blob: 061e2ffdd1dee3613bedb671dbdbb9256350aac8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
26#include "xfs_sb.h"
27#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_dir2.h"
29#include "xfs_dmapi.h"
30#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_dir2_sf.h"
36#include "xfs_attr_sf.h"
37#include "xfs_dinode.h"
38#include "xfs_inode.h"
39#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_itable.h"
41#include "xfs_btree.h"
42#include "xfs_ialloc.h"
43#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include "xfs_attr.h"
46#include "xfs_rw.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include "xfs_quota.h"
49#include "xfs_utils.h"
Nathan Scotta844f452005-11-02 14:38:42 +110050#include "xfs_rtalloc.h"
51#include "xfs_refcache.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include "xfs_log_priv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110054#include "xfs_mac.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056STATIC int
57xfs_open(
58 bhv_desc_t *bdp,
59 cred_t *credp)
60{
61 int mode;
Nathan Scott67fcaa72006-06-09 17:00:52 +100062 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
Nathan Scott7d4fb402006-06-09 15:27:16 +100063 xfs_inode_t *ip = XFS_BHVTOI(bdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
65 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
66 return XFS_ERROR(EIO);
67
68 /*
69 * If it's a directory with any blocks, read-ahead block 0
70 * as we're almost certain to have the next operation be a read there.
71 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +100072 if (VN_ISDIR(vp) && ip->i_d.di_nextents > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 mode = xfs_ilock_map_shared(ip);
74 if (ip->i_d.di_nextents > 0)
75 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
76 xfs_iunlock(ip, mode);
77 }
78 return 0;
79}
80
Nathan Scott7d4fb402006-06-09 15:27:16 +100081STATIC int
82xfs_close(
83 bhv_desc_t *bdp,
84 int flags,
85 lastclose_t lastclose,
86 cred_t *credp)
87{
Nathan Scott67fcaa72006-06-09 17:00:52 +100088 bhv_vnode_t *vp = BHV_TO_VNODE(bdp);
Nathan Scott7d4fb402006-06-09 15:27:16 +100089 xfs_inode_t *ip = XFS_BHVTOI(bdp);
Nathan Scott7d4fb402006-06-09 15:27:16 +100090
91 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
92 return XFS_ERROR(EIO);
93
94 if (lastclose != L_TRUE || !VN_ISREG(vp))
95 return 0;
96
97 /*
98 * If we previously truncated this file and removed old data in
99 * the process, we want to initiate "early" writeout on the last
100 * close. This is an attempt to combat the notorious NULL files
101 * problem which is particularly noticable from a truncate down,
102 * buffered (re-)write (delalloc), followed by a crash. What we
103 * are effectively doing here is significantly reducing the time
104 * window where we'd otherwise be exposed to that problem.
105 */
106 if (VUNTRUNCATE(vp) && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
Nathan Scott67fcaa72006-06-09 17:00:52 +1000107 return bhv_vop_flush_pages(vp, 0, -1, XFS_B_ASYNC, FI_NONE);
108 return 0;
Nathan Scott7d4fb402006-06-09 15:27:16 +1000109}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
111/*
112 * xfs_getattr
113 */
114STATIC int
115xfs_getattr(
116 bhv_desc_t *bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +1000117 bhv_vattr_t *vap,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 int flags,
119 cred_t *credp)
120{
121 xfs_inode_t *ip;
122 xfs_mount_t *mp;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000123 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
125 vp = BHV_TO_VNODE(bdp);
126 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
127
128 ip = XFS_BHVTOI(bdp);
129 mp = ip->i_mount;
130
131 if (XFS_FORCED_SHUTDOWN(mp))
132 return XFS_ERROR(EIO);
133
134 if (!(flags & ATTR_LAZY))
135 xfs_ilock(ip, XFS_ILOCK_SHARED);
136
137 vap->va_size = ip->i_d.di_size;
138 if (vap->va_mask == XFS_AT_SIZE)
139 goto all_done;
140
141 vap->va_nblocks =
142 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
143 vap->va_nodeid = ip->i_ino;
144#if XFS_BIG_INUMS
145 vap->va_nodeid += mp->m_inoadd;
146#endif
147 vap->va_nlink = ip->i_d.di_nlink;
148
149 /*
150 * Quick exit for non-stat callers
151 */
152 if ((vap->va_mask &
153 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
154 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
155 goto all_done;
156
157 /*
158 * Copy from in-core inode.
159 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000160 vap->va_mode = ip->i_d.di_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 vap->va_uid = ip->i_d.di_uid;
162 vap->va_gid = ip->i_d.di_gid;
163 vap->va_projid = ip->i_d.di_projid;
164
165 /*
166 * Check vnode type block/char vs. everything else.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000168 switch (ip->i_d.di_mode & S_IFMT) {
169 case S_IFBLK:
170 case S_IFCHR:
171 vap->va_rdev = ip->i_df.if_u2.if_rdev;
172 vap->va_blocksize = BLKDEV_IOSIZE;
173 break;
174 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 vap->va_rdev = 0;
176
177 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
David Chinnere8c8b3a2005-11-02 10:33:05 +1100178 vap->va_blocksize = xfs_preferred_iosize(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 } else {
180
181 /*
182 * If the file blocks are being allocated from a
183 * realtime partition, then return the inode's
184 * realtime extent size or the realtime volume's
185 * extent size.
186 */
187 vap->va_blocksize = ip->i_d.di_extsize ?
188 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
189 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
190 }
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000191 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 }
193
Nathan Scottca5ccbf2006-01-11 21:03:04 +1100194 vn_atime_to_timespec(vp, &vap->va_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
196 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
197 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
198 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
199
200 /*
201 * Exit for stat callers. See if any of the rest of the fields
202 * to be filled in are needed.
203 */
204 if ((vap->va_mask &
205 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
206 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
207 goto all_done;
208
209 /*
210 * Convert di_flags to xflags.
211 */
212 vap->va_xflags = xfs_ip2xflags(ip);
213
214 /*
215 * Exit for inode revalidate. See if any of the rest of
216 * the fields to be filled in are needed.
217 */
218 if ((vap->va_mask &
219 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
220 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
221 goto all_done;
222
223 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
224 vap->va_nextents =
225 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
226 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
227 ip->i_d.di_nextents;
228 if (ip->i_afp)
229 vap->va_anextents =
230 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
231 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
232 ip->i_d.di_anextents;
233 else
234 vap->va_anextents = 0;
235 vap->va_gen = ip->i_d.di_gen;
236
237 all_done:
238 if (!(flags & ATTR_LAZY))
239 xfs_iunlock(ip, XFS_ILOCK_SHARED);
240 return 0;
241}
242
243
244/*
245 * xfs_setattr
246 */
247int
248xfs_setattr(
249 bhv_desc_t *bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +1000250 bhv_vattr_t *vap,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 int flags,
252 cred_t *credp)
253{
254 xfs_inode_t *ip;
255 xfs_trans_t *tp;
256 xfs_mount_t *mp;
257 int mask;
258 int code;
259 uint lock_flags;
260 uint commit_flags=0;
261 uid_t uid=0, iuid=0;
262 gid_t gid=0, igid=0;
263 int timeflags = 0;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000264 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 xfs_prid_t projid=0, iprojid=0;
266 int mandlock_before, mandlock_after;
267 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
268 int file_owner;
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700269 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
271 vp = BHV_TO_VNODE(bdp);
272 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
273
274 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
275 return XFS_ERROR(EROFS);
276
277 /*
278 * Cannot set certain attributes.
279 */
280 mask = vap->va_mask;
281 if (mask & XFS_AT_NOSET) {
282 return XFS_ERROR(EINVAL);
283 }
284
285 ip = XFS_BHVTOI(bdp);
286 mp = ip->i_mount;
287
288 if (XFS_FORCED_SHUTDOWN(mp))
289 return XFS_ERROR(EIO);
290
291 /*
292 * Timestamps do not need to be logged and hence do not
293 * need to be done within a transaction.
294 */
295 if (mask & XFS_AT_UPDTIMES) {
296 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
297 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
298 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
299 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
300 xfs_ichgtime(ip, timeflags);
301 return 0;
302 }
303
304 olddquot1 = olddquot2 = NULL;
305 udqp = gdqp = NULL;
306
307 /*
308 * If disk quotas is on, we make sure that the dquots do exist on disk,
309 * before we start any other transactions. Trying to do this later
310 * is messy. We don't care to take a readlock to look at the ids
311 * in inode here, because we can't hold it across the trans_reserve.
312 * If the IDs do change before we take the ilock, we're covered
313 * because the i_*dquot fields will get updated anyway.
314 */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000315 if (XFS_IS_QUOTA_ON(mp) &&
316 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 uint qflags = 0;
318
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000319 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 uid = vap->va_uid;
321 qflags |= XFS_QMOPT_UQUOTA;
322 } else {
323 uid = ip->i_d.di_uid;
324 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000325 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 gid = vap->va_gid;
327 qflags |= XFS_QMOPT_GQUOTA;
328 } else {
329 gid = ip->i_d.di_gid;
330 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000331 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
332 projid = vap->va_projid;
333 qflags |= XFS_QMOPT_PQUOTA;
334 } else {
335 projid = ip->i_d.di_projid;
336 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 /*
338 * We take a reference when we initialize udqp and gdqp,
339 * so it is important that we never blindly double trip on
340 * the same variable. See xfs_create() for an example.
341 */
342 ASSERT(udqp == NULL);
343 ASSERT(gdqp == NULL);
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000344 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
345 &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 if (code)
Jesper Juhl014c2542006-01-15 02:37:08 +0100347 return code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 }
349
350 /*
351 * For the other attributes, we acquire the inode lock and
352 * first do an error checking pass.
353 */
354 tp = NULL;
355 lock_flags = XFS_ILOCK_EXCL;
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700356 if (flags & ATTR_NOLOCK)
357 need_iolock = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 if (!(mask & XFS_AT_SIZE)) {
359 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
360 (mp->m_flags & XFS_MOUNT_WSYNC)) {
361 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
362 commit_flags = 0;
363 if ((code = xfs_trans_reserve(tp, 0,
364 XFS_ICHANGE_LOG_RES(mp), 0,
365 0, 0))) {
366 lock_flags = 0;
367 goto error_return;
368 }
369 }
370 } else {
371 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
372 !(flags & ATTR_DMI)) {
373 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
374 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
375 vap->va_size, 0, dmflags, NULL);
376 if (code) {
377 lock_flags = 0;
378 goto error_return;
379 }
380 }
381 if (need_iolock)
382 lock_flags |= XFS_IOLOCK_EXCL;
383 }
384
385 xfs_ilock(ip, lock_flags);
386
387 /* boolean: are we the file owner? */
388 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
389
390 /*
391 * Change various properties of a file.
392 * Only the owner or users with CAP_FOWNER
393 * capability may do these things.
394 */
395 if (mask &
396 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
397 XFS_AT_GID|XFS_AT_PROJID)) {
398 /*
399 * CAP_FOWNER overrides the following restrictions:
400 *
401 * The user ID of the calling process must be equal
402 * to the file owner ID, except in cases where the
403 * CAP_FSETID capability is applicable.
404 */
405 if (!file_owner && !capable(CAP_FOWNER)) {
406 code = XFS_ERROR(EPERM);
407 goto error_return;
408 }
409
410 /*
411 * CAP_FSETID overrides the following restrictions:
412 *
413 * The effective user ID of the calling process shall match
414 * the file owner when setting the set-user-ID and
415 * set-group-ID bits on that file.
416 *
417 * The effective group ID or one of the supplementary group
418 * IDs of the calling process shall match the group owner of
419 * the file when setting the set-group-ID bit on that file
420 */
421 if (mask & XFS_AT_MODE) {
422 mode_t m = 0;
423
424 if ((vap->va_mode & S_ISUID) && !file_owner)
425 m |= S_ISUID;
426 if ((vap->va_mode & S_ISGID) &&
427 !in_group_p((gid_t)ip->i_d.di_gid))
428 m |= S_ISGID;
429#if 0
430 /* Linux allows this, Irix doesn't. */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000431 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 m |= S_ISVTX;
433#endif
434 if (m && !capable(CAP_FSETID))
435 vap->va_mode &= ~m;
436 }
437 }
438
439 /*
440 * Change file ownership. Must be the owner or privileged.
441 * If the system was configured with the "restricted_chown"
442 * option, the owner is not permitted to give away the file,
443 * and can change the group id only to a group of which he
444 * or she is a member.
445 */
446 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
447 /*
448 * These IDs could have changed since we last looked at them.
449 * But, we're assured that if the ownership did change
450 * while we didn't have the inode locked, inode's dquot(s)
451 * would have changed also.
452 */
453 iuid = ip->i_d.di_uid;
454 iprojid = ip->i_d.di_projid;
455 igid = ip->i_d.di_gid;
456 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
457 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
458 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
459 iprojid;
460
461 /*
462 * CAP_CHOWN overrides the following restrictions:
463 *
464 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
465 * shall override the restriction that a process cannot
466 * change the user ID of a file it owns and the restriction
467 * that the group ID supplied to the chown() function
468 * shall be equal to either the group ID or one of the
469 * supplementary group IDs of the calling process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 */
471 if (restricted_chown &&
472 (iuid != uid || (igid != gid &&
473 !in_group_p((gid_t)gid))) &&
474 !capable(CAP_CHOWN)) {
475 code = XFS_ERROR(EPERM);
476 goto error_return;
477 }
478 /*
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000479 * Do a quota reservation only if uid/projid/gid is actually
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 * going to change.
481 */
482 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000483 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
485 ASSERT(tp);
486 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
487 capable(CAP_FOWNER) ?
488 XFS_QMOPT_FORCE_RES : 0);
489 if (code) /* out of quota */
490 goto error_return;
491 }
492 }
493
494 /*
495 * Truncate file. Must have write permission and not be a directory.
496 */
497 if (mask & XFS_AT_SIZE) {
498 /* Short circuit the truncate case for zero length files */
499 if ((vap->va_size == 0) &&
500 (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
501 xfs_iunlock(ip, XFS_ILOCK_EXCL);
502 lock_flags &= ~XFS_ILOCK_EXCL;
503 if (mask & XFS_AT_CTIME)
504 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
505 code = 0;
506 goto error_return;
507 }
508
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000509 if (VN_ISDIR(vp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 code = XFS_ERROR(EISDIR);
511 goto error_return;
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000512 } else if (!VN_ISREG(vp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 code = XFS_ERROR(EINVAL);
514 goto error_return;
515 }
516 /*
517 * Make sure that the dquots are attached to the inode.
518 */
519 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
520 goto error_return;
521 }
522
523 /*
524 * Change file access or modified times.
525 */
526 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
527 if (!file_owner) {
528 if ((flags & ATTR_UTIME) &&
529 !capable(CAP_FOWNER)) {
530 code = XFS_ERROR(EPERM);
531 goto error_return;
532 }
533 }
534 }
535
536 /*
537 * Change extent size or realtime flag.
538 */
539 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
540 /*
541 * Can't change extent size if any extents are allocated.
542 */
Eric Sandeene94af022005-11-02 15:10:41 +1100543 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
545 vap->va_extsize) ) {
546 code = XFS_ERROR(EINVAL); /* EFBIG? */
547 goto error_return;
548 }
549
550 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 * Can't change realtime flag if any extents are allocated.
552 */
Eric Sandeene94af022005-11-02 15:10:41 +1100553 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
554 (mask & XFS_AT_XFLAGS) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
556 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
557 code = XFS_ERROR(EINVAL); /* EFBIG? */
558 goto error_return;
559 }
560 /*
561 * Extent size must be a multiple of the appropriate block
562 * size, if set at all.
563 */
564 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
565 xfs_extlen_t size;
566
567 if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
568 ((mask & XFS_AT_XFLAGS) &&
569 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
570 size = mp->m_sb.sb_rextsize <<
571 mp->m_sb.sb_blocklog;
572 } else {
573 size = mp->m_sb.sb_blocksize;
574 }
575 if (vap->va_extsize % size) {
576 code = XFS_ERROR(EINVAL);
577 goto error_return;
578 }
579 }
580 /*
581 * If realtime flag is set then must have realtime data.
582 */
583 if ((mask & XFS_AT_XFLAGS) &&
584 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
585 if ((mp->m_sb.sb_rblocks == 0) ||
586 (mp->m_sb.sb_rextsize == 0) ||
587 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
588 code = XFS_ERROR(EINVAL);
589 goto error_return;
590 }
591 }
592
593 /*
594 * Can't modify an immutable/append-only file unless
595 * we have appropriate permission.
596 */
597 if ((mask & XFS_AT_XFLAGS) &&
598 (ip->i_d.di_flags &
599 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
600 (vap->va_xflags &
601 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
602 !capable(CAP_LINUX_IMMUTABLE)) {
603 code = XFS_ERROR(EPERM);
604 goto error_return;
605 }
606 }
607
608 /*
609 * Now we can make the changes. Before we join the inode
610 * to the transaction, if XFS_AT_SIZE is set then take care of
611 * the part of the truncation that must be done without the
612 * inode lock. This needs to be done before joining the inode
613 * to the transaction, because the inode cannot be unlocked
614 * once it is a part of the transaction.
615 */
616 if (mask & XFS_AT_SIZE) {
617 code = 0;
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100618 if ((vap->va_size > ip->i_d.di_size) &&
619 (flags & ATTR_NOSIZETOK) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 code = xfs_igrow_start(ip, vap->va_size, credp);
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100621 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Yingping Lu9fa80462006-03-22 12:44:35 +1100623 vn_iowait(vp); /* wait for the completion of any pending DIOs */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 if (!code)
625 code = xfs_itruncate_data(ip, vap->va_size);
626 if (code) {
627 ASSERT(tp == NULL);
628 lock_flags &= ~XFS_ILOCK_EXCL;
629 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
630 goto error_return;
631 }
632 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
633 if ((code = xfs_trans_reserve(tp, 0,
634 XFS_ITRUNCATE_LOG_RES(mp), 0,
635 XFS_TRANS_PERM_LOG_RES,
636 XFS_ITRUNCATE_LOG_COUNT))) {
637 xfs_trans_cancel(tp, 0);
638 if (need_iolock)
639 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
640 return code;
641 }
642 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
643 xfs_ilock(ip, XFS_ILOCK_EXCL);
644 }
645
646 if (tp) {
647 xfs_trans_ijoin(tp, ip, lock_flags);
648 xfs_trans_ihold(tp, ip);
649 }
650
651 /* determine whether mandatory locking mode changes */
652 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
653
654 /*
655 * Truncate file. Must have write permission and not be a directory.
656 */
657 if (mask & XFS_AT_SIZE) {
658 if (vap->va_size > ip->i_d.di_size) {
659 xfs_igrow_finish(tp, ip, vap->va_size,
660 !(flags & ATTR_DMI));
661 } else if ((vap->va_size <= ip->i_d.di_size) ||
662 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
663 /*
664 * signal a sync transaction unless
665 * we're truncating an already unlinked
666 * file on a wsync filesystem
667 */
668 code = xfs_itruncate_finish(&tp, ip,
669 (xfs_fsize_t)vap->va_size,
670 XFS_DATA_FORK,
671 ((ip->i_d.di_nlink != 0 ||
672 !(mp->m_flags & XFS_MOUNT_WSYNC))
673 ? 1 : 0));
Nathan Scott7d4fb402006-06-09 15:27:16 +1000674 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 goto abort_return;
Nathan Scott7d4fb402006-06-09 15:27:16 +1000676 /*
677 * Truncated "down", so we're removing references
678 * to old data here - if we now delay flushing for
679 * a long time, we expose ourselves unduly to the
680 * notorious NULL files problem. So, we mark this
681 * vnode and flush it when the file is closed, and
682 * do not wait the usual (long) time for writeout.
683 */
684 VTRUNCATE(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 }
686 /*
687 * Have to do this even if the file's size doesn't change.
688 */
689 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
690 }
691
692 /*
693 * Change file access modes.
694 */
695 if (mask & XFS_AT_MODE) {
696 ip->i_d.di_mode &= S_IFMT;
697 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
698
699 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
700 timeflags |= XFS_ICHGTIME_CHG;
701 }
702
703 /*
704 * Change file ownership. Must be the owner or privileged.
705 * If the system was configured with the "restricted_chown"
706 * option, the owner is not permitted to give away the file,
707 * and can change the group id only to a group of which he
708 * or she is a member.
709 */
710 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
711 /*
712 * CAP_FSETID overrides the following restrictions:
713 *
714 * The set-user-ID and set-group-ID bits of a file will be
715 * cleared upon successful return from chown()
716 */
717 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
718 !capable(CAP_FSETID)) {
719 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
720 }
721
722 /*
723 * Change the ownerships and register quota modifications
724 * in the transaction.
725 */
726 if (iuid != uid) {
727 if (XFS_IS_UQUOTA_ON(mp)) {
728 ASSERT(mask & XFS_AT_UID);
729 ASSERT(udqp);
730 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
731 &ip->i_udquot, udqp);
732 }
733 ip->i_d.di_uid = uid;
734 }
735 if (igid != gid) {
736 if (XFS_IS_GQUOTA_ON(mp)) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000737 ASSERT(!XFS_IS_PQUOTA_ON(mp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 ASSERT(mask & XFS_AT_GID);
739 ASSERT(gdqp);
740 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
741 &ip->i_gdquot, gdqp);
742 }
743 ip->i_d.di_gid = gid;
744 }
745 if (iprojid != projid) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000746 if (XFS_IS_PQUOTA_ON(mp)) {
747 ASSERT(!XFS_IS_GQUOTA_ON(mp));
748 ASSERT(mask & XFS_AT_PROJID);
749 ASSERT(gdqp);
750 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
751 &ip->i_gdquot, gdqp);
752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 ip->i_d.di_projid = projid;
754 /*
755 * We may have to rev the inode as well as
756 * the superblock version number since projids didn't
757 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
758 */
759 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
760 xfs_bump_ino_vers2(tp, ip);
761 }
762
763 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
764 timeflags |= XFS_ICHGTIME_CHG;
765 }
766
767
768 /*
769 * Change file access or modified times.
770 */
771 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
772 if (mask & XFS_AT_ATIME) {
773 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
774 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
775 ip->i_update_core = 1;
776 timeflags &= ~XFS_ICHGTIME_ACC;
777 }
778 if (mask & XFS_AT_MTIME) {
779 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
780 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
781 timeflags &= ~XFS_ICHGTIME_MOD;
782 timeflags |= XFS_ICHGTIME_CHG;
783 }
784 if (tp && (flags & ATTR_UTIME))
785 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
786 }
787
788 /*
789 * Change XFS-added attributes.
790 */
791 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
792 if (mask & XFS_AT_EXTSIZE) {
793 /*
794 * Converting bytes to fs blocks.
795 */
796 ip->i_d.di_extsize = vap->va_extsize >>
797 mp->m_sb.sb_blocklog;
798 }
799 if (mask & XFS_AT_XFLAGS) {
800 uint di_flags;
801
802 /* can't set PREALLOC this way, just preserve it */
803 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
804 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
805 di_flags |= XFS_DIFLAG_IMMUTABLE;
806 if (vap->va_xflags & XFS_XFLAG_APPEND)
807 di_flags |= XFS_DIFLAG_APPEND;
808 if (vap->va_xflags & XFS_XFLAG_SYNC)
809 di_flags |= XFS_DIFLAG_SYNC;
810 if (vap->va_xflags & XFS_XFLAG_NOATIME)
811 di_flags |= XFS_DIFLAG_NOATIME;
812 if (vap->va_xflags & XFS_XFLAG_NODUMP)
813 di_flags |= XFS_DIFLAG_NODUMP;
Nathan Scott365ca832005-06-21 15:39:12 +1000814 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
815 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000816 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
817 di_flags |= XFS_DIFLAG_NODEFRAG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
819 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
820 di_flags |= XFS_DIFLAG_RTINHERIT;
821 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
822 di_flags |= XFS_DIFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100823 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
824 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
825 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
827 di_flags |= XFS_DIFLAG_REALTIME;
828 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
829 } else {
830 ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
831 }
Nathan Scottdd9f4382006-01-11 15:28:28 +1100832 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
833 di_flags |= XFS_DIFLAG_EXTSIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 }
835 ip->i_d.di_flags = di_flags;
836 }
837 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
838 timeflags |= XFS_ICHGTIME_CHG;
839 }
840
841 /*
842 * Change file inode change time only if XFS_AT_CTIME set
843 * AND we have been called by a DMI function.
844 */
845
846 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
847 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
848 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
849 ip->i_update_core = 1;
850 timeflags &= ~XFS_ICHGTIME_CHG;
851 }
852
853 /*
854 * Send out timestamp changes that need to be set to the
855 * current time. Not done when called by a DMI function.
856 */
857 if (timeflags && !(flags & ATTR_DMI))
858 xfs_ichgtime(ip, timeflags);
859
860 XFS_STATS_INC(xs_ig_attrchg);
861
862 /*
863 * If this is a synchronous mount, make sure that the
864 * transaction goes to disk before returning to the user.
865 * This is slightly sub-optimal in that truncates require
Nathan Scottc41564b2006-03-29 08:55:14 +1000866 * two sync transactions instead of one for wsync filesystems.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 * One for the truncate and one for the timestamps since we
868 * don't want to change the timestamps unless we're sure the
869 * truncate worked. Truncates are less than 1% of the laddis
870 * mix so this probably isn't worth the trouble to optimize.
871 */
872 code = 0;
873 if (tp) {
874 if (mp->m_flags & XFS_MOUNT_WSYNC)
875 xfs_trans_set_sync(tp);
876
877 code = xfs_trans_commit(tp, commit_flags, NULL);
878 }
879
880 /*
881 * If the (regular) file's mandatory locking mode changed, then
882 * notify the vnode. We do this under the inode lock to prevent
883 * racing calls to vop_vnode_change.
884 */
885 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
886 if (mandlock_before != mandlock_after) {
Nathan Scott67fcaa72006-06-09 17:00:52 +1000887 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_ENF_LOCKING,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 mandlock_after);
889 }
890
891 xfs_iunlock(ip, lock_flags);
892
893 /*
894 * Release any dquot(s) the inode had kept before chown.
895 */
896 XFS_QM_DQRELE(mp, olddquot1);
897 XFS_QM_DQRELE(mp, olddquot2);
898 XFS_QM_DQRELE(mp, udqp);
899 XFS_QM_DQRELE(mp, gdqp);
900
901 if (code) {
902 return code;
903 }
904
905 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
906 !(flags & ATTR_DMI)) {
907 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
908 NULL, DM_RIGHT_NULL, NULL, NULL,
909 0, 0, AT_DELAY_FLAG(flags));
910 }
911 return 0;
912
913 abort_return:
914 commit_flags |= XFS_TRANS_ABORT;
915 /* FALLTHROUGH */
916 error_return:
917 XFS_QM_DQRELE(mp, udqp);
918 XFS_QM_DQRELE(mp, gdqp);
919 if (tp) {
920 xfs_trans_cancel(tp, commit_flags);
921 }
922 if (lock_flags != 0) {
923 xfs_iunlock(ip, lock_flags);
924 }
925 return code;
926}
927
928
929/*
930 * xfs_access
931 * Null conversion from vnode mode bits to inode mode bits, as in efs.
932 */
933STATIC int
934xfs_access(
935 bhv_desc_t *bdp,
936 int mode,
937 cred_t *credp)
938{
939 xfs_inode_t *ip;
940 int error;
941
942 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
943 (inst_t *)__return_address);
944
945 ip = XFS_BHVTOI(bdp);
946 xfs_ilock(ip, XFS_ILOCK_SHARED);
947 error = xfs_iaccess(ip, mode, credp);
948 xfs_iunlock(ip, XFS_ILOCK_SHARED);
949 return error;
950}
951
952
953/*
Nathan Scott7d4fb402006-06-09 15:27:16 +1000954 * The maximum pathlen is 1024 bytes. Since the minimum file system
955 * blocksize is 512 bytes, we can get a max of 2 extents back from
956 * bmapi.
957 */
958#define SYMLINK_MAPS 2
959
960/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 * xfs_readlink
962 *
963 */
964STATIC int
965xfs_readlink(
966 bhv_desc_t *bdp,
967 uio_t *uiop,
968 int ioflags,
969 cred_t *credp)
970{
971 xfs_inode_t *ip;
972 int count;
973 xfs_off_t offset;
974 int pathlen;
Nathan Scott67fcaa72006-06-09 17:00:52 +1000975 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 int error = 0;
977 xfs_mount_t *mp;
978 int nmaps;
979 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
980 xfs_daddr_t d;
981 int byte_cnt;
982 int n;
983 xfs_buf_t *bp;
984
985 vp = BHV_TO_VNODE(bdp);
986 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
987
988 ip = XFS_BHVTOI(bdp);
989 mp = ip->i_mount;
990
991 if (XFS_FORCED_SHUTDOWN(mp))
992 return XFS_ERROR(EIO);
993
994 xfs_ilock(ip, XFS_ILOCK_SHARED);
995
996 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
997
998 offset = uiop->uio_offset;
999 count = uiop->uio_resid;
1000
1001 if (offset < 0) {
1002 error = XFS_ERROR(EINVAL);
1003 goto error_return;
1004 }
1005 if (count <= 0) {
1006 error = 0;
1007 goto error_return;
1008 }
1009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 /*
1011 * See if the symlink is stored inline.
1012 */
1013 pathlen = (int)ip->i_d.di_size;
1014
1015 if (ip->i_df.if_flags & XFS_IFINLINE) {
1016 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1017 }
1018 else {
1019 /*
1020 * Symlink not inline. Call bmap to get it in.
1021 */
1022 nmaps = SYMLINK_MAPS;
1023
1024 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001025 0, NULL, 0, mval, &nmaps, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
1027 if (error) {
1028 goto error_return;
1029 }
1030
1031 for (n = 0; n < nmaps; n++) {
1032 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1033 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1034 bp = xfs_buf_read(mp->m_ddev_targp, d,
1035 BTOBB(byte_cnt), 0);
1036 error = XFS_BUF_GETERROR(bp);
1037 if (error) {
1038 xfs_ioerror_alert("xfs_readlink",
1039 ip->i_mount, bp, XFS_BUF_ADDR(bp));
1040 xfs_buf_relse(bp);
1041 goto error_return;
1042 }
1043 if (pathlen < byte_cnt)
1044 byte_cnt = pathlen;
1045 pathlen -= byte_cnt;
1046
1047 error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1048 xfs_buf_relse (bp);
1049 }
1050
1051 }
1052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 return error;
1056}
1057
1058
1059/*
1060 * xfs_fsync
1061 *
1062 * This is called to sync the inode and its data out to disk.
1063 * We need to hold the I/O lock while flushing the data, and
1064 * the inode lock while flushing the inode. The inode lock CANNOT
1065 * be held while flushing the data, so acquire after we're done
1066 * with that.
1067 */
1068STATIC int
1069xfs_fsync(
1070 bhv_desc_t *bdp,
1071 int flag,
1072 cred_t *credp,
1073 xfs_off_t start,
1074 xfs_off_t stop)
1075{
1076 xfs_inode_t *ip;
1077 xfs_trans_t *tp;
1078 int error;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001079 int log_flushed = 0, changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 vn_trace_entry(BHV_TO_VNODE(bdp),
1082 __FUNCTION__, (inst_t *)__return_address);
1083
1084 ip = XFS_BHVTOI(bdp);
1085
1086 ASSERT(start >= 0 && stop >= -1);
1087
1088 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1089 return XFS_ERROR(EIO);
1090
1091 /*
1092 * We always need to make sure that the required inode state
1093 * is safe on disk. The vnode might be clean but because
1094 * of committed transactions that haven't hit the disk yet.
1095 * Likewise, there could be unflushed non-transactional
1096 * changes to the inode core that have to go to disk.
1097 *
1098 * The following code depends on one assumption: that
1099 * any transaction that changes an inode logs the core
1100 * because it has to change some field in the inode core
1101 * (typically nextents or nblocks). That assumption
1102 * implies that any transactions against an inode will
1103 * catch any non-transactional updates. If inode-altering
1104 * transactions exist that violate this assumption, the
1105 * code breaks. Right now, it figures that if the involved
1106 * update_* field is clear and the inode is unpinned, the
1107 * inode is clean. Either it's been flushed or it's been
1108 * committed and the commit has hit the disk unpinning the inode.
1109 * (Note that xfs_inode_item_format() called at commit clears
1110 * the update_* fields.)
1111 */
1112 xfs_ilock(ip, XFS_ILOCK_SHARED);
1113
1114 /* If we are flushing data then we care about update_size
1115 * being set, otherwise we care about update_core
1116 */
1117 if ((flag & FSYNC_DATA) ?
1118 (ip->i_update_size == 0) :
1119 (ip->i_update_core == 0)) {
1120 /*
1121 * Timestamps/size haven't changed since last inode
1122 * flush or inode transaction commit. That means
1123 * either nothing got written or a transaction
1124 * committed which caught the updates. If the
1125 * latter happened and the transaction hasn't
1126 * hit the disk yet, the inode will be still
1127 * be pinned. If it is, force the log.
1128 */
1129
1130 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1131
1132 if (xfs_ipincount(ip)) {
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001133 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 XFS_LOG_FORCE |
1135 ((flag & FSYNC_WAIT)
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001136 ? XFS_LOG_SYNC : 0),
1137 &log_flushed);
1138 } else {
1139 /*
1140 * If the inode is not pinned and nothing
1141 * has changed we don't need to flush the
1142 * cache.
1143 */
1144 changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 }
1146 error = 0;
1147 } else {
1148 /*
1149 * Kick off a transaction to log the inode
1150 * core to get the updates. Make it
1151 * sync if FSYNC_WAIT is passed in (which
1152 * is done by everybody but specfs). The
1153 * sync transaction will also force the log.
1154 */
1155 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1156 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1157 if ((error = xfs_trans_reserve(tp, 0,
1158 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1159 0, 0, 0))) {
1160 xfs_trans_cancel(tp, 0);
1161 return error;
1162 }
1163 xfs_ilock(ip, XFS_ILOCK_EXCL);
1164
1165 /*
1166 * Note - it's possible that we might have pushed
1167 * ourselves out of the way during trans_reserve
1168 * which would flush the inode. But there's no
1169 * guarantee that the inode buffer has actually
1170 * gone out yet (it's delwri). Plus the buffer
1171 * could be pinned anyway if it's part of an
1172 * inode in another recent transaction. So we
1173 * play it safe and fire off the transaction anyway.
1174 */
1175 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1176 xfs_trans_ihold(tp, ip);
1177 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1178 if (flag & FSYNC_WAIT)
1179 xfs_trans_set_sync(tp);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001180 error = _xfs_trans_commit(tp, 0, NULL, &log_flushed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1183 }
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001184
1185 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1186 /*
1187 * If the log write didn't issue an ordered tag we need
1188 * to flush the disk cache for the data device now.
1189 */
1190 if (!log_flushed)
1191 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1192
1193 /*
1194 * If this inode is on the RT dev we need to flush that
Nathan Scottc41564b2006-03-29 08:55:14 +10001195 * cache as well.
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001196 */
1197 if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)
1198 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1199 }
1200
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 return error;
1202}
1203
1204/*
1205 * This is called by xfs_inactive to free any blocks beyond eof,
1206 * when the link count isn't zero.
1207 */
1208STATIC int
1209xfs_inactive_free_eofblocks(
1210 xfs_mount_t *mp,
1211 xfs_inode_t *ip)
1212{
1213 xfs_trans_t *tp;
1214 int error;
1215 xfs_fileoff_t end_fsb;
1216 xfs_fileoff_t last_fsb;
1217 xfs_filblks_t map_len;
1218 int nimaps;
1219 xfs_bmbt_irec_t imap;
1220
1221 /*
1222 * Figure out if there are any blocks beyond the end
1223 * of the file. If not, then there is nothing to do.
1224 */
1225 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1226 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1227 map_len = last_fsb - end_fsb;
1228 if (map_len <= 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01001229 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
1231 nimaps = 1;
1232 xfs_ilock(ip, XFS_ILOCK_SHARED);
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001233 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, end_fsb, map_len, 0,
1234 NULL, 0, &imap, &nimaps, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1236
1237 if (!error && (nimaps != 0) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001238 (imap.br_startblock != HOLESTARTBLOCK ||
1239 ip->i_delayed_blks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 /*
1241 * Attach the dquots to the inode up front.
1242 */
1243 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001244 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
1246 /*
1247 * There are blocks after the end of file.
1248 * Free them up now by truncating the file to
1249 * its current size.
1250 */
1251 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1252
1253 /*
1254 * Do the xfs_itruncate_start() call before
1255 * reserving any log space because
1256 * itruncate_start will call into the buffer
1257 * cache and we can't
1258 * do that within a transaction.
1259 */
1260 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1261 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1262 ip->i_d.di_size);
1263
1264 error = xfs_trans_reserve(tp, 0,
1265 XFS_ITRUNCATE_LOG_RES(mp),
1266 0, XFS_TRANS_PERM_LOG_RES,
1267 XFS_ITRUNCATE_LOG_COUNT);
1268 if (error) {
1269 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1270 xfs_trans_cancel(tp, 0);
1271 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001272 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
1274
1275 xfs_ilock(ip, XFS_ILOCK_EXCL);
1276 xfs_trans_ijoin(tp, ip,
1277 XFS_IOLOCK_EXCL |
1278 XFS_ILOCK_EXCL);
1279 xfs_trans_ihold(tp, ip);
1280
1281 error = xfs_itruncate_finish(&tp, ip,
1282 ip->i_d.di_size,
1283 XFS_DATA_FORK,
1284 0);
1285 /*
1286 * If we get an error at this point we
1287 * simply don't bother truncating the file.
1288 */
1289 if (error) {
1290 xfs_trans_cancel(tp,
1291 (XFS_TRANS_RELEASE_LOG_RES |
1292 XFS_TRANS_ABORT));
1293 } else {
1294 error = xfs_trans_commit(tp,
1295 XFS_TRANS_RELEASE_LOG_RES,
1296 NULL);
1297 }
1298 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1299 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001300 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301}
1302
1303/*
1304 * Free a symlink that has blocks associated with it.
1305 */
1306STATIC int
1307xfs_inactive_symlink_rmt(
1308 xfs_inode_t *ip,
1309 xfs_trans_t **tpp)
1310{
1311 xfs_buf_t *bp;
1312 int committed;
1313 int done;
1314 int error;
1315 xfs_fsblock_t first_block;
1316 xfs_bmap_free_t free_list;
1317 int i;
1318 xfs_mount_t *mp;
1319 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1320 int nmaps;
1321 xfs_trans_t *ntp;
1322 int size;
1323 xfs_trans_t *tp;
1324
1325 tp = *tpp;
1326 mp = ip->i_mount;
1327 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1328 /*
1329 * We're freeing a symlink that has some
1330 * blocks allocated to it. Free the
1331 * blocks here. We know that we've got
1332 * either 1 or 2 extents and that we can
1333 * free them all in one bunmapi call.
1334 */
1335 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1336 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1337 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1338 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1339 xfs_trans_cancel(tp, 0);
1340 *tpp = NULL;
1341 return error;
1342 }
1343 /*
1344 * Lock the inode, fix the size, and join it to the transaction.
1345 * Hold it so in the normal path, we still have it locked for
1346 * the second transaction. In the error paths we need it
1347 * held so the cancel won't rele it, see below.
1348 */
1349 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1350 size = (int)ip->i_d.di_size;
1351 ip->i_d.di_size = 0;
1352 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1353 xfs_trans_ihold(tp, ip);
1354 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1355 /*
1356 * Find the block(s) so we can inval and unmap them.
1357 */
1358 done = 0;
1359 XFS_BMAP_INIT(&free_list, &first_block);
Tobias Klausere8c96f82006-03-24 03:15:34 -08001360 nmaps = ARRAY_SIZE(mval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1362 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001363 &free_list, NULL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 goto error0;
1365 /*
1366 * Invalidate the block(s).
1367 */
1368 for (i = 0; i < nmaps; i++) {
1369 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1370 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1371 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1372 xfs_trans_binval(tp, bp);
1373 }
1374 /*
1375 * Unmap the dead block(s) to the free_list.
1376 */
1377 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001378 &first_block, &free_list, NULL, &done)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 goto error1;
1380 ASSERT(done);
1381 /*
1382 * Commit the first transaction. This logs the EFI and the inode.
1383 */
1384 if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1385 goto error1;
1386 /*
1387 * The transaction must have been committed, since there were
1388 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1389 * The new tp has the extent freeing and EFDs.
1390 */
1391 ASSERT(committed);
1392 /*
1393 * The first xact was committed, so add the inode to the new one.
1394 * Mark it dirty so it will be logged and moved forward in the log as
1395 * part of every commit.
1396 */
1397 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1398 xfs_trans_ihold(tp, ip);
1399 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1400 /*
1401 * Get a new, empty transaction to return to our caller.
1402 */
1403 ntp = xfs_trans_dup(tp);
1404 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001405 * Commit the transaction containing extent freeing and EFDs.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 * If we get an error on the commit here or on the reserve below,
1407 * we need to unlock the inode since the new transaction doesn't
1408 * have the inode attached.
1409 */
1410 error = xfs_trans_commit(tp, 0, NULL);
1411 tp = ntp;
1412 if (error) {
1413 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1414 goto error0;
1415 }
1416 /*
1417 * Remove the memory for extent descriptions (just bookkeeping).
1418 */
1419 if (ip->i_df.if_bytes)
1420 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1421 ASSERT(ip->i_df.if_bytes == 0);
1422 /*
1423 * Put an itruncate log reservation in the new transaction
1424 * for our caller.
1425 */
1426 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1427 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1428 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1429 goto error0;
1430 }
1431 /*
1432 * Return with the inode locked but not joined to the transaction.
1433 */
1434 *tpp = tp;
1435 return 0;
1436
1437 error1:
1438 xfs_bmap_cancel(&free_list);
1439 error0:
1440 /*
1441 * Have to come here with the inode locked and either
1442 * (held and in the transaction) or (not in the transaction).
1443 * If the inode isn't held then cancel would iput it, but
1444 * that's wrong since this is inactive and the vnode ref
1445 * count is 0 already.
1446 * Cancel won't do anything to the inode if held, but it still
1447 * needs to be locked until the cancel is done, if it was
1448 * joined to the transaction.
1449 */
1450 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1451 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1452 *tpp = NULL;
1453 return error;
1454
1455}
1456
1457STATIC int
1458xfs_inactive_symlink_local(
1459 xfs_inode_t *ip,
1460 xfs_trans_t **tpp)
1461{
1462 int error;
1463
1464 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1465 /*
1466 * We're freeing a symlink which fit into
1467 * the inode. Just free the memory used
1468 * to hold the old symlink.
1469 */
1470 error = xfs_trans_reserve(*tpp, 0,
1471 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1472 0, XFS_TRANS_PERM_LOG_RES,
1473 XFS_ITRUNCATE_LOG_COUNT);
1474
1475 if (error) {
1476 xfs_trans_cancel(*tpp, 0);
1477 *tpp = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +01001478 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 }
1480 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1481
1482 /*
1483 * Zero length symlinks _can_ exist.
1484 */
1485 if (ip->i_df.if_bytes > 0) {
1486 xfs_idata_realloc(ip,
1487 -(ip->i_df.if_bytes),
1488 XFS_DATA_FORK);
1489 ASSERT(ip->i_df.if_bytes == 0);
1490 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001491 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492}
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494STATIC int
1495xfs_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);
Jesper Juhl014c2542006-01-15 02:37:08 +01001514 return error; /* goto out */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 }
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);
Jesper Juhl014c2542006-01-15 02:37:08 +01001527 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 }
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;
Jesper Juhl014c2542006-01-15 02:37:08 +01001538 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539}
1540
1541STATIC int
1542xfs_release(
1543 bhv_desc_t *bdp)
1544{
1545 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001546 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 xfs_mount_t *mp;
1548 int error;
1549
1550 vp = BHV_TO_VNODE(bdp);
1551 ip = XFS_BHVTOI(bdp);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001552 mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
Nathan Scott7d4fb402006-06-09 15:27:16 +10001554 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
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
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 if (ip->i_d.di_nlink != 0) {
1568 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001569 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1570 ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001572 (!(ip->i_d.di_flags &
1573 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001575 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 /* Update linux inode block count after free above */
Nathan Scottec86dc02006-03-17 17:25:36 +11001577 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 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 */
1593STATIC int
1594xfs_inactive(
1595 bhv_desc_t *bdp,
1596 cred_t *credp)
1597{
1598 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001599 bhv_vnode_t *vp;
1600 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 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) &&
Yingping Lu1661dc82006-01-11 15:29:39 +11001630 ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0) ||
1631 (ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1633
1634 mp = ip->i_mount;
1635
1636 if (ip->i_d.di_nlink == 0 &&
1637 DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1638 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1639 }
1640
1641 error = 0;
1642
1643 /* If this is a read-only mount, don't do this (would generate I/O) */
1644 if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1645 goto out;
1646
1647 if (ip->i_d.di_nlink != 0) {
1648 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001649 ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0 ||
1650 ip->i_delayed_blks > 0)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001651 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1652 (!(ip->i_d.di_flags &
1653 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1654 (ip->i_delayed_blks != 0)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 if ((error = xfs_inactive_free_eofblocks(mp, ip)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001656 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 /* Update linux inode block count after free above */
Nathan Scottec86dc02006-03-17 17:25:36 +11001658 vn_to_inode(vp)->i_blocks = XFS_FSB_TO_BB(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 ip->i_d.di_nblocks + ip->i_delayed_blks);
1660 }
1661 goto out;
1662 }
1663
1664 ASSERT(ip->i_d.di_nlink == 0);
1665
1666 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001667 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1670 if (truncate) {
1671 /*
1672 * Do the xfs_itruncate_start() call before
1673 * reserving any log space because itruncate_start
1674 * will call into the buffer cache and we can't
1675 * do that within a transaction.
1676 */
1677 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1678
1679 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1680
1681 error = xfs_trans_reserve(tp, 0,
1682 XFS_ITRUNCATE_LOG_RES(mp),
1683 0, XFS_TRANS_PERM_LOG_RES,
1684 XFS_ITRUNCATE_LOG_COUNT);
1685 if (error) {
1686 /* Don't call itruncate_cleanup */
1687 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1688 xfs_trans_cancel(tp, 0);
1689 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001690 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 }
1692
1693 xfs_ilock(ip, XFS_ILOCK_EXCL);
1694 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1695 xfs_trans_ihold(tp, ip);
1696
1697 /*
1698 * normally, we have to run xfs_itruncate_finish sync.
1699 * But if filesystem is wsync and we're in the inactive
1700 * path, then we know that nlink == 0, and that the
1701 * xaction that made nlink == 0 is permanently committed
1702 * since xfs_remove runs as a synchronous transaction.
1703 */
1704 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1705 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1706
1707 if (error) {
1708 xfs_trans_cancel(tp,
1709 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1710 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001711 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 }
1713 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1714
1715 /*
1716 * If we get an error while cleaning up a
1717 * symlink we bail out.
1718 */
1719 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1720 xfs_inactive_symlink_rmt(ip, &tp) :
1721 xfs_inactive_symlink_local(ip, &tp);
1722
1723 if (error) {
1724 ASSERT(tp == NULL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001725 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 }
1727
1728 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1729 xfs_trans_ihold(tp, ip);
1730 } else {
1731 error = xfs_trans_reserve(tp, 0,
1732 XFS_IFREE_LOG_RES(mp),
1733 0, XFS_TRANS_PERM_LOG_RES,
1734 XFS_INACTIVE_LOG_COUNT);
1735 if (error) {
1736 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1737 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001738 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 }
1740
1741 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1742 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1743 xfs_trans_ihold(tp, ip);
1744 }
1745
1746 /*
1747 * If there are attributes associated with the file
1748 * then blow them away now. The code calls a routine
1749 * that recursively deconstructs the attribute fork.
1750 * We need to just commit the current transaction
1751 * because we can't use it for xfs_attr_inactive().
1752 */
1753 if (ip->i_d.di_anextents > 0) {
1754 error = xfs_inactive_attrs(ip, &tp);
1755 /*
1756 * If we got an error, the transaction is already
1757 * cancelled, and the inode is unlocked. Just get out.
1758 */
1759 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001760 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 } else if (ip->i_afp) {
1762 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1763 }
1764
1765 /*
1766 * Free the inode.
1767 */
1768 XFS_BMAP_INIT(&free_list, &first_block);
1769 error = xfs_ifree(tp, ip, &free_list);
1770 if (error) {
1771 /*
1772 * If we fail to free the inode, shut down. The cancel
1773 * might do that, we need to make sure. Otherwise the
1774 * inode might be lost for a long time or forever.
1775 */
1776 if (!XFS_FORCED_SHUTDOWN(mp)) {
1777 cmn_err(CE_NOTE,
1778 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1779 error, mp->m_fsname);
Nathan Scott7d04a332006-06-09 14:58:38 +10001780 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 }
1782 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1783 } else {
1784 /*
1785 * Credit the quota account(s). The inode is gone.
1786 */
1787 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1788
1789 /*
1790 * Just ignore errors at this point. There is
1791 * nothing we can do except to try to keep going.
1792 */
1793 (void) xfs_bmap_finish(&tp, &free_list, first_block,
1794 &committed);
1795 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1796 }
1797 /*
1798 * Release the dquots held by inode, if any.
1799 */
1800 XFS_QM_DQDETACH(mp, ip);
1801
1802 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1803
1804 out:
1805 return VN_INACTIVE_CACHE;
1806}
1807
1808
1809/*
1810 * xfs_lookup
1811 */
1812STATIC int
1813xfs_lookup(
1814 bhv_desc_t *dir_bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +10001815 bhv_vname_t *dentry,
Nathan Scott67fcaa72006-06-09 17:00:52 +10001816 bhv_vnode_t **vpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 int flags,
Nathan Scott67fcaa72006-06-09 17:00:52 +10001818 bhv_vnode_t *rdir,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 cred_t *credp)
1820{
1821 xfs_inode_t *dp, *ip;
1822 xfs_ino_t e_inum;
1823 int error;
1824 uint lock_mode;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001825 bhv_vnode_t *dir_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827 dir_vp = BHV_TO_VNODE(dir_bdp);
1828 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1829
1830 dp = XFS_BHVTOI(dir_bdp);
1831
1832 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1833 return XFS_ERROR(EIO);
1834
1835 lock_mode = xfs_ilock_map_shared(dp);
1836 error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1837 if (!error) {
1838 *vpp = XFS_ITOV(ip);
1839 ITRACE(ip);
1840 }
1841 xfs_iunlock_map_shared(dp, lock_mode);
1842 return error;
1843}
1844
1845
1846/*
1847 * xfs_create (create a new file).
1848 */
1849STATIC int
1850xfs_create(
1851 bhv_desc_t *dir_bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +10001852 bhv_vname_t *dentry,
1853 bhv_vattr_t *vap,
Nathan Scott67fcaa72006-06-09 17:00:52 +10001854 bhv_vnode_t **vpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 cred_t *credp)
1856{
1857 char *name = VNAME(dentry);
Nathan Scott67fcaa72006-06-09 17:00:52 +10001858 bhv_vnode_t *dir_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 xfs_inode_t *dp, *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001860 bhv_vnode_t *vp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 xfs_trans_t *tp;
1862 xfs_mount_t *mp;
1863 xfs_dev_t rdev;
1864 int error;
1865 xfs_bmap_free_t free_list;
1866 xfs_fsblock_t first_block;
1867 boolean_t dp_joined_to_trans;
1868 int dm_event_sent = 0;
1869 uint cancel_flags;
1870 int committed;
1871 xfs_prid_t prid;
1872 struct xfs_dquot *udqp, *gdqp;
1873 uint resblks;
1874 int dm_di_mode;
1875 int namelen;
1876
1877 ASSERT(!*vpp);
1878 dir_vp = BHV_TO_VNODE(dir_bdp);
1879 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1880
1881 dp = XFS_BHVTOI(dir_bdp);
1882 mp = dp->i_mount;
1883
Christoph Hellwig0432dab2005-09-02 16:46:51 +10001884 dm_di_mode = vap->va_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 namelen = VNAMELEN(dentry);
1886
1887 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1888 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1889 dir_vp, DM_RIGHT_NULL, NULL,
1890 DM_RIGHT_NULL, name, NULL,
1891 dm_di_mode, 0, 0);
1892
1893 if (error)
1894 return error;
1895 dm_event_sent = 1;
1896 }
1897
1898 if (XFS_FORCED_SHUTDOWN(mp))
1899 return XFS_ERROR(EIO);
1900
1901 /* Return through std_return after this point. */
1902
1903 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10001904 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1905 prid = dp->i_d.di_projid;
1906 else if (vap->va_mask & XFS_AT_PROJID)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 prid = (xfs_prid_t)vap->va_projid;
1908 else
1909 prid = (xfs_prid_t)dfltprid;
1910
1911 /*
1912 * Make sure that we have allocated dquot(s) on disk.
1913 */
1914 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001915 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1917 if (error)
1918 goto std_return;
1919
1920 ip = NULL;
1921 dp_joined_to_trans = B_FALSE;
1922
1923 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1924 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1925 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1926 /*
1927 * Initially assume that the file does not exist and
1928 * reserve the resources for that case. If that is not
1929 * the case we'll drop the one we have and get a more
1930 * appropriate transaction later.
1931 */
1932 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1933 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1934 if (error == ENOSPC) {
1935 resblks = 0;
1936 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1937 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1938 }
1939 if (error) {
1940 cancel_flags = 0;
1941 dp = NULL;
1942 goto error_return;
1943 }
1944
1945 xfs_ilock(dp, XFS_ILOCK_EXCL);
1946
1947 XFS_BMAP_INIT(&free_list, &first_block);
1948
1949 ASSERT(ip == NULL);
1950
1951 /*
1952 * Reserve disk quota and the inode.
1953 */
1954 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1955 if (error)
1956 goto error_return;
1957
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001958 if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 goto error_return;
1960 rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
Christoph Hellwig0432dab2005-09-02 16:46:51 +10001961 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 rdev, credp, prid, resblks > 0,
1963 &ip, &committed);
1964 if (error) {
1965 if (error == ENOSPC)
1966 goto error_return;
1967 goto abort_return;
1968 }
1969 ITRACE(ip);
1970
1971 /*
1972 * At this point, we've gotten a newly allocated inode.
1973 * It is locked (and joined to the transaction).
1974 */
1975
1976 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1977
1978 /*
1979 * Now we join the directory inode to the transaction.
1980 * We do not do it earlier because xfs_dir_ialloc
1981 * might commit the previous transaction (and release
1982 * all the locks).
1983 */
1984
1985 VN_HOLD(dir_vp);
1986 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1987 dp_joined_to_trans = B_TRUE;
1988
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001989 error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino,
1990 &first_block, &free_list, resblks ?
1991 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 if (error) {
1993 ASSERT(error != ENOSPC);
1994 goto abort_return;
1995 }
1996 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1997 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1998
1999 /*
2000 * If this is a synchronous mount, make sure that the
2001 * create transaction goes to disk before returning to
2002 * the user.
2003 */
2004 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2005 xfs_trans_set_sync(tp);
2006 }
2007
2008 dp->i_gen++;
2009
2010 /*
2011 * Attach the dquot(s) to the inodes and modify them incore.
2012 * These ids of the inode couldn't have changed since the new
2013 * inode has been locked ever since it was created.
2014 */
2015 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2016
2017 /*
2018 * xfs_trans_commit normally decrements the vnode ref count
2019 * when it unlocks the inode. Since we want to return the
2020 * vnode to the caller, we bump the vnode ref count now.
2021 */
2022 IHOLD(ip);
2023 vp = XFS_ITOV(ip);
2024
2025 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2026 if (error) {
2027 xfs_bmap_cancel(&free_list);
2028 goto abort_rele;
2029 }
2030
2031 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2032 if (error) {
2033 IRELE(ip);
2034 tp = NULL;
2035 goto error_return;
2036 }
2037
2038 XFS_QM_DQRELE(mp, udqp);
2039 XFS_QM_DQRELE(mp, gdqp);
2040
2041 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10002042 * Propagate the fact that the vnode changed after the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 * xfs_inode locks have been released.
2044 */
Nathan Scott67fcaa72006-06-09 17:00:52 +10002045 bhv_vop_vnode_change(vp, VCHANGE_FLAGS_TRUNCATED, 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047 *vpp = vp;
2048
2049 /* Fallthrough to std_return with error = 0 */
2050
2051std_return:
2052 if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2053 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2054 DM_EVENT_POSTCREATE)) {
2055 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2056 dir_vp, DM_RIGHT_NULL,
2057 *vpp ? vp:NULL,
2058 DM_RIGHT_NULL, name, NULL,
2059 dm_di_mode, error, 0);
2060 }
2061 return error;
2062
2063 abort_return:
2064 cancel_flags |= XFS_TRANS_ABORT;
2065 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
Jesper Juhl014c2542006-01-15 02:37:08 +01002067 error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 if (tp != NULL)
2069 xfs_trans_cancel(tp, cancel_flags);
2070
2071 if (!dp_joined_to_trans && (dp != NULL))
2072 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2073 XFS_QM_DQRELE(mp, udqp);
2074 XFS_QM_DQRELE(mp, gdqp);
2075
2076 goto std_return;
2077
2078 abort_rele:
2079 /*
2080 * Wait until after the current transaction is aborted to
2081 * release the inode. This prevents recursive transactions
2082 * and deadlocks from xfs_inactive.
2083 */
2084 cancel_flags |= XFS_TRANS_ABORT;
2085 xfs_trans_cancel(tp, cancel_flags);
2086 IRELE(ip);
2087
2088 XFS_QM_DQRELE(mp, udqp);
2089 XFS_QM_DQRELE(mp, gdqp);
2090
2091 goto std_return;
2092}
2093
2094#ifdef DEBUG
2095/*
2096 * Some counters to see if (and how often) we are hitting some deadlock
2097 * prevention code paths.
2098 */
2099
2100int xfs_rm_locks;
2101int xfs_rm_lock_delays;
2102int xfs_rm_attempts;
2103#endif
2104
2105/*
2106 * The following routine will lock the inodes associated with the
2107 * directory and the named entry in the directory. The locks are
2108 * acquired in increasing inode number.
2109 *
2110 * If the entry is "..", then only the directory is locked. The
2111 * vnode ref count will still include that from the .. entry in
2112 * this case.
2113 *
2114 * There is a deadlock we need to worry about. If the locked directory is
2115 * in the AIL, it might be blocking up the log. The next inode we lock
2116 * could be already locked by another thread waiting for log space (e.g
2117 * a permanent log reservation with a long running transaction (see
2118 * xfs_itruncate_finish)). To solve this, we must check if the directory
2119 * is in the ail and use lock_nowait. If we can't lock, we need to
2120 * drop the inode lock on the directory and try again. xfs_iunlock will
2121 * potentially push the tail if we were holding up the log.
2122 */
2123STATIC int
2124xfs_lock_dir_and_entry(
2125 xfs_inode_t *dp,
Nathan Scott8285fb52006-06-09 17:07:12 +10002126 bhv_vname_t *dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 xfs_inode_t *ip) /* inode of entry 'name' */
2128{
2129 int attempts;
2130 xfs_ino_t e_inum;
2131 xfs_inode_t *ips[2];
2132 xfs_log_item_t *lp;
2133
2134#ifdef DEBUG
2135 xfs_rm_locks++;
2136#endif
2137 attempts = 0;
2138
2139again:
2140 xfs_ilock(dp, XFS_ILOCK_EXCL);
2141
2142 e_inum = ip->i_ino;
2143
2144 ITRACE(ip);
2145
2146 /*
2147 * We want to lock in increasing inum. Since we've already
2148 * acquired the lock on the directory, we may need to release
2149 * if if the inum of the entry turns out to be less.
2150 */
2151 if (e_inum > dp->i_ino) {
2152 /*
2153 * We are already in the right order, so just
2154 * lock on the inode of the entry.
2155 * We need to use nowait if dp is in the AIL.
2156 */
2157
2158 lp = (xfs_log_item_t *)dp->i_itemp;
2159 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2160 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2161 attempts++;
2162#ifdef DEBUG
2163 xfs_rm_attempts++;
2164#endif
2165
2166 /*
2167 * Unlock dp and try again.
2168 * xfs_iunlock will try to push the tail
2169 * if the inode is in the AIL.
2170 */
2171
2172 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2173
2174 if ((attempts % 5) == 0) {
2175 delay(1); /* Don't just spin the CPU */
2176#ifdef DEBUG
2177 xfs_rm_lock_delays++;
2178#endif
2179 }
2180 goto again;
2181 }
2182 } else {
2183 xfs_ilock(ip, XFS_ILOCK_EXCL);
2184 }
2185 } else if (e_inum < dp->i_ino) {
2186 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2187
2188 ips[0] = ip;
2189 ips[1] = dp;
2190 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2191 }
2192 /* else e_inum == dp->i_ino */
2193 /* This can happen if we're asked to lock /x/..
2194 * the entry is "..", which is also the parent directory.
2195 */
2196
2197 return 0;
2198}
2199
2200#ifdef DEBUG
2201int xfs_locked_n;
2202int xfs_small_retries;
2203int xfs_middle_retries;
2204int xfs_lots_retries;
2205int xfs_lock_delays;
2206#endif
2207
2208/*
2209 * The following routine will lock n inodes in exclusive mode.
2210 * We assume the caller calls us with the inodes in i_ino order.
2211 *
2212 * We need to detect deadlock where an inode that we lock
2213 * is in the AIL and we start waiting for another inode that is locked
2214 * by a thread in a long running transaction (such as truncate). This can
2215 * result in deadlock since the long running trans might need to wait
2216 * for the inode we just locked in order to push the tail and free space
2217 * in the log.
2218 */
2219void
2220xfs_lock_inodes(
2221 xfs_inode_t **ips,
2222 int inodes,
2223 int first_locked,
2224 uint lock_mode)
2225{
2226 int attempts = 0, i, j, try_lock;
2227 xfs_log_item_t *lp;
2228
2229 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2230
2231 if (first_locked) {
2232 try_lock = 1;
2233 i = 1;
2234 } else {
2235 try_lock = 0;
2236 i = 0;
2237 }
2238
2239again:
2240 for (; i < inodes; i++) {
2241 ASSERT(ips[i]);
2242
2243 if (i && (ips[i] == ips[i-1])) /* Already locked */
2244 continue;
2245
2246 /*
2247 * If try_lock is not set yet, make sure all locked inodes
2248 * are not in the AIL.
2249 * If any are, set try_lock to be used later.
2250 */
2251
2252 if (!try_lock) {
2253 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2254 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2255 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2256 try_lock++;
2257 }
2258 }
2259 }
2260
2261 /*
2262 * If any of the previous locks we have locked is in the AIL,
2263 * we must TRY to get the second and subsequent locks. If
2264 * we can't get any, we must release all we have
2265 * and try again.
2266 */
2267
2268 if (try_lock) {
2269 /* try_lock must be 0 if i is 0. */
2270 /*
2271 * try_lock means we have an inode locked
2272 * that is in the AIL.
2273 */
2274 ASSERT(i != 0);
2275 if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2276 attempts++;
2277
2278 /*
2279 * Unlock all previous guys and try again.
2280 * xfs_iunlock will try to push the tail
2281 * if the inode is in the AIL.
2282 */
2283
2284 for(j = i - 1; j >= 0; j--) {
2285
2286 /*
2287 * Check to see if we've already
2288 * unlocked this one.
2289 * Not the first one going back,
2290 * and the inode ptr is the same.
2291 */
2292 if ((j != (i - 1)) && ips[j] ==
2293 ips[j+1])
2294 continue;
2295
2296 xfs_iunlock(ips[j], lock_mode);
2297 }
2298
2299 if ((attempts % 5) == 0) {
2300 delay(1); /* Don't just spin the CPU */
2301#ifdef DEBUG
2302 xfs_lock_delays++;
2303#endif
2304 }
2305 i = 0;
2306 try_lock = 0;
2307 goto again;
2308 }
2309 } else {
2310 xfs_ilock(ips[i], lock_mode);
2311 }
2312 }
2313
2314#ifdef DEBUG
2315 if (attempts) {
2316 if (attempts < 5) xfs_small_retries++;
2317 else if (attempts < 100) xfs_middle_retries++;
2318 else xfs_lots_retries++;
2319 } else {
2320 xfs_locked_n++;
2321 }
2322#endif
2323}
2324
2325#ifdef DEBUG
2326#define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2327int remove_which_error_return = 0;
2328#else /* ! DEBUG */
2329#define REMOVE_DEBUG_TRACE(x)
2330#endif /* ! DEBUG */
2331
2332
2333/*
2334 * xfs_remove
2335 *
2336 */
2337STATIC int
2338xfs_remove(
2339 bhv_desc_t *dir_bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +10002340 bhv_vname_t *dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 cred_t *credp)
2342{
Nathan Scott67fcaa72006-06-09 17:00:52 +10002343 bhv_vnode_t *dir_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 char *name = VNAME(dentry);
2345 xfs_inode_t *dp, *ip;
2346 xfs_trans_t *tp = NULL;
2347 xfs_mount_t *mp;
2348 int error = 0;
2349 xfs_bmap_free_t free_list;
2350 xfs_fsblock_t first_block;
2351 int cancel_flags;
2352 int committed;
2353 int dm_di_mode = 0;
2354 int link_zero;
2355 uint resblks;
2356 int namelen;
2357
2358 dir_vp = BHV_TO_VNODE(dir_bdp);
2359 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2360
2361 dp = XFS_BHVTOI(dir_bdp);
2362 mp = dp->i_mount;
2363
2364 if (XFS_FORCED_SHUTDOWN(mp))
2365 return XFS_ERROR(EIO);
2366
2367 namelen = VNAMELEN(dentry);
2368
Vlad Apostolov17370092006-09-28 11:02:30 +10002369 if (!xfs_get_dir_entry(dentry, &ip)) {
2370 dm_di_mode = ip->i_d.di_mode;
2371 IRELE(ip);
2372 }
2373
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2375 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2376 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
Vlad Apostolov17370092006-09-28 11:02:30 +10002377 name, NULL, dm_di_mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 if (error)
2379 return error;
2380 }
2381
2382 /* From this point on, return through std_return */
2383 ip = NULL;
2384
2385 /*
2386 * We need to get a reference to ip before we get our log
2387 * reservation. The reason for this is that we cannot call
2388 * xfs_iget for an inode for which we do not have a reference
2389 * once we've acquired a log reservation. This is because the
2390 * inode we are trying to get might be in xfs_inactive going
2391 * for a log reservation. Since we'll have to wait for the
2392 * inactive code to complete before returning from xfs_iget,
2393 * we need to make sure that we don't have log space reserved
Nathan Scottc41564b2006-03-29 08:55:14 +10002394 * when we call xfs_iget. Instead we get an unlocked reference
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 * to the inode before getting our log reservation.
2396 */
2397 error = xfs_get_dir_entry(dentry, &ip);
2398 if (error) {
2399 REMOVE_DEBUG_TRACE(__LINE__);
2400 goto std_return;
2401 }
2402
2403 dm_di_mode = ip->i_d.di_mode;
2404
2405 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2406
2407 ITRACE(ip);
2408
2409 error = XFS_QM_DQATTACH(mp, dp, 0);
2410 if (!error && dp != ip)
2411 error = XFS_QM_DQATTACH(mp, ip, 0);
2412 if (error) {
2413 REMOVE_DEBUG_TRACE(__LINE__);
2414 IRELE(ip);
2415 goto std_return;
2416 }
2417
2418 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2419 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2420 /*
2421 * We try to get the real space reservation first,
2422 * allowing for directory btree deletion(s) implying
2423 * possible bmap insert(s). If we can't get the space
2424 * reservation then we use 0 instead, and avoid the bmap
2425 * btree insert(s) in the directory code by, if the bmap
2426 * insert tries to happen, instead trimming the LAST
2427 * block from the directory.
2428 */
2429 resblks = XFS_REMOVE_SPACE_RES(mp);
2430 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2431 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2432 if (error == ENOSPC) {
2433 resblks = 0;
2434 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2435 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2436 }
2437 if (error) {
2438 ASSERT(error != ENOSPC);
2439 REMOVE_DEBUG_TRACE(__LINE__);
2440 xfs_trans_cancel(tp, 0);
2441 IRELE(ip);
2442 return error;
2443 }
2444
2445 error = xfs_lock_dir_and_entry(dp, dentry, ip);
2446 if (error) {
2447 REMOVE_DEBUG_TRACE(__LINE__);
2448 xfs_trans_cancel(tp, cancel_flags);
2449 IRELE(ip);
2450 goto std_return;
2451 }
2452
2453 /*
2454 * At this point, we've gotten both the directory and the entry
2455 * inodes locked.
2456 */
2457 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2458 if (dp != ip) {
2459 /*
2460 * Increment vnode ref count only in this case since
2461 * there's an extra vnode reference in the case where
2462 * dp == ip.
2463 */
2464 IHOLD(dp);
2465 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2466 }
2467
2468 /*
2469 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2470 */
2471 XFS_BMAP_INIT(&free_list, &first_block);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002472 error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino,
2473 &first_block, &free_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 if (error) {
2475 ASSERT(error != ENOENT);
2476 REMOVE_DEBUG_TRACE(__LINE__);
2477 goto error1;
2478 }
2479 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2480
2481 dp->i_gen++;
2482 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2483
2484 error = xfs_droplink(tp, ip);
2485 if (error) {
2486 REMOVE_DEBUG_TRACE(__LINE__);
2487 goto error1;
2488 }
2489
2490 /* Determine if this is the last link while
2491 * we are in the transaction.
2492 */
2493 link_zero = (ip)->i_d.di_nlink==0;
2494
2495 /*
2496 * Take an extra ref on the inode so that it doesn't
2497 * go to xfs_inactive() from within the commit.
2498 */
2499 IHOLD(ip);
2500
2501 /*
2502 * If this is a synchronous mount, make sure that the
2503 * remove transaction goes to disk before returning to
2504 * the user.
2505 */
2506 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2507 xfs_trans_set_sync(tp);
2508 }
2509
2510 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2511 if (error) {
2512 REMOVE_DEBUG_TRACE(__LINE__);
2513 goto error_rele;
2514 }
2515
2516 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2517 if (error) {
2518 IRELE(ip);
2519 goto std_return;
2520 }
2521
2522 /*
2523 * Before we drop our extra reference to the inode, purge it
2524 * from the refcache if it is there. By waiting until afterwards
2525 * to do the IRELE, we ensure that we won't go inactive in the
2526 * xfs_refcache_purge_ip routine (although that would be OK).
2527 */
2528 xfs_refcache_purge_ip(ip);
2529
2530 vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2531
2532 /*
2533 * Let interposed file systems know about removed links.
2534 */
Nathan Scott67fcaa72006-06-09 17:00:52 +10002535 bhv_vop_link_removed(XFS_ITOV(ip), dir_vp, link_zero);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
2537 IRELE(ip);
2538
2539/* Fall through to std_return with error = 0 */
2540 std_return:
2541 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2542 DM_EVENT_POSTREMOVE)) {
2543 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2544 dir_vp, DM_RIGHT_NULL,
2545 NULL, DM_RIGHT_NULL,
2546 name, NULL, dm_di_mode, error, 0);
2547 }
2548 return error;
2549
2550 error1:
2551 xfs_bmap_cancel(&free_list);
2552 cancel_flags |= XFS_TRANS_ABORT;
2553 xfs_trans_cancel(tp, cancel_flags);
2554 goto std_return;
2555
2556 error_rele:
2557 /*
2558 * In this case make sure to not release the inode until after
2559 * the current transaction is aborted. Releasing it beforehand
2560 * can cause us to go to xfs_inactive and start a recursive
2561 * transaction which can easily deadlock with the current one.
2562 */
2563 xfs_bmap_cancel(&free_list);
2564 cancel_flags |= XFS_TRANS_ABORT;
2565 xfs_trans_cancel(tp, cancel_flags);
2566
2567 /*
2568 * Before we drop our extra reference to the inode, purge it
2569 * from the refcache if it is there. By waiting until afterwards
2570 * to do the IRELE, we ensure that we won't go inactive in the
2571 * xfs_refcache_purge_ip routine (although that would be OK).
2572 */
2573 xfs_refcache_purge_ip(ip);
2574
2575 IRELE(ip);
2576
2577 goto std_return;
2578}
2579
2580
2581/*
2582 * xfs_link
2583 *
2584 */
2585STATIC int
2586xfs_link(
2587 bhv_desc_t *target_dir_bdp,
Nathan Scott67fcaa72006-06-09 17:00:52 +10002588 bhv_vnode_t *src_vp,
Nathan Scott8285fb52006-06-09 17:07:12 +10002589 bhv_vname_t *dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 cred_t *credp)
2591{
2592 xfs_inode_t *tdp, *sip;
2593 xfs_trans_t *tp;
2594 xfs_mount_t *mp;
2595 xfs_inode_t *ips[2];
2596 int error;
2597 xfs_bmap_free_t free_list;
2598 xfs_fsblock_t first_block;
2599 int cancel_flags;
2600 int committed;
Nathan Scott67fcaa72006-06-09 17:00:52 +10002601 bhv_vnode_t *target_dir_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 int resblks;
2603 char *target_name = VNAME(dentry);
2604 int target_namelen;
2605
2606 target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2607 vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2608 vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2609
2610 target_namelen = VNAMELEN(dentry);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002611 ASSERT(!VN_ISDIR(src_vp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612
Christoph Hellwig75e17b32006-01-11 20:58:44 +11002613 sip = xfs_vtoi(src_vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 tdp = XFS_BHVTOI(target_dir_bdp);
2615 mp = tdp->i_mount;
2616 if (XFS_FORCED_SHUTDOWN(mp))
2617 return XFS_ERROR(EIO);
2618
2619 if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2620 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2621 target_dir_vp, DM_RIGHT_NULL,
2622 src_vp, DM_RIGHT_NULL,
2623 target_name, NULL, 0, 0, 0);
2624 if (error)
2625 return error;
2626 }
2627
2628 /* Return through std_return after this point. */
2629
2630 error = XFS_QM_DQATTACH(mp, sip, 0);
2631 if (!error && sip != tdp)
2632 error = XFS_QM_DQATTACH(mp, tdp, 0);
2633 if (error)
2634 goto std_return;
2635
2636 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2637 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2638 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2639 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2640 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2641 if (error == ENOSPC) {
2642 resblks = 0;
2643 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2644 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2645 }
2646 if (error) {
2647 cancel_flags = 0;
2648 goto error_return;
2649 }
2650
2651 if (sip->i_ino < tdp->i_ino) {
2652 ips[0] = sip;
2653 ips[1] = tdp;
2654 } else {
2655 ips[0] = tdp;
2656 ips[1] = sip;
2657 }
2658
2659 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2660
2661 /*
2662 * Increment vnode ref counts since xfs_trans_commit &
2663 * xfs_trans_cancel will both unlock the inodes and
2664 * decrement the associated ref counts.
2665 */
2666 VN_HOLD(src_vp);
2667 VN_HOLD(target_dir_vp);
2668 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2669 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2670
2671 /*
2672 * If the source has too many links, we can't make any more to it.
2673 */
2674 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2675 error = XFS_ERROR(EMLINK);
2676 goto error_return;
2677 }
2678
Nathan Scott365ca832005-06-21 15:39:12 +10002679 /*
2680 * If we are using project inheritance, we only allow hard link
2681 * creation in our tree when the project IDs are the same; else
2682 * the tree quota mechanism could be circumvented.
2683 */
2684 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2685 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
Nathan Scottb1ecdda2006-05-08 19:51:42 +10002686 error = XFS_ERROR(EXDEV);
Nathan Scott365ca832005-06-21 15:39:12 +10002687 goto error_return;
2688 }
2689
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 if (resblks == 0 &&
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002691 (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 goto error_return;
2693
2694 XFS_BMAP_INIT(&free_list, &first_block);
2695
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002696 error = xfs_dir_createname(tp, tdp, target_name, target_namelen,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 sip->i_ino, &first_block, &free_list,
2698 resblks);
2699 if (error)
2700 goto abort_return;
2701 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2702 tdp->i_gen++;
2703 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2704
2705 error = xfs_bumplink(tp, sip);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002706 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 goto abort_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708
2709 /*
2710 * If this is a synchronous mount, make sure that the
2711 * link transaction goes to disk before returning to
2712 * the user.
2713 */
2714 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2715 xfs_trans_set_sync(tp);
2716 }
2717
2718 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2719 if (error) {
2720 xfs_bmap_cancel(&free_list);
2721 goto abort_return;
2722 }
2723
2724 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002725 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
2728 /* Fall through to std_return with error = 0. */
2729std_return:
2730 if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2731 DM_EVENT_POSTLINK)) {
2732 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2733 target_dir_vp, DM_RIGHT_NULL,
2734 src_vp, DM_RIGHT_NULL,
2735 target_name, NULL, 0, error, 0);
2736 }
2737 return error;
2738
2739 abort_return:
2740 cancel_flags |= XFS_TRANS_ABORT;
2741 /* FALLTHROUGH */
Jesper Juhl014c2542006-01-15 02:37:08 +01002742
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 error_return:
2744 xfs_trans_cancel(tp, cancel_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 goto std_return;
2746}
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002747
2748
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749/*
2750 * xfs_mkdir
2751 *
2752 */
2753STATIC int
2754xfs_mkdir(
2755 bhv_desc_t *dir_bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +10002756 bhv_vname_t *dentry,
2757 bhv_vattr_t *vap,
Nathan Scott67fcaa72006-06-09 17:00:52 +10002758 bhv_vnode_t **vpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 cred_t *credp)
2760{
2761 char *dir_name = VNAME(dentry);
2762 xfs_inode_t *dp;
2763 xfs_inode_t *cdp; /* inode of created dir */
Nathan Scott67fcaa72006-06-09 17:00:52 +10002764 bhv_vnode_t *cvp; /* vnode of created dir */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 xfs_trans_t *tp;
2766 xfs_mount_t *mp;
2767 int cancel_flags;
2768 int error;
2769 int committed;
2770 xfs_bmap_free_t free_list;
2771 xfs_fsblock_t first_block;
Nathan Scott67fcaa72006-06-09 17:00:52 +10002772 bhv_vnode_t *dir_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 boolean_t dp_joined_to_trans;
2774 boolean_t created = B_FALSE;
2775 int dm_event_sent = 0;
2776 xfs_prid_t prid;
2777 struct xfs_dquot *udqp, *gdqp;
2778 uint resblks;
2779 int dm_di_mode;
2780 int dir_namelen;
2781
2782 dir_vp = BHV_TO_VNODE(dir_bdp);
2783 dp = XFS_BHVTOI(dir_bdp);
2784 mp = dp->i_mount;
2785
2786 if (XFS_FORCED_SHUTDOWN(mp))
2787 return XFS_ERROR(EIO);
2788
2789 dir_namelen = VNAMELEN(dentry);
2790
2791 tp = NULL;
2792 dp_joined_to_trans = B_FALSE;
Christoph Hellwig0432dab2005-09-02 16:46:51 +10002793 dm_di_mode = vap->va_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
2795 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2796 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2797 dir_vp, DM_RIGHT_NULL, NULL,
2798 DM_RIGHT_NULL, dir_name, NULL,
2799 dm_di_mode, 0, 0);
2800 if (error)
2801 return error;
2802 dm_event_sent = 1;
2803 }
2804
2805 /* Return through std_return after this point. */
2806
2807 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2808
2809 mp = dp->i_mount;
2810 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10002811 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2812 prid = dp->i_d.di_projid;
2813 else if (vap->va_mask & XFS_AT_PROJID)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 prid = (xfs_prid_t)vap->va_projid;
2815 else
2816 prid = (xfs_prid_t)dfltprid;
2817
2818 /*
2819 * Make sure that we have allocated dquot(s) on disk.
2820 */
2821 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002822 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2824 if (error)
2825 goto std_return;
2826
2827 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2828 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2829 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2830 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2831 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2832 if (error == ENOSPC) {
2833 resblks = 0;
2834 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2835 XFS_TRANS_PERM_LOG_RES,
2836 XFS_MKDIR_LOG_COUNT);
2837 }
2838 if (error) {
2839 cancel_flags = 0;
2840 dp = NULL;
2841 goto error_return;
2842 }
2843
2844 xfs_ilock(dp, XFS_ILOCK_EXCL);
2845
2846 /*
2847 * Check for directory link count overflow.
2848 */
2849 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2850 error = XFS_ERROR(EMLINK);
2851 goto error_return;
2852 }
2853
2854 /*
2855 * Reserve disk quota and the inode.
2856 */
2857 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2858 if (error)
2859 goto error_return;
2860
2861 if (resblks == 0 &&
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002862 (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 goto error_return;
2864 /*
2865 * create the directory inode.
2866 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +10002867 error = xfs_dir_ialloc(&tp, dp, vap->va_mode, 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 0, credp, prid, resblks > 0,
2869 &cdp, NULL);
2870 if (error) {
2871 if (error == ENOSPC)
2872 goto error_return;
2873 goto abort_return;
2874 }
2875 ITRACE(cdp);
2876
2877 /*
2878 * Now we add the directory inode to the transaction.
2879 * We waited until now since xfs_dir_ialloc might start
2880 * a new transaction. Had we joined the transaction
2881 * earlier, the locks might have gotten released.
2882 */
2883 VN_HOLD(dir_vp);
2884 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2885 dp_joined_to_trans = B_TRUE;
2886
2887 XFS_BMAP_INIT(&free_list, &first_block);
2888
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002889 error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino,
2890 &first_block, &free_list, resblks ?
2891 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 if (error) {
2893 ASSERT(error != ENOSPC);
2894 goto error1;
2895 }
2896 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2897
2898 /*
2899 * Bump the in memory version number of the parent directory
2900 * so that other processes accessing it will recognize that
2901 * the directory has changed.
2902 */
2903 dp->i_gen++;
2904
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002905 error = xfs_dir_init(tp, cdp, dp);
2906 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908
2909 cdp->i_gen = 1;
2910 error = xfs_bumplink(tp, dp);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002911 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913
2914 cvp = XFS_ITOV(cdp);
2915
2916 created = B_TRUE;
2917
2918 *vpp = cvp;
2919 IHOLD(cdp);
2920
2921 /*
2922 * Attach the dquots to the new inode and modify the icount incore.
2923 */
2924 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2925
2926 /*
2927 * If this is a synchronous mount, make sure that the
2928 * mkdir transaction goes to disk before returning to
2929 * the user.
2930 */
2931 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2932 xfs_trans_set_sync(tp);
2933 }
2934
2935 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2936 if (error) {
2937 IRELE(cdp);
2938 goto error2;
2939 }
2940
2941 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2942 XFS_QM_DQRELE(mp, udqp);
2943 XFS_QM_DQRELE(mp, gdqp);
2944 if (error) {
2945 IRELE(cdp);
2946 }
2947
2948 /* Fall through to std_return with error = 0 or errno from
2949 * xfs_trans_commit. */
2950
2951std_return:
2952 if ( (created || (error != 0 && dm_event_sent != 0)) &&
2953 DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2954 DM_EVENT_POSTCREATE)) {
2955 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2956 dir_vp, DM_RIGHT_NULL,
2957 created ? XFS_ITOV(cdp):NULL,
2958 DM_RIGHT_NULL,
2959 dir_name, NULL,
2960 dm_di_mode, error, 0);
2961 }
2962 return error;
2963
2964 error2:
2965 error1:
2966 xfs_bmap_cancel(&free_list);
2967 abort_return:
2968 cancel_flags |= XFS_TRANS_ABORT;
2969 error_return:
2970 xfs_trans_cancel(tp, cancel_flags);
2971 XFS_QM_DQRELE(mp, udqp);
2972 XFS_QM_DQRELE(mp, gdqp);
2973
2974 if (!dp_joined_to_trans && (dp != NULL)) {
2975 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2976 }
2977
2978 goto std_return;
2979}
2980
2981
2982/*
2983 * xfs_rmdir
2984 *
2985 */
2986STATIC int
2987xfs_rmdir(
2988 bhv_desc_t *dir_bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +10002989 bhv_vname_t *dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 cred_t *credp)
2991{
2992 char *name = VNAME(dentry);
2993 xfs_inode_t *dp;
2994 xfs_inode_t *cdp; /* child directory */
2995 xfs_trans_t *tp;
2996 xfs_mount_t *mp;
2997 int error;
2998 xfs_bmap_free_t free_list;
2999 xfs_fsblock_t first_block;
3000 int cancel_flags;
3001 int committed;
Nathan Scott67fcaa72006-06-09 17:00:52 +10003002 bhv_vnode_t *dir_vp;
Vlad Apostolov17370092006-09-28 11:02:30 +10003003 int dm_di_mode = S_IFDIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 int last_cdp_link;
3005 int namelen;
3006 uint resblks;
3007
3008 dir_vp = BHV_TO_VNODE(dir_bdp);
3009 dp = XFS_BHVTOI(dir_bdp);
3010 mp = dp->i_mount;
3011
3012 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3013
3014 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3015 return XFS_ERROR(EIO);
3016 namelen = VNAMELEN(dentry);
3017
Vlad Apostolov17370092006-09-28 11:02:30 +10003018 if (!xfs_get_dir_entry(dentry, &cdp)) {
3019 dm_di_mode = cdp->i_d.di_mode;
3020 IRELE(cdp);
3021 }
3022
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3024 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3025 dir_vp, DM_RIGHT_NULL,
3026 NULL, DM_RIGHT_NULL,
Vlad Apostolov17370092006-09-28 11:02:30 +10003027 name, NULL, dm_di_mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 if (error)
3029 return XFS_ERROR(error);
3030 }
3031
3032 /* Return through std_return after this point. */
3033
3034 cdp = NULL;
3035
3036 /*
3037 * We need to get a reference to cdp before we get our log
3038 * reservation. The reason for this is that we cannot call
3039 * xfs_iget for an inode for which we do not have a reference
3040 * once we've acquired a log reservation. This is because the
3041 * inode we are trying to get might be in xfs_inactive going
3042 * for a log reservation. Since we'll have to wait for the
3043 * inactive code to complete before returning from xfs_iget,
3044 * we need to make sure that we don't have log space reserved
Nathan Scottc41564b2006-03-29 08:55:14 +10003045 * when we call xfs_iget. Instead we get an unlocked reference
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 * to the inode before getting our log reservation.
3047 */
3048 error = xfs_get_dir_entry(dentry, &cdp);
3049 if (error) {
3050 REMOVE_DEBUG_TRACE(__LINE__);
3051 goto std_return;
3052 }
3053 mp = dp->i_mount;
3054 dm_di_mode = cdp->i_d.di_mode;
3055
3056 /*
3057 * Get the dquots for the inodes.
3058 */
3059 error = XFS_QM_DQATTACH(mp, dp, 0);
3060 if (!error && dp != cdp)
3061 error = XFS_QM_DQATTACH(mp, cdp, 0);
3062 if (error) {
3063 IRELE(cdp);
3064 REMOVE_DEBUG_TRACE(__LINE__);
3065 goto std_return;
3066 }
3067
3068 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3069 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3070 /*
3071 * We try to get the real space reservation first,
3072 * allowing for directory btree deletion(s) implying
3073 * possible bmap insert(s). If we can't get the space
3074 * reservation then we use 0 instead, and avoid the bmap
3075 * btree insert(s) in the directory code by, if the bmap
3076 * insert tries to happen, instead trimming the LAST
3077 * block from the directory.
3078 */
3079 resblks = XFS_REMOVE_SPACE_RES(mp);
3080 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3081 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3082 if (error == ENOSPC) {
3083 resblks = 0;
3084 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3085 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3086 }
3087 if (error) {
3088 ASSERT(error != ENOSPC);
3089 cancel_flags = 0;
3090 IRELE(cdp);
3091 goto error_return;
3092 }
3093 XFS_BMAP_INIT(&free_list, &first_block);
3094
3095 /*
3096 * Now lock the child directory inode and the parent directory
3097 * inode in the proper order. This will take care of validating
3098 * that the directory entry for the child directory inode has
3099 * not changed while we were obtaining a log reservation.
3100 */
3101 error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3102 if (error) {
3103 xfs_trans_cancel(tp, cancel_flags);
3104 IRELE(cdp);
3105 goto std_return;
3106 }
3107
3108 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3109 if (dp != cdp) {
3110 /*
3111 * Only increment the parent directory vnode count if
3112 * we didn't bump it in looking up cdp. The only time
3113 * we don't bump it is when we're looking up ".".
3114 */
3115 VN_HOLD(dir_vp);
3116 }
3117
3118 ITRACE(cdp);
3119 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3120
3121 ASSERT(cdp->i_d.di_nlink >= 2);
3122 if (cdp->i_d.di_nlink != 2) {
3123 error = XFS_ERROR(ENOTEMPTY);
3124 goto error_return;
3125 }
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003126 if (!xfs_dir_isempty(cdp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 error = XFS_ERROR(ENOTEMPTY);
3128 goto error_return;
3129 }
3130
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003131 error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino,
3132 &first_block, &free_list, resblks);
3133 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135
3136 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3137
3138 /*
3139 * Bump the in memory generation count on the parent
3140 * directory so that other can know that it has changed.
3141 */
3142 dp->i_gen++;
3143
3144 /*
3145 * Drop the link from cdp's "..".
3146 */
3147 error = xfs_droplink(tp, dp);
3148 if (error) {
3149 goto error1;
3150 }
3151
3152 /*
3153 * Drop the link from dp to cdp.
3154 */
3155 error = xfs_droplink(tp, cdp);
3156 if (error) {
3157 goto error1;
3158 }
3159
3160 /*
3161 * Drop the "." link from cdp to self.
3162 */
3163 error = xfs_droplink(tp, cdp);
3164 if (error) {
3165 goto error1;
3166 }
3167
3168 /* Determine these before committing transaction */
3169 last_cdp_link = (cdp)->i_d.di_nlink==0;
3170
3171 /*
3172 * Take an extra ref on the child vnode so that it
3173 * does not go to xfs_inactive() from within the commit.
3174 */
3175 IHOLD(cdp);
3176
3177 /*
3178 * If this is a synchronous mount, make sure that the
3179 * rmdir transaction goes to disk before returning to
3180 * the user.
3181 */
3182 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3183 xfs_trans_set_sync(tp);
3184 }
3185
3186 error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3187 if (error) {
3188 xfs_bmap_cancel(&free_list);
3189 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3190 XFS_TRANS_ABORT));
3191 IRELE(cdp);
3192 goto std_return;
3193 }
3194
3195 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3196 if (error) {
3197 IRELE(cdp);
3198 goto std_return;
3199 }
3200
3201
3202 /*
3203 * Let interposed file systems know about removed links.
3204 */
Nathan Scott67fcaa72006-06-09 17:00:52 +10003205 bhv_vop_link_removed(XFS_ITOV(cdp), dir_vp, last_cdp_link);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206
3207 IRELE(cdp);
3208
3209 /* Fall through to std_return with error = 0 or the errno
3210 * from xfs_trans_commit. */
Nathan Scottbb19fba2006-03-22 14:12:12 +11003211 std_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3213 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3214 dir_vp, DM_RIGHT_NULL,
3215 NULL, DM_RIGHT_NULL,
3216 name, NULL, dm_di_mode,
3217 error, 0);
3218 }
3219 return error;
3220
Nathan Scottbb19fba2006-03-22 14:12:12 +11003221 error1:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 xfs_bmap_cancel(&free_list);
3223 cancel_flags |= XFS_TRANS_ABORT;
Jesper Juhl014c2542006-01-15 02:37:08 +01003224 /* FALLTHROUGH */
3225
Nathan Scottbb19fba2006-03-22 14:12:12 +11003226 error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 xfs_trans_cancel(tp, cancel_flags);
3228 goto std_return;
3229}
3230
3231
3232/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 * Read dp's entries starting at uiop->uio_offset and translate them into
3234 * bufsize bytes worth of struct dirents starting at bufbase.
3235 */
3236STATIC int
3237xfs_readdir(
3238 bhv_desc_t *dir_bdp,
3239 uio_t *uiop,
3240 cred_t *credp,
3241 int *eofp)
3242{
3243 xfs_inode_t *dp;
3244 xfs_trans_t *tp = NULL;
3245 int error = 0;
3246 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247
3248 vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3249 (inst_t *)__return_address);
3250 dp = XFS_BHVTOI(dir_bdp);
3251
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003252 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254
3255 lock_mode = xfs_ilock_map_shared(dp);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003256 error = xfs_dir_getdents(tp, dp, uiop, eofp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 xfs_iunlock_map_shared(dp, lock_mode);
3258 return error;
3259}
3260
3261
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262STATIC int
3263xfs_symlink(
3264 bhv_desc_t *dir_bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +10003265 bhv_vname_t *dentry,
3266 bhv_vattr_t *vap,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267 char *target_path,
Nathan Scott67fcaa72006-06-09 17:00:52 +10003268 bhv_vnode_t **vpp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 cred_t *credp)
3270{
3271 xfs_trans_t *tp;
3272 xfs_mount_t *mp;
3273 xfs_inode_t *dp;
3274 xfs_inode_t *ip;
3275 int error;
3276 int pathlen;
3277 xfs_bmap_free_t free_list;
3278 xfs_fsblock_t first_block;
3279 boolean_t dp_joined_to_trans;
Nathan Scott67fcaa72006-06-09 17:00:52 +10003280 bhv_vnode_t *dir_vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 uint cancel_flags;
3282 int committed;
3283 xfs_fileoff_t first_fsb;
3284 xfs_filblks_t fs_blocks;
3285 int nmaps;
3286 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3287 xfs_daddr_t d;
3288 char *cur_chunk;
3289 int byte_cnt;
3290 int n;
3291 xfs_buf_t *bp;
3292 xfs_prid_t prid;
3293 struct xfs_dquot *udqp, *gdqp;
3294 uint resblks;
3295 char *link_name = VNAME(dentry);
3296 int link_namelen;
3297
3298 *vpp = NULL;
3299 dir_vp = BHV_TO_VNODE(dir_bdp);
3300 dp = XFS_BHVTOI(dir_bdp);
3301 dp_joined_to_trans = B_FALSE;
3302 error = 0;
3303 ip = NULL;
3304 tp = NULL;
3305
3306 vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3307
3308 mp = dp->i_mount;
3309
3310 if (XFS_FORCED_SHUTDOWN(mp))
3311 return XFS_ERROR(EIO);
3312
3313 link_namelen = VNAMELEN(dentry);
3314
3315 /*
3316 * Check component lengths of the target path name.
3317 */
3318 pathlen = strlen(target_path);
3319 if (pathlen >= MAXPATHLEN) /* total string too long */
3320 return XFS_ERROR(ENAMETOOLONG);
3321 if (pathlen >= MAXNAMELEN) { /* is any component too long? */
3322 int len, total;
3323 char *path;
3324
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003325 for (total = 0, path = target_path; total < pathlen;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 /*
3327 * Skip any slashes.
3328 */
3329 while(*path == '/') {
3330 total++;
3331 path++;
3332 }
3333
3334 /*
3335 * Count up to the next slash or end of path.
3336 * Error out if the component is bigger than MAXNAMELEN.
3337 */
3338 for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3339 if (++len >= MAXNAMELEN) {
3340 error = ENAMETOOLONG;
3341 return error;
3342 }
3343 }
3344 }
3345 }
3346
3347 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3348 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3349 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3350 link_name, target_path, 0, 0, 0);
3351 if (error)
3352 return error;
3353 }
3354
3355 /* Return through std_return after this point. */
3356
3357 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10003358 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3359 prid = dp->i_d.di_projid;
3360 else if (vap->va_mask & XFS_AT_PROJID)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361 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,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10003369 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 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 &&
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003419 (error = xfs_dir_canenter(tp, dp, link_name, link_namelen)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420 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,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003474 &free_list, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 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 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003506 error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino,
3507 &first_block, &free_list, resblks);
3508 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3511 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3512
3513 /*
3514 * Bump the in memory version number of the parent directory
3515 * so that other processes accessing it will recognize that
3516 * the directory has changed.
3517 */
3518 dp->i_gen++;
3519
3520 /*
3521 * If this is a synchronous mount, make sure that the
3522 * symlink transaction goes to disk before returning to
3523 * the user.
3524 */
3525 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3526 xfs_trans_set_sync(tp);
3527 }
3528
3529 /*
3530 * xfs_trans_commit normally decrements the vnode ref count
3531 * when it unlocks the inode. Since we want to return the
3532 * vnode to the caller, we bump the vnode ref count now.
3533 */
3534 IHOLD(ip);
3535
3536 error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3537 if (error) {
3538 goto error2;
3539 }
3540 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3541 XFS_QM_DQRELE(mp, udqp);
3542 XFS_QM_DQRELE(mp, gdqp);
3543
3544 /* Fall through to std_return with error = 0 or errno from
3545 * xfs_trans_commit */
3546std_return:
3547 if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3548 DM_EVENT_POSTSYMLINK)) {
3549 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3550 dir_vp, DM_RIGHT_NULL,
3551 error ? NULL : XFS_ITOV(ip),
3552 DM_RIGHT_NULL, link_name, target_path,
3553 0, error, 0);
3554 }
3555
3556 if (!error) {
Nathan Scott67fcaa72006-06-09 17:00:52 +10003557 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558
3559 ASSERT(ip);
3560 vp = XFS_ITOV(ip);
3561 *vpp = vp;
3562 }
3563 return error;
3564
3565 error2:
3566 IRELE(ip);
3567 error1:
3568 xfs_bmap_cancel(&free_list);
3569 cancel_flags |= XFS_TRANS_ABORT;
3570 error_return:
3571 xfs_trans_cancel(tp, cancel_flags);
3572 XFS_QM_DQRELE(mp, udqp);
3573 XFS_QM_DQRELE(mp, gdqp);
3574
3575 if (!dp_joined_to_trans && (dp != NULL)) {
3576 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3577 }
3578
3579 goto std_return;
3580}
3581
3582
3583/*
3584 * xfs_fid2
3585 *
3586 * A fid routine that takes a pointer to a previously allocated
3587 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3588 */
3589STATIC int
3590xfs_fid2(
3591 bhv_desc_t *bdp,
3592 fid_t *fidp)
3593{
3594 xfs_inode_t *ip;
3595 xfs_fid2_t *xfid;
3596
3597 vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3598 (inst_t *)__return_address);
3599 ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3600
3601 xfid = (xfs_fid2_t *)fidp;
3602 ip = XFS_BHVTOI(bdp);
3603 xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3604 xfid->fid_pad = 0;
3605 /*
3606 * use memcpy because the inode is a long long and there's no
3607 * assurance that xfid->fid_ino is properly aligned.
3608 */
3609 memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3610 xfid->fid_gen = ip->i_d.di_gen;
3611
3612 return 0;
3613}
3614
3615
3616/*
3617 * xfs_rwlock
3618 */
3619int
3620xfs_rwlock(
3621 bhv_desc_t *bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +10003622 bhv_vrwlock_t locktype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623{
3624 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10003625 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626
3627 vp = BHV_TO_VNODE(bdp);
Christoph Hellwig0432dab2005-09-02 16:46:51 +10003628 if (VN_ISDIR(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 return 1;
3630 ip = XFS_BHVTOI(bdp);
3631 if (locktype == VRWLOCK_WRITE) {
3632 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3633 } else if (locktype == VRWLOCK_TRY_READ) {
Jesper Juhl014c2542006-01-15 02:37:08 +01003634 return xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635 } else if (locktype == VRWLOCK_TRY_WRITE) {
Jesper Juhl014c2542006-01-15 02:37:08 +01003636 return xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003637 } else {
3638 ASSERT((locktype == VRWLOCK_READ) ||
3639 (locktype == VRWLOCK_WRITE_DIRECT));
3640 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3641 }
3642
3643 return 1;
3644}
3645
3646
3647/*
3648 * xfs_rwunlock
3649 */
3650void
3651xfs_rwunlock(
3652 bhv_desc_t *bdp,
Nathan Scott8285fb52006-06-09 17:07:12 +10003653 bhv_vrwlock_t locktype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654{
3655 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10003656 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657
3658 vp = BHV_TO_VNODE(bdp);
Christoph Hellwig0432dab2005-09-02 16:46:51 +10003659 if (VN_ISDIR(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 return;
3661 ip = XFS_BHVTOI(bdp);
3662 if (locktype == VRWLOCK_WRITE) {
3663 /*
3664 * In the write case, we may have added a new entry to
3665 * the reference cache. This might store a pointer to
3666 * an inode to be released in this inode. If it is there,
3667 * clear the pointer and release the inode after unlocking
3668 * this one.
3669 */
3670 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3671 } else {
3672 ASSERT((locktype == VRWLOCK_READ) ||
3673 (locktype == VRWLOCK_WRITE_DIRECT));
3674 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3675 }
3676 return;
3677}
3678
3679STATIC int
3680xfs_inode_flush(
3681 bhv_desc_t *bdp,
3682 int flags)
3683{
3684 xfs_inode_t *ip;
3685 xfs_mount_t *mp;
3686 xfs_inode_log_item_t *iip;
3687 int error = 0;
3688
3689 ip = XFS_BHVTOI(bdp);
3690 mp = ip->i_mount;
3691 iip = ip->i_itemp;
3692
3693 if (XFS_FORCED_SHUTDOWN(mp))
3694 return XFS_ERROR(EIO);
3695
3696 /*
3697 * Bypass inodes which have already been cleaned by
3698 * the inode flush clustering code inside xfs_iflush
3699 */
3700 if ((ip->i_update_core == 0) &&
3701 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3702 return 0;
3703
3704 if (flags & FLUSH_LOG) {
3705 if (iip && iip->ili_last_lsn) {
3706 xlog_t *log = mp->m_log;
3707 xfs_lsn_t sync_lsn;
3708 int s, log_flags = XFS_LOG_FORCE;
3709
3710 s = GRANT_LOCK(log);
3711 sync_lsn = log->l_last_sync_lsn;
3712 GRANT_UNLOCK(log, s);
3713
3714 if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3715 return 0;
3716
3717 if (flags & FLUSH_SYNC)
3718 log_flags |= XFS_LOG_SYNC;
3719 return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3720 }
3721 }
3722
3723 /*
3724 * We make this non-blocking if the inode is contended,
3725 * return EAGAIN to indicate to the caller that they
3726 * did not succeed. This prevents the flush path from
3727 * blocking on inodes inside another operation right
3728 * now, they get caught later by xfs_sync.
3729 */
3730 if (flags & FLUSH_INODE) {
3731 int flush_flags;
3732
3733 if (xfs_ipincount(ip))
3734 return EAGAIN;
3735
3736 if (flags & FLUSH_SYNC) {
3737 xfs_ilock(ip, XFS_ILOCK_SHARED);
3738 xfs_iflock(ip);
3739 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3740 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3741 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3742 return EAGAIN;
3743 }
3744 } else {
3745 return EAGAIN;
3746 }
3747
3748 if (flags & FLUSH_SYNC)
3749 flush_flags = XFS_IFLUSH_SYNC;
3750 else
3751 flush_flags = XFS_IFLUSH_ASYNC;
3752
3753 error = xfs_iflush(ip, flush_flags);
3754 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3755 }
3756
3757 return error;
3758}
3759
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760int
3761xfs_set_dmattrs (
3762 bhv_desc_t *bdp,
3763 u_int evmask,
3764 u_int16_t state,
3765 cred_t *credp)
3766{
3767 xfs_inode_t *ip;
3768 xfs_trans_t *tp;
3769 xfs_mount_t *mp;
3770 int error;
3771
3772 if (!capable(CAP_SYS_ADMIN))
3773 return XFS_ERROR(EPERM);
3774
3775 ip = XFS_BHVTOI(bdp);
3776 mp = ip->i_mount;
3777
3778 if (XFS_FORCED_SHUTDOWN(mp))
3779 return XFS_ERROR(EIO);
3780
3781 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3782 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3783 if (error) {
3784 xfs_trans_cancel(tp, 0);
3785 return error;
3786 }
3787 xfs_ilock(ip, XFS_ILOCK_EXCL);
3788 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3789
3790 ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3791 ip->i_iocore.io_dmstate = ip->i_d.di_dmstate = state;
3792
3793 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3794 IHOLD(ip);
3795 error = xfs_trans_commit(tp, 0, NULL);
3796
3797 return error;
3798}
3799
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800STATIC int
3801xfs_reclaim(
3802 bhv_desc_t *bdp)
3803{
3804 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10003805 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806
3807 vp = BHV_TO_VNODE(bdp);
3808 ip = XFS_BHVTOI(bdp);
3809
3810 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3811
3812 ASSERT(!VN_MAPPED(vp));
3813
3814 /* bad inode, get out here ASAP */
3815 if (VN_BAD(vp)) {
3816 xfs_ireclaim(ip);
3817 return 0;
3818 }
3819
Christoph Hellwig51c91ed2005-09-02 16:58:38 +10003820 vn_iowait(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821
Christoph Hellwig51c91ed2005-09-02 16:58:38 +10003822 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11003824 /*
3825 * Make sure the atime in the XFS inode is correct before freeing the
3826 * Linux inode.
3827 */
3828 xfs_synchronize_atime(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829
3830 /* If we have nothing to flush with this inode then complete the
3831 * teardown now, otherwise break the link between the xfs inode
3832 * and the linux inode and clean up the xfs inode later. This
3833 * avoids flushing the inode to disk during the delete operation
3834 * itself.
3835 */
3836 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3837 xfs_ilock(ip, XFS_ILOCK_EXCL);
3838 xfs_iflock(ip);
3839 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3840 } else {
3841 xfs_mount_t *mp = ip->i_mount;
3842
3843 /* Protect sync from us */
3844 XFS_MOUNT_ILOCK(mp);
3845 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3846 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
David Chinnerf273ab82006-09-28 11:06:03 +10003847 spin_lock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848 ip->i_flags |= XFS_IRECLAIMABLE;
David Chinnerf273ab82006-09-28 11:06:03 +10003849 spin_unlock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850 XFS_MOUNT_IUNLOCK(mp);
3851 }
3852 return 0;
3853}
3854
3855int
3856xfs_finish_reclaim(
3857 xfs_inode_t *ip,
3858 int locked,
3859 int sync_mode)
3860{
3861 xfs_ihash_t *ih = ip->i_hash;
Nathan Scott67fcaa72006-06-09 17:00:52 +10003862 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 int error;
3864
3865 if (vp && VN_BAD(vp))
3866 goto reclaim;
3867
3868 /* The hash lock here protects a thread in xfs_iget_core from
3869 * racing with us on linking the inode back with a vnode.
3870 * Once we have the XFS_IRECLAIM flag set it will not touch
3871 * us.
3872 */
3873 write_lock(&ih->ih_lock);
David Chinnerf273ab82006-09-28 11:06:03 +10003874 spin_lock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 if ((ip->i_flags & XFS_IRECLAIM) ||
3876 (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
David Chinnerf273ab82006-09-28 11:06:03 +10003877 spin_unlock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 write_unlock(&ih->ih_lock);
3879 if (locked) {
3880 xfs_ifunlock(ip);
3881 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3882 }
Jesper Juhl014c2542006-01-15 02:37:08 +01003883 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884 }
3885 ip->i_flags |= XFS_IRECLAIM;
David Chinnerf273ab82006-09-28 11:06:03 +10003886 spin_unlock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 write_unlock(&ih->ih_lock);
3888
3889 /*
3890 * If the inode is still dirty, then flush it out. If the inode
3891 * is not in the AIL, then it will be OK to flush it delwri as
3892 * long as xfs_iflush() does not keep any references to the inode.
3893 * We leave that decision up to xfs_iflush() since it has the
3894 * knowledge of whether it's OK to simply do a delwri flush of
3895 * the inode or whether we need to wait until the inode is
3896 * pulled from the AIL.
3897 * We get the flush lock regardless, though, just to make sure
3898 * we don't free it while it is being flushed.
3899 */
3900 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3901 if (!locked) {
3902 xfs_ilock(ip, XFS_ILOCK_EXCL);
3903 xfs_iflock(ip);
3904 }
3905
3906 if (ip->i_update_core ||
3907 ((ip->i_itemp != NULL) &&
3908 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3909 error = xfs_iflush(ip, sync_mode);
3910 /*
3911 * If we hit an error, typically because of filesystem
3912 * shutdown, we don't need to let vn_reclaim to know
3913 * because we're gonna reclaim the inode anyway.
3914 */
3915 if (error) {
3916 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3917 goto reclaim;
3918 }
3919 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3920 }
3921
3922 ASSERT(ip->i_update_core == 0);
3923 ASSERT(ip->i_itemp == NULL ||
3924 ip->i_itemp->ili_format.ilf_fields == 0);
3925 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3926 } else if (locked) {
3927 /*
3928 * We are not interested in doing an iflush if we're
3929 * in the process of shutting down the filesystem forcibly.
3930 * So, just reclaim the inode.
3931 */
3932 xfs_ifunlock(ip);
3933 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3934 }
3935
3936 reclaim:
3937 xfs_ireclaim(ip);
3938 return 0;
3939}
3940
3941int
3942xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3943{
3944 int purged;
3945 xfs_inode_t *ip, *n;
3946 int done = 0;
3947
3948 while (!done) {
3949 purged = 0;
3950 XFS_MOUNT_ILOCK(mp);
3951 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3952 if (noblock) {
3953 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3954 continue;
3955 if (xfs_ipincount(ip) ||
3956 !xfs_iflock_nowait(ip)) {
3957 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3958 continue;
3959 }
3960 }
3961 XFS_MOUNT_IUNLOCK(mp);
Felix Blyakher6b2cf612005-11-25 16:42:13 +11003962 if (xfs_finish_reclaim(ip, noblock,
3963 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3964 delay(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003965 purged = 1;
3966 break;
3967 }
3968
3969 done = !purged;
3970 }
3971
3972 XFS_MOUNT_IUNLOCK(mp);
3973 return 0;
3974}
3975
3976/*
3977 * xfs_alloc_file_space()
3978 * This routine allocates disk space for the given file.
3979 *
3980 * If alloc_type == 0, this request is for an ALLOCSP type
3981 * request which will change the file size. In this case, no
3982 * DMAPI event will be generated by the call. A TRUNCATE event
3983 * will be generated later by xfs_setattr.
3984 *
3985 * If alloc_type != 0, this request is for a RESVSP type
3986 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3987 * lower block boundary byte address is less than the file's
3988 * length.
3989 *
3990 * RETURNS:
3991 * 0 on success
3992 * errno on error
3993 *
3994 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003995STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996xfs_alloc_file_space(
3997 xfs_inode_t *ip,
3998 xfs_off_t offset,
3999 xfs_off_t len,
4000 int alloc_type,
4001 int attr_flags)
4002{
Nathan Scottdd9f4382006-01-11 15:28:28 +11004003 xfs_mount_t *mp = ip->i_mount;
4004 xfs_off_t count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005 xfs_filblks_t allocated_fsb;
4006 xfs_filblks_t allocatesize_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11004007 xfs_extlen_t extsz, temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008 xfs_fileoff_t startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11004009 xfs_fsblock_t firstfsb;
4010 int nimaps;
4011 int bmapi_flag;
4012 int quota_flag;
4013 int rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014 xfs_trans_t *tp;
Nathan Scottdd9f4382006-01-11 15:28:28 +11004015 xfs_bmbt_irec_t imaps[1], *imapp;
4016 xfs_bmap_free_t free_list;
4017 uint qblocks, resblks, resrtextents;
4018 int committed;
4019 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004020
4021 vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022
4023 if (XFS_FORCED_SHUTDOWN(mp))
4024 return XFS_ERROR(EIO);
4025
Nathan Scottdd9f4382006-01-11 15:28:28 +11004026 rt = XFS_IS_REALTIME_INODE(ip);
4027 if (unlikely(rt)) {
4028 if (!(extsz = ip->i_d.di_extsize))
4029 extsz = mp->m_sb.sb_rextsize;
4030 } else {
4031 extsz = ip->i_d.di_extsize;
4032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033
4034 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4035 return error;
4036
4037 if (len <= 0)
4038 return XFS_ERROR(EINVAL);
4039
4040 count = len;
4041 error = 0;
4042 imapp = &imaps[0];
Nathan Scottdd9f4382006-01-11 15:28:28 +11004043 nimaps = 1;
4044 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004045 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4046 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4047
4048 /* Generate a DMAPI event if needed. */
4049 if (alloc_type != 0 && offset < ip->i_d.di_size &&
4050 (attr_flags&ATTR_DMI) == 0 &&
4051 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4052 xfs_off_t end_dmi_offset;
4053
4054 end_dmi_offset = offset+len;
4055 if (end_dmi_offset > ip->i_d.di_size)
4056 end_dmi_offset = ip->i_d.di_size;
4057 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4058 offset, end_dmi_offset - offset,
4059 0, NULL);
4060 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01004061 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004062 }
4063
4064 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004065 * Allocate file space until done or until there is an error
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066 */
4067retry:
4068 while (allocatesize_fsb && !error) {
Nathan Scottdd9f4382006-01-11 15:28:28 +11004069 xfs_fileoff_t s, e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070
Nathan Scottdd9f4382006-01-11 15:28:28 +11004071 /*
Nathan Scott3ddb8fa2006-01-11 15:33:02 +11004072 * Determine space reservations for data/realtime.
Nathan Scottdd9f4382006-01-11 15:28:28 +11004073 */
4074 if (unlikely(extsz)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 s = startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11004076 do_div(s, extsz);
4077 s *= extsz;
4078 e = startoffset_fsb + allocatesize_fsb;
4079 if ((temp = do_mod(startoffset_fsb, extsz)))
4080 e += temp;
4081 if ((temp = do_mod(e, extsz)))
4082 e += extsz - temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004083 } else {
Nathan Scottdd9f4382006-01-11 15:28:28 +11004084 s = 0;
4085 e = allocatesize_fsb;
4086 }
4087
4088 if (unlikely(rt)) {
4089 resrtextents = qblocks = (uint)(e - s);
4090 resrtextents /= mp->m_sb.sb_rextsize;
4091 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4092 quota_flag = XFS_QMOPT_RES_RTBLKS;
4093 } else {
4094 resrtextents = 0;
4095 resblks = qblocks = \
4096 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
4097 quota_flag = XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004098 }
4099
4100 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004101 * Allocate and setup the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102 */
4103 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Nathan Scottdd9f4382006-01-11 15:28:28 +11004104 error = xfs_trans_reserve(tp, resblks,
4105 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106 XFS_TRANS_PERM_LOG_RES,
4107 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004108 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004109 * Check for running out of space
Linus Torvalds1da177e2005-04-16 15:20:36 -07004110 */
4111 if (error) {
4112 /*
4113 * Free the transaction structure.
4114 */
4115 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4116 xfs_trans_cancel(tp, 0);
4117 break;
4118 }
4119 xfs_ilock(ip, XFS_ILOCK_EXCL);
Nathan Scottdd9f4382006-01-11 15:28:28 +11004120 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
4121 qblocks, 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004122 if (error)
4123 goto error1;
4124
4125 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4126 xfs_trans_ihold(tp, ip);
4127
4128 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004129 * Issue the xfs_bmapi() call to allocate the blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130 */
4131 XFS_BMAP_INIT(&free_list, &firstfsb);
Olaf Weber3e57ecf2006-06-09 14:48:12 +10004132 error = XFS_BMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
Nathan Scottdd9f4382006-01-11 15:28:28 +11004133 allocatesize_fsb, bmapi_flag,
4134 &firstfsb, 0, imapp, &nimaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10004135 &free_list, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004136 if (error) {
4137 goto error0;
4138 }
4139
4140 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11004141 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142 */
4143 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4144 if (error) {
4145 goto error0;
4146 }
4147
4148 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4149 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4150 if (error) {
4151 break;
4152 }
4153
4154 allocated_fsb = imapp->br_blockcount;
4155
Nathan Scottdd9f4382006-01-11 15:28:28 +11004156 if (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004157 error = XFS_ERROR(ENOSPC);
4158 break;
4159 }
4160
4161 startoffset_fsb += allocated_fsb;
4162 allocatesize_fsb -= allocated_fsb;
4163 }
4164dmapi_enospc_check:
4165 if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4166 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4167
4168 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4169 XFS_ITOV(ip), DM_RIGHT_NULL,
4170 XFS_ITOV(ip), DM_RIGHT_NULL,
4171 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4172 if (error == 0)
4173 goto retry; /* Maybe DMAPI app. has made space */
4174 /* else fall through with error from XFS_SEND_DATA */
4175 }
4176
4177 return error;
4178
Nathan Scottdd9f4382006-01-11 15:28:28 +11004179error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004180 xfs_bmap_cancel(&free_list);
Nathan Scottdd9f4382006-01-11 15:28:28 +11004181 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
4182
4183error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004184 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4185 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4186 goto dmapi_enospc_check;
4187}
4188
4189/*
4190 * Zero file bytes between startoff and endoff inclusive.
4191 * The iolock is held exclusive and no blocks are buffered.
4192 */
4193STATIC int
4194xfs_zero_remaining_bytes(
4195 xfs_inode_t *ip,
4196 xfs_off_t startoff,
4197 xfs_off_t endoff)
4198{
4199 xfs_bmbt_irec_t imap;
4200 xfs_fileoff_t offset_fsb;
4201 xfs_off_t lastoffset;
4202 xfs_off_t offset;
4203 xfs_buf_t *bp;
4204 xfs_mount_t *mp = ip->i_mount;
4205 int nimap;
4206 int error = 0;
4207
4208 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4209 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4210 mp->m_rtdev_targp : mp->m_ddev_targp);
4211
4212 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4213 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4214 nimap = 1;
Olaf Weber3e57ecf2006-06-09 14:48:12 +10004215 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, offset_fsb, 1, 0,
4216 NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217 if (error || nimap < 1)
4218 break;
4219 ASSERT(imap.br_blockcount >= 1);
4220 ASSERT(imap.br_startoff == offset_fsb);
4221 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4222 if (lastoffset > endoff)
4223 lastoffset = endoff;
4224 if (imap.br_startblock == HOLESTARTBLOCK)
4225 continue;
4226 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4227 if (imap.br_state == XFS_EXT_UNWRITTEN)
4228 continue;
4229 XFS_BUF_UNDONE(bp);
4230 XFS_BUF_UNWRITE(bp);
4231 XFS_BUF_READ(bp);
4232 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4233 xfsbdstrat(mp, bp);
4234 if ((error = xfs_iowait(bp))) {
4235 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4236 mp, bp, XFS_BUF_ADDR(bp));
4237 break;
4238 }
4239 memset(XFS_BUF_PTR(bp) +
4240 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4241 0, lastoffset - offset + 1);
4242 XFS_BUF_UNDONE(bp);
4243 XFS_BUF_UNREAD(bp);
4244 XFS_BUF_WRITE(bp);
4245 xfsbdstrat(mp, bp);
4246 if ((error = xfs_iowait(bp))) {
4247 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4248 mp, bp, XFS_BUF_ADDR(bp));
4249 break;
4250 }
4251 }
4252 xfs_buf_free(bp);
4253 return error;
4254}
4255
4256/*
4257 * xfs_free_file_space()
4258 * This routine frees disk space for the given file.
4259 *
4260 * This routine is only called by xfs_change_file_space
4261 * for an UNRESVSP type call.
4262 *
4263 * RETURNS:
4264 * 0 on success
4265 * errno on error
4266 *
4267 */
4268STATIC int
4269xfs_free_file_space(
4270 xfs_inode_t *ip,
4271 xfs_off_t offset,
4272 xfs_off_t len,
4273 int attr_flags)
4274{
Nathan Scott67fcaa72006-06-09 17:00:52 +10004275 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004276 int committed;
4277 int done;
4278 xfs_off_t end_dmi_offset;
4279 xfs_fileoff_t endoffset_fsb;
4280 int error;
4281 xfs_fsblock_t firstfsb;
4282 xfs_bmap_free_t free_list;
4283 xfs_off_t ilen;
4284 xfs_bmbt_irec_t imap;
4285 xfs_off_t ioffset;
4286 xfs_extlen_t mod=0;
4287 xfs_mount_t *mp;
4288 int nimap;
4289 uint resblks;
Nathan Scott673cdf52006-09-28 10:56:26 +10004290 uint rounding;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004291 int rt;
4292 xfs_fileoff_t startoffset_fsb;
4293 xfs_trans_t *tp;
Dean Roehrich5fcbab32005-05-05 13:27:19 -07004294 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004295
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004296 vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004297 mp = ip->i_mount;
4298
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004299 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4300
Linus Torvalds1da177e2005-04-16 15:20:36 -07004301 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4302 return error;
4303
4304 error = 0;
4305 if (len <= 0) /* if nothing being freed */
4306 return error;
4307 rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4308 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4309 end_dmi_offset = offset + len;
4310 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4311
4312 if (offset < ip->i_d.di_size &&
4313 (attr_flags & ATTR_DMI) == 0 &&
4314 DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4315 if (end_dmi_offset > ip->i_d.di_size)
4316 end_dmi_offset = ip->i_d.di_size;
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004317 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, vp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004318 offset, end_dmi_offset - offset,
4319 AT_DELAY_FLAG(attr_flags), NULL);
4320 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01004321 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322 }
4323
Dean Roehrich5fcbab32005-05-05 13:27:19 -07004324 if (attr_flags & ATTR_NOLOCK)
4325 need_iolock = 0;
Yingping Lu9fa80462006-03-22 12:44:35 +11004326 if (need_iolock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004327 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Yingping Lu9fa80462006-03-22 12:44:35 +11004328 vn_iowait(vp); /* wait for the completion of any pending DIOs */
4329 }
Dean Roehrich5fcbab32005-05-05 13:27:19 -07004330
Nathan Scott673cdf52006-09-28 10:56:26 +10004331 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, NBPP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004332 ilen = len + (offset & (rounding - 1));
4333 ioffset = offset & ~(rounding - 1);
4334 if (ilen & (rounding - 1))
4335 ilen = (ilen + rounding) & ~(rounding - 1);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004336
4337 if (VN_CACHED(vp) != 0) {
4338 xfs_inval_cached_trace(&ip->i_iocore, ioffset, -1,
4339 ctooff(offtoct(ioffset)), -1);
Nathan Scott67fcaa72006-06-09 17:00:52 +10004340 bhv_vop_flushinval_pages(vp, ctooff(offtoct(ioffset)),
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10004341 -1, FI_REMAPF_LOCKED);
4342 }
4343
Linus Torvalds1da177e2005-04-16 15:20:36 -07004344 /*
4345 * Need to zero the stuff we're not freeing, on disk.
4346 * If its a realtime file & can't use unwritten extents then we
4347 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4348 * will take care of it for us.
4349 */
4350 if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4351 nimap = 1;
Olaf Weber3e57ecf2006-06-09 14:48:12 +10004352 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, startoffset_fsb,
4353 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004354 if (error)
4355 goto out_unlock_iolock;
4356 ASSERT(nimap == 0 || nimap == 1);
4357 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4358 xfs_daddr_t block;
4359
4360 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4361 block = imap.br_startblock;
4362 mod = do_div(block, mp->m_sb.sb_rextsize);
4363 if (mod)
4364 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4365 }
4366 nimap = 1;
Olaf Weber3e57ecf2006-06-09 14:48:12 +10004367 error = XFS_BMAPI(mp, NULL, &ip->i_iocore, endoffset_fsb - 1,
4368 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004369 if (error)
4370 goto out_unlock_iolock;
4371 ASSERT(nimap == 0 || nimap == 1);
4372 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4373 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4374 mod++;
4375 if (mod && (mod != mp->m_sb.sb_rextsize))
4376 endoffset_fsb -= mod;
4377 }
4378 }
4379 if ((done = (endoffset_fsb <= startoffset_fsb)))
4380 /*
4381 * One contiguous piece to clear
4382 */
4383 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4384 else {
4385 /*
4386 * Some full blocks, possibly two pieces to clear
4387 */
4388 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4389 error = xfs_zero_remaining_bytes(ip, offset,
4390 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4391 if (!error &&
4392 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4393 error = xfs_zero_remaining_bytes(ip,
4394 XFS_FSB_TO_B(mp, endoffset_fsb),
4395 offset + len - 1);
4396 }
4397
4398 /*
4399 * free file space until done or until there is an error
4400 */
4401 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4402 while (!error && !done) {
4403
4404 /*
4405 * allocate and setup the transaction
4406 */
4407 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4408 error = xfs_trans_reserve(tp,
4409 resblks,
4410 XFS_WRITE_LOG_RES(mp),
4411 0,
4412 XFS_TRANS_PERM_LOG_RES,
4413 XFS_WRITE_LOG_COUNT);
4414
4415 /*
4416 * check for running out of space
4417 */
4418 if (error) {
4419 /*
4420 * Free the transaction structure.
4421 */
4422 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4423 xfs_trans_cancel(tp, 0);
4424 break;
4425 }
4426 xfs_ilock(ip, XFS_ILOCK_EXCL);
4427 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
Nathan Scottdd9f4382006-01-11 15:28:28 +11004428 ip->i_udquot, ip->i_gdquot, resblks, 0,
4429 XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004430 if (error)
4431 goto error1;
4432
4433 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4434 xfs_trans_ihold(tp, ip);
4435
4436 /*
4437 * issue the bunmapi() call to free the blocks
4438 */
4439 XFS_BMAP_INIT(&free_list, &firstfsb);
Olaf Weber3e57ecf2006-06-09 14:48:12 +10004440 error = XFS_BUNMAPI(mp, tp, &ip->i_iocore, startoffset_fsb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441 endoffset_fsb - startoffset_fsb,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10004442 0, 2, &firstfsb, &free_list, NULL, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004443 if (error) {
4444 goto error0;
4445 }
4446
4447 /*
4448 * complete the transaction
4449 */
4450 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4451 if (error) {
4452 goto error0;
4453 }
4454
4455 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4456 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4457 }
4458
4459 out_unlock_iolock:
4460 if (need_iolock)
4461 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4462 return error;
4463
4464 error0:
4465 xfs_bmap_cancel(&free_list);
4466 error1:
4467 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4468 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4469 XFS_ILOCK_EXCL);
4470 return error;
4471}
4472
4473/*
4474 * xfs_change_file_space()
4475 * This routine allocates or frees disk space for the given file.
4476 * The user specified parameters are checked for alignment and size
4477 * limitations.
4478 *
4479 * RETURNS:
4480 * 0 on success
4481 * errno on error
4482 *
4483 */
4484int
4485xfs_change_file_space(
4486 bhv_desc_t *bdp,
4487 int cmd,
4488 xfs_flock64_t *bf,
4489 xfs_off_t offset,
4490 cred_t *credp,
4491 int attr_flags)
4492{
4493 int clrprealloc;
4494 int error;
4495 xfs_fsize_t fsize;
4496 xfs_inode_t *ip;
4497 xfs_mount_t *mp;
4498 int setprealloc;
4499 xfs_off_t startoffset;
4500 xfs_off_t llen;
4501 xfs_trans_t *tp;
Nathan Scott8285fb52006-06-09 17:07:12 +10004502 bhv_vattr_t va;
Nathan Scott67fcaa72006-06-09 17:00:52 +10004503 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004504
4505 vp = BHV_TO_VNODE(bdp);
4506 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4507
4508 ip = XFS_BHVTOI(bdp);
4509 mp = ip->i_mount;
4510
4511 /*
4512 * must be a regular file and have write permission
4513 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +10004514 if (!VN_ISREG(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004515 return XFS_ERROR(EINVAL);
4516
4517 xfs_ilock(ip, XFS_ILOCK_SHARED);
4518
4519 if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4520 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4521 return error;
4522 }
4523
4524 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4525
4526 switch (bf->l_whence) {
4527 case 0: /*SEEK_SET*/
4528 break;
4529 case 1: /*SEEK_CUR*/
4530 bf->l_start += offset;
4531 break;
4532 case 2: /*SEEK_END*/
4533 bf->l_start += ip->i_d.di_size;
4534 break;
4535 default:
4536 return XFS_ERROR(EINVAL);
4537 }
4538
4539 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4540
4541 if ( (bf->l_start < 0)
4542 || (bf->l_start > XFS_MAXIOFFSET(mp))
4543 || (bf->l_start + llen < 0)
4544 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4545 return XFS_ERROR(EINVAL);
4546
4547 bf->l_whence = 0;
4548
4549 startoffset = bf->l_start;
4550 fsize = ip->i_d.di_size;
4551
4552 /*
4553 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4554 * file space.
4555 * These calls do NOT zero the data space allocated to the file,
4556 * nor do they change the file size.
4557 *
4558 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4559 * space.
4560 * These calls cause the new file data to be zeroed and the file
4561 * size to be changed.
4562 */
4563 setprealloc = clrprealloc = 0;
4564
4565 switch (cmd) {
4566 case XFS_IOC_RESVSP:
4567 case XFS_IOC_RESVSP64:
4568 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4569 1, attr_flags);
4570 if (error)
4571 return error;
4572 setprealloc = 1;
4573 break;
4574
4575 case XFS_IOC_UNRESVSP:
4576 case XFS_IOC_UNRESVSP64:
4577 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4578 attr_flags)))
4579 return error;
4580 break;
4581
4582 case XFS_IOC_ALLOCSP:
4583 case XFS_IOC_ALLOCSP64:
4584 case XFS_IOC_FREESP:
4585 case XFS_IOC_FREESP64:
4586 if (startoffset > fsize) {
4587 error = xfs_alloc_file_space(ip, fsize,
4588 startoffset - fsize, 0, attr_flags);
4589 if (error)
4590 break;
4591 }
4592
4593 va.va_mask = XFS_AT_SIZE;
4594 va.va_size = startoffset;
4595
4596 error = xfs_setattr(bdp, &va, attr_flags, credp);
4597
4598 if (error)
4599 return error;
4600
4601 clrprealloc = 1;
4602 break;
4603
4604 default:
4605 ASSERT(0);
4606 return XFS_ERROR(EINVAL);
4607 }
4608
4609 /*
4610 * update the inode timestamp, mode, and prealloc flag bits
4611 */
4612 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4613
4614 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4615 0, 0, 0))) {
4616 /* ASSERT(0); */
4617 xfs_trans_cancel(tp, 0);
4618 return error;
4619 }
4620
4621 xfs_ilock(ip, XFS_ILOCK_EXCL);
4622
4623 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4624 xfs_trans_ihold(tp, ip);
4625
4626 if ((attr_flags & ATTR_DMI) == 0) {
4627 ip->i_d.di_mode &= ~S_ISUID;
4628
4629 /*
4630 * Note that we don't have to worry about mandatory
4631 * file locking being disabled here because we only
4632 * clear the S_ISGID bit if the Group execute bit is
4633 * on, but if it was on then mandatory locking wouldn't
4634 * have been enabled.
4635 */
4636 if (ip->i_d.di_mode & S_IXGRP)
4637 ip->i_d.di_mode &= ~S_ISGID;
4638
4639 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4640 }
4641 if (setprealloc)
4642 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4643 else if (clrprealloc)
4644 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4645
4646 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4647 xfs_trans_set_sync(tp);
4648
4649 error = xfs_trans_commit(tp, 0, NULL);
4650
4651 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4652
4653 return error;
4654}
4655
Nathan Scott67fcaa72006-06-09 17:00:52 +10004656bhv_vnodeops_t xfs_vnodeops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004657 BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4658 .vop_open = xfs_open,
Nathan Scott7d4fb402006-06-09 15:27:16 +10004659 .vop_close = xfs_close,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004660 .vop_read = xfs_read,
4661#ifdef HAVE_SENDFILE
4662 .vop_sendfile = xfs_sendfile,
4663#endif
Nathan Scott1b895842006-03-31 13:08:59 +10004664#ifdef HAVE_SPLICE
4665 .vop_splice_read = xfs_splice_read,
4666 .vop_splice_write = xfs_splice_write,
4667#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004668 .vop_write = xfs_write,
4669 .vop_ioctl = xfs_ioctl,
4670 .vop_getattr = xfs_getattr,
4671 .vop_setattr = xfs_setattr,
4672 .vop_access = xfs_access,
4673 .vop_lookup = xfs_lookup,
4674 .vop_create = xfs_create,
4675 .vop_remove = xfs_remove,
4676 .vop_link = xfs_link,
4677 .vop_rename = xfs_rename,
4678 .vop_mkdir = xfs_mkdir,
4679 .vop_rmdir = xfs_rmdir,
4680 .vop_readdir = xfs_readdir,
4681 .vop_symlink = xfs_symlink,
4682 .vop_readlink = xfs_readlink,
4683 .vop_fsync = xfs_fsync,
4684 .vop_inactive = xfs_inactive,
4685 .vop_fid2 = xfs_fid2,
4686 .vop_rwlock = xfs_rwlock,
4687 .vop_rwunlock = xfs_rwunlock,
4688 .vop_bmap = xfs_bmap,
4689 .vop_reclaim = xfs_reclaim,
4690 .vop_attr_get = xfs_attr_get,
4691 .vop_attr_set = xfs_attr_set,
4692 .vop_attr_remove = xfs_attr_remove,
4693 .vop_attr_list = xfs_attr_list,
4694 .vop_link_removed = (vop_link_removed_t)fs_noval,
4695 .vop_vnode_change = (vop_vnode_change_t)fs_noval,
4696 .vop_tosspages = fs_tosspages,
4697 .vop_flushinval_pages = fs_flushinval_pages,
4698 .vop_flush_pages = fs_flush_pages,
4699 .vop_release = xfs_release,
4700 .vop_iflush = xfs_inode_flush,
4701};