Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
3e57ecf6 | 3 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 4 | * All Rights Reserved. |
1da177e4 | 5 | */ |
40ebd81d | 6 | #include <linux/log2.h> |
f0e28280 | 7 | #include <linux/iversion.h> |
40ebd81d | 8 | |
1da177e4 | 9 | #include "xfs.h" |
a844f451 | 10 | #include "xfs_fs.h" |
70a9883c | 11 | #include "xfs_shared.h" |
239880ef DC |
12 | #include "xfs_format.h" |
13 | #include "xfs_log_format.h" | |
14 | #include "xfs_trans_resv.h" | |
1da177e4 | 15 | #include "xfs_sb.h" |
1da177e4 | 16 | #include "xfs_mount.h" |
3ab78df2 | 17 | #include "xfs_defer.h" |
a4fbe6ab | 18 | #include "xfs_inode.h" |
57062787 | 19 | #include "xfs_da_format.h" |
c24b5dfa | 20 | #include "xfs_da_btree.h" |
c24b5dfa | 21 | #include "xfs_dir2.h" |
a844f451 | 22 | #include "xfs_attr_sf.h" |
c24b5dfa | 23 | #include "xfs_attr.h" |
239880ef DC |
24 | #include "xfs_trans_space.h" |
25 | #include "xfs_trans.h" | |
1da177e4 | 26 | #include "xfs_buf_item.h" |
a844f451 | 27 | #include "xfs_inode_item.h" |
a844f451 NS |
28 | #include "xfs_ialloc.h" |
29 | #include "xfs_bmap.h" | |
68988114 | 30 | #include "xfs_bmap_util.h" |
e9e899a2 | 31 | #include "xfs_errortag.h" |
1da177e4 | 32 | #include "xfs_error.h" |
1da177e4 | 33 | #include "xfs_quota.h" |
2a82b8be | 34 | #include "xfs_filestream.h" |
93848a99 | 35 | #include "xfs_cksum.h" |
0b1b213f | 36 | #include "xfs_trace.h" |
33479e05 | 37 | #include "xfs_icache.h" |
c24b5dfa | 38 | #include "xfs_symlink.h" |
239880ef DC |
39 | #include "xfs_trans_priv.h" |
40 | #include "xfs_log.h" | |
a4fbe6ab | 41 | #include "xfs_bmap_btree.h" |
aa8968f2 | 42 | #include "xfs_reflink.h" |
005c5db8 | 43 | #include "xfs_dir2_priv.h" |
1da177e4 | 44 | |
1da177e4 | 45 | kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
46 | |
47 | /* | |
8f04c47a | 48 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
49 | * freed from a file in a single transaction. |
50 | */ | |
51 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
52 | ||
54d7b5c1 DC |
53 | STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); |
54 | STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); | |
55 | STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); | |
ab297431 | 56 | |
2a0ec1d9 DC |
57 | /* |
58 | * helper function to extract extent size hint from inode | |
59 | */ | |
60 | xfs_extlen_t | |
61 | xfs_get_extsz_hint( | |
62 | struct xfs_inode *ip) | |
63 | { | |
64 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
65 | return ip->i_d.di_extsize; | |
66 | if (XFS_IS_REALTIME_INODE(ip)) | |
67 | return ip->i_mount->m_sb.sb_rextsize; | |
68 | return 0; | |
69 | } | |
70 | ||
f7ca3522 DW |
71 | /* |
72 | * Helper function to extract CoW extent size hint from inode. | |
73 | * Between the extent size hint and the CoW extent size hint, we | |
e153aa79 DW |
74 | * return the greater of the two. If the value is zero (automatic), |
75 | * use the default size. | |
f7ca3522 DW |
76 | */ |
77 | xfs_extlen_t | |
78 | xfs_get_cowextsz_hint( | |
79 | struct xfs_inode *ip) | |
80 | { | |
81 | xfs_extlen_t a, b; | |
82 | ||
83 | a = 0; | |
84 | if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) | |
85 | a = ip->i_d.di_cowextsize; | |
86 | b = xfs_get_extsz_hint(ip); | |
87 | ||
e153aa79 DW |
88 | a = max(a, b); |
89 | if (a == 0) | |
90 | return XFS_DEFAULT_COWEXTSZ_HINT; | |
91 | return a; | |
f7ca3522 DW |
92 | } |
93 | ||
fa96acad | 94 | /* |
efa70be1 CH |
95 | * These two are wrapper routines around the xfs_ilock() routine used to |
96 | * centralize some grungy code. They are used in places that wish to lock the | |
97 | * inode solely for reading the extents. The reason these places can't just | |
98 | * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to | |
99 | * bringing in of the extents from disk for a file in b-tree format. If the | |
100 | * inode is in b-tree format, then we need to lock the inode exclusively until | |
101 | * the extents are read in. Locking it exclusively all the time would limit | |
102 | * our parallelism unnecessarily, though. What we do instead is check to see | |
103 | * if the extents have been read in yet, and only lock the inode exclusively | |
104 | * if they have not. | |
fa96acad | 105 | * |
efa70be1 | 106 | * The functions return a value which should be given to the corresponding |
01f4f327 | 107 | * xfs_iunlock() call. |
fa96acad DC |
108 | */ |
109 | uint | |
309ecac8 CH |
110 | xfs_ilock_data_map_shared( |
111 | struct xfs_inode *ip) | |
fa96acad | 112 | { |
309ecac8 | 113 | uint lock_mode = XFS_ILOCK_SHARED; |
fa96acad | 114 | |
309ecac8 CH |
115 | if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && |
116 | (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) | |
fa96acad | 117 | lock_mode = XFS_ILOCK_EXCL; |
fa96acad | 118 | xfs_ilock(ip, lock_mode); |
fa96acad DC |
119 | return lock_mode; |
120 | } | |
121 | ||
efa70be1 CH |
122 | uint |
123 | xfs_ilock_attr_map_shared( | |
124 | struct xfs_inode *ip) | |
fa96acad | 125 | { |
efa70be1 CH |
126 | uint lock_mode = XFS_ILOCK_SHARED; |
127 | ||
128 | if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && | |
129 | (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) | |
130 | lock_mode = XFS_ILOCK_EXCL; | |
131 | xfs_ilock(ip, lock_mode); | |
132 | return lock_mode; | |
fa96acad DC |
133 | } |
134 | ||
135 | /* | |
65523218 CH |
136 | * In addition to i_rwsem in the VFS inode, the xfs inode contains 2 |
137 | * multi-reader locks: i_mmap_lock and the i_lock. This routine allows | |
138 | * various combinations of the locks to be obtained. | |
fa96acad | 139 | * |
653c60b6 DC |
140 | * The 3 locks should always be ordered so that the IO lock is obtained first, |
141 | * the mmap lock second and the ilock last in order to prevent deadlock. | |
fa96acad | 142 | * |
653c60b6 DC |
143 | * Basic locking order: |
144 | * | |
65523218 | 145 | * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock |
653c60b6 DC |
146 | * |
147 | * mmap_sem locking order: | |
148 | * | |
65523218 | 149 | * i_rwsem -> page lock -> mmap_sem |
653c60b6 DC |
150 | * mmap_sem -> i_mmap_lock -> page_lock |
151 | * | |
152 | * The difference in mmap_sem locking order mean that we cannot hold the | |
153 | * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can | |
154 | * fault in pages during copy in/out (for buffered IO) or require the mmap_sem | |
155 | * in get_user_pages() to map the user pages into the kernel address space for | |
65523218 | 156 | * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because |
653c60b6 DC |
157 | * page faults already hold the mmap_sem. |
158 | * | |
159 | * Hence to serialise fully against both syscall and mmap based IO, we need to | |
65523218 | 160 | * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both |
653c60b6 DC |
161 | * taken in places where we need to invalidate the page cache in a race |
162 | * free manner (e.g. truncate, hole punch and other extent manipulation | |
163 | * functions). | |
fa96acad DC |
164 | */ |
165 | void | |
166 | xfs_ilock( | |
167 | xfs_inode_t *ip, | |
168 | uint lock_flags) | |
169 | { | |
170 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
171 | ||
172 | /* | |
173 | * You can't set both SHARED and EXCL for the same lock, | |
174 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
175 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
176 | */ | |
177 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
178 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
179 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
180 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
181 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
182 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 183 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad | 184 | |
65523218 CH |
185 | if (lock_flags & XFS_IOLOCK_EXCL) { |
186 | down_write_nested(&VFS_I(ip)->i_rwsem, | |
187 | XFS_IOLOCK_DEP(lock_flags)); | |
188 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
189 | down_read_nested(&VFS_I(ip)->i_rwsem, | |
190 | XFS_IOLOCK_DEP(lock_flags)); | |
191 | } | |
fa96acad | 192 | |
653c60b6 DC |
193 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
194 | mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
195 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
196 | mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); | |
197 | ||
fa96acad DC |
198 | if (lock_flags & XFS_ILOCK_EXCL) |
199 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
200 | else if (lock_flags & XFS_ILOCK_SHARED) | |
201 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
202 | } | |
203 | ||
204 | /* | |
205 | * This is just like xfs_ilock(), except that the caller | |
206 | * is guaranteed not to sleep. It returns 1 if it gets | |
207 | * the requested locks and 0 otherwise. If the IO lock is | |
208 | * obtained but the inode lock cannot be, then the IO lock | |
209 | * is dropped before returning. | |
210 | * | |
211 | * ip -- the inode being locked | |
212 | * lock_flags -- this parameter indicates the inode's locks to be | |
213 | * to be locked. See the comment for xfs_ilock() for a list | |
214 | * of valid values. | |
215 | */ | |
216 | int | |
217 | xfs_ilock_nowait( | |
218 | xfs_inode_t *ip, | |
219 | uint lock_flags) | |
220 | { | |
221 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
222 | ||
223 | /* | |
224 | * You can't set both SHARED and EXCL for the same lock, | |
225 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
226 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
227 | */ | |
228 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
229 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
230 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
231 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
232 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
233 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 234 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
235 | |
236 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
65523218 | 237 | if (!down_write_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
238 | goto out; |
239 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
65523218 | 240 | if (!down_read_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
241 | goto out; |
242 | } | |
653c60b6 DC |
243 | |
244 | if (lock_flags & XFS_MMAPLOCK_EXCL) { | |
245 | if (!mrtryupdate(&ip->i_mmaplock)) | |
246 | goto out_undo_iolock; | |
247 | } else if (lock_flags & XFS_MMAPLOCK_SHARED) { | |
248 | if (!mrtryaccess(&ip->i_mmaplock)) | |
249 | goto out_undo_iolock; | |
250 | } | |
251 | ||
fa96acad DC |
252 | if (lock_flags & XFS_ILOCK_EXCL) { |
253 | if (!mrtryupdate(&ip->i_lock)) | |
653c60b6 | 254 | goto out_undo_mmaplock; |
fa96acad DC |
255 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
256 | if (!mrtryaccess(&ip->i_lock)) | |
653c60b6 | 257 | goto out_undo_mmaplock; |
fa96acad DC |
258 | } |
259 | return 1; | |
260 | ||
653c60b6 DC |
261 | out_undo_mmaplock: |
262 | if (lock_flags & XFS_MMAPLOCK_EXCL) | |
263 | mrunlock_excl(&ip->i_mmaplock); | |
264 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
265 | mrunlock_shared(&ip->i_mmaplock); | |
266 | out_undo_iolock: | |
fa96acad | 267 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 268 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 269 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 270 | up_read(&VFS_I(ip)->i_rwsem); |
653c60b6 | 271 | out: |
fa96acad DC |
272 | return 0; |
273 | } | |
274 | ||
275 | /* | |
276 | * xfs_iunlock() is used to drop the inode locks acquired with | |
277 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
278 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
279 | * that we know which locks to drop. | |
280 | * | |
281 | * ip -- the inode being unlocked | |
282 | * lock_flags -- this parameter indicates the inode's locks to be | |
283 | * to be unlocked. See the comment for xfs_ilock() for a list | |
284 | * of valid values for this parameter. | |
285 | * | |
286 | */ | |
287 | void | |
288 | xfs_iunlock( | |
289 | xfs_inode_t *ip, | |
290 | uint lock_flags) | |
291 | { | |
292 | /* | |
293 | * You can't set both SHARED and EXCL for the same lock, | |
294 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
295 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
296 | */ | |
297 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
298 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
653c60b6 DC |
299 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != |
300 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
fa96acad DC |
301 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
302 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
0952c818 | 303 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); |
fa96acad DC |
304 | ASSERT(lock_flags != 0); |
305 | ||
306 | if (lock_flags & XFS_IOLOCK_EXCL) | |
65523218 | 307 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 308 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 309 | up_read(&VFS_I(ip)->i_rwsem); |
fa96acad | 310 | |
653c60b6 DC |
311 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
312 | mrunlock_excl(&ip->i_mmaplock); | |
313 | else if (lock_flags & XFS_MMAPLOCK_SHARED) | |
314 | mrunlock_shared(&ip->i_mmaplock); | |
315 | ||
fa96acad DC |
316 | if (lock_flags & XFS_ILOCK_EXCL) |
317 | mrunlock_excl(&ip->i_lock); | |
318 | else if (lock_flags & XFS_ILOCK_SHARED) | |
319 | mrunlock_shared(&ip->i_lock); | |
320 | ||
321 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
322 | } | |
323 | ||
324 | /* | |
325 | * give up write locks. the i/o lock cannot be held nested | |
326 | * if it is being demoted. | |
327 | */ | |
328 | void | |
329 | xfs_ilock_demote( | |
330 | xfs_inode_t *ip, | |
331 | uint lock_flags) | |
332 | { | |
653c60b6 DC |
333 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); |
334 | ASSERT((lock_flags & | |
335 | ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
fa96acad DC |
336 | |
337 | if (lock_flags & XFS_ILOCK_EXCL) | |
338 | mrdemote(&ip->i_lock); | |
653c60b6 DC |
339 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
340 | mrdemote(&ip->i_mmaplock); | |
fa96acad | 341 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 342 | downgrade_write(&VFS_I(ip)->i_rwsem); |
fa96acad DC |
343 | |
344 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
345 | } | |
346 | ||
742ae1e3 | 347 | #if defined(DEBUG) || defined(XFS_WARN) |
fa96acad DC |
348 | int |
349 | xfs_isilocked( | |
350 | xfs_inode_t *ip, | |
351 | uint lock_flags) | |
352 | { | |
353 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
354 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
355 | return !!ip->i_lock.mr_writer; | |
356 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
357 | } | |
358 | ||
653c60b6 DC |
359 | if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { |
360 | if (!(lock_flags & XFS_MMAPLOCK_SHARED)) | |
361 | return !!ip->i_mmaplock.mr_writer; | |
362 | return rwsem_is_locked(&ip->i_mmaplock.mr_lock); | |
363 | } | |
364 | ||
fa96acad DC |
365 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
366 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
65523218 CH |
367 | return !debug_locks || |
368 | lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0); | |
369 | return rwsem_is_locked(&VFS_I(ip)->i_rwsem); | |
fa96acad DC |
370 | } |
371 | ||
372 | ASSERT(0); | |
373 | return 0; | |
374 | } | |
375 | #endif | |
376 | ||
b6a9947e DC |
377 | /* |
378 | * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when | |
379 | * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined | |
380 | * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build | |
381 | * errors and warnings. | |
382 | */ | |
383 | #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) | |
3403ccc0 DC |
384 | static bool |
385 | xfs_lockdep_subclass_ok( | |
386 | int subclass) | |
387 | { | |
388 | return subclass < MAX_LOCKDEP_SUBCLASSES; | |
389 | } | |
390 | #else | |
391 | #define xfs_lockdep_subclass_ok(subclass) (true) | |
392 | #endif | |
393 | ||
c24b5dfa | 394 | /* |
653c60b6 | 395 | * Bump the subclass so xfs_lock_inodes() acquires each lock with a different |
0952c818 DC |
396 | * value. This can be called for any type of inode lock combination, including |
397 | * parent locking. Care must be taken to ensure we don't overrun the subclass | |
398 | * storage fields in the class mask we build. | |
c24b5dfa DC |
399 | */ |
400 | static inline int | |
401 | xfs_lock_inumorder(int lock_mode, int subclass) | |
402 | { | |
0952c818 DC |
403 | int class = 0; |
404 | ||
405 | ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | | |
406 | XFS_ILOCK_RTSUM))); | |
3403ccc0 | 407 | ASSERT(xfs_lockdep_subclass_ok(subclass)); |
0952c818 | 408 | |
653c60b6 | 409 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { |
0952c818 | 410 | ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); |
0952c818 | 411 | class += subclass << XFS_IOLOCK_SHIFT; |
653c60b6 DC |
412 | } |
413 | ||
414 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { | |
0952c818 DC |
415 | ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); |
416 | class += subclass << XFS_MMAPLOCK_SHIFT; | |
653c60b6 DC |
417 | } |
418 | ||
0952c818 DC |
419 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { |
420 | ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); | |
421 | class += subclass << XFS_ILOCK_SHIFT; | |
422 | } | |
c24b5dfa | 423 | |
0952c818 | 424 | return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; |
c24b5dfa DC |
425 | } |
426 | ||
427 | /* | |
95afcf5c DC |
428 | * The following routine will lock n inodes in exclusive mode. We assume the |
429 | * caller calls us with the inodes in i_ino order. | |
c24b5dfa | 430 | * |
95afcf5c DC |
431 | * We need to detect deadlock where an inode that we lock is in the AIL and we |
432 | * start waiting for another inode that is locked by a thread in a long running | |
433 | * transaction (such as truncate). This can result in deadlock since the long | |
434 | * running trans might need to wait for the inode we just locked in order to | |
435 | * push the tail and free space in the log. | |
0952c818 DC |
436 | * |
437 | * xfs_lock_inodes() can only be used to lock one type of lock at a time - | |
438 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
439 | * lock more than one at a time, lockdep will report false positives saying we | |
440 | * have violated locking orders. | |
c24b5dfa | 441 | */ |
0d5a75e9 | 442 | static void |
c24b5dfa DC |
443 | xfs_lock_inodes( |
444 | xfs_inode_t **ips, | |
445 | int inodes, | |
446 | uint lock_mode) | |
447 | { | |
448 | int attempts = 0, i, j, try_lock; | |
449 | xfs_log_item_t *lp; | |
450 | ||
0952c818 DC |
451 | /* |
452 | * Currently supports between 2 and 5 inodes with exclusive locking. We | |
453 | * support an arbitrary depth of locking here, but absolute limits on | |
454 | * inodes depend on the the type of locking and the limits placed by | |
455 | * lockdep annotations in xfs_lock_inumorder. These are all checked by | |
456 | * the asserts. | |
457 | */ | |
95afcf5c | 458 | ASSERT(ips && inodes >= 2 && inodes <= 5); |
0952c818 DC |
459 | ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | |
460 | XFS_ILOCK_EXCL)); | |
461 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | | |
462 | XFS_ILOCK_SHARED))); | |
0952c818 DC |
463 | ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || |
464 | inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); | |
465 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || | |
466 | inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); | |
467 | ||
468 | if (lock_mode & XFS_IOLOCK_EXCL) { | |
469 | ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); | |
470 | } else if (lock_mode & XFS_MMAPLOCK_EXCL) | |
471 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); | |
c24b5dfa DC |
472 | |
473 | try_lock = 0; | |
474 | i = 0; | |
c24b5dfa DC |
475 | again: |
476 | for (; i < inodes; i++) { | |
477 | ASSERT(ips[i]); | |
478 | ||
95afcf5c | 479 | if (i && (ips[i] == ips[i - 1])) /* Already locked */ |
c24b5dfa DC |
480 | continue; |
481 | ||
482 | /* | |
95afcf5c DC |
483 | * If try_lock is not set yet, make sure all locked inodes are |
484 | * not in the AIL. If any are, set try_lock to be used later. | |
c24b5dfa | 485 | */ |
c24b5dfa DC |
486 | if (!try_lock) { |
487 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
488 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
22525c17 | 489 | if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) |
c24b5dfa | 490 | try_lock++; |
c24b5dfa DC |
491 | } |
492 | } | |
493 | ||
494 | /* | |
495 | * If any of the previous locks we have locked is in the AIL, | |
496 | * we must TRY to get the second and subsequent locks. If | |
497 | * we can't get any, we must release all we have | |
498 | * and try again. | |
499 | */ | |
95afcf5c DC |
500 | if (!try_lock) { |
501 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); | |
502 | continue; | |
503 | } | |
504 | ||
505 | /* try_lock means we have an inode locked that is in the AIL. */ | |
506 | ASSERT(i != 0); | |
507 | if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) | |
508 | continue; | |
c24b5dfa | 509 | |
95afcf5c DC |
510 | /* |
511 | * Unlock all previous guys and try again. xfs_iunlock will try | |
512 | * to push the tail if the inode is in the AIL. | |
513 | */ | |
514 | attempts++; | |
515 | for (j = i - 1; j >= 0; j--) { | |
c24b5dfa | 516 | /* |
95afcf5c DC |
517 | * Check to see if we've already unlocked this one. Not |
518 | * the first one going back, and the inode ptr is the | |
519 | * same. | |
c24b5dfa | 520 | */ |
95afcf5c DC |
521 | if (j != (i - 1) && ips[j] == ips[j + 1]) |
522 | continue; | |
c24b5dfa | 523 | |
95afcf5c DC |
524 | xfs_iunlock(ips[j], lock_mode); |
525 | } | |
c24b5dfa | 526 | |
95afcf5c DC |
527 | if ((attempts % 5) == 0) { |
528 | delay(1); /* Don't just spin the CPU */ | |
c24b5dfa | 529 | } |
95afcf5c DC |
530 | i = 0; |
531 | try_lock = 0; | |
532 | goto again; | |
c24b5dfa | 533 | } |
c24b5dfa DC |
534 | } |
535 | ||
536 | /* | |
653c60b6 | 537 | * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - |
7c2d238a DW |
538 | * the mmaplock or the ilock, but not more than one type at a time. If we lock |
539 | * more than one at a time, lockdep will report false positives saying we have | |
540 | * violated locking orders. The iolock must be double-locked separately since | |
541 | * we use i_rwsem for that. We now support taking one lock EXCL and the other | |
542 | * SHARED. | |
c24b5dfa DC |
543 | */ |
544 | void | |
545 | xfs_lock_two_inodes( | |
7c2d238a DW |
546 | struct xfs_inode *ip0, |
547 | uint ip0_mode, | |
548 | struct xfs_inode *ip1, | |
549 | uint ip1_mode) | |
c24b5dfa | 550 | { |
7c2d238a DW |
551 | struct xfs_inode *temp; |
552 | uint mode_temp; | |
c24b5dfa DC |
553 | int attempts = 0; |
554 | xfs_log_item_t *lp; | |
555 | ||
7c2d238a DW |
556 | ASSERT(hweight32(ip0_mode) == 1); |
557 | ASSERT(hweight32(ip1_mode) == 1); | |
558 | ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); | |
559 | ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); | |
560 | ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) || | |
561 | !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); | |
562 | ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) || | |
563 | !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); | |
564 | ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) || | |
565 | !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); | |
566 | ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) || | |
567 | !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); | |
653c60b6 | 568 | |
c24b5dfa DC |
569 | ASSERT(ip0->i_ino != ip1->i_ino); |
570 | ||
571 | if (ip0->i_ino > ip1->i_ino) { | |
572 | temp = ip0; | |
573 | ip0 = ip1; | |
574 | ip1 = temp; | |
7c2d238a DW |
575 | mode_temp = ip0_mode; |
576 | ip0_mode = ip1_mode; | |
577 | ip1_mode = mode_temp; | |
c24b5dfa DC |
578 | } |
579 | ||
580 | again: | |
7c2d238a | 581 | xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0)); |
c24b5dfa DC |
582 | |
583 | /* | |
584 | * If the first lock we have locked is in the AIL, we must TRY to get | |
585 | * the second lock. If we can't get it, we must release the first one | |
586 | * and try again. | |
587 | */ | |
588 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
22525c17 | 589 | if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) { |
7c2d238a DW |
590 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) { |
591 | xfs_iunlock(ip0, ip0_mode); | |
c24b5dfa DC |
592 | if ((++attempts % 5) == 0) |
593 | delay(1); /* Don't just spin the CPU */ | |
594 | goto again; | |
595 | } | |
596 | } else { | |
7c2d238a | 597 | xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1)); |
c24b5dfa DC |
598 | } |
599 | } | |
600 | ||
fa96acad DC |
601 | void |
602 | __xfs_iflock( | |
603 | struct xfs_inode *ip) | |
604 | { | |
605 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
606 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
607 | ||
608 | do { | |
21417136 | 609 | prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
fa96acad DC |
610 | if (xfs_isiflocked(ip)) |
611 | io_schedule(); | |
612 | } while (!xfs_iflock_nowait(ip)); | |
613 | ||
21417136 | 614 | finish_wait(wq, &wait.wq_entry); |
fa96acad DC |
615 | } |
616 | ||
1da177e4 LT |
617 | STATIC uint |
618 | _xfs_dic2xflags( | |
c8ce540d | 619 | uint16_t di_flags, |
58f88ca2 DC |
620 | uint64_t di_flags2, |
621 | bool has_attr) | |
1da177e4 LT |
622 | { |
623 | uint flags = 0; | |
624 | ||
625 | if (di_flags & XFS_DIFLAG_ANY) { | |
626 | if (di_flags & XFS_DIFLAG_REALTIME) | |
e7b89481 | 627 | flags |= FS_XFLAG_REALTIME; |
1da177e4 | 628 | if (di_flags & XFS_DIFLAG_PREALLOC) |
e7b89481 | 629 | flags |= FS_XFLAG_PREALLOC; |
1da177e4 | 630 | if (di_flags & XFS_DIFLAG_IMMUTABLE) |
e7b89481 | 631 | flags |= FS_XFLAG_IMMUTABLE; |
1da177e4 | 632 | if (di_flags & XFS_DIFLAG_APPEND) |
e7b89481 | 633 | flags |= FS_XFLAG_APPEND; |
1da177e4 | 634 | if (di_flags & XFS_DIFLAG_SYNC) |
e7b89481 | 635 | flags |= FS_XFLAG_SYNC; |
1da177e4 | 636 | if (di_flags & XFS_DIFLAG_NOATIME) |
e7b89481 | 637 | flags |= FS_XFLAG_NOATIME; |
1da177e4 | 638 | if (di_flags & XFS_DIFLAG_NODUMP) |
e7b89481 | 639 | flags |= FS_XFLAG_NODUMP; |
1da177e4 | 640 | if (di_flags & XFS_DIFLAG_RTINHERIT) |
e7b89481 | 641 | flags |= FS_XFLAG_RTINHERIT; |
1da177e4 | 642 | if (di_flags & XFS_DIFLAG_PROJINHERIT) |
e7b89481 | 643 | flags |= FS_XFLAG_PROJINHERIT; |
1da177e4 | 644 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) |
e7b89481 | 645 | flags |= FS_XFLAG_NOSYMLINKS; |
dd9f438e | 646 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
e7b89481 | 647 | flags |= FS_XFLAG_EXTSIZE; |
dd9f438e | 648 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) |
e7b89481 | 649 | flags |= FS_XFLAG_EXTSZINHERIT; |
d3446eac | 650 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
e7b89481 | 651 | flags |= FS_XFLAG_NODEFRAG; |
2a82b8be | 652 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
e7b89481 | 653 | flags |= FS_XFLAG_FILESTREAM; |
1da177e4 LT |
654 | } |
655 | ||
58f88ca2 DC |
656 | if (di_flags2 & XFS_DIFLAG2_ANY) { |
657 | if (di_flags2 & XFS_DIFLAG2_DAX) | |
658 | flags |= FS_XFLAG_DAX; | |
f7ca3522 DW |
659 | if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE) |
660 | flags |= FS_XFLAG_COWEXTSIZE; | |
58f88ca2 DC |
661 | } |
662 | ||
663 | if (has_attr) | |
664 | flags |= FS_XFLAG_HASATTR; | |
665 | ||
1da177e4 LT |
666 | return flags; |
667 | } | |
668 | ||
669 | uint | |
670 | xfs_ip2xflags( | |
58f88ca2 | 671 | struct xfs_inode *ip) |
1da177e4 | 672 | { |
58f88ca2 | 673 | struct xfs_icdinode *dic = &ip->i_d; |
1da177e4 | 674 | |
58f88ca2 | 675 | return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); |
1da177e4 LT |
676 | } |
677 | ||
c24b5dfa DC |
678 | /* |
679 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
680 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
681 | * ci_name->name will point to a the actual name (caller must free) or | |
682 | * will be set to NULL if an exact match is found. | |
683 | */ | |
684 | int | |
685 | xfs_lookup( | |
686 | xfs_inode_t *dp, | |
687 | struct xfs_name *name, | |
688 | xfs_inode_t **ipp, | |
689 | struct xfs_name *ci_name) | |
690 | { | |
691 | xfs_ino_t inum; | |
692 | int error; | |
c24b5dfa DC |
693 | |
694 | trace_xfs_lookup(dp, name); | |
695 | ||
696 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
2451337d | 697 | return -EIO; |
c24b5dfa | 698 | |
c24b5dfa | 699 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
c24b5dfa | 700 | if (error) |
dbad7c99 | 701 | goto out_unlock; |
c24b5dfa DC |
702 | |
703 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); | |
704 | if (error) | |
705 | goto out_free_name; | |
706 | ||
707 | return 0; | |
708 | ||
709 | out_free_name: | |
710 | if (ci_name) | |
711 | kmem_free(ci_name->name); | |
dbad7c99 | 712 | out_unlock: |
c24b5dfa DC |
713 | *ipp = NULL; |
714 | return error; | |
715 | } | |
716 | ||
1da177e4 LT |
717 | /* |
718 | * Allocate an inode on disk and return a copy of its in-core version. | |
719 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
720 | * appropriately within the inode. The uid and gid for the inode are | |
721 | * set according to the contents of the given cred structure. | |
722 | * | |
723 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
724 | * has a free inode available, call xfs_iget() to obtain the in-core |
725 | * version of the allocated inode. Finally, fill in the inode and | |
726 | * log its initial contents. In this case, ialloc_context would be | |
727 | * set to NULL. | |
1da177e4 | 728 | * |
cd856db6 CM |
729 | * If xfs_dialloc() does not have an available inode, it will replenish |
730 | * its supply by doing an allocation. Since we can only do one | |
731 | * allocation within a transaction without deadlocks, we must commit | |
732 | * the current transaction before returning the inode itself. | |
733 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
734 | * The caller should then commit the current transaction, start a new |
735 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
736 | * | |
737 | * To ensure that some other process does not grab the inode that | |
738 | * was allocated during the first call to xfs_ialloc(), this routine | |
739 | * also returns the [locked] bp pointing to the head of the freelist | |
740 | * as ialloc_context. The caller should hold this buffer across | |
741 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
742 | * |
743 | * If we are allocating quota inodes, we do not have a parent inode | |
744 | * to attach to or associate with (i.e. pip == NULL) because they | |
745 | * are not linked into the directory structure - they are attached | |
746 | * directly to the superblock - and so have no parent. | |
1da177e4 | 747 | */ |
0d5a75e9 | 748 | static int |
1da177e4 LT |
749 | xfs_ialloc( |
750 | xfs_trans_t *tp, | |
751 | xfs_inode_t *pip, | |
576b1d67 | 752 | umode_t mode, |
31b084ae | 753 | xfs_nlink_t nlink, |
66f36464 | 754 | dev_t rdev, |
6743099c | 755 | prid_t prid, |
1da177e4 | 756 | xfs_buf_t **ialloc_context, |
1da177e4 LT |
757 | xfs_inode_t **ipp) |
758 | { | |
93848a99 | 759 | struct xfs_mount *mp = tp->t_mountp; |
1da177e4 LT |
760 | xfs_ino_t ino; |
761 | xfs_inode_t *ip; | |
1da177e4 LT |
762 | uint flags; |
763 | int error; | |
95582b00 | 764 | struct timespec64 tv; |
3987848c | 765 | struct inode *inode; |
1da177e4 LT |
766 | |
767 | /* | |
768 | * Call the space management code to pick | |
769 | * the on-disk inode to be allocated. | |
770 | */ | |
f59cf5c2 | 771 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, |
08358906 | 772 | ialloc_context, &ino); |
bf904248 | 773 | if (error) |
1da177e4 | 774 | return error; |
08358906 | 775 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
776 | *ipp = NULL; |
777 | return 0; | |
778 | } | |
779 | ASSERT(*ialloc_context == NULL); | |
780 | ||
8b26984d DC |
781 | /* |
782 | * Protect against obviously corrupt allocation btree records. Later | |
783 | * xfs_iget checks will catch re-allocation of other active in-memory | |
784 | * and on-disk inodes. If we don't catch reallocating the parent inode | |
785 | * here we will deadlock in xfs_iget() so we have to do these checks | |
786 | * first. | |
787 | */ | |
788 | if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) { | |
789 | xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino); | |
790 | return -EFSCORRUPTED; | |
791 | } | |
792 | ||
1da177e4 LT |
793 | /* |
794 | * Get the in-core inode with the lock held exclusively. | |
795 | * This is because we're setting fields here we need | |
796 | * to prevent others from looking at until we're done. | |
797 | */ | |
93848a99 | 798 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
ec3ba85f | 799 | XFS_ILOCK_EXCL, &ip); |
bf904248 | 800 | if (error) |
1da177e4 | 801 | return error; |
1da177e4 | 802 | ASSERT(ip != NULL); |
3987848c | 803 | inode = VFS_I(ip); |
1da177e4 | 804 | |
263997a6 DC |
805 | /* |
806 | * We always convert v1 inodes to v2 now - we only support filesystems | |
807 | * with >= v2 inode capability, so there is no reason for ever leaving | |
808 | * an inode in v1 format. | |
809 | */ | |
810 | if (ip->i_d.di_version == 1) | |
811 | ip->i_d.di_version = 2; | |
812 | ||
c19b3b05 | 813 | inode->i_mode = mode; |
54d7b5c1 | 814 | set_nlink(inode, nlink); |
7aab1b28 DE |
815 | ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); |
816 | ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); | |
66f36464 | 817 | inode->i_rdev = rdev; |
6743099c | 818 | xfs_set_projid(ip, prid); |
1da177e4 | 819 | |
bd186aa9 | 820 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 821 | ip->i_d.di_gid = pip->i_d.di_gid; |
c19b3b05 DC |
822 | if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) |
823 | inode->i_mode |= S_ISGID; | |
1da177e4 LT |
824 | } |
825 | ||
826 | /* | |
827 | * If the group ID of the new file does not match the effective group | |
828 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
829 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
830 | */ | |
831 | if ((irix_sgid_inherit) && | |
c19b3b05 DC |
832 | (inode->i_mode & S_ISGID) && |
833 | (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) | |
834 | inode->i_mode &= ~S_ISGID; | |
1da177e4 LT |
835 | |
836 | ip->i_d.di_size = 0; | |
837 | ip->i_d.di_nextents = 0; | |
838 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 | 839 | |
c2050a45 | 840 | tv = current_time(inode); |
3987848c DC |
841 | inode->i_mtime = tv; |
842 | inode->i_atime = tv; | |
843 | inode->i_ctime = tv; | |
dff35fd4 | 844 | |
1da177e4 LT |
845 | ip->i_d.di_extsize = 0; |
846 | ip->i_d.di_dmevmask = 0; | |
847 | ip->i_d.di_dmstate = 0; | |
848 | ip->i_d.di_flags = 0; | |
93848a99 CH |
849 | |
850 | if (ip->i_d.di_version == 3) { | |
f0e28280 | 851 | inode_set_iversion(inode, 1); |
93848a99 | 852 | ip->i_d.di_flags2 = 0; |
f7ca3522 | 853 | ip->i_d.di_cowextsize = 0; |
c8ce540d DW |
854 | ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec; |
855 | ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec; | |
93848a99 CH |
856 | } |
857 | ||
858 | ||
1da177e4 LT |
859 | flags = XFS_ILOG_CORE; |
860 | switch (mode & S_IFMT) { | |
861 | case S_IFIFO: | |
862 | case S_IFCHR: | |
863 | case S_IFBLK: | |
864 | case S_IFSOCK: | |
865 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1da177e4 LT |
866 | ip->i_df.if_flags = 0; |
867 | flags |= XFS_ILOG_DEV; | |
868 | break; | |
869 | case S_IFREG: | |
870 | case S_IFDIR: | |
b11f94d5 | 871 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
58f88ca2 | 872 | uint di_flags = 0; |
365ca83d | 873 | |
abbede1b | 874 | if (S_ISDIR(mode)) { |
365ca83d NS |
875 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
876 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
877 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
878 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
879 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
880 | } | |
9336e3a7 DC |
881 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
882 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
abbede1b | 883 | } else if (S_ISREG(mode)) { |
613d7043 | 884 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 885 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
886 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
887 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
888 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
889 | } | |
1da177e4 LT |
890 | } |
891 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
892 | xfs_inherit_noatime) | |
365ca83d | 893 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
894 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
895 | xfs_inherit_nodump) | |
365ca83d | 896 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
897 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
898 | xfs_inherit_sync) | |
365ca83d | 899 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
900 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
901 | xfs_inherit_nosymlinks) | |
365ca83d | 902 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
d3446eac BN |
903 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
904 | xfs_inherit_nodefrag) | |
905 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
906 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
907 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
58f88ca2 | 908 | |
365ca83d | 909 | ip->i_d.di_flags |= di_flags; |
1da177e4 | 910 | } |
f7ca3522 DW |
911 | if (pip && |
912 | (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) && | |
913 | pip->i_d.di_version == 3 && | |
914 | ip->i_d.di_version == 3) { | |
56bdf855 LC |
915 | uint64_t di_flags2 = 0; |
916 | ||
f7ca3522 | 917 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) { |
56bdf855 | 918 | di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; |
f7ca3522 DW |
919 | ip->i_d.di_cowextsize = pip->i_d.di_cowextsize; |
920 | } | |
56bdf855 LC |
921 | if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) |
922 | di_flags2 |= XFS_DIFLAG2_DAX; | |
923 | ||
924 | ip->i_d.di_flags2 |= di_flags2; | |
f7ca3522 | 925 | } |
1da177e4 LT |
926 | /* FALLTHROUGH */ |
927 | case S_IFLNK: | |
928 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
929 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
930 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
6bdcf26a | 931 | ip->i_df.if_u1.if_root = NULL; |
1da177e4 LT |
932 | break; |
933 | default: | |
934 | ASSERT(0); | |
935 | } | |
936 | /* | |
937 | * Attribute fork settings for new inode. | |
938 | */ | |
939 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
940 | ip->i_d.di_anextents = 0; | |
941 | ||
942 | /* | |
943 | * Log the new values stuffed into the inode. | |
944 | */ | |
ddc3415a | 945 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
946 | xfs_trans_log_inode(tp, ip, flags); |
947 | ||
58c90473 | 948 | /* now that we have an i_mode we can setup the inode structure */ |
41be8bed | 949 | xfs_setup_inode(ip); |
1da177e4 LT |
950 | |
951 | *ipp = ip; | |
952 | return 0; | |
953 | } | |
954 | ||
e546cb79 DC |
955 | /* |
956 | * Allocates a new inode from disk and return a pointer to the | |
957 | * incore copy. This routine will internally commit the current | |
958 | * transaction and allocate a new one if the Space Manager needed | |
959 | * to do an allocation to replenish the inode free-list. | |
960 | * | |
961 | * This routine is designed to be called from xfs_create and | |
962 | * xfs_create_dir. | |
963 | * | |
964 | */ | |
965 | int | |
966 | xfs_dir_ialloc( | |
967 | xfs_trans_t **tpp, /* input: current transaction; | |
968 | output: may be a new transaction. */ | |
969 | xfs_inode_t *dp, /* directory within whose allocate | |
970 | the inode. */ | |
971 | umode_t mode, | |
972 | xfs_nlink_t nlink, | |
66f36464 | 973 | dev_t rdev, |
e546cb79 | 974 | prid_t prid, /* project id */ |
c959025e | 975 | xfs_inode_t **ipp) /* pointer to inode; it will be |
e546cb79 | 976 | locked. */ |
e546cb79 DC |
977 | { |
978 | xfs_trans_t *tp; | |
e546cb79 DC |
979 | xfs_inode_t *ip; |
980 | xfs_buf_t *ialloc_context = NULL; | |
981 | int code; | |
e546cb79 DC |
982 | void *dqinfo; |
983 | uint tflags; | |
984 | ||
985 | tp = *tpp; | |
986 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
987 | ||
988 | /* | |
989 | * xfs_ialloc will return a pointer to an incore inode if | |
990 | * the Space Manager has an available inode on the free | |
991 | * list. Otherwise, it will do an allocation and replenish | |
992 | * the freelist. Since we can only do one allocation per | |
993 | * transaction without deadlocks, we will need to commit the | |
994 | * current transaction and start a new one. We will then | |
995 | * need to call xfs_ialloc again to get the inode. | |
996 | * | |
997 | * If xfs_ialloc did an allocation to replenish the freelist, | |
998 | * it returns the bp containing the head of the freelist as | |
999 | * ialloc_context. We will hold a lock on it across the | |
1000 | * transaction commit so that no other process can steal | |
1001 | * the inode(s) that we've just allocated. | |
1002 | */ | |
f59cf5c2 CH |
1003 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context, |
1004 | &ip); | |
e546cb79 DC |
1005 | |
1006 | /* | |
1007 | * Return an error if we were unable to allocate a new inode. | |
1008 | * This should only happen if we run out of space on disk or | |
1009 | * encounter a disk error. | |
1010 | */ | |
1011 | if (code) { | |
1012 | *ipp = NULL; | |
1013 | return code; | |
1014 | } | |
1015 | if (!ialloc_context && !ip) { | |
1016 | *ipp = NULL; | |
2451337d | 1017 | return -ENOSPC; |
e546cb79 DC |
1018 | } |
1019 | ||
1020 | /* | |
1021 | * If the AGI buffer is non-NULL, then we were unable to get an | |
1022 | * inode in one operation. We need to commit the current | |
1023 | * transaction and call xfs_ialloc() again. It is guaranteed | |
1024 | * to succeed the second time. | |
1025 | */ | |
1026 | if (ialloc_context) { | |
1027 | /* | |
1028 | * Normally, xfs_trans_commit releases all the locks. | |
1029 | * We call bhold to hang on to the ialloc_context across | |
1030 | * the commit. Holding this buffer prevents any other | |
1031 | * processes from doing any allocations in this | |
1032 | * allocation group. | |
1033 | */ | |
1034 | xfs_trans_bhold(tp, ialloc_context); | |
e546cb79 DC |
1035 | |
1036 | /* | |
1037 | * We want the quota changes to be associated with the next | |
1038 | * transaction, NOT this one. So, detach the dqinfo from this | |
1039 | * and attach it to the next transaction. | |
1040 | */ | |
1041 | dqinfo = NULL; | |
1042 | tflags = 0; | |
1043 | if (tp->t_dqinfo) { | |
1044 | dqinfo = (void *)tp->t_dqinfo; | |
1045 | tp->t_dqinfo = NULL; | |
1046 | tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; | |
1047 | tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); | |
1048 | } | |
1049 | ||
411350df | 1050 | code = xfs_trans_roll(&tp); |
3d3c8b52 | 1051 | |
e546cb79 DC |
1052 | /* |
1053 | * Re-attach the quota info that we detached from prev trx. | |
1054 | */ | |
1055 | if (dqinfo) { | |
1056 | tp->t_dqinfo = dqinfo; | |
1057 | tp->t_flags |= tflags; | |
1058 | } | |
1059 | ||
1060 | if (code) { | |
1061 | xfs_buf_relse(ialloc_context); | |
2e6db6c4 | 1062 | *tpp = tp; |
e546cb79 DC |
1063 | *ipp = NULL; |
1064 | return code; | |
1065 | } | |
1066 | xfs_trans_bjoin(tp, ialloc_context); | |
1067 | ||
1068 | /* | |
1069 | * Call ialloc again. Since we've locked out all | |
1070 | * other allocations in this allocation group, | |
1071 | * this call should always succeed. | |
1072 | */ | |
1073 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, | |
f59cf5c2 | 1074 | &ialloc_context, &ip); |
e546cb79 DC |
1075 | |
1076 | /* | |
1077 | * If we get an error at this point, return to the caller | |
1078 | * so that the current transaction can be aborted. | |
1079 | */ | |
1080 | if (code) { | |
1081 | *tpp = tp; | |
1082 | *ipp = NULL; | |
1083 | return code; | |
1084 | } | |
1085 | ASSERT(!ialloc_context && ip); | |
1086 | ||
e546cb79 DC |
1087 | } |
1088 | ||
1089 | *ipp = ip; | |
1090 | *tpp = tp; | |
1091 | ||
1092 | return 0; | |
1093 | } | |
1094 | ||
1095 | /* | |
54d7b5c1 DC |
1096 | * Decrement the link count on an inode & log the change. If this causes the |
1097 | * link count to go to zero, move the inode to AGI unlinked list so that it can | |
1098 | * be freed when the last active reference goes away via xfs_inactive(). | |
e546cb79 | 1099 | */ |
0d5a75e9 | 1100 | static int /* error */ |
e546cb79 DC |
1101 | xfs_droplink( |
1102 | xfs_trans_t *tp, | |
1103 | xfs_inode_t *ip) | |
1104 | { | |
e546cb79 DC |
1105 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); |
1106 | ||
e546cb79 DC |
1107 | drop_nlink(VFS_I(ip)); |
1108 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1109 | ||
54d7b5c1 DC |
1110 | if (VFS_I(ip)->i_nlink) |
1111 | return 0; | |
1112 | ||
1113 | return xfs_iunlink(tp, ip); | |
e546cb79 DC |
1114 | } |
1115 | ||
e546cb79 DC |
1116 | /* |
1117 | * Increment the link count on an inode & log the change. | |
1118 | */ | |
0d5a75e9 | 1119 | static int |
e546cb79 DC |
1120 | xfs_bumplink( |
1121 | xfs_trans_t *tp, | |
1122 | xfs_inode_t *ip) | |
1123 | { | |
1124 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); | |
1125 | ||
263997a6 | 1126 | ASSERT(ip->i_d.di_version > 1); |
e546cb79 | 1127 | inc_nlink(VFS_I(ip)); |
e546cb79 DC |
1128 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1129 | return 0; | |
1130 | } | |
1131 | ||
c24b5dfa DC |
1132 | int |
1133 | xfs_create( | |
1134 | xfs_inode_t *dp, | |
1135 | struct xfs_name *name, | |
1136 | umode_t mode, | |
66f36464 | 1137 | dev_t rdev, |
c24b5dfa DC |
1138 | xfs_inode_t **ipp) |
1139 | { | |
1140 | int is_dir = S_ISDIR(mode); | |
1141 | struct xfs_mount *mp = dp->i_mount; | |
1142 | struct xfs_inode *ip = NULL; | |
1143 | struct xfs_trans *tp = NULL; | |
1144 | int error; | |
2c3234d1 | 1145 | struct xfs_defer_ops dfops; |
c24b5dfa | 1146 | bool unlock_dp_on_error = false; |
c24b5dfa DC |
1147 | prid_t prid; |
1148 | struct xfs_dquot *udqp = NULL; | |
1149 | struct xfs_dquot *gdqp = NULL; | |
1150 | struct xfs_dquot *pdqp = NULL; | |
062647a8 | 1151 | struct xfs_trans_res *tres; |
c24b5dfa | 1152 | uint resblks; |
c24b5dfa DC |
1153 | |
1154 | trace_xfs_create(dp, name); | |
1155 | ||
1156 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1157 | return -EIO; |
c24b5dfa | 1158 | |
163467d3 | 1159 | prid = xfs_get_initial_prid(dp); |
c24b5dfa DC |
1160 | |
1161 | /* | |
1162 | * Make sure that we have allocated dquot(s) on disk. | |
1163 | */ | |
7aab1b28 DE |
1164 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), |
1165 | xfs_kgid_to_gid(current_fsgid()), prid, | |
c24b5dfa DC |
1166 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
1167 | &udqp, &gdqp, &pdqp); | |
1168 | if (error) | |
1169 | return error; | |
1170 | ||
1171 | if (is_dir) { | |
c24b5dfa | 1172 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); |
062647a8 | 1173 | tres = &M_RES(mp)->tr_mkdir; |
c24b5dfa DC |
1174 | } else { |
1175 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
062647a8 | 1176 | tres = &M_RES(mp)->tr_create; |
c24b5dfa DC |
1177 | } |
1178 | ||
c24b5dfa DC |
1179 | /* |
1180 | * Initially assume that the file does not exist and | |
1181 | * reserve the resources for that case. If that is not | |
1182 | * the case we'll drop the one we have and get a more | |
1183 | * appropriate transaction later. | |
1184 | */ | |
253f4911 | 1185 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); |
2451337d | 1186 | if (error == -ENOSPC) { |
c24b5dfa DC |
1187 | /* flush outstanding delalloc blocks and retry */ |
1188 | xfs_flush_inodes(mp); | |
253f4911 | 1189 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); |
c24b5dfa | 1190 | } |
4906e215 | 1191 | if (error) |
253f4911 | 1192 | goto out_release_inode; |
c24b5dfa | 1193 | |
65523218 | 1194 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
c24b5dfa DC |
1195 | unlock_dp_on_error = true; |
1196 | ||
f16dea54 | 1197 | xfs_defer_init(tp, &dfops, &tp->t_firstblock); |
c24b5dfa DC |
1198 | |
1199 | /* | |
1200 | * Reserve disk quota and the inode. | |
1201 | */ | |
1202 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1203 | pdqp, resblks, 1, 0); | |
1204 | if (error) | |
1205 | goto out_trans_cancel; | |
1206 | ||
c24b5dfa DC |
1207 | /* |
1208 | * A newly created regular or special file just has one directory | |
1209 | * entry pointing to them, but a directory also the "." entry | |
1210 | * pointing to itself. | |
1211 | */ | |
c959025e | 1212 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip); |
d6077aa3 | 1213 | if (error) |
4906e215 | 1214 | goto out_trans_cancel; |
c24b5dfa DC |
1215 | |
1216 | /* | |
1217 | * Now we join the directory inode to the transaction. We do not do it | |
1218 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1219 | * (and release all the locks). An error from here on will result in | |
1220 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1221 | * error path. | |
1222 | */ | |
65523218 | 1223 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1224 | unlock_dp_on_error = false; |
1225 | ||
381eee69 | 1226 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
c9cfdb38 | 1227 | resblks ? |
c24b5dfa DC |
1228 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); |
1229 | if (error) { | |
2451337d | 1230 | ASSERT(error != -ENOSPC); |
4906e215 | 1231 | goto out_trans_cancel; |
c24b5dfa DC |
1232 | } |
1233 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1234 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1235 | ||
1236 | if (is_dir) { | |
1237 | error = xfs_dir_init(tp, ip, dp); | |
1238 | if (error) | |
1239 | goto out_bmap_cancel; | |
1240 | ||
1241 | error = xfs_bumplink(tp, dp); | |
1242 | if (error) | |
1243 | goto out_bmap_cancel; | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * If this is a synchronous mount, make sure that the | |
1248 | * create transaction goes to disk before returning to | |
1249 | * the user. | |
1250 | */ | |
1251 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
1252 | xfs_trans_set_sync(tp); | |
1253 | ||
1254 | /* | |
1255 | * Attach the dquot(s) to the inodes and modify them incore. | |
1256 | * These ids of the inode couldn't have changed since the new | |
1257 | * inode has been locked ever since it was created. | |
1258 | */ | |
1259 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1260 | ||
8ad7c629 | 1261 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa DC |
1262 | if (error) |
1263 | goto out_bmap_cancel; | |
1264 | ||
70393313 | 1265 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
1266 | if (error) |
1267 | goto out_release_inode; | |
1268 | ||
1269 | xfs_qm_dqrele(udqp); | |
1270 | xfs_qm_dqrele(gdqp); | |
1271 | xfs_qm_dqrele(pdqp); | |
1272 | ||
1273 | *ipp = ip; | |
1274 | return 0; | |
1275 | ||
1276 | out_bmap_cancel: | |
2c3234d1 | 1277 | xfs_defer_cancel(&dfops); |
c24b5dfa | 1278 | out_trans_cancel: |
4906e215 | 1279 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1280 | out_release_inode: |
1281 | /* | |
58c90473 DC |
1282 | * Wait until after the current transaction is aborted to finish the |
1283 | * setup of the inode and release the inode. This prevents recursive | |
1284 | * transactions and deadlocks from xfs_inactive. | |
c24b5dfa | 1285 | */ |
58c90473 DC |
1286 | if (ip) { |
1287 | xfs_finish_inode_setup(ip); | |
c24b5dfa | 1288 | IRELE(ip); |
58c90473 | 1289 | } |
c24b5dfa DC |
1290 | |
1291 | xfs_qm_dqrele(udqp); | |
1292 | xfs_qm_dqrele(gdqp); | |
1293 | xfs_qm_dqrele(pdqp); | |
1294 | ||
1295 | if (unlock_dp_on_error) | |
65523218 | 1296 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1297 | return error; |
1298 | } | |
1299 | ||
99b6436b ZYW |
1300 | int |
1301 | xfs_create_tmpfile( | |
1302 | struct xfs_inode *dp, | |
330033d6 BF |
1303 | umode_t mode, |
1304 | struct xfs_inode **ipp) | |
99b6436b ZYW |
1305 | { |
1306 | struct xfs_mount *mp = dp->i_mount; | |
1307 | struct xfs_inode *ip = NULL; | |
1308 | struct xfs_trans *tp = NULL; | |
1309 | int error; | |
99b6436b ZYW |
1310 | prid_t prid; |
1311 | struct xfs_dquot *udqp = NULL; | |
1312 | struct xfs_dquot *gdqp = NULL; | |
1313 | struct xfs_dquot *pdqp = NULL; | |
1314 | struct xfs_trans_res *tres; | |
1315 | uint resblks; | |
1316 | ||
1317 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1318 | return -EIO; |
99b6436b ZYW |
1319 | |
1320 | prid = xfs_get_initial_prid(dp); | |
1321 | ||
1322 | /* | |
1323 | * Make sure that we have allocated dquot(s) on disk. | |
1324 | */ | |
1325 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), | |
1326 | xfs_kgid_to_gid(current_fsgid()), prid, | |
1327 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, | |
1328 | &udqp, &gdqp, &pdqp); | |
1329 | if (error) | |
1330 | return error; | |
1331 | ||
1332 | resblks = XFS_IALLOC_SPACE_RES(mp); | |
99b6436b | 1333 | tres = &M_RES(mp)->tr_create_tmpfile; |
253f4911 CH |
1334 | |
1335 | error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); | |
4906e215 | 1336 | if (error) |
253f4911 | 1337 | goto out_release_inode; |
99b6436b ZYW |
1338 | |
1339 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1340 | pdqp, resblks, 1, 0); | |
1341 | if (error) | |
1342 | goto out_trans_cancel; | |
1343 | ||
c959025e | 1344 | error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, prid, &ip); |
d6077aa3 | 1345 | if (error) |
4906e215 | 1346 | goto out_trans_cancel; |
99b6436b ZYW |
1347 | |
1348 | if (mp->m_flags & XFS_MOUNT_WSYNC) | |
1349 | xfs_trans_set_sync(tp); | |
1350 | ||
1351 | /* | |
1352 | * Attach the dquot(s) to the inodes and modify them incore. | |
1353 | * These ids of the inode couldn't have changed since the new | |
1354 | * inode has been locked ever since it was created. | |
1355 | */ | |
1356 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1357 | ||
99b6436b ZYW |
1358 | error = xfs_iunlink(tp, ip); |
1359 | if (error) | |
4906e215 | 1360 | goto out_trans_cancel; |
99b6436b | 1361 | |
70393313 | 1362 | error = xfs_trans_commit(tp); |
99b6436b ZYW |
1363 | if (error) |
1364 | goto out_release_inode; | |
1365 | ||
1366 | xfs_qm_dqrele(udqp); | |
1367 | xfs_qm_dqrele(gdqp); | |
1368 | xfs_qm_dqrele(pdqp); | |
1369 | ||
330033d6 | 1370 | *ipp = ip; |
99b6436b ZYW |
1371 | return 0; |
1372 | ||
99b6436b | 1373 | out_trans_cancel: |
4906e215 | 1374 | xfs_trans_cancel(tp); |
99b6436b ZYW |
1375 | out_release_inode: |
1376 | /* | |
58c90473 DC |
1377 | * Wait until after the current transaction is aborted to finish the |
1378 | * setup of the inode and release the inode. This prevents recursive | |
1379 | * transactions and deadlocks from xfs_inactive. | |
99b6436b | 1380 | */ |
58c90473 DC |
1381 | if (ip) { |
1382 | xfs_finish_inode_setup(ip); | |
99b6436b | 1383 | IRELE(ip); |
58c90473 | 1384 | } |
99b6436b ZYW |
1385 | |
1386 | xfs_qm_dqrele(udqp); | |
1387 | xfs_qm_dqrele(gdqp); | |
1388 | xfs_qm_dqrele(pdqp); | |
1389 | ||
1390 | return error; | |
1391 | } | |
1392 | ||
c24b5dfa DC |
1393 | int |
1394 | xfs_link( | |
1395 | xfs_inode_t *tdp, | |
1396 | xfs_inode_t *sip, | |
1397 | struct xfs_name *target_name) | |
1398 | { | |
1399 | xfs_mount_t *mp = tdp->i_mount; | |
1400 | xfs_trans_t *tp; | |
1401 | int error; | |
2c3234d1 | 1402 | struct xfs_defer_ops dfops; |
c24b5dfa DC |
1403 | int resblks; |
1404 | ||
1405 | trace_xfs_link(tdp, target_name); | |
1406 | ||
c19b3b05 | 1407 | ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); |
c24b5dfa DC |
1408 | |
1409 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 1410 | return -EIO; |
c24b5dfa | 1411 | |
c14cfcca | 1412 | error = xfs_qm_dqattach(sip); |
c24b5dfa DC |
1413 | if (error) |
1414 | goto std_return; | |
1415 | ||
c14cfcca | 1416 | error = xfs_qm_dqattach(tdp); |
c24b5dfa DC |
1417 | if (error) |
1418 | goto std_return; | |
1419 | ||
c24b5dfa | 1420 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
253f4911 | 1421 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); |
2451337d | 1422 | if (error == -ENOSPC) { |
c24b5dfa | 1423 | resblks = 0; |
253f4911 | 1424 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); |
c24b5dfa | 1425 | } |
4906e215 | 1426 | if (error) |
253f4911 | 1427 | goto std_return; |
c24b5dfa | 1428 | |
7c2d238a | 1429 | xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1430 | |
1431 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
65523218 | 1432 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1433 | |
1434 | /* | |
1435 | * If we are using project inheritance, we only allow hard link | |
1436 | * creation in our tree when the project IDs are the same; else | |
1437 | * the tree quota mechanism could be circumvented. | |
1438 | */ | |
1439 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
1440 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { | |
2451337d | 1441 | error = -EXDEV; |
c24b5dfa DC |
1442 | goto error_return; |
1443 | } | |
1444 | ||
94f3cad5 ES |
1445 | if (!resblks) { |
1446 | error = xfs_dir_canenter(tp, tdp, target_name); | |
1447 | if (error) | |
1448 | goto error_return; | |
1449 | } | |
c24b5dfa | 1450 | |
f16dea54 | 1451 | xfs_defer_init(tp, &dfops, &tp->t_firstblock); |
c24b5dfa | 1452 | |
54d7b5c1 DC |
1453 | /* |
1454 | * Handle initial link state of O_TMPFILE inode | |
1455 | */ | |
1456 | if (VFS_I(sip)->i_nlink == 0) { | |
ab297431 ZYW |
1457 | error = xfs_iunlink_remove(tp, sip); |
1458 | if (error) | |
4906e215 | 1459 | goto error_return; |
ab297431 ZYW |
1460 | } |
1461 | ||
c24b5dfa | 1462 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
381eee69 | 1463 | resblks); |
c24b5dfa | 1464 | if (error) |
4906e215 | 1465 | goto error_return; |
c24b5dfa DC |
1466 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1467 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
1468 | ||
1469 | error = xfs_bumplink(tp, sip); | |
1470 | if (error) | |
4906e215 | 1471 | goto error_return; |
c24b5dfa DC |
1472 | |
1473 | /* | |
1474 | * If this is a synchronous mount, make sure that the | |
1475 | * link transaction goes to disk before returning to | |
1476 | * the user. | |
1477 | */ | |
f6106efa | 1478 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
c24b5dfa | 1479 | xfs_trans_set_sync(tp); |
c24b5dfa | 1480 | |
8ad7c629 | 1481 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa | 1482 | if (error) { |
2c3234d1 | 1483 | xfs_defer_cancel(&dfops); |
4906e215 | 1484 | goto error_return; |
c24b5dfa DC |
1485 | } |
1486 | ||
70393313 | 1487 | return xfs_trans_commit(tp); |
c24b5dfa | 1488 | |
c24b5dfa | 1489 | error_return: |
4906e215 | 1490 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1491 | std_return: |
1492 | return error; | |
1493 | } | |
1494 | ||
363e59ba DW |
1495 | /* Clear the reflink flag and the cowblocks tag if possible. */ |
1496 | static void | |
1497 | xfs_itruncate_clear_reflink_flags( | |
1498 | struct xfs_inode *ip) | |
1499 | { | |
1500 | struct xfs_ifork *dfork; | |
1501 | struct xfs_ifork *cfork; | |
1502 | ||
1503 | if (!xfs_is_reflink_inode(ip)) | |
1504 | return; | |
1505 | dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
1506 | cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
1507 | if (dfork->if_bytes == 0 && cfork->if_bytes == 0) | |
1508 | ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; | |
1509 | if (cfork->if_bytes == 0) | |
1510 | xfs_inode_clear_cowblocks_tag(ip); | |
1511 | } | |
1512 | ||
1da177e4 | 1513 | /* |
8f04c47a CH |
1514 | * Free up the underlying blocks past new_size. The new size must be smaller |
1515 | * than the current size. This routine can be used both for the attribute and | |
1516 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1517 | * |
f6485057 DC |
1518 | * The transaction passed to this routine must have made a permanent log |
1519 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1520 | * given transaction and start new ones, so make sure everything involved in | |
1521 | * the transaction is tidy before calling here. Some transaction will be | |
1522 | * returned to the caller to be committed. The incoming transaction must | |
1523 | * already include the inode, and both inode locks must be held exclusively. | |
1524 | * The inode must also be "held" within the transaction. On return the inode | |
1525 | * will be "held" within the returned transaction. This routine does NOT | |
1526 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1527 | * |
f6485057 DC |
1528 | * If we get an error, we must return with the inode locked and linked into the |
1529 | * current transaction. This keeps things simple for the higher level code, | |
1530 | * because it always knows that the inode is locked and held in the transaction | |
1531 | * that returns to it whether errors occur or not. We don't mark the inode | |
1532 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1533 | */ |
1534 | int | |
4e529339 | 1535 | xfs_itruncate_extents_flags( |
8f04c47a CH |
1536 | struct xfs_trans **tpp, |
1537 | struct xfs_inode *ip, | |
1538 | int whichfork, | |
13b86fc3 | 1539 | xfs_fsize_t new_size, |
4e529339 | 1540 | int flags) |
1da177e4 | 1541 | { |
8f04c47a CH |
1542 | struct xfs_mount *mp = ip->i_mount; |
1543 | struct xfs_trans *tp = *tpp; | |
4bcfa613 | 1544 | struct xfs_defer_ops *odfops = tp->t_dfops; |
2c3234d1 | 1545 | struct xfs_defer_ops dfops; |
8f04c47a CH |
1546 | xfs_fileoff_t first_unmap_block; |
1547 | xfs_fileoff_t last_block; | |
1548 | xfs_filblks_t unmap_len; | |
8f04c47a CH |
1549 | int error = 0; |
1550 | int done = 0; | |
1da177e4 | 1551 | |
0b56185b CH |
1552 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1553 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1554 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1555 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1556 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1557 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1558 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1559 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1560 | |
673e8e59 CH |
1561 | trace_xfs_itruncate_extents_start(ip, new_size); |
1562 | ||
4e529339 | 1563 | flags |= xfs_bmapi_aflag(whichfork); |
13b86fc3 | 1564 | |
1da177e4 LT |
1565 | /* |
1566 | * Since it is possible for space to become allocated beyond | |
1567 | * the end of the file (in a crash where the space is allocated | |
1568 | * but the inode size is not yet updated), simply remove any | |
1569 | * blocks which show up between the new EOF and the maximum | |
1570 | * possible file size. If the first block to be removed is | |
1571 | * beyond the maximum file size (ie it is the same as last_block), | |
1572 | * then there is nothing to do. | |
1573 | */ | |
8f04c47a | 1574 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1575 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1576 | if (first_unmap_block == last_block) |
1577 | return 0; | |
1578 | ||
1579 | ASSERT(first_unmap_block < last_block); | |
1580 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1581 | while (!done) { |
37283797 | 1582 | xfs_defer_init(tp, &dfops, &tp->t_firstblock); |
13b86fc3 | 1583 | error = xfs_bunmapi(tp, ip, first_unmap_block, unmap_len, flags, |
37283797 | 1584 | XFS_ITRUNC_MAX_EXTENTS, &tp->t_firstblock, |
ccd9d911 | 1585 | &done); |
8f04c47a CH |
1586 | if (error) |
1587 | goto out_bmap_cancel; | |
1da177e4 LT |
1588 | |
1589 | /* | |
1590 | * Duplicate the transaction that has the permanent | |
1591 | * reservation and commit the old transaction. | |
1592 | */ | |
4bcfa613 BF |
1593 | xfs_defer_ijoin(tp->t_dfops, ip); |
1594 | error = xfs_defer_finish(&tp, tp->t_dfops); | |
8f04c47a CH |
1595 | if (error) |
1596 | goto out_bmap_cancel; | |
1da177e4 | 1597 | |
411350df | 1598 | error = xfs_trans_roll_inode(&tp, ip); |
f6485057 | 1599 | if (error) |
8f04c47a | 1600 | goto out; |
1da177e4 | 1601 | } |
8f04c47a | 1602 | |
4919d42a DW |
1603 | if (whichfork == XFS_DATA_FORK) { |
1604 | /* Remove all pending CoW reservations. */ | |
1605 | error = xfs_reflink_cancel_cow_blocks(ip, &tp, | |
1606 | first_unmap_block, last_block, true); | |
1607 | if (error) | |
1608 | goto out; | |
aa8968f2 | 1609 | |
4919d42a DW |
1610 | xfs_itruncate_clear_reflink_flags(ip); |
1611 | } | |
aa8968f2 | 1612 | |
673e8e59 CH |
1613 | /* |
1614 | * Always re-log the inode so that our permanent transaction can keep | |
1615 | * on rolling it forward in the log. | |
1616 | */ | |
1617 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1618 | ||
1619 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1620 | ||
8f04c47a | 1621 | out: |
4bcfa613 BF |
1622 | /* ->t_dfops points to local stack, don't leak it! */ |
1623 | tp->t_dfops = odfops; | |
8f04c47a CH |
1624 | *tpp = tp; |
1625 | return error; | |
1626 | out_bmap_cancel: | |
1da177e4 | 1627 | /* |
8f04c47a CH |
1628 | * If the bunmapi call encounters an error, return to the caller where |
1629 | * the transaction can be properly aborted. We just need to make sure | |
1630 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1631 | */ |
4bcfa613 | 1632 | xfs_defer_cancel(tp->t_dfops); |
8f04c47a CH |
1633 | goto out; |
1634 | } | |
1635 | ||
c24b5dfa DC |
1636 | int |
1637 | xfs_release( | |
1638 | xfs_inode_t *ip) | |
1639 | { | |
1640 | xfs_mount_t *mp = ip->i_mount; | |
1641 | int error; | |
1642 | ||
c19b3b05 | 1643 | if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) |
c24b5dfa DC |
1644 | return 0; |
1645 | ||
1646 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1647 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1648 | return 0; | |
1649 | ||
1650 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1651 | int truncated; | |
1652 | ||
c24b5dfa DC |
1653 | /* |
1654 | * If we previously truncated this file and removed old data | |
1655 | * in the process, we want to initiate "early" writeout on | |
1656 | * the last close. This is an attempt to combat the notorious | |
1657 | * NULL files problem which is particularly noticeable from a | |
1658 | * truncate down, buffered (re-)write (delalloc), followed by | |
1659 | * a crash. What we are effectively doing here is | |
1660 | * significantly reducing the time window where we'd otherwise | |
1661 | * be exposed to that problem. | |
1662 | */ | |
1663 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); | |
1664 | if (truncated) { | |
1665 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
eac152b4 | 1666 | if (ip->i_delayed_blks > 0) { |
2451337d | 1667 | error = filemap_flush(VFS_I(ip)->i_mapping); |
c24b5dfa DC |
1668 | if (error) |
1669 | return error; | |
1670 | } | |
1671 | } | |
1672 | } | |
1673 | ||
54d7b5c1 | 1674 | if (VFS_I(ip)->i_nlink == 0) |
c24b5dfa DC |
1675 | return 0; |
1676 | ||
1677 | if (xfs_can_free_eofblocks(ip, false)) { | |
1678 | ||
a36b9261 BF |
1679 | /* |
1680 | * Check if the inode is being opened, written and closed | |
1681 | * frequently and we have delayed allocation blocks outstanding | |
1682 | * (e.g. streaming writes from the NFS server), truncating the | |
1683 | * blocks past EOF will cause fragmentation to occur. | |
1684 | * | |
1685 | * In this case don't do the truncation, but we have to be | |
1686 | * careful how we detect this case. Blocks beyond EOF show up as | |
1687 | * i_delayed_blks even when the inode is clean, so we need to | |
1688 | * truncate them away first before checking for a dirty release. | |
1689 | * Hence on the first dirty close we will still remove the | |
1690 | * speculative allocation, but after that we will leave it in | |
1691 | * place. | |
1692 | */ | |
1693 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
1694 | return 0; | |
c24b5dfa DC |
1695 | /* |
1696 | * If we can't get the iolock just skip truncating the blocks | |
1697 | * past EOF because we could deadlock with the mmap_sem | |
a36b9261 | 1698 | * otherwise. We'll get another chance to drop them once the |
c24b5dfa DC |
1699 | * last reference to the inode is dropped, so we'll never leak |
1700 | * blocks permanently. | |
c24b5dfa | 1701 | */ |
a36b9261 BF |
1702 | if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { |
1703 | error = xfs_free_eofblocks(ip); | |
1704 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1705 | if (error) | |
1706 | return error; | |
1707 | } | |
c24b5dfa DC |
1708 | |
1709 | /* delalloc blocks after truncation means it really is dirty */ | |
1710 | if (ip->i_delayed_blks) | |
1711 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
1712 | } | |
1713 | return 0; | |
1714 | } | |
1715 | ||
f7be2d7f BF |
1716 | /* |
1717 | * xfs_inactive_truncate | |
1718 | * | |
1719 | * Called to perform a truncate when an inode becomes unlinked. | |
1720 | */ | |
1721 | STATIC int | |
1722 | xfs_inactive_truncate( | |
1723 | struct xfs_inode *ip) | |
1724 | { | |
1725 | struct xfs_mount *mp = ip->i_mount; | |
1726 | struct xfs_trans *tp; | |
1727 | int error; | |
1728 | ||
253f4911 | 1729 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
f7be2d7f BF |
1730 | if (error) { |
1731 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
f7be2d7f BF |
1732 | return error; |
1733 | } | |
1734 | ||
1735 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1736 | xfs_trans_ijoin(tp, ip, 0); | |
1737 | ||
1738 | /* | |
1739 | * Log the inode size first to prevent stale data exposure in the event | |
1740 | * of a system crash before the truncate completes. See the related | |
69bca807 | 1741 | * comment in xfs_vn_setattr_size() for details. |
f7be2d7f BF |
1742 | */ |
1743 | ip->i_d.di_size = 0; | |
1744 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1745 | ||
1746 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
1747 | if (error) | |
1748 | goto error_trans_cancel; | |
1749 | ||
1750 | ASSERT(ip->i_d.di_nextents == 0); | |
1751 | ||
70393313 | 1752 | error = xfs_trans_commit(tp); |
f7be2d7f BF |
1753 | if (error) |
1754 | goto error_unlock; | |
1755 | ||
1756 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1757 | return 0; | |
1758 | ||
1759 | error_trans_cancel: | |
4906e215 | 1760 | xfs_trans_cancel(tp); |
f7be2d7f BF |
1761 | error_unlock: |
1762 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1763 | return error; | |
1764 | } | |
1765 | ||
88877d2b BF |
1766 | /* |
1767 | * xfs_inactive_ifree() | |
1768 | * | |
1769 | * Perform the inode free when an inode is unlinked. | |
1770 | */ | |
1771 | STATIC int | |
1772 | xfs_inactive_ifree( | |
1773 | struct xfs_inode *ip) | |
1774 | { | |
2c3234d1 | 1775 | struct xfs_defer_ops dfops; |
88877d2b | 1776 | xfs_fsblock_t first_block; |
88877d2b BF |
1777 | struct xfs_mount *mp = ip->i_mount; |
1778 | struct xfs_trans *tp; | |
1779 | int error; | |
1780 | ||
9d43b180 | 1781 | /* |
76d771b4 CH |
1782 | * We try to use a per-AG reservation for any block needed by the finobt |
1783 | * tree, but as the finobt feature predates the per-AG reservation | |
1784 | * support a degraded file system might not have enough space for the | |
1785 | * reservation at mount time. In that case try to dip into the reserved | |
1786 | * pool and pray. | |
9d43b180 BF |
1787 | * |
1788 | * Send a warning if the reservation does happen to fail, as the inode | |
1789 | * now remains allocated and sits on the unlinked list until the fs is | |
1790 | * repaired. | |
1791 | */ | |
76d771b4 CH |
1792 | if (unlikely(mp->m_inotbt_nores)) { |
1793 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, | |
1794 | XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, | |
1795 | &tp); | |
1796 | } else { | |
1797 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); | |
1798 | } | |
88877d2b | 1799 | if (error) { |
2451337d | 1800 | if (error == -ENOSPC) { |
9d43b180 BF |
1801 | xfs_warn_ratelimited(mp, |
1802 | "Failed to remove inode(s) from unlinked list. " | |
1803 | "Please free space, unmount and run xfs_repair."); | |
1804 | } else { | |
1805 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1806 | } | |
88877d2b BF |
1807 | return error; |
1808 | } | |
1809 | ||
1810 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1811 | xfs_trans_ijoin(tp, ip, 0); | |
1812 | ||
bcd2c9f3 | 1813 | xfs_defer_init(tp, &dfops, &first_block); |
0e0417f3 | 1814 | error = xfs_ifree(tp, ip); |
88877d2b BF |
1815 | if (error) { |
1816 | /* | |
1817 | * If we fail to free the inode, shut down. The cancel | |
1818 | * might do that, we need to make sure. Otherwise the | |
1819 | * inode might be lost for a long time or forever. | |
1820 | */ | |
1821 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1822 | xfs_notice(mp, "%s: xfs_ifree returned error %d", | |
1823 | __func__, error); | |
1824 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1825 | } | |
4906e215 | 1826 | xfs_trans_cancel(tp); |
88877d2b BF |
1827 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1828 | return error; | |
1829 | } | |
1830 | ||
1831 | /* | |
1832 | * Credit the quota account(s). The inode is gone. | |
1833 | */ | |
1834 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1835 | ||
1836 | /* | |
d4a97a04 BF |
1837 | * Just ignore errors at this point. There is nothing we can do except |
1838 | * to try to keep going. Make sure it's not a silent error. | |
88877d2b | 1839 | */ |
8ad7c629 | 1840 | error = xfs_defer_finish(&tp, &dfops); |
d4a97a04 | 1841 | if (error) { |
310a75a3 | 1842 | xfs_notice(mp, "%s: xfs_defer_finish returned error %d", |
88877d2b | 1843 | __func__, error); |
2c3234d1 | 1844 | xfs_defer_cancel(&dfops); |
d4a97a04 | 1845 | } |
70393313 | 1846 | error = xfs_trans_commit(tp); |
88877d2b BF |
1847 | if (error) |
1848 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", | |
1849 | __func__, error); | |
1850 | ||
1851 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1852 | return 0; | |
1853 | } | |
1854 | ||
c24b5dfa DC |
1855 | /* |
1856 | * xfs_inactive | |
1857 | * | |
1858 | * This is called when the vnode reference count for the vnode | |
1859 | * goes to zero. If the file has been unlinked, then it must | |
1860 | * now be truncated. Also, we clear all of the read-ahead state | |
1861 | * kept for the inode here since the file is now closed. | |
1862 | */ | |
74564fb4 | 1863 | void |
c24b5dfa DC |
1864 | xfs_inactive( |
1865 | xfs_inode_t *ip) | |
1866 | { | |
3d3c8b52 | 1867 | struct xfs_mount *mp; |
6231848c | 1868 | struct xfs_ifork *cow_ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); |
3d3c8b52 JL |
1869 | int error; |
1870 | int truncate = 0; | |
c24b5dfa DC |
1871 | |
1872 | /* | |
1873 | * If the inode is already free, then there can be nothing | |
1874 | * to clean up here. | |
1875 | */ | |
c19b3b05 | 1876 | if (VFS_I(ip)->i_mode == 0) { |
c24b5dfa DC |
1877 | ASSERT(ip->i_df.if_real_bytes == 0); |
1878 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
74564fb4 | 1879 | return; |
c24b5dfa DC |
1880 | } |
1881 | ||
1882 | mp = ip->i_mount; | |
17c12bcd | 1883 | ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); |
c24b5dfa | 1884 | |
c24b5dfa DC |
1885 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
1886 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
74564fb4 | 1887 | return; |
c24b5dfa | 1888 | |
6231848c DW |
1889 | /* Try to clean out the cow blocks if there are any. */ |
1890 | if (xfs_is_reflink_inode(ip) && cow_ifp->if_bytes > 0) | |
1891 | xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true); | |
1892 | ||
54d7b5c1 | 1893 | if (VFS_I(ip)->i_nlink != 0) { |
c24b5dfa DC |
1894 | /* |
1895 | * force is true because we are evicting an inode from the | |
1896 | * cache. Post-eof blocks must be freed, lest we end up with | |
1897 | * broken free space accounting. | |
3b4683c2 BF |
1898 | * |
1899 | * Note: don't bother with iolock here since lockdep complains | |
1900 | * about acquiring it in reclaim context. We have the only | |
1901 | * reference to the inode at this point anyways. | |
c24b5dfa | 1902 | */ |
3b4683c2 | 1903 | if (xfs_can_free_eofblocks(ip, true)) |
a36b9261 | 1904 | xfs_free_eofblocks(ip); |
74564fb4 BF |
1905 | |
1906 | return; | |
c24b5dfa DC |
1907 | } |
1908 | ||
c19b3b05 | 1909 | if (S_ISREG(VFS_I(ip)->i_mode) && |
c24b5dfa DC |
1910 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || |
1911 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) | |
1912 | truncate = 1; | |
1913 | ||
c14cfcca | 1914 | error = xfs_qm_dqattach(ip); |
c24b5dfa | 1915 | if (error) |
74564fb4 | 1916 | return; |
c24b5dfa | 1917 | |
c19b3b05 | 1918 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
36b21dde | 1919 | error = xfs_inactive_symlink(ip); |
f7be2d7f BF |
1920 | else if (truncate) |
1921 | error = xfs_inactive_truncate(ip); | |
1922 | if (error) | |
74564fb4 | 1923 | return; |
c24b5dfa DC |
1924 | |
1925 | /* | |
1926 | * If there are attributes associated with the file then blow them away | |
1927 | * now. The code calls a routine that recursively deconstructs the | |
6dfe5a04 | 1928 | * attribute fork. If also blows away the in-core attribute fork. |
c24b5dfa | 1929 | */ |
6dfe5a04 | 1930 | if (XFS_IFORK_Q(ip)) { |
c24b5dfa DC |
1931 | error = xfs_attr_inactive(ip); |
1932 | if (error) | |
74564fb4 | 1933 | return; |
c24b5dfa DC |
1934 | } |
1935 | ||
6dfe5a04 | 1936 | ASSERT(!ip->i_afp); |
c24b5dfa | 1937 | ASSERT(ip->i_d.di_anextents == 0); |
6dfe5a04 | 1938 | ASSERT(ip->i_d.di_forkoff == 0); |
c24b5dfa DC |
1939 | |
1940 | /* | |
1941 | * Free the inode. | |
1942 | */ | |
88877d2b BF |
1943 | error = xfs_inactive_ifree(ip); |
1944 | if (error) | |
74564fb4 | 1945 | return; |
c24b5dfa DC |
1946 | |
1947 | /* | |
1948 | * Release the dquots held by inode, if any. | |
1949 | */ | |
1950 | xfs_qm_dqdetach(ip); | |
c24b5dfa DC |
1951 | } |
1952 | ||
1da177e4 | 1953 | /* |
54d7b5c1 DC |
1954 | * This is called when the inode's link count goes to 0 or we are creating a |
1955 | * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be | |
1956 | * set to true as the link count is dropped to zero by the VFS after we've | |
1957 | * created the file successfully, so we have to add it to the unlinked list | |
1958 | * while the link count is non-zero. | |
1959 | * | |
1960 | * We place the on-disk inode on a list in the AGI. It will be pulled from this | |
1961 | * list when the inode is freed. | |
1da177e4 | 1962 | */ |
54d7b5c1 | 1963 | STATIC int |
1da177e4 | 1964 | xfs_iunlink( |
54d7b5c1 DC |
1965 | struct xfs_trans *tp, |
1966 | struct xfs_inode *ip) | |
1da177e4 | 1967 | { |
54d7b5c1 | 1968 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
1969 | xfs_agi_t *agi; |
1970 | xfs_dinode_t *dip; | |
1971 | xfs_buf_t *agibp; | |
1972 | xfs_buf_t *ibp; | |
1da177e4 LT |
1973 | xfs_agino_t agino; |
1974 | short bucket_index; | |
1975 | int offset; | |
1976 | int error; | |
1da177e4 | 1977 | |
c19b3b05 | 1978 | ASSERT(VFS_I(ip)->i_mode != 0); |
1da177e4 | 1979 | |
1da177e4 LT |
1980 | /* |
1981 | * Get the agi buffer first. It ensures lock ordering | |
1982 | * on the list. | |
1983 | */ | |
5e1be0fb | 1984 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 1985 | if (error) |
1da177e4 | 1986 | return error; |
1da177e4 | 1987 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1988 | |
1da177e4 LT |
1989 | /* |
1990 | * Get the index into the agi hash table for the | |
1991 | * list this inode will go on. | |
1992 | */ | |
1993 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1994 | ASSERT(agino != 0); | |
1995 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1996 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1997 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1998 | |
69ef921b | 1999 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
2000 | /* |
2001 | * There is already another inode in the bucket we need | |
2002 | * to add ourselves to. Add us at the front of the list. | |
2003 | * Here we put the head pointer into our next pointer, | |
2004 | * and then we fall through to point the head at us. | |
2005 | */ | |
475ee413 CH |
2006 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2007 | 0, 0); | |
c319b58b VA |
2008 | if (error) |
2009 | return error; | |
2010 | ||
69ef921b | 2011 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 2012 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 2013 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2014 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2015 | |
2016 | /* need to recalc the inode CRC if appropriate */ | |
2017 | xfs_dinode_calc_crc(mp, dip); | |
2018 | ||
1da177e4 LT |
2019 | xfs_trans_inode_buf(tp, ibp); |
2020 | xfs_trans_log_buf(tp, ibp, offset, | |
2021 | (offset + sizeof(xfs_agino_t) - 1)); | |
2022 | xfs_inobp_check(mp, ibp); | |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * Point the bucket head pointer at the inode being inserted. | |
2027 | */ | |
2028 | ASSERT(agino != 0); | |
16259e7d | 2029 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
2030 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2031 | (sizeof(xfs_agino_t) * bucket_index); | |
2032 | xfs_trans_log_buf(tp, agibp, offset, | |
2033 | (offset + sizeof(xfs_agino_t) - 1)); | |
2034 | return 0; | |
2035 | } | |
2036 | ||
2037 | /* | |
2038 | * Pull the on-disk inode from the AGI unlinked list. | |
2039 | */ | |
2040 | STATIC int | |
2041 | xfs_iunlink_remove( | |
2042 | xfs_trans_t *tp, | |
2043 | xfs_inode_t *ip) | |
2044 | { | |
2045 | xfs_ino_t next_ino; | |
2046 | xfs_mount_t *mp; | |
2047 | xfs_agi_t *agi; | |
2048 | xfs_dinode_t *dip; | |
2049 | xfs_buf_t *agibp; | |
2050 | xfs_buf_t *ibp; | |
2051 | xfs_agnumber_t agno; | |
1da177e4 LT |
2052 | xfs_agino_t agino; |
2053 | xfs_agino_t next_agino; | |
2054 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 2055 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 2056 | short bucket_index; |
6fdf8ccc | 2057 | int offset, last_offset = 0; |
1da177e4 | 2058 | int error; |
1da177e4 | 2059 | |
1da177e4 | 2060 | mp = tp->t_mountp; |
1da177e4 | 2061 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
2062 | |
2063 | /* | |
2064 | * Get the agi buffer first. It ensures lock ordering | |
2065 | * on the list. | |
2066 | */ | |
5e1be0fb CH |
2067 | error = xfs_read_agi(mp, tp, agno, &agibp); |
2068 | if (error) | |
1da177e4 | 2069 | return error; |
5e1be0fb | 2070 | |
1da177e4 | 2071 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 2072 | |
1da177e4 LT |
2073 | /* |
2074 | * Get the index into the agi hash table for the | |
2075 | * list this inode will go on. | |
2076 | */ | |
2077 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
d2e73665 DW |
2078 | if (!xfs_verify_agino(mp, agno, agino)) |
2079 | return -EFSCORRUPTED; | |
1da177e4 | 2080 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; |
d2e73665 DW |
2081 | if (!xfs_verify_agino(mp, agno, |
2082 | be32_to_cpu(agi->agi_unlinked[bucket_index]))) { | |
2083 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, | |
2084 | agi, sizeof(*agi)); | |
2085 | return -EFSCORRUPTED; | |
2086 | } | |
1da177e4 | 2087 | |
16259e7d | 2088 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 2089 | /* |
475ee413 CH |
2090 | * We're at the head of the list. Get the inode's on-disk |
2091 | * buffer to see if there is anyone after us on the list. | |
2092 | * Only modify our next pointer if it is not already NULLAGINO. | |
2093 | * This saves us the overhead of dealing with the buffer when | |
2094 | * there is no need to change it. | |
1da177e4 | 2095 | */ |
475ee413 CH |
2096 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2097 | 0, 0); | |
1da177e4 | 2098 | if (error) { |
475ee413 | 2099 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2100 | __func__, error); |
1da177e4 LT |
2101 | return error; |
2102 | } | |
347d1c01 | 2103 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2104 | ASSERT(next_agino != 0); |
2105 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2106 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2107 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2108 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2109 | |
2110 | /* need to recalc the inode CRC if appropriate */ | |
2111 | xfs_dinode_calc_crc(mp, dip); | |
2112 | ||
1da177e4 LT |
2113 | xfs_trans_inode_buf(tp, ibp); |
2114 | xfs_trans_log_buf(tp, ibp, offset, | |
2115 | (offset + sizeof(xfs_agino_t) - 1)); | |
2116 | xfs_inobp_check(mp, ibp); | |
2117 | } else { | |
2118 | xfs_trans_brelse(tp, ibp); | |
2119 | } | |
2120 | /* | |
2121 | * Point the bucket head pointer at the next inode. | |
2122 | */ | |
2123 | ASSERT(next_agino != 0); | |
2124 | ASSERT(next_agino != agino); | |
16259e7d | 2125 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2126 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2127 | (sizeof(xfs_agino_t) * bucket_index); | |
2128 | xfs_trans_log_buf(tp, agibp, offset, | |
2129 | (offset + sizeof(xfs_agino_t) - 1)); | |
2130 | } else { | |
2131 | /* | |
2132 | * We need to search the list for the inode being freed. | |
2133 | */ | |
16259e7d | 2134 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2135 | last_ibp = NULL; |
2136 | while (next_agino != agino) { | |
129dbc9a CH |
2137 | struct xfs_imap imap; |
2138 | ||
2139 | if (last_ibp) | |
1da177e4 | 2140 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
2141 | |
2142 | imap.im_blkno = 0; | |
1da177e4 | 2143 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
2144 | |
2145 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
2146 | if (error) { | |
2147 | xfs_warn(mp, | |
2148 | "%s: xfs_imap returned error %d.", | |
2149 | __func__, error); | |
2150 | return error; | |
2151 | } | |
2152 | ||
2153 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
2154 | &last_ibp, 0, 0); | |
1da177e4 | 2155 | if (error) { |
0b932ccc | 2156 | xfs_warn(mp, |
129dbc9a | 2157 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2158 | __func__, error); |
1da177e4 LT |
2159 | return error; |
2160 | } | |
129dbc9a CH |
2161 | |
2162 | last_offset = imap.im_boffset; | |
347d1c01 | 2163 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
d2e73665 DW |
2164 | if (!xfs_verify_agino(mp, agno, next_agino)) { |
2165 | XFS_CORRUPTION_ERROR(__func__, | |
2166 | XFS_ERRLEVEL_LOW, mp, | |
2167 | last_dip, sizeof(*last_dip)); | |
2168 | return -EFSCORRUPTED; | |
2169 | } | |
1da177e4 | 2170 | } |
475ee413 | 2171 | |
1da177e4 | 2172 | /* |
475ee413 CH |
2173 | * Now last_ibp points to the buffer previous to us on the |
2174 | * unlinked list. Pull us from the list. | |
1da177e4 | 2175 | */ |
475ee413 CH |
2176 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2177 | 0, 0); | |
1da177e4 | 2178 | if (error) { |
475ee413 | 2179 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 2180 | __func__, error); |
1da177e4 LT |
2181 | return error; |
2182 | } | |
347d1c01 | 2183 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2184 | ASSERT(next_agino != 0); |
2185 | ASSERT(next_agino != agino); | |
2186 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2187 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2188 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2189 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2190 | |
2191 | /* need to recalc the inode CRC if appropriate */ | |
2192 | xfs_dinode_calc_crc(mp, dip); | |
2193 | ||
1da177e4 LT |
2194 | xfs_trans_inode_buf(tp, ibp); |
2195 | xfs_trans_log_buf(tp, ibp, offset, | |
2196 | (offset + sizeof(xfs_agino_t) - 1)); | |
2197 | xfs_inobp_check(mp, ibp); | |
2198 | } else { | |
2199 | xfs_trans_brelse(tp, ibp); | |
2200 | } | |
2201 | /* | |
2202 | * Point the previous inode on the list to the next inode. | |
2203 | */ | |
347d1c01 | 2204 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
2205 | ASSERT(next_agino != 0); |
2206 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
0a32c26e DC |
2207 | |
2208 | /* need to recalc the inode CRC if appropriate */ | |
2209 | xfs_dinode_calc_crc(mp, last_dip); | |
2210 | ||
1da177e4 LT |
2211 | xfs_trans_inode_buf(tp, last_ibp); |
2212 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2213 | (offset + sizeof(xfs_agino_t) - 1)); | |
2214 | xfs_inobp_check(mp, last_ibp); | |
2215 | } | |
2216 | return 0; | |
2217 | } | |
2218 | ||
5b3eed75 | 2219 | /* |
0b8182db | 2220 | * A big issue when freeing the inode cluster is that we _cannot_ skip any |
5b3eed75 DC |
2221 | * inodes that are in memory - they all must be marked stale and attached to |
2222 | * the cluster buffer. | |
2223 | */ | |
2a30f36d | 2224 | STATIC int |
1da177e4 | 2225 | xfs_ifree_cluster( |
09b56604 BF |
2226 | xfs_inode_t *free_ip, |
2227 | xfs_trans_t *tp, | |
2228 | struct xfs_icluster *xic) | |
1da177e4 LT |
2229 | { |
2230 | xfs_mount_t *mp = free_ip->i_mount; | |
2231 | int blks_per_cluster; | |
982e939e | 2232 | int inodes_per_cluster; |
1da177e4 | 2233 | int nbufs; |
5b257b4a | 2234 | int i, j; |
3cdaa189 | 2235 | int ioffset; |
1da177e4 LT |
2236 | xfs_daddr_t blkno; |
2237 | xfs_buf_t *bp; | |
5b257b4a | 2238 | xfs_inode_t *ip; |
1da177e4 | 2239 | xfs_inode_log_item_t *iip; |
643c8c05 | 2240 | struct xfs_log_item *lip; |
5017e97d | 2241 | struct xfs_perag *pag; |
09b56604 | 2242 | xfs_ino_t inum; |
1da177e4 | 2243 | |
09b56604 | 2244 | inum = xic->first_ino; |
5017e97d | 2245 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
982e939e JL |
2246 | blks_per_cluster = xfs_icluster_size_fsb(mp); |
2247 | inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; | |
2248 | nbufs = mp->m_ialloc_blks / blks_per_cluster; | |
1da177e4 | 2249 | |
982e939e | 2250 | for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { |
09b56604 BF |
2251 | /* |
2252 | * The allocation bitmap tells us which inodes of the chunk were | |
2253 | * physically allocated. Skip the cluster if an inode falls into | |
2254 | * a sparse region. | |
2255 | */ | |
3cdaa189 BF |
2256 | ioffset = inum - xic->first_ino; |
2257 | if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { | |
0703a8e1 | 2258 | ASSERT(ioffset % inodes_per_cluster == 0); |
09b56604 BF |
2259 | continue; |
2260 | } | |
2261 | ||
1da177e4 LT |
2262 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), |
2263 | XFS_INO_TO_AGBNO(mp, inum)); | |
2264 | ||
5b257b4a DC |
2265 | /* |
2266 | * We obtain and lock the backing buffer first in the process | |
2267 | * here, as we have to ensure that any dirty inode that we | |
2268 | * can't get the flush lock on is attached to the buffer. | |
2269 | * If we scan the in-memory inodes first, then buffer IO can | |
2270 | * complete before we get a lock on it, and hence we may fail | |
2271 | * to mark all the active inodes on the buffer stale. | |
2272 | */ | |
2273 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
2274 | mp->m_bsize * blks_per_cluster, |
2275 | XBF_UNMAPPED); | |
5b257b4a | 2276 | |
2a30f36d | 2277 | if (!bp) |
2451337d | 2278 | return -ENOMEM; |
b0f539de DC |
2279 | |
2280 | /* | |
2281 | * This buffer may not have been correctly initialised as we | |
2282 | * didn't read it from disk. That's not important because we are | |
2283 | * only using to mark the buffer as stale in the log, and to | |
2284 | * attach stale cached inodes on it. That means it will never be | |
2285 | * dispatched for IO. If it is, we want to know about it, and we | |
2286 | * want it to fail. We can acheive this by adding a write | |
2287 | * verifier to the buffer. | |
2288 | */ | |
1813dd64 | 2289 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 2290 | |
5b257b4a DC |
2291 | /* |
2292 | * Walk the inodes already attached to the buffer and mark them | |
2293 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
2294 | * in-memory inode walk can't lock them. By marking them all |
2295 | * stale first, we will not attempt to lock them in the loop | |
2296 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 2297 | */ |
643c8c05 | 2298 | list_for_each_entry(lip, &bp->b_li_list, li_bio_list) { |
5b257b4a DC |
2299 | if (lip->li_type == XFS_LI_INODE) { |
2300 | iip = (xfs_inode_log_item_t *)lip; | |
2301 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 2302 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
2303 | xfs_trans_ail_copy_lsn(mp->m_ail, |
2304 | &iip->ili_flush_lsn, | |
2305 | &iip->ili_item.li_lsn); | |
2306 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a | 2307 | } |
5b257b4a | 2308 | } |
1da177e4 | 2309 | |
5b3eed75 | 2310 | |
1da177e4 | 2311 | /* |
5b257b4a DC |
2312 | * For each inode in memory attempt to add it to the inode |
2313 | * buffer and set it up for being staled on buffer IO | |
2314 | * completion. This is safe as we've locked out tail pushing | |
2315 | * and flushing by locking the buffer. | |
1da177e4 | 2316 | * |
5b257b4a DC |
2317 | * We have already marked every inode that was part of a |
2318 | * transaction stale above, which means there is no point in | |
2319 | * even trying to lock them. | |
1da177e4 | 2320 | */ |
982e939e | 2321 | for (i = 0; i < inodes_per_cluster; i++) { |
5b3eed75 | 2322 | retry: |
1a3e8f3d | 2323 | rcu_read_lock(); |
da353b0d DC |
2324 | ip = radix_tree_lookup(&pag->pag_ici_root, |
2325 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 2326 | |
1a3e8f3d DC |
2327 | /* Inode not in memory, nothing to do */ |
2328 | if (!ip) { | |
2329 | rcu_read_unlock(); | |
1da177e4 LT |
2330 | continue; |
2331 | } | |
2332 | ||
1a3e8f3d DC |
2333 | /* |
2334 | * because this is an RCU protected lookup, we could | |
2335 | * find a recently freed or even reallocated inode | |
2336 | * during the lookup. We need to check under the | |
2337 | * i_flags_lock for a valid inode here. Skip it if it | |
2338 | * is not valid, the wrong inode or stale. | |
2339 | */ | |
2340 | spin_lock(&ip->i_flags_lock); | |
2341 | if (ip->i_ino != inum + i || | |
2342 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
2343 | spin_unlock(&ip->i_flags_lock); | |
2344 | rcu_read_unlock(); | |
2345 | continue; | |
2346 | } | |
2347 | spin_unlock(&ip->i_flags_lock); | |
2348 | ||
5b3eed75 DC |
2349 | /* |
2350 | * Don't try to lock/unlock the current inode, but we | |
2351 | * _cannot_ skip the other inodes that we did not find | |
2352 | * in the list attached to the buffer and are not | |
2353 | * already marked stale. If we can't lock it, back off | |
2354 | * and retry. | |
2355 | */ | |
f2e9ad21 OS |
2356 | if (ip != free_ip) { |
2357 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2358 | rcu_read_unlock(); | |
2359 | delay(1); | |
2360 | goto retry; | |
2361 | } | |
2362 | ||
2363 | /* | |
2364 | * Check the inode number again in case we're | |
2365 | * racing with freeing in xfs_reclaim_inode(). | |
2366 | * See the comments in that function for more | |
2367 | * information as to why the initial check is | |
2368 | * not sufficient. | |
2369 | */ | |
2370 | if (ip->i_ino != inum + i) { | |
2371 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
962cc1ad | 2372 | rcu_read_unlock(); |
f2e9ad21 OS |
2373 | continue; |
2374 | } | |
1da177e4 | 2375 | } |
1a3e8f3d | 2376 | rcu_read_unlock(); |
1da177e4 | 2377 | |
5b3eed75 | 2378 | xfs_iflock(ip); |
5b257b4a | 2379 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 2380 | |
5b3eed75 DC |
2381 | /* |
2382 | * we don't need to attach clean inodes or those only | |
2383 | * with unlogged changes (which we throw away, anyway). | |
2384 | */ | |
1da177e4 | 2385 | iip = ip->i_itemp; |
5b3eed75 | 2386 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 2387 | ASSERT(ip != free_ip); |
1da177e4 LT |
2388 | xfs_ifunlock(ip); |
2389 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2390 | continue; | |
2391 | } | |
2392 | ||
f5d8d5c4 CH |
2393 | iip->ili_last_fields = iip->ili_fields; |
2394 | iip->ili_fields = 0; | |
fc0561ce | 2395 | iip->ili_fsync_fields = 0; |
1da177e4 | 2396 | iip->ili_logged = 1; |
7b2e2a31 DC |
2397 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2398 | &iip->ili_item.li_lsn); | |
1da177e4 | 2399 | |
ca30b2a7 CH |
2400 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
2401 | &iip->ili_item); | |
5b257b4a DC |
2402 | |
2403 | if (ip != free_ip) | |
1da177e4 | 2404 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
2405 | } |
2406 | ||
5b3eed75 | 2407 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
2408 | xfs_trans_binval(tp, bp); |
2409 | } | |
2410 | ||
5017e97d | 2411 | xfs_perag_put(pag); |
2a30f36d | 2412 | return 0; |
1da177e4 LT |
2413 | } |
2414 | ||
98c4f78d DW |
2415 | /* |
2416 | * Free any local-format buffers sitting around before we reset to | |
2417 | * extents format. | |
2418 | */ | |
2419 | static inline void | |
2420 | xfs_ifree_local_data( | |
2421 | struct xfs_inode *ip, | |
2422 | int whichfork) | |
2423 | { | |
2424 | struct xfs_ifork *ifp; | |
2425 | ||
2426 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL) | |
2427 | return; | |
2428 | ||
2429 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2430 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); | |
2431 | } | |
2432 | ||
1da177e4 LT |
2433 | /* |
2434 | * This is called to return an inode to the inode free list. | |
2435 | * The inode should already be truncated to 0 length and have | |
2436 | * no pages associated with it. This routine also assumes that | |
2437 | * the inode is already a part of the transaction. | |
2438 | * | |
2439 | * The on-disk copy of the inode will have been added to the list | |
2440 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2441 | * that list atomically with respect to freeing it here. | |
2442 | */ | |
2443 | int | |
2444 | xfs_ifree( | |
0e0417f3 BF |
2445 | struct xfs_trans *tp, |
2446 | struct xfs_inode *ip) | |
1da177e4 LT |
2447 | { |
2448 | int error; | |
09b56604 | 2449 | struct xfs_icluster xic = { 0 }; |
1da177e4 | 2450 | |
579aa9ca | 2451 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
54d7b5c1 | 2452 | ASSERT(VFS_I(ip)->i_nlink == 0); |
1da177e4 LT |
2453 | ASSERT(ip->i_d.di_nextents == 0); |
2454 | ASSERT(ip->i_d.di_anextents == 0); | |
c19b3b05 | 2455 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); |
1da177e4 LT |
2456 | ASSERT(ip->i_d.di_nblocks == 0); |
2457 | ||
2458 | /* | |
2459 | * Pull the on-disk inode from the AGI unlinked list. | |
2460 | */ | |
2461 | error = xfs_iunlink_remove(tp, ip); | |
1baaed8f | 2462 | if (error) |
1da177e4 | 2463 | return error; |
1da177e4 | 2464 | |
0e0417f3 | 2465 | error = xfs_difree(tp, ip->i_ino, &xic); |
1baaed8f | 2466 | if (error) |
1da177e4 | 2467 | return error; |
1baaed8f | 2468 | |
98c4f78d DW |
2469 | xfs_ifree_local_data(ip, XFS_DATA_FORK); |
2470 | xfs_ifree_local_data(ip, XFS_ATTR_FORK); | |
2471 | ||
c19b3b05 | 2472 | VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ |
1da177e4 | 2473 | ip->i_d.di_flags = 0; |
beaae8cd | 2474 | ip->i_d.di_flags2 = 0; |
1da177e4 LT |
2475 | ip->i_d.di_dmevmask = 0; |
2476 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
2477 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
2478 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
dc1baa71 ES |
2479 | |
2480 | /* Don't attempt to replay owner changes for a deleted inode */ | |
2481 | ip->i_itemp->ili_fields &= ~(XFS_ILOG_AOWNER|XFS_ILOG_DOWNER); | |
2482 | ||
1da177e4 LT |
2483 | /* |
2484 | * Bump the generation count so no one will be confused | |
2485 | * by reincarnations of this inode. | |
2486 | */ | |
9e9a2674 | 2487 | VFS_I(ip)->i_generation++; |
1da177e4 LT |
2488 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
2489 | ||
09b56604 BF |
2490 | if (xic.deleted) |
2491 | error = xfs_ifree_cluster(ip, tp, &xic); | |
1da177e4 | 2492 | |
2a30f36d | 2493 | return error; |
1da177e4 LT |
2494 | } |
2495 | ||
1da177e4 | 2496 | /* |
60ec6783 CH |
2497 | * This is called to unpin an inode. The caller must have the inode locked |
2498 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2499 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2500 | */ |
60ec6783 | 2501 | static void |
f392e631 | 2502 | xfs_iunpin( |
60ec6783 | 2503 | struct xfs_inode *ip) |
1da177e4 | 2504 | { |
579aa9ca | 2505 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2506 | |
4aaf15d1 DC |
2507 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2508 | ||
a3f74ffb | 2509 | /* Give the log a push to start the unpinning I/O */ |
656de4ff | 2510 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL); |
a14a348b | 2511 | |
a3f74ffb | 2512 | } |
1da177e4 | 2513 | |
f392e631 CH |
2514 | static void |
2515 | __xfs_iunpin_wait( | |
2516 | struct xfs_inode *ip) | |
2517 | { | |
2518 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2519 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2520 | ||
2521 | xfs_iunpin(ip); | |
2522 | ||
2523 | do { | |
21417136 | 2524 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
f392e631 CH |
2525 | if (xfs_ipincount(ip)) |
2526 | io_schedule(); | |
2527 | } while (xfs_ipincount(ip)); | |
21417136 | 2528 | finish_wait(wq, &wait.wq_entry); |
f392e631 CH |
2529 | } |
2530 | ||
777df5af | 2531 | void |
a3f74ffb | 2532 | xfs_iunpin_wait( |
60ec6783 | 2533 | struct xfs_inode *ip) |
a3f74ffb | 2534 | { |
f392e631 CH |
2535 | if (xfs_ipincount(ip)) |
2536 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2537 | } |
2538 | ||
27320369 DC |
2539 | /* |
2540 | * Removing an inode from the namespace involves removing the directory entry | |
2541 | * and dropping the link count on the inode. Removing the directory entry can | |
2542 | * result in locking an AGF (directory blocks were freed) and removing a link | |
2543 | * count can result in placing the inode on an unlinked list which results in | |
2544 | * locking an AGI. | |
2545 | * | |
2546 | * The big problem here is that we have an ordering constraint on AGF and AGI | |
2547 | * locking - inode allocation locks the AGI, then can allocate a new extent for | |
2548 | * new inodes, locking the AGF after the AGI. Similarly, freeing the inode | |
2549 | * removes the inode from the unlinked list, requiring that we lock the AGI | |
2550 | * first, and then freeing the inode can result in an inode chunk being freed | |
2551 | * and hence freeing disk space requiring that we lock an AGF. | |
2552 | * | |
2553 | * Hence the ordering that is imposed by other parts of the code is AGI before | |
2554 | * AGF. This means we cannot remove the directory entry before we drop the inode | |
2555 | * reference count and put it on the unlinked list as this results in a lock | |
2556 | * order of AGF then AGI, and this can deadlock against inode allocation and | |
2557 | * freeing. Therefore we must drop the link counts before we remove the | |
2558 | * directory entry. | |
2559 | * | |
2560 | * This is still safe from a transactional point of view - it is not until we | |
310a75a3 | 2561 | * get to xfs_defer_finish() that we have the possibility of multiple |
27320369 DC |
2562 | * transactions in this operation. Hence as long as we remove the directory |
2563 | * entry and drop the link count in the first transaction of the remove | |
2564 | * operation, there are no transactional constraints on the ordering here. | |
2565 | */ | |
c24b5dfa DC |
2566 | int |
2567 | xfs_remove( | |
2568 | xfs_inode_t *dp, | |
2569 | struct xfs_name *name, | |
2570 | xfs_inode_t *ip) | |
2571 | { | |
2572 | xfs_mount_t *mp = dp->i_mount; | |
2573 | xfs_trans_t *tp = NULL; | |
c19b3b05 | 2574 | int is_dir = S_ISDIR(VFS_I(ip)->i_mode); |
c24b5dfa | 2575 | int error = 0; |
2c3234d1 | 2576 | struct xfs_defer_ops dfops; |
c24b5dfa | 2577 | uint resblks; |
c24b5dfa DC |
2578 | |
2579 | trace_xfs_remove(dp, name); | |
2580 | ||
2581 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 2582 | return -EIO; |
c24b5dfa | 2583 | |
c14cfcca | 2584 | error = xfs_qm_dqattach(dp); |
c24b5dfa DC |
2585 | if (error) |
2586 | goto std_return; | |
2587 | ||
c14cfcca | 2588 | error = xfs_qm_dqattach(ip); |
c24b5dfa DC |
2589 | if (error) |
2590 | goto std_return; | |
2591 | ||
c24b5dfa DC |
2592 | /* |
2593 | * We try to get the real space reservation first, | |
2594 | * allowing for directory btree deletion(s) implying | |
2595 | * possible bmap insert(s). If we can't get the space | |
2596 | * reservation then we use 0 instead, and avoid the bmap | |
2597 | * btree insert(s) in the directory code by, if the bmap | |
2598 | * insert tries to happen, instead trimming the LAST | |
2599 | * block from the directory. | |
2600 | */ | |
2601 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
253f4911 | 2602 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); |
2451337d | 2603 | if (error == -ENOSPC) { |
c24b5dfa | 2604 | resblks = 0; |
253f4911 CH |
2605 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, |
2606 | &tp); | |
c24b5dfa DC |
2607 | } |
2608 | if (error) { | |
2451337d | 2609 | ASSERT(error != -ENOSPC); |
253f4911 | 2610 | goto std_return; |
c24b5dfa DC |
2611 | } |
2612 | ||
7c2d238a | 2613 | xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL); |
c24b5dfa | 2614 | |
65523218 | 2615 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
2616 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
2617 | ||
2618 | /* | |
2619 | * If we're removing a directory perform some additional validation. | |
2620 | */ | |
2621 | if (is_dir) { | |
54d7b5c1 DC |
2622 | ASSERT(VFS_I(ip)->i_nlink >= 2); |
2623 | if (VFS_I(ip)->i_nlink != 2) { | |
2451337d | 2624 | error = -ENOTEMPTY; |
c24b5dfa DC |
2625 | goto out_trans_cancel; |
2626 | } | |
2627 | if (!xfs_dir_isempty(ip)) { | |
2451337d | 2628 | error = -ENOTEMPTY; |
c24b5dfa DC |
2629 | goto out_trans_cancel; |
2630 | } | |
c24b5dfa | 2631 | |
27320369 | 2632 | /* Drop the link from ip's "..". */ |
c24b5dfa DC |
2633 | error = xfs_droplink(tp, dp); |
2634 | if (error) | |
27320369 | 2635 | goto out_trans_cancel; |
c24b5dfa | 2636 | |
27320369 | 2637 | /* Drop the "." link from ip to self. */ |
c24b5dfa DC |
2638 | error = xfs_droplink(tp, ip); |
2639 | if (error) | |
27320369 | 2640 | goto out_trans_cancel; |
c24b5dfa DC |
2641 | } else { |
2642 | /* | |
2643 | * When removing a non-directory we need to log the parent | |
2644 | * inode here. For a directory this is done implicitly | |
2645 | * by the xfs_droplink call for the ".." entry. | |
2646 | */ | |
2647 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2648 | } | |
27320369 | 2649 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
c24b5dfa | 2650 | |
27320369 | 2651 | /* Drop the link from dp to ip. */ |
c24b5dfa DC |
2652 | error = xfs_droplink(tp, ip); |
2653 | if (error) | |
27320369 | 2654 | goto out_trans_cancel; |
c24b5dfa | 2655 | |
f16dea54 | 2656 | xfs_defer_init(tp, &dfops, &tp->t_firstblock); |
381eee69 | 2657 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks); |
27320369 | 2658 | if (error) { |
2451337d | 2659 | ASSERT(error != -ENOENT); |
27320369 DC |
2660 | goto out_bmap_cancel; |
2661 | } | |
2662 | ||
c24b5dfa DC |
2663 | /* |
2664 | * If this is a synchronous mount, make sure that the | |
2665 | * remove transaction goes to disk before returning to | |
2666 | * the user. | |
2667 | */ | |
2668 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2669 | xfs_trans_set_sync(tp); | |
2670 | ||
8ad7c629 | 2671 | error = xfs_defer_finish(&tp, &dfops); |
c24b5dfa DC |
2672 | if (error) |
2673 | goto out_bmap_cancel; | |
2674 | ||
70393313 | 2675 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
2676 | if (error) |
2677 | goto std_return; | |
2678 | ||
2cd2ef6a | 2679 | if (is_dir && xfs_inode_is_filestream(ip)) |
c24b5dfa DC |
2680 | xfs_filestream_deassociate(ip); |
2681 | ||
2682 | return 0; | |
2683 | ||
2684 | out_bmap_cancel: | |
2c3234d1 | 2685 | xfs_defer_cancel(&dfops); |
c24b5dfa | 2686 | out_trans_cancel: |
4906e215 | 2687 | xfs_trans_cancel(tp); |
c24b5dfa DC |
2688 | std_return: |
2689 | return error; | |
2690 | } | |
2691 | ||
f6bba201 DC |
2692 | /* |
2693 | * Enter all inodes for a rename transaction into a sorted array. | |
2694 | */ | |
95afcf5c | 2695 | #define __XFS_SORT_INODES 5 |
f6bba201 DC |
2696 | STATIC void |
2697 | xfs_sort_for_rename( | |
95afcf5c DC |
2698 | struct xfs_inode *dp1, /* in: old (source) directory inode */ |
2699 | struct xfs_inode *dp2, /* in: new (target) directory inode */ | |
2700 | struct xfs_inode *ip1, /* in: inode of old entry */ | |
2701 | struct xfs_inode *ip2, /* in: inode of new entry */ | |
2702 | struct xfs_inode *wip, /* in: whiteout inode */ | |
2703 | struct xfs_inode **i_tab,/* out: sorted array of inodes */ | |
2704 | int *num_inodes) /* in/out: inodes in array */ | |
f6bba201 | 2705 | { |
f6bba201 DC |
2706 | int i, j; |
2707 | ||
95afcf5c DC |
2708 | ASSERT(*num_inodes == __XFS_SORT_INODES); |
2709 | memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); | |
2710 | ||
f6bba201 DC |
2711 | /* |
2712 | * i_tab contains a list of pointers to inodes. We initialize | |
2713 | * the table here & we'll sort it. We will then use it to | |
2714 | * order the acquisition of the inode locks. | |
2715 | * | |
2716 | * Note that the table may contain duplicates. e.g., dp1 == dp2. | |
2717 | */ | |
95afcf5c DC |
2718 | i = 0; |
2719 | i_tab[i++] = dp1; | |
2720 | i_tab[i++] = dp2; | |
2721 | i_tab[i++] = ip1; | |
2722 | if (ip2) | |
2723 | i_tab[i++] = ip2; | |
2724 | if (wip) | |
2725 | i_tab[i++] = wip; | |
2726 | *num_inodes = i; | |
f6bba201 DC |
2727 | |
2728 | /* | |
2729 | * Sort the elements via bubble sort. (Remember, there are at | |
95afcf5c | 2730 | * most 5 elements to sort, so this is adequate.) |
f6bba201 DC |
2731 | */ |
2732 | for (i = 0; i < *num_inodes; i++) { | |
2733 | for (j = 1; j < *num_inodes; j++) { | |
2734 | if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { | |
95afcf5c | 2735 | struct xfs_inode *temp = i_tab[j]; |
f6bba201 DC |
2736 | i_tab[j] = i_tab[j-1]; |
2737 | i_tab[j-1] = temp; | |
2738 | } | |
2739 | } | |
2740 | } | |
2741 | } | |
2742 | ||
310606b0 DC |
2743 | static int |
2744 | xfs_finish_rename( | |
c9cfdb38 | 2745 | struct xfs_trans *tp) |
310606b0 | 2746 | { |
c9cfdb38 | 2747 | struct xfs_defer_ops *dfops = tp->t_dfops; |
310606b0 DC |
2748 | int error; |
2749 | ||
2750 | /* | |
2751 | * If this is a synchronous mount, make sure that the rename transaction | |
2752 | * goes to disk before returning to the user. | |
2753 | */ | |
2754 | if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2755 | xfs_trans_set_sync(tp); | |
2756 | ||
8ad7c629 | 2757 | error = xfs_defer_finish(&tp, dfops); |
310606b0 | 2758 | if (error) { |
2c3234d1 | 2759 | xfs_defer_cancel(dfops); |
4906e215 | 2760 | xfs_trans_cancel(tp); |
310606b0 DC |
2761 | return error; |
2762 | } | |
2763 | ||
70393313 | 2764 | return xfs_trans_commit(tp); |
310606b0 DC |
2765 | } |
2766 | ||
d31a1825 CM |
2767 | /* |
2768 | * xfs_cross_rename() | |
2769 | * | |
2770 | * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall | |
2771 | */ | |
2772 | STATIC int | |
2773 | xfs_cross_rename( | |
2774 | struct xfs_trans *tp, | |
2775 | struct xfs_inode *dp1, | |
2776 | struct xfs_name *name1, | |
2777 | struct xfs_inode *ip1, | |
2778 | struct xfs_inode *dp2, | |
2779 | struct xfs_name *name2, | |
2780 | struct xfs_inode *ip2, | |
d31a1825 CM |
2781 | int spaceres) |
2782 | { | |
2783 | int error = 0; | |
2784 | int ip1_flags = 0; | |
2785 | int ip2_flags = 0; | |
2786 | int dp2_flags = 0; | |
2787 | ||
2788 | /* Swap inode number for dirent in first parent */ | |
381eee69 | 2789 | error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres); |
d31a1825 | 2790 | if (error) |
eeacd321 | 2791 | goto out_trans_abort; |
d31a1825 CM |
2792 | |
2793 | /* Swap inode number for dirent in second parent */ | |
381eee69 | 2794 | error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres); |
d31a1825 | 2795 | if (error) |
eeacd321 | 2796 | goto out_trans_abort; |
d31a1825 CM |
2797 | |
2798 | /* | |
2799 | * If we're renaming one or more directories across different parents, | |
2800 | * update the respective ".." entries (and link counts) to match the new | |
2801 | * parents. | |
2802 | */ | |
2803 | if (dp1 != dp2) { | |
2804 | dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2805 | ||
c19b3b05 | 2806 | if (S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 | 2807 | error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, |
381eee69 | 2808 | dp1->i_ino, spaceres); |
d31a1825 | 2809 | if (error) |
eeacd321 | 2810 | goto out_trans_abort; |
d31a1825 CM |
2811 | |
2812 | /* transfer ip2 ".." reference to dp1 */ | |
c19b3b05 | 2813 | if (!S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 CM |
2814 | error = xfs_droplink(tp, dp2); |
2815 | if (error) | |
eeacd321 | 2816 | goto out_trans_abort; |
d31a1825 CM |
2817 | error = xfs_bumplink(tp, dp1); |
2818 | if (error) | |
eeacd321 | 2819 | goto out_trans_abort; |
d31a1825 CM |
2820 | } |
2821 | ||
2822 | /* | |
2823 | * Although ip1 isn't changed here, userspace needs | |
2824 | * to be warned about the change, so that applications | |
2825 | * relying on it (like backup ones), will properly | |
2826 | * notify the change | |
2827 | */ | |
2828 | ip1_flags |= XFS_ICHGTIME_CHG; | |
2829 | ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2830 | } | |
2831 | ||
c19b3b05 | 2832 | if (S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 | 2833 | error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, |
381eee69 | 2834 | dp2->i_ino, spaceres); |
d31a1825 | 2835 | if (error) |
eeacd321 | 2836 | goto out_trans_abort; |
d31a1825 CM |
2837 | |
2838 | /* transfer ip1 ".." reference to dp2 */ | |
c19b3b05 | 2839 | if (!S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 CM |
2840 | error = xfs_droplink(tp, dp1); |
2841 | if (error) | |
eeacd321 | 2842 | goto out_trans_abort; |
d31a1825 CM |
2843 | error = xfs_bumplink(tp, dp2); |
2844 | if (error) | |
eeacd321 | 2845 | goto out_trans_abort; |
d31a1825 CM |
2846 | } |
2847 | ||
2848 | /* | |
2849 | * Although ip2 isn't changed here, userspace needs | |
2850 | * to be warned about the change, so that applications | |
2851 | * relying on it (like backup ones), will properly | |
2852 | * notify the change | |
2853 | */ | |
2854 | ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2855 | ip2_flags |= XFS_ICHGTIME_CHG; | |
2856 | } | |
2857 | } | |
2858 | ||
2859 | if (ip1_flags) { | |
2860 | xfs_trans_ichgtime(tp, ip1, ip1_flags); | |
2861 | xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); | |
2862 | } | |
2863 | if (ip2_flags) { | |
2864 | xfs_trans_ichgtime(tp, ip2, ip2_flags); | |
2865 | xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); | |
2866 | } | |
2867 | if (dp2_flags) { | |
2868 | xfs_trans_ichgtime(tp, dp2, dp2_flags); | |
2869 | xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); | |
2870 | } | |
2871 | xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2872 | xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); | |
c9cfdb38 | 2873 | return xfs_finish_rename(tp); |
eeacd321 DC |
2874 | |
2875 | out_trans_abort: | |
c9cfdb38 | 2876 | xfs_defer_cancel(tp->t_dfops); |
4906e215 | 2877 | xfs_trans_cancel(tp); |
d31a1825 CM |
2878 | return error; |
2879 | } | |
2880 | ||
7dcf5c3e DC |
2881 | /* |
2882 | * xfs_rename_alloc_whiteout() | |
2883 | * | |
2884 | * Return a referenced, unlinked, unlocked inode that that can be used as a | |
2885 | * whiteout in a rename transaction. We use a tmpfile inode here so that if we | |
2886 | * crash between allocating the inode and linking it into the rename transaction | |
2887 | * recovery will free the inode and we won't leak it. | |
2888 | */ | |
2889 | static int | |
2890 | xfs_rename_alloc_whiteout( | |
2891 | struct xfs_inode *dp, | |
2892 | struct xfs_inode **wip) | |
2893 | { | |
2894 | struct xfs_inode *tmpfile; | |
2895 | int error; | |
2896 | ||
a1f69417 | 2897 | error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile); |
7dcf5c3e DC |
2898 | if (error) |
2899 | return error; | |
2900 | ||
22419ac9 BF |
2901 | /* |
2902 | * Prepare the tmpfile inode as if it were created through the VFS. | |
2903 | * Otherwise, the link increment paths will complain about nlink 0->1. | |
2904 | * Drop the link count as done by d_tmpfile(), complete the inode setup | |
2905 | * and flag it as linkable. | |
2906 | */ | |
2907 | drop_nlink(VFS_I(tmpfile)); | |
2b3d1d41 | 2908 | xfs_setup_iops(tmpfile); |
7dcf5c3e DC |
2909 | xfs_finish_inode_setup(tmpfile); |
2910 | VFS_I(tmpfile)->i_state |= I_LINKABLE; | |
2911 | ||
2912 | *wip = tmpfile; | |
2913 | return 0; | |
2914 | } | |
2915 | ||
f6bba201 DC |
2916 | /* |
2917 | * xfs_rename | |
2918 | */ | |
2919 | int | |
2920 | xfs_rename( | |
7dcf5c3e DC |
2921 | struct xfs_inode *src_dp, |
2922 | struct xfs_name *src_name, | |
2923 | struct xfs_inode *src_ip, | |
2924 | struct xfs_inode *target_dp, | |
2925 | struct xfs_name *target_name, | |
2926 | struct xfs_inode *target_ip, | |
2927 | unsigned int flags) | |
f6bba201 | 2928 | { |
7dcf5c3e DC |
2929 | struct xfs_mount *mp = src_dp->i_mount; |
2930 | struct xfs_trans *tp; | |
2c3234d1 | 2931 | struct xfs_defer_ops dfops; |
7dcf5c3e DC |
2932 | struct xfs_inode *wip = NULL; /* whiteout inode */ |
2933 | struct xfs_inode *inodes[__XFS_SORT_INODES]; | |
2934 | int num_inodes = __XFS_SORT_INODES; | |
2b93681f | 2935 | bool new_parent = (src_dp != target_dp); |
c19b3b05 | 2936 | bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); |
7dcf5c3e DC |
2937 | int spaceres; |
2938 | int error; | |
f6bba201 DC |
2939 | |
2940 | trace_xfs_rename(src_dp, target_dp, src_name, target_name); | |
2941 | ||
eeacd321 DC |
2942 | if ((flags & RENAME_EXCHANGE) && !target_ip) |
2943 | return -EINVAL; | |
2944 | ||
7dcf5c3e DC |
2945 | /* |
2946 | * If we are doing a whiteout operation, allocate the whiteout inode | |
2947 | * we will be placing at the target and ensure the type is set | |
2948 | * appropriately. | |
2949 | */ | |
2950 | if (flags & RENAME_WHITEOUT) { | |
2951 | ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); | |
2952 | error = xfs_rename_alloc_whiteout(target_dp, &wip); | |
2953 | if (error) | |
2954 | return error; | |
2955 | ||
2956 | /* setup target dirent info as whiteout */ | |
2957 | src_name->type = XFS_DIR3_FT_CHRDEV; | |
2958 | } | |
f6bba201 | 2959 | |
7dcf5c3e | 2960 | xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, |
f6bba201 DC |
2961 | inodes, &num_inodes); |
2962 | ||
f6bba201 | 2963 | spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); |
253f4911 | 2964 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); |
2451337d | 2965 | if (error == -ENOSPC) { |
f6bba201 | 2966 | spaceres = 0; |
253f4911 CH |
2967 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, |
2968 | &tp); | |
f6bba201 | 2969 | } |
445883e8 | 2970 | if (error) |
253f4911 | 2971 | goto out_release_wip; |
f6bba201 DC |
2972 | |
2973 | /* | |
2974 | * Attach the dquots to the inodes | |
2975 | */ | |
2976 | error = xfs_qm_vop_rename_dqattach(inodes); | |
445883e8 DC |
2977 | if (error) |
2978 | goto out_trans_cancel; | |
f6bba201 DC |
2979 | |
2980 | /* | |
2981 | * Lock all the participating inodes. Depending upon whether | |
2982 | * the target_name exists in the target directory, and | |
2983 | * whether the target directory is the same as the source | |
2984 | * directory, we can lock from 2 to 4 inodes. | |
2985 | */ | |
2986 | xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); | |
2987 | ||
2988 | /* | |
2989 | * Join all the inodes to the transaction. From this point on, | |
2990 | * we can rely on either trans_commit or trans_cancel to unlock | |
2991 | * them. | |
2992 | */ | |
65523218 | 2993 | xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); |
f6bba201 | 2994 | if (new_parent) |
65523218 | 2995 | xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); |
f6bba201 DC |
2996 | xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); |
2997 | if (target_ip) | |
2998 | xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); | |
7dcf5c3e DC |
2999 | if (wip) |
3000 | xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); | |
f6bba201 DC |
3001 | |
3002 | /* | |
3003 | * If we are using project inheritance, we only allow renames | |
3004 | * into our tree when the project IDs are the same; else the | |
3005 | * tree quota mechanism would be circumvented. | |
3006 | */ | |
3007 | if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
3008 | (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { | |
2451337d | 3009 | error = -EXDEV; |
445883e8 | 3010 | goto out_trans_cancel; |
f6bba201 DC |
3011 | } |
3012 | ||
f16dea54 | 3013 | xfs_defer_init(tp, &dfops, &tp->t_firstblock); |
445883e8 | 3014 | |
eeacd321 DC |
3015 | /* RENAME_EXCHANGE is unique from here on. */ |
3016 | if (flags & RENAME_EXCHANGE) | |
3017 | return xfs_cross_rename(tp, src_dp, src_name, src_ip, | |
3018 | target_dp, target_name, target_ip, | |
f16dea54 | 3019 | spaceres); |
d31a1825 | 3020 | |
f6bba201 DC |
3021 | /* |
3022 | * Set up the target. | |
3023 | */ | |
3024 | if (target_ip == NULL) { | |
3025 | /* | |
3026 | * If there's no space reservation, check the entry will | |
3027 | * fit before actually inserting it. | |
3028 | */ | |
94f3cad5 ES |
3029 | if (!spaceres) { |
3030 | error = xfs_dir_canenter(tp, target_dp, target_name); | |
3031 | if (error) | |
445883e8 | 3032 | goto out_trans_cancel; |
94f3cad5 | 3033 | } |
f6bba201 DC |
3034 | /* |
3035 | * If target does not exist and the rename crosses | |
3036 | * directories, adjust the target directory link count | |
3037 | * to account for the ".." reference from the new entry. | |
3038 | */ | |
3039 | error = xfs_dir_createname(tp, target_dp, target_name, | |
381eee69 | 3040 | src_ip->i_ino, spaceres); |
f6bba201 | 3041 | if (error) |
4906e215 | 3042 | goto out_bmap_cancel; |
f6bba201 DC |
3043 | |
3044 | xfs_trans_ichgtime(tp, target_dp, | |
3045 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3046 | ||
3047 | if (new_parent && src_is_directory) { | |
3048 | error = xfs_bumplink(tp, target_dp); | |
3049 | if (error) | |
4906e215 | 3050 | goto out_bmap_cancel; |
f6bba201 DC |
3051 | } |
3052 | } else { /* target_ip != NULL */ | |
3053 | /* | |
3054 | * If target exists and it's a directory, check that both | |
3055 | * target and source are directories and that target can be | |
3056 | * destroyed, or that neither is a directory. | |
3057 | */ | |
c19b3b05 | 3058 | if (S_ISDIR(VFS_I(target_ip)->i_mode)) { |
f6bba201 DC |
3059 | /* |
3060 | * Make sure target dir is empty. | |
3061 | */ | |
3062 | if (!(xfs_dir_isempty(target_ip)) || | |
54d7b5c1 | 3063 | (VFS_I(target_ip)->i_nlink > 2)) { |
2451337d | 3064 | error = -EEXIST; |
445883e8 | 3065 | goto out_trans_cancel; |
f6bba201 DC |
3066 | } |
3067 | } | |
3068 | ||
3069 | /* | |
3070 | * Link the source inode under the target name. | |
3071 | * If the source inode is a directory and we are moving | |
3072 | * it across directories, its ".." entry will be | |
3073 | * inconsistent until we replace that down below. | |
3074 | * | |
3075 | * In case there is already an entry with the same | |
3076 | * name at the destination directory, remove it first. | |
3077 | */ | |
3078 | error = xfs_dir_replace(tp, target_dp, target_name, | |
381eee69 | 3079 | src_ip->i_ino, spaceres); |
f6bba201 | 3080 | if (error) |
4906e215 | 3081 | goto out_bmap_cancel; |
f6bba201 DC |
3082 | |
3083 | xfs_trans_ichgtime(tp, target_dp, | |
3084 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
3085 | ||
3086 | /* | |
3087 | * Decrement the link count on the target since the target | |
3088 | * dir no longer points to it. | |
3089 | */ | |
3090 | error = xfs_droplink(tp, target_ip); | |
3091 | if (error) | |
4906e215 | 3092 | goto out_bmap_cancel; |
f6bba201 DC |
3093 | |
3094 | if (src_is_directory) { | |
3095 | /* | |
3096 | * Drop the link from the old "." entry. | |
3097 | */ | |
3098 | error = xfs_droplink(tp, target_ip); | |
3099 | if (error) | |
4906e215 | 3100 | goto out_bmap_cancel; |
f6bba201 DC |
3101 | } |
3102 | } /* target_ip != NULL */ | |
3103 | ||
3104 | /* | |
3105 | * Remove the source. | |
3106 | */ | |
3107 | if (new_parent && src_is_directory) { | |
3108 | /* | |
3109 | * Rewrite the ".." entry to point to the new | |
3110 | * directory. | |
3111 | */ | |
3112 | error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, | |
381eee69 | 3113 | target_dp->i_ino, spaceres); |
2451337d | 3114 | ASSERT(error != -EEXIST); |
f6bba201 | 3115 | if (error) |
4906e215 | 3116 | goto out_bmap_cancel; |
f6bba201 DC |
3117 | } |
3118 | ||
3119 | /* | |
3120 | * We always want to hit the ctime on the source inode. | |
3121 | * | |
3122 | * This isn't strictly required by the standards since the source | |
3123 | * inode isn't really being changed, but old unix file systems did | |
3124 | * it and some incremental backup programs won't work without it. | |
3125 | */ | |
3126 | xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); | |
3127 | xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); | |
3128 | ||
3129 | /* | |
3130 | * Adjust the link count on src_dp. This is necessary when | |
3131 | * renaming a directory, either within one parent when | |
3132 | * the target existed, or across two parent directories. | |
3133 | */ | |
3134 | if (src_is_directory && (new_parent || target_ip != NULL)) { | |
3135 | ||
3136 | /* | |
3137 | * Decrement link count on src_directory since the | |
3138 | * entry that's moved no longer points to it. | |
3139 | */ | |
3140 | error = xfs_droplink(tp, src_dp); | |
3141 | if (error) | |
4906e215 | 3142 | goto out_bmap_cancel; |
f6bba201 DC |
3143 | } |
3144 | ||
7dcf5c3e DC |
3145 | /* |
3146 | * For whiteouts, we only need to update the source dirent with the | |
3147 | * inode number of the whiteout inode rather than removing it | |
3148 | * altogether. | |
3149 | */ | |
3150 | if (wip) { | |
3151 | error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, | |
381eee69 | 3152 | spaceres); |
7dcf5c3e DC |
3153 | } else |
3154 | error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, | |
381eee69 | 3155 | spaceres); |
f6bba201 | 3156 | if (error) |
4906e215 | 3157 | goto out_bmap_cancel; |
f6bba201 DC |
3158 | |
3159 | /* | |
7dcf5c3e DC |
3160 | * For whiteouts, we need to bump the link count on the whiteout inode. |
3161 | * This means that failures all the way up to this point leave the inode | |
3162 | * on the unlinked list and so cleanup is a simple matter of dropping | |
3163 | * the remaining reference to it. If we fail here after bumping the link | |
3164 | * count, we're shutting down the filesystem so we'll never see the | |
3165 | * intermediate state on disk. | |
f6bba201 | 3166 | */ |
7dcf5c3e | 3167 | if (wip) { |
54d7b5c1 | 3168 | ASSERT(VFS_I(wip)->i_nlink == 0); |
7dcf5c3e DC |
3169 | error = xfs_bumplink(tp, wip); |
3170 | if (error) | |
4906e215 | 3171 | goto out_bmap_cancel; |
7dcf5c3e DC |
3172 | error = xfs_iunlink_remove(tp, wip); |
3173 | if (error) | |
4906e215 | 3174 | goto out_bmap_cancel; |
7dcf5c3e | 3175 | xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); |
f6bba201 | 3176 | |
7dcf5c3e DC |
3177 | /* |
3178 | * Now we have a real link, clear the "I'm a tmpfile" state | |
3179 | * flag from the inode so it doesn't accidentally get misused in | |
3180 | * future. | |
3181 | */ | |
3182 | VFS_I(wip)->i_state &= ~I_LINKABLE; | |
f6bba201 DC |
3183 | } |
3184 | ||
f6bba201 DC |
3185 | xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3186 | xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); | |
3187 | if (new_parent) | |
3188 | xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); | |
f6bba201 | 3189 | |
c9cfdb38 | 3190 | error = xfs_finish_rename(tp); |
7dcf5c3e DC |
3191 | if (wip) |
3192 | IRELE(wip); | |
3193 | return error; | |
f6bba201 | 3194 | |
445883e8 | 3195 | out_bmap_cancel: |
2c3234d1 | 3196 | xfs_defer_cancel(&dfops); |
445883e8 | 3197 | out_trans_cancel: |
4906e215 | 3198 | xfs_trans_cancel(tp); |
253f4911 | 3199 | out_release_wip: |
7dcf5c3e DC |
3200 | if (wip) |
3201 | IRELE(wip); | |
f6bba201 DC |
3202 | return error; |
3203 | } | |
3204 | ||
5c4d97d0 DC |
3205 | STATIC int |
3206 | xfs_iflush_cluster( | |
19429363 DC |
3207 | struct xfs_inode *ip, |
3208 | struct xfs_buf *bp) | |
1da177e4 | 3209 | { |
19429363 | 3210 | struct xfs_mount *mp = ip->i_mount; |
5c4d97d0 DC |
3211 | struct xfs_perag *pag; |
3212 | unsigned long first_index, mask; | |
3213 | unsigned long inodes_per_cluster; | |
19429363 DC |
3214 | int cilist_size; |
3215 | struct xfs_inode **cilist; | |
3216 | struct xfs_inode *cip; | |
5c4d97d0 DC |
3217 | int nr_found; |
3218 | int clcount = 0; | |
1da177e4 | 3219 | int i; |
1da177e4 | 3220 | |
5c4d97d0 | 3221 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1da177e4 | 3222 | |
0f49efd8 | 3223 | inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; |
19429363 DC |
3224 | cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); |
3225 | cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); | |
3226 | if (!cilist) | |
5c4d97d0 | 3227 | goto out_put; |
1da177e4 | 3228 | |
0f49efd8 | 3229 | mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); |
5c4d97d0 DC |
3230 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; |
3231 | rcu_read_lock(); | |
3232 | /* really need a gang lookup range call here */ | |
19429363 | 3233 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, |
5c4d97d0 DC |
3234 | first_index, inodes_per_cluster); |
3235 | if (nr_found == 0) | |
3236 | goto out_free; | |
3237 | ||
3238 | for (i = 0; i < nr_found; i++) { | |
19429363 DC |
3239 | cip = cilist[i]; |
3240 | if (cip == ip) | |
bad55843 | 3241 | continue; |
1a3e8f3d DC |
3242 | |
3243 | /* | |
3244 | * because this is an RCU protected lookup, we could find a | |
3245 | * recently freed or even reallocated inode during the lookup. | |
3246 | * We need to check under the i_flags_lock for a valid inode | |
3247 | * here. Skip it if it is not valid or the wrong inode. | |
3248 | */ | |
19429363 DC |
3249 | spin_lock(&cip->i_flags_lock); |
3250 | if (!cip->i_ino || | |
3251 | __xfs_iflags_test(cip, XFS_ISTALE)) { | |
3252 | spin_unlock(&cip->i_flags_lock); | |
1a3e8f3d DC |
3253 | continue; |
3254 | } | |
5a90e53e DC |
3255 | |
3256 | /* | |
3257 | * Once we fall off the end of the cluster, no point checking | |
3258 | * any more inodes in the list because they will also all be | |
3259 | * outside the cluster. | |
3260 | */ | |
19429363 DC |
3261 | if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { |
3262 | spin_unlock(&cip->i_flags_lock); | |
5a90e53e DC |
3263 | break; |
3264 | } | |
19429363 | 3265 | spin_unlock(&cip->i_flags_lock); |
1a3e8f3d | 3266 | |
bad55843 DC |
3267 | /* |
3268 | * Do an un-protected check to see if the inode is dirty and | |
3269 | * is a candidate for flushing. These checks will be repeated | |
3270 | * later after the appropriate locks are acquired. | |
3271 | */ | |
19429363 | 3272 | if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) |
bad55843 | 3273 | continue; |
bad55843 DC |
3274 | |
3275 | /* | |
3276 | * Try to get locks. If any are unavailable or it is pinned, | |
3277 | * then this inode cannot be flushed and is skipped. | |
3278 | */ | |
3279 | ||
19429363 | 3280 | if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) |
bad55843 | 3281 | continue; |
19429363 DC |
3282 | if (!xfs_iflock_nowait(cip)) { |
3283 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3284 | continue; |
3285 | } | |
19429363 DC |
3286 | if (xfs_ipincount(cip)) { |
3287 | xfs_ifunlock(cip); | |
3288 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3289 | continue; |
3290 | } | |
3291 | ||
8a17d7dd DC |
3292 | |
3293 | /* | |
3294 | * Check the inode number again, just to be certain we are not | |
3295 | * racing with freeing in xfs_reclaim_inode(). See the comments | |
3296 | * in that function for more information as to why the initial | |
3297 | * check is not sufficient. | |
3298 | */ | |
19429363 DC |
3299 | if (!cip->i_ino) { |
3300 | xfs_ifunlock(cip); | |
3301 | xfs_iunlock(cip, XFS_ILOCK_SHARED); | |
bad55843 DC |
3302 | continue; |
3303 | } | |
3304 | ||
3305 | /* | |
3306 | * arriving here means that this inode can be flushed. First | |
3307 | * re-check that it's dirty before flushing. | |
3308 | */ | |
19429363 | 3309 | if (!xfs_inode_clean(cip)) { |
33540408 | 3310 | int error; |
19429363 | 3311 | error = xfs_iflush_int(cip, bp); |
bad55843 | 3312 | if (error) { |
19429363 | 3313 | xfs_iunlock(cip, XFS_ILOCK_SHARED); |
bad55843 DC |
3314 | goto cluster_corrupt_out; |
3315 | } | |
3316 | clcount++; | |
3317 | } else { | |
19429363 | 3318 | xfs_ifunlock(cip); |
bad55843 | 3319 | } |
19429363 | 3320 | xfs_iunlock(cip, XFS_ILOCK_SHARED); |
bad55843 DC |
3321 | } |
3322 | ||
3323 | if (clcount) { | |
ff6d6af2 BD |
3324 | XFS_STATS_INC(mp, xs_icluster_flushcnt); |
3325 | XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); | |
bad55843 DC |
3326 | } |
3327 | ||
3328 | out_free: | |
1a3e8f3d | 3329 | rcu_read_unlock(); |
19429363 | 3330 | kmem_free(cilist); |
44b56e0a DC |
3331 | out_put: |
3332 | xfs_perag_put(pag); | |
bad55843 DC |
3333 | return 0; |
3334 | ||
3335 | ||
3336 | cluster_corrupt_out: | |
3337 | /* | |
3338 | * Corruption detected in the clustering loop. Invalidate the | |
3339 | * inode buffer and shut down the filesystem. | |
3340 | */ | |
1a3e8f3d | 3341 | rcu_read_unlock(); |
bad55843 DC |
3342 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
3343 | ||
bad55843 | 3344 | /* |
e53946db DC |
3345 | * We'll always have an inode attached to the buffer for completion |
3346 | * process by the time we are called from xfs_iflush(). Hence we have | |
3347 | * always need to do IO completion processing to abort the inodes | |
3348 | * attached to the buffer. handle them just like the shutdown case in | |
3349 | * xfs_buf_submit(). | |
bad55843 | 3350 | */ |
e53946db DC |
3351 | ASSERT(bp->b_iodone); |
3352 | bp->b_flags &= ~XBF_DONE; | |
3353 | xfs_buf_stale(bp); | |
3354 | xfs_buf_ioerror(bp, -EIO); | |
3355 | xfs_buf_ioend(bp); | |
3356 | ||
3357 | /* abort the corrupt inode, as it was not attached to the buffer */ | |
19429363 DC |
3358 | xfs_iflush_abort(cip, false); |
3359 | kmem_free(cilist); | |
44b56e0a | 3360 | xfs_perag_put(pag); |
2451337d | 3361 | return -EFSCORRUPTED; |
bad55843 DC |
3362 | } |
3363 | ||
1da177e4 | 3364 | /* |
4c46819a CH |
3365 | * Flush dirty inode metadata into the backing buffer. |
3366 | * | |
3367 | * The caller must have the inode lock and the inode flush lock held. The | |
3368 | * inode lock will still be held upon return to the caller, and the inode | |
3369 | * flush lock will be released after the inode has reached the disk. | |
3370 | * | |
3371 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
3372 | */ |
3373 | int | |
3374 | xfs_iflush( | |
4c46819a CH |
3375 | struct xfs_inode *ip, |
3376 | struct xfs_buf **bpp) | |
1da177e4 | 3377 | { |
4c46819a | 3378 | struct xfs_mount *mp = ip->i_mount; |
b1438f47 | 3379 | struct xfs_buf *bp = NULL; |
4c46819a | 3380 | struct xfs_dinode *dip; |
1da177e4 | 3381 | int error; |
1da177e4 | 3382 | |
ff6d6af2 | 3383 | XFS_STATS_INC(mp, xs_iflush_count); |
1da177e4 | 3384 | |
579aa9ca | 3385 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3386 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3387 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3388 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 3389 | |
4c46819a | 3390 | *bpp = NULL; |
1da177e4 | 3391 | |
1da177e4 LT |
3392 | xfs_iunpin_wait(ip); |
3393 | ||
4b6a4688 DC |
3394 | /* |
3395 | * For stale inodes we cannot rely on the backing buffer remaining | |
3396 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 3397 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
3398 | * inodes below. We have to check this after ensuring the inode is |
3399 | * unpinned so that it is safe to reclaim the stale inode after the | |
3400 | * flush call. | |
3401 | */ | |
3402 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
3403 | xfs_ifunlock(ip); | |
3404 | return 0; | |
3405 | } | |
3406 | ||
1da177e4 LT |
3407 | /* |
3408 | * This may have been unpinned because the filesystem is shutting | |
3409 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
3410 | * to disk, because the log record didn't make it to disk. |
3411 | * | |
3412 | * We also have to remove the log item from the AIL in this case, | |
3413 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
3414 | */ |
3415 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
2451337d | 3416 | error = -EIO; |
32ce90a4 | 3417 | goto abort_out; |
1da177e4 LT |
3418 | } |
3419 | ||
a3f74ffb | 3420 | /* |
b1438f47 DC |
3421 | * Get the buffer containing the on-disk inode. We are doing a try-lock |
3422 | * operation here, so we may get an EAGAIN error. In that case, we | |
3423 | * simply want to return with the inode still dirty. | |
3424 | * | |
3425 | * If we get any other error, we effectively have a corruption situation | |
3426 | * and we cannot flush the inode, so we treat it the same as failing | |
3427 | * xfs_iflush_int(). | |
a3f74ffb | 3428 | */ |
475ee413 CH |
3429 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
3430 | 0); | |
b1438f47 | 3431 | if (error == -EAGAIN) { |
a3f74ffb DC |
3432 | xfs_ifunlock(ip); |
3433 | return error; | |
3434 | } | |
b1438f47 DC |
3435 | if (error) |
3436 | goto corrupt_out; | |
a3f74ffb | 3437 | |
1da177e4 LT |
3438 | /* |
3439 | * First flush out the inode that xfs_iflush was called with. | |
3440 | */ | |
3441 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 3442 | if (error) |
1da177e4 | 3443 | goto corrupt_out; |
1da177e4 | 3444 | |
a3f74ffb DC |
3445 | /* |
3446 | * If the buffer is pinned then push on the log now so we won't | |
3447 | * get stuck waiting in the write for too long. | |
3448 | */ | |
811e64c7 | 3449 | if (xfs_buf_ispinned(bp)) |
a14a348b | 3450 | xfs_log_force(mp, 0); |
a3f74ffb | 3451 | |
1da177e4 | 3452 | /* |
e53946db DC |
3453 | * inode clustering: try to gather other inodes into this write |
3454 | * | |
3455 | * Note: Any error during clustering will result in the filesystem | |
3456 | * being shut down and completion callbacks run on the cluster buffer. | |
3457 | * As we have already flushed and attached this inode to the buffer, | |
3458 | * it has already been aborted and released by xfs_iflush_cluster() and | |
3459 | * so we have no further error handling to do here. | |
1da177e4 | 3460 | */ |
bad55843 DC |
3461 | error = xfs_iflush_cluster(ip, bp); |
3462 | if (error) | |
e53946db | 3463 | return error; |
1da177e4 | 3464 | |
4c46819a CH |
3465 | *bpp = bp; |
3466 | return 0; | |
1da177e4 LT |
3467 | |
3468 | corrupt_out: | |
b1438f47 DC |
3469 | if (bp) |
3470 | xfs_buf_relse(bp); | |
7d04a335 | 3471 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
32ce90a4 | 3472 | abort_out: |
e53946db | 3473 | /* abort the corrupt inode, as it was not attached to the buffer */ |
04913fdd | 3474 | xfs_iflush_abort(ip, false); |
32ce90a4 | 3475 | return error; |
1da177e4 LT |
3476 | } |
3477 | ||
9cfb9b47 DW |
3478 | /* |
3479 | * If there are inline format data / attr forks attached to this inode, | |
3480 | * make sure they're not corrupt. | |
3481 | */ | |
3482 | bool | |
3483 | xfs_inode_verify_forks( | |
3484 | struct xfs_inode *ip) | |
3485 | { | |
22431bf3 | 3486 | struct xfs_ifork *ifp; |
9cfb9b47 DW |
3487 | xfs_failaddr_t fa; |
3488 | ||
3489 | fa = xfs_ifork_verify_data(ip, &xfs_default_ifork_ops); | |
3490 | if (fa) { | |
22431bf3 DW |
3491 | ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); |
3492 | xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork", | |
3493 | ifp->if_u1.if_data, ifp->if_bytes, fa); | |
9cfb9b47 DW |
3494 | return false; |
3495 | } | |
3496 | ||
3497 | fa = xfs_ifork_verify_attr(ip, &xfs_default_ifork_ops); | |
3498 | if (fa) { | |
22431bf3 DW |
3499 | ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK); |
3500 | xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork", | |
3501 | ifp ? ifp->if_u1.if_data : NULL, | |
3502 | ifp ? ifp->if_bytes : 0, fa); | |
9cfb9b47 DW |
3503 | return false; |
3504 | } | |
3505 | return true; | |
3506 | } | |
3507 | ||
1da177e4 LT |
3508 | STATIC int |
3509 | xfs_iflush_int( | |
93848a99 CH |
3510 | struct xfs_inode *ip, |
3511 | struct xfs_buf *bp) | |
1da177e4 | 3512 | { |
93848a99 CH |
3513 | struct xfs_inode_log_item *iip = ip->i_itemp; |
3514 | struct xfs_dinode *dip; | |
3515 | struct xfs_mount *mp = ip->i_mount; | |
1da177e4 | 3516 | |
579aa9ca | 3517 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3518 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3519 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3520 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
93848a99 | 3521 | ASSERT(iip != NULL && iip->ili_fields != 0); |
263997a6 | 3522 | ASSERT(ip->i_d.di_version > 1); |
1da177e4 | 3523 | |
1da177e4 | 3524 | /* set *dip = inode's place in the buffer */ |
88ee2df7 | 3525 | dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 3526 | |
69ef921b | 3527 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
9e24cfd0 | 3528 | mp, XFS_ERRTAG_IFLUSH_1)) { |
6a19d939 | 3529 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
c9690043 | 3530 | "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT, |
6a19d939 | 3531 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); |
1da177e4 LT |
3532 | goto corrupt_out; |
3533 | } | |
c19b3b05 | 3534 | if (S_ISREG(VFS_I(ip)->i_mode)) { |
1da177e4 LT |
3535 | if (XFS_TEST_ERROR( |
3536 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3537 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 3538 | mp, XFS_ERRTAG_IFLUSH_3)) { |
6a19d939 | 3539 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
c9690043 | 3540 | "%s: Bad regular inode %Lu, ptr "PTR_FMT, |
6a19d939 | 3541 | __func__, ip->i_ino, ip); |
1da177e4 LT |
3542 | goto corrupt_out; |
3543 | } | |
c19b3b05 | 3544 | } else if (S_ISDIR(VFS_I(ip)->i_mode)) { |
1da177e4 LT |
3545 | if (XFS_TEST_ERROR( |
3546 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3547 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3548 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
9e24cfd0 | 3549 | mp, XFS_ERRTAG_IFLUSH_4)) { |
6a19d939 | 3550 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
c9690043 | 3551 | "%s: Bad directory inode %Lu, ptr "PTR_FMT, |
6a19d939 | 3552 | __func__, ip->i_ino, ip); |
1da177e4 LT |
3553 | goto corrupt_out; |
3554 | } | |
3555 | } | |
3556 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
9e24cfd0 | 3557 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { |
6a19d939 DC |
3558 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3559 | "%s: detected corrupt incore inode %Lu, " | |
c9690043 | 3560 | "total extents = %d, nblocks = %Ld, ptr "PTR_FMT, |
6a19d939 | 3561 | __func__, ip->i_ino, |
1da177e4 | 3562 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 3563 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
3564 | goto corrupt_out; |
3565 | } | |
3566 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
9e24cfd0 | 3567 | mp, XFS_ERRTAG_IFLUSH_6)) { |
6a19d939 | 3568 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
c9690043 | 3569 | "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT, |
6a19d939 | 3570 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); |
1da177e4 LT |
3571 | goto corrupt_out; |
3572 | } | |
e60896d8 | 3573 | |
1da177e4 | 3574 | /* |
263997a6 | 3575 | * Inode item log recovery for v2 inodes are dependent on the |
e60896d8 DC |
3576 | * di_flushiter count for correct sequencing. We bump the flush |
3577 | * iteration count so we can detect flushes which postdate a log record | |
3578 | * during recovery. This is redundant as we now log every change and | |
3579 | * hence this can't happen but we need to still do it to ensure | |
3580 | * backwards compatibility with old kernels that predate logging all | |
3581 | * inode changes. | |
1da177e4 | 3582 | */ |
e60896d8 DC |
3583 | if (ip->i_d.di_version < 3) |
3584 | ip->i_d.di_flushiter++; | |
1da177e4 | 3585 | |
9cfb9b47 DW |
3586 | /* Check the inline fork data before we write out. */ |
3587 | if (!xfs_inode_verify_forks(ip)) | |
005c5db8 DW |
3588 | goto corrupt_out; |
3589 | ||
1da177e4 | 3590 | /* |
3987848c DC |
3591 | * Copy the dirty parts of the inode into the on-disk inode. We always |
3592 | * copy out the core of the inode, because if the inode is dirty at all | |
3593 | * the core must be. | |
1da177e4 | 3594 | */ |
93f958f9 | 3595 | xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); |
1da177e4 LT |
3596 | |
3597 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3598 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3599 | ip->i_d.di_flushiter = 0; | |
3600 | ||
005c5db8 DW |
3601 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); |
3602 | if (XFS_IFORK_Q(ip)) | |
3603 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); | |
1da177e4 LT |
3604 | xfs_inobp_check(mp, bp); |
3605 | ||
3606 | /* | |
f5d8d5c4 CH |
3607 | * We've recorded everything logged in the inode, so we'd like to clear |
3608 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
3609 | * However, we can't stop logging all this information until the data | |
3610 | * we've copied into the disk buffer is written to disk. If we did we | |
3611 | * might overwrite the copy of the inode in the log with all the data | |
3612 | * after re-logging only part of it, and in the face of a crash we | |
3613 | * wouldn't have all the data we need to recover. | |
1da177e4 | 3614 | * |
f5d8d5c4 CH |
3615 | * What we do is move the bits to the ili_last_fields field. When |
3616 | * logging the inode, these bits are moved back to the ili_fields field. | |
3617 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
3618 | * know that the information those bits represent is permanently on | |
3619 | * disk. As long as the flush completes before the inode is logged | |
3620 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 3621 | * |
f5d8d5c4 CH |
3622 | * We can play with the ili_fields bits here, because the inode lock |
3623 | * must be held exclusively in order to set bits there and the flush | |
3624 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
3625 | * done routine can tell whether or not to look in the AIL. Also, store | |
3626 | * the current LSN of the inode so that we can tell whether the item has | |
3627 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
3628 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
3629 | * atomically. | |
1da177e4 | 3630 | */ |
93848a99 CH |
3631 | iip->ili_last_fields = iip->ili_fields; |
3632 | iip->ili_fields = 0; | |
fc0561ce | 3633 | iip->ili_fsync_fields = 0; |
93848a99 | 3634 | iip->ili_logged = 1; |
1da177e4 | 3635 | |
93848a99 CH |
3636 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
3637 | &iip->ili_item.li_lsn); | |
1da177e4 | 3638 | |
93848a99 CH |
3639 | /* |
3640 | * Attach the function xfs_iflush_done to the inode's | |
3641 | * buffer. This will remove the inode from the AIL | |
3642 | * and unlock the inode's flush lock when the inode is | |
3643 | * completely written to disk. | |
3644 | */ | |
3645 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); | |
1da177e4 | 3646 | |
93848a99 CH |
3647 | /* generate the checksum. */ |
3648 | xfs_dinode_calc_crc(mp, dip); | |
1da177e4 | 3649 | |
643c8c05 | 3650 | ASSERT(!list_empty(&bp->b_li_list)); |
93848a99 | 3651 | ASSERT(bp->b_iodone != NULL); |
1da177e4 LT |
3652 | return 0; |
3653 | ||
3654 | corrupt_out: | |
2451337d | 3655 | return -EFSCORRUPTED; |
1da177e4 | 3656 | } |