Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * 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. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * 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 | |
1da177e4 | 17 | */ |
40ebd81d RD |
18 | #include <linux/log2.h> |
19 | ||
1da177e4 | 20 | #include "xfs.h" |
a844f451 | 21 | #include "xfs_fs.h" |
70a9883c | 22 | #include "xfs_shared.h" |
239880ef DC |
23 | #include "xfs_format.h" |
24 | #include "xfs_log_format.h" | |
25 | #include "xfs_trans_resv.h" | |
1da177e4 | 26 | #include "xfs_sb.h" |
1da177e4 | 27 | #include "xfs_mount.h" |
3ab78df2 | 28 | #include "xfs_defer.h" |
a4fbe6ab | 29 | #include "xfs_inode.h" |
57062787 | 30 | #include "xfs_da_format.h" |
c24b5dfa | 31 | #include "xfs_da_btree.h" |
c24b5dfa | 32 | #include "xfs_dir2.h" |
a844f451 | 33 | #include "xfs_attr_sf.h" |
c24b5dfa | 34 | #include "xfs_attr.h" |
239880ef DC |
35 | #include "xfs_trans_space.h" |
36 | #include "xfs_trans.h" | |
1da177e4 | 37 | #include "xfs_buf_item.h" |
a844f451 | 38 | #include "xfs_inode_item.h" |
a844f451 NS |
39 | #include "xfs_ialloc.h" |
40 | #include "xfs_bmap.h" | |
68988114 | 41 | #include "xfs_bmap_util.h" |
e9e899a2 | 42 | #include "xfs_errortag.h" |
1da177e4 | 43 | #include "xfs_error.h" |
1da177e4 | 44 | #include "xfs_quota.h" |
2a82b8be | 45 | #include "xfs_filestream.h" |
93848a99 | 46 | #include "xfs_cksum.h" |
0b1b213f | 47 | #include "xfs_trace.h" |
33479e05 | 48 | #include "xfs_icache.h" |
c24b5dfa | 49 | #include "xfs_symlink.h" |
239880ef DC |
50 | #include "xfs_trans_priv.h" |
51 | #include "xfs_log.h" | |
a4fbe6ab | 52 | #include "xfs_bmap_btree.h" |
aa8968f2 | 53 | #include "xfs_reflink.h" |
005c5db8 | 54 | #include "xfs_dir2_priv.h" |
1da177e4 | 55 | |
1da177e4 | 56 | kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
57 | |
58 | /* | |
8f04c47a | 59 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
60 | * freed from a file in a single transaction. |
61 | */ | |
62 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
63 | ||
54d7b5c1 DC |
64 | STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); |
65 | STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); | |
66 | STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); | |
ab297431 | 67 | |
2a0ec1d9 DC |
68 | /* |
69 | * helper function to extract extent size hint from inode | |
70 | */ | |
71 | xfs_extlen_t | |
72 | xfs_get_extsz_hint( | |
73 | struct xfs_inode *ip) | |
74 | { | |
75 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
76 | return ip->i_d.di_extsize; | |
77 | if (XFS_IS_REALTIME_INODE(ip)) | |
78 | return ip->i_mount->m_sb.sb_rextsize; | |
79 | return 0; | |
80 | } | |
81 | ||
f7ca3522 DW |
82 | /* |
83 | * Helper function to extract CoW extent size hint from inode. | |
84 | * Between the extent size hint and the CoW extent size hint, we | |
e153aa79 DW |
85 | * return the greater of the two. If the value is zero (automatic), |
86 | * use the default size. | |
f7ca3522 DW |
87 | */ |
88 | xfs_extlen_t | |
89 | xfs_get_cowextsz_hint( | |
90 | struct xfs_inode *ip) | |
91 | { | |
92 | xfs_extlen_t a, b; | |
93 | ||
94 | a = 0; | |
95 | if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) | |
96 | a = ip->i_d.di_cowextsize; | |
97 | b = xfs_get_extsz_hint(ip); | |
98 | ||
e153aa79 DW |
99 | a = max(a, b); |
100 | if (a == 0) | |
101 | return XFS_DEFAULT_COWEXTSZ_HINT; | |
102 | return a; | |
f7ca3522 DW |
103 | } |
104 | ||
fa96acad | 105 | /* |
efa70be1 CH |
106 | * These two are wrapper routines around the xfs_ilock() routine used to |
107 | * centralize some grungy code. They are used in places that wish to lock the | |
108 | * inode solely for reading the extents. The reason these places can't just | |
109 | * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to | |
110 | * bringing in of the extents from disk for a file in b-tree format. If the | |
111 | * inode is in b-tree format, then we need to lock the inode exclusively until | |
112 | * the extents are read in. Locking it exclusively all the time would limit | |
113 | * our parallelism unnecessarily, though. What we do instead is check to see | |
114 | * if the extents have been read in yet, and only lock the inode exclusively | |
115 | * if they have not. | |
fa96acad | 116 | * |
efa70be1 | 117 | * The functions return a value which should be given to the corresponding |
01f4f327 | 118 | * xfs_iunlock() call. |
fa96acad DC |
119 | */ |
120 | uint | |
309ecac8 CH |
121 | xfs_ilock_data_map_shared( |
122 | struct xfs_inode *ip) | |
fa96acad | 123 | { |
309ecac8 | 124 | uint lock_mode = XFS_ILOCK_SHARED; |
fa96acad | 125 | |
309ecac8 CH |
126 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && |
127 | (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) | |
fa96acad | 128 | lock_mode = XFS_ILOCK_EXCL; |
fa96acad | 129 | xfs_ilock(ip, lock_mode); |
fa96acad DC |
130 | return lock_mode; |
131 | } | |
132 | ||
efa70be1 CH |
133 | uint |
134 | xfs_ilock_attr_map_shared( | |
135 | struct xfs_inode *ip) | |
fa96acad | 136 | { |
efa70be1 CH |
137 | uint lock_mode = XFS_ILOCK_SHARED; |
138 | ||
139 | if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && | |
140 | (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) | |
141 | lock_mode = XFS_ILOCK_EXCL; | |
142 | xfs_ilock(ip, lock_mode); | |
143 | return lock_mode; | |
fa96acad DC |
144 | } |
145 | ||
146 | /* | |
65523218 CH |
147 | * In addition to i_rwsem in the VFS inode, the xfs inode contains 2 |
148 | * multi-reader locks: i_mmap_lock and the i_lock. This routine allows | |
149 | * various combinations of the locks to be obtained. | |
fa96acad | 150 | * |
653c60b6 DC |
151 | * The 3 locks should always be ordered so that the IO lock is obtained first, |
152 | * the mmap lock second and the ilock last in order to prevent deadlock. | |
fa96acad | 153 | * |
653c60b6 DC |
154 | * Basic locking order: |
155 | * | |
65523218 | 156 | * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock |
653c60b6 DC |
157 | * |
158 | * mmap_sem locking order: | |
159 | * | |
65523218 | 160 | * i_rwsem -> page lock -> mmap_sem |
653c60b6 DC |
161 | * mmap_sem -> i_mmap_lock -> page_lock |
162 | * | |
163 | * The difference in mmap_sem locking order mean that we cannot hold the | |
164 | * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can | |
165 | * fault in pages during copy in/out (for buffered IO) or require the mmap_sem | |
166 | * in get_user_pages() to map the user pages into the kernel address space for | |
65523218 | 167 | * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because |
653c60b6 DC |
168 | * page faults already hold the mmap_sem. |
169 | * | |
170 | * Hence to serialise fully against both syscall and mmap based IO, we need to | |
65523218 | 171 | * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both |
653c60b6 DC |
172 | * taken in places where we need to invalidate the page cache in a race |
173 | * free manner (e.g. truncate, hole punch and other extent manipulation | |
174 | * functions). | |
fa96acad DC |
175 | */ |
176 | void | |
177 | xfs_ilock( | |
178 | xfs_inode_t *ip, | |
179 | uint lock_flags) | |
180 | { | |
181 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
182 | ||
183 | /* | |
184 | * You can't set both SHARED and EXCL for the same lock, | |
185 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
186 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
187 | */ | |
188 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
189 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
190 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
191 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
192 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
193 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 194 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad | 195 | |
65523218 CH |
196 | if (lock_flags & XFS_IOLOCK_EXCL) { |
197 | down_write_nested(&VFS_I(ip)->i_rwsem, | |
198 | XFS_IOLOCK_DEP(lock_flags)); | |
199 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
200 | down_read_nested(&VFS_I(ip)->i_rwsem, | |
201 | XFS_IOLOCK_DEP(lock_flags)); | |
202 | } | |
fa96acad | 203 | |
653c60b6 DC |
204 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
205 | mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
206 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
207 | mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
208 | ||
fa96acad DC |
209 | if (lock_flags & XFS_ILOCK_EXCL) |
210 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
211 | else if (lock_flags & XFS_ILOCK_SHARED) | |
212 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
213 | } | |
214 | ||
215 | /* | |
216 | * This is just like xfs_ilock(), except that the caller | |
217 | * is guaranteed not to sleep. It returns 1 if it gets | |
218 | * the requested locks and 0 otherwise. If the IO lock is | |
219 | * obtained but the inode lock cannot be, then the IO lock | |
220 | * is dropped before returning. | |
221 | * | |
222 | * ip -- the inode being locked | |
223 | * lock_flags -- this parameter indicates the inode's locks to be | |
224 | * to be locked. See the comment for xfs_ilock() for a list | |
225 | * of valid values. | |
226 | */ | |
227 | int | |
228 | xfs_ilock_nowait( | |
229 | xfs_inode_t *ip, | |
230 | uint lock_flags) | |
231 | { | |
232 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
233 | ||
234 | /* | |
235 | * You can't set both SHARED and EXCL for the same lock, | |
236 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
237 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
238 | */ | |
239 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
240 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
241 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
242 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
243 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
244 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 245 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
246 | |
247 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
65523218 | 248 | if (!down_write_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
249 | goto out; |
250 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
65523218 | 251 | if (!down_read_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
252 | goto out; |
253 | } | |
653c60b6 DC |
254 | |
255 | if (lock_flags & XFS_MMAPLOCK_EXCL) { | |
256 | if (!mrtryupdate(&ip->i_mmaplock)) | |
257 | goto out_undo_iolock; | |
258 | } else if (lock_flags & XFS_MMAPLOCK_SHARED) { | |
259 | if (!mrtryaccess(&ip->i_mmaplock)) | |
260 | goto out_undo_iolock; | |
261 | } | |
262 | ||
fa96acad DC |
263 | if (lock_flags & XFS_ILOCK_EXCL) { |
264 | if (!mrtryupdate(&ip->i_lock)) | |
653c60b6 | 265 | goto out_undo_mmaplock; |
fa96acad DC |
266 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
267 | if (!mrtryaccess(&ip->i_lock)) | |
653c60b6 | 268 | goto out_undo_mmaplock; |
fa96acad DC |
269 | } |
270 | return 1; | |
271 | ||
653c60b6 DC |
272 | out_undo_mmaplock: |
273 | if (lock_flags & XFS_MMAPLOCK_EXCL) | |
274 | mrunlock_excl(&ip->i_mmaplock); | |
275 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
276 | mrunlock_shared(&ip->i_mmaplock); | |
277 | out_undo_iolock: | |
fa96acad | 278 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 279 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 280 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 281 | up_read(&VFS_I(ip)->i_rwsem); |
653c60b6 | 282 | out: |
fa96acad DC |
283 | return 0; |
284 | } | |
285 | ||
286 | /* | |
287 | * xfs_iunlock() is used to drop the inode locks acquired with | |
288 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
289 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
290 | * that we know which locks to drop. | |
291 | * | |
292 | * ip -- the inode being unlocked | |
293 | * lock_flags -- this parameter indicates the inode's locks to be | |
294 | * to be unlocked. See the comment for xfs_ilock() for a list | |
295 | * of valid values for this parameter. | |
296 | * | |
297 | */ | |
298 | void | |
299 | xfs_iunlock( | |
300 | xfs_inode_t *ip, | |
301 | uint lock_flags) | |
302 | { | |
303 | /* | |
304 | * You can't set both SHARED and EXCL for the same lock, | |
305 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
306 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
307 | */ | |
308 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
309 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
310 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
311 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
312 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
313 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 314 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
315 | ASSERT(lock_flags != 0); |
316 | ||
317 | if (lock_flags & XFS_IOLOCK_EXCL) | |
65523218 | 318 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 319 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 320 | up_read(&VFS_I(ip)->i_rwsem); |
fa96acad | 321 | |
653c60b6 DC |
322 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
323 | mrunlock_excl(&ip->i_mmaplock); | |
324 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
325 | mrunlock_shared(&ip->i_mmaplock); | |
326 | ||
fa96acad DC |
327 | if (lock_flags & XFS_ILOCK_EXCL) |
328 | mrunlock_excl(&ip->i_lock); | |
329 | else if (lock_flags & XFS_ILOCK_SHARED) | |
330 | mrunlock_shared(&ip->i_lock); | |
331 | ||
332 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
333 | } | |
334 | ||
335 | /* | |
336 | * give up write locks. the i/o lock cannot be held nested | |
337 | * if it is being demoted. | |
338 | */ | |
339 | void | |
340 | xfs_ilock_demote( | |
341 | xfs_inode_t *ip, | |
342 | uint lock_flags) | |
343 | { | |
653c60b6 DC |
344 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); |
345 | ASSERT((lock_flags & | |
346 | ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
fa96acad DC |
347 | |
348 | if (lock_flags & XFS_ILOCK_EXCL) | |
349 | mrdemote(&ip->i_lock); | |
653c60b6 DC |
350 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
351 | mrdemote(&ip->i_mmaplock); | |
fa96acad | 352 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 353 | downgrade_write(&VFS_I(ip)->i_rwsem); |
fa96acad DC |
354 | |
355 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
356 | } | |
357 | ||
742ae1e3 | 358 | #if defined(DEBUG) || defined(XFS_WARN) |
fa96acad DC |
359 | int |
360 | xfs_isilocked( | |
361 | xfs_inode_t *ip, | |
362 | uint lock_flags) | |
363 | { | |
364 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
365 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
366 | return !!ip->i_lock.mr_writer; | |
367 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
368 | } | |
369 | ||
653c60b6 DC |
370 | if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { |
371 | if (!(lock_flags & XFS_MMAPLOCK_SHARED)) | |
372 | return !!ip->i_mmaplock.mr_writer; | |
373 | return rwsem_is_locked(&ip->i_mmaplock.mr_lock); | |
374 | } | |
375 | ||
fa96acad DC |
376 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
377 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
65523218 CH |
378 | return !debug_locks || |
379 | lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0); | |
380 | return rwsem_is_locked(&VFS_I(ip)->i_rwsem); | |
fa96acad DC |
381 | } |
382 | ||
383 | ASSERT(0); | |
384 | return 0; | |
385 | } | |
386 | #endif | |
387 | ||
b6a9947e DC |
388 | /* |
389 | * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when | |
390 | * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined | |
391 | * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build | |
392 | * errors and warnings. | |
393 | */ | |
394 | #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) | |
3403ccc0 DC |
395 | static bool |
396 | xfs_lockdep_subclass_ok( | |
397 | int subclass) | |
398 | { | |
399 | return subclass < MAX_LOCKDEP_SUBCLASSES; | |
400 | } | |
401 | #else | |
402 | #define xfs_lockdep_subclass_ok(subclass) (true) | |
403 | #endif | |
404 | ||
c24b5dfa | 405 | /* |
653c60b6 | 406 | * Bump the subclass so xfs_lock_inodes() acquires each lock with a different |
0952c818 DC |
407 | * value. This can be called for any type of inode lock combination, including |
408 | * parent locking. Care must be taken to ensure we don't overrun the subclass | |
409 | * storage fields in the class mask we build. | |
c24b5dfa DC |
410 | */ |
411 | static inline int | |
412 | xfs_lock_inumorder(int lock_mode, int subclass) | |
413 | { | |
0952c818 DC |
414 | int class = 0; |
415 | ||
416 | ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | | |
417 | XFS_ILOCK_RTSUM))); | |
3403ccc0 | 418 | ASSERT(xfs_lockdep_subclass_ok(subclass)); |
0952c818 | 419 | |
653c60b6 | 420 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { |
0952c818 | 421 | ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); |
0952c818 | 422 | class += subclass << XFS_IOLOCK_SHIFT; |
653c60b6 DC |
423 | } |
424 | ||
425 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { | |
0952c818 DC |
426 | ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); |
427 | class += subclass << XFS_MMAPLOCK_SHIFT; | |
653c60b6 DC |
428 | } |
429 | ||
0952c818 DC |
430 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { |
431 | ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); | |
432 | class += subclass << XFS_ILOCK_SHIFT; | |
433 | } | |
c24b5dfa | 434 | |
0952c818 | 435 | return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; |
c24b5dfa DC |
436 | } |
437 | ||
438 | /* | |
95afcf5c DC |
439 | * The following routine will lock n inodes in exclusive mode. We assume the |
440 | * caller calls us with the inodes in i_ino order. | |
c24b5dfa | 441 | * |
95afcf5c DC |
442 | * We need to detect deadlock where an inode that we lock is in the AIL and we |
443 | * start waiting for another inode that is locked by a thread in a long running | |
444 | * transaction (such as truncate). This can result in deadlock since the long | |
445 | * running trans might need to wait for the inode we just locked in order to | |
446 | * push the tail and free space in the log. | |
0952c818 DC |
447 | * |
448 | * xfs_lock_inodes() can only be used to lock one type of lock at a time - | |
449 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
450 | * lock more than one at a time, lockdep will report false positives saying we | |
451 | * have violated locking orders. | |
c24b5dfa | 452 | */ |
0d5a75e9 | 453 | static void |
c24b5dfa DC |
454 | xfs_lock_inodes( |
455 | xfs_inode_t **ips, | |
456 | int inodes, | |
457 | uint lock_mode) | |
458 | { | |
459 | int attempts = 0, i, j, try_lock; | |
460 | xfs_log_item_t *lp; | |
461 | ||
0952c818 DC |
462 | /* |
463 | * Currently supports between 2 and 5 inodes with exclusive locking. We | |
464 | * support an arbitrary depth of locking here, but absolute limits on | |
465 | * inodes depend on the the type of locking and the limits placed by | |
466 | * lockdep annotations in xfs_lock_inumorder. These are all checked by | |
467 | * the asserts. | |
468 | */ | |
95afcf5c | 469 | ASSERT(ips && inodes >= 2 && inodes <= 5); |
0952c818 DC |
470 | ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | |
471 | XFS_ILOCK_EXCL)); | |
472 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | | |
473 | XFS_ILOCK_SHARED))); | |
0952c818 DC |
474 | ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || |
475 | inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); | |
476 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || | |
477 | inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); | |
478 | ||
479 | if (lock_mode & XFS_IOLOCK_EXCL) { | |
480 | ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); | |
481 | } else if (lock_mode & XFS_MMAPLOCK_EXCL) | |
482 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); | |
c24b5dfa DC |
483 | |
484 | try_lock = 0; | |
485 | i = 0; | |
c24b5dfa DC |
486 | again: |
487 | for (; i < inodes; i++) { | |
488 | ASSERT(ips[i]); | |
489 | ||
95afcf5c | 490 | if (i && (ips[i] == ips[i - 1])) /* Already locked */ |
c24b5dfa DC |
491 | continue; |
492 | ||
493 | /* | |
95afcf5c DC |
494 | * If try_lock is not set yet, make sure all locked inodes are |
495 | * not in the AIL. If any are, set try_lock to be used later. | |
c24b5dfa | 496 | */ |
c24b5dfa DC |
497 | if (!try_lock) { |
498 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
499 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
95afcf5c | 500 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) |
c24b5dfa | 501 | try_lock++; |
c24b5dfa DC |
502 | } |
503 | } | |
504 | ||
505 | /* | |
506 | * If any of the previous locks we have locked is in the AIL, | |
507 | * we must TRY to get the second and subsequent locks. If | |
508 | * we can't get any, we must release all we have | |
509 | * and try again. | |
510 | */ | |
95afcf5c DC |
511 | if (!try_lock) { |
512 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); | |
513 | continue; | |
514 | } | |
515 | ||
516 | /* try_lock means we have an inode locked that is in the AIL. */ | |
517 | ASSERT(i != 0); | |
518 | if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) | |
519 | continue; | |
c24b5dfa | 520 | |
95afcf5c DC |
521 | /* |
522 | * Unlock all previous guys and try again. xfs_iunlock will try | |
523 | * to push the tail if the inode is in the AIL. | |
524 | */ | |
525 | attempts++; | |
526 | for (j = i - 1; j >= 0; j--) { | |
c24b5dfa | 527 | /* |
95afcf5c DC |
528 | * Check to see if we've already unlocked this one. Not |
529 | * the first one going back, and the inode ptr is the | |
530 | * same. | |
c24b5dfa | 531 | */ |
95afcf5c DC |
532 | if (j != (i - 1) && ips[j] == ips[j + 1]) |
533 | continue; | |
c24b5dfa | 534 | |
95afcf5c DC |
535 | xfs_iunlock(ips[j], lock_mode); |
536 | } | |
c24b5dfa | 537 | |
95afcf5c DC |
538 | if ((attempts % 5) == 0) { |
539 | delay(1); /* Don't just spin the CPU */ | |
c24b5dfa | 540 | } |
95afcf5c DC |
541 | i = 0; |
542 | try_lock = 0; | |
543 | goto again; | |
c24b5dfa | 544 | } |
c24b5dfa DC |
545 | } |
546 | ||
547 | /* | |
653c60b6 DC |
548 | * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - |
549 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
550 | * lock more than one at a time, lockdep will report false positives saying we | |
551 | * have violated locking orders. | |
c24b5dfa DC |
552 | */ |
553 | void | |
554 | xfs_lock_two_inodes( | |
555 | xfs_inode_t *ip0, | |
556 | xfs_inode_t *ip1, | |
557 | uint lock_mode) | |
558 | { | |
559 | xfs_inode_t *temp; | |
560 | int attempts = 0; | |
561 | xfs_log_item_t *lp; | |
562 | ||
65523218 CH |
563 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); |
564 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) | |
653c60b6 DC |
565 | ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); |
566 | ||
c24b5dfa DC |
567 | ASSERT(ip0->i_ino != ip1->i_ino); |
568 | ||
569 | if (ip0->i_ino > ip1->i_ino) { | |
570 | temp = ip0; | |
571 | ip0 = ip1; | |
572 | ip1 = temp; | |
573 | } | |
574 | ||
575 | again: | |
576 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
577 | ||
578 | /* | |
579 | * If the first lock we have locked is in the AIL, we must TRY to get | |
580 | * the second lock. If we can't get it, we must release the first one | |
581 | * and try again. | |
582 | */ | |
583 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
584 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
585 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
586 | xfs_iunlock(ip0, lock_mode); | |
587 | if ((++attempts % 5) == 0) | |
588 | delay(1); /* Don't just spin the CPU */ | |
589 | goto again; | |
590 | } | |
591 | } else { | |
592 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
593 | } | |
594 | } | |
595 | ||
596 | ||
fa96acad DC |
597 | void |
598 | __xfs_iflock( | |
599 | struct xfs_inode *ip) | |
600 | { | |
601 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
602 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
603 | ||
604 | do { | |
21417136 | 605 | prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
fa96acad DC |
606 | if (xfs_isiflocked(ip)) |
607 | io_schedule(); | |
608 | } while (!xfs_iflock_nowait(ip)); | |
609 | ||
21417136 | 610 | finish_wait(wq, &wait.wq_entry); |
fa96acad DC |
611 | } |
612 | ||
1da177e4 LT |
613 | STATIC uint |
614 | _xfs_dic2xflags( | |
c8ce540d | 615 | uint16_t di_flags, |
58f88ca2 DC |
616 | uint64_t di_flags2, |
617 | bool has_attr) | |
1da177e4 LT |
618 | { |
619 | uint flags = 0; | |
620 | ||
621 | if (di_flags & XFS_DIFLAG_ANY) { | |
622 | if (di_flags & XFS_DIFLAG_REALTIME) | |
e7b89481 | 623 | flags |= FS_XFLAG_REALTIME; |
1da177e4 | 624 | if (di_flags & XFS_DIFLAG_PREALLOC) |
e7b89481 | 625 | flags |= FS_XFLAG_PREALLOC; |
1da177e4 | 626 | if (di_flags & XFS_DIFLAG_IMMUTABLE) |
e7b89481 | 627 | flags |= FS_XFLAG_IMMUTABLE; |
1da177e4 | 628 | if (di_flags & XFS_DIFLAG_APPEND) |
e7b89481 | 629 | flags |= FS_XFLAG_APPEND; |
1da177e4 | 630 | if (di_flags & XFS_DIFLAG_SYNC) |
e7b89481 | 631 | flags |= FS_XFLAG_SYNC; |
1da177e4 | 632 | if (di_flags & XFS_DIFLAG_NOATIME) |
e7b89481 | 633 | flags |= FS_XFLAG_NOATIME; |
1da177e4 | 634 | if (di_flags & XFS_DIFLAG_NODUMP) |
e7b89481 | 635 | flags |= FS_XFLAG_NODUMP; |
1da177e4 | 636 | if (di_flags & XFS_DIFLAG_RTINHERIT) |
e7b89481 | 637 | flags |= FS_XFLAG_RTINHERIT; |
1da177e4 | 638 | if (di_flags & XFS_DIFLAG_PROJINHERIT) |
e7b89481 | 639 | flags |= FS_XFLAG_PROJINHERIT; |
1da177e4 | 640 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) |
e7b89481 | 641 | flags |= FS_XFLAG_NOSYMLINKS; |
dd9f438e | 642 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
e7b89481 | 643 | flags |= FS_XFLAG_EXTSIZE; |
dd9f438e | 644 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) |
e7b89481 | 645 | flags |= FS_XFLAG_EXTSZINHERIT; |
d3446eac | 646 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
e7b89481 | 647 | flags |= FS_XFLAG_NODEFRAG; |
2a82b8be | 648 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
e7b89481 | 649 | flags |= FS_XFLAG_FILESTREAM; |
1da177e4 LT |
650 | } |
651 | ||
58f88ca2 DC |
652 | if (di_flags2 & XFS_DIFLAG2_ANY) { |
653 | if (di_flags2 & XFS_DIFLAG2_DAX) | |
654 | flags |= FS_XFLAG_DAX; | |
f7ca3522 DW |
655 | if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE) |
656 | flags |= FS_XFLAG_COWEXTSIZE; | |
58f88ca2 DC |
657 | } |
658 | ||
659 | if (has_attr) | |
660 | flags |= FS_XFLAG_HASATTR; | |
661 | ||
1da177e4 LT |
662 | return flags; |
663 | } | |
664 | ||
665 | uint | |
666 | xfs_ip2xflags( | |
58f88ca2 | 667 | struct xfs_inode *ip) |
1da177e4 | 668 | { |
58f88ca2 | 669 | struct xfs_icdinode *dic = &ip->i_d; |
1da177e4 | 670 | |
58f88ca2 | 671 | return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); |
1da177e4 LT |
672 | } |
673 | ||
c24b5dfa DC |
674 | /* |
675 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
676 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
677 | * ci_name->name will point to a the actual name (caller must free) or | |
678 | * will be set to NULL if an exact match is found. | |
679 | */ | |
680 | int | |
681 | xfs_lookup( | |
682 | xfs_inode_t *dp, | |
683 | struct xfs_name *name, | |
684 | xfs_inode_t **ipp, | |
685 | struct xfs_name *ci_name) | |
686 | { | |
687 | xfs_ino_t inum; | |
688 | int error; | |
c24b5dfa DC |
689 | |
690 | trace_xfs_lookup(dp, name); | |
691 | ||
692 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
2451337d | 693 | return -EIO; |
c24b5dfa | 694 | |
c24b5dfa | 695 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
c24b5dfa | 696 | if (error) |
dbad7c99 | 697 | goto out_unlock; |
c24b5dfa DC |
698 | |
699 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); | |
700 | if (error) | |
701 | goto out_free_name; | |
702 | ||
703 | return 0; | |
704 | ||
705 | out_free_name: | |
706 | if (ci_name) | |
707 | kmem_free(ci_name->name); | |
dbad7c99 | 708 | out_unlock: |
c24b5dfa DC |
709 | *ipp = NULL; |
710 | return error; | |
711 | } | |
712 | ||
1da177e4 LT |
713 | /* |
714 | * Allocate an inode on disk and return a copy of its in-core version. | |
715 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
716 | * appropriately within the inode. The uid and gid for the inode are | |
717 | * set according to the contents of the given cred structure. | |
718 | * | |
719 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
720 | * has a free inode available, call xfs_iget() to obtain the in-core |
721 | * version of the allocated inode. Finally, fill in the inode and | |
722 | * log its initial contents. In this case, ialloc_context would be | |
723 | * set to NULL. | |
1da177e4 | 724 | * |
cd856db6 CM |
725 | * If xfs_dialloc() does not have an available inode, it will replenish |
726 | * its supply by doing an allocation. Since we can only do one | |
727 | * allocation within a transaction without deadlocks, we must commit | |
728 | * the current transaction before returning the inode itself. | |
729 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
730 | * The caller should then commit the current transaction, start a new |
731 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
732 | * | |
733 | * To ensure that some other process does not grab the inode that | |
734 | * was allocated during the first call to xfs_ialloc(), this routine | |
735 | * also returns the [locked] bp pointing to the head of the freelist | |
736 | * as ialloc_context. The caller should hold this buffer across | |
737 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
738 | * |
739 | * If we are allocating quota inodes, we do not have a parent inode | |
740 | * to attach to or associate with (i.e. pip == NULL) because they | |
741 | * are not linked into the directory structure - they are attached | |
742 | * directly to the superblock - and so have no parent. | |
1da177e4 | 743 | */ |
0d5a75e9 | 744 | static int |
1da177e4 LT |
745 | xfs_ialloc( |
746 | xfs_trans_t *tp, | |
747 | xfs_inode_t *pip, | |
576b1d67 | 748 | umode_t mode, |
31b084ae | 749 | xfs_nlink_t nlink, |
66f36464 | 750 | dev_t rdev, |
6743099c | 751 | prid_t prid, |
1da177e4 | 752 | xfs_buf_t **ialloc_context, |
1da177e4 LT |
753 | xfs_inode_t **ipp) |
754 | { | |
93848a99 | 755 | struct xfs_mount *mp = tp->t_mountp; |
1da177e4 LT |
756 | xfs_ino_t ino; |
757 | xfs_inode_t *ip; | |
1da177e4 LT |
758 | uint flags; |
759 | int error; | |
e076b0f3 | 760 | struct timespec tv; |
3987848c | 761 | struct inode *inode; |
1da177e4 LT |
762 | |
763 | /* | |
764 | * Call the space management code to pick | |
765 | * the on-disk inode to be allocated. | |
766 | */ | |
f59cf5c2 | 767 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, |
08358906 | 768 | ialloc_context, &ino); |
bf904248 | 769 | if (error) |
1da177e4 | 770 | return error; |
08358906 | 771 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
772 | *ipp = NULL; |
773 | return 0; | |
774 | } | |
775 | ASSERT(*ialloc_context == NULL); | |
776 | ||
777 | /* | |
778 | * Get the in-core inode with the lock held exclusively. | |
779 | * This is because we're setting fields here we need | |
780 | * to prevent others from looking at until we're done. | |
781 | */ | |
93848a99 | 782 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
ec3ba85f | 783 | XFS_ILOCK_EXCL, &ip); |
bf904248 | 784 | if (error) |
1da177e4 | 785 | return error; |
1da177e4 | 786 | ASSERT(ip != NULL); |
3987848c | 787 | inode = VFS_I(ip); |
1da177e4 | 788 | |
263997a6 DC |
789 | /* |
790 | * We always convert v1 inodes to v2 now - we only support filesystems | |
791 | * with >= v2 inode capability, so there is no reason for ever leaving | |
792 | * an inode in v1 format. | |
793 | */ | |
794 | if (ip->i_d.di_version == 1) | |
795 | ip->i_d.di_version = 2; | |
796 | ||
c19b3b05 | 797 | inode->i_mode = mode; |
54d7b5c1 | 798 | set_nlink(inode, nlink); |
7aab1b28 DE |
799 | ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); |
800 | ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); | |
66f36464 | 801 | inode->i_rdev = rdev; |
6743099c | 802 | xfs_set_projid(ip, prid); |
1da177e4 | 803 | |
bd186aa9 | 804 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 805 | ip->i_d.di_gid = pip->i_d.di_gid; |
c19b3b05 DC |
806 | if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) |
807 | inode->i_mode |= S_ISGID; | |
1da177e4 LT |
808 | } |
809 | ||
810 | /* | |
811 | * If the group ID of the new file does not match the effective group | |
812 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
813 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
814 | */ | |
815 | if ((irix_sgid_inherit) && | |
c19b3b05 DC |
816 | (inode->i_mode & S_ISGID) && |
817 | (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) | |
818 | inode->i_mode &= ~S_ISGID; | |
1da177e4 LT |
819 | |
820 | ip->i_d.di_size = 0; | |
821 | ip->i_d.di_nextents = 0; | |
822 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 | 823 | |
c2050a45 | 824 | tv = current_time(inode); |
3987848c DC |
825 | inode->i_mtime = tv; |
826 | inode->i_atime = tv; | |
827 | inode->i_ctime = tv; | |
dff35fd4 | 828 | |
1da177e4 LT |
829 | ip->i_d.di_extsize = 0; |
830 | ip->i_d.di_dmevmask = 0; | |
831 | ip->i_d.di_dmstate = 0; | |
832 | ip->i_d.di_flags = 0; | |
93848a99 CH |
833 | |
834 | if (ip->i_d.di_version == 3) { | |
83e06f21 | 835 | inode->i_version = 1; |
93848a99 | 836 | ip->i_d.di_flags2 = 0; |
f7ca3522 | 837 | ip->i_d.di_cowextsize = 0; |
c8ce540d DW |
838 | ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec; |
839 | ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec; | |
93848a99 CH |
840 | } |
841 | ||
842 | ||
1da177e4 LT |
843 | flags = XFS_ILOG_CORE; |
844 | switch (mode & S_IFMT) { | |
845 | case S_IFIFO: | |
846 | case S_IFCHR: | |
847 | case S_IFBLK: | |
848 | case S_IFSOCK: | |
849 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1da177e4 LT |
850 | ip->i_df.if_flags = 0; |
851 | flags |= XFS_ILOG_DEV; | |
852 | break; | |
853 | case S_IFREG: | |
854 | case S_IFDIR: | |
b11f94d5 | 855 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
58f88ca2 | 856 | uint di_flags = 0; |
365ca83d | 857 | |
abbede1b | 858 | if (S_ISDIR(mode)) { |
365ca83d NS |
859 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
860 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
861 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
862 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
863 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
864 | } | |
9336e3a7 DC |
865 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
866 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
abbede1b | 867 | } else if (S_ISREG(mode)) { |
613d7043 | 868 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 869 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
870 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
871 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
872 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
873 | } | |
1da177e4 LT |
874 | } |
875 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
876 | xfs_inherit_noatime) | |
365ca83d | 877 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
878 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
879 | xfs_inherit_nodump) | |
365ca83d | 880 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
881 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
882 | xfs_inherit_sync) | |
365ca83d | 883 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
884 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
885 | xfs_inherit_nosymlinks) | |
365ca83d | 886 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
d3446eac BN |
887 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
888 | xfs_inherit_nodefrag) | |
889 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
890 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
891 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
58f88ca2 | 892 | |
365ca83d | 893 | ip->i_d.di_flags |= di_flags; |
1da177e4 | 894 | } |
f7ca3522 DW |
895 | if (pip && |
896 | (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) && | |
897 | pip->i_d.di_version == 3 && | |
898 | ip->i_d.di_version == 3) { | |
56bdf855 LC |
899 | uint64_t di_flags2 = 0; |
900 | ||
f7ca3522 | 901 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) { |
56bdf855 | 902 | di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; |
f7ca3522 DW |
903 | ip->i_d.di_cowextsize = pip->i_d.di_cowextsize; |
904 | } | |
56bdf855 LC |
905 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) |
906 | di_flags2 |= XFS_DIFLAG2_DAX; | |
907 | ||
908 | ip->i_d.di_flags2 |= di_flags2; | |
f7ca3522 | 909 | } |
1da177e4 LT |
910 | /* FALLTHROUGH */ |
911 | case S_IFLNK: | |
912 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
913 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
914 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
6bdcf26a | 915 | ip->i_df.if_u1.if_root = NULL; |
1da177e4 LT |
916 | break; |
917 | default: | |
918 | ASSERT(0); | |
919 | } | |
920 | /* | |
921 | * Attribute fork settings for new inode. | |
922 | */ | |
923 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
924 | ip->i_d.di_anextents = 0; | |
925 | ||
926 | /* | |
927 | * Log the new values stuffed into the inode. | |
928 | */ | |
ddc3415a | 929 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
930 | xfs_trans_log_inode(tp, ip, flags); |
931 | ||
58c90473 | 932 | /* now that we have an i_mode we can setup the inode structure */ |
41be8bed | 933 | xfs_setup_inode(ip); |
1da177e4 LT |
934 | |
935 | *ipp = ip; | |
936 | return 0; | |
937 | } | |
938 | ||
e546cb79 DC |
939 | /* |
940 | * Allocates a new inode from disk and return a pointer to the | |
941 | * incore copy. This routine will internally commit the current | |
942 | * transaction and allocate a new one if the Space Manager needed | |
943 | * to do an allocation to replenish the inode free-list. | |
944 | * | |
945 | * This routine is designed to be called from xfs_create and | |
946 | * xfs_create_dir. | |
947 | * | |
948 | */ | |
949 | int | |
950 | xfs_dir_ialloc( | |
951 | xfs_trans_t **tpp, /* input: current transaction; | |
952 | output: may be a new transaction. */ | |
953 | xfs_inode_t *dp, /* directory within whose allocate | |
954 | the inode. */ | |
955 | umode_t mode, | |
956 | xfs_nlink_t nlink, | |
66f36464 | 957 | dev_t rdev, |
e546cb79 | 958 | prid_t prid, /* project id */ |
e546cb79 DC |
959 | xfs_inode_t **ipp, /* pointer to inode; it will be |
960 | locked. */ | |
961 | int *committed) | |
962 | ||
963 | { | |
964 | xfs_trans_t *tp; | |
e546cb79 DC |
965 | xfs_inode_t *ip; |
966 | xfs_buf_t *ialloc_context = NULL; | |
967 | int code; | |
e546cb79 DC |
968 | void *dqinfo; |
969 | uint tflags; | |
970 | ||
971 | tp = *tpp; | |
972 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
973 | ||
974 | /* | |
975 | * xfs_ialloc will return a pointer to an incore inode if | |
976 | * the Space Manager has an available inode on the free | |
977 | * list. Otherwise, it will do an allocation and replenish | |
978 | * the freelist. Since we can only do one allocation per | |
979 | * transaction without deadlocks, we will need to commit the | |
980 | * current transaction and start a new one. We will then | |
981 | * need to call xfs_ialloc again to get the inode. | |
982 | * | |
983 | * If xfs_ialloc did an allocation to replenish the freelist, | |
984 | * it returns the bp containing the head of the freelist as | |
985 | * ialloc_context. We will hold a lock on it across the | |
986 | * transaction commit so that no other process can steal | |
987 | * the inode(s) that we've just allocated. | |
988 | */ | |
f59cf5c2 CH |
989 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context, |
990 | &ip); | |
e546cb79 DC |
991 | |
992 | /* | |
993 | * Return an error if we were unable to allocate a new inode. | |
994 | * This should only happen if we run out of space on disk or | |
995 | * encounter a disk error. | |
996 | */ | |
997 | if (code) { | |
998 | *ipp = NULL; | |
999 | return code; | |
1000 | } | |
1001 | if (!ialloc_context && !ip) { | |
1002 | *ipp = NULL; | |
2451337d | 1003 | return -ENOSPC; |
e546cb79 DC |
1004 | } |
1005 | ||
1006 | /* | |
1007 | * If the AGI buffer is non-NULL, then we were unable to get an | |
1008 | * inode in one operation. We need to commit the current | |
1009 | * transaction and call xfs_ialloc() again. It is guaranteed | |
1010 | * to succeed the second time. | |
1011 | */ | |
1012 | if (ialloc_context) { | |
1013 | /* | |
1014 | * Normally, xfs_trans_commit releases all the locks. | |
1015 | * We call bhold to hang on to the ialloc_context across | |
1016 | * the commit. Holding this buffer prevents any other | |
1017 | * processes from doing any allocations in this | |
1018 | * allocation group. | |
1019 | */ | |
1020 | xfs_trans_bhold(tp, ialloc_context); | |
e546cb79 DC |
1021 | |
1022 | /* | |
1023 | * We want the quota changes to be associated with the next | |
1024 | * transaction, NOT this one. So, detach the dqinfo from this | |
1025 | * and attach it to the next transaction. | |
1026 | */ | |
1027 | dqinfo = NULL; | |
1028 | tflags = 0; | |
1029 | if (tp->t_dqinfo) { | |
1030 | dqinfo = (void *)tp->t_dqinfo; | |
1031 | tp->t_dqinfo = NULL; | |
1032 | tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; | |
1033 | tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); | |
1034 | } | |
1035 | ||
411350df | 1036 | code = xfs_trans_roll(&tp); |
2e6db6c4 | 1037 | if (committed != NULL) |
e546cb79 | 1038 | *committed = 1; |
3d3c8b52 | 1039 | |
e546cb79 DC |
1040 | /* |
1041 | * Re-attach the quota info that we detached from prev trx. | |
1042 | */ | |
1043 | if (dqinfo) { | |
1044 | tp->t_dqinfo = dqinfo; | |
1045 | tp->t_flags |= tflags; | |
1046 | } | |
1047 | ||
1048 | if (code) { | |
1049 | xfs_buf_relse(ialloc_context); | |
2e6db6c4 | 1050 | *tpp = tp; |
e546cb79 DC |
1051 | *ipp = NULL; |
1052 | return code; | |
1053 | } | |
1054 | xfs_trans_bjoin(tp, ialloc_context); | |
1055 | ||
1056 | /* | |
1057 | * Call ialloc again. Since we've locked out all | |
1058 | * other allocations in this allocation group, | |
1059 | * this call should always succeed. | |
1060 | */ | |
1061 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, | |
f59cf5c2 | 1062 | &ialloc_context, &ip); |
e546cb79 DC |
1063 | |
1064 | /* | |
1065 | * If we get an error at this point, return to the caller | |
1066 | * so that the current transaction can be aborted. | |
1067 | */ | |
1068 | if (code) { | |
1069 | *tpp = tp; | |
1070 | *ipp = NULL; | |
1071 | return code; | |
1072 | } | |
1073 | ASSERT(!ialloc_context && ip); | |
1074 | ||
1075 | } else { | |
1076 | if (committed != NULL) | |
1077 | *committed = 0; | |
1078 | } | |
1079 | ||
1080 | *ipp = ip; | |
1081 | *tpp = tp; | |
1082 | ||
1083 | return 0; | |
1084 | } | |
1085 | ||
1086 | /* | |
54d7b5c1 DC |
1087 | * Decrement the link count on an inode & log the change. If this causes the |
1088 | * link count to go to zero, move the inode to AGI unlinked list so that it can | |
1089 | * be freed when the last active reference goes away via xfs_inactive(). | |
e546cb79 | 1090 | */ |
0d5a75e9 | 1091 | static int /* error */ |
e546cb79 DC |
1092 | xfs_droplink( |
1093 | xfs_trans_t *tp, | |
1094 | xfs_inode_t *ip) | |
1095 | { | |
e546cb79 DC |
1096 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); |
1097 | ||
e546cb79 DC |
1098 | drop_nlink(VFS_I(ip)); |
1099 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1100 | ||
54d7b5c1 DC |
1101 | if (VFS_I(ip)->i_nlink) |
1102 | return 0; | |
1103 | ||
1104 | return xfs_iunlink(tp, ip); | |
e546cb79 DC |
1105 | } |
1106 | ||
e546cb79 DC |
1107 | /* |
1108 | * Increment the link count on an inode & log the change. | |
1109 | */ | |
0d5a75e9 | 1110 | static int |
e546cb79 DC |
1111 | xfs_bumplink( |
1112 | xfs_trans_t *tp, | |
1113 | xfs_inode_t *ip) | |
1114 | { | |
1115 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); | |
1116 | ||
263997a6 | 1117 | ASSERT(ip->i_d.di_version > 1); |
e546cb79 | 1118 | inc_nlink(VFS_I(ip)); |
e546cb79 DC |
1119 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1120 | return 0; | |
1121 | } | |
1122 | ||
c24b5dfa DC |
1123 | int |
1124 | xfs_create( | |
1125 | xfs_inode_t *dp, | |
1126 | struct xfs_name *name, | |
1127 | umode_t mode, | |
66f36464 | 1128 | dev_t rdev, |
c24b5dfa DC |
1129 | xfs_inode_t **ipp) |
1130 | { | |
1131 | int is_dir = S_ISDIR(mode); | |
1132 | struct xfs_mount *mp = dp->i_mount; | |
1133 | struct xfs_inode *ip = NULL; | |
1134 | struct xfs_trans *tp = NULL; | |
1135 | int error; | |
2c3234d1 | 1136 | struct xfs_defer_ops dfops; |
c24b5dfa DC |
1137 | xfs_fsblock_t first_block; |
1138 | bool unlock_dp_on_error = false; | |
c24b5dfa DC |
1139 | prid_t prid; |
1140 | struct xfs_dquot *udqp = NULL; | |
1141 | struct xfs_dquot *gdqp = NULL; | |
1142 | struct xfs_dquot *pdqp = NULL; | |
062647a8 | 1143 | struct xfs_trans_res *tres; |
c24b5dfa | 1144 | uint resblks; |
c24b5dfa DC |
1145 | |
1146 | trace_xfs_create(dp, name); | |
1147 | ||
1148 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1149 | return -EIO; |
c24b5dfa | 1150 | |
163467d3 | 1151 | prid = xfs_get_initial_prid(dp); |
c24b5dfa DC |
1152 | |
1153 | /* | |
1154 | * Make sure that we have allocated dquot(s) on disk. | |
1155 | */ | |
7aab1b28 DE |
1156 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), |
1157 | xfs_kgid_to_gid(current_fsgid()), prid, | |
c24b5dfa DC |
1158 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
1159 | &udqp, &gdqp, &pdqp); | |
1160 | if (error) | |
1161 | return error; | |
1162 | ||
1163 | if (is_dir) { | |
c24b5dfa | 1164 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); |
062647a8 | 1165 | tres = &M_RES(mp)->tr_mkdir; |
c24b5dfa DC |
1166 | } else { |
1167 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
062647a8 | 1168 | tres = &M_RES(mp)->tr_create; |
c24b5dfa DC |
1169 | } |
1170 | ||
c24b5dfa DC |
1171 | /* |
1172 | * Initially assume that the file does not exist and | |
1173 | * reserve the resources for that case. If that is not | |
1174 | * the case we'll drop the one we have and get a more | |
1175 | * appropriate transaction later. | |
1176 | */ | |
253f4911 | 1177 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); |
2451337d | 1178 | if (error == -ENOSPC) { |
c24b5dfa DC |
1179 | /* flush outstanding delalloc blocks and retry */ |
1180 | xfs_flush_inodes(mp); | |
253f4911 | 1181 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); |
c24b5dfa | 1182 | } |
4906e215 | 1183 | if (error) |
253f4911 | 1184 | goto out_release_inode; |
c24b5dfa | 1185 | |
65523218 | 1186 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
c24b5dfa DC |
1187 | unlock_dp_on_error = true; |
1188 | ||
2c3234d1 | 1189 | xfs_defer_init(&dfops, &first_block); |
c24b5dfa DC |
1190 | |
1191 | /* | |
1192 | * Reserve disk quota and the inode. | |
1193 | */ | |
1194 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1195 | pdqp, resblks, 1, 0); | |
1196 | if (error) | |
1197 | goto out_trans_cancel; | |
1198 | ||
c24b5dfa DC |
1199 | /* |
1200 | * A newly created regular or special file just has one directory | |
1201 | * entry pointing to them, but a directory also the "." entry | |
1202 | * pointing to itself. | |
1203 | */ | |
f59cf5c2 CH |
1204 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip, |
1205 | NULL); | |
d6077aa3 | 1206 | if (error) |
4906e215 | 1207 | goto out_trans_cancel; |
c24b5dfa DC |
1208 | |
1209 | /* | |
1210 | * Now we join the directory inode to the transaction. We do not do it | |
1211 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1212 | * (and release all the locks). An error from here on will result in | |
1213 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1214 | * error path. | |
1215 | */ | |
65523218 | 1216 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1217 | unlock_dp_on_error = false; |
1218 | ||
1219 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, | |
2c3234d1 | 1220 | &first_block, &dfops, resblks ? |
c24b5dfa DC |
1221 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); |
1222 | if (error) { | |
2451337d | 1223 | ASSERT(error != -ENOSPC); |
4906e215 | 1224 | goto out_trans_cancel; |
c24b5dfa DC |
1225 | } |
1226 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1227 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1228 | ||
1229 | if (is_dir) { | |
1230 | error = xfs_dir_init(tp, ip, dp); | |
1231 | if (error) | |
1232 | goto out_bmap_cancel; | |
1233 | ||
1234 | error = xfs_bumplink(tp, dp); | |
1235 | if (error) | |
1236 | goto out_bmap_cancel; | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * If this is a synchronous mount, make sure that the | |
1241 | * create transaction goes to disk before returning to | |
1242 | * the user. | |
1243 | */ | |
1244 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
1245 | xfs_trans_set_sync(tp); | |
1246 | ||
1247 | /* | |
1248 | * Attach the dquot(s) to the inodes and modify them incore. | |
1249 | * These ids of the inode couldn't have changed since the new | |
1250 | * inode has been locked ever since it was created. | |
1251 | */ | |
1252 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1253 | ||
8ad7c629 | 1254 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa DC |
1255 | if (error) |
1256 | goto out_bmap_cancel; | |
1257 | ||
70393313 | 1258 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
1259 | if (error) |
1260 | goto out_release_inode; | |
1261 | ||
1262 | xfs_qm_dqrele(udqp); | |
1263 | xfs_qm_dqrele(gdqp); | |
1264 | xfs_qm_dqrele(pdqp); | |
1265 | ||
1266 | *ipp = ip; | |
1267 | return 0; | |
1268 | ||
1269 | out_bmap_cancel: | |
2c3234d1 | 1270 | xfs_defer_cancel(&dfops); |
c24b5dfa | 1271 | out_trans_cancel: |
4906e215 | 1272 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1273 | out_release_inode: |
1274 | /* | |
58c90473 DC |
1275 | * Wait until after the current transaction is aborted to finish the |
1276 | * setup of the inode and release the inode. This prevents recursive | |
1277 | * transactions and deadlocks from xfs_inactive. | |
c24b5dfa | 1278 | */ |
58c90473 DC |
1279 | if (ip) { |
1280 | xfs_finish_inode_setup(ip); | |
c24b5dfa | 1281 | IRELE(ip); |
58c90473 | 1282 | } |
c24b5dfa DC |
1283 | |
1284 | xfs_qm_dqrele(udqp); | |
1285 | xfs_qm_dqrele(gdqp); | |
1286 | xfs_qm_dqrele(pdqp); | |
1287 | ||
1288 | if (unlock_dp_on_error) | |
65523218 | 1289 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1290 | return error; |
1291 | } | |
1292 | ||
99b6436b ZYW |
1293 | int |
1294 | xfs_create_tmpfile( | |
1295 | struct xfs_inode *dp, | |
1296 | struct dentry *dentry, | |
330033d6 BF |
1297 | umode_t mode, |
1298 | struct xfs_inode **ipp) | |
99b6436b ZYW |
1299 | { |
1300 | struct xfs_mount *mp = dp->i_mount; | |
1301 | struct xfs_inode *ip = NULL; | |
1302 | struct xfs_trans *tp = NULL; | |
1303 | int error; | |
99b6436b ZYW |
1304 | prid_t prid; |
1305 | struct xfs_dquot *udqp = NULL; | |
1306 | struct xfs_dquot *gdqp = NULL; | |
1307 | struct xfs_dquot *pdqp = NULL; | |
1308 | struct xfs_trans_res *tres; | |
1309 | uint resblks; | |
1310 | ||
1311 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1312 | return -EIO; |
99b6436b ZYW |
1313 | |
1314 | prid = xfs_get_initial_prid(dp); | |
1315 | ||
1316 | /* | |
1317 | * Make sure that we have allocated dquot(s) on disk. | |
1318 | */ | |
1319 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), | |
1320 | xfs_kgid_to_gid(current_fsgid()), prid, | |
1321 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, | |
1322 | &udqp, &gdqp, &pdqp); | |
1323 | if (error) | |
1324 | return error; | |
1325 | ||
1326 | resblks = XFS_IALLOC_SPACE_RES(mp); | |
99b6436b | 1327 | tres = &M_RES(mp)->tr_create_tmpfile; |
253f4911 CH |
1328 | |
1329 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); | |
4906e215 | 1330 | if (error) |
253f4911 | 1331 | goto out_release_inode; |
99b6436b ZYW |
1332 | |
1333 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1334 | pdqp, resblks, 1, 0); | |
1335 | if (error) | |
1336 | goto out_trans_cancel; | |
1337 | ||
f59cf5c2 | 1338 | error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, prid, &ip, NULL); |
d6077aa3 | 1339 | if (error) |
4906e215 | 1340 | goto out_trans_cancel; |
99b6436b ZYW |
1341 | |
1342 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1343 | xfs_trans_set_sync(tp); | |
1344 | ||
1345 | /* | |
1346 | * Attach the dquot(s) to the inodes and modify them incore. | |
1347 | * These ids of the inode couldn't have changed since the new | |
1348 | * inode has been locked ever since it was created. | |
1349 | */ | |
1350 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1351 | ||
99b6436b ZYW |
1352 | error = xfs_iunlink(tp, ip); |
1353 | if (error) | |
4906e215 | 1354 | goto out_trans_cancel; |
99b6436b | 1355 | |
70393313 | 1356 | error = xfs_trans_commit(tp); |
99b6436b ZYW |
1357 | if (error) |
1358 | goto out_release_inode; | |
1359 | ||
1360 | xfs_qm_dqrele(udqp); | |
1361 | xfs_qm_dqrele(gdqp); | |
1362 | xfs_qm_dqrele(pdqp); | |
1363 | ||
330033d6 | 1364 | *ipp = ip; |
99b6436b ZYW |
1365 | return 0; |
1366 | ||
99b6436b | 1367 | out_trans_cancel: |
4906e215 | 1368 | xfs_trans_cancel(tp); |
99b6436b ZYW |
1369 | out_release_inode: |
1370 | /* | |
58c90473 DC |
1371 | * Wait until after the current transaction is aborted to finish the |
1372 | * setup of the inode and release the inode. This prevents recursive | |
1373 | * transactions and deadlocks from xfs_inactive. | |
99b6436b | 1374 | */ |
58c90473 DC |
1375 | if (ip) { |
1376 | xfs_finish_inode_setup(ip); | |
99b6436b | 1377 | IRELE(ip); |
58c90473 | 1378 | } |
99b6436b ZYW |
1379 | |
1380 | xfs_qm_dqrele(udqp); | |
1381 | xfs_qm_dqrele(gdqp); | |
1382 | xfs_qm_dqrele(pdqp); | |
1383 | ||
1384 | return error; | |
1385 | } | |
1386 | ||
c24b5dfa DC |
1387 | int |
1388 | xfs_link( | |
1389 | xfs_inode_t *tdp, | |
1390 | xfs_inode_t *sip, | |
1391 | struct xfs_name *target_name) | |
1392 | { | |
1393 | xfs_mount_t *mp = tdp->i_mount; | |
1394 | xfs_trans_t *tp; | |
1395 | int error; | |
2c3234d1 | 1396 | struct xfs_defer_ops dfops; |
c24b5dfa | 1397 | xfs_fsblock_t first_block; |
c24b5dfa DC |
1398 | int resblks; |
1399 | ||
1400 | trace_xfs_link(tdp, target_name); | |
1401 | ||
c19b3b05 | 1402 | ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); |
c24b5dfa DC |
1403 | |
1404 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1405 | return -EIO; |
c24b5dfa DC |
1406 | |
1407 | error = xfs_qm_dqattach(sip, 0); | |
1408 | if (error) | |
1409 | goto std_return; | |
1410 | ||
1411 | error = xfs_qm_dqattach(tdp, 0); | |
1412 | if (error) | |
1413 | goto std_return; | |
1414 | ||
c24b5dfa | 1415 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
253f4911 | 1416 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); |
2451337d | 1417 | if (error == -ENOSPC) { |
c24b5dfa | 1418 | resblks = 0; |
253f4911 | 1419 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); |
c24b5dfa | 1420 | } |
4906e215 | 1421 | if (error) |
253f4911 | 1422 | goto std_return; |
c24b5dfa DC |
1423 | |
1424 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); | |
1425 | ||
1426 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
65523218 | 1427 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1428 | |
1429 | /* | |
1430 | * If we are using project inheritance, we only allow hard link | |
1431 | * creation in our tree when the project IDs are the same; else | |
1432 | * the tree quota mechanism could be circumvented. | |
1433 | */ | |
1434 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
1435 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { | |
2451337d | 1436 | error = -EXDEV; |
c24b5dfa DC |
1437 | goto error_return; |
1438 | } | |
1439 | ||
94f3cad5 ES |
1440 | if (!resblks) { |
1441 | error = xfs_dir_canenter(tp, tdp, target_name); | |
1442 | if (error) | |
1443 | goto error_return; | |
1444 | } | |
c24b5dfa | 1445 | |
2c3234d1 | 1446 | xfs_defer_init(&dfops, &first_block); |
c24b5dfa | 1447 | |
54d7b5c1 DC |
1448 | /* |
1449 | * Handle initial link state of O_TMPFILE inode | |
1450 | */ | |
1451 | if (VFS_I(sip)->i_nlink == 0) { | |
ab297431 ZYW |
1452 | error = xfs_iunlink_remove(tp, sip); |
1453 | if (error) | |
4906e215 | 1454 | goto error_return; |
ab297431 ZYW |
1455 | } |
1456 | ||
c24b5dfa | 1457 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
2c3234d1 | 1458 | &first_block, &dfops, resblks); |
c24b5dfa | 1459 | if (error) |
4906e215 | 1460 | goto error_return; |
c24b5dfa DC |
1461 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1462 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
1463 | ||
1464 | error = xfs_bumplink(tp, sip); | |
1465 | if (error) | |
4906e215 | 1466 | goto error_return; |
c24b5dfa DC |
1467 | |
1468 | /* | |
1469 | * If this is a synchronous mount, make sure that the | |
1470 | * link transaction goes to disk before returning to | |
1471 | * the user. | |
1472 | */ | |
f6106efa | 1473 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
c24b5dfa | 1474 | xfs_trans_set_sync(tp); |
c24b5dfa | 1475 | |
8ad7c629 | 1476 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa | 1477 | if (error) { |
2c3234d1 | 1478 | xfs_defer_cancel(&dfops); |
4906e215 | 1479 | goto error_return; |
c24b5dfa DC |
1480 | } |
1481 | ||
70393313 | 1482 | return xfs_trans_commit(tp); |
c24b5dfa | 1483 | |
c24b5dfa | 1484 | error_return: |
4906e215 | 1485 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1486 | std_return: |
1487 | return error; | |
1488 | } | |
1489 | ||
363e59ba DW |
1490 | /* Clear the reflink flag and the cowblocks tag if possible. */ |
1491 | static void | |
1492 | xfs_itruncate_clear_reflink_flags( | |
1493 | struct xfs_inode *ip) | |
1494 | { | |
1495 | struct xfs_ifork *dfork; | |
1496 | struct xfs_ifork *cfork; | |
1497 | ||
1498 | if (!xfs_is_reflink_inode(ip)) | |
1499 | return; | |
1500 | dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
1501 | cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
1502 | if (dfork->if_bytes == 0 && cfork->if_bytes == 0) | |
1503 | ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; | |
1504 | if (cfork->if_bytes == 0) | |
1505 | xfs_inode_clear_cowblocks_tag(ip); | |
1506 | } | |
1507 | ||
1da177e4 | 1508 | /* |
8f04c47a CH |
1509 | * Free up the underlying blocks past new_size. The new size must be smaller |
1510 | * than the current size. This routine can be used both for the attribute and | |
1511 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1512 | * |
f6485057 DC |
1513 | * The transaction passed to this routine must have made a permanent log |
1514 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1515 | * given transaction and start new ones, so make sure everything involved in | |
1516 | * the transaction is tidy before calling here. Some transaction will be | |
1517 | * returned to the caller to be committed. The incoming transaction must | |
1518 | * already include the inode, and both inode locks must be held exclusively. | |
1519 | * The inode must also be "held" within the transaction. On return the inode | |
1520 | * will be "held" within the returned transaction. This routine does NOT | |
1521 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1522 | * |
f6485057 DC |
1523 | * If we get an error, we must return with the inode locked and linked into the |
1524 | * current transaction. This keeps things simple for the higher level code, | |
1525 | * because it always knows that the inode is locked and held in the transaction | |
1526 | * that returns to it whether errors occur or not. We don't mark the inode | |
1527 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1528 | */ |
1529 | int | |
8f04c47a CH |
1530 | xfs_itruncate_extents( |
1531 | struct xfs_trans **tpp, | |
1532 | struct xfs_inode *ip, | |
1533 | int whichfork, | |
1534 | xfs_fsize_t new_size) | |
1da177e4 | 1535 | { |
8f04c47a CH |
1536 | struct xfs_mount *mp = ip->i_mount; |
1537 | struct xfs_trans *tp = *tpp; | |
2c3234d1 | 1538 | struct xfs_defer_ops dfops; |
8f04c47a CH |
1539 | xfs_fsblock_t first_block; |
1540 | xfs_fileoff_t first_unmap_block; | |
1541 | xfs_fileoff_t last_block; | |
1542 | xfs_filblks_t unmap_len; | |
8f04c47a CH |
1543 | int error = 0; |
1544 | int done = 0; | |
1da177e4 | 1545 | |
0b56185b CH |
1546 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1547 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1548 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1549 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1550 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1551 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1552 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1553 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1554 | |
673e8e59 CH |
1555 | trace_xfs_itruncate_extents_start(ip, new_size); |
1556 | ||
1da177e4 LT |
1557 | /* |
1558 | * Since it is possible for space to become allocated beyond | |
1559 | * the end of the file (in a crash where the space is allocated | |
1560 | * but the inode size is not yet updated), simply remove any | |
1561 | * blocks which show up between the new EOF and the maximum | |
1562 | * possible file size. If the first block to be removed is | |
1563 | * beyond the maximum file size (ie it is the same as last_block), | |
1564 | * then there is nothing to do. | |
1565 | */ | |
8f04c47a | 1566 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1567 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1568 | if (first_unmap_block == last_block) |
1569 | return 0; | |
1570 | ||
1571 | ASSERT(first_unmap_block < last_block); | |
1572 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1573 | while (!done) { |
2c3234d1 | 1574 | xfs_defer_init(&dfops, &first_block); |
8f04c47a | 1575 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 1576 | first_unmap_block, unmap_len, |
8f04c47a | 1577 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 1578 | XFS_ITRUNC_MAX_EXTENTS, |
2c3234d1 | 1579 | &first_block, &dfops, |
b4e9181e | 1580 | &done); |
8f04c47a CH |
1581 | if (error) |
1582 | goto out_bmap_cancel; | |
1da177e4 LT |
1583 | |
1584 | /* | |
1585 | * Duplicate the transaction that has the permanent | |
1586 | * reservation and commit the old transaction. | |
1587 | */ | |
8ad7c629 CH |
1588 | xfs_defer_ijoin(&dfops, ip); |
1589 | error = xfs_defer_finish(&tp, &dfops); | |
8f04c47a CH |
1590 | if (error) |
1591 | goto out_bmap_cancel; | |
1da177e4 | 1592 | |
411350df | 1593 | error = xfs_trans_roll_inode(&tp, ip); |
f6485057 | 1594 | if (error) |
8f04c47a | 1595 | goto out; |
1da177e4 | 1596 | } |
8f04c47a | 1597 | |
aa8968f2 DW |
1598 | /* Remove all pending CoW reservations. */ |
1599 | error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block, | |
3802a345 | 1600 | last_block, true); |
aa8968f2 DW |
1601 | if (error) |
1602 | goto out; | |
1603 | ||
363e59ba | 1604 | xfs_itruncate_clear_reflink_flags(ip); |
aa8968f2 | 1605 | |
673e8e59 CH |
1606 | /* |
1607 | * Always re-log the inode so that our permanent transaction can keep | |
1608 | * on rolling it forward in the log. | |
1609 | */ | |
1610 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1611 | ||
1612 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1613 | ||
8f04c47a CH |
1614 | out: |
1615 | *tpp = tp; | |
1616 | return error; | |
1617 | out_bmap_cancel: | |
1da177e4 | 1618 | /* |
8f04c47a CH |
1619 | * If the bunmapi call encounters an error, return to the caller where |
1620 | * the transaction can be properly aborted. We just need to make sure | |
1621 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1622 | */ |
2c3234d1 | 1623 | xfs_defer_cancel(&dfops); |
8f04c47a CH |
1624 | goto out; |
1625 | } | |
1626 | ||
c24b5dfa DC |
1627 | int |
1628 | xfs_release( | |
1629 | xfs_inode_t *ip) | |
1630 | { | |
1631 | xfs_mount_t *mp = ip->i_mount; | |
1632 | int error; | |
1633 | ||
c19b3b05 | 1634 | if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) |
c24b5dfa DC |
1635 | return 0; |
1636 | ||
1637 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1638 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1639 | return 0; | |
1640 | ||
1641 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1642 | int truncated; | |
1643 | ||
c24b5dfa DC |
1644 | /* |
1645 | * If we previously truncated this file and removed old data | |
1646 | * in the process, we want to initiate "early" writeout on | |
1647 | * the last close. This is an attempt to combat the notorious | |
1648 | * NULL files problem which is particularly noticeable from a | |
1649 | * truncate down, buffered (re-)write (delalloc), followed by | |
1650 | * a crash. What we are effectively doing here is | |
1651 | * significantly reducing the time window where we'd otherwise | |
1652 | * be exposed to that problem. | |
1653 | */ | |
1654 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); | |
1655 | if (truncated) { | |
1656 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
eac152b4 | 1657 | if (ip->i_delayed_blks > 0) { |
2451337d | 1658 | error = filemap_flush(VFS_I(ip)->i_mapping); |
c24b5dfa DC |
1659 | if (error) |
1660 | return error; | |
1661 | } | |
1662 | } | |
1663 | } | |
1664 | ||
54d7b5c1 | 1665 | if (VFS_I(ip)->i_nlink == 0) |
c24b5dfa DC |
1666 | return 0; |
1667 | ||
1668 | if (xfs_can_free_eofblocks(ip, false)) { | |
1669 | ||
a36b9261 BF |
1670 | /* |
1671 | * Check if the inode is being opened, written and closed | |
1672 | * frequently and we have delayed allocation blocks outstanding | |
1673 | * (e.g. streaming writes from the NFS server), truncating the | |
1674 | * blocks past EOF will cause fragmentation to occur. | |
1675 | * | |
1676 | * In this case don't do the truncation, but we have to be | |
1677 | * careful how we detect this case. Blocks beyond EOF show up as | |
1678 | * i_delayed_blks even when the inode is clean, so we need to | |
1679 | * truncate them away first before checking for a dirty release. | |
1680 | * Hence on the first dirty close we will still remove the | |
1681 | * speculative allocation, but after that we will leave it in | |
1682 | * place. | |
1683 | */ | |
1684 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
1685 | return 0; | |
c24b5dfa DC |
1686 | /* |
1687 | * If we can't get the iolock just skip truncating the blocks | |
1688 | * past EOF because we could deadlock with the mmap_sem | |
a36b9261 | 1689 | * otherwise. We'll get another chance to drop them once the |
c24b5dfa DC |
1690 | * last reference to the inode is dropped, so we'll never leak |
1691 | * blocks permanently. | |
c24b5dfa | 1692 | */ |
a36b9261 BF |
1693 | if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
1694 | error = xfs_free_eofblocks(ip); | |
1695 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1696 | if (error) | |
1697 | return error; | |
1698 | } | |
c24b5dfa DC |
1699 | |
1700 | /* delalloc blocks after truncation means it really is dirty */ | |
1701 | if (ip->i_delayed_blks) | |
1702 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
1703 | } | |
1704 | return 0; | |
1705 | } | |
1706 | ||
f7be2d7f BF |
1707 | /* |
1708 | * xfs_inactive_truncate | |
1709 | * | |
1710 | * Called to perform a truncate when an inode becomes unlinked. | |
1711 | */ | |
1712 | STATIC int | |
1713 | xfs_inactive_truncate( | |
1714 | struct xfs_inode *ip) | |
1715 | { | |
1716 | struct xfs_mount *mp = ip->i_mount; | |
1717 | struct xfs_trans *tp; | |
1718 | int error; | |
1719 | ||
253f4911 | 1720 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
f7be2d7f BF |
1721 | if (error) { |
1722 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
f7be2d7f BF |
1723 | return error; |
1724 | } | |
1725 | ||
1726 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1727 | xfs_trans_ijoin(tp, ip, 0); | |
1728 | ||
1729 | /* | |
1730 | * Log the inode size first to prevent stale data exposure in the event | |
1731 | * of a system crash before the truncate completes. See the related | |
69bca807 | 1732 | * comment in xfs_vn_setattr_size() for details. |
f7be2d7f BF |
1733 | */ |
1734 | ip->i_d.di_size = 0; | |
1735 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1736 | ||
1737 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
1738 | if (error) | |
1739 | goto error_trans_cancel; | |
1740 | ||
1741 | ASSERT(ip->i_d.di_nextents == 0); | |
1742 | ||
70393313 | 1743 | error = xfs_trans_commit(tp); |
f7be2d7f BF |
1744 | if (error) |
1745 | goto error_unlock; | |
1746 | ||
1747 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1748 | return 0; | |
1749 | ||
1750 | error_trans_cancel: | |
4906e215 | 1751 | xfs_trans_cancel(tp); |
f7be2d7f BF |
1752 | error_unlock: |
1753 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1754 | return error; | |
1755 | } | |
1756 | ||
88877d2b BF |
1757 | /* |
1758 | * xfs_inactive_ifree() | |
1759 | * | |
1760 | * Perform the inode free when an inode is unlinked. | |
1761 | */ | |
1762 | STATIC int | |
1763 | xfs_inactive_ifree( | |
1764 | struct xfs_inode *ip) | |
1765 | { | |
2c3234d1 | 1766 | struct xfs_defer_ops dfops; |
88877d2b | 1767 | xfs_fsblock_t first_block; |
88877d2b BF |
1768 | struct xfs_mount *mp = ip->i_mount; |
1769 | struct xfs_trans *tp; | |
1770 | int error; | |
1771 | ||
9d43b180 | 1772 | /* |
76d771b4 CH |
1773 | * We try to use a per-AG reservation for any block needed by the finobt |
1774 | * tree, but as the finobt feature predates the per-AG reservation | |
1775 | * support a degraded file system might not have enough space for the | |
1776 | * reservation at mount time. In that case try to dip into the reserved | |
1777 | * pool and pray. | |
9d43b180 BF |
1778 | * |
1779 | * Send a warning if the reservation does happen to fail, as the inode | |
1780 | * now remains allocated and sits on the unlinked list until the fs is | |
1781 | * repaired. | |
1782 | */ | |
76d771b4 CH |
1783 | if (unlikely(mp->m_inotbt_nores)) { |
1784 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, | |
1785 | XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, | |
1786 | &tp); | |
1787 | } else { | |
1788 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); | |
1789 | } | |
88877d2b | 1790 | if (error) { |
2451337d | 1791 | if (error == -ENOSPC) { |
9d43b180 BF |
1792 | xfs_warn_ratelimited(mp, |
1793 | "Failed to remove inode(s) from unlinked list. " | |
1794 | "Please free space, unmount and run xfs_repair."); | |
1795 | } else { | |
1796 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1797 | } | |
88877d2b BF |
1798 | return error; |
1799 | } | |
1800 | ||
1801 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1802 | xfs_trans_ijoin(tp, ip, 0); | |
1803 | ||
2c3234d1 DW |
1804 | xfs_defer_init(&dfops, &first_block); |
1805 | error = xfs_ifree(tp, ip, &dfops); | |
88877d2b BF |
1806 | if (error) { |
1807 | /* | |
1808 | * If we fail to free the inode, shut down. The cancel | |
1809 | * might do that, we need to make sure. Otherwise the | |
1810 | * inode might be lost for a long time or forever. | |
1811 | */ | |
1812 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1813 | xfs_notice(mp, "%s: xfs_ifree returned error %d", | |
1814 | __func__, error); | |
1815 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1816 | } | |
4906e215 | 1817 | xfs_trans_cancel(tp); |
88877d2b BF |
1818 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1819 | return error; | |
1820 | } | |
1821 | ||
1822 | /* | |
1823 | * Credit the quota account(s). The inode is gone. | |
1824 | */ | |
1825 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1826 | ||
1827 | /* | |
d4a97a04 BF |
1828 | * Just ignore errors at this point. There is nothing we can do except |
1829 | * to try to keep going. Make sure it's not a silent error. | |
88877d2b | 1830 | */ |
8ad7c629 | 1831 | error = xfs_defer_finish(&tp, &dfops); |
d4a97a04 | 1832 | if (error) { |
310a75a3 | 1833 | xfs_notice(mp, "%s: xfs_defer_finish returned error %d", |
88877d2b | 1834 | __func__, error); |
2c3234d1 | 1835 | xfs_defer_cancel(&dfops); |
d4a97a04 | 1836 | } |
70393313 | 1837 | error = xfs_trans_commit(tp); |
88877d2b BF |
1838 | if (error) |
1839 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", | |
1840 | __func__, error); | |
1841 | ||
1842 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1843 | return 0; | |
1844 | } | |
1845 | ||
c24b5dfa DC |
1846 | /* |
1847 | * xfs_inactive | |
1848 | * | |
1849 | * This is called when the vnode reference count for the vnode | |
1850 | * goes to zero. If the file has been unlinked, then it must | |
1851 | * now be truncated. Also, we clear all of the read-ahead state | |
1852 | * kept for the inode here since the file is now closed. | |
1853 | */ | |
74564fb4 | 1854 | void |
c24b5dfa DC |
1855 | xfs_inactive( |
1856 | xfs_inode_t *ip) | |
1857 | { | |
3d3c8b52 | 1858 | struct xfs_mount *mp; |
3d3c8b52 JL |
1859 | int error; |
1860 | int truncate = 0; | |
c24b5dfa DC |
1861 | |
1862 | /* | |
1863 | * If the inode is already free, then there can be nothing | |
1864 | * to clean up here. | |
1865 | */ | |
c19b3b05 | 1866 | if (VFS_I(ip)->i_mode == 0) { |
c24b5dfa DC |
1867 | ASSERT(ip->i_df.if_real_bytes == 0); |
1868 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
74564fb4 | 1869 | return; |
c24b5dfa DC |
1870 | } |
1871 | ||
1872 | mp = ip->i_mount; | |
17c12bcd | 1873 | ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); |
c24b5dfa | 1874 | |
c24b5dfa DC |
1875 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
1876 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
74564fb4 | 1877 | return; |
c24b5dfa | 1878 | |
54d7b5c1 | 1879 | if (VFS_I(ip)->i_nlink != 0) { |
c24b5dfa DC |
1880 | /* |
1881 | * force is true because we are evicting an inode from the | |
1882 | * cache. Post-eof blocks must be freed, lest we end up with | |
1883 | * broken free space accounting. | |
3b4683c2 BF |
1884 | * |
1885 | * Note: don't bother with iolock here since lockdep complains | |
1886 | * about acquiring it in reclaim context. We have the only | |
1887 | * reference to the inode at this point anyways. | |
c24b5dfa | 1888 | */ |
3b4683c2 | 1889 | if (xfs_can_free_eofblocks(ip, true)) |
a36b9261 | 1890 | xfs_free_eofblocks(ip); |
74564fb4 BF |
1891 | |
1892 | return; | |
c24b5dfa DC |
1893 | } |
1894 | ||
c19b3b05 | 1895 | if (S_ISREG(VFS_I(ip)->i_mode) && |
c24b5dfa DC |
1896 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || |
1897 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) | |
1898 | truncate = 1; | |
1899 | ||
1900 | error = xfs_qm_dqattach(ip, 0); | |
1901 | if (error) | |
74564fb4 | 1902 | return; |
c24b5dfa | 1903 | |
c19b3b05 | 1904 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
36b21dde | 1905 | error = xfs_inactive_symlink(ip); |
f7be2d7f BF |
1906 | else if (truncate) |
1907 | error = xfs_inactive_truncate(ip); | |
1908 | if (error) | |
74564fb4 | 1909 | return; |
c24b5dfa DC |
1910 | |
1911 | /* | |
1912 | * If there are attributes associated with the file then blow them away | |
1913 | * now. The code calls a routine that recursively deconstructs the | |
6dfe5a04 | 1914 | * attribute fork. If also blows away the in-core attribute fork. |
c24b5dfa | 1915 | */ |
6dfe5a04 | 1916 | if (XFS_IFORK_Q(ip)) { |
c24b5dfa DC |
1917 | error = xfs_attr_inactive(ip); |
1918 | if (error) | |
74564fb4 | 1919 | return; |
c24b5dfa DC |
1920 | } |
1921 | ||
6dfe5a04 | 1922 | ASSERT(!ip->i_afp); |
c24b5dfa | 1923 | ASSERT(ip->i_d.di_anextents == 0); |
6dfe5a04 | 1924 | ASSERT(ip->i_d.di_forkoff == 0); |
c24b5dfa DC |
1925 | |
1926 | /* | |
1927 | * Free the inode. | |
1928 | */ | |
88877d2b BF |
1929 | error = xfs_inactive_ifree(ip); |
1930 | if (error) | |
74564fb4 | 1931 | return; |
c24b5dfa DC |
1932 | |
1933 | /* | |
1934 | * Release the dquots held by inode, if any. | |
1935 | */ | |
1936 | xfs_qm_dqdetach(ip); | |
c24b5dfa DC |
1937 | } |
1938 | ||
1da177e4 | 1939 | /* |
54d7b5c1 DC |
1940 | * This is called when the inode's link count goes to 0 or we are creating a |
1941 | * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be | |
1942 | * set to true as the link count is dropped to zero by the VFS after we've | |
1943 | * created the file successfully, so we have to add it to the unlinked list | |
1944 | * while the link count is non-zero. | |
1945 | * | |
1946 | * We place the on-disk inode on a list in the AGI. It will be pulled from this | |
1947 | * list when the inode is freed. | |
1da177e4 | 1948 | */ |
54d7b5c1 | 1949 | STATIC int |
1da177e4 | 1950 | xfs_iunlink( |
54d7b5c1 DC |
1951 | struct xfs_trans *tp, |
1952 | struct xfs_inode *ip) | |
1da177e4 | 1953 | { |
54d7b5c1 | 1954 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1955 | xfs_agi_t *agi; |
1956 | xfs_dinode_t *dip; | |
1957 | xfs_buf_t *agibp; | |
1958 | xfs_buf_t *ibp; | |
1da177e4 LT |
1959 | xfs_agino_t agino; |
1960 | short bucket_index; | |
1961 | int offset; | |
1962 | int error; | |
1da177e4 | 1963 | |
c19b3b05 | 1964 | ASSERT(VFS_I(ip)->i_mode != 0); |
1da177e4 | 1965 | |
1da177e4 LT |
1966 | /* |
1967 | * Get the agi buffer first. It ensures lock ordering | |
1968 | * on the list. | |
1969 | */ | |
5e1be0fb | 1970 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 1971 | if (error) |
1da177e4 | 1972 | return error; |
1da177e4 | 1973 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1974 | |
1da177e4 LT |
1975 | /* |
1976 | * Get the index into the agi hash table for the | |
1977 | * list this inode will go on. | |
1978 | */ | |
1979 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1980 | ASSERT(agino != 0); | |
1981 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1982 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1983 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1984 | |
69ef921b | 1985 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
1986 | /* |
1987 | * There is already another inode in the bucket we need | |
1988 | * to add ourselves to. Add us at the front of the list. | |
1989 | * Here we put the head pointer into our next pointer, | |
1990 | * and then we fall through to point the head at us. | |
1991 | */ | |
475ee413 CH |
1992 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1993 | 0, 0); | |
c319b58b VA |
1994 | if (error) |
1995 | return error; | |
1996 | ||
69ef921b | 1997 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 1998 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 1999 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2000 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2001 | |
2002 | /* need to recalc the inode CRC if appropriate */ | |
2003 | xfs_dinode_calc_crc(mp, dip); | |
2004 | ||
1da177e4 LT |
2005 | xfs_trans_inode_buf(tp, ibp); |
2006 | xfs_trans_log_buf(tp, ibp, offset, | |
2007 | (offset + sizeof(xfs_agino_t) - 1)); | |
2008 | xfs_inobp_check(mp, ibp); | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * Point the bucket head pointer at the inode being inserted. | |
2013 | */ | |
2014 | ASSERT(agino != 0); | |
16259e7d | 2015 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
2016 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2017 | (sizeof(xfs_agino_t) * bucket_index); | |
2018 | xfs_trans_log_buf(tp, agibp, offset, | |
2019 | (offset + sizeof(xfs_agino_t) - 1)); | |
2020 | return 0; | |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * Pull the on-disk inode from the AGI unlinked list. | |
2025 | */ | |
2026 | STATIC int | |
2027 | xfs_iunlink_remove( | |
2028 | xfs_trans_t *tp, | |
2029 | xfs_inode_t *ip) | |
2030 | { | |
2031 | xfs_ino_t next_ino; | |
2032 | xfs_mount_t *mp; | |
2033 | xfs_agi_t *agi; | |
2034 | xfs_dinode_t *dip; | |
2035 | xfs_buf_t *agibp; | |
2036 | xfs_buf_t *ibp; | |
2037 | xfs_agnumber_t agno; | |
1da177e4 LT |
2038 | xfs_agino_t agino; |
2039 | xfs_agino_t next_agino; | |
2040 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 2041 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 2042 | short bucket_index; |
6fdf8ccc | 2043 | int offset, last_offset = 0; |
1da177e4 | 2044 | int error; |
1da177e4 | 2045 | |
1da177e4 | 2046 | mp = tp->t_mountp; |
1da177e4 | 2047 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
2048 | |
2049 | /* | |
2050 | * Get the agi buffer first. It ensures lock ordering | |
2051 | * on the list. | |
2052 | */ | |
5e1be0fb CH |
2053 | error = xfs_read_agi(mp, tp, agno, &agibp); |
2054 | if (error) | |
1da177e4 | 2055 | return error; |
5e1be0fb | 2056 | |
1da177e4 | 2057 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 2058 | |
1da177e4 LT |
2059 | /* |
2060 | * Get the index into the agi hash table for the | |
2061 | * list this inode will go on. | |
2062 | */ | |
2063 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
2064 | ASSERT(agino != 0); | |
2065 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 2066 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
2067 | ASSERT(agi->agi_unlinked[bucket_index]); |
2068 | ||
16259e7d | 2069 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 2070 | /* |
475ee413 CH |
2071 | * We're at the head of the list. Get the inode's on-disk |
2072 | * buffer to see if there is anyone after us on the list. | |
2073 | * Only modify our next pointer if it is not already NULLAGINO. | |
2074 | * This saves us the overhead of dealing with the buffer when | |
2075 | * there is no need to change it. | |
1da177e4 | 2076 | */ |
475ee413 CH |
2077 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2078 | 0, 0); | |
1da177e4 | 2079 | if (error) { |
475ee413 | 2080 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2081 | __func__, error); |
1da177e4 LT |
2082 | return error; |
2083 | } | |
347d1c01 | 2084 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2085 | ASSERT(next_agino != 0); |
2086 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2087 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2088 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2089 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2090 | |
2091 | /* need to recalc the inode CRC if appropriate */ | |
2092 | xfs_dinode_calc_crc(mp, dip); | |
2093 | ||
1da177e4 LT |
2094 | xfs_trans_inode_buf(tp, ibp); |
2095 | xfs_trans_log_buf(tp, ibp, offset, | |
2096 | (offset + sizeof(xfs_agino_t) - 1)); | |
2097 | xfs_inobp_check(mp, ibp); | |
2098 | } else { | |
2099 | xfs_trans_brelse(tp, ibp); | |
2100 | } | |
2101 | /* | |
2102 | * Point the bucket head pointer at the next inode. | |
2103 | */ | |
2104 | ASSERT(next_agino != 0); | |
2105 | ASSERT(next_agino != agino); | |
16259e7d | 2106 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2107 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2108 | (sizeof(xfs_agino_t) * bucket_index); | |
2109 | xfs_trans_log_buf(tp, agibp, offset, | |
2110 | (offset + sizeof(xfs_agino_t) - 1)); | |
2111 | } else { | |
2112 | /* | |
2113 | * We need to search the list for the inode being freed. | |
2114 | */ | |
16259e7d | 2115 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2116 | last_ibp = NULL; |
2117 | while (next_agino != agino) { | |
129dbc9a CH |
2118 | struct xfs_imap imap; |
2119 | ||
2120 | if (last_ibp) | |
1da177e4 | 2121 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
2122 | |
2123 | imap.im_blkno = 0; | |
1da177e4 | 2124 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
2125 | |
2126 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
2127 | if (error) { | |
2128 | xfs_warn(mp, | |
2129 | "%s: xfs_imap returned error %d.", | |
2130 | __func__, error); | |
2131 | return error; | |
2132 | } | |
2133 | ||
2134 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
2135 | &last_ibp, 0, 0); | |
1da177e4 | 2136 | if (error) { |
0b932ccc | 2137 | xfs_warn(mp, |
129dbc9a | 2138 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2139 | __func__, error); |
1da177e4 LT |
2140 | return error; |
2141 | } | |
129dbc9a CH |
2142 | |
2143 | last_offset = imap.im_boffset; | |
347d1c01 | 2144 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
2145 | ASSERT(next_agino != NULLAGINO); |
2146 | ASSERT(next_agino != 0); | |
2147 | } | |
475ee413 | 2148 | |
1da177e4 | 2149 | /* |
475ee413 CH |
2150 | * Now last_ibp points to the buffer previous to us on the |
2151 | * unlinked list. Pull us from the list. | |
1da177e4 | 2152 | */ |
475ee413 CH |
2153 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2154 | 0, 0); | |
1da177e4 | 2155 | if (error) { |
475ee413 | 2156 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 2157 | __func__, error); |
1da177e4 LT |
2158 | return error; |
2159 | } | |
347d1c01 | 2160 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2161 | ASSERT(next_agino != 0); |
2162 | ASSERT(next_agino != agino); | |
2163 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2164 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2165 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2166 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2167 | |
2168 | /* need to recalc the inode CRC if appropriate */ | |
2169 | xfs_dinode_calc_crc(mp, dip); | |
2170 | ||
1da177e4 LT |
2171 | xfs_trans_inode_buf(tp, ibp); |
2172 | xfs_trans_log_buf(tp, ibp, offset, | |
2173 | (offset + sizeof(xfs_agino_t) - 1)); | |
2174 | xfs_inobp_check(mp, ibp); | |
2175 | } else { | |
2176 | xfs_trans_brelse(tp, ibp); | |
2177 | } | |
2178 | /* | |
2179 | * Point the previous inode on the list to the next inode. | |
2180 | */ | |
347d1c01 | 2181 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
2182 | ASSERT(next_agino != 0); |
2183 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
0a32c26e DC |
2184 | |
2185 | /* need to recalc the inode CRC if appropriate */ | |
2186 | xfs_dinode_calc_crc(mp, last_dip); | |
2187 | ||
1da177e4 LT |
2188 | xfs_trans_inode_buf(tp, last_ibp); |
2189 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2190 | (offset + sizeof(xfs_agino_t) - 1)); | |
2191 | xfs_inobp_check(mp, last_ibp); | |
2192 | } | |
2193 | return 0; | |
2194 | } | |
2195 | ||
5b3eed75 | 2196 | /* |
0b8182db | 2197 | * A big issue when freeing the inode cluster is that we _cannot_ skip any |
5b3eed75 DC |
2198 | * inodes that are in memory - they all must be marked stale and attached to |
2199 | * the cluster buffer. | |
2200 | */ | |
2a30f36d | 2201 | STATIC int |
1da177e4 | 2202 | xfs_ifree_cluster( |
09b56604 BF |
2203 | xfs_inode_t *free_ip, |
2204 | xfs_trans_t *tp, | |
2205 | struct xfs_icluster *xic) | |
1da177e4 LT |
2206 | { |
2207 | xfs_mount_t *mp = free_ip->i_mount; | |
2208 | int blks_per_cluster; | |
982e939e | 2209 | int inodes_per_cluster; |
1da177e4 | 2210 | int nbufs; |
5b257b4a | 2211 | int i, j; |
3cdaa189 | 2212 | int ioffset; |
1da177e4 LT |
2213 | xfs_daddr_t blkno; |
2214 | xfs_buf_t *bp; | |
5b257b4a | 2215 | xfs_inode_t *ip; |
1da177e4 | 2216 | xfs_inode_log_item_t *iip; |
643c8c05 | 2217 | struct xfs_log_item *lip; |
5017e97d | 2218 | struct xfs_perag *pag; |
09b56604 | 2219 | xfs_ino_t inum; |
1da177e4 | 2220 | |
09b56604 | 2221 | inum = xic->first_ino; |
5017e97d | 2222 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
982e939e JL |
2223 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
2224 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
2225 | nbufs = mp->m_ialloc_blks / blks_per_cluster; | |
1da177e4 | 2226 | |
982e939e | 2227 | for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { |
09b56604 BF |
2228 | /* |
2229 | * The allocation bitmap tells us which inodes of the chunk were | |
2230 | * physically allocated. Skip the cluster if an inode falls into | |
2231 | * a sparse region. | |
2232 | */ | |
3cdaa189 BF |
2233 | ioffset = inum - xic->first_ino; |
2234 | if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { | |
2235 | ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); | |
09b56604 BF |
2236 | continue; |
2237 | } | |
2238 | ||
1da177e4 LT |
2239 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), |
2240 | XFS_INO_TO_AGBNO(mp, inum)); | |
2241 | ||
5b257b4a DC |
2242 | /* |
2243 | * We obtain and lock the backing buffer first in the process | |
2244 | * here, as we have to ensure that any dirty inode that we | |
2245 | * can't get the flush lock on is attached to the buffer. | |
2246 | * If we scan the in-memory inodes first, then buffer IO can | |
2247 | * complete before we get a lock on it, and hence we may fail | |
2248 | * to mark all the active inodes on the buffer stale. | |
2249 | */ | |
2250 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
2251 | mp->m_bsize * blks_per_cluster, |
2252 | XBF_UNMAPPED); | |
5b257b4a | 2253 | |
2a30f36d | 2254 | if (!bp) |
2451337d | 2255 | return -ENOMEM; |
b0f539de DC |
2256 | |
2257 | /* | |
2258 | * This buffer may not have been correctly initialised as we | |
2259 | * didn't read it from disk. That's not important because we are | |
2260 | * only using to mark the buffer as stale in the log, and to | |
2261 | * attach stale cached inodes on it. That means it will never be | |
2262 | * dispatched for IO. If it is, we want to know about it, and we | |
2263 | * want it to fail. We can acheive this by adding a write | |
2264 | * verifier to the buffer. | |
2265 | */ | |
1813dd64 | 2266 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 2267 | |
5b257b4a DC |
2268 | /* |
2269 | * Walk the inodes already attached to the buffer and mark them | |
2270 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
2271 | * in-memory inode walk can't lock them. By marking them all |
2272 | * stale first, we will not attempt to lock them in the loop | |
2273 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 2274 | */ |
643c8c05 | 2275 | list_for_each_entry(lip, &bp->b_li_list, li_bio_list) { |
5b257b4a DC |
2276 | if (lip->li_type == XFS_LI_INODE) { |
2277 | iip = (xfs_inode_log_item_t *)lip; | |
2278 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 2279 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
2280 | xfs_trans_ail_copy_lsn(mp->m_ail, |
2281 | &iip->ili_flush_lsn, | |
2282 | &iip->ili_item.li_lsn); | |
2283 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a | 2284 | } |
5b257b4a | 2285 | } |
1da177e4 | 2286 | |
5b3eed75 | 2287 | |
1da177e4 | 2288 | /* |
5b257b4a DC |
2289 | * For each inode in memory attempt to add it to the inode |
2290 | * buffer and set it up for being staled on buffer IO | |
2291 | * completion. This is safe as we've locked out tail pushing | |
2292 | * and flushing by locking the buffer. | |
1da177e4 | 2293 | * |
5b257b4a DC |
2294 | * We have already marked every inode that was part of a |
2295 | * transaction stale above, which means there is no point in | |
2296 | * even trying to lock them. | |
1da177e4 | 2297 | */ |
982e939e | 2298 | for (i = 0; i < inodes_per_cluster; i++) { |
5b3eed75 | 2299 | retry: |
1a3e8f3d | 2300 | rcu_read_lock(); |
da353b0d DC |
2301 | ip = radix_tree_lookup(&pag->pag_ici_root, |
2302 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 2303 | |
1a3e8f3d DC |
2304 | /* Inode not in memory, nothing to do */ |
2305 | if (!ip) { | |
2306 | rcu_read_unlock(); | |
1da177e4 LT |
2307 | continue; |
2308 | } | |
2309 | ||
1a3e8f3d DC |
2310 | /* |
2311 | * because this is an RCU protected lookup, we could | |
2312 | * find a recently freed or even reallocated inode | |
2313 | * during the lookup. We need to check under the | |
2314 | * i_flags_lock for a valid inode here. Skip it if it | |
2315 | * is not valid, the wrong inode or stale. | |
2316 | */ | |
2317 | spin_lock(&ip->i_flags_lock); | |
2318 | if (ip->i_ino != inum + i || | |
2319 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
2320 | spin_unlock(&ip->i_flags_lock); | |
2321 | rcu_read_unlock(); | |
2322 | continue; | |
2323 | } | |
2324 | spin_unlock(&ip->i_flags_lock); | |
2325 | ||
5b3eed75 DC |
2326 | /* |
2327 | * Don't try to lock/unlock the current inode, but we | |
2328 | * _cannot_ skip the other inodes that we did not find | |
2329 | * in the list attached to the buffer and are not | |
2330 | * already marked stale. If we can't lock it, back off | |
2331 | * and retry. | |
2332 | */ | |
f2e9ad21 OS |
2333 | if (ip != free_ip) { |
2334 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2335 | rcu_read_unlock(); | |
2336 | delay(1); | |
2337 | goto retry; | |
2338 | } | |
2339 | ||
2340 | /* | |
2341 | * Check the inode number again in case we're | |
2342 | * racing with freeing in xfs_reclaim_inode(). | |
2343 | * See the comments in that function for more | |
2344 | * information as to why the initial check is | |
2345 | * not sufficient. | |
2346 | */ | |
2347 | if (ip->i_ino != inum + i) { | |
2348 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
962cc1ad | 2349 | rcu_read_unlock(); |
f2e9ad21 OS |
2350 | continue; |
2351 | } | |
1da177e4 | 2352 | } |
1a3e8f3d | 2353 | rcu_read_unlock(); |
1da177e4 | 2354 | |
5b3eed75 | 2355 | xfs_iflock(ip); |
5b257b4a | 2356 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 2357 | |
5b3eed75 DC |
2358 | /* |
2359 | * we don't need to attach clean inodes or those only | |
2360 | * with unlogged changes (which we throw away, anyway). | |
2361 | */ | |
1da177e4 | 2362 | iip = ip->i_itemp; |
5b3eed75 | 2363 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 2364 | ASSERT(ip != free_ip); |
1da177e4 LT |
2365 | xfs_ifunlock(ip); |
2366 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2367 | continue; | |
2368 | } | |
2369 | ||
f5d8d5c4 CH |
2370 | iip->ili_last_fields = iip->ili_fields; |
2371 | iip->ili_fields = 0; | |
fc0561ce | 2372 | iip->ili_fsync_fields = 0; |
1da177e4 | 2373 | iip->ili_logged = 1; |
7b2e2a31 DC |
2374 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2375 | &iip->ili_item.li_lsn); | |
1da177e4 | 2376 | |
ca30b2a7 CH |
2377 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
2378 | &iip->ili_item); | |
5b257b4a DC |
2379 | |
2380 | if (ip != free_ip) | |
1da177e4 | 2381 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
2382 | } |
2383 | ||
5b3eed75 | 2384 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
2385 | xfs_trans_binval(tp, bp); |
2386 | } | |
2387 | ||
5017e97d | 2388 | xfs_perag_put(pag); |
2a30f36d | 2389 | return 0; |
1da177e4 LT |
2390 | } |
2391 | ||
98c4f78d DW |
2392 | /* |
2393 | * Free any local-format buffers sitting around before we reset to | |
2394 | * extents format. | |
2395 | */ | |
2396 | static inline void | |
2397 | xfs_ifree_local_data( | |
2398 | struct xfs_inode *ip, | |
2399 | int whichfork) | |
2400 | { | |
2401 | struct xfs_ifork *ifp; | |
2402 | ||
2403 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL) | |
2404 | return; | |
2405 | ||
2406 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2407 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); | |
2408 | } | |
2409 | ||
1da177e4 LT |
2410 | /* |
2411 | * This is called to return an inode to the inode free list. | |
2412 | * The inode should already be truncated to 0 length and have | |
2413 | * no pages associated with it. This routine also assumes that | |
2414 | * the inode is already a part of the transaction. | |
2415 | * | |
2416 | * The on-disk copy of the inode will have been added to the list | |
2417 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2418 | * that list atomically with respect to freeing it here. | |
2419 | */ | |
2420 | int | |
2421 | xfs_ifree( | |
2422 | xfs_trans_t *tp, | |
2423 | xfs_inode_t *ip, | |
2c3234d1 | 2424 | struct xfs_defer_ops *dfops) |
1da177e4 LT |
2425 | { |
2426 | int error; | |
09b56604 | 2427 | struct xfs_icluster xic = { 0 }; |
1da177e4 | 2428 | |
579aa9ca | 2429 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
54d7b5c1 | 2430 | ASSERT(VFS_I(ip)->i_nlink == 0); |
1da177e4 LT |
2431 | ASSERT(ip->i_d.di_nextents == 0); |
2432 | ASSERT(ip->i_d.di_anextents == 0); | |
c19b3b05 | 2433 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); |
1da177e4 LT |
2434 | ASSERT(ip->i_d.di_nblocks == 0); |
2435 | ||
2436 | /* | |
2437 | * Pull the on-disk inode from the AGI unlinked list. | |
2438 | */ | |
2439 | error = xfs_iunlink_remove(tp, ip); | |
1baaed8f | 2440 | if (error) |
1da177e4 | 2441 | return error; |
1da177e4 | 2442 | |
2c3234d1 | 2443 | error = xfs_difree(tp, ip->i_ino, dfops, &xic); |
1baaed8f | 2444 | if (error) |
1da177e4 | 2445 | return error; |
1baaed8f | 2446 | |
98c4f78d DW |
2447 | xfs_ifree_local_data(ip, XFS_DATA_FORK); |
2448 | xfs_ifree_local_data(ip, XFS_ATTR_FORK); | |
2449 | ||
c19b3b05 | 2450 | VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ |
1da177e4 | 2451 | ip->i_d.di_flags = 0; |
beaae8cd | 2452 | ip->i_d.di_flags2 = 0; |
1da177e4 LT |
2453 | ip->i_d.di_dmevmask = 0; |
2454 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
2455 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
2456 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2457 | /* | |
2458 | * Bump the generation count so no one will be confused | |
2459 | * by reincarnations of this inode. | |
2460 | */ | |
9e9a2674 | 2461 | VFS_I(ip)->i_generation++; |
1da177e4 LT |
2462 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
2463 | ||
09b56604 BF |
2464 | if (xic.deleted) |
2465 | error = xfs_ifree_cluster(ip, tp, &xic); | |
1da177e4 | 2466 | |
2a30f36d | 2467 | return error; |
1da177e4 LT |
2468 | } |
2469 | ||
1da177e4 | 2470 | /* |
60ec6783 CH |
2471 | * This is called to unpin an inode. The caller must have the inode locked |
2472 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2473 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2474 | */ |
60ec6783 | 2475 | static void |
f392e631 | 2476 | xfs_iunpin( |
60ec6783 | 2477 | struct xfs_inode *ip) |
1da177e4 | 2478 | { |
579aa9ca | 2479 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2480 | |
4aaf15d1 DC |
2481 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2482 | ||
a3f74ffb | 2483 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 2484 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 2485 | |
a3f74ffb | 2486 | } |
1da177e4 | 2487 | |
f392e631 CH |
2488 | static void |
2489 | __xfs_iunpin_wait( | |
2490 | struct xfs_inode *ip) | |
2491 | { | |
2492 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2493 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2494 | ||
2495 | xfs_iunpin(ip); | |
2496 | ||
2497 | do { | |
21417136 | 2498 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
f392e631 CH |
2499 | if (xfs_ipincount(ip)) |
2500 | io_schedule(); | |
2501 | } while (xfs_ipincount(ip)); | |
21417136 | 2502 | finish_wait(wq, &wait.wq_entry); |
f392e631 CH |
2503 | } |
2504 | ||
777df5af | 2505 | void |
a3f74ffb | 2506 | xfs_iunpin_wait( |
60ec6783 | 2507 | struct xfs_inode *ip) |
a3f74ffb | 2508 | { |
f392e631 CH |
2509 | if (xfs_ipincount(ip)) |
2510 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2511 | } |
2512 | ||
27320369 DC |
2513 | /* |
2514 | * Removing an inode from the namespace involves removing the directory entry | |
2515 | * and dropping the link count on the inode. Removing the directory entry can | |
2516 | * result in locking an AGF (directory blocks were freed) and removing a link | |
2517 | * count can result in placing the inode on an unlinked list which results in | |
2518 | * locking an AGI. | |
2519 | * | |
2520 | * The big problem here is that we have an ordering constraint on AGF and AGI | |
2521 | * locking - inode allocation locks the AGI, then can allocate a new extent for | |
2522 | * new inodes, locking the AGF after the AGI. Similarly, freeing the inode | |
2523 | * removes the inode from the unlinked list, requiring that we lock the AGI | |
2524 | * first, and then freeing the inode can result in an inode chunk being freed | |
2525 | * and hence freeing disk space requiring that we lock an AGF. | |
2526 | * | |
2527 | * Hence the ordering that is imposed by other parts of the code is AGI before | |
2528 | * AGF. This means we cannot remove the directory entry before we drop the inode | |
2529 | * reference count and put it on the unlinked list as this results in a lock | |
2530 | * order of AGF then AGI, and this can deadlock against inode allocation and | |
2531 | * freeing. Therefore we must drop the link counts before we remove the | |
2532 | * directory entry. | |
2533 | * | |
2534 | * This is still safe from a transactional point of view - it is not until we | |
310a75a3 | 2535 | * get to xfs_defer_finish() that we have the possibility of multiple |
27320369 DC |
2536 | * transactions in this operation. Hence as long as we remove the directory |
2537 | * entry and drop the link count in the first transaction of the remove | |
2538 | * operation, there are no transactional constraints on the ordering here. | |
2539 | */ | |
c24b5dfa DC |
2540 | int |
2541 | xfs_remove( | |
2542 | xfs_inode_t *dp, | |
2543 | struct xfs_name *name, | |
2544 | xfs_inode_t *ip) | |
2545 | { | |
2546 | xfs_mount_t *mp = dp->i_mount; | |
2547 | xfs_trans_t *tp = NULL; | |
c19b3b05 | 2548 | int is_dir = S_ISDIR(VFS_I(ip)->i_mode); |
c24b5dfa | 2549 | int error = 0; |
2c3234d1 | 2550 | struct xfs_defer_ops dfops; |
c24b5dfa | 2551 | xfs_fsblock_t first_block; |
c24b5dfa | 2552 | uint resblks; |
c24b5dfa DC |
2553 | |
2554 | trace_xfs_remove(dp, name); | |
2555 | ||
2556 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 2557 | return -EIO; |
c24b5dfa DC |
2558 | |
2559 | error = xfs_qm_dqattach(dp, 0); | |
2560 | if (error) | |
2561 | goto std_return; | |
2562 | ||
2563 | error = xfs_qm_dqattach(ip, 0); | |
2564 | if (error) | |
2565 | goto std_return; | |
2566 | ||
c24b5dfa DC |
2567 | /* |
2568 | * We try to get the real space reservation first, | |
2569 | * allowing for directory btree deletion(s) implying | |
2570 | * possible bmap insert(s). If we can't get the space | |
2571 | * reservation then we use 0 instead, and avoid the bmap | |
2572 | * btree insert(s) in the directory code by, if the bmap | |
2573 | * insert tries to happen, instead trimming the LAST | |
2574 | * block from the directory. | |
2575 | */ | |
2576 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
253f4911 | 2577 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); |
2451337d | 2578 | if (error == -ENOSPC) { |
c24b5dfa | 2579 | resblks = 0; |
253f4911 CH |
2580 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, |
2581 | &tp); | |
c24b5dfa DC |
2582 | } |
2583 | if (error) { | |
2451337d | 2584 | ASSERT(error != -ENOSPC); |
253f4911 | 2585 | goto std_return; |
c24b5dfa DC |
2586 | } |
2587 | ||
2588 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); | |
2589 | ||
65523218 | 2590 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
2591 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
2592 | ||
2593 | /* | |
2594 | * If we're removing a directory perform some additional validation. | |
2595 | */ | |
2596 | if (is_dir) { | |
54d7b5c1 DC |
2597 | ASSERT(VFS_I(ip)->i_nlink >= 2); |
2598 | if (VFS_I(ip)->i_nlink != 2) { | |
2451337d | 2599 | error = -ENOTEMPTY; |
c24b5dfa DC |
2600 | goto out_trans_cancel; |
2601 | } | |
2602 | if (!xfs_dir_isempty(ip)) { | |
2451337d | 2603 | error = -ENOTEMPTY; |
c24b5dfa DC |
2604 | goto out_trans_cancel; |
2605 | } | |
c24b5dfa | 2606 | |
27320369 | 2607 | /* Drop the link from ip's "..". */ |
c24b5dfa DC |
2608 | error = xfs_droplink(tp, dp); |
2609 | if (error) | |
27320369 | 2610 | goto out_trans_cancel; |
c24b5dfa | 2611 | |
27320369 | 2612 | /* Drop the "." link from ip to self. */ |
c24b5dfa DC |
2613 | error = xfs_droplink(tp, ip); |
2614 | if (error) | |
27320369 | 2615 | goto out_trans_cancel; |
c24b5dfa DC |
2616 | } else { |
2617 | /* | |
2618 | * When removing a non-directory we need to log the parent | |
2619 | * inode here. For a directory this is done implicitly | |
2620 | * by the xfs_droplink call for the ".." entry. | |
2621 | */ | |
2622 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2623 | } | |
27320369 | 2624 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
c24b5dfa | 2625 | |
27320369 | 2626 | /* Drop the link from dp to ip. */ |
c24b5dfa DC |
2627 | error = xfs_droplink(tp, ip); |
2628 | if (error) | |
27320369 | 2629 | goto out_trans_cancel; |
c24b5dfa | 2630 | |
2c3234d1 | 2631 | xfs_defer_init(&dfops, &first_block); |
27320369 | 2632 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
2c3234d1 | 2633 | &first_block, &dfops, resblks); |
27320369 | 2634 | if (error) { |
2451337d | 2635 | ASSERT(error != -ENOENT); |
27320369 DC |
2636 | goto out_bmap_cancel; |
2637 | } | |
2638 | ||
c24b5dfa DC |
2639 | /* |
2640 | * If this is a synchronous mount, make sure that the | |
2641 | * remove transaction goes to disk before returning to | |
2642 | * the user. | |
2643 | */ | |
2644 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2645 | xfs_trans_set_sync(tp); | |
2646 | ||
8ad7c629 | 2647 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa DC |
2648 | if (error) |
2649 | goto out_bmap_cancel; | |
2650 | ||
70393313 | 2651 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
2652 | if (error) |
2653 | goto std_return; | |
2654 | ||
2cd2ef6a | 2655 | if (is_dir && xfs_inode_is_filestream(ip)) |
c24b5dfa DC |
2656 | xfs_filestream_deassociate(ip); |
2657 | ||
2658 | return 0; | |
2659 | ||
2660 | out_bmap_cancel: | |
2c3234d1 | 2661 | xfs_defer_cancel(&dfops); |
c24b5dfa | 2662 | out_trans_cancel: |
4906e215 | 2663 | xfs_trans_cancel(tp); |
c24b5dfa DC |
2664 | std_return: |
2665 | return error; | |
2666 | } | |
2667 | ||
f6bba201 DC |
2668 | /* |
2669 | * Enter all inodes for a rename transaction into a sorted array. | |
2670 | */ | |
95afcf5c | 2671 | #define __XFS_SORT_INODES 5 |
f6bba201 DC |
2672 | STATIC void |
2673 | xfs_sort_for_rename( | |
95afcf5c DC |
2674 | struct xfs_inode *dp1, /* in: old (source) directory inode */ |
2675 | struct xfs_inode *dp2, /* in: new (target) directory inode */ | |
2676 | struct xfs_inode *ip1, /* in: inode of old entry */ | |
2677 | struct xfs_inode *ip2, /* in: inode of new entry */ | |
2678 | struct xfs_inode *wip, /* in: whiteout inode */ | |
2679 | struct xfs_inode **i_tab,/* out: sorted array of inodes */ | |
2680 | int *num_inodes) /* in/out: inodes in array */ | |
f6bba201 | 2681 | { |
f6bba201 DC |
2682 | int i, j; |
2683 | ||
95afcf5c DC |
2684 | ASSERT(*num_inodes == __XFS_SORT_INODES); |
2685 | memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); | |
2686 | ||
f6bba201 DC |
2687 | /* |
2688 | * i_tab contains a list of pointers to inodes. We initialize | |
2689 | * the table here & we'll sort it. We will then use it to | |
2690 | * order the acquisition of the inode locks. | |
2691 | * | |
2692 | * Note that the table may contain duplicates. e.g., dp1 == dp2. | |
2693 | */ | |
95afcf5c DC |
2694 | i = 0; |
2695 | i_tab[i++] = dp1; | |
2696 | i_tab[i++] = dp2; | |
2697 | i_tab[i++] = ip1; | |
2698 | if (ip2) | |
2699 | i_tab[i++] = ip2; | |
2700 | if (wip) | |
2701 | i_tab[i++] = wip; | |
2702 | *num_inodes = i; | |
f6bba201 DC |
2703 | |
2704 | /* | |
2705 | * Sort the elements via bubble sort. (Remember, there are at | |
95afcf5c | 2706 | * most 5 elements to sort, so this is adequate.) |
f6bba201 DC |
2707 | */ |
2708 | for (i = 0; i < *num_inodes; i++) { | |
2709 | for (j = 1; j < *num_inodes; j++) { | |
2710 | if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { | |
95afcf5c | 2711 | struct xfs_inode *temp = i_tab[j]; |
f6bba201 DC |
2712 | i_tab[j] = i_tab[j-1]; |
2713 | i_tab[j-1] = temp; | |
2714 | } | |
2715 | } | |
2716 | } | |
2717 | } | |
2718 | ||
310606b0 DC |
2719 | static int |
2720 | xfs_finish_rename( | |
2721 | struct xfs_trans *tp, | |
2c3234d1 | 2722 | struct xfs_defer_ops *dfops) |
310606b0 | 2723 | { |
310606b0 DC |
2724 | int error; |
2725 | ||
2726 | /* | |
2727 | * If this is a synchronous mount, make sure that the rename transaction | |
2728 | * goes to disk before returning to the user. | |
2729 | */ | |
2730 | if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2731 | xfs_trans_set_sync(tp); | |
2732 | ||
8ad7c629 | 2733 | error = xfs_defer_finish(&tp, dfops); |
310606b0 | 2734 | if (error) { |
2c3234d1 | 2735 | xfs_defer_cancel(dfops); |
4906e215 | 2736 | xfs_trans_cancel(tp); |
310606b0 DC |
2737 | return error; |
2738 | } | |
2739 | ||
70393313 | 2740 | return xfs_trans_commit(tp); |
310606b0 DC |
2741 | } |
2742 | ||
d31a1825 CM |
2743 | /* |
2744 | * xfs_cross_rename() | |
2745 | * | |
2746 | * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall | |
2747 | */ | |
2748 | STATIC int | |
2749 | xfs_cross_rename( | |
2750 | struct xfs_trans *tp, | |
2751 | struct xfs_inode *dp1, | |
2752 | struct xfs_name *name1, | |
2753 | struct xfs_inode *ip1, | |
2754 | struct xfs_inode *dp2, | |
2755 | struct xfs_name *name2, | |
2756 | struct xfs_inode *ip2, | |
2c3234d1 | 2757 | struct xfs_defer_ops *dfops, |
d31a1825 CM |
2758 | xfs_fsblock_t *first_block, |
2759 | int spaceres) | |
2760 | { | |
2761 | int error = 0; | |
2762 | int ip1_flags = 0; | |
2763 | int ip2_flags = 0; | |
2764 | int dp2_flags = 0; | |
2765 | ||
2766 | /* Swap inode number for dirent in first parent */ | |
2767 | error = xfs_dir_replace(tp, dp1, name1, | |
2768 | ip2->i_ino, | |
2c3234d1 | 2769 | first_block, dfops, spaceres); |
d31a1825 | 2770 | if (error) |
eeacd321 | 2771 | goto out_trans_abort; |
d31a1825 CM |
2772 | |
2773 | /* Swap inode number for dirent in second parent */ | |
2774 | error = xfs_dir_replace(tp, dp2, name2, | |
2775 | ip1->i_ino, | |
2c3234d1 | 2776 | first_block, dfops, spaceres); |
d31a1825 | 2777 | if (error) |
eeacd321 | 2778 | goto out_trans_abort; |
d31a1825 CM |
2779 | |
2780 | /* | |
2781 | * If we're renaming one or more directories across different parents, | |
2782 | * update the respective ".." entries (and link counts) to match the new | |
2783 | * parents. | |
2784 | */ | |
2785 | if (dp1 != dp2) { | |
2786 | dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2787 | ||
c19b3b05 | 2788 | if (S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 CM |
2789 | error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, |
2790 | dp1->i_ino, first_block, | |
2c3234d1 | 2791 | dfops, spaceres); |
d31a1825 | 2792 | if (error) |
eeacd321 | 2793 | goto out_trans_abort; |
d31a1825 CM |
2794 | |
2795 | /* transfer ip2 ".." reference to dp1 */ | |
c19b3b05 | 2796 | if (!S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 CM |
2797 | error = xfs_droplink(tp, dp2); |
2798 | if (error) | |
eeacd321 | 2799 | goto out_trans_abort; |
d31a1825 CM |
2800 | error = xfs_bumplink(tp, dp1); |
2801 | if (error) | |
eeacd321 | 2802 | goto out_trans_abort; |
d31a1825 CM |
2803 | } |
2804 | ||
2805 | /* | |
2806 | * Although ip1 isn't changed here, userspace needs | |
2807 | * to be warned about the change, so that applications | |
2808 | * relying on it (like backup ones), will properly | |
2809 | * notify the change | |
2810 | */ | |
2811 | ip1_flags |= XFS_ICHGTIME_CHG; | |
2812 | ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2813 | } | |
2814 | ||
c19b3b05 | 2815 | if (S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 CM |
2816 | error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, |
2817 | dp2->i_ino, first_block, | |
2c3234d1 | 2818 | dfops, spaceres); |
d31a1825 | 2819 | if (error) |
eeacd321 | 2820 | goto out_trans_abort; |
d31a1825 CM |
2821 | |
2822 | /* transfer ip1 ".." reference to dp2 */ | |
c19b3b05 | 2823 | if (!S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 CM |
2824 | error = xfs_droplink(tp, dp1); |
2825 | if (error) | |
eeacd321 | 2826 | goto out_trans_abort; |
d31a1825 CM |
2827 | error = xfs_bumplink(tp, dp2); |
2828 | if (error) | |
eeacd321 | 2829 | goto out_trans_abort; |
d31a1825 CM |
2830 | } |
2831 | ||
2832 | /* | |
2833 | * Although ip2 isn't changed here, userspace needs | |
2834 | * to be warned about the change, so that applications | |
2835 | * relying on it (like backup ones), will properly | |
2836 | * notify the change | |
2837 | */ | |
2838 | ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2839 | ip2_flags |= XFS_ICHGTIME_CHG; | |
2840 | } | |
2841 | } | |
2842 | ||
2843 | if (ip1_flags) { | |
2844 | xfs_trans_ichgtime(tp, ip1, ip1_flags); | |
2845 | xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); | |
2846 | } | |
2847 | if (ip2_flags) { | |
2848 | xfs_trans_ichgtime(tp, ip2, ip2_flags); | |
2849 | xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); | |
2850 | } | |
2851 | if (dp2_flags) { | |
2852 | xfs_trans_ichgtime(tp, dp2, dp2_flags); | |
2853 | xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); | |
2854 | } | |
2855 | xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2856 | xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); | |
2c3234d1 | 2857 | return xfs_finish_rename(tp, dfops); |
eeacd321 DC |
2858 | |
2859 | out_trans_abort: | |
2c3234d1 | 2860 | xfs_defer_cancel(dfops); |
4906e215 | 2861 | xfs_trans_cancel(tp); |
d31a1825 CM |
2862 | return error; |
2863 | } | |
2864 | ||
7dcf5c3e DC |
2865 | /* |
2866 | * xfs_rename_alloc_whiteout() | |
2867 | * | |
2868 | * Return a referenced, unlinked, unlocked inode that that can be used as a | |
2869 | * whiteout in a rename transaction. We use a tmpfile inode here so that if we | |
2870 | * crash between allocating the inode and linking it into the rename transaction | |
2871 | * recovery will free the inode and we won't leak it. | |
2872 | */ | |
2873 | static int | |
2874 | xfs_rename_alloc_whiteout( | |
2875 | struct xfs_inode *dp, | |
2876 | struct xfs_inode **wip) | |
2877 | { | |
2878 | struct xfs_inode *tmpfile; | |
2879 | int error; | |
2880 | ||
2881 | error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); | |
2882 | if (error) | |
2883 | return error; | |
2884 | ||
22419ac9 BF |
2885 | /* |
2886 | * Prepare the tmpfile inode as if it were created through the VFS. | |
2887 | * Otherwise, the link increment paths will complain about nlink 0->1. | |
2888 | * Drop the link count as done by d_tmpfile(), complete the inode setup | |
2889 | * and flag it as linkable. | |
2890 | */ | |
2891 | drop_nlink(VFS_I(tmpfile)); | |
2b3d1d41 | 2892 | xfs_setup_iops(tmpfile); |
7dcf5c3e DC |
2893 | xfs_finish_inode_setup(tmpfile); |
2894 | VFS_I(tmpfile)->i_state |= I_LINKABLE; | |
2895 | ||
2896 | *wip = tmpfile; | |
2897 | return 0; | |
2898 | } | |
2899 | ||
f6bba201 DC |
2900 | /* |
2901 | * xfs_rename | |
2902 | */ | |
2903 | int | |
2904 | xfs_rename( | |
7dcf5c3e DC |
2905 | struct xfs_inode *src_dp, |
2906 | struct xfs_name *src_name, | |
2907 | struct xfs_inode *src_ip, | |
2908 | struct xfs_inode *target_dp, | |
2909 | struct xfs_name *target_name, | |
2910 | struct xfs_inode *target_ip, | |
2911 | unsigned int flags) | |
f6bba201 | 2912 | { |
7dcf5c3e DC |
2913 | struct xfs_mount *mp = src_dp->i_mount; |
2914 | struct xfs_trans *tp; | |
2c3234d1 | 2915 | struct xfs_defer_ops dfops; |
7dcf5c3e DC |
2916 | xfs_fsblock_t first_block; |
2917 | struct xfs_inode *wip = NULL; /* whiteout inode */ | |
2918 | struct xfs_inode *inodes[__XFS_SORT_INODES]; | |
2919 | int num_inodes = __XFS_SORT_INODES; | |
2b93681f | 2920 | bool new_parent = (src_dp != target_dp); |
c19b3b05 | 2921 | bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); |
7dcf5c3e DC |
2922 | int spaceres; |
2923 | int error; | |
f6bba201 DC |
2924 | |
2925 | trace_xfs_rename(src_dp, target_dp, src_name, target_name); | |
2926 | ||
eeacd321 DC |
2927 | if ((flags & RENAME_EXCHANGE) && !target_ip) |
2928 | return -EINVAL; | |
2929 | ||
7dcf5c3e DC |
2930 | /* |
2931 | * If we are doing a whiteout operation, allocate the whiteout inode | |
2932 | * we will be placing at the target and ensure the type is set | |
2933 | * appropriately. | |
2934 | */ | |
2935 | if (flags & RENAME_WHITEOUT) { | |
2936 | ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); | |
2937 | error = xfs_rename_alloc_whiteout(target_dp, &wip); | |
2938 | if (error) | |
2939 | return error; | |
2940 | ||
2941 | /* setup target dirent info as whiteout */ | |
2942 | src_name->type = XFS_DIR3_FT_CHRDEV; | |
2943 | } | |
f6bba201 | 2944 | |
7dcf5c3e | 2945 | xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, |
f6bba201 DC |
2946 | inodes, &num_inodes); |
2947 | ||
f6bba201 | 2948 | spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); |
253f4911 | 2949 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); |
2451337d | 2950 | if (error == -ENOSPC) { |
f6bba201 | 2951 | spaceres = 0; |
253f4911 CH |
2952 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, |
2953 | &tp); | |
f6bba201 | 2954 | } |
445883e8 | 2955 | if (error) |
253f4911 | 2956 | goto out_release_wip; |
f6bba201 DC |
2957 | |
2958 | /* | |
2959 | * Attach the dquots to the inodes | |
2960 | */ | |
2961 | error = xfs_qm_vop_rename_dqattach(inodes); | |
445883e8 DC |
2962 | if (error) |
2963 | goto out_trans_cancel; | |
f6bba201 DC |
2964 | |
2965 | /* | |
2966 | * Lock all the participating inodes. Depending upon whether | |
2967 | * the target_name exists in the target directory, and | |
2968 | * whether the target directory is the same as the source | |
2969 | * directory, we can lock from 2 to 4 inodes. | |
2970 | */ | |
2971 | xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); | |
2972 | ||
2973 | /* | |
2974 | * Join all the inodes to the transaction. From this point on, | |
2975 | * we can rely on either trans_commit or trans_cancel to unlock | |
2976 | * them. | |
2977 | */ | |
65523218 | 2978 | xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); |
f6bba201 | 2979 | if (new_parent) |
65523218 | 2980 | xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); |
f6bba201 DC |
2981 | xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); |
2982 | if (target_ip) | |
2983 | xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); | |
7dcf5c3e DC |
2984 | if (wip) |
2985 | xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); | |
f6bba201 DC |
2986 | |
2987 | /* | |
2988 | * If we are using project inheritance, we only allow renames | |
2989 | * into our tree when the project IDs are the same; else the | |
2990 | * tree quota mechanism would be circumvented. | |
2991 | */ | |
2992 | if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2993 | (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { | |
2451337d | 2994 | error = -EXDEV; |
445883e8 | 2995 | goto out_trans_cancel; |
f6bba201 DC |
2996 | } |
2997 | ||
2c3234d1 | 2998 | xfs_defer_init(&dfops, &first_block); |
445883e8 | 2999 | |
eeacd321 DC |
3000 | /* RENAME_EXCHANGE is unique from here on. */ |
3001 | if (flags & RENAME_EXCHANGE) | |
3002 | return xfs_cross_rename(tp, src_dp, src_name, src_ip, | |
3003 | target_dp, target_name, target_ip, | |
2c3234d1 | 3004 | &dfops, &first_block, spaceres); |
d31a1825 | 3005 | |
f6bba201 DC |
3006 | /* |
3007 | * Set up the target. | |
3008 | */ | |
3009 | if (target_ip == NULL) { | |
3010 | /* | |
3011 | * If there's no space reservation, check the entry will | |
3012 | * fit before actually inserting it. | |
3013 | */ | |
94f3cad5 ES |
3014 | if (!spaceres) { |
3015 | error = xfs_dir_canenter(tp, target_dp, target_name); | |
3016 | if (error) | |
445883e8 | 3017 | goto out_trans_cancel; |
94f3cad5 | 3018 | } |
f6bba201 DC |
3019 | /* |
3020 | * If target does not exist and the rename crosses | |
3021 | * directories, adjust the target directory link count | |
3022 | * to account for the ".." reference from the new entry. | |
3023 | */ | |
3024 | error = xfs_dir_createname(tp, target_dp, target_name, | |
3025 | src_ip->i_ino, &first_block, | |
2c3234d1 | 3026 | &dfops, spaceres); |
f6bba201 | 3027 | if (error) |
4906e215 | 3028 | goto out_bmap_cancel; |
f6bba201 DC |
3029 | |
3030 | xfs_trans_ichgtime(tp, target_dp, | |
3031 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3032 | ||
3033 | if (new_parent && src_is_directory) { | |
3034 | error = xfs_bumplink(tp, target_dp); | |
3035 | if (error) | |
4906e215 | 3036 | goto out_bmap_cancel; |
f6bba201 DC |
3037 | } |
3038 | } else { /* target_ip != NULL */ | |
3039 | /* | |
3040 | * If target exists and it's a directory, check that both | |
3041 | * target and source are directories and that target can be | |
3042 | * destroyed, or that neither is a directory. | |
3043 | */ | |
c19b3b05 | 3044 | if (S_ISDIR(VFS_I(target_ip)->i_mode)) { |
f6bba201 DC |
3045 | /* |
3046 | * Make sure target dir is empty. | |
3047 | */ | |
3048 | if (!(xfs_dir_isempty(target_ip)) || | |
54d7b5c1 | 3049 | (VFS_I(target_ip)->i_nlink > 2)) { |
2451337d | 3050 | error = -EEXIST; |
445883e8 | 3051 | goto out_trans_cancel; |
f6bba201 DC |
3052 | } |
3053 | } | |
3054 | ||
3055 | /* | |
3056 | * Link the source inode under the target name. | |
3057 | * If the source inode is a directory and we are moving | |
3058 | * it across directories, its ".." entry will be | |
3059 | * inconsistent until we replace that down below. | |
3060 | * | |
3061 | * In case there is already an entry with the same | |
3062 | * name at the destination directory, remove it first. | |
3063 | */ | |
3064 | error = xfs_dir_replace(tp, target_dp, target_name, | |
3065 | src_ip->i_ino, | |
2c3234d1 | 3066 | &first_block, &dfops, spaceres); |
f6bba201 | 3067 | if (error) |
4906e215 | 3068 | goto out_bmap_cancel; |
f6bba201 DC |
3069 | |
3070 | xfs_trans_ichgtime(tp, target_dp, | |
3071 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3072 | ||
3073 | /* | |
3074 | * Decrement the link count on the target since the target | |
3075 | * dir no longer points to it. | |
3076 | */ | |
3077 | error = xfs_droplink(tp, target_ip); | |
3078 | if (error) | |
4906e215 | 3079 | goto out_bmap_cancel; |
f6bba201 DC |
3080 | |
3081 | if (src_is_directory) { | |
3082 | /* | |
3083 | * Drop the link from the old "." entry. | |
3084 | */ | |
3085 | error = xfs_droplink(tp, target_ip); | |
3086 | if (error) | |
4906e215 | 3087 | goto out_bmap_cancel; |
f6bba201 DC |
3088 | } |
3089 | } /* target_ip != NULL */ | |
3090 | ||
3091 | /* | |
3092 | * Remove the source. | |
3093 | */ | |
3094 | if (new_parent && src_is_directory) { | |
3095 | /* | |
3096 | * Rewrite the ".." entry to point to the new | |
3097 | * directory. | |
3098 | */ | |
3099 | error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, | |
3100 | target_dp->i_ino, | |
2c3234d1 | 3101 | &first_block, &dfops, spaceres); |
2451337d | 3102 | ASSERT(error != -EEXIST); |
f6bba201 | 3103 | if (error) |
4906e215 | 3104 | goto out_bmap_cancel; |
f6bba201 DC |
3105 | } |
3106 | ||
3107 | /* | |
3108 | * We always want to hit the ctime on the source inode. | |
3109 | * | |
3110 | * This isn't strictly required by the standards since the source | |
3111 | * inode isn't really being changed, but old unix file systems did | |
3112 | * it and some incremental backup programs won't work without it. | |
3113 | */ | |
3114 | xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); | |
3115 | xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); | |
3116 | ||
3117 | /* | |
3118 | * Adjust the link count on src_dp. This is necessary when | |
3119 | * renaming a directory, either within one parent when | |
3120 | * the target existed, or across two parent directories. | |
3121 | */ | |
3122 | if (src_is_directory && (new_parent || target_ip != NULL)) { | |
3123 | ||
3124 | /* | |
3125 | * Decrement link count on src_directory since the | |
3126 | * entry that's moved no longer points to it. | |
3127 | */ | |
3128 | error = xfs_droplink(tp, src_dp); | |
3129 | if (error) | |
4906e215 | 3130 | goto out_bmap_cancel; |
f6bba201 DC |
3131 | } |
3132 | ||
7dcf5c3e DC |
3133 | /* |
3134 | * For whiteouts, we only need to update the source dirent with the | |
3135 | * inode number of the whiteout inode rather than removing it | |
3136 | * altogether. | |
3137 | */ | |
3138 | if (wip) { | |
3139 | error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, | |
2c3234d1 | 3140 | &first_block, &dfops, spaceres); |
7dcf5c3e DC |
3141 | } else |
3142 | error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, | |
2c3234d1 | 3143 | &first_block, &dfops, spaceres); |
f6bba201 | 3144 | if (error) |
4906e215 | 3145 | goto out_bmap_cancel; |
f6bba201 DC |
3146 | |
3147 | /* | |
7dcf5c3e DC |
3148 | * For whiteouts, we need to bump the link count on the whiteout inode. |
3149 | * This means that failures all the way up to this point leave the inode | |
3150 | * on the unlinked list and so cleanup is a simple matter of dropping | |
3151 | * the remaining reference to it. If we fail here after bumping the link | |
3152 | * count, we're shutting down the filesystem so we'll never see the | |
3153 | * intermediate state on disk. | |
f6bba201 | 3154 | */ |
7dcf5c3e | 3155 | if (wip) { |
54d7b5c1 | 3156 | ASSERT(VFS_I(wip)->i_nlink == 0); |
7dcf5c3e DC |
3157 | error = xfs_bumplink(tp, wip); |
3158 | if (error) | |
4906e215 | 3159 | goto out_bmap_cancel; |
7dcf5c3e DC |
3160 | error = xfs_iunlink_remove(tp, wip); |
3161 | if (error) | |
4906e215 | 3162 | goto out_bmap_cancel; |
7dcf5c3e | 3163 | xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); |
f6bba201 | 3164 | |
7dcf5c3e DC |
3165 | /* |
3166 | * Now we have a real link, clear the "I'm a tmpfile" state | |
3167 | * flag from the inode so it doesn't accidentally get misused in | |
3168 | * future. | |
3169 | */ | |
3170 | VFS_I(wip)->i_state &= ~I_LINKABLE; | |
f6bba201 DC |
3171 | } |
3172 | ||
f6bba201 DC |
3173 | xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3174 | xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); | |
3175 | if (new_parent) | |
3176 | xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); | |
f6bba201 | 3177 | |
2c3234d1 | 3178 | error = xfs_finish_rename(tp, &dfops); |
7dcf5c3e DC |
3179 | if (wip) |
3180 | IRELE(wip); | |
3181 | return error; | |
f6bba201 | 3182 | |
445883e8 | 3183 | out_bmap_cancel: |
2c3234d1 | 3184 | xfs_defer_cancel(&dfops); |
445883e8 | 3185 | out_trans_cancel: |
4906e215 | 3186 | xfs_trans_cancel(tp); |
253f4911 | 3187 | out_release_wip: |
7dcf5c3e DC |
3188 | if (wip) |
3189 | IRELE(wip); | |
f6bba201 DC |
3190 | return error; |
3191 | } | |
3192 | ||
5c4d97d0 DC |
3193 | STATIC int |
3194 | xfs_iflush_cluster( | |
19429363 DC |
3195 | struct xfs_inode *ip, |
3196 | struct xfs_buf *bp) | |
1da177e4 | 3197 | { |
19429363 | 3198 | struct xfs_mount *mp = ip->i_mount; |
5c4d97d0 DC |
3199 | struct xfs_perag *pag; |
3200 | unsigned long first_index, mask; | |
3201 | unsigned long inodes_per_cluster; | |
19429363 DC |
3202 | int cilist_size; |
3203 | struct xfs_inode **cilist; | |
3204 | struct xfs_inode *cip; | |
5c4d97d0 DC |
3205 | int nr_found; |
3206 | int clcount = 0; | |
3207 | int bufwasdelwri; | |
1da177e4 | 3208 | int i; |
1da177e4 | 3209 | |
5c4d97d0 | 3210 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1da177e4 | 3211 | |
0f49efd8 | 3212 | inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; |
19429363 DC |
3213 | cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); |
3214 | cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); | |
3215 | if (!cilist) | |
5c4d97d0 | 3216 | goto out_put; |
1da177e4 | 3217 | |
0f49efd8 | 3218 | mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); |
5c4d97d0 DC |
3219 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; |
3220 | rcu_read_lock(); | |
3221 | /* really need a gang lookup range call here */ | |
19429363 | 3222 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, |
5c4d97d0 DC |
3223 | first_index, inodes_per_cluster); |
3224 | if (nr_found == 0) | |
3225 | goto out_free; | |
3226 | ||
3227 | for (i = 0; i < nr_found; i++) { | |
19429363 DC |
3228 | cip = cilist[i]; |
3229 | if (cip == ip) | |
bad55843 | 3230 | continue; |
1a3e8f3d DC |
3231 | |
3232 | /* | |
3233 | * because this is an RCU protected lookup, we could find a | |
3234 | * recently freed or even reallocated inode during the lookup. | |
3235 | * We need to check under the i_flags_lock for a valid inode | |
3236 | * here. Skip it if it is not valid or the wrong inode. | |
3237 | */ | |
19429363 DC |
3238 | spin_lock(&cip->i_flags_lock); |
3239 | if (!cip->i_ino || | |
3240 | __xfs_iflags_test(cip, XFS_ISTALE)) { | |
3241 | spin_unlock(&cip->i_flags_lock); | |
1a3e8f3d DC |
3242 | continue; |
3243 | } | |
5a90e53e DC |
3244 | |
3245 | /* | |
3246 | * Once we fall off the end of the cluster, no point checking | |
3247 | * any more inodes in the list because they will also all be | |
3248 | * outside the cluster. | |
3249 | */ | |
19429363 DC |
3250 | if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { |
3251 | spin_unlock(&cip->i_flags_lock); | |
5a90e53e DC |
3252 | break; |
3253 | } | |
19429363 | 3254 | spin_unlock(&cip->i_flags_lock); |
1a3e8f3d | 3255 | |
bad55843 DC |
3256 | /* |
3257 | * Do an un-protected check to see if the inode is dirty and | |
3258 | * is a candidate for flushing. These checks will be repeated | |
3259 | * later after the appropriate locks are acquired. | |
3260 | */ | |
19429363 | 3261 | if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) |
bad55843 | 3262 | continue; |
bad55843 DC |
3263 | |
3264 | /* | |
3265 | * Try to get locks. If any are unavailable or it is pinned, | |
3266 | * then this inode cannot be flushed and is skipped. | |
3267 | */ | |
3268 | ||
19429363 | 3269 | if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) |
bad55843 | 3270 | continue; |
19429363 DC |
3271 | if (!xfs_iflock_nowait(cip)) { |
3272 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3273 | continue; |
3274 | } | |
19429363 DC |
3275 | if (xfs_ipincount(cip)) { |
3276 | xfs_ifunlock(cip); | |
3277 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3278 | continue; |
3279 | } | |
3280 | ||
8a17d7dd DC |
3281 | |
3282 | /* | |
3283 | * Check the inode number again, just to be certain we are not | |
3284 | * racing with freeing in xfs_reclaim_inode(). See the comments | |
3285 | * in that function for more information as to why the initial | |
3286 | * check is not sufficient. | |
3287 | */ | |
19429363 DC |
3288 | if (!cip->i_ino) { |
3289 | xfs_ifunlock(cip); | |
3290 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3291 | continue; |
3292 | } | |
3293 | ||
3294 | /* | |
3295 | * arriving here means that this inode can be flushed. First | |
3296 | * re-check that it's dirty before flushing. | |
3297 | */ | |
19429363 | 3298 | if (!xfs_inode_clean(cip)) { |
33540408 | 3299 | int error; |
19429363 | 3300 | error = xfs_iflush_int(cip, bp); |
bad55843 | 3301 | if (error) { |
19429363 | 3302 | xfs_iunlock(cip, XFS_ILOCK_SHARED); |
bad55843 DC |
3303 | goto cluster_corrupt_out; |
3304 | } | |
3305 | clcount++; | |
3306 | } else { | |
19429363 | 3307 | xfs_ifunlock(cip); |
bad55843 | 3308 | } |
19429363 | 3309 | xfs_iunlock(cip, XFS_ILOCK_SHARED); |
bad55843 DC |
3310 | } |
3311 | ||
3312 | if (clcount) { | |
ff6d6af2 BD |
3313 | XFS_STATS_INC(mp, xs_icluster_flushcnt); |
3314 | XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); | |
bad55843 DC |
3315 | } |
3316 | ||
3317 | out_free: | |
1a3e8f3d | 3318 | rcu_read_unlock(); |
19429363 | 3319 | kmem_free(cilist); |
44b56e0a DC |
3320 | out_put: |
3321 | xfs_perag_put(pag); | |
bad55843 DC |
3322 | return 0; |
3323 | ||
3324 | ||
3325 | cluster_corrupt_out: | |
3326 | /* | |
3327 | * Corruption detected in the clustering loop. Invalidate the | |
3328 | * inode buffer and shut down the filesystem. | |
3329 | */ | |
1a3e8f3d | 3330 | rcu_read_unlock(); |
bad55843 | 3331 | /* |
43ff2122 | 3332 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
3333 | * brelse can handle it with no problems. If not, shut down the |
3334 | * filesystem before releasing the buffer. | |
3335 | */ | |
43ff2122 | 3336 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
3337 | if (bufwasdelwri) |
3338 | xfs_buf_relse(bp); | |
3339 | ||
3340 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
3341 | ||
3342 | if (!bufwasdelwri) { | |
3343 | /* | |
3344 | * Just like incore_relse: if we have b_iodone functions, | |
3345 | * mark the buffer as an error and call them. Otherwise | |
3346 | * mark it as stale and brelse. | |
3347 | */ | |
cb669ca5 | 3348 | if (bp->b_iodone) { |
b0388bf1 | 3349 | bp->b_flags &= ~XBF_DONE; |
c867cb61 | 3350 | xfs_buf_stale(bp); |
2451337d | 3351 | xfs_buf_ioerror(bp, -EIO); |
e8aaba9a | 3352 | xfs_buf_ioend(bp); |
bad55843 | 3353 | } else { |
c867cb61 | 3354 | xfs_buf_stale(bp); |
bad55843 DC |
3355 | xfs_buf_relse(bp); |
3356 | } | |
3357 | } | |
3358 | ||
3359 | /* | |
3360 | * Unlocks the flush lock | |
3361 | */ | |
19429363 DC |
3362 | xfs_iflush_abort(cip, false); |
3363 | kmem_free(cilist); | |
44b56e0a | 3364 | xfs_perag_put(pag); |
2451337d | 3365 | return -EFSCORRUPTED; |
bad55843 DC |
3366 | } |
3367 | ||
1da177e4 | 3368 | /* |
4c46819a CH |
3369 | * Flush dirty inode metadata into the backing buffer. |
3370 | * | |
3371 | * The caller must have the inode lock and the inode flush lock held. The | |
3372 | * inode lock will still be held upon return to the caller, and the inode | |
3373 | * flush lock will be released after the inode has reached the disk. | |
3374 | * | |
3375 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
3376 | */ |
3377 | int | |
3378 | xfs_iflush( | |
4c46819a CH |
3379 | struct xfs_inode *ip, |
3380 | struct xfs_buf **bpp) | |
1da177e4 | 3381 | { |
4c46819a | 3382 | struct xfs_mount *mp = ip->i_mount; |
b1438f47 | 3383 | struct xfs_buf *bp = NULL; |
4c46819a | 3384 | struct xfs_dinode *dip; |
1da177e4 | 3385 | int error; |
1da177e4 | 3386 | |
ff6d6af2 | 3387 | XFS_STATS_INC(mp, xs_iflush_count); |
1da177e4 | 3388 | |
579aa9ca | 3389 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3390 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3391 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3392 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 3393 | |
4c46819a | 3394 | *bpp = NULL; |
1da177e4 | 3395 | |
1da177e4 LT |
3396 | xfs_iunpin_wait(ip); |
3397 | ||
4b6a4688 DC |
3398 | /* |
3399 | * For stale inodes we cannot rely on the backing buffer remaining | |
3400 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 3401 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
3402 | * inodes below. We have to check this after ensuring the inode is |
3403 | * unpinned so that it is safe to reclaim the stale inode after the | |
3404 | * flush call. | |
3405 | */ | |
3406 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
3407 | xfs_ifunlock(ip); | |
3408 | return 0; | |
3409 | } | |
3410 | ||
1da177e4 LT |
3411 | /* |
3412 | * This may have been unpinned because the filesystem is shutting | |
3413 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
3414 | * to disk, because the log record didn't make it to disk. |
3415 | * | |
3416 | * We also have to remove the log item from the AIL in this case, | |
3417 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
3418 | */ |
3419 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
2451337d | 3420 | error = -EIO; |
32ce90a4 | 3421 | goto abort_out; |
1da177e4 LT |
3422 | } |
3423 | ||
a3f74ffb | 3424 | /* |
b1438f47 DC |
3425 | * Get the buffer containing the on-disk inode. We are doing a try-lock |
3426 | * operation here, so we may get an EAGAIN error. In that case, we | |
3427 | * simply want to return with the inode still dirty. | |
3428 | * | |
3429 | * If we get any other error, we effectively have a corruption situation | |
3430 | * and we cannot flush the inode, so we treat it the same as failing | |
3431 | * xfs_iflush_int(). | |
a3f74ffb | 3432 | */ |
475ee413 CH |
3433 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
3434 | 0); | |
b1438f47 | 3435 | if (error == -EAGAIN) { |
a3f74ffb DC |
3436 | xfs_ifunlock(ip); |
3437 | return error; | |
3438 | } | |
b1438f47 DC |
3439 | if (error) |
3440 | goto corrupt_out; | |
a3f74ffb | 3441 | |
1da177e4 LT |
3442 | /* |
3443 | * First flush out the inode that xfs_iflush was called with. | |
3444 | */ | |
3445 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 3446 | if (error) |
1da177e4 | 3447 | goto corrupt_out; |
1da177e4 | 3448 | |
a3f74ffb DC |
3449 | /* |
3450 | * If the buffer is pinned then push on the log now so we won't | |
3451 | * get stuck waiting in the write for too long. | |
3452 | */ | |
811e64c7 | 3453 | if (xfs_buf_ispinned(bp)) |
a14a348b | 3454 | xfs_log_force(mp, 0); |
a3f74ffb | 3455 | |
1da177e4 LT |
3456 | /* |
3457 | * inode clustering: | |
3458 | * see if other inodes can be gathered into this write | |
3459 | */ | |
bad55843 DC |
3460 | error = xfs_iflush_cluster(ip, bp); |
3461 | if (error) | |
3462 | goto cluster_corrupt_out; | |
1da177e4 | 3463 | |
4c46819a CH |
3464 | *bpp = bp; |
3465 | return 0; | |
1da177e4 LT |
3466 | |
3467 | corrupt_out: | |
b1438f47 DC |
3468 | if (bp) |
3469 | xfs_buf_relse(bp); | |
7d04a335 | 3470 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 3471 | cluster_corrupt_out: |
2451337d | 3472 | error = -EFSCORRUPTED; |
32ce90a4 | 3473 | abort_out: |
1da177e4 LT |
3474 | /* |
3475 | * Unlocks the flush lock | |
3476 | */ | |
04913fdd | 3477 | xfs_iflush_abort(ip, false); |
32ce90a4 | 3478 | return error; |
1da177e4 LT |
3479 | } |
3480 | ||
9cfb9b47 DW |
3481 | /* |
3482 | * If there are inline format data / attr forks attached to this inode, | |
3483 | * make sure they're not corrupt. | |
3484 | */ | |
3485 | bool | |
3486 | xfs_inode_verify_forks( | |
3487 | struct xfs_inode *ip) | |
3488 | { | |
3489 | xfs_failaddr_t fa; | |
3490 | ||
3491 | fa = xfs_ifork_verify_data(ip, &xfs_default_ifork_ops); | |
3492 | if (fa) { | |
3493 | xfs_alert(ip->i_mount, | |
aff68a55 | 3494 | "%s: bad inode %llu inline data fork at %pS", |
9cfb9b47 DW |
3495 | __func__, ip->i_ino, fa); |
3496 | return false; | |
3497 | } | |
3498 | ||
3499 | fa = xfs_ifork_verify_attr(ip, &xfs_default_ifork_ops); | |
3500 | if (fa) { | |
3501 | xfs_alert(ip->i_mount, | |
aff68a55 | 3502 | "%s: bad inode %llu inline attr fork at %pS", |
9cfb9b47 DW |
3503 | __func__, ip->i_ino, fa); |
3504 | return false; | |
3505 | } | |
3506 | return true; | |
3507 | } | |
3508 | ||
1da177e4 LT |
3509 | STATIC int |
3510 | xfs_iflush_int( | |
93848a99 CH |
3511 | struct xfs_inode *ip, |
3512 | struct xfs_buf *bp) | |
1da177e4 | 3513 | { |
93848a99 CH |
3514 | struct xfs_inode_log_item *iip = ip->i_itemp; |
3515 | struct xfs_dinode *dip; | |
3516 | struct xfs_mount *mp = ip->i_mount; | |
1da177e4 | 3517 | |
579aa9ca | 3518 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3519 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3520 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3521 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
93848a99 | 3522 | ASSERT(iip != NULL && iip->ili_fields != 0); |
263997a6 | 3523 | ASSERT(ip->i_d.di_version > 1); |
1da177e4 | 3524 | |
1da177e4 | 3525 | /* set *dip = inode's place in the buffer */ |
88ee2df7 | 3526 | dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 3527 | |
69ef921b | 3528 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
9e24cfd0 | 3529 | mp, XFS_ERRTAG_IFLUSH_1)) { |
6a19d939 | 3530 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
c9690043 | 3531 | "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT, |
6a19d939 | 3532 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); |
1da177e4 LT |
3533 | goto corrupt_out; |
3534 | } | |
c19b3b05 | 3535 | if (S_ISREG(VFS_I(ip)->i_mode)) { |
1da177e4 LT |
3536 | if (XFS_TEST_ERROR( |
3537 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3538 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 3539 | mp, XFS_ERRTAG_IFLUSH_3)) { |
6a19d939 | 3540 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
c9690043 | 3541 | "%s: Bad regular inode %Lu, ptr "PTR_FMT, |
6a19d939 | 3542 | __func__, ip->i_ino, ip); |
1da177e4 LT |
3543 | goto corrupt_out; |
3544 | } | |
c19b3b05 | 3545 | } else if (S_ISDIR(VFS_I(ip)->i_mode)) { |
1da177e4 LT |
3546 | if (XFS_TEST_ERROR( |
3547 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3548 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3549 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
9e24cfd0 | 3550 | mp, XFS_ERRTAG_IFLUSH_4)) { |
6a19d939 | 3551 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
c9690043 | 3552 | "%s: Bad directory inode %Lu, ptr "PTR_FMT, |
6a19d939 | 3553 | __func__, ip->i_ino, ip); |
1da177e4 LT |
3554 | goto corrupt_out; |
3555 | } | |
3556 | } | |
3557 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
9e24cfd0 | 3558 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { |
6a19d939 DC |
3559 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3560 | "%s: detected corrupt incore inode %Lu, " | |
c9690043 | 3561 | "total extents = %d, nblocks = %Ld, ptr "PTR_FMT, |
6a19d939 | 3562 | __func__, ip->i_ino, |
1da177e4 | 3563 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 3564 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
3565 | goto corrupt_out; |
3566 | } | |
3567 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
9e24cfd0 | 3568 | mp, XFS_ERRTAG_IFLUSH_6)) { |
6a19d939 | 3569 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
c9690043 | 3570 | "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT, |
6a19d939 | 3571 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); |
1da177e4 LT |
3572 | goto corrupt_out; |
3573 | } | |
e60896d8 | 3574 | |
1da177e4 | 3575 | /* |
263997a6 | 3576 | * Inode item log recovery for v2 inodes are dependent on the |
e60896d8 DC |
3577 | * di_flushiter count for correct sequencing. We bump the flush |
3578 | * iteration count so we can detect flushes which postdate a log record | |
3579 | * during recovery. This is redundant as we now log every change and | |
3580 | * hence this can't happen but we need to still do it to ensure | |
3581 | * backwards compatibility with old kernels that predate logging all | |
3582 | * inode changes. | |
1da177e4 | 3583 | */ |
e60896d8 DC |
3584 | if (ip->i_d.di_version < 3) |
3585 | ip->i_d.di_flushiter++; | |
1da177e4 | 3586 | |
9cfb9b47 DW |
3587 | /* Check the inline fork data before we write out. */ |
3588 | if (!xfs_inode_verify_forks(ip)) | |
005c5db8 DW |
3589 | goto corrupt_out; |
3590 | ||
1da177e4 | 3591 | /* |
3987848c DC |
3592 | * Copy the dirty parts of the inode into the on-disk inode. We always |
3593 | * copy out the core of the inode, because if the inode is dirty at all | |
3594 | * the core must be. | |
1da177e4 | 3595 | */ |
93f958f9 | 3596 | xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); |
1da177e4 LT |
3597 | |
3598 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3599 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3600 | ip->i_d.di_flushiter = 0; | |
3601 | ||
005c5db8 DW |
3602 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); |
3603 | if (XFS_IFORK_Q(ip)) | |
3604 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); | |
1da177e4 LT |
3605 | xfs_inobp_check(mp, bp); |
3606 | ||
3607 | /* | |
f5d8d5c4 CH |
3608 | * We've recorded everything logged in the inode, so we'd like to clear |
3609 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
3610 | * However, we can't stop logging all this information until the data | |
3611 | * we've copied into the disk buffer is written to disk. If we did we | |
3612 | * might overwrite the copy of the inode in the log with all the data | |
3613 | * after re-logging only part of it, and in the face of a crash we | |
3614 | * wouldn't have all the data we need to recover. | |
1da177e4 | 3615 | * |
f5d8d5c4 CH |
3616 | * What we do is move the bits to the ili_last_fields field. When |
3617 | * logging the inode, these bits are moved back to the ili_fields field. | |
3618 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
3619 | * know that the information those bits represent is permanently on | |
3620 | * disk. As long as the flush completes before the inode is logged | |
3621 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 3622 | * |
f5d8d5c4 CH |
3623 | * We can play with the ili_fields bits here, because the inode lock |
3624 | * must be held exclusively in order to set bits there and the flush | |
3625 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
3626 | * done routine can tell whether or not to look in the AIL. Also, store | |
3627 | * the current LSN of the inode so that we can tell whether the item has | |
3628 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
3629 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
3630 | * atomically. | |
1da177e4 | 3631 | */ |
93848a99 CH |
3632 | iip->ili_last_fields = iip->ili_fields; |
3633 | iip->ili_fields = 0; | |
fc0561ce | 3634 | iip->ili_fsync_fields = 0; |
93848a99 | 3635 | iip->ili_logged = 1; |
1da177e4 | 3636 | |
93848a99 CH |
3637 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
3638 | &iip->ili_item.li_lsn); | |
1da177e4 | 3639 | |
93848a99 CH |
3640 | /* |
3641 | * Attach the function xfs_iflush_done to the inode's | |
3642 | * buffer. This will remove the inode from the AIL | |
3643 | * and unlock the inode's flush lock when the inode is | |
3644 | * completely written to disk. | |
3645 | */ | |
3646 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); | |
1da177e4 | 3647 | |
93848a99 CH |
3648 | /* generate the checksum. */ |
3649 | xfs_dinode_calc_crc(mp, dip); | |
1da177e4 | 3650 | |
643c8c05 | 3651 | ASSERT(!list_empty(&bp->b_li_list)); |
93848a99 | 3652 | ASSERT(bp->b_iodone != NULL); |
1da177e4 LT |
3653 | return 0; |
3654 | ||
3655 | corrupt_out: | |
2451337d | 3656 | return -EFSCORRUPTED; |
1da177e4 | 3657 | } |