]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
16f7e0fe | 18 | |
1da177e4 | 19 | #include "xfs.h" |
a844f451 | 20 | #include "xfs_fs.h" |
1da177e4 | 21 | #include "xfs_types.h" |
a844f451 | 22 | #include "xfs_bit.h" |
1da177e4 | 23 | #include "xfs_log.h" |
a844f451 | 24 | #include "xfs_inum.h" |
1da177e4 LT |
25 | #include "xfs_trans.h" |
26 | #include "xfs_sb.h" | |
27 | #include "xfs_ag.h" | |
1da177e4 LT |
28 | #include "xfs_dir2.h" |
29 | #include "xfs_dmapi.h" | |
30 | #include "xfs_mount.h" | |
a844f451 | 31 | #include "xfs_da_btree.h" |
1da177e4 | 32 | #include "xfs_bmap_btree.h" |
a844f451 | 33 | #include "xfs_alloc_btree.h" |
1da177e4 | 34 | #include "xfs_ialloc_btree.h" |
1da177e4 | 35 | #include "xfs_dir2_sf.h" |
a844f451 | 36 | #include "xfs_attr_sf.h" |
1da177e4 | 37 | #include "xfs_dinode.h" |
1da177e4 | 38 | #include "xfs_inode.h" |
a844f451 | 39 | #include "xfs_inode_item.h" |
a844f451 NS |
40 | #include "xfs_itable.h" |
41 | #include "xfs_btree.h" | |
42 | #include "xfs_ialloc.h" | |
43 | #include "xfs_alloc.h" | |
1da177e4 | 44 | #include "xfs_bmap.h" |
1da177e4 LT |
45 | #include "xfs_attr.h" |
46 | #include "xfs_rw.h" | |
1da177e4 | 47 | #include "xfs_error.h" |
1da177e4 LT |
48 | #include "xfs_quota.h" |
49 | #include "xfs_utils.h" | |
a844f451 NS |
50 | #include "xfs_rtalloc.h" |
51 | #include "xfs_refcache.h" | |
1da177e4 | 52 | #include "xfs_trans_space.h" |
1da177e4 LT |
53 | #include "xfs_log_priv.h" |
54 | ||
1da177e4 LT |
55 | STATIC int |
56 | xfs_open( | |
57 | bhv_desc_t *bdp, | |
58 | cred_t *credp) | |
59 | { | |
60 | int mode; | |
67fcaa73 | 61 | bhv_vnode_t *vp = BHV_TO_VNODE(bdp); |
7d4fb40a | 62 | xfs_inode_t *ip = XFS_BHVTOI(bdp); |
1da177e4 LT |
63 | |
64 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
65 | return XFS_ERROR(EIO); | |
66 | ||
67 | /* | |
68 | * If it's a directory with any blocks, read-ahead block 0 | |
69 | * as we're almost certain to have the next operation be a read there. | |
70 | */ | |
0432dab2 | 71 | if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) { |
1da177e4 LT |
72 | mode = xfs_ilock_map_shared(ip); |
73 | if (ip->i_d.di_nextents > 0) | |
74 | (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK); | |
75 | xfs_iunlock(ip, mode); | |
76 | } | |
77 | return 0; | |
78 | } | |
79 | ||
7d4fb40a NS |
80 | STATIC int |
81 | xfs_close( | |
82 | bhv_desc_t *bdp, | |
83 | int flags, | |
84 | lastclose_t lastclose, | |
85 | cred_t *credp) | |
86 | { | |
67fcaa73 | 87 | bhv_vnode_t *vp = BHV_TO_VNODE(bdp); |
7d4fb40a | 88 | xfs_inode_t *ip = XFS_BHVTOI(bdp); |
7d4fb40a NS |
89 | |
90 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
91 | return XFS_ERROR(EIO); | |
92 | ||
93 | if (lastclose != L_TRUE || !VN_ISREG(vp)) | |
94 | return 0; | |
95 | ||
96 | /* | |
97 | * If we previously truncated this file and removed old data in | |
98 | * the process, we want to initiate "early" writeout on the last | |
99 | * close. This is an attempt to combat the notorious NULL files | |
100 | * problem which is particularly noticable from a truncate down, | |
101 | * buffered (re-)write (delalloc), followed by a crash. What we | |
102 | * are effectively doing here is significantly reducing the time | |
103 | * window where we'd otherwise be exposed to that problem. | |
104 | */ | |
105 | if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0) | |
67fcaa73 NS |
106 | return bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE); |
107 | return 0; | |
7d4fb40a | 108 | } |
1da177e4 LT |
109 | |
110 | /* | |
111 | * xfs_getattr | |
112 | */ | |
113 | STATIC int | |
114 | xfs_getattr( | |
115 | bhv_desc_t *bdp, | |
8285fb58 | 116 | bhv_vattr_t *vap, |
1da177e4 LT |
117 | int flags, |
118 | cred_t *credp) | |
119 | { | |
120 | xfs_inode_t *ip; | |
121 | xfs_mount_t *mp; | |
67fcaa73 | 122 | bhv_vnode_t *vp; |
1da177e4 LT |
123 | |
124 | vp = BHV_TO_VNODE(bdp); | |
125 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
126 | ||
127 | ip = XFS_BHVTOI(bdp); | |
128 | mp = ip->i_mount; | |
129 | ||
130 | if (XFS_FORCED_SHUTDOWN(mp)) | |
131 | return XFS_ERROR(EIO); | |
132 | ||
133 | if (!(flags & ATTR_LAZY)) | |
134 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
135 | ||
136 | vap->va_size = ip->i_d.di_size; | |
137 | if (vap->va_mask == XFS_AT_SIZE) | |
138 | goto all_done; | |
139 | ||
140 | vap->va_nblocks = | |
141 | XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); | |
142 | vap->va_nodeid = ip->i_ino; | |
143 | #if XFS_BIG_INUMS | |
144 | vap->va_nodeid += mp->m_inoadd; | |
145 | #endif | |
146 | vap->va_nlink = ip->i_d.di_nlink; | |
147 | ||
148 | /* | |
149 | * Quick exit for non-stat callers | |
150 | */ | |
151 | if ((vap->va_mask & | |
152 | ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID| | |
153 | XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0) | |
154 | goto all_done; | |
155 | ||
156 | /* | |
157 | * Copy from in-core inode. | |
158 | */ | |
0432dab2 | 159 | vap->va_mode = ip->i_d.di_mode; |
1da177e4 LT |
160 | vap->va_uid = ip->i_d.di_uid; |
161 | vap->va_gid = ip->i_d.di_gid; | |
162 | vap->va_projid = ip->i_d.di_projid; | |
163 | ||
164 | /* | |
165 | * Check vnode type block/char vs. everything else. | |
1da177e4 | 166 | */ |
0432dab2 CH |
167 | switch (ip->i_d.di_mode & S_IFMT) { |
168 | case S_IFBLK: | |
169 | case S_IFCHR: | |
170 | vap->va_rdev = ip->i_df.if_u2.if_rdev; | |
171 | vap->va_blocksize = BLKDEV_IOSIZE; | |
172 | break; | |
173 | default: | |
1da177e4 LT |
174 | vap->va_rdev = 0; |
175 | ||
176 | if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) { | |
e8c8b3a7 | 177 | vap->va_blocksize = xfs_preferred_iosize(mp); |
1da177e4 LT |
178 | } else { |
179 | ||
180 | /* | |
181 | * If the file blocks are being allocated from a | |
182 | * realtime partition, then return the inode's | |
183 | * realtime extent size or the realtime volume's | |
184 | * extent size. | |
185 | */ | |
186 | vap->va_blocksize = ip->i_d.di_extsize ? | |
187 | (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) : | |
188 | (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog); | |
189 | } | |
0432dab2 | 190 | break; |
1da177e4 LT |
191 | } |
192 | ||
ca5ccbf9 | 193 | vn_atime_to_timespec(vp, &vap->va_atime); |
1da177e4 LT |
194 | vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec; |
195 | vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec; | |
196 | vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec; | |
197 | vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec; | |
198 | ||
199 | /* | |
200 | * Exit for stat callers. See if any of the rest of the fields | |
201 | * to be filled in are needed. | |
202 | */ | |
203 | if ((vap->va_mask & | |
204 | (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS| | |
205 | XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0) | |
206 | goto all_done; | |
207 | ||
208 | /* | |
209 | * Convert di_flags to xflags. | |
210 | */ | |
211 | vap->va_xflags = xfs_ip2xflags(ip); | |
212 | ||
213 | /* | |
214 | * Exit for inode revalidate. See if any of the rest of | |
215 | * the fields to be filled in are needed. | |
216 | */ | |
217 | if ((vap->va_mask & | |
218 | (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS| | |
219 | XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0) | |
220 | goto all_done; | |
221 | ||
222 | vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog; | |
223 | vap->va_nextents = | |
224 | (ip->i_df.if_flags & XFS_IFEXTENTS) ? | |
225 | ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) : | |
226 | ip->i_d.di_nextents; | |
227 | if (ip->i_afp) | |
228 | vap->va_anextents = | |
229 | (ip->i_afp->if_flags & XFS_IFEXTENTS) ? | |
230 | ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) : | |
231 | ip->i_d.di_anextents; | |
232 | else | |
233 | vap->va_anextents = 0; | |
234 | vap->va_gen = ip->i_d.di_gen; | |
235 | ||
236 | all_done: | |
237 | if (!(flags & ATTR_LAZY)) | |
238 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
239 | return 0; | |
240 | } | |
241 | ||
242 | ||
243 | /* | |
244 | * xfs_setattr | |
245 | */ | |
246 | int | |
247 | xfs_setattr( | |
248 | bhv_desc_t *bdp, | |
8285fb58 | 249 | bhv_vattr_t *vap, |
1da177e4 LT |
250 | int flags, |
251 | cred_t *credp) | |
252 | { | |
253 | xfs_inode_t *ip; | |
254 | xfs_trans_t *tp; | |
255 | xfs_mount_t *mp; | |
256 | int mask; | |
257 | int code; | |
258 | uint lock_flags; | |
259 | uint commit_flags=0; | |
260 | uid_t uid=0, iuid=0; | |
261 | gid_t gid=0, igid=0; | |
262 | int timeflags = 0; | |
67fcaa73 | 263 | bhv_vnode_t *vp; |
1da177e4 LT |
264 | xfs_prid_t projid=0, iprojid=0; |
265 | int mandlock_before, mandlock_after; | |
266 | struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2; | |
267 | int file_owner; | |
5fcbab35 | 268 | int need_iolock = 1; |
1da177e4 LT |
269 | |
270 | vp = BHV_TO_VNODE(bdp); | |
271 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
272 | ||
273 | if (vp->v_vfsp->vfs_flag & VFS_RDONLY) | |
274 | return XFS_ERROR(EROFS); | |
275 | ||
276 | /* | |
277 | * Cannot set certain attributes. | |
278 | */ | |
279 | mask = vap->va_mask; | |
280 | if (mask & XFS_AT_NOSET) { | |
281 | return XFS_ERROR(EINVAL); | |
282 | } | |
283 | ||
284 | ip = XFS_BHVTOI(bdp); | |
285 | mp = ip->i_mount; | |
286 | ||
287 | if (XFS_FORCED_SHUTDOWN(mp)) | |
288 | return XFS_ERROR(EIO); | |
289 | ||
290 | /* | |
291 | * Timestamps do not need to be logged and hence do not | |
292 | * need to be done within a transaction. | |
293 | */ | |
294 | if (mask & XFS_AT_UPDTIMES) { | |
295 | ASSERT((mask & ~XFS_AT_UPDTIMES) == 0); | |
296 | timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) | | |
297 | ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) | | |
298 | ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0); | |
299 | xfs_ichgtime(ip, timeflags); | |
300 | return 0; | |
301 | } | |
302 | ||
303 | olddquot1 = olddquot2 = NULL; | |
304 | udqp = gdqp = NULL; | |
305 | ||
306 | /* | |
307 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
308 | * before we start any other transactions. Trying to do this later | |
309 | * is messy. We don't care to take a readlock to look at the ids | |
310 | * in inode here, because we can't hold it across the trans_reserve. | |
311 | * If the IDs do change before we take the ilock, we're covered | |
312 | * because the i_*dquot fields will get updated anyway. | |
313 | */ | |
c8ad20ff NS |
314 | if (XFS_IS_QUOTA_ON(mp) && |
315 | (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) { | |
1da177e4 LT |
316 | uint qflags = 0; |
317 | ||
c8ad20ff | 318 | if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) { |
1da177e4 LT |
319 | uid = vap->va_uid; |
320 | qflags |= XFS_QMOPT_UQUOTA; | |
321 | } else { | |
322 | uid = ip->i_d.di_uid; | |
323 | } | |
c8ad20ff | 324 | if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) { |
1da177e4 LT |
325 | gid = vap->va_gid; |
326 | qflags |= XFS_QMOPT_GQUOTA; | |
327 | } else { | |
328 | gid = ip->i_d.di_gid; | |
329 | } | |
c8ad20ff NS |
330 | if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) { |
331 | projid = vap->va_projid; | |
332 | qflags |= XFS_QMOPT_PQUOTA; | |
333 | } else { | |
334 | projid = ip->i_d.di_projid; | |
335 | } | |
1da177e4 LT |
336 | /* |
337 | * We take a reference when we initialize udqp and gdqp, | |
338 | * so it is important that we never blindly double trip on | |
339 | * the same variable. See xfs_create() for an example. | |
340 | */ | |
341 | ASSERT(udqp == NULL); | |
342 | ASSERT(gdqp == NULL); | |
c8ad20ff NS |
343 | code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags, |
344 | &udqp, &gdqp); | |
1da177e4 | 345 | if (code) |
014c2544 | 346 | return code; |
1da177e4 LT |
347 | } |
348 | ||
349 | /* | |
350 | * For the other attributes, we acquire the inode lock and | |
351 | * first do an error checking pass. | |
352 | */ | |
353 | tp = NULL; | |
354 | lock_flags = XFS_ILOCK_EXCL; | |
5fcbab35 DR |
355 | if (flags & ATTR_NOLOCK) |
356 | need_iolock = 0; | |
1da177e4 LT |
357 | if (!(mask & XFS_AT_SIZE)) { |
358 | if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) || | |
359 | (mp->m_flags & XFS_MOUNT_WSYNC)) { | |
360 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); | |
361 | commit_flags = 0; | |
362 | if ((code = xfs_trans_reserve(tp, 0, | |
363 | XFS_ICHANGE_LOG_RES(mp), 0, | |
364 | 0, 0))) { | |
365 | lock_flags = 0; | |
366 | goto error_return; | |
367 | } | |
368 | } | |
369 | } else { | |
370 | if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) && | |
371 | !(flags & ATTR_DMI)) { | |
372 | int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR; | |
373 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp, | |
374 | vap->va_size, 0, dmflags, NULL); | |
375 | if (code) { | |
376 | lock_flags = 0; | |
377 | goto error_return; | |
378 | } | |
379 | } | |
380 | if (need_iolock) | |
381 | lock_flags |= XFS_IOLOCK_EXCL; | |
382 | } | |
383 | ||
384 | xfs_ilock(ip, lock_flags); | |
385 | ||
386 | /* boolean: are we the file owner? */ | |
387 | file_owner = (current_fsuid(credp) == ip->i_d.di_uid); | |
388 | ||
389 | /* | |
390 | * Change various properties of a file. | |
391 | * Only the owner or users with CAP_FOWNER | |
392 | * capability may do these things. | |
393 | */ | |
394 | if (mask & | |
395 | (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID| | |
396 | XFS_AT_GID|XFS_AT_PROJID)) { | |
397 | /* | |
398 | * CAP_FOWNER overrides the following restrictions: | |
399 | * | |
400 | * The user ID of the calling process must be equal | |
401 | * to the file owner ID, except in cases where the | |
402 | * CAP_FSETID capability is applicable. | |
403 | */ | |
404 | if (!file_owner && !capable(CAP_FOWNER)) { | |
405 | code = XFS_ERROR(EPERM); | |
406 | goto error_return; | |
407 | } | |
408 | ||
409 | /* | |
410 | * CAP_FSETID overrides the following restrictions: | |
411 | * | |
412 | * The effective user ID of the calling process shall match | |
413 | * the file owner when setting the set-user-ID and | |
414 | * set-group-ID bits on that file. | |
415 | * | |
416 | * The effective group ID or one of the supplementary group | |
417 | * IDs of the calling process shall match the group owner of | |
418 | * the file when setting the set-group-ID bit on that file | |
419 | */ | |
420 | if (mask & XFS_AT_MODE) { | |
421 | mode_t m = 0; | |
422 | ||
423 | if ((vap->va_mode & S_ISUID) && !file_owner) | |
424 | m |= S_ISUID; | |
425 | if ((vap->va_mode & S_ISGID) && | |
426 | !in_group_p((gid_t)ip->i_d.di_gid)) | |
427 | m |= S_ISGID; | |
428 | #if 0 | |
429 | /* Linux allows this, Irix doesn't. */ | |
0432dab2 | 430 | if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp)) |
1da177e4 LT |
431 | m |= S_ISVTX; |
432 | #endif | |
433 | if (m && !capable(CAP_FSETID)) | |
434 | vap->va_mode &= ~m; | |
435 | } | |
436 | } | |
437 | ||
438 | /* | |
439 | * Change file ownership. Must be the owner or privileged. | |
440 | * If the system was configured with the "restricted_chown" | |
441 | * option, the owner is not permitted to give away the file, | |
442 | * and can change the group id only to a group of which he | |
443 | * or she is a member. | |
444 | */ | |
445 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
446 | /* | |
447 | * These IDs could have changed since we last looked at them. | |
448 | * But, we're assured that if the ownership did change | |
449 | * while we didn't have the inode locked, inode's dquot(s) | |
450 | * would have changed also. | |
451 | */ | |
452 | iuid = ip->i_d.di_uid; | |
453 | iprojid = ip->i_d.di_projid; | |
454 | igid = ip->i_d.di_gid; | |
455 | gid = (mask & XFS_AT_GID) ? vap->va_gid : igid; | |
456 | uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid; | |
457 | projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid : | |
458 | iprojid; | |
459 | ||
460 | /* | |
461 | * CAP_CHOWN overrides the following restrictions: | |
462 | * | |
463 | * If _POSIX_CHOWN_RESTRICTED is defined, this capability | |
464 | * shall override the restriction that a process cannot | |
465 | * change the user ID of a file it owns and the restriction | |
466 | * that the group ID supplied to the chown() function | |
467 | * shall be equal to either the group ID or one of the | |
468 | * supplementary group IDs of the calling process. | |
1da177e4 LT |
469 | */ |
470 | if (restricted_chown && | |
471 | (iuid != uid || (igid != gid && | |
472 | !in_group_p((gid_t)gid))) && | |
473 | !capable(CAP_CHOWN)) { | |
474 | code = XFS_ERROR(EPERM); | |
475 | goto error_return; | |
476 | } | |
477 | /* | |
c8ad20ff | 478 | * Do a quota reservation only if uid/projid/gid is actually |
1da177e4 LT |
479 | * going to change. |
480 | */ | |
481 | if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || | |
c8ad20ff | 482 | (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) || |
1da177e4 LT |
483 | (XFS_IS_GQUOTA_ON(mp) && igid != gid)) { |
484 | ASSERT(tp); | |
485 | code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp, | |
486 | capable(CAP_FOWNER) ? | |
487 | XFS_QMOPT_FORCE_RES : 0); | |
488 | if (code) /* out of quota */ | |
489 | goto error_return; | |
490 | } | |
491 | } | |
492 | ||
493 | /* | |
494 | * Truncate file. Must have write permission and not be a directory. | |
495 | */ | |
496 | if (mask & XFS_AT_SIZE) { | |
497 | /* Short circuit the truncate case for zero length files */ | |
498 | if ((vap->va_size == 0) && | |
499 | (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) { | |
500 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
501 | lock_flags &= ~XFS_ILOCK_EXCL; | |
502 | if (mask & XFS_AT_CTIME) | |
503 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
504 | code = 0; | |
505 | goto error_return; | |
506 | } | |
507 | ||
0432dab2 | 508 | if (VN_ISDIR(vp)) { |
1da177e4 LT |
509 | code = XFS_ERROR(EISDIR); |
510 | goto error_return; | |
0432dab2 | 511 | } else if (!VN_ISREG(vp)) { |
1da177e4 LT |
512 | code = XFS_ERROR(EINVAL); |
513 | goto error_return; | |
514 | } | |
515 | /* | |
516 | * Make sure that the dquots are attached to the inode. | |
517 | */ | |
518 | if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED))) | |
519 | goto error_return; | |
520 | } | |
521 | ||
522 | /* | |
523 | * Change file access or modified times. | |
524 | */ | |
525 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
526 | if (!file_owner) { | |
527 | if ((flags & ATTR_UTIME) && | |
528 | !capable(CAP_FOWNER)) { | |
529 | code = XFS_ERROR(EPERM); | |
530 | goto error_return; | |
531 | } | |
532 | } | |
533 | } | |
534 | ||
535 | /* | |
536 | * Change extent size or realtime flag. | |
537 | */ | |
538 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
539 | /* | |
540 | * Can't change extent size if any extents are allocated. | |
541 | */ | |
e94af02a | 542 | if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) && |
1da177e4 LT |
543 | ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != |
544 | vap->va_extsize) ) { | |
545 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
546 | goto error_return; | |
547 | } | |
548 | ||
1da177e4 LT |
549 | /* |
550 | * Can't change realtime flag if any extents are allocated. | |
551 | */ | |
e94af02a ES |
552 | if ((ip->i_d.di_nextents || ip->i_delayed_blks) && |
553 | (mask & XFS_AT_XFLAGS) && | |
1da177e4 LT |
554 | (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != |
555 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { | |
556 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
557 | goto error_return; | |
558 | } | |
559 | /* | |
560 | * Extent size must be a multiple of the appropriate block | |
561 | * size, if set at all. | |
562 | */ | |
563 | if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) { | |
564 | xfs_extlen_t size; | |
565 | ||
566 | if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) || | |
567 | ((mask & XFS_AT_XFLAGS) && | |
568 | (vap->va_xflags & XFS_XFLAG_REALTIME))) { | |
569 | size = mp->m_sb.sb_rextsize << | |
570 | mp->m_sb.sb_blocklog; | |
571 | } else { | |
572 | size = mp->m_sb.sb_blocksize; | |
573 | } | |
574 | if (vap->va_extsize % size) { | |
575 | code = XFS_ERROR(EINVAL); | |
576 | goto error_return; | |
577 | } | |
578 | } | |
579 | /* | |
580 | * If realtime flag is set then must have realtime data. | |
581 | */ | |
582 | if ((mask & XFS_AT_XFLAGS) && | |
583 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { | |
584 | if ((mp->m_sb.sb_rblocks == 0) || | |
585 | (mp->m_sb.sb_rextsize == 0) || | |
586 | (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) { | |
587 | code = XFS_ERROR(EINVAL); | |
588 | goto error_return; | |
589 | } | |
590 | } | |
591 | ||
592 | /* | |
593 | * Can't modify an immutable/append-only file unless | |
594 | * we have appropriate permission. | |
595 | */ | |
596 | if ((mask & XFS_AT_XFLAGS) && | |
597 | (ip->i_d.di_flags & | |
598 | (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) || | |
599 | (vap->va_xflags & | |
600 | (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) && | |
601 | !capable(CAP_LINUX_IMMUTABLE)) { | |
602 | code = XFS_ERROR(EPERM); | |
603 | goto error_return; | |
604 | } | |
605 | } | |
606 | ||
607 | /* | |
608 | * Now we can make the changes. Before we join the inode | |
609 | * to the transaction, if XFS_AT_SIZE is set then take care of | |
610 | * the part of the truncation that must be done without the | |
611 | * inode lock. This needs to be done before joining the inode | |
612 | * to the transaction, because the inode cannot be unlocked | |
613 | * once it is a part of the transaction. | |
614 | */ | |
615 | if (mask & XFS_AT_SIZE) { | |
616 | code = 0; | |
374e2ac3 ES |
617 | if ((vap->va_size > ip->i_d.di_size) && |
618 | (flags & ATTR_NOSIZETOK) == 0) { | |
1da177e4 | 619 | code = xfs_igrow_start(ip, vap->va_size, credp); |
374e2ac3 | 620 | } |
1da177e4 | 621 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
9fa8046f | 622 | vn_iowait(vp); /* wait for the completion of any pending DIOs */ |
1da177e4 LT |
623 | if (!code) |
624 | code = xfs_itruncate_data(ip, vap->va_size); | |
625 | if (code) { | |
626 | ASSERT(tp == NULL); | |
627 | lock_flags &= ~XFS_ILOCK_EXCL; | |
628 | ASSERT(lock_flags == XFS_IOLOCK_EXCL); | |
629 | goto error_return; | |
630 | } | |
631 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); | |
632 | if ((code = xfs_trans_reserve(tp, 0, | |
633 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
634 | XFS_TRANS_PERM_LOG_RES, | |
635 | XFS_ITRUNCATE_LOG_COUNT))) { | |
636 | xfs_trans_cancel(tp, 0); | |
637 | if (need_iolock) | |
638 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
639 | return code; | |
640 | } | |
641 | commit_flags = XFS_TRANS_RELEASE_LOG_RES; | |
642 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
643 | } | |
644 | ||
645 | if (tp) { | |
646 | xfs_trans_ijoin(tp, ip, lock_flags); | |
647 | xfs_trans_ihold(tp, ip); | |
648 | } | |
649 | ||
650 | /* determine whether mandatory locking mode changes */ | |
651 | mandlock_before = MANDLOCK(vp, ip->i_d.di_mode); | |
652 | ||
653 | /* | |
654 | * Truncate file. Must have write permission and not be a directory. | |
655 | */ | |
656 | if (mask & XFS_AT_SIZE) { | |
657 | if (vap->va_size > ip->i_d.di_size) { | |
658 | xfs_igrow_finish(tp, ip, vap->va_size, | |
659 | !(flags & ATTR_DMI)); | |
660 | } else if ((vap->va_size <= ip->i_d.di_size) || | |
661 | ((vap->va_size == 0) && ip->i_d.di_nextents)) { | |
662 | /* | |
663 | * signal a sync transaction unless | |
664 | * we're truncating an already unlinked | |
665 | * file on a wsync filesystem | |
666 | */ | |
667 | code = xfs_itruncate_finish(&tp, ip, | |
668 | (xfs_fsize_t)vap->va_size, | |
669 | XFS_DATA_FORK, | |
670 | ((ip->i_d.di_nlink != 0 || | |
671 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
672 | ? 1 : 0)); | |
7d4fb40a | 673 | if (code) |
1da177e4 | 674 | goto abort_return; |
7d4fb40a NS |
675 | /* |
676 | * Truncated "down", so we're removing references | |
677 | * to old data here - if we now delay flushing for | |
678 | * a long time, we expose ourselves unduly to the | |
679 | * notorious NULL files problem. So, we mark this | |
680 | * vnode and flush it when the file is closed, and | |
681 | * do not wait the usual (long) time for writeout. | |
682 | */ | |
683 | VTRUNCATE(vp); | |
1da177e4 LT |
684 | } |
685 | /* | |
686 | * Have to do this even if the file's size doesn't change. | |
687 | */ | |
688 | timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
689 | } | |
690 | ||
691 | /* | |
692 | * Change file access modes. | |
693 | */ | |
694 | if (mask & XFS_AT_MODE) { | |
695 | ip->i_d.di_mode &= S_IFMT; | |
696 | ip->i_d.di_mode |= vap->va_mode & ~S_IFMT; | |
697 | ||
698 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
699 | timeflags |= XFS_ICHGTIME_CHG; | |
700 | } | |
701 | ||
702 | /* | |
703 | * Change file ownership. Must be the owner or privileged. | |
704 | * If the system was configured with the "restricted_chown" | |
705 | * option, the owner is not permitted to give away the file, | |
706 | * and can change the group id only to a group of which he | |
707 | * or she is a member. | |
708 | */ | |
709 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
710 | /* | |
711 | * CAP_FSETID overrides the following restrictions: | |
712 | * | |
713 | * The set-user-ID and set-group-ID bits of a file will be | |
714 | * cleared upon successful return from chown() | |
715 | */ | |
716 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
717 | !capable(CAP_FSETID)) { | |
718 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
719 | } | |
720 | ||
721 | /* | |
722 | * Change the ownerships and register quota modifications | |
723 | * in the transaction. | |
724 | */ | |
725 | if (iuid != uid) { | |
726 | if (XFS_IS_UQUOTA_ON(mp)) { | |
727 | ASSERT(mask & XFS_AT_UID); | |
728 | ASSERT(udqp); | |
729 | olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
730 | &ip->i_udquot, udqp); | |
731 | } | |
732 | ip->i_d.di_uid = uid; | |
733 | } | |
734 | if (igid != gid) { | |
735 | if (XFS_IS_GQUOTA_ON(mp)) { | |
c8ad20ff | 736 | ASSERT(!XFS_IS_PQUOTA_ON(mp)); |
1da177e4 LT |
737 | ASSERT(mask & XFS_AT_GID); |
738 | ASSERT(gdqp); | |
739 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
740 | &ip->i_gdquot, gdqp); | |
741 | } | |
742 | ip->i_d.di_gid = gid; | |
743 | } | |
744 | if (iprojid != projid) { | |
c8ad20ff NS |
745 | if (XFS_IS_PQUOTA_ON(mp)) { |
746 | ASSERT(!XFS_IS_GQUOTA_ON(mp)); | |
747 | ASSERT(mask & XFS_AT_PROJID); | |
748 | ASSERT(gdqp); | |
749 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
750 | &ip->i_gdquot, gdqp); | |
751 | } | |
1da177e4 LT |
752 | ip->i_d.di_projid = projid; |
753 | /* | |
754 | * We may have to rev the inode as well as | |
755 | * the superblock version number since projids didn't | |
756 | * exist before DINODE_VERSION_2 and SB_VERSION_NLINK. | |
757 | */ | |
758 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) | |
759 | xfs_bump_ino_vers2(tp, ip); | |
760 | } | |
761 | ||
762 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
763 | timeflags |= XFS_ICHGTIME_CHG; | |
764 | } | |
765 | ||
766 | ||
767 | /* | |
768 | * Change file access or modified times. | |
769 | */ | |
770 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
771 | if (mask & XFS_AT_ATIME) { | |
772 | ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec; | |
773 | ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec; | |
774 | ip->i_update_core = 1; | |
775 | timeflags &= ~XFS_ICHGTIME_ACC; | |
776 | } | |
777 | if (mask & XFS_AT_MTIME) { | |
778 | ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec; | |
779 | ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec; | |
780 | timeflags &= ~XFS_ICHGTIME_MOD; | |
781 | timeflags |= XFS_ICHGTIME_CHG; | |
782 | } | |
783 | if (tp && (flags & ATTR_UTIME)) | |
784 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
785 | } | |
786 | ||
787 | /* | |
788 | * Change XFS-added attributes. | |
789 | */ | |
790 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
791 | if (mask & XFS_AT_EXTSIZE) { | |
792 | /* | |
793 | * Converting bytes to fs blocks. | |
794 | */ | |
795 | ip->i_d.di_extsize = vap->va_extsize >> | |
796 | mp->m_sb.sb_blocklog; | |
797 | } | |
798 | if (mask & XFS_AT_XFLAGS) { | |
799 | uint di_flags; | |
800 | ||
801 | /* can't set PREALLOC this way, just preserve it */ | |
802 | di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC); | |
803 | if (vap->va_xflags & XFS_XFLAG_IMMUTABLE) | |
804 | di_flags |= XFS_DIFLAG_IMMUTABLE; | |
805 | if (vap->va_xflags & XFS_XFLAG_APPEND) | |
806 | di_flags |= XFS_DIFLAG_APPEND; | |
807 | if (vap->va_xflags & XFS_XFLAG_SYNC) | |
808 | di_flags |= XFS_DIFLAG_SYNC; | |
809 | if (vap->va_xflags & XFS_XFLAG_NOATIME) | |
810 | di_flags |= XFS_DIFLAG_NOATIME; | |
811 | if (vap->va_xflags & XFS_XFLAG_NODUMP) | |
812 | di_flags |= XFS_DIFLAG_NODUMP; | |
365ca83d NS |
813 | if (vap->va_xflags & XFS_XFLAG_PROJINHERIT) |
814 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
815 | if (vap->va_xflags & XFS_XFLAG_NODEFRAG) |
816 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
1da177e4 LT |
817 | if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { |
818 | if (vap->va_xflags & XFS_XFLAG_RTINHERIT) | |
819 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
820 | if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS) | |
821 | di_flags |= XFS_DIFLAG_NOSYMLINKS; | |
dd9f438e NS |
822 | if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT) |
823 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
824 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
1da177e4 LT |
825 | if (vap->va_xflags & XFS_XFLAG_REALTIME) { |
826 | di_flags |= XFS_DIFLAG_REALTIME; | |
827 | ip->i_iocore.io_flags |= XFS_IOCORE_RT; | |
828 | } else { | |
829 | ip->i_iocore.io_flags &= ~XFS_IOCORE_RT; | |
830 | } | |
dd9f438e NS |
831 | if (vap->va_xflags & XFS_XFLAG_EXTSIZE) |
832 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1da177e4 LT |
833 | } |
834 | ip->i_d.di_flags = di_flags; | |
835 | } | |
836 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
837 | timeflags |= XFS_ICHGTIME_CHG; | |
838 | } | |
839 | ||
840 | /* | |
841 | * Change file inode change time only if XFS_AT_CTIME set | |
842 | * AND we have been called by a DMI function. | |
843 | */ | |
844 | ||
845 | if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) { | |
846 | ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec; | |
847 | ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec; | |
848 | ip->i_update_core = 1; | |
849 | timeflags &= ~XFS_ICHGTIME_CHG; | |
850 | } | |
851 | ||
852 | /* | |
853 | * Send out timestamp changes that need to be set to the | |
854 | * current time. Not done when called by a DMI function. | |
855 | */ | |
856 | if (timeflags && !(flags & ATTR_DMI)) | |
857 | xfs_ichgtime(ip, timeflags); | |
858 | ||
859 | XFS_STATS_INC(xs_ig_attrchg); | |
860 | ||
861 | /* | |
862 | * If this is a synchronous mount, make sure that the | |
863 | * transaction goes to disk before returning to the user. | |
864 | * This is slightly sub-optimal in that truncates require | |
c41564b5 | 865 | * two sync transactions instead of one for wsync filesystems. |
1da177e4 LT |
866 | * One for the truncate and one for the timestamps since we |
867 | * don't want to change the timestamps unless we're sure the | |
868 | * truncate worked. Truncates are less than 1% of the laddis | |
869 | * mix so this probably isn't worth the trouble to optimize. | |
870 | */ | |
871 | code = 0; | |
872 | if (tp) { | |
873 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
874 | xfs_trans_set_sync(tp); | |
875 | ||
876 | code = xfs_trans_commit(tp, commit_flags, NULL); | |
877 | } | |
878 | ||
879 | /* | |
880 | * If the (regular) file's mandatory locking mode changed, then | |
881 | * notify the vnode. We do this under the inode lock to prevent | |
882 | * racing calls to vop_vnode_change. | |
883 | */ | |
884 | mandlock_after = MANDLOCK(vp, ip->i_d.di_mode); | |
885 | if (mandlock_before != mandlock_after) { | |
67fcaa73 | 886 | bhv_vop_vnode_change(vp, VCHANGE_FLAGS_ENF_LOCKING, |
1da177e4 LT |
887 | mandlock_after); |
888 | } | |
889 | ||
890 | xfs_iunlock(ip, lock_flags); | |
891 | ||
892 | /* | |
893 | * Release any dquot(s) the inode had kept before chown. | |
894 | */ | |
895 | XFS_QM_DQRELE(mp, olddquot1); | |
896 | XFS_QM_DQRELE(mp, olddquot2); | |
897 | XFS_QM_DQRELE(mp, udqp); | |
898 | XFS_QM_DQRELE(mp, gdqp); | |
899 | ||
900 | if (code) { | |
901 | return code; | |
902 | } | |
903 | ||
904 | if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) && | |
905 | !(flags & ATTR_DMI)) { | |
906 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL, | |
907 | NULL, DM_RIGHT_NULL, NULL, NULL, | |
908 | 0, 0, AT_DELAY_FLAG(flags)); | |
909 | } | |
910 | return 0; | |
911 | ||
912 | abort_return: | |
913 | commit_flags |= XFS_TRANS_ABORT; | |
914 | /* FALLTHROUGH */ | |
915 | error_return: | |
916 | XFS_QM_DQRELE(mp, udqp); | |
917 | XFS_QM_DQRELE(mp, gdqp); | |
918 | if (tp) { | |
919 | xfs_trans_cancel(tp, commit_flags); | |
920 | } | |
921 | if (lock_flags != 0) { | |
922 | xfs_iunlock(ip, lock_flags); | |
923 | } | |
924 | return code; | |
925 | } | |
926 | ||
927 | ||
928 | /* | |
929 | * xfs_access | |
930 | * Null conversion from vnode mode bits to inode mode bits, as in efs. | |
931 | */ | |
932 | STATIC int | |
933 | xfs_access( | |
934 | bhv_desc_t *bdp, | |
935 | int mode, | |
936 | cred_t *credp) | |
937 | { | |
938 | xfs_inode_t *ip; | |
939 | int error; | |
940 | ||
941 | vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__, | |
942 | (inst_t *)__return_address); | |
943 | ||
944 | ip = XFS_BHVTOI(bdp); | |
945 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
946 | error = xfs_iaccess(ip, mode, credp); | |
947 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
948 | return error; | |
949 | } | |
950 | ||
951 | ||
7d4fb40a NS |
952 | /* |
953 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
954 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
955 | * bmapi. | |
956 | */ | |
957 | #define SYMLINK_MAPS 2 | |
958 | ||
1da177e4 LT |
959 | /* |
960 | * xfs_readlink | |
961 | * | |
962 | */ | |
963 | STATIC int | |
964 | xfs_readlink( | |
965 | bhv_desc_t *bdp, | |
966 | uio_t *uiop, | |
967 | int ioflags, | |
968 | cred_t *credp) | |
969 | { | |
970 | xfs_inode_t *ip; | |
971 | int count; | |
972 | xfs_off_t offset; | |
973 | int pathlen; | |
67fcaa73 | 974 | bhv_vnode_t *vp; |
1da177e4 LT |
975 | int error = 0; |
976 | xfs_mount_t *mp; | |
977 | int nmaps; | |
978 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
979 | xfs_daddr_t d; | |
980 | int byte_cnt; | |
981 | int n; | |
982 | xfs_buf_t *bp; | |
983 | ||
984 | vp = BHV_TO_VNODE(bdp); | |
985 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
986 | ||
987 | ip = XFS_BHVTOI(bdp); | |
988 | mp = ip->i_mount; | |
989 | ||
990 | if (XFS_FORCED_SHUTDOWN(mp)) | |
991 | return XFS_ERROR(EIO); | |
992 | ||
993 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
994 | ||
995 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
996 | ||
997 | offset = uiop->uio_offset; | |
998 | count = uiop->uio_resid; | |
999 | ||
1000 | if (offset < 0) { | |
1001 | error = XFS_ERROR(EINVAL); | |
1002 | goto error_return; | |
1003 | } | |
1004 | if (count <= 0) { | |
1005 | error = 0; | |
1006 | goto error_return; | |
1007 | } | |
1008 | ||
1da177e4 LT |
1009 | /* |
1010 | * See if the symlink is stored inline. | |
1011 | */ | |
1012 | pathlen = (int)ip->i_d.di_size; | |
1013 | ||
1014 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
70a50528 | 1015 | error = xfs_uio_read(ip->i_df.if_u1.if_data, pathlen, uiop); |
1da177e4 LT |
1016 | } |
1017 | else { | |
1018 | /* | |
1019 | * Symlink not inline. Call bmap to get it in. | |
1020 | */ | |
1021 | nmaps = SYMLINK_MAPS; | |
1022 | ||
1023 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), | |
3e57ecf6 | 1024 | 0, NULL, 0, mval, &nmaps, NULL, NULL); |
1da177e4 LT |
1025 | |
1026 | if (error) { | |
1027 | goto error_return; | |
1028 | } | |
1029 | ||
1030 | for (n = 0; n < nmaps; n++) { | |
1031 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
1032 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
1033 | bp = xfs_buf_read(mp->m_ddev_targp, d, | |
1034 | BTOBB(byte_cnt), 0); | |
1035 | error = XFS_BUF_GETERROR(bp); | |
1036 | if (error) { | |
1037 | xfs_ioerror_alert("xfs_readlink", | |
1038 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
1039 | xfs_buf_relse(bp); | |
1040 | goto error_return; | |
1041 | } | |
1042 | if (pathlen < byte_cnt) | |
1043 | byte_cnt = pathlen; | |
1044 | pathlen -= byte_cnt; | |
1045 | ||
70a50528 | 1046 | error = xfs_uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop); |
1da177e4 LT |
1047 | xfs_buf_relse (bp); |
1048 | } | |
1049 | ||
1050 | } | |
1051 | ||
1da177e4 | 1052 | error_return: |
1da177e4 | 1053 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 LT |
1054 | return error; |
1055 | } | |
1056 | ||
1057 | ||
1058 | /* | |
1059 | * xfs_fsync | |
1060 | * | |
1061 | * This is called to sync the inode and its data out to disk. | |
1062 | * We need to hold the I/O lock while flushing the data, and | |
1063 | * the inode lock while flushing the inode. The inode lock CANNOT | |
1064 | * be held while flushing the data, so acquire after we're done | |
1065 | * with that. | |
1066 | */ | |
1067 | STATIC int | |
1068 | xfs_fsync( | |
1069 | bhv_desc_t *bdp, | |
1070 | int flag, | |
1071 | cred_t *credp, | |
1072 | xfs_off_t start, | |
1073 | xfs_off_t stop) | |
1074 | { | |
1075 | xfs_inode_t *ip; | |
1076 | xfs_trans_t *tp; | |
1077 | int error; | |
f538d4da | 1078 | int log_flushed = 0, changed = 1; |
1da177e4 LT |
1079 | |
1080 | vn_trace_entry(BHV_TO_VNODE(bdp), | |
1081 | __FUNCTION__, (inst_t *)__return_address); | |
1082 | ||
1083 | ip = XFS_BHVTOI(bdp); | |
1084 | ||
1085 | ASSERT(start >= 0 && stop >= -1); | |
1086 | ||
1087 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
1088 | return XFS_ERROR(EIO); | |
1089 | ||
1090 | /* | |
1091 | * We always need to make sure that the required inode state | |
1092 | * is safe on disk. The vnode might be clean but because | |
1093 | * of committed transactions that haven't hit the disk yet. | |
1094 | * Likewise, there could be unflushed non-transactional | |
1095 | * changes to the inode core that have to go to disk. | |
1096 | * | |
1097 | * The following code depends on one assumption: that | |
1098 | * any transaction that changes an inode logs the core | |
1099 | * because it has to change some field in the inode core | |
1100 | * (typically nextents or nblocks). That assumption | |
1101 | * implies that any transactions against an inode will | |
1102 | * catch any non-transactional updates. If inode-altering | |
1103 | * transactions exist that violate this assumption, the | |
1104 | * code breaks. Right now, it figures that if the involved | |
1105 | * update_* field is clear and the inode is unpinned, the | |
1106 | * inode is clean. Either it's been flushed or it's been | |
1107 | * committed and the commit has hit the disk unpinning the inode. | |
1108 | * (Note that xfs_inode_item_format() called at commit clears | |
1109 | * the update_* fields.) | |
1110 | */ | |
1111 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1112 | ||
1113 | /* If we are flushing data then we care about update_size | |
1114 | * being set, otherwise we care about update_core | |
1115 | */ | |
1116 | if ((flag & FSYNC_DATA) ? | |
1117 | (ip->i_update_size == 0) : | |
1118 | (ip->i_update_core == 0)) { | |
1119 | /* | |
1120 | * Timestamps/size haven't changed since last inode | |
1121 | * flush or inode transaction commit. That means | |
1122 | * either nothing got written or a transaction | |
1123 | * committed which caught the updates. If the | |
1124 | * latter happened and the transaction hasn't | |
1125 | * hit the disk yet, the inode will be still | |
1126 | * be pinned. If it is, force the log. | |
1127 | */ | |
1128 | ||
1129 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1130 | ||
1131 | if (xfs_ipincount(ip)) { | |
f538d4da | 1132 | _xfs_log_force(ip->i_mount, (xfs_lsn_t)0, |
1da177e4 LT |
1133 | XFS_LOG_FORCE | |
1134 | ((flag & FSYNC_WAIT) | |
f538d4da CH |
1135 | ? XFS_LOG_SYNC : 0), |
1136 | &log_flushed); | |
1137 | } else { | |
1138 | /* | |
1139 | * If the inode is not pinned and nothing | |
1140 | * has changed we don't need to flush the | |
1141 | * cache. | |
1142 | */ | |
1143 | changed = 0; | |
1da177e4 LT |
1144 | } |
1145 | error = 0; | |
1146 | } else { | |
1147 | /* | |
1148 | * Kick off a transaction to log the inode | |
1149 | * core to get the updates. Make it | |
1150 | * sync if FSYNC_WAIT is passed in (which | |
1151 | * is done by everybody but specfs). The | |
1152 | * sync transaction will also force the log. | |
1153 | */ | |
1154 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1155 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
1156 | if ((error = xfs_trans_reserve(tp, 0, | |
1157 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), | |
1158 | 0, 0, 0))) { | |
1159 | xfs_trans_cancel(tp, 0); | |
1160 | return error; | |
1161 | } | |
1162 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1163 | ||
1164 | /* | |
1165 | * Note - it's possible that we might have pushed | |
1166 | * ourselves out of the way during trans_reserve | |
1167 | * which would flush the inode. But there's no | |
1168 | * guarantee that the inode buffer has actually | |
1169 | * gone out yet (it's delwri). Plus the buffer | |
1170 | * could be pinned anyway if it's part of an | |
1171 | * inode in another recent transaction. So we | |
1172 | * play it safe and fire off the transaction anyway. | |
1173 | */ | |
1174 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1175 | xfs_trans_ihold(tp, ip); | |
1176 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1177 | if (flag & FSYNC_WAIT) | |
1178 | xfs_trans_set_sync(tp); | |
f538d4da | 1179 | error = _xfs_trans_commit(tp, 0, NULL, &log_flushed); |
1da177e4 LT |
1180 | |
1181 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1182 | } | |
f538d4da CH |
1183 | |
1184 | if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) { | |
1185 | /* | |
1186 | * If the log write didn't issue an ordered tag we need | |
1187 | * to flush the disk cache for the data device now. | |
1188 | */ | |
1189 | if (!log_flushed) | |
1190 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
1191 | ||
1192 | /* | |
1193 | * If this inode is on the RT dev we need to flush that | |
c41564b5 | 1194 | * cache as well. |
f538d4da CH |
1195 | */ |
1196 | if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) | |
1197 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); | |
1198 | } | |
1199 | ||
1da177e4 LT |
1200 | return error; |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * This is called by xfs_inactive to free any blocks beyond eof, | |
1205 | * when the link count isn't zero. | |
1206 | */ | |
1207 | STATIC int | |
1208 | xfs_inactive_free_eofblocks( | |
1209 | xfs_mount_t *mp, | |
1210 | xfs_inode_t *ip) | |
1211 | { | |
1212 | xfs_trans_t *tp; | |
1213 | int error; | |
1214 | xfs_fileoff_t end_fsb; | |
1215 | xfs_fileoff_t last_fsb; | |
1216 | xfs_filblks_t map_len; | |
1217 | int nimaps; | |
1218 | xfs_bmbt_irec_t imap; | |
1219 | ||
1220 | /* | |
1221 | * Figure out if there are any blocks beyond the end | |
1222 | * of the file. If not, then there is nothing to do. | |
1223 | */ | |
1224 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size)); | |
1225 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); | |
1226 | map_len = last_fsb - end_fsb; | |
1227 | if (map_len <= 0) | |
014c2544 | 1228 | return 0; |
1da177e4 LT |
1229 | |
1230 | nimaps = 1; | |
1231 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3e57ecf6 OW |
1232 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0, |
1233 | NULL, 0, &imap, &nimaps, NULL, NULL); | |
1da177e4 LT |
1234 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1235 | ||
1236 | if (!error && (nimaps != 0) && | |
68bdb6ea YL |
1237 | (imap.br_startblock != HOLESTARTBLOCK || |
1238 | ip->i_delayed_blks)) { | |
1da177e4 LT |
1239 | /* |
1240 | * Attach the dquots to the inode up front. | |
1241 | */ | |
1242 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1243 | return error; |
1da177e4 LT |
1244 | |
1245 | /* | |
1246 | * There are blocks after the end of file. | |
1247 | * Free them up now by truncating the file to | |
1248 | * its current size. | |
1249 | */ | |
1250 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1251 | ||
1252 | /* | |
1253 | * Do the xfs_itruncate_start() call before | |
1254 | * reserving any log space because | |
1255 | * itruncate_start will call into the buffer | |
1256 | * cache and we can't | |
1257 | * do that within a transaction. | |
1258 | */ | |
1259 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1260 | xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, | |
1261 | ip->i_d.di_size); | |
1262 | ||
1263 | error = xfs_trans_reserve(tp, 0, | |
1264 | XFS_ITRUNCATE_LOG_RES(mp), | |
1265 | 0, XFS_TRANS_PERM_LOG_RES, | |
1266 | XFS_ITRUNCATE_LOG_COUNT); | |
1267 | if (error) { | |
1268 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1269 | xfs_trans_cancel(tp, 0); | |
1270 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1271 | return error; |
1da177e4 LT |
1272 | } |
1273 | ||
1274 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1275 | xfs_trans_ijoin(tp, ip, | |
1276 | XFS_IOLOCK_EXCL | | |
1277 | XFS_ILOCK_EXCL); | |
1278 | xfs_trans_ihold(tp, ip); | |
1279 | ||
1280 | error = xfs_itruncate_finish(&tp, ip, | |
1281 | ip->i_d.di_size, | |
1282 | XFS_DATA_FORK, | |
1283 | 0); | |
1284 | /* | |
1285 | * If we get an error at this point we | |
1286 | * simply don't bother truncating the file. | |
1287 | */ | |
1288 | if (error) { | |
1289 | xfs_trans_cancel(tp, | |
1290 | (XFS_TRANS_RELEASE_LOG_RES | | |
1291 | XFS_TRANS_ABORT)); | |
1292 | } else { | |
1293 | error = xfs_trans_commit(tp, | |
1294 | XFS_TRANS_RELEASE_LOG_RES, | |
1295 | NULL); | |
1296 | } | |
1297 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1298 | } | |
014c2544 | 1299 | return error; |
1da177e4 LT |
1300 | } |
1301 | ||
1302 | /* | |
1303 | * Free a symlink that has blocks associated with it. | |
1304 | */ | |
1305 | STATIC int | |
1306 | xfs_inactive_symlink_rmt( | |
1307 | xfs_inode_t *ip, | |
1308 | xfs_trans_t **tpp) | |
1309 | { | |
1310 | xfs_buf_t *bp; | |
1311 | int committed; | |
1312 | int done; | |
1313 | int error; | |
1314 | xfs_fsblock_t first_block; | |
1315 | xfs_bmap_free_t free_list; | |
1316 | int i; | |
1317 | xfs_mount_t *mp; | |
1318 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1319 | int nmaps; | |
1320 | xfs_trans_t *ntp; | |
1321 | int size; | |
1322 | xfs_trans_t *tp; | |
1323 | ||
1324 | tp = *tpp; | |
1325 | mp = ip->i_mount; | |
1326 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
1327 | /* | |
1328 | * We're freeing a symlink that has some | |
1329 | * blocks allocated to it. Free the | |
1330 | * blocks here. We know that we've got | |
1331 | * either 1 or 2 extents and that we can | |
1332 | * free them all in one bunmapi call. | |
1333 | */ | |
1334 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
1335 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1336 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1337 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1338 | xfs_trans_cancel(tp, 0); | |
1339 | *tpp = NULL; | |
1340 | return error; | |
1341 | } | |
1342 | /* | |
1343 | * Lock the inode, fix the size, and join it to the transaction. | |
1344 | * Hold it so in the normal path, we still have it locked for | |
1345 | * the second transaction. In the error paths we need it | |
1346 | * held so the cancel won't rele it, see below. | |
1347 | */ | |
1348 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1349 | size = (int)ip->i_d.di_size; | |
1350 | ip->i_d.di_size = 0; | |
1351 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1352 | xfs_trans_ihold(tp, ip); | |
1353 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1354 | /* | |
1355 | * Find the block(s) so we can inval and unmap them. | |
1356 | */ | |
1357 | done = 0; | |
1358 | XFS_BMAP_INIT(&free_list, &first_block); | |
e8c96f8c | 1359 | nmaps = ARRAY_SIZE(mval); |
1da177e4 LT |
1360 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), |
1361 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
3e57ecf6 | 1362 | &free_list, NULL))) |
1da177e4 LT |
1363 | goto error0; |
1364 | /* | |
1365 | * Invalidate the block(s). | |
1366 | */ | |
1367 | for (i = 0; i < nmaps; i++) { | |
1368 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
1369 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
1370 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
1371 | xfs_trans_binval(tp, bp); | |
1372 | } | |
1373 | /* | |
1374 | * Unmap the dead block(s) to the free_list. | |
1375 | */ | |
1376 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
3e57ecf6 | 1377 | &first_block, &free_list, NULL, &done))) |
1da177e4 LT |
1378 | goto error1; |
1379 | ASSERT(done); | |
1380 | /* | |
1381 | * Commit the first transaction. This logs the EFI and the inode. | |
1382 | */ | |
f7c99b6f | 1383 | if ((error = xfs_bmap_finish(&tp, &free_list, &committed))) |
1da177e4 LT |
1384 | goto error1; |
1385 | /* | |
1386 | * The transaction must have been committed, since there were | |
1387 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
1388 | * The new tp has the extent freeing and EFDs. | |
1389 | */ | |
1390 | ASSERT(committed); | |
1391 | /* | |
1392 | * The first xact was committed, so add the inode to the new one. | |
1393 | * Mark it dirty so it will be logged and moved forward in the log as | |
1394 | * part of every commit. | |
1395 | */ | |
1396 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1397 | xfs_trans_ihold(tp, ip); | |
1398 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1399 | /* | |
1400 | * Get a new, empty transaction to return to our caller. | |
1401 | */ | |
1402 | ntp = xfs_trans_dup(tp); | |
1403 | /* | |
c41564b5 | 1404 | * Commit the transaction containing extent freeing and EFDs. |
1da177e4 LT |
1405 | * If we get an error on the commit here or on the reserve below, |
1406 | * we need to unlock the inode since the new transaction doesn't | |
1407 | * have the inode attached. | |
1408 | */ | |
1409 | error = xfs_trans_commit(tp, 0, NULL); | |
1410 | tp = ntp; | |
1411 | if (error) { | |
1412 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1413 | goto error0; | |
1414 | } | |
1415 | /* | |
1416 | * Remove the memory for extent descriptions (just bookkeeping). | |
1417 | */ | |
1418 | if (ip->i_df.if_bytes) | |
1419 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
1420 | ASSERT(ip->i_df.if_bytes == 0); | |
1421 | /* | |
1422 | * Put an itruncate log reservation in the new transaction | |
1423 | * for our caller. | |
1424 | */ | |
1425 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1426 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1427 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1428 | goto error0; | |
1429 | } | |
1430 | /* | |
1431 | * Return with the inode locked but not joined to the transaction. | |
1432 | */ | |
1433 | *tpp = tp; | |
1434 | return 0; | |
1435 | ||
1436 | error1: | |
1437 | xfs_bmap_cancel(&free_list); | |
1438 | error0: | |
1439 | /* | |
1440 | * Have to come here with the inode locked and either | |
1441 | * (held and in the transaction) or (not in the transaction). | |
1442 | * If the inode isn't held then cancel would iput it, but | |
1443 | * that's wrong since this is inactive and the vnode ref | |
1444 | * count is 0 already. | |
1445 | * Cancel won't do anything to the inode if held, but it still | |
1446 | * needs to be locked until the cancel is done, if it was | |
1447 | * joined to the transaction. | |
1448 | */ | |
1449 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1450 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1451 | *tpp = NULL; | |
1452 | return error; | |
1453 | ||
1454 | } | |
1455 | ||
1456 | STATIC int | |
1457 | xfs_inactive_symlink_local( | |
1458 | xfs_inode_t *ip, | |
1459 | xfs_trans_t **tpp) | |
1460 | { | |
1461 | int error; | |
1462 | ||
1463 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
1464 | /* | |
1465 | * We're freeing a symlink which fit into | |
1466 | * the inode. Just free the memory used | |
1467 | * to hold the old symlink. | |
1468 | */ | |
1469 | error = xfs_trans_reserve(*tpp, 0, | |
1470 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
1471 | 0, XFS_TRANS_PERM_LOG_RES, | |
1472 | XFS_ITRUNCATE_LOG_COUNT); | |
1473 | ||
1474 | if (error) { | |
1475 | xfs_trans_cancel(*tpp, 0); | |
1476 | *tpp = NULL; | |
014c2544 | 1477 | return error; |
1da177e4 LT |
1478 | } |
1479 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1480 | ||
1481 | /* | |
1482 | * Zero length symlinks _can_ exist. | |
1483 | */ | |
1484 | if (ip->i_df.if_bytes > 0) { | |
1485 | xfs_idata_realloc(ip, | |
1486 | -(ip->i_df.if_bytes), | |
1487 | XFS_DATA_FORK); | |
1488 | ASSERT(ip->i_df.if_bytes == 0); | |
1489 | } | |
014c2544 | 1490 | return 0; |
1da177e4 LT |
1491 | } |
1492 | ||
1da177e4 LT |
1493 | STATIC int |
1494 | xfs_inactive_attrs( | |
1495 | xfs_inode_t *ip, | |
1496 | xfs_trans_t **tpp) | |
1497 | { | |
1498 | xfs_trans_t *tp; | |
1499 | int error; | |
1500 | xfs_mount_t *mp; | |
1501 | ||
1502 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); | |
1503 | tp = *tpp; | |
1504 | mp = ip->i_mount; | |
1505 | ASSERT(ip->i_d.di_forkoff != 0); | |
1506 | xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
1507 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1508 | ||
1509 | error = xfs_attr_inactive(ip); | |
1510 | if (error) { | |
1511 | *tpp = NULL; | |
1512 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1513 | return error; /* goto out */ |
1da177e4 LT |
1514 | } |
1515 | ||
1516 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1517 | error = xfs_trans_reserve(tp, 0, | |
1518 | XFS_IFREE_LOG_RES(mp), | |
1519 | 0, XFS_TRANS_PERM_LOG_RES, | |
1520 | XFS_INACTIVE_LOG_COUNT); | |
1521 | if (error) { | |
1522 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1523 | xfs_trans_cancel(tp, 0); | |
1524 | *tpp = NULL; | |
1525 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1526 | return error; |
1da177e4 LT |
1527 | } |
1528 | ||
1529 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1530 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1531 | xfs_trans_ihold(tp, ip); | |
1532 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1533 | ||
1534 | ASSERT(ip->i_d.di_anextents == 0); | |
1535 | ||
1536 | *tpp = tp; | |
014c2544 | 1537 | return 0; |
1da177e4 LT |
1538 | } |
1539 | ||
1540 | STATIC int | |
1541 | xfs_release( | |
1542 | bhv_desc_t *bdp) | |
1543 | { | |
1544 | xfs_inode_t *ip; | |
67fcaa73 | 1545 | bhv_vnode_t *vp; |
1da177e4 LT |
1546 | xfs_mount_t *mp; |
1547 | int error; | |
1548 | ||
1549 | vp = BHV_TO_VNODE(bdp); | |
1550 | ip = XFS_BHVTOI(bdp); | |
7d4fb40a | 1551 | mp = ip->i_mount; |
1da177e4 | 1552 | |
7d4fb40a | 1553 | if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0)) |
1da177e4 | 1554 | return 0; |
1da177e4 LT |
1555 | |
1556 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1557 | if (vp->v_vfsp->vfs_flag & VFS_RDONLY) | |
1558 | return 0; | |
1559 | ||
1560 | #ifdef HAVE_REFCACHE | |
1561 | /* If we are in the NFS reference cache then don't do this now */ | |
1562 | if (ip->i_refcache) | |
1563 | return 0; | |
1564 | #endif | |
1565 | ||
1da177e4 LT |
1566 | if (ip->i_d.di_nlink != 0) { |
1567 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
68bdb6ea YL |
1568 | ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 || |
1569 | ip->i_delayed_blks > 0)) && | |
1da177e4 | 1570 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && |
dd9f438e NS |
1571 | (!(ip->i_d.di_flags & |
1572 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) { | |
1da177e4 | 1573 | if ((error = xfs_inactive_free_eofblocks(mp, ip))) |
014c2544 | 1574 | return error; |
1da177e4 | 1575 | /* Update linux inode block count after free above */ |
ec86dc02 | 1576 | vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp, |
1da177e4 LT |
1577 | ip->i_d.di_nblocks + ip->i_delayed_blks); |
1578 | } | |
1579 | } | |
1580 | ||
1581 | return 0; | |
1582 | } | |
1583 | ||
1584 | /* | |
1585 | * xfs_inactive | |
1586 | * | |
1587 | * This is called when the vnode reference count for the vnode | |
1588 | * goes to zero. If the file has been unlinked, then it must | |
1589 | * now be truncated. Also, we clear all of the read-ahead state | |
1590 | * kept for the inode here since the file is now closed. | |
1591 | */ | |
1592 | STATIC int | |
1593 | xfs_inactive( | |
1594 | bhv_desc_t *bdp, | |
1595 | cred_t *credp) | |
1596 | { | |
1597 | xfs_inode_t *ip; | |
67fcaa73 NS |
1598 | bhv_vnode_t *vp; |
1599 | xfs_bmap_free_t free_list; | |
1da177e4 LT |
1600 | xfs_fsblock_t first_block; |
1601 | int committed; | |
1602 | xfs_trans_t *tp; | |
1603 | xfs_mount_t *mp; | |
1604 | int error; | |
1605 | int truncate; | |
1606 | ||
1607 | vp = BHV_TO_VNODE(bdp); | |
1608 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
1609 | ||
1610 | ip = XFS_BHVTOI(bdp); | |
1611 | ||
1612 | /* | |
1613 | * If the inode is already free, then there can be nothing | |
1614 | * to clean up here. | |
1615 | */ | |
1616 | if (ip->i_d.di_mode == 0 || VN_BAD(vp)) { | |
1617 | ASSERT(ip->i_df.if_real_bytes == 0); | |
1618 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1619 | return VN_INACTIVE_CACHE; | |
1620 | } | |
1621 | ||
1622 | /* | |
1623 | * Only do a truncate if it's a regular file with | |
1624 | * some actual space in it. It's OK to look at the | |
1625 | * inode's fields without the lock because we're the | |
1626 | * only one with a reference to the inode. | |
1627 | */ | |
1628 | truncate = ((ip->i_d.di_nlink == 0) && | |
1661dc8e YL |
1629 | ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) || |
1630 | (ip->i_delayed_blks > 0)) && | |
1da177e4 LT |
1631 | ((ip->i_d.di_mode & S_IFMT) == S_IFREG)); |
1632 | ||
1633 | mp = ip->i_mount; | |
1634 | ||
1635 | if (ip->i_d.di_nlink == 0 && | |
1636 | DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) { | |
1637 | (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL); | |
1638 | } | |
1639 | ||
1640 | error = 0; | |
1641 | ||
1642 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1643 | if (vp->v_vfsp->vfs_flag & VFS_RDONLY) | |
1644 | goto out; | |
1645 | ||
1646 | if (ip->i_d.di_nlink != 0) { | |
1647 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
68bdb6ea YL |
1648 | ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 || |
1649 | ip->i_delayed_blks > 0)) && | |
dd9f438e NS |
1650 | (ip->i_df.if_flags & XFS_IFEXTENTS) && |
1651 | (!(ip->i_d.di_flags & | |
1652 | (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) || | |
1653 | (ip->i_delayed_blks != 0)))) { | |
1da177e4 | 1654 | if ((error = xfs_inactive_free_eofblocks(mp, ip))) |
014c2544 | 1655 | return VN_INACTIVE_CACHE; |
1da177e4 | 1656 | /* Update linux inode block count after free above */ |
ec86dc02 | 1657 | vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp, |
1da177e4 LT |
1658 | ip->i_d.di_nblocks + ip->i_delayed_blks); |
1659 | } | |
1660 | goto out; | |
1661 | } | |
1662 | ||
1663 | ASSERT(ip->i_d.di_nlink == 0); | |
1664 | ||
1665 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
014c2544 | 1666 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1667 | |
1668 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1669 | if (truncate) { | |
1670 | /* | |
1671 | * Do the xfs_itruncate_start() call before | |
1672 | * reserving any log space because itruncate_start | |
1673 | * will call into the buffer cache and we can't | |
1674 | * do that within a transaction. | |
1675 | */ | |
1676 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1677 | ||
1678 | xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); | |
1679 | ||
1680 | error = xfs_trans_reserve(tp, 0, | |
1681 | XFS_ITRUNCATE_LOG_RES(mp), | |
1682 | 0, XFS_TRANS_PERM_LOG_RES, | |
1683 | XFS_ITRUNCATE_LOG_COUNT); | |
1684 | if (error) { | |
1685 | /* Don't call itruncate_cleanup */ | |
1686 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1687 | xfs_trans_cancel(tp, 0); | |
1688 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
014c2544 | 1689 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1690 | } |
1691 | ||
1692 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1693 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1694 | xfs_trans_ihold(tp, ip); | |
1695 | ||
1696 | /* | |
1697 | * normally, we have to run xfs_itruncate_finish sync. | |
1698 | * But if filesystem is wsync and we're in the inactive | |
1699 | * path, then we know that nlink == 0, and that the | |
1700 | * xaction that made nlink == 0 is permanently committed | |
1701 | * since xfs_remove runs as a synchronous transaction. | |
1702 | */ | |
1703 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1704 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1705 | ||
1706 | if (error) { | |
1707 | xfs_trans_cancel(tp, | |
1708 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1709 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
014c2544 | 1710 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1711 | } |
1712 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1713 | ||
1714 | /* | |
1715 | * If we get an error while cleaning up a | |
1716 | * symlink we bail out. | |
1717 | */ | |
1718 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1719 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1720 | xfs_inactive_symlink_local(ip, &tp); | |
1721 | ||
1722 | if (error) { | |
1723 | ASSERT(tp == NULL); | |
014c2544 | 1724 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1725 | } |
1726 | ||
1727 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1728 | xfs_trans_ihold(tp, ip); | |
1729 | } else { | |
1730 | error = xfs_trans_reserve(tp, 0, | |
1731 | XFS_IFREE_LOG_RES(mp), | |
1732 | 0, XFS_TRANS_PERM_LOG_RES, | |
1733 | XFS_INACTIVE_LOG_COUNT); | |
1734 | if (error) { | |
1735 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1736 | xfs_trans_cancel(tp, 0); | |
014c2544 | 1737 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1738 | } |
1739 | ||
1740 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1741 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1742 | xfs_trans_ihold(tp, ip); | |
1743 | } | |
1744 | ||
1745 | /* | |
1746 | * If there are attributes associated with the file | |
1747 | * then blow them away now. The code calls a routine | |
1748 | * that recursively deconstructs the attribute fork. | |
1749 | * We need to just commit the current transaction | |
1750 | * because we can't use it for xfs_attr_inactive(). | |
1751 | */ | |
1752 | if (ip->i_d.di_anextents > 0) { | |
1753 | error = xfs_inactive_attrs(ip, &tp); | |
1754 | /* | |
1755 | * If we got an error, the transaction is already | |
1756 | * cancelled, and the inode is unlocked. Just get out. | |
1757 | */ | |
1758 | if (error) | |
014c2544 | 1759 | return VN_INACTIVE_CACHE; |
1da177e4 LT |
1760 | } else if (ip->i_afp) { |
1761 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1762 | } | |
1763 | ||
1764 | /* | |
1765 | * Free the inode. | |
1766 | */ | |
1767 | XFS_BMAP_INIT(&free_list, &first_block); | |
1768 | error = xfs_ifree(tp, ip, &free_list); | |
1769 | if (error) { | |
1770 | /* | |
1771 | * If we fail to free the inode, shut down. The cancel | |
1772 | * might do that, we need to make sure. Otherwise the | |
1773 | * inode might be lost for a long time or forever. | |
1774 | */ | |
1775 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1776 | cmn_err(CE_NOTE, | |
1777 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1778 | error, mp->m_fsname); | |
7d04a335 | 1779 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1da177e4 LT |
1780 | } |
1781 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1782 | } else { | |
1783 | /* | |
1784 | * Credit the quota account(s). The inode is gone. | |
1785 | */ | |
1786 | XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1787 | ||
1788 | /* | |
1789 | * Just ignore errors at this point. There is | |
1790 | * nothing we can do except to try to keep going. | |
1791 | */ | |
f7c99b6f | 1792 | (void) xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
1793 | (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); |
1794 | } | |
1795 | /* | |
1796 | * Release the dquots held by inode, if any. | |
1797 | */ | |
1798 | XFS_QM_DQDETACH(mp, ip); | |
1799 | ||
1800 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1801 | ||
1802 | out: | |
1803 | return VN_INACTIVE_CACHE; | |
1804 | } | |
1805 | ||
1806 | ||
1807 | /* | |
1808 | * xfs_lookup | |
1809 | */ | |
1810 | STATIC int | |
1811 | xfs_lookup( | |
1812 | bhv_desc_t *dir_bdp, | |
8285fb58 | 1813 | bhv_vname_t *dentry, |
67fcaa73 | 1814 | bhv_vnode_t **vpp, |
1da177e4 | 1815 | int flags, |
67fcaa73 | 1816 | bhv_vnode_t *rdir, |
1da177e4 LT |
1817 | cred_t *credp) |
1818 | { | |
1819 | xfs_inode_t *dp, *ip; | |
1820 | xfs_ino_t e_inum; | |
1821 | int error; | |
1822 | uint lock_mode; | |
67fcaa73 | 1823 | bhv_vnode_t *dir_vp; |
1da177e4 LT |
1824 | |
1825 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
1826 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
1827 | ||
1828 | dp = XFS_BHVTOI(dir_bdp); | |
1829 | ||
1830 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1831 | return XFS_ERROR(EIO); | |
1832 | ||
1833 | lock_mode = xfs_ilock_map_shared(dp); | |
1834 | error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip); | |
1835 | if (!error) { | |
1836 | *vpp = XFS_ITOV(ip); | |
1837 | ITRACE(ip); | |
1838 | } | |
1839 | xfs_iunlock_map_shared(dp, lock_mode); | |
1840 | return error; | |
1841 | } | |
1842 | ||
1843 | ||
1844 | /* | |
1845 | * xfs_create (create a new file). | |
1846 | */ | |
1847 | STATIC int | |
1848 | xfs_create( | |
1849 | bhv_desc_t *dir_bdp, | |
8285fb58 NS |
1850 | bhv_vname_t *dentry, |
1851 | bhv_vattr_t *vap, | |
67fcaa73 | 1852 | bhv_vnode_t **vpp, |
1da177e4 LT |
1853 | cred_t *credp) |
1854 | { | |
1855 | char *name = VNAME(dentry); | |
67fcaa73 | 1856 | bhv_vnode_t *dir_vp; |
1da177e4 | 1857 | xfs_inode_t *dp, *ip; |
67fcaa73 | 1858 | bhv_vnode_t *vp = NULL; |
1da177e4 LT |
1859 | xfs_trans_t *tp; |
1860 | xfs_mount_t *mp; | |
1861 | xfs_dev_t rdev; | |
1862 | int error; | |
1863 | xfs_bmap_free_t free_list; | |
1864 | xfs_fsblock_t first_block; | |
1865 | boolean_t dp_joined_to_trans; | |
1866 | int dm_event_sent = 0; | |
1867 | uint cancel_flags; | |
1868 | int committed; | |
1869 | xfs_prid_t prid; | |
1870 | struct xfs_dquot *udqp, *gdqp; | |
1871 | uint resblks; | |
1872 | int dm_di_mode; | |
1873 | int namelen; | |
1874 | ||
1875 | ASSERT(!*vpp); | |
1876 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
1877 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
1878 | ||
1879 | dp = XFS_BHVTOI(dir_bdp); | |
1880 | mp = dp->i_mount; | |
1881 | ||
0432dab2 | 1882 | dm_di_mode = vap->va_mode; |
1da177e4 LT |
1883 | namelen = VNAMELEN(dentry); |
1884 | ||
1885 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) { | |
1886 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, | |
1887 | dir_vp, DM_RIGHT_NULL, NULL, | |
1888 | DM_RIGHT_NULL, name, NULL, | |
1889 | dm_di_mode, 0, 0); | |
1890 | ||
1891 | if (error) | |
1892 | return error; | |
1893 | dm_event_sent = 1; | |
1894 | } | |
1895 | ||
1896 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1897 | return XFS_ERROR(EIO); | |
1898 | ||
1899 | /* Return through std_return after this point. */ | |
1900 | ||
1901 | udqp = gdqp = NULL; | |
365ca83d NS |
1902 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
1903 | prid = dp->i_d.di_projid; | |
1904 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
1905 | prid = (xfs_prid_t)vap->va_projid; |
1906 | else | |
1907 | prid = (xfs_prid_t)dfltprid; | |
1908 | ||
1909 | /* | |
1910 | * Make sure that we have allocated dquot(s) on disk. | |
1911 | */ | |
1912 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 1913 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
1914 | XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp); |
1915 | if (error) | |
1916 | goto std_return; | |
1917 | ||
1918 | ip = NULL; | |
1919 | dp_joined_to_trans = B_FALSE; | |
1920 | ||
1921 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1922 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1923 | resblks = XFS_CREATE_SPACE_RES(mp, namelen); | |
1924 | /* | |
1925 | * Initially assume that the file does not exist and | |
1926 | * reserve the resources for that case. If that is not | |
1927 | * the case we'll drop the one we have and get a more | |
1928 | * appropriate transaction later. | |
1929 | */ | |
1930 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1931 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1932 | if (error == ENOSPC) { | |
1933 | resblks = 0; | |
1934 | error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0, | |
1935 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1936 | } | |
1937 | if (error) { | |
1938 | cancel_flags = 0; | |
1939 | dp = NULL; | |
1940 | goto error_return; | |
1941 | } | |
1942 | ||
1943 | xfs_ilock(dp, XFS_ILOCK_EXCL); | |
1944 | ||
1945 | XFS_BMAP_INIT(&free_list, &first_block); | |
1946 | ||
1947 | ASSERT(ip == NULL); | |
1948 | ||
1949 | /* | |
1950 | * Reserve disk quota and the inode. | |
1951 | */ | |
1952 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
1953 | if (error) | |
1954 | goto error_return; | |
1955 | ||
f6c2d1fa | 1956 | if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen))) |
1da177e4 LT |
1957 | goto error_return; |
1958 | rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0; | |
0432dab2 | 1959 | error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1, |
1da177e4 LT |
1960 | rdev, credp, prid, resblks > 0, |
1961 | &ip, &committed); | |
1962 | if (error) { | |
1963 | if (error == ENOSPC) | |
1964 | goto error_return; | |
1965 | goto abort_return; | |
1966 | } | |
1967 | ITRACE(ip); | |
1968 | ||
1969 | /* | |
1970 | * At this point, we've gotten a newly allocated inode. | |
1971 | * It is locked (and joined to the transaction). | |
1972 | */ | |
1973 | ||
1974 | ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE)); | |
1975 | ||
1976 | /* | |
1977 | * Now we join the directory inode to the transaction. | |
1978 | * We do not do it earlier because xfs_dir_ialloc | |
1979 | * might commit the previous transaction (and release | |
1980 | * all the locks). | |
1981 | */ | |
1982 | ||
1983 | VN_HOLD(dir_vp); | |
1984 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
1985 | dp_joined_to_trans = B_TRUE; | |
1986 | ||
f6c2d1fa NS |
1987 | error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino, |
1988 | &first_block, &free_list, resblks ? | |
1989 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
1990 | if (error) { |
1991 | ASSERT(error != ENOSPC); | |
1992 | goto abort_return; | |
1993 | } | |
1994 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1995 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1996 | ||
1997 | /* | |
1998 | * If this is a synchronous mount, make sure that the | |
1999 | * create transaction goes to disk before returning to | |
2000 | * the user. | |
2001 | */ | |
2002 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2003 | xfs_trans_set_sync(tp); | |
2004 | } | |
2005 | ||
2006 | dp->i_gen++; | |
2007 | ||
2008 | /* | |
2009 | * Attach the dquot(s) to the inodes and modify them incore. | |
2010 | * These ids of the inode couldn't have changed since the new | |
2011 | * inode has been locked ever since it was created. | |
2012 | */ | |
2013 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
2014 | ||
2015 | /* | |
2016 | * xfs_trans_commit normally decrements the vnode ref count | |
2017 | * when it unlocks the inode. Since we want to return the | |
2018 | * vnode to the caller, we bump the vnode ref count now. | |
2019 | */ | |
2020 | IHOLD(ip); | |
2021 | vp = XFS_ITOV(ip); | |
2022 | ||
f7c99b6f | 2023 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2024 | if (error) { |
2025 | xfs_bmap_cancel(&free_list); | |
2026 | goto abort_rele; | |
2027 | } | |
2028 | ||
2029 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
2030 | if (error) { | |
2031 | IRELE(ip); | |
2032 | tp = NULL; | |
2033 | goto error_return; | |
2034 | } | |
2035 | ||
2036 | XFS_QM_DQRELE(mp, udqp); | |
2037 | XFS_QM_DQRELE(mp, gdqp); | |
2038 | ||
2039 | /* | |
c41564b5 | 2040 | * Propagate the fact that the vnode changed after the |
1da177e4 LT |
2041 | * xfs_inode locks have been released. |
2042 | */ | |
67fcaa73 | 2043 | bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3); |
1da177e4 LT |
2044 | |
2045 | *vpp = vp; | |
2046 | ||
2047 | /* Fallthrough to std_return with error = 0 */ | |
2048 | ||
2049 | std_return: | |
2050 | if ( (*vpp || (error != 0 && dm_event_sent != 0)) && | |
2051 | DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp), | |
2052 | DM_EVENT_POSTCREATE)) { | |
2053 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, | |
2054 | dir_vp, DM_RIGHT_NULL, | |
2055 | *vpp ? vp:NULL, | |
2056 | DM_RIGHT_NULL, name, NULL, | |
2057 | dm_di_mode, error, 0); | |
2058 | } | |
2059 | return error; | |
2060 | ||
2061 | abort_return: | |
2062 | cancel_flags |= XFS_TRANS_ABORT; | |
2063 | /* FALLTHROUGH */ | |
1da177e4 | 2064 | |
014c2544 | 2065 | error_return: |
1da177e4 LT |
2066 | if (tp != NULL) |
2067 | xfs_trans_cancel(tp, cancel_flags); | |
2068 | ||
2069 | if (!dp_joined_to_trans && (dp != NULL)) | |
2070 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2071 | XFS_QM_DQRELE(mp, udqp); | |
2072 | XFS_QM_DQRELE(mp, gdqp); | |
2073 | ||
2074 | goto std_return; | |
2075 | ||
2076 | abort_rele: | |
2077 | /* | |
2078 | * Wait until after the current transaction is aborted to | |
2079 | * release the inode. This prevents recursive transactions | |
2080 | * and deadlocks from xfs_inactive. | |
2081 | */ | |
2082 | cancel_flags |= XFS_TRANS_ABORT; | |
2083 | xfs_trans_cancel(tp, cancel_flags); | |
2084 | IRELE(ip); | |
2085 | ||
2086 | XFS_QM_DQRELE(mp, udqp); | |
2087 | XFS_QM_DQRELE(mp, gdqp); | |
2088 | ||
2089 | goto std_return; | |
2090 | } | |
2091 | ||
2092 | #ifdef DEBUG | |
2093 | /* | |
2094 | * Some counters to see if (and how often) we are hitting some deadlock | |
2095 | * prevention code paths. | |
2096 | */ | |
2097 | ||
2098 | int xfs_rm_locks; | |
2099 | int xfs_rm_lock_delays; | |
2100 | int xfs_rm_attempts; | |
2101 | #endif | |
2102 | ||
2103 | /* | |
2104 | * The following routine will lock the inodes associated with the | |
2105 | * directory and the named entry in the directory. The locks are | |
2106 | * acquired in increasing inode number. | |
2107 | * | |
2108 | * If the entry is "..", then only the directory is locked. The | |
2109 | * vnode ref count will still include that from the .. entry in | |
2110 | * this case. | |
2111 | * | |
2112 | * There is a deadlock we need to worry about. If the locked directory is | |
2113 | * in the AIL, it might be blocking up the log. The next inode we lock | |
2114 | * could be already locked by another thread waiting for log space (e.g | |
2115 | * a permanent log reservation with a long running transaction (see | |
2116 | * xfs_itruncate_finish)). To solve this, we must check if the directory | |
2117 | * is in the ail and use lock_nowait. If we can't lock, we need to | |
2118 | * drop the inode lock on the directory and try again. xfs_iunlock will | |
2119 | * potentially push the tail if we were holding up the log. | |
2120 | */ | |
2121 | STATIC int | |
2122 | xfs_lock_dir_and_entry( | |
2123 | xfs_inode_t *dp, | |
8285fb58 | 2124 | bhv_vname_t *dentry, |
1da177e4 LT |
2125 | xfs_inode_t *ip) /* inode of entry 'name' */ |
2126 | { | |
2127 | int attempts; | |
2128 | xfs_ino_t e_inum; | |
2129 | xfs_inode_t *ips[2]; | |
2130 | xfs_log_item_t *lp; | |
2131 | ||
2132 | #ifdef DEBUG | |
2133 | xfs_rm_locks++; | |
2134 | #endif | |
2135 | attempts = 0; | |
2136 | ||
2137 | again: | |
2138 | xfs_ilock(dp, XFS_ILOCK_EXCL); | |
2139 | ||
2140 | e_inum = ip->i_ino; | |
2141 | ||
2142 | ITRACE(ip); | |
2143 | ||
2144 | /* | |
2145 | * We want to lock in increasing inum. Since we've already | |
2146 | * acquired the lock on the directory, we may need to release | |
2147 | * if if the inum of the entry turns out to be less. | |
2148 | */ | |
2149 | if (e_inum > dp->i_ino) { | |
2150 | /* | |
2151 | * We are already in the right order, so just | |
2152 | * lock on the inode of the entry. | |
2153 | * We need to use nowait if dp is in the AIL. | |
2154 | */ | |
2155 | ||
2156 | lp = (xfs_log_item_t *)dp->i_itemp; | |
2157 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2158 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2159 | attempts++; | |
2160 | #ifdef DEBUG | |
2161 | xfs_rm_attempts++; | |
2162 | #endif | |
2163 | ||
2164 | /* | |
2165 | * Unlock dp and try again. | |
2166 | * xfs_iunlock will try to push the tail | |
2167 | * if the inode is in the AIL. | |
2168 | */ | |
2169 | ||
2170 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2171 | ||
2172 | if ((attempts % 5) == 0) { | |
2173 | delay(1); /* Don't just spin the CPU */ | |
2174 | #ifdef DEBUG | |
2175 | xfs_rm_lock_delays++; | |
2176 | #endif | |
2177 | } | |
2178 | goto again; | |
2179 | } | |
2180 | } else { | |
2181 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2182 | } | |
2183 | } else if (e_inum < dp->i_ino) { | |
2184 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2185 | ||
2186 | ips[0] = ip; | |
2187 | ips[1] = dp; | |
2188 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2189 | } | |
2190 | /* else e_inum == dp->i_ino */ | |
2191 | /* This can happen if we're asked to lock /x/.. | |
2192 | * the entry is "..", which is also the parent directory. | |
2193 | */ | |
2194 | ||
2195 | return 0; | |
2196 | } | |
2197 | ||
2198 | #ifdef DEBUG | |
2199 | int xfs_locked_n; | |
2200 | int xfs_small_retries; | |
2201 | int xfs_middle_retries; | |
2202 | int xfs_lots_retries; | |
2203 | int xfs_lock_delays; | |
2204 | #endif | |
2205 | ||
2206 | /* | |
2207 | * The following routine will lock n inodes in exclusive mode. | |
2208 | * We assume the caller calls us with the inodes in i_ino order. | |
2209 | * | |
2210 | * We need to detect deadlock where an inode that we lock | |
2211 | * is in the AIL and we start waiting for another inode that is locked | |
2212 | * by a thread in a long running transaction (such as truncate). This can | |
2213 | * result in deadlock since the long running trans might need to wait | |
2214 | * for the inode we just locked in order to push the tail and free space | |
2215 | * in the log. | |
2216 | */ | |
2217 | void | |
2218 | xfs_lock_inodes( | |
2219 | xfs_inode_t **ips, | |
2220 | int inodes, | |
2221 | int first_locked, | |
2222 | uint lock_mode) | |
2223 | { | |
2224 | int attempts = 0, i, j, try_lock; | |
2225 | xfs_log_item_t *lp; | |
2226 | ||
2227 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
2228 | ||
2229 | if (first_locked) { | |
2230 | try_lock = 1; | |
2231 | i = 1; | |
2232 | } else { | |
2233 | try_lock = 0; | |
2234 | i = 0; | |
2235 | } | |
2236 | ||
2237 | again: | |
2238 | for (; i < inodes; i++) { | |
2239 | ASSERT(ips[i]); | |
2240 | ||
2241 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
2242 | continue; | |
2243 | ||
2244 | /* | |
2245 | * If try_lock is not set yet, make sure all locked inodes | |
2246 | * are not in the AIL. | |
2247 | * If any are, set try_lock to be used later. | |
2248 | */ | |
2249 | ||
2250 | if (!try_lock) { | |
2251 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
2252 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
2253 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2254 | try_lock++; | |
2255 | } | |
2256 | } | |
2257 | } | |
2258 | ||
2259 | /* | |
2260 | * If any of the previous locks we have locked is in the AIL, | |
2261 | * we must TRY to get the second and subsequent locks. If | |
2262 | * we can't get any, we must release all we have | |
2263 | * and try again. | |
2264 | */ | |
2265 | ||
2266 | if (try_lock) { | |
2267 | /* try_lock must be 0 if i is 0. */ | |
2268 | /* | |
2269 | * try_lock means we have an inode locked | |
2270 | * that is in the AIL. | |
2271 | */ | |
2272 | ASSERT(i != 0); | |
2273 | if (!xfs_ilock_nowait(ips[i], lock_mode)) { | |
2274 | attempts++; | |
2275 | ||
2276 | /* | |
2277 | * Unlock all previous guys and try again. | |
2278 | * xfs_iunlock will try to push the tail | |
2279 | * if the inode is in the AIL. | |
2280 | */ | |
2281 | ||
2282 | for(j = i - 1; j >= 0; j--) { | |
2283 | ||
2284 | /* | |
2285 | * Check to see if we've already | |
2286 | * unlocked this one. | |
2287 | * Not the first one going back, | |
2288 | * and the inode ptr is the same. | |
2289 | */ | |
2290 | if ((j != (i - 1)) && ips[j] == | |
2291 | ips[j+1]) | |
2292 | continue; | |
2293 | ||
2294 | xfs_iunlock(ips[j], lock_mode); | |
2295 | } | |
2296 | ||
2297 | if ((attempts % 5) == 0) { | |
2298 | delay(1); /* Don't just spin the CPU */ | |
2299 | #ifdef DEBUG | |
2300 | xfs_lock_delays++; | |
2301 | #endif | |
2302 | } | |
2303 | i = 0; | |
2304 | try_lock = 0; | |
2305 | goto again; | |
2306 | } | |
2307 | } else { | |
2308 | xfs_ilock(ips[i], lock_mode); | |
2309 | } | |
2310 | } | |
2311 | ||
2312 | #ifdef DEBUG | |
2313 | if (attempts) { | |
2314 | if (attempts < 5) xfs_small_retries++; | |
2315 | else if (attempts < 100) xfs_middle_retries++; | |
2316 | else xfs_lots_retries++; | |
2317 | } else { | |
2318 | xfs_locked_n++; | |
2319 | } | |
2320 | #endif | |
2321 | } | |
2322 | ||
2323 | #ifdef DEBUG | |
2324 | #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);} | |
2325 | int remove_which_error_return = 0; | |
2326 | #else /* ! DEBUG */ | |
2327 | #define REMOVE_DEBUG_TRACE(x) | |
2328 | #endif /* ! DEBUG */ | |
2329 | ||
2330 | ||
2331 | /* | |
2332 | * xfs_remove | |
2333 | * | |
2334 | */ | |
2335 | STATIC int | |
2336 | xfs_remove( | |
2337 | bhv_desc_t *dir_bdp, | |
8285fb58 | 2338 | bhv_vname_t *dentry, |
1da177e4 LT |
2339 | cred_t *credp) |
2340 | { | |
67fcaa73 | 2341 | bhv_vnode_t *dir_vp; |
1da177e4 LT |
2342 | char *name = VNAME(dentry); |
2343 | xfs_inode_t *dp, *ip; | |
2344 | xfs_trans_t *tp = NULL; | |
2345 | xfs_mount_t *mp; | |
2346 | int error = 0; | |
2347 | xfs_bmap_free_t free_list; | |
2348 | xfs_fsblock_t first_block; | |
2349 | int cancel_flags; | |
2350 | int committed; | |
2351 | int dm_di_mode = 0; | |
2352 | int link_zero; | |
2353 | uint resblks; | |
2354 | int namelen; | |
2355 | ||
2356 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
2357 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
2358 | ||
2359 | dp = XFS_BHVTOI(dir_bdp); | |
2360 | mp = dp->i_mount; | |
2361 | ||
2362 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2363 | return XFS_ERROR(EIO); | |
2364 | ||
2365 | namelen = VNAMELEN(dentry); | |
2366 | ||
17370097 VA |
2367 | if (!xfs_get_dir_entry(dentry, &ip)) { |
2368 | dm_di_mode = ip->i_d.di_mode; | |
2369 | IRELE(ip); | |
2370 | } | |
2371 | ||
1da177e4 LT |
2372 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) { |
2373 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp, | |
2374 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
17370097 | 2375 | name, NULL, dm_di_mode, 0, 0); |
1da177e4 LT |
2376 | if (error) |
2377 | return error; | |
2378 | } | |
2379 | ||
2380 | /* From this point on, return through std_return */ | |
2381 | ip = NULL; | |
2382 | ||
2383 | /* | |
2384 | * We need to get a reference to ip before we get our log | |
2385 | * reservation. The reason for this is that we cannot call | |
2386 | * xfs_iget for an inode for which we do not have a reference | |
2387 | * once we've acquired a log reservation. This is because the | |
2388 | * inode we are trying to get might be in xfs_inactive going | |
2389 | * for a log reservation. Since we'll have to wait for the | |
2390 | * inactive code to complete before returning from xfs_iget, | |
2391 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 2392 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
2393 | * to the inode before getting our log reservation. |
2394 | */ | |
2395 | error = xfs_get_dir_entry(dentry, &ip); | |
2396 | if (error) { | |
2397 | REMOVE_DEBUG_TRACE(__LINE__); | |
2398 | goto std_return; | |
2399 | } | |
2400 | ||
2401 | dm_di_mode = ip->i_d.di_mode; | |
2402 | ||
2403 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
2404 | ||
2405 | ITRACE(ip); | |
2406 | ||
2407 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2408 | if (!error && dp != ip) | |
2409 | error = XFS_QM_DQATTACH(mp, ip, 0); | |
2410 | if (error) { | |
2411 | REMOVE_DEBUG_TRACE(__LINE__); | |
2412 | IRELE(ip); | |
2413 | goto std_return; | |
2414 | } | |
2415 | ||
2416 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2417 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2418 | /* | |
2419 | * We try to get the real space reservation first, | |
2420 | * allowing for directory btree deletion(s) implying | |
2421 | * possible bmap insert(s). If we can't get the space | |
2422 | * reservation then we use 0 instead, and avoid the bmap | |
2423 | * btree insert(s) in the directory code by, if the bmap | |
2424 | * insert tries to happen, instead trimming the LAST | |
2425 | * block from the directory. | |
2426 | */ | |
2427 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2428 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2429 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2430 | if (error == ENOSPC) { | |
2431 | resblks = 0; | |
2432 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2433 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2434 | } | |
2435 | if (error) { | |
2436 | ASSERT(error != ENOSPC); | |
2437 | REMOVE_DEBUG_TRACE(__LINE__); | |
2438 | xfs_trans_cancel(tp, 0); | |
2439 | IRELE(ip); | |
2440 | return error; | |
2441 | } | |
2442 | ||
2443 | error = xfs_lock_dir_and_entry(dp, dentry, ip); | |
2444 | if (error) { | |
2445 | REMOVE_DEBUG_TRACE(__LINE__); | |
2446 | xfs_trans_cancel(tp, cancel_flags); | |
2447 | IRELE(ip); | |
2448 | goto std_return; | |
2449 | } | |
2450 | ||
2451 | /* | |
2452 | * At this point, we've gotten both the directory and the entry | |
2453 | * inodes locked. | |
2454 | */ | |
2455 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2456 | if (dp != ip) { | |
2457 | /* | |
2458 | * Increment vnode ref count only in this case since | |
2459 | * there's an extra vnode reference in the case where | |
2460 | * dp == ip. | |
2461 | */ | |
2462 | IHOLD(dp); | |
2463 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2464 | } | |
2465 | ||
2466 | /* | |
2467 | * Entry must exist since we did a lookup in xfs_lock_dir_and_entry. | |
2468 | */ | |
2469 | XFS_BMAP_INIT(&free_list, &first_block); | |
f6c2d1fa NS |
2470 | error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino, |
2471 | &first_block, &free_list, 0); | |
1da177e4 LT |
2472 | if (error) { |
2473 | ASSERT(error != ENOENT); | |
2474 | REMOVE_DEBUG_TRACE(__LINE__); | |
2475 | goto error1; | |
2476 | } | |
2477 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2478 | ||
2479 | dp->i_gen++; | |
2480 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2481 | ||
2482 | error = xfs_droplink(tp, ip); | |
2483 | if (error) { | |
2484 | REMOVE_DEBUG_TRACE(__LINE__); | |
2485 | goto error1; | |
2486 | } | |
2487 | ||
2488 | /* Determine if this is the last link while | |
2489 | * we are in the transaction. | |
2490 | */ | |
2491 | link_zero = (ip)->i_d.di_nlink==0; | |
2492 | ||
2493 | /* | |
2494 | * Take an extra ref on the inode so that it doesn't | |
2495 | * go to xfs_inactive() from within the commit. | |
2496 | */ | |
2497 | IHOLD(ip); | |
2498 | ||
2499 | /* | |
2500 | * If this is a synchronous mount, make sure that the | |
2501 | * remove transaction goes to disk before returning to | |
2502 | * the user. | |
2503 | */ | |
2504 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2505 | xfs_trans_set_sync(tp); | |
2506 | } | |
2507 | ||
f7c99b6f | 2508 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2509 | if (error) { |
2510 | REMOVE_DEBUG_TRACE(__LINE__); | |
2511 | goto error_rele; | |
2512 | } | |
2513 | ||
2514 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
2515 | if (error) { | |
2516 | IRELE(ip); | |
2517 | goto std_return; | |
2518 | } | |
2519 | ||
2520 | /* | |
2521 | * Before we drop our extra reference to the inode, purge it | |
2522 | * from the refcache if it is there. By waiting until afterwards | |
2523 | * to do the IRELE, we ensure that we won't go inactive in the | |
2524 | * xfs_refcache_purge_ip routine (although that would be OK). | |
2525 | */ | |
2526 | xfs_refcache_purge_ip(ip); | |
2527 | ||
2528 | vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
2529 | ||
2530 | /* | |
2531 | * Let interposed file systems know about removed links. | |
2532 | */ | |
67fcaa73 | 2533 | bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero); |
1da177e4 LT |
2534 | |
2535 | IRELE(ip); | |
2536 | ||
2537 | /* Fall through to std_return with error = 0 */ | |
2538 | std_return: | |
2539 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, | |
2540 | DM_EVENT_POSTREMOVE)) { | |
2541 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, | |
2542 | dir_vp, DM_RIGHT_NULL, | |
2543 | NULL, DM_RIGHT_NULL, | |
2544 | name, NULL, dm_di_mode, error, 0); | |
2545 | } | |
2546 | return error; | |
2547 | ||
2548 | error1: | |
2549 | xfs_bmap_cancel(&free_list); | |
2550 | cancel_flags |= XFS_TRANS_ABORT; | |
2551 | xfs_trans_cancel(tp, cancel_flags); | |
2552 | goto std_return; | |
2553 | ||
2554 | error_rele: | |
2555 | /* | |
2556 | * In this case make sure to not release the inode until after | |
2557 | * the current transaction is aborted. Releasing it beforehand | |
2558 | * can cause us to go to xfs_inactive and start a recursive | |
2559 | * transaction which can easily deadlock with the current one. | |
2560 | */ | |
2561 | xfs_bmap_cancel(&free_list); | |
2562 | cancel_flags |= XFS_TRANS_ABORT; | |
2563 | xfs_trans_cancel(tp, cancel_flags); | |
2564 | ||
2565 | /* | |
2566 | * Before we drop our extra reference to the inode, purge it | |
2567 | * from the refcache if it is there. By waiting until afterwards | |
2568 | * to do the IRELE, we ensure that we won't go inactive in the | |
2569 | * xfs_refcache_purge_ip routine (although that would be OK). | |
2570 | */ | |
2571 | xfs_refcache_purge_ip(ip); | |
2572 | ||
2573 | IRELE(ip); | |
2574 | ||
2575 | goto std_return; | |
2576 | } | |
2577 | ||
2578 | ||
2579 | /* | |
2580 | * xfs_link | |
2581 | * | |
2582 | */ | |
2583 | STATIC int | |
2584 | xfs_link( | |
2585 | bhv_desc_t *target_dir_bdp, | |
67fcaa73 | 2586 | bhv_vnode_t *src_vp, |
8285fb58 | 2587 | bhv_vname_t *dentry, |
1da177e4 LT |
2588 | cred_t *credp) |
2589 | { | |
2590 | xfs_inode_t *tdp, *sip; | |
2591 | xfs_trans_t *tp; | |
2592 | xfs_mount_t *mp; | |
2593 | xfs_inode_t *ips[2]; | |
2594 | int error; | |
2595 | xfs_bmap_free_t free_list; | |
2596 | xfs_fsblock_t first_block; | |
2597 | int cancel_flags; | |
2598 | int committed; | |
67fcaa73 | 2599 | bhv_vnode_t *target_dir_vp; |
1da177e4 LT |
2600 | int resblks; |
2601 | char *target_name = VNAME(dentry); | |
2602 | int target_namelen; | |
2603 | ||
2604 | target_dir_vp = BHV_TO_VNODE(target_dir_bdp); | |
2605 | vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
2606 | vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address); | |
2607 | ||
2608 | target_namelen = VNAMELEN(dentry); | |
b71d300c | 2609 | ASSERT(!VN_ISDIR(src_vp)); |
1da177e4 | 2610 | |
75e17b3c | 2611 | sip = xfs_vtoi(src_vp); |
1da177e4 LT |
2612 | tdp = XFS_BHVTOI(target_dir_bdp); |
2613 | mp = tdp->i_mount; | |
2614 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2615 | return XFS_ERROR(EIO); | |
2616 | ||
2617 | if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) { | |
2618 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, | |
2619 | target_dir_vp, DM_RIGHT_NULL, | |
2620 | src_vp, DM_RIGHT_NULL, | |
2621 | target_name, NULL, 0, 0, 0); | |
2622 | if (error) | |
2623 | return error; | |
2624 | } | |
2625 | ||
2626 | /* Return through std_return after this point. */ | |
2627 | ||
2628 | error = XFS_QM_DQATTACH(mp, sip, 0); | |
2629 | if (!error && sip != tdp) | |
2630 | error = XFS_QM_DQATTACH(mp, tdp, 0); | |
2631 | if (error) | |
2632 | goto std_return; | |
2633 | ||
2634 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2635 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2636 | resblks = XFS_LINK_SPACE_RES(mp, target_namelen); | |
2637 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, | |
2638 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2639 | if (error == ENOSPC) { | |
2640 | resblks = 0; | |
2641 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2642 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2643 | } | |
2644 | if (error) { | |
2645 | cancel_flags = 0; | |
2646 | goto error_return; | |
2647 | } | |
2648 | ||
2649 | if (sip->i_ino < tdp->i_ino) { | |
2650 | ips[0] = sip; | |
2651 | ips[1] = tdp; | |
2652 | } else { | |
2653 | ips[0] = tdp; | |
2654 | ips[1] = sip; | |
2655 | } | |
2656 | ||
2657 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2658 | ||
2659 | /* | |
2660 | * Increment vnode ref counts since xfs_trans_commit & | |
2661 | * xfs_trans_cancel will both unlock the inodes and | |
2662 | * decrement the associated ref counts. | |
2663 | */ | |
2664 | VN_HOLD(src_vp); | |
2665 | VN_HOLD(target_dir_vp); | |
2666 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
2667 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2668 | ||
2669 | /* | |
2670 | * If the source has too many links, we can't make any more to it. | |
2671 | */ | |
2672 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2673 | error = XFS_ERROR(EMLINK); | |
2674 | goto error_return; | |
2675 | } | |
2676 | ||
365ca83d NS |
2677 | /* |
2678 | * If we are using project inheritance, we only allow hard link | |
2679 | * creation in our tree when the project IDs are the same; else | |
2680 | * the tree quota mechanism could be circumvented. | |
2681 | */ | |
2682 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2683 | (tdp->i_d.di_projid != sip->i_d.di_projid))) { | |
b1ecdda9 | 2684 | error = XFS_ERROR(EXDEV); |
365ca83d NS |
2685 | goto error_return; |
2686 | } | |
2687 | ||
1da177e4 | 2688 | if (resblks == 0 && |
f6c2d1fa | 2689 | (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen))) |
1da177e4 LT |
2690 | goto error_return; |
2691 | ||
2692 | XFS_BMAP_INIT(&free_list, &first_block); | |
2693 | ||
f6c2d1fa | 2694 | error = xfs_dir_createname(tp, tdp, target_name, target_namelen, |
1da177e4 LT |
2695 | sip->i_ino, &first_block, &free_list, |
2696 | resblks); | |
2697 | if (error) | |
2698 | goto abort_return; | |
2699 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2700 | tdp->i_gen++; | |
2701 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
2702 | ||
2703 | error = xfs_bumplink(tp, sip); | |
b71d300c | 2704 | if (error) |
1da177e4 | 2705 | goto abort_return; |
1da177e4 LT |
2706 | |
2707 | /* | |
2708 | * If this is a synchronous mount, make sure that the | |
2709 | * link transaction goes to disk before returning to | |
2710 | * the user. | |
2711 | */ | |
2712 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2713 | xfs_trans_set_sync(tp); | |
2714 | } | |
2715 | ||
f7c99b6f | 2716 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
2717 | if (error) { |
2718 | xfs_bmap_cancel(&free_list); | |
2719 | goto abort_return; | |
2720 | } | |
2721 | ||
2722 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
b71d300c | 2723 | if (error) |
1da177e4 | 2724 | goto std_return; |
1da177e4 LT |
2725 | |
2726 | /* Fall through to std_return with error = 0. */ | |
2727 | std_return: | |
2728 | if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip, | |
2729 | DM_EVENT_POSTLINK)) { | |
2730 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, | |
2731 | target_dir_vp, DM_RIGHT_NULL, | |
2732 | src_vp, DM_RIGHT_NULL, | |
2733 | target_name, NULL, 0, error, 0); | |
2734 | } | |
2735 | return error; | |
2736 | ||
2737 | abort_return: | |
2738 | cancel_flags |= XFS_TRANS_ABORT; | |
2739 | /* FALLTHROUGH */ | |
014c2544 | 2740 | |
1da177e4 LT |
2741 | error_return: |
2742 | xfs_trans_cancel(tp, cancel_flags); | |
1da177e4 LT |
2743 | goto std_return; |
2744 | } | |
b71d300c AD |
2745 | |
2746 | ||
1da177e4 LT |
2747 | /* |
2748 | * xfs_mkdir | |
2749 | * | |
2750 | */ | |
2751 | STATIC int | |
2752 | xfs_mkdir( | |
2753 | bhv_desc_t *dir_bdp, | |
8285fb58 NS |
2754 | bhv_vname_t *dentry, |
2755 | bhv_vattr_t *vap, | |
67fcaa73 | 2756 | bhv_vnode_t **vpp, |
1da177e4 LT |
2757 | cred_t *credp) |
2758 | { | |
2759 | char *dir_name = VNAME(dentry); | |
2760 | xfs_inode_t *dp; | |
2761 | xfs_inode_t *cdp; /* inode of created dir */ | |
67fcaa73 | 2762 | bhv_vnode_t *cvp; /* vnode of created dir */ |
1da177e4 LT |
2763 | xfs_trans_t *tp; |
2764 | xfs_mount_t *mp; | |
2765 | int cancel_flags; | |
2766 | int error; | |
2767 | int committed; | |
2768 | xfs_bmap_free_t free_list; | |
2769 | xfs_fsblock_t first_block; | |
67fcaa73 | 2770 | bhv_vnode_t *dir_vp; |
1da177e4 LT |
2771 | boolean_t dp_joined_to_trans; |
2772 | boolean_t created = B_FALSE; | |
2773 | int dm_event_sent = 0; | |
2774 | xfs_prid_t prid; | |
2775 | struct xfs_dquot *udqp, *gdqp; | |
2776 | uint resblks; | |
2777 | int dm_di_mode; | |
2778 | int dir_namelen; | |
2779 | ||
2780 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
2781 | dp = XFS_BHVTOI(dir_bdp); | |
2782 | mp = dp->i_mount; | |
2783 | ||
2784 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2785 | return XFS_ERROR(EIO); | |
2786 | ||
2787 | dir_namelen = VNAMELEN(dentry); | |
2788 | ||
2789 | tp = NULL; | |
2790 | dp_joined_to_trans = B_FALSE; | |
0432dab2 | 2791 | dm_di_mode = vap->va_mode; |
1da177e4 LT |
2792 | |
2793 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) { | |
2794 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, | |
2795 | dir_vp, DM_RIGHT_NULL, NULL, | |
2796 | DM_RIGHT_NULL, dir_name, NULL, | |
2797 | dm_di_mode, 0, 0); | |
2798 | if (error) | |
2799 | return error; | |
2800 | dm_event_sent = 1; | |
2801 | } | |
2802 | ||
2803 | /* Return through std_return after this point. */ | |
2804 | ||
2805 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
2806 | ||
2807 | mp = dp->i_mount; | |
2808 | udqp = gdqp = NULL; | |
365ca83d NS |
2809 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
2810 | prid = dp->i_d.di_projid; | |
2811 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
2812 | prid = (xfs_prid_t)vap->va_projid; |
2813 | else | |
2814 | prid = (xfs_prid_t)dfltprid; | |
2815 | ||
2816 | /* | |
2817 | * Make sure that we have allocated dquot(s) on disk. | |
2818 | */ | |
2819 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 2820 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
2821 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
2822 | if (error) | |
2823 | goto std_return; | |
2824 | ||
2825 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
2826 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2827 | resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen); | |
2828 | error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0, | |
2829 | XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT); | |
2830 | if (error == ENOSPC) { | |
2831 | resblks = 0; | |
2832 | error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0, | |
2833 | XFS_TRANS_PERM_LOG_RES, | |
2834 | XFS_MKDIR_LOG_COUNT); | |
2835 | } | |
2836 | if (error) { | |
2837 | cancel_flags = 0; | |
2838 | dp = NULL; | |
2839 | goto error_return; | |
2840 | } | |
2841 | ||
2842 | xfs_ilock(dp, XFS_ILOCK_EXCL); | |
2843 | ||
2844 | /* | |
2845 | * Check for directory link count overflow. | |
2846 | */ | |
2847 | if (dp->i_d.di_nlink >= XFS_MAXLINK) { | |
2848 | error = XFS_ERROR(EMLINK); | |
2849 | goto error_return; | |
2850 | } | |
2851 | ||
2852 | /* | |
2853 | * Reserve disk quota and the inode. | |
2854 | */ | |
2855 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
2856 | if (error) | |
2857 | goto error_return; | |
2858 | ||
2859 | if (resblks == 0 && | |
f6c2d1fa | 2860 | (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen))) |
1da177e4 LT |
2861 | goto error_return; |
2862 | /* | |
2863 | * create the directory inode. | |
2864 | */ | |
0432dab2 | 2865 | error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2, |
1da177e4 LT |
2866 | 0, credp, prid, resblks > 0, |
2867 | &cdp, NULL); | |
2868 | if (error) { | |
2869 | if (error == ENOSPC) | |
2870 | goto error_return; | |
2871 | goto abort_return; | |
2872 | } | |
2873 | ITRACE(cdp); | |
2874 | ||
2875 | /* | |
2876 | * Now we add the directory inode to the transaction. | |
2877 | * We waited until now since xfs_dir_ialloc might start | |
2878 | * a new transaction. Had we joined the transaction | |
2879 | * earlier, the locks might have gotten released. | |
2880 | */ | |
2881 | VN_HOLD(dir_vp); | |
2882 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2883 | dp_joined_to_trans = B_TRUE; | |
2884 | ||
2885 | XFS_BMAP_INIT(&free_list, &first_block); | |
2886 | ||
f6c2d1fa NS |
2887 | error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino, |
2888 | &first_block, &free_list, resblks ? | |
2889 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1da177e4 LT |
2890 | if (error) { |
2891 | ASSERT(error != ENOSPC); | |
2892 | goto error1; | |
2893 | } | |
2894 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2895 | ||
2896 | /* | |
2897 | * Bump the in memory version number of the parent directory | |
2898 | * so that other processes accessing it will recognize that | |
2899 | * the directory has changed. | |
2900 | */ | |
2901 | dp->i_gen++; | |
2902 | ||
f6c2d1fa NS |
2903 | error = xfs_dir_init(tp, cdp, dp); |
2904 | if (error) | |
1da177e4 | 2905 | goto error2; |
1da177e4 LT |
2906 | |
2907 | cdp->i_gen = 1; | |
2908 | error = xfs_bumplink(tp, dp); | |
f6c2d1fa | 2909 | if (error) |
1da177e4 | 2910 | goto error2; |
1da177e4 LT |
2911 | |
2912 | cvp = XFS_ITOV(cdp); | |
2913 | ||
2914 | created = B_TRUE; | |
2915 | ||
2916 | *vpp = cvp; | |
2917 | IHOLD(cdp); | |
2918 | ||
2919 | /* | |
2920 | * Attach the dquots to the new inode and modify the icount incore. | |
2921 | */ | |
2922 | XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp); | |
2923 | ||
2924 | /* | |
2925 | * If this is a synchronous mount, make sure that the | |
2926 | * mkdir transaction goes to disk before returning to | |
2927 | * the user. | |
2928 | */ | |
2929 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2930 | xfs_trans_set_sync(tp); | |
2931 | } | |
2932 | ||
f7c99b6f | 2933 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
2934 | if (error) { |
2935 | IRELE(cdp); | |
2936 | goto error2; | |
2937 | } | |
2938 | ||
2939 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
2940 | XFS_QM_DQRELE(mp, udqp); | |
2941 | XFS_QM_DQRELE(mp, gdqp); | |
2942 | if (error) { | |
2943 | IRELE(cdp); | |
2944 | } | |
2945 | ||
2946 | /* Fall through to std_return with error = 0 or errno from | |
2947 | * xfs_trans_commit. */ | |
2948 | ||
2949 | std_return: | |
2950 | if ( (created || (error != 0 && dm_event_sent != 0)) && | |
2951 | DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp), | |
2952 | DM_EVENT_POSTCREATE)) { | |
2953 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, | |
2954 | dir_vp, DM_RIGHT_NULL, | |
2955 | created ? XFS_ITOV(cdp):NULL, | |
2956 | DM_RIGHT_NULL, | |
2957 | dir_name, NULL, | |
2958 | dm_di_mode, error, 0); | |
2959 | } | |
2960 | return error; | |
2961 | ||
2962 | error2: | |
2963 | error1: | |
2964 | xfs_bmap_cancel(&free_list); | |
2965 | abort_return: | |
2966 | cancel_flags |= XFS_TRANS_ABORT; | |
2967 | error_return: | |
2968 | xfs_trans_cancel(tp, cancel_flags); | |
2969 | XFS_QM_DQRELE(mp, udqp); | |
2970 | XFS_QM_DQRELE(mp, gdqp); | |
2971 | ||
2972 | if (!dp_joined_to_trans && (dp != NULL)) { | |
2973 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2974 | } | |
2975 | ||
2976 | goto std_return; | |
2977 | } | |
2978 | ||
2979 | ||
2980 | /* | |
2981 | * xfs_rmdir | |
2982 | * | |
2983 | */ | |
2984 | STATIC int | |
2985 | xfs_rmdir( | |
2986 | bhv_desc_t *dir_bdp, | |
8285fb58 | 2987 | bhv_vname_t *dentry, |
1da177e4 LT |
2988 | cred_t *credp) |
2989 | { | |
2990 | char *name = VNAME(dentry); | |
2991 | xfs_inode_t *dp; | |
2992 | xfs_inode_t *cdp; /* child directory */ | |
2993 | xfs_trans_t *tp; | |
2994 | xfs_mount_t *mp; | |
2995 | int error; | |
2996 | xfs_bmap_free_t free_list; | |
2997 | xfs_fsblock_t first_block; | |
2998 | int cancel_flags; | |
2999 | int committed; | |
67fcaa73 | 3000 | bhv_vnode_t *dir_vp; |
17370097 | 3001 | int dm_di_mode = S_IFDIR; |
1da177e4 LT |
3002 | int last_cdp_link; |
3003 | int namelen; | |
3004 | uint resblks; | |
3005 | ||
3006 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
3007 | dp = XFS_BHVTOI(dir_bdp); | |
3008 | mp = dp->i_mount; | |
3009 | ||
3010 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
3011 | ||
3012 | if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount)) | |
3013 | return XFS_ERROR(EIO); | |
3014 | namelen = VNAMELEN(dentry); | |
3015 | ||
17370097 VA |
3016 | if (!xfs_get_dir_entry(dentry, &cdp)) { |
3017 | dm_di_mode = cdp->i_d.di_mode; | |
3018 | IRELE(cdp); | |
3019 | } | |
3020 | ||
1da177e4 LT |
3021 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) { |
3022 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, | |
3023 | dir_vp, DM_RIGHT_NULL, | |
3024 | NULL, DM_RIGHT_NULL, | |
17370097 | 3025 | name, NULL, dm_di_mode, 0, 0); |
1da177e4 LT |
3026 | if (error) |
3027 | return XFS_ERROR(error); | |
3028 | } | |
3029 | ||
3030 | /* Return through std_return after this point. */ | |
3031 | ||
3032 | cdp = NULL; | |
3033 | ||
3034 | /* | |
3035 | * We need to get a reference to cdp before we get our log | |
3036 | * reservation. The reason for this is that we cannot call | |
3037 | * xfs_iget for an inode for which we do not have a reference | |
3038 | * once we've acquired a log reservation. This is because the | |
3039 | * inode we are trying to get might be in xfs_inactive going | |
3040 | * for a log reservation. Since we'll have to wait for the | |
3041 | * inactive code to complete before returning from xfs_iget, | |
3042 | * we need to make sure that we don't have log space reserved | |
c41564b5 | 3043 | * when we call xfs_iget. Instead we get an unlocked reference |
1da177e4 LT |
3044 | * to the inode before getting our log reservation. |
3045 | */ | |
3046 | error = xfs_get_dir_entry(dentry, &cdp); | |
3047 | if (error) { | |
3048 | REMOVE_DEBUG_TRACE(__LINE__); | |
3049 | goto std_return; | |
3050 | } | |
3051 | mp = dp->i_mount; | |
3052 | dm_di_mode = cdp->i_d.di_mode; | |
3053 | ||
3054 | /* | |
3055 | * Get the dquots for the inodes. | |
3056 | */ | |
3057 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
3058 | if (!error && dp != cdp) | |
3059 | error = XFS_QM_DQATTACH(mp, cdp, 0); | |
3060 | if (error) { | |
3061 | IRELE(cdp); | |
3062 | REMOVE_DEBUG_TRACE(__LINE__); | |
3063 | goto std_return; | |
3064 | } | |
3065 | ||
3066 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
3067 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3068 | /* | |
3069 | * We try to get the real space reservation first, | |
3070 | * allowing for directory btree deletion(s) implying | |
3071 | * possible bmap insert(s). If we can't get the space | |
3072 | * reservation then we use 0 instead, and avoid the bmap | |
3073 | * btree insert(s) in the directory code by, if the bmap | |
3074 | * insert tries to happen, instead trimming the LAST | |
3075 | * block from the directory. | |
3076 | */ | |
3077 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
3078 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
3079 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
3080 | if (error == ENOSPC) { | |
3081 | resblks = 0; | |
3082 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
3083 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
3084 | } | |
3085 | if (error) { | |
3086 | ASSERT(error != ENOSPC); | |
3087 | cancel_flags = 0; | |
3088 | IRELE(cdp); | |
3089 | goto error_return; | |
3090 | } | |
3091 | XFS_BMAP_INIT(&free_list, &first_block); | |
3092 | ||
3093 | /* | |
3094 | * Now lock the child directory inode and the parent directory | |
3095 | * inode in the proper order. This will take care of validating | |
3096 | * that the directory entry for the child directory inode has | |
3097 | * not changed while we were obtaining a log reservation. | |
3098 | */ | |
3099 | error = xfs_lock_dir_and_entry(dp, dentry, cdp); | |
3100 | if (error) { | |
3101 | xfs_trans_cancel(tp, cancel_flags); | |
3102 | IRELE(cdp); | |
3103 | goto std_return; | |
3104 | } | |
3105 | ||
3106 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
3107 | if (dp != cdp) { | |
3108 | /* | |
3109 | * Only increment the parent directory vnode count if | |
3110 | * we didn't bump it in looking up cdp. The only time | |
3111 | * we don't bump it is when we're looking up ".". | |
3112 | */ | |
3113 | VN_HOLD(dir_vp); | |
3114 | } | |
3115 | ||
3116 | ITRACE(cdp); | |
3117 | xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL); | |
3118 | ||
3119 | ASSERT(cdp->i_d.di_nlink >= 2); | |
3120 | if (cdp->i_d.di_nlink != 2) { | |
3121 | error = XFS_ERROR(ENOTEMPTY); | |
3122 | goto error_return; | |
3123 | } | |
f6c2d1fa | 3124 | if (!xfs_dir_isempty(cdp)) { |
1da177e4 LT |
3125 | error = XFS_ERROR(ENOTEMPTY); |
3126 | goto error_return; | |
3127 | } | |
3128 | ||
f6c2d1fa NS |
3129 | error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino, |
3130 | &first_block, &free_list, resblks); | |
3131 | if (error) | |
1da177e4 | 3132 | goto error1; |
1da177e4 LT |
3133 | |
3134 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3135 | ||
3136 | /* | |
3137 | * Bump the in memory generation count on the parent | |
3138 | * directory so that other can know that it has changed. | |
3139 | */ | |
3140 | dp->i_gen++; | |
3141 | ||
3142 | /* | |
3143 | * Drop the link from cdp's "..". | |
3144 | */ | |
3145 | error = xfs_droplink(tp, dp); | |
3146 | if (error) { | |
3147 | goto error1; | |
3148 | } | |
3149 | ||
3150 | /* | |
3151 | * Drop the link from dp to cdp. | |
3152 | */ | |
3153 | error = xfs_droplink(tp, cdp); | |
3154 | if (error) { | |
3155 | goto error1; | |
3156 | } | |
3157 | ||
3158 | /* | |
3159 | * Drop the "." link from cdp to self. | |
3160 | */ | |
3161 | error = xfs_droplink(tp, cdp); | |
3162 | if (error) { | |
3163 | goto error1; | |
3164 | } | |
3165 | ||
3166 | /* Determine these before committing transaction */ | |
3167 | last_cdp_link = (cdp)->i_d.di_nlink==0; | |
3168 | ||
3169 | /* | |
3170 | * Take an extra ref on the child vnode so that it | |
3171 | * does not go to xfs_inactive() from within the commit. | |
3172 | */ | |
3173 | IHOLD(cdp); | |
3174 | ||
3175 | /* | |
3176 | * If this is a synchronous mount, make sure that the | |
3177 | * rmdir transaction goes to disk before returning to | |
3178 | * the user. | |
3179 | */ | |
3180 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3181 | xfs_trans_set_sync(tp); | |
3182 | } | |
3183 | ||
f7c99b6f | 3184 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
1da177e4 LT |
3185 | if (error) { |
3186 | xfs_bmap_cancel(&free_list); | |
3187 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | | |
3188 | XFS_TRANS_ABORT)); | |
3189 | IRELE(cdp); | |
3190 | goto std_return; | |
3191 | } | |
3192 | ||
3193 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
3194 | if (error) { | |
3195 | IRELE(cdp); | |
3196 | goto std_return; | |
3197 | } | |
3198 | ||
3199 | ||
3200 | /* | |
3201 | * Let interposed file systems know about removed links. | |
3202 | */ | |
67fcaa73 | 3203 | bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link); |
1da177e4 LT |
3204 | |
3205 | IRELE(cdp); | |
3206 | ||
3207 | /* Fall through to std_return with error = 0 or the errno | |
3208 | * from xfs_trans_commit. */ | |
bb19fba1 | 3209 | std_return: |
1da177e4 LT |
3210 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) { |
3211 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, | |
3212 | dir_vp, DM_RIGHT_NULL, | |
3213 | NULL, DM_RIGHT_NULL, | |
3214 | name, NULL, dm_di_mode, | |
3215 | error, 0); | |
3216 | } | |
3217 | return error; | |
3218 | ||
bb19fba1 | 3219 | error1: |
1da177e4 LT |
3220 | xfs_bmap_cancel(&free_list); |
3221 | cancel_flags |= XFS_TRANS_ABORT; | |
014c2544 JJ |
3222 | /* FALLTHROUGH */ |
3223 | ||
bb19fba1 | 3224 | error_return: |
1da177e4 LT |
3225 | xfs_trans_cancel(tp, cancel_flags); |
3226 | goto std_return; | |
3227 | } | |
3228 | ||
3229 | ||
3230 | /* | |
1da177e4 LT |
3231 | * Read dp's entries starting at uiop->uio_offset and translate them into |
3232 | * bufsize bytes worth of struct dirents starting at bufbase. | |
3233 | */ | |
3234 | STATIC int | |
3235 | xfs_readdir( | |
3236 | bhv_desc_t *dir_bdp, | |
3237 | uio_t *uiop, | |
3238 | cred_t *credp, | |
3239 | int *eofp) | |
3240 | { | |
3241 | xfs_inode_t *dp; | |
3242 | xfs_trans_t *tp = NULL; | |
3243 | int error = 0; | |
3244 | uint lock_mode; | |
1da177e4 LT |
3245 | |
3246 | vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__, | |
3247 | (inst_t *)__return_address); | |
3248 | dp = XFS_BHVTOI(dir_bdp); | |
3249 | ||
f6c2d1fa | 3250 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) |
1da177e4 | 3251 | return XFS_ERROR(EIO); |
1da177e4 LT |
3252 | |
3253 | lock_mode = xfs_ilock_map_shared(dp); | |
f6c2d1fa | 3254 | error = xfs_dir_getdents(tp, dp, uiop, eofp); |
1da177e4 LT |
3255 | xfs_iunlock_map_shared(dp, lock_mode); |
3256 | return error; | |
3257 | } | |
3258 | ||
3259 | ||
1da177e4 LT |
3260 | STATIC int |
3261 | xfs_symlink( | |
3262 | bhv_desc_t *dir_bdp, | |
8285fb58 NS |
3263 | bhv_vname_t *dentry, |
3264 | bhv_vattr_t *vap, | |
1da177e4 | 3265 | char *target_path, |
67fcaa73 | 3266 | bhv_vnode_t **vpp, |
1da177e4 LT |
3267 | cred_t *credp) |
3268 | { | |
3269 | xfs_trans_t *tp; | |
3270 | xfs_mount_t *mp; | |
3271 | xfs_inode_t *dp; | |
3272 | xfs_inode_t *ip; | |
3273 | int error; | |
3274 | int pathlen; | |
3275 | xfs_bmap_free_t free_list; | |
3276 | xfs_fsblock_t first_block; | |
3277 | boolean_t dp_joined_to_trans; | |
67fcaa73 | 3278 | bhv_vnode_t *dir_vp; |
1da177e4 LT |
3279 | uint cancel_flags; |
3280 | int committed; | |
3281 | xfs_fileoff_t first_fsb; | |
3282 | xfs_filblks_t fs_blocks; | |
3283 | int nmaps; | |
3284 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
3285 | xfs_daddr_t d; | |
3286 | char *cur_chunk; | |
3287 | int byte_cnt; | |
3288 | int n; | |
3289 | xfs_buf_t *bp; | |
3290 | xfs_prid_t prid; | |
3291 | struct xfs_dquot *udqp, *gdqp; | |
3292 | uint resblks; | |
3293 | char *link_name = VNAME(dentry); | |
3294 | int link_namelen; | |
3295 | ||
3296 | *vpp = NULL; | |
3297 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
3298 | dp = XFS_BHVTOI(dir_bdp); | |
3299 | dp_joined_to_trans = B_FALSE; | |
3300 | error = 0; | |
3301 | ip = NULL; | |
3302 | tp = NULL; | |
3303 | ||
3304 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
3305 | ||
3306 | mp = dp->i_mount; | |
3307 | ||
3308 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3309 | return XFS_ERROR(EIO); | |
3310 | ||
3311 | link_namelen = VNAMELEN(dentry); | |
3312 | ||
3313 | /* | |
3314 | * Check component lengths of the target path name. | |
3315 | */ | |
3316 | pathlen = strlen(target_path); | |
3317 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
3318 | return XFS_ERROR(ENAMETOOLONG); | |
3319 | if (pathlen >= MAXNAMELEN) { /* is any component too long? */ | |
3320 | int len, total; | |
3321 | char *path; | |
3322 | ||
f6c2d1fa | 3323 | for (total = 0, path = target_path; total < pathlen;) { |
1da177e4 LT |
3324 | /* |
3325 | * Skip any slashes. | |
3326 | */ | |
3327 | while(*path == '/') { | |
3328 | total++; | |
3329 | path++; | |
3330 | } | |
3331 | ||
3332 | /* | |
3333 | * Count up to the next slash or end of path. | |
3334 | * Error out if the component is bigger than MAXNAMELEN. | |
3335 | */ | |
3336 | for(len = 0; *path != '/' && total < pathlen;total++, path++) { | |
3337 | if (++len >= MAXNAMELEN) { | |
3338 | error = ENAMETOOLONG; | |
3339 | return error; | |
3340 | } | |
3341 | } | |
3342 | } | |
3343 | } | |
3344 | ||
3345 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) { | |
3346 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp, | |
3347 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
3348 | link_name, target_path, 0, 0, 0); | |
3349 | if (error) | |
3350 | return error; | |
3351 | } | |
3352 | ||
3353 | /* Return through std_return after this point. */ | |
3354 | ||
3355 | udqp = gdqp = NULL; | |
365ca83d NS |
3356 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
3357 | prid = dp->i_d.di_projid; | |
3358 | else if (vap->va_mask & XFS_AT_PROJID) | |
1da177e4 LT |
3359 | prid = (xfs_prid_t)vap->va_projid; |
3360 | else | |
3361 | prid = (xfs_prid_t)dfltprid; | |
3362 | ||
3363 | /* | |
3364 | * Make sure that we have allocated dquot(s) on disk. | |
3365 | */ | |
3366 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
c8ad20ff | 3367 | current_fsuid(credp), current_fsgid(credp), prid, |
1da177e4 LT |
3368 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); |
3369 | if (error) | |
3370 | goto std_return; | |
3371 | ||
3372 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
3373 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3374 | /* | |
3375 | * The symlink will fit into the inode data fork? | |
3376 | * There can't be any attributes so we get the whole variable part. | |
3377 | */ | |
3378 | if (pathlen <= XFS_LITINO(mp)) | |
3379 | fs_blocks = 0; | |
3380 | else | |
3381 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
3382 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks); | |
3383 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, | |
3384 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3385 | if (error == ENOSPC && fs_blocks == 0) { | |
3386 | resblks = 0; | |
3387 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
3388 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3389 | } | |
3390 | if (error) { | |
3391 | cancel_flags = 0; | |
3392 | dp = NULL; | |
3393 | goto error_return; | |
3394 | } | |
3395 | ||
3396 | xfs_ilock(dp, XFS_ILOCK_EXCL); | |
3397 | ||
3398 | /* | |
3399 | * Check whether the directory allows new symlinks or not. | |
3400 | */ | |
3401 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
3402 | error = XFS_ERROR(EPERM); | |
3403 | goto error_return; | |
3404 | } | |
3405 | ||
3406 | /* | |
3407 | * Reserve disk quota : blocks and inode. | |
3408 | */ | |
3409 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
3410 | if (error) | |
3411 | goto error_return; | |
3412 | ||
3413 | /* | |
3414 | * Check for ability to enter directory entry, if no space reserved. | |
3415 | */ | |
3416 | if (resblks == 0 && | |
f6c2d1fa | 3417 | (error = xfs_dir_canenter(tp, dp, link_name, link_namelen))) |
1da177e4 LT |
3418 | goto error_return; |
3419 | /* | |
3420 | * Initialize the bmap freelist prior to calling either | |
3421 | * bmapi or the directory create code. | |
3422 | */ | |
3423 | XFS_BMAP_INIT(&free_list, &first_block); | |
3424 | ||
3425 | /* | |
3426 | * Allocate an inode for the symlink. | |
3427 | */ | |
3428 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT), | |
3429 | 1, 0, credp, prid, resblks > 0, &ip, NULL); | |
3430 | if (error) { | |
3431 | if (error == ENOSPC) | |
3432 | goto error_return; | |
3433 | goto error1; | |
3434 | } | |
3435 | ITRACE(ip); | |
3436 | ||
3437 | VN_HOLD(dir_vp); | |
3438 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
3439 | dp_joined_to_trans = B_TRUE; | |
3440 | ||
3441 | /* | |
3442 | * Also attach the dquot(s) to it, if applicable. | |
3443 | */ | |
3444 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
3445 | ||
3446 | if (resblks) | |
3447 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
3448 | /* | |
3449 | * If the symlink will fit into the inode, write it inline. | |
3450 | */ | |
3451 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
3452 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
3453 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
3454 | ip->i_d.di_size = pathlen; | |
3455 | ||
3456 | /* | |
3457 | * The inode was initially created in extent format. | |
3458 | */ | |
3459 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
3460 | ip->i_df.if_flags |= XFS_IFINLINE; | |
3461 | ||
3462 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
3463 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
3464 | ||
3465 | } else { | |
3466 | first_fsb = 0; | |
3467 | nmaps = SYMLINK_MAPS; | |
3468 | ||
3469 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
3470 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
3471 | &first_block, resblks, mval, &nmaps, | |
3e57ecf6 | 3472 | &free_list, NULL); |
1da177e4 LT |
3473 | if (error) { |
3474 | goto error1; | |
3475 | } | |
3476 | ||
3477 | if (resblks) | |
3478 | resblks -= fs_blocks; | |
3479 | ip->i_d.di_size = pathlen; | |
3480 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3481 | ||
3482 | cur_chunk = target_path; | |
3483 | for (n = 0; n < nmaps; n++) { | |
3484 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
3485 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
3486 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
3487 | BTOBB(byte_cnt), 0); | |
3488 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
3489 | if (pathlen < byte_cnt) { | |
3490 | byte_cnt = pathlen; | |
3491 | } | |
3492 | pathlen -= byte_cnt; | |
3493 | ||
3494 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
3495 | cur_chunk += byte_cnt; | |
3496 | ||
3497 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
3498 | } | |
3499 | } | |
3500 | ||
3501 | /* | |
3502 | * Create the directory entry for the symlink. | |
3503 | */ | |
f6c2d1fa NS |
3504 | error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino, |
3505 | &first_block, &free_list, resblks); | |
3506 | if (error) | |
1da177e4 | 3507 | goto error1; |
1da177e4 LT |
3508 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3509 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
3510 | ||
3511 | /* | |
3512 | * Bump the in memory version number of the parent directory | |
3513 | * so that other processes accessing it will recognize that | |
3514 | * the directory has changed. | |
3515 | */ | |
3516 | dp->i_gen++; | |
3517 | ||
3518 | /* | |
3519 | * If this is a synchronous mount, make sure that the | |
3520 | * symlink transaction goes to disk before returning to | |
3521 | * the user. | |
3522 | */ | |
3523 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3524 | xfs_trans_set_sync(tp); | |
3525 | } | |
3526 | ||
3527 | /* | |
3528 | * xfs_trans_commit normally decrements the vnode ref count | |
3529 | * when it unlocks the inode. Since we want to return the | |
3530 | * vnode to the caller, we bump the vnode ref count now. | |
3531 | */ | |
3532 | IHOLD(ip); | |
3533 | ||
f7c99b6f | 3534 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
3535 | if (error) { |
3536 | goto error2; | |
3537 | } | |
3538 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
3539 | XFS_QM_DQRELE(mp, udqp); | |
3540 | XFS_QM_DQRELE(mp, gdqp); | |
3541 | ||
3542 | /* Fall through to std_return with error = 0 or errno from | |
3543 | * xfs_trans_commit */ | |
3544 | std_return: | |
3545 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp), | |
3546 | DM_EVENT_POSTSYMLINK)) { | |
3547 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, | |
3548 | dir_vp, DM_RIGHT_NULL, | |
3549 | error ? NULL : XFS_ITOV(ip), | |
3550 | DM_RIGHT_NULL, link_name, target_path, | |
3551 | 0, error, 0); | |
3552 | } | |
3553 | ||
3554 | if (!error) { | |
67fcaa73 | 3555 | bhv_vnode_t *vp; |
1da177e4 LT |
3556 | |
3557 | ASSERT(ip); | |
3558 | vp = XFS_ITOV(ip); | |
3559 | *vpp = vp; | |
3560 | } | |
3561 | return error; | |
3562 | ||
3563 | error2: | |
3564 | IRELE(ip); | |
3565 | error1: | |
3566 | xfs_bmap_cancel(&free_list); | |
3567 | cancel_flags |= XFS_TRANS_ABORT; | |
3568 | error_return: | |
3569 | xfs_trans_cancel(tp, cancel_flags); | |
3570 | XFS_QM_DQRELE(mp, udqp); | |
3571 | XFS_QM_DQRELE(mp, gdqp); | |
3572 | ||
3573 | if (!dp_joined_to_trans && (dp != NULL)) { | |
3574 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
3575 | } | |
3576 | ||
3577 | goto std_return; | |
3578 | } | |
3579 | ||
3580 | ||
3581 | /* | |
3582 | * xfs_fid2 | |
3583 | * | |
3584 | * A fid routine that takes a pointer to a previously allocated | |
3585 | * fid structure (like xfs_fast_fid) but uses a 64 bit inode number. | |
3586 | */ | |
3587 | STATIC int | |
3588 | xfs_fid2( | |
3589 | bhv_desc_t *bdp, | |
3590 | fid_t *fidp) | |
3591 | { | |
3592 | xfs_inode_t *ip; | |
3593 | xfs_fid2_t *xfid; | |
3594 | ||
3595 | vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__, | |
3596 | (inst_t *)__return_address); | |
3597 | ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t)); | |
3598 | ||
3599 | xfid = (xfs_fid2_t *)fidp; | |
3600 | ip = XFS_BHVTOI(bdp); | |
3601 | xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len); | |
3602 | xfid->fid_pad = 0; | |
3603 | /* | |
3604 | * use memcpy because the inode is a long long and there's no | |
3605 | * assurance that xfid->fid_ino is properly aligned. | |
3606 | */ | |
3607 | memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino)); | |
3608 | xfid->fid_gen = ip->i_d.di_gen; | |
3609 | ||
3610 | return 0; | |
3611 | } | |
3612 | ||
3613 | ||
3614 | /* | |
3615 | * xfs_rwlock | |
3616 | */ | |
3617 | int | |
3618 | xfs_rwlock( | |
3619 | bhv_desc_t *bdp, | |
8285fb58 | 3620 | bhv_vrwlock_t locktype) |
1da177e4 LT |
3621 | { |
3622 | xfs_inode_t *ip; | |
67fcaa73 | 3623 | bhv_vnode_t *vp; |
1da177e4 LT |
3624 | |
3625 | vp = BHV_TO_VNODE(bdp); | |
0432dab2 | 3626 | if (VN_ISDIR(vp)) |
1da177e4 LT |
3627 | return 1; |
3628 | ip = XFS_BHVTOI(bdp); | |
3629 | if (locktype == VRWLOCK_WRITE) { | |
3630 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
3631 | } else if (locktype == VRWLOCK_TRY_READ) { | |
014c2544 | 3632 | return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED); |
1da177e4 | 3633 | } else if (locktype == VRWLOCK_TRY_WRITE) { |
014c2544 | 3634 | return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL); |
1da177e4 LT |
3635 | } else { |
3636 | ASSERT((locktype == VRWLOCK_READ) || | |
3637 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3638 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
3639 | } | |
3640 | ||
3641 | return 1; | |
3642 | } | |
3643 | ||
3644 | ||
3645 | /* | |
3646 | * xfs_rwunlock | |
3647 | */ | |
3648 | void | |
3649 | xfs_rwunlock( | |
3650 | bhv_desc_t *bdp, | |
8285fb58 | 3651 | bhv_vrwlock_t locktype) |
1da177e4 LT |
3652 | { |
3653 | xfs_inode_t *ip; | |
67fcaa73 | 3654 | bhv_vnode_t *vp; |
1da177e4 LT |
3655 | |
3656 | vp = BHV_TO_VNODE(bdp); | |
0432dab2 | 3657 | if (VN_ISDIR(vp)) |
1da177e4 LT |
3658 | return; |
3659 | ip = XFS_BHVTOI(bdp); | |
3660 | if (locktype == VRWLOCK_WRITE) { | |
3661 | /* | |
3662 | * In the write case, we may have added a new entry to | |
3663 | * the reference cache. This might store a pointer to | |
3664 | * an inode to be released in this inode. If it is there, | |
3665 | * clear the pointer and release the inode after unlocking | |
3666 | * this one. | |
3667 | */ | |
3668 | xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL); | |
3669 | } else { | |
3670 | ASSERT((locktype == VRWLOCK_READ) || | |
3671 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3672 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
3673 | } | |
3674 | return; | |
3675 | } | |
3676 | ||
3677 | STATIC int | |
3678 | xfs_inode_flush( | |
3679 | bhv_desc_t *bdp, | |
3680 | int flags) | |
3681 | { | |
3682 | xfs_inode_t *ip; | |
3683 | xfs_mount_t *mp; | |
3684 | xfs_inode_log_item_t *iip; | |
3685 | int error = 0; | |
3686 | ||
3687 | ip = XFS_BHVTOI(bdp); | |
3688 | mp = ip->i_mount; | |
3689 | iip = ip->i_itemp; | |
3690 | ||
3691 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3692 | return XFS_ERROR(EIO); | |
3693 | ||
3694 | /* | |
3695 | * Bypass inodes which have already been cleaned by | |
3696 | * the inode flush clustering code inside xfs_iflush | |
3697 | */ | |
3698 | if ((ip->i_update_core == 0) && | |
3699 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) | |
3700 | return 0; | |
3701 | ||
3702 | if (flags & FLUSH_LOG) { | |
3703 | if (iip && iip->ili_last_lsn) { | |
3704 | xlog_t *log = mp->m_log; | |
3705 | xfs_lsn_t sync_lsn; | |
3706 | int s, log_flags = XFS_LOG_FORCE; | |
3707 | ||
3708 | s = GRANT_LOCK(log); | |
3709 | sync_lsn = log->l_last_sync_lsn; | |
3710 | GRANT_UNLOCK(log, s); | |
3711 | ||
3712 | if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0)) | |
3713 | return 0; | |
3714 | ||
3715 | if (flags & FLUSH_SYNC) | |
3716 | log_flags |= XFS_LOG_SYNC; | |
3717 | return xfs_log_force(mp, iip->ili_last_lsn, log_flags); | |
3718 | } | |
3719 | } | |
3720 | ||
3721 | /* | |
3722 | * We make this non-blocking if the inode is contended, | |
3723 | * return EAGAIN to indicate to the caller that they | |
3724 | * did not succeed. This prevents the flush path from | |
3725 | * blocking on inodes inside another operation right | |
3726 | * now, they get caught later by xfs_sync. | |
3727 | */ | |
3728 | if (flags & FLUSH_INODE) { | |
3729 | int flush_flags; | |
3730 | ||
3731 | if (xfs_ipincount(ip)) | |
3732 | return EAGAIN; | |
3733 | ||
3734 | if (flags & FLUSH_SYNC) { | |
3735 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3736 | xfs_iflock(ip); | |
3737 | } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
3738 | if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) { | |
3739 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3740 | return EAGAIN; | |
3741 | } | |
3742 | } else { | |
3743 | return EAGAIN; | |
3744 | } | |
3745 | ||
3746 | if (flags & FLUSH_SYNC) | |
3747 | flush_flags = XFS_IFLUSH_SYNC; | |
3748 | else | |
3749 | flush_flags = XFS_IFLUSH_ASYNC; | |
3750 | ||
3751 | error = xfs_iflush(ip, flush_flags); | |
3752 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3753 | } | |
3754 | ||
3755 | return error; | |
3756 | } | |
3757 | ||
1da177e4 LT |
3758 | int |
3759 | xfs_set_dmattrs ( | |
3760 | bhv_desc_t *bdp, | |
3761 | u_int evmask, | |
3762 | u_int16_t state, | |
3763 | cred_t *credp) | |
3764 | { | |
3765 | xfs_inode_t *ip; | |
3766 | xfs_trans_t *tp; | |
3767 | xfs_mount_t *mp; | |
3768 | int error; | |
3769 | ||
3770 | if (!capable(CAP_SYS_ADMIN)) | |
3771 | return XFS_ERROR(EPERM); | |
3772 | ||
3773 | ip = XFS_BHVTOI(bdp); | |
3774 | mp = ip->i_mount; | |
3775 | ||
3776 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3777 | return XFS_ERROR(EIO); | |
3778 | ||
3779 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
3780 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
3781 | if (error) { | |
3782 | xfs_trans_cancel(tp, 0); | |
3783 | return error; | |
3784 | } | |
3785 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3786 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3787 | ||
3788 | ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask; | |
3789 | ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state; | |
3790 | ||
3791 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3792 | IHOLD(ip); | |
3793 | error = xfs_trans_commit(tp, 0, NULL); | |
3794 | ||
3795 | return error; | |
3796 | } | |
3797 | ||
1da177e4 LT |
3798 | STATIC int |
3799 | xfs_reclaim( | |
3800 | bhv_desc_t *bdp) | |
3801 | { | |
3802 | xfs_inode_t *ip; | |
67fcaa73 | 3803 | bhv_vnode_t *vp; |
1da177e4 LT |
3804 | |
3805 | vp = BHV_TO_VNODE(bdp); | |
3806 | ip = XFS_BHVTOI(bdp); | |
3807 | ||
3808 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
3809 | ||
3810 | ASSERT(!VN_MAPPED(vp)); | |
3811 | ||
3812 | /* bad inode, get out here ASAP */ | |
3813 | if (VN_BAD(vp)) { | |
3814 | xfs_ireclaim(ip); | |
3815 | return 0; | |
3816 | } | |
3817 | ||
51c91ed5 | 3818 | vn_iowait(vp); |
1da177e4 | 3819 | |
51c91ed5 | 3820 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0); |
1da177e4 | 3821 | |
42fe2b1f CH |
3822 | /* |
3823 | * Make sure the atime in the XFS inode is correct before freeing the | |
3824 | * Linux inode. | |
3825 | */ | |
3826 | xfs_synchronize_atime(ip); | |
1da177e4 | 3827 | |
4c60658e DC |
3828 | /* |
3829 | * If we have nothing to flush with this inode then complete the | |
3830 | * teardown now, otherwise break the link between the xfs inode and the | |
3831 | * linux inode and clean up the xfs inode later. This avoids flushing | |
3832 | * the inode to disk during the delete operation itself. | |
3833 | * | |
3834 | * When breaking the link, we need to set the XFS_IRECLAIMABLE flag | |
3835 | * first to ensure that xfs_iunpin() will never see an xfs inode | |
3836 | * that has a linux inode being reclaimed. Synchronisation is provided | |
3837 | * by the i_flags_lock. | |
1da177e4 LT |
3838 | */ |
3839 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
3840 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3841 | xfs_iflock(ip); | |
3842 | return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); | |
3843 | } else { | |
3844 | xfs_mount_t *mp = ip->i_mount; | |
3845 | ||
4c60658e | 3846 | /* Protect sync and unpin from us */ |
1da177e4 | 3847 | XFS_MOUNT_ILOCK(mp); |
4c60658e DC |
3848 | spin_lock(&ip->i_flags_lock); |
3849 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); | |
1da177e4 | 3850 | vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip)); |
4c60658e | 3851 | spin_unlock(&ip->i_flags_lock); |
1da177e4 | 3852 | list_add_tail(&ip->i_reclaim, &mp->m_del_inodes); |
1da177e4 LT |
3853 | XFS_MOUNT_IUNLOCK(mp); |
3854 | } | |
3855 | return 0; | |
3856 | } | |
3857 | ||
3858 | int | |
3859 | xfs_finish_reclaim( | |
3860 | xfs_inode_t *ip, | |
3861 | int locked, | |
3862 | int sync_mode) | |
3863 | { | |
3864 | xfs_ihash_t *ih = ip->i_hash; | |
67fcaa73 | 3865 | bhv_vnode_t *vp = XFS_ITOV_NULL(ip); |
1da177e4 LT |
3866 | int error; |
3867 | ||
3868 | if (vp && VN_BAD(vp)) | |
3869 | goto reclaim; | |
3870 | ||
3871 | /* The hash lock here protects a thread in xfs_iget_core from | |
3872 | * racing with us on linking the inode back with a vnode. | |
3873 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
3874 | * us. | |
3875 | */ | |
3876 | write_lock(&ih->ih_lock); | |
f273ab84 | 3877 | spin_lock(&ip->i_flags_lock); |
7a18c386 DC |
3878 | if (__xfs_iflags_test(ip, XFS_IRECLAIM) || |
3879 | (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) { | |
f273ab84 | 3880 | spin_unlock(&ip->i_flags_lock); |
1da177e4 LT |
3881 | write_unlock(&ih->ih_lock); |
3882 | if (locked) { | |
3883 | xfs_ifunlock(ip); | |
3884 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3885 | } | |
014c2544 | 3886 | return 1; |
1da177e4 | 3887 | } |
7a18c386 | 3888 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
f273ab84 | 3889 | spin_unlock(&ip->i_flags_lock); |
1da177e4 LT |
3890 | write_unlock(&ih->ih_lock); |
3891 | ||
3892 | /* | |
3893 | * If the inode is still dirty, then flush it out. If the inode | |
3894 | * is not in the AIL, then it will be OK to flush it delwri as | |
3895 | * long as xfs_iflush() does not keep any references to the inode. | |
3896 | * We leave that decision up to xfs_iflush() since it has the | |
3897 | * knowledge of whether it's OK to simply do a delwri flush of | |
3898 | * the inode or whether we need to wait until the inode is | |
3899 | * pulled from the AIL. | |
3900 | * We get the flush lock regardless, though, just to make sure | |
3901 | * we don't free it while it is being flushed. | |
3902 | */ | |
3903 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
3904 | if (!locked) { | |
3905 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3906 | xfs_iflock(ip); | |
3907 | } | |
3908 | ||
3909 | if (ip->i_update_core || | |
3910 | ((ip->i_itemp != NULL) && | |
3911 | (ip->i_itemp->ili_format.ilf_fields != 0))) { | |
3912 | error = xfs_iflush(ip, sync_mode); | |
3913 | /* | |
3914 | * If we hit an error, typically because of filesystem | |
3915 | * shutdown, we don't need to let vn_reclaim to know | |
3916 | * because we're gonna reclaim the inode anyway. | |
3917 | */ | |
3918 | if (error) { | |
3919 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3920 | goto reclaim; | |
3921 | } | |
3922 | xfs_iflock(ip); /* synchronize with xfs_iflush_done */ | |
3923 | } | |
3924 | ||
3925 | ASSERT(ip->i_update_core == 0); | |
3926 | ASSERT(ip->i_itemp == NULL || | |
3927 | ip->i_itemp->ili_format.ilf_fields == 0); | |
3928 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3929 | } else if (locked) { | |
3930 | /* | |
3931 | * We are not interested in doing an iflush if we're | |
3932 | * in the process of shutting down the filesystem forcibly. | |
3933 | * So, just reclaim the inode. | |
3934 | */ | |
3935 | xfs_ifunlock(ip); | |
3936 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3937 | } | |
3938 | ||
3939 | reclaim: | |
3940 | xfs_ireclaim(ip); | |
3941 | return 0; | |
3942 | } | |
3943 | ||
3944 | int | |
3945 | xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock) | |
3946 | { | |
3947 | int purged; | |
3948 | xfs_inode_t *ip, *n; | |
3949 | int done = 0; | |
3950 | ||
3951 | while (!done) { | |
3952 | purged = 0; | |
3953 | XFS_MOUNT_ILOCK(mp); | |
3954 | list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) { | |
3955 | if (noblock) { | |
3956 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) | |
3957 | continue; | |
3958 | if (xfs_ipincount(ip) || | |
3959 | !xfs_iflock_nowait(ip)) { | |
3960 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3961 | continue; | |
3962 | } | |
3963 | } | |
3964 | XFS_MOUNT_IUNLOCK(mp); | |
6b2cf618 FB |
3965 | if (xfs_finish_reclaim(ip, noblock, |
3966 | XFS_IFLUSH_DELWRI_ELSE_ASYNC)) | |
3967 | delay(1); | |
1da177e4 LT |
3968 | purged = 1; |
3969 | break; | |
3970 | } | |
3971 | ||
3972 | done = !purged; | |
3973 | } | |
3974 | ||
3975 | XFS_MOUNT_IUNLOCK(mp); | |
3976 | return 0; | |
3977 | } | |
3978 | ||
3979 | /* | |
3980 | * xfs_alloc_file_space() | |
3981 | * This routine allocates disk space for the given file. | |
3982 | * | |
3983 | * If alloc_type == 0, this request is for an ALLOCSP type | |
3984 | * request which will change the file size. In this case, no | |
3985 | * DMAPI event will be generated by the call. A TRUNCATE event | |
3986 | * will be generated later by xfs_setattr. | |
3987 | * | |
3988 | * If alloc_type != 0, this request is for a RESVSP type | |
3989 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
3990 | * lower block boundary byte address is less than the file's | |
3991 | * length. | |
3992 | * | |
3993 | * RETURNS: | |
3994 | * 0 on success | |
3995 | * errno on error | |
3996 | * | |
3997 | */ | |
ba0f32d4 | 3998 | STATIC int |
1da177e4 LT |
3999 | xfs_alloc_file_space( |
4000 | xfs_inode_t *ip, | |
4001 | xfs_off_t offset, | |
4002 | xfs_off_t len, | |
4003 | int alloc_type, | |
4004 | int attr_flags) | |
4005 | { | |
dd9f438e NS |
4006 | xfs_mount_t *mp = ip->i_mount; |
4007 | xfs_off_t count; | |
1da177e4 LT |
4008 | xfs_filblks_t allocated_fsb; |
4009 | xfs_filblks_t allocatesize_fsb; | |
dd9f438e NS |
4010 | xfs_extlen_t extsz, temp; |
4011 | xfs_fileoff_t startoffset_fsb; | |
1da177e4 | 4012 | xfs_fsblock_t firstfsb; |
dd9f438e NS |
4013 | int nimaps; |
4014 | int bmapi_flag; | |
4015 | int quota_flag; | |
1da177e4 | 4016 | int rt; |
1da177e4 | 4017 | xfs_trans_t *tp; |
dd9f438e NS |
4018 | xfs_bmbt_irec_t imaps[1], *imapp; |
4019 | xfs_bmap_free_t free_list; | |
4020 | uint qblocks, resblks, resrtextents; | |
4021 | int committed; | |
4022 | int error; | |
1da177e4 LT |
4023 | |
4024 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
1da177e4 LT |
4025 | |
4026 | if (XFS_FORCED_SHUTDOWN(mp)) | |
4027 | return XFS_ERROR(EIO); | |
4028 | ||
dd9f438e NS |
4029 | rt = XFS_IS_REALTIME_INODE(ip); |
4030 | if (unlikely(rt)) { | |
4031 | if (!(extsz = ip->i_d.di_extsize)) | |
4032 | extsz = mp->m_sb.sb_rextsize; | |
4033 | } else { | |
4034 | extsz = ip->i_d.di_extsize; | |
4035 | } | |
1da177e4 LT |
4036 | |
4037 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
4038 | return error; | |
4039 | ||
4040 | if (len <= 0) | |
4041 | return XFS_ERROR(EINVAL); | |
4042 | ||
4043 | count = len; | |
4044 | error = 0; | |
4045 | imapp = &imaps[0]; | |
dd9f438e NS |
4046 | nimaps = 1; |
4047 | bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
1da177e4 LT |
4048 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); |
4049 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
4050 | ||
4051 | /* Generate a DMAPI event if needed. */ | |
4052 | if (alloc_type != 0 && offset < ip->i_d.di_size && | |
4053 | (attr_flags&ATTR_DMI) == 0 && | |
4054 | DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) { | |
4055 | xfs_off_t end_dmi_offset; | |
4056 | ||
4057 | end_dmi_offset = offset+len; | |
4058 | if (end_dmi_offset > ip->i_d.di_size) | |
4059 | end_dmi_offset = ip->i_d.di_size; | |
4060 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip), | |
4061 | offset, end_dmi_offset - offset, | |
4062 | 0, NULL); | |
4063 | if (error) | |
014c2544 | 4064 | return error; |
1da177e4 LT |
4065 | } |
4066 | ||
4067 | /* | |
dd9f438e | 4068 | * Allocate file space until done or until there is an error |
1da177e4 LT |
4069 | */ |
4070 | retry: | |
4071 | while (allocatesize_fsb && !error) { | |
dd9f438e NS |
4072 | xfs_fileoff_t s, e; |
4073 | ||
1da177e4 | 4074 | /* |
3ddb8fa9 | 4075 | * Determine space reservations for data/realtime. |
1da177e4 | 4076 | */ |
dd9f438e | 4077 | if (unlikely(extsz)) { |
1da177e4 | 4078 | s = startoffset_fsb; |
dd9f438e NS |
4079 | do_div(s, extsz); |
4080 | s *= extsz; | |
4081 | e = startoffset_fsb + allocatesize_fsb; | |
4082 | if ((temp = do_mod(startoffset_fsb, extsz))) | |
4083 | e += temp; | |
4084 | if ((temp = do_mod(e, extsz))) | |
4085 | e += extsz - temp; | |
4086 | } else { | |
4087 | s = 0; | |
4088 | e = allocatesize_fsb; | |
4089 | } | |
4090 | ||
4091 | if (unlikely(rt)) { | |
4092 | resrtextents = qblocks = (uint)(e - s); | |
4093 | resrtextents /= mp->m_sb.sb_rextsize; | |
4094 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
4095 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
1da177e4 | 4096 | } else { |
dd9f438e NS |
4097 | resrtextents = 0; |
4098 | resblks = qblocks = \ | |
4099 | XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s)); | |
4100 | quota_flag = XFS_QMOPT_RES_REGBLKS; | |
1da177e4 LT |
4101 | } |
4102 | ||
4103 | /* | |
dd9f438e | 4104 | * Allocate and setup the transaction. |
1da177e4 LT |
4105 | */ |
4106 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
dd9f438e NS |
4107 | error = xfs_trans_reserve(tp, resblks, |
4108 | XFS_WRITE_LOG_RES(mp), resrtextents, | |
1da177e4 LT |
4109 | XFS_TRANS_PERM_LOG_RES, |
4110 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 4111 | /* |
dd9f438e | 4112 | * Check for running out of space |
1da177e4 LT |
4113 | */ |
4114 | if (error) { | |
4115 | /* | |
4116 | * Free the transaction structure. | |
4117 | */ | |
4118 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
4119 | xfs_trans_cancel(tp, 0); | |
4120 | break; | |
4121 | } | |
4122 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
dd9f438e NS |
4123 | error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, |
4124 | qblocks, 0, quota_flag); | |
1da177e4 LT |
4125 | if (error) |
4126 | goto error1; | |
4127 | ||
4128 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4129 | xfs_trans_ihold(tp, ip); | |
4130 | ||
4131 | /* | |
dd9f438e | 4132 | * Issue the xfs_bmapi() call to allocate the blocks |
1da177e4 LT |
4133 | */ |
4134 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
3e57ecf6 | 4135 | error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb, |
dd9f438e NS |
4136 | allocatesize_fsb, bmapi_flag, |
4137 | &firstfsb, 0, imapp, &nimaps, | |
3e57ecf6 | 4138 | &free_list, NULL); |
1da177e4 LT |
4139 | if (error) { |
4140 | goto error0; | |
4141 | } | |
4142 | ||
4143 | /* | |
dd9f438e | 4144 | * Complete the transaction |
1da177e4 | 4145 | */ |
f7c99b6f | 4146 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
4147 | if (error) { |
4148 | goto error0; | |
4149 | } | |
4150 | ||
4151 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
4152 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4153 | if (error) { | |
4154 | break; | |
4155 | } | |
4156 | ||
4157 | allocated_fsb = imapp->br_blockcount; | |
4158 | ||
dd9f438e | 4159 | if (nimaps == 0) { |
1da177e4 LT |
4160 | error = XFS_ERROR(ENOSPC); |
4161 | break; | |
4162 | } | |
4163 | ||
4164 | startoffset_fsb += allocated_fsb; | |
4165 | allocatesize_fsb -= allocated_fsb; | |
4166 | } | |
4167 | dmapi_enospc_check: | |
4168 | if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 && | |
4169 | DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) { | |
4170 | ||
4171 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, | |
4172 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
4173 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
4174 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ | |
4175 | if (error == 0) | |
4176 | goto retry; /* Maybe DMAPI app. has made space */ | |
4177 | /* else fall through with error from XFS_SEND_DATA */ | |
4178 | } | |
4179 | ||
4180 | return error; | |
4181 | ||
dd9f438e | 4182 | error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ |
1da177e4 | 4183 | xfs_bmap_cancel(&free_list); |
dd9f438e NS |
4184 | XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag); |
4185 | ||
4186 | error1: /* Just cancel transaction */ | |
1da177e4 LT |
4187 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
4188 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4189 | goto dmapi_enospc_check; | |
4190 | } | |
4191 | ||
4192 | /* | |
4193 | * Zero file bytes between startoff and endoff inclusive. | |
4194 | * The iolock is held exclusive and no blocks are buffered. | |
4195 | */ | |
4196 | STATIC int | |
4197 | xfs_zero_remaining_bytes( | |
4198 | xfs_inode_t *ip, | |
4199 | xfs_off_t startoff, | |
4200 | xfs_off_t endoff) | |
4201 | { | |
4202 | xfs_bmbt_irec_t imap; | |
4203 | xfs_fileoff_t offset_fsb; | |
4204 | xfs_off_t lastoffset; | |
4205 | xfs_off_t offset; | |
4206 | xfs_buf_t *bp; | |
4207 | xfs_mount_t *mp = ip->i_mount; | |
4208 | int nimap; | |
4209 | int error = 0; | |
4210 | ||
4211 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, | |
4212 | ip->i_d.di_flags & XFS_DIFLAG_REALTIME ? | |
4213 | mp->m_rtdev_targp : mp->m_ddev_targp); | |
4214 | ||
4215 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
4216 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
4217 | nimap = 1; | |
3e57ecf6 OW |
4218 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0, |
4219 | NULL, 0, &imap, &nimap, NULL, NULL); | |
1da177e4 LT |
4220 | if (error || nimap < 1) |
4221 | break; | |
4222 | ASSERT(imap.br_blockcount >= 1); | |
4223 | ASSERT(imap.br_startoff == offset_fsb); | |
4224 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
4225 | if (lastoffset > endoff) | |
4226 | lastoffset = endoff; | |
4227 | if (imap.br_startblock == HOLESTARTBLOCK) | |
4228 | continue; | |
4229 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4230 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
4231 | continue; | |
4232 | XFS_BUF_UNDONE(bp); | |
4233 | XFS_BUF_UNWRITE(bp); | |
4234 | XFS_BUF_READ(bp); | |
4235 | XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock)); | |
4236 | xfsbdstrat(mp, bp); | |
4237 | if ((error = xfs_iowait(bp))) { | |
4238 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", | |
4239 | mp, bp, XFS_BUF_ADDR(bp)); | |
4240 | break; | |
4241 | } | |
4242 | memset(XFS_BUF_PTR(bp) + | |
4243 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
4244 | 0, lastoffset - offset + 1); | |
4245 | XFS_BUF_UNDONE(bp); | |
4246 | XFS_BUF_UNREAD(bp); | |
4247 | XFS_BUF_WRITE(bp); | |
4248 | xfsbdstrat(mp, bp); | |
4249 | if ((error = xfs_iowait(bp))) { | |
4250 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", | |
4251 | mp, bp, XFS_BUF_ADDR(bp)); | |
4252 | break; | |
4253 | } | |
4254 | } | |
4255 | xfs_buf_free(bp); | |
4256 | return error; | |
4257 | } | |
4258 | ||
4259 | /* | |
4260 | * xfs_free_file_space() | |
4261 | * This routine frees disk space for the given file. | |
4262 | * | |
4263 | * This routine is only called by xfs_change_file_space | |
4264 | * for an UNRESVSP type call. | |
4265 | * | |
4266 | * RETURNS: | |
4267 | * 0 on success | |
4268 | * errno on error | |
4269 | * | |
4270 | */ | |
4271 | STATIC int | |
4272 | xfs_free_file_space( | |
4273 | xfs_inode_t *ip, | |
4274 | xfs_off_t offset, | |
4275 | xfs_off_t len, | |
4276 | int attr_flags) | |
4277 | { | |
67fcaa73 | 4278 | bhv_vnode_t *vp; |
1da177e4 LT |
4279 | int committed; |
4280 | int done; | |
4281 | xfs_off_t end_dmi_offset; | |
4282 | xfs_fileoff_t endoffset_fsb; | |
4283 | int error; | |
4284 | xfs_fsblock_t firstfsb; | |
4285 | xfs_bmap_free_t free_list; | |
4286 | xfs_off_t ilen; | |
4287 | xfs_bmbt_irec_t imap; | |
4288 | xfs_off_t ioffset; | |
4289 | xfs_extlen_t mod=0; | |
4290 | xfs_mount_t *mp; | |
4291 | int nimap; | |
4292 | uint resblks; | |
673cdf5c | 4293 | uint rounding; |
1da177e4 LT |
4294 | int rt; |
4295 | xfs_fileoff_t startoffset_fsb; | |
4296 | xfs_trans_t *tp; | |
5fcbab35 | 4297 | int need_iolock = 1; |
1da177e4 | 4298 | |
bd5a876a | 4299 | vp = XFS_ITOV(ip); |
1da177e4 LT |
4300 | mp = ip->i_mount; |
4301 | ||
bd5a876a CH |
4302 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
4303 | ||
1da177e4 LT |
4304 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) |
4305 | return error; | |
4306 | ||
4307 | error = 0; | |
4308 | if (len <= 0) /* if nothing being freed */ | |
4309 | return error; | |
4310 | rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME); | |
4311 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); | |
4312 | end_dmi_offset = offset + len; | |
4313 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
4314 | ||
4315 | if (offset < ip->i_d.di_size && | |
4316 | (attr_flags & ATTR_DMI) == 0 && | |
4317 | DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) { | |
4318 | if (end_dmi_offset > ip->i_d.di_size) | |
4319 | end_dmi_offset = ip->i_d.di_size; | |
bd5a876a | 4320 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp, |
1da177e4 LT |
4321 | offset, end_dmi_offset - offset, |
4322 | AT_DELAY_FLAG(attr_flags), NULL); | |
4323 | if (error) | |
014c2544 | 4324 | return error; |
1da177e4 LT |
4325 | } |
4326 | ||
5fcbab35 DR |
4327 | if (attr_flags & ATTR_NOLOCK) |
4328 | need_iolock = 0; | |
9fa8046f | 4329 | if (need_iolock) { |
1da177e4 | 4330 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
9fa8046f YL |
4331 | vn_iowait(vp); /* wait for the completion of any pending DIOs */ |
4332 | } | |
5fcbab35 | 4333 | |
673cdf5c | 4334 | rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, NBPP); |
1da177e4 LT |
4335 | ilen = len + (offset & (rounding - 1)); |
4336 | ioffset = offset & ~(rounding - 1); | |
4337 | if (ilen & (rounding - 1)) | |
4338 | ilen = (ilen + rounding) & ~(rounding - 1); | |
bd5a876a CH |
4339 | |
4340 | if (VN_CACHED(vp) != 0) { | |
4341 | xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1, | |
4342 | ctooff(offtoct(ioffset)), -1); | |
67fcaa73 | 4343 | bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)), |
bd5a876a CH |
4344 | -1, FI_REMAPF_LOCKED); |
4345 | } | |
4346 | ||
1da177e4 LT |
4347 | /* |
4348 | * Need to zero the stuff we're not freeing, on disk. | |
4349 | * If its a realtime file & can't use unwritten extents then we | |
4350 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
4351 | * will take care of it for us. | |
4352 | */ | |
4353 | if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) { | |
4354 | nimap = 1; | |
3e57ecf6 OW |
4355 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb, |
4356 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); | |
1da177e4 LT |
4357 | if (error) |
4358 | goto out_unlock_iolock; | |
4359 | ASSERT(nimap == 0 || nimap == 1); | |
4360 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4361 | xfs_daddr_t block; | |
4362 | ||
4363 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4364 | block = imap.br_startblock; | |
4365 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
4366 | if (mod) | |
4367 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
4368 | } | |
4369 | nimap = 1; | |
3e57ecf6 OW |
4370 | error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1, |
4371 | 1, 0, NULL, 0, &imap, &nimap, NULL, NULL); | |
1da177e4 LT |
4372 | if (error) |
4373 | goto out_unlock_iolock; | |
4374 | ASSERT(nimap == 0 || nimap == 1); | |
4375 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4376 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4377 | mod++; | |
4378 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
4379 | endoffset_fsb -= mod; | |
4380 | } | |
4381 | } | |
4382 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
4383 | /* | |
4384 | * One contiguous piece to clear | |
4385 | */ | |
4386 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
4387 | else { | |
4388 | /* | |
4389 | * Some full blocks, possibly two pieces to clear | |
4390 | */ | |
4391 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
4392 | error = xfs_zero_remaining_bytes(ip, offset, | |
4393 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
4394 | if (!error && | |
4395 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
4396 | error = xfs_zero_remaining_bytes(ip, | |
4397 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
4398 | offset + len - 1); | |
4399 | } | |
4400 | ||
4401 | /* | |
4402 | * free file space until done or until there is an error | |
4403 | */ | |
4404 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
4405 | while (!error && !done) { | |
4406 | ||
4407 | /* | |
4408 | * allocate and setup the transaction | |
4409 | */ | |
4410 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
4411 | error = xfs_trans_reserve(tp, | |
4412 | resblks, | |
4413 | XFS_WRITE_LOG_RES(mp), | |
4414 | 0, | |
4415 | XFS_TRANS_PERM_LOG_RES, | |
4416 | XFS_WRITE_LOG_COUNT); | |
4417 | ||
4418 | /* | |
4419 | * check for running out of space | |
4420 | */ | |
4421 | if (error) { | |
4422 | /* | |
4423 | * Free the transaction structure. | |
4424 | */ | |
4425 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
4426 | xfs_trans_cancel(tp, 0); | |
4427 | break; | |
4428 | } | |
4429 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4430 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, | |
dd9f438e NS |
4431 | ip->i_udquot, ip->i_gdquot, resblks, 0, |
4432 | XFS_QMOPT_RES_REGBLKS); | |
1da177e4 LT |
4433 | if (error) |
4434 | goto error1; | |
4435 | ||
4436 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4437 | xfs_trans_ihold(tp, ip); | |
4438 | ||
4439 | /* | |
4440 | * issue the bunmapi() call to free the blocks | |
4441 | */ | |
4442 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
3e57ecf6 | 4443 | error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb, |
1da177e4 | 4444 | endoffset_fsb - startoffset_fsb, |
3e57ecf6 | 4445 | 0, 2, &firstfsb, &free_list, NULL, &done); |
1da177e4 LT |
4446 | if (error) { |
4447 | goto error0; | |
4448 | } | |
4449 | ||
4450 | /* | |
4451 | * complete the transaction | |
4452 | */ | |
f7c99b6f | 4453 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
4454 | if (error) { |
4455 | goto error0; | |
4456 | } | |
4457 | ||
4458 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
4459 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4460 | } | |
4461 | ||
4462 | out_unlock_iolock: | |
4463 | if (need_iolock) | |
4464 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
4465 | return error; | |
4466 | ||
4467 | error0: | |
4468 | xfs_bmap_cancel(&free_list); | |
4469 | error1: | |
4470 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
4471 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
4472 | XFS_ILOCK_EXCL); | |
4473 | return error; | |
4474 | } | |
4475 | ||
4476 | /* | |
4477 | * xfs_change_file_space() | |
4478 | * This routine allocates or frees disk space for the given file. | |
4479 | * The user specified parameters are checked for alignment and size | |
4480 | * limitations. | |
4481 | * | |
4482 | * RETURNS: | |
4483 | * 0 on success | |
4484 | * errno on error | |
4485 | * | |
4486 | */ | |
4487 | int | |
4488 | xfs_change_file_space( | |
4489 | bhv_desc_t *bdp, | |
4490 | int cmd, | |
4491 | xfs_flock64_t *bf, | |
4492 | xfs_off_t offset, | |
4493 | cred_t *credp, | |
4494 | int attr_flags) | |
4495 | { | |
4496 | int clrprealloc; | |
4497 | int error; | |
4498 | xfs_fsize_t fsize; | |
4499 | xfs_inode_t *ip; | |
4500 | xfs_mount_t *mp; | |
4501 | int setprealloc; | |
4502 | xfs_off_t startoffset; | |
4503 | xfs_off_t llen; | |
4504 | xfs_trans_t *tp; | |
8285fb58 | 4505 | bhv_vattr_t va; |
67fcaa73 | 4506 | bhv_vnode_t *vp; |
1da177e4 LT |
4507 | |
4508 | vp = BHV_TO_VNODE(bdp); | |
4509 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
4510 | ||
4511 | ip = XFS_BHVTOI(bdp); | |
4512 | mp = ip->i_mount; | |
4513 | ||
4514 | /* | |
4515 | * must be a regular file and have write permission | |
4516 | */ | |
0432dab2 | 4517 | if (!VN_ISREG(vp)) |
1da177e4 LT |
4518 | return XFS_ERROR(EINVAL); |
4519 | ||
4520 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
4521 | ||
4522 | if ((error = xfs_iaccess(ip, S_IWUSR, credp))) { | |
4523 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
4524 | return error; | |
4525 | } | |
4526 | ||
4527 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
4528 | ||
4529 | switch (bf->l_whence) { | |
4530 | case 0: /*SEEK_SET*/ | |
4531 | break; | |
4532 | case 1: /*SEEK_CUR*/ | |
4533 | bf->l_start += offset; | |
4534 | break; | |
4535 | case 2: /*SEEK_END*/ | |
4536 | bf->l_start += ip->i_d.di_size; | |
4537 | break; | |
4538 | default: | |
4539 | return XFS_ERROR(EINVAL); | |
4540 | } | |
4541 | ||
4542 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
4543 | ||
4544 | if ( (bf->l_start < 0) | |
4545 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
4546 | || (bf->l_start + llen < 0) | |
4547 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
4548 | return XFS_ERROR(EINVAL); | |
4549 | ||
4550 | bf->l_whence = 0; | |
4551 | ||
4552 | startoffset = bf->l_start; | |
4553 | fsize = ip->i_d.di_size; | |
4554 | ||
4555 | /* | |
4556 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
4557 | * file space. | |
4558 | * These calls do NOT zero the data space allocated to the file, | |
4559 | * nor do they change the file size. | |
4560 | * | |
4561 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
4562 | * space. | |
4563 | * These calls cause the new file data to be zeroed and the file | |
4564 | * size to be changed. | |
4565 | */ | |
4566 | setprealloc = clrprealloc = 0; | |
4567 | ||
4568 | switch (cmd) { | |
4569 | case XFS_IOC_RESVSP: | |
4570 | case XFS_IOC_RESVSP64: | |
4571 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
4572 | 1, attr_flags); | |
4573 | if (error) | |
4574 | return error; | |
4575 | setprealloc = 1; | |
4576 | break; | |
4577 | ||
4578 | case XFS_IOC_UNRESVSP: | |
4579 | case XFS_IOC_UNRESVSP64: | |
4580 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
4581 | attr_flags))) | |
4582 | return error; | |
4583 | break; | |
4584 | ||
4585 | case XFS_IOC_ALLOCSP: | |
4586 | case XFS_IOC_ALLOCSP64: | |
4587 | case XFS_IOC_FREESP: | |
4588 | case XFS_IOC_FREESP64: | |
4589 | if (startoffset > fsize) { | |
4590 | error = xfs_alloc_file_space(ip, fsize, | |
4591 | startoffset - fsize, 0, attr_flags); | |
4592 | if (error) | |
4593 | break; | |
4594 | } | |
4595 | ||
4596 | va.va_mask = XFS_AT_SIZE; | |
4597 | va.va_size = startoffset; | |
4598 | ||
4599 | error = xfs_setattr(bdp, &va, attr_flags, credp); | |
4600 | ||
4601 | if (error) | |
4602 | return error; | |
4603 | ||
4604 | clrprealloc = 1; | |
4605 | break; | |
4606 | ||
4607 | default: | |
4608 | ASSERT(0); | |
4609 | return XFS_ERROR(EINVAL); | |
4610 | } | |
4611 | ||
4612 | /* | |
4613 | * update the inode timestamp, mode, and prealloc flag bits | |
4614 | */ | |
4615 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
4616 | ||
4617 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
4618 | 0, 0, 0))) { | |
4619 | /* ASSERT(0); */ | |
4620 | xfs_trans_cancel(tp, 0); | |
4621 | return error; | |
4622 | } | |
4623 | ||
4624 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4625 | ||
4626 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4627 | xfs_trans_ihold(tp, ip); | |
4628 | ||
4629 | if ((attr_flags & ATTR_DMI) == 0) { | |
4630 | ip->i_d.di_mode &= ~S_ISUID; | |
4631 | ||
4632 | /* | |
4633 | * Note that we don't have to worry about mandatory | |
4634 | * file locking being disabled here because we only | |
4635 | * clear the S_ISGID bit if the Group execute bit is | |
4636 | * on, but if it was on then mandatory locking wouldn't | |
4637 | * have been enabled. | |
4638 | */ | |
4639 | if (ip->i_d.di_mode & S_IXGRP) | |
4640 | ip->i_d.di_mode &= ~S_ISGID; | |
4641 | ||
4642 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
4643 | } | |
4644 | if (setprealloc) | |
4645 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
4646 | else if (clrprealloc) | |
4647 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
4648 | ||
4649 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
4650 | xfs_trans_set_sync(tp); | |
4651 | ||
4652 | error = xfs_trans_commit(tp, 0, NULL); | |
4653 | ||
4654 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4655 | ||
4656 | return error; | |
4657 | } | |
4658 | ||
67fcaa73 | 4659 | bhv_vnodeops_t xfs_vnodeops = { |
1da177e4 LT |
4660 | BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS), |
4661 | .vop_open = xfs_open, | |
7d4fb40a | 4662 | .vop_close = xfs_close, |
1da177e4 LT |
4663 | .vop_read = xfs_read, |
4664 | #ifdef HAVE_SENDFILE | |
4665 | .vop_sendfile = xfs_sendfile, | |
1b895840 NS |
4666 | #endif |
4667 | #ifdef HAVE_SPLICE | |
4668 | .vop_splice_read = xfs_splice_read, | |
4669 | .vop_splice_write = xfs_splice_write, | |
1da177e4 LT |
4670 | #endif |
4671 | .vop_write = xfs_write, | |
4672 | .vop_ioctl = xfs_ioctl, | |
4673 | .vop_getattr = xfs_getattr, | |
4674 | .vop_setattr = xfs_setattr, | |
4675 | .vop_access = xfs_access, | |
4676 | .vop_lookup = xfs_lookup, | |
4677 | .vop_create = xfs_create, | |
4678 | .vop_remove = xfs_remove, | |
4679 | .vop_link = xfs_link, | |
4680 | .vop_rename = xfs_rename, | |
4681 | .vop_mkdir = xfs_mkdir, | |
4682 | .vop_rmdir = xfs_rmdir, | |
4683 | .vop_readdir = xfs_readdir, | |
4684 | .vop_symlink = xfs_symlink, | |
4685 | .vop_readlink = xfs_readlink, | |
4686 | .vop_fsync = xfs_fsync, | |
4687 | .vop_inactive = xfs_inactive, | |
4688 | .vop_fid2 = xfs_fid2, | |
4689 | .vop_rwlock = xfs_rwlock, | |
4690 | .vop_rwunlock = xfs_rwunlock, | |
4691 | .vop_bmap = xfs_bmap, | |
4692 | .vop_reclaim = xfs_reclaim, | |
4693 | .vop_attr_get = xfs_attr_get, | |
4694 | .vop_attr_set = xfs_attr_set, | |
4695 | .vop_attr_remove = xfs_attr_remove, | |
4696 | .vop_attr_list = xfs_attr_list, | |
4697 | .vop_link_removed = (vop_link_removed_t)fs_noval, | |
4698 | .vop_vnode_change = (vop_vnode_change_t)fs_noval, | |
4699 | .vop_tosspages = fs_tosspages, | |
4700 | .vop_flushinval_pages = fs_flushinval_pages, | |
4701 | .vop_flush_pages = fs_flush_pages, | |
4702 | .vop_release = xfs_release, | |
4703 | .vop_iflush = xfs_inode_flush, | |
4704 | }; |