & VI_FREE)
2693 strlcat(buf, "|VI_FREE", sizeof(buf));
2694 if (vp->v_iflag & VI_OBJDIRTY)
2695 strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2696 if (vp->v_iflag & VI_DOINGINACT)
2697 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2698 if (vp->v_iflag & VI_OWEINACT)
2699 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2700 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2701 VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2702 if (flags != 0) {
2703 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2704 strlcat(buf, buf2, sizeof(buf));
2705 }
2706 printf(" flags (%s)\n", buf + 1);
2707 if (mtx_owned(VI_MTX(vp)))
2708 printf(" VI_LOCKed");
2709 if (vp->v_object != NULL)
2710 printf(" v_object %p ref %d pages %d\n",
2711 vp->v_object, vp->v_object->ref_count,
2712 vp->v_object->resident_page_count);
2713 printf(" ");
2714 lockmgr_printinfo(vp->v_vnlock);
2715 printf("\n");
2716 if (vp->v_data != NULL)
2717 VOP_PRINT(vp);
2718 }
2719
2720 #ifdef DDB
2721 /*
2722 * List all of the locked vnodes in the system.
2723 * Called when debugging the kernel.
2724 */
2725 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2726 {
2727 struct mount *mp, *nmp;
2728 struct vnode *vp;
2729
2730 /*
2731 * Note: because this is DDB, we can't obey the locking semantics
2732 * for these structures, which means we could catch an inconsistent
2733 * state and dereference a nasty pointer. Not much to be done
2734 * about that.
2735 */
2736 db_printf("Locked vnodes\n");
2737 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2738 nmp = TAILQ_NEXT(mp, mnt_list);
2739 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2740 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp, NULL))
2741 vprint("", vp);
2742 }
2743 nmp = TAILQ_NEXT(mp, mnt_list);
2744 }
2745 }
2746
2747 /*
2748 * Show details about the given vnode.
2749 */
2750 DB_SHOW_COMMAND(vnode, db_show_vnode)
2751 {
2752 struct vnode *vp;
2753
2754 if (!have_addr)
2755 return;
2756 vp = (struct vnode *)addr;
2757 vn_printf(vp, "vnode ");
2758 }
2759
2760 /*
2761 * Show details about the given mount point.
2762 */
2763 DB_SHOW_COMMAND(mount, db_show_mount)
2764 {
2765 struct mount *mp;
2766 struct statfs *sp;
2767 struct vnode *vp;
2768 char buf[512];
2769 u_int flags;
2770
2771 if (!have_addr) {
2772 /* No address given, print short info about all mount points. */
2773 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2774 db_printf("%p %s on %s (%s)\n", mp,
2775 mp->mnt_stat.f_mntfromname,
2776 mp->mnt_stat.f_mntonname,
2777 mp->mnt_stat.f_fstypename);
2778 if (db_pager_quit)
2779 break;
2780 }
2781 db_printf("\nMore info: show mount <addr>\n");
2782 return;
2783 }
2784
2785 mp = (struct mount *)addr;
2786 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2787 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2788
2789 buf[0] = '\0';
2790 flags = mp->mnt_flag;
2791 #define MNT_FLAG(flag) do { \
2792 if (flags & (flag)) { \
2793 if (buf[0] != '\0') \
2794 strlcat(buf, ", ", sizeof(buf)); \
2795 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2796 flags &= ~(flag); \
2797 } \
2798 } while (0)
2799 MNT_FLAG(MNT_RDONLY);
2800 MNT_FLAG(MNT_SYNCHRONOUS);
2801 MNT_FLAG(MNT_NOEXEC);
2802 MNT_FLAG(MNT_NOSUID);
2803 MNT_FLAG(MNT_UNION);
2804 MNT_FLAG(MNT_ASYNC);
2805 MNT_FLAG(MNT_SUIDDIR);
2806 MNT_FLAG(MNT_SOFTDEP);
2807 MNT_FLAG(MNT_NOSYMFOLLOW);
2808 MNT_FLAG(MNT_GJOURNAL);
2809 MNT_FLAG(MNT_MULTILABEL);
2810 MNT_FLAG(MNT_ACLS);
2811 MNT_FLAG(MNT_NOATIME);
2812 MNT_FLAG(MNT_NOCLUSTERR);
2813 MNT_FLAG(MNT_NOCLUSTERW);
2814 MNT_FLAG(MNT_EXRDONLY);
2815 MNT_FLAG(MNT_EXPORTED);
2816 MNT_FLAG(MNT_DEFEXPORTED);
2817 MNT_FLAG(MNT_EXPORTANON);
2818 MNT_FLAG(MNT_EXKERB);
2819 MNT_FLAG(MNT_EXPUBLIC);
2820 MNT_FLAG(MNT_LOCAL);
2821 MNT_FLAG(MNT_QUOTA);
2822 MNT_FLAG(MNT_ROOTFS);
2823 MNT_FLAG(MNT_USER);
2824 MNT_FLAG(MNT_IGNORE);
2825 MNT_FLAG(MNT_UPDATE);
2826 MNT_FLAG(MNT_DELEXPORT);
2827 MNT_FLAG(MNT_RELOAD);
2828 MNT_FLAG(MNT_FORCE);
2829 MNT_FLAG(MNT_SNAPSHOT);
2830 MNT_FLAG(MNT_BYFSID);
2831 #undef MNT_FLAG
2832 if (flags != 0) {
2833 if (buf[0] != '\0')
2834 strlcat(buf, ", ", sizeof(buf));
2835 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2836 "0x%08x", flags);
2837 }
2838 db_printf(" mnt_flag = %s\n", buf);
2839
2840 buf[0] = '\0';
2841 flags = mp->mnt_kern_flag;
2842 #define MNT_KERN_FLAG(flag) do { \
2843 if (flags & (flag)) { \
2844 if (buf[0] != '\0') \
2845 strlcat(buf, ", ", sizeof(buf)); \
2846 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2847 flags &= ~(flag); \
2848 } \
2849 } while (0)
2850 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2851 MNT_KERN_FLAG(MNTK_ASYNC);
2852 MNT_KERN_FLAG(MNTK_SOFTDEP);
2853 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2854 MNT_KERN_FLAG(MNTK_UNMOUNT);
2855 MNT_KERN_FLAG(MNTK_MWAIT);
2856 MNT_KERN_FLAG(MNTK_SUSPEND);
2857 MNT_KERN_FLAG(MNTK_SUSPEND2);
2858 MNT_KERN_FLAG(MNTK_SUSPENDED);
2859 MNT_KERN_FLAG(MNTK_MPSAFE);
2860 MNT_KERN_FLAG(MNTK_NOKNOTE);
2861 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2862 #undef MNT_KERN_FLAG
2863 if (flags != 0) {
2864 if (buf[0] != '\0')
2865 strlcat(buf, ", ", sizeof(buf));
2866 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2867 "0x%08x", flags);
2868 }
2869 db_printf(" mnt_kern_flag = %s\n", buf);
2870
2871 sp = &mp->mnt_stat;
2872 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2873 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2874 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2875 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2876 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2877 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2878 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2879 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2880 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2881 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2882 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2883 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2884
2885 db_printf(" mnt_cred = { uid=%u ruid=%u",
2886 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2887 if (mp->mnt_cred->cr_prison != NULL)
2888 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2889 db_printf(" }\n");
2890 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2891 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2892 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2893 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2894 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
2895 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2896 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2897 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2898 db_printf(" mnt_markercnt = %d\n", mp->mnt_markercnt);
2899 db_printf(" mnt_holdcnt = %d\n", mp->mnt_holdcnt);
2900 db_printf(" mnt_holdcntwaiters = %d\n", mp->mnt_holdcntwaiters);
2901 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2902 db_printf(" mnt_secondary_accwrites = %d\n",
2903 mp->mnt_secondary_accwrites);
2904 db_printf(" mnt_gjprovider = %s\n",
2905 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2906 db_printf("\n");
2907
2908 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2909 if (vp->v_type != VMARKER) {
2910 vn_printf(vp, "vnode ");
2911 if (db_pager_quit)
2912 break;
2913 }
2914 }
2915 }
2916 #endif /* DDB */
2917
2918 /*
2919 * Fill in a struct xvfsconf based on a struct vfsconf.
2920 */
2921 static void
2922 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2923 {
2924
2925 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2926 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2927 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2928 xvfsp->vfc_flags = vfsp->vfc_flags;
2929 /*
2930 * These are unused in userland, we keep them
2931 * to not break binary compatibility.
2932 */
2933 xvfsp->vfc_vfsops = NULL;
2934 xvfsp->vfc_next = NULL;
2935 }
2936
2937 /*
2938 * Top level filesystem related information gathering.
2939 */
2940 static int
2941 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2942 {
2943 struct vfsconf *vfsp;
2944 struct xvfsconf xvfsp;
2945 int error;
2946
2947 error = 0;
2948 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2949 bzero(&xvfsp, sizeof(xvfsp));
2950 vfsconf2x(vfsp, &xvfsp);
2951 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2952 if (error)
2953 break;
2954 }
2955 return (error);
2956 }
2957
2958 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2959 "S,xvfsconf", "List of all configured filesystems");
2960
2961 #ifndef BURN_BRIDGES
2962 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2963
2964 static int
2965 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2966 {
2967 int *name = (int *)arg1 - 1; /* XXX */
2968 u_int namelen = arg2 + 1; /* XXX */
2969 struct vfsconf *vfsp;
2970 struct xvfsconf xvfsp;
2971
2972 printf("WARNING: userland calling deprecated sysctl, "
2973 "please rebuild world\n");
2974
2975 #if 1 || defined(COMPAT_PRELITE2)
2976 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2977 if (namelen == 1)
2978 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2979 #endif
2980
2981 switch (name[1]) {
2982 case VFS_MAXTYPENUM:
2983 if (namelen != 2)
2984 return (ENOTDIR);
2985 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2986 case VFS_CONF:
2987 if (namelen != 3)
2988 return (ENOTDIR); /* overloaded */
2989 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2990 if (vfsp->vfc_typenum == name[2])
2991 break;
2992 if (vfsp == NULL)
2993 return (EOPNOTSUPP);
2994 bzero(&xvfsp, sizeof(xvfsp));
2995 vfsconf2x(vfsp, &xvfsp);
2996 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2997 }
2998 return (EOPNOTSUPP);
2999 }
3000
3001 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3002 vfs_sysctl, "Generic filesystem");
3003
3004 #if 1 || defined(COMPAT_PRELITE2)
3005
3006 static int
3007 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3008 {
3009 int error;
3010 struct vfsconf *vfsp;
3011 struct ovfsconf ovfs;
3012
3013 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3014 bzero(&ovfs, sizeof(ovfs));
3015 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3016 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3017 ovfs.vfc_index = vfsp->vfc_typenum;
3018 ovfs.vfc_refcount = vfsp->vfc_refcount;
3019 ovfs.vfc_flags = vfsp->vfc_flags;
3020 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3021 if (error)
3022 return error;
3023 }
3024 return 0;
3025 }
3026
3027 #endif /* 1 || COMPAT_PRELITE2 */
3028 #endif /* !BURN_BRIDGES */
3029
3030 #define KINFO_VNODESLOP 10
3031 #ifdef notyet
3032 /*
3033 * Dump vnode list (via sysctl).
3034 */
3035 /* ARGSUSED */
3036 static int
3037 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3038 {
3039 struct xvnode *xvn;
3040 struct thread *td = req->td;
3041 struct mount *mp;
3042 struct vnode *vp;
3043 int error, len, n;
3044
3045 /*
3046 * Stale numvnodes access is not fatal here.
3047 */
3048 req->lock = 0;
3049 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3050 if (!req->oldptr)
3051 /* Make an estimate */
3052 return (SYSCTL_OUT(req, 0, len));
3053
3054 error = sysctl_wire_old_buffer(req, 0);
3055 if (error != 0)
3056 return (error);
3057 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3058 n = 0;
3059 mtx_lock(&mountlist_mtx);
3060 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3061 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3062 continue;
3063 MNT_ILOCK(mp);
3064 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3065 if (n == len)
3066 break;
3067 vref(vp);
3068 xvn[n].xv_size = sizeof *xvn;
3069 xvn[n].xv_vnode = vp;
3070 xvn[n].xv_id = 0; /* XXX compat */
3071 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3072 XV_COPY(usecount);
3073 XV_COPY(writecount);
3074 XV_COPY(holdcnt);
3075 XV_COPY(mount);
3076 XV_COPY(numoutput);
3077 XV_COPY(type);
3078 #undef XV_COPY
3079 xvn[n].xv_flag = vp->v_vflag;
3080
3081 switch (vp->v_type) {
3082 case VREG:
3083 case VDIR:
3084 case VLNK:
3085 break;
3086 case VBLK:
3087 case VCHR:
3088 if (vp->v_rdev == NULL) {
3089 vrele(vp);
3090 continue;
3091 }
3092 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3093 break;
3094 case VSOCK:
3095 xvn[n].xv_socket = vp->v_socket;
3096 break;
3097 case VFIFO:
3098 xvn[n].xv_fifo = vp->v_fifoinfo;
3099 break;
3100 case VNON:
3101 case VBAD:
3102 default:
3103 /* shouldn't happen? */
3104 vrele(vp);
3105 continue;
3106 }
3107 vrele(vp);
3108 ++n;
3109 }
3110 MNT_IUNLOCK(mp);
3111 mtx_lock(&mountlist_mtx);
3112 vfs_unbusy(mp, td);
3113 if (n == len)
3114 break;
3115 }
3116 mtx_unlock(&mountlist_mtx);
3117
3118 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3119 free(xvn, M_TEMP);
3120 return (error);
3121 }
3122
3123 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3124 0, 0, sysctl_vnode, "S,xvnode", "");
3125 #endif
3126
3127 /*
3128 * Unmount all filesystems. The list is traversed in reverse order
3129 * of mounting to avoid dependencies.
3130 */
3131 void
3132 vfs_unmountall(void)
3133 {
3134 struct mount *mp;
3135 struct thread *td;
3136 int error;
3137
3138 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3139 td = curthread;
3140 /*
3141 * Since this only runs when rebooting, it is not interlocked.
3142 */
3143 while(!TAILQ_EMPTY(&mountlist)) {
3144 mp = TAILQ_LAST(&mountlist, mntlist);
3145 error = dounmount(mp, MNT_FORCE, td);
3146 if (error) {
3147 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3148 /*
3149 * XXX: Due to the way in which we mount the root
3150 * file system off of devfs, devfs will generate a
3151 * "busy" warning when we try to unmount it before
3152 * the root. Don't print a warning as a result in
3153 * order to avoid false positive errors that may
3154 * cause needless upset.
3155 */
3156 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3157 printf("unmount of %s failed (",
3158 mp->mnt_stat.f_mntonname);
3159 if (error == EBUSY)
3160 printf("BUSY)\n");
3161 else
3162 printf("%d)\n", error);
3163 }
3164 } else {
3165 /* The unmount has removed mp from the mountlist */
3166 }
3167 }
3168 }
3169
3170 /*
3171 * perform msync on all vnodes under a mount point
3172 * the mount point must be locked.
3173 */
3174 void
3175 vfs_msync(struct mount *mp, int flags)
3176 {
3177 struct vnode *vp, *mvp;
3178 struct vm_object *obj;
3179
3180 MNT_ILOCK(mp);
3181 MNT_VNODE_FOREACH(vp, mp, mvp) {
3182 VI_LOCK(vp);
3183 if ((vp->v_iflag & VI_OBJDIRTY) &&
3184 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3185 MNT_IUNLOCK(mp);
3186 if (!vget(vp,
3187 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3188 curthread)) {
3189 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3190 vput(vp);
3191 MNT_ILOCK(mp);
3192 continue;
3193 }
3194
3195 obj = vp->v_object;
3196 if (obj != NULL) {
3197 VM_OBJECT_LOCK(obj);
3198 vm_object_page_clean(obj, 0, 0,
3199 flags == MNT_WAIT ?
3200 OBJPC_SYNC : OBJPC_NOSYNC);
3201 VM_OBJECT_UNLOCK(obj);
3202 }
3203 vput(vp);
3204 }
3205 MNT_ILOCK(mp);
3206 } else
3207 VI_UNLOCK(vp);
3208 }
3209 MNT_IUNLOCK(mp);
3210 }
3211
3212 /*
3213 * Mark a vnode as free, putting it up for recycling.
3214 */
3215 static void
3216 vfree(struct vnode *vp)
3217 {
3218
3219 CTR1(KTR_VFS, "vfree vp %p", vp);
3220 ASSERT_VI_LOCKED(vp, "vfree");
3221 mtx_lock(&vnode_free_list_mtx);
3222 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3223 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3224 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3225 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3226 ("vfree: Freeing doomed vnode"));
3227 if (vp->v_iflag & VI_AGE) {
3228 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3229 } else {
3230 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3231 }
3232 freevnodes++;
3233 vp->v_iflag &= ~VI_AGE;
3234 vp->v_iflag |= VI_FREE;
3235 mtx_unlock(&vnode_free_list_mtx);
3236 }
3237
3238 /*
3239 * Opposite of vfree() - mark a vnode as in use.
3240 */
3241 static void
3242 vbusy(struct vnode *vp)
3243 {
3244 CTR1(KTR_VFS, "vbusy vp %p", vp);
3245 ASSERT_VI_LOCKED(vp, "vbusy");
3246 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3247 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3248
3249 mtx_lock(&vnode_free_list_mtx);
3250 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3251 freevnodes--;
3252 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3253 mtx_unlock(&vnode_free_list_mtx);
3254 }
3255
3256 static void
3257 destroy_vpollinfo(struct vpollinfo *vi)
3258 {
3259 knlist_destroy(&vi->vpi_selinfo.si_note);
3260 mtx_destroy(&vi->vpi_lock);
3261 uma_zfree(vnodepoll_zone, vi);
3262 }
3263
3264 /*
3265 * Initalize per-vnode helper structure to hold poll-related state.
3266 */
3267 void
3268 v_addpollinfo(struct vnode *vp)
3269 {
3270 struct vpollinfo *vi;
3271
3272 if (vp->v_pollinfo != NULL)
3273 return;
3274 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3275 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3276 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3277 vfs_knlunlock, NULL);
3278 VI_LOCK(vp);
3279 if (vp->v_pollinfo != NULL) {
3280 VI_UNLOCK(vp);
3281 destroy_vpollinfo(vi);
3282 return;
3283 }
3284 vp->v_pollinfo = vi;
3285 VI_UNLOCK(vp);
3286 }
3287
3288 /*
3289 * Record a process's interest in events which might happen to
3290 * a vnode. Because poll uses the historic select-style interface
3291 * internally, this routine serves as both the ``check for any
3292 * pending events'' and the ``record my interest in future events''
3293 * functions. (These are done together, while the lock is held,
3294 * to avoid race conditions.)
3295 */
3296 int
3297 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3298 {
3299
3300 v_addpollinfo(vp);
3301 mtx_lock(&vp->v_pollinfo->vpi_lock);
3302 if (vp->v_pollinfo->vpi_revents & events) {
3303 /*
3304 * This leaves events we are not interested
3305 * in available for the other process which
3306 * which presumably had requested them
3307 * (otherwise they would never have been
3308 * recorded).
3309 */
3310 events &= vp->v_pollinfo->vpi_revents;
3311 vp->v_pollinfo->vpi_revents &= ~events;
3312
3313 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3314 return events;
3315 }
3316 vp->v_pollinfo->vpi_events |= events;
3317 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3318 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3319 return 0;
3320 }
3321
3322 /*
3323 * Routine to create and manage a filesystem syncer vnode.
3324 */
3325 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3326 static int sync_fsync(struct vop_fsync_args *);
3327 static int sync_inactive(struct vop_inactive_args *);
3328 static int sync_reclaim(struct vop_reclaim_args *);
3329
3330 static struct vop_vector sync_vnodeops = {
3331 .vop_bypass = VOP_EOPNOTSUPP,
3332 .vop_close = sync_close, /* close */
3333 .vop_fsync = sync_fsync, /* fsync */
3334 .vop_inactive = sync_inactive, /* inactive */
3335 .vop_reclaim = sync_reclaim, /* reclaim */
3336 .vop_lock1 = vop_stdlock, /* lock */
3337 .vop_unlock = vop_stdunlock, /* unlock */
3338 .vop_islocked = vop_stdislocked, /* islocked */
3339 };
3340
3341 /*
3342 * Create a new filesystem syncer vnode for the specified mount point.
3343 */
3344 int
3345 vfs_allocate_syncvnode(struct mount *mp)
3346 {
3347 struct vnode *vp;
3348 static long start, incr, next;
3349 int error;
3350
3351 /* Allocate a new vnode */
3352 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3353 mp->mnt_syncer = NULL;
3354 return (error);
3355 }
3356 vp->v_type = VNON;
3357 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
3358 vp->v_vflag |= VV_FORCEINSMQ;
3359 error = insmntque(vp, mp);
3360 if (error != 0)
3361 panic("vfs_allocate_syncvnode: insmntque failed");
3362 vp->v_vflag &= ~VV_FORCEINSMQ;
3363 VOP_UNLOCK(vp, 0, curthread);
3364 /*
3365 * Place the vnode onto the syncer worklist. We attempt to
3366 * scatter them about on the list so that they will go off
3367 * at evenly distributed times even if all the filesystems
3368 * are mounted at once.
3369 */
3370 next += incr;
3371 if (next == 0 || next > syncer_maxdelay) {
3372 start /= 2;
3373 incr /= 2;
3374 if (start == 0) {
3375 start = syncer_maxdelay / 2;
3376 incr = syncer_maxdelay;
3377 }
3378 next = start;
3379 }
3380 VI_LOCK(vp);
3381 vn_syncer_add_to_worklist(&vp->v_bufobj,
3382 syncdelay > 0 ? next % syncdelay : 0);
3383 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3384 mtx_lock(&sync_mtx);
3385 sync_vnode_count++;
3386 mtx_unlock(&sync_mtx);
3387 VI_UNLOCK(vp);
3388 mp->mnt_syncer = vp;
3389 return (0);
3390 }
3391
3392 /*
3393 * Do a lazy sync of the filesystem.
3394 */
3395 static int
3396 sync_fsync(struct vop_fsync_args *ap)
3397 {
3398 struct vnode *syncvp = ap->a_vp;
3399 struct mount *mp = syncvp->v_mount;
3400 struct thread *td = ap->a_td;
3401 int error;
3402 struct bufobj *bo;
3403
3404 /*
3405 * We only need to do something if this is a lazy evaluation.
3406 */
3407 if (ap->a_waitfor != MNT_LAZY)
3408 return (0);
3409
3410 /*
3411 * Move ourselves to the back of the sync list.
3412 */
3413 bo = &syncvp->v_bufobj;
3414 BO_LOCK(bo);
3415 vn_syncer_add_to_worklist(bo, syncdelay);
3416 BO_UNLOCK(bo);
3417
3418 /*
3419 * Walk the list of vnodes pushing all that are dirty and
3420 * not already on the sync list.
3421 */
3422 mtx_lock(&mountlist_mtx);
3423 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3424 mtx_unlock(&mountlist_mtx);
3425 return (0);
3426 }
3427 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3428 vfs_unbusy(mp, td);
3429 return (0);
3430 }
3431 MNT_ILOCK(mp);
3432 mp->mnt_noasync++;
3433 mp->mnt_kern_flag &= ~MNTK_ASYNC;
3434 MNT_IUNLOCK(mp);
3435 vfs_msync(mp, MNT_NOWAIT);
3436 error = VFS_SYNC(mp, MNT_LAZY, td);
3437 MNT_ILOCK(mp);
3438 mp->mnt_noasync--;
3439 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3440 mp->mnt_kern_flag |= MNTK_ASYNC;
3441 MNT_IUNLOCK(mp);
3442 vn_finished_write(mp);
3443 vfs_unbusy(mp, td);
3444 return (error);
3445 }
3446
3447 /*
3448 * The syncer vnode is no referenced.
3449 */
3450 static int
3451 sync_inactive(struct vop_inactive_args *ap)
3452 {
3453
3454 vgone(ap->a_vp);
3455 return (0);
3456 }
3457
3458 /*
3459 * The syncer vnode is no longer needed and is being decommissioned.
3460 *
3461 * Modifications to the worklist must be protected by sync_mtx.
3462 */
3463 static int
3464 sync_reclaim(struct vop_reclaim_args *ap)
3465 {
3466 struct vnode *vp = ap->a_vp;
3467 struct bufobj *bo;
3468
3469 VI_LOCK(vp);
3470 bo = &vp->v_bufobj;
3471 vp->v_mount->mnt_syncer = NULL;
3472 if (bo->bo_flag & BO_ONWORKLST) {
3473 mtx_lock(&sync_mtx);
3474 LIST_REMOVE(bo, bo_synclist);
3475 syncer_worklist_len--;
3476 sync_vnode_count--;
3477 mtx_unlock(&sync_mtx);
3478 bo->bo_flag &= ~BO_ONWORKLST;
3479 }
3480 VI_UNLOCK(vp);
3481
3482 return (0);
3483 }
3484
3485 /*
3486 * Check if vnode represents a disk device
3487 */
3488 int
3489 vn_isdisk(struct vnode *vp, int *errp)
3490 {
3491 int error;
3492
3493 error = 0;
3494 dev_lock();
3495 if (vp->v_type != VCHR)
3496 error = ENOTBLK;
3497 else if (vp->v_rdev == NULL)
3498 error = ENXIO;
3499 else if (vp->v_rdev->si_devsw == NULL)
3500 error = ENXIO;
3501 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3502 error = ENOTBLK;
3503 dev_unlock();
3504 if (errp != NULL)
3505 *errp = error;
3506 return (error == 0);
3507 }
3508
3509 /*
3510 * Common filesystem object access control check routine. Accepts a
3511 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3512 * and optional call-by-reference privused argument allowing vaccess()
3513 * to indicate to the caller whether privilege was used to satisfy the
3514 * request (obsoleted). Returns 0 on success, or an errno on failure.
3515 */
3516 int
3517 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3518 mode_t acc_mode, struct ucred *cred, int *privused)
3519 {
3520 mode_t dac_granted;
3521 mode_t priv_granted;
3522
3523 /*
3524 * Look for a normal, non-privileged way to access the file/directory
3525 * as requested. If it exists, go with that.
3526 */
3527
3528 if (privused != NULL)
3529 *privused = 0;
3530
3531 dac_granted = 0;
3532
3533 /* Check the owner. */
3534 if (cred->cr_uid == file_uid) {
3535 dac_granted |= VADMIN;
3536 if (file_mode & S_IXUSR)
3537 dac_granted |= VEXEC;
3538 if (file_mode & S_IRUSR)
3539 dac_granted |= VREAD;
3540 if (file_mode & S_IWUSR)
3541 dac_granted |= (VWRITE | VAPPEND);
3542
3543 if ((acc_mode & dac_granted) == acc_mode)
3544 return (0);
3545
3546 goto privcheck;
3547 }
3548
3549 /* Otherwise, check the groups (first match) */
3550 if (groupmember(file_gid, cred)) {
3551 if (file_mode & S_IXGRP)
3552 dac_granted |= VEXEC;
3553 if (file_mode & S_IRGRP)
3554 dac_granted |= VREAD;
3555 if (file_mode & S_IWGRP)
3556 dac_granted |= (VWRITE | VAPPEND);
3557
3558 if ((acc_mode & dac_granted) == acc_mode)
3559 return (0);
3560
3561 goto privcheck;
3562 }
3563
3564 /* Otherwise, check everyone else. */
3565 if (file_mode & S_IXOTH)
3566 dac_granted |= VEXEC;
3567 if (file_mode & S_IROTH)
3568 dac_granted |= VREAD;
3569 if (file_mode & S_IWOTH)
3570 dac_granted |= (VWRITE | VAPPEND);
3571 if ((acc_mode & dac_granted) == acc_mode)
3572 return (0);
3573
3574 privcheck:
3575 /*
3576 * Build a privilege mask to determine if the set of privileges
3577 * satisfies the requirements when combined with the granted mask
3578 * from above. For each privilege, if the privilege is required,
3579 * bitwise or the request type onto the priv_granted mask.
3580 */
3581 priv_granted = 0;
3582
3583 if (type == VDIR) {
3584 /*
3585 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3586 * requests, instead of PRIV_VFS_EXEC.
3587 */
3588 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3589 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3590 priv_granted |= VEXEC;
3591 } else {
3592 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3593 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3594 priv_granted |= VEXEC;
3595 }
3596
3597 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3598 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3599 priv_granted |= VREAD;
3600
3601 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3602 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3603 priv_granted |= (VWRITE | VAPPEND);
3604
3605 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3606 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3607 priv_granted |= VADMIN;
3608
3609 if ((acc_mode & (priv_granted | dac_granted)) == acc_mode) {
3610 /* XXX audit: privilege used */
3611 if (privused != NULL)
3612 *privused = 1;
3613 return (0);
3614 }
3615
3616 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3617 }
3618
3619 /*
3620 * Credential check based on process requesting service, and per-attribute
3621 * permissions.
3622 */
3623 int
3624 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3625 struct thread *td, int access)
3626 {
3627
3628 /*
3629 * Kernel-invoked always succeeds.
3630 */
3631 if (cred == NOCRED)
3632 return (0);
3633
3634 /*
3635 * Do not allow privileged processes in jail to directly manipulate
3636 * system attributes.
3637 */
3638 switch (attrnamespace) {
3639 case EXTATTR_NAMESPACE_SYSTEM:
3640 /* Potentially should be: return (EPERM); */
3641 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3642 case EXTATTR_NAMESPACE_USER:
3643 return (VOP_ACCESS(vp, access, cred, td));
3644 default:
3645 return (EPERM);
3646 }
3647 }
3648
3649 #ifdef DEBUG_VFS_LOCKS
3650 /*
3651 * This only exists to supress warnings from unlocked specfs accesses. It is
3652 * no longer ok to have an unlocked VFS.
3653 */
3654 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3655
3656 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3657 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3658
3659 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3660 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3661
3662 int vfs_badlock_print = 1; /* Print lock violations. */
3663 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3664
3665 #ifdef KDB
3666 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3667 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3668 #endif
3669
3670 static void
3671 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3672 {
3673
3674 #ifdef KDB
3675 if (vfs_badlock_backtrace)
3676 kdb_backtrace();
3677 #endif
3678 if (vfs_badlock_print)
3679 printf("%s: %p %s\n", str, (void *)vp, msg);
3680 if (vfs_badlock_ddb)
3681 kdb_enter_why(KDB_WHY_VFSLOCK, "lock violation");
3682 }
3683
3684 void
3685 assert_vi_locked(struct vnode *vp, const char *str)
3686 {
3687
3688 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3689 vfs_badlock("interlock is not locked but should be", str, vp);
3690 }
3691
3692 void
3693 assert_vi_unlocked(struct vnode *vp, const char *str)
3694 {
3695
3696 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3697 vfs_badlock("interlock is locked but should not be", str, vp);
3698 }
3699
3700 void
3701 assert_vop_locked(struct vnode *vp, const char *str)
3702 {
3703
3704 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
3705 vfs_badlock("is not locked but should be", str, vp);
3706 }
3707
3708 void
3709 assert_vop_unlocked(struct vnode *vp, const char *str)
3710 {
3711
3712 if (vp && !IGNORE_LOCK(vp) &&
3713 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
3714 vfs_badlock("is locked but should not be", str, vp);
3715 }
3716
3717 void
3718 assert_vop_elocked(struct vnode *vp, const char *str)
3719 {
3720
3721 if (vp && !IGNORE_LOCK(vp) &&
3722 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
3723 vfs_badlock("is not exclusive locked but should be", str, vp);
3724 }
3725
3726 #if 0
3727 void
3728 assert_vop_elocked_other(struct vnode *vp, const char *str)
3729 {
3730
3731 if (vp && !IGNORE_LOCK(vp) &&
3732 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
3733 vfs_badlock("is not exclusive locked by another thread",
3734 str, vp);
3735 }
3736
3737 void
3738 assert_vop_slocked(struct vnode *vp, const char *str)
3739 {
3740
3741 if (vp && !IGNORE_LOCK(vp) &&
3742 VOP_ISLOCKED(vp, curthread) != LK_SHARED)
3743 vfs_badlock("is not locked shared but should be", str, vp);
3744 }
3745 #endif /* 0 */
3746 #endif /* DEBUG_VFS_LOCKS */
3747
3748 void
3749 vop_rename_pre(void *ap)
3750 {
3751 struct vop_rename_args *a = ap;
3752
3753 #ifdef DEBUG_VFS_LOCKS
3754 if (a->a_tvp)
3755 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3756 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3757 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3758 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3759
3760 /* Check the source (from). */
3761 if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3762 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3763 if (a->a_tvp != a->a_fvp)
3764 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3765
3766 /* Check the target. */
3767 if (a->a_tvp)
3768 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3769 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3770 #endif
3771 if (a->a_tdvp != a->a_fdvp)
3772 vhold(a->a_fdvp);
3773 if (a->a_tvp != a->a_fvp)
3774 vhold(a->a_fvp);
3775 vhold(a->a_tdvp);
3776 if (a->a_tvp)
3777 vhold(a->a_tvp);
3778 }
3779
3780 void
3781 vop_strategy_pre(void *ap)
3782 {
3783 #ifdef DEBUG_VFS_LOCKS
3784 struct vop_strategy_args *a;
3785 struct buf *bp;
3786
3787 a = ap;
3788 bp = a->a_bp;
3789
3790 /*
3791 * Cluster ops lock their component buffers but not the IO container.
3792 */
3793 if ((bp->b_flags & B_CLUSTER) != 0)
3794 return;
3795
3796 if (BUF_REFCNT(bp) < 1) {
3797 if (vfs_badlock_print)
3798 printf(
3799 "VOP_STRATEGY: bp is not locked but should be\n");
3800 if (vfs_badlock_ddb)
3801 kdb_enter_why(KDB_WHY_VFSLOCK, "lock violation");
3802 }
3803 #endif
3804 }
3805
3806 void
3807 vop_lookup_pre(void *ap)
3808 {
3809 #ifdef DEBUG_VFS_LOCKS
3810 struct vop_lookup_args *a;
3811 struct vnode *dvp;
3812
3813 a = ap;
3814 dvp = a->a_dvp;
3815 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3816 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3817 #endif
3818 }
3819
3820 void
3821 vop_lookup_post(void *ap, int rc)
3822 {
3823 #ifdef DEBUG_VFS_LOCKS
3824 struct vop_lookup_args *a;
3825 struct vnode *dvp;
3826 struct vnode *vp;
3827
3828 a = ap;
3829 dvp = a->a_dvp;
3830 vp = *(a->a_vpp);
3831
3832 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3833 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3834
3835 if (!rc)
3836 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3837 #endif
3838 }
3839
3840 void
3841 vop_lock_pre(void *ap)
3842 {
3843 #ifdef DEBUG_VFS_LOCKS
3844 struct vop_lock1_args *a = ap;
3845
3846 if ((a->a_flags & LK_INTERLOCK) == 0)
3847 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3848 else
3849 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3850 #endif
3851 }
3852
3853 void
3854 vop_lock_post(void *ap, int rc)
3855 {
3856 #ifdef DEBUG_VFS_LOCKS
3857 struct vop_lock1_args *a = ap;
3858
3859 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3860 if (rc == 0)
3861 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3862 #endif
3863 }
3864
3865 void
3866 vop_unlock_pre(void *ap)
3867 {
3868 #ifdef DEBUG_VFS_LOCKS
3869 struct vop_unlock_args *a = ap;
3870
3871 if (a->a_flags & LK_INTERLOCK)
3872 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3873 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3874 #endif
3875 }
3876
3877 void
3878 vop_unlock_post(void *ap, int rc)
3879 {
3880 #ifdef DEBUG_VFS_LOCKS
3881 struct vop_unlock_args *a = ap;
3882
3883 if (a->a_flags & LK_INTERLOCK)
3884 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3885 #endif
3886 }
3887
3888 void
3889 vop_create_post(void *ap, int rc)
3890 {
3891 struct vop_create_args *a = ap;
3892
3893 if (!rc)
3894 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3895 }
3896
3897 void
3898 vop_link_post(void *ap, int rc)
3899 {
3900 struct vop_link_args *a = ap;
3901
3902 if (!rc) {
3903 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3904 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3905 }
3906 }
3907
3908 void
3909 vop_mkdir_post(void *ap, int rc)
3910 {
3911 struct vop_mkdir_args *a = ap;
3912
3913 if (!rc)
3914 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3915 }
3916
3917 void
3918 vop_mknod_post(void *ap, int rc)
3919 {
3920 struct vop_mknod_args *a = ap;
3921
3922 if (!rc)
3923 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3924 }
3925
3926 void
3927 vop_remove_post(void *ap, int rc)
3928 {
3929 struct vop_remove_args *a = ap;
3930
3931 if (!rc) {
3932 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3933 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3934 }
3935 }
3936
3937 void
3938 vop_rename_post(void *ap, int rc)
3939 {
3940 struct vop_rename_args *a = ap;
3941
3942 if (!rc) {
3943 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3944 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3945 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3946 if (a->a_tvp)
3947 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3948 }
3949 if (a->a_tdvp != a->a_fdvp)
3950 vdrop(a->a_fdvp);
3951 if (a->a_tvp != a->a_fvp)
3952 vdrop(a->a_fvp);
3953 vdrop(a->a_tdvp);
3954 if (a->a_tvp)
3955 vdrop(a->a_tvp);
3956 }
3957
3958 void
3959 vop_rmdir_post(void *ap, int rc)
3960 {
3961 struct vop_rmdir_args *a = ap;
3962
3963 if (!rc) {
3964 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3965 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3966 }
3967 }
3968
3969 void
3970 vop_setattr_post(void *ap, int rc)
3971 {
3972 struct vop_setattr_args *a = ap;
3973
3974 if (!rc)
3975 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3976 }
3977
3978 void
3979 vop_symlink_post(void *ap, int rc)
3980 {
3981 struct vop_symlink_args *a = ap;
3982
3983 if (!rc)
3984 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3985 }
3986
3987 static struct knlist fs_knlist;
3988
3989 static void
3990 vfs_event_init(void *arg)
3991 {
3992 knlist_init_mtx(&fs_knlist, NULL);
3993 }
3994 /* XXX - correct order? */
3995 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3996
3997 void
3998 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3999 {
4000
4001 KNOTE_UNLOCKED(&fs_knlist, event);
4002 }
4003
4004 static int filt_fsattach(struct knote *kn);
4005 static void filt_fsdetach(struct knote *kn);
4006 static int filt_fsevent(struct knote *kn, long hint);
4007
4008 struct filterops fs_filtops =
4009 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
4010
4011 static int
4012 filt_fsattach(struct knote *kn)
4013 {
4014
4015 kn->kn_flags |= EV_CLEAR;
4016 knlist_add(&fs_knlist, kn, 0);
4017 return (0);
4018 }
4019
4020 static void
4021 filt_fsdetach(struct knote *kn)
4022 {
4023
4024 knlist_remove(&fs_knlist, kn, 0);
4025 }
4026
4027 static int
4028 filt_fsevent(struct knote *kn, long hint)
4029 {
4030
4031 kn->kn_fflags |= hint;
4032 return (kn->kn_fflags != 0);
4033 }
4034
4035 static int
4036 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4037 {
4038 struct vfsidctl vc;
4039 int error;
4040 struct mount *mp;
4041
4042 error = SYSCTL_IN(req, &vc, sizeof(vc));
4043 if (error)
4044 return (error);
4045 if (vc.vc_vers != VFS_CTL_VERS1)
4046 return (EINVAL);
4047 mp = vfs_getvfs(&vc.vc_fsid);
4048 if (mp == NULL)
4049 return (ENOENT);
4050 /* ensure that a specific sysctl goes to the right filesystem. */
4051 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4052 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4053 vfs_rel(mp);
4054 return (EINVAL);
4055 }
4056 VCTLTOREQ(&vc, req);
4057 error = VFS_SYSCTL(mp, vc.vc_op, req);
4058 vfs_rel(mp);
4059 return (error);
4060 }
4061
4062 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4063 "Sysctl by fsid");
4064
4065 /*
4066 * Function to initialize a va_filerev field sensibly.
4067 * XXX: Wouldn't a random number make a lot more sense ??
4068 */
4069 u_quad_t
4070 init_va_filerev(void)
4071 {
4072 struct bintime bt;
4073
4074 getbinuptime(&bt);
4075 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4076 }
4077
4078 static int filt_vfsread(struct knote *kn, long hint);
4079 static int filt_vfswrite(struct knote *kn, long hint);
4080 static int filt_vfsvnode(struct knote *kn, long hint);
4081 static void filt_vfsdetach(struct knote *kn);
4082 static struct filterops vfsread_filtops =
4083 { 1, NULL, filt_vfsdetach, filt_vfsread };
4084 static struct filterops vfswrite_filtops =
4085 { 1, NULL, filt_vfsdetach, filt_vfswrite };
4086 static struct filterops vfsvnode_filtops =
4087 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
4088
4089 static void
4090 vfs_knllock(void *arg)
4091 {
4092 struct vnode *vp = arg;
4093
4094 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
4095 }
4096
4097 static void
4098 vfs_knlunlock(void *arg)
4099 {
4100 struct vnode *vp = arg;
4101
4102 VOP_UNLOCK(vp, 0, curthread);
4103 }
4104
4105 int
4106 vfs_kqfilter(struct vop_kqfilter_args *ap)
4107 {
4108 struct vnode *vp = ap->a_vp;
4109 struct knote *kn = ap->a_kn;
4110 struct knlist *knl;
4111
4112 switch (kn->kn_filter) {
4113 case EVFILT_READ:
4114 kn->kn_fop = &vfsread_filtops;
4115 break;
4116 case EVFILT_WRITE:
4117 kn->kn_fop = &vfswrite_filtops;
4118 break;
4119 case EVFILT_VNODE:
4120 kn->kn_fop = &vfsvnode_filtops;
4121 break;
4122 default:
4123 return (EINVAL);
4124 }
4125
4126 kn->kn_hook = (caddr_t)vp;
4127
4128 v_addpollinfo(vp);
4129 if (vp->v_pollinfo == NULL)
4130 return (ENOMEM);
4131 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4132 knlist_add(knl, kn, 0);
4133
4134 return (0);
4135 }
4136
4137 /*
4138 * Detach knote from vnode
4139 */
4140 static void
4141 filt_vfsdetach(struct knote *kn)
4142 {
4143 struct vnode *vp = (struct vnode *)kn->kn_hook;
4144
4145 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4146 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4147 }
4148
4149 /*ARGSUSED*/
4150 static int
4151 filt_vfsread(struct knote *kn, long hint)
4152 {
4153 struct vnode *vp = (struct vnode *)kn->kn_hook;
4154 struct vattr va;
4155 int res;
4156
4157 /*
4158 * filesystem is gone, so set the EOF flag and schedule
4159 * the knote for deletion.
4160 */
4161 if (hint == NOTE_REVOKE) {
4162 VI_LOCK(vp);
4163 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4164 VI_UNLOCK(vp);
4165 return (1);
4166 }
4167
4168 if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
4169 return (0);
4170
4171 VI_LOCK(vp);
4172 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4173 res = (kn->kn_data != 0);
4174 VI_UNLOCK(vp);
4175 return (res);
4176 }
4177
4178 /*ARGSUSED*/
4179 static int
4180 filt_vfswrite(struct knote *kn, long hint)
4181 {
4182 struct vnode *vp = (struct vnode *)kn->kn_hook;
4183
4184 VI_LOCK(vp);
4185
4186 /*
4187 * filesystem is gone, so set the EOF flag and schedule
4188 * the knote for deletion.
4189 */
4190 if (hint == NOTE_REVOKE)
4191 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4192
4193 kn->kn_data = 0;
4194 VI_UNLOCK(vp);
4195 return (1);
4196 }
4197
4198 static int
4199 filt_vfsvnode(struct knote *kn, long hint)
4200 {
4201 struct vnode *vp = (struct vnode *)kn->kn_hook;
4202 int res;
4203
4204 VI_LOCK(vp);
4205 if (kn->kn_sfflags & hint)
4206 kn->kn_fflags |= hint;
4207 if (hint == NOTE_REVOKE) {
4208 kn->kn_flags |= EV_EOF;
4209 VI_UNLOCK(vp);
4210 return (1);
4211 }
4212 res = (kn->kn_fflags != 0);
4213 VI_UNLOCK(vp);
4214 return (res);
4215 }
4216
4217 int
4218 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4219 {
4220 int error;
4221
4222 if (dp->d_reclen > ap->a_uio->uio_resid)
4223 return (ENAMETOOLONG);
4224 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4225 if (error) {
4226 if (ap->a_ncookies != NULL) {
4227 if (ap->a_cookies != NULL)
4228 free(ap->a_cookies, M_TEMP);
4229 ap->a_cookies = NULL;
4230 *ap->a_ncookies = 0;
4231 }
4232 return (error);
4233 }
4234 if (ap->a_ncookies == NULL)
4235 return (0);
4236
4237 KASSERT(ap->a_cookies,
4238 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4239
4240 *ap->a_cookies = realloc(*ap->a_cookies,
4241 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4242 (*ap->a_cookies)[*ap->a_ncookies] = off;
4243 return (0);
4244 }
4245
4246 /*
4247 * Mark for update the access time of the file if the filesystem
4248 * supports VA_MARK_ATIME. This functionality is used by execve
4249 * and mmap, so we want to avoid the synchronous I/O implied by
4250 * directly setting va_atime for the sake of efficiency.
4251 */
4252 void
4253 vfs_mark_atime(struct vnode *vp, struct thread *td)
4254 {
4255 struct vattr atimeattr;
4256
4257 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
4258 VATTR_NULL(&atimeattr);
4259 atimeattr.va_vaflags |= VA_MARK_ATIME;
4260 (void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td);
4261 }
4262 }
Cache object: 85e6ef4e59102fd0e389d04e1c3fd1b3
|