Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2000-2004 Silicon Graphics, Inc. All Rights Reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms of version 2 of the GNU General Public License as | |
6 | * published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it would be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. | |
11 | * | |
12 | * Further, this software is distributed without any warranty that it is | |
13 | * free of the rightful claim of any third person regarding infringement | |
14 | * or the like. Any license provided herein, whether implied or | |
15 | * otherwise, applies only to this software file. Patent licenses, if | |
16 | * any, provided herein do not apply to combinations of this program with | |
17 | * other software, or any other product whatsoever. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License along | |
20 | * with this program; if not, write the Free Software Foundation, Inc., 59 | |
21 | * Temple Place - Suite 330, Boston MA 02111-1307, USA. | |
22 | * | |
23 | * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy, | |
24 | * Mountain View, CA 94043, or: | |
25 | * | |
26 | * http://www.sgi.com | |
27 | * | |
28 | * For further information regarding this notice, see: | |
29 | * | |
30 | * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/ | |
31 | */ | |
32 | ||
33 | #include "xfs.h" | |
34 | #include "xfs_macros.h" | |
35 | #include "xfs_types.h" | |
36 | #include "xfs_inum.h" | |
37 | #include "xfs_log.h" | |
38 | #include "xfs_trans.h" | |
39 | #include "xfs_sb.h" | |
40 | #include "xfs_ag.h" | |
41 | #include "xfs_dir.h" | |
42 | #include "xfs_dir2.h" | |
43 | #include "xfs_dmapi.h" | |
44 | #include "xfs_mount.h" | |
45 | #include "xfs_alloc_btree.h" | |
46 | #include "xfs_bmap_btree.h" | |
47 | #include "xfs_ialloc_btree.h" | |
48 | #include "xfs_itable.h" | |
49 | #include "xfs_btree.h" | |
50 | #include "xfs_ialloc.h" | |
51 | #include "xfs_alloc.h" | |
52 | #include "xfs_attr_sf.h" | |
53 | #include "xfs_dir_sf.h" | |
54 | #include "xfs_dir2_sf.h" | |
55 | #include "xfs_dinode.h" | |
56 | #include "xfs_inode_item.h" | |
57 | #include "xfs_inode.h" | |
58 | #include "xfs_bmap.h" | |
59 | #include "xfs_da_btree.h" | |
60 | #include "xfs_attr.h" | |
61 | #include "xfs_rw.h" | |
62 | #include "xfs_refcache.h" | |
63 | #include "xfs_error.h" | |
64 | #include "xfs_bit.h" | |
65 | #include "xfs_rtalloc.h" | |
66 | #include "xfs_quota.h" | |
67 | #include "xfs_utils.h" | |
68 | #include "xfs_trans_space.h" | |
69 | #include "xfs_dir_leaf.h" | |
70 | #include "xfs_mac.h" | |
71 | #include "xfs_log_priv.h" | |
72 | ||
73 | ||
74 | /* | |
75 | * The maximum pathlen is 1024 bytes. Since the minimum file system | |
76 | * blocksize is 512 bytes, we can get a max of 2 extents back from | |
77 | * bmapi. | |
78 | */ | |
79 | #define SYMLINK_MAPS 2 | |
80 | ||
81 | /* | |
82 | * For xfs, we check that the file isn't too big to be opened by this kernel. | |
83 | * No other open action is required for regular files. Devices are handled | |
84 | * through the specfs file system, pipes through fifofs. Device and | |
85 | * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively, | |
86 | * when a new vnode is first looked up or created. | |
87 | */ | |
88 | STATIC int | |
89 | xfs_open( | |
90 | bhv_desc_t *bdp, | |
91 | cred_t *credp) | |
92 | { | |
93 | int mode; | |
94 | vnode_t *vp; | |
95 | xfs_inode_t *ip; | |
96 | ||
97 | vp = BHV_TO_VNODE(bdp); | |
98 | ip = XFS_BHVTOI(bdp); | |
99 | ||
100 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
101 | return XFS_ERROR(EIO); | |
102 | ||
103 | /* | |
104 | * If it's a directory with any blocks, read-ahead block 0 | |
105 | * as we're almost certain to have the next operation be a read there. | |
106 | */ | |
107 | if (vp->v_type == VDIR && ip->i_d.di_nextents > 0) { | |
108 | mode = xfs_ilock_map_shared(ip); | |
109 | if (ip->i_d.di_nextents > 0) | |
110 | (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK); | |
111 | xfs_iunlock(ip, mode); | |
112 | } | |
113 | return 0; | |
114 | } | |
115 | ||
116 | ||
117 | /* | |
118 | * xfs_getattr | |
119 | */ | |
120 | STATIC int | |
121 | xfs_getattr( | |
122 | bhv_desc_t *bdp, | |
123 | vattr_t *vap, | |
124 | int flags, | |
125 | cred_t *credp) | |
126 | { | |
127 | xfs_inode_t *ip; | |
128 | xfs_mount_t *mp; | |
129 | vnode_t *vp; | |
130 | ||
131 | vp = BHV_TO_VNODE(bdp); | |
132 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
133 | ||
134 | ip = XFS_BHVTOI(bdp); | |
135 | mp = ip->i_mount; | |
136 | ||
137 | if (XFS_FORCED_SHUTDOWN(mp)) | |
138 | return XFS_ERROR(EIO); | |
139 | ||
140 | if (!(flags & ATTR_LAZY)) | |
141 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
142 | ||
143 | vap->va_size = ip->i_d.di_size; | |
144 | if (vap->va_mask == XFS_AT_SIZE) | |
145 | goto all_done; | |
146 | ||
147 | vap->va_nblocks = | |
148 | XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); | |
149 | vap->va_nodeid = ip->i_ino; | |
150 | #if XFS_BIG_INUMS | |
151 | vap->va_nodeid += mp->m_inoadd; | |
152 | #endif | |
153 | vap->va_nlink = ip->i_d.di_nlink; | |
154 | ||
155 | /* | |
156 | * Quick exit for non-stat callers | |
157 | */ | |
158 | if ((vap->va_mask & | |
159 | ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID| | |
160 | XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0) | |
161 | goto all_done; | |
162 | ||
163 | /* | |
164 | * Copy from in-core inode. | |
165 | */ | |
166 | vap->va_type = vp->v_type; | |
167 | vap->va_mode = ip->i_d.di_mode & MODEMASK; | |
168 | vap->va_uid = ip->i_d.di_uid; | |
169 | vap->va_gid = ip->i_d.di_gid; | |
170 | vap->va_projid = ip->i_d.di_projid; | |
171 | ||
172 | /* | |
173 | * Check vnode type block/char vs. everything else. | |
174 | * Do it with bitmask because that's faster than looking | |
175 | * for multiple values individually. | |
176 | */ | |
177 | if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) { | |
178 | vap->va_rdev = 0; | |
179 | ||
180 | if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) { | |
181 | ||
182 | #if 0 | |
183 | /* Large block sizes confuse various | |
184 | * user space programs, so letting the | |
185 | * stripe size through is not a good | |
186 | * idea for now. | |
187 | */ | |
188 | vap->va_blocksize = mp->m_swidth ? | |
189 | /* | |
190 | * If the underlying volume is a stripe, then | |
191 | * return the stripe width in bytes as the | |
192 | * recommended I/O size. | |
193 | */ | |
194 | (mp->m_swidth << mp->m_sb.sb_blocklog) : | |
195 | /* | |
196 | * Return the largest of the preferred buffer | |
197 | * sizes since doing small I/Os into larger | |
198 | * buffers causes buffers to be decommissioned. | |
199 | * The value returned is in bytes. | |
200 | */ | |
201 | (1 << (int)MAX(mp->m_readio_log, | |
202 | mp->m_writeio_log)); | |
203 | ||
204 | #else | |
205 | vap->va_blocksize = | |
206 | /* | |
207 | * Return the largest of the preferred buffer | |
208 | * sizes since doing small I/Os into larger | |
209 | * buffers causes buffers to be decommissioned. | |
210 | * The value returned is in bytes. | |
211 | */ | |
212 | 1 << (int)MAX(mp->m_readio_log, | |
213 | mp->m_writeio_log); | |
214 | #endif | |
215 | } else { | |
216 | ||
217 | /* | |
218 | * If the file blocks are being allocated from a | |
219 | * realtime partition, then return the inode's | |
220 | * realtime extent size or the realtime volume's | |
221 | * extent size. | |
222 | */ | |
223 | vap->va_blocksize = ip->i_d.di_extsize ? | |
224 | (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) : | |
225 | (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog); | |
226 | } | |
227 | } else { | |
228 | vap->va_rdev = ip->i_df.if_u2.if_rdev; | |
229 | vap->va_blocksize = BLKDEV_IOSIZE; | |
230 | } | |
231 | ||
232 | vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec; | |
233 | vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec; | |
234 | vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec; | |
235 | vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec; | |
236 | vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec; | |
237 | vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec; | |
238 | ||
239 | /* | |
240 | * Exit for stat callers. See if any of the rest of the fields | |
241 | * to be filled in are needed. | |
242 | */ | |
243 | if ((vap->va_mask & | |
244 | (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS| | |
245 | XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0) | |
246 | goto all_done; | |
247 | ||
248 | /* | |
249 | * Convert di_flags to xflags. | |
250 | */ | |
251 | vap->va_xflags = xfs_ip2xflags(ip); | |
252 | ||
253 | /* | |
254 | * Exit for inode revalidate. See if any of the rest of | |
255 | * the fields to be filled in are needed. | |
256 | */ | |
257 | if ((vap->va_mask & | |
258 | (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS| | |
259 | XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0) | |
260 | goto all_done; | |
261 | ||
262 | vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog; | |
263 | vap->va_nextents = | |
264 | (ip->i_df.if_flags & XFS_IFEXTENTS) ? | |
265 | ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) : | |
266 | ip->i_d.di_nextents; | |
267 | if (ip->i_afp) | |
268 | vap->va_anextents = | |
269 | (ip->i_afp->if_flags & XFS_IFEXTENTS) ? | |
270 | ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) : | |
271 | ip->i_d.di_anextents; | |
272 | else | |
273 | vap->va_anextents = 0; | |
274 | vap->va_gen = ip->i_d.di_gen; | |
275 | ||
276 | all_done: | |
277 | if (!(flags & ATTR_LAZY)) | |
278 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
279 | return 0; | |
280 | } | |
281 | ||
282 | ||
283 | /* | |
284 | * xfs_setattr | |
285 | */ | |
286 | int | |
287 | xfs_setattr( | |
288 | bhv_desc_t *bdp, | |
289 | vattr_t *vap, | |
290 | int flags, | |
291 | cred_t *credp) | |
292 | { | |
293 | xfs_inode_t *ip; | |
294 | xfs_trans_t *tp; | |
295 | xfs_mount_t *mp; | |
296 | int mask; | |
297 | int code; | |
298 | uint lock_flags; | |
299 | uint commit_flags=0; | |
300 | uid_t uid=0, iuid=0; | |
301 | gid_t gid=0, igid=0; | |
302 | int timeflags = 0; | |
303 | vnode_t *vp; | |
304 | xfs_prid_t projid=0, iprojid=0; | |
305 | int mandlock_before, mandlock_after; | |
306 | struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2; | |
307 | int file_owner; | |
5fcbab35 | 308 | int need_iolock = 1; |
1da177e4 LT |
309 | |
310 | vp = BHV_TO_VNODE(bdp); | |
311 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
312 | ||
313 | if (vp->v_vfsp->vfs_flag & VFS_RDONLY) | |
314 | return XFS_ERROR(EROFS); | |
315 | ||
316 | /* | |
317 | * Cannot set certain attributes. | |
318 | */ | |
319 | mask = vap->va_mask; | |
320 | if (mask & XFS_AT_NOSET) { | |
321 | return XFS_ERROR(EINVAL); | |
322 | } | |
323 | ||
324 | ip = XFS_BHVTOI(bdp); | |
325 | mp = ip->i_mount; | |
326 | ||
327 | if (XFS_FORCED_SHUTDOWN(mp)) | |
328 | return XFS_ERROR(EIO); | |
329 | ||
330 | /* | |
331 | * Timestamps do not need to be logged and hence do not | |
332 | * need to be done within a transaction. | |
333 | */ | |
334 | if (mask & XFS_AT_UPDTIMES) { | |
335 | ASSERT((mask & ~XFS_AT_UPDTIMES) == 0); | |
336 | timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) | | |
337 | ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) | | |
338 | ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0); | |
339 | xfs_ichgtime(ip, timeflags); | |
340 | return 0; | |
341 | } | |
342 | ||
343 | olddquot1 = olddquot2 = NULL; | |
344 | udqp = gdqp = NULL; | |
345 | ||
346 | /* | |
347 | * If disk quotas is on, we make sure that the dquots do exist on disk, | |
348 | * before we start any other transactions. Trying to do this later | |
349 | * is messy. We don't care to take a readlock to look at the ids | |
350 | * in inode here, because we can't hold it across the trans_reserve. | |
351 | * If the IDs do change before we take the ilock, we're covered | |
352 | * because the i_*dquot fields will get updated anyway. | |
353 | */ | |
354 | if (XFS_IS_QUOTA_ON(mp) && (mask & (XFS_AT_UID|XFS_AT_GID))) { | |
355 | uint qflags = 0; | |
356 | ||
357 | if (mask & XFS_AT_UID) { | |
358 | uid = vap->va_uid; | |
359 | qflags |= XFS_QMOPT_UQUOTA; | |
360 | } else { | |
361 | uid = ip->i_d.di_uid; | |
362 | } | |
363 | if (mask & XFS_AT_GID) { | |
364 | gid = vap->va_gid; | |
365 | qflags |= XFS_QMOPT_GQUOTA; | |
366 | } else { | |
367 | gid = ip->i_d.di_gid; | |
368 | } | |
369 | /* | |
370 | * We take a reference when we initialize udqp and gdqp, | |
371 | * so it is important that we never blindly double trip on | |
372 | * the same variable. See xfs_create() for an example. | |
373 | */ | |
374 | ASSERT(udqp == NULL); | |
375 | ASSERT(gdqp == NULL); | |
376 | code = XFS_QM_DQVOPALLOC(mp, ip, uid,gid, qflags, &udqp, &gdqp); | |
377 | if (code) | |
378 | return (code); | |
379 | } | |
380 | ||
381 | /* | |
382 | * For the other attributes, we acquire the inode lock and | |
383 | * first do an error checking pass. | |
384 | */ | |
385 | tp = NULL; | |
386 | lock_flags = XFS_ILOCK_EXCL; | |
5fcbab35 DR |
387 | ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1); |
388 | if (flags & ATTR_NOLOCK) | |
389 | need_iolock = 0; | |
1da177e4 LT |
390 | if (!(mask & XFS_AT_SIZE)) { |
391 | if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) || | |
392 | (mp->m_flags & XFS_MOUNT_WSYNC)) { | |
393 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); | |
394 | commit_flags = 0; | |
395 | if ((code = xfs_trans_reserve(tp, 0, | |
396 | XFS_ICHANGE_LOG_RES(mp), 0, | |
397 | 0, 0))) { | |
398 | lock_flags = 0; | |
399 | goto error_return; | |
400 | } | |
401 | } | |
402 | } else { | |
403 | if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) && | |
404 | !(flags & ATTR_DMI)) { | |
405 | int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR; | |
406 | code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp, | |
407 | vap->va_size, 0, dmflags, NULL); | |
408 | if (code) { | |
409 | lock_flags = 0; | |
410 | goto error_return; | |
411 | } | |
412 | } | |
413 | if (need_iolock) | |
414 | lock_flags |= XFS_IOLOCK_EXCL; | |
415 | } | |
416 | ||
417 | xfs_ilock(ip, lock_flags); | |
418 | ||
419 | /* boolean: are we the file owner? */ | |
420 | file_owner = (current_fsuid(credp) == ip->i_d.di_uid); | |
421 | ||
422 | /* | |
423 | * Change various properties of a file. | |
424 | * Only the owner or users with CAP_FOWNER | |
425 | * capability may do these things. | |
426 | */ | |
427 | if (mask & | |
428 | (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID| | |
429 | XFS_AT_GID|XFS_AT_PROJID)) { | |
430 | /* | |
431 | * CAP_FOWNER overrides the following restrictions: | |
432 | * | |
433 | * The user ID of the calling process must be equal | |
434 | * to the file owner ID, except in cases where the | |
435 | * CAP_FSETID capability is applicable. | |
436 | */ | |
437 | if (!file_owner && !capable(CAP_FOWNER)) { | |
438 | code = XFS_ERROR(EPERM); | |
439 | goto error_return; | |
440 | } | |
441 | ||
442 | /* | |
443 | * CAP_FSETID overrides the following restrictions: | |
444 | * | |
445 | * The effective user ID of the calling process shall match | |
446 | * the file owner when setting the set-user-ID and | |
447 | * set-group-ID bits on that file. | |
448 | * | |
449 | * The effective group ID or one of the supplementary group | |
450 | * IDs of the calling process shall match the group owner of | |
451 | * the file when setting the set-group-ID bit on that file | |
452 | */ | |
453 | if (mask & XFS_AT_MODE) { | |
454 | mode_t m = 0; | |
455 | ||
456 | if ((vap->va_mode & S_ISUID) && !file_owner) | |
457 | m |= S_ISUID; | |
458 | if ((vap->va_mode & S_ISGID) && | |
459 | !in_group_p((gid_t)ip->i_d.di_gid)) | |
460 | m |= S_ISGID; | |
461 | #if 0 | |
462 | /* Linux allows this, Irix doesn't. */ | |
463 | if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR) | |
464 | m |= S_ISVTX; | |
465 | #endif | |
466 | if (m && !capable(CAP_FSETID)) | |
467 | vap->va_mode &= ~m; | |
468 | } | |
469 | } | |
470 | ||
471 | /* | |
472 | * Change file ownership. Must be the owner or privileged. | |
473 | * If the system was configured with the "restricted_chown" | |
474 | * option, the owner is not permitted to give away the file, | |
475 | * and can change the group id only to a group of which he | |
476 | * or she is a member. | |
477 | */ | |
478 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
479 | /* | |
480 | * These IDs could have changed since we last looked at them. | |
481 | * But, we're assured that if the ownership did change | |
482 | * while we didn't have the inode locked, inode's dquot(s) | |
483 | * would have changed also. | |
484 | */ | |
485 | iuid = ip->i_d.di_uid; | |
486 | iprojid = ip->i_d.di_projid; | |
487 | igid = ip->i_d.di_gid; | |
488 | gid = (mask & XFS_AT_GID) ? vap->va_gid : igid; | |
489 | uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid; | |
490 | projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid : | |
491 | iprojid; | |
492 | ||
493 | /* | |
494 | * CAP_CHOWN overrides the following restrictions: | |
495 | * | |
496 | * If _POSIX_CHOWN_RESTRICTED is defined, this capability | |
497 | * shall override the restriction that a process cannot | |
498 | * change the user ID of a file it owns and the restriction | |
499 | * that the group ID supplied to the chown() function | |
500 | * shall be equal to either the group ID or one of the | |
501 | * supplementary group IDs of the calling process. | |
502 | * | |
503 | * XXX: How does restricted_chown affect projid? | |
504 | */ | |
505 | if (restricted_chown && | |
506 | (iuid != uid || (igid != gid && | |
507 | !in_group_p((gid_t)gid))) && | |
508 | !capable(CAP_CHOWN)) { | |
509 | code = XFS_ERROR(EPERM); | |
510 | goto error_return; | |
511 | } | |
512 | /* | |
513 | * Do a quota reservation only if uid or gid is actually | |
514 | * going to change. | |
515 | */ | |
516 | if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || | |
517 | (XFS_IS_GQUOTA_ON(mp) && igid != gid)) { | |
518 | ASSERT(tp); | |
519 | code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp, | |
520 | capable(CAP_FOWNER) ? | |
521 | XFS_QMOPT_FORCE_RES : 0); | |
522 | if (code) /* out of quota */ | |
523 | goto error_return; | |
524 | } | |
525 | } | |
526 | ||
527 | /* | |
528 | * Truncate file. Must have write permission and not be a directory. | |
529 | */ | |
530 | if (mask & XFS_AT_SIZE) { | |
531 | /* Short circuit the truncate case for zero length files */ | |
532 | if ((vap->va_size == 0) && | |
533 | (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) { | |
534 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
535 | lock_flags &= ~XFS_ILOCK_EXCL; | |
536 | if (mask & XFS_AT_CTIME) | |
537 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
538 | code = 0; | |
539 | goto error_return; | |
540 | } | |
541 | ||
542 | if (vp->v_type == VDIR) { | |
543 | code = XFS_ERROR(EISDIR); | |
544 | goto error_return; | |
545 | } else if (vp->v_type != VREG) { | |
546 | code = XFS_ERROR(EINVAL); | |
547 | goto error_return; | |
548 | } | |
549 | /* | |
550 | * Make sure that the dquots are attached to the inode. | |
551 | */ | |
552 | if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED))) | |
553 | goto error_return; | |
554 | } | |
555 | ||
556 | /* | |
557 | * Change file access or modified times. | |
558 | */ | |
559 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
560 | if (!file_owner) { | |
561 | if ((flags & ATTR_UTIME) && | |
562 | !capable(CAP_FOWNER)) { | |
563 | code = XFS_ERROR(EPERM); | |
564 | goto error_return; | |
565 | } | |
566 | } | |
567 | } | |
568 | ||
569 | /* | |
570 | * Change extent size or realtime flag. | |
571 | */ | |
572 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
573 | /* | |
574 | * Can't change extent size if any extents are allocated. | |
575 | */ | |
576 | if ((ip->i_d.di_nextents || ip->i_delayed_blks) && | |
577 | (mask & XFS_AT_EXTSIZE) && | |
578 | ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != | |
579 | vap->va_extsize) ) { | |
580 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
581 | goto error_return; | |
582 | } | |
583 | ||
584 | /* | |
585 | * Can't set extent size unless the file is marked, or | |
586 | * about to be marked as a realtime file. | |
587 | * | |
588 | * This check will be removed when fixed size extents | |
589 | * with buffered data writes is implemented. | |
590 | * | |
591 | */ | |
592 | if ((mask & XFS_AT_EXTSIZE) && | |
593 | ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) != | |
594 | vap->va_extsize) && | |
595 | (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) || | |
596 | ((mask & XFS_AT_XFLAGS) && | |
597 | (vap->va_xflags & XFS_XFLAG_REALTIME))))) { | |
598 | code = XFS_ERROR(EINVAL); | |
599 | goto error_return; | |
600 | } | |
601 | ||
602 | /* | |
603 | * Can't change realtime flag if any extents are allocated. | |
604 | */ | |
605 | if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) && | |
606 | (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != | |
607 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { | |
608 | code = XFS_ERROR(EINVAL); /* EFBIG? */ | |
609 | goto error_return; | |
610 | } | |
611 | /* | |
612 | * Extent size must be a multiple of the appropriate block | |
613 | * size, if set at all. | |
614 | */ | |
615 | if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) { | |
616 | xfs_extlen_t size; | |
617 | ||
618 | if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) || | |
619 | ((mask & XFS_AT_XFLAGS) && | |
620 | (vap->va_xflags & XFS_XFLAG_REALTIME))) { | |
621 | size = mp->m_sb.sb_rextsize << | |
622 | mp->m_sb.sb_blocklog; | |
623 | } else { | |
624 | size = mp->m_sb.sb_blocksize; | |
625 | } | |
626 | if (vap->va_extsize % size) { | |
627 | code = XFS_ERROR(EINVAL); | |
628 | goto error_return; | |
629 | } | |
630 | } | |
631 | /* | |
632 | * If realtime flag is set then must have realtime data. | |
633 | */ | |
634 | if ((mask & XFS_AT_XFLAGS) && | |
635 | (vap->va_xflags & XFS_XFLAG_REALTIME)) { | |
636 | if ((mp->m_sb.sb_rblocks == 0) || | |
637 | (mp->m_sb.sb_rextsize == 0) || | |
638 | (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) { | |
639 | code = XFS_ERROR(EINVAL); | |
640 | goto error_return; | |
641 | } | |
642 | } | |
643 | ||
644 | /* | |
645 | * Can't modify an immutable/append-only file unless | |
646 | * we have appropriate permission. | |
647 | */ | |
648 | if ((mask & XFS_AT_XFLAGS) && | |
649 | (ip->i_d.di_flags & | |
650 | (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) || | |
651 | (vap->va_xflags & | |
652 | (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) && | |
653 | !capable(CAP_LINUX_IMMUTABLE)) { | |
654 | code = XFS_ERROR(EPERM); | |
655 | goto error_return; | |
656 | } | |
657 | } | |
658 | ||
659 | /* | |
660 | * Now we can make the changes. Before we join the inode | |
661 | * to the transaction, if XFS_AT_SIZE is set then take care of | |
662 | * the part of the truncation that must be done without the | |
663 | * inode lock. This needs to be done before joining the inode | |
664 | * to the transaction, because the inode cannot be unlocked | |
665 | * once it is a part of the transaction. | |
666 | */ | |
667 | if (mask & XFS_AT_SIZE) { | |
668 | code = 0; | |
669 | if (vap->va_size > ip->i_d.di_size) | |
670 | code = xfs_igrow_start(ip, vap->va_size, credp); | |
671 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
672 | if (!code) | |
673 | code = xfs_itruncate_data(ip, vap->va_size); | |
674 | if (code) { | |
675 | ASSERT(tp == NULL); | |
676 | lock_flags &= ~XFS_ILOCK_EXCL; | |
677 | ASSERT(lock_flags == XFS_IOLOCK_EXCL); | |
678 | goto error_return; | |
679 | } | |
680 | tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); | |
681 | if ((code = xfs_trans_reserve(tp, 0, | |
682 | XFS_ITRUNCATE_LOG_RES(mp), 0, | |
683 | XFS_TRANS_PERM_LOG_RES, | |
684 | XFS_ITRUNCATE_LOG_COUNT))) { | |
685 | xfs_trans_cancel(tp, 0); | |
686 | if (need_iolock) | |
687 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
688 | return code; | |
689 | } | |
690 | commit_flags = XFS_TRANS_RELEASE_LOG_RES; | |
691 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
692 | } | |
693 | ||
694 | if (tp) { | |
695 | xfs_trans_ijoin(tp, ip, lock_flags); | |
696 | xfs_trans_ihold(tp, ip); | |
697 | } | |
698 | ||
699 | /* determine whether mandatory locking mode changes */ | |
700 | mandlock_before = MANDLOCK(vp, ip->i_d.di_mode); | |
701 | ||
702 | /* | |
703 | * Truncate file. Must have write permission and not be a directory. | |
704 | */ | |
705 | if (mask & XFS_AT_SIZE) { | |
706 | if (vap->va_size > ip->i_d.di_size) { | |
707 | xfs_igrow_finish(tp, ip, vap->va_size, | |
708 | !(flags & ATTR_DMI)); | |
709 | } else if ((vap->va_size <= ip->i_d.di_size) || | |
710 | ((vap->va_size == 0) && ip->i_d.di_nextents)) { | |
711 | /* | |
712 | * signal a sync transaction unless | |
713 | * we're truncating an already unlinked | |
714 | * file on a wsync filesystem | |
715 | */ | |
716 | code = xfs_itruncate_finish(&tp, ip, | |
717 | (xfs_fsize_t)vap->va_size, | |
718 | XFS_DATA_FORK, | |
719 | ((ip->i_d.di_nlink != 0 || | |
720 | !(mp->m_flags & XFS_MOUNT_WSYNC)) | |
721 | ? 1 : 0)); | |
722 | if (code) { | |
723 | goto abort_return; | |
724 | } | |
725 | } | |
726 | /* | |
727 | * Have to do this even if the file's size doesn't change. | |
728 | */ | |
729 | timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
730 | } | |
731 | ||
732 | /* | |
733 | * Change file access modes. | |
734 | */ | |
735 | if (mask & XFS_AT_MODE) { | |
736 | ip->i_d.di_mode &= S_IFMT; | |
737 | ip->i_d.di_mode |= vap->va_mode & ~S_IFMT; | |
738 | ||
739 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
740 | timeflags |= XFS_ICHGTIME_CHG; | |
741 | } | |
742 | ||
743 | /* | |
744 | * Change file ownership. Must be the owner or privileged. | |
745 | * If the system was configured with the "restricted_chown" | |
746 | * option, the owner is not permitted to give away the file, | |
747 | * and can change the group id only to a group of which he | |
748 | * or she is a member. | |
749 | */ | |
750 | if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) { | |
751 | /* | |
752 | * CAP_FSETID overrides the following restrictions: | |
753 | * | |
754 | * The set-user-ID and set-group-ID bits of a file will be | |
755 | * cleared upon successful return from chown() | |
756 | */ | |
757 | if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && | |
758 | !capable(CAP_FSETID)) { | |
759 | ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); | |
760 | } | |
761 | ||
762 | /* | |
763 | * Change the ownerships and register quota modifications | |
764 | * in the transaction. | |
765 | */ | |
766 | if (iuid != uid) { | |
767 | if (XFS_IS_UQUOTA_ON(mp)) { | |
768 | ASSERT(mask & XFS_AT_UID); | |
769 | ASSERT(udqp); | |
770 | olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
771 | &ip->i_udquot, udqp); | |
772 | } | |
773 | ip->i_d.di_uid = uid; | |
774 | } | |
775 | if (igid != gid) { | |
776 | if (XFS_IS_GQUOTA_ON(mp)) { | |
777 | ASSERT(mask & XFS_AT_GID); | |
778 | ASSERT(gdqp); | |
779 | olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip, | |
780 | &ip->i_gdquot, gdqp); | |
781 | } | |
782 | ip->i_d.di_gid = gid; | |
783 | } | |
784 | if (iprojid != projid) { | |
785 | ip->i_d.di_projid = projid; | |
786 | /* | |
787 | * We may have to rev the inode as well as | |
788 | * the superblock version number since projids didn't | |
789 | * exist before DINODE_VERSION_2 and SB_VERSION_NLINK. | |
790 | */ | |
791 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) | |
792 | xfs_bump_ino_vers2(tp, ip); | |
793 | } | |
794 | ||
795 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
796 | timeflags |= XFS_ICHGTIME_CHG; | |
797 | } | |
798 | ||
799 | ||
800 | /* | |
801 | * Change file access or modified times. | |
802 | */ | |
803 | if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) { | |
804 | if (mask & XFS_AT_ATIME) { | |
805 | ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec; | |
806 | ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec; | |
807 | ip->i_update_core = 1; | |
808 | timeflags &= ~XFS_ICHGTIME_ACC; | |
809 | } | |
810 | if (mask & XFS_AT_MTIME) { | |
811 | ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec; | |
812 | ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec; | |
813 | timeflags &= ~XFS_ICHGTIME_MOD; | |
814 | timeflags |= XFS_ICHGTIME_CHG; | |
815 | } | |
816 | if (tp && (flags & ATTR_UTIME)) | |
817 | xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE); | |
818 | } | |
819 | ||
820 | /* | |
821 | * Change XFS-added attributes. | |
822 | */ | |
823 | if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) { | |
824 | if (mask & XFS_AT_EXTSIZE) { | |
825 | /* | |
826 | * Converting bytes to fs blocks. | |
827 | */ | |
828 | ip->i_d.di_extsize = vap->va_extsize >> | |
829 | mp->m_sb.sb_blocklog; | |
830 | } | |
831 | if (mask & XFS_AT_XFLAGS) { | |
832 | uint di_flags; | |
833 | ||
834 | /* can't set PREALLOC this way, just preserve it */ | |
835 | di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC); | |
836 | if (vap->va_xflags & XFS_XFLAG_IMMUTABLE) | |
837 | di_flags |= XFS_DIFLAG_IMMUTABLE; | |
838 | if (vap->va_xflags & XFS_XFLAG_APPEND) | |
839 | di_flags |= XFS_DIFLAG_APPEND; | |
840 | if (vap->va_xflags & XFS_XFLAG_SYNC) | |
841 | di_flags |= XFS_DIFLAG_SYNC; | |
842 | if (vap->va_xflags & XFS_XFLAG_NOATIME) | |
843 | di_flags |= XFS_DIFLAG_NOATIME; | |
844 | if (vap->va_xflags & XFS_XFLAG_NODUMP) | |
845 | di_flags |= XFS_DIFLAG_NODUMP; | |
846 | if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { | |
847 | if (vap->va_xflags & XFS_XFLAG_RTINHERIT) | |
848 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
849 | if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS) | |
850 | di_flags |= XFS_DIFLAG_NOSYMLINKS; | |
851 | } else { | |
852 | if (vap->va_xflags & XFS_XFLAG_REALTIME) { | |
853 | di_flags |= XFS_DIFLAG_REALTIME; | |
854 | ip->i_iocore.io_flags |= XFS_IOCORE_RT; | |
855 | } else { | |
856 | ip->i_iocore.io_flags &= ~XFS_IOCORE_RT; | |
857 | } | |
858 | } | |
859 | ip->i_d.di_flags = di_flags; | |
860 | } | |
861 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
862 | timeflags |= XFS_ICHGTIME_CHG; | |
863 | } | |
864 | ||
865 | /* | |
866 | * Change file inode change time only if XFS_AT_CTIME set | |
867 | * AND we have been called by a DMI function. | |
868 | */ | |
869 | ||
870 | if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) { | |
871 | ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec; | |
872 | ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec; | |
873 | ip->i_update_core = 1; | |
874 | timeflags &= ~XFS_ICHGTIME_CHG; | |
875 | } | |
876 | ||
877 | /* | |
878 | * Send out timestamp changes that need to be set to the | |
879 | * current time. Not done when called by a DMI function. | |
880 | */ | |
881 | if (timeflags && !(flags & ATTR_DMI)) | |
882 | xfs_ichgtime(ip, timeflags); | |
883 | ||
884 | XFS_STATS_INC(xs_ig_attrchg); | |
885 | ||
886 | /* | |
887 | * If this is a synchronous mount, make sure that the | |
888 | * transaction goes to disk before returning to the user. | |
889 | * This is slightly sub-optimal in that truncates require | |
890 | * two sync transactions instead of one for wsync filesytems. | |
891 | * One for the truncate and one for the timestamps since we | |
892 | * don't want to change the timestamps unless we're sure the | |
893 | * truncate worked. Truncates are less than 1% of the laddis | |
894 | * mix so this probably isn't worth the trouble to optimize. | |
895 | */ | |
896 | code = 0; | |
897 | if (tp) { | |
898 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
899 | xfs_trans_set_sync(tp); | |
900 | ||
901 | code = xfs_trans_commit(tp, commit_flags, NULL); | |
902 | } | |
903 | ||
904 | /* | |
905 | * If the (regular) file's mandatory locking mode changed, then | |
906 | * notify the vnode. We do this under the inode lock to prevent | |
907 | * racing calls to vop_vnode_change. | |
908 | */ | |
909 | mandlock_after = MANDLOCK(vp, ip->i_d.di_mode); | |
910 | if (mandlock_before != mandlock_after) { | |
911 | VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING, | |
912 | mandlock_after); | |
913 | } | |
914 | ||
915 | xfs_iunlock(ip, lock_flags); | |
916 | ||
917 | /* | |
918 | * Release any dquot(s) the inode had kept before chown. | |
919 | */ | |
920 | XFS_QM_DQRELE(mp, olddquot1); | |
921 | XFS_QM_DQRELE(mp, olddquot2); | |
922 | XFS_QM_DQRELE(mp, udqp); | |
923 | XFS_QM_DQRELE(mp, gdqp); | |
924 | ||
925 | if (code) { | |
926 | return code; | |
927 | } | |
928 | ||
929 | if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) && | |
930 | !(flags & ATTR_DMI)) { | |
931 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL, | |
932 | NULL, DM_RIGHT_NULL, NULL, NULL, | |
933 | 0, 0, AT_DELAY_FLAG(flags)); | |
934 | } | |
935 | return 0; | |
936 | ||
937 | abort_return: | |
938 | commit_flags |= XFS_TRANS_ABORT; | |
939 | /* FALLTHROUGH */ | |
940 | error_return: | |
941 | XFS_QM_DQRELE(mp, udqp); | |
942 | XFS_QM_DQRELE(mp, gdqp); | |
943 | if (tp) { | |
944 | xfs_trans_cancel(tp, commit_flags); | |
945 | } | |
946 | if (lock_flags != 0) { | |
947 | xfs_iunlock(ip, lock_flags); | |
948 | } | |
949 | return code; | |
950 | } | |
951 | ||
952 | ||
953 | /* | |
954 | * xfs_access | |
955 | * Null conversion from vnode mode bits to inode mode bits, as in efs. | |
956 | */ | |
957 | STATIC int | |
958 | xfs_access( | |
959 | bhv_desc_t *bdp, | |
960 | int mode, | |
961 | cred_t *credp) | |
962 | { | |
963 | xfs_inode_t *ip; | |
964 | int error; | |
965 | ||
966 | vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__, | |
967 | (inst_t *)__return_address); | |
968 | ||
969 | ip = XFS_BHVTOI(bdp); | |
970 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
971 | error = xfs_iaccess(ip, mode, credp); | |
972 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
973 | return error; | |
974 | } | |
975 | ||
976 | ||
977 | /* | |
978 | * xfs_readlink | |
979 | * | |
980 | */ | |
981 | STATIC int | |
982 | xfs_readlink( | |
983 | bhv_desc_t *bdp, | |
984 | uio_t *uiop, | |
985 | int ioflags, | |
986 | cred_t *credp) | |
987 | { | |
988 | xfs_inode_t *ip; | |
989 | int count; | |
990 | xfs_off_t offset; | |
991 | int pathlen; | |
992 | vnode_t *vp; | |
993 | int error = 0; | |
994 | xfs_mount_t *mp; | |
995 | int nmaps; | |
996 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
997 | xfs_daddr_t d; | |
998 | int byte_cnt; | |
999 | int n; | |
1000 | xfs_buf_t *bp; | |
1001 | ||
1002 | vp = BHV_TO_VNODE(bdp); | |
1003 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
1004 | ||
1005 | ip = XFS_BHVTOI(bdp); | |
1006 | mp = ip->i_mount; | |
1007 | ||
1008 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1009 | return XFS_ERROR(EIO); | |
1010 | ||
1011 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1012 | ||
1013 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK); | |
1014 | ||
1015 | offset = uiop->uio_offset; | |
1016 | count = uiop->uio_resid; | |
1017 | ||
1018 | if (offset < 0) { | |
1019 | error = XFS_ERROR(EINVAL); | |
1020 | goto error_return; | |
1021 | } | |
1022 | if (count <= 0) { | |
1023 | error = 0; | |
1024 | goto error_return; | |
1025 | } | |
1026 | ||
1027 | if (!(ioflags & IO_INVIS)) { | |
1028 | xfs_ichgtime(ip, XFS_ICHGTIME_ACC); | |
1029 | } | |
1030 | ||
1031 | /* | |
1032 | * See if the symlink is stored inline. | |
1033 | */ | |
1034 | pathlen = (int)ip->i_d.di_size; | |
1035 | ||
1036 | if (ip->i_df.if_flags & XFS_IFINLINE) { | |
1037 | error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop); | |
1038 | } | |
1039 | else { | |
1040 | /* | |
1041 | * Symlink not inline. Call bmap to get it in. | |
1042 | */ | |
1043 | nmaps = SYMLINK_MAPS; | |
1044 | ||
1045 | error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), | |
1046 | 0, NULL, 0, mval, &nmaps, NULL); | |
1047 | ||
1048 | if (error) { | |
1049 | goto error_return; | |
1050 | } | |
1051 | ||
1052 | for (n = 0; n < nmaps; n++) { | |
1053 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
1054 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
1055 | bp = xfs_buf_read(mp->m_ddev_targp, d, | |
1056 | BTOBB(byte_cnt), 0); | |
1057 | error = XFS_BUF_GETERROR(bp); | |
1058 | if (error) { | |
1059 | xfs_ioerror_alert("xfs_readlink", | |
1060 | ip->i_mount, bp, XFS_BUF_ADDR(bp)); | |
1061 | xfs_buf_relse(bp); | |
1062 | goto error_return; | |
1063 | } | |
1064 | if (pathlen < byte_cnt) | |
1065 | byte_cnt = pathlen; | |
1066 | pathlen -= byte_cnt; | |
1067 | ||
1068 | error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop); | |
1069 | xfs_buf_relse (bp); | |
1070 | } | |
1071 | ||
1072 | } | |
1073 | ||
1074 | ||
1075 | error_return: | |
1076 | ||
1077 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1078 | ||
1079 | return error; | |
1080 | } | |
1081 | ||
1082 | ||
1083 | /* | |
1084 | * xfs_fsync | |
1085 | * | |
1086 | * This is called to sync the inode and its data out to disk. | |
1087 | * We need to hold the I/O lock while flushing the data, and | |
1088 | * the inode lock while flushing the inode. The inode lock CANNOT | |
1089 | * be held while flushing the data, so acquire after we're done | |
1090 | * with that. | |
1091 | */ | |
1092 | STATIC int | |
1093 | xfs_fsync( | |
1094 | bhv_desc_t *bdp, | |
1095 | int flag, | |
1096 | cred_t *credp, | |
1097 | xfs_off_t start, | |
1098 | xfs_off_t stop) | |
1099 | { | |
1100 | xfs_inode_t *ip; | |
1101 | xfs_trans_t *tp; | |
1102 | int error; | |
1103 | ||
1104 | vn_trace_entry(BHV_TO_VNODE(bdp), | |
1105 | __FUNCTION__, (inst_t *)__return_address); | |
1106 | ||
1107 | ip = XFS_BHVTOI(bdp); | |
1108 | ||
1109 | ASSERT(start >= 0 && stop >= -1); | |
1110 | ||
1111 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
1112 | return XFS_ERROR(EIO); | |
1113 | ||
1114 | /* | |
1115 | * We always need to make sure that the required inode state | |
1116 | * is safe on disk. The vnode might be clean but because | |
1117 | * of committed transactions that haven't hit the disk yet. | |
1118 | * Likewise, there could be unflushed non-transactional | |
1119 | * changes to the inode core that have to go to disk. | |
1120 | * | |
1121 | * The following code depends on one assumption: that | |
1122 | * any transaction that changes an inode logs the core | |
1123 | * because it has to change some field in the inode core | |
1124 | * (typically nextents or nblocks). That assumption | |
1125 | * implies that any transactions against an inode will | |
1126 | * catch any non-transactional updates. If inode-altering | |
1127 | * transactions exist that violate this assumption, the | |
1128 | * code breaks. Right now, it figures that if the involved | |
1129 | * update_* field is clear and the inode is unpinned, the | |
1130 | * inode is clean. Either it's been flushed or it's been | |
1131 | * committed and the commit has hit the disk unpinning the inode. | |
1132 | * (Note that xfs_inode_item_format() called at commit clears | |
1133 | * the update_* fields.) | |
1134 | */ | |
1135 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1136 | ||
1137 | /* If we are flushing data then we care about update_size | |
1138 | * being set, otherwise we care about update_core | |
1139 | */ | |
1140 | if ((flag & FSYNC_DATA) ? | |
1141 | (ip->i_update_size == 0) : | |
1142 | (ip->i_update_core == 0)) { | |
1143 | /* | |
1144 | * Timestamps/size haven't changed since last inode | |
1145 | * flush or inode transaction commit. That means | |
1146 | * either nothing got written or a transaction | |
1147 | * committed which caught the updates. If the | |
1148 | * latter happened and the transaction hasn't | |
1149 | * hit the disk yet, the inode will be still | |
1150 | * be pinned. If it is, force the log. | |
1151 | */ | |
1152 | ||
1153 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1154 | ||
1155 | if (xfs_ipincount(ip)) { | |
1156 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, | |
1157 | XFS_LOG_FORCE | | |
1158 | ((flag & FSYNC_WAIT) | |
1159 | ? XFS_LOG_SYNC : 0)); | |
1160 | } | |
1161 | error = 0; | |
1162 | } else { | |
1163 | /* | |
1164 | * Kick off a transaction to log the inode | |
1165 | * core to get the updates. Make it | |
1166 | * sync if FSYNC_WAIT is passed in (which | |
1167 | * is done by everybody but specfs). The | |
1168 | * sync transaction will also force the log. | |
1169 | */ | |
1170 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1171 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
1172 | if ((error = xfs_trans_reserve(tp, 0, | |
1173 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), | |
1174 | 0, 0, 0))) { | |
1175 | xfs_trans_cancel(tp, 0); | |
1176 | return error; | |
1177 | } | |
1178 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1179 | ||
1180 | /* | |
1181 | * Note - it's possible that we might have pushed | |
1182 | * ourselves out of the way during trans_reserve | |
1183 | * which would flush the inode. But there's no | |
1184 | * guarantee that the inode buffer has actually | |
1185 | * gone out yet (it's delwri). Plus the buffer | |
1186 | * could be pinned anyway if it's part of an | |
1187 | * inode in another recent transaction. So we | |
1188 | * play it safe and fire off the transaction anyway. | |
1189 | */ | |
1190 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1191 | xfs_trans_ihold(tp, ip); | |
1192 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1193 | if (flag & FSYNC_WAIT) | |
1194 | xfs_trans_set_sync(tp); | |
1195 | error = xfs_trans_commit(tp, 0, NULL); | |
1196 | ||
1197 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1198 | } | |
1199 | return error; | |
1200 | } | |
1201 | ||
1202 | /* | |
1203 | * This is called by xfs_inactive to free any blocks beyond eof, | |
1204 | * when the link count isn't zero. | |
1205 | */ | |
1206 | STATIC int | |
1207 | xfs_inactive_free_eofblocks( | |
1208 | xfs_mount_t *mp, | |
1209 | xfs_inode_t *ip) | |
1210 | { | |
1211 | xfs_trans_t *tp; | |
1212 | int error; | |
1213 | xfs_fileoff_t end_fsb; | |
1214 | xfs_fileoff_t last_fsb; | |
1215 | xfs_filblks_t map_len; | |
1216 | int nimaps; | |
1217 | xfs_bmbt_irec_t imap; | |
1218 | ||
1219 | /* | |
1220 | * Figure out if there are any blocks beyond the end | |
1221 | * of the file. If not, then there is nothing to do. | |
1222 | */ | |
1223 | end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size)); | |
1224 | last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); | |
1225 | map_len = last_fsb - end_fsb; | |
1226 | if (map_len <= 0) | |
1227 | return (0); | |
1228 | ||
1229 | nimaps = 1; | |
1230 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1231 | error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0, | |
1232 | NULL, 0, &imap, &nimaps, NULL); | |
1233 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
1234 | ||
1235 | if (!error && (nimaps != 0) && | |
1236 | (imap.br_startblock != HOLESTARTBLOCK)) { | |
1237 | /* | |
1238 | * Attach the dquots to the inode up front. | |
1239 | */ | |
1240 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
1241 | return (error); | |
1242 | ||
1243 | /* | |
1244 | * There are blocks after the end of file. | |
1245 | * Free them up now by truncating the file to | |
1246 | * its current size. | |
1247 | */ | |
1248 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1249 | ||
1250 | /* | |
1251 | * Do the xfs_itruncate_start() call before | |
1252 | * reserving any log space because | |
1253 | * itruncate_start will call into the buffer | |
1254 | * cache and we can't | |
1255 | * do that within a transaction. | |
1256 | */ | |
1257 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1258 | xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, | |
1259 | ip->i_d.di_size); | |
1260 | ||
1261 | error = xfs_trans_reserve(tp, 0, | |
1262 | XFS_ITRUNCATE_LOG_RES(mp), | |
1263 | 0, XFS_TRANS_PERM_LOG_RES, | |
1264 | XFS_ITRUNCATE_LOG_COUNT); | |
1265 | if (error) { | |
1266 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1267 | xfs_trans_cancel(tp, 0); | |
1268 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1269 | return (error); | |
1270 | } | |
1271 | ||
1272 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1273 | xfs_trans_ijoin(tp, ip, | |
1274 | XFS_IOLOCK_EXCL | | |
1275 | XFS_ILOCK_EXCL); | |
1276 | xfs_trans_ihold(tp, ip); | |
1277 | ||
1278 | error = xfs_itruncate_finish(&tp, ip, | |
1279 | ip->i_d.di_size, | |
1280 | XFS_DATA_FORK, | |
1281 | 0); | |
1282 | /* | |
1283 | * If we get an error at this point we | |
1284 | * simply don't bother truncating the file. | |
1285 | */ | |
1286 | if (error) { | |
1287 | xfs_trans_cancel(tp, | |
1288 | (XFS_TRANS_RELEASE_LOG_RES | | |
1289 | XFS_TRANS_ABORT)); | |
1290 | } else { | |
1291 | error = xfs_trans_commit(tp, | |
1292 | XFS_TRANS_RELEASE_LOG_RES, | |
1293 | NULL); | |
1294 | } | |
1295 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1296 | } | |
1297 | return (error); | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Free a symlink that has blocks associated with it. | |
1302 | */ | |
1303 | STATIC int | |
1304 | xfs_inactive_symlink_rmt( | |
1305 | xfs_inode_t *ip, | |
1306 | xfs_trans_t **tpp) | |
1307 | { | |
1308 | xfs_buf_t *bp; | |
1309 | int committed; | |
1310 | int done; | |
1311 | int error; | |
1312 | xfs_fsblock_t first_block; | |
1313 | xfs_bmap_free_t free_list; | |
1314 | int i; | |
1315 | xfs_mount_t *mp; | |
1316 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
1317 | int nmaps; | |
1318 | xfs_trans_t *ntp; | |
1319 | int size; | |
1320 | xfs_trans_t *tp; | |
1321 | ||
1322 | tp = *tpp; | |
1323 | mp = ip->i_mount; | |
1324 | ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip)); | |
1325 | /* | |
1326 | * We're freeing a symlink that has some | |
1327 | * blocks allocated to it. Free the | |
1328 | * blocks here. We know that we've got | |
1329 | * either 1 or 2 extents and that we can | |
1330 | * free them all in one bunmapi call. | |
1331 | */ | |
1332 | ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); | |
1333 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1334 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1335 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1336 | xfs_trans_cancel(tp, 0); | |
1337 | *tpp = NULL; | |
1338 | return error; | |
1339 | } | |
1340 | /* | |
1341 | * Lock the inode, fix the size, and join it to the transaction. | |
1342 | * Hold it so in the normal path, we still have it locked for | |
1343 | * the second transaction. In the error paths we need it | |
1344 | * held so the cancel won't rele it, see below. | |
1345 | */ | |
1346 | xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1347 | size = (int)ip->i_d.di_size; | |
1348 | ip->i_d.di_size = 0; | |
1349 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1350 | xfs_trans_ihold(tp, ip); | |
1351 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1352 | /* | |
1353 | * Find the block(s) so we can inval and unmap them. | |
1354 | */ | |
1355 | done = 0; | |
1356 | XFS_BMAP_INIT(&free_list, &first_block); | |
1357 | nmaps = sizeof(mval) / sizeof(mval[0]); | |
1358 | if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size), | |
1359 | XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps, | |
1360 | &free_list))) | |
1361 | goto error0; | |
1362 | /* | |
1363 | * Invalidate the block(s). | |
1364 | */ | |
1365 | for (i = 0; i < nmaps; i++) { | |
1366 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, | |
1367 | XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), | |
1368 | XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); | |
1369 | xfs_trans_binval(tp, bp); | |
1370 | } | |
1371 | /* | |
1372 | * Unmap the dead block(s) to the free_list. | |
1373 | */ | |
1374 | if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps, | |
1375 | &first_block, &free_list, &done))) | |
1376 | goto error1; | |
1377 | ASSERT(done); | |
1378 | /* | |
1379 | * Commit the first transaction. This logs the EFI and the inode. | |
1380 | */ | |
1381 | if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed))) | |
1382 | goto error1; | |
1383 | /* | |
1384 | * The transaction must have been committed, since there were | |
1385 | * actually extents freed by xfs_bunmapi. See xfs_bmap_finish. | |
1386 | * The new tp has the extent freeing and EFDs. | |
1387 | */ | |
1388 | ASSERT(committed); | |
1389 | /* | |
1390 | * The first xact was committed, so add the inode to the new one. | |
1391 | * Mark it dirty so it will be logged and moved forward in the log as | |
1392 | * part of every commit. | |
1393 | */ | |
1394 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1395 | xfs_trans_ihold(tp, ip); | |
1396 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1397 | /* | |
1398 | * Get a new, empty transaction to return to our caller. | |
1399 | */ | |
1400 | ntp = xfs_trans_dup(tp); | |
1401 | /* | |
1402 | * Commit the transaction containing extent freeing and EFD's. | |
1403 | * If we get an error on the commit here or on the reserve below, | |
1404 | * we need to unlock the inode since the new transaction doesn't | |
1405 | * have the inode attached. | |
1406 | */ | |
1407 | error = xfs_trans_commit(tp, 0, NULL); | |
1408 | tp = ntp; | |
1409 | if (error) { | |
1410 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1411 | goto error0; | |
1412 | } | |
1413 | /* | |
1414 | * Remove the memory for extent descriptions (just bookkeeping). | |
1415 | */ | |
1416 | if (ip->i_df.if_bytes) | |
1417 | xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); | |
1418 | ASSERT(ip->i_df.if_bytes == 0); | |
1419 | /* | |
1420 | * Put an itruncate log reservation in the new transaction | |
1421 | * for our caller. | |
1422 | */ | |
1423 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1424 | XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) { | |
1425 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1426 | goto error0; | |
1427 | } | |
1428 | /* | |
1429 | * Return with the inode locked but not joined to the transaction. | |
1430 | */ | |
1431 | *tpp = tp; | |
1432 | return 0; | |
1433 | ||
1434 | error1: | |
1435 | xfs_bmap_cancel(&free_list); | |
1436 | error0: | |
1437 | /* | |
1438 | * Have to come here with the inode locked and either | |
1439 | * (held and in the transaction) or (not in the transaction). | |
1440 | * If the inode isn't held then cancel would iput it, but | |
1441 | * that's wrong since this is inactive and the vnode ref | |
1442 | * count is 0 already. | |
1443 | * Cancel won't do anything to the inode if held, but it still | |
1444 | * needs to be locked until the cancel is done, if it was | |
1445 | * joined to the transaction. | |
1446 | */ | |
1447 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1448 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1449 | *tpp = NULL; | |
1450 | return error; | |
1451 | ||
1452 | } | |
1453 | ||
1454 | STATIC int | |
1455 | xfs_inactive_symlink_local( | |
1456 | xfs_inode_t *ip, | |
1457 | xfs_trans_t **tpp) | |
1458 | { | |
1459 | int error; | |
1460 | ||
1461 | ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip)); | |
1462 | /* | |
1463 | * We're freeing a symlink which fit into | |
1464 | * the inode. Just free the memory used | |
1465 | * to hold the old symlink. | |
1466 | */ | |
1467 | error = xfs_trans_reserve(*tpp, 0, | |
1468 | XFS_ITRUNCATE_LOG_RES(ip->i_mount), | |
1469 | 0, XFS_TRANS_PERM_LOG_RES, | |
1470 | XFS_ITRUNCATE_LOG_COUNT); | |
1471 | ||
1472 | if (error) { | |
1473 | xfs_trans_cancel(*tpp, 0); | |
1474 | *tpp = NULL; | |
1475 | return (error); | |
1476 | } | |
1477 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1478 | ||
1479 | /* | |
1480 | * Zero length symlinks _can_ exist. | |
1481 | */ | |
1482 | if (ip->i_df.if_bytes > 0) { | |
1483 | xfs_idata_realloc(ip, | |
1484 | -(ip->i_df.if_bytes), | |
1485 | XFS_DATA_FORK); | |
1486 | ASSERT(ip->i_df.if_bytes == 0); | |
1487 | } | |
1488 | return (0); | |
1489 | } | |
1490 | ||
1491 | /* | |
1492 | * | |
1493 | */ | |
1494 | STATIC int | |
1495 | xfs_inactive_attrs( | |
1496 | xfs_inode_t *ip, | |
1497 | xfs_trans_t **tpp) | |
1498 | { | |
1499 | xfs_trans_t *tp; | |
1500 | int error; | |
1501 | xfs_mount_t *mp; | |
1502 | ||
1503 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); | |
1504 | tp = *tpp; | |
1505 | mp = ip->i_mount; | |
1506 | ASSERT(ip->i_d.di_forkoff != 0); | |
1507 | xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
1508 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1509 | ||
1510 | error = xfs_attr_inactive(ip); | |
1511 | if (error) { | |
1512 | *tpp = NULL; | |
1513 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1514 | return (error); /* goto out*/ | |
1515 | } | |
1516 | ||
1517 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1518 | error = xfs_trans_reserve(tp, 0, | |
1519 | XFS_IFREE_LOG_RES(mp), | |
1520 | 0, XFS_TRANS_PERM_LOG_RES, | |
1521 | XFS_INACTIVE_LOG_COUNT); | |
1522 | if (error) { | |
1523 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1524 | xfs_trans_cancel(tp, 0); | |
1525 | *tpp = NULL; | |
1526 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1527 | return (error); | |
1528 | } | |
1529 | ||
1530 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1531 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1532 | xfs_trans_ihold(tp, ip); | |
1533 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1534 | ||
1535 | ASSERT(ip->i_d.di_anextents == 0); | |
1536 | ||
1537 | *tpp = tp; | |
1538 | return (0); | |
1539 | } | |
1540 | ||
1541 | STATIC int | |
1542 | xfs_release( | |
1543 | bhv_desc_t *bdp) | |
1544 | { | |
1545 | xfs_inode_t *ip; | |
1546 | vnode_t *vp; | |
1547 | xfs_mount_t *mp; | |
1548 | int error; | |
1549 | ||
1550 | vp = BHV_TO_VNODE(bdp); | |
1551 | ip = XFS_BHVTOI(bdp); | |
1552 | ||
1553 | if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) { | |
1554 | return 0; | |
1555 | } | |
1556 | ||
1557 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1558 | if (vp->v_vfsp->vfs_flag & VFS_RDONLY) | |
1559 | return 0; | |
1560 | ||
1561 | #ifdef HAVE_REFCACHE | |
1562 | /* If we are in the NFS reference cache then don't do this now */ | |
1563 | if (ip->i_refcache) | |
1564 | return 0; | |
1565 | #endif | |
1566 | ||
1567 | mp = ip->i_mount; | |
1568 | ||
1569 | if (ip->i_d.di_nlink != 0) { | |
1570 | if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) && | |
1571 | ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) && | |
1572 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && | |
1573 | (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) { | |
1574 | if ((error = xfs_inactive_free_eofblocks(mp, ip))) | |
1575 | return (error); | |
1576 | /* Update linux inode block count after free above */ | |
1577 | LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp, | |
1578 | ip->i_d.di_nblocks + ip->i_delayed_blks); | |
1579 | } | |
1580 | } | |
1581 | ||
1582 | return 0; | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * xfs_inactive | |
1587 | * | |
1588 | * This is called when the vnode reference count for the vnode | |
1589 | * goes to zero. If the file has been unlinked, then it must | |
1590 | * now be truncated. Also, we clear all of the read-ahead state | |
1591 | * kept for the inode here since the file is now closed. | |
1592 | */ | |
1593 | STATIC int | |
1594 | xfs_inactive( | |
1595 | bhv_desc_t *bdp, | |
1596 | cred_t *credp) | |
1597 | { | |
1598 | xfs_inode_t *ip; | |
1599 | vnode_t *vp; | |
1600 | xfs_bmap_free_t free_list; | |
1601 | xfs_fsblock_t first_block; | |
1602 | int committed; | |
1603 | xfs_trans_t *tp; | |
1604 | xfs_mount_t *mp; | |
1605 | int error; | |
1606 | int truncate; | |
1607 | ||
1608 | vp = BHV_TO_VNODE(bdp); | |
1609 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
1610 | ||
1611 | ip = XFS_BHVTOI(bdp); | |
1612 | ||
1613 | /* | |
1614 | * If the inode is already free, then there can be nothing | |
1615 | * to clean up here. | |
1616 | */ | |
1617 | if (ip->i_d.di_mode == 0 || VN_BAD(vp)) { | |
1618 | ASSERT(ip->i_df.if_real_bytes == 0); | |
1619 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1620 | return VN_INACTIVE_CACHE; | |
1621 | } | |
1622 | ||
1623 | /* | |
1624 | * Only do a truncate if it's a regular file with | |
1625 | * some actual space in it. It's OK to look at the | |
1626 | * inode's fields without the lock because we're the | |
1627 | * only one with a reference to the inode. | |
1628 | */ | |
1629 | truncate = ((ip->i_d.di_nlink == 0) && | |
1630 | ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) && | |
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) && | |
1648 | ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) && | |
1649 | (ip->i_df.if_flags & XFS_IFEXTENTS)) && | |
1650 | (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) || | |
1651 | (ip->i_delayed_blks != 0))) { | |
1652 | if ((error = xfs_inactive_free_eofblocks(mp, ip))) | |
1653 | return (VN_INACTIVE_CACHE); | |
1654 | /* Update linux inode block count after free above */ | |
1655 | LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp, | |
1656 | ip->i_d.di_nblocks + ip->i_delayed_blks); | |
1657 | } | |
1658 | goto out; | |
1659 | } | |
1660 | ||
1661 | ASSERT(ip->i_d.di_nlink == 0); | |
1662 | ||
1663 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
1664 | return (VN_INACTIVE_CACHE); | |
1665 | ||
1666 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1667 | if (truncate) { | |
1668 | /* | |
1669 | * Do the xfs_itruncate_start() call before | |
1670 | * reserving any log space because itruncate_start | |
1671 | * will call into the buffer cache and we can't | |
1672 | * do that within a transaction. | |
1673 | */ | |
1674 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
1675 | ||
1676 | xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0); | |
1677 | ||
1678 | error = xfs_trans_reserve(tp, 0, | |
1679 | XFS_ITRUNCATE_LOG_RES(mp), | |
1680 | 0, XFS_TRANS_PERM_LOG_RES, | |
1681 | XFS_ITRUNCATE_LOG_COUNT); | |
1682 | if (error) { | |
1683 | /* Don't call itruncate_cleanup */ | |
1684 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1685 | xfs_trans_cancel(tp, 0); | |
1686 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1687 | return (VN_INACTIVE_CACHE); | |
1688 | } | |
1689 | ||
1690 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1691 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1692 | xfs_trans_ihold(tp, ip); | |
1693 | ||
1694 | /* | |
1695 | * normally, we have to run xfs_itruncate_finish sync. | |
1696 | * But if filesystem is wsync and we're in the inactive | |
1697 | * path, then we know that nlink == 0, and that the | |
1698 | * xaction that made nlink == 0 is permanently committed | |
1699 | * since xfs_remove runs as a synchronous transaction. | |
1700 | */ | |
1701 | error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK, | |
1702 | (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0)); | |
1703 | ||
1704 | if (error) { | |
1705 | xfs_trans_cancel(tp, | |
1706 | XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1707 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1708 | return (VN_INACTIVE_CACHE); | |
1709 | } | |
1710 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) { | |
1711 | ||
1712 | /* | |
1713 | * If we get an error while cleaning up a | |
1714 | * symlink we bail out. | |
1715 | */ | |
1716 | error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ? | |
1717 | xfs_inactive_symlink_rmt(ip, &tp) : | |
1718 | xfs_inactive_symlink_local(ip, &tp); | |
1719 | ||
1720 | if (error) { | |
1721 | ASSERT(tp == NULL); | |
1722 | return (VN_INACTIVE_CACHE); | |
1723 | } | |
1724 | ||
1725 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1726 | xfs_trans_ihold(tp, ip); | |
1727 | } else { | |
1728 | error = xfs_trans_reserve(tp, 0, | |
1729 | XFS_IFREE_LOG_RES(mp), | |
1730 | 0, XFS_TRANS_PERM_LOG_RES, | |
1731 | XFS_INACTIVE_LOG_COUNT); | |
1732 | if (error) { | |
1733 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1734 | xfs_trans_cancel(tp, 0); | |
1735 | return (VN_INACTIVE_CACHE); | |
1736 | } | |
1737 | ||
1738 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1739 | xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1740 | xfs_trans_ihold(tp, ip); | |
1741 | } | |
1742 | ||
1743 | /* | |
1744 | * If there are attributes associated with the file | |
1745 | * then blow them away now. The code calls a routine | |
1746 | * that recursively deconstructs the attribute fork. | |
1747 | * We need to just commit the current transaction | |
1748 | * because we can't use it for xfs_attr_inactive(). | |
1749 | */ | |
1750 | if (ip->i_d.di_anextents > 0) { | |
1751 | error = xfs_inactive_attrs(ip, &tp); | |
1752 | /* | |
1753 | * If we got an error, the transaction is already | |
1754 | * cancelled, and the inode is unlocked. Just get out. | |
1755 | */ | |
1756 | if (error) | |
1757 | return (VN_INACTIVE_CACHE); | |
1758 | } else if (ip->i_afp) { | |
1759 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1760 | } | |
1761 | ||
1762 | /* | |
1763 | * Free the inode. | |
1764 | */ | |
1765 | XFS_BMAP_INIT(&free_list, &first_block); | |
1766 | error = xfs_ifree(tp, ip, &free_list); | |
1767 | if (error) { | |
1768 | /* | |
1769 | * If we fail to free the inode, shut down. The cancel | |
1770 | * might do that, we need to make sure. Otherwise the | |
1771 | * inode might be lost for a long time or forever. | |
1772 | */ | |
1773 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1774 | cmn_err(CE_NOTE, | |
1775 | "xfs_inactive: xfs_ifree() returned an error = %d on %s", | |
1776 | error, mp->m_fsname); | |
1777 | xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR); | |
1778 | } | |
1779 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1780 | } else { | |
1781 | /* | |
1782 | * Credit the quota account(s). The inode is gone. | |
1783 | */ | |
1784 | XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1785 | ||
1786 | /* | |
1787 | * Just ignore errors at this point. There is | |
1788 | * nothing we can do except to try to keep going. | |
1789 | */ | |
1790 | (void) xfs_bmap_finish(&tp, &free_list, first_block, | |
1791 | &committed); | |
1792 | (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
1793 | } | |
1794 | /* | |
1795 | * Release the dquots held by inode, if any. | |
1796 | */ | |
1797 | XFS_QM_DQDETACH(mp, ip); | |
1798 | ||
1799 | xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); | |
1800 | ||
1801 | out: | |
1802 | return VN_INACTIVE_CACHE; | |
1803 | } | |
1804 | ||
1805 | ||
1806 | /* | |
1807 | * xfs_lookup | |
1808 | */ | |
1809 | STATIC int | |
1810 | xfs_lookup( | |
1811 | bhv_desc_t *dir_bdp, | |
1812 | vname_t *dentry, | |
1813 | vnode_t **vpp, | |
1814 | int flags, | |
1815 | vnode_t *rdir, | |
1816 | cred_t *credp) | |
1817 | { | |
1818 | xfs_inode_t *dp, *ip; | |
1819 | xfs_ino_t e_inum; | |
1820 | int error; | |
1821 | uint lock_mode; | |
1822 | vnode_t *dir_vp; | |
1823 | ||
1824 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
1825 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
1826 | ||
1827 | dp = XFS_BHVTOI(dir_bdp); | |
1828 | ||
1829 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
1830 | return XFS_ERROR(EIO); | |
1831 | ||
1832 | lock_mode = xfs_ilock_map_shared(dp); | |
1833 | error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip); | |
1834 | if (!error) { | |
1835 | *vpp = XFS_ITOV(ip); | |
1836 | ITRACE(ip); | |
1837 | } | |
1838 | xfs_iunlock_map_shared(dp, lock_mode); | |
1839 | return error; | |
1840 | } | |
1841 | ||
1842 | ||
1843 | /* | |
1844 | * xfs_create (create a new file). | |
1845 | */ | |
1846 | STATIC int | |
1847 | xfs_create( | |
1848 | bhv_desc_t *dir_bdp, | |
1849 | vname_t *dentry, | |
1850 | vattr_t *vap, | |
1851 | vnode_t **vpp, | |
1852 | cred_t *credp) | |
1853 | { | |
1854 | char *name = VNAME(dentry); | |
1855 | vnode_t *dir_vp; | |
1856 | xfs_inode_t *dp, *ip; | |
1857 | vnode_t *vp=NULL; | |
1858 | xfs_trans_t *tp; | |
1859 | xfs_mount_t *mp; | |
1860 | xfs_dev_t rdev; | |
1861 | int error; | |
1862 | xfs_bmap_free_t free_list; | |
1863 | xfs_fsblock_t first_block; | |
1864 | boolean_t dp_joined_to_trans; | |
1865 | int dm_event_sent = 0; | |
1866 | uint cancel_flags; | |
1867 | int committed; | |
1868 | xfs_prid_t prid; | |
1869 | struct xfs_dquot *udqp, *gdqp; | |
1870 | uint resblks; | |
1871 | int dm_di_mode; | |
1872 | int namelen; | |
1873 | ||
1874 | ASSERT(!*vpp); | |
1875 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
1876 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
1877 | ||
1878 | dp = XFS_BHVTOI(dir_bdp); | |
1879 | mp = dp->i_mount; | |
1880 | ||
1881 | dm_di_mode = vap->va_mode|VTTOIF(vap->va_type); | |
1882 | namelen = VNAMELEN(dentry); | |
1883 | ||
1884 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) { | |
1885 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, | |
1886 | dir_vp, DM_RIGHT_NULL, NULL, | |
1887 | DM_RIGHT_NULL, name, NULL, | |
1888 | dm_di_mode, 0, 0); | |
1889 | ||
1890 | if (error) | |
1891 | return error; | |
1892 | dm_event_sent = 1; | |
1893 | } | |
1894 | ||
1895 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1896 | return XFS_ERROR(EIO); | |
1897 | ||
1898 | /* Return through std_return after this point. */ | |
1899 | ||
1900 | udqp = gdqp = NULL; | |
1901 | if (vap->va_mask & XFS_AT_PROJID) | |
1902 | prid = (xfs_prid_t)vap->va_projid; | |
1903 | else | |
1904 | prid = (xfs_prid_t)dfltprid; | |
1905 | ||
1906 | /* | |
1907 | * Make sure that we have allocated dquot(s) on disk. | |
1908 | */ | |
1909 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
1910 | current_fsuid(credp), current_fsgid(credp), | |
1911 | XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp); | |
1912 | if (error) | |
1913 | goto std_return; | |
1914 | ||
1915 | ip = NULL; | |
1916 | dp_joined_to_trans = B_FALSE; | |
1917 | ||
1918 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
1919 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1920 | resblks = XFS_CREATE_SPACE_RES(mp, namelen); | |
1921 | /* | |
1922 | * Initially assume that the file does not exist and | |
1923 | * reserve the resources for that case. If that is not | |
1924 | * the case we'll drop the one we have and get a more | |
1925 | * appropriate transaction later. | |
1926 | */ | |
1927 | error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0, | |
1928 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1929 | if (error == ENOSPC) { | |
1930 | resblks = 0; | |
1931 | error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0, | |
1932 | XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT); | |
1933 | } | |
1934 | if (error) { | |
1935 | cancel_flags = 0; | |
1936 | dp = NULL; | |
1937 | goto error_return; | |
1938 | } | |
1939 | ||
1940 | xfs_ilock(dp, XFS_ILOCK_EXCL); | |
1941 | ||
1942 | XFS_BMAP_INIT(&free_list, &first_block); | |
1943 | ||
1944 | ASSERT(ip == NULL); | |
1945 | ||
1946 | /* | |
1947 | * Reserve disk quota and the inode. | |
1948 | */ | |
1949 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
1950 | if (error) | |
1951 | goto error_return; | |
1952 | ||
1953 | if (resblks == 0 && | |
1954 | (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen))) | |
1955 | goto error_return; | |
1956 | rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0; | |
1957 | error = xfs_dir_ialloc(&tp, dp, | |
1958 | MAKEIMODE(vap->va_type,vap->va_mode), 1, | |
1959 | rdev, credp, prid, resblks > 0, | |
1960 | &ip, &committed); | |
1961 | if (error) { | |
1962 | if (error == ENOSPC) | |
1963 | goto error_return; | |
1964 | goto abort_return; | |
1965 | } | |
1966 | ITRACE(ip); | |
1967 | ||
1968 | /* | |
1969 | * At this point, we've gotten a newly allocated inode. | |
1970 | * It is locked (and joined to the transaction). | |
1971 | */ | |
1972 | ||
1973 | ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE)); | |
1974 | ||
1975 | /* | |
1976 | * Now we join the directory inode to the transaction. | |
1977 | * We do not do it earlier because xfs_dir_ialloc | |
1978 | * might commit the previous transaction (and release | |
1979 | * all the locks). | |
1980 | */ | |
1981 | ||
1982 | VN_HOLD(dir_vp); | |
1983 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
1984 | dp_joined_to_trans = B_TRUE; | |
1985 | ||
1986 | error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino, | |
1987 | &first_block, &free_list, | |
1988 | resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1989 | if (error) { | |
1990 | ASSERT(error != ENOSPC); | |
1991 | goto abort_return; | |
1992 | } | |
1993 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1994 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1995 | ||
1996 | /* | |
1997 | * If this is a synchronous mount, make sure that the | |
1998 | * create transaction goes to disk before returning to | |
1999 | * the user. | |
2000 | */ | |
2001 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2002 | xfs_trans_set_sync(tp); | |
2003 | } | |
2004 | ||
2005 | dp->i_gen++; | |
2006 | ||
2007 | /* | |
2008 | * Attach the dquot(s) to the inodes and modify them incore. | |
2009 | * These ids of the inode couldn't have changed since the new | |
2010 | * inode has been locked ever since it was created. | |
2011 | */ | |
2012 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
2013 | ||
2014 | /* | |
2015 | * xfs_trans_commit normally decrements the vnode ref count | |
2016 | * when it unlocks the inode. Since we want to return the | |
2017 | * vnode to the caller, we bump the vnode ref count now. | |
2018 | */ | |
2019 | IHOLD(ip); | |
2020 | vp = XFS_ITOV(ip); | |
2021 | ||
2022 | error = xfs_bmap_finish(&tp, &free_list, first_block, &committed); | |
2023 | if (error) { | |
2024 | xfs_bmap_cancel(&free_list); | |
2025 | goto abort_rele; | |
2026 | } | |
2027 | ||
2028 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
2029 | if (error) { | |
2030 | IRELE(ip); | |
2031 | tp = NULL; | |
2032 | goto error_return; | |
2033 | } | |
2034 | ||
2035 | XFS_QM_DQRELE(mp, udqp); | |
2036 | XFS_QM_DQRELE(mp, gdqp); | |
2037 | ||
2038 | /* | |
2039 | * Propogate the fact that the vnode changed after the | |
2040 | * xfs_inode locks have been released. | |
2041 | */ | |
2042 | VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3); | |
2043 | ||
2044 | *vpp = vp; | |
2045 | ||
2046 | /* Fallthrough to std_return with error = 0 */ | |
2047 | ||
2048 | std_return: | |
2049 | if ( (*vpp || (error != 0 && dm_event_sent != 0)) && | |
2050 | DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp), | |
2051 | DM_EVENT_POSTCREATE)) { | |
2052 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, | |
2053 | dir_vp, DM_RIGHT_NULL, | |
2054 | *vpp ? vp:NULL, | |
2055 | DM_RIGHT_NULL, name, NULL, | |
2056 | dm_di_mode, error, 0); | |
2057 | } | |
2058 | return error; | |
2059 | ||
2060 | abort_return: | |
2061 | cancel_flags |= XFS_TRANS_ABORT; | |
2062 | /* FALLTHROUGH */ | |
2063 | error_return: | |
2064 | ||
2065 | if (tp != NULL) | |
2066 | xfs_trans_cancel(tp, cancel_flags); | |
2067 | ||
2068 | if (!dp_joined_to_trans && (dp != NULL)) | |
2069 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2070 | XFS_QM_DQRELE(mp, udqp); | |
2071 | XFS_QM_DQRELE(mp, gdqp); | |
2072 | ||
2073 | goto std_return; | |
2074 | ||
2075 | abort_rele: | |
2076 | /* | |
2077 | * Wait until after the current transaction is aborted to | |
2078 | * release the inode. This prevents recursive transactions | |
2079 | * and deadlocks from xfs_inactive. | |
2080 | */ | |
2081 | cancel_flags |= XFS_TRANS_ABORT; | |
2082 | xfs_trans_cancel(tp, cancel_flags); | |
2083 | IRELE(ip); | |
2084 | ||
2085 | XFS_QM_DQRELE(mp, udqp); | |
2086 | XFS_QM_DQRELE(mp, gdqp); | |
2087 | ||
2088 | goto std_return; | |
2089 | } | |
2090 | ||
2091 | #ifdef DEBUG | |
2092 | /* | |
2093 | * Some counters to see if (and how often) we are hitting some deadlock | |
2094 | * prevention code paths. | |
2095 | */ | |
2096 | ||
2097 | int xfs_rm_locks; | |
2098 | int xfs_rm_lock_delays; | |
2099 | int xfs_rm_attempts; | |
2100 | #endif | |
2101 | ||
2102 | /* | |
2103 | * The following routine will lock the inodes associated with the | |
2104 | * directory and the named entry in the directory. The locks are | |
2105 | * acquired in increasing inode number. | |
2106 | * | |
2107 | * If the entry is "..", then only the directory is locked. The | |
2108 | * vnode ref count will still include that from the .. entry in | |
2109 | * this case. | |
2110 | * | |
2111 | * There is a deadlock we need to worry about. If the locked directory is | |
2112 | * in the AIL, it might be blocking up the log. The next inode we lock | |
2113 | * could be already locked by another thread waiting for log space (e.g | |
2114 | * a permanent log reservation with a long running transaction (see | |
2115 | * xfs_itruncate_finish)). To solve this, we must check if the directory | |
2116 | * is in the ail and use lock_nowait. If we can't lock, we need to | |
2117 | * drop the inode lock on the directory and try again. xfs_iunlock will | |
2118 | * potentially push the tail if we were holding up the log. | |
2119 | */ | |
2120 | STATIC int | |
2121 | xfs_lock_dir_and_entry( | |
2122 | xfs_inode_t *dp, | |
2123 | vname_t *dentry, | |
2124 | xfs_inode_t *ip) /* inode of entry 'name' */ | |
2125 | { | |
2126 | int attempts; | |
2127 | xfs_ino_t e_inum; | |
2128 | xfs_inode_t *ips[2]; | |
2129 | xfs_log_item_t *lp; | |
2130 | ||
2131 | #ifdef DEBUG | |
2132 | xfs_rm_locks++; | |
2133 | #endif | |
2134 | attempts = 0; | |
2135 | ||
2136 | again: | |
2137 | xfs_ilock(dp, XFS_ILOCK_EXCL); | |
2138 | ||
2139 | e_inum = ip->i_ino; | |
2140 | ||
2141 | ITRACE(ip); | |
2142 | ||
2143 | /* | |
2144 | * We want to lock in increasing inum. Since we've already | |
2145 | * acquired the lock on the directory, we may need to release | |
2146 | * if if the inum of the entry turns out to be less. | |
2147 | */ | |
2148 | if (e_inum > dp->i_ino) { | |
2149 | /* | |
2150 | * We are already in the right order, so just | |
2151 | * lock on the inode of the entry. | |
2152 | * We need to use nowait if dp is in the AIL. | |
2153 | */ | |
2154 | ||
2155 | lp = (xfs_log_item_t *)dp->i_itemp; | |
2156 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2157 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2158 | attempts++; | |
2159 | #ifdef DEBUG | |
2160 | xfs_rm_attempts++; | |
2161 | #endif | |
2162 | ||
2163 | /* | |
2164 | * Unlock dp and try again. | |
2165 | * xfs_iunlock will try to push the tail | |
2166 | * if the inode is in the AIL. | |
2167 | */ | |
2168 | ||
2169 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2170 | ||
2171 | if ((attempts % 5) == 0) { | |
2172 | delay(1); /* Don't just spin the CPU */ | |
2173 | #ifdef DEBUG | |
2174 | xfs_rm_lock_delays++; | |
2175 | #endif | |
2176 | } | |
2177 | goto again; | |
2178 | } | |
2179 | } else { | |
2180 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
2181 | } | |
2182 | } else if (e_inum < dp->i_ino) { | |
2183 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2184 | ||
2185 | ips[0] = ip; | |
2186 | ips[1] = dp; | |
2187 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2188 | } | |
2189 | /* else e_inum == dp->i_ino */ | |
2190 | /* This can happen if we're asked to lock /x/.. | |
2191 | * the entry is "..", which is also the parent directory. | |
2192 | */ | |
2193 | ||
2194 | return 0; | |
2195 | } | |
2196 | ||
2197 | #ifdef DEBUG | |
2198 | int xfs_locked_n; | |
2199 | int xfs_small_retries; | |
2200 | int xfs_middle_retries; | |
2201 | int xfs_lots_retries; | |
2202 | int xfs_lock_delays; | |
2203 | #endif | |
2204 | ||
2205 | /* | |
2206 | * The following routine will lock n inodes in exclusive mode. | |
2207 | * We assume the caller calls us with the inodes in i_ino order. | |
2208 | * | |
2209 | * We need to detect deadlock where an inode that we lock | |
2210 | * is in the AIL and we start waiting for another inode that is locked | |
2211 | * by a thread in a long running transaction (such as truncate). This can | |
2212 | * result in deadlock since the long running trans might need to wait | |
2213 | * for the inode we just locked in order to push the tail and free space | |
2214 | * in the log. | |
2215 | */ | |
2216 | void | |
2217 | xfs_lock_inodes( | |
2218 | xfs_inode_t **ips, | |
2219 | int inodes, | |
2220 | int first_locked, | |
2221 | uint lock_mode) | |
2222 | { | |
2223 | int attempts = 0, i, j, try_lock; | |
2224 | xfs_log_item_t *lp; | |
2225 | ||
2226 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
2227 | ||
2228 | if (first_locked) { | |
2229 | try_lock = 1; | |
2230 | i = 1; | |
2231 | } else { | |
2232 | try_lock = 0; | |
2233 | i = 0; | |
2234 | } | |
2235 | ||
2236 | again: | |
2237 | for (; i < inodes; i++) { | |
2238 | ASSERT(ips[i]); | |
2239 | ||
2240 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
2241 | continue; | |
2242 | ||
2243 | /* | |
2244 | * If try_lock is not set yet, make sure all locked inodes | |
2245 | * are not in the AIL. | |
2246 | * If any are, set try_lock to be used later. | |
2247 | */ | |
2248 | ||
2249 | if (!try_lock) { | |
2250 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
2251 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
2252 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
2253 | try_lock++; | |
2254 | } | |
2255 | } | |
2256 | } | |
2257 | ||
2258 | /* | |
2259 | * If any of the previous locks we have locked is in the AIL, | |
2260 | * we must TRY to get the second and subsequent locks. If | |
2261 | * we can't get any, we must release all we have | |
2262 | * and try again. | |
2263 | */ | |
2264 | ||
2265 | if (try_lock) { | |
2266 | /* try_lock must be 0 if i is 0. */ | |
2267 | /* | |
2268 | * try_lock means we have an inode locked | |
2269 | * that is in the AIL. | |
2270 | */ | |
2271 | ASSERT(i != 0); | |
2272 | if (!xfs_ilock_nowait(ips[i], lock_mode)) { | |
2273 | attempts++; | |
2274 | ||
2275 | /* | |
2276 | * Unlock all previous guys and try again. | |
2277 | * xfs_iunlock will try to push the tail | |
2278 | * if the inode is in the AIL. | |
2279 | */ | |
2280 | ||
2281 | for(j = i - 1; j >= 0; j--) { | |
2282 | ||
2283 | /* | |
2284 | * Check to see if we've already | |
2285 | * unlocked this one. | |
2286 | * Not the first one going back, | |
2287 | * and the inode ptr is the same. | |
2288 | */ | |
2289 | if ((j != (i - 1)) && ips[j] == | |
2290 | ips[j+1]) | |
2291 | continue; | |
2292 | ||
2293 | xfs_iunlock(ips[j], lock_mode); | |
2294 | } | |
2295 | ||
2296 | if ((attempts % 5) == 0) { | |
2297 | delay(1); /* Don't just spin the CPU */ | |
2298 | #ifdef DEBUG | |
2299 | xfs_lock_delays++; | |
2300 | #endif | |
2301 | } | |
2302 | i = 0; | |
2303 | try_lock = 0; | |
2304 | goto again; | |
2305 | } | |
2306 | } else { | |
2307 | xfs_ilock(ips[i], lock_mode); | |
2308 | } | |
2309 | } | |
2310 | ||
2311 | #ifdef DEBUG | |
2312 | if (attempts) { | |
2313 | if (attempts < 5) xfs_small_retries++; | |
2314 | else if (attempts < 100) xfs_middle_retries++; | |
2315 | else xfs_lots_retries++; | |
2316 | } else { | |
2317 | xfs_locked_n++; | |
2318 | } | |
2319 | #endif | |
2320 | } | |
2321 | ||
2322 | #ifdef DEBUG | |
2323 | #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);} | |
2324 | int remove_which_error_return = 0; | |
2325 | #else /* ! DEBUG */ | |
2326 | #define REMOVE_DEBUG_TRACE(x) | |
2327 | #endif /* ! DEBUG */ | |
2328 | ||
2329 | ||
2330 | /* | |
2331 | * xfs_remove | |
2332 | * | |
2333 | */ | |
2334 | STATIC int | |
2335 | xfs_remove( | |
2336 | bhv_desc_t *dir_bdp, | |
2337 | vname_t *dentry, | |
2338 | cred_t *credp) | |
2339 | { | |
2340 | vnode_t *dir_vp; | |
2341 | char *name = VNAME(dentry); | |
2342 | xfs_inode_t *dp, *ip; | |
2343 | xfs_trans_t *tp = NULL; | |
2344 | xfs_mount_t *mp; | |
2345 | int error = 0; | |
2346 | xfs_bmap_free_t free_list; | |
2347 | xfs_fsblock_t first_block; | |
2348 | int cancel_flags; | |
2349 | int committed; | |
2350 | int dm_di_mode = 0; | |
2351 | int link_zero; | |
2352 | uint resblks; | |
2353 | int namelen; | |
2354 | ||
2355 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
2356 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
2357 | ||
2358 | dp = XFS_BHVTOI(dir_bdp); | |
2359 | mp = dp->i_mount; | |
2360 | ||
2361 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2362 | return XFS_ERROR(EIO); | |
2363 | ||
2364 | namelen = VNAMELEN(dentry); | |
2365 | ||
2366 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) { | |
2367 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp, | |
2368 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
2369 | name, NULL, 0, 0, 0); | |
2370 | if (error) | |
2371 | return error; | |
2372 | } | |
2373 | ||
2374 | /* From this point on, return through std_return */ | |
2375 | ip = NULL; | |
2376 | ||
2377 | /* | |
2378 | * We need to get a reference to ip before we get our log | |
2379 | * reservation. The reason for this is that we cannot call | |
2380 | * xfs_iget for an inode for which we do not have a reference | |
2381 | * once we've acquired a log reservation. This is because the | |
2382 | * inode we are trying to get might be in xfs_inactive going | |
2383 | * for a log reservation. Since we'll have to wait for the | |
2384 | * inactive code to complete before returning from xfs_iget, | |
2385 | * we need to make sure that we don't have log space reserved | |
2386 | * when we call xfs_iget. Instead we get an unlocked referece | |
2387 | * to the inode before getting our log reservation. | |
2388 | */ | |
2389 | error = xfs_get_dir_entry(dentry, &ip); | |
2390 | if (error) { | |
2391 | REMOVE_DEBUG_TRACE(__LINE__); | |
2392 | goto std_return; | |
2393 | } | |
2394 | ||
2395 | dm_di_mode = ip->i_d.di_mode; | |
2396 | ||
2397 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
2398 | ||
2399 | ITRACE(ip); | |
2400 | ||
2401 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
2402 | if (!error && dp != ip) | |
2403 | error = XFS_QM_DQATTACH(mp, ip, 0); | |
2404 | if (error) { | |
2405 | REMOVE_DEBUG_TRACE(__LINE__); | |
2406 | IRELE(ip); | |
2407 | goto std_return; | |
2408 | } | |
2409 | ||
2410 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2411 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2412 | /* | |
2413 | * We try to get the real space reservation first, | |
2414 | * allowing for directory btree deletion(s) implying | |
2415 | * possible bmap insert(s). If we can't get the space | |
2416 | * reservation then we use 0 instead, and avoid the bmap | |
2417 | * btree insert(s) in the directory code by, if the bmap | |
2418 | * insert tries to happen, instead trimming the LAST | |
2419 | * block from the directory. | |
2420 | */ | |
2421 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2422 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2423 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2424 | if (error == ENOSPC) { | |
2425 | resblks = 0; | |
2426 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2427 | XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT); | |
2428 | } | |
2429 | if (error) { | |
2430 | ASSERT(error != ENOSPC); | |
2431 | REMOVE_DEBUG_TRACE(__LINE__); | |
2432 | xfs_trans_cancel(tp, 0); | |
2433 | IRELE(ip); | |
2434 | return error; | |
2435 | } | |
2436 | ||
2437 | error = xfs_lock_dir_and_entry(dp, dentry, ip); | |
2438 | if (error) { | |
2439 | REMOVE_DEBUG_TRACE(__LINE__); | |
2440 | xfs_trans_cancel(tp, cancel_flags); | |
2441 | IRELE(ip); | |
2442 | goto std_return; | |
2443 | } | |
2444 | ||
2445 | /* | |
2446 | * At this point, we've gotten both the directory and the entry | |
2447 | * inodes locked. | |
2448 | */ | |
2449 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2450 | if (dp != ip) { | |
2451 | /* | |
2452 | * Increment vnode ref count only in this case since | |
2453 | * there's an extra vnode reference in the case where | |
2454 | * dp == ip. | |
2455 | */ | |
2456 | IHOLD(dp); | |
2457 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * Entry must exist since we did a lookup in xfs_lock_dir_and_entry. | |
2462 | */ | |
2463 | XFS_BMAP_INIT(&free_list, &first_block); | |
2464 | error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino, | |
2465 | &first_block, &free_list, 0); | |
2466 | if (error) { | |
2467 | ASSERT(error != ENOENT); | |
2468 | REMOVE_DEBUG_TRACE(__LINE__); | |
2469 | goto error1; | |
2470 | } | |
2471 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2472 | ||
2473 | dp->i_gen++; | |
2474 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2475 | ||
2476 | error = xfs_droplink(tp, ip); | |
2477 | if (error) { | |
2478 | REMOVE_DEBUG_TRACE(__LINE__); | |
2479 | goto error1; | |
2480 | } | |
2481 | ||
2482 | /* Determine if this is the last link while | |
2483 | * we are in the transaction. | |
2484 | */ | |
2485 | link_zero = (ip)->i_d.di_nlink==0; | |
2486 | ||
2487 | /* | |
2488 | * Take an extra ref on the inode so that it doesn't | |
2489 | * go to xfs_inactive() from within the commit. | |
2490 | */ | |
2491 | IHOLD(ip); | |
2492 | ||
2493 | /* | |
2494 | * If this is a synchronous mount, make sure that the | |
2495 | * remove transaction goes to disk before returning to | |
2496 | * the user. | |
2497 | */ | |
2498 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2499 | xfs_trans_set_sync(tp); | |
2500 | } | |
2501 | ||
2502 | error = xfs_bmap_finish(&tp, &free_list, first_block, &committed); | |
2503 | if (error) { | |
2504 | REMOVE_DEBUG_TRACE(__LINE__); | |
2505 | goto error_rele; | |
2506 | } | |
2507 | ||
2508 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
2509 | if (error) { | |
2510 | IRELE(ip); | |
2511 | goto std_return; | |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | * Before we drop our extra reference to the inode, purge it | |
2516 | * from the refcache if it is there. By waiting until afterwards | |
2517 | * to do the IRELE, we ensure that we won't go inactive in the | |
2518 | * xfs_refcache_purge_ip routine (although that would be OK). | |
2519 | */ | |
2520 | xfs_refcache_purge_ip(ip); | |
2521 | ||
2522 | vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
2523 | ||
2524 | /* | |
2525 | * Let interposed file systems know about removed links. | |
2526 | */ | |
2527 | VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero); | |
2528 | ||
2529 | IRELE(ip); | |
2530 | ||
2531 | /* Fall through to std_return with error = 0 */ | |
2532 | std_return: | |
2533 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, | |
2534 | DM_EVENT_POSTREMOVE)) { | |
2535 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, | |
2536 | dir_vp, DM_RIGHT_NULL, | |
2537 | NULL, DM_RIGHT_NULL, | |
2538 | name, NULL, dm_di_mode, error, 0); | |
2539 | } | |
2540 | return error; | |
2541 | ||
2542 | error1: | |
2543 | xfs_bmap_cancel(&free_list); | |
2544 | cancel_flags |= XFS_TRANS_ABORT; | |
2545 | xfs_trans_cancel(tp, cancel_flags); | |
2546 | goto std_return; | |
2547 | ||
2548 | error_rele: | |
2549 | /* | |
2550 | * In this case make sure to not release the inode until after | |
2551 | * the current transaction is aborted. Releasing it beforehand | |
2552 | * can cause us to go to xfs_inactive and start a recursive | |
2553 | * transaction which can easily deadlock with the current one. | |
2554 | */ | |
2555 | xfs_bmap_cancel(&free_list); | |
2556 | cancel_flags |= XFS_TRANS_ABORT; | |
2557 | xfs_trans_cancel(tp, cancel_flags); | |
2558 | ||
2559 | /* | |
2560 | * Before we drop our extra reference to the inode, purge it | |
2561 | * from the refcache if it is there. By waiting until afterwards | |
2562 | * to do the IRELE, we ensure that we won't go inactive in the | |
2563 | * xfs_refcache_purge_ip routine (although that would be OK). | |
2564 | */ | |
2565 | xfs_refcache_purge_ip(ip); | |
2566 | ||
2567 | IRELE(ip); | |
2568 | ||
2569 | goto std_return; | |
2570 | } | |
2571 | ||
2572 | ||
2573 | /* | |
2574 | * xfs_link | |
2575 | * | |
2576 | */ | |
2577 | STATIC int | |
2578 | xfs_link( | |
2579 | bhv_desc_t *target_dir_bdp, | |
2580 | vnode_t *src_vp, | |
2581 | vname_t *dentry, | |
2582 | cred_t *credp) | |
2583 | { | |
2584 | xfs_inode_t *tdp, *sip; | |
2585 | xfs_trans_t *tp; | |
2586 | xfs_mount_t *mp; | |
2587 | xfs_inode_t *ips[2]; | |
2588 | int error; | |
2589 | xfs_bmap_free_t free_list; | |
2590 | xfs_fsblock_t first_block; | |
2591 | int cancel_flags; | |
2592 | int committed; | |
2593 | vnode_t *target_dir_vp; | |
2594 | bhv_desc_t *src_bdp; | |
2595 | int resblks; | |
2596 | char *target_name = VNAME(dentry); | |
2597 | int target_namelen; | |
2598 | ||
2599 | target_dir_vp = BHV_TO_VNODE(target_dir_bdp); | |
2600 | vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
2601 | vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address); | |
2602 | ||
2603 | target_namelen = VNAMELEN(dentry); | |
2604 | if (src_vp->v_type == VDIR) | |
2605 | return XFS_ERROR(EPERM); | |
2606 | ||
2607 | /* | |
2608 | * For now, manually find the XFS behavior descriptor for | |
2609 | * the source vnode. If it doesn't exist then something | |
2610 | * is wrong and we should just return an error. | |
2611 | * Eventually we need to figure out how link is going to | |
2612 | * work in the face of stacked vnodes. | |
2613 | */ | |
2614 | src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops); | |
2615 | if (src_bdp == NULL) { | |
2616 | return XFS_ERROR(EXDEV); | |
2617 | } | |
2618 | sip = XFS_BHVTOI(src_bdp); | |
2619 | tdp = XFS_BHVTOI(target_dir_bdp); | |
2620 | mp = tdp->i_mount; | |
2621 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2622 | return XFS_ERROR(EIO); | |
2623 | ||
2624 | if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) { | |
2625 | error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK, | |
2626 | target_dir_vp, DM_RIGHT_NULL, | |
2627 | src_vp, DM_RIGHT_NULL, | |
2628 | target_name, NULL, 0, 0, 0); | |
2629 | if (error) | |
2630 | return error; | |
2631 | } | |
2632 | ||
2633 | /* Return through std_return after this point. */ | |
2634 | ||
2635 | error = XFS_QM_DQATTACH(mp, sip, 0); | |
2636 | if (!error && sip != tdp) | |
2637 | error = XFS_QM_DQATTACH(mp, tdp, 0); | |
2638 | if (error) | |
2639 | goto std_return; | |
2640 | ||
2641 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
2642 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2643 | resblks = XFS_LINK_SPACE_RES(mp, target_namelen); | |
2644 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, | |
2645 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2646 | if (error == ENOSPC) { | |
2647 | resblks = 0; | |
2648 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
2649 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
2650 | } | |
2651 | if (error) { | |
2652 | cancel_flags = 0; | |
2653 | goto error_return; | |
2654 | } | |
2655 | ||
2656 | if (sip->i_ino < tdp->i_ino) { | |
2657 | ips[0] = sip; | |
2658 | ips[1] = tdp; | |
2659 | } else { | |
2660 | ips[0] = tdp; | |
2661 | ips[1] = sip; | |
2662 | } | |
2663 | ||
2664 | xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL); | |
2665 | ||
2666 | /* | |
2667 | * Increment vnode ref counts since xfs_trans_commit & | |
2668 | * xfs_trans_cancel will both unlock the inodes and | |
2669 | * decrement the associated ref counts. | |
2670 | */ | |
2671 | VN_HOLD(src_vp); | |
2672 | VN_HOLD(target_dir_vp); | |
2673 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
2674 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
2675 | ||
2676 | /* | |
2677 | * If the source has too many links, we can't make any more to it. | |
2678 | */ | |
2679 | if (sip->i_d.di_nlink >= XFS_MAXLINK) { | |
2680 | error = XFS_ERROR(EMLINK); | |
2681 | goto error_return; | |
2682 | } | |
2683 | ||
2684 | if (resblks == 0 && | |
2685 | (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name, | |
2686 | target_namelen))) | |
2687 | goto error_return; | |
2688 | ||
2689 | XFS_BMAP_INIT(&free_list, &first_block); | |
2690 | ||
2691 | error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen, | |
2692 | sip->i_ino, &first_block, &free_list, | |
2693 | resblks); | |
2694 | if (error) | |
2695 | goto abort_return; | |
2696 | xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2697 | tdp->i_gen++; | |
2698 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
2699 | ||
2700 | error = xfs_bumplink(tp, sip); | |
2701 | if (error) { | |
2702 | goto abort_return; | |
2703 | } | |
2704 | ||
2705 | /* | |
2706 | * If this is a synchronous mount, make sure that the | |
2707 | * link transaction goes to disk before returning to | |
2708 | * the user. | |
2709 | */ | |
2710 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2711 | xfs_trans_set_sync(tp); | |
2712 | } | |
2713 | ||
2714 | error = xfs_bmap_finish (&tp, &free_list, first_block, &committed); | |
2715 | if (error) { | |
2716 | xfs_bmap_cancel(&free_list); | |
2717 | goto abort_return; | |
2718 | } | |
2719 | ||
2720 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
2721 | if (error) { | |
2722 | goto std_return; | |
2723 | } | |
2724 | ||
2725 | /* Fall through to std_return with error = 0. */ | |
2726 | std_return: | |
2727 | if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip, | |
2728 | DM_EVENT_POSTLINK)) { | |
2729 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK, | |
2730 | target_dir_vp, DM_RIGHT_NULL, | |
2731 | src_vp, DM_RIGHT_NULL, | |
2732 | target_name, NULL, 0, error, 0); | |
2733 | } | |
2734 | return error; | |
2735 | ||
2736 | abort_return: | |
2737 | cancel_flags |= XFS_TRANS_ABORT; | |
2738 | /* FALLTHROUGH */ | |
2739 | error_return: | |
2740 | xfs_trans_cancel(tp, cancel_flags); | |
2741 | ||
2742 | goto std_return; | |
2743 | } | |
2744 | /* | |
2745 | * xfs_mkdir | |
2746 | * | |
2747 | */ | |
2748 | STATIC int | |
2749 | xfs_mkdir( | |
2750 | bhv_desc_t *dir_bdp, | |
2751 | vname_t *dentry, | |
2752 | vattr_t *vap, | |
2753 | vnode_t **vpp, | |
2754 | cred_t *credp) | |
2755 | { | |
2756 | char *dir_name = VNAME(dentry); | |
2757 | xfs_inode_t *dp; | |
2758 | xfs_inode_t *cdp; /* inode of created dir */ | |
2759 | vnode_t *cvp; /* vnode of created dir */ | |
2760 | xfs_trans_t *tp; | |
2761 | xfs_mount_t *mp; | |
2762 | int cancel_flags; | |
2763 | int error; | |
2764 | int committed; | |
2765 | xfs_bmap_free_t free_list; | |
2766 | xfs_fsblock_t first_block; | |
2767 | vnode_t *dir_vp; | |
2768 | boolean_t dp_joined_to_trans; | |
2769 | boolean_t created = B_FALSE; | |
2770 | int dm_event_sent = 0; | |
2771 | xfs_prid_t prid; | |
2772 | struct xfs_dquot *udqp, *gdqp; | |
2773 | uint resblks; | |
2774 | int dm_di_mode; | |
2775 | int dir_namelen; | |
2776 | ||
2777 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
2778 | dp = XFS_BHVTOI(dir_bdp); | |
2779 | mp = dp->i_mount; | |
2780 | ||
2781 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2782 | return XFS_ERROR(EIO); | |
2783 | ||
2784 | dir_namelen = VNAMELEN(dentry); | |
2785 | ||
2786 | tp = NULL; | |
2787 | dp_joined_to_trans = B_FALSE; | |
2788 | dm_di_mode = vap->va_mode|VTTOIF(vap->va_type); | |
2789 | ||
2790 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) { | |
2791 | error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE, | |
2792 | dir_vp, DM_RIGHT_NULL, NULL, | |
2793 | DM_RIGHT_NULL, dir_name, NULL, | |
2794 | dm_di_mode, 0, 0); | |
2795 | if (error) | |
2796 | return error; | |
2797 | dm_event_sent = 1; | |
2798 | } | |
2799 | ||
2800 | /* Return through std_return after this point. */ | |
2801 | ||
2802 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
2803 | ||
2804 | mp = dp->i_mount; | |
2805 | udqp = gdqp = NULL; | |
2806 | if (vap->va_mask & XFS_AT_PROJID) | |
2807 | prid = (xfs_prid_t)vap->va_projid; | |
2808 | else | |
2809 | prid = (xfs_prid_t)dfltprid; | |
2810 | ||
2811 | /* | |
2812 | * Make sure that we have allocated dquot(s) on disk. | |
2813 | */ | |
2814 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
2815 | current_fsuid(credp), current_fsgid(credp), | |
2816 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); | |
2817 | if (error) | |
2818 | goto std_return; | |
2819 | ||
2820 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
2821 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2822 | resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen); | |
2823 | error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0, | |
2824 | XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT); | |
2825 | if (error == ENOSPC) { | |
2826 | resblks = 0; | |
2827 | error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0, | |
2828 | XFS_TRANS_PERM_LOG_RES, | |
2829 | XFS_MKDIR_LOG_COUNT); | |
2830 | } | |
2831 | if (error) { | |
2832 | cancel_flags = 0; | |
2833 | dp = NULL; | |
2834 | goto error_return; | |
2835 | } | |
2836 | ||
2837 | xfs_ilock(dp, XFS_ILOCK_EXCL); | |
2838 | ||
2839 | /* | |
2840 | * Check for directory link count overflow. | |
2841 | */ | |
2842 | if (dp->i_d.di_nlink >= XFS_MAXLINK) { | |
2843 | error = XFS_ERROR(EMLINK); | |
2844 | goto error_return; | |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * Reserve disk quota and the inode. | |
2849 | */ | |
2850 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
2851 | if (error) | |
2852 | goto error_return; | |
2853 | ||
2854 | if (resblks == 0 && | |
2855 | (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen))) | |
2856 | goto error_return; | |
2857 | /* | |
2858 | * create the directory inode. | |
2859 | */ | |
2860 | error = xfs_dir_ialloc(&tp, dp, | |
2861 | MAKEIMODE(vap->va_type,vap->va_mode), 2, | |
2862 | 0, credp, prid, resblks > 0, | |
2863 | &cdp, NULL); | |
2864 | if (error) { | |
2865 | if (error == ENOSPC) | |
2866 | goto error_return; | |
2867 | goto abort_return; | |
2868 | } | |
2869 | ITRACE(cdp); | |
2870 | ||
2871 | /* | |
2872 | * Now we add the directory inode to the transaction. | |
2873 | * We waited until now since xfs_dir_ialloc might start | |
2874 | * a new transaction. Had we joined the transaction | |
2875 | * earlier, the locks might have gotten released. | |
2876 | */ | |
2877 | VN_HOLD(dir_vp); | |
2878 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2879 | dp_joined_to_trans = B_TRUE; | |
2880 | ||
2881 | XFS_BMAP_INIT(&free_list, &first_block); | |
2882 | ||
2883 | error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen, | |
2884 | cdp->i_ino, &first_block, &free_list, | |
2885 | resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
2886 | if (error) { | |
2887 | ASSERT(error != ENOSPC); | |
2888 | goto error1; | |
2889 | } | |
2890 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2891 | ||
2892 | /* | |
2893 | * Bump the in memory version number of the parent directory | |
2894 | * so that other processes accessing it will recognize that | |
2895 | * the directory has changed. | |
2896 | */ | |
2897 | dp->i_gen++; | |
2898 | ||
2899 | error = XFS_DIR_INIT(mp, tp, cdp, dp); | |
2900 | if (error) { | |
2901 | goto error2; | |
2902 | } | |
2903 | ||
2904 | cdp->i_gen = 1; | |
2905 | error = xfs_bumplink(tp, dp); | |
2906 | if (error) { | |
2907 | goto error2; | |
2908 | } | |
2909 | ||
2910 | cvp = XFS_ITOV(cdp); | |
2911 | ||
2912 | created = B_TRUE; | |
2913 | ||
2914 | *vpp = cvp; | |
2915 | IHOLD(cdp); | |
2916 | ||
2917 | /* | |
2918 | * Attach the dquots to the new inode and modify the icount incore. | |
2919 | */ | |
2920 | XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp); | |
2921 | ||
2922 | /* | |
2923 | * If this is a synchronous mount, make sure that the | |
2924 | * mkdir transaction goes to disk before returning to | |
2925 | * the user. | |
2926 | */ | |
2927 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2928 | xfs_trans_set_sync(tp); | |
2929 | } | |
2930 | ||
2931 | error = xfs_bmap_finish(&tp, &free_list, first_block, &committed); | |
2932 | if (error) { | |
2933 | IRELE(cdp); | |
2934 | goto error2; | |
2935 | } | |
2936 | ||
2937 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
2938 | XFS_QM_DQRELE(mp, udqp); | |
2939 | XFS_QM_DQRELE(mp, gdqp); | |
2940 | if (error) { | |
2941 | IRELE(cdp); | |
2942 | } | |
2943 | ||
2944 | /* Fall through to std_return with error = 0 or errno from | |
2945 | * xfs_trans_commit. */ | |
2946 | ||
2947 | std_return: | |
2948 | if ( (created || (error != 0 && dm_event_sent != 0)) && | |
2949 | DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp), | |
2950 | DM_EVENT_POSTCREATE)) { | |
2951 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE, | |
2952 | dir_vp, DM_RIGHT_NULL, | |
2953 | created ? XFS_ITOV(cdp):NULL, | |
2954 | DM_RIGHT_NULL, | |
2955 | dir_name, NULL, | |
2956 | dm_di_mode, error, 0); | |
2957 | } | |
2958 | return error; | |
2959 | ||
2960 | error2: | |
2961 | error1: | |
2962 | xfs_bmap_cancel(&free_list); | |
2963 | abort_return: | |
2964 | cancel_flags |= XFS_TRANS_ABORT; | |
2965 | error_return: | |
2966 | xfs_trans_cancel(tp, cancel_flags); | |
2967 | XFS_QM_DQRELE(mp, udqp); | |
2968 | XFS_QM_DQRELE(mp, gdqp); | |
2969 | ||
2970 | if (!dp_joined_to_trans && (dp != NULL)) { | |
2971 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
2972 | } | |
2973 | ||
2974 | goto std_return; | |
2975 | } | |
2976 | ||
2977 | ||
2978 | /* | |
2979 | * xfs_rmdir | |
2980 | * | |
2981 | */ | |
2982 | STATIC int | |
2983 | xfs_rmdir( | |
2984 | bhv_desc_t *dir_bdp, | |
2985 | vname_t *dentry, | |
2986 | cred_t *credp) | |
2987 | { | |
2988 | char *name = VNAME(dentry); | |
2989 | xfs_inode_t *dp; | |
2990 | xfs_inode_t *cdp; /* child directory */ | |
2991 | xfs_trans_t *tp; | |
2992 | xfs_mount_t *mp; | |
2993 | int error; | |
2994 | xfs_bmap_free_t free_list; | |
2995 | xfs_fsblock_t first_block; | |
2996 | int cancel_flags; | |
2997 | int committed; | |
2998 | vnode_t *dir_vp; | |
2999 | int dm_di_mode = 0; | |
3000 | int last_cdp_link; | |
3001 | int namelen; | |
3002 | uint resblks; | |
3003 | ||
3004 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
3005 | dp = XFS_BHVTOI(dir_bdp); | |
3006 | mp = dp->i_mount; | |
3007 | ||
3008 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
3009 | ||
3010 | if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount)) | |
3011 | return XFS_ERROR(EIO); | |
3012 | namelen = VNAMELEN(dentry); | |
3013 | ||
3014 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) { | |
3015 | error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, | |
3016 | dir_vp, DM_RIGHT_NULL, | |
3017 | NULL, DM_RIGHT_NULL, | |
3018 | name, NULL, 0, 0, 0); | |
3019 | if (error) | |
3020 | return XFS_ERROR(error); | |
3021 | } | |
3022 | ||
3023 | /* Return through std_return after this point. */ | |
3024 | ||
3025 | cdp = NULL; | |
3026 | ||
3027 | /* | |
3028 | * We need to get a reference to cdp before we get our log | |
3029 | * reservation. The reason for this is that we cannot call | |
3030 | * xfs_iget for an inode for which we do not have a reference | |
3031 | * once we've acquired a log reservation. This is because the | |
3032 | * inode we are trying to get might be in xfs_inactive going | |
3033 | * for a log reservation. Since we'll have to wait for the | |
3034 | * inactive code to complete before returning from xfs_iget, | |
3035 | * we need to make sure that we don't have log space reserved | |
3036 | * when we call xfs_iget. Instead we get an unlocked referece | |
3037 | * to the inode before getting our log reservation. | |
3038 | */ | |
3039 | error = xfs_get_dir_entry(dentry, &cdp); | |
3040 | if (error) { | |
3041 | REMOVE_DEBUG_TRACE(__LINE__); | |
3042 | goto std_return; | |
3043 | } | |
3044 | mp = dp->i_mount; | |
3045 | dm_di_mode = cdp->i_d.di_mode; | |
3046 | ||
3047 | /* | |
3048 | * Get the dquots for the inodes. | |
3049 | */ | |
3050 | error = XFS_QM_DQATTACH(mp, dp, 0); | |
3051 | if (!error && dp != cdp) | |
3052 | error = XFS_QM_DQATTACH(mp, cdp, 0); | |
3053 | if (error) { | |
3054 | IRELE(cdp); | |
3055 | REMOVE_DEBUG_TRACE(__LINE__); | |
3056 | goto std_return; | |
3057 | } | |
3058 | ||
3059 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
3060 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3061 | /* | |
3062 | * We try to get the real space reservation first, | |
3063 | * allowing for directory btree deletion(s) implying | |
3064 | * possible bmap insert(s). If we can't get the space | |
3065 | * reservation then we use 0 instead, and avoid the bmap | |
3066 | * btree insert(s) in the directory code by, if the bmap | |
3067 | * insert tries to happen, instead trimming the LAST | |
3068 | * block from the directory. | |
3069 | */ | |
3070 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
3071 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
3072 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
3073 | if (error == ENOSPC) { | |
3074 | resblks = 0; | |
3075 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
3076 | XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT); | |
3077 | } | |
3078 | if (error) { | |
3079 | ASSERT(error != ENOSPC); | |
3080 | cancel_flags = 0; | |
3081 | IRELE(cdp); | |
3082 | goto error_return; | |
3083 | } | |
3084 | XFS_BMAP_INIT(&free_list, &first_block); | |
3085 | ||
3086 | /* | |
3087 | * Now lock the child directory inode and the parent directory | |
3088 | * inode in the proper order. This will take care of validating | |
3089 | * that the directory entry for the child directory inode has | |
3090 | * not changed while we were obtaining a log reservation. | |
3091 | */ | |
3092 | error = xfs_lock_dir_and_entry(dp, dentry, cdp); | |
3093 | if (error) { | |
3094 | xfs_trans_cancel(tp, cancel_flags); | |
3095 | IRELE(cdp); | |
3096 | goto std_return; | |
3097 | } | |
3098 | ||
3099 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
3100 | if (dp != cdp) { | |
3101 | /* | |
3102 | * Only increment the parent directory vnode count if | |
3103 | * we didn't bump it in looking up cdp. The only time | |
3104 | * we don't bump it is when we're looking up ".". | |
3105 | */ | |
3106 | VN_HOLD(dir_vp); | |
3107 | } | |
3108 | ||
3109 | ITRACE(cdp); | |
3110 | xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL); | |
3111 | ||
3112 | ASSERT(cdp->i_d.di_nlink >= 2); | |
3113 | if (cdp->i_d.di_nlink != 2) { | |
3114 | error = XFS_ERROR(ENOTEMPTY); | |
3115 | goto error_return; | |
3116 | } | |
3117 | if (!XFS_DIR_ISEMPTY(mp, cdp)) { | |
3118 | error = XFS_ERROR(ENOTEMPTY); | |
3119 | goto error_return; | |
3120 | } | |
3121 | ||
3122 | error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino, | |
3123 | &first_block, &free_list, resblks); | |
3124 | if (error) { | |
3125 | goto error1; | |
3126 | } | |
3127 | ||
3128 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3129 | ||
3130 | /* | |
3131 | * Bump the in memory generation count on the parent | |
3132 | * directory so that other can know that it has changed. | |
3133 | */ | |
3134 | dp->i_gen++; | |
3135 | ||
3136 | /* | |
3137 | * Drop the link from cdp's "..". | |
3138 | */ | |
3139 | error = xfs_droplink(tp, dp); | |
3140 | if (error) { | |
3141 | goto error1; | |
3142 | } | |
3143 | ||
3144 | /* | |
3145 | * Drop the link from dp to cdp. | |
3146 | */ | |
3147 | error = xfs_droplink(tp, cdp); | |
3148 | if (error) { | |
3149 | goto error1; | |
3150 | } | |
3151 | ||
3152 | /* | |
3153 | * Drop the "." link from cdp to self. | |
3154 | */ | |
3155 | error = xfs_droplink(tp, cdp); | |
3156 | if (error) { | |
3157 | goto error1; | |
3158 | } | |
3159 | ||
3160 | /* Determine these before committing transaction */ | |
3161 | last_cdp_link = (cdp)->i_d.di_nlink==0; | |
3162 | ||
3163 | /* | |
3164 | * Take an extra ref on the child vnode so that it | |
3165 | * does not go to xfs_inactive() from within the commit. | |
3166 | */ | |
3167 | IHOLD(cdp); | |
3168 | ||
3169 | /* | |
3170 | * If this is a synchronous mount, make sure that the | |
3171 | * rmdir transaction goes to disk before returning to | |
3172 | * the user. | |
3173 | */ | |
3174 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3175 | xfs_trans_set_sync(tp); | |
3176 | } | |
3177 | ||
3178 | error = xfs_bmap_finish (&tp, &free_list, first_block, &committed); | |
3179 | if (error) { | |
3180 | xfs_bmap_cancel(&free_list); | |
3181 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | | |
3182 | XFS_TRANS_ABORT)); | |
3183 | IRELE(cdp); | |
3184 | goto std_return; | |
3185 | } | |
3186 | ||
3187 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
3188 | if (error) { | |
3189 | IRELE(cdp); | |
3190 | goto std_return; | |
3191 | } | |
3192 | ||
3193 | ||
3194 | /* | |
3195 | * Let interposed file systems know about removed links. | |
3196 | */ | |
3197 | VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link); | |
3198 | ||
3199 | IRELE(cdp); | |
3200 | ||
3201 | /* Fall through to std_return with error = 0 or the errno | |
3202 | * from xfs_trans_commit. */ | |
3203 | std_return: | |
3204 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) { | |
3205 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE, | |
3206 | dir_vp, DM_RIGHT_NULL, | |
3207 | NULL, DM_RIGHT_NULL, | |
3208 | name, NULL, dm_di_mode, | |
3209 | error, 0); | |
3210 | } | |
3211 | return error; | |
3212 | ||
3213 | error1: | |
3214 | xfs_bmap_cancel(&free_list); | |
3215 | cancel_flags |= XFS_TRANS_ABORT; | |
3216 | error_return: | |
3217 | xfs_trans_cancel(tp, cancel_flags); | |
3218 | goto std_return; | |
3219 | } | |
3220 | ||
3221 | ||
3222 | /* | |
3223 | * xfs_readdir | |
3224 | * | |
3225 | * Read dp's entries starting at uiop->uio_offset and translate them into | |
3226 | * bufsize bytes worth of struct dirents starting at bufbase. | |
3227 | */ | |
3228 | STATIC int | |
3229 | xfs_readdir( | |
3230 | bhv_desc_t *dir_bdp, | |
3231 | uio_t *uiop, | |
3232 | cred_t *credp, | |
3233 | int *eofp) | |
3234 | { | |
3235 | xfs_inode_t *dp; | |
3236 | xfs_trans_t *tp = NULL; | |
3237 | int error = 0; | |
3238 | uint lock_mode; | |
3239 | xfs_off_t start_offset; | |
3240 | ||
3241 | vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__, | |
3242 | (inst_t *)__return_address); | |
3243 | dp = XFS_BHVTOI(dir_bdp); | |
3244 | ||
3245 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) { | |
3246 | return XFS_ERROR(EIO); | |
3247 | } | |
3248 | ||
3249 | lock_mode = xfs_ilock_map_shared(dp); | |
3250 | start_offset = uiop->uio_offset; | |
3251 | error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp); | |
3252 | if (start_offset != uiop->uio_offset) { | |
3253 | xfs_ichgtime(dp, XFS_ICHGTIME_ACC); | |
3254 | } | |
3255 | xfs_iunlock_map_shared(dp, lock_mode); | |
3256 | return error; | |
3257 | } | |
3258 | ||
3259 | ||
3260 | /* | |
3261 | * xfs_symlink | |
3262 | * | |
3263 | */ | |
3264 | STATIC int | |
3265 | xfs_symlink( | |
3266 | bhv_desc_t *dir_bdp, | |
3267 | vname_t *dentry, | |
3268 | vattr_t *vap, | |
3269 | char *target_path, | |
3270 | vnode_t **vpp, | |
3271 | cred_t *credp) | |
3272 | { | |
3273 | xfs_trans_t *tp; | |
3274 | xfs_mount_t *mp; | |
3275 | xfs_inode_t *dp; | |
3276 | xfs_inode_t *ip; | |
3277 | int error; | |
3278 | int pathlen; | |
3279 | xfs_bmap_free_t free_list; | |
3280 | xfs_fsblock_t first_block; | |
3281 | boolean_t dp_joined_to_trans; | |
3282 | vnode_t *dir_vp; | |
3283 | uint cancel_flags; | |
3284 | int committed; | |
3285 | xfs_fileoff_t first_fsb; | |
3286 | xfs_filblks_t fs_blocks; | |
3287 | int nmaps; | |
3288 | xfs_bmbt_irec_t mval[SYMLINK_MAPS]; | |
3289 | xfs_daddr_t d; | |
3290 | char *cur_chunk; | |
3291 | int byte_cnt; | |
3292 | int n; | |
3293 | xfs_buf_t *bp; | |
3294 | xfs_prid_t prid; | |
3295 | struct xfs_dquot *udqp, *gdqp; | |
3296 | uint resblks; | |
3297 | char *link_name = VNAME(dentry); | |
3298 | int link_namelen; | |
3299 | ||
3300 | *vpp = NULL; | |
3301 | dir_vp = BHV_TO_VNODE(dir_bdp); | |
3302 | dp = XFS_BHVTOI(dir_bdp); | |
3303 | dp_joined_to_trans = B_FALSE; | |
3304 | error = 0; | |
3305 | ip = NULL; | |
3306 | tp = NULL; | |
3307 | ||
3308 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); | |
3309 | ||
3310 | mp = dp->i_mount; | |
3311 | ||
3312 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3313 | return XFS_ERROR(EIO); | |
3314 | ||
3315 | link_namelen = VNAMELEN(dentry); | |
3316 | ||
3317 | /* | |
3318 | * Check component lengths of the target path name. | |
3319 | */ | |
3320 | pathlen = strlen(target_path); | |
3321 | if (pathlen >= MAXPATHLEN) /* total string too long */ | |
3322 | return XFS_ERROR(ENAMETOOLONG); | |
3323 | if (pathlen >= MAXNAMELEN) { /* is any component too long? */ | |
3324 | int len, total; | |
3325 | char *path; | |
3326 | ||
3327 | for(total = 0, path = target_path; total < pathlen;) { | |
3328 | /* | |
3329 | * Skip any slashes. | |
3330 | */ | |
3331 | while(*path == '/') { | |
3332 | total++; | |
3333 | path++; | |
3334 | } | |
3335 | ||
3336 | /* | |
3337 | * Count up to the next slash or end of path. | |
3338 | * Error out if the component is bigger than MAXNAMELEN. | |
3339 | */ | |
3340 | for(len = 0; *path != '/' && total < pathlen;total++, path++) { | |
3341 | if (++len >= MAXNAMELEN) { | |
3342 | error = ENAMETOOLONG; | |
3343 | return error; | |
3344 | } | |
3345 | } | |
3346 | } | |
3347 | } | |
3348 | ||
3349 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) { | |
3350 | error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp, | |
3351 | DM_RIGHT_NULL, NULL, DM_RIGHT_NULL, | |
3352 | link_name, target_path, 0, 0, 0); | |
3353 | if (error) | |
3354 | return error; | |
3355 | } | |
3356 | ||
3357 | /* Return through std_return after this point. */ | |
3358 | ||
3359 | udqp = gdqp = NULL; | |
3360 | if (vap->va_mask & XFS_AT_PROJID) | |
3361 | prid = (xfs_prid_t)vap->va_projid; | |
3362 | else | |
3363 | prid = (xfs_prid_t)dfltprid; | |
3364 | ||
3365 | /* | |
3366 | * Make sure that we have allocated dquot(s) on disk. | |
3367 | */ | |
3368 | error = XFS_QM_DQVOPALLOC(mp, dp, | |
3369 | current_fsuid(credp), current_fsgid(credp), | |
3370 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp); | |
3371 | if (error) | |
3372 | goto std_return; | |
3373 | ||
3374 | tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK); | |
3375 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
3376 | /* | |
3377 | * The symlink will fit into the inode data fork? | |
3378 | * There can't be any attributes so we get the whole variable part. | |
3379 | */ | |
3380 | if (pathlen <= XFS_LITINO(mp)) | |
3381 | fs_blocks = 0; | |
3382 | else | |
3383 | fs_blocks = XFS_B_TO_FSB(mp, pathlen); | |
3384 | resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks); | |
3385 | error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0, | |
3386 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3387 | if (error == ENOSPC && fs_blocks == 0) { | |
3388 | resblks = 0; | |
3389 | error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0, | |
3390 | XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT); | |
3391 | } | |
3392 | if (error) { | |
3393 | cancel_flags = 0; | |
3394 | dp = NULL; | |
3395 | goto error_return; | |
3396 | } | |
3397 | ||
3398 | xfs_ilock(dp, XFS_ILOCK_EXCL); | |
3399 | ||
3400 | /* | |
3401 | * Check whether the directory allows new symlinks or not. | |
3402 | */ | |
3403 | if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { | |
3404 | error = XFS_ERROR(EPERM); | |
3405 | goto error_return; | |
3406 | } | |
3407 | ||
3408 | /* | |
3409 | * Reserve disk quota : blocks and inode. | |
3410 | */ | |
3411 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0); | |
3412 | if (error) | |
3413 | goto error_return; | |
3414 | ||
3415 | /* | |
3416 | * Check for ability to enter directory entry, if no space reserved. | |
3417 | */ | |
3418 | if (resblks == 0 && | |
3419 | (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen))) | |
3420 | goto error_return; | |
3421 | /* | |
3422 | * Initialize the bmap freelist prior to calling either | |
3423 | * bmapi or the directory create code. | |
3424 | */ | |
3425 | XFS_BMAP_INIT(&free_list, &first_block); | |
3426 | ||
3427 | /* | |
3428 | * Allocate an inode for the symlink. | |
3429 | */ | |
3430 | error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT), | |
3431 | 1, 0, credp, prid, resblks > 0, &ip, NULL); | |
3432 | if (error) { | |
3433 | if (error == ENOSPC) | |
3434 | goto error_return; | |
3435 | goto error1; | |
3436 | } | |
3437 | ITRACE(ip); | |
3438 | ||
3439 | VN_HOLD(dir_vp); | |
3440 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
3441 | dp_joined_to_trans = B_TRUE; | |
3442 | ||
3443 | /* | |
3444 | * Also attach the dquot(s) to it, if applicable. | |
3445 | */ | |
3446 | XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp); | |
3447 | ||
3448 | if (resblks) | |
3449 | resblks -= XFS_IALLOC_SPACE_RES(mp); | |
3450 | /* | |
3451 | * If the symlink will fit into the inode, write it inline. | |
3452 | */ | |
3453 | if (pathlen <= XFS_IFORK_DSIZE(ip)) { | |
3454 | xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK); | |
3455 | memcpy(ip->i_df.if_u1.if_data, target_path, pathlen); | |
3456 | ip->i_d.di_size = pathlen; | |
3457 | ||
3458 | /* | |
3459 | * The inode was initially created in extent format. | |
3460 | */ | |
3461 | ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT); | |
3462 | ip->i_df.if_flags |= XFS_IFINLINE; | |
3463 | ||
3464 | ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; | |
3465 | xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); | |
3466 | ||
3467 | } else { | |
3468 | first_fsb = 0; | |
3469 | nmaps = SYMLINK_MAPS; | |
3470 | ||
3471 | error = xfs_bmapi(tp, ip, first_fsb, fs_blocks, | |
3472 | XFS_BMAPI_WRITE | XFS_BMAPI_METADATA, | |
3473 | &first_block, resblks, mval, &nmaps, | |
3474 | &free_list); | |
3475 | if (error) { | |
3476 | goto error1; | |
3477 | } | |
3478 | ||
3479 | if (resblks) | |
3480 | resblks -= fs_blocks; | |
3481 | ip->i_d.di_size = pathlen; | |
3482 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3483 | ||
3484 | cur_chunk = target_path; | |
3485 | for (n = 0; n < nmaps; n++) { | |
3486 | d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); | |
3487 | byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); | |
3488 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, | |
3489 | BTOBB(byte_cnt), 0); | |
3490 | ASSERT(bp && !XFS_BUF_GETERROR(bp)); | |
3491 | if (pathlen < byte_cnt) { | |
3492 | byte_cnt = pathlen; | |
3493 | } | |
3494 | pathlen -= byte_cnt; | |
3495 | ||
3496 | memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt); | |
3497 | cur_chunk += byte_cnt; | |
3498 | ||
3499 | xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1); | |
3500 | } | |
3501 | } | |
3502 | ||
3503 | /* | |
3504 | * Create the directory entry for the symlink. | |
3505 | */ | |
3506 | error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen, | |
3507 | ip->i_ino, &first_block, &free_list, resblks); | |
3508 | if (error) { | |
3509 | goto error1; | |
3510 | } | |
3511 | xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3512 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
3513 | ||
3514 | /* | |
3515 | * Bump the in memory version number of the parent directory | |
3516 | * so that other processes accessing it will recognize that | |
3517 | * the directory has changed. | |
3518 | */ | |
3519 | dp->i_gen++; | |
3520 | ||
3521 | /* | |
3522 | * If this is a synchronous mount, make sure that the | |
3523 | * symlink transaction goes to disk before returning to | |
3524 | * the user. | |
3525 | */ | |
3526 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
3527 | xfs_trans_set_sync(tp); | |
3528 | } | |
3529 | ||
3530 | /* | |
3531 | * xfs_trans_commit normally decrements the vnode ref count | |
3532 | * when it unlocks the inode. Since we want to return the | |
3533 | * vnode to the caller, we bump the vnode ref count now. | |
3534 | */ | |
3535 | IHOLD(ip); | |
3536 | ||
3537 | error = xfs_bmap_finish(&tp, &free_list, first_block, &committed); | |
3538 | if (error) { | |
3539 | goto error2; | |
3540 | } | |
3541 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
3542 | XFS_QM_DQRELE(mp, udqp); | |
3543 | XFS_QM_DQRELE(mp, gdqp); | |
3544 | ||
3545 | /* Fall through to std_return with error = 0 or errno from | |
3546 | * xfs_trans_commit */ | |
3547 | std_return: | |
3548 | if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp), | |
3549 | DM_EVENT_POSTSYMLINK)) { | |
3550 | (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK, | |
3551 | dir_vp, DM_RIGHT_NULL, | |
3552 | error ? NULL : XFS_ITOV(ip), | |
3553 | DM_RIGHT_NULL, link_name, target_path, | |
3554 | 0, error, 0); | |
3555 | } | |
3556 | ||
3557 | if (!error) { | |
3558 | vnode_t *vp; | |
3559 | ||
3560 | ASSERT(ip); | |
3561 | vp = XFS_ITOV(ip); | |
3562 | *vpp = vp; | |
3563 | } | |
3564 | return error; | |
3565 | ||
3566 | error2: | |
3567 | IRELE(ip); | |
3568 | error1: | |
3569 | xfs_bmap_cancel(&free_list); | |
3570 | cancel_flags |= XFS_TRANS_ABORT; | |
3571 | error_return: | |
3572 | xfs_trans_cancel(tp, cancel_flags); | |
3573 | XFS_QM_DQRELE(mp, udqp); | |
3574 | XFS_QM_DQRELE(mp, gdqp); | |
3575 | ||
3576 | if (!dp_joined_to_trans && (dp != NULL)) { | |
3577 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
3578 | } | |
3579 | ||
3580 | goto std_return; | |
3581 | } | |
3582 | ||
3583 | ||
3584 | /* | |
3585 | * xfs_fid2 | |
3586 | * | |
3587 | * A fid routine that takes a pointer to a previously allocated | |
3588 | * fid structure (like xfs_fast_fid) but uses a 64 bit inode number. | |
3589 | */ | |
3590 | STATIC int | |
3591 | xfs_fid2( | |
3592 | bhv_desc_t *bdp, | |
3593 | fid_t *fidp) | |
3594 | { | |
3595 | xfs_inode_t *ip; | |
3596 | xfs_fid2_t *xfid; | |
3597 | ||
3598 | vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__, | |
3599 | (inst_t *)__return_address); | |
3600 | ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t)); | |
3601 | ||
3602 | xfid = (xfs_fid2_t *)fidp; | |
3603 | ip = XFS_BHVTOI(bdp); | |
3604 | xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len); | |
3605 | xfid->fid_pad = 0; | |
3606 | /* | |
3607 | * use memcpy because the inode is a long long and there's no | |
3608 | * assurance that xfid->fid_ino is properly aligned. | |
3609 | */ | |
3610 | memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino)); | |
3611 | xfid->fid_gen = ip->i_d.di_gen; | |
3612 | ||
3613 | return 0; | |
3614 | } | |
3615 | ||
3616 | ||
3617 | /* | |
3618 | * xfs_rwlock | |
3619 | */ | |
3620 | int | |
3621 | xfs_rwlock( | |
3622 | bhv_desc_t *bdp, | |
3623 | vrwlock_t locktype) | |
3624 | { | |
3625 | xfs_inode_t *ip; | |
3626 | vnode_t *vp; | |
3627 | ||
3628 | vp = BHV_TO_VNODE(bdp); | |
3629 | if (vp->v_type == VDIR) | |
3630 | return 1; | |
3631 | ip = XFS_BHVTOI(bdp); | |
3632 | if (locktype == VRWLOCK_WRITE) { | |
3633 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
3634 | } else if (locktype == VRWLOCK_TRY_READ) { | |
3635 | return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)); | |
3636 | } else if (locktype == VRWLOCK_TRY_WRITE) { | |
3637 | return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)); | |
3638 | } else { | |
3639 | ASSERT((locktype == VRWLOCK_READ) || | |
3640 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3641 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
3642 | } | |
3643 | ||
3644 | return 1; | |
3645 | } | |
3646 | ||
3647 | ||
3648 | /* | |
3649 | * xfs_rwunlock | |
3650 | */ | |
3651 | void | |
3652 | xfs_rwunlock( | |
3653 | bhv_desc_t *bdp, | |
3654 | vrwlock_t locktype) | |
3655 | { | |
3656 | xfs_inode_t *ip; | |
3657 | vnode_t *vp; | |
3658 | ||
3659 | vp = BHV_TO_VNODE(bdp); | |
3660 | if (vp->v_type == VDIR) | |
3661 | return; | |
3662 | ip = XFS_BHVTOI(bdp); | |
3663 | if (locktype == VRWLOCK_WRITE) { | |
3664 | /* | |
3665 | * In the write case, we may have added a new entry to | |
3666 | * the reference cache. This might store a pointer to | |
3667 | * an inode to be released in this inode. If it is there, | |
3668 | * clear the pointer and release the inode after unlocking | |
3669 | * this one. | |
3670 | */ | |
3671 | xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL); | |
3672 | } else { | |
3673 | ASSERT((locktype == VRWLOCK_READ) || | |
3674 | (locktype == VRWLOCK_WRITE_DIRECT)); | |
3675 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
3676 | } | |
3677 | return; | |
3678 | } | |
3679 | ||
3680 | STATIC int | |
3681 | xfs_inode_flush( | |
3682 | bhv_desc_t *bdp, | |
3683 | int flags) | |
3684 | { | |
3685 | xfs_inode_t *ip; | |
3686 | xfs_mount_t *mp; | |
3687 | xfs_inode_log_item_t *iip; | |
3688 | int error = 0; | |
3689 | ||
3690 | ip = XFS_BHVTOI(bdp); | |
3691 | mp = ip->i_mount; | |
3692 | iip = ip->i_itemp; | |
3693 | ||
3694 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3695 | return XFS_ERROR(EIO); | |
3696 | ||
3697 | /* | |
3698 | * Bypass inodes which have already been cleaned by | |
3699 | * the inode flush clustering code inside xfs_iflush | |
3700 | */ | |
3701 | if ((ip->i_update_core == 0) && | |
3702 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) | |
3703 | return 0; | |
3704 | ||
3705 | if (flags & FLUSH_LOG) { | |
3706 | if (iip && iip->ili_last_lsn) { | |
3707 | xlog_t *log = mp->m_log; | |
3708 | xfs_lsn_t sync_lsn; | |
3709 | int s, log_flags = XFS_LOG_FORCE; | |
3710 | ||
3711 | s = GRANT_LOCK(log); | |
3712 | sync_lsn = log->l_last_sync_lsn; | |
3713 | GRANT_UNLOCK(log, s); | |
3714 | ||
3715 | if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0)) | |
3716 | return 0; | |
3717 | ||
3718 | if (flags & FLUSH_SYNC) | |
3719 | log_flags |= XFS_LOG_SYNC; | |
3720 | return xfs_log_force(mp, iip->ili_last_lsn, log_flags); | |
3721 | } | |
3722 | } | |
3723 | ||
3724 | /* | |
3725 | * We make this non-blocking if the inode is contended, | |
3726 | * return EAGAIN to indicate to the caller that they | |
3727 | * did not succeed. This prevents the flush path from | |
3728 | * blocking on inodes inside another operation right | |
3729 | * now, they get caught later by xfs_sync. | |
3730 | */ | |
3731 | if (flags & FLUSH_INODE) { | |
3732 | int flush_flags; | |
3733 | ||
3734 | if (xfs_ipincount(ip)) | |
3735 | return EAGAIN; | |
3736 | ||
3737 | if (flags & FLUSH_SYNC) { | |
3738 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3739 | xfs_iflock(ip); | |
3740 | } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
3741 | if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) { | |
3742 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3743 | return EAGAIN; | |
3744 | } | |
3745 | } else { | |
3746 | return EAGAIN; | |
3747 | } | |
3748 | ||
3749 | if (flags & FLUSH_SYNC) | |
3750 | flush_flags = XFS_IFLUSH_SYNC; | |
3751 | else | |
3752 | flush_flags = XFS_IFLUSH_ASYNC; | |
3753 | ||
3754 | error = xfs_iflush(ip, flush_flags); | |
3755 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3756 | } | |
3757 | ||
3758 | return error; | |
3759 | } | |
3760 | ||
3761 | ||
3762 | int | |
3763 | xfs_set_dmattrs ( | |
3764 | bhv_desc_t *bdp, | |
3765 | u_int evmask, | |
3766 | u_int16_t state, | |
3767 | cred_t *credp) | |
3768 | { | |
3769 | xfs_inode_t *ip; | |
3770 | xfs_trans_t *tp; | |
3771 | xfs_mount_t *mp; | |
3772 | int error; | |
3773 | ||
3774 | if (!capable(CAP_SYS_ADMIN)) | |
3775 | return XFS_ERROR(EPERM); | |
3776 | ||
3777 | ip = XFS_BHVTOI(bdp); | |
3778 | mp = ip->i_mount; | |
3779 | ||
3780 | if (XFS_FORCED_SHUTDOWN(mp)) | |
3781 | return XFS_ERROR(EIO); | |
3782 | ||
3783 | tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS); | |
3784 | error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0); | |
3785 | if (error) { | |
3786 | xfs_trans_cancel(tp, 0); | |
3787 | return error; | |
3788 | } | |
3789 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3790 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
3791 | ||
3792 | ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask; | |
3793 | ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state; | |
3794 | ||
3795 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
3796 | IHOLD(ip); | |
3797 | error = xfs_trans_commit(tp, 0, NULL); | |
3798 | ||
3799 | return error; | |
3800 | } | |
3801 | ||
3802 | ||
3803 | /* | |
3804 | * xfs_reclaim | |
3805 | */ | |
3806 | STATIC int | |
3807 | xfs_reclaim( | |
3808 | bhv_desc_t *bdp) | |
3809 | { | |
3810 | xfs_inode_t *ip; | |
3811 | vnode_t *vp; | |
3812 | ||
3813 | vp = BHV_TO_VNODE(bdp); | |
3814 | ip = XFS_BHVTOI(bdp); | |
3815 | ||
3816 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
3817 | ||
3818 | ASSERT(!VN_MAPPED(vp)); | |
3819 | ||
3820 | /* bad inode, get out here ASAP */ | |
3821 | if (VN_BAD(vp)) { | |
3822 | xfs_ireclaim(ip); | |
3823 | return 0; | |
3824 | } | |
3825 | ||
3826 | if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
3827 | if (ip->i_d.di_size > 0) { | |
3828 | /* | |
3829 | * Flush and invalidate any data left around that is | |
3830 | * a part of this file. | |
3831 | * | |
3832 | * Get the inode's i/o lock so that buffers are pushed | |
3833 | * out while holding the proper lock. We can't hold | |
3834 | * the inode lock here since flushing out buffers may | |
3835 | * cause us to try to get the lock in xfs_strategy(). | |
3836 | * | |
3837 | * We don't have to call remapf() here, because there | |
3838 | * cannot be any mapped file references to this vnode | |
3839 | * since it is being reclaimed. | |
3840 | */ | |
3841 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
3842 | ||
3843 | /* | |
3844 | * If we hit an IO error, we need to make sure that the | |
3845 | * buffer and page caches of file data for | |
3846 | * the file are tossed away. We don't want to use | |
3847 | * VOP_FLUSHINVAL_PAGES here because we don't want dirty | |
3848 | * pages to stay attached to the vnode, but be | |
3849 | * marked P_BAD. pdflush/vnode_pagebad | |
3850 | * hates that. | |
3851 | */ | |
3852 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
3853 | VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE); | |
3854 | } else { | |
3855 | VOP_TOSS_PAGES(vp, 0, -1, FI_NONE); | |
3856 | } | |
3857 | ||
3858 | ASSERT(VN_CACHED(vp) == 0); | |
3859 | ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || | |
3860 | ip->i_delayed_blks == 0); | |
3861 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
3862 | } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
3863 | /* | |
3864 | * di_size field may not be quite accurate if we're | |
3865 | * shutting down. | |
3866 | */ | |
3867 | VOP_TOSS_PAGES(vp, 0, -1, FI_NONE); | |
3868 | ASSERT(VN_CACHED(vp) == 0); | |
3869 | } | |
3870 | } | |
3871 | ||
3872 | /* If we have nothing to flush with this inode then complete the | |
3873 | * teardown now, otherwise break the link between the xfs inode | |
3874 | * and the linux inode and clean up the xfs inode later. This | |
3875 | * avoids flushing the inode to disk during the delete operation | |
3876 | * itself. | |
3877 | */ | |
3878 | if (!ip->i_update_core && (ip->i_itemp == NULL)) { | |
3879 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3880 | xfs_iflock(ip); | |
3881 | return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC); | |
3882 | } else { | |
3883 | xfs_mount_t *mp = ip->i_mount; | |
3884 | ||
3885 | /* Protect sync from us */ | |
3886 | XFS_MOUNT_ILOCK(mp); | |
3887 | vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip)); | |
3888 | list_add_tail(&ip->i_reclaim, &mp->m_del_inodes); | |
3889 | ip->i_flags |= XFS_IRECLAIMABLE; | |
3890 | XFS_MOUNT_IUNLOCK(mp); | |
3891 | } | |
3892 | return 0; | |
3893 | } | |
3894 | ||
3895 | int | |
3896 | xfs_finish_reclaim( | |
3897 | xfs_inode_t *ip, | |
3898 | int locked, | |
3899 | int sync_mode) | |
3900 | { | |
3901 | xfs_ihash_t *ih = ip->i_hash; | |
3902 | vnode_t *vp = XFS_ITOV_NULL(ip); | |
3903 | int error; | |
3904 | ||
3905 | if (vp && VN_BAD(vp)) | |
3906 | goto reclaim; | |
3907 | ||
3908 | /* The hash lock here protects a thread in xfs_iget_core from | |
3909 | * racing with us on linking the inode back with a vnode. | |
3910 | * Once we have the XFS_IRECLAIM flag set it will not touch | |
3911 | * us. | |
3912 | */ | |
3913 | write_lock(&ih->ih_lock); | |
3914 | if ((ip->i_flags & XFS_IRECLAIM) || | |
3915 | (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) { | |
3916 | write_unlock(&ih->ih_lock); | |
3917 | if (locked) { | |
3918 | xfs_ifunlock(ip); | |
3919 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3920 | } | |
3921 | return(1); | |
3922 | } | |
3923 | ip->i_flags |= XFS_IRECLAIM; | |
3924 | write_unlock(&ih->ih_lock); | |
3925 | ||
3926 | /* | |
3927 | * If the inode is still dirty, then flush it out. If the inode | |
3928 | * is not in the AIL, then it will be OK to flush it delwri as | |
3929 | * long as xfs_iflush() does not keep any references to the inode. | |
3930 | * We leave that decision up to xfs_iflush() since it has the | |
3931 | * knowledge of whether it's OK to simply do a delwri flush of | |
3932 | * the inode or whether we need to wait until the inode is | |
3933 | * pulled from the AIL. | |
3934 | * We get the flush lock regardless, though, just to make sure | |
3935 | * we don't free it while it is being flushed. | |
3936 | */ | |
3937 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
3938 | if (!locked) { | |
3939 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
3940 | xfs_iflock(ip); | |
3941 | } | |
3942 | ||
3943 | if (ip->i_update_core || | |
3944 | ((ip->i_itemp != NULL) && | |
3945 | (ip->i_itemp->ili_format.ilf_fields != 0))) { | |
3946 | error = xfs_iflush(ip, sync_mode); | |
3947 | /* | |
3948 | * If we hit an error, typically because of filesystem | |
3949 | * shutdown, we don't need to let vn_reclaim to know | |
3950 | * because we're gonna reclaim the inode anyway. | |
3951 | */ | |
3952 | if (error) { | |
3953 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3954 | goto reclaim; | |
3955 | } | |
3956 | xfs_iflock(ip); /* synchronize with xfs_iflush_done */ | |
3957 | } | |
3958 | ||
3959 | ASSERT(ip->i_update_core == 0); | |
3960 | ASSERT(ip->i_itemp == NULL || | |
3961 | ip->i_itemp->ili_format.ilf_fields == 0); | |
3962 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3963 | } else if (locked) { | |
3964 | /* | |
3965 | * We are not interested in doing an iflush if we're | |
3966 | * in the process of shutting down the filesystem forcibly. | |
3967 | * So, just reclaim the inode. | |
3968 | */ | |
3969 | xfs_ifunlock(ip); | |
3970 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3971 | } | |
3972 | ||
3973 | reclaim: | |
3974 | xfs_ireclaim(ip); | |
3975 | return 0; | |
3976 | } | |
3977 | ||
3978 | int | |
3979 | xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock) | |
3980 | { | |
3981 | int purged; | |
3982 | xfs_inode_t *ip, *n; | |
3983 | int done = 0; | |
3984 | ||
3985 | while (!done) { | |
3986 | purged = 0; | |
3987 | XFS_MOUNT_ILOCK(mp); | |
3988 | list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) { | |
3989 | if (noblock) { | |
3990 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) | |
3991 | continue; | |
3992 | if (xfs_ipincount(ip) || | |
3993 | !xfs_iflock_nowait(ip)) { | |
3994 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
3995 | continue; | |
3996 | } | |
3997 | } | |
3998 | XFS_MOUNT_IUNLOCK(mp); | |
3999 | xfs_finish_reclaim(ip, noblock, | |
4000 | XFS_IFLUSH_DELWRI_ELSE_ASYNC); | |
4001 | purged = 1; | |
4002 | break; | |
4003 | } | |
4004 | ||
4005 | done = !purged; | |
4006 | } | |
4007 | ||
4008 | XFS_MOUNT_IUNLOCK(mp); | |
4009 | return 0; | |
4010 | } | |
4011 | ||
4012 | /* | |
4013 | * xfs_alloc_file_space() | |
4014 | * This routine allocates disk space for the given file. | |
4015 | * | |
4016 | * If alloc_type == 0, this request is for an ALLOCSP type | |
4017 | * request which will change the file size. In this case, no | |
4018 | * DMAPI event will be generated by the call. A TRUNCATE event | |
4019 | * will be generated later by xfs_setattr. | |
4020 | * | |
4021 | * If alloc_type != 0, this request is for a RESVSP type | |
4022 | * request, and a DMAPI DM_EVENT_WRITE will be generated if the | |
4023 | * lower block boundary byte address is less than the file's | |
4024 | * length. | |
4025 | * | |
4026 | * RETURNS: | |
4027 | * 0 on success | |
4028 | * errno on error | |
4029 | * | |
4030 | */ | |
4031 | int | |
4032 | xfs_alloc_file_space( | |
4033 | xfs_inode_t *ip, | |
4034 | xfs_off_t offset, | |
4035 | xfs_off_t len, | |
4036 | int alloc_type, | |
4037 | int attr_flags) | |
4038 | { | |
4039 | xfs_filblks_t allocated_fsb; | |
4040 | xfs_filblks_t allocatesize_fsb; | |
4041 | int committed; | |
4042 | xfs_off_t count; | |
4043 | xfs_filblks_t datablocks; | |
4044 | int error; | |
4045 | xfs_fsblock_t firstfsb; | |
4046 | xfs_bmap_free_t free_list; | |
4047 | xfs_bmbt_irec_t *imapp; | |
4048 | xfs_bmbt_irec_t imaps[1]; | |
4049 | xfs_mount_t *mp; | |
4050 | int numrtextents; | |
4051 | int reccount; | |
4052 | uint resblks; | |
4053 | int rt; | |
4054 | int rtextsize; | |
4055 | xfs_fileoff_t startoffset_fsb; | |
4056 | xfs_trans_t *tp; | |
4057 | int xfs_bmapi_flags; | |
4058 | ||
4059 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
4060 | mp = ip->i_mount; | |
4061 | ||
4062 | if (XFS_FORCED_SHUTDOWN(mp)) | |
4063 | return XFS_ERROR(EIO); | |
4064 | ||
4065 | /* | |
4066 | * determine if this is a realtime file | |
4067 | */ | |
4068 | if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) { | |
4069 | if (ip->i_d.di_extsize) | |
4070 | rtextsize = ip->i_d.di_extsize; | |
4071 | else | |
4072 | rtextsize = mp->m_sb.sb_rextsize; | |
4073 | } else | |
4074 | rtextsize = 0; | |
4075 | ||
4076 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
4077 | return error; | |
4078 | ||
4079 | if (len <= 0) | |
4080 | return XFS_ERROR(EINVAL); | |
4081 | ||
4082 | count = len; | |
4083 | error = 0; | |
4084 | imapp = &imaps[0]; | |
4085 | reccount = 1; | |
4086 | xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0); | |
4087 | startoffset_fsb = XFS_B_TO_FSBT(mp, offset); | |
4088 | allocatesize_fsb = XFS_B_TO_FSB(mp, count); | |
4089 | ||
4090 | /* Generate a DMAPI event if needed. */ | |
4091 | if (alloc_type != 0 && offset < ip->i_d.di_size && | |
4092 | (attr_flags&ATTR_DMI) == 0 && | |
4093 | DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) { | |
4094 | xfs_off_t end_dmi_offset; | |
4095 | ||
4096 | end_dmi_offset = offset+len; | |
4097 | if (end_dmi_offset > ip->i_d.di_size) | |
4098 | end_dmi_offset = ip->i_d.di_size; | |
4099 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip), | |
4100 | offset, end_dmi_offset - offset, | |
4101 | 0, NULL); | |
4102 | if (error) | |
4103 | return(error); | |
4104 | } | |
4105 | ||
4106 | /* | |
4107 | * allocate file space until done or until there is an error | |
4108 | */ | |
4109 | retry: | |
4110 | while (allocatesize_fsb && !error) { | |
4111 | /* | |
4112 | * determine if reserving space on | |
4113 | * the data or realtime partition. | |
4114 | */ | |
4115 | if (rt) { | |
4116 | xfs_fileoff_t s, e; | |
4117 | ||
4118 | s = startoffset_fsb; | |
4119 | do_div(s, rtextsize); | |
4120 | s *= rtextsize; | |
4121 | e = roundup_64(startoffset_fsb + allocatesize_fsb, | |
4122 | rtextsize); | |
4123 | numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize; | |
4124 | datablocks = 0; | |
4125 | } else { | |
4126 | datablocks = allocatesize_fsb; | |
4127 | numrtextents = 0; | |
4128 | } | |
4129 | ||
4130 | /* | |
4131 | * allocate and setup the transaction | |
4132 | */ | |
4133 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
4134 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks); | |
4135 | error = xfs_trans_reserve(tp, | |
4136 | resblks, | |
4137 | XFS_WRITE_LOG_RES(mp), | |
4138 | numrtextents, | |
4139 | XFS_TRANS_PERM_LOG_RES, | |
4140 | XFS_WRITE_LOG_COUNT); | |
4141 | ||
4142 | /* | |
4143 | * check for running out of space | |
4144 | */ | |
4145 | if (error) { | |
4146 | /* | |
4147 | * Free the transaction structure. | |
4148 | */ | |
4149 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
4150 | xfs_trans_cancel(tp, 0); | |
4151 | break; | |
4152 | } | |
4153 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4154 | error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp, | |
4155 | ip->i_udquot, ip->i_gdquot, resblks, 0, rt ? | |
4156 | XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
4157 | if (error) | |
4158 | goto error1; | |
4159 | ||
4160 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4161 | xfs_trans_ihold(tp, ip); | |
4162 | ||
4163 | /* | |
4164 | * issue the bmapi() call to allocate the blocks | |
4165 | */ | |
4166 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
4167 | error = xfs_bmapi(tp, ip, startoffset_fsb, | |
4168 | allocatesize_fsb, xfs_bmapi_flags, | |
4169 | &firstfsb, 0, imapp, &reccount, | |
4170 | &free_list); | |
4171 | if (error) { | |
4172 | goto error0; | |
4173 | } | |
4174 | ||
4175 | /* | |
4176 | * complete the transaction | |
4177 | */ | |
4178 | error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed); | |
4179 | if (error) { | |
4180 | goto error0; | |
4181 | } | |
4182 | ||
4183 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
4184 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4185 | if (error) { | |
4186 | break; | |
4187 | } | |
4188 | ||
4189 | allocated_fsb = imapp->br_blockcount; | |
4190 | ||
4191 | if (reccount == 0) { | |
4192 | error = XFS_ERROR(ENOSPC); | |
4193 | break; | |
4194 | } | |
4195 | ||
4196 | startoffset_fsb += allocated_fsb; | |
4197 | allocatesize_fsb -= allocated_fsb; | |
4198 | } | |
4199 | dmapi_enospc_check: | |
4200 | if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 && | |
4201 | DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) { | |
4202 | ||
4203 | error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE, | |
4204 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
4205 | XFS_ITOV(ip), DM_RIGHT_NULL, | |
4206 | NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */ | |
4207 | if (error == 0) | |
4208 | goto retry; /* Maybe DMAPI app. has made space */ | |
4209 | /* else fall through with error from XFS_SEND_DATA */ | |
4210 | } | |
4211 | ||
4212 | return error; | |
4213 | ||
4214 | error0: | |
4215 | xfs_bmap_cancel(&free_list); | |
4216 | error1: | |
4217 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
4218 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4219 | goto dmapi_enospc_check; | |
4220 | } | |
4221 | ||
4222 | /* | |
4223 | * Zero file bytes between startoff and endoff inclusive. | |
4224 | * The iolock is held exclusive and no blocks are buffered. | |
4225 | */ | |
4226 | STATIC int | |
4227 | xfs_zero_remaining_bytes( | |
4228 | xfs_inode_t *ip, | |
4229 | xfs_off_t startoff, | |
4230 | xfs_off_t endoff) | |
4231 | { | |
4232 | xfs_bmbt_irec_t imap; | |
4233 | xfs_fileoff_t offset_fsb; | |
4234 | xfs_off_t lastoffset; | |
4235 | xfs_off_t offset; | |
4236 | xfs_buf_t *bp; | |
4237 | xfs_mount_t *mp = ip->i_mount; | |
4238 | int nimap; | |
4239 | int error = 0; | |
4240 | ||
4241 | bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize, | |
4242 | ip->i_d.di_flags & XFS_DIFLAG_REALTIME ? | |
4243 | mp->m_rtdev_targp : mp->m_ddev_targp); | |
4244 | ||
4245 | for (offset = startoff; offset <= endoff; offset = lastoffset + 1) { | |
4246 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
4247 | nimap = 1; | |
4248 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap, | |
4249 | &nimap, NULL); | |
4250 | if (error || nimap < 1) | |
4251 | break; | |
4252 | ASSERT(imap.br_blockcount >= 1); | |
4253 | ASSERT(imap.br_startoff == offset_fsb); | |
4254 | lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; | |
4255 | if (lastoffset > endoff) | |
4256 | lastoffset = endoff; | |
4257 | if (imap.br_startblock == HOLESTARTBLOCK) | |
4258 | continue; | |
4259 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4260 | if (imap.br_state == XFS_EXT_UNWRITTEN) | |
4261 | continue; | |
4262 | XFS_BUF_UNDONE(bp); | |
4263 | XFS_BUF_UNWRITE(bp); | |
4264 | XFS_BUF_READ(bp); | |
4265 | XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock)); | |
4266 | xfsbdstrat(mp, bp); | |
4267 | if ((error = xfs_iowait(bp))) { | |
4268 | xfs_ioerror_alert("xfs_zero_remaining_bytes(read)", | |
4269 | mp, bp, XFS_BUF_ADDR(bp)); | |
4270 | break; | |
4271 | } | |
4272 | memset(XFS_BUF_PTR(bp) + | |
4273 | (offset - XFS_FSB_TO_B(mp, imap.br_startoff)), | |
4274 | 0, lastoffset - offset + 1); | |
4275 | XFS_BUF_UNDONE(bp); | |
4276 | XFS_BUF_UNREAD(bp); | |
4277 | XFS_BUF_WRITE(bp); | |
4278 | xfsbdstrat(mp, bp); | |
4279 | if ((error = xfs_iowait(bp))) { | |
4280 | xfs_ioerror_alert("xfs_zero_remaining_bytes(write)", | |
4281 | mp, bp, XFS_BUF_ADDR(bp)); | |
4282 | break; | |
4283 | } | |
4284 | } | |
4285 | xfs_buf_free(bp); | |
4286 | return error; | |
4287 | } | |
4288 | ||
4289 | /* | |
4290 | * xfs_free_file_space() | |
4291 | * This routine frees disk space for the given file. | |
4292 | * | |
4293 | * This routine is only called by xfs_change_file_space | |
4294 | * for an UNRESVSP type call. | |
4295 | * | |
4296 | * RETURNS: | |
4297 | * 0 on success | |
4298 | * errno on error | |
4299 | * | |
4300 | */ | |
4301 | STATIC int | |
4302 | xfs_free_file_space( | |
4303 | xfs_inode_t *ip, | |
4304 | xfs_off_t offset, | |
4305 | xfs_off_t len, | |
4306 | int attr_flags) | |
4307 | { | |
4308 | int committed; | |
4309 | int done; | |
4310 | xfs_off_t end_dmi_offset; | |
4311 | xfs_fileoff_t endoffset_fsb; | |
4312 | int error; | |
4313 | xfs_fsblock_t firstfsb; | |
4314 | xfs_bmap_free_t free_list; | |
4315 | xfs_off_t ilen; | |
4316 | xfs_bmbt_irec_t imap; | |
4317 | xfs_off_t ioffset; | |
4318 | xfs_extlen_t mod=0; | |
4319 | xfs_mount_t *mp; | |
4320 | int nimap; | |
4321 | uint resblks; | |
4322 | int rounding; | |
4323 | int rt; | |
4324 | xfs_fileoff_t startoffset_fsb; | |
4325 | xfs_trans_t *tp; | |
5fcbab35 | 4326 | int need_iolock = 1; |
1da177e4 LT |
4327 | |
4328 | vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address); | |
4329 | mp = ip->i_mount; | |
4330 | ||
4331 | if ((error = XFS_QM_DQATTACH(mp, ip, 0))) | |
4332 | return error; | |
4333 | ||
4334 | error = 0; | |
4335 | if (len <= 0) /* if nothing being freed */ | |
4336 | return error; | |
4337 | rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME); | |
4338 | startoffset_fsb = XFS_B_TO_FSB(mp, offset); | |
4339 | end_dmi_offset = offset + len; | |
4340 | endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset); | |
4341 | ||
4342 | if (offset < ip->i_d.di_size && | |
4343 | (attr_flags & ATTR_DMI) == 0 && | |
4344 | DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) { | |
4345 | if (end_dmi_offset > ip->i_d.di_size) | |
4346 | end_dmi_offset = ip->i_d.di_size; | |
4347 | error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip), | |
4348 | offset, end_dmi_offset - offset, | |
4349 | AT_DELAY_FLAG(attr_flags), NULL); | |
4350 | if (error) | |
4351 | return(error); | |
4352 | } | |
4353 | ||
5fcbab35 DR |
4354 | ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1); |
4355 | if (attr_flags & ATTR_NOLOCK) | |
4356 | need_iolock = 0; | |
1da177e4 LT |
4357 | if (need_iolock) |
4358 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
5fcbab35 | 4359 | |
1da177e4 LT |
4360 | rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog), |
4361 | (__uint8_t)NBPP); | |
4362 | ilen = len + (offset & (rounding - 1)); | |
4363 | ioffset = offset & ~(rounding - 1); | |
4364 | if (ilen & (rounding - 1)) | |
4365 | ilen = (ilen + rounding) & ~(rounding - 1); | |
4366 | xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0); | |
4367 | /* | |
4368 | * Need to zero the stuff we're not freeing, on disk. | |
4369 | * If its a realtime file & can't use unwritten extents then we | |
4370 | * actually need to zero the extent edges. Otherwise xfs_bunmapi | |
4371 | * will take care of it for us. | |
4372 | */ | |
4373 | if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) { | |
4374 | nimap = 1; | |
4375 | error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0, | |
4376 | &imap, &nimap, NULL); | |
4377 | if (error) | |
4378 | goto out_unlock_iolock; | |
4379 | ASSERT(nimap == 0 || nimap == 1); | |
4380 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4381 | xfs_daddr_t block; | |
4382 | ||
4383 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4384 | block = imap.br_startblock; | |
4385 | mod = do_div(block, mp->m_sb.sb_rextsize); | |
4386 | if (mod) | |
4387 | startoffset_fsb += mp->m_sb.sb_rextsize - mod; | |
4388 | } | |
4389 | nimap = 1; | |
4390 | error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0, | |
4391 | &imap, &nimap, NULL); | |
4392 | if (error) | |
4393 | goto out_unlock_iolock; | |
4394 | ASSERT(nimap == 0 || nimap == 1); | |
4395 | if (nimap && imap.br_startblock != HOLESTARTBLOCK) { | |
4396 | ASSERT(imap.br_startblock != DELAYSTARTBLOCK); | |
4397 | mod++; | |
4398 | if (mod && (mod != mp->m_sb.sb_rextsize)) | |
4399 | endoffset_fsb -= mod; | |
4400 | } | |
4401 | } | |
4402 | if ((done = (endoffset_fsb <= startoffset_fsb))) | |
4403 | /* | |
4404 | * One contiguous piece to clear | |
4405 | */ | |
4406 | error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1); | |
4407 | else { | |
4408 | /* | |
4409 | * Some full blocks, possibly two pieces to clear | |
4410 | */ | |
4411 | if (offset < XFS_FSB_TO_B(mp, startoffset_fsb)) | |
4412 | error = xfs_zero_remaining_bytes(ip, offset, | |
4413 | XFS_FSB_TO_B(mp, startoffset_fsb) - 1); | |
4414 | if (!error && | |
4415 | XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len) | |
4416 | error = xfs_zero_remaining_bytes(ip, | |
4417 | XFS_FSB_TO_B(mp, endoffset_fsb), | |
4418 | offset + len - 1); | |
4419 | } | |
4420 | ||
4421 | /* | |
4422 | * free file space until done or until there is an error | |
4423 | */ | |
4424 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); | |
4425 | while (!error && !done) { | |
4426 | ||
4427 | /* | |
4428 | * allocate and setup the transaction | |
4429 | */ | |
4430 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); | |
4431 | error = xfs_trans_reserve(tp, | |
4432 | resblks, | |
4433 | XFS_WRITE_LOG_RES(mp), | |
4434 | 0, | |
4435 | XFS_TRANS_PERM_LOG_RES, | |
4436 | XFS_WRITE_LOG_COUNT); | |
4437 | ||
4438 | /* | |
4439 | * check for running out of space | |
4440 | */ | |
4441 | if (error) { | |
4442 | /* | |
4443 | * Free the transaction structure. | |
4444 | */ | |
4445 | ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp)); | |
4446 | xfs_trans_cancel(tp, 0); | |
4447 | break; | |
4448 | } | |
4449 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4450 | error = XFS_TRANS_RESERVE_QUOTA(mp, tp, | |
4451 | ip->i_udquot, ip->i_gdquot, resblks, 0, rt ? | |
4452 | XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
4453 | if (error) | |
4454 | goto error1; | |
4455 | ||
4456 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4457 | xfs_trans_ihold(tp, ip); | |
4458 | ||
4459 | /* | |
4460 | * issue the bunmapi() call to free the blocks | |
4461 | */ | |
4462 | XFS_BMAP_INIT(&free_list, &firstfsb); | |
4463 | error = xfs_bunmapi(tp, ip, startoffset_fsb, | |
4464 | endoffset_fsb - startoffset_fsb, | |
4465 | 0, 2, &firstfsb, &free_list, &done); | |
4466 | if (error) { | |
4467 | goto error0; | |
4468 | } | |
4469 | ||
4470 | /* | |
4471 | * complete the transaction | |
4472 | */ | |
4473 | error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed); | |
4474 | if (error) { | |
4475 | goto error0; | |
4476 | } | |
4477 | ||
4478 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL); | |
4479 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4480 | } | |
4481 | ||
4482 | out_unlock_iolock: | |
4483 | if (need_iolock) | |
4484 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
4485 | return error; | |
4486 | ||
4487 | error0: | |
4488 | xfs_bmap_cancel(&free_list); | |
4489 | error1: | |
4490 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
4491 | xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) : | |
4492 | XFS_ILOCK_EXCL); | |
4493 | return error; | |
4494 | } | |
4495 | ||
4496 | /* | |
4497 | * xfs_change_file_space() | |
4498 | * This routine allocates or frees disk space for the given file. | |
4499 | * The user specified parameters are checked for alignment and size | |
4500 | * limitations. | |
4501 | * | |
4502 | * RETURNS: | |
4503 | * 0 on success | |
4504 | * errno on error | |
4505 | * | |
4506 | */ | |
4507 | int | |
4508 | xfs_change_file_space( | |
4509 | bhv_desc_t *bdp, | |
4510 | int cmd, | |
4511 | xfs_flock64_t *bf, | |
4512 | xfs_off_t offset, | |
4513 | cred_t *credp, | |
4514 | int attr_flags) | |
4515 | { | |
4516 | int clrprealloc; | |
4517 | int error; | |
4518 | xfs_fsize_t fsize; | |
4519 | xfs_inode_t *ip; | |
4520 | xfs_mount_t *mp; | |
4521 | int setprealloc; | |
4522 | xfs_off_t startoffset; | |
4523 | xfs_off_t llen; | |
4524 | xfs_trans_t *tp; | |
4525 | vattr_t va; | |
4526 | vnode_t *vp; | |
4527 | ||
4528 | vp = BHV_TO_VNODE(bdp); | |
4529 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); | |
4530 | ||
4531 | ip = XFS_BHVTOI(bdp); | |
4532 | mp = ip->i_mount; | |
4533 | ||
4534 | /* | |
4535 | * must be a regular file and have write permission | |
4536 | */ | |
4537 | if (vp->v_type != VREG) | |
4538 | return XFS_ERROR(EINVAL); | |
4539 | ||
4540 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
4541 | ||
4542 | if ((error = xfs_iaccess(ip, S_IWUSR, credp))) { | |
4543 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
4544 | return error; | |
4545 | } | |
4546 | ||
4547 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
4548 | ||
4549 | switch (bf->l_whence) { | |
4550 | case 0: /*SEEK_SET*/ | |
4551 | break; | |
4552 | case 1: /*SEEK_CUR*/ | |
4553 | bf->l_start += offset; | |
4554 | break; | |
4555 | case 2: /*SEEK_END*/ | |
4556 | bf->l_start += ip->i_d.di_size; | |
4557 | break; | |
4558 | default: | |
4559 | return XFS_ERROR(EINVAL); | |
4560 | } | |
4561 | ||
4562 | llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len; | |
4563 | ||
4564 | if ( (bf->l_start < 0) | |
4565 | || (bf->l_start > XFS_MAXIOFFSET(mp)) | |
4566 | || (bf->l_start + llen < 0) | |
4567 | || (bf->l_start + llen > XFS_MAXIOFFSET(mp))) | |
4568 | return XFS_ERROR(EINVAL); | |
4569 | ||
4570 | bf->l_whence = 0; | |
4571 | ||
4572 | startoffset = bf->l_start; | |
4573 | fsize = ip->i_d.di_size; | |
4574 | ||
4575 | /* | |
4576 | * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve | |
4577 | * file space. | |
4578 | * These calls do NOT zero the data space allocated to the file, | |
4579 | * nor do they change the file size. | |
4580 | * | |
4581 | * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file | |
4582 | * space. | |
4583 | * These calls cause the new file data to be zeroed and the file | |
4584 | * size to be changed. | |
4585 | */ | |
4586 | setprealloc = clrprealloc = 0; | |
4587 | ||
4588 | switch (cmd) { | |
4589 | case XFS_IOC_RESVSP: | |
4590 | case XFS_IOC_RESVSP64: | |
4591 | error = xfs_alloc_file_space(ip, startoffset, bf->l_len, | |
4592 | 1, attr_flags); | |
4593 | if (error) | |
4594 | return error; | |
4595 | setprealloc = 1; | |
4596 | break; | |
4597 | ||
4598 | case XFS_IOC_UNRESVSP: | |
4599 | case XFS_IOC_UNRESVSP64: | |
4600 | if ((error = xfs_free_file_space(ip, startoffset, bf->l_len, | |
4601 | attr_flags))) | |
4602 | return error; | |
4603 | break; | |
4604 | ||
4605 | case XFS_IOC_ALLOCSP: | |
4606 | case XFS_IOC_ALLOCSP64: | |
4607 | case XFS_IOC_FREESP: | |
4608 | case XFS_IOC_FREESP64: | |
4609 | if (startoffset > fsize) { | |
4610 | error = xfs_alloc_file_space(ip, fsize, | |
4611 | startoffset - fsize, 0, attr_flags); | |
4612 | if (error) | |
4613 | break; | |
4614 | } | |
4615 | ||
4616 | va.va_mask = XFS_AT_SIZE; | |
4617 | va.va_size = startoffset; | |
4618 | ||
4619 | error = xfs_setattr(bdp, &va, attr_flags, credp); | |
4620 | ||
4621 | if (error) | |
4622 | return error; | |
4623 | ||
4624 | clrprealloc = 1; | |
4625 | break; | |
4626 | ||
4627 | default: | |
4628 | ASSERT(0); | |
4629 | return XFS_ERROR(EINVAL); | |
4630 | } | |
4631 | ||
4632 | /* | |
4633 | * update the inode timestamp, mode, and prealloc flag bits | |
4634 | */ | |
4635 | tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID); | |
4636 | ||
4637 | if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp), | |
4638 | 0, 0, 0))) { | |
4639 | /* ASSERT(0); */ | |
4640 | xfs_trans_cancel(tp, 0); | |
4641 | return error; | |
4642 | } | |
4643 | ||
4644 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4645 | ||
4646 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
4647 | xfs_trans_ihold(tp, ip); | |
4648 | ||
4649 | if ((attr_flags & ATTR_DMI) == 0) { | |
4650 | ip->i_d.di_mode &= ~S_ISUID; | |
4651 | ||
4652 | /* | |
4653 | * Note that we don't have to worry about mandatory | |
4654 | * file locking being disabled here because we only | |
4655 | * clear the S_ISGID bit if the Group execute bit is | |
4656 | * on, but if it was on then mandatory locking wouldn't | |
4657 | * have been enabled. | |
4658 | */ | |
4659 | if (ip->i_d.di_mode & S_IXGRP) | |
4660 | ip->i_d.di_mode &= ~S_ISGID; | |
4661 | ||
4662 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
4663 | } | |
4664 | if (setprealloc) | |
4665 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
4666 | else if (clrprealloc) | |
4667 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
4668 | ||
4669 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
4670 | xfs_trans_set_sync(tp); | |
4671 | ||
4672 | error = xfs_trans_commit(tp, 0, NULL); | |
4673 | ||
4674 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4675 | ||
4676 | return error; | |
4677 | } | |
4678 | ||
4679 | vnodeops_t xfs_vnodeops = { | |
4680 | BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS), | |
4681 | .vop_open = xfs_open, | |
4682 | .vop_read = xfs_read, | |
4683 | #ifdef HAVE_SENDFILE | |
4684 | .vop_sendfile = xfs_sendfile, | |
4685 | #endif | |
4686 | .vop_write = xfs_write, | |
4687 | .vop_ioctl = xfs_ioctl, | |
4688 | .vop_getattr = xfs_getattr, | |
4689 | .vop_setattr = xfs_setattr, | |
4690 | .vop_access = xfs_access, | |
4691 | .vop_lookup = xfs_lookup, | |
4692 | .vop_create = xfs_create, | |
4693 | .vop_remove = xfs_remove, | |
4694 | .vop_link = xfs_link, | |
4695 | .vop_rename = xfs_rename, | |
4696 | .vop_mkdir = xfs_mkdir, | |
4697 | .vop_rmdir = xfs_rmdir, | |
4698 | .vop_readdir = xfs_readdir, | |
4699 | .vop_symlink = xfs_symlink, | |
4700 | .vop_readlink = xfs_readlink, | |
4701 | .vop_fsync = xfs_fsync, | |
4702 | .vop_inactive = xfs_inactive, | |
4703 | .vop_fid2 = xfs_fid2, | |
4704 | .vop_rwlock = xfs_rwlock, | |
4705 | .vop_rwunlock = xfs_rwunlock, | |
4706 | .vop_bmap = xfs_bmap, | |
4707 | .vop_reclaim = xfs_reclaim, | |
4708 | .vop_attr_get = xfs_attr_get, | |
4709 | .vop_attr_set = xfs_attr_set, | |
4710 | .vop_attr_remove = xfs_attr_remove, | |
4711 | .vop_attr_list = xfs_attr_list, | |
4712 | .vop_link_removed = (vop_link_removed_t)fs_noval, | |
4713 | .vop_vnode_change = (vop_vnode_change_t)fs_noval, | |
4714 | .vop_tosspages = fs_tosspages, | |
4715 | .vop_flushinval_pages = fs_flushinval_pages, | |
4716 | .vop_flush_pages = fs_flush_pages, | |
4717 | .vop_release = xfs_release, | |
4718 | .vop_iflush = xfs_inode_flush, | |
4719 | }; |