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 NS |
50 | #include "xfs_rtalloc.h" |
51 | #include "xfs_refcache.h" | |
1da177e4 | 52 | #include "xfs_trans_space.h" |
1da177e4 | 53 | #include "xfs_log_priv.h" |
2a82b8be | 54 | #include "xfs_filestream.h" |
993386c1 | 55 | #include "xfs_vnodeops.h" |
1da177e4 | 56 | |
993386c1 | 57 | int |
1da177e4 | 58 | xfs_open( |
993386c1 | 59 | xfs_inode_t *ip) |
1da177e4 LT |
60 | { |
61 | int mode; | |
1da177e4 LT |
62 | |
63 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
64 | return XFS_ERROR(EIO); | |
65 | ||
66 | /* | |
67 | * If it's a directory with any blocks, read-ahead block 0 | |
68 | * as we're almost certain to have the next operation be a read there. | |
69 | */ | |
993386c1 | 70 | if (S_ISDIR(ip->i_d.di_mode) && ip->i_d.di_nextents > 0) { |
1da177e4 LT |
71 | mode = xfs_ilock_map_shared(ip); |
72 | if (ip->i_d.di_nextents > 0) | |
73 | (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK); | |
74 | xfs_iunlock(ip, mode); | |
75 | } | |
76 | return 0; | |
77 | } | |
78 | ||
1da177e4 LT |
79 | /* |
80 | * xfs_getattr | |
81 | */ | |
993386c1 | 82 | int |
1da177e4 | 83 | xfs_getattr( |
993386c1 | 84 | xfs_inode_t *ip, |
8285fb58 | 85 | bhv_vattr_t *vap, |
993386c1 | 86 | int flags) |
1da177e4 | 87 | { |
993386c1 CH |
88 | bhv_vnode_t *vp = XFS_ITOV(ip); |
89 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 | 90 | |
1543d79c | 91 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 92 | |
1da177e4 LT |
93 | if (XFS_FORCED_SHUTDOWN(mp)) |
94 | return XFS_ERROR(EIO); | |
95 | ||
96 | if (!(flags & ATTR_LAZY)) | |
97 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
98 | ||
ba87ea69 | 99 | vap->va_size = XFS_ISIZE(ip); |
1da177e4 LT |
100 | if (vap->va_mask == XFS_AT_SIZE) |
101 | goto all_done; | |
102 | ||
103 | vap->va_nblocks = | |
104 | XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); | |
105 | vap->va_nodeid = ip->i_ino; | |
106 | #if XFS_BIG_INUMS | |
107 | vap->va_nodeid += mp->m_inoadd; | |
108 | #endif | |
109 | vap->va_nlink = ip->i_d.di_nlink; | |
110 | ||
111 | /* | |
112 | * Quick exit for non-stat callers | |
113 | */ | |
114 | if ((vap->va_mask & | |
115 | ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID| | |
116 | XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0) | |
117 | goto all_done; | |
118 | ||
119 | /* | |
120 | * Copy from in-core inode. | |
121 | */ | |
0432dab2 | 122 | vap->va_mode = ip->i_d.di_mode; |
1da177e4 LT |
123 | vap->va_uid = ip->i_d.di_uid; |
124 | vap->va_gid = ip->i_d.di_gid; | |
125 | vap->va_projid = ip->i_d.di_projid; | |
126 | ||
127 | /* | |
128 | * Check vnode type block/char vs. everything else. | |
1da177e4 | 129 | */ |
0432dab2 CH |
130 | switch (ip->i_d.di_mode & S_IFMT) { |
131 | case S_IFBLK: | |
132 | case S_IFCHR: | |
133 | vap->va_rdev = ip->i_df.if_u2.if_rdev; | |
134 | vap->va_blocksize = BLKDEV_IOSIZE; | |
135 | break; | |
136 | default: | |
1da177e4 LT |
137 | vap->va_rdev = 0; |
138 | ||
139 | if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) { | |
e8c8b3a7 | 140 | vap->va_blocksize = xfs_preferred_iosize(mp); |
1da177e4 LT |
141 | } else { |
142 | ||
143 | /* | |
144 | * If the file blocks are being allocated from a | |
145 | * realtime partition, then return the inode's | |
146 | * realtime extent size or the realtime volume's | |
147 | * extent size. | |
148 | */ | |
957d0ebe DC |
149 | vap->va_blocksize = |
150 | xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog; | |
1da177e4 | 151 | } |
0432dab2 | 152 | break; |
1da177e4 LT |
153 | } |
154 | ||
ca5ccbf9 | 155 | vn_atime_to_timespec(vp, &vap->va_atime); |
1da177e4 LT |
156 | vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec; |
157 | vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec; | |
158 | vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec; | |
159 | vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec; | |
160 | ||
161 | /* | |
162 | * Exit for stat callers. See if any of the rest of the fields | |
163 | * to be filled in are needed. | |
164 | */ | |
165 | if ((vap->va_mask & | |
166 | (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS| | |
167 | XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0) | |
168 | goto all_done; | |
169 | ||
170 | /* | |
171 | * Convert di_flags to xflags. | |
172 | */ | |
173 | vap->va_xflags = xfs_ip2xflags(ip); | |
174 | ||
175 | /* | |
176 | * Exit for inode revalidate. See if any of the rest of | |
177 | * the fields to be filled in are needed. | |
178 | */ | |
179 | if ((vap->va_mask & | |
180 | (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS| | |
181 | XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0) | |
182 | goto all_done; | |
183 | ||
184 | vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog; | |
185 | vap->va_nextents = | |
186 | (ip->i_df.if_flags & XFS_IFEXTENTS) ? | |
187 | ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) : | |
188 | ip->i_d.di_nextents; | |
189 | if (ip->i_afp) | |
190 | vap->va_anextents = | |
191 | (ip->i_afp->if_flags & XFS_IFEXTENTS) ? | |
192 | ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) : | |
193 | ip->i_d.di_anextents; | |
194 | else | |
195 | vap->va_anextents = 0; | |
196 | vap->va_gen = ip->i_d.di_gen; | |
197 | ||
198 | all_done: | |
199 | if (!(flags & ATTR_LAZY)) | |
200 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
201 | return 0; | |
202 | } | |
203 | ||
204 | ||
205 | /* | |
206 | * xfs_setattr | |
207 | */ | |
208 | int | |
209 | xfs_setattr( | |
993386c1 | 210 | xfs_inode_t *ip, |
8285fb58 | 211 | bhv_vattr_t *vap, |
1da177e4 LT |
212 | int flags, |
213 | cred_t *credp) | |
214 | { | |
993386c1 CH |
215 | bhv_vnode_t *vp = XFS_ITOV(ip); |
216 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 | 217 | xfs_trans_t *tp; |
1da177e4 LT |
218 | int mask; |
219 | int code; | |
220 | uint lock_flags; | |
221 | uint commit_flags=0; | |
222 | uid_t uid=0, iuid=0; | |
223 | gid_t gid=0, igid=0; | |
224 | int timeflags = 0; | |
1da177e4 LT |
225 | xfs_prid_t projid=0, iprojid=0; |
226 | int mandlock_before, mandlock_after; | |
227 | struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2; | |
228 | int file_owner; | |
5fcbab35 | 229 | int need_iolock = 1; |
1da177e4 | 230 | |
1543d79c | 231 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 232 | |
bd186aa9 | 233 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
234 | return XFS_ERROR(EROFS); |
235 | ||
236 | /* | |
237 | * Cannot set certain attributes. | |
238 | */ | |
239 | mask = vap->va_mask; | |
240 | if (mask & XFS_AT_NOSET) { | |
241 | return XFS_ERROR(EINVAL); | |
242 | } | |
243 | ||
1da177e4 LT |
244 | if (XFS_FORCED_SHUTDOWN(mp)) |
245 | return XFS_ERROR(EIO); | |
246 | ||
247 | /* | |
248 | * Timestamps do not need to be logged and hence do not | |
249 | * need to be done within a transaction. | |
250 | */ | |
251 | if (mask & XFS_AT_UPDTIMES) { | |
252 | ASSERT((mask & ~XFS_AT_UPDTIMES) == 0); | |
253 | timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) | | |
254 | ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) | | |
255 | ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0); | |
256 | xfs_ichgtime(ip, timeflags); | |
257 | return 0; | |
258 | } | |
259 | ||
260 | olddquot1 = olddquot2 = NULL; | |
261 | udqp = gdqp = NULL; | |
262 | ||
263 | /* | |
264 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
265 | * before we start any other transactions. Trying to do this later | |
266 | * is messy. We don't care to take a readlock to look at the ids | |
267 | * in inode here, because we can't hold it across the trans_reserve. | |
268 | * If the IDs do change before we take the ilock, we're covered | |
269 | * because the i_*dquot fields will get updated anyway. | |
270 | */ | |
c8ad20ff NS |
271 | if (XFS_IS_QUOTA_ON(mp) && |
272 | (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) { | |
1da177e4 LT |
273 | uint qflags = 0; |
274 | ||
c8ad20ff | 275 | if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) { |
1da177e4 LT |
276 | uid = vap->va_uid; |
277 | qflags |= XFS_QMOPT_UQUOTA; | |
278 | } else { | |
279 | uid = ip->i_d.di_uid; | |
280 | } | |
c8ad20ff | 281 | if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) { |
1da177e4 LT |
282 | gid = vap->va_gid; |
283 | qflags |= XFS_QMOPT_GQUOTA; | |
284 | } else { | |
285 | gid = ip->i_d.di_gid; | |
286 | } | |
c8ad20ff NS |
287 | if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) { |
288 | projid = vap->va_projid; | |
289 | qflags |= XFS_QMOPT_PQUOTA; | |
290 | } else { | |
291 | projid = ip->i_d.di_projid; | |
292 | } | |
1da177e4 LT |
293 | /* |
294 | * We take a reference when we initialize udqp and gdqp, | |
295 | * so it is important that we never blindly double trip on | |
296 | * the same variable. See xfs_create() for an example. | |
297 | */ | |
298 | ASSERT(udqp == NULL); | |
299 | ASSERT(gdqp == NULL); | |
c8ad20ff NS |
300 | code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags, |
301 | &udqp, &gdqp); | |
1da177e4 | 302 | if (code) |
014c2544 | 303 | return code; |
1da177e4 LT |
304 | } |
305 | ||
306 | /* | |
307 | * For the other attributes, we acquire the inode lock and | |
308 | * first do an error checking pass. | |
309 | */ | |
310 | tp = NULL; | |
311 | lock_flags = XFS_ILOCK_EXCL; | |
5fcbab35 DR |
312 | if (flags & ATTR_NOLOCK) |
313 | need_iolock = 0; | |
1da177e4 LT |
314 | if (!(mask & XFS_AT_SIZE)) { |
315 | if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) || | |
316 | (mp->m_flags & XFS_MOUNT_WSYNC)) { | |
317 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); | |
318 | commit_flags = 0; | |
319 | if ((code = xfs_trans_reserve(tp, 0, | |
320 | XFS_ICHANGE_LOG_RES(mp), 0, | |
321 | 0, 0))) { | |
322 | lock_flags = 0; | |
323 | goto error_return; | |
324 | } | |
325 | } | |
326 | } else { | |
eb9df39d | 327 | if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) && |
1da177e4 LT |
328 | !(flags & ATTR_DMI)) { |
329 | int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR; | |
330 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp, | |
331 | vap->va_size, 0, dmflags, NULL); | |
332 | if (code) { | |
333 | lock_flags = 0; | |
334 | goto error_return; | |
335 | } | |
336 | } | |
337 | if (need_iolock) | |
338 | lock_flags |= XFS_IOLOCK_EXCL; | |
339 | } | |
340 | ||
341 | xfs_ilock(ip, lock_flags); | |
342 | ||
343 | /* boolean: are we the file owner? */ | |
344 | file_owner = (current_fsuid(credp) == ip->i_d.di_uid); | |
345 | ||
346 | /* | |
347 | * Change various properties of a file. | |
348 | * Only the owner or users with CAP_FOWNER | |
349 | * capability may do these things. | |
350 | */ | |
351 | if (mask & | |
352 | (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID| | |
353 | XFS_AT_GID|XFS_AT_PROJID)) { | |
354 | /* | |
355 | * CAP_FOWNER overrides the following restrictions: | |
356 | * | |
357 | * The user ID of the calling process must be equal | |
358 | * to the file owner ID, except in cases where the | |
359 | * CAP_FSETID capability is applicable. | |
360 | */ | |
361 | if (!file_owner && !capable(CAP_FOWNER)) { | |
362 | code = XFS_ERROR(EPERM); | |
363 | goto error_return; | |
364 | } | |
365 | ||
366 | /* | |
367 | * CAP_FSETID overrides the following restrictions: | |
368 | * | |
369 | * The effective user ID of the calling process shall match | |
370 | * the file owner when setting the set-user-ID and | |
371 | * set-group-ID bits on that file. | |
372 | * | |
373 | * The effective group ID or one of the supplementary group | |
374 | * IDs of the calling process shall match the group owner of | |
375 | * the file when setting the set-group-ID bit on that file | |
376 | */ | |
377 | if (mask & XFS_AT_MODE) { | |
378 | mode_t m = 0; | |
379 | ||
380 | if ((vap->va_mode & S_ISUID) && !file_owner) | |
381 | m |= S_ISUID; | |
382 | if ((vap->va_mode & S_ISGID) && | |
383 | !in_group_p((gid_t)ip->i_d.di_gid)) | |
384 | m |= S_ISGID; | |
385 | #if 0 | |
386 | /* Linux allows this, Irix doesn't. */ | |
0432dab2 | 387 | if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp)) |
1da177e4 LT |
388 | m |= S_ISVTX; |
389 | #endif | |
390 | if (m && !capable(CAP_FSETID)) | |
391 | vap->va_mode &= ~m; | |
392 | } | |
393 | } | |
394 | ||
395 | /* | |
396 | * Change file ownership. Must be the owner or privileged. | |
397 | * If the system was configured with the "restricted_chown" | |
398 | * option, the owner is not permitted to give away the file, | |
399 | * and can change the group id only to a group of which he | |
400 | * or she is a member. | |
401 | */ | |
402 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
403 | /* | |
404 | * These IDs could have changed since we last looked at them. | |
405 | * But, we're assured that if the ownership did change | |
406 | * while we didn't have the inode locked, inode's dquot(s) | |
407 | * would have changed also. | |
408 | */ | |
409 | iuid = ip->i_d.di_uid; | |
410 | iprojid = ip->i_d.di_projid; | |
411 | igid = ip->i_d.di_gid; | |
412 | gid = (mask & XFS_AT_GID) ? vap->va_gid : igid; | |
413 | uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid; | |
414 | projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid : | |
415 | iprojid; | |
416 | ||
417 | /* | |
418 | * CAP_CHOWN overrides the following restrictions: | |
419 | * | |
420 | * If _POSIX_CHOWN_RESTRICTED is defined, this capability | |
421 | * shall override the restriction that a process cannot | |
422 | * change the user ID of a file it owns and the restriction | |
423 | * that the group ID supplied to the chown() function | |
424 | * shall be equal to either the group ID or one of the | |
425 | * supplementary group IDs of the calling process. | |
1da177e4 LT |
426 | */ |
427 | if (restricted_chown && | |
428 | (iuid != uid || (igid != gid && | |
429 | !in_group_p((gid_t)gid))) && | |
430 | !capable(CAP_CHOWN)) { | |
431 | code = XFS_ERROR(EPERM); | |
432 | goto error_return; | |
433 | } | |
434 | /* | |
c8ad20ff | 435 | * Do a quota reservation only if uid/projid/gid is actually |
1da177e4 LT |
436 | * going to change. |
437 | */ | |
438 | if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || | |
c8ad20ff | 439 | (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) || |
1da177e4 LT |
440 | (XFS_IS_GQUOTA_ON(mp) && igid != gid)) { |
441 | ASSERT(tp); | |
442 | code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp, | |
443 | capable(CAP_FOWNER) ? | |
444 | XFS_QMOPT_FORCE_RES : 0); | |
445 | if (code) /* out of quota */ | |
446 | goto error_return; | |
447 | } | |
448 | } | |
449 | ||
450 | /* | |
451 | * Truncate file. Must have write permission and not be a directory. | |
452 | */ | |
453 | if (mask & XFS_AT_SIZE) { | |
454 | /* Short circuit the truncate case for zero length files */ | |
455 | if ((vap->va_size == 0) && | |
ba87ea69 | 456 | (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) { |
1da177e4 LT |
457 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
458 | lock_flags &= ~XFS_ILOCK_EXCL; | |
459 | if (mask & XFS_AT_CTIME) | |
460 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
461 | code = 0; | |
462 | goto error_return; | |
463 | } | |
464 | ||
0432dab2 | 465 | if (VN_ISDIR(vp)) { |
1da177e4 LT |
466 | code = XFS_ERROR(EISDIR); |
467 | goto error_return; | |
0432dab2 | 468 | } else if (!VN_ISREG(vp)) { |
1da177e4 LT |
469 | code = XFS_ERROR(EINVAL); |
470 | goto error_return; | |
471 | } | |
472 | /* | |
473 | * Make sure that the dquots are attached to the inode. | |
474 | */ | |
475 | if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED))) | |
476 | goto error_return; | |
477 | } | |
478 | ||
479 | /* | |
480 | * Change file access or modified times. | |
481 | */ | |
482 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
483 | if (!file_owner) { | |
484 | if ((flags & ATTR_UTIME) && | |
485 | !capable(CAP_FOWNER)) { | |
486 | code = XFS_ERROR(EPERM); | |
487 | goto error_return; | |
488 | } | |
489 | } | |
490 | } | |
491 | ||
492 | /* | |
493 | * Change extent size or realtime flag. | |
494 | */ | |
495 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
496 | /* | |
497 | * Can't change extent size if any extents are allocated. | |
498 | */ | |
e94af02a | 499 | if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) && |
1da177e4 LT |
500 | ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != |
501 | vap->va_extsize) ) { | |
502 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
503 | goto error_return; | |
504 | } | |
505 | ||
1da177e4 LT |
506 | /* |
507 | * Can't change realtime flag if any extents are allocated. | |
508 | */ | |
e94af02a ES |
509 | if ((ip->i_d.di_nextents || ip->i_delayed_blks) && |
510 | (mask & XFS_AT_XFLAGS) && | |
1da177e4 LT |
511 | (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != |
512 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { | |
513 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
514 | goto error_return; | |
515 | } | |
516 | /* | |
517 | * Extent size must be a multiple of the appropriate block | |
518 | * size, if set at all. | |
519 | */ | |
520 | if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) { | |
521 | xfs_extlen_t size; | |
522 | ||
523 | if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) || | |
524 | ((mask & XFS_AT_XFLAGS) && | |
525 | (vap->va_xflags & XFS_XFLAG_REALTIME))) { | |
526 | size = mp->m_sb.sb_rextsize << | |
527 | mp->m_sb.sb_blocklog; | |
528 | } else { | |
529 | size = mp->m_sb.sb_blocksize; | |
530 | } | |
531 | if (vap->va_extsize % size) { | |
532 | code = XFS_ERROR(EINVAL); | |
533 | goto error_return; | |
534 | } | |
535 | } | |
536 | /* | |
537 | * If realtime flag is set then must have realtime data. | |
538 | */ | |
539 | if ((mask & XFS_AT_XFLAGS) && | |
540 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { | |
541 | if ((mp->m_sb.sb_rblocks == 0) || | |
542 | (mp->m_sb.sb_rextsize == 0) || | |
543 | (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) { | |
544 | code = XFS_ERROR(EINVAL); | |
545 | goto error_return; | |
546 | } | |
547 | } | |
548 | ||
549 | /* | |
550 | * Can't modify an immutable/append-only file unless | |
551 | * we have appropriate permission. | |
552 | */ | |
553 | if ((mask & XFS_AT_XFLAGS) && | |
554 | (ip->i_d.di_flags & | |
555 | (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) || | |
556 | (vap->va_xflags & | |
557 | (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) && | |
558 | !capable(CAP_LINUX_IMMUTABLE)) { | |
559 | code = XFS_ERROR(EPERM); | |
560 | goto error_return; | |
561 | } | |
562 | } | |
563 | ||
564 | /* | |
565 | * Now we can make the changes. Before we join the inode | |
566 | * to the transaction, if XFS_AT_SIZE is set then take care of | |
567 | * the part of the truncation that must be done without the | |
568 | * inode lock. This needs to be done before joining the inode | |
569 | * to the transaction, because the inode cannot be unlocked | |
570 | * once it is a part of the transaction. | |
571 | */ | |
572 | if (mask & XFS_AT_SIZE) { | |
573 | code = 0; | |
ba87ea69 | 574 | if ((vap->va_size > ip->i_size) && |
374e2ac3 | 575 | (flags & ATTR_NOSIZETOK) == 0) { |
1da177e4 | 576 | code = xfs_igrow_start(ip, vap->va_size, credp); |
374e2ac3 | 577 | } |
1da177e4 | 578 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
c32676ee DC |
579 | |
580 | /* | |
581 | * We are going to log the inode size change in this | |
582 | * transaction so any previous writes that are beyond the on | |
583 | * disk EOF and the new EOF that have not been written out need | |
584 | * to be written here. If we do not write the data out, we | |
585 | * expose ourselves to the null files problem. | |
586 | * | |
587 | * Only flush from the on disk size to the smaller of the in | |
588 | * memory file size or the new size as that's the range we | |
589 | * really care about here and prevents waiting for other data | |
590 | * not within the range we care about here. | |
591 | */ | |
592 | if (!code && | |
593 | (ip->i_size != ip->i_d.di_size) && | |
594 | (vap->va_size > ip->i_d.di_size)) { | |
739bfb2a | 595 | code = xfs_flush_pages(ip, |
c32676ee DC |
596 | ip->i_d.di_size, vap->va_size, |
597 | XFS_B_ASYNC, FI_NONE); | |
598 | } | |
599 | ||
600 | /* wait for all I/O to complete */ | |
b677c210 | 601 | vn_iowait(ip); |
c32676ee | 602 | |
1da177e4 LT |
603 | if (!code) |
604 | code = xfs_itruncate_data(ip, vap->va_size); | |
605 | if (code) { | |
606 | ASSERT(tp == NULL); | |
607 | lock_flags &= ~XFS_ILOCK_EXCL; | |
608 | ASSERT(lock_flags == XFS_IOLOCK_EXCL); | |
609 | goto error_return; | |
610 | } | |
611 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); | |
612 | if ((code = xfs_trans_reserve(tp, 0, | |
613 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
614 | XFS_TRANS_PERM_LOG_RES, | |
615 | XFS_ITRUNCATE_LOG_COUNT))) { | |
616 | xfs_trans_cancel(tp, 0); | |
617 | if (need_iolock) | |
618 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
619 | return code; | |
620 | } | |
621 | commit_flags = XFS_TRANS_RELEASE_LOG_RES; | |
622 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
623 | } | |
624 | ||
625 | if (tp) { | |
626 | xfs_trans_ijoin(tp, ip, lock_flags); | |
627 | xfs_trans_ihold(tp, ip); | |
628 | } | |
629 | ||
630 | /* determine whether mandatory locking mode changes */ | |
631 | mandlock_before = MANDLOCK(vp, ip->i_d.di_mode); | |
632 | ||
633 | /* | |
634 | * Truncate file. Must have write permission and not be a directory. | |
635 | */ | |
636 | if (mask & XFS_AT_SIZE) { | |
ba87ea69 | 637 | if (vap->va_size > ip->i_size) { |
1da177e4 LT |
638 | xfs_igrow_finish(tp, ip, vap->va_size, |
639 | !(flags & ATTR_DMI)); | |
ba87ea69 | 640 | } else if ((vap->va_size <= ip->i_size) || |
1da177e4 LT |
641 | ((vap->va_size == 0) && ip->i_d.di_nextents)) { |
642 | /* | |
643 | * signal a sync transaction unless | |
644 | * we're truncating an already unlinked | |
645 | * file on a wsync filesystem | |
646 | */ | |
647 | code = xfs_itruncate_finish(&tp, ip, | |
648 | (xfs_fsize_t)vap->va_size, | |
649 | XFS_DATA_FORK, | |
650 | ((ip->i_d.di_nlink != 0 || | |
651 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
652 | ? 1 : 0)); | |
7d4fb40a | 653 | if (code) |
1da177e4 | 654 | goto abort_return; |
7d4fb40a NS |
655 | /* |
656 | * Truncated "down", so we're removing references | |
657 | * to old data here - if we now delay flushing for | |
658 | * a long time, we expose ourselves unduly to the | |
659 | * notorious NULL files problem. So, we mark this | |
660 | * vnode and flush it when the file is closed, and | |
661 | * do not wait the usual (long) time for writeout. | |
662 | */ | |
b3aea4ed | 663 | xfs_iflags_set(ip, XFS_ITRUNCATED); |
1da177e4 LT |
664 | } |
665 | /* | |
666 | * Have to do this even if the file's size doesn't change. | |
667 | */ | |
668 | timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
669 | } | |
670 | ||
671 | /* | |
672 | * Change file access modes. | |
673 | */ | |
674 | if (mask & XFS_AT_MODE) { | |
675 | ip->i_d.di_mode &= S_IFMT; | |
676 | ip->i_d.di_mode |= vap->va_mode & ~S_IFMT; | |
677 | ||
678 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
679 | timeflags |= XFS_ICHGTIME_CHG; | |
680 | } | |
681 | ||
682 | /* | |
683 | * Change file ownership. Must be the owner or privileged. | |
684 | * If the system was configured with the "restricted_chown" | |
685 | * option, the owner is not permitted to give away the file, | |
686 | * and can change the group id only to a group of which he | |
687 | * or she is a member. | |
688 | */ | |
689 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
690 | /* | |
691 | * CAP_FSETID overrides the following restrictions: | |
692 | * | |
693 | * The set-user-ID and set-group-ID bits of a file will be | |
694 | * cleared upon successful return from chown() | |
695 | */ | |
696 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
697 | !capable(CAP_FSETID)) { | |
698 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
699 | } | |
700 | ||
701 | /* | |
702 | * Change the ownerships and register quota modifications | |
703 | * in the transaction. | |
704 | */ | |
705 | if (iuid != uid) { | |
706 | if (XFS_IS_UQUOTA_ON(mp)) { | |
707 | ASSERT(mask & XFS_AT_UID); | |
708 | ASSERT(udqp); | |
709 | olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
710 | &ip->i_udquot, udqp); | |
711 | } | |
712 | ip->i_d.di_uid = uid; | |
713 | } | |
714 | if (igid != gid) { | |
715 | if (XFS_IS_GQUOTA_ON(mp)) { | |
c8ad20ff | 716 | ASSERT(!XFS_IS_PQUOTA_ON(mp)); |
1da177e4 LT |
717 | ASSERT(mask & XFS_AT_GID); |
718 | ASSERT(gdqp); | |
719 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
720 | &ip->i_gdquot, gdqp); | |
721 | } | |
722 | ip->i_d.di_gid = gid; | |
723 | } | |
724 | if (iprojid != projid) { | |
c8ad20ff NS |
725 | if (XFS_IS_PQUOTA_ON(mp)) { |
726 | ASSERT(!XFS_IS_GQUOTA_ON(mp)); | |
727 | ASSERT(mask & XFS_AT_PROJID); | |
728 | ASSERT(gdqp); | |
729 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
730 | &ip->i_gdquot, gdqp); | |
731 | } | |
1da177e4 LT |
732 | ip->i_d.di_projid = projid; |
733 | /* | |
734 | * We may have to rev the inode as well as | |
735 | * the superblock version number since projids didn't | |
736 | * exist before DINODE_VERSION_2 and SB_VERSION_NLINK. | |
737 | */ | |
738 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) | |
739 | xfs_bump_ino_vers2(tp, ip); | |
740 | } | |
741 | ||
742 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
743 | timeflags |= XFS_ICHGTIME_CHG; | |
744 | } | |
745 | ||
746 | ||
747 | /* | |
748 | * Change file access or modified times. | |
749 | */ | |
750 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
751 | if (mask & XFS_AT_ATIME) { | |
752 | ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec; | |
753 | ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec; | |
754 | ip->i_update_core = 1; | |
755 | timeflags &= ~XFS_ICHGTIME_ACC; | |
756 | } | |
757 | if (mask & XFS_AT_MTIME) { | |
758 | ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec; | |
759 | ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec; | |
760 | timeflags &= ~XFS_ICHGTIME_MOD; | |
761 | timeflags |= XFS_ICHGTIME_CHG; | |
762 | } | |
763 | if (tp && (flags & ATTR_UTIME)) | |
764 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
765 | } | |
766 | ||
767 | /* | |
768 | * Change XFS-added attributes. | |
769 | */ | |
770 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
771 | if (mask & XFS_AT_EXTSIZE) { | |
772 | /* | |
773 | * Converting bytes to fs blocks. | |
774 | */ | |
775 | ip->i_d.di_extsize = vap->va_extsize >> | |
776 | mp->m_sb.sb_blocklog; | |
777 | } | |
778 | if (mask & XFS_AT_XFLAGS) { | |
779 | uint di_flags; | |
780 | ||
781 | /* can't set PREALLOC this way, just preserve it */ | |
782 | di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC); | |
783 | if (vap->va_xflags & XFS_XFLAG_IMMUTABLE) | |
784 | di_flags |= XFS_DIFLAG_IMMUTABLE; | |
785 | if (vap->va_xflags & XFS_XFLAG_APPEND) | |
786 | di_flags |= XFS_DIFLAG_APPEND; | |
787 | if (vap->va_xflags & XFS_XFLAG_SYNC) | |
788 | di_flags |= XFS_DIFLAG_SYNC; | |
789 | if (vap->va_xflags & XFS_XFLAG_NOATIME) | |
790 | di_flags |= XFS_DIFLAG_NOATIME; | |
791 | if (vap->va_xflags & XFS_XFLAG_NODUMP) | |
792 | di_flags |= XFS_DIFLAG_NODUMP; | |
365ca83d NS |
793 | if (vap->va_xflags & XFS_XFLAG_PROJINHERIT) |
794 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
795 | if (vap->va_xflags & XFS_XFLAG_NODEFRAG) |
796 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
797 | if (vap->va_xflags & XFS_XFLAG_FILESTREAM) |
798 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
1da177e4 LT |
799 | if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { |
800 | if (vap->va_xflags & XFS_XFLAG_RTINHERIT) | |
801 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
802 | if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS) | |
803 | di_flags |= XFS_DIFLAG_NOSYMLINKS; | |
dd9f438e NS |
804 | if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT) |
805 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
806 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
1da177e4 LT |
807 | if (vap->va_xflags & XFS_XFLAG_REALTIME) { |
808 | di_flags |= XFS_DIFLAG_REALTIME; | |
809 | ip->i_iocore.io_flags |= XFS_IOCORE_RT; | |
810 | } else { | |
811 | ip->i_iocore.io_flags &= ~XFS_IOCORE_RT; | |
812 | } | |
dd9f438e NS |
813 | if (vap->va_xflags & XFS_XFLAG_EXTSIZE) |
814 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1da177e4 LT |
815 | } |
816 | ip->i_d.di_flags = di_flags; | |
817 | } | |
818 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
819 | timeflags |= XFS_ICHGTIME_CHG; | |
820 | } | |
821 | ||
822 | /* | |
823 | * Change file inode change time only if XFS_AT_CTIME set | |
824 | * AND we have been called by a DMI function. | |
825 | */ | |
826 | ||
827 | if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) { | |
828 | ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec; | |
829 | ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec; | |
830 | ip->i_update_core = 1; | |
831 | timeflags &= ~XFS_ICHGTIME_CHG; | |
832 | } | |
833 | ||
834 | /* | |
835 | * Send out timestamp changes that need to be set to the | |
836 | * current time. Not done when called by a DMI function. | |
837 | */ | |
838 | if (timeflags && !(flags & ATTR_DMI)) | |
839 | xfs_ichgtime(ip, timeflags); | |
840 | ||
841 | XFS_STATS_INC(xs_ig_attrchg); | |
842 | ||
843 | /* | |
844 | * If this is a synchronous mount, make sure that the | |
845 | * transaction goes to disk before returning to the user. | |
846 | * This is slightly sub-optimal in that truncates require | |
c41564b5 | 847 | * two sync transactions instead of one for wsync filesystems. |
1da177e4 LT |
848 | * One for the truncate and one for the timestamps since we |
849 | * don't want to change the timestamps unless we're sure the | |
850 | * truncate worked. Truncates are less than 1% of the laddis | |
851 | * mix so this probably isn't worth the trouble to optimize. | |
852 | */ | |
853 | code = 0; | |
854 | if (tp) { | |
855 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
856 | xfs_trans_set_sync(tp); | |
857 | ||
1c72bf90 | 858 | code = xfs_trans_commit(tp, commit_flags); |
1da177e4 LT |
859 | } |
860 | ||
861 | /* | |
862 | * If the (regular) file's mandatory locking mode changed, then | |
863 | * notify the vnode. We do this under the inode lock to prevent | |
864 | * racing calls to vop_vnode_change. | |
865 | */ | |
866 | mandlock_after = MANDLOCK(vp, ip->i_d.di_mode); | |
1da177e4 LT |
867 | |
868 | xfs_iunlock(ip, lock_flags); | |
869 | ||
870 | /* | |
871 | * Release any dquot(s) the inode had kept before chown. | |
872 | */ | |
873 | XFS_QM_DQRELE(mp, olddquot1); | |
874 | XFS_QM_DQRELE(mp, olddquot2); | |
875 | XFS_QM_DQRELE(mp, udqp); | |
876 | XFS_QM_DQRELE(mp, gdqp); | |
877 | ||
878 | if (code) { | |
879 | return code; | |
880 | } | |
881 | ||
eb9df39d | 882 | if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) && |
1da177e4 LT |
883 | !(flags & ATTR_DMI)) { |
884 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL, | |
885 | NULL, DM_RIGHT_NULL, NULL, NULL, | |
886 | 0, 0, AT_DELAY_FLAG(flags)); | |
887 | } | |
888 | return 0; | |
889 | ||
890 | abort_return: | |
891 | commit_flags |= XFS_TRANS_ABORT; | |
892 | /* FALLTHROUGH */ | |
893 | error_return: | |
894 | XFS_QM_DQRELE(mp, udqp); | |
895 | XFS_QM_DQRELE(mp, gdqp); | |
896 | if (tp) { | |
897 | xfs_trans_cancel(tp, commit_flags); | |
898 | } | |
899 | if (lock_flags != 0) { | |
900 | xfs_iunlock(ip, lock_flags); | |
901 | } | |
902 | return code; | |
903 | } | |
904 | ||
905 | ||
906 | /* | |
907 | * xfs_access | |
908 | * Null conversion from vnode mode bits to inode mode bits, as in efs. | |
909 | */ | |
993386c1 | 910 | int |
1da177e4 | 911 | xfs_access( |
993386c1 | 912 | xfs_inode_t *ip, |
1da177e4 LT |
913 | int mode, |
914 | cred_t *credp) | |
915 | { | |
1da177e4 LT |
916 | int error; |
917 | ||
1543d79c | 918 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 919 | |
1da177e4 LT |
920 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
921 | error = xfs_iaccess(ip, mode, credp); | |
922 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
923 | return error; | |
924 | } | |
925 | ||
926 | ||
7d4fb40a NS |
927 | /* |
928 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
929 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
930 | * bmapi. | |
931 | */ | |
932 | #define SYMLINK_MAPS 2 | |
933 | ||
804c83c3 CH |
934 | STATIC int |
935 | xfs_readlink_bmap( | |
936 | xfs_inode_t *ip, | |
937 | char *link) | |
938 | { | |
939 | xfs_mount_t *mp = ip->i_mount; | |
940 | int pathlen = ip->i_d.di_size; | |
941 | int nmaps = SYMLINK_MAPS; | |
942 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
943 | xfs_daddr_t d; | |
944 | int byte_cnt; | |
945 | int n; | |
946 | xfs_buf_t *bp; | |
947 | int error = 0; | |
948 | ||
949 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0, | |
950 | mval, &nmaps, NULL, NULL); | |
951 | if (error) | |
952 | goto out; | |
953 | ||
954 | for (n = 0; n < nmaps; n++) { | |
955 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
956 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
957 | ||
958 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); | |
959 | error = XFS_BUF_GETERROR(bp); | |
960 | if (error) { | |
961 | xfs_ioerror_alert("xfs_readlink", | |
962 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
963 | xfs_buf_relse(bp); | |
964 | goto out; | |
965 | } | |
966 | if (pathlen < byte_cnt) | |
967 | byte_cnt = pathlen; | |
968 | pathlen -= byte_cnt; | |
969 | ||
970 | memcpy(link, XFS_BUF_PTR(bp), byte_cnt); | |
971 | xfs_buf_relse(bp); | |
972 | } | |
973 | ||
974 | link[ip->i_d.di_size] = '\0'; | |
975 | error = 0; | |
976 | ||
977 | out: | |
978 | return error; | |
979 | } | |
980 | ||
993386c1 | 981 | int |
1da177e4 | 982 | xfs_readlink( |
993386c1 | 983 | xfs_inode_t *ip, |
804c83c3 | 984 | char *link) |
1da177e4 | 985 | { |
804c83c3 | 986 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 987 | int pathlen; |
1da177e4 | 988 | int error = 0; |
1da177e4 | 989 | |
1543d79c | 990 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
991 | |
992 | if (XFS_FORCED_SHUTDOWN(mp)) | |
993 | return XFS_ERROR(EIO); | |
994 | ||
995 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
996 | ||
997 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
804c83c3 | 998 | ASSERT(ip->i_d.di_size <= MAXPATHLEN); |
1da177e4 | 999 | |
804c83c3 CH |
1000 | pathlen = ip->i_d.di_size; |
1001 | if (!pathlen) | |
1002 | goto out; | |
1da177e4 LT |
1003 | |
1004 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
804c83c3 CH |
1005 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
1006 | link[pathlen] = '\0'; | |
1007 | } else { | |
1008 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
1009 | } |
1010 | ||
804c83c3 | 1011 | out: |
1da177e4 | 1012 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
1013 | return error; |
1014 | } | |
1015 | ||
1da177e4 LT |
1016 | /* |
1017 | * xfs_fsync | |
1018 | * | |
1019 | * This is called to sync the inode and its data out to disk. | |
1020 | * We need to hold the I/O lock while flushing the data, and | |
1021 | * the inode lock while flushing the inode. The inode lock CANNOT | |
1022 | * be held while flushing the data, so acquire after we're done | |
1023 | * with that. | |
1024 | */ | |
993386c1 | 1025 | int |
1da177e4 | 1026 | xfs_fsync( |
993386c1 | 1027 | xfs_inode_t *ip, |
1da177e4 | 1028 | int flag, |
1da177e4 LT |
1029 | xfs_off_t start, |
1030 | xfs_off_t stop) | |
1031 | { | |
1da177e4 LT |
1032 | xfs_trans_t *tp; |
1033 | int error; | |
f538d4da | 1034 | int log_flushed = 0, changed = 1; |
1da177e4 | 1035 | |
1543d79c | 1036 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
1037 | |
1038 | ASSERT(start >= 0 && stop >= -1); | |
1039 | ||
1040 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
1041 | return XFS_ERROR(EIO); | |
1042 | ||
776a75fa LM |
1043 | if (flag & FSYNC_DATA) |
1044 | filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping); | |
1045 | ||
1da177e4 LT |
1046 | /* |
1047 | * We always need to make sure that the required inode state | |
1048 | * is safe on disk. The vnode might be clean but because | |
1049 | * of committed transactions that haven't hit the disk yet. | |
1050 | * Likewise, there could be unflushed non-transactional | |
1051 | * changes to the inode core that have to go to disk. | |
1052 | * | |
1053 | * The following code depends on one assumption: that | |
1054 | * any transaction that changes an inode logs the core | |
1055 | * because it has to change some field in the inode core | |
1056 | * (typically nextents or nblocks). That assumption | |
1057 | * implies that any transactions against an inode will | |
1058 | * catch any non-transactional updates. If inode-altering | |
1059 | * transactions exist that violate this assumption, the | |
1060 | * code breaks. Right now, it figures that if the involved | |
1061 | * update_* field is clear and the inode is unpinned, the | |
1062 | * inode is clean. Either it's been flushed or it's been | |
1063 | * committed and the commit has hit the disk unpinning the inode. | |
1064 | * (Note that xfs_inode_item_format() called at commit clears | |
1065 | * the update_* fields.) | |
1066 | */ | |
1067 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1068 | ||
1069 | /* If we are flushing data then we care about update_size | |
1070 | * being set, otherwise we care about update_core | |
1071 | */ | |
1072 | if ((flag & FSYNC_DATA) ? | |
1073 | (ip->i_update_size == 0) : | |
1074 | (ip->i_update_core == 0)) { | |
1075 | /* | |
1076 | * Timestamps/size haven't changed since last inode | |
1077 | * flush or inode transaction commit. That means | |
1078 | * either nothing got written or a transaction | |
1079 | * committed which caught the updates. If the | |
1080 | * latter happened and the transaction hasn't | |
1081 | * hit the disk yet, the inode will be still | |
1082 | * be pinned. If it is, force the log. | |
1083 | */ | |
1084 | ||
1085 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1086 | ||
1087 | if (xfs_ipincount(ip)) { | |
f538d4da | 1088 | _xfs_log_force(ip->i_mount, (xfs_lsn_t)0, |
1da177e4 LT |
1089 | XFS_LOG_FORCE | |
1090 | ((flag & FSYNC_WAIT) | |
f538d4da CH |
1091 | ? XFS_LOG_SYNC : 0), |
1092 | &log_flushed); | |
1093 | } else { | |
1094 | /* | |
1095 | * If the inode is not pinned and nothing | |
1096 | * has changed we don't need to flush the | |
1097 | * cache. | |
1098 | */ | |
1099 | changed = 0; | |
1da177e4 LT |
1100 | } |
1101 | error = 0; | |
1102 | } else { | |
1103 | /* | |
1104 | * Kick off a transaction to log the inode | |
1105 | * core to get the updates. Make it | |
1106 | * sync if FSYNC_WAIT is passed in (which | |
1107 | * is done by everybody but specfs). The | |
1108 | * sync transaction will also force the log. | |
1109 | */ | |
1110 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1111 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
1112 | if ((error = xfs_trans_reserve(tp, 0, | |
1113 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), | |
1114 | 0, 0, 0))) { | |
1115 | xfs_trans_cancel(tp, 0); | |
1116 | return error; | |
1117 | } | |
1118 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1119 | ||
1120 | /* | |
1121 | * Note - it's possible that we might have pushed | |
1122 | * ourselves out of the way during trans_reserve | |
1123 | * which would flush the inode. But there's no | |
1124 | * guarantee that the inode buffer has actually | |
1125 | * gone out yet (it's delwri). Plus the buffer | |
1126 | * could be pinned anyway if it's part of an | |
1127 | * inode in another recent transaction. So we | |
1128 | * play it safe and fire off the transaction anyway. | |
1129 | */ | |
1130 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1131 | xfs_trans_ihold(tp, ip); | |
1132 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1133 | if (flag & FSYNC_WAIT) | |
1134 | xfs_trans_set_sync(tp); | |
1c72bf90 | 1135 | error = _xfs_trans_commit(tp, 0, &log_flushed); |
1da177e4 LT |
1136 | |
1137 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1138 | } | |
f538d4da CH |
1139 | |
1140 | if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) { | |
1141 | /* | |
1142 | * If the log write didn't issue an ordered tag we need | |
1143 | * to flush the disk cache for the data device now. | |
1144 | */ | |
1145 | if (!log_flushed) | |
1146 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
1147 | ||
1148 | /* | |
1149 | * If this inode is on the RT dev we need to flush that | |
c41564b5 | 1150 | * cache as well. |
f538d4da CH |
1151 | */ |
1152 | if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) | |
1153 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); | |
1154 | } | |
1155 | ||
1da177e4 LT |
1156 | return error; |
1157 | } | |
1158 | ||
1159 | /* | |
92dfe8d2 DC |
1160 | * This is called by xfs_inactive to free any blocks beyond eof |
1161 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
1162 | * punching a hole to EOF. | |
1da177e4 | 1163 | */ |
92dfe8d2 DC |
1164 | int |
1165 | xfs_free_eofblocks( | |
1da177e4 | 1166 | xfs_mount_t *mp, |
92dfe8d2 DC |
1167 | xfs_inode_t *ip, |
1168 | int flags) | |
1da177e4 LT |
1169 | { |
1170 | xfs_trans_t *tp; | |
1171 | int error; | |
1172 | xfs_fileoff_t end_fsb; | |
1173 | xfs_fileoff_t last_fsb; | |
1174 | xfs_filblks_t map_len; | |
1175 | int nimaps; | |
1176 | xfs_bmbt_irec_t imap; | |
92dfe8d2 | 1177 | int use_iolock = (flags & XFS_FREE_EOF_LOCK); |
1da177e4 LT |
1178 | |
1179 | /* | |
1180 | * Figure out if there are any blocks beyond the end | |
1181 | * of the file. If not, then there is nothing to do. | |
1182 | */ | |
ba87ea69 | 1183 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size)); |
1da177e4 LT |
1184 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); |
1185 | map_len = last_fsb - end_fsb; | |
1186 | if (map_len <= 0) | |
014c2544 | 1187 | return 0; |
1da177e4 LT |
1188 | |
1189 | nimaps = 1; | |
1190 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3e57ecf6 OW |
1191 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0, |
1192 | NULL, 0, &imap, &nimaps, NULL, NULL); | |
1da177e4 LT |
1193 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1194 | ||
1195 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
1196 | (imap.br_startblock != HOLESTARTBLOCK || |
1197 | ip->i_delayed_blks)) { | |
1da177e4 LT |
1198 | /* |
1199 | * Attach the dquots to the inode up front. | |
1200 | */ | |
1201 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1202 | return error; |
1da177e4 LT |
1203 | |
1204 | /* | |
1205 | * There are blocks after the end of file. | |
1206 | * Free them up now by truncating the file to | |
1207 | * its current size. | |
1208 | */ | |
1209 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1210 | ||
1211 | /* | |
1212 | * Do the xfs_itruncate_start() call before | |
1213 | * reserving any log space because | |
1214 | * itruncate_start will call into the buffer | |
1215 | * cache and we can't | |
1216 | * do that within a transaction. | |
1217 | */ | |
92dfe8d2 DC |
1218 | if (use_iolock) |
1219 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 | 1220 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, |
ba87ea69 | 1221 | ip->i_size); |
d3cf2094 | 1222 | if (error) { |
87ae3c24 | 1223 | xfs_trans_cancel(tp, 0); |
92dfe8d2 DC |
1224 | if (use_iolock) |
1225 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 LM |
1226 | return error; |
1227 | } | |
1da177e4 LT |
1228 | |
1229 | error = xfs_trans_reserve(tp, 0, | |
1230 | XFS_ITRUNCATE_LOG_RES(mp), | |
1231 | 0, XFS_TRANS_PERM_LOG_RES, | |
1232 | XFS_ITRUNCATE_LOG_COUNT); | |
1233 | if (error) { | |
1234 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1235 | xfs_trans_cancel(tp, 0); | |
1236 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1237 | return error; |
1da177e4 LT |
1238 | } |
1239 | ||
1240 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1241 | xfs_trans_ijoin(tp, ip, | |
1242 | XFS_IOLOCK_EXCL | | |
1243 | XFS_ILOCK_EXCL); | |
1244 | xfs_trans_ihold(tp, ip); | |
1245 | ||
1246 | error = xfs_itruncate_finish(&tp, ip, | |
ba87ea69 | 1247 | ip->i_size, |
1da177e4 LT |
1248 | XFS_DATA_FORK, |
1249 | 0); | |
1250 | /* | |
1251 | * If we get an error at this point we | |
1252 | * simply don't bother truncating the file. | |
1253 | */ | |
1254 | if (error) { | |
1255 | xfs_trans_cancel(tp, | |
1256 | (XFS_TRANS_RELEASE_LOG_RES | | |
1257 | XFS_TRANS_ABORT)); | |
1258 | } else { | |
1259 | error = xfs_trans_commit(tp, | |
1c72bf90 | 1260 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 1261 | } |
92dfe8d2 DC |
1262 | xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL) |
1263 | : XFS_ILOCK_EXCL)); | |
1da177e4 | 1264 | } |
014c2544 | 1265 | return error; |
1da177e4 LT |
1266 | } |
1267 | ||
1268 | /* | |
1269 | * Free a symlink that has blocks associated with it. | |
1270 | */ | |
1271 | STATIC int | |
1272 | xfs_inactive_symlink_rmt( | |
1273 | xfs_inode_t *ip, | |
1274 | xfs_trans_t **tpp) | |
1275 | { | |
1276 | xfs_buf_t *bp; | |
1277 | int committed; | |
1278 | int done; | |
1279 | int error; | |
1280 | xfs_fsblock_t first_block; | |
1281 | xfs_bmap_free_t free_list; | |
1282 | int i; | |
1283 | xfs_mount_t *mp; | |
1284 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1285 | int nmaps; | |
1286 | xfs_trans_t *ntp; | |
1287 | int size; | |
1288 | xfs_trans_t *tp; | |
1289 | ||
1290 | tp = *tpp; | |
1291 | mp = ip->i_mount; | |
1292 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
1293 | /* | |
1294 | * We're freeing a symlink that has some | |
1295 | * blocks allocated to it. Free the | |
1296 | * blocks here. We know that we've got | |
1297 | * either 1 or 2 extents and that we can | |
1298 | * free them all in one bunmapi call. | |
1299 | */ | |
1300 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
1301 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1302 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1303 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1304 | xfs_trans_cancel(tp, 0); | |
1305 | *tpp = NULL; | |
1306 | return error; | |
1307 | } | |
1308 | /* | |
1309 | * Lock the inode, fix the size, and join it to the transaction. | |
1310 | * Hold it so in the normal path, we still have it locked for | |
1311 | * the second transaction. In the error paths we need it | |
1312 | * held so the cancel won't rele it, see below. | |
1313 | */ | |
1314 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1315 | size = (int)ip->i_d.di_size; | |
1316 | ip->i_d.di_size = 0; | |
1317 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1318 | xfs_trans_ihold(tp, ip); | |
1319 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1320 | /* | |
1321 | * Find the block(s) so we can inval and unmap them. | |
1322 | */ | |
1323 | done = 0; | |
1324 | XFS_BMAP_INIT(&free_list, &first_block); | |
e8c96f8c | 1325 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
1326 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
1327 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 1328 | &free_list, NULL))) |
1da177e4 LT |
1329 | goto error0; |
1330 | /* | |
1331 | * Invalidate the block(s). | |
1332 | */ | |
1333 | for (i = 0; i < nmaps; i++) { | |
1334 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
1335 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
1336 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
1337 | xfs_trans_binval(tp, bp); | |
1338 | } | |
1339 | /* | |
1340 | * Unmap the dead block(s) to the free_list. | |
1341 | */ | |
1342 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 1343 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
1344 | goto error1; |
1345 | ASSERT(done); | |
1346 | /* | |
1347 | * Commit the first transaction. This logs the EFI and the inode. | |
1348 | */ | |
f7c99b6f | 1349 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
1350 | goto error1; |
1351 | /* | |
1352 | * The transaction must have been committed, since there were | |
1353 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
1354 | * The new tp has the extent freeing and EFDs. | |
1355 | */ | |
1356 | ASSERT(committed); | |
1357 | /* | |
1358 | * The first xact was committed, so add the inode to the new one. | |
1359 | * Mark it dirty so it will be logged and moved forward in the log as | |
1360 | * part of every commit. | |
1361 | */ | |
1362 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1363 | xfs_trans_ihold(tp, ip); | |
1364 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1365 | /* | |
1366 | * Get a new, empty transaction to return to our caller. | |
1367 | */ | |
1368 | ntp = xfs_trans_dup(tp); | |
1369 | /* | |
c41564b5 | 1370 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
1371 | * If we get an error on the commit here or on the reserve below, |
1372 | * we need to unlock the inode since the new transaction doesn't | |
1373 | * have the inode attached. | |
1374 | */ | |
1c72bf90 | 1375 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1376 | tp = ntp; |
1377 | if (error) { | |
1378 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1379 | goto error0; | |
1380 | } | |
1381 | /* | |
1382 | * Remove the memory for extent descriptions (just bookkeeping). | |
1383 | */ | |
1384 | if (ip->i_df.if_bytes) | |
1385 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
1386 | ASSERT(ip->i_df.if_bytes == 0); | |
1387 | /* | |
1388 | * Put an itruncate log reservation in the new transaction | |
1389 | * for our caller. | |
1390 | */ | |
1391 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1392 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1393 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1394 | goto error0; | |
1395 | } | |
1396 | /* | |
1397 | * Return with the inode locked but not joined to the transaction. | |
1398 | */ | |
1399 | *tpp = tp; | |
1400 | return 0; | |
1401 | ||
1402 | error1: | |
1403 | xfs_bmap_cancel(&free_list); | |
1404 | error0: | |
1405 | /* | |
1406 | * Have to come here with the inode locked and either | |
1407 | * (held and in the transaction) or (not in the transaction). | |
1408 | * If the inode isn't held then cancel would iput it, but | |
1409 | * that's wrong since this is inactive and the vnode ref | |
1410 | * count is 0 already. | |
1411 | * Cancel won't do anything to the inode if held, but it still | |
1412 | * needs to be locked until the cancel is done, if it was | |
1413 | * joined to the transaction. | |
1414 | */ | |
1415 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1416 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1417 | *tpp = NULL; | |
1418 | return error; | |
1419 | ||
1420 | } | |
1421 | ||
1422 | STATIC int | |
1423 | xfs_inactive_symlink_local( | |
1424 | xfs_inode_t *ip, | |
1425 | xfs_trans_t **tpp) | |
1426 | { | |
1427 | int error; | |
1428 | ||
1429 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
1430 | /* | |
1431 | * We're freeing a symlink which fit into | |
1432 | * the inode. Just free the memory used | |
1433 | * to hold the old symlink. | |
1434 | */ | |
1435 | error = xfs_trans_reserve(*tpp, 0, | |
1436 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
1437 | 0, XFS_TRANS_PERM_LOG_RES, | |
1438 | XFS_ITRUNCATE_LOG_COUNT); | |
1439 | ||
1440 | if (error) { | |
1441 | xfs_trans_cancel(*tpp, 0); | |
1442 | *tpp = NULL; | |
014c2544 | 1443 | return error; |
1da177e4 LT |
1444 | } |
1445 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1446 | ||
1447 | /* | |
1448 | * Zero length symlinks _can_ exist. | |
1449 | */ | |
1450 | if (ip->i_df.if_bytes > 0) { | |
1451 | xfs_idata_realloc(ip, | |
1452 | -(ip->i_df.if_bytes), | |
1453 | XFS_DATA_FORK); | |
1454 | ASSERT(ip->i_df.if_bytes == 0); | |
1455 | } | |
014c2544 | 1456 | return 0; |
1da177e4 LT |
1457 | } |
1458 | ||
1da177e4 LT |
1459 | STATIC int |
1460 | xfs_inactive_attrs( | |
1461 | xfs_inode_t *ip, | |
1462 | xfs_trans_t **tpp) | |
1463 | { | |
1464 | xfs_trans_t *tp; | |
1465 | int error; | |
1466 | xfs_mount_t *mp; | |
1467 | ||
1468 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); | |
1469 | tp = *tpp; | |
1470 | mp = ip->i_mount; | |
1471 | ASSERT(ip->i_d.di_forkoff != 0); | |
1c72bf90 | 1472 | xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1473 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1474 | ||
1475 | error = xfs_attr_inactive(ip); | |
1476 | if (error) { | |
1477 | *tpp = NULL; | |
1478 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1479 | return error; /* goto out */ |
1da177e4 LT |
1480 | } |
1481 | ||
1482 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1483 | error = xfs_trans_reserve(tp, 0, | |
1484 | XFS_IFREE_LOG_RES(mp), | |
1485 | 0, XFS_TRANS_PERM_LOG_RES, | |
1486 | XFS_INACTIVE_LOG_COUNT); | |
1487 | if (error) { | |
1488 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1489 | xfs_trans_cancel(tp, 0); | |
1490 | *tpp = NULL; | |
1491 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1492 | return error; |
1da177e4 LT |
1493 | } |
1494 | ||
1495 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1496 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1497 | xfs_trans_ihold(tp, ip); | |
1498 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1499 | ||
1500 | ASSERT(ip->i_d.di_anextents == 0); | |
1501 | ||
1502 | *tpp = tp; | |
014c2544 | 1503 | return 0; |
1da177e4 LT |
1504 | } |
1505 | ||
993386c1 | 1506 | int |
1da177e4 | 1507 | xfs_release( |
993386c1 | 1508 | xfs_inode_t *ip) |
1da177e4 | 1509 | { |
993386c1 CH |
1510 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1511 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
1512 | int error; |
1513 | ||
7d4fb40a | 1514 | if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) |
1da177e4 | 1515 | return 0; |
1da177e4 LT |
1516 | |
1517 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1518 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1519 | return 0; |
1520 | ||
2a82b8be | 1521 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
1522 | int truncated; |
1523 | ||
2a82b8be DC |
1524 | /* |
1525 | * If we are using filestreams, and we have an unlinked | |
1526 | * file that we are processing the last close on, then nothing | |
1527 | * will be able to reopen and write to this file. Purge this | |
1528 | * inode from the filestreams cache so that it doesn't delay | |
1529 | * teardown of the inode. | |
1530 | */ | |
1531 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1532 | xfs_filestream_deassociate(ip); | |
1533 | ||
fbf3ce8d CH |
1534 | /* |
1535 | * If we previously truncated this file and removed old data | |
1536 | * in the process, we want to initiate "early" writeout on | |
1537 | * the last close. This is an attempt to combat the notorious | |
1538 | * NULL files problem which is particularly noticable from a | |
1539 | * truncate down, buffered (re-)write (delalloc), followed by | |
1540 | * a crash. What we are effectively doing here is | |
1541 | * significantly reducing the time window where we'd otherwise | |
1542 | * be exposed to that problem. | |
1543 | */ | |
09262b43 | 1544 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
b3aea4ed | 1545 | if (truncated && VN_DIRTY(vp) && ip->i_delayed_blks > 0) |
739bfb2a | 1546 | xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE); |
fbf3ce8d CH |
1547 | } |
1548 | ||
1da177e4 LT |
1549 | #ifdef HAVE_REFCACHE |
1550 | /* If we are in the NFS reference cache then don't do this now */ | |
1551 | if (ip->i_refcache) | |
1552 | return 0; | |
1553 | #endif | |
1554 | ||
1da177e4 LT |
1555 | if (ip->i_d.di_nlink != 0) { |
1556 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1557 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1558 | ip->i_delayed_blks > 0)) && |
1da177e4 | 1559 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
1560 | (!(ip->i_d.di_flags & |
1561 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
92dfe8d2 DC |
1562 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1563 | if (error) | |
014c2544 | 1564 | return error; |
1da177e4 | 1565 | /* Update linux inode block count after free above */ |
ec86dc02 | 1566 | vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp, |
1da177e4 LT |
1567 | ip->i_d.di_nblocks + ip->i_delayed_blks); |
1568 | } | |
1569 | } | |
1570 | ||
1571 | return 0; | |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * xfs_inactive | |
1576 | * | |
1577 | * This is called when the vnode reference count for the vnode | |
1578 | * goes to zero. If the file has been unlinked, then it must | |
1579 | * now be truncated. Also, we clear all of the read-ahead state | |
1580 | * kept for the inode here since the file is now closed. | |
1581 | */ | |
993386c1 | 1582 | int |
1da177e4 | 1583 | xfs_inactive( |
993386c1 | 1584 | xfs_inode_t *ip) |
1da177e4 | 1585 | { |
993386c1 | 1586 | bhv_vnode_t *vp = XFS_ITOV(ip); |
67fcaa73 | 1587 | xfs_bmap_free_t free_list; |
1da177e4 LT |
1588 | xfs_fsblock_t first_block; |
1589 | int committed; | |
1590 | xfs_trans_t *tp; | |
1591 | xfs_mount_t *mp; | |
1592 | int error; | |
1593 | int truncate; | |
1594 | ||
1543d79c | 1595 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 1596 | |
1da177e4 LT |
1597 | /* |
1598 | * If the inode is already free, then there can be nothing | |
1599 | * to clean up here. | |
1600 | */ | |
1601 | if (ip->i_d.di_mode == 0 || VN_BAD(vp)) { | |
1602 | ASSERT(ip->i_df.if_real_bytes == 0); | |
1603 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1604 | return VN_INACTIVE_CACHE; | |
1605 | } | |
1606 | ||
1607 | /* | |
1608 | * Only do a truncate if it's a regular file with | |
1609 | * some actual space in it. It's OK to look at the | |
1610 | * inode's fields without the lock because we're the | |
1611 | * only one with a reference to the inode. | |
1612 | */ | |
1613 | truncate = ((ip->i_d.di_nlink == 0) && | |
ba87ea69 LM |
1614 | ((ip->i_d.di_size != 0) || (ip->i_size != 0) || |
1615 | (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1616 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1617 | ||
1618 | mp = ip->i_mount; | |
1619 | ||
eb9df39d | 1620 | if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) { |
1da177e4 LT |
1621 | (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL); |
1622 | } | |
1623 | ||
1624 | error = 0; | |
1625 | ||
1626 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1627 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1628 | goto out; |
1629 | ||
1630 | if (ip->i_d.di_nlink != 0) { | |
1631 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1632 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1633 | ip->i_delayed_blks > 0)) && |
dd9f438e NS |
1634 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1635 | (!(ip->i_d.di_flags & | |
1636 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1637 | (ip->i_delayed_blks != 0)))) { | |
92dfe8d2 DC |
1638 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1639 | if (error) | |
014c2544 | 1640 | return VN_INACTIVE_CACHE; |
1da177e4 | 1641 | /* Update linux inode block count after free above */ |
ec86dc02 | 1642 | vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp, |
1da177e4 LT |
1643 | ip->i_d.di_nblocks + ip->i_delayed_blks); |
1644 | } | |
1645 | goto out; | |
1646 | } | |
1647 | ||
1648 | ASSERT(ip->i_d.di_nlink == 0); | |
1649 | ||
1650 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1651 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1652 | |
1653 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1654 | if (truncate) { | |
1655 | /* | |
1656 | * Do the xfs_itruncate_start() call before | |
1657 | * reserving any log space because itruncate_start | |
1658 | * will call into the buffer cache and we can't | |
1659 | * do that within a transaction. | |
1660 | */ | |
1661 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1662 | ||
d3cf2094 LM |
1663 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); |
1664 | if (error) { | |
87ae3c24 | 1665 | xfs_trans_cancel(tp, 0); |
d3cf2094 LM |
1666 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1667 | return VN_INACTIVE_CACHE; | |
1668 | } | |
1da177e4 LT |
1669 | |
1670 | error = xfs_trans_reserve(tp, 0, | |
1671 | XFS_ITRUNCATE_LOG_RES(mp), | |
1672 | 0, XFS_TRANS_PERM_LOG_RES, | |
1673 | XFS_ITRUNCATE_LOG_COUNT); | |
1674 | if (error) { | |
1675 | /* Don't call itruncate_cleanup */ | |
1676 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1677 | xfs_trans_cancel(tp, 0); | |
1678 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1679 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1680 | } |
1681 | ||
1682 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1683 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1684 | xfs_trans_ihold(tp, ip); | |
1685 | ||
1686 | /* | |
1687 | * normally, we have to run xfs_itruncate_finish sync. | |
1688 | * But if filesystem is wsync and we're in the inactive | |
1689 | * path, then we know that nlink == 0, and that the | |
1690 | * xaction that made nlink == 0 is permanently committed | |
1691 | * since xfs_remove runs as a synchronous transaction. | |
1692 | */ | |
1693 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1694 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1695 | ||
1696 | if (error) { | |
1697 | xfs_trans_cancel(tp, | |
1698 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1699 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1700 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1701 | } |
1702 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1703 | ||
1704 | /* | |
1705 | * If we get an error while cleaning up a | |
1706 | * symlink we bail out. | |
1707 | */ | |
1708 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1709 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1710 | xfs_inactive_symlink_local(ip, &tp); | |
1711 | ||
1712 | if (error) { | |
1713 | ASSERT(tp == NULL); | |
014c2544 | 1714 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1715 | } |
1716 | ||
1717 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1718 | xfs_trans_ihold(tp, ip); | |
1719 | } else { | |
1720 | error = xfs_trans_reserve(tp, 0, | |
1721 | XFS_IFREE_LOG_RES(mp), | |
1722 | 0, XFS_TRANS_PERM_LOG_RES, | |
1723 | XFS_INACTIVE_LOG_COUNT); | |
1724 | if (error) { | |
1725 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1726 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1727 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1728 | } |
1729 | ||
1730 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1731 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1732 | xfs_trans_ihold(tp, ip); | |
1733 | } | |
1734 | ||
1735 | /* | |
1736 | * If there are attributes associated with the file | |
1737 | * then blow them away now. The code calls a routine | |
1738 | * that recursively deconstructs the attribute fork. | |
1739 | * We need to just commit the current transaction | |
1740 | * because we can't use it for xfs_attr_inactive(). | |
1741 | */ | |
1742 | if (ip->i_d.di_anextents > 0) { | |
1743 | error = xfs_inactive_attrs(ip, &tp); | |
1744 | /* | |
1745 | * If we got an error, the transaction is already | |
1746 | * cancelled, and the inode is unlocked. Just get out. | |
1747 | */ | |
1748 | if (error) | |
014c2544 | 1749 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1750 | } else if (ip->i_afp) { |
1751 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1752 | } | |
1753 | ||
1754 | /* | |
1755 | * Free the inode. | |
1756 | */ | |
1757 | XFS_BMAP_INIT(&free_list, &first_block); | |
1758 | error = xfs_ifree(tp, ip, &free_list); | |
1759 | if (error) { | |
1760 | /* | |
1761 | * If we fail to free the inode, shut down. The cancel | |
1762 | * might do that, we need to make sure. Otherwise the | |
1763 | * inode might be lost for a long time or forever. | |
1764 | */ | |
1765 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1766 | cmn_err(CE_NOTE, | |
1767 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1768 | error, mp->m_fsname); | |
7d04a335 | 1769 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1770 | } |
1771 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1772 | } else { | |
1773 | /* | |
1774 | * Credit the quota account(s). The inode is gone. | |
1775 | */ | |
1776 | XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1777 | ||
1778 | /* | |
1779 | * Just ignore errors at this point. There is | |
1780 | * nothing we can do except to try to keep going. | |
1781 | */ | |
f7c99b6f | 1782 | (void) xfs_bmap_finish(&tp, &free_list, &committed); |
1c72bf90 | 1783 | (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1784 | } |
1785 | /* | |
1786 | * Release the dquots held by inode, if any. | |
1787 | */ | |
1788 | XFS_QM_DQDETACH(mp, ip); | |
1789 | ||
1790 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1791 | ||
1792 | out: | |
1793 | return VN_INACTIVE_CACHE; | |
1794 | } | |
1795 | ||
1796 | ||
993386c1 | 1797 | int |
1da177e4 | 1798 | xfs_lookup( |
993386c1 | 1799 | xfs_inode_t *dp, |
8285fb58 | 1800 | bhv_vname_t *dentry, |
993386c1 | 1801 | bhv_vnode_t **vpp) |
1da177e4 | 1802 | { |
993386c1 | 1803 | xfs_inode_t *ip; |
1da177e4 LT |
1804 | xfs_ino_t e_inum; |
1805 | int error; | |
1806 | uint lock_mode; | |
1da177e4 | 1807 | |
1543d79c | 1808 | vn_trace_entry(dp, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
1809 | |
1810 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1811 | return XFS_ERROR(EIO); | |
1812 | ||
1813 | lock_mode = xfs_ilock_map_shared(dp); | |
993386c1 | 1814 | error = xfs_dir_lookup_int(dp, lock_mode, dentry, &e_inum, &ip); |
1da177e4 LT |
1815 | if (!error) { |
1816 | *vpp = XFS_ITOV(ip); | |
1817 | ITRACE(ip); | |
1818 | } | |
1819 | xfs_iunlock_map_shared(dp, lock_mode); | |
1820 | return error; | |
1821 | } | |
1822 | ||
993386c1 | 1823 | int |
1da177e4 | 1824 | xfs_create( |
993386c1 | 1825 | xfs_inode_t *dp, |
8285fb58 NS |
1826 | bhv_vname_t *dentry, |
1827 | bhv_vattr_t *vap, | |
67fcaa73 | 1828 | bhv_vnode_t **vpp, |
1da177e4 LT |
1829 | cred_t *credp) |
1830 | { | |
1831 | char *name = VNAME(dentry); | |
993386c1 CH |
1832 | xfs_mount_t *mp = dp->i_mount; |
1833 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); | |
1834 | xfs_inode_t *ip; | |
67fcaa73 | 1835 | bhv_vnode_t *vp = NULL; |
1da177e4 | 1836 | xfs_trans_t *tp; |
1da177e4 LT |
1837 | xfs_dev_t rdev; |
1838 | int error; | |
1839 | xfs_bmap_free_t free_list; | |
1840 | xfs_fsblock_t first_block; | |
993386c1 | 1841 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1842 | int dm_event_sent = 0; |
1843 | uint cancel_flags; | |
1844 | int committed; | |
1845 | xfs_prid_t prid; | |
1846 | struct xfs_dquot *udqp, *gdqp; | |
1847 | uint resblks; | |
1848 | int dm_di_mode; | |
1849 | int namelen; | |
1850 | ||
1851 | ASSERT(!*vpp); | |
1543d79c | 1852 | vn_trace_entry(dp, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 1853 | |
0432dab2 | 1854 | dm_di_mode = vap->va_mode; |
1da177e4 LT |
1855 | namelen = VNAMELEN(dentry); |
1856 | ||
eb9df39d | 1857 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 LT |
1858 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
1859 | dir_vp, DM_RIGHT_NULL, NULL, | |
1860 | DM_RIGHT_NULL, name, NULL, | |
1861 | dm_di_mode, 0, 0); | |
1862 | ||
1863 | if (error) | |
1864 | return error; | |
1865 | dm_event_sent = 1; | |
1866 | } | |
1867 | ||
1868 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1869 | return XFS_ERROR(EIO); | |
1870 | ||
1871 | /* Return through std_return after this point. */ | |
1872 | ||
1873 | udqp = gdqp = NULL; | |
365ca83d NS |
1874 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1875 | prid = dp->i_d.di_projid; | |
1876 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
1877 | prid = (xfs_prid_t)vap->va_projid; |
1878 | else | |
1879 | prid = (xfs_prid_t)dfltprid; | |
1880 | ||
1881 | /* | |
1882 | * Make sure that we have allocated dquot(s) on disk. | |
1883 | */ | |
1884 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 1885 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
1886 | XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1887 | if (error) | |
1888 | goto std_return; | |
1889 | ||
1890 | ip = NULL; | |
1da177e4 LT |
1891 | |
1892 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1893 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1894 | resblks = XFS_CREATE_SPACE_RES(mp, namelen); | |
1895 | /* | |
1896 | * Initially assume that the file does not exist and | |
1897 | * reserve the resources for that case. If that is not | |
1898 | * the case we'll drop the one we have and get a more | |
1899 | * appropriate transaction later. | |
1900 | */ | |
1901 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1902 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1903 | if (error == ENOSPC) { | |
1904 | resblks = 0; | |
1905 | error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0, | |
1906 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1907 | } | |
1908 | if (error) { | |
1909 | cancel_flags = 0; | |
1da177e4 LT |
1910 | goto error_return; |
1911 | } | |
1912 | ||
f7c66ce3 | 1913 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1914 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
1915 | |
1916 | XFS_BMAP_INIT(&free_list, &first_block); | |
1917 | ||
1918 | ASSERT(ip == NULL); | |
1919 | ||
1920 | /* | |
1921 | * Reserve disk quota and the inode. | |
1922 | */ | |
1923 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
1924 | if (error) | |
1925 | goto error_return; | |
1926 | ||
f6c2d1fa | 1927 | if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen))) |
1da177e4 LT |
1928 | goto error_return; |
1929 | rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0; | |
0432dab2 | 1930 | error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1, |
1da177e4 LT |
1931 | rdev, credp, prid, resblks > 0, |
1932 | &ip, &committed); | |
1933 | if (error) { | |
1934 | if (error == ENOSPC) | |
1935 | goto error_return; | |
1936 | goto abort_return; | |
1937 | } | |
1938 | ITRACE(ip); | |
1939 | ||
1940 | /* | |
1941 | * At this point, we've gotten a newly allocated inode. | |
1942 | * It is locked (and joined to the transaction). | |
1943 | */ | |
1944 | ||
1945 | ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE)); | |
1946 | ||
1947 | /* | |
993386c1 CH |
1948 | * Now we join the directory inode to the transaction. We do not do it |
1949 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1950 | * (and release all the locks). An error from here on will result in | |
1951 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1952 | * error path. | |
1da177e4 | 1953 | */ |
1da177e4 LT |
1954 | VN_HOLD(dir_vp); |
1955 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 1956 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 1957 | |
f6c2d1fa NS |
1958 | error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino, |
1959 | &first_block, &free_list, resblks ? | |
1960 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1961 | if (error) { |
1962 | ASSERT(error != ENOSPC); | |
1963 | goto abort_return; | |
1964 | } | |
1965 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1966 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1967 | ||
1968 | /* | |
1969 | * If this is a synchronous mount, make sure that the | |
1970 | * create transaction goes to disk before returning to | |
1971 | * the user. | |
1972 | */ | |
1973 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1974 | xfs_trans_set_sync(tp); | |
1975 | } | |
1976 | ||
1977 | dp->i_gen++; | |
1978 | ||
1979 | /* | |
1980 | * Attach the dquot(s) to the inodes and modify them incore. | |
1981 | * These ids of the inode couldn't have changed since the new | |
1982 | * inode has been locked ever since it was created. | |
1983 | */ | |
1984 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
1985 | ||
1986 | /* | |
1987 | * xfs_trans_commit normally decrements the vnode ref count | |
1988 | * when it unlocks the inode. Since we want to return the | |
1989 | * vnode to the caller, we bump the vnode ref count now. | |
1990 | */ | |
1991 | IHOLD(ip); | |
1992 | vp = XFS_ITOV(ip); | |
1993 | ||
f7c99b6f | 1994 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1995 | if (error) { |
1996 | xfs_bmap_cancel(&free_list); | |
1997 | goto abort_rele; | |
1998 | } | |
1999 | ||
1c72bf90 | 2000 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2001 | if (error) { |
2002 | IRELE(ip); | |
2003 | tp = NULL; | |
2004 | goto error_return; | |
2005 | } | |
2006 | ||
2007 | XFS_QM_DQRELE(mp, udqp); | |
2008 | XFS_QM_DQRELE(mp, gdqp); | |
2009 | ||
1da177e4 LT |
2010 | *vpp = vp; |
2011 | ||
2012 | /* Fallthrough to std_return with error = 0 */ | |
2013 | ||
2014 | std_return: | |
eb9df39d | 2015 | if ((*vpp || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 2016 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 LT |
2017 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
2018 | dir_vp, DM_RIGHT_NULL, | |
2019 | *vpp ? vp:NULL, | |
2020 | DM_RIGHT_NULL, name, NULL, | |
2021 | dm_di_mode, error, 0); | |
2022 | } | |
2023 | return error; | |
2024 | ||
2025 | abort_return: | |
2026 | cancel_flags |= XFS_TRANS_ABORT; | |
2027 | /* FALLTHROUGH */ | |
1da177e4 | 2028 | |
014c2544 | 2029 | error_return: |
1da177e4 LT |
2030 | if (tp != NULL) |
2031 | xfs_trans_cancel(tp, cancel_flags); | |
2032 | ||
1da177e4 LT |
2033 | XFS_QM_DQRELE(mp, udqp); |
2034 | XFS_QM_DQRELE(mp, gdqp); | |
2035 | ||
993386c1 CH |
2036 | if (unlock_dp_on_error) |
2037 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2038 | ||
1da177e4 LT |
2039 | goto std_return; |
2040 | ||
2041 | abort_rele: | |
2042 | /* | |
2043 | * Wait until after the current transaction is aborted to | |
2044 | * release the inode. This prevents recursive transactions | |
2045 | * and deadlocks from xfs_inactive. | |
2046 | */ | |
2047 | cancel_flags |= XFS_TRANS_ABORT; | |
2048 | xfs_trans_cancel(tp, cancel_flags); | |
2049 | IRELE(ip); | |
2050 | ||
2051 | XFS_QM_DQRELE(mp, udqp); | |
2052 | XFS_QM_DQRELE(mp, gdqp); | |
2053 | ||
2054 | goto std_return; | |
2055 | } | |
2056 | ||
2057 | #ifdef DEBUG | |
2058 | /* | |
2059 | * Some counters to see if (and how often) we are hitting some deadlock | |
2060 | * prevention code paths. | |
2061 | */ | |
2062 | ||
2063 | int xfs_rm_locks; | |
2064 | int xfs_rm_lock_delays; | |
2065 | int xfs_rm_attempts; | |
2066 | #endif | |
2067 | ||
2068 | /* | |
2069 | * The following routine will lock the inodes associated with the | |
2070 | * directory and the named entry in the directory. The locks are | |
2071 | * acquired in increasing inode number. | |
2072 | * | |
2073 | * If the entry is "..", then only the directory is locked. The | |
2074 | * vnode ref count will still include that from the .. entry in | |
2075 | * this case. | |
2076 | * | |
2077 | * There is a deadlock we need to worry about. If the locked directory is | |
2078 | * in the AIL, it might be blocking up the log. The next inode we lock | |
2079 | * could be already locked by another thread waiting for log space (e.g | |
2080 | * a permanent log reservation with a long running transaction (see | |
2081 | * xfs_itruncate_finish)). To solve this, we must check if the directory | |
2082 | * is in the ail and use lock_nowait. If we can't lock, we need to | |
2083 | * drop the inode lock on the directory and try again. xfs_iunlock will | |
2084 | * potentially push the tail if we were holding up the log. | |
2085 | */ | |
2086 | STATIC int | |
2087 | xfs_lock_dir_and_entry( | |
2088 | xfs_inode_t *dp, | |
1da177e4 LT |
2089 | xfs_inode_t *ip) /* inode of entry 'name' */ |
2090 | { | |
2091 | int attempts; | |
2092 | xfs_ino_t e_inum; | |
2093 | xfs_inode_t *ips[2]; | |
2094 | xfs_log_item_t *lp; | |
2095 | ||
2096 | #ifdef DEBUG | |
2097 | xfs_rm_locks++; | |
2098 | #endif | |
2099 | attempts = 0; | |
2100 | ||
2101 | again: | |
f7c66ce3 | 2102 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
1da177e4 LT |
2103 | |
2104 | e_inum = ip->i_ino; | |
2105 | ||
2106 | ITRACE(ip); | |
2107 | ||
2108 | /* | |
2109 | * We want to lock in increasing inum. Since we've already | |
2110 | * acquired the lock on the directory, we may need to release | |
2111 | * if if the inum of the entry turns out to be less. | |
2112 | */ | |
2113 | if (e_inum > dp->i_ino) { | |
2114 | /* | |
2115 | * We are already in the right order, so just | |
2116 | * lock on the inode of the entry. | |
2117 | * We need to use nowait if dp is in the AIL. | |
2118 | */ | |
2119 | ||
2120 | lp = (xfs_log_item_t *)dp->i_itemp; | |
2121 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2122 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2123 | attempts++; | |
2124 | #ifdef DEBUG | |
2125 | xfs_rm_attempts++; | |
2126 | #endif | |
2127 | ||
2128 | /* | |
2129 | * Unlock dp and try again. | |
2130 | * xfs_iunlock will try to push the tail | |
2131 | * if the inode is in the AIL. | |
2132 | */ | |
2133 | ||
2134 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2135 | ||
2136 | if ((attempts % 5) == 0) { | |
2137 | delay(1); /* Don't just spin the CPU */ | |
2138 | #ifdef DEBUG | |
2139 | xfs_rm_lock_delays++; | |
2140 | #endif | |
2141 | } | |
2142 | goto again; | |
2143 | } | |
2144 | } else { | |
2145 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2146 | } | |
2147 | } else if (e_inum < dp->i_ino) { | |
2148 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2149 | ||
2150 | ips[0] = ip; | |
2151 | ips[1] = dp; | |
2152 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2153 | } | |
2154 | /* else e_inum == dp->i_ino */ | |
2155 | /* This can happen if we're asked to lock /x/.. | |
2156 | * the entry is "..", which is also the parent directory. | |
2157 | */ | |
2158 | ||
2159 | return 0; | |
2160 | } | |
2161 | ||
2162 | #ifdef DEBUG | |
2163 | int xfs_locked_n; | |
2164 | int xfs_small_retries; | |
2165 | int xfs_middle_retries; | |
2166 | int xfs_lots_retries; | |
2167 | int xfs_lock_delays; | |
2168 | #endif | |
2169 | ||
f7c66ce3 LM |
2170 | /* |
2171 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
2172 | * a different value | |
2173 | */ | |
2174 | static inline int | |
2175 | xfs_lock_inumorder(int lock_mode, int subclass) | |
2176 | { | |
2177 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 2178 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 2179 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 2180 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
2181 | |
2182 | return lock_mode; | |
2183 | } | |
2184 | ||
1da177e4 LT |
2185 | /* |
2186 | * The following routine will lock n inodes in exclusive mode. | |
2187 | * We assume the caller calls us with the inodes in i_ino order. | |
2188 | * | |
2189 | * We need to detect deadlock where an inode that we lock | |
2190 | * is in the AIL and we start waiting for another inode that is locked | |
2191 | * by a thread in a long running transaction (such as truncate). This can | |
2192 | * result in deadlock since the long running trans might need to wait | |
2193 | * for the inode we just locked in order to push the tail and free space | |
2194 | * in the log. | |
2195 | */ | |
2196 | void | |
2197 | xfs_lock_inodes( | |
2198 | xfs_inode_t **ips, | |
2199 | int inodes, | |
2200 | int first_locked, | |
2201 | uint lock_mode) | |
2202 | { | |
2203 | int attempts = 0, i, j, try_lock; | |
2204 | xfs_log_item_t *lp; | |
2205 | ||
2206 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
2207 | ||
2208 | if (first_locked) { | |
2209 | try_lock = 1; | |
2210 | i = 1; | |
2211 | } else { | |
2212 | try_lock = 0; | |
2213 | i = 0; | |
2214 | } | |
2215 | ||
2216 | again: | |
2217 | for (; i < inodes; i++) { | |
2218 | ASSERT(ips[i]); | |
2219 | ||
2220 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
2221 | continue; | |
2222 | ||
2223 | /* | |
2224 | * If try_lock is not set yet, make sure all locked inodes | |
2225 | * are not in the AIL. | |
2226 | * If any are, set try_lock to be used later. | |
2227 | */ | |
2228 | ||
2229 | if (!try_lock) { | |
2230 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
2231 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
2232 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2233 | try_lock++; | |
2234 | } | |
2235 | } | |
2236 | } | |
2237 | ||
2238 | /* | |
2239 | * If any of the previous locks we have locked is in the AIL, | |
2240 | * we must TRY to get the second and subsequent locks. If | |
2241 | * we can't get any, we must release all we have | |
2242 | * and try again. | |
2243 | */ | |
2244 | ||
2245 | if (try_lock) { | |
2246 | /* try_lock must be 0 if i is 0. */ | |
2247 | /* | |
2248 | * try_lock means we have an inode locked | |
2249 | * that is in the AIL. | |
2250 | */ | |
2251 | ASSERT(i != 0); | |
f7c66ce3 | 2252 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
2253 | attempts++; |
2254 | ||
2255 | /* | |
2256 | * Unlock all previous guys and try again. | |
2257 | * xfs_iunlock will try to push the tail | |
2258 | * if the inode is in the AIL. | |
2259 | */ | |
2260 | ||
2261 | for(j = i - 1; j >= 0; j--) { | |
2262 | ||
2263 | /* | |
2264 | * Check to see if we've already | |
2265 | * unlocked this one. | |
2266 | * Not the first one going back, | |
2267 | * and the inode ptr is the same. | |
2268 | */ | |
2269 | if ((j != (i - 1)) && ips[j] == | |
2270 | ips[j+1]) | |
2271 | continue; | |
2272 | ||
2273 | xfs_iunlock(ips[j], lock_mode); | |
2274 | } | |
2275 | ||
2276 | if ((attempts % 5) == 0) { | |
2277 | delay(1); /* Don't just spin the CPU */ | |
2278 | #ifdef DEBUG | |
2279 | xfs_lock_delays++; | |
2280 | #endif | |
2281 | } | |
2282 | i = 0; | |
2283 | try_lock = 0; | |
2284 | goto again; | |
2285 | } | |
2286 | } else { | |
f7c66ce3 | 2287 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
2288 | } |
2289 | } | |
2290 | ||
2291 | #ifdef DEBUG | |
2292 | if (attempts) { | |
2293 | if (attempts < 5) xfs_small_retries++; | |
2294 | else if (attempts < 100) xfs_middle_retries++; | |
2295 | else xfs_lots_retries++; | |
2296 | } else { | |
2297 | xfs_locked_n++; | |
2298 | } | |
2299 | #endif | |
2300 | } | |
2301 | ||
2302 | #ifdef DEBUG | |
2303 | #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);} | |
2304 | int remove_which_error_return = 0; | |
2305 | #else /* ! DEBUG */ | |
2306 | #define REMOVE_DEBUG_TRACE(x) | |
2307 | #endif /* ! DEBUG */ | |
2308 | ||
993386c1 | 2309 | int |
1da177e4 | 2310 | xfs_remove( |
993386c1 CH |
2311 | xfs_inode_t *dp, |
2312 | bhv_vname_t *dentry) | |
1da177e4 | 2313 | { |
993386c1 | 2314 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2315 | char *name = VNAME(dentry); |
993386c1 CH |
2316 | xfs_mount_t *mp = dp->i_mount; |
2317 | xfs_inode_t *ip; | |
1da177e4 | 2318 | xfs_trans_t *tp = NULL; |
1da177e4 LT |
2319 | int error = 0; |
2320 | xfs_bmap_free_t free_list; | |
2321 | xfs_fsblock_t first_block; | |
2322 | int cancel_flags; | |
2323 | int committed; | |
2324 | int dm_di_mode = 0; | |
2325 | int link_zero; | |
2326 | uint resblks; | |
2327 | int namelen; | |
2328 | ||
1543d79c | 2329 | vn_trace_entry(dp, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 2330 | |
1da177e4 LT |
2331 | if (XFS_FORCED_SHUTDOWN(mp)) |
2332 | return XFS_ERROR(EIO); | |
2333 | ||
2334 | namelen = VNAMELEN(dentry); | |
2335 | ||
17370097 VA |
2336 | if (!xfs_get_dir_entry(dentry, &ip)) { |
2337 | dm_di_mode = ip->i_d.di_mode; | |
2338 | IRELE(ip); | |
2339 | } | |
2340 | ||
eb9df39d | 2341 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
1da177e4 LT |
2342 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp, |
2343 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
17370097 | 2344 | name, NULL, dm_di_mode, 0, 0); |
1da177e4 LT |
2345 | if (error) |
2346 | return error; | |
2347 | } | |
2348 | ||
2349 | /* From this point on, return through std_return */ | |
2350 | ip = NULL; | |
2351 | ||
2352 | /* | |
2353 | * We need to get a reference to ip before we get our log | |
2354 | * reservation. The reason for this is that we cannot call | |
2355 | * xfs_iget for an inode for which we do not have a reference | |
2356 | * once we've acquired a log reservation. This is because the | |
2357 | * inode we are trying to get might be in xfs_inactive going | |
2358 | * for a log reservation. Since we'll have to wait for the | |
2359 | * inactive code to complete before returning from xfs_iget, | |
2360 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2361 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2362 | * to the inode before getting our log reservation. |
2363 | */ | |
2364 | error = xfs_get_dir_entry(dentry, &ip); | |
2365 | if (error) { | |
2366 | REMOVE_DEBUG_TRACE(__LINE__); | |
2367 | goto std_return; | |
2368 | } | |
2369 | ||
2370 | dm_di_mode = ip->i_d.di_mode; | |
2371 | ||
1543d79c | 2372 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
2373 | |
2374 | ITRACE(ip); | |
2375 | ||
2376 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2377 | if (!error && dp != ip) | |
2378 | error = XFS_QM_DQATTACH(mp, ip, 0); | |
2379 | if (error) { | |
2380 | REMOVE_DEBUG_TRACE(__LINE__); | |
2381 | IRELE(ip); | |
2382 | goto std_return; | |
2383 | } | |
2384 | ||
2385 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2386 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2387 | /* | |
2388 | * We try to get the real space reservation first, | |
2389 | * allowing for directory btree deletion(s) implying | |
2390 | * possible bmap insert(s). If we can't get the space | |
2391 | * reservation then we use 0 instead, and avoid the bmap | |
2392 | * btree insert(s) in the directory code by, if the bmap | |
2393 | * insert tries to happen, instead trimming the LAST | |
2394 | * block from the directory. | |
2395 | */ | |
2396 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2397 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2398 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2399 | if (error == ENOSPC) { | |
2400 | resblks = 0; | |
2401 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2402 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2403 | } | |
2404 | if (error) { | |
2405 | ASSERT(error != ENOSPC); | |
2406 | REMOVE_DEBUG_TRACE(__LINE__); | |
2407 | xfs_trans_cancel(tp, 0); | |
2408 | IRELE(ip); | |
2409 | return error; | |
2410 | } | |
2411 | ||
e9ed9d22 | 2412 | error = xfs_lock_dir_and_entry(dp, ip); |
1da177e4 LT |
2413 | if (error) { |
2414 | REMOVE_DEBUG_TRACE(__LINE__); | |
2415 | xfs_trans_cancel(tp, cancel_flags); | |
2416 | IRELE(ip); | |
2417 | goto std_return; | |
2418 | } | |
2419 | ||
2420 | /* | |
2421 | * At this point, we've gotten both the directory and the entry | |
2422 | * inodes locked. | |
2423 | */ | |
2424 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2425 | if (dp != ip) { | |
2426 | /* | |
2427 | * Increment vnode ref count only in this case since | |
2428 | * there's an extra vnode reference in the case where | |
2429 | * dp == ip. | |
2430 | */ | |
2431 | IHOLD(dp); | |
2432 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2433 | } | |
2434 | ||
2435 | /* | |
2436 | * Entry must exist since we did a lookup in xfs_lock_dir_and_entry. | |
2437 | */ | |
2438 | XFS_BMAP_INIT(&free_list, &first_block); | |
f6c2d1fa NS |
2439 | error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino, |
2440 | &first_block, &free_list, 0); | |
1da177e4 LT |
2441 | if (error) { |
2442 | ASSERT(error != ENOENT); | |
2443 | REMOVE_DEBUG_TRACE(__LINE__); | |
2444 | goto error1; | |
2445 | } | |
2446 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2447 | ||
2448 | dp->i_gen++; | |
2449 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2450 | ||
2451 | error = xfs_droplink(tp, ip); | |
2452 | if (error) { | |
2453 | REMOVE_DEBUG_TRACE(__LINE__); | |
2454 | goto error1; | |
2455 | } | |
2456 | ||
2457 | /* Determine if this is the last link while | |
2458 | * we are in the transaction. | |
2459 | */ | |
2460 | link_zero = (ip)->i_d.di_nlink==0; | |
2461 | ||
2462 | /* | |
2463 | * Take an extra ref on the inode so that it doesn't | |
2464 | * go to xfs_inactive() from within the commit. | |
2465 | */ | |
2466 | IHOLD(ip); | |
2467 | ||
2468 | /* | |
2469 | * If this is a synchronous mount, make sure that the | |
2470 | * remove transaction goes to disk before returning to | |
2471 | * the user. | |
2472 | */ | |
2473 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2474 | xfs_trans_set_sync(tp); | |
2475 | } | |
2476 | ||
f7c99b6f | 2477 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2478 | if (error) { |
2479 | REMOVE_DEBUG_TRACE(__LINE__); | |
2480 | goto error_rele; | |
2481 | } | |
2482 | ||
1c72bf90 | 2483 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2484 | if (error) { |
2485 | IRELE(ip); | |
2486 | goto std_return; | |
2487 | } | |
2488 | ||
2489 | /* | |
2490 | * Before we drop our extra reference to the inode, purge it | |
2491 | * from the refcache if it is there. By waiting until afterwards | |
2492 | * to do the IRELE, we ensure that we won't go inactive in the | |
2493 | * xfs_refcache_purge_ip routine (although that would be OK). | |
2494 | */ | |
2495 | xfs_refcache_purge_ip(ip); | |
2496 | ||
2a82b8be DC |
2497 | /* |
2498 | * If we are using filestreams, kill the stream association. | |
2499 | * If the file is still open it may get a new one but that | |
2500 | * will get killed on last close in xfs_close() so we don't | |
2501 | * have to worry about that. | |
2502 | */ | |
2503 | if (link_zero && xfs_inode_is_filestream(ip)) | |
2504 | xfs_filestream_deassociate(ip); | |
2505 | ||
1543d79c | 2506 | vn_trace_exit(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 2507 | |
1da177e4 LT |
2508 | IRELE(ip); |
2509 | ||
2510 | /* Fall through to std_return with error = 0 */ | |
2511 | std_return: | |
eb9df39d | 2512 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 LT |
2513 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
2514 | dir_vp, DM_RIGHT_NULL, | |
2515 | NULL, DM_RIGHT_NULL, | |
2516 | name, NULL, dm_di_mode, error, 0); | |
2517 | } | |
2518 | return error; | |
2519 | ||
2520 | error1: | |
2521 | xfs_bmap_cancel(&free_list); | |
2522 | cancel_flags |= XFS_TRANS_ABORT; | |
2523 | xfs_trans_cancel(tp, cancel_flags); | |
2524 | goto std_return; | |
2525 | ||
2526 | error_rele: | |
2527 | /* | |
2528 | * In this case make sure to not release the inode until after | |
2529 | * the current transaction is aborted. Releasing it beforehand | |
2530 | * can cause us to go to xfs_inactive and start a recursive | |
2531 | * transaction which can easily deadlock with the current one. | |
2532 | */ | |
2533 | xfs_bmap_cancel(&free_list); | |
2534 | cancel_flags |= XFS_TRANS_ABORT; | |
2535 | xfs_trans_cancel(tp, cancel_flags); | |
2536 | ||
2537 | /* | |
2538 | * Before we drop our extra reference to the inode, purge it | |
2539 | * from the refcache if it is there. By waiting until afterwards | |
2540 | * to do the IRELE, we ensure that we won't go inactive in the | |
2541 | * xfs_refcache_purge_ip routine (although that would be OK). | |
2542 | */ | |
2543 | xfs_refcache_purge_ip(ip); | |
2544 | ||
2545 | IRELE(ip); | |
2546 | ||
2547 | goto std_return; | |
2548 | } | |
2549 | ||
993386c1 | 2550 | int |
1da177e4 | 2551 | xfs_link( |
993386c1 | 2552 | xfs_inode_t *tdp, |
67fcaa73 | 2553 | bhv_vnode_t *src_vp, |
993386c1 | 2554 | bhv_vname_t *dentry) |
1da177e4 | 2555 | { |
993386c1 CH |
2556 | bhv_vnode_t *target_dir_vp = XFS_ITOV(tdp); |
2557 | xfs_mount_t *mp = tdp->i_mount; | |
2558 | xfs_inode_t *sip = xfs_vtoi(src_vp); | |
1da177e4 | 2559 | xfs_trans_t *tp; |
1da177e4 LT |
2560 | xfs_inode_t *ips[2]; |
2561 | int error; | |
2562 | xfs_bmap_free_t free_list; | |
2563 | xfs_fsblock_t first_block; | |
2564 | int cancel_flags; | |
2565 | int committed; | |
1da177e4 LT |
2566 | int resblks; |
2567 | char *target_name = VNAME(dentry); | |
2568 | int target_namelen; | |
2569 | ||
1543d79c CH |
2570 | vn_trace_entry(tdp, __FUNCTION__, (inst_t *)__return_address); |
2571 | vn_trace_entry(xfs_vtoi(src_vp), __FUNCTION__, (inst_t *)__return_address); | |
1da177e4 LT |
2572 | |
2573 | target_namelen = VNAMELEN(dentry); | |
b71d300c | 2574 | ASSERT(!VN_ISDIR(src_vp)); |
1da177e4 | 2575 | |
1da177e4 LT |
2576 | if (XFS_FORCED_SHUTDOWN(mp)) |
2577 | return XFS_ERROR(EIO); | |
2578 | ||
eb9df39d | 2579 | if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) { |
1da177e4 LT |
2580 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, |
2581 | target_dir_vp, DM_RIGHT_NULL, | |
2582 | src_vp, DM_RIGHT_NULL, | |
2583 | target_name, NULL, 0, 0, 0); | |
2584 | if (error) | |
2585 | return error; | |
2586 | } | |
2587 | ||
2588 | /* Return through std_return after this point. */ | |
2589 | ||
2590 | error = XFS_QM_DQATTACH(mp, sip, 0); | |
2591 | if (!error && sip != tdp) | |
2592 | error = XFS_QM_DQATTACH(mp, tdp, 0); | |
2593 | if (error) | |
2594 | goto std_return; | |
2595 | ||
2596 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2597 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2598 | resblks = XFS_LINK_SPACE_RES(mp, target_namelen); | |
2599 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, | |
2600 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2601 | if (error == ENOSPC) { | |
2602 | resblks = 0; | |
2603 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2604 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2605 | } | |
2606 | if (error) { | |
2607 | cancel_flags = 0; | |
2608 | goto error_return; | |
2609 | } | |
2610 | ||
2611 | if (sip->i_ino < tdp->i_ino) { | |
2612 | ips[0] = sip; | |
2613 | ips[1] = tdp; | |
2614 | } else { | |
2615 | ips[0] = tdp; | |
2616 | ips[1] = sip; | |
2617 | } | |
2618 | ||
2619 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2620 | ||
2621 | /* | |
2622 | * Increment vnode ref counts since xfs_trans_commit & | |
2623 | * xfs_trans_cancel will both unlock the inodes and | |
2624 | * decrement the associated ref counts. | |
2625 | */ | |
2626 | VN_HOLD(src_vp); | |
2627 | VN_HOLD(target_dir_vp); | |
2628 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
2629 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2630 | ||
2631 | /* | |
2632 | * If the source has too many links, we can't make any more to it. | |
2633 | */ | |
2634 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2635 | error = XFS_ERROR(EMLINK); | |
2636 | goto error_return; | |
2637 | } | |
2638 | ||
365ca83d NS |
2639 | /* |
2640 | * If we are using project inheritance, we only allow hard link | |
2641 | * creation in our tree when the project IDs are the same; else | |
2642 | * the tree quota mechanism could be circumvented. | |
2643 | */ | |
2644 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2645 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 2646 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
2647 | goto error_return; |
2648 | } | |
2649 | ||
1da177e4 | 2650 | if (resblks == 0 && |
f6c2d1fa | 2651 | (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen))) |
1da177e4 LT |
2652 | goto error_return; |
2653 | ||
2654 | XFS_BMAP_INIT(&free_list, &first_block); | |
2655 | ||
f6c2d1fa | 2656 | error = xfs_dir_createname(tp, tdp, target_name, target_namelen, |
1da177e4 LT |
2657 | sip->i_ino, &first_block, &free_list, |
2658 | resblks); | |
2659 | if (error) | |
2660 | goto abort_return; | |
2661 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2662 | tdp->i_gen++; | |
2663 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
2664 | ||
2665 | error = xfs_bumplink(tp, sip); | |
b71d300c | 2666 | if (error) |
1da177e4 | 2667 | goto abort_return; |
1da177e4 LT |
2668 | |
2669 | /* | |
2670 | * If this is a synchronous mount, make sure that the | |
2671 | * link transaction goes to disk before returning to | |
2672 | * the user. | |
2673 | */ | |
2674 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2675 | xfs_trans_set_sync(tp); | |
2676 | } | |
2677 | ||
f7c99b6f | 2678 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2679 | if (error) { |
2680 | xfs_bmap_cancel(&free_list); | |
2681 | goto abort_return; | |
2682 | } | |
2683 | ||
1c72bf90 | 2684 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
b71d300c | 2685 | if (error) |
1da177e4 | 2686 | goto std_return; |
1da177e4 LT |
2687 | |
2688 | /* Fall through to std_return with error = 0. */ | |
2689 | std_return: | |
eb9df39d | 2690 | if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) { |
1da177e4 LT |
2691 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, |
2692 | target_dir_vp, DM_RIGHT_NULL, | |
2693 | src_vp, DM_RIGHT_NULL, | |
2694 | target_name, NULL, 0, error, 0); | |
2695 | } | |
2696 | return error; | |
2697 | ||
2698 | abort_return: | |
2699 | cancel_flags |= XFS_TRANS_ABORT; | |
2700 | /* FALLTHROUGH */ | |
014c2544 | 2701 | |
1da177e4 LT |
2702 | error_return: |
2703 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2704 | goto std_return; |
2705 | } | |
b71d300c AD |
2706 | |
2707 | ||
993386c1 | 2708 | int |
1da177e4 | 2709 | xfs_mkdir( |
993386c1 | 2710 | xfs_inode_t *dp, |
8285fb58 NS |
2711 | bhv_vname_t *dentry, |
2712 | bhv_vattr_t *vap, | |
67fcaa73 | 2713 | bhv_vnode_t **vpp, |
1da177e4 LT |
2714 | cred_t *credp) |
2715 | { | |
993386c1 | 2716 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2717 | char *dir_name = VNAME(dentry); |
993386c1 CH |
2718 | int dir_namelen = VNAMELEN(dentry); |
2719 | xfs_mount_t *mp = dp->i_mount; | |
1da177e4 | 2720 | xfs_inode_t *cdp; /* inode of created dir */ |
67fcaa73 | 2721 | bhv_vnode_t *cvp; /* vnode of created dir */ |
1da177e4 | 2722 | xfs_trans_t *tp; |
1da177e4 LT |
2723 | int cancel_flags; |
2724 | int error; | |
2725 | int committed; | |
2726 | xfs_bmap_free_t free_list; | |
2727 | xfs_fsblock_t first_block; | |
993386c1 | 2728 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2729 | boolean_t created = B_FALSE; |
2730 | int dm_event_sent = 0; | |
2731 | xfs_prid_t prid; | |
2732 | struct xfs_dquot *udqp, *gdqp; | |
2733 | uint resblks; | |
2734 | int dm_di_mode; | |
1da177e4 LT |
2735 | |
2736 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2737 | return XFS_ERROR(EIO); | |
2738 | ||
1da177e4 | 2739 | tp = NULL; |
0432dab2 | 2740 | dm_di_mode = vap->va_mode; |
1da177e4 | 2741 | |
eb9df39d | 2742 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 LT |
2743 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
2744 | dir_vp, DM_RIGHT_NULL, NULL, | |
2745 | DM_RIGHT_NULL, dir_name, NULL, | |
2746 | dm_di_mode, 0, 0); | |
2747 | if (error) | |
2748 | return error; | |
2749 | dm_event_sent = 1; | |
2750 | } | |
2751 | ||
2752 | /* Return through std_return after this point. */ | |
2753 | ||
1543d79c | 2754 | vn_trace_entry(dp, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
2755 | |
2756 | mp = dp->i_mount; | |
2757 | udqp = gdqp = NULL; | |
365ca83d NS |
2758 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2759 | prid = dp->i_d.di_projid; | |
2760 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
2761 | prid = (xfs_prid_t)vap->va_projid; |
2762 | else | |
2763 | prid = (xfs_prid_t)dfltprid; | |
2764 | ||
2765 | /* | |
2766 | * Make sure that we have allocated dquot(s) on disk. | |
2767 | */ | |
2768 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 2769 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
2770 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2771 | if (error) | |
2772 | goto std_return; | |
2773 | ||
2774 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
2775 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2776 | resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen); | |
2777 | error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0, | |
2778 | XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT); | |
2779 | if (error == ENOSPC) { | |
2780 | resblks = 0; | |
2781 | error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0, | |
2782 | XFS_TRANS_PERM_LOG_RES, | |
2783 | XFS_MKDIR_LOG_COUNT); | |
2784 | } | |
2785 | if (error) { | |
2786 | cancel_flags = 0; | |
1da177e4 LT |
2787 | goto error_return; |
2788 | } | |
2789 | ||
f7c66ce3 | 2790 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2791 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2792 | |
2793 | /* | |
2794 | * Check for directory link count overflow. | |
2795 | */ | |
2796 | if (dp->i_d.di_nlink >= XFS_MAXLINK) { | |
2797 | error = XFS_ERROR(EMLINK); | |
2798 | goto error_return; | |
2799 | } | |
2800 | ||
2801 | /* | |
2802 | * Reserve disk quota and the inode. | |
2803 | */ | |
2804 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
2805 | if (error) | |
2806 | goto error_return; | |
2807 | ||
2808 | if (resblks == 0 && | |
f6c2d1fa | 2809 | (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen))) |
1da177e4 LT |
2810 | goto error_return; |
2811 | /* | |
2812 | * create the directory inode. | |
2813 | */ | |
0432dab2 | 2814 | error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2, |
1da177e4 LT |
2815 | 0, credp, prid, resblks > 0, |
2816 | &cdp, NULL); | |
2817 | if (error) { | |
2818 | if (error == ENOSPC) | |
2819 | goto error_return; | |
2820 | goto abort_return; | |
2821 | } | |
2822 | ITRACE(cdp); | |
2823 | ||
2824 | /* | |
2825 | * Now we add the directory inode to the transaction. | |
2826 | * We waited until now since xfs_dir_ialloc might start | |
2827 | * a new transaction. Had we joined the transaction | |
993386c1 CH |
2828 | * earlier, the locks might have gotten released. An error |
2829 | * from here on will result in the transaction cancel | |
2830 | * unlocking dp so don't do it explicitly in the error path. | |
1da177e4 LT |
2831 | */ |
2832 | VN_HOLD(dir_vp); | |
2833 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 2834 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2835 | |
2836 | XFS_BMAP_INIT(&free_list, &first_block); | |
2837 | ||
f6c2d1fa NS |
2838 | error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino, |
2839 | &first_block, &free_list, resblks ? | |
2840 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
2841 | if (error) { |
2842 | ASSERT(error != ENOSPC); | |
2843 | goto error1; | |
2844 | } | |
2845 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2846 | ||
2847 | /* | |
2848 | * Bump the in memory version number of the parent directory | |
2849 | * so that other processes accessing it will recognize that | |
2850 | * the directory has changed. | |
2851 | */ | |
2852 | dp->i_gen++; | |
2853 | ||
f6c2d1fa NS |
2854 | error = xfs_dir_init(tp, cdp, dp); |
2855 | if (error) | |
1da177e4 | 2856 | goto error2; |
1da177e4 LT |
2857 | |
2858 | cdp->i_gen = 1; | |
2859 | error = xfs_bumplink(tp, dp); | |
f6c2d1fa | 2860 | if (error) |
1da177e4 | 2861 | goto error2; |
1da177e4 LT |
2862 | |
2863 | cvp = XFS_ITOV(cdp); | |
2864 | ||
2865 | created = B_TRUE; | |
2866 | ||
2867 | *vpp = cvp; | |
2868 | IHOLD(cdp); | |
2869 | ||
2870 | /* | |
2871 | * Attach the dquots to the new inode and modify the icount incore. | |
2872 | */ | |
2873 | XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp); | |
2874 | ||
2875 | /* | |
2876 | * If this is a synchronous mount, make sure that the | |
2877 | * mkdir transaction goes to disk before returning to | |
2878 | * the user. | |
2879 | */ | |
2880 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2881 | xfs_trans_set_sync(tp); | |
2882 | } | |
2883 | ||
f7c99b6f | 2884 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2885 | if (error) { |
2886 | IRELE(cdp); | |
2887 | goto error2; | |
2888 | } | |
2889 | ||
1c72bf90 | 2890 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2891 | XFS_QM_DQRELE(mp, udqp); |
2892 | XFS_QM_DQRELE(mp, gdqp); | |
2893 | if (error) { | |
2894 | IRELE(cdp); | |
2895 | } | |
2896 | ||
2897 | /* Fall through to std_return with error = 0 or errno from | |
2898 | * xfs_trans_commit. */ | |
2899 | ||
2900 | std_return: | |
eb9df39d | 2901 | if ((created || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 2902 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 LT |
2903 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
2904 | dir_vp, DM_RIGHT_NULL, | |
2905 | created ? XFS_ITOV(cdp):NULL, | |
2906 | DM_RIGHT_NULL, | |
2907 | dir_name, NULL, | |
2908 | dm_di_mode, error, 0); | |
2909 | } | |
2910 | return error; | |
2911 | ||
2912 | error2: | |
2913 | error1: | |
2914 | xfs_bmap_cancel(&free_list); | |
2915 | abort_return: | |
2916 | cancel_flags |= XFS_TRANS_ABORT; | |
2917 | error_return: | |
2918 | xfs_trans_cancel(tp, cancel_flags); | |
2919 | XFS_QM_DQRELE(mp, udqp); | |
2920 | XFS_QM_DQRELE(mp, gdqp); | |
2921 | ||
993386c1 | 2922 | if (unlock_dp_on_error) |
1da177e4 | 2923 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2924 | |
2925 | goto std_return; | |
2926 | } | |
2927 | ||
993386c1 | 2928 | int |
1da177e4 | 2929 | xfs_rmdir( |
993386c1 CH |
2930 | xfs_inode_t *dp, |
2931 | bhv_vname_t *dentry) | |
1da177e4 | 2932 | { |
993386c1 | 2933 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2934 | char *name = VNAME(dentry); |
993386c1 CH |
2935 | int namelen = VNAMELEN(dentry); |
2936 | xfs_mount_t *mp = dp->i_mount; | |
2937 | xfs_inode_t *cdp; /* child directory */ | |
1da177e4 | 2938 | xfs_trans_t *tp; |
1da177e4 LT |
2939 | int error; |
2940 | xfs_bmap_free_t free_list; | |
2941 | xfs_fsblock_t first_block; | |
2942 | int cancel_flags; | |
2943 | int committed; | |
17370097 | 2944 | int dm_di_mode = S_IFDIR; |
1da177e4 | 2945 | int last_cdp_link; |
1da177e4 LT |
2946 | uint resblks; |
2947 | ||
1543d79c | 2948 | vn_trace_entry(dp, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 2949 | |
993386c1 | 2950 | if (XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 | 2951 | return XFS_ERROR(EIO); |
1da177e4 | 2952 | |
17370097 VA |
2953 | if (!xfs_get_dir_entry(dentry, &cdp)) { |
2954 | dm_di_mode = cdp->i_d.di_mode; | |
2955 | IRELE(cdp); | |
2956 | } | |
2957 | ||
eb9df39d | 2958 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
1da177e4 LT |
2959 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, |
2960 | dir_vp, DM_RIGHT_NULL, | |
2961 | NULL, DM_RIGHT_NULL, | |
17370097 | 2962 | name, NULL, dm_di_mode, 0, 0); |
1da177e4 LT |
2963 | if (error) |
2964 | return XFS_ERROR(error); | |
2965 | } | |
2966 | ||
2967 | /* Return through std_return after this point. */ | |
2968 | ||
2969 | cdp = NULL; | |
2970 | ||
2971 | /* | |
2972 | * We need to get a reference to cdp before we get our log | |
2973 | * reservation. The reason for this is that we cannot call | |
2974 | * xfs_iget for an inode for which we do not have a reference | |
2975 | * once we've acquired a log reservation. This is because the | |
2976 | * inode we are trying to get might be in xfs_inactive going | |
2977 | * for a log reservation. Since we'll have to wait for the | |
2978 | * inactive code to complete before returning from xfs_iget, | |
2979 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2980 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2981 | * to the inode before getting our log reservation. |
2982 | */ | |
2983 | error = xfs_get_dir_entry(dentry, &cdp); | |
2984 | if (error) { | |
2985 | REMOVE_DEBUG_TRACE(__LINE__); | |
2986 | goto std_return; | |
2987 | } | |
2988 | mp = dp->i_mount; | |
2989 | dm_di_mode = cdp->i_d.di_mode; | |
2990 | ||
2991 | /* | |
2992 | * Get the dquots for the inodes. | |
2993 | */ | |
2994 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2995 | if (!error && dp != cdp) | |
2996 | error = XFS_QM_DQATTACH(mp, cdp, 0); | |
2997 | if (error) { | |
2998 | IRELE(cdp); | |
2999 | REMOVE_DEBUG_TRACE(__LINE__); | |
3000 | goto std_return; | |
3001 | } | |
3002 | ||
3003 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
3004 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3005 | /* | |
3006 | * We try to get the real space reservation first, | |
3007 | * allowing for directory btree deletion(s) implying | |
3008 | * possible bmap insert(s). If we can't get the space | |
3009 | * reservation then we use 0 instead, and avoid the bmap | |
3010 | * btree insert(s) in the directory code by, if the bmap | |
3011 | * insert tries to happen, instead trimming the LAST | |
3012 | * block from the directory. | |
3013 | */ | |
3014 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
3015 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
3016 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
3017 | if (error == ENOSPC) { | |
3018 | resblks = 0; | |
3019 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
3020 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
3021 | } | |
3022 | if (error) { | |
3023 | ASSERT(error != ENOSPC); | |
3024 | cancel_flags = 0; | |
3025 | IRELE(cdp); | |
3026 | goto error_return; | |
3027 | } | |
3028 | XFS_BMAP_INIT(&free_list, &first_block); | |
3029 | ||
3030 | /* | |
3031 | * Now lock the child directory inode and the parent directory | |
3032 | * inode in the proper order. This will take care of validating | |
3033 | * that the directory entry for the child directory inode has | |
3034 | * not changed while we were obtaining a log reservation. | |
3035 | */ | |
e9ed9d22 | 3036 | error = xfs_lock_dir_and_entry(dp, cdp); |
1da177e4 LT |
3037 | if (error) { |
3038 | xfs_trans_cancel(tp, cancel_flags); | |
3039 | IRELE(cdp); | |
3040 | goto std_return; | |
3041 | } | |
3042 | ||
3043 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
3044 | if (dp != cdp) { | |
3045 | /* | |
3046 | * Only increment the parent directory vnode count if | |
3047 | * we didn't bump it in looking up cdp. The only time | |
3048 | * we don't bump it is when we're looking up ".". | |
3049 | */ | |
3050 | VN_HOLD(dir_vp); | |
3051 | } | |
3052 | ||
3053 | ITRACE(cdp); | |
3054 | xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL); | |
3055 | ||
3056 | ASSERT(cdp->i_d.di_nlink >= 2); | |
3057 | if (cdp->i_d.di_nlink != 2) { | |
3058 | error = XFS_ERROR(ENOTEMPTY); | |
3059 | goto error_return; | |
3060 | } | |
f6c2d1fa | 3061 | if (!xfs_dir_isempty(cdp)) { |
1da177e4 LT |
3062 | error = XFS_ERROR(ENOTEMPTY); |
3063 | goto error_return; | |
3064 | } | |
3065 | ||
f6c2d1fa NS |
3066 | error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino, |
3067 | &first_block, &free_list, resblks); | |
3068 | if (error) | |
1da177e4 | 3069 | goto error1; |
1da177e4 LT |
3070 | |
3071 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3072 | ||
3073 | /* | |
3074 | * Bump the in memory generation count on the parent | |
3075 | * directory so that other can know that it has changed. | |
3076 | */ | |
3077 | dp->i_gen++; | |
3078 | ||
3079 | /* | |
3080 | * Drop the link from cdp's "..". | |
3081 | */ | |
3082 | error = xfs_droplink(tp, dp); | |
3083 | if (error) { | |
3084 | goto error1; | |
3085 | } | |
3086 | ||
3087 | /* | |
3088 | * Drop the link from dp to cdp. | |
3089 | */ | |
3090 | error = xfs_droplink(tp, cdp); | |
3091 | if (error) { | |
3092 | goto error1; | |
3093 | } | |
3094 | ||
3095 | /* | |
3096 | * Drop the "." link from cdp to self. | |
3097 | */ | |
3098 | error = xfs_droplink(tp, cdp); | |
3099 | if (error) { | |
3100 | goto error1; | |
3101 | } | |
3102 | ||
3103 | /* Determine these before committing transaction */ | |
3104 | last_cdp_link = (cdp)->i_d.di_nlink==0; | |
3105 | ||
3106 | /* | |
3107 | * Take an extra ref on the child vnode so that it | |
3108 | * does not go to xfs_inactive() from within the commit. | |
3109 | */ | |
3110 | IHOLD(cdp); | |
3111 | ||
3112 | /* | |
3113 | * If this is a synchronous mount, make sure that the | |
3114 | * rmdir transaction goes to disk before returning to | |
3115 | * the user. | |
3116 | */ | |
3117 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3118 | xfs_trans_set_sync(tp); | |
3119 | } | |
3120 | ||
f7c99b6f | 3121 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
3122 | if (error) { |
3123 | xfs_bmap_cancel(&free_list); | |
3124 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | | |
3125 | XFS_TRANS_ABORT)); | |
3126 | IRELE(cdp); | |
3127 | goto std_return; | |
3128 | } | |
3129 | ||
1c72bf90 | 3130 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3131 | if (error) { |
3132 | IRELE(cdp); | |
3133 | goto std_return; | |
3134 | } | |
3135 | ||
3136 | ||
1da177e4 LT |
3137 | IRELE(cdp); |
3138 | ||
3139 | /* Fall through to std_return with error = 0 or the errno | |
3140 | * from xfs_trans_commit. */ | |
bb19fba1 | 3141 | std_return: |
eb9df39d | 3142 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 LT |
3143 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
3144 | dir_vp, DM_RIGHT_NULL, | |
3145 | NULL, DM_RIGHT_NULL, | |
3146 | name, NULL, dm_di_mode, | |
3147 | error, 0); | |
3148 | } | |
3149 | return error; | |
3150 | ||
bb19fba1 | 3151 | error1: |
1da177e4 LT |
3152 | xfs_bmap_cancel(&free_list); |
3153 | cancel_flags |= XFS_TRANS_ABORT; | |
014c2544 JJ |
3154 | /* FALLTHROUGH */ |
3155 | ||
bb19fba1 | 3156 | error_return: |
1da177e4 LT |
3157 | xfs_trans_cancel(tp, cancel_flags); |
3158 | goto std_return; | |
3159 | } | |
3160 | ||
993386c1 | 3161 | int |
1da177e4 | 3162 | xfs_symlink( |
993386c1 | 3163 | xfs_inode_t *dp, |
8285fb58 NS |
3164 | bhv_vname_t *dentry, |
3165 | bhv_vattr_t *vap, | |
1da177e4 | 3166 | char *target_path, |
67fcaa73 | 3167 | bhv_vnode_t **vpp, |
1da177e4 LT |
3168 | cred_t *credp) |
3169 | { | |
993386c1 CH |
3170 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
3171 | xfs_mount_t *mp = dp->i_mount; | |
1da177e4 | 3172 | xfs_trans_t *tp; |
1da177e4 LT |
3173 | xfs_inode_t *ip; |
3174 | int error; | |
3175 | int pathlen; | |
3176 | xfs_bmap_free_t free_list; | |
3177 | xfs_fsblock_t first_block; | |
993386c1 | 3178 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
3179 | uint cancel_flags; |
3180 | int committed; | |
3181 | xfs_fileoff_t first_fsb; | |
3182 | xfs_filblks_t fs_blocks; | |
3183 | int nmaps; | |
3184 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
3185 | xfs_daddr_t d; | |
3186 | char *cur_chunk; | |
3187 | int byte_cnt; | |
3188 | int n; | |
3189 | xfs_buf_t *bp; | |
3190 | xfs_prid_t prid; | |
3191 | struct xfs_dquot *udqp, *gdqp; | |
3192 | uint resblks; | |
3193 | char *link_name = VNAME(dentry); | |
3194 | int link_namelen; | |
3195 | ||
3196 | *vpp = NULL; | |
1da177e4 LT |
3197 | error = 0; |
3198 | ip = NULL; | |
3199 | tp = NULL; | |
3200 | ||
1543d79c | 3201 | vn_trace_entry(dp, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 3202 | |
1da177e4 LT |
3203 | |
3204 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3205 | return XFS_ERROR(EIO); | |
3206 | ||
3207 | link_namelen = VNAMELEN(dentry); | |
3208 | ||
3209 | /* | |
3210 | * Check component lengths of the target path name. | |
3211 | */ | |
3212 | pathlen = strlen(target_path); | |
3213 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
3214 | return XFS_ERROR(ENAMETOOLONG); | |
3215 | if (pathlen >= MAXNAMELEN) { /* is any component too long? */ | |
3216 | int len, total; | |
3217 | char *path; | |
3218 | ||
f6c2d1fa | 3219 | for (total = 0, path = target_path; total < pathlen;) { |
1da177e4 LT |
3220 | /* |
3221 | * Skip any slashes. | |
3222 | */ | |
3223 | while(*path == '/') { | |
3224 | total++; | |
3225 | path++; | |
3226 | } | |
3227 | ||
3228 | /* | |
3229 | * Count up to the next slash or end of path. | |
3230 | * Error out if the component is bigger than MAXNAMELEN. | |
3231 | */ | |
3232 | for(len = 0; *path != '/' && total < pathlen;total++, path++) { | |
3233 | if (++len >= MAXNAMELEN) { | |
3234 | error = ENAMETOOLONG; | |
3235 | return error; | |
3236 | } | |
3237 | } | |
3238 | } | |
3239 | } | |
3240 | ||
eb9df39d | 3241 | if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) { |
1da177e4 LT |
3242 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp, |
3243 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
3244 | link_name, target_path, 0, 0, 0); | |
3245 | if (error) | |
3246 | return error; | |
3247 | } | |
3248 | ||
3249 | /* Return through std_return after this point. */ | |
3250 | ||
3251 | udqp = gdqp = NULL; | |
365ca83d NS |
3252 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
3253 | prid = dp->i_d.di_projid; | |
3254 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
3255 | prid = (xfs_prid_t)vap->va_projid; |
3256 | else | |
3257 | prid = (xfs_prid_t)dfltprid; | |
3258 | ||
3259 | /* | |
3260 | * Make sure that we have allocated dquot(s) on disk. | |
3261 | */ | |
3262 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 3263 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
3264 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
3265 | if (error) | |
3266 | goto std_return; | |
3267 | ||
3268 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
3269 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3270 | /* | |
3271 | * The symlink will fit into the inode data fork? | |
3272 | * There can't be any attributes so we get the whole variable part. | |
3273 | */ | |
3274 | if (pathlen <= XFS_LITINO(mp)) | |
3275 | fs_blocks = 0; | |
3276 | else | |
3277 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
3278 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks); | |
3279 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, | |
3280 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3281 | if (error == ENOSPC && fs_blocks == 0) { | |
3282 | resblks = 0; | |
3283 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
3284 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3285 | } | |
3286 | if (error) { | |
3287 | cancel_flags = 0; | |
1da177e4 LT |
3288 | goto error_return; |
3289 | } | |
3290 | ||
f7c66ce3 | 3291 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 3292 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
3293 | |
3294 | /* | |
3295 | * Check whether the directory allows new symlinks or not. | |
3296 | */ | |
3297 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
3298 | error = XFS_ERROR(EPERM); | |
3299 | goto error_return; | |
3300 | } | |
3301 | ||
3302 | /* | |
3303 | * Reserve disk quota : blocks and inode. | |
3304 | */ | |
3305 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
3306 | if (error) | |
3307 | goto error_return; | |
3308 | ||
3309 | /* | |
3310 | * Check for ability to enter directory entry, if no space reserved. | |
3311 | */ | |
3312 | if (resblks == 0 && | |
f6c2d1fa | 3313 | (error = xfs_dir_canenter(tp, dp, link_name, link_namelen))) |
1da177e4 LT |
3314 | goto error_return; |
3315 | /* | |
3316 | * Initialize the bmap freelist prior to calling either | |
3317 | * bmapi or the directory create code. | |
3318 | */ | |
3319 | XFS_BMAP_INIT(&free_list, &first_block); | |
3320 | ||
3321 | /* | |
3322 | * Allocate an inode for the symlink. | |
3323 | */ | |
3324 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT), | |
3325 | 1, 0, credp, prid, resblks > 0, &ip, NULL); | |
3326 | if (error) { | |
3327 | if (error == ENOSPC) | |
3328 | goto error_return; | |
3329 | goto error1; | |
3330 | } | |
3331 | ITRACE(ip); | |
3332 | ||
993386c1 CH |
3333 | /* |
3334 | * An error after we've joined dp to the transaction will result in the | |
3335 | * transaction cancel unlocking dp so don't do it explicitly in the | |
3336 | * error path. | |
3337 | */ | |
1da177e4 LT |
3338 | VN_HOLD(dir_vp); |
3339 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 3340 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
3341 | |
3342 | /* | |
3343 | * Also attach the dquot(s) to it, if applicable. | |
3344 | */ | |
3345 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
3346 | ||
3347 | if (resblks) | |
3348 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
3349 | /* | |
3350 | * If the symlink will fit into the inode, write it inline. | |
3351 | */ | |
3352 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
3353 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
3354 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
3355 | ip->i_d.di_size = pathlen; | |
3356 | ||
3357 | /* | |
3358 | * The inode was initially created in extent format. | |
3359 | */ | |
3360 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
3361 | ip->i_df.if_flags |= XFS_IFINLINE; | |
3362 | ||
3363 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
3364 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
3365 | ||
3366 | } else { | |
3367 | first_fsb = 0; | |
3368 | nmaps = SYMLINK_MAPS; | |
3369 | ||
3370 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
3371 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
3372 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 3373 | &free_list, NULL); |
1da177e4 LT |
3374 | if (error) { |
3375 | goto error1; | |
3376 | } | |
3377 | ||
3378 | if (resblks) | |
3379 | resblks -= fs_blocks; | |
3380 | ip->i_d.di_size = pathlen; | |
3381 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3382 | ||
3383 | cur_chunk = target_path; | |
3384 | for (n = 0; n < nmaps; n++) { | |
3385 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
3386 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
3387 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
3388 | BTOBB(byte_cnt), 0); | |
3389 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
3390 | if (pathlen < byte_cnt) { | |
3391 | byte_cnt = pathlen; | |
3392 | } | |
3393 | pathlen -= byte_cnt; | |
3394 | ||
3395 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
3396 | cur_chunk += byte_cnt; | |
3397 | ||
3398 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
3399 | } | |
3400 | } | |
3401 | ||
3402 | /* | |
3403 | * Create the directory entry for the symlink. | |
3404 | */ | |
f6c2d1fa NS |
3405 | error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino, |
3406 | &first_block, &free_list, resblks); | |
3407 | if (error) | |
1da177e4 | 3408 | goto error1; |
1da177e4 LT |
3409 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3410 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
3411 | ||
3412 | /* | |
3413 | * Bump the in memory version number of the parent directory | |
3414 | * so that other processes accessing it will recognize that | |
3415 | * the directory has changed. | |
3416 | */ | |
3417 | dp->i_gen++; | |
3418 | ||
3419 | /* | |
3420 | * If this is a synchronous mount, make sure that the | |
3421 | * symlink transaction goes to disk before returning to | |
3422 | * the user. | |
3423 | */ | |
3424 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3425 | xfs_trans_set_sync(tp); | |
3426 | } | |
3427 | ||
3428 | /* | |
3429 | * xfs_trans_commit normally decrements the vnode ref count | |
3430 | * when it unlocks the inode. Since we want to return the | |
3431 | * vnode to the caller, we bump the vnode ref count now. | |
3432 | */ | |
3433 | IHOLD(ip); | |
3434 | ||
f7c99b6f | 3435 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3436 | if (error) { |
3437 | goto error2; | |
3438 | } | |
1c72bf90 | 3439 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3440 | XFS_QM_DQRELE(mp, udqp); |
3441 | XFS_QM_DQRELE(mp, gdqp); | |
3442 | ||
3443 | /* Fall through to std_return with error = 0 or errno from | |
3444 | * xfs_trans_commit */ | |
3445 | std_return: | |
993386c1 | 3446 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) { |
1da177e4 LT |
3447 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, |
3448 | dir_vp, DM_RIGHT_NULL, | |
3449 | error ? NULL : XFS_ITOV(ip), | |
3450 | DM_RIGHT_NULL, link_name, target_path, | |
3451 | 0, error, 0); | |
3452 | } | |
3453 | ||
3454 | if (!error) { | |
67fcaa73 | 3455 | bhv_vnode_t *vp; |
1da177e4 LT |
3456 | |
3457 | ASSERT(ip); | |
3458 | vp = XFS_ITOV(ip); | |
3459 | *vpp = vp; | |
3460 | } | |
3461 | return error; | |
3462 | ||
3463 | error2: | |
3464 | IRELE(ip); | |
3465 | error1: | |
3466 | xfs_bmap_cancel(&free_list); | |
3467 | cancel_flags |= XFS_TRANS_ABORT; | |
3468 | error_return: | |
3469 | xfs_trans_cancel(tp, cancel_flags); | |
3470 | XFS_QM_DQRELE(mp, udqp); | |
3471 | XFS_QM_DQRELE(mp, gdqp); | |
3472 | ||
993386c1 | 3473 | if (unlock_dp_on_error) |
1da177e4 | 3474 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
3475 | |
3476 | goto std_return; | |
3477 | } | |
3478 | ||
3479 | ||
3480 | /* | |
3481 | * xfs_fid2 | |
3482 | * | |
3483 | * A fid routine that takes a pointer to a previously allocated | |
3484 | * fid structure (like xfs_fast_fid) but uses a 64 bit inode number. | |
3485 | */ | |
993386c1 | 3486 | int |
1da177e4 | 3487 | xfs_fid2( |
993386c1 | 3488 | xfs_inode_t *ip, |
1da177e4 LT |
3489 | fid_t *fidp) |
3490 | { | |
993386c1 | 3491 | xfs_fid2_t *xfid = (xfs_fid2_t *)fidp; |
1da177e4 | 3492 | |
1543d79c | 3493 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
3494 | ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t)); |
3495 | ||
1da177e4 LT |
3496 | xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len); |
3497 | xfid->fid_pad = 0; | |
3498 | /* | |
3499 | * use memcpy because the inode is a long long and there's no | |
3500 | * assurance that xfid->fid_ino is properly aligned. | |
3501 | */ | |
3502 | memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino)); | |
3503 | xfid->fid_gen = ip->i_d.di_gen; | |
3504 | ||
3505 | return 0; | |
3506 | } | |
3507 | ||
3508 | ||
1da177e4 LT |
3509 | int |
3510 | xfs_rwlock( | |
993386c1 | 3511 | xfs_inode_t *ip, |
8285fb58 | 3512 | bhv_vrwlock_t locktype) |
1da177e4 | 3513 | { |
993386c1 | 3514 | if (S_ISDIR(ip->i_d.di_mode)) |
1da177e4 | 3515 | return 1; |
1da177e4 LT |
3516 | if (locktype == VRWLOCK_WRITE) { |
3517 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
3518 | } else if (locktype == VRWLOCK_TRY_READ) { | |
014c2544 | 3519 | return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED); |
1da177e4 | 3520 | } else if (locktype == VRWLOCK_TRY_WRITE) { |
014c2544 | 3521 | return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL); |
1da177e4 LT |
3522 | } else { |
3523 | ASSERT((locktype == VRWLOCK_READ) || | |
3524 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3525 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
3526 | } | |
3527 | ||
3528 | return 1; | |
3529 | } | |
3530 | ||
3531 | ||
1da177e4 LT |
3532 | void |
3533 | xfs_rwunlock( | |
993386c1 | 3534 | xfs_inode_t *ip, |
8285fb58 | 3535 | bhv_vrwlock_t locktype) |
1da177e4 | 3536 | { |
993386c1 CH |
3537 | if (S_ISDIR(ip->i_d.di_mode)) |
3538 | return; | |
1da177e4 LT |
3539 | if (locktype == VRWLOCK_WRITE) { |
3540 | /* | |
3541 | * In the write case, we may have added a new entry to | |
3542 | * the reference cache. This might store a pointer to | |
3543 | * an inode to be released in this inode. If it is there, | |
3544 | * clear the pointer and release the inode after unlocking | |
3545 | * this one. | |
3546 | */ | |
3547 | xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL); | |
3548 | } else { | |
3549 | ASSERT((locktype == VRWLOCK_READ) || | |
3550 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3551 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
3552 | } | |
3553 | return; | |
3554 | } | |
3555 | ||
993386c1 CH |
3556 | |
3557 | int | |
1da177e4 | 3558 | xfs_inode_flush( |
993386c1 | 3559 | xfs_inode_t *ip, |
1da177e4 LT |
3560 | int flags) |
3561 | { | |
993386c1 CH |
3562 | xfs_mount_t *mp = ip->i_mount; |
3563 | xfs_inode_log_item_t *iip = ip->i_itemp; | |
1da177e4 LT |
3564 | int error = 0; |
3565 | ||
1da177e4 LT |
3566 | if (XFS_FORCED_SHUTDOWN(mp)) |
3567 | return XFS_ERROR(EIO); | |
3568 | ||
3569 | /* | |
3570 | * Bypass inodes which have already been cleaned by | |
3571 | * the inode flush clustering code inside xfs_iflush | |
3572 | */ | |
3573 | if ((ip->i_update_core == 0) && | |
3574 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) | |
3575 | return 0; | |
3576 | ||
3577 | if (flags & FLUSH_LOG) { | |
3578 | if (iip && iip->ili_last_lsn) { | |
3579 | xlog_t *log = mp->m_log; | |
3580 | xfs_lsn_t sync_lsn; | |
3581 | int s, log_flags = XFS_LOG_FORCE; | |
3582 | ||
3583 | s = GRANT_LOCK(log); | |
3584 | sync_lsn = log->l_last_sync_lsn; | |
3585 | GRANT_UNLOCK(log, s); | |
3586 | ||
776a75fa LM |
3587 | if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) > 0)) { |
3588 | if (flags & FLUSH_SYNC) | |
3589 | log_flags |= XFS_LOG_SYNC; | |
3590 | error = xfs_log_force(mp, iip->ili_last_lsn, log_flags); | |
3591 | if (error) | |
3592 | return error; | |
3593 | } | |
1da177e4 | 3594 | |
776a75fa LM |
3595 | if (ip->i_update_core == 0) |
3596 | return 0; | |
1da177e4 LT |
3597 | } |
3598 | } | |
3599 | ||
3600 | /* | |
3601 | * We make this non-blocking if the inode is contended, | |
3602 | * return EAGAIN to indicate to the caller that they | |
3603 | * did not succeed. This prevents the flush path from | |
3604 | * blocking on inodes inside another operation right | |
3605 | * now, they get caught later by xfs_sync. | |
3606 | */ | |
3607 | if (flags & FLUSH_INODE) { | |
3608 | int flush_flags; | |
3609 | ||
1da177e4 LT |
3610 | if (flags & FLUSH_SYNC) { |
3611 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3612 | xfs_iflock(ip); | |
3613 | } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
3614 | if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) { | |
3615 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3616 | return EAGAIN; | |
3617 | } | |
3618 | } else { | |
3619 | return EAGAIN; | |
3620 | } | |
3621 | ||
3622 | if (flags & FLUSH_SYNC) | |
3623 | flush_flags = XFS_IFLUSH_SYNC; | |
3624 | else | |
3625 | flush_flags = XFS_IFLUSH_ASYNC; | |
3626 | ||
3627 | error = xfs_iflush(ip, flush_flags); | |
3628 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3629 | } | |
3630 | ||
3631 | return error; | |
3632 | } | |
3633 | ||
993386c1 | 3634 | |
1da177e4 | 3635 | int |
993386c1 CH |
3636 | xfs_set_dmattrs( |
3637 | xfs_inode_t *ip, | |
1da177e4 | 3638 | u_int evmask, |
993386c1 | 3639 | u_int16_t state) |
1da177e4 | 3640 | { |
993386c1 | 3641 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 3642 | xfs_trans_t *tp; |
1da177e4 LT |
3643 | int error; |
3644 | ||
3645 | if (!capable(CAP_SYS_ADMIN)) | |
3646 | return XFS_ERROR(EPERM); | |
3647 | ||
1da177e4 LT |
3648 | if (XFS_FORCED_SHUTDOWN(mp)) |
3649 | return XFS_ERROR(EIO); | |
3650 | ||
3651 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
3652 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
3653 | if (error) { | |
3654 | xfs_trans_cancel(tp, 0); | |
3655 | return error; | |
3656 | } | |
3657 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3658 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3659 | ||
3660 | ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask; | |
3661 | ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state; | |
3662 | ||
3663 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3664 | IHOLD(ip); | |
1c72bf90 | 3665 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
3666 | |
3667 | return error; | |
3668 | } | |
3669 | ||
993386c1 | 3670 | int |
1da177e4 | 3671 | xfs_reclaim( |
993386c1 | 3672 | xfs_inode_t *ip) |
1da177e4 | 3673 | { |
993386c1 | 3674 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1da177e4 | 3675 | |
1543d79c | 3676 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
3677 | |
3678 | ASSERT(!VN_MAPPED(vp)); | |
3679 | ||
3680 | /* bad inode, get out here ASAP */ | |
3681 | if (VN_BAD(vp)) { | |
3682 | xfs_ireclaim(ip); | |
3683 | return 0; | |
3684 | } | |
3685 | ||
b677c210 | 3686 | vn_iowait(ip); |
1da177e4 | 3687 | |
51c91ed5 | 3688 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); |
1da177e4 | 3689 | |
42fe2b1f CH |
3690 | /* |
3691 | * Make sure the atime in the XFS inode is correct before freeing the | |
3692 | * Linux inode. | |
3693 | */ | |
3694 | xfs_synchronize_atime(ip); | |
1da177e4 | 3695 | |
4c60658e DC |
3696 | /* |
3697 | * If we have nothing to flush with this inode then complete the | |
3698 | * teardown now, otherwise break the link between the xfs inode and the | |
3699 | * linux inode and clean up the xfs inode later. This avoids flushing | |
3700 | * the inode to disk during the delete operation itself. | |
3701 | * | |
3702 | * When breaking the link, we need to set the XFS_IRECLAIMABLE flag | |
3703 | * first to ensure that xfs_iunpin() will never see an xfs inode | |
3704 | * that has a linux inode being reclaimed. Synchronisation is provided | |
3705 | * by the i_flags_lock. | |
1da177e4 LT |
3706 | */ |
3707 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
3708 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3709 | xfs_iflock(ip); | |
3710 | return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); | |
3711 | } else { | |
3712 | xfs_mount_t *mp = ip->i_mount; | |
3713 | ||
4c60658e | 3714 | /* Protect sync and unpin from us */ |
1da177e4 | 3715 | XFS_MOUNT_ILOCK(mp); |
4c60658e DC |
3716 | spin_lock(&ip->i_flags_lock); |
3717 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
739bfb2a CH |
3718 | vn_to_inode(vp)->i_private = NULL; |
3719 | ip->i_vnode = NULL; | |
4c60658e | 3720 | spin_unlock(&ip->i_flags_lock); |
1da177e4 | 3721 | list_add_tail(&ip->i_reclaim, &mp->m_del_inodes); |
1da177e4 LT |
3722 | XFS_MOUNT_IUNLOCK(mp); |
3723 | } | |
3724 | return 0; | |
3725 | } | |
3726 | ||
3727 | int | |
3728 | xfs_finish_reclaim( | |
3729 | xfs_inode_t *ip, | |
3730 | int locked, | |
3731 | int sync_mode) | |
3732 | { | |
da353b0d | 3733 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); |
67fcaa73 | 3734 | bhv_vnode_t *vp = XFS_ITOV_NULL(ip); |
1da177e4 LT |
3735 | int error; |
3736 | ||
3737 | if (vp && VN_BAD(vp)) | |
3738 | goto reclaim; | |
3739 | ||
3740 | /* The hash lock here protects a thread in xfs_iget_core from | |
3741 | * racing with us on linking the inode back with a vnode. | |
3742 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
3743 | * us. | |
3744 | */ | |
da353b0d | 3745 | write_lock(&pag->pag_ici_lock); |
f273ab84 | 3746 | spin_lock(&ip->i_flags_lock); |
7a18c386 DC |
3747 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || |
3748 | (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) { | |
f273ab84 | 3749 | spin_unlock(&ip->i_flags_lock); |
da353b0d | 3750 | write_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
3751 | if (locked) { |
3752 | xfs_ifunlock(ip); | |
3753 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3754 | } | |
014c2544 | 3755 | return 1; |
1da177e4 | 3756 | } |
7a18c386 | 3757 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
f273ab84 | 3758 | spin_unlock(&ip->i_flags_lock); |
da353b0d DC |
3759 | write_unlock(&pag->pag_ici_lock); |
3760 | xfs_put_perag(ip->i_mount, pag); | |
1da177e4 LT |
3761 | |
3762 | /* | |
3763 | * If the inode is still dirty, then flush it out. If the inode | |
3764 | * is not in the AIL, then it will be OK to flush it delwri as | |
3765 | * long as xfs_iflush() does not keep any references to the inode. | |
3766 | * We leave that decision up to xfs_iflush() since it has the | |
3767 | * knowledge of whether it's OK to simply do a delwri flush of | |
3768 | * the inode or whether we need to wait until the inode is | |
3769 | * pulled from the AIL. | |
3770 | * We get the flush lock regardless, though, just to make sure | |
3771 | * we don't free it while it is being flushed. | |
3772 | */ | |
3773 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
3774 | if (!locked) { | |
3775 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3776 | xfs_iflock(ip); | |
3777 | } | |
3778 | ||
3779 | if (ip->i_update_core || | |
3780 | ((ip->i_itemp != NULL) && | |
3781 | (ip->i_itemp->ili_format.ilf_fields != 0))) { | |
3782 | error = xfs_iflush(ip, sync_mode); | |
3783 | /* | |
3784 | * If we hit an error, typically because of filesystem | |
3785 | * shutdown, we don't need to let vn_reclaim to know | |
3786 | * because we're gonna reclaim the inode anyway. | |
3787 | */ | |
3788 | if (error) { | |
3789 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3790 | goto reclaim; | |
3791 | } | |
3792 | xfs_iflock(ip); /* synchronize with xfs_iflush_done */ | |
3793 | } | |
3794 | ||
3795 | ASSERT(ip->i_update_core == 0); | |
3796 | ASSERT(ip->i_itemp == NULL || | |
3797 | ip->i_itemp->ili_format.ilf_fields == 0); | |
3798 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3799 | } else if (locked) { | |
3800 | /* | |
3801 | * We are not interested in doing an iflush if we're | |
3802 | * in the process of shutting down the filesystem forcibly. | |
3803 | * So, just reclaim the inode. | |
3804 | */ | |
3805 | xfs_ifunlock(ip); | |
3806 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3807 | } | |
3808 | ||
3809 | reclaim: | |
3810 | xfs_ireclaim(ip); | |
3811 | return 0; | |
3812 | } | |
3813 | ||
3814 | int | |
3815 | xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock) | |
3816 | { | |
3817 | int purged; | |
3818 | xfs_inode_t *ip, *n; | |
3819 | int done = 0; | |
3820 | ||
3821 | while (!done) { | |
3822 | purged = 0; | |
3823 | XFS_MOUNT_ILOCK(mp); | |
3824 | list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) { | |
3825 | if (noblock) { | |
3826 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) | |
3827 | continue; | |
3828 | if (xfs_ipincount(ip) || | |
3829 | !xfs_iflock_nowait(ip)) { | |
3830 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3831 | continue; | |
3832 | } | |
3833 | } | |
3834 | XFS_MOUNT_IUNLOCK(mp); | |
6b2cf618 FB |
3835 | if (xfs_finish_reclaim(ip, noblock, |
3836 | XFS_IFLUSH_DELWRI_ELSE_ASYNC)) | |
3837 | delay(1); | |
1da177e4 LT |
3838 | purged = 1; |
3839 | break; | |
3840 | } | |
3841 | ||
3842 | done = !purged; | |
3843 | } | |
3844 | ||
3845 | XFS_MOUNT_IUNLOCK(mp); | |
3846 | return 0; | |
3847 | } | |
3848 | ||
3849 | /* | |
3850 | * xfs_alloc_file_space() | |
3851 | * This routine allocates disk space for the given file. | |
3852 | * | |
3853 | * If alloc_type == 0, this request is for an ALLOCSP type | |
3854 | * request which will change the file size. In this case, no | |
3855 | * DMAPI event will be generated by the call. A TRUNCATE event | |
3856 | * will be generated later by xfs_setattr. | |
3857 | * | |
3858 | * If alloc_type != 0, this request is for a RESVSP type | |
3859 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
3860 | * lower block boundary byte address is less than the file's | |
3861 | * length. | |
3862 | * | |
3863 | * RETURNS: | |
3864 | * 0 on success | |
3865 | * errno on error | |
3866 | * | |
3867 | */ | |
ba0f32d4 | 3868 | STATIC int |
1da177e4 LT |
3869 | xfs_alloc_file_space( |
3870 | xfs_inode_t *ip, | |
3871 | xfs_off_t offset, | |
3872 | xfs_off_t len, | |
3873 | int alloc_type, | |
3874 | int attr_flags) | |
3875 | { | |
dd9f438e NS |
3876 | xfs_mount_t *mp = ip->i_mount; |
3877 | xfs_off_t count; | |
1da177e4 LT |
3878 | xfs_filblks_t allocated_fsb; |
3879 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
3880 | xfs_extlen_t extsz, temp; |
3881 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 3882 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
3883 | int nimaps; |
3884 | int bmapi_flag; | |
3885 | int quota_flag; | |
1da177e4 | 3886 | int rt; |
1da177e4 | 3887 | xfs_trans_t *tp; |
dd9f438e NS |
3888 | xfs_bmbt_irec_t imaps[1], *imapp; |
3889 | xfs_bmap_free_t free_list; | |
3890 | uint qblocks, resblks, resrtextents; | |
3891 | int committed; | |
3892 | int error; | |
1da177e4 | 3893 | |
1543d79c | 3894 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
3895 | |
3896 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3897 | return XFS_ERROR(EIO); | |
3898 | ||
1da177e4 LT |
3899 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
3900 | return error; | |
3901 | ||
3902 | if (len <= 0) | |
3903 | return XFS_ERROR(EINVAL); | |
3904 | ||
957d0ebe DC |
3905 | rt = XFS_IS_REALTIME_INODE(ip); |
3906 | extsz = xfs_get_extsz_hint(ip); | |
3907 | ||
1da177e4 | 3908 | count = len; |
1da177e4 | 3909 | imapp = &imaps[0]; |
dd9f438e NS |
3910 | nimaps = 1; |
3911 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
3912 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
3913 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
3914 | ||
3915 | /* Generate a DMAPI event if needed. */ | |
ba87ea69 | 3916 | if (alloc_type != 0 && offset < ip->i_size && |
1da177e4 | 3917 | (attr_flags&ATTR_DMI) == 0 && |
eb9df39d | 3918 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
1da177e4 LT |
3919 | xfs_off_t end_dmi_offset; |
3920 | ||
3921 | end_dmi_offset = offset+len; | |
ba87ea69 LM |
3922 | if (end_dmi_offset > ip->i_size) |
3923 | end_dmi_offset = ip->i_size; | |
1da177e4 LT |
3924 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip), |
3925 | offset, end_dmi_offset - offset, | |
3926 | 0, NULL); | |
3927 | if (error) | |
014c2544 | 3928 | return error; |
1da177e4 LT |
3929 | } |
3930 | ||
3931 | /* | |
dd9f438e | 3932 | * Allocate file space until done or until there is an error |
1da177e4 LT |
3933 | */ |
3934 | retry: | |
3935 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
3936 | xfs_fileoff_t s, e; |
3937 | ||
1da177e4 | 3938 | /* |
3ddb8fa9 | 3939 | * Determine space reservations for data/realtime. |
1da177e4 | 3940 | */ |
dd9f438e | 3941 | if (unlikely(extsz)) { |
1da177e4 | 3942 | s = startoffset_fsb; |
dd9f438e NS |
3943 | do_div(s, extsz); |
3944 | s *= extsz; | |
3945 | e = startoffset_fsb + allocatesize_fsb; | |
3946 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
3947 | e += temp; | |
3948 | if ((temp = do_mod(e, extsz))) | |
3949 | e += extsz - temp; | |
3950 | } else { | |
3951 | s = 0; | |
3952 | e = allocatesize_fsb; | |
3953 | } | |
3954 | ||
3955 | if (unlikely(rt)) { | |
3956 | resrtextents = qblocks = (uint)(e - s); | |
3957 | resrtextents /= mp->m_sb.sb_rextsize; | |
3958 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
3959 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 3960 | } else { |
dd9f438e NS |
3961 | resrtextents = 0; |
3962 | resblks = qblocks = \ | |
3963 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
3964 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
3965 | } |
3966 | ||
3967 | /* | |
dd9f438e | 3968 | * Allocate and setup the transaction. |
1da177e4 LT |
3969 | */ |
3970 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
3971 | error = xfs_trans_reserve(tp, resblks, |
3972 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
3973 | XFS_TRANS_PERM_LOG_RES, |
3974 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 3975 | /* |
dd9f438e | 3976 | * Check for running out of space |
1da177e4 LT |
3977 | */ |
3978 | if (error) { | |
3979 | /* | |
3980 | * Free the transaction structure. | |
3981 | */ | |
3982 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
3983 | xfs_trans_cancel(tp, 0); | |
3984 | break; | |
3985 | } | |
3986 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
dd9f438e NS |
3987 | error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, |
3988 | qblocks, 0, quota_flag); | |
1da177e4 LT |
3989 | if (error) |
3990 | goto error1; | |
3991 | ||
3992 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3993 | xfs_trans_ihold(tp, ip); | |
3994 | ||
3995 | /* | |
dd9f438e | 3996 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 LT |
3997 | */ |
3998 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
3e57ecf6 | 3999 | error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb, |
dd9f438e NS |
4000 | allocatesize_fsb, bmapi_flag, |
4001 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 4002 | &free_list, NULL); |
1da177e4 LT |
4003 | if (error) { |
4004 | goto error0; | |
4005 | } | |
4006 | ||
4007 | /* | |
dd9f438e | 4008 | * Complete the transaction |
1da177e4 | 4009 | */ |
f7c99b6f | 4010 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
4011 | if (error) { |
4012 | goto error0; | |
4013 | } | |
4014 | ||
1c72bf90 | 4015 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
4016 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
4017 | if (error) { | |
4018 | break; | |
4019 | } | |
4020 | ||
4021 | allocated_fsb = imapp->br_blockcount; | |
4022 | ||
dd9f438e | 4023 | if (nimaps == 0) { |
1da177e4 LT |
4024 | error = XFS_ERROR(ENOSPC); |
4025 | break; | |
4026 | } | |
4027 | ||
4028 | startoffset_fsb += allocated_fsb; | |
4029 | allocatesize_fsb -= allocated_fsb; | |
4030 | } | |
4031 | dmapi_enospc_check: | |
eb9df39d CH |
4032 | if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 && |
4033 | DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) { | |
1da177e4 LT |
4034 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, |
4035 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
4036 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
4037 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ | |
4038 | if (error == 0) | |
4039 | goto retry; /* Maybe DMAPI app. has made space */ | |
4040 | /* else fall through with error from XFS_SEND_DATA */ | |
4041 | } | |
4042 | ||
4043 | return error; | |
4044 | ||
dd9f438e | 4045 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 4046 | xfs_bmap_cancel(&free_list); |
dd9f438e NS |
4047 | XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag); |
4048 | ||
4049 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
4050 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
4051 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4052 | goto dmapi_enospc_check; | |
4053 | } | |
4054 | ||
4055 | /* | |
4056 | * Zero file bytes between startoff and endoff inclusive. | |
4057 | * The iolock is held exclusive and no blocks are buffered. | |
4058 | */ | |
4059 | STATIC int | |
4060 | xfs_zero_remaining_bytes( | |
4061 | xfs_inode_t *ip, | |
4062 | xfs_off_t startoff, | |
4063 | xfs_off_t endoff) | |
4064 | { | |
4065 | xfs_bmbt_irec_t imap; | |
4066 | xfs_fileoff_t offset_fsb; | |
4067 | xfs_off_t lastoffset; | |
4068 | xfs_off_t offset; | |
4069 | xfs_buf_t *bp; | |
4070 | xfs_mount_t *mp = ip->i_mount; | |
4071 | int nimap; | |
4072 | int error = 0; | |
4073 | ||
4074 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, | |
4075 | ip->i_d.di_flags & XFS_DIFLAG_REALTIME ? | |
4076 | mp->m_rtdev_targp : mp->m_ddev_targp); | |
4077 | ||
4078 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
4079 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
4080 | nimap = 1; | |
3e57ecf6 OW |
4081 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0, |
4082 | NULL, 0, &imap, &nimap, NULL, NULL); | |
1da177e4 LT |
4083 | if (error || nimap < 1) |
4084 | break; | |
4085 | ASSERT(imap.br_blockcount >= 1); | |
4086 | ASSERT(imap.br_startoff == offset_fsb); | |
4087 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
4088 | if (lastoffset > endoff) | |
4089 | lastoffset = endoff; | |
4090 | if (imap.br_startblock == HOLESTARTBLOCK) | |
4091 | continue; | |
4092 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4093 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
4094 | continue; | |
4095 | XFS_BUF_UNDONE(bp); | |
4096 | XFS_BUF_UNWRITE(bp); | |
4097 | XFS_BUF_READ(bp); | |
4098 | XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock)); | |
4099 | xfsbdstrat(mp, bp); | |
4100 | if ((error = xfs_iowait(bp))) { | |
4101 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", | |
4102 | mp, bp, XFS_BUF_ADDR(bp)); | |
4103 | break; | |
4104 | } | |
4105 | memset(XFS_BUF_PTR(bp) + | |
4106 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
4107 | 0, lastoffset - offset + 1); | |
4108 | XFS_BUF_UNDONE(bp); | |
4109 | XFS_BUF_UNREAD(bp); | |
4110 | XFS_BUF_WRITE(bp); | |
4111 | xfsbdstrat(mp, bp); | |
4112 | if ((error = xfs_iowait(bp))) { | |
4113 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", | |
4114 | mp, bp, XFS_BUF_ADDR(bp)); | |
4115 | break; | |
4116 | } | |
4117 | } | |
4118 | xfs_buf_free(bp); | |
4119 | return error; | |
4120 | } | |
4121 | ||
4122 | /* | |
4123 | * xfs_free_file_space() | |
4124 | * This routine frees disk space for the given file. | |
4125 | * | |
4126 | * This routine is only called by xfs_change_file_space | |
4127 | * for an UNRESVSP type call. | |
4128 | * | |
4129 | * RETURNS: | |
4130 | * 0 on success | |
4131 | * errno on error | |
4132 | * | |
4133 | */ | |
4134 | STATIC int | |
4135 | xfs_free_file_space( | |
4136 | xfs_inode_t *ip, | |
4137 | xfs_off_t offset, | |
4138 | xfs_off_t len, | |
4139 | int attr_flags) | |
4140 | { | |
67fcaa73 | 4141 | bhv_vnode_t *vp; |
1da177e4 LT |
4142 | int committed; |
4143 | int done; | |
4144 | xfs_off_t end_dmi_offset; | |
4145 | xfs_fileoff_t endoffset_fsb; | |
4146 | int error; | |
4147 | xfs_fsblock_t firstfsb; | |
4148 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
4149 | xfs_bmbt_irec_t imap; |
4150 | xfs_off_t ioffset; | |
4151 | xfs_extlen_t mod=0; | |
4152 | xfs_mount_t *mp; | |
4153 | int nimap; | |
4154 | uint resblks; | |
673cdf5c | 4155 | uint rounding; |
1da177e4 LT |
4156 | int rt; |
4157 | xfs_fileoff_t startoffset_fsb; | |
4158 | xfs_trans_t *tp; | |
5fcbab35 | 4159 | int need_iolock = 1; |
1da177e4 | 4160 | |
bd5a876a | 4161 | vp = XFS_ITOV(ip); |
1da177e4 LT |
4162 | mp = ip->i_mount; |
4163 | ||
1543d79c | 4164 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
bd5a876a | 4165 | |
1da177e4 LT |
4166 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
4167 | return error; | |
4168 | ||
4169 | error = 0; | |
4170 | if (len <= 0) /* if nothing being freed */ | |
4171 | return error; | |
4172 | rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME); | |
4173 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); | |
4174 | end_dmi_offset = offset + len; | |
4175 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
4176 | ||
eb9df39d CH |
4177 | if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 && |
4178 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { | |
ba87ea69 LM |
4179 | if (end_dmi_offset > ip->i_size) |
4180 | end_dmi_offset = ip->i_size; | |
bd5a876a | 4181 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp, |
1da177e4 LT |
4182 | offset, end_dmi_offset - offset, |
4183 | AT_DELAY_FLAG(attr_flags), NULL); | |
4184 | if (error) | |
014c2544 | 4185 | return error; |
1da177e4 LT |
4186 | } |
4187 | ||
5fcbab35 DR |
4188 | if (attr_flags & ATTR_NOLOCK) |
4189 | need_iolock = 0; | |
9fa8046f | 4190 | if (need_iolock) { |
1da177e4 | 4191 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
b677c210 | 4192 | vn_iowait(ip); /* wait for the completion of any pending DIOs */ |
9fa8046f | 4193 | } |
5fcbab35 | 4194 | |
673cdf5c | 4195 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, NBPP); |
1da177e4 | 4196 | ioffset = offset & ~(rounding - 1); |
bd5a876a CH |
4197 | |
4198 | if (VN_CACHED(vp) != 0) { | |
4199 | xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1, | |
4200 | ctooff(offtoct(ioffset)), -1); | |
739bfb2a CH |
4201 | error = xfs_flushinval_pages(ip, |
4202 | ctooff(offtoct(ioffset)), | |
bd5a876a | 4203 | -1, FI_REMAPF_LOCKED); |
d3cf2094 LM |
4204 | if (error) |
4205 | goto out_unlock_iolock; | |
bd5a876a CH |
4206 | } |
4207 | ||
1da177e4 LT |
4208 | /* |
4209 | * Need to zero the stuff we're not freeing, on disk. | |
4210 | * If its a realtime file & can't use unwritten extents then we | |
4211 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
4212 | * will take care of it for us. | |
4213 | */ | |
4214 | if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) { | |
4215 | nimap = 1; | |
3e57ecf6 OW |
4216 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb, |
4217 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); | |
1da177e4 LT |
4218 | if (error) |
4219 | goto out_unlock_iolock; | |
4220 | ASSERT(nimap == 0 || nimap == 1); | |
4221 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4222 | xfs_daddr_t block; | |
4223 | ||
4224 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4225 | block = imap.br_startblock; | |
4226 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
4227 | if (mod) | |
4228 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
4229 | } | |
4230 | nimap = 1; | |
3e57ecf6 OW |
4231 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1, |
4232 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); | |
1da177e4 LT |
4233 | if (error) |
4234 | goto out_unlock_iolock; | |
4235 | ASSERT(nimap == 0 || nimap == 1); | |
4236 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4237 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4238 | mod++; | |
4239 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
4240 | endoffset_fsb -= mod; | |
4241 | } | |
4242 | } | |
4243 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
4244 | /* | |
4245 | * One contiguous piece to clear | |
4246 | */ | |
4247 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
4248 | else { | |
4249 | /* | |
4250 | * Some full blocks, possibly two pieces to clear | |
4251 | */ | |
4252 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
4253 | error = xfs_zero_remaining_bytes(ip, offset, | |
4254 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
4255 | if (!error && | |
4256 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
4257 | error = xfs_zero_remaining_bytes(ip, | |
4258 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
4259 | offset + len - 1); | |
4260 | } | |
4261 | ||
4262 | /* | |
4263 | * free file space until done or until there is an error | |
4264 | */ | |
4265 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
4266 | while (!error && !done) { | |
4267 | ||
4268 | /* | |
91ebecc7 DC |
4269 | * allocate and setup the transaction. Allow this |
4270 | * transaction to dip into the reserve blocks to ensure | |
4271 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
4272 | */ |
4273 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 4274 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
4275 | error = xfs_trans_reserve(tp, |
4276 | resblks, | |
4277 | XFS_WRITE_LOG_RES(mp), | |
4278 | 0, | |
4279 | XFS_TRANS_PERM_LOG_RES, | |
4280 | XFS_WRITE_LOG_COUNT); | |
4281 | ||
4282 | /* | |
4283 | * check for running out of space | |
4284 | */ | |
4285 | if (error) { | |
4286 | /* | |
4287 | * Free the transaction structure. | |
4288 | */ | |
4289 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
4290 | xfs_trans_cancel(tp, 0); | |
4291 | break; | |
4292 | } | |
4293 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4294 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, | |
dd9f438e NS |
4295 | ip->i_udquot, ip->i_gdquot, resblks, 0, |
4296 | XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
4297 | if (error) |
4298 | goto error1; | |
4299 | ||
4300 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4301 | xfs_trans_ihold(tp, ip); | |
4302 | ||
4303 | /* | |
4304 | * issue the bunmapi() call to free the blocks | |
4305 | */ | |
4306 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
3e57ecf6 | 4307 | error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb, |
1da177e4 | 4308 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 4309 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
4310 | if (error) { |
4311 | goto error0; | |
4312 | } | |
4313 | ||
4314 | /* | |
4315 | * complete the transaction | |
4316 | */ | |
f7c99b6f | 4317 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
4318 | if (error) { |
4319 | goto error0; | |
4320 | } | |
4321 | ||
1c72bf90 | 4322 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
4323 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
4324 | } | |
4325 | ||
4326 | out_unlock_iolock: | |
4327 | if (need_iolock) | |
4328 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
4329 | return error; | |
4330 | ||
4331 | error0: | |
4332 | xfs_bmap_cancel(&free_list); | |
4333 | error1: | |
4334 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
4335 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
4336 | XFS_ILOCK_EXCL); | |
4337 | return error; | |
4338 | } | |
4339 | ||
4340 | /* | |
4341 | * xfs_change_file_space() | |
4342 | * This routine allocates or frees disk space for the given file. | |
4343 | * The user specified parameters are checked for alignment and size | |
4344 | * limitations. | |
4345 | * | |
4346 | * RETURNS: | |
4347 | * 0 on success | |
4348 | * errno on error | |
4349 | * | |
4350 | */ | |
4351 | int | |
4352 | xfs_change_file_space( | |
993386c1 | 4353 | xfs_inode_t *ip, |
1da177e4 LT |
4354 | int cmd, |
4355 | xfs_flock64_t *bf, | |
4356 | xfs_off_t offset, | |
4357 | cred_t *credp, | |
4358 | int attr_flags) | |
4359 | { | |
993386c1 | 4360 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
4361 | int clrprealloc; |
4362 | int error; | |
4363 | xfs_fsize_t fsize; | |
1da177e4 LT |
4364 | int setprealloc; |
4365 | xfs_off_t startoffset; | |
4366 | xfs_off_t llen; | |
4367 | xfs_trans_t *tp; | |
8285fb58 | 4368 | bhv_vattr_t va; |
1da177e4 | 4369 | |
1543d79c | 4370 | vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
4371 | |
4372 | /* | |
4373 | * must be a regular file and have write permission | |
4374 | */ | |
993386c1 | 4375 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
4376 | return XFS_ERROR(EINVAL); |
4377 | ||
4378 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
4379 | ||
4380 | if ((error = xfs_iaccess(ip, S_IWUSR, credp))) { | |
4381 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
4382 | return error; | |
4383 | } | |
4384 | ||
4385 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
4386 | ||
4387 | switch (bf->l_whence) { | |
4388 | case 0: /*SEEK_SET*/ | |
4389 | break; | |
4390 | case 1: /*SEEK_CUR*/ | |
4391 | bf->l_start += offset; | |
4392 | break; | |
4393 | case 2: /*SEEK_END*/ | |
ba87ea69 | 4394 | bf->l_start += ip->i_size; |
1da177e4 LT |
4395 | break; |
4396 | default: | |
4397 | return XFS_ERROR(EINVAL); | |
4398 | } | |
4399 | ||
4400 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
4401 | ||
4402 | if ( (bf->l_start < 0) | |
4403 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
4404 | || (bf->l_start + llen < 0) | |
4405 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
4406 | return XFS_ERROR(EINVAL); | |
4407 | ||
4408 | bf->l_whence = 0; | |
4409 | ||
4410 | startoffset = bf->l_start; | |
ba87ea69 | 4411 | fsize = ip->i_size; |
1da177e4 LT |
4412 | |
4413 | /* | |
4414 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
4415 | * file space. | |
4416 | * These calls do NOT zero the data space allocated to the file, | |
4417 | * nor do they change the file size. | |
4418 | * | |
4419 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
4420 | * space. | |
4421 | * These calls cause the new file data to be zeroed and the file | |
4422 | * size to be changed. | |
4423 | */ | |
4424 | setprealloc = clrprealloc = 0; | |
4425 | ||
4426 | switch (cmd) { | |
4427 | case XFS_IOC_RESVSP: | |
4428 | case XFS_IOC_RESVSP64: | |
4429 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
4430 | 1, attr_flags); | |
4431 | if (error) | |
4432 | return error; | |
4433 | setprealloc = 1; | |
4434 | break; | |
4435 | ||
4436 | case XFS_IOC_UNRESVSP: | |
4437 | case XFS_IOC_UNRESVSP64: | |
4438 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
4439 | attr_flags))) | |
4440 | return error; | |
4441 | break; | |
4442 | ||
4443 | case XFS_IOC_ALLOCSP: | |
4444 | case XFS_IOC_ALLOCSP64: | |
4445 | case XFS_IOC_FREESP: | |
4446 | case XFS_IOC_FREESP64: | |
4447 | if (startoffset > fsize) { | |
4448 | error = xfs_alloc_file_space(ip, fsize, | |
4449 | startoffset - fsize, 0, attr_flags); | |
4450 | if (error) | |
4451 | break; | |
4452 | } | |
4453 | ||
4454 | va.va_mask = XFS_AT_SIZE; | |
4455 | va.va_size = startoffset; | |
4456 | ||
993386c1 | 4457 | error = xfs_setattr(ip, &va, attr_flags, credp); |
1da177e4 LT |
4458 | |
4459 | if (error) | |
4460 | return error; | |
4461 | ||
4462 | clrprealloc = 1; | |
4463 | break; | |
4464 | ||
4465 | default: | |
4466 | ASSERT(0); | |
4467 | return XFS_ERROR(EINVAL); | |
4468 | } | |
4469 | ||
4470 | /* | |
4471 | * update the inode timestamp, mode, and prealloc flag bits | |
4472 | */ | |
4473 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
4474 | ||
4475 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
4476 | 0, 0, 0))) { | |
4477 | /* ASSERT(0); */ | |
4478 | xfs_trans_cancel(tp, 0); | |
4479 | return error; | |
4480 | } | |
4481 | ||
4482 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4483 | ||
4484 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4485 | xfs_trans_ihold(tp, ip); | |
4486 | ||
4487 | if ((attr_flags & ATTR_DMI) == 0) { | |
4488 | ip->i_d.di_mode &= ~S_ISUID; | |
4489 | ||
4490 | /* | |
4491 | * Note that we don't have to worry about mandatory | |
4492 | * file locking being disabled here because we only | |
4493 | * clear the S_ISGID bit if the Group execute bit is | |
4494 | * on, but if it was on then mandatory locking wouldn't | |
4495 | * have been enabled. | |
4496 | */ | |
4497 | if (ip->i_d.di_mode & S_IXGRP) | |
4498 | ip->i_d.di_mode &= ~S_ISGID; | |
4499 | ||
4500 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
4501 | } | |
4502 | if (setprealloc) | |
4503 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
4504 | else if (clrprealloc) | |
4505 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
4506 | ||
4507 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
4508 | xfs_trans_set_sync(tp); | |
4509 | ||
1c72bf90 | 4510 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
4511 | |
4512 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4513 | ||
4514 | return error; | |
4515 | } |