FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_subr.c
1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
35 */
36
37 /*
38 * External virtual filesystem routines
39 */
40
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD: releng/6.1/sys/kern/vfs_subr.c 158286 2006-05-04 07:42:52Z scottl $");
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/kdb.h>
59 #include <sys/kernel.h>
60 #include <sys/kthread.h>
61 #include <sys/mac.h>
62 #include <sys/malloc.h>
63 #include <sys/mount.h>
64 #include <sys/namei.h>
65 #include <sys/reboot.h>
66 #include <sys/sleepqueue.h>
67 #include <sys/stat.h>
68 #include <sys/sysctl.h>
69 #include <sys/syslog.h>
70 #include <sys/vmmeter.h>
71 #include <sys/vnode.h>
72
73 #include <machine/stdarg.h>
74
75 #include <vm/vm.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_extern.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_kern.h>
82 #include <vm/uma.h>
83
84 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
85
86 static void delmntque(struct vnode *vp);
87 static void insmntque(struct vnode *vp, struct mount *mp);
88 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
89 int slpflag, int slptimeo);
90 static void syncer_shutdown(void *arg, int howto);
91 static int vtryrecycle(struct vnode *vp);
92 static void vbusy(struct vnode *vp);
93 static void vdropl(struct vnode *vp);
94 static void vinactive(struct vnode *, struct thread *);
95 static void v_incr_usecount(struct vnode *);
96 static void v_decr_usecount(struct vnode *);
97 static void v_decr_useonly(struct vnode *);
98 static void v_upgrade_usecount(struct vnode *);
99 static void vfree(struct vnode *);
100 static void vnlru_free(int);
101 static void vdestroy(struct vnode *);
102 static void vgonel(struct vnode *);
103 static void vfs_knllock(void *arg);
104 static void vfs_knlunlock(void *arg);
105 static int vfs_knllocked(void *arg);
106
107
108 /*
109 * Enable Giant pushdown based on whether or not the vm is mpsafe in this
110 * build. Without mpsafevm the buffer cache can not run Giant free.
111 */
112 #if defined(__alpha__) || defined(__amd64__) || defined(__i386__) || \
113 defined(__sparc64__)
114 int mpsafe_vfs = 1;
115 #else
116 int mpsafe_vfs;
117 #endif
118 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
119 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
120 "MPSAFE VFS");
121
122 /*
123 * Number of vnodes in existence. Increased whenever getnewvnode()
124 * allocates a new vnode, never decreased.
125 */
126 static unsigned long numvnodes;
127
128 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
129
130 /*
131 * Conversion tables for conversion from vnode types to inode formats
132 * and back.
133 */
134 enum vtype iftovt_tab[16] = {
135 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
137 };
138 int vttoif_tab[10] = {
139 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
141 };
142
143 /*
144 * List of vnodes that are ready for recycling.
145 */
146 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
147
148 /*
149 * Free vnode target. Free vnodes may simply be files which have been stat'd
150 * but not read. This is somewhat common, and a small cache of such files
151 * should be kept to avoid recreation costs.
152 */
153 static u_long wantfreevnodes;
154 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
155 /* Number of vnodes in the free list. */
156 static u_long freevnodes;
157 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
158
159 /*
160 * Various variables used for debugging the new implementation of
161 * reassignbuf().
162 * XXX these are probably of (very) limited utility now.
163 */
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
166
167 /*
168 * Cache for the mount type id assigned to NFS. This is used for
169 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
170 */
171 int nfs_mount_type = -1;
172
173 /* To keep more than one thread at a time from running vfs_getnewfsid */
174 static struct mtx mntid_mtx;
175
176 /*
177 * Lock for any access to the following:
178 * vnode_free_list
179 * numvnodes
180 * freevnodes
181 */
182 static struct mtx vnode_free_list_mtx;
183
184 /* Publicly exported FS */
185 struct nfs_public nfs_pub;
186
187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
188 static uma_zone_t vnode_zone;
189 static uma_zone_t vnodepoll_zone;
190
191 /* Set to 1 to print out reclaim of active vnodes */
192 int prtactive;
193
194 /*
195 * The workitem queue.
196 *
197 * It is useful to delay writes of file data and filesystem metadata
198 * for tens of seconds so that quickly created and deleted files need
199 * not waste disk bandwidth being created and removed. To realize this,
200 * we append vnodes to a "workitem" queue. When running with a soft
201 * updates implementation, most pending metadata dependencies should
202 * not wait for more than a few seconds. Thus, mounted on block devices
203 * are delayed only about a half the time that file data is delayed.
204 * Similarly, directory updates are more critical, so are only delayed
205 * about a third the time that file data is delayed. Thus, there are
206 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207 * one each second (driven off the filesystem syncer process). The
208 * syncer_delayno variable indicates the next queue that is to be processed.
209 * Items that need to be processed soon are placed in this queue:
210 *
211 * syncer_workitem_pending[syncer_delayno]
212 *
213 * A delay of fifteen seconds is done by placing the request fifteen
214 * entries later in the queue:
215 *
216 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
217 *
218 */
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending;
223 /*
224 * The sync_mtx protects:
225 * bo->bo_synclist
226 * sync_vnode_count
227 * syncer_delayno
228 * syncer_state
229 * syncer_workitem_pending
230 * syncer_worklist_len
231 * rushjob
232 */
233 static struct mtx sync_mtx;
234
235 #define SYNCER_MAXDELAY 32
236 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
237 static int syncdelay = 30; /* max time to delay syncing data */
238 static int filedelay = 30; /* time to delay syncing files */
239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
240 static int dirdelay = 29; /* time to delay syncing directories */
241 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
242 static int metadelay = 28; /* time to delay syncing metadata */
243 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
244 static int rushjob; /* number of slots to run ASAP */
245 static int stat_rush_requests; /* number of times I/O speeded up */
246 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
247
248 /*
249 * When shutting down the syncer, run it at four times normal speed.
250 */
251 #define SYNCER_SHUTDOWN_SPEEDUP 4
252 static int sync_vnode_count;
253 static int syncer_worklist_len;
254 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
255 syncer_state;
256
257 /*
258 * Number of vnodes we want to exist at any one time. This is mostly used
259 * to size hash tables in vnode-related code. It is normally not used in
260 * getnewvnode(), as wantfreevnodes is normally nonzero.)
261 *
262 * XXX desiredvnodes is historical cruft and should not exist.
263 */
264 int desiredvnodes;
265 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
266 &desiredvnodes, 0, "Maximum number of vnodes");
267 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
268 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
269 static int vnlru_nowhere;
270 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
271 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
272
273 /*
274 * Macros to control when a vnode is freed and recycled. All require
275 * the vnode interlock.
276 */
277 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
278 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
280
281
282 /*
283 * Initialize the vnode management data structures.
284 */
285 #ifndef MAXVNODES_MAX
286 #define MAXVNODES_MAX 100000
287 #endif
288 static void
289 vntblinit(void *dummy __unused)
290 {
291
292 /*
293 * Desiredvnodes is a function of the physical memory size and
294 * the kernel's heap size. Specifically, desiredvnodes scales
295 * in proportion to the physical memory size until two fifths
296 * of the kernel's heap size is consumed by vnodes and vm
297 * objects.
298 */
299 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
300 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
301 if (desiredvnodes > MAXVNODES_MAX) {
302 if (bootverbose)
303 printf("Reducing kern.maxvnodes %d -> %d\n",
304 desiredvnodes, MAXVNODES_MAX);
305 desiredvnodes = MAXVNODES_MAX;
306 }
307 wantfreevnodes = desiredvnodes / 4;
308 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
309 TAILQ_INIT(&vnode_free_list);
310 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
311 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
312 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
313 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
314 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
315 /*
316 * Initialize the filesystem syncer.
317 */
318 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
319 &syncer_mask);
320 syncer_maxdelay = syncer_mask + 1;
321 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
322 }
323 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
324
325
326 /*
327 * Mark a mount point as busy. Used to synchronize access and to delay
328 * unmounting. Interlock is not released on failure.
329 */
330 int
331 vfs_busy(mp, flags, interlkp, td)
332 struct mount *mp;
333 int flags;
334 struct mtx *interlkp;
335 struct thread *td;
336 {
337 int lkflags;
338
339 MNT_ILOCK(mp);
340 MNT_REF(mp);
341 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
342 if (flags & LK_NOWAIT) {
343 MNT_REL(mp);
344 MNT_IUNLOCK(mp);
345 return (ENOENT);
346 }
347 if (interlkp)
348 mtx_unlock(interlkp);
349 mp->mnt_kern_flag |= MNTK_MWAIT;
350 /*
351 * Since all busy locks are shared except the exclusive
352 * lock granted when unmounting, the only place that a
353 * wakeup needs to be done is at the release of the
354 * exclusive lock at the end of dounmount.
355 */
356 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
357 MNT_REL(mp);
358 MNT_IUNLOCK(mp);
359 if (interlkp)
360 mtx_lock(interlkp);
361 return (ENOENT);
362 }
363 if (interlkp)
364 mtx_unlock(interlkp);
365 lkflags = LK_SHARED | LK_INTERLOCK;
366 if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp), td))
367 panic("vfs_busy: unexpected lock failure");
368 vfs_rel(mp);
369 return (0);
370 }
371
372 /*
373 * Free a busy filesystem.
374 */
375 void
376 vfs_unbusy(mp, td)
377 struct mount *mp;
378 struct thread *td;
379 {
380
381 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
382 }
383
384 /*
385 * Lookup a mount point by filesystem identifier.
386 */
387 struct mount *
388 vfs_getvfs(fsid)
389 fsid_t *fsid;
390 {
391 struct mount *mp;
392
393 mtx_lock(&mountlist_mtx);
394 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
395 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
396 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
397 mtx_unlock(&mountlist_mtx);
398 return (mp);
399 }
400 }
401 mtx_unlock(&mountlist_mtx);
402 return ((struct mount *) 0);
403 }
404
405 /*
406 * Check if a user can access priveledged mount options.
407 */
408 int
409 vfs_suser(struct mount *mp, struct thread *td)
410 {
411 int error;
412
413 if ((mp->mnt_flag & MNT_USER) == 0 ||
414 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
415 if ((error = suser(td)) != 0)
416 return (error);
417 }
418 return (0);
419 }
420
421 /*
422 * Get a new unique fsid. Try to make its val[0] unique, since this value
423 * will be used to create fake device numbers for stat(). Also try (but
424 * not so hard) make its val[0] unique mod 2^16, since some emulators only
425 * support 16-bit device numbers. We end up with unique val[0]'s for the
426 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
427 *
428 * Keep in mind that several mounts may be running in parallel. Starting
429 * the search one past where the previous search terminated is both a
430 * micro-optimization and a defense against returning the same fsid to
431 * different mounts.
432 */
433 void
434 vfs_getnewfsid(mp)
435 struct mount *mp;
436 {
437 static u_int16_t mntid_base;
438 fsid_t tfsid;
439 int mtype;
440
441 mtx_lock(&mntid_mtx);
442 mtype = mp->mnt_vfc->vfc_typenum;
443 tfsid.val[1] = mtype;
444 mtype = (mtype & 0xFF) << 24;
445 for (;;) {
446 tfsid.val[0] = makedev(255,
447 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
448 mntid_base++;
449 if (vfs_getvfs(&tfsid) == NULL)
450 break;
451 }
452 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
453 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
454 mtx_unlock(&mntid_mtx);
455 }
456
457 /*
458 * Knob to control the precision of file timestamps:
459 *
460 * 0 = seconds only; nanoseconds zeroed.
461 * 1 = seconds and nanoseconds, accurate within 1/HZ.
462 * 2 = seconds and nanoseconds, truncated to microseconds.
463 * >=3 = seconds and nanoseconds, maximum precision.
464 */
465 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
466
467 static int timestamp_precision = TSP_SEC;
468 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
469 ×tamp_precision, 0, "");
470
471 /*
472 * Get a current timestamp.
473 */
474 void
475 vfs_timestamp(tsp)
476 struct timespec *tsp;
477 {
478 struct timeval tv;
479
480 switch (timestamp_precision) {
481 case TSP_SEC:
482 tsp->tv_sec = time_second;
483 tsp->tv_nsec = 0;
484 break;
485 case TSP_HZ:
486 getnanotime(tsp);
487 break;
488 case TSP_USEC:
489 microtime(&tv);
490 TIMEVAL_TO_TIMESPEC(&tv, tsp);
491 break;
492 case TSP_NSEC:
493 default:
494 nanotime(tsp);
495 break;
496 }
497 }
498
499 /*
500 * Set vnode attributes to VNOVAL
501 */
502 void
503 vattr_null(vap)
504 struct vattr *vap;
505 {
506
507 vap->va_type = VNON;
508 vap->va_size = VNOVAL;
509 vap->va_bytes = VNOVAL;
510 vap->va_mode = VNOVAL;
511 vap->va_nlink = VNOVAL;
512 vap->va_uid = VNOVAL;
513 vap->va_gid = VNOVAL;
514 vap->va_fsid = VNOVAL;
515 vap->va_fileid = VNOVAL;
516 vap->va_blocksize = VNOVAL;
517 vap->va_rdev = VNOVAL;
518 vap->va_atime.tv_sec = VNOVAL;
519 vap->va_atime.tv_nsec = VNOVAL;
520 vap->va_mtime.tv_sec = VNOVAL;
521 vap->va_mtime.tv_nsec = VNOVAL;
522 vap->va_ctime.tv_sec = VNOVAL;
523 vap->va_ctime.tv_nsec = VNOVAL;
524 vap->va_birthtime.tv_sec = VNOVAL;
525 vap->va_birthtime.tv_nsec = VNOVAL;
526 vap->va_flags = VNOVAL;
527 vap->va_gen = VNOVAL;
528 vap->va_vaflags = 0;
529 }
530
531 /*
532 * This routine is called when we have too many vnodes. It attempts
533 * to free <count> vnodes and will potentially free vnodes that still
534 * have VM backing store (VM backing store is typically the cause
535 * of a vnode blowout so we want to do this). Therefore, this operation
536 * is not considered cheap.
537 *
538 * A number of conditions may prevent a vnode from being reclaimed.
539 * the buffer cache may have references on the vnode, a directory
540 * vnode may still have references due to the namei cache representing
541 * underlying files, or the vnode may be in active use. It is not
542 * desireable to reuse such vnodes. These conditions may cause the
543 * number of vnodes to reach some minimum value regardless of what
544 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
545 */
546 static int
547 vlrureclaim(struct mount *mp)
548 {
549 struct thread *td;
550 struct vnode *vp;
551 int done;
552 int trigger;
553 int usevnodes;
554 int count;
555
556 /*
557 * Calculate the trigger point, don't allow user
558 * screwups to blow us up. This prevents us from
559 * recycling vnodes with lots of resident pages. We
560 * aren't trying to free memory, we are trying to
561 * free vnodes.
562 */
563 usevnodes = desiredvnodes;
564 if (usevnodes <= 0)
565 usevnodes = 1;
566 trigger = cnt.v_page_count * 2 / usevnodes;
567 done = 0;
568 td = curthread;
569 vn_start_write(NULL, &mp, V_WAIT);
570 MNT_ILOCK(mp);
571 count = mp->mnt_nvnodelistsize / 10 + 1;
572 while (count != 0) {
573 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
574 while (vp != NULL && vp->v_type == VMARKER)
575 vp = TAILQ_NEXT(vp, v_nmntvnodes);
576 if (vp == NULL)
577 break;
578 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
579 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
580 --count;
581 if (!VI_TRYLOCK(vp))
582 goto next_iter;
583 /*
584 * If it's been deconstructed already, it's still
585 * referenced, or it exceeds the trigger, skip it.
586 */
587 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
588 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
589 vp->v_object->resident_page_count > trigger)) {
590 VI_UNLOCK(vp);
591 goto next_iter;
592 }
593 MNT_IUNLOCK(mp);
594 vholdl(vp);
595 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT, td)) {
596 vdrop(vp);
597 goto next_iter_mntunlocked;
598 }
599 VI_LOCK(vp);
600 /*
601 * v_usecount may have been bumped after VOP_LOCK() dropped
602 * the vnode interlock and before it was locked again.
603 *
604 * It is not necessary to recheck VI_DOOMED because it can
605 * only be set by another thread that holds both the vnode
606 * lock and vnode interlock. If another thread has the
607 * vnode lock before we get to VOP_LOCK() and obtains the
608 * vnode interlock after VOP_LOCK() drops the vnode
609 * interlock, the other thread will be unable to drop the
610 * vnode lock before our VOP_LOCK() call fails.
611 */
612 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
613 (vp->v_object != NULL &&
614 vp->v_object->resident_page_count > trigger)) {
615 VOP_UNLOCK(vp, LK_INTERLOCK, td);
616 goto next_iter_mntunlocked;
617 }
618 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
619 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
620 vgonel(vp);
621 VOP_UNLOCK(vp, 0, td);
622 vdropl(vp);
623 done++;
624 next_iter_mntunlocked:
625 if ((count % 256) != 0)
626 goto relock_mnt;
627 goto yield;
628 next_iter:
629 if ((count % 256) != 0)
630 continue;
631 MNT_IUNLOCK(mp);
632 yield:
633 uio_yield();
634 relock_mnt:
635 MNT_ILOCK(mp);
636 }
637 MNT_IUNLOCK(mp);
638 vn_finished_write(mp);
639 return done;
640 }
641
642 /*
643 * Attempt to keep the free list at wantfreevnodes length.
644 */
645 static void
646 vnlru_free(int count)
647 {
648 struct vnode *vp;
649 int vfslocked;
650
651 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
652 for (; count > 0; count--) {
653 vp = TAILQ_FIRST(&vnode_free_list);
654 /*
655 * The list can be modified while the free_list_mtx
656 * has been dropped and vp could be NULL here.
657 */
658 if (!vp)
659 break;
660 VNASSERT(vp->v_op != NULL, vp,
661 ("vnlru_free: vnode already reclaimed."));
662 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
663 /*
664 * Don't recycle if we can't get the interlock.
665 */
666 if (!VI_TRYLOCK(vp)) {
667 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
668 continue;
669 }
670 VNASSERT(VCANRECYCLE(vp), vp,
671 ("vp inconsistent on freelist"));
672 freevnodes--;
673 vp->v_iflag &= ~VI_FREE;
674 vholdl(vp);
675 mtx_unlock(&vnode_free_list_mtx);
676 VI_UNLOCK(vp);
677 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
678 vtryrecycle(vp);
679 VFS_UNLOCK_GIANT(vfslocked);
680 /*
681 * If the recycled succeeded this vdrop will actually free
682 * the vnode. If not it will simply place it back on
683 * the free list.
684 */
685 vdrop(vp);
686 mtx_lock(&vnode_free_list_mtx);
687 }
688 }
689 /*
690 * Attempt to recycle vnodes in a context that is always safe to block.
691 * Calling vlrurecycle() from the bowels of filesystem code has some
692 * interesting deadlock problems.
693 */
694 static struct proc *vnlruproc;
695 static int vnlruproc_sig;
696
697 static void
698 vnlru_proc(void)
699 {
700 struct mount *mp, *nmp;
701 int done;
702 struct proc *p = vnlruproc;
703 struct thread *td = FIRST_THREAD_IN_PROC(p);
704
705 mtx_lock(&Giant);
706
707 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
708 SHUTDOWN_PRI_FIRST);
709
710 for (;;) {
711 kthread_suspend_check(p);
712 mtx_lock(&vnode_free_list_mtx);
713 if (freevnodes > wantfreevnodes)
714 vnlru_free(freevnodes - wantfreevnodes);
715 if (numvnodes <= desiredvnodes * 9 / 10) {
716 vnlruproc_sig = 0;
717 wakeup(&vnlruproc_sig);
718 msleep(vnlruproc, &vnode_free_list_mtx,
719 PVFS|PDROP, "vlruwt", hz);
720 continue;
721 }
722 mtx_unlock(&vnode_free_list_mtx);
723 done = 0;
724 mtx_lock(&mountlist_mtx);
725 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
726 int vfsunlocked;
727 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
728 nmp = TAILQ_NEXT(mp, mnt_list);
729 continue;
730 }
731 if (!VFS_NEEDSGIANT(mp)) {
732 mtx_unlock(&Giant);
733 vfsunlocked = 1;
734 } else
735 vfsunlocked = 0;
736 done += vlrureclaim(mp);
737 if (vfsunlocked)
738 mtx_lock(&Giant);
739 mtx_lock(&mountlist_mtx);
740 nmp = TAILQ_NEXT(mp, mnt_list);
741 vfs_unbusy(mp, td);
742 }
743 mtx_unlock(&mountlist_mtx);
744 if (done == 0) {
745 #if 0
746 /* These messages are temporary debugging aids */
747 if (vnlru_nowhere < 5)
748 printf("vnlru process getting nowhere..\n");
749 else if (vnlru_nowhere == 5)
750 printf("vnlru process messages stopped.\n");
751 #endif
752 vnlru_nowhere++;
753 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
754 } else
755 uio_yield();
756 }
757 }
758
759 static struct kproc_desc vnlru_kp = {
760 "vnlru",
761 vnlru_proc,
762 &vnlruproc
763 };
764 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
765
766 /*
767 * Routines having to do with the management of the vnode table.
768 */
769
770 static void
771 vdestroy(struct vnode *vp)
772 {
773 struct bufobj *bo;
774
775 CTR1(KTR_VFS, "vdestroy vp %p", vp);
776 mtx_lock(&vnode_free_list_mtx);
777 numvnodes--;
778 mtx_unlock(&vnode_free_list_mtx);
779 bo = &vp->v_bufobj;
780 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
781 ("cleaned vnode still on the free list."));
782 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
783 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
784 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
785 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
786 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
787 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
788 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
789 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
790 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
791 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
792 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
793 #ifdef MAC
794 mac_destroy_vnode(vp);
795 #endif
796 if (vp->v_pollinfo != NULL) {
797 knlist_destroy(&vp->v_pollinfo->vpi_selinfo.si_note);
798 mtx_destroy(&vp->v_pollinfo->vpi_lock);
799 uma_zfree(vnodepoll_zone, vp->v_pollinfo);
800 }
801 #ifdef INVARIANTS
802 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
803 vp->v_op = NULL;
804 #endif
805 lockdestroy(vp->v_vnlock);
806 mtx_destroy(&vp->v_interlock);
807 uma_zfree(vnode_zone, vp);
808 }
809
810 /*
811 * Try to recycle a freed vnode. We abort if anyone picks up a reference
812 * before we actually vgone(). This function must be called with the vnode
813 * held to prevent the vnode from being returned to the free list midway
814 * through vgone().
815 */
816 static int
817 vtryrecycle(struct vnode *vp)
818 {
819 struct thread *td = curthread;
820 struct mount *vnmp;
821
822 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
823 VNASSERT(vp->v_holdcnt, vp,
824 ("vtryrecycle: Recycling vp %p without a reference.", vp));
825 /*
826 * This vnode may found and locked via some other list, if so we
827 * can't recycle it yet.
828 */
829 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
830 return (EWOULDBLOCK);
831 /*
832 * Don't recycle if its filesystem is being suspended.
833 */
834 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
835 VOP_UNLOCK(vp, 0, td);
836 return (EBUSY);
837 }
838 /*
839 * If we got this far, we need to acquire the interlock and see if
840 * anyone picked up this vnode from another list. If not, we will
841 * mark it with DOOMED via vgonel() so that anyone who does find it
842 * will skip over it.
843 */
844 VI_LOCK(vp);
845 if (vp->v_usecount) {
846 VOP_UNLOCK(vp, LK_INTERLOCK, td);
847 vn_finished_write(vnmp);
848 return (EBUSY);
849 }
850 if ((vp->v_iflag & VI_DOOMED) == 0)
851 vgonel(vp);
852 VOP_UNLOCK(vp, LK_INTERLOCK, td);
853 vn_finished_write(vnmp);
854 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
855 return (0);
856 }
857
858 /*
859 * Return the next vnode from the free list.
860 */
861 int
862 getnewvnode(tag, mp, vops, vpp)
863 const char *tag;
864 struct mount *mp;
865 struct vop_vector *vops;
866 struct vnode **vpp;
867 {
868 struct vnode *vp = NULL;
869 struct bufobj *bo;
870
871 mtx_lock(&vnode_free_list_mtx);
872 /*
873 * Lend our context to reclaim vnodes if they've exceeded the max.
874 */
875 if (freevnodes > wantfreevnodes)
876 vnlru_free(1);
877 /*
878 * Wait for available vnodes.
879 */
880 if (numvnodes > desiredvnodes) {
881 if (vnlruproc_sig == 0) {
882 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
883 wakeup(vnlruproc);
884 }
885 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
886 "vlruwk", hz);
887 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
888 if (numvnodes > desiredvnodes) {
889 mtx_unlock(&vnode_free_list_mtx);
890 return (ENFILE);
891 }
892 #endif
893 }
894 numvnodes++;
895 mtx_unlock(&vnode_free_list_mtx);
896 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
897 /*
898 * Setup locks.
899 */
900 vp->v_vnlock = &vp->v_lock;
901 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
902 /*
903 * By default, don't allow shared locks unless filesystems
904 * opt-in.
905 */
906 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
907 /*
908 * Initialize bufobj.
909 */
910 bo = &vp->v_bufobj;
911 bo->__bo_vnode = vp;
912 bo->bo_mtx = &vp->v_interlock;
913 bo->bo_ops = &buf_ops_bio;
914 bo->bo_private = vp;
915 TAILQ_INIT(&bo->bo_clean.bv_hd);
916 TAILQ_INIT(&bo->bo_dirty.bv_hd);
917 /*
918 * Initialize namecache.
919 */
920 LIST_INIT(&vp->v_cache_src);
921 TAILQ_INIT(&vp->v_cache_dst);
922 /*
923 * Finalize various vnode identity bits.
924 */
925 vp->v_type = VNON;
926 vp->v_tag = tag;
927 vp->v_op = vops;
928 v_incr_usecount(vp);
929 vp->v_data = 0;
930 #ifdef MAC
931 mac_init_vnode(vp);
932 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
933 mac_associate_vnode_singlelabel(mp, vp);
934 else if (mp == NULL)
935 printf("NULL mp in getnewvnode()\n");
936 #endif
937 if (mp != NULL) {
938 insmntque(vp, mp);
939 bo->bo_bsize = mp->mnt_stat.f_iosize;
940 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
941 vp->v_vflag |= VV_NOKNOTE;
942 }
943
944 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
945 *vpp = vp;
946 return (0);
947 }
948
949 /*
950 * Delete from old mount point vnode list, if on one.
951 */
952 static void
953 delmntque(struct vnode *vp)
954 {
955 struct mount *mp;
956
957 mp = vp->v_mount;
958 if (mp == NULL)
959 return;
960 MNT_ILOCK(mp);
961 vp->v_mount = NULL;
962 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
963 ("bad mount point vnode list size"));
964 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
965 mp->mnt_nvnodelistsize--;
966 MNT_REL(mp);
967 MNT_IUNLOCK(mp);
968 }
969
970 /*
971 * Insert into list of vnodes for the new mount point, if available.
972 */
973 static void
974 insmntque(struct vnode *vp, struct mount *mp)
975 {
976
977 vp->v_mount = mp;
978 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
979 MNT_ILOCK(mp);
980 MNT_REF(mp);
981 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
982 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
983 ("neg mount point vnode list size"));
984 mp->mnt_nvnodelistsize++;
985 MNT_IUNLOCK(mp);
986 }
987
988 /*
989 * Flush out and invalidate all buffers associated with a bufobj
990 * Called with the underlying object locked.
991 */
992 int
993 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag, int slptimeo)
994 {
995 int error;
996
997 BO_LOCK(bo);
998 if (flags & V_SAVE) {
999 error = bufobj_wwait(bo, slpflag, slptimeo);
1000 if (error) {
1001 BO_UNLOCK(bo);
1002 return (error);
1003 }
1004 if (bo->bo_dirty.bv_cnt > 0) {
1005 BO_UNLOCK(bo);
1006 if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1007 return (error);
1008 /*
1009 * XXX We could save a lock/unlock if this was only
1010 * enabled under INVARIANTS
1011 */
1012 BO_LOCK(bo);
1013 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1014 panic("vinvalbuf: dirty bufs");
1015 }
1016 }
1017 /*
1018 * If you alter this loop please notice that interlock is dropped and
1019 * reacquired in flushbuflist. Special care is needed to ensure that
1020 * no race conditions occur from this.
1021 */
1022 do {
1023 error = flushbuflist(&bo->bo_clean,
1024 flags, bo, slpflag, slptimeo);
1025 if (error == 0)
1026 error = flushbuflist(&bo->bo_dirty,
1027 flags, bo, slpflag, slptimeo);
1028 if (error != 0 && error != EAGAIN) {
1029 BO_UNLOCK(bo);
1030 return (error);
1031 }
1032 } while (error != 0);
1033
1034 /*
1035 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1036 * have write I/O in-progress but if there is a VM object then the
1037 * VM object can also have read-I/O in-progress.
1038 */
1039 do {
1040 bufobj_wwait(bo, 0, 0);
1041 BO_UNLOCK(bo);
1042 if (bo->bo_object != NULL) {
1043 VM_OBJECT_LOCK(bo->bo_object);
1044 vm_object_pip_wait(bo->bo_object, "bovlbx");
1045 VM_OBJECT_UNLOCK(bo->bo_object);
1046 }
1047 BO_LOCK(bo);
1048 } while (bo->bo_numoutput > 0);
1049 BO_UNLOCK(bo);
1050
1051 /*
1052 * Destroy the copy in the VM cache, too.
1053 */
1054 if (bo->bo_object != NULL) {
1055 VM_OBJECT_LOCK(bo->bo_object);
1056 vm_object_page_remove(bo->bo_object, 0, 0,
1057 (flags & V_SAVE) ? TRUE : FALSE);
1058 VM_OBJECT_UNLOCK(bo->bo_object);
1059 }
1060
1061 #ifdef INVARIANTS
1062 BO_LOCK(bo);
1063 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1064 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1065 panic("vinvalbuf: flush failed");
1066 BO_UNLOCK(bo);
1067 #endif
1068 return (0);
1069 }
1070
1071 /*
1072 * Flush out and invalidate all buffers associated with a vnode.
1073 * Called with the underlying object locked.
1074 */
1075 int
1076 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag, int slptimeo)
1077 {
1078
1079 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1080 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1081 return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1082 }
1083
1084 /*
1085 * Flush out buffers on the specified list.
1086 *
1087 */
1088 static int
1089 flushbuflist(bufv, flags, bo, slpflag, slptimeo)
1090 struct bufv *bufv;
1091 int flags;
1092 struct bufobj *bo;
1093 int slpflag, slptimeo;
1094 {
1095 struct buf *bp, *nbp;
1096 int retval, error;
1097 daddr_t lblkno;
1098 b_xflags_t xflags;
1099
1100 ASSERT_BO_LOCKED(bo);
1101
1102 retval = 0;
1103 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1104 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1105 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1106 continue;
1107 }
1108 lblkno = 0;
1109 xflags = 0;
1110 if (nbp != NULL) {
1111 lblkno = nbp->b_lblkno;
1112 xflags = nbp->b_xflags &
1113 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1114 }
1115 retval = EAGAIN;
1116 error = BUF_TIMELOCK(bp,
1117 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1118 "flushbuf", slpflag, slptimeo);
1119 if (error) {
1120 BO_LOCK(bo);
1121 return (error != ENOLCK ? error : EAGAIN);
1122 }
1123 KASSERT(bp->b_bufobj == bo,
1124 ("bp %p wrong b_bufobj %p should be %p",
1125 bp, bp->b_bufobj, bo));
1126 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1127 BUF_UNLOCK(bp);
1128 BO_LOCK(bo);
1129 return (EAGAIN);
1130 }
1131 /*
1132 * XXX Since there are no node locks for NFS, I
1133 * believe there is a slight chance that a delayed
1134 * write will occur while sleeping just above, so
1135 * check for it.
1136 */
1137 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1138 (flags & V_SAVE)) {
1139 bremfree(bp);
1140 bp->b_flags |= B_ASYNC;
1141 bwrite(bp);
1142 BO_LOCK(bo);
1143 return (EAGAIN); /* XXX: why not loop ? */
1144 }
1145 bremfree(bp);
1146 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1147 bp->b_flags &= ~B_ASYNC;
1148 brelse(bp);
1149 BO_LOCK(bo);
1150 if (nbp != NULL &&
1151 (nbp->b_bufobj != bo ||
1152 nbp->b_lblkno != lblkno ||
1153 (nbp->b_xflags &
1154 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1155 break; /* nbp invalid */
1156 }
1157 return (retval);
1158 }
1159
1160 /*
1161 * Truncate a file's buffer and pages to a specified length. This
1162 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1163 * sync activity.
1164 */
1165 int
1166 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td, off_t length, int blksize)
1167 {
1168 struct buf *bp, *nbp;
1169 int anyfreed;
1170 int trunclbn;
1171 struct bufobj *bo;
1172
1173 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1174 /*
1175 * Round up to the *next* lbn.
1176 */
1177 trunclbn = (length + blksize - 1) / blksize;
1178
1179 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1180 restart:
1181 VI_LOCK(vp);
1182 bo = &vp->v_bufobj;
1183 anyfreed = 1;
1184 for (;anyfreed;) {
1185 anyfreed = 0;
1186 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1187 if (bp->b_lblkno < trunclbn)
1188 continue;
1189 if (BUF_LOCK(bp,
1190 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1191 VI_MTX(vp)) == ENOLCK)
1192 goto restart;
1193
1194 bremfree(bp);
1195 bp->b_flags |= (B_INVAL | B_RELBUF);
1196 bp->b_flags &= ~B_ASYNC;
1197 brelse(bp);
1198 anyfreed = 1;
1199
1200 if (nbp != NULL &&
1201 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1202 (nbp->b_vp != vp) ||
1203 (nbp->b_flags & B_DELWRI))) {
1204 goto restart;
1205 }
1206 VI_LOCK(vp);
1207 }
1208
1209 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1210 if (bp->b_lblkno < trunclbn)
1211 continue;
1212 if (BUF_LOCK(bp,
1213 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1214 VI_MTX(vp)) == ENOLCK)
1215 goto restart;
1216 bremfree(bp);
1217 bp->b_flags |= (B_INVAL | B_RELBUF);
1218 bp->b_flags &= ~B_ASYNC;
1219 brelse(bp);
1220 anyfreed = 1;
1221 if (nbp != NULL &&
1222 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1223 (nbp->b_vp != vp) ||
1224 (nbp->b_flags & B_DELWRI) == 0)) {
1225 goto restart;
1226 }
1227 VI_LOCK(vp);
1228 }
1229 }
1230
1231 if (length > 0) {
1232 restartsync:
1233 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1234 if (bp->b_lblkno > 0)
1235 continue;
1236 /*
1237 * Since we hold the vnode lock this should only
1238 * fail if we're racing with the buf daemon.
1239 */
1240 if (BUF_LOCK(bp,
1241 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1242 VI_MTX(vp)) == ENOLCK) {
1243 goto restart;
1244 }
1245 VNASSERT((bp->b_flags & B_DELWRI), vp,
1246 ("buf(%p) on dirty queue without DELWRI", bp));
1247
1248 bremfree(bp);
1249 bawrite(bp);
1250 VI_LOCK(vp);
1251 goto restartsync;
1252 }
1253 }
1254
1255 bufobj_wwait(bo, 0, 0);
1256 VI_UNLOCK(vp);
1257 vnode_pager_setsize(vp, length);
1258
1259 return (0);
1260 }
1261
1262 /*
1263 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1264 * a vnode.
1265 *
1266 * NOTE: We have to deal with the special case of a background bitmap
1267 * buffer, a situation where two buffers will have the same logical
1268 * block offset. We want (1) only the foreground buffer to be accessed
1269 * in a lookup and (2) must differentiate between the foreground and
1270 * background buffer in the splay tree algorithm because the splay
1271 * tree cannot normally handle multiple entities with the same 'index'.
1272 * We accomplish this by adding differentiating flags to the splay tree's
1273 * numerical domain.
1274 */
1275 static
1276 struct buf *
1277 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1278 {
1279 struct buf dummy;
1280 struct buf *lefttreemax, *righttreemin, *y;
1281
1282 if (root == NULL)
1283 return (NULL);
1284 lefttreemax = righttreemin = &dummy;
1285 for (;;) {
1286 if (lblkno < root->b_lblkno ||
1287 (lblkno == root->b_lblkno &&
1288 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1289 if ((y = root->b_left) == NULL)
1290 break;
1291 if (lblkno < y->b_lblkno) {
1292 /* Rotate right. */
1293 root->b_left = y->b_right;
1294 y->b_right = root;
1295 root = y;
1296 if ((y = root->b_left) == NULL)
1297 break;
1298 }
1299 /* Link into the new root's right tree. */
1300 righttreemin->b_left = root;
1301 righttreemin = root;
1302 } else if (lblkno > root->b_lblkno ||
1303 (lblkno == root->b_lblkno &&
1304 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1305 if ((y = root->b_right) == NULL)
1306 break;
1307 if (lblkno > y->b_lblkno) {
1308 /* Rotate left. */
1309 root->b_right = y->b_left;
1310 y->b_left = root;
1311 root = y;
1312 if ((y = root->b_right) == NULL)
1313 break;
1314 }
1315 /* Link into the new root's left tree. */
1316 lefttreemax->b_right = root;
1317 lefttreemax = root;
1318 } else {
1319 break;
1320 }
1321 root = y;
1322 }
1323 /* Assemble the new root. */
1324 lefttreemax->b_right = root->b_left;
1325 righttreemin->b_left = root->b_right;
1326 root->b_left = dummy.b_right;
1327 root->b_right = dummy.b_left;
1328 return (root);
1329 }
1330
1331 static void
1332 buf_vlist_remove(struct buf *bp)
1333 {
1334 struct buf *root;
1335 struct bufv *bv;
1336
1337 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1338 ASSERT_BO_LOCKED(bp->b_bufobj);
1339 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1340 (BX_VNDIRTY|BX_VNCLEAN),
1341 ("buf_vlist_remove: Buf %p is on two lists", bp));
1342 if (bp->b_xflags & BX_VNDIRTY)
1343 bv = &bp->b_bufobj->bo_dirty;
1344 else
1345 bv = &bp->b_bufobj->bo_clean;
1346 if (bp != bv->bv_root) {
1347 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1348 KASSERT(root == bp, ("splay lookup failed in remove"));
1349 }
1350 if (bp->b_left == NULL) {
1351 root = bp->b_right;
1352 } else {
1353 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1354 root->b_right = bp->b_right;
1355 }
1356 bv->bv_root = root;
1357 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1358 bv->bv_cnt--;
1359 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1360 }
1361
1362 /*
1363 * Add the buffer to the sorted clean or dirty block list using a
1364 * splay tree algorithm.
1365 *
1366 * NOTE: xflags is passed as a constant, optimizing this inline function!
1367 */
1368 static void
1369 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1370 {
1371 struct buf *root;
1372 struct bufv *bv;
1373
1374 ASSERT_BO_LOCKED(bo);
1375 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1376 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1377 bp->b_xflags |= xflags;
1378 if (xflags & BX_VNDIRTY)
1379 bv = &bo->bo_dirty;
1380 else
1381 bv = &bo->bo_clean;
1382
1383 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1384 if (root == NULL) {
1385 bp->b_left = NULL;
1386 bp->b_right = NULL;
1387 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1388 } else if (bp->b_lblkno < root->b_lblkno ||
1389 (bp->b_lblkno == root->b_lblkno &&
1390 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1391 bp->b_left = root->b_left;
1392 bp->b_right = root;
1393 root->b_left = NULL;
1394 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1395 } else {
1396 bp->b_right = root->b_right;
1397 bp->b_left = root;
1398 root->b_right = NULL;
1399 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1400 }
1401 bv->bv_cnt++;
1402 bv->bv_root = bp;
1403 }
1404
1405 /*
1406 * Lookup a buffer using the splay tree. Note that we specifically avoid
1407 * shadow buffers used in background bitmap writes.
1408 *
1409 * This code isn't quite efficient as it could be because we are maintaining
1410 * two sorted lists and do not know which list the block resides in.
1411 *
1412 * During a "make buildworld" the desired buffer is found at one of
1413 * the roots more than 60% of the time. Thus, checking both roots
1414 * before performing either splay eliminates unnecessary splays on the
1415 * first tree splayed.
1416 */
1417 struct buf *
1418 gbincore(struct bufobj *bo, daddr_t lblkno)
1419 {
1420 struct buf *bp;
1421
1422 ASSERT_BO_LOCKED(bo);
1423 if ((bp = bo->bo_clean.bv_root) != NULL &&
1424 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1425 return (bp);
1426 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1427 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1428 return (bp);
1429 if ((bp = bo->bo_clean.bv_root) != NULL) {
1430 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1431 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1432 return (bp);
1433 }
1434 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1435 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1436 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1437 return (bp);
1438 }
1439 return (NULL);
1440 }
1441
1442 /*
1443 * Associate a buffer with a vnode.
1444 */
1445 void
1446 bgetvp(struct vnode *vp, struct buf *bp)
1447 {
1448
1449 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1450
1451 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1452 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1453 ("bgetvp: bp already attached! %p", bp));
1454
1455 ASSERT_VI_LOCKED(vp, "bgetvp");
1456 vholdl(vp);
1457 bp->b_vp = vp;
1458 bp->b_bufobj = &vp->v_bufobj;
1459 /*
1460 * Insert onto list for new vnode.
1461 */
1462 buf_vlist_add(bp, &vp->v_bufobj, BX_VNCLEAN);
1463 }
1464
1465 /*
1466 * Disassociate a buffer from a vnode.
1467 */
1468 void
1469 brelvp(struct buf *bp)
1470 {
1471 struct bufobj *bo;
1472 struct vnode *vp;
1473
1474 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1475 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1476
1477 /*
1478 * Delete from old vnode list, if on one.
1479 */
1480 vp = bp->b_vp; /* XXX */
1481 bo = bp->b_bufobj;
1482 BO_LOCK(bo);
1483 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1484 buf_vlist_remove(bp);
1485 else
1486 panic("brelvp: Buffer %p not on queue.", bp);
1487 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1488 bo->bo_flag &= ~BO_ONWORKLST;
1489 mtx_lock(&sync_mtx);
1490 LIST_REMOVE(bo, bo_synclist);
1491 syncer_worklist_len--;
1492 mtx_unlock(&sync_mtx);
1493 }
1494 bp->b_vp = NULL;
1495 bp->b_bufobj = NULL;
1496 vdropl(vp);
1497 }
1498
1499 /*
1500 * Add an item to the syncer work queue.
1501 */
1502 static void
1503 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1504 {
1505 int slot;
1506
1507 ASSERT_BO_LOCKED(bo);
1508
1509 mtx_lock(&sync_mtx);
1510 if (bo->bo_flag & BO_ONWORKLST)
1511 LIST_REMOVE(bo, bo_synclist);
1512 else {
1513 bo->bo_flag |= BO_ONWORKLST;
1514 syncer_worklist_len++;
1515 }
1516
1517 if (delay > syncer_maxdelay - 2)
1518 delay = syncer_maxdelay - 2;
1519 slot = (syncer_delayno + delay) & syncer_mask;
1520
1521 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1522 mtx_unlock(&sync_mtx);
1523 }
1524
1525 static int
1526 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1527 {
1528 int error, len;
1529
1530 mtx_lock(&sync_mtx);
1531 len = syncer_worklist_len - sync_vnode_count;
1532 mtx_unlock(&sync_mtx);
1533 error = SYSCTL_OUT(req, &len, sizeof(len));
1534 return (error);
1535 }
1536
1537 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1538 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1539
1540 static struct proc *updateproc;
1541 static void sched_sync(void);
1542 static struct kproc_desc up_kp = {
1543 "syncer",
1544 sched_sync,
1545 &updateproc
1546 };
1547 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1548
1549 static int
1550 sync_vnode(struct bufobj *bo, struct thread *td)
1551 {
1552 struct vnode *vp;
1553 struct mount *mp;
1554
1555 vp = bo->__bo_vnode; /* XXX */
1556 if (VOP_ISLOCKED(vp, NULL) != 0)
1557 return (1);
1558 if (VI_TRYLOCK(vp) == 0)
1559 return (1);
1560 /*
1561 * We use vhold in case the vnode does not
1562 * successfully sync. vhold prevents the vnode from
1563 * going away when we unlock the sync_mtx so that
1564 * we can acquire the vnode interlock.
1565 */
1566 vholdl(vp);
1567 mtx_unlock(&sync_mtx);
1568 VI_UNLOCK(vp);
1569 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1570 vdrop(vp);
1571 mtx_lock(&sync_mtx);
1572 return (1);
1573 }
1574 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1575 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1576 VOP_UNLOCK(vp, 0, td);
1577 vn_finished_write(mp);
1578 VI_LOCK(vp);
1579 if ((bo->bo_flag & BO_ONWORKLST) != 0) {
1580 /*
1581 * Put us back on the worklist. The worklist
1582 * routine will remove us from our current
1583 * position and then add us back in at a later
1584 * position.
1585 */
1586 vn_syncer_add_to_worklist(bo, syncdelay);
1587 }
1588 vdropl(vp);
1589 mtx_lock(&sync_mtx);
1590 return (0);
1591 }
1592
1593 /*
1594 * System filesystem synchronizer daemon.
1595 */
1596 static void
1597 sched_sync(void)
1598 {
1599 struct synclist *next;
1600 struct synclist *slp;
1601 struct bufobj *bo;
1602 long starttime;
1603 struct thread *td = FIRST_THREAD_IN_PROC(updateproc);
1604 static int dummychan;
1605 int last_work_seen;
1606 int net_worklist_len;
1607 int syncer_final_iter;
1608 int first_printf;
1609 int error;
1610
1611 mtx_lock(&Giant);
1612 last_work_seen = 0;
1613 syncer_final_iter = 0;
1614 first_printf = 1;
1615 syncer_state = SYNCER_RUNNING;
1616 starttime = time_second;
1617 td->td_pflags |= TDP_NORUNNINGBUF;
1618
1619 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1620 SHUTDOWN_PRI_LAST);
1621
1622 for (;;) {
1623 mtx_lock(&sync_mtx);
1624 if (syncer_state == SYNCER_FINAL_DELAY &&
1625 syncer_final_iter == 0) {
1626 mtx_unlock(&sync_mtx);
1627 kthread_suspend_check(td->td_proc);
1628 mtx_lock(&sync_mtx);
1629 }
1630 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1631 if (syncer_state != SYNCER_RUNNING &&
1632 starttime != time_second) {
1633 if (first_printf) {
1634 printf("\nSyncing disks, vnodes remaining...");
1635 first_printf = 0;
1636 }
1637 printf("%d ", net_worklist_len);
1638 }
1639 starttime = time_second;
1640
1641 /*
1642 * Push files whose dirty time has expired. Be careful
1643 * of interrupt race on slp queue.
1644 *
1645 * Skip over empty worklist slots when shutting down.
1646 */
1647 do {
1648 slp = &syncer_workitem_pending[syncer_delayno];
1649 syncer_delayno += 1;
1650 if (syncer_delayno == syncer_maxdelay)
1651 syncer_delayno = 0;
1652 next = &syncer_workitem_pending[syncer_delayno];
1653 /*
1654 * If the worklist has wrapped since the
1655 * it was emptied of all but syncer vnodes,
1656 * switch to the FINAL_DELAY state and run
1657 * for one more second.
1658 */
1659 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1660 net_worklist_len == 0 &&
1661 last_work_seen == syncer_delayno) {
1662 syncer_state = SYNCER_FINAL_DELAY;
1663 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1664 }
1665 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1666 syncer_worklist_len > 0);
1667
1668 /*
1669 * Keep track of the last time there was anything
1670 * on the worklist other than syncer vnodes.
1671 * Return to the SHUTTING_DOWN state if any
1672 * new work appears.
1673 */
1674 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1675 last_work_seen = syncer_delayno;
1676 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1677 syncer_state = SYNCER_SHUTTING_DOWN;
1678 while ((bo = LIST_FIRST(slp)) != NULL) {
1679 error = sync_vnode(bo, td);
1680 if (error == 1) {
1681 LIST_REMOVE(bo, bo_synclist);
1682 LIST_INSERT_HEAD(next, bo, bo_synclist);
1683 continue;
1684 }
1685 }
1686 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1687 syncer_final_iter--;
1688 mtx_unlock(&sync_mtx);
1689 /*
1690 * The variable rushjob allows the kernel to speed up the
1691 * processing of the filesystem syncer process. A rushjob
1692 * value of N tells the filesystem syncer to process the next
1693 * N seconds worth of work on its queue ASAP. Currently rushjob
1694 * is used by the soft update code to speed up the filesystem
1695 * syncer process when the incore state is getting so far
1696 * ahead of the disk that the kernel memory pool is being
1697 * threatened with exhaustion.
1698 */
1699 mtx_lock(&sync_mtx);
1700 if (rushjob > 0) {
1701 rushjob -= 1;
1702 mtx_unlock(&sync_mtx);
1703 continue;
1704 }
1705 mtx_unlock(&sync_mtx);
1706 /*
1707 * Just sleep for a short period if time between
1708 * iterations when shutting down to allow some I/O
1709 * to happen.
1710 *
1711 * If it has taken us less than a second to process the
1712 * current work, then wait. Otherwise start right over
1713 * again. We can still lose time if any single round
1714 * takes more than two seconds, but it does not really
1715 * matter as we are just trying to generally pace the
1716 * filesystem activity.
1717 */
1718 if (syncer_state != SYNCER_RUNNING)
1719 tsleep(&dummychan, PPAUSE, "syncfnl",
1720 hz / SYNCER_SHUTDOWN_SPEEDUP);
1721 else if (time_second == starttime)
1722 tsleep(&lbolt, PPAUSE, "syncer", 0);
1723 }
1724 }
1725
1726 /*
1727 * Request the syncer daemon to speed up its work.
1728 * We never push it to speed up more than half of its
1729 * normal turn time, otherwise it could take over the cpu.
1730 */
1731 int
1732 speedup_syncer()
1733 {
1734 struct thread *td;
1735 int ret = 0;
1736
1737 td = FIRST_THREAD_IN_PROC(updateproc);
1738 sleepq_remove(td, &lbolt);
1739 mtx_lock(&sync_mtx);
1740 if (rushjob < syncdelay / 2) {
1741 rushjob += 1;
1742 stat_rush_requests += 1;
1743 ret = 1;
1744 }
1745 mtx_unlock(&sync_mtx);
1746 return (ret);
1747 }
1748
1749 /*
1750 * Tell the syncer to speed up its work and run though its work
1751 * list several times, then tell it to shut down.
1752 */
1753 static void
1754 syncer_shutdown(void *arg, int howto)
1755 {
1756 struct thread *td;
1757
1758 if (howto & RB_NOSYNC)
1759 return;
1760 td = FIRST_THREAD_IN_PROC(updateproc);
1761 sleepq_remove(td, &lbolt);
1762 mtx_lock(&sync_mtx);
1763 syncer_state = SYNCER_SHUTTING_DOWN;
1764 rushjob = 0;
1765 mtx_unlock(&sync_mtx);
1766 kproc_shutdown(arg, howto);
1767 }
1768
1769 /*
1770 * Reassign a buffer from one vnode to another.
1771 * Used to assign file specific control information
1772 * (indirect blocks) to the vnode to which they belong.
1773 */
1774 void
1775 reassignbuf(struct buf *bp)
1776 {
1777 struct vnode *vp;
1778 struct bufobj *bo;
1779 int delay;
1780 #ifdef INVARIANTS
1781 struct bufv *bv;
1782 #endif
1783
1784 vp = bp->b_vp;
1785 bo = bp->b_bufobj;
1786 ++reassignbufcalls;
1787
1788 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1789 bp, bp->b_vp, bp->b_flags);
1790 /*
1791 * B_PAGING flagged buffers cannot be reassigned because their vp
1792 * is not fully linked in.
1793 */
1794 if (bp->b_flags & B_PAGING)
1795 panic("cannot reassign paging buffer");
1796
1797 /*
1798 * Delete from old vnode list, if on one.
1799 */
1800 VI_LOCK(vp);
1801 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1802 buf_vlist_remove(bp);
1803 else
1804 panic("reassignbuf: Buffer %p not on queue.", bp);
1805 /*
1806 * If dirty, put on list of dirty buffers; otherwise insert onto list
1807 * of clean buffers.
1808 */
1809 if (bp->b_flags & B_DELWRI) {
1810 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1811 switch (vp->v_type) {
1812 case VDIR:
1813 delay = dirdelay;
1814 break;
1815 case VCHR:
1816 delay = metadelay;
1817 break;
1818 default:
1819 delay = filedelay;
1820 }
1821 vn_syncer_add_to_worklist(bo, delay);
1822 }
1823 buf_vlist_add(bp, bo, BX_VNDIRTY);
1824 } else {
1825 buf_vlist_add(bp, bo, BX_VNCLEAN);
1826
1827 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1828 mtx_lock(&sync_mtx);
1829 LIST_REMOVE(bo, bo_synclist);
1830 syncer_worklist_len--;
1831 mtx_unlock(&sync_mtx);
1832 bo->bo_flag &= ~BO_ONWORKLST;
1833 }
1834 }
1835 #ifdef INVARIANTS
1836 bv = &bo->bo_clean;
1837 bp = TAILQ_FIRST(&bv->bv_hd);
1838 KASSERT(bp == NULL || bp->b_bufobj == bo,
1839 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1840 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1841 KASSERT(bp == NULL || bp->b_bufobj == bo,
1842 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1843 bv = &bo->bo_dirty;
1844 bp = TAILQ_FIRST(&bv->bv_hd);
1845 KASSERT(bp == NULL || bp->b_bufobj == bo,
1846 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1847 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1848 KASSERT(bp == NULL || bp->b_bufobj == bo,
1849 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1850 #endif
1851 VI_UNLOCK(vp);
1852 }
1853
1854 /*
1855 * Increment the use and hold counts on the vnode, taking care to reference
1856 * the driver's usecount if this is a chardev. The vholdl() will remove
1857 * the vnode from the free list if it is presently free. Requires the
1858 * vnode interlock and returns with it held.
1859 */
1860 static void
1861 v_incr_usecount(struct vnode *vp)
1862 {
1863
1864 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1865 vp, vp->v_holdcnt, vp->v_usecount);
1866 vp->v_usecount++;
1867 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1868 dev_lock();
1869 vp->v_rdev->si_usecount++;
1870 dev_unlock();
1871 }
1872 vholdl(vp);
1873 }
1874
1875 /*
1876 * Turn a holdcnt into a use+holdcnt such that only one call to
1877 * v_decr_usecount is needed.
1878 */
1879 static void
1880 v_upgrade_usecount(struct vnode *vp)
1881 {
1882
1883 CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
1884 vp, vp->v_holdcnt, vp->v_usecount);
1885 vp->v_usecount++;
1886 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1887 dev_lock();
1888 vp->v_rdev->si_usecount++;
1889 dev_unlock();
1890 }
1891 }
1892
1893 /*
1894 * Decrement the vnode use and hold count along with the driver's usecount
1895 * if this is a chardev. The vdropl() below releases the vnode interlock
1896 * as it may free the vnode.
1897 */
1898 static void
1899 v_decr_usecount(struct vnode *vp)
1900 {
1901
1902 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
1903 vp, vp->v_holdcnt, vp->v_usecount);
1904 ASSERT_VI_LOCKED(vp, __FUNCTION__);
1905 VNASSERT(vp->v_usecount > 0, vp,
1906 ("v_decr_usecount: negative usecount"));
1907 vp->v_usecount--;
1908 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1909 dev_lock();
1910 vp->v_rdev->si_usecount--;
1911 dev_unlock();
1912 }
1913 vdropl(vp);
1914 }
1915
1916 /*
1917 * Decrement only the use count and driver use count. This is intended to
1918 * be paired with a follow on vdropl() to release the remaining hold count.
1919 * In this way we may vgone() a vnode with a 0 usecount without risk of
1920 * having it end up on a free list because the hold count is kept above 0.
1921 */
1922 static void
1923 v_decr_useonly(struct vnode *vp)
1924 {
1925
1926 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
1927 vp, vp->v_holdcnt, vp->v_usecount);
1928 ASSERT_VI_LOCKED(vp, __FUNCTION__);
1929 VNASSERT(vp->v_usecount > 0, vp,
1930 ("v_decr_useonly: negative usecount"));
1931 vp->v_usecount--;
1932 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1933 dev_lock();
1934 vp->v_rdev->si_usecount--;
1935 dev_unlock();
1936 }
1937 }
1938
1939 /*
1940 * Grab a particular vnode from the free list, increment its
1941 * reference count and lock it. The vnode lock bit is set if the
1942 * vnode is being eliminated in vgone. The process is awakened
1943 * when the transition is completed, and an error returned to
1944 * indicate that the vnode is no longer usable (possibly having
1945 * been changed to a new filesystem type).
1946 */
1947 int
1948 vget(struct vnode *vp, int flags, struct thread *td)
1949 {
1950 int oweinact;
1951 int oldflags;
1952 int error;
1953
1954 error = 0;
1955 oldflags = flags;
1956 oweinact = 0;
1957 VFS_ASSERT_GIANT(vp->v_mount);
1958 if ((flags & LK_INTERLOCK) == 0)
1959 VI_LOCK(vp);
1960 /*
1961 * If the inactive call was deferred because vput() was called
1962 * with a shared lock, we have to do it here before another thread
1963 * gets a reference to data that should be dead.
1964 */
1965 if (vp->v_iflag & VI_OWEINACT) {
1966 if (flags & LK_NOWAIT) {
1967 VI_UNLOCK(vp);
1968 return (EBUSY);
1969 }
1970 flags &= ~LK_TYPE_MASK;
1971 flags |= LK_EXCLUSIVE;
1972 oweinact = 1;
1973 }
1974 vholdl(vp);
1975 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1976 vdrop(vp);
1977 return (error);
1978 }
1979 VI_LOCK(vp);
1980 /* Upgrade our holdcnt to a usecount. */
1981 v_upgrade_usecount(vp);
1982 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
1983 panic("vget: vn_lock failed to return ENOENT\n");
1984 if (oweinact) {
1985 if (vp->v_iflag & VI_OWEINACT)
1986 vinactive(vp, td);
1987 VI_UNLOCK(vp);
1988 if ((oldflags & LK_TYPE_MASK) == 0)
1989 VOP_UNLOCK(vp, 0, td);
1990 } else
1991 VI_UNLOCK(vp);
1992 return (0);
1993 }
1994
1995 /*
1996 * Increase the reference count of a vnode.
1997 */
1998 void
1999 vref(struct vnode *vp)
2000 {
2001
2002 VI_LOCK(vp);
2003 v_incr_usecount(vp);
2004 VI_UNLOCK(vp);
2005 }
2006
2007 /*
2008 * Return reference count of a vnode.
2009 *
2010 * The results of this call are only guaranteed when some mechanism other
2011 * than the VI lock is used to stop other processes from gaining references
2012 * to the vnode. This may be the case if the caller holds the only reference.
2013 * This is also useful when stale data is acceptable as race conditions may
2014 * be accounted for by some other means.
2015 */
2016 int
2017 vrefcnt(struct vnode *vp)
2018 {
2019 int usecnt;
2020
2021 VI_LOCK(vp);
2022 usecnt = vp->v_usecount;
2023 VI_UNLOCK(vp);
2024
2025 return (usecnt);
2026 }
2027
2028
2029 /*
2030 * Vnode put/release.
2031 * If count drops to zero, call inactive routine and return to freelist.
2032 */
2033 void
2034 vrele(vp)
2035 struct vnode *vp;
2036 {
2037 struct thread *td = curthread; /* XXX */
2038
2039 KASSERT(vp != NULL, ("vrele: null vp"));
2040 VFS_ASSERT_GIANT(vp->v_mount);
2041
2042 VI_LOCK(vp);
2043
2044 /* Skip this v_writecount check if we're going to panic below. */
2045 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2046 ("vrele: missed vn_close"));
2047
2048 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2049 vp->v_usecount == 1)) {
2050 v_decr_usecount(vp);
2051 return;
2052 }
2053 if (vp->v_usecount != 1) {
2054 #ifdef DIAGNOSTIC
2055 vprint("vrele: negative ref count", vp);
2056 #endif
2057 VI_UNLOCK(vp);
2058 panic("vrele: negative ref cnt");
2059 }
2060 /*
2061 * We want to hold the vnode until the inactive finishes to
2062 * prevent vgone() races. We drop the use count here and the
2063 * hold count below when we're done.
2064 */
2065 v_decr_useonly(vp);
2066 /*
2067 * We must call VOP_INACTIVE with the node locked. Mark
2068 * as VI_DOINGINACT to avoid recursion.
2069 */
2070 vp->v_iflag |= VI_OWEINACT;
2071 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2072 VI_LOCK(vp);
2073 if (vp->v_usecount > 0)
2074 vp->v_iflag &= ~VI_OWEINACT;
2075 if (vp->v_iflag & VI_OWEINACT)
2076 vinactive(vp, td);
2077 VOP_UNLOCK(vp, 0, td);
2078 } else {
2079 VI_LOCK(vp);
2080 if (vp->v_usecount > 0)
2081 vp->v_iflag &= ~VI_OWEINACT;
2082 }
2083 vdropl(vp);
2084 }
2085
2086 /*
2087 * Release an already locked vnode. This give the same effects as
2088 * unlock+vrele(), but takes less time and avoids releasing and
2089 * re-aquiring the lock (as vrele() aquires the lock internally.)
2090 */
2091 void
2092 vput(vp)
2093 struct vnode *vp;
2094 {
2095 struct thread *td = curthread; /* XXX */
2096 int error;
2097
2098 KASSERT(vp != NULL, ("vput: null vp"));
2099 ASSERT_VOP_LOCKED(vp, "vput");
2100 VFS_ASSERT_GIANT(vp->v_mount);
2101 VI_LOCK(vp);
2102 /* Skip this v_writecount check if we're going to panic below. */
2103 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2104 ("vput: missed vn_close"));
2105 error = 0;
2106
2107 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2108 vp->v_usecount == 1)) {
2109 VOP_UNLOCK(vp, 0, td);
2110 v_decr_usecount(vp);
2111 return;
2112 }
2113
2114 if (vp->v_usecount != 1) {
2115 #ifdef DIAGNOSTIC
2116 vprint("vput: negative ref count", vp);
2117 #endif
2118 panic("vput: negative ref cnt");
2119 }
2120 /*
2121 * We want to hold the vnode until the inactive finishes to
2122 * prevent vgone() races. We drop the use count here and the
2123 * hold count below when we're done.
2124 */
2125 v_decr_useonly(vp);
2126 vp->v_iflag |= VI_OWEINACT;
2127 if (VOP_ISLOCKED(vp, NULL) != LK_EXCLUSIVE) {
2128 error = VOP_LOCK(vp, LK_EXCLUPGRADE|LK_INTERLOCK|LK_NOWAIT, td);
2129 VI_LOCK(vp);
2130 if (error) {
2131 if (vp->v_usecount > 0)
2132 vp->v_iflag &= ~VI_OWEINACT;
2133 goto done;
2134 }
2135 }
2136 if (vp->v_usecount > 0)
2137 vp->v_iflag &= ~VI_OWEINACT;
2138 if (vp->v_iflag & VI_OWEINACT)
2139 vinactive(vp, td);
2140 VOP_UNLOCK(vp, 0, td);
2141 done:
2142 vdropl(vp);
2143 }
2144
2145 /*
2146 * Somebody doesn't want the vnode recycled.
2147 */
2148 void
2149 vhold(struct vnode *vp)
2150 {
2151
2152 VI_LOCK(vp);
2153 vholdl(vp);
2154 VI_UNLOCK(vp);
2155 }
2156
2157 void
2158 vholdl(struct vnode *vp)
2159 {
2160
2161 vp->v_holdcnt++;
2162 if (VSHOULDBUSY(vp))
2163 vbusy(vp);
2164 }
2165
2166 /*
2167 * Note that there is one less who cares about this vnode. vdrop() is the
2168 * opposite of vhold().
2169 */
2170 void
2171 vdrop(struct vnode *vp)
2172 {
2173
2174 VI_LOCK(vp);
2175 vdropl(vp);
2176 }
2177
2178 /*
2179 * Drop the hold count of the vnode. If this is the last reference to
2180 * the vnode we will free it if it has been vgone'd otherwise it is
2181 * placed on the free list.
2182 */
2183 static void
2184 vdropl(struct vnode *vp)
2185 {
2186
2187 if (vp->v_holdcnt <= 0)
2188 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2189 vp->v_holdcnt--;
2190 if (vp->v_holdcnt == 0) {
2191 if (vp->v_iflag & VI_DOOMED) {
2192 vdestroy(vp);
2193 return;
2194 } else
2195 vfree(vp);
2196 }
2197 VI_UNLOCK(vp);
2198 }
2199
2200 /*
2201 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2202 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2203 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2204 * failed lock upgrade.
2205 */
2206 static void
2207 vinactive(struct vnode *vp, struct thread *td)
2208 {
2209
2210 ASSERT_VOP_LOCKED(vp, "vinactive");
2211 ASSERT_VI_LOCKED(vp, "vinactive");
2212 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2213 ("vinactive: recursed on VI_DOINGINACT"));
2214 vp->v_iflag |= VI_DOINGINACT;
2215 vp->v_iflag &= ~VI_OWEINACT;
2216 VI_UNLOCK(vp);
2217 VOP_INACTIVE(vp, td);
2218 VI_LOCK(vp);
2219 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2220 ("vinactive: lost VI_DOINGINACT"));
2221 vp->v_iflag &= ~VI_DOINGINACT;
2222 }
2223
2224 /*
2225 * Remove any vnodes in the vnode table belonging to mount point mp.
2226 *
2227 * If FORCECLOSE is not specified, there should not be any active ones,
2228 * return error if any are found (nb: this is a user error, not a
2229 * system error). If FORCECLOSE is specified, detach any active vnodes
2230 * that are found.
2231 *
2232 * If WRITECLOSE is set, only flush out regular file vnodes open for
2233 * writing.
2234 *
2235 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2236 *
2237 * `rootrefs' specifies the base reference count for the root vnode
2238 * of this filesystem. The root vnode is considered busy if its
2239 * v_usecount exceeds this value. On a successful return, vflush(, td)
2240 * will call vrele() on the root vnode exactly rootrefs times.
2241 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2242 * be zero.
2243 */
2244 #ifdef DIAGNOSTIC
2245 static int busyprt = 0; /* print out busy vnodes */
2246 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2247 #endif
2248
2249 int
2250 vflush(mp, rootrefs, flags, td)
2251 struct mount *mp;
2252 int rootrefs;
2253 int flags;
2254 struct thread *td;
2255 {
2256 struct vnode *vp, *mvp, *rootvp = NULL;
2257 struct vattr vattr;
2258 int busy = 0, error;
2259
2260 CTR1(KTR_VFS, "vflush: mp %p", mp);
2261 if (rootrefs > 0) {
2262 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2263 ("vflush: bad args"));
2264 /*
2265 * Get the filesystem root vnode. We can vput() it
2266 * immediately, since with rootrefs > 0, it won't go away.
2267 */
2268 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2269 return (error);
2270 vput(rootvp);
2271
2272 }
2273 MNT_ILOCK(mp);
2274 loop:
2275 MNT_VNODE_FOREACH(vp, mp, mvp) {
2276
2277 VI_LOCK(vp);
2278 vholdl(vp);
2279 MNT_IUNLOCK(mp);
2280 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2281 if (error) {
2282 vdrop(vp);
2283 MNT_ILOCK(mp);
2284 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2285 goto loop;
2286 }
2287 /*
2288 * Skip over a vnodes marked VV_SYSTEM.
2289 */
2290 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2291 VOP_UNLOCK(vp, 0, td);
2292 vdrop(vp);
2293 MNT_ILOCK(mp);
2294 continue;
2295 }
2296 /*
2297 * If WRITECLOSE is set, flush out unlinked but still open
2298 * files (even if open only for reading) and regular file
2299 * vnodes open for writing.
2300 */
2301 if (flags & WRITECLOSE) {
2302 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2303 VI_LOCK(vp);
2304
2305 if ((vp->v_type == VNON ||
2306 (error == 0 && vattr.va_nlink > 0)) &&
2307 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2308 VOP_UNLOCK(vp, 0, td);
2309 vdropl(vp);
2310 MNT_ILOCK(mp);
2311 continue;
2312 }
2313 } else
2314 VI_LOCK(vp);
2315 /*
2316 * With v_usecount == 0, all we need to do is clear out the
2317 * vnode data structures and we are done.
2318 *
2319 * If FORCECLOSE is set, forcibly close the vnode.
2320 */
2321 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2322 VNASSERT(vp->v_usecount == 0 ||
2323 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2324 ("device VNODE %p is FORCECLOSED", vp));
2325 vgonel(vp);
2326 } else {
2327 busy++;
2328 #ifdef DIAGNOSTIC
2329 if (busyprt)
2330 vprint("vflush: busy vnode", vp);
2331 #endif
2332 }
2333 VOP_UNLOCK(vp, 0, td);
2334 vdropl(vp);
2335 MNT_ILOCK(mp);
2336 }
2337 MNT_IUNLOCK(mp);
2338 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2339 /*
2340 * If just the root vnode is busy, and if its refcount
2341 * is equal to `rootrefs', then go ahead and kill it.
2342 */
2343 VI_LOCK(rootvp);
2344 KASSERT(busy > 0, ("vflush: not busy"));
2345 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2346 ("vflush: usecount %d < rootrefs %d",
2347 rootvp->v_usecount, rootrefs));
2348 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2349 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK, td);
2350 vgone(rootvp);
2351 VOP_UNLOCK(rootvp, 0, td);
2352 busy = 0;
2353 } else
2354 VI_UNLOCK(rootvp);
2355 }
2356 if (busy)
2357 return (EBUSY);
2358 for (; rootrefs > 0; rootrefs--)
2359 vrele(rootvp);
2360 return (0);
2361 }
2362
2363 /*
2364 * Recycle an unused vnode to the front of the free list.
2365 */
2366 int
2367 vrecycle(struct vnode *vp, struct thread *td)
2368 {
2369 int recycled;
2370
2371 ASSERT_VOP_LOCKED(vp, "vrecycle");
2372 recycled = 0;
2373 VI_LOCK(vp);
2374 if (vp->v_usecount == 0) {
2375 recycled = 1;
2376 vgonel(vp);
2377 }
2378 VI_UNLOCK(vp);
2379 return (recycled);
2380 }
2381
2382 /*
2383 * Eliminate all activity associated with a vnode
2384 * in preparation for reuse.
2385 */
2386 void
2387 vgone(struct vnode *vp)
2388 {
2389 VI_LOCK(vp);
2390 vgonel(vp);
2391 VI_UNLOCK(vp);
2392 }
2393
2394 /*
2395 * vgone, with the vp interlock held.
2396 */
2397 void
2398 vgonel(struct vnode *vp)
2399 {
2400 struct thread *td;
2401 int oweinact;
2402 int active;
2403 struct mount *mp;
2404
2405 CTR1(KTR_VFS, "vgonel: vp %p", vp);
2406 ASSERT_VOP_LOCKED(vp, "vgonel");
2407 ASSERT_VI_LOCKED(vp, "vgonel");
2408 #if 0
2409 /* XXX Need to fix ttyvp before I enable this. */
2410 VNASSERT(vp->v_holdcnt, vp,
2411 ("vgonel: vp %p has no reference.", vp));
2412 #endif
2413 td = curthread;
2414
2415 /*
2416 * Don't vgonel if we're already doomed.
2417 */
2418 if (vp->v_iflag & VI_DOOMED)
2419 return;
2420 vp->v_iflag |= VI_DOOMED;
2421 /*
2422 * Check to see if the vnode is in use. If so, we have to call
2423 * VOP_CLOSE() and VOP_INACTIVE().
2424 */
2425 active = vp->v_usecount;
2426 oweinact = (vp->v_iflag & VI_OWEINACT);
2427 VI_UNLOCK(vp);
2428 /*
2429 * Clean out any buffers associated with the vnode.
2430 * If the flush fails, just toss the buffers.
2431 */
2432 mp = NULL;
2433 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2434 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2435 if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2436 vinvalbuf(vp, 0, td, 0, 0);
2437
2438 /*
2439 * If purging an active vnode, it must be closed and
2440 * deactivated before being reclaimed.
2441 */
2442 if (active)
2443 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2444 if (oweinact || active) {
2445 VI_LOCK(vp);
2446 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2447 vinactive(vp, td);
2448 VI_UNLOCK(vp);
2449 }
2450 /*
2451 * Reclaim the vnode.
2452 */
2453 if (VOP_RECLAIM(vp, td))
2454 panic("vgone: cannot reclaim");
2455 if (mp != NULL)
2456 vn_finished_secondary_write(mp);
2457 VNASSERT(vp->v_object == NULL, vp,
2458 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2459 /*
2460 * Delete from old mount point vnode list.
2461 */
2462 delmntque(vp);
2463 cache_purge(vp);
2464 /*
2465 * Done with purge, reset to the standard lock and invalidate
2466 * the vnode.
2467 */
2468 VI_LOCK(vp);
2469 vp->v_vnlock = &vp->v_lock;
2470 vp->v_op = &dead_vnodeops;
2471 vp->v_tag = "none";
2472 vp->v_type = VBAD;
2473 }
2474
2475 /*
2476 * Calculate the total number of references to a special device.
2477 */
2478 int
2479 vcount(vp)
2480 struct vnode *vp;
2481 {
2482 int count;
2483
2484 dev_lock();
2485 count = vp->v_rdev->si_usecount;
2486 dev_unlock();
2487 return (count);
2488 }
2489
2490 /*
2491 * Same as above, but using the struct cdev *as argument
2492 */
2493 int
2494 count_dev(dev)
2495 struct cdev *dev;
2496 {
2497 int count;
2498
2499 dev_lock();
2500 count = dev->si_usecount;
2501 dev_unlock();
2502 return(count);
2503 }
2504
2505 /*
2506 * Print out a description of a vnode.
2507 */
2508 static char *typename[] =
2509 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2510 "VMARKER"};
2511
2512 void
2513 vn_printf(struct vnode *vp, const char *fmt, ...)
2514 {
2515 va_list ap;
2516 char buf[96];
2517
2518 va_start(ap, fmt);
2519 vprintf(fmt, ap);
2520 va_end(ap);
2521 printf("%p: ", (void *)vp);
2522 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2523 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2524 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2525 buf[0] = '\0';
2526 buf[1] = '\0';
2527 if (vp->v_vflag & VV_ROOT)
2528 strcat(buf, "|VV_ROOT");
2529 if (vp->v_vflag & VV_TEXT)
2530 strcat(buf, "|VV_TEXT");
2531 if (vp->v_vflag & VV_SYSTEM)
2532 strcat(buf, "|VV_SYSTEM");
2533 if (vp->v_iflag & VI_DOOMED)
2534 strcat(buf, "|VI_DOOMED");
2535 if (vp->v_iflag & VI_FREE)
2536 strcat(buf, "|VI_FREE");
2537 printf(" flags (%s)\n", buf + 1);
2538 if (mtx_owned(VI_MTX(vp)))
2539 printf(" VI_LOCKed");
2540 if (vp->v_object != NULL)
2541 printf(" v_object %p ref %d pages %d\n",
2542 vp->v_object, vp->v_object->ref_count,
2543 vp->v_object->resident_page_count);
2544 printf(" ");
2545 lockmgr_printinfo(vp->v_vnlock);
2546 printf("\n");
2547 if (vp->v_data != NULL)
2548 VOP_PRINT(vp);
2549 }
2550
2551 #ifdef DDB
2552 #include <ddb/ddb.h>
2553 /*
2554 * List all of the locked vnodes in the system.
2555 * Called when debugging the kernel.
2556 */
2557 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2558 {
2559 struct mount *mp, *nmp;
2560 struct vnode *vp;
2561
2562 /*
2563 * Note: because this is DDB, we can't obey the locking semantics
2564 * for these structures, which means we could catch an inconsistent
2565 * state and dereference a nasty pointer. Not much to be done
2566 * about that.
2567 */
2568 printf("Locked vnodes\n");
2569 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2570 nmp = TAILQ_NEXT(mp, mnt_list);
2571 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2572 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp, NULL))
2573 vprint("", vp);
2574 }
2575 nmp = TAILQ_NEXT(mp, mnt_list);
2576 }
2577 }
2578 #endif
2579
2580 /*
2581 * Fill in a struct xvfsconf based on a struct vfsconf.
2582 */
2583 static void
2584 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2585 {
2586
2587 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2588 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2589 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2590 xvfsp->vfc_flags = vfsp->vfc_flags;
2591 /*
2592 * These are unused in userland, we keep them
2593 * to not break binary compatibility.
2594 */
2595 xvfsp->vfc_vfsops = NULL;
2596 xvfsp->vfc_next = NULL;
2597 }
2598
2599 /*
2600 * Top level filesystem related information gathering.
2601 */
2602 static int
2603 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2604 {
2605 struct vfsconf *vfsp;
2606 struct xvfsconf xvfsp;
2607 int error;
2608
2609 error = 0;
2610 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2611 bzero(&xvfsp, sizeof(xvfsp));
2612 vfsconf2x(vfsp, &xvfsp);
2613 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2614 if (error)
2615 break;
2616 }
2617 return (error);
2618 }
2619
2620 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2621 "S,xvfsconf", "List of all configured filesystems");
2622
2623 #ifndef BURN_BRIDGES
2624 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2625
2626 static int
2627 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2628 {
2629 int *name = (int *)arg1 - 1; /* XXX */
2630 u_int namelen = arg2 + 1; /* XXX */
2631 struct vfsconf *vfsp;
2632 struct xvfsconf xvfsp;
2633
2634 printf("WARNING: userland calling deprecated sysctl, "
2635 "please rebuild world\n");
2636
2637 #if 1 || defined(COMPAT_PRELITE2)
2638 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2639 if (namelen == 1)
2640 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2641 #endif
2642
2643 switch (name[1]) {
2644 case VFS_MAXTYPENUM:
2645 if (namelen != 2)
2646 return (ENOTDIR);
2647 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2648 case VFS_CONF:
2649 if (namelen != 3)
2650 return (ENOTDIR); /* overloaded */
2651 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2652 if (vfsp->vfc_typenum == name[2])
2653 break;
2654 if (vfsp == NULL)
2655 return (EOPNOTSUPP);
2656 bzero(&xvfsp, sizeof(xvfsp));
2657 vfsconf2x(vfsp, &xvfsp);
2658 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2659 }
2660 return (EOPNOTSUPP);
2661 }
2662
2663 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2664 vfs_sysctl, "Generic filesystem");
2665
2666 #if 1 || defined(COMPAT_PRELITE2)
2667
2668 static int
2669 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2670 {
2671 int error;
2672 struct vfsconf *vfsp;
2673 struct ovfsconf ovfs;
2674
2675 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2676 bzero(&ovfs, sizeof(ovfs));
2677 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2678 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2679 ovfs.vfc_index = vfsp->vfc_typenum;
2680 ovfs.vfc_refcount = vfsp->vfc_refcount;
2681 ovfs.vfc_flags = vfsp->vfc_flags;
2682 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2683 if (error)
2684 return error;
2685 }
2686 return 0;
2687 }
2688
2689 #endif /* 1 || COMPAT_PRELITE2 */
2690 #endif /* !BURN_BRIDGES */
2691
2692 #define KINFO_VNODESLOP 10
2693 #ifdef notyet
2694 /*
2695 * Dump vnode list (via sysctl).
2696 */
2697 /* ARGSUSED */
2698 static int
2699 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2700 {
2701 struct xvnode *xvn;
2702 struct thread *td = req->td;
2703 struct mount *mp;
2704 struct vnode *vp;
2705 int error, len, n;
2706
2707 /*
2708 * Stale numvnodes access is not fatal here.
2709 */
2710 req->lock = 0;
2711 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2712 if (!req->oldptr)
2713 /* Make an estimate */
2714 return (SYSCTL_OUT(req, 0, len));
2715
2716 error = sysctl_wire_old_buffer(req, 0);
2717 if (error != 0)
2718 return (error);
2719 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2720 n = 0;
2721 mtx_lock(&mountlist_mtx);
2722 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2723 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
2724 continue;
2725 MNT_ILOCK(mp);
2726 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2727 if (n == len)
2728 break;
2729 vref(vp);
2730 xvn[n].xv_size = sizeof *xvn;
2731 xvn[n].xv_vnode = vp;
2732 xvn[n].xv_id = 0; /* XXX compat */
2733 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
2734 XV_COPY(usecount);
2735 XV_COPY(writecount);
2736 XV_COPY(holdcnt);
2737 XV_COPY(mount);
2738 XV_COPY(numoutput);
2739 XV_COPY(type);
2740 #undef XV_COPY
2741 xvn[n].xv_flag = vp->v_vflag;
2742
2743 switch (vp->v_type) {
2744 case VREG:
2745 case VDIR:
2746 case VLNK:
2747 break;
2748 case VBLK:
2749 case VCHR:
2750 if (vp->v_rdev == NULL) {
2751 vrele(vp);
2752 continue;
2753 }
2754 xvn[n].xv_dev = dev2udev(vp->v_rdev);
2755 break;
2756 case VSOCK:
2757 xvn[n].xv_socket = vp->v_socket;
2758 break;
2759 case VFIFO:
2760 xvn[n].xv_fifo = vp->v_fifoinfo;
2761 break;
2762 case VNON:
2763 case VBAD:
2764 default:
2765 /* shouldn't happen? */
2766 vrele(vp);
2767 continue;
2768 }
2769 vrele(vp);
2770 ++n;
2771 }
2772 MNT_IUNLOCK(mp);
2773 mtx_lock(&mountlist_mtx);
2774 vfs_unbusy(mp, td);
2775 if (n == len)
2776 break;
2777 }
2778 mtx_unlock(&mountlist_mtx);
2779
2780 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
2781 free(xvn, M_TEMP);
2782 return (error);
2783 }
2784
2785 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2786 0, 0, sysctl_vnode, "S,xvnode", "");
2787 #endif
2788
2789 /*
2790 * Unmount all filesystems. The list is traversed in reverse order
2791 * of mounting to avoid dependencies.
2792 */
2793 void
2794 vfs_unmountall()
2795 {
2796 struct mount *mp;
2797 struct thread *td;
2798 int error;
2799
2800 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
2801 td = curthread;
2802 /*
2803 * Since this only runs when rebooting, it is not interlocked.
2804 */
2805 while(!TAILQ_EMPTY(&mountlist)) {
2806 mp = TAILQ_LAST(&mountlist, mntlist);
2807 error = dounmount(mp, MNT_FORCE, td);
2808 if (error) {
2809 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2810 /*
2811 * XXX: Due to the way in which we mount the root
2812 * file system off of devfs, devfs will generate a
2813 * "busy" warning when we try to unmount it before
2814 * the root. Don't print a warning as a result in
2815 * order to avoid false positive errors that may
2816 * cause needless upset.
2817 */
2818 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
2819 printf("unmount of %s failed (",
2820 mp->mnt_stat.f_mntonname);
2821 if (error == EBUSY)
2822 printf("BUSY)\n");
2823 else
2824 printf("%d)\n", error);
2825 }
2826 } else {
2827 /* The unmount has removed mp from the mountlist */
2828 }
2829 }
2830 }
2831
2832 /*
2833 * perform msync on all vnodes under a mount point
2834 * the mount point must be locked.
2835 */
2836 void
2837 vfs_msync(struct mount *mp, int flags)
2838 {
2839 struct vnode *vp, *mvp;
2840 struct vm_object *obj;
2841
2842 (void) vn_start_write(NULL, &mp, V_WAIT);
2843 MNT_ILOCK(mp);
2844 MNT_VNODE_FOREACH(vp, mp, mvp) {
2845 VI_LOCK(vp);
2846 if ((vp->v_iflag & VI_OBJDIRTY) &&
2847 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2848 MNT_IUNLOCK(mp);
2849 if (!vget(vp,
2850 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
2851 curthread)) {
2852 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
2853 vput(vp);
2854 MNT_ILOCK(mp);
2855 continue;
2856 }
2857
2858 obj = vp->v_object;
2859 if (obj != NULL) {
2860 VM_OBJECT_LOCK(obj);
2861 vm_object_page_clean(obj, 0, 0,
2862 flags == MNT_WAIT ?
2863 OBJPC_SYNC : OBJPC_NOSYNC);
2864 VM_OBJECT_UNLOCK(obj);
2865 }
2866 vput(vp);
2867 }
2868 MNT_ILOCK(mp);
2869 } else
2870 VI_UNLOCK(vp);
2871 }
2872 MNT_IUNLOCK(mp);
2873 vn_finished_write(mp);
2874 }
2875
2876 /*
2877 * Mark a vnode as free, putting it up for recycling.
2878 */
2879 static void
2880 vfree(struct vnode *vp)
2881 {
2882
2883 CTR1(KTR_VFS, "vfree vp %p", vp);
2884 ASSERT_VI_LOCKED(vp, "vfree");
2885 mtx_lock(&vnode_free_list_mtx);
2886 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
2887 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
2888 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
2889 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
2890 ("vfree: Freeing doomed vnode"));
2891 if (vp->v_iflag & VI_AGE) {
2892 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2893 } else {
2894 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2895 }
2896 freevnodes++;
2897 vp->v_iflag &= ~VI_AGE;
2898 vp->v_iflag |= VI_FREE;
2899 mtx_unlock(&vnode_free_list_mtx);
2900 }
2901
2902 /*
2903 * Opposite of vfree() - mark a vnode as in use.
2904 */
2905 static void
2906 vbusy(struct vnode *vp)
2907 {
2908 CTR1(KTR_VFS, "vbusy vp %p", vp);
2909 ASSERT_VI_LOCKED(vp, "vbusy");
2910 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2911 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
2912
2913 mtx_lock(&vnode_free_list_mtx);
2914 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2915 freevnodes--;
2916 vp->v_iflag &= ~(VI_FREE|VI_AGE);
2917 mtx_unlock(&vnode_free_list_mtx);
2918 }
2919
2920 /*
2921 * Initalize per-vnode helper structure to hold poll-related state.
2922 */
2923 void
2924 v_addpollinfo(struct vnode *vp)
2925 {
2926 struct vpollinfo *vi;
2927
2928 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
2929 if (vp->v_pollinfo != NULL) {
2930 uma_zfree(vnodepoll_zone, vi);
2931 return;
2932 }
2933 vp->v_pollinfo = vi;
2934 mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
2935 knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note, vp, vfs_knllock,
2936 vfs_knlunlock, vfs_knllocked);
2937 }
2938
2939 /*
2940 * Record a process's interest in events which might happen to
2941 * a vnode. Because poll uses the historic select-style interface
2942 * internally, this routine serves as both the ``check for any
2943 * pending events'' and the ``record my interest in future events''
2944 * functions. (These are done together, while the lock is held,
2945 * to avoid race conditions.)
2946 */
2947 int
2948 vn_pollrecord(vp, td, events)
2949 struct vnode *vp;
2950 struct thread *td;
2951 short events;
2952 {
2953
2954 if (vp->v_pollinfo == NULL)
2955 v_addpollinfo(vp);
2956 mtx_lock(&vp->v_pollinfo->vpi_lock);
2957 if (vp->v_pollinfo->vpi_revents & events) {
2958 /*
2959 * This leaves events we are not interested
2960 * in available for the other process which
2961 * which presumably had requested them
2962 * (otherwise they would never have been
2963 * recorded).
2964 */
2965 events &= vp->v_pollinfo->vpi_revents;
2966 vp->v_pollinfo->vpi_revents &= ~events;
2967
2968 mtx_unlock(&vp->v_pollinfo->vpi_lock);
2969 return events;
2970 }
2971 vp->v_pollinfo->vpi_events |= events;
2972 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
2973 mtx_unlock(&vp->v_pollinfo->vpi_lock);
2974 return 0;
2975 }
2976
2977 /*
2978 * Routine to create and manage a filesystem syncer vnode.
2979 */
2980 #define sync_close ((int (*)(struct vop_close_args *))nullop)
2981 static int sync_fsync(struct vop_fsync_args *);
2982 static int sync_inactive(struct vop_inactive_args *);
2983 static int sync_reclaim(struct vop_reclaim_args *);
2984
2985 static struct vop_vector sync_vnodeops = {
2986 .vop_bypass = VOP_EOPNOTSUPP,
2987 .vop_close = sync_close, /* close */
2988 .vop_fsync = sync_fsync, /* fsync */
2989 .vop_inactive = sync_inactive, /* inactive */
2990 .vop_reclaim = sync_reclaim, /* reclaim */
2991 .vop_lock = vop_stdlock, /* lock */
2992 .vop_unlock = vop_stdunlock, /* unlock */
2993 .vop_islocked = vop_stdislocked, /* islocked */
2994 };
2995
2996 /*
2997 * Create a new filesystem syncer vnode for the specified mount point.
2998 */
2999 int
3000 vfs_allocate_syncvnode(mp)
3001 struct mount *mp;
3002 {
3003 struct vnode *vp;
3004 static long start, incr, next;
3005 int error;
3006
3007 /* Allocate a new vnode */
3008 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3009 mp->mnt_syncer = NULL;
3010 return (error);
3011 }
3012 vp->v_type = VNON;
3013 /*
3014 * Place the vnode onto the syncer worklist. We attempt to
3015 * scatter them about on the list so that they will go off
3016 * at evenly distributed times even if all the filesystems
3017 * are mounted at once.
3018 */
3019 next += incr;
3020 if (next == 0 || next > syncer_maxdelay) {
3021 start /= 2;
3022 incr /= 2;
3023 if (start == 0) {
3024 start = syncer_maxdelay / 2;
3025 incr = syncer_maxdelay;
3026 }
3027 next = start;
3028 }
3029 VI_LOCK(vp);
3030 vn_syncer_add_to_worklist(&vp->v_bufobj,
3031 syncdelay > 0 ? next % syncdelay : 0);
3032 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3033 mtx_lock(&sync_mtx);
3034 sync_vnode_count++;
3035 mtx_unlock(&sync_mtx);
3036 VI_UNLOCK(vp);
3037 mp->mnt_syncer = vp;
3038 return (0);
3039 }
3040
3041 /*
3042 * Do a lazy sync of the filesystem.
3043 */
3044 static int
3045 sync_fsync(ap)
3046 struct vop_fsync_args /* {
3047 struct vnode *a_vp;
3048 struct ucred *a_cred;
3049 int a_waitfor;
3050 struct thread *a_td;
3051 } */ *ap;
3052 {
3053 struct vnode *syncvp = ap->a_vp;
3054 struct mount *mp = syncvp->v_mount;
3055 struct thread *td = ap->a_td;
3056 int error, asyncflag;
3057 struct bufobj *bo;
3058
3059 /*
3060 * We only need to do something if this is a lazy evaluation.
3061 */
3062 if (ap->a_waitfor != MNT_LAZY)
3063 return (0);
3064
3065 /*
3066 * Move ourselves to the back of the sync list.
3067 */
3068 bo = &syncvp->v_bufobj;
3069 BO_LOCK(bo);
3070 vn_syncer_add_to_worklist(bo, syncdelay);
3071 BO_UNLOCK(bo);
3072
3073 /*
3074 * Walk the list of vnodes pushing all that are dirty and
3075 * not already on the sync list.
3076 */
3077 mtx_lock(&mountlist_mtx);
3078 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3079 mtx_unlock(&mountlist_mtx);
3080 return (0);
3081 }
3082 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3083 vfs_unbusy(mp, td);
3084 return (0);
3085 }
3086 asyncflag = mp->mnt_flag & MNT_ASYNC;
3087 mp->mnt_flag &= ~MNT_ASYNC;
3088 vfs_msync(mp, MNT_NOWAIT);
3089 error = VFS_SYNC(mp, MNT_LAZY, td);
3090 if (asyncflag)
3091 mp->mnt_flag |= MNT_ASYNC;
3092 vn_finished_write(mp);
3093 vfs_unbusy(mp, td);
3094 return (error);
3095 }
3096
3097 /*
3098 * The syncer vnode is no referenced.
3099 */
3100 static int
3101 sync_inactive(ap)
3102 struct vop_inactive_args /* {
3103 struct vnode *a_vp;
3104 struct thread *a_td;
3105 } */ *ap;
3106 {
3107
3108 vgone(ap->a_vp);
3109 return (0);
3110 }
3111
3112 /*
3113 * The syncer vnode is no longer needed and is being decommissioned.
3114 *
3115 * Modifications to the worklist must be protected by sync_mtx.
3116 */
3117 static int
3118 sync_reclaim(ap)
3119 struct vop_reclaim_args /* {
3120 struct vnode *a_vp;
3121 } */ *ap;
3122 {
3123 struct vnode *vp = ap->a_vp;
3124 struct bufobj *bo;
3125
3126 VI_LOCK(vp);
3127 bo = &vp->v_bufobj;
3128 vp->v_mount->mnt_syncer = NULL;
3129 if (bo->bo_flag & BO_ONWORKLST) {
3130 mtx_lock(&sync_mtx);
3131 LIST_REMOVE(bo, bo_synclist);
3132 syncer_worklist_len--;
3133 sync_vnode_count--;
3134 mtx_unlock(&sync_mtx);
3135 bo->bo_flag &= ~BO_ONWORKLST;
3136 }
3137 VI_UNLOCK(vp);
3138
3139 return (0);
3140 }
3141
3142 /*
3143 * Check if vnode represents a disk device
3144 */
3145 int
3146 vn_isdisk(vp, errp)
3147 struct vnode *vp;
3148 int *errp;
3149 {
3150 int error;
3151
3152 error = 0;
3153 dev_lock();
3154 if (vp->v_type != VCHR)
3155 error = ENOTBLK;
3156 else if (vp->v_rdev == NULL)
3157 error = ENXIO;
3158 else if (vp->v_rdev->si_devsw == NULL)
3159 error = ENXIO;
3160 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3161 error = ENOTBLK;
3162 dev_unlock();
3163 if (errp != NULL)
3164 *errp = error;
3165 return (error == 0);
3166 }
3167
3168 /*
3169 * Common filesystem object access control check routine. Accepts a
3170 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3171 * and optional call-by-reference privused argument allowing vaccess()
3172 * to indicate to the caller whether privilege was used to satisfy the
3173 * request (obsoleted). Returns 0 on success, or an errno on failure.
3174 */
3175 int
3176 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3177 enum vtype type;
3178 mode_t file_mode;
3179 uid_t file_uid;
3180 gid_t file_gid;
3181 mode_t acc_mode;
3182 struct ucred *cred;
3183 int *privused;
3184 {
3185 mode_t dac_granted;
3186 #ifdef CAPABILITIES
3187 mode_t cap_granted;
3188 #endif
3189
3190 /*
3191 * Look for a normal, non-privileged way to access the file/directory
3192 * as requested. If it exists, go with that.
3193 */
3194
3195 if (privused != NULL)
3196 *privused = 0;
3197
3198 dac_granted = 0;
3199
3200 /* Check the owner. */
3201 if (cred->cr_uid == file_uid) {
3202 dac_granted |= VADMIN;
3203 if (file_mode & S_IXUSR)
3204 dac_granted |= VEXEC;
3205 if (file_mode & S_IRUSR)
3206 dac_granted |= VREAD;
3207 if (file_mode & S_IWUSR)
3208 dac_granted |= (VWRITE | VAPPEND);
3209
3210 if ((acc_mode & dac_granted) == acc_mode)
3211 return (0);
3212
3213 goto privcheck;
3214 }
3215
3216 /* Otherwise, check the groups (first match) */
3217 if (groupmember(file_gid, cred)) {
3218 if (file_mode & S_IXGRP)
3219 dac_granted |= VEXEC;
3220 if (file_mode & S_IRGRP)
3221 dac_granted |= VREAD;
3222 if (file_mode & S_IWGRP)
3223 dac_granted |= (VWRITE | VAPPEND);
3224
3225 if ((acc_mode & dac_granted) == acc_mode)
3226 return (0);
3227
3228 goto privcheck;
3229 }
3230
3231 /* Otherwise, check everyone else. */
3232 if (file_mode & S_IXOTH)
3233 dac_granted |= VEXEC;
3234 if (file_mode & S_IROTH)
3235 dac_granted |= VREAD;
3236 if (file_mode & S_IWOTH)
3237 dac_granted |= (VWRITE | VAPPEND);
3238 if ((acc_mode & dac_granted) == acc_mode)
3239 return (0);
3240
3241 privcheck:
3242 if (!suser_cred(cred, SUSER_ALLOWJAIL)) {
3243 /* XXX audit: privilege used */
3244 if (privused != NULL)
3245 *privused = 1;
3246 return (0);
3247 }
3248
3249 #ifdef CAPABILITIES
3250 /*
3251 * Build a capability mask to determine if the set of capabilities
3252 * satisfies the requirements when combined with the granted mask
3253 * from above.
3254 * For each capability, if the capability is required, bitwise
3255 * or the request type onto the cap_granted mask.
3256 */
3257 cap_granted = 0;
3258
3259 if (type == VDIR) {
3260 /*
3261 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3262 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3263 */
3264 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3265 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, SUSER_ALLOWJAIL))
3266 cap_granted |= VEXEC;
3267 } else {
3268 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3269 !cap_check(cred, NULL, CAP_DAC_EXECUTE, SUSER_ALLOWJAIL))
3270 cap_granted |= VEXEC;
3271 }
3272
3273 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3274 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, SUSER_ALLOWJAIL))
3275 cap_granted |= VREAD;
3276
3277 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3278 !cap_check(cred, NULL, CAP_DAC_WRITE, SUSER_ALLOWJAIL))
3279 cap_granted |= (VWRITE | VAPPEND);
3280
3281 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3282 !cap_check(cred, NULL, CAP_FOWNER, SUSER_ALLOWJAIL))
3283 cap_granted |= VADMIN;
3284
3285 if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3286 /* XXX audit: privilege used */
3287 if (privused != NULL)
3288 *privused = 1;
3289 return (0);
3290 }
3291 #endif
3292
3293 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3294 }
3295
3296 /*
3297 * Credential check based on process requesting service, and per-attribute
3298 * permissions.
3299 */
3300 int
3301 extattr_check_cred(struct vnode *vp, int attrnamespace,
3302 struct ucred *cred, struct thread *td, int access)
3303 {
3304
3305 /*
3306 * Kernel-invoked always succeeds.
3307 */
3308 if (cred == NOCRED)
3309 return (0);
3310
3311 /*
3312 * Do not allow privileged processes in jail to directly
3313 * manipulate system attributes.
3314 *
3315 * XXX What capability should apply here?
3316 * Probably CAP_SYS_SETFFLAG.
3317 */
3318 switch (attrnamespace) {
3319 case EXTATTR_NAMESPACE_SYSTEM:
3320 /* Potentially should be: return (EPERM); */
3321 return (suser_cred(cred, 0));
3322 case EXTATTR_NAMESPACE_USER:
3323 return (VOP_ACCESS(vp, access, cred, td));
3324 default:
3325 return (EPERM);
3326 }
3327 }
3328
3329 #ifdef DEBUG_VFS_LOCKS
3330 /*
3331 * This only exists to supress warnings from unlocked specfs accesses. It is
3332 * no longer ok to have an unlocked VFS.
3333 */
3334 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3335
3336 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3337 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3338
3339 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3340 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3341
3342 int vfs_badlock_print = 1; /* Print lock violations. */
3343 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3344
3345 #ifdef KDB
3346 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3347 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3348 #endif
3349
3350 static void
3351 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3352 {
3353
3354 #ifdef KDB
3355 if (vfs_badlock_backtrace)
3356 kdb_backtrace();
3357 #endif
3358 if (vfs_badlock_print)
3359 printf("%s: %p %s\n", str, (void *)vp, msg);
3360 if (vfs_badlock_ddb)
3361 kdb_enter("lock violation");
3362 }
3363
3364 void
3365 assert_vi_locked(struct vnode *vp, const char *str)
3366 {
3367
3368 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3369 vfs_badlock("interlock is not locked but should be", str, vp);
3370 }
3371
3372 void
3373 assert_vi_unlocked(struct vnode *vp, const char *str)
3374 {
3375
3376 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3377 vfs_badlock("interlock is locked but should not be", str, vp);
3378 }
3379
3380 void
3381 assert_vop_locked(struct vnode *vp, const char *str)
3382 {
3383
3384 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
3385 vfs_badlock("is not locked but should be", str, vp);
3386 }
3387
3388 void
3389 assert_vop_unlocked(struct vnode *vp, const char *str)
3390 {
3391
3392 if (vp && !IGNORE_LOCK(vp) &&
3393 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
3394 vfs_badlock("is locked but should not be", str, vp);
3395 }
3396
3397 void
3398 assert_vop_elocked(struct vnode *vp, const char *str)
3399 {
3400
3401 if (vp && !IGNORE_LOCK(vp) &&
3402 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
3403 vfs_badlock("is not exclusive locked but should be", str, vp);
3404 }
3405
3406 #if 0
3407 void
3408 assert_vop_elocked_other(struct vnode *vp, const char *str)
3409 {
3410
3411 if (vp && !IGNORE_LOCK(vp) &&
3412 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
3413 vfs_badlock("is not exclusive locked by another thread",
3414 str, vp);
3415 }
3416
3417 void
3418 assert_vop_slocked(struct vnode *vp, const char *str)
3419 {
3420
3421 if (vp && !IGNORE_LOCK(vp) &&
3422 VOP_ISLOCKED(vp, curthread) != LK_SHARED)
3423 vfs_badlock("is not locked shared but should be", str, vp);
3424 }
3425 #endif /* 0 */
3426 #endif /* DEBUG_VFS_LOCKS */
3427
3428 void
3429 vop_rename_pre(void *ap)
3430 {
3431 struct vop_rename_args *a = ap;
3432
3433 #ifdef DEBUG_VFS_LOCKS
3434 if (a->a_tvp)
3435 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3436 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3437 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3438 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3439
3440 /* Check the source (from). */
3441 if (a->a_tdvp != a->a_fdvp)
3442 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3443 if (a->a_tvp != a->a_fvp)
3444 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked");
3445
3446 /* Check the target. */
3447 if (a->a_tvp)
3448 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3449 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3450 #endif
3451 if (a->a_tdvp != a->a_fdvp)
3452 vhold(a->a_fdvp);
3453 if (a->a_tvp != a->a_fvp)
3454 vhold(a->a_fvp);
3455 vhold(a->a_tdvp);
3456 if (a->a_tvp)
3457 vhold(a->a_tvp);
3458 }
3459
3460 void
3461 vop_strategy_pre(void *ap)
3462 {
3463 #ifdef DEBUG_VFS_LOCKS
3464 struct vop_strategy_args *a;
3465 struct buf *bp;
3466
3467 a = ap;
3468 bp = a->a_bp;
3469
3470 /*
3471 * Cluster ops lock their component buffers but not the IO container.
3472 */
3473 if ((bp->b_flags & B_CLUSTER) != 0)
3474 return;
3475
3476 if (BUF_REFCNT(bp) < 1) {
3477 if (vfs_badlock_print)
3478 printf(
3479 "VOP_STRATEGY: bp is not locked but should be\n");
3480 if (vfs_badlock_ddb)
3481 kdb_enter("lock violation");
3482 }
3483 #endif
3484 }
3485
3486 void
3487 vop_lookup_pre(void *ap)
3488 {
3489 #ifdef DEBUG_VFS_LOCKS
3490 struct vop_lookup_args *a;
3491 struct vnode *dvp;
3492
3493 a = ap;
3494 dvp = a->a_dvp;
3495 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3496 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3497 #endif
3498 }
3499
3500 void
3501 vop_lookup_post(void *ap, int rc)
3502 {
3503 #ifdef DEBUG_VFS_LOCKS
3504 struct vop_lookup_args *a;
3505 struct vnode *dvp;
3506 struct vnode *vp;
3507
3508 a = ap;
3509 dvp = a->a_dvp;
3510 vp = *(a->a_vpp);
3511
3512 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3513 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3514
3515 if (!rc)
3516 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3517 #endif
3518 }
3519
3520 void
3521 vop_lock_pre(void *ap)
3522 {
3523 #ifdef DEBUG_VFS_LOCKS
3524 struct vop_lock_args *a = ap;
3525
3526 if ((a->a_flags & LK_INTERLOCK) == 0)
3527 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3528 else
3529 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3530 #endif
3531 }
3532
3533 void
3534 vop_lock_post(void *ap, int rc)
3535 {
3536 #ifdef DEBUG_VFS_LOCKS
3537 struct vop_lock_args *a = ap;
3538
3539 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3540 if (rc == 0)
3541 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3542 #endif
3543 }
3544
3545 void
3546 vop_unlock_pre(void *ap)
3547 {
3548 #ifdef DEBUG_VFS_LOCKS
3549 struct vop_unlock_args *a = ap;
3550
3551 if (a->a_flags & LK_INTERLOCK)
3552 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3553 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3554 #endif
3555 }
3556
3557 void
3558 vop_unlock_post(void *ap, int rc)
3559 {
3560 #ifdef DEBUG_VFS_LOCKS
3561 struct vop_unlock_args *a = ap;
3562
3563 if (a->a_flags & LK_INTERLOCK)
3564 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3565 #endif
3566 }
3567
3568 void
3569 vop_create_post(void *ap, int rc)
3570 {
3571 struct vop_create_args *a = ap;
3572
3573 if (!rc)
3574 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3575 }
3576
3577 void
3578 vop_link_post(void *ap, int rc)
3579 {
3580 struct vop_link_args *a = ap;
3581
3582 if (!rc) {
3583 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3584 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3585 }
3586 }
3587
3588 void
3589 vop_mkdir_post(void *ap, int rc)
3590 {
3591 struct vop_mkdir_args *a = ap;
3592
3593 if (!rc)
3594 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3595 }
3596
3597 void
3598 vop_mknod_post(void *ap, int rc)
3599 {
3600 struct vop_mknod_args *a = ap;
3601
3602 if (!rc)
3603 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3604 }
3605
3606 void
3607 vop_remove_post(void *ap, int rc)
3608 {
3609 struct vop_remove_args *a = ap;
3610
3611 if (!rc) {
3612 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3613 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3614 }
3615 }
3616
3617 void
3618 vop_rename_post(void *ap, int rc)
3619 {
3620 struct vop_rename_args *a = ap;
3621
3622 if (!rc) {
3623 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3624 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3625 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3626 if (a->a_tvp)
3627 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3628 }
3629 if (a->a_tdvp != a->a_fdvp)
3630 vdrop(a->a_fdvp);
3631 if (a->a_tvp != a->a_fvp)
3632 vdrop(a->a_fvp);
3633 vdrop(a->a_tdvp);
3634 if (a->a_tvp)
3635 vdrop(a->a_tvp);
3636 }
3637
3638 void
3639 vop_rmdir_post(void *ap, int rc)
3640 {
3641 struct vop_rmdir_args *a = ap;
3642
3643 if (!rc) {
3644 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3645 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3646 }
3647 }
3648
3649 void
3650 vop_setattr_post(void *ap, int rc)
3651 {
3652 struct vop_setattr_args *a = ap;
3653
3654 if (!rc)
3655 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3656 }
3657
3658 void
3659 vop_symlink_post(void *ap, int rc)
3660 {
3661 struct vop_symlink_args *a = ap;
3662
3663 if (!rc)
3664 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3665 }
3666
3667 static struct knlist fs_knlist;
3668
3669 static void
3670 vfs_event_init(void *arg)
3671 {
3672 knlist_init(&fs_knlist, NULL, NULL, NULL, NULL);
3673 }
3674 /* XXX - correct order? */
3675 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3676
3677 void
3678 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3679 {
3680
3681 KNOTE_UNLOCKED(&fs_knlist, event);
3682 }
3683
3684 static int filt_fsattach(struct knote *kn);
3685 static void filt_fsdetach(struct knote *kn);
3686 static int filt_fsevent(struct knote *kn, long hint);
3687
3688 struct filterops fs_filtops =
3689 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3690
3691 static int
3692 filt_fsattach(struct knote *kn)
3693 {
3694
3695 kn->kn_flags |= EV_CLEAR;
3696 knlist_add(&fs_knlist, kn, 0);
3697 return (0);
3698 }
3699
3700 static void
3701 filt_fsdetach(struct knote *kn)
3702 {
3703
3704 knlist_remove(&fs_knlist, kn, 0);
3705 }
3706
3707 static int
3708 filt_fsevent(struct knote *kn, long hint)
3709 {
3710
3711 kn->kn_fflags |= hint;
3712 return (kn->kn_fflags != 0);
3713 }
3714
3715 static int
3716 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
3717 {
3718 struct vfsidctl vc;
3719 int error;
3720 struct mount *mp;
3721
3722 error = SYSCTL_IN(req, &vc, sizeof(vc));
3723 if (error)
3724 return (error);
3725 if (vc.vc_vers != VFS_CTL_VERS1)
3726 return (EINVAL);
3727 mp = vfs_getvfs(&vc.vc_fsid);
3728 if (mp == NULL)
3729 return (ENOENT);
3730 /* ensure that a specific sysctl goes to the right filesystem. */
3731 if (strcmp(vc.vc_fstypename, "*") != 0 &&
3732 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
3733 return (EINVAL);
3734 }
3735 VCTLTOREQ(&vc, req);
3736 return (VFS_SYSCTL(mp, vc.vc_op, req));
3737 }
3738
3739 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR,
3740 NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid");
3741
3742 /*
3743 * Function to initialize a va_filerev field sensibly.
3744 * XXX: Wouldn't a random number make a lot more sense ??
3745 */
3746 u_quad_t
3747 init_va_filerev(void)
3748 {
3749 struct bintime bt;
3750
3751 getbinuptime(&bt);
3752 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
3753 }
3754
3755 static int filt_vfsread(struct knote *kn, long hint);
3756 static int filt_vfswrite(struct knote *kn, long hint);
3757 static int filt_vfsvnode(struct knote *kn, long hint);
3758 static void filt_vfsdetach(struct knote *kn);
3759 static struct filterops vfsread_filtops =
3760 { 1, NULL, filt_vfsdetach, filt_vfsread };
3761 static struct filterops vfswrite_filtops =
3762 { 1, NULL, filt_vfsdetach, filt_vfswrite };
3763 static struct filterops vfsvnode_filtops =
3764 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
3765
3766 static void
3767 vfs_knllock(void *arg)
3768 {
3769 struct vnode *vp = arg;
3770
3771 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
3772 }
3773
3774 static void
3775 vfs_knlunlock(void *arg)
3776 {
3777 struct vnode *vp = arg;
3778
3779 VOP_UNLOCK(vp, 0, curthread);
3780 }
3781
3782 static int
3783 vfs_knllocked(void *arg)
3784 {
3785 struct vnode *vp = arg;
3786
3787 return (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE);
3788 }
3789
3790 int
3791 vfs_kqfilter(struct vop_kqfilter_args *ap)
3792 {
3793 struct vnode *vp = ap->a_vp;
3794 struct knote *kn = ap->a_kn;
3795 struct knlist *knl;
3796
3797 switch (kn->kn_filter) {
3798 case EVFILT_READ:
3799 kn->kn_fop = &vfsread_filtops;
3800 break;
3801 case EVFILT_WRITE:
3802 kn->kn_fop = &vfswrite_filtops;
3803 break;
3804 case EVFILT_VNODE:
3805 kn->kn_fop = &vfsvnode_filtops;
3806 break;
3807 default:
3808 return (EINVAL);
3809 }
3810
3811 kn->kn_hook = (caddr_t)vp;
3812
3813 if (vp->v_pollinfo == NULL)
3814 v_addpollinfo(vp);
3815 if (vp->v_pollinfo == NULL)
3816 return (ENOMEM);
3817 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
3818 knlist_add(knl, kn, 0);
3819
3820 return (0);
3821 }
3822
3823 /*
3824 * Detach knote from vnode
3825 */
3826 static void
3827 filt_vfsdetach(struct knote *kn)
3828 {
3829 struct vnode *vp = (struct vnode *)kn->kn_hook;
3830
3831 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
3832 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
3833 }
3834
3835 /*ARGSUSED*/
3836 static int
3837 filt_vfsread(struct knote *kn, long hint)
3838 {
3839 struct vnode *vp = (struct vnode *)kn->kn_hook;
3840 struct vattr va;
3841
3842 /*
3843 * filesystem is gone, so set the EOF flag and schedule
3844 * the knote for deletion.
3845 */
3846 if (hint == NOTE_REVOKE) {
3847 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
3848 return (1);
3849 }
3850
3851 if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
3852 return (0);
3853
3854 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
3855 return (kn->kn_data != 0);
3856 }
3857
3858 /*ARGSUSED*/
3859 static int
3860 filt_vfswrite(struct knote *kn, long hint)
3861 {
3862 /*
3863 * filesystem is gone, so set the EOF flag and schedule
3864 * the knote for deletion.
3865 */
3866 if (hint == NOTE_REVOKE)
3867 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
3868
3869 kn->kn_data = 0;
3870 return (1);
3871 }
3872
3873 static int
3874 filt_vfsvnode(struct knote *kn, long hint)
3875 {
3876 if (kn->kn_sfflags & hint)
3877 kn->kn_fflags |= hint;
3878 if (hint == NOTE_REVOKE) {
3879 kn->kn_flags |= EV_EOF;
3880 return (1);
3881 }
3882 return (kn->kn_fflags != 0);
3883 }
3884
3885 int
3886 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
3887 {
3888 int error;
3889
3890 if (dp->d_reclen > ap->a_uio->uio_resid)
3891 return (ENAMETOOLONG);
3892 error = uiomove(dp, dp->d_reclen, ap->a_uio);
3893 if (error) {
3894 if (ap->a_ncookies != NULL) {
3895 if (ap->a_cookies != NULL)
3896 free(ap->a_cookies, M_TEMP);
3897 ap->a_cookies = NULL;
3898 *ap->a_ncookies = 0;
3899 }
3900 return (error);
3901 }
3902 if (ap->a_ncookies == NULL)
3903 return (0);
3904
3905 KASSERT(ap->a_cookies,
3906 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
3907
3908 *ap->a_cookies = realloc(*ap->a_cookies,
3909 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
3910 (*ap->a_cookies)[*ap->a_ncookies] = off;
3911 return (0);
3912 }
3913
3914 /*
3915 * Mark for update the access time of the file if the filesystem
3916 * supports VA_MARK_ATIME. This functionality is used by execve
3917 * and mmap, so we want to avoid the synchronous I/O implied by
3918 * directly setting va_atime for the sake of efficiency.
3919 */
3920 void
3921 vfs_mark_atime(struct vnode *vp, struct thread *td)
3922 {
3923 struct vattr atimeattr;
3924
3925 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
3926 VATTR_NULL(&atimeattr);
3927 atimeattr.va_vaflags |= VA_MARK_ATIME;
3928 (void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td);
3929 }
3930 }
Cache object: 15929e1b48c1e1f8e956a550676e535e
|