aboutsummaryrefslogtreecommitdiffstats
path: root/fs/xfs/xfs_vnodeops.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /fs/xfs/xfs_vnodeops.c
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'fs/xfs/xfs_vnodeops.c')
-rw-r--r--fs/xfs/xfs_vnodeops.c4712
1 files changed, 4712 insertions, 0 deletions
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
new file mode 100644
index 000000000000..70092963ca9e
--- /dev/null
+++ b/fs/xfs/xfs_vnodeops.c
@@ -0,0 +1,4712 @@
1/*
2 * Copyright (c) 2000-2004 Silicon Graphics, Inc. All Rights Reserved.
3 *
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.
7 *
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.
11 *
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.
18 *
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.
22 *
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
25 *
26 * http://www.sgi.com
27 *
28 * For further information regarding this notice, see:
29 *
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31 */
32
33#include "xfs.h"
34#include "xfs_macros.h"
35#include "xfs_types.h"
36#include "xfs_inum.h"
37#include "xfs_log.h"
38#include "xfs_trans.h"
39#include "xfs_sb.h"
40#include "xfs_ag.h"
41#include "xfs_dir.h"
42#include "xfs_dir2.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"
58#include "xfs_bmap.h"
59#include "xfs_da_btree.h"
60#include "xfs_attr.h"
61#include "xfs_rw.h"
62#include "xfs_refcache.h"
63#include "xfs_error.h"
64#include "xfs_bit.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"
70#include "xfs_mac.h"
71#include "xfs_log_priv.h"
72
73
74/*
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
77 * bmapi.
78 */
79#define SYMLINK_MAPS 2
80
81/*
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.
87 */
88STATIC int
89xfs_open(
90 bhv_desc_t *bdp,
91 cred_t *credp)
92{
93 int mode;
94 vnode_t *vp;
95 xfs_inode_t *ip;
96
97 vp = BHV_TO_VNODE(bdp);
98 ip = XFS_BHVTOI(bdp);
99
100 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
101 return XFS_ERROR(EIO);
102
103 /*
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.
106 */
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);
112 }
113 return 0;
114}
115
116
117/*
118 * xfs_getattr
119 */
120STATIC int
121xfs_getattr(
122 bhv_desc_t *bdp,
123 vattr_t *vap,
124 int flags,
125 cred_t *credp)
126{
127 xfs_inode_t *ip;
128 xfs_mount_t *mp;
129 vnode_t *vp;
130
131 vp = BHV_TO_VNODE(bdp);
132 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
133
134 ip = XFS_BHVTOI(bdp);
135 mp = ip->i_mount;
136
137 if (XFS_FORCED_SHUTDOWN(mp))
138 return XFS_ERROR(EIO);
139
140 if (!(flags & ATTR_LAZY))
141 xfs_ilock(ip, XFS_ILOCK_SHARED);
142
143 vap->va_size = ip->i_d.di_size;
144 if (vap->va_mask == XFS_AT_SIZE)
145 goto all_done;
146
147 vap->va_nblocks =
148 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
149 vap->va_nodeid = ip->i_ino;
150#if XFS_BIG_INUMS
151 vap->va_nodeid += mp->m_inoadd;
152#endif
153 vap->va_nlink = ip->i_d.di_nlink;
154
155 /*
156 * Quick exit for non-stat callers
157 */
158 if ((vap->va_mask &
159 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
160 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
161 goto all_done;
162
163 /*
164 * Copy from in-core inode.
165 */
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;
171
172 /*
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.
176 */
177 if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
178 vap->va_rdev = 0;
179
180 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
181
182#if 0
183 /* Large block sizes confuse various
184 * user space programs, so letting the
185 * stripe size through is not a good
186 * idea for now.
187 */
188 vap->va_blocksize = mp->m_swidth ?
189 /*
190 * If the underlying volume is a stripe, then
191 * return the stripe width in bytes as the
192 * recommended I/O size.
193 */
194 (mp->m_swidth << mp->m_sb.sb_blocklog) :
195 /*
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.
200 */
201 (1 << (int)MAX(mp->m_readio_log,
202 mp->m_writeio_log));
203
204#else
205 vap->va_blocksize =
206 /*
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.
211 */
212 1 << (int)MAX(mp->m_readio_log,
213 mp->m_writeio_log);
214#endif
215 } else {
216
217 /*
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
221 * extent size.
222 */
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);
226 }
227 } else {
228 vap->va_rdev = ip->i_df.if_u2.if_rdev;
229 vap->va_blocksize = BLKDEV_IOSIZE;
230 }
231
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;
238
239 /*
240 * Exit for stat callers. See if any of the rest of the fields
241 * to be filled in are needed.
242 */
243 if ((vap->va_mask &
244 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
245 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
246 goto all_done;
247
248 /*
249 * Convert di_flags to xflags.
250 */
251 vap->va_xflags = xfs_ip2xflags(ip);
252
253 /*
254 * Exit for inode revalidate. See if any of the rest of
255 * the fields to be filled in are needed.
256 */
257 if ((vap->va_mask &
258 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
259 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
260 goto all_done;
261
262 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
263 vap->va_nextents =
264 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
265 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
266 ip->i_d.di_nextents;
267 if (ip->i_afp)
268 vap->va_anextents =
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;
272 else
273 vap->va_anextents = 0;
274 vap->va_gen = ip->i_d.di_gen;
275
276 all_done:
277 if (!(flags & ATTR_LAZY))
278 xfs_iunlock(ip, XFS_ILOCK_SHARED);
279 return 0;
280}
281
282
283/*
284 * xfs_setattr
285 */
286int
287xfs_setattr(
288 bhv_desc_t *bdp,
289 vattr_t *vap,
290 int flags,
291 cred_t *credp)
292{
293 xfs_inode_t *ip;
294 xfs_trans_t *tp;
295 xfs_mount_t *mp;
296 int mask;
297 int code;
298 uint lock_flags;
299 uint commit_flags=0;
300 uid_t uid=0, iuid=0;
301 gid_t gid=0, igid=0;
302 int timeflags = 0;
303 vnode_t *vp;
304 xfs_prid_t projid=0, iprojid=0;
305 int mandlock_before, mandlock_after;
306 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
307 int file_owner;
308 int need_iolock = (flags & ATTR_DMI) == 0;
309
310 vp = BHV_TO_VNODE(bdp);
311 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
312
313 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
314 return XFS_ERROR(EROFS);
315
316 /*
317 * Cannot set certain attributes.
318 */
319 mask = vap->va_mask;
320 if (mask & XFS_AT_NOSET) {
321 return XFS_ERROR(EINVAL);
322 }
323
324 ip = XFS_BHVTOI(bdp);
325 mp = ip->i_mount;
326
327 if (XFS_FORCED_SHUTDOWN(mp))
328 return XFS_ERROR(EIO);
329
330 /*
331 * Timestamps do not need to be logged and hence do not
332 * need to be done within a transaction.
333 */
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);
340 return 0;
341 }
342
343 olddquot1 = olddquot2 = NULL;
344 udqp = gdqp = NULL;
345
346 /*
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.
353 */
354 if (XFS_IS_QUOTA_ON(mp) && (mask & (XFS_AT_UID|XFS_AT_GID))) {
355 uint qflags = 0;
356
357 if (mask & XFS_AT_UID) {
358 uid = vap->va_uid;
359 qflags |= XFS_QMOPT_UQUOTA;
360 } else {
361 uid = ip->i_d.di_uid;
362 }
363 if (mask & XFS_AT_GID) {
364 gid = vap->va_gid;
365 qflags |= XFS_QMOPT_GQUOTA;
366 } else {
367 gid = ip->i_d.di_gid;
368 }
369 /*
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.
373 */
374 ASSERT(udqp == NULL);
375 ASSERT(gdqp == NULL);
376 code = XFS_QM_DQVOPALLOC(mp, ip, uid,gid, qflags, &udqp, &gdqp);
377 if (code)
378 return (code);
379 }
380
381 /*
382 * For the other attributes, we acquire the inode lock and
383 * first do an error checking pass.
384 */
385 tp = NULL;
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);
391 commit_flags = 0;
392 if ((code = xfs_trans_reserve(tp, 0,
393 XFS_ICHANGE_LOG_RES(mp), 0,
394 0, 0))) {
395 lock_flags = 0;
396 goto error_return;
397 }
398 }
399 } else {
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);
405 if (code) {
406 lock_flags = 0;
407 goto error_return;
408 }
409 }
410 if (need_iolock)
411 lock_flags |= XFS_IOLOCK_EXCL;
412 }
413
414 xfs_ilock(ip, lock_flags);
415
416 /* boolean: are we the file owner? */
417 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
418
419 /*
420 * Change various properties of a file.
421 * Only the owner or users with CAP_FOWNER
422 * capability may do these things.
423 */
424 if (mask &
425 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
426 XFS_AT_GID|XFS_AT_PROJID)) {
427 /*
428 * CAP_FOWNER overrides the following restrictions:
429 *
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.
433 */
434 if (!file_owner && !capable(CAP_FOWNER)) {
435 code = XFS_ERROR(EPERM);
436 goto error_return;
437 }
438
439 /*
440 * CAP_FSETID overrides the following restrictions:
441 *
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.
445 *
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
449 */
450 if (mask & XFS_AT_MODE) {
451 mode_t m = 0;
452
453 if ((vap->va_mode & S_ISUID) && !file_owner)
454 m |= S_ISUID;
455 if ((vap->va_mode & S_ISGID) &&
456 !in_group_p((gid_t)ip->i_d.di_gid))
457 m |= S_ISGID;
458#if 0
459 /* Linux allows this, Irix doesn't. */
460 if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
461 m |= S_ISVTX;
462#endif
463 if (m && !capable(CAP_FSETID))
464 vap->va_mode &= ~m;
465 }
466 }
467
468 /*
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.
474 */
475 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
476 /*
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.
481 */
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 :
488 iprojid;
489
490 /*
491 * CAP_CHOWN overrides the following restrictions:
492 *
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.
499 *
500 * XXX: How does restricted_chown affect projid?
501 */
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);
507 goto error_return;
508 }
509 /*
510 * Do a quota reservation only if uid or gid is actually
511 * going to change.
512 */
513 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
514 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
515 ASSERT(tp);
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 */
520 goto error_return;
521 }
522 }
523
524 /*
525 * Truncate file. Must have write permission and not be a directory.
526 */
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);
535 code = 0;
536 goto error_return;
537 }
538
539 if (vp->v_type == VDIR) {
540 code = XFS_ERROR(EISDIR);
541 goto error_return;
542 } else if (vp->v_type != VREG) {
543 code = XFS_ERROR(EINVAL);
544 goto error_return;
545 }
546 /*
547 * Make sure that the dquots are attached to the inode.
548 */
549 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
550 goto error_return;
551 }
552
553 /*
554 * Change file access or modified times.
555 */
556 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
557 if (!file_owner) {
558 if ((flags & ATTR_UTIME) &&
559 !capable(CAP_FOWNER)) {
560 code = XFS_ERROR(EPERM);
561 goto error_return;
562 }
563 }
564 }
565
566 /*
567 * Change extent size or realtime flag.
568 */
569 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
570 /*
571 * Can't change extent size if any extents are allocated.
572 */
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) !=
576 vap->va_extsize) ) {
577 code = XFS_ERROR(EINVAL); /* EFBIG? */
578 goto error_return;
579 }
580
581 /*
582 * Can't set extent size unless the file is marked, or
583 * about to be marked as a realtime file.
584 *
585 * This check will be removed when fixed size extents
586 * with buffered data writes is implemented.
587 *
588 */
589 if ((mask & XFS_AT_EXTSIZE) &&
590 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
591 vap->va_extsize) &&
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);
596 goto error_return;
597 }
598
599 /*
600 * Can't change realtime flag if any extents are allocated.
601 */
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? */
606 goto error_return;
607 }
608 /*
609 * Extent size must be a multiple of the appropriate block
610 * size, if set at all.
611 */
612 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
613 xfs_extlen_t size;
614
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;
620 } else {
621 size = mp->m_sb.sb_blocksize;
622 }
623 if (vap->va_extsize % size) {
624 code = XFS_ERROR(EINVAL);
625 goto error_return;
626 }
627 }
628 /*
629 * If realtime flag is set then must have realtime data.
630 */
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);
637 goto error_return;
638 }
639 }
640
641 /*
642 * Can't modify an immutable/append-only file unless
643 * we have appropriate permission.
644 */
645 if ((mask & XFS_AT_XFLAGS) &&
646 (ip->i_d.di_flags &
647 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
648 (vap->va_xflags &
649 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
650 !capable(CAP_LINUX_IMMUTABLE)) {
651 code = XFS_ERROR(EPERM);
652 goto error_return;
653 }
654 }
655
656 /*
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.
663 */
664 if (mask & XFS_AT_SIZE) {
665 code = 0;
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);
669 if (!code)
670 code = xfs_itruncate_data(ip, vap->va_size);
671 if (code) {
672 ASSERT(tp == NULL);
673 lock_flags &= ~XFS_ILOCK_EXCL;
674 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
675 goto error_return;
676 }
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);
683 if (need_iolock)
684 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
685 return code;
686 }
687 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
688 xfs_ilock(ip, XFS_ILOCK_EXCL);
689 }
690
691 if (tp) {
692 xfs_trans_ijoin(tp, ip, lock_flags);
693 xfs_trans_ihold(tp, ip);
694 }
695
696 /* determine whether mandatory locking mode changes */
697 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
698
699 /*
700 * Truncate file. Must have write permission and not be a directory.
701 */
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)) {
708 /*
709 * signal a sync transaction unless
710 * we're truncating an already unlinked
711 * file on a wsync filesystem
712 */
713 code = xfs_itruncate_finish(&tp, ip,
714 (xfs_fsize_t)vap->va_size,
715 XFS_DATA_FORK,
716 ((ip->i_d.di_nlink != 0 ||
717 !(mp->m_flags & XFS_MOUNT_WSYNC))
718 ? 1 : 0));
719 if (code) {
720 goto abort_return;
721 }
722 }
723 /*
724 * Have to do this even if the file's size doesn't change.
725 */
726 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
727 }
728
729 /*
730 * Change file access modes.
731 */
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;
735
736 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
737 timeflags |= XFS_ICHGTIME_CHG;
738 }
739
740 /*
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.
746 */
747 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
748 /*
749 * CAP_FSETID overrides the following restrictions:
750 *
751 * The set-user-ID and set-group-ID bits of a file will be
752 * cleared upon successful return from chown()
753 */
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);
757 }
758
759 /*
760 * Change the ownerships and register quota modifications
761 * in the transaction.
762 */
763 if (iuid != uid) {
764 if (XFS_IS_UQUOTA_ON(mp)) {
765 ASSERT(mask & XFS_AT_UID);
766 ASSERT(udqp);
767 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
768 &ip->i_udquot, udqp);
769 }
770 ip->i_d.di_uid = uid;
771 }
772 if (igid != gid) {
773 if (XFS_IS_GQUOTA_ON(mp)) {
774 ASSERT(mask & XFS_AT_GID);
775 ASSERT(gdqp);
776 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
777 &ip->i_gdquot, gdqp);
778 }
779 ip->i_d.di_gid = gid;
780 }
781 if (iprojid != projid) {
782 ip->i_d.di_projid = projid;
783 /*
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.
787 */
788 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
789 xfs_bump_ino_vers2(tp, ip);
790 }
791
792 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
793 timeflags |= XFS_ICHGTIME_CHG;
794 }
795
796
797 /*
798 * Change file access or modified times.
799 */
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;
806 }
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;
812 }
813 if (tp && (flags & ATTR_UTIME))
814 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
815 }
816
817 /*
818 * Change XFS-added attributes.
819 */
820 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
821 if (mask & XFS_AT_EXTSIZE) {
822 /*
823 * Converting bytes to fs blocks.
824 */
825 ip->i_d.di_extsize = vap->va_extsize >>
826 mp->m_sb.sb_blocklog;
827 }
828 if (mask & XFS_AT_XFLAGS) {
829 uint di_flags;
830
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_APPEND)
836 di_flags |= XFS_DIFLAG_APPEND;
837 if (vap->va_xflags & XFS_XFLAG_SYNC)
838 di_flags |= XFS_DIFLAG_SYNC;
839 if (vap->va_xflags & XFS_XFLAG_NOATIME)
840 di_flags |= XFS_DIFLAG_NOATIME;
841 if (vap->va_xflags & XFS_XFLAG_NODUMP)
842 di_flags |= XFS_DIFLAG_NODUMP;
843 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
844 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
845 di_flags |= XFS_DIFLAG_RTINHERIT;
846 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
847 di_flags |= XFS_DIFLAG_NOSYMLINKS;
848 } else {
849 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
850 di_flags |= XFS_DIFLAG_REALTIME;
851 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
852 } else {
853 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
854 }
855 }
856 ip->i_d.di_flags = di_flags;
857 }
858 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
859 timeflags |= XFS_ICHGTIME_CHG;
860 }
861
862 /*
863 * Change file inode change time only if XFS_AT_CTIME set
864 * AND we have been called by a DMI function.
865 */
866
867 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
868 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
869 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
870 ip->i_update_core = 1;
871 timeflags &= ~XFS_ICHGTIME_CHG;
872 }
873
874 /*
875 * Send out timestamp changes that need to be set to the
876 * current time. Not done when called by a DMI function.
877 */
878 if (timeflags && !(flags & ATTR_DMI))
879 xfs_ichgtime(ip, timeflags);
880
881 XFS_STATS_INC(xs_ig_attrchg);
882
883 /*
884 * If this is a synchronous mount, make sure that the
885 * transaction goes to disk before returning to the user.
886 * This is slightly sub-optimal in that truncates require
887 * two sync transactions instead of one for wsync filesytems.
888 * One for the truncate and one for the timestamps since we
889 * don't want to change the timestamps unless we're sure the
890 * truncate worked. Truncates are less than 1% of the laddis
891 * mix so this probably isn't worth the trouble to optimize.
892 */
893 code = 0;
894 if (tp) {
895 if (mp->m_flags & XFS_MOUNT_WSYNC)
896 xfs_trans_set_sync(tp);
897
898 code = xfs_trans_commit(tp, commit_flags, NULL);
899 }
900
901 /*
902 * If the (regular) file's mandatory locking mode changed, then
903 * notify the vnode. We do this under the inode lock to prevent
904 * racing calls to vop_vnode_change.
905 */
906 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
907 if (mandlock_before != mandlock_after) {
908 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
909 mandlock_after);
910 }
911
912 xfs_iunlock(ip, lock_flags);
913
914 /*
915 * Release any dquot(s) the inode had kept before chown.
916 */
917 XFS_QM_DQRELE(mp, olddquot1);
918 XFS_QM_DQRELE(mp, olddquot2);
919 XFS_QM_DQRELE(mp, udqp);
920 XFS_QM_DQRELE(mp, gdqp);
921
922 if (code) {
923 return code;
924 }
925
926 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
927 !(flags & ATTR_DMI)) {
928 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
929 NULL, DM_RIGHT_NULL, NULL, NULL,
930 0, 0, AT_DELAY_FLAG(flags));
931 }
932 return 0;
933
934 abort_return:
935 commit_flags |= XFS_TRANS_ABORT;
936 /* FALLTHROUGH */
937 error_return:
938 XFS_QM_DQRELE(mp, udqp);
939 XFS_QM_DQRELE(mp, gdqp);
940 if (tp) {
941 xfs_trans_cancel(tp, commit_flags);
942 }
943 if (lock_flags != 0) {
944 xfs_iunlock(ip, lock_flags);
945 }
946 return code;
947}
948
949
950/*
951 * xfs_access
952 * Null conversion from vnode mode bits to inode mode bits, as in efs.
953 */
954STATIC int
955xfs_access(
956 bhv_desc_t *bdp,
957 int mode,
958 cred_t *credp)
959{
960 xfs_inode_t *ip;
961 int error;
962
963 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
964 (inst_t *)__return_address);
965
966 ip = XFS_BHVTOI(bdp);
967 xfs_ilock(ip, XFS_ILOCK_SHARED);
968 error = xfs_iaccess(ip, mode, credp);
969 xfs_iunlock(ip, XFS_ILOCK_SHARED);
970 return error;
971}
972
973
974/*
975 * xfs_readlink
976 *
977 */
978STATIC int
979xfs_readlink(
980 bhv_desc_t *bdp,
981 uio_t *uiop,
982 int ioflags,
983 cred_t *credp)
984{
985 xfs_inode_t *ip;
986 int count;
987 xfs_off_t offset;
988 int pathlen;
989 vnode_t *vp;
990 int error = 0;
991 xfs_mount_t *mp;
992 int nmaps;
993 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
994 xfs_daddr_t d;
995 int byte_cnt;
996 int n;
997 xfs_buf_t *bp;
998
999 vp = BHV_TO_VNODE(bdp);
1000 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1001
1002 ip = XFS_BHVTOI(bdp);
1003 mp = ip->i_mount;
1004
1005 if (XFS_FORCED_SHUTDOWN(mp))
1006 return XFS_ERROR(EIO);
1007
1008 xfs_ilock(ip, XFS_ILOCK_SHARED);
1009
1010 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1011
1012 offset = uiop->uio_offset;
1013 count = uiop->uio_resid;
1014
1015 if (offset < 0) {
1016 error = XFS_ERROR(EINVAL);
1017 goto error_return;
1018 }
1019 if (count <= 0) {
1020 error = 0;
1021 goto error_return;
1022 }
1023
1024 if (!(ioflags & IO_INVIS)) {
1025 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1026 }
1027
1028 /*
1029 * See if the symlink is stored inline.
1030 */
1031 pathlen = (int)ip->i_d.di_size;
1032
1033 if (ip->i_df.if_flags & XFS_IFINLINE) {
1034 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1035 }
1036 else {
1037 /*
1038 * Symlink not inline. Call bmap to get it in.
1039 */
1040 nmaps = SYMLINK_MAPS;
1041
1042 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1043 0, NULL, 0, mval, &nmaps, NULL);
1044
1045 if (error) {
1046 goto error_return;
1047 }
1048
1049 for (n = 0; n < nmaps; n++) {
1050 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1051 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1052 bp = xfs_buf_read(mp->m_ddev_targp, d,
1053 BTOBB(byte_cnt), 0);
1054 error = XFS_BUF_GETERROR(bp);
1055 if (error) {
1056 xfs_ioerror_alert("xfs_readlink",
1057 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1058 xfs_buf_relse(bp);
1059 goto error_return;
1060 }
1061 if (pathlen < byte_cnt)
1062 byte_cnt = pathlen;
1063 pathlen -= byte_cnt;
1064
1065 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1066 xfs_buf_relse (bp);
1067 }
1068
1069 }
1070
1071
1072error_return:
1073
1074 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1075
1076 return error;
1077}
1078
1079
1080/*
1081 * xfs_fsync
1082 *
1083 * This is called to sync the inode and its data out to disk.
1084 * We need to hold the I/O lock while flushing the data, and
1085 * the inode lock while flushing the inode. The inode lock CANNOT
1086 * be held while flushing the data, so acquire after we're done
1087 * with that.
1088 */
1089STATIC int
1090xfs_fsync(
1091 bhv_desc_t *bdp,
1092 int flag,
1093 cred_t *credp,
1094 xfs_off_t start,
1095 xfs_off_t stop)
1096{
1097 xfs_inode_t *ip;
1098 xfs_trans_t *tp;
1099 int error;
1100
1101 vn_trace_entry(BHV_TO_VNODE(bdp),
1102 __FUNCTION__, (inst_t *)__return_address);
1103
1104 ip = XFS_BHVTOI(bdp);
1105
1106 ASSERT(start >= 0 && stop >= -1);
1107
1108 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1109 return XFS_ERROR(EIO);
1110
1111 /*
1112 * We always need to make sure that the required inode state
1113 * is safe on disk. The vnode might be clean but because
1114 * of committed transactions that haven't hit the disk yet.
1115 * Likewise, there could be unflushed non-transactional
1116 * changes to the inode core that have to go to disk.
1117 *
1118 * The following code depends on one assumption: that
1119 * any transaction that changes an inode logs the core
1120 * because it has to change some field in the inode core
1121 * (typically nextents or nblocks). That assumption
1122 * implies that any transactions against an inode will
1123 * catch any non-transactional updates. If inode-altering
1124 * transactions exist that violate this assumption, the
1125 * code breaks. Right now, it figures that if the involved
1126 * update_* field is clear and the inode is unpinned, the
1127 * inode is clean. Either it's been flushed or it's been
1128 * committed and the commit has hit the disk unpinning the inode.
1129 * (Note that xfs_inode_item_format() called at commit clears
1130 * the update_* fields.)
1131 */
1132 xfs_ilock(ip, XFS_ILOCK_SHARED);
1133
1134 /* If we are flushing data then we care about update_size
1135 * being set, otherwise we care about update_core
1136 */
1137 if ((flag & FSYNC_DATA) ?
1138 (ip->i_update_size == 0) :
1139 (ip->i_update_core == 0)) {
1140 /*
1141 * Timestamps/size haven't changed since last inode
1142 * flush or inode transaction commit. That means
1143 * either nothing got written or a transaction
1144 * committed which caught the updates. If the
1145 * latter happened and the transaction hasn't
1146 * hit the disk yet, the inode will be still
1147 * be pinned. If it is, force the log.
1148 */
1149
1150 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1151
1152 if (xfs_ipincount(ip)) {
1153 xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1154 XFS_LOG_FORCE |
1155 ((flag & FSYNC_WAIT)
1156 ? XFS_LOG_SYNC : 0));
1157 }
1158 error = 0;
1159 } else {
1160 /*
1161 * Kick off a transaction to log the inode
1162 * core to get the updates. Make it
1163 * sync if FSYNC_WAIT is passed in (which
1164 * is done by everybody but specfs). The
1165 * sync transaction will also force the log.
1166 */
1167 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1168 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1169 if ((error = xfs_trans_reserve(tp, 0,
1170 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1171 0, 0, 0))) {
1172 xfs_trans_cancel(tp, 0);
1173 return error;
1174 }
1175 xfs_ilock(ip, XFS_ILOCK_EXCL);
1176
1177 /*
1178 * Note - it's possible that we might have pushed
1179 * ourselves out of the way during trans_reserve
1180 * which would flush the inode. But there's no
1181 * guarantee that the inode buffer has actually
1182 * gone out yet (it's delwri). Plus the buffer
1183 * could be pinned anyway if it's part of an
1184 * inode in another recent transaction. So we
1185 * play it safe and fire off the transaction anyway.
1186 */
1187 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1188 xfs_trans_ihold(tp, ip);
1189 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1190 if (flag & FSYNC_WAIT)
1191 xfs_trans_set_sync(tp);
1192 error = xfs_trans_commit(tp, 0, NULL);
1193
1194 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1195 }
1196 return error;
1197}
1198
1199/*
1200 * This is called by xfs_inactive to free any blocks beyond eof,
1201 * when the link count isn't zero.
1202 */
1203STATIC int
1204xfs_inactive_free_eofblocks(
1205 xfs_mount_t *mp,
1206 xfs_inode_t *ip)
1207{
1208 xfs_trans_t *tp;
1209 int error;
1210 xfs_fileoff_t end_fsb;
1211 xfs_fileoff_t last_fsb;
1212 xfs_filblks_t map_len;
1213 int nimaps;
1214 xfs_bmbt_irec_t imap;
1215
1216 /*
1217 * Figure out if there are any blocks beyond the end
1218 * of the file. If not, then there is nothing to do.
1219 */
1220 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1221 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1222 map_len = last_fsb - end_fsb;
1223 if (map_len <= 0)
1224 return (0);
1225
1226 nimaps = 1;
1227 xfs_ilock(ip, XFS_ILOCK_SHARED);
1228 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1229 NULL, 0, &imap, &nimaps, NULL);
1230 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1231
1232 if (!error && (nimaps != 0) &&
1233 (imap.br_startblock != HOLESTARTBLOCK)) {
1234 /*
1235 * Attach the dquots to the inode up front.
1236 */
1237 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1238 return (error);
1239
1240 /*
1241 * There are blocks after the end of file.
1242 * Free them up now by truncating the file to
1243 * its current size.
1244 */
1245 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1246
1247 /*
1248 * Do the xfs_itruncate_start() call before
1249 * reserving any log space because
1250 * itruncate_start will call into the buffer
1251 * cache and we can't
1252 * do that within a transaction.
1253 */
1254 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1255 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1256 ip->i_d.di_size);
1257
1258 error = xfs_trans_reserve(tp, 0,
1259 XFS_ITRUNCATE_LOG_RES(mp),
1260 0, XFS_TRANS_PERM_LOG_RES,
1261 XFS_ITRUNCATE_LOG_COUNT);
1262 if (error) {
1263 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1264 xfs_trans_cancel(tp, 0);
1265 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1266 return (error);
1267 }
1268
1269 xfs_ilock(ip, XFS_ILOCK_EXCL);
1270 xfs_trans_ijoin(tp, ip,
1271 XFS_IOLOCK_EXCL |
1272 XFS_ILOCK_EXCL);
1273 xfs_trans_ihold(tp, ip);
1274
1275 error = xfs_itruncate_finish(&tp, ip,
1276 ip->i_d.di_size,
1277 XFS_DATA_FORK,
1278 0);
1279 /*
1280 * If we get an error at this point we
1281 * simply don't bother truncating the file.
1282 */
1283 if (error) {
1284 xfs_trans_cancel(tp,
1285 (XFS_TRANS_RELEASE_LOG_RES |
1286 XFS_TRANS_ABORT));
1287 } else {
1288 error = xfs_trans_commit(tp,
1289 XFS_TRANS_RELEASE_LOG_RES,
1290 NULL);
1291 }
1292 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1293 }
1294 return (error);
1295}
1296
1297/*
1298 * Free a symlink that has blocks associated with it.
1299 */
1300STATIC int
1301xfs_inactive_symlink_rmt(
1302 xfs_inode_t *ip,
1303 xfs_trans_t **tpp)
1304{
1305 xfs_buf_t *bp;
1306 int committed;
1307 int done;
1308 int error;
1309 xfs_fsblock_t first_block;
1310 xfs_bmap_free_t free_list;
1311 int i;
1312 xfs_mount_t *mp;
1313 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1314 int nmaps;
1315 xfs_trans_t *ntp;
1316 int size;
1317 xfs_trans_t *tp;
1318
1319 tp = *tpp;
1320 mp = ip->i_mount;
1321 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1322 /*
1323 * We're freeing a symlink that has some
1324 * blocks allocated to it. Free the
1325 * blocks here. We know that we've got
1326 * either 1 or 2 extents and that we can
1327 * free them all in one bunmapi call.
1328 */
1329 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1330 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1331 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1332 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1333 xfs_trans_cancel(tp, 0);
1334 *tpp = NULL;
1335 return error;
1336 }
1337 /*
1338 * Lock the inode, fix the size, and join it to the transaction.
1339 * Hold it so in the normal path, we still have it locked for
1340 * the second transaction. In the error paths we need it
1341 * held so the cancel won't rele it, see below.
1342 */
1343 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1344 size = (int)ip->i_d.di_size;
1345 ip->i_d.di_size = 0;
1346 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1347 xfs_trans_ihold(tp, ip);
1348 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1349 /*
1350 * Find the block(s) so we can inval and unmap them.
1351 */
1352 done = 0;
1353 XFS_BMAP_INIT(&free_list, &first_block);
1354 nmaps = sizeof(mval) / sizeof(mval[0]);
1355 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1356 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1357 &free_list)))
1358 goto error0;
1359 /*
1360 * Invalidate the block(s).
1361 */
1362 for (i = 0; i < nmaps; i++) {
1363 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1364 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1365 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1366 xfs_trans_binval(tp, bp);
1367 }
1368 /*
1369 * Unmap the dead block(s) to the free_list.
1370 */
1371 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1372 &first_block, &free_list, &done)))
1373 goto error1;
1374 ASSERT(done);
1375 /*
1376 * Commit the first transaction. This logs the EFI and the inode.
1377 */
1378 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1379 goto error1;
1380 /*
1381 * The transaction must have been committed, since there were
1382 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1383 * The new tp has the extent freeing and EFDs.
1384 */
1385 ASSERT(committed);
1386 /*
1387 * The first xact was committed, so add the inode to the new one.
1388 * Mark it dirty so it will be logged and moved forward in the log as
1389 * part of every commit.
1390 */
1391 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1392 xfs_trans_ihold(tp, ip);
1393 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1394 /*
1395 * Get a new, empty transaction to return to our caller.
1396 */
1397 ntp = xfs_trans_dup(tp);
1398 /*
1399 * Commit the transaction containing extent freeing and EFD's.
1400 * If we get an error on the commit here or on the reserve below,
1401 * we need to unlock the inode since the new transaction doesn't
1402 * have the inode attached.
1403 */
1404 error = xfs_trans_commit(tp, 0, NULL);
1405 tp = ntp;
1406 if (error) {
1407 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1408 goto error0;
1409 }
1410 /*
1411 * Remove the memory for extent descriptions (just bookkeeping).
1412 */
1413 if (ip->i_df.if_bytes)
1414 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1415 ASSERT(ip->i_df.if_bytes == 0);
1416 /*
1417 * Put an itruncate log reservation in the new transaction
1418 * for our caller.
1419 */
1420 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1421 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1422 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1423 goto error0;
1424 }
1425 /*
1426 * Return with the inode locked but not joined to the transaction.
1427 */
1428 *tpp = tp;
1429 return 0;
1430
1431 error1:
1432 xfs_bmap_cancel(&free_list);
1433 error0:
1434 /*
1435 * Have to come here with the inode locked and either
1436 * (held and in the transaction) or (not in the transaction).
1437 * If the inode isn't held then cancel would iput it, but
1438 * that's wrong since this is inactive and the vnode ref
1439 * count is 0 already.
1440 * Cancel won't do anything to the inode if held, but it still
1441 * needs to be locked until the cancel is done, if it was
1442 * joined to the transaction.
1443 */
1444 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1445 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1446 *tpp = NULL;
1447 return error;
1448
1449}
1450
1451STATIC int
1452xfs_inactive_symlink_local(
1453 xfs_inode_t *ip,
1454 xfs_trans_t **tpp)
1455{
1456 int error;
1457
1458 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1459 /*
1460 * We're freeing a symlink which fit into
1461 * the inode. Just free the memory used
1462 * to hold the old symlink.
1463 */
1464 error = xfs_trans_reserve(*tpp, 0,
1465 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1466 0, XFS_TRANS_PERM_LOG_RES,
1467 XFS_ITRUNCATE_LOG_COUNT);
1468
1469 if (error) {
1470 xfs_trans_cancel(*tpp, 0);
1471 *tpp = NULL;
1472 return (error);
1473 }
1474 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1475
1476 /*
1477 * Zero length symlinks _can_ exist.
1478 */
1479 if (ip->i_df.if_bytes > 0) {
1480 xfs_idata_realloc(ip,
1481 -(ip->i_df.if_bytes),
1482 XFS_DATA_FORK);
1483 ASSERT(ip->i_df.if_bytes == 0);
1484 }
1485 return (0);
1486}
1487
1488/*
1489 *
1490 */
1491STATIC int
1492xfs_inactive_attrs(
1493 xfs_inode_t *ip,
1494 xfs_trans_t **tpp)
1495{
1496 xfs_trans_t *tp;
1497 int error;
1498 xfs_mount_t *mp;
1499
1500 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1501 tp = *tpp;
1502 mp = ip->i_mount;
1503 ASSERT(ip->i_d.di_forkoff != 0);
1504 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1505 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1506
1507 error = xfs_attr_inactive(ip);
1508 if (error) {
1509 *tpp = NULL;
1510 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1511 return (error); /* goto out*/
1512 }
1513
1514 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1515 error = xfs_trans_reserve(tp, 0,
1516 XFS_IFREE_LOG_RES(mp),
1517 0, XFS_TRANS_PERM_LOG_RES,
1518 XFS_INACTIVE_LOG_COUNT);
1519 if (error) {
1520 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1521 xfs_trans_cancel(tp, 0);
1522 *tpp = NULL;
1523 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1524 return (error);
1525 }
1526
1527 xfs_ilock(ip, XFS_ILOCK_EXCL);
1528 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1529 xfs_trans_ihold(tp, ip);
1530 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1531
1532 ASSERT(ip->i_d.di_anextents == 0);
1533
1534 *tpp = tp;
1535 return (0);
1536}
1537
1538STATIC int
1539xfs_release(
1540 bhv_desc_t *bdp)
1541{
1542 xfs_inode_t *ip;
1543 vnode_t *vp;
1544 xfs_mount_t *mp;
1545 int error;
1546
1547 vp = BHV_TO_VNODE(bdp);
1548 ip = XFS_BHVTOI(bdp);
1549
1550 if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1551 return 0;
1552 }
1553
1554 /* If this is a read-only mount, don't do this (would generate I/O) */
1555 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1556 return 0;
1557
1558#ifdef HAVE_REFCACHE
1559 /* If we are in the NFS reference cache then don't do this now */
1560 if (ip->i_refcache)
1561 return 0;
1562#endif
1563
1564 mp = ip->i_mount;
1565
1566 if (ip->i_d.di_nlink != 0) {
1567 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1568 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1569 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1570 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1571 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1572 return (error);
1573 /* Update linux inode block count after free above */
1574 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1575 ip->i_d.di_nblocks + ip->i_delayed_blks);
1576 }
1577 }
1578
1579 return 0;
1580}
1581
1582/*
1583 * xfs_inactive
1584 *
1585 * This is called when the vnode reference count for the vnode
1586 * goes to zero. If the file has been unlinked, then it must
1587 * now be truncated. Also, we clear all of the read-ahead state
1588 * kept for the inode here since the file is now closed.
1589 */
1590STATIC int
1591xfs_inactive(
1592 bhv_desc_t *bdp,
1593 cred_t *credp)
1594{
1595 xfs_inode_t *ip;
1596 vnode_t *vp;
1597 xfs_bmap_free_t free_list;
1598 xfs_fsblock_t first_block;
1599 int committed;
1600 xfs_trans_t *tp;
1601 xfs_mount_t *mp;
1602 int error;
1603 int truncate;
1604
1605 vp = BHV_TO_VNODE(bdp);
1606 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1607
1608 ip = XFS_BHVTOI(bdp);
1609
1610 /*
1611 * If the inode is already free, then there can be nothing
1612 * to clean up here.
1613 */
1614 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1615 ASSERT(ip->i_df.if_real_bytes == 0);
1616 ASSERT(ip->i_df.if_broot_bytes == 0);
1617 return VN_INACTIVE_CACHE;
1618 }
1619
1620 /*
1621 * Only do a truncate if it's a regular file with
1622 * some actual space in it. It's OK to look at the
1623 * inode's fields without the lock because we're the
1624 * only one with a reference to the inode.
1625 */
1626 truncate = ((ip->i_d.di_nlink == 0) &&
1627 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1628 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1629
1630 mp = ip->i_mount;
1631
1632 if (ip->i_d.di_nlink == 0 &&
1633 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1634 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1635 }
1636
1637 error = 0;
1638
1639 /* If this is a read-only mount, don't do this (would generate I/O) */
1640 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1641 goto out;
1642
1643 if (ip->i_d.di_nlink != 0) {
1644 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1645 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1646 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
1647 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1648 (ip->i_delayed_blks != 0))) {
1649 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1650 return (VN_INACTIVE_CACHE);
1651 /* Update linux inode block count after free above */
1652 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1653 ip->i_d.di_nblocks + ip->i_delayed_blks);
1654 }
1655 goto out;
1656 }
1657
1658 ASSERT(ip->i_d.di_nlink == 0);
1659
1660 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1661 return (VN_INACTIVE_CACHE);
1662
1663 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1664 if (truncate) {
1665 /*
1666 * Do the xfs_itruncate_start() call before
1667 * reserving any log space because itruncate_start
1668 * will call into the buffer cache and we can't
1669 * do that within a transaction.
1670 */
1671 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1672
1673 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1674
1675 error = xfs_trans_reserve(tp, 0,
1676 XFS_ITRUNCATE_LOG_RES(mp),
1677 0, XFS_TRANS_PERM_LOG_RES,
1678 XFS_ITRUNCATE_LOG_COUNT);
1679 if (error) {
1680 /* Don't call itruncate_cleanup */
1681 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1682 xfs_trans_cancel(tp, 0);
1683 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1684 return (VN_INACTIVE_CACHE);
1685 }
1686
1687 xfs_ilock(ip, XFS_ILOCK_EXCL);
1688 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1689 xfs_trans_ihold(tp, ip);
1690
1691 /*
1692 * normally, we have to run xfs_itruncate_finish sync.
1693 * But if filesystem is wsync and we're in the inactive
1694 * path, then we know that nlink == 0, and that the
1695 * xaction that made nlink == 0 is permanently committed
1696 * since xfs_remove runs as a synchronous transaction.
1697 */
1698 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1699 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1700
1701 if (error) {
1702 xfs_trans_cancel(tp,
1703 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1704 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1705 return (VN_INACTIVE_CACHE);
1706 }
1707 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1708
1709 /*
1710 * If we get an error while cleaning up a
1711 * symlink we bail out.
1712 */
1713 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1714 xfs_inactive_symlink_rmt(ip, &tp) :
1715 xfs_inactive_symlink_local(ip, &tp);
1716
1717 if (error) {
1718 ASSERT(tp == NULL);
1719 return (VN_INACTIVE_CACHE);
1720 }
1721
1722 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1723 xfs_trans_ihold(tp, ip);
1724 } else {
1725 error = xfs_trans_reserve(tp, 0,
1726 XFS_IFREE_LOG_RES(mp),
1727 0, XFS_TRANS_PERM_LOG_RES,
1728 XFS_INACTIVE_LOG_COUNT);
1729 if (error) {
1730 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1731 xfs_trans_cancel(tp, 0);
1732 return (VN_INACTIVE_CACHE);
1733 }
1734
1735 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1736 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1737 xfs_trans_ihold(tp, ip);
1738 }
1739
1740 /*
1741 * If there are attributes associated with the file
1742 * then blow them away now. The code calls a routine
1743 * that recursively deconstructs the attribute fork.
1744 * We need to just commit the current transaction
1745 * because we can't use it for xfs_attr_inactive().
1746 */
1747 if (ip->i_d.di_anextents > 0) {
1748 error = xfs_inactive_attrs(ip, &tp);
1749 /*
1750 * If we got an error, the transaction is already
1751 * cancelled, and the inode is unlocked. Just get out.
1752 */
1753 if (error)
1754 return (VN_INACTIVE_CACHE);
1755 } else if (ip->i_afp) {
1756 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1757 }
1758
1759 /*
1760 * Free the inode.
1761 */
1762 XFS_BMAP_INIT(&free_list, &first_block);
1763 error = xfs_ifree(tp, ip, &free_list);
1764 if (error) {
1765 /*
1766 * If we fail to free the inode, shut down. The cancel
1767 * might do that, we need to make sure. Otherwise the
1768 * inode might be lost for a long time or forever.
1769 */
1770 if (!XFS_FORCED_SHUTDOWN(mp)) {
1771 cmn_err(CE_NOTE,
1772 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1773 error, mp->m_fsname);
1774 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1775 }
1776 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1777 } else {
1778 /*
1779 * Credit the quota account(s). The inode is gone.
1780 */
1781 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1782
1783 /*
1784 * Just ignore errors at this point. There is
1785 * nothing we can do except to try to keep going.
1786 */
1787 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1788 &committed);
1789 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1790 }
1791 /*
1792 * Release the dquots held by inode, if any.
1793 */
1794 XFS_QM_DQDETACH(mp, ip);
1795
1796 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1797
1798 out:
1799 return VN_INACTIVE_CACHE;
1800}
1801
1802
1803/*
1804 * xfs_lookup
1805 */
1806STATIC int
1807xfs_lookup(
1808 bhv_desc_t *dir_bdp,
1809 vname_t *dentry,
1810 vnode_t **vpp,
1811 int flags,
1812 vnode_t *rdir,
1813 cred_t *credp)
1814{
1815 xfs_inode_t *dp, *ip;
1816 xfs_ino_t e_inum;
1817 int error;
1818 uint lock_mode;
1819 vnode_t *dir_vp;
1820
1821 dir_vp = BHV_TO_VNODE(dir_bdp);
1822 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1823
1824 dp = XFS_BHVTOI(dir_bdp);
1825
1826 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1827 return XFS_ERROR(EIO);
1828
1829 lock_mode = xfs_ilock_map_shared(dp);
1830 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1831 if (!error) {
1832 *vpp = XFS_ITOV(ip);
1833 ITRACE(ip);
1834 }
1835 xfs_iunlock_map_shared(dp, lock_mode);
1836 return error;
1837}
1838
1839
1840/*
1841 * xfs_create (create a new file).
1842 */
1843STATIC int
1844xfs_create(
1845 bhv_desc_t *dir_bdp,
1846 vname_t *dentry,
1847 vattr_t *vap,
1848 vnode_t **vpp,
1849 cred_t *credp)
1850{
1851 char *name = VNAME(dentry);
1852 vnode_t *dir_vp;
1853 xfs_inode_t *dp, *ip;
1854 vnode_t *vp=NULL;
1855 xfs_trans_t *tp;
1856 xfs_mount_t *mp;
1857 xfs_dev_t rdev;
1858 int error;
1859 xfs_bmap_free_t free_list;
1860 xfs_fsblock_t first_block;
1861 boolean_t dp_joined_to_trans;
1862 int dm_event_sent = 0;
1863 uint cancel_flags;
1864 int committed;
1865 xfs_prid_t prid;
1866 struct xfs_dquot *udqp, *gdqp;
1867 uint resblks;
1868 int dm_di_mode;
1869 int namelen;
1870
1871 ASSERT(!*vpp);
1872 dir_vp = BHV_TO_VNODE(dir_bdp);
1873 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1874
1875 dp = XFS_BHVTOI(dir_bdp);
1876 mp = dp->i_mount;
1877
1878 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1879 namelen = VNAMELEN(dentry);
1880
1881 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1882 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1883 dir_vp, DM_RIGHT_NULL, NULL,
1884 DM_RIGHT_NULL, name, NULL,
1885 dm_di_mode, 0, 0);
1886
1887 if (error)
1888 return error;
1889 dm_event_sent = 1;
1890 }
1891
1892 if (XFS_FORCED_SHUTDOWN(mp))
1893 return XFS_ERROR(EIO);
1894
1895 /* Return through std_return after this point. */
1896
1897 udqp = gdqp = NULL;
1898 if (vap->va_mask & XFS_AT_PROJID)
1899 prid = (xfs_prid_t)vap->va_projid;
1900 else
1901 prid = (xfs_prid_t)dfltprid;
1902
1903 /*
1904 * Make sure that we have allocated dquot(s) on disk.
1905 */
1906 error = XFS_QM_DQVOPALLOC(mp, dp,
1907 current_fsuid(credp), current_fsgid(credp),
1908 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1909 if (error)
1910 goto std_return;
1911
1912 ip = NULL;
1913 dp_joined_to_trans = B_FALSE;
1914
1915 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1916 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1917 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1918 /*
1919 * Initially assume that the file does not exist and
1920 * reserve the resources for that case. If that is not
1921 * the case we'll drop the one we have and get a more
1922 * appropriate transaction later.
1923 */
1924 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1925 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1926 if (error == ENOSPC) {
1927 resblks = 0;
1928 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1929 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1930 }
1931 if (error) {
1932 cancel_flags = 0;
1933 dp = NULL;
1934 goto error_return;
1935 }
1936
1937 xfs_ilock(dp, XFS_ILOCK_EXCL);
1938
1939 XFS_BMAP_INIT(&free_list, &first_block);
1940
1941 ASSERT(ip == NULL);
1942
1943 /*
1944 * Reserve disk quota and the inode.
1945 */
1946 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1947 if (error)
1948 goto error_return;
1949
1950 if (resblks == 0 &&
1951 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1952 goto error_return;
1953 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1954 error = xfs_dir_ialloc(&tp, dp,
1955 MAKEIMODE(vap->va_type,vap->va_mode), 1,
1956 rdev, credp, prid, resblks > 0,
1957 &ip, &committed);
1958 if (error) {
1959 if (error == ENOSPC)
1960 goto error_return;
1961 goto abort_return;
1962 }
1963 ITRACE(ip);
1964
1965 /*
1966 * At this point, we've gotten a newly allocated inode.
1967 * It is locked (and joined to the transaction).
1968 */
1969
1970 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1971
1972 /*
1973 * Now we join the directory inode to the transaction.
1974 * We do not do it earlier because xfs_dir_ialloc
1975 * might commit the previous transaction (and release
1976 * all the locks).
1977 */
1978
1979 VN_HOLD(dir_vp);
1980 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1981 dp_joined_to_trans = B_TRUE;
1982
1983 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1984 &first_block, &free_list,
1985 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1986 if (error) {
1987 ASSERT(error != ENOSPC);
1988 goto abort_return;
1989 }
1990 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1991 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1992
1993 /*
1994 * If this is a synchronous mount, make sure that the
1995 * create transaction goes to disk before returning to
1996 * the user.
1997 */
1998 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1999 xfs_trans_set_sync(tp);
2000 }
2001
2002 dp->i_gen++;
2003
2004 /*
2005 * Attach the dquot(s) to the inodes and modify them incore.
2006 * These ids of the inode couldn't have changed since the new
2007 * inode has been locked ever since it was created.
2008 */
2009 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2010
2011 /*
2012 * xfs_trans_commit normally decrements the vnode ref count
2013 * when it unlocks the inode. Since we want to return the
2014 * vnode to the caller, we bump the vnode ref count now.
2015 */
2016 IHOLD(ip);
2017 vp = XFS_ITOV(ip);
2018
2019 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2020 if (error) {
2021 xfs_bmap_cancel(&free_list);
2022 goto abort_rele;
2023 }
2024
2025 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2026 if (error) {
2027 IRELE(ip);
2028 tp = NULL;
2029 goto error_return;
2030 }
2031
2032 XFS_QM_DQRELE(mp, udqp);
2033 XFS_QM_DQRELE(mp, gdqp);
2034
2035 /*
2036 * Propogate the fact that the vnode changed after the
2037 * xfs_inode locks have been released.
2038 */
2039 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2040
2041 *vpp = vp;
2042
2043 /* Fallthrough to std_return with error = 0 */
2044
2045std_return:
2046 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2047 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2048 DM_EVENT_POSTCREATE)) {
2049 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2050 dir_vp, DM_RIGHT_NULL,
2051 *vpp ? vp:NULL,
2052 DM_RIGHT_NULL, name, NULL,
2053 dm_di_mode, error, 0);
2054 }
2055 return error;
2056
2057 abort_return:
2058 cancel_flags |= XFS_TRANS_ABORT;
2059 /* FALLTHROUGH */
2060 error_return:
2061
2062 if (tp != NULL)
2063 xfs_trans_cancel(tp, cancel_flags);
2064
2065 if (!dp_joined_to_trans && (dp != NULL))
2066 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2067 XFS_QM_DQRELE(mp, udqp);
2068 XFS_QM_DQRELE(mp, gdqp);
2069
2070 goto std_return;
2071
2072 abort_rele:
2073 /*
2074 * Wait until after the current transaction is aborted to
2075 * release the inode. This prevents recursive transactions
2076 * and deadlocks from xfs_inactive.
2077 */
2078 cancel_flags |= XFS_TRANS_ABORT;
2079 xfs_trans_cancel(tp, cancel_flags);
2080 IRELE(ip);
2081
2082 XFS_QM_DQRELE(mp, udqp);
2083 XFS_QM_DQRELE(mp, gdqp);
2084
2085 goto std_return;
2086}
2087
2088#ifdef DEBUG
2089/*
2090 * Some counters to see if (and how often) we are hitting some deadlock
2091 * prevention code paths.
2092 */
2093
2094int xfs_rm_locks;
2095int xfs_rm_lock_delays;
2096int xfs_rm_attempts;
2097#endif
2098
2099/*
2100 * The following routine will lock the inodes associated with the
2101 * directory and the named entry in the directory. The locks are
2102 * acquired in increasing inode number.
2103 *
2104 * If the entry is "..", then only the directory is locked. The
2105 * vnode ref count will still include that from the .. entry in
2106 * this case.
2107 *
2108 * There is a deadlock we need to worry about. If the locked directory is
2109 * in the AIL, it might be blocking up the log. The next inode we lock
2110 * could be already locked by another thread waiting for log space (e.g
2111 * a permanent log reservation with a long running transaction (see
2112 * xfs_itruncate_finish)). To solve this, we must check if the directory
2113 * is in the ail and use lock_nowait. If we can't lock, we need to
2114 * drop the inode lock on the directory and try again. xfs_iunlock will
2115 * potentially push the tail if we were holding up the log.
2116 */
2117STATIC int
2118xfs_lock_dir_and_entry(
2119 xfs_inode_t *dp,
2120 vname_t *dentry,
2121 xfs_inode_t *ip) /* inode of entry 'name' */
2122{
2123 int attempts;
2124 xfs_ino_t e_inum;
2125 xfs_inode_t *ips[2];
2126 xfs_log_item_t *lp;
2127
2128#ifdef DEBUG
2129 xfs_rm_locks++;
2130#endif
2131 attempts = 0;
2132
2133again:
2134 xfs_ilock(dp, XFS_ILOCK_EXCL);
2135
2136 e_inum = ip->i_ino;
2137
2138 ITRACE(ip);
2139
2140 /*
2141 * We want to lock in increasing inum. Since we've already
2142 * acquired the lock on the directory, we may need to release
2143 * if if the inum of the entry turns out to be less.
2144 */
2145 if (e_inum > dp->i_ino) {
2146 /*
2147 * We are already in the right order, so just
2148 * lock on the inode of the entry.
2149 * We need to use nowait if dp is in the AIL.
2150 */
2151
2152 lp = (xfs_log_item_t *)dp->i_itemp;
2153 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2154 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2155 attempts++;
2156#ifdef DEBUG
2157 xfs_rm_attempts++;
2158#endif
2159
2160 /*
2161 * Unlock dp and try again.
2162 * xfs_iunlock will try to push the tail
2163 * if the inode is in the AIL.
2164 */
2165
2166 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2167
2168 if ((attempts % 5) == 0) {
2169 delay(1); /* Don't just spin the CPU */
2170#ifdef DEBUG
2171 xfs_rm_lock_delays++;
2172#endif
2173 }
2174 goto again;
2175 }
2176 } else {
2177 xfs_ilock(ip, XFS_ILOCK_EXCL);
2178 }
2179 } else if (e_inum < dp->i_ino) {
2180 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2181
2182 ips[0] = ip;
2183 ips[1] = dp;
2184 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2185 }
2186 /* else e_inum == dp->i_ino */
2187 /* This can happen if we're asked to lock /x/..
2188 * the entry is "..", which is also the parent directory.
2189 */
2190
2191 return 0;
2192}
2193
2194#ifdef DEBUG
2195int xfs_locked_n;
2196int xfs_small_retries;
2197int xfs_middle_retries;
2198int xfs_lots_retries;
2199int xfs_lock_delays;
2200#endif
2201
2202/*
2203 * The following routine will lock n inodes in exclusive mode.
2204 * We assume the caller calls us with the inodes in i_ino order.
2205 *
2206 * We need to detect deadlock where an inode that we lock
2207 * is in the AIL and we start waiting for another inode that is locked
2208 * by a thread in a long running transaction (such as truncate). This can
2209 * result in deadlock since the long running trans might need to wait
2210 * for the inode we just locked in order to push the tail and free space
2211 * in the log.
2212 */
2213void
2214xfs_lock_inodes(
2215 xfs_inode_t **ips,
2216 int inodes,
2217 int first_locked,
2218 uint lock_mode)
2219{
2220 int attempts = 0, i, j, try_lock;
2221 xfs_log_item_t *lp;
2222
2223 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2224
2225 if (first_locked) {
2226 try_lock = 1;
2227 i = 1;
2228 } else {
2229 try_lock = 0;
2230 i = 0;
2231 }
2232
2233again:
2234 for (; i < inodes; i++) {
2235 ASSERT(ips[i]);
2236
2237 if (i && (ips[i] == ips[i-1])) /* Already locked */
2238 continue;
2239
2240 /*
2241 * If try_lock is not set yet, make sure all locked inodes
2242 * are not in the AIL.
2243 * If any are, set try_lock to be used later.
2244 */
2245
2246 if (!try_lock) {
2247 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2248 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2249 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2250 try_lock++;
2251 }
2252 }
2253 }
2254
2255 /*
2256 * If any of the previous locks we have locked is in the AIL,
2257 * we must TRY to get the second and subsequent locks. If
2258 * we can't get any, we must release all we have
2259 * and try again.
2260 */
2261
2262 if (try_lock) {
2263 /* try_lock must be 0 if i is 0. */
2264 /*
2265 * try_lock means we have an inode locked
2266 * that is in the AIL.
2267 */
2268 ASSERT(i != 0);
2269 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2270 attempts++;
2271
2272 /*
2273 * Unlock all previous guys and try again.
2274 * xfs_iunlock will try to push the tail
2275 * if the inode is in the AIL.
2276 */
2277
2278 for(j = i - 1; j >= 0; j--) {
2279
2280 /*
2281 * Check to see if we've already
2282 * unlocked this one.
2283 * Not the first one going back,
2284 * and the inode ptr is the same.
2285 */
2286 if ((j != (i - 1)) && ips[j] ==
2287 ips[j+1])
2288 continue;
2289
2290 xfs_iunlock(ips[j], lock_mode);
2291 }
2292
2293 if ((attempts % 5) == 0) {
2294 delay(1); /* Don't just spin the CPU */
2295#ifdef DEBUG
2296 xfs_lock_delays++;
2297#endif
2298 }
2299 i = 0;
2300 try_lock = 0;
2301 goto again;
2302 }
2303 } else {
2304 xfs_ilock(ips[i], lock_mode);
2305 }
2306 }
2307
2308#ifdef DEBUG
2309 if (attempts) {
2310 if (attempts < 5) xfs_small_retries++;
2311 else if (attempts < 100) xfs_middle_retries++;
2312 else xfs_lots_retries++;
2313 } else {
2314 xfs_locked_n++;
2315 }
2316#endif
2317}
2318
2319#ifdef DEBUG
2320#define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2321int remove_which_error_return = 0;
2322#else /* ! DEBUG */
2323#define REMOVE_DEBUG_TRACE(x)
2324#endif /* ! DEBUG */
2325
2326
2327/*
2328 * xfs_remove
2329 *
2330 */
2331STATIC int
2332xfs_remove(
2333 bhv_desc_t *dir_bdp,
2334 vname_t *dentry,
2335 cred_t *credp)
2336{
2337 vnode_t *dir_vp;
2338 char *name = VNAME(dentry);
2339 xfs_inode_t *dp, *ip;
2340 xfs_trans_t *tp = NULL;
2341 xfs_mount_t *mp;
2342 int error = 0;
2343 xfs_bmap_free_t free_list;
2344 xfs_fsblock_t first_block;
2345 int cancel_flags;
2346 int committed;
2347 int dm_di_mode = 0;
2348 int link_zero;
2349 uint resblks;
2350 int namelen;
2351
2352 dir_vp = BHV_TO_VNODE(dir_bdp);
2353 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2354
2355 dp = XFS_BHVTOI(dir_bdp);
2356 mp = dp->i_mount;
2357
2358 if (XFS_FORCED_SHUTDOWN(mp))
2359 return XFS_ERROR(EIO);
2360
2361 namelen = VNAMELEN(dentry);
2362
2363 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2364 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2365 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2366 name, NULL, 0, 0, 0);
2367 if (error)
2368 return error;
2369 }
2370
2371 /* From this point on, return through std_return */
2372 ip = NULL;
2373
2374 /*
2375 * We need to get a reference to ip before we get our log
2376 * reservation. The reason for this is that we cannot call
2377 * xfs_iget for an inode for which we do not have a reference
2378 * once we've acquired a log reservation. This is because the
2379 * inode we are trying to get might be in xfs_inactive going
2380 * for a log reservation. Since we'll have to wait for the
2381 * inactive code to complete before returning from xfs_iget,
2382 * we need to make sure that we don't have log space reserved
2383 * when we call xfs_iget. Instead we get an unlocked referece
2384 * to the inode before getting our log reservation.
2385 */
2386 error = xfs_get_dir_entry(dentry, &ip);
2387 if (error) {
2388 REMOVE_DEBUG_TRACE(__LINE__);
2389 goto std_return;
2390 }
2391
2392 dm_di_mode = ip->i_d.di_mode;
2393
2394 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2395
2396 ITRACE(ip);
2397
2398 error = XFS_QM_DQATTACH(mp, dp, 0);
2399 if (!error && dp != ip)
2400 error = XFS_QM_DQATTACH(mp, ip, 0);
2401 if (error) {
2402 REMOVE_DEBUG_TRACE(__LINE__);
2403 IRELE(ip);
2404 goto std_return;
2405 }
2406
2407 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2408 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2409 /*
2410 * We try to get the real space reservation first,
2411 * allowing for directory btree deletion(s) implying
2412 * possible bmap insert(s). If we can't get the space
2413 * reservation then we use 0 instead, and avoid the bmap
2414 * btree insert(s) in the directory code by, if the bmap
2415 * insert tries to happen, instead trimming the LAST
2416 * block from the directory.
2417 */
2418 resblks = XFS_REMOVE_SPACE_RES(mp);
2419 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2420 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2421 if (error == ENOSPC) {
2422 resblks = 0;
2423 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2424 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2425 }
2426 if (error) {
2427 ASSERT(error != ENOSPC);
2428 REMOVE_DEBUG_TRACE(__LINE__);
2429 xfs_trans_cancel(tp, 0);
2430 IRELE(ip);
2431 return error;
2432 }
2433
2434 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2435 if (error) {
2436 REMOVE_DEBUG_TRACE(__LINE__);
2437 xfs_trans_cancel(tp, cancel_flags);
2438 IRELE(ip);
2439 goto std_return;
2440 }
2441
2442 /*
2443 * At this point, we've gotten both the directory and the entry
2444 * inodes locked.
2445 */
2446 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2447 if (dp != ip) {
2448 /*
2449 * Increment vnode ref count only in this case since
2450 * there's an extra vnode reference in the case where
2451 * dp == ip.
2452 */
2453 IHOLD(dp);
2454 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2455 }
2456
2457 /*
2458 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2459 */
2460 XFS_BMAP_INIT(&free_list, &first_block);
2461 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2462 &first_block, &free_list, 0);
2463 if (error) {
2464 ASSERT(error != ENOENT);
2465 REMOVE_DEBUG_TRACE(__LINE__);
2466 goto error1;
2467 }
2468 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2469
2470 dp->i_gen++;
2471 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2472
2473 error = xfs_droplink(tp, ip);
2474 if (error) {
2475 REMOVE_DEBUG_TRACE(__LINE__);
2476 goto error1;
2477 }
2478
2479 /* Determine if this is the last link while
2480 * we are in the transaction.
2481 */
2482 link_zero = (ip)->i_d.di_nlink==0;
2483
2484 /*
2485 * Take an extra ref on the inode so that it doesn't
2486 * go to xfs_inactive() from within the commit.
2487 */
2488 IHOLD(ip);
2489
2490 /*
2491 * If this is a synchronous mount, make sure that the
2492 * remove transaction goes to disk before returning to
2493 * the user.
2494 */
2495 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2496 xfs_trans_set_sync(tp);
2497 }
2498
2499 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2500 if (error) {
2501 REMOVE_DEBUG_TRACE(__LINE__);
2502 goto error_rele;
2503 }
2504
2505 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2506 if (error) {
2507 IRELE(ip);
2508 goto std_return;
2509 }
2510
2511 /*
2512 * Before we drop our extra reference to the inode, purge it
2513 * from the refcache if it is there. By waiting until afterwards
2514 * to do the IRELE, we ensure that we won't go inactive in the
2515 * xfs_refcache_purge_ip routine (although that would be OK).
2516 */
2517 xfs_refcache_purge_ip(ip);
2518
2519 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2520
2521 /*
2522 * Let interposed file systems know about removed links.
2523 */
2524 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2525
2526 IRELE(ip);
2527
2528/* Fall through to std_return with error = 0 */
2529 std_return:
2530 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2531 DM_EVENT_POSTREMOVE)) {
2532 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2533 dir_vp, DM_RIGHT_NULL,
2534 NULL, DM_RIGHT_NULL,
2535 name, NULL, dm_di_mode, error, 0);
2536 }
2537 return error;
2538
2539 error1:
2540 xfs_bmap_cancel(&free_list);
2541 cancel_flags |= XFS_TRANS_ABORT;
2542 xfs_trans_cancel(tp, cancel_flags);
2543 goto std_return;
2544
2545 error_rele:
2546 /*
2547 * In this case make sure to not release the inode until after
2548 * the current transaction is aborted. Releasing it beforehand
2549 * can cause us to go to xfs_inactive and start a recursive
2550 * transaction which can easily deadlock with the current one.
2551 */
2552 xfs_bmap_cancel(&free_list);
2553 cancel_flags |= XFS_TRANS_ABORT;
2554 xfs_trans_cancel(tp, cancel_flags);
2555
2556 /*
2557 * Before we drop our extra reference to the inode, purge it
2558 * from the refcache if it is there. By waiting until afterwards
2559 * to do the IRELE, we ensure that we won't go inactive in the
2560 * xfs_refcache_purge_ip routine (although that would be OK).
2561 */
2562 xfs_refcache_purge_ip(ip);
2563
2564 IRELE(ip);
2565
2566 goto std_return;
2567}
2568
2569
2570/*
2571 * xfs_link
2572 *
2573 */
2574STATIC int
2575xfs_link(
2576 bhv_desc_t *target_dir_bdp,
2577 vnode_t *src_vp,
2578 vname_t *dentry,
2579 cred_t *credp)
2580{
2581 xfs_inode_t *tdp, *sip;
2582 xfs_trans_t *tp;
2583 xfs_mount_t *mp;
2584 xfs_inode_t *ips[2];
2585 int error;
2586 xfs_bmap_free_t free_list;
2587 xfs_fsblock_t first_block;
2588 int cancel_flags;
2589 int committed;
2590 vnode_t *target_dir_vp;
2591 bhv_desc_t *src_bdp;
2592 int resblks;
2593 char *target_name = VNAME(dentry);
2594 int target_namelen;
2595
2596 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2597 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2598 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2599
2600 target_namelen = VNAMELEN(dentry);
2601 if (src_vp->v_type == VDIR)
2602 return XFS_ERROR(EPERM);
2603
2604 /*
2605 * For now, manually find the XFS behavior descriptor for
2606 * the source vnode. If it doesn't exist then something
2607 * is wrong and we should just return an error.
2608 * Eventually we need to figure out how link is going to
2609 * work in the face of stacked vnodes.
2610 */
2611 src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2612 if (src_bdp == NULL) {
2613 return XFS_ERROR(EXDEV);
2614 }
2615 sip = XFS_BHVTOI(src_bdp);
2616 tdp = XFS_BHVTOI(target_dir_bdp);
2617 mp = tdp->i_mount;
2618 if (XFS_FORCED_SHUTDOWN(mp))
2619 return XFS_ERROR(EIO);
2620
2621 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2622 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2623 target_dir_vp, DM_RIGHT_NULL,
2624 src_vp, DM_RIGHT_NULL,
2625 target_name, NULL, 0, 0, 0);
2626 if (error)
2627 return error;
2628 }
2629
2630 /* Return through std_return after this point. */
2631
2632 error = XFS_QM_DQATTACH(mp, sip, 0);
2633 if (!error && sip != tdp)
2634 error = XFS_QM_DQATTACH(mp, tdp, 0);
2635 if (error)
2636 goto std_return;
2637
2638 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2639 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2640 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2641 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2642 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2643 if (error == ENOSPC) {
2644 resblks = 0;
2645 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2646 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2647 }
2648 if (error) {
2649 cancel_flags = 0;
2650 goto error_return;
2651 }
2652
2653 if (sip->i_ino < tdp->i_ino) {
2654 ips[0] = sip;
2655 ips[1] = tdp;
2656 } else {
2657 ips[0] = tdp;
2658 ips[1] = sip;
2659 }
2660
2661 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2662
2663 /*
2664 * Increment vnode ref counts since xfs_trans_commit &
2665 * xfs_trans_cancel will both unlock the inodes and
2666 * decrement the associated ref counts.
2667 */
2668 VN_HOLD(src_vp);
2669 VN_HOLD(target_dir_vp);
2670 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2671 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2672
2673 /*
2674 * If the source has too many links, we can't make any more to it.
2675 */
2676 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2677 error = XFS_ERROR(EMLINK);
2678 goto error_return;
2679 }
2680
2681 if (resblks == 0 &&
2682 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2683 target_namelen)))
2684 goto error_return;
2685
2686 XFS_BMAP_INIT(&free_list, &first_block);
2687
2688 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2689 sip->i_ino, &first_block, &free_list,
2690 resblks);
2691 if (error)
2692 goto abort_return;
2693 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2694 tdp->i_gen++;
2695 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2696
2697 error = xfs_bumplink(tp, sip);
2698 if (error) {
2699 goto abort_return;
2700 }
2701
2702 /*
2703 * If this is a synchronous mount, make sure that the
2704 * link transaction goes to disk before returning to
2705 * the user.
2706 */
2707 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2708 xfs_trans_set_sync(tp);
2709 }
2710
2711 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2712 if (error) {
2713 xfs_bmap_cancel(&free_list);
2714 goto abort_return;
2715 }
2716
2717 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2718 if (error) {
2719 goto std_return;
2720 }
2721
2722 /* Fall through to std_return with error = 0. */
2723std_return:
2724 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2725 DM_EVENT_POSTLINK)) {
2726 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2727 target_dir_vp, DM_RIGHT_NULL,
2728 src_vp, DM_RIGHT_NULL,
2729 target_name, NULL, 0, error, 0);
2730 }
2731 return error;
2732
2733 abort_return:
2734 cancel_flags |= XFS_TRANS_ABORT;
2735 /* FALLTHROUGH */
2736 error_return:
2737 xfs_trans_cancel(tp, cancel_flags);
2738
2739 goto std_return;
2740}
2741/*
2742 * xfs_mkdir
2743 *
2744 */
2745STATIC int
2746xfs_mkdir(
2747 bhv_desc_t *dir_bdp,
2748 vname_t *dentry,
2749 vattr_t *vap,
2750 vnode_t **vpp,
2751 cred_t *credp)
2752{
2753 char *dir_name = VNAME(dentry);
2754 xfs_inode_t *dp;
2755 xfs_inode_t *cdp; /* inode of created dir */
2756 vnode_t *cvp; /* vnode of created dir */
2757 xfs_trans_t *tp;
2758 xfs_mount_t *mp;
2759 int cancel_flags;
2760 int error;
2761 int committed;
2762 xfs_bmap_free_t free_list;
2763 xfs_fsblock_t first_block;
2764 vnode_t *dir_vp;
2765 boolean_t dp_joined_to_trans;
2766 boolean_t created = B_FALSE;
2767 int dm_event_sent = 0;
2768 xfs_prid_t prid;
2769 struct xfs_dquot *udqp, *gdqp;
2770 uint resblks;
2771 int dm_di_mode;
2772 int dir_namelen;
2773
2774 dir_vp = BHV_TO_VNODE(dir_bdp);
2775 dp = XFS_BHVTOI(dir_bdp);
2776 mp = dp->i_mount;
2777
2778 if (XFS_FORCED_SHUTDOWN(mp))
2779 return XFS_ERROR(EIO);
2780
2781 dir_namelen = VNAMELEN(dentry);
2782
2783 tp = NULL;
2784 dp_joined_to_trans = B_FALSE;
2785 dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2786
2787 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2788 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2789 dir_vp, DM_RIGHT_NULL, NULL,
2790 DM_RIGHT_NULL, dir_name, NULL,
2791 dm_di_mode, 0, 0);
2792 if (error)
2793 return error;
2794 dm_event_sent = 1;
2795 }
2796
2797 /* Return through std_return after this point. */
2798
2799 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2800
2801 mp = dp->i_mount;
2802 udqp = gdqp = NULL;
2803 if (vap->va_mask & XFS_AT_PROJID)
2804 prid = (xfs_prid_t)vap->va_projid;
2805 else
2806 prid = (xfs_prid_t)dfltprid;
2807
2808 /*
2809 * Make sure that we have allocated dquot(s) on disk.
2810 */
2811 error = XFS_QM_DQVOPALLOC(mp, dp,
2812 current_fsuid(credp), current_fsgid(credp),
2813 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2814 if (error)
2815 goto std_return;
2816
2817 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2818 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2819 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2820 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2821 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2822 if (error == ENOSPC) {
2823 resblks = 0;
2824 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2825 XFS_TRANS_PERM_LOG_RES,
2826 XFS_MKDIR_LOG_COUNT);
2827 }
2828 if (error) {
2829 cancel_flags = 0;
2830 dp = NULL;
2831 goto error_return;
2832 }
2833
2834 xfs_ilock(dp, XFS_ILOCK_EXCL);
2835
2836 /*
2837 * Check for directory link count overflow.
2838 */
2839 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2840 error = XFS_ERROR(EMLINK);
2841 goto error_return;
2842 }
2843
2844 /*
2845 * Reserve disk quota and the inode.
2846 */
2847 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2848 if (error)
2849 goto error_return;
2850
2851 if (resblks == 0 &&
2852 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2853 goto error_return;
2854 /*
2855 * create the directory inode.
2856 */
2857 error = xfs_dir_ialloc(&tp, dp,
2858 MAKEIMODE(vap->va_type,vap->va_mode), 2,
2859 0, credp, prid, resblks > 0,
2860 &cdp, NULL);
2861 if (error) {
2862 if (error == ENOSPC)
2863 goto error_return;
2864 goto abort_return;
2865 }
2866 ITRACE(cdp);
2867
2868 /*
2869 * Now we add the directory inode to the transaction.
2870 * We waited until now since xfs_dir_ialloc might start
2871 * a new transaction. Had we joined the transaction
2872 * earlier, the locks might have gotten released.
2873 */
2874 VN_HOLD(dir_vp);
2875 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2876 dp_joined_to_trans = B_TRUE;
2877
2878 XFS_BMAP_INIT(&free_list, &first_block);
2879
2880 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2881 cdp->i_ino, &first_block, &free_list,
2882 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2883 if (error) {
2884 ASSERT(error != ENOSPC);
2885 goto error1;
2886 }
2887 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2888
2889 /*
2890 * Bump the in memory version number of the parent directory
2891 * so that other processes accessing it will recognize that
2892 * the directory has changed.
2893 */
2894 dp->i_gen++;
2895
2896 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2897 if (error) {
2898 goto error2;
2899 }
2900
2901 cdp->i_gen = 1;
2902 error = xfs_bumplink(tp, dp);
2903 if (error) {
2904 goto error2;
2905 }
2906
2907 cvp = XFS_ITOV(cdp);
2908
2909 created = B_TRUE;
2910
2911 *vpp = cvp;
2912 IHOLD(cdp);
2913
2914 /*
2915 * Attach the dquots to the new inode and modify the icount incore.
2916 */
2917 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2918
2919 /*
2920 * If this is a synchronous mount, make sure that the
2921 * mkdir transaction goes to disk before returning to
2922 * the user.
2923 */
2924 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2925 xfs_trans_set_sync(tp);
2926 }
2927
2928 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2929 if (error) {
2930 IRELE(cdp);
2931 goto error2;
2932 }
2933
2934 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2935 XFS_QM_DQRELE(mp, udqp);
2936 XFS_QM_DQRELE(mp, gdqp);
2937 if (error) {
2938 IRELE(cdp);
2939 }
2940
2941 /* Fall through to std_return with error = 0 or errno from
2942 * xfs_trans_commit. */
2943
2944std_return:
2945 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2946 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2947 DM_EVENT_POSTCREATE)) {
2948 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2949 dir_vp, DM_RIGHT_NULL,
2950 created ? XFS_ITOV(cdp):NULL,
2951 DM_RIGHT_NULL,
2952 dir_name, NULL,
2953 dm_di_mode, error, 0);
2954 }
2955 return error;
2956
2957 error2:
2958 error1:
2959 xfs_bmap_cancel(&free_list);
2960 abort_return:
2961 cancel_flags |= XFS_TRANS_ABORT;
2962 error_return:
2963 xfs_trans_cancel(tp, cancel_flags);
2964 XFS_QM_DQRELE(mp, udqp);
2965 XFS_QM_DQRELE(mp, gdqp);
2966
2967 if (!dp_joined_to_trans && (dp != NULL)) {
2968 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2969 }
2970
2971 goto std_return;
2972}
2973
2974
2975/*
2976 * xfs_rmdir
2977 *
2978 */
2979STATIC int
2980xfs_rmdir(
2981 bhv_desc_t *dir_bdp,
2982 vname_t *dentry,
2983 cred_t *credp)
2984{
2985 char *name = VNAME(dentry);
2986 xfs_inode_t *dp;
2987 xfs_inode_t *cdp; /* child directory */
2988 xfs_trans_t *tp;
2989 xfs_mount_t *mp;
2990 int error;
2991 xfs_bmap_free_t free_list;
2992 xfs_fsblock_t first_block;
2993 int cancel_flags;
2994 int committed;
2995 vnode_t *dir_vp;
2996 int dm_di_mode = 0;
2997 int last_cdp_link;
2998 int namelen;
2999 uint resblks;
3000
3001 dir_vp = BHV_TO_VNODE(dir_bdp);
3002 dp = XFS_BHVTOI(dir_bdp);
3003 mp = dp->i_mount;
3004
3005 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3006
3007 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3008 return XFS_ERROR(EIO);
3009 namelen = VNAMELEN(dentry);
3010
3011 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3012 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3013 dir_vp, DM_RIGHT_NULL,
3014 NULL, DM_RIGHT_NULL,
3015 name, NULL, 0, 0, 0);
3016 if (error)
3017 return XFS_ERROR(error);
3018 }
3019
3020 /* Return through std_return after this point. */
3021
3022 cdp = NULL;
3023
3024 /*
3025 * We need to get a reference to cdp before we get our log
3026 * reservation. The reason for this is that we cannot call
3027 * xfs_iget for an inode for which we do not have a reference
3028 * once we've acquired a log reservation. This is because the
3029 * inode we are trying to get might be in xfs_inactive going
3030 * for a log reservation. Since we'll have to wait for the
3031 * inactive code to complete before returning from xfs_iget,
3032 * we need to make sure that we don't have log space reserved
3033 * when we call xfs_iget. Instead we get an unlocked referece
3034 * to the inode before getting our log reservation.
3035 */
3036 error = xfs_get_dir_entry(dentry, &cdp);
3037 if (error) {
3038 REMOVE_DEBUG_TRACE(__LINE__);
3039 goto std_return;
3040 }
3041 mp = dp->i_mount;
3042 dm_di_mode = cdp->i_d.di_mode;
3043
3044 /*
3045 * Get the dquots for the inodes.
3046 */
3047 error = XFS_QM_DQATTACH(mp, dp, 0);
3048 if (!error && dp != cdp)
3049 error = XFS_QM_DQATTACH(mp, cdp, 0);
3050 if (error) {
3051 IRELE(cdp);
3052 REMOVE_DEBUG_TRACE(__LINE__);
3053 goto std_return;
3054 }
3055
3056 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3057 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3058 /*
3059 * We try to get the real space reservation first,
3060 * allowing for directory btree deletion(s) implying
3061 * possible bmap insert(s). If we can't get the space
3062 * reservation then we use 0 instead, and avoid the bmap
3063 * btree insert(s) in the directory code by, if the bmap
3064 * insert tries to happen, instead trimming the LAST
3065 * block from the directory.
3066 */
3067 resblks = XFS_REMOVE_SPACE_RES(mp);
3068 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3069 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3070 if (error == ENOSPC) {
3071 resblks = 0;
3072 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3073 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3074 }
3075 if (error) {
3076 ASSERT(error != ENOSPC);
3077 cancel_flags = 0;
3078 IRELE(cdp);
3079 goto error_return;
3080 }
3081 XFS_BMAP_INIT(&free_list, &first_block);
3082
3083 /*
3084 * Now lock the child directory inode and the parent directory
3085 * inode in the proper order. This will take care of validating
3086 * that the directory entry for the child directory inode has
3087 * not changed while we were obtaining a log reservation.
3088 */
3089 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3090 if (error) {
3091 xfs_trans_cancel(tp, cancel_flags);
3092 IRELE(cdp);
3093 goto std_return;
3094 }
3095
3096 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3097 if (dp != cdp) {
3098 /*
3099 * Only increment the parent directory vnode count if
3100 * we didn't bump it in looking up cdp. The only time
3101 * we don't bump it is when we're looking up ".".
3102 */
3103 VN_HOLD(dir_vp);
3104 }
3105
3106 ITRACE(cdp);
3107 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3108
3109 ASSERT(cdp->i_d.di_nlink >= 2);
3110 if (cdp->i_d.di_nlink != 2) {
3111 error = XFS_ERROR(ENOTEMPTY);
3112 goto error_return;
3113 }
3114 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3115 error = XFS_ERROR(ENOTEMPTY);
3116 goto error_return;
3117 }
3118
3119 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3120 &first_block, &free_list, resblks);
3121 if (error) {
3122 goto error1;
3123 }
3124
3125 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3126
3127 /*
3128 * Bump the in memory generation count on the parent
3129 * directory so that other can know that it has changed.
3130 */
3131 dp->i_gen++;
3132
3133 /*
3134 * Drop the link from cdp's "..".
3135 */
3136 error = xfs_droplink(tp, dp);
3137 if (error) {
3138 goto error1;
3139 }
3140
3141 /*
3142 * Drop the link from dp to cdp.
3143 */
3144 error = xfs_droplink(tp, cdp);
3145 if (error) {
3146 goto error1;
3147 }
3148
3149 /*
3150 * Drop the "." link from cdp to self.
3151 */
3152 error = xfs_droplink(tp, cdp);
3153 if (error) {
3154 goto error1;
3155 }
3156
3157 /* Determine these before committing transaction */
3158 last_cdp_link = (cdp)->i_d.di_nlink==0;
3159
3160 /*
3161 * Take an extra ref on the child vnode so that it
3162 * does not go to xfs_inactive() from within the commit.
3163 */
3164 IHOLD(cdp);
3165
3166 /*
3167 * If this is a synchronous mount, make sure that the
3168 * rmdir transaction goes to disk before returning to
3169 * the user.
3170 */
3171 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3172 xfs_trans_set_sync(tp);
3173 }
3174
3175 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3176 if (error) {
3177 xfs_bmap_cancel(&free_list);
3178 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3179 XFS_TRANS_ABORT));
3180 IRELE(cdp);
3181 goto std_return;
3182 }
3183
3184 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3185 if (error) {
3186 IRELE(cdp);
3187 goto std_return;
3188 }
3189
3190
3191 /*
3192 * Let interposed file systems know about removed links.
3193 */
3194 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3195
3196 IRELE(cdp);
3197
3198 /* Fall through to std_return with error = 0 or the errno
3199 * from xfs_trans_commit. */
3200std_return:
3201 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3202 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3203 dir_vp, DM_RIGHT_NULL,
3204 NULL, DM_RIGHT_NULL,
3205 name, NULL, dm_di_mode,
3206 error, 0);
3207 }
3208 return error;
3209
3210 error1:
3211 xfs_bmap_cancel(&free_list);
3212 cancel_flags |= XFS_TRANS_ABORT;
3213 error_return:
3214 xfs_trans_cancel(tp, cancel_flags);
3215 goto std_return;
3216}
3217
3218
3219/*
3220 * xfs_readdir
3221 *
3222 * Read dp's entries starting at uiop->uio_offset and translate them into
3223 * bufsize bytes worth of struct dirents starting at bufbase.
3224 */
3225STATIC int
3226xfs_readdir(
3227 bhv_desc_t *dir_bdp,
3228 uio_t *uiop,
3229 cred_t *credp,
3230 int *eofp)
3231{
3232 xfs_inode_t *dp;
3233 xfs_trans_t *tp = NULL;
3234 int error = 0;
3235 uint lock_mode;
3236 xfs_off_t start_offset;
3237
3238 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3239 (inst_t *)__return_address);
3240 dp = XFS_BHVTOI(dir_bdp);
3241
3242 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3243 return XFS_ERROR(EIO);
3244 }
3245
3246 lock_mode = xfs_ilock_map_shared(dp);
3247 start_offset = uiop->uio_offset;
3248 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3249 if (start_offset != uiop->uio_offset) {
3250 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3251 }
3252 xfs_iunlock_map_shared(dp, lock_mode);
3253 return error;
3254}
3255
3256
3257/*
3258 * xfs_symlink
3259 *
3260 */
3261STATIC int
3262xfs_symlink(
3263 bhv_desc_t *dir_bdp,
3264 vname_t *dentry,
3265 vattr_t *vap,
3266 char *target_path,
3267 vnode_t **vpp,
3268 cred_t *credp)
3269{
3270 xfs_trans_t *tp;
3271 xfs_mount_t *mp;
3272 xfs_inode_t *dp;
3273 xfs_inode_t *ip;
3274 int error;
3275 int pathlen;
3276 xfs_bmap_free_t free_list;
3277 xfs_fsblock_t first_block;
3278 boolean_t dp_joined_to_trans;
3279 vnode_t *dir_vp;
3280 uint cancel_flags;
3281 int committed;
3282 xfs_fileoff_t first_fsb;
3283 xfs_filblks_t fs_blocks;
3284 int nmaps;
3285 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3286 xfs_daddr_t d;
3287 char *cur_chunk;
3288 int byte_cnt;
3289 int n;
3290 xfs_buf_t *bp;
3291 xfs_prid_t prid;
3292 struct xfs_dquot *udqp, *gdqp;
3293 uint resblks;
3294 char *link_name = VNAME(dentry);
3295 int link_namelen;
3296
3297 *vpp = NULL;
3298 dir_vp = BHV_TO_VNODE(dir_bdp);
3299 dp = XFS_BHVTOI(dir_bdp);
3300 dp_joined_to_trans = B_FALSE;
3301 error = 0;
3302 ip = NULL;
3303 tp = NULL;
3304
3305 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3306
3307 mp = dp->i_mount;
3308
3309 if (XFS_FORCED_SHUTDOWN(mp))
3310 return XFS_ERROR(EIO);
3311
3312 link_namelen = VNAMELEN(dentry);
3313
3314 /*
3315 * Check component lengths of the target path name.
3316 */
3317 pathlen = strlen(target_path);
3318 if (pathlen >= MAXPATHLEN) /* total string too long */
3319 return XFS_ERROR(ENAMETOOLONG);
3320 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3321 int len, total;
3322 char *path;
3323
3324 for(total = 0, path = target_path; total < pathlen;) {
3325 /*
3326 * Skip any slashes.
3327 */
3328 while(*path == '/') {
3329 total++;
3330 path++;
3331 }
3332
3333 /*
3334 * Count up to the next slash or end of path.
3335 * Error out if the component is bigger than MAXNAMELEN.
3336 */
3337 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3338 if (++len >= MAXNAMELEN) {
3339 error = ENAMETOOLONG;
3340 return error;
3341 }
3342 }
3343 }
3344 }
3345
3346 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3347 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3348 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3349 link_name, target_path, 0, 0, 0);
3350 if (error)
3351 return error;
3352 }
3353
3354 /* Return through std_return after this point. */
3355
3356 udqp = gdqp = NULL;
3357 if (vap->va_mask & XFS_AT_PROJID)
3358 prid = (xfs_prid_t)vap->va_projid;
3359 else
3360 prid = (xfs_prid_t)dfltprid;
3361
3362 /*
3363 * Make sure that we have allocated dquot(s) on disk.
3364 */
3365 error = XFS_QM_DQVOPALLOC(mp, dp,
3366 current_fsuid(credp), current_fsgid(credp),
3367 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3368 if (error)
3369 goto std_return;
3370
3371 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3372 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3373 /*
3374 * The symlink will fit into the inode data fork?
3375 * There can't be any attributes so we get the whole variable part.
3376 */
3377 if (pathlen <= XFS_LITINO(mp))
3378 fs_blocks = 0;
3379 else
3380 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3381 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3382 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3383 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3384 if (error == ENOSPC && fs_blocks == 0) {
3385 resblks = 0;
3386 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3387 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3388 }
3389 if (error) {
3390 cancel_flags = 0;
3391 dp = NULL;
3392 goto error_return;
3393 }
3394
3395 xfs_ilock(dp, XFS_ILOCK_EXCL);
3396
3397 /*
3398 * Check whether the directory allows new symlinks or not.
3399 */
3400 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3401 error = XFS_ERROR(EPERM);
3402 goto error_return;
3403 }
3404
3405 /*
3406 * Reserve disk quota : blocks and inode.
3407 */
3408 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3409 if (error)
3410 goto error_return;
3411
3412 /*
3413 * Check for ability to enter directory entry, if no space reserved.
3414 */
3415 if (resblks == 0 &&
3416 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3417 goto error_return;
3418 /*
3419 * Initialize the bmap freelist prior to calling either
3420 * bmapi or the directory create code.
3421 */
3422 XFS_BMAP_INIT(&free_list, &first_block);
3423
3424 /*
3425 * Allocate an inode for the symlink.
3426 */
3427 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3428 1, 0, credp, prid, resblks > 0, &ip, NULL);
3429 if (error) {
3430 if (error == ENOSPC)
3431 goto error_return;
3432 goto error1;
3433 }
3434 ITRACE(ip);
3435
3436 VN_HOLD(dir_vp);
3437 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3438 dp_joined_to_trans = B_TRUE;
3439
3440 /*
3441 * Also attach the dquot(s) to it, if applicable.
3442 */
3443 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3444
3445 if (resblks)
3446 resblks -= XFS_IALLOC_SPACE_RES(mp);
3447 /*
3448 * If the symlink will fit into the inode, write it inline.
3449 */
3450 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3451 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3452 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3453 ip->i_d.di_size = pathlen;
3454
3455 /*
3456 * The inode was initially created in extent format.
3457 */
3458 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3459 ip->i_df.if_flags |= XFS_IFINLINE;
3460
3461 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3462 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3463
3464 } else {
3465 first_fsb = 0;
3466 nmaps = SYMLINK_MAPS;
3467
3468 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3469 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3470 &first_block, resblks, mval, &nmaps,
3471 &free_list);
3472 if (error) {
3473 goto error1;
3474 }
3475
3476 if (resblks)
3477 resblks -= fs_blocks;
3478 ip->i_d.di_size = pathlen;
3479 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3480
3481 cur_chunk = target_path;
3482 for (n = 0; n < nmaps; n++) {
3483 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3484 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3485 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3486 BTOBB(byte_cnt), 0);
3487 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3488 if (pathlen < byte_cnt) {
3489 byte_cnt = pathlen;
3490 }
3491 pathlen -= byte_cnt;
3492
3493 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3494 cur_chunk += byte_cnt;
3495
3496 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3497 }
3498 }
3499
3500 /*
3501 * Create the directory entry for the symlink.
3502 */
3503 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3504 ip->i_ino, &first_block, &free_list, resblks);
3505 if (error) {
3506 goto error1;
3507 }
3508 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3509 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3510
3511 /*
3512 * Bump the in memory version number of the parent directory
3513 * so that other processes accessing it will recognize that
3514 * the directory has changed.
3515 */
3516 dp->i_gen++;
3517
3518 /*
3519 * If this is a synchronous mount, make sure that the
3520 * symlink transaction goes to disk before returning to
3521 * the user.
3522 */
3523 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3524 xfs_trans_set_sync(tp);
3525 }
3526
3527 /*
3528 * xfs_trans_commit normally decrements the vnode ref count
3529 * when it unlocks the inode. Since we want to return the
3530 * vnode to the caller, we bump the vnode ref count now.
3531 */
3532 IHOLD(ip);
3533
3534 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3535 if (error) {
3536 goto error2;
3537 }
3538 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3539 XFS_QM_DQRELE(mp, udqp);
3540 XFS_QM_DQRELE(mp, gdqp);
3541
3542 /* Fall through to std_return with error = 0 or errno from
3543 * xfs_trans_commit */
3544std_return:
3545 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3546 DM_EVENT_POSTSYMLINK)) {
3547 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3548 dir_vp, DM_RIGHT_NULL,
3549 error ? NULL : XFS_ITOV(ip),
3550 DM_RIGHT_NULL, link_name, target_path,
3551 0, error, 0);
3552 }
3553
3554 if (!error) {
3555 vnode_t *vp;
3556
3557 ASSERT(ip);
3558 vp = XFS_ITOV(ip);
3559 *vpp = vp;
3560 }
3561 return error;
3562
3563 error2:
3564 IRELE(ip);
3565 error1:
3566 xfs_bmap_cancel(&free_list);
3567 cancel_flags |= XFS_TRANS_ABORT;
3568 error_return:
3569 xfs_trans_cancel(tp, cancel_flags);
3570 XFS_QM_DQRELE(mp, udqp);
3571 XFS_QM_DQRELE(mp, gdqp);
3572
3573 if (!dp_joined_to_trans && (dp != NULL)) {
3574 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3575 }
3576
3577 goto std_return;
3578}
3579
3580
3581/*
3582 * xfs_fid2
3583 *
3584 * A fid routine that takes a pointer to a previously allocated
3585 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3586 */
3587STATIC int
3588xfs_fid2(
3589 bhv_desc_t *bdp,
3590 fid_t *fidp)
3591{
3592 xfs_inode_t *ip;
3593 xfs_fid2_t *xfid;
3594
3595 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3596 (inst_t *)__return_address);
3597 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3598
3599 xfid = (xfs_fid2_t *)fidp;
3600 ip = XFS_BHVTOI(bdp);
3601 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3602 xfid->fid_pad = 0;
3603 /*
3604 * use memcpy because the inode is a long long and there's no
3605 * assurance that xfid->fid_ino is properly aligned.
3606 */
3607 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3608 xfid->fid_gen = ip->i_d.di_gen;
3609
3610 return 0;
3611}
3612
3613
3614/*
3615 * xfs_rwlock
3616 */
3617int
3618xfs_rwlock(
3619 bhv_desc_t *bdp,
3620 vrwlock_t locktype)
3621{
3622 xfs_inode_t *ip;
3623 vnode_t *vp;
3624
3625 vp = BHV_TO_VNODE(bdp);
3626 if (vp->v_type == VDIR)
3627 return 1;
3628 ip = XFS_BHVTOI(bdp);
3629 if (locktype == VRWLOCK_WRITE) {
3630 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3631 } else if (locktype == VRWLOCK_TRY_READ) {
3632 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3633 } else if (locktype == VRWLOCK_TRY_WRITE) {
3634 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3635 } else {
3636 ASSERT((locktype == VRWLOCK_READ) ||
3637 (locktype == VRWLOCK_WRITE_DIRECT));
3638 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3639 }
3640
3641 return 1;
3642}
3643
3644
3645/*
3646 * xfs_rwunlock
3647 */
3648void
3649xfs_rwunlock(
3650 bhv_desc_t *bdp,
3651 vrwlock_t locktype)
3652{
3653 xfs_inode_t *ip;
3654 vnode_t *vp;
3655
3656 vp = BHV_TO_VNODE(bdp);
3657 if (vp->v_type == VDIR)
3658 return;
3659 ip = XFS_BHVTOI(bdp);
3660 if (locktype == VRWLOCK_WRITE) {
3661 /*
3662 * In the write case, we may have added a new entry to
3663 * the reference cache. This might store a pointer to
3664 * an inode to be released in this inode. If it is there,
3665 * clear the pointer and release the inode after unlocking
3666 * this one.
3667 */
3668 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3669 } else {
3670 ASSERT((locktype == VRWLOCK_READ) ||
3671 (locktype == VRWLOCK_WRITE_DIRECT));
3672 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3673 }
3674 return;
3675}
3676
3677STATIC int
3678xfs_inode_flush(
3679 bhv_desc_t *bdp,
3680 int flags)
3681{
3682 xfs_inode_t *ip;
3683 xfs_mount_t *mp;
3684 xfs_inode_log_item_t *iip;
3685 int error = 0;
3686
3687 ip = XFS_BHVTOI(bdp);
3688 mp = ip->i_mount;
3689 iip = ip->i_itemp;
3690
3691 if (XFS_FORCED_SHUTDOWN(mp))
3692 return XFS_ERROR(EIO);
3693
3694 /*
3695 * Bypass inodes which have already been cleaned by
3696 * the inode flush clustering code inside xfs_iflush
3697 */
3698 if ((ip->i_update_core == 0) &&
3699 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3700 return 0;
3701
3702 if (flags & FLUSH_LOG) {
3703 if (iip && iip->ili_last_lsn) {
3704 xlog_t *log = mp->m_log;
3705 xfs_lsn_t sync_lsn;
3706 int s, log_flags = XFS_LOG_FORCE;
3707
3708 s = GRANT_LOCK(log);
3709 sync_lsn = log->l_last_sync_lsn;
3710 GRANT_UNLOCK(log, s);
3711
3712 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3713 return 0;
3714
3715 if (flags & FLUSH_SYNC)
3716 log_flags |= XFS_LOG_SYNC;
3717 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3718 }
3719 }
3720
3721 /*
3722 * We make this non-blocking if the inode is contended,
3723 * return EAGAIN to indicate to the caller that they
3724 * did not succeed. This prevents the flush path from
3725 * blocking on inodes inside another operation right
3726 * now, they get caught later by xfs_sync.
3727 */
3728 if (flags & FLUSH_INODE) {
3729 int flush_flags;
3730
3731 if (xfs_ipincount(ip))
3732 return EAGAIN;
3733
3734 if (flags & FLUSH_SYNC) {
3735 xfs_ilock(ip, XFS_ILOCK_SHARED);
3736 xfs_iflock(ip);
3737 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3738 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3739 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3740 return EAGAIN;
3741 }
3742 } else {
3743 return EAGAIN;
3744 }
3745
3746 if (flags & FLUSH_SYNC)
3747 flush_flags = XFS_IFLUSH_SYNC;
3748 else
3749 flush_flags = XFS_IFLUSH_ASYNC;
3750
3751 error = xfs_iflush(ip, flush_flags);
3752 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3753 }
3754
3755 return error;
3756}
3757
3758
3759int
3760xfs_set_dmattrs (
3761 bhv_desc_t *bdp,
3762 u_int evmask,
3763 u_int16_t state,
3764 cred_t *credp)
3765{
3766 xfs_inode_t *ip;
3767 xfs_trans_t *tp;
3768 xfs_mount_t *mp;
3769 int error;
3770
3771 if (!capable(CAP_SYS_ADMIN))
3772 return XFS_ERROR(EPERM);
3773
3774 ip = XFS_BHVTOI(bdp);
3775 mp = ip->i_mount;
3776
3777 if (XFS_FORCED_SHUTDOWN(mp))
3778 return XFS_ERROR(EIO);
3779
3780 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3781 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3782 if (error) {
3783 xfs_trans_cancel(tp, 0);
3784 return error;
3785 }
3786 xfs_ilock(ip, XFS_ILOCK_EXCL);
3787 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3788
3789 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3790 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3791
3792 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3793 IHOLD(ip);
3794 error = xfs_trans_commit(tp, 0, NULL);
3795
3796 return error;
3797}
3798
3799
3800/*
3801 * xfs_reclaim
3802 */
3803STATIC int
3804xfs_reclaim(
3805 bhv_desc_t *bdp)
3806{
3807 xfs_inode_t *ip;
3808 vnode_t *vp;
3809
3810 vp = BHV_TO_VNODE(bdp);
3811 ip = XFS_BHVTOI(bdp);
3812
3813 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3814
3815 ASSERT(!VN_MAPPED(vp));
3816
3817 /* bad inode, get out here ASAP */
3818 if (VN_BAD(vp)) {
3819 xfs_ireclaim(ip);
3820 return 0;
3821 }
3822
3823 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3824 if (ip->i_d.di_size > 0) {
3825 /*
3826 * Flush and invalidate any data left around that is
3827 * a part of this file.
3828 *
3829 * Get the inode's i/o lock so that buffers are pushed
3830 * out while holding the proper lock. We can't hold
3831 * the inode lock here since flushing out buffers may
3832 * cause us to try to get the lock in xfs_strategy().
3833 *
3834 * We don't have to call remapf() here, because there
3835 * cannot be any mapped file references to this vnode
3836 * since it is being reclaimed.
3837 */
3838 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3839
3840 /*
3841 * If we hit an IO error, we need to make sure that the
3842 * buffer and page caches of file data for
3843 * the file are tossed away. We don't want to use
3844 * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3845 * pages to stay attached to the vnode, but be
3846 * marked P_BAD. pdflush/vnode_pagebad
3847 * hates that.
3848 */
3849 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3850 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3851 } else {
3852 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3853 }
3854
3855 ASSERT(VN_CACHED(vp) == 0);
3856 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3857 ip->i_delayed_blks == 0);
3858 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3859 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3860 /*
3861 * di_size field may not be quite accurate if we're
3862 * shutting down.
3863 */
3864 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3865 ASSERT(VN_CACHED(vp) == 0);
3866 }
3867 }
3868
3869 /* If we have nothing to flush with this inode then complete the
3870 * teardown now, otherwise break the link between the xfs inode
3871 * and the linux inode and clean up the xfs inode later. This
3872 * avoids flushing the inode to disk during the delete operation
3873 * itself.
3874 */
3875 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3876 xfs_ilock(ip, XFS_ILOCK_EXCL);
3877 xfs_iflock(ip);
3878 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3879 } else {
3880 xfs_mount_t *mp = ip->i_mount;
3881
3882 /* Protect sync from us */
3883 XFS_MOUNT_ILOCK(mp);
3884 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3885 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3886 ip->i_flags |= XFS_IRECLAIMABLE;
3887 XFS_MOUNT_IUNLOCK(mp);
3888 }
3889 return 0;
3890}
3891
3892int
3893xfs_finish_reclaim(
3894 xfs_inode_t *ip,
3895 int locked,
3896 int sync_mode)
3897{
3898 xfs_ihash_t *ih = ip->i_hash;
3899 vnode_t *vp = XFS_ITOV_NULL(ip);
3900 int error;
3901
3902 if (vp && VN_BAD(vp))
3903 goto reclaim;
3904
3905 /* The hash lock here protects a thread in xfs_iget_core from
3906 * racing with us on linking the inode back with a vnode.
3907 * Once we have the XFS_IRECLAIM flag set it will not touch
3908 * us.
3909 */
3910 write_lock(&ih->ih_lock);
3911 if ((ip->i_flags & XFS_IRECLAIM) ||
3912 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3913 write_unlock(&ih->ih_lock);
3914 if (locked) {
3915 xfs_ifunlock(ip);
3916 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3917 }
3918 return(1);
3919 }
3920 ip->i_flags |= XFS_IRECLAIM;
3921 write_unlock(&ih->ih_lock);
3922
3923 /*
3924 * If the inode is still dirty, then flush it out. If the inode
3925 * is not in the AIL, then it will be OK to flush it delwri as
3926 * long as xfs_iflush() does not keep any references to the inode.
3927 * We leave that decision up to xfs_iflush() since it has the
3928 * knowledge of whether it's OK to simply do a delwri flush of
3929 * the inode or whether we need to wait until the inode is
3930 * pulled from the AIL.
3931 * We get the flush lock regardless, though, just to make sure
3932 * we don't free it while it is being flushed.
3933 */
3934 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3935 if (!locked) {
3936 xfs_ilock(ip, XFS_ILOCK_EXCL);
3937 xfs_iflock(ip);
3938 }
3939
3940 if (ip->i_update_core ||
3941 ((ip->i_itemp != NULL) &&
3942 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3943 error = xfs_iflush(ip, sync_mode);
3944 /*
3945 * If we hit an error, typically because of filesystem
3946 * shutdown, we don't need to let vn_reclaim to know
3947 * because we're gonna reclaim the inode anyway.
3948 */
3949 if (error) {
3950 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3951 goto reclaim;
3952 }
3953 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3954 }
3955
3956 ASSERT(ip->i_update_core == 0);
3957 ASSERT(ip->i_itemp == NULL ||
3958 ip->i_itemp->ili_format.ilf_fields == 0);
3959 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3960 } else if (locked) {
3961 /*
3962 * We are not interested in doing an iflush if we're
3963 * in the process of shutting down the filesystem forcibly.
3964 * So, just reclaim the inode.
3965 */
3966 xfs_ifunlock(ip);
3967 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3968 }
3969
3970 reclaim:
3971 xfs_ireclaim(ip);
3972 return 0;
3973}
3974
3975int
3976xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3977{
3978 int purged;
3979 xfs_inode_t *ip, *n;
3980 int done = 0;
3981
3982 while (!done) {
3983 purged = 0;
3984 XFS_MOUNT_ILOCK(mp);
3985 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3986 if (noblock) {
3987 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3988 continue;
3989 if (xfs_ipincount(ip) ||
3990 !xfs_iflock_nowait(ip)) {
3991 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3992 continue;
3993 }
3994 }
3995 XFS_MOUNT_IUNLOCK(mp);
3996 xfs_finish_reclaim(ip, noblock,
3997 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
3998 purged = 1;
3999 break;
4000 }
4001
4002 done = !purged;
4003 }
4004
4005 XFS_MOUNT_IUNLOCK(mp);
4006 return 0;
4007}
4008
4009/*
4010 * xfs_alloc_file_space()
4011 * This routine allocates disk space for the given file.
4012 *
4013 * If alloc_type == 0, this request is for an ALLOCSP type
4014 * request which will change the file size. In this case, no
4015 * DMAPI event will be generated by the call. A TRUNCATE event
4016 * will be generated later by xfs_setattr.
4017 *
4018 * If alloc_type != 0, this request is for a RESVSP type
4019 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4020 * lower block boundary byte address is less than the file's
4021 * length.
4022 *
4023 * RETURNS:
4024 * 0 on success
4025 * errno on error
4026 *
4027 */
4028int
4029xfs_alloc_file_space(
4030 xfs_inode_t *ip,
4031 xfs_off_t offset,
4032 xfs_off_t len,
4033 int alloc_type,
4034 int attr_flags)
4035{
4036 xfs_filblks_t allocated_fsb;
4037 xfs_filblks_t allocatesize_fsb;
4038 int committed;
4039 xfs_off_t count;
4040 xfs_filblks_t datablocks;
4041 int error;
4042 xfs_fsblock_t firstfsb;
4043 xfs_bmap_free_t free_list;
4044 xfs_bmbt_irec_t *imapp;
4045 xfs_bmbt_irec_t imaps[1];
4046 xfs_mount_t *mp;
4047 int numrtextents;
4048 int reccount;
4049 uint resblks;
4050 int rt;
4051 int rtextsize;
4052 xfs_fileoff_t startoffset_fsb;
4053 xfs_trans_t *tp;
4054 int xfs_bmapi_flags;
4055
4056 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4057 mp = ip->i_mount;
4058
4059 if (XFS_FORCED_SHUTDOWN(mp))
4060 return XFS_ERROR(EIO);
4061
4062 /*
4063 * determine if this is a realtime file
4064 */
4065 if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4066 if (ip->i_d.di_extsize)
4067 rtextsize = ip->i_d.di_extsize;
4068 else
4069 rtextsize = mp->m_sb.sb_rextsize;
4070 } else
4071 rtextsize = 0;
4072
4073 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4074 return error;
4075
4076 if (len <= 0)
4077 return XFS_ERROR(EINVAL);
4078
4079 count = len;
4080 error = 0;
4081 imapp = &imaps[0];
4082 reccount = 1;
4083 xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4084 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4085 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4086
4087 /* Generate a DMAPI event if needed. */
4088 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4089 (attr_flags&ATTR_DMI) == 0 &&
4090 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4091 xfs_off_t end_dmi_offset;
4092
4093 end_dmi_offset = offset+len;
4094 if (end_dmi_offset > ip->i_d.di_size)
4095 end_dmi_offset = ip->i_d.di_size;
4096 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4097 offset, end_dmi_offset - offset,
4098 0, NULL);
4099 if (error)
4100 return(error);
4101 }
4102
4103 /*
4104 * allocate file space until done or until there is an error
4105 */
4106retry:
4107 while (allocatesize_fsb && !error) {
4108 /*
4109 * determine if reserving space on
4110 * the data or realtime partition.
4111 */
4112 if (rt) {
4113 xfs_fileoff_t s, e;
4114
4115 s = startoffset_fsb;
4116 do_div(s, rtextsize);
4117 s *= rtextsize;
4118 e = roundup_64(startoffset_fsb + allocatesize_fsb,
4119 rtextsize);
4120 numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4121 datablocks = 0;
4122 } else {
4123 datablocks = allocatesize_fsb;
4124 numrtextents = 0;
4125 }
4126
4127 /*
4128 * allocate and setup the transaction
4129 */
4130 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4131 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4132 error = xfs_trans_reserve(tp,
4133 resblks,
4134 XFS_WRITE_LOG_RES(mp),
4135 numrtextents,
4136 XFS_TRANS_PERM_LOG_RES,
4137 XFS_WRITE_LOG_COUNT);
4138
4139 /*
4140 * check for running out of space
4141 */
4142 if (error) {
4143 /*
4144 * Free the transaction structure.
4145 */
4146 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4147 xfs_trans_cancel(tp, 0);
4148 break;
4149 }
4150 xfs_ilock(ip, XFS_ILOCK_EXCL);
4151 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4152 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4153 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4154 if (error)
4155 goto error1;
4156
4157 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4158 xfs_trans_ihold(tp, ip);
4159
4160 /*
4161 * issue the bmapi() call to allocate the blocks
4162 */
4163 XFS_BMAP_INIT(&free_list, &firstfsb);
4164 error = xfs_bmapi(tp, ip, startoffset_fsb,
4165 allocatesize_fsb, xfs_bmapi_flags,
4166 &firstfsb, 0, imapp, &reccount,
4167 &free_list);
4168 if (error) {
4169 goto error0;
4170 }
4171
4172 /*
4173 * complete the transaction
4174 */
4175 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4176 if (error) {
4177 goto error0;
4178 }
4179
4180 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4181 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4182 if (error) {
4183 break;
4184 }
4185
4186 allocated_fsb = imapp->br_blockcount;
4187
4188 if (reccount == 0) {
4189 error = XFS_ERROR(ENOSPC);
4190 break;
4191 }
4192
4193 startoffset_fsb += allocated_fsb;
4194 allocatesize_fsb -= allocated_fsb;
4195 }
4196dmapi_enospc_check:
4197 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4198 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4199
4200 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4201 XFS_ITOV(ip), DM_RIGHT_NULL,
4202 XFS_ITOV(ip), DM_RIGHT_NULL,
4203 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4204 if (error == 0)
4205 goto retry; /* Maybe DMAPI app. has made space */
4206 /* else fall through with error from XFS_SEND_DATA */
4207 }
4208
4209 return error;
4210
4211 error0:
4212 xfs_bmap_cancel(&free_list);
4213 error1:
4214 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4215 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4216 goto dmapi_enospc_check;
4217}
4218
4219/*
4220 * Zero file bytes between startoff and endoff inclusive.
4221 * The iolock is held exclusive and no blocks are buffered.
4222 */
4223STATIC int
4224xfs_zero_remaining_bytes(
4225 xfs_inode_t *ip,
4226 xfs_off_t startoff,
4227 xfs_off_t endoff)
4228{
4229 xfs_bmbt_irec_t imap;
4230 xfs_fileoff_t offset_fsb;
4231 xfs_off_t lastoffset;
4232 xfs_off_t offset;
4233 xfs_buf_t *bp;
4234 xfs_mount_t *mp = ip->i_mount;
4235 int nimap;
4236 int error = 0;
4237
4238 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4239 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4240 mp->m_rtdev_targp : mp->m_ddev_targp);
4241
4242 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4243 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4244 nimap = 1;
4245 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4246 &nimap, NULL);
4247 if (error || nimap < 1)
4248 break;
4249 ASSERT(imap.br_blockcount >= 1);
4250 ASSERT(imap.br_startoff == offset_fsb);
4251 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4252 if (lastoffset > endoff)
4253 lastoffset = endoff;
4254 if (imap.br_startblock == HOLESTARTBLOCK)
4255 continue;
4256 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4257 if (imap.br_state == XFS_EXT_UNWRITTEN)
4258 continue;
4259 XFS_BUF_UNDONE(bp);
4260 XFS_BUF_UNWRITE(bp);
4261 XFS_BUF_READ(bp);
4262 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4263 xfsbdstrat(mp, bp);
4264 if ((error = xfs_iowait(bp))) {
4265 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4266 mp, bp, XFS_BUF_ADDR(bp));
4267 break;
4268 }
4269 memset(XFS_BUF_PTR(bp) +
4270 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4271 0, lastoffset - offset + 1);
4272 XFS_BUF_UNDONE(bp);
4273 XFS_BUF_UNREAD(bp);
4274 XFS_BUF_WRITE(bp);
4275 xfsbdstrat(mp, bp);
4276 if ((error = xfs_iowait(bp))) {
4277 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4278 mp, bp, XFS_BUF_ADDR(bp));
4279 break;
4280 }
4281 }
4282 xfs_buf_free(bp);
4283 return error;
4284}
4285
4286/*
4287 * xfs_free_file_space()
4288 * This routine frees disk space for the given file.
4289 *
4290 * This routine is only called by xfs_change_file_space
4291 * for an UNRESVSP type call.
4292 *
4293 * RETURNS:
4294 * 0 on success
4295 * errno on error
4296 *
4297 */
4298STATIC int
4299xfs_free_file_space(
4300 xfs_inode_t *ip,
4301 xfs_off_t offset,
4302 xfs_off_t len,
4303 int attr_flags)
4304{
4305 int committed;
4306 int done;
4307 xfs_off_t end_dmi_offset;
4308 xfs_fileoff_t endoffset_fsb;
4309 int error;
4310 xfs_fsblock_t firstfsb;
4311 xfs_bmap_free_t free_list;
4312 xfs_off_t ilen;
4313 xfs_bmbt_irec_t imap;
4314 xfs_off_t ioffset;
4315 xfs_extlen_t mod=0;
4316 xfs_mount_t *mp;
4317 int nimap;
4318 uint resblks;
4319 int rounding;
4320 int rt;
4321 xfs_fileoff_t startoffset_fsb;
4322 xfs_trans_t *tp;
4323 int need_iolock = (attr_flags & ATTR_DMI) == 0;
4324
4325 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4326 mp = ip->i_mount;
4327
4328 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4329 return error;
4330
4331 error = 0;
4332 if (len <= 0) /* if nothing being freed */
4333 return error;
4334 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4335 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4336 end_dmi_offset = offset + len;
4337 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4338
4339 if (offset < ip->i_d.di_size &&
4340 (attr_flags & ATTR_DMI) == 0 &&
4341 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4342 if (end_dmi_offset > ip->i_d.di_size)
4343 end_dmi_offset = ip->i_d.di_size;
4344 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4345 offset, end_dmi_offset - offset,
4346 AT_DELAY_FLAG(attr_flags), NULL);
4347 if (error)
4348 return(error);
4349 }
4350
4351 if (need_iolock)
4352 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4353 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4354 (__uint8_t)NBPP);
4355 ilen = len + (offset & (rounding - 1));
4356 ioffset = offset & ~(rounding - 1);
4357 if (ilen & (rounding - 1))
4358 ilen = (ilen + rounding) & ~(rounding - 1);
4359 xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4360 /*
4361 * Need to zero the stuff we're not freeing, on disk.
4362 * If its a realtime file & can't use unwritten extents then we
4363 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4364 * will take care of it for us.
4365 */
4366 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4367 nimap = 1;
4368 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4369 &imap, &nimap, NULL);
4370 if (error)
4371 goto out_unlock_iolock;
4372 ASSERT(nimap == 0 || nimap == 1);
4373 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4374 xfs_daddr_t block;
4375
4376 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4377 block = imap.br_startblock;
4378 mod = do_div(block, mp->m_sb.sb_rextsize);
4379 if (mod)
4380 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4381 }
4382 nimap = 1;
4383 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4384 &imap, &nimap, NULL);
4385 if (error)
4386 goto out_unlock_iolock;
4387 ASSERT(nimap == 0 || nimap == 1);
4388 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4389 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4390 mod++;
4391 if (mod && (mod != mp->m_sb.sb_rextsize))
4392 endoffset_fsb -= mod;
4393 }
4394 }
4395 if ((done = (endoffset_fsb <= startoffset_fsb)))
4396 /*
4397 * One contiguous piece to clear
4398 */
4399 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4400 else {
4401 /*
4402 * Some full blocks, possibly two pieces to clear
4403 */
4404 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4405 error = xfs_zero_remaining_bytes(ip, offset,
4406 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4407 if (!error &&
4408 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4409 error = xfs_zero_remaining_bytes(ip,
4410 XFS_FSB_TO_B(mp, endoffset_fsb),
4411 offset + len - 1);
4412 }
4413
4414 /*
4415 * free file space until done or until there is an error
4416 */
4417 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4418 while (!error && !done) {
4419
4420 /*
4421 * allocate and setup the transaction
4422 */
4423 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4424 error = xfs_trans_reserve(tp,
4425 resblks,
4426 XFS_WRITE_LOG_RES(mp),
4427 0,
4428 XFS_TRANS_PERM_LOG_RES,
4429 XFS_WRITE_LOG_COUNT);
4430
4431 /*
4432 * check for running out of space
4433 */
4434 if (error) {
4435 /*
4436 * Free the transaction structure.
4437 */
4438 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4439 xfs_trans_cancel(tp, 0);
4440 break;
4441 }
4442 xfs_ilock(ip, XFS_ILOCK_EXCL);
4443 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4444 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4445 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4446 if (error)
4447 goto error1;
4448
4449 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4450 xfs_trans_ihold(tp, ip);
4451
4452 /*
4453 * issue the bunmapi() call to free the blocks
4454 */
4455 XFS_BMAP_INIT(&free_list, &firstfsb);
4456 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4457 endoffset_fsb - startoffset_fsb,
4458 0, 2, &firstfsb, &free_list, &done);
4459 if (error) {
4460 goto error0;
4461 }
4462
4463 /*
4464 * complete the transaction
4465 */
4466 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4467 if (error) {
4468 goto error0;
4469 }
4470
4471 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4472 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4473 }
4474
4475 out_unlock_iolock:
4476 if (need_iolock)
4477 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4478 return error;
4479
4480 error0:
4481 xfs_bmap_cancel(&free_list);
4482 error1:
4483 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4484 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4485 XFS_ILOCK_EXCL);
4486 return error;
4487}
4488
4489/*
4490 * xfs_change_file_space()
4491 * This routine allocates or frees disk space for the given file.
4492 * The user specified parameters are checked for alignment and size
4493 * limitations.
4494 *
4495 * RETURNS:
4496 * 0 on success
4497 * errno on error
4498 *
4499 */
4500int
4501xfs_change_file_space(
4502 bhv_desc_t *bdp,
4503 int cmd,
4504 xfs_flock64_t *bf,
4505 xfs_off_t offset,
4506 cred_t *credp,
4507 int attr_flags)
4508{
4509 int clrprealloc;
4510 int error;
4511 xfs_fsize_t fsize;
4512 xfs_inode_t *ip;
4513 xfs_mount_t *mp;
4514 int setprealloc;
4515 xfs_off_t startoffset;
4516 xfs_off_t llen;
4517 xfs_trans_t *tp;
4518 vattr_t va;
4519 vnode_t *vp;
4520
4521 vp = BHV_TO_VNODE(bdp);
4522 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4523
4524 ip = XFS_BHVTOI(bdp);
4525 mp = ip->i_mount;
4526
4527 /*
4528 * must be a regular file and have write permission
4529 */
4530 if (vp->v_type != VREG)
4531 return XFS_ERROR(EINVAL);
4532
4533 xfs_ilock(ip, XFS_ILOCK_SHARED);
4534
4535 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4536 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4537 return error;
4538 }
4539
4540 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4541
4542 switch (bf->l_whence) {
4543 case 0: /*SEEK_SET*/
4544 break;
4545 case 1: /*SEEK_CUR*/
4546 bf->l_start += offset;
4547 break;
4548 case 2: /*SEEK_END*/
4549 bf->l_start += ip->i_d.di_size;
4550 break;
4551 default:
4552 return XFS_ERROR(EINVAL);
4553 }
4554
4555 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4556
4557 if ( (bf->l_start < 0)
4558 || (bf->l_start > XFS_MAXIOFFSET(mp))
4559 || (bf->l_start + llen < 0)
4560 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4561 return XFS_ERROR(EINVAL);
4562
4563 bf->l_whence = 0;
4564
4565 startoffset = bf->l_start;
4566 fsize = ip->i_d.di_size;
4567
4568 /*
4569 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4570 * file space.
4571 * These calls do NOT zero the data space allocated to the file,
4572 * nor do they change the file size.
4573 *
4574 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4575 * space.
4576 * These calls cause the new file data to be zeroed and the file
4577 * size to be changed.
4578 */
4579 setprealloc = clrprealloc = 0;
4580
4581 switch (cmd) {
4582 case XFS_IOC_RESVSP:
4583 case XFS_IOC_RESVSP64:
4584 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4585 1, attr_flags);
4586 if (error)
4587 return error;
4588 setprealloc = 1;
4589 break;
4590
4591 case XFS_IOC_UNRESVSP:
4592 case XFS_IOC_UNRESVSP64:
4593 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4594 attr_flags)))
4595 return error;
4596 break;
4597
4598 case XFS_IOC_ALLOCSP:
4599 case XFS_IOC_ALLOCSP64:
4600 case XFS_IOC_FREESP:
4601 case XFS_IOC_FREESP64:
4602 if (startoffset > fsize) {
4603 error = xfs_alloc_file_space(ip, fsize,
4604 startoffset - fsize, 0, attr_flags);
4605 if (error)
4606 break;
4607 }
4608
4609 va.va_mask = XFS_AT_SIZE;
4610 va.va_size = startoffset;
4611
4612 error = xfs_setattr(bdp, &va, attr_flags, credp);
4613
4614 if (error)
4615 return error;
4616
4617 clrprealloc = 1;
4618 break;
4619
4620 default:
4621 ASSERT(0);
4622 return XFS_ERROR(EINVAL);
4623 }
4624
4625 /*
4626 * update the inode timestamp, mode, and prealloc flag bits
4627 */
4628 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4629
4630 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4631 0, 0, 0))) {
4632 /* ASSERT(0); */
4633 xfs_trans_cancel(tp, 0);
4634 return error;
4635 }
4636
4637 xfs_ilock(ip, XFS_ILOCK_EXCL);
4638
4639 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4640 xfs_trans_ihold(tp, ip);
4641
4642 if ((attr_flags & ATTR_DMI) == 0) {
4643 ip->i_d.di_mode &= ~S_ISUID;
4644
4645 /*
4646 * Note that we don't have to worry about mandatory
4647 * file locking being disabled here because we only
4648 * clear the S_ISGID bit if the Group execute bit is
4649 * on, but if it was on then mandatory locking wouldn't
4650 * have been enabled.
4651 */
4652 if (ip->i_d.di_mode & S_IXGRP)
4653 ip->i_d.di_mode &= ~S_ISGID;
4654
4655 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4656 }
4657 if (setprealloc)
4658 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4659 else if (clrprealloc)
4660 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4661
4662 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4663 xfs_trans_set_sync(tp);
4664
4665 error = xfs_trans_commit(tp, 0, NULL);
4666
4667 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4668
4669 return error;
4670}
4671
4672vnodeops_t xfs_vnodeops = {
4673 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4674 .vop_open = xfs_open,
4675 .vop_read = xfs_read,
4676#ifdef HAVE_SENDFILE
4677 .vop_sendfile = xfs_sendfile,
4678#endif
4679 .vop_write = xfs_write,
4680 .vop_ioctl = xfs_ioctl,
4681 .vop_getattr = xfs_getattr,
4682 .vop_setattr = xfs_setattr,
4683 .vop_access = xfs_access,
4684 .vop_lookup = xfs_lookup,
4685 .vop_create = xfs_create,
4686 .vop_remove = xfs_remove,
4687 .vop_link = xfs_link,
4688 .vop_rename = xfs_rename,
4689 .vop_mkdir = xfs_mkdir,
4690 .vop_rmdir = xfs_rmdir,
4691 .vop_readdir = xfs_readdir,
4692 .vop_symlink = xfs_symlink,
4693 .vop_readlink = xfs_readlink,
4694 .vop_fsync = xfs_fsync,
4695 .vop_inactive = xfs_inactive,
4696 .vop_fid2 = xfs_fid2,
4697 .vop_rwlock = xfs_rwlock,
4698 .vop_rwunlock = xfs_rwunlock,
4699 .vop_bmap = xfs_bmap,
4700 .vop_reclaim = xfs_reclaim,
4701 .vop_attr_get = xfs_attr_get,
4702 .vop_attr_set = xfs_attr_set,
4703 .vop_attr_remove = xfs_attr_remove,
4704 .vop_attr_list = xfs_attr_list,
4705 .vop_link_removed = (vop_link_removed_t)fs_noval,
4706 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4707 .vop_tosspages = fs_tosspages,
4708 .vop_flushinval_pages = fs_flushinval_pages,
4709 .vop_flush_pages = fs_flush_pages,
4710 .vop_release = xfs_release,
4711 .vop_iflush = xfs_inode_flush,
4712};