FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_subr.c
1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
35 */
36
37 /*
38 * External virtual filesystem routines
39 */
40
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD: releng/7.4/sys/kern/vfs_subr.c 213652 2010-10-09 09:24:16Z avg $");
43
44 #include "opt_ddb.h"
45 #include "opt_mac.h"
46
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/dirent.h>
53 #include <sys/event.h>
54 #include <sys/eventhandler.h>
55 #include <sys/extattr.h>
56 #include <sys/file.h>
57 #include <sys/fcntl.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/lockf.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/namei.h>
66 #include <sys/priv.h>
67 #include <sys/reboot.h>
68 #include <sys/sleepqueue.h>
69 #include <sys/stat.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/vmmeter.h>
73 #include <sys/vnode.h>
74
75 #include <machine/stdarg.h>
76
77 #include <security/mac/mac_framework.h>
78
79 #include <vm/vm.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_kern.h>
86 #include <vm/uma.h>
87
88 #ifdef DDB
89 #include <ddb/ddb.h>
90 #endif
91
92 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
93
94 static void delmntque(struct vnode *vp);
95 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
96 int slpflag, int slptimeo);
97 static void syncer_shutdown(void *arg, int howto);
98 static int vtryrecycle(struct vnode *vp);
99 static void vbusy(struct vnode *vp);
100 static void vinactive(struct vnode *, struct thread *);
101 static void v_incr_usecount(struct vnode *);
102 static void v_decr_usecount(struct vnode *);
103 static void v_decr_useonly(struct vnode *);
104 static void v_upgrade_usecount(struct vnode *);
105 static void vfree(struct vnode *);
106 static void vnlru_free(int);
107 static void vgonel(struct vnode *);
108 static void vfs_knllock(void *arg);
109 static void vfs_knlunlock(void *arg);
110 static void destroy_vpollinfo(struct vpollinfo *vi);
111
112 /*
113 * Enable Giant pushdown based on whether or not the vm is mpsafe in this
114 * build. Without mpsafevm the buffer cache can not run Giant free.
115 */
116 int mpsafe_vfs = 1;
117 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
118 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
119 "MPSAFE VFS");
120
121 /*
122 * Number of vnodes in existence. Increased whenever getnewvnode()
123 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
124 * vnode.
125 */
126 static unsigned long numvnodes;
127
128 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
129
130 /*
131 * Conversion tables for conversion from vnode types to inode formats
132 * and back.
133 */
134 enum vtype iftovt_tab[16] = {
135 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
137 };
138 int vttoif_tab[10] = {
139 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
141 };
142
143 /*
144 * List of vnodes that are ready for recycling.
145 */
146 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
147
148 /*
149 * Free vnode target. Free vnodes may simply be files which have been stat'd
150 * but not read. This is somewhat common, and a small cache of such files
151 * should be kept to avoid recreation costs.
152 */
153 static u_long wantfreevnodes;
154 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
155 /* Number of vnodes in the free list. */
156 static u_long freevnodes;
157 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
158
159 /*
160 * Various variables used for debugging the new implementation of
161 * reassignbuf().
162 * XXX these are probably of (very) limited utility now.
163 */
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
166
167 /*
168 * Cache for the mount type id assigned to NFS. This is used for
169 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
170 */
171 int nfs_mount_type = -1;
172
173 /* To keep more than one thread at a time from running vfs_getnewfsid */
174 static struct mtx mntid_mtx;
175
176 /*
177 * Lock for any access to the following:
178 * vnode_free_list
179 * numvnodes
180 * freevnodes
181 */
182 static struct mtx vnode_free_list_mtx;
183
184 /* Publicly exported FS */
185 struct nfs_public nfs_pub;
186
187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
188 static uma_zone_t vnode_zone;
189 static uma_zone_t vnodepoll_zone;
190
191 /* Set to 1 to print out reclaim of active vnodes */
192 int prtactive;
193
194 /*
195 * The workitem queue.
196 *
197 * It is useful to delay writes of file data and filesystem metadata
198 * for tens of seconds so that quickly created and deleted files need
199 * not waste disk bandwidth being created and removed. To realize this,
200 * we append vnodes to a "workitem" queue. When running with a soft
201 * updates implementation, most pending metadata dependencies should
202 * not wait for more than a few seconds. Thus, mounted on block devices
203 * are delayed only about a half the time that file data is delayed.
204 * Similarly, directory updates are more critical, so are only delayed
205 * about a third the time that file data is delayed. Thus, there are
206 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207 * one each second (driven off the filesystem syncer process). The
208 * syncer_delayno variable indicates the next queue that is to be processed.
209 * Items that need to be processed soon are placed in this queue:
210 *
211 * syncer_workitem_pending[syncer_delayno]
212 *
213 * A delay of fifteen seconds is done by placing the request fifteen
214 * entries later in the queue:
215 *
216 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
217 *
218 */
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending;
223 /*
224 * The sync_mtx protects:
225 * bo->bo_synclist
226 * sync_vnode_count
227 * syncer_delayno
228 * syncer_state
229 * syncer_workitem_pending
230 * syncer_worklist_len
231 * rushjob
232 */
233 static struct mtx sync_mtx;
234
235 #define SYNCER_MAXDELAY 32
236 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
237 static int syncdelay = 30; /* max time to delay syncing data */
238 static int filedelay = 30; /* time to delay syncing files */
239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
240 static int dirdelay = 29; /* time to delay syncing directories */
241 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
242 static int metadelay = 28; /* time to delay syncing metadata */
243 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
244 static int rushjob; /* number of slots to run ASAP */
245 static int stat_rush_requests; /* number of times I/O speeded up */
246 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
247
248 /*
249 * When shutting down the syncer, run it at four times normal speed.
250 */
251 #define SYNCER_SHUTDOWN_SPEEDUP 4
252 static int sync_vnode_count;
253 static int syncer_worklist_len;
254 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
255 syncer_state;
256
257 /*
258 * Number of vnodes we want to exist at any one time. This is mostly used
259 * to size hash tables in vnode-related code. It is normally not used in
260 * getnewvnode(), as wantfreevnodes is normally nonzero.)
261 *
262 * XXX desiredvnodes is historical cruft and should not exist.
263 */
264 int desiredvnodes;
265 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
266 &desiredvnodes, 0, "Maximum number of vnodes");
267 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
268 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
269 static int vnlru_nowhere;
270 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
271 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
272
273 /*
274 * Macros to control when a vnode is freed and recycled. All require
275 * the vnode interlock.
276 */
277 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
278 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
280
281
282 /*
283 * Initialize the vnode management data structures.
284 *
285 * Reevaluate the following cap on the number of vnodes after the physical
286 * memory size exceeds 512GB. In the limit, as the physical memory size
287 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
288 */
289 #ifndef MAXVNODES_MAX
290 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
291 #endif
292 static void
293 vntblinit(void *dummy __unused)
294 {
295 int physvnodes, virtvnodes;
296
297 /*
298 * Desiredvnodes is a function of the physical memory size and the
299 * kernel's heap size. Generally speaking, it scales with the
300 * physical memory size. The ratio of desiredvnodes to physical pages
301 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
302 * marginal ratio of desiredvnodes to physical pages is one to
303 * sixteen. However, desiredvnodes is limited by the kernel's heap
304 * size. The memory required by desiredvnodes vnodes and vm objects
305 * may not exceed one seventh of the kernel's heap size.
306 */
307 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
308 cnt.v_page_count) / 16;
309 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
310 sizeof(struct vnode)));
311 desiredvnodes = min(physvnodes, virtvnodes);
312 if (desiredvnodes > MAXVNODES_MAX) {
313 if (bootverbose)
314 printf("Reducing kern.maxvnodes %d -> %d\n",
315 desiredvnodes, MAXVNODES_MAX);
316 desiredvnodes = MAXVNODES_MAX;
317 }
318 wantfreevnodes = desiredvnodes / 4;
319 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
320 TAILQ_INIT(&vnode_free_list);
321 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
322 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
323 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
324 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
325 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
326 /*
327 * Initialize the filesystem syncer.
328 */
329 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
330 &syncer_mask);
331 syncer_maxdelay = syncer_mask + 1;
332 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
333 }
334 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
335
336
337 /*
338 * Mark a mount point as busy. Used to synchronize access and to delay
339 * unmounting. Interlock is not released on failure.
340 */
341 int
342 vfs_busy(struct mount *mp, int flags, struct mtx *interlkp,
343 struct thread *td)
344 {
345 int lkflags;
346
347 MNT_ILOCK(mp);
348 MNT_REF(mp);
349 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
350 if (flags & LK_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
351 MNT_REL(mp);
352 MNT_IUNLOCK(mp);
353 return (ENOENT);
354 }
355 if (interlkp)
356 mtx_unlock(interlkp);
357 mp->mnt_kern_flag |= MNTK_MWAIT;
358 /*
359 * Since all busy locks are shared except the exclusive
360 * lock granted when unmounting, the only place that a
361 * wakeup needs to be done is at the release of the
362 * exclusive lock at the end of dounmount.
363 */
364 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
365 if (interlkp)
366 mtx_lock(interlkp);
367 }
368 if (interlkp)
369 mtx_unlock(interlkp);
370 lkflags = LK_SHARED | LK_INTERLOCK | LK_NOWAIT;
371 if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp), td))
372 panic("vfs_busy: unexpected lock failure");
373 return (0);
374 }
375
376 /*
377 * Free a busy filesystem.
378 */
379 void
380 vfs_unbusy(struct mount *mp, struct thread *td)
381 {
382
383 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
384 vfs_rel(mp);
385 }
386
387 /*
388 * Lookup a mount point by filesystem identifier.
389 */
390 struct mount *
391 vfs_getvfs(fsid_t *fsid)
392 {
393 struct mount *mp;
394
395 mtx_lock(&mountlist_mtx);
396 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
397 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
398 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
399 vfs_ref(mp);
400 mtx_unlock(&mountlist_mtx);
401 return (mp);
402 }
403 }
404 mtx_unlock(&mountlist_mtx);
405 return ((struct mount *) 0);
406 }
407
408 /*
409 * Check if a user can access privileged mount options.
410 */
411 int
412 vfs_suser(struct mount *mp, struct thread *td)
413 {
414 int error;
415
416 /*
417 * If the thread is jailed, but this is not a jail-friendly file
418 * system, deny immediately.
419 */
420 if (jailed(td->td_ucred) && !(mp->mnt_vfc->vfc_flags & VFCF_JAIL))
421 return (EPERM);
422
423 /*
424 * If the file system was mounted outside a jail and a jailed thread
425 * tries to access it, deny immediately.
426 */
427 if (!jailed(mp->mnt_cred) && jailed(td->td_ucred))
428 return (EPERM);
429
430 /*
431 * If the file system was mounted inside different jail that the jail of
432 * the calling thread, deny immediately.
433 */
434 if (jailed(mp->mnt_cred) && jailed(td->td_ucred) &&
435 mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) {
436 return (EPERM);
437 }
438
439 if ((mp->mnt_flag & MNT_USER) == 0 ||
440 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
441 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
442 return (error);
443 }
444 return (0);
445 }
446
447 /*
448 * Get a new unique fsid. Try to make its val[0] unique, since this value
449 * will be used to create fake device numbers for stat(). Also try (but
450 * not so hard) make its val[0] unique mod 2^16, since some emulators only
451 * support 16-bit device numbers. We end up with unique val[0]'s for the
452 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
453 *
454 * Keep in mind that several mounts may be running in parallel. Starting
455 * the search one past where the previous search terminated is both a
456 * micro-optimization and a defense against returning the same fsid to
457 * different mounts.
458 */
459 void
460 vfs_getnewfsid(struct mount *mp)
461 {
462 static u_int16_t mntid_base;
463 struct mount *nmp;
464 fsid_t tfsid;
465 int mtype;
466
467 mtx_lock(&mntid_mtx);
468 mtype = mp->mnt_vfc->vfc_typenum;
469 tfsid.val[1] = mtype;
470 mtype = (mtype & 0xFF) << 24;
471 for (;;) {
472 tfsid.val[0] = makedev(255,
473 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
474 mntid_base++;
475 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
476 break;
477 vfs_rel(nmp);
478 }
479 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
480 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
481 mtx_unlock(&mntid_mtx);
482 }
483
484 /*
485 * Knob to control the precision of file timestamps:
486 *
487 * 0 = seconds only; nanoseconds zeroed.
488 * 1 = seconds and nanoseconds, accurate within 1/HZ.
489 * 2 = seconds and nanoseconds, truncated to microseconds.
490 * >=3 = seconds and nanoseconds, maximum precision.
491 */
492 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
493
494 static int timestamp_precision = TSP_SEC;
495 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
496 ×tamp_precision, 0, "");
497
498 /*
499 * Get a current timestamp.
500 */
501 void
502 vfs_timestamp(struct timespec *tsp)
503 {
504 struct timeval tv;
505
506 switch (timestamp_precision) {
507 case TSP_SEC:
508 tsp->tv_sec = time_second;
509 tsp->tv_nsec = 0;
510 break;
511 case TSP_HZ:
512 getnanotime(tsp);
513 break;
514 case TSP_USEC:
515 microtime(&tv);
516 TIMEVAL_TO_TIMESPEC(&tv, tsp);
517 break;
518 case TSP_NSEC:
519 default:
520 nanotime(tsp);
521 break;
522 }
523 }
524
525 /*
526 * Set vnode attributes to VNOVAL
527 */
528 void
529 vattr_null(struct vattr *vap)
530 {
531
532 vap->va_type = VNON;
533 vap->va_size = VNOVAL;
534 vap->va_bytes = VNOVAL;
535 vap->va_mode = VNOVAL;
536 vap->va_nlink = VNOVAL;
537 vap->va_uid = VNOVAL;
538 vap->va_gid = VNOVAL;
539 vap->va_fsid = VNOVAL;
540 vap->va_fileid = VNOVAL;
541 vap->va_blocksize = VNOVAL;
542 vap->va_rdev = VNOVAL;
543 vap->va_atime.tv_sec = VNOVAL;
544 vap->va_atime.tv_nsec = VNOVAL;
545 vap->va_mtime.tv_sec = VNOVAL;
546 vap->va_mtime.tv_nsec = VNOVAL;
547 vap->va_ctime.tv_sec = VNOVAL;
548 vap->va_ctime.tv_nsec = VNOVAL;
549 vap->va_birthtime.tv_sec = VNOVAL;
550 vap->va_birthtime.tv_nsec = VNOVAL;
551 vap->va_flags = VNOVAL;
552 vap->va_gen = VNOVAL;
553 vap->va_vaflags = 0;
554 }
555
556 /*
557 * This routine is called when we have too many vnodes. It attempts
558 * to free <count> vnodes and will potentially free vnodes that still
559 * have VM backing store (VM backing store is typically the cause
560 * of a vnode blowout so we want to do this). Therefore, this operation
561 * is not considered cheap.
562 *
563 * A number of conditions may prevent a vnode from being reclaimed.
564 * the buffer cache may have references on the vnode, a directory
565 * vnode may still have references due to the namei cache representing
566 * underlying files, or the vnode may be in active use. It is not
567 * desireable to reuse such vnodes. These conditions may cause the
568 * number of vnodes to reach some minimum value regardless of what
569 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
570 */
571 static int
572 vlrureclaim(struct mount *mp)
573 {
574 struct thread *td;
575 struct vnode *vp;
576 int done;
577 int trigger;
578 int usevnodes;
579 int count;
580
581 /*
582 * Calculate the trigger point, don't allow user
583 * screwups to blow us up. This prevents us from
584 * recycling vnodes with lots of resident pages. We
585 * aren't trying to free memory, we are trying to
586 * free vnodes.
587 */
588 usevnodes = desiredvnodes;
589 if (usevnodes <= 0)
590 usevnodes = 1;
591 trigger = cnt.v_page_count * 2 / usevnodes;
592 done = 0;
593 td = curthread;
594 vn_start_write(NULL, &mp, V_WAIT);
595 MNT_ILOCK(mp);
596 count = mp->mnt_nvnodelistsize / 10 + 1;
597 while (count != 0) {
598 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
599 while (vp != NULL && vp->v_type == VMARKER)
600 vp = TAILQ_NEXT(vp, v_nmntvnodes);
601 if (vp == NULL)
602 break;
603 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
604 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
605 --count;
606 if (!VI_TRYLOCK(vp))
607 goto next_iter;
608 /*
609 * If it's been deconstructed already, it's still
610 * referenced, or it exceeds the trigger, skip it.
611 */
612 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
613 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
614 vp->v_object->resident_page_count > trigger)) {
615 VI_UNLOCK(vp);
616 goto next_iter;
617 }
618 MNT_IUNLOCK(mp);
619 vholdl(vp);
620 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT, td)) {
621 vdrop(vp);
622 goto next_iter_mntunlocked;
623 }
624 VI_LOCK(vp);
625 /*
626 * v_usecount may have been bumped after VOP_LOCK() dropped
627 * the vnode interlock and before it was locked again.
628 *
629 * It is not necessary to recheck VI_DOOMED because it can
630 * only be set by another thread that holds both the vnode
631 * lock and vnode interlock. If another thread has the
632 * vnode lock before we get to VOP_LOCK() and obtains the
633 * vnode interlock after VOP_LOCK() drops the vnode
634 * interlock, the other thread will be unable to drop the
635 * vnode lock before our VOP_LOCK() call fails.
636 */
637 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
638 (vp->v_object != NULL &&
639 vp->v_object->resident_page_count > trigger)) {
640 VOP_UNLOCK(vp, LK_INTERLOCK, td);
641 goto next_iter_mntunlocked;
642 }
643 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
644 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
645 vgonel(vp);
646 VOP_UNLOCK(vp, 0, td);
647 vdropl(vp);
648 done++;
649 next_iter_mntunlocked:
650 if ((count % 256) != 0)
651 goto relock_mnt;
652 goto yield;
653 next_iter:
654 if ((count % 256) != 0)
655 continue;
656 MNT_IUNLOCK(mp);
657 yield:
658 uio_yield();
659 relock_mnt:
660 MNT_ILOCK(mp);
661 }
662 MNT_IUNLOCK(mp);
663 vn_finished_write(mp);
664 return done;
665 }
666
667 /*
668 * Attempt to keep the free list at wantfreevnodes length.
669 */
670 static void
671 vnlru_free(int count)
672 {
673 struct vnode *vp;
674 int vfslocked;
675
676 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
677 for (; count > 0; count--) {
678 vp = TAILQ_FIRST(&vnode_free_list);
679 /*
680 * The list can be modified while the free_list_mtx
681 * has been dropped and vp could be NULL here.
682 */
683 if (!vp)
684 break;
685 VNASSERT(vp->v_op != NULL, vp,
686 ("vnlru_free: vnode already reclaimed."));
687 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
688 /*
689 * Don't recycle if we can't get the interlock.
690 */
691 if (!VI_TRYLOCK(vp)) {
692 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
693 continue;
694 }
695 VNASSERT(VCANRECYCLE(vp), vp,
696 ("vp inconsistent on freelist"));
697 freevnodes--;
698 vp->v_iflag &= ~VI_FREE;
699 vholdl(vp);
700 mtx_unlock(&vnode_free_list_mtx);
701 VI_UNLOCK(vp);
702 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
703 vtryrecycle(vp);
704 VFS_UNLOCK_GIANT(vfslocked);
705 /*
706 * If the recycled succeeded this vdrop will actually free
707 * the vnode. If not it will simply place it back on
708 * the free list.
709 */
710 vdrop(vp);
711 mtx_lock(&vnode_free_list_mtx);
712 }
713 }
714 /*
715 * Attempt to recycle vnodes in a context that is always safe to block.
716 * Calling vlrurecycle() from the bowels of filesystem code has some
717 * interesting deadlock problems.
718 */
719 static struct proc *vnlruproc;
720 static int vnlruproc_sig;
721
722 static void
723 vnlru_proc(void)
724 {
725 struct mount *mp, *nmp;
726 int done, vfslocked;
727 struct proc *p = vnlruproc;
728 struct thread *td = curthread;
729
730 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
731 SHUTDOWN_PRI_FIRST);
732
733 for (;;) {
734 kthread_suspend_check(p);
735 mtx_lock(&vnode_free_list_mtx);
736 if (freevnodes > wantfreevnodes)
737 vnlru_free(freevnodes - wantfreevnodes);
738 if (numvnodes <= desiredvnodes * 9 / 10) {
739 vnlruproc_sig = 0;
740 wakeup(&vnlruproc_sig);
741 msleep(vnlruproc, &vnode_free_list_mtx,
742 PVFS|PDROP, "vlruwt", hz);
743 continue;
744 }
745 mtx_unlock(&vnode_free_list_mtx);
746 done = 0;
747 mtx_lock(&mountlist_mtx);
748 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
749 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
750 nmp = TAILQ_NEXT(mp, mnt_list);
751 continue;
752 }
753 vfslocked = VFS_LOCK_GIANT(mp);
754 done += vlrureclaim(mp);
755 VFS_UNLOCK_GIANT(vfslocked);
756 mtx_lock(&mountlist_mtx);
757 nmp = TAILQ_NEXT(mp, mnt_list);
758 vfs_unbusy(mp, td);
759 }
760 mtx_unlock(&mountlist_mtx);
761 if (done == 0) {
762 EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
763 #if 0
764 /* These messages are temporary debugging aids */
765 if (vnlru_nowhere < 5)
766 printf("vnlru process getting nowhere..\n");
767 else if (vnlru_nowhere == 5)
768 printf("vnlru process messages stopped.\n");
769 #endif
770 vnlru_nowhere++;
771 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
772 } else
773 uio_yield();
774 }
775 }
776
777 static struct kproc_desc vnlru_kp = {
778 "vnlru",
779 vnlru_proc,
780 &vnlruproc
781 };
782 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
783 &vnlru_kp);
784
785 /*
786 * Routines having to do with the management of the vnode table.
787 */
788
789 void
790 vdestroy(struct vnode *vp)
791 {
792 struct bufobj *bo;
793
794 CTR1(KTR_VFS, "vdestroy vp %p", vp);
795 mtx_lock(&vnode_free_list_mtx);
796 numvnodes--;
797 mtx_unlock(&vnode_free_list_mtx);
798 bo = &vp->v_bufobj;
799 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
800 ("cleaned vnode still on the free list."));
801 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
802 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
803 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
804 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
805 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
806 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
807 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
808 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
809 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
810 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
811 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
812 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
813 VI_UNLOCK(vp);
814 #ifdef MAC
815 mac_destroy_vnode(vp);
816 #endif
817 if (vp->v_pollinfo != NULL)
818 destroy_vpollinfo(vp->v_pollinfo);
819 #ifdef INVARIANTS
820 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
821 vp->v_op = NULL;
822 #endif
823 lockdestroy(vp->v_vnlock);
824 mtx_destroy(&vp->v_interlock);
825 uma_zfree(vnode_zone, vp);
826 }
827
828 /*
829 * Try to recycle a freed vnode. We abort if anyone picks up a reference
830 * before we actually vgone(). This function must be called with the vnode
831 * held to prevent the vnode from being returned to the free list midway
832 * through vgone().
833 */
834 static int
835 vtryrecycle(struct vnode *vp)
836 {
837 struct thread *td = curthread;
838 struct mount *vnmp;
839
840 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
841 VNASSERT(vp->v_holdcnt, vp,
842 ("vtryrecycle: Recycling vp %p without a reference.", vp));
843 /*
844 * This vnode may found and locked via some other list, if so we
845 * can't recycle it yet.
846 */
847 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
848 return (EWOULDBLOCK);
849 /*
850 * Don't recycle if its filesystem is being suspended.
851 */
852 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
853 VOP_UNLOCK(vp, 0, td);
854 return (EBUSY);
855 }
856 /*
857 * If we got this far, we need to acquire the interlock and see if
858 * anyone picked up this vnode from another list. If not, we will
859 * mark it with DOOMED via vgonel() so that anyone who does find it
860 * will skip over it.
861 */
862 VI_LOCK(vp);
863 if (vp->v_usecount) {
864 VOP_UNLOCK(vp, LK_INTERLOCK, td);
865 vn_finished_write(vnmp);
866 return (EBUSY);
867 }
868 if ((vp->v_iflag & VI_DOOMED) == 0)
869 vgonel(vp);
870 VOP_UNLOCK(vp, LK_INTERLOCK, td);
871 vn_finished_write(vnmp);
872 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
873 return (0);
874 }
875
876 /*
877 * Return the next vnode from the free list.
878 */
879 int
880 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
881 struct vnode **vpp)
882 {
883 struct vnode *vp = NULL;
884 struct bufobj *bo;
885
886 mtx_lock(&vnode_free_list_mtx);
887 /*
888 * Lend our context to reclaim vnodes if they've exceeded the max.
889 */
890 if (freevnodes > wantfreevnodes)
891 vnlru_free(1);
892 /*
893 * Wait for available vnodes.
894 */
895 if (numvnodes > desiredvnodes) {
896 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
897 /*
898 * File system is beeing suspended, we cannot risk a
899 * deadlock here, so allocate new vnode anyway.
900 */
901 if (freevnodes > wantfreevnodes)
902 vnlru_free(freevnodes - wantfreevnodes);
903 goto alloc;
904 }
905 if (vnlruproc_sig == 0) {
906 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
907 wakeup(vnlruproc);
908 }
909 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
910 "vlruwk", hz);
911 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
912 if (numvnodes > desiredvnodes) {
913 mtx_unlock(&vnode_free_list_mtx);
914 return (ENFILE);
915 }
916 #endif
917 }
918 alloc:
919 numvnodes++;
920 mtx_unlock(&vnode_free_list_mtx);
921 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
922 /*
923 * Setup locks.
924 */
925 vp->v_vnlock = &vp->v_lock;
926 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
927 /*
928 * By default, don't allow shared locks unless filesystems
929 * opt-in.
930 */
931 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
932 /*
933 * Initialize bufobj.
934 */
935 bo = &vp->v_bufobj;
936 bo->__bo_vnode = vp;
937 bo->bo_mtx = &vp->v_interlock;
938 bo->bo_ops = &buf_ops_bio;
939 bo->bo_private = vp;
940 TAILQ_INIT(&bo->bo_clean.bv_hd);
941 TAILQ_INIT(&bo->bo_dirty.bv_hd);
942 /*
943 * Initialize namecache.
944 */
945 LIST_INIT(&vp->v_cache_src);
946 TAILQ_INIT(&vp->v_cache_dst);
947 /*
948 * Finalize various vnode identity bits.
949 */
950 vp->v_type = VNON;
951 vp->v_tag = tag;
952 vp->v_op = vops;
953 v_incr_usecount(vp);
954 vp->v_data = 0;
955 #ifdef MAC
956 mac_init_vnode(vp);
957 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
958 mac_associate_vnode_singlelabel(mp, vp);
959 else if (mp == NULL && vops != &dead_vnodeops)
960 printf("NULL mp in getnewvnode()\n");
961 #endif
962 if (mp != NULL) {
963 bo->bo_bsize = mp->mnt_stat.f_iosize;
964 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
965 vp->v_vflag |= VV_NOKNOTE;
966 }
967
968 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
969 *vpp = vp;
970 return (0);
971 }
972
973 /*
974 * Delete from old mount point vnode list, if on one.
975 */
976 static void
977 delmntque(struct vnode *vp)
978 {
979 struct mount *mp;
980
981 mp = vp->v_mount;
982 if (mp == NULL)
983 return;
984 MNT_ILOCK(mp);
985 vp->v_mount = NULL;
986 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
987 ("bad mount point vnode list size"));
988 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
989 mp->mnt_nvnodelistsize--;
990 MNT_REL(mp);
991 MNT_IUNLOCK(mp);
992 }
993
994 static void
995 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
996 {
997 struct thread *td;
998
999 td = curthread; /* XXX ? */
1000 vp->v_data = NULL;
1001 vp->v_op = &dead_vnodeops;
1002 /* XXX non mp-safe fs may still call insmntque with vnode
1003 unlocked */
1004 if (!VOP_ISLOCKED(vp, td))
1005 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1006 vgone(vp);
1007 vput(vp);
1008 }
1009
1010 /*
1011 * Insert into list of vnodes for the new mount point, if available.
1012 */
1013 int
1014 insmntque1(struct vnode *vp, struct mount *mp,
1015 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1016 {
1017 int locked;
1018
1019 KASSERT(vp->v_mount == NULL,
1020 ("insmntque: vnode already on per mount vnode list"));
1021 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1022 #if 0
1023 #ifdef DEBUG_VFS_LOCKS
1024 if (!VFS_NEEDSGIANT(mp))
1025 ASSERT_VOP_ELOCKED(vp,
1026 "insmntque: mp-safe fs and non-locked vp");
1027 #endif
1028 #endif
1029 MNT_ILOCK(mp);
1030 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1031 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1032 mp->mnt_nvnodelistsize == 0)) {
1033 locked = VOP_ISLOCKED(vp, curthread);
1034 if (!locked || (locked == LK_EXCLUSIVE &&
1035 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1036 MNT_IUNLOCK(mp);
1037 if (dtr != NULL)
1038 dtr(vp, dtr_arg);
1039 return (EBUSY);
1040 }
1041 }
1042 vp->v_mount = mp;
1043 MNT_REF(mp);
1044 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1045 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1046 ("neg mount point vnode list size"));
1047 mp->mnt_nvnodelistsize++;
1048 MNT_IUNLOCK(mp);
1049 return (0);
1050 }
1051
1052 int
1053 insmntque(struct vnode *vp, struct mount *mp)
1054 {
1055
1056 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1057 }
1058
1059 /*
1060 * Flush out and invalidate all buffers associated with a bufobj
1061 * Called with the underlying object locked.
1062 */
1063 int
1064 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag,
1065 int slptimeo)
1066 {
1067 int error;
1068
1069 BO_LOCK(bo);
1070 if (flags & V_SAVE) {
1071 error = bufobj_wwait(bo, slpflag, slptimeo);
1072 if (error) {
1073 BO_UNLOCK(bo);
1074 return (error);
1075 }
1076 if (bo->bo_dirty.bv_cnt > 0) {
1077 BO_UNLOCK(bo);
1078 if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1079 return (error);
1080 /*
1081 * XXX We could save a lock/unlock if this was only
1082 * enabled under INVARIANTS
1083 */
1084 BO_LOCK(bo);
1085 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1086 panic("vinvalbuf: dirty bufs");
1087 }
1088 }
1089 /*
1090 * If you alter this loop please notice that interlock is dropped and
1091 * reacquired in flushbuflist. Special care is needed to ensure that
1092 * no race conditions occur from this.
1093 */
1094 do {
1095 error = flushbuflist(&bo->bo_clean,
1096 flags, bo, slpflag, slptimeo);
1097 if (error == 0)
1098 error = flushbuflist(&bo->bo_dirty,
1099 flags, bo, slpflag, slptimeo);
1100 if (error != 0 && error != EAGAIN) {
1101 BO_UNLOCK(bo);
1102 return (error);
1103 }
1104 } while (error != 0);
1105
1106 /*
1107 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1108 * have write I/O in-progress but if there is a VM object then the
1109 * VM object can also have read-I/O in-progress.
1110 */
1111 do {
1112 bufobj_wwait(bo, 0, 0);
1113 BO_UNLOCK(bo);
1114 if (bo->bo_object != NULL) {
1115 VM_OBJECT_LOCK(bo->bo_object);
1116 vm_object_pip_wait(bo->bo_object, "bovlbx");
1117 VM_OBJECT_UNLOCK(bo->bo_object);
1118 }
1119 BO_LOCK(bo);
1120 } while (bo->bo_numoutput > 0);
1121 BO_UNLOCK(bo);
1122
1123 /*
1124 * Destroy the copy in the VM cache, too.
1125 */
1126 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1127 VM_OBJECT_LOCK(bo->bo_object);
1128 vm_object_page_remove(bo->bo_object, 0, 0,
1129 (flags & V_SAVE) ? TRUE : FALSE);
1130 VM_OBJECT_UNLOCK(bo->bo_object);
1131 }
1132
1133 #ifdef INVARIANTS
1134 BO_LOCK(bo);
1135 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1136 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1137 panic("vinvalbuf: flush failed");
1138 BO_UNLOCK(bo);
1139 #endif
1140 return (0);
1141 }
1142
1143 /*
1144 * Flush out and invalidate all buffers associated with a vnode.
1145 * Called with the underlying object locked.
1146 */
1147 int
1148 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag,
1149 int slptimeo)
1150 {
1151
1152 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1153 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1154 return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1155 }
1156
1157 /*
1158 * Flush out buffers on the specified list.
1159 *
1160 */
1161 static int
1162 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1163 int slptimeo)
1164 {
1165 struct buf *bp, *nbp;
1166 int retval, error;
1167 daddr_t lblkno;
1168 b_xflags_t xflags;
1169
1170 ASSERT_BO_LOCKED(bo);
1171
1172 retval = 0;
1173 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1174 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1175 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1176 continue;
1177 }
1178 lblkno = 0;
1179 xflags = 0;
1180 if (nbp != NULL) {
1181 lblkno = nbp->b_lblkno;
1182 xflags = nbp->b_xflags &
1183 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1184 }
1185 retval = EAGAIN;
1186 error = BUF_TIMELOCK(bp,
1187 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1188 "flushbuf", slpflag, slptimeo);
1189 if (error) {
1190 BO_LOCK(bo);
1191 return (error != ENOLCK ? error : EAGAIN);
1192 }
1193 KASSERT(bp->b_bufobj == bo,
1194 ("bp %p wrong b_bufobj %p should be %p",
1195 bp, bp->b_bufobj, bo));
1196 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1197 BUF_UNLOCK(bp);
1198 BO_LOCK(bo);
1199 return (EAGAIN);
1200 }
1201 /*
1202 * XXX Since there are no node locks for NFS, I
1203 * believe there is a slight chance that a delayed
1204 * write will occur while sleeping just above, so
1205 * check for it.
1206 */
1207 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1208 (flags & V_SAVE)) {
1209 bremfree(bp);
1210 bp->b_flags |= B_ASYNC;
1211 bwrite(bp);
1212 BO_LOCK(bo);
1213 return (EAGAIN); /* XXX: why not loop ? */
1214 }
1215 bremfree(bp);
1216 bp->b_flags |= (B_INVAL | B_RELBUF);
1217 bp->b_flags &= ~B_ASYNC;
1218 brelse(bp);
1219 BO_LOCK(bo);
1220 if (nbp != NULL &&
1221 (nbp->b_bufobj != bo ||
1222 nbp->b_lblkno != lblkno ||
1223 (nbp->b_xflags &
1224 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1225 break; /* nbp invalid */
1226 }
1227 return (retval);
1228 }
1229
1230 /*
1231 * Truncate a file's buffer and pages to a specified length. This
1232 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1233 * sync activity.
1234 */
1235 int
1236 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1237 off_t length, int blksize)
1238 {
1239 struct buf *bp, *nbp;
1240 int anyfreed;
1241 int trunclbn;
1242 struct bufobj *bo;
1243
1244 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1245 /*
1246 * Round up to the *next* lbn.
1247 */
1248 trunclbn = (length + blksize - 1) / blksize;
1249
1250 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1251 restart:
1252 VI_LOCK(vp);
1253 bo = &vp->v_bufobj;
1254 anyfreed = 1;
1255 for (;anyfreed;) {
1256 anyfreed = 0;
1257 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1258 if (bp->b_lblkno < trunclbn)
1259 continue;
1260 if (BUF_LOCK(bp,
1261 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1262 VI_MTX(vp)) == ENOLCK)
1263 goto restart;
1264
1265 bremfree(bp);
1266 bp->b_flags |= (B_INVAL | B_RELBUF);
1267 bp->b_flags &= ~B_ASYNC;
1268 brelse(bp);
1269 anyfreed = 1;
1270
1271 if (nbp != NULL &&
1272 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1273 (nbp->b_vp != vp) ||
1274 (nbp->b_flags & B_DELWRI))) {
1275 goto restart;
1276 }
1277 VI_LOCK(vp);
1278 }
1279
1280 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1281 if (bp->b_lblkno < trunclbn)
1282 continue;
1283 if (BUF_LOCK(bp,
1284 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1285 VI_MTX(vp)) == ENOLCK)
1286 goto restart;
1287 bremfree(bp);
1288 bp->b_flags |= (B_INVAL | B_RELBUF);
1289 bp->b_flags &= ~B_ASYNC;
1290 brelse(bp);
1291 anyfreed = 1;
1292 if (nbp != NULL &&
1293 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1294 (nbp->b_vp != vp) ||
1295 (nbp->b_flags & B_DELWRI) == 0)) {
1296 goto restart;
1297 }
1298 VI_LOCK(vp);
1299 }
1300 }
1301
1302 if (length > 0) {
1303 restartsync:
1304 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1305 if (bp->b_lblkno > 0)
1306 continue;
1307 /*
1308 * Since we hold the vnode lock this should only
1309 * fail if we're racing with the buf daemon.
1310 */
1311 if (BUF_LOCK(bp,
1312 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1313 VI_MTX(vp)) == ENOLCK) {
1314 goto restart;
1315 }
1316 VNASSERT((bp->b_flags & B_DELWRI), vp,
1317 ("buf(%p) on dirty queue without DELWRI", bp));
1318
1319 bremfree(bp);
1320 bawrite(bp);
1321 VI_LOCK(vp);
1322 goto restartsync;
1323 }
1324 }
1325
1326 bufobj_wwait(bo, 0, 0);
1327 VI_UNLOCK(vp);
1328 vnode_pager_setsize(vp, length);
1329
1330 return (0);
1331 }
1332
1333 /*
1334 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1335 * a vnode.
1336 *
1337 * NOTE: We have to deal with the special case of a background bitmap
1338 * buffer, a situation where two buffers will have the same logical
1339 * block offset. We want (1) only the foreground buffer to be accessed
1340 * in a lookup and (2) must differentiate between the foreground and
1341 * background buffer in the splay tree algorithm because the splay
1342 * tree cannot normally handle multiple entities with the same 'index'.
1343 * We accomplish this by adding differentiating flags to the splay tree's
1344 * numerical domain.
1345 */
1346 static
1347 struct buf *
1348 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1349 {
1350 struct buf dummy;
1351 struct buf *lefttreemax, *righttreemin, *y;
1352
1353 if (root == NULL)
1354 return (NULL);
1355 lefttreemax = righttreemin = &dummy;
1356 for (;;) {
1357 if (lblkno < root->b_lblkno ||
1358 (lblkno == root->b_lblkno &&
1359 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1360 if ((y = root->b_left) == NULL)
1361 break;
1362 if (lblkno < y->b_lblkno) {
1363 /* Rotate right. */
1364 root->b_left = y->b_right;
1365 y->b_right = root;
1366 root = y;
1367 if ((y = root->b_left) == NULL)
1368 break;
1369 }
1370 /* Link into the new root's right tree. */
1371 righttreemin->b_left = root;
1372 righttreemin = root;
1373 } else if (lblkno > root->b_lblkno ||
1374 (lblkno == root->b_lblkno &&
1375 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1376 if ((y = root->b_right) == NULL)
1377 break;
1378 if (lblkno > y->b_lblkno) {
1379 /* Rotate left. */
1380 root->b_right = y->b_left;
1381 y->b_left = root;
1382 root = y;
1383 if ((y = root->b_right) == NULL)
1384 break;
1385 }
1386 /* Link into the new root's left tree. */
1387 lefttreemax->b_right = root;
1388 lefttreemax = root;
1389 } else {
1390 break;
1391 }
1392 root = y;
1393 }
1394 /* Assemble the new root. */
1395 lefttreemax->b_right = root->b_left;
1396 righttreemin->b_left = root->b_right;
1397 root->b_left = dummy.b_right;
1398 root->b_right = dummy.b_left;
1399 return (root);
1400 }
1401
1402 static void
1403 buf_vlist_remove(struct buf *bp)
1404 {
1405 struct buf *root;
1406 struct bufv *bv;
1407
1408 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1409 ASSERT_BO_LOCKED(bp->b_bufobj);
1410 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1411 (BX_VNDIRTY|BX_VNCLEAN),
1412 ("buf_vlist_remove: Buf %p is on two lists", bp));
1413 if (bp->b_xflags & BX_VNDIRTY)
1414 bv = &bp->b_bufobj->bo_dirty;
1415 else
1416 bv = &bp->b_bufobj->bo_clean;
1417 if (bp != bv->bv_root) {
1418 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1419 KASSERT(root == bp, ("splay lookup failed in remove"));
1420 }
1421 if (bp->b_left == NULL) {
1422 root = bp->b_right;
1423 } else {
1424 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1425 root->b_right = bp->b_right;
1426 }
1427 bv->bv_root = root;
1428 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1429 bv->bv_cnt--;
1430 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1431 }
1432
1433 /*
1434 * Add the buffer to the sorted clean or dirty block list using a
1435 * splay tree algorithm.
1436 *
1437 * NOTE: xflags is passed as a constant, optimizing this inline function!
1438 */
1439 static void
1440 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1441 {
1442 struct buf *root;
1443 struct bufv *bv;
1444
1445 ASSERT_BO_LOCKED(bo);
1446 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1447 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1448 bp->b_xflags |= xflags;
1449 if (xflags & BX_VNDIRTY)
1450 bv = &bo->bo_dirty;
1451 else
1452 bv = &bo->bo_clean;
1453
1454 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1455 if (root == NULL) {
1456 bp->b_left = NULL;
1457 bp->b_right = NULL;
1458 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1459 } else if (bp->b_lblkno < root->b_lblkno ||
1460 (bp->b_lblkno == root->b_lblkno &&
1461 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1462 bp->b_left = root->b_left;
1463 bp->b_right = root;
1464 root->b_left = NULL;
1465 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1466 } else {
1467 bp->b_right = root->b_right;
1468 bp->b_left = root;
1469 root->b_right = NULL;
1470 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1471 }
1472 bv->bv_cnt++;
1473 bv->bv_root = bp;
1474 }
1475
1476 /*
1477 * Lookup a buffer using the splay tree. Note that we specifically avoid
1478 * shadow buffers used in background bitmap writes.
1479 *
1480 * This code isn't quite efficient as it could be because we are maintaining
1481 * two sorted lists and do not know which list the block resides in.
1482 *
1483 * During a "make buildworld" the desired buffer is found at one of
1484 * the roots more than 60% of the time. Thus, checking both roots
1485 * before performing either splay eliminates unnecessary splays on the
1486 * first tree splayed.
1487 */
1488 struct buf *
1489 gbincore(struct bufobj *bo, daddr_t lblkno)
1490 {
1491 struct buf *bp;
1492
1493 ASSERT_BO_LOCKED(bo);
1494 if ((bp = bo->bo_clean.bv_root) != NULL &&
1495 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1496 return (bp);
1497 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1498 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1499 return (bp);
1500 if ((bp = bo->bo_clean.bv_root) != NULL) {
1501 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1502 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1503 return (bp);
1504 }
1505 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1506 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1507 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1508 return (bp);
1509 }
1510 return (NULL);
1511 }
1512
1513 /*
1514 * Associate a buffer with a vnode.
1515 */
1516 void
1517 bgetvp(struct vnode *vp, struct buf *bp)
1518 {
1519
1520 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1521
1522 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1523 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1524 ("bgetvp: bp already attached! %p", bp));
1525
1526 ASSERT_VI_LOCKED(vp, "bgetvp");
1527 vholdl(vp);
1528 if (VFS_NEEDSGIANT(vp->v_mount) ||
1529 vp->v_bufobj.bo_flag & BO_NEEDSGIANT)
1530 bp->b_flags |= B_NEEDSGIANT;
1531 bp->b_vp = vp;
1532 bp->b_bufobj = &vp->v_bufobj;
1533 /*
1534 * Insert onto list for new vnode.
1535 */
1536 buf_vlist_add(bp, &vp->v_bufobj, BX_VNCLEAN);
1537 }
1538
1539 /*
1540 * Disassociate a buffer from a vnode.
1541 */
1542 void
1543 brelvp(struct buf *bp)
1544 {
1545 struct bufobj *bo;
1546 struct vnode *vp;
1547
1548 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1549 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1550
1551 /*
1552 * Delete from old vnode list, if on one.
1553 */
1554 vp = bp->b_vp; /* XXX */
1555 bo = bp->b_bufobj;
1556 BO_LOCK(bo);
1557 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1558 buf_vlist_remove(bp);
1559 else
1560 panic("brelvp: Buffer %p not on queue.", bp);
1561 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1562 bo->bo_flag &= ~BO_ONWORKLST;
1563 mtx_lock(&sync_mtx);
1564 LIST_REMOVE(bo, bo_synclist);
1565 syncer_worklist_len--;
1566 mtx_unlock(&sync_mtx);
1567 }
1568 bp->b_flags &= ~B_NEEDSGIANT;
1569 bp->b_vp = NULL;
1570 bp->b_bufobj = NULL;
1571 vdropl(vp);
1572 }
1573
1574 /*
1575 * Add an item to the syncer work queue.
1576 */
1577 static void
1578 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1579 {
1580 int slot;
1581
1582 ASSERT_BO_LOCKED(bo);
1583
1584 mtx_lock(&sync_mtx);
1585 if (bo->bo_flag & BO_ONWORKLST)
1586 LIST_REMOVE(bo, bo_synclist);
1587 else {
1588 bo->bo_flag |= BO_ONWORKLST;
1589 syncer_worklist_len++;
1590 }
1591
1592 if (delay > syncer_maxdelay - 2)
1593 delay = syncer_maxdelay - 2;
1594 slot = (syncer_delayno + delay) & syncer_mask;
1595
1596 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1597 mtx_unlock(&sync_mtx);
1598 }
1599
1600 static int
1601 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1602 {
1603 int error, len;
1604
1605 mtx_lock(&sync_mtx);
1606 len = syncer_worklist_len - sync_vnode_count;
1607 mtx_unlock(&sync_mtx);
1608 error = SYSCTL_OUT(req, &len, sizeof(len));
1609 return (error);
1610 }
1611
1612 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1613 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1614
1615 static struct proc *updateproc;
1616 static void sched_sync(void);
1617 static struct kproc_desc up_kp = {
1618 "syncer",
1619 sched_sync,
1620 &updateproc
1621 };
1622 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1623
1624 static int
1625 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1626 {
1627 struct vnode *vp;
1628 struct mount *mp;
1629 int vfslocked;
1630
1631 vfslocked = 0;
1632 restart:
1633 *bo = LIST_FIRST(slp);
1634 if (*bo == NULL) {
1635 VFS_UNLOCK_GIANT(vfslocked);
1636 return (0);
1637 }
1638 vp = (*bo)->__bo_vnode; /* XXX */
1639 if (VFS_NEEDSGIANT(vp->v_mount)) {
1640 if (!vfslocked) {
1641 vfslocked = 1;
1642 if (mtx_trylock(&Giant) == 0) {
1643 mtx_unlock(&sync_mtx);
1644 mtx_lock(&Giant);
1645 mtx_lock(&sync_mtx);
1646 goto restart;
1647 }
1648 }
1649 } else {
1650 VFS_UNLOCK_GIANT(vfslocked);
1651 vfslocked = 0;
1652 }
1653 if (VOP_ISLOCKED(vp, NULL) != 0) {
1654 VFS_UNLOCK_GIANT(vfslocked);
1655 return (1);
1656 }
1657 if (VI_TRYLOCK(vp) == 0) {
1658 VFS_UNLOCK_GIANT(vfslocked);
1659 return (1);
1660 }
1661 /*
1662 * We use vhold in case the vnode does not
1663 * successfully sync. vhold prevents the vnode from
1664 * going away when we unlock the sync_mtx so that
1665 * we can acquire the vnode interlock.
1666 */
1667 vholdl(vp);
1668 mtx_unlock(&sync_mtx);
1669 VI_UNLOCK(vp);
1670 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1671 vdrop(vp);
1672 VFS_UNLOCK_GIANT(vfslocked);
1673 mtx_lock(&sync_mtx);
1674 return (*bo == LIST_FIRST(slp));
1675 }
1676 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1677 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1678 VOP_UNLOCK(vp, 0, td);
1679 vn_finished_write(mp);
1680 VI_LOCK(vp);
1681 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1682 /*
1683 * Put us back on the worklist. The worklist
1684 * routine will remove us from our current
1685 * position and then add us back in at a later
1686 * position.
1687 */
1688 vn_syncer_add_to_worklist(*bo, syncdelay);
1689 }
1690 vdropl(vp);
1691 VFS_UNLOCK_GIANT(vfslocked);
1692 mtx_lock(&sync_mtx);
1693 return (0);
1694 }
1695
1696 /*
1697 * System filesystem synchronizer daemon.
1698 */
1699 static void
1700 sched_sync(void)
1701 {
1702 struct synclist *next;
1703 struct synclist *slp;
1704 struct bufobj *bo;
1705 long starttime;
1706 struct thread *td = curthread;
1707 static int dummychan;
1708 int last_work_seen;
1709 int net_worklist_len;
1710 int syncer_final_iter;
1711 int first_printf;
1712 int error;
1713
1714 last_work_seen = 0;
1715 syncer_final_iter = 0;
1716 first_printf = 1;
1717 syncer_state = SYNCER_RUNNING;
1718 starttime = time_uptime;
1719 td->td_pflags |= TDP_NORUNNINGBUF;
1720
1721 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1722 SHUTDOWN_PRI_LAST);
1723
1724 mtx_lock(&sync_mtx);
1725 for (;;) {
1726 if (syncer_state == SYNCER_FINAL_DELAY &&
1727 syncer_final_iter == 0) {
1728 mtx_unlock(&sync_mtx);
1729 kthread_suspend_check(td->td_proc);
1730 mtx_lock(&sync_mtx);
1731 }
1732 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1733 if (syncer_state != SYNCER_RUNNING &&
1734 starttime != time_uptime) {
1735 if (first_printf) {
1736 printf("\nSyncing disks, vnodes remaining...");
1737 first_printf = 0;
1738 }
1739 printf("%d ", net_worklist_len);
1740 }
1741 starttime = time_uptime;
1742
1743 /*
1744 * Push files whose dirty time has expired. Be careful
1745 * of interrupt race on slp queue.
1746 *
1747 * Skip over empty worklist slots when shutting down.
1748 */
1749 do {
1750 slp = &syncer_workitem_pending[syncer_delayno];
1751 syncer_delayno += 1;
1752 if (syncer_delayno == syncer_maxdelay)
1753 syncer_delayno = 0;
1754 next = &syncer_workitem_pending[syncer_delayno];
1755 /*
1756 * If the worklist has wrapped since the
1757 * it was emptied of all but syncer vnodes,
1758 * switch to the FINAL_DELAY state and run
1759 * for one more second.
1760 */
1761 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1762 net_worklist_len == 0 &&
1763 last_work_seen == syncer_delayno) {
1764 syncer_state = SYNCER_FINAL_DELAY;
1765 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1766 }
1767 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1768 syncer_worklist_len > 0);
1769
1770 /*
1771 * Keep track of the last time there was anything
1772 * on the worklist other than syncer vnodes.
1773 * Return to the SHUTTING_DOWN state if any
1774 * new work appears.
1775 */
1776 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1777 last_work_seen = syncer_delayno;
1778 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1779 syncer_state = SYNCER_SHUTTING_DOWN;
1780 while (!LIST_EMPTY(slp)) {
1781 error = sync_vnode(slp, &bo, td);
1782 if (error == 1) {
1783 LIST_REMOVE(bo, bo_synclist);
1784 LIST_INSERT_HEAD(next, bo, bo_synclist);
1785 continue;
1786 }
1787 }
1788 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1789 syncer_final_iter--;
1790 /*
1791 * The variable rushjob allows the kernel to speed up the
1792 * processing of the filesystem syncer process. A rushjob
1793 * value of N tells the filesystem syncer to process the next
1794 * N seconds worth of work on its queue ASAP. Currently rushjob
1795 * is used by the soft update code to speed up the filesystem
1796 * syncer process when the incore state is getting so far
1797 * ahead of the disk that the kernel memory pool is being
1798 * threatened with exhaustion.
1799 */
1800 if (rushjob > 0) {
1801 rushjob -= 1;
1802 continue;
1803 }
1804 /*
1805 * Just sleep for a short period of time between
1806 * iterations when shutting down to allow some I/O
1807 * to happen.
1808 *
1809 * If it has taken us less than a second to process the
1810 * current work, then wait. Otherwise start right over
1811 * again. We can still lose time if any single round
1812 * takes more than two seconds, but it does not really
1813 * matter as we are just trying to generally pace the
1814 * filesystem activity.
1815 */
1816 if (syncer_state != SYNCER_RUNNING)
1817 msleep(&dummychan, &sync_mtx, PPAUSE, "syncfnl",
1818 hz / SYNCER_SHUTDOWN_SPEEDUP);
1819 else if (time_uptime == starttime)
1820 msleep(&lbolt, &sync_mtx, PPAUSE, "syncer", 0);
1821 }
1822 }
1823
1824 /*
1825 * Request the syncer daemon to speed up its work.
1826 * We never push it to speed up more than half of its
1827 * normal turn time, otherwise it could take over the cpu.
1828 */
1829 int
1830 speedup_syncer(void)
1831 {
1832 struct thread *td;
1833 int ret = 0;
1834
1835 td = FIRST_THREAD_IN_PROC(updateproc);
1836 mtx_lock(&sync_mtx);
1837 if (rushjob < syncdelay / 2) {
1838 rushjob += 1;
1839 stat_rush_requests += 1;
1840 ret = 1;
1841 }
1842 mtx_unlock(&sync_mtx);
1843 sleepq_remove(td, &lbolt);
1844 return (ret);
1845 }
1846
1847 /*
1848 * Tell the syncer to speed up its work and run though its work
1849 * list several times, then tell it to shut down.
1850 */
1851 static void
1852 syncer_shutdown(void *arg, int howto)
1853 {
1854 struct thread *td;
1855
1856 if (howto & RB_NOSYNC)
1857 return;
1858 td = FIRST_THREAD_IN_PROC(updateproc);
1859 mtx_lock(&sync_mtx);
1860 syncer_state = SYNCER_SHUTTING_DOWN;
1861 rushjob = 0;
1862 mtx_unlock(&sync_mtx);
1863 sleepq_remove(td, &lbolt);
1864 kproc_shutdown(arg, howto);
1865 }
1866
1867 /*
1868 * Reassign a buffer from one vnode to another.
1869 * Used to assign file specific control information
1870 * (indirect blocks) to the vnode to which they belong.
1871 */
1872 void
1873 reassignbuf(struct buf *bp)
1874 {
1875 struct vnode *vp;
1876 struct bufobj *bo;
1877 int delay;
1878 #ifdef INVARIANTS
1879 struct bufv *bv;
1880 #endif
1881
1882 vp = bp->b_vp;
1883 bo = bp->b_bufobj;
1884 ++reassignbufcalls;
1885
1886 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1887 bp, bp->b_vp, bp->b_flags);
1888 /*
1889 * B_PAGING flagged buffers cannot be reassigned because their vp
1890 * is not fully linked in.
1891 */
1892 if (bp->b_flags & B_PAGING)
1893 panic("cannot reassign paging buffer");
1894
1895 /*
1896 * Delete from old vnode list, if on one.
1897 */
1898 VI_LOCK(vp);
1899 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1900 buf_vlist_remove(bp);
1901 else
1902 panic("reassignbuf: Buffer %p not on queue.", bp);
1903 /*
1904 * If dirty, put on list of dirty buffers; otherwise insert onto list
1905 * of clean buffers.
1906 */
1907 if (bp->b_flags & B_DELWRI) {
1908 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1909 switch (vp->v_type) {
1910 case VDIR:
1911 delay = dirdelay;
1912 break;
1913 case VCHR:
1914 delay = metadelay;
1915 break;
1916 default:
1917 delay = filedelay;
1918 }
1919 vn_syncer_add_to_worklist(bo, delay);
1920 }
1921 buf_vlist_add(bp, bo, BX_VNDIRTY);
1922 } else {
1923 buf_vlist_add(bp, bo, BX_VNCLEAN);
1924
1925 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1926 mtx_lock(&sync_mtx);
1927 LIST_REMOVE(bo, bo_synclist);
1928 syncer_worklist_len--;
1929 mtx_unlock(&sync_mtx);
1930 bo->bo_flag &= ~BO_ONWORKLST;
1931 }
1932 }
1933 #ifdef INVARIANTS
1934 bv = &bo->bo_clean;
1935 bp = TAILQ_FIRST(&bv->bv_hd);
1936 KASSERT(bp == NULL || bp->b_bufobj == bo,
1937 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1938 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1939 KASSERT(bp == NULL || bp->b_bufobj == bo,
1940 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1941 bv = &bo->bo_dirty;
1942 bp = TAILQ_FIRST(&bv->bv_hd);
1943 KASSERT(bp == NULL || bp->b_bufobj == bo,
1944 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1945 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1946 KASSERT(bp == NULL || bp->b_bufobj == bo,
1947 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1948 #endif
1949 VI_UNLOCK(vp);
1950 }
1951
1952 /*
1953 * Increment the use and hold counts on the vnode, taking care to reference
1954 * the driver's usecount if this is a chardev. The vholdl() will remove
1955 * the vnode from the free list if it is presently free. Requires the
1956 * vnode interlock and returns with it held.
1957 */
1958 static void
1959 v_incr_usecount(struct vnode *vp)
1960 {
1961
1962 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1963 vp, vp->v_holdcnt, vp->v_usecount);
1964 vp->v_usecount++;
1965 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1966 dev_lock();
1967 vp->v_rdev->si_usecount++;
1968 dev_unlock();
1969 }
1970 vholdl(vp);
1971 }
1972
1973 /*
1974 * Turn a holdcnt into a use+holdcnt such that only one call to
1975 * v_decr_usecount is needed.
1976 */
1977 static void
1978 v_upgrade_usecount(struct vnode *vp)
1979 {
1980
1981 CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
1982 vp, vp->v_holdcnt, vp->v_usecount);
1983 vp->v_usecount++;
1984 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1985 dev_lock();
1986 vp->v_rdev->si_usecount++;
1987 dev_unlock();
1988 }
1989 }
1990
1991 /*
1992 * Decrement the vnode use and hold count along with the driver's usecount
1993 * if this is a chardev. The vdropl() below releases the vnode interlock
1994 * as it may free the vnode.
1995 */
1996 static void
1997 v_decr_usecount(struct vnode *vp)
1998 {
1999
2000 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
2001 vp, vp->v_holdcnt, vp->v_usecount);
2002 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2003 VNASSERT(vp->v_usecount > 0, vp,
2004 ("v_decr_usecount: negative usecount"));
2005 vp->v_usecount--;
2006 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2007 dev_lock();
2008 vp->v_rdev->si_usecount--;
2009 dev_unlock();
2010 }
2011 vdropl(vp);
2012 }
2013
2014 /*
2015 * Decrement only the use count and driver use count. This is intended to
2016 * be paired with a follow on vdropl() to release the remaining hold count.
2017 * In this way we may vgone() a vnode with a 0 usecount without risk of
2018 * having it end up on a free list because the hold count is kept above 0.
2019 */
2020 static void
2021 v_decr_useonly(struct vnode *vp)
2022 {
2023
2024 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
2025 vp, vp->v_holdcnt, vp->v_usecount);
2026 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2027 VNASSERT(vp->v_usecount > 0, vp,
2028 ("v_decr_useonly: negative usecount"));
2029 vp->v_usecount--;
2030 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2031 dev_lock();
2032 vp->v_rdev->si_usecount--;
2033 dev_unlock();
2034 }
2035 }
2036
2037 /*
2038 * Grab a particular vnode from the free list, increment its
2039 * reference count and lock it. The vnode lock bit is set if the
2040 * vnode is being eliminated in vgone. The process is awakened
2041 * when the transition is completed, and an error returned to
2042 * indicate that the vnode is no longer usable (possibly having
2043 * been changed to a new filesystem type).
2044 */
2045 int
2046 vget(struct vnode *vp, int flags, struct thread *td)
2047 {
2048 int oweinact;
2049 int oldflags;
2050 int error;
2051
2052 error = 0;
2053 oldflags = flags;
2054 oweinact = 0;
2055 VFS_ASSERT_GIANT(vp->v_mount);
2056 if ((flags & LK_INTERLOCK) == 0)
2057 VI_LOCK(vp);
2058 /*
2059 * If the inactive call was deferred because vput() was called
2060 * with a shared lock, we have to do it here before another thread
2061 * gets a reference to data that should be dead.
2062 */
2063 if (vp->v_iflag & VI_OWEINACT) {
2064 if (flags & LK_NOWAIT) {
2065 VI_UNLOCK(vp);
2066 return (EBUSY);
2067 }
2068 flags &= ~LK_TYPE_MASK;
2069 flags |= LK_EXCLUSIVE;
2070 oweinact = 1;
2071 }
2072 vholdl(vp);
2073 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2074 vdrop(vp);
2075 return (error);
2076 }
2077 VI_LOCK(vp);
2078 /*
2079 * Deal with a timing window when the interlock is not held
2080 * and VI_DOOMED can be set, since we only have a holdcnt,
2081 * not a usecount.
2082 */
2083 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) {
2084 KASSERT((flags & LK_TYPE_MASK) == 0, ("Unexpected flags %x", flags));
2085 vdropl(vp);
2086 return (ENOENT);
2087 }
2088 /* Upgrade our holdcnt to a usecount. */
2089 v_upgrade_usecount(vp);
2090 if (oweinact) {
2091 if (vp->v_iflag & VI_OWEINACT)
2092 vinactive(vp, td);
2093 VI_UNLOCK(vp);
2094 if ((oldflags & LK_TYPE_MASK) == 0)
2095 VOP_UNLOCK(vp, 0, td);
2096 } else
2097 VI_UNLOCK(vp);
2098 return (0);
2099 }
2100
2101 /*
2102 * Increase the reference count of a vnode.
2103 */
2104 void
2105 vref(struct vnode *vp)
2106 {
2107
2108 VI_LOCK(vp);
2109 v_incr_usecount(vp);
2110 VI_UNLOCK(vp);
2111 }
2112
2113 /*
2114 * Return reference count of a vnode.
2115 *
2116 * The results of this call are only guaranteed when some mechanism other
2117 * than the VI lock is used to stop other processes from gaining references
2118 * to the vnode. This may be the case if the caller holds the only reference.
2119 * This is also useful when stale data is acceptable as race conditions may
2120 * be accounted for by some other means.
2121 */
2122 int
2123 vrefcnt(struct vnode *vp)
2124 {
2125 int usecnt;
2126
2127 VI_LOCK(vp);
2128 usecnt = vp->v_usecount;
2129 VI_UNLOCK(vp);
2130
2131 return (usecnt);
2132 }
2133
2134
2135 /*
2136 * Vnode put/release.
2137 * If count drops to zero, call inactive routine and return to freelist.
2138 */
2139 void
2140 vrele(struct vnode *vp)
2141 {
2142 struct thread *td = curthread; /* XXX */
2143
2144 KASSERT(vp != NULL, ("vrele: null vp"));
2145 VFS_ASSERT_GIANT(vp->v_mount);
2146
2147 VI_LOCK(vp);
2148
2149 /* Skip this v_writecount check if we're going to panic below. */
2150 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2151 ("vrele: missed vn_close"));
2152
2153 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2154 vp->v_usecount == 1)) {
2155 v_decr_usecount(vp);
2156 return;
2157 }
2158 if (vp->v_usecount != 1) {
2159 #ifdef DIAGNOSTIC
2160 vprint("vrele: negative ref count", vp);
2161 #endif
2162 VI_UNLOCK(vp);
2163 panic("vrele: negative ref cnt");
2164 }
2165 /*
2166 * We want to hold the vnode until the inactive finishes to
2167 * prevent vgone() races. We drop the use count here and the
2168 * hold count below when we're done.
2169 */
2170 v_decr_useonly(vp);
2171 /*
2172 * We must call VOP_INACTIVE with the node locked. Mark
2173 * as VI_DOINGINACT to avoid recursion.
2174 */
2175 vp->v_iflag |= VI_OWEINACT;
2176 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2177 VI_LOCK(vp);
2178 if (vp->v_usecount > 0)
2179 vp->v_iflag &= ~VI_OWEINACT;
2180 if (vp->v_iflag & VI_OWEINACT)
2181 vinactive(vp, td);
2182 VOP_UNLOCK(vp, 0, td);
2183 } else {
2184 VI_LOCK(vp);
2185 if (vp->v_usecount > 0)
2186 vp->v_iflag &= ~VI_OWEINACT;
2187 }
2188 vdropl(vp);
2189 }
2190
2191 /*
2192 * Release an already locked vnode. This give the same effects as
2193 * unlock+vrele(), but takes less time and avoids releasing and
2194 * re-aquiring the lock (as vrele() acquires the lock internally.)
2195 */
2196 void
2197 vput(struct vnode *vp)
2198 {
2199 struct thread *td = curthread; /* XXX */
2200 int error;
2201
2202 KASSERT(vp != NULL, ("vput: null vp"));
2203 ASSERT_VOP_LOCKED(vp, "vput");
2204 VFS_ASSERT_GIANT(vp->v_mount);
2205 VI_LOCK(vp);
2206 /* Skip this v_writecount check if we're going to panic below. */
2207 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2208 ("vput: missed vn_close"));
2209 error = 0;
2210
2211 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2212 vp->v_usecount == 1)) {
2213 VOP_UNLOCK(vp, 0, td);
2214 v_decr_usecount(vp);
2215 return;
2216 }
2217
2218 if (vp->v_usecount != 1) {
2219 #ifdef DIAGNOSTIC
2220 vprint("vput: negative ref count", vp);
2221 #endif
2222 panic("vput: negative ref cnt");
2223 }
2224 /*
2225 * We want to hold the vnode until the inactive finishes to
2226 * prevent vgone() races. We drop the use count here and the
2227 * hold count below when we're done.
2228 */
2229 v_decr_useonly(vp);
2230 vp->v_iflag |= VI_OWEINACT;
2231 if (VOP_ISLOCKED(vp, NULL) != LK_EXCLUSIVE) {
2232 error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT, td);
2233 VI_LOCK(vp);
2234 if (error) {
2235 if (vp->v_usecount > 0)
2236 vp->v_iflag &= ~VI_OWEINACT;
2237 goto done;
2238 }
2239 }
2240 if (vp->v_usecount > 0)
2241 vp->v_iflag &= ~VI_OWEINACT;
2242 if (vp->v_iflag & VI_OWEINACT)
2243 vinactive(vp, td);
2244 VOP_UNLOCK(vp, 0, td);
2245 done:
2246 vdropl(vp);
2247 }
2248
2249 /*
2250 * Somebody doesn't want the vnode recycled.
2251 */
2252 void
2253 vhold(struct vnode *vp)
2254 {
2255
2256 VI_LOCK(vp);
2257 vholdl(vp);
2258 VI_UNLOCK(vp);
2259 }
2260
2261 void
2262 vholdl(struct vnode *vp)
2263 {
2264
2265 vp->v_holdcnt++;
2266 if (VSHOULDBUSY(vp))
2267 vbusy(vp);
2268 }
2269
2270 /*
2271 * Note that there is one less who cares about this vnode. vdrop() is the
2272 * opposite of vhold().
2273 */
2274 void
2275 vdrop(struct vnode *vp)
2276 {
2277
2278 VI_LOCK(vp);
2279 vdropl(vp);
2280 }
2281
2282 /*
2283 * Drop the hold count of the vnode. If this is the last reference to
2284 * the vnode we will free it if it has been vgone'd otherwise it is
2285 * placed on the free list.
2286 */
2287 void
2288 vdropl(struct vnode *vp)
2289 {
2290
2291 ASSERT_VI_LOCKED(vp, "vdropl");
2292 if (vp->v_holdcnt <= 0)
2293 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2294 vp->v_holdcnt--;
2295 if (vp->v_holdcnt == 0) {
2296 if (vp->v_iflag & VI_DOOMED) {
2297 vdestroy(vp);
2298 return;
2299 } else
2300 vfree(vp);
2301 }
2302 VI_UNLOCK(vp);
2303 }
2304
2305 /*
2306 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2307 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2308 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2309 * failed lock upgrade.
2310 */
2311 static void
2312 vinactive(struct vnode *vp, struct thread *td)
2313 {
2314
2315 ASSERT_VOP_LOCKED(vp, "vinactive");
2316 ASSERT_VI_LOCKED(vp, "vinactive");
2317 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2318 ("vinactive: recursed on VI_DOINGINACT"));
2319 vp->v_iflag |= VI_DOINGINACT;
2320 vp->v_iflag &= ~VI_OWEINACT;
2321 VI_UNLOCK(vp);
2322 VOP_INACTIVE(vp, td);
2323 VI_LOCK(vp);
2324 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2325 ("vinactive: lost VI_DOINGINACT"));
2326 vp->v_iflag &= ~VI_DOINGINACT;
2327 }
2328
2329 /*
2330 * Remove any vnodes in the vnode table belonging to mount point mp.
2331 *
2332 * If FORCECLOSE is not specified, there should not be any active ones,
2333 * return error if any are found (nb: this is a user error, not a
2334 * system error). If FORCECLOSE is specified, detach any active vnodes
2335 * that are found.
2336 *
2337 * If WRITECLOSE is set, only flush out regular file vnodes open for
2338 * writing.
2339 *
2340 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2341 *
2342 * `rootrefs' specifies the base reference count for the root vnode
2343 * of this filesystem. The root vnode is considered busy if its
2344 * v_usecount exceeds this value. On a successful return, vflush(, td)
2345 * will call vrele() on the root vnode exactly rootrefs times.
2346 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2347 * be zero.
2348 */
2349 #ifdef DIAGNOSTIC
2350 static int busyprt = 0; /* print out busy vnodes */
2351 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2352 #endif
2353
2354 int
2355 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2356 {
2357 struct vnode *vp, *mvp, *rootvp = NULL;
2358 struct vattr vattr;
2359 int busy = 0, error;
2360
2361 CTR1(KTR_VFS, "vflush: mp %p", mp);
2362 if (rootrefs > 0) {
2363 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2364 ("vflush: bad args"));
2365 /*
2366 * Get the filesystem root vnode. We can vput() it
2367 * immediately, since with rootrefs > 0, it won't go away.
2368 */
2369 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2370 return (error);
2371 vput(rootvp);
2372
2373 }
2374 MNT_ILOCK(mp);
2375 loop:
2376 MNT_VNODE_FOREACH(vp, mp, mvp) {
2377
2378 VI_LOCK(vp);
2379 vholdl(vp);
2380 MNT_IUNLOCK(mp);
2381 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2382 if (error) {
2383 vdrop(vp);
2384 MNT_ILOCK(mp);
2385 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2386 goto loop;
2387 }
2388 /*
2389 * Skip over a vnodes marked VV_SYSTEM.
2390 */
2391 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2392 VOP_UNLOCK(vp, 0, td);
2393 vdrop(vp);
2394 MNT_ILOCK(mp);
2395 continue;
2396 }
2397 /*
2398 * If WRITECLOSE is set, flush out unlinked but still open
2399 * files (even if open only for reading) and regular file
2400 * vnodes open for writing.
2401 */
2402 if (flags & WRITECLOSE) {
2403 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2404 VI_LOCK(vp);
2405
2406 if ((vp->v_type == VNON ||
2407 (error == 0 && vattr.va_nlink > 0)) &&
2408 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2409 VOP_UNLOCK(vp, 0, td);
2410 vdropl(vp);
2411 MNT_ILOCK(mp);
2412 continue;
2413 }
2414 } else
2415 VI_LOCK(vp);
2416 /*
2417 * With v_usecount == 0, all we need to do is clear out the
2418 * vnode data structures and we are done.
2419 *
2420 * If FORCECLOSE is set, forcibly close the vnode.
2421 */
2422 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2423 VNASSERT(vp->v_usecount == 0 ||
2424 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2425 ("device VNODE %p is FORCECLOSED", vp));
2426 vgonel(vp);
2427 } else {
2428 busy++;
2429 #ifdef DIAGNOSTIC
2430 if (busyprt)
2431 vprint("vflush: busy vnode", vp);
2432 #endif
2433 }
2434 VOP_UNLOCK(vp, 0, td);
2435 vdropl(vp);
2436 MNT_ILOCK(mp);
2437 }
2438 MNT_IUNLOCK(mp);
2439 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2440 /*
2441 * If just the root vnode is busy, and if its refcount
2442 * is equal to `rootrefs', then go ahead and kill it.
2443 */
2444 VI_LOCK(rootvp);
2445 KASSERT(busy > 0, ("vflush: not busy"));
2446 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2447 ("vflush: usecount %d < rootrefs %d",
2448 rootvp->v_usecount, rootrefs));
2449 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2450 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK, td);
2451 vgone(rootvp);
2452 VOP_UNLOCK(rootvp, 0, td);
2453 busy = 0;
2454 } else
2455 VI_UNLOCK(rootvp);
2456 }
2457 if (busy)
2458 return (EBUSY);
2459 for (; rootrefs > 0; rootrefs--)
2460 vrele(rootvp);
2461 return (0);
2462 }
2463
2464 /*
2465 * Recycle an unused vnode to the front of the free list.
2466 */
2467 int
2468 vrecycle(struct vnode *vp, struct thread *td)
2469 {
2470 int recycled;
2471
2472 ASSERT_VOP_LOCKED(vp, "vrecycle");
2473 recycled = 0;
2474 VI_LOCK(vp);
2475 if (vp->v_usecount == 0) {
2476 recycled = 1;
2477 vgonel(vp);
2478 }
2479 VI_UNLOCK(vp);
2480 return (recycled);
2481 }
2482
2483 /*
2484 * Eliminate all activity associated with a vnode
2485 * in preparation for reuse.
2486 */
2487 void
2488 vgone(struct vnode *vp)
2489 {
2490 VI_LOCK(vp);
2491 vgonel(vp);
2492 VI_UNLOCK(vp);
2493 }
2494
2495 /*
2496 * vgone, with the vp interlock held.
2497 */
2498 void
2499 vgonel(struct vnode *vp)
2500 {
2501 struct thread *td;
2502 int oweinact;
2503 int active;
2504 struct mount *mp;
2505
2506 CTR1(KTR_VFS, "vgonel: vp %p", vp);
2507 ASSERT_VOP_LOCKED(vp, "vgonel");
2508 ASSERT_VI_LOCKED(vp, "vgonel");
2509 VNASSERT(vp->v_holdcnt, vp,
2510 ("vgonel: vp %p has no reference.", vp));
2511 td = curthread;
2512
2513 /*
2514 * Don't vgonel if we're already doomed.
2515 */
2516 if (vp->v_iflag & VI_DOOMED)
2517 return;
2518 vp->v_iflag |= VI_DOOMED;
2519 /*
2520 * Check to see if the vnode is in use. If so, we have to call
2521 * VOP_CLOSE() and VOP_INACTIVE().
2522 */
2523 active = vp->v_usecount;
2524 oweinact = (vp->v_iflag & VI_OWEINACT);
2525 VI_UNLOCK(vp);
2526 /*
2527 * Clean out any buffers associated with the vnode.
2528 * If the flush fails, just toss the buffers.
2529 */
2530 mp = NULL;
2531 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2532 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2533 if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2534 vinvalbuf(vp, 0, td, 0, 0);
2535
2536 /*
2537 * If purging an active vnode, it must be closed and
2538 * deactivated before being reclaimed.
2539 */
2540 if (active)
2541 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2542 if (oweinact || active) {
2543 VI_LOCK(vp);
2544 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2545 vinactive(vp, td);
2546 VI_UNLOCK(vp);
2547 }
2548 /*
2549 * Reclaim the vnode.
2550 */
2551 if (VOP_RECLAIM(vp, td))
2552 panic("vgone: cannot reclaim");
2553 if (mp != NULL)
2554 vn_finished_secondary_write(mp);
2555 VNASSERT(vp->v_object == NULL, vp,
2556 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2557 /*
2558 * Clear the advisory locks and wake up waiting threads.
2559 */
2560 lf_purgelocks(vp, &(vp->v_lockf));
2561 /*
2562 * Delete from old mount point vnode list.
2563 */
2564 delmntque(vp);
2565 cache_purge(vp);
2566 /*
2567 * Done with purge, reset to the standard lock and invalidate
2568 * the vnode.
2569 */
2570 VI_LOCK(vp);
2571 vp->v_vnlock = &vp->v_lock;
2572 vp->v_op = &dead_vnodeops;
2573 vp->v_tag = "none";
2574 vp->v_type = VBAD;
2575 }
2576
2577 /*
2578 * Calculate the total number of references to a special device.
2579 */
2580 int
2581 vcount(struct vnode *vp)
2582 {
2583 int count;
2584
2585 dev_lock();
2586 count = vp->v_rdev->si_usecount;
2587 dev_unlock();
2588 return (count);
2589 }
2590
2591 /*
2592 * Same as above, but using the struct cdev *as argument
2593 */
2594 int
2595 count_dev(struct cdev *dev)
2596 {
2597 int count;
2598
2599 dev_lock();
2600 count = dev->si_usecount;
2601 dev_unlock();
2602 return(count);
2603 }
2604
2605 /*
2606 * Print out a description of a vnode.
2607 */
2608 static char *typename[] =
2609 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2610 "VMARKER"};
2611
2612 void
2613 vn_printf(struct vnode *vp, const char *fmt, ...)
2614 {
2615 va_list ap;
2616 char buf[256], buf2[16];
2617 u_long flags;
2618
2619 va_start(ap, fmt);
2620 vprintf(fmt, ap);
2621 va_end(ap);
2622 printf("%p: ", (void *)vp);
2623 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2624 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2625 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2626 buf[0] = '\0';
2627 buf[1] = '\0';
2628 if (vp->v_vflag & VV_ROOT)
2629 strlcat(buf, "|VV_ROOT", sizeof(buf));
2630 if (vp->v_vflag & VV_ISTTY)
2631 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2632 if (vp->v_vflag & VV_NOSYNC)
2633 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2634 if (vp->v_vflag & VV_CACHEDLABEL)
2635 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2636 if (vp->v_vflag & VV_TEXT)
2637 strlcat(buf, "|VV_TEXT", sizeof(buf));
2638 if (vp->v_vflag & VV_COPYONWRITE)
2639 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2640 if (vp->v_vflag & VV_SYSTEM)
2641 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2642 if (vp->v_vflag & VV_PROCDEP)
2643 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2644 if (vp->v_vflag & VV_NOKNOTE)
2645 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2646 if (vp->v_vflag & VV_DELETED)
2647 strlcat(buf, "|VV_DELETED", sizeof(buf));
2648 if (vp->v_vflag & VV_MD)
2649 strlcat(buf, "|VV_MD", sizeof(buf));
2650 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2651 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2652 VV_NOKNOTE | VV_DELETED | VV_MD);
2653 if (flags != 0) {
2654 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2655 strlcat(buf, buf2, sizeof(buf));
2656 }
2657 if (vp->v_iflag & VI_MOUNT)
2658 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2659 if (vp->v_iflag & VI_AGE)
2660 strlcat(buf, "|VI_AGE", sizeof(buf));
2661 if (vp->v_iflag & VI_DOOMED)
2662 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2663 if (vp->v_iflag & VI_FREE)
2664 strlcat(buf, "|VI_FREE", sizeof(buf));
2665 if (vp->v_iflag & VI_OBJDIRTY)
2666 strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2667 if (vp->v_iflag & VI_DOINGINACT)
2668 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2669 if (vp->v_iflag & VI_OWEINACT)
2670 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2671 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2672 VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2673 if (flags != 0) {
2674 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2675 strlcat(buf, buf2, sizeof(buf));
2676 }
2677 printf(" flags (%s)\n", buf + 1);
2678 if (mtx_owned(VI_MTX(vp)))
2679 printf(" VI_LOCKed");
2680 if (vp->v_object != NULL)
2681 printf(" v_object %p ref %d pages %d\n",
2682 vp->v_object, vp->v_object->ref_count,
2683 vp->v_object->resident_page_count);
2684 printf(" ");
2685 lockmgr_printinfo(vp->v_vnlock);
2686 printf("\n");
2687 if (vp->v_data != NULL)
2688 VOP_PRINT(vp);
2689 }
2690
2691 #ifdef DDB
2692 /*
2693 * List all of the locked vnodes in the system.
2694 * Called when debugging the kernel.
2695 */
2696 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2697 {
2698 struct mount *mp, *nmp;
2699 struct vnode *vp;
2700
2701 /*
2702 * Note: because this is DDB, we can't obey the locking semantics
2703 * for these structures, which means we could catch an inconsistent
2704 * state and dereference a nasty pointer. Not much to be done
2705 * about that.
2706 */
2707 db_printf("Locked vnodes\n");
2708 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2709 nmp = TAILQ_NEXT(mp, mnt_list);
2710 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2711 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp, NULL))
2712 vprint("", vp);
2713 }
2714 nmp = TAILQ_NEXT(mp, mnt_list);
2715 }
2716 }
2717
2718 /*
2719 * Show details about the given vnode.
2720 */
2721 DB_SHOW_COMMAND(vnode, db_show_vnode)
2722 {
2723 struct vnode *vp;
2724
2725 if (!have_addr)
2726 return;
2727 vp = (struct vnode *)addr;
2728 vn_printf(vp, "vnode ");
2729 }
2730
2731 /*
2732 * Show details about the given mount point.
2733 */
2734 DB_SHOW_COMMAND(mount, db_show_mount)
2735 {
2736 struct mount *mp;
2737 struct statfs *sp;
2738 struct vnode *vp;
2739 char buf[512];
2740 u_int flags;
2741
2742 if (!have_addr) {
2743 /* No address given, print short info about all mount points. */
2744 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2745 db_printf("%p %s on %s (%s)\n", mp,
2746 mp->mnt_stat.f_mntfromname,
2747 mp->mnt_stat.f_mntonname,
2748 mp->mnt_stat.f_fstypename);
2749 if (db_pager_quit)
2750 break;
2751 }
2752 db_printf("\nMore info: show mount <addr>\n");
2753 return;
2754 }
2755
2756 mp = (struct mount *)addr;
2757 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2758 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2759
2760 buf[0] = '\0';
2761 flags = mp->mnt_flag;
2762 #define MNT_FLAG(flag) do { \
2763 if (flags & (flag)) { \
2764 if (buf[0] != '\0') \
2765 strlcat(buf, ", ", sizeof(buf)); \
2766 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2767 flags &= ~(flag); \
2768 } \
2769 } while (0)
2770 MNT_FLAG(MNT_RDONLY);
2771 MNT_FLAG(MNT_SYNCHRONOUS);
2772 MNT_FLAG(MNT_NOEXEC);
2773 MNT_FLAG(MNT_NOSUID);
2774 MNT_FLAG(MNT_UNION);
2775 MNT_FLAG(MNT_ASYNC);
2776 MNT_FLAG(MNT_SUIDDIR);
2777 MNT_FLAG(MNT_SOFTDEP);
2778 MNT_FLAG(MNT_NOSYMFOLLOW);
2779 MNT_FLAG(MNT_GJOURNAL);
2780 MNT_FLAG(MNT_MULTILABEL);
2781 MNT_FLAG(MNT_ACLS);
2782 MNT_FLAG(MNT_NOATIME);
2783 MNT_FLAG(MNT_NOCLUSTERR);
2784 MNT_FLAG(MNT_NOCLUSTERW);
2785 MNT_FLAG(MNT_EXRDONLY);
2786 MNT_FLAG(MNT_EXPORTED);
2787 MNT_FLAG(MNT_DEFEXPORTED);
2788 MNT_FLAG(MNT_EXPORTANON);
2789 MNT_FLAG(MNT_EXKERB);
2790 MNT_FLAG(MNT_EXPUBLIC);
2791 MNT_FLAG(MNT_LOCAL);
2792 MNT_FLAG(MNT_QUOTA);
2793 MNT_FLAG(MNT_ROOTFS);
2794 MNT_FLAG(MNT_USER);
2795 MNT_FLAG(MNT_IGNORE);
2796 MNT_FLAG(MNT_UPDATE);
2797 MNT_FLAG(MNT_DELEXPORT);
2798 MNT_FLAG(MNT_RELOAD);
2799 MNT_FLAG(MNT_FORCE);
2800 MNT_FLAG(MNT_SNAPSHOT);
2801 MNT_FLAG(MNT_BYFSID);
2802 #undef MNT_FLAG
2803 if (flags != 0) {
2804 if (buf[0] != '\0')
2805 strlcat(buf, ", ", sizeof(buf));
2806 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2807 "0x%08x", flags);
2808 }
2809 db_printf(" mnt_flag = %s\n", buf);
2810
2811 buf[0] = '\0';
2812 flags = mp->mnt_kern_flag;
2813 #define MNT_KERN_FLAG(flag) do { \
2814 if (flags & (flag)) { \
2815 if (buf[0] != '\0') \
2816 strlcat(buf, ", ", sizeof(buf)); \
2817 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2818 flags &= ~(flag); \
2819 } \
2820 } while (0)
2821 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2822 MNT_KERN_FLAG(MNTK_ASYNC);
2823 MNT_KERN_FLAG(MNTK_SOFTDEP);
2824 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2825 MNT_KERN_FLAG(MNTK_UNMOUNT);
2826 MNT_KERN_FLAG(MNTK_MWAIT);
2827 MNT_KERN_FLAG(MNTK_SUSPEND);
2828 MNT_KERN_FLAG(MNTK_SUSPEND2);
2829 MNT_KERN_FLAG(MNTK_SUSPENDED);
2830 MNT_KERN_FLAG(MNTK_MPSAFE);
2831 MNT_KERN_FLAG(MNTK_NOKNOTE);
2832 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2833 #undef MNT_KERN_FLAG
2834 if (flags != 0) {
2835 if (buf[0] != '\0')
2836 strlcat(buf, ", ", sizeof(buf));
2837 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2838 "0x%08x", flags);
2839 }
2840 db_printf(" mnt_kern_flag = %s\n", buf);
2841
2842 sp = &mp->mnt_stat;
2843 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2844 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2845 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2846 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2847 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2848 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2849 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2850 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2851 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2852 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2853 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2854 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2855
2856 db_printf(" mnt_cred = { uid=%u ruid=%u",
2857 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2858 if (mp->mnt_cred->cr_prison != NULL)
2859 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2860 db_printf(" }\n");
2861 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2862 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2863 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2864 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2865 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
2866 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2867 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2868 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2869 db_printf(" mnt_markercnt = %d\n", mp->mnt_markercnt);
2870 db_printf(" mnt_holdcnt = %d\n", mp->mnt_holdcnt);
2871 db_printf(" mnt_holdcntwaiters = %d\n", mp->mnt_holdcntwaiters);
2872 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2873 db_printf(" mnt_secondary_accwrites = %d\n",
2874 mp->mnt_secondary_accwrites);
2875 db_printf(" mnt_gjprovider = %s\n",
2876 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2877 db_printf("\n");
2878
2879 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2880 if (vp->v_type != VMARKER) {
2881 vn_printf(vp, "vnode ");
2882 if (db_pager_quit)
2883 break;
2884 }
2885 }
2886 }
2887 #endif /* DDB */
2888
2889 /*
2890 * Fill in a struct xvfsconf based on a struct vfsconf.
2891 */
2892 static void
2893 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2894 {
2895
2896 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2897 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2898 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2899 xvfsp->vfc_flags = vfsp->vfc_flags;
2900 /*
2901 * These are unused in userland, we keep them
2902 * to not break binary compatibility.
2903 */
2904 xvfsp->vfc_vfsops = NULL;
2905 xvfsp->vfc_next = NULL;
2906 }
2907
2908 /*
2909 * Top level filesystem related information gathering.
2910 */
2911 static int
2912 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2913 {
2914 struct vfsconf *vfsp;
2915 struct xvfsconf xvfsp;
2916 int error;
2917
2918 error = 0;
2919 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2920 bzero(&xvfsp, sizeof(xvfsp));
2921 vfsconf2x(vfsp, &xvfsp);
2922 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2923 if (error)
2924 break;
2925 }
2926 return (error);
2927 }
2928
2929 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2930 "S,xvfsconf", "List of all configured filesystems");
2931
2932 #ifndef BURN_BRIDGES
2933 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2934
2935 static int
2936 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2937 {
2938 int *name = (int *)arg1 - 1; /* XXX */
2939 u_int namelen = arg2 + 1; /* XXX */
2940 struct vfsconf *vfsp;
2941 struct xvfsconf xvfsp;
2942
2943 printf("WARNING: userland calling deprecated sysctl, "
2944 "please rebuild world\n");
2945
2946 #if 1 || defined(COMPAT_PRELITE2)
2947 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2948 if (namelen == 1)
2949 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2950 #endif
2951
2952 switch (name[1]) {
2953 case VFS_MAXTYPENUM:
2954 if (namelen != 2)
2955 return (ENOTDIR);
2956 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2957 case VFS_CONF:
2958 if (namelen != 3)
2959 return (ENOTDIR); /* overloaded */
2960 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2961 if (vfsp->vfc_typenum == name[2])
2962 break;
2963 if (vfsp == NULL)
2964 return (EOPNOTSUPP);
2965 bzero(&xvfsp, sizeof(xvfsp));
2966 vfsconf2x(vfsp, &xvfsp);
2967 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2968 }
2969 return (EOPNOTSUPP);
2970 }
2971
2972 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2973 vfs_sysctl, "Generic filesystem");
2974
2975 #if 1 || defined(COMPAT_PRELITE2)
2976
2977 static int
2978 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2979 {
2980 int error;
2981 struct vfsconf *vfsp;
2982 struct ovfsconf ovfs;
2983
2984 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2985 bzero(&ovfs, sizeof(ovfs));
2986 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2987 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2988 ovfs.vfc_index = vfsp->vfc_typenum;
2989 ovfs.vfc_refcount = vfsp->vfc_refcount;
2990 ovfs.vfc_flags = vfsp->vfc_flags;
2991 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2992 if (error)
2993 return error;
2994 }
2995 return 0;
2996 }
2997
2998 #endif /* 1 || COMPAT_PRELITE2 */
2999 #endif /* !BURN_BRIDGES */
3000
3001 #define KINFO_VNODESLOP 10
3002 #ifdef notyet
3003 /*
3004 * Dump vnode list (via sysctl).
3005 */
3006 /* ARGSUSED */
3007 static int
3008 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3009 {
3010 struct xvnode *xvn;
3011 struct thread *td = req->td;
3012 struct mount *mp;
3013 struct vnode *vp;
3014 int error, len, n;
3015
3016 /*
3017 * Stale numvnodes access is not fatal here.
3018 */
3019 req->lock = 0;
3020 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3021 if (!req->oldptr)
3022 /* Make an estimate */
3023 return (SYSCTL_OUT(req, 0, len));
3024
3025 error = sysctl_wire_old_buffer(req, 0);
3026 if (error != 0)
3027 return (error);
3028 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3029 n = 0;
3030 mtx_lock(&mountlist_mtx);
3031 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3032 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3033 continue;
3034 MNT_ILOCK(mp);
3035 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3036 if (n == len)
3037 break;
3038 vref(vp);
3039 xvn[n].xv_size = sizeof *xvn;
3040 xvn[n].xv_vnode = vp;
3041 xvn[n].xv_id = 0; /* XXX compat */
3042 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3043 XV_COPY(usecount);
3044 XV_COPY(writecount);
3045 XV_COPY(holdcnt);
3046 XV_COPY(mount);
3047 XV_COPY(numoutput);
3048 XV_COPY(type);
3049 #undef XV_COPY
3050 xvn[n].xv_flag = vp->v_vflag;
3051
3052 switch (vp->v_type) {
3053 case VREG:
3054 case VDIR:
3055 case VLNK:
3056 break;
3057 case VBLK:
3058 case VCHR:
3059 if (vp->v_rdev == NULL) {
3060 vrele(vp);
3061 continue;
3062 }
3063 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3064 break;
3065 case VSOCK:
3066 xvn[n].xv_socket = vp->v_socket;
3067 break;
3068 case VFIFO:
3069 xvn[n].xv_fifo = vp->v_fifoinfo;
3070 break;
3071 case VNON:
3072 case VBAD:
3073 default:
3074 /* shouldn't happen? */
3075 vrele(vp);
3076 continue;
3077 }
3078 vrele(vp);
3079 ++n;
3080 }
3081 MNT_IUNLOCK(mp);
3082 mtx_lock(&mountlist_mtx);
3083 vfs_unbusy(mp, td);
3084 if (n == len)
3085 break;
3086 }
3087 mtx_unlock(&mountlist_mtx);
3088
3089 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3090 free(xvn, M_TEMP);
3091 return (error);
3092 }
3093
3094 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3095 0, 0, sysctl_vnode, "S,xvnode", "");
3096 #endif
3097
3098 /*
3099 * Unmount all filesystems. The list is traversed in reverse order
3100 * of mounting to avoid dependencies.
3101 */
3102 void
3103 vfs_unmountall(void)
3104 {
3105 struct mount *mp;
3106 struct thread *td;
3107 int error;
3108
3109 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3110 td = curthread;
3111 /*
3112 * Since this only runs when rebooting, it is not interlocked.
3113 */
3114 while(!TAILQ_EMPTY(&mountlist)) {
3115 mp = TAILQ_LAST(&mountlist, mntlist);
3116 error = dounmount(mp, MNT_FORCE, td);
3117 if (error) {
3118 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3119 /*
3120 * XXX: Due to the way in which we mount the root
3121 * file system off of devfs, devfs will generate a
3122 * "busy" warning when we try to unmount it before
3123 * the root. Don't print a warning as a result in
3124 * order to avoid false positive errors that may
3125 * cause needless upset.
3126 */
3127 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3128 printf("unmount of %s failed (",
3129 mp->mnt_stat.f_mntonname);
3130 if (error == EBUSY)
3131 printf("BUSY)\n");
3132 else
3133 printf("%d)\n", error);
3134 }
3135 } else {
3136 /* The unmount has removed mp from the mountlist */
3137 }
3138 }
3139 }
3140
3141 /*
3142 * perform msync on all vnodes under a mount point
3143 * the mount point must be locked.
3144 */
3145 void
3146 vfs_msync(struct mount *mp, int flags)
3147 {
3148 struct vnode *vp, *mvp;
3149 struct vm_object *obj;
3150
3151 MNT_ILOCK(mp);
3152 MNT_VNODE_FOREACH(vp, mp, mvp) {
3153 VI_LOCK(vp);
3154 if ((vp->v_iflag & VI_OBJDIRTY) &&
3155 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3156 MNT_IUNLOCK(mp);
3157 if (!vget(vp,
3158 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3159 curthread)) {
3160 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3161 vput(vp);
3162 MNT_ILOCK(mp);
3163 continue;
3164 }
3165
3166 obj = vp->v_object;
3167 if (obj != NULL) {
3168 VM_OBJECT_LOCK(obj);
3169 vm_object_page_clean(obj, 0, 0,
3170 flags == MNT_WAIT ?
3171 OBJPC_SYNC : OBJPC_NOSYNC);
3172 VM_OBJECT_UNLOCK(obj);
3173 }
3174 vput(vp);
3175 }
3176 MNT_ILOCK(mp);
3177 } else
3178 VI_UNLOCK(vp);
3179 }
3180 MNT_IUNLOCK(mp);
3181 }
3182
3183 /*
3184 * Mark a vnode as free, putting it up for recycling.
3185 */
3186 static void
3187 vfree(struct vnode *vp)
3188 {
3189
3190 CTR1(KTR_VFS, "vfree vp %p", vp);
3191 ASSERT_VI_LOCKED(vp, "vfree");
3192 mtx_lock(&vnode_free_list_mtx);
3193 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3194 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3195 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3196 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3197 ("vfree: Freeing doomed vnode"));
3198 if (vp->v_iflag & VI_AGE) {
3199 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3200 } else {
3201 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3202 }
3203 freevnodes++;
3204 vp->v_iflag &= ~VI_AGE;
3205 vp->v_iflag |= VI_FREE;
3206 mtx_unlock(&vnode_free_list_mtx);
3207 }
3208
3209 /*
3210 * Opposite of vfree() - mark a vnode as in use.
3211 */
3212 static void
3213 vbusy(struct vnode *vp)
3214 {
3215 CTR1(KTR_VFS, "vbusy vp %p", vp);
3216 ASSERT_VI_LOCKED(vp, "vbusy");
3217 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3218 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3219
3220 mtx_lock(&vnode_free_list_mtx);
3221 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3222 freevnodes--;
3223 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3224 mtx_unlock(&vnode_free_list_mtx);
3225 }
3226
3227 static void
3228 destroy_vpollinfo(struct vpollinfo *vi)
3229 {
3230 knlist_destroy(&vi->vpi_selinfo.si_note);
3231 mtx_destroy(&vi->vpi_lock);
3232 uma_zfree(vnodepoll_zone, vi);
3233 }
3234
3235 /*
3236 * Initalize per-vnode helper structure to hold poll-related state.
3237 */
3238 void
3239 v_addpollinfo(struct vnode *vp)
3240 {
3241 struct vpollinfo *vi;
3242
3243 if (vp->v_pollinfo != NULL)
3244 return;
3245 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3246 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3247 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3248 vfs_knlunlock, NULL);
3249 VI_LOCK(vp);
3250 if (vp->v_pollinfo != NULL) {
3251 VI_UNLOCK(vp);
3252 destroy_vpollinfo(vi);
3253 return;
3254 }
3255 vp->v_pollinfo = vi;
3256 VI_UNLOCK(vp);
3257 }
3258
3259 /*
3260 * Record a process's interest in events which might happen to
3261 * a vnode. Because poll uses the historic select-style interface
3262 * internally, this routine serves as both the ``check for any
3263 * pending events'' and the ``record my interest in future events''
3264 * functions. (These are done together, while the lock is held,
3265 * to avoid race conditions.)
3266 */
3267 int
3268 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3269 {
3270
3271 v_addpollinfo(vp);
3272 mtx_lock(&vp->v_pollinfo->vpi_lock);
3273 if (vp->v_pollinfo->vpi_revents & events) {
3274 /*
3275 * This leaves events we are not interested
3276 * in available for the other process which
3277 * which presumably had requested them
3278 * (otherwise they would never have been
3279 * recorded).
3280 */
3281 events &= vp->v_pollinfo->vpi_revents;
3282 vp->v_pollinfo->vpi_revents &= ~events;
3283
3284 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3285 return events;
3286 }
3287 vp->v_pollinfo->vpi_events |= events;
3288 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3289 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3290 return 0;
3291 }
3292
3293 /*
3294 * Routine to create and manage a filesystem syncer vnode.
3295 */
3296 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3297 static int sync_fsync(struct vop_fsync_args *);
3298 static int sync_inactive(struct vop_inactive_args *);
3299 static int sync_reclaim(struct vop_reclaim_args *);
3300
3301 static struct vop_vector sync_vnodeops = {
3302 .vop_bypass = VOP_EOPNOTSUPP,
3303 .vop_close = sync_close, /* close */
3304 .vop_fsync = sync_fsync, /* fsync */
3305 .vop_inactive = sync_inactive, /* inactive */
3306 .vop_reclaim = sync_reclaim, /* reclaim */
3307 .vop_lock1 = vop_stdlock, /* lock */
3308 .vop_unlock = vop_stdunlock, /* unlock */
3309 .vop_islocked = vop_stdislocked, /* islocked */
3310 };
3311
3312 /*
3313 * Create a new filesystem syncer vnode for the specified mount point.
3314 */
3315 int
3316 vfs_allocate_syncvnode(struct mount *mp)
3317 {
3318 struct vnode *vp;
3319 static long start, incr, next;
3320 int error;
3321
3322 /* Allocate a new vnode */
3323 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3324 mp->mnt_syncer = NULL;
3325 return (error);
3326 }
3327 vp->v_type = VNON;
3328 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
3329 vp->v_vflag |= VV_FORCEINSMQ;
3330 error = insmntque(vp, mp);
3331 if (error != 0)
3332 panic("vfs_allocate_syncvnode: insmntque failed");
3333 vp->v_vflag &= ~VV_FORCEINSMQ;
3334 VOP_UNLOCK(vp, 0, curthread);
3335 /*
3336 * Place the vnode onto the syncer worklist. We attempt to
3337 * scatter them about on the list so that they will go off
3338 * at evenly distributed times even if all the filesystems
3339 * are mounted at once.
3340 */
3341 next += incr;
3342 if (next == 0 || next > syncer_maxdelay) {
3343 start /= 2;
3344 incr /= 2;
3345 if (start == 0) {
3346 start = syncer_maxdelay / 2;
3347 incr = syncer_maxdelay;
3348 }
3349 next = start;
3350 }
3351 VI_LOCK(vp);
3352 vn_syncer_add_to_worklist(&vp->v_bufobj,
3353 syncdelay > 0 ? next % syncdelay : 0);
3354 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3355 mtx_lock(&sync_mtx);
3356 sync_vnode_count++;
3357 mtx_unlock(&sync_mtx);
3358 VI_UNLOCK(vp);
3359 mp->mnt_syncer = vp;
3360 return (0);
3361 }
3362
3363 /*
3364 * Do a lazy sync of the filesystem.
3365 */
3366 static int
3367 sync_fsync(struct vop_fsync_args *ap)
3368 {
3369 struct vnode *syncvp = ap->a_vp;
3370 struct mount *mp = syncvp->v_mount;
3371 struct thread *td = ap->a_td;
3372 int error;
3373 struct bufobj *bo;
3374
3375 /*
3376 * We only need to do something if this is a lazy evaluation.
3377 */
3378 if (ap->a_waitfor != MNT_LAZY)
3379 return (0);
3380
3381 /*
3382 * Move ourselves to the back of the sync list.
3383 */
3384 bo = &syncvp->v_bufobj;
3385 BO_LOCK(bo);
3386 vn_syncer_add_to_worklist(bo, syncdelay);
3387 BO_UNLOCK(bo);
3388
3389 /*
3390 * Walk the list of vnodes pushing all that are dirty and
3391 * not already on the sync list.
3392 */
3393 mtx_lock(&mountlist_mtx);
3394 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3395 mtx_unlock(&mountlist_mtx);
3396 return (0);
3397 }
3398 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3399 vfs_unbusy(mp, td);
3400 return (0);
3401 }
3402 MNT_ILOCK(mp);
3403 mp->mnt_noasync++;
3404 mp->mnt_kern_flag &= ~MNTK_ASYNC;
3405 MNT_IUNLOCK(mp);
3406 vfs_msync(mp, MNT_NOWAIT);
3407 error = VFS_SYNC(mp, MNT_LAZY, td);
3408 MNT_ILOCK(mp);
3409 mp->mnt_noasync--;
3410 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3411 mp->mnt_kern_flag |= MNTK_ASYNC;
3412 MNT_IUNLOCK(mp);
3413 vn_finished_write(mp);
3414 vfs_unbusy(mp, td);
3415 return (error);
3416 }
3417
3418 /*
3419 * The syncer vnode is no referenced.
3420 */
3421 static int
3422 sync_inactive(struct vop_inactive_args *ap)
3423 {
3424
3425 vgone(ap->a_vp);
3426 return (0);
3427 }
3428
3429 /*
3430 * The syncer vnode is no longer needed and is being decommissioned.
3431 *
3432 * Modifications to the worklist must be protected by sync_mtx.
3433 */
3434 static int
3435 sync_reclaim(struct vop_reclaim_args *ap)
3436 {
3437 struct vnode *vp = ap->a_vp;
3438 struct bufobj *bo;
3439
3440 VI_LOCK(vp);
3441 bo = &vp->v_bufobj;
3442 vp->v_mount->mnt_syncer = NULL;
3443 if (bo->bo_flag & BO_ONWORKLST) {
3444 mtx_lock(&sync_mtx);
3445 LIST_REMOVE(bo, bo_synclist);
3446 syncer_worklist_len--;
3447 sync_vnode_count--;
3448 mtx_unlock(&sync_mtx);
3449 bo->bo_flag &= ~BO_ONWORKLST;
3450 }
3451 VI_UNLOCK(vp);
3452
3453 return (0);
3454 }
3455
3456 /*
3457 * Check if vnode represents a disk device
3458 */
3459 int
3460 vn_isdisk(struct vnode *vp, int *errp)
3461 {
3462 int error;
3463
3464 error = 0;
3465 dev_lock();
3466 if (vp->v_type != VCHR)
3467 error = ENOTBLK;
3468 else if (vp->v_rdev == NULL)
3469 error = ENXIO;
3470 else if (vp->v_rdev->si_devsw == NULL)
3471 error = ENXIO;
3472 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3473 error = ENOTBLK;
3474 dev_unlock();
3475 if (errp != NULL)
3476 *errp = error;
3477 return (error == 0);
3478 }
3479
3480 /*
3481 * Common filesystem object access control check routine. Accepts a
3482 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3483 * and optional call-by-reference privused argument allowing vaccess()
3484 * to indicate to the caller whether privilege was used to satisfy the
3485 * request (obsoleted). Returns 0 on success, or an errno on failure.
3486 *
3487 * The ifdef'd CAPABILITIES version is here for reference, but is not
3488 * actually used.
3489 */
3490 int
3491 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3492 mode_t acc_mode, struct ucred *cred, int *privused)
3493 {
3494 mode_t dac_granted;
3495 mode_t priv_granted;
3496
3497 /*
3498 * Look for a normal, non-privileged way to access the file/directory
3499 * as requested. If it exists, go with that.
3500 */
3501
3502 if (privused != NULL)
3503 *privused = 0;
3504
3505 dac_granted = 0;
3506
3507 /* Check the owner. */
3508 if (cred->cr_uid == file_uid) {
3509 dac_granted |= VADMIN;
3510 if (file_mode & S_IXUSR)
3511 dac_granted |= VEXEC;
3512 if (file_mode & S_IRUSR)
3513 dac_granted |= VREAD;
3514 if (file_mode & S_IWUSR)
3515 dac_granted |= (VWRITE | VAPPEND);
3516
3517 if ((acc_mode & dac_granted) == acc_mode)
3518 return (0);
3519
3520 goto privcheck;
3521 }
3522
3523 /* Otherwise, check the groups (first match) */
3524 if (groupmember(file_gid, cred)) {
3525 if (file_mode & S_IXGRP)
3526 dac_granted |= VEXEC;
3527 if (file_mode & S_IRGRP)
3528 dac_granted |= VREAD;
3529 if (file_mode & S_IWGRP)
3530 dac_granted |= (VWRITE | VAPPEND);
3531
3532 if ((acc_mode & dac_granted) == acc_mode)
3533 return (0);
3534
3535 goto privcheck;
3536 }
3537
3538 /* Otherwise, check everyone else. */
3539 if (file_mode & S_IXOTH)
3540 dac_granted |= VEXEC;
3541 if (file_mode & S_IROTH)
3542 dac_granted |= VREAD;
3543 if (file_mode & S_IWOTH)
3544 dac_granted |= (VWRITE | VAPPEND);
3545 if ((acc_mode & dac_granted) == acc_mode)
3546 return (0);
3547
3548 privcheck:
3549 /*
3550 * Build a privilege mask to determine if the set of privileges
3551 * satisfies the requirements when combined with the granted mask
3552 * from above. For each privilege, if the privilege is required,
3553 * bitwise or the request type onto the priv_granted mask.
3554 */
3555 priv_granted = 0;
3556
3557 if (type == VDIR) {
3558 /*
3559 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3560 * requests, instead of PRIV_VFS_EXEC.
3561 */
3562 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3563 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3564 priv_granted |= VEXEC;
3565 } else {
3566 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3567 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3568 priv_granted |= VEXEC;
3569 }
3570
3571 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3572 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3573 priv_granted |= VREAD;
3574
3575 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3576 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3577 priv_granted |= (VWRITE | VAPPEND);
3578
3579 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3580 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3581 priv_granted |= VADMIN;
3582
3583 if ((acc_mode & (priv_granted | dac_granted)) == acc_mode) {
3584 /* XXX audit: privilege used */
3585 if (privused != NULL)
3586 *privused = 1;
3587 return (0);
3588 }
3589
3590 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3591 }
3592
3593 /*
3594 * Credential check based on process requesting service, and per-attribute
3595 * permissions.
3596 */
3597 int
3598 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3599 struct thread *td, int access)
3600 {
3601
3602 /*
3603 * Kernel-invoked always succeeds.
3604 */
3605 if (cred == NOCRED)
3606 return (0);
3607
3608 /*
3609 * Do not allow privileged processes in jail to directly manipulate
3610 * system attributes.
3611 */
3612 switch (attrnamespace) {
3613 case EXTATTR_NAMESPACE_SYSTEM:
3614 /* Potentially should be: return (EPERM); */
3615 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3616 case EXTATTR_NAMESPACE_USER:
3617 return (VOP_ACCESS(vp, access, cred, td));
3618 default:
3619 return (EPERM);
3620 }
3621 }
3622
3623 #ifdef DEBUG_VFS_LOCKS
3624 /*
3625 * This only exists to supress warnings from unlocked specfs accesses. It is
3626 * no longer ok to have an unlocked VFS.
3627 */
3628 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3629
3630 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3631 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3632
3633 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3634 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3635
3636 int vfs_badlock_print = 1; /* Print lock violations. */
3637 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3638
3639 #ifdef KDB
3640 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3641 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3642 #endif
3643
3644 static void
3645 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3646 {
3647
3648 #ifdef KDB
3649 if (vfs_badlock_backtrace)
3650 kdb_backtrace();
3651 #endif
3652 if (vfs_badlock_print)
3653 printf("%s: %p %s\n", str, (void *)vp, msg);
3654 if (vfs_badlock_ddb)
3655 kdb_enter_why(KDB_WHY_VFSLOCK, "lock violation");
3656 }
3657
3658 void
3659 assert_vi_locked(struct vnode *vp, const char *str)
3660 {
3661
3662 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3663 vfs_badlock("interlock is not locked but should be", str, vp);
3664 }
3665
3666 void
3667 assert_vi_unlocked(struct vnode *vp, const char *str)
3668 {
3669
3670 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3671 vfs_badlock("interlock is locked but should not be", str, vp);
3672 }
3673
3674 void
3675 assert_vop_locked(struct vnode *vp, const char *str)
3676 {
3677
3678 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
3679 vfs_badlock("is not locked but should be", str, vp);
3680 }
3681
3682 void
3683 assert_vop_unlocked(struct vnode *vp, const char *str)
3684 {
3685
3686 if (vp && !IGNORE_LOCK(vp) &&
3687 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
3688 vfs_badlock("is locked but should not be", str, vp);
3689 }
3690
3691 void
3692 assert_vop_elocked(struct vnode *vp, const char *str)
3693 {
3694
3695 if (vp && !IGNORE_LOCK(vp) &&
3696 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
3697 vfs_badlock("is not exclusive locked but should be", str, vp);
3698 }
3699
3700 #if 0
3701 void
3702 assert_vop_elocked_other(struct vnode *vp, const char *str)
3703 {
3704
3705 if (vp && !IGNORE_LOCK(vp) &&
3706 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
3707 vfs_badlock("is not exclusive locked by another thread",
3708 str, vp);
3709 }
3710
3711 void
3712 assert_vop_slocked(struct vnode *vp, const char *str)
3713 {
3714
3715 if (vp && !IGNORE_LOCK(vp) &&
3716 VOP_ISLOCKED(vp, curthread) != LK_SHARED)
3717 vfs_badlock("is not locked shared but should be", str, vp);
3718 }
3719 #endif /* 0 */
3720 #endif /* DEBUG_VFS_LOCKS */
3721
3722 void
3723 vop_rename_pre(void *ap)
3724 {
3725 struct vop_rename_args *a = ap;
3726
3727 #ifdef DEBUG_VFS_LOCKS
3728 if (a->a_tvp)
3729 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3730 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3731 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3732 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3733
3734 /* Check the source (from). */
3735 if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3736 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3737 if (a->a_tvp != a->a_fvp)
3738 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3739
3740 /* Check the target. */
3741 if (a->a_tvp)
3742 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3743 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3744 #endif
3745 if (a->a_tdvp != a->a_fdvp)
3746 vhold(a->a_fdvp);
3747 if (a->a_tvp != a->a_fvp)
3748 vhold(a->a_fvp);
3749 vhold(a->a_tdvp);
3750 if (a->a_tvp)
3751 vhold(a->a_tvp);
3752 }
3753
3754 void
3755 vop_strategy_pre(void *ap)
3756 {
3757 #ifdef DEBUG_VFS_LOCKS
3758 struct vop_strategy_args *a;
3759 struct buf *bp;
3760
3761 a = ap;
3762 bp = a->a_bp;
3763
3764 /*
3765 * Cluster ops lock their component buffers but not the IO container.
3766 */
3767 if ((bp->b_flags & B_CLUSTER) != 0)
3768 return;
3769
3770 if (BUF_REFCNT(bp) < 1) {
3771 if (vfs_badlock_print)
3772 printf(
3773 "VOP_STRATEGY: bp is not locked but should be\n");
3774 if (vfs_badlock_ddb)
3775 kdb_enter_why(KDB_WHY_VFSLOCK, "lock violation");
3776 }
3777 #endif
3778 }
3779
3780 void
3781 vop_lookup_pre(void *ap)
3782 {
3783 #ifdef DEBUG_VFS_LOCKS
3784 struct vop_lookup_args *a;
3785 struct vnode *dvp;
3786
3787 a = ap;
3788 dvp = a->a_dvp;
3789 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3790 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3791 #endif
3792 }
3793
3794 void
3795 vop_lookup_post(void *ap, int rc)
3796 {
3797 #ifdef DEBUG_VFS_LOCKS
3798 struct vop_lookup_args *a;
3799 struct vnode *dvp;
3800 struct vnode *vp;
3801
3802 a = ap;
3803 dvp = a->a_dvp;
3804 vp = *(a->a_vpp);
3805
3806 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3807 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3808
3809 if (!rc)
3810 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3811 #endif
3812 }
3813
3814 void
3815 vop_lock_pre(void *ap)
3816 {
3817 #ifdef DEBUG_VFS_LOCKS
3818 struct vop_lock1_args *a = ap;
3819
3820 if ((a->a_flags & LK_INTERLOCK) == 0)
3821 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3822 else
3823 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3824 #endif
3825 }
3826
3827 void
3828 vop_lock_post(void *ap, int rc)
3829 {
3830 #ifdef DEBUG_VFS_LOCKS
3831 struct vop_lock1_args *a = ap;
3832
3833 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3834 if (rc == 0)
3835 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3836 #endif
3837 }
3838
3839 void
3840 vop_unlock_pre(void *ap)
3841 {
3842 #ifdef DEBUG_VFS_LOCKS
3843 struct vop_unlock_args *a = ap;
3844
3845 if (a->a_flags & LK_INTERLOCK)
3846 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3847 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3848 #endif
3849 }
3850
3851 void
3852 vop_unlock_post(void *ap, int rc)
3853 {
3854 #ifdef DEBUG_VFS_LOCKS
3855 struct vop_unlock_args *a = ap;
3856
3857 if (a->a_flags & LK_INTERLOCK)
3858 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3859 #endif
3860 }
3861
3862 void
3863 vop_create_post(void *ap, int rc)
3864 {
3865 struct vop_create_args *a = ap;
3866
3867 if (!rc)
3868 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3869 }
3870
3871 void
3872 vop_link_post(void *ap, int rc)
3873 {
3874 struct vop_link_args *a = ap;
3875
3876 if (!rc) {
3877 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3878 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3879 }
3880 }
3881
3882 void
3883 vop_mkdir_post(void *ap, int rc)
3884 {
3885 struct vop_mkdir_args *a = ap;
3886
3887 if (!rc)
3888 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3889 }
3890
3891 void
3892 vop_mknod_post(void *ap, int rc)
3893 {
3894 struct vop_mknod_args *a = ap;
3895
3896 if (!rc)
3897 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3898 }
3899
3900 void
3901 vop_remove_post(void *ap, int rc)
3902 {
3903 struct vop_remove_args *a = ap;
3904
3905 if (!rc) {
3906 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3907 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3908 }
3909 }
3910
3911 void
3912 vop_rename_post(void *ap, int rc)
3913 {
3914 struct vop_rename_args *a = ap;
3915
3916 if (!rc) {
3917 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3918 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3919 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3920 if (a->a_tvp)
3921 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3922 }
3923 if (a->a_tdvp != a->a_fdvp)
3924 vdrop(a->a_fdvp);
3925 if (a->a_tvp != a->a_fvp)
3926 vdrop(a->a_fvp);
3927 vdrop(a->a_tdvp);
3928 if (a->a_tvp)
3929 vdrop(a->a_tvp);
3930 }
3931
3932 void
3933 vop_rmdir_post(void *ap, int rc)
3934 {
3935 struct vop_rmdir_args *a = ap;
3936
3937 if (!rc) {
3938 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3939 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3940 }
3941 }
3942
3943 void
3944 vop_setattr_post(void *ap, int rc)
3945 {
3946 struct vop_setattr_args *a = ap;
3947
3948 if (!rc)
3949 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3950 }
3951
3952 void
3953 vop_symlink_post(void *ap, int rc)
3954 {
3955 struct vop_symlink_args *a = ap;
3956
3957 if (!rc)
3958 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3959 }
3960
3961 static struct knlist fs_knlist;
3962
3963 static void
3964 vfs_event_init(void *arg)
3965 {
3966 knlist_init_mtx(&fs_knlist, NULL);
3967 }
3968 /* XXX - correct order? */
3969 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3970
3971 void
3972 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3973 {
3974
3975 KNOTE_UNLOCKED(&fs_knlist, event);
3976 }
3977
3978 static int filt_fsattach(struct knote *kn);
3979 static void filt_fsdetach(struct knote *kn);
3980 static int filt_fsevent(struct knote *kn, long hint);
3981
3982 struct filterops fs_filtops =
3983 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3984
3985 static int
3986 filt_fsattach(struct knote *kn)
3987 {
3988
3989 kn->kn_flags |= EV_CLEAR;
3990 knlist_add(&fs_knlist, kn, 0);
3991 return (0);
3992 }
3993
3994 static void
3995 filt_fsdetach(struct knote *kn)
3996 {
3997
3998 knlist_remove(&fs_knlist, kn, 0);
3999 }
4000
4001 static int
4002 filt_fsevent(struct knote *kn, long hint)
4003 {
4004
4005 kn->kn_fflags |= hint;
4006 return (kn->kn_fflags != 0);
4007 }
4008
4009 static int
4010 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4011 {
4012 struct vfsidctl vc;
4013 int error;
4014 struct mount *mp;
4015
4016 error = SYSCTL_IN(req, &vc, sizeof(vc));
4017 if (error)
4018 return (error);
4019 if (vc.vc_vers != VFS_CTL_VERS1)
4020 return (EINVAL);
4021 mp = vfs_getvfs(&vc.vc_fsid);
4022 if (mp == NULL)
4023 return (ENOENT);
4024 /* ensure that a specific sysctl goes to the right filesystem. */
4025 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4026 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4027 vfs_rel(mp);
4028 return (EINVAL);
4029 }
4030 VCTLTOREQ(&vc, req);
4031 error = VFS_SYSCTL(mp, vc.vc_op, req);
4032 vfs_rel(mp);
4033 return (error);
4034 }
4035
4036 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4037 "Sysctl by fsid");
4038
4039 /*
4040 * Function to initialize a va_filerev field sensibly.
4041 * XXX: Wouldn't a random number make a lot more sense ??
4042 */
4043 u_quad_t
4044 init_va_filerev(void)
4045 {
4046 struct bintime bt;
4047
4048 getbinuptime(&bt);
4049 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4050 }
4051
4052 static int filt_vfsread(struct knote *kn, long hint);
4053 static int filt_vfswrite(struct knote *kn, long hint);
4054 static int filt_vfsvnode(struct knote *kn, long hint);
4055 static void filt_vfsdetach(struct knote *kn);
4056 static struct filterops vfsread_filtops =
4057 { 1, NULL, filt_vfsdetach, filt_vfsread };
4058 static struct filterops vfswrite_filtops =
4059 { 1, NULL, filt_vfsdetach, filt_vfswrite };
4060 static struct filterops vfsvnode_filtops =
4061 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
4062
4063 static void
4064 vfs_knllock(void *arg)
4065 {
4066 struct vnode *vp = arg;
4067
4068 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
4069 }
4070
4071 static void
4072 vfs_knlunlock(void *arg)
4073 {
4074 struct vnode *vp = arg;
4075
4076 VOP_UNLOCK(vp, 0, curthread);
4077 }
4078
4079 int
4080 vfs_kqfilter(struct vop_kqfilter_args *ap)
4081 {
4082 struct vnode *vp = ap->a_vp;
4083 struct knote *kn = ap->a_kn;
4084 struct knlist *knl;
4085
4086 switch (kn->kn_filter) {
4087 case EVFILT_READ:
4088 kn->kn_fop = &vfsread_filtops;
4089 break;
4090 case EVFILT_WRITE:
4091 kn->kn_fop = &vfswrite_filtops;
4092 break;
4093 case EVFILT_VNODE:
4094 kn->kn_fop = &vfsvnode_filtops;
4095 break;
4096 default:
4097 return (EINVAL);
4098 }
4099
4100 kn->kn_hook = (caddr_t)vp;
4101
4102 v_addpollinfo(vp);
4103 if (vp->v_pollinfo == NULL)
4104 return (ENOMEM);
4105 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4106 knlist_add(knl, kn, 0);
4107
4108 return (0);
4109 }
4110
4111 /*
4112 * Detach knote from vnode
4113 */
4114 static void
4115 filt_vfsdetach(struct knote *kn)
4116 {
4117 struct vnode *vp = (struct vnode *)kn->kn_hook;
4118
4119 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4120 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4121 }
4122
4123 /*ARGSUSED*/
4124 static int
4125 filt_vfsread(struct knote *kn, long hint)
4126 {
4127 struct vnode *vp = (struct vnode *)kn->kn_hook;
4128 struct vattr va;
4129 int res;
4130
4131 /*
4132 * filesystem is gone, so set the EOF flag and schedule
4133 * the knote for deletion.
4134 */
4135 if (hint == NOTE_REVOKE) {
4136 VI_LOCK(vp);
4137 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4138 VI_UNLOCK(vp);
4139 return (1);
4140 }
4141
4142 if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
4143 return (0);
4144
4145 VI_LOCK(vp);
4146 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4147 res = (kn->kn_data != 0);
4148 VI_UNLOCK(vp);
4149 return (res);
4150 }
4151
4152 /*ARGSUSED*/
4153 static int
4154 filt_vfswrite(struct knote *kn, long hint)
4155 {
4156 struct vnode *vp = (struct vnode *)kn->kn_hook;
4157
4158 VI_LOCK(vp);
4159
4160 /*
4161 * filesystem is gone, so set the EOF flag and schedule
4162 * the knote for deletion.
4163 */
4164 if (hint == NOTE_REVOKE)
4165 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4166
4167 kn->kn_data = 0;
4168 VI_UNLOCK(vp);
4169 return (1);
4170 }
4171
4172 static int
4173 filt_vfsvnode(struct knote *kn, long hint)
4174 {
4175 struct vnode *vp = (struct vnode *)kn->kn_hook;
4176 int res;
4177
4178 VI_LOCK(vp);
4179 if (kn->kn_sfflags & hint)
4180 kn->kn_fflags |= hint;
4181 if (hint == NOTE_REVOKE) {
4182 kn->kn_flags |= EV_EOF;
4183 VI_UNLOCK(vp);
4184 return (1);
4185 }
4186 res = (kn->kn_fflags != 0);
4187 VI_UNLOCK(vp);
4188 return (res);
4189 }
4190
4191 int
4192 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4193 {
4194 int error;
4195
4196 if (dp->d_reclen > ap->a_uio->uio_resid)
4197 return (ENAMETOOLONG);
4198 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4199 if (error) {
4200 if (ap->a_ncookies != NULL) {
4201 if (ap->a_cookies != NULL)
4202 free(ap->a_cookies, M_TEMP);
4203 ap->a_cookies = NULL;
4204 *ap->a_ncookies = 0;
4205 }
4206 return (error);
4207 }
4208 if (ap->a_ncookies == NULL)
4209 return (0);
4210
4211 KASSERT(ap->a_cookies,
4212 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4213
4214 *ap->a_cookies = realloc(*ap->a_cookies,
4215 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4216 (*ap->a_cookies)[*ap->a_ncookies] = off;
4217 return (0);
4218 }
4219
4220 /*
4221 * Mark for update the access time of the file if the filesystem
4222 * supports VA_MARK_ATIME. This functionality is used by execve
4223 * and mmap, so we want to avoid the synchronous I/O implied by
4224 * directly setting va_atime for the sake of efficiency.
4225 */
4226 void
4227 vfs_mark_atime(struct vnode *vp, struct thread *td)
4228 {
4229 struct vattr atimeattr;
4230
4231 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
4232 VATTR_NULL(&atimeattr);
4233 atimeattr.va_vaflags |= VA_MARK_ATIME;
4234 (void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td);
4235 }
4236 }
Cache object: 6076cc9edb2ad9b27d0a46db4bd53059
|