1 /*-
2 * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause)
3 *
4 * Copyright (c) 2002, 2003 Networks Associates Technology, Inc.
5 * All rights reserved.
6 *
7 * This software was developed for the FreeBSD Project by Marshall
8 * Kirk McKusick and Network Associates Laboratories, the Security
9 * Research Division of Network Associates, Inc. under DARPA/SPAWAR
10 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
11 * research program
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 *
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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 * Copyright (c) 1982, 1986, 1989, 1993
35 * The Regents of the University of California. All rights reserved.
36 *
37 * Redistribution and use in source and binary forms, with or without
38 * modification, are permitted provided that the following conditions
39 * are met:
40 * 1. Redistributions of source code must retain the above copyright
41 * notice, this list of conditions and the following disclaimer.
42 * 2. Redistributions in binary form must reproduce the above copyright
43 * notice, this list of conditions and the following disclaimer in the
44 * documentation and/or other materials provided with the distribution.
45 * 3. Neither the name of the University nor the names of its contributors
46 * may be used to endorse or promote products derived from this software
47 * without specific prior written permission.
48 *
49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59 * SUCH DAMAGE.
60 *
61 * from: @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95
62 * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ...
63 * @(#)ffs_vnops.c 8.15 (Berkeley) 5/14/95
64 */
65
66 #include <sys/cdefs.h>
67 __FBSDID("$FreeBSD$");
68
69 #include "opt_directio.h"
70 #include "opt_ffs.h"
71 #include "opt_ufs.h"
72
73 #include <sys/param.h>
74 #include <sys/bio.h>
75 #include <sys/systm.h>
76 #include <sys/buf.h>
77 #include <sys/conf.h>
78 #include <sys/extattr.h>
79 #include <sys/kernel.h>
80 #include <sys/limits.h>
81 #include <sys/malloc.h>
82 #include <sys/mount.h>
83 #include <sys/priv.h>
84 #include <sys/rwlock.h>
85 #include <sys/stat.h>
86 #include <sys/sysctl.h>
87 #include <sys/vmmeter.h>
88 #include <sys/vnode.h>
89
90 #include <vm/vm.h>
91 #include <vm/vm_param.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_object.h>
94 #include <vm/vm_page.h>
95 #include <vm/vm_pager.h>
96 #include <vm/vnode_pager.h>
97
98 #include <ufs/ufs/extattr.h>
99 #include <ufs/ufs/quota.h>
100 #include <ufs/ufs/inode.h>
101 #include <ufs/ufs/ufs_extern.h>
102 #include <ufs/ufs/ufsmount.h>
103 #include <ufs/ufs/dir.h>
104 #ifdef UFS_DIRHASH
105 #include <ufs/ufs/dirhash.h>
106 #endif
107
108 #include <ufs/ffs/fs.h>
109 #include <ufs/ffs/ffs_extern.h>
110
111 #define ALIGNED_TO(ptr, s) \
112 (((uintptr_t)(ptr) & (_Alignof(s) - 1)) == 0)
113
114 #ifdef DIRECTIO
115 extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
116 #endif
117 static vop_fdatasync_t ffs_fdatasync;
118 static vop_fsync_t ffs_fsync;
119 static vop_getpages_t ffs_getpages;
120 static vop_getpages_async_t ffs_getpages_async;
121 static vop_lock1_t ffs_lock;
122 #ifdef INVARIANTS
123 static vop_unlock_t ffs_unlock_debug;
124 #endif
125 static vop_read_t ffs_read;
126 static vop_write_t ffs_write;
127 static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag);
128 static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag,
129 struct ucred *cred);
130 static vop_strategy_t ffsext_strategy;
131 static vop_closeextattr_t ffs_closeextattr;
132 static vop_deleteextattr_t ffs_deleteextattr;
133 static vop_getextattr_t ffs_getextattr;
134 static vop_listextattr_t ffs_listextattr;
135 static vop_openextattr_t ffs_openextattr;
136 static vop_setextattr_t ffs_setextattr;
137 static vop_vptofh_t ffs_vptofh;
138 static vop_vput_pair_t ffs_vput_pair;
139
140 /* Global vfs data structures for ufs. */
141 struct vop_vector ffs_vnodeops1 = {
142 .vop_default = &ufs_vnodeops,
143 .vop_fsync = ffs_fsync,
144 .vop_fdatasync = ffs_fdatasync,
145 .vop_getpages = ffs_getpages,
146 .vop_getpages_async = ffs_getpages_async,
147 .vop_lock1 = ffs_lock,
148 #ifdef INVARIANTS
149 .vop_unlock = ffs_unlock_debug,
150 #endif
151 .vop_read = ffs_read,
152 .vop_reallocblks = ffs_reallocblks,
153 .vop_write = ffs_write,
154 .vop_vptofh = ffs_vptofh,
155 .vop_vput_pair = ffs_vput_pair,
156 };
157 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops1);
158
159 struct vop_vector ffs_fifoops1 = {
160 .vop_default = &ufs_fifoops,
161 .vop_fsync = ffs_fsync,
162 .vop_fdatasync = ffs_fdatasync,
163 .vop_lock1 = ffs_lock,
164 #ifdef INVARIANTS
165 .vop_unlock = ffs_unlock_debug,
166 #endif
167 .vop_vptofh = ffs_vptofh,
168 };
169 VFS_VOP_VECTOR_REGISTER(ffs_fifoops1);
170
171 /* Global vfs data structures for ufs. */
172 struct vop_vector ffs_vnodeops2 = {
173 .vop_default = &ufs_vnodeops,
174 .vop_fsync = ffs_fsync,
175 .vop_fdatasync = ffs_fdatasync,
176 .vop_getpages = ffs_getpages,
177 .vop_getpages_async = ffs_getpages_async,
178 .vop_lock1 = ffs_lock,
179 #ifdef INVARIANTS
180 .vop_unlock = ffs_unlock_debug,
181 #endif
182 .vop_read = ffs_read,
183 .vop_reallocblks = ffs_reallocblks,
184 .vop_write = ffs_write,
185 .vop_closeextattr = ffs_closeextattr,
186 .vop_deleteextattr = ffs_deleteextattr,
187 .vop_getextattr = ffs_getextattr,
188 .vop_listextattr = ffs_listextattr,
189 .vop_openextattr = ffs_openextattr,
190 .vop_setextattr = ffs_setextattr,
191 .vop_vptofh = ffs_vptofh,
192 .vop_vput_pair = ffs_vput_pair,
193 };
194 VFS_VOP_VECTOR_REGISTER(ffs_vnodeops2);
195
196 struct vop_vector ffs_fifoops2 = {
197 .vop_default = &ufs_fifoops,
198 .vop_fsync = ffs_fsync,
199 .vop_fdatasync = ffs_fdatasync,
200 .vop_lock1 = ffs_lock,
201 #ifdef INVARIANTS
202 .vop_unlock = ffs_unlock_debug,
203 #endif
204 .vop_reallocblks = ffs_reallocblks,
205 .vop_strategy = ffsext_strategy,
206 .vop_closeextattr = ffs_closeextattr,
207 .vop_deleteextattr = ffs_deleteextattr,
208 .vop_getextattr = ffs_getextattr,
209 .vop_listextattr = ffs_listextattr,
210 .vop_openextattr = ffs_openextattr,
211 .vop_setextattr = ffs_setextattr,
212 .vop_vptofh = ffs_vptofh,
213 };
214 VFS_VOP_VECTOR_REGISTER(ffs_fifoops2);
215
216 /*
217 * Synch an open file.
218 */
219 /* ARGSUSED */
220 static int
221 ffs_fsync(struct vop_fsync_args *ap)
222 {
223 struct vnode *vp;
224 struct bufobj *bo;
225 int error;
226
227 vp = ap->a_vp;
228 bo = &vp->v_bufobj;
229 retry:
230 error = ffs_syncvnode(vp, ap->a_waitfor, 0);
231 if (error)
232 return (error);
233 if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) {
234 error = softdep_fsync(vp);
235 if (error)
236 return (error);
237
238 /*
239 * The softdep_fsync() function may drop vp lock,
240 * allowing for dirty buffers to reappear on the
241 * bo_dirty list. Recheck and resync as needed.
242 */
243 BO_LOCK(bo);
244 if ((vp->v_type == VREG || vp->v_type == VDIR) &&
245 (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
246 BO_UNLOCK(bo);
247 goto retry;
248 }
249 BO_UNLOCK(bo);
250 }
251 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), 0))
252 return (ENXIO);
253 return (0);
254 }
255
256 int
257 ffs_syncvnode(struct vnode *vp, int waitfor, int flags)
258 {
259 struct inode *ip;
260 struct bufobj *bo;
261 struct ufsmount *ump;
262 struct buf *bp, *nbp;
263 ufs_lbn_t lbn;
264 int error, passes, wflag;
265 bool still_dirty, unlocked, wait;
266
267 ip = VTOI(vp);
268 bo = &vp->v_bufobj;
269 ump = VFSTOUFS(vp->v_mount);
270 #ifdef WITNESS
271 wflag = IS_SNAPSHOT(ip) ? LK_NOWITNESS : 0;
272 #else
273 wflag = 0;
274 #endif
275
276 /*
277 * When doing MNT_WAIT we must first flush all dependencies
278 * on the inode.
279 */
280 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
281 (error = softdep_sync_metadata(vp)) != 0) {
282 if (ffs_fsfail_cleanup(ump, error))
283 error = 0;
284 return (error);
285 }
286
287 /*
288 * Flush all dirty buffers associated with a vnode.
289 */
290 error = 0;
291 passes = 0;
292 wait = false; /* Always do an async pass first. */
293 unlocked = false;
294 lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1));
295 BO_LOCK(bo);
296 loop:
297 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
298 bp->b_vflags &= ~BV_SCANNED;
299 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
300 /*
301 * Reasons to skip this buffer: it has already been considered
302 * on this pass, the buffer has dependencies that will cause
303 * it to be redirtied and it has not already been deferred,
304 * or it is already being written.
305 */
306 if ((bp->b_vflags & BV_SCANNED) != 0)
307 continue;
308 bp->b_vflags |= BV_SCANNED;
309 /*
310 * Flush indirects in order, if requested.
311 *
312 * Note that if only datasync is requested, we can
313 * skip indirect blocks when softupdates are not
314 * active. Otherwise we must flush them with data,
315 * since dependencies prevent data block writes.
316 */
317 if (waitfor == MNT_WAIT && bp->b_lblkno <= -UFS_NDADDR &&
318 (lbn_level(bp->b_lblkno) >= passes ||
319 ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp))))
320 continue;
321 if (bp->b_lblkno > lbn)
322 panic("ffs_syncvnode: syncing truncated data.");
323 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) {
324 BO_UNLOCK(bo);
325 } else if (wait) {
326 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
327 LK_INTERLOCK | wflag, BO_LOCKPTR(bo)) != 0) {
328 BO_LOCK(bo);
329 bp->b_vflags &= ~BV_SCANNED;
330 goto next_locked;
331 }
332 } else
333 continue;
334 if ((bp->b_flags & B_DELWRI) == 0)
335 panic("ffs_fsync: not dirty");
336 /*
337 * Check for dependencies and potentially complete them.
338 */
339 if (!LIST_EMPTY(&bp->b_dep) &&
340 (error = softdep_sync_buf(vp, bp,
341 wait ? MNT_WAIT : MNT_NOWAIT)) != 0) {
342 /*
343 * Lock order conflict, buffer was already unlocked,
344 * and vnode possibly unlocked.
345 */
346 if (error == ERELOOKUP) {
347 if (vp->v_data == NULL)
348 return (EBADF);
349 unlocked = true;
350 if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT &&
351 (error = softdep_sync_metadata(vp)) != 0) {
352 if (ffs_fsfail_cleanup(ump, error))
353 error = 0;
354 return (unlocked && error == 0 ?
355 ERELOOKUP : error);
356 }
357 /* Re-evaluate inode size */
358 lbn = lblkno(ITOFS(ip), (ip->i_size +
359 ITOFS(ip)->fs_bsize - 1));
360 goto next;
361 }
362 /* I/O error. */
363 if (error != EBUSY) {
364 BUF_UNLOCK(bp);
365 return (error);
366 }
367 /* If we deferred once, don't defer again. */
368 if ((bp->b_flags & B_DEFERRED) == 0) {
369 bp->b_flags |= B_DEFERRED;
370 BUF_UNLOCK(bp);
371 goto next;
372 }
373 }
374 if (wait) {
375 bremfree(bp);
376 error = bwrite(bp);
377 if (ffs_fsfail_cleanup(ump, error))
378 error = 0;
379 if (error != 0)
380 return (error);
381 } else if ((bp->b_flags & B_CLUSTEROK)) {
382 (void) vfs_bio_awrite(bp);
383 } else {
384 bremfree(bp);
385 (void) bawrite(bp);
386 }
387 next:
388 /*
389 * Since we may have slept during the I/O, we need
390 * to start from a known point.
391 */
392 BO_LOCK(bo);
393 next_locked:
394 nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd);
395 }
396 if (waitfor != MNT_WAIT) {
397 BO_UNLOCK(bo);
398 if ((flags & NO_INO_UPDT) != 0)
399 return (unlocked ? ERELOOKUP : 0);
400 error = ffs_update(vp, 0);
401 if (error == 0 && unlocked)
402 error = ERELOOKUP;
403 return (error);
404 }
405 /* Drain IO to see if we're done. */
406 bufobj_wwait(bo, 0, 0);
407 /*
408 * Block devices associated with filesystems may have new I/O
409 * requests posted for them even if the vnode is locked, so no
410 * amount of trying will get them clean. We make several passes
411 * as a best effort.
412 *
413 * Regular files may need multiple passes to flush all dependency
414 * work as it is possible that we must write once per indirect
415 * level, once for the leaf, and once for the inode and each of
416 * these will be done with one sync and one async pass.
417 */
418 if (bo->bo_dirty.bv_cnt > 0) {
419 if ((flags & DATA_ONLY) == 0) {
420 still_dirty = true;
421 } else {
422 /*
423 * For data-only sync, dirty indirect buffers
424 * are ignored.
425 */
426 still_dirty = false;
427 TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
428 if (bp->b_lblkno > -UFS_NDADDR) {
429 still_dirty = true;
430 break;
431 }
432 }
433 }
434
435 if (still_dirty) {
436 /* Write the inode after sync passes to flush deps. */
437 if (wait && DOINGSOFTDEP(vp) &&
438 (flags & NO_INO_UPDT) == 0) {
439 BO_UNLOCK(bo);
440 ffs_update(vp, 1);
441 BO_LOCK(bo);
442 }
443 /* switch between sync/async. */
444 wait = !wait;
445 if (wait || ++passes < UFS_NIADDR + 2)
446 goto loop;
447 }
448 }
449 BO_UNLOCK(bo);
450 error = 0;
451 if ((flags & DATA_ONLY) == 0) {
452 if ((flags & NO_INO_UPDT) == 0)
453 error = ffs_update(vp, 1);
454 if (DOINGSUJ(vp))
455 softdep_journal_fsync(VTOI(vp));
456 } else if ((ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)) != 0) {
457 error = ffs_update(vp, 1);
458 }
459 if (error == 0 && unlocked)
460 error = ERELOOKUP;
461 if (error == 0)
462 ip->i_flag &= ~IN_NEEDSYNC;
463 return (error);
464 }
465
466 static int
467 ffs_fdatasync(struct vop_fdatasync_args *ap)
468 {
469
470 return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY));
471 }
472
473 static int
474 ffs_lock(
475 struct vop_lock1_args /* {
476 struct vnode *a_vp;
477 int a_flags;
478 char *file;
479 int line;
480 } */ *ap)
481 {
482 #if !defined(NO_FFS_SNAPSHOT) || defined(DIAGNOSTIC)
483 struct vnode *vp = ap->a_vp;
484 #endif /* !NO_FFS_SNAPSHOT || DIAGNOSTIC */
485 #ifdef DIAGNOSTIC
486 struct inode *ip;
487 #endif /* DIAGNOSTIC */
488 int result;
489 #ifndef NO_FFS_SNAPSHOT
490 int flags;
491 struct lock *lkp;
492
493 /*
494 * Adaptive spinning mixed with SU leads to trouble. use a giant hammer
495 * and only use it when LK_NODDLKTREAT is set. Currently this means it
496 * is only used during path lookup.
497 */
498 if ((ap->a_flags & LK_NODDLKTREAT) != 0)
499 ap->a_flags |= LK_ADAPTIVE;
500 switch (ap->a_flags & LK_TYPE_MASK) {
501 case LK_SHARED:
502 case LK_UPGRADE:
503 case LK_EXCLUSIVE:
504 flags = ap->a_flags;
505 for (;;) {
506 #ifdef DEBUG_VFS_LOCKS
507 VNPASS(vp->v_holdcnt != 0, vp);
508 #endif /* DEBUG_VFS_LOCKS */
509 lkp = vp->v_vnlock;
510 result = lockmgr_lock_flags(lkp, flags,
511 &VI_MTX(vp)->lock_object, ap->a_file, ap->a_line);
512 if (lkp == vp->v_vnlock || result != 0)
513 break;
514 /*
515 * Apparent success, except that the vnode
516 * mutated between snapshot file vnode and
517 * regular file vnode while this process
518 * slept. The lock currently held is not the
519 * right lock. Release it, and try to get the
520 * new lock.
521 */
522 lockmgr_unlock(lkp);
523 if ((flags & (LK_INTERLOCK | LK_NOWAIT)) ==
524 (LK_INTERLOCK | LK_NOWAIT))
525 return (EBUSY);
526 if ((flags & LK_TYPE_MASK) == LK_UPGRADE)
527 flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE;
528 flags &= ~LK_INTERLOCK;
529 }
530 #ifdef DIAGNOSTIC
531 switch (ap->a_flags & LK_TYPE_MASK) {
532 case LK_UPGRADE:
533 case LK_EXCLUSIVE:
534 if (result == 0 && vp->v_vnlock->lk_recurse == 0) {
535 ip = VTOI(vp);
536 if (ip != NULL)
537 ip->i_lock_gen++;
538 }
539 }
540 #endif /* DIAGNOSTIC */
541 break;
542 default:
543 #ifdef DIAGNOSTIC
544 if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) {
545 ip = VTOI(vp);
546 if (ip != NULL)
547 ufs_unlock_tracker(ip);
548 }
549 #endif /* DIAGNOSTIC */
550 result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
551 break;
552 }
553 #else /* NO_FFS_SNAPSHOT */
554 /*
555 * See above for an explanation.
556 */
557 if ((ap->a_flags & LK_NODDLKTREAT) != 0)
558 ap->a_flags |= LK_ADAPTIVE;
559 #ifdef DIAGNOSTIC
560 if ((ap->a_flags & LK_TYPE_MASK) == LK_DOWNGRADE) {
561 ip = VTOI(vp);
562 if (ip != NULL)
563 ufs_unlock_tracker(ip);
564 }
565 #endif /* DIAGNOSTIC */
566 result = VOP_LOCK1_APV(&ufs_vnodeops, ap);
567 #endif /* NO_FFS_SNAPSHOT */
568 #ifdef DIAGNOSTIC
569 switch (ap->a_flags & LK_TYPE_MASK) {
570 case LK_UPGRADE:
571 case LK_EXCLUSIVE:
572 if (result == 0 && vp->v_vnlock->lk_recurse == 0) {
573 ip = VTOI(vp);
574 if (ip != NULL)
575 ip->i_lock_gen++;
576 }
577 }
578 #endif /* DIAGNOSTIC */
579 return (result);
580 }
581
582 #ifdef INVARIANTS
583 static int
584 ffs_unlock_debug(struct vop_unlock_args *ap)
585 {
586 struct vnode *vp;
587 struct inode *ip;
588
589 vp = ap->a_vp;
590 ip = VTOI(vp);
591 if (ip->i_flag & UFS_INODE_FLAG_LAZY_MASK_ASSERTABLE) {
592 if ((vp->v_mflag & VMP_LAZYLIST) == 0) {
593 VI_LOCK(vp);
594 VNASSERT((vp->v_mflag & VMP_LAZYLIST), vp,
595 ("%s: modified vnode (%x) not on lazy list",
596 __func__, ip->i_flag));
597 VI_UNLOCK(vp);
598 }
599 }
600 KASSERT(vp->v_type != VDIR || vp->v_vnlock->lk_recurse != 0 ||
601 (ip->i_flag & IN_ENDOFF) == 0,
602 ("ufs dir vp %p ip %p flags %#x", vp, ip, ip->i_flag));
603 #ifdef DIAGNOSTIC
604 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && ip != NULL &&
605 vp->v_vnlock->lk_recurse == 0)
606 ufs_unlock_tracker(ip);
607 #endif
608 return (VOP_UNLOCK_APV(&ufs_vnodeops, ap));
609 }
610 #endif
611
612 static int
613 ffs_read_hole(struct uio *uio, long xfersize, long *size)
614 {
615 ssize_t saved_resid, tlen;
616 int error;
617
618 while (xfersize > 0) {
619 tlen = min(xfersize, ZERO_REGION_SIZE);
620 saved_resid = uio->uio_resid;
621 error = vn_io_fault_uiomove(__DECONST(void *, zero_region),
622 tlen, uio);
623 if (error != 0)
624 return (error);
625 tlen = saved_resid - uio->uio_resid;
626 xfersize -= tlen;
627 *size -= tlen;
628 }
629 return (0);
630 }
631
632 /*
633 * Vnode op for reading.
634 */
635 static int
636 ffs_read(
637 struct vop_read_args /* {
638 struct vnode *a_vp;
639 struct uio *a_uio;
640 int a_ioflag;
641 struct ucred *a_cred;
642 } */ *ap)
643 {
644 struct vnode *vp;
645 struct inode *ip;
646 struct uio *uio;
647 struct fs *fs;
648 struct buf *bp;
649 ufs_lbn_t lbn, nextlbn;
650 off_t bytesinfile;
651 long size, xfersize, blkoffset;
652 ssize_t orig_resid;
653 int bflag, error, ioflag, seqcount;
654
655 vp = ap->a_vp;
656 uio = ap->a_uio;
657 ioflag = ap->a_ioflag;
658 if (ap->a_ioflag & IO_EXT)
659 #ifdef notyet
660 return (ffs_extread(vp, uio, ioflag));
661 #else
662 panic("ffs_read+IO_EXT");
663 #endif
664 #ifdef DIRECTIO
665 if ((ioflag & IO_DIRECT) != 0) {
666 int workdone;
667
668 error = ffs_rawread(vp, uio, &workdone);
669 if (error != 0 || workdone != 0)
670 return error;
671 }
672 #endif
673
674 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
675 ip = VTOI(vp);
676
677 #ifdef INVARIANTS
678 if (uio->uio_rw != UIO_READ)
679 panic("ffs_read: mode");
680
681 if (vp->v_type == VLNK) {
682 if ((int)ip->i_size < VFSTOUFS(vp->v_mount)->um_maxsymlinklen)
683 panic("ffs_read: short symlink");
684 } else if (vp->v_type != VREG && vp->v_type != VDIR)
685 panic("ffs_read: type %d", vp->v_type);
686 #endif
687 orig_resid = uio->uio_resid;
688 KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0"));
689 if (orig_resid == 0)
690 return (0);
691 KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0"));
692 fs = ITOFS(ip);
693 if (uio->uio_offset < ip->i_size &&
694 uio->uio_offset >= fs->fs_maxfilesize)
695 return (EOVERFLOW);
696
697 bflag = GB_UNMAPPED | (uio->uio_segflg == UIO_NOCOPY ? 0 : GB_NOSPARSE);
698 #ifdef WITNESS
699 bflag |= IS_SNAPSHOT(ip) ? GB_NOWITNESS : 0;
700 #endif
701 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
702 if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0)
703 break;
704 lbn = lblkno(fs, uio->uio_offset);
705 nextlbn = lbn + 1;
706
707 /*
708 * size of buffer. The buffer representing the
709 * end of the file is rounded up to the size of
710 * the block type ( fragment or full block,
711 * depending ).
712 */
713 size = blksize(fs, ip, lbn);
714 blkoffset = blkoff(fs, uio->uio_offset);
715
716 /*
717 * The amount we want to transfer in this iteration is
718 * one FS block less the amount of the data before
719 * our startpoint (duh!)
720 */
721 xfersize = fs->fs_bsize - blkoffset;
722
723 /*
724 * But if we actually want less than the block,
725 * or the file doesn't have a whole block more of data,
726 * then use the lesser number.
727 */
728 if (uio->uio_resid < xfersize)
729 xfersize = uio->uio_resid;
730 if (bytesinfile < xfersize)
731 xfersize = bytesinfile;
732
733 if (lblktosize(fs, nextlbn) >= ip->i_size) {
734 /*
735 * Don't do readahead if this is the end of the file.
736 */
737 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
738 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
739 /*
740 * Otherwise if we are allowed to cluster,
741 * grab as much as we can.
742 *
743 * XXX This may not be a win if we are not
744 * doing sequential access.
745 */
746 error = cluster_read(vp, ip->i_size, lbn,
747 size, NOCRED, blkoffset + uio->uio_resid,
748 seqcount, bflag, &bp);
749 } else if (seqcount > 1) {
750 /*
751 * If we are NOT allowed to cluster, then
752 * if we appear to be acting sequentially,
753 * fire off a request for a readahead
754 * as well as a read. Note that the 4th and 5th
755 * arguments point to arrays of the size specified in
756 * the 6th argument.
757 */
758 u_int nextsize = blksize(fs, ip, nextlbn);
759 error = breadn_flags(vp, lbn, lbn, size, &nextlbn,
760 &nextsize, 1, NOCRED, bflag, NULL, &bp);
761 } else {
762 /*
763 * Failing all of the above, just read what the
764 * user asked for. Interestingly, the same as
765 * the first option above.
766 */
767 error = bread_gb(vp, lbn, size, NOCRED, bflag, &bp);
768 }
769 if (error == EJUSTRETURN) {
770 error = ffs_read_hole(uio, xfersize, &size);
771 if (error == 0)
772 continue;
773 }
774 if (error != 0) {
775 brelse(bp);
776 bp = NULL;
777 break;
778 }
779
780 /*
781 * We should only get non-zero b_resid when an I/O error
782 * has occurred, which should cause us to break above.
783 * However, if the short read did not cause an error,
784 * then we want to ensure that we do not uiomove bad
785 * or uninitialized data.
786 */
787 size -= bp->b_resid;
788 if (size < xfersize) {
789 if (size == 0)
790 break;
791 xfersize = size;
792 }
793
794 if (buf_mapped(bp)) {
795 error = vn_io_fault_uiomove((char *)bp->b_data +
796 blkoffset, (int)xfersize, uio);
797 } else {
798 error = vn_io_fault_pgmove(bp->b_pages,
799 blkoffset + (bp->b_offset & PAGE_MASK),
800 (int)xfersize, uio);
801 }
802 if (error)
803 break;
804
805 vfs_bio_brelse(bp, ioflag);
806 }
807
808 /*
809 * This can only happen in the case of an error
810 * because the loop above resets bp to NULL on each iteration
811 * and on normal completion has not set a new value into it.
812 * so it must have come from a 'break' statement
813 */
814 if (bp != NULL)
815 vfs_bio_brelse(bp, ioflag);
816
817 if ((error == 0 || uio->uio_resid != orig_resid) &&
818 (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
819 UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS);
820 return (error);
821 }
822
823 /*
824 * Vnode op for writing.
825 */
826 static int
827 ffs_write(
828 struct vop_write_args /* {
829 struct vnode *a_vp;
830 struct uio *a_uio;
831 int a_ioflag;
832 struct ucred *a_cred;
833 } */ *ap)
834 {
835 struct vnode *vp;
836 struct uio *uio;
837 struct inode *ip;
838 struct fs *fs;
839 struct buf *bp;
840 ufs_lbn_t lbn;
841 off_t osize;
842 ssize_t resid, r;
843 int seqcount;
844 int blkoffset, error, flags, ioflag, size, xfersize;
845
846 vp = ap->a_vp;
847 if (DOINGSUJ(vp))
848 softdep_prealloc(vp, MNT_WAIT);
849 if (vp->v_data == NULL)
850 return (EBADF);
851
852 uio = ap->a_uio;
853 ioflag = ap->a_ioflag;
854 if (ap->a_ioflag & IO_EXT)
855 #ifdef notyet
856 return (ffs_extwrite(vp, uio, ioflag, ap->a_cred));
857 #else
858 panic("ffs_write+IO_EXT");
859 #endif
860
861 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
862 ip = VTOI(vp);
863
864 #ifdef INVARIANTS
865 if (uio->uio_rw != UIO_WRITE)
866 panic("ffs_write: mode");
867 #endif
868
869 switch (vp->v_type) {
870 case VREG:
871 if (ioflag & IO_APPEND)
872 uio->uio_offset = ip->i_size;
873 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
874 return (EPERM);
875 /* FALLTHROUGH */
876 case VLNK:
877 break;
878 case VDIR:
879 panic("ffs_write: dir write");
880 break;
881 default:
882 panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type,
883 (int)uio->uio_offset,
884 (int)uio->uio_resid
885 );
886 }
887
888 KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0"));
889 KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0"));
890 fs = ITOFS(ip);
891
892 /*
893 * Maybe this should be above the vnode op call, but so long as
894 * file servers have no limits, I don't think it matters.
895 */
896 error = vn_rlimit_fsizex(vp, uio, fs->fs_maxfilesize, &r,
897 uio->uio_td);
898 if (error != 0) {
899 vn_rlimit_fsizex_res(uio, r);
900 return (error);
901 }
902
903 resid = uio->uio_resid;
904 osize = ip->i_size;
905 if (seqcount > BA_SEQMAX)
906 flags = BA_SEQMAX << BA_SEQSHIFT;
907 else
908 flags = seqcount << BA_SEQSHIFT;
909 if (ioflag & IO_SYNC)
910 flags |= IO_SYNC;
911 flags |= BA_UNMAPPED;
912
913 for (error = 0; uio->uio_resid > 0;) {
914 lbn = lblkno(fs, uio->uio_offset);
915 blkoffset = blkoff(fs, uio->uio_offset);
916 xfersize = fs->fs_bsize - blkoffset;
917 if (uio->uio_resid < xfersize)
918 xfersize = uio->uio_resid;
919 if (uio->uio_offset + xfersize > ip->i_size)
920 vnode_pager_setsize(vp, uio->uio_offset + xfersize);
921
922 /*
923 * We must perform a read-before-write if the transfer size
924 * does not cover the entire buffer.
925 */
926 if (fs->fs_bsize > xfersize)
927 flags |= BA_CLRBUF;
928 else
929 flags &= ~BA_CLRBUF;
930 /* XXX is uio->uio_offset the right thing here? */
931 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
932 ap->a_cred, flags, &bp);
933 if (error != 0) {
934 vnode_pager_setsize(vp, ip->i_size);
935 break;
936 }
937 if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
938 bp->b_flags |= B_NOCACHE;
939
940 if (uio->uio_offset + xfersize > ip->i_size) {
941 ip->i_size = uio->uio_offset + xfersize;
942 DIP_SET(ip, i_size, ip->i_size);
943 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
944 }
945
946 size = blksize(fs, ip, lbn) - bp->b_resid;
947 if (size < xfersize)
948 xfersize = size;
949
950 if (buf_mapped(bp)) {
951 error = vn_io_fault_uiomove((char *)bp->b_data +
952 blkoffset, (int)xfersize, uio);
953 } else {
954 error = vn_io_fault_pgmove(bp->b_pages,
955 blkoffset + (bp->b_offset & PAGE_MASK),
956 (int)xfersize, uio);
957 }
958 /*
959 * If the buffer is not already filled and we encounter an
960 * error while trying to fill it, we have to clear out any
961 * garbage data from the pages instantiated for the buffer.
962 * If we do not, a failed uiomove() during a write can leave
963 * the prior contents of the pages exposed to a userland mmap.
964 *
965 * Note that we need only clear buffers with a transfer size
966 * equal to the block size because buffers with a shorter
967 * transfer size were cleared above by the call to UFS_BALLOC()
968 * with the BA_CLRBUF flag set.
969 *
970 * If the source region for uiomove identically mmaps the
971 * buffer, uiomove() performed the NOP copy, and the buffer
972 * content remains valid because the page fault handler
973 * validated the pages.
974 */
975 if (error != 0 && (bp->b_flags & B_CACHE) == 0 &&
976 fs->fs_bsize == xfersize)
977 vfs_bio_clrbuf(bp);
978
979 vfs_bio_set_flags(bp, ioflag);
980
981 /*
982 * If IO_SYNC each buffer is written synchronously. Otherwise
983 * if we have a severe page deficiency write the buffer
984 * asynchronously. Otherwise try to cluster, and if that
985 * doesn't do it then either do an async write (if O_DIRECT),
986 * or a delayed write (if not).
987 */
988 if (ioflag & IO_SYNC) {
989 (void)bwrite(bp);
990 } else if (vm_page_count_severe() ||
991 buf_dirty_count_severe() ||
992 (ioflag & IO_ASYNC)) {
993 bp->b_flags |= B_CLUSTEROK;
994 bawrite(bp);
995 } else if (xfersize + blkoffset == fs->fs_bsize) {
996 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
997 bp->b_flags |= B_CLUSTEROK;
998 cluster_write(vp, &ip->i_clusterw, bp,
999 ip->i_size, seqcount, GB_UNMAPPED);
1000 } else {
1001 bawrite(bp);
1002 }
1003 } else if (ioflag & IO_DIRECT) {
1004 bp->b_flags |= B_CLUSTEROK;
1005 bawrite(bp);
1006 } else {
1007 bp->b_flags |= B_CLUSTEROK;
1008 bdwrite(bp);
1009 }
1010 if (error || xfersize == 0)
1011 break;
1012 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE);
1013 }
1014 /*
1015 * If we successfully wrote any data, and we are not the superuser
1016 * we clear the setuid and setgid bits as a precaution against
1017 * tampering.
1018 */
1019 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
1020 ap->a_cred) {
1021 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID)) {
1022 vn_seqc_write_begin(vp);
1023 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1024 DIP_SET(ip, i_mode, ip->i_mode);
1025 vn_seqc_write_end(vp);
1026 }
1027 }
1028 if (error) {
1029 if (ioflag & IO_UNIT) {
1030 (void)ffs_truncate(vp, osize,
1031 IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred);
1032 uio->uio_offset -= resid - uio->uio_resid;
1033 uio->uio_resid = resid;
1034 }
1035 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
1036 if (!(ioflag & IO_DATASYNC) ||
1037 (ip->i_flags & (IN_SIZEMOD | IN_IBLKDATA)))
1038 error = ffs_update(vp, 1);
1039 if (ffs_fsfail_cleanup(VFSTOUFS(vp->v_mount), error))
1040 error = ENXIO;
1041 }
1042 vn_rlimit_fsizex_res(uio, r);
1043 return (error);
1044 }
1045
1046 /*
1047 * Extended attribute area reading.
1048 */
1049 static int
1050 ffs_extread(struct vnode *vp, struct uio *uio, int ioflag)
1051 {
1052 struct inode *ip;
1053 struct ufs2_dinode *dp;
1054 struct fs *fs;
1055 struct buf *bp;
1056 ufs_lbn_t lbn, nextlbn;
1057 off_t bytesinfile;
1058 long size, xfersize, blkoffset;
1059 ssize_t orig_resid;
1060 int error;
1061
1062 ip = VTOI(vp);
1063 fs = ITOFS(ip);
1064 dp = ip->i_din2;
1065
1066 #ifdef INVARIANTS
1067 if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC)
1068 panic("ffs_extread: mode");
1069
1070 #endif
1071 orig_resid = uio->uio_resid;
1072 KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0"));
1073 if (orig_resid == 0)
1074 return (0);
1075 KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0"));
1076
1077 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1078 if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0)
1079 break;
1080 lbn = lblkno(fs, uio->uio_offset);
1081 nextlbn = lbn + 1;
1082
1083 /*
1084 * size of buffer. The buffer representing the
1085 * end of the file is rounded up to the size of
1086 * the block type ( fragment or full block,
1087 * depending ).
1088 */
1089 size = sblksize(fs, dp->di_extsize, lbn);
1090 blkoffset = blkoff(fs, uio->uio_offset);
1091
1092 /*
1093 * The amount we want to transfer in this iteration is
1094 * one FS block less the amount of the data before
1095 * our startpoint (duh!)
1096 */
1097 xfersize = fs->fs_bsize - blkoffset;
1098
1099 /*
1100 * But if we actually want less than the block,
1101 * or the file doesn't have a whole block more of data,
1102 * then use the lesser number.
1103 */
1104 if (uio->uio_resid < xfersize)
1105 xfersize = uio->uio_resid;
1106 if (bytesinfile < xfersize)
1107 xfersize = bytesinfile;
1108
1109 if (lblktosize(fs, nextlbn) >= dp->di_extsize) {
1110 /*
1111 * Don't do readahead if this is the end of the info.
1112 */
1113 error = bread(vp, -1 - lbn, size, NOCRED, &bp);
1114 } else {
1115 /*
1116 * If we have a second block, then
1117 * fire off a request for a readahead
1118 * as well as a read. Note that the 4th and 5th
1119 * arguments point to arrays of the size specified in
1120 * the 6th argument.
1121 */
1122 u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn);
1123
1124 nextlbn = -1 - nextlbn;
1125 error = breadn(vp, -1 - lbn,
1126 size, &nextlbn, &nextsize, 1, NOCRED, &bp);
1127 }
1128 if (error) {
1129 brelse(bp);
1130 bp = NULL;
1131 break;
1132 }
1133
1134 /*
1135 * We should only get non-zero b_resid when an I/O error
1136 * has occurred, which should cause us to break above.
1137 * However, if the short read did not cause an error,
1138 * then we want to ensure that we do not uiomove bad
1139 * or uninitialized data.
1140 */
1141 size -= bp->b_resid;
1142 if (size < xfersize) {
1143 if (size == 0)
1144 break;
1145 xfersize = size;
1146 }
1147
1148 error = uiomove((char *)bp->b_data + blkoffset,
1149 (int)xfersize, uio);
1150 if (error)
1151 break;
1152 vfs_bio_brelse(bp, ioflag);
1153 }
1154
1155 /*
1156 * This can only happen in the case of an error
1157 * because the loop above resets bp to NULL on each iteration
1158 * and on normal completion has not set a new value into it.
1159 * so it must have come from a 'break' statement
1160 */
1161 if (bp != NULL)
1162 vfs_bio_brelse(bp, ioflag);
1163 return (error);
1164 }
1165
1166 /*
1167 * Extended attribute area writing.
1168 */
1169 static int
1170 ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred)
1171 {
1172 struct inode *ip;
1173 struct ufs2_dinode *dp;
1174 struct fs *fs;
1175 struct buf *bp;
1176 ufs_lbn_t lbn;
1177 off_t osize;
1178 ssize_t resid;
1179 int blkoffset, error, flags, size, xfersize;
1180
1181 ip = VTOI(vp);
1182 fs = ITOFS(ip);
1183 dp = ip->i_din2;
1184
1185 #ifdef INVARIANTS
1186 if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC)
1187 panic("ffs_extwrite: mode");
1188 #endif
1189
1190 if (ioflag & IO_APPEND)
1191 uio->uio_offset = dp->di_extsize;
1192 KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0"));
1193 KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0"));
1194 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1195 UFS_NXADDR * fs->fs_bsize)
1196 return (EFBIG);
1197
1198 resid = uio->uio_resid;
1199 osize = dp->di_extsize;
1200 flags = IO_EXT;
1201 if (ioflag & IO_SYNC)
1202 flags |= IO_SYNC;
1203
1204 for (error = 0; uio->uio_resid > 0;) {
1205 lbn = lblkno(fs, uio->uio_offset);
1206 blkoffset = blkoff(fs, uio->uio_offset);
1207 xfersize = fs->fs_bsize - blkoffset;
1208 if (uio->uio_resid < xfersize)
1209 xfersize = uio->uio_resid;
1210
1211 /*
1212 * We must perform a read-before-write if the transfer size
1213 * does not cover the entire buffer.
1214 */
1215 if (fs->fs_bsize > xfersize)
1216 flags |= BA_CLRBUF;
1217 else
1218 flags &= ~BA_CLRBUF;
1219 error = UFS_BALLOC(vp, uio->uio_offset, xfersize,
1220 ucred, flags, &bp);
1221 if (error != 0)
1222 break;
1223 /*
1224 * If the buffer is not valid we have to clear out any
1225 * garbage data from the pages instantiated for the buffer.
1226 * If we do not, a failed uiomove() during a write can leave
1227 * the prior contents of the pages exposed to a userland
1228 * mmap(). XXX deal with uiomove() errors a better way.
1229 */
1230 if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize)
1231 vfs_bio_clrbuf(bp);
1232
1233 if (uio->uio_offset + xfersize > dp->di_extsize) {
1234 dp->di_extsize = uio->uio_offset + xfersize;
1235 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE);
1236 }
1237
1238 size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid;
1239 if (size < xfersize)
1240 xfersize = size;
1241
1242 error =
1243 uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
1244
1245 vfs_bio_set_flags(bp, ioflag);
1246
1247 /*
1248 * If IO_SYNC each buffer is written synchronously. Otherwise
1249 * if we have a severe page deficiency write the buffer
1250 * asynchronously. Otherwise try to cluster, and if that
1251 * doesn't do it then either do an async write (if O_DIRECT),
1252 * or a delayed write (if not).
1253 */
1254 if (ioflag & IO_SYNC) {
1255 (void)bwrite(bp);
1256 } else if (vm_page_count_severe() ||
1257 buf_dirty_count_severe() ||
1258 xfersize + blkoffset == fs->fs_bsize ||
1259 (ioflag & (IO_ASYNC | IO_DIRECT)))
1260 bawrite(bp);
1261 else
1262 bdwrite(bp);
1263 if (error || xfersize == 0)
1264 break;
1265 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1266 }
1267 /*
1268 * If we successfully wrote any data, and we are not the superuser
1269 * we clear the setuid and setgid bits as a precaution against
1270 * tampering.
1271 */
1272 if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) {
1273 if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID)) {
1274 vn_seqc_write_begin(vp);
1275 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1276 dp->di_mode = ip->i_mode;
1277 vn_seqc_write_end(vp);
1278 }
1279 }
1280 if (error) {
1281 if (ioflag & IO_UNIT) {
1282 (void)ffs_truncate(vp, osize,
1283 IO_EXT | (ioflag&IO_SYNC), ucred);
1284 uio->uio_offset -= resid - uio->uio_resid;
1285 uio->uio_resid = resid;
1286 }
1287 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC))
1288 error = ffs_update(vp, 1);
1289 return (error);
1290 }
1291
1292 /*
1293 * Vnode operating to retrieve a named extended attribute.
1294 *
1295 * Locate a particular EA (nspace:name) in the area (ptr:length), and return
1296 * the length of the EA, and possibly the pointer to the entry and to the data.
1297 */
1298 static int
1299 ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name,
1300 struct extattr **eapp, u_char **eac)
1301 {
1302 struct extattr *eap, *eaend;
1303 size_t nlen;
1304
1305 nlen = strlen(name);
1306 KASSERT(ALIGNED_TO(ptr, struct extattr), ("unaligned"));
1307 eap = (struct extattr *)ptr;
1308 eaend = (struct extattr *)(ptr + length);
1309 for (; eap < eaend; eap = EXTATTR_NEXT(eap)) {
1310 KASSERT(EXTATTR_NEXT(eap) <= eaend,
1311 ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend));
1312 if (eap->ea_namespace != nspace || eap->ea_namelength != nlen
1313 || memcmp(eap->ea_name, name, nlen) != 0)
1314 continue;
1315 if (eapp != NULL)
1316 *eapp = eap;
1317 if (eac != NULL)
1318 *eac = EXTATTR_CONTENT(eap);
1319 return (EXTATTR_CONTENT_SIZE(eap));
1320 }
1321 return (-1);
1322 }
1323
1324 static int
1325 ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td)
1326 {
1327 const struct extattr *eap, *eaend, *eapnext;
1328 struct inode *ip;
1329 struct ufs2_dinode *dp;
1330 struct fs *fs;
1331 struct uio luio;
1332 struct iovec liovec;
1333 u_int easize;
1334 int error;
1335 u_char *eae;
1336
1337 ip = VTOI(vp);
1338 fs = ITOFS(ip);
1339 dp = ip->i_din2;
1340 easize = dp->di_extsize;
1341 if ((uoff_t)easize > UFS_NXADDR * fs->fs_bsize)
1342 return (EFBIG);
1343
1344 eae = malloc(easize, M_TEMP, M_WAITOK);
1345
1346 liovec.iov_base = eae;
1347 liovec.iov_len = easize;
1348 luio.uio_iov = &liovec;
1349 luio.uio_iovcnt = 1;
1350 luio.uio_offset = 0;
1351 luio.uio_resid = easize;
1352 luio.uio_segflg = UIO_SYSSPACE;
1353 luio.uio_rw = UIO_READ;
1354 luio.uio_td = td;
1355
1356 error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC);
1357 if (error) {
1358 free(eae, M_TEMP);
1359 return (error);
1360 }
1361 /* Validate disk xattrfile contents. */
1362 for (eap = (void *)eae, eaend = (void *)(eae + easize); eap < eaend;
1363 eap = eapnext) {
1364 /* Detect zeroed out tail */
1365 if (eap->ea_length < sizeof(*eap) || eap->ea_length == 0) {
1366 easize = (const u_char *)eap - eae;
1367 break;
1368 }
1369
1370 eapnext = EXTATTR_NEXT(eap);
1371 /* Bogusly long entry. */
1372 if (eapnext > eaend) {
1373 free(eae, M_TEMP);
1374 return (EINTEGRITY);
1375 }
1376 }
1377 ip->i_ea_len = easize;
1378 *p = eae;
1379 return (0);
1380 }
1381
1382 static void
1383 ffs_lock_ea(struct vnode *vp)
1384 {
1385 struct inode *ip;
1386
1387 ip = VTOI(vp);
1388 VI_LOCK(vp);
1389 while (ip->i_flag & IN_EA_LOCKED) {
1390 UFS_INODE_SET_FLAG(ip, IN_EA_LOCKWAIT);
1391 msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea",
1392 0);
1393 }
1394 UFS_INODE_SET_FLAG(ip, IN_EA_LOCKED);
1395 VI_UNLOCK(vp);
1396 }
1397
1398 static void
1399 ffs_unlock_ea(struct vnode *vp)
1400 {
1401 struct inode *ip;
1402
1403 ip = VTOI(vp);
1404 VI_LOCK(vp);
1405 if (ip->i_flag & IN_EA_LOCKWAIT)
1406 wakeup(&ip->i_ea_refs);
1407 ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT);
1408 VI_UNLOCK(vp);
1409 }
1410
1411 static int
1412 ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td)
1413 {
1414 struct inode *ip;
1415 int error;
1416
1417 ip = VTOI(vp);
1418
1419 ffs_lock_ea(vp);
1420 if (ip->i_ea_area != NULL) {
1421 ip->i_ea_refs++;
1422 ffs_unlock_ea(vp);
1423 return (0);
1424 }
1425 error = ffs_rdextattr(&ip->i_ea_area, vp, td);
1426 if (error) {
1427 ffs_unlock_ea(vp);
1428 return (error);
1429 }
1430 ip->i_ea_error = 0;
1431 ip->i_ea_refs++;
1432 ffs_unlock_ea(vp);
1433 return (0);
1434 }
1435
1436 /*
1437 * Vnode extattr transaction commit/abort
1438 */
1439 static int
1440 ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td)
1441 {
1442 struct inode *ip;
1443 struct uio luio;
1444 struct iovec *liovec;
1445 struct ufs2_dinode *dp;
1446 size_t ea_len, tlen;
1447 int error, i, lcnt;
1448 bool truncate;
1449
1450 ip = VTOI(vp);
1451
1452 ffs_lock_ea(vp);
1453 if (ip->i_ea_area == NULL) {
1454 ffs_unlock_ea(vp);
1455 return (EINVAL);
1456 }
1457 dp = ip->i_din2;
1458 error = ip->i_ea_error;
1459 truncate = false;
1460 if (commit && error == 0) {
1461 ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit");
1462 if (cred == NOCRED)
1463 cred = vp->v_mount->mnt_cred;
1464
1465 ea_len = MAX(ip->i_ea_len, dp->di_extsize);
1466 for (lcnt = 1, tlen = ea_len - ip->i_ea_len; tlen > 0;) {
1467 tlen -= MIN(ZERO_REGION_SIZE, tlen);
1468 lcnt++;
1469 }
1470
1471 liovec = __builtin_alloca(lcnt * sizeof(struct iovec));
1472 luio.uio_iovcnt = lcnt;
1473
1474 liovec[0].iov_base = ip->i_ea_area;
1475 liovec[0].iov_len = ip->i_ea_len;
1476 for (i = 1, tlen = ea_len - ip->i_ea_len; i < lcnt; i++) {
1477 liovec[i].iov_base = __DECONST(void *, zero_region);
1478 liovec[i].iov_len = MIN(ZERO_REGION_SIZE, tlen);
1479 tlen -= liovec[i].iov_len;
1480 }
1481 MPASS(tlen == 0);
1482
1483 luio.uio_iov = liovec;
1484 luio.uio_offset = 0;
1485 luio.uio_resid = ea_len;
1486 luio.uio_segflg = UIO_SYSSPACE;
1487 luio.uio_rw = UIO_WRITE;
1488 luio.uio_td = td;
1489 error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred);
1490 if (error == 0 && ip->i_ea_len == 0)
1491 truncate = true;
1492 }
1493 if (--ip->i_ea_refs == 0) {
1494 free(ip->i_ea_area, M_TEMP);
1495 ip->i_ea_area = NULL;
1496 ip->i_ea_len = 0;
1497 ip->i_ea_error = 0;
1498 }
1499 ffs_unlock_ea(vp);
1500
1501 if (truncate)
1502 ffs_truncate(vp, 0, IO_EXT, cred);
1503 return (error);
1504 }
1505
1506 /*
1507 * Vnode extattr strategy routine for fifos.
1508 *
1509 * We need to check for a read or write of the external attributes.
1510 * Otherwise we just fall through and do the usual thing.
1511 */
1512 static int
1513 ffsext_strategy(
1514 struct vop_strategy_args /* {
1515 struct vnodeop_desc *a_desc;
1516 struct vnode *a_vp;
1517 struct buf *a_bp;
1518 } */ *ap)
1519 {
1520 struct vnode *vp;
1521 daddr_t lbn;
1522
1523 vp = ap->a_vp;
1524 lbn = ap->a_bp->b_lblkno;
1525 if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -UFS_NXADDR)
1526 return (VOP_STRATEGY_APV(&ufs_vnodeops, ap));
1527 if (vp->v_type == VFIFO)
1528 return (VOP_STRATEGY_APV(&ufs_fifoops, ap));
1529 panic("spec nodes went here");
1530 }
1531
1532 /*
1533 * Vnode extattr transaction commit/abort
1534 */
1535 static int
1536 ffs_openextattr(
1537 struct vop_openextattr_args /* {
1538 struct vnodeop_desc *a_desc;
1539 struct vnode *a_vp;
1540 IN struct ucred *a_cred;
1541 IN struct thread *a_td;
1542 } */ *ap)
1543 {
1544
1545 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1546 return (EOPNOTSUPP);
1547
1548 return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td));
1549 }
1550
1551 /*
1552 * Vnode extattr transaction commit/abort
1553 */
1554 static int
1555 ffs_closeextattr(
1556 struct vop_closeextattr_args /* {
1557 struct vnodeop_desc *a_desc;
1558 struct vnode *a_vp;
1559 int a_commit;
1560 IN struct ucred *a_cred;
1561 IN struct thread *a_td;
1562 } */ *ap)
1563 {
1564 struct vnode *vp;
1565
1566 vp = ap->a_vp;
1567 if (vp->v_type == VCHR || vp->v_type == VBLK)
1568 return (EOPNOTSUPP);
1569 if (ap->a_commit && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0)
1570 return (EROFS);
1571
1572 if (ap->a_commit && DOINGSUJ(vp)) {
1573 ASSERT_VOP_ELOCKED(vp, "ffs_closeextattr commit");
1574 softdep_prealloc(vp, MNT_WAIT);
1575 if (vp->v_data == NULL)
1576 return (EBADF);
1577 }
1578 return (ffs_close_ea(vp, ap->a_commit, ap->a_cred, ap->a_td));
1579 }
1580
1581 /*
1582 * Vnode operation to remove a named attribute.
1583 */
1584 static int
1585 ffs_deleteextattr(
1586 struct vop_deleteextattr_args /* {
1587 IN struct vnode *a_vp;
1588 IN int a_attrnamespace;
1589 IN const char *a_name;
1590 IN struct ucred *a_cred;
1591 IN struct thread *a_td;
1592 } */ *ap)
1593 {
1594 struct vnode *vp;
1595 struct inode *ip;
1596 struct extattr *eap;
1597 uint32_t ul;
1598 int olen, error, i, easize;
1599 u_char *eae;
1600 void *tmp;
1601
1602 vp = ap->a_vp;
1603 ip = VTOI(vp);
1604
1605 if (vp->v_type == VCHR || vp->v_type == VBLK)
1606 return (EOPNOTSUPP);
1607 if (strlen(ap->a_name) == 0)
1608 return (EINVAL);
1609 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1610 return (EROFS);
1611
1612 error = extattr_check_cred(vp, ap->a_attrnamespace,
1613 ap->a_cred, ap->a_td, VWRITE);
1614 if (error) {
1615 /*
1616 * ffs_lock_ea is not needed there, because the vnode
1617 * must be exclusively locked.
1618 */
1619 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1620 ip->i_ea_error = error;
1621 return (error);
1622 }
1623
1624 if (DOINGSUJ(vp)) {
1625 ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr");
1626 softdep_prealloc(vp, MNT_WAIT);
1627 if (vp->v_data == NULL)
1628 return (EBADF);
1629 }
1630
1631 error = ffs_open_ea(vp, ap->a_cred, ap->a_td);
1632 if (error)
1633 return (error);
1634
1635 /* CEM: delete could be done in-place instead */
1636 eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK);
1637 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1638 easize = ip->i_ea_len;
1639
1640 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1641 &eap, NULL);
1642 if (olen == -1) {
1643 /* delete but nonexistent */
1644 free(eae, M_TEMP);
1645 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1646 return (ENOATTR);
1647 }
1648 ul = eap->ea_length;
1649 i = (u_char *)EXTATTR_NEXT(eap) - eae;
1650 bcopy(EXTATTR_NEXT(eap), eap, easize - i);
1651 easize -= ul;
1652
1653 tmp = ip->i_ea_area;
1654 ip->i_ea_area = eae;
1655 ip->i_ea_len = easize;
1656 free(tmp, M_TEMP);
1657 error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td);
1658 return (error);
1659 }
1660
1661 /*
1662 * Vnode operation to retrieve a named extended attribute.
1663 */
1664 static int
1665 ffs_getextattr(
1666 struct vop_getextattr_args /* {
1667 IN struct vnode *a_vp;
1668 IN int a_attrnamespace;
1669 IN const char *a_name;
1670 INOUT struct uio *a_uio;
1671 OUT size_t *a_size;
1672 IN struct ucred *a_cred;
1673 IN struct thread *a_td;
1674 } */ *ap)
1675 {
1676 struct inode *ip;
1677 u_char *eae, *p;
1678 unsigned easize;
1679 int error, ealen;
1680
1681 ip = VTOI(ap->a_vp);
1682
1683 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1684 return (EOPNOTSUPP);
1685
1686 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1687 ap->a_cred, ap->a_td, VREAD);
1688 if (error)
1689 return (error);
1690
1691 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1692 if (error)
1693 return (error);
1694
1695 eae = ip->i_ea_area;
1696 easize = ip->i_ea_len;
1697
1698 ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1699 NULL, &p);
1700 if (ealen >= 0) {
1701 error = 0;
1702 if (ap->a_size != NULL)
1703 *ap->a_size = ealen;
1704 else if (ap->a_uio != NULL)
1705 error = uiomove(p, ealen, ap->a_uio);
1706 } else
1707 error = ENOATTR;
1708
1709 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1710 return (error);
1711 }
1712
1713 /*
1714 * Vnode operation to retrieve extended attributes on a vnode.
1715 */
1716 static int
1717 ffs_listextattr(
1718 struct vop_listextattr_args /* {
1719 IN struct vnode *a_vp;
1720 IN int a_attrnamespace;
1721 INOUT struct uio *a_uio;
1722 OUT size_t *a_size;
1723 IN struct ucred *a_cred;
1724 IN struct thread *a_td;
1725 } */ *ap)
1726 {
1727 struct inode *ip;
1728 struct extattr *eap, *eaend;
1729 int error, ealen;
1730
1731 ip = VTOI(ap->a_vp);
1732
1733 if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK)
1734 return (EOPNOTSUPP);
1735
1736 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
1737 ap->a_cred, ap->a_td, VREAD);
1738 if (error)
1739 return (error);
1740
1741 error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td);
1742 if (error)
1743 return (error);
1744
1745 error = 0;
1746 if (ap->a_size != NULL)
1747 *ap->a_size = 0;
1748
1749 KASSERT(ALIGNED_TO(ip->i_ea_area, struct extattr), ("unaligned"));
1750 eap = (struct extattr *)ip->i_ea_area;
1751 eaend = (struct extattr *)(ip->i_ea_area + ip->i_ea_len);
1752 for (; error == 0 && eap < eaend; eap = EXTATTR_NEXT(eap)) {
1753 KASSERT(EXTATTR_NEXT(eap) <= eaend,
1754 ("extattr next %p beyond %p", EXTATTR_NEXT(eap), eaend));
1755 if (eap->ea_namespace != ap->a_attrnamespace)
1756 continue;
1757
1758 ealen = eap->ea_namelength;
1759 if (ap->a_size != NULL)
1760 *ap->a_size += ealen + 1;
1761 else if (ap->a_uio != NULL)
1762 error = uiomove(&eap->ea_namelength, ealen + 1,
1763 ap->a_uio);
1764 }
1765
1766 ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td);
1767 return (error);
1768 }
1769
1770 /*
1771 * Vnode operation to set a named attribute.
1772 */
1773 static int
1774 ffs_setextattr(
1775 struct vop_setextattr_args /* {
1776 IN struct vnode *a_vp;
1777 IN int a_attrnamespace;
1778 IN const char *a_name;
1779 INOUT struct uio *a_uio;
1780 IN struct ucred *a_cred;
1781 IN struct thread *a_td;
1782 } */ *ap)
1783 {
1784 struct vnode *vp;
1785 struct inode *ip;
1786 struct fs *fs;
1787 struct extattr *eap;
1788 uint32_t ealength, ul;
1789 ssize_t ealen;
1790 int olen, eapad1, eapad2, error, i, easize;
1791 u_char *eae;
1792 void *tmp;
1793
1794 vp = ap->a_vp;
1795 ip = VTOI(vp);
1796 fs = ITOFS(ip);
1797
1798 if (vp->v_type == VCHR || vp->v_type == VBLK)
1799 return (EOPNOTSUPP);
1800 if (strlen(ap->a_name) == 0)
1801 return (EINVAL);
1802
1803 /* XXX Now unsupported API to delete EAs using NULL uio. */
1804 if (ap->a_uio == NULL)
1805 return (EOPNOTSUPP);
1806
1807 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1808 return (EROFS);
1809
1810 ealen = ap->a_uio->uio_resid;
1811 if (ealen < 0 || ealen > lblktosize(fs, UFS_NXADDR))
1812 return (EINVAL);
1813
1814 error = extattr_check_cred(vp, ap->a_attrnamespace,
1815 ap->a_cred, ap->a_td, VWRITE);
1816 if (error) {
1817 /*
1818 * ffs_lock_ea is not needed there, because the vnode
1819 * must be exclusively locked.
1820 */
1821 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1822 ip->i_ea_error = error;
1823 return (error);
1824 }
1825
1826 if (DOINGSUJ(vp)) {
1827 ASSERT_VOP_ELOCKED(vp, "ffs_deleteextattr");
1828 softdep_prealloc(vp, MNT_WAIT);
1829 if (vp->v_data == NULL)
1830 return (EBADF);
1831 }
1832
1833 error = ffs_open_ea(vp, ap->a_cred, ap->a_td);
1834 if (error)
1835 return (error);
1836
1837 ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name);
1838 eapad1 = roundup2(ealength, 8) - ealength;
1839 eapad2 = roundup2(ealen, 8) - ealen;
1840 ealength += eapad1 + ealen + eapad2;
1841
1842 /*
1843 * CEM: rewrites of the same size or smaller could be done in-place
1844 * instead. (We don't acquire any fine-grained locks in here either,
1845 * so we could also do bigger writes in-place.)
1846 */
1847 eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK);
1848 bcopy(ip->i_ea_area, eae, ip->i_ea_len);
1849 easize = ip->i_ea_len;
1850
1851 olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name,
1852 &eap, NULL);
1853 if (olen == -1) {
1854 /* new, append at end */
1855 KASSERT(ALIGNED_TO(eae + easize, struct extattr),
1856 ("unaligned"));
1857 eap = (struct extattr *)(eae + easize);
1858 easize += ealength;
1859 } else {
1860 ul = eap->ea_length;
1861 i = (u_char *)EXTATTR_NEXT(eap) - eae;
1862 if (ul != ealength) {
1863 bcopy(EXTATTR_NEXT(eap), (u_char *)eap + ealength,
1864 easize - i);
1865 easize += (ealength - ul);
1866 }
1867 }
1868 if (easize > lblktosize(fs, UFS_NXADDR)) {
1869 free(eae, M_TEMP);
1870 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1871 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1872 ip->i_ea_error = ENOSPC;
1873 return (ENOSPC);
1874 }
1875 eap->ea_length = ealength;
1876 eap->ea_namespace = ap->a_attrnamespace;
1877 eap->ea_contentpadlen = eapad2;
1878 eap->ea_namelength = strlen(ap->a_name);
1879 memcpy(eap->ea_name, ap->a_name, strlen(ap->a_name));
1880 bzero(&eap->ea_name[strlen(ap->a_name)], eapad1);
1881 error = uiomove(EXTATTR_CONTENT(eap), ealen, ap->a_uio);
1882 if (error) {
1883 free(eae, M_TEMP);
1884 ffs_close_ea(vp, 0, ap->a_cred, ap->a_td);
1885 if (ip->i_ea_area != NULL && ip->i_ea_error == 0)
1886 ip->i_ea_error = error;
1887 return (error);
1888 }
1889 bzero((u_char *)EXTATTR_CONTENT(eap) + ealen, eapad2);
1890
1891 tmp = ip->i_ea_area;
1892 ip->i_ea_area = eae;
1893 ip->i_ea_len = easize;
1894 free(tmp, M_TEMP);
1895 error = ffs_close_ea(vp, 1, ap->a_cred, ap->a_td);
1896 return (error);
1897 }
1898
1899 /*
1900 * Vnode pointer to File handle
1901 */
1902 static int
1903 ffs_vptofh(
1904 struct vop_vptofh_args /* {
1905 IN struct vnode *a_vp;
1906 IN struct fid *a_fhp;
1907 } */ *ap)
1908 {
1909 struct inode *ip;
1910 struct ufid *ufhp;
1911
1912 ip = VTOI(ap->a_vp);
1913 ufhp = (struct ufid *)ap->a_fhp;
1914 ufhp->ufid_len = sizeof(struct ufid);
1915 ufhp->ufid_ino = ip->i_number;
1916 ufhp->ufid_gen = ip->i_gen;
1917 return (0);
1918 }
1919
1920 SYSCTL_DECL(_vfs_ffs);
1921 static int use_buf_pager = 1;
1922 SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0,
1923 "Always use buffer pager instead of bmap");
1924
1925 static daddr_t
1926 ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1927 {
1928
1929 return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off));
1930 }
1931
1932 static int
1933 ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *sz)
1934 {
1935
1936 *sz = blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn);
1937 return (0);
1938 }
1939
1940 static int
1941 ffs_getpages(struct vop_getpages_args *ap)
1942 {
1943 struct vnode *vp;
1944 struct ufsmount *um;
1945
1946 vp = ap->a_vp;
1947 um = VFSTOUFS(vp->v_mount);
1948
1949 if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE)
1950 return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1951 ap->a_rbehind, ap->a_rahead, NULL, NULL));
1952 return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1953 ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz));
1954 }
1955
1956 static int
1957 ffs_getpages_async(struct vop_getpages_async_args *ap)
1958 {
1959 struct vnode *vp;
1960 struct ufsmount *um;
1961 bool do_iodone;
1962 int error;
1963
1964 vp = ap->a_vp;
1965 um = VFSTOUFS(vp->v_mount);
1966 do_iodone = true;
1967
1968 if (um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) {
1969 error = vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count,
1970 ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg);
1971 if (error == 0)
1972 do_iodone = false;
1973 } else {
1974 error = vfs_bio_getpages(vp, ap->a_m, ap->a_count,
1975 ap->a_rbehind, ap->a_rahead, ffs_gbp_getblkno,
1976 ffs_gbp_getblksz);
1977 }
1978 if (do_iodone && ap->a_iodone != NULL)
1979 ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
1980
1981 return (error);
1982 }
1983
1984 static int
1985 ffs_vput_pair(struct vop_vput_pair_args *ap)
1986 {
1987 struct mount *mp;
1988 struct vnode *dvp, *vp, *vp1, **vpp;
1989 struct inode *dp, *ip;
1990 ino_t ip_ino;
1991 u_int64_t ip_gen;
1992 int error, vp_locked;
1993
1994 dvp = ap->a_dvp;
1995 dp = VTOI(dvp);
1996 vpp = ap->a_vpp;
1997 vp = vpp != NULL ? *vpp : NULL;
1998
1999 if ((dp->i_flag & (IN_NEEDSYNC | IN_ENDOFF)) == 0) {
2000 vput(dvp);
2001 if (vp != NULL && ap->a_unlock_vp)
2002 vput(vp);
2003 return (0);
2004 }
2005
2006 mp = dvp->v_mount;
2007 if (vp != NULL) {
2008 if (ap->a_unlock_vp) {
2009 vput(vp);
2010 } else {
2011 MPASS(vp->v_type != VNON);
2012 vp_locked = VOP_ISLOCKED(vp);
2013 ip = VTOI(vp);
2014 ip_ino = ip->i_number;
2015 ip_gen = ip->i_gen;
2016 VOP_UNLOCK(vp);
2017 }
2018 }
2019
2020 /*
2021 * If compaction or fsync was requested do it in ffs_vput_pair()
2022 * now that other locks are no longer held.
2023 */
2024 if ((dp->i_flag & IN_ENDOFF) != 0) {
2025 VNASSERT(I_ENDOFF(dp) != 0 && I_ENDOFF(dp) < dp->i_size, dvp,
2026 ("IN_ENDOFF set but I_ENDOFF() is not"));
2027 dp->i_flag &= ~IN_ENDOFF;
2028 error = UFS_TRUNCATE(dvp, (off_t)I_ENDOFF(dp), IO_NORMAL |
2029 (DOINGASYNC(dvp) ? 0 : IO_SYNC), curthread->td_ucred);
2030 if (error != 0 && error != ERELOOKUP) {
2031 if (!ffs_fsfail_cleanup(VFSTOUFS(mp), error)) {
2032 vn_printf(dvp,
2033 "IN_ENDOFF: failed to truncate, "
2034 "error %d\n", error);
2035 }
2036 #ifdef UFS_DIRHASH
2037 ufsdirhash_free(dp);
2038 #endif
2039 }
2040 SET_I_ENDOFF(dp, 0);
2041 }
2042 if ((dp->i_flag & IN_NEEDSYNC) != 0) {
2043 do {
2044 error = ffs_syncvnode(dvp, MNT_WAIT, 0);
2045 } while (error == ERELOOKUP);
2046 }
2047
2048 vput(dvp);
2049
2050 if (vp == NULL || ap->a_unlock_vp)
2051 return (0);
2052 MPASS(mp != NULL);
2053
2054 /*
2055 * It is possible that vp is reclaimed at this point. Only
2056 * routines that call us with a_unlock_vp == false can find
2057 * that their vp has been reclaimed. There are three areas
2058 * that are affected:
2059 * 1) vn_open_cred() - later VOPs could fail, but
2060 * dead_open() returns 0 to simulate successful open.
2061 * 2) ffs_snapshot() - creation of snapshot fails with EBADF.
2062 * 3) NFS server (several places) - code is prepared to detect
2063 * and respond to dead vnodes by returning ESTALE.
2064 */
2065 VOP_LOCK(vp, vp_locked | LK_RETRY);
2066 if (IS_UFS(vp))
2067 return (0);
2068
2069 /*
2070 * Try harder to recover from reclaimed vp if reclaim was not
2071 * because underlying inode was cleared. We saved inode
2072 * number and inode generation, so we can try to reinstantiate
2073 * exactly same version of inode. If this fails, return
2074 * original doomed vnode and let caller to handle
2075 * consequences.
2076 *
2077 * Note that callers must keep write started around
2078 * VOP_VPUT_PAIR() calls, so it is safe to use mp without
2079 * busying it.
2080 */
2081 VOP_UNLOCK(vp);
2082 error = ffs_inotovp(mp, ip_ino, ip_gen, LK_EXCLUSIVE, &vp1,
2083 FFSV_REPLACE_DOOMED);
2084 if (error != 0) {
2085 VOP_LOCK(vp, vp_locked | LK_RETRY);
2086 } else {
2087 vrele(vp);
2088 *vpp = vp1;
2089 }
2090 return (error);
2091 }
Cache object: bc1b17d2270b54f6d6d1aaa8eb23e23b
|