1 /* $OpenBSD: vfs_subr.c,v 1.322 2024/07/13 14:37:55 beck Exp $ */
2 /* $NetBSD: vfs_subr.c,v 1.53 1996/04/22 01:39:13 christos Exp $ */
3
4 /*
5 * Copyright (c) 1989, 1993
6 * The Regents of the University of California. All rights reserved.
7 * (c) UNIX System Laboratories, Inc.
8 * All or some portions of this file are derived from material licensed
9 * to the University of California by American Telephone and Telegraph
10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11 * the permission of UNIX System Laboratories, Inc.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94
38 */
39
40 /*
41 * External virtual filesystem routines
42 */
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/proc.h>
47 #include <sys/sysctl.h>
48 #include <sys/mount.h>
49 #include <sys/fcntl.h>
50 #include <sys/conf.h>
51 #include <sys/vnode.h>
52 #include <sys/lock.h>
53 #include <sys/lockf.h>
54 #include <sys/stat.h>
55 #include <sys/acct.h>
56 #include <sys/namei.h>
57 #include <sys/ucred.h>
58 #include <sys/buf.h>
59 #include <sys/errno.h>
60 #include <sys/malloc.h>
61 #include <sys/mbuf.h>
62 #include <sys/syscallargs.h>
63 #include <sys/pool.h>
64 #include <sys/tree.h>
65 #include <sys/specdev.h>
66 #include <sys/atomic.h>
67
68 #include <netinet/in.h>
69
70 #include <uvm/uvm_extern.h>
71 #include <uvm/uvm_vnode.h>
72
73 #include "softraid.h"
74
75 void sr_quiesce(void);
76
77 enum vtype iftovt_tab[16] = {
78 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
79 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
80 };
81
82 int vttoif_tab[9] = {
83 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
84 S_IFSOCK, S_IFIFO, S_IFMT,
85 };
86
87 int prtactive = 0; /* 1 => print out reclaim of active vnodes */
88 int suid_clear = 1; /* 1 => clear SUID / SGID on owner change */
89
90 /*
91 * Insq/Remq for the vnode usage lists.
92 */
93 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs)
94 #define bufremvn(bp) { \
95 LIST_REMOVE(bp, b_vnbufs); \
96 LIST_NEXT(bp, b_vnbufs) = NOLIST; \
97 }
98
99 TAILQ_HEAD(freelst, vnode);
100 struct freelst vnode_hold_list; /* list of vnodes referencing buffers */
101 struct freelst vnode_free_list; /* vnode free list */
102
103 struct mntlist mountlist; /* mounted filesystem list */
104
105 void vclean(struct vnode *, int, struct proc *);
106
107 void insmntque(struct vnode *, struct mount *);
108 int getdevvp(dev_t, struct vnode **, enum vtype);
109
110 int vfs_hang_addrlist(struct mount *, struct netexport *,
111 struct export_args *);
112 int vfs_free_netcred(struct radix_node *, void *, u_int);
113 void vfs_free_addrlist(struct netexport *);
114 void vputonfreelist(struct vnode *);
115
116 int vflush_vnode(struct vnode *, void *);
117 int maxvnodes;
118
119 struct mutex vnode_mtx = MUTEX_INITIALIZER(IPL_BIO);
120
121 void vfs_unmountall(void);
122
123 #ifdef DEBUG
124 void printlockedvnodes(void);
125 #endif
126
127 struct pool vnode_pool;
128 struct pool uvm_vnode_pool;
129
130 static inline int rb_buf_compare(const struct buf *b1, const struct buf *b2);
131 RBT_GENERATE(buf_rb_bufs, buf, b_rbbufs, rb_buf_compare);
132
133 static inline int
rb_buf_compare(const struct buf * b1,const struct buf * b2)134 rb_buf_compare(const struct buf *b1, const struct buf *b2)
135 {
136 if (b1->b_lblkno < b2->b_lblkno)
137 return(-1);
138 if (b1->b_lblkno > b2->b_lblkno)
139 return(1);
140 return(0);
141 }
142
143 /*
144 * Initialize the vnode management data structures.
145 */
146 void
vntblinit(void)147 vntblinit(void)
148 {
149 /* buffer cache may need a vnode for each buffer */
150 maxvnodes = 2 * initialvnodes;
151 pool_init(&vnode_pool, sizeof(struct vnode), 0, IPL_NONE,
152 PR_WAITOK, "vnodes", NULL);
153 pool_init(&uvm_vnode_pool, sizeof(struct uvm_vnode), 0, IPL_NONE,
154 PR_WAITOK, "uvmvnodes", NULL);
155 TAILQ_INIT(&vnode_hold_list);
156 TAILQ_INIT(&vnode_free_list);
157 TAILQ_INIT(&mountlist);
158 /*
159 * Initialize the filesystem syncer.
160 */
161 vn_initialize_syncerd();
162
163 #ifdef NFSSERVER
164 rn_init(sizeof(struct sockaddr_in));
165 #endif /* NFSSERVER */
166 }
167
168 /*
169 * Allocate a mount point.
170 *
171 * The returned mount point is marked as busy.
172 */
173 struct mount *
vfs_mount_alloc(struct vnode * vp,struct vfsconf * vfsp)174 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp)
175 {
176 struct mount *mp;
177
178 mp = malloc(sizeof(*mp), M_MOUNT, M_WAITOK|M_ZERO);
179 rw_init_flags(&mp->mnt_lock, "vfslock", RWL_IS_VNODE);
180 (void)vfs_busy(mp, VB_READ|VB_NOWAIT);
181
182 TAILQ_INIT(&mp->mnt_vnodelist);
183 mp->mnt_vnodecovered = vp;
184
185 atomic_inc_int(&vfsp->vfc_refcount);
186 mp->mnt_vfc = vfsp;
187 mp->mnt_op = vfsp->vfc_vfsops;
188 mp->mnt_flag = vfsp->vfc_flags;
189 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
190
191 return (mp);
192 }
193
194 /*
195 * Release a mount point.
196 */
197 void
vfs_mount_free(struct mount * mp)198 vfs_mount_free(struct mount *mp)
199 {
200 atomic_dec_int(&mp->mnt_vfc->vfc_refcount);
201 free(mp, M_MOUNT, sizeof(*mp));
202 }
203
204 /*
205 * Mark a mount point as busy. Used to synchronize access and to delay
206 * unmounting.
207 *
208 * Default behaviour is to attempt getting a READ lock and in case of an
209 * ongoing unmount, to wait for it to finish and then return failure.
210 */
211 int
vfs_busy(struct mount * mp,int flags)212 vfs_busy(struct mount *mp, int flags)
213 {
214 int rwflags = 0;
215
216 if (flags & VB_WRITE)
217 rwflags |= RW_WRITE;
218 else
219 rwflags |= RW_READ;
220
221 if (flags & VB_WAIT)
222 rwflags |= RW_SLEEPFAIL;
223 else
224 rwflags |= RW_NOSLEEP;
225
226 #ifdef WITNESS
227 if (flags & VB_DUPOK)
228 rwflags |= RW_DUPOK;
229 #endif
230
231 if (rw_enter(&mp->mnt_lock, rwflags))
232 return (EBUSY);
233
234 return (0);
235 }
236
237 /*
238 * Free a busy file system
239 */
240 void
vfs_unbusy(struct mount * mp)241 vfs_unbusy(struct mount *mp)
242 {
243 rw_exit(&mp->mnt_lock);
244 }
245
246 int
vfs_isbusy(struct mount * mp)247 vfs_isbusy(struct mount *mp)
248 {
249 if (RWLOCK_OWNER(&mp->mnt_lock) > 0)
250 return (1);
251 else
252 return (0);
253 }
254
255 /*
256 * Lookup a filesystem type, and if found allocate and initialize
257 * a mount structure for it.
258 *
259 * Devname is usually updated by mount(8) after booting.
260 */
261 int
vfs_rootmountalloc(char * fstypename,char * devname,struct mount ** mpp)262 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
263 {
264 struct vfsconf *vfsp;
265 struct mount *mp;
266
267 vfsp = vfs_byname(fstypename);
268 if (vfsp == NULL)
269 return (ENODEV);
270 mp = vfs_mount_alloc(NULLVP, vfsp);
271 mp->mnt_flag |= MNT_RDONLY;
272 mp->mnt_stat.f_mntonname[0] = '/';
273 strlcpy(mp->mnt_stat.f_mntfromname, devname, MNAMELEN);
274 strlcpy(mp->mnt_stat.f_mntfromspec, devname, MNAMELEN);
275 *mpp = mp;
276 return (0);
277 }
278
279 /*
280 * Lookup a mount point by filesystem identifier.
281 */
282 struct mount *
vfs_getvfs(fsid_t * fsid)283 vfs_getvfs(fsid_t *fsid)
284 {
285 struct mount *mp;
286
287 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
288 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
289 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
290 return (mp);
291 }
292 }
293
294 return (NULL);
295 }
296
297
298 /*
299 * Get a new unique fsid
300 */
301 void
vfs_getnewfsid(struct mount * mp)302 vfs_getnewfsid(struct mount *mp)
303 {
304 static u_short xxxfs_mntid;
305
306 fsid_t tfsid;
307 int mtype;
308
309 mtype = mp->mnt_vfc->vfc_typenum;
310 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
311 mp->mnt_stat.f_fsid.val[1] = mtype;
312 if (xxxfs_mntid == 0)
313 ++xxxfs_mntid;
314 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
315 tfsid.val[1] = mtype;
316 if (!TAILQ_EMPTY(&mountlist)) {
317 while (vfs_getvfs(&tfsid)) {
318 tfsid.val[0]++;
319 xxxfs_mntid++;
320 }
321 }
322 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
323 }
324
325 /*
326 * Set vnode attributes to VNOVAL
327 */
328 void
vattr_null(struct vattr * vap)329 vattr_null(struct vattr *vap)
330 {
331
332 vap->va_type = VNON;
333 /*
334 * Don't get fancy: u_quad_t = u_int = VNOVAL leaves the u_quad_t
335 * with 2^31-1 instead of 2^64-1. Just write'm out and let
336 * the compiler do its job.
337 */
338 vap->va_mode = VNOVAL;
339 vap->va_nlink = VNOVAL;
340 vap->va_uid = VNOVAL;
341 vap->va_gid = VNOVAL;
342 vap->va_fsid = VNOVAL;
343 vap->va_fileid = VNOVAL;
344 vap->va_size = VNOVAL;
345 vap->va_blocksize = VNOVAL;
346 vap->va_atime.tv_sec = VNOVAL;
347 vap->va_atime.tv_nsec = VNOVAL;
348 vap->va_mtime.tv_sec = VNOVAL;
349 vap->va_mtime.tv_nsec = VNOVAL;
350 vap->va_ctime.tv_sec = VNOVAL;
351 vap->va_ctime.tv_nsec = VNOVAL;
352 vap->va_gen = VNOVAL;
353 vap->va_flags = VNOVAL;
354 vap->va_rdev = VNOVAL;
355 vap->va_bytes = VNOVAL;
356 vap->va_filerev = VNOVAL;
357 vap->va_vaflags = 0;
358 }
359
360 /*
361 * Routines having to do with the management of the vnode table.
362 */
363 long numvnodes;
364
365 /*
366 * Return the next vnode from the free list.
367 */
368 int
getnewvnode(enum vtagtype tag,struct mount * mp,const struct vops * vops,struct vnode ** vpp)369 getnewvnode(enum vtagtype tag, struct mount *mp, const struct vops *vops,
370 struct vnode **vpp)
371 {
372 struct proc *p = curproc;
373 struct freelst *listhd;
374 static int toggle;
375 struct vnode *vp;
376 int s;
377
378 /*
379 * allow maxvnodes to increase if the buffer cache itself
380 * is big enough to justify it. (we don't shrink it ever)
381 */
382 maxvnodes = maxvnodes < bcstats.numbufs ? bcstats.numbufs
383 : maxvnodes;
384
385 /*
386 * We must choose whether to allocate a new vnode or recycle an
387 * existing one. The criterion for allocating a new one is that
388 * the total number of vnodes is less than the number desired or
389 * there are no vnodes on either free list. Generally we only
390 * want to recycle vnodes that have no buffers associated with
391 * them, so we look first on the vnode_free_list. If it is empty,
392 * we next consider vnodes with referencing buffers on the
393 * vnode_hold_list. The toggle ensures that half the time we
394 * will use a buffer from the vnode_hold_list, and half the time
395 * we will allocate a new one unless the list has grown to twice
396 * the desired size. We are reticent to recycle vnodes from the
397 * vnode_hold_list because we will lose the identity of all its
398 * referencing buffers.
399 */
400 toggle ^= 1;
401 if (numvnodes / 2 > maxvnodes)
402 toggle = 0;
403
404 s = splbio();
405 if ((numvnodes < maxvnodes) ||
406 ((TAILQ_FIRST(listhd = &vnode_free_list) == NULL) &&
407 ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) {
408 splx(s);
409 vp = pool_get(&vnode_pool, PR_WAITOK | PR_ZERO);
410 vp->v_uvm = pool_get(&uvm_vnode_pool, PR_WAITOK | PR_ZERO);
411 vp->v_uvm->u_vnode = vp;
412 uvm_obj_init(&vp->v_uvm->u_obj, &uvm_vnodeops, 0);
413 RBT_INIT(buf_rb_bufs, &vp->v_bufs_tree);
414 cache_tree_init(&vp->v_nc_tree);
415 TAILQ_INIT(&vp->v_cache_dst);
416 numvnodes++;
417 } else {
418 TAILQ_FOREACH(vp, listhd, v_freelist) {
419 if (VOP_ISLOCKED(vp) == 0)
420 break;
421 }
422 /*
423 * Unless this is a bad time of the month, at most
424 * the first NCPUS items on the free list are
425 * locked, so this is close enough to being empty.
426 */
427 if (vp == NULL) {
428 splx(s);
429 tablefull("vnode");
430 *vpp = NULL;
431 return (ENFILE);
432 }
433
434 #ifdef DIAGNOSTIC
435 if (vp->v_usecount) {
436 vprint("free vnode", vp);
437 panic("free vnode isn't");
438 }
439 #endif
440
441 TAILQ_REMOVE(listhd, vp, v_freelist);
442 vp->v_bioflag &= ~VBIOONFREELIST;
443 splx(s);
444
445 if (vp->v_type != VBAD)
446 vgonel(vp, p);
447 #ifdef DIAGNOSTIC
448 if (vp->v_data) {
449 vprint("cleaned vnode", vp);
450 panic("cleaned vnode isn't");
451 }
452 s = splbio();
453 if (vp->v_numoutput)
454 panic("Clean vnode has pending I/O's");
455 splx(s);
456 #endif
457 vp->v_flag = 0;
458 vp->v_socket = NULL;
459 }
460 cache_purge(vp);
461 vp->v_type = VNON;
462 vp->v_tag = tag;
463 vp->v_op = vops;
464 insmntque(vp, mp);
465 *vpp = vp;
466 vp->v_usecount = 1;
467 vp->v_data = NULL;
468 return (0);
469 }
470
471 /*
472 * Move a vnode from one mount queue to another.
473 */
474 void
insmntque(struct vnode * vp,struct mount * mp)475 insmntque(struct vnode *vp, struct mount *mp)
476 {
477 /*
478 * Delete from old mount point vnode list, if on one.
479 */
480 if (vp->v_mount != NULL)
481 TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes);
482 /*
483 * Insert into list of vnodes for the new mount point, if available.
484 */
485 if ((vp->v_mount = mp) != NULL)
486 TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes);
487 }
488
489 /*
490 * Create a vnode for a block device.
491 * Used for root filesystem, argdev, and swap areas.
492 * Also used for memory file system special devices.
493 */
494 int
bdevvp(dev_t dev,struct vnode ** vpp)495 bdevvp(dev_t dev, struct vnode **vpp)
496 {
497 return (getdevvp(dev, vpp, VBLK));
498 }
499
500 /*
501 * Create a vnode for a character device.
502 * Used for console handling.
503 */
504 int
cdevvp(dev_t dev,struct vnode ** vpp)505 cdevvp(dev_t dev, struct vnode **vpp)
506 {
507 return (getdevvp(dev, vpp, VCHR));
508 }
509
510 /*
511 * Create a vnode for a device.
512 * Used by bdevvp (block device) for root file system etc.,
513 * and by cdevvp (character device) for console.
514 */
515 int
getdevvp(dev_t dev,struct vnode ** vpp,enum vtype type)516 getdevvp(dev_t dev, struct vnode **vpp, enum vtype type)
517 {
518 struct vnode *vp;
519 struct vnode *nvp;
520 int error;
521
522 if (dev == NODEV) {
523 *vpp = NULLVP;
524 return (0);
525 }
526 error = getnewvnode(VT_NON, NULL, &spec_vops, &nvp);
527 if (error) {
528 *vpp = NULLVP;
529 return (error);
530 }
531 vp = nvp;
532 vp->v_type = type;
533 if ((nvp = checkalias(vp, dev, NULL)) != NULL) {
534 vput(vp);
535 vp = nvp;
536 }
537 if (vp->v_type == VCHR && cdevsw[major(vp->v_rdev)].d_type == D_TTY)
538 vp->v_flag |= VISTTY;
539 *vpp = vp;
540 return (0);
541 }
542
543 /*
544 * Check to see if the new vnode represents a special device
545 * for which we already have a vnode (either because of
546 * bdevvp() or because of a different vnode representing
547 * the same block device). If such an alias exists, deallocate
548 * the existing contents and return the aliased vnode. The
549 * caller is responsible for filling it with its new contents.
550 */
551 struct vnode *
checkalias(struct vnode * nvp,dev_t nvp_rdev,struct mount * mp)552 checkalias(struct vnode *nvp, dev_t nvp_rdev, struct mount *mp)
553 {
554 struct proc *p = curproc;
555 struct vnode *vp;
556 struct vnodechain *vchain;
557
558 if (nvp->v_type != VBLK && nvp->v_type != VCHR)
559 return (NULLVP);
560
561 vchain = &speclisth[SPECHASH(nvp_rdev)];
562 loop:
563 SLIST_FOREACH(vp, vchain, v_specnext) {
564 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) {
565 continue;
566 }
567 /*
568 * Alias, but not in use, so flush it out.
569 */
570 if (vp->v_usecount == 0) {
571 vgonel(vp, p);
572 goto loop;
573 }
574 if (vget(vp, LK_EXCLUSIVE)) {
575 goto loop;
576 }
577 break;
578 }
579
580 /*
581 * Common case is actually in the if statement
582 */
583 if (vp == NULL || !(vp->v_tag == VT_NON && vp->v_type == VBLK)) {
584 nvp->v_specinfo = malloc(sizeof(struct specinfo), M_VNODE,
585 M_WAITOK);
586 nvp->v_rdev = nvp_rdev;
587 nvp->v_hashchain = vchain;
588 nvp->v_specmountpoint = NULL;
589 nvp->v_speclockf = NULL;
590 nvp->v_specbitmap = NULL;
591 if (nvp->v_type == VCHR &&
592 (cdevsw[major(nvp_rdev)].d_flags & D_CLONE) &&
593 (minor(nvp_rdev) >> CLONE_SHIFT == 0)) {
594 if (vp != NULLVP)
595 nvp->v_specbitmap = vp->v_specbitmap;
596 else
597 nvp->v_specbitmap = malloc(CLONE_MAPSZ,
598 M_VNODE, M_WAITOK | M_ZERO);
599 }
600 SLIST_INSERT_HEAD(vchain, nvp, v_specnext);
601 if (vp != NULLVP) {
602 nvp->v_flag |= VALIASED;
603 vp->v_flag |= VALIASED;
604 vput(vp);
605 }
606 return (NULLVP);
607 }
608
609 /*
610 * This code is the uncommon case. It is called in case
611 * we found an alias that was VT_NON && vtype of VBLK
612 * This means we found a block device that was created
613 * using bdevvp.
614 * An example of such a vnode is the root partition device vnode
615 * created in ffs_mountroot.
616 *
617 * The vnodes created by bdevvp should not be aliased (why?).
618 */
619
620 VOP_UNLOCK(vp);
621 vclean(vp, 0, p);
622 vp->v_op = nvp->v_op;
623 vp->v_tag = nvp->v_tag;
624 nvp->v_type = VNON;
625 insmntque(vp, mp);
626 return (vp);
627 }
628
629 /*
630 * Grab a particular vnode from the free list, increment its
631 * reference count and lock it. If the vnode lock bit is set,
632 * the vnode is being eliminated in vgone. In that case, we
633 * cannot grab it, so the process is awakened when the
634 * transition is completed, and an error code is returned to
635 * indicate that the vnode is no longer usable, possibly
636 * having been changed to a new file system type.
637 */
638 int
vget(struct vnode * vp,int flags)639 vget(struct vnode *vp, int flags)
640 {
641 int error, s, onfreelist;
642
643 /*
644 * If the vnode is in the process of being cleaned out for
645 * another use, we wait for the cleaning to finish and then
646 * return failure. Cleaning is determined by checking that
647 * the VXLOCK flag is set.
648 */
649 mtx_enter(&vnode_mtx);
650 if (vp->v_lflag & VXLOCK) {
651 if (flags & LK_NOWAIT) {
652 mtx_leave(&vnode_mtx);
653 return (EBUSY);
654 }
655
656 vp->v_lflag |= VXWANT;
657 msleep_nsec(vp, &vnode_mtx, PINOD, "vget", INFSLP);
658 mtx_leave(&vnode_mtx);
659 return (ENOENT);
660 }
661 mtx_leave(&vnode_mtx);
662
663 s = splbio();
664 onfreelist = vp->v_bioflag & VBIOONFREELIST;
665 if (vp->v_usecount == 0 && onfreelist) {
666 if (vp->v_holdcnt > 0)
667 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
668 else
669 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
670 vp->v_bioflag &= ~VBIOONFREELIST;
671 }
672 splx(s);
673
674 vp->v_usecount++;
675 if (flags & LK_TYPE_MASK) {
676 if ((error = vn_lock(vp, flags)) != 0) {
677 vp->v_usecount--;
678 if (vp->v_usecount == 0 && onfreelist)
679 vputonfreelist(vp);
680 }
681 return (error);
682 }
683
684 return (0);
685 }
686
687
688 /* Vnode reference. */
689 void
vref(struct vnode * vp)690 vref(struct vnode *vp)
691 {
692 KERNEL_ASSERT_LOCKED();
693
694 #ifdef DIAGNOSTIC
695 if (vp->v_usecount == 0)
696 panic("vref used where vget required");
697 if (vp->v_type == VNON)
698 panic("vref on a VNON vnode");
699 #endif
700 vp->v_usecount++;
701 }
702
703 void
vputonfreelist(struct vnode * vp)704 vputonfreelist(struct vnode *vp)
705 {
706 int s;
707 struct freelst *lst;
708
709 s = splbio();
710 #ifdef DIAGNOSTIC
711 if (vp->v_usecount != 0)
712 panic("Use count is not zero!");
713
714 /*
715 * If the hold count is still positive, one or many threads could still
716 * be waiting on the vnode lock inside uvn_io().
717 */
718 if (vp->v_holdcnt == 0 && vp->v_lockcount != 0)
719 panic("%s: lock count is not zero", __func__);
720
721 if (vp->v_bioflag & VBIOONFREELIST) {
722 vprint("vnode already on free list: ", vp);
723 panic("vnode already on free list");
724 }
725 #endif
726
727 vp->v_bioflag |= VBIOONFREELIST;
728 vp->v_bioflag &= ~VBIOERROR;
729
730 if (vp->v_holdcnt > 0)
731 lst = &vnode_hold_list;
732 else
733 lst = &vnode_free_list;
734
735 if (vp->v_type == VBAD)
736 TAILQ_INSERT_HEAD(lst, vp, v_freelist);
737 else
738 TAILQ_INSERT_TAIL(lst, vp, v_freelist);
739
740 splx(s);
741 }
742
743 /*
744 * vput(), just unlock and vrele()
745 */
746 void
vput(struct vnode * vp)747 vput(struct vnode *vp)
748 {
749 struct proc *p = curproc;
750 int s;
751
752 #ifdef DIAGNOSTIC
753 if (vp == NULL)
754 panic("vput: null vp");
755 #endif
756
757 #ifdef DIAGNOSTIC
758 if (vp->v_usecount == 0) {
759 vprint("vput: bad ref count", vp);
760 panic("vput: ref cnt");
761 }
762 #endif
763 vp->v_usecount--;
764 KASSERT(vp->v_usecount > 0 || vp->v_uvcount == 0);
765 if (vp->v_usecount > 0) {
766 VOP_UNLOCK(vp);
767 return;
768 }
769
770 #ifdef DIAGNOSTIC
771 if (vp->v_writecount != 0) {
772 vprint("vput: bad writecount", vp);
773 panic("vput: v_writecount != 0");
774 }
775 #endif
776
777 VOP_INACTIVE(vp, p);
778
779 s = splbio();
780 if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
781 vputonfreelist(vp);
782 splx(s);
783 }
784
785 /*
786 * Vnode release - use for active VNODES.
787 * If count drops to zero, call inactive routine and return to freelist.
788 * Returns 0 if it did not sleep.
789 */
790 int
vrele(struct vnode * vp)791 vrele(struct vnode *vp)
792 {
793 struct proc *p = curproc;
794 int s;
795
796 #ifdef DIAGNOSTIC
797 if (vp == NULL)
798 panic("vrele: null vp");
799 #endif
800 #ifdef DIAGNOSTIC
801 if (vp->v_usecount == 0) {
802 vprint("vrele: bad ref count", vp);
803 panic("vrele: ref cnt");
804 }
805 #endif
806 vp->v_usecount--;
807 if (vp->v_usecount > 0) {
808 return (0);
809 }
810
811 #ifdef DIAGNOSTIC
812 if (vp->v_writecount != 0) {
813 vprint("vrele: bad writecount", vp);
814 panic("vrele: v_writecount != 0");
815 }
816 #endif
817
818 if (vn_lock(vp, LK_EXCLUSIVE)) {
819 #ifdef DIAGNOSTIC
820 vprint("vrele: cannot lock", vp);
821 #endif
822 return (1);
823 }
824
825 VOP_INACTIVE(vp, p);
826
827 s = splbio();
828 if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
829 vputonfreelist(vp);
830 splx(s);
831 return (1);
832 }
833
834 /* Page or buffer structure gets a reference. */
835 void
vhold(struct vnode * vp)836 vhold(struct vnode *vp)
837 {
838 int s;
839
840 s = splbio();
841
842 /*
843 * If it is on the freelist and the hold count is currently
844 * zero, move it to the hold list.
845 */
846 if ((vp->v_bioflag & VBIOONFREELIST) &&
847 vp->v_holdcnt == 0 && vp->v_usecount == 0) {
848 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
849 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
850 }
851 vp->v_holdcnt++;
852
853 splx(s);
854 }
855
856 /* Lose interest in a vnode. */
857 void
vdrop(struct vnode * vp)858 vdrop(struct vnode *vp)
859 {
860 int s;
861
862 s = splbio();
863
864 #ifdef DIAGNOSTIC
865 if (vp->v_holdcnt == 0)
866 panic("vdrop: zero holdcnt");
867 #endif
868
869 vp->v_holdcnt--;
870
871 /*
872 * If it is on the holdlist and the hold count drops to
873 * zero, move it to the free list.
874 */
875 if ((vp->v_bioflag & VBIOONFREELIST) &&
876 vp->v_holdcnt == 0 && vp->v_usecount == 0) {
877 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
878 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
879 }
880
881 splx(s);
882 }
883
884 /*
885 * Remove any vnodes in the vnode table belonging to mount point mp.
886 *
887 * If MNT_NOFORCE is specified, there should not be any active ones,
888 * return error if any are found (nb: this is a user error, not a
889 * system error). If MNT_FORCE is specified, detach any active vnodes
890 * that are found.
891 */
892 #ifdef DEBUG_SYSCTL
893 int busyprt = 0; /* print out busy vnodes */
894 struct ctldebug debug_vfs_busyprt = { "vfs_busyprt", &busyprt };
895 #endif
896
897 int
vfs_mount_foreach_vnode(struct mount * mp,int (* func)(struct vnode *,void *),void * arg)898 vfs_mount_foreach_vnode(struct mount *mp,
899 int (*func)(struct vnode *, void *), void *arg) {
900 struct vnode *vp, *nvp;
901 int error = 0;
902
903 loop:
904 TAILQ_FOREACH_SAFE(vp , &mp->mnt_vnodelist, v_mntvnodes, nvp) {
905 if (vp->v_mount != mp)
906 goto loop;
907
908 error = func(vp, arg);
909
910 if (error != 0)
911 break;
912 }
913
914 return (error);
915 }
916
917 struct vflush_args {
918 struct vnode *skipvp;
919 int busy;
920 int flags;
921 };
922
923 int
vflush_vnode(struct vnode * vp,void * arg)924 vflush_vnode(struct vnode *vp, void *arg)
925 {
926 struct vflush_args *va = arg;
927 struct proc *p = curproc;
928 int empty, s;
929
930 if (vp == va->skipvp) {
931 return (0);
932 }
933
934 if ((va->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
935 return (0);
936 }
937
938 /*
939 * If WRITECLOSE is set, only flush out regular file
940 * vnodes open for writing.
941 */
942 if ((va->flags & WRITECLOSE) &&
943 (vp->v_writecount == 0 || vp->v_type != VREG)) {
944 return (0);
945 }
946
947 /*
948 * With v_usecount == 0, all we need to do is clear
949 * out the vnode data structures and we are done.
950 */
951 if (vp->v_usecount == 0) {
952 vgonel(vp, p);
953 return (0);
954 }
955
956 /*
957 * If FORCECLOSE is set, forcibly close the vnode.
958 * For block or character devices, revert to an
959 * anonymous device. For all other files, just kill them.
960 */
961 if (va->flags & FORCECLOSE) {
962 if (vp->v_type != VBLK && vp->v_type != VCHR) {
963 vgonel(vp, p);
964 } else {
965 vclean(vp, 0, p);
966 vp->v_op = &spec_vops;
967 insmntque(vp, NULL);
968 }
969 return (0);
970 }
971
972 /*
973 * If set, this is allowed to ignore vnodes which don't
974 * have changes pending to disk.
975 * XXX Might be nice to check per-fs "inode" flags, but
976 * generally the filesystem is sync'd already, right?
977 */
978 s = splbio();
979 empty = (va->flags & IGNORECLEAN) && LIST_EMPTY(&vp->v_dirtyblkhd);
980 splx(s);
981
982 if (empty)
983 return (0);
984
985 #ifdef DEBUG_SYSCTL
986 if (busyprt)
987 vprint("vflush: busy vnode", vp);
988 #endif
989 va->busy++;
990 return (0);
991 }
992
993 int
vflush(struct mount * mp,struct vnode * skipvp,int flags)994 vflush(struct mount *mp, struct vnode *skipvp, int flags)
995 {
996 struct vflush_args va;
997 va.skipvp = skipvp;
998 va.busy = 0;
999 va.flags = flags;
1000
1001 vfs_mount_foreach_vnode(mp, vflush_vnode, &va);
1002
1003 if (va.busy)
1004 return (EBUSY);
1005 return (0);
1006 }
1007
1008 /*
1009 * Disassociate the underlying file system from a vnode.
1010 */
1011 void
vclean(struct vnode * vp,int flags,struct proc * p)1012 vclean(struct vnode *vp, int flags, struct proc *p)
1013 {
1014 int active, do_wakeup = 0;
1015 int s;
1016
1017 /*
1018 * Check to see if the vnode is in use.
1019 * If so we have to reference it before we clean it out
1020 * so that its count cannot fall to zero and generate a
1021 * race against ourselves to recycle it.
1022 */
1023 if ((active = vp->v_usecount) != 0)
1024 vp->v_usecount++;
1025
1026 /*
1027 * Prevent the vnode from being recycled or
1028 * brought into use while we clean it out.
1029 */
1030 mtx_enter(&vnode_mtx);
1031 if (vp->v_lflag & VXLOCK)
1032 panic("vclean: deadlock");
1033 vp->v_lflag |= VXLOCK;
1034
1035 if (vp->v_lockcount > 0) {
1036 /*
1037 * Ensure that any thread currently waiting on the same lock has
1038 * observed that the vnode is about to be exclusively locked
1039 * before continuing.
1040 */
1041 msleep_nsec(&vp->v_lockcount, &vnode_mtx, PINOD, "vop_lock",
1042 INFSLP);
1043 KASSERT(vp->v_lockcount == 0);
1044 }
1045 mtx_leave(&vnode_mtx);
1046
1047 /*
1048 * Even if the count is zero, the VOP_INACTIVE routine may still
1049 * have the object locked while it cleans it out. The VOP_LOCK
1050 * ensures that the VOP_INACTIVE routine is done with its work.
1051 * For active vnodes, it ensures that no other activity can
1052 * occur while the underlying object is being cleaned out.
1053 */
1054 VOP_LOCK(vp, LK_EXCLUSIVE | LK_DRAIN);
1055
1056 /*
1057 * Clean out any VM data associated with the vnode.
1058 */
1059 uvm_vnp_terminate(vp);
1060 /*
1061 * Clean out any buffers associated with the vnode.
1062 */
1063 if (flags & DOCLOSE)
1064 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, INFSLP);
1065 /*
1066 * If purging an active vnode, it must be closed and
1067 * deactivated before being reclaimed. Note that the
1068 * VOP_INACTIVE will unlock the vnode
1069 */
1070 if (active) {
1071 if (flags & DOCLOSE)
1072 VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1073 VOP_INACTIVE(vp, p);
1074 } else {
1075 /*
1076 * Any other processes trying to obtain this lock must first
1077 * wait for VXLOCK to clear, then call the new lock operation.
1078 */
1079 VOP_UNLOCK(vp);
1080 }
1081
1082 /*
1083 * Reclaim the vnode.
1084 */
1085 if (VOP_RECLAIM(vp, p))
1086 panic("vclean: cannot reclaim");
1087 if (active) {
1088 vp->v_usecount--;
1089 if (vp->v_usecount == 0) {
1090 s = splbio();
1091 if (vp->v_holdcnt > 0)
1092 panic("vclean: not clean");
1093 vputonfreelist(vp);
1094 splx(s);
1095 }
1096 }
1097 cache_purge(vp);
1098
1099 /*
1100 * Done with purge, notify sleepers of the grim news.
1101 */
1102 vp->v_op = &dead_vops;
1103 VN_KNOTE(vp, NOTE_REVOKE);
1104 vp->v_tag = VT_NON;
1105 #ifdef VFSLCKDEBUG
1106 vp->v_flag &= ~VLOCKSWORK;
1107 #endif
1108 mtx_enter(&vnode_mtx);
1109 vp->v_lflag &= ~VXLOCK;
1110 if (vp->v_lflag & VXWANT) {
1111 vp->v_lflag &= ~VXWANT;
1112 do_wakeup = 1;
1113 }
1114 mtx_leave(&vnode_mtx);
1115 if (do_wakeup)
1116 wakeup(vp);
1117 }
1118
1119 /*
1120 * Recycle an unused vnode to the front of the free list.
1121 */
1122 int
vrecycle(struct vnode * vp,struct proc * p)1123 vrecycle(struct vnode *vp, struct proc *p)
1124 {
1125 if (vp->v_usecount == 0) {
1126 vgonel(vp, p);
1127 return (1);
1128 }
1129 return (0);
1130 }
1131
1132 /*
1133 * Eliminate all activity associated with a vnode
1134 * in preparation for reuse.
1135 */
1136 void
vgone(struct vnode * vp)1137 vgone(struct vnode *vp)
1138 {
1139 struct proc *p = curproc;
1140 vgonel(vp, p);
1141 }
1142
1143 /*
1144 * vgone, with struct proc.
1145 */
1146 void
vgonel(struct vnode * vp,struct proc * p)1147 vgonel(struct vnode *vp, struct proc *p)
1148 {
1149 struct vnode *vq;
1150 struct vnode *vx;
1151 int s;
1152
1153 KASSERT(vp->v_uvcount == 0);
1154
1155 /*
1156 * If a vgone (or vclean) is already in progress,
1157 * wait until it is done and return.
1158 */
1159 mtx_enter(&vnode_mtx);
1160 if (vp->v_lflag & VXLOCK) {
1161 vp->v_lflag |= VXWANT;
1162 msleep_nsec(vp, &vnode_mtx, PINOD, "vgone", INFSLP);
1163 mtx_leave(&vnode_mtx);
1164 return;
1165 }
1166 mtx_leave(&vnode_mtx);
1167
1168 /*
1169 * Clean out the filesystem specific data.
1170 */
1171 vclean(vp, DOCLOSE, p);
1172 /*
1173 * Delete from old mount point vnode list, if on one.
1174 */
1175 if (vp->v_mount != NULL)
1176 insmntque(vp, NULL);
1177 /*
1178 * If special device, remove it from special device alias list
1179 * if it is on one.
1180 */
1181 if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1182 vp->v_specinfo != NULL) {
1183 if ((vp->v_flag & VALIASED) == 0 && vp->v_type == VCHR &&
1184 (cdevsw[major(vp->v_rdev)].d_flags & D_CLONE) &&
1185 (minor(vp->v_rdev) >> CLONE_SHIFT == 0)) {
1186 free(vp->v_specbitmap, M_VNODE, CLONE_MAPSZ);
1187 }
1188 SLIST_REMOVE(vp->v_hashchain, vp, vnode, v_specnext);
1189 if (vp->v_flag & VALIASED) {
1190 vx = NULL;
1191 SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1192 if (vq->v_rdev != vp->v_rdev ||
1193 vq->v_type != vp->v_type)
1194 continue;
1195 if (vx)
1196 break;
1197 vx = vq;
1198 }
1199 if (vx == NULL)
1200 panic("missing alias");
1201 if (vq == NULL)
1202 vx->v_flag &= ~VALIASED;
1203 vp->v_flag &= ~VALIASED;
1204 }
1205 lf_purgelocks(&vp->v_speclockf);
1206 free(vp->v_specinfo, M_VNODE, sizeof(struct specinfo));
1207 vp->v_specinfo = NULL;
1208 }
1209 /*
1210 * If it is on the freelist and not already at the head,
1211 * move it to the head of the list.
1212 */
1213 vp->v_type = VBAD;
1214
1215 /*
1216 * Move onto the free list, unless we were called from
1217 * getnewvnode and we're not on any free list
1218 */
1219 s = splbio();
1220 if (vp->v_usecount == 0 &&
1221 (vp->v_bioflag & VBIOONFREELIST)) {
1222 if (vp->v_holdcnt > 0)
1223 panic("vgonel: not clean");
1224
1225 if (TAILQ_FIRST(&vnode_free_list) != vp) {
1226 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1227 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1228 }
1229 }
1230 splx(s);
1231 }
1232
1233 /*
1234 * Lookup a vnode by device number.
1235 */
1236 int
vfinddev(dev_t dev,enum vtype type,struct vnode ** vpp)1237 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1238 {
1239 struct vnode *vp;
1240 int rc =0;
1241
1242 SLIST_FOREACH(vp, &speclisth[SPECHASH(dev)], v_specnext) {
1243 if (dev != vp->v_rdev || type != vp->v_type)
1244 continue;
1245 *vpp = vp;
1246 rc = 1;
1247 break;
1248 }
1249 return (rc);
1250 }
1251
1252 /*
1253 * Revoke all the vnodes corresponding to the specified minor number
1254 * range (endpoints inclusive) of the specified major.
1255 */
1256 void
vdevgone(int maj,int minl,int minh,enum vtype type)1257 vdevgone(int maj, int minl, int minh, enum vtype type)
1258 {
1259 struct vnode *vp;
1260 int mn;
1261
1262 for (mn = minl; mn <= minh; mn++)
1263 if (vfinddev(makedev(maj, mn), type, &vp))
1264 VOP_REVOKE(vp, REVOKEALL);
1265 }
1266
1267 /*
1268 * Calculate the total number of references to a special device.
1269 */
1270 int
vcount(struct vnode * vp)1271 vcount(struct vnode *vp)
1272 {
1273 struct vnode *vq;
1274 int count;
1275
1276 loop:
1277 if ((vp->v_flag & VALIASED) == 0)
1278 return (vp->v_usecount);
1279 count = 0;
1280 SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1281 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1282 continue;
1283 /*
1284 * Alias, but not in use, so flush it out.
1285 */
1286 if (vq->v_usecount == 0 && vq != vp) {
1287 vgone(vq);
1288 goto loop;
1289 }
1290 count += vq->v_usecount;
1291 }
1292 return (count);
1293 }
1294
1295 #if defined(DEBUG) || defined(DIAGNOSTIC)
1296 /*
1297 * Print out a description of a vnode.
1298 */
1299 static char *typename[] =
1300 { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1301
1302 void
vprint(char * label,struct vnode * vp)1303 vprint(char *label, struct vnode *vp)
1304 {
1305 char buf[64];
1306
1307 if (label != NULL)
1308 printf("%s: ", label);
1309 printf("%p, type %s, use %u, write %u, hold %u,",
1310 vp, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1311 vp->v_holdcnt);
1312 buf[0] = '\0';
1313 if (vp->v_flag & VROOT)
1314 strlcat(buf, "|VROOT", sizeof buf);
1315 if (vp->v_flag & VTEXT)
1316 strlcat(buf, "|VTEXT", sizeof buf);
1317 if (vp->v_flag & VSYSTEM)
1318 strlcat(buf, "|VSYSTEM", sizeof buf);
1319 if (vp->v_lflag & VXLOCK)
1320 strlcat(buf, "|VXLOCK", sizeof buf);
1321 if (vp->v_lflag & VXWANT)
1322 strlcat(buf, "|VXWANT", sizeof buf);
1323 if (vp->v_bioflag & VBIOWAIT)
1324 strlcat(buf, "|VBIOWAIT", sizeof buf);
1325 if (vp->v_bioflag & VBIOONFREELIST)
1326 strlcat(buf, "|VBIOONFREELIST", sizeof buf);
1327 if (vp->v_bioflag & VBIOONSYNCLIST)
1328 strlcat(buf, "|VBIOONSYNCLIST", sizeof buf);
1329 if (vp->v_flag & VALIASED)
1330 strlcat(buf, "|VALIASED", sizeof buf);
1331 if (buf[0] != '\0')
1332 printf(" flags (%s)", &buf[1]);
1333 if (vp->v_data == NULL) {
1334 printf("\n");
1335 } else {
1336 printf("\n\t");
1337 VOP_PRINT(vp);
1338 }
1339 }
1340 #endif /* DEBUG || DIAGNOSTIC */
1341
1342 #ifdef DEBUG
1343 /*
1344 * List all of the locked vnodes in the system.
1345 * Called when debugging the kernel.
1346 */
1347 void
printlockedvnodes(void)1348 printlockedvnodes(void)
1349 {
1350 struct mount *mp;
1351 struct vnode *vp;
1352
1353 printf("Locked vnodes\n");
1354
1355 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1356 if (vfs_busy(mp, VB_READ|VB_NOWAIT))
1357 continue;
1358 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1359 if (VOP_ISLOCKED(vp))
1360 vprint(NULL, vp);
1361 }
1362 vfs_unbusy(mp);
1363 }
1364
1365 }
1366 #endif
1367
1368 /*
1369 * Top level filesystem related information gathering.
1370 */
1371 int
vfs_sysctl(int * name,u_int namelen,void * oldp,size_t * oldlenp,void * newp,size_t newlen,struct proc * p)1372 vfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1373 size_t newlen, struct proc *p)
1374 {
1375 struct vfsconf *vfsp, *tmpvfsp;
1376 int ret;
1377
1378 /* all sysctl names at this level are at least name and field */
1379 if (namelen < 2)
1380 return (ENOTDIR); /* overloaded */
1381
1382 if (name[0] != VFS_GENERIC) {
1383 vfsp = vfs_bytypenum(name[0]);
1384 if (vfsp == NULL || vfsp->vfc_vfsops->vfs_sysctl == NULL)
1385 return (EOPNOTSUPP);
1386
1387 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1388 oldp, oldlenp, newp, newlen, p));
1389 }
1390
1391 switch (name[1]) {
1392 case VFS_MAXTYPENUM:
1393 return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf));
1394
1395 case VFS_CONF:
1396 if (namelen < 3)
1397 return (ENOTDIR); /* overloaded */
1398
1399 vfsp = vfs_bytypenum(name[2]);
1400 if (vfsp == NULL)
1401 return (EOPNOTSUPP);
1402
1403 /* Make a copy, clear out kernel pointers */
1404 tmpvfsp = malloc(sizeof(*tmpvfsp), M_TEMP, M_WAITOK|M_ZERO);
1405 memcpy(tmpvfsp, vfsp, sizeof(*tmpvfsp));
1406 tmpvfsp->vfc_vfsops = NULL;
1407
1408 ret = sysctl_rdstruct(oldp, oldlenp, newp, tmpvfsp,
1409 sizeof(struct vfsconf));
1410
1411 free(tmpvfsp, M_TEMP, sizeof(*tmpvfsp));
1412 return (ret);
1413 case VFS_BCACHESTAT: /* buffer cache statistics */
1414 ret = sysctl_rdstruct(oldp, oldlenp, newp, &bcstats,
1415 sizeof(struct bcachestats));
1416 return(ret);
1417 }
1418 return (EOPNOTSUPP);
1419 }
1420
1421 /*
1422 * Check to see if a filesystem is mounted on a block device.
1423 */
1424 int
vfs_mountedon(struct vnode * vp)1425 vfs_mountedon(struct vnode *vp)
1426 {
1427 struct vnode *vq;
1428 int error = 0;
1429
1430 if (vp->v_specmountpoint != NULL)
1431 return (EBUSY);
1432 if (vp->v_flag & VALIASED) {
1433 SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1434 if (vq->v_rdev != vp->v_rdev ||
1435 vq->v_type != vp->v_type)
1436 continue;
1437 if (vq->v_specmountpoint != NULL) {
1438 error = EBUSY;
1439 break;
1440 }
1441 }
1442 }
1443 return (error);
1444 }
1445
1446 #ifdef NFSSERVER
1447 /*
1448 * Build hash lists of net addresses and hang them off the mount point.
1449 * Called by vfs_export() to set up the lists of export addresses.
1450 */
1451 int
vfs_hang_addrlist(struct mount * mp,struct netexport * nep,struct export_args * argp)1452 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1453 struct export_args *argp)
1454 {
1455 struct netcred *np;
1456 struct radix_node_head *rnh;
1457 int nplen, i;
1458 struct radix_node *rn;
1459 struct sockaddr *saddr, *smask = NULL;
1460 int error;
1461
1462 if (argp->ex_addrlen == 0) {
1463 if (mp->mnt_flag & MNT_DEFEXPORTED)
1464 return (EPERM);
1465 np = &nep->ne_defexported;
1466 /* fill in the kernel's ucred from userspace's xucred */
1467 if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1468 return (error);
1469 mp->mnt_flag |= MNT_DEFEXPORTED;
1470 goto finish;
1471 }
1472 if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN ||
1473 argp->ex_addrlen < 0 || argp->ex_masklen < 0)
1474 return (EINVAL);
1475 nplen = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1476 np = (struct netcred *)malloc(nplen, M_NETADDR, M_WAITOK|M_ZERO);
1477 np->netc_len = nplen;
1478 saddr = (struct sockaddr *)(np + 1);
1479 error = copyin(argp->ex_addr, saddr, argp->ex_addrlen);
1480 if (error)
1481 goto out;
1482 if (saddr->sa_len > argp->ex_addrlen)
1483 saddr->sa_len = argp->ex_addrlen;
1484 if (argp->ex_masklen) {
1485 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1486 error = copyin(argp->ex_mask, smask, argp->ex_masklen);
1487 if (error)
1488 goto out;
1489 if (smask->sa_len > argp->ex_masklen)
1490 smask->sa_len = argp->ex_masklen;
1491 }
1492 /* fill in the kernel's ucred from userspace's xucred */
1493 if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1494 goto out;
1495 i = saddr->sa_family;
1496 switch (i) {
1497 case AF_INET:
1498 if ((rnh = nep->ne_rtable_inet) == NULL) {
1499 if (!rn_inithead((void **)&nep->ne_rtable_inet,
1500 offsetof(struct sockaddr_in, sin_addr))) {
1501 error = ENOBUFS;
1502 goto out;
1503 }
1504 rnh = nep->ne_rtable_inet;
1505 }
1506 break;
1507 default:
1508 error = EINVAL;
1509 goto out;
1510 }
1511 rn = rn_addroute(saddr, smask, rnh, np->netc_rnodes, 0);
1512 if (rn == NULL || np != (struct netcred *)rn) { /* already exists */
1513 error = EPERM;
1514 goto out;
1515 }
1516 finish:
1517 np->netc_exflags = argp->ex_flags;
1518 return (0);
1519 out:
1520 free(np, M_NETADDR, np->netc_len);
1521 return (error);
1522 }
1523
1524 int
vfs_free_netcred(struct radix_node * rn,void * w,u_int id)1525 vfs_free_netcred(struct radix_node *rn, void *w, u_int id)
1526 {
1527 struct radix_node_head *rnh = (struct radix_node_head *)w;
1528 struct netcred * np = (struct netcred *)rn;
1529
1530 rn_delete(rn->rn_key, rn->rn_mask, rnh, NULL);
1531 free(np, M_NETADDR, np->netc_len);
1532 return (0);
1533 }
1534
1535 /*
1536 * Free the net address hash lists that are hanging off the mount points.
1537 */
1538 void
vfs_free_addrlist(struct netexport * nep)1539 vfs_free_addrlist(struct netexport *nep)
1540 {
1541 struct radix_node_head *rnh;
1542
1543 if ((rnh = nep->ne_rtable_inet) != NULL) {
1544 rn_walktree(rnh, vfs_free_netcred, rnh);
1545 free(rnh, M_RTABLE, sizeof(*rnh));
1546 nep->ne_rtable_inet = NULL;
1547 }
1548 }
1549 #endif /* NFSSERVER */
1550
1551 int
vfs_export(struct mount * mp,struct netexport * nep,struct export_args * argp)1552 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1553 {
1554 #ifdef NFSSERVER
1555 int error;
1556
1557 if (argp->ex_flags & MNT_DELEXPORT) {
1558 vfs_free_addrlist(nep);
1559 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1560 }
1561 if (argp->ex_flags & MNT_EXPORTED) {
1562 if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
1563 return (error);
1564 mp->mnt_flag |= MNT_EXPORTED;
1565 }
1566 return (0);
1567 #else
1568 return (ENOTSUP);
1569 #endif /* NFSSERVER */
1570 }
1571
1572 struct netcred *
vfs_export_lookup(struct mount * mp,struct netexport * nep,struct mbuf * nam)1573 vfs_export_lookup(struct mount *mp, struct netexport *nep, struct mbuf *nam)
1574 {
1575 #ifdef NFSSERVER
1576 struct netcred *np;
1577 struct radix_node_head *rnh;
1578 struct sockaddr *saddr;
1579
1580 np = NULL;
1581 if (mp->mnt_flag & MNT_EXPORTED) {
1582 /*
1583 * Lookup in the export list first.
1584 */
1585 if (nam != NULL) {
1586 saddr = mtod(nam, struct sockaddr *);
1587 switch(saddr->sa_family) {
1588 case AF_INET:
1589 rnh = nep->ne_rtable_inet;
1590 break;
1591 default:
1592 rnh = NULL;
1593 break;
1594 }
1595 if (rnh != NULL)
1596 np = (struct netcred *)rn_match(saddr, rnh);
1597 }
1598 /*
1599 * If no address match, use the default if it exists.
1600 */
1601 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1602 np = &nep->ne_defexported;
1603 }
1604 return (np);
1605 #else
1606 return (NULL);
1607 #endif /* NFSSERVER */
1608 }
1609
1610 /*
1611 * Do the usual access checking.
1612 * file_mode, uid and gid are from the vnode in question,
1613 * while acc_mode and cred are from the VOP_ACCESS parameter list
1614 */
1615 int
vaccess(enum vtype type,mode_t file_mode,uid_t uid,gid_t gid,mode_t acc_mode,struct ucred * cred)1616 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1617 mode_t acc_mode, struct ucred *cred)
1618 {
1619 mode_t mask;
1620
1621 /* User id 0 always gets read/write access. */
1622 if (cred->cr_uid == 0) {
1623 /* For VEXEC, at least one of the execute bits must be set. */
1624 if ((acc_mode & VEXEC) && type != VDIR &&
1625 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1626 return EACCES;
1627 return 0;
1628 }
1629
1630 mask = 0;
1631
1632 /* Otherwise, check the owner. */
1633 if (cred->cr_uid == uid) {
1634 if (acc_mode & VEXEC)
1635 mask |= S_IXUSR;
1636 if (acc_mode & VREAD)
1637 mask |= S_IRUSR;
1638 if (acc_mode & VWRITE)
1639 mask |= S_IWUSR;
1640 return (file_mode & mask) == mask ? 0 : EACCES;
1641 }
1642
1643 /* Otherwise, check the groups. */
1644 if (groupmember(gid, cred)) {
1645 if (acc_mode & VEXEC)
1646 mask |= S_IXGRP;
1647 if (acc_mode & VREAD)
1648 mask |= S_IRGRP;
1649 if (acc_mode & VWRITE)
1650 mask |= S_IWGRP;
1651 return (file_mode & mask) == mask ? 0 : EACCES;
1652 }
1653
1654 /* Otherwise, check everyone else. */
1655 if (acc_mode & VEXEC)
1656 mask |= S_IXOTH;
1657 if (acc_mode & VREAD)
1658 mask |= S_IROTH;
1659 if (acc_mode & VWRITE)
1660 mask |= S_IWOTH;
1661 return (file_mode & mask) == mask ? 0 : EACCES;
1662 }
1663
1664 int
vnoperm(struct vnode * vp)1665 vnoperm(struct vnode *vp)
1666 {
1667 if (vp->v_flag & VROOT || vp->v_mount == NULL)
1668 return 0;
1669
1670 return (vp->v_mount->mnt_flag & MNT_NOPERM);
1671 }
1672
1673 struct rwlock vfs_stall_lock = RWLOCK_INITIALIZER("vfs_stall");
1674 unsigned int vfs_stalling = 0;
1675
1676 int
vfs_stall(struct proc * p,int stall)1677 vfs_stall(struct proc *p, int stall)
1678 {
1679 struct mount *mp;
1680 int allerror = 0, error;
1681
1682 if (stall) {
1683 atomic_inc_int(&vfs_stalling);
1684 rw_enter_write(&vfs_stall_lock);
1685 }
1686
1687 /*
1688 * The loop variable mp is protected by vfs_busy() so that it cannot
1689 * be unmounted while VFS_SYNC() sleeps. Traverse forward to keep the
1690 * lock order consistent with dounmount().
1691 */
1692 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1693 if (stall) {
1694 error = vfs_busy(mp, VB_WRITE|VB_WAIT|VB_DUPOK);
1695 if (error) {
1696 printf("%s: busy\n", mp->mnt_stat.f_mntonname);
1697 allerror = error;
1698 continue;
1699 }
1700 uvm_vnp_sync(mp);
1701 error = VFS_SYNC(mp, MNT_WAIT, stall, p->p_ucred, p);
1702 if (error) {
1703 printf("%s: failed to sync\n",
1704 mp->mnt_stat.f_mntonname);
1705 vfs_unbusy(mp);
1706 allerror = error;
1707 continue;
1708 }
1709 mp->mnt_flag |= MNT_STALLED;
1710 } else {
1711 if (mp->mnt_flag & MNT_STALLED) {
1712 vfs_unbusy(mp);
1713 mp->mnt_flag &= ~MNT_STALLED;
1714 }
1715 }
1716 }
1717
1718 if (!stall) {
1719 rw_exit_write(&vfs_stall_lock);
1720 atomic_dec_int(&vfs_stalling);
1721 }
1722
1723 return (allerror);
1724 }
1725
1726 void
vfs_stall_barrier(void)1727 vfs_stall_barrier(void)
1728 {
1729 if (__predict_false(vfs_stalling)) {
1730 rw_enter_read(&vfs_stall_lock);
1731 rw_exit_read(&vfs_stall_lock);
1732 }
1733 }
1734
1735 /*
1736 * Unmount all file systems.
1737 * We traverse the list in reverse order under the assumption that doing so
1738 * will avoid needing to worry about dependencies.
1739 */
1740 void
vfs_unmountall(void)1741 vfs_unmountall(void)
1742 {
1743 struct mount *mp, *nmp;
1744 int allerror, error, again = 1;
1745
1746 retry:
1747 allerror = 0;
1748 TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, nmp) {
1749 if (vfs_busy(mp, VB_WRITE|VB_NOWAIT))
1750 continue;
1751 /* XXX Here is a race, the next pointer is not locked. */
1752 if ((error = dounmount(mp, MNT_FORCE, curproc)) != 0) {
1753 printf("unmount of %s failed with error %d\n",
1754 mp->mnt_stat.f_mntonname, error);
1755 allerror = 1;
1756 }
1757 }
1758
1759 if (allerror) {
1760 printf("WARNING: some file systems would not unmount\n");
1761 if (again) {
1762 printf("retrying\n");
1763 again = 0;
1764 goto retry;
1765 }
1766 }
1767 }
1768
1769 /*
1770 * Sync and unmount file systems before shutting down.
1771 */
1772 void
vfs_shutdown(struct proc * p)1773 vfs_shutdown(struct proc *p)
1774 {
1775 #ifdef ACCOUNTING
1776 acct_shutdown();
1777 #endif
1778
1779 printf("syncing disks...");
1780
1781 if (panicstr == NULL) {
1782 /* Sync before unmount, in case we hang on something. */
1783 sys_sync(p, NULL, NULL);
1784 vfs_unmountall();
1785 }
1786
1787 #if NSOFTRAID > 0
1788 sr_quiesce();
1789 #endif
1790
1791 if (vfs_syncwait(p, 1))
1792 printf(" giving up\n");
1793 else
1794 printf(" done\n");
1795 }
1796
1797 /*
1798 * perform sync() operation and wait for buffers to flush.
1799 */
1800 int
vfs_syncwait(struct proc * p,int verbose)1801 vfs_syncwait(struct proc *p, int verbose)
1802 {
1803 struct buf *bp;
1804 int iter, nbusy, dcount, s;
1805 #ifdef MULTIPROCESSOR
1806 int hold_count;
1807 #endif
1808
1809 sys_sync(p, NULL, NULL);
1810
1811 /* Wait for sync to finish. */
1812 dcount = 10000;
1813 for (iter = 0; iter < 20; iter++) {
1814 nbusy = 0;
1815 LIST_FOREACH(bp, &bufhead, b_list) {
1816 if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
1817 nbusy++;
1818 /*
1819 * With soft updates, some buffers that are
1820 * written will be remarked as dirty until other
1821 * buffers are written.
1822 *
1823 * XXX here be dragons. this should really go away
1824 * but should be carefully made to go away on it's
1825 * own with testing.. XXX
1826 */
1827 if (bp->b_flags & B_DELWRI) {
1828 s = splbio();
1829 bremfree(bp);
1830 buf_acquire(bp);
1831 splx(s);
1832 nbusy++;
1833 bawrite(bp);
1834 if (dcount-- <= 0) {
1835 if (verbose)
1836 printf("softdep ");
1837 return 1;
1838 }
1839 }
1840 }
1841 if (nbusy == 0)
1842 break;
1843 if (verbose)
1844 printf("%d ", nbusy);
1845 #ifdef MULTIPROCESSOR
1846 if (_kernel_lock_held())
1847 hold_count = __mp_release_all(&kernel_lock);
1848 else
1849 hold_count = 0;
1850 #endif
1851 DELAY(40000 * iter);
1852 #ifdef MULTIPROCESSOR
1853 if (hold_count)
1854 __mp_acquire_count(&kernel_lock, hold_count);
1855 #endif
1856 }
1857
1858 return nbusy;
1859 }
1860
1861 /*
1862 * posix file system related system variables.
1863 */
1864 int
fs_posix_sysctl(int * name,u_int namelen,void * oldp,size_t * oldlenp,void * newp,size_t newlen,struct proc * p)1865 fs_posix_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
1866 void *newp, size_t newlen, struct proc *p)
1867 {
1868 /* all sysctl names at this level are terminal */
1869 if (namelen != 1)
1870 return (ENOTDIR);
1871
1872 switch (name[0]) {
1873 case FS_POSIX_SETUID:
1874 return (sysctl_securelevel_int(oldp, oldlenp, newp, newlen,
1875 &suid_clear));
1876 default:
1877 return (EOPNOTSUPP);
1878 }
1879 /* NOTREACHED */
1880 }
1881
1882 /*
1883 * file system related system variables.
1884 */
1885 int
fs_sysctl(int * name,u_int namelen,void * oldp,size_t * oldlenp,void * newp,size_t newlen,struct proc * p)1886 fs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1887 size_t newlen, struct proc *p)
1888 {
1889 sysctlfn *fn;
1890
1891 switch (name[0]) {
1892 case FS_POSIX:
1893 fn = fs_posix_sysctl;
1894 break;
1895 default:
1896 return (EOPNOTSUPP);
1897 }
1898 return (*fn)(name + 1, namelen - 1, oldp, oldlenp, newp, newlen, p);
1899 }
1900
1901
1902 /*
1903 * Routines dealing with vnodes and buffers
1904 */
1905
1906 /*
1907 * Wait for all outstanding I/Os to complete
1908 *
1909 * Manipulates v_numoutput. Must be called at splbio()
1910 */
1911 int
vwaitforio(struct vnode * vp,int slpflag,char * wmesg,uint64_t timeo)1912 vwaitforio(struct vnode *vp, int slpflag, char *wmesg, uint64_t timeo)
1913 {
1914 int error = 0;
1915
1916 splassert(IPL_BIO);
1917
1918 while (vp->v_numoutput) {
1919 vp->v_bioflag |= VBIOWAIT;
1920 error = tsleep_nsec(&vp->v_numoutput,
1921 slpflag | (PRIBIO + 1), wmesg, timeo);
1922 if (error)
1923 break;
1924 }
1925
1926 return (error);
1927 }
1928
1929 /*
1930 * Update outstanding I/O count and do wakeup if requested.
1931 *
1932 * Manipulates v_numoutput. Must be called at splbio()
1933 */
1934 void
vwakeup(struct vnode * vp)1935 vwakeup(struct vnode *vp)
1936 {
1937 splassert(IPL_BIO);
1938
1939 if (vp != NULL) {
1940 if (vp->v_numoutput-- == 0)
1941 panic("vwakeup: neg numoutput");
1942 if ((vp->v_bioflag & VBIOWAIT) && vp->v_numoutput == 0) {
1943 vp->v_bioflag &= ~VBIOWAIT;
1944 wakeup(&vp->v_numoutput);
1945 }
1946 }
1947 }
1948
1949 /*
1950 * Flush out and invalidate all buffers associated with a vnode.
1951 * Called with the underlying object locked.
1952 */
1953 int
vinvalbuf(struct vnode * vp,int flags,struct ucred * cred,struct proc * p,int slpflag,uint64_t slptimeo)1954 vinvalbuf(struct vnode *vp, int flags, struct ucred *cred, struct proc *p,
1955 int slpflag, uint64_t slptimeo)
1956 {
1957 struct buf *bp;
1958 struct buf *nbp, *blist;
1959 int s, error;
1960
1961 #ifdef VFSLCKDEBUG
1962 if ((vp->v_flag & VLOCKSWORK) && !VOP_ISLOCKED(vp))
1963 panic("%s: vp isn't locked, vp %p", __func__, vp);
1964 #endif
1965
1966 if (flags & V_SAVE) {
1967 s = splbio();
1968 vwaitforio(vp, 0, "vinvalbuf", INFSLP);
1969 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
1970 splx(s);
1971 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
1972 return (error);
1973 s = splbio();
1974 if (vp->v_numoutput > 0 ||
1975 !LIST_EMPTY(&vp->v_dirtyblkhd))
1976 panic("%s: dirty bufs, vp %p", __func__, vp);
1977 }
1978 splx(s);
1979 }
1980 loop:
1981 s = splbio();
1982 for (;;) {
1983 int count = 0;
1984 if ((blist = LIST_FIRST(&vp->v_cleanblkhd)) &&
1985 (flags & V_SAVEMETA))
1986 while (blist && blist->b_lblkno < 0)
1987 blist = LIST_NEXT(blist, b_vnbufs);
1988 if (blist == NULL &&
1989 (blist = LIST_FIRST(&vp->v_dirtyblkhd)) &&
1990 (flags & V_SAVEMETA))
1991 while (blist && blist->b_lblkno < 0)
1992 blist = LIST_NEXT(blist, b_vnbufs);
1993 if (!blist)
1994 break;
1995
1996 for (bp = blist; bp; bp = nbp) {
1997 nbp = LIST_NEXT(bp, b_vnbufs);
1998 if (flags & V_SAVEMETA && bp->b_lblkno < 0)
1999 continue;
2000 if (bp->b_flags & B_BUSY) {
2001 bp->b_flags |= B_WANTED;
2002 error = tsleep_nsec(bp, slpflag | (PRIBIO + 1),
2003 "vinvalbuf", slptimeo);
2004 if (error) {
2005 splx(s);
2006 return (error);
2007 }
2008 break;
2009 }
2010 bremfree(bp);
2011 /*
2012 * XXX Since there are no node locks for NFS, I believe
2013 * there is a slight chance that a delayed write will
2014 * occur while sleeping just above, so check for it.
2015 */
2016 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
2017 buf_acquire(bp);
2018 splx(s);
2019 (void) VOP_BWRITE(bp);
2020 goto loop;
2021 }
2022 buf_acquire_nomap(bp);
2023 bp->b_flags |= B_INVAL;
2024 brelse(bp);
2025 count++;
2026 /*
2027 * XXX Temporary workaround XXX
2028 *
2029 * If this is a gigantisch vnode and we are
2030 * trashing a ton of buffers, drop the lock
2031 * and yield every so often. The longer term
2032 * fix is to add a separate list for these
2033 * invalid buffers so we don't have to do the
2034 * work to free these here.
2035 */
2036 if (count > 100) {
2037 splx(s);
2038 sched_pause(yield);
2039 goto loop;
2040 }
2041 }
2042 }
2043 if (!(flags & V_SAVEMETA) &&
2044 (!LIST_EMPTY(&vp->v_dirtyblkhd) || !LIST_EMPTY(&vp->v_cleanblkhd)))
2045 panic("%s: flush failed, vp %p", __func__, vp);
2046 splx(s);
2047 return (0);
2048 }
2049
2050 void
vflushbuf(struct vnode * vp,int sync)2051 vflushbuf(struct vnode *vp, int sync)
2052 {
2053 struct buf *bp, *nbp;
2054 int s;
2055
2056 loop:
2057 s = splbio();
2058 LIST_FOREACH_SAFE(bp, &vp->v_dirtyblkhd, b_vnbufs, nbp) {
2059 if ((bp->b_flags & B_BUSY))
2060 continue;
2061 if ((bp->b_flags & B_DELWRI) == 0)
2062 panic("vflushbuf: not dirty");
2063 bremfree(bp);
2064 buf_acquire(bp);
2065 splx(s);
2066 /*
2067 * Wait for I/O associated with indirect blocks to complete,
2068 * since there is no way to quickly wait for them below.
2069 */
2070 if (bp->b_vp == vp || sync == 0)
2071 (void) bawrite(bp);
2072 else
2073 (void) bwrite(bp);
2074 goto loop;
2075 }
2076 if (sync == 0) {
2077 splx(s);
2078 return;
2079 }
2080 vwaitforio(vp, 0, "vflushbuf", INFSLP);
2081 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
2082 splx(s);
2083 #ifdef DIAGNOSTIC
2084 vprint("vflushbuf: dirty", vp);
2085 #endif
2086 goto loop;
2087 }
2088 splx(s);
2089 }
2090
2091 /*
2092 * Associate a buffer with a vnode.
2093 *
2094 * Manipulates buffer vnode queues. Must be called at splbio().
2095 */
2096 void
bgetvp(struct vnode * vp,struct buf * bp)2097 bgetvp(struct vnode *vp, struct buf *bp)
2098 {
2099 splassert(IPL_BIO);
2100
2101
2102 if (bp->b_vp)
2103 panic("bgetvp: not free");
2104 vhold(vp);
2105 bp->b_vp = vp;
2106 if (vp->v_type == VBLK || vp->v_type == VCHR)
2107 bp->b_dev = vp->v_rdev;
2108 else
2109 bp->b_dev = NODEV;
2110 /*
2111 * Insert onto list for new vnode.
2112 */
2113 bufinsvn(bp, &vp->v_cleanblkhd);
2114 }
2115
2116 /*
2117 * Disassociate a buffer from a vnode.
2118 *
2119 * Manipulates vnode buffer queues. Must be called at splbio().
2120 */
2121 void
brelvp(struct buf * bp)2122 brelvp(struct buf *bp)
2123 {
2124 struct vnode *vp;
2125
2126 splassert(IPL_BIO);
2127
2128 if ((vp = bp->b_vp) == (struct vnode *) 0)
2129 panic("brelvp: NULL");
2130 /*
2131 * Delete from old vnode list, if on one.
2132 */
2133 if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2134 bufremvn(bp);
2135 if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2136 LIST_EMPTY(&vp->v_dirtyblkhd)) {
2137 vp->v_bioflag &= ~VBIOONSYNCLIST;
2138 LIST_REMOVE(vp, v_synclist);
2139 }
2140 bp->b_vp = NULL;
2141
2142 vdrop(vp);
2143 }
2144
2145 /*
2146 * Replaces the current vnode associated with the buffer, if any,
2147 * with a new vnode.
2148 *
2149 * If an output I/O is pending on the buffer, the old vnode
2150 * I/O count is adjusted.
2151 *
2152 * Ignores vnode buffer queues. Must be called at splbio().
2153 */
2154 void
buf_replacevnode(struct buf * bp,struct vnode * newvp)2155 buf_replacevnode(struct buf *bp, struct vnode *newvp)
2156 {
2157 struct vnode *oldvp = bp->b_vp;
2158
2159 splassert(IPL_BIO);
2160
2161 if (oldvp)
2162 brelvp(bp);
2163
2164 if ((bp->b_flags & (B_READ | B_DONE)) == 0) {
2165 newvp->v_numoutput++; /* put it on swapdev */
2166 vwakeup(oldvp);
2167 }
2168
2169 bgetvp(newvp, bp);
2170 bufremvn(bp);
2171 }
2172
2173 /*
2174 * Used to assign buffers to the appropriate clean or dirty list on
2175 * the vnode and to add newly dirty vnodes to the appropriate
2176 * filesystem syncer list.
2177 *
2178 * Manipulates vnode buffer queues. Must be called at splbio().
2179 */
2180 void
reassignbuf(struct buf * bp)2181 reassignbuf(struct buf *bp)
2182 {
2183 struct buflists *listheadp;
2184 int delay;
2185 struct vnode *vp = bp->b_vp;
2186
2187 splassert(IPL_BIO);
2188
2189 /*
2190 * Delete from old vnode list, if on one.
2191 */
2192 if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2193 bufremvn(bp);
2194
2195 /*
2196 * If dirty, put on list of dirty buffers;
2197 * otherwise insert onto list of clean buffers.
2198 */
2199 if ((bp->b_flags & B_DELWRI) == 0) {
2200 listheadp = &vp->v_cleanblkhd;
2201 if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2202 LIST_EMPTY(&vp->v_dirtyblkhd)) {
2203 vp->v_bioflag &= ~VBIOONSYNCLIST;
2204 LIST_REMOVE(vp, v_synclist);
2205 }
2206 } else {
2207 listheadp = &vp->v_dirtyblkhd;
2208 if ((vp->v_bioflag & VBIOONSYNCLIST) == 0) {
2209 switch (vp->v_type) {
2210 case VDIR:
2211 delay = syncdelay / 2;
2212 break;
2213 case VBLK:
2214 if (vp->v_specmountpoint != NULL) {
2215 delay = syncdelay / 3;
2216 break;
2217 }
2218 /* FALLTHROUGH */
2219 default:
2220 delay = syncdelay;
2221 }
2222 vn_syncer_add_to_worklist(vp, delay);
2223 }
2224 }
2225 bufinsvn(bp, listheadp);
2226 }
2227
2228 #ifdef DDB
2229 #include <machine/db_machdep.h>
2230 #include <ddb/db_interface.h>
2231
2232 void
vfs_buf_print(void * b,int full,int (* pr)(const char *,...))2233 vfs_buf_print(void *b, int full,
2234 int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2235 {
2236 struct buf *bp = b;
2237
2238 (*pr)(" vp %p lblkno 0x%llx blkno 0x%llx dev 0x%x\n"
2239 " proc %p error %d flags %lb\n",
2240 bp->b_vp, (int64_t)bp->b_lblkno, (int64_t)bp->b_blkno, bp->b_dev,
2241 bp->b_proc, bp->b_error, bp->b_flags, B_BITS);
2242
2243 (*pr)(" bufsize 0x%lx bcount 0x%lx resid 0x%lx\n"
2244 " data %p saveaddr %p iodone %p\n",
2245 bp->b_bufsize, bp->b_bcount, (long)bp->b_resid,
2246 bp->b_data, bp->b_saveaddr,
2247 bp->b_iodone);
2248
2249 (*pr)(" dirty {off 0x%x end 0x%x} valid {off 0x%x end 0x%x}\n",
2250 bp->b_dirtyoff, bp->b_dirtyend, bp->b_validoff, bp->b_validend);
2251
2252 }
2253
2254 const char *vtypes[] = { VTYPE_NAMES };
2255 const char *vtags[] = { VTAG_NAMES };
2256
2257 void
vfs_vnode_print(void * v,int full,int (* pr)(const char *,...))2258 vfs_vnode_print(void *v, int full,
2259 int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2260 {
2261 struct vnode *vp = v;
2262
2263 (*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
2264 (u_int)vp->v_tag >= nitems(vtags)? "<unk>":vtags[vp->v_tag],
2265 vp->v_tag,
2266 (u_int)vp->v_type >= nitems(vtypes)? "<unk>":vtypes[vp->v_type],
2267 vp->v_type, vp->v_mount, vp->v_mountedhere);
2268
2269 (*pr)("data %p usecount %d writecount %d holdcnt %d numoutput %d\n",
2270 vp->v_data, vp->v_usecount, vp->v_writecount,
2271 vp->v_holdcnt, vp->v_numoutput);
2272
2273 /* uvm_object_printit(&vp->v_uobj, full, pr); */
2274
2275 if (full) {
2276 struct buf *bp;
2277
2278 (*pr)("clean bufs:\n");
2279 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2280 (*pr)(" bp %p\n", bp);
2281 vfs_buf_print(bp, full, pr);
2282 }
2283
2284 (*pr)("dirty bufs:\n");
2285 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2286 (*pr)(" bp %p\n", bp);
2287 vfs_buf_print(bp, full, pr);
2288 }
2289 }
2290 }
2291
2292 void
vfs_mount_print(struct mount * mp,int full,int (* pr)(const char *,...))2293 vfs_mount_print(struct mount *mp, int full,
2294 int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2295 {
2296 struct vfsconf *vfc = mp->mnt_vfc;
2297 struct vnode *vp;
2298 int cnt;
2299
2300 (*pr)("flags %b\nvnodecovered %p syncer %p data %p\n",
2301 mp->mnt_flag, MNT_BITS,
2302 mp->mnt_vnodecovered, mp->mnt_syncer, mp->mnt_data);
2303
2304 (*pr)("vfsconf: ops %p name \"%s\" num %d ref %u flags 0x%x\n",
2305 vfc->vfc_vfsops, vfc->vfc_name, vfc->vfc_typenum,
2306 vfc->vfc_refcount, vfc->vfc_flags);
2307
2308 (*pr)("statvfs cache: bsize %x iosize %x\n"
2309 "blocks %llu free %llu avail %lld\n",
2310 mp->mnt_stat.f_bsize, mp->mnt_stat.f_iosize, mp->mnt_stat.f_blocks,
2311 mp->mnt_stat.f_bfree, mp->mnt_stat.f_bavail);
2312
2313 (*pr)(" files %llu ffiles %llu favail %lld\n", mp->mnt_stat.f_files,
2314 mp->mnt_stat.f_ffree, mp->mnt_stat.f_favail);
2315
2316 (*pr)(" f_fsidx {0x%x, 0x%x} owner %u ctime 0x%llx\n",
2317 mp->mnt_stat.f_fsid.val[0], mp->mnt_stat.f_fsid.val[1],
2318 mp->mnt_stat.f_owner, mp->mnt_stat.f_ctime);
2319
2320 (*pr)(" syncwrites %llu asyncwrites = %llu\n",
2321 mp->mnt_stat.f_syncwrites, mp->mnt_stat.f_asyncwrites);
2322
2323 (*pr)(" syncreads %llu asyncreads = %llu\n",
2324 mp->mnt_stat.f_syncreads, mp->mnt_stat.f_asyncreads);
2325
2326 (*pr)(" fstype \"%s\" mnton \"%s\" mntfrom \"%s\" mntspec \"%s\"\n",
2327 mp->mnt_stat.f_fstypename, mp->mnt_stat.f_mntonname,
2328 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntfromspec);
2329
2330 (*pr)("locked vnodes:");
2331 /* XXX would take mountlist lock, except ddb has no context */
2332 cnt = 0;
2333 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2334 if (VOP_ISLOCKED(vp)) {
2335 if (cnt == 0)
2336 (*pr)("\n %p", vp);
2337 else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0)
2338 (*pr)(",\n %p", vp);
2339 else
2340 (*pr)(", %p", vp);
2341 cnt++;
2342 }
2343 }
2344 (*pr)("\n");
2345
2346 if (full) {
2347 (*pr)("all vnodes:");
2348 /* XXX would take mountlist lock, except ddb has no context */
2349 cnt = 0;
2350 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2351 if (cnt == 0)
2352 (*pr)("\n %p", vp);
2353 else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0)
2354 (*pr)(",\n %p", vp);
2355 else
2356 (*pr)(", %p", vp);
2357 cnt++;
2358 }
2359 (*pr)("\n");
2360 }
2361 }
2362 #endif /* DDB */
2363
2364 void
copy_statfs_info(struct statfs * sbp,const struct mount * mp)2365 copy_statfs_info(struct statfs *sbp, const struct mount *mp)
2366 {
2367 const struct statfs *mbp;
2368
2369 strncpy(sbp->f_fstypename, mp->mnt_vfc->vfc_name, MFSNAMELEN);
2370
2371 if (sbp == (mbp = &mp->mnt_stat))
2372 return;
2373
2374 sbp->f_fsid = mbp->f_fsid;
2375 sbp->f_owner = mbp->f_owner;
2376 sbp->f_flags = mbp->f_flags;
2377 sbp->f_syncwrites = mbp->f_syncwrites;
2378 sbp->f_asyncwrites = mbp->f_asyncwrites;
2379 sbp->f_syncreads = mbp->f_syncreads;
2380 sbp->f_asyncreads = mbp->f_asyncreads;
2381 sbp->f_namemax = mbp->f_namemax;
2382 memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
2383 memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
2384 memcpy(sbp->f_mntfromspec, mp->mnt_stat.f_mntfromspec, MNAMELEN);
2385 memcpy(&sbp->mount_info, &mp->mnt_stat.mount_info,
2386 sizeof(union mount_info));
2387 }
2388