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; | |
bc4ac74a | 329 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip, |
1da177e4 LT |
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) { | |
44d814ce DC |
636 | /* |
637 | * Only change the c/mtime if we are changing the size | |
638 | * or we are explicitly asked to change it. This handles | |
639 | * the semantic difference between truncate() and ftruncate() | |
640 | * as implemented in the VFS. | |
641 | */ | |
642 | if (vap->va_size != ip->i_size || (mask & XFS_AT_CTIME)) | |
643 | timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
644 | ||
ba87ea69 | 645 | if (vap->va_size > ip->i_size) { |
1da177e4 LT |
646 | xfs_igrow_finish(tp, ip, vap->va_size, |
647 | !(flags & ATTR_DMI)); | |
ba87ea69 | 648 | } else if ((vap->va_size <= ip->i_size) || |
1da177e4 LT |
649 | ((vap->va_size == 0) && ip->i_d.di_nextents)) { |
650 | /* | |
651 | * signal a sync transaction unless | |
652 | * we're truncating an already unlinked | |
653 | * file on a wsync filesystem | |
654 | */ | |
655 | code = xfs_itruncate_finish(&tp, ip, | |
656 | (xfs_fsize_t)vap->va_size, | |
657 | XFS_DATA_FORK, | |
658 | ((ip->i_d.di_nlink != 0 || | |
659 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
660 | ? 1 : 0)); | |
7d4fb40a | 661 | if (code) |
1da177e4 | 662 | goto abort_return; |
7d4fb40a NS |
663 | /* |
664 | * Truncated "down", so we're removing references | |
665 | * to old data here - if we now delay flushing for | |
666 | * a long time, we expose ourselves unduly to the | |
667 | * notorious NULL files problem. So, we mark this | |
668 | * vnode and flush it when the file is closed, and | |
669 | * do not wait the usual (long) time for writeout. | |
670 | */ | |
b3aea4ed | 671 | xfs_iflags_set(ip, XFS_ITRUNCATED); |
1da177e4 | 672 | } |
1da177e4 LT |
673 | } |
674 | ||
675 | /* | |
676 | * Change file access modes. | |
677 | */ | |
678 | if (mask & XFS_AT_MODE) { | |
679 | ip->i_d.di_mode &= S_IFMT; | |
680 | ip->i_d.di_mode |= vap->va_mode & ~S_IFMT; | |
681 | ||
682 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
683 | timeflags |= XFS_ICHGTIME_CHG; | |
684 | } | |
685 | ||
686 | /* | |
687 | * Change file ownership. Must be the owner or privileged. | |
688 | * If the system was configured with the "restricted_chown" | |
689 | * option, the owner is not permitted to give away the file, | |
690 | * and can change the group id only to a group of which he | |
691 | * or she is a member. | |
692 | */ | |
693 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
694 | /* | |
695 | * CAP_FSETID overrides the following restrictions: | |
696 | * | |
697 | * The set-user-ID and set-group-ID bits of a file will be | |
698 | * cleared upon successful return from chown() | |
699 | */ | |
700 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
701 | !capable(CAP_FSETID)) { | |
702 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
703 | } | |
704 | ||
705 | /* | |
706 | * Change the ownerships and register quota modifications | |
707 | * in the transaction. | |
708 | */ | |
709 | if (iuid != uid) { | |
710 | if (XFS_IS_UQUOTA_ON(mp)) { | |
711 | ASSERT(mask & XFS_AT_UID); | |
712 | ASSERT(udqp); | |
713 | olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
714 | &ip->i_udquot, udqp); | |
715 | } | |
716 | ip->i_d.di_uid = uid; | |
717 | } | |
718 | if (igid != gid) { | |
719 | if (XFS_IS_GQUOTA_ON(mp)) { | |
c8ad20ff | 720 | ASSERT(!XFS_IS_PQUOTA_ON(mp)); |
1da177e4 LT |
721 | ASSERT(mask & XFS_AT_GID); |
722 | ASSERT(gdqp); | |
723 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
724 | &ip->i_gdquot, gdqp); | |
725 | } | |
726 | ip->i_d.di_gid = gid; | |
727 | } | |
728 | if (iprojid != projid) { | |
c8ad20ff NS |
729 | if (XFS_IS_PQUOTA_ON(mp)) { |
730 | ASSERT(!XFS_IS_GQUOTA_ON(mp)); | |
731 | ASSERT(mask & XFS_AT_PROJID); | |
732 | ASSERT(gdqp); | |
733 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
734 | &ip->i_gdquot, gdqp); | |
735 | } | |
1da177e4 LT |
736 | ip->i_d.di_projid = projid; |
737 | /* | |
738 | * We may have to rev the inode as well as | |
739 | * the superblock version number since projids didn't | |
740 | * exist before DINODE_VERSION_2 and SB_VERSION_NLINK. | |
741 | */ | |
742 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) | |
743 | xfs_bump_ino_vers2(tp, ip); | |
744 | } | |
745 | ||
746 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
747 | timeflags |= XFS_ICHGTIME_CHG; | |
748 | } | |
749 | ||
750 | ||
751 | /* | |
752 | * Change file access or modified times. | |
753 | */ | |
754 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
755 | if (mask & XFS_AT_ATIME) { | |
756 | ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec; | |
757 | ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec; | |
758 | ip->i_update_core = 1; | |
759 | timeflags &= ~XFS_ICHGTIME_ACC; | |
760 | } | |
761 | if (mask & XFS_AT_MTIME) { | |
762 | ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec; | |
763 | ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec; | |
764 | timeflags &= ~XFS_ICHGTIME_MOD; | |
765 | timeflags |= XFS_ICHGTIME_CHG; | |
766 | } | |
767 | if (tp && (flags & ATTR_UTIME)) | |
768 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
769 | } | |
770 | ||
771 | /* | |
772 | * Change XFS-added attributes. | |
773 | */ | |
774 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
775 | if (mask & XFS_AT_EXTSIZE) { | |
776 | /* | |
777 | * Converting bytes to fs blocks. | |
778 | */ | |
779 | ip->i_d.di_extsize = vap->va_extsize >> | |
780 | mp->m_sb.sb_blocklog; | |
781 | } | |
782 | if (mask & XFS_AT_XFLAGS) { | |
783 | uint di_flags; | |
784 | ||
785 | /* can't set PREALLOC this way, just preserve it */ | |
786 | di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC); | |
787 | if (vap->va_xflags & XFS_XFLAG_IMMUTABLE) | |
788 | di_flags |= XFS_DIFLAG_IMMUTABLE; | |
789 | if (vap->va_xflags & XFS_XFLAG_APPEND) | |
790 | di_flags |= XFS_DIFLAG_APPEND; | |
791 | if (vap->va_xflags & XFS_XFLAG_SYNC) | |
792 | di_flags |= XFS_DIFLAG_SYNC; | |
793 | if (vap->va_xflags & XFS_XFLAG_NOATIME) | |
794 | di_flags |= XFS_DIFLAG_NOATIME; | |
795 | if (vap->va_xflags & XFS_XFLAG_NODUMP) | |
796 | di_flags |= XFS_DIFLAG_NODUMP; | |
365ca83d NS |
797 | if (vap->va_xflags & XFS_XFLAG_PROJINHERIT) |
798 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
799 | if (vap->va_xflags & XFS_XFLAG_NODEFRAG) |
800 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
801 | if (vap->va_xflags & XFS_XFLAG_FILESTREAM) |
802 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
1da177e4 LT |
803 | if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { |
804 | if (vap->va_xflags & XFS_XFLAG_RTINHERIT) | |
805 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
806 | if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS) | |
807 | di_flags |= XFS_DIFLAG_NOSYMLINKS; | |
dd9f438e NS |
808 | if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT) |
809 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
810 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
613d7043 | 811 | if (vap->va_xflags & XFS_XFLAG_REALTIME) |
1da177e4 | 812 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
813 | if (vap->va_xflags & XFS_XFLAG_EXTSIZE) |
814 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1da177e4 LT |
815 | } |
816 | ip->i_d.di_flags = di_flags; | |
817 | } | |
818 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
819 | timeflags |= XFS_ICHGTIME_CHG; | |
820 | } | |
821 | ||
822 | /* | |
823 | * Change file inode change time only if XFS_AT_CTIME set | |
824 | * AND we have been called by a DMI function. | |
825 | */ | |
826 | ||
827 | if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) { | |
828 | ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec; | |
829 | ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec; | |
830 | ip->i_update_core = 1; | |
831 | timeflags &= ~XFS_ICHGTIME_CHG; | |
832 | } | |
833 | ||
834 | /* | |
835 | * Send out timestamp changes that need to be set to the | |
836 | * current time. Not done when called by a DMI function. | |
837 | */ | |
838 | if (timeflags && !(flags & ATTR_DMI)) | |
839 | xfs_ichgtime(ip, timeflags); | |
840 | ||
841 | XFS_STATS_INC(xs_ig_attrchg); | |
842 | ||
843 | /* | |
844 | * If this is a synchronous mount, make sure that the | |
845 | * transaction goes to disk before returning to the user. | |
846 | * This is slightly sub-optimal in that truncates require | |
c41564b5 | 847 | * two sync transactions instead of one for wsync filesystems. |
1da177e4 LT |
848 | * One for the truncate and one for the timestamps since we |
849 | * don't want to change the timestamps unless we're sure the | |
850 | * truncate worked. Truncates are less than 1% of the laddis | |
851 | * mix so this probably isn't worth the trouble to optimize. | |
852 | */ | |
853 | code = 0; | |
854 | if (tp) { | |
855 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
856 | xfs_trans_set_sync(tp); | |
857 | ||
1c72bf90 | 858 | code = xfs_trans_commit(tp, commit_flags); |
1da177e4 LT |
859 | } |
860 | ||
861 | /* | |
862 | * If the (regular) file's mandatory locking mode changed, then | |
863 | * notify the vnode. We do this under the inode lock to prevent | |
864 | * racing calls to vop_vnode_change. | |
865 | */ | |
866 | mandlock_after = MANDLOCK(vp, ip->i_d.di_mode); | |
1da177e4 LT |
867 | |
868 | xfs_iunlock(ip, lock_flags); | |
869 | ||
870 | /* | |
871 | * Release any dquot(s) the inode had kept before chown. | |
872 | */ | |
873 | XFS_QM_DQRELE(mp, olddquot1); | |
874 | XFS_QM_DQRELE(mp, olddquot2); | |
875 | XFS_QM_DQRELE(mp, udqp); | |
876 | XFS_QM_DQRELE(mp, gdqp); | |
877 | ||
878 | if (code) { | |
879 | return code; | |
880 | } | |
881 | ||
eb9df39d | 882 | if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) && |
1da177e4 | 883 | !(flags & ATTR_DMI)) { |
bc4ac74a | 884 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL, |
1da177e4 LT |
885 | NULL, DM_RIGHT_NULL, NULL, NULL, |
886 | 0, 0, AT_DELAY_FLAG(flags)); | |
887 | } | |
888 | return 0; | |
889 | ||
890 | abort_return: | |
891 | commit_flags |= XFS_TRANS_ABORT; | |
892 | /* FALLTHROUGH */ | |
893 | error_return: | |
894 | XFS_QM_DQRELE(mp, udqp); | |
895 | XFS_QM_DQRELE(mp, gdqp); | |
896 | if (tp) { | |
897 | xfs_trans_cancel(tp, commit_flags); | |
898 | } | |
899 | if (lock_flags != 0) { | |
900 | xfs_iunlock(ip, lock_flags); | |
901 | } | |
902 | return code; | |
903 | } | |
904 | ||
7d4fb40a NS |
905 | /* |
906 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
907 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
908 | * bmapi. | |
909 | */ | |
910 | #define SYMLINK_MAPS 2 | |
911 | ||
804c83c3 CH |
912 | STATIC int |
913 | xfs_readlink_bmap( | |
914 | xfs_inode_t *ip, | |
915 | char *link) | |
916 | { | |
917 | xfs_mount_t *mp = ip->i_mount; | |
918 | int pathlen = ip->i_d.di_size; | |
919 | int nmaps = SYMLINK_MAPS; | |
920 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
921 | xfs_daddr_t d; | |
922 | int byte_cnt; | |
923 | int n; | |
924 | xfs_buf_t *bp; | |
925 | int error = 0; | |
926 | ||
927 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0, | |
928 | mval, &nmaps, NULL, NULL); | |
929 | if (error) | |
930 | goto out; | |
931 | ||
932 | for (n = 0; n < nmaps; n++) { | |
933 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
934 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
935 | ||
936 | bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0); | |
937 | error = XFS_BUF_GETERROR(bp); | |
938 | if (error) { | |
939 | xfs_ioerror_alert("xfs_readlink", | |
940 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
941 | xfs_buf_relse(bp); | |
942 | goto out; | |
943 | } | |
944 | if (pathlen < byte_cnt) | |
945 | byte_cnt = pathlen; | |
946 | pathlen -= byte_cnt; | |
947 | ||
948 | memcpy(link, XFS_BUF_PTR(bp), byte_cnt); | |
949 | xfs_buf_relse(bp); | |
950 | } | |
951 | ||
952 | link[ip->i_d.di_size] = '\0'; | |
953 | error = 0; | |
954 | ||
955 | out: | |
956 | return error; | |
957 | } | |
958 | ||
993386c1 | 959 | int |
1da177e4 | 960 | xfs_readlink( |
993386c1 | 961 | xfs_inode_t *ip, |
804c83c3 | 962 | char *link) |
1da177e4 | 963 | { |
804c83c3 | 964 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 965 | int pathlen; |
1da177e4 | 966 | int error = 0; |
1da177e4 | 967 | |
cf441eeb | 968 | xfs_itrace_entry(ip); |
1da177e4 LT |
969 | |
970 | if (XFS_FORCED_SHUTDOWN(mp)) | |
971 | return XFS_ERROR(EIO); | |
972 | ||
973 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
974 | ||
975 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
804c83c3 | 976 | ASSERT(ip->i_d.di_size <= MAXPATHLEN); |
1da177e4 | 977 | |
804c83c3 CH |
978 | pathlen = ip->i_d.di_size; |
979 | if (!pathlen) | |
980 | goto out; | |
1da177e4 LT |
981 | |
982 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
804c83c3 CH |
983 | memcpy(link, ip->i_df.if_u1.if_data, pathlen); |
984 | link[pathlen] = '\0'; | |
985 | } else { | |
986 | error = xfs_readlink_bmap(ip, link); | |
1da177e4 LT |
987 | } |
988 | ||
804c83c3 | 989 | out: |
1da177e4 | 990 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
991 | return error; |
992 | } | |
993 | ||
1da177e4 LT |
994 | /* |
995 | * xfs_fsync | |
996 | * | |
997 | * This is called to sync the inode and its data out to disk. | |
998 | * We need to hold the I/O lock while flushing the data, and | |
999 | * the inode lock while flushing the inode. The inode lock CANNOT | |
1000 | * be held while flushing the data, so acquire after we're done | |
1001 | * with that. | |
1002 | */ | |
993386c1 | 1003 | int |
1da177e4 | 1004 | xfs_fsync( |
993386c1 | 1005 | xfs_inode_t *ip, |
1da177e4 | 1006 | int flag, |
1da177e4 LT |
1007 | xfs_off_t start, |
1008 | xfs_off_t stop) | |
1009 | { | |
1da177e4 LT |
1010 | xfs_trans_t *tp; |
1011 | int error; | |
f538d4da | 1012 | int log_flushed = 0, changed = 1; |
1da177e4 | 1013 | |
cf441eeb | 1014 | xfs_itrace_entry(ip); |
1da177e4 LT |
1015 | |
1016 | ASSERT(start >= 0 && stop >= -1); | |
1017 | ||
1018 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
1019 | return XFS_ERROR(EIO); | |
1020 | ||
776a75fa LM |
1021 | if (flag & FSYNC_DATA) |
1022 | filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping); | |
1023 | ||
1da177e4 LT |
1024 | /* |
1025 | * We always need to make sure that the required inode state | |
1026 | * is safe on disk. The vnode might be clean but because | |
1027 | * of committed transactions that haven't hit the disk yet. | |
1028 | * Likewise, there could be unflushed non-transactional | |
1029 | * changes to the inode core that have to go to disk. | |
1030 | * | |
1031 | * The following code depends on one assumption: that | |
1032 | * any transaction that changes an inode logs the core | |
1033 | * because it has to change some field in the inode core | |
1034 | * (typically nextents or nblocks). That assumption | |
1035 | * implies that any transactions against an inode will | |
1036 | * catch any non-transactional updates. If inode-altering | |
1037 | * transactions exist that violate this assumption, the | |
1038 | * code breaks. Right now, it figures that if the involved | |
1039 | * update_* field is clear and the inode is unpinned, the | |
1040 | * inode is clean. Either it's been flushed or it's been | |
1041 | * committed and the commit has hit the disk unpinning the inode. | |
1042 | * (Note that xfs_inode_item_format() called at commit clears | |
1043 | * the update_* fields.) | |
1044 | */ | |
1045 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1046 | ||
1047 | /* If we are flushing data then we care about update_size | |
1048 | * being set, otherwise we care about update_core | |
1049 | */ | |
1050 | if ((flag & FSYNC_DATA) ? | |
1051 | (ip->i_update_size == 0) : | |
1052 | (ip->i_update_core == 0)) { | |
1053 | /* | |
1054 | * Timestamps/size haven't changed since last inode | |
1055 | * flush or inode transaction commit. That means | |
1056 | * either nothing got written or a transaction | |
1057 | * committed which caught the updates. If the | |
1058 | * latter happened and the transaction hasn't | |
1059 | * hit the disk yet, the inode will be still | |
1060 | * be pinned. If it is, force the log. | |
1061 | */ | |
1062 | ||
1063 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1064 | ||
1065 | if (xfs_ipincount(ip)) { | |
f538d4da | 1066 | _xfs_log_force(ip->i_mount, (xfs_lsn_t)0, |
1da177e4 LT |
1067 | XFS_LOG_FORCE | |
1068 | ((flag & FSYNC_WAIT) | |
f538d4da CH |
1069 | ? XFS_LOG_SYNC : 0), |
1070 | &log_flushed); | |
1071 | } else { | |
1072 | /* | |
1073 | * If the inode is not pinned and nothing | |
1074 | * has changed we don't need to flush the | |
1075 | * cache. | |
1076 | */ | |
1077 | changed = 0; | |
1da177e4 LT |
1078 | } |
1079 | error = 0; | |
1080 | } else { | |
1081 | /* | |
1082 | * Kick off a transaction to log the inode | |
1083 | * core to get the updates. Make it | |
1084 | * sync if FSYNC_WAIT is passed in (which | |
1085 | * is done by everybody but specfs). The | |
1086 | * sync transaction will also force the log. | |
1087 | */ | |
1088 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1089 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
1090 | if ((error = xfs_trans_reserve(tp, 0, | |
1091 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), | |
1092 | 0, 0, 0))) { | |
1093 | xfs_trans_cancel(tp, 0); | |
1094 | return error; | |
1095 | } | |
1096 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1097 | ||
1098 | /* | |
1099 | * Note - it's possible that we might have pushed | |
1100 | * ourselves out of the way during trans_reserve | |
1101 | * which would flush the inode. But there's no | |
1102 | * guarantee that the inode buffer has actually | |
1103 | * gone out yet (it's delwri). Plus the buffer | |
1104 | * could be pinned anyway if it's part of an | |
1105 | * inode in another recent transaction. So we | |
1106 | * play it safe and fire off the transaction anyway. | |
1107 | */ | |
1108 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1109 | xfs_trans_ihold(tp, ip); | |
1110 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1111 | if (flag & FSYNC_WAIT) | |
1112 | xfs_trans_set_sync(tp); | |
1c72bf90 | 1113 | error = _xfs_trans_commit(tp, 0, &log_flushed); |
1da177e4 LT |
1114 | |
1115 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1116 | } | |
f538d4da CH |
1117 | |
1118 | if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) { | |
1119 | /* | |
1120 | * If the log write didn't issue an ordered tag we need | |
1121 | * to flush the disk cache for the data device now. | |
1122 | */ | |
1123 | if (!log_flushed) | |
1124 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
1125 | ||
1126 | /* | |
1127 | * If this inode is on the RT dev we need to flush that | |
c41564b5 | 1128 | * cache as well. |
f538d4da | 1129 | */ |
71ddabb9 | 1130 | if (XFS_IS_REALTIME_INODE(ip)) |
f538d4da CH |
1131 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); |
1132 | } | |
1133 | ||
1da177e4 LT |
1134 | return error; |
1135 | } | |
1136 | ||
1137 | /* | |
92dfe8d2 DC |
1138 | * This is called by xfs_inactive to free any blocks beyond eof |
1139 | * when the link count isn't zero and by xfs_dm_punch_hole() when | |
1140 | * punching a hole to EOF. | |
1da177e4 | 1141 | */ |
92dfe8d2 DC |
1142 | int |
1143 | xfs_free_eofblocks( | |
1da177e4 | 1144 | xfs_mount_t *mp, |
92dfe8d2 DC |
1145 | xfs_inode_t *ip, |
1146 | int flags) | |
1da177e4 LT |
1147 | { |
1148 | xfs_trans_t *tp; | |
1149 | int error; | |
1150 | xfs_fileoff_t end_fsb; | |
1151 | xfs_fileoff_t last_fsb; | |
1152 | xfs_filblks_t map_len; | |
1153 | int nimaps; | |
1154 | xfs_bmbt_irec_t imap; | |
92dfe8d2 | 1155 | int use_iolock = (flags & XFS_FREE_EOF_LOCK); |
1da177e4 LT |
1156 | |
1157 | /* | |
1158 | * Figure out if there are any blocks beyond the end | |
1159 | * of the file. If not, then there is nothing to do. | |
1160 | */ | |
ba87ea69 | 1161 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size)); |
1da177e4 LT |
1162 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); |
1163 | map_len = last_fsb - end_fsb; | |
1164 | if (map_len <= 0) | |
014c2544 | 1165 | return 0; |
1da177e4 LT |
1166 | |
1167 | nimaps = 1; | |
1168 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
541d7d3c | 1169 | error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0, |
3e57ecf6 | 1170 | NULL, 0, &imap, &nimaps, NULL, NULL); |
1da177e4 LT |
1171 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1172 | ||
1173 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
1174 | (imap.br_startblock != HOLESTARTBLOCK || |
1175 | ip->i_delayed_blks)) { | |
1da177e4 LT |
1176 | /* |
1177 | * Attach the dquots to the inode up front. | |
1178 | */ | |
1179 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1180 | return error; |
1da177e4 LT |
1181 | |
1182 | /* | |
1183 | * There are blocks after the end of file. | |
1184 | * Free them up now by truncating the file to | |
1185 | * its current size. | |
1186 | */ | |
1187 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1188 | ||
1189 | /* | |
1190 | * Do the xfs_itruncate_start() call before | |
1191 | * reserving any log space because | |
1192 | * itruncate_start will call into the buffer | |
1193 | * cache and we can't | |
1194 | * do that within a transaction. | |
1195 | */ | |
92dfe8d2 DC |
1196 | if (use_iolock) |
1197 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 | 1198 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, |
ba87ea69 | 1199 | ip->i_size); |
d3cf2094 | 1200 | if (error) { |
87ae3c24 | 1201 | xfs_trans_cancel(tp, 0); |
92dfe8d2 DC |
1202 | if (use_iolock) |
1203 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
d3cf2094 LM |
1204 | return error; |
1205 | } | |
1da177e4 LT |
1206 | |
1207 | error = xfs_trans_reserve(tp, 0, | |
1208 | XFS_ITRUNCATE_LOG_RES(mp), | |
1209 | 0, XFS_TRANS_PERM_LOG_RES, | |
1210 | XFS_ITRUNCATE_LOG_COUNT); | |
1211 | if (error) { | |
1212 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1213 | xfs_trans_cancel(tp, 0); | |
1214 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1215 | return error; |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1219 | xfs_trans_ijoin(tp, ip, | |
1220 | XFS_IOLOCK_EXCL | | |
1221 | XFS_ILOCK_EXCL); | |
1222 | xfs_trans_ihold(tp, ip); | |
1223 | ||
1224 | error = xfs_itruncate_finish(&tp, ip, | |
ba87ea69 | 1225 | ip->i_size, |
1da177e4 LT |
1226 | XFS_DATA_FORK, |
1227 | 0); | |
1228 | /* | |
1229 | * If we get an error at this point we | |
1230 | * simply don't bother truncating the file. | |
1231 | */ | |
1232 | if (error) { | |
1233 | xfs_trans_cancel(tp, | |
1234 | (XFS_TRANS_RELEASE_LOG_RES | | |
1235 | XFS_TRANS_ABORT)); | |
1236 | } else { | |
1237 | error = xfs_trans_commit(tp, | |
1c72bf90 | 1238 | XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 | 1239 | } |
92dfe8d2 DC |
1240 | xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL) |
1241 | : XFS_ILOCK_EXCL)); | |
1da177e4 | 1242 | } |
014c2544 | 1243 | return error; |
1da177e4 LT |
1244 | } |
1245 | ||
1246 | /* | |
1247 | * Free a symlink that has blocks associated with it. | |
1248 | */ | |
1249 | STATIC int | |
1250 | xfs_inactive_symlink_rmt( | |
1251 | xfs_inode_t *ip, | |
1252 | xfs_trans_t **tpp) | |
1253 | { | |
1254 | xfs_buf_t *bp; | |
1255 | int committed; | |
1256 | int done; | |
1257 | int error; | |
1258 | xfs_fsblock_t first_block; | |
1259 | xfs_bmap_free_t free_list; | |
1260 | int i; | |
1261 | xfs_mount_t *mp; | |
1262 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1263 | int nmaps; | |
1264 | xfs_trans_t *ntp; | |
1265 | int size; | |
1266 | xfs_trans_t *tp; | |
1267 | ||
1268 | tp = *tpp; | |
1269 | mp = ip->i_mount; | |
1270 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
1271 | /* | |
1272 | * We're freeing a symlink that has some | |
1273 | * blocks allocated to it. Free the | |
1274 | * blocks here. We know that we've got | |
1275 | * either 1 or 2 extents and that we can | |
1276 | * free them all in one bunmapi call. | |
1277 | */ | |
1278 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
1279 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1280 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1281 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1282 | xfs_trans_cancel(tp, 0); | |
1283 | *tpp = NULL; | |
1284 | return error; | |
1285 | } | |
1286 | /* | |
1287 | * Lock the inode, fix the size, and join it to the transaction. | |
1288 | * Hold it so in the normal path, we still have it locked for | |
1289 | * the second transaction. In the error paths we need it | |
1290 | * held so the cancel won't rele it, see below. | |
1291 | */ | |
1292 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1293 | size = (int)ip->i_d.di_size; | |
1294 | ip->i_d.di_size = 0; | |
1295 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1296 | xfs_trans_ihold(tp, ip); | |
1297 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1298 | /* | |
1299 | * Find the block(s) so we can inval and unmap them. | |
1300 | */ | |
1301 | done = 0; | |
1302 | XFS_BMAP_INIT(&free_list, &first_block); | |
e8c96f8c | 1303 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
1304 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
1305 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 1306 | &free_list, NULL))) |
1da177e4 LT |
1307 | goto error0; |
1308 | /* | |
1309 | * Invalidate the block(s). | |
1310 | */ | |
1311 | for (i = 0; i < nmaps; i++) { | |
1312 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
1313 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
1314 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
1315 | xfs_trans_binval(tp, bp); | |
1316 | } | |
1317 | /* | |
1318 | * Unmap the dead block(s) to the free_list. | |
1319 | */ | |
1320 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 1321 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
1322 | goto error1; |
1323 | ASSERT(done); | |
1324 | /* | |
1325 | * Commit the first transaction. This logs the EFI and the inode. | |
1326 | */ | |
f7c99b6f | 1327 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
1328 | goto error1; |
1329 | /* | |
1330 | * The transaction must have been committed, since there were | |
1331 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
1332 | * The new tp has the extent freeing and EFDs. | |
1333 | */ | |
1334 | ASSERT(committed); | |
1335 | /* | |
1336 | * The first xact was committed, so add the inode to the new one. | |
1337 | * Mark it dirty so it will be logged and moved forward in the log as | |
1338 | * part of every commit. | |
1339 | */ | |
1340 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1341 | xfs_trans_ihold(tp, ip); | |
1342 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1343 | /* | |
1344 | * Get a new, empty transaction to return to our caller. | |
1345 | */ | |
1346 | ntp = xfs_trans_dup(tp); | |
1347 | /* | |
c41564b5 | 1348 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
1349 | * If we get an error on the commit here or on the reserve below, |
1350 | * we need to unlock the inode since the new transaction doesn't | |
1351 | * have the inode attached. | |
1352 | */ | |
1c72bf90 | 1353 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1354 | tp = ntp; |
1355 | if (error) { | |
1356 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1357 | goto error0; | |
1358 | } | |
1359 | /* | |
1360 | * Remove the memory for extent descriptions (just bookkeeping). | |
1361 | */ | |
1362 | if (ip->i_df.if_bytes) | |
1363 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
1364 | ASSERT(ip->i_df.if_bytes == 0); | |
1365 | /* | |
1366 | * Put an itruncate log reservation in the new transaction | |
1367 | * for our caller. | |
1368 | */ | |
1369 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1370 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1371 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1372 | goto error0; | |
1373 | } | |
1374 | /* | |
1375 | * Return with the inode locked but not joined to the transaction. | |
1376 | */ | |
1377 | *tpp = tp; | |
1378 | return 0; | |
1379 | ||
1380 | error1: | |
1381 | xfs_bmap_cancel(&free_list); | |
1382 | error0: | |
1383 | /* | |
1384 | * Have to come here with the inode locked and either | |
1385 | * (held and in the transaction) or (not in the transaction). | |
1386 | * If the inode isn't held then cancel would iput it, but | |
1387 | * that's wrong since this is inactive and the vnode ref | |
1388 | * count is 0 already. | |
1389 | * Cancel won't do anything to the inode if held, but it still | |
1390 | * needs to be locked until the cancel is done, if it was | |
1391 | * joined to the transaction. | |
1392 | */ | |
1393 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1394 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1395 | *tpp = NULL; | |
1396 | return error; | |
1397 | ||
1398 | } | |
1399 | ||
1400 | STATIC int | |
1401 | xfs_inactive_symlink_local( | |
1402 | xfs_inode_t *ip, | |
1403 | xfs_trans_t **tpp) | |
1404 | { | |
1405 | int error; | |
1406 | ||
1407 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
1408 | /* | |
1409 | * We're freeing a symlink which fit into | |
1410 | * the inode. Just free the memory used | |
1411 | * to hold the old symlink. | |
1412 | */ | |
1413 | error = xfs_trans_reserve(*tpp, 0, | |
1414 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
1415 | 0, XFS_TRANS_PERM_LOG_RES, | |
1416 | XFS_ITRUNCATE_LOG_COUNT); | |
1417 | ||
1418 | if (error) { | |
1419 | xfs_trans_cancel(*tpp, 0); | |
1420 | *tpp = NULL; | |
014c2544 | 1421 | return error; |
1da177e4 LT |
1422 | } |
1423 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1424 | ||
1425 | /* | |
1426 | * Zero length symlinks _can_ exist. | |
1427 | */ | |
1428 | if (ip->i_df.if_bytes > 0) { | |
1429 | xfs_idata_realloc(ip, | |
1430 | -(ip->i_df.if_bytes), | |
1431 | XFS_DATA_FORK); | |
1432 | ASSERT(ip->i_df.if_bytes == 0); | |
1433 | } | |
014c2544 | 1434 | return 0; |
1da177e4 LT |
1435 | } |
1436 | ||
1da177e4 LT |
1437 | STATIC int |
1438 | xfs_inactive_attrs( | |
1439 | xfs_inode_t *ip, | |
1440 | xfs_trans_t **tpp) | |
1441 | { | |
1442 | xfs_trans_t *tp; | |
1443 | int error; | |
1444 | xfs_mount_t *mp; | |
1445 | ||
1446 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); | |
1447 | tp = *tpp; | |
1448 | mp = ip->i_mount; | |
1449 | ASSERT(ip->i_d.di_forkoff != 0); | |
1c72bf90 | 1450 | xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1451 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1452 | ||
1453 | error = xfs_attr_inactive(ip); | |
1454 | if (error) { | |
1455 | *tpp = NULL; | |
1456 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1457 | return error; /* goto out */ |
1da177e4 LT |
1458 | } |
1459 | ||
1460 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1461 | error = xfs_trans_reserve(tp, 0, | |
1462 | XFS_IFREE_LOG_RES(mp), | |
1463 | 0, XFS_TRANS_PERM_LOG_RES, | |
1464 | XFS_INACTIVE_LOG_COUNT); | |
1465 | if (error) { | |
1466 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1467 | xfs_trans_cancel(tp, 0); | |
1468 | *tpp = NULL; | |
1469 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1470 | return error; |
1da177e4 LT |
1471 | } |
1472 | ||
1473 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1474 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1475 | xfs_trans_ihold(tp, ip); | |
1476 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1477 | ||
1478 | ASSERT(ip->i_d.di_anextents == 0); | |
1479 | ||
1480 | *tpp = tp; | |
014c2544 | 1481 | return 0; |
1da177e4 LT |
1482 | } |
1483 | ||
993386c1 | 1484 | int |
1da177e4 | 1485 | xfs_release( |
993386c1 | 1486 | xfs_inode_t *ip) |
1da177e4 | 1487 | { |
993386c1 CH |
1488 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1489 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
1490 | int error; |
1491 | ||
7d4fb40a | 1492 | if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) |
1da177e4 | 1493 | return 0; |
1da177e4 LT |
1494 | |
1495 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1496 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1497 | return 0; |
1498 | ||
2a82b8be | 1499 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
b3aea4ed CH |
1500 | int truncated; |
1501 | ||
2a82b8be DC |
1502 | /* |
1503 | * If we are using filestreams, and we have an unlinked | |
1504 | * file that we are processing the last close on, then nothing | |
1505 | * will be able to reopen and write to this file. Purge this | |
1506 | * inode from the filestreams cache so that it doesn't delay | |
1507 | * teardown of the inode. | |
1508 | */ | |
1509 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1510 | xfs_filestream_deassociate(ip); | |
1511 | ||
fbf3ce8d CH |
1512 | /* |
1513 | * If we previously truncated this file and removed old data | |
1514 | * in the process, we want to initiate "early" writeout on | |
1515 | * the last close. This is an attempt to combat the notorious | |
1516 | * NULL files problem which is particularly noticable from a | |
1517 | * truncate down, buffered (re-)write (delalloc), followed by | |
1518 | * a crash. What we are effectively doing here is | |
1519 | * significantly reducing the time window where we'd otherwise | |
1520 | * be exposed to that problem. | |
1521 | */ | |
09262b43 | 1522 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
b3aea4ed | 1523 | if (truncated && VN_DIRTY(vp) && ip->i_delayed_blks > 0) |
739bfb2a | 1524 | xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE); |
fbf3ce8d CH |
1525 | } |
1526 | ||
1da177e4 LT |
1527 | if (ip->i_d.di_nlink != 0) { |
1528 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1529 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1530 | ip->i_delayed_blks > 0)) && |
1da177e4 | 1531 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
1532 | (!(ip->i_d.di_flags & |
1533 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
92dfe8d2 DC |
1534 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1535 | if (error) | |
014c2544 | 1536 | return error; |
1da177e4 LT |
1537 | } |
1538 | } | |
1539 | ||
1540 | return 0; | |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * xfs_inactive | |
1545 | * | |
1546 | * This is called when the vnode reference count for the vnode | |
1547 | * goes to zero. If the file has been unlinked, then it must | |
1548 | * now be truncated. Also, we clear all of the read-ahead state | |
1549 | * kept for the inode here since the file is now closed. | |
1550 | */ | |
993386c1 | 1551 | int |
1da177e4 | 1552 | xfs_inactive( |
993386c1 | 1553 | xfs_inode_t *ip) |
1da177e4 | 1554 | { |
993386c1 | 1555 | bhv_vnode_t *vp = XFS_ITOV(ip); |
67fcaa73 | 1556 | xfs_bmap_free_t free_list; |
1da177e4 LT |
1557 | xfs_fsblock_t first_block; |
1558 | int committed; | |
1559 | xfs_trans_t *tp; | |
1560 | xfs_mount_t *mp; | |
1561 | int error; | |
1562 | int truncate; | |
1563 | ||
cf441eeb | 1564 | xfs_itrace_entry(ip); |
1da177e4 | 1565 | |
1da177e4 LT |
1566 | /* |
1567 | * If the inode is already free, then there can be nothing | |
1568 | * to clean up here. | |
1569 | */ | |
1570 | if (ip->i_d.di_mode == 0 || VN_BAD(vp)) { | |
1571 | ASSERT(ip->i_df.if_real_bytes == 0); | |
1572 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1573 | return VN_INACTIVE_CACHE; | |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * Only do a truncate if it's a regular file with | |
1578 | * some actual space in it. It's OK to look at the | |
1579 | * inode's fields without the lock because we're the | |
1580 | * only one with a reference to the inode. | |
1581 | */ | |
1582 | truncate = ((ip->i_d.di_nlink == 0) && | |
ba87ea69 LM |
1583 | ((ip->i_d.di_size != 0) || (ip->i_size != 0) || |
1584 | (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1585 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1586 | ||
1587 | mp = ip->i_mount; | |
1588 | ||
bc4ac74a CH |
1589 | if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY)) |
1590 | XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL); | |
1da177e4 LT |
1591 | |
1592 | error = 0; | |
1593 | ||
1594 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
bd186aa9 | 1595 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
1596 | goto out; |
1597 | ||
1598 | if (ip->i_d.di_nlink != 0) { | |
1599 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
ba87ea69 | 1600 | ((ip->i_size > 0) || (VN_CACHED(vp) > 0 || |
68bdb6ea | 1601 | ip->i_delayed_blks > 0)) && |
dd9f438e NS |
1602 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1603 | (!(ip->i_d.di_flags & | |
1604 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1605 | (ip->i_delayed_blks != 0)))) { | |
92dfe8d2 DC |
1606 | error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK); |
1607 | if (error) | |
014c2544 | 1608 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1609 | } |
1610 | goto out; | |
1611 | } | |
1612 | ||
1613 | ASSERT(ip->i_d.di_nlink == 0); | |
1614 | ||
1615 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1616 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1617 | |
1618 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1619 | if (truncate) { | |
1620 | /* | |
1621 | * Do the xfs_itruncate_start() call before | |
1622 | * reserving any log space because itruncate_start | |
1623 | * will call into the buffer cache and we can't | |
1624 | * do that within a transaction. | |
1625 | */ | |
1626 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1627 | ||
d3cf2094 LM |
1628 | error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); |
1629 | if (error) { | |
87ae3c24 | 1630 | xfs_trans_cancel(tp, 0); |
d3cf2094 LM |
1631 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
1632 | return VN_INACTIVE_CACHE; | |
1633 | } | |
1da177e4 LT |
1634 | |
1635 | error = xfs_trans_reserve(tp, 0, | |
1636 | XFS_ITRUNCATE_LOG_RES(mp), | |
1637 | 0, XFS_TRANS_PERM_LOG_RES, | |
1638 | XFS_ITRUNCATE_LOG_COUNT); | |
1639 | if (error) { | |
1640 | /* Don't call itruncate_cleanup */ | |
1641 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1642 | xfs_trans_cancel(tp, 0); | |
1643 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1644 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1645 | } |
1646 | ||
1647 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1648 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1649 | xfs_trans_ihold(tp, ip); | |
1650 | ||
1651 | /* | |
1652 | * normally, we have to run xfs_itruncate_finish sync. | |
1653 | * But if filesystem is wsync and we're in the inactive | |
1654 | * path, then we know that nlink == 0, and that the | |
1655 | * xaction that made nlink == 0 is permanently committed | |
1656 | * since xfs_remove runs as a synchronous transaction. | |
1657 | */ | |
1658 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1659 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1660 | ||
1661 | if (error) { | |
1662 | xfs_trans_cancel(tp, | |
1663 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1664 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1665 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1666 | } |
1667 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1668 | ||
1669 | /* | |
1670 | * If we get an error while cleaning up a | |
1671 | * symlink we bail out. | |
1672 | */ | |
1673 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1674 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1675 | xfs_inactive_symlink_local(ip, &tp); | |
1676 | ||
1677 | if (error) { | |
1678 | ASSERT(tp == NULL); | |
014c2544 | 1679 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1680 | } |
1681 | ||
1682 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1683 | xfs_trans_ihold(tp, ip); | |
1684 | } else { | |
1685 | error = xfs_trans_reserve(tp, 0, | |
1686 | XFS_IFREE_LOG_RES(mp), | |
1687 | 0, XFS_TRANS_PERM_LOG_RES, | |
1688 | XFS_INACTIVE_LOG_COUNT); | |
1689 | if (error) { | |
1690 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1691 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1692 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1693 | } |
1694 | ||
1695 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1696 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1697 | xfs_trans_ihold(tp, ip); | |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * If there are attributes associated with the file | |
1702 | * then blow them away now. The code calls a routine | |
1703 | * that recursively deconstructs the attribute fork. | |
1704 | * We need to just commit the current transaction | |
1705 | * because we can't use it for xfs_attr_inactive(). | |
1706 | */ | |
1707 | if (ip->i_d.di_anextents > 0) { | |
1708 | error = xfs_inactive_attrs(ip, &tp); | |
1709 | /* | |
1710 | * If we got an error, the transaction is already | |
1711 | * cancelled, and the inode is unlocked. Just get out. | |
1712 | */ | |
1713 | if (error) | |
014c2544 | 1714 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1715 | } else if (ip->i_afp) { |
1716 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1717 | } | |
1718 | ||
1719 | /* | |
1720 | * Free the inode. | |
1721 | */ | |
1722 | XFS_BMAP_INIT(&free_list, &first_block); | |
1723 | error = xfs_ifree(tp, ip, &free_list); | |
1724 | if (error) { | |
1725 | /* | |
1726 | * If we fail to free the inode, shut down. The cancel | |
1727 | * might do that, we need to make sure. Otherwise the | |
1728 | * inode might be lost for a long time or forever. | |
1729 | */ | |
1730 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1731 | cmn_err(CE_NOTE, | |
1732 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1733 | error, mp->m_fsname); | |
7d04a335 | 1734 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1735 | } |
1736 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1737 | } else { | |
1738 | /* | |
1739 | * Credit the quota account(s). The inode is gone. | |
1740 | */ | |
1741 | XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1742 | ||
1743 | /* | |
1744 | * Just ignore errors at this point. There is | |
1745 | * nothing we can do except to try to keep going. | |
1746 | */ | |
f7c99b6f | 1747 | (void) xfs_bmap_finish(&tp, &free_list, &committed); |
1c72bf90 | 1748 | (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1749 | } |
1750 | /* | |
1751 | * Release the dquots held by inode, if any. | |
1752 | */ | |
1753 | XFS_QM_DQDETACH(mp, ip); | |
1754 | ||
1755 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1756 | ||
1757 | out: | |
1758 | return VN_INACTIVE_CACHE; | |
1759 | } | |
1760 | ||
1761 | ||
993386c1 | 1762 | int |
1da177e4 | 1763 | xfs_lookup( |
993386c1 | 1764 | xfs_inode_t *dp, |
8285fb58 | 1765 | bhv_vname_t *dentry, |
993386c1 | 1766 | bhv_vnode_t **vpp) |
1da177e4 | 1767 | { |
993386c1 | 1768 | xfs_inode_t *ip; |
1da177e4 LT |
1769 | xfs_ino_t e_inum; |
1770 | int error; | |
1771 | uint lock_mode; | |
1da177e4 | 1772 | |
cf441eeb | 1773 | xfs_itrace_entry(dp); |
1da177e4 LT |
1774 | |
1775 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1776 | return XFS_ERROR(EIO); | |
1777 | ||
1778 | lock_mode = xfs_ilock_map_shared(dp); | |
993386c1 | 1779 | error = xfs_dir_lookup_int(dp, lock_mode, dentry, &e_inum, &ip); |
1da177e4 LT |
1780 | if (!error) { |
1781 | *vpp = XFS_ITOV(ip); | |
cf441eeb | 1782 | xfs_itrace_ref(ip); |
1da177e4 LT |
1783 | } |
1784 | xfs_iunlock_map_shared(dp, lock_mode); | |
1785 | return error; | |
1786 | } | |
1787 | ||
993386c1 | 1788 | int |
1da177e4 | 1789 | xfs_create( |
993386c1 | 1790 | xfs_inode_t *dp, |
8285fb58 | 1791 | bhv_vname_t *dentry, |
3e5daf05 CH |
1792 | mode_t mode, |
1793 | xfs_dev_t rdev, | |
979ebab1 | 1794 | xfs_inode_t **ipp, |
1da177e4 LT |
1795 | cred_t *credp) |
1796 | { | |
1797 | char *name = VNAME(dentry); | |
993386c1 | 1798 | xfs_mount_t *mp = dp->i_mount; |
993386c1 | 1799 | xfs_inode_t *ip; |
1da177e4 | 1800 | xfs_trans_t *tp; |
1da177e4 LT |
1801 | int error; |
1802 | xfs_bmap_free_t free_list; | |
1803 | xfs_fsblock_t first_block; | |
993386c1 | 1804 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
1805 | int dm_event_sent = 0; |
1806 | uint cancel_flags; | |
1807 | int committed; | |
1808 | xfs_prid_t prid; | |
1809 | struct xfs_dquot *udqp, *gdqp; | |
1810 | uint resblks; | |
1da177e4 LT |
1811 | int namelen; |
1812 | ||
979ebab1 | 1813 | ASSERT(!*ipp); |
cf441eeb | 1814 | xfs_itrace_entry(dp); |
1da177e4 | 1815 | |
1da177e4 LT |
1816 | namelen = VNAMELEN(dentry); |
1817 | ||
eb9df39d | 1818 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 | 1819 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
bc4ac74a | 1820 | dp, DM_RIGHT_NULL, NULL, |
1da177e4 | 1821 | DM_RIGHT_NULL, name, NULL, |
3e5daf05 | 1822 | mode, 0, 0); |
1da177e4 LT |
1823 | |
1824 | if (error) | |
1825 | return error; | |
1826 | dm_event_sent = 1; | |
1827 | } | |
1828 | ||
1829 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1830 | return XFS_ERROR(EIO); | |
1831 | ||
1832 | /* Return through std_return after this point. */ | |
1833 | ||
1834 | udqp = gdqp = NULL; | |
365ca83d NS |
1835 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1836 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
1837 | else |
1838 | prid = (xfs_prid_t)dfltprid; | |
1839 | ||
1840 | /* | |
1841 | * Make sure that we have allocated dquot(s) on disk. | |
1842 | */ | |
1843 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 1844 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
1845 | XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1846 | if (error) | |
1847 | goto std_return; | |
1848 | ||
1849 | ip = NULL; | |
1da177e4 LT |
1850 | |
1851 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1852 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1853 | resblks = XFS_CREATE_SPACE_RES(mp, namelen); | |
1854 | /* | |
1855 | * Initially assume that the file does not exist and | |
1856 | * reserve the resources for that case. If that is not | |
1857 | * the case we'll drop the one we have and get a more | |
1858 | * appropriate transaction later. | |
1859 | */ | |
1860 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1861 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1862 | if (error == ENOSPC) { | |
1863 | resblks = 0; | |
1864 | error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0, | |
1865 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1866 | } | |
1867 | if (error) { | |
1868 | cancel_flags = 0; | |
1da177e4 LT |
1869 | goto error_return; |
1870 | } | |
1871 | ||
f7c66ce3 | 1872 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 1873 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
1874 | |
1875 | XFS_BMAP_INIT(&free_list, &first_block); | |
1876 | ||
1877 | ASSERT(ip == NULL); | |
1878 | ||
1879 | /* | |
1880 | * Reserve disk quota and the inode. | |
1881 | */ | |
1882 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
1883 | if (error) | |
1884 | goto error_return; | |
1885 | ||
f6c2d1fa | 1886 | if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen))) |
1da177e4 | 1887 | goto error_return; |
3e5daf05 | 1888 | error = xfs_dir_ialloc(&tp, dp, mode, 1, |
1da177e4 LT |
1889 | rdev, credp, prid, resblks > 0, |
1890 | &ip, &committed); | |
1891 | if (error) { | |
1892 | if (error == ENOSPC) | |
1893 | goto error_return; | |
1894 | goto abort_return; | |
1895 | } | |
cf441eeb | 1896 | xfs_itrace_ref(ip); |
1da177e4 LT |
1897 | |
1898 | /* | |
1899 | * At this point, we've gotten a newly allocated inode. | |
1900 | * It is locked (and joined to the transaction). | |
1901 | */ | |
1902 | ||
1903 | ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE)); | |
1904 | ||
1905 | /* | |
993386c1 CH |
1906 | * Now we join the directory inode to the transaction. We do not do it |
1907 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1908 | * (and release all the locks). An error from here on will result in | |
1909 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1910 | * error path. | |
1da177e4 | 1911 | */ |
979ebab1 | 1912 | IHOLD(dp); |
1da177e4 | 1913 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 1914 | unlock_dp_on_error = B_FALSE; |
1da177e4 | 1915 | |
f6c2d1fa NS |
1916 | error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino, |
1917 | &first_block, &free_list, resblks ? | |
1918 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1919 | if (error) { |
1920 | ASSERT(error != ENOSPC); | |
1921 | goto abort_return; | |
1922 | } | |
1923 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1924 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1925 | ||
1926 | /* | |
1927 | * If this is a synchronous mount, make sure that the | |
1928 | * create transaction goes to disk before returning to | |
1929 | * the user. | |
1930 | */ | |
1931 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1932 | xfs_trans_set_sync(tp); | |
1933 | } | |
1934 | ||
1935 | dp->i_gen++; | |
1936 | ||
1937 | /* | |
1938 | * Attach the dquot(s) to the inodes and modify them incore. | |
1939 | * These ids of the inode couldn't have changed since the new | |
1940 | * inode has been locked ever since it was created. | |
1941 | */ | |
1942 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
1943 | ||
1944 | /* | |
1945 | * xfs_trans_commit normally decrements the vnode ref count | |
1946 | * when it unlocks the inode. Since we want to return the | |
1947 | * vnode to the caller, we bump the vnode ref count now. | |
1948 | */ | |
1949 | IHOLD(ip); | |
1da177e4 | 1950 | |
f7c99b6f | 1951 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1952 | if (error) { |
1953 | xfs_bmap_cancel(&free_list); | |
1954 | goto abort_rele; | |
1955 | } | |
1956 | ||
1c72bf90 | 1957 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
1958 | if (error) { |
1959 | IRELE(ip); | |
1960 | tp = NULL; | |
1961 | goto error_return; | |
1962 | } | |
1963 | ||
1964 | XFS_QM_DQRELE(mp, udqp); | |
1965 | XFS_QM_DQRELE(mp, gdqp); | |
1966 | ||
979ebab1 | 1967 | *ipp = ip; |
1da177e4 LT |
1968 | |
1969 | /* Fallthrough to std_return with error = 0 */ | |
1970 | ||
1971 | std_return: | |
979ebab1 | 1972 | if ((*ipp || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 1973 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 | 1974 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
bc4ac74a | 1975 | dp, DM_RIGHT_NULL, |
979ebab1 | 1976 | *ipp ? ip : NULL, |
1da177e4 | 1977 | DM_RIGHT_NULL, name, NULL, |
3e5daf05 | 1978 | mode, error, 0); |
1da177e4 LT |
1979 | } |
1980 | return error; | |
1981 | ||
1982 | abort_return: | |
1983 | cancel_flags |= XFS_TRANS_ABORT; | |
1984 | /* FALLTHROUGH */ | |
1da177e4 | 1985 | |
014c2544 | 1986 | error_return: |
1da177e4 LT |
1987 | if (tp != NULL) |
1988 | xfs_trans_cancel(tp, cancel_flags); | |
1989 | ||
1da177e4 LT |
1990 | XFS_QM_DQRELE(mp, udqp); |
1991 | XFS_QM_DQRELE(mp, gdqp); | |
1992 | ||
993386c1 CH |
1993 | if (unlock_dp_on_error) |
1994 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1995 | ||
1da177e4 LT |
1996 | goto std_return; |
1997 | ||
1998 | abort_rele: | |
1999 | /* | |
2000 | * Wait until after the current transaction is aborted to | |
2001 | * release the inode. This prevents recursive transactions | |
2002 | * and deadlocks from xfs_inactive. | |
2003 | */ | |
2004 | cancel_flags |= XFS_TRANS_ABORT; | |
2005 | xfs_trans_cancel(tp, cancel_flags); | |
2006 | IRELE(ip); | |
2007 | ||
2008 | XFS_QM_DQRELE(mp, udqp); | |
2009 | XFS_QM_DQRELE(mp, gdqp); | |
2010 | ||
2011 | goto std_return; | |
2012 | } | |
2013 | ||
2014 | #ifdef DEBUG | |
2015 | /* | |
2016 | * Some counters to see if (and how often) we are hitting some deadlock | |
2017 | * prevention code paths. | |
2018 | */ | |
2019 | ||
2020 | int xfs_rm_locks; | |
2021 | int xfs_rm_lock_delays; | |
2022 | int xfs_rm_attempts; | |
2023 | #endif | |
2024 | ||
2025 | /* | |
2026 | * The following routine will lock the inodes associated with the | |
2027 | * directory and the named entry in the directory. The locks are | |
2028 | * acquired in increasing inode number. | |
2029 | * | |
2030 | * If the entry is "..", then only the directory is locked. The | |
2031 | * vnode ref count will still include that from the .. entry in | |
2032 | * this case. | |
2033 | * | |
2034 | * There is a deadlock we need to worry about. If the locked directory is | |
2035 | * in the AIL, it might be blocking up the log. The next inode we lock | |
2036 | * could be already locked by another thread waiting for log space (e.g | |
2037 | * a permanent log reservation with a long running transaction (see | |
2038 | * xfs_itruncate_finish)). To solve this, we must check if the directory | |
2039 | * is in the ail and use lock_nowait. If we can't lock, we need to | |
2040 | * drop the inode lock on the directory and try again. xfs_iunlock will | |
2041 | * potentially push the tail if we were holding up the log. | |
2042 | */ | |
2043 | STATIC int | |
2044 | xfs_lock_dir_and_entry( | |
2045 | xfs_inode_t *dp, | |
1da177e4 LT |
2046 | xfs_inode_t *ip) /* inode of entry 'name' */ |
2047 | { | |
2048 | int attempts; | |
2049 | xfs_ino_t e_inum; | |
2050 | xfs_inode_t *ips[2]; | |
2051 | xfs_log_item_t *lp; | |
2052 | ||
2053 | #ifdef DEBUG | |
2054 | xfs_rm_locks++; | |
2055 | #endif | |
2056 | attempts = 0; | |
2057 | ||
2058 | again: | |
f7c66ce3 | 2059 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
1da177e4 LT |
2060 | |
2061 | e_inum = ip->i_ino; | |
2062 | ||
cf441eeb | 2063 | xfs_itrace_ref(ip); |
1da177e4 LT |
2064 | |
2065 | /* | |
2066 | * We want to lock in increasing inum. Since we've already | |
2067 | * acquired the lock on the directory, we may need to release | |
2068 | * if if the inum of the entry turns out to be less. | |
2069 | */ | |
2070 | if (e_inum > dp->i_ino) { | |
2071 | /* | |
2072 | * We are already in the right order, so just | |
2073 | * lock on the inode of the entry. | |
2074 | * We need to use nowait if dp is in the AIL. | |
2075 | */ | |
2076 | ||
2077 | lp = (xfs_log_item_t *)dp->i_itemp; | |
2078 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2079 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2080 | attempts++; | |
2081 | #ifdef DEBUG | |
2082 | xfs_rm_attempts++; | |
2083 | #endif | |
2084 | ||
2085 | /* | |
2086 | * Unlock dp and try again. | |
2087 | * xfs_iunlock will try to push the tail | |
2088 | * if the inode is in the AIL. | |
2089 | */ | |
2090 | ||
2091 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2092 | ||
2093 | if ((attempts % 5) == 0) { | |
2094 | delay(1); /* Don't just spin the CPU */ | |
2095 | #ifdef DEBUG | |
2096 | xfs_rm_lock_delays++; | |
2097 | #endif | |
2098 | } | |
2099 | goto again; | |
2100 | } | |
2101 | } else { | |
2102 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2103 | } | |
2104 | } else if (e_inum < dp->i_ino) { | |
2105 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2106 | ||
2107 | ips[0] = ip; | |
2108 | ips[1] = dp; | |
2109 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2110 | } | |
2111 | /* else e_inum == dp->i_ino */ | |
2112 | /* This can happen if we're asked to lock /x/.. | |
2113 | * the entry is "..", which is also the parent directory. | |
2114 | */ | |
2115 | ||
2116 | return 0; | |
2117 | } | |
2118 | ||
2119 | #ifdef DEBUG | |
2120 | int xfs_locked_n; | |
2121 | int xfs_small_retries; | |
2122 | int xfs_middle_retries; | |
2123 | int xfs_lots_retries; | |
2124 | int xfs_lock_delays; | |
2125 | #endif | |
2126 | ||
f7c66ce3 LM |
2127 | /* |
2128 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
2129 | * a different value | |
2130 | */ | |
2131 | static inline int | |
2132 | xfs_lock_inumorder(int lock_mode, int subclass) | |
2133 | { | |
2134 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
0f1145cc | 2135 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
f7c66ce3 | 2136 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
0f1145cc | 2137 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
f7c66ce3 LM |
2138 | |
2139 | return lock_mode; | |
2140 | } | |
2141 | ||
1da177e4 LT |
2142 | /* |
2143 | * The following routine will lock n inodes in exclusive mode. | |
2144 | * We assume the caller calls us with the inodes in i_ino order. | |
2145 | * | |
2146 | * We need to detect deadlock where an inode that we lock | |
2147 | * is in the AIL and we start waiting for another inode that is locked | |
2148 | * by a thread in a long running transaction (such as truncate). This can | |
2149 | * result in deadlock since the long running trans might need to wait | |
2150 | * for the inode we just locked in order to push the tail and free space | |
2151 | * in the log. | |
2152 | */ | |
2153 | void | |
2154 | xfs_lock_inodes( | |
2155 | xfs_inode_t **ips, | |
2156 | int inodes, | |
2157 | int first_locked, | |
2158 | uint lock_mode) | |
2159 | { | |
2160 | int attempts = 0, i, j, try_lock; | |
2161 | xfs_log_item_t *lp; | |
2162 | ||
2163 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
2164 | ||
2165 | if (first_locked) { | |
2166 | try_lock = 1; | |
2167 | i = 1; | |
2168 | } else { | |
2169 | try_lock = 0; | |
2170 | i = 0; | |
2171 | } | |
2172 | ||
2173 | again: | |
2174 | for (; i < inodes; i++) { | |
2175 | ASSERT(ips[i]); | |
2176 | ||
2177 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
2178 | continue; | |
2179 | ||
2180 | /* | |
2181 | * If try_lock is not set yet, make sure all locked inodes | |
2182 | * are not in the AIL. | |
2183 | * If any are, set try_lock to be used later. | |
2184 | */ | |
2185 | ||
2186 | if (!try_lock) { | |
2187 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
2188 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
2189 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2190 | try_lock++; | |
2191 | } | |
2192 | } | |
2193 | } | |
2194 | ||
2195 | /* | |
2196 | * If any of the previous locks we have locked is in the AIL, | |
2197 | * we must TRY to get the second and subsequent locks. If | |
2198 | * we can't get any, we must release all we have | |
2199 | * and try again. | |
2200 | */ | |
2201 | ||
2202 | if (try_lock) { | |
2203 | /* try_lock must be 0 if i is 0. */ | |
2204 | /* | |
2205 | * try_lock means we have an inode locked | |
2206 | * that is in the AIL. | |
2207 | */ | |
2208 | ASSERT(i != 0); | |
f7c66ce3 | 2209 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
1da177e4 LT |
2210 | attempts++; |
2211 | ||
2212 | /* | |
2213 | * Unlock all previous guys and try again. | |
2214 | * xfs_iunlock will try to push the tail | |
2215 | * if the inode is in the AIL. | |
2216 | */ | |
2217 | ||
2218 | for(j = i - 1; j >= 0; j--) { | |
2219 | ||
2220 | /* | |
2221 | * Check to see if we've already | |
2222 | * unlocked this one. | |
2223 | * Not the first one going back, | |
2224 | * and the inode ptr is the same. | |
2225 | */ | |
2226 | if ((j != (i - 1)) && ips[j] == | |
2227 | ips[j+1]) | |
2228 | continue; | |
2229 | ||
2230 | xfs_iunlock(ips[j], lock_mode); | |
2231 | } | |
2232 | ||
2233 | if ((attempts % 5) == 0) { | |
2234 | delay(1); /* Don't just spin the CPU */ | |
2235 | #ifdef DEBUG | |
2236 | xfs_lock_delays++; | |
2237 | #endif | |
2238 | } | |
2239 | i = 0; | |
2240 | try_lock = 0; | |
2241 | goto again; | |
2242 | } | |
2243 | } else { | |
f7c66ce3 | 2244 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
1da177e4 LT |
2245 | } |
2246 | } | |
2247 | ||
2248 | #ifdef DEBUG | |
2249 | if (attempts) { | |
2250 | if (attempts < 5) xfs_small_retries++; | |
2251 | else if (attempts < 100) xfs_middle_retries++; | |
2252 | else xfs_lots_retries++; | |
2253 | } else { | |
2254 | xfs_locked_n++; | |
2255 | } | |
2256 | #endif | |
2257 | } | |
2258 | ||
2259 | #ifdef DEBUG | |
2260 | #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);} | |
2261 | int remove_which_error_return = 0; | |
2262 | #else /* ! DEBUG */ | |
2263 | #define REMOVE_DEBUG_TRACE(x) | |
2264 | #endif /* ! DEBUG */ | |
2265 | ||
993386c1 | 2266 | int |
1da177e4 | 2267 | xfs_remove( |
993386c1 CH |
2268 | xfs_inode_t *dp, |
2269 | bhv_vname_t *dentry) | |
1da177e4 | 2270 | { |
1da177e4 | 2271 | char *name = VNAME(dentry); |
993386c1 | 2272 | xfs_mount_t *mp = dp->i_mount; |
43973964 CH |
2273 | xfs_inode_t *ip = VNAME_TO_INODE(dentry); |
2274 | int namelen = VNAMELEN(dentry); | |
1da177e4 | 2275 | xfs_trans_t *tp = NULL; |
1da177e4 LT |
2276 | int error = 0; |
2277 | xfs_bmap_free_t free_list; | |
2278 | xfs_fsblock_t first_block; | |
2279 | int cancel_flags; | |
2280 | int committed; | |
1da177e4 LT |
2281 | int link_zero; |
2282 | uint resblks; | |
1da177e4 | 2283 | |
cf441eeb | 2284 | xfs_itrace_entry(dp); |
1da177e4 | 2285 | |
1da177e4 LT |
2286 | if (XFS_FORCED_SHUTDOWN(mp)) |
2287 | return XFS_ERROR(EIO); | |
2288 | ||
eb9df39d | 2289 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
bc4ac74a | 2290 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp, |
1da177e4 | 2291 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, |
43973964 | 2292 | name, NULL, ip->i_d.di_mode, 0, 0); |
1da177e4 LT |
2293 | if (error) |
2294 | return error; | |
2295 | } | |
2296 | ||
1da177e4 LT |
2297 | /* |
2298 | * We need to get a reference to ip before we get our log | |
2299 | * reservation. The reason for this is that we cannot call | |
2300 | * xfs_iget for an inode for which we do not have a reference | |
2301 | * once we've acquired a log reservation. This is because the | |
2302 | * inode we are trying to get might be in xfs_inactive going | |
2303 | * for a log reservation. Since we'll have to wait for the | |
2304 | * inactive code to complete before returning from xfs_iget, | |
2305 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2306 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2307 | * to the inode before getting our log reservation. |
2308 | */ | |
43973964 | 2309 | IHOLD(ip); |
1da177e4 | 2310 | |
cf441eeb LM |
2311 | xfs_itrace_entry(ip); |
2312 | xfs_itrace_ref(ip); | |
1da177e4 LT |
2313 | |
2314 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2315 | if (!error && dp != ip) | |
2316 | error = XFS_QM_DQATTACH(mp, ip, 0); | |
2317 | if (error) { | |
2318 | REMOVE_DEBUG_TRACE(__LINE__); | |
2319 | IRELE(ip); | |
2320 | goto std_return; | |
2321 | } | |
2322 | ||
2323 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2324 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2325 | /* | |
2326 | * We try to get the real space reservation first, | |
2327 | * allowing for directory btree deletion(s) implying | |
2328 | * possible bmap insert(s). If we can't get the space | |
2329 | * reservation then we use 0 instead, and avoid the bmap | |
2330 | * btree insert(s) in the directory code by, if the bmap | |
2331 | * insert tries to happen, instead trimming the LAST | |
2332 | * block from the directory. | |
2333 | */ | |
2334 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2335 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2336 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2337 | if (error == ENOSPC) { | |
2338 | resblks = 0; | |
2339 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2340 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2341 | } | |
2342 | if (error) { | |
2343 | ASSERT(error != ENOSPC); | |
2344 | REMOVE_DEBUG_TRACE(__LINE__); | |
2345 | xfs_trans_cancel(tp, 0); | |
2346 | IRELE(ip); | |
2347 | return error; | |
2348 | } | |
2349 | ||
e9ed9d22 | 2350 | error = xfs_lock_dir_and_entry(dp, ip); |
1da177e4 LT |
2351 | if (error) { |
2352 | REMOVE_DEBUG_TRACE(__LINE__); | |
2353 | xfs_trans_cancel(tp, cancel_flags); | |
2354 | IRELE(ip); | |
2355 | goto std_return; | |
2356 | } | |
2357 | ||
2358 | /* | |
2359 | * At this point, we've gotten both the directory and the entry | |
2360 | * inodes locked. | |
2361 | */ | |
2362 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2363 | if (dp != ip) { | |
2364 | /* | |
2365 | * Increment vnode ref count only in this case since | |
2366 | * there's an extra vnode reference in the case where | |
2367 | * dp == ip. | |
2368 | */ | |
2369 | IHOLD(dp); | |
2370 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2371 | } | |
2372 | ||
2373 | /* | |
2374 | * Entry must exist since we did a lookup in xfs_lock_dir_and_entry. | |
2375 | */ | |
2376 | XFS_BMAP_INIT(&free_list, &first_block); | |
f6c2d1fa NS |
2377 | error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino, |
2378 | &first_block, &free_list, 0); | |
1da177e4 LT |
2379 | if (error) { |
2380 | ASSERT(error != ENOENT); | |
2381 | REMOVE_DEBUG_TRACE(__LINE__); | |
2382 | goto error1; | |
2383 | } | |
2384 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2385 | ||
2386 | dp->i_gen++; | |
2387 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2388 | ||
2389 | error = xfs_droplink(tp, ip); | |
2390 | if (error) { | |
2391 | REMOVE_DEBUG_TRACE(__LINE__); | |
2392 | goto error1; | |
2393 | } | |
2394 | ||
2395 | /* Determine if this is the last link while | |
2396 | * we are in the transaction. | |
2397 | */ | |
2398 | link_zero = (ip)->i_d.di_nlink==0; | |
2399 | ||
2400 | /* | |
2401 | * Take an extra ref on the inode so that it doesn't | |
2402 | * go to xfs_inactive() from within the commit. | |
2403 | */ | |
2404 | IHOLD(ip); | |
2405 | ||
2406 | /* | |
2407 | * If this is a synchronous mount, make sure that the | |
2408 | * remove transaction goes to disk before returning to | |
2409 | * the user. | |
2410 | */ | |
2411 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2412 | xfs_trans_set_sync(tp); | |
2413 | } | |
2414 | ||
f7c99b6f | 2415 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2416 | if (error) { |
2417 | REMOVE_DEBUG_TRACE(__LINE__); | |
2418 | goto error_rele; | |
2419 | } | |
2420 | ||
1c72bf90 | 2421 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2422 | if (error) { |
2423 | IRELE(ip); | |
2424 | goto std_return; | |
2425 | } | |
2426 | ||
2a82b8be DC |
2427 | /* |
2428 | * If we are using filestreams, kill the stream association. | |
2429 | * If the file is still open it may get a new one but that | |
2430 | * will get killed on last close in xfs_close() so we don't | |
2431 | * have to worry about that. | |
2432 | */ | |
2433 | if (link_zero && xfs_inode_is_filestream(ip)) | |
2434 | xfs_filestream_deassociate(ip); | |
2435 | ||
cf441eeb | 2436 | xfs_itrace_exit(ip); |
1da177e4 LT |
2437 | IRELE(ip); |
2438 | ||
2439 | /* Fall through to std_return with error = 0 */ | |
2440 | std_return: | |
eb9df39d | 2441 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 | 2442 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
bc4ac74a | 2443 | dp, DM_RIGHT_NULL, |
1da177e4 | 2444 | NULL, DM_RIGHT_NULL, |
43973964 | 2445 | name, NULL, ip->i_d.di_mode, error, 0); |
1da177e4 LT |
2446 | } |
2447 | return error; | |
2448 | ||
2449 | error1: | |
2450 | xfs_bmap_cancel(&free_list); | |
2451 | cancel_flags |= XFS_TRANS_ABORT; | |
2452 | xfs_trans_cancel(tp, cancel_flags); | |
2453 | goto std_return; | |
2454 | ||
2455 | error_rele: | |
2456 | /* | |
2457 | * In this case make sure to not release the inode until after | |
2458 | * the current transaction is aborted. Releasing it beforehand | |
2459 | * can cause us to go to xfs_inactive and start a recursive | |
2460 | * transaction which can easily deadlock with the current one. | |
2461 | */ | |
2462 | xfs_bmap_cancel(&free_list); | |
2463 | cancel_flags |= XFS_TRANS_ABORT; | |
2464 | xfs_trans_cancel(tp, cancel_flags); | |
2465 | ||
1da177e4 LT |
2466 | IRELE(ip); |
2467 | ||
2468 | goto std_return; | |
2469 | } | |
2470 | ||
993386c1 | 2471 | int |
1da177e4 | 2472 | xfs_link( |
993386c1 | 2473 | xfs_inode_t *tdp, |
67fcaa73 | 2474 | bhv_vnode_t *src_vp, |
993386c1 | 2475 | bhv_vname_t *dentry) |
1da177e4 | 2476 | { |
993386c1 CH |
2477 | bhv_vnode_t *target_dir_vp = XFS_ITOV(tdp); |
2478 | xfs_mount_t *mp = tdp->i_mount; | |
2479 | xfs_inode_t *sip = xfs_vtoi(src_vp); | |
1da177e4 | 2480 | xfs_trans_t *tp; |
1da177e4 LT |
2481 | xfs_inode_t *ips[2]; |
2482 | int error; | |
2483 | xfs_bmap_free_t free_list; | |
2484 | xfs_fsblock_t first_block; | |
2485 | int cancel_flags; | |
2486 | int committed; | |
1da177e4 LT |
2487 | int resblks; |
2488 | char *target_name = VNAME(dentry); | |
2489 | int target_namelen; | |
2490 | ||
cf441eeb LM |
2491 | xfs_itrace_entry(tdp); |
2492 | xfs_itrace_entry(xfs_vtoi(src_vp)); | |
1da177e4 LT |
2493 | |
2494 | target_namelen = VNAMELEN(dentry); | |
b71d300c | 2495 | ASSERT(!VN_ISDIR(src_vp)); |
1da177e4 | 2496 | |
1da177e4 LT |
2497 | if (XFS_FORCED_SHUTDOWN(mp)) |
2498 | return XFS_ERROR(EIO); | |
2499 | ||
eb9df39d | 2500 | if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) { |
1da177e4 | 2501 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, |
bc4ac74a CH |
2502 | tdp, DM_RIGHT_NULL, |
2503 | sip, DM_RIGHT_NULL, | |
1da177e4 LT |
2504 | target_name, NULL, 0, 0, 0); |
2505 | if (error) | |
2506 | return error; | |
2507 | } | |
2508 | ||
2509 | /* Return through std_return after this point. */ | |
2510 | ||
2511 | error = XFS_QM_DQATTACH(mp, sip, 0); | |
2512 | if (!error && sip != tdp) | |
2513 | error = XFS_QM_DQATTACH(mp, tdp, 0); | |
2514 | if (error) | |
2515 | goto std_return; | |
2516 | ||
2517 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2518 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2519 | resblks = XFS_LINK_SPACE_RES(mp, target_namelen); | |
2520 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, | |
2521 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2522 | if (error == ENOSPC) { | |
2523 | resblks = 0; | |
2524 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2525 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2526 | } | |
2527 | if (error) { | |
2528 | cancel_flags = 0; | |
2529 | goto error_return; | |
2530 | } | |
2531 | ||
2532 | if (sip->i_ino < tdp->i_ino) { | |
2533 | ips[0] = sip; | |
2534 | ips[1] = tdp; | |
2535 | } else { | |
2536 | ips[0] = tdp; | |
2537 | ips[1] = sip; | |
2538 | } | |
2539 | ||
2540 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2541 | ||
2542 | /* | |
2543 | * Increment vnode ref counts since xfs_trans_commit & | |
2544 | * xfs_trans_cancel will both unlock the inodes and | |
2545 | * decrement the associated ref counts. | |
2546 | */ | |
2547 | VN_HOLD(src_vp); | |
2548 | VN_HOLD(target_dir_vp); | |
2549 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
2550 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2551 | ||
2552 | /* | |
2553 | * If the source has too many links, we can't make any more to it. | |
2554 | */ | |
2555 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2556 | error = XFS_ERROR(EMLINK); | |
2557 | goto error_return; | |
2558 | } | |
2559 | ||
365ca83d NS |
2560 | /* |
2561 | * If we are using project inheritance, we only allow hard link | |
2562 | * creation in our tree when the project IDs are the same; else | |
2563 | * the tree quota mechanism could be circumvented. | |
2564 | */ | |
2565 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2566 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 2567 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
2568 | goto error_return; |
2569 | } | |
2570 | ||
1da177e4 | 2571 | if (resblks == 0 && |
f6c2d1fa | 2572 | (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen))) |
1da177e4 LT |
2573 | goto error_return; |
2574 | ||
2575 | XFS_BMAP_INIT(&free_list, &first_block); | |
2576 | ||
f6c2d1fa | 2577 | error = xfs_dir_createname(tp, tdp, target_name, target_namelen, |
1da177e4 LT |
2578 | sip->i_ino, &first_block, &free_list, |
2579 | resblks); | |
2580 | if (error) | |
2581 | goto abort_return; | |
2582 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2583 | tdp->i_gen++; | |
2584 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
2585 | ||
2586 | error = xfs_bumplink(tp, sip); | |
b71d300c | 2587 | if (error) |
1da177e4 | 2588 | goto abort_return; |
1da177e4 LT |
2589 | |
2590 | /* | |
2591 | * If this is a synchronous mount, make sure that the | |
2592 | * link transaction goes to disk before returning to | |
2593 | * the user. | |
2594 | */ | |
2595 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2596 | xfs_trans_set_sync(tp); | |
2597 | } | |
2598 | ||
f7c99b6f | 2599 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2600 | if (error) { |
2601 | xfs_bmap_cancel(&free_list); | |
2602 | goto abort_return; | |
2603 | } | |
2604 | ||
1c72bf90 | 2605 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
b71d300c | 2606 | if (error) |
1da177e4 | 2607 | goto std_return; |
1da177e4 LT |
2608 | |
2609 | /* Fall through to std_return with error = 0. */ | |
2610 | std_return: | |
eb9df39d | 2611 | if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) { |
1da177e4 | 2612 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, |
bc4ac74a CH |
2613 | tdp, DM_RIGHT_NULL, |
2614 | sip, DM_RIGHT_NULL, | |
1da177e4 LT |
2615 | target_name, NULL, 0, error, 0); |
2616 | } | |
2617 | return error; | |
2618 | ||
2619 | abort_return: | |
2620 | cancel_flags |= XFS_TRANS_ABORT; | |
2621 | /* FALLTHROUGH */ | |
014c2544 | 2622 | |
1da177e4 LT |
2623 | error_return: |
2624 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2625 | goto std_return; |
2626 | } | |
b71d300c AD |
2627 | |
2628 | ||
993386c1 | 2629 | int |
1da177e4 | 2630 | xfs_mkdir( |
993386c1 | 2631 | xfs_inode_t *dp, |
8285fb58 | 2632 | bhv_vname_t *dentry, |
3e5daf05 | 2633 | mode_t mode, |
979ebab1 | 2634 | xfs_inode_t **ipp, |
1da177e4 LT |
2635 | cred_t *credp) |
2636 | { | |
2637 | char *dir_name = VNAME(dentry); | |
993386c1 CH |
2638 | int dir_namelen = VNAMELEN(dentry); |
2639 | xfs_mount_t *mp = dp->i_mount; | |
1da177e4 | 2640 | xfs_inode_t *cdp; /* inode of created dir */ |
1da177e4 | 2641 | xfs_trans_t *tp; |
1da177e4 LT |
2642 | int cancel_flags; |
2643 | int error; | |
2644 | int committed; | |
2645 | xfs_bmap_free_t free_list; | |
2646 | xfs_fsblock_t first_block; | |
993386c1 | 2647 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2648 | boolean_t created = B_FALSE; |
2649 | int dm_event_sent = 0; | |
2650 | xfs_prid_t prid; | |
2651 | struct xfs_dquot *udqp, *gdqp; | |
2652 | uint resblks; | |
1da177e4 LT |
2653 | |
2654 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2655 | return XFS_ERROR(EIO); | |
2656 | ||
1da177e4 | 2657 | tp = NULL; |
1da177e4 | 2658 | |
eb9df39d | 2659 | if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) { |
1da177e4 | 2660 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, |
bc4ac74a | 2661 | dp, DM_RIGHT_NULL, NULL, |
1da177e4 | 2662 | DM_RIGHT_NULL, dir_name, NULL, |
3e5daf05 | 2663 | mode, 0, 0); |
1da177e4 LT |
2664 | if (error) |
2665 | return error; | |
2666 | dm_event_sent = 1; | |
2667 | } | |
2668 | ||
2669 | /* Return through std_return after this point. */ | |
2670 | ||
cf441eeb | 2671 | xfs_itrace_entry(dp); |
1da177e4 LT |
2672 | |
2673 | mp = dp->i_mount; | |
2674 | udqp = gdqp = NULL; | |
365ca83d NS |
2675 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2676 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
2677 | else |
2678 | prid = (xfs_prid_t)dfltprid; | |
2679 | ||
2680 | /* | |
2681 | * Make sure that we have allocated dquot(s) on disk. | |
2682 | */ | |
2683 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 2684 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
2685 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2686 | if (error) | |
2687 | goto std_return; | |
2688 | ||
2689 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
2690 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2691 | resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen); | |
2692 | error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0, | |
2693 | XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT); | |
2694 | if (error == ENOSPC) { | |
2695 | resblks = 0; | |
2696 | error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0, | |
2697 | XFS_TRANS_PERM_LOG_RES, | |
2698 | XFS_MKDIR_LOG_COUNT); | |
2699 | } | |
2700 | if (error) { | |
2701 | cancel_flags = 0; | |
1da177e4 LT |
2702 | goto error_return; |
2703 | } | |
2704 | ||
f7c66ce3 | 2705 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 2706 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
2707 | |
2708 | /* | |
2709 | * Check for directory link count overflow. | |
2710 | */ | |
2711 | if (dp->i_d.di_nlink >= XFS_MAXLINK) { | |
2712 | error = XFS_ERROR(EMLINK); | |
2713 | goto error_return; | |
2714 | } | |
2715 | ||
2716 | /* | |
2717 | * Reserve disk quota and the inode. | |
2718 | */ | |
2719 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
2720 | if (error) | |
2721 | goto error_return; | |
2722 | ||
2723 | if (resblks == 0 && | |
f6c2d1fa | 2724 | (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen))) |
1da177e4 LT |
2725 | goto error_return; |
2726 | /* | |
2727 | * create the directory inode. | |
2728 | */ | |
3e5daf05 | 2729 | error = xfs_dir_ialloc(&tp, dp, mode, 2, |
1da177e4 LT |
2730 | 0, credp, prid, resblks > 0, |
2731 | &cdp, NULL); | |
2732 | if (error) { | |
2733 | if (error == ENOSPC) | |
2734 | goto error_return; | |
2735 | goto abort_return; | |
2736 | } | |
cf441eeb | 2737 | xfs_itrace_ref(cdp); |
1da177e4 LT |
2738 | |
2739 | /* | |
2740 | * Now we add the directory inode to the transaction. | |
2741 | * We waited until now since xfs_dir_ialloc might start | |
2742 | * a new transaction. Had we joined the transaction | |
993386c1 CH |
2743 | * earlier, the locks might have gotten released. An error |
2744 | * from here on will result in the transaction cancel | |
2745 | * unlocking dp so don't do it explicitly in the error path. | |
1da177e4 | 2746 | */ |
979ebab1 | 2747 | IHOLD(dp); |
1da177e4 | 2748 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
993386c1 | 2749 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
2750 | |
2751 | XFS_BMAP_INIT(&free_list, &first_block); | |
2752 | ||
f6c2d1fa NS |
2753 | error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino, |
2754 | &first_block, &free_list, resblks ? | |
2755 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
2756 | if (error) { |
2757 | ASSERT(error != ENOSPC); | |
2758 | goto error1; | |
2759 | } | |
2760 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2761 | ||
2762 | /* | |
2763 | * Bump the in memory version number of the parent directory | |
2764 | * so that other processes accessing it will recognize that | |
2765 | * the directory has changed. | |
2766 | */ | |
2767 | dp->i_gen++; | |
2768 | ||
f6c2d1fa NS |
2769 | error = xfs_dir_init(tp, cdp, dp); |
2770 | if (error) | |
1da177e4 | 2771 | goto error2; |
1da177e4 LT |
2772 | |
2773 | cdp->i_gen = 1; | |
2774 | error = xfs_bumplink(tp, dp); | |
f6c2d1fa | 2775 | if (error) |
1da177e4 | 2776 | goto error2; |
1da177e4 | 2777 | |
1da177e4 LT |
2778 | created = B_TRUE; |
2779 | ||
979ebab1 | 2780 | *ipp = cdp; |
1da177e4 LT |
2781 | IHOLD(cdp); |
2782 | ||
2783 | /* | |
2784 | * Attach the dquots to the new inode and modify the icount incore. | |
2785 | */ | |
2786 | XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp); | |
2787 | ||
2788 | /* | |
2789 | * If this is a synchronous mount, make sure that the | |
2790 | * mkdir transaction goes to disk before returning to | |
2791 | * the user. | |
2792 | */ | |
2793 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2794 | xfs_trans_set_sync(tp); | |
2795 | } | |
2796 | ||
f7c99b6f | 2797 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2798 | if (error) { |
2799 | IRELE(cdp); | |
2800 | goto error2; | |
2801 | } | |
2802 | ||
1c72bf90 | 2803 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
2804 | XFS_QM_DQRELE(mp, udqp); |
2805 | XFS_QM_DQRELE(mp, gdqp); | |
2806 | if (error) { | |
2807 | IRELE(cdp); | |
2808 | } | |
2809 | ||
2810 | /* Fall through to std_return with error = 0 or errno from | |
2811 | * xfs_trans_commit. */ | |
2812 | ||
2813 | std_return: | |
eb9df39d | 2814 | if ((created || (error != 0 && dm_event_sent != 0)) && |
993386c1 | 2815 | DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) { |
1da177e4 | 2816 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, |
bc4ac74a CH |
2817 | dp, DM_RIGHT_NULL, |
2818 | created ? cdp : NULL, | |
1da177e4 LT |
2819 | DM_RIGHT_NULL, |
2820 | dir_name, NULL, | |
3e5daf05 | 2821 | mode, error, 0); |
1da177e4 LT |
2822 | } |
2823 | return error; | |
2824 | ||
2825 | error2: | |
2826 | error1: | |
2827 | xfs_bmap_cancel(&free_list); | |
2828 | abort_return: | |
2829 | cancel_flags |= XFS_TRANS_ABORT; | |
2830 | error_return: | |
2831 | xfs_trans_cancel(tp, cancel_flags); | |
2832 | XFS_QM_DQRELE(mp, udqp); | |
2833 | XFS_QM_DQRELE(mp, gdqp); | |
2834 | ||
993386c1 | 2835 | if (unlock_dp_on_error) |
1da177e4 | 2836 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
2837 | |
2838 | goto std_return; | |
2839 | } | |
2840 | ||
993386c1 | 2841 | int |
1da177e4 | 2842 | xfs_rmdir( |
993386c1 CH |
2843 | xfs_inode_t *dp, |
2844 | bhv_vname_t *dentry) | |
1da177e4 | 2845 | { |
993386c1 | 2846 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
1da177e4 | 2847 | char *name = VNAME(dentry); |
993386c1 CH |
2848 | int namelen = VNAMELEN(dentry); |
2849 | xfs_mount_t *mp = dp->i_mount; | |
43973964 | 2850 | xfs_inode_t *cdp = VNAME_TO_INODE(dentry); |
1da177e4 | 2851 | xfs_trans_t *tp; |
1da177e4 LT |
2852 | int error; |
2853 | xfs_bmap_free_t free_list; | |
2854 | xfs_fsblock_t first_block; | |
2855 | int cancel_flags; | |
2856 | int committed; | |
1da177e4 | 2857 | int last_cdp_link; |
1da177e4 LT |
2858 | uint resblks; |
2859 | ||
cf441eeb | 2860 | xfs_itrace_entry(dp); |
1da177e4 | 2861 | |
993386c1 | 2862 | if (XFS_FORCED_SHUTDOWN(mp)) |
1da177e4 | 2863 | return XFS_ERROR(EIO); |
1da177e4 | 2864 | |
eb9df39d | 2865 | if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) { |
1da177e4 | 2866 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, |
bc4ac74a | 2867 | dp, DM_RIGHT_NULL, |
1da177e4 | 2868 | NULL, DM_RIGHT_NULL, |
43973964 | 2869 | name, NULL, cdp->i_d.di_mode, 0, 0); |
1da177e4 LT |
2870 | if (error) |
2871 | return XFS_ERROR(error); | |
2872 | } | |
2873 | ||
1da177e4 LT |
2874 | /* |
2875 | * We need to get a reference to cdp before we get our log | |
2876 | * reservation. The reason for this is that we cannot call | |
2877 | * xfs_iget for an inode for which we do not have a reference | |
2878 | * once we've acquired a log reservation. This is because the | |
2879 | * inode we are trying to get might be in xfs_inactive going | |
2880 | * for a log reservation. Since we'll have to wait for the | |
2881 | * inactive code to complete before returning from xfs_iget, | |
2882 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2883 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2884 | * to the inode before getting our log reservation. |
2885 | */ | |
43973964 | 2886 | IHOLD(cdp); |
1da177e4 LT |
2887 | |
2888 | /* | |
2889 | * Get the dquots for the inodes. | |
2890 | */ | |
2891 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2892 | if (!error && dp != cdp) | |
2893 | error = XFS_QM_DQATTACH(mp, cdp, 0); | |
2894 | if (error) { | |
2895 | IRELE(cdp); | |
2896 | REMOVE_DEBUG_TRACE(__LINE__); | |
2897 | goto std_return; | |
2898 | } | |
2899 | ||
2900 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
2901 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2902 | /* | |
2903 | * We try to get the real space reservation first, | |
2904 | * allowing for directory btree deletion(s) implying | |
2905 | * possible bmap insert(s). If we can't get the space | |
2906 | * reservation then we use 0 instead, and avoid the bmap | |
2907 | * btree insert(s) in the directory code by, if the bmap | |
2908 | * insert tries to happen, instead trimming the LAST | |
2909 | * block from the directory. | |
2910 | */ | |
2911 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2912 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2913 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
2914 | if (error == ENOSPC) { | |
2915 | resblks = 0; | |
2916 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2917 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
2918 | } | |
2919 | if (error) { | |
2920 | ASSERT(error != ENOSPC); | |
2921 | cancel_flags = 0; | |
2922 | IRELE(cdp); | |
2923 | goto error_return; | |
2924 | } | |
2925 | XFS_BMAP_INIT(&free_list, &first_block); | |
2926 | ||
2927 | /* | |
2928 | * Now lock the child directory inode and the parent directory | |
2929 | * inode in the proper order. This will take care of validating | |
2930 | * that the directory entry for the child directory inode has | |
2931 | * not changed while we were obtaining a log reservation. | |
2932 | */ | |
e9ed9d22 | 2933 | error = xfs_lock_dir_and_entry(dp, cdp); |
1da177e4 LT |
2934 | if (error) { |
2935 | xfs_trans_cancel(tp, cancel_flags); | |
2936 | IRELE(cdp); | |
2937 | goto std_return; | |
2938 | } | |
2939 | ||
2940 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2941 | if (dp != cdp) { | |
2942 | /* | |
2943 | * Only increment the parent directory vnode count if | |
2944 | * we didn't bump it in looking up cdp. The only time | |
2945 | * we don't bump it is when we're looking up ".". | |
2946 | */ | |
2947 | VN_HOLD(dir_vp); | |
2948 | } | |
2949 | ||
cf441eeb | 2950 | xfs_itrace_ref(cdp); |
1da177e4 LT |
2951 | xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL); |
2952 | ||
2953 | ASSERT(cdp->i_d.di_nlink >= 2); | |
2954 | if (cdp->i_d.di_nlink != 2) { | |
2955 | error = XFS_ERROR(ENOTEMPTY); | |
2956 | goto error_return; | |
2957 | } | |
f6c2d1fa | 2958 | if (!xfs_dir_isempty(cdp)) { |
1da177e4 LT |
2959 | error = XFS_ERROR(ENOTEMPTY); |
2960 | goto error_return; | |
2961 | } | |
2962 | ||
f6c2d1fa NS |
2963 | error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino, |
2964 | &first_block, &free_list, resblks); | |
2965 | if (error) | |
1da177e4 | 2966 | goto error1; |
1da177e4 LT |
2967 | |
2968 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2969 | ||
2970 | /* | |
2971 | * Bump the in memory generation count on the parent | |
2972 | * directory so that other can know that it has changed. | |
2973 | */ | |
2974 | dp->i_gen++; | |
2975 | ||
2976 | /* | |
2977 | * Drop the link from cdp's "..". | |
2978 | */ | |
2979 | error = xfs_droplink(tp, dp); | |
2980 | if (error) { | |
2981 | goto error1; | |
2982 | } | |
2983 | ||
2984 | /* | |
2985 | * Drop the link from dp to cdp. | |
2986 | */ | |
2987 | error = xfs_droplink(tp, cdp); | |
2988 | if (error) { | |
2989 | goto error1; | |
2990 | } | |
2991 | ||
2992 | /* | |
2993 | * Drop the "." link from cdp to self. | |
2994 | */ | |
2995 | error = xfs_droplink(tp, cdp); | |
2996 | if (error) { | |
2997 | goto error1; | |
2998 | } | |
2999 | ||
3000 | /* Determine these before committing transaction */ | |
3001 | last_cdp_link = (cdp)->i_d.di_nlink==0; | |
3002 | ||
3003 | /* | |
3004 | * Take an extra ref on the child vnode so that it | |
3005 | * does not go to xfs_inactive() from within the commit. | |
3006 | */ | |
3007 | IHOLD(cdp); | |
3008 | ||
3009 | /* | |
3010 | * If this is a synchronous mount, make sure that the | |
3011 | * rmdir transaction goes to disk before returning to | |
3012 | * the user. | |
3013 | */ | |
3014 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3015 | xfs_trans_set_sync(tp); | |
3016 | } | |
3017 | ||
f7c99b6f | 3018 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
3019 | if (error) { |
3020 | xfs_bmap_cancel(&free_list); | |
3021 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | | |
3022 | XFS_TRANS_ABORT)); | |
3023 | IRELE(cdp); | |
3024 | goto std_return; | |
3025 | } | |
3026 | ||
1c72bf90 | 3027 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3028 | if (error) { |
3029 | IRELE(cdp); | |
3030 | goto std_return; | |
3031 | } | |
3032 | ||
3033 | ||
1da177e4 LT |
3034 | IRELE(cdp); |
3035 | ||
3036 | /* Fall through to std_return with error = 0 or the errno | |
3037 | * from xfs_trans_commit. */ | |
bb19fba1 | 3038 | std_return: |
eb9df39d | 3039 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) { |
1da177e4 | 3040 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, |
bc4ac74a | 3041 | dp, DM_RIGHT_NULL, |
1da177e4 | 3042 | NULL, DM_RIGHT_NULL, |
43973964 | 3043 | name, NULL, cdp->i_d.di_mode, |
1da177e4 LT |
3044 | error, 0); |
3045 | } | |
3046 | return error; | |
3047 | ||
bb19fba1 | 3048 | error1: |
1da177e4 LT |
3049 | xfs_bmap_cancel(&free_list); |
3050 | cancel_flags |= XFS_TRANS_ABORT; | |
014c2544 JJ |
3051 | /* FALLTHROUGH */ |
3052 | ||
bb19fba1 | 3053 | error_return: |
1da177e4 LT |
3054 | xfs_trans_cancel(tp, cancel_flags); |
3055 | goto std_return; | |
3056 | } | |
3057 | ||
993386c1 | 3058 | int |
1da177e4 | 3059 | xfs_symlink( |
993386c1 | 3060 | xfs_inode_t *dp, |
8285fb58 | 3061 | bhv_vname_t *dentry, |
1da177e4 | 3062 | char *target_path, |
3e5daf05 | 3063 | mode_t mode, |
67fcaa73 | 3064 | bhv_vnode_t **vpp, |
1da177e4 LT |
3065 | cred_t *credp) |
3066 | { | |
993386c1 CH |
3067 | bhv_vnode_t *dir_vp = XFS_ITOV(dp); |
3068 | xfs_mount_t *mp = dp->i_mount; | |
1da177e4 | 3069 | xfs_trans_t *tp; |
1da177e4 LT |
3070 | xfs_inode_t *ip; |
3071 | int error; | |
3072 | int pathlen; | |
3073 | xfs_bmap_free_t free_list; | |
3074 | xfs_fsblock_t first_block; | |
993386c1 | 3075 | boolean_t unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
3076 | uint cancel_flags; |
3077 | int committed; | |
3078 | xfs_fileoff_t first_fsb; | |
3079 | xfs_filblks_t fs_blocks; | |
3080 | int nmaps; | |
3081 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
3082 | xfs_daddr_t d; | |
3083 | char *cur_chunk; | |
3084 | int byte_cnt; | |
3085 | int n; | |
3086 | xfs_buf_t *bp; | |
3087 | xfs_prid_t prid; | |
3088 | struct xfs_dquot *udqp, *gdqp; | |
3089 | uint resblks; | |
3090 | char *link_name = VNAME(dentry); | |
3091 | int link_namelen; | |
3092 | ||
3093 | *vpp = NULL; | |
1da177e4 LT |
3094 | error = 0; |
3095 | ip = NULL; | |
3096 | tp = NULL; | |
3097 | ||
cf441eeb | 3098 | xfs_itrace_entry(dp); |
1da177e4 LT |
3099 | |
3100 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3101 | return XFS_ERROR(EIO); | |
3102 | ||
3103 | link_namelen = VNAMELEN(dentry); | |
3104 | ||
3105 | /* | |
3106 | * Check component lengths of the target path name. | |
3107 | */ | |
3108 | pathlen = strlen(target_path); | |
3109 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
3110 | return XFS_ERROR(ENAMETOOLONG); | |
3111 | if (pathlen >= MAXNAMELEN) { /* is any component too long? */ | |
3112 | int len, total; | |
3113 | char *path; | |
3114 | ||
f6c2d1fa | 3115 | for (total = 0, path = target_path; total < pathlen;) { |
1da177e4 LT |
3116 | /* |
3117 | * Skip any slashes. | |
3118 | */ | |
3119 | while(*path == '/') { | |
3120 | total++; | |
3121 | path++; | |
3122 | } | |
3123 | ||
3124 | /* | |
3125 | * Count up to the next slash or end of path. | |
3126 | * Error out if the component is bigger than MAXNAMELEN. | |
3127 | */ | |
3128 | for(len = 0; *path != '/' && total < pathlen;total++, path++) { | |
3129 | if (++len >= MAXNAMELEN) { | |
3130 | error = ENAMETOOLONG; | |
3131 | return error; | |
3132 | } | |
3133 | } | |
3134 | } | |
3135 | } | |
3136 | ||
eb9df39d | 3137 | if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) { |
bc4ac74a | 3138 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp, |
1da177e4 LT |
3139 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, |
3140 | link_name, target_path, 0, 0, 0); | |
3141 | if (error) | |
3142 | return error; | |
3143 | } | |
3144 | ||
3145 | /* Return through std_return after this point. */ | |
3146 | ||
3147 | udqp = gdqp = NULL; | |
365ca83d NS |
3148 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
3149 | prid = dp->i_d.di_projid; | |
1da177e4 LT |
3150 | else |
3151 | prid = (xfs_prid_t)dfltprid; | |
3152 | ||
3153 | /* | |
3154 | * Make sure that we have allocated dquot(s) on disk. | |
3155 | */ | |
3156 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 3157 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
3158 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
3159 | if (error) | |
3160 | goto std_return; | |
3161 | ||
3162 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
3163 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3164 | /* | |
3165 | * The symlink will fit into the inode data fork? | |
3166 | * There can't be any attributes so we get the whole variable part. | |
3167 | */ | |
3168 | if (pathlen <= XFS_LITINO(mp)) | |
3169 | fs_blocks = 0; | |
3170 | else | |
3171 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
3172 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks); | |
3173 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, | |
3174 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3175 | if (error == ENOSPC && fs_blocks == 0) { | |
3176 | resblks = 0; | |
3177 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
3178 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3179 | } | |
3180 | if (error) { | |
3181 | cancel_flags = 0; | |
1da177e4 LT |
3182 | goto error_return; |
3183 | } | |
3184 | ||
f7c66ce3 | 3185 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
993386c1 | 3186 | unlock_dp_on_error = B_TRUE; |
1da177e4 LT |
3187 | |
3188 | /* | |
3189 | * Check whether the directory allows new symlinks or not. | |
3190 | */ | |
3191 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
3192 | error = XFS_ERROR(EPERM); | |
3193 | goto error_return; | |
3194 | } | |
3195 | ||
3196 | /* | |
3197 | * Reserve disk quota : blocks and inode. | |
3198 | */ | |
3199 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
3200 | if (error) | |
3201 | goto error_return; | |
3202 | ||
3203 | /* | |
3204 | * Check for ability to enter directory entry, if no space reserved. | |
3205 | */ | |
3206 | if (resblks == 0 && | |
f6c2d1fa | 3207 | (error = xfs_dir_canenter(tp, dp, link_name, link_namelen))) |
1da177e4 LT |
3208 | goto error_return; |
3209 | /* | |
3210 | * Initialize the bmap freelist prior to calling either | |
3211 | * bmapi or the directory create code. | |
3212 | */ | |
3213 | XFS_BMAP_INIT(&free_list, &first_block); | |
3214 | ||
3215 | /* | |
3216 | * Allocate an inode for the symlink. | |
3217 | */ | |
3e5daf05 | 3218 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), |
1da177e4 LT |
3219 | 1, 0, credp, prid, resblks > 0, &ip, NULL); |
3220 | if (error) { | |
3221 | if (error == ENOSPC) | |
3222 | goto error_return; | |
3223 | goto error1; | |
3224 | } | |
cf441eeb | 3225 | xfs_itrace_ref(ip); |
1da177e4 | 3226 | |
993386c1 CH |
3227 | /* |
3228 | * An error after we've joined dp to the transaction will result in the | |
3229 | * transaction cancel unlocking dp so don't do it explicitly in the | |
3230 | * error path. | |
3231 | */ | |
1da177e4 LT |
3232 | VN_HOLD(dir_vp); |
3233 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
993386c1 | 3234 | unlock_dp_on_error = B_FALSE; |
1da177e4 LT |
3235 | |
3236 | /* | |
3237 | * Also attach the dquot(s) to it, if applicable. | |
3238 | */ | |
3239 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
3240 | ||
3241 | if (resblks) | |
3242 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
3243 | /* | |
3244 | * If the symlink will fit into the inode, write it inline. | |
3245 | */ | |
3246 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
3247 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
3248 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
3249 | ip->i_d.di_size = pathlen; | |
3250 | ||
3251 | /* | |
3252 | * The inode was initially created in extent format. | |
3253 | */ | |
3254 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
3255 | ip->i_df.if_flags |= XFS_IFINLINE; | |
3256 | ||
3257 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
3258 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
3259 | ||
3260 | } else { | |
3261 | first_fsb = 0; | |
3262 | nmaps = SYMLINK_MAPS; | |
3263 | ||
3264 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
3265 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
3266 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 3267 | &free_list, NULL); |
1da177e4 LT |
3268 | if (error) { |
3269 | goto error1; | |
3270 | } | |
3271 | ||
3272 | if (resblks) | |
3273 | resblks -= fs_blocks; | |
3274 | ip->i_d.di_size = pathlen; | |
3275 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3276 | ||
3277 | cur_chunk = target_path; | |
3278 | for (n = 0; n < nmaps; n++) { | |
3279 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
3280 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
3281 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
3282 | BTOBB(byte_cnt), 0); | |
3283 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
3284 | if (pathlen < byte_cnt) { | |
3285 | byte_cnt = pathlen; | |
3286 | } | |
3287 | pathlen -= byte_cnt; | |
3288 | ||
3289 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
3290 | cur_chunk += byte_cnt; | |
3291 | ||
3292 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
3293 | } | |
3294 | } | |
3295 | ||
3296 | /* | |
3297 | * Create the directory entry for the symlink. | |
3298 | */ | |
f6c2d1fa NS |
3299 | error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino, |
3300 | &first_block, &free_list, resblks); | |
3301 | if (error) | |
1da177e4 | 3302 | goto error1; |
1da177e4 LT |
3303 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3304 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
3305 | ||
3306 | /* | |
3307 | * Bump the in memory version number of the parent directory | |
3308 | * so that other processes accessing it will recognize that | |
3309 | * the directory has changed. | |
3310 | */ | |
3311 | dp->i_gen++; | |
3312 | ||
3313 | /* | |
3314 | * If this is a synchronous mount, make sure that the | |
3315 | * symlink transaction goes to disk before returning to | |
3316 | * the user. | |
3317 | */ | |
3318 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3319 | xfs_trans_set_sync(tp); | |
3320 | } | |
3321 | ||
3322 | /* | |
3323 | * xfs_trans_commit normally decrements the vnode ref count | |
3324 | * when it unlocks the inode. Since we want to return the | |
3325 | * vnode to the caller, we bump the vnode ref count now. | |
3326 | */ | |
3327 | IHOLD(ip); | |
3328 | ||
f7c99b6f | 3329 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3330 | if (error) { |
3331 | goto error2; | |
3332 | } | |
1c72bf90 | 3333 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3334 | XFS_QM_DQRELE(mp, udqp); |
3335 | XFS_QM_DQRELE(mp, gdqp); | |
3336 | ||
3337 | /* Fall through to std_return with error = 0 or errno from | |
3338 | * xfs_trans_commit */ | |
3339 | std_return: | |
993386c1 | 3340 | if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) { |
1da177e4 | 3341 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, |
bc4ac74a CH |
3342 | dp, DM_RIGHT_NULL, |
3343 | error ? NULL : ip, | |
1da177e4 LT |
3344 | DM_RIGHT_NULL, link_name, target_path, |
3345 | 0, error, 0); | |
3346 | } | |
3347 | ||
3348 | if (!error) { | |
67fcaa73 | 3349 | bhv_vnode_t *vp; |
1da177e4 LT |
3350 | |
3351 | ASSERT(ip); | |
3352 | vp = XFS_ITOV(ip); | |
3353 | *vpp = vp; | |
3354 | } | |
3355 | return error; | |
3356 | ||
3357 | error2: | |
3358 | IRELE(ip); | |
3359 | error1: | |
3360 | xfs_bmap_cancel(&free_list); | |
3361 | cancel_flags |= XFS_TRANS_ABORT; | |
3362 | error_return: | |
3363 | xfs_trans_cancel(tp, cancel_flags); | |
3364 | XFS_QM_DQRELE(mp, udqp); | |
3365 | XFS_QM_DQRELE(mp, gdqp); | |
3366 | ||
993386c1 | 3367 | if (unlock_dp_on_error) |
1da177e4 | 3368 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
1da177e4 LT |
3369 | |
3370 | goto std_return; | |
3371 | } | |
3372 | ||
993386c1 | 3373 | int |
1da177e4 | 3374 | xfs_inode_flush( |
993386c1 | 3375 | xfs_inode_t *ip, |
1da177e4 LT |
3376 | int flags) |
3377 | { | |
993386c1 | 3378 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
3379 | int error = 0; |
3380 | ||
1da177e4 LT |
3381 | if (XFS_FORCED_SHUTDOWN(mp)) |
3382 | return XFS_ERROR(EIO); | |
3383 | ||
3384 | /* | |
3385 | * Bypass inodes which have already been cleaned by | |
3386 | * the inode flush clustering code inside xfs_iflush | |
3387 | */ | |
33540408 | 3388 | if (xfs_inode_clean(ip)) |
1da177e4 LT |
3389 | return 0; |
3390 | ||
1da177e4 LT |
3391 | /* |
3392 | * We make this non-blocking if the inode is contended, | |
3393 | * return EAGAIN to indicate to the caller that they | |
3394 | * did not succeed. This prevents the flush path from | |
3395 | * blocking on inodes inside another operation right | |
3396 | * now, they get caught later by xfs_sync. | |
3397 | */ | |
a3f74ffb DC |
3398 | if (flags & FLUSH_SYNC) { |
3399 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3400 | xfs_iflock(ip); | |
3401 | } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
3402 | if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) { | |
3403 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1da177e4 LT |
3404 | return EAGAIN; |
3405 | } | |
a3f74ffb DC |
3406 | } else { |
3407 | return EAGAIN; | |
1da177e4 LT |
3408 | } |
3409 | ||
a3f74ffb DC |
3410 | error = xfs_iflush(ip, (flags & FLUSH_SYNC) ? XFS_IFLUSH_SYNC |
3411 | : XFS_IFLUSH_ASYNC_NOBLOCK); | |
3412 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3413 | ||
1da177e4 LT |
3414 | return error; |
3415 | } | |
3416 | ||
993386c1 | 3417 | |
1da177e4 | 3418 | int |
993386c1 CH |
3419 | xfs_set_dmattrs( |
3420 | xfs_inode_t *ip, | |
1da177e4 | 3421 | u_int evmask, |
993386c1 | 3422 | u_int16_t state) |
1da177e4 | 3423 | { |
993386c1 | 3424 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 | 3425 | xfs_trans_t *tp; |
1da177e4 LT |
3426 | int error; |
3427 | ||
3428 | if (!capable(CAP_SYS_ADMIN)) | |
3429 | return XFS_ERROR(EPERM); | |
3430 | ||
1da177e4 LT |
3431 | if (XFS_FORCED_SHUTDOWN(mp)) |
3432 | return XFS_ERROR(EIO); | |
3433 | ||
3434 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
3435 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
3436 | if (error) { | |
3437 | xfs_trans_cancel(tp, 0); | |
3438 | return error; | |
3439 | } | |
3440 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3441 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3442 | ||
613d7043 CH |
3443 | ip->i_d.di_dmevmask = evmask; |
3444 | ip->i_d.di_dmstate = state; | |
1da177e4 LT |
3445 | |
3446 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3447 | IHOLD(ip); | |
1c72bf90 | 3448 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
3449 | |
3450 | return error; | |
3451 | } | |
3452 | ||
993386c1 | 3453 | int |
1da177e4 | 3454 | xfs_reclaim( |
993386c1 | 3455 | xfs_inode_t *ip) |
1da177e4 | 3456 | { |
993386c1 | 3457 | bhv_vnode_t *vp = XFS_ITOV(ip); |
1da177e4 | 3458 | |
cf441eeb | 3459 | xfs_itrace_entry(ip); |
1da177e4 LT |
3460 | |
3461 | ASSERT(!VN_MAPPED(vp)); | |
3462 | ||
3463 | /* bad inode, get out here ASAP */ | |
3464 | if (VN_BAD(vp)) { | |
3465 | xfs_ireclaim(ip); | |
3466 | return 0; | |
3467 | } | |
3468 | ||
b677c210 | 3469 | vn_iowait(ip); |
1da177e4 | 3470 | |
51c91ed5 | 3471 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); |
1da177e4 | 3472 | |
42fe2b1f CH |
3473 | /* |
3474 | * Make sure the atime in the XFS inode is correct before freeing the | |
3475 | * Linux inode. | |
3476 | */ | |
3477 | xfs_synchronize_atime(ip); | |
1da177e4 | 3478 | |
4c60658e DC |
3479 | /* |
3480 | * If we have nothing to flush with this inode then complete the | |
3481 | * teardown now, otherwise break the link between the xfs inode and the | |
3482 | * linux inode and clean up the xfs inode later. This avoids flushing | |
3483 | * the inode to disk during the delete operation itself. | |
3484 | * | |
3485 | * When breaking the link, we need to set the XFS_IRECLAIMABLE flag | |
3486 | * first to ensure that xfs_iunpin() will never see an xfs inode | |
3487 | * that has a linux inode being reclaimed. Synchronisation is provided | |
3488 | * by the i_flags_lock. | |
1da177e4 LT |
3489 | */ |
3490 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
3491 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3492 | xfs_iflock(ip); | |
3493 | return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); | |
3494 | } else { | |
3495 | xfs_mount_t *mp = ip->i_mount; | |
3496 | ||
4c60658e | 3497 | /* Protect sync and unpin from us */ |
1da177e4 | 3498 | XFS_MOUNT_ILOCK(mp); |
4c60658e DC |
3499 | spin_lock(&ip->i_flags_lock); |
3500 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
739bfb2a CH |
3501 | vn_to_inode(vp)->i_private = NULL; |
3502 | ip->i_vnode = NULL; | |
4c60658e | 3503 | spin_unlock(&ip->i_flags_lock); |
1da177e4 | 3504 | list_add_tail(&ip->i_reclaim, &mp->m_del_inodes); |
1da177e4 LT |
3505 | XFS_MOUNT_IUNLOCK(mp); |
3506 | } | |
3507 | return 0; | |
3508 | } | |
3509 | ||
3510 | int | |
3511 | xfs_finish_reclaim( | |
3512 | xfs_inode_t *ip, | |
3513 | int locked, | |
3514 | int sync_mode) | |
3515 | { | |
da353b0d | 3516 | xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino); |
67fcaa73 | 3517 | bhv_vnode_t *vp = XFS_ITOV_NULL(ip); |
1da177e4 LT |
3518 | int error; |
3519 | ||
3520 | if (vp && VN_BAD(vp)) | |
3521 | goto reclaim; | |
3522 | ||
3523 | /* The hash lock here protects a thread in xfs_iget_core from | |
3524 | * racing with us on linking the inode back with a vnode. | |
3525 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
3526 | * us. | |
3527 | */ | |
da353b0d | 3528 | write_lock(&pag->pag_ici_lock); |
f273ab84 | 3529 | spin_lock(&ip->i_flags_lock); |
7a18c386 DC |
3530 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || |
3531 | (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) { | |
f273ab84 | 3532 | spin_unlock(&ip->i_flags_lock); |
da353b0d | 3533 | write_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
3534 | if (locked) { |
3535 | xfs_ifunlock(ip); | |
3536 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3537 | } | |
014c2544 | 3538 | return 1; |
1da177e4 | 3539 | } |
7a18c386 | 3540 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
f273ab84 | 3541 | spin_unlock(&ip->i_flags_lock); |
da353b0d DC |
3542 | write_unlock(&pag->pag_ici_lock); |
3543 | xfs_put_perag(ip->i_mount, pag); | |
1da177e4 LT |
3544 | |
3545 | /* | |
3546 | * If the inode is still dirty, then flush it out. If the inode | |
3547 | * is not in the AIL, then it will be OK to flush it delwri as | |
3548 | * long as xfs_iflush() does not keep any references to the inode. | |
3549 | * We leave that decision up to xfs_iflush() since it has the | |
3550 | * knowledge of whether it's OK to simply do a delwri flush of | |
3551 | * the inode or whether we need to wait until the inode is | |
3552 | * pulled from the AIL. | |
3553 | * We get the flush lock regardless, though, just to make sure | |
3554 | * we don't free it while it is being flushed. | |
3555 | */ | |
461aa8a2 LM |
3556 | if (!locked) { |
3557 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3558 | xfs_iflock(ip); | |
3559 | } | |
1da177e4 | 3560 | |
461aa8a2 | 3561 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1da177e4 LT |
3562 | if (ip->i_update_core || |
3563 | ((ip->i_itemp != NULL) && | |
3564 | (ip->i_itemp->ili_format.ilf_fields != 0))) { | |
3565 | error = xfs_iflush(ip, sync_mode); | |
3566 | /* | |
3567 | * If we hit an error, typically because of filesystem | |
3568 | * shutdown, we don't need to let vn_reclaim to know | |
3569 | * because we're gonna reclaim the inode anyway. | |
3570 | */ | |
3571 | if (error) { | |
3572 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3573 | goto reclaim; | |
3574 | } | |
3575 | xfs_iflock(ip); /* synchronize with xfs_iflush_done */ | |
3576 | } | |
3577 | ||
3578 | ASSERT(ip->i_update_core == 0); | |
3579 | ASSERT(ip->i_itemp == NULL || | |
3580 | ip->i_itemp->ili_format.ilf_fields == 0); | |
1da177e4 LT |
3581 | } |
3582 | ||
461aa8a2 LM |
3583 | xfs_ifunlock(ip); |
3584 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3585 | ||
1da177e4 LT |
3586 | reclaim: |
3587 | xfs_ireclaim(ip); | |
3588 | return 0; | |
3589 | } | |
3590 | ||
3591 | int | |
3592 | xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock) | |
3593 | { | |
3594 | int purged; | |
3595 | xfs_inode_t *ip, *n; | |
3596 | int done = 0; | |
3597 | ||
3598 | while (!done) { | |
3599 | purged = 0; | |
3600 | XFS_MOUNT_ILOCK(mp); | |
3601 | list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) { | |
3602 | if (noblock) { | |
3603 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) | |
3604 | continue; | |
3605 | if (xfs_ipincount(ip) || | |
3606 | !xfs_iflock_nowait(ip)) { | |
3607 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3608 | continue; | |
3609 | } | |
3610 | } | |
3611 | XFS_MOUNT_IUNLOCK(mp); | |
6b2cf618 FB |
3612 | if (xfs_finish_reclaim(ip, noblock, |
3613 | XFS_IFLUSH_DELWRI_ELSE_ASYNC)) | |
3614 | delay(1); | |
1da177e4 LT |
3615 | purged = 1; |
3616 | break; | |
3617 | } | |
3618 | ||
3619 | done = !purged; | |
3620 | } | |
3621 | ||
3622 | XFS_MOUNT_IUNLOCK(mp); | |
3623 | return 0; | |
3624 | } | |
3625 | ||
3626 | /* | |
3627 | * xfs_alloc_file_space() | |
3628 | * This routine allocates disk space for the given file. | |
3629 | * | |
3630 | * If alloc_type == 0, this request is for an ALLOCSP type | |
3631 | * request which will change the file size. In this case, no | |
3632 | * DMAPI event will be generated by the call. A TRUNCATE event | |
3633 | * will be generated later by xfs_setattr. | |
3634 | * | |
3635 | * If alloc_type != 0, this request is for a RESVSP type | |
3636 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
3637 | * lower block boundary byte address is less than the file's | |
3638 | * length. | |
3639 | * | |
3640 | * RETURNS: | |
3641 | * 0 on success | |
3642 | * errno on error | |
3643 | * | |
3644 | */ | |
ba0f32d4 | 3645 | STATIC int |
1da177e4 LT |
3646 | xfs_alloc_file_space( |
3647 | xfs_inode_t *ip, | |
3648 | xfs_off_t offset, | |
3649 | xfs_off_t len, | |
3650 | int alloc_type, | |
3651 | int attr_flags) | |
3652 | { | |
dd9f438e NS |
3653 | xfs_mount_t *mp = ip->i_mount; |
3654 | xfs_off_t count; | |
1da177e4 LT |
3655 | xfs_filblks_t allocated_fsb; |
3656 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
3657 | xfs_extlen_t extsz, temp; |
3658 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 3659 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
3660 | int nimaps; |
3661 | int bmapi_flag; | |
3662 | int quota_flag; | |
1da177e4 | 3663 | int rt; |
1da177e4 | 3664 | xfs_trans_t *tp; |
dd9f438e NS |
3665 | xfs_bmbt_irec_t imaps[1], *imapp; |
3666 | xfs_bmap_free_t free_list; | |
3667 | uint qblocks, resblks, resrtextents; | |
3668 | int committed; | |
3669 | int error; | |
1da177e4 | 3670 | |
cf441eeb | 3671 | xfs_itrace_entry(ip); |
1da177e4 LT |
3672 | |
3673 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3674 | return XFS_ERROR(EIO); | |
3675 | ||
1da177e4 LT |
3676 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
3677 | return error; | |
3678 | ||
3679 | if (len <= 0) | |
3680 | return XFS_ERROR(EINVAL); | |
3681 | ||
957d0ebe DC |
3682 | rt = XFS_IS_REALTIME_INODE(ip); |
3683 | extsz = xfs_get_extsz_hint(ip); | |
3684 | ||
1da177e4 | 3685 | count = len; |
1da177e4 | 3686 | imapp = &imaps[0]; |
dd9f438e NS |
3687 | nimaps = 1; |
3688 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
3689 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
3690 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
3691 | ||
3692 | /* Generate a DMAPI event if needed. */ | |
ba87ea69 | 3693 | if (alloc_type != 0 && offset < ip->i_size && |
1da177e4 | 3694 | (attr_flags&ATTR_DMI) == 0 && |
eb9df39d | 3695 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { |
1da177e4 LT |
3696 | xfs_off_t end_dmi_offset; |
3697 | ||
3698 | end_dmi_offset = offset+len; | |
ba87ea69 LM |
3699 | if (end_dmi_offset > ip->i_size) |
3700 | end_dmi_offset = ip->i_size; | |
bc4ac74a CH |
3701 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset, |
3702 | end_dmi_offset - offset, 0, NULL); | |
1da177e4 | 3703 | if (error) |
014c2544 | 3704 | return error; |
1da177e4 LT |
3705 | } |
3706 | ||
3707 | /* | |
dd9f438e | 3708 | * Allocate file space until done or until there is an error |
1da177e4 LT |
3709 | */ |
3710 | retry: | |
3711 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
3712 | xfs_fileoff_t s, e; |
3713 | ||
1da177e4 | 3714 | /* |
3ddb8fa9 | 3715 | * Determine space reservations for data/realtime. |
1da177e4 | 3716 | */ |
dd9f438e | 3717 | if (unlikely(extsz)) { |
1da177e4 | 3718 | s = startoffset_fsb; |
dd9f438e NS |
3719 | do_div(s, extsz); |
3720 | s *= extsz; | |
3721 | e = startoffset_fsb + allocatesize_fsb; | |
3722 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
3723 | e += temp; | |
3724 | if ((temp = do_mod(e, extsz))) | |
3725 | e += extsz - temp; | |
3726 | } else { | |
3727 | s = 0; | |
3728 | e = allocatesize_fsb; | |
3729 | } | |
3730 | ||
3731 | if (unlikely(rt)) { | |
3732 | resrtextents = qblocks = (uint)(e - s); | |
3733 | resrtextents /= mp->m_sb.sb_rextsize; | |
3734 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
3735 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 3736 | } else { |
dd9f438e NS |
3737 | resrtextents = 0; |
3738 | resblks = qblocks = \ | |
3739 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
3740 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
3741 | } |
3742 | ||
3743 | /* | |
dd9f438e | 3744 | * Allocate and setup the transaction. |
1da177e4 LT |
3745 | */ |
3746 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
3747 | error = xfs_trans_reserve(tp, resblks, |
3748 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
3749 | XFS_TRANS_PERM_LOG_RES, |
3750 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 3751 | /* |
dd9f438e | 3752 | * Check for running out of space |
1da177e4 LT |
3753 | */ |
3754 | if (error) { | |
3755 | /* | |
3756 | * Free the transaction structure. | |
3757 | */ | |
3758 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
3759 | xfs_trans_cancel(tp, 0); | |
3760 | break; | |
3761 | } | |
3762 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
dd9f438e NS |
3763 | error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, |
3764 | qblocks, 0, quota_flag); | |
1da177e4 LT |
3765 | if (error) |
3766 | goto error1; | |
3767 | ||
3768 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3769 | xfs_trans_ihold(tp, ip); | |
3770 | ||
3771 | /* | |
dd9f438e | 3772 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 LT |
3773 | */ |
3774 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
541d7d3c | 3775 | error = xfs_bmapi(tp, ip, startoffset_fsb, |
dd9f438e NS |
3776 | allocatesize_fsb, bmapi_flag, |
3777 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 3778 | &free_list, NULL); |
1da177e4 LT |
3779 | if (error) { |
3780 | goto error0; | |
3781 | } | |
3782 | ||
3783 | /* | |
dd9f438e | 3784 | * Complete the transaction |
1da177e4 | 3785 | */ |
f7c99b6f | 3786 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3787 | if (error) { |
3788 | goto error0; | |
3789 | } | |
3790 | ||
1c72bf90 | 3791 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
3792 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
3793 | if (error) { | |
3794 | break; | |
3795 | } | |
3796 | ||
3797 | allocated_fsb = imapp->br_blockcount; | |
3798 | ||
dd9f438e | 3799 | if (nimaps == 0) { |
1da177e4 LT |
3800 | error = XFS_ERROR(ENOSPC); |
3801 | break; | |
3802 | } | |
3803 | ||
3804 | startoffset_fsb += allocated_fsb; | |
3805 | allocatesize_fsb -= allocated_fsb; | |
3806 | } | |
3807 | dmapi_enospc_check: | |
eb9df39d CH |
3808 | if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 && |
3809 | DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) { | |
1da177e4 | 3810 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, |
bc4ac74a CH |
3811 | ip, DM_RIGHT_NULL, |
3812 | ip, DM_RIGHT_NULL, | |
1da177e4 LT |
3813 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ |
3814 | if (error == 0) | |
3815 | goto retry; /* Maybe DMAPI app. has made space */ | |
3816 | /* else fall through with error from XFS_SEND_DATA */ | |
3817 | } | |
3818 | ||
3819 | return error; | |
3820 | ||
dd9f438e | 3821 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 3822 | xfs_bmap_cancel(&free_list); |
dd9f438e NS |
3823 | XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag); |
3824 | ||
3825 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
3826 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
3827 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3828 | goto dmapi_enospc_check; | |
3829 | } | |
3830 | ||
3831 | /* | |
3832 | * Zero file bytes between startoff and endoff inclusive. | |
3833 | * The iolock is held exclusive and no blocks are buffered. | |
3834 | */ | |
3835 | STATIC int | |
3836 | xfs_zero_remaining_bytes( | |
3837 | xfs_inode_t *ip, | |
3838 | xfs_off_t startoff, | |
3839 | xfs_off_t endoff) | |
3840 | { | |
3841 | xfs_bmbt_irec_t imap; | |
3842 | xfs_fileoff_t offset_fsb; | |
3843 | xfs_off_t lastoffset; | |
3844 | xfs_off_t offset; | |
3845 | xfs_buf_t *bp; | |
3846 | xfs_mount_t *mp = ip->i_mount; | |
3847 | int nimap; | |
3848 | int error = 0; | |
3849 | ||
3850 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, | |
71ddabb9 | 3851 | XFS_IS_REALTIME_INODE(ip) ? |
1da177e4 LT |
3852 | mp->m_rtdev_targp : mp->m_ddev_targp); |
3853 | ||
3854 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
3855 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
3856 | nimap = 1; | |
541d7d3c | 3857 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, |
3e57ecf6 | 3858 | NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
3859 | if (error || nimap < 1) |
3860 | break; | |
3861 | ASSERT(imap.br_blockcount >= 1); | |
3862 | ASSERT(imap.br_startoff == offset_fsb); | |
3863 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
3864 | if (lastoffset > endoff) | |
3865 | lastoffset = endoff; | |
3866 | if (imap.br_startblock == HOLESTARTBLOCK) | |
3867 | continue; | |
3868 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
3869 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
3870 | continue; | |
3871 | XFS_BUF_UNDONE(bp); | |
3872 | XFS_BUF_UNWRITE(bp); | |
3873 | XFS_BUF_READ(bp); | |
3874 | XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock)); | |
3875 | xfsbdstrat(mp, bp); | |
3876 | if ((error = xfs_iowait(bp))) { | |
3877 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", | |
3878 | mp, bp, XFS_BUF_ADDR(bp)); | |
3879 | break; | |
3880 | } | |
3881 | memset(XFS_BUF_PTR(bp) + | |
3882 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
3883 | 0, lastoffset - offset + 1); | |
3884 | XFS_BUF_UNDONE(bp); | |
3885 | XFS_BUF_UNREAD(bp); | |
3886 | XFS_BUF_WRITE(bp); | |
3887 | xfsbdstrat(mp, bp); | |
3888 | if ((error = xfs_iowait(bp))) { | |
3889 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", | |
3890 | mp, bp, XFS_BUF_ADDR(bp)); | |
3891 | break; | |
3892 | } | |
3893 | } | |
3894 | xfs_buf_free(bp); | |
3895 | return error; | |
3896 | } | |
3897 | ||
3898 | /* | |
3899 | * xfs_free_file_space() | |
3900 | * This routine frees disk space for the given file. | |
3901 | * | |
3902 | * This routine is only called by xfs_change_file_space | |
3903 | * for an UNRESVSP type call. | |
3904 | * | |
3905 | * RETURNS: | |
3906 | * 0 on success | |
3907 | * errno on error | |
3908 | * | |
3909 | */ | |
3910 | STATIC int | |
3911 | xfs_free_file_space( | |
3912 | xfs_inode_t *ip, | |
3913 | xfs_off_t offset, | |
3914 | xfs_off_t len, | |
3915 | int attr_flags) | |
3916 | { | |
67fcaa73 | 3917 | bhv_vnode_t *vp; |
1da177e4 LT |
3918 | int committed; |
3919 | int done; | |
3920 | xfs_off_t end_dmi_offset; | |
3921 | xfs_fileoff_t endoffset_fsb; | |
3922 | int error; | |
3923 | xfs_fsblock_t firstfsb; | |
3924 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
3925 | xfs_bmbt_irec_t imap; |
3926 | xfs_off_t ioffset; | |
3927 | xfs_extlen_t mod=0; | |
3928 | xfs_mount_t *mp; | |
3929 | int nimap; | |
3930 | uint resblks; | |
673cdf5c | 3931 | uint rounding; |
1da177e4 LT |
3932 | int rt; |
3933 | xfs_fileoff_t startoffset_fsb; | |
3934 | xfs_trans_t *tp; | |
5fcbab35 | 3935 | int need_iolock = 1; |
1da177e4 | 3936 | |
bd5a876a | 3937 | vp = XFS_ITOV(ip); |
1da177e4 LT |
3938 | mp = ip->i_mount; |
3939 | ||
cf441eeb | 3940 | xfs_itrace_entry(ip); |
bd5a876a | 3941 | |
1da177e4 LT |
3942 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
3943 | return error; | |
3944 | ||
3945 | error = 0; | |
3946 | if (len <= 0) /* if nothing being freed */ | |
3947 | return error; | |
71ddabb9 | 3948 | rt = XFS_IS_REALTIME_INODE(ip); |
1da177e4 LT |
3949 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); |
3950 | end_dmi_offset = offset + len; | |
3951 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
3952 | ||
eb9df39d CH |
3953 | if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 && |
3954 | DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) { | |
ba87ea69 LM |
3955 | if (end_dmi_offset > ip->i_size) |
3956 | end_dmi_offset = ip->i_size; | |
bc4ac74a | 3957 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, |
1da177e4 LT |
3958 | offset, end_dmi_offset - offset, |
3959 | AT_DELAY_FLAG(attr_flags), NULL); | |
3960 | if (error) | |
014c2544 | 3961 | return error; |
1da177e4 LT |
3962 | } |
3963 | ||
5fcbab35 DR |
3964 | if (attr_flags & ATTR_NOLOCK) |
3965 | need_iolock = 0; | |
9fa8046f | 3966 | if (need_iolock) { |
1da177e4 | 3967 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
b677c210 | 3968 | vn_iowait(ip); /* wait for the completion of any pending DIOs */ |
9fa8046f | 3969 | } |
5fcbab35 | 3970 | |
e6a4b37f | 3971 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE); |
1da177e4 | 3972 | ioffset = offset & ~(rounding - 1); |
bd5a876a CH |
3973 | |
3974 | if (VN_CACHED(vp) != 0) { | |
e6a4b37f TS |
3975 | xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1); |
3976 | error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED); | |
d3cf2094 LM |
3977 | if (error) |
3978 | goto out_unlock_iolock; | |
bd5a876a CH |
3979 | } |
3980 | ||
1da177e4 LT |
3981 | /* |
3982 | * Need to zero the stuff we're not freeing, on disk. | |
3983 | * If its a realtime file & can't use unwritten extents then we | |
3984 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
3985 | * will take care of it for us. | |
3986 | */ | |
62118709 | 3987 | if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
1da177e4 | 3988 | nimap = 1; |
541d7d3c | 3989 | error = xfs_bmapi(NULL, ip, startoffset_fsb, |
3e57ecf6 | 3990 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
3991 | if (error) |
3992 | goto out_unlock_iolock; | |
3993 | ASSERT(nimap == 0 || nimap == 1); | |
3994 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
3995 | xfs_daddr_t block; | |
3996 | ||
3997 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
3998 | block = imap.br_startblock; | |
3999 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
4000 | if (mod) | |
4001 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
4002 | } | |
4003 | nimap = 1; | |
541d7d3c | 4004 | error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, |
3e57ecf6 | 4005 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); |
1da177e4 LT |
4006 | if (error) |
4007 | goto out_unlock_iolock; | |
4008 | ASSERT(nimap == 0 || nimap == 1); | |
4009 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4010 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4011 | mod++; | |
4012 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
4013 | endoffset_fsb -= mod; | |
4014 | } | |
4015 | } | |
4016 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
4017 | /* | |
4018 | * One contiguous piece to clear | |
4019 | */ | |
4020 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
4021 | else { | |
4022 | /* | |
4023 | * Some full blocks, possibly two pieces to clear | |
4024 | */ | |
4025 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
4026 | error = xfs_zero_remaining_bytes(ip, offset, | |
4027 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
4028 | if (!error && | |
4029 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
4030 | error = xfs_zero_remaining_bytes(ip, | |
4031 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
4032 | offset + len - 1); | |
4033 | } | |
4034 | ||
4035 | /* | |
4036 | * free file space until done or until there is an error | |
4037 | */ | |
4038 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
4039 | while (!error && !done) { | |
4040 | ||
4041 | /* | |
91ebecc7 DC |
4042 | * allocate and setup the transaction. Allow this |
4043 | * transaction to dip into the reserve blocks to ensure | |
4044 | * the freeing of the space succeeds at ENOSPC. | |
1da177e4 LT |
4045 | */ |
4046 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
91ebecc7 | 4047 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
4048 | error = xfs_trans_reserve(tp, |
4049 | resblks, | |
4050 | XFS_WRITE_LOG_RES(mp), | |
4051 | 0, | |
4052 | XFS_TRANS_PERM_LOG_RES, | |
4053 | XFS_WRITE_LOG_COUNT); | |
4054 | ||
4055 | /* | |
4056 | * check for running out of space | |
4057 | */ | |
4058 | if (error) { | |
4059 | /* | |
4060 | * Free the transaction structure. | |
4061 | */ | |
4062 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
4063 | xfs_trans_cancel(tp, 0); | |
4064 | break; | |
4065 | } | |
4066 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4067 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, | |
dd9f438e NS |
4068 | ip->i_udquot, ip->i_gdquot, resblks, 0, |
4069 | XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
4070 | if (error) |
4071 | goto error1; | |
4072 | ||
4073 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4074 | xfs_trans_ihold(tp, ip); | |
4075 | ||
4076 | /* | |
4077 | * issue the bunmapi() call to free the blocks | |
4078 | */ | |
4079 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
541d7d3c | 4080 | error = xfs_bunmapi(tp, ip, startoffset_fsb, |
1da177e4 | 4081 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 4082 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
4083 | if (error) { |
4084 | goto error0; | |
4085 | } | |
4086 | ||
4087 | /* | |
4088 | * complete the transaction | |
4089 | */ | |
f7c99b6f | 4090 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
4091 | if (error) { |
4092 | goto error0; | |
4093 | } | |
4094 | ||
1c72bf90 | 4095 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
4096 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
4097 | } | |
4098 | ||
4099 | out_unlock_iolock: | |
4100 | if (need_iolock) | |
4101 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
4102 | return error; | |
4103 | ||
4104 | error0: | |
4105 | xfs_bmap_cancel(&free_list); | |
4106 | error1: | |
4107 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
4108 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
4109 | XFS_ILOCK_EXCL); | |
4110 | return error; | |
4111 | } | |
4112 | ||
4113 | /* | |
4114 | * xfs_change_file_space() | |
4115 | * This routine allocates or frees disk space for the given file. | |
4116 | * The user specified parameters are checked for alignment and size | |
4117 | * limitations. | |
4118 | * | |
4119 | * RETURNS: | |
4120 | * 0 on success | |
4121 | * errno on error | |
4122 | * | |
4123 | */ | |
4124 | int | |
4125 | xfs_change_file_space( | |
993386c1 | 4126 | xfs_inode_t *ip, |
1da177e4 LT |
4127 | int cmd, |
4128 | xfs_flock64_t *bf, | |
4129 | xfs_off_t offset, | |
4130 | cred_t *credp, | |
4131 | int attr_flags) | |
4132 | { | |
993386c1 | 4133 | xfs_mount_t *mp = ip->i_mount; |
1da177e4 LT |
4134 | int clrprealloc; |
4135 | int error; | |
4136 | xfs_fsize_t fsize; | |
1da177e4 LT |
4137 | int setprealloc; |
4138 | xfs_off_t startoffset; | |
4139 | xfs_off_t llen; | |
4140 | xfs_trans_t *tp; | |
8285fb58 | 4141 | bhv_vattr_t va; |
1da177e4 | 4142 | |
cf441eeb | 4143 | xfs_itrace_entry(ip); |
1da177e4 | 4144 | |
993386c1 | 4145 | if (!S_ISREG(ip->i_d.di_mode)) |
1da177e4 LT |
4146 | return XFS_ERROR(EINVAL); |
4147 | ||
1da177e4 LT |
4148 | switch (bf->l_whence) { |
4149 | case 0: /*SEEK_SET*/ | |
4150 | break; | |
4151 | case 1: /*SEEK_CUR*/ | |
4152 | bf->l_start += offset; | |
4153 | break; | |
4154 | case 2: /*SEEK_END*/ | |
ba87ea69 | 4155 | bf->l_start += ip->i_size; |
1da177e4 LT |
4156 | break; |
4157 | default: | |
4158 | return XFS_ERROR(EINVAL); | |
4159 | } | |
4160 | ||
4161 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
4162 | ||
4163 | if ( (bf->l_start < 0) | |
4164 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
4165 | || (bf->l_start + llen < 0) | |
4166 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
4167 | return XFS_ERROR(EINVAL); | |
4168 | ||
4169 | bf->l_whence = 0; | |
4170 | ||
4171 | startoffset = bf->l_start; | |
ba87ea69 | 4172 | fsize = ip->i_size; |
1da177e4 LT |
4173 | |
4174 | /* | |
4175 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
4176 | * file space. | |
4177 | * These calls do NOT zero the data space allocated to the file, | |
4178 | * nor do they change the file size. | |
4179 | * | |
4180 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
4181 | * space. | |
4182 | * These calls cause the new file data to be zeroed and the file | |
4183 | * size to be changed. | |
4184 | */ | |
4185 | setprealloc = clrprealloc = 0; | |
4186 | ||
4187 | switch (cmd) { | |
4188 | case XFS_IOC_RESVSP: | |
4189 | case XFS_IOC_RESVSP64: | |
4190 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
4191 | 1, attr_flags); | |
4192 | if (error) | |
4193 | return error; | |
4194 | setprealloc = 1; | |
4195 | break; | |
4196 | ||
4197 | case XFS_IOC_UNRESVSP: | |
4198 | case XFS_IOC_UNRESVSP64: | |
4199 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
4200 | attr_flags))) | |
4201 | return error; | |
4202 | break; | |
4203 | ||
4204 | case XFS_IOC_ALLOCSP: | |
4205 | case XFS_IOC_ALLOCSP64: | |
4206 | case XFS_IOC_FREESP: | |
4207 | case XFS_IOC_FREESP64: | |
4208 | if (startoffset > fsize) { | |
4209 | error = xfs_alloc_file_space(ip, fsize, | |
4210 | startoffset - fsize, 0, attr_flags); | |
4211 | if (error) | |
4212 | break; | |
4213 | } | |
4214 | ||
4215 | va.va_mask = XFS_AT_SIZE; | |
4216 | va.va_size = startoffset; | |
4217 | ||
993386c1 | 4218 | error = xfs_setattr(ip, &va, attr_flags, credp); |
1da177e4 LT |
4219 | |
4220 | if (error) | |
4221 | return error; | |
4222 | ||
4223 | clrprealloc = 1; | |
4224 | break; | |
4225 | ||
4226 | default: | |
4227 | ASSERT(0); | |
4228 | return XFS_ERROR(EINVAL); | |
4229 | } | |
4230 | ||
4231 | /* | |
4232 | * update the inode timestamp, mode, and prealloc flag bits | |
4233 | */ | |
4234 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
4235 | ||
4236 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
4237 | 0, 0, 0))) { | |
4238 | /* ASSERT(0); */ | |
4239 | xfs_trans_cancel(tp, 0); | |
4240 | return error; | |
4241 | } | |
4242 | ||
4243 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4244 | ||
4245 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4246 | xfs_trans_ihold(tp, ip); | |
4247 | ||
4248 | if ((attr_flags & ATTR_DMI) == 0) { | |
4249 | ip->i_d.di_mode &= ~S_ISUID; | |
4250 | ||
4251 | /* | |
4252 | * Note that we don't have to worry about mandatory | |
4253 | * file locking being disabled here because we only | |
4254 | * clear the S_ISGID bit if the Group execute bit is | |
4255 | * on, but if it was on then mandatory locking wouldn't | |
4256 | * have been enabled. | |
4257 | */ | |
4258 | if (ip->i_d.di_mode & S_IXGRP) | |
4259 | ip->i_d.di_mode &= ~S_ISGID; | |
4260 | ||
4261 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
4262 | } | |
4263 | if (setprealloc) | |
4264 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
4265 | else if (clrprealloc) | |
4266 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
4267 | ||
4268 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
4269 | xfs_trans_set_sync(tp); | |
4270 | ||
1c72bf90 | 4271 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
4272 | |
4273 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4274 | ||
4275 | return error; | |
4276 | } |