blob: ae3758a90ed63ba6b714ccbdafdf007207a8be07 [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 */
Robert P. J. Day40ebd812007-11-23 16:30:51 +110018#include <linux/log2.h>
19
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_fs.h"
Dave Chinner70a9883c2013-10-23 10:36:05 +110022#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110023#include "xfs_format.h"
24#include "xfs_log_format.h"
25#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_sb.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110028#include "xfs_inode.h"
Dave Chinner57062782013-10-15 09:17:51 +110029#include "xfs_da_format.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100030#include "xfs_da_btree.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100031#include "xfs_dir2.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_attr_sf.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100033#include "xfs_attr.h"
Dave Chinner239880e2013-10-23 10:50:10 +110034#include "xfs_trans_space.h"
35#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_ialloc.h"
39#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100040#include "xfs_bmap_util.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_quota.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100043#include "xfs_filestream.h"
Christoph Hellwig93848a92013-04-03 16:11:17 +110044#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000045#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110046#include "xfs_icache.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100047#include "xfs_symlink.h"
Dave Chinner239880e2013-10-23 10:50:10 +110048#include "xfs_trans_priv.h"
49#include "xfs_log.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110050#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Linus Torvalds1da177e2005-04-16 15:20:36 -070052kmem_zone_t *xfs_inode_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +020055 * Used in xfs_itruncate_extents(). This is the maximum number of extents
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 * freed from a file in a single transaction.
57 */
58#define XFS_ITRUNC_MAX_EXTENTS 2
59
60STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Zhi Yong Wuab297432013-12-18 08:22:41 +080062STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *);
63
Dave Chinner2a0ec1d2012-04-23 15:59:02 +100064/*
65 * helper function to extract extent size hint from inode
66 */
67xfs_extlen_t
68xfs_get_extsz_hint(
69 struct xfs_inode *ip)
70{
71 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
72 return ip->i_d.di_extsize;
73 if (XFS_IS_REALTIME_INODE(ip))
74 return ip->i_mount->m_sb.sb_rextsize;
75 return 0;
76}
77
Dave Chinnerfa96aca2012-10-08 21:56:10 +110078/*
Christoph Hellwigefa70be2013-12-18 02:14:39 -080079 * These two are wrapper routines around the xfs_ilock() routine used to
80 * centralize some grungy code. They are used in places that wish to lock the
81 * inode solely for reading the extents. The reason these places can't just
82 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
83 * bringing in of the extents from disk for a file in b-tree format. If the
84 * inode is in b-tree format, then we need to lock the inode exclusively until
85 * the extents are read in. Locking it exclusively all the time would limit
86 * our parallelism unnecessarily, though. What we do instead is check to see
87 * if the extents have been read in yet, and only lock the inode exclusively
88 * if they have not.
Dave Chinnerfa96aca2012-10-08 21:56:10 +110089 *
Christoph Hellwigefa70be2013-12-18 02:14:39 -080090 * The functions return a value which should be given to the corresponding
Christoph Hellwig01f4f322013-12-06 12:30:08 -080091 * xfs_iunlock() call.
Dave Chinnerfa96aca2012-10-08 21:56:10 +110092 */
93uint
Christoph Hellwig309ecac2013-12-06 12:30:09 -080094xfs_ilock_data_map_shared(
95 struct xfs_inode *ip)
Dave Chinnerfa96aca2012-10-08 21:56:10 +110096{
Christoph Hellwig309ecac2013-12-06 12:30:09 -080097 uint lock_mode = XFS_ILOCK_SHARED;
Dave Chinnerfa96aca2012-10-08 21:56:10 +110098
Christoph Hellwig309ecac2013-12-06 12:30:09 -080099 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE &&
100 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100101 lock_mode = XFS_ILOCK_EXCL;
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100102 xfs_ilock(ip, lock_mode);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100103 return lock_mode;
104}
105
Christoph Hellwigefa70be2013-12-18 02:14:39 -0800106uint
107xfs_ilock_attr_map_shared(
108 struct xfs_inode *ip)
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100109{
Christoph Hellwigefa70be2013-12-18 02:14:39 -0800110 uint lock_mode = XFS_ILOCK_SHARED;
111
112 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
113 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
114 lock_mode = XFS_ILOCK_EXCL;
115 xfs_ilock(ip, lock_mode);
116 return lock_mode;
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100117}
118
119/*
Dave Chinner653c60b2015-02-23 21:43:37 +1100120 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and
121 * the i_lock. This routine allows various combinations of the locks to be
122 * obtained.
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100123 *
Dave Chinner653c60b2015-02-23 21:43:37 +1100124 * The 3 locks should always be ordered so that the IO lock is obtained first,
125 * the mmap lock second and the ilock last in order to prevent deadlock.
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100126 *
Dave Chinner653c60b2015-02-23 21:43:37 +1100127 * Basic locking order:
128 *
129 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock
130 *
131 * mmap_sem locking order:
132 *
133 * i_iolock -> page lock -> mmap_sem
134 * mmap_sem -> i_mmap_lock -> page_lock
135 *
136 * The difference in mmap_sem locking order mean that we cannot hold the
137 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
138 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem
139 * in get_user_pages() to map the user pages into the kernel address space for
140 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because
141 * page faults already hold the mmap_sem.
142 *
143 * Hence to serialise fully against both syscall and mmap based IO, we need to
144 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both
145 * taken in places where we need to invalidate the page cache in a race
146 * free manner (e.g. truncate, hole punch and other extent manipulation
147 * functions).
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100148 */
149void
150xfs_ilock(
151 xfs_inode_t *ip,
152 uint lock_flags)
153{
154 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
155
156 /*
157 * You can't set both SHARED and EXCL for the same lock,
158 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
159 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
160 */
161 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
162 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
Dave Chinner653c60b2015-02-23 21:43:37 +1100163 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
164 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100165 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
166 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Dave Chinner0952c812015-08-19 10:32:49 +1000167 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100168
169 if (lock_flags & XFS_IOLOCK_EXCL)
170 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
171 else if (lock_flags & XFS_IOLOCK_SHARED)
172 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
173
Dave Chinner653c60b2015-02-23 21:43:37 +1100174 if (lock_flags & XFS_MMAPLOCK_EXCL)
175 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
176 else if (lock_flags & XFS_MMAPLOCK_SHARED)
177 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
178
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100179 if (lock_flags & XFS_ILOCK_EXCL)
180 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
181 else if (lock_flags & XFS_ILOCK_SHARED)
182 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
183}
184
185/*
186 * This is just like xfs_ilock(), except that the caller
187 * is guaranteed not to sleep. It returns 1 if it gets
188 * the requested locks and 0 otherwise. If the IO lock is
189 * obtained but the inode lock cannot be, then the IO lock
190 * is dropped before returning.
191 *
192 * ip -- the inode being locked
193 * lock_flags -- this parameter indicates the inode's locks to be
194 * to be locked. See the comment for xfs_ilock() for a list
195 * of valid values.
196 */
197int
198xfs_ilock_nowait(
199 xfs_inode_t *ip,
200 uint lock_flags)
201{
202 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
203
204 /*
205 * You can't set both SHARED and EXCL for the same lock,
206 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
207 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
208 */
209 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
210 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
Dave Chinner653c60b2015-02-23 21:43:37 +1100211 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
212 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100213 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
214 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Dave Chinner0952c812015-08-19 10:32:49 +1000215 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100216
217 if (lock_flags & XFS_IOLOCK_EXCL) {
218 if (!mrtryupdate(&ip->i_iolock))
219 goto out;
220 } else if (lock_flags & XFS_IOLOCK_SHARED) {
221 if (!mrtryaccess(&ip->i_iolock))
222 goto out;
223 }
Dave Chinner653c60b2015-02-23 21:43:37 +1100224
225 if (lock_flags & XFS_MMAPLOCK_EXCL) {
226 if (!mrtryupdate(&ip->i_mmaplock))
227 goto out_undo_iolock;
228 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
229 if (!mrtryaccess(&ip->i_mmaplock))
230 goto out_undo_iolock;
231 }
232
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100233 if (lock_flags & XFS_ILOCK_EXCL) {
234 if (!mrtryupdate(&ip->i_lock))
Dave Chinner653c60b2015-02-23 21:43:37 +1100235 goto out_undo_mmaplock;
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100236 } else if (lock_flags & XFS_ILOCK_SHARED) {
237 if (!mrtryaccess(&ip->i_lock))
Dave Chinner653c60b2015-02-23 21:43:37 +1100238 goto out_undo_mmaplock;
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100239 }
240 return 1;
241
Dave Chinner653c60b2015-02-23 21:43:37 +1100242out_undo_mmaplock:
243 if (lock_flags & XFS_MMAPLOCK_EXCL)
244 mrunlock_excl(&ip->i_mmaplock);
245 else if (lock_flags & XFS_MMAPLOCK_SHARED)
246 mrunlock_shared(&ip->i_mmaplock);
247out_undo_iolock:
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100248 if (lock_flags & XFS_IOLOCK_EXCL)
249 mrunlock_excl(&ip->i_iolock);
250 else if (lock_flags & XFS_IOLOCK_SHARED)
251 mrunlock_shared(&ip->i_iolock);
Dave Chinner653c60b2015-02-23 21:43:37 +1100252out:
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100253 return 0;
254}
255
256/*
257 * xfs_iunlock() is used to drop the inode locks acquired with
258 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
259 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
260 * that we know which locks to drop.
261 *
262 * ip -- the inode being unlocked
263 * lock_flags -- this parameter indicates the inode's locks to be
264 * to be unlocked. See the comment for xfs_ilock() for a list
265 * of valid values for this parameter.
266 *
267 */
268void
269xfs_iunlock(
270 xfs_inode_t *ip,
271 uint lock_flags)
272{
273 /*
274 * You can't set both SHARED and EXCL for the same lock,
275 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
276 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
277 */
278 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
279 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
Dave Chinner653c60b2015-02-23 21:43:37 +1100280 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
281 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100282 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
283 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
Dave Chinner0952c812015-08-19 10:32:49 +1000284 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100285 ASSERT(lock_flags != 0);
286
287 if (lock_flags & XFS_IOLOCK_EXCL)
288 mrunlock_excl(&ip->i_iolock);
289 else if (lock_flags & XFS_IOLOCK_SHARED)
290 mrunlock_shared(&ip->i_iolock);
291
Dave Chinner653c60b2015-02-23 21:43:37 +1100292 if (lock_flags & XFS_MMAPLOCK_EXCL)
293 mrunlock_excl(&ip->i_mmaplock);
294 else if (lock_flags & XFS_MMAPLOCK_SHARED)
295 mrunlock_shared(&ip->i_mmaplock);
296
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100297 if (lock_flags & XFS_ILOCK_EXCL)
298 mrunlock_excl(&ip->i_lock);
299 else if (lock_flags & XFS_ILOCK_SHARED)
300 mrunlock_shared(&ip->i_lock);
301
302 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
303}
304
305/*
306 * give up write locks. the i/o lock cannot be held nested
307 * if it is being demoted.
308 */
309void
310xfs_ilock_demote(
311 xfs_inode_t *ip,
312 uint lock_flags)
313{
Dave Chinner653c60b2015-02-23 21:43:37 +1100314 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
315 ASSERT((lock_flags &
316 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100317
318 if (lock_flags & XFS_ILOCK_EXCL)
319 mrdemote(&ip->i_lock);
Dave Chinner653c60b2015-02-23 21:43:37 +1100320 if (lock_flags & XFS_MMAPLOCK_EXCL)
321 mrdemote(&ip->i_mmaplock);
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100322 if (lock_flags & XFS_IOLOCK_EXCL)
323 mrdemote(&ip->i_iolock);
324
325 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
326}
327
Dave Chinner742ae1e2013-04-30 21:39:34 +1000328#if defined(DEBUG) || defined(XFS_WARN)
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100329int
330xfs_isilocked(
331 xfs_inode_t *ip,
332 uint lock_flags)
333{
334 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
335 if (!(lock_flags & XFS_ILOCK_SHARED))
336 return !!ip->i_lock.mr_writer;
337 return rwsem_is_locked(&ip->i_lock.mr_lock);
338 }
339
Dave Chinner653c60b2015-02-23 21:43:37 +1100340 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
341 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
342 return !!ip->i_mmaplock.mr_writer;
343 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
344 }
345
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100346 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
347 if (!(lock_flags & XFS_IOLOCK_SHARED))
348 return !!ip->i_iolock.mr_writer;
349 return rwsem_is_locked(&ip->i_iolock.mr_lock);
350 }
351
352 ASSERT(0);
353 return 0;
354}
355#endif
356
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000357#ifdef DEBUG
358int xfs_locked_n;
359int xfs_small_retries;
360int xfs_middle_retries;
361int xfs_lots_retries;
362int xfs_lock_delays;
363#endif
364
Dave Chinnerb6a99472015-08-25 10:05:13 +1000365/*
366 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
367 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
368 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
369 * errors and warnings.
370 */
371#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
Dave Chinner3403ccc2015-08-20 09:27:49 +1000372static bool
373xfs_lockdep_subclass_ok(
374 int subclass)
375{
376 return subclass < MAX_LOCKDEP_SUBCLASSES;
377}
378#else
379#define xfs_lockdep_subclass_ok(subclass) (true)
380#endif
381
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000382/*
Dave Chinner653c60b2015-02-23 21:43:37 +1100383 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
Dave Chinner0952c812015-08-19 10:32:49 +1000384 * value. This can be called for any type of inode lock combination, including
385 * parent locking. Care must be taken to ensure we don't overrun the subclass
386 * storage fields in the class mask we build.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000387 */
388static inline int
389xfs_lock_inumorder(int lock_mode, int subclass)
390{
Dave Chinner0952c812015-08-19 10:32:49 +1000391 int class = 0;
392
393 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
394 XFS_ILOCK_RTSUM)));
Dave Chinner3403ccc2015-08-20 09:27:49 +1000395 ASSERT(xfs_lockdep_subclass_ok(subclass));
Dave Chinner0952c812015-08-19 10:32:49 +1000396
Dave Chinner653c60b2015-02-23 21:43:37 +1100397 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
Dave Chinner0952c812015-08-19 10:32:49 +1000398 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
Dave Chinner3403ccc2015-08-20 09:27:49 +1000399 ASSERT(xfs_lockdep_subclass_ok(subclass +
400 XFS_IOLOCK_PARENT_VAL));
Dave Chinner0952c812015-08-19 10:32:49 +1000401 class += subclass << XFS_IOLOCK_SHIFT;
402 if (lock_mode & XFS_IOLOCK_PARENT)
403 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT;
Dave Chinner653c60b2015-02-23 21:43:37 +1100404 }
405
406 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
Dave Chinner0952c812015-08-19 10:32:49 +1000407 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
408 class += subclass << XFS_MMAPLOCK_SHIFT;
Dave Chinner653c60b2015-02-23 21:43:37 +1100409 }
410
Dave Chinner0952c812015-08-19 10:32:49 +1000411 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
412 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
413 class += subclass << XFS_ILOCK_SHIFT;
414 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000415
Dave Chinner0952c812015-08-19 10:32:49 +1000416 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000417}
418
419/*
Dave Chinner95afcf52015-03-25 14:03:32 +1100420 * The following routine will lock n inodes in exclusive mode. We assume the
421 * caller calls us with the inodes in i_ino order.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000422 *
Dave Chinner95afcf52015-03-25 14:03:32 +1100423 * We need to detect deadlock where an inode that we lock is in the AIL and we
424 * start waiting for another inode that is locked by a thread in a long running
425 * transaction (such as truncate). This can result in deadlock since the long
426 * running trans might need to wait for the inode we just locked in order to
427 * push the tail and free space in the log.
Dave Chinner0952c812015-08-19 10:32:49 +1000428 *
429 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
430 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
431 * lock more than one at a time, lockdep will report false positives saying we
432 * have violated locking orders.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000433 */
434void
435xfs_lock_inodes(
436 xfs_inode_t **ips,
437 int inodes,
438 uint lock_mode)
439{
440 int attempts = 0, i, j, try_lock;
441 xfs_log_item_t *lp;
442
Dave Chinner0952c812015-08-19 10:32:49 +1000443 /*
444 * Currently supports between 2 and 5 inodes with exclusive locking. We
445 * support an arbitrary depth of locking here, but absolute limits on
446 * inodes depend on the the type of locking and the limits placed by
447 * lockdep annotations in xfs_lock_inumorder. These are all checked by
448 * the asserts.
449 */
Dave Chinner95afcf52015-03-25 14:03:32 +1100450 ASSERT(ips && inodes >= 2 && inodes <= 5);
Dave Chinner0952c812015-08-19 10:32:49 +1000451 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
452 XFS_ILOCK_EXCL));
453 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
454 XFS_ILOCK_SHARED)));
455 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) ||
456 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1);
457 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
458 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
459 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
460 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
461
462 if (lock_mode & XFS_IOLOCK_EXCL) {
463 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
464 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
465 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000466
467 try_lock = 0;
468 i = 0;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000469again:
470 for (; i < inodes; i++) {
471 ASSERT(ips[i]);
472
Dave Chinner95afcf52015-03-25 14:03:32 +1100473 if (i && (ips[i] == ips[i - 1])) /* Already locked */
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000474 continue;
475
476 /*
Dave Chinner95afcf52015-03-25 14:03:32 +1100477 * If try_lock is not set yet, make sure all locked inodes are
478 * not in the AIL. If any are, set try_lock to be used later.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000479 */
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000480 if (!try_lock) {
481 for (j = (i - 1); j >= 0 && !try_lock; j--) {
482 lp = (xfs_log_item_t *)ips[j]->i_itemp;
Dave Chinner95afcf52015-03-25 14:03:32 +1100483 if (lp && (lp->li_flags & XFS_LI_IN_AIL))
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000484 try_lock++;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000485 }
486 }
487
488 /*
489 * If any of the previous locks we have locked is in the AIL,
490 * we must TRY to get the second and subsequent locks. If
491 * we can't get any, we must release all we have
492 * and try again.
493 */
Dave Chinner95afcf52015-03-25 14:03:32 +1100494 if (!try_lock) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000495 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
Dave Chinner95afcf52015-03-25 14:03:32 +1100496 continue;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000497 }
Dave Chinner95afcf52015-03-25 14:03:32 +1100498
499 /* try_lock means we have an inode locked that is in the AIL. */
500 ASSERT(i != 0);
501 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
502 continue;
503
504 /*
505 * Unlock all previous guys and try again. xfs_iunlock will try
506 * to push the tail if the inode is in the AIL.
507 */
508 attempts++;
509 for (j = i - 1; j >= 0; j--) {
510 /*
511 * Check to see if we've already unlocked this one. Not
512 * the first one going back, and the inode ptr is the
513 * same.
514 */
515 if (j != (i - 1) && ips[j] == ips[j + 1])
516 continue;
517
518 xfs_iunlock(ips[j], lock_mode);
519 }
520
521 if ((attempts % 5) == 0) {
522 delay(1); /* Don't just spin the CPU */
523#ifdef DEBUG
524 xfs_lock_delays++;
525#endif
526 }
527 i = 0;
528 try_lock = 0;
529 goto again;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000530 }
531
532#ifdef DEBUG
533 if (attempts) {
534 if (attempts < 5) xfs_small_retries++;
535 else if (attempts < 100) xfs_middle_retries++;
536 else xfs_lots_retries++;
537 } else {
538 xfs_locked_n++;
539 }
540#endif
541}
542
543/*
Dave Chinner653c60b2015-02-23 21:43:37 +1100544 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
545 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
546 * lock more than one at a time, lockdep will report false positives saying we
547 * have violated locking orders.
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000548 */
549void
550xfs_lock_two_inodes(
551 xfs_inode_t *ip0,
552 xfs_inode_t *ip1,
553 uint lock_mode)
554{
555 xfs_inode_t *temp;
556 int attempts = 0;
557 xfs_log_item_t *lp;
558
Dave Chinner653c60b2015-02-23 21:43:37 +1100559 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
560 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
561 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
562 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))
563 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
564
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000565 ASSERT(ip0->i_ino != ip1->i_ino);
566
567 if (ip0->i_ino > ip1->i_ino) {
568 temp = ip0;
569 ip0 = ip1;
570 ip1 = temp;
571 }
572
573 again:
574 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
575
576 /*
577 * If the first lock we have locked is in the AIL, we must TRY to get
578 * the second lock. If we can't get it, we must release the first one
579 * and try again.
580 */
581 lp = (xfs_log_item_t *)ip0->i_itemp;
582 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
583 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
584 xfs_iunlock(ip0, lock_mode);
585 if ((++attempts % 5) == 0)
586 delay(1); /* Don't just spin the CPU */
587 goto again;
588 }
589 } else {
590 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
591 }
592}
593
594
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100595void
596__xfs_iflock(
597 struct xfs_inode *ip)
598{
599 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
600 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
601
602 do {
603 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
604 if (xfs_isiflocked(ip))
605 io_schedule();
606 } while (!xfs_iflock_nowait(ip));
607
608 finish_wait(wq, &wait.wait);
609}
610
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611STATIC uint
612_xfs_dic2xflags(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 __uint16_t di_flags)
614{
615 uint flags = 0;
616
617 if (di_flags & XFS_DIFLAG_ANY) {
618 if (di_flags & XFS_DIFLAG_REALTIME)
619 flags |= XFS_XFLAG_REALTIME;
620 if (di_flags & XFS_DIFLAG_PREALLOC)
621 flags |= XFS_XFLAG_PREALLOC;
622 if (di_flags & XFS_DIFLAG_IMMUTABLE)
623 flags |= XFS_XFLAG_IMMUTABLE;
624 if (di_flags & XFS_DIFLAG_APPEND)
625 flags |= XFS_XFLAG_APPEND;
626 if (di_flags & XFS_DIFLAG_SYNC)
627 flags |= XFS_XFLAG_SYNC;
628 if (di_flags & XFS_DIFLAG_NOATIME)
629 flags |= XFS_XFLAG_NOATIME;
630 if (di_flags & XFS_DIFLAG_NODUMP)
631 flags |= XFS_XFLAG_NODUMP;
632 if (di_flags & XFS_DIFLAG_RTINHERIT)
633 flags |= XFS_XFLAG_RTINHERIT;
634 if (di_flags & XFS_DIFLAG_PROJINHERIT)
635 flags |= XFS_XFLAG_PROJINHERIT;
636 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
637 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100638 if (di_flags & XFS_DIFLAG_EXTSIZE)
639 flags |= XFS_XFLAG_EXTSIZE;
640 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
641 flags |= XFS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000642 if (di_flags & XFS_DIFLAG_NODEFRAG)
643 flags |= XFS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000644 if (di_flags & XFS_DIFLAG_FILESTREAM)
645 flags |= XFS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 }
647
648 return flags;
649}
650
651uint
652xfs_ip2xflags(
653 xfs_inode_t *ip)
654{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000655 xfs_icdinode_t *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Nathan Scotta916e2b2006-06-09 17:12:17 +1000657 return _xfs_dic2xflags(dic->di_flags) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100658 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}
660
661uint
662xfs_dic2xflags(
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100663 xfs_dinode_t *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100665 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100666 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669/*
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000670 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
671 * is allowed, otherwise it has to be an exact match. If a CI match is found,
672 * ci_name->name will point to a the actual name (caller must free) or
673 * will be set to NULL if an exact match is found.
674 */
675int
676xfs_lookup(
677 xfs_inode_t *dp,
678 struct xfs_name *name,
679 xfs_inode_t **ipp,
680 struct xfs_name *ci_name)
681{
682 xfs_ino_t inum;
683 int error;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000684
685 trace_xfs_lookup(dp, name);
686
687 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
Dave Chinner24513372014-06-25 14:58:08 +1000688 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000689
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000690 xfs_ilock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000691 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000692 if (error)
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000693 goto out_unlock;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000694
695 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
696 if (error)
697 goto out_free_name;
698
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000699 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000700 return 0;
701
702out_free_name:
703 if (ci_name)
704 kmem_free(ci_name->name);
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000705out_unlock:
706 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000707 *ipp = NULL;
708 return error;
709}
710
711/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 * Allocate an inode on disk and return a copy of its in-core version.
713 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
714 * appropriately within the inode. The uid and gid for the inode are
715 * set according to the contents of the given cred structure.
716 *
717 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300718 * has a free inode available, call xfs_iget() to obtain the in-core
719 * version of the allocated inode. Finally, fill in the inode and
720 * log its initial contents. In this case, ialloc_context would be
721 * set to NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 *
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300723 * If xfs_dialloc() does not have an available inode, it will replenish
724 * its supply by doing an allocation. Since we can only do one
725 * allocation within a transaction without deadlocks, we must commit
726 * the current transaction before returning the inode itself.
727 * In this case, therefore, we will set ialloc_context and return.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 * The caller should then commit the current transaction, start a new
729 * transaction, and call xfs_ialloc() again to actually get the inode.
730 *
731 * To ensure that some other process does not grab the inode that
732 * was allocated during the first call to xfs_ialloc(), this routine
733 * also returns the [locked] bp pointing to the head of the freelist
734 * as ialloc_context. The caller should hold this buffer across
735 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +1000736 *
737 * If we are allocating quota inodes, we do not have a parent inode
738 * to attach to or associate with (i.e. pip == NULL) because they
739 * are not linked into the directory structure - they are attached
740 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 */
742int
743xfs_ialloc(
744 xfs_trans_t *tp,
745 xfs_inode_t *pip,
Al Viro576b1d62011-07-26 02:50:15 -0400746 umode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700747 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 xfs_dev_t rdev,
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000749 prid_t prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 int okalloc,
751 xfs_buf_t **ialloc_context,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 xfs_inode_t **ipp)
753{
Christoph Hellwig93848a92013-04-03 16:11:17 +1100754 struct xfs_mount *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 xfs_ino_t ino;
756 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 uint flags;
758 int error;
Dave Chinnere076b0f2014-10-02 09:18:13 +1000759 struct timespec tv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
761 /*
762 * Call the space management code to pick
763 * the on-disk inode to be allocated.
764 */
David Chinnerb11f94d2007-07-11 11:09:33 +1000765 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Christoph Hellwig08358902012-07-04 10:54:47 -0400766 ialloc_context, &ino);
David Chinnerbf904242008-10-30 17:36:14 +1100767 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 return error;
Christoph Hellwig08358902012-07-04 10:54:47 -0400769 if (*ialloc_context || ino == NULLFSINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 *ipp = NULL;
771 return 0;
772 }
773 ASSERT(*ialloc_context == NULL);
774
775 /*
776 * Get the in-core inode with the lock held exclusively.
777 * This is because we're setting fields here we need
778 * to prevent others from looking at until we're done.
779 */
Christoph Hellwig93848a92013-04-03 16:11:17 +1100780 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
Christoph Hellwigec3ba852011-02-13 13:26:42 +0000781 XFS_ILOCK_EXCL, &ip);
David Chinnerbf904242008-10-30 17:36:14 +1100782 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 ASSERT(ip != NULL);
785
Dave Chinner263997a2014-05-20 07:46:40 +1000786 /*
787 * We always convert v1 inodes to v2 now - we only support filesystems
788 * with >= v2 inode capability, so there is no reason for ever leaving
789 * an inode in v1 format.
790 */
791 if (ip->i_d.di_version == 1)
792 ip->i_d.di_version = 2;
793
Al Viro576b1d62011-07-26 02:50:15 -0400794 ip->i_d.di_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 ip->i_d.di_onlink = 0;
796 ip->i_d.di_nlink = nlink;
797 ASSERT(ip->i_d.di_nlink == nlink);
Dwight Engen7aab1b22013-08-15 14:08:01 -0400798 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
799 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000800 xfs_set_projid(ip, prid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
802
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000803 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 ip->i_d.di_gid = pip->i_d.di_gid;
Al Viroabbede12011-07-26 02:31:30 -0400805 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 ip->i_d.di_mode |= S_ISGID;
807 }
808 }
809
810 /*
811 * If the group ID of the new file does not match the effective group
812 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
813 * (and only if the irix_sgid_inherit compatibility variable is set).
814 */
815 if ((irix_sgid_inherit) &&
816 (ip->i_d.di_mode & S_ISGID) &&
Dwight Engen7aab1b22013-08-15 14:08:01 -0400817 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 ip->i_d.di_mode &= ~S_ISGID;
819 }
820
821 ip->i_d.di_size = 0;
822 ip->i_d.di_nextents = 0;
823 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000824
Dave Chinnere076b0f2014-10-02 09:18:13 +1000825 tv = current_fs_time(mp->m_super);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000826 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
827 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
828 ip->i_d.di_atime = ip->i_d.di_mtime;
829 ip->i_d.di_ctime = ip->i_d.di_mtime;
830
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 /*
832 * di_gen will have been taken care of in xfs_iread.
833 */
834 ip->i_d.di_extsize = 0;
835 ip->i_d.di_dmevmask = 0;
836 ip->i_d.di_dmstate = 0;
837 ip->i_d.di_flags = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100838
839 if (ip->i_d.di_version == 3) {
840 ASSERT(ip->i_d.di_ino == ino);
Dave Chinnerbbf155a2015-08-19 10:31:18 +1000841 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_meta_uuid));
Christoph Hellwig93848a92013-04-03 16:11:17 +1100842 ip->i_d.di_crc = 0;
843 ip->i_d.di_changecount = 1;
844 ip->i_d.di_lsn = 0;
845 ip->i_d.di_flags2 = 0;
846 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
847 ip->i_d.di_crtime = ip->i_d.di_mtime;
848 }
849
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 flags = XFS_ILOG_CORE;
852 switch (mode & S_IFMT) {
853 case S_IFIFO:
854 case S_IFCHR:
855 case S_IFBLK:
856 case S_IFSOCK:
857 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
858 ip->i_df.if_u2.if_rdev = rdev;
859 ip->i_df.if_flags = 0;
860 flags |= XFS_ILOG_DEV;
861 break;
862 case S_IFREG:
863 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +1000864 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +1000865 uint di_flags = 0;
866
Al Viroabbede12011-07-26 02:31:30 -0400867 if (S_ISDIR(mode)) {
Nathan Scott365ca832005-06-21 15:39:12 +1000868 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
869 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100870 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
871 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
872 ip->i_d.di_extsize = pip->i_d.di_extsize;
873 }
Dave Chinner9336e3a2014-10-02 09:18:40 +1000874 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
875 di_flags |= XFS_DIFLAG_PROJINHERIT;
Al Viroabbede12011-07-26 02:31:30 -0400876 } else if (S_ISREG(mode)) {
Christoph Hellwig613d7042007-10-11 17:44:08 +1000877 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +1000878 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100879 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
880 di_flags |= XFS_DIFLAG_EXTSIZE;
881 ip->i_d.di_extsize = pip->i_d.di_extsize;
882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 }
884 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
885 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +1000886 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
888 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +1000889 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
891 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +1000892 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
894 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +1000895 di_flags |= XFS_DIFLAG_NOSYMLINKS;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000896 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
897 xfs_inherit_nodefrag)
898 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000899 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
900 di_flags |= XFS_DIFLAG_FILESTREAM;
Nathan Scott365ca832005-06-21 15:39:12 +1000901 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
903 /* FALLTHROUGH */
904 case S_IFLNK:
905 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
906 ip->i_df.if_flags = XFS_IFEXTENTS;
907 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
908 ip->i_df.if_u1.if_extents = NULL;
909 break;
910 default:
911 ASSERT(0);
912 }
913 /*
914 * Attribute fork settings for new inode.
915 */
916 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
917 ip->i_d.di_anextents = 0;
918
919 /*
920 * Log the new values stuffed into the inode.
921 */
Christoph Hellwigddc34152011-09-19 15:00:54 +0000922 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 xfs_trans_log_inode(tp, ip, flags);
924
Dave Chinner58c90472015-02-23 22:38:08 +1100925 /* now that we have an i_mode we can setup the inode structure */
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000926 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
928 *ipp = ip;
929 return 0;
930}
931
Dave Chinnere546cb72013-08-12 20:49:47 +1000932/*
933 * Allocates a new inode from disk and return a pointer to the
934 * incore copy. This routine will internally commit the current
935 * transaction and allocate a new one if the Space Manager needed
936 * to do an allocation to replenish the inode free-list.
937 *
938 * This routine is designed to be called from xfs_create and
939 * xfs_create_dir.
940 *
941 */
942int
943xfs_dir_ialloc(
944 xfs_trans_t **tpp, /* input: current transaction;
945 output: may be a new transaction. */
946 xfs_inode_t *dp, /* directory within whose allocate
947 the inode. */
948 umode_t mode,
949 xfs_nlink_t nlink,
950 xfs_dev_t rdev,
951 prid_t prid, /* project id */
952 int okalloc, /* ok to allocate new space */
953 xfs_inode_t **ipp, /* pointer to inode; it will be
954 locked. */
955 int *committed)
956
957{
958 xfs_trans_t *tp;
Dave Chinnere546cb72013-08-12 20:49:47 +1000959 xfs_inode_t *ip;
960 xfs_buf_t *ialloc_context = NULL;
961 int code;
Dave Chinnere546cb72013-08-12 20:49:47 +1000962 void *dqinfo;
963 uint tflags;
964
965 tp = *tpp;
966 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
967
968 /*
969 * xfs_ialloc will return a pointer to an incore inode if
970 * the Space Manager has an available inode on the free
971 * list. Otherwise, it will do an allocation and replenish
972 * the freelist. Since we can only do one allocation per
973 * transaction without deadlocks, we will need to commit the
974 * current transaction and start a new one. We will then
975 * need to call xfs_ialloc again to get the inode.
976 *
977 * If xfs_ialloc did an allocation to replenish the freelist,
978 * it returns the bp containing the head of the freelist as
979 * ialloc_context. We will hold a lock on it across the
980 * transaction commit so that no other process can steal
981 * the inode(s) that we've just allocated.
982 */
983 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
984 &ialloc_context, &ip);
985
986 /*
987 * Return an error if we were unable to allocate a new inode.
988 * This should only happen if we run out of space on disk or
989 * encounter a disk error.
990 */
991 if (code) {
992 *ipp = NULL;
993 return code;
994 }
995 if (!ialloc_context && !ip) {
996 *ipp = NULL;
Dave Chinner24513372014-06-25 14:58:08 +1000997 return -ENOSPC;
Dave Chinnere546cb72013-08-12 20:49:47 +1000998 }
999
1000 /*
1001 * If the AGI buffer is non-NULL, then we were unable to get an
1002 * inode in one operation. We need to commit the current
1003 * transaction and call xfs_ialloc() again. It is guaranteed
1004 * to succeed the second time.
1005 */
1006 if (ialloc_context) {
1007 /*
1008 * Normally, xfs_trans_commit releases all the locks.
1009 * We call bhold to hang on to the ialloc_context across
1010 * the commit. Holding this buffer prevents any other
1011 * processes from doing any allocations in this
1012 * allocation group.
1013 */
1014 xfs_trans_bhold(tp, ialloc_context);
Dave Chinnere546cb72013-08-12 20:49:47 +10001015
1016 /*
1017 * We want the quota changes to be associated with the next
1018 * transaction, NOT this one. So, detach the dqinfo from this
1019 * and attach it to the next transaction.
1020 */
1021 dqinfo = NULL;
1022 tflags = 0;
1023 if (tp->t_dqinfo) {
1024 dqinfo = (void *)tp->t_dqinfo;
1025 tp->t_dqinfo = NULL;
1026 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1027 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1028 }
1029
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001030 code = xfs_trans_roll(&tp, 0);
1031 if (committed != NULL)
Dave Chinnere546cb72013-08-12 20:49:47 +10001032 *committed = 1;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001033
Dave Chinnere546cb72013-08-12 20:49:47 +10001034 /*
1035 * Re-attach the quota info that we detached from prev trx.
1036 */
1037 if (dqinfo) {
1038 tp->t_dqinfo = dqinfo;
1039 tp->t_flags |= tflags;
1040 }
1041
1042 if (code) {
1043 xfs_buf_relse(ialloc_context);
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001044 *tpp = tp;
Dave Chinnere546cb72013-08-12 20:49:47 +10001045 *ipp = NULL;
1046 return code;
1047 }
1048 xfs_trans_bjoin(tp, ialloc_context);
1049
1050 /*
1051 * Call ialloc again. Since we've locked out all
1052 * other allocations in this allocation group,
1053 * this call should always succeed.
1054 */
1055 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1056 okalloc, &ialloc_context, &ip);
1057
1058 /*
1059 * If we get an error at this point, return to the caller
1060 * so that the current transaction can be aborted.
1061 */
1062 if (code) {
1063 *tpp = tp;
1064 *ipp = NULL;
1065 return code;
1066 }
1067 ASSERT(!ialloc_context && ip);
1068
1069 } else {
1070 if (committed != NULL)
1071 *committed = 0;
1072 }
1073
1074 *ipp = ip;
1075 *tpp = tp;
1076
1077 return 0;
1078}
1079
1080/*
1081 * Decrement the link count on an inode & log the change.
1082 * If this causes the link count to go to zero, initiate the
1083 * logging activity required to truncate a file.
1084 */
1085int /* error */
1086xfs_droplink(
1087 xfs_trans_t *tp,
1088 xfs_inode_t *ip)
1089{
1090 int error;
1091
1092 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1093
1094 ASSERT (ip->i_d.di_nlink > 0);
1095 ip->i_d.di_nlink--;
1096 drop_nlink(VFS_I(ip));
1097 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1098
1099 error = 0;
1100 if (ip->i_d.di_nlink == 0) {
1101 /*
1102 * We're dropping the last link to this file.
1103 * Move the on-disk inode to the AGI unlinked list.
1104 * From xfs_inactive() we will pull the inode from
1105 * the list and free it.
1106 */
1107 error = xfs_iunlink(tp, ip);
1108 }
1109 return error;
1110}
1111
1112/*
Dave Chinnere546cb72013-08-12 20:49:47 +10001113 * Increment the link count on an inode & log the change.
1114 */
1115int
1116xfs_bumplink(
1117 xfs_trans_t *tp,
1118 xfs_inode_t *ip)
1119{
1120 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1121
Dave Chinner263997a2014-05-20 07:46:40 +10001122 ASSERT(ip->i_d.di_version > 1);
Zhi Yong Wuab297432013-12-18 08:22:41 +08001123 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
Dave Chinnere546cb72013-08-12 20:49:47 +10001124 ip->i_d.di_nlink++;
1125 inc_nlink(VFS_I(ip));
Dave Chinnere546cb72013-08-12 20:49:47 +10001126 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1127 return 0;
1128}
1129
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001130int
1131xfs_create(
1132 xfs_inode_t *dp,
1133 struct xfs_name *name,
1134 umode_t mode,
1135 xfs_dev_t rdev,
1136 xfs_inode_t **ipp)
1137{
1138 int is_dir = S_ISDIR(mode);
1139 struct xfs_mount *mp = dp->i_mount;
1140 struct xfs_inode *ip = NULL;
1141 struct xfs_trans *tp = NULL;
1142 int error;
1143 xfs_bmap_free_t free_list;
1144 xfs_fsblock_t first_block;
1145 bool unlock_dp_on_error = false;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001146 prid_t prid;
1147 struct xfs_dquot *udqp = NULL;
1148 struct xfs_dquot *gdqp = NULL;
1149 struct xfs_dquot *pdqp = NULL;
Brian Foster062647a2014-11-28 14:00:16 +11001150 struct xfs_trans_res *tres;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001151 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001152
1153 trace_xfs_create(dp, name);
1154
1155 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001156 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001157
Zhi Yong Wu163467d2013-12-18 08:22:39 +08001158 prid = xfs_get_initial_prid(dp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001159
1160 /*
1161 * Make sure that we have allocated dquot(s) on disk.
1162 */
Dwight Engen7aab1b22013-08-15 14:08:01 -04001163 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1164 xfs_kgid_to_gid(current_fsgid()), prid,
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001165 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1166 &udqp, &gdqp, &pdqp);
1167 if (error)
1168 return error;
1169
1170 if (is_dir) {
1171 rdev = 0;
1172 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
Brian Foster062647a2014-11-28 14:00:16 +11001173 tres = &M_RES(mp)->tr_mkdir;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001174 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1175 } else {
1176 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
Brian Foster062647a2014-11-28 14:00:16 +11001177 tres = &M_RES(mp)->tr_create;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001178 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1179 }
1180
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001181 /*
1182 * Initially assume that the file does not exist and
1183 * reserve the resources for that case. If that is not
1184 * the case we'll drop the one we have and get a more
1185 * appropriate transaction later.
1186 */
Brian Foster062647a2014-11-28 14:00:16 +11001187 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001188 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001189 /* flush outstanding delalloc blocks and retry */
1190 xfs_flush_inodes(mp);
Brian Foster062647a2014-11-28 14:00:16 +11001191 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001192 }
Dave Chinner24513372014-06-25 14:58:08 +10001193 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001194 /* No space at all so try a "no-allocation" reservation */
1195 resblks = 0;
Brian Foster062647a2014-11-28 14:00:16 +11001196 error = xfs_trans_reserve(tp, tres, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001197 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001198 if (error)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001199 goto out_trans_cancel;
Christoph Hellwig4906e212015-06-04 13:47:56 +10001200
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001201
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001202 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1203 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001204 unlock_dp_on_error = true;
1205
1206 xfs_bmap_init(&free_list, &first_block);
1207
1208 /*
1209 * Reserve disk quota and the inode.
1210 */
1211 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1212 pdqp, resblks, 1, 0);
1213 if (error)
1214 goto out_trans_cancel;
1215
Eric Sandeen94f3cad2014-09-09 11:57:52 +10001216 if (!resblks) {
1217 error = xfs_dir_canenter(tp, dp, name);
1218 if (error)
1219 goto out_trans_cancel;
1220 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001221
1222 /*
1223 * A newly created regular or special file just has one directory
1224 * entry pointing to them, but a directory also the "." entry
1225 * pointing to itself.
1226 */
1227 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001228 prid, resblks > 0, &ip, NULL);
Jan Karad6077aa2015-07-29 11:52:08 +10001229 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001230 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001231
1232 /*
1233 * Now we join the directory inode to the transaction. We do not do it
1234 * earlier because xfs_dir_ialloc might commit the previous transaction
1235 * (and release all the locks). An error from here on will result in
1236 * the transaction cancel unlocking dp so don't do it explicitly in the
1237 * error path.
1238 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001239 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001240 unlock_dp_on_error = false;
1241
1242 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1243 &first_block, &free_list, resblks ?
1244 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1245 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10001246 ASSERT(error != -ENOSPC);
Christoph Hellwig4906e212015-06-04 13:47:56 +10001247 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001248 }
1249 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1250 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1251
1252 if (is_dir) {
1253 error = xfs_dir_init(tp, ip, dp);
1254 if (error)
1255 goto out_bmap_cancel;
1256
1257 error = xfs_bumplink(tp, dp);
1258 if (error)
1259 goto out_bmap_cancel;
1260 }
1261
1262 /*
1263 * If this is a synchronous mount, make sure that the
1264 * create transaction goes to disk before returning to
1265 * the user.
1266 */
1267 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1268 xfs_trans_set_sync(tp);
1269
1270 /*
1271 * Attach the dquot(s) to the inodes and modify them incore.
1272 * These ids of the inode couldn't have changed since the new
1273 * inode has been locked ever since it was created.
1274 */
1275 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1276
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001277 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001278 if (error)
1279 goto out_bmap_cancel;
1280
Christoph Hellwig70393312015-06-04 13:48:08 +10001281 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001282 if (error)
1283 goto out_release_inode;
1284
1285 xfs_qm_dqrele(udqp);
1286 xfs_qm_dqrele(gdqp);
1287 xfs_qm_dqrele(pdqp);
1288
1289 *ipp = ip;
1290 return 0;
1291
1292 out_bmap_cancel:
1293 xfs_bmap_cancel(&free_list);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001294 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001295 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001296 out_release_inode:
1297 /*
Dave Chinner58c90472015-02-23 22:38:08 +11001298 * Wait until after the current transaction is aborted to finish the
1299 * setup of the inode and release the inode. This prevents recursive
1300 * transactions and deadlocks from xfs_inactive.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001301 */
Dave Chinner58c90472015-02-23 22:38:08 +11001302 if (ip) {
1303 xfs_finish_inode_setup(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001304 IRELE(ip);
Dave Chinner58c90472015-02-23 22:38:08 +11001305 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001306
1307 xfs_qm_dqrele(udqp);
1308 xfs_qm_dqrele(gdqp);
1309 xfs_qm_dqrele(pdqp);
1310
1311 if (unlock_dp_on_error)
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001312 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001313 return error;
1314}
1315
1316int
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001317xfs_create_tmpfile(
1318 struct xfs_inode *dp,
1319 struct dentry *dentry,
Brian Foster330033d2014-04-17 08:15:30 +10001320 umode_t mode,
1321 struct xfs_inode **ipp)
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001322{
1323 struct xfs_mount *mp = dp->i_mount;
1324 struct xfs_inode *ip = NULL;
1325 struct xfs_trans *tp = NULL;
1326 int error;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001327 prid_t prid;
1328 struct xfs_dquot *udqp = NULL;
1329 struct xfs_dquot *gdqp = NULL;
1330 struct xfs_dquot *pdqp = NULL;
1331 struct xfs_trans_res *tres;
1332 uint resblks;
1333
1334 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001335 return -EIO;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001336
1337 prid = xfs_get_initial_prid(dp);
1338
1339 /*
1340 * Make sure that we have allocated dquot(s) on disk.
1341 */
1342 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1343 xfs_kgid_to_gid(current_fsgid()), prid,
1344 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1345 &udqp, &gdqp, &pdqp);
1346 if (error)
1347 return error;
1348
1349 resblks = XFS_IALLOC_SPACE_RES(mp);
1350 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1351
1352 tres = &M_RES(mp)->tr_create_tmpfile;
1353 error = xfs_trans_reserve(tp, tres, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001354 if (error == -ENOSPC) {
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001355 /* No space at all so try a "no-allocation" reservation */
1356 resblks = 0;
1357 error = xfs_trans_reserve(tp, tres, 0, 0);
1358 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001359 if (error)
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001360 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001361
1362 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1363 pdqp, resblks, 1, 0);
1364 if (error)
1365 goto out_trans_cancel;
1366
1367 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1368 prid, resblks > 0, &ip, NULL);
Jan Karad6077aa2015-07-29 11:52:08 +10001369 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001370 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001371
1372 if (mp->m_flags & XFS_MOUNT_WSYNC)
1373 xfs_trans_set_sync(tp);
1374
1375 /*
1376 * Attach the dquot(s) to the inodes and modify them incore.
1377 * These ids of the inode couldn't have changed since the new
1378 * inode has been locked ever since it was created.
1379 */
1380 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1381
1382 ip->i_d.di_nlink--;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001383 error = xfs_iunlink(tp, ip);
1384 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001385 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001386
Christoph Hellwig70393312015-06-04 13:48:08 +10001387 error = xfs_trans_commit(tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001388 if (error)
1389 goto out_release_inode;
1390
1391 xfs_qm_dqrele(udqp);
1392 xfs_qm_dqrele(gdqp);
1393 xfs_qm_dqrele(pdqp);
1394
Brian Foster330033d2014-04-17 08:15:30 +10001395 *ipp = ip;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001396 return 0;
1397
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001398 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001399 xfs_trans_cancel(tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001400 out_release_inode:
1401 /*
Dave Chinner58c90472015-02-23 22:38:08 +11001402 * Wait until after the current transaction is aborted to finish the
1403 * setup of the inode and release the inode. This prevents recursive
1404 * transactions and deadlocks from xfs_inactive.
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001405 */
Dave Chinner58c90472015-02-23 22:38:08 +11001406 if (ip) {
1407 xfs_finish_inode_setup(ip);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001408 IRELE(ip);
Dave Chinner58c90472015-02-23 22:38:08 +11001409 }
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001410
1411 xfs_qm_dqrele(udqp);
1412 xfs_qm_dqrele(gdqp);
1413 xfs_qm_dqrele(pdqp);
1414
1415 return error;
1416}
1417
1418int
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001419xfs_link(
1420 xfs_inode_t *tdp,
1421 xfs_inode_t *sip,
1422 struct xfs_name *target_name)
1423{
1424 xfs_mount_t *mp = tdp->i_mount;
1425 xfs_trans_t *tp;
1426 int error;
1427 xfs_bmap_free_t free_list;
1428 xfs_fsblock_t first_block;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001429 int resblks;
1430
1431 trace_xfs_link(tdp, target_name);
1432
1433 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1434
1435 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001436 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001437
1438 error = xfs_qm_dqattach(sip, 0);
1439 if (error)
1440 goto std_return;
1441
1442 error = xfs_qm_dqattach(tdp, 0);
1443 if (error)
1444 goto std_return;
1445
1446 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001447 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001448 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10001449 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001450 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001451 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001452 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001453 if (error)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001454 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001455
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001456 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001457 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1458
1459 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001460 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001461
1462 /*
1463 * If we are using project inheritance, we only allow hard link
1464 * creation in our tree when the project IDs are the same; else
1465 * the tree quota mechanism could be circumvented.
1466 */
1467 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1468 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
Dave Chinner24513372014-06-25 14:58:08 +10001469 error = -EXDEV;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001470 goto error_return;
1471 }
1472
Eric Sandeen94f3cad2014-09-09 11:57:52 +10001473 if (!resblks) {
1474 error = xfs_dir_canenter(tp, tdp, target_name);
1475 if (error)
1476 goto error_return;
1477 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001478
1479 xfs_bmap_init(&free_list, &first_block);
1480
Zhi Yong Wuab297432013-12-18 08:22:41 +08001481 if (sip->i_d.di_nlink == 0) {
1482 error = xfs_iunlink_remove(tp, sip);
1483 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001484 goto error_return;
Zhi Yong Wuab297432013-12-18 08:22:41 +08001485 }
1486
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001487 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1488 &first_block, &free_list, resblks);
1489 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001490 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001491 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1492 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1493
1494 error = xfs_bumplink(tp, sip);
1495 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001496 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001497
1498 /*
1499 * If this is a synchronous mount, make sure that the
1500 * link transaction goes to disk before returning to
1501 * the user.
1502 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001503 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001504 xfs_trans_set_sync(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001505
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001506 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001507 if (error) {
1508 xfs_bmap_cancel(&free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10001509 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001510 }
1511
Christoph Hellwig70393312015-06-04 13:48:08 +10001512 return xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001513
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001514 error_return:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001515 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001516 std_return:
1517 return error;
1518}
1519
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001521 * Free up the underlying blocks past new_size. The new size must be smaller
1522 * than the current size. This routine can be used both for the attribute and
1523 * data fork, and does not modify the inode size, which is left to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 *
David Chinnerf6485052008-04-17 16:50:04 +10001525 * The transaction passed to this routine must have made a permanent log
1526 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1527 * given transaction and start new ones, so make sure everything involved in
1528 * the transaction is tidy before calling here. Some transaction will be
1529 * returned to the caller to be committed. The incoming transaction must
1530 * already include the inode, and both inode locks must be held exclusively.
1531 * The inode must also be "held" within the transaction. On return the inode
1532 * will be "held" within the returned transaction. This routine does NOT
1533 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 *
David Chinnerf6485052008-04-17 16:50:04 +10001535 * If we get an error, we must return with the inode locked and linked into the
1536 * current transaction. This keeps things simple for the higher level code,
1537 * because it always knows that the inode is locked and held in the transaction
1538 * that returns to it whether errors occur or not. We don't mark the inode
1539 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 */
1541int
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001542xfs_itruncate_extents(
1543 struct xfs_trans **tpp,
1544 struct xfs_inode *ip,
1545 int whichfork,
1546 xfs_fsize_t new_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547{
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001548 struct xfs_mount *mp = ip->i_mount;
1549 struct xfs_trans *tp = *tpp;
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001550 xfs_bmap_free_t free_list;
1551 xfs_fsblock_t first_block;
1552 xfs_fileoff_t first_unmap_block;
1553 xfs_fileoff_t last_block;
1554 xfs_filblks_t unmap_len;
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001555 int error = 0;
1556 int done = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Christoph Hellwig0b561852012-07-04 11:13:31 -04001558 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1559 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1560 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001561 ASSERT(new_size <= XFS_ISIZE(ip));
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001562 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 ASSERT(ip->i_itemp != NULL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001564 ASSERT(ip->i_itemp->ili_lock_flags == 0);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001565 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001567 trace_xfs_itruncate_extents_start(ip, new_size);
1568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 /*
1570 * Since it is possible for space to become allocated beyond
1571 * the end of the file (in a crash where the space is allocated
1572 * but the inode size is not yet updated), simply remove any
1573 * blocks which show up between the new EOF and the maximum
1574 * possible file size. If the first block to be removed is
1575 * beyond the maximum file size (ie it is the same as last_block),
1576 * then there is nothing to do.
1577 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001578 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
Dave Chinner32972382012-06-08 15:44:54 +10001579 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001580 if (first_unmap_block == last_block)
1581 return 0;
1582
1583 ASSERT(first_unmap_block < last_block);
1584 unmap_len = last_block - first_unmap_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 while (!done) {
Eric Sandeen9d87c312009-01-14 23:22:07 -06001586 xfs_bmap_init(&free_list, &first_block);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001587 error = xfs_bunmapi(tp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001588 first_unmap_block, unmap_len,
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001589 xfs_bmapi_aflag(whichfork),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001591 &first_block, &free_list,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001592 &done);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001593 if (error)
1594 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
1596 /*
1597 * Duplicate the transaction that has the permanent
1598 * reservation and commit the old transaction.
1599 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001600 error = xfs_bmap_finish(&tp, &free_list, ip);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001601 if (error)
1602 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001604 error = xfs_trans_roll(&tp, ip);
David Chinnerf6485052008-04-17 16:50:04 +10001605 if (error)
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001606 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 }
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001608
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001609 /*
1610 * Always re-log the inode so that our permanent transaction can keep
1611 * on rolling it forward in the log.
1612 */
1613 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1614
1615 trace_xfs_itruncate_extents_end(ip, new_size);
1616
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001617out:
1618 *tpp = tp;
1619 return error;
1620out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 /*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001622 * If the bunmapi call encounters an error, return to the caller where
1623 * the transaction can be properly aborted. We just need to make sure
1624 * we're not holding any resources that we were not when we came in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001626 xfs_bmap_cancel(&free_list);
1627 goto out;
1628}
1629
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001630int
1631xfs_release(
1632 xfs_inode_t *ip)
1633{
1634 xfs_mount_t *mp = ip->i_mount;
1635 int error;
1636
1637 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1638 return 0;
1639
1640 /* If this is a read-only mount, don't do this (would generate I/O) */
1641 if (mp->m_flags & XFS_MOUNT_RDONLY)
1642 return 0;
1643
1644 if (!XFS_FORCED_SHUTDOWN(mp)) {
1645 int truncated;
1646
1647 /*
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001648 * If we previously truncated this file and removed old data
1649 * in the process, we want to initiate "early" writeout on
1650 * the last close. This is an attempt to combat the notorious
1651 * NULL files problem which is particularly noticeable from a
1652 * truncate down, buffered (re-)write (delalloc), followed by
1653 * a crash. What we are effectively doing here is
1654 * significantly reducing the time window where we'd otherwise
1655 * be exposed to that problem.
1656 */
1657 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1658 if (truncated) {
1659 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
Dave Chinnereac152b2014-08-04 13:22:49 +10001660 if (ip->i_delayed_blks > 0) {
Dave Chinner24513372014-06-25 14:58:08 +10001661 error = filemap_flush(VFS_I(ip)->i_mapping);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001662 if (error)
1663 return error;
1664 }
1665 }
1666 }
1667
1668 if (ip->i_d.di_nlink == 0)
1669 return 0;
1670
1671 if (xfs_can_free_eofblocks(ip, false)) {
1672
1673 /*
1674 * If we can't get the iolock just skip truncating the blocks
1675 * past EOF because we could deadlock with the mmap_sem
1676 * otherwise. We'll get another chance to drop them once the
1677 * last reference to the inode is dropped, so we'll never leak
1678 * blocks permanently.
1679 *
1680 * Further, check if the inode is being opened, written and
1681 * closed frequently and we have delayed allocation blocks
1682 * outstanding (e.g. streaming writes from the NFS server),
1683 * truncating the blocks past EOF will cause fragmentation to
1684 * occur.
1685 *
1686 * In this case don't do the truncation, either, but we have to
1687 * be careful how we detect this case. Blocks beyond EOF show
1688 * up as i_delayed_blks even when the inode is clean, so we
1689 * need to truncate them away first before checking for a dirty
1690 * release. Hence on the first dirty close we will still remove
1691 * the speculative allocation, but after that we will leave it
1692 * in place.
1693 */
1694 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1695 return 0;
1696
1697 error = xfs_free_eofblocks(mp, ip, true);
Dave Chinner24513372014-06-25 14:58:08 +10001698 if (error && error != -EAGAIN)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001699 return error;
1700
1701 /* delalloc blocks after truncation means it really is dirty */
1702 if (ip->i_delayed_blks)
1703 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1704 }
1705 return 0;
1706}
1707
1708/*
Brian Fosterf7be2d72013-09-20 11:06:10 -04001709 * xfs_inactive_truncate
1710 *
1711 * Called to perform a truncate when an inode becomes unlinked.
1712 */
1713STATIC int
1714xfs_inactive_truncate(
1715 struct xfs_inode *ip)
1716{
1717 struct xfs_mount *mp = ip->i_mount;
1718 struct xfs_trans *tp;
1719 int error;
1720
1721 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1722 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1723 if (error) {
1724 ASSERT(XFS_FORCED_SHUTDOWN(mp));
Christoph Hellwig4906e212015-06-04 13:47:56 +10001725 xfs_trans_cancel(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001726 return error;
1727 }
1728
1729 xfs_ilock(ip, XFS_ILOCK_EXCL);
1730 xfs_trans_ijoin(tp, ip, 0);
1731
1732 /*
1733 * Log the inode size first to prevent stale data exposure in the event
1734 * of a system crash before the truncate completes. See the related
1735 * comment in xfs_setattr_size() for details.
1736 */
1737 ip->i_d.di_size = 0;
1738 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1739
1740 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1741 if (error)
1742 goto error_trans_cancel;
1743
1744 ASSERT(ip->i_d.di_nextents == 0);
1745
Christoph Hellwig70393312015-06-04 13:48:08 +10001746 error = xfs_trans_commit(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001747 if (error)
1748 goto error_unlock;
1749
1750 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1751 return 0;
1752
1753error_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001754 xfs_trans_cancel(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001755error_unlock:
1756 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1757 return error;
1758}
1759
1760/*
Brian Foster88877d22013-09-20 11:06:11 -04001761 * xfs_inactive_ifree()
1762 *
1763 * Perform the inode free when an inode is unlinked.
1764 */
1765STATIC int
1766xfs_inactive_ifree(
1767 struct xfs_inode *ip)
1768{
1769 xfs_bmap_free_t free_list;
1770 xfs_fsblock_t first_block;
Brian Foster88877d22013-09-20 11:06:11 -04001771 struct xfs_mount *mp = ip->i_mount;
1772 struct xfs_trans *tp;
1773 int error;
1774
1775 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
Brian Foster9d43b182014-04-24 16:00:52 +10001776
1777 /*
1778 * The ifree transaction might need to allocate blocks for record
1779 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1780 * allow ifree to dip into the reserved block pool if necessary.
1781 *
1782 * Freeing large sets of inodes generally means freeing inode chunks,
1783 * directory and file data blocks, so this should be relatively safe.
1784 * Only under severe circumstances should it be possible to free enough
1785 * inodes to exhaust the reserve block pool via finobt expansion while
1786 * at the same time not creating free space in the filesystem.
1787 *
1788 * Send a warning if the reservation does happen to fail, as the inode
1789 * now remains allocated and sits on the unlinked list until the fs is
1790 * repaired.
1791 */
1792 tp->t_flags |= XFS_TRANS_RESERVE;
1793 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree,
1794 XFS_IFREE_SPACE_RES(mp), 0);
Brian Foster88877d22013-09-20 11:06:11 -04001795 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10001796 if (error == -ENOSPC) {
Brian Foster9d43b182014-04-24 16:00:52 +10001797 xfs_warn_ratelimited(mp,
1798 "Failed to remove inode(s) from unlinked list. "
1799 "Please free space, unmount and run xfs_repair.");
1800 } else {
1801 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1802 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001803 xfs_trans_cancel(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001804 return error;
1805 }
1806
1807 xfs_ilock(ip, XFS_ILOCK_EXCL);
1808 xfs_trans_ijoin(tp, ip, 0);
1809
1810 xfs_bmap_init(&free_list, &first_block);
1811 error = xfs_ifree(tp, ip, &free_list);
1812 if (error) {
1813 /*
1814 * If we fail to free the inode, shut down. The cancel
1815 * might do that, we need to make sure. Otherwise the
1816 * inode might be lost for a long time or forever.
1817 */
1818 if (!XFS_FORCED_SHUTDOWN(mp)) {
1819 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1820 __func__, error);
1821 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1822 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001823 xfs_trans_cancel(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001824 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1825 return error;
1826 }
1827
1828 /*
1829 * Credit the quota account(s). The inode is gone.
1830 */
1831 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1832
1833 /*
Brian Fosterd4a97a02015-08-19 10:01:40 +10001834 * Just ignore errors at this point. There is nothing we can do except
1835 * to try to keep going. Make sure it's not a silent error.
Brian Foster88877d22013-09-20 11:06:11 -04001836 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001837 error = xfs_bmap_finish(&tp, &free_list, NULL);
Brian Fosterd4a97a02015-08-19 10:01:40 +10001838 if (error) {
Brian Foster88877d22013-09-20 11:06:11 -04001839 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1840 __func__, error);
Brian Fosterd4a97a02015-08-19 10:01:40 +10001841 xfs_bmap_cancel(&free_list);
1842 }
Christoph Hellwig70393312015-06-04 13:48:08 +10001843 error = xfs_trans_commit(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001844 if (error)
1845 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1846 __func__, error);
1847
1848 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1849 return 0;
1850}
1851
1852/*
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001853 * xfs_inactive
1854 *
1855 * This is called when the vnode reference count for the vnode
1856 * goes to zero. If the file has been unlinked, then it must
1857 * now be truncated. Also, we clear all of the read-ahead state
1858 * kept for the inode here since the file is now closed.
1859 */
Brian Foster74564fb2013-09-20 11:06:12 -04001860void
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001861xfs_inactive(
1862 xfs_inode_t *ip)
1863{
Jie Liu3d3c8b52013-08-12 20:49:59 +10001864 struct xfs_mount *mp;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001865 int error;
1866 int truncate = 0;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001867
1868 /*
1869 * If the inode is already free, then there can be nothing
1870 * to clean up here.
1871 */
Ben Myersd9487092013-09-10 18:11:22 -05001872 if (ip->i_d.di_mode == 0) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001873 ASSERT(ip->i_df.if_real_bytes == 0);
1874 ASSERT(ip->i_df.if_broot_bytes == 0);
Brian Foster74564fb2013-09-20 11:06:12 -04001875 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001876 }
1877
1878 mp = ip->i_mount;
1879
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001880 /* If this is a read-only mount, don't do this (would generate I/O) */
1881 if (mp->m_flags & XFS_MOUNT_RDONLY)
Brian Foster74564fb2013-09-20 11:06:12 -04001882 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001883
1884 if (ip->i_d.di_nlink != 0) {
1885 /*
1886 * force is true because we are evicting an inode from the
1887 * cache. Post-eof blocks must be freed, lest we end up with
1888 * broken free space accounting.
1889 */
Brian Foster74564fb2013-09-20 11:06:12 -04001890 if (xfs_can_free_eofblocks(ip, true))
1891 xfs_free_eofblocks(mp, ip, false);
1892
1893 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001894 }
1895
1896 if (S_ISREG(ip->i_d.di_mode) &&
1897 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1898 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1899 truncate = 1;
1900
1901 error = xfs_qm_dqattach(ip, 0);
1902 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001903 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001904
Brian Fosterf7be2d72013-09-20 11:06:10 -04001905 if (S_ISLNK(ip->i_d.di_mode))
Brian Foster36b21dd2013-09-20 11:06:09 -04001906 error = xfs_inactive_symlink(ip);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001907 else if (truncate)
1908 error = xfs_inactive_truncate(ip);
1909 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001910 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001911
1912 /*
1913 * If there are attributes associated with the file then blow them away
1914 * now. The code calls a routine that recursively deconstructs the
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001915 * attribute fork. If also blows away the in-core attribute fork.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001916 */
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001917 if (XFS_IFORK_Q(ip)) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001918 error = xfs_attr_inactive(ip);
1919 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001920 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001921 }
1922
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001923 ASSERT(!ip->i_afp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001924 ASSERT(ip->i_d.di_anextents == 0);
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001925 ASSERT(ip->i_d.di_forkoff == 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001926
1927 /*
1928 * Free the inode.
1929 */
Brian Foster88877d22013-09-20 11:06:11 -04001930 error = xfs_inactive_ifree(ip);
1931 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001932 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001933
1934 /*
1935 * Release the dquots held by inode, if any.
1936 */
1937 xfs_qm_dqdetach(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001938}
1939
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940/*
1941 * This is called when the inode's link count goes to 0.
1942 * We place the on-disk inode on a list in the AGI. It
1943 * will be pulled from this list when the inode is freed.
1944 */
1945int
1946xfs_iunlink(
1947 xfs_trans_t *tp,
1948 xfs_inode_t *ip)
1949{
1950 xfs_mount_t *mp;
1951 xfs_agi_t *agi;
1952 xfs_dinode_t *dip;
1953 xfs_buf_t *agibp;
1954 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 xfs_agino_t agino;
1956 short bucket_index;
1957 int offset;
1958 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
1960 ASSERT(ip->i_d.di_nlink == 0);
1961 ASSERT(ip->i_d.di_mode != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 mp = tp->t_mountp;
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 /*
1966 * Get the agi buffer first. It ensures lock ordering
1967 * on the list.
1968 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001969 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001970 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001973
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 /*
1975 * Get the index into the agi hash table for the
1976 * list this inode will go on.
1977 */
1978 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1979 ASSERT(agino != 0);
1980 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1981 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001982 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001984 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 /*
1986 * There is already another inode in the bucket we need
1987 * to add ourselves to. Add us at the front of the list.
1988 * Here we put the head pointer into our next pointer,
1989 * and then we fall through to point the head at us.
1990 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04001991 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1992 0, 0);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001993 if (error)
1994 return error;
1995
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001996 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001998 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002000
2001 /* need to recalc the inode CRC if appropriate */
2002 xfs_dinode_calc_crc(mp, dip);
2003
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 xfs_trans_inode_buf(tp, ibp);
2005 xfs_trans_log_buf(tp, ibp, offset,
2006 (offset + sizeof(xfs_agino_t) - 1));
2007 xfs_inobp_check(mp, ibp);
2008 }
2009
2010 /*
2011 * Point the bucket head pointer at the inode being inserted.
2012 */
2013 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002014 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 offset = offsetof(xfs_agi_t, agi_unlinked) +
2016 (sizeof(xfs_agino_t) * bucket_index);
Dave Chinnerf19b8722015-01-22 09:29:40 +11002017 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 xfs_trans_log_buf(tp, agibp, offset,
2019 (offset + sizeof(xfs_agino_t) - 1));
2020 return 0;
2021}
2022
2023/*
2024 * Pull the on-disk inode from the AGI unlinked list.
2025 */
2026STATIC int
2027xfs_iunlink_remove(
2028 xfs_trans_t *tp,
2029 xfs_inode_t *ip)
2030{
2031 xfs_ino_t next_ino;
2032 xfs_mount_t *mp;
2033 xfs_agi_t *agi;
2034 xfs_dinode_t *dip;
2035 xfs_buf_t *agibp;
2036 xfs_buf_t *ibp;
2037 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 xfs_agino_t agino;
2039 xfs_agino_t next_agino;
2040 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002041 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002043 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
2049 /*
2050 * Get the agi buffer first. It ensures lock ordering
2051 * on the list.
2052 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002053 error = xfs_read_agi(mp, tp, agno, &agibp);
2054 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002056
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 /*
2060 * Get the index into the agi hash table for the
2061 * list this inode will go on.
2062 */
2063 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2064 ASSERT(agino != 0);
2065 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002066 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 ASSERT(agi->agi_unlinked[bucket_index]);
2068
Christoph Hellwig16259e72005-11-02 15:11:25 +11002069 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002071 * We're at the head of the list. Get the inode's on-disk
2072 * buffer to see if there is anyone after us on the list.
2073 * Only modify our next pointer if it is not already NULLAGINO.
2074 * This saves us the overhead of dealing with the buffer when
2075 * there is no need to change it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002077 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2078 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002080 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002081 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 return error;
2083 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002084 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 ASSERT(next_agino != 0);
2086 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002087 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002088 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002090
2091 /* need to recalc the inode CRC if appropriate */
2092 xfs_dinode_calc_crc(mp, dip);
2093
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 xfs_trans_inode_buf(tp, ibp);
2095 xfs_trans_log_buf(tp, ibp, offset,
2096 (offset + sizeof(xfs_agino_t) - 1));
2097 xfs_inobp_check(mp, ibp);
2098 } else {
2099 xfs_trans_brelse(tp, ibp);
2100 }
2101 /*
2102 * Point the bucket head pointer at the next inode.
2103 */
2104 ASSERT(next_agino != 0);
2105 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002106 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 offset = offsetof(xfs_agi_t, agi_unlinked) +
2108 (sizeof(xfs_agino_t) * bucket_index);
Dave Chinnerf19b8722015-01-22 09:29:40 +11002109 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 xfs_trans_log_buf(tp, agibp, offset,
2111 (offset + sizeof(xfs_agino_t) - 1));
2112 } else {
2113 /*
2114 * We need to search the list for the inode being freed.
2115 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002116 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 last_ibp = NULL;
2118 while (next_agino != agino) {
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002119 struct xfs_imap imap;
2120
2121 if (last_ibp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 xfs_trans_brelse(tp, last_ibp);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002123
2124 imap.im_blkno = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002126
2127 error = xfs_imap(mp, tp, next_ino, &imap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11002129 xfs_warn(mp,
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002130 "%s: xfs_imap returned error %d.",
2131 __func__, error);
2132 return error;
2133 }
2134
2135 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2136 &last_ibp, 0, 0);
2137 if (error) {
2138 xfs_warn(mp,
2139 "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002140 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 return error;
2142 }
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002143
2144 last_offset = imap.im_boffset;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002145 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 ASSERT(next_agino != NULLAGINO);
2147 ASSERT(next_agino != 0);
2148 }
Christoph Hellwig475ee412012-07-03 12:21:22 -04002149
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002151 * Now last_ibp points to the buffer previous to us on the
2152 * unlinked list. Pull us from the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002154 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2155 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002157 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002158 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 return error;
2160 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002161 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 ASSERT(next_agino != 0);
2163 ASSERT(next_agino != agino);
2164 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002165 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002166 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002168
2169 /* need to recalc the inode CRC if appropriate */
2170 xfs_dinode_calc_crc(mp, dip);
2171
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 xfs_trans_inode_buf(tp, ibp);
2173 xfs_trans_log_buf(tp, ibp, offset,
2174 (offset + sizeof(xfs_agino_t) - 1));
2175 xfs_inobp_check(mp, ibp);
2176 } else {
2177 xfs_trans_brelse(tp, ibp);
2178 }
2179 /*
2180 * Point the previous inode on the list to the next inode.
2181 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002182 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 ASSERT(next_agino != 0);
2184 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002185
2186 /* need to recalc the inode CRC if appropriate */
2187 xfs_dinode_calc_crc(mp, last_dip);
2188
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 xfs_trans_inode_buf(tp, last_ibp);
2190 xfs_trans_log_buf(tp, last_ibp, offset,
2191 (offset + sizeof(xfs_agino_t) - 1));
2192 xfs_inobp_check(mp, last_ibp);
2193 }
2194 return 0;
2195}
2196
Dave Chinner5b3eed72010-08-24 11:42:41 +10002197/*
Zhi Yong Wu0b8182d2013-08-12 03:14:59 +00002198 * A big issue when freeing the inode cluster is that we _cannot_ skip any
Dave Chinner5b3eed72010-08-24 11:42:41 +10002199 * inodes that are in memory - they all must be marked stale and attached to
2200 * the cluster buffer.
2201 */
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002202STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203xfs_ifree_cluster(
Brian Foster09b56602015-05-29 09:26:03 +10002204 xfs_inode_t *free_ip,
2205 xfs_trans_t *tp,
2206 struct xfs_icluster *xic)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207{
2208 xfs_mount_t *mp = free_ip->i_mount;
2209 int blks_per_cluster;
Jie Liu982e9392013-12-13 15:51:49 +11002210 int inodes_per_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 int nbufs;
Dave Chinner5b257b42010-06-03 16:22:29 +10002212 int i, j;
Brian Foster3cdaa182015-06-04 13:03:34 +10002213 int ioffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 xfs_daddr_t blkno;
2215 xfs_buf_t *bp;
Dave Chinner5b257b42010-06-03 16:22:29 +10002216 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 xfs_inode_log_item_t *iip;
2218 xfs_log_item_t *lip;
Dave Chinner5017e972010-01-11 11:47:40 +00002219 struct xfs_perag *pag;
Brian Foster09b56602015-05-29 09:26:03 +10002220 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Brian Foster09b56602015-05-29 09:26:03 +10002222 inum = xic->first_ino;
Dave Chinner5017e972010-01-11 11:47:40 +00002223 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
Jie Liu982e9392013-12-13 15:51:49 +11002224 blks_per_cluster = xfs_icluster_size_fsb(mp);
2225 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2226 nbufs = mp->m_ialloc_blks / blks_per_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
Jie Liu982e9392013-12-13 15:51:49 +11002228 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
Brian Foster09b56602015-05-29 09:26:03 +10002229 /*
2230 * The allocation bitmap tells us which inodes of the chunk were
2231 * physically allocated. Skip the cluster if an inode falls into
2232 * a sparse region.
2233 */
Brian Foster3cdaa182015-06-04 13:03:34 +10002234 ioffset = inum - xic->first_ino;
2235 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2236 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
Brian Foster09b56602015-05-29 09:26:03 +10002237 continue;
2238 }
2239
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2241 XFS_INO_TO_AGBNO(mp, inum));
2242
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 /*
Dave Chinner5b257b42010-06-03 16:22:29 +10002244 * We obtain and lock the backing buffer first in the process
2245 * here, as we have to ensure that any dirty inode that we
2246 * can't get the flush lock on is attached to the buffer.
2247 * If we scan the in-memory inodes first, then buffer IO can
2248 * complete before we get a lock on it, and hence we may fail
2249 * to mark all the active inodes on the buffer stale.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002251 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
Dave Chinnerb6aff292012-11-02 11:38:42 +11002252 mp->m_bsize * blks_per_cluster,
2253 XBF_UNMAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002255 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002256 return -ENOMEM;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002257
2258 /*
2259 * This buffer may not have been correctly initialised as we
2260 * didn't read it from disk. That's not important because we are
2261 * only using to mark the buffer as stale in the log, and to
2262 * attach stale cached inodes on it. That means it will never be
2263 * dispatched for IO. If it is, we want to know about it, and we
2264 * want it to fail. We can acheive this by adding a write
2265 * verifier to the buffer.
2266 */
Dave Chinner1813dd62012-11-14 17:54:40 +11002267 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002268
Dave Chinner5b257b42010-06-03 16:22:29 +10002269 /*
2270 * Walk the inodes already attached to the buffer and mark them
2271 * stale. These will all have the flush locks held, so an
Dave Chinner5b3eed72010-08-24 11:42:41 +10002272 * in-memory inode walk can't lock them. By marking them all
2273 * stale first, we will not attempt to lock them in the loop
2274 * below as the XFS_ISTALE flag will be set.
Dave Chinner5b257b42010-06-03 16:22:29 +10002275 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +02002276 lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 while (lip) {
2278 if (lip->li_type == XFS_LI_INODE) {
2279 iip = (xfs_inode_log_item_t *)lip;
2280 ASSERT(iip->ili_logged == 1);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002281 lip->li_cb = xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11002282 xfs_trans_ail_copy_lsn(mp->m_ail,
2283 &iip->ili_flush_lsn,
2284 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11002285 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 }
2287 lip = lip->li_bio_list;
2288 }
2289
Dave Chinner5b3eed72010-08-24 11:42:41 +10002290
Dave Chinner5b257b42010-06-03 16:22:29 +10002291 /*
2292 * For each inode in memory attempt to add it to the inode
2293 * buffer and set it up for being staled on buffer IO
2294 * completion. This is safe as we've locked out tail pushing
2295 * and flushing by locking the buffer.
2296 *
2297 * We have already marked every inode that was part of a
2298 * transaction stale above, which means there is no point in
2299 * even trying to lock them.
2300 */
Jie Liu982e9392013-12-13 15:51:49 +11002301 for (i = 0; i < inodes_per_cluster; i++) {
Dave Chinner5b3eed72010-08-24 11:42:41 +10002302retry:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002303 rcu_read_lock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002304 ip = radix_tree_lookup(&pag->pag_ici_root,
2305 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002307 /* Inode not in memory, nothing to do */
2308 if (!ip) {
2309 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002310 continue;
2311 }
2312
Dave Chinner5b3eed72010-08-24 11:42:41 +10002313 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002314 * because this is an RCU protected lookup, we could
2315 * find a recently freed or even reallocated inode
2316 * during the lookup. We need to check under the
2317 * i_flags_lock for a valid inode here. Skip it if it
2318 * is not valid, the wrong inode or stale.
2319 */
2320 spin_lock(&ip->i_flags_lock);
2321 if (ip->i_ino != inum + i ||
2322 __xfs_iflags_test(ip, XFS_ISTALE)) {
2323 spin_unlock(&ip->i_flags_lock);
2324 rcu_read_unlock();
2325 continue;
2326 }
2327 spin_unlock(&ip->i_flags_lock);
2328
2329 /*
Dave Chinner5b3eed72010-08-24 11:42:41 +10002330 * Don't try to lock/unlock the current inode, but we
2331 * _cannot_ skip the other inodes that we did not find
2332 * in the list attached to the buffer and are not
2333 * already marked stale. If we can't lock it, back off
2334 * and retry.
2335 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002336 if (ip != free_ip &&
2337 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002338 rcu_read_unlock();
Dave Chinner5b3eed72010-08-24 11:42:41 +10002339 delay(1);
2340 goto retry;
Dave Chinner5b257b42010-06-03 16:22:29 +10002341 }
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002342 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002343
Dave Chinner5b3eed72010-08-24 11:42:41 +10002344 xfs_iflock(ip);
Dave Chinner5b257b42010-06-03 16:22:29 +10002345 xfs_iflags_set(ip, XFS_ISTALE);
Dave Chinner5b257b42010-06-03 16:22:29 +10002346
Dave Chinner5b3eed72010-08-24 11:42:41 +10002347 /*
2348 * we don't need to attach clean inodes or those only
2349 * with unlogged changes (which we throw away, anyway).
2350 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002351 iip = ip->i_itemp;
Dave Chinner5b3eed72010-08-24 11:42:41 +10002352 if (!iip || xfs_inode_clean(ip)) {
Dave Chinner5b257b42010-06-03 16:22:29 +10002353 ASSERT(ip != free_ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 xfs_ifunlock(ip);
2355 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2356 continue;
2357 }
2358
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002359 iip->ili_last_fields = iip->ili_fields;
2360 iip->ili_fields = 0;
Dave Chinnerfc0561c2015-11-03 13:14:59 +11002361 iip->ili_fsync_fields = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11002363 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2364 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002366 xfs_buf_attach_iodone(bp, xfs_istale_done,
2367 &iip->ili_item);
Dave Chinner5b257b42010-06-03 16:22:29 +10002368
2369 if (ip != free_ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 }
2372
Dave Chinner5b3eed72010-08-24 11:42:41 +10002373 xfs_trans_stale_inode_buf(tp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 xfs_trans_binval(tp, bp);
2375 }
2376
Dave Chinner5017e972010-01-11 11:47:40 +00002377 xfs_perag_put(pag);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002378 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379}
2380
2381/*
2382 * This is called to return an inode to the inode free list.
2383 * The inode should already be truncated to 0 length and have
2384 * no pages associated with it. This routine also assumes that
2385 * the inode is already a part of the transaction.
2386 *
2387 * The on-disk copy of the inode will have been added to the list
2388 * of unlinked inodes in the AGI. We need to remove the inode from
2389 * that list atomically with respect to freeing it here.
2390 */
2391int
2392xfs_ifree(
2393 xfs_trans_t *tp,
2394 xfs_inode_t *ip,
2395 xfs_bmap_free_t *flist)
2396{
2397 int error;
Brian Foster09b56602015-05-29 09:26:03 +10002398 struct xfs_icluster xic = { 0 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002400 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 ASSERT(ip->i_d.di_nlink == 0);
2402 ASSERT(ip->i_d.di_nextents == 0);
2403 ASSERT(ip->i_d.di_anextents == 0);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00002404 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 ASSERT(ip->i_d.di_nblocks == 0);
2406
2407 /*
2408 * Pull the on-disk inode from the AGI unlinked list.
2409 */
2410 error = xfs_iunlink_remove(tp, ip);
Dave Chinner1baaed82013-06-27 16:04:50 +10002411 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413
Brian Foster09b56602015-05-29 09:26:03 +10002414 error = xfs_difree(tp, ip->i_ino, flist, &xic);
Dave Chinner1baaed82013-06-27 16:04:50 +10002415 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 return error;
Dave Chinner1baaed82013-06-27 16:04:50 +10002417
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 ip->i_d.di_mode = 0; /* mark incore inode as free */
2419 ip->i_d.di_flags = 0;
2420 ip->i_d.di_dmevmask = 0;
2421 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2423 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2424 /*
2425 * Bump the generation count so no one will be confused
2426 * by reincarnations of this inode.
2427 */
2428 ip->i_d.di_gen++;
2429 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2430
Brian Foster09b56602015-05-29 09:26:03 +10002431 if (xic.deleted)
2432 error = xfs_ifree_cluster(ip, tp, &xic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002434 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435}
2436
2437/*
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002438 * This is called to unpin an inode. The caller must have the inode locked
2439 * in at least shared mode so that the buffer cannot be subsequently pinned
2440 * once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002442static void
Christoph Hellwigf392e632011-12-18 20:00:10 +00002443xfs_iunpin(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002444 struct xfs_inode *ip)
David Chinnera3f74ff2008-03-06 13:43:42 +11002445{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002446 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002447
Dave Chinner4aaf15d2010-03-08 11:24:07 +11002448 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2449
David Chinnera3f74ff2008-03-06 13:43:42 +11002450 /* Give the log a push to start the unpinning I/O */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002451 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002452
David Chinnera3f74ff2008-03-06 13:43:42 +11002453}
2454
Christoph Hellwigf392e632011-12-18 20:00:10 +00002455static void
2456__xfs_iunpin_wait(
2457 struct xfs_inode *ip)
2458{
2459 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2460 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2461
2462 xfs_iunpin(ip);
2463
2464 do {
2465 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2466 if (xfs_ipincount(ip))
2467 io_schedule();
2468 } while (xfs_ipincount(ip));
2469 finish_wait(wq, &wait.wait);
2470}
2471
Dave Chinner777df5a2010-02-06 12:37:26 +11002472void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473xfs_iunpin_wait(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002474 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475{
Christoph Hellwigf392e632011-12-18 20:00:10 +00002476 if (xfs_ipincount(ip))
2477 __xfs_iunpin_wait(ip);
David Chinnera3f74ff2008-03-06 13:43:42 +11002478}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479
Dave Chinner27320362013-10-29 22:11:44 +11002480/*
2481 * Removing an inode from the namespace involves removing the directory entry
2482 * and dropping the link count on the inode. Removing the directory entry can
2483 * result in locking an AGF (directory blocks were freed) and removing a link
2484 * count can result in placing the inode on an unlinked list which results in
2485 * locking an AGI.
2486 *
2487 * The big problem here is that we have an ordering constraint on AGF and AGI
2488 * locking - inode allocation locks the AGI, then can allocate a new extent for
2489 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2490 * removes the inode from the unlinked list, requiring that we lock the AGI
2491 * first, and then freeing the inode can result in an inode chunk being freed
2492 * and hence freeing disk space requiring that we lock an AGF.
2493 *
2494 * Hence the ordering that is imposed by other parts of the code is AGI before
2495 * AGF. This means we cannot remove the directory entry before we drop the inode
2496 * reference count and put it on the unlinked list as this results in a lock
2497 * order of AGF then AGI, and this can deadlock against inode allocation and
2498 * freeing. Therefore we must drop the link counts before we remove the
2499 * directory entry.
2500 *
2501 * This is still safe from a transactional point of view - it is not until we
2502 * get to xfs_bmap_finish() that we have the possibility of multiple
2503 * transactions in this operation. Hence as long as we remove the directory
2504 * entry and drop the link count in the first transaction of the remove
2505 * operation, there are no transactional constraints on the ordering here.
2506 */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002507int
2508xfs_remove(
2509 xfs_inode_t *dp,
2510 struct xfs_name *name,
2511 xfs_inode_t *ip)
2512{
2513 xfs_mount_t *mp = dp->i_mount;
2514 xfs_trans_t *tp = NULL;
2515 int is_dir = S_ISDIR(ip->i_d.di_mode);
2516 int error = 0;
2517 xfs_bmap_free_t free_list;
2518 xfs_fsblock_t first_block;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002519 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002520
2521 trace_xfs_remove(dp, name);
2522
2523 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10002524 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002525
2526 error = xfs_qm_dqattach(dp, 0);
2527 if (error)
2528 goto std_return;
2529
2530 error = xfs_qm_dqattach(ip, 0);
2531 if (error)
2532 goto std_return;
2533
Dave Chinner32296f82014-12-04 09:43:17 +11002534 if (is_dir)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002535 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
Dave Chinner32296f82014-12-04 09:43:17 +11002536 else
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002537 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002538
2539 /*
2540 * We try to get the real space reservation first,
2541 * allowing for directory btree deletion(s) implying
2542 * possible bmap insert(s). If we can't get the space
2543 * reservation then we use 0 instead, and avoid the bmap
2544 * btree insert(s) in the directory code by, if the bmap
2545 * insert tries to happen, instead trimming the LAST
2546 * block from the directory.
2547 */
2548 resblks = XFS_REMOVE_SPACE_RES(mp);
Jie Liu3d3c8b52013-08-12 20:49:59 +10002549 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
Dave Chinner24513372014-06-25 14:58:08 +10002550 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002551 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10002552 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002553 }
2554 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10002555 ASSERT(error != -ENOSPC);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002556 goto out_trans_cancel;
2557 }
2558
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002559 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002560 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2561
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002562 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002563 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2564
2565 /*
2566 * If we're removing a directory perform some additional validation.
2567 */
2568 if (is_dir) {
2569 ASSERT(ip->i_d.di_nlink >= 2);
2570 if (ip->i_d.di_nlink != 2) {
Dave Chinner24513372014-06-25 14:58:08 +10002571 error = -ENOTEMPTY;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002572 goto out_trans_cancel;
2573 }
2574 if (!xfs_dir_isempty(ip)) {
Dave Chinner24513372014-06-25 14:58:08 +10002575 error = -ENOTEMPTY;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002576 goto out_trans_cancel;
2577 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002578
Dave Chinner27320362013-10-29 22:11:44 +11002579 /* Drop the link from ip's "..". */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002580 error = xfs_droplink(tp, dp);
2581 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002582 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002583
Dave Chinner27320362013-10-29 22:11:44 +11002584 /* Drop the "." link from ip to self. */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002585 error = xfs_droplink(tp, ip);
2586 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002587 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002588 } else {
2589 /*
2590 * When removing a non-directory we need to log the parent
2591 * inode here. For a directory this is done implicitly
2592 * by the xfs_droplink call for the ".." entry.
2593 */
2594 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2595 }
Dave Chinner27320362013-10-29 22:11:44 +11002596 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002597
Dave Chinner27320362013-10-29 22:11:44 +11002598 /* Drop the link from dp to ip. */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002599 error = xfs_droplink(tp, ip);
2600 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002601 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002602
Dave Chinner27320362013-10-29 22:11:44 +11002603 xfs_bmap_init(&free_list, &first_block);
2604 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2605 &first_block, &free_list, resblks);
2606 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10002607 ASSERT(error != -ENOENT);
Dave Chinner27320362013-10-29 22:11:44 +11002608 goto out_bmap_cancel;
2609 }
2610
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002611 /*
2612 * If this is a synchronous mount, make sure that the
2613 * remove transaction goes to disk before returning to
2614 * the user.
2615 */
2616 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2617 xfs_trans_set_sync(tp);
2618
Eric Sandeenf6106ef2016-01-11 11:34:01 +11002619 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002620 if (error)
2621 goto out_bmap_cancel;
2622
Christoph Hellwig70393312015-06-04 13:48:08 +10002623 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002624 if (error)
2625 goto std_return;
2626
Christoph Hellwig2cd2ef62014-04-23 07:11:51 +10002627 if (is_dir && xfs_inode_is_filestream(ip))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002628 xfs_filestream_deassociate(ip);
2629
2630 return 0;
2631
2632 out_bmap_cancel:
2633 xfs_bmap_cancel(&free_list);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002634 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10002635 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002636 std_return:
2637 return error;
2638}
2639
Dave Chinnerf6bba202013-08-12 20:49:46 +10002640/*
2641 * Enter all inodes for a rename transaction into a sorted array.
2642 */
Dave Chinner95afcf52015-03-25 14:03:32 +11002643#define __XFS_SORT_INODES 5
Dave Chinnerf6bba202013-08-12 20:49:46 +10002644STATIC void
2645xfs_sort_for_rename(
Dave Chinner95afcf52015-03-25 14:03:32 +11002646 struct xfs_inode *dp1, /* in: old (source) directory inode */
2647 struct xfs_inode *dp2, /* in: new (target) directory inode */
2648 struct xfs_inode *ip1, /* in: inode of old entry */
2649 struct xfs_inode *ip2, /* in: inode of new entry */
2650 struct xfs_inode *wip, /* in: whiteout inode */
2651 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2652 int *num_inodes) /* in/out: inodes in array */
Dave Chinnerf6bba202013-08-12 20:49:46 +10002653{
Dave Chinnerf6bba202013-08-12 20:49:46 +10002654 int i, j;
2655
Dave Chinner95afcf52015-03-25 14:03:32 +11002656 ASSERT(*num_inodes == __XFS_SORT_INODES);
2657 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2658
Dave Chinnerf6bba202013-08-12 20:49:46 +10002659 /*
2660 * i_tab contains a list of pointers to inodes. We initialize
2661 * the table here & we'll sort it. We will then use it to
2662 * order the acquisition of the inode locks.
2663 *
2664 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2665 */
Dave Chinner95afcf52015-03-25 14:03:32 +11002666 i = 0;
2667 i_tab[i++] = dp1;
2668 i_tab[i++] = dp2;
2669 i_tab[i++] = ip1;
2670 if (ip2)
2671 i_tab[i++] = ip2;
2672 if (wip)
2673 i_tab[i++] = wip;
2674 *num_inodes = i;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002675
2676 /*
2677 * Sort the elements via bubble sort. (Remember, there are at
Dave Chinner95afcf52015-03-25 14:03:32 +11002678 * most 5 elements to sort, so this is adequate.)
Dave Chinnerf6bba202013-08-12 20:49:46 +10002679 */
2680 for (i = 0; i < *num_inodes; i++) {
2681 for (j = 1; j < *num_inodes; j++) {
2682 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
Dave Chinner95afcf52015-03-25 14:03:32 +11002683 struct xfs_inode *temp = i_tab[j];
Dave Chinnerf6bba202013-08-12 20:49:46 +10002684 i_tab[j] = i_tab[j-1];
2685 i_tab[j-1] = temp;
2686 }
2687 }
2688 }
2689}
2690
Dave Chinner310606b2015-03-25 14:06:07 +11002691static int
2692xfs_finish_rename(
2693 struct xfs_trans *tp,
2694 struct xfs_bmap_free *free_list)
2695{
Dave Chinner310606b2015-03-25 14:06:07 +11002696 int error;
2697
2698 /*
2699 * If this is a synchronous mount, make sure that the rename transaction
2700 * goes to disk before returning to the user.
2701 */
2702 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2703 xfs_trans_set_sync(tp);
2704
Eric Sandeenf6106ef2016-01-11 11:34:01 +11002705 error = xfs_bmap_finish(&tp, free_list, NULL);
Dave Chinner310606b2015-03-25 14:06:07 +11002706 if (error) {
2707 xfs_bmap_cancel(free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10002708 xfs_trans_cancel(tp);
Dave Chinner310606b2015-03-25 14:06:07 +11002709 return error;
2710 }
2711
Christoph Hellwig70393312015-06-04 13:48:08 +10002712 return xfs_trans_commit(tp);
Dave Chinner310606b2015-03-25 14:06:07 +11002713}
2714
Dave Chinnerf6bba202013-08-12 20:49:46 +10002715/*
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002716 * xfs_cross_rename()
2717 *
2718 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2719 */
2720STATIC int
2721xfs_cross_rename(
2722 struct xfs_trans *tp,
2723 struct xfs_inode *dp1,
2724 struct xfs_name *name1,
2725 struct xfs_inode *ip1,
2726 struct xfs_inode *dp2,
2727 struct xfs_name *name2,
2728 struct xfs_inode *ip2,
2729 struct xfs_bmap_free *free_list,
2730 xfs_fsblock_t *first_block,
2731 int spaceres)
2732{
2733 int error = 0;
2734 int ip1_flags = 0;
2735 int ip2_flags = 0;
2736 int dp2_flags = 0;
2737
2738 /* Swap inode number for dirent in first parent */
2739 error = xfs_dir_replace(tp, dp1, name1,
2740 ip2->i_ino,
2741 first_block, free_list, spaceres);
2742 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002743 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002744
2745 /* Swap inode number for dirent in second parent */
2746 error = xfs_dir_replace(tp, dp2, name2,
2747 ip1->i_ino,
2748 first_block, free_list, spaceres);
2749 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002750 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002751
2752 /*
2753 * If we're renaming one or more directories across different parents,
2754 * update the respective ".." entries (and link counts) to match the new
2755 * parents.
2756 */
2757 if (dp1 != dp2) {
2758 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2759
2760 if (S_ISDIR(ip2->i_d.di_mode)) {
2761 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2762 dp1->i_ino, first_block,
2763 free_list, spaceres);
2764 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002765 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002766
2767 /* transfer ip2 ".." reference to dp1 */
2768 if (!S_ISDIR(ip1->i_d.di_mode)) {
2769 error = xfs_droplink(tp, dp2);
2770 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002771 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002772 error = xfs_bumplink(tp, dp1);
2773 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002774 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002775 }
2776
2777 /*
2778 * Although ip1 isn't changed here, userspace needs
2779 * to be warned about the change, so that applications
2780 * relying on it (like backup ones), will properly
2781 * notify the change
2782 */
2783 ip1_flags |= XFS_ICHGTIME_CHG;
2784 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2785 }
2786
2787 if (S_ISDIR(ip1->i_d.di_mode)) {
2788 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2789 dp2->i_ino, first_block,
2790 free_list, spaceres);
2791 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002792 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002793
2794 /* transfer ip1 ".." reference to dp2 */
2795 if (!S_ISDIR(ip2->i_d.di_mode)) {
2796 error = xfs_droplink(tp, dp1);
2797 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002798 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002799 error = xfs_bumplink(tp, dp2);
2800 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002801 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002802 }
2803
2804 /*
2805 * Although ip2 isn't changed here, userspace needs
2806 * to be warned about the change, so that applications
2807 * relying on it (like backup ones), will properly
2808 * notify the change
2809 */
2810 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2811 ip2_flags |= XFS_ICHGTIME_CHG;
2812 }
2813 }
2814
2815 if (ip1_flags) {
2816 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2817 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2818 }
2819 if (ip2_flags) {
2820 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2821 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2822 }
2823 if (dp2_flags) {
2824 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2825 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2826 }
2827 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2828 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
Dave Chinnereeacd322015-03-25 14:08:07 +11002829 return xfs_finish_rename(tp, free_list);
2830
2831out_trans_abort:
2832 xfs_bmap_cancel(free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10002833 xfs_trans_cancel(tp);
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002834 return error;
2835}
2836
2837/*
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002838 * xfs_rename_alloc_whiteout()
2839 *
2840 * Return a referenced, unlinked, unlocked inode that that can be used as a
2841 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2842 * crash between allocating the inode and linking it into the rename transaction
2843 * recovery will free the inode and we won't leak it.
2844 */
2845static int
2846xfs_rename_alloc_whiteout(
2847 struct xfs_inode *dp,
2848 struct xfs_inode **wip)
2849{
2850 struct xfs_inode *tmpfile;
2851 int error;
2852
2853 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2854 if (error)
2855 return error;
2856
Brian Foster22419ac2015-05-29 08:14:55 +10002857 /*
2858 * Prepare the tmpfile inode as if it were created through the VFS.
2859 * Otherwise, the link increment paths will complain about nlink 0->1.
2860 * Drop the link count as done by d_tmpfile(), complete the inode setup
2861 * and flag it as linkable.
2862 */
2863 drop_nlink(VFS_I(tmpfile));
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002864 xfs_finish_inode_setup(tmpfile);
2865 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2866
2867 *wip = tmpfile;
2868 return 0;
2869}
2870
2871/*
Dave Chinnerf6bba202013-08-12 20:49:46 +10002872 * xfs_rename
2873 */
2874int
2875xfs_rename(
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002876 struct xfs_inode *src_dp,
2877 struct xfs_name *src_name,
2878 struct xfs_inode *src_ip,
2879 struct xfs_inode *target_dp,
2880 struct xfs_name *target_name,
2881 struct xfs_inode *target_ip,
2882 unsigned int flags)
Dave Chinnerf6bba202013-08-12 20:49:46 +10002883{
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002884 struct xfs_mount *mp = src_dp->i_mount;
2885 struct xfs_trans *tp;
2886 struct xfs_bmap_free free_list;
2887 xfs_fsblock_t first_block;
2888 struct xfs_inode *wip = NULL; /* whiteout inode */
2889 struct xfs_inode *inodes[__XFS_SORT_INODES];
2890 int num_inodes = __XFS_SORT_INODES;
Dave Chinner2b936812015-03-25 15:12:30 +11002891 bool new_parent = (src_dp != target_dp);
2892 bool src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002893 int spaceres;
2894 int error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002895
2896 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2897
Dave Chinnereeacd322015-03-25 14:08:07 +11002898 if ((flags & RENAME_EXCHANGE) && !target_ip)
2899 return -EINVAL;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002900
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002901 /*
2902 * If we are doing a whiteout operation, allocate the whiteout inode
2903 * we will be placing at the target and ensure the type is set
2904 * appropriately.
2905 */
2906 if (flags & RENAME_WHITEOUT) {
2907 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2908 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2909 if (error)
2910 return error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002911
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002912 /* setup target dirent info as whiteout */
2913 src_name->type = XFS_DIR3_FT_CHRDEV;
2914 }
2915
2916 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
Dave Chinnerf6bba202013-08-12 20:49:46 +10002917 inodes, &num_inodes);
2918
Dave Chinnerf6bba202013-08-12 20:49:46 +10002919 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002920 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10002921 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
Dave Chinner24513372014-06-25 14:58:08 +10002922 if (error == -ENOSPC) {
Dave Chinnerf6bba202013-08-12 20:49:46 +10002923 spaceres = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10002924 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002925 }
Dave Chinner445883e2015-03-25 14:05:43 +11002926 if (error)
2927 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002928
2929 /*
2930 * Attach the dquots to the inodes
2931 */
2932 error = xfs_qm_vop_rename_dqattach(inodes);
Dave Chinner445883e2015-03-25 14:05:43 +11002933 if (error)
2934 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002935
2936 /*
2937 * Lock all the participating inodes. Depending upon whether
2938 * the target_name exists in the target directory, and
2939 * whether the target directory is the same as the source
2940 * directory, we can lock from 2 to 4 inodes.
2941 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002942 if (!new_parent)
2943 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2944 else
2945 xfs_lock_two_inodes(src_dp, target_dp,
2946 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2947
Dave Chinnerf6bba202013-08-12 20:49:46 +10002948 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2949
2950 /*
2951 * Join all the inodes to the transaction. From this point on,
2952 * we can rely on either trans_commit or trans_cancel to unlock
2953 * them.
2954 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002955 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002956 if (new_parent)
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002957 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002958 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2959 if (target_ip)
2960 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002961 if (wip)
2962 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002963
2964 /*
2965 * If we are using project inheritance, we only allow renames
2966 * into our tree when the project IDs are the same; else the
2967 * tree quota mechanism would be circumvented.
2968 */
2969 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2970 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
Dave Chinner24513372014-06-25 14:58:08 +10002971 error = -EXDEV;
Dave Chinner445883e2015-03-25 14:05:43 +11002972 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002973 }
2974
Dave Chinner445883e2015-03-25 14:05:43 +11002975 xfs_bmap_init(&free_list, &first_block);
2976
Dave Chinnereeacd322015-03-25 14:08:07 +11002977 /* RENAME_EXCHANGE is unique from here on. */
2978 if (flags & RENAME_EXCHANGE)
2979 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2980 target_dp, target_name, target_ip,
2981 &free_list, &first_block, spaceres);
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002982
2983 /*
Dave Chinnerf6bba202013-08-12 20:49:46 +10002984 * Set up the target.
2985 */
2986 if (target_ip == NULL) {
2987 /*
2988 * If there's no space reservation, check the entry will
2989 * fit before actually inserting it.
2990 */
Eric Sandeen94f3cad2014-09-09 11:57:52 +10002991 if (!spaceres) {
2992 error = xfs_dir_canenter(tp, target_dp, target_name);
2993 if (error)
Dave Chinner445883e2015-03-25 14:05:43 +11002994 goto out_trans_cancel;
Eric Sandeen94f3cad2014-09-09 11:57:52 +10002995 }
Dave Chinnerf6bba202013-08-12 20:49:46 +10002996 /*
2997 * If target does not exist and the rename crosses
2998 * directories, adjust the target directory link count
2999 * to account for the ".." reference from the new entry.
3000 */
3001 error = xfs_dir_createname(tp, target_dp, target_name,
3002 src_ip->i_ino, &first_block,
3003 &free_list, spaceres);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003004 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003005 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003006
3007 xfs_trans_ichgtime(tp, target_dp,
3008 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3009
3010 if (new_parent && src_is_directory) {
3011 error = xfs_bumplink(tp, target_dp);
3012 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003013 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003014 }
3015 } else { /* target_ip != NULL */
3016 /*
3017 * If target exists and it's a directory, check that both
3018 * target and source are directories and that target can be
3019 * destroyed, or that neither is a directory.
3020 */
3021 if (S_ISDIR(target_ip->i_d.di_mode)) {
3022 /*
3023 * Make sure target dir is empty.
3024 */
3025 if (!(xfs_dir_isempty(target_ip)) ||
3026 (target_ip->i_d.di_nlink > 2)) {
Dave Chinner24513372014-06-25 14:58:08 +10003027 error = -EEXIST;
Dave Chinner445883e2015-03-25 14:05:43 +11003028 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003029 }
3030 }
3031
3032 /*
3033 * Link the source inode under the target name.
3034 * If the source inode is a directory and we are moving
3035 * it across directories, its ".." entry will be
3036 * inconsistent until we replace that down below.
3037 *
3038 * In case there is already an entry with the same
3039 * name at the destination directory, remove it first.
3040 */
3041 error = xfs_dir_replace(tp, target_dp, target_name,
3042 src_ip->i_ino,
3043 &first_block, &free_list, spaceres);
3044 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003045 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003046
3047 xfs_trans_ichgtime(tp, target_dp,
3048 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3049
3050 /*
3051 * Decrement the link count on the target since the target
3052 * dir no longer points to it.
3053 */
3054 error = xfs_droplink(tp, target_ip);
3055 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003056 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003057
3058 if (src_is_directory) {
3059 /*
3060 * Drop the link from the old "." entry.
3061 */
3062 error = xfs_droplink(tp, target_ip);
3063 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003064 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003065 }
3066 } /* target_ip != NULL */
3067
3068 /*
3069 * Remove the source.
3070 */
3071 if (new_parent && src_is_directory) {
3072 /*
3073 * Rewrite the ".." entry to point to the new
3074 * directory.
3075 */
3076 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3077 target_dp->i_ino,
3078 &first_block, &free_list, spaceres);
Dave Chinner24513372014-06-25 14:58:08 +10003079 ASSERT(error != -EEXIST);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003080 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003081 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003082 }
3083
3084 /*
3085 * We always want to hit the ctime on the source inode.
3086 *
3087 * This isn't strictly required by the standards since the source
3088 * inode isn't really being changed, but old unix file systems did
3089 * it and some incremental backup programs won't work without it.
3090 */
3091 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3092 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3093
3094 /*
3095 * Adjust the link count on src_dp. This is necessary when
3096 * renaming a directory, either within one parent when
3097 * the target existed, or across two parent directories.
3098 */
3099 if (src_is_directory && (new_parent || target_ip != NULL)) {
3100
3101 /*
3102 * Decrement link count on src_directory since the
3103 * entry that's moved no longer points to it.
3104 */
3105 error = xfs_droplink(tp, src_dp);
3106 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003107 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003108 }
3109
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003110 /*
3111 * For whiteouts, we only need to update the source dirent with the
3112 * inode number of the whiteout inode rather than removing it
3113 * altogether.
3114 */
3115 if (wip) {
3116 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
Dave Chinnerf6bba202013-08-12 20:49:46 +10003117 &first_block, &free_list, spaceres);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003118 } else
3119 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3120 &first_block, &free_list, spaceres);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003121 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003122 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003123
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003124 /*
3125 * For whiteouts, we need to bump the link count on the whiteout inode.
3126 * This means that failures all the way up to this point leave the inode
3127 * on the unlinked list and so cleanup is a simple matter of dropping
3128 * the remaining reference to it. If we fail here after bumping the link
3129 * count, we're shutting down the filesystem so we'll never see the
3130 * intermediate state on disk.
3131 */
3132 if (wip) {
Brian Foster22419ac2015-05-29 08:14:55 +10003133 ASSERT(VFS_I(wip)->i_nlink == 0 && wip->i_d.di_nlink == 0);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003134 error = xfs_bumplink(tp, wip);
3135 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003136 goto out_bmap_cancel;
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003137 error = xfs_iunlink_remove(tp, wip);
3138 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003139 goto out_bmap_cancel;
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003140 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3141
3142 /*
3143 * Now we have a real link, clear the "I'm a tmpfile" state
3144 * flag from the inode so it doesn't accidentally get misused in
3145 * future.
3146 */
3147 VFS_I(wip)->i_state &= ~I_LINKABLE;
3148 }
Dave Chinnerf6bba202013-08-12 20:49:46 +10003149
3150 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3151 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3152 if (new_parent)
3153 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3154
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003155 error = xfs_finish_rename(tp, &free_list);
3156 if (wip)
3157 IRELE(wip);
3158 return error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003159
Dave Chinner445883e2015-03-25 14:05:43 +11003160out_bmap_cancel:
Dave Chinnerf6bba202013-08-12 20:49:46 +10003161 xfs_bmap_cancel(&free_list);
Dave Chinner445883e2015-03-25 14:05:43 +11003162out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10003163 xfs_trans_cancel(tp);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003164 if (wip)
3165 IRELE(wip);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003166 return error;
3167}
3168
David Chinnerbad55842008-03-06 13:43:49 +11003169STATIC int
3170xfs_iflush_cluster(
3171 xfs_inode_t *ip,
3172 xfs_buf_t *bp)
3173{
3174 xfs_mount_t *mp = ip->i_mount;
Dave Chinner5017e972010-01-11 11:47:40 +00003175 struct xfs_perag *pag;
David Chinnerbad55842008-03-06 13:43:49 +11003176 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10003177 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11003178 int ilist_size;
3179 xfs_inode_t **ilist;
3180 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11003181 int nr_found;
3182 int clcount = 0;
3183 int bufwasdelwri;
3184 int i;
3185
Dave Chinner5017e972010-01-11 11:47:40 +00003186 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
David Chinnerbad55842008-03-06 13:43:49 +11003187
Jie Liu0f49efd2013-12-13 15:51:48 +11003188 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
David Chinnerc8f5f122008-05-20 11:30:15 +10003189 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10003190 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11003191 if (!ilist)
Dave Chinner44b56e02010-01-11 11:47:43 +00003192 goto out_put;
David Chinnerbad55842008-03-06 13:43:49 +11003193
Jie Liu0f49efd2013-12-13 15:51:48 +11003194 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
David Chinnerbad55842008-03-06 13:43:49 +11003195 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003196 rcu_read_lock();
David Chinnerbad55842008-03-06 13:43:49 +11003197 /* really need a gang lookup range call here */
3198 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10003199 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11003200 if (nr_found == 0)
3201 goto out_free;
3202
3203 for (i = 0; i < nr_found; i++) {
3204 iq = ilist[i];
3205 if (iq == ip)
3206 continue;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003207
3208 /*
3209 * because this is an RCU protected lookup, we could find a
3210 * recently freed or even reallocated inode during the lookup.
3211 * We need to check under the i_flags_lock for a valid inode
3212 * here. Skip it if it is not valid or the wrong inode.
3213 */
3214 spin_lock(&ip->i_flags_lock);
3215 if (!ip->i_ino ||
3216 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3217 spin_unlock(&ip->i_flags_lock);
3218 continue;
3219 }
3220 spin_unlock(&ip->i_flags_lock);
3221
David Chinnerbad55842008-03-06 13:43:49 +11003222 /*
3223 * Do an un-protected check to see if the inode is dirty and
3224 * is a candidate for flushing. These checks will be repeated
3225 * later after the appropriate locks are acquired.
3226 */
David Chinner33540402008-03-06 13:43:59 +11003227 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11003228 continue;
David Chinnerbad55842008-03-06 13:43:49 +11003229
3230 /*
3231 * Try to get locks. If any are unavailable or it is pinned,
3232 * then this inode cannot be flushed and is skipped.
3233 */
3234
3235 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3236 continue;
3237 if (!xfs_iflock_nowait(iq)) {
3238 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3239 continue;
3240 }
3241 if (xfs_ipincount(iq)) {
3242 xfs_ifunlock(iq);
3243 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3244 continue;
3245 }
3246
3247 /*
3248 * arriving here means that this inode can be flushed. First
3249 * re-check that it's dirty before flushing.
3250 */
David Chinner33540402008-03-06 13:43:59 +11003251 if (!xfs_inode_clean(iq)) {
3252 int error;
David Chinnerbad55842008-03-06 13:43:49 +11003253 error = xfs_iflush_int(iq, bp);
3254 if (error) {
3255 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3256 goto cluster_corrupt_out;
3257 }
3258 clcount++;
3259 } else {
3260 xfs_ifunlock(iq);
3261 }
3262 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3263 }
3264
3265 if (clcount) {
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11003266 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3267 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
David Chinnerbad55842008-03-06 13:43:49 +11003268 }
3269
3270out_free:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003271 rcu_read_unlock();
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003272 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00003273out_put:
3274 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11003275 return 0;
3276
3277
3278cluster_corrupt_out:
3279 /*
3280 * Corruption detected in the clustering loop. Invalidate the
3281 * inode buffer and shut down the filesystem.
3282 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003283 rcu_read_unlock();
David Chinnerbad55842008-03-06 13:43:49 +11003284 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003285 * Clean up the buffer. If it was delwri, just release it --
David Chinnerbad55842008-03-06 13:43:49 +11003286 * brelse can handle it with no problems. If not, shut down the
3287 * filesystem before releasing the buffer.
3288 */
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003289 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
David Chinnerbad55842008-03-06 13:43:49 +11003290 if (bufwasdelwri)
3291 xfs_buf_relse(bp);
3292
3293 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3294
3295 if (!bufwasdelwri) {
3296 /*
3297 * Just like incore_relse: if we have b_iodone functions,
3298 * mark the buffer as an error and call them. Otherwise
3299 * mark it as stale and brelse.
3300 */
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02003301 if (bp->b_iodone) {
David Chinnerbad55842008-03-06 13:43:49 +11003302 XFS_BUF_UNDONE(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00003303 xfs_buf_stale(bp);
Dave Chinner24513372014-06-25 14:58:08 +10003304 xfs_buf_ioerror(bp, -EIO);
Dave Chinnere8aaba92014-10-02 09:04:22 +10003305 xfs_buf_ioend(bp);
David Chinnerbad55842008-03-06 13:43:49 +11003306 } else {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00003307 xfs_buf_stale(bp);
David Chinnerbad55842008-03-06 13:43:49 +11003308 xfs_buf_relse(bp);
3309 }
3310 }
3311
3312 /*
3313 * Unlocks the flush lock
3314 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003315 xfs_iflush_abort(iq, false);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003316 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00003317 xfs_perag_put(pag);
Dave Chinner24513372014-06-25 14:58:08 +10003318 return -EFSCORRUPTED;
David Chinnerbad55842008-03-06 13:43:49 +11003319}
3320
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321/*
Christoph Hellwig4c468192012-04-23 15:58:36 +10003322 * Flush dirty inode metadata into the backing buffer.
3323 *
3324 * The caller must have the inode lock and the inode flush lock held. The
3325 * inode lock will still be held upon return to the caller, and the inode
3326 * flush lock will be released after the inode has reached the disk.
3327 *
3328 * The caller must write out the buffer returned in *bpp and release it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 */
3330int
3331xfs_iflush(
Christoph Hellwig4c468192012-04-23 15:58:36 +10003332 struct xfs_inode *ip,
3333 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334{
Christoph Hellwig4c468192012-04-23 15:58:36 +10003335 struct xfs_mount *mp = ip->i_mount;
3336 struct xfs_buf *bp;
3337 struct xfs_dinode *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11003340 XFS_STATS_INC(mp, xs_iflush_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003342 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003343 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003345 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346
Christoph Hellwig4c468192012-04-23 15:58:36 +10003347 *bpp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 xfs_iunpin_wait(ip);
3350
3351 /*
Dave Chinner4b6a4682010-01-11 11:45:21 +00003352 * For stale inodes we cannot rely on the backing buffer remaining
3353 * stale in cache for the remaining life of the stale inode and so
Christoph Hellwig475ee412012-07-03 12:21:22 -04003354 * xfs_imap_to_bp() below may give us a buffer that no longer contains
Dave Chinner4b6a4682010-01-11 11:45:21 +00003355 * inodes below. We have to check this after ensuring the inode is
3356 * unpinned so that it is safe to reclaim the stale inode after the
3357 * flush call.
3358 */
3359 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3360 xfs_ifunlock(ip);
3361 return 0;
3362 }
3363
3364 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 * This may have been unpinned because the filesystem is shutting
3366 * down forcibly. If that's the case we must not write this inode
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003367 * to disk, because the log record didn't make it to disk.
3368 *
3369 * We also have to remove the log item from the AIL in this case,
3370 * as we wait for an empty AIL as part of the unmount process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 */
3372 if (XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10003373 error = -EIO;
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003374 goto abort_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 }
3376
3377 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003378 * Get the buffer containing the on-disk inode.
3379 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003380 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3381 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003382 if (error || !bp) {
3383 xfs_ifunlock(ip);
3384 return error;
3385 }
3386
3387 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388 * First flush out the inode that xfs_iflush was called with.
3389 */
3390 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11003391 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393
3394 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003395 * If the buffer is pinned then push on the log now so we won't
3396 * get stuck waiting in the write for too long.
3397 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00003398 if (xfs_buf_ispinned(bp))
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003399 xfs_log_force(mp, 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003400
3401 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 * inode clustering:
3403 * see if other inodes can be gathered into this write
3404 */
David Chinnerbad55842008-03-06 13:43:49 +11003405 error = xfs_iflush_cluster(ip, bp);
3406 if (error)
3407 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408
Christoph Hellwig4c468192012-04-23 15:58:36 +10003409 *bpp = bp;
3410 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411
3412corrupt_out:
3413 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10003414 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415cluster_corrupt_out:
Dave Chinner24513372014-06-25 14:58:08 +10003416 error = -EFSCORRUPTED;
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003417abort_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 /*
3419 * Unlocks the flush lock
3420 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003421 xfs_iflush_abort(ip, false);
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003422 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423}
3424
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425STATIC int
3426xfs_iflush_int(
Christoph Hellwig93848a92013-04-03 16:11:17 +11003427 struct xfs_inode *ip,
3428 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429{
Christoph Hellwig93848a92013-04-03 16:11:17 +11003430 struct xfs_inode_log_item *iip = ip->i_itemp;
3431 struct xfs_dinode *dip;
3432 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003434 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003435 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003437 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Christoph Hellwig93848a92013-04-03 16:11:17 +11003438 ASSERT(iip != NULL && iip->ili_fields != 0);
Dave Chinner263997a2014-05-20 07:46:40 +10003439 ASSERT(ip->i_d.di_version > 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441 /* set *dip = inode's place in the buffer */
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10003442 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003444 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003446 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3447 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3448 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 goto corrupt_out;
3450 }
3451 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3452 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003453 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3454 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3455 __func__, ip->i_ino, ip, ip->i_d.di_magic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456 goto corrupt_out;
3457 }
Al Viroabbede12011-07-26 02:31:30 -04003458 if (S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 if (XFS_TEST_ERROR(
3460 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3461 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3462 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003463 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3464 "%s: Bad regular inode %Lu, ptr 0x%p",
3465 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466 goto corrupt_out;
3467 }
Al Viroabbede12011-07-26 02:31:30 -04003468 } else if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469 if (XFS_TEST_ERROR(
3470 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3471 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3472 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3473 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003474 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3475 "%s: Bad directory inode %Lu, ptr 0x%p",
3476 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477 goto corrupt_out;
3478 }
3479 }
3480 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3481 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3482 XFS_RANDOM_IFLUSH_5)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003483 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3484 "%s: detected corrupt incore inode %Lu, "
3485 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3486 __func__, ip->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 ip->i_d.di_nextents + ip->i_d.di_anextents,
Dave Chinner6a19d932011-03-07 10:02:35 +11003488 ip->i_d.di_nblocks, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 goto corrupt_out;
3490 }
3491 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3492 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003493 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3494 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3495 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 goto corrupt_out;
3497 }
Dave Chinnere60896d2013-07-24 15:47:30 +10003498
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499 /*
Dave Chinner263997a2014-05-20 07:46:40 +10003500 * Inode item log recovery for v2 inodes are dependent on the
Dave Chinnere60896d2013-07-24 15:47:30 +10003501 * di_flushiter count for correct sequencing. We bump the flush
3502 * iteration count so we can detect flushes which postdate a log record
3503 * during recovery. This is redundant as we now log every change and
3504 * hence this can't happen but we need to still do it to ensure
3505 * backwards compatibility with old kernels that predate logging all
3506 * inode changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507 */
Dave Chinnere60896d2013-07-24 15:47:30 +10003508 if (ip->i_d.di_version < 3)
3509 ip->i_d.di_flushiter++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510
3511 /*
3512 * Copy the dirty parts of the inode into the on-disk
3513 * inode. We always copy out the core of the inode,
3514 * because if the inode is dirty at all the core must
3515 * be.
3516 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003517 xfs_dinode_to_disk(dip, &ip->i_d);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003518
3519 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3520 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3521 ip->i_d.di_flushiter = 0;
3522
Eric Sandeenfd9fdba2014-04-14 19:04:46 +10003523 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10003524 if (XFS_IFORK_Q(ip))
Eric Sandeenfd9fdba2014-04-14 19:04:46 +10003525 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526 xfs_inobp_check(mp, bp);
3527
3528 /*
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003529 * We've recorded everything logged in the inode, so we'd like to clear
3530 * the ili_fields bits so we don't log and flush things unnecessarily.
3531 * However, we can't stop logging all this information until the data
3532 * we've copied into the disk buffer is written to disk. If we did we
3533 * might overwrite the copy of the inode in the log with all the data
3534 * after re-logging only part of it, and in the face of a crash we
3535 * wouldn't have all the data we need to recover.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003537 * What we do is move the bits to the ili_last_fields field. When
3538 * logging the inode, these bits are moved back to the ili_fields field.
3539 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3540 * know that the information those bits represent is permanently on
3541 * disk. As long as the flush completes before the inode is logged
3542 * again, then both ili_fields and ili_last_fields will be cleared.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003544 * We can play with the ili_fields bits here, because the inode lock
3545 * must be held exclusively in order to set bits there and the flush
3546 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3547 * done routine can tell whether or not to look in the AIL. Also, store
3548 * the current LSN of the inode so that we can tell whether the item has
3549 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3550 * need the AIL lock, because it is a 64 bit value that cannot be read
3551 * atomically.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552 */
Christoph Hellwig93848a92013-04-03 16:11:17 +11003553 iip->ili_last_fields = iip->ili_fields;
3554 iip->ili_fields = 0;
Dave Chinnerfc0561c2015-11-03 13:14:59 +11003555 iip->ili_fsync_fields = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +11003556 iip->ili_logged = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557
Christoph Hellwig93848a92013-04-03 16:11:17 +11003558 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3559 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
Christoph Hellwig93848a92013-04-03 16:11:17 +11003561 /*
3562 * Attach the function xfs_iflush_done to the inode's
3563 * buffer. This will remove the inode from the AIL
3564 * and unlock the inode's flush lock when the inode is
3565 * completely written to disk.
3566 */
3567 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003568
Christoph Hellwig93848a92013-04-03 16:11:17 +11003569 /* update the lsn in the on disk inode if required */
3570 if (ip->i_d.di_version == 3)
3571 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572
Christoph Hellwig93848a92013-04-03 16:11:17 +11003573 /* generate the checksum. */
3574 xfs_dinode_calc_crc(mp, dip);
3575
3576 ASSERT(bp->b_fspriv != NULL);
3577 ASSERT(bp->b_iodone != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578 return 0;
3579
3580corrupt_out:
Dave Chinner24513372014-06-25 14:58:08 +10003581 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582}