sched: fix CONFIG_SCHED_DEBUG dependency of lockdep sysctls
[usb.git] / fs / xfs / xfs_vnodeops.c
blob1a5ad8cd97b00d3d8cd24c04961ea075cb5e1635
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_types.h"
22 #include "xfs_bit.h"
23 #include "xfs_log.h"
24 #include "xfs_inum.h"
25 #include "xfs_trans.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_itable.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
44 #include "xfs_bmap.h"
45 #include "xfs_attr.h"
46 #include "xfs_rw.h"
47 #include "xfs_error.h"
48 #include "xfs_quota.h"
49 #include "xfs_utils.h"
50 #include "xfs_rtalloc.h"
51 #include "xfs_refcache.h"
52 #include "xfs_trans_space.h"
53 #include "xfs_log_priv.h"
54 #include "xfs_filestream.h"
56 STATIC int
57 xfs_open(
58 bhv_desc_t *bdp,
59 cred_t *credp)
61 int mode;
62 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
63 xfs_inode_t *ip = XFS_BHVTOI(bdp);
65 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
66 return XFS_ERROR(EIO);
69 * If it's a directory with any blocks, read-ahead block 0
70 * as we're almost certain to have the next operation be a read there.
72 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
73 mode = xfs_ilock_map_shared(ip);
74 if (ip->i_d.di_nextents > 0)
75 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
76 xfs_iunlock(ip, mode);
78 return 0;
82 * xfs_getattr
84 STATIC int
85 xfs_getattr(
86 bhv_desc_t *bdp,
87 bhv_vattr_t *vap,
88 int flags,
89 cred_t *credp)
91 xfs_inode_t *ip;
92 xfs_mount_t *mp;
93 bhv_vnode_t *vp;
95 vp = BHV_TO_VNODE(bdp);
96 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
98 ip = XFS_BHVTOI(bdp);
99 mp = ip->i_mount;
101 if (XFS_FORCED_SHUTDOWN(mp))
102 return XFS_ERROR(EIO);
104 if (!(flags & ATTR_LAZY))
105 xfs_ilock(ip, XFS_ILOCK_SHARED);
107 vap->va_size = XFS_ISIZE(ip);
108 if (vap->va_mask == XFS_AT_SIZE)
109 goto all_done;
111 vap->va_nblocks =
112 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
113 vap->va_nodeid = ip->i_ino;
114 #if XFS_BIG_INUMS
115 vap->va_nodeid += mp->m_inoadd;
116 #endif
117 vap->va_nlink = ip->i_d.di_nlink;
120 * Quick exit for non-stat callers
122 if ((vap->va_mask &
123 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
124 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
125 goto all_done;
128 * Copy from in-core inode.
130 vap->va_mode = ip->i_d.di_mode;
131 vap->va_uid = ip->i_d.di_uid;
132 vap->va_gid = ip->i_d.di_gid;
133 vap->va_projid = ip->i_d.di_projid;
136 * Check vnode type block/char vs. everything else.
138 switch (ip->i_d.di_mode & S_IFMT) {
139 case S_IFBLK:
140 case S_IFCHR:
141 vap->va_rdev = ip->i_df.if_u2.if_rdev;
142 vap->va_blocksize = BLKDEV_IOSIZE;
143 break;
144 default:
145 vap->va_rdev = 0;
147 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
148 vap->va_blocksize = xfs_preferred_iosize(mp);
149 } else {
152 * If the file blocks are being allocated from a
153 * realtime partition, then return the inode's
154 * realtime extent size or the realtime volume's
155 * extent size.
157 vap->va_blocksize =
158 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
160 break;
163 vn_atime_to_timespec(vp, &vap->va_atime);
164 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
165 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
166 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
167 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
170 * Exit for stat callers. See if any of the rest of the fields
171 * to be filled in are needed.
173 if ((vap->va_mask &
174 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
175 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
176 goto all_done;
179 * Convert di_flags to xflags.
181 vap->va_xflags = xfs_ip2xflags(ip);
184 * Exit for inode revalidate. See if any of the rest of
185 * the fields to be filled in are needed.
187 if ((vap->va_mask &
188 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
189 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
190 goto all_done;
192 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
193 vap->va_nextents =
194 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
195 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
196 ip->i_d.di_nextents;
197 if (ip->i_afp)
198 vap->va_anextents =
199 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
200 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
201 ip->i_d.di_anextents;
202 else
203 vap->va_anextents = 0;
204 vap->va_gen = ip->i_d.di_gen;
206 all_done:
207 if (!(flags & ATTR_LAZY))
208 xfs_iunlock(ip, XFS_ILOCK_SHARED);
209 return 0;
214 * xfs_setattr
217 xfs_setattr(
218 bhv_desc_t *bdp,
219 bhv_vattr_t *vap,
220 int flags,
221 cred_t *credp)
223 xfs_inode_t *ip;
224 xfs_trans_t *tp;
225 xfs_mount_t *mp;
226 int mask;
227 int code;
228 uint lock_flags;
229 uint commit_flags=0;
230 uid_t uid=0, iuid=0;
231 gid_t gid=0, igid=0;
232 int timeflags = 0;
233 bhv_vnode_t *vp;
234 xfs_prid_t projid=0, iprojid=0;
235 int mandlock_before, mandlock_after;
236 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
237 int file_owner;
238 int need_iolock = 1;
240 vp = BHV_TO_VNODE(bdp);
241 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
243 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
244 return XFS_ERROR(EROFS);
247 * Cannot set certain attributes.
249 mask = vap->va_mask;
250 if (mask & XFS_AT_NOSET) {
251 return XFS_ERROR(EINVAL);
254 ip = XFS_BHVTOI(bdp);
255 mp = ip->i_mount;
257 if (XFS_FORCED_SHUTDOWN(mp))
258 return XFS_ERROR(EIO);
261 * Timestamps do not need to be logged and hence do not
262 * need to be done within a transaction.
264 if (mask & XFS_AT_UPDTIMES) {
265 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
266 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
267 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
268 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
269 xfs_ichgtime(ip, timeflags);
270 return 0;
273 olddquot1 = olddquot2 = NULL;
274 udqp = gdqp = NULL;
277 * If disk quotas is on, we make sure that the dquots do exist on disk,
278 * before we start any other transactions. Trying to do this later
279 * is messy. We don't care to take a readlock to look at the ids
280 * in inode here, because we can't hold it across the trans_reserve.
281 * If the IDs do change before we take the ilock, we're covered
282 * because the i_*dquot fields will get updated anyway.
284 if (XFS_IS_QUOTA_ON(mp) &&
285 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
286 uint qflags = 0;
288 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
289 uid = vap->va_uid;
290 qflags |= XFS_QMOPT_UQUOTA;
291 } else {
292 uid = ip->i_d.di_uid;
294 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
295 gid = vap->va_gid;
296 qflags |= XFS_QMOPT_GQUOTA;
297 } else {
298 gid = ip->i_d.di_gid;
300 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
301 projid = vap->va_projid;
302 qflags |= XFS_QMOPT_PQUOTA;
303 } else {
304 projid = ip->i_d.di_projid;
307 * We take a reference when we initialize udqp and gdqp,
308 * so it is important that we never blindly double trip on
309 * the same variable. See xfs_create() for an example.
311 ASSERT(udqp == NULL);
312 ASSERT(gdqp == NULL);
313 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
314 &udqp, &gdqp);
315 if (code)
316 return code;
320 * For the other attributes, we acquire the inode lock and
321 * first do an error checking pass.
323 tp = NULL;
324 lock_flags = XFS_ILOCK_EXCL;
325 if (flags & ATTR_NOLOCK)
326 need_iolock = 0;
327 if (!(mask & XFS_AT_SIZE)) {
328 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
329 (mp->m_flags & XFS_MOUNT_WSYNC)) {
330 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
331 commit_flags = 0;
332 if ((code = xfs_trans_reserve(tp, 0,
333 XFS_ICHANGE_LOG_RES(mp), 0,
334 0, 0))) {
335 lock_flags = 0;
336 goto error_return;
339 } else {
340 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
341 !(flags & ATTR_DMI)) {
342 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
343 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
344 vap->va_size, 0, dmflags, NULL);
345 if (code) {
346 lock_flags = 0;
347 goto error_return;
350 if (need_iolock)
351 lock_flags |= XFS_IOLOCK_EXCL;
354 xfs_ilock(ip, lock_flags);
356 /* boolean: are we the file owner? */
357 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
360 * Change various properties of a file.
361 * Only the owner or users with CAP_FOWNER
362 * capability may do these things.
364 if (mask &
365 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
366 XFS_AT_GID|XFS_AT_PROJID)) {
368 * CAP_FOWNER overrides the following restrictions:
370 * The user ID of the calling process must be equal
371 * to the file owner ID, except in cases where the
372 * CAP_FSETID capability is applicable.
374 if (!file_owner && !capable(CAP_FOWNER)) {
375 code = XFS_ERROR(EPERM);
376 goto error_return;
380 * CAP_FSETID overrides the following restrictions:
382 * The effective user ID of the calling process shall match
383 * the file owner when setting the set-user-ID and
384 * set-group-ID bits on that file.
386 * The effective group ID or one of the supplementary group
387 * IDs of the calling process shall match the group owner of
388 * the file when setting the set-group-ID bit on that file
390 if (mask & XFS_AT_MODE) {
391 mode_t m = 0;
393 if ((vap->va_mode & S_ISUID) && !file_owner)
394 m |= S_ISUID;
395 if ((vap->va_mode & S_ISGID) &&
396 !in_group_p((gid_t)ip->i_d.di_gid))
397 m |= S_ISGID;
398 #if 0
399 /* Linux allows this, Irix doesn't. */
400 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
401 m |= S_ISVTX;
402 #endif
403 if (m && !capable(CAP_FSETID))
404 vap->va_mode &= ~m;
409 * Change file ownership. Must be the owner or privileged.
410 * If the system was configured with the "restricted_chown"
411 * option, the owner is not permitted to give away the file,
412 * and can change the group id only to a group of which he
413 * or she is a member.
415 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
417 * These IDs could have changed since we last looked at them.
418 * But, we're assured that if the ownership did change
419 * while we didn't have the inode locked, inode's dquot(s)
420 * would have changed also.
422 iuid = ip->i_d.di_uid;
423 iprojid = ip->i_d.di_projid;
424 igid = ip->i_d.di_gid;
425 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
426 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
427 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
428 iprojid;
431 * CAP_CHOWN overrides the following restrictions:
433 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
434 * shall override the restriction that a process cannot
435 * change the user ID of a file it owns and the restriction
436 * that the group ID supplied to the chown() function
437 * shall be equal to either the group ID or one of the
438 * supplementary group IDs of the calling process.
440 if (restricted_chown &&
441 (iuid != uid || (igid != gid &&
442 !in_group_p((gid_t)gid))) &&
443 !capable(CAP_CHOWN)) {
444 code = XFS_ERROR(EPERM);
445 goto error_return;
448 * Do a quota reservation only if uid/projid/gid is actually
449 * going to change.
451 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
452 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
453 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
454 ASSERT(tp);
455 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
456 capable(CAP_FOWNER) ?
457 XFS_QMOPT_FORCE_RES : 0);
458 if (code) /* out of quota */
459 goto error_return;
464 * Truncate file. Must have write permission and not be a directory.
466 if (mask & XFS_AT_SIZE) {
467 /* Short circuit the truncate case for zero length files */
468 if ((vap->va_size == 0) &&
469 (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) {
470 xfs_iunlock(ip, XFS_ILOCK_EXCL);
471 lock_flags &= ~XFS_ILOCK_EXCL;
472 if (mask & XFS_AT_CTIME)
473 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
474 code = 0;
475 goto error_return;
478 if (VN_ISDIR(vp)) {
479 code = XFS_ERROR(EISDIR);
480 goto error_return;
481 } else if (!VN_ISREG(vp)) {
482 code = XFS_ERROR(EINVAL);
483 goto error_return;
486 * Make sure that the dquots are attached to the inode.
488 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
489 goto error_return;
493 * Change file access or modified times.
495 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
496 if (!file_owner) {
497 if ((flags & ATTR_UTIME) &&
498 !capable(CAP_FOWNER)) {
499 code = XFS_ERROR(EPERM);
500 goto error_return;
506 * Change extent size or realtime flag.
508 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
510 * Can't change extent size if any extents are allocated.
512 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
513 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
514 vap->va_extsize) ) {
515 code = XFS_ERROR(EINVAL); /* EFBIG? */
516 goto error_return;
520 * Can't change realtime flag if any extents are allocated.
522 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
523 (mask & XFS_AT_XFLAGS) &&
524 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
525 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
526 code = XFS_ERROR(EINVAL); /* EFBIG? */
527 goto error_return;
530 * Extent size must be a multiple of the appropriate block
531 * size, if set at all.
533 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
534 xfs_extlen_t size;
536 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
537 ((mask & XFS_AT_XFLAGS) &&
538 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
539 size = mp->m_sb.sb_rextsize <<
540 mp->m_sb.sb_blocklog;
541 } else {
542 size = mp->m_sb.sb_blocksize;
544 if (vap->va_extsize % size) {
545 code = XFS_ERROR(EINVAL);
546 goto error_return;
550 * If realtime flag is set then must have realtime data.
552 if ((mask & XFS_AT_XFLAGS) &&
553 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
554 if ((mp->m_sb.sb_rblocks == 0) ||
555 (mp->m_sb.sb_rextsize == 0) ||
556 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
557 code = XFS_ERROR(EINVAL);
558 goto error_return;
563 * Can't modify an immutable/append-only file unless
564 * we have appropriate permission.
566 if ((mask & XFS_AT_XFLAGS) &&
567 (ip->i_d.di_flags &
568 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
569 (vap->va_xflags &
570 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
571 !capable(CAP_LINUX_IMMUTABLE)) {
572 code = XFS_ERROR(EPERM);
573 goto error_return;
578 * Now we can make the changes. Before we join the inode
579 * to the transaction, if XFS_AT_SIZE is set then take care of
580 * the part of the truncation that must be done without the
581 * inode lock. This needs to be done before joining the inode
582 * to the transaction, because the inode cannot be unlocked
583 * once it is a part of the transaction.
585 if (mask & XFS_AT_SIZE) {
586 code = 0;
587 if ((vap->va_size > ip->i_size) &&
588 (flags & ATTR_NOSIZETOK) == 0) {
589 code = xfs_igrow_start(ip, vap->va_size, credp);
591 xfs_iunlock(ip, XFS_ILOCK_EXCL);
594 * We are going to log the inode size change in this
595 * transaction so any previous writes that are beyond the on
596 * disk EOF and the new EOF that have not been written out need
597 * to be written here. If we do not write the data out, we
598 * expose ourselves to the null files problem.
600 * Only flush from the on disk size to the smaller of the in
601 * memory file size or the new size as that's the range we
602 * really care about here and prevents waiting for other data
603 * not within the range we care about here.
605 if (!code &&
606 (ip->i_size != ip->i_d.di_size) &&
607 (vap->va_size > ip->i_d.di_size)) {
608 code = bhv_vop_flush_pages(XFS_ITOV(ip),
609 ip->i_d.di_size, vap->va_size,
610 XFS_B_ASYNC, FI_NONE);
613 /* wait for all I/O to complete */
614 vn_iowait(vp);
616 if (!code)
617 code = xfs_itruncate_data(ip, vap->va_size);
618 if (code) {
619 ASSERT(tp == NULL);
620 lock_flags &= ~XFS_ILOCK_EXCL;
621 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
622 goto error_return;
624 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
625 if ((code = xfs_trans_reserve(tp, 0,
626 XFS_ITRUNCATE_LOG_RES(mp), 0,
627 XFS_TRANS_PERM_LOG_RES,
628 XFS_ITRUNCATE_LOG_COUNT))) {
629 xfs_trans_cancel(tp, 0);
630 if (need_iolock)
631 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
632 return code;
634 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
635 xfs_ilock(ip, XFS_ILOCK_EXCL);
638 if (tp) {
639 xfs_trans_ijoin(tp, ip, lock_flags);
640 xfs_trans_ihold(tp, ip);
643 /* determine whether mandatory locking mode changes */
644 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
647 * Truncate file. Must have write permission and not be a directory.
649 if (mask & XFS_AT_SIZE) {
650 if (vap->va_size > ip->i_size) {
651 xfs_igrow_finish(tp, ip, vap->va_size,
652 !(flags & ATTR_DMI));
653 } else if ((vap->va_size <= ip->i_size) ||
654 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
656 * signal a sync transaction unless
657 * we're truncating an already unlinked
658 * file on a wsync filesystem
660 code = xfs_itruncate_finish(&tp, ip,
661 (xfs_fsize_t)vap->va_size,
662 XFS_DATA_FORK,
663 ((ip->i_d.di_nlink != 0 ||
664 !(mp->m_flags & XFS_MOUNT_WSYNC))
665 ? 1 : 0));
666 if (code)
667 goto abort_return;
669 * Truncated "down", so we're removing references
670 * to old data here - if we now delay flushing for
671 * a long time, we expose ourselves unduly to the
672 * notorious NULL files problem. So, we mark this
673 * vnode and flush it when the file is closed, and
674 * do not wait the usual (long) time for writeout.
676 VTRUNCATE(vp);
679 * Have to do this even if the file's size doesn't change.
681 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
685 * Change file access modes.
687 if (mask & XFS_AT_MODE) {
688 ip->i_d.di_mode &= S_IFMT;
689 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
691 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
692 timeflags |= XFS_ICHGTIME_CHG;
696 * Change file ownership. Must be the owner or privileged.
697 * If the system was configured with the "restricted_chown"
698 * option, the owner is not permitted to give away the file,
699 * and can change the group id only to a group of which he
700 * or she is a member.
702 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
704 * CAP_FSETID overrides the following restrictions:
706 * The set-user-ID and set-group-ID bits of a file will be
707 * cleared upon successful return from chown()
709 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
710 !capable(CAP_FSETID)) {
711 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
715 * Change the ownerships and register quota modifications
716 * in the transaction.
718 if (iuid != uid) {
719 if (XFS_IS_UQUOTA_ON(mp)) {
720 ASSERT(mask & XFS_AT_UID);
721 ASSERT(udqp);
722 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
723 &ip->i_udquot, udqp);
725 ip->i_d.di_uid = uid;
727 if (igid != gid) {
728 if (XFS_IS_GQUOTA_ON(mp)) {
729 ASSERT(!XFS_IS_PQUOTA_ON(mp));
730 ASSERT(mask & XFS_AT_GID);
731 ASSERT(gdqp);
732 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
733 &ip->i_gdquot, gdqp);
735 ip->i_d.di_gid = gid;
737 if (iprojid != projid) {
738 if (XFS_IS_PQUOTA_ON(mp)) {
739 ASSERT(!XFS_IS_GQUOTA_ON(mp));
740 ASSERT(mask & XFS_AT_PROJID);
741 ASSERT(gdqp);
742 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
743 &ip->i_gdquot, gdqp);
745 ip->i_d.di_projid = projid;
747 * We may have to rev the inode as well as
748 * the superblock version number since projids didn't
749 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
751 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
752 xfs_bump_ino_vers2(tp, ip);
755 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
756 timeflags |= XFS_ICHGTIME_CHG;
761 * Change file access or modified times.
763 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
764 if (mask & XFS_AT_ATIME) {
765 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
766 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
767 ip->i_update_core = 1;
768 timeflags &= ~XFS_ICHGTIME_ACC;
770 if (mask & XFS_AT_MTIME) {
771 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
772 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
773 timeflags &= ~XFS_ICHGTIME_MOD;
774 timeflags |= XFS_ICHGTIME_CHG;
776 if (tp && (flags & ATTR_UTIME))
777 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
781 * Change XFS-added attributes.
783 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
784 if (mask & XFS_AT_EXTSIZE) {
786 * Converting bytes to fs blocks.
788 ip->i_d.di_extsize = vap->va_extsize >>
789 mp->m_sb.sb_blocklog;
791 if (mask & XFS_AT_XFLAGS) {
792 uint di_flags;
794 /* can't set PREALLOC this way, just preserve it */
795 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
796 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
797 di_flags |= XFS_DIFLAG_IMMUTABLE;
798 if (vap->va_xflags & XFS_XFLAG_APPEND)
799 di_flags |= XFS_DIFLAG_APPEND;
800 if (vap->va_xflags & XFS_XFLAG_SYNC)
801 di_flags |= XFS_DIFLAG_SYNC;
802 if (vap->va_xflags & XFS_XFLAG_NOATIME)
803 di_flags |= XFS_DIFLAG_NOATIME;
804 if (vap->va_xflags & XFS_XFLAG_NODUMP)
805 di_flags |= XFS_DIFLAG_NODUMP;
806 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
807 di_flags |= XFS_DIFLAG_PROJINHERIT;
808 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
809 di_flags |= XFS_DIFLAG_NODEFRAG;
810 if (vap->va_xflags & XFS_XFLAG_FILESTREAM)
811 di_flags |= XFS_DIFLAG_FILESTREAM;
812 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
813 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
814 di_flags |= XFS_DIFLAG_RTINHERIT;
815 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
816 di_flags |= XFS_DIFLAG_NOSYMLINKS;
817 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
818 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
819 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
820 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
821 di_flags |= XFS_DIFLAG_REALTIME;
822 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
823 } else {
824 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
826 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
827 di_flags |= XFS_DIFLAG_EXTSIZE;
829 ip->i_d.di_flags = di_flags;
831 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
832 timeflags |= XFS_ICHGTIME_CHG;
836 * Change file inode change time only if XFS_AT_CTIME set
837 * AND we have been called by a DMI function.
840 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
841 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
842 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
843 ip->i_update_core = 1;
844 timeflags &= ~XFS_ICHGTIME_CHG;
848 * Send out timestamp changes that need to be set to the
849 * current time. Not done when called by a DMI function.
851 if (timeflags && !(flags & ATTR_DMI))
852 xfs_ichgtime(ip, timeflags);
854 XFS_STATS_INC(xs_ig_attrchg);
857 * If this is a synchronous mount, make sure that the
858 * transaction goes to disk before returning to the user.
859 * This is slightly sub-optimal in that truncates require
860 * two sync transactions instead of one for wsync filesystems.
861 * One for the truncate and one for the timestamps since we
862 * don't want to change the timestamps unless we're sure the
863 * truncate worked. Truncates are less than 1% of the laddis
864 * mix so this probably isn't worth the trouble to optimize.
866 code = 0;
867 if (tp) {
868 if (mp->m_flags & XFS_MOUNT_WSYNC)
869 xfs_trans_set_sync(tp);
871 code = xfs_trans_commit(tp, commit_flags);
875 * If the (regular) file's mandatory locking mode changed, then
876 * notify the vnode. We do this under the inode lock to prevent
877 * racing calls to vop_vnode_change.
879 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
880 if (mandlock_before != mandlock_after) {
881 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_ENF_LOCKING,
882 mandlock_after);
885 xfs_iunlock(ip, lock_flags);
888 * Release any dquot(s) the inode had kept before chown.
890 XFS_QM_DQRELE(mp, olddquot1);
891 XFS_QM_DQRELE(mp, olddquot2);
892 XFS_QM_DQRELE(mp, udqp);
893 XFS_QM_DQRELE(mp, gdqp);
895 if (code) {
896 return code;
899 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
900 !(flags & ATTR_DMI)) {
901 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
902 NULL, DM_RIGHT_NULL, NULL, NULL,
903 0, 0, AT_DELAY_FLAG(flags));
905 return 0;
907 abort_return:
908 commit_flags |= XFS_TRANS_ABORT;
909 /* FALLTHROUGH */
910 error_return:
911 XFS_QM_DQRELE(mp, udqp);
912 XFS_QM_DQRELE(mp, gdqp);
913 if (tp) {
914 xfs_trans_cancel(tp, commit_flags);
916 if (lock_flags != 0) {
917 xfs_iunlock(ip, lock_flags);
919 return code;
924 * xfs_access
925 * Null conversion from vnode mode bits to inode mode bits, as in efs.
927 STATIC int
928 xfs_access(
929 bhv_desc_t *bdp,
930 int mode,
931 cred_t *credp)
933 xfs_inode_t *ip;
934 int error;
936 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
937 (inst_t *)__return_address);
939 ip = XFS_BHVTOI(bdp);
940 xfs_ilock(ip, XFS_ILOCK_SHARED);
941 error = xfs_iaccess(ip, mode, credp);
942 xfs_iunlock(ip, XFS_ILOCK_SHARED);
943 return error;
948 * The maximum pathlen is 1024 bytes. Since the minimum file system
949 * blocksize is 512 bytes, we can get a max of 2 extents back from
950 * bmapi.
952 #define SYMLINK_MAPS 2
955 * xfs_readlink
958 STATIC int
959 xfs_readlink(
960 bhv_desc_t *bdp,
961 uio_t *uiop,
962 int ioflags,
963 cred_t *credp)
965 xfs_inode_t *ip;
966 int count;
967 xfs_off_t offset;
968 int pathlen;
969 bhv_vnode_t *vp;
970 int error = 0;
971 xfs_mount_t *mp;
972 int nmaps;
973 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
974 xfs_daddr_t d;
975 int byte_cnt;
976 int n;
977 xfs_buf_t *bp;
979 vp = BHV_TO_VNODE(bdp);
980 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
982 ip = XFS_BHVTOI(bdp);
983 mp = ip->i_mount;
985 if (XFS_FORCED_SHUTDOWN(mp))
986 return XFS_ERROR(EIO);
988 xfs_ilock(ip, XFS_ILOCK_SHARED);
990 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
992 offset = uiop->uio_offset;
993 count = uiop->uio_resid;
995 if (offset < 0) {
996 error = XFS_ERROR(EINVAL);
997 goto error_return;
999 if (count <= 0) {
1000 error = 0;
1001 goto error_return;
1005 * See if the symlink is stored inline.
1007 pathlen = (int)ip->i_d.di_size;
1009 if (ip->i_df.if_flags & XFS_IFINLINE) {
1010 error = xfs_uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1012 else {
1014 * Symlink not inline. Call bmap to get it in.
1016 nmaps = SYMLINK_MAPS;
1018 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1019 0, NULL, 0, mval, &nmaps, NULL, NULL);
1021 if (error) {
1022 goto error_return;
1025 for (n = 0; n < nmaps; n++) {
1026 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1027 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1028 bp = xfs_buf_read(mp->m_ddev_targp, d,
1029 BTOBB(byte_cnt), 0);
1030 error = XFS_BUF_GETERROR(bp);
1031 if (error) {
1032 xfs_ioerror_alert("xfs_readlink",
1033 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1034 xfs_buf_relse(bp);
1035 goto error_return;
1037 if (pathlen < byte_cnt)
1038 byte_cnt = pathlen;
1039 pathlen -= byte_cnt;
1041 error = xfs_uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1042 xfs_buf_relse (bp);
1047 error_return:
1048 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1049 return error;
1054 * xfs_fsync
1056 * This is called to sync the inode and its data out to disk.
1057 * We need to hold the I/O lock while flushing the data, and
1058 * the inode lock while flushing the inode. The inode lock CANNOT
1059 * be held while flushing the data, so acquire after we're done
1060 * with that.
1062 STATIC int
1063 xfs_fsync(
1064 bhv_desc_t *bdp,
1065 int flag,
1066 cred_t *credp,
1067 xfs_off_t start,
1068 xfs_off_t stop)
1070 xfs_inode_t *ip;
1071 xfs_trans_t *tp;
1072 int error;
1073 int log_flushed = 0, changed = 1;
1075 vn_trace_entry(BHV_TO_VNODE(bdp),
1076 __FUNCTION__, (inst_t *)__return_address);
1078 ip = XFS_BHVTOI(bdp);
1080 ASSERT(start >= 0 && stop >= -1);
1082 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1083 return XFS_ERROR(EIO);
1086 * We always need to make sure that the required inode state
1087 * is safe on disk. The vnode might be clean but because
1088 * of committed transactions that haven't hit the disk yet.
1089 * Likewise, there could be unflushed non-transactional
1090 * changes to the inode core that have to go to disk.
1092 * The following code depends on one assumption: that
1093 * any transaction that changes an inode logs the core
1094 * because it has to change some field in the inode core
1095 * (typically nextents or nblocks). That assumption
1096 * implies that any transactions against an inode will
1097 * catch any non-transactional updates. If inode-altering
1098 * transactions exist that violate this assumption, the
1099 * code breaks. Right now, it figures that if the involved
1100 * update_* field is clear and the inode is unpinned, the
1101 * inode is clean. Either it's been flushed or it's been
1102 * committed and the commit has hit the disk unpinning the inode.
1103 * (Note that xfs_inode_item_format() called at commit clears
1104 * the update_* fields.)
1106 xfs_ilock(ip, XFS_ILOCK_SHARED);
1108 /* If we are flushing data then we care about update_size
1109 * being set, otherwise we care about update_core
1111 if ((flag & FSYNC_DATA) ?
1112 (ip->i_update_size == 0) :
1113 (ip->i_update_core == 0)) {
1115 * Timestamps/size haven't changed since last inode
1116 * flush or inode transaction commit. That means
1117 * either nothing got written or a transaction
1118 * committed which caught the updates. If the
1119 * latter happened and the transaction hasn't
1120 * hit the disk yet, the inode will be still
1121 * be pinned. If it is, force the log.
1124 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1126 if (xfs_ipincount(ip)) {
1127 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1128 XFS_LOG_FORCE |
1129 ((flag & FSYNC_WAIT)
1130 ? XFS_LOG_SYNC : 0),
1131 &log_flushed);
1132 } else {
1134 * If the inode is not pinned and nothing
1135 * has changed we don't need to flush the
1136 * cache.
1138 changed = 0;
1140 error = 0;
1141 } else {
1143 * Kick off a transaction to log the inode
1144 * core to get the updates. Make it
1145 * sync if FSYNC_WAIT is passed in (which
1146 * is done by everybody but specfs). The
1147 * sync transaction will also force the log.
1149 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1150 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1151 if ((error = xfs_trans_reserve(tp, 0,
1152 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1153 0, 0, 0))) {
1154 xfs_trans_cancel(tp, 0);
1155 return error;
1157 xfs_ilock(ip, XFS_ILOCK_EXCL);
1160 * Note - it's possible that we might have pushed
1161 * ourselves out of the way during trans_reserve
1162 * which would flush the inode. But there's no
1163 * guarantee that the inode buffer has actually
1164 * gone out yet (it's delwri). Plus the buffer
1165 * could be pinned anyway if it's part of an
1166 * inode in another recent transaction. So we
1167 * play it safe and fire off the transaction anyway.
1169 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1170 xfs_trans_ihold(tp, ip);
1171 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1172 if (flag & FSYNC_WAIT)
1173 xfs_trans_set_sync(tp);
1174 error = _xfs_trans_commit(tp, 0, &log_flushed);
1176 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1179 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1181 * If the log write didn't issue an ordered tag we need
1182 * to flush the disk cache for the data device now.
1184 if (!log_flushed)
1185 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1188 * If this inode is on the RT dev we need to flush that
1189 * cache as well.
1191 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1192 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1195 return error;
1199 * This is called by xfs_inactive to free any blocks beyond eof
1200 * when the link count isn't zero and by xfs_dm_punch_hole() when
1201 * punching a hole to EOF.
1204 xfs_free_eofblocks(
1205 xfs_mount_t *mp,
1206 xfs_inode_t *ip,
1207 int flags)
1209 xfs_trans_t *tp;
1210 int error;
1211 xfs_fileoff_t end_fsb;
1212 xfs_fileoff_t last_fsb;
1213 xfs_filblks_t map_len;
1214 int nimaps;
1215 xfs_bmbt_irec_t imap;
1216 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
1219 * Figure out if there are any blocks beyond the end
1220 * of the file. If not, then there is nothing to do.
1222 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
1223 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1224 map_len = last_fsb - end_fsb;
1225 if (map_len <= 0)
1226 return 0;
1228 nimaps = 1;
1229 xfs_ilock(ip, XFS_ILOCK_SHARED);
1230 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0,
1231 NULL, 0, &imap, &nimaps, NULL, NULL);
1232 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1234 if (!error && (nimaps != 0) &&
1235 (imap.br_startblock != HOLESTARTBLOCK ||
1236 ip->i_delayed_blks)) {
1238 * Attach the dquots to the inode up front.
1240 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1241 return error;
1244 * There are blocks after the end of file.
1245 * Free them up now by truncating the file to
1246 * its current size.
1248 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1251 * Do the xfs_itruncate_start() call before
1252 * reserving any log space because
1253 * itruncate_start will call into the buffer
1254 * cache and we can't
1255 * do that within a transaction.
1257 if (use_iolock)
1258 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1259 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1260 ip->i_size);
1261 if (error) {
1262 xfs_trans_cancel(tp, 0);
1263 if (use_iolock)
1264 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1265 return error;
1268 error = xfs_trans_reserve(tp, 0,
1269 XFS_ITRUNCATE_LOG_RES(mp),
1270 0, XFS_TRANS_PERM_LOG_RES,
1271 XFS_ITRUNCATE_LOG_COUNT);
1272 if (error) {
1273 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1274 xfs_trans_cancel(tp, 0);
1275 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1276 return error;
1279 xfs_ilock(ip, XFS_ILOCK_EXCL);
1280 xfs_trans_ijoin(tp, ip,
1281 XFS_IOLOCK_EXCL |
1282 XFS_ILOCK_EXCL);
1283 xfs_trans_ihold(tp, ip);
1285 error = xfs_itruncate_finish(&tp, ip,
1286 ip->i_size,
1287 XFS_DATA_FORK,
1290 * If we get an error at this point we
1291 * simply don't bother truncating the file.
1293 if (error) {
1294 xfs_trans_cancel(tp,
1295 (XFS_TRANS_RELEASE_LOG_RES |
1296 XFS_TRANS_ABORT));
1297 } else {
1298 error = xfs_trans_commit(tp,
1299 XFS_TRANS_RELEASE_LOG_RES);
1301 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
1302 : XFS_ILOCK_EXCL));
1304 return error;
1308 * Free a symlink that has blocks associated with it.
1310 STATIC int
1311 xfs_inactive_symlink_rmt(
1312 xfs_inode_t *ip,
1313 xfs_trans_t **tpp)
1315 xfs_buf_t *bp;
1316 int committed;
1317 int done;
1318 int error;
1319 xfs_fsblock_t first_block;
1320 xfs_bmap_free_t free_list;
1321 int i;
1322 xfs_mount_t *mp;
1323 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1324 int nmaps;
1325 xfs_trans_t *ntp;
1326 int size;
1327 xfs_trans_t *tp;
1329 tp = *tpp;
1330 mp = ip->i_mount;
1331 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1333 * We're freeing a symlink that has some
1334 * blocks allocated to it. Free the
1335 * blocks here. We know that we've got
1336 * either 1 or 2 extents and that we can
1337 * free them all in one bunmapi call.
1339 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1340 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1341 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1342 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1343 xfs_trans_cancel(tp, 0);
1344 *tpp = NULL;
1345 return error;
1348 * Lock the inode, fix the size, and join it to the transaction.
1349 * Hold it so in the normal path, we still have it locked for
1350 * the second transaction. In the error paths we need it
1351 * held so the cancel won't rele it, see below.
1353 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1354 size = (int)ip->i_d.di_size;
1355 ip->i_d.di_size = 0;
1356 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1357 xfs_trans_ihold(tp, ip);
1358 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1360 * Find the block(s) so we can inval and unmap them.
1362 done = 0;
1363 XFS_BMAP_INIT(&free_list, &first_block);
1364 nmaps = ARRAY_SIZE(mval);
1365 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1366 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1367 &free_list, NULL)))
1368 goto error0;
1370 * Invalidate the block(s).
1372 for (i = 0; i < nmaps; i++) {
1373 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1374 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1375 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1376 xfs_trans_binval(tp, bp);
1379 * Unmap the dead block(s) to the free_list.
1381 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1382 &first_block, &free_list, NULL, &done)))
1383 goto error1;
1384 ASSERT(done);
1386 * Commit the first transaction. This logs the EFI and the inode.
1388 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
1389 goto error1;
1391 * The transaction must have been committed, since there were
1392 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1393 * The new tp has the extent freeing and EFDs.
1395 ASSERT(committed);
1397 * The first xact was committed, so add the inode to the new one.
1398 * Mark it dirty so it will be logged and moved forward in the log as
1399 * part of every commit.
1401 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1402 xfs_trans_ihold(tp, ip);
1403 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1405 * Get a new, empty transaction to return to our caller.
1407 ntp = xfs_trans_dup(tp);
1409 * Commit the transaction containing extent freeing and EFDs.
1410 * If we get an error on the commit here or on the reserve below,
1411 * we need to unlock the inode since the new transaction doesn't
1412 * have the inode attached.
1414 error = xfs_trans_commit(tp, 0);
1415 tp = ntp;
1416 if (error) {
1417 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1418 goto error0;
1421 * Remove the memory for extent descriptions (just bookkeeping).
1423 if (ip->i_df.if_bytes)
1424 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1425 ASSERT(ip->i_df.if_bytes == 0);
1427 * Put an itruncate log reservation in the new transaction
1428 * for our caller.
1430 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1431 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1432 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1433 goto error0;
1436 * Return with the inode locked but not joined to the transaction.
1438 *tpp = tp;
1439 return 0;
1441 error1:
1442 xfs_bmap_cancel(&free_list);
1443 error0:
1445 * Have to come here with the inode locked and either
1446 * (held and in the transaction) or (not in the transaction).
1447 * If the inode isn't held then cancel would iput it, but
1448 * that's wrong since this is inactive and the vnode ref
1449 * count is 0 already.
1450 * Cancel won't do anything to the inode if held, but it still
1451 * needs to be locked until the cancel is done, if it was
1452 * joined to the transaction.
1454 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1455 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1456 *tpp = NULL;
1457 return error;
1461 STATIC int
1462 xfs_inactive_symlink_local(
1463 xfs_inode_t *ip,
1464 xfs_trans_t **tpp)
1466 int error;
1468 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1470 * We're freeing a symlink which fit into
1471 * the inode. Just free the memory used
1472 * to hold the old symlink.
1474 error = xfs_trans_reserve(*tpp, 0,
1475 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1476 0, XFS_TRANS_PERM_LOG_RES,
1477 XFS_ITRUNCATE_LOG_COUNT);
1479 if (error) {
1480 xfs_trans_cancel(*tpp, 0);
1481 *tpp = NULL;
1482 return error;
1484 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1487 * Zero length symlinks _can_ exist.
1489 if (ip->i_df.if_bytes > 0) {
1490 xfs_idata_realloc(ip,
1491 -(ip->i_df.if_bytes),
1492 XFS_DATA_FORK);
1493 ASSERT(ip->i_df.if_bytes == 0);
1495 return 0;
1498 STATIC int
1499 xfs_inactive_attrs(
1500 xfs_inode_t *ip,
1501 xfs_trans_t **tpp)
1503 xfs_trans_t *tp;
1504 int error;
1505 xfs_mount_t *mp;
1507 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1508 tp = *tpp;
1509 mp = ip->i_mount;
1510 ASSERT(ip->i_d.di_forkoff != 0);
1511 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1512 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1514 error = xfs_attr_inactive(ip);
1515 if (error) {
1516 *tpp = NULL;
1517 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1518 return error; /* goto out */
1521 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1522 error = xfs_trans_reserve(tp, 0,
1523 XFS_IFREE_LOG_RES(mp),
1524 0, XFS_TRANS_PERM_LOG_RES,
1525 XFS_INACTIVE_LOG_COUNT);
1526 if (error) {
1527 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1528 xfs_trans_cancel(tp, 0);
1529 *tpp = NULL;
1530 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1531 return error;
1534 xfs_ilock(ip, XFS_ILOCK_EXCL);
1535 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1536 xfs_trans_ihold(tp, ip);
1537 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1539 ASSERT(ip->i_d.di_anextents == 0);
1541 *tpp = tp;
1542 return 0;
1545 STATIC int
1546 xfs_release(
1547 bhv_desc_t *bdp)
1549 xfs_inode_t *ip;
1550 bhv_vnode_t *vp;
1551 xfs_mount_t *mp;
1552 int error;
1554 vp = BHV_TO_VNODE(bdp);
1555 ip = XFS_BHVTOI(bdp);
1556 mp = ip->i_mount;
1558 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
1559 return 0;
1561 /* If this is a read-only mount, don't do this (would generate I/O) */
1562 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1563 return 0;
1565 if (!XFS_FORCED_SHUTDOWN(mp)) {
1567 * If we are using filestreams, and we have an unlinked
1568 * file that we are processing the last close on, then nothing
1569 * will be able to reopen and write to this file. Purge this
1570 * inode from the filestreams cache so that it doesn't delay
1571 * teardown of the inode.
1573 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1574 xfs_filestream_deassociate(ip);
1577 * If we previously truncated this file and removed old data
1578 * in the process, we want to initiate "early" writeout on
1579 * the last close. This is an attempt to combat the notorious
1580 * NULL files problem which is particularly noticable from a
1581 * truncate down, buffered (re-)write (delalloc), followed by
1582 * a crash. What we are effectively doing here is
1583 * significantly reducing the time window where we'd otherwise
1584 * be exposed to that problem.
1586 if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
1587 bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE);
1590 #ifdef HAVE_REFCACHE
1591 /* If we are in the NFS reference cache then don't do this now */
1592 if (ip->i_refcache)
1593 return 0;
1594 #endif
1596 if (ip->i_d.di_nlink != 0) {
1597 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1598 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1599 ip->i_delayed_blks > 0)) &&
1600 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1601 (!(ip->i_d.di_flags &
1602 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
1603 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1604 if (error)
1605 return error;
1606 /* Update linux inode block count after free above */
1607 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1608 ip->i_d.di_nblocks + ip->i_delayed_blks);
1612 return 0;
1616 * xfs_inactive
1618 * This is called when the vnode reference count for the vnode
1619 * goes to zero. If the file has been unlinked, then it must
1620 * now be truncated. Also, we clear all of the read-ahead state
1621 * kept for the inode here since the file is now closed.
1623 STATIC int
1624 xfs_inactive(
1625 bhv_desc_t *bdp,
1626 cred_t *credp)
1628 xfs_inode_t *ip;
1629 bhv_vnode_t *vp;
1630 xfs_bmap_free_t free_list;
1631 xfs_fsblock_t first_block;
1632 int committed;
1633 xfs_trans_t *tp;
1634 xfs_mount_t *mp;
1635 int error;
1636 int truncate;
1638 vp = BHV_TO_VNODE(bdp);
1639 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1641 ip = XFS_BHVTOI(bdp);
1644 * If the inode is already free, then there can be nothing
1645 * to clean up here.
1647 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1648 ASSERT(ip->i_df.if_real_bytes == 0);
1649 ASSERT(ip->i_df.if_broot_bytes == 0);
1650 return VN_INACTIVE_CACHE;
1654 * Only do a truncate if it's a regular file with
1655 * some actual space in it. It's OK to look at the
1656 * inode's fields without the lock because we're the
1657 * only one with a reference to the inode.
1659 truncate = ((ip->i_d.di_nlink == 0) &&
1660 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1661 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
1662 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1664 mp = ip->i_mount;
1666 if (ip->i_d.di_nlink == 0 &&
1667 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1668 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1671 error = 0;
1673 /* If this is a read-only mount, don't do this (would generate I/O) */
1674 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1675 goto out;
1677 if (ip->i_d.di_nlink != 0) {
1678 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1679 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
1680 ip->i_delayed_blks > 0)) &&
1681 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1682 (!(ip->i_d.di_flags &
1683 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1684 (ip->i_delayed_blks != 0)))) {
1685 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1686 if (error)
1687 return VN_INACTIVE_CACHE;
1688 /* Update linux inode block count after free above */
1689 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1690 ip->i_d.di_nblocks + ip->i_delayed_blks);
1692 goto out;
1695 ASSERT(ip->i_d.di_nlink == 0);
1697 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1698 return VN_INACTIVE_CACHE;
1700 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1701 if (truncate) {
1703 * Do the xfs_itruncate_start() call before
1704 * reserving any log space because itruncate_start
1705 * will call into the buffer cache and we can't
1706 * do that within a transaction.
1708 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1710 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1711 if (error) {
1712 xfs_trans_cancel(tp, 0);
1713 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1714 return VN_INACTIVE_CACHE;
1717 error = xfs_trans_reserve(tp, 0,
1718 XFS_ITRUNCATE_LOG_RES(mp),
1719 0, XFS_TRANS_PERM_LOG_RES,
1720 XFS_ITRUNCATE_LOG_COUNT);
1721 if (error) {
1722 /* Don't call itruncate_cleanup */
1723 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1724 xfs_trans_cancel(tp, 0);
1725 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1726 return VN_INACTIVE_CACHE;
1729 xfs_ilock(ip, XFS_ILOCK_EXCL);
1730 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1731 xfs_trans_ihold(tp, ip);
1734 * normally, we have to run xfs_itruncate_finish sync.
1735 * But if filesystem is wsync and we're in the inactive
1736 * path, then we know that nlink == 0, and that the
1737 * xaction that made nlink == 0 is permanently committed
1738 * since xfs_remove runs as a synchronous transaction.
1740 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1741 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1743 if (error) {
1744 xfs_trans_cancel(tp,
1745 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1746 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1747 return VN_INACTIVE_CACHE;
1749 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1752 * If we get an error while cleaning up a
1753 * symlink we bail out.
1755 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1756 xfs_inactive_symlink_rmt(ip, &tp) :
1757 xfs_inactive_symlink_local(ip, &tp);
1759 if (error) {
1760 ASSERT(tp == NULL);
1761 return VN_INACTIVE_CACHE;
1764 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1765 xfs_trans_ihold(tp, ip);
1766 } else {
1767 error = xfs_trans_reserve(tp, 0,
1768 XFS_IFREE_LOG_RES(mp),
1769 0, XFS_TRANS_PERM_LOG_RES,
1770 XFS_INACTIVE_LOG_COUNT);
1771 if (error) {
1772 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1773 xfs_trans_cancel(tp, 0);
1774 return VN_INACTIVE_CACHE;
1777 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1778 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1779 xfs_trans_ihold(tp, ip);
1783 * If there are attributes associated with the file
1784 * then blow them away now. The code calls a routine
1785 * that recursively deconstructs the attribute fork.
1786 * We need to just commit the current transaction
1787 * because we can't use it for xfs_attr_inactive().
1789 if (ip->i_d.di_anextents > 0) {
1790 error = xfs_inactive_attrs(ip, &tp);
1792 * If we got an error, the transaction is already
1793 * cancelled, and the inode is unlocked. Just get out.
1795 if (error)
1796 return VN_INACTIVE_CACHE;
1797 } else if (ip->i_afp) {
1798 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1802 * Free the inode.
1804 XFS_BMAP_INIT(&free_list, &first_block);
1805 error = xfs_ifree(tp, ip, &free_list);
1806 if (error) {
1808 * If we fail to free the inode, shut down. The cancel
1809 * might do that, we need to make sure. Otherwise the
1810 * inode might be lost for a long time or forever.
1812 if (!XFS_FORCED_SHUTDOWN(mp)) {
1813 cmn_err(CE_NOTE,
1814 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1815 error, mp->m_fsname);
1816 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1818 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1819 } else {
1821 * Credit the quota account(s). The inode is gone.
1823 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1826 * Just ignore errors at this point. There is
1827 * nothing we can do except to try to keep going.
1829 (void) xfs_bmap_finish(&tp, &free_list, &committed);
1830 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1833 * Release the dquots held by inode, if any.
1835 XFS_QM_DQDETACH(mp, ip);
1837 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1839 out:
1840 return VN_INACTIVE_CACHE;
1845 * xfs_lookup
1847 STATIC int
1848 xfs_lookup(
1849 bhv_desc_t *dir_bdp,
1850 bhv_vname_t *dentry,
1851 bhv_vnode_t **vpp,
1852 int flags,
1853 bhv_vnode_t *rdir,
1854 cred_t *credp)
1856 xfs_inode_t *dp, *ip;
1857 xfs_ino_t e_inum;
1858 int error;
1859 uint lock_mode;
1860 bhv_vnode_t *dir_vp;
1862 dir_vp = BHV_TO_VNODE(dir_bdp);
1863 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1865 dp = XFS_BHVTOI(dir_bdp);
1867 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1868 return XFS_ERROR(EIO);
1870 lock_mode = xfs_ilock_map_shared(dp);
1871 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1872 if (!error) {
1873 *vpp = XFS_ITOV(ip);
1874 ITRACE(ip);
1876 xfs_iunlock_map_shared(dp, lock_mode);
1877 return error;
1882 * xfs_create (create a new file).
1884 STATIC int
1885 xfs_create(
1886 bhv_desc_t *dir_bdp,
1887 bhv_vname_t *dentry,
1888 bhv_vattr_t *vap,
1889 bhv_vnode_t **vpp,
1890 cred_t *credp)
1892 char *name = VNAME(dentry);
1893 bhv_vnode_t *dir_vp;
1894 xfs_inode_t *dp, *ip;
1895 bhv_vnode_t *vp = NULL;
1896 xfs_trans_t *tp;
1897 xfs_mount_t *mp;
1898 xfs_dev_t rdev;
1899 int error;
1900 xfs_bmap_free_t free_list;
1901 xfs_fsblock_t first_block;
1902 boolean_t dp_joined_to_trans;
1903 int dm_event_sent = 0;
1904 uint cancel_flags;
1905 int committed;
1906 xfs_prid_t prid;
1907 struct xfs_dquot *udqp, *gdqp;
1908 uint resblks;
1909 int dm_di_mode;
1910 int namelen;
1912 ASSERT(!*vpp);
1913 dir_vp = BHV_TO_VNODE(dir_bdp);
1914 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1916 dp = XFS_BHVTOI(dir_bdp);
1917 mp = dp->i_mount;
1919 dm_di_mode = vap->va_mode;
1920 namelen = VNAMELEN(dentry);
1922 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1923 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1924 dir_vp, DM_RIGHT_NULL, NULL,
1925 DM_RIGHT_NULL, name, NULL,
1926 dm_di_mode, 0, 0);
1928 if (error)
1929 return error;
1930 dm_event_sent = 1;
1933 if (XFS_FORCED_SHUTDOWN(mp))
1934 return XFS_ERROR(EIO);
1936 /* Return through std_return after this point. */
1938 udqp = gdqp = NULL;
1939 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1940 prid = dp->i_d.di_projid;
1941 else if (vap->va_mask & XFS_AT_PROJID)
1942 prid = (xfs_prid_t)vap->va_projid;
1943 else
1944 prid = (xfs_prid_t)dfltprid;
1947 * Make sure that we have allocated dquot(s) on disk.
1949 error = XFS_QM_DQVOPALLOC(mp, dp,
1950 current_fsuid(credp), current_fsgid(credp), prid,
1951 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1952 if (error)
1953 goto std_return;
1955 ip = NULL;
1956 dp_joined_to_trans = B_FALSE;
1958 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1959 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1960 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1962 * Initially assume that the file does not exist and
1963 * reserve the resources for that case. If that is not
1964 * the case we'll drop the one we have and get a more
1965 * appropriate transaction later.
1967 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1968 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1969 if (error == ENOSPC) {
1970 resblks = 0;
1971 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1972 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1974 if (error) {
1975 cancel_flags = 0;
1976 dp = NULL;
1977 goto error_return;
1980 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1982 XFS_BMAP_INIT(&free_list, &first_block);
1984 ASSERT(ip == NULL);
1987 * Reserve disk quota and the inode.
1989 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1990 if (error)
1991 goto error_return;
1993 if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen)))
1994 goto error_return;
1995 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1996 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1997 rdev, credp, prid, resblks > 0,
1998 &ip, &committed);
1999 if (error) {
2000 if (error == ENOSPC)
2001 goto error_return;
2002 goto abort_return;
2004 ITRACE(ip);
2007 * At this point, we've gotten a newly allocated inode.
2008 * It is locked (and joined to the transaction).
2011 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
2014 * Now we join the directory inode to the transaction.
2015 * We do not do it earlier because xfs_dir_ialloc
2016 * might commit the previous transaction (and release
2017 * all the locks).
2020 VN_HOLD(dir_vp);
2021 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2022 dp_joined_to_trans = B_TRUE;
2024 error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino,
2025 &first_block, &free_list, resblks ?
2026 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2027 if (error) {
2028 ASSERT(error != ENOSPC);
2029 goto abort_return;
2031 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2032 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2035 * If this is a synchronous mount, make sure that the
2036 * create transaction goes to disk before returning to
2037 * the user.
2039 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2040 xfs_trans_set_sync(tp);
2043 dp->i_gen++;
2046 * Attach the dquot(s) to the inodes and modify them incore.
2047 * These ids of the inode couldn't have changed since the new
2048 * inode has been locked ever since it was created.
2050 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2053 * xfs_trans_commit normally decrements the vnode ref count
2054 * when it unlocks the inode. Since we want to return the
2055 * vnode to the caller, we bump the vnode ref count now.
2057 IHOLD(ip);
2058 vp = XFS_ITOV(ip);
2060 error = xfs_bmap_finish(&tp, &free_list, &committed);
2061 if (error) {
2062 xfs_bmap_cancel(&free_list);
2063 goto abort_rele;
2066 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2067 if (error) {
2068 IRELE(ip);
2069 tp = NULL;
2070 goto error_return;
2073 XFS_QM_DQRELE(mp, udqp);
2074 XFS_QM_DQRELE(mp, gdqp);
2077 * Propagate the fact that the vnode changed after the
2078 * xfs_inode locks have been released.
2080 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2082 *vpp = vp;
2084 /* Fallthrough to std_return with error = 0 */
2086 std_return:
2087 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2088 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2089 DM_EVENT_POSTCREATE)) {
2090 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2091 dir_vp, DM_RIGHT_NULL,
2092 *vpp ? vp:NULL,
2093 DM_RIGHT_NULL, name, NULL,
2094 dm_di_mode, error, 0);
2096 return error;
2098 abort_return:
2099 cancel_flags |= XFS_TRANS_ABORT;
2100 /* FALLTHROUGH */
2102 error_return:
2103 if (tp != NULL)
2104 xfs_trans_cancel(tp, cancel_flags);
2106 if (!dp_joined_to_trans && (dp != NULL))
2107 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2108 XFS_QM_DQRELE(mp, udqp);
2109 XFS_QM_DQRELE(mp, gdqp);
2111 goto std_return;
2113 abort_rele:
2115 * Wait until after the current transaction is aborted to
2116 * release the inode. This prevents recursive transactions
2117 * and deadlocks from xfs_inactive.
2119 cancel_flags |= XFS_TRANS_ABORT;
2120 xfs_trans_cancel(tp, cancel_flags);
2121 IRELE(ip);
2123 XFS_QM_DQRELE(mp, udqp);
2124 XFS_QM_DQRELE(mp, gdqp);
2126 goto std_return;
2129 #ifdef DEBUG
2131 * Some counters to see if (and how often) we are hitting some deadlock
2132 * prevention code paths.
2135 int xfs_rm_locks;
2136 int xfs_rm_lock_delays;
2137 int xfs_rm_attempts;
2138 #endif
2141 * The following routine will lock the inodes associated with the
2142 * directory and the named entry in the directory. The locks are
2143 * acquired in increasing inode number.
2145 * If the entry is "..", then only the directory is locked. The
2146 * vnode ref count will still include that from the .. entry in
2147 * this case.
2149 * There is a deadlock we need to worry about. If the locked directory is
2150 * in the AIL, it might be blocking up the log. The next inode we lock
2151 * could be already locked by another thread waiting for log space (e.g
2152 * a permanent log reservation with a long running transaction (see
2153 * xfs_itruncate_finish)). To solve this, we must check if the directory
2154 * is in the ail and use lock_nowait. If we can't lock, we need to
2155 * drop the inode lock on the directory and try again. xfs_iunlock will
2156 * potentially push the tail if we were holding up the log.
2158 STATIC int
2159 xfs_lock_dir_and_entry(
2160 xfs_inode_t *dp,
2161 xfs_inode_t *ip) /* inode of entry 'name' */
2163 int attempts;
2164 xfs_ino_t e_inum;
2165 xfs_inode_t *ips[2];
2166 xfs_log_item_t *lp;
2168 #ifdef DEBUG
2169 xfs_rm_locks++;
2170 #endif
2171 attempts = 0;
2173 again:
2174 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2176 e_inum = ip->i_ino;
2178 ITRACE(ip);
2181 * We want to lock in increasing inum. Since we've already
2182 * acquired the lock on the directory, we may need to release
2183 * if if the inum of the entry turns out to be less.
2185 if (e_inum > dp->i_ino) {
2187 * We are already in the right order, so just
2188 * lock on the inode of the entry.
2189 * We need to use nowait if dp is in the AIL.
2192 lp = (xfs_log_item_t *)dp->i_itemp;
2193 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2194 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2195 attempts++;
2196 #ifdef DEBUG
2197 xfs_rm_attempts++;
2198 #endif
2201 * Unlock dp and try again.
2202 * xfs_iunlock will try to push the tail
2203 * if the inode is in the AIL.
2206 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2208 if ((attempts % 5) == 0) {
2209 delay(1); /* Don't just spin the CPU */
2210 #ifdef DEBUG
2211 xfs_rm_lock_delays++;
2212 #endif
2214 goto again;
2216 } else {
2217 xfs_ilock(ip, XFS_ILOCK_EXCL);
2219 } else if (e_inum < dp->i_ino) {
2220 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2222 ips[0] = ip;
2223 ips[1] = dp;
2224 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2226 /* else e_inum == dp->i_ino */
2227 /* This can happen if we're asked to lock /x/..
2228 * the entry is "..", which is also the parent directory.
2231 return 0;
2234 #ifdef DEBUG
2235 int xfs_locked_n;
2236 int xfs_small_retries;
2237 int xfs_middle_retries;
2238 int xfs_lots_retries;
2239 int xfs_lock_delays;
2240 #endif
2243 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2244 * a different value
2246 static inline int
2247 xfs_lock_inumorder(int lock_mode, int subclass)
2249 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
2250 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
2251 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
2252 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
2254 return lock_mode;
2258 * The following routine will lock n inodes in exclusive mode.
2259 * We assume the caller calls us with the inodes in i_ino order.
2261 * We need to detect deadlock where an inode that we lock
2262 * is in the AIL and we start waiting for another inode that is locked
2263 * by a thread in a long running transaction (such as truncate). This can
2264 * result in deadlock since the long running trans might need to wait
2265 * for the inode we just locked in order to push the tail and free space
2266 * in the log.
2268 void
2269 xfs_lock_inodes(
2270 xfs_inode_t **ips,
2271 int inodes,
2272 int first_locked,
2273 uint lock_mode)
2275 int attempts = 0, i, j, try_lock;
2276 xfs_log_item_t *lp;
2278 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2280 if (first_locked) {
2281 try_lock = 1;
2282 i = 1;
2283 } else {
2284 try_lock = 0;
2285 i = 0;
2288 again:
2289 for (; i < inodes; i++) {
2290 ASSERT(ips[i]);
2292 if (i && (ips[i] == ips[i-1])) /* Already locked */
2293 continue;
2296 * If try_lock is not set yet, make sure all locked inodes
2297 * are not in the AIL.
2298 * If any are, set try_lock to be used later.
2301 if (!try_lock) {
2302 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2303 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2304 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2305 try_lock++;
2311 * If any of the previous locks we have locked is in the AIL,
2312 * we must TRY to get the second and subsequent locks. If
2313 * we can't get any, we must release all we have
2314 * and try again.
2317 if (try_lock) {
2318 /* try_lock must be 0 if i is 0. */
2320 * try_lock means we have an inode locked
2321 * that is in the AIL.
2323 ASSERT(i != 0);
2324 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
2325 attempts++;
2328 * Unlock all previous guys and try again.
2329 * xfs_iunlock will try to push the tail
2330 * if the inode is in the AIL.
2333 for(j = i - 1; j >= 0; j--) {
2336 * Check to see if we've already
2337 * unlocked this one.
2338 * Not the first one going back,
2339 * and the inode ptr is the same.
2341 if ((j != (i - 1)) && ips[j] ==
2342 ips[j+1])
2343 continue;
2345 xfs_iunlock(ips[j], lock_mode);
2348 if ((attempts % 5) == 0) {
2349 delay(1); /* Don't just spin the CPU */
2350 #ifdef DEBUG
2351 xfs_lock_delays++;
2352 #endif
2354 i = 0;
2355 try_lock = 0;
2356 goto again;
2358 } else {
2359 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
2363 #ifdef DEBUG
2364 if (attempts) {
2365 if (attempts < 5) xfs_small_retries++;
2366 else if (attempts < 100) xfs_middle_retries++;
2367 else xfs_lots_retries++;
2368 } else {
2369 xfs_locked_n++;
2371 #endif
2374 #ifdef DEBUG
2375 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2376 int remove_which_error_return = 0;
2377 #else /* ! DEBUG */
2378 #define REMOVE_DEBUG_TRACE(x)
2379 #endif /* ! DEBUG */
2383 * xfs_remove
2386 STATIC int
2387 xfs_remove(
2388 bhv_desc_t *dir_bdp,
2389 bhv_vname_t *dentry,
2390 cred_t *credp)
2392 bhv_vnode_t *dir_vp;
2393 char *name = VNAME(dentry);
2394 xfs_inode_t *dp, *ip;
2395 xfs_trans_t *tp = NULL;
2396 xfs_mount_t *mp;
2397 int error = 0;
2398 xfs_bmap_free_t free_list;
2399 xfs_fsblock_t first_block;
2400 int cancel_flags;
2401 int committed;
2402 int dm_di_mode = 0;
2403 int link_zero;
2404 uint resblks;
2405 int namelen;
2407 dir_vp = BHV_TO_VNODE(dir_bdp);
2408 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2410 dp = XFS_BHVTOI(dir_bdp);
2411 mp = dp->i_mount;
2413 if (XFS_FORCED_SHUTDOWN(mp))
2414 return XFS_ERROR(EIO);
2416 namelen = VNAMELEN(dentry);
2418 if (!xfs_get_dir_entry(dentry, &ip)) {
2419 dm_di_mode = ip->i_d.di_mode;
2420 IRELE(ip);
2423 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2424 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2425 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2426 name, NULL, dm_di_mode, 0, 0);
2427 if (error)
2428 return error;
2431 /* From this point on, return through std_return */
2432 ip = NULL;
2435 * We need to get a reference to ip before we get our log
2436 * reservation. The reason for this is that we cannot call
2437 * xfs_iget for an inode for which we do not have a reference
2438 * once we've acquired a log reservation. This is because the
2439 * inode we are trying to get might be in xfs_inactive going
2440 * for a log reservation. Since we'll have to wait for the
2441 * inactive code to complete before returning from xfs_iget,
2442 * we need to make sure that we don't have log space reserved
2443 * when we call xfs_iget. Instead we get an unlocked reference
2444 * to the inode before getting our log reservation.
2446 error = xfs_get_dir_entry(dentry, &ip);
2447 if (error) {
2448 REMOVE_DEBUG_TRACE(__LINE__);
2449 goto std_return;
2452 dm_di_mode = ip->i_d.di_mode;
2454 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2456 ITRACE(ip);
2458 error = XFS_QM_DQATTACH(mp, dp, 0);
2459 if (!error && dp != ip)
2460 error = XFS_QM_DQATTACH(mp, ip, 0);
2461 if (error) {
2462 REMOVE_DEBUG_TRACE(__LINE__);
2463 IRELE(ip);
2464 goto std_return;
2467 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2468 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2470 * We try to get the real space reservation first,
2471 * allowing for directory btree deletion(s) implying
2472 * possible bmap insert(s). If we can't get the space
2473 * reservation then we use 0 instead, and avoid the bmap
2474 * btree insert(s) in the directory code by, if the bmap
2475 * insert tries to happen, instead trimming the LAST
2476 * block from the directory.
2478 resblks = XFS_REMOVE_SPACE_RES(mp);
2479 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2480 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2481 if (error == ENOSPC) {
2482 resblks = 0;
2483 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2484 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2486 if (error) {
2487 ASSERT(error != ENOSPC);
2488 REMOVE_DEBUG_TRACE(__LINE__);
2489 xfs_trans_cancel(tp, 0);
2490 IRELE(ip);
2491 return error;
2494 error = xfs_lock_dir_and_entry(dp, ip);
2495 if (error) {
2496 REMOVE_DEBUG_TRACE(__LINE__);
2497 xfs_trans_cancel(tp, cancel_flags);
2498 IRELE(ip);
2499 goto std_return;
2503 * At this point, we've gotten both the directory and the entry
2504 * inodes locked.
2506 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2507 if (dp != ip) {
2509 * Increment vnode ref count only in this case since
2510 * there's an extra vnode reference in the case where
2511 * dp == ip.
2513 IHOLD(dp);
2514 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2518 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2520 XFS_BMAP_INIT(&free_list, &first_block);
2521 error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino,
2522 &first_block, &free_list, 0);
2523 if (error) {
2524 ASSERT(error != ENOENT);
2525 REMOVE_DEBUG_TRACE(__LINE__);
2526 goto error1;
2528 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2530 dp->i_gen++;
2531 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2533 error = xfs_droplink(tp, ip);
2534 if (error) {
2535 REMOVE_DEBUG_TRACE(__LINE__);
2536 goto error1;
2539 /* Determine if this is the last link while
2540 * we are in the transaction.
2542 link_zero = (ip)->i_d.di_nlink==0;
2545 * Take an extra ref on the inode so that it doesn't
2546 * go to xfs_inactive() from within the commit.
2548 IHOLD(ip);
2551 * If this is a synchronous mount, make sure that the
2552 * remove transaction goes to disk before returning to
2553 * the user.
2555 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2556 xfs_trans_set_sync(tp);
2559 error = xfs_bmap_finish(&tp, &free_list, &committed);
2560 if (error) {
2561 REMOVE_DEBUG_TRACE(__LINE__);
2562 goto error_rele;
2565 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2566 if (error) {
2567 IRELE(ip);
2568 goto std_return;
2572 * Before we drop our extra reference to the inode, purge it
2573 * from the refcache if it is there. By waiting until afterwards
2574 * to do the IRELE, we ensure that we won't go inactive in the
2575 * xfs_refcache_purge_ip routine (although that would be OK).
2577 xfs_refcache_purge_ip(ip);
2580 * If we are using filestreams, kill the stream association.
2581 * If the file is still open it may get a new one but that
2582 * will get killed on last close in xfs_close() so we don't
2583 * have to worry about that.
2585 if (link_zero && xfs_inode_is_filestream(ip))
2586 xfs_filestream_deassociate(ip);
2588 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2591 * Let interposed file systems know about removed links.
2593 bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero);
2595 IRELE(ip);
2597 /* Fall through to std_return with error = 0 */
2598 std_return:
2599 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2600 DM_EVENT_POSTREMOVE)) {
2601 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2602 dir_vp, DM_RIGHT_NULL,
2603 NULL, DM_RIGHT_NULL,
2604 name, NULL, dm_di_mode, error, 0);
2606 return error;
2608 error1:
2609 xfs_bmap_cancel(&free_list);
2610 cancel_flags |= XFS_TRANS_ABORT;
2611 xfs_trans_cancel(tp, cancel_flags);
2612 goto std_return;
2614 error_rele:
2616 * In this case make sure to not release the inode until after
2617 * the current transaction is aborted. Releasing it beforehand
2618 * can cause us to go to xfs_inactive and start a recursive
2619 * transaction which can easily deadlock with the current one.
2621 xfs_bmap_cancel(&free_list);
2622 cancel_flags |= XFS_TRANS_ABORT;
2623 xfs_trans_cancel(tp, cancel_flags);
2626 * Before we drop our extra reference to the inode, purge it
2627 * from the refcache if it is there. By waiting until afterwards
2628 * to do the IRELE, we ensure that we won't go inactive in the
2629 * xfs_refcache_purge_ip routine (although that would be OK).
2631 xfs_refcache_purge_ip(ip);
2633 IRELE(ip);
2635 goto std_return;
2640 * xfs_link
2643 STATIC int
2644 xfs_link(
2645 bhv_desc_t *target_dir_bdp,
2646 bhv_vnode_t *src_vp,
2647 bhv_vname_t *dentry,
2648 cred_t *credp)
2650 xfs_inode_t *tdp, *sip;
2651 xfs_trans_t *tp;
2652 xfs_mount_t *mp;
2653 xfs_inode_t *ips[2];
2654 int error;
2655 xfs_bmap_free_t free_list;
2656 xfs_fsblock_t first_block;
2657 int cancel_flags;
2658 int committed;
2659 bhv_vnode_t *target_dir_vp;
2660 int resblks;
2661 char *target_name = VNAME(dentry);
2662 int target_namelen;
2664 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2665 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2666 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2668 target_namelen = VNAMELEN(dentry);
2669 ASSERT(!VN_ISDIR(src_vp));
2671 sip = xfs_vtoi(src_vp);
2672 tdp = XFS_BHVTOI(target_dir_bdp);
2673 mp = tdp->i_mount;
2674 if (XFS_FORCED_SHUTDOWN(mp))
2675 return XFS_ERROR(EIO);
2677 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2678 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2679 target_dir_vp, DM_RIGHT_NULL,
2680 src_vp, DM_RIGHT_NULL,
2681 target_name, NULL, 0, 0, 0);
2682 if (error)
2683 return error;
2686 /* Return through std_return after this point. */
2688 error = XFS_QM_DQATTACH(mp, sip, 0);
2689 if (!error && sip != tdp)
2690 error = XFS_QM_DQATTACH(mp, tdp, 0);
2691 if (error)
2692 goto std_return;
2694 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2695 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2696 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2697 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2698 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2699 if (error == ENOSPC) {
2700 resblks = 0;
2701 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2702 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2704 if (error) {
2705 cancel_flags = 0;
2706 goto error_return;
2709 if (sip->i_ino < tdp->i_ino) {
2710 ips[0] = sip;
2711 ips[1] = tdp;
2712 } else {
2713 ips[0] = tdp;
2714 ips[1] = sip;
2717 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2720 * Increment vnode ref counts since xfs_trans_commit &
2721 * xfs_trans_cancel will both unlock the inodes and
2722 * decrement the associated ref counts.
2724 VN_HOLD(src_vp);
2725 VN_HOLD(target_dir_vp);
2726 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2727 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2730 * If the source has too many links, we can't make any more to it.
2732 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2733 error = XFS_ERROR(EMLINK);
2734 goto error_return;
2738 * If we are using project inheritance, we only allow hard link
2739 * creation in our tree when the project IDs are the same; else
2740 * the tree quota mechanism could be circumvented.
2742 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2743 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2744 error = XFS_ERROR(EXDEV);
2745 goto error_return;
2748 if (resblks == 0 &&
2749 (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen)))
2750 goto error_return;
2752 XFS_BMAP_INIT(&free_list, &first_block);
2754 error = xfs_dir_createname(tp, tdp, target_name, target_namelen,
2755 sip->i_ino, &first_block, &free_list,
2756 resblks);
2757 if (error)
2758 goto abort_return;
2759 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2760 tdp->i_gen++;
2761 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2763 error = xfs_bumplink(tp, sip);
2764 if (error)
2765 goto abort_return;
2768 * If this is a synchronous mount, make sure that the
2769 * link transaction goes to disk before returning to
2770 * the user.
2772 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2773 xfs_trans_set_sync(tp);
2776 error = xfs_bmap_finish (&tp, &free_list, &committed);
2777 if (error) {
2778 xfs_bmap_cancel(&free_list);
2779 goto abort_return;
2782 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2783 if (error)
2784 goto std_return;
2786 /* Fall through to std_return with error = 0. */
2787 std_return:
2788 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2789 DM_EVENT_POSTLINK)) {
2790 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2791 target_dir_vp, DM_RIGHT_NULL,
2792 src_vp, DM_RIGHT_NULL,
2793 target_name, NULL, 0, error, 0);
2795 return error;
2797 abort_return:
2798 cancel_flags |= XFS_TRANS_ABORT;
2799 /* FALLTHROUGH */
2801 error_return:
2802 xfs_trans_cancel(tp, cancel_flags);
2803 goto std_return;
2808 * xfs_mkdir
2811 STATIC int
2812 xfs_mkdir(
2813 bhv_desc_t *dir_bdp,
2814 bhv_vname_t *dentry,
2815 bhv_vattr_t *vap,
2816 bhv_vnode_t **vpp,
2817 cred_t *credp)
2819 char *dir_name = VNAME(dentry);
2820 xfs_inode_t *dp;
2821 xfs_inode_t *cdp; /* inode of created dir */
2822 bhv_vnode_t *cvp; /* vnode of created dir */
2823 xfs_trans_t *tp;
2824 xfs_mount_t *mp;
2825 int cancel_flags;
2826 int error;
2827 int committed;
2828 xfs_bmap_free_t free_list;
2829 xfs_fsblock_t first_block;
2830 bhv_vnode_t *dir_vp;
2831 boolean_t dp_joined_to_trans;
2832 boolean_t created = B_FALSE;
2833 int dm_event_sent = 0;
2834 xfs_prid_t prid;
2835 struct xfs_dquot *udqp, *gdqp;
2836 uint resblks;
2837 int dm_di_mode;
2838 int dir_namelen;
2840 dir_vp = BHV_TO_VNODE(dir_bdp);
2841 dp = XFS_BHVTOI(dir_bdp);
2842 mp = dp->i_mount;
2844 if (XFS_FORCED_SHUTDOWN(mp))
2845 return XFS_ERROR(EIO);
2847 dir_namelen = VNAMELEN(dentry);
2849 tp = NULL;
2850 dp_joined_to_trans = B_FALSE;
2851 dm_di_mode = vap->va_mode;
2853 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2854 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2855 dir_vp, DM_RIGHT_NULL, NULL,
2856 DM_RIGHT_NULL, dir_name, NULL,
2857 dm_di_mode, 0, 0);
2858 if (error)
2859 return error;
2860 dm_event_sent = 1;
2863 /* Return through std_return after this point. */
2865 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2867 mp = dp->i_mount;
2868 udqp = gdqp = NULL;
2869 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2870 prid = dp->i_d.di_projid;
2871 else if (vap->va_mask & XFS_AT_PROJID)
2872 prid = (xfs_prid_t)vap->va_projid;
2873 else
2874 prid = (xfs_prid_t)dfltprid;
2877 * Make sure that we have allocated dquot(s) on disk.
2879 error = XFS_QM_DQVOPALLOC(mp, dp,
2880 current_fsuid(credp), current_fsgid(credp), prid,
2881 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2882 if (error)
2883 goto std_return;
2885 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2886 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2887 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2888 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2889 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2890 if (error == ENOSPC) {
2891 resblks = 0;
2892 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2893 XFS_TRANS_PERM_LOG_RES,
2894 XFS_MKDIR_LOG_COUNT);
2896 if (error) {
2897 cancel_flags = 0;
2898 dp = NULL;
2899 goto error_return;
2902 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
2905 * Check for directory link count overflow.
2907 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2908 error = XFS_ERROR(EMLINK);
2909 goto error_return;
2913 * Reserve disk quota and the inode.
2915 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2916 if (error)
2917 goto error_return;
2919 if (resblks == 0 &&
2920 (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen)))
2921 goto error_return;
2923 * create the directory inode.
2925 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2926 0, credp, prid, resblks > 0,
2927 &cdp, NULL);
2928 if (error) {
2929 if (error == ENOSPC)
2930 goto error_return;
2931 goto abort_return;
2933 ITRACE(cdp);
2936 * Now we add the directory inode to the transaction.
2937 * We waited until now since xfs_dir_ialloc might start
2938 * a new transaction. Had we joined the transaction
2939 * earlier, the locks might have gotten released.
2941 VN_HOLD(dir_vp);
2942 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2943 dp_joined_to_trans = B_TRUE;
2945 XFS_BMAP_INIT(&free_list, &first_block);
2947 error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino,
2948 &first_block, &free_list, resblks ?
2949 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2950 if (error) {
2951 ASSERT(error != ENOSPC);
2952 goto error1;
2954 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2957 * Bump the in memory version number of the parent directory
2958 * so that other processes accessing it will recognize that
2959 * the directory has changed.
2961 dp->i_gen++;
2963 error = xfs_dir_init(tp, cdp, dp);
2964 if (error)
2965 goto error2;
2967 cdp->i_gen = 1;
2968 error = xfs_bumplink(tp, dp);
2969 if (error)
2970 goto error2;
2972 cvp = XFS_ITOV(cdp);
2974 created = B_TRUE;
2976 *vpp = cvp;
2977 IHOLD(cdp);
2980 * Attach the dquots to the new inode and modify the icount incore.
2982 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2985 * If this is a synchronous mount, make sure that the
2986 * mkdir transaction goes to disk before returning to
2987 * the user.
2989 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2990 xfs_trans_set_sync(tp);
2993 error = xfs_bmap_finish(&tp, &free_list, &committed);
2994 if (error) {
2995 IRELE(cdp);
2996 goto error2;
2999 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3000 XFS_QM_DQRELE(mp, udqp);
3001 XFS_QM_DQRELE(mp, gdqp);
3002 if (error) {
3003 IRELE(cdp);
3006 /* Fall through to std_return with error = 0 or errno from
3007 * xfs_trans_commit. */
3009 std_return:
3010 if ( (created || (error != 0 && dm_event_sent != 0)) &&
3011 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3012 DM_EVENT_POSTCREATE)) {
3013 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
3014 dir_vp, DM_RIGHT_NULL,
3015 created ? XFS_ITOV(cdp):NULL,
3016 DM_RIGHT_NULL,
3017 dir_name, NULL,
3018 dm_di_mode, error, 0);
3020 return error;
3022 error2:
3023 error1:
3024 xfs_bmap_cancel(&free_list);
3025 abort_return:
3026 cancel_flags |= XFS_TRANS_ABORT;
3027 error_return:
3028 xfs_trans_cancel(tp, cancel_flags);
3029 XFS_QM_DQRELE(mp, udqp);
3030 XFS_QM_DQRELE(mp, gdqp);
3032 if (!dp_joined_to_trans && (dp != NULL)) {
3033 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3036 goto std_return;
3041 * xfs_rmdir
3044 STATIC int
3045 xfs_rmdir(
3046 bhv_desc_t *dir_bdp,
3047 bhv_vname_t *dentry,
3048 cred_t *credp)
3050 char *name = VNAME(dentry);
3051 xfs_inode_t *dp;
3052 xfs_inode_t *cdp; /* child directory */
3053 xfs_trans_t *tp;
3054 xfs_mount_t *mp;
3055 int error;
3056 xfs_bmap_free_t free_list;
3057 xfs_fsblock_t first_block;
3058 int cancel_flags;
3059 int committed;
3060 bhv_vnode_t *dir_vp;
3061 int dm_di_mode = S_IFDIR;
3062 int last_cdp_link;
3063 int namelen;
3064 uint resblks;
3066 dir_vp = BHV_TO_VNODE(dir_bdp);
3067 dp = XFS_BHVTOI(dir_bdp);
3068 mp = dp->i_mount;
3070 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3072 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3073 return XFS_ERROR(EIO);
3074 namelen = VNAMELEN(dentry);
3076 if (!xfs_get_dir_entry(dentry, &cdp)) {
3077 dm_di_mode = cdp->i_d.di_mode;
3078 IRELE(cdp);
3081 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3082 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3083 dir_vp, DM_RIGHT_NULL,
3084 NULL, DM_RIGHT_NULL,
3085 name, NULL, dm_di_mode, 0, 0);
3086 if (error)
3087 return XFS_ERROR(error);
3090 /* Return through std_return after this point. */
3092 cdp = NULL;
3095 * We need to get a reference to cdp before we get our log
3096 * reservation. The reason for this is that we cannot call
3097 * xfs_iget for an inode for which we do not have a reference
3098 * once we've acquired a log reservation. This is because the
3099 * inode we are trying to get might be in xfs_inactive going
3100 * for a log reservation. Since we'll have to wait for the
3101 * inactive code to complete before returning from xfs_iget,
3102 * we need to make sure that we don't have log space reserved
3103 * when we call xfs_iget. Instead we get an unlocked reference
3104 * to the inode before getting our log reservation.
3106 error = xfs_get_dir_entry(dentry, &cdp);
3107 if (error) {
3108 REMOVE_DEBUG_TRACE(__LINE__);
3109 goto std_return;
3111 mp = dp->i_mount;
3112 dm_di_mode = cdp->i_d.di_mode;
3115 * Get the dquots for the inodes.
3117 error = XFS_QM_DQATTACH(mp, dp, 0);
3118 if (!error && dp != cdp)
3119 error = XFS_QM_DQATTACH(mp, cdp, 0);
3120 if (error) {
3121 IRELE(cdp);
3122 REMOVE_DEBUG_TRACE(__LINE__);
3123 goto std_return;
3126 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3127 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3129 * We try to get the real space reservation first,
3130 * allowing for directory btree deletion(s) implying
3131 * possible bmap insert(s). If we can't get the space
3132 * reservation then we use 0 instead, and avoid the bmap
3133 * btree insert(s) in the directory code by, if the bmap
3134 * insert tries to happen, instead trimming the LAST
3135 * block from the directory.
3137 resblks = XFS_REMOVE_SPACE_RES(mp);
3138 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3139 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3140 if (error == ENOSPC) {
3141 resblks = 0;
3142 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3143 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3145 if (error) {
3146 ASSERT(error != ENOSPC);
3147 cancel_flags = 0;
3148 IRELE(cdp);
3149 goto error_return;
3151 XFS_BMAP_INIT(&free_list, &first_block);
3154 * Now lock the child directory inode and the parent directory
3155 * inode in the proper order. This will take care of validating
3156 * that the directory entry for the child directory inode has
3157 * not changed while we were obtaining a log reservation.
3159 error = xfs_lock_dir_and_entry(dp, cdp);
3160 if (error) {
3161 xfs_trans_cancel(tp, cancel_flags);
3162 IRELE(cdp);
3163 goto std_return;
3166 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3167 if (dp != cdp) {
3169 * Only increment the parent directory vnode count if
3170 * we didn't bump it in looking up cdp. The only time
3171 * we don't bump it is when we're looking up ".".
3173 VN_HOLD(dir_vp);
3176 ITRACE(cdp);
3177 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3179 ASSERT(cdp->i_d.di_nlink >= 2);
3180 if (cdp->i_d.di_nlink != 2) {
3181 error = XFS_ERROR(ENOTEMPTY);
3182 goto error_return;
3184 if (!xfs_dir_isempty(cdp)) {
3185 error = XFS_ERROR(ENOTEMPTY);
3186 goto error_return;
3189 error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino,
3190 &first_block, &free_list, resblks);
3191 if (error)
3192 goto error1;
3194 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3197 * Bump the in memory generation count on the parent
3198 * directory so that other can know that it has changed.
3200 dp->i_gen++;
3203 * Drop the link from cdp's "..".
3205 error = xfs_droplink(tp, dp);
3206 if (error) {
3207 goto error1;
3211 * Drop the link from dp to cdp.
3213 error = xfs_droplink(tp, cdp);
3214 if (error) {
3215 goto error1;
3219 * Drop the "." link from cdp to self.
3221 error = xfs_droplink(tp, cdp);
3222 if (error) {
3223 goto error1;
3226 /* Determine these before committing transaction */
3227 last_cdp_link = (cdp)->i_d.di_nlink==0;
3230 * Take an extra ref on the child vnode so that it
3231 * does not go to xfs_inactive() from within the commit.
3233 IHOLD(cdp);
3236 * If this is a synchronous mount, make sure that the
3237 * rmdir transaction goes to disk before returning to
3238 * the user.
3240 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3241 xfs_trans_set_sync(tp);
3244 error = xfs_bmap_finish (&tp, &free_list, &committed);
3245 if (error) {
3246 xfs_bmap_cancel(&free_list);
3247 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3248 XFS_TRANS_ABORT));
3249 IRELE(cdp);
3250 goto std_return;
3253 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3254 if (error) {
3255 IRELE(cdp);
3256 goto std_return;
3261 * Let interposed file systems know about removed links.
3263 bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3265 IRELE(cdp);
3267 /* Fall through to std_return with error = 0 or the errno
3268 * from xfs_trans_commit. */
3269 std_return:
3270 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3271 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3272 dir_vp, DM_RIGHT_NULL,
3273 NULL, DM_RIGHT_NULL,
3274 name, NULL, dm_di_mode,
3275 error, 0);
3277 return error;
3279 error1:
3280 xfs_bmap_cancel(&free_list);
3281 cancel_flags |= XFS_TRANS_ABORT;
3282 /* FALLTHROUGH */
3284 error_return:
3285 xfs_trans_cancel(tp, cancel_flags);
3286 goto std_return;
3291 * Read dp's entries starting at uiop->uio_offset and translate them into
3292 * bufsize bytes worth of struct dirents starting at bufbase.
3294 STATIC int
3295 xfs_readdir(
3296 bhv_desc_t *dir_bdp,
3297 uio_t *uiop,
3298 cred_t *credp,
3299 int *eofp)
3301 xfs_inode_t *dp;
3302 xfs_trans_t *tp = NULL;
3303 int error = 0;
3304 uint lock_mode;
3306 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3307 (inst_t *)__return_address);
3308 dp = XFS_BHVTOI(dir_bdp);
3310 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
3311 return XFS_ERROR(EIO);
3313 lock_mode = xfs_ilock_map_shared(dp);
3314 error = xfs_dir_getdents(tp, dp, uiop, eofp);
3315 xfs_iunlock_map_shared(dp, lock_mode);
3316 return error;
3320 STATIC int
3321 xfs_symlink(
3322 bhv_desc_t *dir_bdp,
3323 bhv_vname_t *dentry,
3324 bhv_vattr_t *vap,
3325 char *target_path,
3326 bhv_vnode_t **vpp,
3327 cred_t *credp)
3329 xfs_trans_t *tp;
3330 xfs_mount_t *mp;
3331 xfs_inode_t *dp;
3332 xfs_inode_t *ip;
3333 int error;
3334 int pathlen;
3335 xfs_bmap_free_t free_list;
3336 xfs_fsblock_t first_block;
3337 boolean_t dp_joined_to_trans;
3338 bhv_vnode_t *dir_vp;
3339 uint cancel_flags;
3340 int committed;
3341 xfs_fileoff_t first_fsb;
3342 xfs_filblks_t fs_blocks;
3343 int nmaps;
3344 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3345 xfs_daddr_t d;
3346 char *cur_chunk;
3347 int byte_cnt;
3348 int n;
3349 xfs_buf_t *bp;
3350 xfs_prid_t prid;
3351 struct xfs_dquot *udqp, *gdqp;
3352 uint resblks;
3353 char *link_name = VNAME(dentry);
3354 int link_namelen;
3356 *vpp = NULL;
3357 dir_vp = BHV_TO_VNODE(dir_bdp);
3358 dp = XFS_BHVTOI(dir_bdp);
3359 dp_joined_to_trans = B_FALSE;
3360 error = 0;
3361 ip = NULL;
3362 tp = NULL;
3364 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3366 mp = dp->i_mount;
3368 if (XFS_FORCED_SHUTDOWN(mp))
3369 return XFS_ERROR(EIO);
3371 link_namelen = VNAMELEN(dentry);
3374 * Check component lengths of the target path name.
3376 pathlen = strlen(target_path);
3377 if (pathlen >= MAXPATHLEN) /* total string too long */
3378 return XFS_ERROR(ENAMETOOLONG);
3379 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3380 int len, total;
3381 char *path;
3383 for (total = 0, path = target_path; total < pathlen;) {
3385 * Skip any slashes.
3387 while(*path == '/') {
3388 total++;
3389 path++;
3393 * Count up to the next slash or end of path.
3394 * Error out if the component is bigger than MAXNAMELEN.
3396 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3397 if (++len >= MAXNAMELEN) {
3398 error = ENAMETOOLONG;
3399 return error;
3405 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3406 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3407 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3408 link_name, target_path, 0, 0, 0);
3409 if (error)
3410 return error;
3413 /* Return through std_return after this point. */
3415 udqp = gdqp = NULL;
3416 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3417 prid = dp->i_d.di_projid;
3418 else if (vap->va_mask & XFS_AT_PROJID)
3419 prid = (xfs_prid_t)vap->va_projid;
3420 else
3421 prid = (xfs_prid_t)dfltprid;
3424 * Make sure that we have allocated dquot(s) on disk.
3426 error = XFS_QM_DQVOPALLOC(mp, dp,
3427 current_fsuid(credp), current_fsgid(credp), prid,
3428 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3429 if (error)
3430 goto std_return;
3432 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3433 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3435 * The symlink will fit into the inode data fork?
3436 * There can't be any attributes so we get the whole variable part.
3438 if (pathlen <= XFS_LITINO(mp))
3439 fs_blocks = 0;
3440 else
3441 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3442 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3443 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3444 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3445 if (error == ENOSPC && fs_blocks == 0) {
3446 resblks = 0;
3447 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3448 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3450 if (error) {
3451 cancel_flags = 0;
3452 dp = NULL;
3453 goto error_return;
3456 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
3459 * Check whether the directory allows new symlinks or not.
3461 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3462 error = XFS_ERROR(EPERM);
3463 goto error_return;
3467 * Reserve disk quota : blocks and inode.
3469 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3470 if (error)
3471 goto error_return;
3474 * Check for ability to enter directory entry, if no space reserved.
3476 if (resblks == 0 &&
3477 (error = xfs_dir_canenter(tp, dp, link_name, link_namelen)))
3478 goto error_return;
3480 * Initialize the bmap freelist prior to calling either
3481 * bmapi or the directory create code.
3483 XFS_BMAP_INIT(&free_list, &first_block);
3486 * Allocate an inode for the symlink.
3488 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3489 1, 0, credp, prid, resblks > 0, &ip, NULL);
3490 if (error) {
3491 if (error == ENOSPC)
3492 goto error_return;
3493 goto error1;
3495 ITRACE(ip);
3497 VN_HOLD(dir_vp);
3498 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3499 dp_joined_to_trans = B_TRUE;
3502 * Also attach the dquot(s) to it, if applicable.
3504 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3506 if (resblks)
3507 resblks -= XFS_IALLOC_SPACE_RES(mp);
3509 * If the symlink will fit into the inode, write it inline.
3511 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3512 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3513 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3514 ip->i_d.di_size = pathlen;
3517 * The inode was initially created in extent format.
3519 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3520 ip->i_df.if_flags |= XFS_IFINLINE;
3522 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3523 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3525 } else {
3526 first_fsb = 0;
3527 nmaps = SYMLINK_MAPS;
3529 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3530 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3531 &first_block, resblks, mval, &nmaps,
3532 &free_list, NULL);
3533 if (error) {
3534 goto error1;
3537 if (resblks)
3538 resblks -= fs_blocks;
3539 ip->i_d.di_size = pathlen;
3540 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3542 cur_chunk = target_path;
3543 for (n = 0; n < nmaps; n++) {
3544 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3545 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3546 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3547 BTOBB(byte_cnt), 0);
3548 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3549 if (pathlen < byte_cnt) {
3550 byte_cnt = pathlen;
3552 pathlen -= byte_cnt;
3554 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3555 cur_chunk += byte_cnt;
3557 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3562 * Create the directory entry for the symlink.
3564 error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino,
3565 &first_block, &free_list, resblks);
3566 if (error)
3567 goto error1;
3568 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3569 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3572 * Bump the in memory version number of the parent directory
3573 * so that other processes accessing it will recognize that
3574 * the directory has changed.
3576 dp->i_gen++;
3579 * If this is a synchronous mount, make sure that the
3580 * symlink transaction goes to disk before returning to
3581 * the user.
3583 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3584 xfs_trans_set_sync(tp);
3588 * xfs_trans_commit normally decrements the vnode ref count
3589 * when it unlocks the inode. Since we want to return the
3590 * vnode to the caller, we bump the vnode ref count now.
3592 IHOLD(ip);
3594 error = xfs_bmap_finish(&tp, &free_list, &committed);
3595 if (error) {
3596 goto error2;
3598 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3599 XFS_QM_DQRELE(mp, udqp);
3600 XFS_QM_DQRELE(mp, gdqp);
3602 /* Fall through to std_return with error = 0 or errno from
3603 * xfs_trans_commit */
3604 std_return:
3605 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3606 DM_EVENT_POSTSYMLINK)) {
3607 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3608 dir_vp, DM_RIGHT_NULL,
3609 error ? NULL : XFS_ITOV(ip),
3610 DM_RIGHT_NULL, link_name, target_path,
3611 0, error, 0);
3614 if (!error) {
3615 bhv_vnode_t *vp;
3617 ASSERT(ip);
3618 vp = XFS_ITOV(ip);
3619 *vpp = vp;
3621 return error;
3623 error2:
3624 IRELE(ip);
3625 error1:
3626 xfs_bmap_cancel(&free_list);
3627 cancel_flags |= XFS_TRANS_ABORT;
3628 error_return:
3629 xfs_trans_cancel(tp, cancel_flags);
3630 XFS_QM_DQRELE(mp, udqp);
3631 XFS_QM_DQRELE(mp, gdqp);
3633 if (!dp_joined_to_trans && (dp != NULL)) {
3634 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3637 goto std_return;
3642 * xfs_fid2
3644 * A fid routine that takes a pointer to a previously allocated
3645 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3647 STATIC int
3648 xfs_fid2(
3649 bhv_desc_t *bdp,
3650 fid_t *fidp)
3652 xfs_inode_t *ip;
3653 xfs_fid2_t *xfid;
3655 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3656 (inst_t *)__return_address);
3657 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3659 xfid = (xfs_fid2_t *)fidp;
3660 ip = XFS_BHVTOI(bdp);
3661 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3662 xfid->fid_pad = 0;
3664 * use memcpy because the inode is a long long and there's no
3665 * assurance that xfid->fid_ino is properly aligned.
3667 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3668 xfid->fid_gen = ip->i_d.di_gen;
3670 return 0;
3675 * xfs_rwlock
3678 xfs_rwlock(
3679 bhv_desc_t *bdp,
3680 bhv_vrwlock_t locktype)
3682 xfs_inode_t *ip;
3683 bhv_vnode_t *vp;
3685 vp = BHV_TO_VNODE(bdp);
3686 if (VN_ISDIR(vp))
3687 return 1;
3688 ip = XFS_BHVTOI(bdp);
3689 if (locktype == VRWLOCK_WRITE) {
3690 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3691 } else if (locktype == VRWLOCK_TRY_READ) {
3692 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
3693 } else if (locktype == VRWLOCK_TRY_WRITE) {
3694 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
3695 } else {
3696 ASSERT((locktype == VRWLOCK_READ) ||
3697 (locktype == VRWLOCK_WRITE_DIRECT));
3698 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3701 return 1;
3706 * xfs_rwunlock
3708 void
3709 xfs_rwunlock(
3710 bhv_desc_t *bdp,
3711 bhv_vrwlock_t locktype)
3713 xfs_inode_t *ip;
3714 bhv_vnode_t *vp;
3716 vp = BHV_TO_VNODE(bdp);
3717 if (VN_ISDIR(vp))
3718 return;
3719 ip = XFS_BHVTOI(bdp);
3720 if (locktype == VRWLOCK_WRITE) {
3722 * In the write case, we may have added a new entry to
3723 * the reference cache. This might store a pointer to
3724 * an inode to be released in this inode. If it is there,
3725 * clear the pointer and release the inode after unlocking
3726 * this one.
3728 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3729 } else {
3730 ASSERT((locktype == VRWLOCK_READ) ||
3731 (locktype == VRWLOCK_WRITE_DIRECT));
3732 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3734 return;
3737 STATIC int
3738 xfs_inode_flush(
3739 bhv_desc_t *bdp,
3740 int flags)
3742 xfs_inode_t *ip;
3743 xfs_mount_t *mp;
3744 xfs_inode_log_item_t *iip;
3745 int error = 0;
3747 ip = XFS_BHVTOI(bdp);
3748 mp = ip->i_mount;
3749 iip = ip->i_itemp;
3751 if (XFS_FORCED_SHUTDOWN(mp))
3752 return XFS_ERROR(EIO);
3755 * Bypass inodes which have already been cleaned by
3756 * the inode flush clustering code inside xfs_iflush
3758 if ((ip->i_update_core == 0) &&
3759 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3760 return 0;
3762 if (flags & FLUSH_LOG) {
3763 if (iip && iip->ili_last_lsn) {
3764 xlog_t *log = mp->m_log;
3765 xfs_lsn_t sync_lsn;
3766 int s, log_flags = XFS_LOG_FORCE;
3768 s = GRANT_LOCK(log);
3769 sync_lsn = log->l_last_sync_lsn;
3770 GRANT_UNLOCK(log, s);
3772 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3773 return 0;
3775 if (flags & FLUSH_SYNC)
3776 log_flags |= XFS_LOG_SYNC;
3777 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3782 * We make this non-blocking if the inode is contended,
3783 * return EAGAIN to indicate to the caller that they
3784 * did not succeed. This prevents the flush path from
3785 * blocking on inodes inside another operation right
3786 * now, they get caught later by xfs_sync.
3788 if (flags & FLUSH_INODE) {
3789 int flush_flags;
3791 if (xfs_ipincount(ip))
3792 return EAGAIN;
3794 if (flags & FLUSH_SYNC) {
3795 xfs_ilock(ip, XFS_ILOCK_SHARED);
3796 xfs_iflock(ip);
3797 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3798 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3799 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3800 return EAGAIN;
3802 } else {
3803 return EAGAIN;
3806 if (flags & FLUSH_SYNC)
3807 flush_flags = XFS_IFLUSH_SYNC;
3808 else
3809 flush_flags = XFS_IFLUSH_ASYNC;
3811 error = xfs_iflush(ip, flush_flags);
3812 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3815 return error;
3819 xfs_set_dmattrs (
3820 bhv_desc_t *bdp,
3821 u_int evmask,
3822 u_int16_t state,
3823 cred_t *credp)
3825 xfs_inode_t *ip;
3826 xfs_trans_t *tp;
3827 xfs_mount_t *mp;
3828 int error;
3830 if (!capable(CAP_SYS_ADMIN))
3831 return XFS_ERROR(EPERM);
3833 ip = XFS_BHVTOI(bdp);
3834 mp = ip->i_mount;
3836 if (XFS_FORCED_SHUTDOWN(mp))
3837 return XFS_ERROR(EIO);
3839 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3840 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3841 if (error) {
3842 xfs_trans_cancel(tp, 0);
3843 return error;
3845 xfs_ilock(ip, XFS_ILOCK_EXCL);
3846 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3848 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3849 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3851 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3852 IHOLD(ip);
3853 error = xfs_trans_commit(tp, 0);
3855 return error;
3858 STATIC int
3859 xfs_reclaim(
3860 bhv_desc_t *bdp)
3862 xfs_inode_t *ip;
3863 bhv_vnode_t *vp;
3865 vp = BHV_TO_VNODE(bdp);
3866 ip = XFS_BHVTOI(bdp);
3868 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3870 ASSERT(!VN_MAPPED(vp));
3872 /* bad inode, get out here ASAP */
3873 if (VN_BAD(vp)) {
3874 xfs_ireclaim(ip);
3875 return 0;
3878 vn_iowait(vp);
3880 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3883 * Make sure the atime in the XFS inode is correct before freeing the
3884 * Linux inode.
3886 xfs_synchronize_atime(ip);
3889 * If we have nothing to flush with this inode then complete the
3890 * teardown now, otherwise break the link between the xfs inode and the
3891 * linux inode and clean up the xfs inode later. This avoids flushing
3892 * the inode to disk during the delete operation itself.
3894 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3895 * first to ensure that xfs_iunpin() will never see an xfs inode
3896 * that has a linux inode being reclaimed. Synchronisation is provided
3897 * by the i_flags_lock.
3899 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3900 xfs_ilock(ip, XFS_ILOCK_EXCL);
3901 xfs_iflock(ip);
3902 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3903 } else {
3904 xfs_mount_t *mp = ip->i_mount;
3906 /* Protect sync and unpin from us */
3907 XFS_MOUNT_ILOCK(mp);
3908 spin_lock(&ip->i_flags_lock);
3909 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
3910 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3911 spin_unlock(&ip->i_flags_lock);
3912 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3913 XFS_MOUNT_IUNLOCK(mp);
3915 return 0;
3919 xfs_finish_reclaim(
3920 xfs_inode_t *ip,
3921 int locked,
3922 int sync_mode)
3924 xfs_ihash_t *ih = ip->i_hash;
3925 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
3926 int error;
3928 if (vp && VN_BAD(vp))
3929 goto reclaim;
3931 /* The hash lock here protects a thread in xfs_iget_core from
3932 * racing with us on linking the inode back with a vnode.
3933 * Once we have the XFS_IRECLAIM flag set it will not touch
3934 * us.
3936 write_lock(&ih->ih_lock);
3937 spin_lock(&ip->i_flags_lock);
3938 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
3939 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
3940 spin_unlock(&ip->i_flags_lock);
3941 write_unlock(&ih->ih_lock);
3942 if (locked) {
3943 xfs_ifunlock(ip);
3944 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3946 return 1;
3948 __xfs_iflags_set(ip, XFS_IRECLAIM);
3949 spin_unlock(&ip->i_flags_lock);
3950 write_unlock(&ih->ih_lock);
3953 * If the inode is still dirty, then flush it out. If the inode
3954 * is not in the AIL, then it will be OK to flush it delwri as
3955 * long as xfs_iflush() does not keep any references to the inode.
3956 * We leave that decision up to xfs_iflush() since it has the
3957 * knowledge of whether it's OK to simply do a delwri flush of
3958 * the inode or whether we need to wait until the inode is
3959 * pulled from the AIL.
3960 * We get the flush lock regardless, though, just to make sure
3961 * we don't free it while it is being flushed.
3963 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3964 if (!locked) {
3965 xfs_ilock(ip, XFS_ILOCK_EXCL);
3966 xfs_iflock(ip);
3969 if (ip->i_update_core ||
3970 ((ip->i_itemp != NULL) &&
3971 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3972 error = xfs_iflush(ip, sync_mode);
3974 * If we hit an error, typically because of filesystem
3975 * shutdown, we don't need to let vn_reclaim to know
3976 * because we're gonna reclaim the inode anyway.
3978 if (error) {
3979 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3980 goto reclaim;
3982 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3985 ASSERT(ip->i_update_core == 0);
3986 ASSERT(ip->i_itemp == NULL ||
3987 ip->i_itemp->ili_format.ilf_fields == 0);
3988 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3989 } else if (locked) {
3991 * We are not interested in doing an iflush if we're
3992 * in the process of shutting down the filesystem forcibly.
3993 * So, just reclaim the inode.
3995 xfs_ifunlock(ip);
3996 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3999 reclaim:
4000 xfs_ireclaim(ip);
4001 return 0;
4005 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
4007 int purged;
4008 xfs_inode_t *ip, *n;
4009 int done = 0;
4011 while (!done) {
4012 purged = 0;
4013 XFS_MOUNT_ILOCK(mp);
4014 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
4015 if (noblock) {
4016 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
4017 continue;
4018 if (xfs_ipincount(ip) ||
4019 !xfs_iflock_nowait(ip)) {
4020 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4021 continue;
4024 XFS_MOUNT_IUNLOCK(mp);
4025 if (xfs_finish_reclaim(ip, noblock,
4026 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
4027 delay(1);
4028 purged = 1;
4029 break;
4032 done = !purged;
4035 XFS_MOUNT_IUNLOCK(mp);
4036 return 0;
4040 * xfs_alloc_file_space()
4041 * This routine allocates disk space for the given file.
4043 * If alloc_type == 0, this request is for an ALLOCSP type
4044 * request which will change the file size. In this case, no
4045 * DMAPI event will be generated by the call. A TRUNCATE event
4046 * will be generated later by xfs_setattr.
4048 * If alloc_type != 0, this request is for a RESVSP type
4049 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4050 * lower block boundary byte address is less than the file's
4051 * length.
4053 * RETURNS:
4054 * 0 on success
4055 * errno on error
4058 STATIC int
4059 xfs_alloc_file_space(
4060 xfs_inode_t *ip,
4061 xfs_off_t offset,
4062 xfs_off_t len,
4063 int alloc_type,
4064 int attr_flags)
4066 xfs_mount_t *mp = ip->i_mount;
4067 xfs_off_t count;
4068 xfs_filblks_t allocated_fsb;
4069 xfs_filblks_t allocatesize_fsb;
4070 xfs_extlen_t extsz, temp;
4071 xfs_fileoff_t startoffset_fsb;
4072 xfs_fsblock_t firstfsb;
4073 int nimaps;
4074 int bmapi_flag;
4075 int quota_flag;
4076 int rt;
4077 xfs_trans_t *tp;
4078 xfs_bmbt_irec_t imaps[1], *imapp;
4079 xfs_bmap_free_t free_list;
4080 uint qblocks, resblks, resrtextents;
4081 int committed;
4082 int error;
4084 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4086 if (XFS_FORCED_SHUTDOWN(mp))
4087 return XFS_ERROR(EIO);
4089 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4090 return error;
4092 if (len <= 0)
4093 return XFS_ERROR(EINVAL);
4095 rt = XFS_IS_REALTIME_INODE(ip);
4096 extsz = xfs_get_extsz_hint(ip);
4098 count = len;
4099 imapp = &imaps[0];
4100 nimaps = 1;
4101 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4102 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4103 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4105 /* Generate a DMAPI event if needed. */
4106 if (alloc_type != 0 && offset < ip->i_size &&
4107 (attr_flags&ATTR_DMI) == 0 &&
4108 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4109 xfs_off_t end_dmi_offset;
4111 end_dmi_offset = offset+len;
4112 if (end_dmi_offset > ip->i_size)
4113 end_dmi_offset = ip->i_size;
4114 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4115 offset, end_dmi_offset - offset,
4116 0, NULL);
4117 if (error)
4118 return error;
4122 * Allocate file space until done or until there is an error
4124 retry:
4125 while (allocatesize_fsb && !error) {
4126 xfs_fileoff_t s, e;
4129 * Determine space reservations for data/realtime.
4131 if (unlikely(extsz)) {
4132 s = startoffset_fsb;
4133 do_div(s, extsz);
4134 s *= extsz;
4135 e = startoffset_fsb + allocatesize_fsb;
4136 if ((temp = do_mod(startoffset_fsb, extsz)))
4137 e += temp;
4138 if ((temp = do_mod(e, extsz)))
4139 e += extsz - temp;
4140 } else {
4141 s = 0;
4142 e = allocatesize_fsb;
4145 if (unlikely(rt)) {
4146 resrtextents = qblocks = (uint)(e - s);
4147 resrtextents /= mp->m_sb.sb_rextsize;
4148 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4149 quota_flag = XFS_QMOPT_RES_RTBLKS;
4150 } else {
4151 resrtextents = 0;
4152 resblks = qblocks = \
4153 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4154 quota_flag = XFS_QMOPT_RES_REGBLKS;
4158 * Allocate and setup the transaction.
4160 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4161 error = xfs_trans_reserve(tp, resblks,
4162 XFS_WRITE_LOG_RES(mp), resrtextents,
4163 XFS_TRANS_PERM_LOG_RES,
4164 XFS_WRITE_LOG_COUNT);
4166 * Check for running out of space
4168 if (error) {
4170 * Free the transaction structure.
4172 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4173 xfs_trans_cancel(tp, 0);
4174 break;
4176 xfs_ilock(ip, XFS_ILOCK_EXCL);
4177 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4178 qblocks, 0, quota_flag);
4179 if (error)
4180 goto error1;
4182 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4183 xfs_trans_ihold(tp, ip);
4186 * Issue the xfs_bmapi() call to allocate the blocks
4188 XFS_BMAP_INIT(&free_list, &firstfsb);
4189 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4190 allocatesize_fsb, bmapi_flag,
4191 &firstfsb, 0, imapp, &nimaps,
4192 &free_list, NULL);
4193 if (error) {
4194 goto error0;
4198 * Complete the transaction
4200 error = xfs_bmap_finish(&tp, &free_list, &committed);
4201 if (error) {
4202 goto error0;
4205 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4206 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4207 if (error) {
4208 break;
4211 allocated_fsb = imapp->br_blockcount;
4213 if (nimaps == 0) {
4214 error = XFS_ERROR(ENOSPC);
4215 break;
4218 startoffset_fsb += allocated_fsb;
4219 allocatesize_fsb -= allocated_fsb;
4221 dmapi_enospc_check:
4222 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4223 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4225 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4226 XFS_ITOV(ip), DM_RIGHT_NULL,
4227 XFS_ITOV(ip), DM_RIGHT_NULL,
4228 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4229 if (error == 0)
4230 goto retry; /* Maybe DMAPI app. has made space */
4231 /* else fall through with error from XFS_SEND_DATA */
4234 return error;
4236 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4237 xfs_bmap_cancel(&free_list);
4238 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4240 error1: /* Just cancel transaction */
4241 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4242 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4243 goto dmapi_enospc_check;
4247 * Zero file bytes between startoff and endoff inclusive.
4248 * The iolock is held exclusive and no blocks are buffered.
4250 STATIC int
4251 xfs_zero_remaining_bytes(
4252 xfs_inode_t *ip,
4253 xfs_off_t startoff,
4254 xfs_off_t endoff)
4256 xfs_bmbt_irec_t imap;
4257 xfs_fileoff_t offset_fsb;
4258 xfs_off_t lastoffset;
4259 xfs_off_t offset;
4260 xfs_buf_t *bp;
4261 xfs_mount_t *mp = ip->i_mount;
4262 int nimap;
4263 int error = 0;
4265 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4266 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4267 mp->m_rtdev_targp : mp->m_ddev_targp);
4269 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4270 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4271 nimap = 1;
4272 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4273 NULL, 0, &imap, &nimap, NULL, NULL);
4274 if (error || nimap < 1)
4275 break;
4276 ASSERT(imap.br_blockcount >= 1);
4277 ASSERT(imap.br_startoff == offset_fsb);
4278 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4279 if (lastoffset > endoff)
4280 lastoffset = endoff;
4281 if (imap.br_startblock == HOLESTARTBLOCK)
4282 continue;
4283 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4284 if (imap.br_state == XFS_EXT_UNWRITTEN)
4285 continue;
4286 XFS_BUF_UNDONE(bp);
4287 XFS_BUF_UNWRITE(bp);
4288 XFS_BUF_READ(bp);
4289 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4290 xfsbdstrat(mp, bp);
4291 if ((error = xfs_iowait(bp))) {
4292 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4293 mp, bp, XFS_BUF_ADDR(bp));
4294 break;
4296 memset(XFS_BUF_PTR(bp) +
4297 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4298 0, lastoffset - offset + 1);
4299 XFS_BUF_UNDONE(bp);
4300 XFS_BUF_UNREAD(bp);
4301 XFS_BUF_WRITE(bp);
4302 xfsbdstrat(mp, bp);
4303 if ((error = xfs_iowait(bp))) {
4304 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4305 mp, bp, XFS_BUF_ADDR(bp));
4306 break;
4309 xfs_buf_free(bp);
4310 return error;
4314 * xfs_free_file_space()
4315 * This routine frees disk space for the given file.
4317 * This routine is only called by xfs_change_file_space
4318 * for an UNRESVSP type call.
4320 * RETURNS:
4321 * 0 on success
4322 * errno on error
4325 STATIC int
4326 xfs_free_file_space(
4327 xfs_inode_t *ip,
4328 xfs_off_t offset,
4329 xfs_off_t len,
4330 int attr_flags)
4332 bhv_vnode_t *vp;
4333 int committed;
4334 int done;
4335 xfs_off_t end_dmi_offset;
4336 xfs_fileoff_t endoffset_fsb;
4337 int error;
4338 xfs_fsblock_t firstfsb;
4339 xfs_bmap_free_t free_list;
4340 xfs_bmbt_irec_t imap;
4341 xfs_off_t ioffset;
4342 xfs_extlen_t mod=0;
4343 xfs_mount_t *mp;
4344 int nimap;
4345 uint resblks;
4346 uint rounding;
4347 int rt;
4348 xfs_fileoff_t startoffset_fsb;
4349 xfs_trans_t *tp;
4350 int need_iolock = 1;
4352 vp = XFS_ITOV(ip);
4353 mp = ip->i_mount;
4355 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4357 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4358 return error;
4360 error = 0;
4361 if (len <= 0) /* if nothing being freed */
4362 return error;
4363 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4364 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4365 end_dmi_offset = offset + len;
4366 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4368 if (offset < ip->i_size &&
4369 (attr_flags & ATTR_DMI) == 0 &&
4370 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4371 if (end_dmi_offset > ip->i_size)
4372 end_dmi_offset = ip->i_size;
4373 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4374 offset, end_dmi_offset - offset,
4375 AT_DELAY_FLAG(attr_flags), NULL);
4376 if (error)
4377 return error;
4380 if (attr_flags & ATTR_NOLOCK)
4381 need_iolock = 0;
4382 if (need_iolock) {
4383 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4384 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4387 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, NBPP);
4388 ioffset = offset & ~(rounding - 1);
4390 if (VN_CACHED(vp) != 0) {
4391 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4392 ctooff(offtoct(ioffset)), -1);
4393 error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
4394 -1, FI_REMAPF_LOCKED);
4395 if (error)
4396 goto out_unlock_iolock;
4400 * Need to zero the stuff we're not freeing, on disk.
4401 * If its a realtime file & can't use unwritten extents then we
4402 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4403 * will take care of it for us.
4405 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4406 nimap = 1;
4407 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4408 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4409 if (error)
4410 goto out_unlock_iolock;
4411 ASSERT(nimap == 0 || nimap == 1);
4412 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4413 xfs_daddr_t block;
4415 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4416 block = imap.br_startblock;
4417 mod = do_div(block, mp->m_sb.sb_rextsize);
4418 if (mod)
4419 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4421 nimap = 1;
4422 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4423 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
4424 if (error)
4425 goto out_unlock_iolock;
4426 ASSERT(nimap == 0 || nimap == 1);
4427 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4428 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4429 mod++;
4430 if (mod && (mod != mp->m_sb.sb_rextsize))
4431 endoffset_fsb -= mod;
4434 if ((done = (endoffset_fsb <= startoffset_fsb)))
4436 * One contiguous piece to clear
4438 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4439 else {
4441 * Some full blocks, possibly two pieces to clear
4443 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4444 error = xfs_zero_remaining_bytes(ip, offset,
4445 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4446 if (!error &&
4447 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4448 error = xfs_zero_remaining_bytes(ip,
4449 XFS_FSB_TO_B(mp, endoffset_fsb),
4450 offset + len - 1);
4454 * free file space until done or until there is an error
4456 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4457 while (!error && !done) {
4460 * allocate and setup the transaction. Allow this
4461 * transaction to dip into the reserve blocks to ensure
4462 * the freeing of the space succeeds at ENOSPC.
4464 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4465 tp->t_flags |= XFS_TRANS_RESERVE;
4466 error = xfs_trans_reserve(tp,
4467 resblks,
4468 XFS_WRITE_LOG_RES(mp),
4470 XFS_TRANS_PERM_LOG_RES,
4471 XFS_WRITE_LOG_COUNT);
4474 * check for running out of space
4476 if (error) {
4478 * Free the transaction structure.
4480 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4481 xfs_trans_cancel(tp, 0);
4482 break;
4484 xfs_ilock(ip, XFS_ILOCK_EXCL);
4485 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4486 ip->i_udquot, ip->i_gdquot, resblks, 0,
4487 XFS_QMOPT_RES_REGBLKS);
4488 if (error)
4489 goto error1;
4491 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4492 xfs_trans_ihold(tp, ip);
4495 * issue the bunmapi() call to free the blocks
4497 XFS_BMAP_INIT(&free_list, &firstfsb);
4498 error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
4499 endoffset_fsb - startoffset_fsb,
4500 0, 2, &firstfsb, &free_list, NULL, &done);
4501 if (error) {
4502 goto error0;
4506 * complete the transaction
4508 error = xfs_bmap_finish(&tp, &free_list, &committed);
4509 if (error) {
4510 goto error0;
4513 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
4514 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4517 out_unlock_iolock:
4518 if (need_iolock)
4519 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4520 return error;
4522 error0:
4523 xfs_bmap_cancel(&free_list);
4524 error1:
4525 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4526 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4527 XFS_ILOCK_EXCL);
4528 return error;
4532 * xfs_change_file_space()
4533 * This routine allocates or frees disk space for the given file.
4534 * The user specified parameters are checked for alignment and size
4535 * limitations.
4537 * RETURNS:
4538 * 0 on success
4539 * errno on error
4543 xfs_change_file_space(
4544 bhv_desc_t *bdp,
4545 int cmd,
4546 xfs_flock64_t *bf,
4547 xfs_off_t offset,
4548 cred_t *credp,
4549 int attr_flags)
4551 int clrprealloc;
4552 int error;
4553 xfs_fsize_t fsize;
4554 xfs_inode_t *ip;
4555 xfs_mount_t *mp;
4556 int setprealloc;
4557 xfs_off_t startoffset;
4558 xfs_off_t llen;
4559 xfs_trans_t *tp;
4560 bhv_vattr_t va;
4561 bhv_vnode_t *vp;
4563 vp = BHV_TO_VNODE(bdp);
4564 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4566 ip = XFS_BHVTOI(bdp);
4567 mp = ip->i_mount;
4570 * must be a regular file and have write permission
4572 if (!VN_ISREG(vp))
4573 return XFS_ERROR(EINVAL);
4575 xfs_ilock(ip, XFS_ILOCK_SHARED);
4577 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4578 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4579 return error;
4582 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4584 switch (bf->l_whence) {
4585 case 0: /*SEEK_SET*/
4586 break;
4587 case 1: /*SEEK_CUR*/
4588 bf->l_start += offset;
4589 break;
4590 case 2: /*SEEK_END*/
4591 bf->l_start += ip->i_size;
4592 break;
4593 default:
4594 return XFS_ERROR(EINVAL);
4597 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4599 if ( (bf->l_start < 0)
4600 || (bf->l_start > XFS_MAXIOFFSET(mp))
4601 || (bf->l_start + llen < 0)
4602 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4603 return XFS_ERROR(EINVAL);
4605 bf->l_whence = 0;
4607 startoffset = bf->l_start;
4608 fsize = ip->i_size;
4611 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4612 * file space.
4613 * These calls do NOT zero the data space allocated to the file,
4614 * nor do they change the file size.
4616 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4617 * space.
4618 * These calls cause the new file data to be zeroed and the file
4619 * size to be changed.
4621 setprealloc = clrprealloc = 0;
4623 switch (cmd) {
4624 case XFS_IOC_RESVSP:
4625 case XFS_IOC_RESVSP64:
4626 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4627 1, attr_flags);
4628 if (error)
4629 return error;
4630 setprealloc = 1;
4631 break;
4633 case XFS_IOC_UNRESVSP:
4634 case XFS_IOC_UNRESVSP64:
4635 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4636 attr_flags)))
4637 return error;
4638 break;
4640 case XFS_IOC_ALLOCSP:
4641 case XFS_IOC_ALLOCSP64:
4642 case XFS_IOC_FREESP:
4643 case XFS_IOC_FREESP64:
4644 if (startoffset > fsize) {
4645 error = xfs_alloc_file_space(ip, fsize,
4646 startoffset - fsize, 0, attr_flags);
4647 if (error)
4648 break;
4651 va.va_mask = XFS_AT_SIZE;
4652 va.va_size = startoffset;
4654 error = xfs_setattr(bdp, &va, attr_flags, credp);
4656 if (error)
4657 return error;
4659 clrprealloc = 1;
4660 break;
4662 default:
4663 ASSERT(0);
4664 return XFS_ERROR(EINVAL);
4668 * update the inode timestamp, mode, and prealloc flag bits
4670 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4672 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4673 0, 0, 0))) {
4674 /* ASSERT(0); */
4675 xfs_trans_cancel(tp, 0);
4676 return error;
4679 xfs_ilock(ip, XFS_ILOCK_EXCL);
4681 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4682 xfs_trans_ihold(tp, ip);
4684 if ((attr_flags & ATTR_DMI) == 0) {
4685 ip->i_d.di_mode &= ~S_ISUID;
4688 * Note that we don't have to worry about mandatory
4689 * file locking being disabled here because we only
4690 * clear the S_ISGID bit if the Group execute bit is
4691 * on, but if it was on then mandatory locking wouldn't
4692 * have been enabled.
4694 if (ip->i_d.di_mode & S_IXGRP)
4695 ip->i_d.di_mode &= ~S_ISGID;
4697 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4699 if (setprealloc)
4700 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4701 else if (clrprealloc)
4702 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4704 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4705 xfs_trans_set_sync(tp);
4707 error = xfs_trans_commit(tp, 0);
4709 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4711 return error;
4714 bhv_vnodeops_t xfs_vnodeops = {
4715 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4716 .vop_open = xfs_open,
4717 .vop_read = xfs_read,
4718 #ifdef HAVE_SPLICE
4719 .vop_splice_read = xfs_splice_read,
4720 .vop_splice_write = xfs_splice_write,
4721 #endif
4722 .vop_write = xfs_write,
4723 .vop_ioctl = xfs_ioctl,
4724 .vop_getattr = xfs_getattr,
4725 .vop_setattr = xfs_setattr,
4726 .vop_access = xfs_access,
4727 .vop_lookup = xfs_lookup,
4728 .vop_create = xfs_create,
4729 .vop_remove = xfs_remove,
4730 .vop_link = xfs_link,
4731 .vop_rename = xfs_rename,
4732 .vop_mkdir = xfs_mkdir,
4733 .vop_rmdir = xfs_rmdir,
4734 .vop_readdir = xfs_readdir,
4735 .vop_symlink = xfs_symlink,
4736 .vop_readlink = xfs_readlink,
4737 .vop_fsync = xfs_fsync,
4738 .vop_inactive = xfs_inactive,
4739 .vop_fid2 = xfs_fid2,
4740 .vop_rwlock = xfs_rwlock,
4741 .vop_rwunlock = xfs_rwunlock,
4742 .vop_bmap = xfs_bmap,
4743 .vop_reclaim = xfs_reclaim,
4744 .vop_attr_get = xfs_attr_get,
4745 .vop_attr_set = xfs_attr_set,
4746 .vop_attr_remove = xfs_attr_remove,
4747 .vop_attr_list = xfs_attr_list,
4748 .vop_link_removed = (vop_link_removed_t)fs_noval,
4749 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4750 .vop_tosspages = fs_tosspages,
4751 .vop_flushinval_pages = fs_flushinval_pages,
4752 .vop_flush_pages = fs_flush_pages,
4753 .vop_release = xfs_release,
4754 .vop_iflush = xfs_inode_flush,