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 | |
1da177e4 LT |
91 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
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 | |
1da177e4 LT |
231 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
232 | ||
233 | if (vp->v_vfsp->vfs_flag & VFS_RDONLY) | |
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)) { | |
595 | code = bhv_vop_flush_pages(XFS_ITOV(ip), | |
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 */ | |
601 | vn_iowait(vp); | |
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 | */ | |
663 | VTRUNCATE(vp); | |
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); | |
867 | if (mandlock_before != mandlock_after) { | |
67fcaa73 | 868 | bhv_vop_vnode_change(vp, VCHANGE_FLAGS_ENF_LOCKING, |
1da177e4 LT |
869 | mandlock_after); |
870 | } | |
871 | ||
872 | xfs_iunlock(ip, lock_flags); | |
873 | ||
874 | /* | |
875 | * Release any dquot(s) the inode had kept before chown. | |
876 | */ | |
877 | XFS_QM_DQRELE(mp, olddquot1); | |
878 | XFS_QM_DQRELE(mp, olddquot2); | |
879 | XFS_QM_DQRELE(mp, udqp); | |
880 | XFS_QM_DQRELE(mp, gdqp); | |
881 | ||
882 | if (code) { | |
883 | return code; | |
884 | } | |
885 | ||
eb9df39d | 886 | if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) && |
1da177e4 LT |
887 | !(flags & ATTR_DMI)) { |
888 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL, | |
889 | NULL, DM_RIGHT_NULL, NULL, NULL, | |
890 | 0, 0, AT_DELAY_FLAG(flags)); | |
891 | } | |
892 | return 0; | |
893 | ||
894 | abort_return: | |
895 | commit_flags |= XFS_TRANS_ABORT; | |
896 | /* FALLTHROUGH */ | |
897 | error_return: | |
898 | XFS_QM_DQRELE(mp, udqp); | |
899 | XFS_QM_DQRELE(mp, gdqp); | |
900 | if (tp) { | |
901 | xfs_trans_cancel(tp, commit_flags); | |
902 | } | |
903 | if (lock_flags != 0) { | |
904 | xfs_iunlock(ip, lock_flags); | |
905 | } | |
906 | return code; | |
907 | } | |
908 | ||
909 | ||
910 | /* | |
911 | * xfs_access | |
912 | * Null conversion from vnode mode bits to inode mode bits, as in efs. | |
913 | */ | |
993386c1 | 914 | int |
1da177e4 | 915 | xfs_access( |
993386c1 | 916 | xfs_inode_t *ip, |
1da177e4 LT |
917 | int mode, |
918 | cred_t *credp) | |
919 | { | |
1da177e4 LT |
920 | int error; |
921 | ||
993386c1 | 922 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); |
1da177e4 | 923 | |
1da177e4 LT |
924 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
925 | error = xfs_iaccess(ip, mode, credp); | |
926 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
927 | return error; | |
928 | } | |
929 | ||
930 | ||
7d4fb40a NS |
931 | /* |
932 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
933 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
934 | * bmapi. | |
935 | */ | |
936 | #define SYMLINK_MAPS 2 | |
937 | ||
804c83c3 CH |
938 | STATIC int |
939 | xfs_readlink_bmap( | |
940 | xfs_inode_t *ip, | |
941 | char *link) | |
942 | { | |
943 | xfs_mount_t *mp = ip->i_mount; | |
944 | int pathlen = ip->i_d.di_size; | |
945 | int nmaps = SYMLINK_MAPS; | |
946 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
947 | xfs_daddr_t d; | |
948 | int byte_cnt; | |
949 | int n; | |
950 | xfs_buf_t *bp; | |
951 | int error = 0; | |
952 | ||
953 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0, | |
954 | mval, &nmaps, NULL, NULL); | |
955 | if (error) | |
956 | goto out; | |
957 | ||
958 | for (n = 0; n < nmaps; n++) { | |
959 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
960 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
961 | ||
962 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); | |
963 | error = XFS_BUF_GETERROR(bp); | |
964 | if (error) { | |
965 | xfs_ioerror_alert("xfs_readlink", | |
966 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
967 | xfs_buf_relse(bp); | |
968 | goto out; | |
969 | } | |
970 | if (pathlen < byte_cnt) | |
971 | byte_cnt = pathlen; | |
972 | pathlen -= byte_cnt; | |
973 | ||
974 | memcpy(link, XFS_BUF_PTR(bp), byte_cnt); | |
975 | xfs_buf_relse(bp); | |
976 | } | |
977 | ||
978 | link[ip->i_d.di_size] = '\0'; | |
979 | error = 0; | |
980 | ||
981 | out: | |
982 | return error; | |
983 | } | |
984 | ||
993386c1 | 985 | int |
1da177e4 | 986 | xfs_readlink( |
993386c1 | 987 | xfs_inode_t *ip, |
804c83c3 | 988 | char *link) |
1da177e4 | 989 | { |
804c83c3 | 990 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 991 | int pathlen; |
1da177e4 | 992 | int error = 0; |
1da177e4 | 993 | |
804c83c3 | 994 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
995 | |
996 | if (XFS_FORCED_SHUTDOWN(mp)) | |
997 | return XFS_ERROR(EIO); | |
998 | ||
999 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1000 | ||
1001 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
804c83c3 | 1002 | ASSERT(ip->i_d.di_size <= MAXPATHLEN); |
1da177e4 | 1003 | |
804c83c3 CH |
1004 | pathlen = ip->i_d.di_size; |
1005 | if (!pathlen) | |
1006 | goto out; | |
1da177e4 LT |
1007 | |
1008 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
804c83c3 CH |
1009 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
1010 | link[pathlen] = '\0'; | |
1011 | } else { | |
1012 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
1013 | } |
1014 | ||
804c83c3 | 1015 | out: |
1da177e4 | 1016 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
1017 | return error; |
1018 | } | |
1019 | ||
1da177e4 LT |
1020 | /* |
1021 | * xfs_fsync | |
1022 | * | |
1023 | * This is called to sync the inode and its data out to disk. | |
1024 | * We need to hold the I/O lock while flushing the data, and | |
1025 | * the inode lock while flushing the inode. The inode lock CANNOT | |
1026 | * be held while flushing the data, so acquire after we're done | |
1027 | * with that. | |
1028 | */ | |
993386c1 | 1029 | int |
1da177e4 | 1030 | xfs_fsync( |
993386c1 | 1031 | xfs_inode_t *ip, |
1da177e4 | 1032 | int flag, |
1da177e4 LT |
1033 | xfs_off_t start, |
1034 | xfs_off_t stop) | |
1035 | { | |
1da177e4 LT |
1036 | xfs_trans_t *tp; |
1037 | int error; | |
f538d4da | 1038 | int log_flushed = 0, changed = 1; |
1da177e4 | 1039 | |
993386c1 | 1040 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
1041 | |
1042 | ASSERT(start >= 0 && stop >= -1); | |
1043 | ||
1044 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
1045 | return XFS_ERROR(EIO); | |
1046 | ||
776a75fa LM |
1047 | if (flag & FSYNC_DATA) |
1048 | filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping); | |
1049 | ||
1da177e4 LT |
1050 | /* |
1051 | * We always need to make sure that the required inode state | |
1052 | * is safe on disk. The vnode might be clean but because | |
1053 | * of committed transactions that haven't hit the disk yet. | |
1054 | * Likewise, there could be unflushed non-transactional | |
1055 | * changes to the inode core that have to go to disk. | |
1056 | * | |
1057 | * The following code depends on one assumption: that | |
1058 | * any transaction that changes an inode logs the core | |
1059 | * because it has to change some field in the inode core | |
1060 | * (typically nextents or nblocks). That assumption | |
1061 | * implies that any transactions against an inode will | |
1062 | * catch any non-transactional updates. If inode-altering | |
1063 | * transactions exist that violate this assumption, the | |
1064 | * code breaks. Right now, it figures that if the involved | |
1065 | * update_* field is clear and the inode is unpinned, the | |
1066 | * inode is clean. Either it's been flushed or it's been | |
1067 | * committed and the commit has hit the disk unpinning the inode. | |
1068 | * (Note that xfs_inode_item_format() called at commit clears | |
1069 | * the update_* fields.) | |
1070 | */ | |
1071 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1072 | ||
1073 | /* If we are flushing data then we care about update_size | |
1074 | * being set, otherwise we care about update_core | |
1075 | */ | |
1076 | if ((flag & FSYNC_DATA) ? | |
1077 | (ip->i_update_size == 0) : | |
1078 | (ip->i_update_core == 0)) { | |
1079 | /* | |
1080 | * Timestamps/size haven't changed since last inode | |
1081 | * flush or inode transaction commit. That means | |
1082 | * either nothing got written or a transaction | |
1083 | * committed which caught the updates. If the | |
1084 | * latter happened and the transaction hasn't | |
1085 | * hit the disk yet, the inode will be still | |
1086 | * be pinned. If it is, force the log. | |
1087 | */ | |
1088 | ||
1089 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1090 | ||
1091 | if (xfs_ipincount(ip)) { | |
f538d4da | 1092 | _xfs_log_force(ip->i_mount, (xfs_lsn_t)0, |
1da177e4 LT |
1093 | XFS_LOG_FORCE | |
1094 | ((flag & FSYNC_WAIT) | |
f538d4da CH |
1095 | ? XFS_LOG_SYNC : 0), |
1096 | &log_flushed); | |
1097 | } else { | |
1098 | /* | |
1099 | * If the inode is not pinned and nothing | |
1100 | * has changed we don't need to flush the | |
1101 | * cache. | |
1102 | */ | |
1103 | changed = 0; | |
1da177e4 LT |
1104 | } |
1105 | error = 0; | |
1106 | } else { | |
1107 | /* | |
1108 | * Kick off a transaction to log the inode | |
1109 | * core to get the updates. Make it | |
1110 | * sync if FSYNC_WAIT is passed in (which | |
1111 | * is done by everybody but specfs). The | |
1112 | * sync transaction will also force the log. | |
1113 | */ | |
1114 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1115 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
1116 | if ((error = xfs_trans_reserve(tp, 0, | |
1117 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), | |
1118 | 0, 0, 0))) { | |
1119 | xfs_trans_cancel(tp, 0); | |
1120 | return error; | |
1121 | } | |
1122 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1123 | ||
1124 | /* | |
1125 | * Note - it's possible that we might have pushed | |
1126 | * ourselves out of the way during trans_reserve | |
1127 | * which would flush the inode. But there's no | |
1128 | * guarantee that the inode buffer has actually | |
1129 | * gone out yet (it's delwri). Plus the buffer | |
1130 | * could be pinned anyway if it's part of an | |
1131 | * inode in another recent transaction. So we | |
1132 | * play it safe and fire off the transaction anyway. | |
1133 | */ | |
1134 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1135 | xfs_trans_ihold(tp, ip); | |
1136 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1137 | if (flag & FSYNC_WAIT) | |
1138 | xfs_trans_set_sync(tp); | |
1c72bf90 | 1139 | error = _xfs_trans_commit(tp, 0, &log_flushed); |
1da177e4 LT |
1140 | |
1141 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1142 | } | |
f538d4da CH |
1143 | |
1144 | if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) { | |
1145 | /* | |
1146 | * If the log write didn't issue an ordered tag we need | |
1147 | * to flush the disk cache for the data device now. | |
1148 | */ | |
1149 | if (!log_flushed) | |
1150 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
1151 | ||
1152 | /* | |
1153 | * If this inode is on the RT dev we need to flush that | |
c41564b5 | 1154 | * cache as well. |
f538d4da CH |
1155 | */ |
1156 | if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) | |
1157 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); | |
1158 | } | |
1159 | ||
1da177e4 LT |
1160 | return error; |
1161 | } | |
1162 | ||
1163 | /* | |
92dfe8d2 DC |
1164 | * This is called by xfs_inactive to free any blocks beyond eof |
1165 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
1166 | * punching a hole to EOF. | |
1da177e4 | 1167 | */ |
92dfe8d2 DC |
1168 | int |
1169 | xfs_free_eofblocks( | |
1da177e4 | 1170 | xfs_mount_t *mp, |
92dfe8d2 DC |
1171 | xfs_inode_t *ip, |
1172 | int flags) | |
1da177e4 LT |
1173 | { |
1174 | xfs_trans_t *tp; | |
1175 | int error; | |
1176 | xfs_fileoff_t end_fsb; | |
1177 | xfs_fileoff_t last_fsb; | |
1178 | xfs_filblks_t map_len; | |
1179 | int nimaps; | |
1180 | xfs_bmbt_irec_t imap; | |
92dfe8d2 | 1181 | int use_iolock = (flags & XFS_FREE_EOF_LOCK); |
1da177e4 LT |
1182 | |
1183 | /* | |
1184 | * Figure out if there are any blocks beyond the end | |
1185 | * of the file. If not, then there is nothing to do. | |
1186 | */ | |
ba87ea69 | 1187 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size)); |
1da177e4 LT |
1188 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); |
1189 | map_len = last_fsb - end_fsb; | |
1190 | if (map_len <= 0) | |
014c2544 | 1191 | return 0; |
1da177e4 LT |
1192 | |
1193 | nimaps = 1; | |
1194 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3e57ecf6 OW |
1195 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0, |
1196 | NULL, 0, &imap, &nimaps, NULL, NULL); | |
1da177e4 LT |
1197 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1198 | ||
1199 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
1200 | (imap.br_startblock != HOLESTARTBLOCK || |
1201 | ip->i_delayed_blks)) { | |
1da177e4 LT |
1202 | /* |
1203 | * Attach the dquots to the inode up front. | |
1204 | */ | |
1205 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1206 | return error; |
1da177e4 LT |
1207 | |
1208 | /* | |
1209 | * There are blocks after the end of file. | |
1210 | * Free them up now by truncating the file to | |
1211 | * its current size. | |
1212 | */ | |
1213 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1214 | ||
1215 | /* | |
1216 | * Do the xfs_itruncate_start() call before | |
1217 | * reserving any log space because | |
1218 | * itruncate_start will call into the buffer | |
1219 | * cache and we can't | |
1220 | * do that within a transaction. | |
1221 | */ | |
92dfe8d2 DC |
1222 | if (use_iolock) |
1223 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 | 1224 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, |
ba87ea69 | 1225 | ip->i_size); |
d3cf2094 | 1226 | if (error) { |
87ae3c24 | 1227 | xfs_trans_cancel(tp, 0); |
92dfe8d2 DC |
1228 | if (use_iolock) |
1229 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 LM |
1230 | return error; |
1231 | } | |
1da177e4 LT |
1232 | |
1233 | error = xfs_trans_reserve(tp, 0, | |
1234 | XFS_ITRUNCATE_LOG_RES(mp), | |
1235 | 0, XFS_TRANS_PERM_LOG_RES, | |
1236 | XFS_ITRUNCATE_LOG_COUNT); | |
1237 | if (error) { | |
1238 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1239 | xfs_trans_cancel(tp, 0); | |
1240 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1241 | return error; |
1da177e4 LT |
1242 | } |
1243 | ||
1244 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1245 | xfs_trans_ijoin(tp, ip, | |
1246 | XFS_IOLOCK_EXCL | | |
1247 | XFS_ILOCK_EXCL); | |
1248 | xfs_trans_ihold(tp, ip); | |
1249 | ||
1250 | error = xfs_itruncate_finish(&tp, ip, | |
ba87ea69 | 1251 | ip->i_size, |
1da177e4 LT |
1252 | XFS_DATA_FORK, |
1253 | 0); | |
1254 | /* | |
1255 | * If we get an error at this point we | |
1256 | * simply don't bother truncating the file. | |
1257 | */ | |
1258 | if (error) { | |
1259 | xfs_trans_cancel(tp, | |
1260 | (XFS_TRANS_RELEASE_LOG_RES | | |
1261 | XFS_TRANS_ABORT)); | |
1262 | } else { | |
1263 | error = xfs_trans_commit(tp, | |
1c72bf90 | 1264 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 1265 | } |
92dfe8d2 DC |
1266 | xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL) |
1267 | : XFS_ILOCK_EXCL)); | |
1da177e4 | 1268 | } |
014c2544 | 1269 | return error; |
1da177e4 LT |
1270 | } |
1271 | ||
1272 | /* | |
1273 | * Free a symlink that has blocks associated with it. | |
1274 | */ | |
1275 | STATIC int | |
1276 | xfs_inactive_symlink_rmt( | |
1277 | xfs_inode_t *ip, | |
1278 | xfs_trans_t **tpp) | |
1279 | { | |
1280 | xfs_buf_t *bp; | |
1281 | int committed; | |
1282 | int done; | |
1283 | int error; | |
1284 | xfs_fsblock_t first_block; | |
1285 | xfs_bmap_free_t free_list; | |
1286 | int i; | |
1287 | xfs_mount_t *mp; | |
1288 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1289 | int nmaps; | |
1290 | xfs_trans_t *ntp; | |
1291 | int size; | |
1292 | xfs_trans_t *tp; | |
1293 | ||
1294 | tp = *tpp; | |
1295 | mp = ip->i_mount; | |
1296 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
1297 | /* | |
1298 | * We're freeing a symlink that has some | |
1299 | * blocks allocated to it. Free the | |
1300 | * blocks here. We know that we've got | |
1301 | * either 1 or 2 extents and that we can | |
1302 | * free them all in one bunmapi call. | |
1303 | */ | |
1304 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
1305 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1306 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1307 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1308 | xfs_trans_cancel(tp, 0); | |
1309 | *tpp = NULL; | |
1310 | return error; | |
1311 | } | |
1312 | /* | |
1313 | * Lock the inode, fix the size, and join it to the transaction. | |
1314 | * Hold it so in the normal path, we still have it locked for | |
1315 | * the second transaction. In the error paths we need it | |
1316 | * held so the cancel won't rele it, see below. | |
1317 | */ | |
1318 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1319 | size = (int)ip->i_d.di_size; | |
1320 | ip->i_d.di_size = 0; | |
1321 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1322 | xfs_trans_ihold(tp, ip); | |
1323 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1324 | /* | |
1325 | * Find the block(s) so we can inval and unmap them. | |
1326 | */ | |
1327 | done = 0; | |
1328 | XFS_BMAP_INIT(&free_list, &first_block); | |
e8c96f8c | 1329 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
1330 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
1331 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 1332 | &free_list, NULL))) |
1da177e4 LT |
1333 | goto error0; |
1334 | /* | |
1335 | * Invalidate the block(s). | |
1336 | */ | |
1337 | for (i = 0; i < nmaps; i++) { | |
1338 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
1339 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
1340 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
1341 | xfs_trans_binval(tp, bp); | |
1342 | } | |
1343 | /* | |
1344 | * Unmap the dead block(s) to the free_list. | |
1345 | */ | |
1346 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 1347 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
1348 | goto error1; |
1349 | ASSERT(done); | |
1350 | /* | |
1351 | * Commit the first transaction. This logs the EFI and the inode. | |
1352 | */ | |
f7c99b6f | 1353 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
1354 | goto error1; |
1355 | /* | |
1356 | * The transaction must have been committed, since there were | |
1357 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
1358 | * The new tp has the extent freeing and EFDs. | |
1359 | */ | |
1360 | ASSERT(committed); | |
1361 | /* | |
1362 | * The first xact was committed, so add the inode to the new one. | |
1363 | * Mark it dirty so it will be logged and moved forward in the log as | |
1364 | * part of every commit. | |
1365 | */ | |
1366 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1367 | xfs_trans_ihold(tp, ip); | |
1368 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1369 | /* | |
1370 | * Get a new, empty transaction to return to our caller. | |
1371 | */ | |
1372 | ntp = xfs_trans_dup(tp); | |
1373 | /* | |
c41564b5 | 1374 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
1375 | * If we get an error on the commit here or on the reserve below, |
1376 | * we need to unlock the inode since the new transaction doesn't | |
1377 | * have the inode attached. | |
1378 | */ | |
1c72bf90 | 1379 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1380 | tp = ntp; |
1381 | if (error) { | |
1382 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1383 | goto error0; | |
1384 | } | |
1385 | /* | |
1386 | * Remove the memory for extent descriptions (just bookkeeping). | |
1387 | */ | |
1388 | if (ip->i_df.if_bytes) | |
1389 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
1390 | ASSERT(ip->i_df.if_bytes == 0); | |
1391 | /* | |
1392 | * Put an itruncate log reservation in the new transaction | |
1393 | * for our caller. | |
1394 | */ | |
1395 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1396 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1397 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1398 | goto error0; | |
1399 | } | |
1400 | /* | |
1401 | * Return with the inode locked but not joined to the transaction. | |
1402 | */ | |
1403 | *tpp = tp; | |
1404 | return 0; | |
1405 | ||
1406 | error1: | |
1407 | xfs_bmap_cancel(&free_list); | |
1408 | error0: | |
1409 | /* | |
1410 | * Have to come here with the inode locked and either | |
1411 | * (held and in the transaction) or (not in the transaction). | |
1412 | * If the inode isn't held then cancel would iput it, but | |
1413 | * that's wrong since this is inactive and the vnode ref | |
1414 | * count is 0 already. | |
1415 | * Cancel won't do anything to the inode if held, but it still | |
1416 | * needs to be locked until the cancel is done, if it was | |
1417 | * joined to the transaction. | |
1418 | */ | |
1419 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1420 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1421 | *tpp = NULL; | |
1422 | return error; | |
1423 | ||
1424 | } | |
1425 | ||
1426 | STATIC int | |
1427 | xfs_inactive_symlink_local( | |
1428 | xfs_inode_t *ip, | |
1429 | xfs_trans_t **tpp) | |
1430 | { | |
1431 | int error; | |
1432 | ||
1433 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
1434 | /* | |
1435 | * We're freeing a symlink which fit into | |
1436 | * the inode. Just free the memory used | |
1437 | * to hold the old symlink. | |
1438 | */ | |
1439 | error = xfs_trans_reserve(*tpp, 0, | |
1440 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
1441 | 0, XFS_TRANS_PERM_LOG_RES, | |
1442 | XFS_ITRUNCATE_LOG_COUNT); | |
1443 | ||
1444 | if (error) { | |
1445 | xfs_trans_cancel(*tpp, 0); | |
1446 | *tpp = NULL; | |
014c2544 | 1447 | return error; |
1da177e4 LT |
1448 | } |
1449 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1450 | ||
1451 | /* | |
1452 | * Zero length symlinks _can_ exist. | |
1453 | */ | |
1454 | if (ip->i_df.if_bytes > 0) { | |
1455 | xfs_idata_realloc(ip, | |
1456 | -(ip->i_df.if_bytes), | |
1457 | XFS_DATA_FORK); | |
1458 | ASSERT(ip->i_df.if_bytes == 0); | |
1459 | } | |
014c2544 | 1460 | return 0; |
1da177e4 LT |
1461 | } |
1462 | ||
1da177e4 LT |
1463 | STATIC int |
1464 | xfs_inactive_attrs( | |
1465 | xfs_inode_t *ip, | |
1466 | xfs_trans_t **tpp) | |
1467 | { | |
1468 | xfs_trans_t *tp; | |
1469 | int error; | |
1470 | xfs_mount_t *mp; | |
1471 | ||
1472 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); | |
1473 | tp = *tpp; | |
1474 | mp = ip->i_mount; | |
1475 | ASSERT(ip->i_d.di_forkoff != 0); | |
1c72bf90 | 1476 | xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1477 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1478 | ||
1479 | error = xfs_attr_inactive(ip); | |
1480 | if (error) { | |
1481 | *tpp = NULL; | |
1482 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1483 | return error; /* goto out */ |
1da177e4 LT |
1484 | } |
1485 | ||
1486 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1487 | error = xfs_trans_reserve(tp, 0, | |
1488 | XFS_IFREE_LOG_RES(mp), | |
1489 | 0, XFS_TRANS_PERM_LOG_RES, | |
1490 | XFS_INACTIVE_LOG_COUNT); | |
1491 | if (error) { | |
1492 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1493 | xfs_trans_cancel(tp, 0); | |
1494 | *tpp = NULL; | |
1495 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1496 | return error; |
1da177e4 LT |
1497 | } |
1498 | ||
1499 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1500 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1501 | xfs_trans_ihold(tp, ip); | |
1502 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1503 | ||
1504 | ASSERT(ip->i_d.di_anextents == 0); | |
1505 | ||
1506 | *tpp = tp; | |
014c2544 | 1507 | return 0; |
1da177e4 LT |
1508 | } |
1509 | ||
993386c1 | 1510 | int |
1da177e4 | 1511 | xfs_release( |
993386c1 | 1512 | xfs_inode_t *ip) |
1da177e4 | 1513 | { |
993386c1 CH |
1514 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1515 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
1516 | int error; |
1517 | ||
7d4fb40a | 1518 | if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) |
1da177e4 | 1519 | return 0; |
1da177e4 LT |
1520 | |
1521 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1522 | if (vp->v_vfsp->vfs_flag & VFS_RDONLY) | |
1523 | return 0; | |
1524 | ||
2a82b8be DC |
1525 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
1526 | /* | |
1527 | * If we are using filestreams, and we have an unlinked | |
1528 | * file that we are processing the last close on, then nothing | |
1529 | * will be able to reopen and write to this file. Purge this | |
1530 | * inode from the filestreams cache so that it doesn't delay | |
1531 | * teardown of the inode. | |
1532 | */ | |
1533 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1534 | xfs_filestream_deassociate(ip); | |
1535 | ||
fbf3ce8d CH |
1536 | /* |
1537 | * If we previously truncated this file and removed old data | |
1538 | * in the process, we want to initiate "early" writeout on | |
1539 | * the last close. This is an attempt to combat the notorious | |
1540 | * NULL files problem which is particularly noticable from a | |
1541 | * truncate down, buffered (re-)write (delalloc), followed by | |
1542 | * a crash. What we are effectively doing here is | |
1543 | * significantly reducing the time window where we'd otherwise | |
1544 | * be exposed to that problem. | |
1545 | */ | |
1546 | if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0) | |
1547 | bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE); | |
1548 | } | |
1549 | ||
1da177e4 LT |
1550 | #ifdef HAVE_REFCACHE |
1551 | /* If we are in the NFS reference cache then don't do this now */ | |
1552 | if (ip->i_refcache) | |
1553 | return 0; | |
1554 | #endif | |
1555 | ||
1da177e4 LT |
1556 | if (ip->i_d.di_nlink != 0) { |
1557 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1558 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1559 | ip->i_delayed_blks > 0)) && |
1da177e4 | 1560 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
1561 | (!(ip->i_d.di_flags & |
1562 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
92dfe8d2 DC |
1563 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1564 | if (error) | |
014c2544 | 1565 | return error; |
1da177e4 | 1566 | /* Update linux inode block count after free above */ |
ec86dc02 | 1567 | vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp, |
1da177e4 LT |
1568 | ip->i_d.di_nblocks + ip->i_delayed_blks); |
1569 | } | |
1570 | } | |
1571 | ||
1572 | return 0; | |
1573 | } | |
1574 | ||
1575 | /* | |
1576 | * xfs_inactive | |
1577 | * | |
1578 | * This is called when the vnode reference count for the vnode | |
1579 | * goes to zero. If the file has been unlinked, then it must | |
1580 | * now be truncated. Also, we clear all of the read-ahead state | |
1581 | * kept for the inode here since the file is now closed. | |
1582 | */ | |
993386c1 | 1583 | int |
1da177e4 | 1584 | xfs_inactive( |
993386c1 | 1585 | xfs_inode_t *ip) |
1da177e4 | 1586 | { |
993386c1 | 1587 | bhv_vnode_t *vp = XFS_ITOV(ip); |
67fcaa73 | 1588 | xfs_bmap_free_t free_list; |
1da177e4 LT |
1589 | xfs_fsblock_t first_block; |
1590 | int committed; | |
1591 | xfs_trans_t *tp; | |
1592 | xfs_mount_t *mp; | |
1593 | int error; | |
1594 | int truncate; | |
1595 | ||
1da177e4 LT |
1596 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
1597 | ||
1da177e4 LT |
1598 | /* |
1599 | * If the inode is already free, then there can be nothing | |
1600 | * to clean up here. | |
1601 | */ | |
1602 | if (ip->i_d.di_mode == 0 || VN_BAD(vp)) { | |
1603 | ASSERT(ip->i_df.if_real_bytes == 0); | |
1604 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1605 | return VN_INACTIVE_CACHE; | |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | * Only do a truncate if it's a regular file with | |
1610 | * some actual space in it. It's OK to look at the | |
1611 | * inode's fields without the lock because we're the | |
1612 | * only one with a reference to the inode. | |
1613 | */ | |
1614 | truncate = ((ip->i_d.di_nlink == 0) && | |
ba87ea69 LM |
1615 | ((ip->i_d.di_size != 0) || (ip->i_size != 0) || |
1616 | (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1617 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1618 | ||
1619 | mp = ip->i_mount; | |
1620 | ||
eb9df39d | 1621 | if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) { |
1da177e4 LT |
1622 | (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL); |
1623 | } | |
1624 | ||
1625 | error = 0; | |
1626 | ||
1627 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1628 | if (vp->v_vfsp->vfs_flag & VFS_RDONLY) | |
1629 | goto out; | |
1630 | ||
1631 | if (ip->i_d.di_nlink != 0) { | |
1632 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1633 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1634 | ip->i_delayed_blks > 0)) && |
dd9f438e NS |
1635 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1636 | (!(ip->i_d.di_flags & | |
1637 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1638 | (ip->i_delayed_blks != 0)))) { | |
92dfe8d2 DC |
1639 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1640 | if (error) | |
014c2544 | 1641 | return VN_INACTIVE_CACHE; |
1da177e4 | 1642 | /* Update linux inode block count after free above */ |
ec86dc02 | 1643 | vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp, |
1da177e4 LT |
1644 | ip->i_d.di_nblocks + ip->i_delayed_blks); |
1645 | } | |
1646 | goto out; | |
1647 | } | |
1648 | ||
1649 | ASSERT(ip->i_d.di_nlink == 0); | |
1650 | ||
1651 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1652 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1653 | |
1654 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1655 | if (truncate) { | |
1656 | /* | |
1657 | * Do the xfs_itruncate_start() call before | |
1658 | * reserving any log space because itruncate_start | |
1659 | * will call into the buffer cache and we can't | |
1660 | * do that within a transaction. | |
1661 | */ | |
1662 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1663 | ||
d3cf2094 LM |
1664 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); |
1665 | if (error) { | |
87ae3c24 | 1666 | xfs_trans_cancel(tp, 0); |
d3cf2094 LM |
1667 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1668 | return VN_INACTIVE_CACHE; | |
1669 | } | |
1da177e4 LT |
1670 | |
1671 | error = xfs_trans_reserve(tp, 0, | |
1672 | XFS_ITRUNCATE_LOG_RES(mp), | |
1673 | 0, XFS_TRANS_PERM_LOG_RES, | |
1674 | XFS_ITRUNCATE_LOG_COUNT); | |
1675 | if (error) { | |
1676 | /* Don't call itruncate_cleanup */ | |
1677 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1678 | xfs_trans_cancel(tp, 0); | |
1679 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1680 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1681 | } |
1682 | ||
1683 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1684 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1685 | xfs_trans_ihold(tp, ip); | |
1686 | ||
1687 | /* | |
1688 | * normally, we have to run xfs_itruncate_finish sync. | |
1689 | * But if filesystem is wsync and we're in the inactive | |
1690 | * path, then we know that nlink == 0, and that the | |
1691 | * xaction that made nlink == 0 is permanently committed | |
1692 | * since xfs_remove runs as a synchronous transaction. | |
1693 | */ | |
1694 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1695 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1696 | ||
1697 | if (error) { | |
1698 | xfs_trans_cancel(tp, | |
1699 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1700 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1701 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1702 | } |
1703 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1704 | ||
1705 | /* | |
1706 | * If we get an error while cleaning up a | |
1707 | * symlink we bail out. | |
1708 | */ | |
1709 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1710 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1711 | xfs_inactive_symlink_local(ip, &tp); | |
1712 | ||
1713 | if (error) { | |
1714 | ASSERT(tp == NULL); | |
014c2544 | 1715 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1716 | } |
1717 | ||
1718 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1719 | xfs_trans_ihold(tp, ip); | |
1720 | } else { | |
1721 | error = xfs_trans_reserve(tp, 0, | |
1722 | XFS_IFREE_LOG_RES(mp), | |
1723 | 0, XFS_TRANS_PERM_LOG_RES, | |
1724 | XFS_INACTIVE_LOG_COUNT); | |
1725 | if (error) { | |
1726 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1727 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1728 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1729 | } |
1730 | ||
1731 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1732 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1733 | xfs_trans_ihold(tp, ip); | |
1734 | } | |
1735 | ||
1736 | /* | |
1737 | * If there are attributes associated with the file | |
1738 | * then blow them away now. The code calls a routine | |
1739 | * that recursively deconstructs the attribute fork. | |
1740 | * We need to just commit the current transaction | |
1741 | * because we can't use it for xfs_attr_inactive(). | |
1742 | */ | |
1743 | if (ip->i_d.di_anextents > 0) { | |
1744 | error = xfs_inactive_attrs(ip, &tp); | |
1745 | /* | |
1746 | * If we got an error, the transaction is already | |
1747 | * cancelled, and the inode is unlocked. Just get out. | |
1748 | */ | |
1749 | if (error) | |
014c2544 | 1750 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1751 | } else if (ip->i_afp) { |
1752 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1753 | } | |
1754 | ||
1755 | /* | |
1756 | * Free the inode. | |
1757 | */ | |
1758 | XFS_BMAP_INIT(&free_list, &first_block); | |
1759 | error = xfs_ifree(tp, ip, &free_list); | |
1760 | if (error) { | |
1761 | /* | |
1762 | * If we fail to free the inode, shut down. The cancel | |
1763 | * might do that, we need to make sure. Otherwise the | |
1764 | * inode might be lost for a long time or forever. | |
1765 | */ | |
1766 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1767 | cmn_err(CE_NOTE, | |
1768 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1769 | error, mp->m_fsname); | |
7d04a335 | 1770 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1771 | } |
1772 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1773 | } else { | |
1774 | /* | |
1775 | * Credit the quota account(s). The inode is gone. | |
1776 | */ | |
1777 | XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1778 | ||
1779 | /* | |
1780 | * Just ignore errors at this point. There is | |
1781 | * nothing we can do except to try to keep going. | |
1782 | */ | |
f7c99b6f | 1783 | (void) xfs_bmap_finish(&tp, &free_list, &committed); |
1c72bf90 | 1784 | (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1785 | } |
1786 | /* | |
1787 | * Release the dquots held by inode, if any. | |
1788 | */ | |
1789 | XFS_QM_DQDETACH(mp, ip); | |
1790 | ||
1791 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1792 | ||
1793 | out: | |
1794 | return VN_INACTIVE_CACHE; | |
1795 | } | |
1796 | ||
1797 | ||
993386c1 | 1798 | int |
1da177e4 | 1799 | xfs_lookup( |
993386c1 | 1800 | xfs_inode_t *dp, |
8285fb58 | 1801 | bhv_vname_t *dentry, |
993386c1 | 1802 | bhv_vnode_t **vpp) |
1da177e4 | 1803 | { |
993386c1 | 1804 | xfs_inode_t *ip; |
1da177e4 LT |
1805 | xfs_ino_t e_inum; |
1806 | int error; | |
1807 | uint lock_mode; | |
1da177e4 | 1808 | |
993386c1 | 1809 | vn_trace_entry(XFS_ITOV(dp), __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
1810 | |
1811 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1812 | return XFS_ERROR(EIO); | |
1813 | ||
1814 | lock_mode = xfs_ilock_map_shared(dp); | |
993386c1 | 1815 | error = xfs_dir_lookup_int(dp, lock_mode, dentry, &e_inum, &ip); |
1da177e4 LT |
1816 | if (!error) { |
1817 | *vpp = XFS_ITOV(ip); | |
1818 | ITRACE(ip); | |
1819 | } | |
1820 | xfs_iunlock_map_shared(dp, lock_mode); | |
1821 | return error; | |
1822 | } | |
1823 | ||
993386c1 | 1824 | int |
1da177e4 | 1825 | xfs_create( |
993386c1 | 1826 | xfs_inode_t *dp, |
8285fb58 NS |
1827 | bhv_vname_t *dentry, |
1828 | bhv_vattr_t *vap, | |
67fcaa73 | 1829 | bhv_vnode_t **vpp, |
1da177e4 LT |
1830 | cred_t *credp) |
1831 | { | |
1832 | char *name = VNAME(dentry); | |
993386c1 CH |
1833 | xfs_mount_t *mp = dp->i_mount; |
1834 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); | |
1835 | xfs_inode_t *ip; | |
67fcaa73 | 1836 | bhv_vnode_t *vp = NULL; |
1da177e4 | 1837 | xfs_trans_t *tp; |
1da177e4 LT |
1838 | xfs_dev_t rdev; |
1839 | int error; | |
1840 | xfs_bmap_free_t free_list; | |
1841 | xfs_fsblock_t first_block; | |
993386c1 | 1842 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1843 | int dm_event_sent = 0; |
1844 | uint cancel_flags; | |
1845 | int committed; | |
1846 | xfs_prid_t prid; | |
1847 | struct xfs_dquot *udqp, *gdqp; | |
1848 | uint resblks; | |
1849 | int dm_di_mode; | |
1850 | int namelen; | |
1851 | ||
1852 | ASSERT(!*vpp); | |
1da177e4 LT |
1853 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); |
1854 | ||
0432dab2 | 1855 | dm_di_mode = vap->va_mode; |
1da177e4 LT |
1856 | namelen = VNAMELEN(dentry); |
1857 | ||
eb9df39d | 1858 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 LT |
1859 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
1860 | dir_vp, DM_RIGHT_NULL, NULL, | |
1861 | DM_RIGHT_NULL, name, NULL, | |
1862 | dm_di_mode, 0, 0); | |
1863 | ||
1864 | if (error) | |
1865 | return error; | |
1866 | dm_event_sent = 1; | |
1867 | } | |
1868 | ||
1869 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1870 | return XFS_ERROR(EIO); | |
1871 | ||
1872 | /* Return through std_return after this point. */ | |
1873 | ||
1874 | udqp = gdqp = NULL; | |
365ca83d NS |
1875 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1876 | prid = dp->i_d.di_projid; | |
1877 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
1878 | prid = (xfs_prid_t)vap->va_projid; |
1879 | else | |
1880 | prid = (xfs_prid_t)dfltprid; | |
1881 | ||
1882 | /* | |
1883 | * Make sure that we have allocated dquot(s) on disk. | |
1884 | */ | |
1885 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 1886 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
1887 | XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1888 | if (error) | |
1889 | goto std_return; | |
1890 | ||
1891 | ip = NULL; | |
1da177e4 LT |
1892 | |
1893 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1894 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1895 | resblks = XFS_CREATE_SPACE_RES(mp, namelen); | |
1896 | /* | |
1897 | * Initially assume that the file does not exist and | |
1898 | * reserve the resources for that case. If that is not | |
1899 | * the case we'll drop the one we have and get a more | |
1900 | * appropriate transaction later. | |
1901 | */ | |
1902 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1903 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1904 | if (error == ENOSPC) { | |
1905 | resblks = 0; | |
1906 | error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0, | |
1907 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1908 | } | |
1909 | if (error) { | |
1910 | cancel_flags = 0; | |
1da177e4 LT |
1911 | goto error_return; |
1912 | } | |
1913 | ||
f7c66ce3 | 1914 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1915 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
1916 | |
1917 | XFS_BMAP_INIT(&free_list, &first_block); | |
1918 | ||
1919 | ASSERT(ip == NULL); | |
1920 | ||
1921 | /* | |
1922 | * Reserve disk quota and the inode. | |
1923 | */ | |
1924 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
1925 | if (error) | |
1926 | goto error_return; | |
1927 | ||
f6c2d1fa | 1928 | if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen))) |
1da177e4 LT |
1929 | goto error_return; |
1930 | rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0; | |
0432dab2 | 1931 | error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1, |
1da177e4 LT |
1932 | rdev, credp, prid, resblks > 0, |
1933 | &ip, &committed); | |
1934 | if (error) { | |
1935 | if (error == ENOSPC) | |
1936 | goto error_return; | |
1937 | goto abort_return; | |
1938 | } | |
1939 | ITRACE(ip); | |
1940 | ||
1941 | /* | |
1942 | * At this point, we've gotten a newly allocated inode. | |
1943 | * It is locked (and joined to the transaction). | |
1944 | */ | |
1945 | ||
1946 | ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE)); | |
1947 | ||
1948 | /* | |
993386c1 CH |
1949 | * Now we join the directory inode to the transaction. We do not do it |
1950 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1951 | * (and release all the locks). An error from here on will result in | |
1952 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1953 | * error path. | |
1da177e4 | 1954 | */ |
1da177e4 LT |
1955 | VN_HOLD(dir_vp); |
1956 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 1957 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 1958 | |
f6c2d1fa NS |
1959 | error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino, |
1960 | &first_block, &free_list, resblks ? | |
1961 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1962 | if (error) { |
1963 | ASSERT(error != ENOSPC); | |
1964 | goto abort_return; | |
1965 | } | |
1966 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1967 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1968 | ||
1969 | /* | |
1970 | * If this is a synchronous mount, make sure that the | |
1971 | * create transaction goes to disk before returning to | |
1972 | * the user. | |
1973 | */ | |
1974 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1975 | xfs_trans_set_sync(tp); | |
1976 | } | |
1977 | ||
1978 | dp->i_gen++; | |
1979 | ||
1980 | /* | |
1981 | * Attach the dquot(s) to the inodes and modify them incore. | |
1982 | * These ids of the inode couldn't have changed since the new | |
1983 | * inode has been locked ever since it was created. | |
1984 | */ | |
1985 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
1986 | ||
1987 | /* | |
1988 | * xfs_trans_commit normally decrements the vnode ref count | |
1989 | * when it unlocks the inode. Since we want to return the | |
1990 | * vnode to the caller, we bump the vnode ref count now. | |
1991 | */ | |
1992 | IHOLD(ip); | |
1993 | vp = XFS_ITOV(ip); | |
1994 | ||
f7c99b6f | 1995 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1996 | if (error) { |
1997 | xfs_bmap_cancel(&free_list); | |
1998 | goto abort_rele; | |
1999 | } | |
2000 | ||
1c72bf90 | 2001 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2002 | if (error) { |
2003 | IRELE(ip); | |
2004 | tp = NULL; | |
2005 | goto error_return; | |
2006 | } | |
2007 | ||
2008 | XFS_QM_DQRELE(mp, udqp); | |
2009 | XFS_QM_DQRELE(mp, gdqp); | |
2010 | ||
2011 | /* | |
c41564b5 | 2012 | * Propagate the fact that the vnode changed after the |
1da177e4 LT |
2013 | * xfs_inode locks have been released. |
2014 | */ | |
67fcaa73 | 2015 | bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3); |
1da177e4 LT |
2016 | |
2017 | *vpp = vp; | |
2018 | ||
2019 | /* Fallthrough to std_return with error = 0 */ | |
2020 | ||
2021 | std_return: | |
eb9df39d | 2022 | if ((*vpp || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 2023 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 LT |
2024 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
2025 | dir_vp, DM_RIGHT_NULL, | |
2026 | *vpp ? vp:NULL, | |
2027 | DM_RIGHT_NULL, name, NULL, | |
2028 | dm_di_mode, error, 0); | |
2029 | } | |
2030 | return error; | |
2031 | ||
2032 | abort_return: | |
2033 | cancel_flags |= XFS_TRANS_ABORT; | |
2034 | /* FALLTHROUGH */ | |
1da177e4 | 2035 | |
014c2544 | 2036 | error_return: |
1da177e4 LT |
2037 | if (tp != NULL) |
2038 | xfs_trans_cancel(tp, cancel_flags); | |
2039 | ||
1da177e4 LT |
2040 | XFS_QM_DQRELE(mp, udqp); |
2041 | XFS_QM_DQRELE(mp, gdqp); | |
2042 | ||
993386c1 CH |
2043 | if (unlock_dp_on_error) |
2044 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2045 | ||
1da177e4 LT |
2046 | goto std_return; |
2047 | ||
2048 | abort_rele: | |
2049 | /* | |
2050 | * Wait until after the current transaction is aborted to | |
2051 | * release the inode. This prevents recursive transactions | |
2052 | * and deadlocks from xfs_inactive. | |
2053 | */ | |
2054 | cancel_flags |= XFS_TRANS_ABORT; | |
2055 | xfs_trans_cancel(tp, cancel_flags); | |
2056 | IRELE(ip); | |
2057 | ||
2058 | XFS_QM_DQRELE(mp, udqp); | |
2059 | XFS_QM_DQRELE(mp, gdqp); | |
2060 | ||
2061 | goto std_return; | |
2062 | } | |
2063 | ||
2064 | #ifdef DEBUG | |
2065 | /* | |
2066 | * Some counters to see if (and how often) we are hitting some deadlock | |
2067 | * prevention code paths. | |
2068 | */ | |
2069 | ||
2070 | int xfs_rm_locks; | |
2071 | int xfs_rm_lock_delays; | |
2072 | int xfs_rm_attempts; | |
2073 | #endif | |
2074 | ||
2075 | /* | |
2076 | * The following routine will lock the inodes associated with the | |
2077 | * directory and the named entry in the directory. The locks are | |
2078 | * acquired in increasing inode number. | |
2079 | * | |
2080 | * If the entry is "..", then only the directory is locked. The | |
2081 | * vnode ref count will still include that from the .. entry in | |
2082 | * this case. | |
2083 | * | |
2084 | * There is a deadlock we need to worry about. If the locked directory is | |
2085 | * in the AIL, it might be blocking up the log. The next inode we lock | |
2086 | * could be already locked by another thread waiting for log space (e.g | |
2087 | * a permanent log reservation with a long running transaction (see | |
2088 | * xfs_itruncate_finish)). To solve this, we must check if the directory | |
2089 | * is in the ail and use lock_nowait. If we can't lock, we need to | |
2090 | * drop the inode lock on the directory and try again. xfs_iunlock will | |
2091 | * potentially push the tail if we were holding up the log. | |
2092 | */ | |
2093 | STATIC int | |
2094 | xfs_lock_dir_and_entry( | |
2095 | xfs_inode_t *dp, | |
1da177e4 LT |
2096 | xfs_inode_t *ip) /* inode of entry 'name' */ |
2097 | { | |
2098 | int attempts; | |
2099 | xfs_ino_t e_inum; | |
2100 | xfs_inode_t *ips[2]; | |
2101 | xfs_log_item_t *lp; | |
2102 | ||
2103 | #ifdef DEBUG | |
2104 | xfs_rm_locks++; | |
2105 | #endif | |
2106 | attempts = 0; | |
2107 | ||
2108 | again: | |
f7c66ce3 | 2109 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
1da177e4 LT |
2110 | |
2111 | e_inum = ip->i_ino; | |
2112 | ||
2113 | ITRACE(ip); | |
2114 | ||
2115 | /* | |
2116 | * We want to lock in increasing inum. Since we've already | |
2117 | * acquired the lock on the directory, we may need to release | |
2118 | * if if the inum of the entry turns out to be less. | |
2119 | */ | |
2120 | if (e_inum > dp->i_ino) { | |
2121 | /* | |
2122 | * We are already in the right order, so just | |
2123 | * lock on the inode of the entry. | |
2124 | * We need to use nowait if dp is in the AIL. | |
2125 | */ | |
2126 | ||
2127 | lp = (xfs_log_item_t *)dp->i_itemp; | |
2128 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2129 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2130 | attempts++; | |
2131 | #ifdef DEBUG | |
2132 | xfs_rm_attempts++; | |
2133 | #endif | |
2134 | ||
2135 | /* | |
2136 | * Unlock dp and try again. | |
2137 | * xfs_iunlock will try to push the tail | |
2138 | * if the inode is in the AIL. | |
2139 | */ | |
2140 | ||
2141 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2142 | ||
2143 | if ((attempts % 5) == 0) { | |
2144 | delay(1); /* Don't just spin the CPU */ | |
2145 | #ifdef DEBUG | |
2146 | xfs_rm_lock_delays++; | |
2147 | #endif | |
2148 | } | |
2149 | goto again; | |
2150 | } | |
2151 | } else { | |
2152 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2153 | } | |
2154 | } else if (e_inum < dp->i_ino) { | |
2155 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2156 | ||
2157 | ips[0] = ip; | |
2158 | ips[1] = dp; | |
2159 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2160 | } | |
2161 | /* else e_inum == dp->i_ino */ | |
2162 | /* This can happen if we're asked to lock /x/.. | |
2163 | * the entry is "..", which is also the parent directory. | |
2164 | */ | |
2165 | ||
2166 | return 0; | |
2167 | } | |
2168 | ||
2169 | #ifdef DEBUG | |
2170 | int xfs_locked_n; | |
2171 | int xfs_small_retries; | |
2172 | int xfs_middle_retries; | |
2173 | int xfs_lots_retries; | |
2174 | int xfs_lock_delays; | |
2175 | #endif | |
2176 | ||
f7c66ce3 LM |
2177 | /* |
2178 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
2179 | * a different value | |
2180 | */ | |
2181 | static inline int | |
2182 | xfs_lock_inumorder(int lock_mode, int subclass) | |
2183 | { | |
2184 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 2185 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 2186 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 2187 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
2188 | |
2189 | return lock_mode; | |
2190 | } | |
2191 | ||
1da177e4 LT |
2192 | /* |
2193 | * The following routine will lock n inodes in exclusive mode. | |
2194 | * We assume the caller calls us with the inodes in i_ino order. | |
2195 | * | |
2196 | * We need to detect deadlock where an inode that we lock | |
2197 | * is in the AIL and we start waiting for another inode that is locked | |
2198 | * by a thread in a long running transaction (such as truncate). This can | |
2199 | * result in deadlock since the long running trans might need to wait | |
2200 | * for the inode we just locked in order to push the tail and free space | |
2201 | * in the log. | |
2202 | */ | |
2203 | void | |
2204 | xfs_lock_inodes( | |
2205 | xfs_inode_t **ips, | |
2206 | int inodes, | |
2207 | int first_locked, | |
2208 | uint lock_mode) | |
2209 | { | |
2210 | int attempts = 0, i, j, try_lock; | |
2211 | xfs_log_item_t *lp; | |
2212 | ||
2213 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
2214 | ||
2215 | if (first_locked) { | |
2216 | try_lock = 1; | |
2217 | i = 1; | |
2218 | } else { | |
2219 | try_lock = 0; | |
2220 | i = 0; | |
2221 | } | |
2222 | ||
2223 | again: | |
2224 | for (; i < inodes; i++) { | |
2225 | ASSERT(ips[i]); | |
2226 | ||
2227 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
2228 | continue; | |
2229 | ||
2230 | /* | |
2231 | * If try_lock is not set yet, make sure all locked inodes | |
2232 | * are not in the AIL. | |
2233 | * If any are, set try_lock to be used later. | |
2234 | */ | |
2235 | ||
2236 | if (!try_lock) { | |
2237 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
2238 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
2239 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2240 | try_lock++; | |
2241 | } | |
2242 | } | |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * If any of the previous locks we have locked is in the AIL, | |
2247 | * we must TRY to get the second and subsequent locks. If | |
2248 | * we can't get any, we must release all we have | |
2249 | * and try again. | |
2250 | */ | |
2251 | ||
2252 | if (try_lock) { | |
2253 | /* try_lock must be 0 if i is 0. */ | |
2254 | /* | |
2255 | * try_lock means we have an inode locked | |
2256 | * that is in the AIL. | |
2257 | */ | |
2258 | ASSERT(i != 0); | |
f7c66ce3 | 2259 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
2260 | attempts++; |
2261 | ||
2262 | /* | |
2263 | * Unlock all previous guys and try again. | |
2264 | * xfs_iunlock will try to push the tail | |
2265 | * if the inode is in the AIL. | |
2266 | */ | |
2267 | ||
2268 | for(j = i - 1; j >= 0; j--) { | |
2269 | ||
2270 | /* | |
2271 | * Check to see if we've already | |
2272 | * unlocked this one. | |
2273 | * Not the first one going back, | |
2274 | * and the inode ptr is the same. | |
2275 | */ | |
2276 | if ((j != (i - 1)) && ips[j] == | |
2277 | ips[j+1]) | |
2278 | continue; | |
2279 | ||
2280 | xfs_iunlock(ips[j], lock_mode); | |
2281 | } | |
2282 | ||
2283 | if ((attempts % 5) == 0) { | |
2284 | delay(1); /* Don't just spin the CPU */ | |
2285 | #ifdef DEBUG | |
2286 | xfs_lock_delays++; | |
2287 | #endif | |
2288 | } | |
2289 | i = 0; | |
2290 | try_lock = 0; | |
2291 | goto again; | |
2292 | } | |
2293 | } else { | |
f7c66ce3 | 2294 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
2295 | } |
2296 | } | |
2297 | ||
2298 | #ifdef DEBUG | |
2299 | if (attempts) { | |
2300 | if (attempts < 5) xfs_small_retries++; | |
2301 | else if (attempts < 100) xfs_middle_retries++; | |
2302 | else xfs_lots_retries++; | |
2303 | } else { | |
2304 | xfs_locked_n++; | |
2305 | } | |
2306 | #endif | |
2307 | } | |
2308 | ||
2309 | #ifdef DEBUG | |
2310 | #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);} | |
2311 | int remove_which_error_return = 0; | |
2312 | #else /* ! DEBUG */ | |
2313 | #define REMOVE_DEBUG_TRACE(x) | |
2314 | #endif /* ! DEBUG */ | |
2315 | ||
993386c1 | 2316 | int |
1da177e4 | 2317 | xfs_remove( |
993386c1 CH |
2318 | xfs_inode_t *dp, |
2319 | bhv_vname_t *dentry) | |
1da177e4 | 2320 | { |
993386c1 | 2321 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2322 | char *name = VNAME(dentry); |
993386c1 CH |
2323 | xfs_mount_t *mp = dp->i_mount; |
2324 | xfs_inode_t *ip; | |
1da177e4 | 2325 | xfs_trans_t *tp = NULL; |
1da177e4 LT |
2326 | int error = 0; |
2327 | xfs_bmap_free_t free_list; | |
2328 | xfs_fsblock_t first_block; | |
2329 | int cancel_flags; | |
2330 | int committed; | |
2331 | int dm_di_mode = 0; | |
2332 | int link_zero; | |
2333 | uint resblks; | |
2334 | int namelen; | |
2335 | ||
1da177e4 LT |
2336 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); |
2337 | ||
1da177e4 LT |
2338 | if (XFS_FORCED_SHUTDOWN(mp)) |
2339 | return XFS_ERROR(EIO); | |
2340 | ||
2341 | namelen = VNAMELEN(dentry); | |
2342 | ||
17370097 VA |
2343 | if (!xfs_get_dir_entry(dentry, &ip)) { |
2344 | dm_di_mode = ip->i_d.di_mode; | |
2345 | IRELE(ip); | |
2346 | } | |
2347 | ||
eb9df39d | 2348 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
1da177e4 LT |
2349 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp, |
2350 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
17370097 | 2351 | name, NULL, dm_di_mode, 0, 0); |
1da177e4 LT |
2352 | if (error) |
2353 | return error; | |
2354 | } | |
2355 | ||
2356 | /* From this point on, return through std_return */ | |
2357 | ip = NULL; | |
2358 | ||
2359 | /* | |
2360 | * We need to get a reference to ip before we get our log | |
2361 | * reservation. The reason for this is that we cannot call | |
2362 | * xfs_iget for an inode for which we do not have a reference | |
2363 | * once we've acquired a log reservation. This is because the | |
2364 | * inode we are trying to get might be in xfs_inactive going | |
2365 | * for a log reservation. Since we'll have to wait for the | |
2366 | * inactive code to complete before returning from xfs_iget, | |
2367 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2368 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2369 | * to the inode before getting our log reservation. |
2370 | */ | |
2371 | error = xfs_get_dir_entry(dentry, &ip); | |
2372 | if (error) { | |
2373 | REMOVE_DEBUG_TRACE(__LINE__); | |
2374 | goto std_return; | |
2375 | } | |
2376 | ||
2377 | dm_di_mode = ip->i_d.di_mode; | |
2378 | ||
2379 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
2380 | ||
2381 | ITRACE(ip); | |
2382 | ||
2383 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2384 | if (!error && dp != ip) | |
2385 | error = XFS_QM_DQATTACH(mp, ip, 0); | |
2386 | if (error) { | |
2387 | REMOVE_DEBUG_TRACE(__LINE__); | |
2388 | IRELE(ip); | |
2389 | goto std_return; | |
2390 | } | |
2391 | ||
2392 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2393 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2394 | /* | |
2395 | * We try to get the real space reservation first, | |
2396 | * allowing for directory btree deletion(s) implying | |
2397 | * possible bmap insert(s). If we can't get the space | |
2398 | * reservation then we use 0 instead, and avoid the bmap | |
2399 | * btree insert(s) in the directory code by, if the bmap | |
2400 | * insert tries to happen, instead trimming the LAST | |
2401 | * block from the directory. | |
2402 | */ | |
2403 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2404 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2405 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2406 | if (error == ENOSPC) { | |
2407 | resblks = 0; | |
2408 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2409 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2410 | } | |
2411 | if (error) { | |
2412 | ASSERT(error != ENOSPC); | |
2413 | REMOVE_DEBUG_TRACE(__LINE__); | |
2414 | xfs_trans_cancel(tp, 0); | |
2415 | IRELE(ip); | |
2416 | return error; | |
2417 | } | |
2418 | ||
e9ed9d22 | 2419 | error = xfs_lock_dir_and_entry(dp, ip); |
1da177e4 LT |
2420 | if (error) { |
2421 | REMOVE_DEBUG_TRACE(__LINE__); | |
2422 | xfs_trans_cancel(tp, cancel_flags); | |
2423 | IRELE(ip); | |
2424 | goto std_return; | |
2425 | } | |
2426 | ||
2427 | /* | |
2428 | * At this point, we've gotten both the directory and the entry | |
2429 | * inodes locked. | |
2430 | */ | |
2431 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2432 | if (dp != ip) { | |
2433 | /* | |
2434 | * Increment vnode ref count only in this case since | |
2435 | * there's an extra vnode reference in the case where | |
2436 | * dp == ip. | |
2437 | */ | |
2438 | IHOLD(dp); | |
2439 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2440 | } | |
2441 | ||
2442 | /* | |
2443 | * Entry must exist since we did a lookup in xfs_lock_dir_and_entry. | |
2444 | */ | |
2445 | XFS_BMAP_INIT(&free_list, &first_block); | |
f6c2d1fa NS |
2446 | error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino, |
2447 | &first_block, &free_list, 0); | |
1da177e4 LT |
2448 | if (error) { |
2449 | ASSERT(error != ENOENT); | |
2450 | REMOVE_DEBUG_TRACE(__LINE__); | |
2451 | goto error1; | |
2452 | } | |
2453 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2454 | ||
2455 | dp->i_gen++; | |
2456 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2457 | ||
2458 | error = xfs_droplink(tp, ip); | |
2459 | if (error) { | |
2460 | REMOVE_DEBUG_TRACE(__LINE__); | |
2461 | goto error1; | |
2462 | } | |
2463 | ||
2464 | /* Determine if this is the last link while | |
2465 | * we are in the transaction. | |
2466 | */ | |
2467 | link_zero = (ip)->i_d.di_nlink==0; | |
2468 | ||
2469 | /* | |
2470 | * Take an extra ref on the inode so that it doesn't | |
2471 | * go to xfs_inactive() from within the commit. | |
2472 | */ | |
2473 | IHOLD(ip); | |
2474 | ||
2475 | /* | |
2476 | * If this is a synchronous mount, make sure that the | |
2477 | * remove transaction goes to disk before returning to | |
2478 | * the user. | |
2479 | */ | |
2480 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2481 | xfs_trans_set_sync(tp); | |
2482 | } | |
2483 | ||
f7c99b6f | 2484 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2485 | if (error) { |
2486 | REMOVE_DEBUG_TRACE(__LINE__); | |
2487 | goto error_rele; | |
2488 | } | |
2489 | ||
1c72bf90 | 2490 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2491 | if (error) { |
2492 | IRELE(ip); | |
2493 | goto std_return; | |
2494 | } | |
2495 | ||
2496 | /* | |
2497 | * Before we drop our extra reference to the inode, purge it | |
2498 | * from the refcache if it is there. By waiting until afterwards | |
2499 | * to do the IRELE, we ensure that we won't go inactive in the | |
2500 | * xfs_refcache_purge_ip routine (although that would be OK). | |
2501 | */ | |
2502 | xfs_refcache_purge_ip(ip); | |
2503 | ||
2a82b8be DC |
2504 | /* |
2505 | * If we are using filestreams, kill the stream association. | |
2506 | * If the file is still open it may get a new one but that | |
2507 | * will get killed on last close in xfs_close() so we don't | |
2508 | * have to worry about that. | |
2509 | */ | |
2510 | if (link_zero && xfs_inode_is_filestream(ip)) | |
2511 | xfs_filestream_deassociate(ip); | |
2512 | ||
1da177e4 LT |
2513 | vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); |
2514 | ||
2515 | /* | |
2516 | * Let interposed file systems know about removed links. | |
2517 | */ | |
67fcaa73 | 2518 | bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero); |
1da177e4 LT |
2519 | |
2520 | IRELE(ip); | |
2521 | ||
2522 | /* Fall through to std_return with error = 0 */ | |
2523 | std_return: | |
eb9df39d | 2524 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 LT |
2525 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
2526 | dir_vp, DM_RIGHT_NULL, | |
2527 | NULL, DM_RIGHT_NULL, | |
2528 | name, NULL, dm_di_mode, error, 0); | |
2529 | } | |
2530 | return error; | |
2531 | ||
2532 | error1: | |
2533 | xfs_bmap_cancel(&free_list); | |
2534 | cancel_flags |= XFS_TRANS_ABORT; | |
2535 | xfs_trans_cancel(tp, cancel_flags); | |
2536 | goto std_return; | |
2537 | ||
2538 | error_rele: | |
2539 | /* | |
2540 | * In this case make sure to not release the inode until after | |
2541 | * the current transaction is aborted. Releasing it beforehand | |
2542 | * can cause us to go to xfs_inactive and start a recursive | |
2543 | * transaction which can easily deadlock with the current one. | |
2544 | */ | |
2545 | xfs_bmap_cancel(&free_list); | |
2546 | cancel_flags |= XFS_TRANS_ABORT; | |
2547 | xfs_trans_cancel(tp, cancel_flags); | |
2548 | ||
2549 | /* | |
2550 | * Before we drop our extra reference to the inode, purge it | |
2551 | * from the refcache if it is there. By waiting until afterwards | |
2552 | * to do the IRELE, we ensure that we won't go inactive in the | |
2553 | * xfs_refcache_purge_ip routine (although that would be OK). | |
2554 | */ | |
2555 | xfs_refcache_purge_ip(ip); | |
2556 | ||
2557 | IRELE(ip); | |
2558 | ||
2559 | goto std_return; | |
2560 | } | |
2561 | ||
993386c1 | 2562 | int |
1da177e4 | 2563 | xfs_link( |
993386c1 | 2564 | xfs_inode_t *tdp, |
67fcaa73 | 2565 | bhv_vnode_t *src_vp, |
993386c1 | 2566 | bhv_vname_t *dentry) |
1da177e4 | 2567 | { |
993386c1 CH |
2568 | bhv_vnode_t *target_dir_vp = XFS_ITOV(tdp); |
2569 | xfs_mount_t *mp = tdp->i_mount; | |
2570 | xfs_inode_t *sip = xfs_vtoi(src_vp); | |
1da177e4 | 2571 | xfs_trans_t *tp; |
1da177e4 LT |
2572 | xfs_inode_t *ips[2]; |
2573 | int error; | |
2574 | xfs_bmap_free_t free_list; | |
2575 | xfs_fsblock_t first_block; | |
2576 | int cancel_flags; | |
2577 | int committed; | |
1da177e4 LT |
2578 | int resblks; |
2579 | char *target_name = VNAME(dentry); | |
2580 | int target_namelen; | |
2581 | ||
1da177e4 LT |
2582 | vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address); |
2583 | vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address); | |
2584 | ||
2585 | target_namelen = VNAMELEN(dentry); | |
b71d300c | 2586 | ASSERT(!VN_ISDIR(src_vp)); |
1da177e4 | 2587 | |
1da177e4 LT |
2588 | if (XFS_FORCED_SHUTDOWN(mp)) |
2589 | return XFS_ERROR(EIO); | |
2590 | ||
eb9df39d | 2591 | if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) { |
1da177e4 LT |
2592 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, |
2593 | target_dir_vp, DM_RIGHT_NULL, | |
2594 | src_vp, DM_RIGHT_NULL, | |
2595 | target_name, NULL, 0, 0, 0); | |
2596 | if (error) | |
2597 | return error; | |
2598 | } | |
2599 | ||
2600 | /* Return through std_return after this point. */ | |
2601 | ||
2602 | error = XFS_QM_DQATTACH(mp, sip, 0); | |
2603 | if (!error && sip != tdp) | |
2604 | error = XFS_QM_DQATTACH(mp, tdp, 0); | |
2605 | if (error) | |
2606 | goto std_return; | |
2607 | ||
2608 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2609 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2610 | resblks = XFS_LINK_SPACE_RES(mp, target_namelen); | |
2611 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, | |
2612 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2613 | if (error == ENOSPC) { | |
2614 | resblks = 0; | |
2615 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2616 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2617 | } | |
2618 | if (error) { | |
2619 | cancel_flags = 0; | |
2620 | goto error_return; | |
2621 | } | |
2622 | ||
2623 | if (sip->i_ino < tdp->i_ino) { | |
2624 | ips[0] = sip; | |
2625 | ips[1] = tdp; | |
2626 | } else { | |
2627 | ips[0] = tdp; | |
2628 | ips[1] = sip; | |
2629 | } | |
2630 | ||
2631 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2632 | ||
2633 | /* | |
2634 | * Increment vnode ref counts since xfs_trans_commit & | |
2635 | * xfs_trans_cancel will both unlock the inodes and | |
2636 | * decrement the associated ref counts. | |
2637 | */ | |
2638 | VN_HOLD(src_vp); | |
2639 | VN_HOLD(target_dir_vp); | |
2640 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
2641 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2642 | ||
2643 | /* | |
2644 | * If the source has too many links, we can't make any more to it. | |
2645 | */ | |
2646 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2647 | error = XFS_ERROR(EMLINK); | |
2648 | goto error_return; | |
2649 | } | |
2650 | ||
365ca83d NS |
2651 | /* |
2652 | * If we are using project inheritance, we only allow hard link | |
2653 | * creation in our tree when the project IDs are the same; else | |
2654 | * the tree quota mechanism could be circumvented. | |
2655 | */ | |
2656 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2657 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 2658 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
2659 | goto error_return; |
2660 | } | |
2661 | ||
1da177e4 | 2662 | if (resblks == 0 && |
f6c2d1fa | 2663 | (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen))) |
1da177e4 LT |
2664 | goto error_return; |
2665 | ||
2666 | XFS_BMAP_INIT(&free_list, &first_block); | |
2667 | ||
f6c2d1fa | 2668 | error = xfs_dir_createname(tp, tdp, target_name, target_namelen, |
1da177e4 LT |
2669 | sip->i_ino, &first_block, &free_list, |
2670 | resblks); | |
2671 | if (error) | |
2672 | goto abort_return; | |
2673 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2674 | tdp->i_gen++; | |
2675 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
2676 | ||
2677 | error = xfs_bumplink(tp, sip); | |
b71d300c | 2678 | if (error) |
1da177e4 | 2679 | goto abort_return; |
1da177e4 LT |
2680 | |
2681 | /* | |
2682 | * If this is a synchronous mount, make sure that the | |
2683 | * link transaction goes to disk before returning to | |
2684 | * the user. | |
2685 | */ | |
2686 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2687 | xfs_trans_set_sync(tp); | |
2688 | } | |
2689 | ||
f7c99b6f | 2690 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2691 | if (error) { |
2692 | xfs_bmap_cancel(&free_list); | |
2693 | goto abort_return; | |
2694 | } | |
2695 | ||
1c72bf90 | 2696 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
b71d300c | 2697 | if (error) |
1da177e4 | 2698 | goto std_return; |
1da177e4 LT |
2699 | |
2700 | /* Fall through to std_return with error = 0. */ | |
2701 | std_return: | |
eb9df39d | 2702 | if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) { |
1da177e4 LT |
2703 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, |
2704 | target_dir_vp, DM_RIGHT_NULL, | |
2705 | src_vp, DM_RIGHT_NULL, | |
2706 | target_name, NULL, 0, error, 0); | |
2707 | } | |
2708 | return error; | |
2709 | ||
2710 | abort_return: | |
2711 | cancel_flags |= XFS_TRANS_ABORT; | |
2712 | /* FALLTHROUGH */ | |
014c2544 | 2713 | |
1da177e4 LT |
2714 | error_return: |
2715 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2716 | goto std_return; |
2717 | } | |
b71d300c AD |
2718 | |
2719 | ||
993386c1 | 2720 | int |
1da177e4 | 2721 | xfs_mkdir( |
993386c1 | 2722 | xfs_inode_t *dp, |
8285fb58 NS |
2723 | bhv_vname_t *dentry, |
2724 | bhv_vattr_t *vap, | |
67fcaa73 | 2725 | bhv_vnode_t **vpp, |
1da177e4 LT |
2726 | cred_t *credp) |
2727 | { | |
993386c1 | 2728 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2729 | char *dir_name = VNAME(dentry); |
993386c1 CH |
2730 | int dir_namelen = VNAMELEN(dentry); |
2731 | xfs_mount_t *mp = dp->i_mount; | |
1da177e4 | 2732 | xfs_inode_t *cdp; /* inode of created dir */ |
67fcaa73 | 2733 | bhv_vnode_t *cvp; /* vnode of created dir */ |
1da177e4 | 2734 | xfs_trans_t *tp; |
1da177e4 LT |
2735 | int cancel_flags; |
2736 | int error; | |
2737 | int committed; | |
2738 | xfs_bmap_free_t free_list; | |
2739 | xfs_fsblock_t first_block; | |
993386c1 | 2740 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2741 | boolean_t created = B_FALSE; |
2742 | int dm_event_sent = 0; | |
2743 | xfs_prid_t prid; | |
2744 | struct xfs_dquot *udqp, *gdqp; | |
2745 | uint resblks; | |
2746 | int dm_di_mode; | |
1da177e4 LT |
2747 | |
2748 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2749 | return XFS_ERROR(EIO); | |
2750 | ||
1da177e4 | 2751 | tp = NULL; |
0432dab2 | 2752 | dm_di_mode = vap->va_mode; |
1da177e4 | 2753 | |
eb9df39d | 2754 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 LT |
2755 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
2756 | dir_vp, DM_RIGHT_NULL, NULL, | |
2757 | DM_RIGHT_NULL, dir_name, NULL, | |
2758 | dm_di_mode, 0, 0); | |
2759 | if (error) | |
2760 | return error; | |
2761 | dm_event_sent = 1; | |
2762 | } | |
2763 | ||
2764 | /* Return through std_return after this point. */ | |
2765 | ||
2766 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
2767 | ||
2768 | mp = dp->i_mount; | |
2769 | udqp = gdqp = NULL; | |
365ca83d NS |
2770 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2771 | prid = dp->i_d.di_projid; | |
2772 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
2773 | prid = (xfs_prid_t)vap->va_projid; |
2774 | else | |
2775 | prid = (xfs_prid_t)dfltprid; | |
2776 | ||
2777 | /* | |
2778 | * Make sure that we have allocated dquot(s) on disk. | |
2779 | */ | |
2780 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 2781 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
2782 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2783 | if (error) | |
2784 | goto std_return; | |
2785 | ||
2786 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
2787 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2788 | resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen); | |
2789 | error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0, | |
2790 | XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT); | |
2791 | if (error == ENOSPC) { | |
2792 | resblks = 0; | |
2793 | error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0, | |
2794 | XFS_TRANS_PERM_LOG_RES, | |
2795 | XFS_MKDIR_LOG_COUNT); | |
2796 | } | |
2797 | if (error) { | |
2798 | cancel_flags = 0; | |
1da177e4 LT |
2799 | goto error_return; |
2800 | } | |
2801 | ||
f7c66ce3 | 2802 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2803 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2804 | |
2805 | /* | |
2806 | * Check for directory link count overflow. | |
2807 | */ | |
2808 | if (dp->i_d.di_nlink >= XFS_MAXLINK) { | |
2809 | error = XFS_ERROR(EMLINK); | |
2810 | goto error_return; | |
2811 | } | |
2812 | ||
2813 | /* | |
2814 | * Reserve disk quota and the inode. | |
2815 | */ | |
2816 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
2817 | if (error) | |
2818 | goto error_return; | |
2819 | ||
2820 | if (resblks == 0 && | |
f6c2d1fa | 2821 | (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen))) |
1da177e4 LT |
2822 | goto error_return; |
2823 | /* | |
2824 | * create the directory inode. | |
2825 | */ | |
0432dab2 | 2826 | error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2, |
1da177e4 LT |
2827 | 0, credp, prid, resblks > 0, |
2828 | &cdp, NULL); | |
2829 | if (error) { | |
2830 | if (error == ENOSPC) | |
2831 | goto error_return; | |
2832 | goto abort_return; | |
2833 | } | |
2834 | ITRACE(cdp); | |
2835 | ||
2836 | /* | |
2837 | * Now we add the directory inode to the transaction. | |
2838 | * We waited until now since xfs_dir_ialloc might start | |
2839 | * a new transaction. Had we joined the transaction | |
993386c1 CH |
2840 | * earlier, the locks might have gotten released. An error |
2841 | * from here on will result in the transaction cancel | |
2842 | * unlocking dp so don't do it explicitly in the error path. | |
1da177e4 LT |
2843 | */ |
2844 | VN_HOLD(dir_vp); | |
2845 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 2846 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2847 | |
2848 | XFS_BMAP_INIT(&free_list, &first_block); | |
2849 | ||
f6c2d1fa NS |
2850 | error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino, |
2851 | &first_block, &free_list, resblks ? | |
2852 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
2853 | if (error) { |
2854 | ASSERT(error != ENOSPC); | |
2855 | goto error1; | |
2856 | } | |
2857 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2858 | ||
2859 | /* | |
2860 | * Bump the in memory version number of the parent directory | |
2861 | * so that other processes accessing it will recognize that | |
2862 | * the directory has changed. | |
2863 | */ | |
2864 | dp->i_gen++; | |
2865 | ||
f6c2d1fa NS |
2866 | error = xfs_dir_init(tp, cdp, dp); |
2867 | if (error) | |
1da177e4 | 2868 | goto error2; |
1da177e4 LT |
2869 | |
2870 | cdp->i_gen = 1; | |
2871 | error = xfs_bumplink(tp, dp); | |
f6c2d1fa | 2872 | if (error) |
1da177e4 | 2873 | goto error2; |
1da177e4 LT |
2874 | |
2875 | cvp = XFS_ITOV(cdp); | |
2876 | ||
2877 | created = B_TRUE; | |
2878 | ||
2879 | *vpp = cvp; | |
2880 | IHOLD(cdp); | |
2881 | ||
2882 | /* | |
2883 | * Attach the dquots to the new inode and modify the icount incore. | |
2884 | */ | |
2885 | XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp); | |
2886 | ||
2887 | /* | |
2888 | * If this is a synchronous mount, make sure that the | |
2889 | * mkdir transaction goes to disk before returning to | |
2890 | * the user. | |
2891 | */ | |
2892 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2893 | xfs_trans_set_sync(tp); | |
2894 | } | |
2895 | ||
f7c99b6f | 2896 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2897 | if (error) { |
2898 | IRELE(cdp); | |
2899 | goto error2; | |
2900 | } | |
2901 | ||
1c72bf90 | 2902 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2903 | XFS_QM_DQRELE(mp, udqp); |
2904 | XFS_QM_DQRELE(mp, gdqp); | |
2905 | if (error) { | |
2906 | IRELE(cdp); | |
2907 | } | |
2908 | ||
2909 | /* Fall through to std_return with error = 0 or errno from | |
2910 | * xfs_trans_commit. */ | |
2911 | ||
2912 | std_return: | |
eb9df39d | 2913 | if ((created || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 2914 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 LT |
2915 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
2916 | dir_vp, DM_RIGHT_NULL, | |
2917 | created ? XFS_ITOV(cdp):NULL, | |
2918 | DM_RIGHT_NULL, | |
2919 | dir_name, NULL, | |
2920 | dm_di_mode, error, 0); | |
2921 | } | |
2922 | return error; | |
2923 | ||
2924 | error2: | |
2925 | error1: | |
2926 | xfs_bmap_cancel(&free_list); | |
2927 | abort_return: | |
2928 | cancel_flags |= XFS_TRANS_ABORT; | |
2929 | error_return: | |
2930 | xfs_trans_cancel(tp, cancel_flags); | |
2931 | XFS_QM_DQRELE(mp, udqp); | |
2932 | XFS_QM_DQRELE(mp, gdqp); | |
2933 | ||
993386c1 | 2934 | if (unlock_dp_on_error) |
1da177e4 | 2935 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2936 | |
2937 | goto std_return; | |
2938 | } | |
2939 | ||
993386c1 | 2940 | int |
1da177e4 | 2941 | xfs_rmdir( |
993386c1 CH |
2942 | xfs_inode_t *dp, |
2943 | bhv_vname_t *dentry) | |
1da177e4 | 2944 | { |
993386c1 | 2945 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2946 | char *name = VNAME(dentry); |
993386c1 CH |
2947 | int namelen = VNAMELEN(dentry); |
2948 | xfs_mount_t *mp = dp->i_mount; | |
2949 | xfs_inode_t *cdp; /* child directory */ | |
1da177e4 | 2950 | xfs_trans_t *tp; |
1da177e4 LT |
2951 | int error; |
2952 | xfs_bmap_free_t free_list; | |
2953 | xfs_fsblock_t first_block; | |
2954 | int cancel_flags; | |
2955 | int committed; | |
17370097 | 2956 | int dm_di_mode = S_IFDIR; |
1da177e4 | 2957 | int last_cdp_link; |
1da177e4 LT |
2958 | uint resblks; |
2959 | ||
1da177e4 LT |
2960 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); |
2961 | ||
993386c1 | 2962 | if (XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 | 2963 | return XFS_ERROR(EIO); |
1da177e4 | 2964 | |
17370097 VA |
2965 | if (!xfs_get_dir_entry(dentry, &cdp)) { |
2966 | dm_di_mode = cdp->i_d.di_mode; | |
2967 | IRELE(cdp); | |
2968 | } | |
2969 | ||
eb9df39d | 2970 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
1da177e4 LT |
2971 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, |
2972 | dir_vp, DM_RIGHT_NULL, | |
2973 | NULL, DM_RIGHT_NULL, | |
17370097 | 2974 | name, NULL, dm_di_mode, 0, 0); |
1da177e4 LT |
2975 | if (error) |
2976 | return XFS_ERROR(error); | |
2977 | } | |
2978 | ||
2979 | /* Return through std_return after this point. */ | |
2980 | ||
2981 | cdp = NULL; | |
2982 | ||
2983 | /* | |
2984 | * We need to get a reference to cdp before we get our log | |
2985 | * reservation. The reason for this is that we cannot call | |
2986 | * xfs_iget for an inode for which we do not have a reference | |
2987 | * once we've acquired a log reservation. This is because the | |
2988 | * inode we are trying to get might be in xfs_inactive going | |
2989 | * for a log reservation. Since we'll have to wait for the | |
2990 | * inactive code to complete before returning from xfs_iget, | |
2991 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2992 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2993 | * to the inode before getting our log reservation. |
2994 | */ | |
2995 | error = xfs_get_dir_entry(dentry, &cdp); | |
2996 | if (error) { | |
2997 | REMOVE_DEBUG_TRACE(__LINE__); | |
2998 | goto std_return; | |
2999 | } | |
3000 | mp = dp->i_mount; | |
3001 | dm_di_mode = cdp->i_d.di_mode; | |
3002 | ||
3003 | /* | |
3004 | * Get the dquots for the inodes. | |
3005 | */ | |
3006 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
3007 | if (!error && dp != cdp) | |
3008 | error = XFS_QM_DQATTACH(mp, cdp, 0); | |
3009 | if (error) { | |
3010 | IRELE(cdp); | |
3011 | REMOVE_DEBUG_TRACE(__LINE__); | |
3012 | goto std_return; | |
3013 | } | |
3014 | ||
3015 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
3016 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3017 | /* | |
3018 | * We try to get the real space reservation first, | |
3019 | * allowing for directory btree deletion(s) implying | |
3020 | * possible bmap insert(s). If we can't get the space | |
3021 | * reservation then we use 0 instead, and avoid the bmap | |
3022 | * btree insert(s) in the directory code by, if the bmap | |
3023 | * insert tries to happen, instead trimming the LAST | |
3024 | * block from the directory. | |
3025 | */ | |
3026 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
3027 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
3028 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
3029 | if (error == ENOSPC) { | |
3030 | resblks = 0; | |
3031 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
3032 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
3033 | } | |
3034 | if (error) { | |
3035 | ASSERT(error != ENOSPC); | |
3036 | cancel_flags = 0; | |
3037 | IRELE(cdp); | |
3038 | goto error_return; | |
3039 | } | |
3040 | XFS_BMAP_INIT(&free_list, &first_block); | |
3041 | ||
3042 | /* | |
3043 | * Now lock the child directory inode and the parent directory | |
3044 | * inode in the proper order. This will take care of validating | |
3045 | * that the directory entry for the child directory inode has | |
3046 | * not changed while we were obtaining a log reservation. | |
3047 | */ | |
e9ed9d22 | 3048 | error = xfs_lock_dir_and_entry(dp, cdp); |
1da177e4 LT |
3049 | if (error) { |
3050 | xfs_trans_cancel(tp, cancel_flags); | |
3051 | IRELE(cdp); | |
3052 | goto std_return; | |
3053 | } | |
3054 | ||
3055 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
3056 | if (dp != cdp) { | |
3057 | /* | |
3058 | * Only increment the parent directory vnode count if | |
3059 | * we didn't bump it in looking up cdp. The only time | |
3060 | * we don't bump it is when we're looking up ".". | |
3061 | */ | |
3062 | VN_HOLD(dir_vp); | |
3063 | } | |
3064 | ||
3065 | ITRACE(cdp); | |
3066 | xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL); | |
3067 | ||
3068 | ASSERT(cdp->i_d.di_nlink >= 2); | |
3069 | if (cdp->i_d.di_nlink != 2) { | |
3070 | error = XFS_ERROR(ENOTEMPTY); | |
3071 | goto error_return; | |
3072 | } | |
f6c2d1fa | 3073 | if (!xfs_dir_isempty(cdp)) { |
1da177e4 LT |
3074 | error = XFS_ERROR(ENOTEMPTY); |
3075 | goto error_return; | |
3076 | } | |
3077 | ||
f6c2d1fa NS |
3078 | error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino, |
3079 | &first_block, &free_list, resblks); | |
3080 | if (error) | |
1da177e4 | 3081 | goto error1; |
1da177e4 LT |
3082 | |
3083 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3084 | ||
3085 | /* | |
3086 | * Bump the in memory generation count on the parent | |
3087 | * directory so that other can know that it has changed. | |
3088 | */ | |
3089 | dp->i_gen++; | |
3090 | ||
3091 | /* | |
3092 | * Drop the link from cdp's "..". | |
3093 | */ | |
3094 | error = xfs_droplink(tp, dp); | |
3095 | if (error) { | |
3096 | goto error1; | |
3097 | } | |
3098 | ||
3099 | /* | |
3100 | * Drop the link from dp to cdp. | |
3101 | */ | |
3102 | error = xfs_droplink(tp, cdp); | |
3103 | if (error) { | |
3104 | goto error1; | |
3105 | } | |
3106 | ||
3107 | /* | |
3108 | * Drop the "." link from cdp to self. | |
3109 | */ | |
3110 | error = xfs_droplink(tp, cdp); | |
3111 | if (error) { | |
3112 | goto error1; | |
3113 | } | |
3114 | ||
3115 | /* Determine these before committing transaction */ | |
3116 | last_cdp_link = (cdp)->i_d.di_nlink==0; | |
3117 | ||
3118 | /* | |
3119 | * Take an extra ref on the child vnode so that it | |
3120 | * does not go to xfs_inactive() from within the commit. | |
3121 | */ | |
3122 | IHOLD(cdp); | |
3123 | ||
3124 | /* | |
3125 | * If this is a synchronous mount, make sure that the | |
3126 | * rmdir transaction goes to disk before returning to | |
3127 | * the user. | |
3128 | */ | |
3129 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3130 | xfs_trans_set_sync(tp); | |
3131 | } | |
3132 | ||
f7c99b6f | 3133 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
3134 | if (error) { |
3135 | xfs_bmap_cancel(&free_list); | |
3136 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | | |
3137 | XFS_TRANS_ABORT)); | |
3138 | IRELE(cdp); | |
3139 | goto std_return; | |
3140 | } | |
3141 | ||
1c72bf90 | 3142 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3143 | if (error) { |
3144 | IRELE(cdp); | |
3145 | goto std_return; | |
3146 | } | |
3147 | ||
3148 | ||
3149 | /* | |
3150 | * Let interposed file systems know about removed links. | |
3151 | */ | |
67fcaa73 | 3152 | bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link); |
1da177e4 LT |
3153 | |
3154 | IRELE(cdp); | |
3155 | ||
3156 | /* Fall through to std_return with error = 0 or the errno | |
3157 | * from xfs_trans_commit. */ | |
bb19fba1 | 3158 | std_return: |
eb9df39d | 3159 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 LT |
3160 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
3161 | dir_vp, DM_RIGHT_NULL, | |
3162 | NULL, DM_RIGHT_NULL, | |
3163 | name, NULL, dm_di_mode, | |
3164 | error, 0); | |
3165 | } | |
3166 | return error; | |
3167 | ||
bb19fba1 | 3168 | error1: |
1da177e4 LT |
3169 | xfs_bmap_cancel(&free_list); |
3170 | cancel_flags |= XFS_TRANS_ABORT; | |
014c2544 JJ |
3171 | /* FALLTHROUGH */ |
3172 | ||
bb19fba1 | 3173 | error_return: |
1da177e4 LT |
3174 | xfs_trans_cancel(tp, cancel_flags); |
3175 | goto std_return; | |
3176 | } | |
3177 | ||
993386c1 | 3178 | int |
1da177e4 | 3179 | xfs_symlink( |
993386c1 | 3180 | xfs_inode_t *dp, |
8285fb58 NS |
3181 | bhv_vname_t *dentry, |
3182 | bhv_vattr_t *vap, | |
1da177e4 | 3183 | char *target_path, |
67fcaa73 | 3184 | bhv_vnode_t **vpp, |
1da177e4 LT |
3185 | cred_t *credp) |
3186 | { | |
993386c1 CH |
3187 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
3188 | xfs_mount_t *mp = dp->i_mount; | |
1da177e4 | 3189 | xfs_trans_t *tp; |
1da177e4 LT |
3190 | xfs_inode_t *ip; |
3191 | int error; | |
3192 | int pathlen; | |
3193 | xfs_bmap_free_t free_list; | |
3194 | xfs_fsblock_t first_block; | |
993386c1 | 3195 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
3196 | uint cancel_flags; |
3197 | int committed; | |
3198 | xfs_fileoff_t first_fsb; | |
3199 | xfs_filblks_t fs_blocks; | |
3200 | int nmaps; | |
3201 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
3202 | xfs_daddr_t d; | |
3203 | char *cur_chunk; | |
3204 | int byte_cnt; | |
3205 | int n; | |
3206 | xfs_buf_t *bp; | |
3207 | xfs_prid_t prid; | |
3208 | struct xfs_dquot *udqp, *gdqp; | |
3209 | uint resblks; | |
3210 | char *link_name = VNAME(dentry); | |
3211 | int link_namelen; | |
3212 | ||
3213 | *vpp = NULL; | |
1da177e4 LT |
3214 | error = 0; |
3215 | ip = NULL; | |
3216 | tp = NULL; | |
3217 | ||
3218 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
3219 | ||
1da177e4 LT |
3220 | |
3221 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3222 | return XFS_ERROR(EIO); | |
3223 | ||
3224 | link_namelen = VNAMELEN(dentry); | |
3225 | ||
3226 | /* | |
3227 | * Check component lengths of the target path name. | |
3228 | */ | |
3229 | pathlen = strlen(target_path); | |
3230 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
3231 | return XFS_ERROR(ENAMETOOLONG); | |
3232 | if (pathlen >= MAXNAMELEN) { /* is any component too long? */ | |
3233 | int len, total; | |
3234 | char *path; | |
3235 | ||
f6c2d1fa | 3236 | for (total = 0, path = target_path; total < pathlen;) { |
1da177e4 LT |
3237 | /* |
3238 | * Skip any slashes. | |
3239 | */ | |
3240 | while(*path == '/') { | |
3241 | total++; | |
3242 | path++; | |
3243 | } | |
3244 | ||
3245 | /* | |
3246 | * Count up to the next slash or end of path. | |
3247 | * Error out if the component is bigger than MAXNAMELEN. | |
3248 | */ | |
3249 | for(len = 0; *path != '/' && total < pathlen;total++, path++) { | |
3250 | if (++len >= MAXNAMELEN) { | |
3251 | error = ENAMETOOLONG; | |
3252 | return error; | |
3253 | } | |
3254 | } | |
3255 | } | |
3256 | } | |
3257 | ||
eb9df39d | 3258 | if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) { |
1da177e4 LT |
3259 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp, |
3260 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
3261 | link_name, target_path, 0, 0, 0); | |
3262 | if (error) | |
3263 | return error; | |
3264 | } | |
3265 | ||
3266 | /* Return through std_return after this point. */ | |
3267 | ||
3268 | udqp = gdqp = NULL; | |
365ca83d NS |
3269 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
3270 | prid = dp->i_d.di_projid; | |
3271 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
3272 | prid = (xfs_prid_t)vap->va_projid; |
3273 | else | |
3274 | prid = (xfs_prid_t)dfltprid; | |
3275 | ||
3276 | /* | |
3277 | * Make sure that we have allocated dquot(s) on disk. | |
3278 | */ | |
3279 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 3280 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
3281 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
3282 | if (error) | |
3283 | goto std_return; | |
3284 | ||
3285 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
3286 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3287 | /* | |
3288 | * The symlink will fit into the inode data fork? | |
3289 | * There can't be any attributes so we get the whole variable part. | |
3290 | */ | |
3291 | if (pathlen <= XFS_LITINO(mp)) | |
3292 | fs_blocks = 0; | |
3293 | else | |
3294 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
3295 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks); | |
3296 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, | |
3297 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3298 | if (error == ENOSPC && fs_blocks == 0) { | |
3299 | resblks = 0; | |
3300 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
3301 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3302 | } | |
3303 | if (error) { | |
3304 | cancel_flags = 0; | |
1da177e4 LT |
3305 | goto error_return; |
3306 | } | |
3307 | ||
f7c66ce3 | 3308 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 3309 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
3310 | |
3311 | /* | |
3312 | * Check whether the directory allows new symlinks or not. | |
3313 | */ | |
3314 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
3315 | error = XFS_ERROR(EPERM); | |
3316 | goto error_return; | |
3317 | } | |
3318 | ||
3319 | /* | |
3320 | * Reserve disk quota : blocks and inode. | |
3321 | */ | |
3322 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
3323 | if (error) | |
3324 | goto error_return; | |
3325 | ||
3326 | /* | |
3327 | * Check for ability to enter directory entry, if no space reserved. | |
3328 | */ | |
3329 | if (resblks == 0 && | |
f6c2d1fa | 3330 | (error = xfs_dir_canenter(tp, dp, link_name, link_namelen))) |
1da177e4 LT |
3331 | goto error_return; |
3332 | /* | |
3333 | * Initialize the bmap freelist prior to calling either | |
3334 | * bmapi or the directory create code. | |
3335 | */ | |
3336 | XFS_BMAP_INIT(&free_list, &first_block); | |
3337 | ||
3338 | /* | |
3339 | * Allocate an inode for the symlink. | |
3340 | */ | |
3341 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT), | |
3342 | 1, 0, credp, prid, resblks > 0, &ip, NULL); | |
3343 | if (error) { | |
3344 | if (error == ENOSPC) | |
3345 | goto error_return; | |
3346 | goto error1; | |
3347 | } | |
3348 | ITRACE(ip); | |
3349 | ||
993386c1 CH |
3350 | /* |
3351 | * An error after we've joined dp to the transaction will result in the | |
3352 | * transaction cancel unlocking dp so don't do it explicitly in the | |
3353 | * error path. | |
3354 | */ | |
1da177e4 LT |
3355 | VN_HOLD(dir_vp); |
3356 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 3357 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
3358 | |
3359 | /* | |
3360 | * Also attach the dquot(s) to it, if applicable. | |
3361 | */ | |
3362 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
3363 | ||
3364 | if (resblks) | |
3365 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
3366 | /* | |
3367 | * If the symlink will fit into the inode, write it inline. | |
3368 | */ | |
3369 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
3370 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
3371 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
3372 | ip->i_d.di_size = pathlen; | |
3373 | ||
3374 | /* | |
3375 | * The inode was initially created in extent format. | |
3376 | */ | |
3377 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
3378 | ip->i_df.if_flags |= XFS_IFINLINE; | |
3379 | ||
3380 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
3381 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
3382 | ||
3383 | } else { | |
3384 | first_fsb = 0; | |
3385 | nmaps = SYMLINK_MAPS; | |
3386 | ||
3387 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
3388 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
3389 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 3390 | &free_list, NULL); |
1da177e4 LT |
3391 | if (error) { |
3392 | goto error1; | |
3393 | } | |
3394 | ||
3395 | if (resblks) | |
3396 | resblks -= fs_blocks; | |
3397 | ip->i_d.di_size = pathlen; | |
3398 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3399 | ||
3400 | cur_chunk = target_path; | |
3401 | for (n = 0; n < nmaps; n++) { | |
3402 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
3403 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
3404 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
3405 | BTOBB(byte_cnt), 0); | |
3406 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
3407 | if (pathlen < byte_cnt) { | |
3408 | byte_cnt = pathlen; | |
3409 | } | |
3410 | pathlen -= byte_cnt; | |
3411 | ||
3412 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
3413 | cur_chunk += byte_cnt; | |
3414 | ||
3415 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
3416 | } | |
3417 | } | |
3418 | ||
3419 | /* | |
3420 | * Create the directory entry for the symlink. | |
3421 | */ | |
f6c2d1fa NS |
3422 | error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino, |
3423 | &first_block, &free_list, resblks); | |
3424 | if (error) | |
1da177e4 | 3425 | goto error1; |
1da177e4 LT |
3426 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3427 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
3428 | ||
3429 | /* | |
3430 | * Bump the in memory version number of the parent directory | |
3431 | * so that other processes accessing it will recognize that | |
3432 | * the directory has changed. | |
3433 | */ | |
3434 | dp->i_gen++; | |
3435 | ||
3436 | /* | |
3437 | * If this is a synchronous mount, make sure that the | |
3438 | * symlink transaction goes to disk before returning to | |
3439 | * the user. | |
3440 | */ | |
3441 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3442 | xfs_trans_set_sync(tp); | |
3443 | } | |
3444 | ||
3445 | /* | |
3446 | * xfs_trans_commit normally decrements the vnode ref count | |
3447 | * when it unlocks the inode. Since we want to return the | |
3448 | * vnode to the caller, we bump the vnode ref count now. | |
3449 | */ | |
3450 | IHOLD(ip); | |
3451 | ||
f7c99b6f | 3452 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3453 | if (error) { |
3454 | goto error2; | |
3455 | } | |
1c72bf90 | 3456 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3457 | XFS_QM_DQRELE(mp, udqp); |
3458 | XFS_QM_DQRELE(mp, gdqp); | |
3459 | ||
3460 | /* Fall through to std_return with error = 0 or errno from | |
3461 | * xfs_trans_commit */ | |
3462 | std_return: | |
993386c1 | 3463 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) { |
1da177e4 LT |
3464 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, |
3465 | dir_vp, DM_RIGHT_NULL, | |
3466 | error ? NULL : XFS_ITOV(ip), | |
3467 | DM_RIGHT_NULL, link_name, target_path, | |
3468 | 0, error, 0); | |
3469 | } | |
3470 | ||
3471 | if (!error) { | |
67fcaa73 | 3472 | bhv_vnode_t *vp; |
1da177e4 LT |
3473 | |
3474 | ASSERT(ip); | |
3475 | vp = XFS_ITOV(ip); | |
3476 | *vpp = vp; | |
3477 | } | |
3478 | return error; | |
3479 | ||
3480 | error2: | |
3481 | IRELE(ip); | |
3482 | error1: | |
3483 | xfs_bmap_cancel(&free_list); | |
3484 | cancel_flags |= XFS_TRANS_ABORT; | |
3485 | error_return: | |
3486 | xfs_trans_cancel(tp, cancel_flags); | |
3487 | XFS_QM_DQRELE(mp, udqp); | |
3488 | XFS_QM_DQRELE(mp, gdqp); | |
3489 | ||
993386c1 | 3490 | if (unlock_dp_on_error) |
1da177e4 | 3491 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
3492 | |
3493 | goto std_return; | |
3494 | } | |
3495 | ||
3496 | ||
3497 | /* | |
3498 | * xfs_fid2 | |
3499 | * | |
3500 | * A fid routine that takes a pointer to a previously allocated | |
3501 | * fid structure (like xfs_fast_fid) but uses a 64 bit inode number. | |
3502 | */ | |
993386c1 | 3503 | int |
1da177e4 | 3504 | xfs_fid2( |
993386c1 | 3505 | xfs_inode_t *ip, |
1da177e4 LT |
3506 | fid_t *fidp) |
3507 | { | |
993386c1 | 3508 | xfs_fid2_t *xfid = (xfs_fid2_t *)fidp; |
1da177e4 | 3509 | |
993386c1 | 3510 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
3511 | ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t)); |
3512 | ||
1da177e4 LT |
3513 | xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len); |
3514 | xfid->fid_pad = 0; | |
3515 | /* | |
3516 | * use memcpy because the inode is a long long and there's no | |
3517 | * assurance that xfid->fid_ino is properly aligned. | |
3518 | */ | |
3519 | memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino)); | |
3520 | xfid->fid_gen = ip->i_d.di_gen; | |
3521 | ||
3522 | return 0; | |
3523 | } | |
3524 | ||
3525 | ||
1da177e4 LT |
3526 | int |
3527 | xfs_rwlock( | |
993386c1 | 3528 | xfs_inode_t *ip, |
8285fb58 | 3529 | bhv_vrwlock_t locktype) |
1da177e4 | 3530 | { |
993386c1 | 3531 | if (S_ISDIR(ip->i_d.di_mode)) |
1da177e4 | 3532 | return 1; |
1da177e4 LT |
3533 | if (locktype == VRWLOCK_WRITE) { |
3534 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
3535 | } else if (locktype == VRWLOCK_TRY_READ) { | |
014c2544 | 3536 | return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED); |
1da177e4 | 3537 | } else if (locktype == VRWLOCK_TRY_WRITE) { |
014c2544 | 3538 | return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL); |
1da177e4 LT |
3539 | } else { |
3540 | ASSERT((locktype == VRWLOCK_READ) || | |
3541 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3542 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
3543 | } | |
3544 | ||
3545 | return 1; | |
3546 | } | |
3547 | ||
3548 | ||
1da177e4 LT |
3549 | void |
3550 | xfs_rwunlock( | |
993386c1 | 3551 | xfs_inode_t *ip, |
8285fb58 | 3552 | bhv_vrwlock_t locktype) |
1da177e4 | 3553 | { |
993386c1 CH |
3554 | if (S_ISDIR(ip->i_d.di_mode)) |
3555 | return; | |
1da177e4 LT |
3556 | if (locktype == VRWLOCK_WRITE) { |
3557 | /* | |
3558 | * In the write case, we may have added a new entry to | |
3559 | * the reference cache. This might store a pointer to | |
3560 | * an inode to be released in this inode. If it is there, | |
3561 | * clear the pointer and release the inode after unlocking | |
3562 | * this one. | |
3563 | */ | |
3564 | xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL); | |
3565 | } else { | |
3566 | ASSERT((locktype == VRWLOCK_READ) || | |
3567 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3568 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
3569 | } | |
3570 | return; | |
3571 | } | |
3572 | ||
993386c1 CH |
3573 | |
3574 | int | |
1da177e4 | 3575 | xfs_inode_flush( |
993386c1 | 3576 | xfs_inode_t *ip, |
1da177e4 LT |
3577 | int flags) |
3578 | { | |
993386c1 CH |
3579 | xfs_mount_t *mp = ip->i_mount; |
3580 | xfs_inode_log_item_t *iip = ip->i_itemp; | |
1da177e4 LT |
3581 | int error = 0; |
3582 | ||
1da177e4 LT |
3583 | if (XFS_FORCED_SHUTDOWN(mp)) |
3584 | return XFS_ERROR(EIO); | |
3585 | ||
3586 | /* | |
3587 | * Bypass inodes which have already been cleaned by | |
3588 | * the inode flush clustering code inside xfs_iflush | |
3589 | */ | |
3590 | if ((ip->i_update_core == 0) && | |
3591 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) | |
3592 | return 0; | |
3593 | ||
3594 | if (flags & FLUSH_LOG) { | |
3595 | if (iip && iip->ili_last_lsn) { | |
3596 | xlog_t *log = mp->m_log; | |
3597 | xfs_lsn_t sync_lsn; | |
3598 | int s, log_flags = XFS_LOG_FORCE; | |
3599 | ||
3600 | s = GRANT_LOCK(log); | |
3601 | sync_lsn = log->l_last_sync_lsn; | |
3602 | GRANT_UNLOCK(log, s); | |
3603 | ||
776a75fa LM |
3604 | if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) > 0)) { |
3605 | if (flags & FLUSH_SYNC) | |
3606 | log_flags |= XFS_LOG_SYNC; | |
3607 | error = xfs_log_force(mp, iip->ili_last_lsn, log_flags); | |
3608 | if (error) | |
3609 | return error; | |
3610 | } | |
1da177e4 | 3611 | |
776a75fa LM |
3612 | if (ip->i_update_core == 0) |
3613 | return 0; | |
1da177e4 LT |
3614 | } |
3615 | } | |
3616 | ||
3617 | /* | |
3618 | * We make this non-blocking if the inode is contended, | |
3619 | * return EAGAIN to indicate to the caller that they | |
3620 | * did not succeed. This prevents the flush path from | |
3621 | * blocking on inodes inside another operation right | |
3622 | * now, they get caught later by xfs_sync. | |
3623 | */ | |
3624 | if (flags & FLUSH_INODE) { | |
3625 | int flush_flags; | |
3626 | ||
1da177e4 LT |
3627 | if (flags & FLUSH_SYNC) { |
3628 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3629 | xfs_iflock(ip); | |
3630 | } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
3631 | if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) { | |
3632 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3633 | return EAGAIN; | |
3634 | } | |
3635 | } else { | |
3636 | return EAGAIN; | |
3637 | } | |
3638 | ||
3639 | if (flags & FLUSH_SYNC) | |
3640 | flush_flags = XFS_IFLUSH_SYNC; | |
3641 | else | |
3642 | flush_flags = XFS_IFLUSH_ASYNC; | |
3643 | ||
3644 | error = xfs_iflush(ip, flush_flags); | |
3645 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3646 | } | |
3647 | ||
3648 | return error; | |
3649 | } | |
3650 | ||
993386c1 | 3651 | |
1da177e4 | 3652 | int |
993386c1 CH |
3653 | xfs_set_dmattrs( |
3654 | xfs_inode_t *ip, | |
1da177e4 | 3655 | u_int evmask, |
993386c1 | 3656 | u_int16_t state) |
1da177e4 | 3657 | { |
993386c1 | 3658 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 3659 | xfs_trans_t *tp; |
1da177e4 LT |
3660 | int error; |
3661 | ||
3662 | if (!capable(CAP_SYS_ADMIN)) | |
3663 | return XFS_ERROR(EPERM); | |
3664 | ||
1da177e4 LT |
3665 | if (XFS_FORCED_SHUTDOWN(mp)) |
3666 | return XFS_ERROR(EIO); | |
3667 | ||
3668 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
3669 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
3670 | if (error) { | |
3671 | xfs_trans_cancel(tp, 0); | |
3672 | return error; | |
3673 | } | |
3674 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3675 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3676 | ||
3677 | ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask; | |
3678 | ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state; | |
3679 | ||
3680 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3681 | IHOLD(ip); | |
1c72bf90 | 3682 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
3683 | |
3684 | return error; | |
3685 | } | |
3686 | ||
993386c1 | 3687 | int |
1da177e4 | 3688 | xfs_reclaim( |
993386c1 | 3689 | xfs_inode_t *ip) |
1da177e4 | 3690 | { |
993386c1 | 3691 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1da177e4 LT |
3692 | |
3693 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
3694 | ||
3695 | ASSERT(!VN_MAPPED(vp)); | |
3696 | ||
3697 | /* bad inode, get out here ASAP */ | |
3698 | if (VN_BAD(vp)) { | |
3699 | xfs_ireclaim(ip); | |
3700 | return 0; | |
3701 | } | |
3702 | ||
51c91ed5 | 3703 | vn_iowait(vp); |
1da177e4 | 3704 | |
51c91ed5 | 3705 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); |
1da177e4 | 3706 | |
42fe2b1f CH |
3707 | /* |
3708 | * Make sure the atime in the XFS inode is correct before freeing the | |
3709 | * Linux inode. | |
3710 | */ | |
3711 | xfs_synchronize_atime(ip); | |
1da177e4 | 3712 | |
4c60658e DC |
3713 | /* |
3714 | * If we have nothing to flush with this inode then complete the | |
3715 | * teardown now, otherwise break the link between the xfs inode and the | |
3716 | * linux inode and clean up the xfs inode later. This avoids flushing | |
3717 | * the inode to disk during the delete operation itself. | |
3718 | * | |
3719 | * When breaking the link, we need to set the XFS_IRECLAIMABLE flag | |
3720 | * first to ensure that xfs_iunpin() will never see an xfs inode | |
3721 | * that has a linux inode being reclaimed. Synchronisation is provided | |
3722 | * by the i_flags_lock. | |
1da177e4 LT |
3723 | */ |
3724 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
3725 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3726 | xfs_iflock(ip); | |
3727 | return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); | |
3728 | } else { | |
3729 | xfs_mount_t *mp = ip->i_mount; | |
3730 | ||
4c60658e | 3731 | /* Protect sync and unpin from us */ |
1da177e4 | 3732 | XFS_MOUNT_ILOCK(mp); |
4c60658e DC |
3733 | spin_lock(&ip->i_flags_lock); |
3734 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
1da177e4 | 3735 | vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip)); |
4c60658e | 3736 | spin_unlock(&ip->i_flags_lock); |
1da177e4 | 3737 | list_add_tail(&ip->i_reclaim, &mp->m_del_inodes); |
1da177e4 LT |
3738 | XFS_MOUNT_IUNLOCK(mp); |
3739 | } | |
3740 | return 0; | |
3741 | } | |
3742 | ||
3743 | int | |
3744 | xfs_finish_reclaim( | |
3745 | xfs_inode_t *ip, | |
3746 | int locked, | |
3747 | int sync_mode) | |
3748 | { | |
da353b0d | 3749 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); |
67fcaa73 | 3750 | bhv_vnode_t *vp = XFS_ITOV_NULL(ip); |
1da177e4 LT |
3751 | int error; |
3752 | ||
3753 | if (vp && VN_BAD(vp)) | |
3754 | goto reclaim; | |
3755 | ||
3756 | /* The hash lock here protects a thread in xfs_iget_core from | |
3757 | * racing with us on linking the inode back with a vnode. | |
3758 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
3759 | * us. | |
3760 | */ | |
da353b0d | 3761 | write_lock(&pag->pag_ici_lock); |
f273ab84 | 3762 | spin_lock(&ip->i_flags_lock); |
7a18c386 DC |
3763 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || |
3764 | (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) { | |
f273ab84 | 3765 | spin_unlock(&ip->i_flags_lock); |
da353b0d | 3766 | write_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
3767 | if (locked) { |
3768 | xfs_ifunlock(ip); | |
3769 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3770 | } | |
014c2544 | 3771 | return 1; |
1da177e4 | 3772 | } |
7a18c386 | 3773 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
f273ab84 | 3774 | spin_unlock(&ip->i_flags_lock); |
da353b0d DC |
3775 | write_unlock(&pag->pag_ici_lock); |
3776 | xfs_put_perag(ip->i_mount, pag); | |
1da177e4 LT |
3777 | |
3778 | /* | |
3779 | * If the inode is still dirty, then flush it out. If the inode | |
3780 | * is not in the AIL, then it will be OK to flush it delwri as | |
3781 | * long as xfs_iflush() does not keep any references to the inode. | |
3782 | * We leave that decision up to xfs_iflush() since it has the | |
3783 | * knowledge of whether it's OK to simply do a delwri flush of | |
3784 | * the inode or whether we need to wait until the inode is | |
3785 | * pulled from the AIL. | |
3786 | * We get the flush lock regardless, though, just to make sure | |
3787 | * we don't free it while it is being flushed. | |
3788 | */ | |
3789 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
3790 | if (!locked) { | |
3791 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3792 | xfs_iflock(ip); | |
3793 | } | |
3794 | ||
3795 | if (ip->i_update_core || | |
3796 | ((ip->i_itemp != NULL) && | |
3797 | (ip->i_itemp->ili_format.ilf_fields != 0))) { | |
3798 | error = xfs_iflush(ip, sync_mode); | |
3799 | /* | |
3800 | * If we hit an error, typically because of filesystem | |
3801 | * shutdown, we don't need to let vn_reclaim to know | |
3802 | * because we're gonna reclaim the inode anyway. | |
3803 | */ | |
3804 | if (error) { | |
3805 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3806 | goto reclaim; | |
3807 | } | |
3808 | xfs_iflock(ip); /* synchronize with xfs_iflush_done */ | |
3809 | } | |
3810 | ||
3811 | ASSERT(ip->i_update_core == 0); | |
3812 | ASSERT(ip->i_itemp == NULL || | |
3813 | ip->i_itemp->ili_format.ilf_fields == 0); | |
3814 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3815 | } else if (locked) { | |
3816 | /* | |
3817 | * We are not interested in doing an iflush if we're | |
3818 | * in the process of shutting down the filesystem forcibly. | |
3819 | * So, just reclaim the inode. | |
3820 | */ | |
3821 | xfs_ifunlock(ip); | |
3822 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3823 | } | |
3824 | ||
3825 | reclaim: | |
3826 | xfs_ireclaim(ip); | |
3827 | return 0; | |
3828 | } | |
3829 | ||
3830 | int | |
3831 | xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock) | |
3832 | { | |
3833 | int purged; | |
3834 | xfs_inode_t *ip, *n; | |
3835 | int done = 0; | |
3836 | ||
3837 | while (!done) { | |
3838 | purged = 0; | |
3839 | XFS_MOUNT_ILOCK(mp); | |
3840 | list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) { | |
3841 | if (noblock) { | |
3842 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) | |
3843 | continue; | |
3844 | if (xfs_ipincount(ip) || | |
3845 | !xfs_iflock_nowait(ip)) { | |
3846 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3847 | continue; | |
3848 | } | |
3849 | } | |
3850 | XFS_MOUNT_IUNLOCK(mp); | |
6b2cf618 FB |
3851 | if (xfs_finish_reclaim(ip, noblock, |
3852 | XFS_IFLUSH_DELWRI_ELSE_ASYNC)) | |
3853 | delay(1); | |
1da177e4 LT |
3854 | purged = 1; |
3855 | break; | |
3856 | } | |
3857 | ||
3858 | done = !purged; | |
3859 | } | |
3860 | ||
3861 | XFS_MOUNT_IUNLOCK(mp); | |
3862 | return 0; | |
3863 | } | |
3864 | ||
3865 | /* | |
3866 | * xfs_alloc_file_space() | |
3867 | * This routine allocates disk space for the given file. | |
3868 | * | |
3869 | * If alloc_type == 0, this request is for an ALLOCSP type | |
3870 | * request which will change the file size. In this case, no | |
3871 | * DMAPI event will be generated by the call. A TRUNCATE event | |
3872 | * will be generated later by xfs_setattr. | |
3873 | * | |
3874 | * If alloc_type != 0, this request is for a RESVSP type | |
3875 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
3876 | * lower block boundary byte address is less than the file's | |
3877 | * length. | |
3878 | * | |
3879 | * RETURNS: | |
3880 | * 0 on success | |
3881 | * errno on error | |
3882 | * | |
3883 | */ | |
ba0f32d4 | 3884 | STATIC int |
1da177e4 LT |
3885 | xfs_alloc_file_space( |
3886 | xfs_inode_t *ip, | |
3887 | xfs_off_t offset, | |
3888 | xfs_off_t len, | |
3889 | int alloc_type, | |
3890 | int attr_flags) | |
3891 | { | |
dd9f438e NS |
3892 | xfs_mount_t *mp = ip->i_mount; |
3893 | xfs_off_t count; | |
1da177e4 LT |
3894 | xfs_filblks_t allocated_fsb; |
3895 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
3896 | xfs_extlen_t extsz, temp; |
3897 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 3898 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
3899 | int nimaps; |
3900 | int bmapi_flag; | |
3901 | int quota_flag; | |
1da177e4 | 3902 | int rt; |
1da177e4 | 3903 | xfs_trans_t *tp; |
dd9f438e NS |
3904 | xfs_bmbt_irec_t imaps[1], *imapp; |
3905 | xfs_bmap_free_t free_list; | |
3906 | uint qblocks, resblks, resrtextents; | |
3907 | int committed; | |
3908 | int error; | |
1da177e4 LT |
3909 | |
3910 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
1da177e4 LT |
3911 | |
3912 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3913 | return XFS_ERROR(EIO); | |
3914 | ||
1da177e4 LT |
3915 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
3916 | return error; | |
3917 | ||
3918 | if (len <= 0) | |
3919 | return XFS_ERROR(EINVAL); | |
3920 | ||
957d0ebe DC |
3921 | rt = XFS_IS_REALTIME_INODE(ip); |
3922 | extsz = xfs_get_extsz_hint(ip); | |
3923 | ||
1da177e4 | 3924 | count = len; |
1da177e4 | 3925 | imapp = &imaps[0]; |
dd9f438e NS |
3926 | nimaps = 1; |
3927 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
3928 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
3929 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
3930 | ||
3931 | /* Generate a DMAPI event if needed. */ | |
ba87ea69 | 3932 | if (alloc_type != 0 && offset < ip->i_size && |
1da177e4 | 3933 | (attr_flags&ATTR_DMI) == 0 && |
eb9df39d | 3934 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
1da177e4 LT |
3935 | xfs_off_t end_dmi_offset; |
3936 | ||
3937 | end_dmi_offset = offset+len; | |
ba87ea69 LM |
3938 | if (end_dmi_offset > ip->i_size) |
3939 | end_dmi_offset = ip->i_size; | |
1da177e4 LT |
3940 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip), |
3941 | offset, end_dmi_offset - offset, | |
3942 | 0, NULL); | |
3943 | if (error) | |
014c2544 | 3944 | return error; |
1da177e4 LT |
3945 | } |
3946 | ||
3947 | /* | |
dd9f438e | 3948 | * Allocate file space until done or until there is an error |
1da177e4 LT |
3949 | */ |
3950 | retry: | |
3951 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
3952 | xfs_fileoff_t s, e; |
3953 | ||
1da177e4 | 3954 | /* |
3ddb8fa9 | 3955 | * Determine space reservations for data/realtime. |
1da177e4 | 3956 | */ |
dd9f438e | 3957 | if (unlikely(extsz)) { |
1da177e4 | 3958 | s = startoffset_fsb; |
dd9f438e NS |
3959 | do_div(s, extsz); |
3960 | s *= extsz; | |
3961 | e = startoffset_fsb + allocatesize_fsb; | |
3962 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
3963 | e += temp; | |
3964 | if ((temp = do_mod(e, extsz))) | |
3965 | e += extsz - temp; | |
3966 | } else { | |
3967 | s = 0; | |
3968 | e = allocatesize_fsb; | |
3969 | } | |
3970 | ||
3971 | if (unlikely(rt)) { | |
3972 | resrtextents = qblocks = (uint)(e - s); | |
3973 | resrtextents /= mp->m_sb.sb_rextsize; | |
3974 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
3975 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 3976 | } else { |
dd9f438e NS |
3977 | resrtextents = 0; |
3978 | resblks = qblocks = \ | |
3979 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
3980 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
3981 | } |
3982 | ||
3983 | /* | |
dd9f438e | 3984 | * Allocate and setup the transaction. |
1da177e4 LT |
3985 | */ |
3986 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
3987 | error = xfs_trans_reserve(tp, resblks, |
3988 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
3989 | XFS_TRANS_PERM_LOG_RES, |
3990 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 3991 | /* |
dd9f438e | 3992 | * Check for running out of space |
1da177e4 LT |
3993 | */ |
3994 | if (error) { | |
3995 | /* | |
3996 | * Free the transaction structure. | |
3997 | */ | |
3998 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
3999 | xfs_trans_cancel(tp, 0); | |
4000 | break; | |
4001 | } | |
4002 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
dd9f438e NS |
4003 | error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, |
4004 | qblocks, 0, quota_flag); | |
1da177e4 LT |
4005 | if (error) |
4006 | goto error1; | |
4007 | ||
4008 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4009 | xfs_trans_ihold(tp, ip); | |
4010 | ||
4011 | /* | |
dd9f438e | 4012 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 LT |
4013 | */ |
4014 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
3e57ecf6 | 4015 | error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb, |
dd9f438e NS |
4016 | allocatesize_fsb, bmapi_flag, |
4017 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 4018 | &free_list, NULL); |
1da177e4 LT |
4019 | if (error) { |
4020 | goto error0; | |
4021 | } | |
4022 | ||
4023 | /* | |
dd9f438e | 4024 | * Complete the transaction |
1da177e4 | 4025 | */ |
f7c99b6f | 4026 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
4027 | if (error) { |
4028 | goto error0; | |
4029 | } | |
4030 | ||
1c72bf90 | 4031 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
4032 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
4033 | if (error) { | |
4034 | break; | |
4035 | } | |
4036 | ||
4037 | allocated_fsb = imapp->br_blockcount; | |
4038 | ||
dd9f438e | 4039 | if (nimaps == 0) { |
1da177e4 LT |
4040 | error = XFS_ERROR(ENOSPC); |
4041 | break; | |
4042 | } | |
4043 | ||
4044 | startoffset_fsb += allocated_fsb; | |
4045 | allocatesize_fsb -= allocated_fsb; | |
4046 | } | |
4047 | dmapi_enospc_check: | |
eb9df39d CH |
4048 | if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 && |
4049 | DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) { | |
1da177e4 LT |
4050 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, |
4051 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
4052 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
4053 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ | |
4054 | if (error == 0) | |
4055 | goto retry; /* Maybe DMAPI app. has made space */ | |
4056 | /* else fall through with error from XFS_SEND_DATA */ | |
4057 | } | |
4058 | ||
4059 | return error; | |
4060 | ||
dd9f438e | 4061 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 4062 | xfs_bmap_cancel(&free_list); |
dd9f438e NS |
4063 | XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag); |
4064 | ||
4065 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
4066 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
4067 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4068 | goto dmapi_enospc_check; | |
4069 | } | |
4070 | ||
4071 | /* | |
4072 | * Zero file bytes between startoff and endoff inclusive. | |
4073 | * The iolock is held exclusive and no blocks are buffered. | |
4074 | */ | |
4075 | STATIC int | |
4076 | xfs_zero_remaining_bytes( | |
4077 | xfs_inode_t *ip, | |
4078 | xfs_off_t startoff, | |
4079 | xfs_off_t endoff) | |
4080 | { | |
4081 | xfs_bmbt_irec_t imap; | |
4082 | xfs_fileoff_t offset_fsb; | |
4083 | xfs_off_t lastoffset; | |
4084 | xfs_off_t offset; | |
4085 | xfs_buf_t *bp; | |
4086 | xfs_mount_t *mp = ip->i_mount; | |
4087 | int nimap; | |
4088 | int error = 0; | |
4089 | ||
4090 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, | |
4091 | ip->i_d.di_flags & XFS_DIFLAG_REALTIME ? | |
4092 | mp->m_rtdev_targp : mp->m_ddev_targp); | |
4093 | ||
4094 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
4095 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
4096 | nimap = 1; | |
3e57ecf6 OW |
4097 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0, |
4098 | NULL, 0, &imap, &nimap, NULL, NULL); | |
1da177e4 LT |
4099 | if (error || nimap < 1) |
4100 | break; | |
4101 | ASSERT(imap.br_blockcount >= 1); | |
4102 | ASSERT(imap.br_startoff == offset_fsb); | |
4103 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
4104 | if (lastoffset > endoff) | |
4105 | lastoffset = endoff; | |
4106 | if (imap.br_startblock == HOLESTARTBLOCK) | |
4107 | continue; | |
4108 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4109 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
4110 | continue; | |
4111 | XFS_BUF_UNDONE(bp); | |
4112 | XFS_BUF_UNWRITE(bp); | |
4113 | XFS_BUF_READ(bp); | |
4114 | XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock)); | |
4115 | xfsbdstrat(mp, bp); | |
4116 | if ((error = xfs_iowait(bp))) { | |
4117 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", | |
4118 | mp, bp, XFS_BUF_ADDR(bp)); | |
4119 | break; | |
4120 | } | |
4121 | memset(XFS_BUF_PTR(bp) + | |
4122 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
4123 | 0, lastoffset - offset + 1); | |
4124 | XFS_BUF_UNDONE(bp); | |
4125 | XFS_BUF_UNREAD(bp); | |
4126 | XFS_BUF_WRITE(bp); | |
4127 | xfsbdstrat(mp, bp); | |
4128 | if ((error = xfs_iowait(bp))) { | |
4129 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", | |
4130 | mp, bp, XFS_BUF_ADDR(bp)); | |
4131 | break; | |
4132 | } | |
4133 | } | |
4134 | xfs_buf_free(bp); | |
4135 | return error; | |
4136 | } | |
4137 | ||
4138 | /* | |
4139 | * xfs_free_file_space() | |
4140 | * This routine frees disk space for the given file. | |
4141 | * | |
4142 | * This routine is only called by xfs_change_file_space | |
4143 | * for an UNRESVSP type call. | |
4144 | * | |
4145 | * RETURNS: | |
4146 | * 0 on success | |
4147 | * errno on error | |
4148 | * | |
4149 | */ | |
4150 | STATIC int | |
4151 | xfs_free_file_space( | |
4152 | xfs_inode_t *ip, | |
4153 | xfs_off_t offset, | |
4154 | xfs_off_t len, | |
4155 | int attr_flags) | |
4156 | { | |
67fcaa73 | 4157 | bhv_vnode_t *vp; |
1da177e4 LT |
4158 | int committed; |
4159 | int done; | |
4160 | xfs_off_t end_dmi_offset; | |
4161 | xfs_fileoff_t endoffset_fsb; | |
4162 | int error; | |
4163 | xfs_fsblock_t firstfsb; | |
4164 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
4165 | xfs_bmbt_irec_t imap; |
4166 | xfs_off_t ioffset; | |
4167 | xfs_extlen_t mod=0; | |
4168 | xfs_mount_t *mp; | |
4169 | int nimap; | |
4170 | uint resblks; | |
673cdf5c | 4171 | uint rounding; |
1da177e4 LT |
4172 | int rt; |
4173 | xfs_fileoff_t startoffset_fsb; | |
4174 | xfs_trans_t *tp; | |
5fcbab35 | 4175 | int need_iolock = 1; |
1da177e4 | 4176 | |
bd5a876a | 4177 | vp = XFS_ITOV(ip); |
1da177e4 LT |
4178 | mp = ip->i_mount; |
4179 | ||
bd5a876a CH |
4180 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
4181 | ||
1da177e4 LT |
4182 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
4183 | return error; | |
4184 | ||
4185 | error = 0; | |
4186 | if (len <= 0) /* if nothing being freed */ | |
4187 | return error; | |
4188 | rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME); | |
4189 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); | |
4190 | end_dmi_offset = offset + len; | |
4191 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
4192 | ||
eb9df39d CH |
4193 | if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 && |
4194 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { | |
ba87ea69 LM |
4195 | if (end_dmi_offset > ip->i_size) |
4196 | end_dmi_offset = ip->i_size; | |
bd5a876a | 4197 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp, |
1da177e4 LT |
4198 | offset, end_dmi_offset - offset, |
4199 | AT_DELAY_FLAG(attr_flags), NULL); | |
4200 | if (error) | |
014c2544 | 4201 | return error; |
1da177e4 LT |
4202 | } |
4203 | ||
5fcbab35 DR |
4204 | if (attr_flags & ATTR_NOLOCK) |
4205 | need_iolock = 0; | |
9fa8046f | 4206 | if (need_iolock) { |
1da177e4 | 4207 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
9fa8046f YL |
4208 | vn_iowait(vp); /* wait for the completion of any pending DIOs */ |
4209 | } | |
5fcbab35 | 4210 | |
673cdf5c | 4211 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, NBPP); |
1da177e4 | 4212 | ioffset = offset & ~(rounding - 1); |
bd5a876a CH |
4213 | |
4214 | if (VN_CACHED(vp) != 0) { | |
4215 | xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1, | |
4216 | ctooff(offtoct(ioffset)), -1); | |
d3cf2094 | 4217 | error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)), |
bd5a876a | 4218 | -1, FI_REMAPF_LOCKED); |
d3cf2094 LM |
4219 | if (error) |
4220 | goto out_unlock_iolock; | |
bd5a876a CH |
4221 | } |
4222 | ||
1da177e4 LT |
4223 | /* |
4224 | * Need to zero the stuff we're not freeing, on disk. | |
4225 | * If its a realtime file & can't use unwritten extents then we | |
4226 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
4227 | * will take care of it for us. | |
4228 | */ | |
4229 | if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) { | |
4230 | nimap = 1; | |
3e57ecf6 OW |
4231 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb, |
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 | xfs_daddr_t block; | |
4238 | ||
4239 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4240 | block = imap.br_startblock; | |
4241 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
4242 | if (mod) | |
4243 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
4244 | } | |
4245 | nimap = 1; | |
3e57ecf6 OW |
4246 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1, |
4247 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); | |
1da177e4 LT |
4248 | if (error) |
4249 | goto out_unlock_iolock; | |
4250 | ASSERT(nimap == 0 || nimap == 1); | |
4251 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4252 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4253 | mod++; | |
4254 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
4255 | endoffset_fsb -= mod; | |
4256 | } | |
4257 | } | |
4258 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
4259 | /* | |
4260 | * One contiguous piece to clear | |
4261 | */ | |
4262 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
4263 | else { | |
4264 | /* | |
4265 | * Some full blocks, possibly two pieces to clear | |
4266 | */ | |
4267 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
4268 | error = xfs_zero_remaining_bytes(ip, offset, | |
4269 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
4270 | if (!error && | |
4271 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
4272 | error = xfs_zero_remaining_bytes(ip, | |
4273 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
4274 | offset + len - 1); | |
4275 | } | |
4276 | ||
4277 | /* | |
4278 | * free file space until done or until there is an error | |
4279 | */ | |
4280 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
4281 | while (!error && !done) { | |
4282 | ||
4283 | /* | |
91ebecc7 DC |
4284 | * allocate and setup the transaction. Allow this |
4285 | * transaction to dip into the reserve blocks to ensure | |
4286 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
4287 | */ |
4288 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 4289 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
4290 | error = xfs_trans_reserve(tp, |
4291 | resblks, | |
4292 | XFS_WRITE_LOG_RES(mp), | |
4293 | 0, | |
4294 | XFS_TRANS_PERM_LOG_RES, | |
4295 | XFS_WRITE_LOG_COUNT); | |
4296 | ||
4297 | /* | |
4298 | * check for running out of space | |
4299 | */ | |
4300 | if (error) { | |
4301 | /* | |
4302 | * Free the transaction structure. | |
4303 | */ | |
4304 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
4305 | xfs_trans_cancel(tp, 0); | |
4306 | break; | |
4307 | } | |
4308 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4309 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, | |
dd9f438e NS |
4310 | ip->i_udquot, ip->i_gdquot, resblks, 0, |
4311 | XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
4312 | if (error) |
4313 | goto error1; | |
4314 | ||
4315 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4316 | xfs_trans_ihold(tp, ip); | |
4317 | ||
4318 | /* | |
4319 | * issue the bunmapi() call to free the blocks | |
4320 | */ | |
4321 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
3e57ecf6 | 4322 | error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb, |
1da177e4 | 4323 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 4324 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
4325 | if (error) { |
4326 | goto error0; | |
4327 | } | |
4328 | ||
4329 | /* | |
4330 | * complete the transaction | |
4331 | */ | |
f7c99b6f | 4332 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
4333 | if (error) { |
4334 | goto error0; | |
4335 | } | |
4336 | ||
1c72bf90 | 4337 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
4338 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
4339 | } | |
4340 | ||
4341 | out_unlock_iolock: | |
4342 | if (need_iolock) | |
4343 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
4344 | return error; | |
4345 | ||
4346 | error0: | |
4347 | xfs_bmap_cancel(&free_list); | |
4348 | error1: | |
4349 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
4350 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
4351 | XFS_ILOCK_EXCL); | |
4352 | return error; | |
4353 | } | |
4354 | ||
4355 | /* | |
4356 | * xfs_change_file_space() | |
4357 | * This routine allocates or frees disk space for the given file. | |
4358 | * The user specified parameters are checked for alignment and size | |
4359 | * limitations. | |
4360 | * | |
4361 | * RETURNS: | |
4362 | * 0 on success | |
4363 | * errno on error | |
4364 | * | |
4365 | */ | |
4366 | int | |
4367 | xfs_change_file_space( | |
993386c1 | 4368 | xfs_inode_t *ip, |
1da177e4 LT |
4369 | int cmd, |
4370 | xfs_flock64_t *bf, | |
4371 | xfs_off_t offset, | |
4372 | cred_t *credp, | |
4373 | int attr_flags) | |
4374 | { | |
993386c1 | 4375 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
4376 | int clrprealloc; |
4377 | int error; | |
4378 | xfs_fsize_t fsize; | |
1da177e4 LT |
4379 | int setprealloc; |
4380 | xfs_off_t startoffset; | |
4381 | xfs_off_t llen; | |
4382 | xfs_trans_t *tp; | |
8285fb58 | 4383 | bhv_vattr_t va; |
1da177e4 | 4384 | |
993386c1 | 4385 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
4386 | |
4387 | /* | |
4388 | * must be a regular file and have write permission | |
4389 | */ | |
993386c1 | 4390 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
4391 | return XFS_ERROR(EINVAL); |
4392 | ||
4393 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
4394 | ||
4395 | if ((error = xfs_iaccess(ip, S_IWUSR, credp))) { | |
4396 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
4397 | return error; | |
4398 | } | |
4399 | ||
4400 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
4401 | ||
4402 | switch (bf->l_whence) { | |
4403 | case 0: /*SEEK_SET*/ | |
4404 | break; | |
4405 | case 1: /*SEEK_CUR*/ | |
4406 | bf->l_start += offset; | |
4407 | break; | |
4408 | case 2: /*SEEK_END*/ | |
ba87ea69 | 4409 | bf->l_start += ip->i_size; |
1da177e4 LT |
4410 | break; |
4411 | default: | |
4412 | return XFS_ERROR(EINVAL); | |
4413 | } | |
4414 | ||
4415 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
4416 | ||
4417 | if ( (bf->l_start < 0) | |
4418 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
4419 | || (bf->l_start + llen < 0) | |
4420 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
4421 | return XFS_ERROR(EINVAL); | |
4422 | ||
4423 | bf->l_whence = 0; | |
4424 | ||
4425 | startoffset = bf->l_start; | |
ba87ea69 | 4426 | fsize = ip->i_size; |
1da177e4 LT |
4427 | |
4428 | /* | |
4429 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
4430 | * file space. | |
4431 | * These calls do NOT zero the data space allocated to the file, | |
4432 | * nor do they change the file size. | |
4433 | * | |
4434 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
4435 | * space. | |
4436 | * These calls cause the new file data to be zeroed and the file | |
4437 | * size to be changed. | |
4438 | */ | |
4439 | setprealloc = clrprealloc = 0; | |
4440 | ||
4441 | switch (cmd) { | |
4442 | case XFS_IOC_RESVSP: | |
4443 | case XFS_IOC_RESVSP64: | |
4444 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
4445 | 1, attr_flags); | |
4446 | if (error) | |
4447 | return error; | |
4448 | setprealloc = 1; | |
4449 | break; | |
4450 | ||
4451 | case XFS_IOC_UNRESVSP: | |
4452 | case XFS_IOC_UNRESVSP64: | |
4453 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
4454 | attr_flags))) | |
4455 | return error; | |
4456 | break; | |
4457 | ||
4458 | case XFS_IOC_ALLOCSP: | |
4459 | case XFS_IOC_ALLOCSP64: | |
4460 | case XFS_IOC_FREESP: | |
4461 | case XFS_IOC_FREESP64: | |
4462 | if (startoffset > fsize) { | |
4463 | error = xfs_alloc_file_space(ip, fsize, | |
4464 | startoffset - fsize, 0, attr_flags); | |
4465 | if (error) | |
4466 | break; | |
4467 | } | |
4468 | ||
4469 | va.va_mask = XFS_AT_SIZE; | |
4470 | va.va_size = startoffset; | |
4471 | ||
993386c1 | 4472 | error = xfs_setattr(ip, &va, attr_flags, credp); |
1da177e4 LT |
4473 | |
4474 | if (error) | |
4475 | return error; | |
4476 | ||
4477 | clrprealloc = 1; | |
4478 | break; | |
4479 | ||
4480 | default: | |
4481 | ASSERT(0); | |
4482 | return XFS_ERROR(EINVAL); | |
4483 | } | |
4484 | ||
4485 | /* | |
4486 | * update the inode timestamp, mode, and prealloc flag bits | |
4487 | */ | |
4488 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
4489 | ||
4490 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
4491 | 0, 0, 0))) { | |
4492 | /* ASSERT(0); */ | |
4493 | xfs_trans_cancel(tp, 0); | |
4494 | return error; | |
4495 | } | |
4496 | ||
4497 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4498 | ||
4499 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4500 | xfs_trans_ihold(tp, ip); | |
4501 | ||
4502 | if ((attr_flags & ATTR_DMI) == 0) { | |
4503 | ip->i_d.di_mode &= ~S_ISUID; | |
4504 | ||
4505 | /* | |
4506 | * Note that we don't have to worry about mandatory | |
4507 | * file locking being disabled here because we only | |
4508 | * clear the S_ISGID bit if the Group execute bit is | |
4509 | * on, but if it was on then mandatory locking wouldn't | |
4510 | * have been enabled. | |
4511 | */ | |
4512 | if (ip->i_d.di_mode & S_IXGRP) | |
4513 | ip->i_d.di_mode &= ~S_ISGID; | |
4514 | ||
4515 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
4516 | } | |
4517 | if (setprealloc) | |
4518 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
4519 | else if (clrprealloc) | |
4520 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
4521 | ||
4522 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
4523 | xfs_trans_set_sync(tp); | |
4524 | ||
1c72bf90 | 4525 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
4526 | |
4527 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4528 | ||
4529 | return error; | |
4530 | } |