blob: b82c729634f6ee30a0f1b19082394ea188f91b9c [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
Dave Chinner54d7b5c2016-02-09 16:54:58 +110060STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *);
61STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
62STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
Zhi Yong Wuab297432013-12-18 08:22:41 +080063
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(
Dave Chinner58f88ca2016-01-04 16:44:15 +1100613 __uint16_t di_flags,
614 uint64_t di_flags2,
615 bool has_attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
617 uint flags = 0;
618
619 if (di_flags & XFS_DIFLAG_ANY) {
620 if (di_flags & XFS_DIFLAG_REALTIME)
Dave Chinnere7b89482016-01-04 16:44:15 +1100621 flags |= FS_XFLAG_REALTIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (di_flags & XFS_DIFLAG_PREALLOC)
Dave Chinnere7b89482016-01-04 16:44:15 +1100623 flags |= FS_XFLAG_PREALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 if (di_flags & XFS_DIFLAG_IMMUTABLE)
Dave Chinnere7b89482016-01-04 16:44:15 +1100625 flags |= FS_XFLAG_IMMUTABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 if (di_flags & XFS_DIFLAG_APPEND)
Dave Chinnere7b89482016-01-04 16:44:15 +1100627 flags |= FS_XFLAG_APPEND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (di_flags & XFS_DIFLAG_SYNC)
Dave Chinnere7b89482016-01-04 16:44:15 +1100629 flags |= FS_XFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 if (di_flags & XFS_DIFLAG_NOATIME)
Dave Chinnere7b89482016-01-04 16:44:15 +1100631 flags |= FS_XFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 if (di_flags & XFS_DIFLAG_NODUMP)
Dave Chinnere7b89482016-01-04 16:44:15 +1100633 flags |= FS_XFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 if (di_flags & XFS_DIFLAG_RTINHERIT)
Dave Chinnere7b89482016-01-04 16:44:15 +1100635 flags |= FS_XFLAG_RTINHERIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (di_flags & XFS_DIFLAG_PROJINHERIT)
Dave Chinnere7b89482016-01-04 16:44:15 +1100637 flags |= FS_XFLAG_PROJINHERIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
Dave Chinnere7b89482016-01-04 16:44:15 +1100639 flags |= FS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100640 if (di_flags & XFS_DIFLAG_EXTSIZE)
Dave Chinnere7b89482016-01-04 16:44:15 +1100641 flags |= FS_XFLAG_EXTSIZE;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100642 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
Dave Chinnere7b89482016-01-04 16:44:15 +1100643 flags |= FS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000644 if (di_flags & XFS_DIFLAG_NODEFRAG)
Dave Chinnere7b89482016-01-04 16:44:15 +1100645 flags |= FS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000646 if (di_flags & XFS_DIFLAG_FILESTREAM)
Dave Chinnere7b89482016-01-04 16:44:15 +1100647 flags |= FS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 }
649
Dave Chinner58f88ca2016-01-04 16:44:15 +1100650 if (di_flags2 & XFS_DIFLAG2_ANY) {
651 if (di_flags2 & XFS_DIFLAG2_DAX)
652 flags |= FS_XFLAG_DAX;
653 }
654
655 if (has_attr)
656 flags |= FS_XFLAG_HASATTR;
657
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return flags;
659}
660
661uint
662xfs_ip2xflags(
Dave Chinner58f88ca2016-01-04 16:44:15 +1100663 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Dave Chinner58f88ca2016-01-04 16:44:15 +1100665 struct xfs_icdinode *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
Dave Chinner58f88ca2016-01-04 16:44:15 +1100667 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
669
670uint
671xfs_dic2xflags(
Dave Chinner58f88ca2016-01-04 16:44:15 +1100672 struct xfs_dinode *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
Dave Chinner58f88ca2016-01-04 16:44:15 +1100674 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags),
675 be64_to_cpu(dip->di_flags2), XFS_DFORK_Q(dip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678/*
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000679 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
680 * is allowed, otherwise it has to be an exact match. If a CI match is found,
681 * ci_name->name will point to a the actual name (caller must free) or
682 * will be set to NULL if an exact match is found.
683 */
684int
685xfs_lookup(
686 xfs_inode_t *dp,
687 struct xfs_name *name,
688 xfs_inode_t **ipp,
689 struct xfs_name *ci_name)
690{
691 xfs_ino_t inum;
692 int error;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000693
694 trace_xfs_lookup(dp, name);
695
696 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
Dave Chinner24513372014-06-25 14:58:08 +1000697 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000698
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000699 xfs_ilock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000700 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000701 if (error)
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000702 goto out_unlock;
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000703
704 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
705 if (error)
706 goto out_free_name;
707
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000708 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000709 return 0;
710
711out_free_name:
712 if (ci_name)
713 kmem_free(ci_name->name);
Dave Chinnerdbad7c92015-08-19 10:33:00 +1000714out_unlock:
715 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000716 *ipp = NULL;
717 return error;
718}
719
720/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 * Allocate an inode on disk and return a copy of its in-core version.
722 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
723 * appropriately within the inode. The uid and gid for the inode are
724 * set according to the contents of the given cred structure.
725 *
726 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300727 * has a free inode available, call xfs_iget() to obtain the in-core
728 * version of the allocated inode. Finally, fill in the inode and
729 * log its initial contents. In this case, ialloc_context would be
730 * set to NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 *
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300732 * If xfs_dialloc() does not have an available inode, it will replenish
733 * its supply by doing an allocation. Since we can only do one
734 * allocation within a transaction without deadlocks, we must commit
735 * the current transaction before returning the inode itself.
736 * In this case, therefore, we will set ialloc_context and return.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 * The caller should then commit the current transaction, start a new
738 * transaction, and call xfs_ialloc() again to actually get the inode.
739 *
740 * To ensure that some other process does not grab the inode that
741 * was allocated during the first call to xfs_ialloc(), this routine
742 * also returns the [locked] bp pointing to the head of the freelist
743 * as ialloc_context. The caller should hold this buffer across
744 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +1000745 *
746 * If we are allocating quota inodes, we do not have a parent inode
747 * to attach to or associate with (i.e. pip == NULL) because they
748 * are not linked into the directory structure - they are attached
749 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 */
751int
752xfs_ialloc(
753 xfs_trans_t *tp,
754 xfs_inode_t *pip,
Al Viro576b1d62011-07-26 02:50:15 -0400755 umode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700756 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 xfs_dev_t rdev,
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000758 prid_t prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 int okalloc,
760 xfs_buf_t **ialloc_context,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 xfs_inode_t **ipp)
762{
Christoph Hellwig93848a92013-04-03 16:11:17 +1100763 struct xfs_mount *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 xfs_ino_t ino;
765 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 uint flags;
767 int error;
Dave Chinnere076b0f2014-10-02 09:18:13 +1000768 struct timespec tv;
Dave Chinner39878482016-02-09 16:54:58 +1100769 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771 /*
772 * Call the space management code to pick
773 * the on-disk inode to be allocated.
774 */
David Chinnerb11f94d2007-07-11 11:09:33 +1000775 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Christoph Hellwig08358902012-07-04 10:54:47 -0400776 ialloc_context, &ino);
David Chinnerbf904242008-10-30 17:36:14 +1100777 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 return error;
Christoph Hellwig08358902012-07-04 10:54:47 -0400779 if (*ialloc_context || ino == NULLFSINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 *ipp = NULL;
781 return 0;
782 }
783 ASSERT(*ialloc_context == NULL);
784
785 /*
786 * Get the in-core inode with the lock held exclusively.
787 * This is because we're setting fields here we need
788 * to prevent others from looking at until we're done.
789 */
Christoph Hellwig93848a92013-04-03 16:11:17 +1100790 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
Christoph Hellwigec3ba852011-02-13 13:26:42 +0000791 XFS_ILOCK_EXCL, &ip);
David Chinnerbf904242008-10-30 17:36:14 +1100792 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 ASSERT(ip != NULL);
Dave Chinner39878482016-02-09 16:54:58 +1100795 inode = VFS_I(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Dave Chinner263997a2014-05-20 07:46:40 +1000797 /*
798 * We always convert v1 inodes to v2 now - we only support filesystems
799 * with >= v2 inode capability, so there is no reason for ever leaving
800 * an inode in v1 format.
801 */
802 if (ip->i_d.di_version == 1)
803 ip->i_d.di_version = 2;
804
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100805 inode->i_mode = mode;
Dave Chinner54d7b5c2016-02-09 16:54:58 +1100806 set_nlink(inode, nlink);
Dwight Engen7aab1b22013-08-15 14:08:01 -0400807 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
808 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000809 xfs_set_projid(ip, prid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000811 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 ip->i_d.di_gid = pip->i_d.di_gid;
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100813 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
814 inode->i_mode |= S_ISGID;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 }
816
817 /*
818 * If the group ID of the new file does not match the effective group
819 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
820 * (and only if the irix_sgid_inherit compatibility variable is set).
821 */
822 if ((irix_sgid_inherit) &&
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100823 (inode->i_mode & S_ISGID) &&
824 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid))))
825 inode->i_mode &= ~S_ISGID;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
827 ip->i_d.di_size = 0;
828 ip->i_d.di_nextents = 0;
829 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000830
Dave Chinnere076b0f2014-10-02 09:18:13 +1000831 tv = current_fs_time(mp->m_super);
Dave Chinner39878482016-02-09 16:54:58 +1100832 inode->i_mtime = tv;
833 inode->i_atime = tv;
834 inode->i_ctime = tv;
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 ip->i_d.di_extsize = 0;
837 ip->i_d.di_dmevmask = 0;
838 ip->i_d.di_dmstate = 0;
839 ip->i_d.di_flags = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100840
841 if (ip->i_d.di_version == 3) {
Dave Chinner83e06f22016-02-09 16:54:58 +1100842 inode->i_version = 1;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100843 ip->i_d.di_flags2 = 0;
Dave Chinner39878482016-02-09 16:54:58 +1100844 ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec;
845 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100846 }
847
848
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 flags = XFS_ILOG_CORE;
850 switch (mode & S_IFMT) {
851 case S_IFIFO:
852 case S_IFCHR:
853 case S_IFBLK:
854 case S_IFSOCK:
855 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
856 ip->i_df.if_u2.if_rdev = rdev;
857 ip->i_df.if_flags = 0;
858 flags |= XFS_ILOG_DEV;
859 break;
860 case S_IFREG:
861 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +1000862 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Dave Chinner58f88ca2016-01-04 16:44:15 +1100863 uint64_t di_flags2 = 0;
864 uint di_flags = 0;
Nathan Scott365ca832005-06-21 15:39:12 +1000865
Al Viroabbede12011-07-26 02:31:30 -0400866 if (S_ISDIR(mode)) {
Nathan Scott365ca832005-06-21 15:39:12 +1000867 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
868 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100869 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
870 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
871 ip->i_d.di_extsize = pip->i_d.di_extsize;
872 }
Dave Chinner9336e3a2014-10-02 09:18:40 +1000873 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
874 di_flags |= XFS_DIFLAG_PROJINHERIT;
Al Viroabbede12011-07-26 02:31:30 -0400875 } else if (S_ISREG(mode)) {
Christoph Hellwig613d7042007-10-11 17:44:08 +1000876 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +1000877 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100878 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
879 di_flags |= XFS_DIFLAG_EXTSIZE;
880 ip->i_d.di_extsize = pip->i_d.di_extsize;
881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 }
883 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
884 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +1000885 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
887 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +1000888 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
890 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +1000891 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
893 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +1000894 di_flags |= XFS_DIFLAG_NOSYMLINKS;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000895 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
896 xfs_inherit_nodefrag)
897 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000898 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
899 di_flags |= XFS_DIFLAG_FILESTREAM;
Dave Chinner58f88ca2016-01-04 16:44:15 +1100900 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
901 di_flags2 |= XFS_DIFLAG2_DAX;
902
Nathan Scott365ca832005-06-21 15:39:12 +1000903 ip->i_d.di_flags |= di_flags;
Dave Chinner58f88ca2016-01-04 16:44:15 +1100904 ip->i_d.di_flags2 |= di_flags2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 }
906 /* FALLTHROUGH */
907 case S_IFLNK:
908 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
909 ip->i_df.if_flags = XFS_IFEXTENTS;
910 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
911 ip->i_df.if_u1.if_extents = NULL;
912 break;
913 default:
914 ASSERT(0);
915 }
916 /*
917 * Attribute fork settings for new inode.
918 */
919 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
920 ip->i_d.di_anextents = 0;
921
922 /*
923 * Log the new values stuffed into the inode.
924 */
Christoph Hellwigddc34152011-09-19 15:00:54 +0000925 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 xfs_trans_log_inode(tp, ip, flags);
927
Dave Chinner58c90472015-02-23 22:38:08 +1100928 /* now that we have an i_mode we can setup the inode structure */
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000929 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
931 *ipp = ip;
932 return 0;
933}
934
Dave Chinnere546cb72013-08-12 20:49:47 +1000935/*
936 * Allocates a new inode from disk and return a pointer to the
937 * incore copy. This routine will internally commit the current
938 * transaction and allocate a new one if the Space Manager needed
939 * to do an allocation to replenish the inode free-list.
940 *
941 * This routine is designed to be called from xfs_create and
942 * xfs_create_dir.
943 *
944 */
945int
946xfs_dir_ialloc(
947 xfs_trans_t **tpp, /* input: current transaction;
948 output: may be a new transaction. */
949 xfs_inode_t *dp, /* directory within whose allocate
950 the inode. */
951 umode_t mode,
952 xfs_nlink_t nlink,
953 xfs_dev_t rdev,
954 prid_t prid, /* project id */
955 int okalloc, /* ok to allocate new space */
956 xfs_inode_t **ipp, /* pointer to inode; it will be
957 locked. */
958 int *committed)
959
960{
961 xfs_trans_t *tp;
Dave Chinnere546cb72013-08-12 20:49:47 +1000962 xfs_inode_t *ip;
963 xfs_buf_t *ialloc_context = NULL;
964 int code;
Dave Chinnere546cb72013-08-12 20:49:47 +1000965 void *dqinfo;
966 uint tflags;
967
968 tp = *tpp;
969 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
970
971 /*
972 * xfs_ialloc will return a pointer to an incore inode if
973 * the Space Manager has an available inode on the free
974 * list. Otherwise, it will do an allocation and replenish
975 * the freelist. Since we can only do one allocation per
976 * transaction without deadlocks, we will need to commit the
977 * current transaction and start a new one. We will then
978 * need to call xfs_ialloc again to get the inode.
979 *
980 * If xfs_ialloc did an allocation to replenish the freelist,
981 * it returns the bp containing the head of the freelist as
982 * ialloc_context. We will hold a lock on it across the
983 * transaction commit so that no other process can steal
984 * the inode(s) that we've just allocated.
985 */
986 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
987 &ialloc_context, &ip);
988
989 /*
990 * Return an error if we were unable to allocate a new inode.
991 * This should only happen if we run out of space on disk or
992 * encounter a disk error.
993 */
994 if (code) {
995 *ipp = NULL;
996 return code;
997 }
998 if (!ialloc_context && !ip) {
999 *ipp = NULL;
Dave Chinner24513372014-06-25 14:58:08 +10001000 return -ENOSPC;
Dave Chinnere546cb72013-08-12 20:49:47 +10001001 }
1002
1003 /*
1004 * If the AGI buffer is non-NULL, then we were unable to get an
1005 * inode in one operation. We need to commit the current
1006 * transaction and call xfs_ialloc() again. It is guaranteed
1007 * to succeed the second time.
1008 */
1009 if (ialloc_context) {
1010 /*
1011 * Normally, xfs_trans_commit releases all the locks.
1012 * We call bhold to hang on to the ialloc_context across
1013 * the commit. Holding this buffer prevents any other
1014 * processes from doing any allocations in this
1015 * allocation group.
1016 */
1017 xfs_trans_bhold(tp, ialloc_context);
Dave Chinnere546cb72013-08-12 20:49:47 +10001018
1019 /*
1020 * We want the quota changes to be associated with the next
1021 * transaction, NOT this one. So, detach the dqinfo from this
1022 * and attach it to the next transaction.
1023 */
1024 dqinfo = NULL;
1025 tflags = 0;
1026 if (tp->t_dqinfo) {
1027 dqinfo = (void *)tp->t_dqinfo;
1028 tp->t_dqinfo = NULL;
1029 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1030 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1031 }
1032
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001033 code = xfs_trans_roll(&tp, 0);
1034 if (committed != NULL)
Dave Chinnere546cb72013-08-12 20:49:47 +10001035 *committed = 1;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001036
Dave Chinnere546cb72013-08-12 20:49:47 +10001037 /*
1038 * Re-attach the quota info that we detached from prev trx.
1039 */
1040 if (dqinfo) {
1041 tp->t_dqinfo = dqinfo;
1042 tp->t_flags |= tflags;
1043 }
1044
1045 if (code) {
1046 xfs_buf_relse(ialloc_context);
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001047 *tpp = tp;
Dave Chinnere546cb72013-08-12 20:49:47 +10001048 *ipp = NULL;
1049 return code;
1050 }
1051 xfs_trans_bjoin(tp, ialloc_context);
1052
1053 /*
1054 * Call ialloc again. Since we've locked out all
1055 * other allocations in this allocation group,
1056 * this call should always succeed.
1057 */
1058 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1059 okalloc, &ialloc_context, &ip);
1060
1061 /*
1062 * If we get an error at this point, return to the caller
1063 * so that the current transaction can be aborted.
1064 */
1065 if (code) {
1066 *tpp = tp;
1067 *ipp = NULL;
1068 return code;
1069 }
1070 ASSERT(!ialloc_context && ip);
1071
1072 } else {
1073 if (committed != NULL)
1074 *committed = 0;
1075 }
1076
1077 *ipp = ip;
1078 *tpp = tp;
1079
1080 return 0;
1081}
1082
1083/*
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001084 * Decrement the link count on an inode & log the change. If this causes the
1085 * link count to go to zero, move the inode to AGI unlinked list so that it can
1086 * be freed when the last active reference goes away via xfs_inactive().
Dave Chinnere546cb72013-08-12 20:49:47 +10001087 */
1088int /* error */
1089xfs_droplink(
1090 xfs_trans_t *tp,
1091 xfs_inode_t *ip)
1092{
Dave Chinnere546cb72013-08-12 20:49:47 +10001093 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1094
Dave Chinnere546cb72013-08-12 20:49:47 +10001095 drop_nlink(VFS_I(ip));
1096 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1097
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001098 if (VFS_I(ip)->i_nlink)
1099 return 0;
1100
1101 return xfs_iunlink(tp, ip);
Dave Chinnere546cb72013-08-12 20:49:47 +10001102}
1103
1104/*
Dave Chinnere546cb72013-08-12 20:49:47 +10001105 * Increment the link count on an inode & log the change.
1106 */
1107int
1108xfs_bumplink(
1109 xfs_trans_t *tp,
1110 xfs_inode_t *ip)
1111{
1112 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1113
Dave Chinner263997a2014-05-20 07:46:40 +10001114 ASSERT(ip->i_d.di_version > 1);
Dave Chinnere546cb72013-08-12 20:49:47 +10001115 inc_nlink(VFS_I(ip));
Dave Chinnere546cb72013-08-12 20:49:47 +10001116 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1117 return 0;
1118}
1119
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001120int
1121xfs_create(
1122 xfs_inode_t *dp,
1123 struct xfs_name *name,
1124 umode_t mode,
1125 xfs_dev_t rdev,
1126 xfs_inode_t **ipp)
1127{
1128 int is_dir = S_ISDIR(mode);
1129 struct xfs_mount *mp = dp->i_mount;
1130 struct xfs_inode *ip = NULL;
1131 struct xfs_trans *tp = NULL;
1132 int error;
1133 xfs_bmap_free_t free_list;
1134 xfs_fsblock_t first_block;
1135 bool unlock_dp_on_error = false;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001136 prid_t prid;
1137 struct xfs_dquot *udqp = NULL;
1138 struct xfs_dquot *gdqp = NULL;
1139 struct xfs_dquot *pdqp = NULL;
Brian Foster062647a2014-11-28 14:00:16 +11001140 struct xfs_trans_res *tres;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001141 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001142
1143 trace_xfs_create(dp, name);
1144
1145 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001146 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001147
Zhi Yong Wu163467d2013-12-18 08:22:39 +08001148 prid = xfs_get_initial_prid(dp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001149
1150 /*
1151 * Make sure that we have allocated dquot(s) on disk.
1152 */
Dwight Engen7aab1b22013-08-15 14:08:01 -04001153 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1154 xfs_kgid_to_gid(current_fsgid()), prid,
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001155 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1156 &udqp, &gdqp, &pdqp);
1157 if (error)
1158 return error;
1159
1160 if (is_dir) {
1161 rdev = 0;
1162 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
Brian Foster062647a2014-11-28 14:00:16 +11001163 tres = &M_RES(mp)->tr_mkdir;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001164 } else {
1165 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
Brian Foster062647a2014-11-28 14:00:16 +11001166 tres = &M_RES(mp)->tr_create;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001167 }
1168
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001169 /*
1170 * Initially assume that the file does not exist and
1171 * reserve the resources for that case. If that is not
1172 * the case we'll drop the one we have and get a more
1173 * appropriate transaction later.
1174 */
Christoph Hellwig253f4912016-04-06 09:19:55 +10001175 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
Dave Chinner24513372014-06-25 14:58:08 +10001176 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001177 /* flush outstanding delalloc blocks and retry */
1178 xfs_flush_inodes(mp);
Christoph Hellwig253f4912016-04-06 09:19:55 +10001179 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001180 }
Dave Chinner24513372014-06-25 14:58:08 +10001181 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001182 /* No space at all so try a "no-allocation" reservation */
1183 resblks = 0;
Christoph Hellwig253f4912016-04-06 09:19:55 +10001184 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001185 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001186 if (error)
Christoph Hellwig253f4912016-04-06 09:19:55 +10001187 goto out_release_inode;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001188
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001189 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
1190 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001191 unlock_dp_on_error = true;
1192
1193 xfs_bmap_init(&free_list, &first_block);
1194
1195 /*
1196 * Reserve disk quota and the inode.
1197 */
1198 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1199 pdqp, resblks, 1, 0);
1200 if (error)
1201 goto out_trans_cancel;
1202
Eric Sandeen94f3cad2014-09-09 11:57:52 +10001203 if (!resblks) {
1204 error = xfs_dir_canenter(tp, dp, name);
1205 if (error)
1206 goto out_trans_cancel;
1207 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001208
1209 /*
1210 * A newly created regular or special file just has one directory
1211 * entry pointing to them, but a directory also the "." entry
1212 * pointing to itself.
1213 */
1214 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001215 prid, resblks > 0, &ip, NULL);
Jan Karad6077aa2015-07-29 11:52:08 +10001216 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001217 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001218
1219 /*
1220 * Now we join the directory inode to the transaction. We do not do it
1221 * earlier because xfs_dir_ialloc might commit the previous transaction
1222 * (and release all the locks). An error from here on will result in
1223 * the transaction cancel unlocking dp so don't do it explicitly in the
1224 * error path.
1225 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001226 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001227 unlock_dp_on_error = false;
1228
1229 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1230 &first_block, &free_list, resblks ?
1231 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1232 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10001233 ASSERT(error != -ENOSPC);
Christoph Hellwig4906e212015-06-04 13:47:56 +10001234 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001235 }
1236 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1237 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1238
1239 if (is_dir) {
1240 error = xfs_dir_init(tp, ip, dp);
1241 if (error)
1242 goto out_bmap_cancel;
1243
1244 error = xfs_bumplink(tp, dp);
1245 if (error)
1246 goto out_bmap_cancel;
1247 }
1248
1249 /*
1250 * If this is a synchronous mount, make sure that the
1251 * create transaction goes to disk before returning to
1252 * the user.
1253 */
1254 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1255 xfs_trans_set_sync(tp);
1256
1257 /*
1258 * Attach the dquot(s) to the inodes and modify them incore.
1259 * These ids of the inode couldn't have changed since the new
1260 * inode has been locked ever since it was created.
1261 */
1262 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1263
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001264 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001265 if (error)
1266 goto out_bmap_cancel;
1267
Christoph Hellwig70393312015-06-04 13:48:08 +10001268 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001269 if (error)
1270 goto out_release_inode;
1271
1272 xfs_qm_dqrele(udqp);
1273 xfs_qm_dqrele(gdqp);
1274 xfs_qm_dqrele(pdqp);
1275
1276 *ipp = ip;
1277 return 0;
1278
1279 out_bmap_cancel:
1280 xfs_bmap_cancel(&free_list);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001281 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001282 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001283 out_release_inode:
1284 /*
Dave Chinner58c90472015-02-23 22:38:08 +11001285 * Wait until after the current transaction is aborted to finish the
1286 * setup of the inode and release the inode. This prevents recursive
1287 * transactions and deadlocks from xfs_inactive.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001288 */
Dave Chinner58c90472015-02-23 22:38:08 +11001289 if (ip) {
1290 xfs_finish_inode_setup(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001291 IRELE(ip);
Dave Chinner58c90472015-02-23 22:38:08 +11001292 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001293
1294 xfs_qm_dqrele(udqp);
1295 xfs_qm_dqrele(gdqp);
1296 xfs_qm_dqrele(pdqp);
1297
1298 if (unlock_dp_on_error)
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001299 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001300 return error;
1301}
1302
1303int
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001304xfs_create_tmpfile(
1305 struct xfs_inode *dp,
1306 struct dentry *dentry,
Brian Foster330033d2014-04-17 08:15:30 +10001307 umode_t mode,
1308 struct xfs_inode **ipp)
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001309{
1310 struct xfs_mount *mp = dp->i_mount;
1311 struct xfs_inode *ip = NULL;
1312 struct xfs_trans *tp = NULL;
1313 int error;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001314 prid_t prid;
1315 struct xfs_dquot *udqp = NULL;
1316 struct xfs_dquot *gdqp = NULL;
1317 struct xfs_dquot *pdqp = NULL;
1318 struct xfs_trans_res *tres;
1319 uint resblks;
1320
1321 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001322 return -EIO;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001323
1324 prid = xfs_get_initial_prid(dp);
1325
1326 /*
1327 * Make sure that we have allocated dquot(s) on disk.
1328 */
1329 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1330 xfs_kgid_to_gid(current_fsgid()), prid,
1331 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1332 &udqp, &gdqp, &pdqp);
1333 if (error)
1334 return error;
1335
1336 resblks = XFS_IALLOC_SPACE_RES(mp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001337 tres = &M_RES(mp)->tr_create_tmpfile;
Christoph Hellwig253f4912016-04-06 09:19:55 +10001338
1339 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
Dave Chinner24513372014-06-25 14:58:08 +10001340 if (error == -ENOSPC) {
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001341 /* No space at all so try a "no-allocation" reservation */
1342 resblks = 0;
Christoph Hellwig253f4912016-04-06 09:19:55 +10001343 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001344 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001345 if (error)
Christoph Hellwig253f4912016-04-06 09:19:55 +10001346 goto out_release_inode;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001347
1348 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1349 pdqp, resblks, 1, 0);
1350 if (error)
1351 goto out_trans_cancel;
1352
1353 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1354 prid, resblks > 0, &ip, NULL);
Jan Karad6077aa2015-07-29 11:52:08 +10001355 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001356 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001357
1358 if (mp->m_flags & XFS_MOUNT_WSYNC)
1359 xfs_trans_set_sync(tp);
1360
1361 /*
1362 * Attach the dquot(s) to the inodes and modify them incore.
1363 * These ids of the inode couldn't have changed since the new
1364 * inode has been locked ever since it was created.
1365 */
1366 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1367
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001368 error = xfs_iunlink(tp, ip);
1369 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001370 goto out_trans_cancel;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001371
Christoph Hellwig70393312015-06-04 13:48:08 +10001372 error = xfs_trans_commit(tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001373 if (error)
1374 goto out_release_inode;
1375
1376 xfs_qm_dqrele(udqp);
1377 xfs_qm_dqrele(gdqp);
1378 xfs_qm_dqrele(pdqp);
1379
Brian Foster330033d2014-04-17 08:15:30 +10001380 *ipp = ip;
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001381 return 0;
1382
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001383 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001384 xfs_trans_cancel(tp);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001385 out_release_inode:
1386 /*
Dave Chinner58c90472015-02-23 22:38:08 +11001387 * Wait until after the current transaction is aborted to finish the
1388 * setup of the inode and release the inode. This prevents recursive
1389 * transactions and deadlocks from xfs_inactive.
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001390 */
Dave Chinner58c90472015-02-23 22:38:08 +11001391 if (ip) {
1392 xfs_finish_inode_setup(ip);
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001393 IRELE(ip);
Dave Chinner58c90472015-02-23 22:38:08 +11001394 }
Zhi Yong Wu99b64362013-12-18 08:22:40 +08001395
1396 xfs_qm_dqrele(udqp);
1397 xfs_qm_dqrele(gdqp);
1398 xfs_qm_dqrele(pdqp);
1399
1400 return error;
1401}
1402
1403int
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001404xfs_link(
1405 xfs_inode_t *tdp,
1406 xfs_inode_t *sip,
1407 struct xfs_name *target_name)
1408{
1409 xfs_mount_t *mp = tdp->i_mount;
1410 xfs_trans_t *tp;
1411 int error;
1412 xfs_bmap_free_t free_list;
1413 xfs_fsblock_t first_block;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001414 int resblks;
1415
1416 trace_xfs_link(tdp, target_name);
1417
Dave Chinnerc19b3b052016-02-09 16:54:58 +11001418 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001419
1420 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10001421 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001422
1423 error = xfs_qm_dqattach(sip, 0);
1424 if (error)
1425 goto std_return;
1426
1427 error = xfs_qm_dqattach(tdp, 0);
1428 if (error)
1429 goto std_return;
1430
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001431 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Christoph Hellwig253f4912016-04-06 09:19:55 +10001432 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
Dave Chinner24513372014-06-25 14:58:08 +10001433 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001434 resblks = 0;
Christoph Hellwig253f4912016-04-06 09:19:55 +10001435 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001436 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001437 if (error)
Christoph Hellwig253f4912016-04-06 09:19:55 +10001438 goto std_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001439
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001440 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001441 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1442
1443 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
Dave Chinnerdbad7c92015-08-19 10:33:00 +10001444 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001445
1446 /*
1447 * If we are using project inheritance, we only allow hard link
1448 * creation in our tree when the project IDs are the same; else
1449 * the tree quota mechanism could be circumvented.
1450 */
1451 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1452 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
Dave Chinner24513372014-06-25 14:58:08 +10001453 error = -EXDEV;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001454 goto error_return;
1455 }
1456
Eric Sandeen94f3cad2014-09-09 11:57:52 +10001457 if (!resblks) {
1458 error = xfs_dir_canenter(tp, tdp, target_name);
1459 if (error)
1460 goto error_return;
1461 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001462
1463 xfs_bmap_init(&free_list, &first_block);
1464
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001465 /*
1466 * Handle initial link state of O_TMPFILE inode
1467 */
1468 if (VFS_I(sip)->i_nlink == 0) {
Zhi Yong Wuab297432013-12-18 08:22:41 +08001469 error = xfs_iunlink_remove(tp, sip);
1470 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001471 goto error_return;
Zhi Yong Wuab297432013-12-18 08:22:41 +08001472 }
1473
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001474 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1475 &first_block, &free_list, resblks);
1476 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001477 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001478 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1479 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1480
1481 error = xfs_bumplink(tp, sip);
1482 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10001483 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001484
1485 /*
1486 * If this is a synchronous mount, make sure that the
1487 * link transaction goes to disk before returning to
1488 * the user.
1489 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001490 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001491 xfs_trans_set_sync(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001492
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001493 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001494 if (error) {
1495 xfs_bmap_cancel(&free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10001496 goto error_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001497 }
1498
Christoph Hellwig70393312015-06-04 13:48:08 +10001499 return xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001500
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001501 error_return:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001502 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001503 std_return:
1504 return error;
1505}
1506
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001508 * Free up the underlying blocks past new_size. The new size must be smaller
1509 * than the current size. This routine can be used both for the attribute and
1510 * data fork, and does not modify the inode size, which is left to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 *
David Chinnerf6485052008-04-17 16:50:04 +10001512 * The transaction passed to this routine must have made a permanent log
1513 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1514 * given transaction and start new ones, so make sure everything involved in
1515 * the transaction is tidy before calling here. Some transaction will be
1516 * returned to the caller to be committed. The incoming transaction must
1517 * already include the inode, and both inode locks must be held exclusively.
1518 * The inode must also be "held" within the transaction. On return the inode
1519 * will be "held" within the returned transaction. This routine does NOT
1520 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 *
David Chinnerf6485052008-04-17 16:50:04 +10001522 * If we get an error, we must return with the inode locked and linked into the
1523 * current transaction. This keeps things simple for the higher level code,
1524 * because it always knows that the inode is locked and held in the transaction
1525 * that returns to it whether errors occur or not. We don't mark the inode
1526 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 */
1528int
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001529xfs_itruncate_extents(
1530 struct xfs_trans **tpp,
1531 struct xfs_inode *ip,
1532 int whichfork,
1533 xfs_fsize_t new_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001535 struct xfs_mount *mp = ip->i_mount;
1536 struct xfs_trans *tp = *tpp;
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001537 xfs_bmap_free_t free_list;
1538 xfs_fsblock_t first_block;
1539 xfs_fileoff_t first_unmap_block;
1540 xfs_fileoff_t last_block;
1541 xfs_filblks_t unmap_len;
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001542 int error = 0;
1543 int done = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
Christoph Hellwig0b561852012-07-04 11:13:31 -04001545 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1546 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1547 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001548 ASSERT(new_size <= XFS_ISIZE(ip));
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001549 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 ASSERT(ip->i_itemp != NULL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001551 ASSERT(ip->i_itemp->ili_lock_flags == 0);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001552 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001554 trace_xfs_itruncate_extents_start(ip, new_size);
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 /*
1557 * Since it is possible for space to become allocated beyond
1558 * the end of the file (in a crash where the space is allocated
1559 * but the inode size is not yet updated), simply remove any
1560 * blocks which show up between the new EOF and the maximum
1561 * possible file size. If the first block to be removed is
1562 * beyond the maximum file size (ie it is the same as last_block),
1563 * then there is nothing to do.
1564 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001565 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
Dave Chinner32972382012-06-08 15:44:54 +10001566 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001567 if (first_unmap_block == last_block)
1568 return 0;
1569
1570 ASSERT(first_unmap_block < last_block);
1571 unmap_len = last_block - first_unmap_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 while (!done) {
Eric Sandeen9d87c312009-01-14 23:22:07 -06001573 xfs_bmap_init(&free_list, &first_block);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001574 error = xfs_bunmapi(tp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001575 first_unmap_block, unmap_len,
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001576 xfs_bmapi_aflag(whichfork),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001578 &first_block, &free_list,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001579 &done);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001580 if (error)
1581 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
1583 /*
1584 * Duplicate the transaction that has the permanent
1585 * reservation and commit the old transaction.
1586 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001587 error = xfs_bmap_finish(&tp, &free_list, ip);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001588 if (error)
1589 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
Christoph Hellwig2e6db6c2015-06-04 13:47:29 +10001591 error = xfs_trans_roll(&tp, ip);
David Chinnerf6485052008-04-17 16:50:04 +10001592 if (error)
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001593 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 }
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001595
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001596 /*
1597 * Always re-log the inode so that our permanent transaction can keep
1598 * on rolling it forward in the log.
1599 */
1600 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1601
1602 trace_xfs_itruncate_extents_end(ip, new_size);
1603
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001604out:
1605 *tpp = tp;
1606 return error;
1607out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 /*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001609 * If the bunmapi call encounters an error, return to the caller where
1610 * the transaction can be properly aborted. We just need to make sure
1611 * we're not holding any resources that we were not when we came in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001613 xfs_bmap_cancel(&free_list);
1614 goto out;
1615}
1616
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001617int
1618xfs_release(
1619 xfs_inode_t *ip)
1620{
1621 xfs_mount_t *mp = ip->i_mount;
1622 int error;
1623
Dave Chinnerc19b3b052016-02-09 16:54:58 +11001624 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001625 return 0;
1626
1627 /* If this is a read-only mount, don't do this (would generate I/O) */
1628 if (mp->m_flags & XFS_MOUNT_RDONLY)
1629 return 0;
1630
1631 if (!XFS_FORCED_SHUTDOWN(mp)) {
1632 int truncated;
1633
1634 /*
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001635 * If we previously truncated this file and removed old data
1636 * in the process, we want to initiate "early" writeout on
1637 * the last close. This is an attempt to combat the notorious
1638 * NULL files problem which is particularly noticeable from a
1639 * truncate down, buffered (re-)write (delalloc), followed by
1640 * a crash. What we are effectively doing here is
1641 * significantly reducing the time window where we'd otherwise
1642 * be exposed to that problem.
1643 */
1644 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1645 if (truncated) {
1646 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
Dave Chinnereac152b2014-08-04 13:22:49 +10001647 if (ip->i_delayed_blks > 0) {
Dave Chinner24513372014-06-25 14:58:08 +10001648 error = filemap_flush(VFS_I(ip)->i_mapping);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001649 if (error)
1650 return error;
1651 }
1652 }
1653 }
1654
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001655 if (VFS_I(ip)->i_nlink == 0)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001656 return 0;
1657
1658 if (xfs_can_free_eofblocks(ip, false)) {
1659
1660 /*
1661 * If we can't get the iolock just skip truncating the blocks
1662 * past EOF because we could deadlock with the mmap_sem
1663 * otherwise. We'll get another chance to drop them once the
1664 * last reference to the inode is dropped, so we'll never leak
1665 * blocks permanently.
1666 *
1667 * Further, check if the inode is being opened, written and
1668 * closed frequently and we have delayed allocation blocks
1669 * outstanding (e.g. streaming writes from the NFS server),
1670 * truncating the blocks past EOF will cause fragmentation to
1671 * occur.
1672 *
1673 * In this case don't do the truncation, either, but we have to
1674 * be careful how we detect this case. Blocks beyond EOF show
1675 * up as i_delayed_blks even when the inode is clean, so we
1676 * need to truncate them away first before checking for a dirty
1677 * release. Hence on the first dirty close we will still remove
1678 * the speculative allocation, but after that we will leave it
1679 * in place.
1680 */
1681 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1682 return 0;
1683
1684 error = xfs_free_eofblocks(mp, ip, true);
Dave Chinner24513372014-06-25 14:58:08 +10001685 if (error && error != -EAGAIN)
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001686 return error;
1687
1688 /* delalloc blocks after truncation means it really is dirty */
1689 if (ip->i_delayed_blks)
1690 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1691 }
1692 return 0;
1693}
1694
1695/*
Brian Fosterf7be2d72013-09-20 11:06:10 -04001696 * xfs_inactive_truncate
1697 *
1698 * Called to perform a truncate when an inode becomes unlinked.
1699 */
1700STATIC int
1701xfs_inactive_truncate(
1702 struct xfs_inode *ip)
1703{
1704 struct xfs_mount *mp = ip->i_mount;
1705 struct xfs_trans *tp;
1706 int error;
1707
Christoph Hellwig253f4912016-04-06 09:19:55 +10001708 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001709 if (error) {
1710 ASSERT(XFS_FORCED_SHUTDOWN(mp));
Brian Fosterf7be2d72013-09-20 11:06:10 -04001711 return error;
1712 }
1713
1714 xfs_ilock(ip, XFS_ILOCK_EXCL);
1715 xfs_trans_ijoin(tp, ip, 0);
1716
1717 /*
1718 * Log the inode size first to prevent stale data exposure in the event
1719 * of a system crash before the truncate completes. See the related
1720 * comment in xfs_setattr_size() for details.
1721 */
1722 ip->i_d.di_size = 0;
1723 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1724
1725 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1726 if (error)
1727 goto error_trans_cancel;
1728
1729 ASSERT(ip->i_d.di_nextents == 0);
1730
Christoph Hellwig70393312015-06-04 13:48:08 +10001731 error = xfs_trans_commit(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001732 if (error)
1733 goto error_unlock;
1734
1735 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1736 return 0;
1737
1738error_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10001739 xfs_trans_cancel(tp);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001740error_unlock:
1741 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1742 return error;
1743}
1744
1745/*
Brian Foster88877d22013-09-20 11:06:11 -04001746 * xfs_inactive_ifree()
1747 *
1748 * Perform the inode free when an inode is unlinked.
1749 */
1750STATIC int
1751xfs_inactive_ifree(
1752 struct xfs_inode *ip)
1753{
1754 xfs_bmap_free_t free_list;
1755 xfs_fsblock_t first_block;
Brian Foster88877d22013-09-20 11:06:11 -04001756 struct xfs_mount *mp = ip->i_mount;
1757 struct xfs_trans *tp;
1758 int error;
1759
Brian Foster9d43b182014-04-24 16:00:52 +10001760 /*
1761 * The ifree transaction might need to allocate blocks for record
1762 * insertion to the finobt. We don't want to fail here at ENOSPC, so
1763 * allow ifree to dip into the reserved block pool if necessary.
1764 *
1765 * Freeing large sets of inodes generally means freeing inode chunks,
1766 * directory and file data blocks, so this should be relatively safe.
1767 * Only under severe circumstances should it be possible to free enough
1768 * inodes to exhaust the reserve block pool via finobt expansion while
1769 * at the same time not creating free space in the filesystem.
1770 *
1771 * Send a warning if the reservation does happen to fail, as the inode
1772 * now remains allocated and sits on the unlinked list until the fs is
1773 * repaired.
1774 */
Christoph Hellwig253f4912016-04-06 09:19:55 +10001775 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1776 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
Brian Foster88877d22013-09-20 11:06:11 -04001777 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10001778 if (error == -ENOSPC) {
Brian Foster9d43b182014-04-24 16:00:52 +10001779 xfs_warn_ratelimited(mp,
1780 "Failed to remove inode(s) from unlinked list. "
1781 "Please free space, unmount and run xfs_repair.");
1782 } else {
1783 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1784 }
Brian Foster88877d22013-09-20 11:06:11 -04001785 return error;
1786 }
1787
1788 xfs_ilock(ip, XFS_ILOCK_EXCL);
1789 xfs_trans_ijoin(tp, ip, 0);
1790
1791 xfs_bmap_init(&free_list, &first_block);
1792 error = xfs_ifree(tp, ip, &free_list);
1793 if (error) {
1794 /*
1795 * If we fail to free the inode, shut down. The cancel
1796 * might do that, we need to make sure. Otherwise the
1797 * inode might be lost for a long time or forever.
1798 */
1799 if (!XFS_FORCED_SHUTDOWN(mp)) {
1800 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1801 __func__, error);
1802 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1803 }
Christoph Hellwig4906e212015-06-04 13:47:56 +10001804 xfs_trans_cancel(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001805 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1806 return error;
1807 }
1808
1809 /*
1810 * Credit the quota account(s). The inode is gone.
1811 */
1812 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1813
1814 /*
Brian Fosterd4a97a02015-08-19 10:01:40 +10001815 * Just ignore errors at this point. There is nothing we can do except
1816 * to try to keep going. Make sure it's not a silent error.
Brian Foster88877d22013-09-20 11:06:11 -04001817 */
Eric Sandeenf6106ef2016-01-11 11:34:01 +11001818 error = xfs_bmap_finish(&tp, &free_list, NULL);
Brian Fosterd4a97a02015-08-19 10:01:40 +10001819 if (error) {
Brian Foster88877d22013-09-20 11:06:11 -04001820 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1821 __func__, error);
Brian Fosterd4a97a02015-08-19 10:01:40 +10001822 xfs_bmap_cancel(&free_list);
1823 }
Christoph Hellwig70393312015-06-04 13:48:08 +10001824 error = xfs_trans_commit(tp);
Brian Foster88877d22013-09-20 11:06:11 -04001825 if (error)
1826 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1827 __func__, error);
1828
1829 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1830 return 0;
1831}
1832
1833/*
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001834 * xfs_inactive
1835 *
1836 * This is called when the vnode reference count for the vnode
1837 * goes to zero. If the file has been unlinked, then it must
1838 * now be truncated. Also, we clear all of the read-ahead state
1839 * kept for the inode here since the file is now closed.
1840 */
Brian Foster74564fb2013-09-20 11:06:12 -04001841void
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001842xfs_inactive(
1843 xfs_inode_t *ip)
1844{
Jie Liu3d3c8b52013-08-12 20:49:59 +10001845 struct xfs_mount *mp;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001846 int error;
1847 int truncate = 0;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001848
1849 /*
1850 * If the inode is already free, then there can be nothing
1851 * to clean up here.
1852 */
Dave Chinnerc19b3b052016-02-09 16:54:58 +11001853 if (VFS_I(ip)->i_mode == 0) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001854 ASSERT(ip->i_df.if_real_bytes == 0);
1855 ASSERT(ip->i_df.if_broot_bytes == 0);
Brian Foster74564fb2013-09-20 11:06:12 -04001856 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001857 }
1858
1859 mp = ip->i_mount;
1860
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001861 /* If this is a read-only mount, don't do this (would generate I/O) */
1862 if (mp->m_flags & XFS_MOUNT_RDONLY)
Brian Foster74564fb2013-09-20 11:06:12 -04001863 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001864
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001865 if (VFS_I(ip)->i_nlink != 0) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001866 /*
1867 * force is true because we are evicting an inode from the
1868 * cache. Post-eof blocks must be freed, lest we end up with
1869 * broken free space accounting.
1870 */
Brian Foster74564fb2013-09-20 11:06:12 -04001871 if (xfs_can_free_eofblocks(ip, true))
1872 xfs_free_eofblocks(mp, ip, false);
1873
1874 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001875 }
1876
Dave Chinnerc19b3b052016-02-09 16:54:58 +11001877 if (S_ISREG(VFS_I(ip)->i_mode) &&
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001878 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1879 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1880 truncate = 1;
1881
1882 error = xfs_qm_dqattach(ip, 0);
1883 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001884 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001885
Dave Chinnerc19b3b052016-02-09 16:54:58 +11001886 if (S_ISLNK(VFS_I(ip)->i_mode))
Brian Foster36b21dd2013-09-20 11:06:09 -04001887 error = xfs_inactive_symlink(ip);
Brian Fosterf7be2d72013-09-20 11:06:10 -04001888 else if (truncate)
1889 error = xfs_inactive_truncate(ip);
1890 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001891 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001892
1893 /*
1894 * If there are attributes associated with the file then blow them away
1895 * now. The code calls a routine that recursively deconstructs the
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001896 * attribute fork. If also blows away the in-core attribute fork.
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001897 */
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001898 if (XFS_IFORK_Q(ip)) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001899 error = xfs_attr_inactive(ip);
1900 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001901 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001902 }
1903
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001904 ASSERT(!ip->i_afp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001905 ASSERT(ip->i_d.di_anextents == 0);
Dave Chinner6dfe5a02015-05-29 07:40:08 +10001906 ASSERT(ip->i_d.di_forkoff == 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001907
1908 /*
1909 * Free the inode.
1910 */
Brian Foster88877d22013-09-20 11:06:11 -04001911 error = xfs_inactive_ifree(ip);
1912 if (error)
Brian Foster74564fb2013-09-20 11:06:12 -04001913 return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001914
1915 /*
1916 * Release the dquots held by inode, if any.
1917 */
1918 xfs_qm_dqdetach(ip);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001919}
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921/*
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001922 * This is called when the inode's link count goes to 0 or we are creating a
1923 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be
1924 * set to true as the link count is dropped to zero by the VFS after we've
1925 * created the file successfully, so we have to add it to the unlinked list
1926 * while the link count is non-zero.
1927 *
1928 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1929 * list when the inode is freed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 */
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001931STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932xfs_iunlink(
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001933 struct xfs_trans *tp,
1934 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935{
Dave Chinner54d7b5c2016-02-09 16:54:58 +11001936 xfs_mount_t *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 xfs_agi_t *agi;
1938 xfs_dinode_t *dip;
1939 xfs_buf_t *agibp;
1940 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 xfs_agino_t agino;
1942 short bucket_index;
1943 int offset;
1944 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
Dave Chinnerc19b3b052016-02-09 16:54:58 +11001946 ASSERT(VFS_I(ip)->i_mode != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 /*
1949 * Get the agi buffer first. It ensures lock ordering
1950 * on the list.
1951 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001952 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001953 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001956
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 /*
1958 * Get the index into the agi hash table for the
1959 * list this inode will go on.
1960 */
1961 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1962 ASSERT(agino != 0);
1963 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1964 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001965 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001967 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 /*
1969 * There is already another inode in the bucket we need
1970 * to add ourselves to. Add us at the front of the list.
1971 * Here we put the head pointer into our next pointer,
1972 * and then we fall through to point the head at us.
1973 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04001974 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1975 0, 0);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001976 if (error)
1977 return error;
1978
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001979 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001981 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10001983
1984 /* need to recalc the inode CRC if appropriate */
1985 xfs_dinode_calc_crc(mp, dip);
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 xfs_trans_inode_buf(tp, ibp);
1988 xfs_trans_log_buf(tp, ibp, offset,
1989 (offset + sizeof(xfs_agino_t) - 1));
1990 xfs_inobp_check(mp, ibp);
1991 }
1992
1993 /*
1994 * Point the bucket head pointer at the inode being inserted.
1995 */
1996 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001997 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 offset = offsetof(xfs_agi_t, agi_unlinked) +
1999 (sizeof(xfs_agino_t) * bucket_index);
Dave Chinnerf19b8722015-01-22 09:29:40 +11002000 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 xfs_trans_log_buf(tp, agibp, offset,
2002 (offset + sizeof(xfs_agino_t) - 1));
2003 return 0;
2004}
2005
2006/*
2007 * Pull the on-disk inode from the AGI unlinked list.
2008 */
2009STATIC int
2010xfs_iunlink_remove(
2011 xfs_trans_t *tp,
2012 xfs_inode_t *ip)
2013{
2014 xfs_ino_t next_ino;
2015 xfs_mount_t *mp;
2016 xfs_agi_t *agi;
2017 xfs_dinode_t *dip;
2018 xfs_buf_t *agibp;
2019 xfs_buf_t *ibp;
2020 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 xfs_agino_t agino;
2022 xfs_agino_t next_agino;
2023 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002024 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10002026 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031
2032 /*
2033 * Get the agi buffer first. It ensures lock ordering
2034 * on the list.
2035 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002036 error = xfs_read_agi(mp, tp, agno, &agibp);
2037 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002039
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002041
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 /*
2043 * Get the index into the agi hash table for the
2044 * list this inode will go on.
2045 */
2046 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2047 ASSERT(agino != 0);
2048 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002049 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 ASSERT(agi->agi_unlinked[bucket_index]);
2051
Christoph Hellwig16259e72005-11-02 15:11:25 +11002052 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002054 * We're at the head of the list. Get the inode's on-disk
2055 * buffer to see if there is anyone after us on the list.
2056 * Only modify our next pointer if it is not already NULLAGINO.
2057 * This saves us the overhead of dealing with the buffer when
2058 * there is no need to change it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002060 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2061 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002063 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002064 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 return error;
2066 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002067 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 ASSERT(next_agino != 0);
2069 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002070 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002071 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002073
2074 /* need to recalc the inode CRC if appropriate */
2075 xfs_dinode_calc_crc(mp, dip);
2076
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 xfs_trans_inode_buf(tp, ibp);
2078 xfs_trans_log_buf(tp, ibp, offset,
2079 (offset + sizeof(xfs_agino_t) - 1));
2080 xfs_inobp_check(mp, ibp);
2081 } else {
2082 xfs_trans_brelse(tp, ibp);
2083 }
2084 /*
2085 * Point the bucket head pointer at the next inode.
2086 */
2087 ASSERT(next_agino != 0);
2088 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002089 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 offset = offsetof(xfs_agi_t, agi_unlinked) +
2091 (sizeof(xfs_agino_t) * bucket_index);
Dave Chinnerf19b8722015-01-22 09:29:40 +11002092 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 xfs_trans_log_buf(tp, agibp, offset,
2094 (offset + sizeof(xfs_agino_t) - 1));
2095 } else {
2096 /*
2097 * We need to search the list for the inode being freed.
2098 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002099 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 last_ibp = NULL;
2101 while (next_agino != agino) {
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002102 struct xfs_imap imap;
2103
2104 if (last_ibp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 xfs_trans_brelse(tp, last_ibp);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002106
2107 imap.im_blkno = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002109
2110 error = xfs_imap(mp, tp, next_ino, &imap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11002112 xfs_warn(mp,
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002113 "%s: xfs_imap returned error %d.",
2114 __func__, error);
2115 return error;
2116 }
2117
2118 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2119 &last_ibp, 0, 0);
2120 if (error) {
2121 xfs_warn(mp,
2122 "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002123 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 return error;
2125 }
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002126
2127 last_offset = imap.im_boffset;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002128 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 ASSERT(next_agino != NULLAGINO);
2130 ASSERT(next_agino != 0);
2131 }
Christoph Hellwig475ee412012-07-03 12:21:22 -04002132
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002134 * Now last_ibp points to the buffer previous to us on the
2135 * unlinked list. Pull us from the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002137 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2138 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002140 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002141 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 return error;
2143 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002144 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 ASSERT(next_agino != 0);
2146 ASSERT(next_agino != agino);
2147 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002148 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002149 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002151
2152 /* need to recalc the inode CRC if appropriate */
2153 xfs_dinode_calc_crc(mp, dip);
2154
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 xfs_trans_inode_buf(tp, ibp);
2156 xfs_trans_log_buf(tp, ibp, offset,
2157 (offset + sizeof(xfs_agino_t) - 1));
2158 xfs_inobp_check(mp, ibp);
2159 } else {
2160 xfs_trans_brelse(tp, ibp);
2161 }
2162 /*
2163 * Point the previous inode on the list to the next inode.
2164 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002165 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 ASSERT(next_agino != 0);
2167 offset = last_offset + 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, last_dip);
2171
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 xfs_trans_inode_buf(tp, last_ibp);
2173 xfs_trans_log_buf(tp, last_ibp, offset,
2174 (offset + sizeof(xfs_agino_t) - 1));
2175 xfs_inobp_check(mp, last_ibp);
2176 }
2177 return 0;
2178}
2179
Dave Chinner5b3eed72010-08-24 11:42:41 +10002180/*
Zhi Yong Wu0b8182d2013-08-12 03:14:59 +00002181 * A big issue when freeing the inode cluster is that we _cannot_ skip any
Dave Chinner5b3eed72010-08-24 11:42:41 +10002182 * inodes that are in memory - they all must be marked stale and attached to
2183 * the cluster buffer.
2184 */
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002185STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186xfs_ifree_cluster(
Brian Foster09b56602015-05-29 09:26:03 +10002187 xfs_inode_t *free_ip,
2188 xfs_trans_t *tp,
2189 struct xfs_icluster *xic)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
2191 xfs_mount_t *mp = free_ip->i_mount;
2192 int blks_per_cluster;
Jie Liu982e9392013-12-13 15:51:49 +11002193 int inodes_per_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 int nbufs;
Dave Chinner5b257b42010-06-03 16:22:29 +10002195 int i, j;
Brian Foster3cdaa182015-06-04 13:03:34 +10002196 int ioffset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 xfs_daddr_t blkno;
2198 xfs_buf_t *bp;
Dave Chinner5b257b42010-06-03 16:22:29 +10002199 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 xfs_inode_log_item_t *iip;
2201 xfs_log_item_t *lip;
Dave Chinner5017e972010-01-11 11:47:40 +00002202 struct xfs_perag *pag;
Brian Foster09b56602015-05-29 09:26:03 +10002203 xfs_ino_t inum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
Brian Foster09b56602015-05-29 09:26:03 +10002205 inum = xic->first_ino;
Dave Chinner5017e972010-01-11 11:47:40 +00002206 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
Jie Liu982e9392013-12-13 15:51:49 +11002207 blks_per_cluster = xfs_icluster_size_fsb(mp);
2208 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2209 nbufs = mp->m_ialloc_blks / blks_per_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
Jie Liu982e9392013-12-13 15:51:49 +11002211 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
Brian Foster09b56602015-05-29 09:26:03 +10002212 /*
2213 * The allocation bitmap tells us which inodes of the chunk were
2214 * physically allocated. Skip the cluster if an inode falls into
2215 * a sparse region.
2216 */
Brian Foster3cdaa182015-06-04 13:03:34 +10002217 ioffset = inum - xic->first_ino;
2218 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2219 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0);
Brian Foster09b56602015-05-29 09:26:03 +10002220 continue;
2221 }
2222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2224 XFS_INO_TO_AGBNO(mp, inum));
2225
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 /*
Dave Chinner5b257b42010-06-03 16:22:29 +10002227 * We obtain and lock the backing buffer first in the process
2228 * here, as we have to ensure that any dirty inode that we
2229 * can't get the flush lock on is attached to the buffer.
2230 * If we scan the in-memory inodes first, then buffer IO can
2231 * complete before we get a lock on it, and hence we may fail
2232 * to mark all the active inodes on the buffer stale.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002234 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
Dave Chinnerb6aff292012-11-02 11:38:42 +11002235 mp->m_bsize * blks_per_cluster,
2236 XBF_UNMAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002238 if (!bp)
Dave Chinner24513372014-06-25 14:58:08 +10002239 return -ENOMEM;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002240
2241 /*
2242 * This buffer may not have been correctly initialised as we
2243 * didn't read it from disk. That's not important because we are
2244 * only using to mark the buffer as stale in the log, and to
2245 * attach stale cached inodes on it. That means it will never be
2246 * dispatched for IO. If it is, we want to know about it, and we
2247 * want it to fail. We can acheive this by adding a write
2248 * verifier to the buffer.
2249 */
Dave Chinner1813dd62012-11-14 17:54:40 +11002250 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002251
Dave Chinner5b257b42010-06-03 16:22:29 +10002252 /*
2253 * Walk the inodes already attached to the buffer and mark them
2254 * stale. These will all have the flush locks held, so an
Dave Chinner5b3eed72010-08-24 11:42:41 +10002255 * in-memory inode walk can't lock them. By marking them all
2256 * stale first, we will not attempt to lock them in the loop
2257 * below as the XFS_ISTALE flag will be set.
Dave Chinner5b257b42010-06-03 16:22:29 +10002258 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +02002259 lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 while (lip) {
2261 if (lip->li_type == XFS_LI_INODE) {
2262 iip = (xfs_inode_log_item_t *)lip;
2263 ASSERT(iip->ili_logged == 1);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002264 lip->li_cb = xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11002265 xfs_trans_ail_copy_lsn(mp->m_ail,
2266 &iip->ili_flush_lsn,
2267 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11002268 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 }
2270 lip = lip->li_bio_list;
2271 }
2272
Dave Chinner5b3eed72010-08-24 11:42:41 +10002273
Dave Chinner5b257b42010-06-03 16:22:29 +10002274 /*
2275 * For each inode in memory attempt to add it to the inode
2276 * buffer and set it up for being staled on buffer IO
2277 * completion. This is safe as we've locked out tail pushing
2278 * and flushing by locking the buffer.
2279 *
2280 * We have already marked every inode that was part of a
2281 * transaction stale above, which means there is no point in
2282 * even trying to lock them.
2283 */
Jie Liu982e9392013-12-13 15:51:49 +11002284 for (i = 0; i < inodes_per_cluster; i++) {
Dave Chinner5b3eed72010-08-24 11:42:41 +10002285retry:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002286 rcu_read_lock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002287 ip = radix_tree_lookup(&pag->pag_ici_root,
2288 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002290 /* Inode not in memory, nothing to do */
2291 if (!ip) {
2292 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002293 continue;
2294 }
2295
Dave Chinner5b3eed72010-08-24 11:42:41 +10002296 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002297 * because this is an RCU protected lookup, we could
2298 * find a recently freed or even reallocated inode
2299 * during the lookup. We need to check under the
2300 * i_flags_lock for a valid inode here. Skip it if it
2301 * is not valid, the wrong inode or stale.
2302 */
2303 spin_lock(&ip->i_flags_lock);
2304 if (ip->i_ino != inum + i ||
2305 __xfs_iflags_test(ip, XFS_ISTALE)) {
2306 spin_unlock(&ip->i_flags_lock);
2307 rcu_read_unlock();
2308 continue;
2309 }
2310 spin_unlock(&ip->i_flags_lock);
2311
2312 /*
Dave Chinner5b3eed72010-08-24 11:42:41 +10002313 * Don't try to lock/unlock the current inode, but we
2314 * _cannot_ skip the other inodes that we did not find
2315 * in the list attached to the buffer and are not
2316 * already marked stale. If we can't lock it, back off
2317 * and retry.
2318 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002319 if (ip != free_ip &&
2320 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002321 rcu_read_unlock();
Dave Chinner5b3eed72010-08-24 11:42:41 +10002322 delay(1);
2323 goto retry;
Dave Chinner5b257b42010-06-03 16:22:29 +10002324 }
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002325 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002326
Dave Chinner5b3eed72010-08-24 11:42:41 +10002327 xfs_iflock(ip);
Dave Chinner5b257b42010-06-03 16:22:29 +10002328 xfs_iflags_set(ip, XFS_ISTALE);
Dave Chinner5b257b42010-06-03 16:22:29 +10002329
Dave Chinner5b3eed72010-08-24 11:42:41 +10002330 /*
2331 * we don't need to attach clean inodes or those only
2332 * with unlogged changes (which we throw away, anyway).
2333 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002334 iip = ip->i_itemp;
Dave Chinner5b3eed72010-08-24 11:42:41 +10002335 if (!iip || xfs_inode_clean(ip)) {
Dave Chinner5b257b42010-06-03 16:22:29 +10002336 ASSERT(ip != free_ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 xfs_ifunlock(ip);
2338 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2339 continue;
2340 }
2341
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002342 iip->ili_last_fields = iip->ili_fields;
2343 iip->ili_fields = 0;
Dave Chinnerfc0561c2015-11-03 13:14:59 +11002344 iip->ili_fsync_fields = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11002346 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2347 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002349 xfs_buf_attach_iodone(bp, xfs_istale_done,
2350 &iip->ili_item);
Dave Chinner5b257b42010-06-03 16:22:29 +10002351
2352 if (ip != free_ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 }
2355
Dave Chinner5b3eed72010-08-24 11:42:41 +10002356 xfs_trans_stale_inode_buf(tp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 xfs_trans_binval(tp, bp);
2358 }
2359
Dave Chinner5017e972010-01-11 11:47:40 +00002360 xfs_perag_put(pag);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002361 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362}
2363
2364/*
2365 * This is called to return an inode to the inode free list.
2366 * The inode should already be truncated to 0 length and have
2367 * no pages associated with it. This routine also assumes that
2368 * the inode is already a part of the transaction.
2369 *
2370 * The on-disk copy of the inode will have been added to the list
2371 * of unlinked inodes in the AGI. We need to remove the inode from
2372 * that list atomically with respect to freeing it here.
2373 */
2374int
2375xfs_ifree(
2376 xfs_trans_t *tp,
2377 xfs_inode_t *ip,
2378 xfs_bmap_free_t *flist)
2379{
2380 int error;
Brian Foster09b56602015-05-29 09:26:03 +10002381 struct xfs_icluster xic = { 0 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002383 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Dave Chinner54d7b5c2016-02-09 16:54:58 +11002384 ASSERT(VFS_I(ip)->i_nlink == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 ASSERT(ip->i_d.di_nextents == 0);
2386 ASSERT(ip->i_d.di_anextents == 0);
Dave Chinnerc19b3b052016-02-09 16:54:58 +11002387 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 ASSERT(ip->i_d.di_nblocks == 0);
2389
2390 /*
2391 * Pull the on-disk inode from the AGI unlinked list.
2392 */
2393 error = xfs_iunlink_remove(tp, ip);
Dave Chinner1baaed82013-06-27 16:04:50 +10002394 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396
Brian Foster09b56602015-05-29 09:26:03 +10002397 error = xfs_difree(tp, ip->i_ino, flist, &xic);
Dave Chinner1baaed82013-06-27 16:04:50 +10002398 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 return error;
Dave Chinner1baaed82013-06-27 16:04:50 +10002400
Dave Chinnerc19b3b052016-02-09 16:54:58 +11002401 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 ip->i_d.di_flags = 0;
2403 ip->i_d.di_dmevmask = 0;
2404 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2406 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2407 /*
2408 * Bump the generation count so no one will be confused
2409 * by reincarnations of this inode.
2410 */
Dave Chinner9e9a2672016-02-09 16:54:58 +11002411 VFS_I(ip)->i_generation++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2413
Brian Foster09b56602015-05-29 09:26:03 +10002414 if (xic.deleted)
2415 error = xfs_ifree_cluster(ip, tp, &xic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002417 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418}
2419
2420/*
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002421 * This is called to unpin an inode. The caller must have the inode locked
2422 * in at least shared mode so that the buffer cannot be subsequently pinned
2423 * once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002425static void
Christoph Hellwigf392e632011-12-18 20:00:10 +00002426xfs_iunpin(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002427 struct xfs_inode *ip)
David Chinnera3f74ff2008-03-06 13:43:42 +11002428{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002429 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002430
Dave Chinner4aaf15d2010-03-08 11:24:07 +11002431 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2432
David Chinnera3f74ff2008-03-06 13:43:42 +11002433 /* Give the log a push to start the unpinning I/O */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002434 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002435
David Chinnera3f74ff2008-03-06 13:43:42 +11002436}
2437
Christoph Hellwigf392e632011-12-18 20:00:10 +00002438static void
2439__xfs_iunpin_wait(
2440 struct xfs_inode *ip)
2441{
2442 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2443 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2444
2445 xfs_iunpin(ip);
2446
2447 do {
2448 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2449 if (xfs_ipincount(ip))
2450 io_schedule();
2451 } while (xfs_ipincount(ip));
2452 finish_wait(wq, &wait.wait);
2453}
2454
Dave Chinner777df5a2010-02-06 12:37:26 +11002455void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456xfs_iunpin_wait(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002457 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458{
Christoph Hellwigf392e632011-12-18 20:00:10 +00002459 if (xfs_ipincount(ip))
2460 __xfs_iunpin_wait(ip);
David Chinnera3f74ff2008-03-06 13:43:42 +11002461}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462
Dave Chinner27320362013-10-29 22:11:44 +11002463/*
2464 * Removing an inode from the namespace involves removing the directory entry
2465 * and dropping the link count on the inode. Removing the directory entry can
2466 * result in locking an AGF (directory blocks were freed) and removing a link
2467 * count can result in placing the inode on an unlinked list which results in
2468 * locking an AGI.
2469 *
2470 * The big problem here is that we have an ordering constraint on AGF and AGI
2471 * locking - inode allocation locks the AGI, then can allocate a new extent for
2472 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2473 * removes the inode from the unlinked list, requiring that we lock the AGI
2474 * first, and then freeing the inode can result in an inode chunk being freed
2475 * and hence freeing disk space requiring that we lock an AGF.
2476 *
2477 * Hence the ordering that is imposed by other parts of the code is AGI before
2478 * AGF. This means we cannot remove the directory entry before we drop the inode
2479 * reference count and put it on the unlinked list as this results in a lock
2480 * order of AGF then AGI, and this can deadlock against inode allocation and
2481 * freeing. Therefore we must drop the link counts before we remove the
2482 * directory entry.
2483 *
2484 * This is still safe from a transactional point of view - it is not until we
2485 * get to xfs_bmap_finish() that we have the possibility of multiple
2486 * transactions in this operation. Hence as long as we remove the directory
2487 * entry and drop the link count in the first transaction of the remove
2488 * operation, there are no transactional constraints on the ordering here.
2489 */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002490int
2491xfs_remove(
2492 xfs_inode_t *dp,
2493 struct xfs_name *name,
2494 xfs_inode_t *ip)
2495{
2496 xfs_mount_t *mp = dp->i_mount;
2497 xfs_trans_t *tp = NULL;
Dave Chinnerc19b3b052016-02-09 16:54:58 +11002498 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002499 int error = 0;
2500 xfs_bmap_free_t free_list;
2501 xfs_fsblock_t first_block;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002502 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002503
2504 trace_xfs_remove(dp, name);
2505
2506 if (XFS_FORCED_SHUTDOWN(mp))
Dave Chinner24513372014-06-25 14:58:08 +10002507 return -EIO;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002508
2509 error = xfs_qm_dqattach(dp, 0);
2510 if (error)
2511 goto std_return;
2512
2513 error = xfs_qm_dqattach(ip, 0);
2514 if (error)
2515 goto std_return;
2516
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002517 /*
2518 * We try to get the real space reservation first,
2519 * allowing for directory btree deletion(s) implying
2520 * possible bmap insert(s). If we can't get the space
2521 * reservation then we use 0 instead, and avoid the bmap
2522 * btree insert(s) in the directory code by, if the bmap
2523 * insert tries to happen, instead trimming the LAST
2524 * block from the directory.
2525 */
2526 resblks = XFS_REMOVE_SPACE_RES(mp);
Christoph Hellwig253f4912016-04-06 09:19:55 +10002527 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
Dave Chinner24513372014-06-25 14:58:08 +10002528 if (error == -ENOSPC) {
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002529 resblks = 0;
Christoph Hellwig253f4912016-04-06 09:19:55 +10002530 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2531 &tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002532 }
2533 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10002534 ASSERT(error != -ENOSPC);
Christoph Hellwig253f4912016-04-06 09:19:55 +10002535 goto std_return;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002536 }
2537
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002538 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002539 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2540
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002541 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002542 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2543
2544 /*
2545 * If we're removing a directory perform some additional validation.
2546 */
2547 if (is_dir) {
Dave Chinner54d7b5c2016-02-09 16:54:58 +11002548 ASSERT(VFS_I(ip)->i_nlink >= 2);
2549 if (VFS_I(ip)->i_nlink != 2) {
Dave Chinner24513372014-06-25 14:58:08 +10002550 error = -ENOTEMPTY;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002551 goto out_trans_cancel;
2552 }
2553 if (!xfs_dir_isempty(ip)) {
Dave Chinner24513372014-06-25 14:58:08 +10002554 error = -ENOTEMPTY;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002555 goto out_trans_cancel;
2556 }
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002557
Dave Chinner27320362013-10-29 22:11:44 +11002558 /* Drop the link from ip's "..". */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002559 error = xfs_droplink(tp, dp);
2560 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002561 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002562
Dave Chinner27320362013-10-29 22:11:44 +11002563 /* Drop the "." link from ip to self. */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002564 error = xfs_droplink(tp, ip);
2565 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002566 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002567 } else {
2568 /*
2569 * When removing a non-directory we need to log the parent
2570 * inode here. For a directory this is done implicitly
2571 * by the xfs_droplink call for the ".." entry.
2572 */
2573 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2574 }
Dave Chinner27320362013-10-29 22:11:44 +11002575 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002576
Dave Chinner27320362013-10-29 22:11:44 +11002577 /* Drop the link from dp to ip. */
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002578 error = xfs_droplink(tp, ip);
2579 if (error)
Dave Chinner27320362013-10-29 22:11:44 +11002580 goto out_trans_cancel;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002581
Dave Chinner27320362013-10-29 22:11:44 +11002582 xfs_bmap_init(&free_list, &first_block);
2583 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2584 &first_block, &free_list, resblks);
2585 if (error) {
Dave Chinner24513372014-06-25 14:58:08 +10002586 ASSERT(error != -ENOENT);
Dave Chinner27320362013-10-29 22:11:44 +11002587 goto out_bmap_cancel;
2588 }
2589
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002590 /*
2591 * If this is a synchronous mount, make sure that the
2592 * remove transaction goes to disk before returning to
2593 * the user.
2594 */
2595 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2596 xfs_trans_set_sync(tp);
2597
Eric Sandeenf6106ef2016-01-11 11:34:01 +11002598 error = xfs_bmap_finish(&tp, &free_list, NULL);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002599 if (error)
2600 goto out_bmap_cancel;
2601
Christoph Hellwig70393312015-06-04 13:48:08 +10002602 error = xfs_trans_commit(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002603 if (error)
2604 goto std_return;
2605
Christoph Hellwig2cd2ef62014-04-23 07:11:51 +10002606 if (is_dir && xfs_inode_is_filestream(ip))
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002607 xfs_filestream_deassociate(ip);
2608
2609 return 0;
2610
2611 out_bmap_cancel:
2612 xfs_bmap_cancel(&free_list);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002613 out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10002614 xfs_trans_cancel(tp);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002615 std_return:
2616 return error;
2617}
2618
Dave Chinnerf6bba202013-08-12 20:49:46 +10002619/*
2620 * Enter all inodes for a rename transaction into a sorted array.
2621 */
Dave Chinner95afcf52015-03-25 14:03:32 +11002622#define __XFS_SORT_INODES 5
Dave Chinnerf6bba202013-08-12 20:49:46 +10002623STATIC void
2624xfs_sort_for_rename(
Dave Chinner95afcf52015-03-25 14:03:32 +11002625 struct xfs_inode *dp1, /* in: old (source) directory inode */
2626 struct xfs_inode *dp2, /* in: new (target) directory inode */
2627 struct xfs_inode *ip1, /* in: inode of old entry */
2628 struct xfs_inode *ip2, /* in: inode of new entry */
2629 struct xfs_inode *wip, /* in: whiteout inode */
2630 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2631 int *num_inodes) /* in/out: inodes in array */
Dave Chinnerf6bba202013-08-12 20:49:46 +10002632{
Dave Chinnerf6bba202013-08-12 20:49:46 +10002633 int i, j;
2634
Dave Chinner95afcf52015-03-25 14:03:32 +11002635 ASSERT(*num_inodes == __XFS_SORT_INODES);
2636 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2637
Dave Chinnerf6bba202013-08-12 20:49:46 +10002638 /*
2639 * i_tab contains a list of pointers to inodes. We initialize
2640 * the table here & we'll sort it. We will then use it to
2641 * order the acquisition of the inode locks.
2642 *
2643 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2644 */
Dave Chinner95afcf52015-03-25 14:03:32 +11002645 i = 0;
2646 i_tab[i++] = dp1;
2647 i_tab[i++] = dp2;
2648 i_tab[i++] = ip1;
2649 if (ip2)
2650 i_tab[i++] = ip2;
2651 if (wip)
2652 i_tab[i++] = wip;
2653 *num_inodes = i;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002654
2655 /*
2656 * Sort the elements via bubble sort. (Remember, there are at
Dave Chinner95afcf52015-03-25 14:03:32 +11002657 * most 5 elements to sort, so this is adequate.)
Dave Chinnerf6bba202013-08-12 20:49:46 +10002658 */
2659 for (i = 0; i < *num_inodes; i++) {
2660 for (j = 1; j < *num_inodes; j++) {
2661 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
Dave Chinner95afcf52015-03-25 14:03:32 +11002662 struct xfs_inode *temp = i_tab[j];
Dave Chinnerf6bba202013-08-12 20:49:46 +10002663 i_tab[j] = i_tab[j-1];
2664 i_tab[j-1] = temp;
2665 }
2666 }
2667 }
2668}
2669
Dave Chinner310606b2015-03-25 14:06:07 +11002670static int
2671xfs_finish_rename(
2672 struct xfs_trans *tp,
2673 struct xfs_bmap_free *free_list)
2674{
Dave Chinner310606b2015-03-25 14:06:07 +11002675 int error;
2676
2677 /*
2678 * If this is a synchronous mount, make sure that the rename transaction
2679 * goes to disk before returning to the user.
2680 */
2681 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2682 xfs_trans_set_sync(tp);
2683
Eric Sandeenf6106ef2016-01-11 11:34:01 +11002684 error = xfs_bmap_finish(&tp, free_list, NULL);
Dave Chinner310606b2015-03-25 14:06:07 +11002685 if (error) {
2686 xfs_bmap_cancel(free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10002687 xfs_trans_cancel(tp);
Dave Chinner310606b2015-03-25 14:06:07 +11002688 return error;
2689 }
2690
Christoph Hellwig70393312015-06-04 13:48:08 +10002691 return xfs_trans_commit(tp);
Dave Chinner310606b2015-03-25 14:06:07 +11002692}
2693
Dave Chinnerf6bba202013-08-12 20:49:46 +10002694/*
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002695 * xfs_cross_rename()
2696 *
2697 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
2698 */
2699STATIC int
2700xfs_cross_rename(
2701 struct xfs_trans *tp,
2702 struct xfs_inode *dp1,
2703 struct xfs_name *name1,
2704 struct xfs_inode *ip1,
2705 struct xfs_inode *dp2,
2706 struct xfs_name *name2,
2707 struct xfs_inode *ip2,
2708 struct xfs_bmap_free *free_list,
2709 xfs_fsblock_t *first_block,
2710 int spaceres)
2711{
2712 int error = 0;
2713 int ip1_flags = 0;
2714 int ip2_flags = 0;
2715 int dp2_flags = 0;
2716
2717 /* Swap inode number for dirent in first parent */
2718 error = xfs_dir_replace(tp, dp1, name1,
2719 ip2->i_ino,
2720 first_block, free_list, spaceres);
2721 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002722 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002723
2724 /* Swap inode number for dirent in second parent */
2725 error = xfs_dir_replace(tp, dp2, name2,
2726 ip1->i_ino,
2727 first_block, free_list, spaceres);
2728 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002729 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002730
2731 /*
2732 * If we're renaming one or more directories across different parents,
2733 * update the respective ".." entries (and link counts) to match the new
2734 * parents.
2735 */
2736 if (dp1 != dp2) {
2737 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2738
Dave Chinnerc19b3b052016-02-09 16:54:58 +11002739 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002740 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2741 dp1->i_ino, first_block,
2742 free_list, spaceres);
2743 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002744 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002745
2746 /* transfer ip2 ".." reference to dp1 */
Dave Chinnerc19b3b052016-02-09 16:54:58 +11002747 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002748 error = xfs_droplink(tp, dp2);
2749 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002750 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002751 error = xfs_bumplink(tp, dp1);
2752 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002753 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002754 }
2755
2756 /*
2757 * Although ip1 isn't changed here, userspace needs
2758 * to be warned about the change, so that applications
2759 * relying on it (like backup ones), will properly
2760 * notify the change
2761 */
2762 ip1_flags |= XFS_ICHGTIME_CHG;
2763 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2764 }
2765
Dave Chinnerc19b3b052016-02-09 16:54:58 +11002766 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002767 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2768 dp2->i_ino, first_block,
2769 free_list, spaceres);
2770 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002771 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002772
2773 /* transfer ip1 ".." reference to dp2 */
Dave Chinnerc19b3b052016-02-09 16:54:58 +11002774 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002775 error = xfs_droplink(tp, dp1);
2776 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002777 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002778 error = xfs_bumplink(tp, dp2);
2779 if (error)
Dave Chinnereeacd322015-03-25 14:08:07 +11002780 goto out_trans_abort;
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002781 }
2782
2783 /*
2784 * Although ip2 isn't changed here, userspace needs
2785 * to be warned about the change, so that applications
2786 * relying on it (like backup ones), will properly
2787 * notify the change
2788 */
2789 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2790 ip2_flags |= XFS_ICHGTIME_CHG;
2791 }
2792 }
2793
2794 if (ip1_flags) {
2795 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2796 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2797 }
2798 if (ip2_flags) {
2799 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2800 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2801 }
2802 if (dp2_flags) {
2803 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2804 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2805 }
2806 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2807 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
Dave Chinnereeacd322015-03-25 14:08:07 +11002808 return xfs_finish_rename(tp, free_list);
2809
2810out_trans_abort:
2811 xfs_bmap_cancel(free_list);
Christoph Hellwig4906e212015-06-04 13:47:56 +10002812 xfs_trans_cancel(tp);
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002813 return error;
2814}
2815
2816/*
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002817 * xfs_rename_alloc_whiteout()
2818 *
2819 * Return a referenced, unlinked, unlocked inode that that can be used as a
2820 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2821 * crash between allocating the inode and linking it into the rename transaction
2822 * recovery will free the inode and we won't leak it.
2823 */
2824static int
2825xfs_rename_alloc_whiteout(
2826 struct xfs_inode *dp,
2827 struct xfs_inode **wip)
2828{
2829 struct xfs_inode *tmpfile;
2830 int error;
2831
2832 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile);
2833 if (error)
2834 return error;
2835
Brian Foster22419ac2015-05-29 08:14:55 +10002836 /*
2837 * Prepare the tmpfile inode as if it were created through the VFS.
2838 * Otherwise, the link increment paths will complain about nlink 0->1.
2839 * Drop the link count as done by d_tmpfile(), complete the inode setup
2840 * and flag it as linkable.
2841 */
2842 drop_nlink(VFS_I(tmpfile));
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002843 xfs_finish_inode_setup(tmpfile);
2844 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2845
2846 *wip = tmpfile;
2847 return 0;
2848}
2849
2850/*
Dave Chinnerf6bba202013-08-12 20:49:46 +10002851 * xfs_rename
2852 */
2853int
2854xfs_rename(
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002855 struct xfs_inode *src_dp,
2856 struct xfs_name *src_name,
2857 struct xfs_inode *src_ip,
2858 struct xfs_inode *target_dp,
2859 struct xfs_name *target_name,
2860 struct xfs_inode *target_ip,
2861 unsigned int flags)
Dave Chinnerf6bba202013-08-12 20:49:46 +10002862{
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002863 struct xfs_mount *mp = src_dp->i_mount;
2864 struct xfs_trans *tp;
2865 struct xfs_bmap_free free_list;
2866 xfs_fsblock_t first_block;
2867 struct xfs_inode *wip = NULL; /* whiteout inode */
2868 struct xfs_inode *inodes[__XFS_SORT_INODES];
2869 int num_inodes = __XFS_SORT_INODES;
Dave Chinner2b936812015-03-25 15:12:30 +11002870 bool new_parent = (src_dp != target_dp);
Dave Chinnerc19b3b052016-02-09 16:54:58 +11002871 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002872 int spaceres;
2873 int error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002874
2875 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2876
Dave Chinnereeacd322015-03-25 14:08:07 +11002877 if ((flags & RENAME_EXCHANGE) && !target_ip)
2878 return -EINVAL;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002879
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002880 /*
2881 * If we are doing a whiteout operation, allocate the whiteout inode
2882 * we will be placing at the target and ensure the type is set
2883 * appropriately.
2884 */
2885 if (flags & RENAME_WHITEOUT) {
2886 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
2887 error = xfs_rename_alloc_whiteout(target_dp, &wip);
2888 if (error)
2889 return error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002890
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002891 /* setup target dirent info as whiteout */
2892 src_name->type = XFS_DIR3_FT_CHRDEV;
2893 }
2894
2895 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
Dave Chinnerf6bba202013-08-12 20:49:46 +10002896 inodes, &num_inodes);
2897
Dave Chinnerf6bba202013-08-12 20:49:46 +10002898 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
Christoph Hellwig253f4912016-04-06 09:19:55 +10002899 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
Dave Chinner24513372014-06-25 14:58:08 +10002900 if (error == -ENOSPC) {
Dave Chinnerf6bba202013-08-12 20:49:46 +10002901 spaceres = 0;
Christoph Hellwig253f4912016-04-06 09:19:55 +10002902 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2903 &tp);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002904 }
Dave Chinner445883e2015-03-25 14:05:43 +11002905 if (error)
Christoph Hellwig253f4912016-04-06 09:19:55 +10002906 goto out_release_wip;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002907
2908 /*
2909 * Attach the dquots to the inodes
2910 */
2911 error = xfs_qm_vop_rename_dqattach(inodes);
Dave Chinner445883e2015-03-25 14:05:43 +11002912 if (error)
2913 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002914
2915 /*
2916 * Lock all the participating inodes. Depending upon whether
2917 * the target_name exists in the target directory, and
2918 * whether the target directory is the same as the source
2919 * directory, we can lock from 2 to 4 inodes.
2920 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002921 if (!new_parent)
2922 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2923 else
2924 xfs_lock_two_inodes(src_dp, target_dp,
2925 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT);
2926
Dave Chinnerf6bba202013-08-12 20:49:46 +10002927 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2928
2929 /*
2930 * Join all the inodes to the transaction. From this point on,
2931 * we can rely on either trans_commit or trans_cancel to unlock
2932 * them.
2933 */
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002934 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002935 if (new_parent)
Dave Chinnerdbad7c92015-08-19 10:33:00 +10002936 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002937 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2938 if (target_ip)
2939 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11002940 if (wip)
2941 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002942
2943 /*
2944 * If we are using project inheritance, we only allow renames
2945 * into our tree when the project IDs are the same; else the
2946 * tree quota mechanism would be circumvented.
2947 */
2948 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2949 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
Dave Chinner24513372014-06-25 14:58:08 +10002950 error = -EXDEV;
Dave Chinner445883e2015-03-25 14:05:43 +11002951 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002952 }
2953
Dave Chinner445883e2015-03-25 14:05:43 +11002954 xfs_bmap_init(&free_list, &first_block);
2955
Dave Chinnereeacd322015-03-25 14:08:07 +11002956 /* RENAME_EXCHANGE is unique from here on. */
2957 if (flags & RENAME_EXCHANGE)
2958 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2959 target_dp, target_name, target_ip,
2960 &free_list, &first_block, spaceres);
Carlos Maiolinod31a1822014-12-24 08:51:42 +11002961
2962 /*
Dave Chinnerf6bba202013-08-12 20:49:46 +10002963 * Set up the target.
2964 */
2965 if (target_ip == NULL) {
2966 /*
2967 * If there's no space reservation, check the entry will
2968 * fit before actually inserting it.
2969 */
Eric Sandeen94f3cad2014-09-09 11:57:52 +10002970 if (!spaceres) {
2971 error = xfs_dir_canenter(tp, target_dp, target_name);
2972 if (error)
Dave Chinner445883e2015-03-25 14:05:43 +11002973 goto out_trans_cancel;
Eric Sandeen94f3cad2014-09-09 11:57:52 +10002974 }
Dave Chinnerf6bba202013-08-12 20:49:46 +10002975 /*
2976 * If target does not exist and the rename crosses
2977 * directories, adjust the target directory link count
2978 * to account for the ".." reference from the new entry.
2979 */
2980 error = xfs_dir_createname(tp, target_dp, target_name,
2981 src_ip->i_ino, &first_block,
2982 &free_list, spaceres);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002983 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10002984 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002985
2986 xfs_trans_ichgtime(tp, target_dp,
2987 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2988
2989 if (new_parent && src_is_directory) {
2990 error = xfs_bumplink(tp, target_dp);
2991 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10002992 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10002993 }
2994 } else { /* target_ip != NULL */
2995 /*
2996 * If target exists and it's a directory, check that both
2997 * target and source are directories and that target can be
2998 * destroyed, or that neither is a directory.
2999 */
Dave Chinnerc19b3b052016-02-09 16:54:58 +11003000 if (S_ISDIR(VFS_I(target_ip)->i_mode)) {
Dave Chinnerf6bba202013-08-12 20:49:46 +10003001 /*
3002 * Make sure target dir is empty.
3003 */
3004 if (!(xfs_dir_isempty(target_ip)) ||
Dave Chinner54d7b5c2016-02-09 16:54:58 +11003005 (VFS_I(target_ip)->i_nlink > 2)) {
Dave Chinner24513372014-06-25 14:58:08 +10003006 error = -EEXIST;
Dave Chinner445883e2015-03-25 14:05:43 +11003007 goto out_trans_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003008 }
3009 }
3010
3011 /*
3012 * Link the source inode under the target name.
3013 * If the source inode is a directory and we are moving
3014 * it across directories, its ".." entry will be
3015 * inconsistent until we replace that down below.
3016 *
3017 * In case there is already an entry with the same
3018 * name at the destination directory, remove it first.
3019 */
3020 error = xfs_dir_replace(tp, target_dp, target_name,
3021 src_ip->i_ino,
3022 &first_block, &free_list, spaceres);
3023 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003024 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003025
3026 xfs_trans_ichgtime(tp, target_dp,
3027 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3028
3029 /*
3030 * Decrement the link count on the target since the target
3031 * dir no longer points to it.
3032 */
3033 error = xfs_droplink(tp, target_ip);
3034 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003035 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003036
3037 if (src_is_directory) {
3038 /*
3039 * Drop the link from the old "." entry.
3040 */
3041 error = xfs_droplink(tp, target_ip);
3042 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003043 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003044 }
3045 } /* target_ip != NULL */
3046
3047 /*
3048 * Remove the source.
3049 */
3050 if (new_parent && src_is_directory) {
3051 /*
3052 * Rewrite the ".." entry to point to the new
3053 * directory.
3054 */
3055 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3056 target_dp->i_ino,
3057 &first_block, &free_list, spaceres);
Dave Chinner24513372014-06-25 14:58:08 +10003058 ASSERT(error != -EEXIST);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003059 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003060 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003061 }
3062
3063 /*
3064 * We always want to hit the ctime on the source inode.
3065 *
3066 * This isn't strictly required by the standards since the source
3067 * inode isn't really being changed, but old unix file systems did
3068 * it and some incremental backup programs won't work without it.
3069 */
3070 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3071 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3072
3073 /*
3074 * Adjust the link count on src_dp. This is necessary when
3075 * renaming a directory, either within one parent when
3076 * the target existed, or across two parent directories.
3077 */
3078 if (src_is_directory && (new_parent || target_ip != NULL)) {
3079
3080 /*
3081 * Decrement link count on src_directory since the
3082 * entry that's moved no longer points to it.
3083 */
3084 error = xfs_droplink(tp, src_dp);
3085 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003086 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003087 }
3088
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003089 /*
3090 * For whiteouts, we only need to update the source dirent with the
3091 * inode number of the whiteout inode rather than removing it
3092 * altogether.
3093 */
3094 if (wip) {
3095 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
Dave Chinnerf6bba202013-08-12 20:49:46 +10003096 &first_block, &free_list, spaceres);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003097 } else
3098 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3099 &first_block, &free_list, spaceres);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003100 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003101 goto out_bmap_cancel;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003102
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003103 /*
3104 * For whiteouts, we need to bump the link count on the whiteout inode.
3105 * This means that failures all the way up to this point leave the inode
3106 * on the unlinked list and so cleanup is a simple matter of dropping
3107 * the remaining reference to it. If we fail here after bumping the link
3108 * count, we're shutting down the filesystem so we'll never see the
3109 * intermediate state on disk.
3110 */
3111 if (wip) {
Dave Chinner54d7b5c2016-02-09 16:54:58 +11003112 ASSERT(VFS_I(wip)->i_nlink == 0);
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003113 error = xfs_bumplink(tp, wip);
3114 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003115 goto out_bmap_cancel;
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003116 error = xfs_iunlink_remove(tp, wip);
3117 if (error)
Christoph Hellwig4906e212015-06-04 13:47:56 +10003118 goto out_bmap_cancel;
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003119 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE);
3120
3121 /*
3122 * Now we have a real link, clear the "I'm a tmpfile" state
3123 * flag from the inode so it doesn't accidentally get misused in
3124 * future.
3125 */
3126 VFS_I(wip)->i_state &= ~I_LINKABLE;
3127 }
Dave Chinnerf6bba202013-08-12 20:49:46 +10003128
3129 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3130 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3131 if (new_parent)
3132 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3133
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003134 error = xfs_finish_rename(tp, &free_list);
3135 if (wip)
3136 IRELE(wip);
3137 return error;
Dave Chinnerf6bba202013-08-12 20:49:46 +10003138
Dave Chinner445883e2015-03-25 14:05:43 +11003139out_bmap_cancel:
Dave Chinnerf6bba202013-08-12 20:49:46 +10003140 xfs_bmap_cancel(&free_list);
Dave Chinner445883e2015-03-25 14:05:43 +11003141out_trans_cancel:
Christoph Hellwig4906e212015-06-04 13:47:56 +10003142 xfs_trans_cancel(tp);
Christoph Hellwig253f4912016-04-06 09:19:55 +10003143out_release_wip:
Dave Chinner7dcf5c32015-03-25 14:08:08 +11003144 if (wip)
3145 IRELE(wip);
Dave Chinnerf6bba202013-08-12 20:49:46 +10003146 return error;
3147}
3148
David Chinnerbad55842008-03-06 13:43:49 +11003149STATIC int
3150xfs_iflush_cluster(
3151 xfs_inode_t *ip,
3152 xfs_buf_t *bp)
3153{
3154 xfs_mount_t *mp = ip->i_mount;
Dave Chinner5017e972010-01-11 11:47:40 +00003155 struct xfs_perag *pag;
David Chinnerbad55842008-03-06 13:43:49 +11003156 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10003157 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11003158 int ilist_size;
3159 xfs_inode_t **ilist;
3160 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11003161 int nr_found;
3162 int clcount = 0;
3163 int bufwasdelwri;
3164 int i;
3165
Dave Chinner5017e972010-01-11 11:47:40 +00003166 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
David Chinnerbad55842008-03-06 13:43:49 +11003167
Jie Liu0f49efd2013-12-13 15:51:48 +11003168 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
David Chinnerc8f5f122008-05-20 11:30:15 +10003169 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10003170 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11003171 if (!ilist)
Dave Chinner44b56e02010-01-11 11:47:43 +00003172 goto out_put;
David Chinnerbad55842008-03-06 13:43:49 +11003173
Jie Liu0f49efd2013-12-13 15:51:48 +11003174 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
David Chinnerbad55842008-03-06 13:43:49 +11003175 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003176 rcu_read_lock();
David Chinnerbad55842008-03-06 13:43:49 +11003177 /* really need a gang lookup range call here */
3178 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10003179 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11003180 if (nr_found == 0)
3181 goto out_free;
3182
3183 for (i = 0; i < nr_found; i++) {
3184 iq = ilist[i];
3185 if (iq == ip)
3186 continue;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003187
3188 /*
3189 * because this is an RCU protected lookup, we could find a
3190 * recently freed or even reallocated inode during the lookup.
3191 * We need to check under the i_flags_lock for a valid inode
3192 * here. Skip it if it is not valid or the wrong inode.
3193 */
3194 spin_lock(&ip->i_flags_lock);
3195 if (!ip->i_ino ||
3196 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3197 spin_unlock(&ip->i_flags_lock);
3198 continue;
3199 }
3200 spin_unlock(&ip->i_flags_lock);
3201
David Chinnerbad55842008-03-06 13:43:49 +11003202 /*
3203 * Do an un-protected check to see if the inode is dirty and
3204 * is a candidate for flushing. These checks will be repeated
3205 * later after the appropriate locks are acquired.
3206 */
David Chinner33540402008-03-06 13:43:59 +11003207 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11003208 continue;
David Chinnerbad55842008-03-06 13:43:49 +11003209
3210 /*
3211 * Try to get locks. If any are unavailable or it is pinned,
3212 * then this inode cannot be flushed and is skipped.
3213 */
3214
3215 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3216 continue;
3217 if (!xfs_iflock_nowait(iq)) {
3218 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3219 continue;
3220 }
3221 if (xfs_ipincount(iq)) {
3222 xfs_ifunlock(iq);
3223 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3224 continue;
3225 }
3226
3227 /*
3228 * arriving here means that this inode can be flushed. First
3229 * re-check that it's dirty before flushing.
3230 */
David Chinner33540402008-03-06 13:43:59 +11003231 if (!xfs_inode_clean(iq)) {
3232 int error;
David Chinnerbad55842008-03-06 13:43:49 +11003233 error = xfs_iflush_int(iq, bp);
3234 if (error) {
3235 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3236 goto cluster_corrupt_out;
3237 }
3238 clcount++;
3239 } else {
3240 xfs_ifunlock(iq);
3241 }
3242 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3243 }
3244
3245 if (clcount) {
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11003246 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3247 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
David Chinnerbad55842008-03-06 13:43:49 +11003248 }
3249
3250out_free:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003251 rcu_read_unlock();
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003252 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00003253out_put:
3254 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11003255 return 0;
3256
3257
3258cluster_corrupt_out:
3259 /*
3260 * Corruption detected in the clustering loop. Invalidate the
3261 * inode buffer and shut down the filesystem.
3262 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11003263 rcu_read_unlock();
David Chinnerbad55842008-03-06 13:43:49 +11003264 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003265 * Clean up the buffer. If it was delwri, just release it --
David Chinnerbad55842008-03-06 13:43:49 +11003266 * brelse can handle it with no problems. If not, shut down the
3267 * filesystem before releasing the buffer.
3268 */
Christoph Hellwig43ff2122012-04-23 15:58:39 +10003269 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
David Chinnerbad55842008-03-06 13:43:49 +11003270 if (bufwasdelwri)
3271 xfs_buf_relse(bp);
3272
3273 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3274
3275 if (!bufwasdelwri) {
3276 /*
3277 * Just like incore_relse: if we have b_iodone functions,
3278 * mark the buffer as an error and call them. Otherwise
3279 * mark it as stale and brelse.
3280 */
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02003281 if (bp->b_iodone) {
Dave Chinnerb0388bf2016-02-10 15:01:11 +11003282 bp->b_flags &= ~XBF_DONE;
Christoph Hellwigc867cb62011-10-10 16:52:46 +00003283 xfs_buf_stale(bp);
Dave Chinner24513372014-06-25 14:58:08 +10003284 xfs_buf_ioerror(bp, -EIO);
Dave Chinnere8aaba92014-10-02 09:04:22 +10003285 xfs_buf_ioend(bp);
David Chinnerbad55842008-03-06 13:43:49 +11003286 } else {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00003287 xfs_buf_stale(bp);
David Chinnerbad55842008-03-06 13:43:49 +11003288 xfs_buf_relse(bp);
3289 }
3290 }
3291
3292 /*
3293 * Unlocks the flush lock
3294 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003295 xfs_iflush_abort(iq, false);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003296 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00003297 xfs_perag_put(pag);
Dave Chinner24513372014-06-25 14:58:08 +10003298 return -EFSCORRUPTED;
David Chinnerbad55842008-03-06 13:43:49 +11003299}
3300
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301/*
Christoph Hellwig4c468192012-04-23 15:58:36 +10003302 * Flush dirty inode metadata into the backing buffer.
3303 *
3304 * The caller must have the inode lock and the inode flush lock held. The
3305 * inode lock will still be held upon return to the caller, and the inode
3306 * flush lock will be released after the inode has reached the disk.
3307 *
3308 * The caller must write out the buffer returned in *bpp and release it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 */
3310int
3311xfs_iflush(
Christoph Hellwig4c468192012-04-23 15:58:36 +10003312 struct xfs_inode *ip,
3313 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314{
Christoph Hellwig4c468192012-04-23 15:58:36 +10003315 struct xfs_mount *mp = ip->i_mount;
3316 struct xfs_buf *bp;
3317 struct xfs_dinode *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319
Bill O'Donnellff6d6af2015-10-12 18:21:22 +11003320 XFS_STATS_INC(mp, xs_iflush_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003322 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003323 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003325 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326
Christoph Hellwig4c468192012-04-23 15:58:36 +10003327 *bpp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 xfs_iunpin_wait(ip);
3330
3331 /*
Dave Chinner4b6a4682010-01-11 11:45:21 +00003332 * For stale inodes we cannot rely on the backing buffer remaining
3333 * stale in cache for the remaining life of the stale inode and so
Christoph Hellwig475ee412012-07-03 12:21:22 -04003334 * xfs_imap_to_bp() below may give us a buffer that no longer contains
Dave Chinner4b6a4682010-01-11 11:45:21 +00003335 * inodes below. We have to check this after ensuring the inode is
3336 * unpinned so that it is safe to reclaim the stale inode after the
3337 * flush call.
3338 */
3339 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3340 xfs_ifunlock(ip);
3341 return 0;
3342 }
3343
3344 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 * This may have been unpinned because the filesystem is shutting
3346 * down forcibly. If that's the case we must not write this inode
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003347 * to disk, because the log record didn't make it to disk.
3348 *
3349 * We also have to remove the log item from the AIL in this case,
3350 * as we wait for an empty AIL as part of the unmount process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351 */
3352 if (XFS_FORCED_SHUTDOWN(mp)) {
Dave Chinner24513372014-06-25 14:58:08 +10003353 error = -EIO;
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003354 goto abort_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355 }
3356
3357 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003358 * Get the buffer containing the on-disk inode.
3359 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003360 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3361 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003362 if (error || !bp) {
3363 xfs_ifunlock(ip);
3364 return error;
3365 }
3366
3367 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 * First flush out the inode that xfs_iflush was called with.
3369 */
3370 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11003371 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373
3374 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003375 * If the buffer is pinned then push on the log now so we won't
3376 * get stuck waiting in the write for too long.
3377 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00003378 if (xfs_buf_ispinned(bp))
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003379 xfs_log_force(mp, 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003380
3381 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382 * inode clustering:
3383 * see if other inodes can be gathered into this write
3384 */
David Chinnerbad55842008-03-06 13:43:49 +11003385 error = xfs_iflush_cluster(ip, bp);
3386 if (error)
3387 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388
Christoph Hellwig4c468192012-04-23 15:58:36 +10003389 *bpp = bp;
3390 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391
3392corrupt_out:
3393 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10003394 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395cluster_corrupt_out:
Dave Chinner24513372014-06-25 14:58:08 +10003396 error = -EFSCORRUPTED;
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003397abort_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 /*
3399 * Unlocks the flush lock
3400 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003401 xfs_iflush_abort(ip, false);
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003402 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403}
3404
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405STATIC int
3406xfs_iflush_int(
Christoph Hellwig93848a92013-04-03 16:11:17 +11003407 struct xfs_inode *ip,
3408 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409{
Christoph Hellwig93848a92013-04-03 16:11:17 +11003410 struct xfs_inode_log_item *iip = ip->i_itemp;
3411 struct xfs_dinode *dip;
3412 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003414 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003415 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003417 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Christoph Hellwig93848a92013-04-03 16:11:17 +11003418 ASSERT(iip != NULL && iip->ili_fields != 0);
Dave Chinner263997a2014-05-20 07:46:40 +10003419 ASSERT(ip->i_d.di_version > 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421 /* set *dip = inode's place in the buffer */
Christoph Hellwig88ee2df2015-06-22 09:44:29 +10003422 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003424 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003426 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3427 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3428 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 goto corrupt_out;
3430 }
Dave Chinnerc19b3b052016-02-09 16:54:58 +11003431 if (S_ISREG(VFS_I(ip)->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 if (XFS_TEST_ERROR(
3433 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3434 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3435 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003436 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3437 "%s: Bad regular inode %Lu, ptr 0x%p",
3438 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003439 goto corrupt_out;
3440 }
Dave Chinnerc19b3b052016-02-09 16:54:58 +11003441 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442 if (XFS_TEST_ERROR(
3443 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3444 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3445 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3446 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003447 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3448 "%s: Bad directory inode %Lu, ptr 0x%p",
3449 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450 goto corrupt_out;
3451 }
3452 }
3453 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3454 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3455 XFS_RANDOM_IFLUSH_5)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003456 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3457 "%s: detected corrupt incore inode %Lu, "
3458 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3459 __func__, ip->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 ip->i_d.di_nextents + ip->i_d.di_anextents,
Dave Chinner6a19d932011-03-07 10:02:35 +11003461 ip->i_d.di_nblocks, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462 goto corrupt_out;
3463 }
3464 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3465 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003466 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3467 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3468 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469 goto corrupt_out;
3470 }
Dave Chinnere60896d2013-07-24 15:47:30 +10003471
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472 /*
Dave Chinner263997a2014-05-20 07:46:40 +10003473 * Inode item log recovery for v2 inodes are dependent on the
Dave Chinnere60896d2013-07-24 15:47:30 +10003474 * di_flushiter count for correct sequencing. We bump the flush
3475 * iteration count so we can detect flushes which postdate a log record
3476 * during recovery. This is redundant as we now log every change and
3477 * hence this can't happen but we need to still do it to ensure
3478 * backwards compatibility with old kernels that predate logging all
3479 * inode changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 */
Dave Chinnere60896d2013-07-24 15:47:30 +10003481 if (ip->i_d.di_version < 3)
3482 ip->i_d.di_flushiter++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483
3484 /*
Dave Chinner39878482016-02-09 16:54:58 +11003485 * Copy the dirty parts of the inode into the on-disk inode. We always
3486 * copy out the core of the inode, because if the inode is dirty at all
3487 * the core must be.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 */
Dave Chinner93f958f2016-02-09 16:54:58 +11003489 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490
3491 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3492 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3493 ip->i_d.di_flushiter = 0;
3494
Eric Sandeenfd9fdba2014-04-14 19:04:46 +10003495 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10003496 if (XFS_IFORK_Q(ip))
Eric Sandeenfd9fdba2014-04-14 19:04:46 +10003497 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498 xfs_inobp_check(mp, bp);
3499
3500 /*
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003501 * We've recorded everything logged in the inode, so we'd like to clear
3502 * the ili_fields bits so we don't log and flush things unnecessarily.
3503 * However, we can't stop logging all this information until the data
3504 * we've copied into the disk buffer is written to disk. If we did we
3505 * might overwrite the copy of the inode in the log with all the data
3506 * after re-logging only part of it, and in the face of a crash we
3507 * wouldn't have all the data we need to recover.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003509 * What we do is move the bits to the ili_last_fields field. When
3510 * logging the inode, these bits are moved back to the ili_fields field.
3511 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3512 * know that the information those bits represent is permanently on
3513 * disk. As long as the flush completes before the inode is logged
3514 * again, then both ili_fields and ili_last_fields will be cleared.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003516 * We can play with the ili_fields bits here, because the inode lock
3517 * must be held exclusively in order to set bits there and the flush
3518 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3519 * done routine can tell whether or not to look in the AIL. Also, store
3520 * the current LSN of the inode so that we can tell whether the item has
3521 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3522 * need the AIL lock, because it is a 64 bit value that cannot be read
3523 * atomically.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 */
Christoph Hellwig93848a92013-04-03 16:11:17 +11003525 iip->ili_last_fields = iip->ili_fields;
3526 iip->ili_fields = 0;
Dave Chinnerfc0561c2015-11-03 13:14:59 +11003527 iip->ili_fsync_fields = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +11003528 iip->ili_logged = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529
Christoph Hellwig93848a92013-04-03 16:11:17 +11003530 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3531 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532
Christoph Hellwig93848a92013-04-03 16:11:17 +11003533 /*
3534 * Attach the function xfs_iflush_done to the inode's
3535 * buffer. This will remove the inode from the AIL
3536 * and unlock the inode's flush lock when the inode is
3537 * completely written to disk.
3538 */
3539 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540
Christoph Hellwig93848a92013-04-03 16:11:17 +11003541 /* generate the checksum. */
3542 xfs_dinode_calc_crc(mp, dip);
3543
3544 ASSERT(bp->b_fspriv != NULL);
3545 ASSERT(bp->b_iodone != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546 return 0;
3547
3548corrupt_out:
Dave Chinner24513372014-06-25 14:58:08 +10003549 return -EFSCORRUPTED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550}