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" |
1da177e4 LT |
45 | #include "xfs_attr.h" |
46 | #include "xfs_rw.h" | |
1da177e4 | 47 | #include "xfs_error.h" |
1da177e4 LT |
48 | #include "xfs_quota.h" |
49 | #include "xfs_utils.h" | |
a844f451 | 50 | #include "xfs_rtalloc.h" |
1da177e4 | 51 | #include "xfs_trans_space.h" |
1da177e4 | 52 | #include "xfs_log_priv.h" |
2a82b8be | 53 | #include "xfs_filestream.h" |
993386c1 | 54 | #include "xfs_vnodeops.h" |
1da177e4 | 55 | |
1da177e4 LT |
56 | int |
57 | xfs_setattr( | |
0f285c8a CH |
58 | struct xfs_inode *ip, |
59 | struct iattr *iattr, | |
ea5a3dc8 | 60 | int flags) |
1da177e4 | 61 | { |
993386c1 | 62 | xfs_mount_t *mp = ip->i_mount; |
e4f75291 | 63 | struct inode *inode = VFS_I(ip); |
0f285c8a | 64 | int mask = iattr->ia_valid; |
1da177e4 | 65 | xfs_trans_t *tp; |
1da177e4 LT |
66 | int code; |
67 | uint lock_flags; | |
68 | uint commit_flags=0; | |
69 | uid_t uid=0, iuid=0; | |
70 | gid_t gid=0, igid=0; | |
71 | int timeflags = 0; | |
1da177e4 | 72 | struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2; |
5fcbab35 | 73 | int need_iolock = 1; |
1da177e4 | 74 | |
cf441eeb | 75 | xfs_itrace_entry(ip); |
1da177e4 | 76 | |
bd186aa9 | 77 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
78 | return XFS_ERROR(EROFS); |
79 | ||
1da177e4 LT |
80 | if (XFS_FORCED_SHUTDOWN(mp)) |
81 | return XFS_ERROR(EIO); | |
82 | ||
c4cd747e CH |
83 | code = -inode_change_ok(inode, iattr); |
84 | if (code) | |
85 | return code; | |
86 | ||
1da177e4 LT |
87 | olddquot1 = olddquot2 = NULL; |
88 | udqp = gdqp = NULL; | |
89 | ||
90 | /* | |
91 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
92 | * before we start any other transactions. Trying to do this later | |
93 | * is messy. We don't care to take a readlock to look at the ids | |
94 | * in inode here, because we can't hold it across the trans_reserve. | |
95 | * If the IDs do change before we take the ilock, we're covered | |
96 | * because the i_*dquot fields will get updated anyway. | |
97 | */ | |
0f285c8a | 98 | if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { |
1da177e4 LT |
99 | uint qflags = 0; |
100 | ||
0f285c8a CH |
101 | if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { |
102 | uid = iattr->ia_uid; | |
1da177e4 LT |
103 | qflags |= XFS_QMOPT_UQUOTA; |
104 | } else { | |
105 | uid = ip->i_d.di_uid; | |
106 | } | |
0f285c8a CH |
107 | if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { |
108 | gid = iattr->ia_gid; | |
1da177e4 LT |
109 | qflags |= XFS_QMOPT_GQUOTA; |
110 | } else { | |
111 | gid = ip->i_d.di_gid; | |
112 | } | |
25fe55e8 | 113 | |
1da177e4 LT |
114 | /* |
115 | * We take a reference when we initialize udqp and gdqp, | |
116 | * so it is important that we never blindly double trip on | |
117 | * the same variable. See xfs_create() for an example. | |
118 | */ | |
119 | ASSERT(udqp == NULL); | |
120 | ASSERT(gdqp == NULL); | |
7d095257 | 121 | code = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid, |
25fe55e8 | 122 | qflags, &udqp, &gdqp); |
1da177e4 | 123 | if (code) |
014c2544 | 124 | return code; |
1da177e4 LT |
125 | } |
126 | ||
127 | /* | |
128 | * For the other attributes, we acquire the inode lock and | |
129 | * first do an error checking pass. | |
130 | */ | |
131 | tp = NULL; | |
132 | lock_flags = XFS_ILOCK_EXCL; | |
0f285c8a | 133 | if (flags & XFS_ATTR_NOLOCK) |
5fcbab35 | 134 | need_iolock = 0; |
0f285c8a CH |
135 | if (!(mask & ATTR_SIZE)) { |
136 | if ((mask != (ATTR_CTIME|ATTR_ATIME|ATTR_MTIME)) || | |
1da177e4 LT |
137 | (mp->m_flags & XFS_MOUNT_WSYNC)) { |
138 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); | |
139 | commit_flags = 0; | |
140 | if ((code = xfs_trans_reserve(tp, 0, | |
141 | XFS_ICHANGE_LOG_RES(mp), 0, | |
142 | 0, 0))) { | |
143 | lock_flags = 0; | |
144 | goto error_return; | |
145 | } | |
146 | } | |
147 | } else { | |
eb9df39d | 148 | if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) && |
0f285c8a | 149 | !(flags & XFS_ATTR_DMI)) { |
1da177e4 | 150 | int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR; |
bc4ac74a | 151 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip, |
0f285c8a | 152 | iattr->ia_size, 0, dmflags, NULL); |
1da177e4 LT |
153 | if (code) { |
154 | lock_flags = 0; | |
155 | goto error_return; | |
156 | } | |
157 | } | |
158 | if (need_iolock) | |
159 | lock_flags |= XFS_IOLOCK_EXCL; | |
160 | } | |
161 | ||
162 | xfs_ilock(ip, lock_flags); | |
163 | ||
1da177e4 LT |
164 | /* |
165 | * Change file ownership. Must be the owner or privileged. | |
1da177e4 | 166 | */ |
0f285c8a | 167 | if (mask & (ATTR_UID|ATTR_GID)) { |
1da177e4 LT |
168 | /* |
169 | * These IDs could have changed since we last looked at them. | |
170 | * But, we're assured that if the ownership did change | |
171 | * while we didn't have the inode locked, inode's dquot(s) | |
172 | * would have changed also. | |
173 | */ | |
174 | iuid = ip->i_d.di_uid; | |
1da177e4 | 175 | igid = ip->i_d.di_gid; |
0f285c8a CH |
176 | gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; |
177 | uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; | |
1da177e4 | 178 | |
1da177e4 | 179 | /* |
25fe55e8 | 180 | * Do a quota reservation only if uid/gid is actually |
1da177e4 LT |
181 | * going to change. |
182 | */ | |
7d095257 CH |
183 | if (XFS_IS_QUOTA_RUNNING(mp) && |
184 | ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || | |
185 | (XFS_IS_GQUOTA_ON(mp) && igid != gid))) { | |
1da177e4 | 186 | ASSERT(tp); |
7d095257 | 187 | code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp, |
1da177e4 LT |
188 | capable(CAP_FOWNER) ? |
189 | XFS_QMOPT_FORCE_RES : 0); | |
190 | if (code) /* out of quota */ | |
191 | goto error_return; | |
192 | } | |
193 | } | |
194 | ||
195 | /* | |
196 | * Truncate file. Must have write permission and not be a directory. | |
197 | */ | |
0f285c8a | 198 | if (mask & ATTR_SIZE) { |
1da177e4 | 199 | /* Short circuit the truncate case for zero length files */ |
0f285c8a CH |
200 | if (iattr->ia_size == 0 && |
201 | ip->i_size == 0 && ip->i_d.di_nextents == 0) { | |
1da177e4 LT |
202 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
203 | lock_flags &= ~XFS_ILOCK_EXCL; | |
0f285c8a | 204 | if (mask & ATTR_CTIME) |
1da177e4 LT |
205 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
206 | code = 0; | |
207 | goto error_return; | |
208 | } | |
209 | ||
42173f68 | 210 | if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
211 | code = XFS_ERROR(EISDIR); |
212 | goto error_return; | |
42173f68 | 213 | } else if (!S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
214 | code = XFS_ERROR(EINVAL); |
215 | goto error_return; | |
216 | } | |
c4cd747e | 217 | |
1da177e4 LT |
218 | /* |
219 | * Make sure that the dquots are attached to the inode. | |
220 | */ | |
7d095257 | 221 | code = xfs_qm_dqattach_locked(ip, 0); |
c4cd747e | 222 | if (code) |
1da177e4 | 223 | goto error_return; |
1da177e4 | 224 | |
c4cd747e CH |
225 | /* |
226 | * Now we can make the changes. Before we join the inode | |
227 | * to the transaction, if ATTR_SIZE is set then take care of | |
228 | * the part of the truncation that must be done without the | |
229 | * inode lock. This needs to be done before joining the inode | |
230 | * to the transaction, because the inode cannot be unlocked | |
231 | * once it is a part of the transaction. | |
232 | */ | |
0f285c8a | 233 | if (iattr->ia_size > ip->i_size) { |
61436feb CH |
234 | /* |
235 | * Do the first part of growing a file: zero any data | |
236 | * in the last block that is beyond the old EOF. We | |
237 | * need to do this before the inode is joined to the | |
238 | * transaction to modify the i_size. | |
239 | */ | |
0f285c8a | 240 | code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size); |
374e2ac3 | 241 | } |
1da177e4 | 242 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
c32676ee DC |
243 | |
244 | /* | |
245 | * We are going to log the inode size change in this | |
246 | * transaction so any previous writes that are beyond the on | |
247 | * disk EOF and the new EOF that have not been written out need | |
248 | * to be written here. If we do not write the data out, we | |
249 | * expose ourselves to the null files problem. | |
250 | * | |
251 | * Only flush from the on disk size to the smaller of the in | |
252 | * memory file size or the new size as that's the range we | |
253 | * really care about here and prevents waiting for other data | |
254 | * not within the range we care about here. | |
255 | */ | |
256 | if (!code && | |
0f285c8a CH |
257 | ip->i_size != ip->i_d.di_size && |
258 | iattr->ia_size > ip->i_d.di_size) { | |
739bfb2a | 259 | code = xfs_flush_pages(ip, |
0f285c8a | 260 | ip->i_d.di_size, iattr->ia_size, |
c32676ee DC |
261 | XFS_B_ASYNC, FI_NONE); |
262 | } | |
263 | ||
264 | /* wait for all I/O to complete */ | |
25e41b3d | 265 | xfs_ioend_wait(ip); |
c32676ee | 266 | |
1da177e4 | 267 | if (!code) |
0f285c8a | 268 | code = xfs_itruncate_data(ip, iattr->ia_size); |
1da177e4 LT |
269 | if (code) { |
270 | ASSERT(tp == NULL); | |
271 | lock_flags &= ~XFS_ILOCK_EXCL; | |
272 | ASSERT(lock_flags == XFS_IOLOCK_EXCL); | |
273 | goto error_return; | |
274 | } | |
275 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); | |
276 | if ((code = xfs_trans_reserve(tp, 0, | |
277 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
278 | XFS_TRANS_PERM_LOG_RES, | |
279 | XFS_ITRUNCATE_LOG_COUNT))) { | |
280 | xfs_trans_cancel(tp, 0); | |
281 | if (need_iolock) | |
282 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
283 | return code; | |
284 | } | |
285 | commit_flags = XFS_TRANS_RELEASE_LOG_RES; | |
286 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1da177e4 | 287 | |
1da177e4 LT |
288 | xfs_trans_ijoin(tp, ip, lock_flags); |
289 | xfs_trans_ihold(tp, ip); | |
1da177e4 | 290 | |
44d814ce DC |
291 | /* |
292 | * Only change the c/mtime if we are changing the size | |
293 | * or we are explicitly asked to change it. This handles | |
294 | * the semantic difference between truncate() and ftruncate() | |
295 | * as implemented in the VFS. | |
296 | */ | |
0f285c8a | 297 | if (iattr->ia_size != ip->i_size || (mask & ATTR_CTIME)) |
44d814ce DC |
298 | timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; |
299 | ||
0f285c8a CH |
300 | if (iattr->ia_size > ip->i_size) { |
301 | ip->i_d.di_size = iattr->ia_size; | |
302 | ip->i_size = iattr->ia_size; | |
303 | if (!(flags & XFS_ATTR_DMI)) | |
61436feb CH |
304 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG); |
305 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
0f285c8a CH |
306 | } else if (iattr->ia_size <= ip->i_size || |
307 | (iattr->ia_size == 0 && ip->i_d.di_nextents)) { | |
1da177e4 LT |
308 | /* |
309 | * signal a sync transaction unless | |
310 | * we're truncating an already unlinked | |
311 | * file on a wsync filesystem | |
312 | */ | |
0f285c8a | 313 | code = xfs_itruncate_finish(&tp, ip, iattr->ia_size, |
1da177e4 LT |
314 | XFS_DATA_FORK, |
315 | ((ip->i_d.di_nlink != 0 || | |
316 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
317 | ? 1 : 0)); | |
7d4fb40a | 318 | if (code) |
1da177e4 | 319 | goto abort_return; |
7d4fb40a NS |
320 | /* |
321 | * Truncated "down", so we're removing references | |
322 | * to old data here - if we now delay flushing for | |
323 | * a long time, we expose ourselves unduly to the | |
324 | * notorious NULL files problem. So, we mark this | |
325 | * vnode and flush it when the file is closed, and | |
326 | * do not wait the usual (long) time for writeout. | |
327 | */ | |
b3aea4ed | 328 | xfs_iflags_set(ip, XFS_ITRUNCATED); |
1da177e4 | 329 | } |
c4cd747e CH |
330 | } else if (tp) { |
331 | xfs_trans_ijoin(tp, ip, lock_flags); | |
332 | xfs_trans_ihold(tp, ip); | |
1da177e4 LT |
333 | } |
334 | ||
335 | /* | |
336 | * Change file ownership. Must be the owner or privileged. | |
1da177e4 | 337 | */ |
0f285c8a | 338 | if (mask & (ATTR_UID|ATTR_GID)) { |
1da177e4 LT |
339 | /* |
340 | * CAP_FSETID overrides the following restrictions: | |
341 | * | |
342 | * The set-user-ID and set-group-ID bits of a file will be | |
343 | * cleared upon successful return from chown() | |
344 | */ | |
345 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
346 | !capable(CAP_FSETID)) { | |
347 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
348 | } | |
349 | ||
350 | /* | |
351 | * Change the ownerships and register quota modifications | |
352 | * in the transaction. | |
353 | */ | |
354 | if (iuid != uid) { | |
7d095257 | 355 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { |
0f285c8a | 356 | ASSERT(mask & ATTR_UID); |
1da177e4 | 357 | ASSERT(udqp); |
7d095257 | 358 | olddquot1 = xfs_qm_vop_chown(tp, ip, |
1da177e4 LT |
359 | &ip->i_udquot, udqp); |
360 | } | |
361 | ip->i_d.di_uid = uid; | |
f13fae2d | 362 | inode->i_uid = uid; |
1da177e4 LT |
363 | } |
364 | if (igid != gid) { | |
7d095257 | 365 | if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { |
c8ad20ff | 366 | ASSERT(!XFS_IS_PQUOTA_ON(mp)); |
0f285c8a | 367 | ASSERT(mask & ATTR_GID); |
1da177e4 | 368 | ASSERT(gdqp); |
7d095257 | 369 | olddquot2 = xfs_qm_vop_chown(tp, ip, |
1da177e4 LT |
370 | &ip->i_gdquot, gdqp); |
371 | } | |
372 | ip->i_d.di_gid = gid; | |
f13fae2d | 373 | inode->i_gid = gid; |
1da177e4 | 374 | } |
1da177e4 LT |
375 | |
376 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
377 | timeflags |= XFS_ICHGTIME_CHG; | |
378 | } | |
379 | ||
c4cd747e CH |
380 | /* |
381 | * Change file access modes. | |
382 | */ | |
383 | if (mask & ATTR_MODE) { | |
384 | umode_t mode = iattr->ia_mode; | |
385 | ||
386 | if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) | |
387 | mode &= ~S_ISGID; | |
388 | ||
389 | ip->i_d.di_mode &= S_IFMT; | |
390 | ip->i_d.di_mode |= mode & ~S_IFMT; | |
391 | ||
392 | inode->i_mode &= S_IFMT; | |
393 | inode->i_mode |= mode & ~S_IFMT; | |
394 | ||
395 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
396 | timeflags |= XFS_ICHGTIME_CHG; | |
397 | } | |
1da177e4 LT |
398 | |
399 | /* | |
400 | * Change file access or modified times. | |
401 | */ | |
0f285c8a CH |
402 | if (mask & (ATTR_ATIME|ATTR_MTIME)) { |
403 | if (mask & ATTR_ATIME) { | |
f13fae2d | 404 | inode->i_atime = iattr->ia_atime; |
0f285c8a CH |
405 | ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec; |
406 | ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec; | |
1da177e4 | 407 | ip->i_update_core = 1; |
1da177e4 | 408 | } |
0f285c8a | 409 | if (mask & ATTR_MTIME) { |
f13fae2d | 410 | inode->i_mtime = iattr->ia_mtime; |
0f285c8a CH |
411 | ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec; |
412 | ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec; | |
1da177e4 LT |
413 | timeflags &= ~XFS_ICHGTIME_MOD; |
414 | timeflags |= XFS_ICHGTIME_CHG; | |
415 | } | |
0f285c8a | 416 | if (tp && (mask & (ATTR_MTIME_SET|ATTR_ATIME_SET))) |
1da177e4 LT |
417 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); |
418 | } | |
419 | ||
1da177e4 | 420 | /* |
0f285c8a | 421 | * Change file inode change time only if ATTR_CTIME set |
1da177e4 LT |
422 | * AND we have been called by a DMI function. |
423 | */ | |
424 | ||
0f285c8a | 425 | if ((flags & XFS_ATTR_DMI) && (mask & ATTR_CTIME)) { |
f13fae2d | 426 | inode->i_ctime = iattr->ia_ctime; |
0f285c8a CH |
427 | ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec; |
428 | ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec; | |
1da177e4 LT |
429 | ip->i_update_core = 1; |
430 | timeflags &= ~XFS_ICHGTIME_CHG; | |
431 | } | |
432 | ||
433 | /* | |
434 | * Send out timestamp changes that need to be set to the | |
435 | * current time. Not done when called by a DMI function. | |
436 | */ | |
0f285c8a | 437 | if (timeflags && !(flags & XFS_ATTR_DMI)) |
1da177e4 LT |
438 | xfs_ichgtime(ip, timeflags); |
439 | ||
440 | XFS_STATS_INC(xs_ig_attrchg); | |
441 | ||
442 | /* | |
443 | * If this is a synchronous mount, make sure that the | |
444 | * transaction goes to disk before returning to the user. | |
445 | * This is slightly sub-optimal in that truncates require | |
c41564b5 | 446 | * two sync transactions instead of one for wsync filesystems. |
1da177e4 LT |
447 | * One for the truncate and one for the timestamps since we |
448 | * don't want to change the timestamps unless we're sure the | |
449 | * truncate worked. Truncates are less than 1% of the laddis | |
450 | * mix so this probably isn't worth the trouble to optimize. | |
451 | */ | |
452 | code = 0; | |
453 | if (tp) { | |
454 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
455 | xfs_trans_set_sync(tp); | |
456 | ||
1c72bf90 | 457 | code = xfs_trans_commit(tp, commit_flags); |
1da177e4 LT |
458 | } |
459 | ||
1da177e4 LT |
460 | xfs_iunlock(ip, lock_flags); |
461 | ||
462 | /* | |
463 | * Release any dquot(s) the inode had kept before chown. | |
464 | */ | |
7d095257 CH |
465 | xfs_qm_dqrele(olddquot1); |
466 | xfs_qm_dqrele(olddquot2); | |
467 | xfs_qm_dqrele(udqp); | |
468 | xfs_qm_dqrele(gdqp); | |
1da177e4 LT |
469 | |
470 | if (code) { | |
471 | return code; | |
472 | } | |
473 | ||
eb9df39d | 474 | if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) && |
0f285c8a | 475 | !(flags & XFS_ATTR_DMI)) { |
bc4ac74a | 476 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL, |
1da177e4 LT |
477 | NULL, DM_RIGHT_NULL, NULL, NULL, |
478 | 0, 0, AT_DELAY_FLAG(flags)); | |
479 | } | |
480 | return 0; | |
481 | ||
482 | abort_return: | |
483 | commit_flags |= XFS_TRANS_ABORT; | |
484 | /* FALLTHROUGH */ | |
485 | error_return: | |
7d095257 CH |
486 | xfs_qm_dqrele(udqp); |
487 | xfs_qm_dqrele(gdqp); | |
1da177e4 LT |
488 | if (tp) { |
489 | xfs_trans_cancel(tp, commit_flags); | |
490 | } | |
491 | if (lock_flags != 0) { | |
492 | xfs_iunlock(ip, lock_flags); | |
493 | } | |
494 | return code; | |
495 | } | |
496 | ||
7d4fb40a NS |
497 | /* |
498 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
499 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
500 | * bmapi. | |
501 | */ | |
502 | #define SYMLINK_MAPS 2 | |
503 | ||
804c83c3 CH |
504 | STATIC int |
505 | xfs_readlink_bmap( | |
506 | xfs_inode_t *ip, | |
507 | char *link) | |
508 | { | |
509 | xfs_mount_t *mp = ip->i_mount; | |
510 | int pathlen = ip->i_d.di_size; | |
511 | int nmaps = SYMLINK_MAPS; | |
512 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
513 | xfs_daddr_t d; | |
514 | int byte_cnt; | |
515 | int n; | |
516 | xfs_buf_t *bp; | |
517 | int error = 0; | |
518 | ||
519 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0, | |
520 | mval, &nmaps, NULL, NULL); | |
521 | if (error) | |
522 | goto out; | |
523 | ||
524 | for (n = 0; n < nmaps; n++) { | |
525 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
526 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
527 | ||
528 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); | |
529 | error = XFS_BUF_GETERROR(bp); | |
530 | if (error) { | |
531 | xfs_ioerror_alert("xfs_readlink", | |
532 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
533 | xfs_buf_relse(bp); | |
534 | goto out; | |
535 | } | |
536 | if (pathlen < byte_cnt) | |
537 | byte_cnt = pathlen; | |
538 | pathlen -= byte_cnt; | |
539 | ||
540 | memcpy(link, XFS_BUF_PTR(bp), byte_cnt); | |
541 | xfs_buf_relse(bp); | |
542 | } | |
543 | ||
544 | link[ip->i_d.di_size] = '\0'; | |
545 | error = 0; | |
546 | ||
547 | out: | |
548 | return error; | |
549 | } | |
550 | ||
993386c1 | 551 | int |
1da177e4 | 552 | xfs_readlink( |
993386c1 | 553 | xfs_inode_t *ip, |
804c83c3 | 554 | char *link) |
1da177e4 | 555 | { |
804c83c3 | 556 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 557 | int pathlen; |
1da177e4 | 558 | int error = 0; |
1da177e4 | 559 | |
cf441eeb | 560 | xfs_itrace_entry(ip); |
1da177e4 LT |
561 | |
562 | if (XFS_FORCED_SHUTDOWN(mp)) | |
563 | return XFS_ERROR(EIO); | |
564 | ||
565 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
566 | ||
567 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
804c83c3 | 568 | ASSERT(ip->i_d.di_size <= MAXPATHLEN); |
1da177e4 | 569 | |
804c83c3 CH |
570 | pathlen = ip->i_d.di_size; |
571 | if (!pathlen) | |
572 | goto out; | |
1da177e4 LT |
573 | |
574 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
804c83c3 CH |
575 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
576 | link[pathlen] = '\0'; | |
577 | } else { | |
578 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
579 | } |
580 | ||
804c83c3 | 581 | out: |
1da177e4 | 582 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
583 | return error; |
584 | } | |
585 | ||
1da177e4 LT |
586 | /* |
587 | * xfs_fsync | |
588 | * | |
978b7237 DC |
589 | * This is called to sync the inode and its data out to disk. We need to hold |
590 | * the I/O lock while flushing the data, and the inode lock while flushing the | |
591 | * inode. The inode lock CANNOT be held while flushing the data, so acquire | |
592 | * after we're done with that. | |
1da177e4 | 593 | */ |
993386c1 | 594 | int |
1da177e4 | 595 | xfs_fsync( |
978b7237 | 596 | xfs_inode_t *ip) |
1da177e4 | 597 | { |
1da177e4 LT |
598 | xfs_trans_t *tp; |
599 | int error; | |
f538d4da | 600 | int log_flushed = 0, changed = 1; |
1da177e4 | 601 | |
cf441eeb | 602 | xfs_itrace_entry(ip); |
1da177e4 | 603 | |
1da177e4 LT |
604 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
605 | return XFS_ERROR(EIO); | |
606 | ||
978b7237 | 607 | /* capture size updates in I/O completion before writing the inode. */ |
2e656092 | 608 | error = xfs_wait_on_pages(ip, 0, -1); |
978b7237 DC |
609 | if (error) |
610 | return XFS_ERROR(error); | |
776a75fa | 611 | |
1da177e4 | 612 | /* |
978b7237 DC |
613 | * We always need to make sure that the required inode state is safe on |
614 | * disk. The vnode might be clean but we still might need to force the | |
615 | * log because of committed transactions that haven't hit the disk yet. | |
616 | * Likewise, there could be unflushed non-transactional changes to the | |
617 | * inode core that have to go to disk and this requires us to issue | |
618 | * a synchronous transaction to capture these changes correctly. | |
1da177e4 | 619 | * |
978b7237 DC |
620 | * This code relies on the assumption that if the update_* fields |
621 | * of the inode are clear and the inode is unpinned then it is clean | |
622 | * and no action is required. | |
1da177e4 LT |
623 | */ |
624 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
625 | ||
978b7237 | 626 | if (!(ip->i_update_size || ip->i_update_core)) { |
1da177e4 | 627 | /* |
978b7237 DC |
628 | * Timestamps/size haven't changed since last inode flush or |
629 | * inode transaction commit. That means either nothing got | |
630 | * written or a transaction committed which caught the updates. | |
631 | * If the latter happened and the transaction hasn't hit the | |
632 | * disk yet, the inode will be still be pinned. If it is, | |
633 | * force the log. | |
1da177e4 LT |
634 | */ |
635 | ||
636 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
637 | ||
638 | if (xfs_ipincount(ip)) { | |
978b7237 DC |
639 | error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0, |
640 | XFS_LOG_FORCE | XFS_LOG_SYNC, | |
f538d4da CH |
641 | &log_flushed); |
642 | } else { | |
643 | /* | |
978b7237 DC |
644 | * If the inode is not pinned and nothing has changed |
645 | * we don't need to flush the cache. | |
f538d4da CH |
646 | */ |
647 | changed = 0; | |
1da177e4 | 648 | } |
1da177e4 LT |
649 | } else { |
650 | /* | |
978b7237 DC |
651 | * Kick off a transaction to log the inode core to get the |
652 | * updates. The sync transaction will also force the log. | |
1da177e4 LT |
653 | */ |
654 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
655 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
978b7237 DC |
656 | error = xfs_trans_reserve(tp, 0, |
657 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0); | |
658 | if (error) { | |
1da177e4 LT |
659 | xfs_trans_cancel(tp, 0); |
660 | return error; | |
661 | } | |
662 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
663 | ||
664 | /* | |
978b7237 DC |
665 | * Note - it's possible that we might have pushed ourselves out |
666 | * of the way during trans_reserve which would flush the inode. | |
667 | * But there's no guarantee that the inode buffer has actually | |
668 | * gone out yet (it's delwri). Plus the buffer could be pinned | |
669 | * anyway if it's part of an inode in another recent | |
670 | * transaction. So we play it safe and fire off the | |
671 | * transaction anyway. | |
1da177e4 LT |
672 | */ |
673 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
674 | xfs_trans_ihold(tp, ip); | |
675 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
978b7237 | 676 | xfs_trans_set_sync(tp); |
1c72bf90 | 677 | error = _xfs_trans_commit(tp, 0, &log_flushed); |
1da177e4 LT |
678 | |
679 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
680 | } | |
f538d4da CH |
681 | |
682 | if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) { | |
683 | /* | |
684 | * If the log write didn't issue an ordered tag we need | |
685 | * to flush the disk cache for the data device now. | |
686 | */ | |
687 | if (!log_flushed) | |
688 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
689 | ||
690 | /* | |
691 | * If this inode is on the RT dev we need to flush that | |
c41564b5 | 692 | * cache as well. |
f538d4da | 693 | */ |
71ddabb9 | 694 | if (XFS_IS_REALTIME_INODE(ip)) |
f538d4da CH |
695 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); |
696 | } | |
697 | ||
1da177e4 LT |
698 | return error; |
699 | } | |
700 | ||
701 | /* | |
92dfe8d2 DC |
702 | * This is called by xfs_inactive to free any blocks beyond eof |
703 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
704 | * punching a hole to EOF. | |
1da177e4 | 705 | */ |
92dfe8d2 DC |
706 | int |
707 | xfs_free_eofblocks( | |
1da177e4 | 708 | xfs_mount_t *mp, |
92dfe8d2 DC |
709 | xfs_inode_t *ip, |
710 | int flags) | |
1da177e4 LT |
711 | { |
712 | xfs_trans_t *tp; | |
713 | int error; | |
714 | xfs_fileoff_t end_fsb; | |
715 | xfs_fileoff_t last_fsb; | |
716 | xfs_filblks_t map_len; | |
717 | int nimaps; | |
718 | xfs_bmbt_irec_t imap; | |
92dfe8d2 | 719 | int use_iolock = (flags & XFS_FREE_EOF_LOCK); |
1da177e4 LT |
720 | |
721 | /* | |
722 | * Figure out if there are any blocks beyond the end | |
723 | * of the file. If not, then there is nothing to do. | |
724 | */ | |
ba87ea69 | 725 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size)); |
1da177e4 LT |
726 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); |
727 | map_len = last_fsb - end_fsb; | |
728 | if (map_len <= 0) | |
014c2544 | 729 | return 0; |
1da177e4 LT |
730 | |
731 | nimaps = 1; | |
732 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
541d7d3c | 733 | error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0, |
3e57ecf6 | 734 | NULL, 0, &imap, &nimaps, NULL, NULL); |
1da177e4 LT |
735 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
736 | ||
737 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
738 | (imap.br_startblock != HOLESTARTBLOCK || |
739 | ip->i_delayed_blks)) { | |
1da177e4 LT |
740 | /* |
741 | * Attach the dquots to the inode up front. | |
742 | */ | |
7d095257 CH |
743 | error = xfs_qm_dqattach(ip, 0); |
744 | if (error) | |
014c2544 | 745 | return error; |
1da177e4 LT |
746 | |
747 | /* | |
748 | * There are blocks after the end of file. | |
749 | * Free them up now by truncating the file to | |
750 | * its current size. | |
751 | */ | |
752 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
753 | ||
754 | /* | |
755 | * Do the xfs_itruncate_start() call before | |
756 | * reserving any log space because | |
757 | * itruncate_start will call into the buffer | |
758 | * cache and we can't | |
759 | * do that within a transaction. | |
760 | */ | |
92dfe8d2 DC |
761 | if (use_iolock) |
762 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 | 763 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, |
ba87ea69 | 764 | ip->i_size); |
d3cf2094 | 765 | if (error) { |
87ae3c24 | 766 | xfs_trans_cancel(tp, 0); |
92dfe8d2 DC |
767 | if (use_iolock) |
768 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 LM |
769 | return error; |
770 | } | |
1da177e4 LT |
771 | |
772 | error = xfs_trans_reserve(tp, 0, | |
773 | XFS_ITRUNCATE_LOG_RES(mp), | |
774 | 0, XFS_TRANS_PERM_LOG_RES, | |
775 | XFS_ITRUNCATE_LOG_COUNT); | |
776 | if (error) { | |
777 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
778 | xfs_trans_cancel(tp, 0); | |
779 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 780 | return error; |
1da177e4 LT |
781 | } |
782 | ||
783 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
784 | xfs_trans_ijoin(tp, ip, | |
785 | XFS_IOLOCK_EXCL | | |
786 | XFS_ILOCK_EXCL); | |
787 | xfs_trans_ihold(tp, ip); | |
788 | ||
789 | error = xfs_itruncate_finish(&tp, ip, | |
ba87ea69 | 790 | ip->i_size, |
1da177e4 LT |
791 | XFS_DATA_FORK, |
792 | 0); | |
793 | /* | |
794 | * If we get an error at this point we | |
795 | * simply don't bother truncating the file. | |
796 | */ | |
797 | if (error) { | |
798 | xfs_trans_cancel(tp, | |
799 | (XFS_TRANS_RELEASE_LOG_RES | | |
800 | XFS_TRANS_ABORT)); | |
801 | } else { | |
802 | error = xfs_trans_commit(tp, | |
1c72bf90 | 803 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 804 | } |
92dfe8d2 DC |
805 | xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL) |
806 | : XFS_ILOCK_EXCL)); | |
1da177e4 | 807 | } |
014c2544 | 808 | return error; |
1da177e4 LT |
809 | } |
810 | ||
811 | /* | |
812 | * Free a symlink that has blocks associated with it. | |
813 | */ | |
814 | STATIC int | |
815 | xfs_inactive_symlink_rmt( | |
816 | xfs_inode_t *ip, | |
817 | xfs_trans_t **tpp) | |
818 | { | |
819 | xfs_buf_t *bp; | |
820 | int committed; | |
821 | int done; | |
822 | int error; | |
823 | xfs_fsblock_t first_block; | |
824 | xfs_bmap_free_t free_list; | |
825 | int i; | |
826 | xfs_mount_t *mp; | |
827 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
828 | int nmaps; | |
829 | xfs_trans_t *ntp; | |
830 | int size; | |
831 | xfs_trans_t *tp; | |
832 | ||
833 | tp = *tpp; | |
834 | mp = ip->i_mount; | |
835 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
836 | /* | |
837 | * We're freeing a symlink that has some | |
838 | * blocks allocated to it. Free the | |
839 | * blocks here. We know that we've got | |
840 | * either 1 or 2 extents and that we can | |
841 | * free them all in one bunmapi call. | |
842 | */ | |
843 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
844 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
845 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
846 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
847 | xfs_trans_cancel(tp, 0); | |
848 | *tpp = NULL; | |
849 | return error; | |
850 | } | |
851 | /* | |
852 | * Lock the inode, fix the size, and join it to the transaction. | |
853 | * Hold it so in the normal path, we still have it locked for | |
854 | * the second transaction. In the error paths we need it | |
855 | * held so the cancel won't rele it, see below. | |
856 | */ | |
857 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
858 | size = (int)ip->i_d.di_size; | |
859 | ip->i_d.di_size = 0; | |
860 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
861 | xfs_trans_ihold(tp, ip); | |
862 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
863 | /* | |
864 | * Find the block(s) so we can inval and unmap them. | |
865 | */ | |
866 | done = 0; | |
9d87c319 | 867 | xfs_bmap_init(&free_list, &first_block); |
e8c96f8c | 868 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
869 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
870 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 871 | &free_list, NULL))) |
1da177e4 LT |
872 | goto error0; |
873 | /* | |
874 | * Invalidate the block(s). | |
875 | */ | |
876 | for (i = 0; i < nmaps; i++) { | |
877 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
878 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
879 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
880 | xfs_trans_binval(tp, bp); | |
881 | } | |
882 | /* | |
883 | * Unmap the dead block(s) to the free_list. | |
884 | */ | |
885 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 886 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
887 | goto error1; |
888 | ASSERT(done); | |
889 | /* | |
890 | * Commit the first transaction. This logs the EFI and the inode. | |
891 | */ | |
f7c99b6f | 892 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
893 | goto error1; |
894 | /* | |
895 | * The transaction must have been committed, since there were | |
896 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
897 | * The new tp has the extent freeing and EFDs. | |
898 | */ | |
899 | ASSERT(committed); | |
900 | /* | |
901 | * The first xact was committed, so add the inode to the new one. | |
902 | * Mark it dirty so it will be logged and moved forward in the log as | |
903 | * part of every commit. | |
904 | */ | |
905 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
906 | xfs_trans_ihold(tp, ip); | |
907 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
908 | /* | |
909 | * Get a new, empty transaction to return to our caller. | |
910 | */ | |
911 | ntp = xfs_trans_dup(tp); | |
912 | /* | |
c41564b5 | 913 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
914 | * If we get an error on the commit here or on the reserve below, |
915 | * we need to unlock the inode since the new transaction doesn't | |
916 | * have the inode attached. | |
917 | */ | |
1c72bf90 | 918 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
919 | tp = ntp; |
920 | if (error) { | |
921 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
922 | goto error0; | |
923 | } | |
cc09c0dc DC |
924 | /* |
925 | * transaction commit worked ok so we can drop the extra ticket | |
926 | * reference that we gained in xfs_trans_dup() | |
927 | */ | |
928 | xfs_log_ticket_put(tp->t_ticket); | |
929 | ||
1da177e4 LT |
930 | /* |
931 | * Remove the memory for extent descriptions (just bookkeeping). | |
932 | */ | |
933 | if (ip->i_df.if_bytes) | |
934 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
935 | ASSERT(ip->i_df.if_bytes == 0); | |
936 | /* | |
937 | * Put an itruncate log reservation in the new transaction | |
938 | * for our caller. | |
939 | */ | |
940 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
941 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
942 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
943 | goto error0; | |
944 | } | |
945 | /* | |
946 | * Return with the inode locked but not joined to the transaction. | |
947 | */ | |
948 | *tpp = tp; | |
949 | return 0; | |
950 | ||
951 | error1: | |
952 | xfs_bmap_cancel(&free_list); | |
953 | error0: | |
954 | /* | |
955 | * Have to come here with the inode locked and either | |
956 | * (held and in the transaction) or (not in the transaction). | |
957 | * If the inode isn't held then cancel would iput it, but | |
958 | * that's wrong since this is inactive and the vnode ref | |
959 | * count is 0 already. | |
960 | * Cancel won't do anything to the inode if held, but it still | |
961 | * needs to be locked until the cancel is done, if it was | |
962 | * joined to the transaction. | |
963 | */ | |
964 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
965 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
966 | *tpp = NULL; | |
967 | return error; | |
968 | ||
969 | } | |
970 | ||
971 | STATIC int | |
972 | xfs_inactive_symlink_local( | |
973 | xfs_inode_t *ip, | |
974 | xfs_trans_t **tpp) | |
975 | { | |
976 | int error; | |
977 | ||
978 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
979 | /* | |
980 | * We're freeing a symlink which fit into | |
981 | * the inode. Just free the memory used | |
982 | * to hold the old symlink. | |
983 | */ | |
984 | error = xfs_trans_reserve(*tpp, 0, | |
985 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
986 | 0, XFS_TRANS_PERM_LOG_RES, | |
987 | XFS_ITRUNCATE_LOG_COUNT); | |
988 | ||
989 | if (error) { | |
990 | xfs_trans_cancel(*tpp, 0); | |
991 | *tpp = NULL; | |
014c2544 | 992 | return error; |
1da177e4 LT |
993 | } |
994 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
995 | ||
996 | /* | |
997 | * Zero length symlinks _can_ exist. | |
998 | */ | |
999 | if (ip->i_df.if_bytes > 0) { | |
1000 | xfs_idata_realloc(ip, | |
1001 | -(ip->i_df.if_bytes), | |
1002 | XFS_DATA_FORK); | |
1003 | ASSERT(ip->i_df.if_bytes == 0); | |
1004 | } | |
014c2544 | 1005 | return 0; |
1da177e4 LT |
1006 | } |
1007 | ||
1da177e4 LT |
1008 | STATIC int |
1009 | xfs_inactive_attrs( | |
1010 | xfs_inode_t *ip, | |
1011 | xfs_trans_t **tpp) | |
1012 | { | |
1013 | xfs_trans_t *tp; | |
1014 | int error; | |
1015 | xfs_mount_t *mp; | |
1016 | ||
579aa9ca | 1017 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
1da177e4 LT |
1018 | tp = *tpp; |
1019 | mp = ip->i_mount; | |
1020 | ASSERT(ip->i_d.di_forkoff != 0); | |
e5720eec | 1021 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 1022 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
e5720eec DC |
1023 | if (error) |
1024 | goto error_unlock; | |
1da177e4 LT |
1025 | |
1026 | error = xfs_attr_inactive(ip); | |
e5720eec DC |
1027 | if (error) |
1028 | goto error_unlock; | |
1da177e4 LT |
1029 | |
1030 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1031 | error = xfs_trans_reserve(tp, 0, | |
1032 | XFS_IFREE_LOG_RES(mp), | |
1033 | 0, XFS_TRANS_PERM_LOG_RES, | |
1034 | XFS_INACTIVE_LOG_COUNT); | |
e5720eec DC |
1035 | if (error) |
1036 | goto error_cancel; | |
1da177e4 LT |
1037 | |
1038 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1039 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1040 | xfs_trans_ihold(tp, ip); | |
1041 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1042 | ||
1043 | ASSERT(ip->i_d.di_anextents == 0); | |
1044 | ||
1045 | *tpp = tp; | |
014c2544 | 1046 | return 0; |
e5720eec DC |
1047 | |
1048 | error_cancel: | |
1049 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1050 | xfs_trans_cancel(tp, 0); | |
1051 | error_unlock: | |
1052 | *tpp = NULL; | |
1053 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1054 | return error; | |
1da177e4 LT |
1055 | } |
1056 | ||
993386c1 | 1057 | int |
1da177e4 | 1058 | xfs_release( |
993386c1 | 1059 | xfs_inode_t *ip) |
1da177e4 | 1060 | { |
993386c1 | 1061 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
1062 | int error; |
1063 | ||
42173f68 | 1064 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) |
1da177e4 | 1065 | return 0; |
1da177e4 LT |
1066 | |
1067 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1068 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1069 | return 0; |
1070 | ||
2a82b8be | 1071 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
1072 | int truncated; |
1073 | ||
2a82b8be DC |
1074 | /* |
1075 | * If we are using filestreams, and we have an unlinked | |
1076 | * file that we are processing the last close on, then nothing | |
1077 | * will be able to reopen and write to this file. Purge this | |
1078 | * inode from the filestreams cache so that it doesn't delay | |
1079 | * teardown of the inode. | |
1080 | */ | |
1081 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1082 | xfs_filestream_deassociate(ip); | |
1083 | ||
fbf3ce8d CH |
1084 | /* |
1085 | * If we previously truncated this file and removed old data | |
1086 | * in the process, we want to initiate "early" writeout on | |
1087 | * the last close. This is an attempt to combat the notorious | |
1088 | * NULL files problem which is particularly noticable from a | |
1089 | * truncate down, buffered (re-)write (delalloc), followed by | |
1090 | * a crash. What we are effectively doing here is | |
1091 | * significantly reducing the time window where we'd otherwise | |
1092 | * be exposed to that problem. | |
1093 | */ | |
09262b43 | 1094 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
df80c933 | 1095 | if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) |
739bfb2a | 1096 | xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE); |
fbf3ce8d CH |
1097 | } |
1098 | ||
1da177e4 LT |
1099 | if (ip->i_d.di_nlink != 0) { |
1100 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
df80c933 | 1101 | ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 || |
68bdb6ea | 1102 | ip->i_delayed_blks > 0)) && |
1da177e4 | 1103 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
1104 | (!(ip->i_d.di_flags & |
1105 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
92dfe8d2 DC |
1106 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1107 | if (error) | |
014c2544 | 1108 | return error; |
1da177e4 LT |
1109 | } |
1110 | } | |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * xfs_inactive | |
1117 | * | |
1118 | * This is called when the vnode reference count for the vnode | |
1119 | * goes to zero. If the file has been unlinked, then it must | |
1120 | * now be truncated. Also, we clear all of the read-ahead state | |
1121 | * kept for the inode here since the file is now closed. | |
1122 | */ | |
993386c1 | 1123 | int |
1da177e4 | 1124 | xfs_inactive( |
993386c1 | 1125 | xfs_inode_t *ip) |
1da177e4 | 1126 | { |
67fcaa73 | 1127 | xfs_bmap_free_t free_list; |
1da177e4 LT |
1128 | xfs_fsblock_t first_block; |
1129 | int committed; | |
1130 | xfs_trans_t *tp; | |
1131 | xfs_mount_t *mp; | |
1132 | int error; | |
1133 | int truncate; | |
1134 | ||
cf441eeb | 1135 | xfs_itrace_entry(ip); |
1da177e4 | 1136 | |
1da177e4 LT |
1137 | /* |
1138 | * If the inode is already free, then there can be nothing | |
1139 | * to clean up here. | |
1140 | */ | |
cb4c8cc1 | 1141 | if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) { |
1da177e4 LT |
1142 | ASSERT(ip->i_df.if_real_bytes == 0); |
1143 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1144 | return VN_INACTIVE_CACHE; | |
1145 | } | |
1146 | ||
1147 | /* | |
1148 | * Only do a truncate if it's a regular file with | |
1149 | * some actual space in it. It's OK to look at the | |
1150 | * inode's fields without the lock because we're the | |
1151 | * only one with a reference to the inode. | |
1152 | */ | |
1153 | truncate = ((ip->i_d.di_nlink == 0) && | |
ba87ea69 LM |
1154 | ((ip->i_d.di_size != 0) || (ip->i_size != 0) || |
1155 | (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1156 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1157 | ||
1158 | mp = ip->i_mount; | |
1159 | ||
bc4ac74a CH |
1160 | if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) |
1161 | XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL); | |
1da177e4 LT |
1162 | |
1163 | error = 0; | |
1164 | ||
1165 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1166 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1167 | goto out; |
1168 | ||
1169 | if (ip->i_d.di_nlink != 0) { | |
1170 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
df80c933 | 1171 | ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 || |
68bdb6ea | 1172 | ip->i_delayed_blks > 0)) && |
dd9f438e NS |
1173 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1174 | (!(ip->i_d.di_flags & | |
1175 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1176 | (ip->i_delayed_blks != 0)))) { | |
92dfe8d2 DC |
1177 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1178 | if (error) | |
014c2544 | 1179 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1180 | } |
1181 | goto out; | |
1182 | } | |
1183 | ||
1184 | ASSERT(ip->i_d.di_nlink == 0); | |
1185 | ||
7d095257 CH |
1186 | error = xfs_qm_dqattach(ip, 0); |
1187 | if (error) | |
014c2544 | 1188 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1189 | |
1190 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1191 | if (truncate) { | |
1192 | /* | |
1193 | * Do the xfs_itruncate_start() call before | |
1194 | * reserving any log space because itruncate_start | |
1195 | * will call into the buffer cache and we can't | |
1196 | * do that within a transaction. | |
1197 | */ | |
1198 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1199 | ||
d3cf2094 LM |
1200 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); |
1201 | if (error) { | |
87ae3c24 | 1202 | xfs_trans_cancel(tp, 0); |
d3cf2094 LM |
1203 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1204 | return VN_INACTIVE_CACHE; | |
1205 | } | |
1da177e4 LT |
1206 | |
1207 | error = xfs_trans_reserve(tp, 0, | |
1208 | XFS_ITRUNCATE_LOG_RES(mp), | |
1209 | 0, XFS_TRANS_PERM_LOG_RES, | |
1210 | XFS_ITRUNCATE_LOG_COUNT); | |
1211 | if (error) { | |
1212 | /* Don't call itruncate_cleanup */ | |
1213 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1214 | xfs_trans_cancel(tp, 0); | |
1215 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1216 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1217 | } |
1218 | ||
1219 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1220 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1221 | xfs_trans_ihold(tp, ip); | |
1222 | ||
1223 | /* | |
1224 | * normally, we have to run xfs_itruncate_finish sync. | |
1225 | * But if filesystem is wsync and we're in the inactive | |
1226 | * path, then we know that nlink == 0, and that the | |
1227 | * xaction that made nlink == 0 is permanently committed | |
1228 | * since xfs_remove runs as a synchronous transaction. | |
1229 | */ | |
1230 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1231 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1232 | ||
1233 | if (error) { | |
1234 | xfs_trans_cancel(tp, | |
1235 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1236 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1237 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1238 | } |
1239 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1240 | ||
1241 | /* | |
1242 | * If we get an error while cleaning up a | |
1243 | * symlink we bail out. | |
1244 | */ | |
1245 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1246 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1247 | xfs_inactive_symlink_local(ip, &tp); | |
1248 | ||
1249 | if (error) { | |
1250 | ASSERT(tp == NULL); | |
014c2544 | 1251 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1252 | } |
1253 | ||
1254 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1255 | xfs_trans_ihold(tp, ip); | |
1256 | } else { | |
1257 | error = xfs_trans_reserve(tp, 0, | |
1258 | XFS_IFREE_LOG_RES(mp), | |
1259 | 0, XFS_TRANS_PERM_LOG_RES, | |
1260 | XFS_INACTIVE_LOG_COUNT); | |
1261 | if (error) { | |
1262 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1263 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1264 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1265 | } |
1266 | ||
1267 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1268 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1269 | xfs_trans_ihold(tp, ip); | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * If there are attributes associated with the file | |
1274 | * then blow them away now. The code calls a routine | |
1275 | * that recursively deconstructs the attribute fork. | |
1276 | * We need to just commit the current transaction | |
1277 | * because we can't use it for xfs_attr_inactive(). | |
1278 | */ | |
1279 | if (ip->i_d.di_anextents > 0) { | |
1280 | error = xfs_inactive_attrs(ip, &tp); | |
1281 | /* | |
1282 | * If we got an error, the transaction is already | |
1283 | * cancelled, and the inode is unlocked. Just get out. | |
1284 | */ | |
1285 | if (error) | |
014c2544 | 1286 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1287 | } else if (ip->i_afp) { |
1288 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * Free the inode. | |
1293 | */ | |
9d87c319 | 1294 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
1295 | error = xfs_ifree(tp, ip, &free_list); |
1296 | if (error) { | |
1297 | /* | |
1298 | * If we fail to free the inode, shut down. The cancel | |
1299 | * might do that, we need to make sure. Otherwise the | |
1300 | * inode might be lost for a long time or forever. | |
1301 | */ | |
1302 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1303 | cmn_err(CE_NOTE, | |
1304 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1305 | error, mp->m_fsname); | |
7d04a335 | 1306 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1307 | } |
1308 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1309 | } else { | |
1310 | /* | |
1311 | * Credit the quota account(s). The inode is gone. | |
1312 | */ | |
7d095257 | 1313 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); |
1da177e4 LT |
1314 | |
1315 | /* | |
78e9da77 DC |
1316 | * Just ignore errors at this point. There is nothing we can |
1317 | * do except to try to keep going. Make sure it's not a silent | |
1318 | * error. | |
1da177e4 | 1319 | */ |
78e9da77 DC |
1320 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1321 | if (error) | |
1322 | xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: " | |
1323 | "xfs_bmap_finish() returned error %d", error); | |
1324 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1325 | if (error) | |
1326 | xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: " | |
1327 | "xfs_trans_commit() returned error %d", error); | |
1da177e4 | 1328 | } |
7d095257 | 1329 | |
1da177e4 LT |
1330 | /* |
1331 | * Release the dquots held by inode, if any. | |
1332 | */ | |
7d095257 | 1333 | xfs_qm_dqdetach(ip); |
1da177e4 LT |
1334 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); |
1335 | ||
1336 | out: | |
1337 | return VN_INACTIVE_CACHE; | |
1338 | } | |
1339 | ||
384f3ced BN |
1340 | /* |
1341 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
1342 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
1343 | * ci_name->name will point to a the actual name (caller must free) or | |
1344 | * will be set to NULL if an exact match is found. | |
1345 | */ | |
993386c1 | 1346 | int |
1da177e4 | 1347 | xfs_lookup( |
993386c1 | 1348 | xfs_inode_t *dp, |
556b8b16 | 1349 | struct xfs_name *name, |
384f3ced BN |
1350 | xfs_inode_t **ipp, |
1351 | struct xfs_name *ci_name) | |
1da177e4 | 1352 | { |
eca450b7 | 1353 | xfs_ino_t inum; |
1da177e4 LT |
1354 | int error; |
1355 | uint lock_mode; | |
1da177e4 | 1356 | |
cf441eeb | 1357 | xfs_itrace_entry(dp); |
1da177e4 LT |
1358 | |
1359 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1360 | return XFS_ERROR(EIO); | |
1361 | ||
1362 | lock_mode = xfs_ilock_map_shared(dp); | |
384f3ced | 1363 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
1da177e4 | 1364 | xfs_iunlock_map_shared(dp, lock_mode); |
eca450b7 CH |
1365 | |
1366 | if (error) | |
1367 | goto out; | |
1368 | ||
1369 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0); | |
1370 | if (error) | |
384f3ced | 1371 | goto out_free_name; |
eca450b7 CH |
1372 | |
1373 | xfs_itrace_ref(*ipp); | |
1374 | return 0; | |
1375 | ||
384f3ced BN |
1376 | out_free_name: |
1377 | if (ci_name) | |
1378 | kmem_free(ci_name->name); | |
1379 | out: | |
eca450b7 | 1380 | *ipp = NULL; |
1da177e4 LT |
1381 | return error; |
1382 | } | |
1383 | ||
993386c1 | 1384 | int |
1da177e4 | 1385 | xfs_create( |
993386c1 | 1386 | xfs_inode_t *dp, |
556b8b16 | 1387 | struct xfs_name *name, |
3e5daf05 CH |
1388 | mode_t mode, |
1389 | xfs_dev_t rdev, | |
979ebab1 | 1390 | xfs_inode_t **ipp, |
1da177e4 LT |
1391 | cred_t *credp) |
1392 | { | |
517b5e8c CH |
1393 | int is_dir = S_ISDIR(mode); |
1394 | struct xfs_mount *mp = dp->i_mount; | |
1395 | struct xfs_inode *ip = NULL; | |
1396 | struct xfs_trans *tp = NULL; | |
556b8b16 | 1397 | int error; |
1da177e4 LT |
1398 | xfs_bmap_free_t free_list; |
1399 | xfs_fsblock_t first_block; | |
993386c1 | 1400 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1401 | uint cancel_flags; |
1402 | int committed; | |
1403 | xfs_prid_t prid; | |
517b5e8c CH |
1404 | struct xfs_dquot *udqp = NULL; |
1405 | struct xfs_dquot *gdqp = NULL; | |
1da177e4 | 1406 | uint resblks; |
517b5e8c CH |
1407 | uint log_res; |
1408 | uint log_count; | |
1da177e4 | 1409 | |
cf441eeb | 1410 | xfs_itrace_entry(dp); |
1da177e4 | 1411 | |
517b5e8c CH |
1412 | if (XFS_FORCED_SHUTDOWN(mp)) |
1413 | return XFS_ERROR(EIO); | |
1414 | ||
eb9df39d | 1415 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 | 1416 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
bc4ac74a | 1417 | dp, DM_RIGHT_NULL, NULL, |
556b8b16 | 1418 | DM_RIGHT_NULL, name->name, NULL, |
3e5daf05 | 1419 | mode, 0, 0); |
1da177e4 LT |
1420 | |
1421 | if (error) | |
1422 | return error; | |
1da177e4 LT |
1423 | } |
1424 | ||
365ca83d NS |
1425 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1426 | prid = dp->i_d.di_projid; | |
1da177e4 | 1427 | else |
517b5e8c | 1428 | prid = dfltprid; |
1da177e4 LT |
1429 | |
1430 | /* | |
1431 | * Make sure that we have allocated dquot(s) on disk. | |
1432 | */ | |
7d095257 | 1433 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
517b5e8c | 1434 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1da177e4 LT |
1435 | if (error) |
1436 | goto std_return; | |
1437 | ||
517b5e8c CH |
1438 | if (is_dir) { |
1439 | rdev = 0; | |
1440 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); | |
1441 | log_res = XFS_MKDIR_LOG_RES(mp); | |
1442 | log_count = XFS_MKDIR_LOG_COUNT; | |
1443 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
1444 | } else { | |
1445 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
1446 | log_res = XFS_CREATE_LOG_RES(mp); | |
1447 | log_count = XFS_CREATE_LOG_COUNT; | |
1448 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1449 | } | |
1da177e4 | 1450 | |
1da177e4 | 1451 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
517b5e8c | 1452 | |
1da177e4 LT |
1453 | /* |
1454 | * Initially assume that the file does not exist and | |
1455 | * reserve the resources for that case. If that is not | |
1456 | * the case we'll drop the one we have and get a more | |
1457 | * appropriate transaction later. | |
1458 | */ | |
517b5e8c CH |
1459 | error = xfs_trans_reserve(tp, resblks, log_res, 0, |
1460 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 | 1461 | if (error == ENOSPC) { |
153fec43 DC |
1462 | /* flush outstanding delalloc blocks and retry */ |
1463 | xfs_flush_inodes(dp); | |
1464 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1465 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1466 | } | |
1467 | if (error == ENOSPC) { | |
1468 | /* No space at all so try a "no-allocation" reservation */ | |
1da177e4 | 1469 | resblks = 0; |
517b5e8c CH |
1470 | error = xfs_trans_reserve(tp, 0, log_res, 0, |
1471 | XFS_TRANS_PERM_LOG_RES, log_count); | |
1da177e4 LT |
1472 | } |
1473 | if (error) { | |
1474 | cancel_flags = 0; | |
517b5e8c | 1475 | goto out_trans_cancel; |
1da177e4 LT |
1476 | } |
1477 | ||
f7c66ce3 | 1478 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1479 | unlock_dp_on_error = B_TRUE; |
1da177e4 | 1480 | |
517b5e8c CH |
1481 | /* |
1482 | * Check for directory link count overflow. | |
1483 | */ | |
1484 | if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) { | |
1485 | error = XFS_ERROR(EMLINK); | |
1486 | goto out_trans_cancel; | |
1487 | } | |
1da177e4 | 1488 | |
517b5e8c | 1489 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
1490 | |
1491 | /* | |
1492 | * Reserve disk quota and the inode. | |
1493 | */ | |
7d095257 | 1494 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 | 1495 | if (error) |
517b5e8c | 1496 | goto out_trans_cancel; |
1da177e4 | 1497 | |
556b8b16 BN |
1498 | error = xfs_dir_canenter(tp, dp, name, resblks); |
1499 | if (error) | |
517b5e8c CH |
1500 | goto out_trans_cancel; |
1501 | ||
1502 | /* | |
1503 | * A newly created regular or special file just has one directory | |
1504 | * entry pointing to them, but a directory also the "." entry | |
1505 | * pointing to itself. | |
1506 | */ | |
1507 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, credp, | |
1508 | prid, resblks > 0, &ip, &committed); | |
1da177e4 LT |
1509 | if (error) { |
1510 | if (error == ENOSPC) | |
517b5e8c CH |
1511 | goto out_trans_cancel; |
1512 | goto out_trans_abort; | |
1da177e4 | 1513 | } |
1da177e4 LT |
1514 | |
1515 | /* | |
1516 | * At this point, we've gotten a newly allocated inode. | |
1517 | * It is locked (and joined to the transaction). | |
1518 | */ | |
517b5e8c | 1519 | xfs_itrace_ref(ip); |
579aa9ca | 1520 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
1521 | |
1522 | /* | |
993386c1 CH |
1523 | * Now we join the directory inode to the transaction. We do not do it |
1524 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1525 | * (and release all the locks). An error from here on will result in | |
1526 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1527 | * error path. | |
1da177e4 | 1528 | */ |
979ebab1 | 1529 | IHOLD(dp); |
1da177e4 | 1530 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 1531 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 1532 | |
556b8b16 | 1533 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
f6c2d1fa NS |
1534 | &first_block, &free_list, resblks ? |
1535 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1536 | if (error) { |
1537 | ASSERT(error != ENOSPC); | |
517b5e8c | 1538 | goto out_trans_abort; |
1da177e4 LT |
1539 | } |
1540 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1541 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1542 | ||
517b5e8c CH |
1543 | if (is_dir) { |
1544 | error = xfs_dir_init(tp, ip, dp); | |
1545 | if (error) | |
1546 | goto out_bmap_cancel; | |
1547 | ||
1548 | error = xfs_bumplink(tp, dp); | |
1549 | if (error) | |
1550 | goto out_bmap_cancel; | |
1551 | } | |
1552 | ||
1da177e4 LT |
1553 | /* |
1554 | * If this is a synchronous mount, make sure that the | |
1555 | * create transaction goes to disk before returning to | |
1556 | * the user. | |
1557 | */ | |
517b5e8c | 1558 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 1559 | xfs_trans_set_sync(tp); |
1da177e4 | 1560 | |
1da177e4 LT |
1561 | /* |
1562 | * Attach the dquot(s) to the inodes and modify them incore. | |
1563 | * These ids of the inode couldn't have changed since the new | |
1564 | * inode has been locked ever since it was created. | |
1565 | */ | |
7d095257 | 1566 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 LT |
1567 | |
1568 | /* | |
1569 | * xfs_trans_commit normally decrements the vnode ref count | |
1570 | * when it unlocks the inode. Since we want to return the | |
1571 | * vnode to the caller, we bump the vnode ref count now. | |
1572 | */ | |
1573 | IHOLD(ip); | |
1da177e4 | 1574 | |
f7c99b6f | 1575 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
517b5e8c CH |
1576 | if (error) |
1577 | goto out_abort_rele; | |
1da177e4 | 1578 | |
1c72bf90 | 1579 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1580 | if (error) { |
1581 | IRELE(ip); | |
517b5e8c | 1582 | goto out_dqrele; |
1da177e4 LT |
1583 | } |
1584 | ||
7d095257 CH |
1585 | xfs_qm_dqrele(udqp); |
1586 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1587 | |
979ebab1 | 1588 | *ipp = ip; |
1da177e4 LT |
1589 | |
1590 | /* Fallthrough to std_return with error = 0 */ | |
517b5e8c CH |
1591 | std_return: |
1592 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { | |
1593 | XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, dp, DM_RIGHT_NULL, | |
1594 | ip, DM_RIGHT_NULL, name->name, NULL, mode, | |
1595 | error, 0); | |
1da177e4 | 1596 | } |
517b5e8c | 1597 | |
1da177e4 LT |
1598 | return error; |
1599 | ||
517b5e8c CH |
1600 | out_bmap_cancel: |
1601 | xfs_bmap_cancel(&free_list); | |
1602 | out_trans_abort: | |
1da177e4 | 1603 | cancel_flags |= XFS_TRANS_ABORT; |
517b5e8c CH |
1604 | out_trans_cancel: |
1605 | xfs_trans_cancel(tp, cancel_flags); | |
1606 | out_dqrele: | |
7d095257 CH |
1607 | xfs_qm_dqrele(udqp); |
1608 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 1609 | |
993386c1 CH |
1610 | if (unlock_dp_on_error) |
1611 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1612 | ||
1da177e4 LT |
1613 | goto std_return; |
1614 | ||
517b5e8c | 1615 | out_abort_rele: |
1da177e4 LT |
1616 | /* |
1617 | * Wait until after the current transaction is aborted to | |
1618 | * release the inode. This prevents recursive transactions | |
1619 | * and deadlocks from xfs_inactive. | |
1620 | */ | |
517b5e8c | 1621 | xfs_bmap_cancel(&free_list); |
1da177e4 LT |
1622 | cancel_flags |= XFS_TRANS_ABORT; |
1623 | xfs_trans_cancel(tp, cancel_flags); | |
1624 | IRELE(ip); | |
517b5e8c CH |
1625 | unlock_dp_on_error = B_FALSE; |
1626 | goto out_dqrele; | |
1da177e4 LT |
1627 | } |
1628 | ||
1da177e4 LT |
1629 | #ifdef DEBUG |
1630 | int xfs_locked_n; | |
1631 | int xfs_small_retries; | |
1632 | int xfs_middle_retries; | |
1633 | int xfs_lots_retries; | |
1634 | int xfs_lock_delays; | |
1635 | #endif | |
1636 | ||
f7c66ce3 LM |
1637 | /* |
1638 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
1639 | * a different value | |
1640 | */ | |
1641 | static inline int | |
1642 | xfs_lock_inumorder(int lock_mode, int subclass) | |
1643 | { | |
1644 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 1645 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 1646 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 1647 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
1648 | |
1649 | return lock_mode; | |
1650 | } | |
1651 | ||
1da177e4 LT |
1652 | /* |
1653 | * The following routine will lock n inodes in exclusive mode. | |
1654 | * We assume the caller calls us with the inodes in i_ino order. | |
1655 | * | |
1656 | * We need to detect deadlock where an inode that we lock | |
1657 | * is in the AIL and we start waiting for another inode that is locked | |
1658 | * by a thread in a long running transaction (such as truncate). This can | |
1659 | * result in deadlock since the long running trans might need to wait | |
1660 | * for the inode we just locked in order to push the tail and free space | |
1661 | * in the log. | |
1662 | */ | |
1663 | void | |
1664 | xfs_lock_inodes( | |
1665 | xfs_inode_t **ips, | |
1666 | int inodes, | |
1da177e4 LT |
1667 | uint lock_mode) |
1668 | { | |
1669 | int attempts = 0, i, j, try_lock; | |
1670 | xfs_log_item_t *lp; | |
1671 | ||
1672 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
1673 | ||
cfa853e4 CH |
1674 | try_lock = 0; |
1675 | i = 0; | |
1da177e4 LT |
1676 | |
1677 | again: | |
1678 | for (; i < inodes; i++) { | |
1679 | ASSERT(ips[i]); | |
1680 | ||
1681 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
1682 | continue; | |
1683 | ||
1684 | /* | |
1685 | * If try_lock is not set yet, make sure all locked inodes | |
1686 | * are not in the AIL. | |
1687 | * If any are, set try_lock to be used later. | |
1688 | */ | |
1689 | ||
1690 | if (!try_lock) { | |
1691 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
1692 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
1693 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1694 | try_lock++; | |
1695 | } | |
1696 | } | |
1697 | } | |
1698 | ||
1699 | /* | |
1700 | * If any of the previous locks we have locked is in the AIL, | |
1701 | * we must TRY to get the second and subsequent locks. If | |
1702 | * we can't get any, we must release all we have | |
1703 | * and try again. | |
1704 | */ | |
1705 | ||
1706 | if (try_lock) { | |
1707 | /* try_lock must be 0 if i is 0. */ | |
1708 | /* | |
1709 | * try_lock means we have an inode locked | |
1710 | * that is in the AIL. | |
1711 | */ | |
1712 | ASSERT(i != 0); | |
f7c66ce3 | 1713 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
1714 | attempts++; |
1715 | ||
1716 | /* | |
1717 | * Unlock all previous guys and try again. | |
1718 | * xfs_iunlock will try to push the tail | |
1719 | * if the inode is in the AIL. | |
1720 | */ | |
1721 | ||
1722 | for(j = i - 1; j >= 0; j--) { | |
1723 | ||
1724 | /* | |
1725 | * Check to see if we've already | |
1726 | * unlocked this one. | |
1727 | * Not the first one going back, | |
1728 | * and the inode ptr is the same. | |
1729 | */ | |
1730 | if ((j != (i - 1)) && ips[j] == | |
1731 | ips[j+1]) | |
1732 | continue; | |
1733 | ||
1734 | xfs_iunlock(ips[j], lock_mode); | |
1735 | } | |
1736 | ||
1737 | if ((attempts % 5) == 0) { | |
1738 | delay(1); /* Don't just spin the CPU */ | |
1739 | #ifdef DEBUG | |
1740 | xfs_lock_delays++; | |
1741 | #endif | |
1742 | } | |
1743 | i = 0; | |
1744 | try_lock = 0; | |
1745 | goto again; | |
1746 | } | |
1747 | } else { | |
f7c66ce3 | 1748 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
1749 | } |
1750 | } | |
1751 | ||
1752 | #ifdef DEBUG | |
1753 | if (attempts) { | |
1754 | if (attempts < 5) xfs_small_retries++; | |
1755 | else if (attempts < 100) xfs_middle_retries++; | |
1756 | else xfs_lots_retries++; | |
1757 | } else { | |
1758 | xfs_locked_n++; | |
1759 | } | |
1760 | #endif | |
1761 | } | |
1762 | ||
f9114eba DC |
1763 | /* |
1764 | * xfs_lock_two_inodes() can only be used to lock one type of lock | |
1765 | * at a time - the iolock or the ilock, but not both at once. If | |
1766 | * we lock both at once, lockdep will report false positives saying | |
1767 | * we have violated locking orders. | |
1768 | */ | |
e1cccd91 CH |
1769 | void |
1770 | xfs_lock_two_inodes( | |
1771 | xfs_inode_t *ip0, | |
1772 | xfs_inode_t *ip1, | |
1773 | uint lock_mode) | |
1774 | { | |
1775 | xfs_inode_t *temp; | |
1776 | int attempts = 0; | |
1777 | xfs_log_item_t *lp; | |
1778 | ||
f9114eba DC |
1779 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) |
1780 | ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0); | |
e1cccd91 CH |
1781 | ASSERT(ip0->i_ino != ip1->i_ino); |
1782 | ||
1783 | if (ip0->i_ino > ip1->i_ino) { | |
1784 | temp = ip0; | |
1785 | ip0 = ip1; | |
1786 | ip1 = temp; | |
1787 | } | |
1788 | ||
1789 | again: | |
1790 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
1791 | ||
1792 | /* | |
1793 | * If the first lock we have locked is in the AIL, we must TRY to get | |
1794 | * the second lock. If we can't get it, we must release the first one | |
1795 | * and try again. | |
1796 | */ | |
1797 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
1798 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1799 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
1800 | xfs_iunlock(ip0, lock_mode); | |
1801 | if ((++attempts % 5) == 0) | |
1802 | delay(1); /* Don't just spin the CPU */ | |
1803 | goto again; | |
1804 | } | |
1805 | } else { | |
1806 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
1807 | } | |
1808 | } | |
1809 | ||
993386c1 | 1810 | int |
1da177e4 | 1811 | xfs_remove( |
993386c1 | 1812 | xfs_inode_t *dp, |
556b8b16 BN |
1813 | struct xfs_name *name, |
1814 | xfs_inode_t *ip) | |
1da177e4 | 1815 | { |
993386c1 | 1816 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 1817 | xfs_trans_t *tp = NULL; |
8f112e3b | 1818 | int is_dir = S_ISDIR(ip->i_d.di_mode); |
1da177e4 LT |
1819 | int error = 0; |
1820 | xfs_bmap_free_t free_list; | |
1821 | xfs_fsblock_t first_block; | |
1822 | int cancel_flags; | |
1823 | int committed; | |
1da177e4 LT |
1824 | int link_zero; |
1825 | uint resblks; | |
8f112e3b | 1826 | uint log_count; |
1da177e4 | 1827 | |
cf441eeb | 1828 | xfs_itrace_entry(dp); |
8f112e3b | 1829 | xfs_itrace_entry(ip); |
1da177e4 | 1830 | |
1da177e4 LT |
1831 | if (XFS_FORCED_SHUTDOWN(mp)) |
1832 | return XFS_ERROR(EIO); | |
1833 | ||
eb9df39d | 1834 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
556b8b16 BN |
1835 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL, |
1836 | NULL, DM_RIGHT_NULL, name->name, NULL, | |
1837 | ip->i_d.di_mode, 0, 0); | |
1da177e4 LT |
1838 | if (error) |
1839 | return error; | |
1840 | } | |
1841 | ||
7d095257 | 1842 | error = xfs_qm_dqattach(dp, 0); |
8f112e3b CH |
1843 | if (error) |
1844 | goto std_return; | |
1845 | ||
7d095257 | 1846 | error = xfs_qm_dqattach(ip, 0); |
8f112e3b | 1847 | if (error) |
1da177e4 | 1848 | goto std_return; |
1da177e4 | 1849 | |
8f112e3b CH |
1850 | if (is_dir) { |
1851 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
1852 | log_count = XFS_DEFAULT_LOG_COUNT; | |
1853 | } else { | |
1854 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
1855 | log_count = XFS_REMOVE_LOG_COUNT; | |
1856 | } | |
1da177e4 | 1857 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
8f112e3b | 1858 | |
1da177e4 LT |
1859 | /* |
1860 | * We try to get the real space reservation first, | |
1861 | * allowing for directory btree deletion(s) implying | |
1862 | * possible bmap insert(s). If we can't get the space | |
1863 | * reservation then we use 0 instead, and avoid the bmap | |
1864 | * btree insert(s) in the directory code by, if the bmap | |
1865 | * insert tries to happen, instead trimming the LAST | |
1866 | * block from the directory. | |
1867 | */ | |
1868 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
1869 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1870 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1871 | if (error == ENOSPC) { |
1872 | resblks = 0; | |
1873 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
8f112e3b | 1874 | XFS_TRANS_PERM_LOG_RES, log_count); |
1da177e4 LT |
1875 | } |
1876 | if (error) { | |
1877 | ASSERT(error != ENOSPC); | |
8f112e3b CH |
1878 | cancel_flags = 0; |
1879 | goto out_trans_cancel; | |
1da177e4 LT |
1880 | } |
1881 | ||
e1cccd91 | 1882 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1883 | |
1884 | /* | |
1885 | * At this point, we've gotten both the directory and the entry | |
1886 | * inodes locked. | |
1887 | */ | |
5df78e73 | 1888 | IHOLD(ip); |
1da177e4 | 1889 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
82dab941 CH |
1890 | |
1891 | IHOLD(dp); | |
1892 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1da177e4 | 1893 | |
8f112e3b CH |
1894 | /* |
1895 | * If we're removing a directory perform some additional validation. | |
1896 | */ | |
1897 | if (is_dir) { | |
1898 | ASSERT(ip->i_d.di_nlink >= 2); | |
1899 | if (ip->i_d.di_nlink != 2) { | |
1900 | error = XFS_ERROR(ENOTEMPTY); | |
1901 | goto out_trans_cancel; | |
1902 | } | |
1903 | if (!xfs_dir_isempty(ip)) { | |
1904 | error = XFS_ERROR(ENOTEMPTY); | |
1905 | goto out_trans_cancel; | |
1906 | } | |
1907 | } | |
1908 | ||
9d87c319 | 1909 | xfs_bmap_init(&free_list, &first_block); |
556b8b16 | 1910 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
d4377d84 | 1911 | &first_block, &free_list, resblks); |
1da177e4 LT |
1912 | if (error) { |
1913 | ASSERT(error != ENOENT); | |
8f112e3b | 1914 | goto out_bmap_cancel; |
1da177e4 LT |
1915 | } |
1916 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1917 | ||
8f112e3b CH |
1918 | if (is_dir) { |
1919 | /* | |
1920 | * Drop the link from ip's "..". | |
1921 | */ | |
1922 | error = xfs_droplink(tp, dp); | |
1923 | if (error) | |
1924 | goto out_bmap_cancel; | |
1925 | ||
1926 | /* | |
2b7035fd | 1927 | * Drop the "." link from ip to self. |
8f112e3b CH |
1928 | */ |
1929 | error = xfs_droplink(tp, ip); | |
1930 | if (error) | |
1931 | goto out_bmap_cancel; | |
1932 | } else { | |
1933 | /* | |
1934 | * When removing a non-directory we need to log the parent | |
26c52951 DC |
1935 | * inode here. For a directory this is done implicitly |
1936 | * by the xfs_droplink call for the ".." entry. | |
8f112e3b CH |
1937 | */ |
1938 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1da177e4 LT |
1939 | } |
1940 | ||
8f112e3b | 1941 | /* |
2b7035fd | 1942 | * Drop the link from dp to ip. |
8f112e3b CH |
1943 | */ |
1944 | error = xfs_droplink(tp, ip); | |
1945 | if (error) | |
1946 | goto out_bmap_cancel; | |
1947 | ||
1948 | /* | |
1949 | * Determine if this is the last link while | |
1da177e4 LT |
1950 | * we are in the transaction. |
1951 | */ | |
8f112e3b | 1952 | link_zero = (ip->i_d.di_nlink == 0); |
1da177e4 | 1953 | |
1da177e4 LT |
1954 | /* |
1955 | * If this is a synchronous mount, make sure that the | |
1956 | * remove transaction goes to disk before returning to | |
1957 | * the user. | |
1958 | */ | |
8f112e3b | 1959 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
1da177e4 | 1960 | xfs_trans_set_sync(tp); |
1da177e4 | 1961 | |
f7c99b6f | 1962 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
8f112e3b CH |
1963 | if (error) |
1964 | goto out_bmap_cancel; | |
1da177e4 | 1965 | |
1c72bf90 | 1966 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
5df78e73 | 1967 | if (error) |
1da177e4 | 1968 | goto std_return; |
1da177e4 | 1969 | |
2a82b8be DC |
1970 | /* |
1971 | * If we are using filestreams, kill the stream association. | |
1972 | * If the file is still open it may get a new one but that | |
1973 | * will get killed on last close in xfs_close() so we don't | |
1974 | * have to worry about that. | |
1975 | */ | |
8f112e3b | 1976 | if (!is_dir && link_zero && xfs_inode_is_filestream(ip)) |
2a82b8be DC |
1977 | xfs_filestream_deassociate(ip); |
1978 | ||
cf441eeb | 1979 | xfs_itrace_exit(ip); |
8f112e3b | 1980 | xfs_itrace_exit(dp); |
1da177e4 | 1981 | |
1da177e4 | 1982 | std_return: |
eb9df39d | 1983 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
8f112e3b CH |
1984 | XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, dp, DM_RIGHT_NULL, |
1985 | NULL, DM_RIGHT_NULL, name->name, NULL, | |
1986 | ip->i_d.di_mode, error, 0); | |
1da177e4 | 1987 | } |
8f112e3b | 1988 | |
1da177e4 LT |
1989 | return error; |
1990 | ||
8f112e3b | 1991 | out_bmap_cancel: |
1da177e4 LT |
1992 | xfs_bmap_cancel(&free_list); |
1993 | cancel_flags |= XFS_TRANS_ABORT; | |
8f112e3b | 1994 | out_trans_cancel: |
1da177e4 LT |
1995 | xfs_trans_cancel(tp, cancel_flags); |
1996 | goto std_return; | |
1da177e4 LT |
1997 | } |
1998 | ||
993386c1 | 1999 | int |
1da177e4 | 2000 | xfs_link( |
993386c1 | 2001 | xfs_inode_t *tdp, |
a3da7896 | 2002 | xfs_inode_t *sip, |
556b8b16 | 2003 | struct xfs_name *target_name) |
1da177e4 | 2004 | { |
993386c1 | 2005 | xfs_mount_t *mp = tdp->i_mount; |
1da177e4 | 2006 | xfs_trans_t *tp; |
1da177e4 LT |
2007 | int error; |
2008 | xfs_bmap_free_t free_list; | |
2009 | xfs_fsblock_t first_block; | |
2010 | int cancel_flags; | |
2011 | int committed; | |
1da177e4 | 2012 | int resblks; |
1da177e4 | 2013 | |
cf441eeb | 2014 | xfs_itrace_entry(tdp); |
a3da7896 | 2015 | xfs_itrace_entry(sip); |
1da177e4 | 2016 | |
a3da7896 | 2017 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); |
1da177e4 | 2018 | |
1da177e4 LT |
2019 | if (XFS_FORCED_SHUTDOWN(mp)) |
2020 | return XFS_ERROR(EIO); | |
2021 | ||
eb9df39d | 2022 | if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) { |
1da177e4 | 2023 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, |
bc4ac74a CH |
2024 | tdp, DM_RIGHT_NULL, |
2025 | sip, DM_RIGHT_NULL, | |
556b8b16 | 2026 | target_name->name, NULL, 0, 0, 0); |
1da177e4 LT |
2027 | if (error) |
2028 | return error; | |
2029 | } | |
2030 | ||
2031 | /* Return through std_return after this point. */ | |
2032 | ||
7d095257 | 2033 | error = xfs_qm_dqattach(sip, 0); |
cb3f35bb CH |
2034 | if (error) |
2035 | goto std_return; | |
2036 | ||
7d095257 | 2037 | error = xfs_qm_dqattach(tdp, 0); |
1da177e4 LT |
2038 | if (error) |
2039 | goto std_return; | |
2040 | ||
2041 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2042 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
556b8b16 | 2043 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
1da177e4 LT |
2044 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, |
2045 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2046 | if (error == ENOSPC) { | |
2047 | resblks = 0; | |
2048 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2049 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2050 | } | |
2051 | if (error) { | |
2052 | cancel_flags = 0; | |
2053 | goto error_return; | |
2054 | } | |
2055 | ||
e1cccd91 | 2056 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2057 | |
2058 | /* | |
2059 | * Increment vnode ref counts since xfs_trans_commit & | |
2060 | * xfs_trans_cancel will both unlock the inodes and | |
2061 | * decrement the associated ref counts. | |
2062 | */ | |
a3da7896 CH |
2063 | IHOLD(sip); |
2064 | IHOLD(tdp); | |
1da177e4 LT |
2065 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); |
2066 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2067 | ||
2068 | /* | |
2069 | * If the source has too many links, we can't make any more to it. | |
2070 | */ | |
2071 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2072 | error = XFS_ERROR(EMLINK); | |
2073 | goto error_return; | |
2074 | } | |
2075 | ||
365ca83d NS |
2076 | /* |
2077 | * If we are using project inheritance, we only allow hard link | |
2078 | * creation in our tree when the project IDs are the same; else | |
2079 | * the tree quota mechanism could be circumvented. | |
2080 | */ | |
2081 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2082 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 2083 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
2084 | goto error_return; |
2085 | } | |
2086 | ||
556b8b16 BN |
2087 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); |
2088 | if (error) | |
1da177e4 LT |
2089 | goto error_return; |
2090 | ||
9d87c319 | 2091 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 2092 | |
556b8b16 BN |
2093 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
2094 | &first_block, &free_list, resblks); | |
1da177e4 LT |
2095 | if (error) |
2096 | goto abort_return; | |
2097 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1da177e4 LT |
2098 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); |
2099 | ||
2100 | error = xfs_bumplink(tp, sip); | |
b71d300c | 2101 | if (error) |
1da177e4 | 2102 | goto abort_return; |
1da177e4 LT |
2103 | |
2104 | /* | |
2105 | * If this is a synchronous mount, make sure that the | |
2106 | * link transaction goes to disk before returning to | |
2107 | * the user. | |
2108 | */ | |
2109 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2110 | xfs_trans_set_sync(tp); | |
2111 | } | |
2112 | ||
f7c99b6f | 2113 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2114 | if (error) { |
2115 | xfs_bmap_cancel(&free_list); | |
2116 | goto abort_return; | |
2117 | } | |
2118 | ||
1c72bf90 | 2119 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
b71d300c | 2120 | if (error) |
1da177e4 | 2121 | goto std_return; |
1da177e4 LT |
2122 | |
2123 | /* Fall through to std_return with error = 0. */ | |
2124 | std_return: | |
eb9df39d | 2125 | if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) { |
1da177e4 | 2126 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, |
bc4ac74a CH |
2127 | tdp, DM_RIGHT_NULL, |
2128 | sip, DM_RIGHT_NULL, | |
556b8b16 | 2129 | target_name->name, NULL, 0, error, 0); |
1da177e4 LT |
2130 | } |
2131 | return error; | |
2132 | ||
2133 | abort_return: | |
2134 | cancel_flags |= XFS_TRANS_ABORT; | |
2135 | /* FALLTHROUGH */ | |
014c2544 | 2136 | |
1da177e4 LT |
2137 | error_return: |
2138 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2139 | goto std_return; |
2140 | } | |
b71d300c | 2141 | |
993386c1 | 2142 | int |
1da177e4 | 2143 | xfs_symlink( |
993386c1 | 2144 | xfs_inode_t *dp, |
556b8b16 BN |
2145 | struct xfs_name *link_name, |
2146 | const char *target_path, | |
3e5daf05 | 2147 | mode_t mode, |
3937be5b | 2148 | xfs_inode_t **ipp, |
1da177e4 LT |
2149 | cred_t *credp) |
2150 | { | |
993386c1 | 2151 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 2152 | xfs_trans_t *tp; |
1da177e4 LT |
2153 | xfs_inode_t *ip; |
2154 | int error; | |
2155 | int pathlen; | |
2156 | xfs_bmap_free_t free_list; | |
2157 | xfs_fsblock_t first_block; | |
993386c1 | 2158 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2159 | uint cancel_flags; |
2160 | int committed; | |
2161 | xfs_fileoff_t first_fsb; | |
2162 | xfs_filblks_t fs_blocks; | |
2163 | int nmaps; | |
2164 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
2165 | xfs_daddr_t d; | |
556b8b16 | 2166 | const char *cur_chunk; |
1da177e4 LT |
2167 | int byte_cnt; |
2168 | int n; | |
2169 | xfs_buf_t *bp; | |
2170 | xfs_prid_t prid; | |
2171 | struct xfs_dquot *udqp, *gdqp; | |
2172 | uint resblks; | |
1da177e4 | 2173 | |
3937be5b | 2174 | *ipp = NULL; |
1da177e4 LT |
2175 | error = 0; |
2176 | ip = NULL; | |
2177 | tp = NULL; | |
2178 | ||
cf441eeb | 2179 | xfs_itrace_entry(dp); |
1da177e4 LT |
2180 | |
2181 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2182 | return XFS_ERROR(EIO); | |
2183 | ||
1da177e4 LT |
2184 | /* |
2185 | * Check component lengths of the target path name. | |
2186 | */ | |
2187 | pathlen = strlen(target_path); | |
2188 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
2189 | return XFS_ERROR(ENAMETOOLONG); | |
1da177e4 | 2190 | |
eb9df39d | 2191 | if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) { |
bc4ac74a | 2192 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp, |
1da177e4 | 2193 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, |
556b8b16 | 2194 | link_name->name, target_path, 0, 0, 0); |
1da177e4 LT |
2195 | if (error) |
2196 | return error; | |
2197 | } | |
2198 | ||
2199 | /* Return through std_return after this point. */ | |
2200 | ||
2201 | udqp = gdqp = NULL; | |
365ca83d NS |
2202 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2203 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
2204 | else |
2205 | prid = (xfs_prid_t)dfltprid; | |
2206 | ||
2207 | /* | |
2208 | * Make sure that we have allocated dquot(s) on disk. | |
2209 | */ | |
7d095257 | 2210 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, |
1da177e4 LT |
2211 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2212 | if (error) | |
2213 | goto std_return; | |
2214 | ||
2215 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
2216 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2217 | /* | |
2218 | * The symlink will fit into the inode data fork? | |
2219 | * There can't be any attributes so we get the whole variable part. | |
2220 | */ | |
2221 | if (pathlen <= XFS_LITINO(mp)) | |
2222 | fs_blocks = 0; | |
2223 | else | |
2224 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
556b8b16 | 2225 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks); |
1da177e4 LT |
2226 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, |
2227 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
2228 | if (error == ENOSPC && fs_blocks == 0) { | |
2229 | resblks = 0; | |
2230 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
2231 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
2232 | } | |
2233 | if (error) { | |
2234 | cancel_flags = 0; | |
1da177e4 LT |
2235 | goto error_return; |
2236 | } | |
2237 | ||
f7c66ce3 | 2238 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2239 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2240 | |
2241 | /* | |
2242 | * Check whether the directory allows new symlinks or not. | |
2243 | */ | |
2244 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
2245 | error = XFS_ERROR(EPERM); | |
2246 | goto error_return; | |
2247 | } | |
2248 | ||
2249 | /* | |
2250 | * Reserve disk quota : blocks and inode. | |
2251 | */ | |
7d095257 | 2252 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0); |
1da177e4 LT |
2253 | if (error) |
2254 | goto error_return; | |
2255 | ||
2256 | /* | |
2257 | * Check for ability to enter directory entry, if no space reserved. | |
2258 | */ | |
556b8b16 BN |
2259 | error = xfs_dir_canenter(tp, dp, link_name, resblks); |
2260 | if (error) | |
1da177e4 LT |
2261 | goto error_return; |
2262 | /* | |
2263 | * Initialize the bmap freelist prior to calling either | |
2264 | * bmapi or the directory create code. | |
2265 | */ | |
9d87c319 | 2266 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 LT |
2267 | |
2268 | /* | |
2269 | * Allocate an inode for the symlink. | |
2270 | */ | |
3e5daf05 | 2271 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), |
1da177e4 LT |
2272 | 1, 0, credp, prid, resblks > 0, &ip, NULL); |
2273 | if (error) { | |
2274 | if (error == ENOSPC) | |
2275 | goto error_return; | |
2276 | goto error1; | |
2277 | } | |
cf441eeb | 2278 | xfs_itrace_ref(ip); |
1da177e4 | 2279 | |
993386c1 CH |
2280 | /* |
2281 | * An error after we've joined dp to the transaction will result in the | |
2282 | * transaction cancel unlocking dp so don't do it explicitly in the | |
2283 | * error path. | |
2284 | */ | |
3937be5b | 2285 | IHOLD(dp); |
1da177e4 | 2286 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 2287 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2288 | |
2289 | /* | |
2290 | * Also attach the dquot(s) to it, if applicable. | |
2291 | */ | |
7d095257 | 2292 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp); |
1da177e4 LT |
2293 | |
2294 | if (resblks) | |
2295 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
2296 | /* | |
2297 | * If the symlink will fit into the inode, write it inline. | |
2298 | */ | |
2299 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
2300 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
2301 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
2302 | ip->i_d.di_size = pathlen; | |
2303 | ||
2304 | /* | |
2305 | * The inode was initially created in extent format. | |
2306 | */ | |
2307 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
2308 | ip->i_df.if_flags |= XFS_IFINLINE; | |
2309 | ||
2310 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
2311 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
2312 | ||
2313 | } else { | |
2314 | first_fsb = 0; | |
2315 | nmaps = SYMLINK_MAPS; | |
2316 | ||
2317 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
2318 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
2319 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 2320 | &free_list, NULL); |
1da177e4 LT |
2321 | if (error) { |
2322 | goto error1; | |
2323 | } | |
2324 | ||
2325 | if (resblks) | |
2326 | resblks -= fs_blocks; | |
2327 | ip->i_d.di_size = pathlen; | |
2328 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2329 | ||
2330 | cur_chunk = target_path; | |
2331 | for (n = 0; n < nmaps; n++) { | |
2332 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
2333 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
2334 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
2335 | BTOBB(byte_cnt), 0); | |
2336 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
2337 | if (pathlen < byte_cnt) { | |
2338 | byte_cnt = pathlen; | |
2339 | } | |
2340 | pathlen -= byte_cnt; | |
2341 | ||
2342 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
2343 | cur_chunk += byte_cnt; | |
2344 | ||
2345 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
2346 | } | |
2347 | } | |
2348 | ||
2349 | /* | |
2350 | * Create the directory entry for the symlink. | |
2351 | */ | |
556b8b16 BN |
2352 | error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, |
2353 | &first_block, &free_list, resblks); | |
f6c2d1fa | 2354 | if (error) |
1da177e4 | 2355 | goto error1; |
1da177e4 LT |
2356 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
2357 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2358 | ||
1da177e4 LT |
2359 | /* |
2360 | * If this is a synchronous mount, make sure that the | |
2361 | * symlink transaction goes to disk before returning to | |
2362 | * the user. | |
2363 | */ | |
2364 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2365 | xfs_trans_set_sync(tp); | |
2366 | } | |
2367 | ||
2368 | /* | |
2369 | * xfs_trans_commit normally decrements the vnode ref count | |
2370 | * when it unlocks the inode. Since we want to return the | |
2371 | * vnode to the caller, we bump the vnode ref count now. | |
2372 | */ | |
2373 | IHOLD(ip); | |
2374 | ||
f7c99b6f | 2375 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2376 | if (error) { |
2377 | goto error2; | |
2378 | } | |
1c72bf90 | 2379 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
7d095257 CH |
2380 | xfs_qm_dqrele(udqp); |
2381 | xfs_qm_dqrele(gdqp); | |
1da177e4 LT |
2382 | |
2383 | /* Fall through to std_return with error = 0 or errno from | |
2384 | * xfs_trans_commit */ | |
2385 | std_return: | |
993386c1 | 2386 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) { |
1da177e4 | 2387 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, |
bc4ac74a CH |
2388 | dp, DM_RIGHT_NULL, |
2389 | error ? NULL : ip, | |
556b8b16 BN |
2390 | DM_RIGHT_NULL, link_name->name, |
2391 | target_path, 0, error, 0); | |
1da177e4 LT |
2392 | } |
2393 | ||
3937be5b CH |
2394 | if (!error) |
2395 | *ipp = ip; | |
1da177e4 LT |
2396 | return error; |
2397 | ||
2398 | error2: | |
2399 | IRELE(ip); | |
2400 | error1: | |
2401 | xfs_bmap_cancel(&free_list); | |
2402 | cancel_flags |= XFS_TRANS_ABORT; | |
2403 | error_return: | |
2404 | xfs_trans_cancel(tp, cancel_flags); | |
7d095257 CH |
2405 | xfs_qm_dqrele(udqp); |
2406 | xfs_qm_dqrele(gdqp); | |
1da177e4 | 2407 | |
993386c1 | 2408 | if (unlock_dp_on_error) |
1da177e4 | 2409 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2410 | |
2411 | goto std_return; | |
2412 | } | |
2413 | ||
1da177e4 | 2414 | int |
993386c1 CH |
2415 | xfs_set_dmattrs( |
2416 | xfs_inode_t *ip, | |
1da177e4 | 2417 | u_int evmask, |
993386c1 | 2418 | u_int16_t state) |
1da177e4 | 2419 | { |
993386c1 | 2420 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 2421 | xfs_trans_t *tp; |
1da177e4 LT |
2422 | int error; |
2423 | ||
2424 | if (!capable(CAP_SYS_ADMIN)) | |
2425 | return XFS_ERROR(EPERM); | |
2426 | ||
1da177e4 LT |
2427 | if (XFS_FORCED_SHUTDOWN(mp)) |
2428 | return XFS_ERROR(EIO); | |
2429 | ||
2430 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
2431 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
2432 | if (error) { | |
2433 | xfs_trans_cancel(tp, 0); | |
2434 | return error; | |
2435 | } | |
2436 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2437 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2438 | ||
613d7043 CH |
2439 | ip->i_d.di_dmevmask = evmask; |
2440 | ip->i_d.di_dmstate = state; | |
1da177e4 LT |
2441 | |
2442 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2443 | IHOLD(ip); | |
1c72bf90 | 2444 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
2445 | |
2446 | return error; | |
2447 | } | |
2448 | ||
993386c1 | 2449 | int |
1da177e4 | 2450 | xfs_reclaim( |
993386c1 | 2451 | xfs_inode_t *ip) |
1da177e4 | 2452 | { |
1da177e4 | 2453 | |
cf441eeb | 2454 | xfs_itrace_entry(ip); |
1da177e4 | 2455 | |
df80c933 | 2456 | ASSERT(!VN_MAPPED(VFS_I(ip))); |
1da177e4 LT |
2457 | |
2458 | /* bad inode, get out here ASAP */ | |
cb4c8cc1 | 2459 | if (is_bad_inode(VFS_I(ip))) { |
1da177e4 LT |
2460 | xfs_ireclaim(ip); |
2461 | return 0; | |
2462 | } | |
2463 | ||
25e41b3d | 2464 | xfs_ioend_wait(ip); |
1da177e4 | 2465 | |
51c91ed5 | 2466 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); |
1da177e4 | 2467 | |
42fe2b1f CH |
2468 | /* |
2469 | * Make sure the atime in the XFS inode is correct before freeing the | |
2470 | * Linux inode. | |
2471 | */ | |
2472 | xfs_synchronize_atime(ip); | |
1da177e4 | 2473 | |
4c60658e DC |
2474 | /* |
2475 | * If we have nothing to flush with this inode then complete the | |
2476 | * teardown now, otherwise break the link between the xfs inode and the | |
2477 | * linux inode and clean up the xfs inode later. This avoids flushing | |
2478 | * the inode to disk during the delete operation itself. | |
2479 | * | |
2480 | * When breaking the link, we need to set the XFS_IRECLAIMABLE flag | |
2481 | * first to ensure that xfs_iunpin() will never see an xfs inode | |
2482 | * that has a linux inode being reclaimed. Synchronisation is provided | |
2483 | * by the i_flags_lock. | |
1da177e4 LT |
2484 | */ |
2485 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
2486 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2487 | xfs_iflock(ip); | |
bf904248 | 2488 | xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
1dc3318a | 2489 | return xfs_reclaim_inode(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); |
1da177e4 | 2490 | } |
11654513 | 2491 | xfs_inode_set_reclaim_tag(ip); |
1da177e4 LT |
2492 | return 0; |
2493 | } | |
1da177e4 LT |
2494 | |
2495 | /* | |
2496 | * xfs_alloc_file_space() | |
2497 | * This routine allocates disk space for the given file. | |
2498 | * | |
2499 | * If alloc_type == 0, this request is for an ALLOCSP type | |
2500 | * request which will change the file size. In this case, no | |
2501 | * DMAPI event will be generated by the call. A TRUNCATE event | |
2502 | * will be generated later by xfs_setattr. | |
2503 | * | |
2504 | * If alloc_type != 0, this request is for a RESVSP type | |
2505 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
2506 | * lower block boundary byte address is less than the file's | |
2507 | * length. | |
2508 | * | |
2509 | * RETURNS: | |
2510 | * 0 on success | |
2511 | * errno on error | |
2512 | * | |
2513 | */ | |
ba0f32d4 | 2514 | STATIC int |
1da177e4 LT |
2515 | xfs_alloc_file_space( |
2516 | xfs_inode_t *ip, | |
2517 | xfs_off_t offset, | |
2518 | xfs_off_t len, | |
2519 | int alloc_type, | |
2520 | int attr_flags) | |
2521 | { | |
dd9f438e NS |
2522 | xfs_mount_t *mp = ip->i_mount; |
2523 | xfs_off_t count; | |
1da177e4 LT |
2524 | xfs_filblks_t allocated_fsb; |
2525 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
2526 | xfs_extlen_t extsz, temp; |
2527 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 2528 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
2529 | int nimaps; |
2530 | int bmapi_flag; | |
2531 | int quota_flag; | |
1da177e4 | 2532 | int rt; |
1da177e4 | 2533 | xfs_trans_t *tp; |
dd9f438e NS |
2534 | xfs_bmbt_irec_t imaps[1], *imapp; |
2535 | xfs_bmap_free_t free_list; | |
2536 | uint qblocks, resblks, resrtextents; | |
2537 | int committed; | |
2538 | int error; | |
1da177e4 | 2539 | |
cf441eeb | 2540 | xfs_itrace_entry(ip); |
1da177e4 LT |
2541 | |
2542 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2543 | return XFS_ERROR(EIO); | |
2544 | ||
7d095257 CH |
2545 | error = xfs_qm_dqattach(ip, 0); |
2546 | if (error) | |
1da177e4 LT |
2547 | return error; |
2548 | ||
2549 | if (len <= 0) | |
2550 | return XFS_ERROR(EINVAL); | |
2551 | ||
957d0ebe DC |
2552 | rt = XFS_IS_REALTIME_INODE(ip); |
2553 | extsz = xfs_get_extsz_hint(ip); | |
2554 | ||
1da177e4 | 2555 | count = len; |
1da177e4 | 2556 | imapp = &imaps[0]; |
dd9f438e NS |
2557 | nimaps = 1; |
2558 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
2559 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
2560 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
2561 | ||
2562 | /* Generate a DMAPI event if needed. */ | |
ba87ea69 | 2563 | if (alloc_type != 0 && offset < ip->i_size && |
0f285c8a | 2564 | (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2565 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
1da177e4 LT |
2566 | xfs_off_t end_dmi_offset; |
2567 | ||
2568 | end_dmi_offset = offset+len; | |
ba87ea69 LM |
2569 | if (end_dmi_offset > ip->i_size) |
2570 | end_dmi_offset = ip->i_size; | |
bc4ac74a CH |
2571 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset, |
2572 | end_dmi_offset - offset, 0, NULL); | |
1da177e4 | 2573 | if (error) |
014c2544 | 2574 | return error; |
1da177e4 LT |
2575 | } |
2576 | ||
2577 | /* | |
dd9f438e | 2578 | * Allocate file space until done or until there is an error |
1da177e4 LT |
2579 | */ |
2580 | retry: | |
2581 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
2582 | xfs_fileoff_t s, e; |
2583 | ||
1da177e4 | 2584 | /* |
3ddb8fa9 | 2585 | * Determine space reservations for data/realtime. |
1da177e4 | 2586 | */ |
dd9f438e | 2587 | if (unlikely(extsz)) { |
1da177e4 | 2588 | s = startoffset_fsb; |
dd9f438e NS |
2589 | do_div(s, extsz); |
2590 | s *= extsz; | |
2591 | e = startoffset_fsb + allocatesize_fsb; | |
2592 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
2593 | e += temp; | |
2594 | if ((temp = do_mod(e, extsz))) | |
2595 | e += extsz - temp; | |
2596 | } else { | |
2597 | s = 0; | |
2598 | e = allocatesize_fsb; | |
2599 | } | |
2600 | ||
2601 | if (unlikely(rt)) { | |
2602 | resrtextents = qblocks = (uint)(e - s); | |
2603 | resrtextents /= mp->m_sb.sb_rextsize; | |
2604 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
2605 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 2606 | } else { |
dd9f438e NS |
2607 | resrtextents = 0; |
2608 | resblks = qblocks = \ | |
2609 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
2610 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
2611 | } |
2612 | ||
2613 | /* | |
dd9f438e | 2614 | * Allocate and setup the transaction. |
1da177e4 LT |
2615 | */ |
2616 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
2617 | error = xfs_trans_reserve(tp, resblks, |
2618 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
2619 | XFS_TRANS_PERM_LOG_RES, |
2620 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 2621 | /* |
dd9f438e | 2622 | * Check for running out of space |
1da177e4 LT |
2623 | */ |
2624 | if (error) { | |
2625 | /* | |
2626 | * Free the transaction structure. | |
2627 | */ | |
2628 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
2629 | xfs_trans_cancel(tp, 0); | |
2630 | break; | |
2631 | } | |
2632 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
2633 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, |
2634 | 0, quota_flag); | |
1da177e4 LT |
2635 | if (error) |
2636 | goto error1; | |
2637 | ||
2638 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2639 | xfs_trans_ihold(tp, ip); | |
2640 | ||
2641 | /* | |
dd9f438e | 2642 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 | 2643 | */ |
9d87c319 | 2644 | xfs_bmap_init(&free_list, &firstfsb); |
541d7d3c | 2645 | error = xfs_bmapi(tp, ip, startoffset_fsb, |
dd9f438e NS |
2646 | allocatesize_fsb, bmapi_flag, |
2647 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 2648 | &free_list, NULL); |
1da177e4 LT |
2649 | if (error) { |
2650 | goto error0; | |
2651 | } | |
2652 | ||
2653 | /* | |
dd9f438e | 2654 | * Complete the transaction |
1da177e4 | 2655 | */ |
f7c99b6f | 2656 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2657 | if (error) { |
2658 | goto error0; | |
2659 | } | |
2660 | ||
1c72bf90 | 2661 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2662 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2663 | if (error) { | |
2664 | break; | |
2665 | } | |
2666 | ||
2667 | allocated_fsb = imapp->br_blockcount; | |
2668 | ||
dd9f438e | 2669 | if (nimaps == 0) { |
1da177e4 LT |
2670 | error = XFS_ERROR(ENOSPC); |
2671 | break; | |
2672 | } | |
2673 | ||
2674 | startoffset_fsb += allocated_fsb; | |
2675 | allocatesize_fsb -= allocated_fsb; | |
2676 | } | |
2677 | dmapi_enospc_check: | |
0f285c8a | 2678 | if (error == ENOSPC && (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2679 | DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) { |
1da177e4 | 2680 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, |
bc4ac74a CH |
2681 | ip, DM_RIGHT_NULL, |
2682 | ip, DM_RIGHT_NULL, | |
1da177e4 LT |
2683 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ |
2684 | if (error == 0) | |
2685 | goto retry; /* Maybe DMAPI app. has made space */ | |
2686 | /* else fall through with error from XFS_SEND_DATA */ | |
2687 | } | |
2688 | ||
2689 | return error; | |
2690 | ||
dd9f438e | 2691 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 2692 | xfs_bmap_cancel(&free_list); |
7d095257 | 2693 | xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e NS |
2694 | |
2695 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
2696 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
2697 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2698 | goto dmapi_enospc_check; | |
2699 | } | |
2700 | ||
2701 | /* | |
2702 | * Zero file bytes between startoff and endoff inclusive. | |
2703 | * The iolock is held exclusive and no blocks are buffered. | |
2fd6f6ec LM |
2704 | * |
2705 | * This function is used by xfs_free_file_space() to zero | |
2706 | * partial blocks when the range to free is not block aligned. | |
2707 | * When unreserving space with boundaries that are not block | |
2708 | * aligned we round up the start and round down the end | |
2709 | * boundaries and then use this function to zero the parts of | |
2710 | * the blocks that got dropped during the rounding. | |
1da177e4 LT |
2711 | */ |
2712 | STATIC int | |
2713 | xfs_zero_remaining_bytes( | |
2714 | xfs_inode_t *ip, | |
2715 | xfs_off_t startoff, | |
2716 | xfs_off_t endoff) | |
2717 | { | |
2718 | xfs_bmbt_irec_t imap; | |
2719 | xfs_fileoff_t offset_fsb; | |
2720 | xfs_off_t lastoffset; | |
2721 | xfs_off_t offset; | |
2722 | xfs_buf_t *bp; | |
2723 | xfs_mount_t *mp = ip->i_mount; | |
2724 | int nimap; | |
2725 | int error = 0; | |
2726 | ||
2fd6f6ec LM |
2727 | /* |
2728 | * Avoid doing I/O beyond eof - it's not necessary | |
2729 | * since nothing can read beyond eof. The space will | |
2730 | * be zeroed when the file is extended anyway. | |
2731 | */ | |
2732 | if (startoff >= ip->i_size) | |
2733 | return 0; | |
2734 | ||
2735 | if (endoff > ip->i_size) | |
2736 | endoff = ip->i_size; | |
2737 | ||
1da177e4 | 2738 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, |
71ddabb9 | 2739 | XFS_IS_REALTIME_INODE(ip) ? |
1da177e4 | 2740 | mp->m_rtdev_targp : mp->m_ddev_targp); |
c6422617 LM |
2741 | if (!bp) |
2742 | return XFS_ERROR(ENOMEM); | |
1da177e4 LT |
2743 | |
2744 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
2745 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
2746 | nimap = 1; | |
541d7d3c | 2747 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, |
3e57ecf6 | 2748 | NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2749 | if (error || nimap < 1) |
2750 | break; | |
2751 | ASSERT(imap.br_blockcount >= 1); | |
2752 | ASSERT(imap.br_startoff == offset_fsb); | |
2753 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
2754 | if (lastoffset > endoff) | |
2755 | lastoffset = endoff; | |
2756 | if (imap.br_startblock == HOLESTARTBLOCK) | |
2757 | continue; | |
2758 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2759 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
2760 | continue; | |
2761 | XFS_BUF_UNDONE(bp); | |
2762 | XFS_BUF_UNWRITE(bp); | |
2763 | XFS_BUF_READ(bp); | |
9d87c319 | 2764 | XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock)); |
1da177e4 | 2765 | xfsbdstrat(mp, bp); |
d64e31a2 DC |
2766 | error = xfs_iowait(bp); |
2767 | if (error) { | |
1da177e4 LT |
2768 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", |
2769 | mp, bp, XFS_BUF_ADDR(bp)); | |
2770 | break; | |
2771 | } | |
2772 | memset(XFS_BUF_PTR(bp) + | |
2773 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
2774 | 0, lastoffset - offset + 1); | |
2775 | XFS_BUF_UNDONE(bp); | |
2776 | XFS_BUF_UNREAD(bp); | |
2777 | XFS_BUF_WRITE(bp); | |
2778 | xfsbdstrat(mp, bp); | |
d64e31a2 DC |
2779 | error = xfs_iowait(bp); |
2780 | if (error) { | |
1da177e4 LT |
2781 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", |
2782 | mp, bp, XFS_BUF_ADDR(bp)); | |
2783 | break; | |
2784 | } | |
2785 | } | |
2786 | xfs_buf_free(bp); | |
2787 | return error; | |
2788 | } | |
2789 | ||
2790 | /* | |
2791 | * xfs_free_file_space() | |
2792 | * This routine frees disk space for the given file. | |
2793 | * | |
2794 | * This routine is only called by xfs_change_file_space | |
2795 | * for an UNRESVSP type call. | |
2796 | * | |
2797 | * RETURNS: | |
2798 | * 0 on success | |
2799 | * errno on error | |
2800 | * | |
2801 | */ | |
2802 | STATIC int | |
2803 | xfs_free_file_space( | |
2804 | xfs_inode_t *ip, | |
2805 | xfs_off_t offset, | |
2806 | xfs_off_t len, | |
2807 | int attr_flags) | |
2808 | { | |
2809 | int committed; | |
2810 | int done; | |
2811 | xfs_off_t end_dmi_offset; | |
2812 | xfs_fileoff_t endoffset_fsb; | |
2813 | int error; | |
2814 | xfs_fsblock_t firstfsb; | |
2815 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
2816 | xfs_bmbt_irec_t imap; |
2817 | xfs_off_t ioffset; | |
2818 | xfs_extlen_t mod=0; | |
2819 | xfs_mount_t *mp; | |
2820 | int nimap; | |
2821 | uint resblks; | |
673cdf5c | 2822 | uint rounding; |
1da177e4 LT |
2823 | int rt; |
2824 | xfs_fileoff_t startoffset_fsb; | |
2825 | xfs_trans_t *tp; | |
5fcbab35 | 2826 | int need_iolock = 1; |
1da177e4 | 2827 | |
1da177e4 LT |
2828 | mp = ip->i_mount; |
2829 | ||
cf441eeb | 2830 | xfs_itrace_entry(ip); |
bd5a876a | 2831 | |
7d095257 CH |
2832 | error = xfs_qm_dqattach(ip, 0); |
2833 | if (error) | |
1da177e4 LT |
2834 | return error; |
2835 | ||
2836 | error = 0; | |
2837 | if (len <= 0) /* if nothing being freed */ | |
2838 | return error; | |
71ddabb9 | 2839 | rt = XFS_IS_REALTIME_INODE(ip); |
1da177e4 LT |
2840 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); |
2841 | end_dmi_offset = offset + len; | |
2842 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
2843 | ||
0f285c8a | 2844 | if (offset < ip->i_size && (attr_flags & XFS_ATTR_DMI) == 0 && |
eb9df39d | 2845 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
ba87ea69 LM |
2846 | if (end_dmi_offset > ip->i_size) |
2847 | end_dmi_offset = ip->i_size; | |
bc4ac74a | 2848 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, |
1da177e4 LT |
2849 | offset, end_dmi_offset - offset, |
2850 | AT_DELAY_FLAG(attr_flags), NULL); | |
2851 | if (error) | |
014c2544 | 2852 | return error; |
1da177e4 LT |
2853 | } |
2854 | ||
0f285c8a | 2855 | if (attr_flags & XFS_ATTR_NOLOCK) |
5fcbab35 | 2856 | need_iolock = 0; |
9fa8046f | 2857 | if (need_iolock) { |
1da177e4 | 2858 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
25e41b3d CH |
2859 | /* wait for the completion of any pending DIOs */ |
2860 | xfs_ioend_wait(ip); | |
9fa8046f | 2861 | } |
5fcbab35 | 2862 | |
e6a4b37f | 2863 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1da177e4 | 2864 | ioffset = offset & ~(rounding - 1); |
bd5a876a | 2865 | |
df80c933 | 2866 | if (VN_CACHED(VFS_I(ip)) != 0) { |
e6a4b37f TS |
2867 | xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1); |
2868 | error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED); | |
d3cf2094 LM |
2869 | if (error) |
2870 | goto out_unlock_iolock; | |
bd5a876a CH |
2871 | } |
2872 | ||
1da177e4 LT |
2873 | /* |
2874 | * Need to zero the stuff we're not freeing, on disk. | |
9da096fd | 2875 | * If it's a realtime file & can't use unwritten extents then we |
1da177e4 LT |
2876 | * actually need to zero the extent edges. Otherwise xfs_bunmapi |
2877 | * will take care of it for us. | |
2878 | */ | |
62118709 | 2879 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
1da177e4 | 2880 | nimap = 1; |
541d7d3c | 2881 | error = xfs_bmapi(NULL, ip, startoffset_fsb, |
3e57ecf6 | 2882 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2883 | if (error) |
2884 | goto out_unlock_iolock; | |
2885 | ASSERT(nimap == 0 || nimap == 1); | |
2886 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
2887 | xfs_daddr_t block; | |
2888 | ||
2889 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2890 | block = imap.br_startblock; | |
2891 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
2892 | if (mod) | |
2893 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
2894 | } | |
2895 | nimap = 1; | |
541d7d3c | 2896 | error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, |
3e57ecf6 | 2897 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
2898 | if (error) |
2899 | goto out_unlock_iolock; | |
2900 | ASSERT(nimap == 0 || nimap == 1); | |
2901 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
2902 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
2903 | mod++; | |
2904 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
2905 | endoffset_fsb -= mod; | |
2906 | } | |
2907 | } | |
2908 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
2909 | /* | |
2910 | * One contiguous piece to clear | |
2911 | */ | |
2912 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
2913 | else { | |
2914 | /* | |
2915 | * Some full blocks, possibly two pieces to clear | |
2916 | */ | |
2917 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
2918 | error = xfs_zero_remaining_bytes(ip, offset, | |
2919 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
2920 | if (!error && | |
2921 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
2922 | error = xfs_zero_remaining_bytes(ip, | |
2923 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
2924 | offset + len - 1); | |
2925 | } | |
2926 | ||
2927 | /* | |
2928 | * free file space until done or until there is an error | |
2929 | */ | |
2930 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
2931 | while (!error && !done) { | |
2932 | ||
2933 | /* | |
91ebecc7 DC |
2934 | * allocate and setup the transaction. Allow this |
2935 | * transaction to dip into the reserve blocks to ensure | |
2936 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
2937 | */ |
2938 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 2939 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
2940 | error = xfs_trans_reserve(tp, |
2941 | resblks, | |
2942 | XFS_WRITE_LOG_RES(mp), | |
2943 | 0, | |
2944 | XFS_TRANS_PERM_LOG_RES, | |
2945 | XFS_WRITE_LOG_COUNT); | |
2946 | ||
2947 | /* | |
2948 | * check for running out of space | |
2949 | */ | |
2950 | if (error) { | |
2951 | /* | |
2952 | * Free the transaction structure. | |
2953 | */ | |
2954 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
2955 | xfs_trans_cancel(tp, 0); | |
2956 | break; | |
2957 | } | |
2958 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
7d095257 CH |
2959 | error = xfs_trans_reserve_quota(tp, mp, |
2960 | ip->i_udquot, ip->i_gdquot, | |
2961 | resblks, 0, XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
2962 | if (error) |
2963 | goto error1; | |
2964 | ||
2965 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2966 | xfs_trans_ihold(tp, ip); | |
2967 | ||
2968 | /* | |
2969 | * issue the bunmapi() call to free the blocks | |
2970 | */ | |
9d87c319 | 2971 | xfs_bmap_init(&free_list, &firstfsb); |
541d7d3c | 2972 | error = xfs_bunmapi(tp, ip, startoffset_fsb, |
1da177e4 | 2973 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 2974 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
2975 | if (error) { |
2976 | goto error0; | |
2977 | } | |
2978 | ||
2979 | /* | |
2980 | * complete the transaction | |
2981 | */ | |
f7c99b6f | 2982 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2983 | if (error) { |
2984 | goto error0; | |
2985 | } | |
2986 | ||
1c72bf90 | 2987 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2988 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
2989 | } | |
2990 | ||
2991 | out_unlock_iolock: | |
2992 | if (need_iolock) | |
2993 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
2994 | return error; | |
2995 | ||
2996 | error0: | |
2997 | xfs_bmap_cancel(&free_list); | |
2998 | error1: | |
2999 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
3000 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
3001 | XFS_ILOCK_EXCL); | |
3002 | return error; | |
3003 | } | |
3004 | ||
3005 | /* | |
3006 | * xfs_change_file_space() | |
3007 | * This routine allocates or frees disk space for the given file. | |
3008 | * The user specified parameters are checked for alignment and size | |
3009 | * limitations. | |
3010 | * | |
3011 | * RETURNS: | |
3012 | * 0 on success | |
3013 | * errno on error | |
3014 | * | |
3015 | */ | |
3016 | int | |
3017 | xfs_change_file_space( | |
993386c1 | 3018 | xfs_inode_t *ip, |
1da177e4 LT |
3019 | int cmd, |
3020 | xfs_flock64_t *bf, | |
3021 | xfs_off_t offset, | |
1da177e4 LT |
3022 | int attr_flags) |
3023 | { | |
993386c1 | 3024 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
3025 | int clrprealloc; |
3026 | int error; | |
3027 | xfs_fsize_t fsize; | |
1da177e4 LT |
3028 | int setprealloc; |
3029 | xfs_off_t startoffset; | |
3030 | xfs_off_t llen; | |
3031 | xfs_trans_t *tp; | |
0f285c8a | 3032 | struct iattr iattr; |
1da177e4 | 3033 | |
cf441eeb | 3034 | xfs_itrace_entry(ip); |
1da177e4 | 3035 | |
993386c1 | 3036 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
3037 | return XFS_ERROR(EINVAL); |
3038 | ||
1da177e4 LT |
3039 | switch (bf->l_whence) { |
3040 | case 0: /*SEEK_SET*/ | |
3041 | break; | |
3042 | case 1: /*SEEK_CUR*/ | |
3043 | bf->l_start += offset; | |
3044 | break; | |
3045 | case 2: /*SEEK_END*/ | |
ba87ea69 | 3046 | bf->l_start += ip->i_size; |
1da177e4 LT |
3047 | break; |
3048 | default: | |
3049 | return XFS_ERROR(EINVAL); | |
3050 | } | |
3051 | ||
3052 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
3053 | ||
3054 | if ( (bf->l_start < 0) | |
3055 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
3056 | || (bf->l_start + llen < 0) | |
3057 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
3058 | return XFS_ERROR(EINVAL); | |
3059 | ||
3060 | bf->l_whence = 0; | |
3061 | ||
3062 | startoffset = bf->l_start; | |
ba87ea69 | 3063 | fsize = ip->i_size; |
1da177e4 LT |
3064 | |
3065 | /* | |
3066 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
3067 | * file space. | |
3068 | * These calls do NOT zero the data space allocated to the file, | |
3069 | * nor do they change the file size. | |
3070 | * | |
3071 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
3072 | * space. | |
3073 | * These calls cause the new file data to be zeroed and the file | |
3074 | * size to be changed. | |
3075 | */ | |
3076 | setprealloc = clrprealloc = 0; | |
3077 | ||
3078 | switch (cmd) { | |
3079 | case XFS_IOC_RESVSP: | |
3080 | case XFS_IOC_RESVSP64: | |
3081 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
3082 | 1, attr_flags); | |
3083 | if (error) | |
3084 | return error; | |
3085 | setprealloc = 1; | |
3086 | break; | |
3087 | ||
3088 | case XFS_IOC_UNRESVSP: | |
3089 | case XFS_IOC_UNRESVSP64: | |
3090 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
3091 | attr_flags))) | |
3092 | return error; | |
3093 | break; | |
3094 | ||
3095 | case XFS_IOC_ALLOCSP: | |
3096 | case XFS_IOC_ALLOCSP64: | |
3097 | case XFS_IOC_FREESP: | |
3098 | case XFS_IOC_FREESP64: | |
3099 | if (startoffset > fsize) { | |
3100 | error = xfs_alloc_file_space(ip, fsize, | |
3101 | startoffset - fsize, 0, attr_flags); | |
3102 | if (error) | |
3103 | break; | |
3104 | } | |
3105 | ||
0f285c8a CH |
3106 | iattr.ia_valid = ATTR_SIZE; |
3107 | iattr.ia_size = startoffset; | |
1da177e4 | 3108 | |
ea5a3dc8 | 3109 | error = xfs_setattr(ip, &iattr, attr_flags); |
1da177e4 LT |
3110 | |
3111 | if (error) | |
3112 | return error; | |
3113 | ||
3114 | clrprealloc = 1; | |
3115 | break; | |
3116 | ||
3117 | default: | |
3118 | ASSERT(0); | |
3119 | return XFS_ERROR(EINVAL); | |
3120 | } | |
3121 | ||
3122 | /* | |
3123 | * update the inode timestamp, mode, and prealloc flag bits | |
3124 | */ | |
3125 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
3126 | ||
3127 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
3128 | 0, 0, 0))) { | |
3129 | /* ASSERT(0); */ | |
3130 | xfs_trans_cancel(tp, 0); | |
3131 | return error; | |
3132 | } | |
3133 | ||
3134 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3135 | ||
3136 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3137 | xfs_trans_ihold(tp, ip); | |
3138 | ||
0f285c8a | 3139 | if ((attr_flags & XFS_ATTR_DMI) == 0) { |
1da177e4 LT |
3140 | ip->i_d.di_mode &= ~S_ISUID; |
3141 | ||
3142 | /* | |
3143 | * Note that we don't have to worry about mandatory | |
3144 | * file locking being disabled here because we only | |
3145 | * clear the S_ISGID bit if the Group execute bit is | |
3146 | * on, but if it was on then mandatory locking wouldn't | |
3147 | * have been enabled. | |
3148 | */ | |
3149 | if (ip->i_d.di_mode & S_IXGRP) | |
3150 | ip->i_d.di_mode &= ~S_ISGID; | |
3151 | ||
3152 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3153 | } | |
3154 | if (setprealloc) | |
3155 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
3156 | else if (clrprealloc) | |
3157 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
3158 | ||
3159 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3160 | xfs_trans_set_sync(tp); | |
3161 | ||
1c72bf90 | 3162 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
3163 | |
3164 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3165 | ||
3166 | return error; | |
3167 | } |