2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
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 <linux/capability.h>
23 #include "xfs_types.h"
27 #include "xfs_trans.h"
32 #include "xfs_dmapi.h"
33 #include "xfs_mount.h"
34 #include "xfs_da_btree.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_alloc_btree.h"
37 #include "xfs_ialloc_btree.h"
38 #include "xfs_dir_sf.h"
39 #include "xfs_dir2_sf.h"
40 #include "xfs_attr_sf.h"
41 #include "xfs_dinode.h"
42 #include "xfs_inode.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_dir_leaf.h"
45 #include "xfs_itable.h"
46 #include "xfs_btree.h"
47 #include "xfs_ialloc.h"
48 #include "xfs_alloc.h"
52 #include "xfs_error.h"
53 #include "xfs_quota.h"
54 #include "xfs_utils.h"
55 #include "xfs_rtalloc.h"
56 #include "xfs_refcache.h"
57 #include "xfs_trans_space.h"
58 #include "xfs_log_priv.h"
63 * The maximum pathlen is 1024 bytes. Since the minimum file system
64 * blocksize is 512 bytes, we can get a max of 2 extents back from
67 #define SYMLINK_MAPS 2
70 * For xfs, we check that the file isn't too big to be opened by this kernel.
71 * No other open action is required for regular files. Devices are handled
72 * through the specfs file system, pipes through fifofs. Device and
73 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
74 * when a new vnode is first looked up or created.
85 vp = BHV_TO_VNODE(bdp);
88 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
89 return XFS_ERROR(EIO);
92 * If it's a directory with any blocks, read-ahead block 0
93 * as we're almost certain to have the next operation be a read there.
95 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
96 mode = xfs_ilock_map_shared(ip);
97 if (ip->i_d.di_nextents > 0)
98 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
99 xfs_iunlock(ip, mode);
119 vp = BHV_TO_VNODE(bdp);
120 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
122 ip = XFS_BHVTOI(bdp);
125 if (XFS_FORCED_SHUTDOWN(mp))
126 return XFS_ERROR(EIO);
128 if (!(flags & ATTR_LAZY))
129 xfs_ilock(ip, XFS_ILOCK_SHARED);
131 vap->va_size = ip->i_d.di_size;
132 if (vap->va_mask == XFS_AT_SIZE)
136 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
137 vap->va_nodeid = ip->i_ino;
139 vap->va_nodeid += mp->m_inoadd;
141 vap->va_nlink = ip->i_d.di_nlink;
144 * Quick exit for non-stat callers
147 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
148 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
152 * Copy from in-core inode.
154 vap->va_mode = ip->i_d.di_mode;
155 vap->va_uid = ip->i_d.di_uid;
156 vap->va_gid = ip->i_d.di_gid;
157 vap->va_projid = ip->i_d.di_projid;
160 * Check vnode type block/char vs. everything else.
162 switch (ip->i_d.di_mode & S_IFMT) {
165 vap->va_rdev = ip->i_df.if_u2.if_rdev;
166 vap->va_blocksize = BLKDEV_IOSIZE;
171 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
172 vap->va_blocksize = xfs_preferred_iosize(mp);
176 * If the file blocks are being allocated from a
177 * realtime partition, then return the inode's
178 * realtime extent size or the realtime volume's
181 vap->va_blocksize = ip->i_d.di_extsize ?
182 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
183 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
188 vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
189 vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
190 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
191 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
192 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
193 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
196 * Exit for stat callers. See if any of the rest of the fields
197 * to be filled in are needed.
200 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
201 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
205 * Convert di_flags to xflags.
207 vap->va_xflags = xfs_ip2xflags(ip);
210 * Exit for inode revalidate. See if any of the rest of
211 * the fields to be filled in are needed.
214 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
215 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
218 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
220 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
221 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
225 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
226 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
227 ip->i_d.di_anextents;
229 vap->va_anextents = 0;
230 vap->va_gen = ip->i_d.di_gen;
233 if (!(flags & ATTR_LAZY))
234 xfs_iunlock(ip, XFS_ILOCK_SHARED);
260 xfs_prid_t projid=0, iprojid=0;
261 int mandlock_before, mandlock_after;
262 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
266 vp = BHV_TO_VNODE(bdp);
267 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
269 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
270 return XFS_ERROR(EROFS);
273 * Cannot set certain attributes.
276 if (mask & XFS_AT_NOSET) {
277 return XFS_ERROR(EINVAL);
280 ip = XFS_BHVTOI(bdp);
283 if (XFS_FORCED_SHUTDOWN(mp))
284 return XFS_ERROR(EIO);
287 * Timestamps do not need to be logged and hence do not
288 * need to be done within a transaction.
290 if (mask & XFS_AT_UPDTIMES) {
291 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
292 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
293 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
294 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
295 xfs_ichgtime(ip, timeflags);
299 olddquot1 = olddquot2 = NULL;
303 * If disk quotas is on, we make sure that the dquots do exist on disk,
304 * before we start any other transactions. Trying to do this later
305 * is messy. We don't care to take a readlock to look at the ids
306 * in inode here, because we can't hold it across the trans_reserve.
307 * If the IDs do change before we take the ilock, we're covered
308 * because the i_*dquot fields will get updated anyway.
310 if (XFS_IS_QUOTA_ON(mp) &&
311 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
314 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
316 qflags |= XFS_QMOPT_UQUOTA;
318 uid = ip->i_d.di_uid;
320 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
322 qflags |= XFS_QMOPT_GQUOTA;
324 gid = ip->i_d.di_gid;
326 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
327 projid = vap->va_projid;
328 qflags |= XFS_QMOPT_PQUOTA;
330 projid = ip->i_d.di_projid;
333 * We take a reference when we initialize udqp and gdqp,
334 * so it is important that we never blindly double trip on
335 * the same variable. See xfs_create() for an example.
337 ASSERT(udqp == NULL);
338 ASSERT(gdqp == NULL);
339 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
346 * For the other attributes, we acquire the inode lock and
347 * first do an error checking pass.
350 lock_flags = XFS_ILOCK_EXCL;
351 ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
352 if (flags & ATTR_NOLOCK)
354 if (!(mask & XFS_AT_SIZE)) {
355 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
356 (mp->m_flags & XFS_MOUNT_WSYNC)) {
357 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
359 if ((code = xfs_trans_reserve(tp, 0,
360 XFS_ICHANGE_LOG_RES(mp), 0,
367 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
368 !(flags & ATTR_DMI)) {
369 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
370 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
371 vap->va_size, 0, dmflags, NULL);
378 lock_flags |= XFS_IOLOCK_EXCL;
381 xfs_ilock(ip, lock_flags);
383 /* boolean: are we the file owner? */
384 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
387 * Change various properties of a file.
388 * Only the owner or users with CAP_FOWNER
389 * capability may do these things.
392 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
393 XFS_AT_GID|XFS_AT_PROJID)) {
395 * CAP_FOWNER overrides the following restrictions:
397 * The user ID of the calling process must be equal
398 * to the file owner ID, except in cases where the
399 * CAP_FSETID capability is applicable.
401 if (!file_owner && !capable(CAP_FOWNER)) {
402 code = XFS_ERROR(EPERM);
407 * CAP_FSETID overrides the following restrictions:
409 * The effective user ID of the calling process shall match
410 * the file owner when setting the set-user-ID and
411 * set-group-ID bits on that file.
413 * The effective group ID or one of the supplementary group
414 * IDs of the calling process shall match the group owner of
415 * the file when setting the set-group-ID bit on that file
417 if (mask & XFS_AT_MODE) {
420 if ((vap->va_mode & S_ISUID) && !file_owner)
422 if ((vap->va_mode & S_ISGID) &&
423 !in_group_p((gid_t)ip->i_d.di_gid))
426 /* Linux allows this, Irix doesn't. */
427 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
430 if (m && !capable(CAP_FSETID))
436 * Change file ownership. Must be the owner or privileged.
437 * If the system was configured with the "restricted_chown"
438 * option, the owner is not permitted to give away the file,
439 * and can change the group id only to a group of which he
440 * or she is a member.
442 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
444 * These IDs could have changed since we last looked at them.
445 * But, we're assured that if the ownership did change
446 * while we didn't have the inode locked, inode's dquot(s)
447 * would have changed also.
449 iuid = ip->i_d.di_uid;
450 iprojid = ip->i_d.di_projid;
451 igid = ip->i_d.di_gid;
452 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
453 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
454 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
458 * CAP_CHOWN overrides the following restrictions:
460 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
461 * shall override the restriction that a process cannot
462 * change the user ID of a file it owns and the restriction
463 * that the group ID supplied to the chown() function
464 * shall be equal to either the group ID or one of the
465 * supplementary group IDs of the calling process.
467 if (restricted_chown &&
468 (iuid != uid || (igid != gid &&
469 !in_group_p((gid_t)gid))) &&
470 !capable(CAP_CHOWN)) {
471 code = XFS_ERROR(EPERM);
475 * Do a quota reservation only if uid/projid/gid is actually
478 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
479 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
480 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
482 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
483 capable(CAP_FOWNER) ?
484 XFS_QMOPT_FORCE_RES : 0);
485 if (code) /* out of quota */
491 * Truncate file. Must have write permission and not be a directory.
493 if (mask & XFS_AT_SIZE) {
494 /* Short circuit the truncate case for zero length files */
495 if ((vap->va_size == 0) &&
496 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
497 xfs_iunlock(ip, XFS_ILOCK_EXCL);
498 lock_flags &= ~XFS_ILOCK_EXCL;
499 if (mask & XFS_AT_CTIME)
500 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
506 code = XFS_ERROR(EISDIR);
508 } else if (!VN_ISREG(vp)) {
509 code = XFS_ERROR(EINVAL);
513 * Make sure that the dquots are attached to the inode.
515 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
520 * Change file access or modified times.
522 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
524 if ((flags & ATTR_UTIME) &&
525 !capable(CAP_FOWNER)) {
526 code = XFS_ERROR(EPERM);
533 * Change extent size or realtime flag.
535 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
537 * Can't change extent size if any extents are allocated.
539 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
540 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
542 code = XFS_ERROR(EINVAL); /* EFBIG? */
547 * Can't set extent size unless the file is marked, or
548 * about to be marked as a realtime file.
550 * This check will be removed when fixed size extents
551 * with buffered data writes is implemented.
554 if ((mask & XFS_AT_EXTSIZE) &&
555 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
557 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
558 ((mask & XFS_AT_XFLAGS) &&
559 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
560 code = XFS_ERROR(EINVAL);
565 * Can't change realtime flag if any extents are allocated.
567 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
568 (mask & XFS_AT_XFLAGS) &&
569 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
570 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
571 code = XFS_ERROR(EINVAL); /* EFBIG? */
575 * Extent size must be a multiple of the appropriate block
576 * size, if set at all.
578 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
581 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
582 ((mask & XFS_AT_XFLAGS) &&
583 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
584 size = mp->m_sb.sb_rextsize <<
585 mp->m_sb.sb_blocklog;
587 size = mp->m_sb.sb_blocksize;
589 if (vap->va_extsize % size) {
590 code = XFS_ERROR(EINVAL);
595 * If realtime flag is set then must have realtime data.
597 if ((mask & XFS_AT_XFLAGS) &&
598 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
599 if ((mp->m_sb.sb_rblocks == 0) ||
600 (mp->m_sb.sb_rextsize == 0) ||
601 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
602 code = XFS_ERROR(EINVAL);
608 * Can't modify an immutable/append-only file unless
609 * we have appropriate permission.
611 if ((mask & XFS_AT_XFLAGS) &&
613 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
615 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
616 !capable(CAP_LINUX_IMMUTABLE)) {
617 code = XFS_ERROR(EPERM);
623 * Now we can make the changes. Before we join the inode
624 * to the transaction, if XFS_AT_SIZE is set then take care of
625 * the part of the truncation that must be done without the
626 * inode lock. This needs to be done before joining the inode
627 * to the transaction, because the inode cannot be unlocked
628 * once it is a part of the transaction.
630 if (mask & XFS_AT_SIZE) {
632 if ((vap->va_size > ip->i_d.di_size) &&
633 (flags & ATTR_NOSIZETOK) == 0) {
634 code = xfs_igrow_start(ip, vap->va_size, credp);
636 xfs_iunlock(ip, XFS_ILOCK_EXCL);
638 code = xfs_itruncate_data(ip, vap->va_size);
641 lock_flags &= ~XFS_ILOCK_EXCL;
642 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
645 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
646 if ((code = xfs_trans_reserve(tp, 0,
647 XFS_ITRUNCATE_LOG_RES(mp), 0,
648 XFS_TRANS_PERM_LOG_RES,
649 XFS_ITRUNCATE_LOG_COUNT))) {
650 xfs_trans_cancel(tp, 0);
652 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
655 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
656 xfs_ilock(ip, XFS_ILOCK_EXCL);
660 xfs_trans_ijoin(tp, ip, lock_flags);
661 xfs_trans_ihold(tp, ip);
664 /* determine whether mandatory locking mode changes */
665 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
668 * Truncate file. Must have write permission and not be a directory.
670 if (mask & XFS_AT_SIZE) {
671 if (vap->va_size > ip->i_d.di_size) {
672 xfs_igrow_finish(tp, ip, vap->va_size,
673 !(flags & ATTR_DMI));
674 } else if ((vap->va_size <= ip->i_d.di_size) ||
675 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
677 * signal a sync transaction unless
678 * we're truncating an already unlinked
679 * file on a wsync filesystem
681 code = xfs_itruncate_finish(&tp, ip,
682 (xfs_fsize_t)vap->va_size,
684 ((ip->i_d.di_nlink != 0 ||
685 !(mp->m_flags & XFS_MOUNT_WSYNC))
692 * Have to do this even if the file's size doesn't change.
694 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
698 * Change file access modes.
700 if (mask & XFS_AT_MODE) {
701 ip->i_d.di_mode &= S_IFMT;
702 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
704 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
705 timeflags |= XFS_ICHGTIME_CHG;
709 * Change file ownership. Must be the owner or privileged.
710 * If the system was configured with the "restricted_chown"
711 * option, the owner is not permitted to give away the file,
712 * and can change the group id only to a group of which he
713 * or she is a member.
715 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
717 * CAP_FSETID overrides the following restrictions:
719 * The set-user-ID and set-group-ID bits of a file will be
720 * cleared upon successful return from chown()
722 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
723 !capable(CAP_FSETID)) {
724 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
728 * Change the ownerships and register quota modifications
729 * in the transaction.
732 if (XFS_IS_UQUOTA_ON(mp)) {
733 ASSERT(mask & XFS_AT_UID);
735 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
736 &ip->i_udquot, udqp);
738 ip->i_d.di_uid = uid;
741 if (XFS_IS_GQUOTA_ON(mp)) {
742 ASSERT(!XFS_IS_PQUOTA_ON(mp));
743 ASSERT(mask & XFS_AT_GID);
745 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
746 &ip->i_gdquot, gdqp);
748 ip->i_d.di_gid = gid;
750 if (iprojid != projid) {
751 if (XFS_IS_PQUOTA_ON(mp)) {
752 ASSERT(!XFS_IS_GQUOTA_ON(mp));
753 ASSERT(mask & XFS_AT_PROJID);
755 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
756 &ip->i_gdquot, gdqp);
758 ip->i_d.di_projid = projid;
760 * We may have to rev the inode as well as
761 * the superblock version number since projids didn't
762 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
764 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
765 xfs_bump_ino_vers2(tp, ip);
768 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
769 timeflags |= XFS_ICHGTIME_CHG;
774 * Change file access or modified times.
776 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
777 if (mask & XFS_AT_ATIME) {
778 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
779 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
780 ip->i_update_core = 1;
781 timeflags &= ~XFS_ICHGTIME_ACC;
783 if (mask & XFS_AT_MTIME) {
784 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
785 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
786 timeflags &= ~XFS_ICHGTIME_MOD;
787 timeflags |= XFS_ICHGTIME_CHG;
789 if (tp && (flags & ATTR_UTIME))
790 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
794 * Change XFS-added attributes.
796 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
797 if (mask & XFS_AT_EXTSIZE) {
799 * Converting bytes to fs blocks.
801 ip->i_d.di_extsize = vap->va_extsize >>
802 mp->m_sb.sb_blocklog;
804 if (mask & XFS_AT_XFLAGS) {
807 /* can't set PREALLOC this way, just preserve it */
808 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
809 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
810 di_flags |= XFS_DIFLAG_IMMUTABLE;
811 if (vap->va_xflags & XFS_XFLAG_APPEND)
812 di_flags |= XFS_DIFLAG_APPEND;
813 if (vap->va_xflags & XFS_XFLAG_SYNC)
814 di_flags |= XFS_DIFLAG_SYNC;
815 if (vap->va_xflags & XFS_XFLAG_NOATIME)
816 di_flags |= XFS_DIFLAG_NOATIME;
817 if (vap->va_xflags & XFS_XFLAG_NODUMP)
818 di_flags |= XFS_DIFLAG_NODUMP;
819 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
820 di_flags |= XFS_DIFLAG_PROJINHERIT;
821 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
822 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
823 di_flags |= XFS_DIFLAG_RTINHERIT;
824 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
825 di_flags |= XFS_DIFLAG_NOSYMLINKS;
827 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
828 di_flags |= XFS_DIFLAG_REALTIME;
829 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
831 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
834 ip->i_d.di_flags = di_flags;
836 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
837 timeflags |= XFS_ICHGTIME_CHG;
841 * Change file inode change time only if XFS_AT_CTIME set
842 * AND we have been called by a DMI function.
845 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
846 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
847 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
848 ip->i_update_core = 1;
849 timeflags &= ~XFS_ICHGTIME_CHG;
853 * Send out timestamp changes that need to be set to the
854 * current time. Not done when called by a DMI function.
856 if (timeflags && !(flags & ATTR_DMI))
857 xfs_ichgtime(ip, timeflags);
859 XFS_STATS_INC(xs_ig_attrchg);
862 * If this is a synchronous mount, make sure that the
863 * transaction goes to disk before returning to the user.
864 * This is slightly sub-optimal in that truncates require
865 * two sync transactions instead of one for wsync filesytems.
866 * One for the truncate and one for the timestamps since we
867 * don't want to change the timestamps unless we're sure the
868 * truncate worked. Truncates are less than 1% of the laddis
869 * mix so this probably isn't worth the trouble to optimize.
873 if (mp->m_flags & XFS_MOUNT_WSYNC)
874 xfs_trans_set_sync(tp);
876 code = xfs_trans_commit(tp, commit_flags, NULL);
880 * If the (regular) file's mandatory locking mode changed, then
881 * notify the vnode. We do this under the inode lock to prevent
882 * racing calls to vop_vnode_change.
884 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
885 if (mandlock_before != mandlock_after) {
886 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
890 xfs_iunlock(ip, lock_flags);
893 * Release any dquot(s) the inode had kept before chown.
895 XFS_QM_DQRELE(mp, olddquot1);
896 XFS_QM_DQRELE(mp, olddquot2);
897 XFS_QM_DQRELE(mp, udqp);
898 XFS_QM_DQRELE(mp, gdqp);
904 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
905 !(flags & ATTR_DMI)) {
906 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
907 NULL, DM_RIGHT_NULL, NULL, NULL,
908 0, 0, AT_DELAY_FLAG(flags));
913 commit_flags |= XFS_TRANS_ABORT;
916 XFS_QM_DQRELE(mp, udqp);
917 XFS_QM_DQRELE(mp, gdqp);
919 xfs_trans_cancel(tp, commit_flags);
921 if (lock_flags != 0) {
922 xfs_iunlock(ip, lock_flags);
930 * Null conversion from vnode mode bits to inode mode bits, as in efs.
941 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
942 (inst_t *)__return_address);
944 ip = XFS_BHVTOI(bdp);
945 xfs_ilock(ip, XFS_ILOCK_SHARED);
946 error = xfs_iaccess(ip, mode, credp);
947 xfs_iunlock(ip, XFS_ILOCK_SHARED);
971 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
977 vp = BHV_TO_VNODE(bdp);
978 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
980 ip = XFS_BHVTOI(bdp);
983 if (XFS_FORCED_SHUTDOWN(mp))
984 return XFS_ERROR(EIO);
986 xfs_ilock(ip, XFS_ILOCK_SHARED);
988 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
990 offset = uiop->uio_offset;
991 count = uiop->uio_resid;
994 error = XFS_ERROR(EINVAL);
1002 if (!(ioflags & IO_INVIS)) {
1003 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1007 * See if the symlink is stored inline.
1009 pathlen = (int)ip->i_d.di_size;
1011 if (ip->i_df.if_flags & XFS_IFINLINE) {
1012 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1016 * Symlink not inline. Call bmap to get it in.
1018 nmaps = SYMLINK_MAPS;
1020 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1021 0, NULL, 0, mval, &nmaps, NULL);
1027 for (n = 0; n < nmaps; n++) {
1028 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1029 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1030 bp = xfs_buf_read(mp->m_ddev_targp, d,
1031 BTOBB(byte_cnt), 0);
1032 error = XFS_BUF_GETERROR(bp);
1034 xfs_ioerror_alert("xfs_readlink",
1035 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1039 if (pathlen < byte_cnt)
1041 pathlen -= byte_cnt;
1043 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1052 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1061 * This is called to sync the inode and its data out to disk.
1062 * We need to hold the I/O lock while flushing the data, and
1063 * the inode lock while flushing the inode. The inode lock CANNOT
1064 * be held while flushing the data, so acquire after we're done
1078 int log_flushed = 0, changed = 1;
1080 vn_trace_entry(BHV_TO_VNODE(bdp),
1081 __FUNCTION__, (inst_t *)__return_address);
1083 ip = XFS_BHVTOI(bdp);
1085 ASSERT(start >= 0 && stop >= -1);
1087 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1088 return XFS_ERROR(EIO);
1091 * We always need to make sure that the required inode state
1092 * is safe on disk. The vnode might be clean but because
1093 * of committed transactions that haven't hit the disk yet.
1094 * Likewise, there could be unflushed non-transactional
1095 * changes to the inode core that have to go to disk.
1097 * The following code depends on one assumption: that
1098 * any transaction that changes an inode logs the core
1099 * because it has to change some field in the inode core
1100 * (typically nextents or nblocks). That assumption
1101 * implies that any transactions against an inode will
1102 * catch any non-transactional updates. If inode-altering
1103 * transactions exist that violate this assumption, the
1104 * code breaks. Right now, it figures that if the involved
1105 * update_* field is clear and the inode is unpinned, the
1106 * inode is clean. Either it's been flushed or it's been
1107 * committed and the commit has hit the disk unpinning the inode.
1108 * (Note that xfs_inode_item_format() called at commit clears
1109 * the update_* fields.)
1111 xfs_ilock(ip, XFS_ILOCK_SHARED);
1113 /* If we are flushing data then we care about update_size
1114 * being set, otherwise we care about update_core
1116 if ((flag & FSYNC_DATA) ?
1117 (ip->i_update_size == 0) :
1118 (ip->i_update_core == 0)) {
1120 * Timestamps/size haven't changed since last inode
1121 * flush or inode transaction commit. That means
1122 * either nothing got written or a transaction
1123 * committed which caught the updates. If the
1124 * latter happened and the transaction hasn't
1125 * hit the disk yet, the inode will be still
1126 * be pinned. If it is, force the log.
1129 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1131 if (xfs_ipincount(ip)) {
1132 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1134 ((flag & FSYNC_WAIT)
1135 ? XFS_LOG_SYNC : 0),
1139 * If the inode is not pinned and nothing
1140 * has changed we don't need to flush the
1148 * Kick off a transaction to log the inode
1149 * core to get the updates. Make it
1150 * sync if FSYNC_WAIT is passed in (which
1151 * is done by everybody but specfs). The
1152 * sync transaction will also force the log.
1154 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1155 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1156 if ((error = xfs_trans_reserve(tp, 0,
1157 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1159 xfs_trans_cancel(tp, 0);
1162 xfs_ilock(ip, XFS_ILOCK_EXCL);
1165 * Note - it's possible that we might have pushed
1166 * ourselves out of the way during trans_reserve
1167 * which would flush the inode. But there's no
1168 * guarantee that the inode buffer has actually
1169 * gone out yet (it's delwri). Plus the buffer
1170 * could be pinned anyway if it's part of an
1171 * inode in another recent transaction. So we
1172 * play it safe and fire off the transaction anyway.
1174 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1175 xfs_trans_ihold(tp, ip);
1176 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1177 if (flag & FSYNC_WAIT)
1178 xfs_trans_set_sync(tp);
1179 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
1181 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1184 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1186 * If the log write didn't issue an ordered tag we need
1187 * to flush the disk cache for the data device now.
1190 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1193 * If this inode is on the RT dev we need to flush that
1196 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1197 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1204 * This is called by xfs_inactive to free any blocks beyond eof,
1205 * when the link count isn't zero.
1208 xfs_inactive_free_eofblocks(
1214 xfs_fileoff_t end_fsb;
1215 xfs_fileoff_t last_fsb;
1216 xfs_filblks_t map_len;
1218 xfs_bmbt_irec_t imap;
1221 * Figure out if there are any blocks beyond the end
1222 * of the file. If not, then there is nothing to do.
1224 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1225 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1226 map_len = last_fsb - end_fsb;
1231 xfs_ilock(ip, XFS_ILOCK_SHARED);
1232 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1233 NULL, 0, &imap, &nimaps, NULL);
1234 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1236 if (!error && (nimaps != 0) &&
1237 (imap.br_startblock != HOLESTARTBLOCK)) {
1239 * Attach the dquots to the inode up front.
1241 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1245 * There are blocks after the end of file.
1246 * Free them up now by truncating the file to
1249 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1252 * Do the xfs_itruncate_start() call before
1253 * reserving any log space because
1254 * itruncate_start will call into the buffer
1255 * cache and we can't
1256 * do that within a transaction.
1258 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1259 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1262 error = xfs_trans_reserve(tp, 0,
1263 XFS_ITRUNCATE_LOG_RES(mp),
1264 0, XFS_TRANS_PERM_LOG_RES,
1265 XFS_ITRUNCATE_LOG_COUNT);
1267 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1268 xfs_trans_cancel(tp, 0);
1269 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1273 xfs_ilock(ip, XFS_ILOCK_EXCL);
1274 xfs_trans_ijoin(tp, ip,
1277 xfs_trans_ihold(tp, ip);
1279 error = xfs_itruncate_finish(&tp, ip,
1284 * If we get an error at this point we
1285 * simply don't bother truncating the file.
1288 xfs_trans_cancel(tp,
1289 (XFS_TRANS_RELEASE_LOG_RES |
1292 error = xfs_trans_commit(tp,
1293 XFS_TRANS_RELEASE_LOG_RES,
1296 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1302 * Free a symlink that has blocks associated with it.
1305 xfs_inactive_symlink_rmt(
1313 xfs_fsblock_t first_block;
1314 xfs_bmap_free_t free_list;
1317 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1325 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1327 * We're freeing a symlink that has some
1328 * blocks allocated to it. Free the
1329 * blocks here. We know that we've got
1330 * either 1 or 2 extents and that we can
1331 * free them all in one bunmapi call.
1333 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1334 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1335 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1336 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1337 xfs_trans_cancel(tp, 0);
1342 * Lock the inode, fix the size, and join it to the transaction.
1343 * Hold it so in the normal path, we still have it locked for
1344 * the second transaction. In the error paths we need it
1345 * held so the cancel won't rele it, see below.
1347 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1348 size = (int)ip->i_d.di_size;
1349 ip->i_d.di_size = 0;
1350 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1351 xfs_trans_ihold(tp, ip);
1352 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1354 * Find the block(s) so we can inval and unmap them.
1357 XFS_BMAP_INIT(&free_list, &first_block);
1358 nmaps = sizeof(mval) / sizeof(mval[0]);
1359 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1360 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1364 * Invalidate the block(s).
1366 for (i = 0; i < nmaps; i++) {
1367 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1368 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1369 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1370 xfs_trans_binval(tp, bp);
1373 * Unmap the dead block(s) to the free_list.
1375 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1376 &first_block, &free_list, &done)))
1380 * Commit the first transaction. This logs the EFI and the inode.
1382 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1385 * The transaction must have been committed, since there were
1386 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1387 * The new tp has the extent freeing and EFDs.
1391 * The first xact was committed, so add the inode to the new one.
1392 * Mark it dirty so it will be logged and moved forward in the log as
1393 * part of every commit.
1395 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1396 xfs_trans_ihold(tp, ip);
1397 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1399 * Get a new, empty transaction to return to our caller.
1401 ntp = xfs_trans_dup(tp);
1403 * Commit the transaction containing extent freeing and EFD's.
1404 * If we get an error on the commit here or on the reserve below,
1405 * we need to unlock the inode since the new transaction doesn't
1406 * have the inode attached.
1408 error = xfs_trans_commit(tp, 0, NULL);
1411 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1415 * Remove the memory for extent descriptions (just bookkeeping).
1417 if (ip->i_df.if_bytes)
1418 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1419 ASSERT(ip->i_df.if_bytes == 0);
1421 * Put an itruncate log reservation in the new transaction
1424 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1425 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1426 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1430 * Return with the inode locked but not joined to the transaction.
1436 xfs_bmap_cancel(&free_list);
1439 * Have to come here with the inode locked and either
1440 * (held and in the transaction) or (not in the transaction).
1441 * If the inode isn't held then cancel would iput it, but
1442 * that's wrong since this is inactive and the vnode ref
1443 * count is 0 already.
1444 * Cancel won't do anything to the inode if held, but it still
1445 * needs to be locked until the cancel is done, if it was
1446 * joined to the transaction.
1448 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1449 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1456 xfs_inactive_symlink_local(
1462 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1464 * We're freeing a symlink which fit into
1465 * the inode. Just free the memory used
1466 * to hold the old symlink.
1468 error = xfs_trans_reserve(*tpp, 0,
1469 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1470 0, XFS_TRANS_PERM_LOG_RES,
1471 XFS_ITRUNCATE_LOG_COUNT);
1474 xfs_trans_cancel(*tpp, 0);
1478 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1481 * Zero length symlinks _can_ exist.
1483 if (ip->i_df.if_bytes > 0) {
1484 xfs_idata_realloc(ip,
1485 -(ip->i_df.if_bytes),
1487 ASSERT(ip->i_df.if_bytes == 0);
1504 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1507 ASSERT(ip->i_d.di_forkoff != 0);
1508 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1509 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1511 error = xfs_attr_inactive(ip);
1514 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1515 return (error); /* goto out*/
1518 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1519 error = xfs_trans_reserve(tp, 0,
1520 XFS_IFREE_LOG_RES(mp),
1521 0, XFS_TRANS_PERM_LOG_RES,
1522 XFS_INACTIVE_LOG_COUNT);
1524 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1525 xfs_trans_cancel(tp, 0);
1527 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1531 xfs_ilock(ip, XFS_ILOCK_EXCL);
1532 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1533 xfs_trans_ihold(tp, ip);
1534 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1536 ASSERT(ip->i_d.di_anextents == 0);
1551 vp = BHV_TO_VNODE(bdp);
1552 ip = XFS_BHVTOI(bdp);
1554 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
1558 /* If this is a read-only mount, don't do this (would generate I/O) */
1559 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1562 #ifdef HAVE_REFCACHE
1563 /* If we are in the NFS reference cache then don't do this now */
1570 if (ip->i_d.di_nlink != 0) {
1571 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1572 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1573 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1574 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1575 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1577 /* Update linux inode block count after free above */
1578 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1579 ip->i_d.di_nblocks + ip->i_delayed_blks);
1589 * This is called when the vnode reference count for the vnode
1590 * goes to zero. If the file has been unlinked, then it must
1591 * now be truncated. Also, we clear all of the read-ahead state
1592 * kept for the inode here since the file is now closed.
1601 xfs_bmap_free_t free_list;
1602 xfs_fsblock_t first_block;
1609 vp = BHV_TO_VNODE(bdp);
1610 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1612 ip = XFS_BHVTOI(bdp);
1615 * If the inode is already free, then there can be nothing
1618 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1619 ASSERT(ip->i_df.if_real_bytes == 0);
1620 ASSERT(ip->i_df.if_broot_bytes == 0);
1621 return VN_INACTIVE_CACHE;
1625 * Only do a truncate if it's a regular file with
1626 * some actual space in it. It's OK to look at the
1627 * inode's fields without the lock because we're the
1628 * only one with a reference to the inode.
1630 truncate = ((ip->i_d.di_nlink == 0) &&
1631 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1632 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1636 if (ip->i_d.di_nlink == 0 &&
1637 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1638 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1643 /* If this is a read-only mount, don't do this (would generate I/O) */
1644 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1647 if (ip->i_d.di_nlink != 0) {
1648 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1649 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1650 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1651 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1652 (ip->i_delayed_blks != 0))) {
1653 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1654 return (VN_INACTIVE_CACHE);
1655 /* Update linux inode block count after free above */
1656 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1657 ip->i_d.di_nblocks + ip->i_delayed_blks);
1662 ASSERT(ip->i_d.di_nlink == 0);
1664 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1665 return (VN_INACTIVE_CACHE);
1667 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1670 * Do the xfs_itruncate_start() call before
1671 * reserving any log space because itruncate_start
1672 * will call into the buffer cache and we can't
1673 * do that within a transaction.
1675 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1677 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1679 error = xfs_trans_reserve(tp, 0,
1680 XFS_ITRUNCATE_LOG_RES(mp),
1681 0, XFS_TRANS_PERM_LOG_RES,
1682 XFS_ITRUNCATE_LOG_COUNT);
1684 /* Don't call itruncate_cleanup */
1685 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1686 xfs_trans_cancel(tp, 0);
1687 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1688 return (VN_INACTIVE_CACHE);
1691 xfs_ilock(ip, XFS_ILOCK_EXCL);
1692 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1693 xfs_trans_ihold(tp, ip);
1696 * normally, we have to run xfs_itruncate_finish sync.
1697 * But if filesystem is wsync and we're in the inactive
1698 * path, then we know that nlink == 0, and that the
1699 * xaction that made nlink == 0 is permanently committed
1700 * since xfs_remove runs as a synchronous transaction.
1702 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1703 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1706 xfs_trans_cancel(tp,
1707 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1708 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1709 return (VN_INACTIVE_CACHE);
1711 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1714 * If we get an error while cleaning up a
1715 * symlink we bail out.
1717 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1718 xfs_inactive_symlink_rmt(ip, &tp) :
1719 xfs_inactive_symlink_local(ip, &tp);
1723 return (VN_INACTIVE_CACHE);
1726 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1727 xfs_trans_ihold(tp, ip);
1729 error = xfs_trans_reserve(tp, 0,
1730 XFS_IFREE_LOG_RES(mp),
1731 0, XFS_TRANS_PERM_LOG_RES,
1732 XFS_INACTIVE_LOG_COUNT);
1734 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1735 xfs_trans_cancel(tp, 0);
1736 return (VN_INACTIVE_CACHE);
1739 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1740 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1741 xfs_trans_ihold(tp, ip);
1745 * If there are attributes associated with the file
1746 * then blow them away now. The code calls a routine
1747 * that recursively deconstructs the attribute fork.
1748 * We need to just commit the current transaction
1749 * because we can't use it for xfs_attr_inactive().
1751 if (ip->i_d.di_anextents > 0) {
1752 error = xfs_inactive_attrs(ip, &tp);
1754 * If we got an error, the transaction is already
1755 * cancelled, and the inode is unlocked. Just get out.
1758 return (VN_INACTIVE_CACHE);
1759 } else if (ip->i_afp) {
1760 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1766 XFS_BMAP_INIT(&free_list, &first_block);
1767 error = xfs_ifree(tp, ip, &free_list);
1770 * If we fail to free the inode, shut down. The cancel
1771 * might do that, we need to make sure. Otherwise the
1772 * inode might be lost for a long time or forever.
1774 if (!XFS_FORCED_SHUTDOWN(mp)) {
1776 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1777 error, mp->m_fsname);
1778 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1780 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1783 * Credit the quota account(s). The inode is gone.
1785 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1788 * Just ignore errors at this point. There is
1789 * nothing we can do except to try to keep going.
1791 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1793 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1796 * Release the dquots held by inode, if any.
1798 XFS_QM_DQDETACH(mp, ip);
1800 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1803 return VN_INACTIVE_CACHE;
1812 bhv_desc_t *dir_bdp,
1819 xfs_inode_t *dp, *ip;
1825 dir_vp = BHV_TO_VNODE(dir_bdp);
1826 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1828 dp = XFS_BHVTOI(dir_bdp);
1830 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1831 return XFS_ERROR(EIO);
1833 lock_mode = xfs_ilock_map_shared(dp);
1834 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1836 *vpp = XFS_ITOV(ip);
1839 xfs_iunlock_map_shared(dp, lock_mode);
1845 * xfs_create (create a new file).
1849 bhv_desc_t *dir_bdp,
1855 char *name = VNAME(dentry);
1857 xfs_inode_t *dp, *ip;
1863 xfs_bmap_free_t free_list;
1864 xfs_fsblock_t first_block;
1865 boolean_t dp_joined_to_trans;
1866 int dm_event_sent = 0;
1870 struct xfs_dquot *udqp, *gdqp;
1876 dir_vp = BHV_TO_VNODE(dir_bdp);
1877 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1879 dp = XFS_BHVTOI(dir_bdp);
1882 dm_di_mode = vap->va_mode;
1883 namelen = VNAMELEN(dentry);
1885 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1886 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1887 dir_vp, DM_RIGHT_NULL, NULL,
1888 DM_RIGHT_NULL, name, NULL,
1896 if (XFS_FORCED_SHUTDOWN(mp))
1897 return XFS_ERROR(EIO);
1899 /* Return through std_return after this point. */
1902 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1903 prid = dp->i_d.di_projid;
1904 else if (vap->va_mask & XFS_AT_PROJID)
1905 prid = (xfs_prid_t)vap->va_projid;
1907 prid = (xfs_prid_t)dfltprid;
1910 * Make sure that we have allocated dquot(s) on disk.
1912 error = XFS_QM_DQVOPALLOC(mp, dp,
1913 current_fsuid(credp), current_fsgid(credp), prid,
1914 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1919 dp_joined_to_trans = B_FALSE;
1921 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1922 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1923 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1925 * Initially assume that the file does not exist and
1926 * reserve the resources for that case. If that is not
1927 * the case we'll drop the one we have and get a more
1928 * appropriate transaction later.
1930 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1931 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1932 if (error == ENOSPC) {
1934 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1935 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1943 xfs_ilock(dp, XFS_ILOCK_EXCL);
1945 XFS_BMAP_INIT(&free_list, &first_block);
1950 * Reserve disk quota and the inode.
1952 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1957 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1959 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1960 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
1961 rdev, credp, prid, resblks > 0,
1964 if (error == ENOSPC)
1971 * At this point, we've gotten a newly allocated inode.
1972 * It is locked (and joined to the transaction).
1975 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1978 * Now we join the directory inode to the transaction.
1979 * We do not do it earlier because xfs_dir_ialloc
1980 * might commit the previous transaction (and release
1985 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1986 dp_joined_to_trans = B_TRUE;
1988 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1989 &first_block, &free_list,
1990 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1992 ASSERT(error != ENOSPC);
1995 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1996 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1999 * If this is a synchronous mount, make sure that the
2000 * create transaction goes to disk before returning to
2003 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2004 xfs_trans_set_sync(tp);
2010 * Attach the dquot(s) to the inodes and modify them incore.
2011 * These ids of the inode couldn't have changed since the new
2012 * inode has been locked ever since it was created.
2014 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2017 * xfs_trans_commit normally decrements the vnode ref count
2018 * when it unlocks the inode. Since we want to return the
2019 * vnode to the caller, we bump the vnode ref count now.
2024 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2026 xfs_bmap_cancel(&free_list);
2030 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2037 XFS_QM_DQRELE(mp, udqp);
2038 XFS_QM_DQRELE(mp, gdqp);
2041 * Propogate the fact that the vnode changed after the
2042 * xfs_inode locks have been released.
2044 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2048 /* Fallthrough to std_return with error = 0 */
2051 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2052 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2053 DM_EVENT_POSTCREATE)) {
2054 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2055 dir_vp, DM_RIGHT_NULL,
2057 DM_RIGHT_NULL, name, NULL,
2058 dm_di_mode, error, 0);
2063 cancel_flags |= XFS_TRANS_ABORT;
2068 xfs_trans_cancel(tp, cancel_flags);
2070 if (!dp_joined_to_trans && (dp != NULL))
2071 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2072 XFS_QM_DQRELE(mp, udqp);
2073 XFS_QM_DQRELE(mp, gdqp);
2079 * Wait until after the current transaction is aborted to
2080 * release the inode. This prevents recursive transactions
2081 * and deadlocks from xfs_inactive.
2083 cancel_flags |= XFS_TRANS_ABORT;
2084 xfs_trans_cancel(tp, cancel_flags);
2087 XFS_QM_DQRELE(mp, udqp);
2088 XFS_QM_DQRELE(mp, gdqp);
2095 * Some counters to see if (and how often) we are hitting some deadlock
2096 * prevention code paths.
2100 int xfs_rm_lock_delays;
2101 int xfs_rm_attempts;
2105 * The following routine will lock the inodes associated with the
2106 * directory and the named entry in the directory. The locks are
2107 * acquired in increasing inode number.
2109 * If the entry is "..", then only the directory is locked. The
2110 * vnode ref count will still include that from the .. entry in
2113 * There is a deadlock we need to worry about. If the locked directory is
2114 * in the AIL, it might be blocking up the log. The next inode we lock
2115 * could be already locked by another thread waiting for log space (e.g
2116 * a permanent log reservation with a long running transaction (see
2117 * xfs_itruncate_finish)). To solve this, we must check if the directory
2118 * is in the ail and use lock_nowait. If we can't lock, we need to
2119 * drop the inode lock on the directory and try again. xfs_iunlock will
2120 * potentially push the tail if we were holding up the log.
2123 xfs_lock_dir_and_entry(
2126 xfs_inode_t *ip) /* inode of entry 'name' */
2130 xfs_inode_t *ips[2];
2139 xfs_ilock(dp, XFS_ILOCK_EXCL);
2146 * We want to lock in increasing inum. Since we've already
2147 * acquired the lock on the directory, we may need to release
2148 * if if the inum of the entry turns out to be less.
2150 if (e_inum > dp->i_ino) {
2152 * We are already in the right order, so just
2153 * lock on the inode of the entry.
2154 * We need to use nowait if dp is in the AIL.
2157 lp = (xfs_log_item_t *)dp->i_itemp;
2158 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2159 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2166 * Unlock dp and try again.
2167 * xfs_iunlock will try to push the tail
2168 * if the inode is in the AIL.
2171 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2173 if ((attempts % 5) == 0) {
2174 delay(1); /* Don't just spin the CPU */
2176 xfs_rm_lock_delays++;
2182 xfs_ilock(ip, XFS_ILOCK_EXCL);
2184 } else if (e_inum < dp->i_ino) {
2185 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2189 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2191 /* else e_inum == dp->i_ino */
2192 /* This can happen if we're asked to lock /x/..
2193 * the entry is "..", which is also the parent directory.
2201 int xfs_small_retries;
2202 int xfs_middle_retries;
2203 int xfs_lots_retries;
2204 int xfs_lock_delays;
2208 * The following routine will lock n inodes in exclusive mode.
2209 * We assume the caller calls us with the inodes in i_ino order.
2211 * We need to detect deadlock where an inode that we lock
2212 * is in the AIL and we start waiting for another inode that is locked
2213 * by a thread in a long running transaction (such as truncate). This can
2214 * result in deadlock since the long running trans might need to wait
2215 * for the inode we just locked in order to push the tail and free space
2225 int attempts = 0, i, j, try_lock;
2228 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2239 for (; i < inodes; i++) {
2242 if (i && (ips[i] == ips[i-1])) /* Already locked */
2246 * If try_lock is not set yet, make sure all locked inodes
2247 * are not in the AIL.
2248 * If any are, set try_lock to be used later.
2252 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2253 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2254 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2261 * If any of the previous locks we have locked is in the AIL,
2262 * we must TRY to get the second and subsequent locks. If
2263 * we can't get any, we must release all we have
2268 /* try_lock must be 0 if i is 0. */
2270 * try_lock means we have an inode locked
2271 * that is in the AIL.
2274 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2278 * Unlock all previous guys and try again.
2279 * xfs_iunlock will try to push the tail
2280 * if the inode is in the AIL.
2283 for(j = i - 1; j >= 0; j--) {
2286 * Check to see if we've already
2287 * unlocked this one.
2288 * Not the first one going back,
2289 * and the inode ptr is the same.
2291 if ((j != (i - 1)) && ips[j] ==
2295 xfs_iunlock(ips[j], lock_mode);
2298 if ((attempts % 5) == 0) {
2299 delay(1); /* Don't just spin the CPU */
2309 xfs_ilock(ips[i], lock_mode);
2315 if (attempts < 5) xfs_small_retries++;
2316 else if (attempts < 100) xfs_middle_retries++;
2317 else xfs_lots_retries++;
2325 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2326 int remove_which_error_return = 0;
2328 #define REMOVE_DEBUG_TRACE(x)
2329 #endif /* ! DEBUG */
2338 bhv_desc_t *dir_bdp,
2343 char *name = VNAME(dentry);
2344 xfs_inode_t *dp, *ip;
2345 xfs_trans_t *tp = NULL;
2348 xfs_bmap_free_t free_list;
2349 xfs_fsblock_t first_block;
2357 dir_vp = BHV_TO_VNODE(dir_bdp);
2358 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2360 dp = XFS_BHVTOI(dir_bdp);
2363 if (XFS_FORCED_SHUTDOWN(mp))
2364 return XFS_ERROR(EIO);
2366 namelen = VNAMELEN(dentry);
2368 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2369 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2370 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2371 name, NULL, 0, 0, 0);
2376 /* From this point on, return through std_return */
2380 * We need to get a reference to ip before we get our log
2381 * reservation. The reason for this is that we cannot call
2382 * xfs_iget for an inode for which we do not have a reference
2383 * once we've acquired a log reservation. This is because the
2384 * inode we are trying to get might be in xfs_inactive going
2385 * for a log reservation. Since we'll have to wait for the
2386 * inactive code to complete before returning from xfs_iget,
2387 * we need to make sure that we don't have log space reserved
2388 * when we call xfs_iget. Instead we get an unlocked referece
2389 * to the inode before getting our log reservation.
2391 error = xfs_get_dir_entry(dentry, &ip);
2393 REMOVE_DEBUG_TRACE(__LINE__);
2397 dm_di_mode = ip->i_d.di_mode;
2399 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2403 error = XFS_QM_DQATTACH(mp, dp, 0);
2404 if (!error && dp != ip)
2405 error = XFS_QM_DQATTACH(mp, ip, 0);
2407 REMOVE_DEBUG_TRACE(__LINE__);
2412 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2413 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2415 * We try to get the real space reservation first,
2416 * allowing for directory btree deletion(s) implying
2417 * possible bmap insert(s). If we can't get the space
2418 * reservation then we use 0 instead, and avoid the bmap
2419 * btree insert(s) in the directory code by, if the bmap
2420 * insert tries to happen, instead trimming the LAST
2421 * block from the directory.
2423 resblks = XFS_REMOVE_SPACE_RES(mp);
2424 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2425 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2426 if (error == ENOSPC) {
2428 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2429 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2432 ASSERT(error != ENOSPC);
2433 REMOVE_DEBUG_TRACE(__LINE__);
2434 xfs_trans_cancel(tp, 0);
2439 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2441 REMOVE_DEBUG_TRACE(__LINE__);
2442 xfs_trans_cancel(tp, cancel_flags);
2448 * At this point, we've gotten both the directory and the entry
2451 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2454 * Increment vnode ref count only in this case since
2455 * there's an extra vnode reference in the case where
2459 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2463 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2465 XFS_BMAP_INIT(&free_list, &first_block);
2466 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2467 &first_block, &free_list, 0);
2469 ASSERT(error != ENOENT);
2470 REMOVE_DEBUG_TRACE(__LINE__);
2473 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2476 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2478 error = xfs_droplink(tp, ip);
2480 REMOVE_DEBUG_TRACE(__LINE__);
2484 /* Determine if this is the last link while
2485 * we are in the transaction.
2487 link_zero = (ip)->i_d.di_nlink==0;
2490 * Take an extra ref on the inode so that it doesn't
2491 * go to xfs_inactive() from within the commit.
2496 * If this is a synchronous mount, make sure that the
2497 * remove transaction goes to disk before returning to
2500 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2501 xfs_trans_set_sync(tp);
2504 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2506 REMOVE_DEBUG_TRACE(__LINE__);
2510 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2517 * Before we drop our extra reference to the inode, purge it
2518 * from the refcache if it is there. By waiting until afterwards
2519 * to do the IRELE, we ensure that we won't go inactive in the
2520 * xfs_refcache_purge_ip routine (although that would be OK).
2522 xfs_refcache_purge_ip(ip);
2524 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2527 * Let interposed file systems know about removed links.
2529 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2533 /* Fall through to std_return with error = 0 */
2535 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2536 DM_EVENT_POSTREMOVE)) {
2537 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2538 dir_vp, DM_RIGHT_NULL,
2539 NULL, DM_RIGHT_NULL,
2540 name, NULL, dm_di_mode, error, 0);
2545 xfs_bmap_cancel(&free_list);
2546 cancel_flags |= XFS_TRANS_ABORT;
2547 xfs_trans_cancel(tp, cancel_flags);
2552 * In this case make sure to not release the inode until after
2553 * the current transaction is aborted. Releasing it beforehand
2554 * can cause us to go to xfs_inactive and start a recursive
2555 * transaction which can easily deadlock with the current one.
2557 xfs_bmap_cancel(&free_list);
2558 cancel_flags |= XFS_TRANS_ABORT;
2559 xfs_trans_cancel(tp, cancel_flags);
2562 * Before we drop our extra reference to the inode, purge it
2563 * from the refcache if it is there. By waiting until afterwards
2564 * to do the IRELE, we ensure that we won't go inactive in the
2565 * xfs_refcache_purge_ip routine (although that would be OK).
2567 xfs_refcache_purge_ip(ip);
2581 bhv_desc_t *target_dir_bdp,
2586 xfs_inode_t *tdp, *sip;
2589 xfs_inode_t *ips[2];
2591 xfs_bmap_free_t free_list;
2592 xfs_fsblock_t first_block;
2595 vnode_t *target_dir_vp;
2596 bhv_desc_t *src_bdp;
2598 char *target_name = VNAME(dentry);
2601 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2602 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2603 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2605 target_namelen = VNAMELEN(dentry);
2606 if (VN_ISDIR(src_vp))
2607 return XFS_ERROR(EPERM);
2609 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2610 sip = XFS_BHVTOI(src_bdp);
2611 tdp = XFS_BHVTOI(target_dir_bdp);
2613 if (XFS_FORCED_SHUTDOWN(mp))
2614 return XFS_ERROR(EIO);
2616 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2617 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2618 target_dir_vp, DM_RIGHT_NULL,
2619 src_vp, DM_RIGHT_NULL,
2620 target_name, NULL, 0, 0, 0);
2625 /* Return through std_return after this point. */
2627 error = XFS_QM_DQATTACH(mp, sip, 0);
2628 if (!error && sip != tdp)
2629 error = XFS_QM_DQATTACH(mp, tdp, 0);
2633 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2634 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2635 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2636 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2637 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2638 if (error == ENOSPC) {
2640 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2641 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2648 if (sip->i_ino < tdp->i_ino) {
2656 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2659 * Increment vnode ref counts since xfs_trans_commit &
2660 * xfs_trans_cancel will both unlock the inodes and
2661 * decrement the associated ref counts.
2664 VN_HOLD(target_dir_vp);
2665 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2666 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2669 * If the source has too many links, we can't make any more to it.
2671 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2672 error = XFS_ERROR(EMLINK);
2677 * If we are using project inheritance, we only allow hard link
2678 * creation in our tree when the project IDs are the same; else
2679 * the tree quota mechanism could be circumvented.
2681 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2682 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2683 error = XFS_ERROR(EPERM);
2688 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2692 XFS_BMAP_INIT(&free_list, &first_block);
2694 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2695 sip->i_ino, &first_block, &free_list,
2699 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2701 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2703 error = xfs_bumplink(tp, sip);
2709 * If this is a synchronous mount, make sure that the
2710 * link transaction goes to disk before returning to
2713 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2714 xfs_trans_set_sync(tp);
2717 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2719 xfs_bmap_cancel(&free_list);
2723 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2728 /* Fall through to std_return with error = 0. */
2730 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2731 DM_EVENT_POSTLINK)) {
2732 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2733 target_dir_vp, DM_RIGHT_NULL,
2734 src_vp, DM_RIGHT_NULL,
2735 target_name, NULL, 0, error, 0);
2740 cancel_flags |= XFS_TRANS_ABORT;
2743 xfs_trans_cancel(tp, cancel_flags);
2753 bhv_desc_t *dir_bdp,
2759 char *dir_name = VNAME(dentry);
2761 xfs_inode_t *cdp; /* inode of created dir */
2762 vnode_t *cvp; /* vnode of created dir */
2768 xfs_bmap_free_t free_list;
2769 xfs_fsblock_t first_block;
2771 boolean_t dp_joined_to_trans;
2772 boolean_t created = B_FALSE;
2773 int dm_event_sent = 0;
2775 struct xfs_dquot *udqp, *gdqp;
2780 dir_vp = BHV_TO_VNODE(dir_bdp);
2781 dp = XFS_BHVTOI(dir_bdp);
2784 if (XFS_FORCED_SHUTDOWN(mp))
2785 return XFS_ERROR(EIO);
2787 dir_namelen = VNAMELEN(dentry);
2790 dp_joined_to_trans = B_FALSE;
2791 dm_di_mode = vap->va_mode;
2793 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2794 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2795 dir_vp, DM_RIGHT_NULL, NULL,
2796 DM_RIGHT_NULL, dir_name, NULL,
2803 /* Return through std_return after this point. */
2805 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2809 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2810 prid = dp->i_d.di_projid;
2811 else if (vap->va_mask & XFS_AT_PROJID)
2812 prid = (xfs_prid_t)vap->va_projid;
2814 prid = (xfs_prid_t)dfltprid;
2817 * Make sure that we have allocated dquot(s) on disk.
2819 error = XFS_QM_DQVOPALLOC(mp, dp,
2820 current_fsuid(credp), current_fsgid(credp), prid,
2821 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2825 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2826 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2827 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2828 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2829 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2830 if (error == ENOSPC) {
2832 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2833 XFS_TRANS_PERM_LOG_RES,
2834 XFS_MKDIR_LOG_COUNT);
2842 xfs_ilock(dp, XFS_ILOCK_EXCL);
2845 * Check for directory link count overflow.
2847 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2848 error = XFS_ERROR(EMLINK);
2853 * Reserve disk quota and the inode.
2855 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2860 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2863 * create the directory inode.
2865 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
2866 0, credp, prid, resblks > 0,
2869 if (error == ENOSPC)
2876 * Now we add the directory inode to the transaction.
2877 * We waited until now since xfs_dir_ialloc might start
2878 * a new transaction. Had we joined the transaction
2879 * earlier, the locks might have gotten released.
2882 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2883 dp_joined_to_trans = B_TRUE;
2885 XFS_BMAP_INIT(&free_list, &first_block);
2887 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2888 cdp->i_ino, &first_block, &free_list,
2889 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2891 ASSERT(error != ENOSPC);
2894 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2897 * Bump the in memory version number of the parent directory
2898 * so that other processes accessing it will recognize that
2899 * the directory has changed.
2903 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2909 error = xfs_bumplink(tp, dp);
2914 cvp = XFS_ITOV(cdp);
2922 * Attach the dquots to the new inode and modify the icount incore.
2924 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2927 * If this is a synchronous mount, make sure that the
2928 * mkdir transaction goes to disk before returning to
2931 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2932 xfs_trans_set_sync(tp);
2935 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2941 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2942 XFS_QM_DQRELE(mp, udqp);
2943 XFS_QM_DQRELE(mp, gdqp);
2948 /* Fall through to std_return with error = 0 or errno from
2949 * xfs_trans_commit. */
2952 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2953 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2954 DM_EVENT_POSTCREATE)) {
2955 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2956 dir_vp, DM_RIGHT_NULL,
2957 created ? XFS_ITOV(cdp):NULL,
2960 dm_di_mode, error, 0);
2966 xfs_bmap_cancel(&free_list);
2968 cancel_flags |= XFS_TRANS_ABORT;
2970 xfs_trans_cancel(tp, cancel_flags);
2971 XFS_QM_DQRELE(mp, udqp);
2972 XFS_QM_DQRELE(mp, gdqp);
2974 if (!dp_joined_to_trans && (dp != NULL)) {
2975 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2988 bhv_desc_t *dir_bdp,
2992 char *name = VNAME(dentry);
2994 xfs_inode_t *cdp; /* child directory */
2998 xfs_bmap_free_t free_list;
2999 xfs_fsblock_t first_block;
3008 dir_vp = BHV_TO_VNODE(dir_bdp);
3009 dp = XFS_BHVTOI(dir_bdp);
3012 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3014 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3015 return XFS_ERROR(EIO);
3016 namelen = VNAMELEN(dentry);
3018 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3019 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3020 dir_vp, DM_RIGHT_NULL,
3021 NULL, DM_RIGHT_NULL,
3022 name, NULL, 0, 0, 0);
3024 return XFS_ERROR(error);
3027 /* Return through std_return after this point. */
3032 * We need to get a reference to cdp before we get our log
3033 * reservation. The reason for this is that we cannot call
3034 * xfs_iget for an inode for which we do not have a reference
3035 * once we've acquired a log reservation. This is because the
3036 * inode we are trying to get might be in xfs_inactive going
3037 * for a log reservation. Since we'll have to wait for the
3038 * inactive code to complete before returning from xfs_iget,
3039 * we need to make sure that we don't have log space reserved
3040 * when we call xfs_iget. Instead we get an unlocked referece
3041 * to the inode before getting our log reservation.
3043 error = xfs_get_dir_entry(dentry, &cdp);
3045 REMOVE_DEBUG_TRACE(__LINE__);
3049 dm_di_mode = cdp->i_d.di_mode;
3052 * Get the dquots for the inodes.
3054 error = XFS_QM_DQATTACH(mp, dp, 0);
3055 if (!error && dp != cdp)
3056 error = XFS_QM_DQATTACH(mp, cdp, 0);
3059 REMOVE_DEBUG_TRACE(__LINE__);
3063 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3064 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3066 * We try to get the real space reservation first,
3067 * allowing for directory btree deletion(s) implying
3068 * possible bmap insert(s). If we can't get the space
3069 * reservation then we use 0 instead, and avoid the bmap
3070 * btree insert(s) in the directory code by, if the bmap
3071 * insert tries to happen, instead trimming the LAST
3072 * block from the directory.
3074 resblks = XFS_REMOVE_SPACE_RES(mp);
3075 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3076 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3077 if (error == ENOSPC) {
3079 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3080 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3083 ASSERT(error != ENOSPC);
3088 XFS_BMAP_INIT(&free_list, &first_block);
3091 * Now lock the child directory inode and the parent directory
3092 * inode in the proper order. This will take care of validating
3093 * that the directory entry for the child directory inode has
3094 * not changed while we were obtaining a log reservation.
3096 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3098 xfs_trans_cancel(tp, cancel_flags);
3103 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3106 * Only increment the parent directory vnode count if
3107 * we didn't bump it in looking up cdp. The only time
3108 * we don't bump it is when we're looking up ".".
3114 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3116 ASSERT(cdp->i_d.di_nlink >= 2);
3117 if (cdp->i_d.di_nlink != 2) {
3118 error = XFS_ERROR(ENOTEMPTY);
3121 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3122 error = XFS_ERROR(ENOTEMPTY);
3126 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3127 &first_block, &free_list, resblks);
3132 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3135 * Bump the in memory generation count on the parent
3136 * directory so that other can know that it has changed.
3141 * Drop the link from cdp's "..".
3143 error = xfs_droplink(tp, dp);
3149 * Drop the link from dp to cdp.
3151 error = xfs_droplink(tp, cdp);
3157 * Drop the "." link from cdp to self.
3159 error = xfs_droplink(tp, cdp);
3164 /* Determine these before committing transaction */
3165 last_cdp_link = (cdp)->i_d.di_nlink==0;
3168 * Take an extra ref on the child vnode so that it
3169 * does not go to xfs_inactive() from within the commit.
3174 * If this is a synchronous mount, make sure that the
3175 * rmdir transaction goes to disk before returning to
3178 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3179 xfs_trans_set_sync(tp);
3182 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3184 xfs_bmap_cancel(&free_list);
3185 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3191 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3199 * Let interposed file systems know about removed links.
3201 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3205 /* Fall through to std_return with error = 0 or the errno
3206 * from xfs_trans_commit. */
3208 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3209 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3210 dir_vp, DM_RIGHT_NULL,
3211 NULL, DM_RIGHT_NULL,
3212 name, NULL, dm_di_mode,
3218 xfs_bmap_cancel(&free_list);
3219 cancel_flags |= XFS_TRANS_ABORT;
3221 xfs_trans_cancel(tp, cancel_flags);
3229 * Read dp's entries starting at uiop->uio_offset and translate them into
3230 * bufsize bytes worth of struct dirents starting at bufbase.
3234 bhv_desc_t *dir_bdp,
3240 xfs_trans_t *tp = NULL;
3243 xfs_off_t start_offset;
3245 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3246 (inst_t *)__return_address);
3247 dp = XFS_BHVTOI(dir_bdp);
3249 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3250 return XFS_ERROR(EIO);
3253 lock_mode = xfs_ilock_map_shared(dp);
3254 start_offset = uiop->uio_offset;
3255 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3256 if (start_offset != uiop->uio_offset) {
3257 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3259 xfs_iunlock_map_shared(dp, lock_mode);
3270 bhv_desc_t *dir_bdp,
3283 xfs_bmap_free_t free_list;
3284 xfs_fsblock_t first_block;
3285 boolean_t dp_joined_to_trans;
3289 xfs_fileoff_t first_fsb;
3290 xfs_filblks_t fs_blocks;
3292 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3299 struct xfs_dquot *udqp, *gdqp;
3301 char *link_name = VNAME(dentry);
3305 dir_vp = BHV_TO_VNODE(dir_bdp);
3306 dp = XFS_BHVTOI(dir_bdp);
3307 dp_joined_to_trans = B_FALSE;
3312 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3316 if (XFS_FORCED_SHUTDOWN(mp))
3317 return XFS_ERROR(EIO);
3319 link_namelen = VNAMELEN(dentry);
3322 * Check component lengths of the target path name.
3324 pathlen = strlen(target_path);
3325 if (pathlen >= MAXPATHLEN) /* total string too long */
3326 return XFS_ERROR(ENAMETOOLONG);
3327 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3331 for(total = 0, path = target_path; total < pathlen;) {
3335 while(*path == '/') {
3341 * Count up to the next slash or end of path.
3342 * Error out if the component is bigger than MAXNAMELEN.
3344 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3345 if (++len >= MAXNAMELEN) {
3346 error = ENAMETOOLONG;
3353 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3354 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3355 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3356 link_name, target_path, 0, 0, 0);
3361 /* Return through std_return after this point. */
3364 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3365 prid = dp->i_d.di_projid;
3366 else if (vap->va_mask & XFS_AT_PROJID)
3367 prid = (xfs_prid_t)vap->va_projid;
3369 prid = (xfs_prid_t)dfltprid;
3372 * Make sure that we have allocated dquot(s) on disk.
3374 error = XFS_QM_DQVOPALLOC(mp, dp,
3375 current_fsuid(credp), current_fsgid(credp), prid,
3376 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3380 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3381 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3383 * The symlink will fit into the inode data fork?
3384 * There can't be any attributes so we get the whole variable part.
3386 if (pathlen <= XFS_LITINO(mp))
3389 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3390 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3391 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3392 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3393 if (error == ENOSPC && fs_blocks == 0) {
3395 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3396 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3404 xfs_ilock(dp, XFS_ILOCK_EXCL);
3407 * Check whether the directory allows new symlinks or not.
3409 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3410 error = XFS_ERROR(EPERM);
3415 * Reserve disk quota : blocks and inode.
3417 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3422 * Check for ability to enter directory entry, if no space reserved.
3425 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3428 * Initialize the bmap freelist prior to calling either
3429 * bmapi or the directory create code.
3431 XFS_BMAP_INIT(&free_list, &first_block);
3434 * Allocate an inode for the symlink.
3436 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3437 1, 0, credp, prid, resblks > 0, &ip, NULL);
3439 if (error == ENOSPC)
3446 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3447 dp_joined_to_trans = B_TRUE;
3450 * Also attach the dquot(s) to it, if applicable.
3452 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3455 resblks -= XFS_IALLOC_SPACE_RES(mp);
3457 * If the symlink will fit into the inode, write it inline.
3459 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3460 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3461 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3462 ip->i_d.di_size = pathlen;
3465 * The inode was initially created in extent format.
3467 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3468 ip->i_df.if_flags |= XFS_IFINLINE;
3470 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3471 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3475 nmaps = SYMLINK_MAPS;
3477 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3478 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3479 &first_block, resblks, mval, &nmaps,
3486 resblks -= fs_blocks;
3487 ip->i_d.di_size = pathlen;
3488 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3490 cur_chunk = target_path;
3491 for (n = 0; n < nmaps; n++) {
3492 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3493 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3494 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3495 BTOBB(byte_cnt), 0);
3496 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3497 if (pathlen < byte_cnt) {
3500 pathlen -= byte_cnt;
3502 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3503 cur_chunk += byte_cnt;
3505 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3510 * Create the directory entry for the symlink.
3512 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3513 ip->i_ino, &first_block, &free_list, resblks);
3517 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3518 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3521 * Bump the in memory version number of the parent directory
3522 * so that other processes accessing it will recognize that
3523 * the directory has changed.
3528 * If this is a synchronous mount, make sure that the
3529 * symlink transaction goes to disk before returning to
3532 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3533 xfs_trans_set_sync(tp);
3537 * xfs_trans_commit normally decrements the vnode ref count
3538 * when it unlocks the inode. Since we want to return the
3539 * vnode to the caller, we bump the vnode ref count now.
3543 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3547 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3548 XFS_QM_DQRELE(mp, udqp);
3549 XFS_QM_DQRELE(mp, gdqp);
3551 /* Fall through to std_return with error = 0 or errno from
3552 * xfs_trans_commit */
3554 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3555 DM_EVENT_POSTSYMLINK)) {
3556 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3557 dir_vp, DM_RIGHT_NULL,
3558 error ? NULL : XFS_ITOV(ip),
3559 DM_RIGHT_NULL, link_name, target_path,
3575 xfs_bmap_cancel(&free_list);
3576 cancel_flags |= XFS_TRANS_ABORT;
3578 xfs_trans_cancel(tp, cancel_flags);
3579 XFS_QM_DQRELE(mp, udqp);
3580 XFS_QM_DQRELE(mp, gdqp);
3582 if (!dp_joined_to_trans && (dp != NULL)) {
3583 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3593 * A fid routine that takes a pointer to a previously allocated
3594 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3604 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3605 (inst_t *)__return_address);
3606 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3608 xfid = (xfs_fid2_t *)fidp;
3609 ip = XFS_BHVTOI(bdp);
3610 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3613 * use memcpy because the inode is a long long and there's no
3614 * assurance that xfid->fid_ino is properly aligned.
3616 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3617 xfid->fid_gen = ip->i_d.di_gen;
3634 vp = BHV_TO_VNODE(bdp);
3637 ip = XFS_BHVTOI(bdp);
3638 if (locktype == VRWLOCK_WRITE) {
3639 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3640 } else if (locktype == VRWLOCK_TRY_READ) {
3641 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3642 } else if (locktype == VRWLOCK_TRY_WRITE) {
3643 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3645 ASSERT((locktype == VRWLOCK_READ) ||
3646 (locktype == VRWLOCK_WRITE_DIRECT));
3647 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3665 vp = BHV_TO_VNODE(bdp);
3668 ip = XFS_BHVTOI(bdp);
3669 if (locktype == VRWLOCK_WRITE) {
3671 * In the write case, we may have added a new entry to
3672 * the reference cache. This might store a pointer to
3673 * an inode to be released in this inode. If it is there,
3674 * clear the pointer and release the inode after unlocking
3677 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3679 ASSERT((locktype == VRWLOCK_READ) ||
3680 (locktype == VRWLOCK_WRITE_DIRECT));
3681 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3693 xfs_inode_log_item_t *iip;
3696 ip = XFS_BHVTOI(bdp);
3700 if (XFS_FORCED_SHUTDOWN(mp))
3701 return XFS_ERROR(EIO);
3704 * Bypass inodes which have already been cleaned by
3705 * the inode flush clustering code inside xfs_iflush
3707 if ((ip->i_update_core == 0) &&
3708 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3711 if (flags & FLUSH_LOG) {
3712 if (iip && iip->ili_last_lsn) {
3713 xlog_t *log = mp->m_log;
3715 int s, log_flags = XFS_LOG_FORCE;
3717 s = GRANT_LOCK(log);
3718 sync_lsn = log->l_last_sync_lsn;
3719 GRANT_UNLOCK(log, s);
3721 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3724 if (flags & FLUSH_SYNC)
3725 log_flags |= XFS_LOG_SYNC;
3726 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3731 * We make this non-blocking if the inode is contended,
3732 * return EAGAIN to indicate to the caller that they
3733 * did not succeed. This prevents the flush path from
3734 * blocking on inodes inside another operation right
3735 * now, they get caught later by xfs_sync.
3737 if (flags & FLUSH_INODE) {
3740 if (xfs_ipincount(ip))
3743 if (flags & FLUSH_SYNC) {
3744 xfs_ilock(ip, XFS_ILOCK_SHARED);
3746 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3747 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3748 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3755 if (flags & FLUSH_SYNC)
3756 flush_flags = XFS_IFLUSH_SYNC;
3758 flush_flags = XFS_IFLUSH_ASYNC;
3760 error = xfs_iflush(ip, flush_flags);
3761 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3780 if (!capable(CAP_SYS_ADMIN))
3781 return XFS_ERROR(EPERM);
3783 ip = XFS_BHVTOI(bdp);
3786 if (XFS_FORCED_SHUTDOWN(mp))
3787 return XFS_ERROR(EIO);
3789 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3790 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3792 xfs_trans_cancel(tp, 0);
3795 xfs_ilock(ip, XFS_ILOCK_EXCL);
3796 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3798 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3799 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3801 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3803 error = xfs_trans_commit(tp, 0, NULL);
3819 vp = BHV_TO_VNODE(bdp);
3820 ip = XFS_BHVTOI(bdp);
3822 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3824 ASSERT(!VN_MAPPED(vp));
3826 /* bad inode, get out here ASAP */
3834 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
3835 ASSERT(VN_CACHED(vp) == 0);
3837 /* If we have nothing to flush with this inode then complete the
3838 * teardown now, otherwise break the link between the xfs inode
3839 * and the linux inode and clean up the xfs inode later. This
3840 * avoids flushing the inode to disk during the delete operation
3843 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3844 xfs_ilock(ip, XFS_ILOCK_EXCL);
3846 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3848 xfs_mount_t *mp = ip->i_mount;
3850 /* Protect sync from us */
3851 XFS_MOUNT_ILOCK(mp);
3852 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3853 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3854 ip->i_flags |= XFS_IRECLAIMABLE;
3855 XFS_MOUNT_IUNLOCK(mp);
3866 xfs_ihash_t *ih = ip->i_hash;
3867 vnode_t *vp = XFS_ITOV_NULL(ip);
3870 if (vp && VN_BAD(vp))
3873 /* The hash lock here protects a thread in xfs_iget_core from
3874 * racing with us on linking the inode back with a vnode.
3875 * Once we have the XFS_IRECLAIM flag set it will not touch
3878 write_lock(&ih->ih_lock);
3879 if ((ip->i_flags & XFS_IRECLAIM) ||
3880 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3881 write_unlock(&ih->ih_lock);
3884 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3888 ip->i_flags |= XFS_IRECLAIM;
3889 write_unlock(&ih->ih_lock);
3892 * If the inode is still dirty, then flush it out. If the inode
3893 * is not in the AIL, then it will be OK to flush it delwri as
3894 * long as xfs_iflush() does not keep any references to the inode.
3895 * We leave that decision up to xfs_iflush() since it has the
3896 * knowledge of whether it's OK to simply do a delwri flush of
3897 * the inode or whether we need to wait until the inode is
3898 * pulled from the AIL.
3899 * We get the flush lock regardless, though, just to make sure
3900 * we don't free it while it is being flushed.
3902 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3904 xfs_ilock(ip, XFS_ILOCK_EXCL);
3908 if (ip->i_update_core ||
3909 ((ip->i_itemp != NULL) &&
3910 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3911 error = xfs_iflush(ip, sync_mode);
3913 * If we hit an error, typically because of filesystem
3914 * shutdown, we don't need to let vn_reclaim to know
3915 * because we're gonna reclaim the inode anyway.
3918 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3921 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3924 ASSERT(ip->i_update_core == 0);
3925 ASSERT(ip->i_itemp == NULL ||
3926 ip->i_itemp->ili_format.ilf_fields == 0);
3927 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3928 } else if (locked) {
3930 * We are not interested in doing an iflush if we're
3931 * in the process of shutting down the filesystem forcibly.
3932 * So, just reclaim the inode.
3935 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3944 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3947 xfs_inode_t *ip, *n;
3952 XFS_MOUNT_ILOCK(mp);
3953 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3955 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3957 if (xfs_ipincount(ip) ||
3958 !xfs_iflock_nowait(ip)) {
3959 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3963 XFS_MOUNT_IUNLOCK(mp);
3964 if (xfs_finish_reclaim(ip, noblock,
3965 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3974 XFS_MOUNT_IUNLOCK(mp);
3979 * xfs_alloc_file_space()
3980 * This routine allocates disk space for the given file.
3982 * If alloc_type == 0, this request is for an ALLOCSP type
3983 * request which will change the file size. In this case, no
3984 * DMAPI event will be generated by the call. A TRUNCATE event
3985 * will be generated later by xfs_setattr.
3987 * If alloc_type != 0, this request is for a RESVSP type
3988 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3989 * lower block boundary byte address is less than the file's
3998 xfs_alloc_file_space(
4005 xfs_filblks_t allocated_fsb;
4006 xfs_filblks_t allocatesize_fsb;
4009 xfs_filblks_t datablocks;
4011 xfs_fsblock_t firstfsb;
4012 xfs_bmap_free_t free_list;
4013 xfs_bmbt_irec_t *imapp;
4014 xfs_bmbt_irec_t imaps[1];
4021 xfs_fileoff_t startoffset_fsb;
4023 int xfs_bmapi_flags;
4025 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4028 if (XFS_FORCED_SHUTDOWN(mp))
4029 return XFS_ERROR(EIO);
4032 * determine if this is a realtime file
4034 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4035 if (ip->i_d.di_extsize)
4036 rtextsize = ip->i_d.di_extsize;
4038 rtextsize = mp->m_sb.sb_rextsize;
4042 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4046 return XFS_ERROR(EINVAL);
4052 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4053 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4054 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4056 /* Generate a DMAPI event if needed. */
4057 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4058 (attr_flags&ATTR_DMI) == 0 &&
4059 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4060 xfs_off_t end_dmi_offset;
4062 end_dmi_offset = offset+len;
4063 if (end_dmi_offset > ip->i_d.di_size)
4064 end_dmi_offset = ip->i_d.di_size;
4065 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4066 offset, end_dmi_offset - offset,
4073 * allocate file space until done or until there is an error
4076 while (allocatesize_fsb && !error) {
4078 * determine if reserving space on
4079 * the data or realtime partition.
4084 s = startoffset_fsb;
4085 do_div(s, rtextsize);
4087 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4089 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4092 datablocks = allocatesize_fsb;
4097 * allocate and setup the transaction
4099 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4100 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4101 error = xfs_trans_reserve(tp,
4103 XFS_WRITE_LOG_RES(mp),
4105 XFS_TRANS_PERM_LOG_RES,
4106 XFS_WRITE_LOG_COUNT);
4109 * check for running out of space
4113 * Free the transaction structure.
4115 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4116 xfs_trans_cancel(tp, 0);
4119 xfs_ilock(ip, XFS_ILOCK_EXCL);
4120 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4121 ip->i_udquot, ip->i_gdquot, resblks, 0, 0);
4125 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4126 xfs_trans_ihold(tp, ip);
4129 * issue the bmapi() call to allocate the blocks
4131 XFS_BMAP_INIT(&free_list, &firstfsb);
4132 error = xfs_bmapi(tp, ip, startoffset_fsb,
4133 allocatesize_fsb, xfs_bmapi_flags,
4134 &firstfsb, 0, imapp, &reccount,
4141 * complete the transaction
4143 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4148 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4149 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4154 allocated_fsb = imapp->br_blockcount;
4156 if (reccount == 0) {
4157 error = XFS_ERROR(ENOSPC);
4161 startoffset_fsb += allocated_fsb;
4162 allocatesize_fsb -= allocated_fsb;
4165 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4166 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4168 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4169 XFS_ITOV(ip), DM_RIGHT_NULL,
4170 XFS_ITOV(ip), DM_RIGHT_NULL,
4171 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4173 goto retry; /* Maybe DMAPI app. has made space */
4174 /* else fall through with error from XFS_SEND_DATA */
4180 xfs_bmap_cancel(&free_list);
4182 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4183 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4184 goto dmapi_enospc_check;
4188 * Zero file bytes between startoff and endoff inclusive.
4189 * The iolock is held exclusive and no blocks are buffered.
4192 xfs_zero_remaining_bytes(
4197 xfs_bmbt_irec_t imap;
4198 xfs_fileoff_t offset_fsb;
4199 xfs_off_t lastoffset;
4202 xfs_mount_t *mp = ip->i_mount;
4206 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4207 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4208 mp->m_rtdev_targp : mp->m_ddev_targp);
4210 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4211 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4213 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4215 if (error || nimap < 1)
4217 ASSERT(imap.br_blockcount >= 1);
4218 ASSERT(imap.br_startoff == offset_fsb);
4219 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4220 if (lastoffset > endoff)
4221 lastoffset = endoff;
4222 if (imap.br_startblock == HOLESTARTBLOCK)
4224 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4225 if (imap.br_state == XFS_EXT_UNWRITTEN)
4228 XFS_BUF_UNWRITE(bp);
4230 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4232 if ((error = xfs_iowait(bp))) {
4233 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4234 mp, bp, XFS_BUF_ADDR(bp));
4237 memset(XFS_BUF_PTR(bp) +
4238 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4239 0, lastoffset - offset + 1);
4244 if ((error = xfs_iowait(bp))) {
4245 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4246 mp, bp, XFS_BUF_ADDR(bp));
4255 * xfs_free_file_space()
4256 * This routine frees disk space for the given file.
4258 * This routine is only called by xfs_change_file_space
4259 * for an UNRESVSP type call.
4267 xfs_free_file_space(
4276 xfs_off_t end_dmi_offset;
4277 xfs_fileoff_t endoffset_fsb;
4279 xfs_fsblock_t firstfsb;
4280 xfs_bmap_free_t free_list;
4282 xfs_bmbt_irec_t imap;
4290 xfs_fileoff_t startoffset_fsb;
4292 int need_iolock = 1;
4297 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4299 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4303 if (len <= 0) /* if nothing being freed */
4305 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4306 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4307 end_dmi_offset = offset + len;
4308 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4310 if (offset < ip->i_d.di_size &&
4311 (attr_flags & ATTR_DMI) == 0 &&
4312 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4313 if (end_dmi_offset > ip->i_d.di_size)
4314 end_dmi_offset = ip->i_d.di_size;
4315 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
4316 offset, end_dmi_offset - offset,
4317 AT_DELAY_FLAG(attr_flags), NULL);
4322 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4323 if (attr_flags & ATTR_NOLOCK)
4326 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4328 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4330 ilen = len + (offset & (rounding - 1));
4331 ioffset = offset & ~(rounding - 1);
4332 if (ilen & (rounding - 1))
4333 ilen = (ilen + rounding) & ~(rounding - 1);
4335 if (VN_CACHED(vp) != 0) {
4336 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4337 ctooff(offtoct(ioffset)), -1);
4338 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4339 -1, FI_REMAPF_LOCKED);
4343 * Need to zero the stuff we're not freeing, on disk.
4344 * If its a realtime file & can't use unwritten extents then we
4345 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4346 * will take care of it for us.
4348 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4350 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4351 &imap, &nimap, NULL);
4353 goto out_unlock_iolock;
4354 ASSERT(nimap == 0 || nimap == 1);
4355 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4358 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4359 block = imap.br_startblock;
4360 mod = do_div(block, mp->m_sb.sb_rextsize);
4362 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4365 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4366 &imap, &nimap, NULL);
4368 goto out_unlock_iolock;
4369 ASSERT(nimap == 0 || nimap == 1);
4370 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4371 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4373 if (mod && (mod != mp->m_sb.sb_rextsize))
4374 endoffset_fsb -= mod;
4377 if ((done = (endoffset_fsb <= startoffset_fsb)))
4379 * One contiguous piece to clear
4381 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4384 * Some full blocks, possibly two pieces to clear
4386 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4387 error = xfs_zero_remaining_bytes(ip, offset,
4388 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4390 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4391 error = xfs_zero_remaining_bytes(ip,
4392 XFS_FSB_TO_B(mp, endoffset_fsb),
4397 * free file space until done or until there is an error
4399 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4400 while (!error && !done) {
4403 * allocate and setup the transaction
4405 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4406 error = xfs_trans_reserve(tp,
4408 XFS_WRITE_LOG_RES(mp),
4410 XFS_TRANS_PERM_LOG_RES,
4411 XFS_WRITE_LOG_COUNT);
4414 * check for running out of space
4418 * Free the transaction structure.
4420 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4421 xfs_trans_cancel(tp, 0);
4424 xfs_ilock(ip, XFS_ILOCK_EXCL);
4425 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4426 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4427 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4431 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4432 xfs_trans_ihold(tp, ip);
4435 * issue the bunmapi() call to free the blocks
4437 XFS_BMAP_INIT(&free_list, &firstfsb);
4438 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4439 endoffset_fsb - startoffset_fsb,
4440 0, 2, &firstfsb, &free_list, &done);
4446 * complete the transaction
4448 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4453 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4454 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4459 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4463 xfs_bmap_cancel(&free_list);
4465 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4466 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4472 * xfs_change_file_space()
4473 * This routine allocates or frees disk space for the given file.
4474 * The user specified parameters are checked for alignment and size
4483 xfs_change_file_space(
4497 xfs_off_t startoffset;
4503 vp = BHV_TO_VNODE(bdp);
4504 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4506 ip = XFS_BHVTOI(bdp);
4510 * must be a regular file and have write permission
4513 return XFS_ERROR(EINVAL);
4515 xfs_ilock(ip, XFS_ILOCK_SHARED);
4517 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4518 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4522 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4524 switch (bf->l_whence) {
4525 case 0: /*SEEK_SET*/
4527 case 1: /*SEEK_CUR*/
4528 bf->l_start += offset;
4530 case 2: /*SEEK_END*/
4531 bf->l_start += ip->i_d.di_size;
4534 return XFS_ERROR(EINVAL);
4537 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4539 if ( (bf->l_start < 0)
4540 || (bf->l_start > XFS_MAXIOFFSET(mp))
4541 || (bf->l_start + llen < 0)
4542 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4543 return XFS_ERROR(EINVAL);
4547 startoffset = bf->l_start;
4548 fsize = ip->i_d.di_size;
4551 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4553 * These calls do NOT zero the data space allocated to the file,
4554 * nor do they change the file size.
4556 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4558 * These calls cause the new file data to be zeroed and the file
4559 * size to be changed.
4561 setprealloc = clrprealloc = 0;
4564 case XFS_IOC_RESVSP:
4565 case XFS_IOC_RESVSP64:
4566 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4573 case XFS_IOC_UNRESVSP:
4574 case XFS_IOC_UNRESVSP64:
4575 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4580 case XFS_IOC_ALLOCSP:
4581 case XFS_IOC_ALLOCSP64:
4582 case XFS_IOC_FREESP:
4583 case XFS_IOC_FREESP64:
4584 if (startoffset > fsize) {
4585 error = xfs_alloc_file_space(ip, fsize,
4586 startoffset - fsize, 0, attr_flags);
4591 va.va_mask = XFS_AT_SIZE;
4592 va.va_size = startoffset;
4594 error = xfs_setattr(bdp, &va, attr_flags, credp);
4604 return XFS_ERROR(EINVAL);
4608 * update the inode timestamp, mode, and prealloc flag bits
4610 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4612 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4615 xfs_trans_cancel(tp, 0);
4619 xfs_ilock(ip, XFS_ILOCK_EXCL);
4621 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4622 xfs_trans_ihold(tp, ip);
4624 if ((attr_flags & ATTR_DMI) == 0) {
4625 ip->i_d.di_mode &= ~S_ISUID;
4628 * Note that we don't have to worry about mandatory
4629 * file locking being disabled here because we only
4630 * clear the S_ISGID bit if the Group execute bit is
4631 * on, but if it was on then mandatory locking wouldn't
4632 * have been enabled.
4634 if (ip->i_d.di_mode & S_IXGRP)
4635 ip->i_d.di_mode &= ~S_ISGID;
4637 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4640 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4641 else if (clrprealloc)
4642 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4644 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4645 xfs_trans_set_sync(tp);
4647 error = xfs_trans_commit(tp, 0, NULL);
4649 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4654 vnodeops_t xfs_vnodeops = {
4655 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4656 .vop_open = xfs_open,
4657 .vop_read = xfs_read,
4658 #ifdef HAVE_SENDFILE
4659 .vop_sendfile = xfs_sendfile,
4661 .vop_write = xfs_write,
4662 .vop_ioctl = xfs_ioctl,
4663 .vop_getattr = xfs_getattr,
4664 .vop_setattr = xfs_setattr,
4665 .vop_access = xfs_access,
4666 .vop_lookup = xfs_lookup,
4667 .vop_create = xfs_create,
4668 .vop_remove = xfs_remove,
4669 .vop_link = xfs_link,
4670 .vop_rename = xfs_rename,
4671 .vop_mkdir = xfs_mkdir,
4672 .vop_rmdir = xfs_rmdir,
4673 .vop_readdir = xfs_readdir,
4674 .vop_symlink = xfs_symlink,
4675 .vop_readlink = xfs_readlink,
4676 .vop_fsync = xfs_fsync,
4677 .vop_inactive = xfs_inactive,
4678 .vop_fid2 = xfs_fid2,
4679 .vop_rwlock = xfs_rwlock,
4680 .vop_rwunlock = xfs_rwunlock,
4681 .vop_bmap = xfs_bmap,
4682 .vop_reclaim = xfs_reclaim,
4683 .vop_attr_get = xfs_attr_get,
4684 .vop_attr_set = xfs_attr_set,
4685 .vop_attr_remove = xfs_attr_remove,
4686 .vop_attr_list = xfs_attr_list,
4687 .vop_link_removed = (vop_link_removed_t)fs_noval,
4688 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4689 .vop_tosspages = fs_tosspages,
4690 .vop_flushinval_pages = fs_flushinval_pages,
4691 .vop_flush_pages = fs_flush_pages,
4692 .vop_release = xfs_release,
4693 .vop_iflush = xfs_inode_flush,