Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * 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. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * 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 | |
1da177e4 | 17 | */ |
16f7e0fe | 18 | |
1da177e4 | 19 | #include "xfs.h" |
a844f451 | 20 | #include "xfs_fs.h" |
1da177e4 | 21 | #include "xfs_types.h" |
a844f451 | 22 | #include "xfs_bit.h" |
1da177e4 | 23 | #include "xfs_log.h" |
a844f451 | 24 | #include "xfs_inum.h" |
1da177e4 LT |
25 | #include "xfs_trans.h" |
26 | #include "xfs_sb.h" | |
27 | #include "xfs_ag.h" | |
1da177e4 LT |
28 | #include "xfs_dir2.h" |
29 | #include "xfs_dmapi.h" | |
30 | #include "xfs_mount.h" | |
a844f451 | 31 | #include "xfs_da_btree.h" |
1da177e4 | 32 | #include "xfs_bmap_btree.h" |
a844f451 | 33 | #include "xfs_alloc_btree.h" |
1da177e4 | 34 | #include "xfs_ialloc_btree.h" |
1da177e4 | 35 | #include "xfs_dir2_sf.h" |
a844f451 | 36 | #include "xfs_attr_sf.h" |
1da177e4 | 37 | #include "xfs_dinode.h" |
1da177e4 | 38 | #include "xfs_inode.h" |
a844f451 | 39 | #include "xfs_inode_item.h" |
a844f451 NS |
40 | #include "xfs_itable.h" |
41 | #include "xfs_btree.h" | |
42 | #include "xfs_ialloc.h" | |
43 | #include "xfs_alloc.h" | |
1da177e4 | 44 | #include "xfs_bmap.h" |
ef14f0c1 | 45 | #include "xfs_acl.h" |
1da177e4 LT |
46 | #include "xfs_attr.h" |
47 | #include "xfs_rw.h" | |
1da177e4 | 48 | #include "xfs_error.h" |
1da177e4 LT |
49 | #include "xfs_quota.h" |
50 | #include "xfs_utils.h" | |
a844f451 | 51 | #include "xfs_rtalloc.h" |
1da177e4 | 52 | #include "xfs_trans_space.h" |
1da177e4 | 53 | #include "xfs_log_priv.h" |
2a82b8be | 54 | #include "xfs_filestream.h" |
993386c1 | 55 | #include "xfs_vnodeops.h" |
0b1b213f | 56 | #include "xfs_trace.h" |
1da177e4 | 57 | |
1da177e4 LT |
58 | int |
59 | xfs_setattr( | |
0f285c8a CH |
60 | struct xfs_inode *ip, |
61 | struct iattr *iattr, | |
ea5a3dc8 | 62 | int flags) |
1da177e4 | 63 | { |
993386c1 | 64 | xfs_mount_t *mp = ip->i_mount; |
e4f75291 | 65 | struct inode *inode = VFS_I(ip); |
0f285c8a | 66 | int mask = iattr->ia_valid; |
1da177e4 | 67 | xfs_trans_t *tp; |
1da177e4 LT |
68 | int code; |
69 | uint lock_flags; | |
70 | uint commit_flags=0; | |
71 | uid_t uid=0, iuid=0; | |
72 | gid_t gid=0, igid=0; | |
1da177e4 | 73 | struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2; |
5fcbab35 | 74 | int need_iolock = 1; |
1da177e4 | 75 | |
cf441eeb | 76 | xfs_itrace_entry(ip); |
1da177e4 | 77 | |
bd186aa9 | 78 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
79 | return XFS_ERROR(EROFS); |
80 | ||
1da177e4 LT |
81 | if (XFS_FORCED_SHUTDOWN(mp)) |
82 | return XFS_ERROR(EIO); | |
83 | ||
c4cd747e CH |
84 | code = -inode_change_ok(inode, iattr); |
85 | if (code) | |
86 | return code; | |
87 | ||
1da177e4 LT |
88 | olddquot1 = olddquot2 = NULL; |
89 | udqp = gdqp = NULL; | |
90 | ||
91 | /* | |
92 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
93 | * before we start any other transactions. Trying to do this later | |
94 | * is messy. We don't care to take a readlock to look at the ids | |
95 | * in inode here, because we can't hold it across the trans_reserve. | |
96 | * If the IDs do change before we take the ilock, we're covered | |
97 | * because the i_*dquot fields will get updated anyway. | |
98 | */ | |
0f285c8a | 99 | if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { |
1da177e4 LT |
100 | uint qflags = 0; |
101 | ||
0f285c8a CH |
102 | if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { |
103 | uid = iattr->ia_uid; | |
1da177e4 LT |
104 | qflags |= XFS_QMOPT_UQUOTA; |
105 | } else { | |
106 | uid = ip->i_d.di_uid; | |
107 | } | |
0f285c8a CH |
108 | if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { |
109 | gid = iattr->ia_gid; | |
1da177e4 LT |
110 | qflags |= XFS_QMOPT_GQUOTA; |
111 | } else { | |
112 | gid = ip->i_d.di_gid; | |
113 | } | |
25fe55e8 | 114 | |
1da177e4 LT |
115 | /* |
116 | * We take a reference when we initialize udqp and gdqp, | |
117 | * so it is important that we never blindly double trip on | |
118 | * the same variable. See xfs_create() for an example. | |
119 | */ | |
120 | ASSERT(udqp == NULL); | |
121 | ASSERT(gdqp == NULL); | |
7d095257 | 122 | code = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid, |
25fe55e8 | 123 | qflags, &udqp, &gdqp); |
1da177e4 | 124 | if (code) |
014c2544 | 125 | return code; |
1da177e4 LT |
126 | } |
127 | ||
128 | /* | |
129 | * For the other attributes, we acquire the inode lock and | |
130 | * first do an error checking pass. | |
131 | */ | |
132 | tp = NULL; | |
133 | lock_flags = XFS_ILOCK_EXCL; | |
0f285c8a | 134 | if (flags & XFS_ATTR_NOLOCK) |
5fcbab35 | 135 | need_iolock = 0; |
0f285c8a | 136 | if (!(mask & ATTR_SIZE)) { |
d6d59bad CH |
137 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); |
138 | commit_flags = 0; | |
139 | code = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), | |
140 | 0, 0, 0); | |
141 | if (code) { | |
142 | lock_flags = 0; | |
143 | goto error_return; | |
1da177e4 LT |
144 | } |
145 | } else { | |
eb9df39d | 146 | if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) && |
0f285c8a | 147 | !(flags & XFS_ATTR_DMI)) { |
1da177e4 | 148 | int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR; |
bc4ac74a | 149 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip, |
0f285c8a | 150 | iattr->ia_size, 0, dmflags, NULL); |
1da177e4 LT |
151 | if (code) { |
152 | lock_flags = 0; | |
153 | goto error_return; | |
154 | } | |
155 | } | |
156 | if (need_iolock) | |
157 | lock_flags |= XFS_IOLOCK_EXCL; | |
158 | } | |
159 | ||
160 | xfs_ilock(ip, lock_flags); | |
161 | ||
1da177e4 LT |
162 | /* |
163 | * Change file ownership. Must be the owner or privileged. | |
1da177e4 | 164 | */ |
0f285c8a | 165 | if (mask & (ATTR_UID|ATTR_GID)) { |
1da177e4 LT |
166 | /* |
167 | * These IDs could have changed since we last looked at them. | |
168 | * But, we're assured that if the ownership did change | |
169 | * while we didn't have the inode locked, inode's dquot(s) | |
170 | * would have changed also. | |
171 | */ | |
172 | iuid = ip->i_d.di_uid; | |
1da177e4 | 173 | igid = ip->i_d.di_gid; |
0f285c8a CH |
174 | gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; |
175 | uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; | |
1da177e4 | 176 | |
1da177e4 | 177 | /* |
25fe55e8 | 178 | * Do a quota reservation only if uid/gid is actually |
1da177e4 LT |
179 | * going to change. |
180 | */ | |
7d095257 CH |
181 | if (XFS_IS_QUOTA_RUNNING(mp) && |
182 | ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || | |
183 | (XFS_IS_GQUOTA_ON(mp) && igid != gid))) { | |
1da177e4 | 184 | ASSERT(tp); |
7d095257 | 185 | code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp, |
1da177e4 LT |
186 | capable(CAP_FOWNER) ? |
187 | XFS_QMOPT_FORCE_RES : 0); | |
188 | if (code) /* out of quota */ | |
189 | goto error_return; | |
190 | } | |
191 | } | |
192 | ||
193 | /* | |
194 | * Truncate file. Must have write permission and not be a directory. | |
195 | */ | |
0f285c8a | 196 | if (mask & ATTR_SIZE) { |
1da177e4 | 197 | /* Short circuit the truncate case for zero length files */ |
0f285c8a CH |
198 | if (iattr->ia_size == 0 && |
199 | ip->i_size == 0 && ip->i_d.di_nextents == 0) { | |
1da177e4 LT |
200 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
201 | lock_flags &= ~XFS_ILOCK_EXCL; | |
0f285c8a | 202 | if (mask & ATTR_CTIME) |
1da177e4 LT |
203 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
204 | code = 0; | |
205 | goto error_return; | |
206 | } | |
207 | ||
42173f68 | 208 | if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
209 | code = XFS_ERROR(EISDIR); |
210 | goto error_return; | |
42173f68 | 211 | } else if (!S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
212 | code = XFS_ERROR(EINVAL); |
213 | goto error_return; | |
214 | } | |
c4cd747e | 215 | |
1da177e4 LT |
216 | /* |
217 | * Make sure that the dquots are attached to the inode. | |
218 | */ | |
7d095257 | 219 | code = xfs_qm_dqattach_locked(ip, 0); |
c4cd747e | 220 | if (code) |
1da177e4 | 221 | goto error_return; |
1da177e4 | 222 | |
c4cd747e CH |
223 | /* |
224 | * Now we can make the changes. Before we join the inode | |
225 | * to the transaction, if ATTR_SIZE is set then take care of | |
226 | * the part of the truncation that must be done without the | |
227 | * inode lock. This needs to be done before joining the inode | |
228 | * to the transaction, because the inode cannot be unlocked | |
229 | * once it is a part of the transaction. | |
230 | */ | |
0f285c8a | 231 | if (iattr->ia_size > ip->i_size) { |
61436feb CH |
232 | /* |
233 | * Do the first part of growing a file: zero any data | |
234 | * in the last block that is beyond the old EOF. We | |
235 | * need to do this before the inode is joined to the | |
236 | * transaction to modify the i_size. | |
237 | */ | |
0f285c8a | 238 | code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size); |
374e2ac3 | 239 | } |
1da177e4 | 240 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
c32676ee DC |
241 | |
242 | /* | |
243 | * We are going to log the inode size change in this | |
244 | * transaction so any previous writes that are beyond the on | |
245 | * disk EOF and the new EOF that have not been written out need | |
246 | * to be written here. If we do not write the data out, we | |
247 | * expose ourselves to the null files problem. | |
248 | * | |
249 | * Only flush from the on disk size to the smaller of the in | |
250 | * memory file size or the new size as that's the range we | |
251 | * really care about here and prevents waiting for other data | |
252 | * not within the range we care about here. | |
253 | */ | |
254 | if (!code && | |
0f285c8a CH |
255 | ip->i_size != ip->i_d.di_size && |
256 | iattr->ia_size > ip->i_d.di_size) { | |
739bfb2a | 257 | code = xfs_flush_pages(ip, |
0f285c8a | 258 | ip->i_d.di_size, iattr->ia_size, |
0cadda1c | 259 | XBF_ASYNC, FI_NONE); |
c32676ee DC |
260 | } |
261 | ||
262 | /* wait for all I/O to complete */ | |
25e41b3d | 263 | xfs_ioend_wait(ip); |
c32676ee | 264 | |
1da177e4 | 265 | if (!code) |
0f285c8a | 266 | code = xfs_itruncate_data(ip, iattr->ia_size); |
1da177e4 LT |
267 | if (code) { |
268 | ASSERT(tp == NULL); | |
269 | lock_flags &= ~XFS_ILOCK_EXCL; | |
07f1a4f5 | 270 | ASSERT(lock_flags == XFS_IOLOCK_EXCL || !need_iolock); |
1da177e4 LT |
271 | goto error_return; |
272 | } | |
273 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); | |
274 | if ((code = xfs_trans_reserve(tp, 0, | |
275 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
276 | XFS_TRANS_PERM_LOG_RES, | |
277 | XFS_ITRUNCATE_LOG_COUNT))) { | |
278 | xfs_trans_cancel(tp, 0); | |
279 | if (need_iolock) | |
280 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
281 | return code; | |
282 | } | |
283 | commit_flags = XFS_TRANS_RELEASE_LOG_RES; | |
284 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 285 | |
1da177e4 LT |
286 | xfs_trans_ijoin(tp, ip, lock_flags); |
287 | xfs_trans_ihold(tp, ip); | |
1da177e4 | 288 | |
44d814ce DC |
289 | /* |
290 | * Only change the c/mtime if we are changing the size | |
291 | * or we are explicitly asked to change it. This handles | |
292 | * the semantic difference between truncate() and ftruncate() | |
293 | * as implemented in the VFS. | |
d6d59bad CH |
294 | * |
295 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME | |
296 | * is a special case where we need to update the times despite | |
297 | * not having these flags set. For all other operations the | |
298 | * VFS set these flags explicitly if it wants a timestamp | |
299 | * update. | |
44d814ce | 300 | */ |
d6d59bad CH |
301 | if (iattr->ia_size != ip->i_size && |
302 | (!(mask & (ATTR_CTIME | ATTR_MTIME)))) { | |
303 | iattr->ia_ctime = iattr->ia_mtime = | |
304 | current_fs_time(inode->i_sb); | |
305 | mask |= ATTR_CTIME | ATTR_MTIME; | |
306 | } | |
44d814ce | 307 | |
0f285c8a CH |
308 | if (iattr->ia_size > ip->i_size) { |
309 | ip->i_d.di_size = iattr->ia_size; | |
310 | ip->i_size = iattr->ia_size; | |
61436feb | 311 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
0f285c8a CH |
312 | } else if (iattr->ia_size <= ip->i_size || |
313 | (iattr->ia_size == 0 && ip->i_d.di_nextents)) { | |
1da177e4 LT |
314 | /* |
315 | * signal a sync transaction unless | |
316 | * we're truncating an already unlinked | |
317 | * file on a wsync filesystem | |
318 | */ | |
0f285c8a | 319 | code = xfs_itruncate_finish(&tp, ip, iattr->ia_size, |
1da177e4 LT |
320 | XFS_DATA_FORK, |
321 | ((ip->i_d.di_nlink != 0 || | |
322 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
323 | ? 1 : 0)); | |
7d4fb40a | 324 | if (code) |
1da177e4 | 325 | goto abort_return; |
7d4fb40a NS |
326 | /* |
327 | * Truncated "down", so we're removing references | |
328 | * to old data here - if we now delay flushing for | |
329 | * a long time, we expose ourselves unduly to the | |
330 | * notorious NULL files problem. So, we mark this | |
331 | * vnode and flush it when the file is closed, and | |
332 | * do not wait the usual (long) time for writeout. | |
333 | */ | |
b3aea4ed | 334 | xfs_iflags_set(ip, XFS_ITRUNCATED); |
1da177e4 | 335 | } |
c4cd747e CH |
336 | } else if (tp) { |
337 | xfs_trans_ijoin(tp, ip, lock_flags); | |
338 | xfs_trans_ihold(tp, ip); | |
1da177e4 LT |
339 | } |
340 | ||
341 | /* | |
342 | * Change file ownership. Must be the owner or privileged. | |
1da177e4 | 343 | */ |
0f285c8a | 344 | if (mask & (ATTR_UID|ATTR_GID)) { |
1da177e4 LT |
345 | /* |
346 | * CAP_FSETID overrides the following restrictions: | |
347 | * | |
348 | * The set-user-ID and set-group-ID bits of a file will be | |
349 | * cleared upon successful return from chown() | |
350 | */ | |
351 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
352 | !capable(CAP_FSETID)) { | |
353 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
354 | } | |
355 | ||
356 | /* | |
357 | * Change the ownerships and register quota modifications | |
358 | * in the transaction. | |
359 | */ | |
360 | if (iuid != uid) { | |
7d095257 | 361 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { |
0f285c8a | 362 | ASSERT(mask & ATTR_UID); |
1da177e4 | 363 | ASSERT(udqp); |
7d095257 | 364 | olddquot1 = xfs_qm_vop_chown(tp, ip, |
1da177e4 LT |
365 | &ip->i_udquot, udqp); |
366 | } | |
367 | ip->i_d.di_uid = uid; | |
f13fae2d | 368 | inode->i_uid = uid; |
1da177e4 LT |
369 | } |
370 | if (igid != gid) { | |
7d095257 | 371 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { |
c8ad20ff | 372 | ASSERT(!XFS_IS_PQUOTA_ON(mp)); |
0f285c8a | 373 | ASSERT(mask & ATTR_GID); |
1da177e4 | 374 | ASSERT(gdqp); |
7d095257 | 375 | olddquot2 = xfs_qm_vop_chown(tp, ip, |
1da177e4 LT |
376 | &ip->i_gdquot, gdqp); |
377 | } | |
378 | ip->i_d.di_gid = gid; | |
f13fae2d | 379 | inode->i_gid = gid; |
1da177e4 | 380 | } |
1da177e4 LT |
381 | } |
382 | ||
c4cd747e CH |
383 | /* |
384 | * Change file access modes. | |
385 | */ | |
386 | if (mask & ATTR_MODE) { | |
387 | umode_t mode = iattr->ia_mode; | |
388 | ||
389 | if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) | |
390 | mode &= ~S_ISGID; | |
391 | ||
392 | ip->i_d.di_mode &= S_IFMT; | |
393 | ip->i_d.di_mode |= mode & ~S_IFMT; | |
394 | ||
395 | inode->i_mode &= S_IFMT; | |
396 | inode->i_mode |= mode & ~S_IFMT; | |
c4cd747e | 397 | } |
1da177e4 LT |
398 | |
399 | /* | |
400 | * Change file access or modified times. | |
401 | */ | |
d6d59bad CH |
402 | if (mask & ATTR_ATIME) { |
403 | inode->i_atime = iattr->ia_atime; | |
404 | ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec; | |
405 | ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec; | |
406 | ip->i_update_core = 1; | |
1da177e4 | 407 | } |
d6d59bad | 408 | if (mask & ATTR_CTIME) { |
f13fae2d | 409 | inode->i_ctime = iattr->ia_ctime; |
0f285c8a CH |
410 | ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec; |
411 | ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec; | |
1da177e4 | 412 | ip->i_update_core = 1; |
d6d59bad CH |
413 | } |
414 | if (mask & ATTR_MTIME) { | |
415 | inode->i_mtime = iattr->ia_mtime; | |
416 | ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec; | |
417 | ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec; | |
418 | ip->i_update_core = 1; | |
1da177e4 LT |
419 | } |
420 | ||
421 | /* | |
d6d59bad CH |
422 | * And finally, log the inode core if any attribute in it |
423 | * has been changed. | |
1da177e4 | 424 | */ |
d6d59bad CH |
425 | if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE| |
426 | ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) | |
427 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1da177e4 LT |
428 | |
429 | XFS_STATS_INC(xs_ig_attrchg); | |
430 | ||
431 | /* | |
432 | * If this is a synchronous mount, make sure that the | |
433 | * transaction goes to disk before returning to the user. | |
434 | * This is slightly sub-optimal in that truncates require | |
c41564b5 | 435 | * two sync transactions instead of one for wsync filesystems. |
1da177e4 LT |
436 | * One for the truncate and one for the timestamps since we |
437 | * don't want to change the timestamps unless we're sure the | |
438 | * truncate worked. Truncates are less than 1% of the laddis | |
439 | * mix so this probably isn't worth the trouble to optimize. | |
440 | */ | |
441 | code = 0; | |
d6d59bad CH |
442 | if (mp->m_flags & XFS_MOUNT_WSYNC) |
443 | xfs_trans_set_sync(tp); | |
1da177e4 | 444 | |
d6d59bad | 445 | code = xfs_trans_commit(tp, commit_flags); |
1da177e4 | 446 | |
1da177e4 LT |
447 | xfs_iunlock(ip, lock_flags); |
448 | ||
449 | /* | |
450 | * Release any dquot(s) the inode had kept before chown. | |
451 | */ | |
7d095257 CH |
452 | xfs_qm_dqrele(olddquot1); |
453 | xfs_qm_dqrele(olddquot2); | |
454 | xfs_qm_dqrele(udqp); | |
455 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 456 | |
ef14f0c1 | 457 | if (code) |
1da177e4 | 458 | return code; |
ef14f0c1 CH |
459 | |
460 | /* | |
461 | * XXX(hch): Updating the ACL entries is not atomic vs the i_mode | |
462 | * update. We could avoid this with linked transactions | |
463 | * and passing down the transaction pointer all the way | |
464 | * to attr_set. No previous user of the generic | |
465 | * Posix ACL code seems to care about this issue either. | |
466 | */ | |
467 | if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) { | |
468 | code = -xfs_acl_chmod(inode); | |
469 | if (code) | |
470 | return XFS_ERROR(code); | |
1da177e4 LT |
471 | } |
472 | ||
eb9df39d | 473 | if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) && |
0f285c8a | 474 | !(flags & XFS_ATTR_DMI)) { |
bc4ac74a | 475 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL, |
1da177e4 LT |
476 | NULL, DM_RIGHT_NULL, NULL, NULL, |
477 | 0, 0, AT_DELAY_FLAG(flags)); | |
478 | } | |
479 | return 0; | |
480 | ||
481 | abort_return: | |
482 | commit_flags |= XFS_TRANS_ABORT; | |
483 | /* FALLTHROUGH */ | |
484 | error_return: | |
7d095257 CH |
485 | xfs_qm_dqrele(udqp); |
486 | xfs_qm_dqrele(gdqp); | |
1da177e4 LT |
487 | if (tp) { |
488 | xfs_trans_cancel(tp, commit_flags); | |
489 | } | |
490 | if (lock_flags != 0) { | |
491 | xfs_iunlock(ip, lock_flags); | |
492 | } | |
493 | return code; | |
494 | } | |
495 | ||
7d4fb40a NS |
496 | /* |
497 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
498 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
499 | * bmapi. | |
500 | */ | |
501 | #define SYMLINK_MAPS 2 | |
502 | ||
804c83c3 CH |
503 | STATIC int |
504 | xfs_readlink_bmap( | |
505 | xfs_inode_t *ip, | |
506 | char *link) | |
507 | { | |
508 | xfs_mount_t *mp = ip->i_mount; | |
509 | int pathlen = ip->i_d.di_size; | |
510 | int nmaps = SYMLINK_MAPS; | |
511 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
512 | xfs_daddr_t d; | |
513 | int byte_cnt; | |
514 | int n; | |
515 | xfs_buf_t *bp; | |
516 | int error = 0; | |
517 | ||
518 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0, | |
519 | mval, &nmaps, NULL, NULL); | |
520 | if (error) | |
521 | goto out; | |
522 | ||
523 | for (n = 0; n < nmaps; n++) { | |
524 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
525 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
526 | ||
6ad112bf CH |
527 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), |
528 | XBF_LOCK | XBF_MAPPED | XBF_DONT_BLOCK); | |
804c83c3 CH |
529 | error = XFS_BUF_GETERROR(bp); |
530 | if (error) { | |
531 | xfs_ioerror_alert("xfs_readlink", | |
532 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
533 | xfs_buf_relse(bp); | |
534 | goto out; | |
535 | } | |
536 | if (pathlen < byte_cnt) | |
537 | byte_cnt = pathlen; | |
538 | pathlen -= byte_cnt; | |
539 | ||
540 | memcpy(link, XFS_BUF_PTR(bp), byte_cnt); | |
541 | xfs_buf_relse(bp); | |
542 | } | |
543 | ||
544 | link[ip->i_d.di_size] = '\0'; | |
545 | error = 0; | |
546 | ||
547 | out: | |
548 | return error; | |
549 | } | |
550 | ||
993386c1 | 551 | int |
1da177e4 | 552 | xfs_readlink( |
993386c1 | 553 | xfs_inode_t *ip, |
804c83c3 | 554 | char *link) |
1da177e4 | 555 | { |
804c83c3 | 556 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 557 | int pathlen; |
1da177e4 | 558 | int error = 0; |
1da177e4 | 559 | |
cf441eeb | 560 | xfs_itrace_entry(ip); |
1da177e4 LT |
561 | |
562 | if (XFS_FORCED_SHUTDOWN(mp)) | |
563 | return XFS_ERROR(EIO); | |
564 | ||
565 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
566 | ||
567 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
804c83c3 | 568 | ASSERT(ip->i_d.di_size <= MAXPATHLEN); |
1da177e4 | 569 | |
804c83c3 CH |
570 | pathlen = ip->i_d.di_size; |
571 | if (!pathlen) | |
572 | goto out; | |
1da177e4 LT |
573 | |
574 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
804c83c3 CH |
575 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
576 | link[pathlen] = '\0'; | |
577 | } else { | |
578 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
579 | } |
580 | ||
804c83c3 | 581 | out: |
1da177e4 | 582 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
583 | return error; |
584 | } | |
585 | ||
c56c9631 CH |
586 | /* |
587 | * Flags for xfs_free_eofblocks | |
588 | */ | |
589 | #define XFS_FREE_EOF_TRYLOCK (1<<0) | |
590 | ||
1da177e4 | 591 | /* |
92dfe8d2 DC |
592 | * This is called by xfs_inactive to free any blocks beyond eof |
593 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
594 | * punching a hole to EOF. | |
1da177e4 | 595 | */ |
d96f8f89 | 596 | STATIC int |
92dfe8d2 | 597 | xfs_free_eofblocks( |
1da177e4 | 598 | xfs_mount_t *mp, |
92dfe8d2 DC |
599 | xfs_inode_t *ip, |
600 | int flags) | |
1da177e4 LT |
601 | { |
602 | xfs_trans_t *tp; | |
603 | int error; | |
604 | xfs_fileoff_t end_fsb; | |
605 | xfs_fileoff_t last_fsb; | |
606 | xfs_filblks_t map_len; | |
607 | int nimaps; | |
608 | xfs_bmbt_irec_t imap; | |
609 | ||
610 | /* | |
611 | * Figure out if there are any blocks beyond the end | |
612 | * of the file. If not, then there is nothing to do. | |
613 | */ | |
ba87ea69 | 614 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size)); |
1da177e4 LT |
615 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); |
616 | map_len = last_fsb - end_fsb; | |
617 | if (map_len <= 0) | |
014c2544 | 618 | return 0; |
1da177e4 LT |
619 | |
620 | nimaps = 1; | |
621 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
541d7d3c | 622 | error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0, |
3e57ecf6 | 623 | NULL, 0, &imap, &nimaps, NULL, NULL); |
1da177e4 LT |
624 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
625 | ||
626 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
627 | (imap.br_startblock != HOLESTARTBLOCK || |
628 | ip->i_delayed_blks)) { | |
1da177e4 LT |
629 | /* |
630 | * Attach the dquots to the inode up front. | |
631 | */ | |
7d095257 CH |
632 | error = xfs_qm_dqattach(ip, 0); |
633 | if (error) | |
014c2544 | 634 | return error; |
1da177e4 LT |
635 | |
636 | /* | |
637 | * There are blocks after the end of file. | |
638 | * Free them up now by truncating the file to | |
639 | * its current size. | |
640 | */ | |
641 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
642 | ||
643 | /* | |
644 | * Do the xfs_itruncate_start() call before | |
645 | * reserving any log space because | |
646 | * itruncate_start will call into the buffer | |
647 | * cache and we can't | |
648 | * do that within a transaction. | |
649 | */ | |
c56c9631 CH |
650 | if (flags & XFS_FREE_EOF_TRYLOCK) { |
651 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { | |
652 | xfs_trans_cancel(tp, 0); | |
653 | return 0; | |
654 | } | |
655 | } else { | |
92dfe8d2 | 656 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
c56c9631 | 657 | } |
d3cf2094 | 658 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, |
ba87ea69 | 659 | ip->i_size); |
d3cf2094 | 660 | if (error) { |
87ae3c24 | 661 | xfs_trans_cancel(tp, 0); |
c56c9631 | 662 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
d3cf2094 LM |
663 | return error; |
664 | } | |
1da177e4 LT |
665 | |
666 | error = xfs_trans_reserve(tp, 0, | |
667 | XFS_ITRUNCATE_LOG_RES(mp), | |
668 | 0, XFS_TRANS_PERM_LOG_RES, | |
669 | XFS_ITRUNCATE_LOG_COUNT); | |
670 | if (error) { | |
671 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
672 | xfs_trans_cancel(tp, 0); | |
673 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 674 | return error; |
1da177e4 LT |
675 | } |
676 | ||
677 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
678 | xfs_trans_ijoin(tp, ip, | |
679 | XFS_IOLOCK_EXCL | | |
680 | XFS_ILOCK_EXCL); | |
681 | xfs_trans_ihold(tp, ip); | |
682 | ||
683 | error = xfs_itruncate_finish(&tp, ip, | |
ba87ea69 | 684 | ip->i_size, |
1da177e4 LT |
685 | XFS_DATA_FORK, |
686 | 0); | |
687 | /* | |
688 | * If we get an error at this point we | |
689 | * simply don't bother truncating the file. | |
690 | */ | |
691 | if (error) { | |
692 | xfs_trans_cancel(tp, | |
693 | (XFS_TRANS_RELEASE_LOG_RES | | |
694 | XFS_TRANS_ABORT)); | |
695 | } else { | |
696 | error = xfs_trans_commit(tp, | |
1c72bf90 | 697 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 698 | } |
c56c9631 | 699 | xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL); |
1da177e4 | 700 | } |
014c2544 | 701 | return error; |
1da177e4 LT |
702 | } |
703 | ||
704 | /* | |
705 | * Free a symlink that has blocks associated with it. | |
706 | */ | |
707 | STATIC int | |
708 | xfs_inactive_symlink_rmt( | |
709 | xfs_inode_t *ip, | |
710 | xfs_trans_t **tpp) | |
711 | { | |
712 | xfs_buf_t *bp; | |
713 | int committed; | |
714 | int done; | |
715 | int error; | |
716 | xfs_fsblock_t first_block; | |
717 | xfs_bmap_free_t free_list; | |
718 | int i; | |
719 | xfs_mount_t *mp; | |
720 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
721 | int nmaps; | |
722 | xfs_trans_t *ntp; | |
723 | int size; | |
724 | xfs_trans_t *tp; | |
725 | ||
726 | tp = *tpp; | |
727 | mp = ip->i_mount; | |
728 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
729 | /* | |
730 | * We're freeing a symlink that has some | |
731 | * blocks allocated to it. Free the | |
732 | * blocks here. We know that we've got | |
733 | * either 1 or 2 extents and that we can | |
734 | * free them all in one bunmapi call. | |
735 | */ | |
736 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
737 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
738 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
739 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
740 | xfs_trans_cancel(tp, 0); | |
741 | *tpp = NULL; | |
742 | return error; | |
743 | } | |
744 | /* | |
745 | * Lock the inode, fix the size, and join it to the transaction. | |
746 | * Hold it so in the normal path, we still have it locked for | |
747 | * the second transaction. In the error paths we need it | |
748 | * held so the cancel won't rele it, see below. | |
749 | */ | |
750 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
751 | size = (int)ip->i_d.di_size; | |
752 | ip->i_d.di_size = 0; | |
753 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
754 | xfs_trans_ihold(tp, ip); | |
755 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
756 | /* | |
757 | * Find the block(s) so we can inval and unmap them. | |
758 | */ | |
759 | done = 0; | |
9d87c319 | 760 | xfs_bmap_init(&free_list, &first_block); |
e8c96f8c | 761 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
762 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
763 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 764 | &free_list, NULL))) |
1da177e4 LT |
765 | goto error0; |
766 | /* | |
767 | * Invalidate the block(s). | |
768 | */ | |
769 | for (i = 0; i < nmaps; i++) { | |
770 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
771 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
772 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
773 | xfs_trans_binval(tp, bp); | |
774 | } | |
775 | /* | |
776 | * Unmap the dead block(s) to the free_list. | |
777 | */ | |
778 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 779 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
780 | goto error1; |
781 | ASSERT(done); | |
782 | /* | |
783 | * Commit the first transaction. This logs the EFI and the inode. | |
784 | */ | |
f7c99b6f | 785 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
786 | goto error1; |
787 | /* | |
788 | * The transaction must have been committed, since there were | |
789 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
790 | * The new tp has the extent freeing and EFDs. | |
791 | */ | |
792 | ASSERT(committed); | |
793 | /* | |
794 | * The first xact was committed, so add the inode to the new one. | |
795 | * Mark it dirty so it will be logged and moved forward in the log as | |
796 | * part of every commit. | |
797 | */ | |
798 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
799 | xfs_trans_ihold(tp, ip); | |
800 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
801 | /* | |
802 | * Get a new, empty transaction to return to our caller. | |
803 | */ | |
804 | ntp = xfs_trans_dup(tp); | |
805 | /* | |
c41564b5 | 806 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
807 | * If we get an error on the commit here or on the reserve below, |
808 | * we need to unlock the inode since the new transaction doesn't | |
809 | * have the inode attached. | |
810 | */ | |
1c72bf90 | 811 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
812 | tp = ntp; |
813 | if (error) { | |
814 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
815 | goto error0; | |
816 | } | |
cc09c0dc DC |
817 | /* |
818 | * transaction commit worked ok so we can drop the extra ticket | |
819 | * reference that we gained in xfs_trans_dup() | |
820 | */ | |
821 | xfs_log_ticket_put(tp->t_ticket); | |
822 | ||
1da177e4 LT |
823 | /* |
824 | * Remove the memory for extent descriptions (just bookkeeping). | |
825 | */ | |
826 | if (ip->i_df.if_bytes) | |
827 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
828 | ASSERT(ip->i_df.if_bytes == 0); | |
829 | /* | |
830 | * Put an itruncate log reservation in the new transaction | |
831 | * for our caller. | |
832 | */ | |
833 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
834 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
835 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
836 | goto error0; | |
837 | } | |
838 | /* | |
839 | * Return with the inode locked but not joined to the transaction. | |
840 | */ | |
841 | *tpp = tp; | |
842 | return 0; | |
843 | ||
844 | error1: | |
845 | xfs_bmap_cancel(&free_list); | |
846 | error0: | |
847 | /* | |
848 | * Have to come here with the inode locked and either | |
849 | * (held and in the transaction) or (not in the transaction). | |
850 | * If the inode isn't held then cancel would iput it, but | |
851 | * that's wrong since this is inactive and the vnode ref | |
852 | * count is 0 already. | |
853 | * Cancel won't do anything to the inode if held, but it still | |
854 | * needs to be locked until the cancel is done, if it was | |
855 | * joined to the transaction. | |
856 | */ | |
857 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
858 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
859 | *tpp = NULL; | |
860 | return error; | |
861 | ||
862 | } | |
863 | ||
864 | STATIC int | |
865 | xfs_inactive_symlink_local( | |
866 | xfs_inode_t *ip, | |
867 | xfs_trans_t **tpp) | |
868 | { | |
869 | int error; | |
870 | ||
871 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
872 | /* | |
873 | * We're freeing a symlink which fit into | |
874 | * the inode. Just free the memory used | |
875 | * to hold the old symlink. | |
876 | */ | |
877 | error = xfs_trans_reserve(*tpp, 0, | |
878 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
879 | 0, XFS_TRANS_PERM_LOG_RES, | |
880 | XFS_ITRUNCATE_LOG_COUNT); | |
881 | ||
882 | if (error) { | |
883 | xfs_trans_cancel(*tpp, 0); | |
884 | *tpp = NULL; | |
014c2544 | 885 | return error; |
1da177e4 LT |
886 | } |
887 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
888 | ||
889 | /* | |
890 | * Zero length symlinks _can_ exist. | |
891 | */ | |
892 | if (ip->i_df.if_bytes > 0) { | |
893 | xfs_idata_realloc(ip, | |
894 | -(ip->i_df.if_bytes), | |
895 | XFS_DATA_FORK); | |
896 | ASSERT(ip->i_df.if_bytes == 0); | |
897 | } | |
014c2544 | 898 | return 0; |
1da177e4 LT |
899 | } |
900 | ||
1da177e4 LT |
901 | STATIC int |
902 | xfs_inactive_attrs( | |
903 | xfs_inode_t *ip, | |
904 | xfs_trans_t **tpp) | |
905 | { | |
906 | xfs_trans_t *tp; | |
907 | int error; | |
908 | xfs_mount_t *mp; | |
909 | ||
579aa9ca | 910 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
1da177e4 LT |
911 | tp = *tpp; |
912 | mp = ip->i_mount; | |
913 | ASSERT(ip->i_d.di_forkoff != 0); | |
e5720eec | 914 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 915 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
e5720eec DC |
916 | if (error) |
917 | goto error_unlock; | |
1da177e4 LT |
918 | |
919 | error = xfs_attr_inactive(ip); | |
e5720eec DC |
920 | if (error) |
921 | goto error_unlock; | |
1da177e4 LT |
922 | |
923 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
924 | error = xfs_trans_reserve(tp, 0, | |
925 | XFS_IFREE_LOG_RES(mp), | |
926 | 0, XFS_TRANS_PERM_LOG_RES, | |
927 | XFS_INACTIVE_LOG_COUNT); | |
e5720eec DC |
928 | if (error) |
929 | goto error_cancel; | |
1da177e4 LT |
930 | |
931 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
932 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
933 | xfs_trans_ihold(tp, ip); | |
934 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
935 | ||
936 | ASSERT(ip->i_d.di_anextents == 0); | |
937 | ||
938 | *tpp = tp; | |
014c2544 | 939 | return 0; |
e5720eec DC |
940 | |
941 | error_cancel: | |
942 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
943 | xfs_trans_cancel(tp, 0); | |
944 | error_unlock: | |
945 | *tpp = NULL; | |
946 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
947 | return error; | |
1da177e4 LT |
948 | } |
949 | ||
993386c1 | 950 | int |
1da177e4 | 951 | xfs_release( |
993386c1 | 952 | xfs_inode_t *ip) |
1da177e4 | 953 | { |
993386c1 | 954 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
955 | int error; |
956 | ||
42173f68 | 957 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) |
1da177e4 | 958 | return 0; |
1da177e4 LT |
959 | |
960 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 961 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
962 | return 0; |
963 | ||
2a82b8be | 964 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
965 | int truncated; |
966 | ||
2a82b8be DC |
967 | /* |
968 | * If we are using filestreams, and we have an unlinked | |
969 | * file that we are processing the last close on, then nothing | |
970 | * will be able to reopen and write to this file. Purge this | |
971 | * inode from the filestreams cache so that it doesn't delay | |
972 | * teardown of the inode. | |
973 | */ | |
974 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
975 | xfs_filestream_deassociate(ip); | |
976 | ||
fbf3ce8d CH |
977 | /* |
978 | * If we previously truncated this file and removed old data | |
979 | * in the process, we want to initiate "early" writeout on | |
980 | * the last close. This is an attempt to combat the notorious | |
981 | * NULL files problem which is particularly noticable from a | |
982 | * truncate down, buffered (re-)write (delalloc), followed by | |
983 | * a crash. What we are effectively doing here is | |
984 | * significantly reducing the time window where we'd otherwise | |
985 | * be exposed to that problem. | |
986 | */ | |
09262b43 | 987 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
df80c933 | 988 | if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) |
0cadda1c | 989 | xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE); |
fbf3ce8d CH |
990 | } |
991 | ||
1da177e4 LT |
992 | if (ip->i_d.di_nlink != 0) { |
993 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
df80c933 | 994 | ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 || |
68bdb6ea | 995 | ip->i_delayed_blks > 0)) && |
1da177e4 | 996 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
997 | (!(ip->i_d.di_flags & |
998 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
c56c9631 CH |
999 | |
1000 | /* | |
1001 | * If we can't get the iolock just skip truncating | |
1002 | * the blocks past EOF because we could deadlock | |
1003 | * with the mmap_sem otherwise. We'll get another | |
1004 | * chance to drop them once the last reference to | |
1005 | * the inode is dropped, so we'll never leak blocks | |
1006 | * permanently. | |
1007 | */ | |
1008 | error = xfs_free_eofblocks(mp, ip, | |
1009 | XFS_FREE_EOF_TRYLOCK); | |
92dfe8d2 | 1010 | if (error) |
014c2544 | 1011 | return error; |
1da177e4 LT |
1012 | } |
1013 | } | |
1014 | ||
1015 | return 0; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * xfs_inactive | |
1020 | * | |
1021 | * This is called when the vnode reference count for the vnode | |
1022 | * goes to zero. If the file has been unlinked, then it must | |
1023 | * now be truncated. Also, we clear all of the read-ahead state | |
1024 | * kept for the inode here since the file is now closed. | |
1025 | */ | |
993386c1 | 1026 | int |
1da177e4 | 1027 | xfs_inactive( |
993386c1 | 1028 | xfs_inode_t *ip) |
1da177e4 | 1029 | { |
67fcaa73 | 1030 | xfs_bmap_free_t free_list; |
1da177e4 LT |
1031 | xfs_fsblock_t first_block; |
1032 | int committed; | |
1033 | xfs_trans_t *tp; | |
1034 | xfs_mount_t *mp; | |
1035 | int error; | |
1036 | int truncate; | |
1037 | ||
cf441eeb | 1038 | xfs_itrace_entry(ip); |
1da177e4 | 1039 | |
1da177e4 LT |
1040 | /* |
1041 | * If the inode is already free, then there can be nothing | |
1042 | * to clean up here. | |
1043 | */ | |
cb4c8cc1 | 1044 | if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) { |
1da177e4 LT |
1045 | ASSERT(ip->i_df.if_real_bytes == 0); |
1046 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1047 | return VN_INACTIVE_CACHE; | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * Only do a truncate if it's a regular file with | |
1052 | * some actual space in it. It's OK to look at the | |
1053 | * inode's fields without the lock because we're the | |
1054 | * only one with a reference to the inode. | |
1055 | */ | |
1056 | truncate = ((ip->i_d.di_nlink == 0) && | |
ba87ea69 LM |
1057 | ((ip->i_d.di_size != 0) || (ip->i_size != 0) || |
1058 | (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1059 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1060 | ||
1061 | mp = ip->i_mount; | |
1062 | ||
bc4ac74a CH |
1063 | if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) |
1064 | XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL); | |
1da177e4 LT |
1065 | |
1066 | error = 0; | |
1067 | ||
1068 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1069 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1070 | goto out; |
1071 | ||
1072 | if (ip->i_d.di_nlink != 0) { | |
1073 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
df80c933 | 1074 | ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 || |
68bdb6ea | 1075 | ip->i_delayed_blks > 0)) && |
dd9f438e NS |
1076 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1077 | (!(ip->i_d.di_flags & | |
1078 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1079 | (ip->i_delayed_blks != 0)))) { | |
c56c9631 | 1080 | error = xfs_free_eofblocks(mp, ip, 0); |
92dfe8d2 | 1081 | if (error) |
014c2544 | 1082 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1083 | } |
1084 | goto out; | |
1085 | } | |
1086 | ||
1087 | ASSERT(ip->i_d.di_nlink == 0); | |
1088 | ||
7d095257 CH |
1089 | error = xfs_qm_dqattach(ip, 0); |
1090 | if (error) | |
014c2544 | 1091 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1092 | |
1093 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1094 | if (truncate) { | |
1095 | /* | |
1096 | * Do the xfs_itruncate_start() call before | |
1097 | * reserving any log space because itruncate_start | |
1098 | * will call into the buffer cache and we can't | |
1099 | * do that within a transaction. | |
1100 | */ | |
1101 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1102 | ||
d3cf2094 LM |
1103 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); |
1104 | if (error) { | |
87ae3c24 | 1105 | xfs_trans_cancel(tp, 0); |
d3cf2094 LM |
1106 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1107 | return VN_INACTIVE_CACHE; | |
1108 | } | |
1da177e4 LT |
1109 | |
1110 | error = xfs_trans_reserve(tp, 0, | |
1111 | XFS_ITRUNCATE_LOG_RES(mp), | |
1112 | 0, XFS_TRANS_PERM_LOG_RES, | |
1113 | XFS_ITRUNCATE_LOG_COUNT); | |
1114 | if (error) { | |
1115 | /* Don't call itruncate_cleanup */ | |
1116 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1117 | xfs_trans_cancel(tp, 0); | |
1118 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1119 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1120 | } |
1121 | ||
1122 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1123 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1124 | xfs_trans_ihold(tp, ip); | |
1125 | ||
1126 | /* | |
1127 | * normally, we have to run xfs_itruncate_finish sync. | |
1128 | * But if filesystem is wsync and we're in the inactive | |
1129 | * path, then we know that nlink == 0, and that the | |
1130 | * xaction that made nlink == 0 is permanently committed | |
1131 | * since xfs_remove runs as a synchronous transaction. | |
1132 | */ | |
1133 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1134 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1135 | ||
1136 | if (error) { | |
1137 | xfs_trans_cancel(tp, | |
1138 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1139 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1140 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1141 | } |
1142 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1143 | ||
1144 | /* | |
1145 | * If we get an error while cleaning up a | |
1146 | * symlink we bail out. | |
1147 | */ | |
1148 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1149 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1150 | xfs_inactive_symlink_local(ip, &tp); | |
1151 | ||
1152 | if (error) { | |
1153 | ASSERT(tp == NULL); | |
014c2544 | 1154 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1155 | } |
1156 | ||
1157 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1158 | xfs_trans_ihold(tp, ip); | |
1159 | } else { | |
1160 | error = xfs_trans_reserve(tp, 0, | |
1161 | XFS_IFREE_LOG_RES(mp), | |
1162 | 0, XFS_TRANS_PERM_LOG_RES, | |
1163 | XFS_INACTIVE_LOG_COUNT); | |
1164 | if (error) { | |
1165 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1166 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1167 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1168 | } |
1169 | ||
1170 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1171 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1172 | xfs_trans_ihold(tp, ip); | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * If there are attributes associated with the file | |
1177 | * then blow them away now. The code calls a routine | |
1178 | * that recursively deconstructs the attribute fork. | |
1179 | * We need to just commit the current transaction | |
1180 | * because we can't use it for xfs_attr_inactive(). | |
1181 | */ | |
1182 | if (ip->i_d.di_anextents > 0) { | |
1183 | error = xfs_inactive_attrs(ip, &tp); | |
1184 | /* | |
1185 | * If we got an error, the transaction is already | |
1186 | * cancelled, and the inode is unlocked. Just get out. | |
1187 | */ | |
1188 | if (error) | |
014c2544 | 1189 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1190 | } else if (ip->i_afp) { |
1191 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * Free the inode. | |
1196 | */ | |
9d87c319 | 1197 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
1198 | error = xfs_ifree(tp, ip, &free_list); |
1199 | if (error) { | |
1200 | /* | |
1201 | * If we fail to free the inode, shut down. The cancel | |
1202 | * might do that, we need to make sure. Otherwise the | |
1203 | * inode might be lost for a long time or forever. | |
1204 | */ | |
1205 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1206 | cmn_err(CE_NOTE, | |
1207 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1208 | error, mp->m_fsname); | |
7d04a335 | 1209 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1210 | } |
1211 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1212 | } else { | |
1213 | /* | |
1214 | * Credit the quota account(s). The inode is gone. | |
1215 | */ | |
7d095257 | 1216 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); |
1da177e4 LT |
1217 | |
1218 | /* | |
78e9da77 DC |
1219 | * Just ignore errors at this point. There is nothing we can |
1220 | * do except to try to keep going. Make sure it's not a silent | |
1221 | * error. | |
1da177e4 | 1222 | */ |
78e9da77 DC |
1223 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1224 | if (error) | |
1225 | xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: " | |
1226 | "xfs_bmap_finish() returned error %d", error); | |
1227 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1228 | if (error) | |
1229 | xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: " | |
1230 | "xfs_trans_commit() returned error %d", error); | |
1da177e4 | 1231 | } |
7d095257 | 1232 | |
1da177e4 LT |
1233 | /* |
1234 | * Release the dquots held by inode, if any. | |
1235 | */ | |
7d095257 | 1236 | xfs_qm_dqdetach(ip); |
1da177e4 LT |
1237 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
1238 | ||
1239 | out: | |
1240 | return VN_INACTIVE_CACHE; | |
1241 | } | |
1242 | ||
384f3ced BN |
1243 | /* |
1244 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
1245 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
1246 | * ci_name->name will point to a the actual name (caller must free) or | |
1247 | * will be set to NULL if an exact match is found. | |
1248 | */ | |
993386c1 | 1249 | int |
1da177e4 | 1250 | xfs_lookup( |
993386c1 | 1251 | xfs_inode_t *dp, |
556b8b16 | 1252 | struct xfs_name *name, |
384f3ced BN |
1253 | xfs_inode_t **ipp, |
1254 | struct xfs_name *ci_name) | |
1da177e4 | 1255 | { |
eca450b7 | 1256 | xfs_ino_t inum; |
1da177e4 LT |
1257 | int error; |
1258 | uint lock_mode; | |
1da177e4 | 1259 | |
cf441eeb | 1260 | xfs_itrace_entry(dp); |
1da177e4 LT |
1261 | |
1262 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1263 | return XFS_ERROR(EIO); | |
1264 | ||
1265 | lock_mode = xfs_ilock_map_shared(dp); | |
384f3ced | 1266 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
1da177e4 | 1267 | xfs_iunlock_map_shared(dp, lock_mode); |
eca450b7 CH |
1268 | |
1269 | if (error) | |
1270 | goto out; | |
1271 | ||
1272 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0); | |
1273 | if (error) | |
384f3ced | 1274 | goto out_free_name; |
eca450b7 | 1275 | |
eca450b7 CH |
1276 | return 0; |
1277 | ||
384f3ced BN |
1278 | out_free_name: |
1279 | if (ci_name) | |
1280 | kmem_free(ci_name->name); | |
1281 | out: | |
eca450b7 | 1282 | *ipp = NULL; |
1da177e4 LT |
1283 | return error; |
1284 | } | |
1285 | ||
993386c1 | 1286 | int |
1da177e4 | 1287 | xfs_create( |
993386c1 | 1288 | xfs_inode_t *dp, |
556b8b16 | 1289 | struct xfs_name *name, |
3e5daf05 CH |
1290 | mode_t mode, |
1291 | xfs_dev_t rdev, | |
979ebab1 | 1292 | xfs_inode_t **ipp, |
1da177e4 LT |
1293 | cred_t *credp) |
1294 | { | |
517b5e8c CH |
1295 | int is_dir = S_ISDIR(mode); |
1296 | struct xfs_mount *mp = dp->i_mount; | |
1297 | struct xfs_inode *ip = NULL; | |
1298 | struct xfs_trans *tp = NULL; | |
556b8b16 | 1299 | int error; |
1da177e4 LT |
1300 | xfs_bmap_free_t free_list; |
1301 | xfs_fsblock_t first_block; | |
993386c1 | 1302 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1303 | uint cancel_flags; |
1304 | int committed; | |
1305 | xfs_prid_t prid; | |
517b5e8c CH |
1306 | struct xfs_dquot *udqp = NULL; |
1307 | struct xfs_dquot *gdqp = NULL; | |
1da177e4 | 1308 | uint resblks; |
517b5e8c CH |
1309 | uint log_res; |
1310 | uint log_count; | |
1da177e4 | 1311 | |
cf441eeb | 1312 | xfs_itrace_entry(dp); |
1da177e4 | 1313 | |
517b5e8c CH |
1314 | if (XFS_FORCED_SHUTDOWN(mp)) |
1315 | return XFS_ERROR(EIO); | |
1316 | ||
eb9df39d | 1317 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 | 1318 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
bc4ac74a | 1319 | dp, DM_RIGHT_NULL, NULL, |
556b8b16 | 1320 | DM_RIGHT_NULL, name->name, NULL, |
3e5daf05 | 1321 | mode, 0, 0); |
1da177e4 LT |
1322 | |
1323 | if (error) | |
1324 | return error; | |
1da177e4 LT |
1325 | } |
1326 | ||
365ca83d NS |
1327 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1328 | prid = dp->i_d.di_projid; | |
1da177e4 | 1329 | else |
517b5e8c | 1330 | prid = dfltprid; |
1da177e4 LT |
1331 | |
1332 | /* | |
1333 | * Make sure that we have allocated dquot(s) on disk. | |
1334 | */ | |
7d095257 | 1335 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
517b5e8c | 1336 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1da177e4 LT |
1337 | if (error) |
1338 | goto std_return; | |
1339 | ||
517b5e8c CH |
1340 | if (is_dir) { |
1341 | rdev = 0; | |
1342 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); | |
1343 | log_res = XFS_MKDIR_LOG_RES(mp); | |
1344 | log_count = XFS_MKDIR_LOG_COUNT; | |
1345 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
1346 | } else { | |
1347 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
1348 | log_res = XFS_CREATE_LOG_RES(mp); | |
1349 | log_count = XFS_CREATE_LOG_COUNT; | |
1350 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1351 | } | |
1da177e4 | 1352 | |
1da177e4 | 1353 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
517b5e8c | 1354 | |
1da177e4 LT |
1355 | /* |
1356 | * Initially assume that the file does not exist and | |
1357 | * reserve the resources for that case. If that is not | |
1358 | * the case we'll drop the one we have and get a more | |
1359 | * appropriate transaction later. | |
1360 | */ | |
517b5e8c CH |
1361 | error = xfs_trans_reserve(tp, resblks, log_res, 0, |
1362 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 | 1363 | if (error == ENOSPC) { |
153fec43 DC |
1364 | /* flush outstanding delalloc blocks and retry */ |
1365 | xfs_flush_inodes(dp); | |
4734d401 CH |
1366 | error = xfs_trans_reserve(tp, resblks, log_res, 0, |
1367 | XFS_TRANS_PERM_LOG_RES, log_count); | |
153fec43 DC |
1368 | } |
1369 | if (error == ENOSPC) { | |
1370 | /* No space at all so try a "no-allocation" reservation */ | |
1da177e4 | 1371 | resblks = 0; |
517b5e8c CH |
1372 | error = xfs_trans_reserve(tp, 0, log_res, 0, |
1373 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 LT |
1374 | } |
1375 | if (error) { | |
1376 | cancel_flags = 0; | |
517b5e8c | 1377 | goto out_trans_cancel; |
1da177e4 LT |
1378 | } |
1379 | ||
f7c66ce3 | 1380 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1381 | unlock_dp_on_error = B_TRUE; |
1da177e4 | 1382 | |
517b5e8c CH |
1383 | /* |
1384 | * Check for directory link count overflow. | |
1385 | */ | |
1386 | if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) { | |
1387 | error = XFS_ERROR(EMLINK); | |
1388 | goto out_trans_cancel; | |
1389 | } | |
1da177e4 | 1390 | |
517b5e8c | 1391 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
1392 | |
1393 | /* | |
1394 | * Reserve disk quota and the inode. | |
1395 | */ | |
7d095257 | 1396 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 | 1397 | if (error) |
517b5e8c | 1398 | goto out_trans_cancel; |
1da177e4 | 1399 | |
556b8b16 BN |
1400 | error = xfs_dir_canenter(tp, dp, name, resblks); |
1401 | if (error) | |
517b5e8c CH |
1402 | goto out_trans_cancel; |
1403 | ||
1404 | /* | |
1405 | * A newly created regular or special file just has one directory | |
1406 | * entry pointing to them, but a directory also the "." entry | |
1407 | * pointing to itself. | |
1408 | */ | |
1409 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, credp, | |
1410 | prid, resblks > 0, &ip, &committed); | |
1da177e4 LT |
1411 | if (error) { |
1412 | if (error == ENOSPC) | |
517b5e8c CH |
1413 | goto out_trans_cancel; |
1414 | goto out_trans_abort; | |
1da177e4 | 1415 | } |
1da177e4 LT |
1416 | |
1417 | /* | |
1418 | * At this point, we've gotten a newly allocated inode. | |
1419 | * It is locked (and joined to the transaction). | |
1420 | */ | |
579aa9ca | 1421 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
1422 | |
1423 | /* | |
993386c1 CH |
1424 | * Now we join the directory inode to the transaction. We do not do it |
1425 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1426 | * (and release all the locks). An error from here on will result in | |
1427 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1428 | * error path. | |
1da177e4 | 1429 | */ |
979ebab1 | 1430 | IHOLD(dp); |
1da177e4 | 1431 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 1432 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 1433 | |
556b8b16 | 1434 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
f6c2d1fa NS |
1435 | &first_block, &free_list, resblks ? |
1436 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1437 | if (error) { |
1438 | ASSERT(error != ENOSPC); | |
517b5e8c | 1439 | goto out_trans_abort; |
1da177e4 LT |
1440 | } |
1441 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1442 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1443 | ||
517b5e8c CH |
1444 | if (is_dir) { |
1445 | error = xfs_dir_init(tp, ip, dp); | |
1446 | if (error) | |
1447 | goto out_bmap_cancel; | |
1448 | ||
1449 | error = xfs_bumplink(tp, dp); | |
1450 | if (error) | |
1451 | goto out_bmap_cancel; | |
1452 | } | |
1453 | ||
1da177e4 LT |
1454 | /* |
1455 | * If this is a synchronous mount, make sure that the | |
1456 | * create transaction goes to disk before returning to | |
1457 | * the user. | |
1458 | */ | |
517b5e8c | 1459 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 1460 | xfs_trans_set_sync(tp); |
1da177e4 | 1461 | |
1da177e4 LT |
1462 | /* |
1463 | * Attach the dquot(s) to the inodes and modify them incore. | |
1464 | * These ids of the inode couldn't have changed since the new | |
1465 | * inode has been locked ever since it was created. | |
1466 | */ | |
7d095257 | 1467 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 LT |
1468 | |
1469 | /* | |
1470 | * xfs_trans_commit normally decrements the vnode ref count | |
1471 | * when it unlocks the inode. Since we want to return the | |
1472 | * vnode to the caller, we bump the vnode ref count now. | |
1473 | */ | |
1474 | IHOLD(ip); | |
1da177e4 | 1475 | |
f7c99b6f | 1476 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
517b5e8c CH |
1477 | if (error) |
1478 | goto out_abort_rele; | |
1da177e4 | 1479 | |
1c72bf90 | 1480 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1481 | if (error) { |
1482 | IRELE(ip); | |
517b5e8c | 1483 | goto out_dqrele; |
1da177e4 LT |
1484 | } |
1485 | ||
7d095257 CH |
1486 | xfs_qm_dqrele(udqp); |
1487 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1488 | |
979ebab1 | 1489 | *ipp = ip; |
1da177e4 LT |
1490 | |
1491 | /* Fallthrough to std_return with error = 0 */ | |
517b5e8c CH |
1492 | std_return: |
1493 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { | |
1494 | XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, dp, DM_RIGHT_NULL, | |
1495 | ip, DM_RIGHT_NULL, name->name, NULL, mode, | |
1496 | error, 0); | |
1da177e4 | 1497 | } |
517b5e8c | 1498 | |
1da177e4 LT |
1499 | return error; |
1500 | ||
517b5e8c CH |
1501 | out_bmap_cancel: |
1502 | xfs_bmap_cancel(&free_list); | |
1503 | out_trans_abort: | |
1da177e4 | 1504 | cancel_flags |= XFS_TRANS_ABORT; |
517b5e8c CH |
1505 | out_trans_cancel: |
1506 | xfs_trans_cancel(tp, cancel_flags); | |
1507 | out_dqrele: | |
7d095257 CH |
1508 | xfs_qm_dqrele(udqp); |
1509 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1510 | |
993386c1 CH |
1511 | if (unlock_dp_on_error) |
1512 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1513 | ||
1da177e4 LT |
1514 | goto std_return; |
1515 | ||
517b5e8c | 1516 | out_abort_rele: |
1da177e4 LT |
1517 | /* |
1518 | * Wait until after the current transaction is aborted to | |
1519 | * release the inode. This prevents recursive transactions | |
1520 | * and deadlocks from xfs_inactive. | |
1521 | */ | |
517b5e8c | 1522 | xfs_bmap_cancel(&free_list); |
1da177e4 LT |
1523 | cancel_flags |= XFS_TRANS_ABORT; |
1524 | xfs_trans_cancel(tp, cancel_flags); | |
1525 | IRELE(ip); | |
517b5e8c CH |
1526 | unlock_dp_on_error = B_FALSE; |
1527 | goto out_dqrele; | |
1da177e4 LT |
1528 | } |
1529 | ||
1da177e4 LT |
1530 | #ifdef DEBUG |
1531 | int xfs_locked_n; | |
1532 | int xfs_small_retries; | |
1533 | int xfs_middle_retries; | |
1534 | int xfs_lots_retries; | |
1535 | int xfs_lock_delays; | |
1536 | #endif | |
1537 | ||
f7c66ce3 LM |
1538 | /* |
1539 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
1540 | * a different value | |
1541 | */ | |
1542 | static inline int | |
1543 | xfs_lock_inumorder(int lock_mode, int subclass) | |
1544 | { | |
1545 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 1546 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 1547 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 1548 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
1549 | |
1550 | return lock_mode; | |
1551 | } | |
1552 | ||
1da177e4 LT |
1553 | /* |
1554 | * The following routine will lock n inodes in exclusive mode. | |
1555 | * We assume the caller calls us with the inodes in i_ino order. | |
1556 | * | |
1557 | * We need to detect deadlock where an inode that we lock | |
1558 | * is in the AIL and we start waiting for another inode that is locked | |
1559 | * by a thread in a long running transaction (such as truncate). This can | |
1560 | * result in deadlock since the long running trans might need to wait | |
1561 | * for the inode we just locked in order to push the tail and free space | |
1562 | * in the log. | |
1563 | */ | |
1564 | void | |
1565 | xfs_lock_inodes( | |
1566 | xfs_inode_t **ips, | |
1567 | int inodes, | |
1da177e4 LT |
1568 | uint lock_mode) |
1569 | { | |
1570 | int attempts = 0, i, j, try_lock; | |
1571 | xfs_log_item_t *lp; | |
1572 | ||
1573 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
1574 | ||
cfa853e4 CH |
1575 | try_lock = 0; |
1576 | i = 0; | |
1da177e4 LT |
1577 | |
1578 | again: | |
1579 | for (; i < inodes; i++) { | |
1580 | ASSERT(ips[i]); | |
1581 | ||
1582 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
1583 | continue; | |
1584 | ||
1585 | /* | |
1586 | * If try_lock is not set yet, make sure all locked inodes | |
1587 | * are not in the AIL. | |
1588 | * If any are, set try_lock to be used later. | |
1589 | */ | |
1590 | ||
1591 | if (!try_lock) { | |
1592 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
1593 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
1594 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1595 | try_lock++; | |
1596 | } | |
1597 | } | |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * If any of the previous locks we have locked is in the AIL, | |
1602 | * we must TRY to get the second and subsequent locks. If | |
1603 | * we can't get any, we must release all we have | |
1604 | * and try again. | |
1605 | */ | |
1606 | ||
1607 | if (try_lock) { | |
1608 | /* try_lock must be 0 if i is 0. */ | |
1609 | /* | |
1610 | * try_lock means we have an inode locked | |
1611 | * that is in the AIL. | |
1612 | */ | |
1613 | ASSERT(i != 0); | |
f7c66ce3 | 1614 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
1615 | attempts++; |
1616 | ||
1617 | /* | |
1618 | * Unlock all previous guys and try again. | |
1619 | * xfs_iunlock will try to push the tail | |
1620 | * if the inode is in the AIL. | |
1621 | */ | |
1622 | ||
1623 | for(j = i - 1; j >= 0; j--) { | |
1624 | ||
1625 | /* | |
1626 | * Check to see if we've already | |
1627 | * unlocked this one. | |
1628 | * Not the first one going back, | |
1629 | * and the inode ptr is the same. | |
1630 | */ | |
1631 | if ((j != (i - 1)) && ips[j] == | |
1632 | ips[j+1]) | |
1633 | continue; | |
1634 | ||
1635 | xfs_iunlock(ips[j], lock_mode); | |
1636 | } | |
1637 | ||
1638 | if ((attempts % 5) == 0) { | |
1639 | delay(1); /* Don't just spin the CPU */ | |
1640 | #ifdef DEBUG | |
1641 | xfs_lock_delays++; | |
1642 | #endif | |
1643 | } | |
1644 | i = 0; | |
1645 | try_lock = 0; | |
1646 | goto again; | |
1647 | } | |
1648 | } else { | |
f7c66ce3 | 1649 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
1650 | } |
1651 | } | |
1652 | ||
1653 | #ifdef DEBUG | |
1654 | if (attempts) { | |
1655 | if (attempts < 5) xfs_small_retries++; | |
1656 | else if (attempts < 100) xfs_middle_retries++; | |
1657 | else xfs_lots_retries++; | |
1658 | } else { | |
1659 | xfs_locked_n++; | |
1660 | } | |
1661 | #endif | |
1662 | } | |
1663 | ||
f9114eba DC |
1664 | /* |
1665 | * xfs_lock_two_inodes() can only be used to lock one type of lock | |
1666 | * at a time - the iolock or the ilock, but not both at once. If | |
1667 | * we lock both at once, lockdep will report false positives saying | |
1668 | * we have violated locking orders. | |
1669 | */ | |
e1cccd91 CH |
1670 | void |
1671 | xfs_lock_two_inodes( | |
1672 | xfs_inode_t *ip0, | |
1673 | xfs_inode_t *ip1, | |
1674 | uint lock_mode) | |
1675 | { | |
1676 | xfs_inode_t *temp; | |
1677 | int attempts = 0; | |
1678 | xfs_log_item_t *lp; | |
1679 | ||
f9114eba DC |
1680 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) |
1681 | ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0); | |
e1cccd91 CH |
1682 | ASSERT(ip0->i_ino != ip1->i_ino); |
1683 | ||
1684 | if (ip0->i_ino > ip1->i_ino) { | |
1685 | temp = ip0; | |
1686 | ip0 = ip1; | |
1687 | ip1 = temp; | |
1688 | } | |
1689 | ||
1690 | again: | |
1691 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
1692 | ||
1693 | /* | |
1694 | * If the first lock we have locked is in the AIL, we must TRY to get | |
1695 | * the second lock. If we can't get it, we must release the first one | |
1696 | * and try again. | |
1697 | */ | |
1698 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
1699 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1700 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
1701 | xfs_iunlock(ip0, lock_mode); | |
1702 | if ((++attempts % 5) == 0) | |
1703 | delay(1); /* Don't just spin the CPU */ | |
1704 | goto again; | |
1705 | } | |
1706 | } else { | |
1707 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
1708 | } | |
1709 | } | |
1710 | ||
993386c1 | 1711 | int |
1da177e4 | 1712 | xfs_remove( |
993386c1 | 1713 | xfs_inode_t *dp, |
556b8b16 BN |
1714 | struct xfs_name *name, |
1715 | xfs_inode_t *ip) | |
1da177e4 | 1716 | { |
993386c1 | 1717 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 1718 | xfs_trans_t *tp = NULL; |
8f112e3b | 1719 | int is_dir = S_ISDIR(ip->i_d.di_mode); |
1da177e4 LT |
1720 | int error = 0; |
1721 | xfs_bmap_free_t free_list; | |
1722 | xfs_fsblock_t first_block; | |
1723 | int cancel_flags; | |
1724 | int committed; | |
1da177e4 LT |
1725 | int link_zero; |
1726 | uint resblks; | |
8f112e3b | 1727 | uint log_count; |
1da177e4 | 1728 | |
cf441eeb | 1729 | xfs_itrace_entry(dp); |
8f112e3b | 1730 | xfs_itrace_entry(ip); |
1da177e4 | 1731 | |
1da177e4 LT |
1732 | if (XFS_FORCED_SHUTDOWN(mp)) |
1733 | return XFS_ERROR(EIO); | |
1734 | ||
eb9df39d | 1735 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
556b8b16 BN |
1736 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL, |
1737 | NULL, DM_RIGHT_NULL, name->name, NULL, | |
1738 | ip->i_d.di_mode, 0, 0); | |
1da177e4 LT |
1739 | if (error) |
1740 | return error; | |
1741 | } | |
1742 | ||
7d095257 | 1743 | error = xfs_qm_dqattach(dp, 0); |
8f112e3b CH |
1744 | if (error) |
1745 | goto std_return; | |
1746 | ||
7d095257 | 1747 | error = xfs_qm_dqattach(ip, 0); |
8f112e3b | 1748 | if (error) |
1da177e4 | 1749 | goto std_return; |
1da177e4 | 1750 | |
8f112e3b CH |
1751 | if (is_dir) { |
1752 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
1753 | log_count = XFS_DEFAULT_LOG_COUNT; | |
1754 | } else { | |
1755 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
1756 | log_count = XFS_REMOVE_LOG_COUNT; | |
1757 | } | |
1da177e4 | 1758 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
8f112e3b | 1759 | |
1da177e4 LT |
1760 | /* |
1761 | * We try to get the real space reservation first, | |
1762 | * allowing for directory btree deletion(s) implying | |
1763 | * possible bmap insert(s). If we can't get the space | |
1764 | * reservation then we use 0 instead, and avoid the bmap | |
1765 | * btree insert(s) in the directory code by, if the bmap | |
1766 | * insert tries to happen, instead trimming the LAST | |
1767 | * block from the directory. | |
1768 | */ | |
1769 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
1770 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1771 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1772 | if (error == ENOSPC) { |
1773 | resblks = 0; | |
1774 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1775 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1776 | } |
1777 | if (error) { | |
1778 | ASSERT(error != ENOSPC); | |
8f112e3b CH |
1779 | cancel_flags = 0; |
1780 | goto out_trans_cancel; | |
1da177e4 LT |
1781 | } |
1782 | ||
e1cccd91 | 1783 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1784 | |
1785 | /* | |
1786 | * At this point, we've gotten both the directory and the entry | |
1787 | * inodes locked. | |
1788 | */ | |
5df78e73 | 1789 | IHOLD(ip); |
1da177e4 | 1790 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
82dab941 CH |
1791 | |
1792 | IHOLD(dp); | |
1793 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1da177e4 | 1794 | |
8f112e3b CH |
1795 | /* |
1796 | * If we're removing a directory perform some additional validation. | |
1797 | */ | |
1798 | if (is_dir) { | |
1799 | ASSERT(ip->i_d.di_nlink >= 2); | |
1800 | if (ip->i_d.di_nlink != 2) { | |
1801 | error = XFS_ERROR(ENOTEMPTY); | |
1802 | goto out_trans_cancel; | |
1803 | } | |
1804 | if (!xfs_dir_isempty(ip)) { | |
1805 | error = XFS_ERROR(ENOTEMPTY); | |
1806 | goto out_trans_cancel; | |
1807 | } | |
1808 | } | |
1809 | ||
9d87c319 | 1810 | xfs_bmap_init(&free_list, &first_block); |
556b8b16 | 1811 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
d4377d84 | 1812 | &first_block, &free_list, resblks); |
1da177e4 LT |
1813 | if (error) { |
1814 | ASSERT(error != ENOENT); | |
8f112e3b | 1815 | goto out_bmap_cancel; |
1da177e4 LT |
1816 | } |
1817 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1818 | ||
8f112e3b CH |
1819 | if (is_dir) { |
1820 | /* | |
1821 | * Drop the link from ip's "..". | |
1822 | */ | |
1823 | error = xfs_droplink(tp, dp); | |
1824 | if (error) | |
1825 | goto out_bmap_cancel; | |
1826 | ||
1827 | /* | |
2b7035fd | 1828 | * Drop the "." link from ip to self. |
8f112e3b CH |
1829 | */ |
1830 | error = xfs_droplink(tp, ip); | |
1831 | if (error) | |
1832 | goto out_bmap_cancel; | |
1833 | } else { | |
1834 | /* | |
1835 | * When removing a non-directory we need to log the parent | |
26c52951 DC |
1836 | * inode here. For a directory this is done implicitly |
1837 | * by the xfs_droplink call for the ".." entry. | |
8f112e3b CH |
1838 | */ |
1839 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1da177e4 LT |
1840 | } |
1841 | ||
8f112e3b | 1842 | /* |
2b7035fd | 1843 | * Drop the link from dp to ip. |
8f112e3b CH |
1844 | */ |
1845 | error = xfs_droplink(tp, ip); | |
1846 | if (error) | |
1847 | goto out_bmap_cancel; | |
1848 | ||
1849 | /* | |
1850 | * Determine if this is the last link while | |
1da177e4 LT |
1851 | * we are in the transaction. |
1852 | */ | |
8f112e3b | 1853 | link_zero = (ip->i_d.di_nlink == 0); |
1da177e4 | 1854 | |
1da177e4 LT |
1855 | /* |
1856 | * If this is a synchronous mount, make sure that the | |
1857 | * remove transaction goes to disk before returning to | |
1858 | * the user. | |
1859 | */ | |
8f112e3b | 1860 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 1861 | xfs_trans_set_sync(tp); |
1da177e4 | 1862 | |
f7c99b6f | 1863 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
8f112e3b CH |
1864 | if (error) |
1865 | goto out_bmap_cancel; | |
1da177e4 | 1866 | |
1c72bf90 | 1867 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
5df78e73 | 1868 | if (error) |
1da177e4 | 1869 | goto std_return; |
1da177e4 | 1870 | |
2a82b8be DC |
1871 | /* |
1872 | * If we are using filestreams, kill the stream association. | |
1873 | * If the file is still open it may get a new one but that | |
1874 | * will get killed on last close in xfs_close() so we don't | |
1875 | * have to worry about that. | |
1876 | */ | |
8f112e3b | 1877 | if (!is_dir && link_zero && xfs_inode_is_filestream(ip)) |
2a82b8be DC |
1878 | xfs_filestream_deassociate(ip); |
1879 | ||
1da177e4 | 1880 | std_return: |
eb9df39d | 1881 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
8f112e3b CH |
1882 | XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, dp, DM_RIGHT_NULL, |
1883 | NULL, DM_RIGHT_NULL, name->name, NULL, | |
1884 | ip->i_d.di_mode, error, 0); | |
1da177e4 | 1885 | } |
8f112e3b | 1886 | |
1da177e4 LT |
1887 | return error; |
1888 | ||
8f112e3b | 1889 | out_bmap_cancel: |
1da177e4 LT |
1890 | xfs_bmap_cancel(&free_list); |
1891 | cancel_flags |= XFS_TRANS_ABORT; | |
8f112e3b | 1892 | out_trans_cancel: |
1da177e4 LT |
1893 | xfs_trans_cancel(tp, cancel_flags); |
1894 | goto std_return; | |
1da177e4 LT |
1895 | } |
1896 | ||
993386c1 | 1897 | int |
1da177e4 | 1898 | xfs_link( |
993386c1 | 1899 | xfs_inode_t *tdp, |
a3da7896 | 1900 | xfs_inode_t *sip, |
556b8b16 | 1901 | struct xfs_name *target_name) |
1da177e4 | 1902 | { |
993386c1 | 1903 | xfs_mount_t *mp = tdp->i_mount; |
1da177e4 | 1904 | xfs_trans_t *tp; |
1da177e4 LT |
1905 | int error; |
1906 | xfs_bmap_free_t free_list; | |
1907 | xfs_fsblock_t first_block; | |
1908 | int cancel_flags; | |
1909 | int committed; | |
1da177e4 | 1910 | int resblks; |
1da177e4 | 1911 | |
cf441eeb | 1912 | xfs_itrace_entry(tdp); |
a3da7896 | 1913 | xfs_itrace_entry(sip); |
1da177e4 | 1914 | |
a3da7896 | 1915 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); |
1da177e4 | 1916 | |
1da177e4 LT |
1917 | if (XFS_FORCED_SHUTDOWN(mp)) |
1918 | return XFS_ERROR(EIO); | |
1919 | ||
eb9df39d | 1920 | if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) { |
1da177e4 | 1921 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, |
bc4ac74a CH |
1922 | tdp, DM_RIGHT_NULL, |
1923 | sip, DM_RIGHT_NULL, | |
556b8b16 | 1924 | target_name->name, NULL, 0, 0, 0); |
1da177e4 LT |
1925 | if (error) |
1926 | return error; | |
1927 | } | |
1928 | ||
1929 | /* Return through std_return after this point. */ | |
1930 | ||
7d095257 | 1931 | error = xfs_qm_dqattach(sip, 0); |
cb3f35bb CH |
1932 | if (error) |
1933 | goto std_return; | |
1934 | ||
7d095257 | 1935 | error = xfs_qm_dqattach(tdp, 0); |
1da177e4 LT |
1936 | if (error) |
1937 | goto std_return; | |
1938 | ||
1939 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
1940 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
556b8b16 | 1941 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
1da177e4 LT |
1942 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, |
1943 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
1944 | if (error == ENOSPC) { | |
1945 | resblks = 0; | |
1946 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
1947 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
1948 | } | |
1949 | if (error) { | |
1950 | cancel_flags = 0; | |
1951 | goto error_return; | |
1952 | } | |
1953 | ||
e1cccd91 | 1954 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); |
1da177e4 LT |
1955 | |
1956 | /* | |
1957 | * Increment vnode ref counts since xfs_trans_commit & | |
1958 | * xfs_trans_cancel will both unlock the inodes and | |
1959 | * decrement the associated ref counts. | |
1960 | */ | |
a3da7896 CH |
1961 | IHOLD(sip); |
1962 | IHOLD(tdp); | |
1da177e4 LT |
1963 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); |
1964 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
1965 | ||
1966 | /* | |
1967 | * If the source has too many links, we can't make any more to it. | |
1968 | */ | |
1969 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
1970 | error = XFS_ERROR(EMLINK); | |
1971 | goto error_return; | |
1972 | } | |
1973 | ||
365ca83d NS |
1974 | /* |
1975 | * If we are using project inheritance, we only allow hard link | |
1976 | * creation in our tree when the project IDs are the same; else | |
1977 | * the tree quota mechanism could be circumvented. | |
1978 | */ | |
1979 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
1980 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 1981 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
1982 | goto error_return; |
1983 | } | |
1984 | ||
556b8b16 BN |
1985 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); |
1986 | if (error) | |
1da177e4 LT |
1987 | goto error_return; |
1988 | ||
9d87c319 | 1989 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 1990 | |
556b8b16 BN |
1991 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
1992 | &first_block, &free_list, resblks); | |
1da177e4 LT |
1993 | if (error) |
1994 | goto abort_return; | |
1995 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1da177e4 LT |
1996 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); |
1997 | ||
1998 | error = xfs_bumplink(tp, sip); | |
b71d300c | 1999 | if (error) |
1da177e4 | 2000 | goto abort_return; |
1da177e4 LT |
2001 | |
2002 | /* | |
2003 | * If this is a synchronous mount, make sure that the | |
2004 | * link transaction goes to disk before returning to | |
2005 | * the user. | |
2006 | */ | |
2007 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2008 | xfs_trans_set_sync(tp); | |
2009 | } | |
2010 | ||
f7c99b6f | 2011 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2012 | if (error) { |
2013 | xfs_bmap_cancel(&free_list); | |
2014 | goto abort_return; | |
2015 | } | |
2016 | ||
1c72bf90 | 2017 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
b71d300c | 2018 | if (error) |
1da177e4 | 2019 | goto std_return; |
1da177e4 LT |
2020 | |
2021 | /* Fall through to std_return with error = 0. */ | |
2022 | std_return: | |
eb9df39d | 2023 | if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) { |
1da177e4 | 2024 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, |
bc4ac74a CH |
2025 | tdp, DM_RIGHT_NULL, |
2026 | sip, DM_RIGHT_NULL, | |
556b8b16 | 2027 | target_name->name, NULL, 0, error, 0); |
1da177e4 LT |
2028 | } |
2029 | return error; | |
2030 | ||
2031 | abort_return: | |
2032 | cancel_flags |= XFS_TRANS_ABORT; | |
2033 | /* FALLTHROUGH */ | |
014c2544 | 2034 | |
1da177e4 LT |
2035 | error_return: |
2036 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2037 | goto std_return; |
2038 | } | |
b71d300c | 2039 | |
993386c1 | 2040 | int |
1da177e4 | 2041 | xfs_symlink( |
993386c1 | 2042 | xfs_inode_t *dp, |
556b8b16 BN |
2043 | struct xfs_name *link_name, |
2044 | const char *target_path, | |
3e5daf05 | 2045 | mode_t mode, |
3937be5b | 2046 | xfs_inode_t **ipp, |
1da177e4 LT |
2047 | cred_t *credp) |
2048 | { | |
993386c1 | 2049 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 2050 | xfs_trans_t *tp; |
1da177e4 LT |
2051 | xfs_inode_t *ip; |
2052 | int error; | |
2053 | int pathlen; | |
2054 | xfs_bmap_free_t free_list; | |
2055 | xfs_fsblock_t first_block; | |
993386c1 | 2056 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2057 | uint cancel_flags; |
2058 | int committed; | |
2059 | xfs_fileoff_t first_fsb; | |
2060 | xfs_filblks_t fs_blocks; | |
2061 | int nmaps; | |
2062 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
2063 | xfs_daddr_t d; | |
556b8b16 | 2064 | const char *cur_chunk; |
1da177e4 LT |
2065 | int byte_cnt; |
2066 | int n; | |
2067 | xfs_buf_t *bp; | |
2068 | xfs_prid_t prid; | |
2069 | struct xfs_dquot *udqp, *gdqp; | |
2070 | uint resblks; | |
1da177e4 | 2071 | |
3937be5b | 2072 | *ipp = NULL; |
1da177e4 LT |
2073 | error = 0; |
2074 | ip = NULL; | |
2075 | tp = NULL; | |
2076 | ||
cf441eeb | 2077 | xfs_itrace_entry(dp); |
1da177e4 LT |
2078 | |
2079 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2080 | return XFS_ERROR(EIO); | |
2081 | ||
1da177e4 LT |
2082 | /* |
2083 | * Check component lengths of the target path name. | |
2084 | */ | |
2085 | pathlen = strlen(target_path); | |
2086 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
2087 | return XFS_ERROR(ENAMETOOLONG); | |
1da177e4 | 2088 | |
eb9df39d | 2089 | if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) { |
bc4ac74a | 2090 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp, |
1da177e4 | 2091 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, |
046ea753 DC |
2092 | link_name->name, |
2093 | (unsigned char *)target_path, 0, 0, 0); | |
1da177e4 LT |
2094 | if (error) |
2095 | return error; | |
2096 | } | |
2097 | ||
2098 | /* Return through std_return after this point. */ | |
2099 | ||
2100 | udqp = gdqp = NULL; | |
365ca83d NS |
2101 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2102 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
2103 | else |
2104 | prid = (xfs_prid_t)dfltprid; | |
2105 | ||
2106 | /* | |
2107 | * Make sure that we have allocated dquot(s) on disk. | |
2108 | */ | |
7d095257 | 2109 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
1da177e4 LT |
2110 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2111 | if (error) | |
2112 | goto std_return; | |
2113 | ||
2114 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
2115 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2116 | /* | |
2117 | * The symlink will fit into the inode data fork? | |
2118 | * There can't be any attributes so we get the whole variable part. | |
2119 | */ | |
2120 | if (pathlen <= XFS_LITINO(mp)) | |
2121 | fs_blocks = 0; | |
2122 | else | |
2123 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
556b8b16 | 2124 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks); |
1da177e4 LT |
2125 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, |
2126 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
2127 | if (error == ENOSPC && fs_blocks == 0) { | |
2128 | resblks = 0; | |
2129 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
2130 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
2131 | } | |
2132 | if (error) { | |
2133 | cancel_flags = 0; | |
1da177e4 LT |
2134 | goto error_return; |
2135 | } | |
2136 | ||
f7c66ce3 | 2137 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2138 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2139 | |
2140 | /* | |
2141 | * Check whether the directory allows new symlinks or not. | |
2142 | */ | |
2143 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
2144 | error = XFS_ERROR(EPERM); | |
2145 | goto error_return; | |
2146 | } | |
2147 | ||
2148 | /* | |
2149 | * Reserve disk quota : blocks and inode. | |
2150 | */ | |
7d095257 | 2151 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 LT |
2152 | if (error) |
2153 | goto error_return; | |
2154 | ||
2155 | /* | |
2156 | * Check for ability to enter directory entry, if no space reserved. | |
2157 | */ | |
556b8b16 BN |
2158 | error = xfs_dir_canenter(tp, dp, link_name, resblks); |
2159 | if (error) | |
1da177e4 LT |
2160 | goto error_return; |
2161 | /* | |
2162 | * Initialize the bmap freelist prior to calling either | |
2163 | * bmapi or the directory create code. | |
2164 | */ | |
9d87c319 | 2165 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
2166 | |
2167 | /* | |
2168 | * Allocate an inode for the symlink. | |
2169 | */ | |
3e5daf05 | 2170 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), |
1da177e4 LT |
2171 | 1, 0, credp, prid, resblks > 0, &ip, NULL); |
2172 | if (error) { | |
2173 | if (error == ENOSPC) | |
2174 | goto error_return; | |
2175 | goto error1; | |
2176 | } | |
1da177e4 | 2177 | |
993386c1 CH |
2178 | /* |
2179 | * An error after we've joined dp to the transaction will result in the | |
2180 | * transaction cancel unlocking dp so don't do it explicitly in the | |
2181 | * error path. | |
2182 | */ | |
3937be5b | 2183 | IHOLD(dp); |
1da177e4 | 2184 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 2185 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2186 | |
2187 | /* | |
2188 | * Also attach the dquot(s) to it, if applicable. | |
2189 | */ | |
7d095257 | 2190 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 LT |
2191 | |
2192 | if (resblks) | |
2193 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
2194 | /* | |
2195 | * If the symlink will fit into the inode, write it inline. | |
2196 | */ | |
2197 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
2198 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
2199 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
2200 | ip->i_d.di_size = pathlen; | |
2201 | ||
2202 | /* | |
2203 | * The inode was initially created in extent format. | |
2204 | */ | |
2205 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
2206 | ip->i_df.if_flags |= XFS_IFINLINE; | |
2207 | ||
2208 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
2209 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
2210 | ||
2211 | } else { | |
2212 | first_fsb = 0; | |
2213 | nmaps = SYMLINK_MAPS; | |
2214 | ||
2215 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
2216 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
2217 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 2218 | &free_list, NULL); |
1da177e4 LT |
2219 | if (error) { |
2220 | goto error1; | |
2221 | } | |
2222 | ||
2223 | if (resblks) | |
2224 | resblks -= fs_blocks; | |
2225 | ip->i_d.di_size = pathlen; | |
2226 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2227 | ||
2228 | cur_chunk = target_path; | |
2229 | for (n = 0; n < nmaps; n++) { | |
2230 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
2231 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
2232 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
2233 | BTOBB(byte_cnt), 0); | |
2234 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
2235 | if (pathlen < byte_cnt) { | |
2236 | byte_cnt = pathlen; | |
2237 | } | |
2238 | pathlen -= byte_cnt; | |
2239 | ||
2240 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
2241 | cur_chunk += byte_cnt; | |
2242 | ||
2243 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
2244 | } | |
2245 | } | |
2246 | ||
2247 | /* | |
2248 | * Create the directory entry for the symlink. | |
2249 | */ | |
556b8b16 BN |
2250 | error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, |
2251 | &first_block, &free_list, resblks); | |
f6c2d1fa | 2252 | if (error) |
1da177e4 | 2253 | goto error1; |
1da177e4 LT |
2254 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
2255 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2256 | ||
1da177e4 LT |
2257 | /* |
2258 | * If this is a synchronous mount, make sure that the | |
2259 | * symlink transaction goes to disk before returning to | |
2260 | * the user. | |
2261 | */ | |
2262 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2263 | xfs_trans_set_sync(tp); | |
2264 | } | |
2265 | ||
2266 | /* | |
2267 | * xfs_trans_commit normally decrements the vnode ref count | |
2268 | * when it unlocks the inode. Since we want to return the | |
2269 | * vnode to the caller, we bump the vnode ref count now. | |
2270 | */ | |
2271 | IHOLD(ip); | |
2272 | ||
f7c99b6f | 2273 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2274 | if (error) { |
2275 | goto error2; | |
2276 | } | |
1c72bf90 | 2277 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
7d095257 CH |
2278 | xfs_qm_dqrele(udqp); |
2279 | xfs_qm_dqrele(gdqp); | |
1da177e4 LT |
2280 | |
2281 | /* Fall through to std_return with error = 0 or errno from | |
2282 | * xfs_trans_commit */ | |
2283 | std_return: | |
993386c1 | 2284 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) { |
1da177e4 | 2285 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, |
bc4ac74a CH |
2286 | dp, DM_RIGHT_NULL, |
2287 | error ? NULL : ip, | |
556b8b16 | 2288 | DM_RIGHT_NULL, link_name->name, |
046ea753 DC |
2289 | (unsigned char *)target_path, |
2290 | 0, error, 0); | |
1da177e4 LT |
2291 | } |
2292 | ||
3937be5b CH |
2293 | if (!error) |
2294 | *ipp = ip; | |
1da177e4 LT |
2295 | return error; |
2296 | ||
2297 | error2: | |
2298 | IRELE(ip); | |
2299 | error1: | |
2300 | xfs_bmap_cancel(&free_list); | |
2301 | cancel_flags |= XFS_TRANS_ABORT; | |
2302 | error_return: | |
2303 | xfs_trans_cancel(tp, cancel_flags); | |
7d095257 CH |
2304 | xfs_qm_dqrele(udqp); |
2305 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 2306 | |
993386c1 | 2307 | if (unlock_dp_on_error) |
1da177e4 | 2308 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2309 | |
2310 | goto std_return; | |
2311 | } | |
2312 | ||
1da177e4 | 2313 | int |
993386c1 CH |
2314 | xfs_set_dmattrs( |
2315 | xfs_inode_t *ip, | |
1da177e4 | 2316 | u_int evmask, |
993386c1 | 2317 | u_int16_t state) |
1da177e4 | 2318 | { |
993386c1 | 2319 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 2320 | xfs_trans_t *tp; |
1da177e4 LT |
2321 | int error; |
2322 | ||
2323 | if (!capable(CAP_SYS_ADMIN)) | |
2324 | return XFS_ERROR(EPERM); | |
2325 | ||
1da177e4 LT |
2326 | if (XFS_FORCED_SHUTDOWN(mp)) |
2327 | return XFS_ERROR(EIO); | |
2328 | ||
2329 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
2330 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
2331 | if (error) { | |
2332 | xfs_trans_cancel(tp, 0); | |
2333 | return error; | |
2334 | } | |
2335 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2336 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2337 | ||
613d7043 CH |
2338 | ip->i_d.di_dmevmask = evmask; |
2339 | ip->i_d.di_dmstate = state; | |
1da177e4 LT |
2340 | |
2341 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2342 | IHOLD(ip); | |
1c72bf90 | 2343 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
2344 | |
2345 | return error; | |
2346 | } | |
2347 | ||
1da177e4 LT |
2348 | /* |
2349 | * xfs_alloc_file_space() | |
2350 | * This routine allocates disk space for the given file. | |
2351 | * | |
2352 | * If alloc_type == 0, this request is for an ALLOCSP type | |
2353 | * request which will change the file size. In this case, no | |
2354 | * DMAPI event will be generated by the call. A TRUNCATE event | |
2355 | * will be generated later by xfs_setattr. | |
2356 | * | |
2357 | * If alloc_type != 0, this request is for a RESVSP type | |
2358 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
2359 | * lower block boundary byte address is less than the file's | |
2360 | * length. | |
2361 | * | |
2362 | * RETURNS: | |
2363 | * 0 on success | |
2364 | * errno on error | |
2365 | * | |
2366 | */ | |
ba0f32d4 | 2367 | STATIC int |
1da177e4 LT |
2368 | xfs_alloc_file_space( |
2369 | xfs_inode_t *ip, | |
2370 | xfs_off_t offset, | |
2371 | xfs_off_t len, | |
2372 | int alloc_type, | |
2373 | int attr_flags) | |
2374 | { | |
dd9f438e NS |
2375 | xfs_mount_t *mp = ip->i_mount; |
2376 | xfs_off_t count; | |
1da177e4 LT |
2377 | xfs_filblks_t allocated_fsb; |
2378 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
2379 | xfs_extlen_t extsz, temp; |
2380 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 2381 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
2382 | int nimaps; |
2383 | int bmapi_flag; | |
2384 | int quota_flag; | |
1da177e4 | 2385 | int rt; |
1da177e4 | 2386 | xfs_trans_t *tp; |
dd9f438e NS |
2387 | xfs_bmbt_irec_t imaps[1], *imapp; |
2388 | xfs_bmap_free_t free_list; | |
2389 | uint qblocks, resblks, resrtextents; | |
2390 | int committed; | |
2391 | int error; | |
1da177e4 | 2392 | |
cf441eeb | 2393 | xfs_itrace_entry(ip); |
1da177e4 LT |
2394 | |
2395 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2396 | return XFS_ERROR(EIO); | |
2397 | ||
7d095257 CH |
2398 | error = xfs_qm_dqattach(ip, 0); |
2399 | if (error) | |
1da177e4 LT |
2400 | return error; |
2401 | ||
2402 | if (len <= 0) | |
2403 | return XFS_ERROR(EINVAL); | |
2404 | ||
957d0ebe DC |
2405 | rt = XFS_IS_REALTIME_INODE(ip); |
2406 | extsz = xfs_get_extsz_hint(ip); | |
2407 | ||
1da177e4 | 2408 | count = len; |
1da177e4 | 2409 | imapp = &imaps[0]; |
dd9f438e NS |
2410 | nimaps = 1; |
2411 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
2412 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
2413 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
2414 | ||
2415 | /* Generate a DMAPI event if needed. */ | |
ba87ea69 | 2416 | if (alloc_type != 0 && offset < ip->i_size && |
0f285c8a | 2417 | (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2418 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
1da177e4 LT |
2419 | xfs_off_t end_dmi_offset; |
2420 | ||
2421 | end_dmi_offset = offset+len; | |
ba87ea69 LM |
2422 | if (end_dmi_offset > ip->i_size) |
2423 | end_dmi_offset = ip->i_size; | |
bc4ac74a CH |
2424 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset, |
2425 | end_dmi_offset - offset, 0, NULL); | |
1da177e4 | 2426 | if (error) |
014c2544 | 2427 | return error; |
1da177e4 LT |
2428 | } |
2429 | ||
2430 | /* | |
dd9f438e | 2431 | * Allocate file space until done or until there is an error |
1da177e4 LT |
2432 | */ |
2433 | retry: | |
2434 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
2435 | xfs_fileoff_t s, e; |
2436 | ||
1da177e4 | 2437 | /* |
3ddb8fa9 | 2438 | * Determine space reservations for data/realtime. |
1da177e4 | 2439 | */ |
dd9f438e | 2440 | if (unlikely(extsz)) { |
1da177e4 | 2441 | s = startoffset_fsb; |
dd9f438e NS |
2442 | do_div(s, extsz); |
2443 | s *= extsz; | |
2444 | e = startoffset_fsb + allocatesize_fsb; | |
2445 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
2446 | e += temp; | |
2447 | if ((temp = do_mod(e, extsz))) | |
2448 | e += extsz - temp; | |
2449 | } else { | |
2450 | s = 0; | |
2451 | e = allocatesize_fsb; | |
2452 | } | |
2453 | ||
2454 | if (unlikely(rt)) { | |
2455 | resrtextents = qblocks = (uint)(e - s); | |
2456 | resrtextents /= mp->m_sb.sb_rextsize; | |
2457 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
2458 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 2459 | } else { |
dd9f438e NS |
2460 | resrtextents = 0; |
2461 | resblks = qblocks = \ | |
2462 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
2463 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
2464 | } |
2465 | ||
2466 | /* | |
dd9f438e | 2467 | * Allocate and setup the transaction. |
1da177e4 LT |
2468 | */ |
2469 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
2470 | error = xfs_trans_reserve(tp, resblks, |
2471 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
2472 | XFS_TRANS_PERM_LOG_RES, |
2473 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 2474 | /* |
dd9f438e | 2475 | * Check for running out of space |
1da177e4 LT |
2476 | */ |
2477 | if (error) { | |
2478 | /* | |
2479 | * Free the transaction structure. | |
2480 | */ | |
2481 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
2482 | xfs_trans_cancel(tp, 0); | |
2483 | break; | |
2484 | } | |
2485 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
2486 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, |
2487 | 0, quota_flag); | |
1da177e4 LT |
2488 | if (error) |
2489 | goto error1; | |
2490 | ||
2491 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2492 | xfs_trans_ihold(tp, ip); | |
2493 | ||
2494 | /* | |
dd9f438e | 2495 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 | 2496 | */ |
9d87c319 | 2497 | xfs_bmap_init(&free_list, &firstfsb); |
541d7d3c | 2498 | error = xfs_bmapi(tp, ip, startoffset_fsb, |
dd9f438e NS |
2499 | allocatesize_fsb, bmapi_flag, |
2500 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 2501 | &free_list, NULL); |
1da177e4 LT |
2502 | if (error) { |
2503 | goto error0; | |
2504 | } | |
2505 | ||
2506 | /* | |
dd9f438e | 2507 | * Complete the transaction |
1da177e4 | 2508 | */ |
f7c99b6f | 2509 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2510 | if (error) { |
2511 | goto error0; | |
2512 | } | |
2513 | ||
1c72bf90 | 2514 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2515 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2516 | if (error) { | |
2517 | break; | |
2518 | } | |
2519 | ||
2520 | allocated_fsb = imapp->br_blockcount; | |
2521 | ||
dd9f438e | 2522 | if (nimaps == 0) { |
1da177e4 LT |
2523 | error = XFS_ERROR(ENOSPC); |
2524 | break; | |
2525 | } | |
2526 | ||
2527 | startoffset_fsb += allocated_fsb; | |
2528 | allocatesize_fsb -= allocated_fsb; | |
2529 | } | |
2530 | dmapi_enospc_check: | |
0f285c8a | 2531 | if (error == ENOSPC && (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2532 | DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) { |
1da177e4 | 2533 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, |
bc4ac74a CH |
2534 | ip, DM_RIGHT_NULL, |
2535 | ip, DM_RIGHT_NULL, | |
1da177e4 LT |
2536 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ |
2537 | if (error == 0) | |
2538 | goto retry; /* Maybe DMAPI app. has made space */ | |
2539 | /* else fall through with error from XFS_SEND_DATA */ | |
2540 | } | |
2541 | ||
2542 | return error; | |
2543 | ||
dd9f438e | 2544 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 2545 | xfs_bmap_cancel(&free_list); |
7d095257 | 2546 | xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e NS |
2547 | |
2548 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
2549 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
2550 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2551 | goto dmapi_enospc_check; | |
2552 | } | |
2553 | ||
2554 | /* | |
2555 | * Zero file bytes between startoff and endoff inclusive. | |
2556 | * The iolock is held exclusive and no blocks are buffered. | |
2fd6f6ec LM |
2557 | * |
2558 | * This function is used by xfs_free_file_space() to zero | |
2559 | * partial blocks when the range to free is not block aligned. | |
2560 | * When unreserving space with boundaries that are not block | |
2561 | * aligned we round up the start and round down the end | |
2562 | * boundaries and then use this function to zero the parts of | |
2563 | * the blocks that got dropped during the rounding. | |
1da177e4 LT |
2564 | */ |
2565 | STATIC int | |
2566 | xfs_zero_remaining_bytes( | |
2567 | xfs_inode_t *ip, | |
2568 | xfs_off_t startoff, | |
2569 | xfs_off_t endoff) | |
2570 | { | |
2571 | xfs_bmbt_irec_t imap; | |
2572 | xfs_fileoff_t offset_fsb; | |
2573 | xfs_off_t lastoffset; | |
2574 | xfs_off_t offset; | |
2575 | xfs_buf_t *bp; | |
2576 | xfs_mount_t *mp = ip->i_mount; | |
2577 | int nimap; | |
2578 | int error = 0; | |
2579 | ||
2fd6f6ec LM |
2580 | /* |
2581 | * Avoid doing I/O beyond eof - it's not necessary | |
2582 | * since nothing can read beyond eof. The space will | |
2583 | * be zeroed when the file is extended anyway. | |
2584 | */ | |
2585 | if (startoff >= ip->i_size) | |
2586 | return 0; | |
2587 | ||
2588 | if (endoff > ip->i_size) | |
2589 | endoff = ip->i_size; | |
2590 | ||
1da177e4 | 2591 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, |
71ddabb9 | 2592 | XFS_IS_REALTIME_INODE(ip) ? |
1da177e4 | 2593 | mp->m_rtdev_targp : mp->m_ddev_targp); |
c6422617 LM |
2594 | if (!bp) |
2595 | return XFS_ERROR(ENOMEM); | |
1da177e4 LT |
2596 | |
2597 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
2598 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
2599 | nimap = 1; | |
541d7d3c | 2600 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, |
3e57ecf6 | 2601 | NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2602 | if (error || nimap < 1) |
2603 | break; | |
2604 | ASSERT(imap.br_blockcount >= 1); | |
2605 | ASSERT(imap.br_startoff == offset_fsb); | |
2606 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
2607 | if (lastoffset > endoff) | |
2608 | lastoffset = endoff; | |
2609 | if (imap.br_startblock == HOLESTARTBLOCK) | |
2610 | continue; | |
2611 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2612 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
2613 | continue; | |
2614 | XFS_BUF_UNDONE(bp); | |
2615 | XFS_BUF_UNWRITE(bp); | |
2616 | XFS_BUF_READ(bp); | |
9d87c319 | 2617 | XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock)); |
1da177e4 | 2618 | xfsbdstrat(mp, bp); |
d64e31a2 DC |
2619 | error = xfs_iowait(bp); |
2620 | if (error) { | |
1da177e4 LT |
2621 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", |
2622 | mp, bp, XFS_BUF_ADDR(bp)); | |
2623 | break; | |
2624 | } | |
2625 | memset(XFS_BUF_PTR(bp) + | |
2626 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
2627 | 0, lastoffset - offset + 1); | |
2628 | XFS_BUF_UNDONE(bp); | |
2629 | XFS_BUF_UNREAD(bp); | |
2630 | XFS_BUF_WRITE(bp); | |
2631 | xfsbdstrat(mp, bp); | |
d64e31a2 DC |
2632 | error = xfs_iowait(bp); |
2633 | if (error) { | |
1da177e4 LT |
2634 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", |
2635 | mp, bp, XFS_BUF_ADDR(bp)); | |
2636 | break; | |
2637 | } | |
2638 | } | |
2639 | xfs_buf_free(bp); | |
2640 | return error; | |
2641 | } | |
2642 | ||
2643 | /* | |
2644 | * xfs_free_file_space() | |
2645 | * This routine frees disk space for the given file. | |
2646 | * | |
2647 | * This routine is only called by xfs_change_file_space | |
2648 | * for an UNRESVSP type call. | |
2649 | * | |
2650 | * RETURNS: | |
2651 | * 0 on success | |
2652 | * errno on error | |
2653 | * | |
2654 | */ | |
2655 | STATIC int | |
2656 | xfs_free_file_space( | |
2657 | xfs_inode_t *ip, | |
2658 | xfs_off_t offset, | |
2659 | xfs_off_t len, | |
2660 | int attr_flags) | |
2661 | { | |
2662 | int committed; | |
2663 | int done; | |
2664 | xfs_off_t end_dmi_offset; | |
2665 | xfs_fileoff_t endoffset_fsb; | |
2666 | int error; | |
2667 | xfs_fsblock_t firstfsb; | |
2668 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
2669 | xfs_bmbt_irec_t imap; |
2670 | xfs_off_t ioffset; | |
2671 | xfs_extlen_t mod=0; | |
2672 | xfs_mount_t *mp; | |
2673 | int nimap; | |
2674 | uint resblks; | |
673cdf5c | 2675 | uint rounding; |
1da177e4 LT |
2676 | int rt; |
2677 | xfs_fileoff_t startoffset_fsb; | |
2678 | xfs_trans_t *tp; | |
5fcbab35 | 2679 | int need_iolock = 1; |
1da177e4 | 2680 | |
1da177e4 LT |
2681 | mp = ip->i_mount; |
2682 | ||
cf441eeb | 2683 | xfs_itrace_entry(ip); |
bd5a876a | 2684 | |
7d095257 CH |
2685 | error = xfs_qm_dqattach(ip, 0); |
2686 | if (error) | |
1da177e4 LT |
2687 | return error; |
2688 | ||
2689 | error = 0; | |
2690 | if (len <= 0) /* if nothing being freed */ | |
2691 | return error; | |
71ddabb9 | 2692 | rt = XFS_IS_REALTIME_INODE(ip); |
1da177e4 LT |
2693 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); |
2694 | end_dmi_offset = offset + len; | |
2695 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
2696 | ||
0f285c8a | 2697 | if (offset < ip->i_size && (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2698 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
ba87ea69 LM |
2699 | if (end_dmi_offset > ip->i_size) |
2700 | end_dmi_offset = ip->i_size; | |
bc4ac74a | 2701 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, |
1da177e4 LT |
2702 | offset, end_dmi_offset - offset, |
2703 | AT_DELAY_FLAG(attr_flags), NULL); | |
2704 | if (error) | |
014c2544 | 2705 | return error; |
1da177e4 LT |
2706 | } |
2707 | ||
0f285c8a | 2708 | if (attr_flags & XFS_ATTR_NOLOCK) |
5fcbab35 | 2709 | need_iolock = 0; |
9fa8046f | 2710 | if (need_iolock) { |
1da177e4 | 2711 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
25e41b3d CH |
2712 | /* wait for the completion of any pending DIOs */ |
2713 | xfs_ioend_wait(ip); | |
9fa8046f | 2714 | } |
5fcbab35 | 2715 | |
e6a4b37f | 2716 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1da177e4 | 2717 | ioffset = offset & ~(rounding - 1); |
bd5a876a | 2718 | |
df80c933 | 2719 | if (VN_CACHED(VFS_I(ip)) != 0) { |
e6a4b37f | 2720 | error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED); |
d3cf2094 LM |
2721 | if (error) |
2722 | goto out_unlock_iolock; | |
bd5a876a CH |
2723 | } |
2724 | ||
1da177e4 LT |
2725 | /* |
2726 | * Need to zero the stuff we're not freeing, on disk. | |
9da096fd | 2727 | * If it's a realtime file & can't use unwritten extents then we |
1da177e4 LT |
2728 | * actually need to zero the extent edges. Otherwise xfs_bunmapi |
2729 | * will take care of it for us. | |
2730 | */ | |
62118709 | 2731 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
1da177e4 | 2732 | nimap = 1; |
541d7d3c | 2733 | error = xfs_bmapi(NULL, ip, startoffset_fsb, |
3e57ecf6 | 2734 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2735 | if (error) |
2736 | goto out_unlock_iolock; | |
2737 | ASSERT(nimap == 0 || nimap == 1); | |
2738 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
2739 | xfs_daddr_t block; | |
2740 | ||
2741 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2742 | block = imap.br_startblock; | |
2743 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
2744 | if (mod) | |
2745 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
2746 | } | |
2747 | nimap = 1; | |
541d7d3c | 2748 | error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, |
3e57ecf6 | 2749 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2750 | if (error) |
2751 | goto out_unlock_iolock; | |
2752 | ASSERT(nimap == 0 || nimap == 1); | |
2753 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
2754 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2755 | mod++; | |
2756 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
2757 | endoffset_fsb -= mod; | |
2758 | } | |
2759 | } | |
2760 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
2761 | /* | |
2762 | * One contiguous piece to clear | |
2763 | */ | |
2764 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
2765 | else { | |
2766 | /* | |
2767 | * Some full blocks, possibly two pieces to clear | |
2768 | */ | |
2769 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
2770 | error = xfs_zero_remaining_bytes(ip, offset, | |
2771 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
2772 | if (!error && | |
2773 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
2774 | error = xfs_zero_remaining_bytes(ip, | |
2775 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
2776 | offset + len - 1); | |
2777 | } | |
2778 | ||
2779 | /* | |
2780 | * free file space until done or until there is an error | |
2781 | */ | |
2782 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
2783 | while (!error && !done) { | |
2784 | ||
2785 | /* | |
91ebecc7 DC |
2786 | * allocate and setup the transaction. Allow this |
2787 | * transaction to dip into the reserve blocks to ensure | |
2788 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
2789 | */ |
2790 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 2791 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
2792 | error = xfs_trans_reserve(tp, |
2793 | resblks, | |
2794 | XFS_WRITE_LOG_RES(mp), | |
2795 | 0, | |
2796 | XFS_TRANS_PERM_LOG_RES, | |
2797 | XFS_WRITE_LOG_COUNT); | |
2798 | ||
2799 | /* | |
2800 | * check for running out of space | |
2801 | */ | |
2802 | if (error) { | |
2803 | /* | |
2804 | * Free the transaction structure. | |
2805 | */ | |
2806 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
2807 | xfs_trans_cancel(tp, 0); | |
2808 | break; | |
2809 | } | |
2810 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
2811 | error = xfs_trans_reserve_quota(tp, mp, |
2812 | ip->i_udquot, ip->i_gdquot, | |
2813 | resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
2814 | if (error) |
2815 | goto error1; | |
2816 | ||
2817 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2818 | xfs_trans_ihold(tp, ip); | |
2819 | ||
2820 | /* | |
2821 | * issue the bunmapi() call to free the blocks | |
2822 | */ | |
9d87c319 | 2823 | xfs_bmap_init(&free_list, &firstfsb); |
541d7d3c | 2824 | error = xfs_bunmapi(tp, ip, startoffset_fsb, |
1da177e4 | 2825 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 2826 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
2827 | if (error) { |
2828 | goto error0; | |
2829 | } | |
2830 | ||
2831 | /* | |
2832 | * complete the transaction | |
2833 | */ | |
f7c99b6f | 2834 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2835 | if (error) { |
2836 | goto error0; | |
2837 | } | |
2838 | ||
1c72bf90 | 2839 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2840 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2841 | } | |
2842 | ||
2843 | out_unlock_iolock: | |
2844 | if (need_iolock) | |
2845 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
2846 | return error; | |
2847 | ||
2848 | error0: | |
2849 | xfs_bmap_cancel(&free_list); | |
2850 | error1: | |
2851 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
2852 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
2853 | XFS_ILOCK_EXCL); | |
2854 | return error; | |
2855 | } | |
2856 | ||
2857 | /* | |
2858 | * xfs_change_file_space() | |
2859 | * This routine allocates or frees disk space for the given file. | |
2860 | * The user specified parameters are checked for alignment and size | |
2861 | * limitations. | |
2862 | * | |
2863 | * RETURNS: | |
2864 | * 0 on success | |
2865 | * errno on error | |
2866 | * | |
2867 | */ | |
2868 | int | |
2869 | xfs_change_file_space( | |
993386c1 | 2870 | xfs_inode_t *ip, |
1da177e4 LT |
2871 | int cmd, |
2872 | xfs_flock64_t *bf, | |
2873 | xfs_off_t offset, | |
1da177e4 LT |
2874 | int attr_flags) |
2875 | { | |
993386c1 | 2876 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
2877 | int clrprealloc; |
2878 | int error; | |
2879 | xfs_fsize_t fsize; | |
1da177e4 LT |
2880 | int setprealloc; |
2881 | xfs_off_t startoffset; | |
2882 | xfs_off_t llen; | |
2883 | xfs_trans_t *tp; | |
0f285c8a | 2884 | struct iattr iattr; |
1da177e4 | 2885 | |
cf441eeb | 2886 | xfs_itrace_entry(ip); |
1da177e4 | 2887 | |
993386c1 | 2888 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
2889 | return XFS_ERROR(EINVAL); |
2890 | ||
1da177e4 LT |
2891 | switch (bf->l_whence) { |
2892 | case 0: /*SEEK_SET*/ | |
2893 | break; | |
2894 | case 1: /*SEEK_CUR*/ | |
2895 | bf->l_start += offset; | |
2896 | break; | |
2897 | case 2: /*SEEK_END*/ | |
ba87ea69 | 2898 | bf->l_start += ip->i_size; |
1da177e4 LT |
2899 | break; |
2900 | default: | |
2901 | return XFS_ERROR(EINVAL); | |
2902 | } | |
2903 | ||
2904 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
2905 | ||
2906 | if ( (bf->l_start < 0) | |
2907 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
2908 | || (bf->l_start + llen < 0) | |
2909 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
2910 | return XFS_ERROR(EINVAL); | |
2911 | ||
2912 | bf->l_whence = 0; | |
2913 | ||
2914 | startoffset = bf->l_start; | |
ba87ea69 | 2915 | fsize = ip->i_size; |
1da177e4 LT |
2916 | |
2917 | /* | |
2918 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
2919 | * file space. | |
2920 | * These calls do NOT zero the data space allocated to the file, | |
2921 | * nor do they change the file size. | |
2922 | * | |
2923 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
2924 | * space. | |
2925 | * These calls cause the new file data to be zeroed and the file | |
2926 | * size to be changed. | |
2927 | */ | |
2928 | setprealloc = clrprealloc = 0; | |
2929 | ||
2930 | switch (cmd) { | |
2931 | case XFS_IOC_RESVSP: | |
2932 | case XFS_IOC_RESVSP64: | |
2933 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
2934 | 1, attr_flags); | |
2935 | if (error) | |
2936 | return error; | |
2937 | setprealloc = 1; | |
2938 | break; | |
2939 | ||
2940 | case XFS_IOC_UNRESVSP: | |
2941 | case XFS_IOC_UNRESVSP64: | |
2942 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
2943 | attr_flags))) | |
2944 | return error; | |
2945 | break; | |
2946 | ||
2947 | case XFS_IOC_ALLOCSP: | |
2948 | case XFS_IOC_ALLOCSP64: | |
2949 | case XFS_IOC_FREESP: | |
2950 | case XFS_IOC_FREESP64: | |
2951 | if (startoffset > fsize) { | |
2952 | error = xfs_alloc_file_space(ip, fsize, | |
2953 | startoffset - fsize, 0, attr_flags); | |
2954 | if (error) | |
2955 | break; | |
2956 | } | |
2957 | ||
0f285c8a CH |
2958 | iattr.ia_valid = ATTR_SIZE; |
2959 | iattr.ia_size = startoffset; | |
1da177e4 | 2960 | |
ea5a3dc8 | 2961 | error = xfs_setattr(ip, &iattr, attr_flags); |
1da177e4 LT |
2962 | |
2963 | if (error) | |
2964 | return error; | |
2965 | ||
2966 | clrprealloc = 1; | |
2967 | break; | |
2968 | ||
2969 | default: | |
2970 | ASSERT(0); | |
2971 | return XFS_ERROR(EINVAL); | |
2972 | } | |
2973 | ||
2974 | /* | |
2975 | * update the inode timestamp, mode, and prealloc flag bits | |
2976 | */ | |
2977 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
2978 | ||
2979 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
2980 | 0, 0, 0))) { | |
2981 | /* ASSERT(0); */ | |
2982 | xfs_trans_cancel(tp, 0); | |
2983 | return error; | |
2984 | } | |
2985 | ||
2986 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2987 | ||
2988 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2989 | xfs_trans_ihold(tp, ip); | |
2990 | ||
0f285c8a | 2991 | if ((attr_flags & XFS_ATTR_DMI) == 0) { |
1da177e4 LT |
2992 | ip->i_d.di_mode &= ~S_ISUID; |
2993 | ||
2994 | /* | |
2995 | * Note that we don't have to worry about mandatory | |
2996 | * file locking being disabled here because we only | |
2997 | * clear the S_ISGID bit if the Group execute bit is | |
2998 | * on, but if it was on then mandatory locking wouldn't | |
2999 | * have been enabled. | |
3000 | */ | |
3001 | if (ip->i_d.di_mode & S_IXGRP) | |
3002 | ip->i_d.di_mode &= ~S_ISGID; | |
3003 | ||
3004 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3005 | } | |
3006 | if (setprealloc) | |
3007 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
3008 | else if (clrprealloc) | |
3009 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
3010 | ||
3011 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3012 | xfs_trans_set_sync(tp); | |
3013 | ||
1c72bf90 | 3014 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
3015 | |
3016 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3017 | ||
3018 | return error; | |
3019 | } |