Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
16f7e0fe | 18 | |
1da177e4 | 19 | #include "xfs.h" |
a844f451 | 20 | #include "xfs_fs.h" |
1da177e4 | 21 | #include "xfs_types.h" |
a844f451 | 22 | #include "xfs_bit.h" |
1da177e4 | 23 | #include "xfs_log.h" |
a844f451 | 24 | #include "xfs_inum.h" |
1da177e4 LT |
25 | #include "xfs_trans.h" |
26 | #include "xfs_sb.h" | |
27 | #include "xfs_ag.h" | |
1da177e4 LT |
28 | #include "xfs_dir2.h" |
29 | #include "xfs_dmapi.h" | |
30 | #include "xfs_mount.h" | |
a844f451 | 31 | #include "xfs_da_btree.h" |
1da177e4 | 32 | #include "xfs_bmap_btree.h" |
a844f451 | 33 | #include "xfs_alloc_btree.h" |
1da177e4 | 34 | #include "xfs_ialloc_btree.h" |
1da177e4 | 35 | #include "xfs_dir2_sf.h" |
a844f451 | 36 | #include "xfs_attr_sf.h" |
1da177e4 | 37 | #include "xfs_dinode.h" |
1da177e4 | 38 | #include "xfs_inode.h" |
a844f451 | 39 | #include "xfs_inode_item.h" |
a844f451 NS |
40 | #include "xfs_itable.h" |
41 | #include "xfs_btree.h" | |
42 | #include "xfs_ialloc.h" | |
43 | #include "xfs_alloc.h" | |
1da177e4 | 44 | #include "xfs_bmap.h" |
1da177e4 LT |
45 | #include "xfs_attr.h" |
46 | #include "xfs_rw.h" | |
1da177e4 | 47 | #include "xfs_error.h" |
1da177e4 LT |
48 | #include "xfs_quota.h" |
49 | #include "xfs_utils.h" | |
a844f451 | 50 | #include "xfs_rtalloc.h" |
1da177e4 | 51 | #include "xfs_trans_space.h" |
1da177e4 | 52 | #include "xfs_log_priv.h" |
2a82b8be | 53 | #include "xfs_filestream.h" |
993386c1 | 54 | #include "xfs_vnodeops.h" |
1da177e4 | 55 | |
993386c1 | 56 | int |
1da177e4 | 57 | xfs_open( |
993386c1 | 58 | xfs_inode_t *ip) |
1da177e4 LT |
59 | { |
60 | int mode; | |
1da177e4 LT |
61 | |
62 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
63 | return XFS_ERROR(EIO); | |
64 | ||
65 | /* | |
66 | * If it's a directory with any blocks, read-ahead block 0 | |
67 | * as we're almost certain to have the next operation be a read there. | |
68 | */ | |
993386c1 | 69 | if (S_ISDIR(ip->i_d.di_mode) && ip->i_d.di_nextents > 0) { |
1da177e4 LT |
70 | mode = xfs_ilock_map_shared(ip); |
71 | if (ip->i_d.di_nextents > 0) | |
72 | (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK); | |
73 | xfs_iunlock(ip, mode); | |
74 | } | |
75 | return 0; | |
76 | } | |
77 | ||
1da177e4 LT |
78 | /* |
79 | * xfs_getattr | |
80 | */ | |
993386c1 | 81 | int |
1da177e4 | 82 | xfs_getattr( |
993386c1 | 83 | xfs_inode_t *ip, |
8285fb58 | 84 | bhv_vattr_t *vap, |
993386c1 | 85 | int flags) |
1da177e4 | 86 | { |
993386c1 CH |
87 | bhv_vnode_t *vp = XFS_ITOV(ip); |
88 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 | 89 | |
cf441eeb | 90 | xfs_itrace_entry(ip); |
1da177e4 | 91 | |
1da177e4 LT |
92 | if (XFS_FORCED_SHUTDOWN(mp)) |
93 | return XFS_ERROR(EIO); | |
94 | ||
95 | if (!(flags & ATTR_LAZY)) | |
96 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
97 | ||
ba87ea69 | 98 | vap->va_size = XFS_ISIZE(ip); |
1da177e4 LT |
99 | if (vap->va_mask == XFS_AT_SIZE) |
100 | goto all_done; | |
101 | ||
102 | vap->va_nblocks = | |
103 | XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); | |
104 | vap->va_nodeid = ip->i_ino; | |
105 | #if XFS_BIG_INUMS | |
106 | vap->va_nodeid += mp->m_inoadd; | |
107 | #endif | |
108 | vap->va_nlink = ip->i_d.di_nlink; | |
109 | ||
110 | /* | |
111 | * Quick exit for non-stat callers | |
112 | */ | |
113 | if ((vap->va_mask & | |
114 | ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID| | |
115 | XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0) | |
116 | goto all_done; | |
117 | ||
118 | /* | |
119 | * Copy from in-core inode. | |
120 | */ | |
0432dab2 | 121 | vap->va_mode = ip->i_d.di_mode; |
1da177e4 LT |
122 | vap->va_uid = ip->i_d.di_uid; |
123 | vap->va_gid = ip->i_d.di_gid; | |
124 | vap->va_projid = ip->i_d.di_projid; | |
125 | ||
126 | /* | |
127 | * Check vnode type block/char vs. everything else. | |
1da177e4 | 128 | */ |
0432dab2 CH |
129 | switch (ip->i_d.di_mode & S_IFMT) { |
130 | case S_IFBLK: | |
131 | case S_IFCHR: | |
132 | vap->va_rdev = ip->i_df.if_u2.if_rdev; | |
133 | vap->va_blocksize = BLKDEV_IOSIZE; | |
134 | break; | |
135 | default: | |
1da177e4 LT |
136 | vap->va_rdev = 0; |
137 | ||
71ddabb9 | 138 | if (!(XFS_IS_REALTIME_INODE(ip))) { |
e8c8b3a7 | 139 | vap->va_blocksize = xfs_preferred_iosize(mp); |
1da177e4 LT |
140 | } else { |
141 | ||
142 | /* | |
143 | * If the file blocks are being allocated from a | |
144 | * realtime partition, then return the inode's | |
145 | * realtime extent size or the realtime volume's | |
146 | * extent size. | |
147 | */ | |
957d0ebe DC |
148 | vap->va_blocksize = |
149 | xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog; | |
1da177e4 | 150 | } |
0432dab2 | 151 | break; |
1da177e4 LT |
152 | } |
153 | ||
ca5ccbf9 | 154 | vn_atime_to_timespec(vp, &vap->va_atime); |
1da177e4 LT |
155 | vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec; |
156 | vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec; | |
157 | vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec; | |
158 | vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec; | |
159 | ||
160 | /* | |
161 | * Exit for stat callers. See if any of the rest of the fields | |
162 | * to be filled in are needed. | |
163 | */ | |
164 | if ((vap->va_mask & | |
165 | (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS| | |
166 | XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0) | |
167 | goto all_done; | |
168 | ||
169 | /* | |
170 | * Convert di_flags to xflags. | |
171 | */ | |
172 | vap->va_xflags = xfs_ip2xflags(ip); | |
173 | ||
174 | /* | |
175 | * Exit for inode revalidate. See if any of the rest of | |
176 | * the fields to be filled in are needed. | |
177 | */ | |
178 | if ((vap->va_mask & | |
179 | (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS| | |
180 | XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0) | |
181 | goto all_done; | |
182 | ||
183 | vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog; | |
184 | vap->va_nextents = | |
185 | (ip->i_df.if_flags & XFS_IFEXTENTS) ? | |
186 | ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) : | |
187 | ip->i_d.di_nextents; | |
188 | if (ip->i_afp) | |
189 | vap->va_anextents = | |
190 | (ip->i_afp->if_flags & XFS_IFEXTENTS) ? | |
191 | ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) : | |
192 | ip->i_d.di_anextents; | |
193 | else | |
194 | vap->va_anextents = 0; | |
195 | vap->va_gen = ip->i_d.di_gen; | |
196 | ||
197 | all_done: | |
198 | if (!(flags & ATTR_LAZY)) | |
199 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
200 | return 0; | |
201 | } | |
202 | ||
203 | ||
204 | /* | |
205 | * xfs_setattr | |
206 | */ | |
207 | int | |
208 | xfs_setattr( | |
993386c1 | 209 | xfs_inode_t *ip, |
8285fb58 | 210 | bhv_vattr_t *vap, |
1da177e4 LT |
211 | int flags, |
212 | cred_t *credp) | |
213 | { | |
993386c1 CH |
214 | bhv_vnode_t *vp = XFS_ITOV(ip); |
215 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 | 216 | xfs_trans_t *tp; |
1da177e4 LT |
217 | int mask; |
218 | int code; | |
219 | uint lock_flags; | |
220 | uint commit_flags=0; | |
221 | uid_t uid=0, iuid=0; | |
222 | gid_t gid=0, igid=0; | |
223 | int timeflags = 0; | |
1da177e4 LT |
224 | xfs_prid_t projid=0, iprojid=0; |
225 | int mandlock_before, mandlock_after; | |
226 | struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2; | |
227 | int file_owner; | |
5fcbab35 | 228 | int need_iolock = 1; |
1da177e4 | 229 | |
cf441eeb | 230 | xfs_itrace_entry(ip); |
1da177e4 | 231 | |
bd186aa9 | 232 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
233 | return XFS_ERROR(EROFS); |
234 | ||
235 | /* | |
236 | * Cannot set certain attributes. | |
237 | */ | |
238 | mask = vap->va_mask; | |
239 | if (mask & XFS_AT_NOSET) { | |
240 | return XFS_ERROR(EINVAL); | |
241 | } | |
242 | ||
1da177e4 LT |
243 | if (XFS_FORCED_SHUTDOWN(mp)) |
244 | return XFS_ERROR(EIO); | |
245 | ||
246 | /* | |
247 | * Timestamps do not need to be logged and hence do not | |
248 | * need to be done within a transaction. | |
249 | */ | |
250 | if (mask & XFS_AT_UPDTIMES) { | |
251 | ASSERT((mask & ~XFS_AT_UPDTIMES) == 0); | |
252 | timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) | | |
253 | ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) | | |
254 | ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0); | |
255 | xfs_ichgtime(ip, timeflags); | |
256 | return 0; | |
257 | } | |
258 | ||
259 | olddquot1 = olddquot2 = NULL; | |
260 | udqp = gdqp = NULL; | |
261 | ||
262 | /* | |
263 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
264 | * before we start any other transactions. Trying to do this later | |
265 | * is messy. We don't care to take a readlock to look at the ids | |
266 | * in inode here, because we can't hold it across the trans_reserve. | |
267 | * If the IDs do change before we take the ilock, we're covered | |
268 | * because the i_*dquot fields will get updated anyway. | |
269 | */ | |
c8ad20ff NS |
270 | if (XFS_IS_QUOTA_ON(mp) && |
271 | (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) { | |
1da177e4 LT |
272 | uint qflags = 0; |
273 | ||
c8ad20ff | 274 | if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) { |
1da177e4 LT |
275 | uid = vap->va_uid; |
276 | qflags |= XFS_QMOPT_UQUOTA; | |
277 | } else { | |
278 | uid = ip->i_d.di_uid; | |
279 | } | |
c8ad20ff | 280 | if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) { |
1da177e4 LT |
281 | gid = vap->va_gid; |
282 | qflags |= XFS_QMOPT_GQUOTA; | |
283 | } else { | |
284 | gid = ip->i_d.di_gid; | |
285 | } | |
c8ad20ff NS |
286 | if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) { |
287 | projid = vap->va_projid; | |
288 | qflags |= XFS_QMOPT_PQUOTA; | |
289 | } else { | |
290 | projid = ip->i_d.di_projid; | |
291 | } | |
1da177e4 LT |
292 | /* |
293 | * We take a reference when we initialize udqp and gdqp, | |
294 | * so it is important that we never blindly double trip on | |
295 | * the same variable. See xfs_create() for an example. | |
296 | */ | |
297 | ASSERT(udqp == NULL); | |
298 | ASSERT(gdqp == NULL); | |
c8ad20ff NS |
299 | code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags, |
300 | &udqp, &gdqp); | |
1da177e4 | 301 | if (code) |
014c2544 | 302 | return code; |
1da177e4 LT |
303 | } |
304 | ||
305 | /* | |
306 | * For the other attributes, we acquire the inode lock and | |
307 | * first do an error checking pass. | |
308 | */ | |
309 | tp = NULL; | |
310 | lock_flags = XFS_ILOCK_EXCL; | |
5fcbab35 DR |
311 | if (flags & ATTR_NOLOCK) |
312 | need_iolock = 0; | |
1da177e4 LT |
313 | if (!(mask & XFS_AT_SIZE)) { |
314 | if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) || | |
315 | (mp->m_flags & XFS_MOUNT_WSYNC)) { | |
316 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); | |
317 | commit_flags = 0; | |
318 | if ((code = xfs_trans_reserve(tp, 0, | |
319 | XFS_ICHANGE_LOG_RES(mp), 0, | |
320 | 0, 0))) { | |
321 | lock_flags = 0; | |
322 | goto error_return; | |
323 | } | |
324 | } | |
325 | } else { | |
eb9df39d | 326 | if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) && |
1da177e4 LT |
327 | !(flags & ATTR_DMI)) { |
328 | int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR; | |
329 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp, | |
330 | vap->va_size, 0, dmflags, NULL); | |
331 | if (code) { | |
332 | lock_flags = 0; | |
333 | goto error_return; | |
334 | } | |
335 | } | |
336 | if (need_iolock) | |
337 | lock_flags |= XFS_IOLOCK_EXCL; | |
338 | } | |
339 | ||
340 | xfs_ilock(ip, lock_flags); | |
341 | ||
342 | /* boolean: are we the file owner? */ | |
343 | file_owner = (current_fsuid(credp) == ip->i_d.di_uid); | |
344 | ||
345 | /* | |
346 | * Change various properties of a file. | |
347 | * Only the owner or users with CAP_FOWNER | |
348 | * capability may do these things. | |
349 | */ | |
350 | if (mask & | |
351 | (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID| | |
352 | XFS_AT_GID|XFS_AT_PROJID)) { | |
353 | /* | |
354 | * CAP_FOWNER overrides the following restrictions: | |
355 | * | |
356 | * The user ID of the calling process must be equal | |
357 | * to the file owner ID, except in cases where the | |
358 | * CAP_FSETID capability is applicable. | |
359 | */ | |
360 | if (!file_owner && !capable(CAP_FOWNER)) { | |
361 | code = XFS_ERROR(EPERM); | |
362 | goto error_return; | |
363 | } | |
364 | ||
365 | /* | |
366 | * CAP_FSETID overrides the following restrictions: | |
367 | * | |
368 | * The effective user ID of the calling process shall match | |
369 | * the file owner when setting the set-user-ID and | |
370 | * set-group-ID bits on that file. | |
371 | * | |
372 | * The effective group ID or one of the supplementary group | |
373 | * IDs of the calling process shall match the group owner of | |
374 | * the file when setting the set-group-ID bit on that file | |
375 | */ | |
376 | if (mask & XFS_AT_MODE) { | |
377 | mode_t m = 0; | |
378 | ||
379 | if ((vap->va_mode & S_ISUID) && !file_owner) | |
380 | m |= S_ISUID; | |
381 | if ((vap->va_mode & S_ISGID) && | |
382 | !in_group_p((gid_t)ip->i_d.di_gid)) | |
383 | m |= S_ISGID; | |
384 | #if 0 | |
385 | /* Linux allows this, Irix doesn't. */ | |
0432dab2 | 386 | if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp)) |
1da177e4 LT |
387 | m |= S_ISVTX; |
388 | #endif | |
389 | if (m && !capable(CAP_FSETID)) | |
390 | vap->va_mode &= ~m; | |
391 | } | |
392 | } | |
393 | ||
394 | /* | |
395 | * Change file ownership. Must be the owner or privileged. | |
396 | * If the system was configured with the "restricted_chown" | |
397 | * option, the owner is not permitted to give away the file, | |
398 | * and can change the group id only to a group of which he | |
399 | * or she is a member. | |
400 | */ | |
401 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
402 | /* | |
403 | * These IDs could have changed since we last looked at them. | |
404 | * But, we're assured that if the ownership did change | |
405 | * while we didn't have the inode locked, inode's dquot(s) | |
406 | * would have changed also. | |
407 | */ | |
408 | iuid = ip->i_d.di_uid; | |
409 | iprojid = ip->i_d.di_projid; | |
410 | igid = ip->i_d.di_gid; | |
411 | gid = (mask & XFS_AT_GID) ? vap->va_gid : igid; | |
412 | uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid; | |
413 | projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid : | |
414 | iprojid; | |
415 | ||
416 | /* | |
417 | * CAP_CHOWN overrides the following restrictions: | |
418 | * | |
419 | * If _POSIX_CHOWN_RESTRICTED is defined, this capability | |
420 | * shall override the restriction that a process cannot | |
421 | * change the user ID of a file it owns and the restriction | |
422 | * that the group ID supplied to the chown() function | |
423 | * shall be equal to either the group ID or one of the | |
424 | * supplementary group IDs of the calling process. | |
1da177e4 LT |
425 | */ |
426 | if (restricted_chown && | |
427 | (iuid != uid || (igid != gid && | |
428 | !in_group_p((gid_t)gid))) && | |
429 | !capable(CAP_CHOWN)) { | |
430 | code = XFS_ERROR(EPERM); | |
431 | goto error_return; | |
432 | } | |
433 | /* | |
c8ad20ff | 434 | * Do a quota reservation only if uid/projid/gid is actually |
1da177e4 LT |
435 | * going to change. |
436 | */ | |
437 | if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || | |
c8ad20ff | 438 | (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) || |
1da177e4 LT |
439 | (XFS_IS_GQUOTA_ON(mp) && igid != gid)) { |
440 | ASSERT(tp); | |
441 | code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp, | |
442 | capable(CAP_FOWNER) ? | |
443 | XFS_QMOPT_FORCE_RES : 0); | |
444 | if (code) /* out of quota */ | |
445 | goto error_return; | |
446 | } | |
447 | } | |
448 | ||
449 | /* | |
450 | * Truncate file. Must have write permission and not be a directory. | |
451 | */ | |
452 | if (mask & XFS_AT_SIZE) { | |
453 | /* Short circuit the truncate case for zero length files */ | |
454 | if ((vap->va_size == 0) && | |
ba87ea69 | 455 | (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) { |
1da177e4 LT |
456 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
457 | lock_flags &= ~XFS_ILOCK_EXCL; | |
458 | if (mask & XFS_AT_CTIME) | |
459 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
460 | code = 0; | |
461 | goto error_return; | |
462 | } | |
463 | ||
0432dab2 | 464 | if (VN_ISDIR(vp)) { |
1da177e4 LT |
465 | code = XFS_ERROR(EISDIR); |
466 | goto error_return; | |
0432dab2 | 467 | } else if (!VN_ISREG(vp)) { |
1da177e4 LT |
468 | code = XFS_ERROR(EINVAL); |
469 | goto error_return; | |
470 | } | |
471 | /* | |
472 | * Make sure that the dquots are attached to the inode. | |
473 | */ | |
474 | if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED))) | |
475 | goto error_return; | |
476 | } | |
477 | ||
478 | /* | |
479 | * Change file access or modified times. | |
480 | */ | |
481 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
482 | if (!file_owner) { | |
483 | if ((flags & ATTR_UTIME) && | |
484 | !capable(CAP_FOWNER)) { | |
485 | code = XFS_ERROR(EPERM); | |
486 | goto error_return; | |
487 | } | |
488 | } | |
489 | } | |
490 | ||
491 | /* | |
492 | * Change extent size or realtime flag. | |
493 | */ | |
494 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
495 | /* | |
496 | * Can't change extent size if any extents are allocated. | |
497 | */ | |
e94af02a | 498 | if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) && |
1da177e4 LT |
499 | ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != |
500 | vap->va_extsize) ) { | |
501 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
502 | goto error_return; | |
503 | } | |
504 | ||
1da177e4 LT |
505 | /* |
506 | * Can't change realtime flag if any extents are allocated. | |
507 | */ | |
e94af02a ES |
508 | if ((ip->i_d.di_nextents || ip->i_delayed_blks) && |
509 | (mask & XFS_AT_XFLAGS) && | |
71ddabb9 | 510 | (XFS_IS_REALTIME_INODE(ip)) != |
1da177e4 LT |
511 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { |
512 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
513 | goto error_return; | |
514 | } | |
515 | /* | |
516 | * Extent size must be a multiple of the appropriate block | |
517 | * size, if set at all. | |
518 | */ | |
519 | if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) { | |
520 | xfs_extlen_t size; | |
521 | ||
71ddabb9 | 522 | if (XFS_IS_REALTIME_INODE(ip) || |
1da177e4 LT |
523 | ((mask & XFS_AT_XFLAGS) && |
524 | (vap->va_xflags & XFS_XFLAG_REALTIME))) { | |
525 | size = mp->m_sb.sb_rextsize << | |
526 | mp->m_sb.sb_blocklog; | |
527 | } else { | |
528 | size = mp->m_sb.sb_blocksize; | |
529 | } | |
530 | if (vap->va_extsize % size) { | |
531 | code = XFS_ERROR(EINVAL); | |
532 | goto error_return; | |
533 | } | |
534 | } | |
535 | /* | |
536 | * If realtime flag is set then must have realtime data. | |
537 | */ | |
538 | if ((mask & XFS_AT_XFLAGS) && | |
539 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { | |
540 | if ((mp->m_sb.sb_rblocks == 0) || | |
541 | (mp->m_sb.sb_rextsize == 0) || | |
542 | (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) { | |
543 | code = XFS_ERROR(EINVAL); | |
544 | goto error_return; | |
545 | } | |
546 | } | |
547 | ||
548 | /* | |
549 | * Can't modify an immutable/append-only file unless | |
550 | * we have appropriate permission. | |
551 | */ | |
552 | if ((mask & XFS_AT_XFLAGS) && | |
553 | (ip->i_d.di_flags & | |
554 | (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) || | |
555 | (vap->va_xflags & | |
556 | (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) && | |
557 | !capable(CAP_LINUX_IMMUTABLE)) { | |
558 | code = XFS_ERROR(EPERM); | |
559 | goto error_return; | |
560 | } | |
561 | } | |
562 | ||
563 | /* | |
564 | * Now we can make the changes. Before we join the inode | |
565 | * to the transaction, if XFS_AT_SIZE is set then take care of | |
566 | * the part of the truncation that must be done without the | |
567 | * inode lock. This needs to be done before joining the inode | |
568 | * to the transaction, because the inode cannot be unlocked | |
569 | * once it is a part of the transaction. | |
570 | */ | |
571 | if (mask & XFS_AT_SIZE) { | |
572 | code = 0; | |
ba87ea69 | 573 | if ((vap->va_size > ip->i_size) && |
374e2ac3 | 574 | (flags & ATTR_NOSIZETOK) == 0) { |
1da177e4 | 575 | code = xfs_igrow_start(ip, vap->va_size, credp); |
374e2ac3 | 576 | } |
1da177e4 | 577 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
c32676ee DC |
578 | |
579 | /* | |
580 | * We are going to log the inode size change in this | |
581 | * transaction so any previous writes that are beyond the on | |
582 | * disk EOF and the new EOF that have not been written out need | |
583 | * to be written here. If we do not write the data out, we | |
584 | * expose ourselves to the null files problem. | |
585 | * | |
586 | * Only flush from the on disk size to the smaller of the in | |
587 | * memory file size or the new size as that's the range we | |
588 | * really care about here and prevents waiting for other data | |
589 | * not within the range we care about here. | |
590 | */ | |
591 | if (!code && | |
592 | (ip->i_size != ip->i_d.di_size) && | |
593 | (vap->va_size > ip->i_d.di_size)) { | |
739bfb2a | 594 | code = xfs_flush_pages(ip, |
c32676ee DC |
595 | ip->i_d.di_size, vap->va_size, |
596 | XFS_B_ASYNC, FI_NONE); | |
597 | } | |
598 | ||
599 | /* wait for all I/O to complete */ | |
b677c210 | 600 | vn_iowait(ip); |
c32676ee | 601 | |
1da177e4 LT |
602 | if (!code) |
603 | code = xfs_itruncate_data(ip, vap->va_size); | |
604 | if (code) { | |
605 | ASSERT(tp == NULL); | |
606 | lock_flags &= ~XFS_ILOCK_EXCL; | |
607 | ASSERT(lock_flags == XFS_IOLOCK_EXCL); | |
608 | goto error_return; | |
609 | } | |
610 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); | |
611 | if ((code = xfs_trans_reserve(tp, 0, | |
612 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
613 | XFS_TRANS_PERM_LOG_RES, | |
614 | XFS_ITRUNCATE_LOG_COUNT))) { | |
615 | xfs_trans_cancel(tp, 0); | |
616 | if (need_iolock) | |
617 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
618 | return code; | |
619 | } | |
620 | commit_flags = XFS_TRANS_RELEASE_LOG_RES; | |
621 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
622 | } | |
623 | ||
624 | if (tp) { | |
625 | xfs_trans_ijoin(tp, ip, lock_flags); | |
626 | xfs_trans_ihold(tp, ip); | |
627 | } | |
628 | ||
629 | /* determine whether mandatory locking mode changes */ | |
630 | mandlock_before = MANDLOCK(vp, ip->i_d.di_mode); | |
631 | ||
632 | /* | |
633 | * Truncate file. Must have write permission and not be a directory. | |
634 | */ | |
635 | if (mask & XFS_AT_SIZE) { | |
ba87ea69 | 636 | if (vap->va_size > ip->i_size) { |
1da177e4 LT |
637 | xfs_igrow_finish(tp, ip, vap->va_size, |
638 | !(flags & ATTR_DMI)); | |
ba87ea69 | 639 | } else if ((vap->va_size <= ip->i_size) || |
1da177e4 LT |
640 | ((vap->va_size == 0) && ip->i_d.di_nextents)) { |
641 | /* | |
642 | * signal a sync transaction unless | |
643 | * we're truncating an already unlinked | |
644 | * file on a wsync filesystem | |
645 | */ | |
646 | code = xfs_itruncate_finish(&tp, ip, | |
647 | (xfs_fsize_t)vap->va_size, | |
648 | XFS_DATA_FORK, | |
649 | ((ip->i_d.di_nlink != 0 || | |
650 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
651 | ? 1 : 0)); | |
7d4fb40a | 652 | if (code) |
1da177e4 | 653 | goto abort_return; |
7d4fb40a NS |
654 | /* |
655 | * Truncated "down", so we're removing references | |
656 | * to old data here - if we now delay flushing for | |
657 | * a long time, we expose ourselves unduly to the | |
658 | * notorious NULL files problem. So, we mark this | |
659 | * vnode and flush it when the file is closed, and | |
660 | * do not wait the usual (long) time for writeout. | |
661 | */ | |
b3aea4ed | 662 | xfs_iflags_set(ip, XFS_ITRUNCATED); |
1da177e4 LT |
663 | } |
664 | /* | |
665 | * Have to do this even if the file's size doesn't change. | |
666 | */ | |
667 | timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
668 | } | |
669 | ||
670 | /* | |
671 | * Change file access modes. | |
672 | */ | |
673 | if (mask & XFS_AT_MODE) { | |
674 | ip->i_d.di_mode &= S_IFMT; | |
675 | ip->i_d.di_mode |= vap->va_mode & ~S_IFMT; | |
676 | ||
677 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
678 | timeflags |= XFS_ICHGTIME_CHG; | |
679 | } | |
680 | ||
681 | /* | |
682 | * Change file ownership. Must be the owner or privileged. | |
683 | * If the system was configured with the "restricted_chown" | |
684 | * option, the owner is not permitted to give away the file, | |
685 | * and can change the group id only to a group of which he | |
686 | * or she is a member. | |
687 | */ | |
688 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
689 | /* | |
690 | * CAP_FSETID overrides the following restrictions: | |
691 | * | |
692 | * The set-user-ID and set-group-ID bits of a file will be | |
693 | * cleared upon successful return from chown() | |
694 | */ | |
695 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
696 | !capable(CAP_FSETID)) { | |
697 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
698 | } | |
699 | ||
700 | /* | |
701 | * Change the ownerships and register quota modifications | |
702 | * in the transaction. | |
703 | */ | |
704 | if (iuid != uid) { | |
705 | if (XFS_IS_UQUOTA_ON(mp)) { | |
706 | ASSERT(mask & XFS_AT_UID); | |
707 | ASSERT(udqp); | |
708 | olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
709 | &ip->i_udquot, udqp); | |
710 | } | |
711 | ip->i_d.di_uid = uid; | |
712 | } | |
713 | if (igid != gid) { | |
714 | if (XFS_IS_GQUOTA_ON(mp)) { | |
c8ad20ff | 715 | ASSERT(!XFS_IS_PQUOTA_ON(mp)); |
1da177e4 LT |
716 | ASSERT(mask & XFS_AT_GID); |
717 | ASSERT(gdqp); | |
718 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
719 | &ip->i_gdquot, gdqp); | |
720 | } | |
721 | ip->i_d.di_gid = gid; | |
722 | } | |
723 | if (iprojid != projid) { | |
c8ad20ff NS |
724 | if (XFS_IS_PQUOTA_ON(mp)) { |
725 | ASSERT(!XFS_IS_GQUOTA_ON(mp)); | |
726 | ASSERT(mask & XFS_AT_PROJID); | |
727 | ASSERT(gdqp); | |
728 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
729 | &ip->i_gdquot, gdqp); | |
730 | } | |
1da177e4 LT |
731 | ip->i_d.di_projid = projid; |
732 | /* | |
733 | * We may have to rev the inode as well as | |
734 | * the superblock version number since projids didn't | |
735 | * exist before DINODE_VERSION_2 and SB_VERSION_NLINK. | |
736 | */ | |
737 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) | |
738 | xfs_bump_ino_vers2(tp, ip); | |
739 | } | |
740 | ||
741 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
742 | timeflags |= XFS_ICHGTIME_CHG; | |
743 | } | |
744 | ||
745 | ||
746 | /* | |
747 | * Change file access or modified times. | |
748 | */ | |
749 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
750 | if (mask & XFS_AT_ATIME) { | |
751 | ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec; | |
752 | ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec; | |
753 | ip->i_update_core = 1; | |
754 | timeflags &= ~XFS_ICHGTIME_ACC; | |
755 | } | |
756 | if (mask & XFS_AT_MTIME) { | |
757 | ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec; | |
758 | ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec; | |
759 | timeflags &= ~XFS_ICHGTIME_MOD; | |
760 | timeflags |= XFS_ICHGTIME_CHG; | |
761 | } | |
762 | if (tp && (flags & ATTR_UTIME)) | |
763 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
764 | } | |
765 | ||
766 | /* | |
767 | * Change XFS-added attributes. | |
768 | */ | |
769 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
770 | if (mask & XFS_AT_EXTSIZE) { | |
771 | /* | |
772 | * Converting bytes to fs blocks. | |
773 | */ | |
774 | ip->i_d.di_extsize = vap->va_extsize >> | |
775 | mp->m_sb.sb_blocklog; | |
776 | } | |
777 | if (mask & XFS_AT_XFLAGS) { | |
778 | uint di_flags; | |
779 | ||
780 | /* can't set PREALLOC this way, just preserve it */ | |
781 | di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC); | |
782 | if (vap->va_xflags & XFS_XFLAG_IMMUTABLE) | |
783 | di_flags |= XFS_DIFLAG_IMMUTABLE; | |
784 | if (vap->va_xflags & XFS_XFLAG_APPEND) | |
785 | di_flags |= XFS_DIFLAG_APPEND; | |
786 | if (vap->va_xflags & XFS_XFLAG_SYNC) | |
787 | di_flags |= XFS_DIFLAG_SYNC; | |
788 | if (vap->va_xflags & XFS_XFLAG_NOATIME) | |
789 | di_flags |= XFS_DIFLAG_NOATIME; | |
790 | if (vap->va_xflags & XFS_XFLAG_NODUMP) | |
791 | di_flags |= XFS_DIFLAG_NODUMP; | |
365ca83d NS |
792 | if (vap->va_xflags & XFS_XFLAG_PROJINHERIT) |
793 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
794 | if (vap->va_xflags & XFS_XFLAG_NODEFRAG) |
795 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
796 | if (vap->va_xflags & XFS_XFLAG_FILESTREAM) |
797 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
1da177e4 LT |
798 | if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { |
799 | if (vap->va_xflags & XFS_XFLAG_RTINHERIT) | |
800 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
801 | if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS) | |
802 | di_flags |= XFS_DIFLAG_NOSYMLINKS; | |
dd9f438e NS |
803 | if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT) |
804 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
805 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
613d7043 | 806 | if (vap->va_xflags & XFS_XFLAG_REALTIME) |
1da177e4 | 807 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
808 | if (vap->va_xflags & XFS_XFLAG_EXTSIZE) |
809 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1da177e4 LT |
810 | } |
811 | ip->i_d.di_flags = di_flags; | |
812 | } | |
813 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
814 | timeflags |= XFS_ICHGTIME_CHG; | |
815 | } | |
816 | ||
817 | /* | |
818 | * Change file inode change time only if XFS_AT_CTIME set | |
819 | * AND we have been called by a DMI function. | |
820 | */ | |
821 | ||
822 | if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) { | |
823 | ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec; | |
824 | ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec; | |
825 | ip->i_update_core = 1; | |
826 | timeflags &= ~XFS_ICHGTIME_CHG; | |
827 | } | |
828 | ||
829 | /* | |
830 | * Send out timestamp changes that need to be set to the | |
831 | * current time. Not done when called by a DMI function. | |
832 | */ | |
833 | if (timeflags && !(flags & ATTR_DMI)) | |
834 | xfs_ichgtime(ip, timeflags); | |
835 | ||
836 | XFS_STATS_INC(xs_ig_attrchg); | |
837 | ||
838 | /* | |
839 | * If this is a synchronous mount, make sure that the | |
840 | * transaction goes to disk before returning to the user. | |
841 | * This is slightly sub-optimal in that truncates require | |
c41564b5 | 842 | * two sync transactions instead of one for wsync filesystems. |
1da177e4 LT |
843 | * One for the truncate and one for the timestamps since we |
844 | * don't want to change the timestamps unless we're sure the | |
845 | * truncate worked. Truncates are less than 1% of the laddis | |
846 | * mix so this probably isn't worth the trouble to optimize. | |
847 | */ | |
848 | code = 0; | |
849 | if (tp) { | |
850 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
851 | xfs_trans_set_sync(tp); | |
852 | ||
1c72bf90 | 853 | code = xfs_trans_commit(tp, commit_flags); |
1da177e4 LT |
854 | } |
855 | ||
856 | /* | |
857 | * If the (regular) file's mandatory locking mode changed, then | |
858 | * notify the vnode. We do this under the inode lock to prevent | |
859 | * racing calls to vop_vnode_change. | |
860 | */ | |
861 | mandlock_after = MANDLOCK(vp, ip->i_d.di_mode); | |
1da177e4 LT |
862 | |
863 | xfs_iunlock(ip, lock_flags); | |
864 | ||
865 | /* | |
866 | * Release any dquot(s) the inode had kept before chown. | |
867 | */ | |
868 | XFS_QM_DQRELE(mp, olddquot1); | |
869 | XFS_QM_DQRELE(mp, olddquot2); | |
870 | XFS_QM_DQRELE(mp, udqp); | |
871 | XFS_QM_DQRELE(mp, gdqp); | |
872 | ||
873 | if (code) { | |
874 | return code; | |
875 | } | |
876 | ||
eb9df39d | 877 | if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) && |
1da177e4 LT |
878 | !(flags & ATTR_DMI)) { |
879 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL, | |
880 | NULL, DM_RIGHT_NULL, NULL, NULL, | |
881 | 0, 0, AT_DELAY_FLAG(flags)); | |
882 | } | |
883 | return 0; | |
884 | ||
885 | abort_return: | |
886 | commit_flags |= XFS_TRANS_ABORT; | |
887 | /* FALLTHROUGH */ | |
888 | error_return: | |
889 | XFS_QM_DQRELE(mp, udqp); | |
890 | XFS_QM_DQRELE(mp, gdqp); | |
891 | if (tp) { | |
892 | xfs_trans_cancel(tp, commit_flags); | |
893 | } | |
894 | if (lock_flags != 0) { | |
895 | xfs_iunlock(ip, lock_flags); | |
896 | } | |
897 | return code; | |
898 | } | |
899 | ||
7d4fb40a NS |
900 | /* |
901 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
902 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
903 | * bmapi. | |
904 | */ | |
905 | #define SYMLINK_MAPS 2 | |
906 | ||
804c83c3 CH |
907 | STATIC int |
908 | xfs_readlink_bmap( | |
909 | xfs_inode_t *ip, | |
910 | char *link) | |
911 | { | |
912 | xfs_mount_t *mp = ip->i_mount; | |
913 | int pathlen = ip->i_d.di_size; | |
914 | int nmaps = SYMLINK_MAPS; | |
915 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
916 | xfs_daddr_t d; | |
917 | int byte_cnt; | |
918 | int n; | |
919 | xfs_buf_t *bp; | |
920 | int error = 0; | |
921 | ||
922 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0, | |
923 | mval, &nmaps, NULL, NULL); | |
924 | if (error) | |
925 | goto out; | |
926 | ||
927 | for (n = 0; n < nmaps; n++) { | |
928 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
929 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
930 | ||
931 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); | |
932 | error = XFS_BUF_GETERROR(bp); | |
933 | if (error) { | |
934 | xfs_ioerror_alert("xfs_readlink", | |
935 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
936 | xfs_buf_relse(bp); | |
937 | goto out; | |
938 | } | |
939 | if (pathlen < byte_cnt) | |
940 | byte_cnt = pathlen; | |
941 | pathlen -= byte_cnt; | |
942 | ||
943 | memcpy(link, XFS_BUF_PTR(bp), byte_cnt); | |
944 | xfs_buf_relse(bp); | |
945 | } | |
946 | ||
947 | link[ip->i_d.di_size] = '\0'; | |
948 | error = 0; | |
949 | ||
950 | out: | |
951 | return error; | |
952 | } | |
953 | ||
993386c1 | 954 | int |
1da177e4 | 955 | xfs_readlink( |
993386c1 | 956 | xfs_inode_t *ip, |
804c83c3 | 957 | char *link) |
1da177e4 | 958 | { |
804c83c3 | 959 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 960 | int pathlen; |
1da177e4 | 961 | int error = 0; |
1da177e4 | 962 | |
cf441eeb | 963 | xfs_itrace_entry(ip); |
1da177e4 LT |
964 | |
965 | if (XFS_FORCED_SHUTDOWN(mp)) | |
966 | return XFS_ERROR(EIO); | |
967 | ||
968 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
969 | ||
970 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
804c83c3 | 971 | ASSERT(ip->i_d.di_size <= MAXPATHLEN); |
1da177e4 | 972 | |
804c83c3 CH |
973 | pathlen = ip->i_d.di_size; |
974 | if (!pathlen) | |
975 | goto out; | |
1da177e4 LT |
976 | |
977 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
804c83c3 CH |
978 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
979 | link[pathlen] = '\0'; | |
980 | } else { | |
981 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
982 | } |
983 | ||
804c83c3 | 984 | out: |
1da177e4 | 985 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
986 | return error; |
987 | } | |
988 | ||
1da177e4 LT |
989 | /* |
990 | * xfs_fsync | |
991 | * | |
992 | * This is called to sync the inode and its data out to disk. | |
993 | * We need to hold the I/O lock while flushing the data, and | |
994 | * the inode lock while flushing the inode. The inode lock CANNOT | |
995 | * be held while flushing the data, so acquire after we're done | |
996 | * with that. | |
997 | */ | |
993386c1 | 998 | int |
1da177e4 | 999 | xfs_fsync( |
993386c1 | 1000 | xfs_inode_t *ip, |
1da177e4 | 1001 | int flag, |
1da177e4 LT |
1002 | xfs_off_t start, |
1003 | xfs_off_t stop) | |
1004 | { | |
1da177e4 LT |
1005 | xfs_trans_t *tp; |
1006 | int error; | |
f538d4da | 1007 | int log_flushed = 0, changed = 1; |
1da177e4 | 1008 | |
cf441eeb | 1009 | xfs_itrace_entry(ip); |
1da177e4 LT |
1010 | |
1011 | ASSERT(start >= 0 && stop >= -1); | |
1012 | ||
1013 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
1014 | return XFS_ERROR(EIO); | |
1015 | ||
776a75fa LM |
1016 | if (flag & FSYNC_DATA) |
1017 | filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping); | |
1018 | ||
1da177e4 LT |
1019 | /* |
1020 | * We always need to make sure that the required inode state | |
1021 | * is safe on disk. The vnode might be clean but because | |
1022 | * of committed transactions that haven't hit the disk yet. | |
1023 | * Likewise, there could be unflushed non-transactional | |
1024 | * changes to the inode core that have to go to disk. | |
1025 | * | |
1026 | * The following code depends on one assumption: that | |
1027 | * any transaction that changes an inode logs the core | |
1028 | * because it has to change some field in the inode core | |
1029 | * (typically nextents or nblocks). That assumption | |
1030 | * implies that any transactions against an inode will | |
1031 | * catch any non-transactional updates. If inode-altering | |
1032 | * transactions exist that violate this assumption, the | |
1033 | * code breaks. Right now, it figures that if the involved | |
1034 | * update_* field is clear and the inode is unpinned, the | |
1035 | * inode is clean. Either it's been flushed or it's been | |
1036 | * committed and the commit has hit the disk unpinning the inode. | |
1037 | * (Note that xfs_inode_item_format() called at commit clears | |
1038 | * the update_* fields.) | |
1039 | */ | |
1040 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1041 | ||
1042 | /* If we are flushing data then we care about update_size | |
1043 | * being set, otherwise we care about update_core | |
1044 | */ | |
1045 | if ((flag & FSYNC_DATA) ? | |
1046 | (ip->i_update_size == 0) : | |
1047 | (ip->i_update_core == 0)) { | |
1048 | /* | |
1049 | * Timestamps/size haven't changed since last inode | |
1050 | * flush or inode transaction commit. That means | |
1051 | * either nothing got written or a transaction | |
1052 | * committed which caught the updates. If the | |
1053 | * latter happened and the transaction hasn't | |
1054 | * hit the disk yet, the inode will be still | |
1055 | * be pinned. If it is, force the log. | |
1056 | */ | |
1057 | ||
1058 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1059 | ||
1060 | if (xfs_ipincount(ip)) { | |
f538d4da | 1061 | _xfs_log_force(ip->i_mount, (xfs_lsn_t)0, |
1da177e4 LT |
1062 | XFS_LOG_FORCE | |
1063 | ((flag & FSYNC_WAIT) | |
f538d4da CH |
1064 | ? XFS_LOG_SYNC : 0), |
1065 | &log_flushed); | |
1066 | } else { | |
1067 | /* | |
1068 | * If the inode is not pinned and nothing | |
1069 | * has changed we don't need to flush the | |
1070 | * cache. | |
1071 | */ | |
1072 | changed = 0; | |
1da177e4 LT |
1073 | } |
1074 | error = 0; | |
1075 | } else { | |
1076 | /* | |
1077 | * Kick off a transaction to log the inode | |
1078 | * core to get the updates. Make it | |
1079 | * sync if FSYNC_WAIT is passed in (which | |
1080 | * is done by everybody but specfs). The | |
1081 | * sync transaction will also force the log. | |
1082 | */ | |
1083 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1084 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
1085 | if ((error = xfs_trans_reserve(tp, 0, | |
1086 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), | |
1087 | 0, 0, 0))) { | |
1088 | xfs_trans_cancel(tp, 0); | |
1089 | return error; | |
1090 | } | |
1091 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1092 | ||
1093 | /* | |
1094 | * Note - it's possible that we might have pushed | |
1095 | * ourselves out of the way during trans_reserve | |
1096 | * which would flush the inode. But there's no | |
1097 | * guarantee that the inode buffer has actually | |
1098 | * gone out yet (it's delwri). Plus the buffer | |
1099 | * could be pinned anyway if it's part of an | |
1100 | * inode in another recent transaction. So we | |
1101 | * play it safe and fire off the transaction anyway. | |
1102 | */ | |
1103 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1104 | xfs_trans_ihold(tp, ip); | |
1105 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1106 | if (flag & FSYNC_WAIT) | |
1107 | xfs_trans_set_sync(tp); | |
1c72bf90 | 1108 | error = _xfs_trans_commit(tp, 0, &log_flushed); |
1da177e4 LT |
1109 | |
1110 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1111 | } | |
f538d4da CH |
1112 | |
1113 | if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) { | |
1114 | /* | |
1115 | * If the log write didn't issue an ordered tag we need | |
1116 | * to flush the disk cache for the data device now. | |
1117 | */ | |
1118 | if (!log_flushed) | |
1119 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
1120 | ||
1121 | /* | |
1122 | * If this inode is on the RT dev we need to flush that | |
c41564b5 | 1123 | * cache as well. |
f538d4da | 1124 | */ |
71ddabb9 | 1125 | if (XFS_IS_REALTIME_INODE(ip)) |
f538d4da CH |
1126 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); |
1127 | } | |
1128 | ||
1da177e4 LT |
1129 | return error; |
1130 | } | |
1131 | ||
1132 | /* | |
92dfe8d2 DC |
1133 | * This is called by xfs_inactive to free any blocks beyond eof |
1134 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
1135 | * punching a hole to EOF. | |
1da177e4 | 1136 | */ |
92dfe8d2 DC |
1137 | int |
1138 | xfs_free_eofblocks( | |
1da177e4 | 1139 | xfs_mount_t *mp, |
92dfe8d2 DC |
1140 | xfs_inode_t *ip, |
1141 | int flags) | |
1da177e4 LT |
1142 | { |
1143 | xfs_trans_t *tp; | |
1144 | int error; | |
1145 | xfs_fileoff_t end_fsb; | |
1146 | xfs_fileoff_t last_fsb; | |
1147 | xfs_filblks_t map_len; | |
1148 | int nimaps; | |
1149 | xfs_bmbt_irec_t imap; | |
92dfe8d2 | 1150 | int use_iolock = (flags & XFS_FREE_EOF_LOCK); |
1da177e4 LT |
1151 | |
1152 | /* | |
1153 | * Figure out if there are any blocks beyond the end | |
1154 | * of the file. If not, then there is nothing to do. | |
1155 | */ | |
ba87ea69 | 1156 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size)); |
1da177e4 LT |
1157 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); |
1158 | map_len = last_fsb - end_fsb; | |
1159 | if (map_len <= 0) | |
014c2544 | 1160 | return 0; |
1da177e4 LT |
1161 | |
1162 | nimaps = 1; | |
1163 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
541d7d3c | 1164 | error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0, |
3e57ecf6 | 1165 | NULL, 0, &imap, &nimaps, NULL, NULL); |
1da177e4 LT |
1166 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1167 | ||
1168 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
1169 | (imap.br_startblock != HOLESTARTBLOCK || |
1170 | ip->i_delayed_blks)) { | |
1da177e4 LT |
1171 | /* |
1172 | * Attach the dquots to the inode up front. | |
1173 | */ | |
1174 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1175 | return error; |
1da177e4 LT |
1176 | |
1177 | /* | |
1178 | * There are blocks after the end of file. | |
1179 | * Free them up now by truncating the file to | |
1180 | * its current size. | |
1181 | */ | |
1182 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1183 | ||
1184 | /* | |
1185 | * Do the xfs_itruncate_start() call before | |
1186 | * reserving any log space because | |
1187 | * itruncate_start will call into the buffer | |
1188 | * cache and we can't | |
1189 | * do that within a transaction. | |
1190 | */ | |
92dfe8d2 DC |
1191 | if (use_iolock) |
1192 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 | 1193 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, |
ba87ea69 | 1194 | ip->i_size); |
d3cf2094 | 1195 | if (error) { |
87ae3c24 | 1196 | xfs_trans_cancel(tp, 0); |
92dfe8d2 DC |
1197 | if (use_iolock) |
1198 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 LM |
1199 | return error; |
1200 | } | |
1da177e4 LT |
1201 | |
1202 | error = xfs_trans_reserve(tp, 0, | |
1203 | XFS_ITRUNCATE_LOG_RES(mp), | |
1204 | 0, XFS_TRANS_PERM_LOG_RES, | |
1205 | XFS_ITRUNCATE_LOG_COUNT); | |
1206 | if (error) { | |
1207 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1208 | xfs_trans_cancel(tp, 0); | |
1209 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1210 | return error; |
1da177e4 LT |
1211 | } |
1212 | ||
1213 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1214 | xfs_trans_ijoin(tp, ip, | |
1215 | XFS_IOLOCK_EXCL | | |
1216 | XFS_ILOCK_EXCL); | |
1217 | xfs_trans_ihold(tp, ip); | |
1218 | ||
1219 | error = xfs_itruncate_finish(&tp, ip, | |
ba87ea69 | 1220 | ip->i_size, |
1da177e4 LT |
1221 | XFS_DATA_FORK, |
1222 | 0); | |
1223 | /* | |
1224 | * If we get an error at this point we | |
1225 | * simply don't bother truncating the file. | |
1226 | */ | |
1227 | if (error) { | |
1228 | xfs_trans_cancel(tp, | |
1229 | (XFS_TRANS_RELEASE_LOG_RES | | |
1230 | XFS_TRANS_ABORT)); | |
1231 | } else { | |
1232 | error = xfs_trans_commit(tp, | |
1c72bf90 | 1233 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 1234 | } |
92dfe8d2 DC |
1235 | xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL) |
1236 | : XFS_ILOCK_EXCL)); | |
1da177e4 | 1237 | } |
014c2544 | 1238 | return error; |
1da177e4 LT |
1239 | } |
1240 | ||
1241 | /* | |
1242 | * Free a symlink that has blocks associated with it. | |
1243 | */ | |
1244 | STATIC int | |
1245 | xfs_inactive_symlink_rmt( | |
1246 | xfs_inode_t *ip, | |
1247 | xfs_trans_t **tpp) | |
1248 | { | |
1249 | xfs_buf_t *bp; | |
1250 | int committed; | |
1251 | int done; | |
1252 | int error; | |
1253 | xfs_fsblock_t first_block; | |
1254 | xfs_bmap_free_t free_list; | |
1255 | int i; | |
1256 | xfs_mount_t *mp; | |
1257 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1258 | int nmaps; | |
1259 | xfs_trans_t *ntp; | |
1260 | int size; | |
1261 | xfs_trans_t *tp; | |
1262 | ||
1263 | tp = *tpp; | |
1264 | mp = ip->i_mount; | |
1265 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
1266 | /* | |
1267 | * We're freeing a symlink that has some | |
1268 | * blocks allocated to it. Free the | |
1269 | * blocks here. We know that we've got | |
1270 | * either 1 or 2 extents and that we can | |
1271 | * free them all in one bunmapi call. | |
1272 | */ | |
1273 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
1274 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1275 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1276 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1277 | xfs_trans_cancel(tp, 0); | |
1278 | *tpp = NULL; | |
1279 | return error; | |
1280 | } | |
1281 | /* | |
1282 | * Lock the inode, fix the size, and join it to the transaction. | |
1283 | * Hold it so in the normal path, we still have it locked for | |
1284 | * the second transaction. In the error paths we need it | |
1285 | * held so the cancel won't rele it, see below. | |
1286 | */ | |
1287 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1288 | size = (int)ip->i_d.di_size; | |
1289 | ip->i_d.di_size = 0; | |
1290 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1291 | xfs_trans_ihold(tp, ip); | |
1292 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1293 | /* | |
1294 | * Find the block(s) so we can inval and unmap them. | |
1295 | */ | |
1296 | done = 0; | |
1297 | XFS_BMAP_INIT(&free_list, &first_block); | |
e8c96f8c | 1298 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
1299 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
1300 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 1301 | &free_list, NULL))) |
1da177e4 LT |
1302 | goto error0; |
1303 | /* | |
1304 | * Invalidate the block(s). | |
1305 | */ | |
1306 | for (i = 0; i < nmaps; i++) { | |
1307 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
1308 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
1309 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
1310 | xfs_trans_binval(tp, bp); | |
1311 | } | |
1312 | /* | |
1313 | * Unmap the dead block(s) to the free_list. | |
1314 | */ | |
1315 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 1316 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
1317 | goto error1; |
1318 | ASSERT(done); | |
1319 | /* | |
1320 | * Commit the first transaction. This logs the EFI and the inode. | |
1321 | */ | |
f7c99b6f | 1322 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
1323 | goto error1; |
1324 | /* | |
1325 | * The transaction must have been committed, since there were | |
1326 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
1327 | * The new tp has the extent freeing and EFDs. | |
1328 | */ | |
1329 | ASSERT(committed); | |
1330 | /* | |
1331 | * The first xact was committed, so add the inode to the new one. | |
1332 | * Mark it dirty so it will be logged and moved forward in the log as | |
1333 | * part of every commit. | |
1334 | */ | |
1335 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1336 | xfs_trans_ihold(tp, ip); | |
1337 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1338 | /* | |
1339 | * Get a new, empty transaction to return to our caller. | |
1340 | */ | |
1341 | ntp = xfs_trans_dup(tp); | |
1342 | /* | |
c41564b5 | 1343 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
1344 | * If we get an error on the commit here or on the reserve below, |
1345 | * we need to unlock the inode since the new transaction doesn't | |
1346 | * have the inode attached. | |
1347 | */ | |
1c72bf90 | 1348 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1349 | tp = ntp; |
1350 | if (error) { | |
1351 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1352 | goto error0; | |
1353 | } | |
1354 | /* | |
1355 | * Remove the memory for extent descriptions (just bookkeeping). | |
1356 | */ | |
1357 | if (ip->i_df.if_bytes) | |
1358 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
1359 | ASSERT(ip->i_df.if_bytes == 0); | |
1360 | /* | |
1361 | * Put an itruncate log reservation in the new transaction | |
1362 | * for our caller. | |
1363 | */ | |
1364 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1365 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1366 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1367 | goto error0; | |
1368 | } | |
1369 | /* | |
1370 | * Return with the inode locked but not joined to the transaction. | |
1371 | */ | |
1372 | *tpp = tp; | |
1373 | return 0; | |
1374 | ||
1375 | error1: | |
1376 | xfs_bmap_cancel(&free_list); | |
1377 | error0: | |
1378 | /* | |
1379 | * Have to come here with the inode locked and either | |
1380 | * (held and in the transaction) or (not in the transaction). | |
1381 | * If the inode isn't held then cancel would iput it, but | |
1382 | * that's wrong since this is inactive and the vnode ref | |
1383 | * count is 0 already. | |
1384 | * Cancel won't do anything to the inode if held, but it still | |
1385 | * needs to be locked until the cancel is done, if it was | |
1386 | * joined to the transaction. | |
1387 | */ | |
1388 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1389 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1390 | *tpp = NULL; | |
1391 | return error; | |
1392 | ||
1393 | } | |
1394 | ||
1395 | STATIC int | |
1396 | xfs_inactive_symlink_local( | |
1397 | xfs_inode_t *ip, | |
1398 | xfs_trans_t **tpp) | |
1399 | { | |
1400 | int error; | |
1401 | ||
1402 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
1403 | /* | |
1404 | * We're freeing a symlink which fit into | |
1405 | * the inode. Just free the memory used | |
1406 | * to hold the old symlink. | |
1407 | */ | |
1408 | error = xfs_trans_reserve(*tpp, 0, | |
1409 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
1410 | 0, XFS_TRANS_PERM_LOG_RES, | |
1411 | XFS_ITRUNCATE_LOG_COUNT); | |
1412 | ||
1413 | if (error) { | |
1414 | xfs_trans_cancel(*tpp, 0); | |
1415 | *tpp = NULL; | |
014c2544 | 1416 | return error; |
1da177e4 LT |
1417 | } |
1418 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1419 | ||
1420 | /* | |
1421 | * Zero length symlinks _can_ exist. | |
1422 | */ | |
1423 | if (ip->i_df.if_bytes > 0) { | |
1424 | xfs_idata_realloc(ip, | |
1425 | -(ip->i_df.if_bytes), | |
1426 | XFS_DATA_FORK); | |
1427 | ASSERT(ip->i_df.if_bytes == 0); | |
1428 | } | |
014c2544 | 1429 | return 0; |
1da177e4 LT |
1430 | } |
1431 | ||
1da177e4 LT |
1432 | STATIC int |
1433 | xfs_inactive_attrs( | |
1434 | xfs_inode_t *ip, | |
1435 | xfs_trans_t **tpp) | |
1436 | { | |
1437 | xfs_trans_t *tp; | |
1438 | int error; | |
1439 | xfs_mount_t *mp; | |
1440 | ||
1441 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); | |
1442 | tp = *tpp; | |
1443 | mp = ip->i_mount; | |
1444 | ASSERT(ip->i_d.di_forkoff != 0); | |
1c72bf90 | 1445 | xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1446 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1447 | ||
1448 | error = xfs_attr_inactive(ip); | |
1449 | if (error) { | |
1450 | *tpp = NULL; | |
1451 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1452 | return error; /* goto out */ |
1da177e4 LT |
1453 | } |
1454 | ||
1455 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1456 | error = xfs_trans_reserve(tp, 0, | |
1457 | XFS_IFREE_LOG_RES(mp), | |
1458 | 0, XFS_TRANS_PERM_LOG_RES, | |
1459 | XFS_INACTIVE_LOG_COUNT); | |
1460 | if (error) { | |
1461 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1462 | xfs_trans_cancel(tp, 0); | |
1463 | *tpp = NULL; | |
1464 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1465 | return error; |
1da177e4 LT |
1466 | } |
1467 | ||
1468 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1469 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1470 | xfs_trans_ihold(tp, ip); | |
1471 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1472 | ||
1473 | ASSERT(ip->i_d.di_anextents == 0); | |
1474 | ||
1475 | *tpp = tp; | |
014c2544 | 1476 | return 0; |
1da177e4 LT |
1477 | } |
1478 | ||
993386c1 | 1479 | int |
1da177e4 | 1480 | xfs_release( |
993386c1 | 1481 | xfs_inode_t *ip) |
1da177e4 | 1482 | { |
993386c1 CH |
1483 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1484 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
1485 | int error; |
1486 | ||
7d4fb40a | 1487 | if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) |
1da177e4 | 1488 | return 0; |
1da177e4 LT |
1489 | |
1490 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1491 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1492 | return 0; |
1493 | ||
2a82b8be | 1494 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
1495 | int truncated; |
1496 | ||
2a82b8be DC |
1497 | /* |
1498 | * If we are using filestreams, and we have an unlinked | |
1499 | * file that we are processing the last close on, then nothing | |
1500 | * will be able to reopen and write to this file. Purge this | |
1501 | * inode from the filestreams cache so that it doesn't delay | |
1502 | * teardown of the inode. | |
1503 | */ | |
1504 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1505 | xfs_filestream_deassociate(ip); | |
1506 | ||
fbf3ce8d CH |
1507 | /* |
1508 | * If we previously truncated this file and removed old data | |
1509 | * in the process, we want to initiate "early" writeout on | |
1510 | * the last close. This is an attempt to combat the notorious | |
1511 | * NULL files problem which is particularly noticable from a | |
1512 | * truncate down, buffered (re-)write (delalloc), followed by | |
1513 | * a crash. What we are effectively doing here is | |
1514 | * significantly reducing the time window where we'd otherwise | |
1515 | * be exposed to that problem. | |
1516 | */ | |
09262b43 | 1517 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
b3aea4ed | 1518 | if (truncated && VN_DIRTY(vp) && ip->i_delayed_blks > 0) |
739bfb2a | 1519 | xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE); |
fbf3ce8d CH |
1520 | } |
1521 | ||
1da177e4 LT |
1522 | if (ip->i_d.di_nlink != 0) { |
1523 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1524 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1525 | ip->i_delayed_blks > 0)) && |
1da177e4 | 1526 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
1527 | (!(ip->i_d.di_flags & |
1528 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
92dfe8d2 DC |
1529 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1530 | if (error) | |
014c2544 | 1531 | return error; |
1da177e4 LT |
1532 | } |
1533 | } | |
1534 | ||
1535 | return 0; | |
1536 | } | |
1537 | ||
1538 | /* | |
1539 | * xfs_inactive | |
1540 | * | |
1541 | * This is called when the vnode reference count for the vnode | |
1542 | * goes to zero. If the file has been unlinked, then it must | |
1543 | * now be truncated. Also, we clear all of the read-ahead state | |
1544 | * kept for the inode here since the file is now closed. | |
1545 | */ | |
993386c1 | 1546 | int |
1da177e4 | 1547 | xfs_inactive( |
993386c1 | 1548 | xfs_inode_t *ip) |
1da177e4 | 1549 | { |
993386c1 | 1550 | bhv_vnode_t *vp = XFS_ITOV(ip); |
67fcaa73 | 1551 | xfs_bmap_free_t free_list; |
1da177e4 LT |
1552 | xfs_fsblock_t first_block; |
1553 | int committed; | |
1554 | xfs_trans_t *tp; | |
1555 | xfs_mount_t *mp; | |
1556 | int error; | |
1557 | int truncate; | |
1558 | ||
cf441eeb | 1559 | xfs_itrace_entry(ip); |
1da177e4 | 1560 | |
1da177e4 LT |
1561 | /* |
1562 | * If the inode is already free, then there can be nothing | |
1563 | * to clean up here. | |
1564 | */ | |
1565 | if (ip->i_d.di_mode == 0 || VN_BAD(vp)) { | |
1566 | ASSERT(ip->i_df.if_real_bytes == 0); | |
1567 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1568 | return VN_INACTIVE_CACHE; | |
1569 | } | |
1570 | ||
1571 | /* | |
1572 | * Only do a truncate if it's a regular file with | |
1573 | * some actual space in it. It's OK to look at the | |
1574 | * inode's fields without the lock because we're the | |
1575 | * only one with a reference to the inode. | |
1576 | */ | |
1577 | truncate = ((ip->i_d.di_nlink == 0) && | |
ba87ea69 LM |
1578 | ((ip->i_d.di_size != 0) || (ip->i_size != 0) || |
1579 | (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1580 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1581 | ||
1582 | mp = ip->i_mount; | |
1583 | ||
eb9df39d | 1584 | if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) { |
1da177e4 LT |
1585 | (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL); |
1586 | } | |
1587 | ||
1588 | error = 0; | |
1589 | ||
1590 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1591 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1592 | goto out; |
1593 | ||
1594 | if (ip->i_d.di_nlink != 0) { | |
1595 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1596 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1597 | ip->i_delayed_blks > 0)) && |
dd9f438e NS |
1598 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1599 | (!(ip->i_d.di_flags & | |
1600 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1601 | (ip->i_delayed_blks != 0)))) { | |
92dfe8d2 DC |
1602 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1603 | if (error) | |
014c2544 | 1604 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1605 | } |
1606 | goto out; | |
1607 | } | |
1608 | ||
1609 | ASSERT(ip->i_d.di_nlink == 0); | |
1610 | ||
1611 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1612 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1613 | |
1614 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1615 | if (truncate) { | |
1616 | /* | |
1617 | * Do the xfs_itruncate_start() call before | |
1618 | * reserving any log space because itruncate_start | |
1619 | * will call into the buffer cache and we can't | |
1620 | * do that within a transaction. | |
1621 | */ | |
1622 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1623 | ||
d3cf2094 LM |
1624 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); |
1625 | if (error) { | |
87ae3c24 | 1626 | xfs_trans_cancel(tp, 0); |
d3cf2094 LM |
1627 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1628 | return VN_INACTIVE_CACHE; | |
1629 | } | |
1da177e4 LT |
1630 | |
1631 | error = xfs_trans_reserve(tp, 0, | |
1632 | XFS_ITRUNCATE_LOG_RES(mp), | |
1633 | 0, XFS_TRANS_PERM_LOG_RES, | |
1634 | XFS_ITRUNCATE_LOG_COUNT); | |
1635 | if (error) { | |
1636 | /* Don't call itruncate_cleanup */ | |
1637 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1638 | xfs_trans_cancel(tp, 0); | |
1639 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1640 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1641 | } |
1642 | ||
1643 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1644 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1645 | xfs_trans_ihold(tp, ip); | |
1646 | ||
1647 | /* | |
1648 | * normally, we have to run xfs_itruncate_finish sync. | |
1649 | * But if filesystem is wsync and we're in the inactive | |
1650 | * path, then we know that nlink == 0, and that the | |
1651 | * xaction that made nlink == 0 is permanently committed | |
1652 | * since xfs_remove runs as a synchronous transaction. | |
1653 | */ | |
1654 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1655 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1656 | ||
1657 | if (error) { | |
1658 | xfs_trans_cancel(tp, | |
1659 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1660 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1661 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1662 | } |
1663 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1664 | ||
1665 | /* | |
1666 | * If we get an error while cleaning up a | |
1667 | * symlink we bail out. | |
1668 | */ | |
1669 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1670 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1671 | xfs_inactive_symlink_local(ip, &tp); | |
1672 | ||
1673 | if (error) { | |
1674 | ASSERT(tp == NULL); | |
014c2544 | 1675 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1676 | } |
1677 | ||
1678 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1679 | xfs_trans_ihold(tp, ip); | |
1680 | } else { | |
1681 | error = xfs_trans_reserve(tp, 0, | |
1682 | XFS_IFREE_LOG_RES(mp), | |
1683 | 0, XFS_TRANS_PERM_LOG_RES, | |
1684 | XFS_INACTIVE_LOG_COUNT); | |
1685 | if (error) { | |
1686 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1687 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1688 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1689 | } |
1690 | ||
1691 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1692 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1693 | xfs_trans_ihold(tp, ip); | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * If there are attributes associated with the file | |
1698 | * then blow them away now. The code calls a routine | |
1699 | * that recursively deconstructs the attribute fork. | |
1700 | * We need to just commit the current transaction | |
1701 | * because we can't use it for xfs_attr_inactive(). | |
1702 | */ | |
1703 | if (ip->i_d.di_anextents > 0) { | |
1704 | error = xfs_inactive_attrs(ip, &tp); | |
1705 | /* | |
1706 | * If we got an error, the transaction is already | |
1707 | * cancelled, and the inode is unlocked. Just get out. | |
1708 | */ | |
1709 | if (error) | |
014c2544 | 1710 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1711 | } else if (ip->i_afp) { |
1712 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * Free the inode. | |
1717 | */ | |
1718 | XFS_BMAP_INIT(&free_list, &first_block); | |
1719 | error = xfs_ifree(tp, ip, &free_list); | |
1720 | if (error) { | |
1721 | /* | |
1722 | * If we fail to free the inode, shut down. The cancel | |
1723 | * might do that, we need to make sure. Otherwise the | |
1724 | * inode might be lost for a long time or forever. | |
1725 | */ | |
1726 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1727 | cmn_err(CE_NOTE, | |
1728 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1729 | error, mp->m_fsname); | |
7d04a335 | 1730 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1731 | } |
1732 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1733 | } else { | |
1734 | /* | |
1735 | * Credit the quota account(s). The inode is gone. | |
1736 | */ | |
1737 | XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1738 | ||
1739 | /* | |
1740 | * Just ignore errors at this point. There is | |
1741 | * nothing we can do except to try to keep going. | |
1742 | */ | |
f7c99b6f | 1743 | (void) xfs_bmap_finish(&tp, &free_list, &committed); |
1c72bf90 | 1744 | (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1745 | } |
1746 | /* | |
1747 | * Release the dquots held by inode, if any. | |
1748 | */ | |
1749 | XFS_QM_DQDETACH(mp, ip); | |
1750 | ||
1751 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1752 | ||
1753 | out: | |
1754 | return VN_INACTIVE_CACHE; | |
1755 | } | |
1756 | ||
1757 | ||
993386c1 | 1758 | int |
1da177e4 | 1759 | xfs_lookup( |
993386c1 | 1760 | xfs_inode_t *dp, |
8285fb58 | 1761 | bhv_vname_t *dentry, |
993386c1 | 1762 | bhv_vnode_t **vpp) |
1da177e4 | 1763 | { |
993386c1 | 1764 | xfs_inode_t *ip; |
1da177e4 LT |
1765 | xfs_ino_t e_inum; |
1766 | int error; | |
1767 | uint lock_mode; | |
1da177e4 | 1768 | |
cf441eeb | 1769 | xfs_itrace_entry(dp); |
1da177e4 LT |
1770 | |
1771 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1772 | return XFS_ERROR(EIO); | |
1773 | ||
1774 | lock_mode = xfs_ilock_map_shared(dp); | |
993386c1 | 1775 | error = xfs_dir_lookup_int(dp, lock_mode, dentry, &e_inum, &ip); |
1da177e4 LT |
1776 | if (!error) { |
1777 | *vpp = XFS_ITOV(ip); | |
cf441eeb | 1778 | xfs_itrace_ref(ip); |
1da177e4 LT |
1779 | } |
1780 | xfs_iunlock_map_shared(dp, lock_mode); | |
1781 | return error; | |
1782 | } | |
1783 | ||
993386c1 | 1784 | int |
1da177e4 | 1785 | xfs_create( |
993386c1 | 1786 | xfs_inode_t *dp, |
8285fb58 | 1787 | bhv_vname_t *dentry, |
3e5daf05 CH |
1788 | mode_t mode, |
1789 | xfs_dev_t rdev, | |
67fcaa73 | 1790 | bhv_vnode_t **vpp, |
1da177e4 LT |
1791 | cred_t *credp) |
1792 | { | |
1793 | char *name = VNAME(dentry); | |
993386c1 CH |
1794 | xfs_mount_t *mp = dp->i_mount; |
1795 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); | |
1796 | xfs_inode_t *ip; | |
67fcaa73 | 1797 | bhv_vnode_t *vp = NULL; |
1da177e4 | 1798 | xfs_trans_t *tp; |
1da177e4 LT |
1799 | int error; |
1800 | xfs_bmap_free_t free_list; | |
1801 | xfs_fsblock_t first_block; | |
993386c1 | 1802 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1803 | int dm_event_sent = 0; |
1804 | uint cancel_flags; | |
1805 | int committed; | |
1806 | xfs_prid_t prid; | |
1807 | struct xfs_dquot *udqp, *gdqp; | |
1808 | uint resblks; | |
1da177e4 LT |
1809 | int namelen; |
1810 | ||
1811 | ASSERT(!*vpp); | |
cf441eeb | 1812 | xfs_itrace_entry(dp); |
1da177e4 | 1813 | |
1da177e4 LT |
1814 | namelen = VNAMELEN(dentry); |
1815 | ||
eb9df39d | 1816 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 LT |
1817 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
1818 | dir_vp, DM_RIGHT_NULL, NULL, | |
1819 | DM_RIGHT_NULL, name, NULL, | |
3e5daf05 | 1820 | mode, 0, 0); |
1da177e4 LT |
1821 | |
1822 | if (error) | |
1823 | return error; | |
1824 | dm_event_sent = 1; | |
1825 | } | |
1826 | ||
1827 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1828 | return XFS_ERROR(EIO); | |
1829 | ||
1830 | /* Return through std_return after this point. */ | |
1831 | ||
1832 | udqp = gdqp = NULL; | |
365ca83d NS |
1833 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1834 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
1835 | else |
1836 | prid = (xfs_prid_t)dfltprid; | |
1837 | ||
1838 | /* | |
1839 | * Make sure that we have allocated dquot(s) on disk. | |
1840 | */ | |
1841 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 1842 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
1843 | XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1844 | if (error) | |
1845 | goto std_return; | |
1846 | ||
1847 | ip = NULL; | |
1da177e4 LT |
1848 | |
1849 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1850 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1851 | resblks = XFS_CREATE_SPACE_RES(mp, namelen); | |
1852 | /* | |
1853 | * Initially assume that the file does not exist and | |
1854 | * reserve the resources for that case. If that is not | |
1855 | * the case we'll drop the one we have and get a more | |
1856 | * appropriate transaction later. | |
1857 | */ | |
1858 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1859 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1860 | if (error == ENOSPC) { | |
1861 | resblks = 0; | |
1862 | error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0, | |
1863 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1864 | } | |
1865 | if (error) { | |
1866 | cancel_flags = 0; | |
1da177e4 LT |
1867 | goto error_return; |
1868 | } | |
1869 | ||
f7c66ce3 | 1870 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1871 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
1872 | |
1873 | XFS_BMAP_INIT(&free_list, &first_block); | |
1874 | ||
1875 | ASSERT(ip == NULL); | |
1876 | ||
1877 | /* | |
1878 | * Reserve disk quota and the inode. | |
1879 | */ | |
1880 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
1881 | if (error) | |
1882 | goto error_return; | |
1883 | ||
f6c2d1fa | 1884 | if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen))) |
1da177e4 | 1885 | goto error_return; |
3e5daf05 | 1886 | error = xfs_dir_ialloc(&tp, dp, mode, 1, |
1da177e4 LT |
1887 | rdev, credp, prid, resblks > 0, |
1888 | &ip, &committed); | |
1889 | if (error) { | |
1890 | if (error == ENOSPC) | |
1891 | goto error_return; | |
1892 | goto abort_return; | |
1893 | } | |
cf441eeb | 1894 | xfs_itrace_ref(ip); |
1da177e4 LT |
1895 | |
1896 | /* | |
1897 | * At this point, we've gotten a newly allocated inode. | |
1898 | * It is locked (and joined to the transaction). | |
1899 | */ | |
1900 | ||
1901 | ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE)); | |
1902 | ||
1903 | /* | |
993386c1 CH |
1904 | * Now we join the directory inode to the transaction. We do not do it |
1905 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1906 | * (and release all the locks). An error from here on will result in | |
1907 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1908 | * error path. | |
1da177e4 | 1909 | */ |
1da177e4 LT |
1910 | VN_HOLD(dir_vp); |
1911 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 1912 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 1913 | |
f6c2d1fa NS |
1914 | error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino, |
1915 | &first_block, &free_list, resblks ? | |
1916 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1917 | if (error) { |
1918 | ASSERT(error != ENOSPC); | |
1919 | goto abort_return; | |
1920 | } | |
1921 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1922 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1923 | ||
1924 | /* | |
1925 | * If this is a synchronous mount, make sure that the | |
1926 | * create transaction goes to disk before returning to | |
1927 | * the user. | |
1928 | */ | |
1929 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1930 | xfs_trans_set_sync(tp); | |
1931 | } | |
1932 | ||
1933 | dp->i_gen++; | |
1934 | ||
1935 | /* | |
1936 | * Attach the dquot(s) to the inodes and modify them incore. | |
1937 | * These ids of the inode couldn't have changed since the new | |
1938 | * inode has been locked ever since it was created. | |
1939 | */ | |
1940 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
1941 | ||
1942 | /* | |
1943 | * xfs_trans_commit normally decrements the vnode ref count | |
1944 | * when it unlocks the inode. Since we want to return the | |
1945 | * vnode to the caller, we bump the vnode ref count now. | |
1946 | */ | |
1947 | IHOLD(ip); | |
1948 | vp = XFS_ITOV(ip); | |
1949 | ||
f7c99b6f | 1950 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1951 | if (error) { |
1952 | xfs_bmap_cancel(&free_list); | |
1953 | goto abort_rele; | |
1954 | } | |
1955 | ||
1c72bf90 | 1956 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1957 | if (error) { |
1958 | IRELE(ip); | |
1959 | tp = NULL; | |
1960 | goto error_return; | |
1961 | } | |
1962 | ||
1963 | XFS_QM_DQRELE(mp, udqp); | |
1964 | XFS_QM_DQRELE(mp, gdqp); | |
1965 | ||
1da177e4 LT |
1966 | *vpp = vp; |
1967 | ||
1968 | /* Fallthrough to std_return with error = 0 */ | |
1969 | ||
1970 | std_return: | |
eb9df39d | 1971 | if ((*vpp || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 1972 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 LT |
1973 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
1974 | dir_vp, DM_RIGHT_NULL, | |
1975 | *vpp ? vp:NULL, | |
1976 | DM_RIGHT_NULL, name, NULL, | |
3e5daf05 | 1977 | mode, error, 0); |
1da177e4 LT |
1978 | } |
1979 | return error; | |
1980 | ||
1981 | abort_return: | |
1982 | cancel_flags |= XFS_TRANS_ABORT; | |
1983 | /* FALLTHROUGH */ | |
1da177e4 | 1984 | |
014c2544 | 1985 | error_return: |
1da177e4 LT |
1986 | if (tp != NULL) |
1987 | xfs_trans_cancel(tp, cancel_flags); | |
1988 | ||
1da177e4 LT |
1989 | XFS_QM_DQRELE(mp, udqp); |
1990 | XFS_QM_DQRELE(mp, gdqp); | |
1991 | ||
993386c1 CH |
1992 | if (unlock_dp_on_error) |
1993 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1994 | ||
1da177e4 LT |
1995 | goto std_return; |
1996 | ||
1997 | abort_rele: | |
1998 | /* | |
1999 | * Wait until after the current transaction is aborted to | |
2000 | * release the inode. This prevents recursive transactions | |
2001 | * and deadlocks from xfs_inactive. | |
2002 | */ | |
2003 | cancel_flags |= XFS_TRANS_ABORT; | |
2004 | xfs_trans_cancel(tp, cancel_flags); | |
2005 | IRELE(ip); | |
2006 | ||
2007 | XFS_QM_DQRELE(mp, udqp); | |
2008 | XFS_QM_DQRELE(mp, gdqp); | |
2009 | ||
2010 | goto std_return; | |
2011 | } | |
2012 | ||
2013 | #ifdef DEBUG | |
2014 | /* | |
2015 | * Some counters to see if (and how often) we are hitting some deadlock | |
2016 | * prevention code paths. | |
2017 | */ | |
2018 | ||
2019 | int xfs_rm_locks; | |
2020 | int xfs_rm_lock_delays; | |
2021 | int xfs_rm_attempts; | |
2022 | #endif | |
2023 | ||
2024 | /* | |
2025 | * The following routine will lock the inodes associated with the | |
2026 | * directory and the named entry in the directory. The locks are | |
2027 | * acquired in increasing inode number. | |
2028 | * | |
2029 | * If the entry is "..", then only the directory is locked. The | |
2030 | * vnode ref count will still include that from the .. entry in | |
2031 | * this case. | |
2032 | * | |
2033 | * There is a deadlock we need to worry about. If the locked directory is | |
2034 | * in the AIL, it might be blocking up the log. The next inode we lock | |
2035 | * could be already locked by another thread waiting for log space (e.g | |
2036 | * a permanent log reservation with a long running transaction (see | |
2037 | * xfs_itruncate_finish)). To solve this, we must check if the directory | |
2038 | * is in the ail and use lock_nowait. If we can't lock, we need to | |
2039 | * drop the inode lock on the directory and try again. xfs_iunlock will | |
2040 | * potentially push the tail if we were holding up the log. | |
2041 | */ | |
2042 | STATIC int | |
2043 | xfs_lock_dir_and_entry( | |
2044 | xfs_inode_t *dp, | |
1da177e4 LT |
2045 | xfs_inode_t *ip) /* inode of entry 'name' */ |
2046 | { | |
2047 | int attempts; | |
2048 | xfs_ino_t e_inum; | |
2049 | xfs_inode_t *ips[2]; | |
2050 | xfs_log_item_t *lp; | |
2051 | ||
2052 | #ifdef DEBUG | |
2053 | xfs_rm_locks++; | |
2054 | #endif | |
2055 | attempts = 0; | |
2056 | ||
2057 | again: | |
f7c66ce3 | 2058 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
1da177e4 LT |
2059 | |
2060 | e_inum = ip->i_ino; | |
2061 | ||
cf441eeb | 2062 | xfs_itrace_ref(ip); |
1da177e4 LT |
2063 | |
2064 | /* | |
2065 | * We want to lock in increasing inum. Since we've already | |
2066 | * acquired the lock on the directory, we may need to release | |
2067 | * if if the inum of the entry turns out to be less. | |
2068 | */ | |
2069 | if (e_inum > dp->i_ino) { | |
2070 | /* | |
2071 | * We are already in the right order, so just | |
2072 | * lock on the inode of the entry. | |
2073 | * We need to use nowait if dp is in the AIL. | |
2074 | */ | |
2075 | ||
2076 | lp = (xfs_log_item_t *)dp->i_itemp; | |
2077 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2078 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2079 | attempts++; | |
2080 | #ifdef DEBUG | |
2081 | xfs_rm_attempts++; | |
2082 | #endif | |
2083 | ||
2084 | /* | |
2085 | * Unlock dp and try again. | |
2086 | * xfs_iunlock will try to push the tail | |
2087 | * if the inode is in the AIL. | |
2088 | */ | |
2089 | ||
2090 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2091 | ||
2092 | if ((attempts % 5) == 0) { | |
2093 | delay(1); /* Don't just spin the CPU */ | |
2094 | #ifdef DEBUG | |
2095 | xfs_rm_lock_delays++; | |
2096 | #endif | |
2097 | } | |
2098 | goto again; | |
2099 | } | |
2100 | } else { | |
2101 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2102 | } | |
2103 | } else if (e_inum < dp->i_ino) { | |
2104 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2105 | ||
2106 | ips[0] = ip; | |
2107 | ips[1] = dp; | |
2108 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2109 | } | |
2110 | /* else e_inum == dp->i_ino */ | |
2111 | /* This can happen if we're asked to lock /x/.. | |
2112 | * the entry is "..", which is also the parent directory. | |
2113 | */ | |
2114 | ||
2115 | return 0; | |
2116 | } | |
2117 | ||
2118 | #ifdef DEBUG | |
2119 | int xfs_locked_n; | |
2120 | int xfs_small_retries; | |
2121 | int xfs_middle_retries; | |
2122 | int xfs_lots_retries; | |
2123 | int xfs_lock_delays; | |
2124 | #endif | |
2125 | ||
f7c66ce3 LM |
2126 | /* |
2127 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
2128 | * a different value | |
2129 | */ | |
2130 | static inline int | |
2131 | xfs_lock_inumorder(int lock_mode, int subclass) | |
2132 | { | |
2133 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 2134 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 2135 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 2136 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
2137 | |
2138 | return lock_mode; | |
2139 | } | |
2140 | ||
1da177e4 LT |
2141 | /* |
2142 | * The following routine will lock n inodes in exclusive mode. | |
2143 | * We assume the caller calls us with the inodes in i_ino order. | |
2144 | * | |
2145 | * We need to detect deadlock where an inode that we lock | |
2146 | * is in the AIL and we start waiting for another inode that is locked | |
2147 | * by a thread in a long running transaction (such as truncate). This can | |
2148 | * result in deadlock since the long running trans might need to wait | |
2149 | * for the inode we just locked in order to push the tail and free space | |
2150 | * in the log. | |
2151 | */ | |
2152 | void | |
2153 | xfs_lock_inodes( | |
2154 | xfs_inode_t **ips, | |
2155 | int inodes, | |
2156 | int first_locked, | |
2157 | uint lock_mode) | |
2158 | { | |
2159 | int attempts = 0, i, j, try_lock; | |
2160 | xfs_log_item_t *lp; | |
2161 | ||
2162 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
2163 | ||
2164 | if (first_locked) { | |
2165 | try_lock = 1; | |
2166 | i = 1; | |
2167 | } else { | |
2168 | try_lock = 0; | |
2169 | i = 0; | |
2170 | } | |
2171 | ||
2172 | again: | |
2173 | for (; i < inodes; i++) { | |
2174 | ASSERT(ips[i]); | |
2175 | ||
2176 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
2177 | continue; | |
2178 | ||
2179 | /* | |
2180 | * If try_lock is not set yet, make sure all locked inodes | |
2181 | * are not in the AIL. | |
2182 | * If any are, set try_lock to be used later. | |
2183 | */ | |
2184 | ||
2185 | if (!try_lock) { | |
2186 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
2187 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
2188 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2189 | try_lock++; | |
2190 | } | |
2191 | } | |
2192 | } | |
2193 | ||
2194 | /* | |
2195 | * If any of the previous locks we have locked is in the AIL, | |
2196 | * we must TRY to get the second and subsequent locks. If | |
2197 | * we can't get any, we must release all we have | |
2198 | * and try again. | |
2199 | */ | |
2200 | ||
2201 | if (try_lock) { | |
2202 | /* try_lock must be 0 if i is 0. */ | |
2203 | /* | |
2204 | * try_lock means we have an inode locked | |
2205 | * that is in the AIL. | |
2206 | */ | |
2207 | ASSERT(i != 0); | |
f7c66ce3 | 2208 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
2209 | attempts++; |
2210 | ||
2211 | /* | |
2212 | * Unlock all previous guys and try again. | |
2213 | * xfs_iunlock will try to push the tail | |
2214 | * if the inode is in the AIL. | |
2215 | */ | |
2216 | ||
2217 | for(j = i - 1; j >= 0; j--) { | |
2218 | ||
2219 | /* | |
2220 | * Check to see if we've already | |
2221 | * unlocked this one. | |
2222 | * Not the first one going back, | |
2223 | * and the inode ptr is the same. | |
2224 | */ | |
2225 | if ((j != (i - 1)) && ips[j] == | |
2226 | ips[j+1]) | |
2227 | continue; | |
2228 | ||
2229 | xfs_iunlock(ips[j], lock_mode); | |
2230 | } | |
2231 | ||
2232 | if ((attempts % 5) == 0) { | |
2233 | delay(1); /* Don't just spin the CPU */ | |
2234 | #ifdef DEBUG | |
2235 | xfs_lock_delays++; | |
2236 | #endif | |
2237 | } | |
2238 | i = 0; | |
2239 | try_lock = 0; | |
2240 | goto again; | |
2241 | } | |
2242 | } else { | |
f7c66ce3 | 2243 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
2244 | } |
2245 | } | |
2246 | ||
2247 | #ifdef DEBUG | |
2248 | if (attempts) { | |
2249 | if (attempts < 5) xfs_small_retries++; | |
2250 | else if (attempts < 100) xfs_middle_retries++; | |
2251 | else xfs_lots_retries++; | |
2252 | } else { | |
2253 | xfs_locked_n++; | |
2254 | } | |
2255 | #endif | |
2256 | } | |
2257 | ||
2258 | #ifdef DEBUG | |
2259 | #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);} | |
2260 | int remove_which_error_return = 0; | |
2261 | #else /* ! DEBUG */ | |
2262 | #define REMOVE_DEBUG_TRACE(x) | |
2263 | #endif /* ! DEBUG */ | |
2264 | ||
993386c1 | 2265 | int |
1da177e4 | 2266 | xfs_remove( |
993386c1 CH |
2267 | xfs_inode_t *dp, |
2268 | bhv_vname_t *dentry) | |
1da177e4 | 2269 | { |
993386c1 | 2270 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2271 | char *name = VNAME(dentry); |
993386c1 CH |
2272 | xfs_mount_t *mp = dp->i_mount; |
2273 | xfs_inode_t *ip; | |
1da177e4 | 2274 | xfs_trans_t *tp = NULL; |
1da177e4 LT |
2275 | int error = 0; |
2276 | xfs_bmap_free_t free_list; | |
2277 | xfs_fsblock_t first_block; | |
2278 | int cancel_flags; | |
2279 | int committed; | |
2280 | int dm_di_mode = 0; | |
2281 | int link_zero; | |
2282 | uint resblks; | |
2283 | int namelen; | |
2284 | ||
cf441eeb | 2285 | xfs_itrace_entry(dp); |
1da177e4 | 2286 | |
1da177e4 LT |
2287 | if (XFS_FORCED_SHUTDOWN(mp)) |
2288 | return XFS_ERROR(EIO); | |
2289 | ||
2290 | namelen = VNAMELEN(dentry); | |
2291 | ||
17370097 VA |
2292 | if (!xfs_get_dir_entry(dentry, &ip)) { |
2293 | dm_di_mode = ip->i_d.di_mode; | |
2294 | IRELE(ip); | |
2295 | } | |
2296 | ||
eb9df39d | 2297 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
1da177e4 LT |
2298 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp, |
2299 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
17370097 | 2300 | name, NULL, dm_di_mode, 0, 0); |
1da177e4 LT |
2301 | if (error) |
2302 | return error; | |
2303 | } | |
2304 | ||
2305 | /* From this point on, return through std_return */ | |
2306 | ip = NULL; | |
2307 | ||
2308 | /* | |
2309 | * We need to get a reference to ip before we get our log | |
2310 | * reservation. The reason for this is that we cannot call | |
2311 | * xfs_iget for an inode for which we do not have a reference | |
2312 | * once we've acquired a log reservation. This is because the | |
2313 | * inode we are trying to get might be in xfs_inactive going | |
2314 | * for a log reservation. Since we'll have to wait for the | |
2315 | * inactive code to complete before returning from xfs_iget, | |
2316 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2317 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2318 | * to the inode before getting our log reservation. |
2319 | */ | |
2320 | error = xfs_get_dir_entry(dentry, &ip); | |
2321 | if (error) { | |
2322 | REMOVE_DEBUG_TRACE(__LINE__); | |
2323 | goto std_return; | |
2324 | } | |
2325 | ||
2326 | dm_di_mode = ip->i_d.di_mode; | |
2327 | ||
cf441eeb LM |
2328 | xfs_itrace_entry(ip); |
2329 | xfs_itrace_ref(ip); | |
1da177e4 LT |
2330 | |
2331 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2332 | if (!error && dp != ip) | |
2333 | error = XFS_QM_DQATTACH(mp, ip, 0); | |
2334 | if (error) { | |
2335 | REMOVE_DEBUG_TRACE(__LINE__); | |
2336 | IRELE(ip); | |
2337 | goto std_return; | |
2338 | } | |
2339 | ||
2340 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2341 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2342 | /* | |
2343 | * We try to get the real space reservation first, | |
2344 | * allowing for directory btree deletion(s) implying | |
2345 | * possible bmap insert(s). If we can't get the space | |
2346 | * reservation then we use 0 instead, and avoid the bmap | |
2347 | * btree insert(s) in the directory code by, if the bmap | |
2348 | * insert tries to happen, instead trimming the LAST | |
2349 | * block from the directory. | |
2350 | */ | |
2351 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2352 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2353 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2354 | if (error == ENOSPC) { | |
2355 | resblks = 0; | |
2356 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2357 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2358 | } | |
2359 | if (error) { | |
2360 | ASSERT(error != ENOSPC); | |
2361 | REMOVE_DEBUG_TRACE(__LINE__); | |
2362 | xfs_trans_cancel(tp, 0); | |
2363 | IRELE(ip); | |
2364 | return error; | |
2365 | } | |
2366 | ||
e9ed9d22 | 2367 | error = xfs_lock_dir_and_entry(dp, ip); |
1da177e4 LT |
2368 | if (error) { |
2369 | REMOVE_DEBUG_TRACE(__LINE__); | |
2370 | xfs_trans_cancel(tp, cancel_flags); | |
2371 | IRELE(ip); | |
2372 | goto std_return; | |
2373 | } | |
2374 | ||
2375 | /* | |
2376 | * At this point, we've gotten both the directory and the entry | |
2377 | * inodes locked. | |
2378 | */ | |
2379 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2380 | if (dp != ip) { | |
2381 | /* | |
2382 | * Increment vnode ref count only in this case since | |
2383 | * there's an extra vnode reference in the case where | |
2384 | * dp == ip. | |
2385 | */ | |
2386 | IHOLD(dp); | |
2387 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2388 | } | |
2389 | ||
2390 | /* | |
2391 | * Entry must exist since we did a lookup in xfs_lock_dir_and_entry. | |
2392 | */ | |
2393 | XFS_BMAP_INIT(&free_list, &first_block); | |
f6c2d1fa NS |
2394 | error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino, |
2395 | &first_block, &free_list, 0); | |
1da177e4 LT |
2396 | if (error) { |
2397 | ASSERT(error != ENOENT); | |
2398 | REMOVE_DEBUG_TRACE(__LINE__); | |
2399 | goto error1; | |
2400 | } | |
2401 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2402 | ||
2403 | dp->i_gen++; | |
2404 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2405 | ||
2406 | error = xfs_droplink(tp, ip); | |
2407 | if (error) { | |
2408 | REMOVE_DEBUG_TRACE(__LINE__); | |
2409 | goto error1; | |
2410 | } | |
2411 | ||
2412 | /* Determine if this is the last link while | |
2413 | * we are in the transaction. | |
2414 | */ | |
2415 | link_zero = (ip)->i_d.di_nlink==0; | |
2416 | ||
2417 | /* | |
2418 | * Take an extra ref on the inode so that it doesn't | |
2419 | * go to xfs_inactive() from within the commit. | |
2420 | */ | |
2421 | IHOLD(ip); | |
2422 | ||
2423 | /* | |
2424 | * If this is a synchronous mount, make sure that the | |
2425 | * remove transaction goes to disk before returning to | |
2426 | * the user. | |
2427 | */ | |
2428 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2429 | xfs_trans_set_sync(tp); | |
2430 | } | |
2431 | ||
f7c99b6f | 2432 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2433 | if (error) { |
2434 | REMOVE_DEBUG_TRACE(__LINE__); | |
2435 | goto error_rele; | |
2436 | } | |
2437 | ||
1c72bf90 | 2438 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2439 | if (error) { |
2440 | IRELE(ip); | |
2441 | goto std_return; | |
2442 | } | |
2443 | ||
2a82b8be DC |
2444 | /* |
2445 | * If we are using filestreams, kill the stream association. | |
2446 | * If the file is still open it may get a new one but that | |
2447 | * will get killed on last close in xfs_close() so we don't | |
2448 | * have to worry about that. | |
2449 | */ | |
2450 | if (link_zero && xfs_inode_is_filestream(ip)) | |
2451 | xfs_filestream_deassociate(ip); | |
2452 | ||
cf441eeb | 2453 | xfs_itrace_exit(ip); |
1da177e4 LT |
2454 | IRELE(ip); |
2455 | ||
2456 | /* Fall through to std_return with error = 0 */ | |
2457 | std_return: | |
eb9df39d | 2458 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 LT |
2459 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
2460 | dir_vp, DM_RIGHT_NULL, | |
2461 | NULL, DM_RIGHT_NULL, | |
2462 | name, NULL, dm_di_mode, error, 0); | |
2463 | } | |
2464 | return error; | |
2465 | ||
2466 | error1: | |
2467 | xfs_bmap_cancel(&free_list); | |
2468 | cancel_flags |= XFS_TRANS_ABORT; | |
2469 | xfs_trans_cancel(tp, cancel_flags); | |
2470 | goto std_return; | |
2471 | ||
2472 | error_rele: | |
2473 | /* | |
2474 | * In this case make sure to not release the inode until after | |
2475 | * the current transaction is aborted. Releasing it beforehand | |
2476 | * can cause us to go to xfs_inactive and start a recursive | |
2477 | * transaction which can easily deadlock with the current one. | |
2478 | */ | |
2479 | xfs_bmap_cancel(&free_list); | |
2480 | cancel_flags |= XFS_TRANS_ABORT; | |
2481 | xfs_trans_cancel(tp, cancel_flags); | |
2482 | ||
1da177e4 LT |
2483 | IRELE(ip); |
2484 | ||
2485 | goto std_return; | |
2486 | } | |
2487 | ||
993386c1 | 2488 | int |
1da177e4 | 2489 | xfs_link( |
993386c1 | 2490 | xfs_inode_t *tdp, |
67fcaa73 | 2491 | bhv_vnode_t *src_vp, |
993386c1 | 2492 | bhv_vname_t *dentry) |
1da177e4 | 2493 | { |
993386c1 CH |
2494 | bhv_vnode_t *target_dir_vp = XFS_ITOV(tdp); |
2495 | xfs_mount_t *mp = tdp->i_mount; | |
2496 | xfs_inode_t *sip = xfs_vtoi(src_vp); | |
1da177e4 | 2497 | xfs_trans_t *tp; |
1da177e4 LT |
2498 | xfs_inode_t *ips[2]; |
2499 | int error; | |
2500 | xfs_bmap_free_t free_list; | |
2501 | xfs_fsblock_t first_block; | |
2502 | int cancel_flags; | |
2503 | int committed; | |
1da177e4 LT |
2504 | int resblks; |
2505 | char *target_name = VNAME(dentry); | |
2506 | int target_namelen; | |
2507 | ||
cf441eeb LM |
2508 | xfs_itrace_entry(tdp); |
2509 | xfs_itrace_entry(xfs_vtoi(src_vp)); | |
1da177e4 LT |
2510 | |
2511 | target_namelen = VNAMELEN(dentry); | |
b71d300c | 2512 | ASSERT(!VN_ISDIR(src_vp)); |
1da177e4 | 2513 | |
1da177e4 LT |
2514 | if (XFS_FORCED_SHUTDOWN(mp)) |
2515 | return XFS_ERROR(EIO); | |
2516 | ||
eb9df39d | 2517 | if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) { |
1da177e4 LT |
2518 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, |
2519 | target_dir_vp, DM_RIGHT_NULL, | |
2520 | src_vp, DM_RIGHT_NULL, | |
2521 | target_name, NULL, 0, 0, 0); | |
2522 | if (error) | |
2523 | return error; | |
2524 | } | |
2525 | ||
2526 | /* Return through std_return after this point. */ | |
2527 | ||
2528 | error = XFS_QM_DQATTACH(mp, sip, 0); | |
2529 | if (!error && sip != tdp) | |
2530 | error = XFS_QM_DQATTACH(mp, tdp, 0); | |
2531 | if (error) | |
2532 | goto std_return; | |
2533 | ||
2534 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2535 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2536 | resblks = XFS_LINK_SPACE_RES(mp, target_namelen); | |
2537 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, | |
2538 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2539 | if (error == ENOSPC) { | |
2540 | resblks = 0; | |
2541 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2542 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2543 | } | |
2544 | if (error) { | |
2545 | cancel_flags = 0; | |
2546 | goto error_return; | |
2547 | } | |
2548 | ||
2549 | if (sip->i_ino < tdp->i_ino) { | |
2550 | ips[0] = sip; | |
2551 | ips[1] = tdp; | |
2552 | } else { | |
2553 | ips[0] = tdp; | |
2554 | ips[1] = sip; | |
2555 | } | |
2556 | ||
2557 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2558 | ||
2559 | /* | |
2560 | * Increment vnode ref counts since xfs_trans_commit & | |
2561 | * xfs_trans_cancel will both unlock the inodes and | |
2562 | * decrement the associated ref counts. | |
2563 | */ | |
2564 | VN_HOLD(src_vp); | |
2565 | VN_HOLD(target_dir_vp); | |
2566 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
2567 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2568 | ||
2569 | /* | |
2570 | * If the source has too many links, we can't make any more to it. | |
2571 | */ | |
2572 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2573 | error = XFS_ERROR(EMLINK); | |
2574 | goto error_return; | |
2575 | } | |
2576 | ||
365ca83d NS |
2577 | /* |
2578 | * If we are using project inheritance, we only allow hard link | |
2579 | * creation in our tree when the project IDs are the same; else | |
2580 | * the tree quota mechanism could be circumvented. | |
2581 | */ | |
2582 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2583 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 2584 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
2585 | goto error_return; |
2586 | } | |
2587 | ||
1da177e4 | 2588 | if (resblks == 0 && |
f6c2d1fa | 2589 | (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen))) |
1da177e4 LT |
2590 | goto error_return; |
2591 | ||
2592 | XFS_BMAP_INIT(&free_list, &first_block); | |
2593 | ||
f6c2d1fa | 2594 | error = xfs_dir_createname(tp, tdp, target_name, target_namelen, |
1da177e4 LT |
2595 | sip->i_ino, &first_block, &free_list, |
2596 | resblks); | |
2597 | if (error) | |
2598 | goto abort_return; | |
2599 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2600 | tdp->i_gen++; | |
2601 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
2602 | ||
2603 | error = xfs_bumplink(tp, sip); | |
b71d300c | 2604 | if (error) |
1da177e4 | 2605 | goto abort_return; |
1da177e4 LT |
2606 | |
2607 | /* | |
2608 | * If this is a synchronous mount, make sure that the | |
2609 | * link transaction goes to disk before returning to | |
2610 | * the user. | |
2611 | */ | |
2612 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2613 | xfs_trans_set_sync(tp); | |
2614 | } | |
2615 | ||
f7c99b6f | 2616 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2617 | if (error) { |
2618 | xfs_bmap_cancel(&free_list); | |
2619 | goto abort_return; | |
2620 | } | |
2621 | ||
1c72bf90 | 2622 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
b71d300c | 2623 | if (error) |
1da177e4 | 2624 | goto std_return; |
1da177e4 LT |
2625 | |
2626 | /* Fall through to std_return with error = 0. */ | |
2627 | std_return: | |
eb9df39d | 2628 | if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) { |
1da177e4 LT |
2629 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, |
2630 | target_dir_vp, DM_RIGHT_NULL, | |
2631 | src_vp, DM_RIGHT_NULL, | |
2632 | target_name, NULL, 0, error, 0); | |
2633 | } | |
2634 | return error; | |
2635 | ||
2636 | abort_return: | |
2637 | cancel_flags |= XFS_TRANS_ABORT; | |
2638 | /* FALLTHROUGH */ | |
014c2544 | 2639 | |
1da177e4 LT |
2640 | error_return: |
2641 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2642 | goto std_return; |
2643 | } | |
b71d300c AD |
2644 | |
2645 | ||
993386c1 | 2646 | int |
1da177e4 | 2647 | xfs_mkdir( |
993386c1 | 2648 | xfs_inode_t *dp, |
8285fb58 | 2649 | bhv_vname_t *dentry, |
3e5daf05 | 2650 | mode_t mode, |
67fcaa73 | 2651 | bhv_vnode_t **vpp, |
1da177e4 LT |
2652 | cred_t *credp) |
2653 | { | |
993386c1 | 2654 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2655 | char *dir_name = VNAME(dentry); |
993386c1 CH |
2656 | int dir_namelen = VNAMELEN(dentry); |
2657 | xfs_mount_t *mp = dp->i_mount; | |
1da177e4 | 2658 | xfs_inode_t *cdp; /* inode of created dir */ |
67fcaa73 | 2659 | bhv_vnode_t *cvp; /* vnode of created dir */ |
1da177e4 | 2660 | xfs_trans_t *tp; |
1da177e4 LT |
2661 | int cancel_flags; |
2662 | int error; | |
2663 | int committed; | |
2664 | xfs_bmap_free_t free_list; | |
2665 | xfs_fsblock_t first_block; | |
993386c1 | 2666 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2667 | boolean_t created = B_FALSE; |
2668 | int dm_event_sent = 0; | |
2669 | xfs_prid_t prid; | |
2670 | struct xfs_dquot *udqp, *gdqp; | |
2671 | uint resblks; | |
1da177e4 LT |
2672 | |
2673 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2674 | return XFS_ERROR(EIO); | |
2675 | ||
1da177e4 | 2676 | tp = NULL; |
1da177e4 | 2677 | |
eb9df39d | 2678 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 LT |
2679 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
2680 | dir_vp, DM_RIGHT_NULL, NULL, | |
2681 | DM_RIGHT_NULL, dir_name, NULL, | |
3e5daf05 | 2682 | mode, 0, 0); |
1da177e4 LT |
2683 | if (error) |
2684 | return error; | |
2685 | dm_event_sent = 1; | |
2686 | } | |
2687 | ||
2688 | /* Return through std_return after this point. */ | |
2689 | ||
cf441eeb | 2690 | xfs_itrace_entry(dp); |
1da177e4 LT |
2691 | |
2692 | mp = dp->i_mount; | |
2693 | udqp = gdqp = NULL; | |
365ca83d NS |
2694 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2695 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
2696 | else |
2697 | prid = (xfs_prid_t)dfltprid; | |
2698 | ||
2699 | /* | |
2700 | * Make sure that we have allocated dquot(s) on disk. | |
2701 | */ | |
2702 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 2703 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
2704 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2705 | if (error) | |
2706 | goto std_return; | |
2707 | ||
2708 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
2709 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2710 | resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen); | |
2711 | error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0, | |
2712 | XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT); | |
2713 | if (error == ENOSPC) { | |
2714 | resblks = 0; | |
2715 | error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0, | |
2716 | XFS_TRANS_PERM_LOG_RES, | |
2717 | XFS_MKDIR_LOG_COUNT); | |
2718 | } | |
2719 | if (error) { | |
2720 | cancel_flags = 0; | |
1da177e4 LT |
2721 | goto error_return; |
2722 | } | |
2723 | ||
f7c66ce3 | 2724 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2725 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2726 | |
2727 | /* | |
2728 | * Check for directory link count overflow. | |
2729 | */ | |
2730 | if (dp->i_d.di_nlink >= XFS_MAXLINK) { | |
2731 | error = XFS_ERROR(EMLINK); | |
2732 | goto error_return; | |
2733 | } | |
2734 | ||
2735 | /* | |
2736 | * Reserve disk quota and the inode. | |
2737 | */ | |
2738 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
2739 | if (error) | |
2740 | goto error_return; | |
2741 | ||
2742 | if (resblks == 0 && | |
f6c2d1fa | 2743 | (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen))) |
1da177e4 LT |
2744 | goto error_return; |
2745 | /* | |
2746 | * create the directory inode. | |
2747 | */ | |
3e5daf05 | 2748 | error = xfs_dir_ialloc(&tp, dp, mode, 2, |
1da177e4 LT |
2749 | 0, credp, prid, resblks > 0, |
2750 | &cdp, NULL); | |
2751 | if (error) { | |
2752 | if (error == ENOSPC) | |
2753 | goto error_return; | |
2754 | goto abort_return; | |
2755 | } | |
cf441eeb | 2756 | xfs_itrace_ref(cdp); |
1da177e4 LT |
2757 | |
2758 | /* | |
2759 | * Now we add the directory inode to the transaction. | |
2760 | * We waited until now since xfs_dir_ialloc might start | |
2761 | * a new transaction. Had we joined the transaction | |
993386c1 CH |
2762 | * earlier, the locks might have gotten released. An error |
2763 | * from here on will result in the transaction cancel | |
2764 | * unlocking dp so don't do it explicitly in the error path. | |
1da177e4 LT |
2765 | */ |
2766 | VN_HOLD(dir_vp); | |
2767 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 2768 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2769 | |
2770 | XFS_BMAP_INIT(&free_list, &first_block); | |
2771 | ||
f6c2d1fa NS |
2772 | error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino, |
2773 | &first_block, &free_list, resblks ? | |
2774 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
2775 | if (error) { |
2776 | ASSERT(error != ENOSPC); | |
2777 | goto error1; | |
2778 | } | |
2779 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2780 | ||
2781 | /* | |
2782 | * Bump the in memory version number of the parent directory | |
2783 | * so that other processes accessing it will recognize that | |
2784 | * the directory has changed. | |
2785 | */ | |
2786 | dp->i_gen++; | |
2787 | ||
f6c2d1fa NS |
2788 | error = xfs_dir_init(tp, cdp, dp); |
2789 | if (error) | |
1da177e4 | 2790 | goto error2; |
1da177e4 LT |
2791 | |
2792 | cdp->i_gen = 1; | |
2793 | error = xfs_bumplink(tp, dp); | |
f6c2d1fa | 2794 | if (error) |
1da177e4 | 2795 | goto error2; |
1da177e4 LT |
2796 | |
2797 | cvp = XFS_ITOV(cdp); | |
2798 | ||
2799 | created = B_TRUE; | |
2800 | ||
2801 | *vpp = cvp; | |
2802 | IHOLD(cdp); | |
2803 | ||
2804 | /* | |
2805 | * Attach the dquots to the new inode and modify the icount incore. | |
2806 | */ | |
2807 | XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp); | |
2808 | ||
2809 | /* | |
2810 | * If this is a synchronous mount, make sure that the | |
2811 | * mkdir transaction goes to disk before returning to | |
2812 | * the user. | |
2813 | */ | |
2814 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2815 | xfs_trans_set_sync(tp); | |
2816 | } | |
2817 | ||
f7c99b6f | 2818 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2819 | if (error) { |
2820 | IRELE(cdp); | |
2821 | goto error2; | |
2822 | } | |
2823 | ||
1c72bf90 | 2824 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2825 | XFS_QM_DQRELE(mp, udqp); |
2826 | XFS_QM_DQRELE(mp, gdqp); | |
2827 | if (error) { | |
2828 | IRELE(cdp); | |
2829 | } | |
2830 | ||
2831 | /* Fall through to std_return with error = 0 or errno from | |
2832 | * xfs_trans_commit. */ | |
2833 | ||
2834 | std_return: | |
eb9df39d | 2835 | if ((created || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 2836 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 LT |
2837 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
2838 | dir_vp, DM_RIGHT_NULL, | |
2839 | created ? XFS_ITOV(cdp):NULL, | |
2840 | DM_RIGHT_NULL, | |
2841 | dir_name, NULL, | |
3e5daf05 | 2842 | mode, error, 0); |
1da177e4 LT |
2843 | } |
2844 | return error; | |
2845 | ||
2846 | error2: | |
2847 | error1: | |
2848 | xfs_bmap_cancel(&free_list); | |
2849 | abort_return: | |
2850 | cancel_flags |= XFS_TRANS_ABORT; | |
2851 | error_return: | |
2852 | xfs_trans_cancel(tp, cancel_flags); | |
2853 | XFS_QM_DQRELE(mp, udqp); | |
2854 | XFS_QM_DQRELE(mp, gdqp); | |
2855 | ||
993386c1 | 2856 | if (unlock_dp_on_error) |
1da177e4 | 2857 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2858 | |
2859 | goto std_return; | |
2860 | } | |
2861 | ||
993386c1 | 2862 | int |
1da177e4 | 2863 | xfs_rmdir( |
993386c1 CH |
2864 | xfs_inode_t *dp, |
2865 | bhv_vname_t *dentry) | |
1da177e4 | 2866 | { |
993386c1 | 2867 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2868 | char *name = VNAME(dentry); |
993386c1 CH |
2869 | int namelen = VNAMELEN(dentry); |
2870 | xfs_mount_t *mp = dp->i_mount; | |
2871 | xfs_inode_t *cdp; /* child directory */ | |
1da177e4 | 2872 | xfs_trans_t *tp; |
1da177e4 LT |
2873 | int error; |
2874 | xfs_bmap_free_t free_list; | |
2875 | xfs_fsblock_t first_block; | |
2876 | int cancel_flags; | |
2877 | int committed; | |
17370097 | 2878 | int dm_di_mode = S_IFDIR; |
1da177e4 | 2879 | int last_cdp_link; |
1da177e4 LT |
2880 | uint resblks; |
2881 | ||
cf441eeb | 2882 | xfs_itrace_entry(dp); |
1da177e4 | 2883 | |
993386c1 | 2884 | if (XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 | 2885 | return XFS_ERROR(EIO); |
1da177e4 | 2886 | |
17370097 VA |
2887 | if (!xfs_get_dir_entry(dentry, &cdp)) { |
2888 | dm_di_mode = cdp->i_d.di_mode; | |
2889 | IRELE(cdp); | |
2890 | } | |
2891 | ||
eb9df39d | 2892 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
1da177e4 LT |
2893 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, |
2894 | dir_vp, DM_RIGHT_NULL, | |
2895 | NULL, DM_RIGHT_NULL, | |
17370097 | 2896 | name, NULL, dm_di_mode, 0, 0); |
1da177e4 LT |
2897 | if (error) |
2898 | return XFS_ERROR(error); | |
2899 | } | |
2900 | ||
2901 | /* Return through std_return after this point. */ | |
2902 | ||
2903 | cdp = NULL; | |
2904 | ||
2905 | /* | |
2906 | * We need to get a reference to cdp before we get our log | |
2907 | * reservation. The reason for this is that we cannot call | |
2908 | * xfs_iget for an inode for which we do not have a reference | |
2909 | * once we've acquired a log reservation. This is because the | |
2910 | * inode we are trying to get might be in xfs_inactive going | |
2911 | * for a log reservation. Since we'll have to wait for the | |
2912 | * inactive code to complete before returning from xfs_iget, | |
2913 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2914 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2915 | * to the inode before getting our log reservation. |
2916 | */ | |
2917 | error = xfs_get_dir_entry(dentry, &cdp); | |
2918 | if (error) { | |
2919 | REMOVE_DEBUG_TRACE(__LINE__); | |
2920 | goto std_return; | |
2921 | } | |
2922 | mp = dp->i_mount; | |
2923 | dm_di_mode = cdp->i_d.di_mode; | |
2924 | ||
2925 | /* | |
2926 | * Get the dquots for the inodes. | |
2927 | */ | |
2928 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2929 | if (!error && dp != cdp) | |
2930 | error = XFS_QM_DQATTACH(mp, cdp, 0); | |
2931 | if (error) { | |
2932 | IRELE(cdp); | |
2933 | REMOVE_DEBUG_TRACE(__LINE__); | |
2934 | goto std_return; | |
2935 | } | |
2936 | ||
2937 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
2938 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2939 | /* | |
2940 | * We try to get the real space reservation first, | |
2941 | * allowing for directory btree deletion(s) implying | |
2942 | * possible bmap insert(s). If we can't get the space | |
2943 | * reservation then we use 0 instead, and avoid the bmap | |
2944 | * btree insert(s) in the directory code by, if the bmap | |
2945 | * insert tries to happen, instead trimming the LAST | |
2946 | * block from the directory. | |
2947 | */ | |
2948 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2949 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2950 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
2951 | if (error == ENOSPC) { | |
2952 | resblks = 0; | |
2953 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2954 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
2955 | } | |
2956 | if (error) { | |
2957 | ASSERT(error != ENOSPC); | |
2958 | cancel_flags = 0; | |
2959 | IRELE(cdp); | |
2960 | goto error_return; | |
2961 | } | |
2962 | XFS_BMAP_INIT(&free_list, &first_block); | |
2963 | ||
2964 | /* | |
2965 | * Now lock the child directory inode and the parent directory | |
2966 | * inode in the proper order. This will take care of validating | |
2967 | * that the directory entry for the child directory inode has | |
2968 | * not changed while we were obtaining a log reservation. | |
2969 | */ | |
e9ed9d22 | 2970 | error = xfs_lock_dir_and_entry(dp, cdp); |
1da177e4 LT |
2971 | if (error) { |
2972 | xfs_trans_cancel(tp, cancel_flags); | |
2973 | IRELE(cdp); | |
2974 | goto std_return; | |
2975 | } | |
2976 | ||
2977 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2978 | if (dp != cdp) { | |
2979 | /* | |
2980 | * Only increment the parent directory vnode count if | |
2981 | * we didn't bump it in looking up cdp. The only time | |
2982 | * we don't bump it is when we're looking up ".". | |
2983 | */ | |
2984 | VN_HOLD(dir_vp); | |
2985 | } | |
2986 | ||
cf441eeb | 2987 | xfs_itrace_ref(cdp); |
1da177e4 LT |
2988 | xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL); |
2989 | ||
2990 | ASSERT(cdp->i_d.di_nlink >= 2); | |
2991 | if (cdp->i_d.di_nlink != 2) { | |
2992 | error = XFS_ERROR(ENOTEMPTY); | |
2993 | goto error_return; | |
2994 | } | |
f6c2d1fa | 2995 | if (!xfs_dir_isempty(cdp)) { |
1da177e4 LT |
2996 | error = XFS_ERROR(ENOTEMPTY); |
2997 | goto error_return; | |
2998 | } | |
2999 | ||
f6c2d1fa NS |
3000 | error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino, |
3001 | &first_block, &free_list, resblks); | |
3002 | if (error) | |
1da177e4 | 3003 | goto error1; |
1da177e4 LT |
3004 | |
3005 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3006 | ||
3007 | /* | |
3008 | * Bump the in memory generation count on the parent | |
3009 | * directory so that other can know that it has changed. | |
3010 | */ | |
3011 | dp->i_gen++; | |
3012 | ||
3013 | /* | |
3014 | * Drop the link from cdp's "..". | |
3015 | */ | |
3016 | error = xfs_droplink(tp, dp); | |
3017 | if (error) { | |
3018 | goto error1; | |
3019 | } | |
3020 | ||
3021 | /* | |
3022 | * Drop the link from dp to cdp. | |
3023 | */ | |
3024 | error = xfs_droplink(tp, cdp); | |
3025 | if (error) { | |
3026 | goto error1; | |
3027 | } | |
3028 | ||
3029 | /* | |
3030 | * Drop the "." link from cdp to self. | |
3031 | */ | |
3032 | error = xfs_droplink(tp, cdp); | |
3033 | if (error) { | |
3034 | goto error1; | |
3035 | } | |
3036 | ||
3037 | /* Determine these before committing transaction */ | |
3038 | last_cdp_link = (cdp)->i_d.di_nlink==0; | |
3039 | ||
3040 | /* | |
3041 | * Take an extra ref on the child vnode so that it | |
3042 | * does not go to xfs_inactive() from within the commit. | |
3043 | */ | |
3044 | IHOLD(cdp); | |
3045 | ||
3046 | /* | |
3047 | * If this is a synchronous mount, make sure that the | |
3048 | * rmdir transaction goes to disk before returning to | |
3049 | * the user. | |
3050 | */ | |
3051 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3052 | xfs_trans_set_sync(tp); | |
3053 | } | |
3054 | ||
f7c99b6f | 3055 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
3056 | if (error) { |
3057 | xfs_bmap_cancel(&free_list); | |
3058 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | | |
3059 | XFS_TRANS_ABORT)); | |
3060 | IRELE(cdp); | |
3061 | goto std_return; | |
3062 | } | |
3063 | ||
1c72bf90 | 3064 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3065 | if (error) { |
3066 | IRELE(cdp); | |
3067 | goto std_return; | |
3068 | } | |
3069 | ||
3070 | ||
1da177e4 LT |
3071 | IRELE(cdp); |
3072 | ||
3073 | /* Fall through to std_return with error = 0 or the errno | |
3074 | * from xfs_trans_commit. */ | |
bb19fba1 | 3075 | std_return: |
eb9df39d | 3076 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 LT |
3077 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
3078 | dir_vp, DM_RIGHT_NULL, | |
3079 | NULL, DM_RIGHT_NULL, | |
3080 | name, NULL, dm_di_mode, | |
3081 | error, 0); | |
3082 | } | |
3083 | return error; | |
3084 | ||
bb19fba1 | 3085 | error1: |
1da177e4 LT |
3086 | xfs_bmap_cancel(&free_list); |
3087 | cancel_flags |= XFS_TRANS_ABORT; | |
014c2544 JJ |
3088 | /* FALLTHROUGH */ |
3089 | ||
bb19fba1 | 3090 | error_return: |
1da177e4 LT |
3091 | xfs_trans_cancel(tp, cancel_flags); |
3092 | goto std_return; | |
3093 | } | |
3094 | ||
993386c1 | 3095 | int |
1da177e4 | 3096 | xfs_symlink( |
993386c1 | 3097 | xfs_inode_t *dp, |
8285fb58 | 3098 | bhv_vname_t *dentry, |
1da177e4 | 3099 | char *target_path, |
3e5daf05 | 3100 | mode_t mode, |
67fcaa73 | 3101 | bhv_vnode_t **vpp, |
1da177e4 LT |
3102 | cred_t *credp) |
3103 | { | |
993386c1 CH |
3104 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
3105 | xfs_mount_t *mp = dp->i_mount; | |
1da177e4 | 3106 | xfs_trans_t *tp; |
1da177e4 LT |
3107 | xfs_inode_t *ip; |
3108 | int error; | |
3109 | int pathlen; | |
3110 | xfs_bmap_free_t free_list; | |
3111 | xfs_fsblock_t first_block; | |
993386c1 | 3112 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
3113 | uint cancel_flags; |
3114 | int committed; | |
3115 | xfs_fileoff_t first_fsb; | |
3116 | xfs_filblks_t fs_blocks; | |
3117 | int nmaps; | |
3118 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
3119 | xfs_daddr_t d; | |
3120 | char *cur_chunk; | |
3121 | int byte_cnt; | |
3122 | int n; | |
3123 | xfs_buf_t *bp; | |
3124 | xfs_prid_t prid; | |
3125 | struct xfs_dquot *udqp, *gdqp; | |
3126 | uint resblks; | |
3127 | char *link_name = VNAME(dentry); | |
3128 | int link_namelen; | |
3129 | ||
3130 | *vpp = NULL; | |
1da177e4 LT |
3131 | error = 0; |
3132 | ip = NULL; | |
3133 | tp = NULL; | |
3134 | ||
cf441eeb | 3135 | xfs_itrace_entry(dp); |
1da177e4 LT |
3136 | |
3137 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3138 | return XFS_ERROR(EIO); | |
3139 | ||
3140 | link_namelen = VNAMELEN(dentry); | |
3141 | ||
3142 | /* | |
3143 | * Check component lengths of the target path name. | |
3144 | */ | |
3145 | pathlen = strlen(target_path); | |
3146 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
3147 | return XFS_ERROR(ENAMETOOLONG); | |
3148 | if (pathlen >= MAXNAMELEN) { /* is any component too long? */ | |
3149 | int len, total; | |
3150 | char *path; | |
3151 | ||
f6c2d1fa | 3152 | for (total = 0, path = target_path; total < pathlen;) { |
1da177e4 LT |
3153 | /* |
3154 | * Skip any slashes. | |
3155 | */ | |
3156 | while(*path == '/') { | |
3157 | total++; | |
3158 | path++; | |
3159 | } | |
3160 | ||
3161 | /* | |
3162 | * Count up to the next slash or end of path. | |
3163 | * Error out if the component is bigger than MAXNAMELEN. | |
3164 | */ | |
3165 | for(len = 0; *path != '/' && total < pathlen;total++, path++) { | |
3166 | if (++len >= MAXNAMELEN) { | |
3167 | error = ENAMETOOLONG; | |
3168 | return error; | |
3169 | } | |
3170 | } | |
3171 | } | |
3172 | } | |
3173 | ||
eb9df39d | 3174 | if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) { |
1da177e4 LT |
3175 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp, |
3176 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
3177 | link_name, target_path, 0, 0, 0); | |
3178 | if (error) | |
3179 | return error; | |
3180 | } | |
3181 | ||
3182 | /* Return through std_return after this point. */ | |
3183 | ||
3184 | udqp = gdqp = NULL; | |
365ca83d NS |
3185 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
3186 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
3187 | else |
3188 | prid = (xfs_prid_t)dfltprid; | |
3189 | ||
3190 | /* | |
3191 | * Make sure that we have allocated dquot(s) on disk. | |
3192 | */ | |
3193 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 3194 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
3195 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
3196 | if (error) | |
3197 | goto std_return; | |
3198 | ||
3199 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
3200 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3201 | /* | |
3202 | * The symlink will fit into the inode data fork? | |
3203 | * There can't be any attributes so we get the whole variable part. | |
3204 | */ | |
3205 | if (pathlen <= XFS_LITINO(mp)) | |
3206 | fs_blocks = 0; | |
3207 | else | |
3208 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
3209 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks); | |
3210 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, | |
3211 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3212 | if (error == ENOSPC && fs_blocks == 0) { | |
3213 | resblks = 0; | |
3214 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
3215 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3216 | } | |
3217 | if (error) { | |
3218 | cancel_flags = 0; | |
1da177e4 LT |
3219 | goto error_return; |
3220 | } | |
3221 | ||
f7c66ce3 | 3222 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 3223 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
3224 | |
3225 | /* | |
3226 | * Check whether the directory allows new symlinks or not. | |
3227 | */ | |
3228 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
3229 | error = XFS_ERROR(EPERM); | |
3230 | goto error_return; | |
3231 | } | |
3232 | ||
3233 | /* | |
3234 | * Reserve disk quota : blocks and inode. | |
3235 | */ | |
3236 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
3237 | if (error) | |
3238 | goto error_return; | |
3239 | ||
3240 | /* | |
3241 | * Check for ability to enter directory entry, if no space reserved. | |
3242 | */ | |
3243 | if (resblks == 0 && | |
f6c2d1fa | 3244 | (error = xfs_dir_canenter(tp, dp, link_name, link_namelen))) |
1da177e4 LT |
3245 | goto error_return; |
3246 | /* | |
3247 | * Initialize the bmap freelist prior to calling either | |
3248 | * bmapi or the directory create code. | |
3249 | */ | |
3250 | XFS_BMAP_INIT(&free_list, &first_block); | |
3251 | ||
3252 | /* | |
3253 | * Allocate an inode for the symlink. | |
3254 | */ | |
3e5daf05 | 3255 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), |
1da177e4 LT |
3256 | 1, 0, credp, prid, resblks > 0, &ip, NULL); |
3257 | if (error) { | |
3258 | if (error == ENOSPC) | |
3259 | goto error_return; | |
3260 | goto error1; | |
3261 | } | |
cf441eeb | 3262 | xfs_itrace_ref(ip); |
1da177e4 | 3263 | |
993386c1 CH |
3264 | /* |
3265 | * An error after we've joined dp to the transaction will result in the | |
3266 | * transaction cancel unlocking dp so don't do it explicitly in the | |
3267 | * error path. | |
3268 | */ | |
1da177e4 LT |
3269 | VN_HOLD(dir_vp); |
3270 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 3271 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
3272 | |
3273 | /* | |
3274 | * Also attach the dquot(s) to it, if applicable. | |
3275 | */ | |
3276 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
3277 | ||
3278 | if (resblks) | |
3279 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
3280 | /* | |
3281 | * If the symlink will fit into the inode, write it inline. | |
3282 | */ | |
3283 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
3284 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
3285 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
3286 | ip->i_d.di_size = pathlen; | |
3287 | ||
3288 | /* | |
3289 | * The inode was initially created in extent format. | |
3290 | */ | |
3291 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
3292 | ip->i_df.if_flags |= XFS_IFINLINE; | |
3293 | ||
3294 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
3295 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
3296 | ||
3297 | } else { | |
3298 | first_fsb = 0; | |
3299 | nmaps = SYMLINK_MAPS; | |
3300 | ||
3301 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
3302 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
3303 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 3304 | &free_list, NULL); |
1da177e4 LT |
3305 | if (error) { |
3306 | goto error1; | |
3307 | } | |
3308 | ||
3309 | if (resblks) | |
3310 | resblks -= fs_blocks; | |
3311 | ip->i_d.di_size = pathlen; | |
3312 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3313 | ||
3314 | cur_chunk = target_path; | |
3315 | for (n = 0; n < nmaps; n++) { | |
3316 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
3317 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
3318 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
3319 | BTOBB(byte_cnt), 0); | |
3320 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
3321 | if (pathlen < byte_cnt) { | |
3322 | byte_cnt = pathlen; | |
3323 | } | |
3324 | pathlen -= byte_cnt; | |
3325 | ||
3326 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
3327 | cur_chunk += byte_cnt; | |
3328 | ||
3329 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
3330 | } | |
3331 | } | |
3332 | ||
3333 | /* | |
3334 | * Create the directory entry for the symlink. | |
3335 | */ | |
f6c2d1fa NS |
3336 | error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino, |
3337 | &first_block, &free_list, resblks); | |
3338 | if (error) | |
1da177e4 | 3339 | goto error1; |
1da177e4 LT |
3340 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3341 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
3342 | ||
3343 | /* | |
3344 | * Bump the in memory version number of the parent directory | |
3345 | * so that other processes accessing it will recognize that | |
3346 | * the directory has changed. | |
3347 | */ | |
3348 | dp->i_gen++; | |
3349 | ||
3350 | /* | |
3351 | * If this is a synchronous mount, make sure that the | |
3352 | * symlink transaction goes to disk before returning to | |
3353 | * the user. | |
3354 | */ | |
3355 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3356 | xfs_trans_set_sync(tp); | |
3357 | } | |
3358 | ||
3359 | /* | |
3360 | * xfs_trans_commit normally decrements the vnode ref count | |
3361 | * when it unlocks the inode. Since we want to return the | |
3362 | * vnode to the caller, we bump the vnode ref count now. | |
3363 | */ | |
3364 | IHOLD(ip); | |
3365 | ||
f7c99b6f | 3366 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3367 | if (error) { |
3368 | goto error2; | |
3369 | } | |
1c72bf90 | 3370 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3371 | XFS_QM_DQRELE(mp, udqp); |
3372 | XFS_QM_DQRELE(mp, gdqp); | |
3373 | ||
3374 | /* Fall through to std_return with error = 0 or errno from | |
3375 | * xfs_trans_commit */ | |
3376 | std_return: | |
993386c1 | 3377 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) { |
1da177e4 LT |
3378 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, |
3379 | dir_vp, DM_RIGHT_NULL, | |
3380 | error ? NULL : XFS_ITOV(ip), | |
3381 | DM_RIGHT_NULL, link_name, target_path, | |
3382 | 0, error, 0); | |
3383 | } | |
3384 | ||
3385 | if (!error) { | |
67fcaa73 | 3386 | bhv_vnode_t *vp; |
1da177e4 LT |
3387 | |
3388 | ASSERT(ip); | |
3389 | vp = XFS_ITOV(ip); | |
3390 | *vpp = vp; | |
3391 | } | |
3392 | return error; | |
3393 | ||
3394 | error2: | |
3395 | IRELE(ip); | |
3396 | error1: | |
3397 | xfs_bmap_cancel(&free_list); | |
3398 | cancel_flags |= XFS_TRANS_ABORT; | |
3399 | error_return: | |
3400 | xfs_trans_cancel(tp, cancel_flags); | |
3401 | XFS_QM_DQRELE(mp, udqp); | |
3402 | XFS_QM_DQRELE(mp, gdqp); | |
3403 | ||
993386c1 | 3404 | if (unlock_dp_on_error) |
1da177e4 | 3405 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
3406 | |
3407 | goto std_return; | |
3408 | } | |
3409 | ||
1da177e4 LT |
3410 | int |
3411 | xfs_rwlock( | |
993386c1 | 3412 | xfs_inode_t *ip, |
8285fb58 | 3413 | bhv_vrwlock_t locktype) |
1da177e4 | 3414 | { |
993386c1 | 3415 | if (S_ISDIR(ip->i_d.di_mode)) |
1da177e4 | 3416 | return 1; |
1da177e4 LT |
3417 | if (locktype == VRWLOCK_WRITE) { |
3418 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
3419 | } else if (locktype == VRWLOCK_TRY_READ) { | |
014c2544 | 3420 | return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED); |
1da177e4 | 3421 | } else if (locktype == VRWLOCK_TRY_WRITE) { |
014c2544 | 3422 | return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL); |
1da177e4 LT |
3423 | } else { |
3424 | ASSERT((locktype == VRWLOCK_READ) || | |
3425 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3426 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
3427 | } | |
3428 | ||
3429 | return 1; | |
3430 | } | |
3431 | ||
3432 | ||
1da177e4 LT |
3433 | void |
3434 | xfs_rwunlock( | |
993386c1 | 3435 | xfs_inode_t *ip, |
8285fb58 | 3436 | bhv_vrwlock_t locktype) |
1da177e4 | 3437 | { |
993386c1 CH |
3438 | if (S_ISDIR(ip->i_d.di_mode)) |
3439 | return; | |
e9a56b7c LM |
3440 | if (locktype == VRWLOCK_WRITE) { |
3441 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
3442 | } else { | |
1da177e4 LT |
3443 | ASSERT((locktype == VRWLOCK_READ) || |
3444 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3445 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
3446 | } | |
3447 | return; | |
3448 | } | |
3449 | ||
993386c1 CH |
3450 | |
3451 | int | |
1da177e4 | 3452 | xfs_inode_flush( |
993386c1 | 3453 | xfs_inode_t *ip, |
1da177e4 LT |
3454 | int flags) |
3455 | { | |
993386c1 | 3456 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
3457 | int error = 0; |
3458 | ||
1da177e4 LT |
3459 | if (XFS_FORCED_SHUTDOWN(mp)) |
3460 | return XFS_ERROR(EIO); | |
3461 | ||
3462 | /* | |
3463 | * Bypass inodes which have already been cleaned by | |
3464 | * the inode flush clustering code inside xfs_iflush | |
3465 | */ | |
33540408 | 3466 | if (xfs_inode_clean(ip)) |
1da177e4 LT |
3467 | return 0; |
3468 | ||
1da177e4 LT |
3469 | /* |
3470 | * We make this non-blocking if the inode is contended, | |
3471 | * return EAGAIN to indicate to the caller that they | |
3472 | * did not succeed. This prevents the flush path from | |
3473 | * blocking on inodes inside another operation right | |
3474 | * now, they get caught later by xfs_sync. | |
3475 | */ | |
a3f74ffb DC |
3476 | if (flags & FLUSH_SYNC) { |
3477 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3478 | xfs_iflock(ip); | |
3479 | } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
3480 | if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) { | |
3481 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1da177e4 LT |
3482 | return EAGAIN; |
3483 | } | |
a3f74ffb DC |
3484 | } else { |
3485 | return EAGAIN; | |
1da177e4 LT |
3486 | } |
3487 | ||
a3f74ffb DC |
3488 | error = xfs_iflush(ip, (flags & FLUSH_SYNC) ? XFS_IFLUSH_SYNC |
3489 | : XFS_IFLUSH_ASYNC_NOBLOCK); | |
3490 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3491 | ||
1da177e4 LT |
3492 | return error; |
3493 | } | |
3494 | ||
993386c1 | 3495 | |
1da177e4 | 3496 | int |
993386c1 CH |
3497 | xfs_set_dmattrs( |
3498 | xfs_inode_t *ip, | |
1da177e4 | 3499 | u_int evmask, |
993386c1 | 3500 | u_int16_t state) |
1da177e4 | 3501 | { |
993386c1 | 3502 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 3503 | xfs_trans_t *tp; |
1da177e4 LT |
3504 | int error; |
3505 | ||
3506 | if (!capable(CAP_SYS_ADMIN)) | |
3507 | return XFS_ERROR(EPERM); | |
3508 | ||
1da177e4 LT |
3509 | if (XFS_FORCED_SHUTDOWN(mp)) |
3510 | return XFS_ERROR(EIO); | |
3511 | ||
3512 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
3513 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
3514 | if (error) { | |
3515 | xfs_trans_cancel(tp, 0); | |
3516 | return error; | |
3517 | } | |
3518 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3519 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3520 | ||
613d7043 CH |
3521 | ip->i_d.di_dmevmask = evmask; |
3522 | ip->i_d.di_dmstate = state; | |
1da177e4 LT |
3523 | |
3524 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3525 | IHOLD(ip); | |
1c72bf90 | 3526 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
3527 | |
3528 | return error; | |
3529 | } | |
3530 | ||
993386c1 | 3531 | int |
1da177e4 | 3532 | xfs_reclaim( |
993386c1 | 3533 | xfs_inode_t *ip) |
1da177e4 | 3534 | { |
993386c1 | 3535 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1da177e4 | 3536 | |
cf441eeb | 3537 | xfs_itrace_entry(ip); |
1da177e4 LT |
3538 | |
3539 | ASSERT(!VN_MAPPED(vp)); | |
3540 | ||
3541 | /* bad inode, get out here ASAP */ | |
3542 | if (VN_BAD(vp)) { | |
3543 | xfs_ireclaim(ip); | |
3544 | return 0; | |
3545 | } | |
3546 | ||
b677c210 | 3547 | vn_iowait(ip); |
1da177e4 | 3548 | |
51c91ed5 | 3549 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); |
1da177e4 | 3550 | |
42fe2b1f CH |
3551 | /* |
3552 | * Make sure the atime in the XFS inode is correct before freeing the | |
3553 | * Linux inode. | |
3554 | */ | |
3555 | xfs_synchronize_atime(ip); | |
1da177e4 | 3556 | |
4c60658e DC |
3557 | /* |
3558 | * If we have nothing to flush with this inode then complete the | |
3559 | * teardown now, otherwise break the link between the xfs inode and the | |
3560 | * linux inode and clean up the xfs inode later. This avoids flushing | |
3561 | * the inode to disk during the delete operation itself. | |
3562 | * | |
3563 | * When breaking the link, we need to set the XFS_IRECLAIMABLE flag | |
3564 | * first to ensure that xfs_iunpin() will never see an xfs inode | |
3565 | * that has a linux inode being reclaimed. Synchronisation is provided | |
3566 | * by the i_flags_lock. | |
1da177e4 LT |
3567 | */ |
3568 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
3569 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3570 | xfs_iflock(ip); | |
3571 | return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); | |
3572 | } else { | |
3573 | xfs_mount_t *mp = ip->i_mount; | |
3574 | ||
4c60658e | 3575 | /* Protect sync and unpin from us */ |
1da177e4 | 3576 | XFS_MOUNT_ILOCK(mp); |
4c60658e DC |
3577 | spin_lock(&ip->i_flags_lock); |
3578 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
739bfb2a CH |
3579 | vn_to_inode(vp)->i_private = NULL; |
3580 | ip->i_vnode = NULL; | |
4c60658e | 3581 | spin_unlock(&ip->i_flags_lock); |
1da177e4 | 3582 | list_add_tail(&ip->i_reclaim, &mp->m_del_inodes); |
1da177e4 LT |
3583 | XFS_MOUNT_IUNLOCK(mp); |
3584 | } | |
3585 | return 0; | |
3586 | } | |
3587 | ||
3588 | int | |
3589 | xfs_finish_reclaim( | |
3590 | xfs_inode_t *ip, | |
3591 | int locked, | |
3592 | int sync_mode) | |
3593 | { | |
da353b0d | 3594 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); |
67fcaa73 | 3595 | bhv_vnode_t *vp = XFS_ITOV_NULL(ip); |
1da177e4 LT |
3596 | int error; |
3597 | ||
3598 | if (vp && VN_BAD(vp)) | |
3599 | goto reclaim; | |
3600 | ||
3601 | /* The hash lock here protects a thread in xfs_iget_core from | |
3602 | * racing with us on linking the inode back with a vnode. | |
3603 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
3604 | * us. | |
3605 | */ | |
da353b0d | 3606 | write_lock(&pag->pag_ici_lock); |
f273ab84 | 3607 | spin_lock(&ip->i_flags_lock); |
7a18c386 DC |
3608 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || |
3609 | (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) { | |
f273ab84 | 3610 | spin_unlock(&ip->i_flags_lock); |
da353b0d | 3611 | write_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
3612 | if (locked) { |
3613 | xfs_ifunlock(ip); | |
3614 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3615 | } | |
014c2544 | 3616 | return 1; |
1da177e4 | 3617 | } |
7a18c386 | 3618 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
f273ab84 | 3619 | spin_unlock(&ip->i_flags_lock); |
da353b0d DC |
3620 | write_unlock(&pag->pag_ici_lock); |
3621 | xfs_put_perag(ip->i_mount, pag); | |
1da177e4 LT |
3622 | |
3623 | /* | |
3624 | * If the inode is still dirty, then flush it out. If the inode | |
3625 | * is not in the AIL, then it will be OK to flush it delwri as | |
3626 | * long as xfs_iflush() does not keep any references to the inode. | |
3627 | * We leave that decision up to xfs_iflush() since it has the | |
3628 | * knowledge of whether it's OK to simply do a delwri flush of | |
3629 | * the inode or whether we need to wait until the inode is | |
3630 | * pulled from the AIL. | |
3631 | * We get the flush lock regardless, though, just to make sure | |
3632 | * we don't free it while it is being flushed. | |
3633 | */ | |
461aa8a2 LM |
3634 | if (!locked) { |
3635 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3636 | xfs_iflock(ip); | |
3637 | } | |
1da177e4 | 3638 | |
461aa8a2 | 3639 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1da177e4 LT |
3640 | if (ip->i_update_core || |
3641 | ((ip->i_itemp != NULL) && | |
3642 | (ip->i_itemp->ili_format.ilf_fields != 0))) { | |
3643 | error = xfs_iflush(ip, sync_mode); | |
3644 | /* | |
3645 | * If we hit an error, typically because of filesystem | |
3646 | * shutdown, we don't need to let vn_reclaim to know | |
3647 | * because we're gonna reclaim the inode anyway. | |
3648 | */ | |
3649 | if (error) { | |
3650 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3651 | goto reclaim; | |
3652 | } | |
3653 | xfs_iflock(ip); /* synchronize with xfs_iflush_done */ | |
3654 | } | |
3655 | ||
3656 | ASSERT(ip->i_update_core == 0); | |
3657 | ASSERT(ip->i_itemp == NULL || | |
3658 | ip->i_itemp->ili_format.ilf_fields == 0); | |
1da177e4 LT |
3659 | } |
3660 | ||
461aa8a2 LM |
3661 | xfs_ifunlock(ip); |
3662 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3663 | ||
1da177e4 LT |
3664 | reclaim: |
3665 | xfs_ireclaim(ip); | |
3666 | return 0; | |
3667 | } | |
3668 | ||
3669 | int | |
3670 | xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock) | |
3671 | { | |
3672 | int purged; | |
3673 | xfs_inode_t *ip, *n; | |
3674 | int done = 0; | |
3675 | ||
3676 | while (!done) { | |
3677 | purged = 0; | |
3678 | XFS_MOUNT_ILOCK(mp); | |
3679 | list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) { | |
3680 | if (noblock) { | |
3681 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) | |
3682 | continue; | |
3683 | if (xfs_ipincount(ip) || | |
3684 | !xfs_iflock_nowait(ip)) { | |
3685 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3686 | continue; | |
3687 | } | |
3688 | } | |
3689 | XFS_MOUNT_IUNLOCK(mp); | |
6b2cf618 FB |
3690 | if (xfs_finish_reclaim(ip, noblock, |
3691 | XFS_IFLUSH_DELWRI_ELSE_ASYNC)) | |
3692 | delay(1); | |
1da177e4 LT |
3693 | purged = 1; |
3694 | break; | |
3695 | } | |
3696 | ||
3697 | done = !purged; | |
3698 | } | |
3699 | ||
3700 | XFS_MOUNT_IUNLOCK(mp); | |
3701 | return 0; | |
3702 | } | |
3703 | ||
3704 | /* | |
3705 | * xfs_alloc_file_space() | |
3706 | * This routine allocates disk space for the given file. | |
3707 | * | |
3708 | * If alloc_type == 0, this request is for an ALLOCSP type | |
3709 | * request which will change the file size. In this case, no | |
3710 | * DMAPI event will be generated by the call. A TRUNCATE event | |
3711 | * will be generated later by xfs_setattr. | |
3712 | * | |
3713 | * If alloc_type != 0, this request is for a RESVSP type | |
3714 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
3715 | * lower block boundary byte address is less than the file's | |
3716 | * length. | |
3717 | * | |
3718 | * RETURNS: | |
3719 | * 0 on success | |
3720 | * errno on error | |
3721 | * | |
3722 | */ | |
ba0f32d4 | 3723 | STATIC int |
1da177e4 LT |
3724 | xfs_alloc_file_space( |
3725 | xfs_inode_t *ip, | |
3726 | xfs_off_t offset, | |
3727 | xfs_off_t len, | |
3728 | int alloc_type, | |
3729 | int attr_flags) | |
3730 | { | |
dd9f438e NS |
3731 | xfs_mount_t *mp = ip->i_mount; |
3732 | xfs_off_t count; | |
1da177e4 LT |
3733 | xfs_filblks_t allocated_fsb; |
3734 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
3735 | xfs_extlen_t extsz, temp; |
3736 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 3737 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
3738 | int nimaps; |
3739 | int bmapi_flag; | |
3740 | int quota_flag; | |
1da177e4 | 3741 | int rt; |
1da177e4 | 3742 | xfs_trans_t *tp; |
dd9f438e NS |
3743 | xfs_bmbt_irec_t imaps[1], *imapp; |
3744 | xfs_bmap_free_t free_list; | |
3745 | uint qblocks, resblks, resrtextents; | |
3746 | int committed; | |
3747 | int error; | |
1da177e4 | 3748 | |
cf441eeb | 3749 | xfs_itrace_entry(ip); |
1da177e4 LT |
3750 | |
3751 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3752 | return XFS_ERROR(EIO); | |
3753 | ||
1da177e4 LT |
3754 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
3755 | return error; | |
3756 | ||
3757 | if (len <= 0) | |
3758 | return XFS_ERROR(EINVAL); | |
3759 | ||
957d0ebe DC |
3760 | rt = XFS_IS_REALTIME_INODE(ip); |
3761 | extsz = xfs_get_extsz_hint(ip); | |
3762 | ||
1da177e4 | 3763 | count = len; |
1da177e4 | 3764 | imapp = &imaps[0]; |
dd9f438e NS |
3765 | nimaps = 1; |
3766 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
3767 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
3768 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
3769 | ||
3770 | /* Generate a DMAPI event if needed. */ | |
ba87ea69 | 3771 | if (alloc_type != 0 && offset < ip->i_size && |
1da177e4 | 3772 | (attr_flags&ATTR_DMI) == 0 && |
eb9df39d | 3773 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
1da177e4 LT |
3774 | xfs_off_t end_dmi_offset; |
3775 | ||
3776 | end_dmi_offset = offset+len; | |
ba87ea69 LM |
3777 | if (end_dmi_offset > ip->i_size) |
3778 | end_dmi_offset = ip->i_size; | |
1da177e4 LT |
3779 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip), |
3780 | offset, end_dmi_offset - offset, | |
3781 | 0, NULL); | |
3782 | if (error) | |
014c2544 | 3783 | return error; |
1da177e4 LT |
3784 | } |
3785 | ||
3786 | /* | |
dd9f438e | 3787 | * Allocate file space until done or until there is an error |
1da177e4 LT |
3788 | */ |
3789 | retry: | |
3790 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
3791 | xfs_fileoff_t s, e; |
3792 | ||
1da177e4 | 3793 | /* |
3ddb8fa9 | 3794 | * Determine space reservations for data/realtime. |
1da177e4 | 3795 | */ |
dd9f438e | 3796 | if (unlikely(extsz)) { |
1da177e4 | 3797 | s = startoffset_fsb; |
dd9f438e NS |
3798 | do_div(s, extsz); |
3799 | s *= extsz; | |
3800 | e = startoffset_fsb + allocatesize_fsb; | |
3801 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
3802 | e += temp; | |
3803 | if ((temp = do_mod(e, extsz))) | |
3804 | e += extsz - temp; | |
3805 | } else { | |
3806 | s = 0; | |
3807 | e = allocatesize_fsb; | |
3808 | } | |
3809 | ||
3810 | if (unlikely(rt)) { | |
3811 | resrtextents = qblocks = (uint)(e - s); | |
3812 | resrtextents /= mp->m_sb.sb_rextsize; | |
3813 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
3814 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 3815 | } else { |
dd9f438e NS |
3816 | resrtextents = 0; |
3817 | resblks = qblocks = \ | |
3818 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
3819 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
3820 | } |
3821 | ||
3822 | /* | |
dd9f438e | 3823 | * Allocate and setup the transaction. |
1da177e4 LT |
3824 | */ |
3825 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
3826 | error = xfs_trans_reserve(tp, resblks, |
3827 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
3828 | XFS_TRANS_PERM_LOG_RES, |
3829 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 3830 | /* |
dd9f438e | 3831 | * Check for running out of space |
1da177e4 LT |
3832 | */ |
3833 | if (error) { | |
3834 | /* | |
3835 | * Free the transaction structure. | |
3836 | */ | |
3837 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
3838 | xfs_trans_cancel(tp, 0); | |
3839 | break; | |
3840 | } | |
3841 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
dd9f438e NS |
3842 | error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, |
3843 | qblocks, 0, quota_flag); | |
1da177e4 LT |
3844 | if (error) |
3845 | goto error1; | |
3846 | ||
3847 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3848 | xfs_trans_ihold(tp, ip); | |
3849 | ||
3850 | /* | |
dd9f438e | 3851 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 LT |
3852 | */ |
3853 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
541d7d3c | 3854 | error = xfs_bmapi(tp, ip, startoffset_fsb, |
dd9f438e NS |
3855 | allocatesize_fsb, bmapi_flag, |
3856 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 3857 | &free_list, NULL); |
1da177e4 LT |
3858 | if (error) { |
3859 | goto error0; | |
3860 | } | |
3861 | ||
3862 | /* | |
dd9f438e | 3863 | * Complete the transaction |
1da177e4 | 3864 | */ |
f7c99b6f | 3865 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3866 | if (error) { |
3867 | goto error0; | |
3868 | } | |
3869 | ||
1c72bf90 | 3870 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3871 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
3872 | if (error) { | |
3873 | break; | |
3874 | } | |
3875 | ||
3876 | allocated_fsb = imapp->br_blockcount; | |
3877 | ||
dd9f438e | 3878 | if (nimaps == 0) { |
1da177e4 LT |
3879 | error = XFS_ERROR(ENOSPC); |
3880 | break; | |
3881 | } | |
3882 | ||
3883 | startoffset_fsb += allocated_fsb; | |
3884 | allocatesize_fsb -= allocated_fsb; | |
3885 | } | |
3886 | dmapi_enospc_check: | |
eb9df39d CH |
3887 | if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 && |
3888 | DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) { | |
1da177e4 LT |
3889 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, |
3890 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
3891 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
3892 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ | |
3893 | if (error == 0) | |
3894 | goto retry; /* Maybe DMAPI app. has made space */ | |
3895 | /* else fall through with error from XFS_SEND_DATA */ | |
3896 | } | |
3897 | ||
3898 | return error; | |
3899 | ||
dd9f438e | 3900 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 3901 | xfs_bmap_cancel(&free_list); |
dd9f438e NS |
3902 | XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag); |
3903 | ||
3904 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
3905 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
3906 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3907 | goto dmapi_enospc_check; | |
3908 | } | |
3909 | ||
3910 | /* | |
3911 | * Zero file bytes between startoff and endoff inclusive. | |
3912 | * The iolock is held exclusive and no blocks are buffered. | |
3913 | */ | |
3914 | STATIC int | |
3915 | xfs_zero_remaining_bytes( | |
3916 | xfs_inode_t *ip, | |
3917 | xfs_off_t startoff, | |
3918 | xfs_off_t endoff) | |
3919 | { | |
3920 | xfs_bmbt_irec_t imap; | |
3921 | xfs_fileoff_t offset_fsb; | |
3922 | xfs_off_t lastoffset; | |
3923 | xfs_off_t offset; | |
3924 | xfs_buf_t *bp; | |
3925 | xfs_mount_t *mp = ip->i_mount; | |
3926 | int nimap; | |
3927 | int error = 0; | |
3928 | ||
3929 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, | |
71ddabb9 | 3930 | XFS_IS_REALTIME_INODE(ip) ? |
1da177e4 LT |
3931 | mp->m_rtdev_targp : mp->m_ddev_targp); |
3932 | ||
3933 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
3934 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
3935 | nimap = 1; | |
541d7d3c | 3936 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, |
3e57ecf6 | 3937 | NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
3938 | if (error || nimap < 1) |
3939 | break; | |
3940 | ASSERT(imap.br_blockcount >= 1); | |
3941 | ASSERT(imap.br_startoff == offset_fsb); | |
3942 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
3943 | if (lastoffset > endoff) | |
3944 | lastoffset = endoff; | |
3945 | if (imap.br_startblock == HOLESTARTBLOCK) | |
3946 | continue; | |
3947 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
3948 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
3949 | continue; | |
3950 | XFS_BUF_UNDONE(bp); | |
3951 | XFS_BUF_UNWRITE(bp); | |
3952 | XFS_BUF_READ(bp); | |
3953 | XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock)); | |
3954 | xfsbdstrat(mp, bp); | |
3955 | if ((error = xfs_iowait(bp))) { | |
3956 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", | |
3957 | mp, bp, XFS_BUF_ADDR(bp)); | |
3958 | break; | |
3959 | } | |
3960 | memset(XFS_BUF_PTR(bp) + | |
3961 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
3962 | 0, lastoffset - offset + 1); | |
3963 | XFS_BUF_UNDONE(bp); | |
3964 | XFS_BUF_UNREAD(bp); | |
3965 | XFS_BUF_WRITE(bp); | |
3966 | xfsbdstrat(mp, bp); | |
3967 | if ((error = xfs_iowait(bp))) { | |
3968 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", | |
3969 | mp, bp, XFS_BUF_ADDR(bp)); | |
3970 | break; | |
3971 | } | |
3972 | } | |
3973 | xfs_buf_free(bp); | |
3974 | return error; | |
3975 | } | |
3976 | ||
3977 | /* | |
3978 | * xfs_free_file_space() | |
3979 | * This routine frees disk space for the given file. | |
3980 | * | |
3981 | * This routine is only called by xfs_change_file_space | |
3982 | * for an UNRESVSP type call. | |
3983 | * | |
3984 | * RETURNS: | |
3985 | * 0 on success | |
3986 | * errno on error | |
3987 | * | |
3988 | */ | |
3989 | STATIC int | |
3990 | xfs_free_file_space( | |
3991 | xfs_inode_t *ip, | |
3992 | xfs_off_t offset, | |
3993 | xfs_off_t len, | |
3994 | int attr_flags) | |
3995 | { | |
67fcaa73 | 3996 | bhv_vnode_t *vp; |
1da177e4 LT |
3997 | int committed; |
3998 | int done; | |
3999 | xfs_off_t end_dmi_offset; | |
4000 | xfs_fileoff_t endoffset_fsb; | |
4001 | int error; | |
4002 | xfs_fsblock_t firstfsb; | |
4003 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
4004 | xfs_bmbt_irec_t imap; |
4005 | xfs_off_t ioffset; | |
4006 | xfs_extlen_t mod=0; | |
4007 | xfs_mount_t *mp; | |
4008 | int nimap; | |
4009 | uint resblks; | |
673cdf5c | 4010 | uint rounding; |
1da177e4 LT |
4011 | int rt; |
4012 | xfs_fileoff_t startoffset_fsb; | |
4013 | xfs_trans_t *tp; | |
5fcbab35 | 4014 | int need_iolock = 1; |
1da177e4 | 4015 | |
bd5a876a | 4016 | vp = XFS_ITOV(ip); |
1da177e4 LT |
4017 | mp = ip->i_mount; |
4018 | ||
cf441eeb | 4019 | xfs_itrace_entry(ip); |
bd5a876a | 4020 | |
1da177e4 LT |
4021 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
4022 | return error; | |
4023 | ||
4024 | error = 0; | |
4025 | if (len <= 0) /* if nothing being freed */ | |
4026 | return error; | |
71ddabb9 | 4027 | rt = XFS_IS_REALTIME_INODE(ip); |
1da177e4 LT |
4028 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); |
4029 | end_dmi_offset = offset + len; | |
4030 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
4031 | ||
eb9df39d CH |
4032 | if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 && |
4033 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { | |
ba87ea69 LM |
4034 | if (end_dmi_offset > ip->i_size) |
4035 | end_dmi_offset = ip->i_size; | |
bd5a876a | 4036 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp, |
1da177e4 LT |
4037 | offset, end_dmi_offset - offset, |
4038 | AT_DELAY_FLAG(attr_flags), NULL); | |
4039 | if (error) | |
014c2544 | 4040 | return error; |
1da177e4 LT |
4041 | } |
4042 | ||
5fcbab35 DR |
4043 | if (attr_flags & ATTR_NOLOCK) |
4044 | need_iolock = 0; | |
9fa8046f | 4045 | if (need_iolock) { |
1da177e4 | 4046 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
b677c210 | 4047 | vn_iowait(ip); /* wait for the completion of any pending DIOs */ |
9fa8046f | 4048 | } |
5fcbab35 | 4049 | |
e6a4b37f | 4050 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1da177e4 | 4051 | ioffset = offset & ~(rounding - 1); |
bd5a876a CH |
4052 | |
4053 | if (VN_CACHED(vp) != 0) { | |
e6a4b37f TS |
4054 | xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1); |
4055 | error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED); | |
d3cf2094 LM |
4056 | if (error) |
4057 | goto out_unlock_iolock; | |
bd5a876a CH |
4058 | } |
4059 | ||
1da177e4 LT |
4060 | /* |
4061 | * Need to zero the stuff we're not freeing, on disk. | |
4062 | * If its a realtime file & can't use unwritten extents then we | |
4063 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
4064 | * will take care of it for us. | |
4065 | */ | |
62118709 | 4066 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
1da177e4 | 4067 | nimap = 1; |
541d7d3c | 4068 | error = xfs_bmapi(NULL, ip, startoffset_fsb, |
3e57ecf6 | 4069 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
4070 | if (error) |
4071 | goto out_unlock_iolock; | |
4072 | ASSERT(nimap == 0 || nimap == 1); | |
4073 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4074 | xfs_daddr_t block; | |
4075 | ||
4076 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4077 | block = imap.br_startblock; | |
4078 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
4079 | if (mod) | |
4080 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
4081 | } | |
4082 | nimap = 1; | |
541d7d3c | 4083 | error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, |
3e57ecf6 | 4084 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
4085 | if (error) |
4086 | goto out_unlock_iolock; | |
4087 | ASSERT(nimap == 0 || nimap == 1); | |
4088 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4089 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4090 | mod++; | |
4091 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
4092 | endoffset_fsb -= mod; | |
4093 | } | |
4094 | } | |
4095 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
4096 | /* | |
4097 | * One contiguous piece to clear | |
4098 | */ | |
4099 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
4100 | else { | |
4101 | /* | |
4102 | * Some full blocks, possibly two pieces to clear | |
4103 | */ | |
4104 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
4105 | error = xfs_zero_remaining_bytes(ip, offset, | |
4106 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
4107 | if (!error && | |
4108 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
4109 | error = xfs_zero_remaining_bytes(ip, | |
4110 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
4111 | offset + len - 1); | |
4112 | } | |
4113 | ||
4114 | /* | |
4115 | * free file space until done or until there is an error | |
4116 | */ | |
4117 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
4118 | while (!error && !done) { | |
4119 | ||
4120 | /* | |
91ebecc7 DC |
4121 | * allocate and setup the transaction. Allow this |
4122 | * transaction to dip into the reserve blocks to ensure | |
4123 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
4124 | */ |
4125 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 4126 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
4127 | error = xfs_trans_reserve(tp, |
4128 | resblks, | |
4129 | XFS_WRITE_LOG_RES(mp), | |
4130 | 0, | |
4131 | XFS_TRANS_PERM_LOG_RES, | |
4132 | XFS_WRITE_LOG_COUNT); | |
4133 | ||
4134 | /* | |
4135 | * check for running out of space | |
4136 | */ | |
4137 | if (error) { | |
4138 | /* | |
4139 | * Free the transaction structure. | |
4140 | */ | |
4141 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
4142 | xfs_trans_cancel(tp, 0); | |
4143 | break; | |
4144 | } | |
4145 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4146 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, | |
dd9f438e NS |
4147 | ip->i_udquot, ip->i_gdquot, resblks, 0, |
4148 | XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
4149 | if (error) |
4150 | goto error1; | |
4151 | ||
4152 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4153 | xfs_trans_ihold(tp, ip); | |
4154 | ||
4155 | /* | |
4156 | * issue the bunmapi() call to free the blocks | |
4157 | */ | |
4158 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
541d7d3c | 4159 | error = xfs_bunmapi(tp, ip, startoffset_fsb, |
1da177e4 | 4160 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 4161 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
4162 | if (error) { |
4163 | goto error0; | |
4164 | } | |
4165 | ||
4166 | /* | |
4167 | * complete the transaction | |
4168 | */ | |
f7c99b6f | 4169 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
4170 | if (error) { |
4171 | goto error0; | |
4172 | } | |
4173 | ||
1c72bf90 | 4174 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
4175 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
4176 | } | |
4177 | ||
4178 | out_unlock_iolock: | |
4179 | if (need_iolock) | |
4180 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
4181 | return error; | |
4182 | ||
4183 | error0: | |
4184 | xfs_bmap_cancel(&free_list); | |
4185 | error1: | |
4186 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
4187 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
4188 | XFS_ILOCK_EXCL); | |
4189 | return error; | |
4190 | } | |
4191 | ||
4192 | /* | |
4193 | * xfs_change_file_space() | |
4194 | * This routine allocates or frees disk space for the given file. | |
4195 | * The user specified parameters are checked for alignment and size | |
4196 | * limitations. | |
4197 | * | |
4198 | * RETURNS: | |
4199 | * 0 on success | |
4200 | * errno on error | |
4201 | * | |
4202 | */ | |
4203 | int | |
4204 | xfs_change_file_space( | |
993386c1 | 4205 | xfs_inode_t *ip, |
1da177e4 LT |
4206 | int cmd, |
4207 | xfs_flock64_t *bf, | |
4208 | xfs_off_t offset, | |
4209 | cred_t *credp, | |
4210 | int attr_flags) | |
4211 | { | |
993386c1 | 4212 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
4213 | int clrprealloc; |
4214 | int error; | |
4215 | xfs_fsize_t fsize; | |
1da177e4 LT |
4216 | int setprealloc; |
4217 | xfs_off_t startoffset; | |
4218 | xfs_off_t llen; | |
4219 | xfs_trans_t *tp; | |
8285fb58 | 4220 | bhv_vattr_t va; |
1da177e4 | 4221 | |
cf441eeb | 4222 | xfs_itrace_entry(ip); |
1da177e4 | 4223 | |
993386c1 | 4224 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
4225 | return XFS_ERROR(EINVAL); |
4226 | ||
1da177e4 LT |
4227 | switch (bf->l_whence) { |
4228 | case 0: /*SEEK_SET*/ | |
4229 | break; | |
4230 | case 1: /*SEEK_CUR*/ | |
4231 | bf->l_start += offset; | |
4232 | break; | |
4233 | case 2: /*SEEK_END*/ | |
ba87ea69 | 4234 | bf->l_start += ip->i_size; |
1da177e4 LT |
4235 | break; |
4236 | default: | |
4237 | return XFS_ERROR(EINVAL); | |
4238 | } | |
4239 | ||
4240 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
4241 | ||
4242 | if ( (bf->l_start < 0) | |
4243 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
4244 | || (bf->l_start + llen < 0) | |
4245 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
4246 | return XFS_ERROR(EINVAL); | |
4247 | ||
4248 | bf->l_whence = 0; | |
4249 | ||
4250 | startoffset = bf->l_start; | |
ba87ea69 | 4251 | fsize = ip->i_size; |
1da177e4 LT |
4252 | |
4253 | /* | |
4254 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
4255 | * file space. | |
4256 | * These calls do NOT zero the data space allocated to the file, | |
4257 | * nor do they change the file size. | |
4258 | * | |
4259 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
4260 | * space. | |
4261 | * These calls cause the new file data to be zeroed and the file | |
4262 | * size to be changed. | |
4263 | */ | |
4264 | setprealloc = clrprealloc = 0; | |
4265 | ||
4266 | switch (cmd) { | |
4267 | case XFS_IOC_RESVSP: | |
4268 | case XFS_IOC_RESVSP64: | |
4269 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
4270 | 1, attr_flags); | |
4271 | if (error) | |
4272 | return error; | |
4273 | setprealloc = 1; | |
4274 | break; | |
4275 | ||
4276 | case XFS_IOC_UNRESVSP: | |
4277 | case XFS_IOC_UNRESVSP64: | |
4278 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
4279 | attr_flags))) | |
4280 | return error; | |
4281 | break; | |
4282 | ||
4283 | case XFS_IOC_ALLOCSP: | |
4284 | case XFS_IOC_ALLOCSP64: | |
4285 | case XFS_IOC_FREESP: | |
4286 | case XFS_IOC_FREESP64: | |
4287 | if (startoffset > fsize) { | |
4288 | error = xfs_alloc_file_space(ip, fsize, | |
4289 | startoffset - fsize, 0, attr_flags); | |
4290 | if (error) | |
4291 | break; | |
4292 | } | |
4293 | ||
4294 | va.va_mask = XFS_AT_SIZE; | |
4295 | va.va_size = startoffset; | |
4296 | ||
993386c1 | 4297 | error = xfs_setattr(ip, &va, attr_flags, credp); |
1da177e4 LT |
4298 | |
4299 | if (error) | |
4300 | return error; | |
4301 | ||
4302 | clrprealloc = 1; | |
4303 | break; | |
4304 | ||
4305 | default: | |
4306 | ASSERT(0); | |
4307 | return XFS_ERROR(EINVAL); | |
4308 | } | |
4309 | ||
4310 | /* | |
4311 | * update the inode timestamp, mode, and prealloc flag bits | |
4312 | */ | |
4313 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
4314 | ||
4315 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
4316 | 0, 0, 0))) { | |
4317 | /* ASSERT(0); */ | |
4318 | xfs_trans_cancel(tp, 0); | |
4319 | return error; | |
4320 | } | |
4321 | ||
4322 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4323 | ||
4324 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4325 | xfs_trans_ihold(tp, ip); | |
4326 | ||
4327 | if ((attr_flags & ATTR_DMI) == 0) { | |
4328 | ip->i_d.di_mode &= ~S_ISUID; | |
4329 | ||
4330 | /* | |
4331 | * Note that we don't have to worry about mandatory | |
4332 | * file locking being disabled here because we only | |
4333 | * clear the S_ISGID bit if the Group execute bit is | |
4334 | * on, but if it was on then mandatory locking wouldn't | |
4335 | * have been enabled. | |
4336 | */ | |
4337 | if (ip->i_d.di_mode & S_IXGRP) | |
4338 | ip->i_d.di_mode &= ~S_ISGID; | |
4339 | ||
4340 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
4341 | } | |
4342 | if (setprealloc) | |
4343 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
4344 | else if (clrprealloc) | |
4345 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
4346 | ||
4347 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
4348 | xfs_trans_set_sync(tp); | |
4349 | ||
1c72bf90 | 4350 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
4351 | |
4352 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4353 | ||
4354 | return error; | |
4355 | } |