blob: 619720705bc6843e4624ce02c27829fa32256ba9 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap_btree.h"
32#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_dinode.h"
34#include "xfs_inode.h"
35#include "xfs_inode_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_itable.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_ialloc.h"
38#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_bmap.h"
Christoph Hellwigef14f0c2009-06-10 17:07:47 +020040#include "xfs_acl.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_attr.h"
42#include "xfs_rw.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_quota.h"
45#include "xfs_utils.h"
Nathan Scotta844f452005-11-02 14:38:42 +110046#include "xfs_rtalloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include "xfs_log_priv.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100049#include "xfs_filestream.h"
Christoph Hellwig993386c2007-08-28 16:12:30 +100050#include "xfs_vnodeops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000051#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Christoph Hellwig993386c2007-08-28 16:12:30 +100053int
Linus Torvalds1da177e2005-04-16 15:20:36 -070054xfs_setattr(
Christoph Hellwig0f285c82008-07-18 17:13:28 +100055 struct xfs_inode *ip,
56 struct iattr *iattr,
Christoph Hellwigea5a3dc82008-10-30 18:27:48 +110057 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058{
Christoph Hellwig993386c2007-08-28 16:12:30 +100059 xfs_mount_t *mp = ip->i_mount;
David Chinnere4f75292008-08-13 16:00:45 +100060 struct inode *inode = VFS_I(ip);
Christoph Hellwig0f285c82008-07-18 17:13:28 +100061 int mask = iattr->ia_valid;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 int code;
64 uint lock_flags;
65 uint commit_flags=0;
66 uid_t uid=0, iuid=0;
67 gid_t gid=0, igid=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 struct xfs_dquot *udqp, *gdqp, *olddquot1, *olddquot2;
Dean Roehrich5fcbab32005-05-05 13:27:19 -070069 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Christoph Hellwigcca28fb2010-06-24 11:57:09 +100071 trace_xfs_setattr(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Christoph Hellwigbd186aa2007-08-30 17:21:12 +100073 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 return XFS_ERROR(EROFS);
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 if (XFS_FORCED_SHUTDOWN(mp))
77 return XFS_ERROR(EIO);
78
Christoph Hellwigc4cd7472008-12-09 04:47:34 -050079 code = -inode_change_ok(inode, iattr);
80 if (code)
81 return code;
82
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 olddquot1 = olddquot2 = NULL;
84 udqp = gdqp = NULL;
85
86 /*
87 * If disk quotas is on, we make sure that the dquots do exist on disk,
88 * before we start any other transactions. Trying to do this later
89 * is messy. We don't care to take a readlock to look at the ids
90 * in inode here, because we can't hold it across the trans_reserve.
91 * If the IDs do change before we take the ilock, we're covered
92 * because the i_*dquot fields will get updated anyway.
93 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +100094 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 uint qflags = 0;
96
Christoph Hellwig0f285c82008-07-18 17:13:28 +100097 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
98 uid = iattr->ia_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 qflags |= XFS_QMOPT_UQUOTA;
100 } else {
101 uid = ip->i_d.di_uid;
102 }
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000103 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
104 gid = iattr->ia_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 qflags |= XFS_QMOPT_GQUOTA;
106 } else {
107 gid = ip->i_d.di_gid;
108 }
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 /*
111 * We take a reference when we initialize udqp and gdqp,
112 * so it is important that we never blindly double trip on
113 * the same variable. See xfs_create() for an example.
114 */
115 ASSERT(udqp == NULL);
116 ASSERT(gdqp == NULL);
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000117 code = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000118 qflags, &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 if (code)
Jesper Juhl014c2542006-01-15 02:37:08 +0100120 return code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 }
122
123 /*
124 * For the other attributes, we acquire the inode lock and
125 * first do an error checking pass.
126 */
127 tp = NULL;
128 lock_flags = XFS_ILOCK_EXCL;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000129 if (flags & XFS_ATTR_NOLOCK)
Dean Roehrich5fcbab32005-05-05 13:27:19 -0700130 need_iolock = 0;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000131 if (!(mask & ATTR_SIZE)) {
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000132 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
133 commit_flags = 0;
134 code = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp),
135 0, 0, 0);
136 if (code) {
137 lock_flags = 0;
138 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 }
140 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 if (need_iolock)
142 lock_flags |= XFS_IOLOCK_EXCL;
143 }
144
145 xfs_ilock(ip, lock_flags);
146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 /*
148 * Change file ownership. Must be the owner or privileged.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000150 if (mask & (ATTR_UID|ATTR_GID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 /*
152 * These IDs could have changed since we last looked at them.
153 * But, we're assured that if the ownership did change
154 * while we didn't have the inode locked, inode's dquot(s)
155 * would have changed also.
156 */
157 iuid = ip->i_d.di_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 igid = ip->i_d.di_gid;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000159 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
160 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 /*
Christoph Hellwig25fe55e2008-07-18 17:13:20 +1000163 * Do a quota reservation only if uid/gid is actually
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 * going to change.
165 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200166 if (XFS_IS_QUOTA_RUNNING(mp) &&
167 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
168 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 ASSERT(tp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200170 code = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 capable(CAP_FOWNER) ?
172 XFS_QMOPT_FORCE_RES : 0);
173 if (code) /* out of quota */
174 goto error_return;
175 }
176 }
177
178 /*
179 * Truncate file. Must have write permission and not be a directory.
180 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000181 if (mask & ATTR_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 /* Short circuit the truncate case for zero length files */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000183 if (iattr->ia_size == 0 &&
184 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 xfs_iunlock(ip, XFS_ILOCK_EXCL);
186 lock_flags &= ~XFS_ILOCK_EXCL;
Dave Chinnerdcd79a12010-09-28 12:27:25 +1000187 if (mask & ATTR_CTIME) {
188 inode->i_mtime = inode->i_ctime =
189 current_fs_time(inode->i_sb);
190 xfs_mark_inode_dirty_sync(ip);
191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 code = 0;
193 goto error_return;
194 }
195
Christoph Hellwig42173f62008-04-22 17:33:25 +1000196 if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 code = XFS_ERROR(EISDIR);
198 goto error_return;
Christoph Hellwig42173f62008-04-22 17:33:25 +1000199 } else if (!S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 code = XFS_ERROR(EINVAL);
201 goto error_return;
202 }
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 /*
205 * Make sure that the dquots are attached to the inode.
206 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200207 code = xfs_qm_dqattach_locked(ip, 0);
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500208 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500211 /*
212 * Now we can make the changes. Before we join the inode
213 * to the transaction, if ATTR_SIZE is set then take care of
214 * the part of the truncation that must be done without the
215 * inode lock. This needs to be done before joining the inode
216 * to the transaction, because the inode cannot be unlocked
217 * once it is a part of the transaction.
218 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000219 if (iattr->ia_size > ip->i_size) {
Christoph Hellwig61436fe2008-05-20 11:30:46 +1000220 /*
221 * Do the first part of growing a file: zero any data
222 * in the last block that is beyond the old EOF. We
223 * need to do this before the inode is joined to the
224 * transaction to modify the i_size.
225 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000226 code = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400227 if (code)
228 goto error_return;
Eric Sandeen374e2ac2005-11-02 15:07:34 +1100229 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400231 lock_flags &= ~XFS_ILOCK_EXCL;
David Chinnerc32676e2007-07-19 16:28:58 +1000232
233 /*
234 * We are going to log the inode size change in this
235 * transaction so any previous writes that are beyond the on
236 * disk EOF and the new EOF that have not been written out need
237 * to be written here. If we do not write the data out, we
238 * expose ourselves to the null files problem.
239 *
240 * Only flush from the on disk size to the smaller of the in
241 * memory file size or the new size as that's the range we
242 * really care about here and prevents waiting for other data
243 * not within the range we care about here.
244 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400245 if (ip->i_size != ip->i_d.di_size &&
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000246 iattr->ia_size > ip->i_d.di_size) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000247 code = xfs_flush_pages(ip,
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000248 ip->i_d.di_size, iattr->ia_size,
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000249 XBF_ASYNC, FI_NONE);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400250 if (code)
251 goto error_return;
David Chinnerc32676e2007-07-19 16:28:58 +1000252 }
253
254 /* wait for all I/O to complete */
Christoph Hellwig25e41b32008-12-03 12:20:39 +0100255 xfs_ioend_wait(ip);
David Chinnerc32676e2007-07-19 16:28:58 +1000256
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400257 code = -block_truncate_page(inode->i_mapping, iattr->ia_size,
258 xfs_get_blocks);
259 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 goto error_return;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400263 code = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
264 XFS_TRANS_PERM_LOG_RES,
265 XFS_ITRUNCATE_LOG_COUNT);
266 if (code)
267 goto error_return;
268
269 truncate_setsize(inode, iattr->ia_size);
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -0400272 lock_flags |= XFS_ILOCK_EXCL;
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 xfs_ilock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Christoph Hellwig898621d2010-06-24 11:36:58 +1000276 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
David Chinner44d814c2008-03-06 13:45:29 +1100278 /*
279 * Only change the c/mtime if we are changing the size
280 * or we are explicitly asked to change it. This handles
281 * the semantic difference between truncate() and ftruncate()
282 * as implemented in the VFS.
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000283 *
284 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME
285 * is a special case where we need to update the times despite
286 * not having these flags set. For all other operations the
287 * VFS set these flags explicitly if it wants a timestamp
288 * update.
David Chinner44d814c2008-03-06 13:45:29 +1100289 */
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000290 if (iattr->ia_size != ip->i_size &&
291 (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
292 iattr->ia_ctime = iattr->ia_mtime =
293 current_fs_time(inode->i_sb);
294 mask |= ATTR_CTIME | ATTR_MTIME;
295 }
David Chinner44d814c2008-03-06 13:45:29 +1100296
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000297 if (iattr->ia_size > ip->i_size) {
298 ip->i_d.di_size = iattr->ia_size;
299 ip->i_size = iattr->ia_size;
Christoph Hellwig61436fe2008-05-20 11:30:46 +1000300 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000301 } else if (iattr->ia_size <= ip->i_size ||
302 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 /*
304 * signal a sync transaction unless
305 * we're truncating an already unlinked
306 * file on a wsync filesystem
307 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000308 code = xfs_itruncate_finish(&tp, ip, iattr->ia_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 XFS_DATA_FORK,
310 ((ip->i_d.di_nlink != 0 ||
311 !(mp->m_flags & XFS_MOUNT_WSYNC))
312 ? 1 : 0));
Nathan Scott7d4fb402006-06-09 15:27:16 +1000313 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 goto abort_return;
Nathan Scott7d4fb402006-06-09 15:27:16 +1000315 /*
316 * Truncated "down", so we're removing references
317 * to old data here - if we now delay flushing for
318 * a long time, we expose ourselves unduly to the
319 * notorious NULL files problem. So, we mark this
320 * vnode and flush it when the file is closed, and
321 * do not wait the usual (long) time for writeout.
322 */
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000323 xfs_iflags_set(ip, XFS_ITRUNCATED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 }
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500325 } else if (tp) {
Christoph Hellwig898621d2010-06-24 11:36:58 +1000326 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 }
328
329 /*
330 * Change file ownership. Must be the owner or privileged.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 */
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000332 if (mask & (ATTR_UID|ATTR_GID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 /*
334 * CAP_FSETID overrides the following restrictions:
335 *
336 * The set-user-ID and set-group-ID bits of a file will be
337 * cleared upon successful return from chown()
338 */
339 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
340 !capable(CAP_FSETID)) {
341 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
342 }
343
344 /*
345 * Change the ownerships and register quota modifications
346 * in the transaction.
347 */
348 if (iuid != uid) {
Christoph Hellwig7d095252009-06-08 15:33:32 +0200349 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000350 ASSERT(mask & ATTR_UID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 ASSERT(udqp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200352 olddquot1 = xfs_qm_vop_chown(tp, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 &ip->i_udquot, udqp);
354 }
355 ip->i_d.di_uid = uid;
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000356 inode->i_uid = uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 }
358 if (igid != gid) {
Christoph Hellwig7d095252009-06-08 15:33:32 +0200359 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000360 ASSERT(!XFS_IS_PQUOTA_ON(mp));
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000361 ASSERT(mask & ATTR_GID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 ASSERT(gdqp);
Christoph Hellwig7d095252009-06-08 15:33:32 +0200363 olddquot2 = xfs_qm_vop_chown(tp, ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 &ip->i_gdquot, gdqp);
365 }
366 ip->i_d.di_gid = gid;
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000367 inode->i_gid = gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
370
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500371 /*
372 * Change file access modes.
373 */
374 if (mask & ATTR_MODE) {
375 umode_t mode = iattr->ia_mode;
376
377 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
378 mode &= ~S_ISGID;
379
380 ip->i_d.di_mode &= S_IFMT;
381 ip->i_d.di_mode |= mode & ~S_IFMT;
382
383 inode->i_mode &= S_IFMT;
384 inode->i_mode |= mode & ~S_IFMT;
Christoph Hellwigc4cd7472008-12-09 04:47:34 -0500385 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
387 /*
388 * Change file access or modified times.
389 */
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000390 if (mask & ATTR_ATIME) {
391 inode->i_atime = iattr->ia_atime;
392 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
393 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
394 ip->i_update_core = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000396 if (mask & ATTR_CTIME) {
Christoph Hellwigf13fae22008-07-21 16:16:15 +1000397 inode->i_ctime = iattr->ia_ctime;
Christoph Hellwig0f285c82008-07-18 17:13:28 +1000398 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
399 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 ip->i_update_core = 1;
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000401 }
402 if (mask & ATTR_MTIME) {
403 inode->i_mtime = iattr->ia_mtime;
404 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
405 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
406 ip->i_update_core = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 }
408
409 /*
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000410 * And finally, log the inode core if any attribute in it
411 * has been changed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 */
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000413 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE|
414 ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
415 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
417 XFS_STATS_INC(xs_ig_attrchg);
418
419 /*
420 * If this is a synchronous mount, make sure that the
421 * transaction goes to disk before returning to the user.
422 * This is slightly sub-optimal in that truncates require
Nathan Scottc41564b2006-03-29 08:55:14 +1000423 * two sync transactions instead of one for wsync filesystems.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 * One for the truncate and one for the timestamps since we
425 * don't want to change the timestamps unless we're sure the
426 * truncate worked. Truncates are less than 1% of the laddis
427 * mix so this probably isn't worth the trouble to optimize.
428 */
429 code = 0;
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000430 if (mp->m_flags & XFS_MOUNT_WSYNC)
431 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Christoph Hellwigd6d59ba2009-12-23 16:09:13 +0000433 code = xfs_trans_commit(tp, commit_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 xfs_iunlock(ip, lock_flags);
436
437 /*
438 * Release any dquot(s) the inode had kept before chown.
439 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200440 xfs_qm_dqrele(olddquot1);
441 xfs_qm_dqrele(olddquot2);
442 xfs_qm_dqrele(udqp);
443 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Christoph Hellwigef14f0c2009-06-10 17:07:47 +0200445 if (code)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 return code;
Christoph Hellwigef14f0c2009-06-10 17:07:47 +0200447
448 /*
449 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
450 * update. We could avoid this with linked transactions
451 * and passing down the transaction pointer all the way
452 * to attr_set. No previous user of the generic
453 * Posix ACL code seems to care about this issue either.
454 */
455 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
456 code = -xfs_acl_chmod(inode);
457 if (code)
458 return XFS_ERROR(code);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 return 0;
462
463 abort_return:
464 commit_flags |= XFS_TRANS_ABORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 error_return:
Christoph Hellwig7d095252009-06-08 15:33:32 +0200466 xfs_qm_dqrele(udqp);
467 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (tp) {
469 xfs_trans_cancel(tp, commit_flags);
470 }
471 if (lock_flags != 0) {
472 xfs_iunlock(ip, lock_flags);
473 }
474 return code;
475}
476
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477/*
Nathan Scott7d4fb402006-06-09 15:27:16 +1000478 * The maximum pathlen is 1024 bytes. Since the minimum file system
479 * blocksize is 512 bytes, we can get a max of 2 extents back from
480 * bmapi.
481 */
482#define SYMLINK_MAPS 2
483
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000484STATIC int
485xfs_readlink_bmap(
486 xfs_inode_t *ip,
487 char *link)
488{
489 xfs_mount_t *mp = ip->i_mount;
490 int pathlen = ip->i_d.di_size;
491 int nmaps = SYMLINK_MAPS;
492 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
493 xfs_daddr_t d;
494 int byte_cnt;
495 int n;
496 xfs_buf_t *bp;
497 int error = 0;
498
499 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen), 0, NULL, 0,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000500 mval, &nmaps, NULL);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000501 if (error)
502 goto out;
503
504 for (n = 0; n < nmaps; n++) {
505 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
506 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
507
Christoph Hellwig6ad112b2009-11-24 18:02:23 +0000508 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt),
509 XBF_LOCK | XBF_MAPPED | XBF_DONT_BLOCK);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000510 error = XFS_BUF_GETERROR(bp);
511 if (error) {
512 xfs_ioerror_alert("xfs_readlink",
513 ip->i_mount, bp, XFS_BUF_ADDR(bp));
514 xfs_buf_relse(bp);
515 goto out;
516 }
517 if (pathlen < byte_cnt)
518 byte_cnt = pathlen;
519 pathlen -= byte_cnt;
520
521 memcpy(link, XFS_BUF_PTR(bp), byte_cnt);
522 xfs_buf_relse(bp);
523 }
524
525 link[ip->i_d.di_size] = '\0';
526 error = 0;
527
528 out:
529 return error;
530}
531
Christoph Hellwig993386c2007-08-28 16:12:30 +1000532int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533xfs_readlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000534 xfs_inode_t *ip,
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000535 char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000537 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 int pathlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000541 trace_xfs_readlink(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
543 if (XFS_FORCED_SHUTDOWN(mp))
544 return XFS_ERROR(EIO);
545
546 xfs_ilock(ip, XFS_ILOCK_SHARED);
547
548 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000549 ASSERT(ip->i_d.di_size <= MAXPATHLEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000551 pathlen = ip->i_d.di_size;
552 if (!pathlen)
553 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 if (ip->i_df.if_flags & XFS_IFINLINE) {
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000556 memcpy(link, ip->i_df.if_u1.if_data, pathlen);
557 link[pathlen] = '\0';
558 } else {
559 error = xfs_readlink_bmap(ip, link);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 }
561
Christoph Hellwig804c83c2007-08-28 13:59:03 +1000562 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return error;
565}
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567/*
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000568 * Flags for xfs_free_eofblocks
569 */
570#define XFS_FREE_EOF_TRYLOCK (1<<0)
571
572/*
David Chinner92dfe8d2007-05-24 15:22:19 +1000573 * This is called by xfs_inactive to free any blocks beyond eof
574 * when the link count isn't zero and by xfs_dm_punch_hole() when
575 * punching a hole to EOF.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 */
Eric Sandeend96f8f82009-07-02 00:09:33 -0500577STATIC int
David Chinner92dfe8d2007-05-24 15:22:19 +1000578xfs_free_eofblocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 xfs_mount_t *mp,
David Chinner92dfe8d2007-05-24 15:22:19 +1000580 xfs_inode_t *ip,
581 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
583 xfs_trans_t *tp;
584 int error;
585 xfs_fileoff_t end_fsb;
586 xfs_fileoff_t last_fsb;
587 xfs_filblks_t map_len;
588 int nimaps;
589 xfs_bmbt_irec_t imap;
590
591 /*
592 * Figure out if there are any blocks beyond the end
593 * of the file. If not, then there is nothing to do.
594 */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000595 end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
Kulikov Vasiliy3f348852010-07-20 17:54:28 +1000597 if (last_fsb <= end_fsb)
Jesper Juhl014c2542006-01-15 02:37:08 +0100598 return 0;
Kulikov Vasiliy3f348852010-07-20 17:54:28 +1000599 map_len = last_fsb - end_fsb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601 nimaps = 1;
602 xfs_ilock(ip, XFS_ILOCK_SHARED);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +1000603 error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000604 NULL, 0, &imap, &nimaps, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 xfs_iunlock(ip, XFS_ILOCK_SHARED);
606
607 if (!error && (nimaps != 0) &&
Yingping Lu68bdb6e2006-01-11 15:38:31 +1100608 (imap.br_startblock != HOLESTARTBLOCK ||
609 ip->i_delayed_blks)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 /*
611 * Attach the dquots to the inode up front.
612 */
Christoph Hellwig7d095252009-06-08 15:33:32 +0200613 error = xfs_qm_dqattach(ip, 0);
614 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +0100615 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617 /*
618 * There are blocks after the end of file.
619 * Free them up now by truncating the file to
620 * its current size.
621 */
622 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
623
624 /*
625 * Do the xfs_itruncate_start() call before
626 * reserving any log space because
627 * itruncate_start will call into the buffer
628 * cache and we can't
629 * do that within a transaction.
630 */
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000631 if (flags & XFS_FREE_EOF_TRYLOCK) {
632 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
633 xfs_trans_cancel(tp, 0);
634 return 0;
635 }
636 } else {
David Chinner92dfe8d2007-05-24 15:22:19 +1000637 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000638 }
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000639 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000640 ip->i_size);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000641 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +1000642 xfs_trans_cancel(tp, 0);
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000643 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +1000644 return error;
645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647 error = xfs_trans_reserve(tp, 0,
648 XFS_ITRUNCATE_LOG_RES(mp),
649 0, XFS_TRANS_PERM_LOG_RES,
650 XFS_ITRUNCATE_LOG_COUNT);
651 if (error) {
652 ASSERT(XFS_FORCED_SHUTDOWN(mp));
653 xfs_trans_cancel(tp, 0);
654 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +0100655 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 }
657
658 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +1000659 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
661 error = xfs_itruncate_finish(&tp, ip,
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000662 ip->i_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 XFS_DATA_FORK,
664 0);
665 /*
666 * If we get an error at this point we
667 * simply don't bother truncating the file.
668 */
669 if (error) {
670 xfs_trans_cancel(tp,
671 (XFS_TRANS_RELEASE_LOG_RES |
672 XFS_TRANS_ABORT));
673 } else {
674 error = xfs_trans_commit(tp,
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000675 XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000677 xfs_iunlock(ip, XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
Jesper Juhl014c2542006-01-15 02:37:08 +0100679 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680}
681
682/*
683 * Free a symlink that has blocks associated with it.
684 */
685STATIC int
686xfs_inactive_symlink_rmt(
687 xfs_inode_t *ip,
688 xfs_trans_t **tpp)
689{
690 xfs_buf_t *bp;
691 int committed;
692 int done;
693 int error;
694 xfs_fsblock_t first_block;
695 xfs_bmap_free_t free_list;
696 int i;
697 xfs_mount_t *mp;
698 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
699 int nmaps;
700 xfs_trans_t *ntp;
701 int size;
702 xfs_trans_t *tp;
703
704 tp = *tpp;
705 mp = ip->i_mount;
706 ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
707 /*
708 * We're freeing a symlink that has some
709 * blocks allocated to it. Free the
710 * blocks here. We know that we've got
711 * either 1 or 2 extents and that we can
712 * free them all in one bunmapi call.
713 */
714 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
715 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
716 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
717 ASSERT(XFS_FORCED_SHUTDOWN(mp));
718 xfs_trans_cancel(tp, 0);
719 *tpp = NULL;
720 return error;
721 }
722 /*
723 * Lock the inode, fix the size, and join it to the transaction.
724 * Hold it so in the normal path, we still have it locked for
725 * the second transaction. In the error paths we need it
726 * held so the cancel won't rele it, see below.
727 */
728 xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
729 size = (int)ip->i_d.di_size;
730 ip->i_d.di_size = 0;
Christoph Hellwig898621d2010-06-24 11:36:58 +1000731 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
733 /*
734 * Find the block(s) so we can inval and unmap them.
735 */
736 done = 0;
Eric Sandeen9d87c312009-01-14 23:22:07 -0600737 xfs_bmap_init(&free_list, &first_block);
Tobias Klausere8c96f82006-03-24 03:15:34 -0800738 nmaps = ARRAY_SIZE(mval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
740 XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000741 &free_list)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 goto error0;
743 /*
744 * Invalidate the block(s).
745 */
746 for (i = 0; i < nmaps; i++) {
747 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
748 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
749 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
750 xfs_trans_binval(tp, bp);
751 }
752 /*
753 * Unmap the dead block(s) to the free_list.
754 */
755 if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +1000756 &first_block, &free_list, &done)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 goto error1;
758 ASSERT(done);
759 /*
760 * Commit the first transaction. This logs the EFI and the inode.
761 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +1100762 if ((error = xfs_bmap_finish(&tp, &free_list, &committed)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 goto error1;
764 /*
765 * The transaction must have been committed, since there were
766 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
767 * The new tp has the extent freeing and EFDs.
768 */
769 ASSERT(committed);
770 /*
771 * The first xact was committed, so add the inode to the new one.
772 * Mark it dirty so it will be logged and moved forward in the log as
773 * part of every commit.
774 */
Christoph Hellwig898621d2010-06-24 11:36:58 +1000775 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
777 /*
778 * Get a new, empty transaction to return to our caller.
779 */
780 ntp = xfs_trans_dup(tp);
781 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000782 * Commit the transaction containing extent freeing and EFDs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 * If we get an error on the commit here or on the reserve below,
784 * we need to unlock the inode since the new transaction doesn't
785 * have the inode attached.
786 */
Eric Sandeen1c72bf92007-05-08 13:48:42 +1000787 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 tp = ntp;
789 if (error) {
790 ASSERT(XFS_FORCED_SHUTDOWN(mp));
791 goto error0;
792 }
793 /*
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100794 * transaction commit worked ok so we can drop the extra ticket
795 * reference that we gained in xfs_trans_dup()
796 */
797 xfs_log_ticket_put(tp->t_ticket);
798
799 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * Remove the memory for extent descriptions (just bookkeeping).
801 */
802 if (ip->i_df.if_bytes)
803 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
804 ASSERT(ip->i_df.if_bytes == 0);
805 /*
806 * Put an itruncate log reservation in the new transaction
807 * for our caller.
808 */
809 if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
810 XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
811 ASSERT(XFS_FORCED_SHUTDOWN(mp));
812 goto error0;
813 }
814 /*
815 * Return with the inode locked but not joined to the transaction.
816 */
817 *tpp = tp;
818 return 0;
819
820 error1:
821 xfs_bmap_cancel(&free_list);
822 error0:
823 /*
824 * Have to come here with the inode locked and either
825 * (held and in the transaction) or (not in the transaction).
826 * If the inode isn't held then cancel would iput it, but
827 * that's wrong since this is inactive and the vnode ref
828 * count is 0 already.
829 * Cancel won't do anything to the inode if held, but it still
830 * needs to be locked until the cancel is done, if it was
831 * joined to the transaction.
832 */
833 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
834 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
835 *tpp = NULL;
836 return error;
837
838}
839
840STATIC int
841xfs_inactive_symlink_local(
842 xfs_inode_t *ip,
843 xfs_trans_t **tpp)
844{
845 int error;
846
847 ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
848 /*
849 * We're freeing a symlink which fit into
850 * the inode. Just free the memory used
851 * to hold the old symlink.
852 */
853 error = xfs_trans_reserve(*tpp, 0,
854 XFS_ITRUNCATE_LOG_RES(ip->i_mount),
855 0, XFS_TRANS_PERM_LOG_RES,
856 XFS_ITRUNCATE_LOG_COUNT);
857
858 if (error) {
859 xfs_trans_cancel(*tpp, 0);
860 *tpp = NULL;
Jesper Juhl014c2542006-01-15 02:37:08 +0100861 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 }
863 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
864
865 /*
866 * Zero length symlinks _can_ exist.
867 */
868 if (ip->i_df.if_bytes > 0) {
869 xfs_idata_realloc(ip,
870 -(ip->i_df.if_bytes),
871 XFS_DATA_FORK);
872 ASSERT(ip->i_df.if_bytes == 0);
873 }
Jesper Juhl014c2542006-01-15 02:37:08 +0100874 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877STATIC int
878xfs_inactive_attrs(
879 xfs_inode_t *ip,
880 xfs_trans_t **tpp)
881{
882 xfs_trans_t *tp;
883 int error;
884 xfs_mount_t *mp;
885
Christoph Hellwig579aa9c2008-04-22 17:34:00 +1000886 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 tp = *tpp;
888 mp = ip->i_mount;
889 ASSERT(ip->i_d.di_forkoff != 0);
David Chinnere5720ee2008-04-10 12:21:18 +1000890 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 xfs_iunlock(ip, XFS_ILOCK_EXCL);
David Chinnere5720ee2008-04-10 12:21:18 +1000892 if (error)
893 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 error = xfs_attr_inactive(ip);
David Chinnere5720ee2008-04-10 12:21:18 +1000896 if (error)
897 goto error_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
900 error = xfs_trans_reserve(tp, 0,
901 XFS_IFREE_LOG_RES(mp),
902 0, XFS_TRANS_PERM_LOG_RES,
903 XFS_INACTIVE_LOG_COUNT);
David Chinnere5720ee2008-04-10 12:21:18 +1000904 if (error)
905 goto error_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
907 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +1000908 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
910
911 ASSERT(ip->i_d.di_anextents == 0);
912
913 *tpp = tp;
Jesper Juhl014c2542006-01-15 02:37:08 +0100914 return 0;
David Chinnere5720ee2008-04-10 12:21:18 +1000915
916error_cancel:
917 ASSERT(XFS_FORCED_SHUTDOWN(mp));
918 xfs_trans_cancel(tp, 0);
919error_unlock:
920 *tpp = NULL;
921 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
922 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
924
Christoph Hellwig993386c2007-08-28 16:12:30 +1000925int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926xfs_release(
Christoph Hellwig993386c2007-08-28 16:12:30 +1000927 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
Christoph Hellwig993386c2007-08-28 16:12:30 +1000929 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 int error;
931
Christoph Hellwig42173f62008-04-22 17:33:25 +1000932 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000936 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 return 0;
938
David Chinner2a82b8b2007-07-11 11:09:12 +1000939 if (!XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +1000940 int truncated;
941
David Chinner2a82b8b2007-07-11 11:09:12 +1000942 /*
943 * If we are using filestreams, and we have an unlinked
944 * file that we are processing the last close on, then nothing
945 * will be able to reopen and write to this file. Purge this
946 * inode from the filestreams cache so that it doesn't delay
947 * teardown of the inode.
948 */
949 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
950 xfs_filestream_deassociate(ip);
951
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000952 /*
953 * If we previously truncated this file and removed old data
954 * in the process, we want to initiate "early" writeout on
955 * the last close. This is an attempt to combat the notorious
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300956 * NULL files problem which is particularly noticeable from a
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000957 * truncate down, buffered (re-)write (delalloc), followed by
958 * a crash. What we are effectively doing here is
959 * significantly reducing the time window where we'd otherwise
960 * be exposed to that problem.
961 */
Christoph Hellwig09262b42007-08-29 11:44:50 +1000962 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
Dave Chinnerdf4368a2011-06-23 01:35:00 +0000963 if (truncated) {
964 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
965 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
966 xfs_flush_pages(ip, 0, -1, XBF_ASYNC, FI_NONE);
967 }
Christoph Hellwigfbf3ce82007-06-28 16:46:47 +1000968 }
969
Dave Chinner6e857562010-12-23 12:02:31 +1100970 if (ip->i_d.di_nlink == 0)
971 return 0;
Christoph Hellwigc56c9632009-10-19 04:03:46 +0000972
Dave Chinner6e857562010-12-23 12:02:31 +1100973 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
974 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
975 ip->i_delayed_blks > 0)) &&
976 (ip->i_df.if_flags & XFS_IFEXTENTS)) &&
977 (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
978
979 /*
980 * If we can't get the iolock just skip truncating the blocks
981 * past EOF because we could deadlock with the mmap_sem
982 * otherwise. We'll get another chance to drop them once the
983 * last reference to the inode is dropped, so we'll never leak
984 * blocks permanently.
985 *
986 * Further, check if the inode is being opened, written and
987 * closed frequently and we have delayed allocation blocks
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300988 * outstanding (e.g. streaming writes from the NFS server),
Dave Chinner6e857562010-12-23 12:02:31 +1100989 * truncating the blocks past EOF will cause fragmentation to
990 * occur.
991 *
992 * In this case don't do the truncation, either, but we have to
993 * be careful how we detect this case. Blocks beyond EOF show
994 * up as i_delayed_blks even when the inode is clean, so we
995 * need to truncate them away first before checking for a dirty
996 * release. Hence on the first dirty close we will still remove
997 * the speculative allocation, but after that we will leave it
998 * in place.
999 */
1000 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1001 return 0;
1002
1003 error = xfs_free_eofblocks(mp, ip,
1004 XFS_FREE_EOF_TRYLOCK);
1005 if (error)
1006 return error;
1007
1008 /* delalloc blocks after truncation means it really is dirty */
1009 if (ip->i_delayed_blks)
1010 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 return 0;
1013}
1014
1015/*
1016 * xfs_inactive
1017 *
1018 * This is called when the vnode reference count for the vnode
1019 * goes to zero. If the file has been unlinked, then it must
1020 * now be truncated. Also, we clear all of the read-ahead state
1021 * kept for the inode here since the file is now closed.
1022 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001023int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024xfs_inactive(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001025 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
Nathan Scott67fcaa72006-06-09 17:00:52 +10001027 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 xfs_fsblock_t first_block;
1029 int committed;
1030 xfs_trans_t *tp;
1031 xfs_mount_t *mp;
1032 int error;
1033 int truncate;
1034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 /*
1036 * If the inode is already free, then there can be nothing
1037 * to clean up here.
1038 */
Christoph Hellwigcb4c8cc2009-03-16 08:25:25 +01001039 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 ASSERT(ip->i_df.if_real_bytes == 0);
1041 ASSERT(ip->i_df.if_broot_bytes == 0);
1042 return VN_INACTIVE_CACHE;
1043 }
1044
1045 /*
1046 * Only do a truncate if it's a regular file with
1047 * some actual space in it. It's OK to look at the
1048 * inode's fields without the lock because we're the
1049 * only one with a reference to the inode.
1050 */
1051 truncate = ((ip->i_d.di_nlink == 0) &&
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001052 ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
1053 (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1055
1056 mp = ip->i_mount;
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 error = 0;
1059
1060 /* If this is a read-only mount, don't do this (would generate I/O) */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001061 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 goto out;
1063
1064 if (ip->i_d.di_nlink != 0) {
1065 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
Christoph Hellwigdf80c932008-08-13 16:22:09 +10001066 ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
Yingping Lu68bdb6e2006-01-11 15:38:31 +11001067 ip->i_delayed_blks > 0)) &&
Nathan Scottdd9f4382006-01-11 15:28:28 +11001068 (ip->i_df.if_flags & XFS_IFEXTENTS) &&
1069 (!(ip->i_d.di_flags &
1070 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
1071 (ip->i_delayed_blks != 0)))) {
Christoph Hellwigc56c9632009-10-19 04:03:46 +00001072 error = xfs_free_eofblocks(mp, ip, 0);
David Chinner92dfe8d2007-05-24 15:22:19 +10001073 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001074 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 }
1076 goto out;
1077 }
1078
1079 ASSERT(ip->i_d.di_nlink == 0);
1080
Christoph Hellwig7d095252009-06-08 15:33:32 +02001081 error = xfs_qm_dqattach(ip, 0);
1082 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001083 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
1085 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1086 if (truncate) {
1087 /*
1088 * Do the xfs_itruncate_start() call before
1089 * reserving any log space because itruncate_start
1090 * will call into the buffer cache and we can't
1091 * do that within a transaction.
1092 */
1093 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1094
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001095 error = xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1096 if (error) {
Jesper Juhl87ae3c22007-06-28 16:43:14 +10001097 xfs_trans_cancel(tp, 0);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001098 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1099 return VN_INACTIVE_CACHE;
1100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102 error = xfs_trans_reserve(tp, 0,
1103 XFS_ITRUNCATE_LOG_RES(mp),
1104 0, XFS_TRANS_PERM_LOG_RES,
1105 XFS_ITRUNCATE_LOG_COUNT);
1106 if (error) {
1107 /* Don't call itruncate_cleanup */
1108 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1109 xfs_trans_cancel(tp, 0);
1110 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001111 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 }
1113
1114 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001115 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117 /*
1118 * normally, we have to run xfs_itruncate_finish sync.
1119 * But if filesystem is wsync and we're in the inactive
1120 * path, then we know that nlink == 0, and that the
1121 * xaction that made nlink == 0 is permanently committed
1122 * since xfs_remove runs as a synchronous transaction.
1123 */
1124 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1125 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1126
1127 if (error) {
1128 xfs_trans_cancel(tp,
1129 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1130 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001131 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 }
1133 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1134
1135 /*
1136 * If we get an error while cleaning up a
1137 * symlink we bail out.
1138 */
1139 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1140 xfs_inactive_symlink_rmt(ip, &tp) :
1141 xfs_inactive_symlink_local(ip, &tp);
1142
1143 if (error) {
1144 ASSERT(tp == NULL);
Jesper Juhl014c2542006-01-15 02:37:08 +01001145 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 }
1147
Christoph Hellwig898621d2010-06-24 11:36:58 +10001148 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 } else {
1150 error = xfs_trans_reserve(tp, 0,
1151 XFS_IFREE_LOG_RES(mp),
1152 0, XFS_TRANS_PERM_LOG_RES,
1153 XFS_INACTIVE_LOG_COUNT);
1154 if (error) {
1155 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1156 xfs_trans_cancel(tp, 0);
Jesper Juhl014c2542006-01-15 02:37:08 +01001157 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 }
1159
1160 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001161 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 }
1163
1164 /*
1165 * If there are attributes associated with the file
1166 * then blow them away now. The code calls a routine
1167 * that recursively deconstructs the attribute fork.
1168 * We need to just commit the current transaction
1169 * because we can't use it for xfs_attr_inactive().
1170 */
1171 if (ip->i_d.di_anextents > 0) {
1172 error = xfs_inactive_attrs(ip, &tp);
1173 /*
1174 * If we got an error, the transaction is already
1175 * cancelled, and the inode is unlocked. Just get out.
1176 */
1177 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01001178 return VN_INACTIVE_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 } else if (ip->i_afp) {
1180 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1181 }
1182
1183 /*
1184 * Free the inode.
1185 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06001186 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 error = xfs_ifree(tp, ip, &free_list);
1188 if (error) {
1189 /*
1190 * If we fail to free the inode, shut down. The cancel
1191 * might do that, we need to make sure. Otherwise the
1192 * inode might be lost for a long time or forever.
1193 */
1194 if (!XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001195 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1196 __func__, error);
Nathan Scott7d04a332006-06-09 14:58:38 +10001197 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 }
1199 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1200 } else {
1201 /*
1202 * Credit the quota account(s). The inode is gone.
1203 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001204 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
1206 /*
David Chinner78e9da72008-04-10 12:24:17 +10001207 * Just ignore errors at this point. There is nothing we can
1208 * do except to try to keep going. Make sure it's not a silent
1209 * error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 */
David Chinner78e9da72008-04-10 12:24:17 +10001211 error = xfs_bmap_finish(&tp, &free_list, &committed);
1212 if (error)
Dave Chinner53487782011-03-07 10:05:35 +11001213 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1214 __func__, error);
David Chinner78e9da72008-04-10 12:24:17 +10001215 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1216 if (error)
Dave Chinner53487782011-03-07 10:05:35 +11001217 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1218 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 }
Christoph Hellwig7d095252009-06-08 15:33:32 +02001220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 /*
1222 * Release the dquots held by inode, if any.
1223 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001224 xfs_qm_dqdetach(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1226
1227 out:
1228 return VN_INACTIVE_CACHE;
1229}
1230
Barry Naujok384f3ce2008-05-21 16:58:22 +10001231/*
1232 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1233 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1234 * ci_name->name will point to a the actual name (caller must free) or
1235 * will be set to NULL if an exact match is found.
1236 */
Christoph Hellwig993386c2007-08-28 16:12:30 +10001237int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238xfs_lookup(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001239 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001240 struct xfs_name *name,
Barry Naujok384f3ce2008-05-21 16:58:22 +10001241 xfs_inode_t **ipp,
1242 struct xfs_name *ci_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001244 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 int error;
1246 uint lock_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001248 trace_xfs_lookup(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1251 return XFS_ERROR(EIO);
1252
1253 lock_mode = xfs_ilock_map_shared(dp);
Barry Naujok384f3ce2008-05-21 16:58:22 +10001254 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 xfs_iunlock_map_shared(dp, lock_mode);
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001256
1257 if (error)
1258 goto out;
1259
Dave Chinner7b6259e2010-06-24 11:35:17 +10001260 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001261 if (error)
Barry Naujok384f3ce2008-05-21 16:58:22 +10001262 goto out_free_name;
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001263
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001264 return 0;
1265
Barry Naujok384f3ce2008-05-21 16:58:22 +10001266out_free_name:
1267 if (ci_name)
1268 kmem_free(ci_name->name);
1269out:
Christoph Hellwigeca450b2008-04-22 17:33:52 +10001270 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 return error;
1272}
1273
Christoph Hellwig993386c2007-08-28 16:12:30 +10001274int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275xfs_create(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001276 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001277 struct xfs_name *name,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001278 mode_t mode,
1279 xfs_dev_t rdev,
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00001280 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281{
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001282 int is_dir = S_ISDIR(mode);
1283 struct xfs_mount *mp = dp->i_mount;
1284 struct xfs_inode *ip = NULL;
1285 struct xfs_trans *tp = NULL;
Barry Naujok556b8b12008-04-10 12:22:07 +10001286 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 xfs_bmap_free_t free_list;
1288 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001289 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 uint cancel_flags;
1291 int committed;
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001292 prid_t prid;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001293 struct xfs_dquot *udqp = NULL;
1294 struct xfs_dquot *gdqp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 uint resblks;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001296 uint log_res;
1297 uint log_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001299 trace_xfs_create(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001301 if (XFS_FORCED_SHUTDOWN(mp))
1302 return XFS_ERROR(EIO);
1303
Nathan Scott365ca832005-06-21 15:39:12 +10001304 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001305 prid = xfs_get_projid(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 else
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001307 prid = XFS_PROJID_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
1309 /*
1310 * Make sure that we have allocated dquot(s) on disk.
1311 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001312 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001313 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001315 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001317 if (is_dir) {
1318 rdev = 0;
1319 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1320 log_res = XFS_MKDIR_LOG_RES(mp);
1321 log_count = XFS_MKDIR_LOG_COUNT;
1322 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1323 } else {
1324 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1325 log_res = XFS_CREATE_LOG_RES(mp);
1326 log_count = XFS_CREATE_LOG_COUNT;
1327 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1328 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 /*
1333 * Initially assume that the file does not exist and
1334 * reserve the resources for that case. If that is not
1335 * the case we'll drop the one we have and get a more
1336 * appropriate transaction later.
1337 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001338 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1339 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (error == ENOSPC) {
Dave Chinner153fec42009-04-06 18:48:30 +02001341 /* flush outstanding delalloc blocks and retry */
1342 xfs_flush_inodes(dp);
Christoph Hellwig4734d402009-09-09 18:19:02 -05001343 error = xfs_trans_reserve(tp, resblks, log_res, 0,
1344 XFS_TRANS_PERM_LOG_RES, log_count);
Dave Chinner153fec42009-04-06 18:48:30 +02001345 }
1346 if (error == ENOSPC) {
1347 /* No space at all so try a "no-allocation" reservation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 resblks = 0;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001349 error = xfs_trans_reserve(tp, 0, log_res, 0,
1350 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
1352 if (error) {
1353 cancel_flags = 0;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001354 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 }
1356
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001357 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001358 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001360 /*
1361 * Check for directory link count overflow.
1362 */
1363 if (is_dir && dp->i_d.di_nlink >= XFS_MAXLINK) {
1364 error = XFS_ERROR(EMLINK);
1365 goto out_trans_cancel;
1366 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001368 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370 /*
1371 * Reserve disk quota and the inode.
1372 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001373 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (error)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001375 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
Barry Naujok556b8b12008-04-10 12:22:07 +10001377 error = xfs_dir_canenter(tp, dp, name, resblks);
1378 if (error)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001379 goto out_trans_cancel;
1380
1381 /*
1382 * A newly created regular or special file just has one directory
1383 * entry pointing to them, but a directory also the "." entry
1384 * pointing to itself.
1385 */
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00001386 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001387 prid, resblks > 0, &ip, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (error) {
1389 if (error == ENOSPC)
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001390 goto out_trans_cancel;
1391 goto out_trans_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
1394 /*
Christoph Hellwig993386c2007-08-28 16:12:30 +10001395 * Now we join the directory inode to the transaction. We do not do it
1396 * earlier because xfs_dir_ialloc might commit the previous transaction
1397 * (and release all the locks). An error from here on will result in
1398 * the transaction cancel unlocking dp so don't do it explicitly in the
1399 * error path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 */
Christoph Hellwig898621d2010-06-24 11:36:58 +10001401 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10001402 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Barry Naujok556b8b12008-04-10 12:22:07 +10001404 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10001405 &first_block, &free_list, resblks ?
1406 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 if (error) {
1408 ASSERT(error != ENOSPC);
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001409 goto out_trans_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 }
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001411 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1413
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001414 if (is_dir) {
1415 error = xfs_dir_init(tp, ip, dp);
1416 if (error)
1417 goto out_bmap_cancel;
1418
1419 error = xfs_bumplink(tp, dp);
1420 if (error)
1421 goto out_bmap_cancel;
1422 }
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 /*
1425 * If this is a synchronous mount, make sure that the
1426 * create transaction goes to disk before returning to
1427 * the user.
1428 */
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001429 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 /*
1433 * Attach the dquot(s) to the inodes and modify them incore.
1434 * These ids of the inode couldn't have changed since the new
1435 * inode has been locked ever since it was created.
1436 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001437 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001439 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001440 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001441 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001443 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001444 if (error)
1445 goto out_release_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
Christoph Hellwig7d095252009-06-08 15:33:32 +02001447 xfs_qm_dqrele(udqp);
1448 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Christoph Hellwig979ebab2008-03-06 13:46:05 +11001450 *ipp = ip;
Christoph Hellwig288699f2010-06-23 18:11:15 +10001451 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001453 out_bmap_cancel:
1454 xfs_bmap_cancel(&free_list);
1455 out_trans_abort:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 cancel_flags |= XFS_TRANS_ABORT;
Christoph Hellwig517b5e82009-02-09 08:38:02 +01001457 out_trans_cancel:
1458 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001459 out_release_inode:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 /*
1461 * Wait until after the current transaction is aborted to
1462 * release the inode. This prevents recursive transactions
1463 * and deadlocks from xfs_inactive.
1464 */
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001465 if (ip)
1466 IRELE(ip);
1467
1468 xfs_qm_dqrele(udqp);
1469 xfs_qm_dqrele(gdqp);
1470
1471 if (unlock_dp_on_error)
1472 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1473 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
1475
1476#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477int xfs_locked_n;
1478int xfs_small_retries;
1479int xfs_middle_retries;
1480int xfs_lots_retries;
1481int xfs_lock_delays;
1482#endif
1483
1484/*
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001485 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1486 * a different value
1487 */
1488static inline int
1489xfs_lock_inumorder(int lock_mode, int subclass)
1490{
1491 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10001492 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001493 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
David Chinner0f1145c2007-06-29 17:26:09 +10001494 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001495
1496 return lock_mode;
1497}
1498
1499/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 * The following routine will lock n inodes in exclusive mode.
1501 * We assume the caller calls us with the inodes in i_ino order.
1502 *
1503 * We need to detect deadlock where an inode that we lock
1504 * is in the AIL and we start waiting for another inode that is locked
1505 * by a thread in a long running transaction (such as truncate). This can
1506 * result in deadlock since the long running trans might need to wait
1507 * for the inode we just locked in order to push the tail and free space
1508 * in the log.
1509 */
1510void
1511xfs_lock_inodes(
1512 xfs_inode_t **ips,
1513 int inodes,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 uint lock_mode)
1515{
1516 int attempts = 0, i, j, try_lock;
1517 xfs_log_item_t *lp;
1518
1519 ASSERT(ips && (inodes >= 2)); /* we need at least two */
1520
Christoph Hellwigcfa853e2008-04-22 17:34:06 +10001521 try_lock = 0;
1522 i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
1524again:
1525 for (; i < inodes; i++) {
1526 ASSERT(ips[i]);
1527
1528 if (i && (ips[i] == ips[i-1])) /* Already locked */
1529 continue;
1530
1531 /*
1532 * If try_lock is not set yet, make sure all locked inodes
1533 * are not in the AIL.
1534 * If any are, set try_lock to be used later.
1535 */
1536
1537 if (!try_lock) {
1538 for (j = (i - 1); j >= 0 && !try_lock; j--) {
1539 lp = (xfs_log_item_t *)ips[j]->i_itemp;
1540 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1541 try_lock++;
1542 }
1543 }
1544 }
1545
1546 /*
1547 * If any of the previous locks we have locked is in the AIL,
1548 * we must TRY to get the second and subsequent locks. If
1549 * we can't get any, we must release all we have
1550 * and try again.
1551 */
1552
1553 if (try_lock) {
1554 /* try_lock must be 0 if i is 0. */
1555 /*
1556 * try_lock means we have an inode locked
1557 * that is in the AIL.
1558 */
1559 ASSERT(i != 0);
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001560 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 attempts++;
1562
1563 /*
1564 * Unlock all previous guys and try again.
1565 * xfs_iunlock will try to push the tail
1566 * if the inode is in the AIL.
1567 */
1568
1569 for(j = i - 1; j >= 0; j--) {
1570
1571 /*
1572 * Check to see if we've already
1573 * unlocked this one.
1574 * Not the first one going back,
1575 * and the inode ptr is the same.
1576 */
1577 if ((j != (i - 1)) && ips[j] ==
1578 ips[j+1])
1579 continue;
1580
1581 xfs_iunlock(ips[j], lock_mode);
1582 }
1583
1584 if ((attempts % 5) == 0) {
1585 delay(1); /* Don't just spin the CPU */
1586#ifdef DEBUG
1587 xfs_lock_delays++;
1588#endif
1589 }
1590 i = 0;
1591 try_lock = 0;
1592 goto again;
1593 }
1594 } else {
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10001595 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
1597 }
1598
1599#ifdef DEBUG
1600 if (attempts) {
1601 if (attempts < 5) xfs_small_retries++;
1602 else if (attempts < 100) xfs_middle_retries++;
1603 else xfs_lots_retries++;
1604 } else {
1605 xfs_locked_n++;
1606 }
1607#endif
1608}
1609
David Chinnerf9114eb2008-09-17 16:51:21 +10001610/*
1611 * xfs_lock_two_inodes() can only be used to lock one type of lock
1612 * at a time - the iolock or the ilock, but not both at once. If
1613 * we lock both at once, lockdep will report false positives saying
1614 * we have violated locking orders.
1615 */
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001616void
1617xfs_lock_two_inodes(
1618 xfs_inode_t *ip0,
1619 xfs_inode_t *ip1,
1620 uint lock_mode)
1621{
1622 xfs_inode_t *temp;
1623 int attempts = 0;
1624 xfs_log_item_t *lp;
1625
David Chinnerf9114eb2008-09-17 16:51:21 +10001626 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
1627 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001628 ASSERT(ip0->i_ino != ip1->i_ino);
1629
1630 if (ip0->i_ino > ip1->i_ino) {
1631 temp = ip0;
1632 ip0 = ip1;
1633 ip1 = temp;
1634 }
1635
1636 again:
1637 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
1638
1639 /*
1640 * If the first lock we have locked is in the AIL, we must TRY to get
1641 * the second lock. If we can't get it, we must release the first one
1642 * and try again.
1643 */
1644 lp = (xfs_log_item_t *)ip0->i_itemp;
1645 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
1646 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
1647 xfs_iunlock(ip0, lock_mode);
1648 if ((++attempts % 5) == 0)
1649 delay(1); /* Don't just spin the CPU */
1650 goto again;
1651 }
1652 } else {
1653 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
1654 }
1655}
1656
Christoph Hellwig993386c2007-08-28 16:12:30 +10001657int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658xfs_remove(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001659 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001660 struct xfs_name *name,
1661 xfs_inode_t *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001663 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 xfs_trans_t *tp = NULL;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001665 int is_dir = S_ISDIR(ip->i_d.di_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 int error = 0;
1667 xfs_bmap_free_t free_list;
1668 xfs_fsblock_t first_block;
1669 int cancel_flags;
1670 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 int link_zero;
1672 uint resblks;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001673 uint log_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001675 trace_xfs_remove(dp, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 if (XFS_FORCED_SHUTDOWN(mp))
1678 return XFS_ERROR(EIO);
1679
Christoph Hellwig7d095252009-06-08 15:33:32 +02001680 error = xfs_qm_dqattach(dp, 0);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001681 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
Christoph Hellwig7d095252009-06-08 15:33:32 +02001684 error = xfs_qm_dqattach(ip, 0);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001685 if (error)
1686 goto std_return;
1687
1688 if (is_dir) {
1689 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
1690 log_count = XFS_DEFAULT_LOG_COUNT;
1691 } else {
1692 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
1693 log_count = XFS_REMOVE_LOG_COUNT;
1694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001696
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 /*
1698 * We try to get the real space reservation first,
1699 * allowing for directory btree deletion(s) implying
1700 * possible bmap insert(s). If we can't get the space
1701 * reservation then we use 0 instead, and avoid the bmap
1702 * btree insert(s) in the directory code by, if the bmap
1703 * insert tries to happen, instead trimming the LAST
1704 * block from the directory.
1705 */
1706 resblks = XFS_REMOVE_SPACE_RES(mp);
1707 error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001708 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 if (error == ENOSPC) {
1710 resblks = 0;
1711 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001712 XFS_TRANS_PERM_LOG_RES, log_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 }
1714 if (error) {
1715 ASSERT(error != ENOSPC);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001716 cancel_flags = 0;
1717 goto out_trans_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
1719
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001720 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
Christoph Hellwig898621d2010-06-24 11:36:58 +10001722 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
1723 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725 /*
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001726 * If we're removing a directory perform some additional validation.
1727 */
1728 if (is_dir) {
1729 ASSERT(ip->i_d.di_nlink >= 2);
1730 if (ip->i_d.di_nlink != 2) {
1731 error = XFS_ERROR(ENOTEMPTY);
1732 goto out_trans_cancel;
1733 }
1734 if (!xfs_dir_isempty(ip)) {
1735 error = XFS_ERROR(ENOTEMPTY);
1736 goto out_trans_cancel;
1737 }
1738 }
1739
Eric Sandeen9d87c312009-01-14 23:22:07 -06001740 xfs_bmap_init(&free_list, &first_block);
Barry Naujok556b8b12008-04-10 12:22:07 +10001741 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
Christoph Hellwigd4377d82008-04-22 17:33:46 +10001742 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 if (error) {
1744 ASSERT(error != ENOENT);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001745 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 }
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001747 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001749 if (is_dir) {
1750 /*
1751 * Drop the link from ip's "..".
1752 */
1753 error = xfs_droplink(tp, dp);
1754 if (error)
1755 goto out_bmap_cancel;
1756
1757 /*
Christoph Hellwig2b7035f2008-10-30 17:55:18 +11001758 * Drop the "." link from ip to self.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001759 */
1760 error = xfs_droplink(tp, ip);
1761 if (error)
1762 goto out_bmap_cancel;
1763 } else {
1764 /*
1765 * When removing a non-directory we need to log the parent
Dave Chinner26c52952008-11-28 14:23:37 +11001766 * inode here. For a directory this is done implicitly
1767 * by the xfs_droplink call for the ".." entry.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001768 */
1769 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 }
1771
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001772 /*
Christoph Hellwig2b7035f2008-10-30 17:55:18 +11001773 * Drop the link from dp to ip.
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001774 */
1775 error = xfs_droplink(tp, ip);
1776 if (error)
1777 goto out_bmap_cancel;
1778
1779 /*
1780 * Determine if this is the last link while
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 * we are in the transaction.
1782 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001783 link_zero = (ip->i_d.di_nlink == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
1785 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 * If this is a synchronous mount, make sure that the
1787 * remove transaction goes to disk before returning to
1788 * the user.
1789 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001790 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001793 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001794 if (error)
1795 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
Eric Sandeen1c72bf92007-05-08 13:48:42 +10001797 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwig5df78e72008-04-22 17:34:24 +10001798 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 goto std_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
1801 /*
David Chinner2a82b8b2007-07-11 11:09:12 +10001802 * If we are using filestreams, kill the stream association.
1803 * If the file is still open it may get a new one but that
1804 * will get killed on last close in xfs_close() so we don't
1805 * have to worry about that.
1806 */
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001807 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
David Chinner2a82b8b2007-07-11 11:09:12 +10001808 xfs_filestream_deassociate(ip);
1809
Christoph Hellwig288699f2010-06-23 18:11:15 +10001810 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001812 out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 xfs_bmap_cancel(&free_list);
1814 cancel_flags |= XFS_TRANS_ABORT;
Christoph Hellwig8f112e32008-06-23 13:25:17 +10001815 out_trans_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig288699f2010-06-23 18:11:15 +10001817 std_return:
1818 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819}
1820
Christoph Hellwig993386c2007-08-28 16:12:30 +10001821int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822xfs_link(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001823 xfs_inode_t *tdp,
Christoph Hellwiga3da7892008-03-06 13:46:12 +11001824 xfs_inode_t *sip,
Barry Naujok556b8b12008-04-10 12:22:07 +10001825 struct xfs_name *target_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001827 xfs_mount_t *mp = tdp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 int error;
1830 xfs_bmap_free_t free_list;
1831 xfs_fsblock_t first_block;
1832 int cancel_flags;
1833 int committed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 int resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001836 trace_xfs_link(tdp, target_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Christoph Hellwiga3da7892008-03-06 13:46:12 +11001838 ASSERT(!S_ISDIR(sip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 if (XFS_FORCED_SHUTDOWN(mp))
1841 return XFS_ERROR(EIO);
1842
Christoph Hellwig7d095252009-06-08 15:33:32 +02001843 error = xfs_qm_dqattach(sip, 0);
Christoph Hellwigcb3f35b2009-02-04 09:34:20 +01001844 if (error)
1845 goto std_return;
1846
Christoph Hellwig7d095252009-06-08 15:33:32 +02001847 error = xfs_qm_dqattach(tdp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 if (error)
1849 goto std_return;
1850
1851 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1852 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
Barry Naujok556b8b12008-04-10 12:22:07 +10001853 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
1855 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1856 if (error == ENOSPC) {
1857 resblks = 0;
1858 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
1859 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
1860 }
1861 if (error) {
1862 cancel_flags = 0;
1863 goto error_return;
1864 }
1865
Christoph Hellwige1cccd92008-08-13 16:18:07 +10001866 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
Christoph Hellwig898621d2010-06-24 11:36:58 +10001868 xfs_trans_ijoin_ref(tp, sip, XFS_ILOCK_EXCL);
1869 xfs_trans_ijoin_ref(tp, tdp, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
1871 /*
1872 * If the source has too many links, we can't make any more to it.
1873 */
1874 if (sip->i_d.di_nlink >= XFS_MAXLINK) {
1875 error = XFS_ERROR(EMLINK);
1876 goto error_return;
1877 }
1878
Nathan Scott365ca832005-06-21 15:39:12 +10001879 /*
1880 * If we are using project inheritance, we only allow hard link
1881 * creation in our tree when the project IDs are the same; else
1882 * the tree quota mechanism could be circumvented.
1883 */
1884 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001885 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
Nathan Scottb1ecdda2006-05-08 19:51:42 +10001886 error = XFS_ERROR(EXDEV);
Nathan Scott365ca832005-06-21 15:39:12 +10001887 goto error_return;
1888 }
1889
Barry Naujok556b8b12008-04-10 12:22:07 +10001890 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1891 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 goto error_return;
1893
Eric Sandeen9d87c312009-01-14 23:22:07 -06001894 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
Barry Naujok556b8b12008-04-10 12:22:07 +10001896 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1897 &first_block, &free_list, resblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 if (error)
1899 goto abort_return;
Dave Chinnerdcd79a12010-09-28 12:27:25 +10001900 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1902
1903 error = xfs_bumplink(tp, sip);
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10001904 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 goto abort_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
1907 /*
1908 * If this is a synchronous mount, make sure that the
1909 * link transaction goes to disk before returning to
1910 * the user.
1911 */
1912 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1913 xfs_trans_set_sync(tp);
1914 }
1915
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001916 error = xfs_bmap_finish (&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 if (error) {
1918 xfs_bmap_cancel(&free_list);
1919 goto abort_return;
1920 }
1921
Christoph Hellwig288699f2010-06-23 18:11:15 +10001922 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 abort_return:
1925 cancel_flags |= XFS_TRANS_ABORT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 error_return:
1927 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig288699f2010-06-23 18:11:15 +10001928 std_return:
1929 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930}
Alexey Dobriyanb71d3002006-06-27 12:45:17 +10001931
Christoph Hellwig993386c2007-08-28 16:12:30 +10001932int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933xfs_symlink(
Christoph Hellwig993386c2007-08-28 16:12:30 +10001934 xfs_inode_t *dp,
Barry Naujok556b8b12008-04-10 12:22:07 +10001935 struct xfs_name *link_name,
1936 const char *target_path,
Christoph Hellwig3e5daf02007-10-11 18:09:12 +10001937 mode_t mode,
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00001938 xfs_inode_t **ipp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939{
Christoph Hellwig993386c2007-08-28 16:12:30 +10001940 xfs_mount_t *mp = dp->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 xfs_inode_t *ip;
1943 int error;
1944 int pathlen;
1945 xfs_bmap_free_t free_list;
1946 xfs_fsblock_t first_block;
Christoph Hellwig993386c2007-08-28 16:12:30 +10001947 boolean_t unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 uint cancel_flags;
1949 int committed;
1950 xfs_fileoff_t first_fsb;
1951 xfs_filblks_t fs_blocks;
1952 int nmaps;
1953 xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1954 xfs_daddr_t d;
Barry Naujok556b8b12008-04-10 12:22:07 +10001955 const char *cur_chunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 int byte_cnt;
1957 int n;
1958 xfs_buf_t *bp;
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001959 prid_t prid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 struct xfs_dquot *udqp, *gdqp;
1961 uint resblks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
Christoph Hellwig3937be52008-03-06 13:46:19 +11001963 *ipp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 error = 0;
1965 ip = NULL;
1966 tp = NULL;
1967
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001968 trace_xfs_symlink(dp, link_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
1970 if (XFS_FORCED_SHUTDOWN(mp))
1971 return XFS_ERROR(EIO);
1972
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 /*
1974 * Check component lengths of the target path name.
1975 */
1976 pathlen = strlen(target_path);
1977 if (pathlen >= MAXPATHLEN) /* total string too long */
1978 return XFS_ERROR(ENAMETOOLONG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 udqp = gdqp = NULL;
Nathan Scott365ca832005-06-21 15:39:12 +10001981 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001982 prid = xfs_get_projid(dp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 else
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001984 prid = XFS_PROJID_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
1986 /*
1987 * Make sure that we have allocated dquot(s) on disk.
1988 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02001989 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
1991 if (error)
1992 goto std_return;
1993
1994 tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
1995 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1996 /*
1997 * The symlink will fit into the inode data fork?
1998 * There can't be any attributes so we get the whole variable part.
1999 */
2000 if (pathlen <= XFS_LITINO(mp))
2001 fs_blocks = 0;
2002 else
2003 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
Barry Naujok556b8b12008-04-10 12:22:07 +10002004 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
2006 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2007 if (error == ENOSPC && fs_blocks == 0) {
2008 resblks = 0;
2009 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
2010 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
2011 }
2012 if (error) {
2013 cancel_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 goto error_return;
2015 }
2016
Lachlan McIlroyf7c66ce2007-05-08 13:50:19 +10002017 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002018 unlock_dp_on_error = B_TRUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
2020 /*
2021 * Check whether the directory allows new symlinks or not.
2022 */
2023 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
2024 error = XFS_ERROR(EPERM);
2025 goto error_return;
2026 }
2027
2028 /*
2029 * Reserve disk quota : blocks and inode.
2030 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02002031 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, resblks, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 if (error)
2033 goto error_return;
2034
2035 /*
2036 * Check for ability to enter directory entry, if no space reserved.
2037 */
Barry Naujok556b8b12008-04-10 12:22:07 +10002038 error = xfs_dir_canenter(tp, dp, link_name, resblks);
2039 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 goto error_return;
2041 /*
2042 * Initialize the bmap freelist prior to calling either
2043 * bmapi or the directory create code.
2044 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002045 xfs_bmap_init(&free_list, &first_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047 /*
2048 * Allocate an inode for the symlink.
2049 */
Christoph Hellwig6c77b0e2010-10-06 18:41:17 +00002050 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0,
2051 prid, resblks > 0, &ip, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 if (error) {
2053 if (error == ENOSPC)
2054 goto error_return;
2055 goto error1;
2056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Christoph Hellwig993386c2007-08-28 16:12:30 +10002058 /*
2059 * An error after we've joined dp to the transaction will result in the
2060 * transaction cancel unlocking dp so don't do it explicitly in the
2061 * error path.
2062 */
Christoph Hellwig898621d2010-06-24 11:36:58 +10002063 xfs_trans_ijoin_ref(tp, dp, XFS_ILOCK_EXCL);
Christoph Hellwig993386c2007-08-28 16:12:30 +10002064 unlock_dp_on_error = B_FALSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
2066 /*
2067 * Also attach the dquot(s) to it, if applicable.
2068 */
Christoph Hellwig7d095252009-06-08 15:33:32 +02002069 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071 if (resblks)
2072 resblks -= XFS_IALLOC_SPACE_RES(mp);
2073 /*
2074 * If the symlink will fit into the inode, write it inline.
2075 */
2076 if (pathlen <= XFS_IFORK_DSIZE(ip)) {
2077 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
2078 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
2079 ip->i_d.di_size = pathlen;
2080
2081 /*
2082 * The inode was initially created in extent format.
2083 */
2084 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
2085 ip->i_df.if_flags |= XFS_IFINLINE;
2086
2087 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
2088 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
2089
2090 } else {
2091 first_fsb = 0;
2092 nmaps = SYMLINK_MAPS;
2093
2094 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
2095 XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
2096 &first_block, resblks, mval, &nmaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002097 &free_list);
Christoph Hellwigec3ba852011-02-13 13:26:42 +00002098 if (error)
2099 goto error2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
2101 if (resblks)
2102 resblks -= fs_blocks;
2103 ip->i_d.di_size = pathlen;
2104 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2105
2106 cur_chunk = target_path;
2107 for (n = 0; n < nmaps; n++) {
2108 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
2109 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
2110 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
2111 BTOBB(byte_cnt), 0);
2112 ASSERT(bp && !XFS_BUF_GETERROR(bp));
2113 if (pathlen < byte_cnt) {
2114 byte_cnt = pathlen;
2115 }
2116 pathlen -= byte_cnt;
2117
2118 memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
2119 cur_chunk += byte_cnt;
2120
2121 xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
2122 }
2123 }
2124
2125 /*
2126 * Create the directory entry for the symlink.
2127 */
Barry Naujok556b8b12008-04-10 12:22:07 +10002128 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino,
2129 &first_block, &free_list, resblks);
Nathan Scottf6c2d1f2006-06-20 13:04:51 +10002130 if (error)
Christoph Hellwigec3ba852011-02-13 13:26:42 +00002131 goto error2;
Dave Chinnerdcd79a12010-09-28 12:27:25 +10002132 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2134
2135 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 * If this is a synchronous mount, make sure that the
2137 * symlink transaction goes to disk before returning to
2138 * the user.
2139 */
2140 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2141 xfs_trans_set_sync(tp);
2142 }
2143
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002144 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 if (error) {
2146 goto error2;
2147 }
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002148 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002149 xfs_qm_dqrele(udqp);
2150 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151
Christoph Hellwig288699f2010-06-23 18:11:15 +10002152 *ipp = ip;
2153 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
2155 error2:
2156 IRELE(ip);
2157 error1:
2158 xfs_bmap_cancel(&free_list);
2159 cancel_flags |= XFS_TRANS_ABORT;
2160 error_return:
2161 xfs_trans_cancel(tp, cancel_flags);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002162 xfs_qm_dqrele(udqp);
2163 xfs_qm_dqrele(gdqp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
Christoph Hellwig993386c2007-08-28 16:12:30 +10002165 if (unlock_dp_on_error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 xfs_iunlock(dp, XFS_ILOCK_EXCL);
Christoph Hellwig288699f2010-06-23 18:11:15 +10002167 std_return:
2168 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169}
2170
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171int
Christoph Hellwig993386c2007-08-28 16:12:30 +10002172xfs_set_dmattrs(
2173 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 u_int evmask,
Christoph Hellwig993386c2007-08-28 16:12:30 +10002175 u_int16_t state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002177 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 xfs_trans_t *tp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 int error;
2180
2181 if (!capable(CAP_SYS_ADMIN))
2182 return XFS_ERROR(EPERM);
2183
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 if (XFS_FORCED_SHUTDOWN(mp))
2185 return XFS_ERROR(EIO);
2186
2187 tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
2188 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
2189 if (error) {
2190 xfs_trans_cancel(tp, 0);
2191 return error;
2192 }
2193 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10002194 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195
Christoph Hellwig613d7042007-10-11 17:44:08 +10002196 ip->i_d.di_dmevmask = evmask;
2197 ip->i_d.di_dmstate = state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
2199 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002200 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201
2202 return error;
2203}
2204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205/*
2206 * xfs_alloc_file_space()
2207 * This routine allocates disk space for the given file.
2208 *
2209 * If alloc_type == 0, this request is for an ALLOCSP type
2210 * request which will change the file size. In this case, no
2211 * DMAPI event will be generated by the call. A TRUNCATE event
2212 * will be generated later by xfs_setattr.
2213 *
2214 * If alloc_type != 0, this request is for a RESVSP type
2215 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2216 * lower block boundary byte address is less than the file's
2217 * length.
2218 *
2219 * RETURNS:
2220 * 0 on success
2221 * errno on error
2222 *
2223 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002224STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225xfs_alloc_file_space(
2226 xfs_inode_t *ip,
2227 xfs_off_t offset,
2228 xfs_off_t len,
2229 int alloc_type,
2230 int attr_flags)
2231{
Nathan Scottdd9f4382006-01-11 15:28:28 +11002232 xfs_mount_t *mp = ip->i_mount;
2233 xfs_off_t count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 xfs_filblks_t allocated_fsb;
2235 xfs_filblks_t allocatesize_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002236 xfs_extlen_t extsz, temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 xfs_fileoff_t startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002238 xfs_fsblock_t firstfsb;
2239 int nimaps;
2240 int bmapi_flag;
2241 int quota_flag;
2242 int rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 xfs_trans_t *tp;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002244 xfs_bmbt_irec_t imaps[1], *imapp;
2245 xfs_bmap_free_t free_list;
2246 uint qblocks, resblks, resrtextents;
2247 int committed;
2248 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10002250 trace_xfs_alloc_file_space(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
2252 if (XFS_FORCED_SHUTDOWN(mp))
2253 return XFS_ERROR(EIO);
2254
Christoph Hellwig7d095252009-06-08 15:33:32 +02002255 error = xfs_qm_dqattach(ip, 0);
2256 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 return error;
2258
2259 if (len <= 0)
2260 return XFS_ERROR(EINVAL);
2261
David Chinner957d0eb2007-06-18 16:50:37 +10002262 rt = XFS_IS_REALTIME_INODE(ip);
2263 extsz = xfs_get_extsz_hint(ip);
2264
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 count = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 imapp = &imaps[0];
Nathan Scottdd9f4382006-01-11 15:28:28 +11002267 nimaps = 1;
Dave Chinner44722352010-08-24 12:02:11 +10002268 bmapi_flag = XFS_BMAPI_WRITE | alloc_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
2270 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
2271
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002273 * Allocate file space until done or until there is an error
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 while (allocatesize_fsb && !error) {
Nathan Scottdd9f4382006-01-11 15:28:28 +11002276 xfs_fileoff_t s, e;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
Nathan Scottdd9f4382006-01-11 15:28:28 +11002278 /*
Nathan Scott3ddb8fa2006-01-11 15:33:02 +11002279 * Determine space reservations for data/realtime.
Nathan Scottdd9f4382006-01-11 15:28:28 +11002280 */
2281 if (unlikely(extsz)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 s = startoffset_fsb;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002283 do_div(s, extsz);
2284 s *= extsz;
2285 e = startoffset_fsb + allocatesize_fsb;
2286 if ((temp = do_mod(startoffset_fsb, extsz)))
2287 e += temp;
2288 if ((temp = do_mod(e, extsz)))
2289 e += extsz - temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 } else {
Nathan Scottdd9f4382006-01-11 15:28:28 +11002291 s = 0;
2292 e = allocatesize_fsb;
2293 }
2294
Dave Chinner72656c42010-09-03 12:19:33 +10002295 /*
2296 * The transaction reservation is limited to a 32-bit block
2297 * count, hence we need to limit the number of blocks we are
2298 * trying to reserve to avoid an overflow. We can't allocate
2299 * more than @nimaps extents, and an extent is limited on disk
2300 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
2301 */
2302 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
Nathan Scottdd9f4382006-01-11 15:28:28 +11002303 if (unlikely(rt)) {
Dave Chinner72656c42010-09-03 12:19:33 +10002304 resrtextents = qblocks = resblks;
Nathan Scottdd9f4382006-01-11 15:28:28 +11002305 resrtextents /= mp->m_sb.sb_rextsize;
2306 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2307 quota_flag = XFS_QMOPT_RES_RTBLKS;
2308 } else {
2309 resrtextents = 0;
Dave Chinner72656c42010-09-03 12:19:33 +10002310 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
Nathan Scottdd9f4382006-01-11 15:28:28 +11002311 quota_flag = XFS_QMOPT_RES_REGBLKS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 }
2313
2314 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002315 * Allocate and setup the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 */
2317 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
Nathan Scottdd9f4382006-01-11 15:28:28 +11002318 error = xfs_trans_reserve(tp, resblks,
2319 XFS_WRITE_LOG_RES(mp), resrtextents,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 XFS_TRANS_PERM_LOG_RES,
2321 XFS_WRITE_LOG_COUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002323 * Check for running out of space
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 */
2325 if (error) {
2326 /*
2327 * Free the transaction structure.
2328 */
2329 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2330 xfs_trans_cancel(tp, 0);
2331 break;
2332 }
2333 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002334 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
2335 0, quota_flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 if (error)
2337 goto error1;
2338
Christoph Hellwig898621d2010-06-24 11:36:58 +10002339 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340
2341 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002342 * Issue the xfs_bmapi() call to allocate the blocks
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002344 xfs_bmap_init(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002345 error = xfs_bmapi(tp, ip, startoffset_fsb,
Nathan Scottdd9f4382006-01-11 15:28:28 +11002346 allocatesize_fsb, bmapi_flag,
2347 &firstfsb, 0, imapp, &nimaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002348 &free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 if (error) {
2350 goto error0;
2351 }
2352
2353 /*
Nathan Scottdd9f4382006-01-11 15:28:28 +11002354 * Complete the transaction
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002356 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 if (error) {
2358 goto error0;
2359 }
2360
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002361 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2363 if (error) {
2364 break;
2365 }
2366
2367 allocated_fsb = imapp->br_blockcount;
2368
Nathan Scottdd9f4382006-01-11 15:28:28 +11002369 if (nimaps == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 error = XFS_ERROR(ENOSPC);
2371 break;
2372 }
2373
2374 startoffset_fsb += allocated_fsb;
2375 allocatesize_fsb -= allocated_fsb;
2376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
2378 return error;
2379
Nathan Scottdd9f4382006-01-11 15:28:28 +11002380error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 xfs_bmap_cancel(&free_list);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002382 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
Nathan Scottdd9f4382006-01-11 15:28:28 +11002383
2384error1: /* Just cancel transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2386 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig288699f2010-06-23 18:11:15 +10002387 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388}
2389
2390/*
2391 * Zero file bytes between startoff and endoff inclusive.
2392 * The iolock is held exclusive and no blocks are buffered.
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10002393 *
2394 * This function is used by xfs_free_file_space() to zero
2395 * partial blocks when the range to free is not block aligned.
2396 * When unreserving space with boundaries that are not block
2397 * aligned we round up the start and round down the end
2398 * boundaries and then use this function to zero the parts of
2399 * the blocks that got dropped during the rounding.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 */
2401STATIC int
2402xfs_zero_remaining_bytes(
2403 xfs_inode_t *ip,
2404 xfs_off_t startoff,
2405 xfs_off_t endoff)
2406{
2407 xfs_bmbt_irec_t imap;
2408 xfs_fileoff_t offset_fsb;
2409 xfs_off_t lastoffset;
2410 xfs_off_t offset;
2411 xfs_buf_t *bp;
2412 xfs_mount_t *mp = ip->i_mount;
2413 int nimap;
2414 int error = 0;
2415
Lachlan McIlroy2fd6f6e2008-09-17 16:52:50 +10002416 /*
2417 * Avoid doing I/O beyond eof - it's not necessary
2418 * since nothing can read beyond eof. The space will
2419 * be zeroed when the file is extended anyway.
2420 */
2421 if (startoff >= ip->i_size)
2422 return 0;
2423
2424 if (endoff > ip->i_size)
2425 endoff = ip->i_size;
2426
Dave Chinner686865f2010-09-24 20:07:47 +10002427 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
2428 mp->m_rtdev_targp : mp->m_ddev_targp,
2429 mp->m_sb.sb_blocksize, XBF_DONT_BLOCK);
Lachlan McIlroyc6422612008-12-05 13:16:15 +11002430 if (!bp)
2431 return XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432
2433 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
2434 offset_fsb = XFS_B_TO_FSBT(mp, offset);
2435 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002436 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002437 NULL, 0, &imap, &nimap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 if (error || nimap < 1)
2439 break;
2440 ASSERT(imap.br_blockcount >= 1);
2441 ASSERT(imap.br_startoff == offset_fsb);
2442 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
2443 if (lastoffset > endoff)
2444 lastoffset = endoff;
2445 if (imap.br_startblock == HOLESTARTBLOCK)
2446 continue;
2447 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2448 if (imap.br_state == XFS_EXT_UNWRITTEN)
2449 continue;
2450 XFS_BUF_UNDONE(bp);
2451 XFS_BUF_UNWRITE(bp);
2452 XFS_BUF_READ(bp);
Eric Sandeen9d87c312009-01-14 23:22:07 -06002453 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 xfsbdstrat(mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002455 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10002456 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2458 mp, bp, XFS_BUF_ADDR(bp));
2459 break;
2460 }
2461 memset(XFS_BUF_PTR(bp) +
2462 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
2463 0, lastoffset - offset + 1);
2464 XFS_BUF_UNDONE(bp);
2465 XFS_BUF_UNREAD(bp);
2466 XFS_BUF_WRITE(bp);
2467 xfsbdstrat(mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002468 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10002469 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2471 mp, bp, XFS_BUF_ADDR(bp));
2472 break;
2473 }
2474 }
2475 xfs_buf_free(bp);
2476 return error;
2477}
2478
2479/*
2480 * xfs_free_file_space()
2481 * This routine frees disk space for the given file.
2482 *
2483 * This routine is only called by xfs_change_file_space
2484 * for an UNRESVSP type call.
2485 *
2486 * RETURNS:
2487 * 0 on success
2488 * errno on error
2489 *
2490 */
2491STATIC int
2492xfs_free_file_space(
2493 xfs_inode_t *ip,
2494 xfs_off_t offset,
2495 xfs_off_t len,
2496 int attr_flags)
2497{
2498 int committed;
2499 int done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 xfs_fileoff_t endoffset_fsb;
2501 int error;
2502 xfs_fsblock_t firstfsb;
2503 xfs_bmap_free_t free_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 xfs_bmbt_irec_t imap;
2505 xfs_off_t ioffset;
2506 xfs_extlen_t mod=0;
2507 xfs_mount_t *mp;
2508 int nimap;
2509 uint resblks;
Nathan Scott673cdf52006-09-28 10:56:26 +10002510 uint rounding;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 int rt;
2512 xfs_fileoff_t startoffset_fsb;
2513 xfs_trans_t *tp;
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002514 int need_iolock = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 mp = ip->i_mount;
2517
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10002518 trace_xfs_free_file_space(ip);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002519
Christoph Hellwig7d095252009-06-08 15:33:32 +02002520 error = xfs_qm_dqattach(ip, 0);
2521 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 return error;
2523
2524 error = 0;
2525 if (len <= 0) /* if nothing being freed */
2526 return error;
Eric Sandeen71ddabb2007-11-23 16:29:42 +11002527 rt = XFS_IS_REALTIME_INODE(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
Christoph Hellwig288699f2010-06-23 18:11:15 +10002529 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002531 if (attr_flags & XFS_ATTR_NOLOCK)
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002532 need_iolock = 0;
Yingping Lu9fa80462006-03-22 12:44:35 +11002533 if (need_iolock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwig25e41b32008-12-03 12:20:39 +01002535 /* wait for the completion of any pending DIOs */
2536 xfs_ioend_wait(ip);
Yingping Lu9fa80462006-03-22 12:44:35 +11002537 }
Dean Roehrich5fcbab32005-05-05 13:27:19 -07002538
Tim Shimmine6a4b372007-11-23 16:30:42 +11002539 rounding = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 ioffset = offset & ~(rounding - 1);
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002541
Christoph Hellwigdf80c932008-08-13 16:22:09 +10002542 if (VN_CACHED(VFS_I(ip)) != 0) {
Tim Shimmine6a4b372007-11-23 16:30:42 +11002543 error = xfs_flushinval_pages(ip, ioffset, -1, FI_REMAPF_LOCKED);
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10002544 if (error)
2545 goto out_unlock_iolock;
Christoph Hellwigbd5a8762005-06-21 15:47:39 +10002546 }
2547
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 /*
2549 * Need to zero the stuff we're not freeing, on disk.
Malcolm Parsons9da096f2009-03-29 09:55:42 +02002550 * If it's a realtime file & can't use unwritten extents then we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2552 * will take care of it for us.
2553 */
Eric Sandeen62118702008-03-06 13:44:28 +11002554 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002556 error = xfs_bmapi(NULL, ip, startoffset_fsb,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002557 1, 0, NULL, 0, &imap, &nimap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 if (error)
2559 goto out_unlock_iolock;
2560 ASSERT(nimap == 0 || nimap == 1);
2561 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2562 xfs_daddr_t block;
2563
2564 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2565 block = imap.br_startblock;
2566 mod = do_div(block, mp->m_sb.sb_rextsize);
2567 if (mod)
2568 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
2569 }
2570 nimap = 1;
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002571 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002572 1, 0, NULL, 0, &imap, &nimap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 if (error)
2574 goto out_unlock_iolock;
2575 ASSERT(nimap == 0 || nimap == 1);
2576 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
2577 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
2578 mod++;
2579 if (mod && (mod != mp->m_sb.sb_rextsize))
2580 endoffset_fsb -= mod;
2581 }
2582 }
2583 if ((done = (endoffset_fsb <= startoffset_fsb)))
2584 /*
2585 * One contiguous piece to clear
2586 */
2587 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
2588 else {
2589 /*
2590 * Some full blocks, possibly two pieces to clear
2591 */
2592 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
2593 error = xfs_zero_remaining_bytes(ip, offset,
2594 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
2595 if (!error &&
2596 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
2597 error = xfs_zero_remaining_bytes(ip,
2598 XFS_FSB_TO_B(mp, endoffset_fsb),
2599 offset + len - 1);
2600 }
2601
2602 /*
2603 * free file space until done or until there is an error
2604 */
2605 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
2606 while (!error && !done) {
2607
2608 /*
David Chinner91ebecc2007-07-19 16:28:30 +10002609 * allocate and setup the transaction. Allow this
2610 * transaction to dip into the reserve blocks to ensure
2611 * the freeing of the space succeeds at ENOSPC.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 */
2613 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
David Chinner91ebecc2007-07-19 16:28:30 +10002614 tp->t_flags |= XFS_TRANS_RESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 error = xfs_trans_reserve(tp,
2616 resblks,
2617 XFS_WRITE_LOG_RES(mp),
2618 0,
2619 XFS_TRANS_PERM_LOG_RES,
2620 XFS_WRITE_LOG_COUNT);
2621
2622 /*
2623 * check for running out of space
2624 */
2625 if (error) {
2626 /*
2627 * Free the transaction structure.
2628 */
2629 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
2630 xfs_trans_cancel(tp, 0);
2631 break;
2632 }
2633 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig7d095252009-06-08 15:33:32 +02002634 error = xfs_trans_reserve_quota(tp, mp,
2635 ip->i_udquot, ip->i_gdquot,
2636 resblks, 0, XFS_QMOPT_RES_REGBLKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 if (error)
2638 goto error1;
2639
Christoph Hellwig898621d2010-06-24 11:36:58 +10002640 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
2642 /*
2643 * issue the bunmapi() call to free the blocks
2644 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06002645 xfs_bmap_init(&free_list, &firstfsb);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10002646 error = xfs_bunmapi(tp, ip, startoffset_fsb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 endoffset_fsb - startoffset_fsb,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10002648 0, 2, &firstfsb, &free_list, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 if (error) {
2650 goto error0;
2651 }
2652
2653 /*
2654 * complete the transaction
2655 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11002656 error = xfs_bmap_finish(&tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 if (error) {
2658 goto error0;
2659 }
2660
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002661 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2663 }
2664
2665 out_unlock_iolock:
2666 if (need_iolock)
2667 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
2668 return error;
2669
2670 error0:
2671 xfs_bmap_cancel(&free_list);
2672 error1:
2673 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
2674 xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
2675 XFS_ILOCK_EXCL);
2676 return error;
2677}
2678
2679/*
2680 * xfs_change_file_space()
2681 * This routine allocates or frees disk space for the given file.
2682 * The user specified parameters are checked for alignment and size
2683 * limitations.
2684 *
2685 * RETURNS:
2686 * 0 on success
2687 * errno on error
2688 *
2689 */
2690int
2691xfs_change_file_space(
Christoph Hellwig993386c2007-08-28 16:12:30 +10002692 xfs_inode_t *ip,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 int cmd,
2694 xfs_flock64_t *bf,
2695 xfs_off_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 int attr_flags)
2697{
Christoph Hellwig993386c2007-08-28 16:12:30 +10002698 xfs_mount_t *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 int clrprealloc;
2700 int error;
2701 xfs_fsize_t fsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 int setprealloc;
2703 xfs_off_t startoffset;
2704 xfs_off_t llen;
2705 xfs_trans_t *tp;
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002706 struct iattr iattr;
Dave Chinner44722352010-08-24 12:02:11 +10002707 int prealloc_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708
Christoph Hellwig993386c2007-08-28 16:12:30 +10002709 if (!S_ISREG(ip->i_d.di_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 return XFS_ERROR(EINVAL);
2711
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 switch (bf->l_whence) {
2713 case 0: /*SEEK_SET*/
2714 break;
2715 case 1: /*SEEK_CUR*/
2716 bf->l_start += offset;
2717 break;
2718 case 2: /*SEEK_END*/
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002719 bf->l_start += ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 break;
2721 default:
2722 return XFS_ERROR(EINVAL);
2723 }
2724
2725 llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
2726
2727 if ( (bf->l_start < 0)
2728 || (bf->l_start > XFS_MAXIOFFSET(mp))
2729 || (bf->l_start + llen < 0)
2730 || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
2731 return XFS_ERROR(EINVAL);
2732
2733 bf->l_whence = 0;
2734
2735 startoffset = bf->l_start;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002736 fsize = ip->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737
2738 /*
2739 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2740 * file space.
2741 * These calls do NOT zero the data space allocated to the file,
2742 * nor do they change the file size.
2743 *
2744 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2745 * space.
2746 * These calls cause the new file data to be zeroed and the file
2747 * size to be changed.
2748 */
2749 setprealloc = clrprealloc = 0;
Dave Chinner44722352010-08-24 12:02:11 +10002750 prealloc_type = XFS_BMAPI_PREALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751
2752 switch (cmd) {
Dave Chinner44722352010-08-24 12:02:11 +10002753 case XFS_IOC_ZERO_RANGE:
2754 prealloc_type |= XFS_BMAPI_CONVERT;
2755 xfs_tosspages(ip, startoffset, startoffset + bf->l_len, 0);
2756 /* FALLTHRU */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 case XFS_IOC_RESVSP:
2758 case XFS_IOC_RESVSP64:
2759 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
Dave Chinner44722352010-08-24 12:02:11 +10002760 prealloc_type, attr_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 if (error)
2762 return error;
2763 setprealloc = 1;
2764 break;
2765
2766 case XFS_IOC_UNRESVSP:
2767 case XFS_IOC_UNRESVSP64:
2768 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
2769 attr_flags)))
2770 return error;
2771 break;
2772
2773 case XFS_IOC_ALLOCSP:
2774 case XFS_IOC_ALLOCSP64:
2775 case XFS_IOC_FREESP:
2776 case XFS_IOC_FREESP64:
2777 if (startoffset > fsize) {
2778 error = xfs_alloc_file_space(ip, fsize,
2779 startoffset - fsize, 0, attr_flags);
2780 if (error)
2781 break;
2782 }
2783
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002784 iattr.ia_valid = ATTR_SIZE;
2785 iattr.ia_size = startoffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786
Christoph Hellwigea5a3dc82008-10-30 18:27:48 +11002787 error = xfs_setattr(ip, &iattr, attr_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
2789 if (error)
2790 return error;
2791
2792 clrprealloc = 1;
2793 break;
2794
2795 default:
2796 ASSERT(0);
2797 return XFS_ERROR(EINVAL);
2798 }
2799
2800 /*
2801 * update the inode timestamp, mode, and prealloc flag bits
2802 */
2803 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
2804
2805 if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
2806 0, 0, 0))) {
2807 /* ASSERT(0); */
2808 xfs_trans_cancel(tp, 0);
2809 return error;
2810 }
2811
2812 xfs_ilock(ip, XFS_ILOCK_EXCL);
2813
Christoph Hellwig898621d2010-06-24 11:36:58 +10002814 xfs_trans_ijoin(tp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
Christoph Hellwig0f285c82008-07-18 17:13:28 +10002816 if ((attr_flags & XFS_ATTR_DMI) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 ip->i_d.di_mode &= ~S_ISUID;
2818
2819 /*
2820 * Note that we don't have to worry about mandatory
2821 * file locking being disabled here because we only
2822 * clear the S_ISGID bit if the Group execute bit is
2823 * on, but if it was on then mandatory locking wouldn't
2824 * have been enabled.
2825 */
2826 if (ip->i_d.di_mode & S_IXGRP)
2827 ip->i_d.di_mode &= ~S_ISGID;
2828
Dave Chinnerdcd79a12010-09-28 12:27:25 +10002829 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 }
2831 if (setprealloc)
2832 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
2833 else if (clrprealloc)
2834 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
2835
2836 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Dave Chinner82878892011-03-26 09:13:08 +11002837 if (attr_flags & XFS_ATTR_SYNC)
2838 xfs_trans_set_sync(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839
Eric Sandeen1c72bf92007-05-08 13:48:42 +10002840 error = xfs_trans_commit(tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841
2842 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2843
2844 return error;
2845}