blob: eaab355f5a892fcd95af7402dddf2845723d59ed [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018
19#include <linux/capability.h>
20
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110026#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_trans.h"
28#include "xfs_sb.h"
29#include "xfs_ag.h"
30#include "xfs_dir.h"
31#include "xfs_dir2.h"
32#include "xfs_dmapi.h"
33#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110036#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_dir_sf.h"
39#include "xfs_dir2_sf.h"
40#include "xfs_attr_sf.h"
41#include "xfs_dinode.h"
42#include "xfs_inode.h"
43#include "xfs_inode_item.h"
44#include "xfs_dir_leaf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include "xfs_itable.h"
46#include "xfs_btree.h"
47#include "xfs_ialloc.h"
48#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include "xfs_attr.h"
51#include "xfs_rw.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include "xfs_quota.h"
54#include "xfs_utils.h"
Nathan Scotta844f452005-11-02 14:38:42 +110055#include "xfs_rtalloc.h"
56#include "xfs_refcache.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include "xfs_log_priv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110059#include "xfs_mac.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
61
62/*
63 * The maximum pathlen is 1024 bytes. Since the minimum file system
64 * blocksize is 512 bytes, we can get a max of 2 extents back from
65 * bmapi.
66 */
67#define SYMLINK_MAPS 2
68
69/*
70 * For xfs, we check that the file isn't too big to be opened by this kernel.
71 * No other open action is required for regular files. Devices are handled
72 * through the specfs file system, pipes through fifofs. Device and
73 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
74 * when a new vnode is first looked up or created.
75 */
76STATIC int
77xfs_open(
78 bhv_desc_t *bdp,
79 cred_t *credp)
80{
81 int mode;
82 vnode_t *vp;
83 xfs_inode_t *ip;
84
85 vp = BHV_TO_VNODE(bdp);
86 ip = XFS_BHVTOI(bdp);
87
88 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
89 return XFS_ERROR(EIO);
90
91 /*
92 * If it's a directory with any blocks, read-ahead block 0
93 * as we're almost certain to have the next operation be a read there.
94 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +100095 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 mode = xfs_ilock_map_shared(ip);
97 if (ip->i_d.di_nextents > 0)
98 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
99 xfs_iunlock(ip, mode);
100 }
101 return 0;
102}
103
104
105/*
106 * xfs_getattr
107 */
108STATIC int
109xfs_getattr(
110 bhv_desc_t *bdp,
111 vattr_t *vap,
112 int flags,
113 cred_t *credp)
114{
115 xfs_inode_t *ip;
116 xfs_mount_t *mp;
117 vnode_t *vp;
118
119 vp = BHV_TO_VNODE(bdp);
120 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
121
122 ip = XFS_BHVTOI(bdp);
123 mp = ip->i_mount;
124
125 if (XFS_FORCED_SHUTDOWN(mp))
126 return XFS_ERROR(EIO);
127
128 if (!(flags & ATTR_LAZY))
129 xfs_ilock(ip, XFS_ILOCK_SHARED);
130
131 vap->va_size = ip->i_d.di_size;
132 if (vap->va_mask == XFS_AT_SIZE)
133 goto all_done;
134
135 vap->va_nblocks =
136 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
137 vap->va_nodeid = ip->i_ino;
138#if XFS_BIG_INUMS
139 vap->va_nodeid += mp->m_inoadd;
140#endif
141 vap->va_nlink = ip->i_d.di_nlink;
142
143 /*
144 * Quick exit for non-stat callers
145 */
146 if ((vap->va_mask &
147 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
148 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
149 goto all_done;
150
151 /*
152 * Copy from in-core inode.
153 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000154 vap->va_mode = ip->i_d.di_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 vap->va_uid = ip->i_d.di_uid;
156 vap->va_gid = ip->i_d.di_gid;
157 vap->va_projid = ip->i_d.di_projid;
158
159 /*
160 * Check vnode type block/char vs. everything else.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000162 switch (ip->i_d.di_mode & S_IFMT) {
163 case S_IFBLK:
164 case S_IFCHR:
165 vap->va_rdev = ip->i_df.if_u2.if_rdev;
166 vap->va_blocksize = BLKDEV_IOSIZE;
167 break;
168 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 vap->va_rdev = 0;
170
171 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
David Chinnere8c8b3a2005-11-02 10:33:05 +1100172 vap->va_blocksize = xfs_preferred_iosize(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 } else {
174
175 /*
176 * If the file blocks are being allocated from a
177 * realtime partition, then return the inode's
178 * realtime extent size or the realtime volume's
179 * extent size.
180 */
181 vap->va_blocksize = ip->i_d.di_extsize ?
182 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
183 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
184 }
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000185 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 }
187
Nathan Scottca5ccbf2006-01-11 21:03:04 +1100188 vn_atime_to_timespec(vp, &vap->va_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
190 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
191 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
192 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
193
194 /*
195 * Exit for stat callers. See if any of the rest of the fields
196 * to be filled in are needed.
197 */
198 if ((vap->va_mask &
199 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
200 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
201 goto all_done;
202
203 /*
204 * Convert di_flags to xflags.
205 */
206 vap->va_xflags = xfs_ip2xflags(ip);
207
208 /*
209 * Exit for inode revalidate. See if any of the rest of
210 * the fields to be filled in are needed.
211 */
212 if ((vap->va_mask &
213 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
214 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
215 goto all_done;
216
217 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
218 vap->va_nextents =
219 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
220 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
221 ip->i_d.di_nextents;
222 if (ip->i_afp)
223 vap->va_anextents =
224 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
225 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
226 ip->i_d.di_anextents;
227 else
228 vap->va_anextents = 0;
229 vap->va_gen = ip->i_d.di_gen;
230
231 all_done:
232 if (!(flags & ATTR_LAZY))
233 xfs_iunlock(ip, XFS_ILOCK_SHARED);
234 return 0;
235}
236
237
238/*
239 * xfs_setattr
240 */
241int
242xfs_setattr(
243 bhv_desc_t *bdp,
244 vattr_t *vap,
245 int flags,
246 cred_t *credp)
247{
248 xfs_inode_t *ip;
249 xfs_trans_t *tp;
250 xfs_mount_t *mp;
251 int mask;
252 int code;
253 uint lock_flags;
254 uint commit_flags=0;
255 uid_t uid=0, iuid=0;
256 gid_t gid=0, igid=0;
257 int timeflags = 0;
258 vnode_t *vp;
259 xfs_prid_t projid=0, iprojid=0;
260 int mandlock_before, mandlock_after;
261 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
262 int file_owner;
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700263 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
265 vp = BHV_TO_VNODE(bdp);
266 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
267
268 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
269 return XFS_ERROR(EROFS);
270
271 /*
272 * Cannot set certain attributes.
273 */
274 mask = vap->va_mask;
275 if (mask & XFS_AT_NOSET) {
276 return XFS_ERROR(EINVAL);
277 }
278
279 ip = XFS_BHVTOI(bdp);
280 mp = ip->i_mount;
281
282 if (XFS_FORCED_SHUTDOWN(mp))
283 return XFS_ERROR(EIO);
284
285 /*
286 * Timestamps do not need to be logged and hence do not
287 * need to be done within a transaction.
288 */
289 if (mask & XFS_AT_UPDTIMES) {
290 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
291 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
292 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
293 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
294 xfs_ichgtime(ip, timeflags);
295 return 0;
296 }
297
298 olddquot1 = olddquot2 = NULL;
299 udqp = gdqp = NULL;
300
301 /*
302 * If disk quotas is on, we make sure that the dquots do exist on disk,
303 * before we start any other transactions. Trying to do this later
304 * is messy. We don't care to take a readlock to look at the ids
305 * in inode here, because we can't hold it across the trans_reserve.
306 * If the IDs do change before we take the ilock, we're covered
307 * because the i_*dquot fields will get updated anyway.
308 */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000309 if (XFS_IS_QUOTA_ON(mp) &&
310 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 uint qflags = 0;
312
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000313 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 uid = vap->va_uid;
315 qflags |= XFS_QMOPT_UQUOTA;
316 } else {
317 uid = ip->i_d.di_uid;
318 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000319 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 gid = vap->va_gid;
321 qflags |= XFS_QMOPT_GQUOTA;
322 } else {
323 gid = ip->i_d.di_gid;
324 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000325 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
326 projid = vap->va_projid;
327 qflags |= XFS_QMOPT_PQUOTA;
328 } else {
329 projid = ip->i_d.di_projid;
330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 /*
332 * We take a reference when we initialize udqp and gdqp,
333 * so it is important that we never blindly double trip on
334 * the same variable. See xfs_create() for an example.
335 */
336 ASSERT(udqp == NULL);
337 ASSERT(gdqp == NULL);
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000338 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
339 &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 if (code)
Jesper Juhl014c2542006-01-15 02:37:08 +0100341 return code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 }
343
344 /*
345 * For the other attributes, we acquire the inode lock and
346 * first do an error checking pass.
347 */
348 tp = NULL;
349 lock_flags = XFS_ILOCK_EXCL;
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700350 ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
351 if (flags & ATTR_NOLOCK)
352 need_iolock = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 if (!(mask & XFS_AT_SIZE)) {
354 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
355 (mp->m_flags & XFS_MOUNT_WSYNC)) {
356 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
357 commit_flags = 0;
358 if ((code = xfs_trans_reserve(tp, 0,
359 XFS_ICHANGE_LOG_RES(mp), 0,
360 0, 0))) {
361 lock_flags = 0;
362 goto error_return;
363 }
364 }
365 } else {
366 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
367 !(flags & ATTR_DMI)) {
368 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
369 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
370 vap->va_size, 0, dmflags, NULL);
371 if (code) {
372 lock_flags = 0;
373 goto error_return;
374 }
375 }
376 if (need_iolock)
377 lock_flags |= XFS_IOLOCK_EXCL;
378 }
379
380 xfs_ilock(ip, lock_flags);
381
382 /* boolean: are we the file owner? */
383 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
384
385 /*
386 * Change various properties of a file.
387 * Only the owner or users with CAP_FOWNER
388 * capability may do these things.
389 */
390 if (mask &
391 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
392 XFS_AT_GID|XFS_AT_PROJID)) {
393 /*
394 * CAP_FOWNER overrides the following restrictions:
395 *
396 * The user ID of the calling process must be equal
397 * to the file owner ID, except in cases where the
398 * CAP_FSETID capability is applicable.
399 */
400 if (!file_owner && !capable(CAP_FOWNER)) {
401 code = XFS_ERROR(EPERM);
402 goto error_return;
403 }
404
405 /*
406 * CAP_FSETID overrides the following restrictions:
407 *
408 * The effective user ID of the calling process shall match
409 * the file owner when setting the set-user-ID and
410 * set-group-ID bits on that file.
411 *
412 * The effective group ID or one of the supplementary group
413 * IDs of the calling process shall match the group owner of
414 * the file when setting the set-group-ID bit on that file
415 */
416 if (mask & XFS_AT_MODE) {
417 mode_t m = 0;
418
419 if ((vap->va_mode & S_ISUID) && !file_owner)
420 m |= S_ISUID;
421 if ((vap->va_mode & S_ISGID) &&
422 !in_group_p((gid_t)ip->i_d.di_gid))
423 m |= S_ISGID;
424#if 0
425 /* Linux allows this, Irix doesn't. */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000426 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 m |= S_ISVTX;
428#endif
429 if (m && !capable(CAP_FSETID))
430 vap->va_mode &= ~m;
431 }
432 }
433
434 /*
435 * Change file ownership. Must be the owner or privileged.
436 * If the system was configured with the "restricted_chown"
437 * option, the owner is not permitted to give away the file,
438 * and can change the group id only to a group of which he
439 * or she is a member.
440 */
441 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
442 /*
443 * These IDs could have changed since we last looked at them.
444 * But, we're assured that if the ownership did change
445 * while we didn't have the inode locked, inode's dquot(s)
446 * would have changed also.
447 */
448 iuid = ip->i_d.di_uid;
449 iprojid = ip->i_d.di_projid;
450 igid = ip->i_d.di_gid;
451 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
452 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
453 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
454 iprojid;
455
456 /*
457 * CAP_CHOWN overrides the following restrictions:
458 *
459 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
460 * shall override the restriction that a process cannot
461 * change the user ID of a file it owns and the restriction
462 * that the group ID supplied to the chown() function
463 * shall be equal to either the group ID or one of the
464 * supplementary group IDs of the calling process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 */
466 if (restricted_chown &&
467 (iuid != uid || (igid != gid &&
468 !in_group_p((gid_t)gid))) &&
469 !capable(CAP_CHOWN)) {
470 code = XFS_ERROR(EPERM);
471 goto error_return;
472 }
473 /*
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000474 * Do a quota reservation only if uid/projid/gid is actually
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 * going to change.
476 */
477 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000478 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
480 ASSERT(tp);
481 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
482 capable(CAP_FOWNER) ?
483 XFS_QMOPT_FORCE_RES : 0);
484 if (code) /* out of quota */
485 goto error_return;
486 }
487 }
488
489 /*
490 * Truncate file. Must have write permission and not be a directory.
491 */
492 if (mask & XFS_AT_SIZE) {
493 /* Short circuit the truncate case for zero length files */
494 if ((vap->va_size == 0) &&
495 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
496 xfs_iunlock(ip, XFS_ILOCK_EXCL);
497 lock_flags &= ~XFS_ILOCK_EXCL;
498 if (mask & XFS_AT_CTIME)
499 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
500 code = 0;
501 goto error_return;
502 }
503
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000504 if (VN_ISDIR(vp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 code = XFS_ERROR(EISDIR);
506 goto error_return;
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000507 } else if (!VN_ISREG(vp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 code = XFS_ERROR(EINVAL);
509 goto error_return;
510 }
511 /*
512 * Make sure that the dquots are attached to the inode.
513 */
514 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
515 goto error_return;
516 }
517
518 /*
519 * Change file access or modified times.
520 */
521 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
522 if (!file_owner) {
523 if ((flags & ATTR_UTIME) &&
524 !capable(CAP_FOWNER)) {
525 code = XFS_ERROR(EPERM);
526 goto error_return;
527 }
528 }
529 }
530
531 /*
532 * Change extent size or realtime flag.
533 */
534 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
535 /*
536 * Can't change extent size if any extents are allocated.
537 */
Eric Sandeene94af022005-11-02 15:10:41 +1100538 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
540 vap->va_extsize) ) {
541 code = XFS_ERROR(EINVAL); /* EFBIG? */
542 goto error_return;
543 }
544
545 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 * Can't change realtime flag if any extents are allocated.
547 */
Eric Sandeene94af022005-11-02 15:10:41 +1100548 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
549 (mask & XFS_AT_XFLAGS) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
551 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
552 code = XFS_ERROR(EINVAL); /* EFBIG? */
553 goto error_return;
554 }
555 /*
556 * Extent size must be a multiple of the appropriate block
557 * size, if set at all.
558 */
559 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
560 xfs_extlen_t size;
561
562 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
563 ((mask & XFS_AT_XFLAGS) &&
564 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
565 size = mp->m_sb.sb_rextsize <<
566 mp->m_sb.sb_blocklog;
567 } else {
568 size = mp->m_sb.sb_blocksize;
569 }
570 if (vap->va_extsize % size) {
571 code = XFS_ERROR(EINVAL);
572 goto error_return;
573 }
574 }
575 /*
576 * If realtime flag is set then must have realtime data.
577 */
578 if ((mask & XFS_AT_XFLAGS) &&
579 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
580 if ((mp->m_sb.sb_rblocks == 0) ||
581 (mp->m_sb.sb_rextsize == 0) ||
582 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
583 code = XFS_ERROR(EINVAL);
584 goto error_return;
585 }
586 }
587
588 /*
589 * Can't modify an immutable/append-only file unless
590 * we have appropriate permission.
591 */
592 if ((mask & XFS_AT_XFLAGS) &&
593 (ip->i_d.di_flags &
594 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
595 (vap->va_xflags &
596 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
597 !capable(CAP_LINUX_IMMUTABLE)) {
598 code = XFS_ERROR(EPERM);
599 goto error_return;
600 }
601 }
602
603 /*
604 * Now we can make the changes. Before we join the inode
605 * to the transaction, if XFS_AT_SIZE is set then take care of
606 * the part of the truncation that must be done without the
607 * inode lock. This needs to be done before joining the inode
608 * to the transaction, because the inode cannot be unlocked
609 * once it is a part of the transaction.
610 */
611 if (mask & XFS_AT_SIZE) {
612 code = 0;
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100613 if ((vap->va_size > ip->i_d.di_size) &&
614 (flags & ATTR_NOSIZETOK) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 code = xfs_igrow_start(ip, vap->va_size, credp);
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100616 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 xfs_iunlock(ip, XFS_ILOCK_EXCL);
618 if (!code)
619 code = xfs_itruncate_data(ip, vap->va_size);
620 if (code) {
621 ASSERT(tp == NULL);
622 lock_flags &= ~XFS_ILOCK_EXCL;
623 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
624 goto error_return;
625 }
626 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
627 if ((code = xfs_trans_reserve(tp, 0,
628 XFS_ITRUNCATE_LOG_RES(mp), 0,
629 XFS_TRANS_PERM_LOG_RES,
630 XFS_ITRUNCATE_LOG_COUNT))) {
631 xfs_trans_cancel(tp, 0);
632 if (need_iolock)
633 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
634 return code;
635 }
636 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
637 xfs_ilock(ip, XFS_ILOCK_EXCL);
638 }
639
640 if (tp) {
641 xfs_trans_ijoin(tp, ip, lock_flags);
642 xfs_trans_ihold(tp, ip);
643 }
644
645 /* determine whether mandatory locking mode changes */
646 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
647
648 /*
649 * Truncate file. Must have write permission and not be a directory.
650 */
651 if (mask & XFS_AT_SIZE) {
652 if (vap->va_size > ip->i_d.di_size) {
653 xfs_igrow_finish(tp, ip, vap->va_size,
654 !(flags & ATTR_DMI));
655 } else if ((vap->va_size <= ip->i_d.di_size) ||
656 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
657 /*
658 * signal a sync transaction unless
659 * we're truncating an already unlinked
660 * file on a wsync filesystem
661 */
662 code = xfs_itruncate_finish(&tp, ip,
663 (xfs_fsize_t)vap->va_size,
664 XFS_DATA_FORK,
665 ((ip->i_d.di_nlink != 0 ||
666 !(mp->m_flags & XFS_MOUNT_WSYNC))
667 ? 1 : 0));
668 if (code) {
669 goto abort_return;
670 }
671 }
672 /*
673 * Have to do this even if the file's size doesn't change.
674 */
675 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
676 }
677
678 /*
679 * Change file access modes.
680 */
681 if (mask & XFS_AT_MODE) {
682 ip->i_d.di_mode &= S_IFMT;
683 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
684
685 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
686 timeflags |= XFS_ICHGTIME_CHG;
687 }
688
689 /*
690 * Change file ownership. Must be the owner or privileged.
691 * If the system was configured with the "restricted_chown"
692 * option, the owner is not permitted to give away the file,
693 * and can change the group id only to a group of which he
694 * or she is a member.
695 */
696 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
697 /*
698 * CAP_FSETID overrides the following restrictions:
699 *
700 * The set-user-ID and set-group-ID bits of a file will be
701 * cleared upon successful return from chown()
702 */
703 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
704 !capable(CAP_FSETID)) {
705 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
706 }
707
708 /*
709 * Change the ownerships and register quota modifications
710 * in the transaction.
711 */
712 if (iuid != uid) {
713 if (XFS_IS_UQUOTA_ON(mp)) {
714 ASSERT(mask & XFS_AT_UID);
715 ASSERT(udqp);
716 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
717 &ip->i_udquot, udqp);
718 }
719 ip->i_d.di_uid = uid;
720 }
721 if (igid != gid) {
722 if (XFS_IS_GQUOTA_ON(mp)) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000723 ASSERT(!XFS_IS_PQUOTA_ON(mp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 ASSERT(mask & XFS_AT_GID);
725 ASSERT(gdqp);
726 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
727 &ip->i_gdquot, gdqp);
728 }
729 ip->i_d.di_gid = gid;
730 }
731 if (iprojid != projid) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000732 if (XFS_IS_PQUOTA_ON(mp)) {
733 ASSERT(!XFS_IS_GQUOTA_ON(mp));
734 ASSERT(mask & XFS_AT_PROJID);
735 ASSERT(gdqp);
736 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
737 &ip->i_gdquot, gdqp);
738 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 ip->i_d.di_projid = projid;
740 /*
741 * We may have to rev the inode as well as
742 * the superblock version number since projids didn't
743 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
744 */
745 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
746 xfs_bump_ino_vers2(tp, ip);
747 }
748
749 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
750 timeflags |= XFS_ICHGTIME_CHG;
751 }
752
753
754 /*
755 * Change file access or modified times.
756 */
757 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
758 if (mask & XFS_AT_ATIME) {
759 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
760 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
761 ip->i_update_core = 1;
762 timeflags &= ~XFS_ICHGTIME_ACC;
763 }
764 if (mask & XFS_AT_MTIME) {
765 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
766 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
767 timeflags &= ~XFS_ICHGTIME_MOD;
768 timeflags |= XFS_ICHGTIME_CHG;
769 }
770 if (tp && (flags & ATTR_UTIME))
771 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
772 }
773
774 /*
775 * Change XFS-added attributes.
776 */
777 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
778 if (mask & XFS_AT_EXTSIZE) {
779 /*
780 * Converting bytes to fs blocks.
781 */
782 ip->i_d.di_extsize = vap->va_extsize >>
783 mp->m_sb.sb_blocklog;
784 }
785 if (mask & XFS_AT_XFLAGS) {
786 uint di_flags;
787
788 /* can't set PREALLOC this way, just preserve it */
789 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
790 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
791 di_flags |= XFS_DIFLAG_IMMUTABLE;
792 if (vap->va_xflags & XFS_XFLAG_APPEND)
793 di_flags |= XFS_DIFLAG_APPEND;
794 if (vap->va_xflags & XFS_XFLAG_SYNC)
795 di_flags |= XFS_DIFLAG_SYNC;
796 if (vap->va_xflags & XFS_XFLAG_NOATIME)
797 di_flags |= XFS_DIFLAG_NOATIME;
798 if (vap->va_xflags & XFS_XFLAG_NODUMP)
799 di_flags |= XFS_DIFLAG_NODUMP;
Nathan Scott365ca832005-06-21 15:39:12 +1000800 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
801 di_flags |= XFS_DIFLAG_PROJINHERIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
803 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
804 di_flags |= XFS_DIFLAG_RTINHERIT;
805 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
806 di_flags |= XFS_DIFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100807 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
808 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
809 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
811 di_flags |= XFS_DIFLAG_REALTIME;
812 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
813 } else {
814 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
815 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100816 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
817 di_flags |= XFS_DIFLAG_EXTSIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 }
819 ip->i_d.di_flags = di_flags;
820 }
821 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
822 timeflags |= XFS_ICHGTIME_CHG;
823 }
824
825 /*
826 * Change file inode change time only if XFS_AT_CTIME set
827 * AND we have been called by a DMI function.
828 */
829
830 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
831 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
832 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
833 ip->i_update_core = 1;
834 timeflags &= ~XFS_ICHGTIME_CHG;
835 }
836
837 /*
838 * Send out timestamp changes that need to be set to the
839 * current time. Not done when called by a DMI function.
840 */
841 if (timeflags && !(flags & ATTR_DMI))
842 xfs_ichgtime(ip, timeflags);
843
844 XFS_STATS_INC(xs_ig_attrchg);
845
846 /*
847 * If this is a synchronous mount, make sure that the
848 * transaction goes to disk before returning to the user.
849 * This is slightly sub-optimal in that truncates require
850 * two sync transactions instead of one for wsync filesytems.
851 * One for the truncate and one for the timestamps since we
852 * don't want to change the timestamps unless we're sure the
853 * truncate worked. Truncates are less than 1% of the laddis
854 * mix so this probably isn't worth the trouble to optimize.
855 */
856 code = 0;
857 if (tp) {
858 if (mp->m_flags & XFS_MOUNT_WSYNC)
859 xfs_trans_set_sync(tp);
860
861 code = xfs_trans_commit(tp, commit_flags, NULL);
862 }
863
864 /*
865 * If the (regular) file's mandatory locking mode changed, then
866 * notify the vnode. We do this under the inode lock to prevent
867 * racing calls to vop_vnode_change.
868 */
869 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
870 if (mandlock_before != mandlock_after) {
871 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
872 mandlock_after);
873 }
874
875 xfs_iunlock(ip, lock_flags);
876
877 /*
878 * Release any dquot(s) the inode had kept before chown.
879 */
880 XFS_QM_DQRELE(mp, olddquot1);
881 XFS_QM_DQRELE(mp, olddquot2);
882 XFS_QM_DQRELE(mp, udqp);
883 XFS_QM_DQRELE(mp, gdqp);
884
885 if (code) {
886 return code;
887 }
888
889 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
890 !(flags & ATTR_DMI)) {
891 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
892 NULL, DM_RIGHT_NULL, NULL, NULL,
893 0, 0, AT_DELAY_FLAG(flags));
894 }
895 return 0;
896
897 abort_return:
898 commit_flags |= XFS_TRANS_ABORT;
899 /* FALLTHROUGH */
900 error_return:
901 XFS_QM_DQRELE(mp, udqp);
902 XFS_QM_DQRELE(mp, gdqp);
903 if (tp) {
904 xfs_trans_cancel(tp, commit_flags);
905 }
906 if (lock_flags != 0) {
907 xfs_iunlock(ip, lock_flags);
908 }
909 return code;
910}
911
912
913/*
914 * xfs_access
915 * Null conversion from vnode mode bits to inode mode bits, as in efs.
916 */
917STATIC int
918xfs_access(
919 bhv_desc_t *bdp,
920 int mode,
921 cred_t *credp)
922{
923 xfs_inode_t *ip;
924 int error;
925
926 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
927 (inst_t *)__return_address);
928
929 ip = XFS_BHVTOI(bdp);
930 xfs_ilock(ip, XFS_ILOCK_SHARED);
931 error = xfs_iaccess(ip, mode, credp);
932 xfs_iunlock(ip, XFS_ILOCK_SHARED);
933 return error;
934}
935
936
937/*
938 * xfs_readlink
939 *
940 */
941STATIC int
942xfs_readlink(
943 bhv_desc_t *bdp,
944 uio_t *uiop,
945 int ioflags,
946 cred_t *credp)
947{
948 xfs_inode_t *ip;
949 int count;
950 xfs_off_t offset;
951 int pathlen;
952 vnode_t *vp;
953 int error = 0;
954 xfs_mount_t *mp;
955 int nmaps;
956 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
957 xfs_daddr_t d;
958 int byte_cnt;
959 int n;
960 xfs_buf_t *bp;
961
962 vp = BHV_TO_VNODE(bdp);
963 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
964
965 ip = XFS_BHVTOI(bdp);
966 mp = ip->i_mount;
967
968 if (XFS_FORCED_SHUTDOWN(mp))
969 return XFS_ERROR(EIO);
970
971 xfs_ilock(ip, XFS_ILOCK_SHARED);
972
973 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
974
975 offset = uiop->uio_offset;
976 count = uiop->uio_resid;
977
978 if (offset < 0) {
979 error = XFS_ERROR(EINVAL);
980 goto error_return;
981 }
982 if (count <= 0) {
983 error = 0;
984 goto error_return;
985 }
986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 /*
988 * See if the symlink is stored inline.
989 */
990 pathlen = (int)ip->i_d.di_size;
991
992 if (ip->i_df.if_flags & XFS_IFINLINE) {
993 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
994 }
995 else {
996 /*
997 * Symlink not inline. Call bmap to get it in.
998 */
999 nmaps = SYMLINK_MAPS;
1000
1001 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1002 0, NULL, 0, mval, &nmaps, NULL);
1003
1004 if (error) {
1005 goto error_return;
1006 }
1007
1008 for (n = 0; n < nmaps; n++) {
1009 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1010 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1011 bp = xfs_buf_read(mp->m_ddev_targp, d,
1012 BTOBB(byte_cnt), 0);
1013 error = XFS_BUF_GETERROR(bp);
1014 if (error) {
1015 xfs_ioerror_alert("xfs_readlink",
1016 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1017 xfs_buf_relse(bp);
1018 goto error_return;
1019 }
1020 if (pathlen < byte_cnt)
1021 byte_cnt = pathlen;
1022 pathlen -= byte_cnt;
1023
1024 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1025 xfs_buf_relse (bp);
1026 }
1027
1028 }
1029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 return error;
1033}
1034
1035
1036/*
1037 * xfs_fsync
1038 *
1039 * This is called to sync the inode and its data out to disk.
1040 * We need to hold the I/O lock while flushing the data, and
1041 * the inode lock while flushing the inode. The inode lock CANNOT
1042 * be held while flushing the data, so acquire after we're done
1043 * with that.
1044 */
1045STATIC int
1046xfs_fsync(
1047 bhv_desc_t *bdp,
1048 int flag,
1049 cred_t *credp,
1050 xfs_off_t start,
1051 xfs_off_t stop)
1052{
1053 xfs_inode_t *ip;
1054 xfs_trans_t *tp;
1055 int error;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001056 int log_flushed = 0, changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
1058 vn_trace_entry(BHV_TO_VNODE(bdp),
1059 __FUNCTION__, (inst_t *)__return_address);
1060
1061 ip = XFS_BHVTOI(bdp);
1062
1063 ASSERT(start >= 0 && stop >= -1);
1064
1065 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1066 return XFS_ERROR(EIO);
1067
1068 /*
1069 * We always need to make sure that the required inode state
1070 * is safe on disk. The vnode might be clean but because
1071 * of committed transactions that haven't hit the disk yet.
1072 * Likewise, there could be unflushed non-transactional
1073 * changes to the inode core that have to go to disk.
1074 *
1075 * The following code depends on one assumption: that
1076 * any transaction that changes an inode logs the core
1077 * because it has to change some field in the inode core
1078 * (typically nextents or nblocks). That assumption
1079 * implies that any transactions against an inode will
1080 * catch any non-transactional updates. If inode-altering
1081 * transactions exist that violate this assumption, the
1082 * code breaks. Right now, it figures that if the involved
1083 * update_* field is clear and the inode is unpinned, the
1084 * inode is clean. Either it's been flushed or it's been
1085 * committed and the commit has hit the disk unpinning the inode.
1086 * (Note that xfs_inode_item_format() called at commit clears
1087 * the update_* fields.)
1088 */
1089 xfs_ilock(ip, XFS_ILOCK_SHARED);
1090
1091 /* If we are flushing data then we care about update_size
1092 * being set, otherwise we care about update_core
1093 */
1094 if ((flag & FSYNC_DATA) ?
1095 (ip->i_update_size == 0) :
1096 (ip->i_update_core == 0)) {
1097 /*
1098 * Timestamps/size haven't changed since last inode
1099 * flush or inode transaction commit. That means
1100 * either nothing got written or a transaction
1101 * committed which caught the updates. If the
1102 * latter happened and the transaction hasn't
1103 * hit the disk yet, the inode will be still
1104 * be pinned. If it is, force the log.
1105 */
1106
1107 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1108
1109 if (xfs_ipincount(ip)) {
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001110 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 XFS_LOG_FORCE |
1112 ((flag & FSYNC_WAIT)
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001113 ? XFS_LOG_SYNC : 0),
1114 &log_flushed);
1115 } else {
1116 /*
1117 * If the inode is not pinned and nothing
1118 * has changed we don't need to flush the
1119 * cache.
1120 */
1121 changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 }
1123 error = 0;
1124 } else {
1125 /*
1126 * Kick off a transaction to log the inode
1127 * core to get the updates. Make it
1128 * sync if FSYNC_WAIT is passed in (which
1129 * is done by everybody but specfs). The
1130 * sync transaction will also force the log.
1131 */
1132 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1133 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1134 if ((error = xfs_trans_reserve(tp, 0,
1135 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1136 0, 0, 0))) {
1137 xfs_trans_cancel(tp, 0);
1138 return error;
1139 }
1140 xfs_ilock(ip, XFS_ILOCK_EXCL);
1141
1142 /*
1143 * Note - it's possible that we might have pushed
1144 * ourselves out of the way during trans_reserve
1145 * which would flush the inode. But there's no
1146 * guarantee that the inode buffer has actually
1147 * gone out yet (it's delwri). Plus the buffer
1148 * could be pinned anyway if it's part of an
1149 * inode in another recent transaction. So we
1150 * play it safe and fire off the transaction anyway.
1151 */
1152 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1153 xfs_trans_ihold(tp, ip);
1154 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1155 if (flag & FSYNC_WAIT)
1156 xfs_trans_set_sync(tp);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001157 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
1159 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1160 }
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001161
1162 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1163 /*
1164 * If the log write didn't issue an ordered tag we need
1165 * to flush the disk cache for the data device now.
1166 */
1167 if (!log_flushed)
1168 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1169
1170 /*
1171 * If this inode is on the RT dev we need to flush that
1172 * cache aswell.
1173 */
1174 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1175 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1176 }
1177
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 return error;
1179}
1180
1181/*
1182 * This is called by xfs_inactive to free any blocks beyond eof,
1183 * when the link count isn't zero.
1184 */
1185STATIC int
1186xfs_inactive_free_eofblocks(
1187 xfs_mount_t *mp,
1188 xfs_inode_t *ip)
1189{
1190 xfs_trans_t *tp;
1191 int error;
1192 xfs_fileoff_t end_fsb;
1193 xfs_fileoff_t last_fsb;
1194 xfs_filblks_t map_len;
1195 int nimaps;
1196 xfs_bmbt_irec_t imap;
1197
1198 /*
1199 * Figure out if there are any blocks beyond the end
1200 * of the file. If not, then there is nothing to do.
1201 */
1202 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1203 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1204 map_len = last_fsb - end_fsb;
1205 if (map_len <= 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01001206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
1208 nimaps = 1;
1209 xfs_ilock(ip, XFS_ILOCK_SHARED);
1210 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1211 NULL, 0, &imap, &nimaps, NULL);
1212 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1213
1214 if (!error && (nimaps != 0) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001215 (imap.br_startblock != HOLESTARTBLOCK ||
1216 ip->i_delayed_blks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 /*
1218 * Attach the dquots to the inode up front.
1219 */
1220 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001221 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
1223 /*
1224 * There are blocks after the end of file.
1225 * Free them up now by truncating the file to
1226 * its current size.
1227 */
1228 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1229
1230 /*
1231 * Do the xfs_itruncate_start() call before
1232 * reserving any log space because
1233 * itruncate_start will call into the buffer
1234 * cache and we can't
1235 * do that within a transaction.
1236 */
1237 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1238 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1239 ip->i_d.di_size);
1240
1241 error = xfs_trans_reserve(tp, 0,
1242 XFS_ITRUNCATE_LOG_RES(mp),
1243 0, XFS_TRANS_PERM_LOG_RES,
1244 XFS_ITRUNCATE_LOG_COUNT);
1245 if (error) {
1246 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1247 xfs_trans_cancel(tp, 0);
1248 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001249 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 }
1251
1252 xfs_ilock(ip, XFS_ILOCK_EXCL);
1253 xfs_trans_ijoin(tp, ip,
1254 XFS_IOLOCK_EXCL |
1255 XFS_ILOCK_EXCL);
1256 xfs_trans_ihold(tp, ip);
1257
1258 error = xfs_itruncate_finish(&tp, ip,
1259 ip->i_d.di_size,
1260 XFS_DATA_FORK,
1261 0);
1262 /*
1263 * If we get an error at this point we
1264 * simply don't bother truncating the file.
1265 */
1266 if (error) {
1267 xfs_trans_cancel(tp,
1268 (XFS_TRANS_RELEASE_LOG_RES |
1269 XFS_TRANS_ABORT));
1270 } else {
1271 error = xfs_trans_commit(tp,
1272 XFS_TRANS_RELEASE_LOG_RES,
1273 NULL);
1274 }
1275 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1276 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001277 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278}
1279
1280/*
1281 * Free a symlink that has blocks associated with it.
1282 */
1283STATIC int
1284xfs_inactive_symlink_rmt(
1285 xfs_inode_t *ip,
1286 xfs_trans_t **tpp)
1287{
1288 xfs_buf_t *bp;
1289 int committed;
1290 int done;
1291 int error;
1292 xfs_fsblock_t first_block;
1293 xfs_bmap_free_t free_list;
1294 int i;
1295 xfs_mount_t *mp;
1296 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1297 int nmaps;
1298 xfs_trans_t *ntp;
1299 int size;
1300 xfs_trans_t *tp;
1301
1302 tp = *tpp;
1303 mp = ip->i_mount;
1304 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1305 /*
1306 * We're freeing a symlink that has some
1307 * blocks allocated to it. Free the
1308 * blocks here. We know that we've got
1309 * either 1 or 2 extents and that we can
1310 * free them all in one bunmapi call.
1311 */
1312 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1313 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1314 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1315 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1316 xfs_trans_cancel(tp, 0);
1317 *tpp = NULL;
1318 return error;
1319 }
1320 /*
1321 * Lock the inode, fix the size, and join it to the transaction.
1322 * Hold it so in the normal path, we still have it locked for
1323 * the second transaction. In the error paths we need it
1324 * held so the cancel won't rele it, see below.
1325 */
1326 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1327 size = (int)ip->i_d.di_size;
1328 ip->i_d.di_size = 0;
1329 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1330 xfs_trans_ihold(tp, ip);
1331 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1332 /*
1333 * Find the block(s) so we can inval and unmap them.
1334 */
1335 done = 0;
1336 XFS_BMAP_INIT(&free_list, &first_block);
1337 nmaps = sizeof(mval) / sizeof(mval[0]);
1338 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1339 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1340 &free_list)))
1341 goto error0;
1342 /*
1343 * Invalidate the block(s).
1344 */
1345 for (i = 0; i < nmaps; i++) {
1346 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1347 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1348 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1349 xfs_trans_binval(tp, bp);
1350 }
1351 /*
1352 * Unmap the dead block(s) to the free_list.
1353 */
1354 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1355 &first_block, &free_list, &done)))
1356 goto error1;
1357 ASSERT(done);
1358 /*
1359 * Commit the first transaction. This logs the EFI and the inode.
1360 */
1361 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1362 goto error1;
1363 /*
1364 * The transaction must have been committed, since there were
1365 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1366 * The new tp has the extent freeing and EFDs.
1367 */
1368 ASSERT(committed);
1369 /*
1370 * The first xact was committed, so add the inode to the new one.
1371 * Mark it dirty so it will be logged and moved forward in the log as
1372 * part of every commit.
1373 */
1374 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1375 xfs_trans_ihold(tp, ip);
1376 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1377 /*
1378 * Get a new, empty transaction to return to our caller.
1379 */
1380 ntp = xfs_trans_dup(tp);
1381 /*
1382 * Commit the transaction containing extent freeing and EFD's.
1383 * If we get an error on the commit here or on the reserve below,
1384 * we need to unlock the inode since the new transaction doesn't
1385 * have the inode attached.
1386 */
1387 error = xfs_trans_commit(tp, 0, NULL);
1388 tp = ntp;
1389 if (error) {
1390 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1391 goto error0;
1392 }
1393 /*
1394 * Remove the memory for extent descriptions (just bookkeeping).
1395 */
1396 if (ip->i_df.if_bytes)
1397 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1398 ASSERT(ip->i_df.if_bytes == 0);
1399 /*
1400 * Put an itruncate log reservation in the new transaction
1401 * for our caller.
1402 */
1403 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1404 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1405 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1406 goto error0;
1407 }
1408 /*
1409 * Return with the inode locked but not joined to the transaction.
1410 */
1411 *tpp = tp;
1412 return 0;
1413
1414 error1:
1415 xfs_bmap_cancel(&free_list);
1416 error0:
1417 /*
1418 * Have to come here with the inode locked and either
1419 * (held and in the transaction) or (not in the transaction).
1420 * If the inode isn't held then cancel would iput it, but
1421 * that's wrong since this is inactive and the vnode ref
1422 * count is 0 already.
1423 * Cancel won't do anything to the inode if held, but it still
1424 * needs to be locked until the cancel is done, if it was
1425 * joined to the transaction.
1426 */
1427 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1428 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1429 *tpp = NULL;
1430 return error;
1431
1432}
1433
1434STATIC int
1435xfs_inactive_symlink_local(
1436 xfs_inode_t *ip,
1437 xfs_trans_t **tpp)
1438{
1439 int error;
1440
1441 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1442 /*
1443 * We're freeing a symlink which fit into
1444 * the inode. Just free the memory used
1445 * to hold the old symlink.
1446 */
1447 error = xfs_trans_reserve(*tpp, 0,
1448 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1449 0, XFS_TRANS_PERM_LOG_RES,
1450 XFS_ITRUNCATE_LOG_COUNT);
1451
1452 if (error) {
1453 xfs_trans_cancel(*tpp, 0);
1454 *tpp = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +01001455 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 }
1457 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1458
1459 /*
1460 * Zero length symlinks _can_ exist.
1461 */
1462 if (ip->i_df.if_bytes > 0) {
1463 xfs_idata_realloc(ip,
1464 -(ip->i_df.if_bytes),
1465 XFS_DATA_FORK);
1466 ASSERT(ip->i_df.if_bytes == 0);
1467 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001468 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469}
1470
1471/*
1472 *
1473 */
1474STATIC int
1475xfs_inactive_attrs(
1476 xfs_inode_t *ip,
1477 xfs_trans_t **tpp)
1478{
1479 xfs_trans_t *tp;
1480 int error;
1481 xfs_mount_t *mp;
1482
1483 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1484 tp = *tpp;
1485 mp = ip->i_mount;
1486 ASSERT(ip->i_d.di_forkoff != 0);
1487 xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1488 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1489
1490 error = xfs_attr_inactive(ip);
1491 if (error) {
1492 *tpp = NULL;
1493 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001494 return error; /* goto out */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 }
1496
1497 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1498 error = xfs_trans_reserve(tp, 0,
1499 XFS_IFREE_LOG_RES(mp),
1500 0, XFS_TRANS_PERM_LOG_RES,
1501 XFS_INACTIVE_LOG_COUNT);
1502 if (error) {
1503 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1504 xfs_trans_cancel(tp, 0);
1505 *tpp = NULL;
1506 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001507 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
1509
1510 xfs_ilock(ip, XFS_ILOCK_EXCL);
1511 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1512 xfs_trans_ihold(tp, ip);
1513 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1514
1515 ASSERT(ip->i_d.di_anextents == 0);
1516
1517 *tpp = tp;
Jesper Juhl014c2542006-01-15 02:37:08 +01001518 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519}
1520
1521STATIC int
1522xfs_release(
1523 bhv_desc_t *bdp)
1524{
1525 xfs_inode_t *ip;
1526 vnode_t *vp;
1527 xfs_mount_t *mp;
1528 int error;
1529
1530 vp = BHV_TO_VNODE(bdp);
1531 ip = XFS_BHVTOI(bdp);
1532
Christoph Hellwig0432dab2005-09-02 16:46:51 +10001533 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 return 0;
1535 }
1536
1537 /* If this is a read-only mount, don't do this (would generate I/O) */
1538 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1539 return 0;
1540
1541#ifdef HAVE_REFCACHE
1542 /* If we are in the NFS reference cache then don't do this now */
1543 if (ip->i_refcache)
1544 return 0;
1545#endif
1546
1547 mp = ip->i_mount;
1548
1549 if (ip->i_d.di_nlink != 0) {
1550 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001551 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1552 ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001554 (!(ip->i_d.di_flags &
1555 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001557 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 /* Update linux inode block count after free above */
1559 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1560 ip->i_d.di_nblocks + ip->i_delayed_blks);
1561 }
1562 }
1563
1564 return 0;
1565}
1566
1567/*
1568 * xfs_inactive
1569 *
1570 * This is called when the vnode reference count for the vnode
1571 * goes to zero. If the file has been unlinked, then it must
1572 * now be truncated. Also, we clear all of the read-ahead state
1573 * kept for the inode here since the file is now closed.
1574 */
1575STATIC int
1576xfs_inactive(
1577 bhv_desc_t *bdp,
1578 cred_t *credp)
1579{
1580 xfs_inode_t *ip;
1581 vnode_t *vp;
1582 xfs_bmap_free_t free_list;
1583 xfs_fsblock_t first_block;
1584 int committed;
1585 xfs_trans_t *tp;
1586 xfs_mount_t *mp;
1587 int error;
1588 int truncate;
1589
1590 vp = BHV_TO_VNODE(bdp);
1591 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1592
1593 ip = XFS_BHVTOI(bdp);
1594
1595 /*
1596 * If the inode is already free, then there can be nothing
1597 * to clean up here.
1598 */
1599 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1600 ASSERT(ip->i_df.if_real_bytes == 0);
1601 ASSERT(ip->i_df.if_broot_bytes == 0);
1602 return VN_INACTIVE_CACHE;
1603 }
1604
1605 /*
1606 * Only do a truncate if it's a regular file with
1607 * some actual space in it. It's OK to look at the
1608 * inode's fields without the lock because we're the
1609 * only one with a reference to the inode.
1610 */
1611 truncate = ((ip->i_d.di_nlink == 0) &&
Yingping Lu1661dc82006-01-11 15:29:39 +11001612 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1613 (ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1615
1616 mp = ip->i_mount;
1617
1618 if (ip->i_d.di_nlink == 0 &&
1619 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1620 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1621 }
1622
1623 error = 0;
1624
1625 /* If this is a read-only mount, don't do this (would generate I/O) */
1626 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1627 goto out;
1628
1629 if (ip->i_d.di_nlink != 0) {
1630 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001631 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1632 ip->i_delayed_blks > 0)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001633 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1634 (!(ip->i_d.di_flags &
1635 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1636 (ip->i_delayed_blks != 0)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001638 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 /* Update linux inode block count after free above */
1640 LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1641 ip->i_d.di_nblocks + ip->i_delayed_blks);
1642 }
1643 goto out;
1644 }
1645
1646 ASSERT(ip->i_d.di_nlink == 0);
1647
1648 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001649 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
1651 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1652 if (truncate) {
1653 /*
1654 * Do the xfs_itruncate_start() call before
1655 * reserving any log space because itruncate_start
1656 * will call into the buffer cache and we can't
1657 * do that within a transaction.
1658 */
1659 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1660
1661 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1662
1663 error = xfs_trans_reserve(tp, 0,
1664 XFS_ITRUNCATE_LOG_RES(mp),
1665 0, XFS_TRANS_PERM_LOG_RES,
1666 XFS_ITRUNCATE_LOG_COUNT);
1667 if (error) {
1668 /* Don't call itruncate_cleanup */
1669 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1670 xfs_trans_cancel(tp, 0);
1671 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001672 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 }
1674
1675 xfs_ilock(ip, XFS_ILOCK_EXCL);
1676 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1677 xfs_trans_ihold(tp, ip);
1678
1679 /*
1680 * normally, we have to run xfs_itruncate_finish sync.
1681 * But if filesystem is wsync and we're in the inactive
1682 * path, then we know that nlink == 0, and that the
1683 * xaction that made nlink == 0 is permanently committed
1684 * since xfs_remove runs as a synchronous transaction.
1685 */
1686 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1687 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1688
1689 if (error) {
1690 xfs_trans_cancel(tp,
1691 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1692 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001693 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 }
1695 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1696
1697 /*
1698 * If we get an error while cleaning up a
1699 * symlink we bail out.
1700 */
1701 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1702 xfs_inactive_symlink_rmt(ip, &tp) :
1703 xfs_inactive_symlink_local(ip, &tp);
1704
1705 if (error) {
1706 ASSERT(tp == NULL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001707 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
1709
1710 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1711 xfs_trans_ihold(tp, ip);
1712 } else {
1713 error = xfs_trans_reserve(tp, 0,
1714 XFS_IFREE_LOG_RES(mp),
1715 0, XFS_TRANS_PERM_LOG_RES,
1716 XFS_INACTIVE_LOG_COUNT);
1717 if (error) {
1718 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1719 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001720 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 }
1722
1723 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1724 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1725 xfs_trans_ihold(tp, ip);
1726 }
1727
1728 /*
1729 * If there are attributes associated with the file
1730 * then blow them away now. The code calls a routine
1731 * that recursively deconstructs the attribute fork.
1732 * We need to just commit the current transaction
1733 * because we can't use it for xfs_attr_inactive().
1734 */
1735 if (ip->i_d.di_anextents > 0) {
1736 error = xfs_inactive_attrs(ip, &tp);
1737 /*
1738 * If we got an error, the transaction is already
1739 * cancelled, and the inode is unlocked. Just get out.
1740 */
1741 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001742 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 } else if (ip->i_afp) {
1744 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1745 }
1746
1747 /*
1748 * Free the inode.
1749 */
1750 XFS_BMAP_INIT(&free_list, &first_block);
1751 error = xfs_ifree(tp, ip, &free_list);
1752 if (error) {
1753 /*
1754 * If we fail to free the inode, shut down. The cancel
1755 * might do that, we need to make sure. Otherwise the
1756 * inode might be lost for a long time or forever.
1757 */
1758 if (!XFS_FORCED_SHUTDOWN(mp)) {
1759 cmn_err(CE_NOTE,
1760 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1761 error, mp->m_fsname);
1762 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1763 }
1764 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1765 } else {
1766 /*
1767 * Credit the quota account(s). The inode is gone.
1768 */
1769 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1770
1771 /*
1772 * Just ignore errors at this point. There is
1773 * nothing we can do except to try to keep going.
1774 */
1775 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1776 &committed);
1777 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1778 }
1779 /*
1780 * Release the dquots held by inode, if any.
1781 */
1782 XFS_QM_DQDETACH(mp, ip);
1783
1784 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1785
1786 out:
1787 return VN_INACTIVE_CACHE;
1788}
1789
1790
1791/*
1792 * xfs_lookup
1793 */
1794STATIC int
1795xfs_lookup(
1796 bhv_desc_t *dir_bdp,
1797 vname_t *dentry,
1798 vnode_t **vpp,
1799 int flags,
1800 vnode_t *rdir,
1801 cred_t *credp)
1802{
1803 xfs_inode_t *dp, *ip;
1804 xfs_ino_t e_inum;
1805 int error;
1806 uint lock_mode;
1807 vnode_t *dir_vp;
1808
1809 dir_vp = BHV_TO_VNODE(dir_bdp);
1810 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1811
1812 dp = XFS_BHVTOI(dir_bdp);
1813
1814 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1815 return XFS_ERROR(EIO);
1816
1817 lock_mode = xfs_ilock_map_shared(dp);
1818 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1819 if (!error) {
1820 *vpp = XFS_ITOV(ip);
1821 ITRACE(ip);
1822 }
1823 xfs_iunlock_map_shared(dp, lock_mode);
1824 return error;
1825}
1826
1827
1828/*
1829 * xfs_create (create a new file).
1830 */
1831STATIC int
1832xfs_create(
1833 bhv_desc_t *dir_bdp,
1834 vname_t *dentry,
1835 vattr_t *vap,
1836 vnode_t **vpp,
1837 cred_t *credp)
1838{
1839 char *name = VNAME(dentry);
1840 vnode_t *dir_vp;
1841 xfs_inode_t *dp, *ip;
1842 vnode_t *vp=NULL;
1843 xfs_trans_t *tp;
1844 xfs_mount_t *mp;
1845 xfs_dev_t rdev;
1846 int error;
1847 xfs_bmap_free_t free_list;
1848 xfs_fsblock_t first_block;
1849 boolean_t dp_joined_to_trans;
1850 int dm_event_sent = 0;
1851 uint cancel_flags;
1852 int committed;
1853 xfs_prid_t prid;
1854 struct xfs_dquot *udqp, *gdqp;
1855 uint resblks;
1856 int dm_di_mode;
1857 int namelen;
1858
1859 ASSERT(!*vpp);
1860 dir_vp = BHV_TO_VNODE(dir_bdp);
1861 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1862
1863 dp = XFS_BHVTOI(dir_bdp);
1864 mp = dp->i_mount;
1865
Christoph Hellwig0432dab2005-09-02 16:46:51 +10001866 dm_di_mode = vap->va_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 namelen = VNAMELEN(dentry);
1868
1869 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1870 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1871 dir_vp, DM_RIGHT_NULL, NULL,
1872 DM_RIGHT_NULL, name, NULL,
1873 dm_di_mode, 0, 0);
1874
1875 if (error)
1876 return error;
1877 dm_event_sent = 1;
1878 }
1879
1880 if (XFS_FORCED_SHUTDOWN(mp))
1881 return XFS_ERROR(EIO);
1882
1883 /* Return through std_return after this point. */
1884
1885 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10001886 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1887 prid = dp->i_d.di_projid;
1888 else if (vap->va_mask & XFS_AT_PROJID)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 prid = (xfs_prid_t)vap->va_projid;
1890 else
1891 prid = (xfs_prid_t)dfltprid;
1892
1893 /*
1894 * Make sure that we have allocated dquot(s) on disk.
1895 */
1896 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001897 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1899 if (error)
1900 goto std_return;
1901
1902 ip = NULL;
1903 dp_joined_to_trans = B_FALSE;
1904
1905 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1906 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1907 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1908 /*
1909 * Initially assume that the file does not exist and
1910 * reserve the resources for that case. If that is not
1911 * the case we'll drop the one we have and get a more
1912 * appropriate transaction later.
1913 */
1914 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1915 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1916 if (error == ENOSPC) {
1917 resblks = 0;
1918 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1919 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1920 }
1921 if (error) {
1922 cancel_flags = 0;
1923 dp = NULL;
1924 goto error_return;
1925 }
1926
1927 xfs_ilock(dp, XFS_ILOCK_EXCL);
1928
1929 XFS_BMAP_INIT(&free_list, &first_block);
1930
1931 ASSERT(ip == NULL);
1932
1933 /*
1934 * Reserve disk quota and the inode.
1935 */
1936 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1937 if (error)
1938 goto error_return;
1939
1940 if (resblks == 0 &&
1941 (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1942 goto error_return;
1943 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
Christoph Hellwig0432dab2005-09-02 16:46:51 +10001944 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 rdev, credp, prid, resblks > 0,
1946 &ip, &committed);
1947 if (error) {
1948 if (error == ENOSPC)
1949 goto error_return;
1950 goto abort_return;
1951 }
1952 ITRACE(ip);
1953
1954 /*
1955 * At this point, we've gotten a newly allocated inode.
1956 * It is locked (and joined to the transaction).
1957 */
1958
1959 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1960
1961 /*
1962 * Now we join the directory inode to the transaction.
1963 * We do not do it earlier because xfs_dir_ialloc
1964 * might commit the previous transaction (and release
1965 * all the locks).
1966 */
1967
1968 VN_HOLD(dir_vp);
1969 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1970 dp_joined_to_trans = B_TRUE;
1971
1972 error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
1973 &first_block, &free_list,
1974 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1975 if (error) {
1976 ASSERT(error != ENOSPC);
1977 goto abort_return;
1978 }
1979 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1980 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1981
1982 /*
1983 * If this is a synchronous mount, make sure that the
1984 * create transaction goes to disk before returning to
1985 * the user.
1986 */
1987 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1988 xfs_trans_set_sync(tp);
1989 }
1990
1991 dp->i_gen++;
1992
1993 /*
1994 * Attach the dquot(s) to the inodes and modify them incore.
1995 * These ids of the inode couldn't have changed since the new
1996 * inode has been locked ever since it was created.
1997 */
1998 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
1999
2000 /*
2001 * xfs_trans_commit normally decrements the vnode ref count
2002 * when it unlocks the inode. Since we want to return the
2003 * vnode to the caller, we bump the vnode ref count now.
2004 */
2005 IHOLD(ip);
2006 vp = XFS_ITOV(ip);
2007
2008 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2009 if (error) {
2010 xfs_bmap_cancel(&free_list);
2011 goto abort_rele;
2012 }
2013
2014 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2015 if (error) {
2016 IRELE(ip);
2017 tp = NULL;
2018 goto error_return;
2019 }
2020
2021 XFS_QM_DQRELE(mp, udqp);
2022 XFS_QM_DQRELE(mp, gdqp);
2023
2024 /*
2025 * Propogate the fact that the vnode changed after the
2026 * xfs_inode locks have been released.
2027 */
2028 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2029
2030 *vpp = vp;
2031
2032 /* Fallthrough to std_return with error = 0 */
2033
2034std_return:
2035 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2036 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2037 DM_EVENT_POSTCREATE)) {
2038 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2039 dir_vp, DM_RIGHT_NULL,
2040 *vpp ? vp:NULL,
2041 DM_RIGHT_NULL, name, NULL,
2042 dm_di_mode, error, 0);
2043 }
2044 return error;
2045
2046 abort_return:
2047 cancel_flags |= XFS_TRANS_ABORT;
2048 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Jesper Juhl014c2542006-01-15 02:37:08 +01002050 error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 if (tp != NULL)
2052 xfs_trans_cancel(tp, cancel_flags);
2053
2054 if (!dp_joined_to_trans && (dp != NULL))
2055 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2056 XFS_QM_DQRELE(mp, udqp);
2057 XFS_QM_DQRELE(mp, gdqp);
2058
2059 goto std_return;
2060
2061 abort_rele:
2062 /*
2063 * Wait until after the current transaction is aborted to
2064 * release the inode. This prevents recursive transactions
2065 * and deadlocks from xfs_inactive.
2066 */
2067 cancel_flags |= XFS_TRANS_ABORT;
2068 xfs_trans_cancel(tp, cancel_flags);
2069 IRELE(ip);
2070
2071 XFS_QM_DQRELE(mp, udqp);
2072 XFS_QM_DQRELE(mp, gdqp);
2073
2074 goto std_return;
2075}
2076
2077#ifdef DEBUG
2078/*
2079 * Some counters to see if (and how often) we are hitting some deadlock
2080 * prevention code paths.
2081 */
2082
2083int xfs_rm_locks;
2084int xfs_rm_lock_delays;
2085int xfs_rm_attempts;
2086#endif
2087
2088/*
2089 * The following routine will lock the inodes associated with the
2090 * directory and the named entry in the directory. The locks are
2091 * acquired in increasing inode number.
2092 *
2093 * If the entry is "..", then only the directory is locked. The
2094 * vnode ref count will still include that from the .. entry in
2095 * this case.
2096 *
2097 * There is a deadlock we need to worry about. If the locked directory is
2098 * in the AIL, it might be blocking up the log. The next inode we lock
2099 * could be already locked by another thread waiting for log space (e.g
2100 * a permanent log reservation with a long running transaction (see
2101 * xfs_itruncate_finish)). To solve this, we must check if the directory
2102 * is in the ail and use lock_nowait. If we can't lock, we need to
2103 * drop the inode lock on the directory and try again. xfs_iunlock will
2104 * potentially push the tail if we were holding up the log.
2105 */
2106STATIC int
2107xfs_lock_dir_and_entry(
2108 xfs_inode_t *dp,
2109 vname_t *dentry,
2110 xfs_inode_t *ip) /* inode of entry 'name' */
2111{
2112 int attempts;
2113 xfs_ino_t e_inum;
2114 xfs_inode_t *ips[2];
2115 xfs_log_item_t *lp;
2116
2117#ifdef DEBUG
2118 xfs_rm_locks++;
2119#endif
2120 attempts = 0;
2121
2122again:
2123 xfs_ilock(dp, XFS_ILOCK_EXCL);
2124
2125 e_inum = ip->i_ino;
2126
2127 ITRACE(ip);
2128
2129 /*
2130 * We want to lock in increasing inum. Since we've already
2131 * acquired the lock on the directory, we may need to release
2132 * if if the inum of the entry turns out to be less.
2133 */
2134 if (e_inum > dp->i_ino) {
2135 /*
2136 * We are already in the right order, so just
2137 * lock on the inode of the entry.
2138 * We need to use nowait if dp is in the AIL.
2139 */
2140
2141 lp = (xfs_log_item_t *)dp->i_itemp;
2142 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2143 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2144 attempts++;
2145#ifdef DEBUG
2146 xfs_rm_attempts++;
2147#endif
2148
2149 /*
2150 * Unlock dp and try again.
2151 * xfs_iunlock will try to push the tail
2152 * if the inode is in the AIL.
2153 */
2154
2155 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2156
2157 if ((attempts % 5) == 0) {
2158 delay(1); /* Don't just spin the CPU */
2159#ifdef DEBUG
2160 xfs_rm_lock_delays++;
2161#endif
2162 }
2163 goto again;
2164 }
2165 } else {
2166 xfs_ilock(ip, XFS_ILOCK_EXCL);
2167 }
2168 } else if (e_inum < dp->i_ino) {
2169 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2170
2171 ips[0] = ip;
2172 ips[1] = dp;
2173 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2174 }
2175 /* else e_inum == dp->i_ino */
2176 /* This can happen if we're asked to lock /x/..
2177 * the entry is "..", which is also the parent directory.
2178 */
2179
2180 return 0;
2181}
2182
2183#ifdef DEBUG
2184int xfs_locked_n;
2185int xfs_small_retries;
2186int xfs_middle_retries;
2187int xfs_lots_retries;
2188int xfs_lock_delays;
2189#endif
2190
2191/*
2192 * The following routine will lock n inodes in exclusive mode.
2193 * We assume the caller calls us with the inodes in i_ino order.
2194 *
2195 * We need to detect deadlock where an inode that we lock
2196 * is in the AIL and we start waiting for another inode that is locked
2197 * by a thread in a long running transaction (such as truncate). This can
2198 * result in deadlock since the long running trans might need to wait
2199 * for the inode we just locked in order to push the tail and free space
2200 * in the log.
2201 */
2202void
2203xfs_lock_inodes(
2204 xfs_inode_t **ips,
2205 int inodes,
2206 int first_locked,
2207 uint lock_mode)
2208{
2209 int attempts = 0, i, j, try_lock;
2210 xfs_log_item_t *lp;
2211
2212 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2213
2214 if (first_locked) {
2215 try_lock = 1;
2216 i = 1;
2217 } else {
2218 try_lock = 0;
2219 i = 0;
2220 }
2221
2222again:
2223 for (; i < inodes; i++) {
2224 ASSERT(ips[i]);
2225
2226 if (i && (ips[i] == ips[i-1])) /* Already locked */
2227 continue;
2228
2229 /*
2230 * If try_lock is not set yet, make sure all locked inodes
2231 * are not in the AIL.
2232 * If any are, set try_lock to be used later.
2233 */
2234
2235 if (!try_lock) {
2236 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2237 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2238 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2239 try_lock++;
2240 }
2241 }
2242 }
2243
2244 /*
2245 * If any of the previous locks we have locked is in the AIL,
2246 * we must TRY to get the second and subsequent locks. If
2247 * we can't get any, we must release all we have
2248 * and try again.
2249 */
2250
2251 if (try_lock) {
2252 /* try_lock must be 0 if i is 0. */
2253 /*
2254 * try_lock means we have an inode locked
2255 * that is in the AIL.
2256 */
2257 ASSERT(i != 0);
2258 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2259 attempts++;
2260
2261 /*
2262 * Unlock all previous guys and try again.
2263 * xfs_iunlock will try to push the tail
2264 * if the inode is in the AIL.
2265 */
2266
2267 for(j = i - 1; j >= 0; j--) {
2268
2269 /*
2270 * Check to see if we've already
2271 * unlocked this one.
2272 * Not the first one going back,
2273 * and the inode ptr is the same.
2274 */
2275 if ((j != (i - 1)) && ips[j] ==
2276 ips[j+1])
2277 continue;
2278
2279 xfs_iunlock(ips[j], lock_mode);
2280 }
2281
2282 if ((attempts % 5) == 0) {
2283 delay(1); /* Don't just spin the CPU */
2284#ifdef DEBUG
2285 xfs_lock_delays++;
2286#endif
2287 }
2288 i = 0;
2289 try_lock = 0;
2290 goto again;
2291 }
2292 } else {
2293 xfs_ilock(ips[i], lock_mode);
2294 }
2295 }
2296
2297#ifdef DEBUG
2298 if (attempts) {
2299 if (attempts < 5) xfs_small_retries++;
2300 else if (attempts < 100) xfs_middle_retries++;
2301 else xfs_lots_retries++;
2302 } else {
2303 xfs_locked_n++;
2304 }
2305#endif
2306}
2307
2308#ifdef DEBUG
2309#define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2310int remove_which_error_return = 0;
2311#else /* ! DEBUG */
2312#define REMOVE_DEBUG_TRACE(x)
2313#endif /* ! DEBUG */
2314
2315
2316/*
2317 * xfs_remove
2318 *
2319 */
2320STATIC int
2321xfs_remove(
2322 bhv_desc_t *dir_bdp,
2323 vname_t *dentry,
2324 cred_t *credp)
2325{
2326 vnode_t *dir_vp;
2327 char *name = VNAME(dentry);
2328 xfs_inode_t *dp, *ip;
2329 xfs_trans_t *tp = NULL;
2330 xfs_mount_t *mp;
2331 int error = 0;
2332 xfs_bmap_free_t free_list;
2333 xfs_fsblock_t first_block;
2334 int cancel_flags;
2335 int committed;
2336 int dm_di_mode = 0;
2337 int link_zero;
2338 uint resblks;
2339 int namelen;
2340
2341 dir_vp = BHV_TO_VNODE(dir_bdp);
2342 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2343
2344 dp = XFS_BHVTOI(dir_bdp);
2345 mp = dp->i_mount;
2346
2347 if (XFS_FORCED_SHUTDOWN(mp))
2348 return XFS_ERROR(EIO);
2349
2350 namelen = VNAMELEN(dentry);
2351
2352 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2353 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2354 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2355 name, NULL, 0, 0, 0);
2356 if (error)
2357 return error;
2358 }
2359
2360 /* From this point on, return through std_return */
2361 ip = NULL;
2362
2363 /*
2364 * We need to get a reference to ip before we get our log
2365 * reservation. The reason for this is that we cannot call
2366 * xfs_iget for an inode for which we do not have a reference
2367 * once we've acquired a log reservation. This is because the
2368 * inode we are trying to get might be in xfs_inactive going
2369 * for a log reservation. Since we'll have to wait for the
2370 * inactive code to complete before returning from xfs_iget,
2371 * we need to make sure that we don't have log space reserved
2372 * when we call xfs_iget. Instead we get an unlocked referece
2373 * to the inode before getting our log reservation.
2374 */
2375 error = xfs_get_dir_entry(dentry, &ip);
2376 if (error) {
2377 REMOVE_DEBUG_TRACE(__LINE__);
2378 goto std_return;
2379 }
2380
2381 dm_di_mode = ip->i_d.di_mode;
2382
2383 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2384
2385 ITRACE(ip);
2386
2387 error = XFS_QM_DQATTACH(mp, dp, 0);
2388 if (!error && dp != ip)
2389 error = XFS_QM_DQATTACH(mp, ip, 0);
2390 if (error) {
2391 REMOVE_DEBUG_TRACE(__LINE__);
2392 IRELE(ip);
2393 goto std_return;
2394 }
2395
2396 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2397 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2398 /*
2399 * We try to get the real space reservation first,
2400 * allowing for directory btree deletion(s) implying
2401 * possible bmap insert(s). If we can't get the space
2402 * reservation then we use 0 instead, and avoid the bmap
2403 * btree insert(s) in the directory code by, if the bmap
2404 * insert tries to happen, instead trimming the LAST
2405 * block from the directory.
2406 */
2407 resblks = XFS_REMOVE_SPACE_RES(mp);
2408 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2409 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2410 if (error == ENOSPC) {
2411 resblks = 0;
2412 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2413 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2414 }
2415 if (error) {
2416 ASSERT(error != ENOSPC);
2417 REMOVE_DEBUG_TRACE(__LINE__);
2418 xfs_trans_cancel(tp, 0);
2419 IRELE(ip);
2420 return error;
2421 }
2422
2423 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2424 if (error) {
2425 REMOVE_DEBUG_TRACE(__LINE__);
2426 xfs_trans_cancel(tp, cancel_flags);
2427 IRELE(ip);
2428 goto std_return;
2429 }
2430
2431 /*
2432 * At this point, we've gotten both the directory and the entry
2433 * inodes locked.
2434 */
2435 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2436 if (dp != ip) {
2437 /*
2438 * Increment vnode ref count only in this case since
2439 * there's an extra vnode reference in the case where
2440 * dp == ip.
2441 */
2442 IHOLD(dp);
2443 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2444 }
2445
2446 /*
2447 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2448 */
2449 XFS_BMAP_INIT(&free_list, &first_block);
2450 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2451 &first_block, &free_list, 0);
2452 if (error) {
2453 ASSERT(error != ENOENT);
2454 REMOVE_DEBUG_TRACE(__LINE__);
2455 goto error1;
2456 }
2457 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2458
2459 dp->i_gen++;
2460 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2461
2462 error = xfs_droplink(tp, ip);
2463 if (error) {
2464 REMOVE_DEBUG_TRACE(__LINE__);
2465 goto error1;
2466 }
2467
2468 /* Determine if this is the last link while
2469 * we are in the transaction.
2470 */
2471 link_zero = (ip)->i_d.di_nlink==0;
2472
2473 /*
2474 * Take an extra ref on the inode so that it doesn't
2475 * go to xfs_inactive() from within the commit.
2476 */
2477 IHOLD(ip);
2478
2479 /*
2480 * If this is a synchronous mount, make sure that the
2481 * remove transaction goes to disk before returning to
2482 * the user.
2483 */
2484 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2485 xfs_trans_set_sync(tp);
2486 }
2487
2488 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2489 if (error) {
2490 REMOVE_DEBUG_TRACE(__LINE__);
2491 goto error_rele;
2492 }
2493
2494 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2495 if (error) {
2496 IRELE(ip);
2497 goto std_return;
2498 }
2499
2500 /*
2501 * Before we drop our extra reference to the inode, purge it
2502 * from the refcache if it is there. By waiting until afterwards
2503 * to do the IRELE, we ensure that we won't go inactive in the
2504 * xfs_refcache_purge_ip routine (although that would be OK).
2505 */
2506 xfs_refcache_purge_ip(ip);
2507
2508 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2509
2510 /*
2511 * Let interposed file systems know about removed links.
2512 */
2513 VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2514
2515 IRELE(ip);
2516
2517/* Fall through to std_return with error = 0 */
2518 std_return:
2519 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2520 DM_EVENT_POSTREMOVE)) {
2521 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2522 dir_vp, DM_RIGHT_NULL,
2523 NULL, DM_RIGHT_NULL,
2524 name, NULL, dm_di_mode, error, 0);
2525 }
2526 return error;
2527
2528 error1:
2529 xfs_bmap_cancel(&free_list);
2530 cancel_flags |= XFS_TRANS_ABORT;
2531 xfs_trans_cancel(tp, cancel_flags);
2532 goto std_return;
2533
2534 error_rele:
2535 /*
2536 * In this case make sure to not release the inode until after
2537 * the current transaction is aborted. Releasing it beforehand
2538 * can cause us to go to xfs_inactive and start a recursive
2539 * transaction which can easily deadlock with the current one.
2540 */
2541 xfs_bmap_cancel(&free_list);
2542 cancel_flags |= XFS_TRANS_ABORT;
2543 xfs_trans_cancel(tp, cancel_flags);
2544
2545 /*
2546 * Before we drop our extra reference to the inode, purge it
2547 * from the refcache if it is there. By waiting until afterwards
2548 * to do the IRELE, we ensure that we won't go inactive in the
2549 * xfs_refcache_purge_ip routine (although that would be OK).
2550 */
2551 xfs_refcache_purge_ip(ip);
2552
2553 IRELE(ip);
2554
2555 goto std_return;
2556}
2557
2558
2559/*
2560 * xfs_link
2561 *
2562 */
2563STATIC int
2564xfs_link(
2565 bhv_desc_t *target_dir_bdp,
2566 vnode_t *src_vp,
2567 vname_t *dentry,
2568 cred_t *credp)
2569{
2570 xfs_inode_t *tdp, *sip;
2571 xfs_trans_t *tp;
2572 xfs_mount_t *mp;
2573 xfs_inode_t *ips[2];
2574 int error;
2575 xfs_bmap_free_t free_list;
2576 xfs_fsblock_t first_block;
2577 int cancel_flags;
2578 int committed;
2579 vnode_t *target_dir_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 int resblks;
2581 char *target_name = VNAME(dentry);
2582 int target_namelen;
2583
2584 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2585 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2586 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2587
2588 target_namelen = VNAMELEN(dentry);
Christoph Hellwig0432dab2005-09-02 16:46:51 +10002589 if (VN_ISDIR(src_vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 return XFS_ERROR(EPERM);
2591
Christoph Hellwig75e17b32006-01-11 20:58:44 +11002592 sip = xfs_vtoi(src_vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 tdp = XFS_BHVTOI(target_dir_bdp);
2594 mp = tdp->i_mount;
2595 if (XFS_FORCED_SHUTDOWN(mp))
2596 return XFS_ERROR(EIO);
2597
2598 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2599 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2600 target_dir_vp, DM_RIGHT_NULL,
2601 src_vp, DM_RIGHT_NULL,
2602 target_name, NULL, 0, 0, 0);
2603 if (error)
2604 return error;
2605 }
2606
2607 /* Return through std_return after this point. */
2608
2609 error = XFS_QM_DQATTACH(mp, sip, 0);
2610 if (!error && sip != tdp)
2611 error = XFS_QM_DQATTACH(mp, tdp, 0);
2612 if (error)
2613 goto std_return;
2614
2615 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2616 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2617 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2618 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2619 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2620 if (error == ENOSPC) {
2621 resblks = 0;
2622 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2623 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2624 }
2625 if (error) {
2626 cancel_flags = 0;
2627 goto error_return;
2628 }
2629
2630 if (sip->i_ino < tdp->i_ino) {
2631 ips[0] = sip;
2632 ips[1] = tdp;
2633 } else {
2634 ips[0] = tdp;
2635 ips[1] = sip;
2636 }
2637
2638 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2639
2640 /*
2641 * Increment vnode ref counts since xfs_trans_commit &
2642 * xfs_trans_cancel will both unlock the inodes and
2643 * decrement the associated ref counts.
2644 */
2645 VN_HOLD(src_vp);
2646 VN_HOLD(target_dir_vp);
2647 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2648 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2649
2650 /*
2651 * If the source has too many links, we can't make any more to it.
2652 */
2653 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2654 error = XFS_ERROR(EMLINK);
2655 goto error_return;
2656 }
2657
Nathan Scott365ca832005-06-21 15:39:12 +10002658 /*
2659 * If we are using project inheritance, we only allow hard link
2660 * creation in our tree when the project IDs are the same; else
2661 * the tree quota mechanism could be circumvented.
2662 */
2663 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2664 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
2665 error = XFS_ERROR(EPERM);
2666 goto error_return;
2667 }
2668
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 if (resblks == 0 &&
2670 (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2671 target_namelen)))
2672 goto error_return;
2673
2674 XFS_BMAP_INIT(&free_list, &first_block);
2675
2676 error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2677 sip->i_ino, &first_block, &free_list,
2678 resblks);
2679 if (error)
2680 goto abort_return;
2681 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2682 tdp->i_gen++;
2683 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2684
2685 error = xfs_bumplink(tp, sip);
2686 if (error) {
2687 goto abort_return;
2688 }
2689
2690 /*
2691 * If this is a synchronous mount, make sure that the
2692 * link transaction goes to disk before returning to
2693 * the user.
2694 */
2695 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2696 xfs_trans_set_sync(tp);
2697 }
2698
2699 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2700 if (error) {
2701 xfs_bmap_cancel(&free_list);
2702 goto abort_return;
2703 }
2704
2705 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2706 if (error) {
2707 goto std_return;
2708 }
2709
2710 /* Fall through to std_return with error = 0. */
2711std_return:
2712 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2713 DM_EVENT_POSTLINK)) {
2714 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2715 target_dir_vp, DM_RIGHT_NULL,
2716 src_vp, DM_RIGHT_NULL,
2717 target_name, NULL, 0, error, 0);
2718 }
2719 return error;
2720
2721 abort_return:
2722 cancel_flags |= XFS_TRANS_ABORT;
2723 /* FALLTHROUGH */
Jesper Juhl014c2542006-01-15 02:37:08 +01002724
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 error_return:
2726 xfs_trans_cancel(tp, cancel_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 goto std_return;
2728}
2729/*
2730 * xfs_mkdir
2731 *
2732 */
2733STATIC int
2734xfs_mkdir(
2735 bhv_desc_t *dir_bdp,
2736 vname_t *dentry,
2737 vattr_t *vap,
2738 vnode_t **vpp,
2739 cred_t *credp)
2740{
2741 char *dir_name = VNAME(dentry);
2742 xfs_inode_t *dp;
2743 xfs_inode_t *cdp; /* inode of created dir */
2744 vnode_t *cvp; /* vnode of created dir */
2745 xfs_trans_t *tp;
2746 xfs_mount_t *mp;
2747 int cancel_flags;
2748 int error;
2749 int committed;
2750 xfs_bmap_free_t free_list;
2751 xfs_fsblock_t first_block;
2752 vnode_t *dir_vp;
2753 boolean_t dp_joined_to_trans;
2754 boolean_t created = B_FALSE;
2755 int dm_event_sent = 0;
2756 xfs_prid_t prid;
2757 struct xfs_dquot *udqp, *gdqp;
2758 uint resblks;
2759 int dm_di_mode;
2760 int dir_namelen;
2761
2762 dir_vp = BHV_TO_VNODE(dir_bdp);
2763 dp = XFS_BHVTOI(dir_bdp);
2764 mp = dp->i_mount;
2765
2766 if (XFS_FORCED_SHUTDOWN(mp))
2767 return XFS_ERROR(EIO);
2768
2769 dir_namelen = VNAMELEN(dentry);
2770
2771 tp = NULL;
2772 dp_joined_to_trans = B_FALSE;
Christoph Hellwig0432dab2005-09-02 16:46:51 +10002773 dm_di_mode = vap->va_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
2775 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2776 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2777 dir_vp, DM_RIGHT_NULL, NULL,
2778 DM_RIGHT_NULL, dir_name, NULL,
2779 dm_di_mode, 0, 0);
2780 if (error)
2781 return error;
2782 dm_event_sent = 1;
2783 }
2784
2785 /* Return through std_return after this point. */
2786
2787 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2788
2789 mp = dp->i_mount;
2790 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10002791 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2792 prid = dp->i_d.di_projid;
2793 else if (vap->va_mask & XFS_AT_PROJID)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 prid = (xfs_prid_t)vap->va_projid;
2795 else
2796 prid = (xfs_prid_t)dfltprid;
2797
2798 /*
2799 * Make sure that we have allocated dquot(s) on disk.
2800 */
2801 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002802 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2804 if (error)
2805 goto std_return;
2806
2807 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2808 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2809 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2810 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2811 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2812 if (error == ENOSPC) {
2813 resblks = 0;
2814 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2815 XFS_TRANS_PERM_LOG_RES,
2816 XFS_MKDIR_LOG_COUNT);
2817 }
2818 if (error) {
2819 cancel_flags = 0;
2820 dp = NULL;
2821 goto error_return;
2822 }
2823
2824 xfs_ilock(dp, XFS_ILOCK_EXCL);
2825
2826 /*
2827 * Check for directory link count overflow.
2828 */
2829 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2830 error = XFS_ERROR(EMLINK);
2831 goto error_return;
2832 }
2833
2834 /*
2835 * Reserve disk quota and the inode.
2836 */
2837 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2838 if (error)
2839 goto error_return;
2840
2841 if (resblks == 0 &&
2842 (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2843 goto error_return;
2844 /*
2845 * create the directory inode.
2846 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +10002847 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 0, credp, prid, resblks > 0,
2849 &cdp, NULL);
2850 if (error) {
2851 if (error == ENOSPC)
2852 goto error_return;
2853 goto abort_return;
2854 }
2855 ITRACE(cdp);
2856
2857 /*
2858 * Now we add the directory inode to the transaction.
2859 * We waited until now since xfs_dir_ialloc might start
2860 * a new transaction. Had we joined the transaction
2861 * earlier, the locks might have gotten released.
2862 */
2863 VN_HOLD(dir_vp);
2864 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2865 dp_joined_to_trans = B_TRUE;
2866
2867 XFS_BMAP_INIT(&free_list, &first_block);
2868
2869 error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2870 cdp->i_ino, &first_block, &free_list,
2871 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2872 if (error) {
2873 ASSERT(error != ENOSPC);
2874 goto error1;
2875 }
2876 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2877
2878 /*
2879 * Bump the in memory version number of the parent directory
2880 * so that other processes accessing it will recognize that
2881 * the directory has changed.
2882 */
2883 dp->i_gen++;
2884
2885 error = XFS_DIR_INIT(mp, tp, cdp, dp);
2886 if (error) {
2887 goto error2;
2888 }
2889
2890 cdp->i_gen = 1;
2891 error = xfs_bumplink(tp, dp);
2892 if (error) {
2893 goto error2;
2894 }
2895
2896 cvp = XFS_ITOV(cdp);
2897
2898 created = B_TRUE;
2899
2900 *vpp = cvp;
2901 IHOLD(cdp);
2902
2903 /*
2904 * Attach the dquots to the new inode and modify the icount incore.
2905 */
2906 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2907
2908 /*
2909 * If this is a synchronous mount, make sure that the
2910 * mkdir transaction goes to disk before returning to
2911 * the user.
2912 */
2913 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2914 xfs_trans_set_sync(tp);
2915 }
2916
2917 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2918 if (error) {
2919 IRELE(cdp);
2920 goto error2;
2921 }
2922
2923 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2924 XFS_QM_DQRELE(mp, udqp);
2925 XFS_QM_DQRELE(mp, gdqp);
2926 if (error) {
2927 IRELE(cdp);
2928 }
2929
2930 /* Fall through to std_return with error = 0 or errno from
2931 * xfs_trans_commit. */
2932
2933std_return:
2934 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2935 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2936 DM_EVENT_POSTCREATE)) {
2937 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2938 dir_vp, DM_RIGHT_NULL,
2939 created ? XFS_ITOV(cdp):NULL,
2940 DM_RIGHT_NULL,
2941 dir_name, NULL,
2942 dm_di_mode, error, 0);
2943 }
2944 return error;
2945
2946 error2:
2947 error1:
2948 xfs_bmap_cancel(&free_list);
2949 abort_return:
2950 cancel_flags |= XFS_TRANS_ABORT;
2951 error_return:
2952 xfs_trans_cancel(tp, cancel_flags);
2953 XFS_QM_DQRELE(mp, udqp);
2954 XFS_QM_DQRELE(mp, gdqp);
2955
2956 if (!dp_joined_to_trans && (dp != NULL)) {
2957 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2958 }
2959
2960 goto std_return;
2961}
2962
2963
2964/*
2965 * xfs_rmdir
2966 *
2967 */
2968STATIC int
2969xfs_rmdir(
2970 bhv_desc_t *dir_bdp,
2971 vname_t *dentry,
2972 cred_t *credp)
2973{
2974 char *name = VNAME(dentry);
2975 xfs_inode_t *dp;
2976 xfs_inode_t *cdp; /* child directory */
2977 xfs_trans_t *tp;
2978 xfs_mount_t *mp;
2979 int error;
2980 xfs_bmap_free_t free_list;
2981 xfs_fsblock_t first_block;
2982 int cancel_flags;
2983 int committed;
2984 vnode_t *dir_vp;
2985 int dm_di_mode = 0;
2986 int last_cdp_link;
2987 int namelen;
2988 uint resblks;
2989
2990 dir_vp = BHV_TO_VNODE(dir_bdp);
2991 dp = XFS_BHVTOI(dir_bdp);
2992 mp = dp->i_mount;
2993
2994 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2995
2996 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
2997 return XFS_ERROR(EIO);
2998 namelen = VNAMELEN(dentry);
2999
3000 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3001 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3002 dir_vp, DM_RIGHT_NULL,
3003 NULL, DM_RIGHT_NULL,
3004 name, NULL, 0, 0, 0);
3005 if (error)
3006 return XFS_ERROR(error);
3007 }
3008
3009 /* Return through std_return after this point. */
3010
3011 cdp = NULL;
3012
3013 /*
3014 * We need to get a reference to cdp before we get our log
3015 * reservation. The reason for this is that we cannot call
3016 * xfs_iget for an inode for which we do not have a reference
3017 * once we've acquired a log reservation. This is because the
3018 * inode we are trying to get might be in xfs_inactive going
3019 * for a log reservation. Since we'll have to wait for the
3020 * inactive code to complete before returning from xfs_iget,
3021 * we need to make sure that we don't have log space reserved
3022 * when we call xfs_iget. Instead we get an unlocked referece
3023 * to the inode before getting our log reservation.
3024 */
3025 error = xfs_get_dir_entry(dentry, &cdp);
3026 if (error) {
3027 REMOVE_DEBUG_TRACE(__LINE__);
3028 goto std_return;
3029 }
3030 mp = dp->i_mount;
3031 dm_di_mode = cdp->i_d.di_mode;
3032
3033 /*
3034 * Get the dquots for the inodes.
3035 */
3036 error = XFS_QM_DQATTACH(mp, dp, 0);
3037 if (!error && dp != cdp)
3038 error = XFS_QM_DQATTACH(mp, cdp, 0);
3039 if (error) {
3040 IRELE(cdp);
3041 REMOVE_DEBUG_TRACE(__LINE__);
3042 goto std_return;
3043 }
3044
3045 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3046 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3047 /*
3048 * We try to get the real space reservation first,
3049 * allowing for directory btree deletion(s) implying
3050 * possible bmap insert(s). If we can't get the space
3051 * reservation then we use 0 instead, and avoid the bmap
3052 * btree insert(s) in the directory code by, if the bmap
3053 * insert tries to happen, instead trimming the LAST
3054 * block from the directory.
3055 */
3056 resblks = XFS_REMOVE_SPACE_RES(mp);
3057 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3058 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3059 if (error == ENOSPC) {
3060 resblks = 0;
3061 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3062 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3063 }
3064 if (error) {
3065 ASSERT(error != ENOSPC);
3066 cancel_flags = 0;
3067 IRELE(cdp);
3068 goto error_return;
3069 }
3070 XFS_BMAP_INIT(&free_list, &first_block);
3071
3072 /*
3073 * Now lock the child directory inode and the parent directory
3074 * inode in the proper order. This will take care of validating
3075 * that the directory entry for the child directory inode has
3076 * not changed while we were obtaining a log reservation.
3077 */
3078 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3079 if (error) {
3080 xfs_trans_cancel(tp, cancel_flags);
3081 IRELE(cdp);
3082 goto std_return;
3083 }
3084
3085 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3086 if (dp != cdp) {
3087 /*
3088 * Only increment the parent directory vnode count if
3089 * we didn't bump it in looking up cdp. The only time
3090 * we don't bump it is when we're looking up ".".
3091 */
3092 VN_HOLD(dir_vp);
3093 }
3094
3095 ITRACE(cdp);
3096 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3097
3098 ASSERT(cdp->i_d.di_nlink >= 2);
3099 if (cdp->i_d.di_nlink != 2) {
3100 error = XFS_ERROR(ENOTEMPTY);
3101 goto error_return;
3102 }
3103 if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3104 error = XFS_ERROR(ENOTEMPTY);
3105 goto error_return;
3106 }
3107
3108 error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3109 &first_block, &free_list, resblks);
3110 if (error) {
3111 goto error1;
3112 }
3113
3114 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3115
3116 /*
3117 * Bump the in memory generation count on the parent
3118 * directory so that other can know that it has changed.
3119 */
3120 dp->i_gen++;
3121
3122 /*
3123 * Drop the link from cdp's "..".
3124 */
3125 error = xfs_droplink(tp, dp);
3126 if (error) {
3127 goto error1;
3128 }
3129
3130 /*
3131 * Drop the link from dp to cdp.
3132 */
3133 error = xfs_droplink(tp, cdp);
3134 if (error) {
3135 goto error1;
3136 }
3137
3138 /*
3139 * Drop the "." link from cdp to self.
3140 */
3141 error = xfs_droplink(tp, cdp);
3142 if (error) {
3143 goto error1;
3144 }
3145
3146 /* Determine these before committing transaction */
3147 last_cdp_link = (cdp)->i_d.di_nlink==0;
3148
3149 /*
3150 * Take an extra ref on the child vnode so that it
3151 * does not go to xfs_inactive() from within the commit.
3152 */
3153 IHOLD(cdp);
3154
3155 /*
3156 * If this is a synchronous mount, make sure that the
3157 * rmdir transaction goes to disk before returning to
3158 * the user.
3159 */
3160 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3161 xfs_trans_set_sync(tp);
3162 }
3163
3164 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3165 if (error) {
3166 xfs_bmap_cancel(&free_list);
3167 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3168 XFS_TRANS_ABORT));
3169 IRELE(cdp);
3170 goto std_return;
3171 }
3172
3173 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3174 if (error) {
3175 IRELE(cdp);
3176 goto std_return;
3177 }
3178
3179
3180 /*
3181 * Let interposed file systems know about removed links.
3182 */
3183 VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3184
3185 IRELE(cdp);
3186
3187 /* Fall through to std_return with error = 0 or the errno
3188 * from xfs_trans_commit. */
3189std_return:
3190 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3191 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3192 dir_vp, DM_RIGHT_NULL,
3193 NULL, DM_RIGHT_NULL,
3194 name, NULL, dm_di_mode,
3195 error, 0);
3196 }
3197 return error;
3198
Jesper Juhl014c2542006-01-15 02:37:08 +01003199error1:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200 xfs_bmap_cancel(&free_list);
3201 cancel_flags |= XFS_TRANS_ABORT;
Jesper Juhl014c2542006-01-15 02:37:08 +01003202 /* FALLTHROUGH */
3203
3204error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205 xfs_trans_cancel(tp, cancel_flags);
3206 goto std_return;
3207}
3208
3209
3210/*
3211 * xfs_readdir
3212 *
3213 * Read dp's entries starting at uiop->uio_offset and translate them into
3214 * bufsize bytes worth of struct dirents starting at bufbase.
3215 */
3216STATIC int
3217xfs_readdir(
3218 bhv_desc_t *dir_bdp,
3219 uio_t *uiop,
3220 cred_t *credp,
3221 int *eofp)
3222{
3223 xfs_inode_t *dp;
3224 xfs_trans_t *tp = NULL;
3225 int error = 0;
3226 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227
3228 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3229 (inst_t *)__return_address);
3230 dp = XFS_BHVTOI(dir_bdp);
3231
3232 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3233 return XFS_ERROR(EIO);
3234 }
3235
3236 lock_mode = xfs_ilock_map_shared(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238 xfs_iunlock_map_shared(dp, lock_mode);
3239 return error;
3240}
3241
3242
3243/*
3244 * xfs_symlink
3245 *
3246 */
3247STATIC int
3248xfs_symlink(
3249 bhv_desc_t *dir_bdp,
3250 vname_t *dentry,
3251 vattr_t *vap,
3252 char *target_path,
3253 vnode_t **vpp,
3254 cred_t *credp)
3255{
3256 xfs_trans_t *tp;
3257 xfs_mount_t *mp;
3258 xfs_inode_t *dp;
3259 xfs_inode_t *ip;
3260 int error;
3261 int pathlen;
3262 xfs_bmap_free_t free_list;
3263 xfs_fsblock_t first_block;
3264 boolean_t dp_joined_to_trans;
3265 vnode_t *dir_vp;
3266 uint cancel_flags;
3267 int committed;
3268 xfs_fileoff_t first_fsb;
3269 xfs_filblks_t fs_blocks;
3270 int nmaps;
3271 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3272 xfs_daddr_t d;
3273 char *cur_chunk;
3274 int byte_cnt;
3275 int n;
3276 xfs_buf_t *bp;
3277 xfs_prid_t prid;
3278 struct xfs_dquot *udqp, *gdqp;
3279 uint resblks;
3280 char *link_name = VNAME(dentry);
3281 int link_namelen;
3282
3283 *vpp = NULL;
3284 dir_vp = BHV_TO_VNODE(dir_bdp);
3285 dp = XFS_BHVTOI(dir_bdp);
3286 dp_joined_to_trans = B_FALSE;
3287 error = 0;
3288 ip = NULL;
3289 tp = NULL;
3290
3291 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3292
3293 mp = dp->i_mount;
3294
3295 if (XFS_FORCED_SHUTDOWN(mp))
3296 return XFS_ERROR(EIO);
3297
3298 link_namelen = VNAMELEN(dentry);
3299
3300 /*
3301 * Check component lengths of the target path name.
3302 */
3303 pathlen = strlen(target_path);
3304 if (pathlen >= MAXPATHLEN) /* total string too long */
3305 return XFS_ERROR(ENAMETOOLONG);
3306 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3307 int len, total;
3308 char *path;
3309
3310 for(total = 0, path = target_path; total < pathlen;) {
3311 /*
3312 * Skip any slashes.
3313 */
3314 while(*path == '/') {
3315 total++;
3316 path++;
3317 }
3318
3319 /*
3320 * Count up to the next slash or end of path.
3321 * Error out if the component is bigger than MAXNAMELEN.
3322 */
3323 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3324 if (++len >= MAXNAMELEN) {
3325 error = ENAMETOOLONG;
3326 return error;
3327 }
3328 }
3329 }
3330 }
3331
3332 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3333 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3334 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3335 link_name, target_path, 0, 0, 0);
3336 if (error)
3337 return error;
3338 }
3339
3340 /* Return through std_return after this point. */
3341
3342 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10003343 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3344 prid = dp->i_d.di_projid;
3345 else if (vap->va_mask & XFS_AT_PROJID)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 prid = (xfs_prid_t)vap->va_projid;
3347 else
3348 prid = (xfs_prid_t)dfltprid;
3349
3350 /*
3351 * Make sure that we have allocated dquot(s) on disk.
3352 */
3353 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10003354 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3356 if (error)
3357 goto std_return;
3358
3359 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3360 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3361 /*
3362 * The symlink will fit into the inode data fork?
3363 * There can't be any attributes so we get the whole variable part.
3364 */
3365 if (pathlen <= XFS_LITINO(mp))
3366 fs_blocks = 0;
3367 else
3368 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3369 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3370 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3371 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3372 if (error == ENOSPC && fs_blocks == 0) {
3373 resblks = 0;
3374 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3375 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3376 }
3377 if (error) {
3378 cancel_flags = 0;
3379 dp = NULL;
3380 goto error_return;
3381 }
3382
3383 xfs_ilock(dp, XFS_ILOCK_EXCL);
3384
3385 /*
3386 * Check whether the directory allows new symlinks or not.
3387 */
3388 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3389 error = XFS_ERROR(EPERM);
3390 goto error_return;
3391 }
3392
3393 /*
3394 * Reserve disk quota : blocks and inode.
3395 */
3396 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3397 if (error)
3398 goto error_return;
3399
3400 /*
3401 * Check for ability to enter directory entry, if no space reserved.
3402 */
3403 if (resblks == 0 &&
3404 (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3405 goto error_return;
3406 /*
3407 * Initialize the bmap freelist prior to calling either
3408 * bmapi or the directory create code.
3409 */
3410 XFS_BMAP_INIT(&free_list, &first_block);
3411
3412 /*
3413 * Allocate an inode for the symlink.
3414 */
3415 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3416 1, 0, credp, prid, resblks > 0, &ip, NULL);
3417 if (error) {
3418 if (error == ENOSPC)
3419 goto error_return;
3420 goto error1;
3421 }
3422 ITRACE(ip);
3423
3424 VN_HOLD(dir_vp);
3425 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3426 dp_joined_to_trans = B_TRUE;
3427
3428 /*
3429 * Also attach the dquot(s) to it, if applicable.
3430 */
3431 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3432
3433 if (resblks)
3434 resblks -= XFS_IALLOC_SPACE_RES(mp);
3435 /*
3436 * If the symlink will fit into the inode, write it inline.
3437 */
3438 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3439 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3440 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3441 ip->i_d.di_size = pathlen;
3442
3443 /*
3444 * The inode was initially created in extent format.
3445 */
3446 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3447 ip->i_df.if_flags |= XFS_IFINLINE;
3448
3449 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3450 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3451
3452 } else {
3453 first_fsb = 0;
3454 nmaps = SYMLINK_MAPS;
3455
3456 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3457 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3458 &first_block, resblks, mval, &nmaps,
3459 &free_list);
3460 if (error) {
3461 goto error1;
3462 }
3463
3464 if (resblks)
3465 resblks -= fs_blocks;
3466 ip->i_d.di_size = pathlen;
3467 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3468
3469 cur_chunk = target_path;
3470 for (n = 0; n < nmaps; n++) {
3471 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3472 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3473 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3474 BTOBB(byte_cnt), 0);
3475 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3476 if (pathlen < byte_cnt) {
3477 byte_cnt = pathlen;
3478 }
3479 pathlen -= byte_cnt;
3480
3481 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3482 cur_chunk += byte_cnt;
3483
3484 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3485 }
3486 }
3487
3488 /*
3489 * Create the directory entry for the symlink.
3490 */
3491 error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3492 ip->i_ino, &first_block, &free_list, resblks);
3493 if (error) {
3494 goto error1;
3495 }
3496 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3497 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3498
3499 /*
3500 * Bump the in memory version number of the parent directory
3501 * so that other processes accessing it will recognize that
3502 * the directory has changed.
3503 */
3504 dp->i_gen++;
3505
3506 /*
3507 * If this is a synchronous mount, make sure that the
3508 * symlink transaction goes to disk before returning to
3509 * the user.
3510 */
3511 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3512 xfs_trans_set_sync(tp);
3513 }
3514
3515 /*
3516 * xfs_trans_commit normally decrements the vnode ref count
3517 * when it unlocks the inode. Since we want to return the
3518 * vnode to the caller, we bump the vnode ref count now.
3519 */
3520 IHOLD(ip);
3521
3522 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3523 if (error) {
3524 goto error2;
3525 }
3526 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3527 XFS_QM_DQRELE(mp, udqp);
3528 XFS_QM_DQRELE(mp, gdqp);
3529
3530 /* Fall through to std_return with error = 0 or errno from
3531 * xfs_trans_commit */
3532std_return:
3533 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3534 DM_EVENT_POSTSYMLINK)) {
3535 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3536 dir_vp, DM_RIGHT_NULL,
3537 error ? NULL : XFS_ITOV(ip),
3538 DM_RIGHT_NULL, link_name, target_path,
3539 0, error, 0);
3540 }
3541
3542 if (!error) {
3543 vnode_t *vp;
3544
3545 ASSERT(ip);
3546 vp = XFS_ITOV(ip);
3547 *vpp = vp;
3548 }
3549 return error;
3550
3551 error2:
3552 IRELE(ip);
3553 error1:
3554 xfs_bmap_cancel(&free_list);
3555 cancel_flags |= XFS_TRANS_ABORT;
3556 error_return:
3557 xfs_trans_cancel(tp, cancel_flags);
3558 XFS_QM_DQRELE(mp, udqp);
3559 XFS_QM_DQRELE(mp, gdqp);
3560
3561 if (!dp_joined_to_trans && (dp != NULL)) {
3562 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3563 }
3564
3565 goto std_return;
3566}
3567
3568
3569/*
3570 * xfs_fid2
3571 *
3572 * A fid routine that takes a pointer to a previously allocated
3573 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3574 */
3575STATIC int
3576xfs_fid2(
3577 bhv_desc_t *bdp,
3578 fid_t *fidp)
3579{
3580 xfs_inode_t *ip;
3581 xfs_fid2_t *xfid;
3582
3583 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3584 (inst_t *)__return_address);
3585 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3586
3587 xfid = (xfs_fid2_t *)fidp;
3588 ip = XFS_BHVTOI(bdp);
3589 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3590 xfid->fid_pad = 0;
3591 /*
3592 * use memcpy because the inode is a long long and there's no
3593 * assurance that xfid->fid_ino is properly aligned.
3594 */
3595 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3596 xfid->fid_gen = ip->i_d.di_gen;
3597
3598 return 0;
3599}
3600
3601
3602/*
3603 * xfs_rwlock
3604 */
3605int
3606xfs_rwlock(
3607 bhv_desc_t *bdp,
3608 vrwlock_t locktype)
3609{
3610 xfs_inode_t *ip;
3611 vnode_t *vp;
3612
3613 vp = BHV_TO_VNODE(bdp);
Christoph Hellwig0432dab2005-09-02 16:46:51 +10003614 if (VN_ISDIR(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615 return 1;
3616 ip = XFS_BHVTOI(bdp);
3617 if (locktype == VRWLOCK_WRITE) {
3618 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3619 } else if (locktype == VRWLOCK_TRY_READ) {
Jesper Juhl014c2542006-01-15 02:37:08 +01003620 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 } else if (locktype == VRWLOCK_TRY_WRITE) {
Jesper Juhl014c2542006-01-15 02:37:08 +01003622 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 } else {
3624 ASSERT((locktype == VRWLOCK_READ) ||
3625 (locktype == VRWLOCK_WRITE_DIRECT));
3626 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3627 }
3628
3629 return 1;
3630}
3631
3632
3633/*
3634 * xfs_rwunlock
3635 */
3636void
3637xfs_rwunlock(
3638 bhv_desc_t *bdp,
3639 vrwlock_t locktype)
3640{
3641 xfs_inode_t *ip;
3642 vnode_t *vp;
3643
3644 vp = BHV_TO_VNODE(bdp);
Christoph Hellwig0432dab2005-09-02 16:46:51 +10003645 if (VN_ISDIR(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646 return;
3647 ip = XFS_BHVTOI(bdp);
3648 if (locktype == VRWLOCK_WRITE) {
3649 /*
3650 * In the write case, we may have added a new entry to
3651 * the reference cache. This might store a pointer to
3652 * an inode to be released in this inode. If it is there,
3653 * clear the pointer and release the inode after unlocking
3654 * this one.
3655 */
3656 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3657 } else {
3658 ASSERT((locktype == VRWLOCK_READ) ||
3659 (locktype == VRWLOCK_WRITE_DIRECT));
3660 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3661 }
3662 return;
3663}
3664
3665STATIC int
3666xfs_inode_flush(
3667 bhv_desc_t *bdp,
3668 int flags)
3669{
3670 xfs_inode_t *ip;
3671 xfs_mount_t *mp;
3672 xfs_inode_log_item_t *iip;
3673 int error = 0;
3674
3675 ip = XFS_BHVTOI(bdp);
3676 mp = ip->i_mount;
3677 iip = ip->i_itemp;
3678
3679 if (XFS_FORCED_SHUTDOWN(mp))
3680 return XFS_ERROR(EIO);
3681
3682 /*
3683 * Bypass inodes which have already been cleaned by
3684 * the inode flush clustering code inside xfs_iflush
3685 */
3686 if ((ip->i_update_core == 0) &&
3687 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3688 return 0;
3689
3690 if (flags & FLUSH_LOG) {
3691 if (iip && iip->ili_last_lsn) {
3692 xlog_t *log = mp->m_log;
3693 xfs_lsn_t sync_lsn;
3694 int s, log_flags = XFS_LOG_FORCE;
3695
3696 s = GRANT_LOCK(log);
3697 sync_lsn = log->l_last_sync_lsn;
3698 GRANT_UNLOCK(log, s);
3699
3700 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3701 return 0;
3702
3703 if (flags & FLUSH_SYNC)
3704 log_flags |= XFS_LOG_SYNC;
3705 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3706 }
3707 }
3708
3709 /*
3710 * We make this non-blocking if the inode is contended,
3711 * return EAGAIN to indicate to the caller that they
3712 * did not succeed. This prevents the flush path from
3713 * blocking on inodes inside another operation right
3714 * now, they get caught later by xfs_sync.
3715 */
3716 if (flags & FLUSH_INODE) {
3717 int flush_flags;
3718
3719 if (xfs_ipincount(ip))
3720 return EAGAIN;
3721
3722 if (flags & FLUSH_SYNC) {
3723 xfs_ilock(ip, XFS_ILOCK_SHARED);
3724 xfs_iflock(ip);
3725 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3726 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3727 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3728 return EAGAIN;
3729 }
3730 } else {
3731 return EAGAIN;
3732 }
3733
3734 if (flags & FLUSH_SYNC)
3735 flush_flags = XFS_IFLUSH_SYNC;
3736 else
3737 flush_flags = XFS_IFLUSH_ASYNC;
3738
3739 error = xfs_iflush(ip, flush_flags);
3740 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3741 }
3742
3743 return error;
3744}
3745
3746
3747int
3748xfs_set_dmattrs (
3749 bhv_desc_t *bdp,
3750 u_int evmask,
3751 u_int16_t state,
3752 cred_t *credp)
3753{
3754 xfs_inode_t *ip;
3755 xfs_trans_t *tp;
3756 xfs_mount_t *mp;
3757 int error;
3758
3759 if (!capable(CAP_SYS_ADMIN))
3760 return XFS_ERROR(EPERM);
3761
3762 ip = XFS_BHVTOI(bdp);
3763 mp = ip->i_mount;
3764
3765 if (XFS_FORCED_SHUTDOWN(mp))
3766 return XFS_ERROR(EIO);
3767
3768 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3769 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3770 if (error) {
3771 xfs_trans_cancel(tp, 0);
3772 return error;
3773 }
3774 xfs_ilock(ip, XFS_ILOCK_EXCL);
3775 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3776
3777 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3778 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3779
3780 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3781 IHOLD(ip);
3782 error = xfs_trans_commit(tp, 0, NULL);
3783
3784 return error;
3785}
3786
3787
3788/*
3789 * xfs_reclaim
3790 */
3791STATIC int
3792xfs_reclaim(
3793 bhv_desc_t *bdp)
3794{
3795 xfs_inode_t *ip;
3796 vnode_t *vp;
3797
3798 vp = BHV_TO_VNODE(bdp);
3799 ip = XFS_BHVTOI(bdp);
3800
3801 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3802
3803 ASSERT(!VN_MAPPED(vp));
3804
3805 /* bad inode, get out here ASAP */
3806 if (VN_BAD(vp)) {
3807 xfs_ireclaim(ip);
3808 return 0;
3809 }
3810
Christoph Hellwig51c91ed2005-09-02 16:58:38 +10003811 vn_iowait(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812
Christoph Hellwig51c91ed2005-09-02 16:58:38 +10003813 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11003815 /*
3816 * Make sure the atime in the XFS inode is correct before freeing the
3817 * Linux inode.
3818 */
3819 xfs_synchronize_atime(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820
3821 /* If we have nothing to flush with this inode then complete the
3822 * teardown now, otherwise break the link between the xfs inode
3823 * and the linux inode and clean up the xfs inode later. This
3824 * avoids flushing the inode to disk during the delete operation
3825 * itself.
3826 */
3827 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3828 xfs_ilock(ip, XFS_ILOCK_EXCL);
3829 xfs_iflock(ip);
3830 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3831 } else {
3832 xfs_mount_t *mp = ip->i_mount;
3833
3834 /* Protect sync from us */
3835 XFS_MOUNT_ILOCK(mp);
3836 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3837 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3838 ip->i_flags |= XFS_IRECLAIMABLE;
3839 XFS_MOUNT_IUNLOCK(mp);
3840 }
3841 return 0;
3842}
3843
3844int
3845xfs_finish_reclaim(
3846 xfs_inode_t *ip,
3847 int locked,
3848 int sync_mode)
3849{
3850 xfs_ihash_t *ih = ip->i_hash;
3851 vnode_t *vp = XFS_ITOV_NULL(ip);
3852 int error;
3853
3854 if (vp && VN_BAD(vp))
3855 goto reclaim;
3856
3857 /* The hash lock here protects a thread in xfs_iget_core from
3858 * racing with us on linking the inode back with a vnode.
3859 * Once we have the XFS_IRECLAIM flag set it will not touch
3860 * us.
3861 */
3862 write_lock(&ih->ih_lock);
3863 if ((ip->i_flags & XFS_IRECLAIM) ||
3864 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3865 write_unlock(&ih->ih_lock);
3866 if (locked) {
3867 xfs_ifunlock(ip);
3868 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3869 }
Jesper Juhl014c2542006-01-15 02:37:08 +01003870 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871 }
3872 ip->i_flags |= XFS_IRECLAIM;
3873 write_unlock(&ih->ih_lock);
3874
3875 /*
3876 * If the inode is still dirty, then flush it out. If the inode
3877 * is not in the AIL, then it will be OK to flush it delwri as
3878 * long as xfs_iflush() does not keep any references to the inode.
3879 * We leave that decision up to xfs_iflush() since it has the
3880 * knowledge of whether it's OK to simply do a delwri flush of
3881 * the inode or whether we need to wait until the inode is
3882 * pulled from the AIL.
3883 * We get the flush lock regardless, though, just to make sure
3884 * we don't free it while it is being flushed.
3885 */
3886 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3887 if (!locked) {
3888 xfs_ilock(ip, XFS_ILOCK_EXCL);
3889 xfs_iflock(ip);
3890 }
3891
3892 if (ip->i_update_core ||
3893 ((ip->i_itemp != NULL) &&
3894 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3895 error = xfs_iflush(ip, sync_mode);
3896 /*
3897 * If we hit an error, typically because of filesystem
3898 * shutdown, we don't need to let vn_reclaim to know
3899 * because we're gonna reclaim the inode anyway.
3900 */
3901 if (error) {
3902 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3903 goto reclaim;
3904 }
3905 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3906 }
3907
3908 ASSERT(ip->i_update_core == 0);
3909 ASSERT(ip->i_itemp == NULL ||
3910 ip->i_itemp->ili_format.ilf_fields == 0);
3911 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3912 } else if (locked) {
3913 /*
3914 * We are not interested in doing an iflush if we're
3915 * in the process of shutting down the filesystem forcibly.
3916 * So, just reclaim the inode.
3917 */
3918 xfs_ifunlock(ip);
3919 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3920 }
3921
3922 reclaim:
3923 xfs_ireclaim(ip);
3924 return 0;
3925}
3926
3927int
3928xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3929{
3930 int purged;
3931 xfs_inode_t *ip, *n;
3932 int done = 0;
3933
3934 while (!done) {
3935 purged = 0;
3936 XFS_MOUNT_ILOCK(mp);
3937 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3938 if (noblock) {
3939 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3940 continue;
3941 if (xfs_ipincount(ip) ||
3942 !xfs_iflock_nowait(ip)) {
3943 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3944 continue;
3945 }
3946 }
3947 XFS_MOUNT_IUNLOCK(mp);
Felix Blyakher6b2cf612005-11-25 16:42:13 +11003948 if (xfs_finish_reclaim(ip, noblock,
3949 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3950 delay(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003951 purged = 1;
3952 break;
3953 }
3954
3955 done = !purged;
3956 }
3957
3958 XFS_MOUNT_IUNLOCK(mp);
3959 return 0;
3960}
3961
3962/*
3963 * xfs_alloc_file_space()
3964 * This routine allocates disk space for the given file.
3965 *
3966 * If alloc_type == 0, this request is for an ALLOCSP type
3967 * request which will change the file size. In this case, no
3968 * DMAPI event will be generated by the call. A TRUNCATE event
3969 * will be generated later by xfs_setattr.
3970 *
3971 * If alloc_type != 0, this request is for a RESVSP type
3972 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3973 * lower block boundary byte address is less than the file's
3974 * length.
3975 *
3976 * RETURNS:
3977 * 0 on success
3978 * errno on error
3979 *
3980 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003981STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982xfs_alloc_file_space(
3983 xfs_inode_t *ip,
3984 xfs_off_t offset,
3985 xfs_off_t len,
3986 int alloc_type,
3987 int attr_flags)
3988{
Nathan Scottdd9f4382006-01-11 15:28:28 +11003989 xfs_mount_t *mp = ip->i_mount;
3990 xfs_off_t count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 xfs_filblks_t allocated_fsb;
3992 xfs_filblks_t allocatesize_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003993 xfs_extlen_t extsz, temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994 xfs_fileoff_t startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003995 xfs_fsblock_t firstfsb;
3996 int nimaps;
3997 int bmapi_flag;
3998 int quota_flag;
3999 int rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 xfs_trans_t *tp;
Nathan Scottdd9f4382006-01-11 15:28:28 +11004001 xfs_bmbt_irec_t imaps[1], *imapp;
4002 xfs_bmap_free_t free_list;
4003 uint qblocks, resblks, resrtextents;
4004 int committed;
4005 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006
4007 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008
4009 if (XFS_FORCED_SHUTDOWN(mp))
4010 return XFS_ERROR(EIO);
4011
Nathan Scottdd9f4382006-01-11 15:28:28 +11004012 rt = XFS_IS_REALTIME_INODE(ip);
4013 if (unlikely(rt)) {
4014 if (!(extsz = ip->i_d.di_extsize))
4015 extsz = mp->m_sb.sb_rextsize;
4016 } else {
4017 extsz = ip->i_d.di_extsize;
4018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019
4020 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4021 return error;
4022
4023 if (len <= 0)
4024 return XFS_ERROR(EINVAL);
4025
4026 count = len;
4027 error = 0;
4028 imapp = &imaps[0];
Nathan Scottdd9f4382006-01-11 15:28:28 +11004029 nimaps = 1;
4030 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004031 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4032 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4033
4034 /* Generate a DMAPI event if needed. */
4035 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4036 (attr_flags&ATTR_DMI) == 0 &&
4037 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4038 xfs_off_t end_dmi_offset;
4039
4040 end_dmi_offset = offset+len;
4041 if (end_dmi_offset > ip->i_d.di_size)
4042 end_dmi_offset = ip->i_d.di_size;
4043 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4044 offset, end_dmi_offset - offset,
4045 0, NULL);
4046 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01004047 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 }
4049
4050 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004051 * Allocate file space until done or until there is an error
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052 */
4053retry:
4054 while (allocatesize_fsb && !error) {
Nathan Scottdd9f4382006-01-11 15:28:28 +11004055 xfs_fileoff_t s, e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004056
Nathan Scottdd9f4382006-01-11 15:28:28 +11004057 /*
Nathan Scott3ddb8fa2006-01-11 15:33:02 +11004058 * Determine space reservations for data/realtime.
Nathan Scottdd9f4382006-01-11 15:28:28 +11004059 */
4060 if (unlikely(extsz)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061 s = startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11004062 do_div(s, extsz);
4063 s *= extsz;
4064 e = startoffset_fsb + allocatesize_fsb;
4065 if ((temp = do_mod(startoffset_fsb, extsz)))
4066 e += temp;
4067 if ((temp = do_mod(e, extsz)))
4068 e += extsz - temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069 } else {
Nathan Scottdd9f4382006-01-11 15:28:28 +11004070 s = 0;
4071 e = allocatesize_fsb;
4072 }
4073
4074 if (unlikely(rt)) {
4075 resrtextents = qblocks = (uint)(e - s);
4076 resrtextents /= mp->m_sb.sb_rextsize;
4077 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4078 quota_flag = XFS_QMOPT_RES_RTBLKS;
4079 } else {
4080 resrtextents = 0;
4081 resblks = qblocks = \
4082 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4083 quota_flag = XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004084 }
4085
4086 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004087 * Allocate and setup the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088 */
4089 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Nathan Scottdd9f4382006-01-11 15:28:28 +11004090 error = xfs_trans_reserve(tp, resblks,
4091 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092 XFS_TRANS_PERM_LOG_RES,
4093 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004095 * Check for running out of space
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096 */
4097 if (error) {
4098 /*
4099 * Free the transaction structure.
4100 */
4101 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4102 xfs_trans_cancel(tp, 0);
4103 break;
4104 }
4105 xfs_ilock(ip, XFS_ILOCK_EXCL);
Nathan Scottdd9f4382006-01-11 15:28:28 +11004106 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4107 qblocks, 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004108 if (error)
4109 goto error1;
4110
4111 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4112 xfs_trans_ihold(tp, ip);
4113
4114 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004115 * Issue the xfs_bmapi() call to allocate the blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116 */
4117 XFS_BMAP_INIT(&free_list, &firstfsb);
4118 error = xfs_bmapi(tp, ip, startoffset_fsb,
Nathan Scottdd9f4382006-01-11 15:28:28 +11004119 allocatesize_fsb, bmapi_flag,
4120 &firstfsb, 0, imapp, &nimaps,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121 &free_list);
4122 if (error) {
4123 goto error0;
4124 }
4125
4126 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004127 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128 */
4129 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4130 if (error) {
4131 goto error0;
4132 }
4133
4134 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4135 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4136 if (error) {
4137 break;
4138 }
4139
4140 allocated_fsb = imapp->br_blockcount;
4141
Nathan Scottdd9f4382006-01-11 15:28:28 +11004142 if (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004143 error = XFS_ERROR(ENOSPC);
4144 break;
4145 }
4146
4147 startoffset_fsb += allocated_fsb;
4148 allocatesize_fsb -= allocated_fsb;
4149 }
4150dmapi_enospc_check:
4151 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4152 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4153
4154 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4155 XFS_ITOV(ip), DM_RIGHT_NULL,
4156 XFS_ITOV(ip), DM_RIGHT_NULL,
4157 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4158 if (error == 0)
4159 goto retry; /* Maybe DMAPI app. has made space */
4160 /* else fall through with error from XFS_SEND_DATA */
4161 }
4162
4163 return error;
4164
Nathan Scottdd9f4382006-01-11 15:28:28 +11004165error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 xfs_bmap_cancel(&free_list);
Nathan Scottdd9f4382006-01-11 15:28:28 +11004167 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4168
4169error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004170 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4171 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4172 goto dmapi_enospc_check;
4173}
4174
4175/*
4176 * Zero file bytes between startoff and endoff inclusive.
4177 * The iolock is held exclusive and no blocks are buffered.
4178 */
4179STATIC int
4180xfs_zero_remaining_bytes(
4181 xfs_inode_t *ip,
4182 xfs_off_t startoff,
4183 xfs_off_t endoff)
4184{
4185 xfs_bmbt_irec_t imap;
4186 xfs_fileoff_t offset_fsb;
4187 xfs_off_t lastoffset;
4188 xfs_off_t offset;
4189 xfs_buf_t *bp;
4190 xfs_mount_t *mp = ip->i_mount;
4191 int nimap;
4192 int error = 0;
4193
4194 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4195 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4196 mp->m_rtdev_targp : mp->m_ddev_targp);
4197
4198 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4199 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4200 nimap = 1;
4201 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4202 &nimap, NULL);
4203 if (error || nimap < 1)
4204 break;
4205 ASSERT(imap.br_blockcount >= 1);
4206 ASSERT(imap.br_startoff == offset_fsb);
4207 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4208 if (lastoffset > endoff)
4209 lastoffset = endoff;
4210 if (imap.br_startblock == HOLESTARTBLOCK)
4211 continue;
4212 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4213 if (imap.br_state == XFS_EXT_UNWRITTEN)
4214 continue;
4215 XFS_BUF_UNDONE(bp);
4216 XFS_BUF_UNWRITE(bp);
4217 XFS_BUF_READ(bp);
4218 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4219 xfsbdstrat(mp, bp);
4220 if ((error = xfs_iowait(bp))) {
4221 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4222 mp, bp, XFS_BUF_ADDR(bp));
4223 break;
4224 }
4225 memset(XFS_BUF_PTR(bp) +
4226 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4227 0, lastoffset - offset + 1);
4228 XFS_BUF_UNDONE(bp);
4229 XFS_BUF_UNREAD(bp);
4230 XFS_BUF_WRITE(bp);
4231 xfsbdstrat(mp, bp);
4232 if ((error = xfs_iowait(bp))) {
4233 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4234 mp, bp, XFS_BUF_ADDR(bp));
4235 break;
4236 }
4237 }
4238 xfs_buf_free(bp);
4239 return error;
4240}
4241
4242/*
4243 * xfs_free_file_space()
4244 * This routine frees disk space for the given file.
4245 *
4246 * This routine is only called by xfs_change_file_space
4247 * for an UNRESVSP type call.
4248 *
4249 * RETURNS:
4250 * 0 on success
4251 * errno on error
4252 *
4253 */
4254STATIC int
4255xfs_free_file_space(
4256 xfs_inode_t *ip,
4257 xfs_off_t offset,
4258 xfs_off_t len,
4259 int attr_flags)
4260{
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004261 vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004262 int committed;
4263 int done;
4264 xfs_off_t end_dmi_offset;
4265 xfs_fileoff_t endoffset_fsb;
4266 int error;
4267 xfs_fsblock_t firstfsb;
4268 xfs_bmap_free_t free_list;
4269 xfs_off_t ilen;
4270 xfs_bmbt_irec_t imap;
4271 xfs_off_t ioffset;
4272 xfs_extlen_t mod=0;
4273 xfs_mount_t *mp;
4274 int nimap;
4275 uint resblks;
4276 int rounding;
4277 int rt;
4278 xfs_fileoff_t startoffset_fsb;
4279 xfs_trans_t *tp;
Dean Roehrich5fcbab32005-05-05 13:27:19 -07004280 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004281
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004282 vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004283 mp = ip->i_mount;
4284
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004285 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4286
Linus Torvalds1da177e2005-04-16 15:20:36 -07004287 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4288 return error;
4289
4290 error = 0;
4291 if (len <= 0) /* if nothing being freed */
4292 return error;
4293 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4294 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4295 end_dmi_offset = offset + len;
4296 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4297
4298 if (offset < ip->i_d.di_size &&
4299 (attr_flags & ATTR_DMI) == 0 &&
4300 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4301 if (end_dmi_offset > ip->i_d.di_size)
4302 end_dmi_offset = ip->i_d.di_size;
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004303 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004304 offset, end_dmi_offset - offset,
4305 AT_DELAY_FLAG(attr_flags), NULL);
4306 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01004307 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004308 }
4309
Dean Roehrich5fcbab32005-05-05 13:27:19 -07004310 ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4311 if (attr_flags & ATTR_NOLOCK)
4312 need_iolock = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004313 if (need_iolock)
4314 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Dean Roehrich5fcbab32005-05-05 13:27:19 -07004315
Linus Torvalds1da177e2005-04-16 15:20:36 -07004316 rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4317 (__uint8_t)NBPP);
4318 ilen = len + (offset & (rounding - 1));
4319 ioffset = offset & ~(rounding - 1);
4320 if (ilen & (rounding - 1))
4321 ilen = (ilen + rounding) & ~(rounding - 1);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004322
4323 if (VN_CACHED(vp) != 0) {
4324 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4325 ctooff(offtoct(ioffset)), -1);
4326 VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(ioffset)),
4327 -1, FI_REMAPF_LOCKED);
4328 }
4329
Linus Torvalds1da177e2005-04-16 15:20:36 -07004330 /*
4331 * Need to zero the stuff we're not freeing, on disk.
4332 * If its a realtime file & can't use unwritten extents then we
4333 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4334 * will take care of it for us.
4335 */
4336 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4337 nimap = 1;
4338 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4339 &imap, &nimap, NULL);
4340 if (error)
4341 goto out_unlock_iolock;
4342 ASSERT(nimap == 0 || nimap == 1);
4343 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4344 xfs_daddr_t block;
4345
4346 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4347 block = imap.br_startblock;
4348 mod = do_div(block, mp->m_sb.sb_rextsize);
4349 if (mod)
4350 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4351 }
4352 nimap = 1;
4353 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4354 &imap, &nimap, NULL);
4355 if (error)
4356 goto out_unlock_iolock;
4357 ASSERT(nimap == 0 || nimap == 1);
4358 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4359 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4360 mod++;
4361 if (mod && (mod != mp->m_sb.sb_rextsize))
4362 endoffset_fsb -= mod;
4363 }
4364 }
4365 if ((done = (endoffset_fsb <= startoffset_fsb)))
4366 /*
4367 * One contiguous piece to clear
4368 */
4369 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4370 else {
4371 /*
4372 * Some full blocks, possibly two pieces to clear
4373 */
4374 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4375 error = xfs_zero_remaining_bytes(ip, offset,
4376 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4377 if (!error &&
4378 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4379 error = xfs_zero_remaining_bytes(ip,
4380 XFS_FSB_TO_B(mp, endoffset_fsb),
4381 offset + len - 1);
4382 }
4383
4384 /*
4385 * free file space until done or until there is an error
4386 */
4387 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4388 while (!error && !done) {
4389
4390 /*
4391 * allocate and setup the transaction
4392 */
4393 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4394 error = xfs_trans_reserve(tp,
4395 resblks,
4396 XFS_WRITE_LOG_RES(mp),
4397 0,
4398 XFS_TRANS_PERM_LOG_RES,
4399 XFS_WRITE_LOG_COUNT);
4400
4401 /*
4402 * check for running out of space
4403 */
4404 if (error) {
4405 /*
4406 * Free the transaction structure.
4407 */
4408 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4409 xfs_trans_cancel(tp, 0);
4410 break;
4411 }
4412 xfs_ilock(ip, XFS_ILOCK_EXCL);
4413 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
Nathan Scottdd9f4382006-01-11 15:28:28 +11004414 ip->i_udquot, ip->i_gdquot, resblks, 0,
4415 XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416 if (error)
4417 goto error1;
4418
4419 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4420 xfs_trans_ihold(tp, ip);
4421
4422 /*
4423 * issue the bunmapi() call to free the blocks
4424 */
4425 XFS_BMAP_INIT(&free_list, &firstfsb);
4426 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4427 endoffset_fsb - startoffset_fsb,
4428 0, 2, &firstfsb, &free_list, &done);
4429 if (error) {
4430 goto error0;
4431 }
4432
4433 /*
4434 * complete the transaction
4435 */
4436 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4437 if (error) {
4438 goto error0;
4439 }
4440
4441 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4442 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4443 }
4444
4445 out_unlock_iolock:
4446 if (need_iolock)
4447 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4448 return error;
4449
4450 error0:
4451 xfs_bmap_cancel(&free_list);
4452 error1:
4453 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4454 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4455 XFS_ILOCK_EXCL);
4456 return error;
4457}
4458
4459/*
4460 * xfs_change_file_space()
4461 * This routine allocates or frees disk space for the given file.
4462 * The user specified parameters are checked for alignment and size
4463 * limitations.
4464 *
4465 * RETURNS:
4466 * 0 on success
4467 * errno on error
4468 *
4469 */
4470int
4471xfs_change_file_space(
4472 bhv_desc_t *bdp,
4473 int cmd,
4474 xfs_flock64_t *bf,
4475 xfs_off_t offset,
4476 cred_t *credp,
4477 int attr_flags)
4478{
4479 int clrprealloc;
4480 int error;
4481 xfs_fsize_t fsize;
4482 xfs_inode_t *ip;
4483 xfs_mount_t *mp;
4484 int setprealloc;
4485 xfs_off_t startoffset;
4486 xfs_off_t llen;
4487 xfs_trans_t *tp;
4488 vattr_t va;
4489 vnode_t *vp;
4490
4491 vp = BHV_TO_VNODE(bdp);
4492 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4493
4494 ip = XFS_BHVTOI(bdp);
4495 mp = ip->i_mount;
4496
4497 /*
4498 * must be a regular file and have write permission
4499 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +10004500 if (!VN_ISREG(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004501 return XFS_ERROR(EINVAL);
4502
4503 xfs_ilock(ip, XFS_ILOCK_SHARED);
4504
4505 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4506 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4507 return error;
4508 }
4509
4510 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4511
4512 switch (bf->l_whence) {
4513 case 0: /*SEEK_SET*/
4514 break;
4515 case 1: /*SEEK_CUR*/
4516 bf->l_start += offset;
4517 break;
4518 case 2: /*SEEK_END*/
4519 bf->l_start += ip->i_d.di_size;
4520 break;
4521 default:
4522 return XFS_ERROR(EINVAL);
4523 }
4524
4525 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4526
4527 if ( (bf->l_start < 0)
4528 || (bf->l_start > XFS_MAXIOFFSET(mp))
4529 || (bf->l_start + llen < 0)
4530 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4531 return XFS_ERROR(EINVAL);
4532
4533 bf->l_whence = 0;
4534
4535 startoffset = bf->l_start;
4536 fsize = ip->i_d.di_size;
4537
4538 /*
4539 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4540 * file space.
4541 * These calls do NOT zero the data space allocated to the file,
4542 * nor do they change the file size.
4543 *
4544 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4545 * space.
4546 * These calls cause the new file data to be zeroed and the file
4547 * size to be changed.
4548 */
4549 setprealloc = clrprealloc = 0;
4550
4551 switch (cmd) {
4552 case XFS_IOC_RESVSP:
4553 case XFS_IOC_RESVSP64:
4554 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4555 1, attr_flags);
4556 if (error)
4557 return error;
4558 setprealloc = 1;
4559 break;
4560
4561 case XFS_IOC_UNRESVSP:
4562 case XFS_IOC_UNRESVSP64:
4563 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4564 attr_flags)))
4565 return error;
4566 break;
4567
4568 case XFS_IOC_ALLOCSP:
4569 case XFS_IOC_ALLOCSP64:
4570 case XFS_IOC_FREESP:
4571 case XFS_IOC_FREESP64:
4572 if (startoffset > fsize) {
4573 error = xfs_alloc_file_space(ip, fsize,
4574 startoffset - fsize, 0, attr_flags);
4575 if (error)
4576 break;
4577 }
4578
4579 va.va_mask = XFS_AT_SIZE;
4580 va.va_size = startoffset;
4581
4582 error = xfs_setattr(bdp, &va, attr_flags, credp);
4583
4584 if (error)
4585 return error;
4586
4587 clrprealloc = 1;
4588 break;
4589
4590 default:
4591 ASSERT(0);
4592 return XFS_ERROR(EINVAL);
4593 }
4594
4595 /*
4596 * update the inode timestamp, mode, and prealloc flag bits
4597 */
4598 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4599
4600 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4601 0, 0, 0))) {
4602 /* ASSERT(0); */
4603 xfs_trans_cancel(tp, 0);
4604 return error;
4605 }
4606
4607 xfs_ilock(ip, XFS_ILOCK_EXCL);
4608
4609 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4610 xfs_trans_ihold(tp, ip);
4611
4612 if ((attr_flags & ATTR_DMI) == 0) {
4613 ip->i_d.di_mode &= ~S_ISUID;
4614
4615 /*
4616 * Note that we don't have to worry about mandatory
4617 * file locking being disabled here because we only
4618 * clear the S_ISGID bit if the Group execute bit is
4619 * on, but if it was on then mandatory locking wouldn't
4620 * have been enabled.
4621 */
4622 if (ip->i_d.di_mode & S_IXGRP)
4623 ip->i_d.di_mode &= ~S_ISGID;
4624
4625 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4626 }
4627 if (setprealloc)
4628 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4629 else if (clrprealloc)
4630 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4631
4632 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4633 xfs_trans_set_sync(tp);
4634
4635 error = xfs_trans_commit(tp, 0, NULL);
4636
4637 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4638
4639 return error;
4640}
4641
4642vnodeops_t xfs_vnodeops = {
4643 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4644 .vop_open = xfs_open,
4645 .vop_read = xfs_read,
4646#ifdef HAVE_SENDFILE
4647 .vop_sendfile = xfs_sendfile,
4648#endif
4649 .vop_write = xfs_write,
4650 .vop_ioctl = xfs_ioctl,
4651 .vop_getattr = xfs_getattr,
4652 .vop_setattr = xfs_setattr,
4653 .vop_access = xfs_access,
4654 .vop_lookup = xfs_lookup,
4655 .vop_create = xfs_create,
4656 .vop_remove = xfs_remove,
4657 .vop_link = xfs_link,
4658 .vop_rename = xfs_rename,
4659 .vop_mkdir = xfs_mkdir,
4660 .vop_rmdir = xfs_rmdir,
4661 .vop_readdir = xfs_readdir,
4662 .vop_symlink = xfs_symlink,
4663 .vop_readlink = xfs_readlink,
4664 .vop_fsync = xfs_fsync,
4665 .vop_inactive = xfs_inactive,
4666 .vop_fid2 = xfs_fid2,
4667 .vop_rwlock = xfs_rwlock,
4668 .vop_rwunlock = xfs_rwunlock,
4669 .vop_bmap = xfs_bmap,
4670 .vop_reclaim = xfs_reclaim,
4671 .vop_attr_get = xfs_attr_get,
4672 .vop_attr_set = xfs_attr_set,
4673 .vop_attr_remove = xfs_attr_remove,
4674 .vop_attr_list = xfs_attr_list,
4675 .vop_link_removed = (vop_link_removed_t)fs_noval,
4676 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4677 .vop_tosspages = fs_tosspages,
4678 .vop_flushinval_pages = fs_flushinval_pages,
4679 .vop_flush_pages = fs_flush_pages,
4680 .vop_release = xfs_release,
4681 .vop_iflush = xfs_inode_flush,
4682};