blob: 637fc1a2bb44f51c49e78517194329a782517eed [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
26#include "xfs_sb.h"
27#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_dir2.h"
29#include "xfs_dmapi.h"
30#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_dir2_sf.h"
36#include "xfs_attr_sf.h"
37#include "xfs_dinode.h"
38#include "xfs_inode.h"
39#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_itable.h"
41#include "xfs_btree.h"
42#include "xfs_ialloc.h"
43#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include "xfs_attr.h"
46#include "xfs_rw.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include "xfs_quota.h"
49#include "xfs_utils.h"
Nathan Scotta844f452005-11-02 14:38:42 +110050#include "xfs_rtalloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include "xfs_log_priv.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100053#include "xfs_filestream.h"
Christoph Hellwig993386c2007-08-28 16:12:30 +100054#include "xfs_vnodeops.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Christoph Hellwig993386c2007-08-28 16:12:30 +100056int
Linus Torvalds1da177e2005-04-16 15:20:36 -070057xfs_open(
Christoph Hellwig993386c2007-08-28 16:12:30 +100058 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -070059{
60 int mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
63 return XFS_ERROR(EIO);
64
65 /*
66 * If it's a directory with any blocks, read-ahead block 0
67 * as we're almost certain to have the next operation be a read there.
68 */
Christoph Hellwig993386c2007-08-28 16:12:30 +100069 if (S_ISDIR(ip->i_d.di_mode) && ip->i_d.di_nextents > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 mode = xfs_ilock_map_shared(ip);
71 if (ip->i_d.di_nextents > 0)
72 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
73 xfs_iunlock(ip, mode);
74 }
75 return 0;
76}
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * xfs_setattr
80 */
81int
82xfs_setattr(
Christoph Hellwig993386c2007-08-28 16:12:30 +100083 xfs_inode_t *ip,
Nathan Scott8285fb52006-06-09 17:07:12 +100084 bhv_vattr_t *vap,
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 int flags,
86 cred_t *credp)
87{
Christoph Hellwig993386c2007-08-28 16:12:30 +100088 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 int mask;
91 int code;
92 uint lock_flags;
93 uint commit_flags=0;
94 uid_t uid=0, iuid=0;
95 gid_t gid=0, igid=0;
96 int timeflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 xfs_prid_t projid=0, iprojid=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
99 int file_owner;
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700100 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100102 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000104 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 return XFS_ERROR(EROFS);
106
107 /*
108 * Cannot set certain attributes.
109 */
110 mask = vap->va_mask;
111 if (mask & XFS_AT_NOSET) {
112 return XFS_ERROR(EINVAL);
113 }
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 if (XFS_FORCED_SHUTDOWN(mp))
116 return XFS_ERROR(EIO);
117
118 /*
119 * Timestamps do not need to be logged and hence do not
120 * need to be done within a transaction.
121 */
122 if (mask & XFS_AT_UPDTIMES) {
123 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
124 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
125 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
126 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
127 xfs_ichgtime(ip, timeflags);
128 return 0;
129 }
130
131 olddquot1 = olddquot2 = NULL;
132 udqp = gdqp = NULL;
133
134 /*
135 * If disk quotas is on, we make sure that the dquots do exist on disk,
136 * before we start any other transactions. Trying to do this later
137 * is messy. We don't care to take a readlock to look at the ids
138 * in inode here, because we can't hold it across the trans_reserve.
139 * If the IDs do change before we take the ilock, we're covered
140 * because the i_*dquot fields will get updated anyway.
141 */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000142 if (XFS_IS_QUOTA_ON(mp) &&
143 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 uint qflags = 0;
145
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000146 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 uid = vap->va_uid;
148 qflags |= XFS_QMOPT_UQUOTA;
149 } else {
150 uid = ip->i_d.di_uid;
151 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000152 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 gid = vap->va_gid;
154 qflags |= XFS_QMOPT_GQUOTA;
155 } else {
156 gid = ip->i_d.di_gid;
157 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000158 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
159 projid = vap->va_projid;
160 qflags |= XFS_QMOPT_PQUOTA;
161 } else {
162 projid = ip->i_d.di_projid;
163 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 /*
165 * We take a reference when we initialize udqp and gdqp,
166 * so it is important that we never blindly double trip on
167 * the same variable. See xfs_create() for an example.
168 */
169 ASSERT(udqp == NULL);
170 ASSERT(gdqp == NULL);
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000171 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
172 &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 if (code)
Jesper Juhl014c2542006-01-15 02:37:08 +0100174 return code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 }
176
177 /*
178 * For the other attributes, we acquire the inode lock and
179 * first do an error checking pass.
180 */
181 tp = NULL;
182 lock_flags = XFS_ILOCK_EXCL;
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700183 if (flags & ATTR_NOLOCK)
184 need_iolock = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 if (!(mask & XFS_AT_SIZE)) {
186 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
187 (mp->m_flags & XFS_MOUNT_WSYNC)) {
188 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
189 commit_flags = 0;
190 if ((code = xfs_trans_reserve(tp, 0,
191 XFS_ICHANGE_LOG_RES(mp), 0,
192 0, 0))) {
193 lock_flags = 0;
194 goto error_return;
195 }
196 }
197 } else {
Christoph Hellwigeb9df392007-08-16 18:42:07 +1000198 if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 !(flags & ATTR_DMI)) {
200 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +1100201 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 vap->va_size, 0, dmflags, NULL);
203 if (code) {
204 lock_flags = 0;
205 goto error_return;
206 }
207 }
208 if (need_iolock)
209 lock_flags |= XFS_IOLOCK_EXCL;
210 }
211
212 xfs_ilock(ip, lock_flags);
213
214 /* boolean: are we the file owner? */
215 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
216
217 /*
218 * Change various properties of a file.
219 * Only the owner or users with CAP_FOWNER
220 * capability may do these things.
221 */
222 if (mask &
223 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
224 XFS_AT_GID|XFS_AT_PROJID)) {
225 /*
226 * CAP_FOWNER overrides the following restrictions:
227 *
228 * The user ID of the calling process must be equal
229 * to the file owner ID, except in cases where the
230 * CAP_FSETID capability is applicable.
231 */
232 if (!file_owner && !capable(CAP_FOWNER)) {
233 code = XFS_ERROR(EPERM);
234 goto error_return;
235 }
236
237 /*
238 * CAP_FSETID overrides the following restrictions:
239 *
240 * The effective user ID of the calling process shall match
241 * the file owner when setting the set-user-ID and
242 * set-group-ID bits on that file.
243 *
244 * The effective group ID or one of the supplementary group
245 * IDs of the calling process shall match the group owner of
246 * the file when setting the set-group-ID bit on that file
247 */
248 if (mask & XFS_AT_MODE) {
249 mode_t m = 0;
250
251 if ((vap->va_mode & S_ISUID) && !file_owner)
252 m |= S_ISUID;
253 if ((vap->va_mode & S_ISGID) &&
254 !in_group_p((gid_t)ip->i_d.di_gid))
255 m |= S_ISGID;
256#if 0
257 /* Linux allows this, Irix doesn't. */
Christoph Hellwig42173f62008-04-22 17:33:25 +1000258 if ((vap->va_mode & S_ISVTX) && !S_ISDIR(ip->i_d.di_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 m |= S_ISVTX;
260#endif
261 if (m && !capable(CAP_FSETID))
262 vap->va_mode &= ~m;
263 }
264 }
265
266 /*
267 * Change file ownership. Must be the owner or privileged.
268 * If the system was configured with the "restricted_chown"
269 * option, the owner is not permitted to give away the file,
270 * and can change the group id only to a group of which he
271 * or she is a member.
272 */
273 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
274 /*
275 * These IDs could have changed since we last looked at them.
276 * But, we're assured that if the ownership did change
277 * while we didn't have the inode locked, inode's dquot(s)
278 * would have changed also.
279 */
280 iuid = ip->i_d.di_uid;
281 iprojid = ip->i_d.di_projid;
282 igid = ip->i_d.di_gid;
283 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
284 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
285 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
286 iprojid;
287
288 /*
289 * CAP_CHOWN overrides the following restrictions:
290 *
291 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
292 * shall override the restriction that a process cannot
293 * change the user ID of a file it owns and the restriction
294 * that the group ID supplied to the chown() function
295 * shall be equal to either the group ID or one of the
296 * supplementary group IDs of the calling process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 */
298 if (restricted_chown &&
299 (iuid != uid || (igid != gid &&
300 !in_group_p((gid_t)gid))) &&
301 !capable(CAP_CHOWN)) {
302 code = XFS_ERROR(EPERM);
303 goto error_return;
304 }
305 /*
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000306 * Do a quota reservation only if uid/projid/gid is actually
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 * going to change.
308 */
309 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000310 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
312 ASSERT(tp);
313 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
314 capable(CAP_FOWNER) ?
315 XFS_QMOPT_FORCE_RES : 0);
316 if (code) /* out of quota */
317 goto error_return;
318 }
319 }
320
321 /*
322 * Truncate file. Must have write permission and not be a directory.
323 */
324 if (mask & XFS_AT_SIZE) {
325 /* Short circuit the truncate case for zero length files */
326 if ((vap->va_size == 0) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000327 (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 xfs_iunlock(ip, XFS_ILOCK_EXCL);
329 lock_flags &= ~XFS_ILOCK_EXCL;
330 if (mask & XFS_AT_CTIME)
331 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
332 code = 0;
333 goto error_return;
334 }
335
Christoph Hellwig42173f62008-04-22 17:33:25 +1000336 if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 code = XFS_ERROR(EISDIR);
338 goto error_return;
Christoph Hellwig42173f62008-04-22 17:33:25 +1000339 } else if (!S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 code = XFS_ERROR(EINVAL);
341 goto error_return;
342 }
343 /*
344 * Make sure that the dquots are attached to the inode.
345 */
346 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
347 goto error_return;
348 }
349
350 /*
351 * Change file access or modified times.
352 */
353 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
354 if (!file_owner) {
355 if ((flags & ATTR_UTIME) &&
356 !capable(CAP_FOWNER)) {
357 code = XFS_ERROR(EPERM);
358 goto error_return;
359 }
360 }
361 }
362
363 /*
364 * Change extent size or realtime flag.
365 */
366 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
367 /*
368 * Can't change extent size if any extents are allocated.
369 */
Eric Sandeene94af022005-11-02 15:10:41 +1100370 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
372 vap->va_extsize) ) {
373 code = XFS_ERROR(EINVAL); /* EFBIG? */
374 goto error_return;
375 }
376
377 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 * Can't change realtime flag if any extents are allocated.
379 */
Eric Sandeene94af022005-11-02 15:10:41 +1100380 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
381 (mask & XFS_AT_XFLAGS) &&
Eric Sandeen71ddabb2007-11-23 16:29:42 +1100382 (XFS_IS_REALTIME_INODE(ip)) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
384 code = XFS_ERROR(EINVAL); /* EFBIG? */
385 goto error_return;
386 }
387 /*
388 * Extent size must be a multiple of the appropriate block
389 * size, if set at all.
390 */
391 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
392 xfs_extlen_t size;
393
Eric Sandeen71ddabb2007-11-23 16:29:42 +1100394 if (XFS_IS_REALTIME_INODE(ip) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 ((mask & XFS_AT_XFLAGS) &&
396 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
397 size = mp->m_sb.sb_rextsize <<
398 mp->m_sb.sb_blocklog;
399 } else {
400 size = mp->m_sb.sb_blocksize;
401 }
402 if (vap->va_extsize % size) {
403 code = XFS_ERROR(EINVAL);
404 goto error_return;
405 }
406 }
407 /*
408 * If realtime flag is set then must have realtime data.
409 */
410 if ((mask & XFS_AT_XFLAGS) &&
411 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
412 if ((mp->m_sb.sb_rblocks == 0) ||
413 (mp->m_sb.sb_rextsize == 0) ||
414 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
415 code = XFS_ERROR(EINVAL);
416 goto error_return;
417 }
418 }
419
420 /*
421 * Can't modify an immutable/append-only file unless
422 * we have appropriate permission.
423 */
424 if ((mask & XFS_AT_XFLAGS) &&
425 (ip->i_d.di_flags &
426 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
427 (vap->va_xflags &
428 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
429 !capable(CAP_LINUX_IMMUTABLE)) {
430 code = XFS_ERROR(EPERM);
431 goto error_return;
432 }
433 }
434
435 /*
436 * Now we can make the changes. Before we join the inode
437 * to the transaction, if XFS_AT_SIZE is set then take care of
438 * the part of the truncation that must be done without the
439 * inode lock. This needs to be done before joining the inode
440 * to the transaction, because the inode cannot be unlocked
441 * once it is a part of the transaction.
442 */
443 if (mask & XFS_AT_SIZE) {
444 code = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000445 if ((vap->va_size > ip->i_size) &&
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100446 (flags & ATTR_NOSIZETOK) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 code = xfs_igrow_start(ip, vap->va_size, credp);
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnerc32676e2007-07-19 16:28:58 +1000450
451 /*
452 * We are going to log the inode size change in this
453 * transaction so any previous writes that are beyond the on
454 * disk EOF and the new EOF that have not been written out need
455 * to be written here. If we do not write the data out, we
456 * expose ourselves to the null files problem.
457 *
458 * Only flush from the on disk size to the smaller of the in
459 * memory file size or the new size as that's the range we
460 * really care about here and prevents waiting for other data
461 * not within the range we care about here.
462 */
463 if (!code &&
464 (ip->i_size != ip->i_d.di_size) &&
465 (vap->va_size > ip->i_d.di_size)) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000466 code = xfs_flush_pages(ip,
David Chinnerc32676e2007-07-19 16:28:58 +1000467 ip->i_d.di_size, vap->va_size,
468 XFS_B_ASYNC, FI_NONE);
469 }
470
471 /* wait for all I/O to complete */
Christoph Hellwigb677c212007-08-29 11:46:28 +1000472 vn_iowait(ip);
David Chinnerc32676e2007-07-19 16:28:58 +1000473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 if (!code)
475 code = xfs_itruncate_data(ip, vap->va_size);
476 if (code) {
477 ASSERT(tp == NULL);
478 lock_flags &= ~XFS_ILOCK_EXCL;
479 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
480 goto error_return;
481 }
482 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
483 if ((code = xfs_trans_reserve(tp, 0,
484 XFS_ITRUNCATE_LOG_RES(mp), 0,
485 XFS_TRANS_PERM_LOG_RES,
486 XFS_ITRUNCATE_LOG_COUNT))) {
487 xfs_trans_cancel(tp, 0);
488 if (need_iolock)
489 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
490 return code;
491 }
492 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
493 xfs_ilock(ip, XFS_ILOCK_EXCL);
494 }
495
496 if (tp) {
497 xfs_trans_ijoin(tp, ip, lock_flags);
498 xfs_trans_ihold(tp, ip);
499 }
500
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 /*
502 * Truncate file. Must have write permission and not be a directory.
503 */
504 if (mask & XFS_AT_SIZE) {
David Chinner44d814c2008-03-06 13:45:29 +1100505 /*
506 * Only change the c/mtime if we are changing the size
507 * or we are explicitly asked to change it. This handles
508 * the semantic difference between truncate() and ftruncate()
509 * as implemented in the VFS.
510 */
511 if (vap->va_size != ip->i_size || (mask & XFS_AT_CTIME))
512 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
513
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000514 if (vap->va_size > ip->i_size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 xfs_igrow_finish(tp, ip, vap->va_size,
516 !(flags & ATTR_DMI));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000517 } else if ((vap->va_size <= ip->i_size) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
519 /*
520 * signal a sync transaction unless
521 * we're truncating an already unlinked
522 * file on a wsync filesystem
523 */
524 code = xfs_itruncate_finish(&tp, ip,
525 (xfs_fsize_t)vap->va_size,
526 XFS_DATA_FORK,
527 ((ip->i_d.di_nlink != 0 ||
528 !(mp->m_flags & XFS_MOUNT_WSYNC))
529 ? 1 : 0));
Nathan Scott7d4fb402006-06-09 15:27:16 +1000530 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 goto abort_return;
Nathan Scott7d4fb402006-06-09 15:27:16 +1000532 /*
533 * Truncated "down", so we're removing references
534 * to old data here - if we now delay flushing for
535 * a long time, we expose ourselves unduly to the
536 * notorious NULL files problem. So, we mark this
537 * vnode and flush it when the file is closed, and
538 * do not wait the usual (long) time for writeout.
539 */
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000540 xfs_iflags_set(ip, XFS_ITRUNCATED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
543
544 /*
545 * Change file access modes.
546 */
547 if (mask & XFS_AT_MODE) {
548 ip->i_d.di_mode &= S_IFMT;
549 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
550
551 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
552 timeflags |= XFS_ICHGTIME_CHG;
553 }
554
555 /*
556 * Change file ownership. Must be the owner or privileged.
557 * If the system was configured with the "restricted_chown"
558 * option, the owner is not permitted to give away the file,
559 * and can change the group id only to a group of which he
560 * or she is a member.
561 */
562 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
563 /*
564 * CAP_FSETID overrides the following restrictions:
565 *
566 * The set-user-ID and set-group-ID bits of a file will be
567 * cleared upon successful return from chown()
568 */
569 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
570 !capable(CAP_FSETID)) {
571 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
572 }
573
574 /*
575 * Change the ownerships and register quota modifications
576 * in the transaction.
577 */
578 if (iuid != uid) {
579 if (XFS_IS_UQUOTA_ON(mp)) {
580 ASSERT(mask & XFS_AT_UID);
581 ASSERT(udqp);
582 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
583 &ip->i_udquot, udqp);
584 }
585 ip->i_d.di_uid = uid;
586 }
587 if (igid != gid) {
588 if (XFS_IS_GQUOTA_ON(mp)) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000589 ASSERT(!XFS_IS_PQUOTA_ON(mp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 ASSERT(mask & XFS_AT_GID);
591 ASSERT(gdqp);
592 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
593 &ip->i_gdquot, gdqp);
594 }
595 ip->i_d.di_gid = gid;
596 }
597 if (iprojid != projid) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000598 if (XFS_IS_PQUOTA_ON(mp)) {
599 ASSERT(!XFS_IS_GQUOTA_ON(mp));
600 ASSERT(mask & XFS_AT_PROJID);
601 ASSERT(gdqp);
602 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
603 &ip->i_gdquot, gdqp);
604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 ip->i_d.di_projid = projid;
606 /*
607 * We may have to rev the inode as well as
608 * the superblock version number since projids didn't
609 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
610 */
611 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
612 xfs_bump_ino_vers2(tp, ip);
613 }
614
615 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
616 timeflags |= XFS_ICHGTIME_CHG;
617 }
618
619
620 /*
621 * Change file access or modified times.
622 */
623 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
624 if (mask & XFS_AT_ATIME) {
625 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
626 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
627 ip->i_update_core = 1;
628 timeflags &= ~XFS_ICHGTIME_ACC;
629 }
630 if (mask & XFS_AT_MTIME) {
631 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
632 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
633 timeflags &= ~XFS_ICHGTIME_MOD;
634 timeflags |= XFS_ICHGTIME_CHG;
635 }
636 if (tp && (flags & ATTR_UTIME))
637 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
638 }
639
640 /*
641 * Change XFS-added attributes.
642 */
643 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
644 if (mask & XFS_AT_EXTSIZE) {
645 /*
646 * Converting bytes to fs blocks.
647 */
648 ip->i_d.di_extsize = vap->va_extsize >>
649 mp->m_sb.sb_blocklog;
650 }
651 if (mask & XFS_AT_XFLAGS) {
652 uint di_flags;
653
654 /* can't set PREALLOC this way, just preserve it */
655 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
656 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
657 di_flags |= XFS_DIFLAG_IMMUTABLE;
658 if (vap->va_xflags & XFS_XFLAG_APPEND)
659 di_flags |= XFS_DIFLAG_APPEND;
660 if (vap->va_xflags & XFS_XFLAG_SYNC)
661 di_flags |= XFS_DIFLAG_SYNC;
662 if (vap->va_xflags & XFS_XFLAG_NOATIME)
663 di_flags |= XFS_DIFLAG_NOATIME;
664 if (vap->va_xflags & XFS_XFLAG_NODUMP)
665 di_flags |= XFS_DIFLAG_NODUMP;
Nathan Scott365ca832005-06-21 15:39:12 +1000666 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
667 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000668 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
669 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000670 if (vap->va_xflags & XFS_XFLAG_FILESTREAM)
671 di_flags |= XFS_DIFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
673 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
674 di_flags |= XFS_DIFLAG_RTINHERIT;
675 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
676 di_flags |= XFS_DIFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100677 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
678 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
679 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
Christoph Hellwig613d7042007-10-11 17:44:08 +1000680 if (vap->va_xflags & XFS_XFLAG_REALTIME)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100682 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
683 di_flags |= XFS_DIFLAG_EXTSIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 }
685 ip->i_d.di_flags = di_flags;
686 }
687 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
688 timeflags |= XFS_ICHGTIME_CHG;
689 }
690
691 /*
692 * Change file inode change time only if XFS_AT_CTIME set
693 * AND we have been called by a DMI function.
694 */
695
696 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
697 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
698 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
699 ip->i_update_core = 1;
700 timeflags &= ~XFS_ICHGTIME_CHG;
701 }
702
703 /*
704 * Send out timestamp changes that need to be set to the
705 * current time. Not done when called by a DMI function.
706 */
707 if (timeflags && !(flags & ATTR_DMI))
708 xfs_ichgtime(ip, timeflags);
709
710 XFS_STATS_INC(xs_ig_attrchg);
711
712 /*
713 * If this is a synchronous mount, make sure that the
714 * transaction goes to disk before returning to the user.
715 * This is slightly sub-optimal in that truncates require
Nathan Scottc41564b2006-03-29 08:55:14 +1000716 * two sync transactions instead of one for wsync filesystems.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 * One for the truncate and one for the timestamps since we
718 * don't want to change the timestamps unless we're sure the
719 * truncate worked. Truncates are less than 1% of the laddis
720 * mix so this probably isn't worth the trouble to optimize.
721 */
722 code = 0;
723 if (tp) {
724 if (mp->m_flags & XFS_MOUNT_WSYNC)
725 xfs_trans_set_sync(tp);
726
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000727 code = xfs_trans_commit(tp, commit_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 xfs_iunlock(ip, lock_flags);
731
732 /*
733 * Release any dquot(s) the inode had kept before chown.
734 */
735 XFS_QM_DQRELE(mp, olddquot1);
736 XFS_QM_DQRELE(mp, olddquot2);
737 XFS_QM_DQRELE(mp, udqp);
738 XFS_QM_DQRELE(mp, gdqp);
739
740 if (code) {
741 return code;
742 }
743
Christoph Hellwigeb9df392007-08-16 18:42:07 +1000744 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 !(flags & ATTR_DMI)) {
Christoph Hellwigbc4ac742008-03-06 13:45:58 +1100746 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 NULL, DM_RIGHT_NULL, NULL, NULL,
748 0, 0, AT_DELAY_FLAG(flags));
749 }
750 return 0;
751
752 abort_return:
753 commit_flags |= XFS_TRANS_ABORT;
754 /* FALLTHROUGH */
755 error_return:
756 XFS_QM_DQRELE(mp, udqp);
757 XFS_QM_DQRELE(mp, gdqp);
758 if (tp) {
759 xfs_trans_cancel(tp, commit_flags);
760 }
761 if (lock_flags != 0) {
762 xfs_iunlock(ip, lock_flags);
763 }
764 return code;
765}
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767/*
Nathan Scott7d4fb402006-06-09 15:27:16 +1000768 * The maximum pathlen is 1024 bytes. Since the minimum file system
769 * blocksize is 512 bytes, we can get a max of 2 extents back from
770 * bmapi.
771 */
772#define SYMLINK_MAPS 2
773
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000774STATIC int
775xfs_readlink_bmap(
776 xfs_inode_t *ip,
777 char *link)
778{
779 xfs_mount_t *mp = ip->i_mount;
780 int pathlen = ip->i_d.di_size;
781 int nmaps = SYMLINK_MAPS;
782 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
783 xfs_daddr_t d;
784 int byte_cnt;
785 int n;
786 xfs_buf_t *bp;
787 int error = 0;
788
789 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
790 mval, &nmaps, NULL, NULL);
791 if (error)
792 goto out;
793
794 for (n = 0; n < nmaps; n++) {
795 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
796 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
797
798 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0);
799 error = XFS_BUF_GETERROR(bp);
800 if (error) {
801 xfs_ioerror_alert("xfs_readlink",
802 ip->i_mount, bp, XFS_BUF_ADDR(bp));
803 xfs_buf_relse(bp);
804 goto out;
805 }
806 if (pathlen < byte_cnt)
807 byte_cnt = pathlen;
808 pathlen -= byte_cnt;
809
810 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
811 xfs_buf_relse(bp);
812 }
813
814 link[ip->i_d.di_size] = '\0';
815 error = 0;
816
817 out:
818 return error;
819}
820
Christoph Hellwig993386c2007-08-28 16:12:30 +1000821int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822xfs_readlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000823 xfs_inode_t *ip,
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000824 char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000826 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 int pathlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100830 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
832 if (XFS_FORCED_SHUTDOWN(mp))
833 return XFS_ERROR(EIO);
834
835 xfs_ilock(ip, XFS_ILOCK_SHARED);
836
837 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000838 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000840 pathlen = ip->i_d.di_size;
841 if (!pathlen)
842 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
844 if (ip->i_df.if_flags & XFS_IFINLINE) {
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000845 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
846 link[pathlen] = '\0';
847 } else {
848 error = xfs_readlink_bmap(ip, link);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 }
850
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000851 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 return error;
854}
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856/*
857 * xfs_fsync
858 *
859 * This is called to sync the inode and its data out to disk.
860 * We need to hold the I/O lock while flushing the data, and
861 * the inode lock while flushing the inode. The inode lock CANNOT
862 * be held while flushing the data, so acquire after we're done
863 * with that.
864 */
Christoph Hellwig993386c2007-08-28 16:12:30 +1000865int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866xfs_fsync(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000867 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 int flag,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 xfs_off_t start,
870 xfs_off_t stop)
871{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 xfs_trans_t *tp;
873 int error;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100874 int log_flushed = 0, changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100876 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
878 ASSERT(start >= 0 && stop >= -1);
879
880 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
881 return XFS_ERROR(EIO);
882
Lachlan McIlroy776a75fa2007-09-14 15:22:50 +1000883 if (flag & FSYNC_DATA)
884 filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping);
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 /*
887 * We always need to make sure that the required inode state
888 * is safe on disk. The vnode might be clean but because
889 * of committed transactions that haven't hit the disk yet.
890 * Likewise, there could be unflushed non-transactional
891 * changes to the inode core that have to go to disk.
892 *
893 * The following code depends on one assumption: that
894 * any transaction that changes an inode logs the core
895 * because it has to change some field in the inode core
896 * (typically nextents or nblocks). That assumption
897 * implies that any transactions against an inode will
898 * catch any non-transactional updates. If inode-altering
899 * transactions exist that violate this assumption, the
900 * code breaks. Right now, it figures that if the involved
901 * update_* field is clear and the inode is unpinned, the
902 * inode is clean. Either it's been flushed or it's been
903 * committed and the commit has hit the disk unpinning the inode.
904 * (Note that xfs_inode_item_format() called at commit clears
905 * the update_* fields.)
906 */
907 xfs_ilock(ip, XFS_ILOCK_SHARED);
908
909 /* If we are flushing data then we care about update_size
910 * being set, otherwise we care about update_core
911 */
912 if ((flag & FSYNC_DATA) ?
913 (ip->i_update_size == 0) :
914 (ip->i_update_core == 0)) {
915 /*
916 * Timestamps/size haven't changed since last inode
917 * flush or inode transaction commit. That means
918 * either nothing got written or a transaction
919 * committed which caught the updates. If the
920 * latter happened and the transaction hasn't
921 * hit the disk yet, the inode will be still
922 * be pinned. If it is, force the log.
923 */
924
925 xfs_iunlock(ip, XFS_ILOCK_SHARED);
926
927 if (xfs_ipincount(ip)) {
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100928 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 XFS_LOG_FORCE |
930 ((flag & FSYNC_WAIT)
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100931 ? XFS_LOG_SYNC : 0),
932 &log_flushed);
933 } else {
934 /*
935 * If the inode is not pinned and nothing
936 * has changed we don't need to flush the
937 * cache.
938 */
939 changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 }
941 error = 0;
942 } else {
943 /*
944 * Kick off a transaction to log the inode
945 * core to get the updates. Make it
946 * sync if FSYNC_WAIT is passed in (which
947 * is done by everybody but specfs). The
948 * sync transaction will also force the log.
949 */
950 xfs_iunlock(ip, XFS_ILOCK_SHARED);
951 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
952 if ((error = xfs_trans_reserve(tp, 0,
953 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
954 0, 0, 0))) {
955 xfs_trans_cancel(tp, 0);
956 return error;
957 }
958 xfs_ilock(ip, XFS_ILOCK_EXCL);
959
960 /*
961 * Note - it's possible that we might have pushed
962 * ourselves out of the way during trans_reserve
963 * which would flush the inode. But there's no
964 * guarantee that the inode buffer has actually
965 * gone out yet (it's delwri). Plus the buffer
966 * could be pinned anyway if it's part of an
967 * inode in another recent transaction. So we
968 * play it safe and fire off the transaction anyway.
969 */
970 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
971 xfs_trans_ihold(tp, ip);
972 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
973 if (flag & FSYNC_WAIT)
974 xfs_trans_set_sync(tp);
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000975 error = _xfs_trans_commit(tp, 0, &log_flushed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
977 xfs_iunlock(ip, XFS_ILOCK_EXCL);
978 }
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100979
980 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
981 /*
982 * If the log write didn't issue an ordered tag we need
983 * to flush the disk cache for the data device now.
984 */
985 if (!log_flushed)
986 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
987
988 /*
989 * If this inode is on the RT dev we need to flush that
Nathan Scottc41564b2006-03-29 08:55:14 +1000990 * cache as well.
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100991 */
Eric Sandeen71ddabb2007-11-23 16:29:42 +1100992 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100993 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
994 }
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 return error;
997}
998
999/*
David Chinner92dfe8d2007-05-24 15:22:19 +10001000 * This is called by xfs_inactive to free any blocks beyond eof
1001 * when the link count isn't zero and by xfs_dm_punch_hole() when
1002 * punching a hole to EOF.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 */
David Chinner92dfe8d2007-05-24 15:22:19 +10001004int
1005xfs_free_eofblocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 xfs_mount_t *mp,
David Chinner92dfe8d2007-05-24 15:22:19 +10001007 xfs_inode_t *ip,
1008 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009{
1010 xfs_trans_t *tp;
1011 int error;
1012 xfs_fileoff_t end_fsb;
1013 xfs_fileoff_t last_fsb;
1014 xfs_filblks_t map_len;
1015 int nimaps;
1016 xfs_bmbt_irec_t imap;
David Chinner92dfe8d2007-05-24 15:22:19 +10001017 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
1019 /*
1020 * Figure out if there are any blocks beyond the end
1021 * of the file. If not, then there is nothing to do.
1022 */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001023 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1025 map_len = last_fsb - end_fsb;
1026 if (map_len <= 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01001027 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 nimaps = 1;
1030 xfs_ilock(ip, XFS_ILOCK_SHARED);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10001031 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001032 NULL, 0, &imap, &nimaps, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1034
1035 if (!error && (nimaps != 0) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001036 (imap.br_startblock != HOLESTARTBLOCK ||
1037 ip->i_delayed_blks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 /*
1039 * Attach the dquots to the inode up front.
1040 */
1041 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001042 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
1044 /*
1045 * There are blocks after the end of file.
1046 * Free them up now by truncating the file to
1047 * its current size.
1048 */
1049 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1050
1051 /*
1052 * Do the xfs_itruncate_start() call before
1053 * reserving any log space because
1054 * itruncate_start will call into the buffer
1055 * cache and we can't
1056 * do that within a transaction.
1057 */
David Chinner92dfe8d2007-05-24 15:22:19 +10001058 if (use_iolock)
1059 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001060 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001061 ip->i_size);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001062 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +10001063 xfs_trans_cancel(tp, 0);
David Chinner92dfe8d2007-05-24 15:22:19 +10001064 if (use_iolock)
1065 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001066 return error;
1067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
1069 error = xfs_trans_reserve(tp, 0,
1070 XFS_ITRUNCATE_LOG_RES(mp),
1071 0, XFS_TRANS_PERM_LOG_RES,
1072 XFS_ITRUNCATE_LOG_COUNT);
1073 if (error) {
1074 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1075 xfs_trans_cancel(tp, 0);
1076 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001077 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 }
1079
1080 xfs_ilock(ip, XFS_ILOCK_EXCL);
1081 xfs_trans_ijoin(tp, ip,
1082 XFS_IOLOCK_EXCL |
1083 XFS_ILOCK_EXCL);
1084 xfs_trans_ihold(tp, ip);
1085
1086 error = xfs_itruncate_finish(&tp, ip,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001087 ip->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 XFS_DATA_FORK,
1089 0);
1090 /*
1091 * If we get an error at this point we
1092 * simply don't bother truncating the file.
1093 */
1094 if (error) {
1095 xfs_trans_cancel(tp,
1096 (XFS_TRANS_RELEASE_LOG_RES |
1097 XFS_TRANS_ABORT));
1098 } else {
1099 error = xfs_trans_commit(tp,
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001100 XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 }
David Chinner92dfe8d2007-05-24 15:22:19 +10001102 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
1103 : XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001105 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106}
1107
1108/*
1109 * Free a symlink that has blocks associated with it.
1110 */
1111STATIC int
1112xfs_inactive_symlink_rmt(
1113 xfs_inode_t *ip,
1114 xfs_trans_t **tpp)
1115{
1116 xfs_buf_t *bp;
1117 int committed;
1118 int done;
1119 int error;
1120 xfs_fsblock_t first_block;
1121 xfs_bmap_free_t free_list;
1122 int i;
1123 xfs_mount_t *mp;
1124 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1125 int nmaps;
1126 xfs_trans_t *ntp;
1127 int size;
1128 xfs_trans_t *tp;
1129
1130 tp = *tpp;
1131 mp = ip->i_mount;
1132 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1133 /*
1134 * We're freeing a symlink that has some
1135 * blocks allocated to it. Free the
1136 * blocks here. We know that we've got
1137 * either 1 or 2 extents and that we can
1138 * free them all in one bunmapi call.
1139 */
1140 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1141 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1142 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1143 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1144 xfs_trans_cancel(tp, 0);
1145 *tpp = NULL;
1146 return error;
1147 }
1148 /*
1149 * Lock the inode, fix the size, and join it to the transaction.
1150 * Hold it so in the normal path, we still have it locked for
1151 * the second transaction. In the error paths we need it
1152 * held so the cancel won't rele it, see below.
1153 */
1154 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1155 size = (int)ip->i_d.di_size;
1156 ip->i_d.di_size = 0;
1157 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1158 xfs_trans_ihold(tp, ip);
1159 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1160 /*
1161 * Find the block(s) so we can inval and unmap them.
1162 */
1163 done = 0;
1164 XFS_BMAP_INIT(&free_list, &first_block);
Tobias Klausere8c96f82006-03-24 03:15:34 -08001165 nmaps = ARRAY_SIZE(mval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1167 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001168 &free_list, NULL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 goto error0;
1170 /*
1171 * Invalidate the block(s).
1172 */
1173 for (i = 0; i < nmaps; i++) {
1174 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1175 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1176 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1177 xfs_trans_binval(tp, bp);
1178 }
1179 /*
1180 * Unmap the dead block(s) to the free_list.
1181 */
1182 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001183 &first_block, &free_list, NULL, &done)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 goto error1;
1185 ASSERT(done);
1186 /*
1187 * Commit the first transaction. This logs the EFI and the inode.
1188 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001189 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 goto error1;
1191 /*
1192 * The transaction must have been committed, since there were
1193 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1194 * The new tp has the extent freeing and EFDs.
1195 */
1196 ASSERT(committed);
1197 /*
1198 * The first xact was committed, so add the inode to the new one.
1199 * Mark it dirty so it will be logged and moved forward in the log as
1200 * part of every commit.
1201 */
1202 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1203 xfs_trans_ihold(tp, ip);
1204 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1205 /*
1206 * Get a new, empty transaction to return to our caller.
1207 */
1208 ntp = xfs_trans_dup(tp);
1209 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001210 * Commit the transaction containing extent freeing and EFDs.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 * If we get an error on the commit here or on the reserve below,
1212 * we need to unlock the inode since the new transaction doesn't
1213 * have the inode attached.
1214 */
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001215 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 tp = ntp;
1217 if (error) {
1218 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1219 goto error0;
1220 }
1221 /*
1222 * Remove the memory for extent descriptions (just bookkeeping).
1223 */
1224 if (ip->i_df.if_bytes)
1225 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1226 ASSERT(ip->i_df.if_bytes == 0);
1227 /*
1228 * Put an itruncate log reservation in the new transaction
1229 * for our caller.
1230 */
1231 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1232 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1233 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1234 goto error0;
1235 }
1236 /*
1237 * Return with the inode locked but not joined to the transaction.
1238 */
1239 *tpp = tp;
1240 return 0;
1241
1242 error1:
1243 xfs_bmap_cancel(&free_list);
1244 error0:
1245 /*
1246 * Have to come here with the inode locked and either
1247 * (held and in the transaction) or (not in the transaction).
1248 * If the inode isn't held then cancel would iput it, but
1249 * that's wrong since this is inactive and the vnode ref
1250 * count is 0 already.
1251 * Cancel won't do anything to the inode if held, but it still
1252 * needs to be locked until the cancel is done, if it was
1253 * joined to the transaction.
1254 */
1255 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1256 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1257 *tpp = NULL;
1258 return error;
1259
1260}
1261
1262STATIC int
1263xfs_inactive_symlink_local(
1264 xfs_inode_t *ip,
1265 xfs_trans_t **tpp)
1266{
1267 int error;
1268
1269 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1270 /*
1271 * We're freeing a symlink which fit into
1272 * the inode. Just free the memory used
1273 * to hold the old symlink.
1274 */
1275 error = xfs_trans_reserve(*tpp, 0,
1276 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1277 0, XFS_TRANS_PERM_LOG_RES,
1278 XFS_ITRUNCATE_LOG_COUNT);
1279
1280 if (error) {
1281 xfs_trans_cancel(*tpp, 0);
1282 *tpp = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +01001283 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 }
1285 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1286
1287 /*
1288 * Zero length symlinks _can_ exist.
1289 */
1290 if (ip->i_df.if_bytes > 0) {
1291 xfs_idata_realloc(ip,
1292 -(ip->i_df.if_bytes),
1293 XFS_DATA_FORK);
1294 ASSERT(ip->i_df.if_bytes == 0);
1295 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001296 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297}
1298
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299STATIC int
1300xfs_inactive_attrs(
1301 xfs_inode_t *ip,
1302 xfs_trans_t **tpp)
1303{
1304 xfs_trans_t *tp;
1305 int error;
1306 xfs_mount_t *mp;
1307
1308 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1309 tp = *tpp;
1310 mp = ip->i_mount;
1311 ASSERT(ip->i_d.di_forkoff != 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001312 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnere5720ee2008-04-10 12:21:18 +10001314 if (error)
1315 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317 error = xfs_attr_inactive(ip);
David Chinnere5720ee2008-04-10 12:21:18 +10001318 if (error)
1319 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
1321 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1322 error = xfs_trans_reserve(tp, 0,
1323 XFS_IFREE_LOG_RES(mp),
1324 0, XFS_TRANS_PERM_LOG_RES,
1325 XFS_INACTIVE_LOG_COUNT);
David Chinnere5720ee2008-04-10 12:21:18 +10001326 if (error)
1327 goto error_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
1329 xfs_ilock(ip, XFS_ILOCK_EXCL);
1330 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1331 xfs_trans_ihold(tp, ip);
1332 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1333
1334 ASSERT(ip->i_d.di_anextents == 0);
1335
1336 *tpp = tp;
Jesper Juhl014c2542006-01-15 02:37:08 +01001337 return 0;
David Chinnere5720ee2008-04-10 12:21:18 +10001338
1339error_cancel:
1340 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1341 xfs_trans_cancel(tp, 0);
1342error_unlock:
1343 *tpp = NULL;
1344 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1345 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346}
1347
Christoph Hellwig993386c2007-08-28 16:12:30 +10001348int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349xfs_release(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001350 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001352 bhv_vnode_t *vp = XFS_ITOV(ip);
1353 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 int error;
1355
Christoph Hellwig42173f62008-04-22 17:33:25 +10001356 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001360 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 return 0;
1362
David Chinner2a82b8b2007-07-11 11:09:12 +10001363 if (!XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001364 int truncated;
1365
David Chinner2a82b8b2007-07-11 11:09:12 +10001366 /*
1367 * If we are using filestreams, and we have an unlinked
1368 * file that we are processing the last close on, then nothing
1369 * will be able to reopen and write to this file. Purge this
1370 * inode from the filestreams cache so that it doesn't delay
1371 * teardown of the inode.
1372 */
1373 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1374 xfs_filestream_deassociate(ip);
1375
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +10001376 /*
1377 * If we previously truncated this file and removed old data
1378 * in the process, we want to initiate "early" writeout on
1379 * the last close. This is an attempt to combat the notorious
1380 * NULL files problem which is particularly noticable from a
1381 * truncate down, buffered (re-)write (delalloc), followed by
1382 * a crash. What we are effectively doing here is
1383 * significantly reducing the time window where we'd otherwise
1384 * be exposed to that problem.
1385 */
Christoph Hellwig09262b42007-08-29 11:44:50 +10001386 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001387 if (truncated && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001388 xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +10001389 }
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 if (ip->i_d.di_nlink != 0) {
1392 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001393 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001394 ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001396 (!(ip->i_d.di_flags &
1397 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
David Chinner92dfe8d2007-05-24 15:22:19 +10001398 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1399 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001400 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 }
1402 }
1403
1404 return 0;
1405}
1406
1407/*
1408 * xfs_inactive
1409 *
1410 * This is called when the vnode reference count for the vnode
1411 * goes to zero. If the file has been unlinked, then it must
1412 * now be truncated. Also, we clear all of the read-ahead state
1413 * kept for the inode here since the file is now closed.
1414 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001415int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416xfs_inactive(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001417 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001419 bhv_vnode_t *vp = XFS_ITOV(ip);
Nathan Scott67fcaa72006-06-09 17:00:52 +10001420 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 xfs_fsblock_t first_block;
1422 int committed;
1423 xfs_trans_t *tp;
1424 xfs_mount_t *mp;
1425 int error;
1426 int truncate;
1427
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001428 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 /*
1431 * If the inode is already free, then there can be nothing
1432 * to clean up here.
1433 */
1434 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1435 ASSERT(ip->i_df.if_real_bytes == 0);
1436 ASSERT(ip->i_df.if_broot_bytes == 0);
1437 return VN_INACTIVE_CACHE;
1438 }
1439
1440 /*
1441 * Only do a truncate if it's a regular file with
1442 * some actual space in it. It's OK to look at the
1443 * inode's fields without the lock because we're the
1444 * only one with a reference to the inode.
1445 */
1446 truncate = ((ip->i_d.di_nlink == 0) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001447 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1448 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1450
1451 mp = ip->i_mount;
1452
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11001453 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY))
1454 XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
1456 error = 0;
1457
1458 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001459 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 goto out;
1461
1462 if (ip->i_d.di_nlink != 0) {
1463 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001464 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001465 ip->i_delayed_blks > 0)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001466 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1467 (!(ip->i_d.di_flags &
1468 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1469 (ip->i_delayed_blks != 0)))) {
David Chinner92dfe8d2007-05-24 15:22:19 +10001470 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1471 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001472 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 }
1474 goto out;
1475 }
1476
1477 ASSERT(ip->i_d.di_nlink == 0);
1478
1479 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001480 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
1482 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1483 if (truncate) {
1484 /*
1485 * Do the xfs_itruncate_start() call before
1486 * reserving any log space because itruncate_start
1487 * will call into the buffer cache and we can't
1488 * do that within a transaction.
1489 */
1490 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1491
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001492 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1493 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +10001494 xfs_trans_cancel(tp, 0);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001495 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1496 return VN_INACTIVE_CACHE;
1497 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
1499 error = xfs_trans_reserve(tp, 0,
1500 XFS_ITRUNCATE_LOG_RES(mp),
1501 0, XFS_TRANS_PERM_LOG_RES,
1502 XFS_ITRUNCATE_LOG_COUNT);
1503 if (error) {
1504 /* Don't call itruncate_cleanup */
1505 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1506 xfs_trans_cancel(tp, 0);
1507 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001508 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 }
1510
1511 xfs_ilock(ip, XFS_ILOCK_EXCL);
1512 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1513 xfs_trans_ihold(tp, ip);
1514
1515 /*
1516 * normally, we have to run xfs_itruncate_finish sync.
1517 * But if filesystem is wsync and we're in the inactive
1518 * path, then we know that nlink == 0, and that the
1519 * xaction that made nlink == 0 is permanently committed
1520 * since xfs_remove runs as a synchronous transaction.
1521 */
1522 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1523 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1524
1525 if (error) {
1526 xfs_trans_cancel(tp,
1527 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1528 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001529 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 }
1531 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1532
1533 /*
1534 * If we get an error while cleaning up a
1535 * symlink we bail out.
1536 */
1537 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1538 xfs_inactive_symlink_rmt(ip, &tp) :
1539 xfs_inactive_symlink_local(ip, &tp);
1540
1541 if (error) {
1542 ASSERT(tp == NULL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001543 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 }
1545
1546 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1547 xfs_trans_ihold(tp, ip);
1548 } else {
1549 error = xfs_trans_reserve(tp, 0,
1550 XFS_IFREE_LOG_RES(mp),
1551 0, XFS_TRANS_PERM_LOG_RES,
1552 XFS_INACTIVE_LOG_COUNT);
1553 if (error) {
1554 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1555 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001556 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 }
1558
1559 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1560 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1561 xfs_trans_ihold(tp, ip);
1562 }
1563
1564 /*
1565 * If there are attributes associated with the file
1566 * then blow them away now. The code calls a routine
1567 * that recursively deconstructs the attribute fork.
1568 * We need to just commit the current transaction
1569 * because we can't use it for xfs_attr_inactive().
1570 */
1571 if (ip->i_d.di_anextents > 0) {
1572 error = xfs_inactive_attrs(ip, &tp);
1573 /*
1574 * If we got an error, the transaction is already
1575 * cancelled, and the inode is unlocked. Just get out.
1576 */
1577 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001578 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 } else if (ip->i_afp) {
1580 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1581 }
1582
1583 /*
1584 * Free the inode.
1585 */
1586 XFS_BMAP_INIT(&free_list, &first_block);
1587 error = xfs_ifree(tp, ip, &free_list);
1588 if (error) {
1589 /*
1590 * If we fail to free the inode, shut down. The cancel
1591 * might do that, we need to make sure. Otherwise the
1592 * inode might be lost for a long time or forever.
1593 */
1594 if (!XFS_FORCED_SHUTDOWN(mp)) {
1595 cmn_err(CE_NOTE,
1596 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1597 error, mp->m_fsname);
Nathan Scott7d04a332006-06-09 14:58:38 +10001598 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 }
1600 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1601 } else {
1602 /*
1603 * Credit the quota account(s). The inode is gone.
1604 */
1605 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1606
1607 /*
David Chinner78e9da72008-04-10 12:24:17 +10001608 * Just ignore errors at this point. There is nothing we can
1609 * do except to try to keep going. Make sure it's not a silent
1610 * error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 */
David Chinner78e9da72008-04-10 12:24:17 +10001612 error = xfs_bmap_finish(&tp, &free_list, &committed);
1613 if (error)
1614 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1615 "xfs_bmap_finish() returned error %d", error);
1616 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1617 if (error)
1618 xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: "
1619 "xfs_trans_commit() returned error %d", error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 }
1621 /*
1622 * Release the dquots held by inode, if any.
1623 */
1624 XFS_QM_DQDETACH(mp, ip);
1625
1626 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1627
1628 out:
1629 return VN_INACTIVE_CACHE;
1630}
1631
1632
Christoph Hellwig993386c2007-08-28 16:12:30 +10001633int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634xfs_lookup(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001635 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001636 struct xfs_name *name,
Christoph Hellwigef1f5e72008-03-06 13:46:25 +11001637 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638{
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001639 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 int error;
1641 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001643 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
1645 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1646 return XFS_ERROR(EIO);
1647
1648 lock_mode = xfs_ilock_map_shared(dp);
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001649 error = xfs_dir_lookup(NULL, dp, name, &inum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 xfs_iunlock_map_shared(dp, lock_mode);
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001651
1652 if (error)
1653 goto out;
1654
1655 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0);
1656 if (error)
1657 goto out;
1658
1659 xfs_itrace_ref(*ipp);
1660 return 0;
1661
1662 out:
1663 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 return error;
1665}
1666
Christoph Hellwig993386c2007-08-28 16:12:30 +10001667int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668xfs_create(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001669 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001670 struct xfs_name *name,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001671 mode_t mode,
1672 xfs_dev_t rdev,
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001673 xfs_inode_t **ipp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 cred_t *credp)
1675{
Barry Naujok556b8b12008-04-10 12:22:07 +10001676 xfs_mount_t *mp = dp->i_mount;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001677 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 xfs_trans_t *tp;
Barry Naujok556b8b12008-04-10 12:22:07 +10001679 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 xfs_bmap_free_t free_list;
1681 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001682 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 int dm_event_sent = 0;
1684 uint cancel_flags;
1685 int committed;
1686 xfs_prid_t prid;
1687 struct xfs_dquot *udqp, *gdqp;
1688 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001690 ASSERT(!*ipp);
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001691 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
Christoph Hellwigeb9df392007-08-16 18:42:07 +10001693 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11001695 dp, DM_RIGHT_NULL, NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10001696 DM_RIGHT_NULL, name->name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001697 mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
1699 if (error)
1700 return error;
1701 dm_event_sent = 1;
1702 }
1703
1704 if (XFS_FORCED_SHUTDOWN(mp))
1705 return XFS_ERROR(EIO);
1706
1707 /* Return through std_return after this point. */
1708
1709 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10001710 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1711 prid = dp->i_d.di_projid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 else
1713 prid = (xfs_prid_t)dfltprid;
1714
1715 /*
1716 * Make sure that we have allocated dquot(s) on disk.
1717 */
1718 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001719 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1721 if (error)
1722 goto std_return;
1723
1724 ip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725
1726 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1727 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Barry Naujok556b8b12008-04-10 12:22:07 +10001728 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 /*
1730 * Initially assume that the file does not exist and
1731 * reserve the resources for that case. If that is not
1732 * the case we'll drop the one we have and get a more
1733 * appropriate transaction later.
1734 */
1735 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1736 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1737 if (error == ENOSPC) {
1738 resblks = 0;
1739 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1740 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1741 }
1742 if (error) {
1743 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 goto error_return;
1745 }
1746
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001747 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001748 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
1750 XFS_BMAP_INIT(&free_list, &first_block);
1751
1752 ASSERT(ip == NULL);
1753
1754 /*
1755 * Reserve disk quota and the inode.
1756 */
1757 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1758 if (error)
1759 goto error_return;
1760
Barry Naujok556b8b12008-04-10 12:22:07 +10001761 error = xfs_dir_canenter(tp, dp, name, resblks);
1762 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 goto error_return;
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001764 error = xfs_dir_ialloc(&tp, dp, mode, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 rdev, credp, prid, resblks > 0,
1766 &ip, &committed);
1767 if (error) {
1768 if (error == ENOSPC)
1769 goto error_return;
1770 goto abort_return;
1771 }
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001772 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774 /*
1775 * At this point, we've gotten a newly allocated inode.
1776 * It is locked (and joined to the transaction).
1777 */
1778
1779 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1780
1781 /*
Christoph Hellwig993386c2007-08-28 16:12:30 +10001782 * Now we join the directory inode to the transaction. We do not do it
1783 * earlier because xfs_dir_ialloc might commit the previous transaction
1784 * (and release all the locks). An error from here on will result in
1785 * the transaction cancel unlocking dp so don't do it explicitly in the
1786 * error path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 */
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001788 IHOLD(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001790 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Barry Naujok556b8b12008-04-10 12:22:07 +10001792 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001793 &first_block, &free_list, resblks ?
1794 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 if (error) {
1796 ASSERT(error != ENOSPC);
1797 goto abort_return;
1798 }
1799 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1800 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1801
1802 /*
1803 * If this is a synchronous mount, make sure that the
1804 * create transaction goes to disk before returning to
1805 * the user.
1806 */
1807 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1808 xfs_trans_set_sync(tp);
1809 }
1810
1811 dp->i_gen++;
1812
1813 /*
1814 * Attach the dquot(s) to the inodes and modify them incore.
1815 * These ids of the inode couldn't have changed since the new
1816 * inode has been locked ever since it was created.
1817 */
1818 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
1819
1820 /*
1821 * xfs_trans_commit normally decrements the vnode ref count
1822 * when it unlocks the inode. Since we want to return the
1823 * vnode to the caller, we bump the vnode ref count now.
1824 */
1825 IHOLD(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001827 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 if (error) {
1829 xfs_bmap_cancel(&free_list);
1830 goto abort_rele;
1831 }
1832
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001833 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 if (error) {
1835 IRELE(ip);
1836 tp = NULL;
1837 goto error_return;
1838 }
1839
1840 XFS_QM_DQRELE(mp, udqp);
1841 XFS_QM_DQRELE(mp, gdqp);
1842
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001843 *ipp = ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844
1845 /* Fallthrough to std_return with error = 0 */
1846
1847std_return:
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001848 if ((*ipp || (error != 0 && dm_event_sent != 0)) &&
Christoph Hellwig993386c2007-08-28 16:12:30 +10001849 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11001851 dp, DM_RIGHT_NULL,
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001852 *ipp ? ip : NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10001853 DM_RIGHT_NULL, name->name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001854 mode, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 }
1856 return error;
1857
1858 abort_return:
1859 cancel_flags |= XFS_TRANS_ABORT;
1860 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
Jesper Juhl014c2542006-01-15 02:37:08 +01001862 error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 if (tp != NULL)
1864 xfs_trans_cancel(tp, cancel_flags);
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 XFS_QM_DQRELE(mp, udqp);
1867 XFS_QM_DQRELE(mp, gdqp);
1868
Christoph Hellwig993386c2007-08-28 16:12:30 +10001869 if (unlock_dp_on_error)
1870 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1871
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 goto std_return;
1873
1874 abort_rele:
1875 /*
1876 * Wait until after the current transaction is aborted to
1877 * release the inode. This prevents recursive transactions
1878 * and deadlocks from xfs_inactive.
1879 */
1880 cancel_flags |= XFS_TRANS_ABORT;
1881 xfs_trans_cancel(tp, cancel_flags);
1882 IRELE(ip);
1883
1884 XFS_QM_DQRELE(mp, udqp);
1885 XFS_QM_DQRELE(mp, gdqp);
1886
1887 goto std_return;
1888}
1889
1890#ifdef DEBUG
1891/*
1892 * Some counters to see if (and how often) we are hitting some deadlock
1893 * prevention code paths.
1894 */
1895
1896int xfs_rm_locks;
1897int xfs_rm_lock_delays;
1898int xfs_rm_attempts;
1899#endif
1900
1901/*
1902 * The following routine will lock the inodes associated with the
1903 * directory and the named entry in the directory. The locks are
1904 * acquired in increasing inode number.
1905 *
1906 * If the entry is "..", then only the directory is locked. The
1907 * vnode ref count will still include that from the .. entry in
1908 * this case.
1909 *
1910 * There is a deadlock we need to worry about. If the locked directory is
1911 * in the AIL, it might be blocking up the log. The next inode we lock
1912 * could be already locked by another thread waiting for log space (e.g
1913 * a permanent log reservation with a long running transaction (see
1914 * xfs_itruncate_finish)). To solve this, we must check if the directory
1915 * is in the ail and use lock_nowait. If we can't lock, we need to
1916 * drop the inode lock on the directory and try again. xfs_iunlock will
1917 * potentially push the tail if we were holding up the log.
1918 */
1919STATIC int
1920xfs_lock_dir_and_entry(
1921 xfs_inode_t *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 xfs_inode_t *ip) /* inode of entry 'name' */
1923{
1924 int attempts;
1925 xfs_ino_t e_inum;
1926 xfs_inode_t *ips[2];
1927 xfs_log_item_t *lp;
1928
1929#ifdef DEBUG
1930 xfs_rm_locks++;
1931#endif
1932 attempts = 0;
1933
1934again:
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001935 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
1937 e_inum = ip->i_ino;
1938
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001939 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
1941 /*
1942 * We want to lock in increasing inum. Since we've already
1943 * acquired the lock on the directory, we may need to release
1944 * if if the inum of the entry turns out to be less.
1945 */
1946 if (e_inum > dp->i_ino) {
1947 /*
1948 * We are already in the right order, so just
1949 * lock on the inode of the entry.
1950 * We need to use nowait if dp is in the AIL.
1951 */
1952
1953 lp = (xfs_log_item_t *)dp->i_itemp;
1954 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1955 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1956 attempts++;
1957#ifdef DEBUG
1958 xfs_rm_attempts++;
1959#endif
1960
1961 /*
1962 * Unlock dp and try again.
1963 * xfs_iunlock will try to push the tail
1964 * if the inode is in the AIL.
1965 */
1966
1967 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1968
1969 if ((attempts % 5) == 0) {
1970 delay(1); /* Don't just spin the CPU */
1971#ifdef DEBUG
1972 xfs_rm_lock_delays++;
1973#endif
1974 }
1975 goto again;
1976 }
1977 } else {
1978 xfs_ilock(ip, XFS_ILOCK_EXCL);
1979 }
1980 } else if (e_inum < dp->i_ino) {
1981 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1982
1983 ips[0] = ip;
1984 ips[1] = dp;
1985 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
1986 }
1987 /* else e_inum == dp->i_ino */
1988 /* This can happen if we're asked to lock /x/..
1989 * the entry is "..", which is also the parent directory.
1990 */
1991
1992 return 0;
1993}
1994
1995#ifdef DEBUG
1996int xfs_locked_n;
1997int xfs_small_retries;
1998int xfs_middle_retries;
1999int xfs_lots_retries;
2000int xfs_lock_delays;
2001#endif
2002
2003/*
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002004 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2005 * a different value
2006 */
2007static inline int
2008xfs_lock_inumorder(int lock_mode, int subclass)
2009{
2010 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10002011 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002012 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10002013 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002014
2015 return lock_mode;
2016}
2017
2018/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 * The following routine will lock n inodes in exclusive mode.
2020 * We assume the caller calls us with the inodes in i_ino order.
2021 *
2022 * We need to detect deadlock where an inode that we lock
2023 * is in the AIL and we start waiting for another inode that is locked
2024 * by a thread in a long running transaction (such as truncate). This can
2025 * result in deadlock since the long running trans might need to wait
2026 * for the inode we just locked in order to push the tail and free space
2027 * in the log.
2028 */
2029void
2030xfs_lock_inodes(
2031 xfs_inode_t **ips,
2032 int inodes,
2033 int first_locked,
2034 uint lock_mode)
2035{
2036 int attempts = 0, i, j, try_lock;
2037 xfs_log_item_t *lp;
2038
2039 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2040
2041 if (first_locked) {
2042 try_lock = 1;
2043 i = 1;
2044 } else {
2045 try_lock = 0;
2046 i = 0;
2047 }
2048
2049again:
2050 for (; i < inodes; i++) {
2051 ASSERT(ips[i]);
2052
2053 if (i && (ips[i] == ips[i-1])) /* Already locked */
2054 continue;
2055
2056 /*
2057 * If try_lock is not set yet, make sure all locked inodes
2058 * are not in the AIL.
2059 * If any are, set try_lock to be used later.
2060 */
2061
2062 if (!try_lock) {
2063 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2064 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2065 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2066 try_lock++;
2067 }
2068 }
2069 }
2070
2071 /*
2072 * If any of the previous locks we have locked is in the AIL,
2073 * we must TRY to get the second and subsequent locks. If
2074 * we can't get any, we must release all we have
2075 * and try again.
2076 */
2077
2078 if (try_lock) {
2079 /* try_lock must be 0 if i is 0. */
2080 /*
2081 * try_lock means we have an inode locked
2082 * that is in the AIL.
2083 */
2084 ASSERT(i != 0);
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002085 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 attempts++;
2087
2088 /*
2089 * Unlock all previous guys and try again.
2090 * xfs_iunlock will try to push the tail
2091 * if the inode is in the AIL.
2092 */
2093
2094 for(j = i - 1; j >= 0; j--) {
2095
2096 /*
2097 * Check to see if we've already
2098 * unlocked this one.
2099 * Not the first one going back,
2100 * and the inode ptr is the same.
2101 */
2102 if ((j != (i - 1)) && ips[j] ==
2103 ips[j+1])
2104 continue;
2105
2106 xfs_iunlock(ips[j], lock_mode);
2107 }
2108
2109 if ((attempts % 5) == 0) {
2110 delay(1); /* Don't just spin the CPU */
2111#ifdef DEBUG
2112 xfs_lock_delays++;
2113#endif
2114 }
2115 i = 0;
2116 try_lock = 0;
2117 goto again;
2118 }
2119 } else {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002120 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 }
2122 }
2123
2124#ifdef DEBUG
2125 if (attempts) {
2126 if (attempts < 5) xfs_small_retries++;
2127 else if (attempts < 100) xfs_middle_retries++;
2128 else xfs_lots_retries++;
2129 } else {
2130 xfs_locked_n++;
2131 }
2132#endif
2133}
2134
2135#ifdef DEBUG
2136#define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2137int remove_which_error_return = 0;
2138#else /* ! DEBUG */
2139#define REMOVE_DEBUG_TRACE(x)
2140#endif /* ! DEBUG */
2141
Christoph Hellwig993386c2007-08-28 16:12:30 +10002142int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143xfs_remove(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002144 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10002145 struct xfs_name *name,
2146 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002148 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 xfs_trans_t *tp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 int error = 0;
2151 xfs_bmap_free_t free_list;
2152 xfs_fsblock_t first_block;
2153 int cancel_flags;
2154 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 int link_zero;
2156 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002158 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 if (XFS_FORCED_SHUTDOWN(mp))
2161 return XFS_ERROR(EIO);
2162
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002163 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
Barry Naujok556b8b12008-04-10 12:22:07 +10002164 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL,
2165 NULL, DM_RIGHT_NULL, name->name, NULL,
2166 ip->i_d.di_mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 if (error)
2168 return error;
2169 }
2170
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 /*
2172 * We need to get a reference to ip before we get our log
2173 * reservation. The reason for this is that we cannot call
2174 * xfs_iget for an inode for which we do not have a reference
2175 * once we've acquired a log reservation. This is because the
2176 * inode we are trying to get might be in xfs_inactive going
2177 * for a log reservation. Since we'll have to wait for the
2178 * inactive code to complete before returning from xfs_iget,
2179 * we need to make sure that we don't have log space reserved
Nathan Scottc41564b2006-03-29 08:55:14 +10002180 * when we call xfs_iget. Instead we get an unlocked reference
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 * to the inode before getting our log reservation.
2182 */
Christoph Hellwig43973962008-03-06 13:44:50 +11002183 IHOLD(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002185 xfs_itrace_entry(ip);
2186 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
2188 error = XFS_QM_DQATTACH(mp, dp, 0);
2189 if (!error && dp != ip)
2190 error = XFS_QM_DQATTACH(mp, ip, 0);
2191 if (error) {
2192 REMOVE_DEBUG_TRACE(__LINE__);
2193 IRELE(ip);
2194 goto std_return;
2195 }
2196
2197 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2198 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2199 /*
2200 * We try to get the real space reservation first,
2201 * allowing for directory btree deletion(s) implying
2202 * possible bmap insert(s). If we can't get the space
2203 * reservation then we use 0 instead, and avoid the bmap
2204 * btree insert(s) in the directory code by, if the bmap
2205 * insert tries to happen, instead trimming the LAST
2206 * block from the directory.
2207 */
2208 resblks = XFS_REMOVE_SPACE_RES(mp);
2209 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2210 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2211 if (error == ENOSPC) {
2212 resblks = 0;
2213 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2214 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2215 }
2216 if (error) {
2217 ASSERT(error != ENOSPC);
2218 REMOVE_DEBUG_TRACE(__LINE__);
2219 xfs_trans_cancel(tp, 0);
2220 IRELE(ip);
2221 return error;
2222 }
2223
Eric Sandeene9ed9d22007-05-08 13:48:56 +10002224 error = xfs_lock_dir_and_entry(dp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 if (error) {
2226 REMOVE_DEBUG_TRACE(__LINE__);
2227 xfs_trans_cancel(tp, cancel_flags);
2228 IRELE(ip);
2229 goto std_return;
2230 }
2231
2232 /*
2233 * At this point, we've gotten both the directory and the entry
2234 * inodes locked.
2235 */
2236 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2237 if (dp != ip) {
2238 /*
2239 * Increment vnode ref count only in this case since
2240 * there's an extra vnode reference in the case where
2241 * dp == ip.
2242 */
2243 IHOLD(dp);
2244 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2245 }
2246
2247 /*
2248 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2249 */
2250 XFS_BMAP_INIT(&free_list, &first_block);
Barry Naujok556b8b12008-04-10 12:22:07 +10002251 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
Christoph Hellwigd4377d82008-04-22 17:33:46 +10002252 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 if (error) {
2254 ASSERT(error != ENOENT);
2255 REMOVE_DEBUG_TRACE(__LINE__);
2256 goto error1;
2257 }
2258 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2259
2260 dp->i_gen++;
2261 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2262
2263 error = xfs_droplink(tp, ip);
2264 if (error) {
2265 REMOVE_DEBUG_TRACE(__LINE__);
2266 goto error1;
2267 }
2268
2269 /* Determine if this is the last link while
2270 * we are in the transaction.
2271 */
2272 link_zero = (ip)->i_d.di_nlink==0;
2273
2274 /*
2275 * Take an extra ref on the inode so that it doesn't
2276 * go to xfs_inactive() from within the commit.
2277 */
2278 IHOLD(ip);
2279
2280 /*
2281 * If this is a synchronous mount, make sure that the
2282 * remove transaction goes to disk before returning to
2283 * the user.
2284 */
2285 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2286 xfs_trans_set_sync(tp);
2287 }
2288
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002289 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 if (error) {
2291 REMOVE_DEBUG_TRACE(__LINE__);
2292 goto error_rele;
2293 }
2294
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002295 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 if (error) {
2297 IRELE(ip);
2298 goto std_return;
2299 }
2300
2301 /*
David Chinner2a82b8b2007-07-11 11:09:12 +10002302 * If we are using filestreams, kill the stream association.
2303 * If the file is still open it may get a new one but that
2304 * will get killed on last close in xfs_close() so we don't
2305 * have to worry about that.
2306 */
2307 if (link_zero && xfs_inode_is_filestream(ip))
2308 xfs_filestream_deassociate(ip);
2309
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002310 xfs_itrace_exit(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 IRELE(ip);
2312
2313/* Fall through to std_return with error = 0 */
2314 std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002315 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002317 dp, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 NULL, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002319 name->name, NULL, ip->i_d.di_mode, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 }
2321 return error;
2322
2323 error1:
2324 xfs_bmap_cancel(&free_list);
2325 cancel_flags |= XFS_TRANS_ABORT;
2326 xfs_trans_cancel(tp, cancel_flags);
2327 goto std_return;
2328
2329 error_rele:
2330 /*
2331 * In this case make sure to not release the inode until after
2332 * the current transaction is aborted. Releasing it beforehand
2333 * can cause us to go to xfs_inactive and start a recursive
2334 * transaction which can easily deadlock with the current one.
2335 */
2336 xfs_bmap_cancel(&free_list);
2337 cancel_flags |= XFS_TRANS_ABORT;
2338 xfs_trans_cancel(tp, cancel_flags);
2339
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 IRELE(ip);
2341
2342 goto std_return;
2343}
2344
Christoph Hellwig993386c2007-08-28 16:12:30 +10002345int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346xfs_link(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002347 xfs_inode_t *tdp,
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002348 xfs_inode_t *sip,
Barry Naujok556b8b12008-04-10 12:22:07 +10002349 struct xfs_name *target_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002351 xfs_mount_t *mp = tdp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 xfs_inode_t *ips[2];
2354 int error;
2355 xfs_bmap_free_t free_list;
2356 xfs_fsblock_t first_block;
2357 int cancel_flags;
2358 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 int resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002361 xfs_itrace_entry(tdp);
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002362 xfs_itrace_entry(sip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002364 ASSERT(!S_ISDIR(sip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 if (XFS_FORCED_SHUTDOWN(mp))
2367 return XFS_ERROR(EIO);
2368
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002369 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002371 tdp, DM_RIGHT_NULL,
2372 sip, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002373 target_name->name, NULL, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 if (error)
2375 return error;
2376 }
2377
2378 /* Return through std_return after this point. */
2379
2380 error = XFS_QM_DQATTACH(mp, sip, 0);
2381 if (!error && sip != tdp)
2382 error = XFS_QM_DQATTACH(mp, tdp, 0);
2383 if (error)
2384 goto std_return;
2385
2386 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2387 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Barry Naujok556b8b12008-04-10 12:22:07 +10002388 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2390 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2391 if (error == ENOSPC) {
2392 resblks = 0;
2393 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2394 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2395 }
2396 if (error) {
2397 cancel_flags = 0;
2398 goto error_return;
2399 }
2400
2401 if (sip->i_ino < tdp->i_ino) {
2402 ips[0] = sip;
2403 ips[1] = tdp;
2404 } else {
2405 ips[0] = tdp;
2406 ips[1] = sip;
2407 }
2408
2409 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2410
2411 /*
2412 * Increment vnode ref counts since xfs_trans_commit &
2413 * xfs_trans_cancel will both unlock the inodes and
2414 * decrement the associated ref counts.
2415 */
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002416 IHOLD(sip);
2417 IHOLD(tdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2419 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2420
2421 /*
2422 * If the source has too many links, we can't make any more to it.
2423 */
2424 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2425 error = XFS_ERROR(EMLINK);
2426 goto error_return;
2427 }
2428
Nathan Scott365ca832005-06-21 15:39:12 +10002429 /*
2430 * If we are using project inheritance, we only allow hard link
2431 * creation in our tree when the project IDs are the same; else
2432 * the tree quota mechanism could be circumvented.
2433 */
2434 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2435 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
Nathan Scottb1ecdda2006-05-08 19:51:42 +10002436 error = XFS_ERROR(EXDEV);
Nathan Scott365ca832005-06-21 15:39:12 +10002437 goto error_return;
2438 }
2439
Barry Naujok556b8b12008-04-10 12:22:07 +10002440 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
2441 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 goto error_return;
2443
2444 XFS_BMAP_INIT(&free_list, &first_block);
2445
Barry Naujok556b8b12008-04-10 12:22:07 +10002446 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
2447 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 if (error)
2449 goto abort_return;
2450 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2451 tdp->i_gen++;
2452 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2453
2454 error = xfs_bumplink(tp, sip);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002455 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 goto abort_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
2458 /*
2459 * If this is a synchronous mount, make sure that the
2460 * link transaction goes to disk before returning to
2461 * the user.
2462 */
2463 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2464 xfs_trans_set_sync(tp);
2465 }
2466
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002467 error = xfs_bmap_finish (&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 if (error) {
2469 xfs_bmap_cancel(&free_list);
2470 goto abort_return;
2471 }
2472
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002473 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002474 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
2477 /* Fall through to std_return with error = 0. */
2478std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002479 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002481 tdp, DM_RIGHT_NULL,
2482 sip, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002483 target_name->name, NULL, 0, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 }
2485 return error;
2486
2487 abort_return:
2488 cancel_flags |= XFS_TRANS_ABORT;
2489 /* FALLTHROUGH */
Jesper Juhl014c2542006-01-15 02:37:08 +01002490
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 error_return:
2492 xfs_trans_cancel(tp, cancel_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 goto std_return;
2494}
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002495
2496
Christoph Hellwig993386c2007-08-28 16:12:30 +10002497int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498xfs_mkdir(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002499 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10002500 struct xfs_name *dir_name,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002501 mode_t mode,
Christoph Hellwig979ebab2008-03-06 13:46:05 +11002502 xfs_inode_t **ipp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 cred_t *credp)
2504{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002505 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 xfs_inode_t *cdp; /* inode of created dir */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 int cancel_flags;
2509 int error;
2510 int committed;
2511 xfs_bmap_free_t free_list;
2512 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10002513 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 boolean_t created = B_FALSE;
2515 int dm_event_sent = 0;
2516 xfs_prid_t prid;
2517 struct xfs_dquot *udqp, *gdqp;
2518 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
2520 if (XFS_FORCED_SHUTDOWN(mp))
2521 return XFS_ERROR(EIO);
2522
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 tp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002525 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002527 dp, DM_RIGHT_NULL, NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002528 DM_RIGHT_NULL, dir_name->name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002529 mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (error)
2531 return error;
2532 dm_event_sent = 1;
2533 }
2534
2535 /* Return through std_return after this point. */
2536
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002537 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
2539 mp = dp->i_mount;
2540 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10002541 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2542 prid = dp->i_d.di_projid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 else
2544 prid = (xfs_prid_t)dfltprid;
2545
2546 /*
2547 * Make sure that we have allocated dquot(s) on disk.
2548 */
2549 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002550 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2552 if (error)
2553 goto std_return;
2554
2555 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2556 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Barry Naujok556b8b12008-04-10 12:22:07 +10002557 resblks = XFS_MKDIR_SPACE_RES(mp, dir_name->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2559 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2560 if (error == ENOSPC) {
2561 resblks = 0;
2562 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2563 XFS_TRANS_PERM_LOG_RES,
2564 XFS_MKDIR_LOG_COUNT);
2565 }
2566 if (error) {
2567 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 goto error_return;
2569 }
2570
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002571 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002572 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573
2574 /*
2575 * Check for directory link count overflow.
2576 */
2577 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2578 error = XFS_ERROR(EMLINK);
2579 goto error_return;
2580 }
2581
2582 /*
2583 * Reserve disk quota and the inode.
2584 */
2585 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2586 if (error)
2587 goto error_return;
2588
Barry Naujok556b8b12008-04-10 12:22:07 +10002589 error = xfs_dir_canenter(tp, dp, dir_name, resblks);
2590 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 goto error_return;
2592 /*
2593 * create the directory inode.
2594 */
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002595 error = xfs_dir_ialloc(&tp, dp, mode, 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 0, credp, prid, resblks > 0,
2597 &cdp, NULL);
2598 if (error) {
2599 if (error == ENOSPC)
2600 goto error_return;
2601 goto abort_return;
2602 }
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002603 xfs_itrace_ref(cdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604
2605 /*
2606 * Now we add the directory inode to the transaction.
2607 * We waited until now since xfs_dir_ialloc might start
2608 * a new transaction. Had we joined the transaction
Christoph Hellwig993386c2007-08-28 16:12:30 +10002609 * earlier, the locks might have gotten released. An error
2610 * from here on will result in the transaction cancel
2611 * unlocking dp so don't do it explicitly in the error path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 */
Christoph Hellwig979ebab2008-03-06 13:46:05 +11002613 IHOLD(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002615 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
2617 XFS_BMAP_INIT(&free_list, &first_block);
2618
Barry Naujok556b8b12008-04-10 12:22:07 +10002619 error = xfs_dir_createname(tp, dp, dir_name, cdp->i_ino,
2620 &first_block, &free_list, resblks ?
2621 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 if (error) {
2623 ASSERT(error != ENOSPC);
2624 goto error1;
2625 }
2626 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2627
2628 /*
2629 * Bump the in memory version number of the parent directory
2630 * so that other processes accessing it will recognize that
2631 * the directory has changed.
2632 */
2633 dp->i_gen++;
2634
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002635 error = xfs_dir_init(tp, cdp, dp);
2636 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
2639 cdp->i_gen = 1;
2640 error = xfs_bumplink(tp, dp);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002641 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 created = B_TRUE;
2645
Christoph Hellwig979ebab2008-03-06 13:46:05 +11002646 *ipp = cdp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 IHOLD(cdp);
2648
2649 /*
2650 * Attach the dquots to the new inode and modify the icount incore.
2651 */
2652 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2653
2654 /*
2655 * If this is a synchronous mount, make sure that the
2656 * mkdir transaction goes to disk before returning to
2657 * the user.
2658 */
2659 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2660 xfs_trans_set_sync(tp);
2661 }
2662
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002663 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 if (error) {
2665 IRELE(cdp);
2666 goto error2;
2667 }
2668
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002669 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 XFS_QM_DQRELE(mp, udqp);
2671 XFS_QM_DQRELE(mp, gdqp);
2672 if (error) {
2673 IRELE(cdp);
2674 }
2675
2676 /* Fall through to std_return with error = 0 or errno from
2677 * xfs_trans_commit. */
2678
2679std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002680 if ((created || (error != 0 && dm_event_sent != 0)) &&
Christoph Hellwig993386c2007-08-28 16:12:30 +10002681 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002683 dp, DM_RIGHT_NULL,
2684 created ? cdp : NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002686 dir_name->name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002687 mode, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 }
2689 return error;
2690
2691 error2:
2692 error1:
2693 xfs_bmap_cancel(&free_list);
2694 abort_return:
2695 cancel_flags |= XFS_TRANS_ABORT;
2696 error_return:
2697 xfs_trans_cancel(tp, cancel_flags);
2698 XFS_QM_DQRELE(mp, udqp);
2699 XFS_QM_DQRELE(mp, gdqp);
2700
Christoph Hellwig993386c2007-08-28 16:12:30 +10002701 if (unlock_dp_on_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 xfs_iunlock(dp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703
2704 goto std_return;
2705}
2706
Christoph Hellwig993386c2007-08-28 16:12:30 +10002707int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708xfs_rmdir(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002709 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10002710 struct xfs_name *name,
2711 xfs_inode_t *cdp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002713 bhv_vnode_t *dir_vp = XFS_ITOV(dp);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002714 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 int error;
2717 xfs_bmap_free_t free_list;
2718 xfs_fsblock_t first_block;
2719 int cancel_flags;
2720 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 int last_cdp_link;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 uint resblks;
2723
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002724 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
Christoph Hellwig993386c2007-08-28 16:12:30 +10002726 if (XFS_FORCED_SHUTDOWN(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002729 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002731 dp, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002732 NULL, DM_RIGHT_NULL, name->name,
2733 NULL, cdp->i_d.di_mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 if (error)
2735 return XFS_ERROR(error);
2736 }
2737
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 /*
2739 * We need to get a reference to cdp before we get our log
2740 * reservation. The reason for this is that we cannot call
2741 * xfs_iget for an inode for which we do not have a reference
2742 * once we've acquired a log reservation. This is because the
2743 * inode we are trying to get might be in xfs_inactive going
2744 * for a log reservation. Since we'll have to wait for the
2745 * inactive code to complete before returning from xfs_iget,
2746 * we need to make sure that we don't have log space reserved
Nathan Scottc41564b2006-03-29 08:55:14 +10002747 * when we call xfs_iget. Instead we get an unlocked reference
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 * to the inode before getting our log reservation.
2749 */
Christoph Hellwig43973962008-03-06 13:44:50 +11002750 IHOLD(cdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751
2752 /*
2753 * Get the dquots for the inodes.
2754 */
2755 error = XFS_QM_DQATTACH(mp, dp, 0);
2756 if (!error && dp != cdp)
2757 error = XFS_QM_DQATTACH(mp, cdp, 0);
2758 if (error) {
2759 IRELE(cdp);
2760 REMOVE_DEBUG_TRACE(__LINE__);
2761 goto std_return;
2762 }
2763
2764 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2765 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2766 /*
2767 * We try to get the real space reservation first,
2768 * allowing for directory btree deletion(s) implying
2769 * possible bmap insert(s). If we can't get the space
2770 * reservation then we use 0 instead, and avoid the bmap
2771 * btree insert(s) in the directory code by, if the bmap
2772 * insert tries to happen, instead trimming the LAST
2773 * block from the directory.
2774 */
2775 resblks = XFS_REMOVE_SPACE_RES(mp);
2776 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2777 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
2778 if (error == ENOSPC) {
2779 resblks = 0;
2780 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2781 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
2782 }
2783 if (error) {
2784 ASSERT(error != ENOSPC);
2785 cancel_flags = 0;
2786 IRELE(cdp);
2787 goto error_return;
2788 }
2789 XFS_BMAP_INIT(&free_list, &first_block);
2790
2791 /*
2792 * Now lock the child directory inode and the parent directory
2793 * inode in the proper order. This will take care of validating
2794 * that the directory entry for the child directory inode has
2795 * not changed while we were obtaining a log reservation.
2796 */
Eric Sandeene9ed9d22007-05-08 13:48:56 +10002797 error = xfs_lock_dir_and_entry(dp, cdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 if (error) {
2799 xfs_trans_cancel(tp, cancel_flags);
2800 IRELE(cdp);
2801 goto std_return;
2802 }
2803
2804 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2805 if (dp != cdp) {
2806 /*
2807 * Only increment the parent directory vnode count if
2808 * we didn't bump it in looking up cdp. The only time
2809 * we don't bump it is when we're looking up ".".
2810 */
2811 VN_HOLD(dir_vp);
2812 }
2813
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002814 xfs_itrace_ref(cdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
2816
2817 ASSERT(cdp->i_d.di_nlink >= 2);
2818 if (cdp->i_d.di_nlink != 2) {
2819 error = XFS_ERROR(ENOTEMPTY);
2820 goto error_return;
2821 }
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002822 if (!xfs_dir_isempty(cdp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 error = XFS_ERROR(ENOTEMPTY);
2824 goto error_return;
2825 }
2826
Barry Naujok556b8b12008-04-10 12:22:07 +10002827 error = xfs_dir_removename(tp, dp, name, cdp->i_ino,
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002828 &first_block, &free_list, resblks);
2829 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831
2832 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2833
2834 /*
2835 * Bump the in memory generation count on the parent
2836 * directory so that other can know that it has changed.
2837 */
2838 dp->i_gen++;
2839
2840 /*
2841 * Drop the link from cdp's "..".
2842 */
2843 error = xfs_droplink(tp, dp);
2844 if (error) {
2845 goto error1;
2846 }
2847
2848 /*
2849 * Drop the link from dp to cdp.
2850 */
2851 error = xfs_droplink(tp, cdp);
2852 if (error) {
2853 goto error1;
2854 }
2855
2856 /*
2857 * Drop the "." link from cdp to self.
2858 */
2859 error = xfs_droplink(tp, cdp);
2860 if (error) {
2861 goto error1;
2862 }
2863
2864 /* Determine these before committing transaction */
2865 last_cdp_link = (cdp)->i_d.di_nlink==0;
2866
2867 /*
2868 * Take an extra ref on the child vnode so that it
2869 * does not go to xfs_inactive() from within the commit.
2870 */
2871 IHOLD(cdp);
2872
2873 /*
2874 * If this is a synchronous mount, make sure that the
2875 * rmdir transaction goes to disk before returning to
2876 * the user.
2877 */
2878 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2879 xfs_trans_set_sync(tp);
2880 }
2881
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002882 error = xfs_bmap_finish (&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 if (error) {
2884 xfs_bmap_cancel(&free_list);
2885 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2886 XFS_TRANS_ABORT));
2887 IRELE(cdp);
2888 goto std_return;
2889 }
2890
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002891 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 if (error) {
2893 IRELE(cdp);
2894 goto std_return;
2895 }
2896
2897
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 IRELE(cdp);
2899
2900 /* Fall through to std_return with error = 0 or the errno
2901 * from xfs_trans_commit. */
Nathan Scottbb19fba2006-03-22 14:12:12 +11002902 std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002903 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002905 dp, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 NULL, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002907 name->name, NULL, cdp->i_d.di_mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 error, 0);
2909 }
2910 return error;
2911
Nathan Scottbb19fba2006-03-22 14:12:12 +11002912 error1:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 xfs_bmap_cancel(&free_list);
2914 cancel_flags |= XFS_TRANS_ABORT;
Jesper Juhl014c2542006-01-15 02:37:08 +01002915 /* FALLTHROUGH */
2916
Nathan Scottbb19fba2006-03-22 14:12:12 +11002917 error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 xfs_trans_cancel(tp, cancel_flags);
2919 goto std_return;
2920}
2921
Christoph Hellwig993386c2007-08-28 16:12:30 +10002922int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923xfs_symlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002924 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10002925 struct xfs_name *link_name,
2926 const char *target_path,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002927 mode_t mode,
Christoph Hellwig3937be52008-03-06 13:46:19 +11002928 xfs_inode_t **ipp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 cred_t *credp)
2930{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002931 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 xfs_inode_t *ip;
2934 int error;
2935 int pathlen;
2936 xfs_bmap_free_t free_list;
2937 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10002938 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 uint cancel_flags;
2940 int committed;
2941 xfs_fileoff_t first_fsb;
2942 xfs_filblks_t fs_blocks;
2943 int nmaps;
2944 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
2945 xfs_daddr_t d;
Barry Naujok556b8b12008-04-10 12:22:07 +10002946 const char *cur_chunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 int byte_cnt;
2948 int n;
2949 xfs_buf_t *bp;
2950 xfs_prid_t prid;
2951 struct xfs_dquot *udqp, *gdqp;
2952 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
Christoph Hellwig3937be52008-03-06 13:46:19 +11002954 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 error = 0;
2956 ip = NULL;
2957 tp = NULL;
2958
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002959 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960
2961 if (XFS_FORCED_SHUTDOWN(mp))
2962 return XFS_ERROR(EIO);
2963
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 /*
2965 * Check component lengths of the target path name.
2966 */
2967 pathlen = strlen(target_path);
2968 if (pathlen >= MAXPATHLEN) /* total string too long */
2969 return XFS_ERROR(ENAMETOOLONG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002971 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002972 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
Barry Naujok556b8b12008-04-10 12:22:07 +10002974 link_name->name, target_path, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 if (error)
2976 return error;
2977 }
2978
2979 /* Return through std_return after this point. */
2980
2981 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10002982 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2983 prid = dp->i_d.di_projid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 else
2985 prid = (xfs_prid_t)dfltprid;
2986
2987 /*
2988 * Make sure that we have allocated dquot(s) on disk.
2989 */
2990 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002991 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2993 if (error)
2994 goto std_return;
2995
2996 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
2997 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2998 /*
2999 * The symlink will fit into the inode data fork?
3000 * There can't be any attributes so we get the whole variable part.
3001 */
3002 if (pathlen <= XFS_LITINO(mp))
3003 fs_blocks = 0;
3004 else
3005 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
Barry Naujok556b8b12008-04-10 12:22:07 +10003006 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3008 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3009 if (error == ENOSPC && fs_blocks == 0) {
3010 resblks = 0;
3011 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3012 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3013 }
3014 if (error) {
3015 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016 goto error_return;
3017 }
3018
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10003019 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10003020 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021
3022 /*
3023 * Check whether the directory allows new symlinks or not.
3024 */
3025 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3026 error = XFS_ERROR(EPERM);
3027 goto error_return;
3028 }
3029
3030 /*
3031 * Reserve disk quota : blocks and inode.
3032 */
3033 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3034 if (error)
3035 goto error_return;
3036
3037 /*
3038 * Check for ability to enter directory entry, if no space reserved.
3039 */
Barry Naujok556b8b12008-04-10 12:22:07 +10003040 error = xfs_dir_canenter(tp, dp, link_name, resblks);
3041 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 goto error_return;
3043 /*
3044 * Initialize the bmap freelist prior to calling either
3045 * bmapi or the directory create code.
3046 */
3047 XFS_BMAP_INIT(&free_list, &first_block);
3048
3049 /*
3050 * Allocate an inode for the symlink.
3051 */
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10003052 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053 1, 0, credp, prid, resblks > 0, &ip, NULL);
3054 if (error) {
3055 if (error == ENOSPC)
3056 goto error_return;
3057 goto error1;
3058 }
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003059 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060
Christoph Hellwig993386c2007-08-28 16:12:30 +10003061 /*
3062 * An error after we've joined dp to the transaction will result in the
3063 * transaction cancel unlocking dp so don't do it explicitly in the
3064 * error path.
3065 */
Christoph Hellwig3937be52008-03-06 13:46:19 +11003066 IHOLD(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10003068 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069
3070 /*
3071 * Also attach the dquot(s) to it, if applicable.
3072 */
3073 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3074
3075 if (resblks)
3076 resblks -= XFS_IALLOC_SPACE_RES(mp);
3077 /*
3078 * If the symlink will fit into the inode, write it inline.
3079 */
3080 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3081 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3082 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3083 ip->i_d.di_size = pathlen;
3084
3085 /*
3086 * The inode was initially created in extent format.
3087 */
3088 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3089 ip->i_df.if_flags |= XFS_IFINLINE;
3090
3091 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3092 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3093
3094 } else {
3095 first_fsb = 0;
3096 nmaps = SYMLINK_MAPS;
3097
3098 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3099 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3100 &first_block, resblks, mval, &nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003101 &free_list, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 if (error) {
3103 goto error1;
3104 }
3105
3106 if (resblks)
3107 resblks -= fs_blocks;
3108 ip->i_d.di_size = pathlen;
3109 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3110
3111 cur_chunk = target_path;
3112 for (n = 0; n < nmaps; n++) {
3113 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3114 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3115 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3116 BTOBB(byte_cnt), 0);
3117 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3118 if (pathlen < byte_cnt) {
3119 byte_cnt = pathlen;
3120 }
3121 pathlen -= byte_cnt;
3122
3123 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3124 cur_chunk += byte_cnt;
3125
3126 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3127 }
3128 }
3129
3130 /*
3131 * Create the directory entry for the symlink.
3132 */
Barry Naujok556b8b12008-04-10 12:22:07 +10003133 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
3134 &first_block, &free_list, resblks);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003135 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3138 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3139
3140 /*
3141 * Bump the in memory version number of the parent directory
3142 * so that other processes accessing it will recognize that
3143 * the directory has changed.
3144 */
3145 dp->i_gen++;
3146
3147 /*
3148 * If this is a synchronous mount, make sure that the
3149 * symlink transaction goes to disk before returning to
3150 * the user.
3151 */
3152 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3153 xfs_trans_set_sync(tp);
3154 }
3155
3156 /*
3157 * xfs_trans_commit normally decrements the vnode ref count
3158 * when it unlocks the inode. Since we want to return the
3159 * vnode to the caller, we bump the vnode ref count now.
3160 */
3161 IHOLD(ip);
3162
Eric Sandeenf7c99b62007-02-10 18:37:16 +11003163 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 if (error) {
3165 goto error2;
3166 }
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003167 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168 XFS_QM_DQRELE(mp, udqp);
3169 XFS_QM_DQRELE(mp, gdqp);
3170
3171 /* Fall through to std_return with error = 0 or errno from
3172 * xfs_trans_commit */
3173std_return:
Christoph Hellwig993386c2007-08-28 16:12:30 +10003174 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003176 dp, DM_RIGHT_NULL,
3177 error ? NULL : ip,
Barry Naujok556b8b12008-04-10 12:22:07 +10003178 DM_RIGHT_NULL, link_name->name,
3179 target_path, 0, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180 }
3181
Christoph Hellwig3937be52008-03-06 13:46:19 +11003182 if (!error)
3183 *ipp = ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 return error;
3185
3186 error2:
3187 IRELE(ip);
3188 error1:
3189 xfs_bmap_cancel(&free_list);
3190 cancel_flags |= XFS_TRANS_ABORT;
3191 error_return:
3192 xfs_trans_cancel(tp, cancel_flags);
3193 XFS_QM_DQRELE(mp, udqp);
3194 XFS_QM_DQRELE(mp, gdqp);
3195
Christoph Hellwig993386c2007-08-28 16:12:30 +10003196 if (unlock_dp_on_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 xfs_iunlock(dp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198
3199 goto std_return;
3200}
3201
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203xfs_inode_flush(
Christoph Hellwig993386c2007-08-28 16:12:30 +10003204 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205 int flags)
3206{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003207 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 int error = 0;
3209
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 if (XFS_FORCED_SHUTDOWN(mp))
3211 return XFS_ERROR(EIO);
3212
3213 /*
3214 * Bypass inodes which have already been cleaned by
3215 * the inode flush clustering code inside xfs_iflush
3216 */
David Chinner33540402008-03-06 13:43:59 +11003217 if (xfs_inode_clean(ip))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218 return 0;
3219
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 /*
3221 * We make this non-blocking if the inode is contended,
3222 * return EAGAIN to indicate to the caller that they
3223 * did not succeed. This prevents the flush path from
3224 * blocking on inodes inside another operation right
3225 * now, they get caught later by xfs_sync.
3226 */
David Chinnera3f74ff2008-03-06 13:43:42 +11003227 if (flags & FLUSH_SYNC) {
3228 xfs_ilock(ip, XFS_ILOCK_SHARED);
3229 xfs_iflock(ip);
3230 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3231 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3232 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 return EAGAIN;
3234 }
David Chinnera3f74ff2008-03-06 13:43:42 +11003235 } else {
3236 return EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 }
3238
David Chinnera3f74ff2008-03-06 13:43:42 +11003239 error = xfs_iflush(ip, (flags & FLUSH_SYNC) ? XFS_IFLUSH_SYNC
3240 : XFS_IFLUSH_ASYNC_NOBLOCK);
3241 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3242
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243 return error;
3244}
3245
Christoph Hellwig993386c2007-08-28 16:12:30 +10003246
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247int
Christoph Hellwig993386c2007-08-28 16:12:30 +10003248xfs_set_dmattrs(
3249 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250 u_int evmask,
Christoph Hellwig993386c2007-08-28 16:12:30 +10003251 u_int16_t state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003253 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 int error;
3256
3257 if (!capable(CAP_SYS_ADMIN))
3258 return XFS_ERROR(EPERM);
3259
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 if (XFS_FORCED_SHUTDOWN(mp))
3261 return XFS_ERROR(EIO);
3262
3263 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3264 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3265 if (error) {
3266 xfs_trans_cancel(tp, 0);
3267 return error;
3268 }
3269 xfs_ilock(ip, XFS_ILOCK_EXCL);
3270 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3271
Christoph Hellwig613d7042007-10-11 17:44:08 +10003272 ip->i_d.di_dmevmask = evmask;
3273 ip->i_d.di_dmstate = state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274
3275 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3276 IHOLD(ip);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003277 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278
3279 return error;
3280}
3281
Christoph Hellwig993386c2007-08-28 16:12:30 +10003282int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283xfs_reclaim(
Christoph Hellwig993386c2007-08-28 16:12:30 +10003284 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003286 bhv_vnode_t *vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003288 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289
3290 ASSERT(!VN_MAPPED(vp));
3291
3292 /* bad inode, get out here ASAP */
3293 if (VN_BAD(vp)) {
3294 xfs_ireclaim(ip);
3295 return 0;
3296 }
3297
Christoph Hellwigb677c212007-08-29 11:46:28 +10003298 vn_iowait(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299
Christoph Hellwig51c91ed2005-09-02 16:58:38 +10003300 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11003302 /*
3303 * Make sure the atime in the XFS inode is correct before freeing the
3304 * Linux inode.
3305 */
3306 xfs_synchronize_atime(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307
David Chinner4c606582006-11-11 18:05:00 +11003308 /*
3309 * If we have nothing to flush with this inode then complete the
3310 * teardown now, otherwise break the link between the xfs inode and the
3311 * linux inode and clean up the xfs inode later. This avoids flushing
3312 * the inode to disk during the delete operation itself.
3313 *
3314 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3315 * first to ensure that xfs_iunpin() will never see an xfs inode
3316 * that has a linux inode being reclaimed. Synchronisation is provided
3317 * by the i_flags_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 */
3319 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3320 xfs_ilock(ip, XFS_ILOCK_EXCL);
3321 xfs_iflock(ip);
3322 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3323 } else {
3324 xfs_mount_t *mp = ip->i_mount;
3325
David Chinner4c606582006-11-11 18:05:00 +11003326 /* Protect sync and unpin from us */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 XFS_MOUNT_ILOCK(mp);
David Chinner4c606582006-11-11 18:05:00 +11003328 spin_lock(&ip->i_flags_lock);
3329 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
Christoph Hellwig739bfb22007-08-29 10:58:01 +10003330 vn_to_inode(vp)->i_private = NULL;
3331 ip->i_vnode = NULL;
David Chinner4c606582006-11-11 18:05:00 +11003332 spin_unlock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 XFS_MOUNT_IUNLOCK(mp);
3335 }
3336 return 0;
3337}
3338
3339int
3340xfs_finish_reclaim(
3341 xfs_inode_t *ip,
3342 int locked,
3343 int sync_mode)
3344{
David Chinnerda353b02007-08-28 14:00:13 +10003345 xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino);
Nathan Scott67fcaa72006-06-09 17:00:52 +10003346 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347 int error;
3348
3349 if (vp && VN_BAD(vp))
3350 goto reclaim;
3351
3352 /* The hash lock here protects a thread in xfs_iget_core from
3353 * racing with us on linking the inode back with a vnode.
3354 * Once we have the XFS_IRECLAIM flag set it will not touch
3355 * us.
3356 */
David Chinnerda353b02007-08-28 14:00:13 +10003357 write_lock(&pag->pag_ici_lock);
David Chinnerf273ab82006-09-28 11:06:03 +10003358 spin_lock(&ip->i_flags_lock);
David Chinner7a18c382006-11-11 18:04:54 +11003359 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
3360 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
David Chinnerf273ab82006-09-28 11:06:03 +10003361 spin_unlock(&ip->i_flags_lock);
David Chinnerda353b02007-08-28 14:00:13 +10003362 write_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363 if (locked) {
3364 xfs_ifunlock(ip);
3365 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3366 }
Jesper Juhl014c2542006-01-15 02:37:08 +01003367 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 }
David Chinner7a18c382006-11-11 18:04:54 +11003369 __xfs_iflags_set(ip, XFS_IRECLAIM);
David Chinnerf273ab82006-09-28 11:06:03 +10003370 spin_unlock(&ip->i_flags_lock);
David Chinnerda353b02007-08-28 14:00:13 +10003371 write_unlock(&pag->pag_ici_lock);
3372 xfs_put_perag(ip->i_mount, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373
3374 /*
3375 * If the inode is still dirty, then flush it out. If the inode
3376 * is not in the AIL, then it will be OK to flush it delwri as
3377 * long as xfs_iflush() does not keep any references to the inode.
3378 * We leave that decision up to xfs_iflush() since it has the
3379 * knowledge of whether it's OK to simply do a delwri flush of
3380 * the inode or whether we need to wait until the inode is
3381 * pulled from the AIL.
3382 * We get the flush lock regardless, though, just to make sure
3383 * we don't free it while it is being flushed.
3384 */
Lachlan McIlroy461aa8a2008-03-06 13:43:11 +11003385 if (!locked) {
3386 xfs_ilock(ip, XFS_ILOCK_EXCL);
3387 xfs_iflock(ip);
3388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389
Lachlan McIlroy461aa8a2008-03-06 13:43:11 +11003390 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 if (ip->i_update_core ||
3392 ((ip->i_itemp != NULL) &&
3393 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3394 error = xfs_iflush(ip, sync_mode);
3395 /*
3396 * If we hit an error, typically because of filesystem
3397 * shutdown, we don't need to let vn_reclaim to know
3398 * because we're gonna reclaim the inode anyway.
3399 */
3400 if (error) {
3401 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3402 goto reclaim;
3403 }
3404 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3405 }
3406
3407 ASSERT(ip->i_update_core == 0);
3408 ASSERT(ip->i_itemp == NULL ||
3409 ip->i_itemp->ili_format.ilf_fields == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410 }
3411
Lachlan McIlroy461aa8a2008-03-06 13:43:11 +11003412 xfs_ifunlock(ip);
3413 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3414
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 reclaim:
3416 xfs_ireclaim(ip);
3417 return 0;
3418}
3419
3420int
3421xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3422{
3423 int purged;
3424 xfs_inode_t *ip, *n;
3425 int done = 0;
3426
3427 while (!done) {
3428 purged = 0;
3429 XFS_MOUNT_ILOCK(mp);
3430 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3431 if (noblock) {
3432 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3433 continue;
3434 if (xfs_ipincount(ip) ||
3435 !xfs_iflock_nowait(ip)) {
3436 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3437 continue;
3438 }
3439 }
3440 XFS_MOUNT_IUNLOCK(mp);
Felix Blyakher6b2cf612005-11-25 16:42:13 +11003441 if (xfs_finish_reclaim(ip, noblock,
3442 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3443 delay(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444 purged = 1;
3445 break;
3446 }
3447
3448 done = !purged;
3449 }
3450
3451 XFS_MOUNT_IUNLOCK(mp);
3452 return 0;
3453}
3454
3455/*
3456 * xfs_alloc_file_space()
3457 * This routine allocates disk space for the given file.
3458 *
3459 * If alloc_type == 0, this request is for an ALLOCSP type
3460 * request which will change the file size. In this case, no
3461 * DMAPI event will be generated by the call. A TRUNCATE event
3462 * will be generated later by xfs_setattr.
3463 *
3464 * If alloc_type != 0, this request is for a RESVSP type
3465 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3466 * lower block boundary byte address is less than the file's
3467 * length.
3468 *
3469 * RETURNS:
3470 * 0 on success
3471 * errno on error
3472 *
3473 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003474STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475xfs_alloc_file_space(
3476 xfs_inode_t *ip,
3477 xfs_off_t offset,
3478 xfs_off_t len,
3479 int alloc_type,
3480 int attr_flags)
3481{
Nathan Scottdd9f4382006-01-11 15:28:28 +11003482 xfs_mount_t *mp = ip->i_mount;
3483 xfs_off_t count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 xfs_filblks_t allocated_fsb;
3485 xfs_filblks_t allocatesize_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003486 xfs_extlen_t extsz, temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 xfs_fileoff_t startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003488 xfs_fsblock_t firstfsb;
3489 int nimaps;
3490 int bmapi_flag;
3491 int quota_flag;
3492 int rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 xfs_trans_t *tp;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003494 xfs_bmbt_irec_t imaps[1], *imapp;
3495 xfs_bmap_free_t free_list;
3496 uint qblocks, resblks, resrtextents;
3497 int committed;
3498 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003500 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501
3502 if (XFS_FORCED_SHUTDOWN(mp))
3503 return XFS_ERROR(EIO);
3504
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3506 return error;
3507
3508 if (len <= 0)
3509 return XFS_ERROR(EINVAL);
3510
David Chinner957d0eb2007-06-18 16:50:37 +10003511 rt = XFS_IS_REALTIME_INODE(ip);
3512 extsz = xfs_get_extsz_hint(ip);
3513
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514 count = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 imapp = &imaps[0];
Nathan Scottdd9f4382006-01-11 15:28:28 +11003516 nimaps = 1;
3517 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
3519 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
3520
3521 /* Generate a DMAPI event if needed. */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003522 if (alloc_type != 0 && offset < ip->i_size &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 (attr_flags&ATTR_DMI) == 0 &&
Christoph Hellwigeb9df392007-08-16 18:42:07 +10003524 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525 xfs_off_t end_dmi_offset;
3526
3527 end_dmi_offset = offset+len;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003528 if (end_dmi_offset > ip->i_size)
3529 end_dmi_offset = ip->i_size;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003530 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset,
3531 end_dmi_offset - offset, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01003533 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 }
3535
3536 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003537 * Allocate file space until done or until there is an error
Linus Torvalds1da177e2005-04-16 15:20:36 -07003538 */
3539retry:
3540 while (allocatesize_fsb && !error) {
Nathan Scottdd9f4382006-01-11 15:28:28 +11003541 xfs_fileoff_t s, e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542
Nathan Scottdd9f4382006-01-11 15:28:28 +11003543 /*
Nathan Scott3ddb8fa2006-01-11 15:33:02 +11003544 * Determine space reservations for data/realtime.
Nathan Scottdd9f4382006-01-11 15:28:28 +11003545 */
3546 if (unlikely(extsz)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 s = startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003548 do_div(s, extsz);
3549 s *= extsz;
3550 e = startoffset_fsb + allocatesize_fsb;
3551 if ((temp = do_mod(startoffset_fsb, extsz)))
3552 e += temp;
3553 if ((temp = do_mod(e, extsz)))
3554 e += extsz - temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 } else {
Nathan Scottdd9f4382006-01-11 15:28:28 +11003556 s = 0;
3557 e = allocatesize_fsb;
3558 }
3559
3560 if (unlikely(rt)) {
3561 resrtextents = qblocks = (uint)(e - s);
3562 resrtextents /= mp->m_sb.sb_rextsize;
3563 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
3564 quota_flag = XFS_QMOPT_RES_RTBLKS;
3565 } else {
3566 resrtextents = 0;
3567 resblks = qblocks = \
3568 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
3569 quota_flag = XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 }
3571
3572 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003573 * Allocate and setup the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574 */
3575 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Nathan Scottdd9f4382006-01-11 15:28:28 +11003576 error = xfs_trans_reserve(tp, resblks,
3577 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578 XFS_TRANS_PERM_LOG_RES,
3579 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003581 * Check for running out of space
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582 */
3583 if (error) {
3584 /*
3585 * Free the transaction structure.
3586 */
3587 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
3588 xfs_trans_cancel(tp, 0);
3589 break;
3590 }
3591 xfs_ilock(ip, XFS_ILOCK_EXCL);
Nathan Scottdd9f4382006-01-11 15:28:28 +11003592 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
3593 qblocks, 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 if (error)
3595 goto error1;
3596
3597 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3598 xfs_trans_ihold(tp, ip);
3599
3600 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003601 * Issue the xfs_bmapi() call to allocate the blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602 */
3603 XFS_BMAP_INIT(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003604 error = xfs_bmapi(tp, ip, startoffset_fsb,
Nathan Scottdd9f4382006-01-11 15:28:28 +11003605 allocatesize_fsb, bmapi_flag,
3606 &firstfsb, 0, imapp, &nimaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003607 &free_list, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 if (error) {
3609 goto error0;
3610 }
3611
3612 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003613 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11003615 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616 if (error) {
3617 goto error0;
3618 }
3619
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003620 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3622 if (error) {
3623 break;
3624 }
3625
3626 allocated_fsb = imapp->br_blockcount;
3627
Nathan Scottdd9f4382006-01-11 15:28:28 +11003628 if (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 error = XFS_ERROR(ENOSPC);
3630 break;
3631 }
3632
3633 startoffset_fsb += allocated_fsb;
3634 allocatesize_fsb -= allocated_fsb;
3635 }
3636dmapi_enospc_check:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10003637 if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 &&
3638 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003640 ip, DM_RIGHT_NULL,
3641 ip, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003642 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
3643 if (error == 0)
3644 goto retry; /* Maybe DMAPI app. has made space */
3645 /* else fall through with error from XFS_SEND_DATA */
3646 }
3647
3648 return error;
3649
Nathan Scottdd9f4382006-01-11 15:28:28 +11003650error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651 xfs_bmap_cancel(&free_list);
Nathan Scottdd9f4382006-01-11 15:28:28 +11003652 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
3653
3654error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3656 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3657 goto dmapi_enospc_check;
3658}
3659
3660/*
3661 * Zero file bytes between startoff and endoff inclusive.
3662 * The iolock is held exclusive and no blocks are buffered.
3663 */
3664STATIC int
3665xfs_zero_remaining_bytes(
3666 xfs_inode_t *ip,
3667 xfs_off_t startoff,
3668 xfs_off_t endoff)
3669{
3670 xfs_bmbt_irec_t imap;
3671 xfs_fileoff_t offset_fsb;
3672 xfs_off_t lastoffset;
3673 xfs_off_t offset;
3674 xfs_buf_t *bp;
3675 xfs_mount_t *mp = ip->i_mount;
3676 int nimap;
3677 int error = 0;
3678
3679 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
Eric Sandeen71ddabb2007-11-23 16:29:42 +11003680 XFS_IS_REALTIME_INODE(ip) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681 mp->m_rtdev_targp : mp->m_ddev_targp);
3682
3683 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
3684 offset_fsb = XFS_B_TO_FSBT(mp, offset);
3685 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003686 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003687 NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 if (error || nimap < 1)
3689 break;
3690 ASSERT(imap.br_blockcount >= 1);
3691 ASSERT(imap.br_startoff == offset_fsb);
3692 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
3693 if (lastoffset > endoff)
3694 lastoffset = endoff;
3695 if (imap.br_startblock == HOLESTARTBLOCK)
3696 continue;
3697 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3698 if (imap.br_state == XFS_EXT_UNWRITTEN)
3699 continue;
3700 XFS_BUF_UNDONE(bp);
3701 XFS_BUF_UNWRITE(bp);
3702 XFS_BUF_READ(bp);
3703 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
3704 xfsbdstrat(mp, bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10003705 error = xfs_iowait(bp);
3706 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
3708 mp, bp, XFS_BUF_ADDR(bp));
3709 break;
3710 }
3711 memset(XFS_BUF_PTR(bp) +
3712 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
3713 0, lastoffset - offset + 1);
3714 XFS_BUF_UNDONE(bp);
3715 XFS_BUF_UNREAD(bp);
3716 XFS_BUF_WRITE(bp);
3717 xfsbdstrat(mp, bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10003718 error = xfs_iowait(bp);
3719 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
3721 mp, bp, XFS_BUF_ADDR(bp));
3722 break;
3723 }
3724 }
3725 xfs_buf_free(bp);
3726 return error;
3727}
3728
3729/*
3730 * xfs_free_file_space()
3731 * This routine frees disk space for the given file.
3732 *
3733 * This routine is only called by xfs_change_file_space
3734 * for an UNRESVSP type call.
3735 *
3736 * RETURNS:
3737 * 0 on success
3738 * errno on error
3739 *
3740 */
3741STATIC int
3742xfs_free_file_space(
3743 xfs_inode_t *ip,
3744 xfs_off_t offset,
3745 xfs_off_t len,
3746 int attr_flags)
3747{
Nathan Scott67fcaa72006-06-09 17:00:52 +10003748 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003749 int committed;
3750 int done;
3751 xfs_off_t end_dmi_offset;
3752 xfs_fileoff_t endoffset_fsb;
3753 int error;
3754 xfs_fsblock_t firstfsb;
3755 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 xfs_bmbt_irec_t imap;
3757 xfs_off_t ioffset;
3758 xfs_extlen_t mod=0;
3759 xfs_mount_t *mp;
3760 int nimap;
3761 uint resblks;
Nathan Scott673cdf52006-09-28 10:56:26 +10003762 uint rounding;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763 int rt;
3764 xfs_fileoff_t startoffset_fsb;
3765 xfs_trans_t *tp;
Dean Roehrich5fcbab32005-05-05 13:27:19 -07003766 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10003768 vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 mp = ip->i_mount;
3770
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003771 xfs_itrace_entry(ip);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10003772
Linus Torvalds1da177e2005-04-16 15:20:36 -07003773 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3774 return error;
3775
3776 error = 0;
3777 if (len <= 0) /* if nothing being freed */
3778 return error;
Eric Sandeen71ddabb2007-11-23 16:29:42 +11003779 rt = XFS_IS_REALTIME_INODE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
3781 end_dmi_offset = offset + len;
3782 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
3783
Christoph Hellwigeb9df392007-08-16 18:42:07 +10003784 if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 &&
3785 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003786 if (end_dmi_offset > ip->i_size)
3787 end_dmi_offset = ip->i_size;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003788 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789 offset, end_dmi_offset - offset,
3790 AT_DELAY_FLAG(attr_flags), NULL);
3791 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01003792 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793 }
3794
Dean Roehrich5fcbab32005-05-05 13:27:19 -07003795 if (attr_flags & ATTR_NOLOCK)
3796 need_iolock = 0;
Yingping Lu9fa80462006-03-22 12:44:35 +11003797 if (need_iolock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigb677c212007-08-29 11:46:28 +10003799 vn_iowait(ip); /* wait for the completion of any pending DIOs */
Yingping Lu9fa80462006-03-22 12:44:35 +11003800 }
Dean Roehrich5fcbab32005-05-05 13:27:19 -07003801
Tim Shimmine6a4b372007-11-23 16:30:42 +11003802 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 ioffset = offset & ~(rounding - 1);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10003804
3805 if (VN_CACHED(vp) != 0) {
Tim Shimmine6a4b372007-11-23 16:30:42 +11003806 xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1);
3807 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10003808 if (error)
3809 goto out_unlock_iolock;
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10003810 }
3811
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 /*
3813 * Need to zero the stuff we're not freeing, on disk.
3814 * If its a realtime file & can't use unwritten extents then we
3815 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3816 * will take care of it for us.
3817 */
Eric Sandeen62118702008-03-06 13:44:28 +11003818 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003820 error = xfs_bmapi(NULL, ip, startoffset_fsb,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003821 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 if (error)
3823 goto out_unlock_iolock;
3824 ASSERT(nimap == 0 || nimap == 1);
3825 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3826 xfs_daddr_t block;
3827
3828 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3829 block = imap.br_startblock;
3830 mod = do_div(block, mp->m_sb.sb_rextsize);
3831 if (mod)
3832 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
3833 }
3834 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003835 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003836 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837 if (error)
3838 goto out_unlock_iolock;
3839 ASSERT(nimap == 0 || nimap == 1);
3840 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3841 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3842 mod++;
3843 if (mod && (mod != mp->m_sb.sb_rextsize))
3844 endoffset_fsb -= mod;
3845 }
3846 }
3847 if ((done = (endoffset_fsb <= startoffset_fsb)))
3848 /*
3849 * One contiguous piece to clear
3850 */
3851 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
3852 else {
3853 /*
3854 * Some full blocks, possibly two pieces to clear
3855 */
3856 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
3857 error = xfs_zero_remaining_bytes(ip, offset,
3858 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
3859 if (!error &&
3860 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
3861 error = xfs_zero_remaining_bytes(ip,
3862 XFS_FSB_TO_B(mp, endoffset_fsb),
3863 offset + len - 1);
3864 }
3865
3866 /*
3867 * free file space until done or until there is an error
3868 */
3869 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
3870 while (!error && !done) {
3871
3872 /*
David Chinner91ebecc2007-07-19 16:28:30 +10003873 * allocate and setup the transaction. Allow this
3874 * transaction to dip into the reserve blocks to ensure
3875 * the freeing of the space succeeds at ENOSPC.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876 */
3877 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
David Chinner91ebecc2007-07-19 16:28:30 +10003878 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879 error = xfs_trans_reserve(tp,
3880 resblks,
3881 XFS_WRITE_LOG_RES(mp),
3882 0,
3883 XFS_TRANS_PERM_LOG_RES,
3884 XFS_WRITE_LOG_COUNT);
3885
3886 /*
3887 * check for running out of space
3888 */
3889 if (error) {
3890 /*
3891 * Free the transaction structure.
3892 */
3893 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
3894 xfs_trans_cancel(tp, 0);
3895 break;
3896 }
3897 xfs_ilock(ip, XFS_ILOCK_EXCL);
3898 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
Nathan Scottdd9f4382006-01-11 15:28:28 +11003899 ip->i_udquot, ip->i_gdquot, resblks, 0,
3900 XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 if (error)
3902 goto error1;
3903
3904 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3905 xfs_trans_ihold(tp, ip);
3906
3907 /*
3908 * issue the bunmapi() call to free the blocks
3909 */
3910 XFS_BMAP_INIT(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003911 error = xfs_bunmapi(tp, ip, startoffset_fsb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912 endoffset_fsb - startoffset_fsb,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003913 0, 2, &firstfsb, &free_list, NULL, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003914 if (error) {
3915 goto error0;
3916 }
3917
3918 /*
3919 * complete the transaction
3920 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11003921 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003922 if (error) {
3923 goto error0;
3924 }
3925
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003926 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3928 }
3929
3930 out_unlock_iolock:
3931 if (need_iolock)
3932 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3933 return error;
3934
3935 error0:
3936 xfs_bmap_cancel(&free_list);
3937 error1:
3938 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3939 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
3940 XFS_ILOCK_EXCL);
3941 return error;
3942}
3943
3944/*
3945 * xfs_change_file_space()
3946 * This routine allocates or frees disk space for the given file.
3947 * The user specified parameters are checked for alignment and size
3948 * limitations.
3949 *
3950 * RETURNS:
3951 * 0 on success
3952 * errno on error
3953 *
3954 */
3955int
3956xfs_change_file_space(
Christoph Hellwig993386c2007-08-28 16:12:30 +10003957 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958 int cmd,
3959 xfs_flock64_t *bf,
3960 xfs_off_t offset,
3961 cred_t *credp,
3962 int attr_flags)
3963{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003964 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 int clrprealloc;
3966 int error;
3967 xfs_fsize_t fsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 int setprealloc;
3969 xfs_off_t startoffset;
3970 xfs_off_t llen;
3971 xfs_trans_t *tp;
Nathan Scott8285fb52006-06-09 17:07:12 +10003972 bhv_vattr_t va;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003974 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975
Christoph Hellwig993386c2007-08-28 16:12:30 +10003976 if (!S_ISREG(ip->i_d.di_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977 return XFS_ERROR(EINVAL);
3978
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979 switch (bf->l_whence) {
3980 case 0: /*SEEK_SET*/
3981 break;
3982 case 1: /*SEEK_CUR*/
3983 bf->l_start += offset;
3984 break;
3985 case 2: /*SEEK_END*/
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003986 bf->l_start += ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 break;
3988 default:
3989 return XFS_ERROR(EINVAL);
3990 }
3991
3992 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
3993
3994 if ( (bf->l_start < 0)
3995 || (bf->l_start > XFS_MAXIOFFSET(mp))
3996 || (bf->l_start + llen < 0)
3997 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
3998 return XFS_ERROR(EINVAL);
3999
4000 bf->l_whence = 0;
4001
4002 startoffset = bf->l_start;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10004003 fsize = ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004
4005 /*
4006 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4007 * file space.
4008 * These calls do NOT zero the data space allocated to the file,
4009 * nor do they change the file size.
4010 *
4011 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4012 * space.
4013 * These calls cause the new file data to be zeroed and the file
4014 * size to be changed.
4015 */
4016 setprealloc = clrprealloc = 0;
4017
4018 switch (cmd) {
4019 case XFS_IOC_RESVSP:
4020 case XFS_IOC_RESVSP64:
4021 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4022 1, attr_flags);
4023 if (error)
4024 return error;
4025 setprealloc = 1;
4026 break;
4027
4028 case XFS_IOC_UNRESVSP:
4029 case XFS_IOC_UNRESVSP64:
4030 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4031 attr_flags)))
4032 return error;
4033 break;
4034
4035 case XFS_IOC_ALLOCSP:
4036 case XFS_IOC_ALLOCSP64:
4037 case XFS_IOC_FREESP:
4038 case XFS_IOC_FREESP64:
4039 if (startoffset > fsize) {
4040 error = xfs_alloc_file_space(ip, fsize,
4041 startoffset - fsize, 0, attr_flags);
4042 if (error)
4043 break;
4044 }
4045
4046 va.va_mask = XFS_AT_SIZE;
4047 va.va_size = startoffset;
4048
Christoph Hellwig993386c2007-08-28 16:12:30 +10004049 error = xfs_setattr(ip, &va, attr_flags, credp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050
4051 if (error)
4052 return error;
4053
4054 clrprealloc = 1;
4055 break;
4056
4057 default:
4058 ASSERT(0);
4059 return XFS_ERROR(EINVAL);
4060 }
4061
4062 /*
4063 * update the inode timestamp, mode, and prealloc flag bits
4064 */
4065 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4066
4067 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4068 0, 0, 0))) {
4069 /* ASSERT(0); */
4070 xfs_trans_cancel(tp, 0);
4071 return error;
4072 }
4073
4074 xfs_ilock(ip, XFS_ILOCK_EXCL);
4075
4076 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4077 xfs_trans_ihold(tp, ip);
4078
4079 if ((attr_flags & ATTR_DMI) == 0) {
4080 ip->i_d.di_mode &= ~S_ISUID;
4081
4082 /*
4083 * Note that we don't have to worry about mandatory
4084 * file locking being disabled here because we only
4085 * clear the S_ISGID bit if the Group execute bit is
4086 * on, but if it was on then mandatory locking wouldn't
4087 * have been enabled.
4088 */
4089 if (ip->i_d.di_mode & S_IXGRP)
4090 ip->i_d.di_mode &= ~S_ISGID;
4091
4092 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4093 }
4094 if (setprealloc)
4095 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4096 else if (clrprealloc)
4097 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4098
4099 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4100 xfs_trans_set_sync(tp);
4101
Eric Sandeen1c72bf92007-05-08 13:48:42 +10004102 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103
4104 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4105
4106 return error;
4107}