blob: bc0a4707189a0147212bf2d46cdbfb59cfd6f58f [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include "xfs_log_priv.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100053#include "xfs_filestream.h"
Christoph Hellwig993386c2007-08-28 16:12:30 +100054#include "xfs_vnodeops.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Christoph Hellwig993386c2007-08-28 16:12:30 +100056int
Linus Torvalds1da177e2005-04-16 15:20:36 -070057xfs_open(
Christoph Hellwig993386c2007-08-28 16:12:30 +100058 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -070059{
60 int mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
63 return XFS_ERROR(EIO);
64
65 /*
66 * If it's a directory with any blocks, read-ahead block 0
67 * as we're almost certain to have the next operation be a read there.
68 */
Christoph Hellwig993386c2007-08-28 16:12:30 +100069 if (S_ISDIR(ip->i_d.di_mode) && ip->i_d.di_nextents > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 mode = xfs_ilock_map_shared(ip);
71 if (ip->i_d.di_nextents > 0)
72 (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
73 xfs_iunlock(ip, mode);
74 }
75 return 0;
76}
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/*
79 * xfs_getattr
80 */
Christoph Hellwig993386c2007-08-28 16:12:30 +100081int
Linus Torvalds1da177e2005-04-16 15:20:36 -070082xfs_getattr(
Christoph Hellwig993386c2007-08-28 16:12:30 +100083 xfs_inode_t *ip,
Nathan Scott8285fb52006-06-09 17:07:12 +100084 bhv_vattr_t *vap,
Christoph Hellwig993386c2007-08-28 16:12:30 +100085 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
Christoph Hellwig993386c2007-08-28 16:12:30 +100087 bhv_vnode_t *vp = XFS_ITOV(ip);
88 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Lachlan McIlroycf441ee2008-02-07 16:42:19 +110090 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 if (XFS_FORCED_SHUTDOWN(mp))
93 return XFS_ERROR(EIO);
94
95 if (!(flags & ATTR_LAZY))
96 xfs_ilock(ip, XFS_ILOCK_SHARED);
97
Lachlan McIlroyba87ea62007-05-08 13:49:46 +100098 vap->va_size = XFS_ISIZE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 if (vap->va_mask == XFS_AT_SIZE)
100 goto all_done;
101
102 vap->va_nblocks =
103 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
104 vap->va_nodeid = ip->i_ino;
105#if XFS_BIG_INUMS
106 vap->va_nodeid += mp->m_inoadd;
107#endif
108 vap->va_nlink = ip->i_d.di_nlink;
109
110 /*
111 * Quick exit for non-stat callers
112 */
113 if ((vap->va_mask &
114 ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
115 XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
116 goto all_done;
117
118 /*
119 * Copy from in-core inode.
120 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000121 vap->va_mode = ip->i_d.di_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 vap->va_uid = ip->i_d.di_uid;
123 vap->va_gid = ip->i_d.di_gid;
124 vap->va_projid = ip->i_d.di_projid;
125
126 /*
127 * Check vnode type block/char vs. everything else.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000129 switch (ip->i_d.di_mode & S_IFMT) {
130 case S_IFBLK:
131 case S_IFCHR:
132 vap->va_rdev = ip->i_df.if_u2.if_rdev;
133 vap->va_blocksize = BLKDEV_IOSIZE;
134 break;
135 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 vap->va_rdev = 0;
137
Eric Sandeen71ddabb2007-11-23 16:29:42 +1100138 if (!(XFS_IS_REALTIME_INODE(ip))) {
David Chinnere8c8b3a2005-11-02 10:33:05 +1100139 vap->va_blocksize = xfs_preferred_iosize(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 } else {
141
142 /*
143 * If the file blocks are being allocated from a
144 * realtime partition, then return the inode's
145 * realtime extent size or the realtime volume's
146 * extent size.
147 */
David Chinner957d0eb2007-06-18 16:50:37 +1000148 vap->va_blocksize =
149 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 }
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000151 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 }
153
Nathan Scottca5ccbf2006-01-11 21:03:04 +1100154 vn_atime_to_timespec(vp, &vap->va_atime);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
156 vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
157 vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
158 vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
159
160 /*
161 * Exit for stat callers. See if any of the rest of the fields
162 * to be filled in are needed.
163 */
164 if ((vap->va_mask &
165 (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
166 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
167 goto all_done;
168
169 /*
170 * Convert di_flags to xflags.
171 */
172 vap->va_xflags = xfs_ip2xflags(ip);
173
174 /*
175 * Exit for inode revalidate. See if any of the rest of
176 * the fields to be filled in are needed.
177 */
178 if ((vap->va_mask &
179 (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
180 XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
181 goto all_done;
182
183 vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
184 vap->va_nextents =
185 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
186 ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
187 ip->i_d.di_nextents;
188 if (ip->i_afp)
189 vap->va_anextents =
190 (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
191 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
192 ip->i_d.di_anextents;
193 else
194 vap->va_anextents = 0;
195 vap->va_gen = ip->i_d.di_gen;
196
197 all_done:
198 if (!(flags & ATTR_LAZY))
199 xfs_iunlock(ip, XFS_ILOCK_SHARED);
200 return 0;
201}
202
203
204/*
205 * xfs_setattr
206 */
207int
208xfs_setattr(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000209 xfs_inode_t *ip,
Nathan Scott8285fb52006-06-09 17:07:12 +1000210 bhv_vattr_t *vap,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 int flags,
212 cred_t *credp)
213{
Christoph Hellwig993386c2007-08-28 16:12:30 +1000214 bhv_vnode_t *vp = XFS_ITOV(ip);
215 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 int mask;
218 int code;
219 uint lock_flags;
220 uint commit_flags=0;
221 uid_t uid=0, iuid=0;
222 gid_t gid=0, igid=0;
223 int timeflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 xfs_prid_t projid=0, iprojid=0;
225 int mandlock_before, mandlock_after;
226 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
227 int file_owner;
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700228 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100230 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000232 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return XFS_ERROR(EROFS);
234
235 /*
236 * Cannot set certain attributes.
237 */
238 mask = vap->va_mask;
239 if (mask & XFS_AT_NOSET) {
240 return XFS_ERROR(EINVAL);
241 }
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 if (XFS_FORCED_SHUTDOWN(mp))
244 return XFS_ERROR(EIO);
245
246 /*
247 * Timestamps do not need to be logged and hence do not
248 * need to be done within a transaction.
249 */
250 if (mask & XFS_AT_UPDTIMES) {
251 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
252 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
253 ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
254 ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
255 xfs_ichgtime(ip, timeflags);
256 return 0;
257 }
258
259 olddquot1 = olddquot2 = NULL;
260 udqp = gdqp = NULL;
261
262 /*
263 * If disk quotas is on, we make sure that the dquots do exist on disk,
264 * before we start any other transactions. Trying to do this later
265 * is messy. We don't care to take a readlock to look at the ids
266 * in inode here, because we can't hold it across the trans_reserve.
267 * If the IDs do change before we take the ilock, we're covered
268 * because the i_*dquot fields will get updated anyway.
269 */
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000270 if (XFS_IS_QUOTA_ON(mp) &&
271 (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 uint qflags = 0;
273
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000274 if ((mask & XFS_AT_UID) && XFS_IS_UQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 uid = vap->va_uid;
276 qflags |= XFS_QMOPT_UQUOTA;
277 } else {
278 uid = ip->i_d.di_uid;
279 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000280 if ((mask & XFS_AT_GID) && XFS_IS_GQUOTA_ON(mp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 gid = vap->va_gid;
282 qflags |= XFS_QMOPT_GQUOTA;
283 } else {
284 gid = ip->i_d.di_gid;
285 }
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000286 if ((mask & XFS_AT_PROJID) && XFS_IS_PQUOTA_ON(mp)) {
287 projid = vap->va_projid;
288 qflags |= XFS_QMOPT_PQUOTA;
289 } else {
290 projid = ip->i_d.di_projid;
291 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 /*
293 * We take a reference when we initialize udqp and gdqp,
294 * so it is important that we never blindly double trip on
295 * the same variable. See xfs_create() for an example.
296 */
297 ASSERT(udqp == NULL);
298 ASSERT(gdqp == NULL);
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000299 code = XFS_QM_DQVOPALLOC(mp, ip, uid, gid, projid, qflags,
300 &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 if (code)
Jesper Juhl014c2542006-01-15 02:37:08 +0100302 return code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 }
304
305 /*
306 * For the other attributes, we acquire the inode lock and
307 * first do an error checking pass.
308 */
309 tp = NULL;
310 lock_flags = XFS_ILOCK_EXCL;
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700311 if (flags & ATTR_NOLOCK)
312 need_iolock = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 if (!(mask & XFS_AT_SIZE)) {
314 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
315 (mp->m_flags & XFS_MOUNT_WSYNC)) {
316 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
317 commit_flags = 0;
318 if ((code = xfs_trans_reserve(tp, 0,
319 XFS_ICHANGE_LOG_RES(mp), 0,
320 0, 0))) {
321 lock_flags = 0;
322 goto error_return;
323 }
324 }
325 } else {
Christoph Hellwigeb9df392007-08-16 18:42:07 +1000326 if (DM_EVENT_ENABLED(ip, DM_EVENT_TRUNCATE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 !(flags & ATTR_DMI)) {
328 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +1100329 code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 vap->va_size, 0, dmflags, NULL);
331 if (code) {
332 lock_flags = 0;
333 goto error_return;
334 }
335 }
336 if (need_iolock)
337 lock_flags |= XFS_IOLOCK_EXCL;
338 }
339
340 xfs_ilock(ip, lock_flags);
341
342 /* boolean: are we the file owner? */
343 file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
344
345 /*
346 * Change various properties of a file.
347 * Only the owner or users with CAP_FOWNER
348 * capability may do these things.
349 */
350 if (mask &
351 (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
352 XFS_AT_GID|XFS_AT_PROJID)) {
353 /*
354 * CAP_FOWNER overrides the following restrictions:
355 *
356 * The user ID of the calling process must be equal
357 * to the file owner ID, except in cases where the
358 * CAP_FSETID capability is applicable.
359 */
360 if (!file_owner && !capable(CAP_FOWNER)) {
361 code = XFS_ERROR(EPERM);
362 goto error_return;
363 }
364
365 /*
366 * CAP_FSETID overrides the following restrictions:
367 *
368 * The effective user ID of the calling process shall match
369 * the file owner when setting the set-user-ID and
370 * set-group-ID bits on that file.
371 *
372 * The effective group ID or one of the supplementary group
373 * IDs of the calling process shall match the group owner of
374 * the file when setting the set-group-ID bit on that file
375 */
376 if (mask & XFS_AT_MODE) {
377 mode_t m = 0;
378
379 if ((vap->va_mode & S_ISUID) && !file_owner)
380 m |= S_ISUID;
381 if ((vap->va_mode & S_ISGID) &&
382 !in_group_p((gid_t)ip->i_d.di_gid))
383 m |= S_ISGID;
384#if 0
385 /* Linux allows this, Irix doesn't. */
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000386 if ((vap->va_mode & S_ISVTX) && !VN_ISDIR(vp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 m |= S_ISVTX;
388#endif
389 if (m && !capable(CAP_FSETID))
390 vap->va_mode &= ~m;
391 }
392 }
393
394 /*
395 * Change file ownership. Must be the owner or privileged.
396 * If the system was configured with the "restricted_chown"
397 * option, the owner is not permitted to give away the file,
398 * and can change the group id only to a group of which he
399 * or she is a member.
400 */
401 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
402 /*
403 * These IDs could have changed since we last looked at them.
404 * But, we're assured that if the ownership did change
405 * while we didn't have the inode locked, inode's dquot(s)
406 * would have changed also.
407 */
408 iuid = ip->i_d.di_uid;
409 iprojid = ip->i_d.di_projid;
410 igid = ip->i_d.di_gid;
411 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
412 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
413 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
414 iprojid;
415
416 /*
417 * CAP_CHOWN overrides the following restrictions:
418 *
419 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
420 * shall override the restriction that a process cannot
421 * change the user ID of a file it owns and the restriction
422 * that the group ID supplied to the chown() function
423 * shall be equal to either the group ID or one of the
424 * supplementary group IDs of the calling process.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 */
426 if (restricted_chown &&
427 (iuid != uid || (igid != gid &&
428 !in_group_p((gid_t)gid))) &&
429 !capable(CAP_CHOWN)) {
430 code = XFS_ERROR(EPERM);
431 goto error_return;
432 }
433 /*
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000434 * Do a quota reservation only if uid/projid/gid is actually
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 * going to change.
436 */
437 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000438 (XFS_IS_PQUOTA_ON(mp) && iprojid != projid) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
440 ASSERT(tp);
441 code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
442 capable(CAP_FOWNER) ?
443 XFS_QMOPT_FORCE_RES : 0);
444 if (code) /* out of quota */
445 goto error_return;
446 }
447 }
448
449 /*
450 * Truncate file. Must have write permission and not be a directory.
451 */
452 if (mask & XFS_AT_SIZE) {
453 /* Short circuit the truncate case for zero length files */
454 if ((vap->va_size == 0) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000455 (ip->i_size == 0) && (ip->i_d.di_nextents == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 xfs_iunlock(ip, XFS_ILOCK_EXCL);
457 lock_flags &= ~XFS_ILOCK_EXCL;
458 if (mask & XFS_AT_CTIME)
459 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
460 code = 0;
461 goto error_return;
462 }
463
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000464 if (VN_ISDIR(vp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 code = XFS_ERROR(EISDIR);
466 goto error_return;
Christoph Hellwig0432dab2005-09-02 16:46:51 +1000467 } else if (!VN_ISREG(vp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 code = XFS_ERROR(EINVAL);
469 goto error_return;
470 }
471 /*
472 * Make sure that the dquots are attached to the inode.
473 */
474 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
475 goto error_return;
476 }
477
478 /*
479 * Change file access or modified times.
480 */
481 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
482 if (!file_owner) {
483 if ((flags & ATTR_UTIME) &&
484 !capable(CAP_FOWNER)) {
485 code = XFS_ERROR(EPERM);
486 goto error_return;
487 }
488 }
489 }
490
491 /*
492 * Change extent size or realtime flag.
493 */
494 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
495 /*
496 * Can't change extent size if any extents are allocated.
497 */
Eric Sandeene94af022005-11-02 15:10:41 +1100498 if (ip->i_d.di_nextents && (mask & XFS_AT_EXTSIZE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
500 vap->va_extsize) ) {
501 code = XFS_ERROR(EINVAL); /* EFBIG? */
502 goto error_return;
503 }
504
505 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 * Can't change realtime flag if any extents are allocated.
507 */
Eric Sandeene94af022005-11-02 15:10:41 +1100508 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
509 (mask & XFS_AT_XFLAGS) &&
Eric Sandeen71ddabb2007-11-23 16:29:42 +1100510 (XFS_IS_REALTIME_INODE(ip)) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
512 code = XFS_ERROR(EINVAL); /* EFBIG? */
513 goto error_return;
514 }
515 /*
516 * Extent size must be a multiple of the appropriate block
517 * size, if set at all.
518 */
519 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
520 xfs_extlen_t size;
521
Eric Sandeen71ddabb2007-11-23 16:29:42 +1100522 if (XFS_IS_REALTIME_INODE(ip) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 ((mask & XFS_AT_XFLAGS) &&
524 (vap->va_xflags & XFS_XFLAG_REALTIME))) {
525 size = mp->m_sb.sb_rextsize <<
526 mp->m_sb.sb_blocklog;
527 } else {
528 size = mp->m_sb.sb_blocksize;
529 }
530 if (vap->va_extsize % size) {
531 code = XFS_ERROR(EINVAL);
532 goto error_return;
533 }
534 }
535 /*
536 * If realtime flag is set then must have realtime data.
537 */
538 if ((mask & XFS_AT_XFLAGS) &&
539 (vap->va_xflags & XFS_XFLAG_REALTIME)) {
540 if ((mp->m_sb.sb_rblocks == 0) ||
541 (mp->m_sb.sb_rextsize == 0) ||
542 (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
543 code = XFS_ERROR(EINVAL);
544 goto error_return;
545 }
546 }
547
548 /*
549 * Can't modify an immutable/append-only file unless
550 * we have appropriate permission.
551 */
552 if ((mask & XFS_AT_XFLAGS) &&
553 (ip->i_d.di_flags &
554 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
555 (vap->va_xflags &
556 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
557 !capable(CAP_LINUX_IMMUTABLE)) {
558 code = XFS_ERROR(EPERM);
559 goto error_return;
560 }
561 }
562
563 /*
564 * Now we can make the changes. Before we join the inode
565 * to the transaction, if XFS_AT_SIZE is set then take care of
566 * the part of the truncation that must be done without the
567 * inode lock. This needs to be done before joining the inode
568 * to the transaction, because the inode cannot be unlocked
569 * once it is a part of the transaction.
570 */
571 if (mask & XFS_AT_SIZE) {
572 code = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000573 if ((vap->va_size > ip->i_size) &&
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100574 (flags & ATTR_NOSIZETOK) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 code = xfs_igrow_start(ip, vap->va_size, credp);
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnerc32676e2007-07-19 16:28:58 +1000578
579 /*
580 * We are going to log the inode size change in this
581 * transaction so any previous writes that are beyond the on
582 * disk EOF and the new EOF that have not been written out need
583 * to be written here. If we do not write the data out, we
584 * expose ourselves to the null files problem.
585 *
586 * Only flush from the on disk size to the smaller of the in
587 * memory file size or the new size as that's the range we
588 * really care about here and prevents waiting for other data
589 * not within the range we care about here.
590 */
591 if (!code &&
592 (ip->i_size != ip->i_d.di_size) &&
593 (vap->va_size > ip->i_d.di_size)) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000594 code = xfs_flush_pages(ip,
David Chinnerc32676e2007-07-19 16:28:58 +1000595 ip->i_d.di_size, vap->va_size,
596 XFS_B_ASYNC, FI_NONE);
597 }
598
599 /* wait for all I/O to complete */
Christoph Hellwigb677c212007-08-29 11:46:28 +1000600 vn_iowait(ip);
David Chinnerc32676e2007-07-19 16:28:58 +1000601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 if (!code)
603 code = xfs_itruncate_data(ip, vap->va_size);
604 if (code) {
605 ASSERT(tp == NULL);
606 lock_flags &= ~XFS_ILOCK_EXCL;
607 ASSERT(lock_flags == XFS_IOLOCK_EXCL);
608 goto error_return;
609 }
610 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
611 if ((code = xfs_trans_reserve(tp, 0,
612 XFS_ITRUNCATE_LOG_RES(mp), 0,
613 XFS_TRANS_PERM_LOG_RES,
614 XFS_ITRUNCATE_LOG_COUNT))) {
615 xfs_trans_cancel(tp, 0);
616 if (need_iolock)
617 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
618 return code;
619 }
620 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
621 xfs_ilock(ip, XFS_ILOCK_EXCL);
622 }
623
624 if (tp) {
625 xfs_trans_ijoin(tp, ip, lock_flags);
626 xfs_trans_ihold(tp, ip);
627 }
628
629 /* determine whether mandatory locking mode changes */
630 mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
631
632 /*
633 * Truncate file. Must have write permission and not be a directory.
634 */
635 if (mask & XFS_AT_SIZE) {
David Chinner44d814c2008-03-06 13:45:29 +1100636 /*
637 * Only change the c/mtime if we are changing the size
638 * or we are explicitly asked to change it. This handles
639 * the semantic difference between truncate() and ftruncate()
640 * as implemented in the VFS.
641 */
642 if (vap->va_size != ip->i_size || (mask & XFS_AT_CTIME))
643 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
644
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000645 if (vap->va_size > ip->i_size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 xfs_igrow_finish(tp, ip, vap->va_size,
647 !(flags & ATTR_DMI));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000648 } else if ((vap->va_size <= ip->i_size) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 ((vap->va_size == 0) && ip->i_d.di_nextents)) {
650 /*
651 * signal a sync transaction unless
652 * we're truncating an already unlinked
653 * file on a wsync filesystem
654 */
655 code = xfs_itruncate_finish(&tp, ip,
656 (xfs_fsize_t)vap->va_size,
657 XFS_DATA_FORK,
658 ((ip->i_d.di_nlink != 0 ||
659 !(mp->m_flags & XFS_MOUNT_WSYNC))
660 ? 1 : 0));
Nathan Scott7d4fb402006-06-09 15:27:16 +1000661 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 goto abort_return;
Nathan Scott7d4fb402006-06-09 15:27:16 +1000663 /*
664 * Truncated "down", so we're removing references
665 * to old data here - if we now delay flushing for
666 * a long time, we expose ourselves unduly to the
667 * notorious NULL files problem. So, we mark this
668 * vnode and flush it when the file is closed, and
669 * do not wait the usual (long) time for writeout.
670 */
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000671 xfs_iflags_set(ip, XFS_ITRUNCATED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 }
674
675 /*
676 * Change file access modes.
677 */
678 if (mask & XFS_AT_MODE) {
679 ip->i_d.di_mode &= S_IFMT;
680 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
681
682 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
683 timeflags |= XFS_ICHGTIME_CHG;
684 }
685
686 /*
687 * Change file ownership. Must be the owner or privileged.
688 * If the system was configured with the "restricted_chown"
689 * option, the owner is not permitted to give away the file,
690 * and can change the group id only to a group of which he
691 * or she is a member.
692 */
693 if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_PROJID)) {
694 /*
695 * CAP_FSETID overrides the following restrictions:
696 *
697 * The set-user-ID and set-group-ID bits of a file will be
698 * cleared upon successful return from chown()
699 */
700 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
701 !capable(CAP_FSETID)) {
702 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
703 }
704
705 /*
706 * Change the ownerships and register quota modifications
707 * in the transaction.
708 */
709 if (iuid != uid) {
710 if (XFS_IS_UQUOTA_ON(mp)) {
711 ASSERT(mask & XFS_AT_UID);
712 ASSERT(udqp);
713 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
714 &ip->i_udquot, udqp);
715 }
716 ip->i_d.di_uid = uid;
717 }
718 if (igid != gid) {
719 if (XFS_IS_GQUOTA_ON(mp)) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000720 ASSERT(!XFS_IS_PQUOTA_ON(mp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 ASSERT(mask & XFS_AT_GID);
722 ASSERT(gdqp);
723 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
724 &ip->i_gdquot, gdqp);
725 }
726 ip->i_d.di_gid = gid;
727 }
728 if (iprojid != projid) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000729 if (XFS_IS_PQUOTA_ON(mp)) {
730 ASSERT(!XFS_IS_GQUOTA_ON(mp));
731 ASSERT(mask & XFS_AT_PROJID);
732 ASSERT(gdqp);
733 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
734 &ip->i_gdquot, gdqp);
735 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 ip->i_d.di_projid = projid;
737 /*
738 * We may have to rev the inode as well as
739 * the superblock version number since projids didn't
740 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
741 */
742 if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
743 xfs_bump_ino_vers2(tp, ip);
744 }
745
746 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
747 timeflags |= XFS_ICHGTIME_CHG;
748 }
749
750
751 /*
752 * Change file access or modified times.
753 */
754 if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
755 if (mask & XFS_AT_ATIME) {
756 ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
757 ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
758 ip->i_update_core = 1;
759 timeflags &= ~XFS_ICHGTIME_ACC;
760 }
761 if (mask & XFS_AT_MTIME) {
762 ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
763 ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
764 timeflags &= ~XFS_ICHGTIME_MOD;
765 timeflags |= XFS_ICHGTIME_CHG;
766 }
767 if (tp && (flags & ATTR_UTIME))
768 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
769 }
770
771 /*
772 * Change XFS-added attributes.
773 */
774 if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
775 if (mask & XFS_AT_EXTSIZE) {
776 /*
777 * Converting bytes to fs blocks.
778 */
779 ip->i_d.di_extsize = vap->va_extsize >>
780 mp->m_sb.sb_blocklog;
781 }
782 if (mask & XFS_AT_XFLAGS) {
783 uint di_flags;
784
785 /* can't set PREALLOC this way, just preserve it */
786 di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
787 if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
788 di_flags |= XFS_DIFLAG_IMMUTABLE;
789 if (vap->va_xflags & XFS_XFLAG_APPEND)
790 di_flags |= XFS_DIFLAG_APPEND;
791 if (vap->va_xflags & XFS_XFLAG_SYNC)
792 di_flags |= XFS_DIFLAG_SYNC;
793 if (vap->va_xflags & XFS_XFLAG_NOATIME)
794 di_flags |= XFS_DIFLAG_NOATIME;
795 if (vap->va_xflags & XFS_XFLAG_NODUMP)
796 di_flags |= XFS_DIFLAG_NODUMP;
Nathan Scott365ca832005-06-21 15:39:12 +1000797 if (vap->va_xflags & XFS_XFLAG_PROJINHERIT)
798 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000799 if (vap->va_xflags & XFS_XFLAG_NODEFRAG)
800 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000801 if (vap->va_xflags & XFS_XFLAG_FILESTREAM)
802 di_flags |= XFS_DIFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
804 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
805 di_flags |= XFS_DIFLAG_RTINHERIT;
806 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
807 di_flags |= XFS_DIFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100808 if (vap->va_xflags & XFS_XFLAG_EXTSZINHERIT)
809 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
810 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
Christoph Hellwig613d7042007-10-11 17:44:08 +1000811 if (vap->va_xflags & XFS_XFLAG_REALTIME)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100813 if (vap->va_xflags & XFS_XFLAG_EXTSIZE)
814 di_flags |= XFS_DIFLAG_EXTSIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 }
816 ip->i_d.di_flags = di_flags;
817 }
818 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
819 timeflags |= XFS_ICHGTIME_CHG;
820 }
821
822 /*
823 * Change file inode change time only if XFS_AT_CTIME set
824 * AND we have been called by a DMI function.
825 */
826
827 if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
828 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
829 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
830 ip->i_update_core = 1;
831 timeflags &= ~XFS_ICHGTIME_CHG;
832 }
833
834 /*
835 * Send out timestamp changes that need to be set to the
836 * current time. Not done when called by a DMI function.
837 */
838 if (timeflags && !(flags & ATTR_DMI))
839 xfs_ichgtime(ip, timeflags);
840
841 XFS_STATS_INC(xs_ig_attrchg);
842
843 /*
844 * If this is a synchronous mount, make sure that the
845 * transaction goes to disk before returning to the user.
846 * This is slightly sub-optimal in that truncates require
Nathan Scottc41564b2006-03-29 08:55:14 +1000847 * two sync transactions instead of one for wsync filesystems.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 * One for the truncate and one for the timestamps since we
849 * don't want to change the timestamps unless we're sure the
850 * truncate worked. Truncates are less than 1% of the laddis
851 * mix so this probably isn't worth the trouble to optimize.
852 */
853 code = 0;
854 if (tp) {
855 if (mp->m_flags & XFS_MOUNT_WSYNC)
856 xfs_trans_set_sync(tp);
857
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000858 code = xfs_trans_commit(tp, commit_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 }
860
861 /*
862 * If the (regular) file's mandatory locking mode changed, then
863 * notify the vnode. We do this under the inode lock to prevent
864 * racing calls to vop_vnode_change.
865 */
866 mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868 xfs_iunlock(ip, lock_flags);
869
870 /*
871 * Release any dquot(s) the inode had kept before chown.
872 */
873 XFS_QM_DQRELE(mp, olddquot1);
874 XFS_QM_DQRELE(mp, olddquot2);
875 XFS_QM_DQRELE(mp, udqp);
876 XFS_QM_DQRELE(mp, gdqp);
877
878 if (code) {
879 return code;
880 }
881
Christoph Hellwigeb9df392007-08-16 18:42:07 +1000882 if (DM_EVENT_ENABLED(ip, DM_EVENT_ATTRIBUTE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 !(flags & ATTR_DMI)) {
Christoph Hellwigbc4ac742008-03-06 13:45:58 +1100884 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, ip, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 NULL, DM_RIGHT_NULL, NULL, NULL,
886 0, 0, AT_DELAY_FLAG(flags));
887 }
888 return 0;
889
890 abort_return:
891 commit_flags |= XFS_TRANS_ABORT;
892 /* FALLTHROUGH */
893 error_return:
894 XFS_QM_DQRELE(mp, udqp);
895 XFS_QM_DQRELE(mp, gdqp);
896 if (tp) {
897 xfs_trans_cancel(tp, commit_flags);
898 }
899 if (lock_flags != 0) {
900 xfs_iunlock(ip, lock_flags);
901 }
902 return code;
903}
904
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905/*
Nathan Scott7d4fb402006-06-09 15:27:16 +1000906 * The maximum pathlen is 1024 bytes. Since the minimum file system
907 * blocksize is 512 bytes, we can get a max of 2 extents back from
908 * bmapi.
909 */
910#define SYMLINK_MAPS 2
911
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000912STATIC int
913xfs_readlink_bmap(
914 xfs_inode_t *ip,
915 char *link)
916{
917 xfs_mount_t *mp = ip->i_mount;
918 int pathlen = ip->i_d.di_size;
919 int nmaps = SYMLINK_MAPS;
920 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
921 xfs_daddr_t d;
922 int byte_cnt;
923 int n;
924 xfs_buf_t *bp;
925 int error = 0;
926
927 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
928 mval, &nmaps, NULL, NULL);
929 if (error)
930 goto out;
931
932 for (n = 0; n < nmaps; n++) {
933 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
934 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
935
936 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0);
937 error = XFS_BUF_GETERROR(bp);
938 if (error) {
939 xfs_ioerror_alert("xfs_readlink",
940 ip->i_mount, bp, XFS_BUF_ADDR(bp));
941 xfs_buf_relse(bp);
942 goto out;
943 }
944 if (pathlen < byte_cnt)
945 byte_cnt = pathlen;
946 pathlen -= byte_cnt;
947
948 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
949 xfs_buf_relse(bp);
950 }
951
952 link[ip->i_d.di_size] = '\0';
953 error = 0;
954
955 out:
956 return error;
957}
958
Christoph Hellwig993386c2007-08-28 16:12:30 +1000959int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960xfs_readlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000961 xfs_inode_t *ip,
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000962 char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000964 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 int pathlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100968 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
970 if (XFS_FORCED_SHUTDOWN(mp))
971 return XFS_ERROR(EIO);
972
973 xfs_ilock(ip, XFS_ILOCK_SHARED);
974
975 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000976 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000978 pathlen = ip->i_d.di_size;
979 if (!pathlen)
980 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982 if (ip->i_df.if_flags & XFS_IFINLINE) {
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000983 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
984 link[pathlen] = '\0';
985 } else {
986 error = xfs_readlink_bmap(ip, link);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 }
988
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000989 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 return error;
992}
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994/*
995 * xfs_fsync
996 *
997 * This is called to sync the inode and its data out to disk.
998 * We need to hold the I/O lock while flushing the data, and
999 * the inode lock while flushing the inode. The inode lock CANNOT
1000 * be held while flushing the data, so acquire after we're done
1001 * with that.
1002 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001003int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004xfs_fsync(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001005 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 int flag,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 xfs_off_t start,
1008 xfs_off_t stop)
1009{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 xfs_trans_t *tp;
1011 int error;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001012 int log_flushed = 0, changed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001014 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016 ASSERT(start >= 0 && stop >= -1);
1017
1018 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1019 return XFS_ERROR(EIO);
1020
Lachlan McIlroy776a75f2007-09-14 15:22:50 +10001021 if (flag & FSYNC_DATA)
1022 filemap_fdatawait(vn_to_inode(XFS_ITOV(ip))->i_mapping);
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 /*
1025 * We always need to make sure that the required inode state
1026 * is safe on disk. The vnode might be clean but because
1027 * of committed transactions that haven't hit the disk yet.
1028 * Likewise, there could be unflushed non-transactional
1029 * changes to the inode core that have to go to disk.
1030 *
1031 * The following code depends on one assumption: that
1032 * any transaction that changes an inode logs the core
1033 * because it has to change some field in the inode core
1034 * (typically nextents or nblocks). That assumption
1035 * implies that any transactions against an inode will
1036 * catch any non-transactional updates. If inode-altering
1037 * transactions exist that violate this assumption, the
1038 * code breaks. Right now, it figures that if the involved
1039 * update_* field is clear and the inode is unpinned, the
1040 * inode is clean. Either it's been flushed or it's been
1041 * committed and the commit has hit the disk unpinning the inode.
1042 * (Note that xfs_inode_item_format() called at commit clears
1043 * the update_* fields.)
1044 */
1045 xfs_ilock(ip, XFS_ILOCK_SHARED);
1046
1047 /* If we are flushing data then we care about update_size
1048 * being set, otherwise we care about update_core
1049 */
1050 if ((flag & FSYNC_DATA) ?
1051 (ip->i_update_size == 0) :
1052 (ip->i_update_core == 0)) {
1053 /*
1054 * Timestamps/size haven't changed since last inode
1055 * flush or inode transaction commit. That means
1056 * either nothing got written or a transaction
1057 * committed which caught the updates. If the
1058 * latter happened and the transaction hasn't
1059 * hit the disk yet, the inode will be still
1060 * be pinned. If it is, force the log.
1061 */
1062
1063 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1064
1065 if (xfs_ipincount(ip)) {
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001066 _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 XFS_LOG_FORCE |
1068 ((flag & FSYNC_WAIT)
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001069 ? XFS_LOG_SYNC : 0),
1070 &log_flushed);
1071 } else {
1072 /*
1073 * If the inode is not pinned and nothing
1074 * has changed we don't need to flush the
1075 * cache.
1076 */
1077 changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 }
1079 error = 0;
1080 } else {
1081 /*
1082 * Kick off a transaction to log the inode
1083 * core to get the updates. Make it
1084 * sync if FSYNC_WAIT is passed in (which
1085 * is done by everybody but specfs). The
1086 * sync transaction will also force the log.
1087 */
1088 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1089 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1090 if ((error = xfs_trans_reserve(tp, 0,
1091 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1092 0, 0, 0))) {
1093 xfs_trans_cancel(tp, 0);
1094 return error;
1095 }
1096 xfs_ilock(ip, XFS_ILOCK_EXCL);
1097
1098 /*
1099 * Note - it's possible that we might have pushed
1100 * ourselves out of the way during trans_reserve
1101 * which would flush the inode. But there's no
1102 * guarantee that the inode buffer has actually
1103 * gone out yet (it's delwri). Plus the buffer
1104 * could be pinned anyway if it's part of an
1105 * inode in another recent transaction. So we
1106 * play it safe and fire off the transaction anyway.
1107 */
1108 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1109 xfs_trans_ihold(tp, ip);
1110 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1111 if (flag & FSYNC_WAIT)
1112 xfs_trans_set_sync(tp);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001113 error = _xfs_trans_commit(tp, 0, &log_flushed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
1115 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1116 }
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001117
1118 if ((ip->i_mount->m_flags & XFS_MOUNT_BARRIER) && changed) {
1119 /*
1120 * If the log write didn't issue an ordered tag we need
1121 * to flush the disk cache for the data device now.
1122 */
1123 if (!log_flushed)
1124 xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp);
1125
1126 /*
1127 * If this inode is on the RT dev we need to flush that
Nathan Scottc41564b2006-03-29 08:55:14 +10001128 * cache as well.
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001129 */
Eric Sandeen71ddabb2007-11-23 16:29:42 +11001130 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001131 xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp);
1132 }
1133
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 return error;
1135}
1136
1137/*
David Chinner92dfe8d2007-05-24 15:22:19 +10001138 * This is called by xfs_inactive to free any blocks beyond eof
1139 * when the link count isn't zero and by xfs_dm_punch_hole() when
1140 * punching a hole to EOF.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 */
David Chinner92dfe8d2007-05-24 15:22:19 +10001142int
1143xfs_free_eofblocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 xfs_mount_t *mp,
David Chinner92dfe8d2007-05-24 15:22:19 +10001145 xfs_inode_t *ip,
1146 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147{
1148 xfs_trans_t *tp;
1149 int error;
1150 xfs_fileoff_t end_fsb;
1151 xfs_fileoff_t last_fsb;
1152 xfs_filblks_t map_len;
1153 int nimaps;
1154 xfs_bmbt_irec_t imap;
David Chinner92dfe8d2007-05-24 15:22:19 +10001155 int use_iolock = (flags & XFS_FREE_EOF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
1157 /*
1158 * Figure out if there are any blocks beyond the end
1159 * of the file. If not, then there is nothing to do.
1160 */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001161 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1163 map_len = last_fsb - end_fsb;
1164 if (map_len <= 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01001165 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
1167 nimaps = 1;
1168 xfs_ilock(ip, XFS_ILOCK_SHARED);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10001169 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001170 NULL, 0, &imap, &nimaps, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1172
1173 if (!error && (nimaps != 0) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001174 (imap.br_startblock != HOLESTARTBLOCK ||
1175 ip->i_delayed_blks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 /*
1177 * Attach the dquots to the inode up front.
1178 */
1179 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001180 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182 /*
1183 * There are blocks after the end of file.
1184 * Free them up now by truncating the file to
1185 * its current size.
1186 */
1187 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1188
1189 /*
1190 * Do the xfs_itruncate_start() call before
1191 * reserving any log space because
1192 * itruncate_start will call into the buffer
1193 * cache and we can't
1194 * do that within a transaction.
1195 */
David Chinner92dfe8d2007-05-24 15:22:19 +10001196 if (use_iolock)
1197 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001198 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001199 ip->i_size);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001200 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +10001201 xfs_trans_cancel(tp, 0);
David Chinner92dfe8d2007-05-24 15:22:19 +10001202 if (use_iolock)
1203 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001204 return error;
1205 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207 error = xfs_trans_reserve(tp, 0,
1208 XFS_ITRUNCATE_LOG_RES(mp),
1209 0, XFS_TRANS_PERM_LOG_RES,
1210 XFS_ITRUNCATE_LOG_COUNT);
1211 if (error) {
1212 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1213 xfs_trans_cancel(tp, 0);
1214 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001215 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
1217
1218 xfs_ilock(ip, XFS_ILOCK_EXCL);
1219 xfs_trans_ijoin(tp, ip,
1220 XFS_IOLOCK_EXCL |
1221 XFS_ILOCK_EXCL);
1222 xfs_trans_ihold(tp, ip);
1223
1224 error = xfs_itruncate_finish(&tp, ip,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001225 ip->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 XFS_DATA_FORK,
1227 0);
1228 /*
1229 * If we get an error at this point we
1230 * simply don't bother truncating the file.
1231 */
1232 if (error) {
1233 xfs_trans_cancel(tp,
1234 (XFS_TRANS_RELEASE_LOG_RES |
1235 XFS_TRANS_ABORT));
1236 } else {
1237 error = xfs_trans_commit(tp,
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001238 XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 }
David Chinner92dfe8d2007-05-24 15:22:19 +10001240 xfs_iunlock(ip, (use_iolock ? (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)
1241 : XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001243 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244}
1245
1246/*
1247 * Free a symlink that has blocks associated with it.
1248 */
1249STATIC int
1250xfs_inactive_symlink_rmt(
1251 xfs_inode_t *ip,
1252 xfs_trans_t **tpp)
1253{
1254 xfs_buf_t *bp;
1255 int committed;
1256 int done;
1257 int error;
1258 xfs_fsblock_t first_block;
1259 xfs_bmap_free_t free_list;
1260 int i;
1261 xfs_mount_t *mp;
1262 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1263 int nmaps;
1264 xfs_trans_t *ntp;
1265 int size;
1266 xfs_trans_t *tp;
1267
1268 tp = *tpp;
1269 mp = ip->i_mount;
1270 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1271 /*
1272 * We're freeing a symlink that has some
1273 * blocks allocated to it. Free the
1274 * blocks here. We know that we've got
1275 * either 1 or 2 extents and that we can
1276 * free them all in one bunmapi call.
1277 */
1278 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1279 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1280 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1281 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1282 xfs_trans_cancel(tp, 0);
1283 *tpp = NULL;
1284 return error;
1285 }
1286 /*
1287 * Lock the inode, fix the size, and join it to the transaction.
1288 * Hold it so in the normal path, we still have it locked for
1289 * the second transaction. In the error paths we need it
1290 * held so the cancel won't rele it, see below.
1291 */
1292 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1293 size = (int)ip->i_d.di_size;
1294 ip->i_d.di_size = 0;
1295 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1296 xfs_trans_ihold(tp, ip);
1297 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1298 /*
1299 * Find the block(s) so we can inval and unmap them.
1300 */
1301 done = 0;
1302 XFS_BMAP_INIT(&free_list, &first_block);
Tobias Klausere8c96f82006-03-24 03:15:34 -08001303 nmaps = ARRAY_SIZE(mval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1305 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001306 &free_list, NULL)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 goto error0;
1308 /*
1309 * Invalidate the block(s).
1310 */
1311 for (i = 0; i < nmaps; i++) {
1312 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1313 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1314 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1315 xfs_trans_binval(tp, bp);
1316 }
1317 /*
1318 * Unmap the dead block(s) to the free_list.
1319 */
1320 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001321 &first_block, &free_list, NULL, &done)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 goto error1;
1323 ASSERT(done);
1324 /*
1325 * Commit the first transaction. This logs the EFI and the inode.
1326 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001327 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 goto error1;
1329 /*
1330 * The transaction must have been committed, since there were
1331 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1332 * The new tp has the extent freeing and EFDs.
1333 */
1334 ASSERT(committed);
1335 /*
1336 * The first xact was committed, so add the inode to the new one.
1337 * Mark it dirty so it will be logged and moved forward in the log as
1338 * part of every commit.
1339 */
1340 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1341 xfs_trans_ihold(tp, ip);
1342 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1343 /*
1344 * Get a new, empty transaction to return to our caller.
1345 */
1346 ntp = xfs_trans_dup(tp);
1347 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001348 * Commit the transaction containing extent freeing and EFDs.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 * If we get an error on the commit here or on the reserve below,
1350 * we need to unlock the inode since the new transaction doesn't
1351 * have the inode attached.
1352 */
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001353 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 tp = ntp;
1355 if (error) {
1356 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1357 goto error0;
1358 }
1359 /*
1360 * Remove the memory for extent descriptions (just bookkeeping).
1361 */
1362 if (ip->i_df.if_bytes)
1363 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1364 ASSERT(ip->i_df.if_bytes == 0);
1365 /*
1366 * Put an itruncate log reservation in the new transaction
1367 * for our caller.
1368 */
1369 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1370 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1371 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1372 goto error0;
1373 }
1374 /*
1375 * Return with the inode locked but not joined to the transaction.
1376 */
1377 *tpp = tp;
1378 return 0;
1379
1380 error1:
1381 xfs_bmap_cancel(&free_list);
1382 error0:
1383 /*
1384 * Have to come here with the inode locked and either
1385 * (held and in the transaction) or (not in the transaction).
1386 * If the inode isn't held then cancel would iput it, but
1387 * that's wrong since this is inactive and the vnode ref
1388 * count is 0 already.
1389 * Cancel won't do anything to the inode if held, but it still
1390 * needs to be locked until the cancel is done, if it was
1391 * joined to the transaction.
1392 */
1393 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1394 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1395 *tpp = NULL;
1396 return error;
1397
1398}
1399
1400STATIC int
1401xfs_inactive_symlink_local(
1402 xfs_inode_t *ip,
1403 xfs_trans_t **tpp)
1404{
1405 int error;
1406
1407 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1408 /*
1409 * We're freeing a symlink which fit into
1410 * the inode. Just free the memory used
1411 * to hold the old symlink.
1412 */
1413 error = xfs_trans_reserve(*tpp, 0,
1414 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1415 0, XFS_TRANS_PERM_LOG_RES,
1416 XFS_ITRUNCATE_LOG_COUNT);
1417
1418 if (error) {
1419 xfs_trans_cancel(*tpp, 0);
1420 *tpp = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +01001421 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
1423 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1424
1425 /*
1426 * Zero length symlinks _can_ exist.
1427 */
1428 if (ip->i_df.if_bytes > 0) {
1429 xfs_idata_realloc(ip,
1430 -(ip->i_df.if_bytes),
1431 XFS_DATA_FORK);
1432 ASSERT(ip->i_df.if_bytes == 0);
1433 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001434 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435}
1436
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437STATIC int
1438xfs_inactive_attrs(
1439 xfs_inode_t *ip,
1440 xfs_trans_t **tpp)
1441{
1442 xfs_trans_t *tp;
1443 int error;
1444 xfs_mount_t *mp;
1445
1446 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1447 tp = *tpp;
1448 mp = ip->i_mount;
1449 ASSERT(ip->i_d.di_forkoff != 0);
David Chinnere5720ee2008-04-10 12:21:18 +10001450 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnere5720ee2008-04-10 12:21:18 +10001452 if (error)
1453 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
1455 error = xfs_attr_inactive(ip);
David Chinnere5720ee2008-04-10 12:21:18 +10001456 if (error)
1457 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1460 error = xfs_trans_reserve(tp, 0,
1461 XFS_IFREE_LOG_RES(mp),
1462 0, XFS_TRANS_PERM_LOG_RES,
1463 XFS_INACTIVE_LOG_COUNT);
David Chinnere5720ee2008-04-10 12:21:18 +10001464 if (error)
1465 goto error_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
1467 xfs_ilock(ip, XFS_ILOCK_EXCL);
1468 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1469 xfs_trans_ihold(tp, ip);
1470 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1471
1472 ASSERT(ip->i_d.di_anextents == 0);
1473
1474 *tpp = tp;
Jesper Juhl014c2542006-01-15 02:37:08 +01001475 return 0;
David Chinnere5720ee2008-04-10 12:21:18 +10001476
1477error_cancel:
1478 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1479 xfs_trans_cancel(tp, 0);
1480error_unlock:
1481 *tpp = NULL;
1482 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1483 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484}
1485
Christoph Hellwig993386c2007-08-28 16:12:30 +10001486int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487xfs_release(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001488 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001490 bhv_vnode_t *vp = XFS_ITOV(ip);
1491 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 int error;
1493
Nathan Scott7d4fb402006-06-09 15:27:16 +10001494 if (!VN_ISREG(vp) || (ip->i_d.di_mode == 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
1497 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001498 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 return 0;
1500
David Chinner2a82b8b2007-07-11 11:09:12 +10001501 if (!XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001502 int truncated;
1503
David Chinner2a82b8b2007-07-11 11:09:12 +10001504 /*
1505 * If we are using filestreams, and we have an unlinked
1506 * file that we are processing the last close on, then nothing
1507 * will be able to reopen and write to this file. Purge this
1508 * inode from the filestreams cache so that it doesn't delay
1509 * teardown of the inode.
1510 */
1511 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1512 xfs_filestream_deassociate(ip);
1513
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +10001514 /*
1515 * If we previously truncated this file and removed old data
1516 * in the process, we want to initiate "early" writeout on
1517 * the last close. This is an attempt to combat the notorious
1518 * NULL files problem which is particularly noticable from a
1519 * truncate down, buffered (re-)write (delalloc), followed by
1520 * a crash. What we are effectively doing here is
1521 * significantly reducing the time window where we'd otherwise
1522 * be exposed to that problem.
1523 */
Christoph Hellwig09262b42007-08-29 11:44:50 +10001524 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001525 if (truncated && VN_DIRTY(vp) && ip->i_delayed_blks > 0)
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001526 xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +10001527 }
1528
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 if (ip->i_d.di_nlink != 0) {
1530 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001531 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001532 ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001534 (!(ip->i_d.di_flags &
1535 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
David Chinner92dfe8d2007-05-24 15:22:19 +10001536 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1537 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001538 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 }
1540 }
1541
1542 return 0;
1543}
1544
1545/*
1546 * xfs_inactive
1547 *
1548 * This is called when the vnode reference count for the vnode
1549 * goes to zero. If the file has been unlinked, then it must
1550 * now be truncated. Also, we clear all of the read-ahead state
1551 * kept for the inode here since the file is now closed.
1552 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001553int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554xfs_inactive(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001555 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001557 bhv_vnode_t *vp = XFS_ITOV(ip);
Nathan Scott67fcaa72006-06-09 17:00:52 +10001558 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 xfs_fsblock_t first_block;
1560 int committed;
1561 xfs_trans_t *tp;
1562 xfs_mount_t *mp;
1563 int error;
1564 int truncate;
1565
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001566 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 /*
1569 * If the inode is already free, then there can be nothing
1570 * to clean up here.
1571 */
1572 if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1573 ASSERT(ip->i_df.if_real_bytes == 0);
1574 ASSERT(ip->i_df.if_broot_bytes == 0);
1575 return VN_INACTIVE_CACHE;
1576 }
1577
1578 /*
1579 * Only do a truncate if it's a regular file with
1580 * some actual space in it. It's OK to look at the
1581 * inode's fields without the lock because we're the
1582 * only one with a reference to the inode.
1583 */
1584 truncate = ((ip->i_d.di_nlink == 0) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001585 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1586 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1588
1589 mp = ip->i_mount;
1590
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11001591 if (ip->i_d.di_nlink == 0 && DM_EVENT_ENABLED(ip, DM_EVENT_DESTROY))
1592 XFS_SEND_DESTROY(mp, ip, DM_RIGHT_NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
1594 error = 0;
1595
1596 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001597 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 goto out;
1599
1600 if (ip->i_d.di_nlink != 0) {
1601 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001602 ((ip->i_size > 0) || (VN_CACHED(vp) > 0 ||
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001603 ip->i_delayed_blks > 0)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001604 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1605 (!(ip->i_d.di_flags &
1606 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1607 (ip->i_delayed_blks != 0)))) {
David Chinner92dfe8d2007-05-24 15:22:19 +10001608 error = xfs_free_eofblocks(mp, ip, XFS_FREE_EOF_LOCK);
1609 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001610 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 }
1612 goto out;
1613 }
1614
1615 ASSERT(ip->i_d.di_nlink == 0);
1616
1617 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001618 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
1620 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1621 if (truncate) {
1622 /*
1623 * Do the xfs_itruncate_start() call before
1624 * reserving any log space because itruncate_start
1625 * will call into the buffer cache and we can't
1626 * do that within a transaction.
1627 */
1628 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1629
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001630 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1631 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +10001632 xfs_trans_cancel(tp, 0);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001633 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1634 return VN_INACTIVE_CACHE;
1635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
1637 error = xfs_trans_reserve(tp, 0,
1638 XFS_ITRUNCATE_LOG_RES(mp),
1639 0, XFS_TRANS_PERM_LOG_RES,
1640 XFS_ITRUNCATE_LOG_COUNT);
1641 if (error) {
1642 /* Don't call itruncate_cleanup */
1643 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1644 xfs_trans_cancel(tp, 0);
1645 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001646 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 }
1648
1649 xfs_ilock(ip, XFS_ILOCK_EXCL);
1650 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1651 xfs_trans_ihold(tp, ip);
1652
1653 /*
1654 * normally, we have to run xfs_itruncate_finish sync.
1655 * But if filesystem is wsync and we're in the inactive
1656 * path, then we know that nlink == 0, and that the
1657 * xaction that made nlink == 0 is permanently committed
1658 * since xfs_remove runs as a synchronous transaction.
1659 */
1660 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1661 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1662
1663 if (error) {
1664 xfs_trans_cancel(tp,
1665 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1666 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001667 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 }
1669 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1670
1671 /*
1672 * If we get an error while cleaning up a
1673 * symlink we bail out.
1674 */
1675 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1676 xfs_inactive_symlink_rmt(ip, &tp) :
1677 xfs_inactive_symlink_local(ip, &tp);
1678
1679 if (error) {
1680 ASSERT(tp == NULL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001681 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683
1684 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1685 xfs_trans_ihold(tp, ip);
1686 } else {
1687 error = xfs_trans_reserve(tp, 0,
1688 XFS_IFREE_LOG_RES(mp),
1689 0, XFS_TRANS_PERM_LOG_RES,
1690 XFS_INACTIVE_LOG_COUNT);
1691 if (error) {
1692 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1693 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001694 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 }
1696
1697 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1698 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1699 xfs_trans_ihold(tp, ip);
1700 }
1701
1702 /*
1703 * If there are attributes associated with the file
1704 * then blow them away now. The code calls a routine
1705 * that recursively deconstructs the attribute fork.
1706 * We need to just commit the current transaction
1707 * because we can't use it for xfs_attr_inactive().
1708 */
1709 if (ip->i_d.di_anextents > 0) {
1710 error = xfs_inactive_attrs(ip, &tp);
1711 /*
1712 * If we got an error, the transaction is already
1713 * cancelled, and the inode is unlocked. Just get out.
1714 */
1715 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001716 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 } else if (ip->i_afp) {
1718 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1719 }
1720
1721 /*
1722 * Free the inode.
1723 */
1724 XFS_BMAP_INIT(&free_list, &first_block);
1725 error = xfs_ifree(tp, ip, &free_list);
1726 if (error) {
1727 /*
1728 * If we fail to free the inode, shut down. The cancel
1729 * might do that, we need to make sure. Otherwise the
1730 * inode might be lost for a long time or forever.
1731 */
1732 if (!XFS_FORCED_SHUTDOWN(mp)) {
1733 cmn_err(CE_NOTE,
1734 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1735 error, mp->m_fsname);
Nathan Scott7d04a332006-06-09 14:58:38 +10001736 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 }
1738 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1739 } else {
1740 /*
1741 * Credit the quota account(s). The inode is gone.
1742 */
1743 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1744
1745 /*
1746 * Just ignore errors at this point. There is
1747 * nothing we can do except to try to keep going.
1748 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001749 (void) xfs_bmap_finish(&tp, &free_list, &committed);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001750 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 }
1752 /*
1753 * Release the dquots held by inode, if any.
1754 */
1755 XFS_QM_DQDETACH(mp, ip);
1756
1757 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1758
1759 out:
1760 return VN_INACTIVE_CACHE;
1761}
1762
1763
Christoph Hellwig993386c2007-08-28 16:12:30 +10001764int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765xfs_lookup(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001766 xfs_inode_t *dp,
Nathan Scott8285fb52006-06-09 17:07:12 +10001767 bhv_vname_t *dentry,
Christoph Hellwigef1f5e72008-03-06 13:46:25 +11001768 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001770 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 xfs_ino_t e_inum;
1772 int error;
1773 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001775 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776
1777 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1778 return XFS_ERROR(EIO);
1779
1780 lock_mode = xfs_ilock_map_shared(dp);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001781 error = xfs_dir_lookup_int(dp, lock_mode, dentry, &e_inum, &ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 if (!error) {
Christoph Hellwigef1f5e72008-03-06 13:46:25 +11001783 *ipp = ip;
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001784 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 }
1786 xfs_iunlock_map_shared(dp, lock_mode);
1787 return error;
1788}
1789
Christoph Hellwig993386c2007-08-28 16:12:30 +10001790int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791xfs_create(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001792 xfs_inode_t *dp,
Nathan Scott8285fb52006-06-09 17:07:12 +10001793 bhv_vname_t *dentry,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001794 mode_t mode,
1795 xfs_dev_t rdev,
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001796 xfs_inode_t **ipp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 cred_t *credp)
1798{
1799 char *name = VNAME(dentry);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001800 xfs_mount_t *mp = dp->i_mount;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001801 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 int error;
1804 xfs_bmap_free_t free_list;
1805 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001806 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 int dm_event_sent = 0;
1808 uint cancel_flags;
1809 int committed;
1810 xfs_prid_t prid;
1811 struct xfs_dquot *udqp, *gdqp;
1812 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 int namelen;
1814
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001815 ASSERT(!*ipp);
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001816 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 namelen = VNAMELEN(dentry);
1819
Christoph Hellwigeb9df392007-08-16 18:42:07 +10001820 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11001822 dp, DM_RIGHT_NULL, NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 DM_RIGHT_NULL, name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001824 mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
1826 if (error)
1827 return error;
1828 dm_event_sent = 1;
1829 }
1830
1831 if (XFS_FORCED_SHUTDOWN(mp))
1832 return XFS_ERROR(EIO);
1833
1834 /* Return through std_return after this point. */
1835
1836 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10001837 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1838 prid = dp->i_d.di_projid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 else
1840 prid = (xfs_prid_t)dfltprid;
1841
1842 /*
1843 * Make sure that we have allocated dquot(s) on disk.
1844 */
1845 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001846 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1848 if (error)
1849 goto std_return;
1850
1851 ip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
1853 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1854 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1855 resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1856 /*
1857 * Initially assume that the file does not exist and
1858 * reserve the resources for that case. If that is not
1859 * the case we'll drop the one we have and get a more
1860 * appropriate transaction later.
1861 */
1862 error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1863 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1864 if (error == ENOSPC) {
1865 resblks = 0;
1866 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1867 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1868 }
1869 if (error) {
1870 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 goto error_return;
1872 }
1873
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001874 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001875 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
1877 XFS_BMAP_INIT(&free_list, &first_block);
1878
1879 ASSERT(ip == NULL);
1880
1881 /*
1882 * Reserve disk quota and the inode.
1883 */
1884 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1885 if (error)
1886 goto error_return;
1887
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001888 if (resblks == 0 && (error = xfs_dir_canenter(tp, dp, name, namelen)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 goto error_return;
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001890 error = xfs_dir_ialloc(&tp, dp, mode, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 rdev, credp, prid, resblks > 0,
1892 &ip, &committed);
1893 if (error) {
1894 if (error == ENOSPC)
1895 goto error_return;
1896 goto abort_return;
1897 }
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11001898 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
1900 /*
1901 * At this point, we've gotten a newly allocated inode.
1902 * It is locked (and joined to the transaction).
1903 */
1904
1905 ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1906
1907 /*
Christoph Hellwig993386c2007-08-28 16:12:30 +10001908 * Now we join the directory inode to the transaction. We do not do it
1909 * earlier because xfs_dir_ialloc might commit the previous transaction
1910 * (and release all the locks). An error from here on will result in
1911 * the transaction cancel unlocking dp so don't do it explicitly in the
1912 * error path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 */
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001914 IHOLD(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001916 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001918 error = xfs_dir_createname(tp, dp, name, namelen, ip->i_ino,
1919 &first_block, &free_list, resblks ?
1920 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 if (error) {
1922 ASSERT(error != ENOSPC);
1923 goto abort_return;
1924 }
1925 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1926 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1927
1928 /*
1929 * If this is a synchronous mount, make sure that the
1930 * create transaction goes to disk before returning to
1931 * the user.
1932 */
1933 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1934 xfs_trans_set_sync(tp);
1935 }
1936
1937 dp->i_gen++;
1938
1939 /*
1940 * Attach the dquot(s) to the inodes and modify them incore.
1941 * These ids of the inode couldn't have changed since the new
1942 * inode has been locked ever since it was created.
1943 */
1944 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
1945
1946 /*
1947 * xfs_trans_commit normally decrements the vnode ref count
1948 * when it unlocks the inode. Since we want to return the
1949 * vnode to the caller, we bump the vnode ref count now.
1950 */
1951 IHOLD(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001953 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 if (error) {
1955 xfs_bmap_cancel(&free_list);
1956 goto abort_rele;
1957 }
1958
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001959 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 if (error) {
1961 IRELE(ip);
1962 tp = NULL;
1963 goto error_return;
1964 }
1965
1966 XFS_QM_DQRELE(mp, udqp);
1967 XFS_QM_DQRELE(mp, gdqp);
1968
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001969 *ipp = ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
1971 /* Fallthrough to std_return with error = 0 */
1972
1973std_return:
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001974 if ((*ipp || (error != 0 && dm_event_sent != 0)) &&
Christoph Hellwig993386c2007-08-28 16:12:30 +10001975 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11001977 dp, DM_RIGHT_NULL,
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001978 *ipp ? ip : NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 DM_RIGHT_NULL, name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001980 mode, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 }
1982 return error;
1983
1984 abort_return:
1985 cancel_flags |= XFS_TRANS_ABORT;
1986 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Jesper Juhl014c2542006-01-15 02:37:08 +01001988 error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 if (tp != NULL)
1990 xfs_trans_cancel(tp, cancel_flags);
1991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 XFS_QM_DQRELE(mp, udqp);
1993 XFS_QM_DQRELE(mp, gdqp);
1994
Christoph Hellwig993386c2007-08-28 16:12:30 +10001995 if (unlock_dp_on_error)
1996 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1997
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 goto std_return;
1999
2000 abort_rele:
2001 /*
2002 * Wait until after the current transaction is aborted to
2003 * release the inode. This prevents recursive transactions
2004 * and deadlocks from xfs_inactive.
2005 */
2006 cancel_flags |= XFS_TRANS_ABORT;
2007 xfs_trans_cancel(tp, cancel_flags);
2008 IRELE(ip);
2009
2010 XFS_QM_DQRELE(mp, udqp);
2011 XFS_QM_DQRELE(mp, gdqp);
2012
2013 goto std_return;
2014}
2015
2016#ifdef DEBUG
2017/*
2018 * Some counters to see if (and how often) we are hitting some deadlock
2019 * prevention code paths.
2020 */
2021
2022int xfs_rm_locks;
2023int xfs_rm_lock_delays;
2024int xfs_rm_attempts;
2025#endif
2026
2027/*
2028 * The following routine will lock the inodes associated with the
2029 * directory and the named entry in the directory. The locks are
2030 * acquired in increasing inode number.
2031 *
2032 * If the entry is "..", then only the directory is locked. The
2033 * vnode ref count will still include that from the .. entry in
2034 * this case.
2035 *
2036 * There is a deadlock we need to worry about. If the locked directory is
2037 * in the AIL, it might be blocking up the log. The next inode we lock
2038 * could be already locked by another thread waiting for log space (e.g
2039 * a permanent log reservation with a long running transaction (see
2040 * xfs_itruncate_finish)). To solve this, we must check if the directory
2041 * is in the ail and use lock_nowait. If we can't lock, we need to
2042 * drop the inode lock on the directory and try again. xfs_iunlock will
2043 * potentially push the tail if we were holding up the log.
2044 */
2045STATIC int
2046xfs_lock_dir_and_entry(
2047 xfs_inode_t *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 xfs_inode_t *ip) /* inode of entry 'name' */
2049{
2050 int attempts;
2051 xfs_ino_t e_inum;
2052 xfs_inode_t *ips[2];
2053 xfs_log_item_t *lp;
2054
2055#ifdef DEBUG
2056 xfs_rm_locks++;
2057#endif
2058 attempts = 0;
2059
2060again:
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002061 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
2063 e_inum = ip->i_ino;
2064
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002065 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
2067 /*
2068 * We want to lock in increasing inum. Since we've already
2069 * acquired the lock on the directory, we may need to release
2070 * if if the inum of the entry turns out to be less.
2071 */
2072 if (e_inum > dp->i_ino) {
2073 /*
2074 * We are already in the right order, so just
2075 * lock on the inode of the entry.
2076 * We need to use nowait if dp is in the AIL.
2077 */
2078
2079 lp = (xfs_log_item_t *)dp->i_itemp;
2080 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2081 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2082 attempts++;
2083#ifdef DEBUG
2084 xfs_rm_attempts++;
2085#endif
2086
2087 /*
2088 * Unlock dp and try again.
2089 * xfs_iunlock will try to push the tail
2090 * if the inode is in the AIL.
2091 */
2092
2093 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2094
2095 if ((attempts % 5) == 0) {
2096 delay(1); /* Don't just spin the CPU */
2097#ifdef DEBUG
2098 xfs_rm_lock_delays++;
2099#endif
2100 }
2101 goto again;
2102 }
2103 } else {
2104 xfs_ilock(ip, XFS_ILOCK_EXCL);
2105 }
2106 } else if (e_inum < dp->i_ino) {
2107 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2108
2109 ips[0] = ip;
2110 ips[1] = dp;
2111 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2112 }
2113 /* else e_inum == dp->i_ino */
2114 /* This can happen if we're asked to lock /x/..
2115 * the entry is "..", which is also the parent directory.
2116 */
2117
2118 return 0;
2119}
2120
2121#ifdef DEBUG
2122int xfs_locked_n;
2123int xfs_small_retries;
2124int xfs_middle_retries;
2125int xfs_lots_retries;
2126int xfs_lock_delays;
2127#endif
2128
2129/*
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002130 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2131 * a different value
2132 */
2133static inline int
2134xfs_lock_inumorder(int lock_mode, int subclass)
2135{
2136 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10002137 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002138 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10002139 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002140
2141 return lock_mode;
2142}
2143
2144/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 * The following routine will lock n inodes in exclusive mode.
2146 * We assume the caller calls us with the inodes in i_ino order.
2147 *
2148 * We need to detect deadlock where an inode that we lock
2149 * is in the AIL and we start waiting for another inode that is locked
2150 * by a thread in a long running transaction (such as truncate). This can
2151 * result in deadlock since the long running trans might need to wait
2152 * for the inode we just locked in order to push the tail and free space
2153 * in the log.
2154 */
2155void
2156xfs_lock_inodes(
2157 xfs_inode_t **ips,
2158 int inodes,
2159 int first_locked,
2160 uint lock_mode)
2161{
2162 int attempts = 0, i, j, try_lock;
2163 xfs_log_item_t *lp;
2164
2165 ASSERT(ips && (inodes >= 2)); /* we need at least two */
2166
2167 if (first_locked) {
2168 try_lock = 1;
2169 i = 1;
2170 } else {
2171 try_lock = 0;
2172 i = 0;
2173 }
2174
2175again:
2176 for (; i < inodes; i++) {
2177 ASSERT(ips[i]);
2178
2179 if (i && (ips[i] == ips[i-1])) /* Already locked */
2180 continue;
2181
2182 /*
2183 * If try_lock is not set yet, make sure all locked inodes
2184 * are not in the AIL.
2185 * If any are, set try_lock to be used later.
2186 */
2187
2188 if (!try_lock) {
2189 for (j = (i - 1); j >= 0 && !try_lock; j--) {
2190 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2191 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2192 try_lock++;
2193 }
2194 }
2195 }
2196
2197 /*
2198 * If any of the previous locks we have locked is in the AIL,
2199 * we must TRY to get the second and subsequent locks. If
2200 * we can't get any, we must release all we have
2201 * and try again.
2202 */
2203
2204 if (try_lock) {
2205 /* try_lock must be 0 if i is 0. */
2206 /*
2207 * try_lock means we have an inode locked
2208 * that is in the AIL.
2209 */
2210 ASSERT(i != 0);
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002211 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 attempts++;
2213
2214 /*
2215 * Unlock all previous guys and try again.
2216 * xfs_iunlock will try to push the tail
2217 * if the inode is in the AIL.
2218 */
2219
2220 for(j = i - 1; j >= 0; j--) {
2221
2222 /*
2223 * Check to see if we've already
2224 * unlocked this one.
2225 * Not the first one going back,
2226 * and the inode ptr is the same.
2227 */
2228 if ((j != (i - 1)) && ips[j] ==
2229 ips[j+1])
2230 continue;
2231
2232 xfs_iunlock(ips[j], lock_mode);
2233 }
2234
2235 if ((attempts % 5) == 0) {
2236 delay(1); /* Don't just spin the CPU */
2237#ifdef DEBUG
2238 xfs_lock_delays++;
2239#endif
2240 }
2241 i = 0;
2242 try_lock = 0;
2243 goto again;
2244 }
2245 } else {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002246 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 }
2248 }
2249
2250#ifdef DEBUG
2251 if (attempts) {
2252 if (attempts < 5) xfs_small_retries++;
2253 else if (attempts < 100) xfs_middle_retries++;
2254 else xfs_lots_retries++;
2255 } else {
2256 xfs_locked_n++;
2257 }
2258#endif
2259}
2260
2261#ifdef DEBUG
2262#define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2263int remove_which_error_return = 0;
2264#else /* ! DEBUG */
2265#define REMOVE_DEBUG_TRACE(x)
2266#endif /* ! DEBUG */
2267
Christoph Hellwig993386c2007-08-28 16:12:30 +10002268int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269xfs_remove(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002270 xfs_inode_t *dp,
2271 bhv_vname_t *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 char *name = VNAME(dentry);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002274 xfs_mount_t *mp = dp->i_mount;
Christoph Hellwig43973962008-03-06 13:44:50 +11002275 xfs_inode_t *ip = VNAME_TO_INODE(dentry);
2276 int namelen = VNAMELEN(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 xfs_trans_t *tp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 int error = 0;
2279 xfs_bmap_free_t free_list;
2280 xfs_fsblock_t first_block;
2281 int cancel_flags;
2282 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 int link_zero;
2284 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002286 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 if (XFS_FORCED_SHUTDOWN(mp))
2289 return XFS_ERROR(EIO);
2290
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002291 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002292 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
Christoph Hellwig43973962008-03-06 13:44:50 +11002294 name, NULL, ip->i_d.di_mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 if (error)
2296 return error;
2297 }
2298
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 /*
2300 * We need to get a reference to ip before we get our log
2301 * reservation. The reason for this is that we cannot call
2302 * xfs_iget for an inode for which we do not have a reference
2303 * once we've acquired a log reservation. This is because the
2304 * inode we are trying to get might be in xfs_inactive going
2305 * for a log reservation. Since we'll have to wait for the
2306 * inactive code to complete before returning from xfs_iget,
2307 * we need to make sure that we don't have log space reserved
Nathan Scottc41564b2006-03-29 08:55:14 +10002308 * when we call xfs_iget. Instead we get an unlocked reference
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 * to the inode before getting our log reservation.
2310 */
Christoph Hellwig43973962008-03-06 13:44:50 +11002311 IHOLD(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002313 xfs_itrace_entry(ip);
2314 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315
2316 error = XFS_QM_DQATTACH(mp, dp, 0);
2317 if (!error && dp != ip)
2318 error = XFS_QM_DQATTACH(mp, ip, 0);
2319 if (error) {
2320 REMOVE_DEBUG_TRACE(__LINE__);
2321 IRELE(ip);
2322 goto std_return;
2323 }
2324
2325 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2326 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2327 /*
2328 * We try to get the real space reservation first,
2329 * allowing for directory btree deletion(s) implying
2330 * possible bmap insert(s). If we can't get the space
2331 * reservation then we use 0 instead, and avoid the bmap
2332 * btree insert(s) in the directory code by, if the bmap
2333 * insert tries to happen, instead trimming the LAST
2334 * block from the directory.
2335 */
2336 resblks = XFS_REMOVE_SPACE_RES(mp);
2337 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2338 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2339 if (error == ENOSPC) {
2340 resblks = 0;
2341 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2342 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2343 }
2344 if (error) {
2345 ASSERT(error != ENOSPC);
2346 REMOVE_DEBUG_TRACE(__LINE__);
2347 xfs_trans_cancel(tp, 0);
2348 IRELE(ip);
2349 return error;
2350 }
2351
Eric Sandeene9ed9d22007-05-08 13:48:56 +10002352 error = xfs_lock_dir_and_entry(dp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 if (error) {
2354 REMOVE_DEBUG_TRACE(__LINE__);
2355 xfs_trans_cancel(tp, cancel_flags);
2356 IRELE(ip);
2357 goto std_return;
2358 }
2359
2360 /*
2361 * At this point, we've gotten both the directory and the entry
2362 * inodes locked.
2363 */
2364 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2365 if (dp != ip) {
2366 /*
2367 * Increment vnode ref count only in this case since
2368 * there's an extra vnode reference in the case where
2369 * dp == ip.
2370 */
2371 IHOLD(dp);
2372 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2373 }
2374
2375 /*
2376 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2377 */
2378 XFS_BMAP_INIT(&free_list, &first_block);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002379 error = xfs_dir_removename(tp, dp, name, namelen, ip->i_ino,
2380 &first_block, &free_list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 if (error) {
2382 ASSERT(error != ENOENT);
2383 REMOVE_DEBUG_TRACE(__LINE__);
2384 goto error1;
2385 }
2386 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2387
2388 dp->i_gen++;
2389 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2390
2391 error = xfs_droplink(tp, ip);
2392 if (error) {
2393 REMOVE_DEBUG_TRACE(__LINE__);
2394 goto error1;
2395 }
2396
2397 /* Determine if this is the last link while
2398 * we are in the transaction.
2399 */
2400 link_zero = (ip)->i_d.di_nlink==0;
2401
2402 /*
2403 * Take an extra ref on the inode so that it doesn't
2404 * go to xfs_inactive() from within the commit.
2405 */
2406 IHOLD(ip);
2407
2408 /*
2409 * If this is a synchronous mount, make sure that the
2410 * remove transaction goes to disk before returning to
2411 * the user.
2412 */
2413 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2414 xfs_trans_set_sync(tp);
2415 }
2416
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002417 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 if (error) {
2419 REMOVE_DEBUG_TRACE(__LINE__);
2420 goto error_rele;
2421 }
2422
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002423 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 if (error) {
2425 IRELE(ip);
2426 goto std_return;
2427 }
2428
2429 /*
David Chinner2a82b8b2007-07-11 11:09:12 +10002430 * If we are using filestreams, kill the stream association.
2431 * If the file is still open it may get a new one but that
2432 * will get killed on last close in xfs_close() so we don't
2433 * have to worry about that.
2434 */
2435 if (link_zero && xfs_inode_is_filestream(ip))
2436 xfs_filestream_deassociate(ip);
2437
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002438 xfs_itrace_exit(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 IRELE(ip);
2440
2441/* Fall through to std_return with error = 0 */
2442 std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002443 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002445 dp, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 NULL, DM_RIGHT_NULL,
Christoph Hellwig43973962008-03-06 13:44:50 +11002447 name, NULL, ip->i_d.di_mode, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 }
2449 return error;
2450
2451 error1:
2452 xfs_bmap_cancel(&free_list);
2453 cancel_flags |= XFS_TRANS_ABORT;
2454 xfs_trans_cancel(tp, cancel_flags);
2455 goto std_return;
2456
2457 error_rele:
2458 /*
2459 * In this case make sure to not release the inode until after
2460 * the current transaction is aborted. Releasing it beforehand
2461 * can cause us to go to xfs_inactive and start a recursive
2462 * transaction which can easily deadlock with the current one.
2463 */
2464 xfs_bmap_cancel(&free_list);
2465 cancel_flags |= XFS_TRANS_ABORT;
2466 xfs_trans_cancel(tp, cancel_flags);
2467
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 IRELE(ip);
2469
2470 goto std_return;
2471}
2472
Christoph Hellwig993386c2007-08-28 16:12:30 +10002473int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474xfs_link(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002475 xfs_inode_t *tdp,
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002476 xfs_inode_t *sip,
Christoph Hellwig993386c2007-08-28 16:12:30 +10002477 bhv_vname_t *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002479 xfs_mount_t *mp = tdp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 xfs_inode_t *ips[2];
2482 int error;
2483 xfs_bmap_free_t free_list;
2484 xfs_fsblock_t first_block;
2485 int cancel_flags;
2486 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 int resblks;
2488 char *target_name = VNAME(dentry);
2489 int target_namelen;
2490
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002491 xfs_itrace_entry(tdp);
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002492 xfs_itrace_entry(sip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493
2494 target_namelen = VNAMELEN(dentry);
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002495 ASSERT(!S_ISDIR(sip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 if (XFS_FORCED_SHUTDOWN(mp))
2498 return XFS_ERROR(EIO);
2499
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002500 if (DM_EVENT_ENABLED(tdp, DM_EVENT_LINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002502 tdp, DM_RIGHT_NULL,
2503 sip, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 target_name, NULL, 0, 0, 0);
2505 if (error)
2506 return error;
2507 }
2508
2509 /* Return through std_return after this point. */
2510
2511 error = XFS_QM_DQATTACH(mp, sip, 0);
2512 if (!error && sip != tdp)
2513 error = XFS_QM_DQATTACH(mp, tdp, 0);
2514 if (error)
2515 goto std_return;
2516
2517 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2518 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2519 resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2520 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2521 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2522 if (error == ENOSPC) {
2523 resblks = 0;
2524 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2525 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2526 }
2527 if (error) {
2528 cancel_flags = 0;
2529 goto error_return;
2530 }
2531
2532 if (sip->i_ino < tdp->i_ino) {
2533 ips[0] = sip;
2534 ips[1] = tdp;
2535 } else {
2536 ips[0] = tdp;
2537 ips[1] = sip;
2538 }
2539
2540 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2541
2542 /*
2543 * Increment vnode ref counts since xfs_trans_commit &
2544 * xfs_trans_cancel will both unlock the inodes and
2545 * decrement the associated ref counts.
2546 */
Christoph Hellwiga3da7892008-03-06 13:46:12 +11002547 IHOLD(sip);
2548 IHOLD(tdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2550 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2551
2552 /*
2553 * If the source has too many links, we can't make any more to it.
2554 */
2555 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2556 error = XFS_ERROR(EMLINK);
2557 goto error_return;
2558 }
2559
Nathan Scott365ca832005-06-21 15:39:12 +10002560 /*
2561 * If we are using project inheritance, we only allow hard link
2562 * creation in our tree when the project IDs are the same; else
2563 * the tree quota mechanism could be circumvented.
2564 */
2565 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2566 (tdp->i_d.di_projid != sip->i_d.di_projid))) {
Nathan Scottb1ecdda2006-05-08 19:51:42 +10002567 error = XFS_ERROR(EXDEV);
Nathan Scott365ca832005-06-21 15:39:12 +10002568 goto error_return;
2569 }
2570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 if (resblks == 0 &&
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002572 (error = xfs_dir_canenter(tp, tdp, target_name, target_namelen)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 goto error_return;
2574
2575 XFS_BMAP_INIT(&free_list, &first_block);
2576
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002577 error = xfs_dir_createname(tp, tdp, target_name, target_namelen,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 sip->i_ino, &first_block, &free_list,
2579 resblks);
2580 if (error)
2581 goto abort_return;
2582 xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2583 tdp->i_gen++;
2584 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2585
2586 error = xfs_bumplink(tp, sip);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002587 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 goto abort_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
2590 /*
2591 * If this is a synchronous mount, make sure that the
2592 * link transaction goes to disk before returning to
2593 * the user.
2594 */
2595 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2596 xfs_trans_set_sync(tp);
2597 }
2598
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002599 error = xfs_bmap_finish (&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 if (error) {
2601 xfs_bmap_cancel(&free_list);
2602 goto abort_return;
2603 }
2604
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002605 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002606 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608
2609 /* Fall through to std_return with error = 0. */
2610std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002611 if (DM_EVENT_ENABLED(sip, DM_EVENT_POSTLINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002613 tdp, DM_RIGHT_NULL,
2614 sip, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 target_name, NULL, 0, error, 0);
2616 }
2617 return error;
2618
2619 abort_return:
2620 cancel_flags |= XFS_TRANS_ABORT;
2621 /* FALLTHROUGH */
Jesper Juhl014c2542006-01-15 02:37:08 +01002622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 error_return:
2624 xfs_trans_cancel(tp, cancel_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 goto std_return;
2626}
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10002627
2628
Christoph Hellwig993386c2007-08-28 16:12:30 +10002629int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630xfs_mkdir(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002631 xfs_inode_t *dp,
Nathan Scott8285fb52006-06-09 17:07:12 +10002632 bhv_vname_t *dentry,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002633 mode_t mode,
Christoph Hellwig979ebab2008-03-06 13:46:05 +11002634 xfs_inode_t **ipp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 cred_t *credp)
2636{
2637 char *dir_name = VNAME(dentry);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002638 int dir_namelen = VNAMELEN(dentry);
2639 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 xfs_inode_t *cdp; /* inode of created dir */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 int cancel_flags;
2643 int error;
2644 int committed;
2645 xfs_bmap_free_t free_list;
2646 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10002647 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 boolean_t created = B_FALSE;
2649 int dm_event_sent = 0;
2650 xfs_prid_t prid;
2651 struct xfs_dquot *udqp, *gdqp;
2652 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653
2654 if (XFS_FORCED_SHUTDOWN(mp))
2655 return XFS_ERROR(EIO);
2656
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 tp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002659 if (DM_EVENT_ENABLED(dp, DM_EVENT_CREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002661 dp, DM_RIGHT_NULL, NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 DM_RIGHT_NULL, dir_name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002663 mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 if (error)
2665 return error;
2666 dm_event_sent = 1;
2667 }
2668
2669 /* Return through std_return after this point. */
2670
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002671 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
2673 mp = dp->i_mount;
2674 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10002675 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
2676 prid = dp->i_d.di_projid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 else
2678 prid = (xfs_prid_t)dfltprid;
2679
2680 /*
2681 * Make sure that we have allocated dquot(s) on disk.
2682 */
2683 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002684 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2686 if (error)
2687 goto std_return;
2688
2689 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2690 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2691 resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2692 error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2693 XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2694 if (error == ENOSPC) {
2695 resblks = 0;
2696 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2697 XFS_TRANS_PERM_LOG_RES,
2698 XFS_MKDIR_LOG_COUNT);
2699 }
2700 if (error) {
2701 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 goto error_return;
2703 }
2704
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002705 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002706 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
2708 /*
2709 * Check for directory link count overflow.
2710 */
2711 if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2712 error = XFS_ERROR(EMLINK);
2713 goto error_return;
2714 }
2715
2716 /*
2717 * Reserve disk quota and the inode.
2718 */
2719 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2720 if (error)
2721 goto error_return;
2722
2723 if (resblks == 0 &&
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002724 (error = xfs_dir_canenter(tp, dp, dir_name, dir_namelen)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 goto error_return;
2726 /*
2727 * create the directory inode.
2728 */
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002729 error = xfs_dir_ialloc(&tp, dp, mode, 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 0, credp, prid, resblks > 0,
2731 &cdp, NULL);
2732 if (error) {
2733 if (error == ENOSPC)
2734 goto error_return;
2735 goto abort_return;
2736 }
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002737 xfs_itrace_ref(cdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738
2739 /*
2740 * Now we add the directory inode to the transaction.
2741 * We waited until now since xfs_dir_ialloc might start
2742 * a new transaction. Had we joined the transaction
Christoph Hellwig993386c2007-08-28 16:12:30 +10002743 * earlier, the locks might have gotten released. An error
2744 * from here on will result in the transaction cancel
2745 * unlocking dp so don't do it explicitly in the error path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 */
Christoph Hellwig979ebab2008-03-06 13:46:05 +11002747 IHOLD(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002749 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
2751 XFS_BMAP_INIT(&free_list, &first_block);
2752
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002753 error = xfs_dir_createname(tp, dp, dir_name, dir_namelen, cdp->i_ino,
2754 &first_block, &free_list, resblks ?
2755 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 if (error) {
2757 ASSERT(error != ENOSPC);
2758 goto error1;
2759 }
2760 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2761
2762 /*
2763 * Bump the in memory version number of the parent directory
2764 * so that other processes accessing it will recognize that
2765 * the directory has changed.
2766 */
2767 dp->i_gen++;
2768
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002769 error = xfs_dir_init(tp, cdp, dp);
2770 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772
2773 cdp->i_gen = 1;
2774 error = xfs_bumplink(tp, dp);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002775 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 created = B_TRUE;
2779
Christoph Hellwig979ebab2008-03-06 13:46:05 +11002780 *ipp = cdp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 IHOLD(cdp);
2782
2783 /*
2784 * Attach the dquots to the new inode and modify the icount incore.
2785 */
2786 XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2787
2788 /*
2789 * If this is a synchronous mount, make sure that the
2790 * mkdir transaction goes to disk before returning to
2791 * the user.
2792 */
2793 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2794 xfs_trans_set_sync(tp);
2795 }
2796
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002797 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 if (error) {
2799 IRELE(cdp);
2800 goto error2;
2801 }
2802
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002803 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 XFS_QM_DQRELE(mp, udqp);
2805 XFS_QM_DQRELE(mp, gdqp);
2806 if (error) {
2807 IRELE(cdp);
2808 }
2809
2810 /* Fall through to std_return with error = 0 or errno from
2811 * xfs_trans_commit. */
2812
2813std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002814 if ((created || (error != 0 && dm_event_sent != 0)) &&
Christoph Hellwig993386c2007-08-28 16:12:30 +10002815 DM_EVENT_ENABLED(dp, DM_EVENT_POSTCREATE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002817 dp, DM_RIGHT_NULL,
2818 created ? cdp : NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 DM_RIGHT_NULL,
2820 dir_name, NULL,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10002821 mode, error, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 }
2823 return error;
2824
2825 error2:
2826 error1:
2827 xfs_bmap_cancel(&free_list);
2828 abort_return:
2829 cancel_flags |= XFS_TRANS_ABORT;
2830 error_return:
2831 xfs_trans_cancel(tp, cancel_flags);
2832 XFS_QM_DQRELE(mp, udqp);
2833 XFS_QM_DQRELE(mp, gdqp);
2834
Christoph Hellwig993386c2007-08-28 16:12:30 +10002835 if (unlock_dp_on_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 xfs_iunlock(dp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
2838 goto std_return;
2839}
2840
Christoph Hellwig993386c2007-08-28 16:12:30 +10002841int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842xfs_rmdir(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002843 xfs_inode_t *dp,
2844 bhv_vname_t *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002846 bhv_vnode_t *dir_vp = XFS_ITOV(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 char *name = VNAME(dentry);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002848 int namelen = VNAMELEN(dentry);
2849 xfs_mount_t *mp = dp->i_mount;
Christoph Hellwig43973962008-03-06 13:44:50 +11002850 xfs_inode_t *cdp = VNAME_TO_INODE(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 int error;
2853 xfs_bmap_free_t free_list;
2854 xfs_fsblock_t first_block;
2855 int cancel_flags;
2856 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 int last_cdp_link;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 uint resblks;
2859
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002860 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861
Christoph Hellwig993386c2007-08-28 16:12:30 +10002862 if (XFS_FORCED_SHUTDOWN(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
Christoph Hellwigeb9df392007-08-16 18:42:07 +10002865 if (DM_EVENT_ENABLED(dp, DM_EVENT_REMOVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11002867 dp, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 NULL, DM_RIGHT_NULL,
Christoph Hellwig43973962008-03-06 13:44:50 +11002869 name, NULL, cdp->i_d.di_mode, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 if (error)
2871 return XFS_ERROR(error);
2872 }
2873
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 /*
2875 * We need to get a reference to cdp before we get our log
2876 * reservation. The reason for this is that we cannot call
2877 * xfs_iget for an inode for which we do not have a reference
2878 * once we've acquired a log reservation. This is because the
2879 * inode we are trying to get might be in xfs_inactive going
2880 * for a log reservation. Since we'll have to wait for the
2881 * inactive code to complete before returning from xfs_iget,
2882 * we need to make sure that we don't have log space reserved
Nathan Scottc41564b2006-03-29 08:55:14 +10002883 * when we call xfs_iget. Instead we get an unlocked reference
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 * to the inode before getting our log reservation.
2885 */
Christoph Hellwig43973962008-03-06 13:44:50 +11002886 IHOLD(cdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887
2888 /*
2889 * Get the dquots for the inodes.
2890 */
2891 error = XFS_QM_DQATTACH(mp, dp, 0);
2892 if (!error && dp != cdp)
2893 error = XFS_QM_DQATTACH(mp, cdp, 0);
2894 if (error) {
2895 IRELE(cdp);
2896 REMOVE_DEBUG_TRACE(__LINE__);
2897 goto std_return;
2898 }
2899
2900 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2901 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2902 /*
2903 * We try to get the real space reservation first,
2904 * allowing for directory btree deletion(s) implying
2905 * possible bmap insert(s). If we can't get the space
2906 * reservation then we use 0 instead, and avoid the bmap
2907 * btree insert(s) in the directory code by, if the bmap
2908 * insert tries to happen, instead trimming the LAST
2909 * block from the directory.
2910 */
2911 resblks = XFS_REMOVE_SPACE_RES(mp);
2912 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2913 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
2914 if (error == ENOSPC) {
2915 resblks = 0;
2916 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2917 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
2918 }
2919 if (error) {
2920 ASSERT(error != ENOSPC);
2921 cancel_flags = 0;
2922 IRELE(cdp);
2923 goto error_return;
2924 }
2925 XFS_BMAP_INIT(&free_list, &first_block);
2926
2927 /*
2928 * Now lock the child directory inode and the parent directory
2929 * inode in the proper order. This will take care of validating
2930 * that the directory entry for the child directory inode has
2931 * not changed while we were obtaining a log reservation.
2932 */
Eric Sandeene9ed9d22007-05-08 13:48:56 +10002933 error = xfs_lock_dir_and_entry(dp, cdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 if (error) {
2935 xfs_trans_cancel(tp, cancel_flags);
2936 IRELE(cdp);
2937 goto std_return;
2938 }
2939
2940 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2941 if (dp != cdp) {
2942 /*
2943 * Only increment the parent directory vnode count if
2944 * we didn't bump it in looking up cdp. The only time
2945 * we don't bump it is when we're looking up ".".
2946 */
2947 VN_HOLD(dir_vp);
2948 }
2949
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002950 xfs_itrace_ref(cdp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
2952
2953 ASSERT(cdp->i_d.di_nlink >= 2);
2954 if (cdp->i_d.di_nlink != 2) {
2955 error = XFS_ERROR(ENOTEMPTY);
2956 goto error_return;
2957 }
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002958 if (!xfs_dir_isempty(cdp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 error = XFS_ERROR(ENOTEMPTY);
2960 goto error_return;
2961 }
2962
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002963 error = xfs_dir_removename(tp, dp, name, namelen, cdp->i_ino,
2964 &first_block, &free_list, resblks);
2965 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967
2968 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2969
2970 /*
2971 * Bump the in memory generation count on the parent
2972 * directory so that other can know that it has changed.
2973 */
2974 dp->i_gen++;
2975
2976 /*
2977 * Drop the link from cdp's "..".
2978 */
2979 error = xfs_droplink(tp, dp);
2980 if (error) {
2981 goto error1;
2982 }
2983
2984 /*
2985 * Drop the link from dp to cdp.
2986 */
2987 error = xfs_droplink(tp, cdp);
2988 if (error) {
2989 goto error1;
2990 }
2991
2992 /*
2993 * Drop the "." link from cdp to self.
2994 */
2995 error = xfs_droplink(tp, cdp);
2996 if (error) {
2997 goto error1;
2998 }
2999
3000 /* Determine these before committing transaction */
3001 last_cdp_link = (cdp)->i_d.di_nlink==0;
3002
3003 /*
3004 * Take an extra ref on the child vnode so that it
3005 * does not go to xfs_inactive() from within the commit.
3006 */
3007 IHOLD(cdp);
3008
3009 /*
3010 * If this is a synchronous mount, make sure that the
3011 * rmdir transaction goes to disk before returning to
3012 * the user.
3013 */
3014 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3015 xfs_trans_set_sync(tp);
3016 }
3017
Eric Sandeenf7c99b62007-02-10 18:37:16 +11003018 error = xfs_bmap_finish (&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 if (error) {
3020 xfs_bmap_cancel(&free_list);
3021 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3022 XFS_TRANS_ABORT));
3023 IRELE(cdp);
3024 goto std_return;
3025 }
3026
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003027 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 if (error) {
3029 IRELE(cdp);
3030 goto std_return;
3031 }
3032
3033
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 IRELE(cdp);
3035
3036 /* Fall through to std_return with error = 0 or the errno
3037 * from xfs_trans_commit. */
Nathan Scottbb19fba2006-03-22 14:12:12 +11003038 std_return:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10003039 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTREMOVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003041 dp, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 NULL, DM_RIGHT_NULL,
Christoph Hellwig43973962008-03-06 13:44:50 +11003043 name, NULL, cdp->i_d.di_mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 error, 0);
3045 }
3046 return error;
3047
Nathan Scottbb19fba2006-03-22 14:12:12 +11003048 error1:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 xfs_bmap_cancel(&free_list);
3050 cancel_flags |= XFS_TRANS_ABORT;
Jesper Juhl014c2542006-01-15 02:37:08 +01003051 /* FALLTHROUGH */
3052
Nathan Scottbb19fba2006-03-22 14:12:12 +11003053 error_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 xfs_trans_cancel(tp, cancel_flags);
3055 goto std_return;
3056}
3057
Christoph Hellwig993386c2007-08-28 16:12:30 +10003058int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059xfs_symlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +10003060 xfs_inode_t *dp,
Nathan Scott8285fb52006-06-09 17:07:12 +10003061 bhv_vname_t *dentry,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062 char *target_path,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10003063 mode_t mode,
Christoph Hellwig3937be52008-03-06 13:46:19 +11003064 xfs_inode_t **ipp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 cred_t *credp)
3066{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003067 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 xfs_inode_t *ip;
3070 int error;
3071 int pathlen;
3072 xfs_bmap_free_t free_list;
3073 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10003074 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 uint cancel_flags;
3076 int committed;
3077 xfs_fileoff_t first_fsb;
3078 xfs_filblks_t fs_blocks;
3079 int nmaps;
3080 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
3081 xfs_daddr_t d;
3082 char *cur_chunk;
3083 int byte_cnt;
3084 int n;
3085 xfs_buf_t *bp;
3086 xfs_prid_t prid;
3087 struct xfs_dquot *udqp, *gdqp;
3088 uint resblks;
3089 char *link_name = VNAME(dentry);
3090 int link_namelen;
3091
Christoph Hellwig3937be52008-03-06 13:46:19 +11003092 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093 error = 0;
3094 ip = NULL;
3095 tp = NULL;
3096
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003097 xfs_itrace_entry(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098
3099 if (XFS_FORCED_SHUTDOWN(mp))
3100 return XFS_ERROR(EIO);
3101
3102 link_namelen = VNAMELEN(dentry);
3103
3104 /*
3105 * Check component lengths of the target path name.
3106 */
3107 pathlen = strlen(target_path);
3108 if (pathlen >= MAXPATHLEN) /* total string too long */
3109 return XFS_ERROR(ENAMETOOLONG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
Christoph Hellwigeb9df392007-08-16 18:42:07 +10003111 if (DM_EVENT_ENABLED(dp, DM_EVENT_SYMLINK)) {
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003112 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3114 link_name, target_path, 0, 0, 0);
3115 if (error)
3116 return error;
3117 }
3118
3119 /* Return through std_return after this point. */
3120
3121 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10003122 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
3123 prid = dp->i_d.di_projid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124 else
3125 prid = (xfs_prid_t)dfltprid;
3126
3127 /*
3128 * Make sure that we have allocated dquot(s) on disk.
3129 */
3130 error = XFS_QM_DQVOPALLOC(mp, dp,
Nathan Scottc8ad20f2005-06-21 15:38:48 +10003131 current_fsuid(credp), current_fsgid(credp), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3133 if (error)
3134 goto std_return;
3135
3136 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3137 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3138 /*
3139 * The symlink will fit into the inode data fork?
3140 * There can't be any attributes so we get the whole variable part.
3141 */
3142 if (pathlen <= XFS_LITINO(mp))
3143 fs_blocks = 0;
3144 else
3145 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3146 resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3147 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3148 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3149 if (error == ENOSPC && fs_blocks == 0) {
3150 resblks = 0;
3151 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3152 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3153 }
3154 if (error) {
3155 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156 goto error_return;
3157 }
3158
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10003159 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10003160 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161
3162 /*
3163 * Check whether the directory allows new symlinks or not.
3164 */
3165 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3166 error = XFS_ERROR(EPERM);
3167 goto error_return;
3168 }
3169
3170 /*
3171 * Reserve disk quota : blocks and inode.
3172 */
3173 error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3174 if (error)
3175 goto error_return;
3176
3177 /*
3178 * Check for ability to enter directory entry, if no space reserved.
3179 */
3180 if (resblks == 0 &&
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003181 (error = xfs_dir_canenter(tp, dp, link_name, link_namelen)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 goto error_return;
3183 /*
3184 * Initialize the bmap freelist prior to calling either
3185 * bmapi or the directory create code.
3186 */
3187 XFS_BMAP_INIT(&free_list, &first_block);
3188
3189 /*
3190 * Allocate an inode for the symlink.
3191 */
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10003192 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193 1, 0, credp, prid, resblks > 0, &ip, NULL);
3194 if (error) {
3195 if (error == ENOSPC)
3196 goto error_return;
3197 goto error1;
3198 }
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003199 xfs_itrace_ref(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200
Christoph Hellwig993386c2007-08-28 16:12:30 +10003201 /*
3202 * An error after we've joined dp to the transaction will result in the
3203 * transaction cancel unlocking dp so don't do it explicitly in the
3204 * error path.
3205 */
Christoph Hellwig3937be52008-03-06 13:46:19 +11003206 IHOLD(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10003208 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209
3210 /*
3211 * Also attach the dquot(s) to it, if applicable.
3212 */
3213 XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3214
3215 if (resblks)
3216 resblks -= XFS_IALLOC_SPACE_RES(mp);
3217 /*
3218 * If the symlink will fit into the inode, write it inline.
3219 */
3220 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3221 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3222 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3223 ip->i_d.di_size = pathlen;
3224
3225 /*
3226 * The inode was initially created in extent format.
3227 */
3228 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3229 ip->i_df.if_flags |= XFS_IFINLINE;
3230
3231 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3232 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3233
3234 } else {
3235 first_fsb = 0;
3236 nmaps = SYMLINK_MAPS;
3237
3238 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3239 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3240 &first_block, resblks, mval, &nmaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003241 &free_list, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 if (error) {
3243 goto error1;
3244 }
3245
3246 if (resblks)
3247 resblks -= fs_blocks;
3248 ip->i_d.di_size = pathlen;
3249 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3250
3251 cur_chunk = target_path;
3252 for (n = 0; n < nmaps; n++) {
3253 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3254 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3255 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3256 BTOBB(byte_cnt), 0);
3257 ASSERT(bp && !XFS_BUF_GETERROR(bp));
3258 if (pathlen < byte_cnt) {
3259 byte_cnt = pathlen;
3260 }
3261 pathlen -= byte_cnt;
3262
3263 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3264 cur_chunk += byte_cnt;
3265
3266 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3267 }
3268 }
3269
3270 /*
3271 * Create the directory entry for the symlink.
3272 */
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10003273 error = xfs_dir_createname(tp, dp, link_name, link_namelen, ip->i_ino,
3274 &first_block, &free_list, resblks);
3275 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277 xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3278 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3279
3280 /*
3281 * Bump the in memory version number of the parent directory
3282 * so that other processes accessing it will recognize that
3283 * the directory has changed.
3284 */
3285 dp->i_gen++;
3286
3287 /*
3288 * If this is a synchronous mount, make sure that the
3289 * symlink transaction goes to disk before returning to
3290 * the user.
3291 */
3292 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3293 xfs_trans_set_sync(tp);
3294 }
3295
3296 /*
3297 * xfs_trans_commit normally decrements the vnode ref count
3298 * when it unlocks the inode. Since we want to return the
3299 * vnode to the caller, we bump the vnode ref count now.
3300 */
3301 IHOLD(ip);
3302
Eric Sandeenf7c99b62007-02-10 18:37:16 +11003303 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304 if (error) {
3305 goto error2;
3306 }
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003307 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 XFS_QM_DQRELE(mp, udqp);
3309 XFS_QM_DQRELE(mp, gdqp);
3310
3311 /* Fall through to std_return with error = 0 or errno from
3312 * xfs_trans_commit */
3313std_return:
Christoph Hellwig993386c2007-08-28 16:12:30 +10003314 if (DM_EVENT_ENABLED(dp, DM_EVENT_POSTSYMLINK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003316 dp, DM_RIGHT_NULL,
3317 error ? NULL : ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 DM_RIGHT_NULL, link_name, target_path,
3319 0, error, 0);
3320 }
3321
Christoph Hellwig3937be52008-03-06 13:46:19 +11003322 if (!error)
3323 *ipp = ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 return error;
3325
3326 error2:
3327 IRELE(ip);
3328 error1:
3329 xfs_bmap_cancel(&free_list);
3330 cancel_flags |= XFS_TRANS_ABORT;
3331 error_return:
3332 xfs_trans_cancel(tp, cancel_flags);
3333 XFS_QM_DQRELE(mp, udqp);
3334 XFS_QM_DQRELE(mp, gdqp);
3335
Christoph Hellwig993386c2007-08-28 16:12:30 +10003336 if (unlock_dp_on_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 xfs_iunlock(dp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338
3339 goto std_return;
3340}
3341
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343xfs_inode_flush(
Christoph Hellwig993386c2007-08-28 16:12:30 +10003344 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 int flags)
3346{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003347 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 int error = 0;
3349
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 if (XFS_FORCED_SHUTDOWN(mp))
3351 return XFS_ERROR(EIO);
3352
3353 /*
3354 * Bypass inodes which have already been cleaned by
3355 * the inode flush clustering code inside xfs_iflush
3356 */
David Chinner33540402008-03-06 13:43:59 +11003357 if (xfs_inode_clean(ip))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 return 0;
3359
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 /*
3361 * We make this non-blocking if the inode is contended,
3362 * return EAGAIN to indicate to the caller that they
3363 * did not succeed. This prevents the flush path from
3364 * blocking on inodes inside another operation right
3365 * now, they get caught later by xfs_sync.
3366 */
David Chinnera3f74ff2008-03-06 13:43:42 +11003367 if (flags & FLUSH_SYNC) {
3368 xfs_ilock(ip, XFS_ILOCK_SHARED);
3369 xfs_iflock(ip);
3370 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3371 if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3372 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 return EAGAIN;
3374 }
David Chinnera3f74ff2008-03-06 13:43:42 +11003375 } else {
3376 return EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 }
3378
David Chinnera3f74ff2008-03-06 13:43:42 +11003379 error = xfs_iflush(ip, (flags & FLUSH_SYNC) ? XFS_IFLUSH_SYNC
3380 : XFS_IFLUSH_ASYNC_NOBLOCK);
3381 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3382
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383 return error;
3384}
3385
Christoph Hellwig993386c2007-08-28 16:12:30 +10003386
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387int
Christoph Hellwig993386c2007-08-28 16:12:30 +10003388xfs_set_dmattrs(
3389 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 u_int evmask,
Christoph Hellwig993386c2007-08-28 16:12:30 +10003391 u_int16_t state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003393 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 int error;
3396
3397 if (!capable(CAP_SYS_ADMIN))
3398 return XFS_ERROR(EPERM);
3399
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 if (XFS_FORCED_SHUTDOWN(mp))
3401 return XFS_ERROR(EIO);
3402
3403 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3404 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3405 if (error) {
3406 xfs_trans_cancel(tp, 0);
3407 return error;
3408 }
3409 xfs_ilock(ip, XFS_ILOCK_EXCL);
3410 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3411
Christoph Hellwig613d7042007-10-11 17:44:08 +10003412 ip->i_d.di_dmevmask = evmask;
3413 ip->i_d.di_dmstate = state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414
3415 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3416 IHOLD(ip);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003417 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418
3419 return error;
3420}
3421
Christoph Hellwig993386c2007-08-28 16:12:30 +10003422int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423xfs_reclaim(
Christoph Hellwig993386c2007-08-28 16:12:30 +10003424 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425{
Christoph Hellwig993386c2007-08-28 16:12:30 +10003426 bhv_vnode_t *vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003428 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429
3430 ASSERT(!VN_MAPPED(vp));
3431
3432 /* bad inode, get out here ASAP */
3433 if (VN_BAD(vp)) {
3434 xfs_ireclaim(ip);
3435 return 0;
3436 }
3437
Christoph Hellwigb677c212007-08-29 11:46:28 +10003438 vn_iowait(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439
Christoph Hellwig51c91ed2005-09-02 16:58:38 +10003440 ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11003442 /*
3443 * Make sure the atime in the XFS inode is correct before freeing the
3444 * Linux inode.
3445 */
3446 xfs_synchronize_atime(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447
David Chinner4c606582006-11-11 18:05:00 +11003448 /*
3449 * If we have nothing to flush with this inode then complete the
3450 * teardown now, otherwise break the link between the xfs inode and the
3451 * linux inode and clean up the xfs inode later. This avoids flushing
3452 * the inode to disk during the delete operation itself.
3453 *
3454 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3455 * first to ensure that xfs_iunpin() will never see an xfs inode
3456 * that has a linux inode being reclaimed. Synchronisation is provided
3457 * by the i_flags_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 */
3459 if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3460 xfs_ilock(ip, XFS_ILOCK_EXCL);
3461 xfs_iflock(ip);
3462 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3463 } else {
3464 xfs_mount_t *mp = ip->i_mount;
3465
David Chinner4c606582006-11-11 18:05:00 +11003466 /* Protect sync and unpin from us */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467 XFS_MOUNT_ILOCK(mp);
David Chinner4c606582006-11-11 18:05:00 +11003468 spin_lock(&ip->i_flags_lock);
3469 __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
Christoph Hellwig739bfb22007-08-29 10:58:01 +10003470 vn_to_inode(vp)->i_private = NULL;
3471 ip->i_vnode = NULL;
David Chinner4c606582006-11-11 18:05:00 +11003472 spin_unlock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 XFS_MOUNT_IUNLOCK(mp);
3475 }
3476 return 0;
3477}
3478
3479int
3480xfs_finish_reclaim(
3481 xfs_inode_t *ip,
3482 int locked,
3483 int sync_mode)
3484{
David Chinnerda353b02007-08-28 14:00:13 +10003485 xfs_perag_t *pag = xfs_get_perag(ip->i_mount, ip->i_ino);
Nathan Scott67fcaa72006-06-09 17:00:52 +10003486 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 int error;
3488
3489 if (vp && VN_BAD(vp))
3490 goto reclaim;
3491
3492 /* The hash lock here protects a thread in xfs_iget_core from
3493 * racing with us on linking the inode back with a vnode.
3494 * Once we have the XFS_IRECLAIM flag set it will not touch
3495 * us.
3496 */
David Chinnerda353b02007-08-28 14:00:13 +10003497 write_lock(&pag->pag_ici_lock);
David Chinnerf273ab82006-09-28 11:06:03 +10003498 spin_lock(&ip->i_flags_lock);
David Chinner7a18c382006-11-11 18:04:54 +11003499 if (__xfs_iflags_test(ip, XFS_IRECLAIM) ||
3500 (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) && vp == NULL)) {
David Chinnerf273ab82006-09-28 11:06:03 +10003501 spin_unlock(&ip->i_flags_lock);
David Chinnerda353b02007-08-28 14:00:13 +10003502 write_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 if (locked) {
3504 xfs_ifunlock(ip);
3505 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3506 }
Jesper Juhl014c2542006-01-15 02:37:08 +01003507 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 }
David Chinner7a18c382006-11-11 18:04:54 +11003509 __xfs_iflags_set(ip, XFS_IRECLAIM);
David Chinnerf273ab82006-09-28 11:06:03 +10003510 spin_unlock(&ip->i_flags_lock);
David Chinnerda353b02007-08-28 14:00:13 +10003511 write_unlock(&pag->pag_ici_lock);
3512 xfs_put_perag(ip->i_mount, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513
3514 /*
3515 * If the inode is still dirty, then flush it out. If the inode
3516 * is not in the AIL, then it will be OK to flush it delwri as
3517 * long as xfs_iflush() does not keep any references to the inode.
3518 * We leave that decision up to xfs_iflush() since it has the
3519 * knowledge of whether it's OK to simply do a delwri flush of
3520 * the inode or whether we need to wait until the inode is
3521 * pulled from the AIL.
3522 * We get the flush lock regardless, though, just to make sure
3523 * we don't free it while it is being flushed.
3524 */
Lachlan McIlroy461aa8a2008-03-06 13:43:11 +11003525 if (!locked) {
3526 xfs_ilock(ip, XFS_ILOCK_EXCL);
3527 xfs_iflock(ip);
3528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529
Lachlan McIlroy461aa8a2008-03-06 13:43:11 +11003530 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531 if (ip->i_update_core ||
3532 ((ip->i_itemp != NULL) &&
3533 (ip->i_itemp->ili_format.ilf_fields != 0))) {
3534 error = xfs_iflush(ip, sync_mode);
3535 /*
3536 * If we hit an error, typically because of filesystem
3537 * shutdown, we don't need to let vn_reclaim to know
3538 * because we're gonna reclaim the inode anyway.
3539 */
3540 if (error) {
3541 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3542 goto reclaim;
3543 }
3544 xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3545 }
3546
3547 ASSERT(ip->i_update_core == 0);
3548 ASSERT(ip->i_itemp == NULL ||
3549 ip->i_itemp->ili_format.ilf_fields == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 }
3551
Lachlan McIlroy461aa8a2008-03-06 13:43:11 +11003552 xfs_ifunlock(ip);
3553 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3554
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 reclaim:
3556 xfs_ireclaim(ip);
3557 return 0;
3558}
3559
3560int
3561xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3562{
3563 int purged;
3564 xfs_inode_t *ip, *n;
3565 int done = 0;
3566
3567 while (!done) {
3568 purged = 0;
3569 XFS_MOUNT_ILOCK(mp);
3570 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
3571 if (noblock) {
3572 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
3573 continue;
3574 if (xfs_ipincount(ip) ||
3575 !xfs_iflock_nowait(ip)) {
3576 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3577 continue;
3578 }
3579 }
3580 XFS_MOUNT_IUNLOCK(mp);
Felix Blyakher6b2cf612005-11-25 16:42:13 +11003581 if (xfs_finish_reclaim(ip, noblock,
3582 XFS_IFLUSH_DELWRI_ELSE_ASYNC))
3583 delay(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584 purged = 1;
3585 break;
3586 }
3587
3588 done = !purged;
3589 }
3590
3591 XFS_MOUNT_IUNLOCK(mp);
3592 return 0;
3593}
3594
3595/*
3596 * xfs_alloc_file_space()
3597 * This routine allocates disk space for the given file.
3598 *
3599 * If alloc_type == 0, this request is for an ALLOCSP type
3600 * request which will change the file size. In this case, no
3601 * DMAPI event will be generated by the call. A TRUNCATE event
3602 * will be generated later by xfs_setattr.
3603 *
3604 * If alloc_type != 0, this request is for a RESVSP type
3605 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3606 * lower block boundary byte address is less than the file's
3607 * length.
3608 *
3609 * RETURNS:
3610 * 0 on success
3611 * errno on error
3612 *
3613 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003614STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615xfs_alloc_file_space(
3616 xfs_inode_t *ip,
3617 xfs_off_t offset,
3618 xfs_off_t len,
3619 int alloc_type,
3620 int attr_flags)
3621{
Nathan Scottdd9f4382006-01-11 15:28:28 +11003622 xfs_mount_t *mp = ip->i_mount;
3623 xfs_off_t count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003624 xfs_filblks_t allocated_fsb;
3625 xfs_filblks_t allocatesize_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003626 xfs_extlen_t extsz, temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 xfs_fileoff_t startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003628 xfs_fsblock_t firstfsb;
3629 int nimaps;
3630 int bmapi_flag;
3631 int quota_flag;
3632 int rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 xfs_trans_t *tp;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003634 xfs_bmbt_irec_t imaps[1], *imapp;
3635 xfs_bmap_free_t free_list;
3636 uint qblocks, resblks, resrtextents;
3637 int committed;
3638 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003640 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641
3642 if (XFS_FORCED_SHUTDOWN(mp))
3643 return XFS_ERROR(EIO);
3644
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3646 return error;
3647
3648 if (len <= 0)
3649 return XFS_ERROR(EINVAL);
3650
David Chinner957d0eb2007-06-18 16:50:37 +10003651 rt = XFS_IS_REALTIME_INODE(ip);
3652 extsz = xfs_get_extsz_hint(ip);
3653
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654 count = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 imapp = &imaps[0];
Nathan Scottdd9f4382006-01-11 15:28:28 +11003656 nimaps = 1;
3657 bmapi_flag = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
3659 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
3660
3661 /* Generate a DMAPI event if needed. */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003662 if (alloc_type != 0 && offset < ip->i_size &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 (attr_flags&ATTR_DMI) == 0 &&
Christoph Hellwigeb9df392007-08-16 18:42:07 +10003664 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665 xfs_off_t end_dmi_offset;
3666
3667 end_dmi_offset = offset+len;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003668 if (end_dmi_offset > ip->i_size)
3669 end_dmi_offset = ip->i_size;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003670 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip, offset,
3671 end_dmi_offset - offset, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01003673 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 }
3675
3676 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003677 * Allocate file space until done or until there is an error
Linus Torvalds1da177e2005-04-16 15:20:36 -07003678 */
3679retry:
3680 while (allocatesize_fsb && !error) {
Nathan Scottdd9f4382006-01-11 15:28:28 +11003681 xfs_fileoff_t s, e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682
Nathan Scottdd9f4382006-01-11 15:28:28 +11003683 /*
Nathan Scott3ddb8fa2006-01-11 15:33:02 +11003684 * Determine space reservations for data/realtime.
Nathan Scottdd9f4382006-01-11 15:28:28 +11003685 */
3686 if (unlikely(extsz)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 s = startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11003688 do_div(s, extsz);
3689 s *= extsz;
3690 e = startoffset_fsb + allocatesize_fsb;
3691 if ((temp = do_mod(startoffset_fsb, extsz)))
3692 e += temp;
3693 if ((temp = do_mod(e, extsz)))
3694 e += extsz - temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695 } else {
Nathan Scottdd9f4382006-01-11 15:28:28 +11003696 s = 0;
3697 e = allocatesize_fsb;
3698 }
3699
3700 if (unlikely(rt)) {
3701 resrtextents = qblocks = (uint)(e - s);
3702 resrtextents /= mp->m_sb.sb_rextsize;
3703 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
3704 quota_flag = XFS_QMOPT_RES_RTBLKS;
3705 } else {
3706 resrtextents = 0;
3707 resblks = qblocks = \
3708 XFS_DIOSTRAT_SPACE_RES(mp, (uint)(e - s));
3709 quota_flag = XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710 }
3711
3712 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003713 * Allocate and setup the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714 */
3715 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Nathan Scottdd9f4382006-01-11 15:28:28 +11003716 error = xfs_trans_reserve(tp, resblks,
3717 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 XFS_TRANS_PERM_LOG_RES,
3719 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003721 * Check for running out of space
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 */
3723 if (error) {
3724 /*
3725 * Free the transaction structure.
3726 */
3727 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
3728 xfs_trans_cancel(tp, 0);
3729 break;
3730 }
3731 xfs_ilock(ip, XFS_ILOCK_EXCL);
Nathan Scottdd9f4382006-01-11 15:28:28 +11003732 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip,
3733 qblocks, 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734 if (error)
3735 goto error1;
3736
3737 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3738 xfs_trans_ihold(tp, ip);
3739
3740 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003741 * Issue the xfs_bmapi() call to allocate the blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 */
3743 XFS_BMAP_INIT(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003744 error = xfs_bmapi(tp, ip, startoffset_fsb,
Nathan Scottdd9f4382006-01-11 15:28:28 +11003745 allocatesize_fsb, bmapi_flag,
3746 &firstfsb, 0, imapp, &nimaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003747 &free_list, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748 if (error) {
3749 goto error0;
3750 }
3751
3752 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11003753 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11003755 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 if (error) {
3757 goto error0;
3758 }
3759
Eric Sandeen1c72bf92007-05-08 13:48:42 +10003760 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003761 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3762 if (error) {
3763 break;
3764 }
3765
3766 allocated_fsb = imapp->br_blockcount;
3767
Nathan Scottdd9f4382006-01-11 15:28:28 +11003768 if (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 error = XFS_ERROR(ENOSPC);
3770 break;
3771 }
3772
3773 startoffset_fsb += allocated_fsb;
3774 allocatesize_fsb -= allocated_fsb;
3775 }
3776dmapi_enospc_check:
Christoph Hellwigeb9df392007-08-16 18:42:07 +10003777 if (error == ENOSPC && (attr_flags & ATTR_DMI) == 0 &&
3778 DM_EVENT_ENABLED(ip, DM_EVENT_NOSPACE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003780 ip, DM_RIGHT_NULL,
3781 ip, DM_RIGHT_NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
3783 if (error == 0)
3784 goto retry; /* Maybe DMAPI app. has made space */
3785 /* else fall through with error from XFS_SEND_DATA */
3786 }
3787
3788 return error;
3789
Nathan Scottdd9f4382006-01-11 15:28:28 +11003790error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 xfs_bmap_cancel(&free_list);
Nathan Scottdd9f4382006-01-11 15:28:28 +11003792 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp, tp, ip, qblocks, 0, quota_flag);
3793
3794error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
3796 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3797 goto dmapi_enospc_check;
3798}
3799
3800/*
3801 * Zero file bytes between startoff and endoff inclusive.
3802 * The iolock is held exclusive and no blocks are buffered.
3803 */
3804STATIC int
3805xfs_zero_remaining_bytes(
3806 xfs_inode_t *ip,
3807 xfs_off_t startoff,
3808 xfs_off_t endoff)
3809{
3810 xfs_bmbt_irec_t imap;
3811 xfs_fileoff_t offset_fsb;
3812 xfs_off_t lastoffset;
3813 xfs_off_t offset;
3814 xfs_buf_t *bp;
3815 xfs_mount_t *mp = ip->i_mount;
3816 int nimap;
3817 int error = 0;
3818
3819 bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
Eric Sandeen71ddabb2007-11-23 16:29:42 +11003820 XFS_IS_REALTIME_INODE(ip) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821 mp->m_rtdev_targp : mp->m_ddev_targp);
3822
3823 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
3824 offset_fsb = XFS_B_TO_FSBT(mp, offset);
3825 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003826 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003827 NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 if (error || nimap < 1)
3829 break;
3830 ASSERT(imap.br_blockcount >= 1);
3831 ASSERT(imap.br_startoff == offset_fsb);
3832 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
3833 if (lastoffset > endoff)
3834 lastoffset = endoff;
3835 if (imap.br_startblock == HOLESTARTBLOCK)
3836 continue;
3837 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3838 if (imap.br_state == XFS_EXT_UNWRITTEN)
3839 continue;
3840 XFS_BUF_UNDONE(bp);
3841 XFS_BUF_UNWRITE(bp);
3842 XFS_BUF_READ(bp);
3843 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
3844 xfsbdstrat(mp, bp);
3845 if ((error = xfs_iowait(bp))) {
3846 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
3847 mp, bp, XFS_BUF_ADDR(bp));
3848 break;
3849 }
3850 memset(XFS_BUF_PTR(bp) +
3851 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
3852 0, lastoffset - offset + 1);
3853 XFS_BUF_UNDONE(bp);
3854 XFS_BUF_UNREAD(bp);
3855 XFS_BUF_WRITE(bp);
3856 xfsbdstrat(mp, bp);
3857 if ((error = xfs_iowait(bp))) {
3858 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
3859 mp, bp, XFS_BUF_ADDR(bp));
3860 break;
3861 }
3862 }
3863 xfs_buf_free(bp);
3864 return error;
3865}
3866
3867/*
3868 * xfs_free_file_space()
3869 * This routine frees disk space for the given file.
3870 *
3871 * This routine is only called by xfs_change_file_space
3872 * for an UNRESVSP type call.
3873 *
3874 * RETURNS:
3875 * 0 on success
3876 * errno on error
3877 *
3878 */
3879STATIC int
3880xfs_free_file_space(
3881 xfs_inode_t *ip,
3882 xfs_off_t offset,
3883 xfs_off_t len,
3884 int attr_flags)
3885{
Nathan Scott67fcaa72006-06-09 17:00:52 +10003886 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 int committed;
3888 int done;
3889 xfs_off_t end_dmi_offset;
3890 xfs_fileoff_t endoffset_fsb;
3891 int error;
3892 xfs_fsblock_t firstfsb;
3893 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894 xfs_bmbt_irec_t imap;
3895 xfs_off_t ioffset;
3896 xfs_extlen_t mod=0;
3897 xfs_mount_t *mp;
3898 int nimap;
3899 uint resblks;
Nathan Scott673cdf52006-09-28 10:56:26 +10003900 uint rounding;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003901 int rt;
3902 xfs_fileoff_t startoffset_fsb;
3903 xfs_trans_t *tp;
Dean Roehrich5fcbab32005-05-05 13:27:19 -07003904 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10003906 vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907 mp = ip->i_mount;
3908
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11003909 xfs_itrace_entry(ip);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10003910
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
3912 return error;
3913
3914 error = 0;
3915 if (len <= 0) /* if nothing being freed */
3916 return error;
Eric Sandeen71ddabb2007-11-23 16:29:42 +11003917 rt = XFS_IS_REALTIME_INODE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
3919 end_dmi_offset = offset + len;
3920 endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
3921
Christoph Hellwigeb9df392007-08-16 18:42:07 +10003922 if (offset < ip->i_size && (attr_flags & ATTR_DMI) == 0 &&
3923 DM_EVENT_ENABLED(ip, DM_EVENT_WRITE)) {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10003924 if (end_dmi_offset > ip->i_size)
3925 end_dmi_offset = ip->i_size;
Christoph Hellwigbc4ac742008-03-06 13:45:58 +11003926 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003927 offset, end_dmi_offset - offset,
3928 AT_DELAY_FLAG(attr_flags), NULL);
3929 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01003930 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931 }
3932
Dean Roehrich5fcbab32005-05-05 13:27:19 -07003933 if (attr_flags & ATTR_NOLOCK)
3934 need_iolock = 0;
Yingping Lu9fa80462006-03-22 12:44:35 +11003935 if (need_iolock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigb677c212007-08-29 11:46:28 +10003937 vn_iowait(ip); /* wait for the completion of any pending DIOs */
Yingping Lu9fa80462006-03-22 12:44:35 +11003938 }
Dean Roehrich5fcbab32005-05-05 13:27:19 -07003939
Tim Shimmine6a4b372007-11-23 16:30:42 +11003940 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003941 ioffset = offset & ~(rounding - 1);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10003942
3943 if (VN_CACHED(vp) != 0) {
Tim Shimmine6a4b372007-11-23 16:30:42 +11003944 xfs_inval_cached_trace(ip, ioffset, -1, ioffset, -1);
3945 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10003946 if (error)
3947 goto out_unlock_iolock;
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10003948 }
3949
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 /*
3951 * Need to zero the stuff we're not freeing, on disk.
3952 * If its a realtime file & can't use unwritten extents then we
3953 * actually need to zero the extent edges. Otherwise xfs_bunmapi
3954 * will take care of it for us.
3955 */
Eric Sandeen62118702008-03-06 13:44:28 +11003956 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003958 error = xfs_bmapi(NULL, ip, startoffset_fsb,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003959 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003960 if (error)
3961 goto out_unlock_iolock;
3962 ASSERT(nimap == 0 || nimap == 1);
3963 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3964 xfs_daddr_t block;
3965
3966 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3967 block = imap.br_startblock;
3968 mod = do_div(block, mp->m_sb.sb_rextsize);
3969 if (mod)
3970 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
3971 }
3972 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10003973 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10003974 1, 0, NULL, 0, &imap, &nimap, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975 if (error)
3976 goto out_unlock_iolock;
3977 ASSERT(nimap == 0 || nimap == 1);
3978 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
3979 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
3980 mod++;
3981 if (mod && (mod != mp->m_sb.sb_rextsize))
3982 endoffset_fsb -= mod;
3983 }
3984 }
3985 if ((done = (endoffset_fsb <= startoffset_fsb)))
3986 /*
3987 * One contiguous piece to clear
3988 */
3989 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
3990 else {
3991 /*
3992 * Some full blocks, possibly two pieces to clear
3993 */
3994 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
3995 error = xfs_zero_remaining_bytes(ip, offset,
3996 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
3997 if (!error &&
3998 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
3999 error = xfs_zero_remaining_bytes(ip,
4000 XFS_FSB_TO_B(mp, endoffset_fsb),
4001 offset + len - 1);
4002 }
4003
4004 /*
4005 * free file space until done or until there is an error
4006 */
4007 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4008 while (!error && !done) {
4009
4010 /*
David Chinner91ebecc2007-07-19 16:28:30 +10004011 * allocate and setup the transaction. Allow this
4012 * transaction to dip into the reserve blocks to ensure
4013 * the freeing of the space succeeds at ENOSPC.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014 */
4015 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
David Chinner91ebecc2007-07-19 16:28:30 +10004016 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004017 error = xfs_trans_reserve(tp,
4018 resblks,
4019 XFS_WRITE_LOG_RES(mp),
4020 0,
4021 XFS_TRANS_PERM_LOG_RES,
4022 XFS_WRITE_LOG_COUNT);
4023
4024 /*
4025 * check for running out of space
4026 */
4027 if (error) {
4028 /*
4029 * Free the transaction structure.
4030 */
4031 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4032 xfs_trans_cancel(tp, 0);
4033 break;
4034 }
4035 xfs_ilock(ip, XFS_ILOCK_EXCL);
4036 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
Nathan Scottdd9f4382006-01-11 15:28:28 +11004037 ip->i_udquot, ip->i_gdquot, resblks, 0,
4038 XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004039 if (error)
4040 goto error1;
4041
4042 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4043 xfs_trans_ihold(tp, ip);
4044
4045 /*
4046 * issue the bunmapi() call to free the blocks
4047 */
4048 XFS_BMAP_INIT(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10004049 error = xfs_bunmapi(tp, ip, startoffset_fsb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050 endoffset_fsb - startoffset_fsb,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10004051 0, 2, &firstfsb, &free_list, NULL, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004052 if (error) {
4053 goto error0;
4054 }
4055
4056 /*
4057 * complete the transaction
4058 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11004059 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060 if (error) {
4061 goto error0;
4062 }
4063
Eric Sandeen1c72bf92007-05-08 13:48:42 +10004064 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4066 }
4067
4068 out_unlock_iolock:
4069 if (need_iolock)
4070 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4071 return error;
4072
4073 error0:
4074 xfs_bmap_cancel(&free_list);
4075 error1:
4076 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4077 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4078 XFS_ILOCK_EXCL);
4079 return error;
4080}
4081
4082/*
4083 * xfs_change_file_space()
4084 * This routine allocates or frees disk space for the given file.
4085 * The user specified parameters are checked for alignment and size
4086 * limitations.
4087 *
4088 * RETURNS:
4089 * 0 on success
4090 * errno on error
4091 *
4092 */
4093int
4094xfs_change_file_space(
Christoph Hellwig993386c2007-08-28 16:12:30 +10004095 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096 int cmd,
4097 xfs_flock64_t *bf,
4098 xfs_off_t offset,
4099 cred_t *credp,
4100 int attr_flags)
4101{
Christoph Hellwig993386c2007-08-28 16:12:30 +10004102 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004103 int clrprealloc;
4104 int error;
4105 xfs_fsize_t fsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004106 int setprealloc;
4107 xfs_off_t startoffset;
4108 xfs_off_t llen;
4109 xfs_trans_t *tp;
Nathan Scott8285fb52006-06-09 17:07:12 +10004110 bhv_vattr_t va;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004111
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11004112 xfs_itrace_entry(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113
Christoph Hellwig993386c2007-08-28 16:12:30 +10004114 if (!S_ISREG(ip->i_d.di_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004115 return XFS_ERROR(EINVAL);
4116
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117 switch (bf->l_whence) {
4118 case 0: /*SEEK_SET*/
4119 break;
4120 case 1: /*SEEK_CUR*/
4121 bf->l_start += offset;
4122 break;
4123 case 2: /*SEEK_END*/
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10004124 bf->l_start += ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125 break;
4126 default:
4127 return XFS_ERROR(EINVAL);
4128 }
4129
4130 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4131
4132 if ( (bf->l_start < 0)
4133 || (bf->l_start > XFS_MAXIOFFSET(mp))
4134 || (bf->l_start + llen < 0)
4135 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4136 return XFS_ERROR(EINVAL);
4137
4138 bf->l_whence = 0;
4139
4140 startoffset = bf->l_start;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10004141 fsize = ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142
4143 /*
4144 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4145 * file space.
4146 * These calls do NOT zero the data space allocated to the file,
4147 * nor do they change the file size.
4148 *
4149 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4150 * space.
4151 * These calls cause the new file data to be zeroed and the file
4152 * size to be changed.
4153 */
4154 setprealloc = clrprealloc = 0;
4155
4156 switch (cmd) {
4157 case XFS_IOC_RESVSP:
4158 case XFS_IOC_RESVSP64:
4159 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4160 1, attr_flags);
4161 if (error)
4162 return error;
4163 setprealloc = 1;
4164 break;
4165
4166 case XFS_IOC_UNRESVSP:
4167 case XFS_IOC_UNRESVSP64:
4168 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4169 attr_flags)))
4170 return error;
4171 break;
4172
4173 case XFS_IOC_ALLOCSP:
4174 case XFS_IOC_ALLOCSP64:
4175 case XFS_IOC_FREESP:
4176 case XFS_IOC_FREESP64:
4177 if (startoffset > fsize) {
4178 error = xfs_alloc_file_space(ip, fsize,
4179 startoffset - fsize, 0, attr_flags);
4180 if (error)
4181 break;
4182 }
4183
4184 va.va_mask = XFS_AT_SIZE;
4185 va.va_size = startoffset;
4186
Christoph Hellwig993386c2007-08-28 16:12:30 +10004187 error = xfs_setattr(ip, &va, attr_flags, credp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188
4189 if (error)
4190 return error;
4191
4192 clrprealloc = 1;
4193 break;
4194
4195 default:
4196 ASSERT(0);
4197 return XFS_ERROR(EINVAL);
4198 }
4199
4200 /*
4201 * update the inode timestamp, mode, and prealloc flag bits
4202 */
4203 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4204
4205 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4206 0, 0, 0))) {
4207 /* ASSERT(0); */
4208 xfs_trans_cancel(tp, 0);
4209 return error;
4210 }
4211
4212 xfs_ilock(ip, XFS_ILOCK_EXCL);
4213
4214 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4215 xfs_trans_ihold(tp, ip);
4216
4217 if ((attr_flags & ATTR_DMI) == 0) {
4218 ip->i_d.di_mode &= ~S_ISUID;
4219
4220 /*
4221 * Note that we don't have to worry about mandatory
4222 * file locking being disabled here because we only
4223 * clear the S_ISGID bit if the Group execute bit is
4224 * on, but if it was on then mandatory locking wouldn't
4225 * have been enabled.
4226 */
4227 if (ip->i_d.di_mode & S_IXGRP)
4228 ip->i_d.di_mode &= ~S_ISGID;
4229
4230 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4231 }
4232 if (setprealloc)
4233 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4234 else if (clrprealloc)
4235 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4236
4237 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4238 xfs_trans_set_sync(tp);
4239
Eric Sandeen1c72bf92007-05-08 13:48:42 +10004240 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241
4242 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4243
4244 return error;
4245}