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 | */ |
f0e28280 | 6 | #include <linux/iversion.h> |
40ebd81d | 7 | |
1da177e4 | 8 | #include "xfs.h" |
a844f451 | 9 | #include "xfs_fs.h" |
70a9883c | 10 | #include "xfs_shared.h" |
239880ef DC |
11 | #include "xfs_format.h" |
12 | #include "xfs_log_format.h" | |
13 | #include "xfs_trans_resv.h" | |
1da177e4 | 14 | #include "xfs_mount.h" |
3ab78df2 | 15 | #include "xfs_defer.h" |
a4fbe6ab | 16 | #include "xfs_inode.h" |
c24b5dfa | 17 | #include "xfs_dir2.h" |
c24b5dfa | 18 | #include "xfs_attr.h" |
239880ef DC |
19 | #include "xfs_trans_space.h" |
20 | #include "xfs_trans.h" | |
1da177e4 | 21 | #include "xfs_buf_item.h" |
a844f451 | 22 | #include "xfs_inode_item.h" |
784eb7d8 | 23 | #include "xfs_iunlink_item.h" |
a844f451 NS |
24 | #include "xfs_ialloc.h" |
25 | #include "xfs_bmap.h" | |
68988114 | 26 | #include "xfs_bmap_util.h" |
e9e899a2 | 27 | #include "xfs_errortag.h" |
1da177e4 | 28 | #include "xfs_error.h" |
1da177e4 | 29 | #include "xfs_quota.h" |
2a82b8be | 30 | #include "xfs_filestream.h" |
0b1b213f | 31 | #include "xfs_trace.h" |
33479e05 | 32 | #include "xfs_icache.h" |
c24b5dfa | 33 | #include "xfs_symlink.h" |
239880ef DC |
34 | #include "xfs_trans_priv.h" |
35 | #include "xfs_log.h" | |
a4fbe6ab | 36 | #include "xfs_bmap_btree.h" |
aa8968f2 | 37 | #include "xfs_reflink.h" |
9bbafc71 | 38 | #include "xfs_ag.h" |
01728b44 | 39 | #include "xfs_log_priv.h" |
1da177e4 | 40 | |
182696fb | 41 | struct kmem_cache *xfs_inode_cache; |
1da177e4 LT |
42 | |
43 | /* | |
8f04c47a | 44 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
45 | * freed from a file in a single transaction. |
46 | */ | |
47 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
48 | ||
54d7b5c1 | 49 | STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); |
f40aadb2 DC |
50 | STATIC int xfs_iunlink_remove(struct xfs_trans *tp, struct xfs_perag *pag, |
51 | struct xfs_inode *); | |
ab297431 | 52 | |
2a0ec1d9 DC |
53 | /* |
54 | * helper function to extract extent size hint from inode | |
55 | */ | |
56 | xfs_extlen_t | |
57 | xfs_get_extsz_hint( | |
58 | struct xfs_inode *ip) | |
59 | { | |
bdb2ed2d CH |
60 | /* |
61 | * No point in aligning allocations if we need to COW to actually | |
62 | * write to them. | |
63 | */ | |
64 | if (xfs_is_always_cow_inode(ip)) | |
65 | return 0; | |
db07349d | 66 | if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize) |
031474c2 | 67 | return ip->i_extsize; |
2a0ec1d9 DC |
68 | if (XFS_IS_REALTIME_INODE(ip)) |
69 | return ip->i_mount->m_sb.sb_rextsize; | |
70 | return 0; | |
71 | } | |
72 | ||
f7ca3522 DW |
73 | /* |
74 | * Helper function to extract CoW extent size hint from inode. | |
75 | * Between the extent size hint and the CoW extent size hint, we | |
e153aa79 DW |
76 | * return the greater of the two. If the value is zero (automatic), |
77 | * use the default size. | |
f7ca3522 DW |
78 | */ |
79 | xfs_extlen_t | |
80 | xfs_get_cowextsz_hint( | |
81 | struct xfs_inode *ip) | |
82 | { | |
83 | xfs_extlen_t a, b; | |
84 | ||
85 | a = 0; | |
3e09ab8f | 86 | if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) |
b33ce57d | 87 | a = ip->i_cowextsize; |
f7ca3522 DW |
88 | b = xfs_get_extsz_hint(ip); |
89 | ||
e153aa79 DW |
90 | a = max(a, b); |
91 | if (a == 0) | |
92 | return XFS_DEFAULT_COWEXTSZ_HINT; | |
93 | return a; | |
f7ca3522 DW |
94 | } |
95 | ||
fa96acad | 96 | /* |
efa70be1 CH |
97 | * These two are wrapper routines around the xfs_ilock() routine used to |
98 | * centralize some grungy code. They are used in places that wish to lock the | |
99 | * inode solely for reading the extents. The reason these places can't just | |
100 | * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to | |
101 | * bringing in of the extents from disk for a file in b-tree format. If the | |
102 | * inode is in b-tree format, then we need to lock the inode exclusively until | |
103 | * the extents are read in. Locking it exclusively all the time would limit | |
104 | * our parallelism unnecessarily, though. What we do instead is check to see | |
105 | * if the extents have been read in yet, and only lock the inode exclusively | |
106 | * if they have not. | |
fa96acad | 107 | * |
efa70be1 | 108 | * The functions return a value which should be given to the corresponding |
01f4f327 | 109 | * xfs_iunlock() call. |
fa96acad DC |
110 | */ |
111 | uint | |
309ecac8 CH |
112 | xfs_ilock_data_map_shared( |
113 | struct xfs_inode *ip) | |
fa96acad | 114 | { |
309ecac8 | 115 | uint lock_mode = XFS_ILOCK_SHARED; |
fa96acad | 116 | |
b2197a36 | 117 | if (xfs_need_iread_extents(&ip->i_df)) |
fa96acad | 118 | lock_mode = XFS_ILOCK_EXCL; |
fa96acad | 119 | xfs_ilock(ip, lock_mode); |
fa96acad DC |
120 | return lock_mode; |
121 | } | |
122 | ||
efa70be1 CH |
123 | uint |
124 | xfs_ilock_attr_map_shared( | |
125 | struct xfs_inode *ip) | |
fa96acad | 126 | { |
efa70be1 CH |
127 | uint lock_mode = XFS_ILOCK_SHARED; |
128 | ||
932b42c6 | 129 | if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af)) |
efa70be1 CH |
130 | lock_mode = XFS_ILOCK_EXCL; |
131 | xfs_ilock(ip, lock_mode); | |
132 | return lock_mode; | |
fa96acad DC |
133 | } |
134 | ||
ca76a761 KX |
135 | /* |
136 | * You can't set both SHARED and EXCL for the same lock, | |
137 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED, | |
138 | * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values | |
139 | * to set in lock_flags. | |
140 | */ | |
141 | static inline void | |
142 | xfs_lock_flags_assert( | |
143 | uint lock_flags) | |
144 | { | |
145 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
146 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
147 | ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != | |
148 | (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); | |
149 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
150 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
151 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); | |
152 | ASSERT(lock_flags != 0); | |
153 | } | |
154 | ||
fa96acad | 155 | /* |
65523218 | 156 | * In addition to i_rwsem in the VFS inode, the xfs inode contains 2 |
2433480a | 157 | * multi-reader locks: invalidate_lock and the i_lock. This routine allows |
65523218 | 158 | * various combinations of the locks to be obtained. |
fa96acad | 159 | * |
653c60b6 DC |
160 | * The 3 locks should always be ordered so that the IO lock is obtained first, |
161 | * the mmap lock second and the ilock last in order to prevent deadlock. | |
fa96acad | 162 | * |
653c60b6 DC |
163 | * Basic locking order: |
164 | * | |
2433480a | 165 | * i_rwsem -> invalidate_lock -> page_lock -> i_ilock |
653c60b6 | 166 | * |
c1e8d7c6 | 167 | * mmap_lock locking order: |
653c60b6 | 168 | * |
c1e8d7c6 | 169 | * i_rwsem -> page lock -> mmap_lock |
2433480a | 170 | * mmap_lock -> invalidate_lock -> page_lock |
653c60b6 | 171 | * |
c1e8d7c6 | 172 | * The difference in mmap_lock locking order mean that we cannot hold the |
2433480a JK |
173 | * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths |
174 | * can fault in pages during copy in/out (for buffered IO) or require the | |
175 | * mmap_lock in get_user_pages() to map the user pages into the kernel address | |
176 | * space for direct IO. Similarly the i_rwsem cannot be taken inside a page | |
177 | * fault because page faults already hold the mmap_lock. | |
653c60b6 DC |
178 | * |
179 | * Hence to serialise fully against both syscall and mmap based IO, we need to | |
2433480a JK |
180 | * take both the i_rwsem and the invalidate_lock. These locks should *only* be |
181 | * both taken in places where we need to invalidate the page cache in a race | |
653c60b6 DC |
182 | * free manner (e.g. truncate, hole punch and other extent manipulation |
183 | * functions). | |
fa96acad DC |
184 | */ |
185 | void | |
186 | xfs_ilock( | |
187 | xfs_inode_t *ip, | |
188 | uint lock_flags) | |
189 | { | |
190 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
191 | ||
ca76a761 | 192 | xfs_lock_flags_assert(lock_flags); |
fa96acad | 193 | |
65523218 CH |
194 | if (lock_flags & XFS_IOLOCK_EXCL) { |
195 | down_write_nested(&VFS_I(ip)->i_rwsem, | |
196 | XFS_IOLOCK_DEP(lock_flags)); | |
197 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
198 | down_read_nested(&VFS_I(ip)->i_rwsem, | |
199 | XFS_IOLOCK_DEP(lock_flags)); | |
200 | } | |
fa96acad | 201 | |
2433480a JK |
202 | if (lock_flags & XFS_MMAPLOCK_EXCL) { |
203 | down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock, | |
204 | XFS_MMAPLOCK_DEP(lock_flags)); | |
205 | } else if (lock_flags & XFS_MMAPLOCK_SHARED) { | |
206 | down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock, | |
207 | XFS_MMAPLOCK_DEP(lock_flags)); | |
208 | } | |
653c60b6 | 209 | |
fa96acad DC |
210 | if (lock_flags & XFS_ILOCK_EXCL) |
211 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
212 | else if (lock_flags & XFS_ILOCK_SHARED) | |
213 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
214 | } | |
215 | ||
216 | /* | |
217 | * This is just like xfs_ilock(), except that the caller | |
218 | * is guaranteed not to sleep. It returns 1 if it gets | |
219 | * the requested locks and 0 otherwise. If the IO lock is | |
220 | * obtained but the inode lock cannot be, then the IO lock | |
221 | * is dropped before returning. | |
222 | * | |
223 | * ip -- the inode being locked | |
224 | * lock_flags -- this parameter indicates the inode's locks to be | |
225 | * to be locked. See the comment for xfs_ilock() for a list | |
226 | * of valid values. | |
227 | */ | |
228 | int | |
229 | xfs_ilock_nowait( | |
230 | xfs_inode_t *ip, | |
231 | uint lock_flags) | |
232 | { | |
233 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
234 | ||
ca76a761 | 235 | xfs_lock_flags_assert(lock_flags); |
fa96acad DC |
236 | |
237 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
65523218 | 238 | if (!down_write_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
239 | goto out; |
240 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
65523218 | 241 | if (!down_read_trylock(&VFS_I(ip)->i_rwsem)) |
fa96acad DC |
242 | goto out; |
243 | } | |
653c60b6 DC |
244 | |
245 | if (lock_flags & XFS_MMAPLOCK_EXCL) { | |
2433480a | 246 | if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock)) |
653c60b6 DC |
247 | goto out_undo_iolock; |
248 | } else if (lock_flags & XFS_MMAPLOCK_SHARED) { | |
2433480a | 249 | if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock)) |
653c60b6 DC |
250 | goto out_undo_iolock; |
251 | } | |
252 | ||
fa96acad DC |
253 | if (lock_flags & XFS_ILOCK_EXCL) { |
254 | if (!mrtryupdate(&ip->i_lock)) | |
653c60b6 | 255 | goto out_undo_mmaplock; |
fa96acad DC |
256 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
257 | if (!mrtryaccess(&ip->i_lock)) | |
653c60b6 | 258 | goto out_undo_mmaplock; |
fa96acad DC |
259 | } |
260 | return 1; | |
261 | ||
653c60b6 DC |
262 | out_undo_mmaplock: |
263 | if (lock_flags & XFS_MMAPLOCK_EXCL) | |
2433480a | 264 | up_write(&VFS_I(ip)->i_mapping->invalidate_lock); |
653c60b6 | 265 | else if (lock_flags & XFS_MMAPLOCK_SHARED) |
2433480a | 266 | up_read(&VFS_I(ip)->i_mapping->invalidate_lock); |
653c60b6 | 267 | out_undo_iolock: |
fa96acad | 268 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 269 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 270 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 271 | up_read(&VFS_I(ip)->i_rwsem); |
653c60b6 | 272 | out: |
fa96acad DC |
273 | return 0; |
274 | } | |
275 | ||
276 | /* | |
277 | * xfs_iunlock() is used to drop the inode locks acquired with | |
278 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
279 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
280 | * that we know which locks to drop. | |
281 | * | |
282 | * ip -- the inode being unlocked | |
283 | * lock_flags -- this parameter indicates the inode's locks to be | |
284 | * to be unlocked. See the comment for xfs_ilock() for a list | |
285 | * of valid values for this parameter. | |
286 | * | |
287 | */ | |
288 | void | |
289 | xfs_iunlock( | |
290 | xfs_inode_t *ip, | |
291 | uint lock_flags) | |
292 | { | |
ca76a761 | 293 | xfs_lock_flags_assert(lock_flags); |
fa96acad DC |
294 | |
295 | if (lock_flags & XFS_IOLOCK_EXCL) | |
65523218 | 296 | up_write(&VFS_I(ip)->i_rwsem); |
fa96acad | 297 | else if (lock_flags & XFS_IOLOCK_SHARED) |
65523218 | 298 | up_read(&VFS_I(ip)->i_rwsem); |
fa96acad | 299 | |
653c60b6 | 300 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
2433480a | 301 | up_write(&VFS_I(ip)->i_mapping->invalidate_lock); |
653c60b6 | 302 | else if (lock_flags & XFS_MMAPLOCK_SHARED) |
2433480a | 303 | up_read(&VFS_I(ip)->i_mapping->invalidate_lock); |
653c60b6 | 304 | |
fa96acad DC |
305 | if (lock_flags & XFS_ILOCK_EXCL) |
306 | mrunlock_excl(&ip->i_lock); | |
307 | else if (lock_flags & XFS_ILOCK_SHARED) | |
308 | mrunlock_shared(&ip->i_lock); | |
309 | ||
310 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
311 | } | |
312 | ||
313 | /* | |
314 | * give up write locks. the i/o lock cannot be held nested | |
315 | * if it is being demoted. | |
316 | */ | |
317 | void | |
318 | xfs_ilock_demote( | |
319 | xfs_inode_t *ip, | |
320 | uint lock_flags) | |
321 | { | |
653c60b6 DC |
322 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); |
323 | ASSERT((lock_flags & | |
324 | ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
fa96acad DC |
325 | |
326 | if (lock_flags & XFS_ILOCK_EXCL) | |
327 | mrdemote(&ip->i_lock); | |
653c60b6 | 328 | if (lock_flags & XFS_MMAPLOCK_EXCL) |
2433480a | 329 | downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock); |
fa96acad | 330 | if (lock_flags & XFS_IOLOCK_EXCL) |
65523218 | 331 | downgrade_write(&VFS_I(ip)->i_rwsem); |
fa96acad DC |
332 | |
333 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
334 | } | |
335 | ||
742ae1e3 | 336 | #if defined(DEBUG) || defined(XFS_WARN) |
e31cbde7 PR |
337 | static inline bool |
338 | __xfs_rwsem_islocked( | |
339 | struct rw_semaphore *rwsem, | |
340 | bool shared) | |
341 | { | |
342 | if (!debug_locks) | |
343 | return rwsem_is_locked(rwsem); | |
344 | ||
345 | if (!shared) | |
346 | return lockdep_is_held_type(rwsem, 0); | |
347 | ||
348 | /* | |
349 | * We are checking that the lock is held at least in shared | |
350 | * mode but don't care that it might be held exclusively | |
351 | * (i.e. shared | excl). Hence we check if the lock is held | |
352 | * in any mode rather than an explicit shared mode. | |
353 | */ | |
354 | return lockdep_is_held_type(rwsem, -1); | |
355 | } | |
356 | ||
357 | bool | |
fa96acad | 358 | xfs_isilocked( |
e31cbde7 | 359 | struct xfs_inode *ip, |
fa96acad DC |
360 | uint lock_flags) |
361 | { | |
362 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
363 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
364 | return !!ip->i_lock.mr_writer; | |
365 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
366 | } | |
367 | ||
653c60b6 | 368 | if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { |
82af8806 KX |
369 | return __xfs_rwsem_islocked(&VFS_I(ip)->i_mapping->invalidate_lock, |
370 | (lock_flags & XFS_MMAPLOCK_SHARED)); | |
653c60b6 DC |
371 | } |
372 | ||
e31cbde7 PR |
373 | if (lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) { |
374 | return __xfs_rwsem_islocked(&VFS_I(ip)->i_rwsem, | |
375 | (lock_flags & XFS_IOLOCK_SHARED)); | |
fa96acad DC |
376 | } |
377 | ||
378 | ASSERT(0); | |
e31cbde7 | 379 | return false; |
fa96acad DC |
380 | } |
381 | #endif | |
382 | ||
b6a9947e DC |
383 | /* |
384 | * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when | |
385 | * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined | |
386 | * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build | |
387 | * errors and warnings. | |
388 | */ | |
389 | #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) | |
3403ccc0 DC |
390 | static bool |
391 | xfs_lockdep_subclass_ok( | |
392 | int subclass) | |
393 | { | |
394 | return subclass < MAX_LOCKDEP_SUBCLASSES; | |
395 | } | |
396 | #else | |
397 | #define xfs_lockdep_subclass_ok(subclass) (true) | |
398 | #endif | |
399 | ||
c24b5dfa | 400 | /* |
653c60b6 | 401 | * Bump the subclass so xfs_lock_inodes() acquires each lock with a different |
0952c818 DC |
402 | * value. This can be called for any type of inode lock combination, including |
403 | * parent locking. Care must be taken to ensure we don't overrun the subclass | |
404 | * storage fields in the class mask we build. | |
c24b5dfa | 405 | */ |
a1033753 DC |
406 | static inline uint |
407 | xfs_lock_inumorder( | |
408 | uint lock_mode, | |
409 | uint subclass) | |
c24b5dfa | 410 | { |
a1033753 | 411 | uint class = 0; |
0952c818 DC |
412 | |
413 | ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | | |
414 | XFS_ILOCK_RTSUM))); | |
3403ccc0 | 415 | ASSERT(xfs_lockdep_subclass_ok(subclass)); |
0952c818 | 416 | |
653c60b6 | 417 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { |
0952c818 | 418 | ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); |
0952c818 | 419 | class += subclass << XFS_IOLOCK_SHIFT; |
653c60b6 DC |
420 | } |
421 | ||
422 | if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { | |
0952c818 DC |
423 | ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); |
424 | class += subclass << XFS_MMAPLOCK_SHIFT; | |
653c60b6 DC |
425 | } |
426 | ||
0952c818 DC |
427 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { |
428 | ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); | |
429 | class += subclass << XFS_ILOCK_SHIFT; | |
430 | } | |
c24b5dfa | 431 | |
0952c818 | 432 | return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; |
c24b5dfa DC |
433 | } |
434 | ||
435 | /* | |
95afcf5c DC |
436 | * The following routine will lock n inodes in exclusive mode. We assume the |
437 | * caller calls us with the inodes in i_ino order. | |
c24b5dfa | 438 | * |
95afcf5c DC |
439 | * We need to detect deadlock where an inode that we lock is in the AIL and we |
440 | * start waiting for another inode that is locked by a thread in a long running | |
441 | * transaction (such as truncate). This can result in deadlock since the long | |
442 | * running trans might need to wait for the inode we just locked in order to | |
443 | * push the tail and free space in the log. | |
0952c818 DC |
444 | * |
445 | * xfs_lock_inodes() can only be used to lock one type of lock at a time - | |
446 | * the iolock, the mmaplock or the ilock, but not more than one at a time. If we | |
447 | * lock more than one at a time, lockdep will report false positives saying we | |
448 | * have violated locking orders. | |
c24b5dfa | 449 | */ |
0d5a75e9 | 450 | static void |
c24b5dfa | 451 | xfs_lock_inodes( |
efe2330f CH |
452 | struct xfs_inode **ips, |
453 | int inodes, | |
454 | uint lock_mode) | |
c24b5dfa | 455 | { |
a1033753 DC |
456 | int attempts = 0; |
457 | uint i; | |
458 | int j; | |
459 | bool try_lock; | |
efe2330f | 460 | struct xfs_log_item *lp; |
c24b5dfa | 461 | |
0952c818 DC |
462 | /* |
463 | * Currently supports between 2 and 5 inodes with exclusive locking. We | |
464 | * support an arbitrary depth of locking here, but absolute limits on | |
b63da6c8 | 465 | * inodes depend on the type of locking and the limits placed by |
0952c818 DC |
466 | * lockdep annotations in xfs_lock_inumorder. These are all checked by |
467 | * the asserts. | |
468 | */ | |
95afcf5c | 469 | ASSERT(ips && inodes >= 2 && inodes <= 5); |
0952c818 DC |
470 | ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | |
471 | XFS_ILOCK_EXCL)); | |
472 | ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | | |
473 | XFS_ILOCK_SHARED))); | |
0952c818 DC |
474 | ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || |
475 | inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); | |
476 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || | |
477 | inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); | |
478 | ||
479 | if (lock_mode & XFS_IOLOCK_EXCL) { | |
480 | ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); | |
481 | } else if (lock_mode & XFS_MMAPLOCK_EXCL) | |
482 | ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); | |
c24b5dfa | 483 | |
c24b5dfa | 484 | again: |
a1033753 DC |
485 | try_lock = false; |
486 | i = 0; | |
c24b5dfa DC |
487 | for (; i < inodes; i++) { |
488 | ASSERT(ips[i]); | |
489 | ||
95afcf5c | 490 | if (i && (ips[i] == ips[i - 1])) /* Already locked */ |
c24b5dfa DC |
491 | continue; |
492 | ||
493 | /* | |
95afcf5c DC |
494 | * If try_lock is not set yet, make sure all locked inodes are |
495 | * not in the AIL. If any are, set try_lock to be used later. | |
c24b5dfa | 496 | */ |
c24b5dfa DC |
497 | if (!try_lock) { |
498 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
b3b14aac | 499 | lp = &ips[j]->i_itemp->ili_item; |
22525c17 | 500 | if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) |
a1033753 | 501 | try_lock = true; |
c24b5dfa DC |
502 | } |
503 | } | |
504 | ||
505 | /* | |
506 | * If any of the previous locks we have locked is in the AIL, | |
507 | * we must TRY to get the second and subsequent locks. If | |
508 | * we can't get any, we must release all we have | |
509 | * and try again. | |
510 | */ | |
95afcf5c DC |
511 | if (!try_lock) { |
512 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); | |
513 | continue; | |
514 | } | |
515 | ||
516 | /* try_lock means we have an inode locked that is in the AIL. */ | |
517 | ASSERT(i != 0); | |
518 | if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) | |
519 | continue; | |
c24b5dfa | 520 | |
95afcf5c DC |
521 | /* |
522 | * Unlock all previous guys and try again. xfs_iunlock will try | |
523 | * to push the tail if the inode is in the AIL. | |
524 | */ | |
525 | attempts++; | |
526 | for (j = i - 1; j >= 0; j--) { | |
c24b5dfa | 527 | /* |
95afcf5c DC |
528 | * Check to see if we've already unlocked this one. Not |
529 | * the first one going back, and the inode ptr is the | |
530 | * same. | |
c24b5dfa | 531 | */ |
95afcf5c DC |
532 | if (j != (i - 1) && ips[j] == ips[j + 1]) |
533 | continue; | |
c24b5dfa | 534 | |
95afcf5c DC |
535 | xfs_iunlock(ips[j], lock_mode); |
536 | } | |
c24b5dfa | 537 | |
95afcf5c DC |
538 | if ((attempts % 5) == 0) { |
539 | delay(1); /* Don't just spin the CPU */ | |
c24b5dfa | 540 | } |
95afcf5c | 541 | goto again; |
c24b5dfa | 542 | } |
c24b5dfa DC |
543 | } |
544 | ||
545 | /* | |
d2c292d8 JK |
546 | * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and |
547 | * mmaplock must be double-locked separately since we use i_rwsem and | |
548 | * invalidate_lock for that. We now support taking one lock EXCL and the | |
549 | * other SHARED. | |
c24b5dfa DC |
550 | */ |
551 | void | |
552 | xfs_lock_two_inodes( | |
7c2d238a DW |
553 | struct xfs_inode *ip0, |
554 | uint ip0_mode, | |
555 | struct xfs_inode *ip1, | |
556 | uint ip1_mode) | |
c24b5dfa | 557 | { |
c24b5dfa | 558 | int attempts = 0; |
efe2330f | 559 | struct xfs_log_item *lp; |
c24b5dfa | 560 | |
7c2d238a DW |
561 | ASSERT(hweight32(ip0_mode) == 1); |
562 | ASSERT(hweight32(ip1_mode) == 1); | |
563 | ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); | |
564 | ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); | |
d2c292d8 JK |
565 | ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))); |
566 | ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))); | |
c24b5dfa DC |
567 | ASSERT(ip0->i_ino != ip1->i_ino); |
568 | ||
569 | if (ip0->i_ino > ip1->i_ino) { | |
2a09b575 CD |
570 | swap(ip0, ip1); |
571 | swap(ip0_mode, ip1_mode); | |
c24b5dfa DC |
572 | } |
573 | ||
574 | again: | |
7c2d238a | 575 | xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0)); |
c24b5dfa DC |
576 | |
577 | /* | |
578 | * If the first lock we have locked is in the AIL, we must TRY to get | |
579 | * the second lock. If we can't get it, we must release the first one | |
580 | * and try again. | |
581 | */ | |
b3b14aac | 582 | lp = &ip0->i_itemp->ili_item; |
22525c17 | 583 | if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) { |
7c2d238a DW |
584 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) { |
585 | xfs_iunlock(ip0, ip0_mode); | |
c24b5dfa DC |
586 | if ((++attempts % 5) == 0) |
587 | delay(1); /* Don't just spin the CPU */ | |
588 | goto again; | |
589 | } | |
590 | } else { | |
7c2d238a | 591 | xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1)); |
c24b5dfa DC |
592 | } |
593 | } | |
594 | ||
4422501d CH |
595 | uint |
596 | xfs_ip2xflags( | |
597 | struct xfs_inode *ip) | |
1da177e4 LT |
598 | { |
599 | uint flags = 0; | |
600 | ||
4422501d CH |
601 | if (ip->i_diflags & XFS_DIFLAG_ANY) { |
602 | if (ip->i_diflags & XFS_DIFLAG_REALTIME) | |
e7b89481 | 603 | flags |= FS_XFLAG_REALTIME; |
4422501d | 604 | if (ip->i_diflags & XFS_DIFLAG_PREALLOC) |
e7b89481 | 605 | flags |= FS_XFLAG_PREALLOC; |
4422501d | 606 | if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) |
e7b89481 | 607 | flags |= FS_XFLAG_IMMUTABLE; |
4422501d | 608 | if (ip->i_diflags & XFS_DIFLAG_APPEND) |
e7b89481 | 609 | flags |= FS_XFLAG_APPEND; |
4422501d | 610 | if (ip->i_diflags & XFS_DIFLAG_SYNC) |
e7b89481 | 611 | flags |= FS_XFLAG_SYNC; |
4422501d | 612 | if (ip->i_diflags & XFS_DIFLAG_NOATIME) |
e7b89481 | 613 | flags |= FS_XFLAG_NOATIME; |
4422501d | 614 | if (ip->i_diflags & XFS_DIFLAG_NODUMP) |
e7b89481 | 615 | flags |= FS_XFLAG_NODUMP; |
4422501d | 616 | if (ip->i_diflags & XFS_DIFLAG_RTINHERIT) |
e7b89481 | 617 | flags |= FS_XFLAG_RTINHERIT; |
4422501d | 618 | if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT) |
e7b89481 | 619 | flags |= FS_XFLAG_PROJINHERIT; |
4422501d | 620 | if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS) |
e7b89481 | 621 | flags |= FS_XFLAG_NOSYMLINKS; |
4422501d | 622 | if (ip->i_diflags & XFS_DIFLAG_EXTSIZE) |
e7b89481 | 623 | flags |= FS_XFLAG_EXTSIZE; |
4422501d | 624 | if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) |
e7b89481 | 625 | flags |= FS_XFLAG_EXTSZINHERIT; |
4422501d | 626 | if (ip->i_diflags & XFS_DIFLAG_NODEFRAG) |
e7b89481 | 627 | flags |= FS_XFLAG_NODEFRAG; |
4422501d | 628 | if (ip->i_diflags & XFS_DIFLAG_FILESTREAM) |
e7b89481 | 629 | flags |= FS_XFLAG_FILESTREAM; |
1da177e4 LT |
630 | } |
631 | ||
4422501d CH |
632 | if (ip->i_diflags2 & XFS_DIFLAG2_ANY) { |
633 | if (ip->i_diflags2 & XFS_DIFLAG2_DAX) | |
58f88ca2 | 634 | flags |= FS_XFLAG_DAX; |
4422501d | 635 | if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) |
f7ca3522 | 636 | flags |= FS_XFLAG_COWEXTSIZE; |
58f88ca2 DC |
637 | } |
638 | ||
932b42c6 | 639 | if (xfs_inode_has_attr_fork(ip)) |
58f88ca2 | 640 | flags |= FS_XFLAG_HASATTR; |
1da177e4 LT |
641 | return flags; |
642 | } | |
643 | ||
c24b5dfa DC |
644 | /* |
645 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
646 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
647 | * ci_name->name will point to a the actual name (caller must free) or | |
648 | * will be set to NULL if an exact match is found. | |
649 | */ | |
650 | int | |
651 | xfs_lookup( | |
996b2329 DW |
652 | struct xfs_inode *dp, |
653 | const struct xfs_name *name, | |
654 | struct xfs_inode **ipp, | |
c24b5dfa DC |
655 | struct xfs_name *ci_name) |
656 | { | |
657 | xfs_ino_t inum; | |
658 | int error; | |
c24b5dfa DC |
659 | |
660 | trace_xfs_lookup(dp, name); | |
661 | ||
75c8c50f | 662 | if (xfs_is_shutdown(dp->i_mount)) |
2451337d | 663 | return -EIO; |
c24b5dfa | 664 | |
c24b5dfa | 665 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
c24b5dfa | 666 | if (error) |
dbad7c99 | 667 | goto out_unlock; |
c24b5dfa DC |
668 | |
669 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); | |
670 | if (error) | |
671 | goto out_free_name; | |
672 | ||
673 | return 0; | |
674 | ||
675 | out_free_name: | |
676 | if (ci_name) | |
677 | kmem_free(ci_name->name); | |
dbad7c99 | 678 | out_unlock: |
c24b5dfa DC |
679 | *ipp = NULL; |
680 | return error; | |
681 | } | |
682 | ||
8a569d71 DW |
683 | /* Propagate di_flags from a parent inode to a child inode. */ |
684 | static void | |
685 | xfs_inode_inherit_flags( | |
686 | struct xfs_inode *ip, | |
687 | const struct xfs_inode *pip) | |
688 | { | |
689 | unsigned int di_flags = 0; | |
603f000b | 690 | xfs_failaddr_t failaddr; |
8a569d71 DW |
691 | umode_t mode = VFS_I(ip)->i_mode; |
692 | ||
693 | if (S_ISDIR(mode)) { | |
db07349d | 694 | if (pip->i_diflags & XFS_DIFLAG_RTINHERIT) |
8a569d71 | 695 | di_flags |= XFS_DIFLAG_RTINHERIT; |
db07349d | 696 | if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) { |
8a569d71 | 697 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; |
031474c2 | 698 | ip->i_extsize = pip->i_extsize; |
8a569d71 | 699 | } |
db07349d | 700 | if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT) |
8a569d71 DW |
701 | di_flags |= XFS_DIFLAG_PROJINHERIT; |
702 | } else if (S_ISREG(mode)) { | |
db07349d | 703 | if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) && |
38c26bfd | 704 | xfs_has_realtime(ip->i_mount)) |
8a569d71 | 705 | di_flags |= XFS_DIFLAG_REALTIME; |
db07349d | 706 | if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) { |
8a569d71 | 707 | di_flags |= XFS_DIFLAG_EXTSIZE; |
031474c2 | 708 | ip->i_extsize = pip->i_extsize; |
8a569d71 DW |
709 | } |
710 | } | |
db07349d | 711 | if ((pip->i_diflags & XFS_DIFLAG_NOATIME) && |
8a569d71 DW |
712 | xfs_inherit_noatime) |
713 | di_flags |= XFS_DIFLAG_NOATIME; | |
db07349d | 714 | if ((pip->i_diflags & XFS_DIFLAG_NODUMP) && |
8a569d71 DW |
715 | xfs_inherit_nodump) |
716 | di_flags |= XFS_DIFLAG_NODUMP; | |
db07349d | 717 | if ((pip->i_diflags & XFS_DIFLAG_SYNC) && |
8a569d71 DW |
718 | xfs_inherit_sync) |
719 | di_flags |= XFS_DIFLAG_SYNC; | |
db07349d | 720 | if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) && |
8a569d71 DW |
721 | xfs_inherit_nosymlinks) |
722 | di_flags |= XFS_DIFLAG_NOSYMLINKS; | |
db07349d | 723 | if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) && |
8a569d71 DW |
724 | xfs_inherit_nodefrag) |
725 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
db07349d | 726 | if (pip->i_diflags & XFS_DIFLAG_FILESTREAM) |
8a569d71 DW |
727 | di_flags |= XFS_DIFLAG_FILESTREAM; |
728 | ||
db07349d | 729 | ip->i_diflags |= di_flags; |
603f000b DW |
730 | |
731 | /* | |
732 | * Inode verifiers on older kernels only check that the extent size | |
733 | * hint is an integer multiple of the rt extent size on realtime files. | |
734 | * They did not check the hint alignment on a directory with both | |
735 | * rtinherit and extszinherit flags set. If the misaligned hint is | |
736 | * propagated from a directory into a new realtime file, new file | |
737 | * allocations will fail due to math errors in the rt allocator and/or | |
738 | * trip the verifiers. Validate the hint settings in the new file so | |
739 | * that we don't let broken hints propagate. | |
740 | */ | |
741 | failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize, | |
742 | VFS_I(ip)->i_mode, ip->i_diflags); | |
743 | if (failaddr) { | |
744 | ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE | | |
745 | XFS_DIFLAG_EXTSZINHERIT); | |
746 | ip->i_extsize = 0; | |
747 | } | |
8a569d71 DW |
748 | } |
749 | ||
750 | /* Propagate di_flags2 from a parent inode to a child inode. */ | |
751 | static void | |
752 | xfs_inode_inherit_flags2( | |
753 | struct xfs_inode *ip, | |
754 | const struct xfs_inode *pip) | |
755 | { | |
603f000b DW |
756 | xfs_failaddr_t failaddr; |
757 | ||
3e09ab8f CH |
758 | if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) { |
759 | ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE; | |
b33ce57d | 760 | ip->i_cowextsize = pip->i_cowextsize; |
8a569d71 | 761 | } |
3e09ab8f CH |
762 | if (pip->i_diflags2 & XFS_DIFLAG2_DAX) |
763 | ip->i_diflags2 |= XFS_DIFLAG2_DAX; | |
603f000b DW |
764 | |
765 | /* Don't let invalid cowextsize hints propagate. */ | |
766 | failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize, | |
767 | VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2); | |
768 | if (failaddr) { | |
769 | ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE; | |
770 | ip->i_cowextsize = 0; | |
771 | } | |
8a569d71 DW |
772 | } |
773 | ||
1da177e4 | 774 | /* |
1abcf261 DC |
775 | * Initialise a newly allocated inode and return the in-core inode to the |
776 | * caller locked exclusively. | |
1da177e4 | 777 | */ |
b652afd9 | 778 | int |
1abcf261 | 779 | xfs_init_new_inode( |
f2d40141 | 780 | struct mnt_idmap *idmap, |
1abcf261 DC |
781 | struct xfs_trans *tp, |
782 | struct xfs_inode *pip, | |
783 | xfs_ino_t ino, | |
784 | umode_t mode, | |
785 | xfs_nlink_t nlink, | |
786 | dev_t rdev, | |
787 | prid_t prid, | |
e6a688c3 | 788 | bool init_xattrs, |
1abcf261 | 789 | struct xfs_inode **ipp) |
1da177e4 | 790 | { |
01ea173e | 791 | struct inode *dir = pip ? VFS_I(pip) : NULL; |
1abcf261 DC |
792 | struct xfs_mount *mp = tp->t_mountp; |
793 | struct xfs_inode *ip; | |
794 | unsigned int flags; | |
795 | int error; | |
796 | struct timespec64 tv; | |
797 | struct inode *inode; | |
1da177e4 | 798 | |
8b26984d DC |
799 | /* |
800 | * Protect against obviously corrupt allocation btree records. Later | |
801 | * xfs_iget checks will catch re-allocation of other active in-memory | |
802 | * and on-disk inodes. If we don't catch reallocating the parent inode | |
803 | * here we will deadlock in xfs_iget() so we have to do these checks | |
804 | * first. | |
805 | */ | |
806 | if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) { | |
807 | xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino); | |
808 | return -EFSCORRUPTED; | |
809 | } | |
810 | ||
1da177e4 | 811 | /* |
1abcf261 DC |
812 | * Get the in-core inode with the lock held exclusively to prevent |
813 | * others from looking at until we're done. | |
1da177e4 | 814 | */ |
1abcf261 | 815 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip); |
bf904248 | 816 | if (error) |
1da177e4 | 817 | return error; |
1abcf261 | 818 | |
1da177e4 | 819 | ASSERT(ip != NULL); |
3987848c | 820 | inode = VFS_I(ip); |
54d7b5c1 | 821 | set_nlink(inode, nlink); |
66f36464 | 822 | inode->i_rdev = rdev; |
ceaf603c | 823 | ip->i_projid = prid; |
1da177e4 | 824 | |
0560f31a | 825 | if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) { |
c14329d3 | 826 | inode_fsuid_set(inode, idmap); |
01ea173e CH |
827 | inode->i_gid = dir->i_gid; |
828 | inode->i_mode = mode; | |
3d8f2821 | 829 | } else { |
f2d40141 | 830 | inode_init_owner(idmap, inode, dir, mode); |
1da177e4 LT |
831 | } |
832 | ||
833 | /* | |
834 | * If the group ID of the new file does not match the effective group | |
835 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
836 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
837 | */ | |
42b7cc11 | 838 | if (irix_sgid_inherit && (inode->i_mode & S_ISGID) && |
e67fe633 | 839 | !vfsgid_in_group_p(i_gid_into_vfsgid(idmap, inode))) |
c19b3b05 | 840 | inode->i_mode &= ~S_ISGID; |
1da177e4 | 841 | |
13d2c10b | 842 | ip->i_disk_size = 0; |
daf83964 | 843 | ip->i_df.if_nextents = 0; |
6e73a545 | 844 | ASSERT(ip->i_nblocks == 0); |
dff35fd4 | 845 | |
c2050a45 | 846 | tv = current_time(inode); |
3987848c DC |
847 | inode->i_mtime = tv; |
848 | inode->i_atime = tv; | |
849 | inode->i_ctime = tv; | |
dff35fd4 | 850 | |
031474c2 | 851 | ip->i_extsize = 0; |
db07349d | 852 | ip->i_diflags = 0; |
93848a99 | 853 | |
38c26bfd | 854 | if (xfs_has_v3inodes(mp)) { |
f0e28280 | 855 | inode_set_iversion(inode, 1); |
b33ce57d | 856 | ip->i_cowextsize = 0; |
e98d5e88 | 857 | ip->i_crtime = tv; |
93848a99 CH |
858 | } |
859 | ||
1da177e4 LT |
860 | flags = XFS_ILOG_CORE; |
861 | switch (mode & S_IFMT) { | |
862 | case S_IFIFO: | |
863 | case S_IFCHR: | |
864 | case S_IFBLK: | |
865 | case S_IFSOCK: | |
f7e67b20 | 866 | ip->i_df.if_format = XFS_DINODE_FMT_DEV; |
1da177e4 LT |
867 | flags |= XFS_ILOG_DEV; |
868 | break; | |
869 | case S_IFREG: | |
870 | case S_IFDIR: | |
db07349d | 871 | if (pip && (pip->i_diflags & XFS_DIFLAG_ANY)) |
8a569d71 | 872 | xfs_inode_inherit_flags(ip, pip); |
3e09ab8f | 873 | if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY)) |
8a569d71 | 874 | xfs_inode_inherit_flags2(ip, pip); |
53004ee7 | 875 | fallthrough; |
1da177e4 | 876 | case S_IFLNK: |
f7e67b20 | 877 | ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS; |
fcacbc3f | 878 | ip->i_df.if_bytes = 0; |
6bdcf26a | 879 | ip->i_df.if_u1.if_root = NULL; |
1da177e4 LT |
880 | break; |
881 | default: | |
882 | ASSERT(0); | |
883 | } | |
1da177e4 | 884 | |
e6a688c3 DC |
885 | /* |
886 | * If we need to create attributes immediately after allocating the | |
887 | * inode, initialise an empty attribute fork right now. We use the | |
888 | * default fork offset for attributes here as we don't know exactly what | |
889 | * size or how many attributes we might be adding. We can do this | |
890 | * safely here because we know the data fork is completely empty and | |
891 | * this saves us from needing to run a separate transaction to set the | |
892 | * fork offset in the immediate future. | |
893 | */ | |
38c26bfd | 894 | if (init_xattrs && xfs_has_attr(mp)) { |
7821ea30 | 895 | ip->i_forkoff = xfs_default_attroffset(ip) >> 3; |
2ed5b09b | 896 | xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0); |
e6a688c3 DC |
897 | } |
898 | ||
1da177e4 LT |
899 | /* |
900 | * Log the new values stuffed into the inode. | |
901 | */ | |
ddc3415a | 902 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
903 | xfs_trans_log_inode(tp, ip, flags); |
904 | ||
58c90473 | 905 | /* now that we have an i_mode we can setup the inode structure */ |
41be8bed | 906 | xfs_setup_inode(ip); |
1da177e4 LT |
907 | |
908 | *ipp = ip; | |
909 | return 0; | |
910 | } | |
911 | ||
e546cb79 | 912 | /* |
54d7b5c1 DC |
913 | * Decrement the link count on an inode & log the change. If this causes the |
914 | * link count to go to zero, move the inode to AGI unlinked list so that it can | |
915 | * be freed when the last active reference goes away via xfs_inactive(). | |
e546cb79 | 916 | */ |
0d5a75e9 | 917 | static int /* error */ |
e546cb79 DC |
918 | xfs_droplink( |
919 | xfs_trans_t *tp, | |
920 | xfs_inode_t *ip) | |
921 | { | |
e546cb79 DC |
922 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); |
923 | ||
e546cb79 DC |
924 | drop_nlink(VFS_I(ip)); |
925 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
926 | ||
54d7b5c1 DC |
927 | if (VFS_I(ip)->i_nlink) |
928 | return 0; | |
929 | ||
930 | return xfs_iunlink(tp, ip); | |
e546cb79 DC |
931 | } |
932 | ||
e546cb79 DC |
933 | /* |
934 | * Increment the link count on an inode & log the change. | |
935 | */ | |
91083269 | 936 | static void |
e546cb79 DC |
937 | xfs_bumplink( |
938 | xfs_trans_t *tp, | |
939 | xfs_inode_t *ip) | |
940 | { | |
941 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); | |
942 | ||
e546cb79 | 943 | inc_nlink(VFS_I(ip)); |
e546cb79 | 944 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
e546cb79 DC |
945 | } |
946 | ||
c24b5dfa DC |
947 | int |
948 | xfs_create( | |
f2d40141 | 949 | struct mnt_idmap *idmap, |
c24b5dfa DC |
950 | xfs_inode_t *dp, |
951 | struct xfs_name *name, | |
952 | umode_t mode, | |
66f36464 | 953 | dev_t rdev, |
e6a688c3 | 954 | bool init_xattrs, |
c24b5dfa DC |
955 | xfs_inode_t **ipp) |
956 | { | |
957 | int is_dir = S_ISDIR(mode); | |
958 | struct xfs_mount *mp = dp->i_mount; | |
959 | struct xfs_inode *ip = NULL; | |
960 | struct xfs_trans *tp = NULL; | |
961 | int error; | |
c24b5dfa | 962 | bool unlock_dp_on_error = false; |
c24b5dfa DC |
963 | prid_t prid; |
964 | struct xfs_dquot *udqp = NULL; | |
965 | struct xfs_dquot *gdqp = NULL; | |
966 | struct xfs_dquot *pdqp = NULL; | |
062647a8 | 967 | struct xfs_trans_res *tres; |
c24b5dfa | 968 | uint resblks; |
b652afd9 | 969 | xfs_ino_t ino; |
c24b5dfa DC |
970 | |
971 | trace_xfs_create(dp, name); | |
972 | ||
75c8c50f | 973 | if (xfs_is_shutdown(mp)) |
2451337d | 974 | return -EIO; |
c24b5dfa | 975 | |
163467d3 | 976 | prid = xfs_get_initial_prid(dp); |
c24b5dfa DC |
977 | |
978 | /* | |
979 | * Make sure that we have allocated dquot(s) on disk. | |
980 | */ | |
c14329d3 CB |
981 | error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns), |
982 | mapped_fsgid(idmap, &init_user_ns), prid, | |
b5a08423 DW |
983 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
984 | &udqp, &gdqp, &pdqp); | |
c24b5dfa DC |
985 | if (error) |
986 | return error; | |
987 | ||
988 | if (is_dir) { | |
c24b5dfa | 989 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); |
062647a8 | 990 | tres = &M_RES(mp)->tr_mkdir; |
c24b5dfa DC |
991 | } else { |
992 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
062647a8 | 993 | tres = &M_RES(mp)->tr_create; |
c24b5dfa DC |
994 | } |
995 | ||
c24b5dfa DC |
996 | /* |
997 | * Initially assume that the file does not exist and | |
998 | * reserve the resources for that case. If that is not | |
999 | * the case we'll drop the one we have and get a more | |
1000 | * appropriate transaction later. | |
1001 | */ | |
f2f7b9ff DW |
1002 | error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks, |
1003 | &tp); | |
2451337d | 1004 | if (error == -ENOSPC) { |
c24b5dfa DC |
1005 | /* flush outstanding delalloc blocks and retry */ |
1006 | xfs_flush_inodes(mp); | |
f2f7b9ff DW |
1007 | error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, |
1008 | resblks, &tp); | |
c24b5dfa | 1009 | } |
4906e215 | 1010 | if (error) |
f2f7b9ff | 1011 | goto out_release_dquots; |
c24b5dfa | 1012 | |
65523218 | 1013 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
c24b5dfa DC |
1014 | unlock_dp_on_error = true; |
1015 | ||
c24b5dfa DC |
1016 | /* |
1017 | * A newly created regular or special file just has one directory | |
1018 | * entry pointing to them, but a directory also the "." entry | |
1019 | * pointing to itself. | |
1020 | */ | |
b652afd9 DC |
1021 | error = xfs_dialloc(&tp, dp->i_ino, mode, &ino); |
1022 | if (!error) | |
f2d40141 | 1023 | error = xfs_init_new_inode(idmap, tp, dp, ino, mode, |
b652afd9 | 1024 | is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip); |
d6077aa3 | 1025 | if (error) |
4906e215 | 1026 | goto out_trans_cancel; |
c24b5dfa DC |
1027 | |
1028 | /* | |
1029 | * Now we join the directory inode to the transaction. We do not do it | |
b652afd9 | 1030 | * earlier because xfs_dialloc might commit the previous transaction |
c24b5dfa DC |
1031 | * (and release all the locks). An error from here on will result in |
1032 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1033 | * error path. | |
1034 | */ | |
65523218 | 1035 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1036 | unlock_dp_on_error = false; |
1037 | ||
381eee69 | 1038 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
63337b63 | 1039 | resblks - XFS_IALLOC_SPACE_RES(mp)); |
c24b5dfa | 1040 | if (error) { |
2451337d | 1041 | ASSERT(error != -ENOSPC); |
4906e215 | 1042 | goto out_trans_cancel; |
c24b5dfa DC |
1043 | } |
1044 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1045 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1046 | ||
1047 | if (is_dir) { | |
1048 | error = xfs_dir_init(tp, ip, dp); | |
1049 | if (error) | |
c8eac49e | 1050 | goto out_trans_cancel; |
c24b5dfa | 1051 | |
91083269 | 1052 | xfs_bumplink(tp, dp); |
c24b5dfa DC |
1053 | } |
1054 | ||
1055 | /* | |
1056 | * If this is a synchronous mount, make sure that the | |
1057 | * create transaction goes to disk before returning to | |
1058 | * the user. | |
1059 | */ | |
0560f31a | 1060 | if (xfs_has_wsync(mp) || xfs_has_dirsync(mp)) |
c24b5dfa DC |
1061 | xfs_trans_set_sync(tp); |
1062 | ||
1063 | /* | |
1064 | * Attach the dquot(s) to the inodes and modify them incore. | |
1065 | * These ids of the inode couldn't have changed since the new | |
1066 | * inode has been locked ever since it was created. | |
1067 | */ | |
1068 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1069 | ||
70393313 | 1070 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
1071 | if (error) |
1072 | goto out_release_inode; | |
1073 | ||
1074 | xfs_qm_dqrele(udqp); | |
1075 | xfs_qm_dqrele(gdqp); | |
1076 | xfs_qm_dqrele(pdqp); | |
1077 | ||
1078 | *ipp = ip; | |
1079 | return 0; | |
1080 | ||
c24b5dfa | 1081 | out_trans_cancel: |
4906e215 | 1082 | xfs_trans_cancel(tp); |
c24b5dfa DC |
1083 | out_release_inode: |
1084 | /* | |
58c90473 DC |
1085 | * Wait until after the current transaction is aborted to finish the |
1086 | * setup of the inode and release the inode. This prevents recursive | |
1087 | * transactions and deadlocks from xfs_inactive. | |
c24b5dfa | 1088 | */ |
58c90473 DC |
1089 | if (ip) { |
1090 | xfs_finish_inode_setup(ip); | |
44a8736b | 1091 | xfs_irele(ip); |
58c90473 | 1092 | } |
f2f7b9ff | 1093 | out_release_dquots: |
c24b5dfa DC |
1094 | xfs_qm_dqrele(udqp); |
1095 | xfs_qm_dqrele(gdqp); | |
1096 | xfs_qm_dqrele(pdqp); | |
1097 | ||
1098 | if (unlock_dp_on_error) | |
65523218 | 1099 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
c24b5dfa DC |
1100 | return error; |
1101 | } | |
1102 | ||
99b6436b ZYW |
1103 | int |
1104 | xfs_create_tmpfile( | |
f2d40141 | 1105 | struct mnt_idmap *idmap, |
99b6436b | 1106 | struct xfs_inode *dp, |
330033d6 BF |
1107 | umode_t mode, |
1108 | struct xfs_inode **ipp) | |
99b6436b ZYW |
1109 | { |
1110 | struct xfs_mount *mp = dp->i_mount; | |
1111 | struct xfs_inode *ip = NULL; | |
1112 | struct xfs_trans *tp = NULL; | |
1113 | int error; | |
99b6436b ZYW |
1114 | prid_t prid; |
1115 | struct xfs_dquot *udqp = NULL; | |
1116 | struct xfs_dquot *gdqp = NULL; | |
1117 | struct xfs_dquot *pdqp = NULL; | |
1118 | struct xfs_trans_res *tres; | |
1119 | uint resblks; | |
b652afd9 | 1120 | xfs_ino_t ino; |
99b6436b | 1121 | |
75c8c50f | 1122 | if (xfs_is_shutdown(mp)) |
2451337d | 1123 | return -EIO; |
99b6436b ZYW |
1124 | |
1125 | prid = xfs_get_initial_prid(dp); | |
1126 | ||
1127 | /* | |
1128 | * Make sure that we have allocated dquot(s) on disk. | |
1129 | */ | |
c14329d3 CB |
1130 | error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns), |
1131 | mapped_fsgid(idmap, &init_user_ns), prid, | |
b5a08423 DW |
1132 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
1133 | &udqp, &gdqp, &pdqp); | |
99b6436b ZYW |
1134 | if (error) |
1135 | return error; | |
1136 | ||
1137 | resblks = XFS_IALLOC_SPACE_RES(mp); | |
99b6436b | 1138 | tres = &M_RES(mp)->tr_create_tmpfile; |
253f4911 | 1139 | |
f2f7b9ff DW |
1140 | error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks, |
1141 | &tp); | |
99b6436b | 1142 | if (error) |
f2f7b9ff | 1143 | goto out_release_dquots; |
99b6436b | 1144 | |
b652afd9 DC |
1145 | error = xfs_dialloc(&tp, dp->i_ino, mode, &ino); |
1146 | if (!error) | |
f2d40141 | 1147 | error = xfs_init_new_inode(idmap, tp, dp, ino, mode, |
b652afd9 | 1148 | 0, 0, prid, false, &ip); |
d6077aa3 | 1149 | if (error) |
4906e215 | 1150 | goto out_trans_cancel; |
99b6436b | 1151 | |
0560f31a | 1152 | if (xfs_has_wsync(mp)) |
99b6436b ZYW |
1153 | xfs_trans_set_sync(tp); |
1154 | ||
1155 | /* | |
1156 | * Attach the dquot(s) to the inodes and modify them incore. | |
1157 | * These ids of the inode couldn't have changed since the new | |
1158 | * inode has been locked ever since it was created. | |
1159 | */ | |
1160 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1161 | ||
99b6436b ZYW |
1162 | error = xfs_iunlink(tp, ip); |
1163 | if (error) | |
4906e215 | 1164 | goto out_trans_cancel; |
99b6436b | 1165 | |
70393313 | 1166 | error = xfs_trans_commit(tp); |
99b6436b ZYW |
1167 | if (error) |
1168 | goto out_release_inode; | |
1169 | ||
1170 | xfs_qm_dqrele(udqp); | |
1171 | xfs_qm_dqrele(gdqp); | |
1172 | xfs_qm_dqrele(pdqp); | |
1173 | ||
330033d6 | 1174 | *ipp = ip; |
99b6436b ZYW |
1175 | return 0; |
1176 | ||
99b6436b | 1177 | out_trans_cancel: |
4906e215 | 1178 | xfs_trans_cancel(tp); |
99b6436b ZYW |
1179 | out_release_inode: |
1180 | /* | |
58c90473 DC |
1181 | * Wait until after the current transaction is aborted to finish the |
1182 | * setup of the inode and release the inode. This prevents recursive | |
1183 | * transactions and deadlocks from xfs_inactive. | |
99b6436b | 1184 | */ |
58c90473 DC |
1185 | if (ip) { |
1186 | xfs_finish_inode_setup(ip); | |
44a8736b | 1187 | xfs_irele(ip); |
58c90473 | 1188 | } |
f2f7b9ff | 1189 | out_release_dquots: |
99b6436b ZYW |
1190 | xfs_qm_dqrele(udqp); |
1191 | xfs_qm_dqrele(gdqp); | |
1192 | xfs_qm_dqrele(pdqp); | |
1193 | ||
1194 | return error; | |
1195 | } | |
1196 | ||
c24b5dfa DC |
1197 | int |
1198 | xfs_link( | |
1199 | xfs_inode_t *tdp, | |
1200 | xfs_inode_t *sip, | |
1201 | struct xfs_name *target_name) | |
1202 | { | |
1203 | xfs_mount_t *mp = tdp->i_mount; | |
1204 | xfs_trans_t *tp; | |
871b9316 | 1205 | int error, nospace_error = 0; |
c24b5dfa DC |
1206 | int resblks; |
1207 | ||
1208 | trace_xfs_link(tdp, target_name); | |
1209 | ||
c19b3b05 | 1210 | ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); |
c24b5dfa | 1211 | |
75c8c50f | 1212 | if (xfs_is_shutdown(mp)) |
2451337d | 1213 | return -EIO; |
c24b5dfa | 1214 | |
c14cfcca | 1215 | error = xfs_qm_dqattach(sip); |
c24b5dfa DC |
1216 | if (error) |
1217 | goto std_return; | |
1218 | ||
c14cfcca | 1219 | error = xfs_qm_dqattach(tdp); |
c24b5dfa DC |
1220 | if (error) |
1221 | goto std_return; | |
1222 | ||
c24b5dfa | 1223 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
871b9316 DW |
1224 | error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks, |
1225 | &tp, &nospace_error); | |
4906e215 | 1226 | if (error) |
253f4911 | 1227 | goto std_return; |
c24b5dfa | 1228 | |
c24b5dfa DC |
1229 | /* |
1230 | * If we are using project inheritance, we only allow hard link | |
1231 | * creation in our tree when the project IDs are the same; else | |
1232 | * the tree quota mechanism could be circumvented. | |
1233 | */ | |
db07349d | 1234 | if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) && |
ceaf603c | 1235 | tdp->i_projid != sip->i_projid)) { |
2451337d | 1236 | error = -EXDEV; |
c24b5dfa DC |
1237 | goto error_return; |
1238 | } | |
1239 | ||
94f3cad5 ES |
1240 | if (!resblks) { |
1241 | error = xfs_dir_canenter(tp, tdp, target_name); | |
1242 | if (error) | |
1243 | goto error_return; | |
1244 | } | |
c24b5dfa | 1245 | |
54d7b5c1 DC |
1246 | /* |
1247 | * Handle initial link state of O_TMPFILE inode | |
1248 | */ | |
1249 | if (VFS_I(sip)->i_nlink == 0) { | |
f40aadb2 DC |
1250 | struct xfs_perag *pag; |
1251 | ||
1252 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino)); | |
1253 | error = xfs_iunlink_remove(tp, pag, sip); | |
1254 | xfs_perag_put(pag); | |
ab297431 | 1255 | if (error) |
4906e215 | 1256 | goto error_return; |
ab297431 ZYW |
1257 | } |
1258 | ||
c24b5dfa | 1259 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
381eee69 | 1260 | resblks); |
c24b5dfa | 1261 | if (error) |
4906e215 | 1262 | goto error_return; |
c24b5dfa DC |
1263 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
1264 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
1265 | ||
91083269 | 1266 | xfs_bumplink(tp, sip); |
c24b5dfa DC |
1267 | |
1268 | /* | |
1269 | * If this is a synchronous mount, make sure that the | |
1270 | * link transaction goes to disk before returning to | |
1271 | * the user. | |
1272 | */ | |
0560f31a | 1273 | if (xfs_has_wsync(mp) || xfs_has_dirsync(mp)) |
c24b5dfa | 1274 | xfs_trans_set_sync(tp); |
c24b5dfa | 1275 | |
70393313 | 1276 | return xfs_trans_commit(tp); |
c24b5dfa | 1277 | |
c24b5dfa | 1278 | error_return: |
4906e215 | 1279 | xfs_trans_cancel(tp); |
c24b5dfa | 1280 | std_return: |
871b9316 DW |
1281 | if (error == -ENOSPC && nospace_error) |
1282 | error = nospace_error; | |
c24b5dfa DC |
1283 | return error; |
1284 | } | |
1285 | ||
363e59ba DW |
1286 | /* Clear the reflink flag and the cowblocks tag if possible. */ |
1287 | static void | |
1288 | xfs_itruncate_clear_reflink_flags( | |
1289 | struct xfs_inode *ip) | |
1290 | { | |
1291 | struct xfs_ifork *dfork; | |
1292 | struct xfs_ifork *cfork; | |
1293 | ||
1294 | if (!xfs_is_reflink_inode(ip)) | |
1295 | return; | |
732436ef DW |
1296 | dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK); |
1297 | cfork = xfs_ifork_ptr(ip, XFS_COW_FORK); | |
363e59ba | 1298 | if (dfork->if_bytes == 0 && cfork->if_bytes == 0) |
3e09ab8f | 1299 | ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK; |
363e59ba DW |
1300 | if (cfork->if_bytes == 0) |
1301 | xfs_inode_clear_cowblocks_tag(ip); | |
1302 | } | |
1303 | ||
1da177e4 | 1304 | /* |
8f04c47a CH |
1305 | * Free up the underlying blocks past new_size. The new size must be smaller |
1306 | * than the current size. This routine can be used both for the attribute and | |
1307 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1308 | * |
f6485057 DC |
1309 | * The transaction passed to this routine must have made a permanent log |
1310 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1311 | * given transaction and start new ones, so make sure everything involved in | |
1312 | * the transaction is tidy before calling here. Some transaction will be | |
1313 | * returned to the caller to be committed. The incoming transaction must | |
1314 | * already include the inode, and both inode locks must be held exclusively. | |
1315 | * The inode must also be "held" within the transaction. On return the inode | |
1316 | * will be "held" within the returned transaction. This routine does NOT | |
1317 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1318 | * |
f6485057 DC |
1319 | * If we get an error, we must return with the inode locked and linked into the |
1320 | * current transaction. This keeps things simple for the higher level code, | |
1321 | * because it always knows that the inode is locked and held in the transaction | |
1322 | * that returns to it whether errors occur or not. We don't mark the inode | |
1323 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1324 | */ |
1325 | int | |
4e529339 | 1326 | xfs_itruncate_extents_flags( |
8f04c47a CH |
1327 | struct xfs_trans **tpp, |
1328 | struct xfs_inode *ip, | |
1329 | int whichfork, | |
13b86fc3 | 1330 | xfs_fsize_t new_size, |
4e529339 | 1331 | int flags) |
1da177e4 | 1332 | { |
8f04c47a CH |
1333 | struct xfs_mount *mp = ip->i_mount; |
1334 | struct xfs_trans *tp = *tpp; | |
8f04c47a | 1335 | xfs_fileoff_t first_unmap_block; |
8f04c47a | 1336 | xfs_filblks_t unmap_len; |
8f04c47a | 1337 | int error = 0; |
1da177e4 | 1338 | |
0b56185b CH |
1339 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1340 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1341 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1342 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1343 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1344 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1345 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1346 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1347 | |
673e8e59 CH |
1348 | trace_xfs_itruncate_extents_start(ip, new_size); |
1349 | ||
4e529339 | 1350 | flags |= xfs_bmapi_aflag(whichfork); |
13b86fc3 | 1351 | |
1da177e4 LT |
1352 | /* |
1353 | * Since it is possible for space to become allocated beyond | |
1354 | * the end of the file (in a crash where the space is allocated | |
1355 | * but the inode size is not yet updated), simply remove any | |
1356 | * blocks which show up between the new EOF and the maximum | |
4bbb04ab DW |
1357 | * possible file size. |
1358 | * | |
1359 | * We have to free all the blocks to the bmbt maximum offset, even if | |
1360 | * the page cache can't scale that far. | |
1da177e4 | 1361 | */ |
8f04c47a | 1362 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
33005fd0 | 1363 | if (!xfs_verify_fileoff(mp, first_unmap_block)) { |
4bbb04ab | 1364 | WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF); |
8f04c47a | 1365 | return 0; |
4bbb04ab | 1366 | } |
8f04c47a | 1367 | |
4bbb04ab DW |
1368 | unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1; |
1369 | while (unmap_len > 0) { | |
692b6cdd | 1370 | ASSERT(tp->t_highest_agno == NULLAGNUMBER); |
4bbb04ab DW |
1371 | error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len, |
1372 | flags, XFS_ITRUNC_MAX_EXTENTS); | |
8f04c47a | 1373 | if (error) |
d5a2e289 | 1374 | goto out; |
1da177e4 | 1375 | |
6dd379c7 | 1376 | /* free the just unmapped extents */ |
9e28a242 | 1377 | error = xfs_defer_finish(&tp); |
8f04c47a | 1378 | if (error) |
9b1f4e98 | 1379 | goto out; |
1da177e4 | 1380 | } |
8f04c47a | 1381 | |
4919d42a DW |
1382 | if (whichfork == XFS_DATA_FORK) { |
1383 | /* Remove all pending CoW reservations. */ | |
1384 | error = xfs_reflink_cancel_cow_blocks(ip, &tp, | |
4bbb04ab | 1385 | first_unmap_block, XFS_MAX_FILEOFF, true); |
4919d42a DW |
1386 | if (error) |
1387 | goto out; | |
aa8968f2 | 1388 | |
4919d42a DW |
1389 | xfs_itruncate_clear_reflink_flags(ip); |
1390 | } | |
aa8968f2 | 1391 | |
673e8e59 CH |
1392 | /* |
1393 | * Always re-log the inode so that our permanent transaction can keep | |
1394 | * on rolling it forward in the log. | |
1395 | */ | |
1396 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1397 | ||
1398 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1399 | ||
8f04c47a CH |
1400 | out: |
1401 | *tpp = tp; | |
1402 | return error; | |
8f04c47a CH |
1403 | } |
1404 | ||
c24b5dfa DC |
1405 | int |
1406 | xfs_release( | |
1407 | xfs_inode_t *ip) | |
1408 | { | |
1409 | xfs_mount_t *mp = ip->i_mount; | |
7d88329e | 1410 | int error = 0; |
c24b5dfa | 1411 | |
c19b3b05 | 1412 | if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) |
c24b5dfa DC |
1413 | return 0; |
1414 | ||
1415 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
2e973b2c | 1416 | if (xfs_is_readonly(mp)) |
c24b5dfa DC |
1417 | return 0; |
1418 | ||
75c8c50f | 1419 | if (!xfs_is_shutdown(mp)) { |
c24b5dfa DC |
1420 | int truncated; |
1421 | ||
c24b5dfa DC |
1422 | /* |
1423 | * If we previously truncated this file and removed old data | |
1424 | * in the process, we want to initiate "early" writeout on | |
1425 | * the last close. This is an attempt to combat the notorious | |
1426 | * NULL files problem which is particularly noticeable from a | |
1427 | * truncate down, buffered (re-)write (delalloc), followed by | |
1428 | * a crash. What we are effectively doing here is | |
1429 | * significantly reducing the time window where we'd otherwise | |
1430 | * be exposed to that problem. | |
1431 | */ | |
1432 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); | |
1433 | if (truncated) { | |
1434 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
eac152b4 | 1435 | if (ip->i_delayed_blks > 0) { |
2451337d | 1436 | error = filemap_flush(VFS_I(ip)->i_mapping); |
c24b5dfa DC |
1437 | if (error) |
1438 | return error; | |
1439 | } | |
1440 | } | |
1441 | } | |
1442 | ||
54d7b5c1 | 1443 | if (VFS_I(ip)->i_nlink == 0) |
c24b5dfa DC |
1444 | return 0; |
1445 | ||
7d88329e DW |
1446 | /* |
1447 | * If we can't get the iolock just skip truncating the blocks past EOF | |
1448 | * because we could deadlock with the mmap_lock otherwise. We'll get | |
1449 | * another chance to drop them once the last reference to the inode is | |
1450 | * dropped, so we'll never leak blocks permanently. | |
1451 | */ | |
1452 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) | |
1453 | return 0; | |
c24b5dfa | 1454 | |
7d88329e | 1455 | if (xfs_can_free_eofblocks(ip, false)) { |
a36b9261 BF |
1456 | /* |
1457 | * Check if the inode is being opened, written and closed | |
1458 | * frequently and we have delayed allocation blocks outstanding | |
1459 | * (e.g. streaming writes from the NFS server), truncating the | |
1460 | * blocks past EOF will cause fragmentation to occur. | |
1461 | * | |
1462 | * In this case don't do the truncation, but we have to be | |
1463 | * careful how we detect this case. Blocks beyond EOF show up as | |
1464 | * i_delayed_blks even when the inode is clean, so we need to | |
1465 | * truncate them away first before checking for a dirty release. | |
1466 | * Hence on the first dirty close we will still remove the | |
1467 | * speculative allocation, but after that we will leave it in | |
1468 | * place. | |
1469 | */ | |
1470 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
7d88329e DW |
1471 | goto out_unlock; |
1472 | ||
1473 | error = xfs_free_eofblocks(ip); | |
1474 | if (error) | |
1475 | goto out_unlock; | |
c24b5dfa DC |
1476 | |
1477 | /* delalloc blocks after truncation means it really is dirty */ | |
1478 | if (ip->i_delayed_blks) | |
1479 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
1480 | } | |
7d88329e DW |
1481 | |
1482 | out_unlock: | |
1483 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
1484 | return error; | |
c24b5dfa DC |
1485 | } |
1486 | ||
f7be2d7f BF |
1487 | /* |
1488 | * xfs_inactive_truncate | |
1489 | * | |
1490 | * Called to perform a truncate when an inode becomes unlinked. | |
1491 | */ | |
1492 | STATIC int | |
1493 | xfs_inactive_truncate( | |
1494 | struct xfs_inode *ip) | |
1495 | { | |
1496 | struct xfs_mount *mp = ip->i_mount; | |
1497 | struct xfs_trans *tp; | |
1498 | int error; | |
1499 | ||
253f4911 | 1500 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); |
f7be2d7f | 1501 | if (error) { |
75c8c50f | 1502 | ASSERT(xfs_is_shutdown(mp)); |
f7be2d7f BF |
1503 | return error; |
1504 | } | |
f7be2d7f BF |
1505 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1506 | xfs_trans_ijoin(tp, ip, 0); | |
1507 | ||
1508 | /* | |
1509 | * Log the inode size first to prevent stale data exposure in the event | |
1510 | * of a system crash before the truncate completes. See the related | |
69bca807 | 1511 | * comment in xfs_vn_setattr_size() for details. |
f7be2d7f | 1512 | */ |
13d2c10b | 1513 | ip->i_disk_size = 0; |
f7be2d7f BF |
1514 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1515 | ||
1516 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
1517 | if (error) | |
1518 | goto error_trans_cancel; | |
1519 | ||
daf83964 | 1520 | ASSERT(ip->i_df.if_nextents == 0); |
f7be2d7f | 1521 | |
70393313 | 1522 | error = xfs_trans_commit(tp); |
f7be2d7f BF |
1523 | if (error) |
1524 | goto error_unlock; | |
1525 | ||
1526 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1527 | return 0; | |
1528 | ||
1529 | error_trans_cancel: | |
4906e215 | 1530 | xfs_trans_cancel(tp); |
f7be2d7f BF |
1531 | error_unlock: |
1532 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1533 | return error; | |
1534 | } | |
1535 | ||
88877d2b BF |
1536 | /* |
1537 | * xfs_inactive_ifree() | |
1538 | * | |
1539 | * Perform the inode free when an inode is unlinked. | |
1540 | */ | |
1541 | STATIC int | |
1542 | xfs_inactive_ifree( | |
1543 | struct xfs_inode *ip) | |
1544 | { | |
88877d2b BF |
1545 | struct xfs_mount *mp = ip->i_mount; |
1546 | struct xfs_trans *tp; | |
1547 | int error; | |
1548 | ||
9d43b180 | 1549 | /* |
76d771b4 CH |
1550 | * We try to use a per-AG reservation for any block needed by the finobt |
1551 | * tree, but as the finobt feature predates the per-AG reservation | |
1552 | * support a degraded file system might not have enough space for the | |
1553 | * reservation at mount time. In that case try to dip into the reserved | |
1554 | * pool and pray. | |
9d43b180 BF |
1555 | * |
1556 | * Send a warning if the reservation does happen to fail, as the inode | |
1557 | * now remains allocated and sits on the unlinked list until the fs is | |
1558 | * repaired. | |
1559 | */ | |
e1f6ca11 | 1560 | if (unlikely(mp->m_finobt_nores)) { |
76d771b4 CH |
1561 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, |
1562 | XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, | |
1563 | &tp); | |
1564 | } else { | |
1565 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); | |
1566 | } | |
88877d2b | 1567 | if (error) { |
2451337d | 1568 | if (error == -ENOSPC) { |
9d43b180 BF |
1569 | xfs_warn_ratelimited(mp, |
1570 | "Failed to remove inode(s) from unlinked list. " | |
1571 | "Please free space, unmount and run xfs_repair."); | |
1572 | } else { | |
75c8c50f | 1573 | ASSERT(xfs_is_shutdown(mp)); |
9d43b180 | 1574 | } |
88877d2b BF |
1575 | return error; |
1576 | } | |
1577 | ||
96355d5a DC |
1578 | /* |
1579 | * We do not hold the inode locked across the entire rolling transaction | |
1580 | * here. We only need to hold it for the first transaction that | |
1581 | * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the | |
1582 | * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode | |
1583 | * here breaks the relationship between cluster buffer invalidation and | |
1584 | * stale inode invalidation on cluster buffer item journal commit | |
1585 | * completion, and can result in leaving dirty stale inodes hanging | |
1586 | * around in memory. | |
1587 | * | |
1588 | * We have no need for serialising this inode operation against other | |
1589 | * operations - we freed the inode and hence reallocation is required | |
1590 | * and that will serialise on reallocating the space the deferops need | |
1591 | * to free. Hence we can unlock the inode on the first commit of | |
1592 | * the transaction rather than roll it right through the deferops. This | |
1593 | * avoids relogging the XFS_ISTALE inode. | |
1594 | * | |
1595 | * We check that xfs_ifree() hasn't grown an internal transaction roll | |
1596 | * by asserting that the inode is still locked when it returns. | |
1597 | */ | |
88877d2b | 1598 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
96355d5a | 1599 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
88877d2b | 1600 | |
0e0417f3 | 1601 | error = xfs_ifree(tp, ip); |
96355d5a | 1602 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
88877d2b BF |
1603 | if (error) { |
1604 | /* | |
1605 | * If we fail to free the inode, shut down. The cancel | |
1606 | * might do that, we need to make sure. Otherwise the | |
1607 | * inode might be lost for a long time or forever. | |
1608 | */ | |
75c8c50f | 1609 | if (!xfs_is_shutdown(mp)) { |
88877d2b BF |
1610 | xfs_notice(mp, "%s: xfs_ifree returned error %d", |
1611 | __func__, error); | |
1612 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1613 | } | |
4906e215 | 1614 | xfs_trans_cancel(tp); |
88877d2b BF |
1615 | return error; |
1616 | } | |
1617 | ||
1618 | /* | |
1619 | * Credit the quota account(s). The inode is gone. | |
1620 | */ | |
1621 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1622 | ||
1623 | /* | |
d4a97a04 BF |
1624 | * Just ignore errors at this point. There is nothing we can do except |
1625 | * to try to keep going. Make sure it's not a silent error. | |
88877d2b | 1626 | */ |
70393313 | 1627 | error = xfs_trans_commit(tp); |
88877d2b BF |
1628 | if (error) |
1629 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", | |
1630 | __func__, error); | |
1631 | ||
88877d2b BF |
1632 | return 0; |
1633 | } | |
1634 | ||
62af7d54 DW |
1635 | /* |
1636 | * Returns true if we need to update the on-disk metadata before we can free | |
1637 | * the memory used by this inode. Updates include freeing post-eof | |
1638 | * preallocations; freeing COW staging extents; and marking the inode free in | |
1639 | * the inobt if it is on the unlinked list. | |
1640 | */ | |
1641 | bool | |
1642 | xfs_inode_needs_inactive( | |
1643 | struct xfs_inode *ip) | |
1644 | { | |
1645 | struct xfs_mount *mp = ip->i_mount; | |
732436ef | 1646 | struct xfs_ifork *cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK); |
62af7d54 DW |
1647 | |
1648 | /* | |
1649 | * If the inode is already free, then there can be nothing | |
1650 | * to clean up here. | |
1651 | */ | |
1652 | if (VFS_I(ip)->i_mode == 0) | |
1653 | return false; | |
1654 | ||
1655 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
2e973b2c | 1656 | if (xfs_is_readonly(mp)) |
62af7d54 DW |
1657 | return false; |
1658 | ||
1659 | /* If the log isn't running, push inodes straight to reclaim. */ | |
75c8c50f | 1660 | if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp)) |
62af7d54 DW |
1661 | return false; |
1662 | ||
1663 | /* Metadata inodes require explicit resource cleanup. */ | |
1664 | if (xfs_is_metadata_inode(ip)) | |
1665 | return false; | |
1666 | ||
1667 | /* Want to clean out the cow blocks if there are any. */ | |
1668 | if (cow_ifp && cow_ifp->if_bytes > 0) | |
1669 | return true; | |
1670 | ||
1671 | /* Unlinked files must be freed. */ | |
1672 | if (VFS_I(ip)->i_nlink == 0) | |
1673 | return true; | |
1674 | ||
1675 | /* | |
1676 | * This file isn't being freed, so check if there are post-eof blocks | |
1677 | * to free. @force is true because we are evicting an inode from the | |
1678 | * cache. Post-eof blocks must be freed, lest we end up with broken | |
1679 | * free space accounting. | |
1680 | * | |
1681 | * Note: don't bother with iolock here since lockdep complains about | |
1682 | * acquiring it in reclaim context. We have the only reference to the | |
1683 | * inode at this point anyways. | |
1684 | */ | |
1685 | return xfs_can_free_eofblocks(ip, true); | |
1686 | } | |
1687 | ||
c24b5dfa DC |
1688 | /* |
1689 | * xfs_inactive | |
1690 | * | |
1691 | * This is called when the vnode reference count for the vnode | |
1692 | * goes to zero. If the file has been unlinked, then it must | |
1693 | * now be truncated. Also, we clear all of the read-ahead state | |
1694 | * kept for the inode here since the file is now closed. | |
1695 | */ | |
74564fb4 | 1696 | void |
c24b5dfa DC |
1697 | xfs_inactive( |
1698 | xfs_inode_t *ip) | |
1699 | { | |
3d3c8b52 | 1700 | struct xfs_mount *mp; |
3d3c8b52 JL |
1701 | int error; |
1702 | int truncate = 0; | |
c24b5dfa DC |
1703 | |
1704 | /* | |
1705 | * If the inode is already free, then there can be nothing | |
1706 | * to clean up here. | |
1707 | */ | |
c19b3b05 | 1708 | if (VFS_I(ip)->i_mode == 0) { |
c24b5dfa | 1709 | ASSERT(ip->i_df.if_broot_bytes == 0); |
3ea06d73 | 1710 | goto out; |
c24b5dfa DC |
1711 | } |
1712 | ||
1713 | mp = ip->i_mount; | |
17c12bcd | 1714 | ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); |
c24b5dfa | 1715 | |
c24b5dfa | 1716 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
2e973b2c | 1717 | if (xfs_is_readonly(mp)) |
3ea06d73 | 1718 | goto out; |
c24b5dfa | 1719 | |
383e32b0 DW |
1720 | /* Metadata inodes require explicit resource cleanup. */ |
1721 | if (xfs_is_metadata_inode(ip)) | |
3ea06d73 | 1722 | goto out; |
383e32b0 | 1723 | |
6231848c | 1724 | /* Try to clean out the cow blocks if there are any. */ |
51d62690 | 1725 | if (xfs_inode_has_cow_data(ip)) |
6231848c DW |
1726 | xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true); |
1727 | ||
54d7b5c1 | 1728 | if (VFS_I(ip)->i_nlink != 0) { |
c24b5dfa DC |
1729 | /* |
1730 | * force is true because we are evicting an inode from the | |
1731 | * cache. Post-eof blocks must be freed, lest we end up with | |
1732 | * broken free space accounting. | |
3b4683c2 BF |
1733 | * |
1734 | * Note: don't bother with iolock here since lockdep complains | |
1735 | * about acquiring it in reclaim context. We have the only | |
1736 | * reference to the inode at this point anyways. | |
c24b5dfa | 1737 | */ |
3b4683c2 | 1738 | if (xfs_can_free_eofblocks(ip, true)) |
a36b9261 | 1739 | xfs_free_eofblocks(ip); |
74564fb4 | 1740 | |
3ea06d73 | 1741 | goto out; |
c24b5dfa DC |
1742 | } |
1743 | ||
c19b3b05 | 1744 | if (S_ISREG(VFS_I(ip)->i_mode) && |
13d2c10b | 1745 | (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 || |
daf83964 | 1746 | ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0)) |
c24b5dfa DC |
1747 | truncate = 1; |
1748 | ||
c14cfcca | 1749 | error = xfs_qm_dqattach(ip); |
c24b5dfa | 1750 | if (error) |
3ea06d73 | 1751 | goto out; |
c24b5dfa | 1752 | |
c19b3b05 | 1753 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
36b21dde | 1754 | error = xfs_inactive_symlink(ip); |
f7be2d7f BF |
1755 | else if (truncate) |
1756 | error = xfs_inactive_truncate(ip); | |
1757 | if (error) | |
3ea06d73 | 1758 | goto out; |
c24b5dfa DC |
1759 | |
1760 | /* | |
1761 | * If there are attributes associated with the file then blow them away | |
1762 | * now. The code calls a routine that recursively deconstructs the | |
6dfe5a04 | 1763 | * attribute fork. If also blows away the in-core attribute fork. |
c24b5dfa | 1764 | */ |
932b42c6 | 1765 | if (xfs_inode_has_attr_fork(ip)) { |
c24b5dfa DC |
1766 | error = xfs_attr_inactive(ip); |
1767 | if (error) | |
3ea06d73 | 1768 | goto out; |
c24b5dfa DC |
1769 | } |
1770 | ||
7821ea30 | 1771 | ASSERT(ip->i_forkoff == 0); |
c24b5dfa DC |
1772 | |
1773 | /* | |
1774 | * Free the inode. | |
1775 | */ | |
3ea06d73 | 1776 | xfs_inactive_ifree(ip); |
c24b5dfa | 1777 | |
3ea06d73 | 1778 | out: |
c24b5dfa | 1779 | /* |
3ea06d73 DW |
1780 | * We're done making metadata updates for this inode, so we can release |
1781 | * the attached dquots. | |
c24b5dfa DC |
1782 | */ |
1783 | xfs_qm_dqdetach(ip); | |
c24b5dfa DC |
1784 | } |
1785 | ||
9b247179 DW |
1786 | /* |
1787 | * In-Core Unlinked List Lookups | |
1788 | * ============================= | |
1789 | * | |
1790 | * Every inode is supposed to be reachable from some other piece of metadata | |
1791 | * with the exception of the root directory. Inodes with a connection to a | |
1792 | * file descriptor but not linked from anywhere in the on-disk directory tree | |
1793 | * are collectively known as unlinked inodes, though the filesystem itself | |
1794 | * maintains links to these inodes so that on-disk metadata are consistent. | |
1795 | * | |
1796 | * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI | |
1797 | * header contains a number of buckets that point to an inode, and each inode | |
1798 | * record has a pointer to the next inode in the hash chain. This | |
1799 | * singly-linked list causes scaling problems in the iunlink remove function | |
1800 | * because we must walk that list to find the inode that points to the inode | |
1801 | * being removed from the unlinked hash bucket list. | |
1802 | * | |
2fd26cc0 DC |
1803 | * Hence we keep an in-memory double linked list to link each inode on an |
1804 | * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer | |
1805 | * based lists would require having 64 list heads in the perag, one for each | |
1806 | * list. This is expensive in terms of memory (think millions of AGs) and cache | |
1807 | * misses on lookups. Instead, use the fact that inodes on the unlinked list | |
1808 | * must be referenced at the VFS level to keep them on the list and hence we | |
1809 | * have an existence guarantee for inodes on the unlinked list. | |
9b247179 | 1810 | * |
2fd26cc0 DC |
1811 | * Given we have an existence guarantee, we can use lockless inode cache lookups |
1812 | * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode | |
1813 | * for the double linked unlinked list, and we don't need any extra locking to | |
1814 | * keep the list safe as all manipulations are done under the AGI buffer lock. | |
1815 | * Keeping the list up to date does not require memory allocation, just finding | |
1816 | * the XFS inode and updating the next/prev unlinked list aginos. | |
9b247179 DW |
1817 | */ |
1818 | ||
9b247179 | 1819 | /* |
a83d5a8b DC |
1820 | * Find an inode on the unlinked list. This does not take references to the |
1821 | * inode as we have existence guarantees by holding the AGI buffer lock and that | |
1822 | * only unlinked, referenced inodes can be on the unlinked inode list. If we | |
1823 | * don't find the inode in cache, then let the caller handle the situation. | |
9b247179 | 1824 | */ |
a83d5a8b DC |
1825 | static struct xfs_inode * |
1826 | xfs_iunlink_lookup( | |
9b247179 DW |
1827 | struct xfs_perag *pag, |
1828 | xfs_agino_t agino) | |
1829 | { | |
a83d5a8b | 1830 | struct xfs_inode *ip; |
9b247179 | 1831 | |
a83d5a8b DC |
1832 | rcu_read_lock(); |
1833 | ip = radix_tree_lookup(&pag->pag_ici_root, agino); | |
9b247179 | 1834 | |
9b247179 | 1835 | /* |
a83d5a8b DC |
1836 | * Inode not in memory or in RCU freeing limbo should not happen. |
1837 | * Warn about this and let the caller handle the failure. | |
9b247179 | 1838 | */ |
a83d5a8b DC |
1839 | if (WARN_ON_ONCE(!ip || !ip->i_ino)) { |
1840 | rcu_read_unlock(); | |
1841 | return NULL; | |
9b247179 | 1842 | } |
a83d5a8b DC |
1843 | ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM)); |
1844 | rcu_read_unlock(); | |
1845 | return ip; | |
9b247179 DW |
1846 | } |
1847 | ||
2fd26cc0 | 1848 | /* Update the prev pointer of the next agino. */ |
9b247179 | 1849 | static int |
2fd26cc0 | 1850 | xfs_iunlink_update_backref( |
9b247179 DW |
1851 | struct xfs_perag *pag, |
1852 | xfs_agino_t prev_agino, | |
2fd26cc0 | 1853 | xfs_agino_t next_agino) |
9b247179 | 1854 | { |
2fd26cc0 | 1855 | struct xfs_inode *ip; |
9b247179 | 1856 | |
2fd26cc0 DC |
1857 | /* No update necessary if we are at the end of the list. */ |
1858 | if (next_agino == NULLAGINO) | |
9b247179 | 1859 | return 0; |
9b247179 | 1860 | |
2fd26cc0 DC |
1861 | ip = xfs_iunlink_lookup(pag, next_agino); |
1862 | if (!ip) | |
1863 | return -EFSCORRUPTED; | |
1864 | ip->i_prev_unlinked = prev_agino; | |
1865 | return 0; | |
9b247179 DW |
1866 | } |
1867 | ||
9a4a5118 DW |
1868 | /* |
1869 | * Point the AGI unlinked bucket at an inode and log the results. The caller | |
1870 | * is responsible for validating the old value. | |
1871 | */ | |
1872 | STATIC int | |
1873 | xfs_iunlink_update_bucket( | |
1874 | struct xfs_trans *tp, | |
f40aadb2 | 1875 | struct xfs_perag *pag, |
9a4a5118 DW |
1876 | struct xfs_buf *agibp, |
1877 | unsigned int bucket_index, | |
1878 | xfs_agino_t new_agino) | |
1879 | { | |
370c782b | 1880 | struct xfs_agi *agi = agibp->b_addr; |
9a4a5118 DW |
1881 | xfs_agino_t old_value; |
1882 | int offset; | |
1883 | ||
2d6ca832 | 1884 | ASSERT(xfs_verify_agino_or_null(pag, new_agino)); |
9a4a5118 DW |
1885 | |
1886 | old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]); | |
f40aadb2 | 1887 | trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index, |
9a4a5118 DW |
1888 | old_value, new_agino); |
1889 | ||
1890 | /* | |
1891 | * We should never find the head of the list already set to the value | |
1892 | * passed in because either we're adding or removing ourselves from the | |
1893 | * head of the list. | |
1894 | */ | |
a5155b87 | 1895 | if (old_value == new_agino) { |
8d57c216 | 1896 | xfs_buf_mark_corrupt(agibp); |
9a4a5118 | 1897 | return -EFSCORRUPTED; |
a5155b87 | 1898 | } |
9a4a5118 DW |
1899 | |
1900 | agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino); | |
1901 | offset = offsetof(struct xfs_agi, agi_unlinked) + | |
1902 | (sizeof(xfs_agino_t) * bucket_index); | |
1903 | xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1); | |
1904 | return 0; | |
1905 | } | |
1906 | ||
a4454cd6 DC |
1907 | static int |
1908 | xfs_iunlink_insert_inode( | |
f2fc16a3 | 1909 | struct xfs_trans *tp, |
f40aadb2 | 1910 | struct xfs_perag *pag, |
a4454cd6 | 1911 | struct xfs_buf *agibp, |
5837f625 | 1912 | struct xfs_inode *ip) |
f2fc16a3 DW |
1913 | { |
1914 | struct xfs_mount *mp = tp->t_mountp; | |
a4454cd6 | 1915 | struct xfs_agi *agi = agibp->b_addr; |
86bfd375 | 1916 | xfs_agino_t next_agino; |
5837f625 DW |
1917 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino); |
1918 | short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
f2fc16a3 DW |
1919 | int error; |
1920 | ||
1da177e4 | 1921 | /* |
86bfd375 DW |
1922 | * Get the index into the agi hash table for the list this inode will |
1923 | * go on. Make sure the pointer isn't garbage and that this inode | |
1924 | * isn't already on the list. | |
1da177e4 | 1925 | */ |
86bfd375 DW |
1926 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1927 | if (next_agino == agino || | |
2d6ca832 | 1928 | !xfs_verify_agino_or_null(pag, next_agino)) { |
8d57c216 | 1929 | xfs_buf_mark_corrupt(agibp); |
a4454cd6 | 1930 | return -EFSCORRUPTED; |
f2fc16a3 DW |
1931 | } |
1932 | ||
1933 | /* | |
2fd26cc0 DC |
1934 | * Update the prev pointer in the next inode to point back to this |
1935 | * inode. | |
f2fc16a3 | 1936 | */ |
2fd26cc0 DC |
1937 | error = xfs_iunlink_update_backref(pag, agino, next_agino); |
1938 | if (error) | |
1939 | return error; | |
1940 | ||
86bfd375 | 1941 | if (next_agino != NULLAGINO) { |
1da177e4 | 1942 | /* |
f2fc16a3 DW |
1943 | * There is already another inode in the bucket, so point this |
1944 | * inode to the current head of the list. | |
1da177e4 | 1945 | */ |
062efdb0 | 1946 | error = xfs_iunlink_log_inode(tp, ip, pag, next_agino); |
c319b58b | 1947 | if (error) |
a4454cd6 | 1948 | return error; |
4fcc94d6 | 1949 | ip->i_next_unlinked = next_agino; |
f2fc16a3 DW |
1950 | } |
1951 | ||
9a4a5118 | 1952 | /* Point the head of the list to point to this inode. */ |
a4454cd6 | 1953 | return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino); |
f2fc16a3 DW |
1954 | } |
1955 | ||
1da177e4 | 1956 | /* |
c4a6bf7f DW |
1957 | * This is called when the inode's link count has gone to 0 or we are creating |
1958 | * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0. | |
54d7b5c1 DC |
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( |
5837f625 DW |
1965 | struct xfs_trans *tp, |
1966 | struct xfs_inode *ip) | |
1da177e4 | 1967 | { |
5837f625 | 1968 | struct xfs_mount *mp = tp->t_mountp; |
f40aadb2 | 1969 | struct xfs_perag *pag; |
5837f625 | 1970 | struct xfs_buf *agibp; |
5837f625 | 1971 | int error; |
1da177e4 | 1972 | |
c4a6bf7f | 1973 | ASSERT(VFS_I(ip)->i_nlink == 0); |
c19b3b05 | 1974 | ASSERT(VFS_I(ip)->i_mode != 0); |
4664c66c | 1975 | trace_xfs_iunlink(ip); |
1da177e4 | 1976 | |
f40aadb2 DC |
1977 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1978 | ||
5837f625 | 1979 | /* Get the agi buffer first. It ensures lock ordering on the list. */ |
61021deb | 1980 | error = xfs_read_agi(pag, tp, &agibp); |
859d7182 | 1981 | if (error) |
f40aadb2 | 1982 | goto out; |
5e1be0fb | 1983 | |
a4454cd6 | 1984 | error = xfs_iunlink_insert_inode(tp, pag, agibp, ip); |
f40aadb2 DC |
1985 | out: |
1986 | xfs_perag_put(pag); | |
1987 | return error; | |
1da177e4 LT |
1988 | } |
1989 | ||
a4454cd6 DC |
1990 | static int |
1991 | xfs_iunlink_remove_inode( | |
5837f625 | 1992 | struct xfs_trans *tp, |
f40aadb2 | 1993 | struct xfs_perag *pag, |
a4454cd6 | 1994 | struct xfs_buf *agibp, |
5837f625 | 1995 | struct xfs_inode *ip) |
1da177e4 | 1996 | { |
5837f625 | 1997 | struct xfs_mount *mp = tp->t_mountp; |
a4454cd6 | 1998 | struct xfs_agi *agi = agibp->b_addr; |
5837f625 | 1999 | xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino); |
b1d2a068 | 2000 | xfs_agino_t head_agino; |
5837f625 | 2001 | short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; |
5837f625 | 2002 | int error; |
1da177e4 | 2003 | |
4664c66c DW |
2004 | trace_xfs_iunlink_remove(ip); |
2005 | ||
1da177e4 | 2006 | /* |
86bfd375 DW |
2007 | * Get the index into the agi hash table for the list this inode will |
2008 | * go on. Make sure the head pointer isn't garbage. | |
1da177e4 | 2009 | */ |
b1d2a068 | 2010 | head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
2d6ca832 | 2011 | if (!xfs_verify_agino(pag, head_agino)) { |
d2e73665 DW |
2012 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, |
2013 | agi, sizeof(*agi)); | |
2014 | return -EFSCORRUPTED; | |
2015 | } | |
1da177e4 | 2016 | |
b1d2a068 DW |
2017 | /* |
2018 | * Set our inode's next_unlinked pointer to NULL and then return | |
2019 | * the old pointer value so that we can update whatever was previous | |
2020 | * to us in the list to point to whatever was next in the list. | |
2021 | */ | |
062efdb0 | 2022 | error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO); |
b1d2a068 DW |
2023 | if (error) |
2024 | return error; | |
9a4a5118 | 2025 | |
9b247179 | 2026 | /* |
2fd26cc0 DC |
2027 | * Update the prev pointer in the next inode to point back to previous |
2028 | * inode in the chain. | |
9b247179 | 2029 | */ |
2fd26cc0 DC |
2030 | error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked, |
2031 | ip->i_next_unlinked); | |
2032 | if (error) | |
2033 | return error; | |
9b247179 | 2034 | |
92a00544 | 2035 | if (head_agino != agino) { |
a83d5a8b | 2036 | struct xfs_inode *prev_ip; |
475ee413 | 2037 | |
2fd26cc0 DC |
2038 | prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked); |
2039 | if (!prev_ip) | |
2040 | return -EFSCORRUPTED; | |
9b247179 | 2041 | |
062efdb0 | 2042 | error = xfs_iunlink_log_inode(tp, prev_ip, pag, |
5301f870 | 2043 | ip->i_next_unlinked); |
a83d5a8b | 2044 | prev_ip->i_next_unlinked = ip->i_next_unlinked; |
2fd26cc0 DC |
2045 | } else { |
2046 | /* Point the head of the list to the next unlinked inode. */ | |
2047 | error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, | |
2048 | ip->i_next_unlinked); | |
1da177e4 | 2049 | } |
9b247179 | 2050 | |
a83d5a8b | 2051 | ip->i_next_unlinked = NULLAGINO; |
2fd26cc0 DC |
2052 | ip->i_prev_unlinked = NULLAGINO; |
2053 | return error; | |
1da177e4 LT |
2054 | } |
2055 | ||
a4454cd6 DC |
2056 | /* |
2057 | * Pull the on-disk inode from the AGI unlinked list. | |
2058 | */ | |
2059 | STATIC int | |
2060 | xfs_iunlink_remove( | |
2061 | struct xfs_trans *tp, | |
2062 | struct xfs_perag *pag, | |
2063 | struct xfs_inode *ip) | |
2064 | { | |
2065 | struct xfs_buf *agibp; | |
2066 | int error; | |
2067 | ||
2068 | trace_xfs_iunlink_remove(ip); | |
2069 | ||
2070 | /* Get the agi buffer first. It ensures lock ordering on the list. */ | |
2071 | error = xfs_read_agi(pag, tp, &agibp); | |
2072 | if (error) | |
2073 | return error; | |
2074 | ||
2075 | return xfs_iunlink_remove_inode(tp, pag, agibp, ip); | |
1da177e4 LT |
2076 | } |
2077 | ||
5806165a | 2078 | /* |
71e3e356 DC |
2079 | * Look up the inode number specified and if it is not already marked XFS_ISTALE |
2080 | * mark it stale. We should only find clean inodes in this lookup that aren't | |
2081 | * already stale. | |
5806165a | 2082 | */ |
71e3e356 DC |
2083 | static void |
2084 | xfs_ifree_mark_inode_stale( | |
f40aadb2 | 2085 | struct xfs_perag *pag, |
5806165a | 2086 | struct xfs_inode *free_ip, |
d9fdd0ad | 2087 | xfs_ino_t inum) |
5806165a | 2088 | { |
f40aadb2 | 2089 | struct xfs_mount *mp = pag->pag_mount; |
71e3e356 | 2090 | struct xfs_inode_log_item *iip; |
5806165a DC |
2091 | struct xfs_inode *ip; |
2092 | ||
2093 | retry: | |
2094 | rcu_read_lock(); | |
2095 | ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum)); | |
2096 | ||
2097 | /* Inode not in memory, nothing to do */ | |
71e3e356 DC |
2098 | if (!ip) { |
2099 | rcu_read_unlock(); | |
2100 | return; | |
2101 | } | |
5806165a DC |
2102 | |
2103 | /* | |
2104 | * because this is an RCU protected lookup, we could find a recently | |
2105 | * freed or even reallocated inode during the lookup. We need to check | |
2106 | * under the i_flags_lock for a valid inode here. Skip it if it is not | |
2107 | * valid, the wrong inode or stale. | |
2108 | */ | |
2109 | spin_lock(&ip->i_flags_lock); | |
718ecc50 DC |
2110 | if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE)) |
2111 | goto out_iflags_unlock; | |
5806165a DC |
2112 | |
2113 | /* | |
2114 | * Don't try to lock/unlock the current inode, but we _cannot_ skip the | |
2115 | * other inodes that we did not find in the list attached to the buffer | |
2116 | * and are not already marked stale. If we can't lock it, back off and | |
2117 | * retry. | |
2118 | */ | |
2119 | if (ip != free_ip) { | |
2120 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
71e3e356 | 2121 | spin_unlock(&ip->i_flags_lock); |
5806165a DC |
2122 | rcu_read_unlock(); |
2123 | delay(1); | |
2124 | goto retry; | |
2125 | } | |
5806165a | 2126 | } |
71e3e356 | 2127 | ip->i_flags |= XFS_ISTALE; |
5806165a | 2128 | |
71e3e356 | 2129 | /* |
718ecc50 | 2130 | * If the inode is flushing, it is already attached to the buffer. All |
71e3e356 DC |
2131 | * we needed to do here is mark the inode stale so buffer IO completion |
2132 | * will remove it from the AIL. | |
2133 | */ | |
2134 | iip = ip->i_itemp; | |
718ecc50 | 2135 | if (__xfs_iflags_test(ip, XFS_IFLUSHING)) { |
71e3e356 DC |
2136 | ASSERT(!list_empty(&iip->ili_item.li_bio_list)); |
2137 | ASSERT(iip->ili_last_fields); | |
2138 | goto out_iunlock; | |
2139 | } | |
5806165a DC |
2140 | |
2141 | /* | |
48d55e2a DC |
2142 | * Inodes not attached to the buffer can be released immediately. |
2143 | * Everything else has to go through xfs_iflush_abort() on journal | |
2144 | * commit as the flock synchronises removal of the inode from the | |
2145 | * cluster buffer against inode reclaim. | |
5806165a | 2146 | */ |
718ecc50 | 2147 | if (!iip || list_empty(&iip->ili_item.li_bio_list)) |
71e3e356 | 2148 | goto out_iunlock; |
718ecc50 DC |
2149 | |
2150 | __xfs_iflags_set(ip, XFS_IFLUSHING); | |
2151 | spin_unlock(&ip->i_flags_lock); | |
2152 | rcu_read_unlock(); | |
5806165a | 2153 | |
71e3e356 | 2154 | /* we have a dirty inode in memory that has not yet been flushed. */ |
71e3e356 DC |
2155 | spin_lock(&iip->ili_lock); |
2156 | iip->ili_last_fields = iip->ili_fields; | |
2157 | iip->ili_fields = 0; | |
2158 | iip->ili_fsync_fields = 0; | |
2159 | spin_unlock(&iip->ili_lock); | |
71e3e356 DC |
2160 | ASSERT(iip->ili_last_fields); |
2161 | ||
718ecc50 DC |
2162 | if (ip != free_ip) |
2163 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2164 | return; | |
2165 | ||
71e3e356 DC |
2166 | out_iunlock: |
2167 | if (ip != free_ip) | |
2168 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
718ecc50 DC |
2169 | out_iflags_unlock: |
2170 | spin_unlock(&ip->i_flags_lock); | |
2171 | rcu_read_unlock(); | |
5806165a DC |
2172 | } |
2173 | ||
5b3eed75 | 2174 | /* |
0b8182db | 2175 | * A big issue when freeing the inode cluster is that we _cannot_ skip any |
5b3eed75 DC |
2176 | * inodes that are in memory - they all must be marked stale and attached to |
2177 | * the cluster buffer. | |
2178 | */ | |
f40aadb2 | 2179 | static int |
1da177e4 | 2180 | xfs_ifree_cluster( |
71e3e356 | 2181 | struct xfs_trans *tp, |
f40aadb2 DC |
2182 | struct xfs_perag *pag, |
2183 | struct xfs_inode *free_ip, | |
09b56604 | 2184 | struct xfs_icluster *xic) |
1da177e4 | 2185 | { |
71e3e356 DC |
2186 | struct xfs_mount *mp = free_ip->i_mount; |
2187 | struct xfs_ino_geometry *igeo = M_IGEO(mp); | |
2188 | struct xfs_buf *bp; | |
2189 | xfs_daddr_t blkno; | |
2190 | xfs_ino_t inum = xic->first_ino; | |
1da177e4 | 2191 | int nbufs; |
5b257b4a | 2192 | int i, j; |
3cdaa189 | 2193 | int ioffset; |
ce92464c | 2194 | int error; |
1da177e4 | 2195 | |
ef325959 | 2196 | nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster; |
1da177e4 | 2197 | |
ef325959 | 2198 | for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) { |
09b56604 BF |
2199 | /* |
2200 | * The allocation bitmap tells us which inodes of the chunk were | |
2201 | * physically allocated. Skip the cluster if an inode falls into | |
2202 | * a sparse region. | |
2203 | */ | |
3cdaa189 BF |
2204 | ioffset = inum - xic->first_ino; |
2205 | if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { | |
ef325959 | 2206 | ASSERT(ioffset % igeo->inodes_per_cluster == 0); |
09b56604 BF |
2207 | continue; |
2208 | } | |
2209 | ||
1da177e4 LT |
2210 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), |
2211 | XFS_INO_TO_AGBNO(mp, inum)); | |
2212 | ||
5b257b4a DC |
2213 | /* |
2214 | * We obtain and lock the backing buffer first in the process | |
718ecc50 DC |
2215 | * here to ensure dirty inodes attached to the buffer remain in |
2216 | * the flushing state while we mark them stale. | |
2217 | * | |
5b257b4a DC |
2218 | * If we scan the in-memory inodes first, then buffer IO can |
2219 | * complete before we get a lock on it, and hence we may fail | |
2220 | * to mark all the active inodes on the buffer stale. | |
2221 | */ | |
ce92464c DW |
2222 | error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, |
2223 | mp->m_bsize * igeo->blocks_per_cluster, | |
2224 | XBF_UNMAPPED, &bp); | |
71e3e356 | 2225 | if (error) |
ce92464c | 2226 | return error; |
b0f539de DC |
2227 | |
2228 | /* | |
2229 | * This buffer may not have been correctly initialised as we | |
2230 | * didn't read it from disk. That's not important because we are | |
2231 | * only using to mark the buffer as stale in the log, and to | |
2232 | * attach stale cached inodes on it. That means it will never be | |
2233 | * dispatched for IO. If it is, we want to know about it, and we | |
2234 | * want it to fail. We can acheive this by adding a write | |
2235 | * verifier to the buffer. | |
2236 | */ | |
8c4ce794 | 2237 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 2238 | |
5b257b4a | 2239 | /* |
71e3e356 DC |
2240 | * Now we need to set all the cached clean inodes as XFS_ISTALE, |
2241 | * too. This requires lookups, and will skip inodes that we've | |
2242 | * already marked XFS_ISTALE. | |
1da177e4 | 2243 | */ |
71e3e356 | 2244 | for (i = 0; i < igeo->inodes_per_cluster; i++) |
f40aadb2 | 2245 | xfs_ifree_mark_inode_stale(pag, free_ip, inum + i); |
1da177e4 | 2246 | |
5b3eed75 | 2247 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
2248 | xfs_trans_binval(tp, bp); |
2249 | } | |
2a30f36d | 2250 | return 0; |
1da177e4 LT |
2251 | } |
2252 | ||
2253 | /* | |
9a5280b3 DC |
2254 | * This is called to return an inode to the inode free list. The inode should |
2255 | * already be truncated to 0 length and have no pages associated with it. This | |
2256 | * routine also assumes that the inode is already a part of the transaction. | |
1da177e4 | 2257 | * |
9a5280b3 DC |
2258 | * The on-disk copy of the inode will have been added to the list of unlinked |
2259 | * inodes in the AGI. We need to remove the inode from that list atomically with | |
2260 | * respect to freeing it here. | |
1da177e4 LT |
2261 | */ |
2262 | int | |
2263 | xfs_ifree( | |
0e0417f3 BF |
2264 | struct xfs_trans *tp, |
2265 | struct xfs_inode *ip) | |
1da177e4 | 2266 | { |
f40aadb2 DC |
2267 | struct xfs_mount *mp = ip->i_mount; |
2268 | struct xfs_perag *pag; | |
09b56604 | 2269 | struct xfs_icluster xic = { 0 }; |
1319ebef | 2270 | struct xfs_inode_log_item *iip = ip->i_itemp; |
f40aadb2 | 2271 | int error; |
1da177e4 | 2272 | |
579aa9ca | 2273 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
54d7b5c1 | 2274 | ASSERT(VFS_I(ip)->i_nlink == 0); |
daf83964 | 2275 | ASSERT(ip->i_df.if_nextents == 0); |
13d2c10b | 2276 | ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); |
6e73a545 | 2277 | ASSERT(ip->i_nblocks == 0); |
1da177e4 | 2278 | |
f40aadb2 DC |
2279 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
2280 | ||
1da177e4 | 2281 | /* |
9a5280b3 DC |
2282 | * Free the inode first so that we guarantee that the AGI lock is going |
2283 | * to be taken before we remove the inode from the unlinked list. This | |
2284 | * makes the AGI lock -> unlinked list modification order the same as | |
2285 | * used in O_TMPFILE creation. | |
1da177e4 | 2286 | */ |
9a5280b3 | 2287 | error = xfs_difree(tp, pag, ip->i_ino, &xic); |
1baaed8f | 2288 | if (error) |
6f5097e3 | 2289 | goto out; |
1da177e4 | 2290 | |
9a5280b3 | 2291 | error = xfs_iunlink_remove(tp, pag, ip); |
1baaed8f | 2292 | if (error) |
f40aadb2 | 2293 | goto out; |
1baaed8f | 2294 | |
b2c20045 CH |
2295 | /* |
2296 | * Free any local-format data sitting around before we reset the | |
2297 | * data fork to extents format. Note that the attr fork data has | |
2298 | * already been freed by xfs_attr_inactive. | |
2299 | */ | |
f7e67b20 | 2300 | if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) { |
b2c20045 CH |
2301 | kmem_free(ip->i_df.if_u1.if_data); |
2302 | ip->i_df.if_u1.if_data = NULL; | |
2303 | ip->i_df.if_bytes = 0; | |
2304 | } | |
98c4f78d | 2305 | |
c19b3b05 | 2306 | VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ |
db07349d | 2307 | ip->i_diflags = 0; |
f40aadb2 | 2308 | ip->i_diflags2 = mp->m_ino_geo.new_diflags2; |
7821ea30 | 2309 | ip->i_forkoff = 0; /* mark the attr fork not in use */ |
f7e67b20 | 2310 | ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS; |
9b3beb02 CH |
2311 | if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS)) |
2312 | xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS); | |
dc1baa71 ES |
2313 | |
2314 | /* Don't attempt to replay owner changes for a deleted inode */ | |
1319ebef DC |
2315 | spin_lock(&iip->ili_lock); |
2316 | iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER); | |
2317 | spin_unlock(&iip->ili_lock); | |
dc1baa71 | 2318 | |
1da177e4 LT |
2319 | /* |
2320 | * Bump the generation count so no one will be confused | |
2321 | * by reincarnations of this inode. | |
2322 | */ | |
9e9a2674 | 2323 | VFS_I(ip)->i_generation++; |
1da177e4 LT |
2324 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
2325 | ||
09b56604 | 2326 | if (xic.deleted) |
f40aadb2 DC |
2327 | error = xfs_ifree_cluster(tp, pag, ip, &xic); |
2328 | out: | |
2329 | xfs_perag_put(pag); | |
2a30f36d | 2330 | return error; |
1da177e4 LT |
2331 | } |
2332 | ||
1da177e4 | 2333 | /* |
60ec6783 CH |
2334 | * This is called to unpin an inode. The caller must have the inode locked |
2335 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2336 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2337 | */ |
60ec6783 | 2338 | static void |
f392e631 | 2339 | xfs_iunpin( |
60ec6783 | 2340 | struct xfs_inode *ip) |
1da177e4 | 2341 | { |
579aa9ca | 2342 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2343 | |
4aaf15d1 DC |
2344 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2345 | ||
a3f74ffb | 2346 | /* Give the log a push to start the unpinning I/O */ |
5f9b4b0d | 2347 | xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL); |
a14a348b | 2348 | |
a3f74ffb | 2349 | } |
1da177e4 | 2350 | |
f392e631 CH |
2351 | static void |
2352 | __xfs_iunpin_wait( | |
2353 | struct xfs_inode *ip) | |
2354 | { | |
2355 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2356 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2357 | ||
2358 | xfs_iunpin(ip); | |
2359 | ||
2360 | do { | |
21417136 | 2361 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
f392e631 CH |
2362 | if (xfs_ipincount(ip)) |
2363 | io_schedule(); | |
2364 | } while (xfs_ipincount(ip)); | |
21417136 | 2365 | finish_wait(wq, &wait.wq_entry); |
f392e631 CH |
2366 | } |
2367 | ||
777df5af | 2368 | void |
a3f74ffb | 2369 | xfs_iunpin_wait( |
60ec6783 | 2370 | struct xfs_inode *ip) |
a3f74ffb | 2371 | { |
f392e631 CH |
2372 | if (xfs_ipincount(ip)) |
2373 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2374 | } |
2375 | ||
27320369 DC |
2376 | /* |
2377 | * Removing an inode from the namespace involves removing the directory entry | |
2378 | * and dropping the link count on the inode. Removing the directory entry can | |
2379 | * result in locking an AGF (directory blocks were freed) and removing a link | |
2380 | * count can result in placing the inode on an unlinked list which results in | |
2381 | * locking an AGI. | |
2382 | * | |
2383 | * The big problem here is that we have an ordering constraint on AGF and AGI | |
2384 | * locking - inode allocation locks the AGI, then can allocate a new extent for | |
2385 | * new inodes, locking the AGF after the AGI. Similarly, freeing the inode | |
2386 | * removes the inode from the unlinked list, requiring that we lock the AGI | |
2387 | * first, and then freeing the inode can result in an inode chunk being freed | |
2388 | * and hence freeing disk space requiring that we lock an AGF. | |
2389 | * | |
2390 | * Hence the ordering that is imposed by other parts of the code is AGI before | |
2391 | * AGF. This means we cannot remove the directory entry before we drop the inode | |
2392 | * reference count and put it on the unlinked list as this results in a lock | |
2393 | * order of AGF then AGI, and this can deadlock against inode allocation and | |
2394 | * freeing. Therefore we must drop the link counts before we remove the | |
2395 | * directory entry. | |
2396 | * | |
2397 | * This is still safe from a transactional point of view - it is not until we | |
310a75a3 | 2398 | * get to xfs_defer_finish() that we have the possibility of multiple |
27320369 DC |
2399 | * transactions in this operation. Hence as long as we remove the directory |
2400 | * entry and drop the link count in the first transaction of the remove | |
2401 | * operation, there are no transactional constraints on the ordering here. | |
2402 | */ | |
c24b5dfa DC |
2403 | int |
2404 | xfs_remove( | |
2405 | xfs_inode_t *dp, | |
2406 | struct xfs_name *name, | |
2407 | xfs_inode_t *ip) | |
2408 | { | |
2409 | xfs_mount_t *mp = dp->i_mount; | |
2410 | xfs_trans_t *tp = NULL; | |
c19b3b05 | 2411 | int is_dir = S_ISDIR(VFS_I(ip)->i_mode); |
871b9316 | 2412 | int dontcare; |
c24b5dfa | 2413 | int error = 0; |
c24b5dfa | 2414 | uint resblks; |
c24b5dfa DC |
2415 | |
2416 | trace_xfs_remove(dp, name); | |
2417 | ||
75c8c50f | 2418 | if (xfs_is_shutdown(mp)) |
2451337d | 2419 | return -EIO; |
c24b5dfa | 2420 | |
c14cfcca | 2421 | error = xfs_qm_dqattach(dp); |
c24b5dfa DC |
2422 | if (error) |
2423 | goto std_return; | |
2424 | ||
c14cfcca | 2425 | error = xfs_qm_dqattach(ip); |
c24b5dfa DC |
2426 | if (error) |
2427 | goto std_return; | |
2428 | ||
c24b5dfa | 2429 | /* |
871b9316 DW |
2430 | * We try to get the real space reservation first, allowing for |
2431 | * directory btree deletion(s) implying possible bmap insert(s). If we | |
2432 | * can't get the space reservation then we use 0 instead, and avoid the | |
2433 | * bmap btree insert(s) in the directory code by, if the bmap insert | |
2434 | * tries to happen, instead trimming the LAST block from the directory. | |
2435 | * | |
2436 | * Ignore EDQUOT and ENOSPC being returned via nospace_error because | |
2437 | * the directory code can handle a reservationless update and we don't | |
2438 | * want to prevent a user from trying to free space by deleting things. | |
c24b5dfa DC |
2439 | */ |
2440 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
871b9316 DW |
2441 | error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks, |
2442 | &tp, &dontcare); | |
c24b5dfa | 2443 | if (error) { |
2451337d | 2444 | ASSERT(error != -ENOSPC); |
253f4911 | 2445 | goto std_return; |
c24b5dfa DC |
2446 | } |
2447 | ||
c24b5dfa DC |
2448 | /* |
2449 | * If we're removing a directory perform some additional validation. | |
2450 | */ | |
2451 | if (is_dir) { | |
54d7b5c1 DC |
2452 | ASSERT(VFS_I(ip)->i_nlink >= 2); |
2453 | if (VFS_I(ip)->i_nlink != 2) { | |
2451337d | 2454 | error = -ENOTEMPTY; |
c24b5dfa DC |
2455 | goto out_trans_cancel; |
2456 | } | |
2457 | if (!xfs_dir_isempty(ip)) { | |
2451337d | 2458 | error = -ENOTEMPTY; |
c24b5dfa DC |
2459 | goto out_trans_cancel; |
2460 | } | |
c24b5dfa | 2461 | |
27320369 | 2462 | /* Drop the link from ip's "..". */ |
c24b5dfa DC |
2463 | error = xfs_droplink(tp, dp); |
2464 | if (error) | |
27320369 | 2465 | goto out_trans_cancel; |
c24b5dfa | 2466 | |
27320369 | 2467 | /* Drop the "." link from ip to self. */ |
c24b5dfa DC |
2468 | error = xfs_droplink(tp, ip); |
2469 | if (error) | |
27320369 | 2470 | goto out_trans_cancel; |
5838d035 DW |
2471 | |
2472 | /* | |
2473 | * Point the unlinked child directory's ".." entry to the root | |
2474 | * directory to eliminate back-references to inodes that may | |
2475 | * get freed before the child directory is closed. If the fs | |
2476 | * gets shrunk, this can lead to dirent inode validation errors. | |
2477 | */ | |
2478 | if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) { | |
2479 | error = xfs_dir_replace(tp, ip, &xfs_name_dotdot, | |
2480 | tp->t_mountp->m_sb.sb_rootino, 0); | |
2481 | if (error) | |
2653d533 | 2482 | goto out_trans_cancel; |
5838d035 | 2483 | } |
c24b5dfa DC |
2484 | } else { |
2485 | /* | |
2486 | * When removing a non-directory we need to log the parent | |
2487 | * inode here. For a directory this is done implicitly | |
2488 | * by the xfs_droplink call for the ".." entry. | |
2489 | */ | |
2490 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2491 | } | |
27320369 | 2492 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
c24b5dfa | 2493 | |
27320369 | 2494 | /* Drop the link from dp to ip. */ |
c24b5dfa DC |
2495 | error = xfs_droplink(tp, ip); |
2496 | if (error) | |
27320369 | 2497 | goto out_trans_cancel; |
c24b5dfa | 2498 | |
381eee69 | 2499 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks); |
27320369 | 2500 | if (error) { |
2451337d | 2501 | ASSERT(error != -ENOENT); |
c8eac49e | 2502 | goto out_trans_cancel; |
27320369 DC |
2503 | } |
2504 | ||
c24b5dfa DC |
2505 | /* |
2506 | * If this is a synchronous mount, make sure that the | |
2507 | * remove transaction goes to disk before returning to | |
2508 | * the user. | |
2509 | */ | |
0560f31a | 2510 | if (xfs_has_wsync(mp) || xfs_has_dirsync(mp)) |
c24b5dfa DC |
2511 | xfs_trans_set_sync(tp); |
2512 | ||
70393313 | 2513 | error = xfs_trans_commit(tp); |
c24b5dfa DC |
2514 | if (error) |
2515 | goto std_return; | |
2516 | ||
2cd2ef6a | 2517 | if (is_dir && xfs_inode_is_filestream(ip)) |
c24b5dfa DC |
2518 | xfs_filestream_deassociate(ip); |
2519 | ||
2520 | return 0; | |
2521 | ||
c24b5dfa | 2522 | out_trans_cancel: |
4906e215 | 2523 | xfs_trans_cancel(tp); |
c24b5dfa DC |
2524 | std_return: |
2525 | return error; | |
2526 | } | |
2527 | ||
f6bba201 DC |
2528 | /* |
2529 | * Enter all inodes for a rename transaction into a sorted array. | |
2530 | */ | |
95afcf5c | 2531 | #define __XFS_SORT_INODES 5 |
f6bba201 DC |
2532 | STATIC void |
2533 | xfs_sort_for_rename( | |
95afcf5c DC |
2534 | struct xfs_inode *dp1, /* in: old (source) directory inode */ |
2535 | struct xfs_inode *dp2, /* in: new (target) directory inode */ | |
2536 | struct xfs_inode *ip1, /* in: inode of old entry */ | |
2537 | struct xfs_inode *ip2, /* in: inode of new entry */ | |
2538 | struct xfs_inode *wip, /* in: whiteout inode */ | |
2539 | struct xfs_inode **i_tab,/* out: sorted array of inodes */ | |
2540 | int *num_inodes) /* in/out: inodes in array */ | |
f6bba201 | 2541 | { |
f6bba201 DC |
2542 | int i, j; |
2543 | ||
95afcf5c DC |
2544 | ASSERT(*num_inodes == __XFS_SORT_INODES); |
2545 | memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); | |
2546 | ||
f6bba201 DC |
2547 | /* |
2548 | * i_tab contains a list of pointers to inodes. We initialize | |
2549 | * the table here & we'll sort it. We will then use it to | |
2550 | * order the acquisition of the inode locks. | |
2551 | * | |
2552 | * Note that the table may contain duplicates. e.g., dp1 == dp2. | |
2553 | */ | |
95afcf5c DC |
2554 | i = 0; |
2555 | i_tab[i++] = dp1; | |
2556 | i_tab[i++] = dp2; | |
2557 | i_tab[i++] = ip1; | |
2558 | if (ip2) | |
2559 | i_tab[i++] = ip2; | |
2560 | if (wip) | |
2561 | i_tab[i++] = wip; | |
2562 | *num_inodes = i; | |
f6bba201 DC |
2563 | |
2564 | /* | |
2565 | * Sort the elements via bubble sort. (Remember, there are at | |
95afcf5c | 2566 | * most 5 elements to sort, so this is adequate.) |
f6bba201 DC |
2567 | */ |
2568 | for (i = 0; i < *num_inodes; i++) { | |
2569 | for (j = 1; j < *num_inodes; j++) { | |
2570 | if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { | |
95afcf5c | 2571 | struct xfs_inode *temp = i_tab[j]; |
f6bba201 DC |
2572 | i_tab[j] = i_tab[j-1]; |
2573 | i_tab[j-1] = temp; | |
2574 | } | |
2575 | } | |
2576 | } | |
2577 | } | |
2578 | ||
310606b0 DC |
2579 | static int |
2580 | xfs_finish_rename( | |
c9cfdb38 | 2581 | struct xfs_trans *tp) |
310606b0 | 2582 | { |
310606b0 DC |
2583 | /* |
2584 | * If this is a synchronous mount, make sure that the rename transaction | |
2585 | * goes to disk before returning to the user. | |
2586 | */ | |
0560f31a | 2587 | if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp)) |
310606b0 DC |
2588 | xfs_trans_set_sync(tp); |
2589 | ||
70393313 | 2590 | return xfs_trans_commit(tp); |
310606b0 DC |
2591 | } |
2592 | ||
d31a1825 CM |
2593 | /* |
2594 | * xfs_cross_rename() | |
2595 | * | |
0145225e | 2596 | * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall |
d31a1825 CM |
2597 | */ |
2598 | STATIC int | |
2599 | xfs_cross_rename( | |
2600 | struct xfs_trans *tp, | |
2601 | struct xfs_inode *dp1, | |
2602 | struct xfs_name *name1, | |
2603 | struct xfs_inode *ip1, | |
2604 | struct xfs_inode *dp2, | |
2605 | struct xfs_name *name2, | |
2606 | struct xfs_inode *ip2, | |
d31a1825 CM |
2607 | int spaceres) |
2608 | { | |
2609 | int error = 0; | |
2610 | int ip1_flags = 0; | |
2611 | int ip2_flags = 0; | |
2612 | int dp2_flags = 0; | |
2613 | ||
2614 | /* Swap inode number for dirent in first parent */ | |
381eee69 | 2615 | error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres); |
d31a1825 | 2616 | if (error) |
eeacd321 | 2617 | goto out_trans_abort; |
d31a1825 CM |
2618 | |
2619 | /* Swap inode number for dirent in second parent */ | |
381eee69 | 2620 | error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres); |
d31a1825 | 2621 | if (error) |
eeacd321 | 2622 | goto out_trans_abort; |
d31a1825 CM |
2623 | |
2624 | /* | |
2625 | * If we're renaming one or more directories across different parents, | |
2626 | * update the respective ".." entries (and link counts) to match the new | |
2627 | * parents. | |
2628 | */ | |
2629 | if (dp1 != dp2) { | |
2630 | dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2631 | ||
c19b3b05 | 2632 | if (S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 | 2633 | error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, |
381eee69 | 2634 | dp1->i_ino, spaceres); |
d31a1825 | 2635 | if (error) |
eeacd321 | 2636 | goto out_trans_abort; |
d31a1825 CM |
2637 | |
2638 | /* transfer ip2 ".." reference to dp1 */ | |
c19b3b05 | 2639 | if (!S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 CM |
2640 | error = xfs_droplink(tp, dp2); |
2641 | if (error) | |
eeacd321 | 2642 | goto out_trans_abort; |
91083269 | 2643 | xfs_bumplink(tp, dp1); |
d31a1825 CM |
2644 | } |
2645 | ||
2646 | /* | |
2647 | * Although ip1 isn't changed here, userspace needs | |
2648 | * to be warned about the change, so that applications | |
2649 | * relying on it (like backup ones), will properly | |
2650 | * notify the change | |
2651 | */ | |
2652 | ip1_flags |= XFS_ICHGTIME_CHG; | |
2653 | ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2654 | } | |
2655 | ||
c19b3b05 | 2656 | if (S_ISDIR(VFS_I(ip1)->i_mode)) { |
d31a1825 | 2657 | error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, |
381eee69 | 2658 | dp2->i_ino, spaceres); |
d31a1825 | 2659 | if (error) |
eeacd321 | 2660 | goto out_trans_abort; |
d31a1825 CM |
2661 | |
2662 | /* transfer ip1 ".." reference to dp2 */ | |
c19b3b05 | 2663 | if (!S_ISDIR(VFS_I(ip2)->i_mode)) { |
d31a1825 CM |
2664 | error = xfs_droplink(tp, dp1); |
2665 | if (error) | |
eeacd321 | 2666 | goto out_trans_abort; |
91083269 | 2667 | xfs_bumplink(tp, dp2); |
d31a1825 CM |
2668 | } |
2669 | ||
2670 | /* | |
2671 | * Although ip2 isn't changed here, userspace needs | |
2672 | * to be warned about the change, so that applications | |
2673 | * relying on it (like backup ones), will properly | |
2674 | * notify the change | |
2675 | */ | |
2676 | ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; | |
2677 | ip2_flags |= XFS_ICHGTIME_CHG; | |
2678 | } | |
2679 | } | |
2680 | ||
2681 | if (ip1_flags) { | |
2682 | xfs_trans_ichgtime(tp, ip1, ip1_flags); | |
2683 | xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); | |
2684 | } | |
2685 | if (ip2_flags) { | |
2686 | xfs_trans_ichgtime(tp, ip2, ip2_flags); | |
2687 | xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); | |
2688 | } | |
2689 | if (dp2_flags) { | |
2690 | xfs_trans_ichgtime(tp, dp2, dp2_flags); | |
2691 | xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); | |
2692 | } | |
2693 | xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2694 | xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); | |
c9cfdb38 | 2695 | return xfs_finish_rename(tp); |
eeacd321 DC |
2696 | |
2697 | out_trans_abort: | |
4906e215 | 2698 | xfs_trans_cancel(tp); |
d31a1825 CM |
2699 | return error; |
2700 | } | |
2701 | ||
7dcf5c3e DC |
2702 | /* |
2703 | * xfs_rename_alloc_whiteout() | |
2704 | * | |
b63da6c8 | 2705 | * Return a referenced, unlinked, unlocked inode that can be used as a |
7dcf5c3e DC |
2706 | * whiteout in a rename transaction. We use a tmpfile inode here so that if we |
2707 | * crash between allocating the inode and linking it into the rename transaction | |
2708 | * recovery will free the inode and we won't leak it. | |
2709 | */ | |
2710 | static int | |
2711 | xfs_rename_alloc_whiteout( | |
f2d40141 | 2712 | struct mnt_idmap *idmap, |
70b589a3 | 2713 | struct xfs_name *src_name, |
7dcf5c3e DC |
2714 | struct xfs_inode *dp, |
2715 | struct xfs_inode **wip) | |
2716 | { | |
2717 | struct xfs_inode *tmpfile; | |
70b589a3 | 2718 | struct qstr name; |
7dcf5c3e DC |
2719 | int error; |
2720 | ||
f2d40141 | 2721 | error = xfs_create_tmpfile(idmap, dp, S_IFCHR | WHITEOUT_MODE, |
f736d93d | 2722 | &tmpfile); |
7dcf5c3e DC |
2723 | if (error) |
2724 | return error; | |
2725 | ||
70b589a3 ES |
2726 | name.name = src_name->name; |
2727 | name.len = src_name->len; | |
2728 | error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name); | |
2729 | if (error) { | |
2730 | xfs_finish_inode_setup(tmpfile); | |
2731 | xfs_irele(tmpfile); | |
2732 | return error; | |
2733 | } | |
2734 | ||
22419ac9 BF |
2735 | /* |
2736 | * Prepare the tmpfile inode as if it were created through the VFS. | |
c4a6bf7f DW |
2737 | * Complete the inode setup and flag it as linkable. nlink is already |
2738 | * zero, so we can skip the drop_nlink. | |
22419ac9 | 2739 | */ |
2b3d1d41 | 2740 | xfs_setup_iops(tmpfile); |
7dcf5c3e DC |
2741 | xfs_finish_inode_setup(tmpfile); |
2742 | VFS_I(tmpfile)->i_state |= I_LINKABLE; | |
2743 | ||
2744 | *wip = tmpfile; | |
2745 | return 0; | |
2746 | } | |
2747 | ||
f6bba201 DC |
2748 | /* |
2749 | * xfs_rename | |
2750 | */ | |
2751 | int | |
2752 | xfs_rename( | |
f2d40141 | 2753 | struct mnt_idmap *idmap, |
7dcf5c3e DC |
2754 | struct xfs_inode *src_dp, |
2755 | struct xfs_name *src_name, | |
2756 | struct xfs_inode *src_ip, | |
2757 | struct xfs_inode *target_dp, | |
2758 | struct xfs_name *target_name, | |
2759 | struct xfs_inode *target_ip, | |
2760 | unsigned int flags) | |
f6bba201 | 2761 | { |
7dcf5c3e DC |
2762 | struct xfs_mount *mp = src_dp->i_mount; |
2763 | struct xfs_trans *tp; | |
7dcf5c3e DC |
2764 | struct xfs_inode *wip = NULL; /* whiteout inode */ |
2765 | struct xfs_inode *inodes[__XFS_SORT_INODES]; | |
6da1b4b1 | 2766 | int i; |
7dcf5c3e | 2767 | int num_inodes = __XFS_SORT_INODES; |
2b93681f | 2768 | bool new_parent = (src_dp != target_dp); |
c19b3b05 | 2769 | bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); |
7dcf5c3e | 2770 | int spaceres; |
41667260 DW |
2771 | bool retried = false; |
2772 | int error, nospace_error = 0; | |
f6bba201 DC |
2773 | |
2774 | trace_xfs_rename(src_dp, target_dp, src_name, target_name); | |
2775 | ||
eeacd321 DC |
2776 | if ((flags & RENAME_EXCHANGE) && !target_ip) |
2777 | return -EINVAL; | |
2778 | ||
7dcf5c3e DC |
2779 | /* |
2780 | * If we are doing a whiteout operation, allocate the whiteout inode | |
2781 | * we will be placing at the target and ensure the type is set | |
2782 | * appropriately. | |
2783 | */ | |
2784 | if (flags & RENAME_WHITEOUT) { | |
f2d40141 | 2785 | error = xfs_rename_alloc_whiteout(idmap, src_name, |
70b589a3 | 2786 | target_dp, &wip); |
7dcf5c3e DC |
2787 | if (error) |
2788 | return error; | |
2789 | ||
2790 | /* setup target dirent info as whiteout */ | |
2791 | src_name->type = XFS_DIR3_FT_CHRDEV; | |
2792 | } | |
f6bba201 | 2793 | |
7dcf5c3e | 2794 | xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, |
f6bba201 DC |
2795 | inodes, &num_inodes); |
2796 | ||
41667260 DW |
2797 | retry: |
2798 | nospace_error = 0; | |
f6bba201 | 2799 | spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); |
253f4911 | 2800 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); |
2451337d | 2801 | if (error == -ENOSPC) { |
41667260 | 2802 | nospace_error = error; |
f6bba201 | 2803 | spaceres = 0; |
253f4911 CH |
2804 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, |
2805 | &tp); | |
f6bba201 | 2806 | } |
445883e8 | 2807 | if (error) |
253f4911 | 2808 | goto out_release_wip; |
f6bba201 DC |
2809 | |
2810 | /* | |
2811 | * Attach the dquots to the inodes | |
2812 | */ | |
2813 | error = xfs_qm_vop_rename_dqattach(inodes); | |
445883e8 DC |
2814 | if (error) |
2815 | goto out_trans_cancel; | |
f6bba201 DC |
2816 | |
2817 | /* | |
2818 | * Lock all the participating inodes. Depending upon whether | |
2819 | * the target_name exists in the target directory, and | |
2820 | * whether the target directory is the same as the source | |
e07ee6fe | 2821 | * directory, we can lock from 2 to 5 inodes. |
f6bba201 DC |
2822 | */ |
2823 | xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); | |
2824 | ||
2825 | /* | |
2826 | * Join all the inodes to the transaction. From this point on, | |
2827 | * we can rely on either trans_commit or trans_cancel to unlock | |
2828 | * them. | |
2829 | */ | |
65523218 | 2830 | xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); |
f6bba201 | 2831 | if (new_parent) |
65523218 | 2832 | xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); |
f6bba201 DC |
2833 | xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); |
2834 | if (target_ip) | |
2835 | xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); | |
7dcf5c3e DC |
2836 | if (wip) |
2837 | xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); | |
f6bba201 DC |
2838 | |
2839 | /* | |
2840 | * If we are using project inheritance, we only allow renames | |
2841 | * into our tree when the project IDs are the same; else the | |
2842 | * tree quota mechanism would be circumvented. | |
2843 | */ | |
db07349d | 2844 | if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) && |
ceaf603c | 2845 | target_dp->i_projid != src_ip->i_projid)) { |
2451337d | 2846 | error = -EXDEV; |
445883e8 | 2847 | goto out_trans_cancel; |
f6bba201 DC |
2848 | } |
2849 | ||
eeacd321 DC |
2850 | /* RENAME_EXCHANGE is unique from here on. */ |
2851 | if (flags & RENAME_EXCHANGE) | |
2852 | return xfs_cross_rename(tp, src_dp, src_name, src_ip, | |
2853 | target_dp, target_name, target_ip, | |
f16dea54 | 2854 | spaceres); |
d31a1825 | 2855 | |
41667260 DW |
2856 | /* |
2857 | * Try to reserve quota to handle an expansion of the target directory. | |
2858 | * We'll allow the rename to continue in reservationless mode if we hit | |
2859 | * a space usage constraint. If we trigger reservationless mode, save | |
2860 | * the errno if there isn't any free space in the target directory. | |
2861 | */ | |
2862 | if (spaceres != 0) { | |
2863 | error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres, | |
2864 | 0, false); | |
2865 | if (error == -EDQUOT || error == -ENOSPC) { | |
2866 | if (!retried) { | |
2867 | xfs_trans_cancel(tp); | |
2868 | xfs_blockgc_free_quota(target_dp, 0); | |
2869 | retried = true; | |
2870 | goto retry; | |
2871 | } | |
2872 | ||
2873 | nospace_error = error; | |
2874 | spaceres = 0; | |
2875 | error = 0; | |
2876 | } | |
2877 | if (error) | |
2878 | goto out_trans_cancel; | |
2879 | } | |
2880 | ||
f6bba201 | 2881 | /* |
bc56ad8c | 2882 | * Check for expected errors before we dirty the transaction |
2883 | * so we can return an error without a transaction abort. | |
f6bba201 DC |
2884 | */ |
2885 | if (target_ip == NULL) { | |
2886 | /* | |
2887 | * If there's no space reservation, check the entry will | |
2888 | * fit before actually inserting it. | |
2889 | */ | |
94f3cad5 ES |
2890 | if (!spaceres) { |
2891 | error = xfs_dir_canenter(tp, target_dp, target_name); | |
2892 | if (error) | |
445883e8 | 2893 | goto out_trans_cancel; |
94f3cad5 | 2894 | } |
bc56ad8c | 2895 | } else { |
2896 | /* | |
2897 | * If target exists and it's a directory, check that whether | |
2898 | * it can be destroyed. | |
2899 | */ | |
2900 | if (S_ISDIR(VFS_I(target_ip)->i_mode) && | |
2901 | (!xfs_dir_isempty(target_ip) || | |
2902 | (VFS_I(target_ip)->i_nlink > 2))) { | |
2903 | error = -EEXIST; | |
2904 | goto out_trans_cancel; | |
2905 | } | |
2906 | } | |
2907 | ||
6da1b4b1 DW |
2908 | /* |
2909 | * Lock the AGI buffers we need to handle bumping the nlink of the | |
2910 | * whiteout inode off the unlinked list and to handle dropping the | |
2911 | * nlink of the target inode. Per locking order rules, do this in | |
2912 | * increasing AG order and before directory block allocation tries to | |
2913 | * grab AGFs because we grab AGIs before AGFs. | |
2914 | * | |
2915 | * The (vfs) caller must ensure that if src is a directory then | |
2916 | * target_ip is either null or an empty directory. | |
2917 | */ | |
2918 | for (i = 0; i < num_inodes && inodes[i] != NULL; i++) { | |
2919 | if (inodes[i] == wip || | |
2920 | (inodes[i] == target_ip && | |
2921 | (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) { | |
61021deb DC |
2922 | struct xfs_perag *pag; |
2923 | struct xfs_buf *bp; | |
6da1b4b1 | 2924 | |
61021deb DC |
2925 | pag = xfs_perag_get(mp, |
2926 | XFS_INO_TO_AGNO(mp, inodes[i]->i_ino)); | |
2927 | error = xfs_read_agi(pag, tp, &bp); | |
2928 | xfs_perag_put(pag); | |
6da1b4b1 DW |
2929 | if (error) |
2930 | goto out_trans_cancel; | |
2931 | } | |
2932 | } | |
2933 | ||
bc56ad8c | 2934 | /* |
2935 | * Directory entry creation below may acquire the AGF. Remove | |
2936 | * the whiteout from the unlinked list first to preserve correct | |
2937 | * AGI/AGF locking order. This dirties the transaction so failures | |
2938 | * after this point will abort and log recovery will clean up the | |
2939 | * mess. | |
2940 | * | |
2941 | * For whiteouts, we need to bump the link count on the whiteout | |
2942 | * inode. After this point, we have a real link, clear the tmpfile | |
2943 | * state flag from the inode so it doesn't accidentally get misused | |
2944 | * in future. | |
2945 | */ | |
2946 | if (wip) { | |
f40aadb2 DC |
2947 | struct xfs_perag *pag; |
2948 | ||
bc56ad8c | 2949 | ASSERT(VFS_I(wip)->i_nlink == 0); |
f40aadb2 DC |
2950 | |
2951 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino)); | |
2952 | error = xfs_iunlink_remove(tp, pag, wip); | |
2953 | xfs_perag_put(pag); | |
bc56ad8c | 2954 | if (error) |
2955 | goto out_trans_cancel; | |
2956 | ||
2957 | xfs_bumplink(tp, wip); | |
bc56ad8c | 2958 | VFS_I(wip)->i_state &= ~I_LINKABLE; |
2959 | } | |
2960 | ||
2961 | /* | |
2962 | * Set up the target. | |
2963 | */ | |
2964 | if (target_ip == NULL) { | |
f6bba201 DC |
2965 | /* |
2966 | * If target does not exist and the rename crosses | |
2967 | * directories, adjust the target directory link count | |
2968 | * to account for the ".." reference from the new entry. | |
2969 | */ | |
2970 | error = xfs_dir_createname(tp, target_dp, target_name, | |
381eee69 | 2971 | src_ip->i_ino, spaceres); |
f6bba201 | 2972 | if (error) |
c8eac49e | 2973 | goto out_trans_cancel; |
f6bba201 DC |
2974 | |
2975 | xfs_trans_ichgtime(tp, target_dp, | |
2976 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2977 | ||
2978 | if (new_parent && src_is_directory) { | |
91083269 | 2979 | xfs_bumplink(tp, target_dp); |
f6bba201 DC |
2980 | } |
2981 | } else { /* target_ip != NULL */ | |
f6bba201 DC |
2982 | /* |
2983 | * Link the source inode under the target name. | |
2984 | * If the source inode is a directory and we are moving | |
2985 | * it across directories, its ".." entry will be | |
2986 | * inconsistent until we replace that down below. | |
2987 | * | |
2988 | * In case there is already an entry with the same | |
2989 | * name at the destination directory, remove it first. | |
2990 | */ | |
2991 | error = xfs_dir_replace(tp, target_dp, target_name, | |
381eee69 | 2992 | src_ip->i_ino, spaceres); |
f6bba201 | 2993 | if (error) |
c8eac49e | 2994 | goto out_trans_cancel; |
f6bba201 DC |
2995 | |
2996 | xfs_trans_ichgtime(tp, target_dp, | |
2997 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2998 | ||
2999 | /* | |
3000 | * Decrement the link count on the target since the target | |
3001 | * dir no longer points to it. | |
3002 | */ | |
3003 | error = xfs_droplink(tp, target_ip); | |
3004 | if (error) | |
c8eac49e | 3005 | goto out_trans_cancel; |
f6bba201 DC |
3006 | |
3007 | if (src_is_directory) { | |
3008 | /* | |
3009 | * Drop the link from the old "." entry. | |
3010 | */ | |
3011 | error = xfs_droplink(tp, target_ip); | |
3012 | if (error) | |
c8eac49e | 3013 | goto out_trans_cancel; |
f6bba201 DC |
3014 | } |
3015 | } /* target_ip != NULL */ | |
3016 | ||
3017 | /* | |
3018 | * Remove the source. | |
3019 | */ | |
3020 | if (new_parent && src_is_directory) { | |
3021 | /* | |
3022 | * Rewrite the ".." entry to point to the new | |
3023 | * directory. | |
3024 | */ | |
3025 | error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, | |
381eee69 | 3026 | target_dp->i_ino, spaceres); |
2451337d | 3027 | ASSERT(error != -EEXIST); |
f6bba201 | 3028 | if (error) |
c8eac49e | 3029 | goto out_trans_cancel; |
f6bba201 DC |
3030 | } |
3031 | ||
3032 | /* | |
3033 | * We always want to hit the ctime on the source inode. | |
3034 | * | |
3035 | * This isn't strictly required by the standards since the source | |
3036 | * inode isn't really being changed, but old unix file systems did | |
3037 | * it and some incremental backup programs won't work without it. | |
3038 | */ | |
3039 | xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); | |
3040 | xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); | |
3041 | ||
3042 | /* | |
3043 | * Adjust the link count on src_dp. This is necessary when | |
3044 | * renaming a directory, either within one parent when | |
3045 | * the target existed, or across two parent directories. | |
3046 | */ | |
3047 | if (src_is_directory && (new_parent || target_ip != NULL)) { | |
3048 | ||
3049 | /* | |
3050 | * Decrement link count on src_directory since the | |
3051 | * entry that's moved no longer points to it. | |
3052 | */ | |
3053 | error = xfs_droplink(tp, src_dp); | |
3054 | if (error) | |
c8eac49e | 3055 | goto out_trans_cancel; |
f6bba201 DC |
3056 | } |
3057 | ||
7dcf5c3e DC |
3058 | /* |
3059 | * For whiteouts, we only need to update the source dirent with the | |
3060 | * inode number of the whiteout inode rather than removing it | |
3061 | * altogether. | |
3062 | */ | |
83a21c18 | 3063 | if (wip) |
7dcf5c3e | 3064 | error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, |
381eee69 | 3065 | spaceres); |
83a21c18 | 3066 | else |
7dcf5c3e | 3067 | error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, |
381eee69 | 3068 | spaceres); |
02092a2f | 3069 | |
f6bba201 | 3070 | if (error) |
c8eac49e | 3071 | goto out_trans_cancel; |
f6bba201 | 3072 | |
f6bba201 DC |
3073 | xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
3074 | xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); | |
3075 | if (new_parent) | |
3076 | xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); | |
f6bba201 | 3077 | |
c9cfdb38 | 3078 | error = xfs_finish_rename(tp); |
7dcf5c3e | 3079 | if (wip) |
44a8736b | 3080 | xfs_irele(wip); |
7dcf5c3e | 3081 | return error; |
f6bba201 | 3082 | |
445883e8 | 3083 | out_trans_cancel: |
4906e215 | 3084 | xfs_trans_cancel(tp); |
253f4911 | 3085 | out_release_wip: |
7dcf5c3e | 3086 | if (wip) |
44a8736b | 3087 | xfs_irele(wip); |
41667260 DW |
3088 | if (error == -ENOSPC && nospace_error) |
3089 | error = nospace_error; | |
f6bba201 DC |
3090 | return error; |
3091 | } | |
3092 | ||
e6187b34 DC |
3093 | static int |
3094 | xfs_iflush( | |
93848a99 CH |
3095 | struct xfs_inode *ip, |
3096 | struct xfs_buf *bp) | |
1da177e4 | 3097 | { |
93848a99 CH |
3098 | struct xfs_inode_log_item *iip = ip->i_itemp; |
3099 | struct xfs_dinode *dip; | |
3100 | struct xfs_mount *mp = ip->i_mount; | |
f2019299 | 3101 | int error; |
1da177e4 | 3102 | |
579aa9ca | 3103 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
718ecc50 | 3104 | ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING)); |
f7e67b20 | 3105 | ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE || |
daf83964 | 3106 | ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
90c60e16 | 3107 | ASSERT(iip->ili_item.li_buf == bp); |
1da177e4 | 3108 | |
88ee2df7 | 3109 | dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 3110 | |
f2019299 BF |
3111 | /* |
3112 | * We don't flush the inode if any of the following checks fail, but we | |
3113 | * do still update the log item and attach to the backing buffer as if | |
3114 | * the flush happened. This is a formality to facilitate predictable | |
3115 | * error handling as the caller will shutdown and fail the buffer. | |
3116 | */ | |
3117 | error = -EFSCORRUPTED; | |
69ef921b | 3118 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
9e24cfd0 | 3119 | mp, XFS_ERRTAG_IFLUSH_1)) { |
6a19d939 | 3120 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
78b0f58b | 3121 | "%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT, |
6a19d939 | 3122 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); |
f2019299 | 3123 | goto flush_out; |
1da177e4 | 3124 | } |
c19b3b05 | 3125 | if (S_ISREG(VFS_I(ip)->i_mode)) { |
1da177e4 | 3126 | if (XFS_TEST_ERROR( |
f7e67b20 CH |
3127 | ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS && |
3128 | ip->i_df.if_format != XFS_DINODE_FMT_BTREE, | |
9e24cfd0 | 3129 | mp, XFS_ERRTAG_IFLUSH_3)) { |
6a19d939 | 3130 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
78b0f58b | 3131 | "%s: Bad regular inode %llu, ptr "PTR_FMT, |
6a19d939 | 3132 | __func__, ip->i_ino, ip); |
f2019299 | 3133 | goto flush_out; |
1da177e4 | 3134 | } |
c19b3b05 | 3135 | } else if (S_ISDIR(VFS_I(ip)->i_mode)) { |
1da177e4 | 3136 | if (XFS_TEST_ERROR( |
f7e67b20 CH |
3137 | ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS && |
3138 | ip->i_df.if_format != XFS_DINODE_FMT_BTREE && | |
3139 | ip->i_df.if_format != XFS_DINODE_FMT_LOCAL, | |
9e24cfd0 | 3140 | mp, XFS_ERRTAG_IFLUSH_4)) { |
6a19d939 | 3141 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
78b0f58b | 3142 | "%s: Bad directory inode %llu, ptr "PTR_FMT, |
6a19d939 | 3143 | __func__, ip->i_ino, ip); |
f2019299 | 3144 | goto flush_out; |
1da177e4 LT |
3145 | } |
3146 | } | |
2ed5b09b | 3147 | if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) > |
6e73a545 | 3148 | ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { |
6a19d939 | 3149 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
755c38ff CB |
3150 | "%s: detected corrupt incore inode %llu, " |
3151 | "total extents = %llu nblocks = %lld, ptr "PTR_FMT, | |
6a19d939 | 3152 | __func__, ip->i_ino, |
2ed5b09b | 3153 | ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af), |
6e73a545 | 3154 | ip->i_nblocks, ip); |
f2019299 | 3155 | goto flush_out; |
1da177e4 | 3156 | } |
7821ea30 | 3157 | if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize, |
9e24cfd0 | 3158 | mp, XFS_ERRTAG_IFLUSH_6)) { |
6a19d939 | 3159 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
78b0f58b | 3160 | "%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT, |
7821ea30 | 3161 | __func__, ip->i_ino, ip->i_forkoff, ip); |
f2019299 | 3162 | goto flush_out; |
1da177e4 | 3163 | } |
e60896d8 | 3164 | |
1da177e4 | 3165 | /* |
965e0a1a CH |
3166 | * Inode item log recovery for v2 inodes are dependent on the flushiter |
3167 | * count for correct sequencing. We bump the flush iteration count so | |
3168 | * we can detect flushes which postdate a log record during recovery. | |
3169 | * This is redundant as we now log every change and hence this can't | |
3170 | * happen but we need to still do it to ensure backwards compatibility | |
3171 | * with old kernels that predate logging all inode changes. | |
1da177e4 | 3172 | */ |
38c26bfd | 3173 | if (!xfs_has_v3inodes(mp)) |
965e0a1a | 3174 | ip->i_flushiter++; |
1da177e4 | 3175 | |
0f45a1b2 CH |
3176 | /* |
3177 | * If there are inline format data / attr forks attached to this inode, | |
3178 | * make sure they are not corrupt. | |
3179 | */ | |
f7e67b20 | 3180 | if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL && |
0f45a1b2 CH |
3181 | xfs_ifork_verify_local_data(ip)) |
3182 | goto flush_out; | |
932b42c6 | 3183 | if (xfs_inode_has_attr_fork(ip) && |
2ed5b09b | 3184 | ip->i_af.if_format == XFS_DINODE_FMT_LOCAL && |
0f45a1b2 | 3185 | xfs_ifork_verify_local_attr(ip)) |
f2019299 | 3186 | goto flush_out; |
005c5db8 | 3187 | |
1da177e4 | 3188 | /* |
3987848c DC |
3189 | * Copy the dirty parts of the inode into the on-disk inode. We always |
3190 | * copy out the core of the inode, because if the inode is dirty at all | |
3191 | * the core must be. | |
1da177e4 | 3192 | */ |
93f958f9 | 3193 | xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); |
1da177e4 LT |
3194 | |
3195 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
38c26bfd | 3196 | if (!xfs_has_v3inodes(mp)) { |
ee7b83fd CH |
3197 | if (ip->i_flushiter == DI_MAX_FLUSH) |
3198 | ip->i_flushiter = 0; | |
3199 | } | |
1da177e4 | 3200 | |
005c5db8 | 3201 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); |
932b42c6 | 3202 | if (xfs_inode_has_attr_fork(ip)) |
005c5db8 | 3203 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); |
1da177e4 LT |
3204 | |
3205 | /* | |
f5d8d5c4 CH |
3206 | * We've recorded everything logged in the inode, so we'd like to clear |
3207 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
3208 | * However, we can't stop logging all this information until the data | |
3209 | * we've copied into the disk buffer is written to disk. If we did we | |
3210 | * might overwrite the copy of the inode in the log with all the data | |
3211 | * after re-logging only part of it, and in the face of a crash we | |
3212 | * wouldn't have all the data we need to recover. | |
1da177e4 | 3213 | * |
f5d8d5c4 CH |
3214 | * What we do is move the bits to the ili_last_fields field. When |
3215 | * logging the inode, these bits are moved back to the ili_fields field. | |
664ffb8a CH |
3216 | * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since |
3217 | * we know that the information those bits represent is permanently on | |
f5d8d5c4 CH |
3218 | * disk. As long as the flush completes before the inode is logged |
3219 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 3220 | */ |
f2019299 BF |
3221 | error = 0; |
3222 | flush_out: | |
1319ebef | 3223 | spin_lock(&iip->ili_lock); |
93848a99 CH |
3224 | iip->ili_last_fields = iip->ili_fields; |
3225 | iip->ili_fields = 0; | |
fc0561ce | 3226 | iip->ili_fsync_fields = 0; |
1319ebef | 3227 | spin_unlock(&iip->ili_lock); |
1da177e4 | 3228 | |
1319ebef DC |
3229 | /* |
3230 | * Store the current LSN of the inode so that we can tell whether the | |
664ffb8a | 3231 | * item has moved in the AIL from xfs_buf_inode_iodone(). |
1319ebef | 3232 | */ |
93848a99 CH |
3233 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
3234 | &iip->ili_item.li_lsn); | |
1da177e4 | 3235 | |
93848a99 CH |
3236 | /* generate the checksum. */ |
3237 | xfs_dinode_calc_crc(mp, dip); | |
f2019299 | 3238 | return error; |
1da177e4 | 3239 | } |
44a8736b | 3240 | |
e6187b34 DC |
3241 | /* |
3242 | * Non-blocking flush of dirty inode metadata into the backing buffer. | |
3243 | * | |
3244 | * The caller must have a reference to the inode and hold the cluster buffer | |
3245 | * locked. The function will walk across all the inodes on the cluster buffer it | |
3246 | * can find and lock without blocking, and flush them to the cluster buffer. | |
3247 | * | |
5717ea4d DC |
3248 | * On successful flushing of at least one inode, the caller must write out the |
3249 | * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and | |
3250 | * the caller needs to release the buffer. On failure, the filesystem will be | |
3251 | * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED | |
3252 | * will be returned. | |
e6187b34 DC |
3253 | */ |
3254 | int | |
3255 | xfs_iflush_cluster( | |
e6187b34 DC |
3256 | struct xfs_buf *bp) |
3257 | { | |
5717ea4d DC |
3258 | struct xfs_mount *mp = bp->b_mount; |
3259 | struct xfs_log_item *lip, *n; | |
3260 | struct xfs_inode *ip; | |
3261 | struct xfs_inode_log_item *iip; | |
e6187b34 | 3262 | int clcount = 0; |
5717ea4d | 3263 | int error = 0; |
e6187b34 | 3264 | |
5717ea4d DC |
3265 | /* |
3266 | * We must use the safe variant here as on shutdown xfs_iflush_abort() | |
d2d7c047 | 3267 | * will remove itself from the list. |
5717ea4d DC |
3268 | */ |
3269 | list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) { | |
3270 | iip = (struct xfs_inode_log_item *)lip; | |
3271 | ip = iip->ili_inode; | |
e6187b34 DC |
3272 | |
3273 | /* | |
5717ea4d | 3274 | * Quick and dirty check to avoid locks if possible. |
e6187b34 | 3275 | */ |
718ecc50 | 3276 | if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) |
5717ea4d DC |
3277 | continue; |
3278 | if (xfs_ipincount(ip)) | |
e6187b34 | 3279 | continue; |
e6187b34 DC |
3280 | |
3281 | /* | |
5717ea4d DC |
3282 | * The inode is still attached to the buffer, which means it is |
3283 | * dirty but reclaim might try to grab it. Check carefully for | |
3284 | * that, and grab the ilock while still holding the i_flags_lock | |
3285 | * to guarantee reclaim will not be able to reclaim this inode | |
3286 | * once we drop the i_flags_lock. | |
e6187b34 | 3287 | */ |
5717ea4d DC |
3288 | spin_lock(&ip->i_flags_lock); |
3289 | ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE)); | |
718ecc50 | 3290 | if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) { |
5717ea4d DC |
3291 | spin_unlock(&ip->i_flags_lock); |
3292 | continue; | |
e6187b34 | 3293 | } |
e6187b34 DC |
3294 | |
3295 | /* | |
5717ea4d DC |
3296 | * ILOCK will pin the inode against reclaim and prevent |
3297 | * concurrent transactions modifying the inode while we are | |
718ecc50 DC |
3298 | * flushing the inode. If we get the lock, set the flushing |
3299 | * state before we drop the i_flags_lock. | |
e6187b34 | 3300 | */ |
5717ea4d DC |
3301 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { |
3302 | spin_unlock(&ip->i_flags_lock); | |
e6187b34 | 3303 | continue; |
5717ea4d | 3304 | } |
718ecc50 | 3305 | __xfs_iflags_set(ip, XFS_IFLUSHING); |
5717ea4d | 3306 | spin_unlock(&ip->i_flags_lock); |
e6187b34 | 3307 | |
e6187b34 | 3308 | /* |
5717ea4d DC |
3309 | * Abort flushing this inode if we are shut down because the |
3310 | * inode may not currently be in the AIL. This can occur when | |
3311 | * log I/O failure unpins the inode without inserting into the | |
3312 | * AIL, leaving a dirty/unpinned inode attached to the buffer | |
3313 | * that otherwise looks like it should be flushed. | |
e6187b34 | 3314 | */ |
01728b44 | 3315 | if (xlog_is_shutdown(mp->m_log)) { |
5717ea4d | 3316 | xfs_iunpin_wait(ip); |
5717ea4d DC |
3317 | xfs_iflush_abort(ip); |
3318 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
3319 | error = -EIO; | |
e6187b34 DC |
3320 | continue; |
3321 | } | |
3322 | ||
5717ea4d DC |
3323 | /* don't block waiting on a log force to unpin dirty inodes */ |
3324 | if (xfs_ipincount(ip)) { | |
718ecc50 | 3325 | xfs_iflags_clear(ip, XFS_IFLUSHING); |
5717ea4d DC |
3326 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
3327 | continue; | |
e6187b34 | 3328 | } |
e6187b34 | 3329 | |
5717ea4d DC |
3330 | if (!xfs_inode_clean(ip)) |
3331 | error = xfs_iflush(ip, bp); | |
3332 | else | |
718ecc50 | 3333 | xfs_iflags_clear(ip, XFS_IFLUSHING); |
5717ea4d DC |
3334 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
3335 | if (error) | |
3336 | break; | |
3337 | clcount++; | |
e6187b34 DC |
3338 | } |
3339 | ||
e6187b34 | 3340 | if (error) { |
01728b44 DC |
3341 | /* |
3342 | * Shutdown first so we kill the log before we release this | |
3343 | * buffer. If it is an INODE_ALLOC buffer and pins the tail | |
3344 | * of the log, failing it before the _log_ is shut down can | |
3345 | * result in the log tail being moved forward in the journal | |
3346 | * on disk because log writes can still be taking place. Hence | |
3347 | * unpinning the tail will allow the ICREATE intent to be | |
3348 | * removed from the log an recovery will fail with uninitialised | |
3349 | * inode cluster buffers. | |
3350 | */ | |
3351 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
e6187b34 DC |
3352 | bp->b_flags |= XBF_ASYNC; |
3353 | xfs_buf_ioend_fail(bp); | |
5717ea4d | 3354 | return error; |
e6187b34 | 3355 | } |
5717ea4d DC |
3356 | |
3357 | if (!clcount) | |
3358 | return -EAGAIN; | |
3359 | ||
3360 | XFS_STATS_INC(mp, xs_icluster_flushcnt); | |
3361 | XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); | |
3362 | return 0; | |
3363 | ||
e6187b34 DC |
3364 | } |
3365 | ||
44a8736b DW |
3366 | /* Release an inode. */ |
3367 | void | |
3368 | xfs_irele( | |
3369 | struct xfs_inode *ip) | |
3370 | { | |
3371 | trace_xfs_irele(ip, _RET_IP_); | |
3372 | iput(VFS_I(ip)); | |
3373 | } | |
54fbdd10 CH |
3374 | |
3375 | /* | |
3376 | * Ensure all commited transactions touching the inode are written to the log. | |
3377 | */ | |
3378 | int | |
3379 | xfs_log_force_inode( | |
3380 | struct xfs_inode *ip) | |
3381 | { | |
5f9b4b0d | 3382 | xfs_csn_t seq = 0; |
54fbdd10 CH |
3383 | |
3384 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
3385 | if (xfs_ipincount(ip)) | |
5f9b4b0d | 3386 | seq = ip->i_itemp->ili_commit_seq; |
54fbdd10 CH |
3387 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
3388 | ||
5f9b4b0d | 3389 | if (!seq) |
54fbdd10 | 3390 | return 0; |
5f9b4b0d | 3391 | return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL); |
54fbdd10 | 3392 | } |
e2aaee9c DW |
3393 | |
3394 | /* | |
3395 | * Grab the exclusive iolock for a data copy from src to dest, making sure to | |
3396 | * abide vfs locking order (lowest pointer value goes first) and breaking the | |
3397 | * layout leases before proceeding. The loop is needed because we cannot call | |
3398 | * the blocking break_layout() with the iolocks held, and therefore have to | |
3399 | * back out both locks. | |
3400 | */ | |
3401 | static int | |
3402 | xfs_iolock_two_inodes_and_break_layout( | |
3403 | struct inode *src, | |
3404 | struct inode *dest) | |
3405 | { | |
3406 | int error; | |
3407 | ||
3408 | if (src > dest) | |
3409 | swap(src, dest); | |
3410 | ||
3411 | retry: | |
3412 | /* Wait to break both inodes' layouts before we start locking. */ | |
3413 | error = break_layout(src, true); | |
3414 | if (error) | |
3415 | return error; | |
3416 | if (src != dest) { | |
3417 | error = break_layout(dest, true); | |
3418 | if (error) | |
3419 | return error; | |
3420 | } | |
3421 | ||
3422 | /* Lock one inode and make sure nobody got in and leased it. */ | |
3423 | inode_lock(src); | |
3424 | error = break_layout(src, false); | |
3425 | if (error) { | |
3426 | inode_unlock(src); | |
3427 | if (error == -EWOULDBLOCK) | |
3428 | goto retry; | |
3429 | return error; | |
3430 | } | |
3431 | ||
3432 | if (src == dest) | |
3433 | return 0; | |
3434 | ||
3435 | /* Lock the other inode and make sure nobody got in and leased it. */ | |
3436 | inode_lock_nested(dest, I_MUTEX_NONDIR2); | |
3437 | error = break_layout(dest, false); | |
3438 | if (error) { | |
3439 | inode_unlock(src); | |
3440 | inode_unlock(dest); | |
3441 | if (error == -EWOULDBLOCK) | |
3442 | goto retry; | |
3443 | return error; | |
3444 | } | |
3445 | ||
3446 | return 0; | |
3447 | } | |
3448 | ||
13f9e267 SR |
3449 | static int |
3450 | xfs_mmaplock_two_inodes_and_break_dax_layout( | |
3451 | struct xfs_inode *ip1, | |
3452 | struct xfs_inode *ip2) | |
3453 | { | |
3454 | int error; | |
3455 | bool retry; | |
3456 | struct page *page; | |
3457 | ||
3458 | if (ip1->i_ino > ip2->i_ino) | |
3459 | swap(ip1, ip2); | |
3460 | ||
3461 | again: | |
3462 | retry = false; | |
3463 | /* Lock the first inode */ | |
3464 | xfs_ilock(ip1, XFS_MMAPLOCK_EXCL); | |
3465 | error = xfs_break_dax_layouts(VFS_I(ip1), &retry); | |
3466 | if (error || retry) { | |
3467 | xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL); | |
3468 | if (error == 0 && retry) | |
3469 | goto again; | |
3470 | return error; | |
3471 | } | |
3472 | ||
3473 | if (ip1 == ip2) | |
3474 | return 0; | |
3475 | ||
3476 | /* Nested lock the second inode */ | |
3477 | xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1)); | |
3478 | /* | |
3479 | * We cannot use xfs_break_dax_layouts() directly here because it may | |
3480 | * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable | |
3481 | * for this nested lock case. | |
3482 | */ | |
3483 | page = dax_layout_busy_page(VFS_I(ip2)->i_mapping); | |
3484 | if (page && page_ref_count(page) != 1) { | |
3485 | xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL); | |
3486 | xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL); | |
3487 | goto again; | |
3488 | } | |
3489 | ||
3490 | return 0; | |
3491 | } | |
3492 | ||
e2aaee9c DW |
3493 | /* |
3494 | * Lock two inodes so that userspace cannot initiate I/O via file syscalls or | |
3495 | * mmap activity. | |
3496 | */ | |
3497 | int | |
3498 | xfs_ilock2_io_mmap( | |
3499 | struct xfs_inode *ip1, | |
3500 | struct xfs_inode *ip2) | |
3501 | { | |
3502 | int ret; | |
3503 | ||
3504 | ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2)); | |
3505 | if (ret) | |
3506 | return ret; | |
13f9e267 SR |
3507 | |
3508 | if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) { | |
3509 | ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2); | |
3510 | if (ret) { | |
3511 | inode_unlock(VFS_I(ip2)); | |
3512 | if (ip1 != ip2) | |
3513 | inode_unlock(VFS_I(ip1)); | |
3514 | return ret; | |
3515 | } | |
3516 | } else | |
3517 | filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping, | |
3518 | VFS_I(ip2)->i_mapping); | |
3519 | ||
e2aaee9c DW |
3520 | return 0; |
3521 | } | |
3522 | ||
3523 | /* Unlock both inodes to allow IO and mmap activity. */ | |
3524 | void | |
3525 | xfs_iunlock2_io_mmap( | |
3526 | struct xfs_inode *ip1, | |
3527 | struct xfs_inode *ip2) | |
3528 | { | |
13f9e267 SR |
3529 | if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) { |
3530 | xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL); | |
3531 | if (ip1 != ip2) | |
3532 | xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL); | |
3533 | } else | |
3534 | filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping, | |
3535 | VFS_I(ip2)->i_mapping); | |
3536 | ||
e2aaee9c | 3537 | inode_unlock(VFS_I(ip2)); |
d2c292d8 | 3538 | if (ip1 != ip2) |
e2aaee9c DW |
3539 | inode_unlock(VFS_I(ip1)); |
3540 | } |