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" |
1da177e4 | 56 | |
1da177e4 LT |
57 | int |
58 | xfs_setattr( | |
0f285c8a CH |
59 | struct xfs_inode *ip, |
60 | struct iattr *iattr, | |
ea5a3dc8 | 61 | int flags) |
1da177e4 | 62 | { |
993386c1 | 63 | xfs_mount_t *mp = ip->i_mount; |
e4f75291 | 64 | struct inode *inode = VFS_I(ip); |
0f285c8a | 65 | int mask = iattr->ia_valid; |
1da177e4 | 66 | xfs_trans_t *tp; |
1da177e4 LT |
67 | int code; |
68 | uint lock_flags; | |
69 | uint commit_flags=0; | |
70 | uid_t uid=0, iuid=0; | |
71 | gid_t gid=0, igid=0; | |
72 | int timeflags = 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 CH |
136 | if (!(mask & ATTR_SIZE)) { |
137 | if ((mask != (ATTR_CTIME|ATTR_ATIME|ATTR_MTIME)) || | |
1da177e4 LT |
138 | (mp->m_flags & XFS_MOUNT_WSYNC)) { |
139 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); | |
140 | commit_flags = 0; | |
141 | if ((code = xfs_trans_reserve(tp, 0, | |
142 | XFS_ICHANGE_LOG_RES(mp), 0, | |
143 | 0, 0))) { | |
144 | lock_flags = 0; | |
145 | goto error_return; | |
146 | } | |
147 | } | |
148 | } else { | |
eb9df39d | 149 | if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) && |
0f285c8a | 150 | !(flags & XFS_ATTR_DMI)) { |
1da177e4 | 151 | int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR; |
bc4ac74a | 152 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip, |
0f285c8a | 153 | iattr->ia_size, 0, dmflags, NULL); |
1da177e4 LT |
154 | if (code) { |
155 | lock_flags = 0; | |
156 | goto error_return; | |
157 | } | |
158 | } | |
159 | if (need_iolock) | |
160 | lock_flags |= XFS_IOLOCK_EXCL; | |
161 | } | |
162 | ||
163 | xfs_ilock(ip, lock_flags); | |
164 | ||
1da177e4 LT |
165 | /* |
166 | * Change file ownership. Must be the owner or privileged. | |
1da177e4 | 167 | */ |
0f285c8a | 168 | if (mask & (ATTR_UID|ATTR_GID)) { |
1da177e4 LT |
169 | /* |
170 | * These IDs could have changed since we last looked at them. | |
171 | * But, we're assured that if the ownership did change | |
172 | * while we didn't have the inode locked, inode's dquot(s) | |
173 | * would have changed also. | |
174 | */ | |
175 | iuid = ip->i_d.di_uid; | |
1da177e4 | 176 | igid = ip->i_d.di_gid; |
0f285c8a CH |
177 | gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; |
178 | uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; | |
1da177e4 | 179 | |
1da177e4 | 180 | /* |
25fe55e8 | 181 | * Do a quota reservation only if uid/gid is actually |
1da177e4 LT |
182 | * going to change. |
183 | */ | |
7d095257 CH |
184 | if (XFS_IS_QUOTA_RUNNING(mp) && |
185 | ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || | |
186 | (XFS_IS_GQUOTA_ON(mp) && igid != gid))) { | |
1da177e4 | 187 | ASSERT(tp); |
7d095257 | 188 | code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp, |
1da177e4 LT |
189 | capable(CAP_FOWNER) ? |
190 | XFS_QMOPT_FORCE_RES : 0); | |
191 | if (code) /* out of quota */ | |
192 | goto error_return; | |
193 | } | |
194 | } | |
195 | ||
196 | /* | |
197 | * Truncate file. Must have write permission and not be a directory. | |
198 | */ | |
0f285c8a | 199 | if (mask & ATTR_SIZE) { |
1da177e4 | 200 | /* Short circuit the truncate case for zero length files */ |
0f285c8a CH |
201 | if (iattr->ia_size == 0 && |
202 | ip->i_size == 0 && ip->i_d.di_nextents == 0) { | |
1da177e4 LT |
203 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
204 | lock_flags &= ~XFS_ILOCK_EXCL; | |
0f285c8a | 205 | if (mask & ATTR_CTIME) |
1da177e4 LT |
206 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
207 | code = 0; | |
208 | goto error_return; | |
209 | } | |
210 | ||
42173f68 | 211 | if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
212 | code = XFS_ERROR(EISDIR); |
213 | goto error_return; | |
42173f68 | 214 | } else if (!S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
215 | code = XFS_ERROR(EINVAL); |
216 | goto error_return; | |
217 | } | |
c4cd747e | 218 | |
1da177e4 LT |
219 | /* |
220 | * Make sure that the dquots are attached to the inode. | |
221 | */ | |
7d095257 | 222 | code = xfs_qm_dqattach_locked(ip, 0); |
c4cd747e | 223 | if (code) |
1da177e4 | 224 | goto error_return; |
1da177e4 | 225 | |
c4cd747e CH |
226 | /* |
227 | * Now we can make the changes. Before we join the inode | |
228 | * to the transaction, if ATTR_SIZE is set then take care of | |
229 | * the part of the truncation that must be done without the | |
230 | * inode lock. This needs to be done before joining the inode | |
231 | * to the transaction, because the inode cannot be unlocked | |
232 | * once it is a part of the transaction. | |
233 | */ | |
0f285c8a | 234 | if (iattr->ia_size > ip->i_size) { |
61436feb CH |
235 | /* |
236 | * Do the first part of growing a file: zero any data | |
237 | * in the last block that is beyond the old EOF. We | |
238 | * need to do this before the inode is joined to the | |
239 | * transaction to modify the i_size. | |
240 | */ | |
0f285c8a | 241 | code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size); |
374e2ac3 | 242 | } |
1da177e4 | 243 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
c32676ee DC |
244 | |
245 | /* | |
246 | * We are going to log the inode size change in this | |
247 | * transaction so any previous writes that are beyond the on | |
248 | * disk EOF and the new EOF that have not been written out need | |
249 | * to be written here. If we do not write the data out, we | |
250 | * expose ourselves to the null files problem. | |
251 | * | |
252 | * Only flush from the on disk size to the smaller of the in | |
253 | * memory file size or the new size as that's the range we | |
254 | * really care about here and prevents waiting for other data | |
255 | * not within the range we care about here. | |
256 | */ | |
257 | if (!code && | |
0f285c8a CH |
258 | ip->i_size != ip->i_d.di_size && |
259 | iattr->ia_size > ip->i_d.di_size) { | |
739bfb2a | 260 | code = xfs_flush_pages(ip, |
0f285c8a | 261 | ip->i_d.di_size, iattr->ia_size, |
c32676ee DC |
262 | XFS_B_ASYNC, FI_NONE); |
263 | } | |
264 | ||
265 | /* wait for all I/O to complete */ | |
25e41b3d | 266 | xfs_ioend_wait(ip); |
c32676ee | 267 | |
1da177e4 | 268 | if (!code) |
0f285c8a | 269 | code = xfs_itruncate_data(ip, iattr->ia_size); |
1da177e4 LT |
270 | if (code) { |
271 | ASSERT(tp == NULL); | |
272 | lock_flags &= ~XFS_ILOCK_EXCL; | |
273 | ASSERT(lock_flags == XFS_IOLOCK_EXCL); | |
274 | goto error_return; | |
275 | } | |
276 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); | |
277 | if ((code = xfs_trans_reserve(tp, 0, | |
278 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
279 | XFS_TRANS_PERM_LOG_RES, | |
280 | XFS_ITRUNCATE_LOG_COUNT))) { | |
281 | xfs_trans_cancel(tp, 0); | |
282 | if (need_iolock) | |
283 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
284 | return code; | |
285 | } | |
286 | commit_flags = XFS_TRANS_RELEASE_LOG_RES; | |
287 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 288 | |
1da177e4 LT |
289 | xfs_trans_ijoin(tp, ip, lock_flags); |
290 | xfs_trans_ihold(tp, ip); | |
1da177e4 | 291 | |
44d814ce DC |
292 | /* |
293 | * Only change the c/mtime if we are changing the size | |
294 | * or we are explicitly asked to change it. This handles | |
295 | * the semantic difference between truncate() and ftruncate() | |
296 | * as implemented in the VFS. | |
297 | */ | |
0f285c8a | 298 | if (iattr->ia_size != ip->i_size || (mask & ATTR_CTIME)) |
44d814ce DC |
299 | timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; |
300 | ||
0f285c8a CH |
301 | if (iattr->ia_size > ip->i_size) { |
302 | ip->i_d.di_size = iattr->ia_size; | |
303 | ip->i_size = iattr->ia_size; | |
304 | if (!(flags & XFS_ATTR_DMI)) | |
61436feb CH |
305 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG); |
306 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
0f285c8a CH |
307 | } else if (iattr->ia_size <= ip->i_size || |
308 | (iattr->ia_size == 0 && ip->i_d.di_nextents)) { | |
1da177e4 LT |
309 | /* |
310 | * signal a sync transaction unless | |
311 | * we're truncating an already unlinked | |
312 | * file on a wsync filesystem | |
313 | */ | |
0f285c8a | 314 | code = xfs_itruncate_finish(&tp, ip, iattr->ia_size, |
1da177e4 LT |
315 | XFS_DATA_FORK, |
316 | ((ip->i_d.di_nlink != 0 || | |
317 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
318 | ? 1 : 0)); | |
7d4fb40a | 319 | if (code) |
1da177e4 | 320 | goto abort_return; |
7d4fb40a NS |
321 | /* |
322 | * Truncated "down", so we're removing references | |
323 | * to old data here - if we now delay flushing for | |
324 | * a long time, we expose ourselves unduly to the | |
325 | * notorious NULL files problem. So, we mark this | |
326 | * vnode and flush it when the file is closed, and | |
327 | * do not wait the usual (long) time for writeout. | |
328 | */ | |
b3aea4ed | 329 | xfs_iflags_set(ip, XFS_ITRUNCATED); |
1da177e4 | 330 | } |
c4cd747e CH |
331 | } else if (tp) { |
332 | xfs_trans_ijoin(tp, ip, lock_flags); | |
333 | xfs_trans_ihold(tp, ip); | |
1da177e4 LT |
334 | } |
335 | ||
336 | /* | |
337 | * Change file ownership. Must be the owner or privileged. | |
1da177e4 | 338 | */ |
0f285c8a | 339 | if (mask & (ATTR_UID|ATTR_GID)) { |
1da177e4 LT |
340 | /* |
341 | * CAP_FSETID overrides the following restrictions: | |
342 | * | |
343 | * The set-user-ID and set-group-ID bits of a file will be | |
344 | * cleared upon successful return from chown() | |
345 | */ | |
346 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
347 | !capable(CAP_FSETID)) { | |
348 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
349 | } | |
350 | ||
351 | /* | |
352 | * Change the ownerships and register quota modifications | |
353 | * in the transaction. | |
354 | */ | |
355 | if (iuid != uid) { | |
7d095257 | 356 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { |
0f285c8a | 357 | ASSERT(mask & ATTR_UID); |
1da177e4 | 358 | ASSERT(udqp); |
7d095257 | 359 | olddquot1 = xfs_qm_vop_chown(tp, ip, |
1da177e4 LT |
360 | &ip->i_udquot, udqp); |
361 | } | |
362 | ip->i_d.di_uid = uid; | |
f13fae2d | 363 | inode->i_uid = uid; |
1da177e4 LT |
364 | } |
365 | if (igid != gid) { | |
7d095257 | 366 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { |
c8ad20ff | 367 | ASSERT(!XFS_IS_PQUOTA_ON(mp)); |
0f285c8a | 368 | ASSERT(mask & ATTR_GID); |
1da177e4 | 369 | ASSERT(gdqp); |
7d095257 | 370 | olddquot2 = xfs_qm_vop_chown(tp, ip, |
1da177e4 LT |
371 | &ip->i_gdquot, gdqp); |
372 | } | |
373 | ip->i_d.di_gid = gid; | |
f13fae2d | 374 | inode->i_gid = gid; |
1da177e4 | 375 | } |
1da177e4 LT |
376 | |
377 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
378 | timeflags |= XFS_ICHGTIME_CHG; | |
379 | } | |
380 | ||
c4cd747e CH |
381 | /* |
382 | * Change file access modes. | |
383 | */ | |
384 | if (mask & ATTR_MODE) { | |
385 | umode_t mode = iattr->ia_mode; | |
386 | ||
387 | if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) | |
388 | mode &= ~S_ISGID; | |
389 | ||
390 | ip->i_d.di_mode &= S_IFMT; | |
391 | ip->i_d.di_mode |= mode & ~S_IFMT; | |
392 | ||
393 | inode->i_mode &= S_IFMT; | |
394 | inode->i_mode |= mode & ~S_IFMT; | |
395 | ||
396 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
397 | timeflags |= XFS_ICHGTIME_CHG; | |
398 | } | |
1da177e4 LT |
399 | |
400 | /* | |
401 | * Change file access or modified times. | |
402 | */ | |
0f285c8a CH |
403 | if (mask & (ATTR_ATIME|ATTR_MTIME)) { |
404 | if (mask & ATTR_ATIME) { | |
f13fae2d | 405 | inode->i_atime = iattr->ia_atime; |
0f285c8a CH |
406 | ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec; |
407 | ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec; | |
1da177e4 | 408 | ip->i_update_core = 1; |
1da177e4 | 409 | } |
0f285c8a | 410 | if (mask & ATTR_MTIME) { |
f13fae2d | 411 | inode->i_mtime = iattr->ia_mtime; |
0f285c8a CH |
412 | ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec; |
413 | ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec; | |
1da177e4 LT |
414 | timeflags &= ~XFS_ICHGTIME_MOD; |
415 | timeflags |= XFS_ICHGTIME_CHG; | |
416 | } | |
0f285c8a | 417 | if (tp && (mask & (ATTR_MTIME_SET|ATTR_ATIME_SET))) |
1da177e4 LT |
418 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); |
419 | } | |
420 | ||
1da177e4 | 421 | /* |
0f285c8a | 422 | * Change file inode change time only if ATTR_CTIME set |
1da177e4 LT |
423 | * AND we have been called by a DMI function. |
424 | */ | |
425 | ||
0f285c8a | 426 | if ((flags & XFS_ATTR_DMI) && (mask & ATTR_CTIME)) { |
f13fae2d | 427 | inode->i_ctime = iattr->ia_ctime; |
0f285c8a CH |
428 | ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec; |
429 | ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec; | |
1da177e4 LT |
430 | ip->i_update_core = 1; |
431 | timeflags &= ~XFS_ICHGTIME_CHG; | |
432 | } | |
433 | ||
434 | /* | |
435 | * Send out timestamp changes that need to be set to the | |
436 | * current time. Not done when called by a DMI function. | |
437 | */ | |
0f285c8a | 438 | if (timeflags && !(flags & XFS_ATTR_DMI)) |
1da177e4 LT |
439 | xfs_ichgtime(ip, timeflags); |
440 | ||
441 | XFS_STATS_INC(xs_ig_attrchg); | |
442 | ||
443 | /* | |
444 | * If this is a synchronous mount, make sure that the | |
445 | * transaction goes to disk before returning to the user. | |
446 | * This is slightly sub-optimal in that truncates require | |
c41564b5 | 447 | * two sync transactions instead of one for wsync filesystems. |
1da177e4 LT |
448 | * One for the truncate and one for the timestamps since we |
449 | * don't want to change the timestamps unless we're sure the | |
450 | * truncate worked. Truncates are less than 1% of the laddis | |
451 | * mix so this probably isn't worth the trouble to optimize. | |
452 | */ | |
453 | code = 0; | |
454 | if (tp) { | |
455 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
456 | xfs_trans_set_sync(tp); | |
457 | ||
1c72bf90 | 458 | code = xfs_trans_commit(tp, commit_flags); |
1da177e4 LT |
459 | } |
460 | ||
1da177e4 LT |
461 | xfs_iunlock(ip, lock_flags); |
462 | ||
463 | /* | |
464 | * Release any dquot(s) the inode had kept before chown. | |
465 | */ | |
7d095257 CH |
466 | xfs_qm_dqrele(olddquot1); |
467 | xfs_qm_dqrele(olddquot2); | |
468 | xfs_qm_dqrele(udqp); | |
469 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 470 | |
ef14f0c1 | 471 | if (code) |
1da177e4 | 472 | return code; |
ef14f0c1 CH |
473 | |
474 | /* | |
475 | * XXX(hch): Updating the ACL entries is not atomic vs the i_mode | |
476 | * update. We could avoid this with linked transactions | |
477 | * and passing down the transaction pointer all the way | |
478 | * to attr_set. No previous user of the generic | |
479 | * Posix ACL code seems to care about this issue either. | |
480 | */ | |
481 | if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) { | |
482 | code = -xfs_acl_chmod(inode); | |
483 | if (code) | |
484 | return XFS_ERROR(code); | |
1da177e4 LT |
485 | } |
486 | ||
eb9df39d | 487 | if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) && |
0f285c8a | 488 | !(flags & XFS_ATTR_DMI)) { |
bc4ac74a | 489 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL, |
1da177e4 LT |
490 | NULL, DM_RIGHT_NULL, NULL, NULL, |
491 | 0, 0, AT_DELAY_FLAG(flags)); | |
492 | } | |
493 | return 0; | |
494 | ||
495 | abort_return: | |
496 | commit_flags |= XFS_TRANS_ABORT; | |
497 | /* FALLTHROUGH */ | |
498 | error_return: | |
7d095257 CH |
499 | xfs_qm_dqrele(udqp); |
500 | xfs_qm_dqrele(gdqp); | |
1da177e4 LT |
501 | if (tp) { |
502 | xfs_trans_cancel(tp, commit_flags); | |
503 | } | |
504 | if (lock_flags != 0) { | |
505 | xfs_iunlock(ip, lock_flags); | |
506 | } | |
507 | return code; | |
508 | } | |
509 | ||
7d4fb40a NS |
510 | /* |
511 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
512 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
513 | * bmapi. | |
514 | */ | |
515 | #define SYMLINK_MAPS 2 | |
516 | ||
804c83c3 CH |
517 | STATIC int |
518 | xfs_readlink_bmap( | |
519 | xfs_inode_t *ip, | |
520 | char *link) | |
521 | { | |
522 | xfs_mount_t *mp = ip->i_mount; | |
523 | int pathlen = ip->i_d.di_size; | |
524 | int nmaps = SYMLINK_MAPS; | |
525 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
526 | xfs_daddr_t d; | |
527 | int byte_cnt; | |
528 | int n; | |
529 | xfs_buf_t *bp; | |
530 | int error = 0; | |
531 | ||
532 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0, | |
533 | mval, &nmaps, NULL, NULL); | |
534 | if (error) | |
535 | goto out; | |
536 | ||
537 | for (n = 0; n < nmaps; n++) { | |
538 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
539 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
540 | ||
541 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); | |
542 | error = XFS_BUF_GETERROR(bp); | |
543 | if (error) { | |
544 | xfs_ioerror_alert("xfs_readlink", | |
545 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
546 | xfs_buf_relse(bp); | |
547 | goto out; | |
548 | } | |
549 | if (pathlen < byte_cnt) | |
550 | byte_cnt = pathlen; | |
551 | pathlen -= byte_cnt; | |
552 | ||
553 | memcpy(link, XFS_BUF_PTR(bp), byte_cnt); | |
554 | xfs_buf_relse(bp); | |
555 | } | |
556 | ||
557 | link[ip->i_d.di_size] = '\0'; | |
558 | error = 0; | |
559 | ||
560 | out: | |
561 | return error; | |
562 | } | |
563 | ||
993386c1 | 564 | int |
1da177e4 | 565 | xfs_readlink( |
993386c1 | 566 | xfs_inode_t *ip, |
804c83c3 | 567 | char *link) |
1da177e4 | 568 | { |
804c83c3 | 569 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 570 | int pathlen; |
1da177e4 | 571 | int error = 0; |
1da177e4 | 572 | |
cf441eeb | 573 | xfs_itrace_entry(ip); |
1da177e4 LT |
574 | |
575 | if (XFS_FORCED_SHUTDOWN(mp)) | |
576 | return XFS_ERROR(EIO); | |
577 | ||
578 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
579 | ||
580 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
804c83c3 | 581 | ASSERT(ip->i_d.di_size <= MAXPATHLEN); |
1da177e4 | 582 | |
804c83c3 CH |
583 | pathlen = ip->i_d.di_size; |
584 | if (!pathlen) | |
585 | goto out; | |
1da177e4 LT |
586 | |
587 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
804c83c3 CH |
588 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
589 | link[pathlen] = '\0'; | |
590 | } else { | |
591 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
592 | } |
593 | ||
804c83c3 | 594 | out: |
1da177e4 | 595 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
596 | return error; |
597 | } | |
598 | ||
1da177e4 LT |
599 | /* |
600 | * xfs_fsync | |
601 | * | |
978b7237 DC |
602 | * This is called to sync the inode and its data out to disk. We need to hold |
603 | * the I/O lock while flushing the data, and the inode lock while flushing the | |
604 | * inode. The inode lock CANNOT be held while flushing the data, so acquire | |
605 | * after we're done with that. | |
1da177e4 | 606 | */ |
993386c1 | 607 | int |
1da177e4 | 608 | xfs_fsync( |
978b7237 | 609 | xfs_inode_t *ip) |
1da177e4 | 610 | { |
1da177e4 LT |
611 | xfs_trans_t *tp; |
612 | int error; | |
f538d4da | 613 | int log_flushed = 0, changed = 1; |
1da177e4 | 614 | |
cf441eeb | 615 | xfs_itrace_entry(ip); |
1da177e4 | 616 | |
1da177e4 LT |
617 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
618 | return XFS_ERROR(EIO); | |
619 | ||
978b7237 | 620 | /* capture size updates in I/O completion before writing the inode. */ |
2e656092 | 621 | error = xfs_wait_on_pages(ip, 0, -1); |
978b7237 DC |
622 | if (error) |
623 | return XFS_ERROR(error); | |
776a75fa | 624 | |
1da177e4 | 625 | /* |
978b7237 DC |
626 | * We always need to make sure that the required inode state is safe on |
627 | * disk. The vnode might be clean but we still might need to force the | |
628 | * log because of committed transactions that haven't hit the disk yet. | |
629 | * Likewise, there could be unflushed non-transactional changes to the | |
630 | * inode core that have to go to disk and this requires us to issue | |
631 | * a synchronous transaction to capture these changes correctly. | |
1da177e4 | 632 | * |
978b7237 DC |
633 | * This code relies on the assumption that if the update_* fields |
634 | * of the inode are clear and the inode is unpinned then it is clean | |
635 | * and no action is required. | |
1da177e4 LT |
636 | */ |
637 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
638 | ||
978b7237 | 639 | if (!(ip->i_update_size || ip->i_update_core)) { |
1da177e4 | 640 | /* |
978b7237 DC |
641 | * Timestamps/size haven't changed since last inode flush or |
642 | * inode transaction commit. That means either nothing got | |
643 | * written or a transaction committed which caught the updates. | |
644 | * If the latter happened and the transaction hasn't hit the | |
645 | * disk yet, the inode will be still be pinned. If it is, | |
646 | * force the log. | |
1da177e4 LT |
647 | */ |
648 | ||
649 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
650 | ||
651 | if (xfs_ipincount(ip)) { | |
978b7237 DC |
652 | error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0, |
653 | XFS_LOG_FORCE | XFS_LOG_SYNC, | |
f538d4da CH |
654 | &log_flushed); |
655 | } else { | |
656 | /* | |
978b7237 DC |
657 | * If the inode is not pinned and nothing has changed |
658 | * we don't need to flush the cache. | |
f538d4da CH |
659 | */ |
660 | changed = 0; | |
1da177e4 | 661 | } |
1da177e4 LT |
662 | } else { |
663 | /* | |
978b7237 DC |
664 | * Kick off a transaction to log the inode core to get the |
665 | * updates. The sync transaction will also force the log. | |
1da177e4 LT |
666 | */ |
667 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
668 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
978b7237 DC |
669 | error = xfs_trans_reserve(tp, 0, |
670 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0); | |
671 | if (error) { | |
1da177e4 LT |
672 | xfs_trans_cancel(tp, 0); |
673 | return error; | |
674 | } | |
675 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
676 | ||
677 | /* | |
978b7237 DC |
678 | * Note - it's possible that we might have pushed ourselves out |
679 | * of the way during trans_reserve which would flush the inode. | |
680 | * But there's no guarantee that the inode buffer has actually | |
681 | * gone out yet (it's delwri). Plus the buffer could be pinned | |
682 | * anyway if it's part of an inode in another recent | |
683 | * transaction. So we play it safe and fire off the | |
684 | * transaction anyway. | |
1da177e4 LT |
685 | */ |
686 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
687 | xfs_trans_ihold(tp, ip); | |
688 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
978b7237 | 689 | xfs_trans_set_sync(tp); |
1c72bf90 | 690 | error = _xfs_trans_commit(tp, 0, &log_flushed); |
1da177e4 LT |
691 | |
692 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
693 | } | |
f538d4da CH |
694 | |
695 | if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) { | |
696 | /* | |
697 | * If the log write didn't issue an ordered tag we need | |
698 | * to flush the disk cache for the data device now. | |
699 | */ | |
700 | if (!log_flushed) | |
701 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
702 | ||
703 | /* | |
704 | * If this inode is on the RT dev we need to flush that | |
c41564b5 | 705 | * cache as well. |
f538d4da | 706 | */ |
71ddabb9 | 707 | if (XFS_IS_REALTIME_INODE(ip)) |
f538d4da CH |
708 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); |
709 | } | |
710 | ||
1da177e4 LT |
711 | return error; |
712 | } | |
713 | ||
714 | /* | |
92dfe8d2 DC |
715 | * This is called by xfs_inactive to free any blocks beyond eof |
716 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
717 | * punching a hole to EOF. | |
1da177e4 | 718 | */ |
92dfe8d2 DC |
719 | int |
720 | xfs_free_eofblocks( | |
1da177e4 | 721 | xfs_mount_t *mp, |
92dfe8d2 DC |
722 | xfs_inode_t *ip, |
723 | int flags) | |
1da177e4 LT |
724 | { |
725 | xfs_trans_t *tp; | |
726 | int error; | |
727 | xfs_fileoff_t end_fsb; | |
728 | xfs_fileoff_t last_fsb; | |
729 | xfs_filblks_t map_len; | |
730 | int nimaps; | |
731 | xfs_bmbt_irec_t imap; | |
92dfe8d2 | 732 | int use_iolock = (flags & XFS_FREE_EOF_LOCK); |
1da177e4 LT |
733 | |
734 | /* | |
735 | * Figure out if there are any blocks beyond the end | |
736 | * of the file. If not, then there is nothing to do. | |
737 | */ | |
ba87ea69 | 738 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size)); |
1da177e4 LT |
739 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); |
740 | map_len = last_fsb - end_fsb; | |
741 | if (map_len <= 0) | |
014c2544 | 742 | return 0; |
1da177e4 LT |
743 | |
744 | nimaps = 1; | |
745 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
541d7d3c | 746 | error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0, |
3e57ecf6 | 747 | NULL, 0, &imap, &nimaps, NULL, NULL); |
1da177e4 LT |
748 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
749 | ||
750 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
751 | (imap.br_startblock != HOLESTARTBLOCK || |
752 | ip->i_delayed_blks)) { | |
1da177e4 LT |
753 | /* |
754 | * Attach the dquots to the inode up front. | |
755 | */ | |
7d095257 CH |
756 | error = xfs_qm_dqattach(ip, 0); |
757 | if (error) | |
014c2544 | 758 | return error; |
1da177e4 LT |
759 | |
760 | /* | |
761 | * There are blocks after the end of file. | |
762 | * Free them up now by truncating the file to | |
763 | * its current size. | |
764 | */ | |
765 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
766 | ||
767 | /* | |
768 | * Do the xfs_itruncate_start() call before | |
769 | * reserving any log space because | |
770 | * itruncate_start will call into the buffer | |
771 | * cache and we can't | |
772 | * do that within a transaction. | |
773 | */ | |
92dfe8d2 DC |
774 | if (use_iolock) |
775 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 | 776 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, |
ba87ea69 | 777 | ip->i_size); |
d3cf2094 | 778 | if (error) { |
87ae3c24 | 779 | xfs_trans_cancel(tp, 0); |
92dfe8d2 DC |
780 | if (use_iolock) |
781 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 LM |
782 | return error; |
783 | } | |
1da177e4 LT |
784 | |
785 | error = xfs_trans_reserve(tp, 0, | |
786 | XFS_ITRUNCATE_LOG_RES(mp), | |
787 | 0, XFS_TRANS_PERM_LOG_RES, | |
788 | XFS_ITRUNCATE_LOG_COUNT); | |
789 | if (error) { | |
790 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
791 | xfs_trans_cancel(tp, 0); | |
792 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 793 | return error; |
1da177e4 LT |
794 | } |
795 | ||
796 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
797 | xfs_trans_ijoin(tp, ip, | |
798 | XFS_IOLOCK_EXCL | | |
799 | XFS_ILOCK_EXCL); | |
800 | xfs_trans_ihold(tp, ip); | |
801 | ||
802 | error = xfs_itruncate_finish(&tp, ip, | |
ba87ea69 | 803 | ip->i_size, |
1da177e4 LT |
804 | XFS_DATA_FORK, |
805 | 0); | |
806 | /* | |
807 | * If we get an error at this point we | |
808 | * simply don't bother truncating the file. | |
809 | */ | |
810 | if (error) { | |
811 | xfs_trans_cancel(tp, | |
812 | (XFS_TRANS_RELEASE_LOG_RES | | |
813 | XFS_TRANS_ABORT)); | |
814 | } else { | |
815 | error = xfs_trans_commit(tp, | |
1c72bf90 | 816 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 817 | } |
92dfe8d2 DC |
818 | xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL) |
819 | : XFS_ILOCK_EXCL)); | |
1da177e4 | 820 | } |
014c2544 | 821 | return error; |
1da177e4 LT |
822 | } |
823 | ||
824 | /* | |
825 | * Free a symlink that has blocks associated with it. | |
826 | */ | |
827 | STATIC int | |
828 | xfs_inactive_symlink_rmt( | |
829 | xfs_inode_t *ip, | |
830 | xfs_trans_t **tpp) | |
831 | { | |
832 | xfs_buf_t *bp; | |
833 | int committed; | |
834 | int done; | |
835 | int error; | |
836 | xfs_fsblock_t first_block; | |
837 | xfs_bmap_free_t free_list; | |
838 | int i; | |
839 | xfs_mount_t *mp; | |
840 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
841 | int nmaps; | |
842 | xfs_trans_t *ntp; | |
843 | int size; | |
844 | xfs_trans_t *tp; | |
845 | ||
846 | tp = *tpp; | |
847 | mp = ip->i_mount; | |
848 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
849 | /* | |
850 | * We're freeing a symlink that has some | |
851 | * blocks allocated to it. Free the | |
852 | * blocks here. We know that we've got | |
853 | * either 1 or 2 extents and that we can | |
854 | * free them all in one bunmapi call. | |
855 | */ | |
856 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
857 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
858 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
859 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
860 | xfs_trans_cancel(tp, 0); | |
861 | *tpp = NULL; | |
862 | return error; | |
863 | } | |
864 | /* | |
865 | * Lock the inode, fix the size, and join it to the transaction. | |
866 | * Hold it so in the normal path, we still have it locked for | |
867 | * the second transaction. In the error paths we need it | |
868 | * held so the cancel won't rele it, see below. | |
869 | */ | |
870 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
871 | size = (int)ip->i_d.di_size; | |
872 | ip->i_d.di_size = 0; | |
873 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
874 | xfs_trans_ihold(tp, ip); | |
875 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
876 | /* | |
877 | * Find the block(s) so we can inval and unmap them. | |
878 | */ | |
879 | done = 0; | |
9d87c319 | 880 | xfs_bmap_init(&free_list, &first_block); |
e8c96f8c | 881 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
882 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
883 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 884 | &free_list, NULL))) |
1da177e4 LT |
885 | goto error0; |
886 | /* | |
887 | * Invalidate the block(s). | |
888 | */ | |
889 | for (i = 0; i < nmaps; i++) { | |
890 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
891 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
892 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
893 | xfs_trans_binval(tp, bp); | |
894 | } | |
895 | /* | |
896 | * Unmap the dead block(s) to the free_list. | |
897 | */ | |
898 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 899 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
900 | goto error1; |
901 | ASSERT(done); | |
902 | /* | |
903 | * Commit the first transaction. This logs the EFI and the inode. | |
904 | */ | |
f7c99b6f | 905 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
906 | goto error1; |
907 | /* | |
908 | * The transaction must have been committed, since there were | |
909 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
910 | * The new tp has the extent freeing and EFDs. | |
911 | */ | |
912 | ASSERT(committed); | |
913 | /* | |
914 | * The first xact was committed, so add the inode to the new one. | |
915 | * Mark it dirty so it will be logged and moved forward in the log as | |
916 | * part of every commit. | |
917 | */ | |
918 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
919 | xfs_trans_ihold(tp, ip); | |
920 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
921 | /* | |
922 | * Get a new, empty transaction to return to our caller. | |
923 | */ | |
924 | ntp = xfs_trans_dup(tp); | |
925 | /* | |
c41564b5 | 926 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
927 | * If we get an error on the commit here or on the reserve below, |
928 | * we need to unlock the inode since the new transaction doesn't | |
929 | * have the inode attached. | |
930 | */ | |
1c72bf90 | 931 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
932 | tp = ntp; |
933 | if (error) { | |
934 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
935 | goto error0; | |
936 | } | |
cc09c0dc DC |
937 | /* |
938 | * transaction commit worked ok so we can drop the extra ticket | |
939 | * reference that we gained in xfs_trans_dup() | |
940 | */ | |
941 | xfs_log_ticket_put(tp->t_ticket); | |
942 | ||
1da177e4 LT |
943 | /* |
944 | * Remove the memory for extent descriptions (just bookkeeping). | |
945 | */ | |
946 | if (ip->i_df.if_bytes) | |
947 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
948 | ASSERT(ip->i_df.if_bytes == 0); | |
949 | /* | |
950 | * Put an itruncate log reservation in the new transaction | |
951 | * for our caller. | |
952 | */ | |
953 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
954 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
955 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
956 | goto error0; | |
957 | } | |
958 | /* | |
959 | * Return with the inode locked but not joined to the transaction. | |
960 | */ | |
961 | *tpp = tp; | |
962 | return 0; | |
963 | ||
964 | error1: | |
965 | xfs_bmap_cancel(&free_list); | |
966 | error0: | |
967 | /* | |
968 | * Have to come here with the inode locked and either | |
969 | * (held and in the transaction) or (not in the transaction). | |
970 | * If the inode isn't held then cancel would iput it, but | |
971 | * that's wrong since this is inactive and the vnode ref | |
972 | * count is 0 already. | |
973 | * Cancel won't do anything to the inode if held, but it still | |
974 | * needs to be locked until the cancel is done, if it was | |
975 | * joined to the transaction. | |
976 | */ | |
977 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
978 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
979 | *tpp = NULL; | |
980 | return error; | |
981 | ||
982 | } | |
983 | ||
984 | STATIC int | |
985 | xfs_inactive_symlink_local( | |
986 | xfs_inode_t *ip, | |
987 | xfs_trans_t **tpp) | |
988 | { | |
989 | int error; | |
990 | ||
991 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
992 | /* | |
993 | * We're freeing a symlink which fit into | |
994 | * the inode. Just free the memory used | |
995 | * to hold the old symlink. | |
996 | */ | |
997 | error = xfs_trans_reserve(*tpp, 0, | |
998 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
999 | 0, XFS_TRANS_PERM_LOG_RES, | |
1000 | XFS_ITRUNCATE_LOG_COUNT); | |
1001 | ||
1002 | if (error) { | |
1003 | xfs_trans_cancel(*tpp, 0); | |
1004 | *tpp = NULL; | |
014c2544 | 1005 | return error; |
1da177e4 LT |
1006 | } |
1007 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1008 | ||
1009 | /* | |
1010 | * Zero length symlinks _can_ exist. | |
1011 | */ | |
1012 | if (ip->i_df.if_bytes > 0) { | |
1013 | xfs_idata_realloc(ip, | |
1014 | -(ip->i_df.if_bytes), | |
1015 | XFS_DATA_FORK); | |
1016 | ASSERT(ip->i_df.if_bytes == 0); | |
1017 | } | |
014c2544 | 1018 | return 0; |
1da177e4 LT |
1019 | } |
1020 | ||
1da177e4 LT |
1021 | STATIC int |
1022 | xfs_inactive_attrs( | |
1023 | xfs_inode_t *ip, | |
1024 | xfs_trans_t **tpp) | |
1025 | { | |
1026 | xfs_trans_t *tp; | |
1027 | int error; | |
1028 | xfs_mount_t *mp; | |
1029 | ||
579aa9ca | 1030 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
1da177e4 LT |
1031 | tp = *tpp; |
1032 | mp = ip->i_mount; | |
1033 | ASSERT(ip->i_d.di_forkoff != 0); | |
e5720eec | 1034 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 1035 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
e5720eec DC |
1036 | if (error) |
1037 | goto error_unlock; | |
1da177e4 LT |
1038 | |
1039 | error = xfs_attr_inactive(ip); | |
e5720eec DC |
1040 | if (error) |
1041 | goto error_unlock; | |
1da177e4 LT |
1042 | |
1043 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1044 | error = xfs_trans_reserve(tp, 0, | |
1045 | XFS_IFREE_LOG_RES(mp), | |
1046 | 0, XFS_TRANS_PERM_LOG_RES, | |
1047 | XFS_INACTIVE_LOG_COUNT); | |
e5720eec DC |
1048 | if (error) |
1049 | goto error_cancel; | |
1da177e4 LT |
1050 | |
1051 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1052 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1053 | xfs_trans_ihold(tp, ip); | |
1054 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1055 | ||
1056 | ASSERT(ip->i_d.di_anextents == 0); | |
1057 | ||
1058 | *tpp = tp; | |
014c2544 | 1059 | return 0; |
e5720eec DC |
1060 | |
1061 | error_cancel: | |
1062 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1063 | xfs_trans_cancel(tp, 0); | |
1064 | error_unlock: | |
1065 | *tpp = NULL; | |
1066 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1067 | return error; | |
1da177e4 LT |
1068 | } |
1069 | ||
993386c1 | 1070 | int |
1da177e4 | 1071 | xfs_release( |
993386c1 | 1072 | xfs_inode_t *ip) |
1da177e4 | 1073 | { |
993386c1 | 1074 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
1075 | int error; |
1076 | ||
42173f68 | 1077 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) |
1da177e4 | 1078 | return 0; |
1da177e4 LT |
1079 | |
1080 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1081 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1082 | return 0; |
1083 | ||
2a82b8be | 1084 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
1085 | int truncated; |
1086 | ||
2a82b8be DC |
1087 | /* |
1088 | * If we are using filestreams, and we have an unlinked | |
1089 | * file that we are processing the last close on, then nothing | |
1090 | * will be able to reopen and write to this file. Purge this | |
1091 | * inode from the filestreams cache so that it doesn't delay | |
1092 | * teardown of the inode. | |
1093 | */ | |
1094 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1095 | xfs_filestream_deassociate(ip); | |
1096 | ||
fbf3ce8d CH |
1097 | /* |
1098 | * If we previously truncated this file and removed old data | |
1099 | * in the process, we want to initiate "early" writeout on | |
1100 | * the last close. This is an attempt to combat the notorious | |
1101 | * NULL files problem which is particularly noticable from a | |
1102 | * truncate down, buffered (re-)write (delalloc), followed by | |
1103 | * a crash. What we are effectively doing here is | |
1104 | * significantly reducing the time window where we'd otherwise | |
1105 | * be exposed to that problem. | |
1106 | */ | |
09262b43 | 1107 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
df80c933 | 1108 | if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) |
739bfb2a | 1109 | xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE); |
fbf3ce8d CH |
1110 | } |
1111 | ||
1da177e4 LT |
1112 | if (ip->i_d.di_nlink != 0) { |
1113 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
df80c933 | 1114 | ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 || |
68bdb6ea | 1115 | ip->i_delayed_blks > 0)) && |
1da177e4 | 1116 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
1117 | (!(ip->i_d.di_flags & |
1118 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
92dfe8d2 DC |
1119 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1120 | if (error) | |
014c2544 | 1121 | return error; |
1da177e4 LT |
1122 | } |
1123 | } | |
1124 | ||
1125 | return 0; | |
1126 | } | |
1127 | ||
1128 | /* | |
1129 | * xfs_inactive | |
1130 | * | |
1131 | * This is called when the vnode reference count for the vnode | |
1132 | * goes to zero. If the file has been unlinked, then it must | |
1133 | * now be truncated. Also, we clear all of the read-ahead state | |
1134 | * kept for the inode here since the file is now closed. | |
1135 | */ | |
993386c1 | 1136 | int |
1da177e4 | 1137 | xfs_inactive( |
993386c1 | 1138 | xfs_inode_t *ip) |
1da177e4 | 1139 | { |
67fcaa73 | 1140 | xfs_bmap_free_t free_list; |
1da177e4 LT |
1141 | xfs_fsblock_t first_block; |
1142 | int committed; | |
1143 | xfs_trans_t *tp; | |
1144 | xfs_mount_t *mp; | |
1145 | int error; | |
1146 | int truncate; | |
1147 | ||
cf441eeb | 1148 | xfs_itrace_entry(ip); |
1da177e4 | 1149 | |
1da177e4 LT |
1150 | /* |
1151 | * If the inode is already free, then there can be nothing | |
1152 | * to clean up here. | |
1153 | */ | |
cb4c8cc1 | 1154 | if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) { |
1da177e4 LT |
1155 | ASSERT(ip->i_df.if_real_bytes == 0); |
1156 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1157 | return VN_INACTIVE_CACHE; | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * Only do a truncate if it's a regular file with | |
1162 | * some actual space in it. It's OK to look at the | |
1163 | * inode's fields without the lock because we're the | |
1164 | * only one with a reference to the inode. | |
1165 | */ | |
1166 | truncate = ((ip->i_d.di_nlink == 0) && | |
ba87ea69 LM |
1167 | ((ip->i_d.di_size != 0) || (ip->i_size != 0) || |
1168 | (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1169 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1170 | ||
1171 | mp = ip->i_mount; | |
1172 | ||
bc4ac74a CH |
1173 | if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) |
1174 | XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL); | |
1da177e4 LT |
1175 | |
1176 | error = 0; | |
1177 | ||
1178 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1179 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1180 | goto out; |
1181 | ||
1182 | if (ip->i_d.di_nlink != 0) { | |
1183 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
df80c933 | 1184 | ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 || |
68bdb6ea | 1185 | ip->i_delayed_blks > 0)) && |
dd9f438e NS |
1186 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1187 | (!(ip->i_d.di_flags & | |
1188 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1189 | (ip->i_delayed_blks != 0)))) { | |
92dfe8d2 DC |
1190 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1191 | if (error) | |
014c2544 | 1192 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1193 | } |
1194 | goto out; | |
1195 | } | |
1196 | ||
1197 | ASSERT(ip->i_d.di_nlink == 0); | |
1198 | ||
7d095257 CH |
1199 | error = xfs_qm_dqattach(ip, 0); |
1200 | if (error) | |
014c2544 | 1201 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1202 | |
1203 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1204 | if (truncate) { | |
1205 | /* | |
1206 | * Do the xfs_itruncate_start() call before | |
1207 | * reserving any log space because itruncate_start | |
1208 | * will call into the buffer cache and we can't | |
1209 | * do that within a transaction. | |
1210 | */ | |
1211 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1212 | ||
d3cf2094 LM |
1213 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); |
1214 | if (error) { | |
87ae3c24 | 1215 | xfs_trans_cancel(tp, 0); |
d3cf2094 LM |
1216 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1217 | return VN_INACTIVE_CACHE; | |
1218 | } | |
1da177e4 LT |
1219 | |
1220 | error = xfs_trans_reserve(tp, 0, | |
1221 | XFS_ITRUNCATE_LOG_RES(mp), | |
1222 | 0, XFS_TRANS_PERM_LOG_RES, | |
1223 | XFS_ITRUNCATE_LOG_COUNT); | |
1224 | if (error) { | |
1225 | /* Don't call itruncate_cleanup */ | |
1226 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1227 | xfs_trans_cancel(tp, 0); | |
1228 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1229 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1233 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1234 | xfs_trans_ihold(tp, ip); | |
1235 | ||
1236 | /* | |
1237 | * normally, we have to run xfs_itruncate_finish sync. | |
1238 | * But if filesystem is wsync and we're in the inactive | |
1239 | * path, then we know that nlink == 0, and that the | |
1240 | * xaction that made nlink == 0 is permanently committed | |
1241 | * since xfs_remove runs as a synchronous transaction. | |
1242 | */ | |
1243 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1244 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1245 | ||
1246 | if (error) { | |
1247 | xfs_trans_cancel(tp, | |
1248 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1249 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1250 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1251 | } |
1252 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1253 | ||
1254 | /* | |
1255 | * If we get an error while cleaning up a | |
1256 | * symlink we bail out. | |
1257 | */ | |
1258 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1259 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1260 | xfs_inactive_symlink_local(ip, &tp); | |
1261 | ||
1262 | if (error) { | |
1263 | ASSERT(tp == NULL); | |
014c2544 | 1264 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1265 | } |
1266 | ||
1267 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1268 | xfs_trans_ihold(tp, ip); | |
1269 | } else { | |
1270 | error = xfs_trans_reserve(tp, 0, | |
1271 | XFS_IFREE_LOG_RES(mp), | |
1272 | 0, XFS_TRANS_PERM_LOG_RES, | |
1273 | XFS_INACTIVE_LOG_COUNT); | |
1274 | if (error) { | |
1275 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1276 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1277 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1278 | } |
1279 | ||
1280 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1281 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1282 | xfs_trans_ihold(tp, ip); | |
1283 | } | |
1284 | ||
1285 | /* | |
1286 | * If there are attributes associated with the file | |
1287 | * then blow them away now. The code calls a routine | |
1288 | * that recursively deconstructs the attribute fork. | |
1289 | * We need to just commit the current transaction | |
1290 | * because we can't use it for xfs_attr_inactive(). | |
1291 | */ | |
1292 | if (ip->i_d.di_anextents > 0) { | |
1293 | error = xfs_inactive_attrs(ip, &tp); | |
1294 | /* | |
1295 | * If we got an error, the transaction is already | |
1296 | * cancelled, and the inode is unlocked. Just get out. | |
1297 | */ | |
1298 | if (error) | |
014c2544 | 1299 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1300 | } else if (ip->i_afp) { |
1301 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Free the inode. | |
1306 | */ | |
9d87c319 | 1307 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
1308 | error = xfs_ifree(tp, ip, &free_list); |
1309 | if (error) { | |
1310 | /* | |
1311 | * If we fail to free the inode, shut down. The cancel | |
1312 | * might do that, we need to make sure. Otherwise the | |
1313 | * inode might be lost for a long time or forever. | |
1314 | */ | |
1315 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1316 | cmn_err(CE_NOTE, | |
1317 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1318 | error, mp->m_fsname); | |
7d04a335 | 1319 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1320 | } |
1321 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1322 | } else { | |
1323 | /* | |
1324 | * Credit the quota account(s). The inode is gone. | |
1325 | */ | |
7d095257 | 1326 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); |
1da177e4 LT |
1327 | |
1328 | /* | |
78e9da77 DC |
1329 | * Just ignore errors at this point. There is nothing we can |
1330 | * do except to try to keep going. Make sure it's not a silent | |
1331 | * error. | |
1da177e4 | 1332 | */ |
78e9da77 DC |
1333 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1334 | if (error) | |
1335 | xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: " | |
1336 | "xfs_bmap_finish() returned error %d", error); | |
1337 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1338 | if (error) | |
1339 | xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: " | |
1340 | "xfs_trans_commit() returned error %d", error); | |
1da177e4 | 1341 | } |
7d095257 | 1342 | |
1da177e4 LT |
1343 | /* |
1344 | * Release the dquots held by inode, if any. | |
1345 | */ | |
7d095257 | 1346 | xfs_qm_dqdetach(ip); |
1da177e4 LT |
1347 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
1348 | ||
1349 | out: | |
1350 | return VN_INACTIVE_CACHE; | |
1351 | } | |
1352 | ||
384f3ced BN |
1353 | /* |
1354 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
1355 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
1356 | * ci_name->name will point to a the actual name (caller must free) or | |
1357 | * will be set to NULL if an exact match is found. | |
1358 | */ | |
993386c1 | 1359 | int |
1da177e4 | 1360 | xfs_lookup( |
993386c1 | 1361 | xfs_inode_t *dp, |
556b8b16 | 1362 | struct xfs_name *name, |
384f3ced BN |
1363 | xfs_inode_t **ipp, |
1364 | struct xfs_name *ci_name) | |
1da177e4 | 1365 | { |
eca450b7 | 1366 | xfs_ino_t inum; |
1da177e4 LT |
1367 | int error; |
1368 | uint lock_mode; | |
1da177e4 | 1369 | |
cf441eeb | 1370 | xfs_itrace_entry(dp); |
1da177e4 LT |
1371 | |
1372 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1373 | return XFS_ERROR(EIO); | |
1374 | ||
1375 | lock_mode = xfs_ilock_map_shared(dp); | |
384f3ced | 1376 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
1da177e4 | 1377 | xfs_iunlock_map_shared(dp, lock_mode); |
eca450b7 CH |
1378 | |
1379 | if (error) | |
1380 | goto out; | |
1381 | ||
1382 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0); | |
1383 | if (error) | |
384f3ced | 1384 | goto out_free_name; |
eca450b7 CH |
1385 | |
1386 | xfs_itrace_ref(*ipp); | |
1387 | return 0; | |
1388 | ||
384f3ced BN |
1389 | out_free_name: |
1390 | if (ci_name) | |
1391 | kmem_free(ci_name->name); | |
1392 | out: | |
eca450b7 | 1393 | *ipp = NULL; |
1da177e4 LT |
1394 | return error; |
1395 | } | |
1396 | ||
993386c1 | 1397 | int |
1da177e4 | 1398 | xfs_create( |
993386c1 | 1399 | xfs_inode_t *dp, |
556b8b16 | 1400 | struct xfs_name *name, |
3e5daf05 CH |
1401 | mode_t mode, |
1402 | xfs_dev_t rdev, | |
979ebab1 | 1403 | xfs_inode_t **ipp, |
1da177e4 LT |
1404 | cred_t *credp) |
1405 | { | |
517b5e8c CH |
1406 | int is_dir = S_ISDIR(mode); |
1407 | struct xfs_mount *mp = dp->i_mount; | |
1408 | struct xfs_inode *ip = NULL; | |
1409 | struct xfs_trans *tp = NULL; | |
556b8b16 | 1410 | int error; |
1da177e4 LT |
1411 | xfs_bmap_free_t free_list; |
1412 | xfs_fsblock_t first_block; | |
993386c1 | 1413 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1414 | uint cancel_flags; |
1415 | int committed; | |
1416 | xfs_prid_t prid; | |
517b5e8c CH |
1417 | struct xfs_dquot *udqp = NULL; |
1418 | struct xfs_dquot *gdqp = NULL; | |
1da177e4 | 1419 | uint resblks; |
517b5e8c CH |
1420 | uint log_res; |
1421 | uint log_count; | |
1da177e4 | 1422 | |
cf441eeb | 1423 | xfs_itrace_entry(dp); |
1da177e4 | 1424 | |
517b5e8c CH |
1425 | if (XFS_FORCED_SHUTDOWN(mp)) |
1426 | return XFS_ERROR(EIO); | |
1427 | ||
eb9df39d | 1428 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 | 1429 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
bc4ac74a | 1430 | dp, DM_RIGHT_NULL, NULL, |
556b8b16 | 1431 | DM_RIGHT_NULL, name->name, NULL, |
3e5daf05 | 1432 | mode, 0, 0); |
1da177e4 LT |
1433 | |
1434 | if (error) | |
1435 | return error; | |
1da177e4 LT |
1436 | } |
1437 | ||
365ca83d NS |
1438 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1439 | prid = dp->i_d.di_projid; | |
1da177e4 | 1440 | else |
517b5e8c | 1441 | prid = dfltprid; |
1da177e4 LT |
1442 | |
1443 | /* | |
1444 | * Make sure that we have allocated dquot(s) on disk. | |
1445 | */ | |
7d095257 | 1446 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
517b5e8c | 1447 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1da177e4 LT |
1448 | if (error) |
1449 | goto std_return; | |
1450 | ||
517b5e8c CH |
1451 | if (is_dir) { |
1452 | rdev = 0; | |
1453 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); | |
1454 | log_res = XFS_MKDIR_LOG_RES(mp); | |
1455 | log_count = XFS_MKDIR_LOG_COUNT; | |
1456 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
1457 | } else { | |
1458 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
1459 | log_res = XFS_CREATE_LOG_RES(mp); | |
1460 | log_count = XFS_CREATE_LOG_COUNT; | |
1461 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1462 | } | |
1da177e4 | 1463 | |
1da177e4 | 1464 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
517b5e8c | 1465 | |
1da177e4 LT |
1466 | /* |
1467 | * Initially assume that the file does not exist and | |
1468 | * reserve the resources for that case. If that is not | |
1469 | * the case we'll drop the one we have and get a more | |
1470 | * appropriate transaction later. | |
1471 | */ | |
517b5e8c CH |
1472 | error = xfs_trans_reserve(tp, resblks, log_res, 0, |
1473 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 | 1474 | if (error == ENOSPC) { |
153fec43 DC |
1475 | /* flush outstanding delalloc blocks and retry */ |
1476 | xfs_flush_inodes(dp); | |
1477 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1478 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1479 | } | |
1480 | if (error == ENOSPC) { | |
1481 | /* No space at all so try a "no-allocation" reservation */ | |
1da177e4 | 1482 | resblks = 0; |
517b5e8c CH |
1483 | error = xfs_trans_reserve(tp, 0, log_res, 0, |
1484 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 LT |
1485 | } |
1486 | if (error) { | |
1487 | cancel_flags = 0; | |
517b5e8c | 1488 | goto out_trans_cancel; |
1da177e4 LT |
1489 | } |
1490 | ||
f7c66ce3 | 1491 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1492 | unlock_dp_on_error = B_TRUE; |
1da177e4 | 1493 | |
517b5e8c CH |
1494 | /* |
1495 | * Check for directory link count overflow. | |
1496 | */ | |
1497 | if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) { | |
1498 | error = XFS_ERROR(EMLINK); | |
1499 | goto out_trans_cancel; | |
1500 | } | |
1da177e4 | 1501 | |
517b5e8c | 1502 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
1503 | |
1504 | /* | |
1505 | * Reserve disk quota and the inode. | |
1506 | */ | |
7d095257 | 1507 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 | 1508 | if (error) |
517b5e8c | 1509 | goto out_trans_cancel; |
1da177e4 | 1510 | |
556b8b16 BN |
1511 | error = xfs_dir_canenter(tp, dp, name, resblks); |
1512 | if (error) | |
517b5e8c CH |
1513 | goto out_trans_cancel; |
1514 | ||
1515 | /* | |
1516 | * A newly created regular or special file just has one directory | |
1517 | * entry pointing to them, but a directory also the "." entry | |
1518 | * pointing to itself. | |
1519 | */ | |
1520 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, credp, | |
1521 | prid, resblks > 0, &ip, &committed); | |
1da177e4 LT |
1522 | if (error) { |
1523 | if (error == ENOSPC) | |
517b5e8c CH |
1524 | goto out_trans_cancel; |
1525 | goto out_trans_abort; | |
1da177e4 | 1526 | } |
1da177e4 LT |
1527 | |
1528 | /* | |
1529 | * At this point, we've gotten a newly allocated inode. | |
1530 | * It is locked (and joined to the transaction). | |
1531 | */ | |
517b5e8c | 1532 | xfs_itrace_ref(ip); |
579aa9ca | 1533 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
1534 | |
1535 | /* | |
993386c1 CH |
1536 | * Now we join the directory inode to the transaction. We do not do it |
1537 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1538 | * (and release all the locks). An error from here on will result in | |
1539 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1540 | * error path. | |
1da177e4 | 1541 | */ |
979ebab1 | 1542 | IHOLD(dp); |
1da177e4 | 1543 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 1544 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 1545 | |
556b8b16 | 1546 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
f6c2d1fa NS |
1547 | &first_block, &free_list, resblks ? |
1548 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1549 | if (error) { |
1550 | ASSERT(error != ENOSPC); | |
517b5e8c | 1551 | goto out_trans_abort; |
1da177e4 LT |
1552 | } |
1553 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1554 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1555 | ||
517b5e8c CH |
1556 | if (is_dir) { |
1557 | error = xfs_dir_init(tp, ip, dp); | |
1558 | if (error) | |
1559 | goto out_bmap_cancel; | |
1560 | ||
1561 | error = xfs_bumplink(tp, dp); | |
1562 | if (error) | |
1563 | goto out_bmap_cancel; | |
1564 | } | |
1565 | ||
1da177e4 LT |
1566 | /* |
1567 | * If this is a synchronous mount, make sure that the | |
1568 | * create transaction goes to disk before returning to | |
1569 | * the user. | |
1570 | */ | |
517b5e8c | 1571 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 1572 | xfs_trans_set_sync(tp); |
1da177e4 | 1573 | |
1da177e4 LT |
1574 | /* |
1575 | * Attach the dquot(s) to the inodes and modify them incore. | |
1576 | * These ids of the inode couldn't have changed since the new | |
1577 | * inode has been locked ever since it was created. | |
1578 | */ | |
7d095257 | 1579 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 LT |
1580 | |
1581 | /* | |
1582 | * xfs_trans_commit normally decrements the vnode ref count | |
1583 | * when it unlocks the inode. Since we want to return the | |
1584 | * vnode to the caller, we bump the vnode ref count now. | |
1585 | */ | |
1586 | IHOLD(ip); | |
1da177e4 | 1587 | |
f7c99b6f | 1588 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
517b5e8c CH |
1589 | if (error) |
1590 | goto out_abort_rele; | |
1da177e4 | 1591 | |
1c72bf90 | 1592 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1593 | if (error) { |
1594 | IRELE(ip); | |
517b5e8c | 1595 | goto out_dqrele; |
1da177e4 LT |
1596 | } |
1597 | ||
7d095257 CH |
1598 | xfs_qm_dqrele(udqp); |
1599 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1600 | |
979ebab1 | 1601 | *ipp = ip; |
1da177e4 LT |
1602 | |
1603 | /* Fallthrough to std_return with error = 0 */ | |
517b5e8c CH |
1604 | std_return: |
1605 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { | |
1606 | XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, dp, DM_RIGHT_NULL, | |
1607 | ip, DM_RIGHT_NULL, name->name, NULL, mode, | |
1608 | error, 0); | |
1da177e4 | 1609 | } |
517b5e8c | 1610 | |
1da177e4 LT |
1611 | return error; |
1612 | ||
517b5e8c CH |
1613 | out_bmap_cancel: |
1614 | xfs_bmap_cancel(&free_list); | |
1615 | out_trans_abort: | |
1da177e4 | 1616 | cancel_flags |= XFS_TRANS_ABORT; |
517b5e8c CH |
1617 | out_trans_cancel: |
1618 | xfs_trans_cancel(tp, cancel_flags); | |
1619 | out_dqrele: | |
7d095257 CH |
1620 | xfs_qm_dqrele(udqp); |
1621 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1622 | |
993386c1 CH |
1623 | if (unlock_dp_on_error) |
1624 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1625 | ||
1da177e4 LT |
1626 | goto std_return; |
1627 | ||
517b5e8c | 1628 | out_abort_rele: |
1da177e4 LT |
1629 | /* |
1630 | * Wait until after the current transaction is aborted to | |
1631 | * release the inode. This prevents recursive transactions | |
1632 | * and deadlocks from xfs_inactive. | |
1633 | */ | |
517b5e8c | 1634 | xfs_bmap_cancel(&free_list); |
1da177e4 LT |
1635 | cancel_flags |= XFS_TRANS_ABORT; |
1636 | xfs_trans_cancel(tp, cancel_flags); | |
1637 | IRELE(ip); | |
517b5e8c CH |
1638 | unlock_dp_on_error = B_FALSE; |
1639 | goto out_dqrele; | |
1da177e4 LT |
1640 | } |
1641 | ||
1da177e4 LT |
1642 | #ifdef DEBUG |
1643 | int xfs_locked_n; | |
1644 | int xfs_small_retries; | |
1645 | int xfs_middle_retries; | |
1646 | int xfs_lots_retries; | |
1647 | int xfs_lock_delays; | |
1648 | #endif | |
1649 | ||
f7c66ce3 LM |
1650 | /* |
1651 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
1652 | * a different value | |
1653 | */ | |
1654 | static inline int | |
1655 | xfs_lock_inumorder(int lock_mode, int subclass) | |
1656 | { | |
1657 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 1658 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 1659 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 1660 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
1661 | |
1662 | return lock_mode; | |
1663 | } | |
1664 | ||
1da177e4 LT |
1665 | /* |
1666 | * The following routine will lock n inodes in exclusive mode. | |
1667 | * We assume the caller calls us with the inodes in i_ino order. | |
1668 | * | |
1669 | * We need to detect deadlock where an inode that we lock | |
1670 | * is in the AIL and we start waiting for another inode that is locked | |
1671 | * by a thread in a long running transaction (such as truncate). This can | |
1672 | * result in deadlock since the long running trans might need to wait | |
1673 | * for the inode we just locked in order to push the tail and free space | |
1674 | * in the log. | |
1675 | */ | |
1676 | void | |
1677 | xfs_lock_inodes( | |
1678 | xfs_inode_t **ips, | |
1679 | int inodes, | |
1da177e4 LT |
1680 | uint lock_mode) |
1681 | { | |
1682 | int attempts = 0, i, j, try_lock; | |
1683 | xfs_log_item_t *lp; | |
1684 | ||
1685 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
1686 | ||
cfa853e4 CH |
1687 | try_lock = 0; |
1688 | i = 0; | |
1da177e4 LT |
1689 | |
1690 | again: | |
1691 | for (; i < inodes; i++) { | |
1692 | ASSERT(ips[i]); | |
1693 | ||
1694 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
1695 | continue; | |
1696 | ||
1697 | /* | |
1698 | * If try_lock is not set yet, make sure all locked inodes | |
1699 | * are not in the AIL. | |
1700 | * If any are, set try_lock to be used later. | |
1701 | */ | |
1702 | ||
1703 | if (!try_lock) { | |
1704 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
1705 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
1706 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1707 | try_lock++; | |
1708 | } | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | /* | |
1713 | * If any of the previous locks we have locked is in the AIL, | |
1714 | * we must TRY to get the second and subsequent locks. If | |
1715 | * we can't get any, we must release all we have | |
1716 | * and try again. | |
1717 | */ | |
1718 | ||
1719 | if (try_lock) { | |
1720 | /* try_lock must be 0 if i is 0. */ | |
1721 | /* | |
1722 | * try_lock means we have an inode locked | |
1723 | * that is in the AIL. | |
1724 | */ | |
1725 | ASSERT(i != 0); | |
f7c66ce3 | 1726 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
1727 | attempts++; |
1728 | ||
1729 | /* | |
1730 | * Unlock all previous guys and try again. | |
1731 | * xfs_iunlock will try to push the tail | |
1732 | * if the inode is in the AIL. | |
1733 | */ | |
1734 | ||
1735 | for(j = i - 1; j >= 0; j--) { | |
1736 | ||
1737 | /* | |
1738 | * Check to see if we've already | |
1739 | * unlocked this one. | |
1740 | * Not the first one going back, | |
1741 | * and the inode ptr is the same. | |
1742 | */ | |
1743 | if ((j != (i - 1)) && ips[j] == | |
1744 | ips[j+1]) | |
1745 | continue; | |
1746 | ||
1747 | xfs_iunlock(ips[j], lock_mode); | |
1748 | } | |
1749 | ||
1750 | if ((attempts % 5) == 0) { | |
1751 | delay(1); /* Don't just spin the CPU */ | |
1752 | #ifdef DEBUG | |
1753 | xfs_lock_delays++; | |
1754 | #endif | |
1755 | } | |
1756 | i = 0; | |
1757 | try_lock = 0; | |
1758 | goto again; | |
1759 | } | |
1760 | } else { | |
f7c66ce3 | 1761 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
1762 | } |
1763 | } | |
1764 | ||
1765 | #ifdef DEBUG | |
1766 | if (attempts) { | |
1767 | if (attempts < 5) xfs_small_retries++; | |
1768 | else if (attempts < 100) xfs_middle_retries++; | |
1769 | else xfs_lots_retries++; | |
1770 | } else { | |
1771 | xfs_locked_n++; | |
1772 | } | |
1773 | #endif | |
1774 | } | |
1775 | ||
f9114eba DC |
1776 | /* |
1777 | * xfs_lock_two_inodes() can only be used to lock one type of lock | |
1778 | * at a time - the iolock or the ilock, but not both at once. If | |
1779 | * we lock both at once, lockdep will report false positives saying | |
1780 | * we have violated locking orders. | |
1781 | */ | |
e1cccd91 CH |
1782 | void |
1783 | xfs_lock_two_inodes( | |
1784 | xfs_inode_t *ip0, | |
1785 | xfs_inode_t *ip1, | |
1786 | uint lock_mode) | |
1787 | { | |
1788 | xfs_inode_t *temp; | |
1789 | int attempts = 0; | |
1790 | xfs_log_item_t *lp; | |
1791 | ||
f9114eba DC |
1792 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) |
1793 | ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0); | |
e1cccd91 CH |
1794 | ASSERT(ip0->i_ino != ip1->i_ino); |
1795 | ||
1796 | if (ip0->i_ino > ip1->i_ino) { | |
1797 | temp = ip0; | |
1798 | ip0 = ip1; | |
1799 | ip1 = temp; | |
1800 | } | |
1801 | ||
1802 | again: | |
1803 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
1804 | ||
1805 | /* | |
1806 | * If the first lock we have locked is in the AIL, we must TRY to get | |
1807 | * the second lock. If we can't get it, we must release the first one | |
1808 | * and try again. | |
1809 | */ | |
1810 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
1811 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1812 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
1813 | xfs_iunlock(ip0, lock_mode); | |
1814 | if ((++attempts % 5) == 0) | |
1815 | delay(1); /* Don't just spin the CPU */ | |
1816 | goto again; | |
1817 | } | |
1818 | } else { | |
1819 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
1820 | } | |
1821 | } | |
1822 | ||
993386c1 | 1823 | int |
1da177e4 | 1824 | xfs_remove( |
993386c1 | 1825 | xfs_inode_t *dp, |
556b8b16 BN |
1826 | struct xfs_name *name, |
1827 | xfs_inode_t *ip) | |
1da177e4 | 1828 | { |
993386c1 | 1829 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 1830 | xfs_trans_t *tp = NULL; |
8f112e3b | 1831 | int is_dir = S_ISDIR(ip->i_d.di_mode); |
1da177e4 LT |
1832 | int error = 0; |
1833 | xfs_bmap_free_t free_list; | |
1834 | xfs_fsblock_t first_block; | |
1835 | int cancel_flags; | |
1836 | int committed; | |
1da177e4 LT |
1837 | int link_zero; |
1838 | uint resblks; | |
8f112e3b | 1839 | uint log_count; |
1da177e4 | 1840 | |
cf441eeb | 1841 | xfs_itrace_entry(dp); |
8f112e3b | 1842 | xfs_itrace_entry(ip); |
1da177e4 | 1843 | |
1da177e4 LT |
1844 | if (XFS_FORCED_SHUTDOWN(mp)) |
1845 | return XFS_ERROR(EIO); | |
1846 | ||
eb9df39d | 1847 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
556b8b16 BN |
1848 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL, |
1849 | NULL, DM_RIGHT_NULL, name->name, NULL, | |
1850 | ip->i_d.di_mode, 0, 0); | |
1da177e4 LT |
1851 | if (error) |
1852 | return error; | |
1853 | } | |
1854 | ||
7d095257 | 1855 | error = xfs_qm_dqattach(dp, 0); |
8f112e3b CH |
1856 | if (error) |
1857 | goto std_return; | |
1858 | ||
7d095257 | 1859 | error = xfs_qm_dqattach(ip, 0); |
8f112e3b | 1860 | if (error) |
1da177e4 | 1861 | goto std_return; |
1da177e4 | 1862 | |
8f112e3b CH |
1863 | if (is_dir) { |
1864 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
1865 | log_count = XFS_DEFAULT_LOG_COUNT; | |
1866 | } else { | |
1867 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
1868 | log_count = XFS_REMOVE_LOG_COUNT; | |
1869 | } | |
1da177e4 | 1870 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
8f112e3b | 1871 | |
1da177e4 LT |
1872 | /* |
1873 | * We try to get the real space reservation first, | |
1874 | * allowing for directory btree deletion(s) implying | |
1875 | * possible bmap insert(s). If we can't get the space | |
1876 | * reservation then we use 0 instead, and avoid the bmap | |
1877 | * btree insert(s) in the directory code by, if the bmap | |
1878 | * insert tries to happen, instead trimming the LAST | |
1879 | * block from the directory. | |
1880 | */ | |
1881 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
1882 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1883 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1884 | if (error == ENOSPC) { |
1885 | resblks = 0; | |
1886 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1887 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1888 | } |
1889 | if (error) { | |
1890 | ASSERT(error != ENOSPC); | |
8f112e3b CH |
1891 | cancel_flags = 0; |
1892 | goto out_trans_cancel; | |
1da177e4 LT |
1893 | } |
1894 | ||
e1cccd91 | 1895 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1896 | |
1897 | /* | |
1898 | * At this point, we've gotten both the directory and the entry | |
1899 | * inodes locked. | |
1900 | */ | |
5df78e73 | 1901 | IHOLD(ip); |
1da177e4 | 1902 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
82dab941 CH |
1903 | |
1904 | IHOLD(dp); | |
1905 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1da177e4 | 1906 | |
8f112e3b CH |
1907 | /* |
1908 | * If we're removing a directory perform some additional validation. | |
1909 | */ | |
1910 | if (is_dir) { | |
1911 | ASSERT(ip->i_d.di_nlink >= 2); | |
1912 | if (ip->i_d.di_nlink != 2) { | |
1913 | error = XFS_ERROR(ENOTEMPTY); | |
1914 | goto out_trans_cancel; | |
1915 | } | |
1916 | if (!xfs_dir_isempty(ip)) { | |
1917 | error = XFS_ERROR(ENOTEMPTY); | |
1918 | goto out_trans_cancel; | |
1919 | } | |
1920 | } | |
1921 | ||
9d87c319 | 1922 | xfs_bmap_init(&free_list, &first_block); |
556b8b16 | 1923 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
d4377d84 | 1924 | &first_block, &free_list, resblks); |
1da177e4 LT |
1925 | if (error) { |
1926 | ASSERT(error != ENOENT); | |
8f112e3b | 1927 | goto out_bmap_cancel; |
1da177e4 LT |
1928 | } |
1929 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1930 | ||
8f112e3b CH |
1931 | if (is_dir) { |
1932 | /* | |
1933 | * Drop the link from ip's "..". | |
1934 | */ | |
1935 | error = xfs_droplink(tp, dp); | |
1936 | if (error) | |
1937 | goto out_bmap_cancel; | |
1938 | ||
1939 | /* | |
2b7035fd | 1940 | * Drop the "." link from ip to self. |
8f112e3b CH |
1941 | */ |
1942 | error = xfs_droplink(tp, ip); | |
1943 | if (error) | |
1944 | goto out_bmap_cancel; | |
1945 | } else { | |
1946 | /* | |
1947 | * When removing a non-directory we need to log the parent | |
26c52951 DC |
1948 | * inode here. For a directory this is done implicitly |
1949 | * by the xfs_droplink call for the ".." entry. | |
8f112e3b CH |
1950 | */ |
1951 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1da177e4 LT |
1952 | } |
1953 | ||
8f112e3b | 1954 | /* |
2b7035fd | 1955 | * Drop the link from dp to ip. |
8f112e3b CH |
1956 | */ |
1957 | error = xfs_droplink(tp, ip); | |
1958 | if (error) | |
1959 | goto out_bmap_cancel; | |
1960 | ||
1961 | /* | |
1962 | * Determine if this is the last link while | |
1da177e4 LT |
1963 | * we are in the transaction. |
1964 | */ | |
8f112e3b | 1965 | link_zero = (ip->i_d.di_nlink == 0); |
1da177e4 | 1966 | |
1da177e4 LT |
1967 | /* |
1968 | * If this is a synchronous mount, make sure that the | |
1969 | * remove transaction goes to disk before returning to | |
1970 | * the user. | |
1971 | */ | |
8f112e3b | 1972 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 1973 | xfs_trans_set_sync(tp); |
1da177e4 | 1974 | |
f7c99b6f | 1975 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
8f112e3b CH |
1976 | if (error) |
1977 | goto out_bmap_cancel; | |
1da177e4 | 1978 | |
1c72bf90 | 1979 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
5df78e73 | 1980 | if (error) |
1da177e4 | 1981 | goto std_return; |
1da177e4 | 1982 | |
2a82b8be DC |
1983 | /* |
1984 | * If we are using filestreams, kill the stream association. | |
1985 | * If the file is still open it may get a new one but that | |
1986 | * will get killed on last close in xfs_close() so we don't | |
1987 | * have to worry about that. | |
1988 | */ | |
8f112e3b | 1989 | if (!is_dir && link_zero && xfs_inode_is_filestream(ip)) |
2a82b8be DC |
1990 | xfs_filestream_deassociate(ip); |
1991 | ||
cf441eeb | 1992 | xfs_itrace_exit(ip); |
8f112e3b | 1993 | xfs_itrace_exit(dp); |
1da177e4 | 1994 | |
1da177e4 | 1995 | std_return: |
eb9df39d | 1996 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
8f112e3b CH |
1997 | XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, dp, DM_RIGHT_NULL, |
1998 | NULL, DM_RIGHT_NULL, name->name, NULL, | |
1999 | ip->i_d.di_mode, error, 0); | |
1da177e4 | 2000 | } |
8f112e3b | 2001 | |
1da177e4 LT |
2002 | return error; |
2003 | ||
8f112e3b | 2004 | out_bmap_cancel: |
1da177e4 LT |
2005 | xfs_bmap_cancel(&free_list); |
2006 | cancel_flags |= XFS_TRANS_ABORT; | |
8f112e3b | 2007 | out_trans_cancel: |
1da177e4 LT |
2008 | xfs_trans_cancel(tp, cancel_flags); |
2009 | goto std_return; | |
1da177e4 LT |
2010 | } |
2011 | ||
993386c1 | 2012 | int |
1da177e4 | 2013 | xfs_link( |
993386c1 | 2014 | xfs_inode_t *tdp, |
a3da7896 | 2015 | xfs_inode_t *sip, |
556b8b16 | 2016 | struct xfs_name *target_name) |
1da177e4 | 2017 | { |
993386c1 | 2018 | xfs_mount_t *mp = tdp->i_mount; |
1da177e4 | 2019 | xfs_trans_t *tp; |
1da177e4 LT |
2020 | int error; |
2021 | xfs_bmap_free_t free_list; | |
2022 | xfs_fsblock_t first_block; | |
2023 | int cancel_flags; | |
2024 | int committed; | |
1da177e4 | 2025 | int resblks; |
1da177e4 | 2026 | |
cf441eeb | 2027 | xfs_itrace_entry(tdp); |
a3da7896 | 2028 | xfs_itrace_entry(sip); |
1da177e4 | 2029 | |
a3da7896 | 2030 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); |
1da177e4 | 2031 | |
1da177e4 LT |
2032 | if (XFS_FORCED_SHUTDOWN(mp)) |
2033 | return XFS_ERROR(EIO); | |
2034 | ||
eb9df39d | 2035 | if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) { |
1da177e4 | 2036 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, |
bc4ac74a CH |
2037 | tdp, DM_RIGHT_NULL, |
2038 | sip, DM_RIGHT_NULL, | |
556b8b16 | 2039 | target_name->name, NULL, 0, 0, 0); |
1da177e4 LT |
2040 | if (error) |
2041 | return error; | |
2042 | } | |
2043 | ||
2044 | /* Return through std_return after this point. */ | |
2045 | ||
7d095257 | 2046 | error = xfs_qm_dqattach(sip, 0); |
cb3f35bb CH |
2047 | if (error) |
2048 | goto std_return; | |
2049 | ||
7d095257 | 2050 | error = xfs_qm_dqattach(tdp, 0); |
1da177e4 LT |
2051 | if (error) |
2052 | goto std_return; | |
2053 | ||
2054 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2055 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
556b8b16 | 2056 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
1da177e4 LT |
2057 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, |
2058 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2059 | if (error == ENOSPC) { | |
2060 | resblks = 0; | |
2061 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2062 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2063 | } | |
2064 | if (error) { | |
2065 | cancel_flags = 0; | |
2066 | goto error_return; | |
2067 | } | |
2068 | ||
e1cccd91 | 2069 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2070 | |
2071 | /* | |
2072 | * Increment vnode ref counts since xfs_trans_commit & | |
2073 | * xfs_trans_cancel will both unlock the inodes and | |
2074 | * decrement the associated ref counts. | |
2075 | */ | |
a3da7896 CH |
2076 | IHOLD(sip); |
2077 | IHOLD(tdp); | |
1da177e4 LT |
2078 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); |
2079 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2080 | ||
2081 | /* | |
2082 | * If the source has too many links, we can't make any more to it. | |
2083 | */ | |
2084 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2085 | error = XFS_ERROR(EMLINK); | |
2086 | goto error_return; | |
2087 | } | |
2088 | ||
365ca83d NS |
2089 | /* |
2090 | * If we are using project inheritance, we only allow hard link | |
2091 | * creation in our tree when the project IDs are the same; else | |
2092 | * the tree quota mechanism could be circumvented. | |
2093 | */ | |
2094 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2095 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 2096 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
2097 | goto error_return; |
2098 | } | |
2099 | ||
556b8b16 BN |
2100 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); |
2101 | if (error) | |
1da177e4 LT |
2102 | goto error_return; |
2103 | ||
9d87c319 | 2104 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 2105 | |
556b8b16 BN |
2106 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
2107 | &first_block, &free_list, resblks); | |
1da177e4 LT |
2108 | if (error) |
2109 | goto abort_return; | |
2110 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1da177e4 LT |
2111 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); |
2112 | ||
2113 | error = xfs_bumplink(tp, sip); | |
b71d300c | 2114 | if (error) |
1da177e4 | 2115 | goto abort_return; |
1da177e4 LT |
2116 | |
2117 | /* | |
2118 | * If this is a synchronous mount, make sure that the | |
2119 | * link transaction goes to disk before returning to | |
2120 | * the user. | |
2121 | */ | |
2122 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2123 | xfs_trans_set_sync(tp); | |
2124 | } | |
2125 | ||
f7c99b6f | 2126 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2127 | if (error) { |
2128 | xfs_bmap_cancel(&free_list); | |
2129 | goto abort_return; | |
2130 | } | |
2131 | ||
1c72bf90 | 2132 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
b71d300c | 2133 | if (error) |
1da177e4 | 2134 | goto std_return; |
1da177e4 LT |
2135 | |
2136 | /* Fall through to std_return with error = 0. */ | |
2137 | std_return: | |
eb9df39d | 2138 | if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) { |
1da177e4 | 2139 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, |
bc4ac74a CH |
2140 | tdp, DM_RIGHT_NULL, |
2141 | sip, DM_RIGHT_NULL, | |
556b8b16 | 2142 | target_name->name, NULL, 0, error, 0); |
1da177e4 LT |
2143 | } |
2144 | return error; | |
2145 | ||
2146 | abort_return: | |
2147 | cancel_flags |= XFS_TRANS_ABORT; | |
2148 | /* FALLTHROUGH */ | |
014c2544 | 2149 | |
1da177e4 LT |
2150 | error_return: |
2151 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2152 | goto std_return; |
2153 | } | |
b71d300c | 2154 | |
993386c1 | 2155 | int |
1da177e4 | 2156 | xfs_symlink( |
993386c1 | 2157 | xfs_inode_t *dp, |
556b8b16 BN |
2158 | struct xfs_name *link_name, |
2159 | const char *target_path, | |
3e5daf05 | 2160 | mode_t mode, |
3937be5b | 2161 | xfs_inode_t **ipp, |
1da177e4 LT |
2162 | cred_t *credp) |
2163 | { | |
993386c1 | 2164 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 2165 | xfs_trans_t *tp; |
1da177e4 LT |
2166 | xfs_inode_t *ip; |
2167 | int error; | |
2168 | int pathlen; | |
2169 | xfs_bmap_free_t free_list; | |
2170 | xfs_fsblock_t first_block; | |
993386c1 | 2171 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2172 | uint cancel_flags; |
2173 | int committed; | |
2174 | xfs_fileoff_t first_fsb; | |
2175 | xfs_filblks_t fs_blocks; | |
2176 | int nmaps; | |
2177 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
2178 | xfs_daddr_t d; | |
556b8b16 | 2179 | const char *cur_chunk; |
1da177e4 LT |
2180 | int byte_cnt; |
2181 | int n; | |
2182 | xfs_buf_t *bp; | |
2183 | xfs_prid_t prid; | |
2184 | struct xfs_dquot *udqp, *gdqp; | |
2185 | uint resblks; | |
1da177e4 | 2186 | |
3937be5b | 2187 | *ipp = NULL; |
1da177e4 LT |
2188 | error = 0; |
2189 | ip = NULL; | |
2190 | tp = NULL; | |
2191 | ||
cf441eeb | 2192 | xfs_itrace_entry(dp); |
1da177e4 LT |
2193 | |
2194 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2195 | return XFS_ERROR(EIO); | |
2196 | ||
1da177e4 LT |
2197 | /* |
2198 | * Check component lengths of the target path name. | |
2199 | */ | |
2200 | pathlen = strlen(target_path); | |
2201 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
2202 | return XFS_ERROR(ENAMETOOLONG); | |
1da177e4 | 2203 | |
eb9df39d | 2204 | if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) { |
bc4ac74a | 2205 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp, |
1da177e4 | 2206 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, |
556b8b16 | 2207 | link_name->name, target_path, 0, 0, 0); |
1da177e4 LT |
2208 | if (error) |
2209 | return error; | |
2210 | } | |
2211 | ||
2212 | /* Return through std_return after this point. */ | |
2213 | ||
2214 | udqp = gdqp = NULL; | |
365ca83d NS |
2215 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2216 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
2217 | else |
2218 | prid = (xfs_prid_t)dfltprid; | |
2219 | ||
2220 | /* | |
2221 | * Make sure that we have allocated dquot(s) on disk. | |
2222 | */ | |
7d095257 | 2223 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
1da177e4 LT |
2224 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2225 | if (error) | |
2226 | goto std_return; | |
2227 | ||
2228 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
2229 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2230 | /* | |
2231 | * The symlink will fit into the inode data fork? | |
2232 | * There can't be any attributes so we get the whole variable part. | |
2233 | */ | |
2234 | if (pathlen <= XFS_LITINO(mp)) | |
2235 | fs_blocks = 0; | |
2236 | else | |
2237 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
556b8b16 | 2238 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks); |
1da177e4 LT |
2239 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, |
2240 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
2241 | if (error == ENOSPC && fs_blocks == 0) { | |
2242 | resblks = 0; | |
2243 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
2244 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
2245 | } | |
2246 | if (error) { | |
2247 | cancel_flags = 0; | |
1da177e4 LT |
2248 | goto error_return; |
2249 | } | |
2250 | ||
f7c66ce3 | 2251 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2252 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2253 | |
2254 | /* | |
2255 | * Check whether the directory allows new symlinks or not. | |
2256 | */ | |
2257 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
2258 | error = XFS_ERROR(EPERM); | |
2259 | goto error_return; | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * Reserve disk quota : blocks and inode. | |
2264 | */ | |
7d095257 | 2265 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 LT |
2266 | if (error) |
2267 | goto error_return; | |
2268 | ||
2269 | /* | |
2270 | * Check for ability to enter directory entry, if no space reserved. | |
2271 | */ | |
556b8b16 BN |
2272 | error = xfs_dir_canenter(tp, dp, link_name, resblks); |
2273 | if (error) | |
1da177e4 LT |
2274 | goto error_return; |
2275 | /* | |
2276 | * Initialize the bmap freelist prior to calling either | |
2277 | * bmapi or the directory create code. | |
2278 | */ | |
9d87c319 | 2279 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
2280 | |
2281 | /* | |
2282 | * Allocate an inode for the symlink. | |
2283 | */ | |
3e5daf05 | 2284 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), |
1da177e4 LT |
2285 | 1, 0, credp, prid, resblks > 0, &ip, NULL); |
2286 | if (error) { | |
2287 | if (error == ENOSPC) | |
2288 | goto error_return; | |
2289 | goto error1; | |
2290 | } | |
cf441eeb | 2291 | xfs_itrace_ref(ip); |
1da177e4 | 2292 | |
993386c1 CH |
2293 | /* |
2294 | * An error after we've joined dp to the transaction will result in the | |
2295 | * transaction cancel unlocking dp so don't do it explicitly in the | |
2296 | * error path. | |
2297 | */ | |
3937be5b | 2298 | IHOLD(dp); |
1da177e4 | 2299 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 2300 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2301 | |
2302 | /* | |
2303 | * Also attach the dquot(s) to it, if applicable. | |
2304 | */ | |
7d095257 | 2305 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 LT |
2306 | |
2307 | if (resblks) | |
2308 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
2309 | /* | |
2310 | * If the symlink will fit into the inode, write it inline. | |
2311 | */ | |
2312 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
2313 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
2314 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
2315 | ip->i_d.di_size = pathlen; | |
2316 | ||
2317 | /* | |
2318 | * The inode was initially created in extent format. | |
2319 | */ | |
2320 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
2321 | ip->i_df.if_flags |= XFS_IFINLINE; | |
2322 | ||
2323 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
2324 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
2325 | ||
2326 | } else { | |
2327 | first_fsb = 0; | |
2328 | nmaps = SYMLINK_MAPS; | |
2329 | ||
2330 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
2331 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
2332 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 2333 | &free_list, NULL); |
1da177e4 LT |
2334 | if (error) { |
2335 | goto error1; | |
2336 | } | |
2337 | ||
2338 | if (resblks) | |
2339 | resblks -= fs_blocks; | |
2340 | ip->i_d.di_size = pathlen; | |
2341 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2342 | ||
2343 | cur_chunk = target_path; | |
2344 | for (n = 0; n < nmaps; n++) { | |
2345 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
2346 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
2347 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
2348 | BTOBB(byte_cnt), 0); | |
2349 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
2350 | if (pathlen < byte_cnt) { | |
2351 | byte_cnt = pathlen; | |
2352 | } | |
2353 | pathlen -= byte_cnt; | |
2354 | ||
2355 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
2356 | cur_chunk += byte_cnt; | |
2357 | ||
2358 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
2359 | } | |
2360 | } | |
2361 | ||
2362 | /* | |
2363 | * Create the directory entry for the symlink. | |
2364 | */ | |
556b8b16 BN |
2365 | error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, |
2366 | &first_block, &free_list, resblks); | |
f6c2d1fa | 2367 | if (error) |
1da177e4 | 2368 | goto error1; |
1da177e4 LT |
2369 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
2370 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2371 | ||
1da177e4 LT |
2372 | /* |
2373 | * If this is a synchronous mount, make sure that the | |
2374 | * symlink transaction goes to disk before returning to | |
2375 | * the user. | |
2376 | */ | |
2377 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2378 | xfs_trans_set_sync(tp); | |
2379 | } | |
2380 | ||
2381 | /* | |
2382 | * xfs_trans_commit normally decrements the vnode ref count | |
2383 | * when it unlocks the inode. Since we want to return the | |
2384 | * vnode to the caller, we bump the vnode ref count now. | |
2385 | */ | |
2386 | IHOLD(ip); | |
2387 | ||
f7c99b6f | 2388 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2389 | if (error) { |
2390 | goto error2; | |
2391 | } | |
1c72bf90 | 2392 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
7d095257 CH |
2393 | xfs_qm_dqrele(udqp); |
2394 | xfs_qm_dqrele(gdqp); | |
1da177e4 LT |
2395 | |
2396 | /* Fall through to std_return with error = 0 or errno from | |
2397 | * xfs_trans_commit */ | |
2398 | std_return: | |
993386c1 | 2399 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) { |
1da177e4 | 2400 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, |
bc4ac74a CH |
2401 | dp, DM_RIGHT_NULL, |
2402 | error ? NULL : ip, | |
556b8b16 BN |
2403 | DM_RIGHT_NULL, link_name->name, |
2404 | target_path, 0, error, 0); | |
1da177e4 LT |
2405 | } |
2406 | ||
3937be5b CH |
2407 | if (!error) |
2408 | *ipp = ip; | |
1da177e4 LT |
2409 | return error; |
2410 | ||
2411 | error2: | |
2412 | IRELE(ip); | |
2413 | error1: | |
2414 | xfs_bmap_cancel(&free_list); | |
2415 | cancel_flags |= XFS_TRANS_ABORT; | |
2416 | error_return: | |
2417 | xfs_trans_cancel(tp, cancel_flags); | |
7d095257 CH |
2418 | xfs_qm_dqrele(udqp); |
2419 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 2420 | |
993386c1 | 2421 | if (unlock_dp_on_error) |
1da177e4 | 2422 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2423 | |
2424 | goto std_return; | |
2425 | } | |
2426 | ||
1da177e4 | 2427 | int |
993386c1 CH |
2428 | xfs_set_dmattrs( |
2429 | xfs_inode_t *ip, | |
1da177e4 | 2430 | u_int evmask, |
993386c1 | 2431 | u_int16_t state) |
1da177e4 | 2432 | { |
993386c1 | 2433 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 2434 | xfs_trans_t *tp; |
1da177e4 LT |
2435 | int error; |
2436 | ||
2437 | if (!capable(CAP_SYS_ADMIN)) | |
2438 | return XFS_ERROR(EPERM); | |
2439 | ||
1da177e4 LT |
2440 | if (XFS_FORCED_SHUTDOWN(mp)) |
2441 | return XFS_ERROR(EIO); | |
2442 | ||
2443 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
2444 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
2445 | if (error) { | |
2446 | xfs_trans_cancel(tp, 0); | |
2447 | return error; | |
2448 | } | |
2449 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2450 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2451 | ||
613d7043 CH |
2452 | ip->i_d.di_dmevmask = evmask; |
2453 | ip->i_d.di_dmstate = state; | |
1da177e4 LT |
2454 | |
2455 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2456 | IHOLD(ip); | |
1c72bf90 | 2457 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
2458 | |
2459 | return error; | |
2460 | } | |
2461 | ||
993386c1 | 2462 | int |
1da177e4 | 2463 | xfs_reclaim( |
993386c1 | 2464 | xfs_inode_t *ip) |
1da177e4 | 2465 | { |
1da177e4 | 2466 | |
cf441eeb | 2467 | xfs_itrace_entry(ip); |
1da177e4 | 2468 | |
df80c933 | 2469 | ASSERT(!VN_MAPPED(VFS_I(ip))); |
1da177e4 LT |
2470 | |
2471 | /* bad inode, get out here ASAP */ | |
cb4c8cc1 | 2472 | if (is_bad_inode(VFS_I(ip))) { |
1da177e4 LT |
2473 | xfs_ireclaim(ip); |
2474 | return 0; | |
2475 | } | |
2476 | ||
25e41b3d | 2477 | xfs_ioend_wait(ip); |
1da177e4 | 2478 | |
51c91ed5 | 2479 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); |
1da177e4 | 2480 | |
42fe2b1f CH |
2481 | /* |
2482 | * Make sure the atime in the XFS inode is correct before freeing the | |
2483 | * Linux inode. | |
2484 | */ | |
2485 | xfs_synchronize_atime(ip); | |
1da177e4 | 2486 | |
4c60658e DC |
2487 | /* |
2488 | * If we have nothing to flush with this inode then complete the | |
2489 | * teardown now, otherwise break the link between the xfs inode and the | |
2490 | * linux inode and clean up the xfs inode later. This avoids flushing | |
2491 | * the inode to disk during the delete operation itself. | |
2492 | * | |
2493 | * When breaking the link, we need to set the XFS_IRECLAIMABLE flag | |
2494 | * first to ensure that xfs_iunpin() will never see an xfs inode | |
2495 | * that has a linux inode being reclaimed. Synchronisation is provided | |
2496 | * by the i_flags_lock. | |
1da177e4 LT |
2497 | */ |
2498 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
2499 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2500 | xfs_iflock(ip); | |
bf904248 | 2501 | xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
1dc3318a | 2502 | return xfs_reclaim_inode(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); |
1da177e4 | 2503 | } |
11654513 | 2504 | xfs_inode_set_reclaim_tag(ip); |
1da177e4 LT |
2505 | return 0; |
2506 | } | |
1da177e4 LT |
2507 | |
2508 | /* | |
2509 | * xfs_alloc_file_space() | |
2510 | * This routine allocates disk space for the given file. | |
2511 | * | |
2512 | * If alloc_type == 0, this request is for an ALLOCSP type | |
2513 | * request which will change the file size. In this case, no | |
2514 | * DMAPI event will be generated by the call. A TRUNCATE event | |
2515 | * will be generated later by xfs_setattr. | |
2516 | * | |
2517 | * If alloc_type != 0, this request is for a RESVSP type | |
2518 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
2519 | * lower block boundary byte address is less than the file's | |
2520 | * length. | |
2521 | * | |
2522 | * RETURNS: | |
2523 | * 0 on success | |
2524 | * errno on error | |
2525 | * | |
2526 | */ | |
ba0f32d4 | 2527 | STATIC int |
1da177e4 LT |
2528 | xfs_alloc_file_space( |
2529 | xfs_inode_t *ip, | |
2530 | xfs_off_t offset, | |
2531 | xfs_off_t len, | |
2532 | int alloc_type, | |
2533 | int attr_flags) | |
2534 | { | |
dd9f438e NS |
2535 | xfs_mount_t *mp = ip->i_mount; |
2536 | xfs_off_t count; | |
1da177e4 LT |
2537 | xfs_filblks_t allocated_fsb; |
2538 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
2539 | xfs_extlen_t extsz, temp; |
2540 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 2541 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
2542 | int nimaps; |
2543 | int bmapi_flag; | |
2544 | int quota_flag; | |
1da177e4 | 2545 | int rt; |
1da177e4 | 2546 | xfs_trans_t *tp; |
dd9f438e NS |
2547 | xfs_bmbt_irec_t imaps[1], *imapp; |
2548 | xfs_bmap_free_t free_list; | |
2549 | uint qblocks, resblks, resrtextents; | |
2550 | int committed; | |
2551 | int error; | |
1da177e4 | 2552 | |
cf441eeb | 2553 | xfs_itrace_entry(ip); |
1da177e4 LT |
2554 | |
2555 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2556 | return XFS_ERROR(EIO); | |
2557 | ||
7d095257 CH |
2558 | error = xfs_qm_dqattach(ip, 0); |
2559 | if (error) | |
1da177e4 LT |
2560 | return error; |
2561 | ||
2562 | if (len <= 0) | |
2563 | return XFS_ERROR(EINVAL); | |
2564 | ||
957d0ebe DC |
2565 | rt = XFS_IS_REALTIME_INODE(ip); |
2566 | extsz = xfs_get_extsz_hint(ip); | |
2567 | ||
1da177e4 | 2568 | count = len; |
1da177e4 | 2569 | imapp = &imaps[0]; |
dd9f438e NS |
2570 | nimaps = 1; |
2571 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
2572 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
2573 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
2574 | ||
2575 | /* Generate a DMAPI event if needed. */ | |
ba87ea69 | 2576 | if (alloc_type != 0 && offset < ip->i_size && |
0f285c8a | 2577 | (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2578 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
1da177e4 LT |
2579 | xfs_off_t end_dmi_offset; |
2580 | ||
2581 | end_dmi_offset = offset+len; | |
ba87ea69 LM |
2582 | if (end_dmi_offset > ip->i_size) |
2583 | end_dmi_offset = ip->i_size; | |
bc4ac74a CH |
2584 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset, |
2585 | end_dmi_offset - offset, 0, NULL); | |
1da177e4 | 2586 | if (error) |
014c2544 | 2587 | return error; |
1da177e4 LT |
2588 | } |
2589 | ||
2590 | /* | |
dd9f438e | 2591 | * Allocate file space until done or until there is an error |
1da177e4 LT |
2592 | */ |
2593 | retry: | |
2594 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
2595 | xfs_fileoff_t s, e; |
2596 | ||
1da177e4 | 2597 | /* |
3ddb8fa9 | 2598 | * Determine space reservations for data/realtime. |
1da177e4 | 2599 | */ |
dd9f438e | 2600 | if (unlikely(extsz)) { |
1da177e4 | 2601 | s = startoffset_fsb; |
dd9f438e NS |
2602 | do_div(s, extsz); |
2603 | s *= extsz; | |
2604 | e = startoffset_fsb + allocatesize_fsb; | |
2605 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
2606 | e += temp; | |
2607 | if ((temp = do_mod(e, extsz))) | |
2608 | e += extsz - temp; | |
2609 | } else { | |
2610 | s = 0; | |
2611 | e = allocatesize_fsb; | |
2612 | } | |
2613 | ||
2614 | if (unlikely(rt)) { | |
2615 | resrtextents = qblocks = (uint)(e - s); | |
2616 | resrtextents /= mp->m_sb.sb_rextsize; | |
2617 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
2618 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 2619 | } else { |
dd9f438e NS |
2620 | resrtextents = 0; |
2621 | resblks = qblocks = \ | |
2622 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
2623 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
2624 | } |
2625 | ||
2626 | /* | |
dd9f438e | 2627 | * Allocate and setup the transaction. |
1da177e4 LT |
2628 | */ |
2629 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
2630 | error = xfs_trans_reserve(tp, resblks, |
2631 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
2632 | XFS_TRANS_PERM_LOG_RES, |
2633 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 2634 | /* |
dd9f438e | 2635 | * Check for running out of space |
1da177e4 LT |
2636 | */ |
2637 | if (error) { | |
2638 | /* | |
2639 | * Free the transaction structure. | |
2640 | */ | |
2641 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
2642 | xfs_trans_cancel(tp, 0); | |
2643 | break; | |
2644 | } | |
2645 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
2646 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, |
2647 | 0, quota_flag); | |
1da177e4 LT |
2648 | if (error) |
2649 | goto error1; | |
2650 | ||
2651 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2652 | xfs_trans_ihold(tp, ip); | |
2653 | ||
2654 | /* | |
dd9f438e | 2655 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 | 2656 | */ |
9d87c319 | 2657 | xfs_bmap_init(&free_list, &firstfsb); |
541d7d3c | 2658 | error = xfs_bmapi(tp, ip, startoffset_fsb, |
dd9f438e NS |
2659 | allocatesize_fsb, bmapi_flag, |
2660 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 2661 | &free_list, NULL); |
1da177e4 LT |
2662 | if (error) { |
2663 | goto error0; | |
2664 | } | |
2665 | ||
2666 | /* | |
dd9f438e | 2667 | * Complete the transaction |
1da177e4 | 2668 | */ |
f7c99b6f | 2669 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2670 | if (error) { |
2671 | goto error0; | |
2672 | } | |
2673 | ||
1c72bf90 | 2674 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2675 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2676 | if (error) { | |
2677 | break; | |
2678 | } | |
2679 | ||
2680 | allocated_fsb = imapp->br_blockcount; | |
2681 | ||
dd9f438e | 2682 | if (nimaps == 0) { |
1da177e4 LT |
2683 | error = XFS_ERROR(ENOSPC); |
2684 | break; | |
2685 | } | |
2686 | ||
2687 | startoffset_fsb += allocated_fsb; | |
2688 | allocatesize_fsb -= allocated_fsb; | |
2689 | } | |
2690 | dmapi_enospc_check: | |
0f285c8a | 2691 | if (error == ENOSPC && (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2692 | DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) { |
1da177e4 | 2693 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, |
bc4ac74a CH |
2694 | ip, DM_RIGHT_NULL, |
2695 | ip, DM_RIGHT_NULL, | |
1da177e4 LT |
2696 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ |
2697 | if (error == 0) | |
2698 | goto retry; /* Maybe DMAPI app. has made space */ | |
2699 | /* else fall through with error from XFS_SEND_DATA */ | |
2700 | } | |
2701 | ||
2702 | return error; | |
2703 | ||
dd9f438e | 2704 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 2705 | xfs_bmap_cancel(&free_list); |
7d095257 | 2706 | xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e NS |
2707 | |
2708 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
2709 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
2710 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2711 | goto dmapi_enospc_check; | |
2712 | } | |
2713 | ||
2714 | /* | |
2715 | * Zero file bytes between startoff and endoff inclusive. | |
2716 | * The iolock is held exclusive and no blocks are buffered. | |
2fd6f6ec LM |
2717 | * |
2718 | * This function is used by xfs_free_file_space() to zero | |
2719 | * partial blocks when the range to free is not block aligned. | |
2720 | * When unreserving space with boundaries that are not block | |
2721 | * aligned we round up the start and round down the end | |
2722 | * boundaries and then use this function to zero the parts of | |
2723 | * the blocks that got dropped during the rounding. | |
1da177e4 LT |
2724 | */ |
2725 | STATIC int | |
2726 | xfs_zero_remaining_bytes( | |
2727 | xfs_inode_t *ip, | |
2728 | xfs_off_t startoff, | |
2729 | xfs_off_t endoff) | |
2730 | { | |
2731 | xfs_bmbt_irec_t imap; | |
2732 | xfs_fileoff_t offset_fsb; | |
2733 | xfs_off_t lastoffset; | |
2734 | xfs_off_t offset; | |
2735 | xfs_buf_t *bp; | |
2736 | xfs_mount_t *mp = ip->i_mount; | |
2737 | int nimap; | |
2738 | int error = 0; | |
2739 | ||
2fd6f6ec LM |
2740 | /* |
2741 | * Avoid doing I/O beyond eof - it's not necessary | |
2742 | * since nothing can read beyond eof. The space will | |
2743 | * be zeroed when the file is extended anyway. | |
2744 | */ | |
2745 | if (startoff >= ip->i_size) | |
2746 | return 0; | |
2747 | ||
2748 | if (endoff > ip->i_size) | |
2749 | endoff = ip->i_size; | |
2750 | ||
1da177e4 | 2751 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, |
71ddabb9 | 2752 | XFS_IS_REALTIME_INODE(ip) ? |
1da177e4 | 2753 | mp->m_rtdev_targp : mp->m_ddev_targp); |
c6422617 LM |
2754 | if (!bp) |
2755 | return XFS_ERROR(ENOMEM); | |
1da177e4 LT |
2756 | |
2757 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
2758 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
2759 | nimap = 1; | |
541d7d3c | 2760 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, |
3e57ecf6 | 2761 | NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2762 | if (error || nimap < 1) |
2763 | break; | |
2764 | ASSERT(imap.br_blockcount >= 1); | |
2765 | ASSERT(imap.br_startoff == offset_fsb); | |
2766 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
2767 | if (lastoffset > endoff) | |
2768 | lastoffset = endoff; | |
2769 | if (imap.br_startblock == HOLESTARTBLOCK) | |
2770 | continue; | |
2771 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2772 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
2773 | continue; | |
2774 | XFS_BUF_UNDONE(bp); | |
2775 | XFS_BUF_UNWRITE(bp); | |
2776 | XFS_BUF_READ(bp); | |
9d87c319 | 2777 | XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock)); |
1da177e4 | 2778 | xfsbdstrat(mp, bp); |
d64e31a2 DC |
2779 | error = xfs_iowait(bp); |
2780 | if (error) { | |
1da177e4 LT |
2781 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", |
2782 | mp, bp, XFS_BUF_ADDR(bp)); | |
2783 | break; | |
2784 | } | |
2785 | memset(XFS_BUF_PTR(bp) + | |
2786 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
2787 | 0, lastoffset - offset + 1); | |
2788 | XFS_BUF_UNDONE(bp); | |
2789 | XFS_BUF_UNREAD(bp); | |
2790 | XFS_BUF_WRITE(bp); | |
2791 | xfsbdstrat(mp, bp); | |
d64e31a2 DC |
2792 | error = xfs_iowait(bp); |
2793 | if (error) { | |
1da177e4 LT |
2794 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", |
2795 | mp, bp, XFS_BUF_ADDR(bp)); | |
2796 | break; | |
2797 | } | |
2798 | } | |
2799 | xfs_buf_free(bp); | |
2800 | return error; | |
2801 | } | |
2802 | ||
2803 | /* | |
2804 | * xfs_free_file_space() | |
2805 | * This routine frees disk space for the given file. | |
2806 | * | |
2807 | * This routine is only called by xfs_change_file_space | |
2808 | * for an UNRESVSP type call. | |
2809 | * | |
2810 | * RETURNS: | |
2811 | * 0 on success | |
2812 | * errno on error | |
2813 | * | |
2814 | */ | |
2815 | STATIC int | |
2816 | xfs_free_file_space( | |
2817 | xfs_inode_t *ip, | |
2818 | xfs_off_t offset, | |
2819 | xfs_off_t len, | |
2820 | int attr_flags) | |
2821 | { | |
2822 | int committed; | |
2823 | int done; | |
2824 | xfs_off_t end_dmi_offset; | |
2825 | xfs_fileoff_t endoffset_fsb; | |
2826 | int error; | |
2827 | xfs_fsblock_t firstfsb; | |
2828 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
2829 | xfs_bmbt_irec_t imap; |
2830 | xfs_off_t ioffset; | |
2831 | xfs_extlen_t mod=0; | |
2832 | xfs_mount_t *mp; | |
2833 | int nimap; | |
2834 | uint resblks; | |
673cdf5c | 2835 | uint rounding; |
1da177e4 LT |
2836 | int rt; |
2837 | xfs_fileoff_t startoffset_fsb; | |
2838 | xfs_trans_t *tp; | |
5fcbab35 | 2839 | int need_iolock = 1; |
1da177e4 | 2840 | |
1da177e4 LT |
2841 | mp = ip->i_mount; |
2842 | ||
cf441eeb | 2843 | xfs_itrace_entry(ip); |
bd5a876a | 2844 | |
7d095257 CH |
2845 | error = xfs_qm_dqattach(ip, 0); |
2846 | if (error) | |
1da177e4 LT |
2847 | return error; |
2848 | ||
2849 | error = 0; | |
2850 | if (len <= 0) /* if nothing being freed */ | |
2851 | return error; | |
71ddabb9 | 2852 | rt = XFS_IS_REALTIME_INODE(ip); |
1da177e4 LT |
2853 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); |
2854 | end_dmi_offset = offset + len; | |
2855 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
2856 | ||
0f285c8a | 2857 | if (offset < ip->i_size && (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2858 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
ba87ea69 LM |
2859 | if (end_dmi_offset > ip->i_size) |
2860 | end_dmi_offset = ip->i_size; | |
bc4ac74a | 2861 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, |
1da177e4 LT |
2862 | offset, end_dmi_offset - offset, |
2863 | AT_DELAY_FLAG(attr_flags), NULL); | |
2864 | if (error) | |
014c2544 | 2865 | return error; |
1da177e4 LT |
2866 | } |
2867 | ||
0f285c8a | 2868 | if (attr_flags & XFS_ATTR_NOLOCK) |
5fcbab35 | 2869 | need_iolock = 0; |
9fa8046f | 2870 | if (need_iolock) { |
1da177e4 | 2871 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
25e41b3d CH |
2872 | /* wait for the completion of any pending DIOs */ |
2873 | xfs_ioend_wait(ip); | |
9fa8046f | 2874 | } |
5fcbab35 | 2875 | |
e6a4b37f | 2876 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1da177e4 | 2877 | ioffset = offset & ~(rounding - 1); |
bd5a876a | 2878 | |
df80c933 | 2879 | if (VN_CACHED(VFS_I(ip)) != 0) { |
e6a4b37f TS |
2880 | xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1); |
2881 | error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED); | |
d3cf2094 LM |
2882 | if (error) |
2883 | goto out_unlock_iolock; | |
bd5a876a CH |
2884 | } |
2885 | ||
1da177e4 LT |
2886 | /* |
2887 | * Need to zero the stuff we're not freeing, on disk. | |
9da096fd | 2888 | * If it's a realtime file & can't use unwritten extents then we |
1da177e4 LT |
2889 | * actually need to zero the extent edges. Otherwise xfs_bunmapi |
2890 | * will take care of it for us. | |
2891 | */ | |
62118709 | 2892 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
1da177e4 | 2893 | nimap = 1; |
541d7d3c | 2894 | error = xfs_bmapi(NULL, ip, startoffset_fsb, |
3e57ecf6 | 2895 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2896 | if (error) |
2897 | goto out_unlock_iolock; | |
2898 | ASSERT(nimap == 0 || nimap == 1); | |
2899 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
2900 | xfs_daddr_t block; | |
2901 | ||
2902 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2903 | block = imap.br_startblock; | |
2904 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
2905 | if (mod) | |
2906 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
2907 | } | |
2908 | nimap = 1; | |
541d7d3c | 2909 | error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, |
3e57ecf6 | 2910 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2911 | if (error) |
2912 | goto out_unlock_iolock; | |
2913 | ASSERT(nimap == 0 || nimap == 1); | |
2914 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
2915 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2916 | mod++; | |
2917 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
2918 | endoffset_fsb -= mod; | |
2919 | } | |
2920 | } | |
2921 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
2922 | /* | |
2923 | * One contiguous piece to clear | |
2924 | */ | |
2925 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
2926 | else { | |
2927 | /* | |
2928 | * Some full blocks, possibly two pieces to clear | |
2929 | */ | |
2930 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
2931 | error = xfs_zero_remaining_bytes(ip, offset, | |
2932 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
2933 | if (!error && | |
2934 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
2935 | error = xfs_zero_remaining_bytes(ip, | |
2936 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
2937 | offset + len - 1); | |
2938 | } | |
2939 | ||
2940 | /* | |
2941 | * free file space until done or until there is an error | |
2942 | */ | |
2943 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
2944 | while (!error && !done) { | |
2945 | ||
2946 | /* | |
91ebecc7 DC |
2947 | * allocate and setup the transaction. Allow this |
2948 | * transaction to dip into the reserve blocks to ensure | |
2949 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
2950 | */ |
2951 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 2952 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
2953 | error = xfs_trans_reserve(tp, |
2954 | resblks, | |
2955 | XFS_WRITE_LOG_RES(mp), | |
2956 | 0, | |
2957 | XFS_TRANS_PERM_LOG_RES, | |
2958 | XFS_WRITE_LOG_COUNT); | |
2959 | ||
2960 | /* | |
2961 | * check for running out of space | |
2962 | */ | |
2963 | if (error) { | |
2964 | /* | |
2965 | * Free the transaction structure. | |
2966 | */ | |
2967 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
2968 | xfs_trans_cancel(tp, 0); | |
2969 | break; | |
2970 | } | |
2971 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
2972 | error = xfs_trans_reserve_quota(tp, mp, |
2973 | ip->i_udquot, ip->i_gdquot, | |
2974 | resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
2975 | if (error) |
2976 | goto error1; | |
2977 | ||
2978 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2979 | xfs_trans_ihold(tp, ip); | |
2980 | ||
2981 | /* | |
2982 | * issue the bunmapi() call to free the blocks | |
2983 | */ | |
9d87c319 | 2984 | xfs_bmap_init(&free_list, &firstfsb); |
541d7d3c | 2985 | error = xfs_bunmapi(tp, ip, startoffset_fsb, |
1da177e4 | 2986 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 2987 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
2988 | if (error) { |
2989 | goto error0; | |
2990 | } | |
2991 | ||
2992 | /* | |
2993 | * complete the transaction | |
2994 | */ | |
f7c99b6f | 2995 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2996 | if (error) { |
2997 | goto error0; | |
2998 | } | |
2999 | ||
1c72bf90 | 3000 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3001 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
3002 | } | |
3003 | ||
3004 | out_unlock_iolock: | |
3005 | if (need_iolock) | |
3006 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
3007 | return error; | |
3008 | ||
3009 | error0: | |
3010 | xfs_bmap_cancel(&free_list); | |
3011 | error1: | |
3012 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
3013 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
3014 | XFS_ILOCK_EXCL); | |
3015 | return error; | |
3016 | } | |
3017 | ||
3018 | /* | |
3019 | * xfs_change_file_space() | |
3020 | * This routine allocates or frees disk space for the given file. | |
3021 | * The user specified parameters are checked for alignment and size | |
3022 | * limitations. | |
3023 | * | |
3024 | * RETURNS: | |
3025 | * 0 on success | |
3026 | * errno on error | |
3027 | * | |
3028 | */ | |
3029 | int | |
3030 | xfs_change_file_space( | |
993386c1 | 3031 | xfs_inode_t *ip, |
1da177e4 LT |
3032 | int cmd, |
3033 | xfs_flock64_t *bf, | |
3034 | xfs_off_t offset, | |
1da177e4 LT |
3035 | int attr_flags) |
3036 | { | |
993386c1 | 3037 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
3038 | int clrprealloc; |
3039 | int error; | |
3040 | xfs_fsize_t fsize; | |
1da177e4 LT |
3041 | int setprealloc; |
3042 | xfs_off_t startoffset; | |
3043 | xfs_off_t llen; | |
3044 | xfs_trans_t *tp; | |
0f285c8a | 3045 | struct iattr iattr; |
1da177e4 | 3046 | |
cf441eeb | 3047 | xfs_itrace_entry(ip); |
1da177e4 | 3048 | |
993386c1 | 3049 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
3050 | return XFS_ERROR(EINVAL); |
3051 | ||
1da177e4 LT |
3052 | switch (bf->l_whence) { |
3053 | case 0: /*SEEK_SET*/ | |
3054 | break; | |
3055 | case 1: /*SEEK_CUR*/ | |
3056 | bf->l_start += offset; | |
3057 | break; | |
3058 | case 2: /*SEEK_END*/ | |
ba87ea69 | 3059 | bf->l_start += ip->i_size; |
1da177e4 LT |
3060 | break; |
3061 | default: | |
3062 | return XFS_ERROR(EINVAL); | |
3063 | } | |
3064 | ||
3065 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
3066 | ||
3067 | if ( (bf->l_start < 0) | |
3068 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
3069 | || (bf->l_start + llen < 0) | |
3070 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
3071 | return XFS_ERROR(EINVAL); | |
3072 | ||
3073 | bf->l_whence = 0; | |
3074 | ||
3075 | startoffset = bf->l_start; | |
ba87ea69 | 3076 | fsize = ip->i_size; |
1da177e4 LT |
3077 | |
3078 | /* | |
3079 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
3080 | * file space. | |
3081 | * These calls do NOT zero the data space allocated to the file, | |
3082 | * nor do they change the file size. | |
3083 | * | |
3084 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
3085 | * space. | |
3086 | * These calls cause the new file data to be zeroed and the file | |
3087 | * size to be changed. | |
3088 | */ | |
3089 | setprealloc = clrprealloc = 0; | |
3090 | ||
3091 | switch (cmd) { | |
3092 | case XFS_IOC_RESVSP: | |
3093 | case XFS_IOC_RESVSP64: | |
3094 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
3095 | 1, attr_flags); | |
3096 | if (error) | |
3097 | return error; | |
3098 | setprealloc = 1; | |
3099 | break; | |
3100 | ||
3101 | case XFS_IOC_UNRESVSP: | |
3102 | case XFS_IOC_UNRESVSP64: | |
3103 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
3104 | attr_flags))) | |
3105 | return error; | |
3106 | break; | |
3107 | ||
3108 | case XFS_IOC_ALLOCSP: | |
3109 | case XFS_IOC_ALLOCSP64: | |
3110 | case XFS_IOC_FREESP: | |
3111 | case XFS_IOC_FREESP64: | |
3112 | if (startoffset > fsize) { | |
3113 | error = xfs_alloc_file_space(ip, fsize, | |
3114 | startoffset - fsize, 0, attr_flags); | |
3115 | if (error) | |
3116 | break; | |
3117 | } | |
3118 | ||
0f285c8a CH |
3119 | iattr.ia_valid = ATTR_SIZE; |
3120 | iattr.ia_size = startoffset; | |
1da177e4 | 3121 | |
ea5a3dc8 | 3122 | error = xfs_setattr(ip, &iattr, attr_flags); |
1da177e4 LT |
3123 | |
3124 | if (error) | |
3125 | return error; | |
3126 | ||
3127 | clrprealloc = 1; | |
3128 | break; | |
3129 | ||
3130 | default: | |
3131 | ASSERT(0); | |
3132 | return XFS_ERROR(EINVAL); | |
3133 | } | |
3134 | ||
3135 | /* | |
3136 | * update the inode timestamp, mode, and prealloc flag bits | |
3137 | */ | |
3138 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
3139 | ||
3140 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
3141 | 0, 0, 0))) { | |
3142 | /* ASSERT(0); */ | |
3143 | xfs_trans_cancel(tp, 0); | |
3144 | return error; | |
3145 | } | |
3146 | ||
3147 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3148 | ||
3149 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3150 | xfs_trans_ihold(tp, ip); | |
3151 | ||
0f285c8a | 3152 | if ((attr_flags & XFS_ATTR_DMI) == 0) { |
1da177e4 LT |
3153 | ip->i_d.di_mode &= ~S_ISUID; |
3154 | ||
3155 | /* | |
3156 | * Note that we don't have to worry about mandatory | |
3157 | * file locking being disabled here because we only | |
3158 | * clear the S_ISGID bit if the Group execute bit is | |
3159 | * on, but if it was on then mandatory locking wouldn't | |
3160 | * have been enabled. | |
3161 | */ | |
3162 | if (ip->i_d.di_mode & S_IXGRP) | |
3163 | ip->i_d.di_mode &= ~S_ISGID; | |
3164 | ||
3165 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3166 | } | |
3167 | if (setprealloc) | |
3168 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
3169 | else if (clrprealloc) | |
3170 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
3171 | ||
3172 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3173 | xfs_trans_set_sync(tp); | |
3174 | ||
1c72bf90 | 3175 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
3176 | |
3177 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3178 | ||
3179 | return error; | |
3180 | } |