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 | |
993386c1 | 56 | int |
1da177e4 | 57 | xfs_open( |
993386c1 | 58 | xfs_inode_t *ip) |
1da177e4 LT |
59 | { |
60 | int mode; | |
1da177e4 LT |
61 | |
62 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
63 | return XFS_ERROR(EIO); | |
64 | ||
65 | /* | |
66 | * If it's a directory with any blocks, read-ahead block 0 | |
67 | * as we're almost certain to have the next operation be a read there. | |
68 | */ | |
993386c1 | 69 | if (S_ISDIR(ip->i_d.di_mode) && ip->i_d.di_nextents > 0) { |
1da177e4 LT |
70 | mode = xfs_ilock_map_shared(ip); |
71 | if (ip->i_d.di_nextents > 0) | |
72 | (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK); | |
73 | xfs_iunlock(ip, mode); | |
74 | } | |
75 | return 0; | |
76 | } | |
77 | ||
1da177e4 LT |
78 | /* |
79 | * xfs_setattr | |
80 | */ | |
81 | int | |
82 | xfs_setattr( | |
993386c1 | 83 | xfs_inode_t *ip, |
8285fb58 | 84 | bhv_vattr_t *vap, |
1da177e4 LT |
85 | int flags, |
86 | cred_t *credp) | |
87 | { | |
993386c1 | 88 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 89 | xfs_trans_t *tp; |
1da177e4 LT |
90 | int mask; |
91 | int code; | |
92 | uint lock_flags; | |
93 | uint commit_flags=0; | |
94 | uid_t uid=0, iuid=0; | |
95 | gid_t gid=0, igid=0; | |
96 | int timeflags = 0; | |
1da177e4 | 97 | xfs_prid_t projid=0, iprojid=0; |
1da177e4 LT |
98 | struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2; |
99 | int file_owner; | |
5fcbab35 | 100 | int need_iolock = 1; |
1da177e4 | 101 | |
cf441eeb | 102 | xfs_itrace_entry(ip); |
1da177e4 | 103 | |
bd186aa9 | 104 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
105 | return XFS_ERROR(EROFS); |
106 | ||
107 | /* | |
108 | * Cannot set certain attributes. | |
109 | */ | |
110 | mask = vap->va_mask; | |
111 | if (mask & XFS_AT_NOSET) { | |
112 | return XFS_ERROR(EINVAL); | |
113 | } | |
114 | ||
1da177e4 LT |
115 | if (XFS_FORCED_SHUTDOWN(mp)) |
116 | return XFS_ERROR(EIO); | |
117 | ||
118 | /* | |
119 | * Timestamps do not need to be logged and hence do not | |
120 | * need to be done within a transaction. | |
121 | */ | |
122 | if (mask & XFS_AT_UPDTIMES) { | |
123 | ASSERT((mask & ~XFS_AT_UPDTIMES) == 0); | |
124 | timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) | | |
125 | ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) | | |
126 | ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0); | |
127 | xfs_ichgtime(ip, timeflags); | |
128 | return 0; | |
129 | } | |
130 | ||
131 | olddquot1 = olddquot2 = NULL; | |
132 | udqp = gdqp = NULL; | |
133 | ||
134 | /* | |
135 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
136 | * before we start any other transactions. Trying to do this later | |
137 | * is messy. We don't care to take a readlock to look at the ids | |
138 | * in inode here, because we can't hold it across the trans_reserve. | |
139 | * If the IDs do change before we take the ilock, we're covered | |
140 | * because the i_*dquot fields will get updated anyway. | |
141 | */ | |
c8ad20ff NS |
142 | if (XFS_IS_QUOTA_ON(mp) && |
143 | (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) { | |
1da177e4 LT |
144 | uint qflags = 0; |
145 | ||
c8ad20ff | 146 | if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) { |
1da177e4 LT |
147 | uid = vap->va_uid; |
148 | qflags |= XFS_QMOPT_UQUOTA; | |
149 | } else { | |
150 | uid = ip->i_d.di_uid; | |
151 | } | |
c8ad20ff | 152 | if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) { |
1da177e4 LT |
153 | gid = vap->va_gid; |
154 | qflags |= XFS_QMOPT_GQUOTA; | |
155 | } else { | |
156 | gid = ip->i_d.di_gid; | |
157 | } | |
c8ad20ff NS |
158 | if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) { |
159 | projid = vap->va_projid; | |
160 | qflags |= XFS_QMOPT_PQUOTA; | |
161 | } else { | |
162 | projid = ip->i_d.di_projid; | |
163 | } | |
1da177e4 LT |
164 | /* |
165 | * We take a reference when we initialize udqp and gdqp, | |
166 | * so it is important that we never blindly double trip on | |
167 | * the same variable. See xfs_create() for an example. | |
168 | */ | |
169 | ASSERT(udqp == NULL); | |
170 | ASSERT(gdqp == NULL); | |
c8ad20ff NS |
171 | code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags, |
172 | &udqp, &gdqp); | |
1da177e4 | 173 | if (code) |
014c2544 | 174 | return code; |
1da177e4 LT |
175 | } |
176 | ||
177 | /* | |
178 | * For the other attributes, we acquire the inode lock and | |
179 | * first do an error checking pass. | |
180 | */ | |
181 | tp = NULL; | |
182 | lock_flags = XFS_ILOCK_EXCL; | |
5fcbab35 DR |
183 | if (flags & ATTR_NOLOCK) |
184 | need_iolock = 0; | |
1da177e4 LT |
185 | if (!(mask & XFS_AT_SIZE)) { |
186 | if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) || | |
187 | (mp->m_flags & XFS_MOUNT_WSYNC)) { | |
188 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); | |
189 | commit_flags = 0; | |
190 | if ((code = xfs_trans_reserve(tp, 0, | |
191 | XFS_ICHANGE_LOG_RES(mp), 0, | |
192 | 0, 0))) { | |
193 | lock_flags = 0; | |
194 | goto error_return; | |
195 | } | |
196 | } | |
197 | } else { | |
eb9df39d | 198 | if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) && |
1da177e4 LT |
199 | !(flags & ATTR_DMI)) { |
200 | int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR; | |
bc4ac74a | 201 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip, |
1da177e4 LT |
202 | vap->va_size, 0, dmflags, NULL); |
203 | if (code) { | |
204 | lock_flags = 0; | |
205 | goto error_return; | |
206 | } | |
207 | } | |
208 | if (need_iolock) | |
209 | lock_flags |= XFS_IOLOCK_EXCL; | |
210 | } | |
211 | ||
212 | xfs_ilock(ip, lock_flags); | |
213 | ||
214 | /* boolean: are we the file owner? */ | |
215 | file_owner = (current_fsuid(credp) == ip->i_d.di_uid); | |
216 | ||
217 | /* | |
218 | * Change various properties of a file. | |
219 | * Only the owner or users with CAP_FOWNER | |
220 | * capability may do these things. | |
221 | */ | |
222 | if (mask & | |
223 | (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID| | |
224 | XFS_AT_GID|XFS_AT_PROJID)) { | |
225 | /* | |
226 | * CAP_FOWNER overrides the following restrictions: | |
227 | * | |
228 | * The user ID of the calling process must be equal | |
229 | * to the file owner ID, except in cases where the | |
230 | * CAP_FSETID capability is applicable. | |
231 | */ | |
232 | if (!file_owner && !capable(CAP_FOWNER)) { | |
233 | code = XFS_ERROR(EPERM); | |
234 | goto error_return; | |
235 | } | |
236 | ||
237 | /* | |
238 | * CAP_FSETID overrides the following restrictions: | |
239 | * | |
240 | * The effective user ID of the calling process shall match | |
241 | * the file owner when setting the set-user-ID and | |
242 | * set-group-ID bits on that file. | |
243 | * | |
244 | * The effective group ID or one of the supplementary group | |
245 | * IDs of the calling process shall match the group owner of | |
246 | * the file when setting the set-group-ID bit on that file | |
247 | */ | |
248 | if (mask & XFS_AT_MODE) { | |
249 | mode_t m = 0; | |
250 | ||
251 | if ((vap->va_mode & S_ISUID) && !file_owner) | |
252 | m |= S_ISUID; | |
253 | if ((vap->va_mode & S_ISGID) && | |
254 | !in_group_p((gid_t)ip->i_d.di_gid)) | |
255 | m |= S_ISGID; | |
256 | #if 0 | |
257 | /* Linux allows this, Irix doesn't. */ | |
42173f68 | 258 | if ((vap->va_mode & S_ISVTX) && !S_ISDIR(ip->i_d.di_mode)) |
1da177e4 LT |
259 | m |= S_ISVTX; |
260 | #endif | |
261 | if (m && !capable(CAP_FSETID)) | |
262 | vap->va_mode &= ~m; | |
263 | } | |
264 | } | |
265 | ||
266 | /* | |
267 | * Change file ownership. Must be the owner or privileged. | |
268 | * If the system was configured with the "restricted_chown" | |
269 | * option, the owner is not permitted to give away the file, | |
270 | * and can change the group id only to a group of which he | |
271 | * or she is a member. | |
272 | */ | |
273 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
274 | /* | |
275 | * These IDs could have changed since we last looked at them. | |
276 | * But, we're assured that if the ownership did change | |
277 | * while we didn't have the inode locked, inode's dquot(s) | |
278 | * would have changed also. | |
279 | */ | |
280 | iuid = ip->i_d.di_uid; | |
281 | iprojid = ip->i_d.di_projid; | |
282 | igid = ip->i_d.di_gid; | |
283 | gid = (mask & XFS_AT_GID) ? vap->va_gid : igid; | |
284 | uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid; | |
285 | projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid : | |
286 | iprojid; | |
287 | ||
288 | /* | |
289 | * CAP_CHOWN overrides the following restrictions: | |
290 | * | |
291 | * If _POSIX_CHOWN_RESTRICTED is defined, this capability | |
292 | * shall override the restriction that a process cannot | |
293 | * change the user ID of a file it owns and the restriction | |
294 | * that the group ID supplied to the chown() function | |
295 | * shall be equal to either the group ID or one of the | |
296 | * supplementary group IDs of the calling process. | |
1da177e4 LT |
297 | */ |
298 | if (restricted_chown && | |
299 | (iuid != uid || (igid != gid && | |
300 | !in_group_p((gid_t)gid))) && | |
301 | !capable(CAP_CHOWN)) { | |
302 | code = XFS_ERROR(EPERM); | |
303 | goto error_return; | |
304 | } | |
305 | /* | |
c8ad20ff | 306 | * Do a quota reservation only if uid/projid/gid is actually |
1da177e4 LT |
307 | * going to change. |
308 | */ | |
309 | if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || | |
c8ad20ff | 310 | (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) || |
1da177e4 LT |
311 | (XFS_IS_GQUOTA_ON(mp) && igid != gid)) { |
312 | ASSERT(tp); | |
313 | code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp, | |
314 | capable(CAP_FOWNER) ? | |
315 | XFS_QMOPT_FORCE_RES : 0); | |
316 | if (code) /* out of quota */ | |
317 | goto error_return; | |
318 | } | |
319 | } | |
320 | ||
321 | /* | |
322 | * Truncate file. Must have write permission and not be a directory. | |
323 | */ | |
324 | if (mask & XFS_AT_SIZE) { | |
325 | /* Short circuit the truncate case for zero length files */ | |
326 | if ((vap->va_size == 0) && | |
ba87ea69 | 327 | (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) { |
1da177e4 LT |
328 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
329 | lock_flags &= ~XFS_ILOCK_EXCL; | |
330 | if (mask & XFS_AT_CTIME) | |
331 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
332 | code = 0; | |
333 | goto error_return; | |
334 | } | |
335 | ||
42173f68 | 336 | if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
337 | code = XFS_ERROR(EISDIR); |
338 | goto error_return; | |
42173f68 | 339 | } else if (!S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
340 | code = XFS_ERROR(EINVAL); |
341 | goto error_return; | |
342 | } | |
343 | /* | |
344 | * Make sure that the dquots are attached to the inode. | |
345 | */ | |
346 | if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED))) | |
347 | goto error_return; | |
348 | } | |
349 | ||
350 | /* | |
351 | * Change file access or modified times. | |
352 | */ | |
353 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
354 | if (!file_owner) { | |
355 | if ((flags & ATTR_UTIME) && | |
356 | !capable(CAP_FOWNER)) { | |
357 | code = XFS_ERROR(EPERM); | |
358 | goto error_return; | |
359 | } | |
360 | } | |
361 | } | |
362 | ||
363 | /* | |
364 | * Change extent size or realtime flag. | |
365 | */ | |
366 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
367 | /* | |
368 | * Can't change extent size if any extents are allocated. | |
369 | */ | |
e94af02a | 370 | if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) && |
1da177e4 LT |
371 | ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != |
372 | vap->va_extsize) ) { | |
373 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
374 | goto error_return; | |
375 | } | |
376 | ||
1da177e4 LT |
377 | /* |
378 | * Can't change realtime flag if any extents are allocated. | |
379 | */ | |
e94af02a ES |
380 | if ((ip->i_d.di_nextents || ip->i_delayed_blks) && |
381 | (mask & XFS_AT_XFLAGS) && | |
71ddabb9 | 382 | (XFS_IS_REALTIME_INODE(ip)) != |
1da177e4 LT |
383 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { |
384 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
385 | goto error_return; | |
386 | } | |
387 | /* | |
388 | * Extent size must be a multiple of the appropriate block | |
389 | * size, if set at all. | |
390 | */ | |
391 | if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) { | |
392 | xfs_extlen_t size; | |
393 | ||
71ddabb9 | 394 | if (XFS_IS_REALTIME_INODE(ip) || |
1da177e4 LT |
395 | ((mask & XFS_AT_XFLAGS) && |
396 | (vap->va_xflags & XFS_XFLAG_REALTIME))) { | |
397 | size = mp->m_sb.sb_rextsize << | |
398 | mp->m_sb.sb_blocklog; | |
399 | } else { | |
400 | size = mp->m_sb.sb_blocksize; | |
401 | } | |
402 | if (vap->va_extsize % size) { | |
403 | code = XFS_ERROR(EINVAL); | |
404 | goto error_return; | |
405 | } | |
406 | } | |
407 | /* | |
408 | * If realtime flag is set then must have realtime data. | |
409 | */ | |
410 | if ((mask & XFS_AT_XFLAGS) && | |
411 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { | |
412 | if ((mp->m_sb.sb_rblocks == 0) || | |
413 | (mp->m_sb.sb_rextsize == 0) || | |
414 | (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) { | |
415 | code = XFS_ERROR(EINVAL); | |
416 | goto error_return; | |
417 | } | |
418 | } | |
419 | ||
420 | /* | |
421 | * Can't modify an immutable/append-only file unless | |
422 | * we have appropriate permission. | |
423 | */ | |
424 | if ((mask & XFS_AT_XFLAGS) && | |
425 | (ip->i_d.di_flags & | |
426 | (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) || | |
427 | (vap->va_xflags & | |
428 | (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) && | |
429 | !capable(CAP_LINUX_IMMUTABLE)) { | |
430 | code = XFS_ERROR(EPERM); | |
431 | goto error_return; | |
432 | } | |
433 | } | |
434 | ||
435 | /* | |
436 | * Now we can make the changes. Before we join the inode | |
437 | * to the transaction, if XFS_AT_SIZE is set then take care of | |
438 | * the part of the truncation that must be done without the | |
439 | * inode lock. This needs to be done before joining the inode | |
440 | * to the transaction, because the inode cannot be unlocked | |
441 | * once it is a part of the transaction. | |
442 | */ | |
443 | if (mask & XFS_AT_SIZE) { | |
444 | code = 0; | |
ba87ea69 | 445 | if ((vap->va_size > ip->i_size) && |
374e2ac3 | 446 | (flags & ATTR_NOSIZETOK) == 0) { |
1da177e4 | 447 | code = xfs_igrow_start(ip, vap->va_size, credp); |
374e2ac3 | 448 | } |
1da177e4 | 449 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
c32676ee DC |
450 | |
451 | /* | |
452 | * We are going to log the inode size change in this | |
453 | * transaction so any previous writes that are beyond the on | |
454 | * disk EOF and the new EOF that have not been written out need | |
455 | * to be written here. If we do not write the data out, we | |
456 | * expose ourselves to the null files problem. | |
457 | * | |
458 | * Only flush from the on disk size to the smaller of the in | |
459 | * memory file size or the new size as that's the range we | |
460 | * really care about here and prevents waiting for other data | |
461 | * not within the range we care about here. | |
462 | */ | |
463 | if (!code && | |
464 | (ip->i_size != ip->i_d.di_size) && | |
465 | (vap->va_size > ip->i_d.di_size)) { | |
739bfb2a | 466 | code = xfs_flush_pages(ip, |
c32676ee DC |
467 | ip->i_d.di_size, vap->va_size, |
468 | XFS_B_ASYNC, FI_NONE); | |
469 | } | |
470 | ||
471 | /* wait for all I/O to complete */ | |
b677c210 | 472 | vn_iowait(ip); |
c32676ee | 473 | |
1da177e4 LT |
474 | if (!code) |
475 | code = xfs_itruncate_data(ip, vap->va_size); | |
476 | if (code) { | |
477 | ASSERT(tp == NULL); | |
478 | lock_flags &= ~XFS_ILOCK_EXCL; | |
479 | ASSERT(lock_flags == XFS_IOLOCK_EXCL); | |
480 | goto error_return; | |
481 | } | |
482 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); | |
483 | if ((code = xfs_trans_reserve(tp, 0, | |
484 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
485 | XFS_TRANS_PERM_LOG_RES, | |
486 | XFS_ITRUNCATE_LOG_COUNT))) { | |
487 | xfs_trans_cancel(tp, 0); | |
488 | if (need_iolock) | |
489 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
490 | return code; | |
491 | } | |
492 | commit_flags = XFS_TRANS_RELEASE_LOG_RES; | |
493 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
494 | } | |
495 | ||
496 | if (tp) { | |
497 | xfs_trans_ijoin(tp, ip, lock_flags); | |
498 | xfs_trans_ihold(tp, ip); | |
499 | } | |
500 | ||
1da177e4 LT |
501 | /* |
502 | * Truncate file. Must have write permission and not be a directory. | |
503 | */ | |
504 | if (mask & XFS_AT_SIZE) { | |
44d814ce DC |
505 | /* |
506 | * Only change the c/mtime if we are changing the size | |
507 | * or we are explicitly asked to change it. This handles | |
508 | * the semantic difference between truncate() and ftruncate() | |
509 | * as implemented in the VFS. | |
510 | */ | |
511 | if (vap->va_size != ip->i_size || (mask & XFS_AT_CTIME)) | |
512 | timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
513 | ||
ba87ea69 | 514 | if (vap->va_size > ip->i_size) { |
1da177e4 LT |
515 | xfs_igrow_finish(tp, ip, vap->va_size, |
516 | !(flags & ATTR_DMI)); | |
ba87ea69 | 517 | } else if ((vap->va_size <= ip->i_size) || |
1da177e4 LT |
518 | ((vap->va_size == 0) && ip->i_d.di_nextents)) { |
519 | /* | |
520 | * signal a sync transaction unless | |
521 | * we're truncating an already unlinked | |
522 | * file on a wsync filesystem | |
523 | */ | |
524 | code = xfs_itruncate_finish(&tp, ip, | |
525 | (xfs_fsize_t)vap->va_size, | |
526 | XFS_DATA_FORK, | |
527 | ((ip->i_d.di_nlink != 0 || | |
528 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
529 | ? 1 : 0)); | |
7d4fb40a | 530 | if (code) |
1da177e4 | 531 | goto abort_return; |
7d4fb40a NS |
532 | /* |
533 | * Truncated "down", so we're removing references | |
534 | * to old data here - if we now delay flushing for | |
535 | * a long time, we expose ourselves unduly to the | |
536 | * notorious NULL files problem. So, we mark this | |
537 | * vnode and flush it when the file is closed, and | |
538 | * do not wait the usual (long) time for writeout. | |
539 | */ | |
b3aea4ed | 540 | xfs_iflags_set(ip, XFS_ITRUNCATED); |
1da177e4 | 541 | } |
1da177e4 LT |
542 | } |
543 | ||
544 | /* | |
545 | * Change file access modes. | |
546 | */ | |
547 | if (mask & XFS_AT_MODE) { | |
548 | ip->i_d.di_mode &= S_IFMT; | |
549 | ip->i_d.di_mode |= vap->va_mode & ~S_IFMT; | |
550 | ||
551 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
552 | timeflags |= XFS_ICHGTIME_CHG; | |
553 | } | |
554 | ||
555 | /* | |
556 | * Change file ownership. Must be the owner or privileged. | |
557 | * If the system was configured with the "restricted_chown" | |
558 | * option, the owner is not permitted to give away the file, | |
559 | * and can change the group id only to a group of which he | |
560 | * or she is a member. | |
561 | */ | |
562 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
563 | /* | |
564 | * CAP_FSETID overrides the following restrictions: | |
565 | * | |
566 | * The set-user-ID and set-group-ID bits of a file will be | |
567 | * cleared upon successful return from chown() | |
568 | */ | |
569 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
570 | !capable(CAP_FSETID)) { | |
571 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
572 | } | |
573 | ||
574 | /* | |
575 | * Change the ownerships and register quota modifications | |
576 | * in the transaction. | |
577 | */ | |
578 | if (iuid != uid) { | |
579 | if (XFS_IS_UQUOTA_ON(mp)) { | |
580 | ASSERT(mask & XFS_AT_UID); | |
581 | ASSERT(udqp); | |
582 | olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
583 | &ip->i_udquot, udqp); | |
584 | } | |
585 | ip->i_d.di_uid = uid; | |
586 | } | |
587 | if (igid != gid) { | |
588 | if (XFS_IS_GQUOTA_ON(mp)) { | |
c8ad20ff | 589 | ASSERT(!XFS_IS_PQUOTA_ON(mp)); |
1da177e4 LT |
590 | ASSERT(mask & XFS_AT_GID); |
591 | ASSERT(gdqp); | |
592 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
593 | &ip->i_gdquot, gdqp); | |
594 | } | |
595 | ip->i_d.di_gid = gid; | |
596 | } | |
597 | if (iprojid != projid) { | |
c8ad20ff NS |
598 | if (XFS_IS_PQUOTA_ON(mp)) { |
599 | ASSERT(!XFS_IS_GQUOTA_ON(mp)); | |
600 | ASSERT(mask & XFS_AT_PROJID); | |
601 | ASSERT(gdqp); | |
602 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
603 | &ip->i_gdquot, gdqp); | |
604 | } | |
1da177e4 LT |
605 | ip->i_d.di_projid = projid; |
606 | /* | |
607 | * We may have to rev the inode as well as | |
608 | * the superblock version number since projids didn't | |
609 | * exist before DINODE_VERSION_2 and SB_VERSION_NLINK. | |
610 | */ | |
611 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) | |
612 | xfs_bump_ino_vers2(tp, ip); | |
613 | } | |
614 | ||
615 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
616 | timeflags |= XFS_ICHGTIME_CHG; | |
617 | } | |
618 | ||
619 | ||
620 | /* | |
621 | * Change file access or modified times. | |
622 | */ | |
623 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
624 | if (mask & XFS_AT_ATIME) { | |
625 | ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec; | |
626 | ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec; | |
627 | ip->i_update_core = 1; | |
628 | timeflags &= ~XFS_ICHGTIME_ACC; | |
629 | } | |
630 | if (mask & XFS_AT_MTIME) { | |
631 | ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec; | |
632 | ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec; | |
633 | timeflags &= ~XFS_ICHGTIME_MOD; | |
634 | timeflags |= XFS_ICHGTIME_CHG; | |
635 | } | |
636 | if (tp && (flags & ATTR_UTIME)) | |
637 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
638 | } | |
639 | ||
640 | /* | |
641 | * Change XFS-added attributes. | |
642 | */ | |
643 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
644 | if (mask & XFS_AT_EXTSIZE) { | |
645 | /* | |
646 | * Converting bytes to fs blocks. | |
647 | */ | |
648 | ip->i_d.di_extsize = vap->va_extsize >> | |
649 | mp->m_sb.sb_blocklog; | |
650 | } | |
651 | if (mask & XFS_AT_XFLAGS) { | |
652 | uint di_flags; | |
653 | ||
654 | /* can't set PREALLOC this way, just preserve it */ | |
655 | di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC); | |
656 | if (vap->va_xflags & XFS_XFLAG_IMMUTABLE) | |
657 | di_flags |= XFS_DIFLAG_IMMUTABLE; | |
658 | if (vap->va_xflags & XFS_XFLAG_APPEND) | |
659 | di_flags |= XFS_DIFLAG_APPEND; | |
660 | if (vap->va_xflags & XFS_XFLAG_SYNC) | |
661 | di_flags |= XFS_DIFLAG_SYNC; | |
662 | if (vap->va_xflags & XFS_XFLAG_NOATIME) | |
663 | di_flags |= XFS_DIFLAG_NOATIME; | |
664 | if (vap->va_xflags & XFS_XFLAG_NODUMP) | |
665 | di_flags |= XFS_DIFLAG_NODUMP; | |
365ca83d NS |
666 | if (vap->va_xflags & XFS_XFLAG_PROJINHERIT) |
667 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
668 | if (vap->va_xflags & XFS_XFLAG_NODEFRAG) |
669 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
670 | if (vap->va_xflags & XFS_XFLAG_FILESTREAM) |
671 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
1da177e4 LT |
672 | if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { |
673 | if (vap->va_xflags & XFS_XFLAG_RTINHERIT) | |
674 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
675 | if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS) | |
676 | di_flags |= XFS_DIFLAG_NOSYMLINKS; | |
dd9f438e NS |
677 | if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT) |
678 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
679 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
613d7043 | 680 | if (vap->va_xflags & XFS_XFLAG_REALTIME) |
1da177e4 | 681 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
682 | if (vap->va_xflags & XFS_XFLAG_EXTSIZE) |
683 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1da177e4 LT |
684 | } |
685 | ip->i_d.di_flags = di_flags; | |
686 | } | |
687 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
688 | timeflags |= XFS_ICHGTIME_CHG; | |
689 | } | |
690 | ||
691 | /* | |
692 | * Change file inode change time only if XFS_AT_CTIME set | |
693 | * AND we have been called by a DMI function. | |
694 | */ | |
695 | ||
696 | if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) { | |
697 | ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec; | |
698 | ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec; | |
699 | ip->i_update_core = 1; | |
700 | timeflags &= ~XFS_ICHGTIME_CHG; | |
701 | } | |
702 | ||
703 | /* | |
704 | * Send out timestamp changes that need to be set to the | |
705 | * current time. Not done when called by a DMI function. | |
706 | */ | |
707 | if (timeflags && !(flags & ATTR_DMI)) | |
708 | xfs_ichgtime(ip, timeflags); | |
709 | ||
710 | XFS_STATS_INC(xs_ig_attrchg); | |
711 | ||
712 | /* | |
713 | * If this is a synchronous mount, make sure that the | |
714 | * transaction goes to disk before returning to the user. | |
715 | * This is slightly sub-optimal in that truncates require | |
c41564b5 | 716 | * two sync transactions instead of one for wsync filesystems. |
1da177e4 LT |
717 | * One for the truncate and one for the timestamps since we |
718 | * don't want to change the timestamps unless we're sure the | |
719 | * truncate worked. Truncates are less than 1% of the laddis | |
720 | * mix so this probably isn't worth the trouble to optimize. | |
721 | */ | |
722 | code = 0; | |
723 | if (tp) { | |
724 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
725 | xfs_trans_set_sync(tp); | |
726 | ||
1c72bf90 | 727 | code = xfs_trans_commit(tp, commit_flags); |
1da177e4 LT |
728 | } |
729 | ||
1da177e4 LT |
730 | xfs_iunlock(ip, lock_flags); |
731 | ||
732 | /* | |
733 | * Release any dquot(s) the inode had kept before chown. | |
734 | */ | |
735 | XFS_QM_DQRELE(mp, olddquot1); | |
736 | XFS_QM_DQRELE(mp, olddquot2); | |
737 | XFS_QM_DQRELE(mp, udqp); | |
738 | XFS_QM_DQRELE(mp, gdqp); | |
739 | ||
740 | if (code) { | |
741 | return code; | |
742 | } | |
743 | ||
eb9df39d | 744 | if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) && |
1da177e4 | 745 | !(flags & ATTR_DMI)) { |
bc4ac74a | 746 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL, |
1da177e4 LT |
747 | NULL, DM_RIGHT_NULL, NULL, NULL, |
748 | 0, 0, AT_DELAY_FLAG(flags)); | |
749 | } | |
750 | return 0; | |
751 | ||
752 | abort_return: | |
753 | commit_flags |= XFS_TRANS_ABORT; | |
754 | /* FALLTHROUGH */ | |
755 | error_return: | |
756 | XFS_QM_DQRELE(mp, udqp); | |
757 | XFS_QM_DQRELE(mp, gdqp); | |
758 | if (tp) { | |
759 | xfs_trans_cancel(tp, commit_flags); | |
760 | } | |
761 | if (lock_flags != 0) { | |
762 | xfs_iunlock(ip, lock_flags); | |
763 | } | |
764 | return code; | |
765 | } | |
766 | ||
7d4fb40a NS |
767 | /* |
768 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
769 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
770 | * bmapi. | |
771 | */ | |
772 | #define SYMLINK_MAPS 2 | |
773 | ||
804c83c3 CH |
774 | STATIC int |
775 | xfs_readlink_bmap( | |
776 | xfs_inode_t *ip, | |
777 | char *link) | |
778 | { | |
779 | xfs_mount_t *mp = ip->i_mount; | |
780 | int pathlen = ip->i_d.di_size; | |
781 | int nmaps = SYMLINK_MAPS; | |
782 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
783 | xfs_daddr_t d; | |
784 | int byte_cnt; | |
785 | int n; | |
786 | xfs_buf_t *bp; | |
787 | int error = 0; | |
788 | ||
789 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0, | |
790 | mval, &nmaps, NULL, NULL); | |
791 | if (error) | |
792 | goto out; | |
793 | ||
794 | for (n = 0; n < nmaps; n++) { | |
795 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
796 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
797 | ||
798 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); | |
799 | error = XFS_BUF_GETERROR(bp); | |
800 | if (error) { | |
801 | xfs_ioerror_alert("xfs_readlink", | |
802 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
803 | xfs_buf_relse(bp); | |
804 | goto out; | |
805 | } | |
806 | if (pathlen < byte_cnt) | |
807 | byte_cnt = pathlen; | |
808 | pathlen -= byte_cnt; | |
809 | ||
810 | memcpy(link, XFS_BUF_PTR(bp), byte_cnt); | |
811 | xfs_buf_relse(bp); | |
812 | } | |
813 | ||
814 | link[ip->i_d.di_size] = '\0'; | |
815 | error = 0; | |
816 | ||
817 | out: | |
818 | return error; | |
819 | } | |
820 | ||
993386c1 | 821 | int |
1da177e4 | 822 | xfs_readlink( |
993386c1 | 823 | xfs_inode_t *ip, |
804c83c3 | 824 | char *link) |
1da177e4 | 825 | { |
804c83c3 | 826 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 827 | int pathlen; |
1da177e4 | 828 | int error = 0; |
1da177e4 | 829 | |
cf441eeb | 830 | xfs_itrace_entry(ip); |
1da177e4 LT |
831 | |
832 | if (XFS_FORCED_SHUTDOWN(mp)) | |
833 | return XFS_ERROR(EIO); | |
834 | ||
835 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
836 | ||
837 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
804c83c3 | 838 | ASSERT(ip->i_d.di_size <= MAXPATHLEN); |
1da177e4 | 839 | |
804c83c3 CH |
840 | pathlen = ip->i_d.di_size; |
841 | if (!pathlen) | |
842 | goto out; | |
1da177e4 LT |
843 | |
844 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
804c83c3 CH |
845 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
846 | link[pathlen] = '\0'; | |
847 | } else { | |
848 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
849 | } |
850 | ||
804c83c3 | 851 | out: |
1da177e4 | 852 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
853 | return error; |
854 | } | |
855 | ||
1da177e4 LT |
856 | /* |
857 | * xfs_fsync | |
858 | * | |
978b7237 DC |
859 | * This is called to sync the inode and its data out to disk. We need to hold |
860 | * the I/O lock while flushing the data, and the inode lock while flushing the | |
861 | * inode. The inode lock CANNOT be held while flushing the data, so acquire | |
862 | * after we're done with that. | |
1da177e4 | 863 | */ |
993386c1 | 864 | int |
1da177e4 | 865 | xfs_fsync( |
978b7237 | 866 | xfs_inode_t *ip) |
1da177e4 | 867 | { |
1da177e4 LT |
868 | xfs_trans_t *tp; |
869 | int error; | |
f538d4da | 870 | int log_flushed = 0, changed = 1; |
1da177e4 | 871 | |
cf441eeb | 872 | xfs_itrace_entry(ip); |
1da177e4 | 873 | |
1da177e4 LT |
874 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
875 | return XFS_ERROR(EIO); | |
876 | ||
978b7237 DC |
877 | /* capture size updates in I/O completion before writing the inode. */ |
878 | error = filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping); | |
879 | if (error) | |
880 | return XFS_ERROR(error); | |
776a75fa | 881 | |
1da177e4 | 882 | /* |
978b7237 DC |
883 | * We always need to make sure that the required inode state is safe on |
884 | * disk. The vnode might be clean but we still might need to force the | |
885 | * log because of committed transactions that haven't hit the disk yet. | |
886 | * Likewise, there could be unflushed non-transactional changes to the | |
887 | * inode core that have to go to disk and this requires us to issue | |
888 | * a synchronous transaction to capture these changes correctly. | |
1da177e4 | 889 | * |
978b7237 DC |
890 | * This code relies on the assumption that if the update_* fields |
891 | * of the inode are clear and the inode is unpinned then it is clean | |
892 | * and no action is required. | |
1da177e4 LT |
893 | */ |
894 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
895 | ||
978b7237 | 896 | if (!(ip->i_update_size || ip->i_update_core)) { |
1da177e4 | 897 | /* |
978b7237 DC |
898 | * Timestamps/size haven't changed since last inode flush or |
899 | * inode transaction commit. That means either nothing got | |
900 | * written or a transaction committed which caught the updates. | |
901 | * If the latter happened and the transaction hasn't hit the | |
902 | * disk yet, the inode will be still be pinned. If it is, | |
903 | * force the log. | |
1da177e4 LT |
904 | */ |
905 | ||
906 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
907 | ||
908 | if (xfs_ipincount(ip)) { | |
978b7237 DC |
909 | error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0, |
910 | XFS_LOG_FORCE | XFS_LOG_SYNC, | |
f538d4da CH |
911 | &log_flushed); |
912 | } else { | |
913 | /* | |
978b7237 DC |
914 | * If the inode is not pinned and nothing has changed |
915 | * we don't need to flush the cache. | |
f538d4da CH |
916 | */ |
917 | changed = 0; | |
1da177e4 | 918 | } |
1da177e4 LT |
919 | } else { |
920 | /* | |
978b7237 DC |
921 | * Kick off a transaction to log the inode core to get the |
922 | * updates. The sync transaction will also force the log. | |
1da177e4 LT |
923 | */ |
924 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
925 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
978b7237 DC |
926 | error = xfs_trans_reserve(tp, 0, |
927 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0); | |
928 | if (error) { | |
1da177e4 LT |
929 | xfs_trans_cancel(tp, 0); |
930 | return error; | |
931 | } | |
932 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
933 | ||
934 | /* | |
978b7237 DC |
935 | * Note - it's possible that we might have pushed ourselves out |
936 | * of the way during trans_reserve which would flush the inode. | |
937 | * But there's no guarantee that the inode buffer has actually | |
938 | * gone out yet (it's delwri). Plus the buffer could be pinned | |
939 | * anyway if it's part of an inode in another recent | |
940 | * transaction. So we play it safe and fire off the | |
941 | * transaction anyway. | |
1da177e4 LT |
942 | */ |
943 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
944 | xfs_trans_ihold(tp, ip); | |
945 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
978b7237 | 946 | xfs_trans_set_sync(tp); |
1c72bf90 | 947 | error = _xfs_trans_commit(tp, 0, &log_flushed); |
1da177e4 LT |
948 | |
949 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
950 | } | |
f538d4da CH |
951 | |
952 | if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) { | |
953 | /* | |
954 | * If the log write didn't issue an ordered tag we need | |
955 | * to flush the disk cache for the data device now. | |
956 | */ | |
957 | if (!log_flushed) | |
958 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
959 | ||
960 | /* | |
961 | * If this inode is on the RT dev we need to flush that | |
c41564b5 | 962 | * cache as well. |
f538d4da | 963 | */ |
71ddabb9 | 964 | if (XFS_IS_REALTIME_INODE(ip)) |
f538d4da CH |
965 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); |
966 | } | |
967 | ||
1da177e4 LT |
968 | return error; |
969 | } | |
970 | ||
971 | /* | |
92dfe8d2 DC |
972 | * This is called by xfs_inactive to free any blocks beyond eof |
973 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
974 | * punching a hole to EOF. | |
1da177e4 | 975 | */ |
92dfe8d2 DC |
976 | int |
977 | xfs_free_eofblocks( | |
1da177e4 | 978 | xfs_mount_t *mp, |
92dfe8d2 DC |
979 | xfs_inode_t *ip, |
980 | int flags) | |
1da177e4 LT |
981 | { |
982 | xfs_trans_t *tp; | |
983 | int error; | |
984 | xfs_fileoff_t end_fsb; | |
985 | xfs_fileoff_t last_fsb; | |
986 | xfs_filblks_t map_len; | |
987 | int nimaps; | |
988 | xfs_bmbt_irec_t imap; | |
92dfe8d2 | 989 | int use_iolock = (flags & XFS_FREE_EOF_LOCK); |
1da177e4 LT |
990 | |
991 | /* | |
992 | * Figure out if there are any blocks beyond the end | |
993 | * of the file. If not, then there is nothing to do. | |
994 | */ | |
ba87ea69 | 995 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size)); |
1da177e4 LT |
996 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); |
997 | map_len = last_fsb - end_fsb; | |
998 | if (map_len <= 0) | |
014c2544 | 999 | return 0; |
1da177e4 LT |
1000 | |
1001 | nimaps = 1; | |
1002 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
541d7d3c | 1003 | error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0, |
3e57ecf6 | 1004 | NULL, 0, &imap, &nimaps, NULL, NULL); |
1da177e4 LT |
1005 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1006 | ||
1007 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
1008 | (imap.br_startblock != HOLESTARTBLOCK || |
1009 | ip->i_delayed_blks)) { | |
1da177e4 LT |
1010 | /* |
1011 | * Attach the dquots to the inode up front. | |
1012 | */ | |
1013 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1014 | return error; |
1da177e4 LT |
1015 | |
1016 | /* | |
1017 | * There are blocks after the end of file. | |
1018 | * Free them up now by truncating the file to | |
1019 | * its current size. | |
1020 | */ | |
1021 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1022 | ||
1023 | /* | |
1024 | * Do the xfs_itruncate_start() call before | |
1025 | * reserving any log space because | |
1026 | * itruncate_start will call into the buffer | |
1027 | * cache and we can't | |
1028 | * do that within a transaction. | |
1029 | */ | |
92dfe8d2 DC |
1030 | if (use_iolock) |
1031 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 | 1032 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, |
ba87ea69 | 1033 | ip->i_size); |
d3cf2094 | 1034 | if (error) { |
87ae3c24 | 1035 | xfs_trans_cancel(tp, 0); |
92dfe8d2 DC |
1036 | if (use_iolock) |
1037 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 LM |
1038 | return error; |
1039 | } | |
1da177e4 LT |
1040 | |
1041 | error = xfs_trans_reserve(tp, 0, | |
1042 | XFS_ITRUNCATE_LOG_RES(mp), | |
1043 | 0, XFS_TRANS_PERM_LOG_RES, | |
1044 | XFS_ITRUNCATE_LOG_COUNT); | |
1045 | if (error) { | |
1046 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1047 | xfs_trans_cancel(tp, 0); | |
1048 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1049 | return error; |
1da177e4 LT |
1050 | } |
1051 | ||
1052 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1053 | xfs_trans_ijoin(tp, ip, | |
1054 | XFS_IOLOCK_EXCL | | |
1055 | XFS_ILOCK_EXCL); | |
1056 | xfs_trans_ihold(tp, ip); | |
1057 | ||
1058 | error = xfs_itruncate_finish(&tp, ip, | |
ba87ea69 | 1059 | ip->i_size, |
1da177e4 LT |
1060 | XFS_DATA_FORK, |
1061 | 0); | |
1062 | /* | |
1063 | * If we get an error at this point we | |
1064 | * simply don't bother truncating the file. | |
1065 | */ | |
1066 | if (error) { | |
1067 | xfs_trans_cancel(tp, | |
1068 | (XFS_TRANS_RELEASE_LOG_RES | | |
1069 | XFS_TRANS_ABORT)); | |
1070 | } else { | |
1071 | error = xfs_trans_commit(tp, | |
1c72bf90 | 1072 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 1073 | } |
92dfe8d2 DC |
1074 | xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL) |
1075 | : XFS_ILOCK_EXCL)); | |
1da177e4 | 1076 | } |
014c2544 | 1077 | return error; |
1da177e4 LT |
1078 | } |
1079 | ||
1080 | /* | |
1081 | * Free a symlink that has blocks associated with it. | |
1082 | */ | |
1083 | STATIC int | |
1084 | xfs_inactive_symlink_rmt( | |
1085 | xfs_inode_t *ip, | |
1086 | xfs_trans_t **tpp) | |
1087 | { | |
1088 | xfs_buf_t *bp; | |
1089 | int committed; | |
1090 | int done; | |
1091 | int error; | |
1092 | xfs_fsblock_t first_block; | |
1093 | xfs_bmap_free_t free_list; | |
1094 | int i; | |
1095 | xfs_mount_t *mp; | |
1096 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1097 | int nmaps; | |
1098 | xfs_trans_t *ntp; | |
1099 | int size; | |
1100 | xfs_trans_t *tp; | |
1101 | ||
1102 | tp = *tpp; | |
1103 | mp = ip->i_mount; | |
1104 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
1105 | /* | |
1106 | * We're freeing a symlink that has some | |
1107 | * blocks allocated to it. Free the | |
1108 | * blocks here. We know that we've got | |
1109 | * either 1 or 2 extents and that we can | |
1110 | * free them all in one bunmapi call. | |
1111 | */ | |
1112 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
1113 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1114 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1115 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1116 | xfs_trans_cancel(tp, 0); | |
1117 | *tpp = NULL; | |
1118 | return error; | |
1119 | } | |
1120 | /* | |
1121 | * Lock the inode, fix the size, and join it to the transaction. | |
1122 | * Hold it so in the normal path, we still have it locked for | |
1123 | * the second transaction. In the error paths we need it | |
1124 | * held so the cancel won't rele it, see below. | |
1125 | */ | |
1126 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1127 | size = (int)ip->i_d.di_size; | |
1128 | ip->i_d.di_size = 0; | |
1129 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1130 | xfs_trans_ihold(tp, ip); | |
1131 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1132 | /* | |
1133 | * Find the block(s) so we can inval and unmap them. | |
1134 | */ | |
1135 | done = 0; | |
1136 | XFS_BMAP_INIT(&free_list, &first_block); | |
e8c96f8c | 1137 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
1138 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
1139 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 1140 | &free_list, NULL))) |
1da177e4 LT |
1141 | goto error0; |
1142 | /* | |
1143 | * Invalidate the block(s). | |
1144 | */ | |
1145 | for (i = 0; i < nmaps; i++) { | |
1146 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
1147 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
1148 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
1149 | xfs_trans_binval(tp, bp); | |
1150 | } | |
1151 | /* | |
1152 | * Unmap the dead block(s) to the free_list. | |
1153 | */ | |
1154 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 1155 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
1156 | goto error1; |
1157 | ASSERT(done); | |
1158 | /* | |
1159 | * Commit the first transaction. This logs the EFI and the inode. | |
1160 | */ | |
f7c99b6f | 1161 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
1162 | goto error1; |
1163 | /* | |
1164 | * The transaction must have been committed, since there were | |
1165 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
1166 | * The new tp has the extent freeing and EFDs. | |
1167 | */ | |
1168 | ASSERT(committed); | |
1169 | /* | |
1170 | * The first xact was committed, so add the inode to the new one. | |
1171 | * Mark it dirty so it will be logged and moved forward in the log as | |
1172 | * part of every commit. | |
1173 | */ | |
1174 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1175 | xfs_trans_ihold(tp, ip); | |
1176 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1177 | /* | |
1178 | * Get a new, empty transaction to return to our caller. | |
1179 | */ | |
1180 | ntp = xfs_trans_dup(tp); | |
1181 | /* | |
c41564b5 | 1182 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
1183 | * If we get an error on the commit here or on the reserve below, |
1184 | * we need to unlock the inode since the new transaction doesn't | |
1185 | * have the inode attached. | |
1186 | */ | |
1c72bf90 | 1187 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1188 | tp = ntp; |
1189 | if (error) { | |
1190 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1191 | goto error0; | |
1192 | } | |
1193 | /* | |
1194 | * Remove the memory for extent descriptions (just bookkeeping). | |
1195 | */ | |
1196 | if (ip->i_df.if_bytes) | |
1197 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
1198 | ASSERT(ip->i_df.if_bytes == 0); | |
1199 | /* | |
1200 | * Put an itruncate log reservation in the new transaction | |
1201 | * for our caller. | |
1202 | */ | |
1203 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1204 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1205 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1206 | goto error0; | |
1207 | } | |
1208 | /* | |
1209 | * Return with the inode locked but not joined to the transaction. | |
1210 | */ | |
1211 | *tpp = tp; | |
1212 | return 0; | |
1213 | ||
1214 | error1: | |
1215 | xfs_bmap_cancel(&free_list); | |
1216 | error0: | |
1217 | /* | |
1218 | * Have to come here with the inode locked and either | |
1219 | * (held and in the transaction) or (not in the transaction). | |
1220 | * If the inode isn't held then cancel would iput it, but | |
1221 | * that's wrong since this is inactive and the vnode ref | |
1222 | * count is 0 already. | |
1223 | * Cancel won't do anything to the inode if held, but it still | |
1224 | * needs to be locked until the cancel is done, if it was | |
1225 | * joined to the transaction. | |
1226 | */ | |
1227 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1228 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1229 | *tpp = NULL; | |
1230 | return error; | |
1231 | ||
1232 | } | |
1233 | ||
1234 | STATIC int | |
1235 | xfs_inactive_symlink_local( | |
1236 | xfs_inode_t *ip, | |
1237 | xfs_trans_t **tpp) | |
1238 | { | |
1239 | int error; | |
1240 | ||
1241 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
1242 | /* | |
1243 | * We're freeing a symlink which fit into | |
1244 | * the inode. Just free the memory used | |
1245 | * to hold the old symlink. | |
1246 | */ | |
1247 | error = xfs_trans_reserve(*tpp, 0, | |
1248 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
1249 | 0, XFS_TRANS_PERM_LOG_RES, | |
1250 | XFS_ITRUNCATE_LOG_COUNT); | |
1251 | ||
1252 | if (error) { | |
1253 | xfs_trans_cancel(*tpp, 0); | |
1254 | *tpp = NULL; | |
014c2544 | 1255 | return error; |
1da177e4 LT |
1256 | } |
1257 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1258 | ||
1259 | /* | |
1260 | * Zero length symlinks _can_ exist. | |
1261 | */ | |
1262 | if (ip->i_df.if_bytes > 0) { | |
1263 | xfs_idata_realloc(ip, | |
1264 | -(ip->i_df.if_bytes), | |
1265 | XFS_DATA_FORK); | |
1266 | ASSERT(ip->i_df.if_bytes == 0); | |
1267 | } | |
014c2544 | 1268 | return 0; |
1da177e4 LT |
1269 | } |
1270 | ||
1da177e4 LT |
1271 | STATIC int |
1272 | xfs_inactive_attrs( | |
1273 | xfs_inode_t *ip, | |
1274 | xfs_trans_t **tpp) | |
1275 | { | |
1276 | xfs_trans_t *tp; | |
1277 | int error; | |
1278 | xfs_mount_t *mp; | |
1279 | ||
579aa9ca | 1280 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
1da177e4 LT |
1281 | tp = *tpp; |
1282 | mp = ip->i_mount; | |
1283 | ASSERT(ip->i_d.di_forkoff != 0); | |
e5720eec | 1284 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 1285 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
e5720eec DC |
1286 | if (error) |
1287 | goto error_unlock; | |
1da177e4 LT |
1288 | |
1289 | error = xfs_attr_inactive(ip); | |
e5720eec DC |
1290 | if (error) |
1291 | goto error_unlock; | |
1da177e4 LT |
1292 | |
1293 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1294 | error = xfs_trans_reserve(tp, 0, | |
1295 | XFS_IFREE_LOG_RES(mp), | |
1296 | 0, XFS_TRANS_PERM_LOG_RES, | |
1297 | XFS_INACTIVE_LOG_COUNT); | |
e5720eec DC |
1298 | if (error) |
1299 | goto error_cancel; | |
1da177e4 LT |
1300 | |
1301 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1302 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1303 | xfs_trans_ihold(tp, ip); | |
1304 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1305 | ||
1306 | ASSERT(ip->i_d.di_anextents == 0); | |
1307 | ||
1308 | *tpp = tp; | |
014c2544 | 1309 | return 0; |
e5720eec DC |
1310 | |
1311 | error_cancel: | |
1312 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1313 | xfs_trans_cancel(tp, 0); | |
1314 | error_unlock: | |
1315 | *tpp = NULL; | |
1316 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1317 | return error; | |
1da177e4 LT |
1318 | } |
1319 | ||
993386c1 | 1320 | int |
1da177e4 | 1321 | xfs_release( |
993386c1 | 1322 | xfs_inode_t *ip) |
1da177e4 | 1323 | { |
993386c1 CH |
1324 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1325 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
1326 | int error; |
1327 | ||
42173f68 | 1328 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) |
1da177e4 | 1329 | return 0; |
1da177e4 LT |
1330 | |
1331 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1332 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1333 | return 0; |
1334 | ||
2a82b8be | 1335 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
1336 | int truncated; |
1337 | ||
2a82b8be DC |
1338 | /* |
1339 | * If we are using filestreams, and we have an unlinked | |
1340 | * file that we are processing the last close on, then nothing | |
1341 | * will be able to reopen and write to this file. Purge this | |
1342 | * inode from the filestreams cache so that it doesn't delay | |
1343 | * teardown of the inode. | |
1344 | */ | |
1345 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1346 | xfs_filestream_deassociate(ip); | |
1347 | ||
fbf3ce8d CH |
1348 | /* |
1349 | * If we previously truncated this file and removed old data | |
1350 | * in the process, we want to initiate "early" writeout on | |
1351 | * the last close. This is an attempt to combat the notorious | |
1352 | * NULL files problem which is particularly noticable from a | |
1353 | * truncate down, buffered (re-)write (delalloc), followed by | |
1354 | * a crash. What we are effectively doing here is | |
1355 | * significantly reducing the time window where we'd otherwise | |
1356 | * be exposed to that problem. | |
1357 | */ | |
09262b43 | 1358 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
b3aea4ed | 1359 | if (truncated && VN_DIRTY(vp) && ip->i_delayed_blks > 0) |
739bfb2a | 1360 | xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE); |
fbf3ce8d CH |
1361 | } |
1362 | ||
1da177e4 LT |
1363 | if (ip->i_d.di_nlink != 0) { |
1364 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1365 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1366 | ip->i_delayed_blks > 0)) && |
1da177e4 | 1367 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
1368 | (!(ip->i_d.di_flags & |
1369 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
92dfe8d2 DC |
1370 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1371 | if (error) | |
014c2544 | 1372 | return error; |
1da177e4 LT |
1373 | } |
1374 | } | |
1375 | ||
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * xfs_inactive | |
1381 | * | |
1382 | * This is called when the vnode reference count for the vnode | |
1383 | * goes to zero. If the file has been unlinked, then it must | |
1384 | * now be truncated. Also, we clear all of the read-ahead state | |
1385 | * kept for the inode here since the file is now closed. | |
1386 | */ | |
993386c1 | 1387 | int |
1da177e4 | 1388 | xfs_inactive( |
993386c1 | 1389 | xfs_inode_t *ip) |
1da177e4 | 1390 | { |
993386c1 | 1391 | bhv_vnode_t *vp = XFS_ITOV(ip); |
67fcaa73 | 1392 | xfs_bmap_free_t free_list; |
1da177e4 LT |
1393 | xfs_fsblock_t first_block; |
1394 | int committed; | |
1395 | xfs_trans_t *tp; | |
1396 | xfs_mount_t *mp; | |
1397 | int error; | |
1398 | int truncate; | |
1399 | ||
cf441eeb | 1400 | xfs_itrace_entry(ip); |
1da177e4 | 1401 | |
1da177e4 LT |
1402 | /* |
1403 | * If the inode is already free, then there can be nothing | |
1404 | * to clean up here. | |
1405 | */ | |
1406 | if (ip->i_d.di_mode == 0 || VN_BAD(vp)) { | |
1407 | ASSERT(ip->i_df.if_real_bytes == 0); | |
1408 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1409 | return VN_INACTIVE_CACHE; | |
1410 | } | |
1411 | ||
1412 | /* | |
1413 | * Only do a truncate if it's a regular file with | |
1414 | * some actual space in it. It's OK to look at the | |
1415 | * inode's fields without the lock because we're the | |
1416 | * only one with a reference to the inode. | |
1417 | */ | |
1418 | truncate = ((ip->i_d.di_nlink == 0) && | |
ba87ea69 LM |
1419 | ((ip->i_d.di_size != 0) || (ip->i_size != 0) || |
1420 | (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1421 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1422 | ||
1423 | mp = ip->i_mount; | |
1424 | ||
bc4ac74a CH |
1425 | if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) |
1426 | XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL); | |
1da177e4 LT |
1427 | |
1428 | error = 0; | |
1429 | ||
1430 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1431 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1432 | goto out; |
1433 | ||
1434 | if (ip->i_d.di_nlink != 0) { | |
1435 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1436 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1437 | ip->i_delayed_blks > 0)) && |
dd9f438e NS |
1438 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1439 | (!(ip->i_d.di_flags & | |
1440 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1441 | (ip->i_delayed_blks != 0)))) { | |
92dfe8d2 DC |
1442 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1443 | if (error) | |
014c2544 | 1444 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1445 | } |
1446 | goto out; | |
1447 | } | |
1448 | ||
1449 | ASSERT(ip->i_d.di_nlink == 0); | |
1450 | ||
1451 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1452 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1453 | |
1454 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1455 | if (truncate) { | |
1456 | /* | |
1457 | * Do the xfs_itruncate_start() call before | |
1458 | * reserving any log space because itruncate_start | |
1459 | * will call into the buffer cache and we can't | |
1460 | * do that within a transaction. | |
1461 | */ | |
1462 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1463 | ||
d3cf2094 LM |
1464 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); |
1465 | if (error) { | |
87ae3c24 | 1466 | xfs_trans_cancel(tp, 0); |
d3cf2094 LM |
1467 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1468 | return VN_INACTIVE_CACHE; | |
1469 | } | |
1da177e4 LT |
1470 | |
1471 | error = xfs_trans_reserve(tp, 0, | |
1472 | XFS_ITRUNCATE_LOG_RES(mp), | |
1473 | 0, XFS_TRANS_PERM_LOG_RES, | |
1474 | XFS_ITRUNCATE_LOG_COUNT); | |
1475 | if (error) { | |
1476 | /* Don't call itruncate_cleanup */ | |
1477 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1478 | xfs_trans_cancel(tp, 0); | |
1479 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1480 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1481 | } |
1482 | ||
1483 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1484 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1485 | xfs_trans_ihold(tp, ip); | |
1486 | ||
1487 | /* | |
1488 | * normally, we have to run xfs_itruncate_finish sync. | |
1489 | * But if filesystem is wsync and we're in the inactive | |
1490 | * path, then we know that nlink == 0, and that the | |
1491 | * xaction that made nlink == 0 is permanently committed | |
1492 | * since xfs_remove runs as a synchronous transaction. | |
1493 | */ | |
1494 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1495 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1496 | ||
1497 | if (error) { | |
1498 | xfs_trans_cancel(tp, | |
1499 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1500 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1501 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1502 | } |
1503 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1504 | ||
1505 | /* | |
1506 | * If we get an error while cleaning up a | |
1507 | * symlink we bail out. | |
1508 | */ | |
1509 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1510 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1511 | xfs_inactive_symlink_local(ip, &tp); | |
1512 | ||
1513 | if (error) { | |
1514 | ASSERT(tp == NULL); | |
014c2544 | 1515 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1516 | } |
1517 | ||
1518 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1519 | xfs_trans_ihold(tp, ip); | |
1520 | } else { | |
1521 | error = xfs_trans_reserve(tp, 0, | |
1522 | XFS_IFREE_LOG_RES(mp), | |
1523 | 0, XFS_TRANS_PERM_LOG_RES, | |
1524 | XFS_INACTIVE_LOG_COUNT); | |
1525 | if (error) { | |
1526 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1527 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1528 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1529 | } |
1530 | ||
1531 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1532 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1533 | xfs_trans_ihold(tp, ip); | |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * If there are attributes associated with the file | |
1538 | * then blow them away now. The code calls a routine | |
1539 | * that recursively deconstructs the attribute fork. | |
1540 | * We need to just commit the current transaction | |
1541 | * because we can't use it for xfs_attr_inactive(). | |
1542 | */ | |
1543 | if (ip->i_d.di_anextents > 0) { | |
1544 | error = xfs_inactive_attrs(ip, &tp); | |
1545 | /* | |
1546 | * If we got an error, the transaction is already | |
1547 | * cancelled, and the inode is unlocked. Just get out. | |
1548 | */ | |
1549 | if (error) | |
014c2544 | 1550 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1551 | } else if (ip->i_afp) { |
1552 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1553 | } | |
1554 | ||
1555 | /* | |
1556 | * Free the inode. | |
1557 | */ | |
1558 | XFS_BMAP_INIT(&free_list, &first_block); | |
1559 | error = xfs_ifree(tp, ip, &free_list); | |
1560 | if (error) { | |
1561 | /* | |
1562 | * If we fail to free the inode, shut down. The cancel | |
1563 | * might do that, we need to make sure. Otherwise the | |
1564 | * inode might be lost for a long time or forever. | |
1565 | */ | |
1566 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1567 | cmn_err(CE_NOTE, | |
1568 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1569 | error, mp->m_fsname); | |
7d04a335 | 1570 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1571 | } |
1572 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1573 | } else { | |
1574 | /* | |
1575 | * Credit the quota account(s). The inode is gone. | |
1576 | */ | |
1577 | XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1578 | ||
1579 | /* | |
78e9da77 DC |
1580 | * Just ignore errors at this point. There is nothing we can |
1581 | * do except to try to keep going. Make sure it's not a silent | |
1582 | * error. | |
1da177e4 | 1583 | */ |
78e9da77 DC |
1584 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1585 | if (error) | |
1586 | xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: " | |
1587 | "xfs_bmap_finish() returned error %d", error); | |
1588 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1589 | if (error) | |
1590 | xfs_fs_cmn_err(CE_NOTE, mp, "xfs_inactive: " | |
1591 | "xfs_trans_commit() returned error %d", error); | |
1da177e4 LT |
1592 | } |
1593 | /* | |
1594 | * Release the dquots held by inode, if any. | |
1595 | */ | |
1596 | XFS_QM_DQDETACH(mp, ip); | |
1597 | ||
1598 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1599 | ||
1600 | out: | |
1601 | return VN_INACTIVE_CACHE; | |
1602 | } | |
1603 | ||
1604 | ||
993386c1 | 1605 | int |
1da177e4 | 1606 | xfs_lookup( |
993386c1 | 1607 | xfs_inode_t *dp, |
556b8b16 | 1608 | struct xfs_name *name, |
ef1f5e7a | 1609 | xfs_inode_t **ipp) |
1da177e4 | 1610 | { |
eca450b7 | 1611 | xfs_ino_t inum; |
1da177e4 LT |
1612 | int error; |
1613 | uint lock_mode; | |
1da177e4 | 1614 | |
cf441eeb | 1615 | xfs_itrace_entry(dp); |
1da177e4 LT |
1616 | |
1617 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1618 | return XFS_ERROR(EIO); | |
1619 | ||
1620 | lock_mode = xfs_ilock_map_shared(dp); | |
eca450b7 | 1621 | error = xfs_dir_lookup(NULL, dp, name, &inum); |
1da177e4 | 1622 | xfs_iunlock_map_shared(dp, lock_mode); |
eca450b7 CH |
1623 | |
1624 | if (error) | |
1625 | goto out; | |
1626 | ||
1627 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0); | |
1628 | if (error) | |
1629 | goto out; | |
1630 | ||
1631 | xfs_itrace_ref(*ipp); | |
1632 | return 0; | |
1633 | ||
1634 | out: | |
1635 | *ipp = NULL; | |
1da177e4 LT |
1636 | return error; |
1637 | } | |
1638 | ||
993386c1 | 1639 | int |
1da177e4 | 1640 | xfs_create( |
993386c1 | 1641 | xfs_inode_t *dp, |
556b8b16 | 1642 | struct xfs_name *name, |
3e5daf05 CH |
1643 | mode_t mode, |
1644 | xfs_dev_t rdev, | |
979ebab1 | 1645 | xfs_inode_t **ipp, |
1da177e4 LT |
1646 | cred_t *credp) |
1647 | { | |
556b8b16 | 1648 | xfs_mount_t *mp = dp->i_mount; |
993386c1 | 1649 | xfs_inode_t *ip; |
1da177e4 | 1650 | xfs_trans_t *tp; |
556b8b16 | 1651 | int error; |
1da177e4 LT |
1652 | xfs_bmap_free_t free_list; |
1653 | xfs_fsblock_t first_block; | |
993386c1 | 1654 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1655 | int dm_event_sent = 0; |
1656 | uint cancel_flags; | |
1657 | int committed; | |
1658 | xfs_prid_t prid; | |
1659 | struct xfs_dquot *udqp, *gdqp; | |
1660 | uint resblks; | |
1da177e4 | 1661 | |
979ebab1 | 1662 | ASSERT(!*ipp); |
cf441eeb | 1663 | xfs_itrace_entry(dp); |
1da177e4 | 1664 | |
eb9df39d | 1665 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 | 1666 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
bc4ac74a | 1667 | dp, DM_RIGHT_NULL, NULL, |
556b8b16 | 1668 | DM_RIGHT_NULL, name->name, NULL, |
3e5daf05 | 1669 | mode, 0, 0); |
1da177e4 LT |
1670 | |
1671 | if (error) | |
1672 | return error; | |
1673 | dm_event_sent = 1; | |
1674 | } | |
1675 | ||
1676 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1677 | return XFS_ERROR(EIO); | |
1678 | ||
1679 | /* Return through std_return after this point. */ | |
1680 | ||
1681 | udqp = gdqp = NULL; | |
365ca83d NS |
1682 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1683 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
1684 | else |
1685 | prid = (xfs_prid_t)dfltprid; | |
1686 | ||
1687 | /* | |
1688 | * Make sure that we have allocated dquot(s) on disk. | |
1689 | */ | |
1690 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 1691 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
1692 | XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1693 | if (error) | |
1694 | goto std_return; | |
1695 | ||
1696 | ip = NULL; | |
1da177e4 LT |
1697 | |
1698 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1699 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
556b8b16 | 1700 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); |
1da177e4 LT |
1701 | /* |
1702 | * Initially assume that the file does not exist and | |
1703 | * reserve the resources for that case. If that is not | |
1704 | * the case we'll drop the one we have and get a more | |
1705 | * appropriate transaction later. | |
1706 | */ | |
1707 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1708 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1709 | if (error == ENOSPC) { | |
1710 | resblks = 0; | |
1711 | error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0, | |
1712 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1713 | } | |
1714 | if (error) { | |
1715 | cancel_flags = 0; | |
1da177e4 LT |
1716 | goto error_return; |
1717 | } | |
1718 | ||
f7c66ce3 | 1719 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1720 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
1721 | |
1722 | XFS_BMAP_INIT(&free_list, &first_block); | |
1723 | ||
1724 | ASSERT(ip == NULL); | |
1725 | ||
1726 | /* | |
1727 | * Reserve disk quota and the inode. | |
1728 | */ | |
1729 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
1730 | if (error) | |
1731 | goto error_return; | |
1732 | ||
556b8b16 BN |
1733 | error = xfs_dir_canenter(tp, dp, name, resblks); |
1734 | if (error) | |
1da177e4 | 1735 | goto error_return; |
3e5daf05 | 1736 | error = xfs_dir_ialloc(&tp, dp, mode, 1, |
1da177e4 LT |
1737 | rdev, credp, prid, resblks > 0, |
1738 | &ip, &committed); | |
1739 | if (error) { | |
1740 | if (error == ENOSPC) | |
1741 | goto error_return; | |
1742 | goto abort_return; | |
1743 | } | |
cf441eeb | 1744 | xfs_itrace_ref(ip); |
1da177e4 LT |
1745 | |
1746 | /* | |
1747 | * At this point, we've gotten a newly allocated inode. | |
1748 | * It is locked (and joined to the transaction). | |
1749 | */ | |
1750 | ||
579aa9ca | 1751 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
1752 | |
1753 | /* | |
993386c1 CH |
1754 | * Now we join the directory inode to the transaction. We do not do it |
1755 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1756 | * (and release all the locks). An error from here on will result in | |
1757 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1758 | * error path. | |
1da177e4 | 1759 | */ |
979ebab1 | 1760 | IHOLD(dp); |
1da177e4 | 1761 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 1762 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 1763 | |
556b8b16 | 1764 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
f6c2d1fa NS |
1765 | &first_block, &free_list, resblks ? |
1766 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1767 | if (error) { |
1768 | ASSERT(error != ENOSPC); | |
1769 | goto abort_return; | |
1770 | } | |
1771 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1772 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1773 | ||
1774 | /* | |
1775 | * If this is a synchronous mount, make sure that the | |
1776 | * create transaction goes to disk before returning to | |
1777 | * the user. | |
1778 | */ | |
1779 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1780 | xfs_trans_set_sync(tp); | |
1781 | } | |
1782 | ||
1783 | dp->i_gen++; | |
1784 | ||
1785 | /* | |
1786 | * Attach the dquot(s) to the inodes and modify them incore. | |
1787 | * These ids of the inode couldn't have changed since the new | |
1788 | * inode has been locked ever since it was created. | |
1789 | */ | |
1790 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
1791 | ||
1792 | /* | |
1793 | * xfs_trans_commit normally decrements the vnode ref count | |
1794 | * when it unlocks the inode. Since we want to return the | |
1795 | * vnode to the caller, we bump the vnode ref count now. | |
1796 | */ | |
1797 | IHOLD(ip); | |
1da177e4 | 1798 | |
f7c99b6f | 1799 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1800 | if (error) { |
1801 | xfs_bmap_cancel(&free_list); | |
1802 | goto abort_rele; | |
1803 | } | |
1804 | ||
1c72bf90 | 1805 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1806 | if (error) { |
1807 | IRELE(ip); | |
1808 | tp = NULL; | |
1809 | goto error_return; | |
1810 | } | |
1811 | ||
1812 | XFS_QM_DQRELE(mp, udqp); | |
1813 | XFS_QM_DQRELE(mp, gdqp); | |
1814 | ||
979ebab1 | 1815 | *ipp = ip; |
1da177e4 LT |
1816 | |
1817 | /* Fallthrough to std_return with error = 0 */ | |
1818 | ||
1819 | std_return: | |
979ebab1 | 1820 | if ((*ipp || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 1821 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 | 1822 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
bc4ac74a | 1823 | dp, DM_RIGHT_NULL, |
979ebab1 | 1824 | *ipp ? ip : NULL, |
556b8b16 | 1825 | DM_RIGHT_NULL, name->name, NULL, |
3e5daf05 | 1826 | mode, error, 0); |
1da177e4 LT |
1827 | } |
1828 | return error; | |
1829 | ||
1830 | abort_return: | |
1831 | cancel_flags |= XFS_TRANS_ABORT; | |
1832 | /* FALLTHROUGH */ | |
1da177e4 | 1833 | |
014c2544 | 1834 | error_return: |
1da177e4 LT |
1835 | if (tp != NULL) |
1836 | xfs_trans_cancel(tp, cancel_flags); | |
1837 | ||
1da177e4 LT |
1838 | XFS_QM_DQRELE(mp, udqp); |
1839 | XFS_QM_DQRELE(mp, gdqp); | |
1840 | ||
993386c1 CH |
1841 | if (unlock_dp_on_error) |
1842 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1843 | ||
1da177e4 LT |
1844 | goto std_return; |
1845 | ||
1846 | abort_rele: | |
1847 | /* | |
1848 | * Wait until after the current transaction is aborted to | |
1849 | * release the inode. This prevents recursive transactions | |
1850 | * and deadlocks from xfs_inactive. | |
1851 | */ | |
1852 | cancel_flags |= XFS_TRANS_ABORT; | |
1853 | xfs_trans_cancel(tp, cancel_flags); | |
1854 | IRELE(ip); | |
1855 | ||
1856 | XFS_QM_DQRELE(mp, udqp); | |
1857 | XFS_QM_DQRELE(mp, gdqp); | |
1858 | ||
1859 | goto std_return; | |
1860 | } | |
1861 | ||
1862 | #ifdef DEBUG | |
1863 | /* | |
1864 | * Some counters to see if (and how often) we are hitting some deadlock | |
1865 | * prevention code paths. | |
1866 | */ | |
1867 | ||
1868 | int xfs_rm_locks; | |
1869 | int xfs_rm_lock_delays; | |
1870 | int xfs_rm_attempts; | |
1871 | #endif | |
1872 | ||
1873 | /* | |
1874 | * The following routine will lock the inodes associated with the | |
1875 | * directory and the named entry in the directory. The locks are | |
1876 | * acquired in increasing inode number. | |
1877 | * | |
1878 | * If the entry is "..", then only the directory is locked. The | |
1879 | * vnode ref count will still include that from the .. entry in | |
1880 | * this case. | |
1881 | * | |
1882 | * There is a deadlock we need to worry about. If the locked directory is | |
1883 | * in the AIL, it might be blocking up the log. The next inode we lock | |
1884 | * could be already locked by another thread waiting for log space (e.g | |
1885 | * a permanent log reservation with a long running transaction (see | |
1886 | * xfs_itruncate_finish)). To solve this, we must check if the directory | |
1887 | * is in the ail and use lock_nowait. If we can't lock, we need to | |
1888 | * drop the inode lock on the directory and try again. xfs_iunlock will | |
1889 | * potentially push the tail if we were holding up the log. | |
1890 | */ | |
1891 | STATIC int | |
1892 | xfs_lock_dir_and_entry( | |
1893 | xfs_inode_t *dp, | |
1da177e4 LT |
1894 | xfs_inode_t *ip) /* inode of entry 'name' */ |
1895 | { | |
1896 | int attempts; | |
1897 | xfs_ino_t e_inum; | |
1898 | xfs_inode_t *ips[2]; | |
1899 | xfs_log_item_t *lp; | |
1900 | ||
1901 | #ifdef DEBUG | |
1902 | xfs_rm_locks++; | |
1903 | #endif | |
1904 | attempts = 0; | |
1905 | ||
1906 | again: | |
f7c66ce3 | 1907 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
1da177e4 LT |
1908 | |
1909 | e_inum = ip->i_ino; | |
1910 | ||
cf441eeb | 1911 | xfs_itrace_ref(ip); |
1da177e4 LT |
1912 | |
1913 | /* | |
1914 | * We want to lock in increasing inum. Since we've already | |
1915 | * acquired the lock on the directory, we may need to release | |
1916 | * if if the inum of the entry turns out to be less. | |
1917 | */ | |
1918 | if (e_inum > dp->i_ino) { | |
1919 | /* | |
1920 | * We are already in the right order, so just | |
1921 | * lock on the inode of the entry. | |
1922 | * We need to use nowait if dp is in the AIL. | |
1923 | */ | |
1924 | ||
1925 | lp = (xfs_log_item_t *)dp->i_itemp; | |
1926 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
1927 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
1928 | attempts++; | |
1929 | #ifdef DEBUG | |
1930 | xfs_rm_attempts++; | |
1931 | #endif | |
1932 | ||
1933 | /* | |
1934 | * Unlock dp and try again. | |
1935 | * xfs_iunlock will try to push the tail | |
1936 | * if the inode is in the AIL. | |
1937 | */ | |
1938 | ||
1939 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1940 | ||
1941 | if ((attempts % 5) == 0) { | |
1942 | delay(1); /* Don't just spin the CPU */ | |
1943 | #ifdef DEBUG | |
1944 | xfs_rm_lock_delays++; | |
1945 | #endif | |
1946 | } | |
1947 | goto again; | |
1948 | } | |
1949 | } else { | |
1950 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1951 | } | |
1952 | } else if (e_inum < dp->i_ino) { | |
1953 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1954 | ||
1955 | ips[0] = ip; | |
1956 | ips[1] = dp; | |
cfa853e4 | 1957 | xfs_lock_inodes(ips, 2, XFS_ILOCK_EXCL); |
1da177e4 LT |
1958 | } |
1959 | /* else e_inum == dp->i_ino */ | |
1960 | /* This can happen if we're asked to lock /x/.. | |
1961 | * the entry is "..", which is also the parent directory. | |
1962 | */ | |
1963 | ||
1964 | return 0; | |
1965 | } | |
1966 | ||
1967 | #ifdef DEBUG | |
1968 | int xfs_locked_n; | |
1969 | int xfs_small_retries; | |
1970 | int xfs_middle_retries; | |
1971 | int xfs_lots_retries; | |
1972 | int xfs_lock_delays; | |
1973 | #endif | |
1974 | ||
f7c66ce3 LM |
1975 | /* |
1976 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
1977 | * a different value | |
1978 | */ | |
1979 | static inline int | |
1980 | xfs_lock_inumorder(int lock_mode, int subclass) | |
1981 | { | |
1982 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 1983 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 1984 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 1985 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
1986 | |
1987 | return lock_mode; | |
1988 | } | |
1989 | ||
1da177e4 LT |
1990 | /* |
1991 | * The following routine will lock n inodes in exclusive mode. | |
1992 | * We assume the caller calls us with the inodes in i_ino order. | |
1993 | * | |
1994 | * We need to detect deadlock where an inode that we lock | |
1995 | * is in the AIL and we start waiting for another inode that is locked | |
1996 | * by a thread in a long running transaction (such as truncate). This can | |
1997 | * result in deadlock since the long running trans might need to wait | |
1998 | * for the inode we just locked in order to push the tail and free space | |
1999 | * in the log. | |
2000 | */ | |
2001 | void | |
2002 | xfs_lock_inodes( | |
2003 | xfs_inode_t **ips, | |
2004 | int inodes, | |
1da177e4 LT |
2005 | uint lock_mode) |
2006 | { | |
2007 | int attempts = 0, i, j, try_lock; | |
2008 | xfs_log_item_t *lp; | |
2009 | ||
2010 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
2011 | ||
cfa853e4 CH |
2012 | try_lock = 0; |
2013 | i = 0; | |
1da177e4 LT |
2014 | |
2015 | again: | |
2016 | for (; i < inodes; i++) { | |
2017 | ASSERT(ips[i]); | |
2018 | ||
2019 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
2020 | continue; | |
2021 | ||
2022 | /* | |
2023 | * If try_lock is not set yet, make sure all locked inodes | |
2024 | * are not in the AIL. | |
2025 | * If any are, set try_lock to be used later. | |
2026 | */ | |
2027 | ||
2028 | if (!try_lock) { | |
2029 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
2030 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
2031 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2032 | try_lock++; | |
2033 | } | |
2034 | } | |
2035 | } | |
2036 | ||
2037 | /* | |
2038 | * If any of the previous locks we have locked is in the AIL, | |
2039 | * we must TRY to get the second and subsequent locks. If | |
2040 | * we can't get any, we must release all we have | |
2041 | * and try again. | |
2042 | */ | |
2043 | ||
2044 | if (try_lock) { | |
2045 | /* try_lock must be 0 if i is 0. */ | |
2046 | /* | |
2047 | * try_lock means we have an inode locked | |
2048 | * that is in the AIL. | |
2049 | */ | |
2050 | ASSERT(i != 0); | |
f7c66ce3 | 2051 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
2052 | attempts++; |
2053 | ||
2054 | /* | |
2055 | * Unlock all previous guys and try again. | |
2056 | * xfs_iunlock will try to push the tail | |
2057 | * if the inode is in the AIL. | |
2058 | */ | |
2059 | ||
2060 | for(j = i - 1; j >= 0; j--) { | |
2061 | ||
2062 | /* | |
2063 | * Check to see if we've already | |
2064 | * unlocked this one. | |
2065 | * Not the first one going back, | |
2066 | * and the inode ptr is the same. | |
2067 | */ | |
2068 | if ((j != (i - 1)) && ips[j] == | |
2069 | ips[j+1]) | |
2070 | continue; | |
2071 | ||
2072 | xfs_iunlock(ips[j], lock_mode); | |
2073 | } | |
2074 | ||
2075 | if ((attempts % 5) == 0) { | |
2076 | delay(1); /* Don't just spin the CPU */ | |
2077 | #ifdef DEBUG | |
2078 | xfs_lock_delays++; | |
2079 | #endif | |
2080 | } | |
2081 | i = 0; | |
2082 | try_lock = 0; | |
2083 | goto again; | |
2084 | } | |
2085 | } else { | |
f7c66ce3 | 2086 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
2087 | } |
2088 | } | |
2089 | ||
2090 | #ifdef DEBUG | |
2091 | if (attempts) { | |
2092 | if (attempts < 5) xfs_small_retries++; | |
2093 | else if (attempts < 100) xfs_middle_retries++; | |
2094 | else xfs_lots_retries++; | |
2095 | } else { | |
2096 | xfs_locked_n++; | |
2097 | } | |
2098 | #endif | |
2099 | } | |
2100 | ||
2101 | #ifdef DEBUG | |
2102 | #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);} | |
2103 | int remove_which_error_return = 0; | |
2104 | #else /* ! DEBUG */ | |
2105 | #define REMOVE_DEBUG_TRACE(x) | |
2106 | #endif /* ! DEBUG */ | |
2107 | ||
993386c1 | 2108 | int |
1da177e4 | 2109 | xfs_remove( |
993386c1 | 2110 | xfs_inode_t *dp, |
556b8b16 BN |
2111 | struct xfs_name *name, |
2112 | xfs_inode_t *ip) | |
1da177e4 | 2113 | { |
993386c1 | 2114 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 2115 | xfs_trans_t *tp = NULL; |
1da177e4 LT |
2116 | int error = 0; |
2117 | xfs_bmap_free_t free_list; | |
2118 | xfs_fsblock_t first_block; | |
2119 | int cancel_flags; | |
2120 | int committed; | |
1da177e4 LT |
2121 | int link_zero; |
2122 | uint resblks; | |
1da177e4 | 2123 | |
cf441eeb | 2124 | xfs_itrace_entry(dp); |
1da177e4 | 2125 | |
1da177e4 LT |
2126 | if (XFS_FORCED_SHUTDOWN(mp)) |
2127 | return XFS_ERROR(EIO); | |
2128 | ||
eb9df39d | 2129 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
556b8b16 BN |
2130 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, DM_RIGHT_NULL, |
2131 | NULL, DM_RIGHT_NULL, name->name, NULL, | |
2132 | ip->i_d.di_mode, 0, 0); | |
1da177e4 LT |
2133 | if (error) |
2134 | return error; | |
2135 | } | |
2136 | ||
cf441eeb LM |
2137 | xfs_itrace_entry(ip); |
2138 | xfs_itrace_ref(ip); | |
1da177e4 LT |
2139 | |
2140 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
82dab941 | 2141 | if (!error) |
1da177e4 LT |
2142 | error = XFS_QM_DQATTACH(mp, ip, 0); |
2143 | if (error) { | |
2144 | REMOVE_DEBUG_TRACE(__LINE__); | |
1da177e4 LT |
2145 | goto std_return; |
2146 | } | |
2147 | ||
2148 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2149 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2150 | /* | |
2151 | * We try to get the real space reservation first, | |
2152 | * allowing for directory btree deletion(s) implying | |
2153 | * possible bmap insert(s). If we can't get the space | |
2154 | * reservation then we use 0 instead, and avoid the bmap | |
2155 | * btree insert(s) in the directory code by, if the bmap | |
2156 | * insert tries to happen, instead trimming the LAST | |
2157 | * block from the directory. | |
2158 | */ | |
2159 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2160 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2161 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2162 | if (error == ENOSPC) { | |
2163 | resblks = 0; | |
2164 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2165 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2166 | } | |
2167 | if (error) { | |
2168 | ASSERT(error != ENOSPC); | |
2169 | REMOVE_DEBUG_TRACE(__LINE__); | |
2170 | xfs_trans_cancel(tp, 0); | |
1da177e4 LT |
2171 | return error; |
2172 | } | |
2173 | ||
e9ed9d22 | 2174 | error = xfs_lock_dir_and_entry(dp, ip); |
1da177e4 LT |
2175 | if (error) { |
2176 | REMOVE_DEBUG_TRACE(__LINE__); | |
2177 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2178 | goto std_return; |
2179 | } | |
2180 | ||
2181 | /* | |
2182 | * At this point, we've gotten both the directory and the entry | |
2183 | * inodes locked. | |
2184 | */ | |
5df78e73 | 2185 | IHOLD(ip); |
1da177e4 | 2186 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
82dab941 CH |
2187 | |
2188 | IHOLD(dp); | |
2189 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1da177e4 LT |
2190 | |
2191 | /* | |
2192 | * Entry must exist since we did a lookup in xfs_lock_dir_and_entry. | |
2193 | */ | |
2194 | XFS_BMAP_INIT(&free_list, &first_block); | |
556b8b16 | 2195 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
d4377d84 | 2196 | &first_block, &free_list, resblks); |
1da177e4 LT |
2197 | if (error) { |
2198 | ASSERT(error != ENOENT); | |
2199 | REMOVE_DEBUG_TRACE(__LINE__); | |
2200 | goto error1; | |
2201 | } | |
2202 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2203 | ||
2204 | dp->i_gen++; | |
2205 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2206 | ||
2207 | error = xfs_droplink(tp, ip); | |
2208 | if (error) { | |
2209 | REMOVE_DEBUG_TRACE(__LINE__); | |
2210 | goto error1; | |
2211 | } | |
2212 | ||
2213 | /* Determine if this is the last link while | |
2214 | * we are in the transaction. | |
2215 | */ | |
2216 | link_zero = (ip)->i_d.di_nlink==0; | |
2217 | ||
1da177e4 LT |
2218 | /* |
2219 | * If this is a synchronous mount, make sure that the | |
2220 | * remove transaction goes to disk before returning to | |
2221 | * the user. | |
2222 | */ | |
2223 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2224 | xfs_trans_set_sync(tp); | |
2225 | } | |
2226 | ||
f7c99b6f | 2227 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2228 | if (error) { |
2229 | REMOVE_DEBUG_TRACE(__LINE__); | |
2230 | goto error_rele; | |
2231 | } | |
2232 | ||
1c72bf90 | 2233 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
5df78e73 | 2234 | if (error) |
1da177e4 | 2235 | goto std_return; |
1da177e4 | 2236 | |
2a82b8be DC |
2237 | /* |
2238 | * If we are using filestreams, kill the stream association. | |
2239 | * If the file is still open it may get a new one but that | |
2240 | * will get killed on last close in xfs_close() so we don't | |
2241 | * have to worry about that. | |
2242 | */ | |
2243 | if (link_zero && xfs_inode_is_filestream(ip)) | |
2244 | xfs_filestream_deassociate(ip); | |
2245 | ||
cf441eeb | 2246 | xfs_itrace_exit(ip); |
1da177e4 LT |
2247 | |
2248 | /* Fall through to std_return with error = 0 */ | |
2249 | std_return: | |
eb9df39d | 2250 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 | 2251 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
bc4ac74a | 2252 | dp, DM_RIGHT_NULL, |
1da177e4 | 2253 | NULL, DM_RIGHT_NULL, |
556b8b16 | 2254 | name->name, NULL, ip->i_d.di_mode, error, 0); |
1da177e4 LT |
2255 | } |
2256 | return error; | |
2257 | ||
2258 | error1: | |
2259 | xfs_bmap_cancel(&free_list); | |
2260 | cancel_flags |= XFS_TRANS_ABORT; | |
2261 | xfs_trans_cancel(tp, cancel_flags); | |
2262 | goto std_return; | |
2263 | ||
2264 | error_rele: | |
2265 | /* | |
2266 | * In this case make sure to not release the inode until after | |
2267 | * the current transaction is aborted. Releasing it beforehand | |
2268 | * can cause us to go to xfs_inactive and start a recursive | |
2269 | * transaction which can easily deadlock with the current one. | |
2270 | */ | |
2271 | xfs_bmap_cancel(&free_list); | |
2272 | cancel_flags |= XFS_TRANS_ABORT; | |
2273 | xfs_trans_cancel(tp, cancel_flags); | |
2274 | ||
1da177e4 LT |
2275 | goto std_return; |
2276 | } | |
2277 | ||
993386c1 | 2278 | int |
1da177e4 | 2279 | xfs_link( |
993386c1 | 2280 | xfs_inode_t *tdp, |
a3da7896 | 2281 | xfs_inode_t *sip, |
556b8b16 | 2282 | struct xfs_name *target_name) |
1da177e4 | 2283 | { |
993386c1 | 2284 | xfs_mount_t *mp = tdp->i_mount; |
1da177e4 | 2285 | xfs_trans_t *tp; |
1da177e4 LT |
2286 | xfs_inode_t *ips[2]; |
2287 | int error; | |
2288 | xfs_bmap_free_t free_list; | |
2289 | xfs_fsblock_t first_block; | |
2290 | int cancel_flags; | |
2291 | int committed; | |
1da177e4 | 2292 | int resblks; |
1da177e4 | 2293 | |
cf441eeb | 2294 | xfs_itrace_entry(tdp); |
a3da7896 | 2295 | xfs_itrace_entry(sip); |
1da177e4 | 2296 | |
a3da7896 | 2297 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); |
1da177e4 | 2298 | |
1da177e4 LT |
2299 | if (XFS_FORCED_SHUTDOWN(mp)) |
2300 | return XFS_ERROR(EIO); | |
2301 | ||
eb9df39d | 2302 | if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) { |
1da177e4 | 2303 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, |
bc4ac74a CH |
2304 | tdp, DM_RIGHT_NULL, |
2305 | sip, DM_RIGHT_NULL, | |
556b8b16 | 2306 | target_name->name, NULL, 0, 0, 0); |
1da177e4 LT |
2307 | if (error) |
2308 | return error; | |
2309 | } | |
2310 | ||
2311 | /* Return through std_return after this point. */ | |
2312 | ||
2313 | error = XFS_QM_DQATTACH(mp, sip, 0); | |
2314 | if (!error && sip != tdp) | |
2315 | error = XFS_QM_DQATTACH(mp, tdp, 0); | |
2316 | if (error) | |
2317 | goto std_return; | |
2318 | ||
2319 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2320 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
556b8b16 | 2321 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
1da177e4 LT |
2322 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, |
2323 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2324 | if (error == ENOSPC) { | |
2325 | resblks = 0; | |
2326 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2327 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2328 | } | |
2329 | if (error) { | |
2330 | cancel_flags = 0; | |
2331 | goto error_return; | |
2332 | } | |
2333 | ||
2334 | if (sip->i_ino < tdp->i_ino) { | |
2335 | ips[0] = sip; | |
2336 | ips[1] = tdp; | |
2337 | } else { | |
2338 | ips[0] = tdp; | |
2339 | ips[1] = sip; | |
2340 | } | |
2341 | ||
cfa853e4 | 2342 | xfs_lock_inodes(ips, 2, XFS_ILOCK_EXCL); |
1da177e4 LT |
2343 | |
2344 | /* | |
2345 | * Increment vnode ref counts since xfs_trans_commit & | |
2346 | * xfs_trans_cancel will both unlock the inodes and | |
2347 | * decrement the associated ref counts. | |
2348 | */ | |
a3da7896 CH |
2349 | IHOLD(sip); |
2350 | IHOLD(tdp); | |
1da177e4 LT |
2351 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); |
2352 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2353 | ||
2354 | /* | |
2355 | * If the source has too many links, we can't make any more to it. | |
2356 | */ | |
2357 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2358 | error = XFS_ERROR(EMLINK); | |
2359 | goto error_return; | |
2360 | } | |
2361 | ||
365ca83d NS |
2362 | /* |
2363 | * If we are using project inheritance, we only allow hard link | |
2364 | * creation in our tree when the project IDs are the same; else | |
2365 | * the tree quota mechanism could be circumvented. | |
2366 | */ | |
2367 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2368 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 2369 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
2370 | goto error_return; |
2371 | } | |
2372 | ||
556b8b16 BN |
2373 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); |
2374 | if (error) | |
1da177e4 LT |
2375 | goto error_return; |
2376 | ||
2377 | XFS_BMAP_INIT(&free_list, &first_block); | |
2378 | ||
556b8b16 BN |
2379 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
2380 | &first_block, &free_list, resblks); | |
1da177e4 LT |
2381 | if (error) |
2382 | goto abort_return; | |
2383 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2384 | tdp->i_gen++; | |
2385 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
2386 | ||
2387 | error = xfs_bumplink(tp, sip); | |
b71d300c | 2388 | if (error) |
1da177e4 | 2389 | goto abort_return; |
1da177e4 LT |
2390 | |
2391 | /* | |
2392 | * If this is a synchronous mount, make sure that the | |
2393 | * link transaction goes to disk before returning to | |
2394 | * the user. | |
2395 | */ | |
2396 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2397 | xfs_trans_set_sync(tp); | |
2398 | } | |
2399 | ||
f7c99b6f | 2400 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2401 | if (error) { |
2402 | xfs_bmap_cancel(&free_list); | |
2403 | goto abort_return; | |
2404 | } | |
2405 | ||
1c72bf90 | 2406 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
b71d300c | 2407 | if (error) |
1da177e4 | 2408 | goto std_return; |
1da177e4 LT |
2409 | |
2410 | /* Fall through to std_return with error = 0. */ | |
2411 | std_return: | |
eb9df39d | 2412 | if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) { |
1da177e4 | 2413 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, |
bc4ac74a CH |
2414 | tdp, DM_RIGHT_NULL, |
2415 | sip, DM_RIGHT_NULL, | |
556b8b16 | 2416 | target_name->name, NULL, 0, error, 0); |
1da177e4 LT |
2417 | } |
2418 | return error; | |
2419 | ||
2420 | abort_return: | |
2421 | cancel_flags |= XFS_TRANS_ABORT; | |
2422 | /* FALLTHROUGH */ | |
014c2544 | 2423 | |
1da177e4 LT |
2424 | error_return: |
2425 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2426 | goto std_return; |
2427 | } | |
b71d300c AD |
2428 | |
2429 | ||
993386c1 | 2430 | int |
1da177e4 | 2431 | xfs_mkdir( |
993386c1 | 2432 | xfs_inode_t *dp, |
556b8b16 | 2433 | struct xfs_name *dir_name, |
3e5daf05 | 2434 | mode_t mode, |
979ebab1 | 2435 | xfs_inode_t **ipp, |
1da177e4 LT |
2436 | cred_t *credp) |
2437 | { | |
993386c1 | 2438 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 2439 | xfs_inode_t *cdp; /* inode of created dir */ |
1da177e4 | 2440 | xfs_trans_t *tp; |
1da177e4 LT |
2441 | int cancel_flags; |
2442 | int error; | |
2443 | int committed; | |
2444 | xfs_bmap_free_t free_list; | |
2445 | xfs_fsblock_t first_block; | |
993386c1 | 2446 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2447 | boolean_t created = B_FALSE; |
2448 | int dm_event_sent = 0; | |
2449 | xfs_prid_t prid; | |
2450 | struct xfs_dquot *udqp, *gdqp; | |
2451 | uint resblks; | |
1da177e4 LT |
2452 | |
2453 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2454 | return XFS_ERROR(EIO); | |
2455 | ||
1da177e4 | 2456 | tp = NULL; |
1da177e4 | 2457 | |
eb9df39d | 2458 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 | 2459 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
bc4ac74a | 2460 | dp, DM_RIGHT_NULL, NULL, |
556b8b16 | 2461 | DM_RIGHT_NULL, dir_name->name, NULL, |
3e5daf05 | 2462 | mode, 0, 0); |
1da177e4 LT |
2463 | if (error) |
2464 | return error; | |
2465 | dm_event_sent = 1; | |
2466 | } | |
2467 | ||
2468 | /* Return through std_return after this point. */ | |
2469 | ||
cf441eeb | 2470 | xfs_itrace_entry(dp); |
1da177e4 LT |
2471 | |
2472 | mp = dp->i_mount; | |
2473 | udqp = gdqp = NULL; | |
365ca83d NS |
2474 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2475 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
2476 | else |
2477 | prid = (xfs_prid_t)dfltprid; | |
2478 | ||
2479 | /* | |
2480 | * Make sure that we have allocated dquot(s) on disk. | |
2481 | */ | |
2482 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 2483 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
2484 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2485 | if (error) | |
2486 | goto std_return; | |
2487 | ||
2488 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
2489 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
556b8b16 | 2490 | resblks = XFS_MKDIR_SPACE_RES(mp, dir_name->len); |
1da177e4 LT |
2491 | error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0, |
2492 | XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT); | |
2493 | if (error == ENOSPC) { | |
2494 | resblks = 0; | |
2495 | error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0, | |
2496 | XFS_TRANS_PERM_LOG_RES, | |
2497 | XFS_MKDIR_LOG_COUNT); | |
2498 | } | |
2499 | if (error) { | |
2500 | cancel_flags = 0; | |
1da177e4 LT |
2501 | goto error_return; |
2502 | } | |
2503 | ||
f7c66ce3 | 2504 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2505 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2506 | |
2507 | /* | |
2508 | * Check for directory link count overflow. | |
2509 | */ | |
2510 | if (dp->i_d.di_nlink >= XFS_MAXLINK) { | |
2511 | error = XFS_ERROR(EMLINK); | |
2512 | goto error_return; | |
2513 | } | |
2514 | ||
2515 | /* | |
2516 | * Reserve disk quota and the inode. | |
2517 | */ | |
2518 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
2519 | if (error) | |
2520 | goto error_return; | |
2521 | ||
556b8b16 BN |
2522 | error = xfs_dir_canenter(tp, dp, dir_name, resblks); |
2523 | if (error) | |
1da177e4 LT |
2524 | goto error_return; |
2525 | /* | |
2526 | * create the directory inode. | |
2527 | */ | |
3e5daf05 | 2528 | error = xfs_dir_ialloc(&tp, dp, mode, 2, |
1da177e4 LT |
2529 | 0, credp, prid, resblks > 0, |
2530 | &cdp, NULL); | |
2531 | if (error) { | |
2532 | if (error == ENOSPC) | |
2533 | goto error_return; | |
2534 | goto abort_return; | |
2535 | } | |
cf441eeb | 2536 | xfs_itrace_ref(cdp); |
1da177e4 LT |
2537 | |
2538 | /* | |
2539 | * Now we add the directory inode to the transaction. | |
2540 | * We waited until now since xfs_dir_ialloc might start | |
2541 | * a new transaction. Had we joined the transaction | |
993386c1 CH |
2542 | * earlier, the locks might have gotten released. An error |
2543 | * from here on will result in the transaction cancel | |
2544 | * unlocking dp so don't do it explicitly in the error path. | |
1da177e4 | 2545 | */ |
979ebab1 | 2546 | IHOLD(dp); |
1da177e4 | 2547 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 2548 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2549 | |
2550 | XFS_BMAP_INIT(&free_list, &first_block); | |
2551 | ||
556b8b16 BN |
2552 | error = xfs_dir_createname(tp, dp, dir_name, cdp->i_ino, |
2553 | &first_block, &free_list, resblks ? | |
2554 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
2555 | if (error) { |
2556 | ASSERT(error != ENOSPC); | |
2557 | goto error1; | |
2558 | } | |
2559 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2560 | ||
2561 | /* | |
2562 | * Bump the in memory version number of the parent directory | |
2563 | * so that other processes accessing it will recognize that | |
2564 | * the directory has changed. | |
2565 | */ | |
2566 | dp->i_gen++; | |
2567 | ||
f6c2d1fa NS |
2568 | error = xfs_dir_init(tp, cdp, dp); |
2569 | if (error) | |
1da177e4 | 2570 | goto error2; |
1da177e4 LT |
2571 | |
2572 | cdp->i_gen = 1; | |
2573 | error = xfs_bumplink(tp, dp); | |
f6c2d1fa | 2574 | if (error) |
1da177e4 | 2575 | goto error2; |
1da177e4 | 2576 | |
1da177e4 LT |
2577 | created = B_TRUE; |
2578 | ||
979ebab1 | 2579 | *ipp = cdp; |
1da177e4 LT |
2580 | IHOLD(cdp); |
2581 | ||
2582 | /* | |
2583 | * Attach the dquots to the new inode and modify the icount incore. | |
2584 | */ | |
2585 | XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp); | |
2586 | ||
2587 | /* | |
2588 | * If this is a synchronous mount, make sure that the | |
2589 | * mkdir transaction goes to disk before returning to | |
2590 | * the user. | |
2591 | */ | |
2592 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2593 | xfs_trans_set_sync(tp); | |
2594 | } | |
2595 | ||
f7c99b6f | 2596 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2597 | if (error) { |
2598 | IRELE(cdp); | |
2599 | goto error2; | |
2600 | } | |
2601 | ||
1c72bf90 | 2602 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2603 | XFS_QM_DQRELE(mp, udqp); |
2604 | XFS_QM_DQRELE(mp, gdqp); | |
2605 | if (error) { | |
2606 | IRELE(cdp); | |
2607 | } | |
2608 | ||
2609 | /* Fall through to std_return with error = 0 or errno from | |
2610 | * xfs_trans_commit. */ | |
2611 | ||
2612 | std_return: | |
eb9df39d | 2613 | if ((created || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 2614 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 | 2615 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
bc4ac74a CH |
2616 | dp, DM_RIGHT_NULL, |
2617 | created ? cdp : NULL, | |
1da177e4 | 2618 | DM_RIGHT_NULL, |
556b8b16 | 2619 | dir_name->name, NULL, |
3e5daf05 | 2620 | mode, error, 0); |
1da177e4 LT |
2621 | } |
2622 | return error; | |
2623 | ||
2624 | error2: | |
2625 | error1: | |
2626 | xfs_bmap_cancel(&free_list); | |
2627 | abort_return: | |
2628 | cancel_flags |= XFS_TRANS_ABORT; | |
2629 | error_return: | |
2630 | xfs_trans_cancel(tp, cancel_flags); | |
2631 | XFS_QM_DQRELE(mp, udqp); | |
2632 | XFS_QM_DQRELE(mp, gdqp); | |
2633 | ||
993386c1 | 2634 | if (unlock_dp_on_error) |
1da177e4 | 2635 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2636 | |
2637 | goto std_return; | |
2638 | } | |
2639 | ||
993386c1 | 2640 | int |
1da177e4 | 2641 | xfs_rmdir( |
993386c1 | 2642 | xfs_inode_t *dp, |
556b8b16 BN |
2643 | struct xfs_name *name, |
2644 | xfs_inode_t *cdp) | |
1da177e4 | 2645 | { |
993386c1 | 2646 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 2647 | xfs_trans_t *tp; |
1da177e4 LT |
2648 | int error; |
2649 | xfs_bmap_free_t free_list; | |
2650 | xfs_fsblock_t first_block; | |
2651 | int cancel_flags; | |
2652 | int committed; | |
1da177e4 | 2653 | int last_cdp_link; |
1da177e4 LT |
2654 | uint resblks; |
2655 | ||
cf441eeb | 2656 | xfs_itrace_entry(dp); |
1da177e4 | 2657 | |
993386c1 | 2658 | if (XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 | 2659 | return XFS_ERROR(EIO); |
1da177e4 | 2660 | |
eb9df39d | 2661 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
1da177e4 | 2662 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, |
bc4ac74a | 2663 | dp, DM_RIGHT_NULL, |
556b8b16 BN |
2664 | NULL, DM_RIGHT_NULL, name->name, |
2665 | NULL, cdp->i_d.di_mode, 0, 0); | |
1da177e4 LT |
2666 | if (error) |
2667 | return XFS_ERROR(error); | |
2668 | } | |
2669 | ||
1da177e4 LT |
2670 | /* |
2671 | * Get the dquots for the inodes. | |
2672 | */ | |
2673 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
82dab941 | 2674 | if (!error) |
1da177e4 LT |
2675 | error = XFS_QM_DQATTACH(mp, cdp, 0); |
2676 | if (error) { | |
1da177e4 LT |
2677 | REMOVE_DEBUG_TRACE(__LINE__); |
2678 | goto std_return; | |
2679 | } | |
2680 | ||
2681 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
2682 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2683 | /* | |
2684 | * We try to get the real space reservation first, | |
2685 | * allowing for directory btree deletion(s) implying | |
2686 | * possible bmap insert(s). If we can't get the space | |
2687 | * reservation then we use 0 instead, and avoid the bmap | |
2688 | * btree insert(s) in the directory code by, if the bmap | |
2689 | * insert tries to happen, instead trimming the LAST | |
2690 | * block from the directory. | |
2691 | */ | |
2692 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2693 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2694 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
2695 | if (error == ENOSPC) { | |
2696 | resblks = 0; | |
2697 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2698 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
2699 | } | |
2700 | if (error) { | |
2701 | ASSERT(error != ENOSPC); | |
2702 | cancel_flags = 0; | |
1da177e4 LT |
2703 | goto error_return; |
2704 | } | |
2705 | XFS_BMAP_INIT(&free_list, &first_block); | |
2706 | ||
2707 | /* | |
2708 | * Now lock the child directory inode and the parent directory | |
2709 | * inode in the proper order. This will take care of validating | |
2710 | * that the directory entry for the child directory inode has | |
2711 | * not changed while we were obtaining a log reservation. | |
2712 | */ | |
e9ed9d22 | 2713 | error = xfs_lock_dir_and_entry(dp, cdp); |
1da177e4 LT |
2714 | if (error) { |
2715 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2716 | goto std_return; |
2717 | } | |
2718 | ||
5df78e73 | 2719 | IHOLD(dp); |
1da177e4 | 2720 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
1da177e4 | 2721 | |
5df78e73 | 2722 | IHOLD(cdp); |
1da177e4 LT |
2723 | xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL); |
2724 | ||
2725 | ASSERT(cdp->i_d.di_nlink >= 2); | |
2726 | if (cdp->i_d.di_nlink != 2) { | |
2727 | error = XFS_ERROR(ENOTEMPTY); | |
2728 | goto error_return; | |
2729 | } | |
f6c2d1fa | 2730 | if (!xfs_dir_isempty(cdp)) { |
1da177e4 LT |
2731 | error = XFS_ERROR(ENOTEMPTY); |
2732 | goto error_return; | |
2733 | } | |
2734 | ||
556b8b16 | 2735 | error = xfs_dir_removename(tp, dp, name, cdp->i_ino, |
f6c2d1fa NS |
2736 | &first_block, &free_list, resblks); |
2737 | if (error) | |
1da177e4 | 2738 | goto error1; |
1da177e4 LT |
2739 | |
2740 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2741 | ||
2742 | /* | |
2743 | * Bump the in memory generation count on the parent | |
2744 | * directory so that other can know that it has changed. | |
2745 | */ | |
2746 | dp->i_gen++; | |
2747 | ||
2748 | /* | |
2749 | * Drop the link from cdp's "..". | |
2750 | */ | |
2751 | error = xfs_droplink(tp, dp); | |
2752 | if (error) { | |
2753 | goto error1; | |
2754 | } | |
2755 | ||
2756 | /* | |
2757 | * Drop the link from dp to cdp. | |
2758 | */ | |
2759 | error = xfs_droplink(tp, cdp); | |
2760 | if (error) { | |
2761 | goto error1; | |
2762 | } | |
2763 | ||
2764 | /* | |
2765 | * Drop the "." link from cdp to self. | |
2766 | */ | |
2767 | error = xfs_droplink(tp, cdp); | |
2768 | if (error) { | |
2769 | goto error1; | |
2770 | } | |
2771 | ||
2772 | /* Determine these before committing transaction */ | |
2773 | last_cdp_link = (cdp)->i_d.di_nlink==0; | |
2774 | ||
1da177e4 LT |
2775 | /* |
2776 | * If this is a synchronous mount, make sure that the | |
2777 | * rmdir transaction goes to disk before returning to | |
2778 | * the user. | |
2779 | */ | |
2780 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2781 | xfs_trans_set_sync(tp); | |
2782 | } | |
2783 | ||
f7c99b6f | 2784 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2785 | if (error) { |
2786 | xfs_bmap_cancel(&free_list); | |
2787 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | | |
2788 | XFS_TRANS_ABORT)); | |
1da177e4 LT |
2789 | goto std_return; |
2790 | } | |
2791 | ||
1c72bf90 | 2792 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 2793 | if (error) { |
1da177e4 LT |
2794 | goto std_return; |
2795 | } | |
2796 | ||
2797 | ||
1da177e4 LT |
2798 | /* Fall through to std_return with error = 0 or the errno |
2799 | * from xfs_trans_commit. */ | |
bb19fba1 | 2800 | std_return: |
eb9df39d | 2801 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 | 2802 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
bc4ac74a | 2803 | dp, DM_RIGHT_NULL, |
1da177e4 | 2804 | NULL, DM_RIGHT_NULL, |
556b8b16 | 2805 | name->name, NULL, cdp->i_d.di_mode, |
1da177e4 LT |
2806 | error, 0); |
2807 | } | |
2808 | return error; | |
2809 | ||
bb19fba1 | 2810 | error1: |
1da177e4 LT |
2811 | xfs_bmap_cancel(&free_list); |
2812 | cancel_flags |= XFS_TRANS_ABORT; | |
014c2544 JJ |
2813 | /* FALLTHROUGH */ |
2814 | ||
bb19fba1 | 2815 | error_return: |
1da177e4 LT |
2816 | xfs_trans_cancel(tp, cancel_flags); |
2817 | goto std_return; | |
2818 | } | |
2819 | ||
993386c1 | 2820 | int |
1da177e4 | 2821 | xfs_symlink( |
993386c1 | 2822 | xfs_inode_t *dp, |
556b8b16 BN |
2823 | struct xfs_name *link_name, |
2824 | const char *target_path, | |
3e5daf05 | 2825 | mode_t mode, |
3937be5b | 2826 | xfs_inode_t **ipp, |
1da177e4 LT |
2827 | cred_t *credp) |
2828 | { | |
993386c1 | 2829 | xfs_mount_t *mp = dp->i_mount; |
1da177e4 | 2830 | xfs_trans_t *tp; |
1da177e4 LT |
2831 | xfs_inode_t *ip; |
2832 | int error; | |
2833 | int pathlen; | |
2834 | xfs_bmap_free_t free_list; | |
2835 | xfs_fsblock_t first_block; | |
993386c1 | 2836 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2837 | uint cancel_flags; |
2838 | int committed; | |
2839 | xfs_fileoff_t first_fsb; | |
2840 | xfs_filblks_t fs_blocks; | |
2841 | int nmaps; | |
2842 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
2843 | xfs_daddr_t d; | |
556b8b16 | 2844 | const char *cur_chunk; |
1da177e4 LT |
2845 | int byte_cnt; |
2846 | int n; | |
2847 | xfs_buf_t *bp; | |
2848 | xfs_prid_t prid; | |
2849 | struct xfs_dquot *udqp, *gdqp; | |
2850 | uint resblks; | |
1da177e4 | 2851 | |
3937be5b | 2852 | *ipp = NULL; |
1da177e4 LT |
2853 | error = 0; |
2854 | ip = NULL; | |
2855 | tp = NULL; | |
2856 | ||
cf441eeb | 2857 | xfs_itrace_entry(dp); |
1da177e4 LT |
2858 | |
2859 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2860 | return XFS_ERROR(EIO); | |
2861 | ||
1da177e4 LT |
2862 | /* |
2863 | * Check component lengths of the target path name. | |
2864 | */ | |
2865 | pathlen = strlen(target_path); | |
2866 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
2867 | return XFS_ERROR(ENAMETOOLONG); | |
1da177e4 | 2868 | |
eb9df39d | 2869 | if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) { |
bc4ac74a | 2870 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp, |
1da177e4 | 2871 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, |
556b8b16 | 2872 | link_name->name, target_path, 0, 0, 0); |
1da177e4 LT |
2873 | if (error) |
2874 | return error; | |
2875 | } | |
2876 | ||
2877 | /* Return through std_return after this point. */ | |
2878 | ||
2879 | udqp = gdqp = NULL; | |
365ca83d NS |
2880 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2881 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
2882 | else |
2883 | prid = (xfs_prid_t)dfltprid; | |
2884 | ||
2885 | /* | |
2886 | * Make sure that we have allocated dquot(s) on disk. | |
2887 | */ | |
2888 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 2889 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
2890 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2891 | if (error) | |
2892 | goto std_return; | |
2893 | ||
2894 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
2895 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2896 | /* | |
2897 | * The symlink will fit into the inode data fork? | |
2898 | * There can't be any attributes so we get the whole variable part. | |
2899 | */ | |
2900 | if (pathlen <= XFS_LITINO(mp)) | |
2901 | fs_blocks = 0; | |
2902 | else | |
2903 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
556b8b16 | 2904 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks); |
1da177e4 LT |
2905 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, |
2906 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
2907 | if (error == ENOSPC && fs_blocks == 0) { | |
2908 | resblks = 0; | |
2909 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
2910 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
2911 | } | |
2912 | if (error) { | |
2913 | cancel_flags = 0; | |
1da177e4 LT |
2914 | goto error_return; |
2915 | } | |
2916 | ||
f7c66ce3 | 2917 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2918 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2919 | |
2920 | /* | |
2921 | * Check whether the directory allows new symlinks or not. | |
2922 | */ | |
2923 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
2924 | error = XFS_ERROR(EPERM); | |
2925 | goto error_return; | |
2926 | } | |
2927 | ||
2928 | /* | |
2929 | * Reserve disk quota : blocks and inode. | |
2930 | */ | |
2931 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
2932 | if (error) | |
2933 | goto error_return; | |
2934 | ||
2935 | /* | |
2936 | * Check for ability to enter directory entry, if no space reserved. | |
2937 | */ | |
556b8b16 BN |
2938 | error = xfs_dir_canenter(tp, dp, link_name, resblks); |
2939 | if (error) | |
1da177e4 LT |
2940 | goto error_return; |
2941 | /* | |
2942 | * Initialize the bmap freelist prior to calling either | |
2943 | * bmapi or the directory create code. | |
2944 | */ | |
2945 | XFS_BMAP_INIT(&free_list, &first_block); | |
2946 | ||
2947 | /* | |
2948 | * Allocate an inode for the symlink. | |
2949 | */ | |
3e5daf05 | 2950 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), |
1da177e4 LT |
2951 | 1, 0, credp, prid, resblks > 0, &ip, NULL); |
2952 | if (error) { | |
2953 | if (error == ENOSPC) | |
2954 | goto error_return; | |
2955 | goto error1; | |
2956 | } | |
cf441eeb | 2957 | xfs_itrace_ref(ip); |
1da177e4 | 2958 | |
993386c1 CH |
2959 | /* |
2960 | * An error after we've joined dp to the transaction will result in the | |
2961 | * transaction cancel unlocking dp so don't do it explicitly in the | |
2962 | * error path. | |
2963 | */ | |
3937be5b | 2964 | IHOLD(dp); |
1da177e4 | 2965 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 2966 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2967 | |
2968 | /* | |
2969 | * Also attach the dquot(s) to it, if applicable. | |
2970 | */ | |
2971 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
2972 | ||
2973 | if (resblks) | |
2974 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
2975 | /* | |
2976 | * If the symlink will fit into the inode, write it inline. | |
2977 | */ | |
2978 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
2979 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
2980 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
2981 | ip->i_d.di_size = pathlen; | |
2982 | ||
2983 | /* | |
2984 | * The inode was initially created in extent format. | |
2985 | */ | |
2986 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
2987 | ip->i_df.if_flags |= XFS_IFINLINE; | |
2988 | ||
2989 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
2990 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
2991 | ||
2992 | } else { | |
2993 | first_fsb = 0; | |
2994 | nmaps = SYMLINK_MAPS; | |
2995 | ||
2996 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
2997 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
2998 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 2999 | &free_list, NULL); |
1da177e4 LT |
3000 | if (error) { |
3001 | goto error1; | |
3002 | } | |
3003 | ||
3004 | if (resblks) | |
3005 | resblks -= fs_blocks; | |
3006 | ip->i_d.di_size = pathlen; | |
3007 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3008 | ||
3009 | cur_chunk = target_path; | |
3010 | for (n = 0; n < nmaps; n++) { | |
3011 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
3012 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
3013 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
3014 | BTOBB(byte_cnt), 0); | |
3015 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
3016 | if (pathlen < byte_cnt) { | |
3017 | byte_cnt = pathlen; | |
3018 | } | |
3019 | pathlen -= byte_cnt; | |
3020 | ||
3021 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
3022 | cur_chunk += byte_cnt; | |
3023 | ||
3024 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
3025 | } | |
3026 | } | |
3027 | ||
3028 | /* | |
3029 | * Create the directory entry for the symlink. | |
3030 | */ | |
556b8b16 BN |
3031 | error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, |
3032 | &first_block, &free_list, resblks); | |
f6c2d1fa | 3033 | if (error) |
1da177e4 | 3034 | goto error1; |
1da177e4 LT |
3035 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3036 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
3037 | ||
3038 | /* | |
3039 | * Bump the in memory version number of the parent directory | |
3040 | * so that other processes accessing it will recognize that | |
3041 | * the directory has changed. | |
3042 | */ | |
3043 | dp->i_gen++; | |
3044 | ||
3045 | /* | |
3046 | * If this is a synchronous mount, make sure that the | |
3047 | * symlink transaction goes to disk before returning to | |
3048 | * the user. | |
3049 | */ | |
3050 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3051 | xfs_trans_set_sync(tp); | |
3052 | } | |
3053 | ||
3054 | /* | |
3055 | * xfs_trans_commit normally decrements the vnode ref count | |
3056 | * when it unlocks the inode. Since we want to return the | |
3057 | * vnode to the caller, we bump the vnode ref count now. | |
3058 | */ | |
3059 | IHOLD(ip); | |
3060 | ||
f7c99b6f | 3061 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3062 | if (error) { |
3063 | goto error2; | |
3064 | } | |
1c72bf90 | 3065 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3066 | XFS_QM_DQRELE(mp, udqp); |
3067 | XFS_QM_DQRELE(mp, gdqp); | |
3068 | ||
3069 | /* Fall through to std_return with error = 0 or errno from | |
3070 | * xfs_trans_commit */ | |
3071 | std_return: | |
993386c1 | 3072 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) { |
1da177e4 | 3073 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, |
bc4ac74a CH |
3074 | dp, DM_RIGHT_NULL, |
3075 | error ? NULL : ip, | |
556b8b16 BN |
3076 | DM_RIGHT_NULL, link_name->name, |
3077 | target_path, 0, error, 0); | |
1da177e4 LT |
3078 | } |
3079 | ||
3937be5b CH |
3080 | if (!error) |
3081 | *ipp = ip; | |
1da177e4 LT |
3082 | return error; |
3083 | ||
3084 | error2: | |
3085 | IRELE(ip); | |
3086 | error1: | |
3087 | xfs_bmap_cancel(&free_list); | |
3088 | cancel_flags |= XFS_TRANS_ABORT; | |
3089 | error_return: | |
3090 | xfs_trans_cancel(tp, cancel_flags); | |
3091 | XFS_QM_DQRELE(mp, udqp); | |
3092 | XFS_QM_DQRELE(mp, gdqp); | |
3093 | ||
993386c1 | 3094 | if (unlock_dp_on_error) |
1da177e4 | 3095 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
3096 | |
3097 | goto std_return; | |
3098 | } | |
3099 | ||
993386c1 | 3100 | int |
1da177e4 | 3101 | xfs_inode_flush( |
993386c1 | 3102 | xfs_inode_t *ip, |
1da177e4 LT |
3103 | int flags) |
3104 | { | |
993386c1 | 3105 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
3106 | int error = 0; |
3107 | ||
1da177e4 LT |
3108 | if (XFS_FORCED_SHUTDOWN(mp)) |
3109 | return XFS_ERROR(EIO); | |
3110 | ||
3111 | /* | |
3112 | * Bypass inodes which have already been cleaned by | |
3113 | * the inode flush clustering code inside xfs_iflush | |
3114 | */ | |
33540408 | 3115 | if (xfs_inode_clean(ip)) |
1da177e4 LT |
3116 | return 0; |
3117 | ||
1da177e4 LT |
3118 | /* |
3119 | * We make this non-blocking if the inode is contended, | |
3120 | * return EAGAIN to indicate to the caller that they | |
3121 | * did not succeed. This prevents the flush path from | |
3122 | * blocking on inodes inside another operation right | |
3123 | * now, they get caught later by xfs_sync. | |
3124 | */ | |
a3f74ffb DC |
3125 | if (flags & FLUSH_SYNC) { |
3126 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3127 | xfs_iflock(ip); | |
3128 | } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
3129 | if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) { | |
3130 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1da177e4 LT |
3131 | return EAGAIN; |
3132 | } | |
a3f74ffb DC |
3133 | } else { |
3134 | return EAGAIN; | |
1da177e4 LT |
3135 | } |
3136 | ||
a3f74ffb DC |
3137 | error = xfs_iflush(ip, (flags & FLUSH_SYNC) ? XFS_IFLUSH_SYNC |
3138 | : XFS_IFLUSH_ASYNC_NOBLOCK); | |
3139 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3140 | ||
1da177e4 LT |
3141 | return error; |
3142 | } | |
3143 | ||
993386c1 | 3144 | |
1da177e4 | 3145 | int |
993386c1 CH |
3146 | xfs_set_dmattrs( |
3147 | xfs_inode_t *ip, | |
1da177e4 | 3148 | u_int evmask, |
993386c1 | 3149 | u_int16_t state) |
1da177e4 | 3150 | { |
993386c1 | 3151 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 3152 | xfs_trans_t *tp; |
1da177e4 LT |
3153 | int error; |
3154 | ||
3155 | if (!capable(CAP_SYS_ADMIN)) | |
3156 | return XFS_ERROR(EPERM); | |
3157 | ||
1da177e4 LT |
3158 | if (XFS_FORCED_SHUTDOWN(mp)) |
3159 | return XFS_ERROR(EIO); | |
3160 | ||
3161 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
3162 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
3163 | if (error) { | |
3164 | xfs_trans_cancel(tp, 0); | |
3165 | return error; | |
3166 | } | |
3167 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3168 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3169 | ||
613d7043 CH |
3170 | ip->i_d.di_dmevmask = evmask; |
3171 | ip->i_d.di_dmstate = state; | |
1da177e4 LT |
3172 | |
3173 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3174 | IHOLD(ip); | |
1c72bf90 | 3175 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
3176 | |
3177 | return error; | |
3178 | } | |
3179 | ||
993386c1 | 3180 | int |
1da177e4 | 3181 | xfs_reclaim( |
993386c1 | 3182 | xfs_inode_t *ip) |
1da177e4 | 3183 | { |
993386c1 | 3184 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1da177e4 | 3185 | |
cf441eeb | 3186 | xfs_itrace_entry(ip); |
1da177e4 LT |
3187 | |
3188 | ASSERT(!VN_MAPPED(vp)); | |
3189 | ||
3190 | /* bad inode, get out here ASAP */ | |
3191 | if (VN_BAD(vp)) { | |
3192 | xfs_ireclaim(ip); | |
3193 | return 0; | |
3194 | } | |
3195 | ||
b677c210 | 3196 | vn_iowait(ip); |
1da177e4 | 3197 | |
51c91ed5 | 3198 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); |
1da177e4 | 3199 | |
42fe2b1f CH |
3200 | /* |
3201 | * Make sure the atime in the XFS inode is correct before freeing the | |
3202 | * Linux inode. | |
3203 | */ | |
3204 | xfs_synchronize_atime(ip); | |
1da177e4 | 3205 | |
4c60658e DC |
3206 | /* |
3207 | * If we have nothing to flush with this inode then complete the | |
3208 | * teardown now, otherwise break the link between the xfs inode and the | |
3209 | * linux inode and clean up the xfs inode later. This avoids flushing | |
3210 | * the inode to disk during the delete operation itself. | |
3211 | * | |
3212 | * When breaking the link, we need to set the XFS_IRECLAIMABLE flag | |
3213 | * first to ensure that xfs_iunpin() will never see an xfs inode | |
3214 | * that has a linux inode being reclaimed. Synchronisation is provided | |
3215 | * by the i_flags_lock. | |
1da177e4 LT |
3216 | */ |
3217 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
3218 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3219 | xfs_iflock(ip); | |
3220 | return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); | |
3221 | } else { | |
3222 | xfs_mount_t *mp = ip->i_mount; | |
3223 | ||
4c60658e | 3224 | /* Protect sync and unpin from us */ |
1da177e4 | 3225 | XFS_MOUNT_ILOCK(mp); |
4c60658e DC |
3226 | spin_lock(&ip->i_flags_lock); |
3227 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
739bfb2a CH |
3228 | vn_to_inode(vp)->i_private = NULL; |
3229 | ip->i_vnode = NULL; | |
4c60658e | 3230 | spin_unlock(&ip->i_flags_lock); |
1da177e4 | 3231 | list_add_tail(&ip->i_reclaim, &mp->m_del_inodes); |
1da177e4 LT |
3232 | XFS_MOUNT_IUNLOCK(mp); |
3233 | } | |
3234 | return 0; | |
3235 | } | |
3236 | ||
3237 | int | |
3238 | xfs_finish_reclaim( | |
3239 | xfs_inode_t *ip, | |
3240 | int locked, | |
3241 | int sync_mode) | |
3242 | { | |
da353b0d | 3243 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); |
67fcaa73 | 3244 | bhv_vnode_t *vp = XFS_ITOV_NULL(ip); |
1da177e4 LT |
3245 | int error; |
3246 | ||
3247 | if (vp && VN_BAD(vp)) | |
3248 | goto reclaim; | |
3249 | ||
3250 | /* The hash lock here protects a thread in xfs_iget_core from | |
3251 | * racing with us on linking the inode back with a vnode. | |
3252 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
3253 | * us. | |
3254 | */ | |
da353b0d | 3255 | write_lock(&pag->pag_ici_lock); |
f273ab84 | 3256 | spin_lock(&ip->i_flags_lock); |
7a18c386 DC |
3257 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || |
3258 | (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) { | |
f273ab84 | 3259 | spin_unlock(&ip->i_flags_lock); |
da353b0d | 3260 | write_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
3261 | if (locked) { |
3262 | xfs_ifunlock(ip); | |
3263 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3264 | } | |
014c2544 | 3265 | return 1; |
1da177e4 | 3266 | } |
7a18c386 | 3267 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
f273ab84 | 3268 | spin_unlock(&ip->i_flags_lock); |
da353b0d DC |
3269 | write_unlock(&pag->pag_ici_lock); |
3270 | xfs_put_perag(ip->i_mount, pag); | |
1da177e4 LT |
3271 | |
3272 | /* | |
3273 | * If the inode is still dirty, then flush it out. If the inode | |
3274 | * is not in the AIL, then it will be OK to flush it delwri as | |
3275 | * long as xfs_iflush() does not keep any references to the inode. | |
3276 | * We leave that decision up to xfs_iflush() since it has the | |
3277 | * knowledge of whether it's OK to simply do a delwri flush of | |
3278 | * the inode or whether we need to wait until the inode is | |
3279 | * pulled from the AIL. | |
3280 | * We get the flush lock regardless, though, just to make sure | |
3281 | * we don't free it while it is being flushed. | |
3282 | */ | |
461aa8a2 LM |
3283 | if (!locked) { |
3284 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3285 | xfs_iflock(ip); | |
3286 | } | |
1da177e4 | 3287 | |
461aa8a2 | 3288 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1da177e4 LT |
3289 | if (ip->i_update_core || |
3290 | ((ip->i_itemp != NULL) && | |
3291 | (ip->i_itemp->ili_format.ilf_fields != 0))) { | |
3292 | error = xfs_iflush(ip, sync_mode); | |
3293 | /* | |
3294 | * If we hit an error, typically because of filesystem | |
3295 | * shutdown, we don't need to let vn_reclaim to know | |
3296 | * because we're gonna reclaim the inode anyway. | |
3297 | */ | |
3298 | if (error) { | |
3299 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3300 | goto reclaim; | |
3301 | } | |
3302 | xfs_iflock(ip); /* synchronize with xfs_iflush_done */ | |
3303 | } | |
3304 | ||
3305 | ASSERT(ip->i_update_core == 0); | |
3306 | ASSERT(ip->i_itemp == NULL || | |
3307 | ip->i_itemp->ili_format.ilf_fields == 0); | |
1da177e4 LT |
3308 | } |
3309 | ||
461aa8a2 LM |
3310 | xfs_ifunlock(ip); |
3311 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3312 | ||
1da177e4 LT |
3313 | reclaim: |
3314 | xfs_ireclaim(ip); | |
3315 | return 0; | |
3316 | } | |
3317 | ||
3318 | int | |
3319 | xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock) | |
3320 | { | |
3321 | int purged; | |
3322 | xfs_inode_t *ip, *n; | |
3323 | int done = 0; | |
3324 | ||
3325 | while (!done) { | |
3326 | purged = 0; | |
3327 | XFS_MOUNT_ILOCK(mp); | |
3328 | list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) { | |
3329 | if (noblock) { | |
3330 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) | |
3331 | continue; | |
3332 | if (xfs_ipincount(ip) || | |
3333 | !xfs_iflock_nowait(ip)) { | |
3334 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3335 | continue; | |
3336 | } | |
3337 | } | |
3338 | XFS_MOUNT_IUNLOCK(mp); | |
6b2cf618 FB |
3339 | if (xfs_finish_reclaim(ip, noblock, |
3340 | XFS_IFLUSH_DELWRI_ELSE_ASYNC)) | |
3341 | delay(1); | |
1da177e4 LT |
3342 | purged = 1; |
3343 | break; | |
3344 | } | |
3345 | ||
3346 | done = !purged; | |
3347 | } | |
3348 | ||
3349 | XFS_MOUNT_IUNLOCK(mp); | |
3350 | return 0; | |
3351 | } | |
3352 | ||
3353 | /* | |
3354 | * xfs_alloc_file_space() | |
3355 | * This routine allocates disk space for the given file. | |
3356 | * | |
3357 | * If alloc_type == 0, this request is for an ALLOCSP type | |
3358 | * request which will change the file size. In this case, no | |
3359 | * DMAPI event will be generated by the call. A TRUNCATE event | |
3360 | * will be generated later by xfs_setattr. | |
3361 | * | |
3362 | * If alloc_type != 0, this request is for a RESVSP type | |
3363 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
3364 | * lower block boundary byte address is less than the file's | |
3365 | * length. | |
3366 | * | |
3367 | * RETURNS: | |
3368 | * 0 on success | |
3369 | * errno on error | |
3370 | * | |
3371 | */ | |
ba0f32d4 | 3372 | STATIC int |
1da177e4 LT |
3373 | xfs_alloc_file_space( |
3374 | xfs_inode_t *ip, | |
3375 | xfs_off_t offset, | |
3376 | xfs_off_t len, | |
3377 | int alloc_type, | |
3378 | int attr_flags) | |
3379 | { | |
dd9f438e NS |
3380 | xfs_mount_t *mp = ip->i_mount; |
3381 | xfs_off_t count; | |
1da177e4 LT |
3382 | xfs_filblks_t allocated_fsb; |
3383 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
3384 | xfs_extlen_t extsz, temp; |
3385 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 3386 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
3387 | int nimaps; |
3388 | int bmapi_flag; | |
3389 | int quota_flag; | |
1da177e4 | 3390 | int rt; |
1da177e4 | 3391 | xfs_trans_t *tp; |
dd9f438e NS |
3392 | xfs_bmbt_irec_t imaps[1], *imapp; |
3393 | xfs_bmap_free_t free_list; | |
3394 | uint qblocks, resblks, resrtextents; | |
3395 | int committed; | |
3396 | int error; | |
1da177e4 | 3397 | |
cf441eeb | 3398 | xfs_itrace_entry(ip); |
1da177e4 LT |
3399 | |
3400 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3401 | return XFS_ERROR(EIO); | |
3402 | ||
1da177e4 LT |
3403 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
3404 | return error; | |
3405 | ||
3406 | if (len <= 0) | |
3407 | return XFS_ERROR(EINVAL); | |
3408 | ||
957d0ebe DC |
3409 | rt = XFS_IS_REALTIME_INODE(ip); |
3410 | extsz = xfs_get_extsz_hint(ip); | |
3411 | ||
1da177e4 | 3412 | count = len; |
1da177e4 | 3413 | imapp = &imaps[0]; |
dd9f438e NS |
3414 | nimaps = 1; |
3415 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
3416 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
3417 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
3418 | ||
3419 | /* Generate a DMAPI event if needed. */ | |
ba87ea69 | 3420 | if (alloc_type != 0 && offset < ip->i_size && |
1da177e4 | 3421 | (attr_flags&ATTR_DMI) == 0 && |
eb9df39d | 3422 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
1da177e4 LT |
3423 | xfs_off_t end_dmi_offset; |
3424 | ||
3425 | end_dmi_offset = offset+len; | |
ba87ea69 LM |
3426 | if (end_dmi_offset > ip->i_size) |
3427 | end_dmi_offset = ip->i_size; | |
bc4ac74a CH |
3428 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset, |
3429 | end_dmi_offset - offset, 0, NULL); | |
1da177e4 | 3430 | if (error) |
014c2544 | 3431 | return error; |
1da177e4 LT |
3432 | } |
3433 | ||
3434 | /* | |
dd9f438e | 3435 | * Allocate file space until done or until there is an error |
1da177e4 LT |
3436 | */ |
3437 | retry: | |
3438 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
3439 | xfs_fileoff_t s, e; |
3440 | ||
1da177e4 | 3441 | /* |
3ddb8fa9 | 3442 | * Determine space reservations for data/realtime. |
1da177e4 | 3443 | */ |
dd9f438e | 3444 | if (unlikely(extsz)) { |
1da177e4 | 3445 | s = startoffset_fsb; |
dd9f438e NS |
3446 | do_div(s, extsz); |
3447 | s *= extsz; | |
3448 | e = startoffset_fsb + allocatesize_fsb; | |
3449 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
3450 | e += temp; | |
3451 | if ((temp = do_mod(e, extsz))) | |
3452 | e += extsz - temp; | |
3453 | } else { | |
3454 | s = 0; | |
3455 | e = allocatesize_fsb; | |
3456 | } | |
3457 | ||
3458 | if (unlikely(rt)) { | |
3459 | resrtextents = qblocks = (uint)(e - s); | |
3460 | resrtextents /= mp->m_sb.sb_rextsize; | |
3461 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
3462 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 3463 | } else { |
dd9f438e NS |
3464 | resrtextents = 0; |
3465 | resblks = qblocks = \ | |
3466 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
3467 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
3468 | } |
3469 | ||
3470 | /* | |
dd9f438e | 3471 | * Allocate and setup the transaction. |
1da177e4 LT |
3472 | */ |
3473 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
3474 | error = xfs_trans_reserve(tp, resblks, |
3475 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
3476 | XFS_TRANS_PERM_LOG_RES, |
3477 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 3478 | /* |
dd9f438e | 3479 | * Check for running out of space |
1da177e4 LT |
3480 | */ |
3481 | if (error) { | |
3482 | /* | |
3483 | * Free the transaction structure. | |
3484 | */ | |
3485 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
3486 | xfs_trans_cancel(tp, 0); | |
3487 | break; | |
3488 | } | |
3489 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
dd9f438e NS |
3490 | error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, |
3491 | qblocks, 0, quota_flag); | |
1da177e4 LT |
3492 | if (error) |
3493 | goto error1; | |
3494 | ||
3495 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3496 | xfs_trans_ihold(tp, ip); | |
3497 | ||
3498 | /* | |
dd9f438e | 3499 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 LT |
3500 | */ |
3501 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
541d7d3c | 3502 | error = xfs_bmapi(tp, ip, startoffset_fsb, |
dd9f438e NS |
3503 | allocatesize_fsb, bmapi_flag, |
3504 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 3505 | &free_list, NULL); |
1da177e4 LT |
3506 | if (error) { |
3507 | goto error0; | |
3508 | } | |
3509 | ||
3510 | /* | |
dd9f438e | 3511 | * Complete the transaction |
1da177e4 | 3512 | */ |
f7c99b6f | 3513 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3514 | if (error) { |
3515 | goto error0; | |
3516 | } | |
3517 | ||
1c72bf90 | 3518 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3519 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
3520 | if (error) { | |
3521 | break; | |
3522 | } | |
3523 | ||
3524 | allocated_fsb = imapp->br_blockcount; | |
3525 | ||
dd9f438e | 3526 | if (nimaps == 0) { |
1da177e4 LT |
3527 | error = XFS_ERROR(ENOSPC); |
3528 | break; | |
3529 | } | |
3530 | ||
3531 | startoffset_fsb += allocated_fsb; | |
3532 | allocatesize_fsb -= allocated_fsb; | |
3533 | } | |
3534 | dmapi_enospc_check: | |
eb9df39d CH |
3535 | if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 && |
3536 | DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) { | |
1da177e4 | 3537 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, |
bc4ac74a CH |
3538 | ip, DM_RIGHT_NULL, |
3539 | ip, DM_RIGHT_NULL, | |
1da177e4 LT |
3540 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ |
3541 | if (error == 0) | |
3542 | goto retry; /* Maybe DMAPI app. has made space */ | |
3543 | /* else fall through with error from XFS_SEND_DATA */ | |
3544 | } | |
3545 | ||
3546 | return error; | |
3547 | ||
dd9f438e | 3548 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 3549 | xfs_bmap_cancel(&free_list); |
dd9f438e NS |
3550 | XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag); |
3551 | ||
3552 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
3553 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
3554 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3555 | goto dmapi_enospc_check; | |
3556 | } | |
3557 | ||
3558 | /* | |
3559 | * Zero file bytes between startoff and endoff inclusive. | |
3560 | * The iolock is held exclusive and no blocks are buffered. | |
3561 | */ | |
3562 | STATIC int | |
3563 | xfs_zero_remaining_bytes( | |
3564 | xfs_inode_t *ip, | |
3565 | xfs_off_t startoff, | |
3566 | xfs_off_t endoff) | |
3567 | { | |
3568 | xfs_bmbt_irec_t imap; | |
3569 | xfs_fileoff_t offset_fsb; | |
3570 | xfs_off_t lastoffset; | |
3571 | xfs_off_t offset; | |
3572 | xfs_buf_t *bp; | |
3573 | xfs_mount_t *mp = ip->i_mount; | |
3574 | int nimap; | |
3575 | int error = 0; | |
3576 | ||
3577 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, | |
71ddabb9 | 3578 | XFS_IS_REALTIME_INODE(ip) ? |
1da177e4 LT |
3579 | mp->m_rtdev_targp : mp->m_ddev_targp); |
3580 | ||
3581 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
3582 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
3583 | nimap = 1; | |
541d7d3c | 3584 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, |
3e57ecf6 | 3585 | NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
3586 | if (error || nimap < 1) |
3587 | break; | |
3588 | ASSERT(imap.br_blockcount >= 1); | |
3589 | ASSERT(imap.br_startoff == offset_fsb); | |
3590 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
3591 | if (lastoffset > endoff) | |
3592 | lastoffset = endoff; | |
3593 | if (imap.br_startblock == HOLESTARTBLOCK) | |
3594 | continue; | |
3595 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
3596 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
3597 | continue; | |
3598 | XFS_BUF_UNDONE(bp); | |
3599 | XFS_BUF_UNWRITE(bp); | |
3600 | XFS_BUF_READ(bp); | |
3601 | XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock)); | |
3602 | xfsbdstrat(mp, bp); | |
d64e31a2 DC |
3603 | error = xfs_iowait(bp); |
3604 | if (error) { | |
1da177e4 LT |
3605 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", |
3606 | mp, bp, XFS_BUF_ADDR(bp)); | |
3607 | break; | |
3608 | } | |
3609 | memset(XFS_BUF_PTR(bp) + | |
3610 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
3611 | 0, lastoffset - offset + 1); | |
3612 | XFS_BUF_UNDONE(bp); | |
3613 | XFS_BUF_UNREAD(bp); | |
3614 | XFS_BUF_WRITE(bp); | |
3615 | xfsbdstrat(mp, bp); | |
d64e31a2 DC |
3616 | error = xfs_iowait(bp); |
3617 | if (error) { | |
1da177e4 LT |
3618 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", |
3619 | mp, bp, XFS_BUF_ADDR(bp)); | |
3620 | break; | |
3621 | } | |
3622 | } | |
3623 | xfs_buf_free(bp); | |
3624 | return error; | |
3625 | } | |
3626 | ||
3627 | /* | |
3628 | * xfs_free_file_space() | |
3629 | * This routine frees disk space for the given file. | |
3630 | * | |
3631 | * This routine is only called by xfs_change_file_space | |
3632 | * for an UNRESVSP type call. | |
3633 | * | |
3634 | * RETURNS: | |
3635 | * 0 on success | |
3636 | * errno on error | |
3637 | * | |
3638 | */ | |
3639 | STATIC int | |
3640 | xfs_free_file_space( | |
3641 | xfs_inode_t *ip, | |
3642 | xfs_off_t offset, | |
3643 | xfs_off_t len, | |
3644 | int attr_flags) | |
3645 | { | |
67fcaa73 | 3646 | bhv_vnode_t *vp; |
1da177e4 LT |
3647 | int committed; |
3648 | int done; | |
3649 | xfs_off_t end_dmi_offset; | |
3650 | xfs_fileoff_t endoffset_fsb; | |
3651 | int error; | |
3652 | xfs_fsblock_t firstfsb; | |
3653 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
3654 | xfs_bmbt_irec_t imap; |
3655 | xfs_off_t ioffset; | |
3656 | xfs_extlen_t mod=0; | |
3657 | xfs_mount_t *mp; | |
3658 | int nimap; | |
3659 | uint resblks; | |
673cdf5c | 3660 | uint rounding; |
1da177e4 LT |
3661 | int rt; |
3662 | xfs_fileoff_t startoffset_fsb; | |
3663 | xfs_trans_t *tp; | |
5fcbab35 | 3664 | int need_iolock = 1; |
1da177e4 | 3665 | |
bd5a876a | 3666 | vp = XFS_ITOV(ip); |
1da177e4 LT |
3667 | mp = ip->i_mount; |
3668 | ||
cf441eeb | 3669 | xfs_itrace_entry(ip); |
bd5a876a | 3670 | |
1da177e4 LT |
3671 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
3672 | return error; | |
3673 | ||
3674 | error = 0; | |
3675 | if (len <= 0) /* if nothing being freed */ | |
3676 | return error; | |
71ddabb9 | 3677 | rt = XFS_IS_REALTIME_INODE(ip); |
1da177e4 LT |
3678 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); |
3679 | end_dmi_offset = offset + len; | |
3680 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
3681 | ||
eb9df39d CH |
3682 | if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 && |
3683 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { | |
ba87ea69 LM |
3684 | if (end_dmi_offset > ip->i_size) |
3685 | end_dmi_offset = ip->i_size; | |
bc4ac74a | 3686 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, |
1da177e4 LT |
3687 | offset, end_dmi_offset - offset, |
3688 | AT_DELAY_FLAG(attr_flags), NULL); | |
3689 | if (error) | |
014c2544 | 3690 | return error; |
1da177e4 LT |
3691 | } |
3692 | ||
5fcbab35 DR |
3693 | if (attr_flags & ATTR_NOLOCK) |
3694 | need_iolock = 0; | |
9fa8046f | 3695 | if (need_iolock) { |
1da177e4 | 3696 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
b677c210 | 3697 | vn_iowait(ip); /* wait for the completion of any pending DIOs */ |
9fa8046f | 3698 | } |
5fcbab35 | 3699 | |
e6a4b37f | 3700 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1da177e4 | 3701 | ioffset = offset & ~(rounding - 1); |
bd5a876a CH |
3702 | |
3703 | if (VN_CACHED(vp) != 0) { | |
e6a4b37f TS |
3704 | xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1); |
3705 | error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED); | |
d3cf2094 LM |
3706 | if (error) |
3707 | goto out_unlock_iolock; | |
bd5a876a CH |
3708 | } |
3709 | ||
1da177e4 LT |
3710 | /* |
3711 | * Need to zero the stuff we're not freeing, on disk. | |
3712 | * If its a realtime file & can't use unwritten extents then we | |
3713 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
3714 | * will take care of it for us. | |
3715 | */ | |
62118709 | 3716 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
1da177e4 | 3717 | nimap = 1; |
541d7d3c | 3718 | error = xfs_bmapi(NULL, ip, startoffset_fsb, |
3e57ecf6 | 3719 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
3720 | if (error) |
3721 | goto out_unlock_iolock; | |
3722 | ASSERT(nimap == 0 || nimap == 1); | |
3723 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
3724 | xfs_daddr_t block; | |
3725 | ||
3726 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
3727 | block = imap.br_startblock; | |
3728 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
3729 | if (mod) | |
3730 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
3731 | } | |
3732 | nimap = 1; | |
541d7d3c | 3733 | error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, |
3e57ecf6 | 3734 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
3735 | if (error) |
3736 | goto out_unlock_iolock; | |
3737 | ASSERT(nimap == 0 || nimap == 1); | |
3738 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
3739 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
3740 | mod++; | |
3741 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
3742 | endoffset_fsb -= mod; | |
3743 | } | |
3744 | } | |
3745 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
3746 | /* | |
3747 | * One contiguous piece to clear | |
3748 | */ | |
3749 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
3750 | else { | |
3751 | /* | |
3752 | * Some full blocks, possibly two pieces to clear | |
3753 | */ | |
3754 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
3755 | error = xfs_zero_remaining_bytes(ip, offset, | |
3756 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
3757 | if (!error && | |
3758 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
3759 | error = xfs_zero_remaining_bytes(ip, | |
3760 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
3761 | offset + len - 1); | |
3762 | } | |
3763 | ||
3764 | /* | |
3765 | * free file space until done or until there is an error | |
3766 | */ | |
3767 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
3768 | while (!error && !done) { | |
3769 | ||
3770 | /* | |
91ebecc7 DC |
3771 | * allocate and setup the transaction. Allow this |
3772 | * transaction to dip into the reserve blocks to ensure | |
3773 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
3774 | */ |
3775 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 3776 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
3777 | error = xfs_trans_reserve(tp, |
3778 | resblks, | |
3779 | XFS_WRITE_LOG_RES(mp), | |
3780 | 0, | |
3781 | XFS_TRANS_PERM_LOG_RES, | |
3782 | XFS_WRITE_LOG_COUNT); | |
3783 | ||
3784 | /* | |
3785 | * check for running out of space | |
3786 | */ | |
3787 | if (error) { | |
3788 | /* | |
3789 | * Free the transaction structure. | |
3790 | */ | |
3791 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
3792 | xfs_trans_cancel(tp, 0); | |
3793 | break; | |
3794 | } | |
3795 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3796 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, | |
dd9f438e NS |
3797 | ip->i_udquot, ip->i_gdquot, resblks, 0, |
3798 | XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
3799 | if (error) |
3800 | goto error1; | |
3801 | ||
3802 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3803 | xfs_trans_ihold(tp, ip); | |
3804 | ||
3805 | /* | |
3806 | * issue the bunmapi() call to free the blocks | |
3807 | */ | |
3808 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
541d7d3c | 3809 | error = xfs_bunmapi(tp, ip, startoffset_fsb, |
1da177e4 | 3810 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 3811 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
3812 | if (error) { |
3813 | goto error0; | |
3814 | } | |
3815 | ||
3816 | /* | |
3817 | * complete the transaction | |
3818 | */ | |
f7c99b6f | 3819 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3820 | if (error) { |
3821 | goto error0; | |
3822 | } | |
3823 | ||
1c72bf90 | 3824 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3825 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
3826 | } | |
3827 | ||
3828 | out_unlock_iolock: | |
3829 | if (need_iolock) | |
3830 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
3831 | return error; | |
3832 | ||
3833 | error0: | |
3834 | xfs_bmap_cancel(&free_list); | |
3835 | error1: | |
3836 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
3837 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
3838 | XFS_ILOCK_EXCL); | |
3839 | return error; | |
3840 | } | |
3841 | ||
3842 | /* | |
3843 | * xfs_change_file_space() | |
3844 | * This routine allocates or frees disk space for the given file. | |
3845 | * The user specified parameters are checked for alignment and size | |
3846 | * limitations. | |
3847 | * | |
3848 | * RETURNS: | |
3849 | * 0 on success | |
3850 | * errno on error | |
3851 | * | |
3852 | */ | |
3853 | int | |
3854 | xfs_change_file_space( | |
993386c1 | 3855 | xfs_inode_t *ip, |
1da177e4 LT |
3856 | int cmd, |
3857 | xfs_flock64_t *bf, | |
3858 | xfs_off_t offset, | |
3859 | cred_t *credp, | |
3860 | int attr_flags) | |
3861 | { | |
993386c1 | 3862 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
3863 | int clrprealloc; |
3864 | int error; | |
3865 | xfs_fsize_t fsize; | |
1da177e4 LT |
3866 | int setprealloc; |
3867 | xfs_off_t startoffset; | |
3868 | xfs_off_t llen; | |
3869 | xfs_trans_t *tp; | |
8285fb58 | 3870 | bhv_vattr_t va; |
1da177e4 | 3871 | |
cf441eeb | 3872 | xfs_itrace_entry(ip); |
1da177e4 | 3873 | |
993386c1 | 3874 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
3875 | return XFS_ERROR(EINVAL); |
3876 | ||
1da177e4 LT |
3877 | switch (bf->l_whence) { |
3878 | case 0: /*SEEK_SET*/ | |
3879 | break; | |
3880 | case 1: /*SEEK_CUR*/ | |
3881 | bf->l_start += offset; | |
3882 | break; | |
3883 | case 2: /*SEEK_END*/ | |
ba87ea69 | 3884 | bf->l_start += ip->i_size; |
1da177e4 LT |
3885 | break; |
3886 | default: | |
3887 | return XFS_ERROR(EINVAL); | |
3888 | } | |
3889 | ||
3890 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
3891 | ||
3892 | if ( (bf->l_start < 0) | |
3893 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
3894 | || (bf->l_start + llen < 0) | |
3895 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
3896 | return XFS_ERROR(EINVAL); | |
3897 | ||
3898 | bf->l_whence = 0; | |
3899 | ||
3900 | startoffset = bf->l_start; | |
ba87ea69 | 3901 | fsize = ip->i_size; |
1da177e4 LT |
3902 | |
3903 | /* | |
3904 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
3905 | * file space. | |
3906 | * These calls do NOT zero the data space allocated to the file, | |
3907 | * nor do they change the file size. | |
3908 | * | |
3909 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
3910 | * space. | |
3911 | * These calls cause the new file data to be zeroed and the file | |
3912 | * size to be changed. | |
3913 | */ | |
3914 | setprealloc = clrprealloc = 0; | |
3915 | ||
3916 | switch (cmd) { | |
3917 | case XFS_IOC_RESVSP: | |
3918 | case XFS_IOC_RESVSP64: | |
3919 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
3920 | 1, attr_flags); | |
3921 | if (error) | |
3922 | return error; | |
3923 | setprealloc = 1; | |
3924 | break; | |
3925 | ||
3926 | case XFS_IOC_UNRESVSP: | |
3927 | case XFS_IOC_UNRESVSP64: | |
3928 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
3929 | attr_flags))) | |
3930 | return error; | |
3931 | break; | |
3932 | ||
3933 | case XFS_IOC_ALLOCSP: | |
3934 | case XFS_IOC_ALLOCSP64: | |
3935 | case XFS_IOC_FREESP: | |
3936 | case XFS_IOC_FREESP64: | |
3937 | if (startoffset > fsize) { | |
3938 | error = xfs_alloc_file_space(ip, fsize, | |
3939 | startoffset - fsize, 0, attr_flags); | |
3940 | if (error) | |
3941 | break; | |
3942 | } | |
3943 | ||
3944 | va.va_mask = XFS_AT_SIZE; | |
3945 | va.va_size = startoffset; | |
3946 | ||
993386c1 | 3947 | error = xfs_setattr(ip, &va, attr_flags, credp); |
1da177e4 LT |
3948 | |
3949 | if (error) | |
3950 | return error; | |
3951 | ||
3952 | clrprealloc = 1; | |
3953 | break; | |
3954 | ||
3955 | default: | |
3956 | ASSERT(0); | |
3957 | return XFS_ERROR(EINVAL); | |
3958 | } | |
3959 | ||
3960 | /* | |
3961 | * update the inode timestamp, mode, and prealloc flag bits | |
3962 | */ | |
3963 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
3964 | ||
3965 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
3966 | 0, 0, 0))) { | |
3967 | /* ASSERT(0); */ | |
3968 | xfs_trans_cancel(tp, 0); | |
3969 | return error; | |
3970 | } | |
3971 | ||
3972 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3973 | ||
3974 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3975 | xfs_trans_ihold(tp, ip); | |
3976 | ||
3977 | if ((attr_flags & ATTR_DMI) == 0) { | |
3978 | ip->i_d.di_mode &= ~S_ISUID; | |
3979 | ||
3980 | /* | |
3981 | * Note that we don't have to worry about mandatory | |
3982 | * file locking being disabled here because we only | |
3983 | * clear the S_ISGID bit if the Group execute bit is | |
3984 | * on, but if it was on then mandatory locking wouldn't | |
3985 | * have been enabled. | |
3986 | */ | |
3987 | if (ip->i_d.di_mode & S_IXGRP) | |
3988 | ip->i_d.di_mode &= ~S_ISGID; | |
3989 | ||
3990 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3991 | } | |
3992 | if (setprealloc) | |
3993 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
3994 | else if (clrprealloc) | |
3995 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
3996 | ||
3997 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3998 | xfs_trans_set_sync(tp); | |
3999 | ||
1c72bf90 | 4000 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
4001 | |
4002 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4003 | ||
4004 | return error; | |
4005 | } |