2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
38 #include "xfs_trans.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_alloc_btree.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_itable.h"
49 #include "xfs_btree.h"
50 #include "xfs_ialloc.h"
51 #include "xfs_alloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode_item.h"
57 #include "xfs_inode.h"
59 #include "xfs_da_btree.h"
62 #include "xfs_refcache.h"
63 #include "xfs_error.h"
65 #include "xfs_rtalloc.h"
66 #include "xfs_quota.h"
67 #include "xfs_utils.h"
68 #include "xfs_trans_space.h"
69 #include "xfs_dir_leaf.h"
71 #include "xfs_log_priv.h"
75 * The maximum pathlen is 1024 bytes. Since the minimum file system
76 * blocksize is 512 bytes, we can get a max of 2 extents back from
79 #define SYMLINK_MAPS 2
82 * For xfs, we check that the file isn't too big to be opened by this kernel.
83 * No other open action is required for regular files. Devices are handled
84 * through the specfs file system, pipes through fifofs. Device and
85 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
86 * when a new vnode is first looked up or created.
97 vp = BHV_TO_VNODE(bdp);
100 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
101 return XFS_ERROR(EIO);
104 * If it's a directory with any blocks, read-ahead block 0
105 * as we're almost certain to have the next operation be a read there.
107 if (vp->v_type == VDIR && ip->i_d.di_nextents > 0) {
108 mode = xfs_ilock_map_shared(ip);
109 if (ip->i_d.di_nextents > 0)
110 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
111 xfs_iunlock(ip, mode);
131 vp = BHV_TO_VNODE(bdp);
132 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
134 ip = XFS_BHVTOI(bdp);
137 if (XFS_FORCED_SHUTDOWN(mp))
138 return XFS_ERROR(EIO);
140 if (!(flags & ATTR_LAZY))
141 xfs_ilock(ip, XFS_ILOCK_SHARED);
143 vap->va_size = ip->i_d.di_size;
144 if (vap->va_mask == XFS_AT_SIZE)
148 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
149 vap->va_nodeid = ip->i_ino;
151 vap->va_nodeid += mp->m_inoadd;
153 vap->va_nlink = ip->i_d.di_nlink;
156 * Quick exit for non-stat callers
159 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
160 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
164 * Copy from in-core inode.
166 vap->va_type = vp->v_type;
167 vap->va_mode = ip->i_d.di_mode & MODEMASK;
168 vap->va_uid = ip->i_d.di_uid;
169 vap->va_gid = ip->i_d.di_gid;
170 vap->va_projid = ip->i_d.di_projid;
173 * Check vnode type block/char vs. everything else.
174 * Do it with bitmask because that's faster than looking
175 * for multiple values individually.
177 if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
180 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
183 /* Large block sizes confuse various
184 * user space programs, so letting the
185 * stripe size through is not a good
188 vap->va_blocksize = mp->m_swidth ?
190 * If the underlying volume is a stripe, then
191 * return the stripe width in bytes as the
192 * recommended I/O size.
194 (mp->m_swidth << mp->m_sb.sb_blocklog) :
196 * Return the largest of the preferred buffer
197 * sizes since doing small I/Os into larger
198 * buffers causes buffers to be decommissioned.
199 * The value returned is in bytes.
201 (1 << (int)MAX(mp->m_readio_log,
207 * Return the largest of the preferred buffer
208 * sizes since doing small I/Os into larger
209 * buffers causes buffers to be decommissioned.
210 * The value returned is in bytes.
212 1 << (int)MAX(mp->m_readio_log,
218 * If the file blocks are being allocated from a
219 * realtime partition, then return the inode's
220 * realtime extent size or the realtime volume's
223 vap->va_blocksize = ip->i_d.di_extsize ?
224 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
225 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
228 vap->va_rdev = ip->i_df.if_u2.if_rdev;
229 vap->va_blocksize = BLKDEV_IOSIZE;
232 vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
233 vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
234 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
235 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
236 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
237 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
240 * Exit for stat callers. See if any of the rest of the fields
241 * to be filled in are needed.
244 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
245 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
249 * convert di_flags to xflags
252 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
253 vap->va_xflags |= XFS_XFLAG_REALTIME;
254 if (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC)
255 vap->va_xflags |= XFS_XFLAG_PREALLOC;
256 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
257 vap->va_xflags |= XFS_XFLAG_IMMUTABLE;
258 if (ip->i_d.di_flags & XFS_DIFLAG_IUNLINK)
259 vap->va_xflags |= XFS_XFLAG_IUNLINK;
260 if (ip->i_d.di_flags & XFS_DIFLAG_BARRIER)
261 vap->va_xflags |= XFS_XFLAG_BARRIER;
262 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
263 vap->va_xflags |= XFS_XFLAG_APPEND;
264 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
265 vap->va_xflags |= XFS_XFLAG_SYNC;
266 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
267 vap->va_xflags |= XFS_XFLAG_NOATIME;
268 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
269 vap->va_xflags |= XFS_XFLAG_NODUMP;
271 vap->va_xflags |= XFS_XFLAG_HASATTR;
273 * Exit for inode revalidate. See if any of the rest of
274 * the fields to be filled in are needed.
277 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
278 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
281 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
283 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
284 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
288 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
289 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
290 ip->i_d.di_anextents;
292 vap->va_anextents = 0;
293 vap->va_gen = ip->i_d.di_gen;
296 if (!(flags & ATTR_LAZY))
297 xfs_iunlock(ip, XFS_ILOCK_SHARED);
323 xfs_prid_t projid=0, iprojid=0;
324 int mandlock_before, mandlock_after;
325 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
328 vp = BHV_TO_VNODE(bdp);
329 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
331 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
332 return XFS_ERROR(EROFS);
335 * Cannot set certain attributes.
338 if (mask & XFS_AT_NOSET) {
339 return XFS_ERROR(EINVAL);
342 ip = XFS_BHVTOI(bdp);
345 if (XFS_FORCED_SHUTDOWN(mp))
346 return XFS_ERROR(EIO);
349 * Timestamps do not need to be logged and hence do not
350 * need to be done within a transaction.
352 if (mask & XFS_AT_UPDTIMES) {
353 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
354 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
355 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
356 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
357 xfs_ichgtime(ip, timeflags);
361 olddquot1 = olddquot2 = NULL;
365 * If disk quotas is on, we make sure that the dquots do exist on disk,
366 * before we start any other transactions. Trying to do this later
367 * is messy. We don't care to take a readlock to look at the ids
368 * in inode here, because we can't hold it across the trans_reserve.
369 * If the IDs do change before we take the ilock, we're covered
370 * because the i_*dquot fields will get updated anyway.
372 if (XFS_IS_QUOTA_ON(mp) && (mask & (XFS_AT_UID|XFS_AT_GID))) {
375 if (mask & XFS_AT_UID) {
377 qflags |= XFS_QMOPT_UQUOTA;
379 uid = ip->i_d.di_uid;
381 if (mask & XFS_AT_GID) {
383 qflags |= XFS_QMOPT_GQUOTA;
385 gid = ip->i_d.di_gid;
388 * We take a reference when we initialize udqp and gdqp,
389 * so it is important that we never blindly double trip on
390 * the same variable. See xfs_create() for an example.
392 ASSERT(udqp == NULL);
393 ASSERT(gdqp == NULL);
394 code = XFS_QM_DQVOPALLOC(mp, ip, uid,gid, qflags, &udqp, &gdqp);
400 * For the other attributes, we acquire the inode lock and
401 * first do an error checking pass.
404 lock_flags = XFS_ILOCK_EXCL;
405 if (!(mask & XFS_AT_SIZE)) {
406 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
407 (mp->m_flags & XFS_MOUNT_WSYNC)) {
408 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
410 if ((code = xfs_trans_reserve(tp, 0,
411 XFS_ICHANGE_LOG_RES(mp), 0,
418 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
419 !(flags & ATTR_DMI)) {
420 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
421 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
422 vap->va_size, 0, dmflags, NULL);
428 lock_flags |= XFS_IOLOCK_EXCL;
431 xfs_ilock(ip, lock_flags);
433 if (_MAC_XFS_IACCESS(ip, MACWRITE, credp)) {
434 code = XFS_ERROR(EACCES);
438 /* boolean: are we the file owner? */
439 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
442 * Change various properties of a file.
443 * Only the owner or users with CAP_FOWNER
444 * capability may do these things.
447 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
448 XFS_AT_GID|XFS_AT_PROJID)) {
450 * CAP_FOWNER overrides the following restrictions:
452 * The user ID of the calling process must be equal
453 * to the file owner ID, except in cases where the
454 * CAP_FSETID capability is applicable.
456 if (!file_owner && !capable(CAP_FOWNER)) {
457 code = XFS_ERROR(EPERM);
462 * CAP_FSETID overrides the following restrictions:
464 * The effective user ID of the calling process shall match
465 * the file owner when setting the set-user-ID and
466 * set-group-ID bits on that file.
468 * The effective group ID or one of the supplementary group
469 * IDs of the calling process shall match the group owner of
470 * the file when setting the set-group-ID bit on that file
472 if (mask & XFS_AT_MODE) {
475 if ((vap->va_mode & S_ISUID) && !file_owner)
477 if ((vap->va_mode & S_ISGID) &&
478 !in_group_p((gid_t)ip->i_d.di_gid))
481 /* Linux allows this, Irix doesn't. */
482 if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
485 if (m && !capable(CAP_FSETID))
491 * Change file ownership. Must be the owner or privileged.
492 * If the system was configured with the "restricted_chown"
493 * option, the owner is not permitted to give away the file,
494 * and can change the group id only to a group of which he
495 * or she is a member.
497 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
499 * These IDs could have changed since we last looked at them.
500 * But, we're assured that if the ownership did change
501 * while we didn't have the inode locked, inode's dquot(s)
502 * would have changed also.
504 iuid = ip->i_d.di_uid;
505 iprojid = ip->i_d.di_projid;
506 igid = ip->i_d.di_gid;
507 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
508 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
509 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
513 * CAP_CHOWN overrides the following restrictions:
515 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
516 * shall override the restriction that a process cannot
517 * change the user ID of a file it owns and the restriction
518 * that the group ID supplied to the chown() function
519 * shall be equal to either the group ID or one of the
520 * supplementary group IDs of the calling process.
522 * XXX: How does restricted_chown affect projid?
524 if (restricted_chown &&
525 (iuid != uid || (igid != gid &&
526 !in_group_p((gid_t)gid))) &&
527 !capable(CAP_CHOWN)) {
528 code = XFS_ERROR(EPERM);
532 * Do a quota reservation only if uid or gid is actually
535 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
536 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
538 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
539 capable(CAP_FOWNER) ?
540 XFS_QMOPT_FORCE_RES : 0);
541 if (code) /* out of quota */
547 * Truncate file. Must have write permission and not be a directory.
549 if (mask & XFS_AT_SIZE) {
550 /* Short circuit the truncate case for zero length files */
551 if ((vap->va_size == 0) &&
552 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
553 xfs_iunlock(ip, XFS_ILOCK_EXCL);
554 lock_flags &= ~XFS_ILOCK_EXCL;
555 if (mask & XFS_AT_CTIME)
556 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
561 if (vp->v_type == VDIR) {
562 code = XFS_ERROR(EISDIR);
564 } else if (vp->v_type != VREG) {
565 code = XFS_ERROR(EINVAL);
569 * Make sure that the dquots are attached to the inode.
571 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
576 * Change file access or modified times.
578 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
580 if ((flags & ATTR_UTIME) &&
581 !capable(CAP_FOWNER)) {
582 code = XFS_ERROR(EPERM);
589 * Change extent size or realtime flag.
591 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
593 * Can't change extent size if any extents are allocated.
595 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
596 (mask & XFS_AT_EXTSIZE) &&
597 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
599 code = XFS_ERROR(EINVAL); /* EFBIG? */
604 * Can't set extent size unless the file is marked, or
605 * about to be marked as a realtime file.
607 * This check will be removed when fixed size extents
608 * with buffered data writes is implemented.
611 if ((mask & XFS_AT_EXTSIZE) &&
612 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
614 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
615 ((mask & XFS_AT_XFLAGS) &&
616 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
617 code = XFS_ERROR(EINVAL);
622 * Can't change realtime flag if any extents are allocated.
624 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
625 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
626 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
627 code = XFS_ERROR(EINVAL); /* EFBIG? */
631 * Extent size must be a multiple of the appropriate block
632 * size, if set at all.
634 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
637 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
638 ((mask & XFS_AT_XFLAGS) &&
639 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
640 size = mp->m_sb.sb_rextsize <<
641 mp->m_sb.sb_blocklog;
643 size = mp->m_sb.sb_blocksize;
645 if (vap->va_extsize % size) {
646 code = XFS_ERROR(EINVAL);
651 * If realtime flag is set then must have realtime data.
653 if ((mask & XFS_AT_XFLAGS) &&
654 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
655 if ((mp->m_sb.sb_rblocks == 0) ||
656 (mp->m_sb.sb_rextsize == 0) ||
657 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
658 code = XFS_ERROR(EINVAL);
664 * Can't modify an immutable/append-only file unless
665 * we have appropriate permission.
667 if ((mask & XFS_AT_XFLAGS) &&
669 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
671 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
672 !capable(CAP_LINUX_IMMUTABLE)) {
673 code = XFS_ERROR(EPERM);
679 * Now we can make the changes. Before we join the inode
680 * to the transaction, if XFS_AT_SIZE is set then take care of
681 * the part of the truncation that must be done without the
682 * inode lock. This needs to be done before joining the inode
683 * to the transaction, because the inode cannot be unlocked
684 * once it is a part of the transaction.
686 if (mask & XFS_AT_SIZE) {
688 if (vap->va_size > ip->i_d.di_size)
689 code = xfs_igrow_start(ip, vap->va_size, credp);
690 xfs_iunlock(ip, XFS_ILOCK_EXCL);
692 code = xfs_itruncate_data(ip, vap->va_size);
695 lock_flags &= ~XFS_ILOCK_EXCL;
696 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
699 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
700 if ((code = xfs_trans_reserve(tp, 0,
701 XFS_ITRUNCATE_LOG_RES(mp), 0,
702 XFS_TRANS_PERM_LOG_RES,
703 XFS_ITRUNCATE_LOG_COUNT))) {
704 xfs_trans_cancel(tp, 0);
705 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
708 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
709 xfs_ilock(ip, XFS_ILOCK_EXCL);
713 xfs_trans_ijoin(tp, ip, lock_flags);
714 xfs_trans_ihold(tp, ip);
717 /* determine whether mandatory locking mode changes */
718 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
721 * Truncate file. Must have write permission and not be a directory.
723 if (mask & XFS_AT_SIZE) {
724 if (vap->va_size > ip->i_d.di_size) {
725 xfs_igrow_finish(tp, ip, vap->va_size,
726 !(flags & ATTR_DMI));
727 } else if ((vap->va_size <= ip->i_d.di_size) ||
728 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
730 * signal a sync transaction unless
731 * we're truncating an already unlinked
732 * file on a wsync filesystem
734 code = xfs_itruncate_finish(&tp, ip,
735 (xfs_fsize_t)vap->va_size,
737 ((ip->i_d.di_nlink != 0 ||
738 !(mp->m_flags & XFS_MOUNT_WSYNC))
745 * Have to do this even if the file's size doesn't change.
747 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
751 * Change file access modes.
753 if (mask & XFS_AT_MODE) {
754 ip->i_d.di_mode &= S_IFMT;
755 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
757 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
758 timeflags |= XFS_ICHGTIME_CHG;
762 * Change file ownership. Must be the owner or privileged.
763 * If the system was configured with the "restricted_chown"
764 * option, the owner is not permitted to give away the file,
765 * and can change the group id only to a group of which he
766 * or she is a member.
768 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
770 * CAP_FSETID overrides the following restrictions:
772 * The set-user-ID and set-group-ID bits of a file will be
773 * cleared upon successful return from chown()
775 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
776 !capable(CAP_FSETID)) {
777 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
781 * Change the ownerships and register quota modifications
782 * in the transaction.
785 if (XFS_IS_UQUOTA_ON(mp)) {
786 ASSERT(mask & XFS_AT_UID);
788 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
789 &ip->i_udquot, udqp);
791 ip->i_d.di_uid = uid;
794 if (XFS_IS_GQUOTA_ON(mp)) {
795 ASSERT(mask & XFS_AT_GID);
797 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
798 &ip->i_gdquot, gdqp);
800 ip->i_d.di_gid = gid;
802 if (iprojid != projid) {
803 ip->i_d.di_projid = projid;
805 * We may have to rev the inode as well as
806 * the superblock version number since projids didn't
807 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
809 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
810 xfs_bump_ino_vers2(tp, ip);
813 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
814 timeflags |= XFS_ICHGTIME_CHG;
819 * Change file access or modified times.
821 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
822 if (mask & XFS_AT_ATIME) {
823 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
824 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
825 ip->i_update_core = 1;
826 timeflags &= ~XFS_ICHGTIME_ACC;
828 if (mask & XFS_AT_MTIME) {
829 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
830 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
831 timeflags &= ~XFS_ICHGTIME_MOD;
832 timeflags |= XFS_ICHGTIME_CHG;
834 if (tp && (flags & ATTR_UTIME))
835 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
839 * Change XFS-added attributes.
841 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
842 if (mask & XFS_AT_EXTSIZE) {
844 * Converting bytes to fs blocks.
846 ip->i_d.di_extsize = vap->va_extsize >>
847 mp->m_sb.sb_blocklog;
849 if (mask & XFS_AT_XFLAGS) {
850 ip->i_d.di_flags = 0;
851 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
852 ip->i_d.di_flags |= XFS_DIFLAG_REALTIME;
853 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
855 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
856 ip->i_d.di_flags |= XFS_DIFLAG_IMMUTABLE;
857 if (vap->va_xflags & XFS_XFLAG_IUNLINK)
858 ip->i_d.di_flags |= XFS_DIFLAG_IUNLINK;
859 if (vap->va_xflags & XFS_XFLAG_BARRIER)
860 ip->i_d.di_flags |= XFS_DIFLAG_BARRIER;
861 if (vap->va_xflags & XFS_XFLAG_APPEND)
862 ip->i_d.di_flags |= XFS_DIFLAG_APPEND;
863 if (vap->va_xflags & XFS_XFLAG_SYNC)
864 ip->i_d.di_flags |= XFS_DIFLAG_SYNC;
865 if (vap->va_xflags & XFS_XFLAG_NOATIME)
866 ip->i_d.di_flags |= XFS_DIFLAG_NOATIME;
867 if (vap->va_xflags & XFS_XFLAG_NODUMP)
868 ip->i_d.di_flags |= XFS_DIFLAG_NODUMP;
869 /* can't set PREALLOC this way, just ignore it */
871 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
872 timeflags |= XFS_ICHGTIME_CHG;
876 * Change file inode change time only if XFS_AT_CTIME set
877 * AND we have been called by a DMI function.
880 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
881 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
882 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
883 ip->i_update_core = 1;
884 timeflags &= ~XFS_ICHGTIME_CHG;
888 * Send out timestamp changes that need to be set to the
889 * current time. Not done when called by a DMI function.
891 if (timeflags && !(flags & ATTR_DMI))
892 xfs_ichgtime(ip, timeflags);
894 XFS_STATS_INC(xs_ig_attrchg);
897 * If this is a synchronous mount, make sure that the
898 * transaction goes to disk before returning to the user.
899 * This is slightly sub-optimal in that truncates require
900 * two sync transactions instead of one for wsync filesytems.
901 * One for the truncate and one for the timestamps since we
902 * don't want to change the timestamps unless we're sure the
903 * truncate worked. Truncates are less than 1% of the laddis
904 * mix so this probably isn't worth the trouble to optimize.
908 if (mp->m_flags & XFS_MOUNT_WSYNC)
909 xfs_trans_set_sync(tp);
911 code = xfs_trans_commit(tp, commit_flags, NULL);
915 * If the (regular) file's mandatory locking mode changed, then
916 * notify the vnode. We do this under the inode lock to prevent
917 * racing calls to vop_vnode_change.
919 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
920 if (mandlock_before != mandlock_after) {
921 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
925 xfs_iunlock(ip, lock_flags);
928 * Release any dquot(s) the inode had kept before chown.
930 XFS_QM_DQRELE(mp, olddquot1);
931 XFS_QM_DQRELE(mp, olddquot2);
932 XFS_QM_DQRELE(mp, udqp);
933 XFS_QM_DQRELE(mp, gdqp);
939 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
940 !(flags & ATTR_DMI)) {
941 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
942 NULL, DM_RIGHT_NULL, NULL, NULL,
943 0, 0, AT_DELAY_FLAG(flags));
948 commit_flags |= XFS_TRANS_ABORT;
951 XFS_QM_DQRELE(mp, udqp);
952 XFS_QM_DQRELE(mp, gdqp);
954 xfs_trans_cancel(tp, commit_flags);
956 if (lock_flags != 0) {
957 xfs_iunlock(ip, lock_flags);
965 * Null conversion from vnode mode bits to inode mode bits, as in efs.
976 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
977 (inst_t *)__return_address);
979 ip = XFS_BHVTOI(bdp);
980 xfs_ilock(ip, XFS_ILOCK_SHARED);
981 error = xfs_iaccess(ip, mode, credp);
982 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1006 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1012 vp = BHV_TO_VNODE(bdp);
1013 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1015 ip = XFS_BHVTOI(bdp);
1018 if (XFS_FORCED_SHUTDOWN(mp))
1019 return XFS_ERROR(EIO);
1021 xfs_ilock(ip, XFS_ILOCK_SHARED);
1023 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1025 offset = uiop->uio_offset;
1026 count = uiop->uio_resid;
1029 error = XFS_ERROR(EINVAL);
1037 if (!(ioflags & IO_INVIS)) {
1038 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1042 * See if the symlink is stored inline.
1044 pathlen = (int)ip->i_d.di_size;
1046 if (ip->i_df.if_flags & XFS_IFINLINE) {
1047 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1051 * Symlink not inline. Call bmap to get it in.
1053 nmaps = SYMLINK_MAPS;
1055 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1056 0, NULL, 0, mval, &nmaps, NULL);
1062 for (n = 0; n < nmaps; n++) {
1063 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1064 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1065 bp = xfs_buf_read(mp->m_ddev_targp, d,
1066 BTOBB(byte_cnt), 0);
1067 error = XFS_BUF_GETERROR(bp);
1069 xfs_ioerror_alert("xfs_readlink",
1070 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1074 if (pathlen < byte_cnt)
1076 pathlen -= byte_cnt;
1078 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1087 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1096 * This is called to sync the inode and its data out to disk.
1097 * We need to hold the I/O lock while flushing the data, and
1098 * the inode lock while flushing the inode. The inode lock CANNOT
1099 * be held while flushing the data, so acquire after we're done
1114 vn_trace_entry(BHV_TO_VNODE(bdp),
1115 __FUNCTION__, (inst_t *)__return_address);
1117 ip = XFS_BHVTOI(bdp);
1119 ASSERT(start >= 0 && stop >= -1);
1121 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1122 return XFS_ERROR(EIO);
1125 * We always need to make sure that the required inode state
1126 * is safe on disk. The vnode might be clean but because
1127 * of committed transactions that haven't hit the disk yet.
1128 * Likewise, there could be unflushed non-transactional
1129 * changes to the inode core that have to go to disk.
1131 * The following code depends on one assumption: that
1132 * any transaction that changes an inode logs the core
1133 * because it has to change some field in the inode core
1134 * (typically nextents or nblocks). That assumption
1135 * implies that any transactions against an inode will
1136 * catch any non-transactional updates. If inode-altering
1137 * transactions exist that violate this assumption, the
1138 * code breaks. Right now, it figures that if the involved
1139 * update_* field is clear and the inode is unpinned, the
1140 * inode is clean. Either it's been flushed or it's been
1141 * committed and the commit has hit the disk unpinning the inode.
1142 * (Note that xfs_inode_item_format() called at commit clears
1143 * the update_* fields.)
1145 xfs_ilock(ip, XFS_ILOCK_SHARED);
1147 /* If we are flushing data then we care about update_size
1148 * being set, otherwise we care about update_core
1150 if ((flag & FSYNC_DATA) ?
1151 (ip->i_update_size == 0) :
1152 (ip->i_update_core == 0)) {
1154 * Timestamps/size haven't changed since last inode
1155 * flush or inode transaction commit. That means
1156 * either nothing got written or a transaction
1157 * committed which caught the updates. If the
1158 * latter happened and the transaction hasn't
1159 * hit the disk yet, the inode will be still
1160 * be pinned. If it is, force the log.
1163 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1165 if (xfs_ipincount(ip)) {
1166 xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1168 ((flag & FSYNC_WAIT)
1169 ? XFS_LOG_SYNC : 0));
1174 * Kick off a transaction to log the inode
1175 * core to get the updates. Make it
1176 * sync if FSYNC_WAIT is passed in (which
1177 * is done by everybody but specfs). The
1178 * sync transaction will also force the log.
1180 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1181 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1182 if ((error = xfs_trans_reserve(tp, 0,
1183 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1185 xfs_trans_cancel(tp, 0);
1188 xfs_ilock(ip, XFS_ILOCK_EXCL);
1191 * Note - it's possible that we might have pushed
1192 * ourselves out of the way during trans_reserve
1193 * which would flush the inode. But there's no
1194 * guarantee that the inode buffer has actually
1195 * gone out yet (it's delwri). Plus the buffer
1196 * could be pinned anyway if it's part of an
1197 * inode in another recent transaction. So we
1198 * play it safe and fire off the transaction anyway.
1200 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1201 xfs_trans_ihold(tp, ip);
1202 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1203 if (flag & FSYNC_WAIT)
1204 xfs_trans_set_sync(tp);
1205 error = xfs_trans_commit(tp, 0, NULL);
1207 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1213 * This is called by xfs_inactive to free any blocks beyond eof,
1214 * when the link count isn't zero.
1217 xfs_inactive_free_eofblocks(
1223 xfs_fileoff_t end_fsb;
1224 xfs_fileoff_t last_fsb;
1225 xfs_filblks_t map_len;
1227 xfs_bmbt_irec_t imap;
1230 * Figure out if there are any blocks beyond the end
1231 * of the file. If not, then there is nothing to do.
1233 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1234 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1235 map_len = last_fsb - end_fsb;
1240 xfs_ilock(ip, XFS_ILOCK_SHARED);
1241 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1242 NULL, 0, &imap, &nimaps, NULL);
1243 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1245 if (!error && (nimaps != 0) &&
1246 (imap.br_startblock != HOLESTARTBLOCK)) {
1248 * Attach the dquots to the inode up front.
1250 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1254 * There are blocks after the end of file.
1255 * Free them up now by truncating the file to
1258 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1261 * Do the xfs_itruncate_start() call before
1262 * reserving any log space because
1263 * itruncate_start will call into the buffer
1264 * cache and we can't
1265 * do that within a transaction.
1267 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1268 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1271 error = xfs_trans_reserve(tp, 0,
1272 XFS_ITRUNCATE_LOG_RES(mp),
1273 0, XFS_TRANS_PERM_LOG_RES,
1274 XFS_ITRUNCATE_LOG_COUNT);
1276 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1277 xfs_trans_cancel(tp, 0);
1278 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1282 xfs_ilock(ip, XFS_ILOCK_EXCL);
1283 xfs_trans_ijoin(tp, ip,
1286 xfs_trans_ihold(tp, ip);
1288 error = xfs_itruncate_finish(&tp, ip,
1293 * If we get an error at this point we
1294 * simply don't bother truncating the file.
1297 xfs_trans_cancel(tp,
1298 (XFS_TRANS_RELEASE_LOG_RES |
1301 error = xfs_trans_commit(tp,
1302 XFS_TRANS_RELEASE_LOG_RES,
1305 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1311 * Free a symlink that has blocks associated with it.
1314 xfs_inactive_symlink_rmt(
1322 xfs_fsblock_t first_block;
1323 xfs_bmap_free_t free_list;
1326 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1334 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1336 * We're freeing a symlink that has some
1337 * blocks allocated to it. Free the
1338 * blocks here. We know that we've got
1339 * either 1 or 2 extents and that we can
1340 * free them all in one bunmapi call.
1342 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1343 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1344 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1345 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1346 xfs_trans_cancel(tp, 0);
1351 * Lock the inode, fix the size, and join it to the transaction.
1352 * Hold it so in the normal path, we still have it locked for
1353 * the second transaction. In the error paths we need it
1354 * held so the cancel won't rele it, see below.
1356 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1357 size = (int)ip->i_d.di_size;
1358 ip->i_d.di_size = 0;
1359 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1360 xfs_trans_ihold(tp, ip);
1361 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1363 * Find the block(s) so we can inval and unmap them.
1366 XFS_BMAP_INIT(&free_list, &first_block);
1367 nmaps = sizeof(mval) / sizeof(mval[0]);
1368 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1369 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1373 * Invalidate the block(s).
1375 for (i = 0; i < nmaps; i++) {
1376 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1377 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1378 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1379 xfs_trans_binval(tp, bp);
1382 * Unmap the dead block(s) to the free_list.
1384 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1385 &first_block, &free_list, &done)))
1389 * Commit the first transaction. This logs the EFI and the inode.
1391 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1394 * The transaction must have been committed, since there were
1395 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1396 * The new tp has the extent freeing and EFDs.
1400 * The first xact was committed, so add the inode to the new one.
1401 * Mark it dirty so it will be logged and moved forward in the log as
1402 * part of every commit.
1404 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1405 xfs_trans_ihold(tp, ip);
1406 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1408 * Get a new, empty transaction to return to our caller.
1410 ntp = xfs_trans_dup(tp);
1412 * Commit the transaction containing extent freeing and EFD's.
1413 * If we get an error on the commit here or on the reserve below,
1414 * we need to unlock the inode since the new transaction doesn't
1415 * have the inode attached.
1417 error = xfs_trans_commit(tp, 0, NULL);
1420 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1424 * Remove the memory for extent descriptions (just bookkeeping).
1426 if (ip->i_df.if_bytes)
1427 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1428 ASSERT(ip->i_df.if_bytes == 0);
1430 * Put an itruncate log reservation in the new transaction
1433 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1434 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1435 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1439 * Return with the inode locked but not joined to the transaction.
1445 xfs_bmap_cancel(&free_list);
1448 * Have to come here with the inode locked and either
1449 * (held and in the transaction) or (not in the transaction).
1450 * If the inode isn't held then cancel would iput it, but
1451 * that's wrong since this is inactive and the vnode ref
1452 * count is 0 already.
1453 * Cancel won't do anything to the inode if held, but it still
1454 * needs to be locked until the cancel is done, if it was
1455 * joined to the transaction.
1457 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1458 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1465 xfs_inactive_symlink_local(
1471 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1473 * We're freeing a symlink which fit into
1474 * the inode. Just free the memory used
1475 * to hold the old symlink.
1477 error = xfs_trans_reserve(*tpp, 0,
1478 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1479 0, XFS_TRANS_PERM_LOG_RES,
1480 XFS_ITRUNCATE_LOG_COUNT);
1483 xfs_trans_cancel(*tpp, 0);
1487 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1490 * Zero length symlinks _can_ exist.
1492 if (ip->i_df.if_bytes > 0) {
1493 xfs_idata_realloc(ip,
1494 -(ip->i_df.if_bytes),
1496 ASSERT(ip->i_df.if_bytes == 0);
1513 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1516 ASSERT(ip->i_d.di_forkoff != 0);
1517 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1518 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1520 error = xfs_attr_inactive(ip);
1523 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1524 return (error); /* goto out*/
1527 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1528 error = xfs_trans_reserve(tp, 0,
1529 XFS_IFREE_LOG_RES(mp),
1530 0, XFS_TRANS_PERM_LOG_RES,
1531 XFS_INACTIVE_LOG_COUNT);
1533 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1534 xfs_trans_cancel(tp, 0);
1536 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1540 xfs_ilock(ip, XFS_ILOCK_EXCL);
1541 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1542 xfs_trans_ihold(tp, ip);
1543 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1545 ASSERT(ip->i_d.di_anextents == 0);
1560 vp = BHV_TO_VNODE(bdp);
1561 ip = XFS_BHVTOI(bdp);
1563 if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1567 /* If this is a read-only mount, don't do this (would generate I/O) */
1568 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1571 #ifdef HAVE_REFCACHE
1572 /* If we are in the NFS reference cache then don't do this now */
1579 if (ip->i_d.di_nlink != 0) {
1580 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1581 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1582 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1583 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1584 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1586 /* Update linux inode block count after free above */
1587 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1588 ip->i_d.di_nblocks + ip->i_delayed_blks);
1598 * This is called when the vnode reference count for the vnode
1599 * goes to zero. If the file has been unlinked, then it must
1600 * now be truncated. Also, we clear all of the read-ahead state
1601 * kept for the inode here since the file is now closed.
1610 xfs_bmap_free_t free_list;
1611 xfs_fsblock_t first_block;
1618 vp = BHV_TO_VNODE(bdp);
1619 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1621 ip = XFS_BHVTOI(bdp);
1624 * If the inode is already free, then there can be nothing
1627 if (ip->i_d.di_mode == 0) {
1628 ASSERT(ip->i_df.if_real_bytes == 0);
1629 ASSERT(ip->i_df.if_broot_bytes == 0);
1630 return VN_INACTIVE_CACHE;
1634 * Only do a truncate if it's a regular file with
1635 * some actual space in it. It's OK to look at the
1636 * inode's fields without the lock because we're the
1637 * only one with a reference to the inode.
1639 truncate = ((ip->i_d.di_nlink == 0) &&
1640 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1641 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1645 if (ip->i_d.di_nlink == 0 &&
1646 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1647 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1652 /* If this is a read-only mount, don't do this (would generate I/O) */
1653 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1656 if (ip->i_d.di_nlink != 0) {
1657 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1658 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1659 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1660 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1661 (ip->i_delayed_blks != 0))) {
1662 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1663 return (VN_INACTIVE_CACHE);
1664 /* Update linux inode block count after free above */
1665 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1666 ip->i_d.di_nblocks + ip->i_delayed_blks);
1671 ASSERT(ip->i_d.di_nlink == 0);
1673 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1674 return (VN_INACTIVE_CACHE);
1676 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1679 * Do the xfs_itruncate_start() call before
1680 * reserving any log space because itruncate_start
1681 * will call into the buffer cache and we can't
1682 * do that within a transaction.
1684 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1686 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1688 error = xfs_trans_reserve(tp, 0,
1689 XFS_ITRUNCATE_LOG_RES(mp),
1690 0, XFS_TRANS_PERM_LOG_RES,
1691 XFS_ITRUNCATE_LOG_COUNT);
1693 /* Don't call itruncate_cleanup */
1694 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1695 xfs_trans_cancel(tp, 0);
1696 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1697 return (VN_INACTIVE_CACHE);
1700 xfs_ilock(ip, XFS_ILOCK_EXCL);
1701 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1702 xfs_trans_ihold(tp, ip);
1705 * normally, we have to run xfs_itruncate_finish sync.
1706 * But if filesystem is wsync and we're in the inactive
1707 * path, then we know that nlink == 0, and that the
1708 * xaction that made nlink == 0 is permanently committed
1709 * since xfs_remove runs as a synchronous transaction.
1711 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1712 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1715 xfs_trans_cancel(tp,
1716 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1717 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1718 return (VN_INACTIVE_CACHE);
1720 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1723 * If we get an error while cleaning up a
1724 * symlink we bail out.
1726 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1727 xfs_inactive_symlink_rmt(ip, &tp) :
1728 xfs_inactive_symlink_local(ip, &tp);
1732 return (VN_INACTIVE_CACHE);
1735 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1736 xfs_trans_ihold(tp, ip);
1738 error = xfs_trans_reserve(tp, 0,
1739 XFS_IFREE_LOG_RES(mp),
1740 0, XFS_TRANS_PERM_LOG_RES,
1741 XFS_INACTIVE_LOG_COUNT);
1743 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1744 xfs_trans_cancel(tp, 0);
1745 return (VN_INACTIVE_CACHE);
1748 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1749 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1750 xfs_trans_ihold(tp, ip);
1754 * If there are attributes associated with the file
1755 * then blow them away now. The code calls a routine
1756 * that recursively deconstructs the attribute fork.
1757 * We need to just commit the current transaction
1758 * because we can't use it for xfs_attr_inactive().
1760 if (ip->i_d.di_anextents > 0) {
1761 error = xfs_inactive_attrs(ip, &tp);
1763 * If we got an error, the transaction is already
1764 * cancelled, and the inode is unlocked. Just get out.
1767 return (VN_INACTIVE_CACHE);
1768 } else if (ip->i_afp) {
1769 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1775 XFS_BMAP_INIT(&free_list, &first_block);
1776 error = xfs_ifree(tp, ip, &free_list);
1779 * If we fail to free the inode, shut down. The cancel
1780 * might do that, we need to make sure. Otherwise the
1781 * inode might be lost for a long time or forever.
1783 if (!XFS_FORCED_SHUTDOWN(mp)) {
1785 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1786 error, mp->m_fsname);
1787 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1789 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1792 * Credit the quota account(s). The inode is gone.
1794 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1797 * Just ignore errors at this point. There is
1798 * nothing we can do except to try to keep going.
1800 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1802 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1805 * Release the dquots held by inode, if any.
1807 XFS_QM_DQDETACH(mp, ip);
1809 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1812 return VN_INACTIVE_CACHE;
1821 bhv_desc_t *dir_bdp,
1828 xfs_inode_t *dp, *ip;
1834 dir_vp = BHV_TO_VNODE(dir_bdp);
1835 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1837 dp = XFS_BHVTOI(dir_bdp);
1839 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1840 return XFS_ERROR(EIO);
1842 lock_mode = xfs_ilock_map_shared(dp);
1843 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1845 *vpp = XFS_ITOV(ip);
1848 xfs_iunlock_map_shared(dp, lock_mode);
1853 #define XFS_CREATE_NEW_MAXTRIES 10000
1856 * xfs_create (create a new file).
1860 bhv_desc_t *dir_bdp,
1866 char *name = VNAME(dentry);
1868 xfs_inode_t *dp, *ip;
1874 xfs_bmap_free_t free_list;
1875 xfs_fsblock_t first_block;
1876 boolean_t dp_joined_to_trans;
1877 int dm_event_sent = 0;
1881 struct xfs_dquot *udqp, *gdqp;
1887 dir_vp = BHV_TO_VNODE(dir_bdp);
1888 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1890 dp = XFS_BHVTOI(dir_bdp);
1893 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1894 namelen = VNAMELEN(dentry);
1896 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1897 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1898 dir_vp, DM_RIGHT_NULL, NULL,
1899 DM_RIGHT_NULL, name, NULL,
1907 if (XFS_FORCED_SHUTDOWN(mp))
1908 return XFS_ERROR(EIO);
1910 /* Return through std_return after this point. */
1913 if (vap->va_mask & XFS_AT_PROJID)
1914 prid = (xfs_prid_t)vap->va_projid;
1916 prid = (xfs_prid_t)dfltprid;
1919 * Make sure that we have allocated dquot(s) on disk.
1921 error = XFS_QM_DQVOPALLOC(mp, dp,
1922 current_fsuid(credp), current_fsgid(credp),
1923 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1928 dp_joined_to_trans = B_FALSE;
1930 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1931 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1932 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1934 * Initially assume that the file does not exist and
1935 * reserve the resources for that case. If that is not
1936 * the case we'll drop the one we have and get a more
1937 * appropriate transaction later.
1939 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1940 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1941 if (error == ENOSPC) {
1943 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1944 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1952 xfs_ilock(dp, XFS_ILOCK_EXCL);
1954 XFS_BMAP_INIT(&free_list, &first_block);
1959 * Reserve disk quota and the inode.
1961 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1966 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1968 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1969 error = xfs_dir_ialloc(&tp, dp,
1970 MAKEIMODE(vap->va_type,vap->va_mode), 1,
1971 rdev, credp, prid, resblks > 0,
1974 if (error == ENOSPC)
1981 * At this point, we've gotten a newly allocated inode.
1982 * It is locked (and joined to the transaction).
1985 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1988 * Now we join the directory inode to the transaction.
1989 * We do not do it earlier because xfs_dir_ialloc
1990 * might commit the previous transaction (and release
1995 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1996 dp_joined_to_trans = B_TRUE;
1998 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1999 &first_block, &free_list,
2000 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2002 ASSERT(error != ENOSPC);
2005 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2006 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2009 * If this is a synchronous mount, make sure that the
2010 * create transaction goes to disk before returning to
2013 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2014 xfs_trans_set_sync(tp);
2020 * Attach the dquot(s) to the inodes and modify them incore.
2021 * These ids of the inode couldn't have changed since the new
2022 * inode has been locked ever since it was created.
2024 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2027 * xfs_trans_commit normally decrements the vnode ref count
2028 * when it unlocks the inode. Since we want to return the
2029 * vnode to the caller, we bump the vnode ref count now.
2034 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2036 xfs_bmap_cancel(&free_list);
2040 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2047 XFS_QM_DQRELE(mp, udqp);
2048 XFS_QM_DQRELE(mp, gdqp);
2051 * Propogate the fact that the vnode changed after the
2052 * xfs_inode locks have been released.
2054 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2058 /* Fallthrough to std_return with error = 0 */
2061 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2062 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2063 DM_EVENT_POSTCREATE)) {
2064 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2065 dir_vp, DM_RIGHT_NULL,
2067 DM_RIGHT_NULL, name, NULL,
2068 dm_di_mode, error, 0);
2073 cancel_flags |= XFS_TRANS_ABORT;
2078 xfs_trans_cancel(tp, cancel_flags);
2080 if (!dp_joined_to_trans && (dp != NULL))
2081 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2082 XFS_QM_DQRELE(mp, udqp);
2083 XFS_QM_DQRELE(mp, gdqp);
2089 * Wait until after the current transaction is aborted to
2090 * release the inode. This prevents recursive transactions
2091 * and deadlocks from xfs_inactive.
2093 cancel_flags |= XFS_TRANS_ABORT;
2094 xfs_trans_cancel(tp, cancel_flags);
2097 XFS_QM_DQRELE(mp, udqp);
2098 XFS_QM_DQRELE(mp, gdqp);
2105 * Some counters to see if (and how often) we are hitting some deadlock
2106 * prevention code paths.
2110 int xfs_rm_lock_delays;
2111 int xfs_rm_attempts;
2115 * The following routine will lock the inodes associated with the
2116 * directory and the named entry in the directory. The locks are
2117 * acquired in increasing inode number.
2119 * If the entry is "..", then only the directory is locked. The
2120 * vnode ref count will still include that from the .. entry in
2123 * There is a deadlock we need to worry about. If the locked directory is
2124 * in the AIL, it might be blocking up the log. The next inode we lock
2125 * could be already locked by another thread waiting for log space (e.g
2126 * a permanent log reservation with a long running transaction (see
2127 * xfs_itruncate_finish)). To solve this, we must check if the directory
2128 * is in the ail and use lock_nowait. If we can't lock, we need to
2129 * drop the inode lock on the directory and try again. xfs_iunlock will
2130 * potentially push the tail if we were holding up the log.
2133 xfs_lock_dir_and_entry(
2136 xfs_inode_t *ip) /* inode of entry 'name' */
2140 xfs_inode_t *ips[2];
2149 xfs_ilock(dp, XFS_ILOCK_EXCL);
2156 * We want to lock in increasing inum. Since we've already
2157 * acquired the lock on the directory, we may need to release
2158 * if if the inum of the entry turns out to be less.
2160 if (e_inum > dp->i_ino) {
2162 * We are already in the right order, so just
2163 * lock on the inode of the entry.
2164 * We need to use nowait if dp is in the AIL.
2167 lp = (xfs_log_item_t *)dp->i_itemp;
2168 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2169 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2176 * Unlock dp and try again.
2177 * xfs_iunlock will try to push the tail
2178 * if the inode is in the AIL.
2181 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2183 if ((attempts % 5) == 0) {
2184 delay(1); /* Don't just spin the CPU */
2186 xfs_rm_lock_delays++;
2192 xfs_ilock(ip, XFS_ILOCK_EXCL);
2194 } else if (e_inum < dp->i_ino) {
2195 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2199 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2201 /* else e_inum == dp->i_ino */
2202 /* This can happen if we're asked to lock /x/..
2203 * the entry is "..", which is also the parent directory.
2211 int xfs_small_retries;
2212 int xfs_middle_retries;
2213 int xfs_lots_retries;
2214 int xfs_lock_delays;
2218 * The following routine will lock n inodes in exclusive mode.
2219 * We assume the caller calls us with the inodes in i_ino order.
2221 * We need to detect deadlock where an inode that we lock
2222 * is in the AIL and we start waiting for another inode that is locked
2223 * by a thread in a long running transaction (such as truncate). This can
2224 * result in deadlock since the long running trans might need to wait
2225 * for the inode we just locked in order to push the tail and free space
2235 int attempts = 0, i, j, try_lock;
2238 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2249 for (; i < inodes; i++) {
2252 if (i && (ips[i] == ips[i-1])) /* Already locked */
2256 * If try_lock is not set yet, make sure all locked inodes
2257 * are not in the AIL.
2258 * If any are, set try_lock to be used later.
2262 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2263 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2264 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2271 * If any of the previous locks we have locked is in the AIL,
2272 * we must TRY to get the second and subsequent locks. If
2273 * we can't get any, we must release all we have
2278 /* try_lock must be 0 if i is 0. */
2280 * try_lock means we have an inode locked
2281 * that is in the AIL.
2284 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2288 * Unlock all previous guys and try again.
2289 * xfs_iunlock will try to push the tail
2290 * if the inode is in the AIL.
2293 for(j = i - 1; j >= 0; j--) {
2296 * Check to see if we've already
2297 * unlocked this one.
2298 * Not the first one going back,
2299 * and the inode ptr is the same.
2301 if ((j != (i - 1)) && ips[j] ==
2305 xfs_iunlock(ips[j], lock_mode);
2308 if ((attempts % 5) == 0) {
2309 delay(1); /* Don't just spin the CPU */
2319 xfs_ilock(ips[i], lock_mode);
2325 if (attempts < 5) xfs_small_retries++;
2326 else if (attempts < 100) xfs_middle_retries++;
2327 else xfs_lots_retries++;
2335 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2336 int remove_which_error_return = 0;
2338 #define REMOVE_DEBUG_TRACE(x)
2339 #endif /* ! DEBUG */
2348 bhv_desc_t *dir_bdp,
2353 char *name = VNAME(dentry);
2354 xfs_inode_t *dp, *ip;
2355 xfs_trans_t *tp = NULL;
2358 xfs_bmap_free_t free_list;
2359 xfs_fsblock_t first_block;
2367 dir_vp = BHV_TO_VNODE(dir_bdp);
2368 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2370 dp = XFS_BHVTOI(dir_bdp);
2373 if (XFS_FORCED_SHUTDOWN(mp))
2374 return XFS_ERROR(EIO);
2376 namelen = VNAMELEN(dentry);
2378 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2379 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2380 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2381 name, NULL, 0, 0, 0);
2386 /* From this point on, return through std_return */
2390 * We need to get a reference to ip before we get our log
2391 * reservation. The reason for this is that we cannot call
2392 * xfs_iget for an inode for which we do not have a reference
2393 * once we've acquired a log reservation. This is because the
2394 * inode we are trying to get might be in xfs_inactive going
2395 * for a log reservation. Since we'll have to wait for the
2396 * inactive code to complete before returning from xfs_iget,
2397 * we need to make sure that we don't have log space reserved
2398 * when we call xfs_iget. Instead we get an unlocked referece
2399 * to the inode before getting our log reservation.
2401 error = xfs_get_dir_entry(dentry, &ip);
2403 REMOVE_DEBUG_TRACE(__LINE__);
2407 dm_di_mode = ip->i_d.di_mode;
2409 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2413 error = XFS_QM_DQATTACH(mp, dp, 0);
2414 if (!error && dp != ip)
2415 error = XFS_QM_DQATTACH(mp, ip, 0);
2417 REMOVE_DEBUG_TRACE(__LINE__);
2422 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2423 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2425 * We try to get the real space reservation first,
2426 * allowing for directory btree deletion(s) implying
2427 * possible bmap insert(s). If we can't get the space
2428 * reservation then we use 0 instead, and avoid the bmap
2429 * btree insert(s) in the directory code by, if the bmap
2430 * insert tries to happen, instead trimming the LAST
2431 * block from the directory.
2433 resblks = XFS_REMOVE_SPACE_RES(mp);
2434 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2435 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2436 if (error == ENOSPC) {
2438 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2439 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2442 ASSERT(error != ENOSPC);
2443 REMOVE_DEBUG_TRACE(__LINE__);
2444 xfs_trans_cancel(tp, 0);
2449 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2451 REMOVE_DEBUG_TRACE(__LINE__);
2452 xfs_trans_cancel(tp, cancel_flags);
2458 * At this point, we've gotten both the directory and the entry
2461 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2464 * Increment vnode ref count only in this case since
2465 * there's an extra vnode reference in the case where
2469 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2472 if ((error = _MAC_XFS_IACCESS(ip, MACWRITE, credp))) {
2473 REMOVE_DEBUG_TRACE(__LINE__);
2478 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2480 XFS_BMAP_INIT(&free_list, &first_block);
2481 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2482 &first_block, &free_list, 0);
2484 ASSERT(error != ENOENT);
2485 REMOVE_DEBUG_TRACE(__LINE__);
2488 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2491 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2493 error = xfs_droplink(tp, ip);
2495 REMOVE_DEBUG_TRACE(__LINE__);
2499 /* Determine if this is the last link while
2500 * we are in the transaction.
2502 link_zero = (ip)->i_d.di_nlink==0;
2505 * Take an extra ref on the inode so that it doesn't
2506 * go to xfs_inactive() from within the commit.
2511 * If this is a synchronous mount, make sure that the
2512 * remove transaction goes to disk before returning to
2515 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2516 xfs_trans_set_sync(tp);
2519 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2521 REMOVE_DEBUG_TRACE(__LINE__);
2525 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2532 * Before we drop our extra reference to the inode, purge it
2533 * from the refcache if it is there. By waiting until afterwards
2534 * to do the IRELE, we ensure that we won't go inactive in the
2535 * xfs_refcache_purge_ip routine (although that would be OK).
2537 xfs_refcache_purge_ip(ip);
2539 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2542 * Let interposed file systems know about removed links.
2544 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2548 /* Fall through to std_return with error = 0 */
2550 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2551 DM_EVENT_POSTREMOVE)) {
2552 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2553 dir_vp, DM_RIGHT_NULL,
2554 NULL, DM_RIGHT_NULL,
2555 name, NULL, dm_di_mode, error, 0);
2560 xfs_bmap_cancel(&free_list);
2561 cancel_flags |= XFS_TRANS_ABORT;
2564 xfs_trans_cancel(tp, cancel_flags);
2569 * In this case make sure to not release the inode until after
2570 * the current transaction is aborted. Releasing it beforehand
2571 * can cause us to go to xfs_inactive and start a recursive
2572 * transaction which can easily deadlock with the current one.
2574 xfs_bmap_cancel(&free_list);
2575 cancel_flags |= XFS_TRANS_ABORT;
2576 xfs_trans_cancel(tp, cancel_flags);
2579 * Before we drop our extra reference to the inode, purge it
2580 * from the refcache if it is there. By waiting until afterwards
2581 * to do the IRELE, we ensure that we won't go inactive in the
2582 * xfs_refcache_purge_ip routine (although that would be OK).
2584 xfs_refcache_purge_ip(ip);
2598 bhv_desc_t *target_dir_bdp,
2603 xfs_inode_t *tdp, *sip;
2606 xfs_inode_t *ips[2];
2608 xfs_bmap_free_t free_list;
2609 xfs_fsblock_t first_block;
2612 vnode_t *target_dir_vp;
2613 bhv_desc_t *src_bdp;
2615 char *target_name = VNAME(dentry);
2618 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2619 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2620 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2622 target_namelen = VNAMELEN(dentry);
2623 if (src_vp->v_type == VDIR)
2624 return XFS_ERROR(EPERM);
2627 * For now, manually find the XFS behavior descriptor for
2628 * the source vnode. If it doesn't exist then something
2629 * is wrong and we should just return an error.
2630 * Eventually we need to figure out how link is going to
2631 * work in the face of stacked vnodes.
2633 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2634 if (src_bdp == NULL) {
2635 return XFS_ERROR(EXDEV);
2637 sip = XFS_BHVTOI(src_bdp);
2638 tdp = XFS_BHVTOI(target_dir_bdp);
2640 if (XFS_FORCED_SHUTDOWN(mp))
2641 return XFS_ERROR(EIO);
2643 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2644 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2645 target_dir_vp, DM_RIGHT_NULL,
2646 src_vp, DM_RIGHT_NULL,
2647 target_name, NULL, 0, 0, 0);
2652 /* Return through std_return after this point. */
2654 error = XFS_QM_DQATTACH(mp, sip, 0);
2655 if (!error && sip != tdp)
2656 error = XFS_QM_DQATTACH(mp, tdp, 0);
2660 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2661 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2662 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2663 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2664 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2665 if (error == ENOSPC) {
2667 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2668 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2675 if (sip->i_ino < tdp->i_ino) {
2683 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2686 * Increment vnode ref counts since xfs_trans_commit &
2687 * xfs_trans_cancel will both unlock the inodes and
2688 * decrement the associated ref counts.
2691 VN_HOLD(target_dir_vp);
2692 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2693 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2696 * If the source has too many links, we can't make any more to it.
2698 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2699 error = XFS_ERROR(EMLINK);
2704 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2708 XFS_BMAP_INIT(&free_list, &first_block);
2710 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2711 sip->i_ino, &first_block, &free_list,
2715 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2717 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2719 error = xfs_bumplink(tp, sip);
2725 * If this is a synchronous mount, make sure that the
2726 * link transaction goes to disk before returning to
2729 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2730 xfs_trans_set_sync(tp);
2733 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2735 xfs_bmap_cancel(&free_list);
2739 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2744 /* Fall through to std_return with error = 0. */
2746 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2747 DM_EVENT_POSTLINK)) {
2748 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2749 target_dir_vp, DM_RIGHT_NULL,
2750 src_vp, DM_RIGHT_NULL,
2751 target_name, NULL, 0, error, 0);
2756 cancel_flags |= XFS_TRANS_ABORT;
2759 xfs_trans_cancel(tp, cancel_flags);
2769 bhv_desc_t *dir_bdp,
2775 char *dir_name = VNAME(dentry);
2777 xfs_inode_t *cdp; /* inode of created dir */
2778 vnode_t *cvp; /* vnode of created dir */
2784 xfs_bmap_free_t free_list;
2785 xfs_fsblock_t first_block;
2787 boolean_t dp_joined_to_trans;
2788 boolean_t created = B_FALSE;
2789 int dm_event_sent = 0;
2791 struct xfs_dquot *udqp, *gdqp;
2796 dir_vp = BHV_TO_VNODE(dir_bdp);
2797 dp = XFS_BHVTOI(dir_bdp);
2800 if (XFS_FORCED_SHUTDOWN(mp))
2801 return XFS_ERROR(EIO);
2803 dir_namelen = VNAMELEN(dentry);
2806 dp_joined_to_trans = B_FALSE;
2807 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2809 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2810 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2811 dir_vp, DM_RIGHT_NULL, NULL,
2812 DM_RIGHT_NULL, dir_name, NULL,
2819 /* Return through std_return after this point. */
2821 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2825 if (vap->va_mask & XFS_AT_PROJID)
2826 prid = (xfs_prid_t)vap->va_projid;
2828 prid = (xfs_prid_t)dfltprid;
2831 * Make sure that we have allocated dquot(s) on disk.
2833 error = XFS_QM_DQVOPALLOC(mp, dp,
2834 current_fsuid(credp), current_fsgid(credp),
2835 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2839 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2840 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2841 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2842 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2843 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2844 if (error == ENOSPC) {
2846 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2847 XFS_TRANS_PERM_LOG_RES,
2848 XFS_MKDIR_LOG_COUNT);
2856 xfs_ilock(dp, XFS_ILOCK_EXCL);
2859 * Check for directory link count overflow.
2861 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2862 error = XFS_ERROR(EMLINK);
2867 * Reserve disk quota and the inode.
2869 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2874 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2877 * create the directory inode.
2879 error = xfs_dir_ialloc(&tp, dp,
2880 MAKEIMODE(vap->va_type,vap->va_mode), 2,
2881 0, credp, prid, resblks > 0,
2884 if (error == ENOSPC)
2891 * Now we add the directory inode to the transaction.
2892 * We waited until now since xfs_dir_ialloc might start
2893 * a new transaction. Had we joined the transaction
2894 * earlier, the locks might have gotten released.
2897 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2898 dp_joined_to_trans = B_TRUE;
2900 XFS_BMAP_INIT(&free_list, &first_block);
2902 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2903 cdp->i_ino, &first_block, &free_list,
2904 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2906 ASSERT(error != ENOSPC);
2909 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2912 * Bump the in memory version number of the parent directory
2913 * so that other processes accessing it will recognize that
2914 * the directory has changed.
2918 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2924 error = xfs_bumplink(tp, dp);
2929 cvp = XFS_ITOV(cdp);
2937 * Attach the dquots to the new inode and modify the icount incore.
2939 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2942 * If this is a synchronous mount, make sure that the
2943 * mkdir transaction goes to disk before returning to
2946 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2947 xfs_trans_set_sync(tp);
2950 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2956 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2957 XFS_QM_DQRELE(mp, udqp);
2958 XFS_QM_DQRELE(mp, gdqp);
2963 /* Fall through to std_return with error = 0 or errno from
2964 * xfs_trans_commit. */
2967 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2968 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2969 DM_EVENT_POSTCREATE)) {
2970 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2971 dir_vp, DM_RIGHT_NULL,
2972 created ? XFS_ITOV(cdp):NULL,
2975 dm_di_mode, error, 0);
2981 xfs_bmap_cancel(&free_list);
2983 cancel_flags |= XFS_TRANS_ABORT;
2985 xfs_trans_cancel(tp, cancel_flags);
2986 XFS_QM_DQRELE(mp, udqp);
2987 XFS_QM_DQRELE(mp, gdqp);
2989 if (!dp_joined_to_trans && (dp != NULL)) {
2990 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3003 bhv_desc_t *dir_bdp,
3007 char *name = VNAME(dentry);
3009 xfs_inode_t *cdp; /* child directory */
3013 xfs_bmap_free_t free_list;
3014 xfs_fsblock_t first_block;
3023 dir_vp = BHV_TO_VNODE(dir_bdp);
3024 dp = XFS_BHVTOI(dir_bdp);
3027 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3029 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3030 return XFS_ERROR(EIO);
3031 namelen = VNAMELEN(dentry);
3033 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3034 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3035 dir_vp, DM_RIGHT_NULL,
3036 NULL, DM_RIGHT_NULL,
3037 name, NULL, 0, 0, 0);
3039 return XFS_ERROR(error);
3042 /* Return through std_return after this point. */
3047 * We need to get a reference to cdp before we get our log
3048 * reservation. The reason for this is that we cannot call
3049 * xfs_iget for an inode for which we do not have a reference
3050 * once we've acquired a log reservation. This is because the
3051 * inode we are trying to get might be in xfs_inactive going
3052 * for a log reservation. Since we'll have to wait for the
3053 * inactive code to complete before returning from xfs_iget,
3054 * we need to make sure that we don't have log space reserved
3055 * when we call xfs_iget. Instead we get an unlocked referece
3056 * to the inode before getting our log reservation.
3058 error = xfs_get_dir_entry(dentry, &cdp);
3060 REMOVE_DEBUG_TRACE(__LINE__);
3064 dm_di_mode = cdp->i_d.di_mode;
3067 * Get the dquots for the inodes.
3069 error = XFS_QM_DQATTACH(mp, dp, 0);
3070 if (!error && dp != cdp)
3071 error = XFS_QM_DQATTACH(mp, cdp, 0);
3074 REMOVE_DEBUG_TRACE(__LINE__);
3078 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3079 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3081 * We try to get the real space reservation first,
3082 * allowing for directory btree deletion(s) implying
3083 * possible bmap insert(s). If we can't get the space
3084 * reservation then we use 0 instead, and avoid the bmap
3085 * btree insert(s) in the directory code by, if the bmap
3086 * insert tries to happen, instead trimming the LAST
3087 * block from the directory.
3089 resblks = XFS_REMOVE_SPACE_RES(mp);
3090 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3091 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3092 if (error == ENOSPC) {
3094 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3095 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3098 ASSERT(error != ENOSPC);
3103 XFS_BMAP_INIT(&free_list, &first_block);
3106 * Now lock the child directory inode and the parent directory
3107 * inode in the proper order. This will take care of validating
3108 * that the directory entry for the child directory inode has
3109 * not changed while we were obtaining a log reservation.
3111 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3113 xfs_trans_cancel(tp, cancel_flags);
3118 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3121 * Only increment the parent directory vnode count if
3122 * we didn't bump it in looking up cdp. The only time
3123 * we don't bump it is when we're looking up ".".
3129 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3131 if ((error = _MAC_XFS_IACCESS(cdp, MACWRITE, credp))) {
3135 ASSERT(cdp->i_d.di_nlink >= 2);
3136 if (cdp->i_d.di_nlink != 2) {
3137 error = XFS_ERROR(ENOTEMPTY);
3140 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3141 error = XFS_ERROR(ENOTEMPTY);
3145 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3146 &first_block, &free_list, resblks);
3151 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3154 * Bump the in memory generation count on the parent
3155 * directory so that other can know that it has changed.
3160 * Drop the link from cdp's "..".
3162 error = xfs_droplink(tp, dp);
3168 * Drop the link from dp to cdp.
3170 error = xfs_droplink(tp, cdp);
3176 * Drop the "." link from cdp to self.
3178 error = xfs_droplink(tp, cdp);
3183 /* Determine these before committing transaction */
3184 last_cdp_link = (cdp)->i_d.di_nlink==0;
3187 * Take an extra ref on the child vnode so that it
3188 * does not go to xfs_inactive() from within the commit.
3193 * If this is a synchronous mount, make sure that the
3194 * rmdir transaction goes to disk before returning to
3197 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3198 xfs_trans_set_sync(tp);
3201 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3203 xfs_bmap_cancel(&free_list);
3204 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3210 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3218 * Let interposed file systems know about removed links.
3220 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3224 /* Fall through to std_return with error = 0 or the errno
3225 * from xfs_trans_commit. */
3227 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3228 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3229 dir_vp, DM_RIGHT_NULL,
3230 NULL, DM_RIGHT_NULL,
3231 name, NULL, dm_di_mode,
3237 xfs_bmap_cancel(&free_list);
3238 cancel_flags |= XFS_TRANS_ABORT;
3240 xfs_trans_cancel(tp, cancel_flags);
3248 * Read dp's entries starting at uiop->uio_offset and translate them into
3249 * bufsize bytes worth of struct dirents starting at bufbase.
3253 bhv_desc_t *dir_bdp,
3259 xfs_trans_t *tp = NULL;
3262 xfs_off_t start_offset;
3264 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3265 (inst_t *)__return_address);
3266 dp = XFS_BHVTOI(dir_bdp);
3268 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3269 return XFS_ERROR(EIO);
3272 lock_mode = xfs_ilock_map_shared(dp);
3273 start_offset = uiop->uio_offset;
3274 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3275 if (start_offset != uiop->uio_offset) {
3276 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3278 xfs_iunlock_map_shared(dp, lock_mode);
3289 bhv_desc_t *dir_bdp,
3302 xfs_bmap_free_t free_list;
3303 xfs_fsblock_t first_block;
3304 boolean_t dp_joined_to_trans;
3308 xfs_fileoff_t first_fsb;
3309 xfs_filblks_t fs_blocks;
3311 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3318 struct xfs_dquot *udqp, *gdqp;
3320 char *link_name = VNAME(dentry);
3324 dir_vp = BHV_TO_VNODE(dir_bdp);
3325 dp = XFS_BHVTOI(dir_bdp);
3326 dp_joined_to_trans = B_FALSE;
3331 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3335 if (XFS_FORCED_SHUTDOWN(mp))
3336 return XFS_ERROR(EIO);
3338 link_namelen = VNAMELEN(dentry);
3341 * Check component lengths of the target path name.
3343 pathlen = strlen(target_path);
3344 if (pathlen >= MAXPATHLEN) /* total string too long */
3345 return XFS_ERROR(ENAMETOOLONG);
3346 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3350 for(total = 0, path = target_path; total < pathlen;) {
3354 while(*path == '/') {
3360 * Count up to the next slash or end of path.
3361 * Error out if the component is bigger than MAXNAMELEN.
3363 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3364 if (++len >= MAXNAMELEN) {
3365 error = ENAMETOOLONG;
3372 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3373 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3374 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3375 link_name, target_path, 0, 0, 0);
3380 /* Return through std_return after this point. */
3383 if (vap->va_mask & XFS_AT_PROJID)
3384 prid = (xfs_prid_t)vap->va_projid;
3386 prid = (xfs_prid_t)dfltprid;
3389 * Make sure that we have allocated dquot(s) on disk.
3391 error = XFS_QM_DQVOPALLOC(mp, dp,
3392 current_fsuid(credp), current_fsgid(credp),
3393 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3397 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3398 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3400 * The symlink will fit into the inode data fork?
3401 * There can't be any attributes so we get the whole variable part.
3403 if (pathlen <= XFS_LITINO(mp))
3406 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3407 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3408 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3409 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3410 if (error == ENOSPC && fs_blocks == 0) {
3412 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3413 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3421 xfs_ilock(dp, XFS_ILOCK_EXCL);
3424 * Reserve disk quota : blocks and inode.
3426 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3431 * Check for ability to enter directory entry, if no space reserved.
3434 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3437 * Initialize the bmap freelist prior to calling either
3438 * bmapi or the directory create code.
3440 XFS_BMAP_INIT(&free_list, &first_block);
3443 * Allocate an inode for the symlink.
3445 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3446 1, 0, credp, prid, resblks > 0, &ip, NULL);
3448 if (error == ENOSPC)
3455 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3456 dp_joined_to_trans = B_TRUE;
3459 * Also attach the dquot(s) to it, if applicable.
3461 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3464 resblks -= XFS_IALLOC_SPACE_RES(mp);
3466 * If the symlink will fit into the inode, write it inline.
3468 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3469 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3470 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3471 ip->i_d.di_size = pathlen;
3474 * The inode was initially created in extent format.
3476 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3477 ip->i_df.if_flags |= XFS_IFINLINE;
3479 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3480 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3484 nmaps = SYMLINK_MAPS;
3486 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3487 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3488 &first_block, resblks, mval, &nmaps,
3495 resblks -= fs_blocks;
3496 ip->i_d.di_size = pathlen;
3497 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3499 cur_chunk = target_path;
3500 for (n = 0; n < nmaps; n++) {
3501 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3502 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3503 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3504 BTOBB(byte_cnt), 0);
3505 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3506 if (pathlen < byte_cnt) {
3509 pathlen -= byte_cnt;
3511 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3512 cur_chunk += byte_cnt;
3514 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3519 * Create the directory entry for the symlink.
3521 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3522 ip->i_ino, &first_block, &free_list, resblks);
3526 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3527 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3530 * Bump the in memory version number of the parent directory
3531 * so that other processes accessing it will recognize that
3532 * the directory has changed.
3537 * If this is a synchronous mount, make sure that the
3538 * symlink transaction goes to disk before returning to
3541 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3542 xfs_trans_set_sync(tp);
3546 * xfs_trans_commit normally decrements the vnode ref count
3547 * when it unlocks the inode. Since we want to return the
3548 * vnode to the caller, we bump the vnode ref count now.
3552 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3556 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3557 XFS_QM_DQRELE(mp, udqp);
3558 XFS_QM_DQRELE(mp, gdqp);
3560 /* Fall through to std_return with error = 0 or errno from
3561 * xfs_trans_commit */
3563 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3564 DM_EVENT_POSTSYMLINK)) {
3565 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3566 dir_vp, DM_RIGHT_NULL,
3567 error ? NULL : XFS_ITOV(ip),
3568 DM_RIGHT_NULL, link_name, target_path,
3584 xfs_bmap_cancel(&free_list);
3585 cancel_flags |= XFS_TRANS_ABORT;
3587 xfs_trans_cancel(tp, cancel_flags);
3588 XFS_QM_DQRELE(mp, udqp);
3589 XFS_QM_DQRELE(mp, gdqp);
3591 if (!dp_joined_to_trans && (dp != NULL)) {
3592 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3602 * A fid routine that takes a pointer to a previously allocated
3603 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3613 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3614 (inst_t *)__return_address);
3615 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3617 xfid = (xfs_fid2_t *)fidp;
3618 ip = XFS_BHVTOI(bdp);
3619 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3622 * use memcpy because the inode is a long long and there's no
3623 * assurance that xfid->fid_ino is properly aligned.
3625 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3626 xfid->fid_gen = ip->i_d.di_gen;
3643 vp = BHV_TO_VNODE(bdp);
3644 if (vp->v_type == VDIR)
3646 ip = XFS_BHVTOI(bdp);
3647 if (locktype == VRWLOCK_WRITE) {
3648 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3649 } else if (locktype == VRWLOCK_TRY_READ) {
3650 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3651 } else if (locktype == VRWLOCK_TRY_WRITE) {
3652 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3654 ASSERT((locktype == VRWLOCK_READ) ||
3655 (locktype == VRWLOCK_WRITE_DIRECT));
3656 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3674 vp = BHV_TO_VNODE(bdp);
3675 if (vp->v_type == VDIR)
3677 ip = XFS_BHVTOI(bdp);
3678 if (locktype == VRWLOCK_WRITE) {
3680 * In the write case, we may have added a new entry to
3681 * the reference cache. This might store a pointer to
3682 * an inode to be released in this inode. If it is there,
3683 * clear the pointer and release the inode after unlocking
3686 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3688 ASSERT((locktype == VRWLOCK_READ) ||
3689 (locktype == VRWLOCK_WRITE_DIRECT));
3690 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3704 ip = XFS_BHVTOI(bdp);
3707 if (XFS_FORCED_SHUTDOWN(mp))
3708 return XFS_ERROR(EIO);
3710 /* Bypass inodes which have already been cleaned by
3711 * the inode flush clustering code inside xfs_iflush
3713 if ((ip->i_update_core == 0) &&
3714 ((ip->i_itemp == NULL) ||
3715 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)))
3718 if (flags & FLUSH_LOG) {
3719 xfs_inode_log_item_t *iip = ip->i_itemp;
3721 if (iip && iip->ili_last_lsn) {
3722 xlog_t *log = mp->m_log;
3724 int s, log_flags = XFS_LOG_FORCE;
3726 s = GRANT_LOCK(log);
3727 sync_lsn = log->l_last_sync_lsn;
3728 GRANT_UNLOCK(log, s);
3730 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3733 if (flags & FLUSH_SYNC)
3734 log_flags |= XFS_LOG_SYNC;
3735 return xfs_log_force(mp, iip->ili_last_lsn,
3740 /* We make this non-blocking if the inode is contended,
3741 * return EAGAIN to indicate to the caller that they
3742 * did not succeed. This prevents the flush path from
3743 * blocking on inodes inside another operation right
3744 * now, they get caught later by xfs_sync.
3746 if (flags & FLUSH_INODE) {
3749 if (xfs_ipincount(ip))
3752 if (flags & FLUSH_SYNC) {
3753 xfs_ilock(ip, XFS_ILOCK_SHARED);
3755 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3756 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3757 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3764 if (flags & FLUSH_SYNC)
3765 flush_flags = XFS_IFLUSH_SYNC;
3767 flush_flags = XFS_IFLUSH_ASYNC;
3769 error = xfs_iflush(ip, flush_flags);
3770 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3789 if (!capable(CAP_SYS_ADMIN))
3790 return XFS_ERROR(EPERM);
3792 ip = XFS_BHVTOI(bdp);
3795 if (XFS_FORCED_SHUTDOWN(mp))
3796 return XFS_ERROR(EIO);
3798 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3799 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3801 xfs_trans_cancel(tp, 0);
3804 xfs_ilock(ip, XFS_ILOCK_EXCL);
3805 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3807 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3808 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3810 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3812 error = xfs_trans_commit(tp, 0, NULL);
3828 vp = BHV_TO_VNODE(bdp);
3830 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3832 ASSERT(!VN_MAPPED(vp));
3833 ip = XFS_BHVTOI(bdp);
3835 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3836 if (ip->i_d.di_size > 0) {
3838 * Flush and invalidate any data left around that is
3839 * a part of this file.
3841 * Get the inode's i/o lock so that buffers are pushed
3842 * out while holding the proper lock. We can't hold
3843 * the inode lock here since flushing out buffers may
3844 * cause us to try to get the lock in xfs_strategy().
3846 * We don't have to call remapf() here, because there
3847 * cannot be any mapped file references to this vnode
3848 * since it is being reclaimed.
3850 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3853 * If we hit an IO error, we need to make sure that the
3854 * buffer and page caches of file data for
3855 * the file are tossed away. We don't want to use
3856 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3857 * pages to stay attached to the vnode, but be
3858 * marked P_BAD. pdflush/vnode_pagebad
3861 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3862 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3864 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3867 ASSERT(VN_CACHED(vp) == 0);
3868 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3869 ip->i_delayed_blks == 0);
3870 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3871 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3873 * di_size field may not be quite accurate if we're
3876 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3877 ASSERT(VN_CACHED(vp) == 0);
3881 /* If we have nothing to flush with this inode then complete the
3882 * teardown now, otherwise break the link between the xfs inode
3883 * and the linux inode and clean up the xfs inode later. This
3884 * avoids flushing the inode to disk during the delete operation
3887 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3888 xfs_ilock(ip, XFS_ILOCK_EXCL);
3890 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3892 xfs_mount_t *mp = ip->i_mount;
3894 /* Protect sync from us */
3895 XFS_MOUNT_ILOCK(mp);
3896 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3897 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3898 ip->i_flags |= XFS_IRECLAIMABLE;
3899 XFS_MOUNT_IUNLOCK(mp);
3910 xfs_ihash_t *ih = ip->i_hash;
3913 /* The hash lock here protects a thread in xfs_iget_core from
3914 * racing with us on linking the inode back with a vnode.
3915 * Once we have the XFS_IRECLAIM flag set it will not touch
3918 write_lock(&ih->ih_lock);
3919 if ((ip->i_flags & XFS_IRECLAIM) ||
3920 (!(ip->i_flags & XFS_IRECLAIMABLE) &&
3921 (XFS_ITOV_NULL(ip) == NULL))) {
3922 write_unlock(&ih->ih_lock);
3925 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3929 ip->i_flags |= XFS_IRECLAIM;
3930 write_unlock(&ih->ih_lock);
3933 * If the inode is still dirty, then flush it out. If the inode
3934 * is not in the AIL, then it will be OK to flush it delwri as
3935 * long as xfs_iflush() does not keep any references to the inode.
3936 * We leave that decision up to xfs_iflush() since it has the
3937 * knowledge of whether it's OK to simply do a delwri flush of
3938 * the inode or whether we need to wait until the inode is
3939 * pulled from the AIL.
3940 * We get the flush lock regardless, though, just to make sure
3941 * we don't free it while it is being flushed.
3943 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3945 xfs_ilock(ip, XFS_ILOCK_EXCL);
3949 if (ip->i_update_core ||
3950 ((ip->i_itemp != NULL) &&
3951 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3952 error = xfs_iflush(ip, sync_mode);
3954 * If we hit an error, typically because of filesystem
3955 * shutdown, we don't need to let vn_reclaim to know
3956 * because we're gonna reclaim the inode anyway.
3959 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3963 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3966 ASSERT(ip->i_update_core == 0);
3967 ASSERT(ip->i_itemp == NULL ||
3968 ip->i_itemp->ili_format.ilf_fields == 0);
3969 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3970 } else if (locked) {
3972 * We are not interested in doing an iflush if we're
3973 * in the process of shutting down the filesystem forcibly.
3974 * So, just reclaim the inode.
3977 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3985 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3988 struct list_head *curr, *next;
3994 XFS_MOUNT_ILOCK(mp);
3995 list_for_each_safe(curr, next, &mp->m_del_inodes) {
3996 ip = list_entry(curr, xfs_inode_t, i_reclaim);
3998 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
4000 if (xfs_ipincount(ip) ||
4001 !xfs_iflock_nowait(ip)) {
4002 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4006 XFS_MOUNT_IUNLOCK(mp);
4007 xfs_finish_reclaim(ip, noblock,
4008 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
4016 XFS_MOUNT_IUNLOCK(mp);
4021 * xfs_alloc_file_space()
4022 * This routine allocates disk space for the given file.
4024 * If alloc_type == 0, this request is for an ALLOCSP type
4025 * request which will change the file size. In this case, no
4026 * DMAPI event will be generated by the call. A TRUNCATE event
4027 * will be generated later by xfs_setattr.
4029 * If alloc_type != 0, this request is for a RESVSP type
4030 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4031 * lower block boundary byte address is less than the file's
4040 xfs_alloc_file_space(
4047 xfs_filblks_t allocated_fsb;
4048 xfs_filblks_t allocatesize_fsb;
4051 xfs_filblks_t datablocks;
4053 xfs_fsblock_t firstfsb;
4054 xfs_bmap_free_t free_list;
4055 xfs_bmbt_irec_t *imapp;
4056 xfs_bmbt_irec_t imaps[1];
4063 xfs_fileoff_t startoffset_fsb;
4065 int xfs_bmapi_flags;
4067 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4070 if (XFS_FORCED_SHUTDOWN(mp))
4071 return XFS_ERROR(EIO);
4074 * determine if this is a realtime file
4076 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4077 if (ip->i_d.di_extsize)
4078 rtextsize = ip->i_d.di_extsize;
4080 rtextsize = mp->m_sb.sb_rextsize;
4084 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4088 return XFS_ERROR(EINVAL);
4094 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4095 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4096 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4098 /* Generate a DMAPI event if needed. */
4099 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4100 (attr_flags&ATTR_DMI) == 0 &&
4101 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4102 xfs_off_t end_dmi_offset;
4104 end_dmi_offset = offset+len;
4105 if (end_dmi_offset > ip->i_d.di_size)
4106 end_dmi_offset = ip->i_d.di_size;
4107 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4108 offset, end_dmi_offset - offset,
4115 * allocate file space until done or until there is an error
4118 while (allocatesize_fsb && !error) {
4120 * determine if reserving space on
4121 * the data or realtime partition.
4126 s = startoffset_fsb;
4127 do_div(s, rtextsize);
4129 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4131 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4134 datablocks = allocatesize_fsb;
4139 * allocate and setup the transaction
4141 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4142 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4143 error = xfs_trans_reserve(tp,
4145 XFS_WRITE_LOG_RES(mp),
4147 XFS_TRANS_PERM_LOG_RES,
4148 XFS_WRITE_LOG_COUNT);
4151 * check for running out of space
4155 * Free the transaction structure.
4157 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4158 xfs_trans_cancel(tp, 0);
4161 xfs_ilock(ip, XFS_ILOCK_EXCL);
4162 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4163 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4164 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4168 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4169 xfs_trans_ihold(tp, ip);
4172 * issue the bmapi() call to allocate the blocks
4174 XFS_BMAP_INIT(&free_list, &firstfsb);
4175 error = xfs_bmapi(tp, ip, startoffset_fsb,
4176 allocatesize_fsb, xfs_bmapi_flags,
4177 &firstfsb, 0, imapp, &reccount,
4184 * complete the transaction
4186 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4191 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4192 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4197 allocated_fsb = imapp->br_blockcount;
4199 if (reccount == 0) {
4200 error = XFS_ERROR(ENOSPC);
4204 startoffset_fsb += allocated_fsb;
4205 allocatesize_fsb -= allocated_fsb;
4208 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4209 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4211 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4212 XFS_ITOV(ip), DM_RIGHT_NULL,
4213 XFS_ITOV(ip), DM_RIGHT_NULL,
4214 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4216 goto retry; /* Maybe DMAPI app. has made space */
4217 /* else fall through with error from XFS_SEND_DATA */
4223 xfs_bmap_cancel(&free_list);
4225 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4226 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4227 goto dmapi_enospc_check;
4231 * Zero file bytes between startoff and endoff inclusive.
4232 * The iolock is held exclusive and no blocks are buffered.
4235 xfs_zero_remaining_bytes(
4240 xfs_bmbt_irec_t imap;
4241 xfs_fileoff_t offset_fsb;
4242 xfs_off_t lastoffset;
4245 xfs_mount_t *mp = ip->i_mount;
4249 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4250 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4251 mp->m_rtdev_targp : mp->m_ddev_targp);
4253 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4254 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4256 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4258 if (error || nimap < 1)
4260 ASSERT(imap.br_blockcount >= 1);
4261 ASSERT(imap.br_startoff == offset_fsb);
4262 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4263 if (lastoffset > endoff)
4264 lastoffset = endoff;
4265 if (imap.br_startblock == HOLESTARTBLOCK)
4267 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4268 if (imap.br_state == XFS_EXT_UNWRITTEN)
4271 XFS_BUF_UNWRITE(bp);
4273 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4275 if ((error = xfs_iowait(bp))) {
4276 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4277 mp, bp, XFS_BUF_ADDR(bp));
4280 memset(XFS_BUF_PTR(bp) +
4281 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4282 0, lastoffset - offset + 1);
4287 if ((error = xfs_iowait(bp))) {
4288 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4289 mp, bp, XFS_BUF_ADDR(bp));
4298 * xfs_free_file_space()
4299 * This routine frees disk space for the given file.
4301 * This routine is only called by xfs_change_file_space
4302 * for an UNRESVSP type call.
4310 xfs_free_file_space(
4318 xfs_off_t end_dmi_offset;
4319 xfs_fileoff_t endoffset_fsb;
4321 xfs_fsblock_t firstfsb;
4322 xfs_bmap_free_t free_list;
4324 xfs_bmbt_irec_t imap;
4332 xfs_fileoff_t startoffset_fsb;
4335 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4338 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4342 if (len <= 0) /* if nothing being freed */
4344 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4345 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4346 end_dmi_offset = offset + len;
4347 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4349 if (offset < ip->i_d.di_size &&
4350 (attr_flags & ATTR_DMI) == 0 &&
4351 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4352 if (end_dmi_offset > ip->i_d.di_size)
4353 end_dmi_offset = ip->i_d.di_size;
4354 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4355 offset, end_dmi_offset - offset,
4356 AT_DELAY_FLAG(attr_flags), NULL);
4361 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4362 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4364 ilen = len + (offset & (rounding - 1));
4365 ioffset = offset & ~(rounding - 1);
4366 if (ilen & (rounding - 1))
4367 ilen = (ilen + rounding) & ~(rounding - 1);
4368 xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4370 * Need to zero the stuff we're not freeing, on disk.
4371 * If its a realtime file & can't use unwritten extents then we
4372 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4373 * will take care of it for us.
4375 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4377 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4378 &imap, &nimap, NULL);
4381 ASSERT(nimap == 0 || nimap == 1);
4382 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4385 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4386 block = imap.br_startblock;
4387 mod = do_div(block, mp->m_sb.sb_rextsize);
4389 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4392 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4393 &imap, &nimap, NULL);
4396 ASSERT(nimap == 0 || nimap == 1);
4397 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4398 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4400 if (mod && (mod != mp->m_sb.sb_rextsize))
4401 endoffset_fsb -= mod;
4404 if ((done = (endoffset_fsb <= startoffset_fsb)))
4406 * One contiguous piece to clear
4408 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4411 * Some full blocks, possibly two pieces to clear
4413 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4414 error = xfs_zero_remaining_bytes(ip, offset,
4415 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4417 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4418 error = xfs_zero_remaining_bytes(ip,
4419 XFS_FSB_TO_B(mp, endoffset_fsb),
4424 * free file space until done or until there is an error
4426 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4427 while (!error && !done) {
4430 * allocate and setup the transaction
4432 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4433 error = xfs_trans_reserve(tp,
4435 XFS_WRITE_LOG_RES(mp),
4437 XFS_TRANS_PERM_LOG_RES,
4438 XFS_WRITE_LOG_COUNT);
4441 * check for running out of space
4445 * Free the transaction structure.
4447 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4448 xfs_trans_cancel(tp, 0);
4451 xfs_ilock(ip, XFS_ILOCK_EXCL);
4452 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4453 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4454 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4458 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4459 xfs_trans_ihold(tp, ip);
4462 * issue the bunmapi() call to free the blocks
4464 XFS_BMAP_INIT(&free_list, &firstfsb);
4465 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4466 endoffset_fsb - startoffset_fsb,
4467 0, 2, &firstfsb, &free_list, &done);
4473 * complete the transaction
4475 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4480 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4481 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4484 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4488 xfs_bmap_cancel(&free_list);
4490 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4491 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
4496 * xfs_change_file_space()
4497 * This routine allocates or frees disk space for the given file.
4498 * The user specified parameters are checked for alignment and size
4507 xfs_change_file_space(
4521 xfs_off_t startoffset;
4527 vp = BHV_TO_VNODE(bdp);
4528 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4530 ip = XFS_BHVTOI(bdp);
4534 * must be a regular file and have write permission
4536 if (vp->v_type != VREG)
4537 return XFS_ERROR(EINVAL);
4539 xfs_ilock(ip, XFS_ILOCK_SHARED);
4541 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4542 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4546 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4548 switch (bf->l_whence) {
4549 case 0: /*SEEK_SET*/
4551 case 1: /*SEEK_CUR*/
4552 bf->l_start += offset;
4554 case 2: /*SEEK_END*/
4555 bf->l_start += ip->i_d.di_size;
4558 return XFS_ERROR(EINVAL);
4561 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4563 if ( (bf->l_start < 0)
4564 || (bf->l_start > XFS_MAXIOFFSET(mp))
4565 || (bf->l_start + llen < 0)
4566 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4567 return XFS_ERROR(EINVAL);
4571 startoffset = bf->l_start;
4572 fsize = ip->i_d.di_size;
4575 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4577 * These calls do NOT zero the data space allocated to the file,
4578 * nor do they change the file size.
4580 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4582 * These calls cause the new file data to be zeroed and the file
4583 * size to be changed.
4585 setprealloc = clrprealloc = 0;
4588 case XFS_IOC_RESVSP:
4589 case XFS_IOC_RESVSP64:
4590 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4597 case XFS_IOC_UNRESVSP:
4598 case XFS_IOC_UNRESVSP64:
4599 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4604 case XFS_IOC_ALLOCSP:
4605 case XFS_IOC_ALLOCSP64:
4606 case XFS_IOC_FREESP:
4607 case XFS_IOC_FREESP64:
4608 if (startoffset > fsize) {
4609 error = xfs_alloc_file_space(ip, fsize,
4610 startoffset - fsize, 0, attr_flags);
4615 va.va_mask = XFS_AT_SIZE;
4616 va.va_size = startoffset;
4618 error = xfs_setattr(bdp, &va, attr_flags, credp);
4628 return XFS_ERROR(EINVAL);
4632 * update the inode timestamp, mode, and prealloc flag bits
4634 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4636 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4639 xfs_trans_cancel(tp, 0);
4643 xfs_ilock(ip, XFS_ILOCK_EXCL);
4645 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4646 xfs_trans_ihold(tp, ip);
4648 ip->i_d.di_mode &= ~S_ISUID;
4651 * Note that we don't have to worry about mandatory
4652 * file locking being disabled here because we only
4653 * clear the S_ISGID bit if the Group execute bit is
4654 * on, but if it was on then mandatory locking wouldn't
4655 * have been enabled.
4657 if (ip->i_d.di_mode & S_IXGRP)
4658 ip->i_d.di_mode &= ~S_ISGID;
4660 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4663 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4664 else if (clrprealloc)
4665 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4667 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4668 xfs_trans_set_sync(tp);
4670 error = xfs_trans_commit(tp, 0, NULL);
4672 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4677 vnodeops_t xfs_vnodeops = {
4678 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4679 .vop_open = xfs_open,
4680 .vop_read = xfs_read,
4681 #ifdef HAVE_SENDFILE
4682 .vop_sendfile = xfs_sendfile,
4684 .vop_write = xfs_write,
4685 .vop_ioctl = xfs_ioctl,
4686 .vop_getattr = xfs_getattr,
4687 .vop_setattr = xfs_setattr,
4688 .vop_access = xfs_access,
4689 .vop_lookup = xfs_lookup,
4690 .vop_create = xfs_create,
4691 .vop_remove = xfs_remove,
4692 .vop_link = xfs_link,
4693 .vop_rename = xfs_rename,
4694 .vop_mkdir = xfs_mkdir,
4695 .vop_rmdir = xfs_rmdir,
4696 .vop_readdir = xfs_readdir,
4697 .vop_symlink = xfs_symlink,
4698 .vop_readlink = xfs_readlink,
4699 .vop_fsync = xfs_fsync,
4700 .vop_inactive = xfs_inactive,
4701 .vop_fid2 = xfs_fid2,
4702 .vop_rwlock = xfs_rwlock,
4703 .vop_rwunlock = xfs_rwunlock,
4704 .vop_bmap = xfs_bmap,
4705 .vop_reclaim = xfs_reclaim,
4706 .vop_attr_get = xfs_attr_get,
4707 .vop_attr_set = xfs_attr_set,
4708 .vop_attr_remove = xfs_attr_remove,
4709 .vop_attr_list = xfs_attr_list,
4710 .vop_link_removed = (vop_link_removed_t)fs_noval,
4711 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4712 .vop_tosspages = fs_tosspages,
4713 .vop_flushinval_pages = fs_flushinval_pages,
4714 .vop_flush_pages = fs_flush_pages,
4715 .vop_release = xfs_release,
4716 .vop_iflush = xfs_inode_flush,