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/6.0/sys/kern/vfs_subr.c 151671 2005-10-25 20:43:25Z kris $");
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 vfree(struct vnode *);
99 static void vnlru_free(int);
100 static void vdestroy(struct vnode *);
101 static void vgonel(struct vnode *);
102 static void vfs_knllock(void *arg);
103 static void vfs_knlunlock(void *arg);
104 static int vfs_knllocked(void *arg);
105
106
107 /*
108 * Enable Giant pushdown based on whether or not the vm is mpsafe in this
109 * build. Without mpsafevm the buffer cache can not run Giant free.
110 */
111 #if defined(__alpha__) || defined(__amd64__) || defined(__i386__) || \
112 defined(__sparc64__)
113 int mpsafe_vfs = 1;
114 #else
115 int mpsafe_vfs;
116 #endif
117 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
118 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
119 "MPSAFE VFS");
120
121 /*
122 * Number of vnodes in existence. Increased whenever getnewvnode()
123 * allocates a new vnode, never decreased.
124 */
125 static unsigned long numvnodes;
126
127 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
128
129 /*
130 * Conversion tables for conversion from vnode types to inode formats
131 * and back.
132 */
133 enum vtype iftovt_tab[16] = {
134 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
135 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
136 };
137 int vttoif_tab[9] = {
138 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
139 S_IFSOCK, S_IFIFO, S_IFMT,
140 };
141
142 /*
143 * List of vnodes that are ready for recycling.
144 */
145 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
146
147 /*
148 * Free vnode target. Free vnodes may simply be files which have been stat'd
149 * but not read. This is somewhat common, and a small cache of such files
150 * should be kept to avoid recreation costs.
151 */
152 static u_long wantfreevnodes;
153 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
154 /* Number of vnodes in the free list. */
155 static u_long freevnodes;
156 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
157
158 /*
159 * Various variables used for debugging the new implementation of
160 * reassignbuf().
161 * XXX these are probably of (very) limited utility now.
162 */
163 static int reassignbufcalls;
164 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
165
166 /*
167 * Cache for the mount type id assigned to NFS. This is used for
168 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
169 */
170 int nfs_mount_type = -1;
171
172 /* To keep more than one thread at a time from running vfs_getnewfsid */
173 static struct mtx mntid_mtx;
174
175 /*
176 * Lock for any access to the following:
177 * vnode_free_list
178 * numvnodes
179 * freevnodes
180 */
181 static struct mtx vnode_free_list_mtx;
182
183 /* Publicly exported FS */
184 struct nfs_public nfs_pub;
185
186 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
187 static uma_zone_t vnode_zone;
188 static uma_zone_t vnodepoll_zone;
189
190 /* Set to 1 to print out reclaim of active vnodes */
191 int prtactive;
192
193 /*
194 * The workitem queue.
195 *
196 * It is useful to delay writes of file data and filesystem metadata
197 * for tens of seconds so that quickly created and deleted files need
198 * not waste disk bandwidth being created and removed. To realize this,
199 * we append vnodes to a "workitem" queue. When running with a soft
200 * updates implementation, most pending metadata dependencies should
201 * not wait for more than a few seconds. Thus, mounted on block devices
202 * are delayed only about a half the time that file data is delayed.
203 * Similarly, directory updates are more critical, so are only delayed
204 * about a third the time that file data is delayed. Thus, there are
205 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
206 * one each second (driven off the filesystem syncer process). The
207 * syncer_delayno variable indicates the next queue that is to be processed.
208 * Items that need to be processed soon are placed in this queue:
209 *
210 * syncer_workitem_pending[syncer_delayno]
211 *
212 * A delay of fifteen seconds is done by placing the request fifteen
213 * entries later in the queue:
214 *
215 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
216 *
217 */
218 static int syncer_delayno;
219 static long syncer_mask;
220 LIST_HEAD(synclist, bufobj);
221 static struct synclist *syncer_workitem_pending;
222 /*
223 * The sync_mtx protects:
224 * bo->bo_synclist
225 * sync_vnode_count
226 * syncer_delayno
227 * syncer_state
228 * syncer_workitem_pending
229 * syncer_worklist_len
230 * rushjob
231 */
232 static struct mtx sync_mtx;
233
234 #define SYNCER_MAXDELAY 32
235 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
236 static int syncdelay = 30; /* max time to delay syncing data */
237 static int filedelay = 30; /* time to delay syncing files */
238 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
239 static int dirdelay = 29; /* time to delay syncing directories */
240 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
241 static int metadelay = 28; /* time to delay syncing metadata */
242 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
243 static int rushjob; /* number of slots to run ASAP */
244 static int stat_rush_requests; /* number of times I/O speeded up */
245 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
246
247 /*
248 * When shutting down the syncer, run it at four times normal speed.
249 */
250 #define SYNCER_SHUTDOWN_SPEEDUP 4
251 static int sync_vnode_count;
252 static int syncer_worklist_len;
253 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
254 syncer_state;
255
256 /*
257 * Number of vnodes we want to exist at any one time. This is mostly used
258 * to size hash tables in vnode-related code. It is normally not used in
259 * getnewvnode(), as wantfreevnodes is normally nonzero.)
260 *
261 * XXX desiredvnodes is historical cruft and should not exist.
262 */
263 int desiredvnodes;
264 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
265 &desiredvnodes, 0, "Maximum number of vnodes");
266 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
267 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
268 static int vnlru_nowhere;
269 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
270 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
271
272 /* Hook for calling soft updates. */
273 int (*softdep_process_worklist_hook)(struct mount *);
274
275 /*
276 * Macros to control when a vnode is freed and recycled. All require
277 * the vnode interlock.
278 */
279 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
280 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
281 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
282
283
284 /*
285 * Initialize the vnode management data structures.
286 */
287 #ifndef MAXVNODES_MAX
288 #define MAXVNODES_MAX 100000
289 #endif
290 static void
291 vntblinit(void *dummy __unused)
292 {
293
294 /*
295 * Desiredvnodes is a function of the physical memory size and
296 * the kernel's heap size. Specifically, desiredvnodes scales
297 * in proportion to the physical memory size until two fifths
298 * of the kernel's heap size is consumed by vnodes and vm
299 * objects.
300 */
301 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
302 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
303 if (desiredvnodes > MAXVNODES_MAX) {
304 if (bootverbose)
305 printf("Reducing kern.maxvnodes %d -> %d\n",
306 desiredvnodes, MAXVNODES_MAX);
307 desiredvnodes = MAXVNODES_MAX;
308 }
309 wantfreevnodes = desiredvnodes / 4;
310 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
311 TAILQ_INIT(&vnode_free_list);
312 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
313 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
314 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
315 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
316 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
317 /*
318 * Initialize the filesystem syncer.
319 */
320 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
321 &syncer_mask);
322 syncer_maxdelay = syncer_mask + 1;
323 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
324 }
325 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
326
327
328 /*
329 * Mark a mount point as busy. Used to synchronize access and to delay
330 * unmounting. Interlock is not released on failure.
331 */
332 int
333 vfs_busy(mp, flags, interlkp, td)
334 struct mount *mp;
335 int flags;
336 struct mtx *interlkp;
337 struct thread *td;
338 {
339 int lkflags;
340
341 MNT_ILOCK(mp);
342 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
343 if (flags & LK_NOWAIT) {
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|PDROP, "vfs_busy", 0);
357 if (interlkp)
358 mtx_lock(interlkp);
359 return (ENOENT);
360 }
361 if (interlkp)
362 mtx_unlock(interlkp);
363 lkflags = LK_SHARED | LK_INTERLOCK;
364 if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp), td))
365 panic("vfs_busy: unexpected lock failure");
366 return (0);
367 }
368
369 /*
370 * Free a busy filesystem.
371 */
372 void
373 vfs_unbusy(mp, td)
374 struct mount *mp;
375 struct thread *td;
376 {
377
378 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
379 }
380
381 /*
382 * Lookup a mount point by filesystem identifier.
383 */
384 struct mount *
385 vfs_getvfs(fsid)
386 fsid_t *fsid;
387 {
388 struct mount *mp;
389
390 mtx_lock(&mountlist_mtx);
391 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
392 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
393 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
394 mtx_unlock(&mountlist_mtx);
395 return (mp);
396 }
397 }
398 mtx_unlock(&mountlist_mtx);
399 return ((struct mount *) 0);
400 }
401
402 /*
403 * Check if a user can access priveledged mount options.
404 */
405 int
406 vfs_suser(struct mount *mp, struct thread *td)
407 {
408 int error;
409
410 if ((mp->mnt_flag & MNT_USER) == 0 ||
411 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
412 if ((error = suser(td)) != 0)
413 return (error);
414 }
415 return (0);
416 }
417
418 /*
419 * Get a new unique fsid. Try to make its val[0] unique, since this value
420 * will be used to create fake device numbers for stat(). Also try (but
421 * not so hard) make its val[0] unique mod 2^16, since some emulators only
422 * support 16-bit device numbers. We end up with unique val[0]'s for the
423 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
424 *
425 * Keep in mind that several mounts may be running in parallel. Starting
426 * the search one past where the previous search terminated is both a
427 * micro-optimization and a defense against returning the same fsid to
428 * different mounts.
429 */
430 void
431 vfs_getnewfsid(mp)
432 struct mount *mp;
433 {
434 static u_int16_t mntid_base;
435 fsid_t tfsid;
436 int mtype;
437
438 mtx_lock(&mntid_mtx);
439 mtype = mp->mnt_vfc->vfc_typenum;
440 tfsid.val[1] = mtype;
441 mtype = (mtype & 0xFF) << 24;
442 for (;;) {
443 tfsid.val[0] = makedev(255,
444 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
445 mntid_base++;
446 if (vfs_getvfs(&tfsid) == NULL)
447 break;
448 }
449 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
450 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
451 mtx_unlock(&mntid_mtx);
452 }
453
454 /*
455 * Knob to control the precision of file timestamps:
456 *
457 * 0 = seconds only; nanoseconds zeroed.
458 * 1 = seconds and nanoseconds, accurate within 1/HZ.
459 * 2 = seconds and nanoseconds, truncated to microseconds.
460 * >=3 = seconds and nanoseconds, maximum precision.
461 */
462 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
463
464 static int timestamp_precision = TSP_SEC;
465 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
466 ×tamp_precision, 0, "");
467
468 /*
469 * Get a current timestamp.
470 */
471 void
472 vfs_timestamp(tsp)
473 struct timespec *tsp;
474 {
475 struct timeval tv;
476
477 switch (timestamp_precision) {
478 case TSP_SEC:
479 tsp->tv_sec = time_second;
480 tsp->tv_nsec = 0;
481 break;
482 case TSP_HZ:
483 getnanotime(tsp);
484 break;
485 case TSP_USEC:
486 microtime(&tv);
487 TIMEVAL_TO_TIMESPEC(&tv, tsp);
488 break;
489 case TSP_NSEC:
490 default:
491 nanotime(tsp);
492 break;
493 }
494 }
495
496 /*
497 * Set vnode attributes to VNOVAL
498 */
499 void
500 vattr_null(vap)
501 struct vattr *vap;
502 {
503
504 vap->va_type = VNON;
505 vap->va_size = VNOVAL;
506 vap->va_bytes = VNOVAL;
507 vap->va_mode = VNOVAL;
508 vap->va_nlink = VNOVAL;
509 vap->va_uid = VNOVAL;
510 vap->va_gid = VNOVAL;
511 vap->va_fsid = VNOVAL;
512 vap->va_fileid = VNOVAL;
513 vap->va_blocksize = VNOVAL;
514 vap->va_rdev = VNOVAL;
515 vap->va_atime.tv_sec = VNOVAL;
516 vap->va_atime.tv_nsec = VNOVAL;
517 vap->va_mtime.tv_sec = VNOVAL;
518 vap->va_mtime.tv_nsec = VNOVAL;
519 vap->va_ctime.tv_sec = VNOVAL;
520 vap->va_ctime.tv_nsec = VNOVAL;
521 vap->va_birthtime.tv_sec = VNOVAL;
522 vap->va_birthtime.tv_nsec = VNOVAL;
523 vap->va_flags = VNOVAL;
524 vap->va_gen = VNOVAL;
525 vap->va_vaflags = 0;
526 }
527
528 /*
529 * This routine is called when we have too many vnodes. It attempts
530 * to free <count> vnodes and will potentially free vnodes that still
531 * have VM backing store (VM backing store is typically the cause
532 * of a vnode blowout so we want to do this). Therefore, this operation
533 * is not considered cheap.
534 *
535 * A number of conditions may prevent a vnode from being reclaimed.
536 * the buffer cache may have references on the vnode, a directory
537 * vnode may still have references due to the namei cache representing
538 * underlying files, or the vnode may be in active use. It is not
539 * desireable to reuse such vnodes. These conditions may cause the
540 * number of vnodes to reach some minimum value regardless of what
541 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
542 */
543 static int
544 vlrureclaim(struct mount *mp)
545 {
546 struct thread *td;
547 struct vnode *vp;
548 int done;
549 int trigger;
550 int usevnodes;
551 int count;
552
553 /*
554 * Calculate the trigger point, don't allow user
555 * screwups to blow us up. This prevents us from
556 * recycling vnodes with lots of resident pages. We
557 * aren't trying to free memory, we are trying to
558 * free vnodes.
559 */
560 usevnodes = desiredvnodes;
561 if (usevnodes <= 0)
562 usevnodes = 1;
563 trigger = cnt.v_page_count * 2 / usevnodes;
564 done = 0;
565 td = curthread;
566 vn_start_write(NULL, &mp, V_WAIT);
567 MNT_ILOCK(mp);
568 count = mp->mnt_nvnodelistsize / 10 + 1;
569 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
570 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
571 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
572 --count;
573 if (!VI_TRYLOCK(vp))
574 goto next_iter;
575 /*
576 * If it's been deconstructed already, it's still
577 * referenced, or it exceeds the trigger, skip it.
578 */
579 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
580 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
581 vp->v_object->resident_page_count > trigger)) {
582 VI_UNLOCK(vp);
583 goto next_iter;
584 }
585 MNT_IUNLOCK(mp);
586 vholdl(vp);
587 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT, td)) {
588 vdrop(vp);
589 goto next_iter_mntunlocked;
590 }
591 VI_LOCK(vp);
592 /*
593 * v_usecount may have been bumped after VOP_LOCK() dropped
594 * the vnode interlock and before it was locked again.
595 *
596 * It is not necessary to recheck VI_DOOMED because it can
597 * only be set by another thread that holds both the vnode
598 * lock and vnode interlock. If another thread has the
599 * vnode lock before we get to VOP_LOCK() and obtains the
600 * vnode interlock after VOP_LOCK() drops the vnode
601 * interlock, the other thread will be unable to drop the
602 * vnode lock before our VOP_LOCK() call fails.
603 */
604 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
605 (vp->v_object != NULL &&
606 vp->v_object->resident_page_count > trigger)) {
607 VOP_UNLOCK(vp, LK_INTERLOCK, td);
608 goto next_iter_mntunlocked;
609 }
610 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
611 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
612 vgonel(vp);
613 VOP_UNLOCK(vp, 0, td);
614 vdropl(vp);
615 done++;
616 next_iter_mntunlocked:
617 if ((count % 256) != 0)
618 goto relock_mnt;
619 goto yield;
620 next_iter:
621 if ((count % 256) != 0)
622 continue;
623 MNT_IUNLOCK(mp);
624 yield:
625 uio_yield();
626 relock_mnt:
627 MNT_ILOCK(mp);
628 }
629 MNT_IUNLOCK(mp);
630 vn_finished_write(mp);
631 return done;
632 }
633
634 /*
635 * Attempt to keep the free list at wantfreevnodes length.
636 */
637 static void
638 vnlru_free(int count)
639 {
640 struct vnode *vp;
641
642 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
643 for (; count > 0; count--) {
644 vp = TAILQ_FIRST(&vnode_free_list);
645 /*
646 * The list can be modified while the free_list_mtx
647 * has been dropped and vp could be NULL here.
648 */
649 if (!vp)
650 break;
651 VNASSERT(vp->v_op != NULL, vp,
652 ("vnlru_free: vnode already reclaimed."));
653 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
654 /*
655 * Don't recycle if we can't get the interlock.
656 */
657 if (!VI_TRYLOCK(vp)) {
658 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
659 continue;
660 }
661 VNASSERT(VCANRECYCLE(vp), vp,
662 ("vp inconsistent on freelist"));
663 freevnodes--;
664 vp->v_iflag &= ~VI_FREE;
665 vholdl(vp);
666 mtx_unlock(&vnode_free_list_mtx);
667 VI_UNLOCK(vp);
668 vtryrecycle(vp);
669 /*
670 * If the recycled succeeded this vdrop will actually free
671 * the vnode. If not it will simply place it back on
672 * the free list.
673 */
674 vdrop(vp);
675 mtx_lock(&vnode_free_list_mtx);
676 }
677 }
678 /*
679 * Attempt to recycle vnodes in a context that is always safe to block.
680 * Calling vlrurecycle() from the bowels of filesystem code has some
681 * interesting deadlock problems.
682 */
683 static struct proc *vnlruproc;
684 static int vnlruproc_sig;
685
686 static void
687 vnlru_proc(void)
688 {
689 struct mount *mp, *nmp;
690 int done;
691 struct proc *p = vnlruproc;
692 struct thread *td = FIRST_THREAD_IN_PROC(p);
693
694 mtx_lock(&Giant);
695
696 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
697 SHUTDOWN_PRI_FIRST);
698
699 for (;;) {
700 kthread_suspend_check(p);
701 mtx_lock(&vnode_free_list_mtx);
702 if (freevnodes > wantfreevnodes)
703 vnlru_free(freevnodes - wantfreevnodes);
704 if (numvnodes <= desiredvnodes * 9 / 10) {
705 vnlruproc_sig = 0;
706 wakeup(&vnlruproc_sig);
707 msleep(vnlruproc, &vnode_free_list_mtx,
708 PVFS|PDROP, "vlruwt", hz);
709 continue;
710 }
711 mtx_unlock(&vnode_free_list_mtx);
712 done = 0;
713 mtx_lock(&mountlist_mtx);
714 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
715 int vfsunlocked;
716 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
717 nmp = TAILQ_NEXT(mp, mnt_list);
718 continue;
719 }
720 if (!VFS_NEEDSGIANT(mp)) {
721 mtx_unlock(&Giant);
722 vfsunlocked = 1;
723 } else
724 vfsunlocked = 0;
725 done += vlrureclaim(mp);
726 if (vfsunlocked)
727 mtx_lock(&Giant);
728 mtx_lock(&mountlist_mtx);
729 nmp = TAILQ_NEXT(mp, mnt_list);
730 vfs_unbusy(mp, td);
731 }
732 mtx_unlock(&mountlist_mtx);
733 if (done == 0) {
734 #if 0
735 /* These messages are temporary debugging aids */
736 if (vnlru_nowhere < 5)
737 printf("vnlru process getting nowhere..\n");
738 else if (vnlru_nowhere == 5)
739 printf("vnlru process messages stopped.\n");
740 #endif
741 vnlru_nowhere++;
742 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
743 } else
744 uio_yield();
745 }
746 }
747
748 static struct kproc_desc vnlru_kp = {
749 "vnlru",
750 vnlru_proc,
751 &vnlruproc
752 };
753 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
754
755 /*
756 * Routines having to do with the management of the vnode table.
757 */
758
759 static void
760 vdestroy(struct vnode *vp)
761 {
762 struct bufobj *bo;
763
764 CTR1(KTR_VFS, "vdestroy vp %p", vp);
765 mtx_lock(&vnode_free_list_mtx);
766 numvnodes--;
767 mtx_unlock(&vnode_free_list_mtx);
768 bo = &vp->v_bufobj;
769 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
770 ("cleaned vnode still on the free list."));
771 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
772 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
773 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
774 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
775 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
776 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
777 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
778 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
779 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
780 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
781 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
782 #ifdef MAC
783 mac_destroy_vnode(vp);
784 #endif
785 if (vp->v_pollinfo != NULL) {
786 knlist_destroy(&vp->v_pollinfo->vpi_selinfo.si_note);
787 mtx_destroy(&vp->v_pollinfo->vpi_lock);
788 uma_zfree(vnodepoll_zone, vp->v_pollinfo);
789 }
790 #ifdef INVARIANTS
791 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
792 vp->v_op = NULL;
793 #endif
794 lockdestroy(vp->v_vnlock);
795 mtx_destroy(&vp->v_interlock);
796 uma_zfree(vnode_zone, vp);
797 }
798
799 /*
800 * Try to recycle a freed vnode. We abort if anyone picks up a reference
801 * before we actually vgone(). This function must be called with the vnode
802 * held to prevent the vnode from being returned to the free list midway
803 * through vgone().
804 */
805 static int
806 vtryrecycle(struct vnode *vp)
807 {
808 struct thread *td = curthread;
809 struct mount *vnmp;
810
811 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
812 VNASSERT(vp->v_holdcnt, vp,
813 ("vtryrecycle: Recycling vp %p without a reference.", vp));
814 /*
815 * This vnode may found and locked via some other list, if so we
816 * can't recycle it yet.
817 */
818 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
819 return (EWOULDBLOCK);
820 /*
821 * Don't recycle if its filesystem is being suspended.
822 */
823 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
824 VOP_UNLOCK(vp, 0, td);
825 return (EBUSY);
826 }
827 /*
828 * If we got this far, we need to acquire the interlock and see if
829 * anyone picked up this vnode from another list. If not, we will
830 * mark it with DOOMED via vgonel() so that anyone who does find it
831 * will skip over it.
832 */
833 VI_LOCK(vp);
834 if (vp->v_usecount) {
835 VOP_UNLOCK(vp, LK_INTERLOCK, td);
836 vn_finished_write(vnmp);
837 return (EBUSY);
838 }
839 if ((vp->v_iflag & VI_DOOMED) == 0)
840 vgonel(vp);
841 VOP_UNLOCK(vp, LK_INTERLOCK, td);
842 vn_finished_write(vnmp);
843 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
844 return (0);
845 }
846
847 /*
848 * Return the next vnode from the free list.
849 */
850 int
851 getnewvnode(tag, mp, vops, vpp)
852 const char *tag;
853 struct mount *mp;
854 struct vop_vector *vops;
855 struct vnode **vpp;
856 {
857 struct vnode *vp = NULL;
858 struct bufobj *bo;
859
860 mtx_lock(&vnode_free_list_mtx);
861 /*
862 * Lend our context to reclaim vnodes if they've exceeded the max.
863 */
864 if (freevnodes > wantfreevnodes)
865 vnlru_free(1);
866 /*
867 * Wait for available vnodes.
868 */
869 if (numvnodes > desiredvnodes) {
870 if (vnlruproc_sig == 0) {
871 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
872 wakeup(vnlruproc);
873 }
874 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
875 "vlruwk", hz);
876 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
877 if (numvnodes > desiredvnodes) {
878 mtx_unlock(&vnode_free_list_mtx);
879 return (ENFILE);
880 }
881 #endif
882 }
883 numvnodes++;
884 mtx_unlock(&vnode_free_list_mtx);
885 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
886 /*
887 * Setup locks.
888 */
889 vp->v_vnlock = &vp->v_lock;
890 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
891 /*
892 * By default, don't allow shared locks unless filesystems
893 * opt-in.
894 */
895 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
896 /*
897 * Initialize bufobj.
898 */
899 bo = &vp->v_bufobj;
900 bo->__bo_vnode = vp;
901 bo->bo_mtx = &vp->v_interlock;
902 bo->bo_ops = &buf_ops_bio;
903 bo->bo_private = vp;
904 TAILQ_INIT(&bo->bo_clean.bv_hd);
905 TAILQ_INIT(&bo->bo_dirty.bv_hd);
906 /*
907 * Initialize namecache.
908 */
909 LIST_INIT(&vp->v_cache_src);
910 TAILQ_INIT(&vp->v_cache_dst);
911 /*
912 * Finalize various vnode identity bits.
913 */
914 vp->v_type = VNON;
915 vp->v_tag = tag;
916 vp->v_op = vops;
917 v_incr_usecount(vp);
918 vp->v_data = 0;
919 #ifdef MAC
920 mac_init_vnode(vp);
921 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
922 mac_associate_vnode_singlelabel(mp, vp);
923 else if (mp == NULL)
924 printf("NULL mp in getnewvnode()\n");
925 #endif
926 delmntque(vp);
927 if (mp != NULL) {
928 insmntque(vp, mp);
929 bo->bo_bsize = mp->mnt_stat.f_iosize;
930 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
931 vp->v_vflag |= VV_NOKNOTE;
932 }
933
934 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
935 *vpp = vp;
936 return (0);
937 }
938
939 /*
940 * Delete from old mount point vnode list, if on one.
941 */
942 static void
943 delmntque(struct vnode *vp)
944 {
945 struct mount *mp;
946
947 if (vp->v_mount == NULL)
948 return;
949 mp = vp->v_mount;
950 MNT_ILOCK(mp);
951 vp->v_mount = NULL;
952 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
953 ("bad mount point vnode list size"));
954 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
955 mp->mnt_nvnodelistsize--;
956 MNT_IUNLOCK(mp);
957 }
958
959 /*
960 * Insert into list of vnodes for the new mount point, if available.
961 */
962 static void
963 insmntque(struct vnode *vp, struct mount *mp)
964 {
965
966 vp->v_mount = mp;
967 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
968 MNT_ILOCK(vp->v_mount);
969 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
970 mp->mnt_nvnodelistsize++;
971 MNT_IUNLOCK(vp->v_mount);
972 }
973
974 /*
975 * Flush out and invalidate all buffers associated with a bufobj
976 * Called with the underlying object locked.
977 */
978 int
979 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag, int slptimeo)
980 {
981 int error;
982
983 BO_LOCK(bo);
984 if (flags & V_SAVE) {
985 error = bufobj_wwait(bo, slpflag, slptimeo);
986 if (error) {
987 BO_UNLOCK(bo);
988 return (error);
989 }
990 if (bo->bo_dirty.bv_cnt > 0) {
991 BO_UNLOCK(bo);
992 if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
993 return (error);
994 /*
995 * XXX We could save a lock/unlock if this was only
996 * enabled under INVARIANTS
997 */
998 BO_LOCK(bo);
999 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1000 panic("vinvalbuf: dirty bufs");
1001 }
1002 }
1003 /*
1004 * If you alter this loop please notice that interlock is dropped and
1005 * reacquired in flushbuflist. Special care is needed to ensure that
1006 * no race conditions occur from this.
1007 */
1008 do {
1009 error = flushbuflist(&bo->bo_clean,
1010 flags, bo, slpflag, slptimeo);
1011 if (error == 0)
1012 error = flushbuflist(&bo->bo_dirty,
1013 flags, bo, slpflag, slptimeo);
1014 if (error != 0 && error != EAGAIN) {
1015 BO_UNLOCK(bo);
1016 return (error);
1017 }
1018 } while (error != 0);
1019
1020 /*
1021 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1022 * have write I/O in-progress but if there is a VM object then the
1023 * VM object can also have read-I/O in-progress.
1024 */
1025 do {
1026 bufobj_wwait(bo, 0, 0);
1027 BO_UNLOCK(bo);
1028 if (bo->bo_object != NULL) {
1029 VM_OBJECT_LOCK(bo->bo_object);
1030 vm_object_pip_wait(bo->bo_object, "bovlbx");
1031 VM_OBJECT_UNLOCK(bo->bo_object);
1032 }
1033 BO_LOCK(bo);
1034 } while (bo->bo_numoutput > 0);
1035 BO_UNLOCK(bo);
1036
1037 /*
1038 * Destroy the copy in the VM cache, too.
1039 */
1040 if (bo->bo_object != NULL) {
1041 VM_OBJECT_LOCK(bo->bo_object);
1042 vm_object_page_remove(bo->bo_object, 0, 0,
1043 (flags & V_SAVE) ? TRUE : FALSE);
1044 VM_OBJECT_UNLOCK(bo->bo_object);
1045 }
1046
1047 #ifdef INVARIANTS
1048 BO_LOCK(bo);
1049 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1050 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1051 panic("vinvalbuf: flush failed");
1052 BO_UNLOCK(bo);
1053 #endif
1054 return (0);
1055 }
1056
1057 /*
1058 * Flush out and invalidate all buffers associated with a vnode.
1059 * Called with the underlying object locked.
1060 */
1061 int
1062 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag, int slptimeo)
1063 {
1064
1065 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1066 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1067 return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1068 }
1069
1070 /*
1071 * Flush out buffers on the specified list.
1072 *
1073 */
1074 static int
1075 flushbuflist(bufv, flags, bo, slpflag, slptimeo)
1076 struct bufv *bufv;
1077 int flags;
1078 struct bufobj *bo;
1079 int slpflag, slptimeo;
1080 {
1081 struct buf *bp, *nbp;
1082 int retval, error;
1083 daddr_t lblkno;
1084 b_xflags_t xflags;
1085
1086 ASSERT_BO_LOCKED(bo);
1087
1088 retval = 0;
1089 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1090 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1091 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1092 continue;
1093 }
1094 lblkno = 0;
1095 xflags = 0;
1096 if (nbp != NULL) {
1097 lblkno = nbp->b_lblkno;
1098 xflags = nbp->b_xflags &
1099 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1100 }
1101 retval = EAGAIN;
1102 error = BUF_TIMELOCK(bp,
1103 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1104 "flushbuf", slpflag, slptimeo);
1105 if (error) {
1106 BO_LOCK(bo);
1107 return (error != ENOLCK ? error : EAGAIN);
1108 }
1109 KASSERT(bp->b_bufobj == bo,
1110 ("bp %p wrong b_bufobj %p should be %p",
1111 bp, bp->b_bufobj, bo));
1112 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1113 BUF_UNLOCK(bp);
1114 BO_LOCK(bo);
1115 return (EAGAIN);
1116 }
1117 /*
1118 * XXX Since there are no node locks for NFS, I
1119 * believe there is a slight chance that a delayed
1120 * write will occur while sleeping just above, so
1121 * check for it.
1122 */
1123 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1124 (flags & V_SAVE)) {
1125 bremfree(bp);
1126 bp->b_flags |= B_ASYNC;
1127 bwrite(bp);
1128 BO_LOCK(bo);
1129 return (EAGAIN); /* XXX: why not loop ? */
1130 }
1131 bremfree(bp);
1132 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1133 bp->b_flags &= ~B_ASYNC;
1134 brelse(bp);
1135 BO_LOCK(bo);
1136 if (nbp != NULL &&
1137 (nbp->b_bufobj != bo ||
1138 nbp->b_lblkno != lblkno ||
1139 (nbp->b_xflags &
1140 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1141 break; /* nbp invalid */
1142 }
1143 return (retval);
1144 }
1145
1146 /*
1147 * Truncate a file's buffer and pages to a specified length. This
1148 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1149 * sync activity.
1150 */
1151 int
1152 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td, off_t length, int blksize)
1153 {
1154 struct buf *bp, *nbp;
1155 int anyfreed;
1156 int trunclbn;
1157 struct bufobj *bo;
1158
1159 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1160 /*
1161 * Round up to the *next* lbn.
1162 */
1163 trunclbn = (length + blksize - 1) / blksize;
1164
1165 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1166 restart:
1167 VI_LOCK(vp);
1168 bo = &vp->v_bufobj;
1169 anyfreed = 1;
1170 for (;anyfreed;) {
1171 anyfreed = 0;
1172 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1173 if (bp->b_lblkno < trunclbn)
1174 continue;
1175 if (BUF_LOCK(bp,
1176 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1177 VI_MTX(vp)) == ENOLCK)
1178 goto restart;
1179
1180 bremfree(bp);
1181 bp->b_flags |= (B_INVAL | B_RELBUF);
1182 bp->b_flags &= ~B_ASYNC;
1183 brelse(bp);
1184 anyfreed = 1;
1185
1186 if (nbp != NULL &&
1187 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1188 (nbp->b_vp != vp) ||
1189 (nbp->b_flags & B_DELWRI))) {
1190 goto restart;
1191 }
1192 VI_LOCK(vp);
1193 }
1194
1195 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1196 if (bp->b_lblkno < trunclbn)
1197 continue;
1198 if (BUF_LOCK(bp,
1199 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1200 VI_MTX(vp)) == ENOLCK)
1201 goto restart;
1202 bremfree(bp);
1203 bp->b_flags |= (B_INVAL | B_RELBUF);
1204 bp->b_flags &= ~B_ASYNC;
1205 brelse(bp);
1206 anyfreed = 1;
1207 if (nbp != NULL &&
1208 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1209 (nbp->b_vp != vp) ||
1210 (nbp->b_flags & B_DELWRI) == 0)) {
1211 goto restart;
1212 }
1213 VI_LOCK(vp);
1214 }
1215 }
1216
1217 if (length > 0) {
1218 restartsync:
1219 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1220 if (bp->b_lblkno > 0)
1221 continue;
1222 /*
1223 * Since we hold the vnode lock this should only
1224 * fail if we're racing with the buf daemon.
1225 */
1226 if (BUF_LOCK(bp,
1227 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1228 VI_MTX(vp)) == ENOLCK) {
1229 goto restart;
1230 }
1231 VNASSERT((bp->b_flags & B_DELWRI), vp,
1232 ("buf(%p) on dirty queue without DELWRI", bp));
1233
1234 bremfree(bp);
1235 bawrite(bp);
1236 VI_LOCK(vp);
1237 goto restartsync;
1238 }
1239 }
1240
1241 bufobj_wwait(bo, 0, 0);
1242 VI_UNLOCK(vp);
1243 vnode_pager_setsize(vp, length);
1244
1245 return (0);
1246 }
1247
1248 /*
1249 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1250 * a vnode.
1251 *
1252 * NOTE: We have to deal with the special case of a background bitmap
1253 * buffer, a situation where two buffers will have the same logical
1254 * block offset. We want (1) only the foreground buffer to be accessed
1255 * in a lookup and (2) must differentiate between the foreground and
1256 * background buffer in the splay tree algorithm because the splay
1257 * tree cannot normally handle multiple entities with the same 'index'.
1258 * We accomplish this by adding differentiating flags to the splay tree's
1259 * numerical domain.
1260 */
1261 static
1262 struct buf *
1263 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1264 {
1265 struct buf dummy;
1266 struct buf *lefttreemax, *righttreemin, *y;
1267
1268 if (root == NULL)
1269 return (NULL);
1270 lefttreemax = righttreemin = &dummy;
1271 for (;;) {
1272 if (lblkno < root->b_lblkno ||
1273 (lblkno == root->b_lblkno &&
1274 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1275 if ((y = root->b_left) == NULL)
1276 break;
1277 if (lblkno < y->b_lblkno) {
1278 /* Rotate right. */
1279 root->b_left = y->b_right;
1280 y->b_right = root;
1281 root = y;
1282 if ((y = root->b_left) == NULL)
1283 break;
1284 }
1285 /* Link into the new root's right tree. */
1286 righttreemin->b_left = root;
1287 righttreemin = root;
1288 } else if (lblkno > root->b_lblkno ||
1289 (lblkno == root->b_lblkno &&
1290 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1291 if ((y = root->b_right) == NULL)
1292 break;
1293 if (lblkno > y->b_lblkno) {
1294 /* Rotate left. */
1295 root->b_right = y->b_left;
1296 y->b_left = root;
1297 root = y;
1298 if ((y = root->b_right) == NULL)
1299 break;
1300 }
1301 /* Link into the new root's left tree. */
1302 lefttreemax->b_right = root;
1303 lefttreemax = root;
1304 } else {
1305 break;
1306 }
1307 root = y;
1308 }
1309 /* Assemble the new root. */
1310 lefttreemax->b_right = root->b_left;
1311 righttreemin->b_left = root->b_right;
1312 root->b_left = dummy.b_right;
1313 root->b_right = dummy.b_left;
1314 return (root);
1315 }
1316
1317 static void
1318 buf_vlist_remove(struct buf *bp)
1319 {
1320 struct buf *root;
1321 struct bufv *bv;
1322
1323 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1324 ASSERT_BO_LOCKED(bp->b_bufobj);
1325 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1326 (BX_VNDIRTY|BX_VNCLEAN),
1327 ("buf_vlist_remove: Buf %p is on two lists", bp));
1328 if (bp->b_xflags & BX_VNDIRTY)
1329 bv = &bp->b_bufobj->bo_dirty;
1330 else
1331 bv = &bp->b_bufobj->bo_clean;
1332 if (bp != bv->bv_root) {
1333 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1334 KASSERT(root == bp, ("splay lookup failed in remove"));
1335 }
1336 if (bp->b_left == NULL) {
1337 root = bp->b_right;
1338 } else {
1339 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1340 root->b_right = bp->b_right;
1341 }
1342 bv->bv_root = root;
1343 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1344 bv->bv_cnt--;
1345 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1346 }
1347
1348 /*
1349 * Add the buffer to the sorted clean or dirty block list using a
1350 * splay tree algorithm.
1351 *
1352 * NOTE: xflags is passed as a constant, optimizing this inline function!
1353 */
1354 static void
1355 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1356 {
1357 struct buf *root;
1358 struct bufv *bv;
1359
1360 ASSERT_BO_LOCKED(bo);
1361 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1362 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1363 bp->b_xflags |= xflags;
1364 if (xflags & BX_VNDIRTY)
1365 bv = &bo->bo_dirty;
1366 else
1367 bv = &bo->bo_clean;
1368
1369 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1370 if (root == NULL) {
1371 bp->b_left = NULL;
1372 bp->b_right = NULL;
1373 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1374 } else if (bp->b_lblkno < root->b_lblkno ||
1375 (bp->b_lblkno == root->b_lblkno &&
1376 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1377 bp->b_left = root->b_left;
1378 bp->b_right = root;
1379 root->b_left = NULL;
1380 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1381 } else {
1382 bp->b_right = root->b_right;
1383 bp->b_left = root;
1384 root->b_right = NULL;
1385 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1386 }
1387 bv->bv_cnt++;
1388 bv->bv_root = bp;
1389 }
1390
1391 /*
1392 * Lookup a buffer using the splay tree. Note that we specifically avoid
1393 * shadow buffers used in background bitmap writes.
1394 *
1395 * This code isn't quite efficient as it could be because we are maintaining
1396 * two sorted lists and do not know which list the block resides in.
1397 *
1398 * During a "make buildworld" the desired buffer is found at one of
1399 * the roots more than 60% of the time. Thus, checking both roots
1400 * before performing either splay eliminates unnecessary splays on the
1401 * first tree splayed.
1402 */
1403 struct buf *
1404 gbincore(struct bufobj *bo, daddr_t lblkno)
1405 {
1406 struct buf *bp;
1407
1408 ASSERT_BO_LOCKED(bo);
1409 if ((bp = bo->bo_clean.bv_root) != NULL &&
1410 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1411 return (bp);
1412 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1413 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1414 return (bp);
1415 if ((bp = bo->bo_clean.bv_root) != NULL) {
1416 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1417 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1418 return (bp);
1419 }
1420 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1421 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1422 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1423 return (bp);
1424 }
1425 return (NULL);
1426 }
1427
1428 /*
1429 * Associate a buffer with a vnode.
1430 */
1431 void
1432 bgetvp(struct vnode *vp, struct buf *bp)
1433 {
1434
1435 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1436
1437 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1438 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1439 ("bgetvp: bp already attached! %p", bp));
1440
1441 ASSERT_VI_LOCKED(vp, "bgetvp");
1442 vholdl(vp);
1443 bp->b_vp = vp;
1444 bp->b_bufobj = &vp->v_bufobj;
1445 /*
1446 * Insert onto list for new vnode.
1447 */
1448 buf_vlist_add(bp, &vp->v_bufobj, BX_VNCLEAN);
1449 }
1450
1451 /*
1452 * Disassociate a buffer from a vnode.
1453 */
1454 void
1455 brelvp(struct buf *bp)
1456 {
1457 struct bufobj *bo;
1458 struct vnode *vp;
1459
1460 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1461 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1462
1463 /*
1464 * Delete from old vnode list, if on one.
1465 */
1466 vp = bp->b_vp; /* XXX */
1467 bo = bp->b_bufobj;
1468 BO_LOCK(bo);
1469 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1470 buf_vlist_remove(bp);
1471 else
1472 panic("brelvp: Buffer %p not on queue.", bp);
1473 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1474 bo->bo_flag &= ~BO_ONWORKLST;
1475 mtx_lock(&sync_mtx);
1476 LIST_REMOVE(bo, bo_synclist);
1477 syncer_worklist_len--;
1478 mtx_unlock(&sync_mtx);
1479 }
1480 bp->b_vp = NULL;
1481 bp->b_bufobj = NULL;
1482 vdropl(vp);
1483 }
1484
1485 /*
1486 * Add an item to the syncer work queue.
1487 */
1488 static void
1489 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1490 {
1491 int slot;
1492
1493 ASSERT_BO_LOCKED(bo);
1494
1495 mtx_lock(&sync_mtx);
1496 if (bo->bo_flag & BO_ONWORKLST)
1497 LIST_REMOVE(bo, bo_synclist);
1498 else {
1499 bo->bo_flag |= BO_ONWORKLST;
1500 syncer_worklist_len++;
1501 }
1502
1503 if (delay > syncer_maxdelay - 2)
1504 delay = syncer_maxdelay - 2;
1505 slot = (syncer_delayno + delay) & syncer_mask;
1506
1507 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1508 mtx_unlock(&sync_mtx);
1509 }
1510
1511 static int
1512 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1513 {
1514 int error, len;
1515
1516 mtx_lock(&sync_mtx);
1517 len = syncer_worklist_len - sync_vnode_count;
1518 mtx_unlock(&sync_mtx);
1519 error = SYSCTL_OUT(req, &len, sizeof(len));
1520 return (error);
1521 }
1522
1523 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1524 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1525
1526 static struct proc *updateproc;
1527 static void sched_sync(void);
1528 static struct kproc_desc up_kp = {
1529 "syncer",
1530 sched_sync,
1531 &updateproc
1532 };
1533 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1534
1535 static int
1536 sync_vnode(struct bufobj *bo, struct thread *td)
1537 {
1538 struct vnode *vp;
1539 struct mount *mp;
1540
1541 vp = bo->__bo_vnode; /* XXX */
1542 if (VOP_ISLOCKED(vp, NULL) != 0)
1543 return (1);
1544 if (VI_TRYLOCK(vp) == 0)
1545 return (1);
1546 /*
1547 * We use vhold in case the vnode does not
1548 * successfully sync. vhold prevents the vnode from
1549 * going away when we unlock the sync_mtx so that
1550 * we can acquire the vnode interlock.
1551 */
1552 vholdl(vp);
1553 mtx_unlock(&sync_mtx);
1554 VI_UNLOCK(vp);
1555 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1556 vdrop(vp);
1557 mtx_lock(&sync_mtx);
1558 return (1);
1559 }
1560 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1561 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1562 VOP_UNLOCK(vp, 0, td);
1563 vn_finished_write(mp);
1564 VI_LOCK(vp);
1565 if ((bo->bo_flag & BO_ONWORKLST) != 0) {
1566 /*
1567 * Put us back on the worklist. The worklist
1568 * routine will remove us from our current
1569 * position and then add us back in at a later
1570 * position.
1571 */
1572 vn_syncer_add_to_worklist(bo, syncdelay);
1573 }
1574 vdropl(vp);
1575 mtx_lock(&sync_mtx);
1576 return (0);
1577 }
1578
1579 /*
1580 * System filesystem synchronizer daemon.
1581 */
1582 static void
1583 sched_sync(void)
1584 {
1585 struct synclist *next;
1586 struct synclist *slp;
1587 struct bufobj *bo;
1588 long starttime;
1589 struct thread *td = FIRST_THREAD_IN_PROC(updateproc);
1590 static int dummychan;
1591 int last_work_seen;
1592 int net_worklist_len;
1593 int syncer_final_iter;
1594 int first_printf;
1595 int error;
1596
1597 mtx_lock(&Giant);
1598 last_work_seen = 0;
1599 syncer_final_iter = 0;
1600 first_printf = 1;
1601 syncer_state = SYNCER_RUNNING;
1602 starttime = time_second;
1603 td->td_pflags |= TDP_NORUNNINGBUF;
1604
1605 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1606 SHUTDOWN_PRI_LAST);
1607
1608 for (;;) {
1609 mtx_lock(&sync_mtx);
1610 if (syncer_state == SYNCER_FINAL_DELAY &&
1611 syncer_final_iter == 0) {
1612 mtx_unlock(&sync_mtx);
1613 kthread_suspend_check(td->td_proc);
1614 mtx_lock(&sync_mtx);
1615 }
1616 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1617 if (syncer_state != SYNCER_RUNNING &&
1618 starttime != time_second) {
1619 if (first_printf) {
1620 printf("\nSyncing disks, vnodes remaining...");
1621 first_printf = 0;
1622 }
1623 printf("%d ", net_worklist_len);
1624 }
1625 starttime = time_second;
1626
1627 /*
1628 * Push files whose dirty time has expired. Be careful
1629 * of interrupt race on slp queue.
1630 *
1631 * Skip over empty worklist slots when shutting down.
1632 */
1633 do {
1634 slp = &syncer_workitem_pending[syncer_delayno];
1635 syncer_delayno += 1;
1636 if (syncer_delayno == syncer_maxdelay)
1637 syncer_delayno = 0;
1638 next = &syncer_workitem_pending[syncer_delayno];
1639 /*
1640 * If the worklist has wrapped since the
1641 * it was emptied of all but syncer vnodes,
1642 * switch to the FINAL_DELAY state and run
1643 * for one more second.
1644 */
1645 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1646 net_worklist_len == 0 &&
1647 last_work_seen == syncer_delayno) {
1648 syncer_state = SYNCER_FINAL_DELAY;
1649 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1650 }
1651 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1652 syncer_worklist_len > 0);
1653
1654 /*
1655 * Keep track of the last time there was anything
1656 * on the worklist other than syncer vnodes.
1657 * Return to the SHUTTING_DOWN state if any
1658 * new work appears.
1659 */
1660 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1661 last_work_seen = syncer_delayno;
1662 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1663 syncer_state = SYNCER_SHUTTING_DOWN;
1664 while ((bo = LIST_FIRST(slp)) != NULL) {
1665 error = sync_vnode(bo, td);
1666 if (error == 1) {
1667 LIST_REMOVE(bo, bo_synclist);
1668 LIST_INSERT_HEAD(next, bo, bo_synclist);
1669 continue;
1670 }
1671 }
1672 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1673 syncer_final_iter--;
1674 mtx_unlock(&sync_mtx);
1675
1676 /*
1677 * Do soft update processing.
1678 */
1679 if (softdep_process_worklist_hook != NULL)
1680 (*softdep_process_worklist_hook)(NULL);
1681
1682 /*
1683 * The variable rushjob allows the kernel to speed up the
1684 * processing of the filesystem syncer process. A rushjob
1685 * value of N tells the filesystem syncer to process the next
1686 * N seconds worth of work on its queue ASAP. Currently rushjob
1687 * is used by the soft update code to speed up the filesystem
1688 * syncer process when the incore state is getting so far
1689 * ahead of the disk that the kernel memory pool is being
1690 * threatened with exhaustion.
1691 */
1692 mtx_lock(&sync_mtx);
1693 if (rushjob > 0) {
1694 rushjob -= 1;
1695 mtx_unlock(&sync_mtx);
1696 continue;
1697 }
1698 mtx_unlock(&sync_mtx);
1699 /*
1700 * Just sleep for a short period if time between
1701 * iterations when shutting down to allow some I/O
1702 * to happen.
1703 *
1704 * If it has taken us less than a second to process the
1705 * current work, then wait. Otherwise start right over
1706 * again. We can still lose time if any single round
1707 * takes more than two seconds, but it does not really
1708 * matter as we are just trying to generally pace the
1709 * filesystem activity.
1710 */
1711 if (syncer_state != SYNCER_RUNNING)
1712 tsleep(&dummychan, PPAUSE, "syncfnl",
1713 hz / SYNCER_SHUTDOWN_SPEEDUP);
1714 else if (time_second == starttime)
1715 tsleep(&lbolt, PPAUSE, "syncer", 0);
1716 }
1717 }
1718
1719 /*
1720 * Request the syncer daemon to speed up its work.
1721 * We never push it to speed up more than half of its
1722 * normal turn time, otherwise it could take over the cpu.
1723 */
1724 int
1725 speedup_syncer()
1726 {
1727 struct thread *td;
1728 int ret = 0;
1729
1730 td = FIRST_THREAD_IN_PROC(updateproc);
1731 sleepq_remove(td, &lbolt);
1732 mtx_lock(&sync_mtx);
1733 if (rushjob < syncdelay / 2) {
1734 rushjob += 1;
1735 stat_rush_requests += 1;
1736 ret = 1;
1737 }
1738 mtx_unlock(&sync_mtx);
1739 return (ret);
1740 }
1741
1742 /*
1743 * Tell the syncer to speed up its work and run though its work
1744 * list several times, then tell it to shut down.
1745 */
1746 static void
1747 syncer_shutdown(void *arg, int howto)
1748 {
1749 struct thread *td;
1750
1751 if (howto & RB_NOSYNC)
1752 return;
1753 td = FIRST_THREAD_IN_PROC(updateproc);
1754 sleepq_remove(td, &lbolt);
1755 mtx_lock(&sync_mtx);
1756 syncer_state = SYNCER_SHUTTING_DOWN;
1757 rushjob = 0;
1758 mtx_unlock(&sync_mtx);
1759 kproc_shutdown(arg, howto);
1760 }
1761
1762 /*
1763 * Reassign a buffer from one vnode to another.
1764 * Used to assign file specific control information
1765 * (indirect blocks) to the vnode to which they belong.
1766 */
1767 void
1768 reassignbuf(struct buf *bp)
1769 {
1770 struct vnode *vp;
1771 struct bufobj *bo;
1772 int delay;
1773 #ifdef INVARIANTS
1774 struct bufv *bv;
1775 #endif
1776
1777 vp = bp->b_vp;
1778 bo = bp->b_bufobj;
1779 ++reassignbufcalls;
1780
1781 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1782 bp, bp->b_vp, bp->b_flags);
1783 /*
1784 * B_PAGING flagged buffers cannot be reassigned because their vp
1785 * is not fully linked in.
1786 */
1787 if (bp->b_flags & B_PAGING)
1788 panic("cannot reassign paging buffer");
1789
1790 /*
1791 * Delete from old vnode list, if on one.
1792 */
1793 VI_LOCK(vp);
1794 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1795 buf_vlist_remove(bp);
1796 else
1797 panic("reassignbuf: Buffer %p not on queue.", bp);
1798 /*
1799 * If dirty, put on list of dirty buffers; otherwise insert onto list
1800 * of clean buffers.
1801 */
1802 if (bp->b_flags & B_DELWRI) {
1803 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1804 switch (vp->v_type) {
1805 case VDIR:
1806 delay = dirdelay;
1807 break;
1808 case VCHR:
1809 delay = metadelay;
1810 break;
1811 default:
1812 delay = filedelay;
1813 }
1814 vn_syncer_add_to_worklist(bo, delay);
1815 }
1816 buf_vlist_add(bp, bo, BX_VNDIRTY);
1817 } else {
1818 buf_vlist_add(bp, bo, BX_VNCLEAN);
1819
1820 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1821 mtx_lock(&sync_mtx);
1822 LIST_REMOVE(bo, bo_synclist);
1823 syncer_worklist_len--;
1824 mtx_unlock(&sync_mtx);
1825 bo->bo_flag &= ~BO_ONWORKLST;
1826 }
1827 }
1828 #ifdef INVARIANTS
1829 bv = &bo->bo_clean;
1830 bp = TAILQ_FIRST(&bv->bv_hd);
1831 KASSERT(bp == NULL || bp->b_bufobj == bo,
1832 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1833 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1834 KASSERT(bp == NULL || bp->b_bufobj == bo,
1835 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1836 bv = &bo->bo_dirty;
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 #endif
1844 VI_UNLOCK(vp);
1845 }
1846
1847 /*
1848 * Increment the use and hold counts on the vnode, taking care to reference
1849 * the driver's usecount if this is a chardev. The vholdl() will remove
1850 * the vnode from the free list if it is presently free. Requires the
1851 * vnode interlock and returns with it held.
1852 */
1853 static void
1854 v_incr_usecount(struct vnode *vp)
1855 {
1856
1857 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1858 vp, vp->v_holdcnt, vp->v_usecount);
1859 vp->v_usecount++;
1860 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1861 dev_lock();
1862 vp->v_rdev->si_usecount++;
1863 dev_unlock();
1864 }
1865 vholdl(vp);
1866 }
1867
1868 /*
1869 * Decrement the vnode use and hold count along with the driver's usecount
1870 * if this is a chardev. The vdropl() below releases the vnode interlock
1871 * as it may free the vnode.
1872 */
1873 static void
1874 v_decr_usecount(struct vnode *vp)
1875 {
1876
1877 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
1878 vp, vp->v_holdcnt, vp->v_usecount);
1879 ASSERT_VI_LOCKED(vp, __FUNCTION__);
1880 VNASSERT(vp->v_usecount > 0, vp,
1881 ("v_decr_usecount: negative usecount"));
1882 vp->v_usecount--;
1883 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1884 dev_lock();
1885 vp->v_rdev->si_usecount--;
1886 dev_unlock();
1887 }
1888 vdropl(vp);
1889 }
1890
1891 /*
1892 * Decrement only the use count and driver use count. This is intended to
1893 * be paired with a follow on vdropl() to release the remaining hold count.
1894 * In this way we may vgone() a vnode with a 0 usecount without risk of
1895 * having it end up on a free list because the hold count is kept above 0.
1896 */
1897 static void
1898 v_decr_useonly(struct vnode *vp)
1899 {
1900
1901 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
1902 vp, vp->v_holdcnt, vp->v_usecount);
1903 ASSERT_VI_LOCKED(vp, __FUNCTION__);
1904 VNASSERT(vp->v_usecount > 0, vp,
1905 ("v_decr_useonly: negative usecount"));
1906 vp->v_usecount--;
1907 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1908 dev_lock();
1909 vp->v_rdev->si_usecount--;
1910 dev_unlock();
1911 }
1912 }
1913
1914 /*
1915 * Grab a particular vnode from the free list, increment its
1916 * reference count and lock it. The vnode lock bit is set if the
1917 * vnode is being eliminated in vgone. The process is awakened
1918 * when the transition is completed, and an error returned to
1919 * indicate that the vnode is no longer usable (possibly having
1920 * been changed to a new filesystem type).
1921 */
1922 int
1923 vget(vp, flags, td)
1924 struct vnode *vp;
1925 int flags;
1926 struct thread *td;
1927 {
1928 int oweinact;
1929 int oldflags;
1930 int error;
1931
1932 error = 0;
1933 oldflags = flags;
1934 oweinact = 0;
1935 if ((flags & LK_INTERLOCK) == 0)
1936 VI_LOCK(vp);
1937 /*
1938 * If the inactive call was deferred because vput() was called
1939 * with a shared lock, we have to do it here before another thread
1940 * gets a reference to data that should be dead.
1941 */
1942 if (vp->v_iflag & VI_OWEINACT) {
1943 if (flags & LK_NOWAIT) {
1944 VI_UNLOCK(vp);
1945 return (EBUSY);
1946 }
1947 flags &= ~LK_TYPE_MASK;
1948 flags |= LK_EXCLUSIVE;
1949 oweinact = 1;
1950 }
1951 v_incr_usecount(vp);
1952 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1953 VI_LOCK(vp);
1954 /*
1955 * must expand vrele here because we do not want
1956 * to call VOP_INACTIVE if the reference count
1957 * drops back to zero since it was never really
1958 * active.
1959 */
1960 v_decr_usecount(vp);
1961 return (error);
1962 }
1963 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
1964 panic("vget: vn_lock failed to return ENOENT\n");
1965 if (oweinact) {
1966 VI_LOCK(vp);
1967 if (vp->v_iflag & VI_OWEINACT)
1968 vinactive(vp, td);
1969 VI_UNLOCK(vp);
1970 if ((oldflags & LK_TYPE_MASK) == 0)
1971 VOP_UNLOCK(vp, 0, td);
1972 }
1973 return (0);
1974 }
1975
1976 /*
1977 * Increase the reference count of a vnode.
1978 */
1979 void
1980 vref(struct vnode *vp)
1981 {
1982
1983 VI_LOCK(vp);
1984 v_incr_usecount(vp);
1985 VI_UNLOCK(vp);
1986 }
1987
1988 /*
1989 * Return reference count of a vnode.
1990 *
1991 * The results of this call are only guaranteed when some mechanism other
1992 * than the VI lock is used to stop other processes from gaining references
1993 * to the vnode. This may be the case if the caller holds the only reference.
1994 * This is also useful when stale data is acceptable as race conditions may
1995 * be accounted for by some other means.
1996 */
1997 int
1998 vrefcnt(struct vnode *vp)
1999 {
2000 int usecnt;
2001
2002 VI_LOCK(vp);
2003 usecnt = vp->v_usecount;
2004 VI_UNLOCK(vp);
2005
2006 return (usecnt);
2007 }
2008
2009
2010 /*
2011 * Vnode put/release.
2012 * If count drops to zero, call inactive routine and return to freelist.
2013 */
2014 void
2015 vrele(vp)
2016 struct vnode *vp;
2017 {
2018 struct thread *td = curthread; /* XXX */
2019
2020 KASSERT(vp != NULL, ("vrele: null vp"));
2021
2022 VI_LOCK(vp);
2023
2024 /* Skip this v_writecount check if we're going to panic below. */
2025 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2026 ("vrele: missed vn_close"));
2027
2028 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2029 vp->v_usecount == 1)) {
2030 v_decr_usecount(vp);
2031 return;
2032 }
2033 if (vp->v_usecount != 1) {
2034 #ifdef DIAGNOSTIC
2035 vprint("vrele: negative ref count", vp);
2036 #endif
2037 VI_UNLOCK(vp);
2038 panic("vrele: negative ref cnt");
2039 }
2040 /*
2041 * We want to hold the vnode until the inactive finishes to
2042 * prevent vgone() races. We drop the use count here and the
2043 * hold count below when we're done.
2044 */
2045 v_decr_useonly(vp);
2046 /*
2047 * We must call VOP_INACTIVE with the node locked. Mark
2048 * as VI_DOINGINACT to avoid recursion.
2049 */
2050 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2051 VI_LOCK(vp);
2052 vinactive(vp, td);
2053 VOP_UNLOCK(vp, 0, td);
2054 } else
2055 VI_LOCK(vp);
2056 vdropl(vp);
2057 }
2058
2059 /*
2060 * Release an already locked vnode. This give the same effects as
2061 * unlock+vrele(), but takes less time and avoids releasing and
2062 * re-aquiring the lock (as vrele() aquires the lock internally.)
2063 */
2064 void
2065 vput(vp)
2066 struct vnode *vp;
2067 {
2068 struct thread *td = curthread; /* XXX */
2069 int error;
2070
2071 KASSERT(vp != NULL, ("vput: null vp"));
2072 ASSERT_VOP_LOCKED(vp, "vput");
2073 VI_LOCK(vp);
2074 /* Skip this v_writecount check if we're going to panic below. */
2075 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2076 ("vput: missed vn_close"));
2077 error = 0;
2078
2079 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2080 vp->v_usecount == 1)) {
2081 VOP_UNLOCK(vp, 0, td);
2082 v_decr_usecount(vp);
2083 return;
2084 }
2085
2086 if (vp->v_usecount != 1) {
2087 #ifdef DIAGNOSTIC
2088 vprint("vput: negative ref count", vp);
2089 #endif
2090 panic("vput: negative ref cnt");
2091 }
2092 /*
2093 * We want to hold the vnode until the inactive finishes to
2094 * prevent vgone() races. We drop the use count here and the
2095 * hold count below when we're done.
2096 */
2097 v_decr_useonly(vp);
2098 vp->v_iflag |= VI_OWEINACT;
2099 if (VOP_ISLOCKED(vp, NULL) != LK_EXCLUSIVE) {
2100 error = VOP_LOCK(vp, LK_EXCLUPGRADE|LK_INTERLOCK|LK_NOWAIT, td);
2101 VI_LOCK(vp);
2102 if (error)
2103 goto done;
2104 }
2105 if (vp->v_iflag & VI_OWEINACT)
2106 vinactive(vp, td);
2107 VOP_UNLOCK(vp, 0, td);
2108 done:
2109 vdropl(vp);
2110 }
2111
2112 /*
2113 * Somebody doesn't want the vnode recycled.
2114 */
2115 void
2116 vhold(struct vnode *vp)
2117 {
2118
2119 VI_LOCK(vp);
2120 vholdl(vp);
2121 VI_UNLOCK(vp);
2122 }
2123
2124 void
2125 vholdl(struct vnode *vp)
2126 {
2127
2128 vp->v_holdcnt++;
2129 if (VSHOULDBUSY(vp))
2130 vbusy(vp);
2131 }
2132
2133 /*
2134 * Note that there is one less who cares about this vnode. vdrop() is the
2135 * opposite of vhold().
2136 */
2137 void
2138 vdrop(struct vnode *vp)
2139 {
2140
2141 VI_LOCK(vp);
2142 vdropl(vp);
2143 }
2144
2145 /*
2146 * Drop the hold count of the vnode. If this is the last reference to
2147 * the vnode we will free it if it has been vgone'd otherwise it is
2148 * placed on the free list.
2149 */
2150 static void
2151 vdropl(struct vnode *vp)
2152 {
2153
2154 if (vp->v_holdcnt <= 0)
2155 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2156 vp->v_holdcnt--;
2157 if (vp->v_holdcnt == 0) {
2158 if (vp->v_iflag & VI_DOOMED) {
2159 vdestroy(vp);
2160 return;
2161 } else
2162 vfree(vp);
2163 }
2164 VI_UNLOCK(vp);
2165 }
2166
2167 /*
2168 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2169 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2170 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2171 * failed lock upgrade.
2172 */
2173 static void
2174 vinactive(struct vnode *vp, struct thread *td)
2175 {
2176
2177 ASSERT_VOP_LOCKED(vp, "vinactive");
2178 ASSERT_VI_LOCKED(vp, "vinactive");
2179 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2180 ("vinactive: recursed on VI_DOINGINACT"));
2181 vp->v_iflag |= VI_DOINGINACT;
2182 vp->v_iflag &= ~VI_OWEINACT;
2183 VI_UNLOCK(vp);
2184 VOP_INACTIVE(vp, td);
2185 VI_LOCK(vp);
2186 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2187 ("vinactive: lost VI_DOINGINACT"));
2188 vp->v_iflag &= ~VI_DOINGINACT;
2189 }
2190
2191 /*
2192 * Remove any vnodes in the vnode table belonging to mount point mp.
2193 *
2194 * If FORCECLOSE is not specified, there should not be any active ones,
2195 * return error if any are found (nb: this is a user error, not a
2196 * system error). If FORCECLOSE is specified, detach any active vnodes
2197 * that are found.
2198 *
2199 * If WRITECLOSE is set, only flush out regular file vnodes open for
2200 * writing.
2201 *
2202 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2203 *
2204 * `rootrefs' specifies the base reference count for the root vnode
2205 * of this filesystem. The root vnode is considered busy if its
2206 * v_usecount exceeds this value. On a successful return, vflush(, td)
2207 * will call vrele() on the root vnode exactly rootrefs times.
2208 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2209 * be zero.
2210 */
2211 #ifdef DIAGNOSTIC
2212 static int busyprt = 0; /* print out busy vnodes */
2213 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2214 #endif
2215
2216 int
2217 vflush(mp, rootrefs, flags, td)
2218 struct mount *mp;
2219 int rootrefs;
2220 int flags;
2221 struct thread *td;
2222 {
2223 struct vnode *vp, *nvp, *rootvp = NULL;
2224 struct vattr vattr;
2225 int busy = 0, error;
2226
2227 CTR1(KTR_VFS, "vflush: mp %p", mp);
2228 if (rootrefs > 0) {
2229 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2230 ("vflush: bad args"));
2231 /*
2232 * Get the filesystem root vnode. We can vput() it
2233 * immediately, since with rootrefs > 0, it won't go away.
2234 */
2235 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2236 return (error);
2237 vput(rootvp);
2238
2239 }
2240 MNT_ILOCK(mp);
2241 loop:
2242 MNT_VNODE_FOREACH(vp, mp, nvp) {
2243
2244 VI_LOCK(vp);
2245 vholdl(vp);
2246 MNT_IUNLOCK(mp);
2247 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2248 if (error) {
2249 vdrop(vp);
2250 MNT_ILOCK(mp);
2251 goto loop;
2252 }
2253 /*
2254 * Skip over a vnodes marked VV_SYSTEM.
2255 */
2256 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2257 VOP_UNLOCK(vp, 0, td);
2258 vdrop(vp);
2259 MNT_ILOCK(mp);
2260 continue;
2261 }
2262 /*
2263 * If WRITECLOSE is set, flush out unlinked but still open
2264 * files (even if open only for reading) and regular file
2265 * vnodes open for writing.
2266 */
2267 if (flags & WRITECLOSE) {
2268 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2269 VI_LOCK(vp);
2270
2271 if ((vp->v_type == VNON ||
2272 (error == 0 && vattr.va_nlink > 0)) &&
2273 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2274 VOP_UNLOCK(vp, 0, td);
2275 vdropl(vp);
2276 MNT_ILOCK(mp);
2277 continue;
2278 }
2279 } else
2280 VI_LOCK(vp);
2281 /*
2282 * With v_usecount == 0, all we need to do is clear out the
2283 * vnode data structures and we are done.
2284 *
2285 * If FORCECLOSE is set, forcibly close the vnode.
2286 */
2287 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2288 VNASSERT(vp->v_usecount == 0 ||
2289 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2290 ("device VNODE %p is FORCECLOSED", vp));
2291 vgonel(vp);
2292 } else {
2293 busy++;
2294 #ifdef DIAGNOSTIC
2295 if (busyprt)
2296 vprint("vflush: busy vnode", vp);
2297 #endif
2298 }
2299 VOP_UNLOCK(vp, 0, td);
2300 vdropl(vp);
2301 MNT_ILOCK(mp);
2302 }
2303 MNT_IUNLOCK(mp);
2304 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2305 /*
2306 * If just the root vnode is busy, and if its refcount
2307 * is equal to `rootrefs', then go ahead and kill it.
2308 */
2309 VI_LOCK(rootvp);
2310 KASSERT(busy > 0, ("vflush: not busy"));
2311 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2312 ("vflush: usecount %d < rootrefs %d",
2313 rootvp->v_usecount, rootrefs));
2314 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2315 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK, td);
2316 vgone(rootvp);
2317 VOP_UNLOCK(rootvp, 0, td);
2318 busy = 0;
2319 } else
2320 VI_UNLOCK(rootvp);
2321 }
2322 if (busy)
2323 return (EBUSY);
2324 for (; rootrefs > 0; rootrefs--)
2325 vrele(rootvp);
2326 return (0);
2327 }
2328
2329 /*
2330 * Recycle an unused vnode to the front of the free list.
2331 */
2332 int
2333 vrecycle(struct vnode *vp, struct thread *td)
2334 {
2335 int recycled;
2336
2337 ASSERT_VOP_LOCKED(vp, "vrecycle");
2338 recycled = 0;
2339 VI_LOCK(vp);
2340 if (vp->v_usecount == 0) {
2341 recycled = 1;
2342 vgonel(vp);
2343 }
2344 VI_UNLOCK(vp);
2345 return (recycled);
2346 }
2347
2348 /*
2349 * Eliminate all activity associated with a vnode
2350 * in preparation for reuse.
2351 */
2352 void
2353 vgone(struct vnode *vp)
2354 {
2355 VI_LOCK(vp);
2356 vgonel(vp);
2357 VI_UNLOCK(vp);
2358 }
2359
2360 /*
2361 * vgone, with the vp interlock held.
2362 */
2363 void
2364 vgonel(struct vnode *vp)
2365 {
2366 struct thread *td;
2367 int oweinact;
2368 int active;
2369
2370 CTR1(KTR_VFS, "vgonel: vp %p", vp);
2371 ASSERT_VOP_LOCKED(vp, "vgonel");
2372 ASSERT_VI_LOCKED(vp, "vgonel");
2373 #if 0
2374 /* XXX Need to fix ttyvp before I enable this. */
2375 VNASSERT(vp->v_holdcnt, vp,
2376 ("vgonel: vp %p has no reference.", vp));
2377 #endif
2378 td = curthread;
2379
2380 /*
2381 * Don't vgonel if we're already doomed.
2382 */
2383 if (vp->v_iflag & VI_DOOMED)
2384 return;
2385 vp->v_iflag |= VI_DOOMED;
2386 /*
2387 * Check to see if the vnode is in use. If so, we have to call
2388 * VOP_CLOSE() and VOP_INACTIVE().
2389 */
2390 active = vp->v_usecount;
2391 oweinact = (vp->v_iflag & VI_OWEINACT);
2392 VI_UNLOCK(vp);
2393 /*
2394 * Clean out any buffers associated with the vnode.
2395 * If the flush fails, just toss the buffers.
2396 */
2397 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2398 (void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2399 if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2400 vinvalbuf(vp, 0, td, 0, 0);
2401
2402 /*
2403 * If purging an active vnode, it must be closed and
2404 * deactivated before being reclaimed.
2405 */
2406 if (active)
2407 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2408 if (oweinact || active) {
2409 VI_LOCK(vp);
2410 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2411 vinactive(vp, td);
2412 VI_UNLOCK(vp);
2413 }
2414 /*
2415 * Reclaim the vnode.
2416 */
2417 if (VOP_RECLAIM(vp, td))
2418 panic("vgone: cannot reclaim");
2419 VNASSERT(vp->v_object == NULL, vp,
2420 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2421 /*
2422 * Delete from old mount point vnode list.
2423 */
2424 delmntque(vp);
2425 cache_purge(vp);
2426 /*
2427 * Done with purge, reset to the standard lock and invalidate
2428 * the vnode.
2429 */
2430 VI_LOCK(vp);
2431 vp->v_vnlock = &vp->v_lock;
2432 vp->v_op = &dead_vnodeops;
2433 vp->v_tag = "none";
2434 vp->v_type = VBAD;
2435 }
2436
2437 /*
2438 * Calculate the total number of references to a special device.
2439 */
2440 int
2441 vcount(vp)
2442 struct vnode *vp;
2443 {
2444 int count;
2445
2446 dev_lock();
2447 count = vp->v_rdev->si_usecount;
2448 dev_unlock();
2449 return (count);
2450 }
2451
2452 /*
2453 * Same as above, but using the struct cdev *as argument
2454 */
2455 int
2456 count_dev(dev)
2457 struct cdev *dev;
2458 {
2459 int count;
2460
2461 dev_lock();
2462 count = dev->si_usecount;
2463 dev_unlock();
2464 return(count);
2465 }
2466
2467 /*
2468 * Print out a description of a vnode.
2469 */
2470 static char *typename[] =
2471 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2472
2473 void
2474 vn_printf(struct vnode *vp, const char *fmt, ...)
2475 {
2476 va_list ap;
2477 char buf[96];
2478
2479 va_start(ap, fmt);
2480 vprintf(fmt, ap);
2481 va_end(ap);
2482 printf("%p: ", (void *)vp);
2483 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2484 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2485 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2486 buf[0] = '\0';
2487 buf[1] = '\0';
2488 if (vp->v_vflag & VV_ROOT)
2489 strcat(buf, "|VV_ROOT");
2490 if (vp->v_vflag & VV_TEXT)
2491 strcat(buf, "|VV_TEXT");
2492 if (vp->v_vflag & VV_SYSTEM)
2493 strcat(buf, "|VV_SYSTEM");
2494 if (vp->v_iflag & VI_DOOMED)
2495 strcat(buf, "|VI_DOOMED");
2496 if (vp->v_iflag & VI_FREE)
2497 strcat(buf, "|VI_FREE");
2498 printf(" flags (%s)\n", buf + 1);
2499 if (mtx_owned(VI_MTX(vp)))
2500 printf(" VI_LOCKed");
2501 if (vp->v_object != NULL)
2502 printf(" v_object %p ref %d pages %d\n",
2503 vp->v_object, vp->v_object->ref_count,
2504 vp->v_object->resident_page_count);
2505 printf(" ");
2506 lockmgr_printinfo(vp->v_vnlock);
2507 printf("\n");
2508 if (vp->v_data != NULL)
2509 VOP_PRINT(vp);
2510 }
2511
2512 #ifdef DDB
2513 #include <ddb/ddb.h>
2514 /*
2515 * List all of the locked vnodes in the system.
2516 * Called when debugging the kernel.
2517 */
2518 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2519 {
2520 struct mount *mp, *nmp;
2521 struct vnode *vp;
2522
2523 /*
2524 * Note: because this is DDB, we can't obey the locking semantics
2525 * for these structures, which means we could catch an inconsistent
2526 * state and dereference a nasty pointer. Not much to be done
2527 * about that.
2528 */
2529 printf("Locked vnodes\n");
2530 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2531 nmp = TAILQ_NEXT(mp, mnt_list);
2532 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2533 if (VOP_ISLOCKED(vp, NULL))
2534 vprint("", vp);
2535 }
2536 nmp = TAILQ_NEXT(mp, mnt_list);
2537 }
2538 }
2539 #endif
2540
2541 /*
2542 * Fill in a struct xvfsconf based on a struct vfsconf.
2543 */
2544 static void
2545 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2546 {
2547
2548 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2549 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2550 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2551 xvfsp->vfc_flags = vfsp->vfc_flags;
2552 /*
2553 * These are unused in userland, we keep them
2554 * to not break binary compatibility.
2555 */
2556 xvfsp->vfc_vfsops = NULL;
2557 xvfsp->vfc_next = NULL;
2558 }
2559
2560 /*
2561 * Top level filesystem related information gathering.
2562 */
2563 static int
2564 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2565 {
2566 struct vfsconf *vfsp;
2567 struct xvfsconf xvfsp;
2568 int error;
2569
2570 error = 0;
2571 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2572 bzero(&xvfsp, sizeof(xvfsp));
2573 vfsconf2x(vfsp, &xvfsp);
2574 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2575 if (error)
2576 break;
2577 }
2578 return (error);
2579 }
2580
2581 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2582 "S,xvfsconf", "List of all configured filesystems");
2583
2584 #ifndef BURN_BRIDGES
2585 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2586
2587 static int
2588 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2589 {
2590 int *name = (int *)arg1 - 1; /* XXX */
2591 u_int namelen = arg2 + 1; /* XXX */
2592 struct vfsconf *vfsp;
2593 struct xvfsconf xvfsp;
2594
2595 printf("WARNING: userland calling deprecated sysctl, "
2596 "please rebuild world\n");
2597
2598 #if 1 || defined(COMPAT_PRELITE2)
2599 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2600 if (namelen == 1)
2601 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2602 #endif
2603
2604 switch (name[1]) {
2605 case VFS_MAXTYPENUM:
2606 if (namelen != 2)
2607 return (ENOTDIR);
2608 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2609 case VFS_CONF:
2610 if (namelen != 3)
2611 return (ENOTDIR); /* overloaded */
2612 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2613 if (vfsp->vfc_typenum == name[2])
2614 break;
2615 if (vfsp == NULL)
2616 return (EOPNOTSUPP);
2617 bzero(&xvfsp, sizeof(xvfsp));
2618 vfsconf2x(vfsp, &xvfsp);
2619 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2620 }
2621 return (EOPNOTSUPP);
2622 }
2623
2624 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2625 vfs_sysctl, "Generic filesystem");
2626
2627 #if 1 || defined(COMPAT_PRELITE2)
2628
2629 static int
2630 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2631 {
2632 int error;
2633 struct vfsconf *vfsp;
2634 struct ovfsconf ovfs;
2635
2636 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2637 bzero(&ovfs, sizeof(ovfs));
2638 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2639 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2640 ovfs.vfc_index = vfsp->vfc_typenum;
2641 ovfs.vfc_refcount = vfsp->vfc_refcount;
2642 ovfs.vfc_flags = vfsp->vfc_flags;
2643 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2644 if (error)
2645 return error;
2646 }
2647 return 0;
2648 }
2649
2650 #endif /* 1 || COMPAT_PRELITE2 */
2651 #endif /* !BURN_BRIDGES */
2652
2653 #define KINFO_VNODESLOP 10
2654 #ifdef notyet
2655 /*
2656 * Dump vnode list (via sysctl).
2657 */
2658 /* ARGSUSED */
2659 static int
2660 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2661 {
2662 struct xvnode *xvn;
2663 struct thread *td = req->td;
2664 struct mount *mp;
2665 struct vnode *vp;
2666 int error, len, n;
2667
2668 /*
2669 * Stale numvnodes access is not fatal here.
2670 */
2671 req->lock = 0;
2672 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2673 if (!req->oldptr)
2674 /* Make an estimate */
2675 return (SYSCTL_OUT(req, 0, len));
2676
2677 error = sysctl_wire_old_buffer(req, 0);
2678 if (error != 0)
2679 return (error);
2680 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2681 n = 0;
2682 mtx_lock(&mountlist_mtx);
2683 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2684 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
2685 continue;
2686 MNT_ILOCK(mp);
2687 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2688 if (n == len)
2689 break;
2690 vref(vp);
2691 xvn[n].xv_size = sizeof *xvn;
2692 xvn[n].xv_vnode = vp;
2693 xvn[n].xv_id = 0; /* XXX compat */
2694 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
2695 XV_COPY(usecount);
2696 XV_COPY(writecount);
2697 XV_COPY(holdcnt);
2698 XV_COPY(mount);
2699 XV_COPY(numoutput);
2700 XV_COPY(type);
2701 #undef XV_COPY
2702 xvn[n].xv_flag = vp->v_vflag;
2703
2704 switch (vp->v_type) {
2705 case VREG:
2706 case VDIR:
2707 case VLNK:
2708 break;
2709 case VBLK:
2710 case VCHR:
2711 if (vp->v_rdev == NULL) {
2712 vrele(vp);
2713 continue;
2714 }
2715 xvn[n].xv_dev = dev2udev(vp->v_rdev);
2716 break;
2717 case VSOCK:
2718 xvn[n].xv_socket = vp->v_socket;
2719 break;
2720 case VFIFO:
2721 xvn[n].xv_fifo = vp->v_fifoinfo;
2722 break;
2723 case VNON:
2724 case VBAD:
2725 default:
2726 /* shouldn't happen? */
2727 vrele(vp);
2728 continue;
2729 }
2730 vrele(vp);
2731 ++n;
2732 }
2733 MNT_IUNLOCK(mp);
2734 mtx_lock(&mountlist_mtx);
2735 vfs_unbusy(mp, td);
2736 if (n == len)
2737 break;
2738 }
2739 mtx_unlock(&mountlist_mtx);
2740
2741 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
2742 free(xvn, M_TEMP);
2743 return (error);
2744 }
2745
2746 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2747 0, 0, sysctl_vnode, "S,xvnode", "");
2748 #endif
2749
2750 /*
2751 * Unmount all filesystems. The list is traversed in reverse order
2752 * of mounting to avoid dependencies.
2753 */
2754 void
2755 vfs_unmountall()
2756 {
2757 struct mount *mp;
2758 struct thread *td;
2759 int error;
2760
2761 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
2762 td = curthread;
2763 /*
2764 * Since this only runs when rebooting, it is not interlocked.
2765 */
2766 while(!TAILQ_EMPTY(&mountlist)) {
2767 mp = TAILQ_LAST(&mountlist, mntlist);
2768 error = dounmount(mp, MNT_FORCE, td);
2769 if (error) {
2770 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2771 /*
2772 * XXX: Due to the way in which we mount the root
2773 * file system off of devfs, devfs will generate a
2774 * "busy" warning when we try to unmount it before
2775 * the root. Don't print a warning as a result in
2776 * order to avoid false positive errors that may
2777 * cause needless upset.
2778 */
2779 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
2780 printf("unmount of %s failed (",
2781 mp->mnt_stat.f_mntonname);
2782 if (error == EBUSY)
2783 printf("BUSY)\n");
2784 else
2785 printf("%d)\n", error);
2786 }
2787 } else {
2788 /* The unmount has removed mp from the mountlist */
2789 }
2790 }
2791 }
2792
2793 /*
2794 * perform msync on all vnodes under a mount point
2795 * the mount point must be locked.
2796 */
2797 void
2798 vfs_msync(struct mount *mp, int flags)
2799 {
2800 struct vnode *vp, *nvp;
2801 struct vm_object *obj;
2802 int tries;
2803
2804 tries = 5;
2805 MNT_ILOCK(mp);
2806 loop:
2807 TAILQ_FOREACH_SAFE(vp, &mp->mnt_nvnodelist, v_nmntvnodes, nvp) {
2808 if (vp->v_mount != mp) {
2809 if (--tries > 0)
2810 goto loop;
2811 break;
2812 }
2813
2814 VI_LOCK(vp);
2815 if ((vp->v_iflag & VI_OBJDIRTY) &&
2816 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2817 MNT_IUNLOCK(mp);
2818 if (!vget(vp,
2819 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
2820 curthread)) {
2821 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
2822 vput(vp);
2823 MNT_ILOCK(mp);
2824 continue;
2825 }
2826
2827 obj = vp->v_object;
2828 if (obj != NULL) {
2829 VM_OBJECT_LOCK(obj);
2830 vm_object_page_clean(obj, 0, 0,
2831 flags == MNT_WAIT ?
2832 OBJPC_SYNC : OBJPC_NOSYNC);
2833 VM_OBJECT_UNLOCK(obj);
2834 }
2835 vput(vp);
2836 }
2837 MNT_ILOCK(mp);
2838 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2839 if (--tries > 0)
2840 goto loop;
2841 break;
2842 }
2843 } else
2844 VI_UNLOCK(vp);
2845 }
2846 MNT_IUNLOCK(mp);
2847 }
2848
2849 /*
2850 * Mark a vnode as free, putting it up for recycling.
2851 */
2852 static void
2853 vfree(struct vnode *vp)
2854 {
2855
2856 CTR1(KTR_VFS, "vfree vp %p", vp);
2857 ASSERT_VI_LOCKED(vp, "vfree");
2858 mtx_lock(&vnode_free_list_mtx);
2859 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
2860 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
2861 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
2862 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
2863 ("vfree: Freeing doomed vnode"));
2864 if (vp->v_iflag & VI_AGE) {
2865 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2866 } else {
2867 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2868 }
2869 freevnodes++;
2870 vp->v_iflag &= ~VI_AGE;
2871 vp->v_iflag |= VI_FREE;
2872 mtx_unlock(&vnode_free_list_mtx);
2873 }
2874
2875 /*
2876 * Opposite of vfree() - mark a vnode as in use.
2877 */
2878 static void
2879 vbusy(struct vnode *vp)
2880 {
2881 CTR1(KTR_VFS, "vbusy vp %p", vp);
2882 ASSERT_VI_LOCKED(vp, "vbusy");
2883 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2884 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
2885
2886 mtx_lock(&vnode_free_list_mtx);
2887 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2888 freevnodes--;
2889 vp->v_iflag &= ~(VI_FREE|VI_AGE);
2890 mtx_unlock(&vnode_free_list_mtx);
2891 }
2892
2893 /*
2894 * Initalize per-vnode helper structure to hold poll-related state.
2895 */
2896 void
2897 v_addpollinfo(struct vnode *vp)
2898 {
2899 struct vpollinfo *vi;
2900
2901 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
2902 if (vp->v_pollinfo != NULL) {
2903 uma_zfree(vnodepoll_zone, vi);
2904 return;
2905 }
2906 vp->v_pollinfo = vi;
2907 mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
2908 knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note, vp, vfs_knllock,
2909 vfs_knlunlock, vfs_knllocked);
2910 }
2911
2912 /*
2913 * Record a process's interest in events which might happen to
2914 * a vnode. Because poll uses the historic select-style interface
2915 * internally, this routine serves as both the ``check for any
2916 * pending events'' and the ``record my interest in future events''
2917 * functions. (These are done together, while the lock is held,
2918 * to avoid race conditions.)
2919 */
2920 int
2921 vn_pollrecord(vp, td, events)
2922 struct vnode *vp;
2923 struct thread *td;
2924 short events;
2925 {
2926
2927 if (vp->v_pollinfo == NULL)
2928 v_addpollinfo(vp);
2929 mtx_lock(&vp->v_pollinfo->vpi_lock);
2930 if (vp->v_pollinfo->vpi_revents & events) {
2931 /*
2932 * This leaves events we are not interested
2933 * in available for the other process which
2934 * which presumably had requested them
2935 * (otherwise they would never have been
2936 * recorded).
2937 */
2938 events &= vp->v_pollinfo->vpi_revents;
2939 vp->v_pollinfo->vpi_revents &= ~events;
2940
2941 mtx_unlock(&vp->v_pollinfo->vpi_lock);
2942 return events;
2943 }
2944 vp->v_pollinfo->vpi_events |= events;
2945 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
2946 mtx_unlock(&vp->v_pollinfo->vpi_lock);
2947 return 0;
2948 }
2949
2950 /*
2951 * Routine to create and manage a filesystem syncer vnode.
2952 */
2953 #define sync_close ((int (*)(struct vop_close_args *))nullop)
2954 static int sync_fsync(struct vop_fsync_args *);
2955 static int sync_inactive(struct vop_inactive_args *);
2956 static int sync_reclaim(struct vop_reclaim_args *);
2957
2958 static struct vop_vector sync_vnodeops = {
2959 .vop_bypass = VOP_EOPNOTSUPP,
2960 .vop_close = sync_close, /* close */
2961 .vop_fsync = sync_fsync, /* fsync */
2962 .vop_inactive = sync_inactive, /* inactive */
2963 .vop_reclaim = sync_reclaim, /* reclaim */
2964 .vop_lock = vop_stdlock, /* lock */
2965 .vop_unlock = vop_stdunlock, /* unlock */
2966 .vop_islocked = vop_stdislocked, /* islocked */
2967 };
2968
2969 /*
2970 * Create a new filesystem syncer vnode for the specified mount point.
2971 */
2972 int
2973 vfs_allocate_syncvnode(mp)
2974 struct mount *mp;
2975 {
2976 struct vnode *vp;
2977 static long start, incr, next;
2978 int error;
2979
2980 /* Allocate a new vnode */
2981 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
2982 mp->mnt_syncer = NULL;
2983 return (error);
2984 }
2985 vp->v_type = VNON;
2986 /*
2987 * Place the vnode onto the syncer worklist. We attempt to
2988 * scatter them about on the list so that they will go off
2989 * at evenly distributed times even if all the filesystems
2990 * are mounted at once.
2991 */
2992 next += incr;
2993 if (next == 0 || next > syncer_maxdelay) {
2994 start /= 2;
2995 incr /= 2;
2996 if (start == 0) {
2997 start = syncer_maxdelay / 2;
2998 incr = syncer_maxdelay;
2999 }
3000 next = start;
3001 }
3002 VI_LOCK(vp);
3003 vn_syncer_add_to_worklist(&vp->v_bufobj,
3004 syncdelay > 0 ? next % syncdelay : 0);
3005 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3006 mtx_lock(&sync_mtx);
3007 sync_vnode_count++;
3008 mtx_unlock(&sync_mtx);
3009 VI_UNLOCK(vp);
3010 mp->mnt_syncer = vp;
3011 return (0);
3012 }
3013
3014 /*
3015 * Do a lazy sync of the filesystem.
3016 */
3017 static int
3018 sync_fsync(ap)
3019 struct vop_fsync_args /* {
3020 struct vnode *a_vp;
3021 struct ucred *a_cred;
3022 int a_waitfor;
3023 struct thread *a_td;
3024 } */ *ap;
3025 {
3026 struct vnode *syncvp = ap->a_vp;
3027 struct mount *mp = syncvp->v_mount;
3028 struct thread *td = ap->a_td;
3029 int error, asyncflag;
3030 struct bufobj *bo;
3031
3032 /*
3033 * We only need to do something if this is a lazy evaluation.
3034 */
3035 if (ap->a_waitfor != MNT_LAZY)
3036 return (0);
3037
3038 /*
3039 * Move ourselves to the back of the sync list.
3040 */
3041 bo = &syncvp->v_bufobj;
3042 BO_LOCK(bo);
3043 vn_syncer_add_to_worklist(bo, syncdelay);
3044 BO_UNLOCK(bo);
3045
3046 /*
3047 * Walk the list of vnodes pushing all that are dirty and
3048 * not already on the sync list.
3049 */
3050 mtx_lock(&mountlist_mtx);
3051 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3052 mtx_unlock(&mountlist_mtx);
3053 return (0);
3054 }
3055 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3056 vfs_unbusy(mp, td);
3057 return (0);
3058 }
3059 asyncflag = mp->mnt_flag & MNT_ASYNC;
3060 mp->mnt_flag &= ~MNT_ASYNC;
3061 vfs_msync(mp, MNT_NOWAIT);
3062 error = VFS_SYNC(mp, MNT_LAZY, td);
3063 if (asyncflag)
3064 mp->mnt_flag |= MNT_ASYNC;
3065 vn_finished_write(mp);
3066 vfs_unbusy(mp, td);
3067 return (error);
3068 }
3069
3070 /*
3071 * The syncer vnode is no referenced.
3072 */
3073 static int
3074 sync_inactive(ap)
3075 struct vop_inactive_args /* {
3076 struct vnode *a_vp;
3077 struct thread *a_td;
3078 } */ *ap;
3079 {
3080
3081 vgone(ap->a_vp);
3082 return (0);
3083 }
3084
3085 /*
3086 * The syncer vnode is no longer needed and is being decommissioned.
3087 *
3088 * Modifications to the worklist must be protected by sync_mtx.
3089 */
3090 static int
3091 sync_reclaim(ap)
3092 struct vop_reclaim_args /* {
3093 struct vnode *a_vp;
3094 } */ *ap;
3095 {
3096 struct vnode *vp = ap->a_vp;
3097 struct bufobj *bo;
3098
3099 VI_LOCK(vp);
3100 bo = &vp->v_bufobj;
3101 vp->v_mount->mnt_syncer = NULL;
3102 if (bo->bo_flag & BO_ONWORKLST) {
3103 mtx_lock(&sync_mtx);
3104 LIST_REMOVE(bo, bo_synclist);
3105 syncer_worklist_len--;
3106 sync_vnode_count--;
3107 mtx_unlock(&sync_mtx);
3108 bo->bo_flag &= ~BO_ONWORKLST;
3109 }
3110 VI_UNLOCK(vp);
3111
3112 return (0);
3113 }
3114
3115 /*
3116 * Check if vnode represents a disk device
3117 */
3118 int
3119 vn_isdisk(vp, errp)
3120 struct vnode *vp;
3121 int *errp;
3122 {
3123 int error;
3124
3125 error = 0;
3126 dev_lock();
3127 if (vp->v_type != VCHR)
3128 error = ENOTBLK;
3129 else if (vp->v_rdev == NULL)
3130 error = ENXIO;
3131 else if (vp->v_rdev->si_devsw == NULL)
3132 error = ENXIO;
3133 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3134 error = ENOTBLK;
3135 dev_unlock();
3136 if (errp != NULL)
3137 *errp = error;
3138 return (error == 0);
3139 }
3140
3141 /*
3142 * Common filesystem object access control check routine. Accepts a
3143 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3144 * and optional call-by-reference privused argument allowing vaccess()
3145 * to indicate to the caller whether privilege was used to satisfy the
3146 * request (obsoleted). Returns 0 on success, or an errno on failure.
3147 */
3148 int
3149 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3150 enum vtype type;
3151 mode_t file_mode;
3152 uid_t file_uid;
3153 gid_t file_gid;
3154 mode_t acc_mode;
3155 struct ucred *cred;
3156 int *privused;
3157 {
3158 mode_t dac_granted;
3159 #ifdef CAPABILITIES
3160 mode_t cap_granted;
3161 #endif
3162
3163 /*
3164 * Look for a normal, non-privileged way to access the file/directory
3165 * as requested. If it exists, go with that.
3166 */
3167
3168 if (privused != NULL)
3169 *privused = 0;
3170
3171 dac_granted = 0;
3172
3173 /* Check the owner. */
3174 if (cred->cr_uid == file_uid) {
3175 dac_granted |= VADMIN;
3176 if (file_mode & S_IXUSR)
3177 dac_granted |= VEXEC;
3178 if (file_mode & S_IRUSR)
3179 dac_granted |= VREAD;
3180 if (file_mode & S_IWUSR)
3181 dac_granted |= (VWRITE | VAPPEND);
3182
3183 if ((acc_mode & dac_granted) == acc_mode)
3184 return (0);
3185
3186 goto privcheck;
3187 }
3188
3189 /* Otherwise, check the groups (first match) */
3190 if (groupmember(file_gid, cred)) {
3191 if (file_mode & S_IXGRP)
3192 dac_granted |= VEXEC;
3193 if (file_mode & S_IRGRP)
3194 dac_granted |= VREAD;
3195 if (file_mode & S_IWGRP)
3196 dac_granted |= (VWRITE | VAPPEND);
3197
3198 if ((acc_mode & dac_granted) == acc_mode)
3199 return (0);
3200
3201 goto privcheck;
3202 }
3203
3204 /* Otherwise, check everyone else. */
3205 if (file_mode & S_IXOTH)
3206 dac_granted |= VEXEC;
3207 if (file_mode & S_IROTH)
3208 dac_granted |= VREAD;
3209 if (file_mode & S_IWOTH)
3210 dac_granted |= (VWRITE | VAPPEND);
3211 if ((acc_mode & dac_granted) == acc_mode)
3212 return (0);
3213
3214 privcheck:
3215 if (!suser_cred(cred, SUSER_ALLOWJAIL)) {
3216 /* XXX audit: privilege used */
3217 if (privused != NULL)
3218 *privused = 1;
3219 return (0);
3220 }
3221
3222 #ifdef CAPABILITIES
3223 /*
3224 * Build a capability mask to determine if the set of capabilities
3225 * satisfies the requirements when combined with the granted mask
3226 * from above.
3227 * For each capability, if the capability is required, bitwise
3228 * or the request type onto the cap_granted mask.
3229 */
3230 cap_granted = 0;
3231
3232 if (type == VDIR) {
3233 /*
3234 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3235 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3236 */
3237 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3238 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, SUSER_ALLOWJAIL))
3239 cap_granted |= VEXEC;
3240 } else {
3241 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3242 !cap_check(cred, NULL, CAP_DAC_EXECUTE, SUSER_ALLOWJAIL))
3243 cap_granted |= VEXEC;
3244 }
3245
3246 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3247 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, SUSER_ALLOWJAIL))
3248 cap_granted |= VREAD;
3249
3250 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3251 !cap_check(cred, NULL, CAP_DAC_WRITE, SUSER_ALLOWJAIL))
3252 cap_granted |= (VWRITE | VAPPEND);
3253
3254 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3255 !cap_check(cred, NULL, CAP_FOWNER, SUSER_ALLOWJAIL))
3256 cap_granted |= VADMIN;
3257
3258 if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3259 /* XXX audit: privilege used */
3260 if (privused != NULL)
3261 *privused = 1;
3262 return (0);
3263 }
3264 #endif
3265
3266 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3267 }
3268
3269 /*
3270 * Credential check based on process requesting service, and per-attribute
3271 * permissions.
3272 */
3273 int
3274 extattr_check_cred(struct vnode *vp, int attrnamespace,
3275 struct ucred *cred, struct thread *td, int access)
3276 {
3277
3278 /*
3279 * Kernel-invoked always succeeds.
3280 */
3281 if (cred == NOCRED)
3282 return (0);
3283
3284 /*
3285 * Do not allow privileged processes in jail to directly
3286 * manipulate system attributes.
3287 *
3288 * XXX What capability should apply here?
3289 * Probably CAP_SYS_SETFFLAG.
3290 */
3291 switch (attrnamespace) {
3292 case EXTATTR_NAMESPACE_SYSTEM:
3293 /* Potentially should be: return (EPERM); */
3294 return (suser_cred(cred, 0));
3295 case EXTATTR_NAMESPACE_USER:
3296 return (VOP_ACCESS(vp, access, cred, td));
3297 default:
3298 return (EPERM);
3299 }
3300 }
3301
3302 #ifdef DEBUG_VFS_LOCKS
3303 /*
3304 * This only exists to supress warnings from unlocked specfs accesses. It is
3305 * no longer ok to have an unlocked VFS.
3306 */
3307 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3308
3309 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3310 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3311
3312 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3313 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3314
3315 int vfs_badlock_print = 1; /* Print lock violations. */
3316 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3317
3318 #ifdef KDB
3319 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3320 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3321 #endif
3322
3323 static void
3324 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3325 {
3326
3327 #ifdef KDB
3328 if (vfs_badlock_backtrace)
3329 kdb_backtrace();
3330 #endif
3331 if (vfs_badlock_print)
3332 printf("%s: %p %s\n", str, (void *)vp, msg);
3333 if (vfs_badlock_ddb)
3334 kdb_enter("lock violation");
3335 }
3336
3337 void
3338 assert_vi_locked(struct vnode *vp, const char *str)
3339 {
3340
3341 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3342 vfs_badlock("interlock is not locked but should be", str, vp);
3343 }
3344
3345 void
3346 assert_vi_unlocked(struct vnode *vp, const char *str)
3347 {
3348
3349 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3350 vfs_badlock("interlock is locked but should not be", str, vp);
3351 }
3352
3353 void
3354 assert_vop_locked(struct vnode *vp, const char *str)
3355 {
3356
3357 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
3358 vfs_badlock("is not locked but should be", str, vp);
3359 }
3360
3361 void
3362 assert_vop_unlocked(struct vnode *vp, const char *str)
3363 {
3364
3365 if (vp && !IGNORE_LOCK(vp) &&
3366 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
3367 vfs_badlock("is locked but should not be", str, vp);
3368 }
3369
3370 void
3371 assert_vop_elocked(struct vnode *vp, const char *str)
3372 {
3373
3374 if (vp && !IGNORE_LOCK(vp) &&
3375 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
3376 vfs_badlock("is not exclusive locked but should be", str, vp);
3377 }
3378
3379 #if 0
3380 void
3381 assert_vop_elocked_other(struct vnode *vp, const char *str)
3382 {
3383
3384 if (vp && !IGNORE_LOCK(vp) &&
3385 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
3386 vfs_badlock("is not exclusive locked by another thread",
3387 str, vp);
3388 }
3389
3390 void
3391 assert_vop_slocked(struct vnode *vp, const char *str)
3392 {
3393
3394 if (vp && !IGNORE_LOCK(vp) &&
3395 VOP_ISLOCKED(vp, curthread) != LK_SHARED)
3396 vfs_badlock("is not locked shared but should be", str, vp);
3397 }
3398 #endif /* 0 */
3399 #endif /* DEBUG_VFS_LOCKS */
3400
3401 void
3402 vop_rename_pre(void *ap)
3403 {
3404 struct vop_rename_args *a = ap;
3405
3406 #ifdef DEBUG_VFS_LOCKS
3407 if (a->a_tvp)
3408 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3409 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3410 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3411 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3412
3413 /* Check the source (from). */
3414 if (a->a_tdvp != a->a_fdvp)
3415 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3416 if (a->a_tvp != a->a_fvp)
3417 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked");
3418
3419 /* Check the target. */
3420 if (a->a_tvp)
3421 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3422 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3423 #endif
3424 if (a->a_tdvp != a->a_fdvp)
3425 vhold(a->a_fdvp);
3426 if (a->a_tvp != a->a_fvp)
3427 vhold(a->a_fvp);
3428 vhold(a->a_tdvp);
3429 if (a->a_tvp)
3430 vhold(a->a_tvp);
3431 }
3432
3433 void
3434 vop_strategy_pre(void *ap)
3435 {
3436 #ifdef DEBUG_VFS_LOCKS
3437 struct vop_strategy_args *a;
3438 struct buf *bp;
3439
3440 a = ap;
3441 bp = a->a_bp;
3442
3443 /*
3444 * Cluster ops lock their component buffers but not the IO container.
3445 */
3446 if ((bp->b_flags & B_CLUSTER) != 0)
3447 return;
3448
3449 if (BUF_REFCNT(bp) < 1) {
3450 if (vfs_badlock_print)
3451 printf(
3452 "VOP_STRATEGY: bp is not locked but should be\n");
3453 if (vfs_badlock_ddb)
3454 kdb_enter("lock violation");
3455 }
3456 #endif
3457 }
3458
3459 void
3460 vop_lookup_pre(void *ap)
3461 {
3462 #ifdef DEBUG_VFS_LOCKS
3463 struct vop_lookup_args *a;
3464 struct vnode *dvp;
3465
3466 a = ap;
3467 dvp = a->a_dvp;
3468 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3469 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3470 #endif
3471 }
3472
3473 void
3474 vop_lookup_post(void *ap, int rc)
3475 {
3476 #ifdef DEBUG_VFS_LOCKS
3477 struct vop_lookup_args *a;
3478 struct vnode *dvp;
3479 struct vnode *vp;
3480
3481 a = ap;
3482 dvp = a->a_dvp;
3483 vp = *(a->a_vpp);
3484
3485 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3486 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3487
3488 if (!rc)
3489 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3490 #endif
3491 }
3492
3493 void
3494 vop_lock_pre(void *ap)
3495 {
3496 #ifdef DEBUG_VFS_LOCKS
3497 struct vop_lock_args *a = ap;
3498
3499 if ((a->a_flags & LK_INTERLOCK) == 0)
3500 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3501 else
3502 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3503 #endif
3504 }
3505
3506 void
3507 vop_lock_post(void *ap, int rc)
3508 {
3509 #ifdef DEBUG_VFS_LOCKS
3510 struct vop_lock_args *a = ap;
3511
3512 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3513 if (rc == 0)
3514 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3515 #endif
3516 }
3517
3518 void
3519 vop_unlock_pre(void *ap)
3520 {
3521 #ifdef DEBUG_VFS_LOCKS
3522 struct vop_unlock_args *a = ap;
3523
3524 if (a->a_flags & LK_INTERLOCK)
3525 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3526 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3527 #endif
3528 }
3529
3530 void
3531 vop_unlock_post(void *ap, int rc)
3532 {
3533 #ifdef DEBUG_VFS_LOCKS
3534 struct vop_unlock_args *a = ap;
3535
3536 if (a->a_flags & LK_INTERLOCK)
3537 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3538 #endif
3539 }
3540
3541 void
3542 vop_create_post(void *ap, int rc)
3543 {
3544 struct vop_create_args *a = ap;
3545
3546 if (!rc)
3547 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3548 }
3549
3550 void
3551 vop_link_post(void *ap, int rc)
3552 {
3553 struct vop_link_args *a = ap;
3554
3555 if (!rc) {
3556 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3557 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3558 }
3559 }
3560
3561 void
3562 vop_mkdir_post(void *ap, int rc)
3563 {
3564 struct vop_mkdir_args *a = ap;
3565
3566 if (!rc)
3567 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3568 }
3569
3570 void
3571 vop_mknod_post(void *ap, int rc)
3572 {
3573 struct vop_mknod_args *a = ap;
3574
3575 if (!rc)
3576 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3577 }
3578
3579 void
3580 vop_remove_post(void *ap, int rc)
3581 {
3582 struct vop_remove_args *a = ap;
3583
3584 if (!rc) {
3585 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3586 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3587 }
3588 }
3589
3590 void
3591 vop_rename_post(void *ap, int rc)
3592 {
3593 struct vop_rename_args *a = ap;
3594
3595 if (!rc) {
3596 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3597 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3598 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3599 if (a->a_tvp)
3600 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3601 }
3602 if (a->a_tdvp != a->a_fdvp)
3603 vdrop(a->a_fdvp);
3604 if (a->a_tvp != a->a_fvp)
3605 vdrop(a->a_fvp);
3606 vdrop(a->a_tdvp);
3607 if (a->a_tvp)
3608 vdrop(a->a_tvp);
3609 }
3610
3611 void
3612 vop_rmdir_post(void *ap, int rc)
3613 {
3614 struct vop_rmdir_args *a = ap;
3615
3616 if (!rc) {
3617 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3618 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3619 }
3620 }
3621
3622 void
3623 vop_setattr_post(void *ap, int rc)
3624 {
3625 struct vop_setattr_args *a = ap;
3626
3627 if (!rc)
3628 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3629 }
3630
3631 void
3632 vop_symlink_post(void *ap, int rc)
3633 {
3634 struct vop_symlink_args *a = ap;
3635
3636 if (!rc)
3637 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3638 }
3639
3640 static struct knlist fs_knlist;
3641
3642 static void
3643 vfs_event_init(void *arg)
3644 {
3645 knlist_init(&fs_knlist, NULL, NULL, NULL, NULL);
3646 }
3647 /* XXX - correct order? */
3648 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3649
3650 void
3651 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3652 {
3653
3654 KNOTE_UNLOCKED(&fs_knlist, event);
3655 }
3656
3657 static int filt_fsattach(struct knote *kn);
3658 static void filt_fsdetach(struct knote *kn);
3659 static int filt_fsevent(struct knote *kn, long hint);
3660
3661 struct filterops fs_filtops =
3662 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3663
3664 static int
3665 filt_fsattach(struct knote *kn)
3666 {
3667
3668 kn->kn_flags |= EV_CLEAR;
3669 knlist_add(&fs_knlist, kn, 0);
3670 return (0);
3671 }
3672
3673 static void
3674 filt_fsdetach(struct knote *kn)
3675 {
3676
3677 knlist_remove(&fs_knlist, kn, 0);
3678 }
3679
3680 static int
3681 filt_fsevent(struct knote *kn, long hint)
3682 {
3683
3684 kn->kn_fflags |= hint;
3685 return (kn->kn_fflags != 0);
3686 }
3687
3688 static int
3689 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
3690 {
3691 struct vfsidctl vc;
3692 int error;
3693 struct mount *mp;
3694
3695 error = SYSCTL_IN(req, &vc, sizeof(vc));
3696 if (error)
3697 return (error);
3698 if (vc.vc_vers != VFS_CTL_VERS1)
3699 return (EINVAL);
3700 mp = vfs_getvfs(&vc.vc_fsid);
3701 if (mp == NULL)
3702 return (ENOENT);
3703 /* ensure that a specific sysctl goes to the right filesystem. */
3704 if (strcmp(vc.vc_fstypename, "*") != 0 &&
3705 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
3706 return (EINVAL);
3707 }
3708 VCTLTOREQ(&vc, req);
3709 return (VFS_SYSCTL(mp, vc.vc_op, req));
3710 }
3711
3712 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR,
3713 NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid");
3714
3715 /*
3716 * Function to initialize a va_filerev field sensibly.
3717 * XXX: Wouldn't a random number make a lot more sense ??
3718 */
3719 u_quad_t
3720 init_va_filerev(void)
3721 {
3722 struct bintime bt;
3723
3724 getbinuptime(&bt);
3725 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
3726 }
3727
3728 static int filt_vfsread(struct knote *kn, long hint);
3729 static int filt_vfswrite(struct knote *kn, long hint);
3730 static int filt_vfsvnode(struct knote *kn, long hint);
3731 static void filt_vfsdetach(struct knote *kn);
3732 static struct filterops vfsread_filtops =
3733 { 1, NULL, filt_vfsdetach, filt_vfsread };
3734 static struct filterops vfswrite_filtops =
3735 { 1, NULL, filt_vfsdetach, filt_vfswrite };
3736 static struct filterops vfsvnode_filtops =
3737 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
3738
3739 static void
3740 vfs_knllock(void *arg)
3741 {
3742 struct vnode *vp = arg;
3743
3744 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
3745 }
3746
3747 static void
3748 vfs_knlunlock(void *arg)
3749 {
3750 struct vnode *vp = arg;
3751
3752 VOP_UNLOCK(vp, 0, curthread);
3753 }
3754
3755 static int
3756 vfs_knllocked(void *arg)
3757 {
3758 struct vnode *vp = arg;
3759
3760 return (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE);
3761 }
3762
3763 int
3764 vfs_kqfilter(struct vop_kqfilter_args *ap)
3765 {
3766 struct vnode *vp = ap->a_vp;
3767 struct knote *kn = ap->a_kn;
3768 struct knlist *knl;
3769
3770 switch (kn->kn_filter) {
3771 case EVFILT_READ:
3772 kn->kn_fop = &vfsread_filtops;
3773 break;
3774 case EVFILT_WRITE:
3775 kn->kn_fop = &vfswrite_filtops;
3776 break;
3777 case EVFILT_VNODE:
3778 kn->kn_fop = &vfsvnode_filtops;
3779 break;
3780 default:
3781 return (EINVAL);
3782 }
3783
3784 kn->kn_hook = (caddr_t)vp;
3785
3786 if (vp->v_pollinfo == NULL)
3787 v_addpollinfo(vp);
3788 if (vp->v_pollinfo == NULL)
3789 return (ENOMEM);
3790 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
3791 knlist_add(knl, kn, 0);
3792
3793 return (0);
3794 }
3795
3796 /*
3797 * Detach knote from vnode
3798 */
3799 static void
3800 filt_vfsdetach(struct knote *kn)
3801 {
3802 struct vnode *vp = (struct vnode *)kn->kn_hook;
3803
3804 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
3805 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
3806 }
3807
3808 /*ARGSUSED*/
3809 static int
3810 filt_vfsread(struct knote *kn, long hint)
3811 {
3812 struct vnode *vp = (struct vnode *)kn->kn_hook;
3813 struct vattr va;
3814
3815 /*
3816 * filesystem is gone, so set the EOF flag and schedule
3817 * the knote for deletion.
3818 */
3819 if (hint == NOTE_REVOKE) {
3820 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
3821 return (1);
3822 }
3823
3824 if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
3825 return (0);
3826
3827 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
3828 return (kn->kn_data != 0);
3829 }
3830
3831 /*ARGSUSED*/
3832 static int
3833 filt_vfswrite(struct knote *kn, long hint)
3834 {
3835 /*
3836 * filesystem is gone, so set the EOF flag and schedule
3837 * the knote for deletion.
3838 */
3839 if (hint == NOTE_REVOKE)
3840 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
3841
3842 kn->kn_data = 0;
3843 return (1);
3844 }
3845
3846 static int
3847 filt_vfsvnode(struct knote *kn, long hint)
3848 {
3849 if (kn->kn_sfflags & hint)
3850 kn->kn_fflags |= hint;
3851 if (hint == NOTE_REVOKE) {
3852 kn->kn_flags |= EV_EOF;
3853 return (1);
3854 }
3855 return (kn->kn_fflags != 0);
3856 }
3857
3858 int
3859 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
3860 {
3861 int error;
3862
3863 if (dp->d_reclen > ap->a_uio->uio_resid)
3864 return (ENAMETOOLONG);
3865 error = uiomove(dp, dp->d_reclen, ap->a_uio);
3866 if (error) {
3867 if (ap->a_ncookies != NULL) {
3868 if (ap->a_cookies != NULL)
3869 free(ap->a_cookies, M_TEMP);
3870 ap->a_cookies = NULL;
3871 *ap->a_ncookies = 0;
3872 }
3873 return (error);
3874 }
3875 if (ap->a_ncookies == NULL)
3876 return (0);
3877 *ap->a_cookies = realloc(*ap->a_cookies,
3878 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
3879 (*ap->a_cookies)[*ap->a_ncookies] = off;
3880 return (0);
3881 }
3882
Cache object: 35e0c3826d5a6a70b79711d04cf68577
|