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.
251 vap->va_xflags = xfs_dic2xflags(&ip->i_d, ARCH_NOCONVERT);
254 * Exit for inode revalidate. See if any of the rest of
255 * the fields to be filled in are needed.
258 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
259 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
262 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
264 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
265 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
269 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
270 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
271 ip->i_d.di_anextents;
273 vap->va_anextents = 0;
274 vap->va_gen = ip->i_d.di_gen;
277 if (!(flags & ATTR_LAZY))
278 xfs_iunlock(ip, XFS_ILOCK_SHARED);
304 xfs_prid_t projid=0, iprojid=0;
305 int mandlock_before, mandlock_after;
306 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
308 int need_iolock = (flags & ATTR_DMI) == 0;
310 vp = BHV_TO_VNODE(bdp);
311 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
313 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
314 return XFS_ERROR(EROFS);
317 * Cannot set certain attributes.
320 if (mask & XFS_AT_NOSET) {
321 return XFS_ERROR(EINVAL);
324 ip = XFS_BHVTOI(bdp);
327 if (XFS_FORCED_SHUTDOWN(mp))
328 return XFS_ERROR(EIO);
331 * Timestamps do not need to be logged and hence do not
332 * need to be done within a transaction.
334 if (mask & XFS_AT_UPDTIMES) {
335 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
336 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
337 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
338 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
339 xfs_ichgtime(ip, timeflags);
343 olddquot1 = olddquot2 = NULL;
347 * If disk quotas is on, we make sure that the dquots do exist on disk,
348 * before we start any other transactions. Trying to do this later
349 * is messy. We don't care to take a readlock to look at the ids
350 * in inode here, because we can't hold it across the trans_reserve.
351 * If the IDs do change before we take the ilock, we're covered
352 * because the i_*dquot fields will get updated anyway.
354 if (XFS_IS_QUOTA_ON(mp) && (mask & (XFS_AT_UID|XFS_AT_GID))) {
357 if (mask & XFS_AT_UID) {
359 qflags |= XFS_QMOPT_UQUOTA;
361 uid = ip->i_d.di_uid;
363 if (mask & XFS_AT_GID) {
365 qflags |= XFS_QMOPT_GQUOTA;
367 gid = ip->i_d.di_gid;
370 * We take a reference when we initialize udqp and gdqp,
371 * so it is important that we never blindly double trip on
372 * the same variable. See xfs_create() for an example.
374 ASSERT(udqp == NULL);
375 ASSERT(gdqp == NULL);
376 code = XFS_QM_DQVOPALLOC(mp, ip, uid,gid, qflags, &udqp, &gdqp);
382 * For the other attributes, we acquire the inode lock and
383 * first do an error checking pass.
386 lock_flags = XFS_ILOCK_EXCL;
387 if (!(mask & XFS_AT_SIZE)) {
388 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
389 (mp->m_flags & XFS_MOUNT_WSYNC)) {
390 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
392 if ((code = xfs_trans_reserve(tp, 0,
393 XFS_ICHANGE_LOG_RES(mp), 0,
400 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
401 !(flags & ATTR_DMI)) {
402 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
403 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
404 vap->va_size, 0, dmflags, NULL);
411 lock_flags |= XFS_IOLOCK_EXCL;
414 xfs_ilock(ip, lock_flags);
416 /* boolean: are we the file owner? */
417 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
420 * Change various properties of a file.
421 * Only the owner or users with CAP_FOWNER
422 * capability may do these things.
425 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
426 XFS_AT_GID|XFS_AT_PROJID)) {
428 * CAP_FOWNER overrides the following restrictions:
430 * The user ID of the calling process must be equal
431 * to the file owner ID, except in cases where the
432 * CAP_FSETID capability is applicable.
434 if (!file_owner && !capable(CAP_FOWNER)) {
435 code = XFS_ERROR(EPERM);
440 * CAP_FSETID overrides the following restrictions:
442 * The effective user ID of the calling process shall match
443 * the file owner when setting the set-user-ID and
444 * set-group-ID bits on that file.
446 * The effective group ID or one of the supplementary group
447 * IDs of the calling process shall match the group owner of
448 * the file when setting the set-group-ID bit on that file
450 if (mask & XFS_AT_MODE) {
453 if ((vap->va_mode & S_ISUID) && !file_owner)
455 if ((vap->va_mode & S_ISGID) &&
456 !in_group_p((gid_t)ip->i_d.di_gid))
459 /* Linux allows this, Irix doesn't. */
460 if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
463 if (m && !capable(CAP_FSETID))
469 * Change file ownership. Must be the owner or privileged.
470 * If the system was configured with the "restricted_chown"
471 * option, the owner is not permitted to give away the file,
472 * and can change the group id only to a group of which he
473 * or she is a member.
475 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
477 * These IDs could have changed since we last looked at them.
478 * But, we're assured that if the ownership did change
479 * while we didn't have the inode locked, inode's dquot(s)
480 * would have changed also.
482 iuid = ip->i_d.di_uid;
483 iprojid = ip->i_d.di_projid;
484 igid = ip->i_d.di_gid;
485 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
486 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
487 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
491 * CAP_CHOWN overrides the following restrictions:
493 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
494 * shall override the restriction that a process cannot
495 * change the user ID of a file it owns and the restriction
496 * that the group ID supplied to the chown() function
497 * shall be equal to either the group ID or one of the
498 * supplementary group IDs of the calling process.
500 * XXX: How does restricted_chown affect projid?
502 if (restricted_chown &&
503 (iuid != uid || (igid != gid &&
504 !in_group_p((gid_t)gid))) &&
505 !capable(CAP_CHOWN)) {
506 code = XFS_ERROR(EPERM);
510 * Do a quota reservation only if uid or gid is actually
513 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
514 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
516 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
517 capable(CAP_FOWNER) ?
518 XFS_QMOPT_FORCE_RES : 0);
519 if (code) /* out of quota */
525 * Truncate file. Must have write permission and not be a directory.
527 if (mask & XFS_AT_SIZE) {
528 /* Short circuit the truncate case for zero length files */
529 if ((vap->va_size == 0) &&
530 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
531 xfs_iunlock(ip, XFS_ILOCK_EXCL);
532 lock_flags &= ~XFS_ILOCK_EXCL;
533 if (mask & XFS_AT_CTIME)
534 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
539 if (vp->v_type == VDIR) {
540 code = XFS_ERROR(EISDIR);
542 } else if (vp->v_type != VREG) {
543 code = XFS_ERROR(EINVAL);
547 * Make sure that the dquots are attached to the inode.
549 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
554 * Change file access or modified times.
556 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
558 if ((flags & ATTR_UTIME) &&
559 !capable(CAP_FOWNER)) {
560 code = XFS_ERROR(EPERM);
567 * Change extent size or realtime flag.
569 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
571 * Can't change extent size if any extents are allocated.
573 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
574 (mask & XFS_AT_EXTSIZE) &&
575 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
577 code = XFS_ERROR(EINVAL); /* EFBIG? */
582 * Can't set extent size unless the file is marked, or
583 * about to be marked as a realtime file.
585 * This check will be removed when fixed size extents
586 * with buffered data writes is implemented.
589 if ((mask & XFS_AT_EXTSIZE) &&
590 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
592 (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
593 ((mask & XFS_AT_XFLAGS) &&
594 (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
595 code = XFS_ERROR(EINVAL);
600 * Can't change realtime flag if any extents are allocated.
602 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
603 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
604 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
605 code = XFS_ERROR(EINVAL); /* EFBIG? */
609 * Extent size must be a multiple of the appropriate block
610 * size, if set at all.
612 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
615 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
616 ((mask & XFS_AT_XFLAGS) &&
617 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
618 size = mp->m_sb.sb_rextsize <<
619 mp->m_sb.sb_blocklog;
621 size = mp->m_sb.sb_blocksize;
623 if (vap->va_extsize % size) {
624 code = XFS_ERROR(EINVAL);
629 * If realtime flag is set then must have realtime data.
631 if ((mask & XFS_AT_XFLAGS) &&
632 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
633 if ((mp->m_sb.sb_rblocks == 0) ||
634 (mp->m_sb.sb_rextsize == 0) ||
635 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
636 code = XFS_ERROR(EINVAL);
642 * Can't modify an immutable/append-only file unless
643 * we have appropriate permission.
645 if ((mask & XFS_AT_XFLAGS) &&
647 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
649 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
650 !capable(CAP_LINUX_IMMUTABLE)) {
651 code = XFS_ERROR(EPERM);
657 * Now we can make the changes. Before we join the inode
658 * to the transaction, if XFS_AT_SIZE is set then take care of
659 * the part of the truncation that must be done without the
660 * inode lock. This needs to be done before joining the inode
661 * to the transaction, because the inode cannot be unlocked
662 * once it is a part of the transaction.
664 if (mask & XFS_AT_SIZE) {
666 if (vap->va_size > ip->i_d.di_size)
667 code = xfs_igrow_start(ip, vap->va_size, credp);
668 xfs_iunlock(ip, XFS_ILOCK_EXCL);
670 code = xfs_itruncate_data(ip, vap->va_size);
673 lock_flags &= ~XFS_ILOCK_EXCL;
674 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
677 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
678 if ((code = xfs_trans_reserve(tp, 0,
679 XFS_ITRUNCATE_LOG_RES(mp), 0,
680 XFS_TRANS_PERM_LOG_RES,
681 XFS_ITRUNCATE_LOG_COUNT))) {
682 xfs_trans_cancel(tp, 0);
684 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
687 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
688 xfs_ilock(ip, XFS_ILOCK_EXCL);
692 xfs_trans_ijoin(tp, ip, lock_flags);
693 xfs_trans_ihold(tp, ip);
696 /* determine whether mandatory locking mode changes */
697 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
700 * Truncate file. Must have write permission and not be a directory.
702 if (mask & XFS_AT_SIZE) {
703 if (vap->va_size > ip->i_d.di_size) {
704 xfs_igrow_finish(tp, ip, vap->va_size,
705 !(flags & ATTR_DMI));
706 } else if ((vap->va_size <= ip->i_d.di_size) ||
707 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
709 * signal a sync transaction unless
710 * we're truncating an already unlinked
711 * file on a wsync filesystem
713 code = xfs_itruncate_finish(&tp, ip,
714 (xfs_fsize_t)vap->va_size,
716 ((ip->i_d.di_nlink != 0 ||
717 !(mp->m_flags & XFS_MOUNT_WSYNC))
724 * Have to do this even if the file's size doesn't change.
726 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
730 * Change file access modes.
732 if (mask & XFS_AT_MODE) {
733 ip->i_d.di_mode &= S_IFMT;
734 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
736 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
737 timeflags |= XFS_ICHGTIME_CHG;
741 * Change file ownership. Must be the owner or privileged.
742 * If the system was configured with the "restricted_chown"
743 * option, the owner is not permitted to give away the file,
744 * and can change the group id only to a group of which he
745 * or she is a member.
747 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
749 * CAP_FSETID overrides the following restrictions:
751 * The set-user-ID and set-group-ID bits of a file will be
752 * cleared upon successful return from chown()
754 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
755 !capable(CAP_FSETID)) {
756 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
760 * Change the ownerships and register quota modifications
761 * in the transaction.
764 if (XFS_IS_UQUOTA_ON(mp)) {
765 ASSERT(mask & XFS_AT_UID);
767 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
768 &ip->i_udquot, udqp);
770 ip->i_d.di_uid = uid;
773 if (XFS_IS_GQUOTA_ON(mp)) {
774 ASSERT(mask & XFS_AT_GID);
776 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
777 &ip->i_gdquot, gdqp);
779 ip->i_d.di_gid = gid;
781 if (iprojid != projid) {
782 ip->i_d.di_projid = projid;
784 * We may have to rev the inode as well as
785 * the superblock version number since projids didn't
786 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
788 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
789 xfs_bump_ino_vers2(tp, ip);
792 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
793 timeflags |= XFS_ICHGTIME_CHG;
798 * Change file access or modified times.
800 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
801 if (mask & XFS_AT_ATIME) {
802 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
803 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
804 ip->i_update_core = 1;
805 timeflags &= ~XFS_ICHGTIME_ACC;
807 if (mask & XFS_AT_MTIME) {
808 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
809 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
810 timeflags &= ~XFS_ICHGTIME_MOD;
811 timeflags |= XFS_ICHGTIME_CHG;
813 if (tp && (flags & ATTR_UTIME))
814 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
818 * Change XFS-added attributes.
820 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
821 if (mask & XFS_AT_EXTSIZE) {
823 * Converting bytes to fs blocks.
825 ip->i_d.di_extsize = vap->va_extsize >>
826 mp->m_sb.sb_blocklog;
828 if (mask & XFS_AT_XFLAGS) {
831 /* can't set PREALLOC this way, just preserve it */
832 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
833 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
834 di_flags |= XFS_DIFLAG_IMMUTABLE;
835 if (vap->va_xflags & XFS_XFLAG_IUNLINK)
836 di_flags |= XFS_DIFLAG_IUNLINK;
837 if (vap->va_xflags & XFS_XFLAG_BARRIER)
838 di_flags |= XFS_DIFLAG_BARRIER;
839 if (vap->va_xflags & XFS_XFLAG_APPEND)
840 di_flags |= XFS_DIFLAG_APPEND;
841 if (vap->va_xflags & XFS_XFLAG_SYNC)
842 di_flags |= XFS_DIFLAG_SYNC;
843 if (vap->va_xflags & XFS_XFLAG_NOATIME)
844 di_flags |= XFS_DIFLAG_NOATIME;
845 if (vap->va_xflags & XFS_XFLAG_NODUMP)
846 di_flags |= XFS_DIFLAG_NODUMP;
847 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
848 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
849 di_flags |= XFS_DIFLAG_RTINHERIT;
850 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
851 di_flags |= XFS_DIFLAG_NOSYMLINKS;
853 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
854 di_flags |= XFS_DIFLAG_REALTIME;
855 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
857 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
860 ip->i_d.di_flags = di_flags;
862 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
863 timeflags |= XFS_ICHGTIME_CHG;
867 * Change file inode change time only if XFS_AT_CTIME set
868 * AND we have been called by a DMI function.
871 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
872 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
873 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
874 ip->i_update_core = 1;
875 timeflags &= ~XFS_ICHGTIME_CHG;
879 * Send out timestamp changes that need to be set to the
880 * current time. Not done when called by a DMI function.
882 if (timeflags && !(flags & ATTR_DMI))
883 xfs_ichgtime(ip, timeflags);
885 XFS_STATS_INC(xs_ig_attrchg);
888 * If this is a synchronous mount, make sure that the
889 * transaction goes to disk before returning to the user.
890 * This is slightly sub-optimal in that truncates require
891 * two sync transactions instead of one for wsync filesytems.
892 * One for the truncate and one for the timestamps since we
893 * don't want to change the timestamps unless we're sure the
894 * truncate worked. Truncates are less than 1% of the laddis
895 * mix so this probably isn't worth the trouble to optimize.
899 if (mp->m_flags & XFS_MOUNT_WSYNC)
900 xfs_trans_set_sync(tp);
902 code = xfs_trans_commit(tp, commit_flags, NULL);
906 * If the (regular) file's mandatory locking mode changed, then
907 * notify the vnode. We do this under the inode lock to prevent
908 * racing calls to vop_vnode_change.
910 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
911 if (mandlock_before != mandlock_after) {
912 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
916 xfs_iunlock(ip, lock_flags);
919 * Release any dquot(s) the inode had kept before chown.
921 XFS_QM_DQRELE(mp, olddquot1);
922 XFS_QM_DQRELE(mp, olddquot2);
923 XFS_QM_DQRELE(mp, udqp);
924 XFS_QM_DQRELE(mp, gdqp);
930 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
931 !(flags & ATTR_DMI)) {
932 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
933 NULL, DM_RIGHT_NULL, NULL, NULL,
934 0, 0, AT_DELAY_FLAG(flags));
939 commit_flags |= XFS_TRANS_ABORT;
942 XFS_QM_DQRELE(mp, udqp);
943 XFS_QM_DQRELE(mp, gdqp);
945 xfs_trans_cancel(tp, commit_flags);
947 if (lock_flags != 0) {
948 xfs_iunlock(ip, lock_flags);
956 * Null conversion from vnode mode bits to inode mode bits, as in efs.
967 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
968 (inst_t *)__return_address);
970 ip = XFS_BHVTOI(bdp);
971 xfs_ilock(ip, XFS_ILOCK_SHARED);
972 error = xfs_iaccess(ip, mode, credp);
973 xfs_iunlock(ip, XFS_ILOCK_SHARED);
997 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1003 vp = BHV_TO_VNODE(bdp);
1004 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1006 ip = XFS_BHVTOI(bdp);
1009 if (XFS_FORCED_SHUTDOWN(mp))
1010 return XFS_ERROR(EIO);
1012 xfs_ilock(ip, XFS_ILOCK_SHARED);
1014 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1016 offset = uiop->uio_offset;
1017 count = uiop->uio_resid;
1020 error = XFS_ERROR(EINVAL);
1028 if (!(ioflags & IO_INVIS)) {
1029 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1033 * See if the symlink is stored inline.
1035 pathlen = (int)ip->i_d.di_size;
1037 if (ip->i_df.if_flags & XFS_IFINLINE) {
1038 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1042 * Symlink not inline. Call bmap to get it in.
1044 nmaps = SYMLINK_MAPS;
1046 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1047 0, NULL, 0, mval, &nmaps, NULL);
1053 for (n = 0; n < nmaps; n++) {
1054 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1055 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1056 bp = xfs_buf_read(mp->m_ddev_targp, d,
1057 BTOBB(byte_cnt), 0);
1058 error = XFS_BUF_GETERROR(bp);
1060 xfs_ioerror_alert("xfs_readlink",
1061 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1065 if (pathlen < byte_cnt)
1067 pathlen -= byte_cnt;
1069 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1078 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1087 * This is called to sync the inode and its data out to disk.
1088 * We need to hold the I/O lock while flushing the data, and
1089 * the inode lock while flushing the inode. The inode lock CANNOT
1090 * be held while flushing the data, so acquire after we're done
1105 vn_trace_entry(BHV_TO_VNODE(bdp),
1106 __FUNCTION__, (inst_t *)__return_address);
1108 ip = XFS_BHVTOI(bdp);
1110 ASSERT(start >= 0 && stop >= -1);
1112 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1113 return XFS_ERROR(EIO);
1116 * We always need to make sure that the required inode state
1117 * is safe on disk. The vnode might be clean but because
1118 * of committed transactions that haven't hit the disk yet.
1119 * Likewise, there could be unflushed non-transactional
1120 * changes to the inode core that have to go to disk.
1122 * The following code depends on one assumption: that
1123 * any transaction that changes an inode logs the core
1124 * because it has to change some field in the inode core
1125 * (typically nextents or nblocks). That assumption
1126 * implies that any transactions against an inode will
1127 * catch any non-transactional updates. If inode-altering
1128 * transactions exist that violate this assumption, the
1129 * code breaks. Right now, it figures that if the involved
1130 * update_* field is clear and the inode is unpinned, the
1131 * inode is clean. Either it's been flushed or it's been
1132 * committed and the commit has hit the disk unpinning the inode.
1133 * (Note that xfs_inode_item_format() called at commit clears
1134 * the update_* fields.)
1136 xfs_ilock(ip, XFS_ILOCK_SHARED);
1138 /* If we are flushing data then we care about update_size
1139 * being set, otherwise we care about update_core
1141 if ((flag & FSYNC_DATA) ?
1142 (ip->i_update_size == 0) :
1143 (ip->i_update_core == 0)) {
1145 * Timestamps/size haven't changed since last inode
1146 * flush or inode transaction commit. That means
1147 * either nothing got written or a transaction
1148 * committed which caught the updates. If the
1149 * latter happened and the transaction hasn't
1150 * hit the disk yet, the inode will be still
1151 * be pinned. If it is, force the log.
1154 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1156 if (xfs_ipincount(ip)) {
1157 xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1159 ((flag & FSYNC_WAIT)
1160 ? XFS_LOG_SYNC : 0));
1165 * Kick off a transaction to log the inode
1166 * core to get the updates. Make it
1167 * sync if FSYNC_WAIT is passed in (which
1168 * is done by everybody but specfs). The
1169 * sync transaction will also force the log.
1171 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1172 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1173 if ((error = xfs_trans_reserve(tp, 0,
1174 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1176 xfs_trans_cancel(tp, 0);
1179 xfs_ilock(ip, XFS_ILOCK_EXCL);
1182 * Note - it's possible that we might have pushed
1183 * ourselves out of the way during trans_reserve
1184 * which would flush the inode. But there's no
1185 * guarantee that the inode buffer has actually
1186 * gone out yet (it's delwri). Plus the buffer
1187 * could be pinned anyway if it's part of an
1188 * inode in another recent transaction. So we
1189 * play it safe and fire off the transaction anyway.
1191 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1192 xfs_trans_ihold(tp, ip);
1193 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1194 if (flag & FSYNC_WAIT)
1195 xfs_trans_set_sync(tp);
1196 error = xfs_trans_commit(tp, 0, NULL);
1198 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1204 * This is called by xfs_inactive to free any blocks beyond eof,
1205 * when the link count isn't zero.
1208 xfs_inactive_free_eofblocks(
1214 xfs_fileoff_t end_fsb;
1215 xfs_fileoff_t last_fsb;
1216 xfs_filblks_t map_len;
1218 xfs_bmbt_irec_t imap;
1221 * Figure out if there are any blocks beyond the end
1222 * of the file. If not, then there is nothing to do.
1224 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1225 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1226 map_len = last_fsb - end_fsb;
1231 xfs_ilock(ip, XFS_ILOCK_SHARED);
1232 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1233 NULL, 0, &imap, &nimaps, NULL);
1234 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1236 if (!error && (nimaps != 0) &&
1237 (imap.br_startblock != HOLESTARTBLOCK)) {
1239 * Attach the dquots to the inode up front.
1241 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1245 * There are blocks after the end of file.
1246 * Free them up now by truncating the file to
1249 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1252 * Do the xfs_itruncate_start() call before
1253 * reserving any log space because
1254 * itruncate_start will call into the buffer
1255 * cache and we can't
1256 * do that within a transaction.
1258 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1259 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1262 error = xfs_trans_reserve(tp, 0,
1263 XFS_ITRUNCATE_LOG_RES(mp),
1264 0, XFS_TRANS_PERM_LOG_RES,
1265 XFS_ITRUNCATE_LOG_COUNT);
1267 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1268 xfs_trans_cancel(tp, 0);
1269 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1273 xfs_ilock(ip, XFS_ILOCK_EXCL);
1274 xfs_trans_ijoin(tp, ip,
1277 xfs_trans_ihold(tp, ip);
1279 error = xfs_itruncate_finish(&tp, ip,
1284 * If we get an error at this point we
1285 * simply don't bother truncating the file.
1288 xfs_trans_cancel(tp,
1289 (XFS_TRANS_RELEASE_LOG_RES |
1292 error = xfs_trans_commit(tp,
1293 XFS_TRANS_RELEASE_LOG_RES,
1296 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1302 * Free a symlink that has blocks associated with it.
1305 xfs_inactive_symlink_rmt(
1313 xfs_fsblock_t first_block;
1314 xfs_bmap_free_t free_list;
1317 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1325 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1327 * We're freeing a symlink that has some
1328 * blocks allocated to it. Free the
1329 * blocks here. We know that we've got
1330 * either 1 or 2 extents and that we can
1331 * free them all in one bunmapi call.
1333 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1334 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1335 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1336 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1337 xfs_trans_cancel(tp, 0);
1342 * Lock the inode, fix the size, and join it to the transaction.
1343 * Hold it so in the normal path, we still have it locked for
1344 * the second transaction. In the error paths we need it
1345 * held so the cancel won't rele it, see below.
1347 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1348 size = (int)ip->i_d.di_size;
1349 ip->i_d.di_size = 0;
1350 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1351 xfs_trans_ihold(tp, ip);
1352 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1354 * Find the block(s) so we can inval and unmap them.
1357 XFS_BMAP_INIT(&free_list, &first_block);
1358 nmaps = sizeof(mval) / sizeof(mval[0]);
1359 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1360 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1364 * Invalidate the block(s).
1366 for (i = 0; i < nmaps; i++) {
1367 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1368 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1369 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1370 xfs_trans_binval(tp, bp);
1373 * Unmap the dead block(s) to the free_list.
1375 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1376 &first_block, &free_list, &done)))
1380 * Commit the first transaction. This logs the EFI and the inode.
1382 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1385 * The transaction must have been committed, since there were
1386 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1387 * The new tp has the extent freeing and EFDs.
1391 * The first xact was committed, so add the inode to the new one.
1392 * Mark it dirty so it will be logged and moved forward in the log as
1393 * part of every commit.
1395 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1396 xfs_trans_ihold(tp, ip);
1397 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1399 * Get a new, empty transaction to return to our caller.
1401 ntp = xfs_trans_dup(tp);
1403 * Commit the transaction containing extent freeing and EFD's.
1404 * If we get an error on the commit here or on the reserve below,
1405 * we need to unlock the inode since the new transaction doesn't
1406 * have the inode attached.
1408 error = xfs_trans_commit(tp, 0, NULL);
1411 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1415 * Remove the memory for extent descriptions (just bookkeeping).
1417 if (ip->i_df.if_bytes)
1418 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1419 ASSERT(ip->i_df.if_bytes == 0);
1421 * Put an itruncate log reservation in the new transaction
1424 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1425 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1426 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1430 * Return with the inode locked but not joined to the transaction.
1436 xfs_bmap_cancel(&free_list);
1439 * Have to come here with the inode locked and either
1440 * (held and in the transaction) or (not in the transaction).
1441 * If the inode isn't held then cancel would iput it, but
1442 * that's wrong since this is inactive and the vnode ref
1443 * count is 0 already.
1444 * Cancel won't do anything to the inode if held, but it still
1445 * needs to be locked until the cancel is done, if it was
1446 * joined to the transaction.
1448 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1449 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1456 xfs_inactive_symlink_local(
1462 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1464 * We're freeing a symlink which fit into
1465 * the inode. Just free the memory used
1466 * to hold the old symlink.
1468 error = xfs_trans_reserve(*tpp, 0,
1469 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1470 0, XFS_TRANS_PERM_LOG_RES,
1471 XFS_ITRUNCATE_LOG_COUNT);
1474 xfs_trans_cancel(*tpp, 0);
1478 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1481 * Zero length symlinks _can_ exist.
1483 if (ip->i_df.if_bytes > 0) {
1484 xfs_idata_realloc(ip,
1485 -(ip->i_df.if_bytes),
1487 ASSERT(ip->i_df.if_bytes == 0);
1504 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1507 ASSERT(ip->i_d.di_forkoff != 0);
1508 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1509 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1511 error = xfs_attr_inactive(ip);
1514 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1515 return (error); /* goto out*/
1518 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1519 error = xfs_trans_reserve(tp, 0,
1520 XFS_IFREE_LOG_RES(mp),
1521 0, XFS_TRANS_PERM_LOG_RES,
1522 XFS_INACTIVE_LOG_COUNT);
1524 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1525 xfs_trans_cancel(tp, 0);
1527 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1531 xfs_ilock(ip, XFS_ILOCK_EXCL);
1532 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1533 xfs_trans_ihold(tp, ip);
1534 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1536 ASSERT(ip->i_d.di_anextents == 0);
1551 vp = BHV_TO_VNODE(bdp);
1552 ip = XFS_BHVTOI(bdp);
1554 if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1558 /* If this is a read-only mount, don't do this (would generate I/O) */
1559 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1562 #ifdef HAVE_REFCACHE
1563 /* If we are in the NFS reference cache then don't do this now */
1570 if (ip->i_d.di_nlink != 0) {
1571 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1572 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1573 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1574 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1575 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1577 /* Update linux inode block count after free above */
1578 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1579 ip->i_d.di_nblocks + ip->i_delayed_blks);
1589 * This is called when the vnode reference count for the vnode
1590 * goes to zero. If the file has been unlinked, then it must
1591 * now be truncated. Also, we clear all of the read-ahead state
1592 * kept for the inode here since the file is now closed.
1601 xfs_bmap_free_t free_list;
1602 xfs_fsblock_t first_block;
1609 vp = BHV_TO_VNODE(bdp);
1610 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1612 ip = XFS_BHVTOI(bdp);
1615 * If the inode is already free, then there can be nothing
1618 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1619 ASSERT(ip->i_df.if_real_bytes == 0);
1620 ASSERT(ip->i_df.if_broot_bytes == 0);
1621 return VN_INACTIVE_CACHE;
1625 * Only do a truncate if it's a regular file with
1626 * some actual space in it. It's OK to look at the
1627 * inode's fields without the lock because we're the
1628 * only one with a reference to the inode.
1630 truncate = ((ip->i_d.di_nlink == 0) &&
1631 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1632 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1636 if (ip->i_d.di_nlink == 0 &&
1637 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1638 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1643 /* If this is a read-only mount, don't do this (would generate I/O) */
1644 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1647 if (ip->i_d.di_nlink != 0) {
1648 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1649 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1650 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1651 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1652 (ip->i_delayed_blks != 0))) {
1653 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1654 return (VN_INACTIVE_CACHE);
1655 /* Update linux inode block count after free above */
1656 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1657 ip->i_d.di_nblocks + ip->i_delayed_blks);
1662 ASSERT(ip->i_d.di_nlink == 0);
1664 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1665 return (VN_INACTIVE_CACHE);
1667 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1670 * Do the xfs_itruncate_start() call before
1671 * reserving any log space because itruncate_start
1672 * will call into the buffer cache and we can't
1673 * do that within a transaction.
1675 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1677 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1679 error = xfs_trans_reserve(tp, 0,
1680 XFS_ITRUNCATE_LOG_RES(mp),
1681 0, XFS_TRANS_PERM_LOG_RES,
1682 XFS_ITRUNCATE_LOG_COUNT);
1684 /* Don't call itruncate_cleanup */
1685 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1686 xfs_trans_cancel(tp, 0);
1687 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1688 return (VN_INACTIVE_CACHE);
1691 xfs_ilock(ip, XFS_ILOCK_EXCL);
1692 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1693 xfs_trans_ihold(tp, ip);
1696 * normally, we have to run xfs_itruncate_finish sync.
1697 * But if filesystem is wsync and we're in the inactive
1698 * path, then we know that nlink == 0, and that the
1699 * xaction that made nlink == 0 is permanently committed
1700 * since xfs_remove runs as a synchronous transaction.
1702 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1703 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1706 xfs_trans_cancel(tp,
1707 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1708 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1709 return (VN_INACTIVE_CACHE);
1711 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1714 * If we get an error while cleaning up a
1715 * symlink we bail out.
1717 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1718 xfs_inactive_symlink_rmt(ip, &tp) :
1719 xfs_inactive_symlink_local(ip, &tp);
1723 return (VN_INACTIVE_CACHE);
1726 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1727 xfs_trans_ihold(tp, ip);
1729 error = xfs_trans_reserve(tp, 0,
1730 XFS_IFREE_LOG_RES(mp),
1731 0, XFS_TRANS_PERM_LOG_RES,
1732 XFS_INACTIVE_LOG_COUNT);
1734 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1735 xfs_trans_cancel(tp, 0);
1736 return (VN_INACTIVE_CACHE);
1739 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1740 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1741 xfs_trans_ihold(tp, ip);
1745 * If there are attributes associated with the file
1746 * then blow them away now. The code calls a routine
1747 * that recursively deconstructs the attribute fork.
1748 * We need to just commit the current transaction
1749 * because we can't use it for xfs_attr_inactive().
1751 if (ip->i_d.di_anextents > 0) {
1752 error = xfs_inactive_attrs(ip, &tp);
1754 * If we got an error, the transaction is already
1755 * cancelled, and the inode is unlocked. Just get out.
1758 return (VN_INACTIVE_CACHE);
1759 } else if (ip->i_afp) {
1760 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1766 XFS_BMAP_INIT(&free_list, &first_block);
1767 error = xfs_ifree(tp, ip, &free_list);
1770 * If we fail to free the inode, shut down. The cancel
1771 * might do that, we need to make sure. Otherwise the
1772 * inode might be lost for a long time or forever.
1774 if (!XFS_FORCED_SHUTDOWN(mp)) {
1776 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1777 error, mp->m_fsname);
1778 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1780 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1783 * Credit the quota account(s). The inode is gone.
1785 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1788 * Just ignore errors at this point. There is
1789 * nothing we can do except to try to keep going.
1791 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1793 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1796 * Release the dquots held by inode, if any.
1798 XFS_QM_DQDETACH(mp, ip);
1800 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1803 return VN_INACTIVE_CACHE;
1812 bhv_desc_t *dir_bdp,
1819 xfs_inode_t *dp, *ip;
1825 dir_vp = BHV_TO_VNODE(dir_bdp);
1826 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1828 dp = XFS_BHVTOI(dir_bdp);
1830 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1831 return XFS_ERROR(EIO);
1833 lock_mode = xfs_ilock_map_shared(dp);
1834 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1836 *vpp = XFS_ITOV(ip);
1839 xfs_iunlock_map_shared(dp, lock_mode);
1845 * xfs_create (create a new file).
1849 bhv_desc_t *dir_bdp,
1855 char *name = VNAME(dentry);
1857 xfs_inode_t *dp, *ip;
1863 xfs_bmap_free_t free_list;
1864 xfs_fsblock_t first_block;
1865 boolean_t dp_joined_to_trans;
1866 int dm_event_sent = 0;
1870 struct xfs_dquot *udqp, *gdqp;
1876 dir_vp = BHV_TO_VNODE(dir_bdp);
1877 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1879 dp = XFS_BHVTOI(dir_bdp);
1882 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1883 namelen = VNAMELEN(dentry);
1885 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1886 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1887 dir_vp, DM_RIGHT_NULL, NULL,
1888 DM_RIGHT_NULL, name, NULL,
1896 if (XFS_FORCED_SHUTDOWN(mp))
1897 return XFS_ERROR(EIO);
1899 /* Return through std_return after this point. */
1902 if (vap->va_mask & XFS_AT_PROJID)
1903 prid = (xfs_prid_t)vap->va_projid;
1905 prid = (xfs_prid_t)dfltprid;
1908 * Make sure that we have allocated dquot(s) on disk.
1910 error = XFS_QM_DQVOPALLOC(mp, dp,
1911 current_fsuid(credp), current_fsgid(credp),
1912 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1917 dp_joined_to_trans = B_FALSE;
1919 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1920 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1921 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1923 * Initially assume that the file does not exist and
1924 * reserve the resources for that case. If that is not
1925 * the case we'll drop the one we have and get a more
1926 * appropriate transaction later.
1928 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1929 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1930 if (error == ENOSPC) {
1932 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1933 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1941 xfs_ilock(dp, XFS_ILOCK_EXCL);
1943 XFS_BMAP_INIT(&free_list, &first_block);
1948 * Reserve disk quota and the inode.
1950 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1955 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1957 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1958 error = xfs_dir_ialloc(&tp, dp,
1959 MAKEIMODE(vap->va_type,vap->va_mode), 1,
1960 rdev, credp, prid, resblks > 0,
1963 if (error == ENOSPC)
1970 * At this point, we've gotten a newly allocated inode.
1971 * It is locked (and joined to the transaction).
1974 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1977 * Now we join the directory inode to the transaction.
1978 * We do not do it earlier because xfs_dir_ialloc
1979 * might commit the previous transaction (and release
1984 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1985 dp_joined_to_trans = B_TRUE;
1987 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1988 &first_block, &free_list,
1989 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1991 ASSERT(error != ENOSPC);
1994 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1995 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1998 * If this is a synchronous mount, make sure that the
1999 * create transaction goes to disk before returning to
2002 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2003 xfs_trans_set_sync(tp);
2009 * Attach the dquot(s) to the inodes and modify them incore.
2010 * These ids of the inode couldn't have changed since the new
2011 * inode has been locked ever since it was created.
2013 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2016 * xfs_trans_commit normally decrements the vnode ref count
2017 * when it unlocks the inode. Since we want to return the
2018 * vnode to the caller, we bump the vnode ref count now.
2023 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2025 xfs_bmap_cancel(&free_list);
2029 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2036 XFS_QM_DQRELE(mp, udqp);
2037 XFS_QM_DQRELE(mp, gdqp);
2040 * Propogate the fact that the vnode changed after the
2041 * xfs_inode locks have been released.
2043 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2047 /* Fallthrough to std_return with error = 0 */
2050 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2051 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2052 DM_EVENT_POSTCREATE)) {
2053 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2054 dir_vp, DM_RIGHT_NULL,
2056 DM_RIGHT_NULL, name, NULL,
2057 dm_di_mode, error, 0);
2062 cancel_flags |= XFS_TRANS_ABORT;
2067 xfs_trans_cancel(tp, cancel_flags);
2069 if (!dp_joined_to_trans && (dp != NULL))
2070 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2071 XFS_QM_DQRELE(mp, udqp);
2072 XFS_QM_DQRELE(mp, gdqp);
2078 * Wait until after the current transaction is aborted to
2079 * release the inode. This prevents recursive transactions
2080 * and deadlocks from xfs_inactive.
2082 cancel_flags |= XFS_TRANS_ABORT;
2083 xfs_trans_cancel(tp, cancel_flags);
2086 XFS_QM_DQRELE(mp, udqp);
2087 XFS_QM_DQRELE(mp, gdqp);
2094 * Some counters to see if (and how often) we are hitting some deadlock
2095 * prevention code paths.
2099 int xfs_rm_lock_delays;
2100 int xfs_rm_attempts;
2104 * The following routine will lock the inodes associated with the
2105 * directory and the named entry in the directory. The locks are
2106 * acquired in increasing inode number.
2108 * If the entry is "..", then only the directory is locked. The
2109 * vnode ref count will still include that from the .. entry in
2112 * There is a deadlock we need to worry about. If the locked directory is
2113 * in the AIL, it might be blocking up the log. The next inode we lock
2114 * could be already locked by another thread waiting for log space (e.g
2115 * a permanent log reservation with a long running transaction (see
2116 * xfs_itruncate_finish)). To solve this, we must check if the directory
2117 * is in the ail and use lock_nowait. If we can't lock, we need to
2118 * drop the inode lock on the directory and try again. xfs_iunlock will
2119 * potentially push the tail if we were holding up the log.
2122 xfs_lock_dir_and_entry(
2125 xfs_inode_t *ip) /* inode of entry 'name' */
2129 xfs_inode_t *ips[2];
2138 xfs_ilock(dp, XFS_ILOCK_EXCL);
2145 * We want to lock in increasing inum. Since we've already
2146 * acquired the lock on the directory, we may need to release
2147 * if if the inum of the entry turns out to be less.
2149 if (e_inum > dp->i_ino) {
2151 * We are already in the right order, so just
2152 * lock on the inode of the entry.
2153 * We need to use nowait if dp is in the AIL.
2156 lp = (xfs_log_item_t *)dp->i_itemp;
2157 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2158 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2165 * Unlock dp and try again.
2166 * xfs_iunlock will try to push the tail
2167 * if the inode is in the AIL.
2170 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2172 if ((attempts % 5) == 0) {
2173 delay(1); /* Don't just spin the CPU */
2175 xfs_rm_lock_delays++;
2181 xfs_ilock(ip, XFS_ILOCK_EXCL);
2183 } else if (e_inum < dp->i_ino) {
2184 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2188 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2190 /* else e_inum == dp->i_ino */
2191 /* This can happen if we're asked to lock /x/..
2192 * the entry is "..", which is also the parent directory.
2200 int xfs_small_retries;
2201 int xfs_middle_retries;
2202 int xfs_lots_retries;
2203 int xfs_lock_delays;
2207 * The following routine will lock n inodes in exclusive mode.
2208 * We assume the caller calls us with the inodes in i_ino order.
2210 * We need to detect deadlock where an inode that we lock
2211 * is in the AIL and we start waiting for another inode that is locked
2212 * by a thread in a long running transaction (such as truncate). This can
2213 * result in deadlock since the long running trans might need to wait
2214 * for the inode we just locked in order to push the tail and free space
2224 int attempts = 0, i, j, try_lock;
2227 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2238 for (; i < inodes; i++) {
2241 if (i && (ips[i] == ips[i-1])) /* Already locked */
2245 * If try_lock is not set yet, make sure all locked inodes
2246 * are not in the AIL.
2247 * If any are, set try_lock to be used later.
2251 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2252 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2253 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2260 * If any of the previous locks we have locked is in the AIL,
2261 * we must TRY to get the second and subsequent locks. If
2262 * we can't get any, we must release all we have
2267 /* try_lock must be 0 if i is 0. */
2269 * try_lock means we have an inode locked
2270 * that is in the AIL.
2273 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2277 * Unlock all previous guys and try again.
2278 * xfs_iunlock will try to push the tail
2279 * if the inode is in the AIL.
2282 for(j = i - 1; j >= 0; j--) {
2285 * Check to see if we've already
2286 * unlocked this one.
2287 * Not the first one going back,
2288 * and the inode ptr is the same.
2290 if ((j != (i - 1)) && ips[j] ==
2294 xfs_iunlock(ips[j], lock_mode);
2297 if ((attempts % 5) == 0) {
2298 delay(1); /* Don't just spin the CPU */
2308 xfs_ilock(ips[i], lock_mode);
2314 if (attempts < 5) xfs_small_retries++;
2315 else if (attempts < 100) xfs_middle_retries++;
2316 else xfs_lots_retries++;
2324 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2325 int remove_which_error_return = 0;
2327 #define REMOVE_DEBUG_TRACE(x)
2328 #endif /* ! DEBUG */
2337 bhv_desc_t *dir_bdp,
2342 char *name = VNAME(dentry);
2343 xfs_inode_t *dp, *ip;
2344 xfs_trans_t *tp = NULL;
2347 xfs_bmap_free_t free_list;
2348 xfs_fsblock_t first_block;
2356 dir_vp = BHV_TO_VNODE(dir_bdp);
2357 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2359 dp = XFS_BHVTOI(dir_bdp);
2362 if (XFS_FORCED_SHUTDOWN(mp))
2363 return XFS_ERROR(EIO);
2365 namelen = VNAMELEN(dentry);
2367 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2368 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2369 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2370 name, NULL, 0, 0, 0);
2375 /* From this point on, return through std_return */
2379 * We need to get a reference to ip before we get our log
2380 * reservation. The reason for this is that we cannot call
2381 * xfs_iget for an inode for which we do not have a reference
2382 * once we've acquired a log reservation. This is because the
2383 * inode we are trying to get might be in xfs_inactive going
2384 * for a log reservation. Since we'll have to wait for the
2385 * inactive code to complete before returning from xfs_iget,
2386 * we need to make sure that we don't have log space reserved
2387 * when we call xfs_iget. Instead we get an unlocked referece
2388 * to the inode before getting our log reservation.
2390 error = xfs_get_dir_entry(dentry, &ip);
2392 REMOVE_DEBUG_TRACE(__LINE__);
2396 dm_di_mode = ip->i_d.di_mode;
2398 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2402 error = XFS_QM_DQATTACH(mp, dp, 0);
2403 if (!error && dp != ip)
2404 error = XFS_QM_DQATTACH(mp, ip, 0);
2406 REMOVE_DEBUG_TRACE(__LINE__);
2411 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2412 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2414 * We try to get the real space reservation first,
2415 * allowing for directory btree deletion(s) implying
2416 * possible bmap insert(s). If we can't get the space
2417 * reservation then we use 0 instead, and avoid the bmap
2418 * btree insert(s) in the directory code by, if the bmap
2419 * insert tries to happen, instead trimming the LAST
2420 * block from the directory.
2422 resblks = XFS_REMOVE_SPACE_RES(mp);
2423 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2424 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2425 if (error == ENOSPC) {
2427 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2428 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2431 ASSERT(error != ENOSPC);
2432 REMOVE_DEBUG_TRACE(__LINE__);
2433 xfs_trans_cancel(tp, 0);
2438 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2440 REMOVE_DEBUG_TRACE(__LINE__);
2441 xfs_trans_cancel(tp, cancel_flags);
2447 * At this point, we've gotten both the directory and the entry
2450 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2453 * Increment vnode ref count only in this case since
2454 * there's an extra vnode reference in the case where
2458 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2462 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2464 XFS_BMAP_INIT(&free_list, &first_block);
2465 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2466 &first_block, &free_list, 0);
2468 ASSERT(error != ENOENT);
2469 REMOVE_DEBUG_TRACE(__LINE__);
2472 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2475 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2477 error = xfs_droplink(tp, ip);
2479 REMOVE_DEBUG_TRACE(__LINE__);
2483 /* Determine if this is the last link while
2484 * we are in the transaction.
2486 link_zero = (ip)->i_d.di_nlink==0;
2489 * Take an extra ref on the inode so that it doesn't
2490 * go to xfs_inactive() from within the commit.
2495 * If this is a synchronous mount, make sure that the
2496 * remove transaction goes to disk before returning to
2499 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2500 xfs_trans_set_sync(tp);
2503 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2505 REMOVE_DEBUG_TRACE(__LINE__);
2509 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2516 * Before we drop our extra reference to the inode, purge it
2517 * from the refcache if it is there. By waiting until afterwards
2518 * to do the IRELE, we ensure that we won't go inactive in the
2519 * xfs_refcache_purge_ip routine (although that would be OK).
2521 xfs_refcache_purge_ip(ip);
2523 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2526 * Let interposed file systems know about removed links.
2528 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2532 /* Fall through to std_return with error = 0 */
2534 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2535 DM_EVENT_POSTREMOVE)) {
2536 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2537 dir_vp, DM_RIGHT_NULL,
2538 NULL, DM_RIGHT_NULL,
2539 name, NULL, dm_di_mode, error, 0);
2544 xfs_bmap_cancel(&free_list);
2545 cancel_flags |= XFS_TRANS_ABORT;
2546 xfs_trans_cancel(tp, cancel_flags);
2551 * In this case make sure to not release the inode until after
2552 * the current transaction is aborted. Releasing it beforehand
2553 * can cause us to go to xfs_inactive and start a recursive
2554 * transaction which can easily deadlock with the current one.
2556 xfs_bmap_cancel(&free_list);
2557 cancel_flags |= XFS_TRANS_ABORT;
2558 xfs_trans_cancel(tp, cancel_flags);
2561 * Before we drop our extra reference to the inode, purge it
2562 * from the refcache if it is there. By waiting until afterwards
2563 * to do the IRELE, we ensure that we won't go inactive in the
2564 * xfs_refcache_purge_ip routine (although that would be OK).
2566 xfs_refcache_purge_ip(ip);
2580 bhv_desc_t *target_dir_bdp,
2585 xfs_inode_t *tdp, *sip;
2588 xfs_inode_t *ips[2];
2590 xfs_bmap_free_t free_list;
2591 xfs_fsblock_t first_block;
2594 vnode_t *target_dir_vp;
2595 bhv_desc_t *src_bdp;
2597 char *target_name = VNAME(dentry);
2600 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2601 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2602 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2604 target_namelen = VNAMELEN(dentry);
2605 if (src_vp->v_type == VDIR)
2606 return XFS_ERROR(EPERM);
2609 * For now, manually find the XFS behavior descriptor for
2610 * the source vnode. If it doesn't exist then something
2611 * is wrong and we should just return an error.
2612 * Eventually we need to figure out how link is going to
2613 * work in the face of stacked vnodes.
2615 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2616 if (src_bdp == NULL) {
2617 return XFS_ERROR(EXDEV);
2619 sip = XFS_BHVTOI(src_bdp);
2620 tdp = XFS_BHVTOI(target_dir_bdp);
2622 if (XFS_FORCED_SHUTDOWN(mp))
2623 return XFS_ERROR(EIO);
2625 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2626 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2627 target_dir_vp, DM_RIGHT_NULL,
2628 src_vp, DM_RIGHT_NULL,
2629 target_name, NULL, 0, 0, 0);
2634 /* Return through std_return after this point. */
2636 error = XFS_QM_DQATTACH(mp, sip, 0);
2637 if (!error && sip != tdp)
2638 error = XFS_QM_DQATTACH(mp, tdp, 0);
2642 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2643 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2644 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2645 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2646 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2647 if (error == ENOSPC) {
2649 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2650 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2657 if (sip->i_ino < tdp->i_ino) {
2665 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2668 * Increment vnode ref counts since xfs_trans_commit &
2669 * xfs_trans_cancel will both unlock the inodes and
2670 * decrement the associated ref counts.
2673 VN_HOLD(target_dir_vp);
2674 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2675 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2678 * If the source has too many links, we can't make any more to it.
2680 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2681 error = XFS_ERROR(EMLINK);
2686 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2690 XFS_BMAP_INIT(&free_list, &first_block);
2692 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2693 sip->i_ino, &first_block, &free_list,
2697 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2699 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2701 error = xfs_bumplink(tp, sip);
2707 * If this is a synchronous mount, make sure that the
2708 * link transaction goes to disk before returning to
2711 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2712 xfs_trans_set_sync(tp);
2715 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2717 xfs_bmap_cancel(&free_list);
2721 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2726 /* Fall through to std_return with error = 0. */
2728 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2729 DM_EVENT_POSTLINK)) {
2730 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2731 target_dir_vp, DM_RIGHT_NULL,
2732 src_vp, DM_RIGHT_NULL,
2733 target_name, NULL, 0, error, 0);
2738 cancel_flags |= XFS_TRANS_ABORT;
2741 xfs_trans_cancel(tp, cancel_flags);
2751 bhv_desc_t *dir_bdp,
2757 char *dir_name = VNAME(dentry);
2759 xfs_inode_t *cdp; /* inode of created dir */
2760 vnode_t *cvp; /* vnode of created dir */
2766 xfs_bmap_free_t free_list;
2767 xfs_fsblock_t first_block;
2769 boolean_t dp_joined_to_trans;
2770 boolean_t created = B_FALSE;
2771 int dm_event_sent = 0;
2773 struct xfs_dquot *udqp, *gdqp;
2778 dir_vp = BHV_TO_VNODE(dir_bdp);
2779 dp = XFS_BHVTOI(dir_bdp);
2782 if (XFS_FORCED_SHUTDOWN(mp))
2783 return XFS_ERROR(EIO);
2785 dir_namelen = VNAMELEN(dentry);
2788 dp_joined_to_trans = B_FALSE;
2789 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2791 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2792 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2793 dir_vp, DM_RIGHT_NULL, NULL,
2794 DM_RIGHT_NULL, dir_name, NULL,
2801 /* Return through std_return after this point. */
2803 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2807 if (vap->va_mask & XFS_AT_PROJID)
2808 prid = (xfs_prid_t)vap->va_projid;
2810 prid = (xfs_prid_t)dfltprid;
2813 * Make sure that we have allocated dquot(s) on disk.
2815 error = XFS_QM_DQVOPALLOC(mp, dp,
2816 current_fsuid(credp), current_fsgid(credp),
2817 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2821 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2822 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2823 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2824 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2825 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2826 if (error == ENOSPC) {
2828 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2829 XFS_TRANS_PERM_LOG_RES,
2830 XFS_MKDIR_LOG_COUNT);
2838 xfs_ilock(dp, XFS_ILOCK_EXCL);
2841 * Check for directory link count overflow.
2843 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2844 error = XFS_ERROR(EMLINK);
2849 * Reserve disk quota and the inode.
2851 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2856 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2859 * create the directory inode.
2861 error = xfs_dir_ialloc(&tp, dp,
2862 MAKEIMODE(vap->va_type,vap->va_mode), 2,
2863 0, credp, prid, resblks > 0,
2866 if (error == ENOSPC)
2873 * Now we add the directory inode to the transaction.
2874 * We waited until now since xfs_dir_ialloc might start
2875 * a new transaction. Had we joined the transaction
2876 * earlier, the locks might have gotten released.
2879 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2880 dp_joined_to_trans = B_TRUE;
2882 XFS_BMAP_INIT(&free_list, &first_block);
2884 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2885 cdp->i_ino, &first_block, &free_list,
2886 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2888 ASSERT(error != ENOSPC);
2891 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2894 * Bump the in memory version number of the parent directory
2895 * so that other processes accessing it will recognize that
2896 * the directory has changed.
2900 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2906 error = xfs_bumplink(tp, dp);
2911 cvp = XFS_ITOV(cdp);
2919 * Attach the dquots to the new inode and modify the icount incore.
2921 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2924 * If this is a synchronous mount, make sure that the
2925 * mkdir transaction goes to disk before returning to
2928 if (mp->m_flags & XFS_MOUNT_WSYNC) {
2929 xfs_trans_set_sync(tp);
2932 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2938 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2939 XFS_QM_DQRELE(mp, udqp);
2940 XFS_QM_DQRELE(mp, gdqp);
2945 /* Fall through to std_return with error = 0 or errno from
2946 * xfs_trans_commit. */
2949 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2950 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2951 DM_EVENT_POSTCREATE)) {
2952 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2953 dir_vp, DM_RIGHT_NULL,
2954 created ? XFS_ITOV(cdp):NULL,
2957 dm_di_mode, error, 0);
2963 xfs_bmap_cancel(&free_list);
2965 cancel_flags |= XFS_TRANS_ABORT;
2967 xfs_trans_cancel(tp, cancel_flags);
2968 XFS_QM_DQRELE(mp, udqp);
2969 XFS_QM_DQRELE(mp, gdqp);
2971 if (!dp_joined_to_trans && (dp != NULL)) {
2972 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2985 bhv_desc_t *dir_bdp,
2989 char *name = VNAME(dentry);
2991 xfs_inode_t *cdp; /* child directory */
2995 xfs_bmap_free_t free_list;
2996 xfs_fsblock_t first_block;
3005 dir_vp = BHV_TO_VNODE(dir_bdp);
3006 dp = XFS_BHVTOI(dir_bdp);
3009 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3011 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3012 return XFS_ERROR(EIO);
3013 namelen = VNAMELEN(dentry);
3015 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3016 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3017 dir_vp, DM_RIGHT_NULL,
3018 NULL, DM_RIGHT_NULL,
3019 name, NULL, 0, 0, 0);
3021 return XFS_ERROR(error);
3024 /* Return through std_return after this point. */
3029 * We need to get a reference to cdp before we get our log
3030 * reservation. The reason for this is that we cannot call
3031 * xfs_iget for an inode for which we do not have a reference
3032 * once we've acquired a log reservation. This is because the
3033 * inode we are trying to get might be in xfs_inactive going
3034 * for a log reservation. Since we'll have to wait for the
3035 * inactive code to complete before returning from xfs_iget,
3036 * we need to make sure that we don't have log space reserved
3037 * when we call xfs_iget. Instead we get an unlocked referece
3038 * to the inode before getting our log reservation.
3040 error = xfs_get_dir_entry(dentry, &cdp);
3042 REMOVE_DEBUG_TRACE(__LINE__);
3046 dm_di_mode = cdp->i_d.di_mode;
3049 * Get the dquots for the inodes.
3051 error = XFS_QM_DQATTACH(mp, dp, 0);
3052 if (!error && dp != cdp)
3053 error = XFS_QM_DQATTACH(mp, cdp, 0);
3056 REMOVE_DEBUG_TRACE(__LINE__);
3060 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3061 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3063 * We try to get the real space reservation first,
3064 * allowing for directory btree deletion(s) implying
3065 * possible bmap insert(s). If we can't get the space
3066 * reservation then we use 0 instead, and avoid the bmap
3067 * btree insert(s) in the directory code by, if the bmap
3068 * insert tries to happen, instead trimming the LAST
3069 * block from the directory.
3071 resblks = XFS_REMOVE_SPACE_RES(mp);
3072 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3073 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3074 if (error == ENOSPC) {
3076 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3077 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3080 ASSERT(error != ENOSPC);
3085 XFS_BMAP_INIT(&free_list, &first_block);
3088 * Now lock the child directory inode and the parent directory
3089 * inode in the proper order. This will take care of validating
3090 * that the directory entry for the child directory inode has
3091 * not changed while we were obtaining a log reservation.
3093 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3095 xfs_trans_cancel(tp, cancel_flags);
3100 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3103 * Only increment the parent directory vnode count if
3104 * we didn't bump it in looking up cdp. The only time
3105 * we don't bump it is when we're looking up ".".
3111 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3113 ASSERT(cdp->i_d.di_nlink >= 2);
3114 if (cdp->i_d.di_nlink != 2) {
3115 error = XFS_ERROR(ENOTEMPTY);
3118 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3119 error = XFS_ERROR(ENOTEMPTY);
3123 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3124 &first_block, &free_list, resblks);
3129 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3132 * Bump the in memory generation count on the parent
3133 * directory so that other can know that it has changed.
3138 * Drop the link from cdp's "..".
3140 error = xfs_droplink(tp, dp);
3146 * Drop the link from dp to cdp.
3148 error = xfs_droplink(tp, cdp);
3154 * Drop the "." link from cdp to self.
3156 error = xfs_droplink(tp, cdp);
3161 /* Determine these before committing transaction */
3162 last_cdp_link = (cdp)->i_d.di_nlink==0;
3165 * Take an extra ref on the child vnode so that it
3166 * does not go to xfs_inactive() from within the commit.
3171 * If this is a synchronous mount, make sure that the
3172 * rmdir transaction goes to disk before returning to
3175 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3176 xfs_trans_set_sync(tp);
3179 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3181 xfs_bmap_cancel(&free_list);
3182 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3188 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3196 * Let interposed file systems know about removed links.
3198 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3202 /* Fall through to std_return with error = 0 or the errno
3203 * from xfs_trans_commit. */
3205 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3206 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3207 dir_vp, DM_RIGHT_NULL,
3208 NULL, DM_RIGHT_NULL,
3209 name, NULL, dm_di_mode,
3215 xfs_bmap_cancel(&free_list);
3216 cancel_flags |= XFS_TRANS_ABORT;
3218 xfs_trans_cancel(tp, cancel_flags);
3226 * Read dp's entries starting at uiop->uio_offset and translate them into
3227 * bufsize bytes worth of struct dirents starting at bufbase.
3231 bhv_desc_t *dir_bdp,
3237 xfs_trans_t *tp = NULL;
3240 xfs_off_t start_offset;
3242 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3243 (inst_t *)__return_address);
3244 dp = XFS_BHVTOI(dir_bdp);
3246 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3247 return XFS_ERROR(EIO);
3250 lock_mode = xfs_ilock_map_shared(dp);
3251 start_offset = uiop->uio_offset;
3252 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3253 if (start_offset != uiop->uio_offset) {
3254 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3256 xfs_iunlock_map_shared(dp, lock_mode);
3267 bhv_desc_t *dir_bdp,
3280 xfs_bmap_free_t free_list;
3281 xfs_fsblock_t first_block;
3282 boolean_t dp_joined_to_trans;
3286 xfs_fileoff_t first_fsb;
3287 xfs_filblks_t fs_blocks;
3289 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3296 struct xfs_dquot *udqp, *gdqp;
3298 char *link_name = VNAME(dentry);
3302 dir_vp = BHV_TO_VNODE(dir_bdp);
3303 dp = XFS_BHVTOI(dir_bdp);
3304 dp_joined_to_trans = B_FALSE;
3309 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3313 if (XFS_FORCED_SHUTDOWN(mp))
3314 return XFS_ERROR(EIO);
3316 link_namelen = VNAMELEN(dentry);
3319 * Check component lengths of the target path name.
3321 pathlen = strlen(target_path);
3322 if (pathlen >= MAXPATHLEN) /* total string too long */
3323 return XFS_ERROR(ENAMETOOLONG);
3324 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3328 for(total = 0, path = target_path; total < pathlen;) {
3332 while(*path == '/') {
3338 * Count up to the next slash or end of path.
3339 * Error out if the component is bigger than MAXNAMELEN.
3341 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3342 if (++len >= MAXNAMELEN) {
3343 error = ENAMETOOLONG;
3350 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3351 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3352 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3353 link_name, target_path, 0, 0, 0);
3358 /* Return through std_return after this point. */
3361 if (vap->va_mask & XFS_AT_PROJID)
3362 prid = (xfs_prid_t)vap->va_projid;
3364 prid = (xfs_prid_t)dfltprid;
3367 * Make sure that we have allocated dquot(s) on disk.
3369 error = XFS_QM_DQVOPALLOC(mp, dp,
3370 current_fsuid(credp), current_fsgid(credp),
3371 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3375 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3376 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3378 * The symlink will fit into the inode data fork?
3379 * There can't be any attributes so we get the whole variable part.
3381 if (pathlen <= XFS_LITINO(mp))
3384 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3385 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3386 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3387 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3388 if (error == ENOSPC && fs_blocks == 0) {
3390 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3391 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3399 xfs_ilock(dp, XFS_ILOCK_EXCL);
3402 * Check whether the directory allows new symlinks or not.
3404 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3405 error = XFS_ERROR(EPERM);
3410 * Reserve disk quota : blocks and inode.
3412 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3417 * Check for ability to enter directory entry, if no space reserved.
3420 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3423 * Initialize the bmap freelist prior to calling either
3424 * bmapi or the directory create code.
3426 XFS_BMAP_INIT(&free_list, &first_block);
3429 * Allocate an inode for the symlink.
3431 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3432 1, 0, credp, prid, resblks > 0, &ip, NULL);
3434 if (error == ENOSPC)
3441 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3442 dp_joined_to_trans = B_TRUE;
3445 * Also attach the dquot(s) to it, if applicable.
3447 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3450 resblks -= XFS_IALLOC_SPACE_RES(mp);
3452 * If the symlink will fit into the inode, write it inline.
3454 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3455 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3456 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3457 ip->i_d.di_size = pathlen;
3460 * The inode was initially created in extent format.
3462 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3463 ip->i_df.if_flags |= XFS_IFINLINE;
3465 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3466 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3470 nmaps = SYMLINK_MAPS;
3472 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3473 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3474 &first_block, resblks, mval, &nmaps,
3481 resblks -= fs_blocks;
3482 ip->i_d.di_size = pathlen;
3483 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3485 cur_chunk = target_path;
3486 for (n = 0; n < nmaps; n++) {
3487 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3488 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3489 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3490 BTOBB(byte_cnt), 0);
3491 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3492 if (pathlen < byte_cnt) {
3495 pathlen -= byte_cnt;
3497 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3498 cur_chunk += byte_cnt;
3500 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3505 * Create the directory entry for the symlink.
3507 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3508 ip->i_ino, &first_block, &free_list, resblks);
3512 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3513 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3516 * Bump the in memory version number of the parent directory
3517 * so that other processes accessing it will recognize that
3518 * the directory has changed.
3523 * If this is a synchronous mount, make sure that the
3524 * symlink transaction goes to disk before returning to
3527 if (mp->m_flags & XFS_MOUNT_WSYNC) {
3528 xfs_trans_set_sync(tp);
3532 * xfs_trans_commit normally decrements the vnode ref count
3533 * when it unlocks the inode. Since we want to return the
3534 * vnode to the caller, we bump the vnode ref count now.
3538 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3542 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3543 XFS_QM_DQRELE(mp, udqp);
3544 XFS_QM_DQRELE(mp, gdqp);
3546 /* Fall through to std_return with error = 0 or errno from
3547 * xfs_trans_commit */
3549 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3550 DM_EVENT_POSTSYMLINK)) {
3551 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3552 dir_vp, DM_RIGHT_NULL,
3553 error ? NULL : XFS_ITOV(ip),
3554 DM_RIGHT_NULL, link_name, target_path,
3570 xfs_bmap_cancel(&free_list);
3571 cancel_flags |= XFS_TRANS_ABORT;
3573 xfs_trans_cancel(tp, cancel_flags);
3574 XFS_QM_DQRELE(mp, udqp);
3575 XFS_QM_DQRELE(mp, gdqp);
3577 if (!dp_joined_to_trans && (dp != NULL)) {
3578 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3588 * A fid routine that takes a pointer to a previously allocated
3589 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3599 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3600 (inst_t *)__return_address);
3601 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3603 xfid = (xfs_fid2_t *)fidp;
3604 ip = XFS_BHVTOI(bdp);
3605 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3608 * use memcpy because the inode is a long long and there's no
3609 * assurance that xfid->fid_ino is properly aligned.
3611 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3612 xfid->fid_gen = ip->i_d.di_gen;
3629 vp = BHV_TO_VNODE(bdp);
3630 if (vp->v_type == VDIR)
3632 ip = XFS_BHVTOI(bdp);
3633 if (locktype == VRWLOCK_WRITE) {
3634 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3635 } else if (locktype == VRWLOCK_TRY_READ) {
3636 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3637 } else if (locktype == VRWLOCK_TRY_WRITE) {
3638 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3640 ASSERT((locktype == VRWLOCK_READ) ||
3641 (locktype == VRWLOCK_WRITE_DIRECT));
3642 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3660 vp = BHV_TO_VNODE(bdp);
3661 if (vp->v_type == VDIR)
3663 ip = XFS_BHVTOI(bdp);
3664 if (locktype == VRWLOCK_WRITE) {
3666 * In the write case, we may have added a new entry to
3667 * the reference cache. This might store a pointer to
3668 * an inode to be released in this inode. If it is there,
3669 * clear the pointer and release the inode after unlocking
3672 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3674 ASSERT((locktype == VRWLOCK_READ) ||
3675 (locktype == VRWLOCK_WRITE_DIRECT));
3676 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3688 xfs_inode_log_item_t *iip;
3691 ip = XFS_BHVTOI(bdp);
3695 if (XFS_FORCED_SHUTDOWN(mp))
3696 return XFS_ERROR(EIO);
3699 * Bypass inodes which have already been cleaned by
3700 * the inode flush clustering code inside xfs_iflush
3702 if ((ip->i_update_core == 0) &&
3703 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3706 if (flags & FLUSH_LOG) {
3707 if (iip && iip->ili_last_lsn) {
3708 xlog_t *log = mp->m_log;
3710 int s, log_flags = XFS_LOG_FORCE;
3712 s = GRANT_LOCK(log);
3713 sync_lsn = log->l_last_sync_lsn;
3714 GRANT_UNLOCK(log, s);
3716 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3719 if (flags & FLUSH_SYNC)
3720 log_flags |= XFS_LOG_SYNC;
3721 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3726 * We make this non-blocking if the inode is contended,
3727 * return EAGAIN to indicate to the caller that they
3728 * did not succeed. This prevents the flush path from
3729 * blocking on inodes inside another operation right
3730 * now, they get caught later by xfs_sync.
3732 if (flags & FLUSH_INODE) {
3735 if (xfs_ipincount(ip))
3738 if (flags & FLUSH_SYNC) {
3739 xfs_ilock(ip, XFS_ILOCK_SHARED);
3741 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3742 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3743 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3750 if (flags & FLUSH_SYNC)
3751 flush_flags = XFS_IFLUSH_SYNC;
3753 flush_flags = XFS_IFLUSH_ASYNC;
3755 error = xfs_iflush(ip, flush_flags);
3756 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3775 if (!capable(CAP_SYS_ADMIN))
3776 return XFS_ERROR(EPERM);
3778 ip = XFS_BHVTOI(bdp);
3781 if (XFS_FORCED_SHUTDOWN(mp))
3782 return XFS_ERROR(EIO);
3784 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3785 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3787 xfs_trans_cancel(tp, 0);
3790 xfs_ilock(ip, XFS_ILOCK_EXCL);
3791 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3793 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3794 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3796 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3798 error = xfs_trans_commit(tp, 0, NULL);
3814 vp = BHV_TO_VNODE(bdp);
3815 ip = XFS_BHVTOI(bdp);
3817 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3819 ASSERT(!VN_MAPPED(vp));
3821 /* bad inode, get out here ASAP */
3827 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3828 if (ip->i_d.di_size > 0) {
3830 * Flush and invalidate any data left around that is
3831 * a part of this file.
3833 * Get the inode's i/o lock so that buffers are pushed
3834 * out while holding the proper lock. We can't hold
3835 * the inode lock here since flushing out buffers may
3836 * cause us to try to get the lock in xfs_strategy().
3838 * We don't have to call remapf() here, because there
3839 * cannot be any mapped file references to this vnode
3840 * since it is being reclaimed.
3842 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3845 * If we hit an IO error, we need to make sure that the
3846 * buffer and page caches of file data for
3847 * the file are tossed away. We don't want to use
3848 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3849 * pages to stay attached to the vnode, but be
3850 * marked P_BAD. pdflush/vnode_pagebad
3853 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3854 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3856 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3859 ASSERT(VN_CACHED(vp) == 0);
3860 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3861 ip->i_delayed_blks == 0);
3862 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3863 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3865 * di_size field may not be quite accurate if we're
3868 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3869 ASSERT(VN_CACHED(vp) == 0);
3873 /* If we have nothing to flush with this inode then complete the
3874 * teardown now, otherwise break the link between the xfs inode
3875 * and the linux inode and clean up the xfs inode later. This
3876 * avoids flushing the inode to disk during the delete operation
3879 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3880 xfs_ilock(ip, XFS_ILOCK_EXCL);
3882 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3884 xfs_mount_t *mp = ip->i_mount;
3886 /* Protect sync from us */
3887 XFS_MOUNT_ILOCK(mp);
3888 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3889 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3890 ip->i_flags |= XFS_IRECLAIMABLE;
3891 XFS_MOUNT_IUNLOCK(mp);
3902 xfs_ihash_t *ih = ip->i_hash;
3903 vnode_t *vp = XFS_ITOV_NULL(ip);
3906 if (vp && VN_BAD(vp))
3909 /* The hash lock here protects a thread in xfs_iget_core from
3910 * racing with us on linking the inode back with a vnode.
3911 * Once we have the XFS_IRECLAIM flag set it will not touch
3914 write_lock(&ih->ih_lock);
3915 if ((ip->i_flags & XFS_IRECLAIM) ||
3916 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3917 write_unlock(&ih->ih_lock);
3920 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3924 ip->i_flags |= XFS_IRECLAIM;
3925 write_unlock(&ih->ih_lock);
3928 * If the inode is still dirty, then flush it out. If the inode
3929 * is not in the AIL, then it will be OK to flush it delwri as
3930 * long as xfs_iflush() does not keep any references to the inode.
3931 * We leave that decision up to xfs_iflush() since it has the
3932 * knowledge of whether it's OK to simply do a delwri flush of
3933 * the inode or whether we need to wait until the inode is
3934 * pulled from the AIL.
3935 * We get the flush lock regardless, though, just to make sure
3936 * we don't free it while it is being flushed.
3938 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3940 xfs_ilock(ip, XFS_ILOCK_EXCL);
3944 if (ip->i_update_core ||
3945 ((ip->i_itemp != NULL) &&
3946 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3947 error = xfs_iflush(ip, sync_mode);
3949 * If we hit an error, typically because of filesystem
3950 * shutdown, we don't need to let vn_reclaim to know
3951 * because we're gonna reclaim the inode anyway.
3954 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3957 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3960 ASSERT(ip->i_update_core == 0);
3961 ASSERT(ip->i_itemp == NULL ||
3962 ip->i_itemp->ili_format.ilf_fields == 0);
3963 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3964 } else if (locked) {
3966 * We are not interested in doing an iflush if we're
3967 * in the process of shutting down the filesystem forcibly.
3968 * So, just reclaim the inode.
3971 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3980 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3983 xfs_inode_t *ip, *n;
3988 XFS_MOUNT_ILOCK(mp);
3989 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3991 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3993 if (xfs_ipincount(ip) ||
3994 !xfs_iflock_nowait(ip)) {
3995 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3999 XFS_MOUNT_IUNLOCK(mp);
4000 xfs_finish_reclaim(ip, noblock,
4001 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
4009 XFS_MOUNT_IUNLOCK(mp);
4014 * xfs_alloc_file_space()
4015 * This routine allocates disk space for the given file.
4017 * If alloc_type == 0, this request is for an ALLOCSP type
4018 * request which will change the file size. In this case, no
4019 * DMAPI event will be generated by the call. A TRUNCATE event
4020 * will be generated later by xfs_setattr.
4022 * If alloc_type != 0, this request is for a RESVSP type
4023 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4024 * lower block boundary byte address is less than the file's
4033 xfs_alloc_file_space(
4040 xfs_filblks_t allocated_fsb;
4041 xfs_filblks_t allocatesize_fsb;
4044 xfs_filblks_t datablocks;
4046 xfs_fsblock_t firstfsb;
4047 xfs_bmap_free_t free_list;
4048 xfs_bmbt_irec_t *imapp;
4049 xfs_bmbt_irec_t imaps[1];
4056 xfs_fileoff_t startoffset_fsb;
4058 int xfs_bmapi_flags;
4060 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4063 if (XFS_FORCED_SHUTDOWN(mp))
4064 return XFS_ERROR(EIO);
4067 * determine if this is a realtime file
4069 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4070 if (ip->i_d.di_extsize)
4071 rtextsize = ip->i_d.di_extsize;
4073 rtextsize = mp->m_sb.sb_rextsize;
4077 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4081 return XFS_ERROR(EINVAL);
4087 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4088 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4089 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4091 /* Generate a DMAPI event if needed. */
4092 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4093 (attr_flags&ATTR_DMI) == 0 &&
4094 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4095 xfs_off_t end_dmi_offset;
4097 end_dmi_offset = offset+len;
4098 if (end_dmi_offset > ip->i_d.di_size)
4099 end_dmi_offset = ip->i_d.di_size;
4100 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4101 offset, end_dmi_offset - offset,
4108 * allocate file space until done or until there is an error
4111 while (allocatesize_fsb && !error) {
4113 * determine if reserving space on
4114 * the data or realtime partition.
4119 s = startoffset_fsb;
4120 do_div(s, rtextsize);
4122 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4124 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4127 datablocks = allocatesize_fsb;
4132 * allocate and setup the transaction
4134 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4135 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4136 error = xfs_trans_reserve(tp,
4138 XFS_WRITE_LOG_RES(mp),
4140 XFS_TRANS_PERM_LOG_RES,
4141 XFS_WRITE_LOG_COUNT);
4144 * check for running out of space
4148 * Free the transaction structure.
4150 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4151 xfs_trans_cancel(tp, 0);
4154 xfs_ilock(ip, XFS_ILOCK_EXCL);
4155 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4156 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4157 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4161 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4162 xfs_trans_ihold(tp, ip);
4165 * issue the bmapi() call to allocate the blocks
4167 XFS_BMAP_INIT(&free_list, &firstfsb);
4168 error = xfs_bmapi(tp, ip, startoffset_fsb,
4169 allocatesize_fsb, xfs_bmapi_flags,
4170 &firstfsb, 0, imapp, &reccount,
4177 * complete the transaction
4179 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4184 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4185 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4190 allocated_fsb = imapp->br_blockcount;
4192 if (reccount == 0) {
4193 error = XFS_ERROR(ENOSPC);
4197 startoffset_fsb += allocated_fsb;
4198 allocatesize_fsb -= allocated_fsb;
4201 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4202 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4204 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4205 XFS_ITOV(ip), DM_RIGHT_NULL,
4206 XFS_ITOV(ip), DM_RIGHT_NULL,
4207 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4209 goto retry; /* Maybe DMAPI app. has made space */
4210 /* else fall through with error from XFS_SEND_DATA */
4216 xfs_bmap_cancel(&free_list);
4218 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4219 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4220 goto dmapi_enospc_check;
4224 * Zero file bytes between startoff and endoff inclusive.
4225 * The iolock is held exclusive and no blocks are buffered.
4228 xfs_zero_remaining_bytes(
4233 xfs_bmbt_irec_t imap;
4234 xfs_fileoff_t offset_fsb;
4235 xfs_off_t lastoffset;
4238 xfs_mount_t *mp = ip->i_mount;
4242 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4243 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4244 mp->m_rtdev_targp : mp->m_ddev_targp);
4246 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4247 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4249 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4251 if (error || nimap < 1)
4253 ASSERT(imap.br_blockcount >= 1);
4254 ASSERT(imap.br_startoff == offset_fsb);
4255 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4256 if (lastoffset > endoff)
4257 lastoffset = endoff;
4258 if (imap.br_startblock == HOLESTARTBLOCK)
4260 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4261 if (imap.br_state == XFS_EXT_UNWRITTEN)
4264 XFS_BUF_UNWRITE(bp);
4266 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4268 if ((error = xfs_iowait(bp))) {
4269 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4270 mp, bp, XFS_BUF_ADDR(bp));
4273 memset(XFS_BUF_PTR(bp) +
4274 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4275 0, lastoffset - offset + 1);
4280 if ((error = xfs_iowait(bp))) {
4281 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4282 mp, bp, XFS_BUF_ADDR(bp));
4291 * xfs_free_file_space()
4292 * This routine frees disk space for the given file.
4294 * This routine is only called by xfs_change_file_space
4295 * for an UNRESVSP type call.
4303 xfs_free_file_space(
4311 xfs_off_t end_dmi_offset;
4312 xfs_fileoff_t endoffset_fsb;
4314 xfs_fsblock_t firstfsb;
4315 xfs_bmap_free_t free_list;
4317 xfs_bmbt_irec_t imap;
4325 xfs_fileoff_t startoffset_fsb;
4327 int need_iolock = (attr_flags & ATTR_DMI) == 0;
4329 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4332 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4336 if (len <= 0) /* if nothing being freed */
4338 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4339 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4340 end_dmi_offset = offset + len;
4341 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4343 if (offset < ip->i_d.di_size &&
4344 (attr_flags & ATTR_DMI) == 0 &&
4345 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4346 if (end_dmi_offset > ip->i_d.di_size)
4347 end_dmi_offset = ip->i_d.di_size;
4348 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4349 offset, end_dmi_offset - offset,
4350 AT_DELAY_FLAG(attr_flags), NULL);
4356 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4357 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4359 ilen = len + (offset & (rounding - 1));
4360 ioffset = offset & ~(rounding - 1);
4361 if (ilen & (rounding - 1))
4362 ilen = (ilen + rounding) & ~(rounding - 1);
4363 xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4365 * Need to zero the stuff we're not freeing, on disk.
4366 * If its a realtime file & can't use unwritten extents then we
4367 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4368 * will take care of it for us.
4370 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4372 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4373 &imap, &nimap, NULL);
4375 goto out_unlock_iolock;
4376 ASSERT(nimap == 0 || nimap == 1);
4377 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4380 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4381 block = imap.br_startblock;
4382 mod = do_div(block, mp->m_sb.sb_rextsize);
4384 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4387 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4388 &imap, &nimap, NULL);
4390 goto out_unlock_iolock;
4391 ASSERT(nimap == 0 || nimap == 1);
4392 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4393 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4395 if (mod && (mod != mp->m_sb.sb_rextsize))
4396 endoffset_fsb -= mod;
4399 if ((done = (endoffset_fsb <= startoffset_fsb)))
4401 * One contiguous piece to clear
4403 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4406 * Some full blocks, possibly two pieces to clear
4408 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4409 error = xfs_zero_remaining_bytes(ip, offset,
4410 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4412 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4413 error = xfs_zero_remaining_bytes(ip,
4414 XFS_FSB_TO_B(mp, endoffset_fsb),
4419 * free file space until done or until there is an error
4421 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4422 while (!error && !done) {
4425 * allocate and setup the transaction
4427 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4428 error = xfs_trans_reserve(tp,
4430 XFS_WRITE_LOG_RES(mp),
4432 XFS_TRANS_PERM_LOG_RES,
4433 XFS_WRITE_LOG_COUNT);
4436 * check for running out of space
4440 * Free the transaction structure.
4442 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4443 xfs_trans_cancel(tp, 0);
4446 xfs_ilock(ip, XFS_ILOCK_EXCL);
4447 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4448 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4449 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4453 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4454 xfs_trans_ihold(tp, ip);
4457 * issue the bunmapi() call to free the blocks
4459 XFS_BMAP_INIT(&free_list, &firstfsb);
4460 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4461 endoffset_fsb - startoffset_fsb,
4462 0, 2, &firstfsb, &free_list, &done);
4468 * complete the transaction
4470 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4475 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4476 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4481 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4485 xfs_bmap_cancel(&free_list);
4487 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4488 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4494 * xfs_change_file_space()
4495 * This routine allocates or frees disk space for the given file.
4496 * The user specified parameters are checked for alignment and size
4505 xfs_change_file_space(
4519 xfs_off_t startoffset;
4525 vp = BHV_TO_VNODE(bdp);
4526 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4528 ip = XFS_BHVTOI(bdp);
4532 * must be a regular file and have write permission
4534 if (vp->v_type != VREG)
4535 return XFS_ERROR(EINVAL);
4537 xfs_ilock(ip, XFS_ILOCK_SHARED);
4539 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4540 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4544 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4546 switch (bf->l_whence) {
4547 case 0: /*SEEK_SET*/
4549 case 1: /*SEEK_CUR*/
4550 bf->l_start += offset;
4552 case 2: /*SEEK_END*/
4553 bf->l_start += ip->i_d.di_size;
4556 return XFS_ERROR(EINVAL);
4559 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4561 if ( (bf->l_start < 0)
4562 || (bf->l_start > XFS_MAXIOFFSET(mp))
4563 || (bf->l_start + llen < 0)
4564 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4565 return XFS_ERROR(EINVAL);
4569 startoffset = bf->l_start;
4570 fsize = ip->i_d.di_size;
4573 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4575 * These calls do NOT zero the data space allocated to the file,
4576 * nor do they change the file size.
4578 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4580 * These calls cause the new file data to be zeroed and the file
4581 * size to be changed.
4583 setprealloc = clrprealloc = 0;
4586 case XFS_IOC_RESVSP:
4587 case XFS_IOC_RESVSP64:
4588 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4595 case XFS_IOC_UNRESVSP:
4596 case XFS_IOC_UNRESVSP64:
4597 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4602 case XFS_IOC_ALLOCSP:
4603 case XFS_IOC_ALLOCSP64:
4604 case XFS_IOC_FREESP:
4605 case XFS_IOC_FREESP64:
4606 if (startoffset > fsize) {
4607 error = xfs_alloc_file_space(ip, fsize,
4608 startoffset - fsize, 0, attr_flags);
4613 va.va_mask = XFS_AT_SIZE;
4614 va.va_size = startoffset;
4616 error = xfs_setattr(bdp, &va, attr_flags, credp);
4626 return XFS_ERROR(EINVAL);
4630 * update the inode timestamp, mode, and prealloc flag bits
4632 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4634 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4637 xfs_trans_cancel(tp, 0);
4641 xfs_ilock(ip, XFS_ILOCK_EXCL);
4643 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4644 xfs_trans_ihold(tp, ip);
4646 if ((attr_flags & ATTR_DMI) == 0) {
4647 ip->i_d.di_mode &= ~S_ISUID;
4650 * Note that we don't have to worry about mandatory
4651 * file locking being disabled here because we only
4652 * clear the S_ISGID bit if the Group execute bit is
4653 * on, but if it was on then mandatory locking wouldn't
4654 * have been enabled.
4656 if (ip->i_d.di_mode & S_IXGRP)
4657 ip->i_d.di_mode &= ~S_ISGID;
4659 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4662 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4663 else if (clrprealloc)
4664 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4666 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4667 xfs_trans_set_sync(tp);
4669 error = xfs_trans_commit(tp, 0, NULL);
4671 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4676 vnodeops_t xfs_vnodeops = {
4677 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4678 .vop_open = xfs_open,
4679 .vop_read = xfs_read,
4680 #ifdef HAVE_SENDFILE
4681 .vop_sendfile = xfs_sendfile,
4683 .vop_write = xfs_write,
4684 .vop_ioctl = xfs_ioctl,
4685 .vop_getattr = xfs_getattr,
4686 .vop_setattr = xfs_setattr,
4687 .vop_access = xfs_access,
4688 .vop_lookup = xfs_lookup,
4689 .vop_create = xfs_create,
4690 .vop_remove = xfs_remove,
4691 .vop_link = xfs_link,
4692 .vop_rename = xfs_rename,
4693 .vop_mkdir = xfs_mkdir,
4694 .vop_rmdir = xfs_rmdir,
4695 .vop_readdir = xfs_readdir,
4696 .vop_symlink = xfs_symlink,
4697 .vop_readlink = xfs_readlink,
4698 .vop_fsync = xfs_fsync,
4699 .vop_inactive = xfs_inactive,
4700 .vop_fid2 = xfs_fid2,
4701 .vop_rwlock = xfs_rwlock,
4702 .vop_rwunlock = xfs_rwunlock,
4703 .vop_bmap = xfs_bmap,
4704 .vop_reclaim = xfs_reclaim,
4705 .vop_attr_get = xfs_attr_get,
4706 .vop_attr_set = xfs_attr_set,
4707 .vop_attr_remove = xfs_attr_remove,
4708 .vop_attr_list = xfs_attr_list,
4709 .vop_link_removed = (vop_link_removed_t)fs_noval,
4710 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4711 .vop_tosspages = fs_tosspages,
4712 .vop_flushinval_pages = fs_flushinval_pages,
4713 .vop_flush_pages = fs_flush_pages,
4714 .vop_release = xfs_release,
4715 .vop_iflush = xfs_inode_flush,