]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
40ebd81d RD |
18 | #include <linux/log2.h> |
19 | ||
1da177e4 | 20 | #include "xfs.h" |
a844f451 | 21 | #include "xfs_fs.h" |
1da177e4 | 22 | #include "xfs_types.h" |
1da177e4 | 23 | #include "xfs_log.h" |
a844f451 | 24 | #include "xfs_inum.h" |
1da177e4 LT |
25 | #include "xfs_trans.h" |
26 | #include "xfs_trans_priv.h" | |
27 | #include "xfs_sb.h" | |
28 | #include "xfs_ag.h" | |
1da177e4 | 29 | #include "xfs_mount.h" |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
a844f451 | 33 | #include "xfs_attr_sf.h" |
1da177e4 | 34 | #include "xfs_dinode.h" |
1da177e4 | 35 | #include "xfs_inode.h" |
1da177e4 | 36 | #include "xfs_buf_item.h" |
a844f451 NS |
37 | #include "xfs_inode_item.h" |
38 | #include "xfs_btree.h" | |
39 | #include "xfs_alloc.h" | |
40 | #include "xfs_ialloc.h" | |
41 | #include "xfs_bmap.h" | |
1da177e4 | 42 | #include "xfs_error.h" |
1da177e4 | 43 | #include "xfs_utils.h" |
1da177e4 | 44 | #include "xfs_quota.h" |
2a82b8be | 45 | #include "xfs_filestream.h" |
739bfb2a | 46 | #include "xfs_vnodeops.h" |
0b1b213f | 47 | #include "xfs_trace.h" |
1da177e4 | 48 | |
1da177e4 LT |
49 | kmem_zone_t *xfs_ifork_zone; |
50 | kmem_zone_t *xfs_inode_zone; | |
1da177e4 LT |
51 | |
52 | /* | |
8f04c47a | 53 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
54 | * freed from a file in a single transaction. |
55 | */ | |
56 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
57 | ||
58 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); | |
59 | STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); | |
60 | STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); | |
61 | STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); | |
62 | ||
2a0ec1d9 DC |
63 | /* |
64 | * helper function to extract extent size hint from inode | |
65 | */ | |
66 | xfs_extlen_t | |
67 | xfs_get_extsz_hint( | |
68 | struct xfs_inode *ip) | |
69 | { | |
70 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
71 | return ip->i_d.di_extsize; | |
72 | if (XFS_IS_REALTIME_INODE(ip)) | |
73 | return ip->i_mount->m_sb.sb_rextsize; | |
74 | return 0; | |
75 | } | |
76 | ||
fa96acad DC |
77 | /* |
78 | * This is a wrapper routine around the xfs_ilock() routine used to centralize | |
79 | * some grungy code. It is used in places that wish to lock the inode solely | |
80 | * for reading the extents. The reason these places can't just call | |
81 | * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the | |
82 | * extents from disk for a file in b-tree format. If the inode is in b-tree | |
83 | * format, then we need to lock the inode exclusively until the extents are read | |
84 | * in. Locking it exclusively all the time would limit our parallelism | |
85 | * unnecessarily, though. What we do instead is check to see if the extents | |
86 | * have been read in yet, and only lock the inode exclusively if they have not. | |
87 | * | |
88 | * The function returns a value which should be given to the corresponding | |
89 | * xfs_iunlock_map_shared(). This value is the mode in which the lock was | |
90 | * actually taken. | |
91 | */ | |
92 | uint | |
93 | xfs_ilock_map_shared( | |
94 | xfs_inode_t *ip) | |
95 | { | |
96 | uint lock_mode; | |
97 | ||
98 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
99 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
100 | lock_mode = XFS_ILOCK_EXCL; | |
101 | } else { | |
102 | lock_mode = XFS_ILOCK_SHARED; | |
103 | } | |
104 | ||
105 | xfs_ilock(ip, lock_mode); | |
106 | ||
107 | return lock_mode; | |
108 | } | |
109 | ||
110 | /* | |
111 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
112 | * All it does is call xfs_iunlock() with the given lock_mode. | |
113 | */ | |
114 | void | |
115 | xfs_iunlock_map_shared( | |
116 | xfs_inode_t *ip, | |
117 | unsigned int lock_mode) | |
118 | { | |
119 | xfs_iunlock(ip, lock_mode); | |
120 | } | |
121 | ||
122 | /* | |
123 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
124 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
125 | * allows either or both of the locks to be obtained. | |
126 | * | |
127 | * The 2 locks should always be ordered so that the IO lock is | |
128 | * obtained first in order to prevent deadlock. | |
129 | * | |
130 | * ip -- the inode being locked | |
131 | * lock_flags -- this parameter indicates the inode's locks | |
132 | * to be locked. It can be: | |
133 | * XFS_IOLOCK_SHARED, | |
134 | * XFS_IOLOCK_EXCL, | |
135 | * XFS_ILOCK_SHARED, | |
136 | * XFS_ILOCK_EXCL, | |
137 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
138 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
139 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
140 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
141 | */ | |
142 | void | |
143 | xfs_ilock( | |
144 | xfs_inode_t *ip, | |
145 | uint lock_flags) | |
146 | { | |
147 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
148 | ||
149 | /* | |
150 | * You can't set both SHARED and EXCL for the same lock, | |
151 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
152 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
153 | */ | |
154 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
155 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
156 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
157 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
158 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
159 | ||
160 | if (lock_flags & XFS_IOLOCK_EXCL) | |
161 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
162 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
163 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
164 | ||
165 | if (lock_flags & XFS_ILOCK_EXCL) | |
166 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
167 | else if (lock_flags & XFS_ILOCK_SHARED) | |
168 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
169 | } | |
170 | ||
171 | /* | |
172 | * This is just like xfs_ilock(), except that the caller | |
173 | * is guaranteed not to sleep. It returns 1 if it gets | |
174 | * the requested locks and 0 otherwise. If the IO lock is | |
175 | * obtained but the inode lock cannot be, then the IO lock | |
176 | * is dropped before returning. | |
177 | * | |
178 | * ip -- the inode being locked | |
179 | * lock_flags -- this parameter indicates the inode's locks to be | |
180 | * to be locked. See the comment for xfs_ilock() for a list | |
181 | * of valid values. | |
182 | */ | |
183 | int | |
184 | xfs_ilock_nowait( | |
185 | xfs_inode_t *ip, | |
186 | uint lock_flags) | |
187 | { | |
188 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
189 | ||
190 | /* | |
191 | * You can't set both SHARED and EXCL for the same lock, | |
192 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
193 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
194 | */ | |
195 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
196 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
197 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
198 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
199 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
200 | ||
201 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
202 | if (!mrtryupdate(&ip->i_iolock)) | |
203 | goto out; | |
204 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
205 | if (!mrtryaccess(&ip->i_iolock)) | |
206 | goto out; | |
207 | } | |
208 | if (lock_flags & XFS_ILOCK_EXCL) { | |
209 | if (!mrtryupdate(&ip->i_lock)) | |
210 | goto out_undo_iolock; | |
211 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
212 | if (!mrtryaccess(&ip->i_lock)) | |
213 | goto out_undo_iolock; | |
214 | } | |
215 | return 1; | |
216 | ||
217 | out_undo_iolock: | |
218 | if (lock_flags & XFS_IOLOCK_EXCL) | |
219 | mrunlock_excl(&ip->i_iolock); | |
220 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
221 | mrunlock_shared(&ip->i_iolock); | |
222 | out: | |
223 | return 0; | |
224 | } | |
225 | ||
226 | /* | |
227 | * xfs_iunlock() is used to drop the inode locks acquired with | |
228 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
229 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
230 | * that we know which locks to drop. | |
231 | * | |
232 | * ip -- the inode being unlocked | |
233 | * lock_flags -- this parameter indicates the inode's locks to be | |
234 | * to be unlocked. See the comment for xfs_ilock() for a list | |
235 | * of valid values for this parameter. | |
236 | * | |
237 | */ | |
238 | void | |
239 | xfs_iunlock( | |
240 | xfs_inode_t *ip, | |
241 | uint lock_flags) | |
242 | { | |
243 | /* | |
244 | * You can't set both SHARED and EXCL for the same lock, | |
245 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
246 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
247 | */ | |
248 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
249 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
250 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
251 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
252 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
253 | ASSERT(lock_flags != 0); | |
254 | ||
255 | if (lock_flags & XFS_IOLOCK_EXCL) | |
256 | mrunlock_excl(&ip->i_iolock); | |
257 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
258 | mrunlock_shared(&ip->i_iolock); | |
259 | ||
260 | if (lock_flags & XFS_ILOCK_EXCL) | |
261 | mrunlock_excl(&ip->i_lock); | |
262 | else if (lock_flags & XFS_ILOCK_SHARED) | |
263 | mrunlock_shared(&ip->i_lock); | |
264 | ||
265 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
266 | } | |
267 | ||
268 | /* | |
269 | * give up write locks. the i/o lock cannot be held nested | |
270 | * if it is being demoted. | |
271 | */ | |
272 | void | |
273 | xfs_ilock_demote( | |
274 | xfs_inode_t *ip, | |
275 | uint lock_flags) | |
276 | { | |
277 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
278 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
279 | ||
280 | if (lock_flags & XFS_ILOCK_EXCL) | |
281 | mrdemote(&ip->i_lock); | |
282 | if (lock_flags & XFS_IOLOCK_EXCL) | |
283 | mrdemote(&ip->i_iolock); | |
284 | ||
285 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
286 | } | |
287 | ||
288 | #ifdef DEBUG | |
289 | int | |
290 | xfs_isilocked( | |
291 | xfs_inode_t *ip, | |
292 | uint lock_flags) | |
293 | { | |
294 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
295 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
296 | return !!ip->i_lock.mr_writer; | |
297 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
298 | } | |
299 | ||
300 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { | |
301 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
302 | return !!ip->i_iolock.mr_writer; | |
303 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
304 | } | |
305 | ||
306 | ASSERT(0); | |
307 | return 0; | |
308 | } | |
309 | #endif | |
310 | ||
311 | void | |
312 | __xfs_iflock( | |
313 | struct xfs_inode *ip) | |
314 | { | |
315 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
316 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
317 | ||
318 | do { | |
319 | prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
320 | if (xfs_isiflocked(ip)) | |
321 | io_schedule(); | |
322 | } while (!xfs_iflock_nowait(ip)); | |
323 | ||
324 | finish_wait(wq, &wait.wait); | |
325 | } | |
326 | ||
1da177e4 LT |
327 | #ifdef DEBUG |
328 | /* | |
329 | * Make sure that the extents in the given memory buffer | |
330 | * are valid. | |
331 | */ | |
332 | STATIC void | |
333 | xfs_validate_extents( | |
4eea22f0 | 334 | xfs_ifork_t *ifp, |
1da177e4 | 335 | int nrecs, |
1da177e4 LT |
336 | xfs_exntfmt_t fmt) |
337 | { | |
338 | xfs_bmbt_irec_t irec; | |
a6f64d4a | 339 | xfs_bmbt_rec_host_t rec; |
1da177e4 LT |
340 | int i; |
341 | ||
342 | for (i = 0; i < nrecs; i++) { | |
a6f64d4a CH |
343 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
344 | rec.l0 = get_unaligned(&ep->l0); | |
345 | rec.l1 = get_unaligned(&ep->l1); | |
346 | xfs_bmbt_get_all(&rec, &irec); | |
1da177e4 LT |
347 | if (fmt == XFS_EXTFMT_NOSTATE) |
348 | ASSERT(irec.br_state == XFS_EXT_NORM); | |
1da177e4 LT |
349 | } |
350 | } | |
351 | #else /* DEBUG */ | |
a6f64d4a | 352 | #define xfs_validate_extents(ifp, nrecs, fmt) |
1da177e4 LT |
353 | #endif /* DEBUG */ |
354 | ||
355 | /* | |
356 | * Check that none of the inode's in the buffer have a next | |
357 | * unlinked field of 0. | |
358 | */ | |
359 | #if defined(DEBUG) | |
360 | void | |
361 | xfs_inobp_check( | |
362 | xfs_mount_t *mp, | |
363 | xfs_buf_t *bp) | |
364 | { | |
365 | int i; | |
366 | int j; | |
367 | xfs_dinode_t *dip; | |
368 | ||
369 | j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; | |
370 | ||
371 | for (i = 0; i < j; i++) { | |
372 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
373 | i * mp->m_sb.sb_inodesize); | |
374 | if (!dip->di_next_unlinked) { | |
53487786 DC |
375 | xfs_alert(mp, |
376 | "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.", | |
1da177e4 LT |
377 | bp); |
378 | ASSERT(dip->di_next_unlinked); | |
379 | } | |
380 | } | |
381 | } | |
382 | #endif | |
383 | ||
4ae29b43 | 384 | /* |
475ee413 CH |
385 | * This routine is called to map an inode to the buffer containing the on-disk |
386 | * version of the inode. It returns a pointer to the buffer containing the | |
387 | * on-disk inode in the bpp parameter, and in the dipp parameter it returns a | |
388 | * pointer to the on-disk inode within that buffer. | |
389 | * | |
390 | * If a non-zero error is returned, then the contents of bpp and dipp are | |
391 | * undefined. | |
4ae29b43 | 392 | */ |
475ee413 | 393 | int |
4ae29b43 | 394 | xfs_imap_to_bp( |
475ee413 CH |
395 | struct xfs_mount *mp, |
396 | struct xfs_trans *tp, | |
397 | struct xfs_imap *imap, | |
398 | struct xfs_dinode **dipp, | |
399 | struct xfs_buf **bpp, | |
400 | uint buf_flags, | |
401 | uint iget_flags) | |
4ae29b43 | 402 | { |
475ee413 CH |
403 | struct xfs_buf *bp; |
404 | int error; | |
405 | int i; | |
406 | int ni; | |
4ae29b43 | 407 | |
611c9946 | 408 | buf_flags |= XBF_UNMAPPED; |
4ae29b43 | 409 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno, |
a3f74ffb | 410 | (int)imap->im_len, buf_flags, &bp); |
4ae29b43 | 411 | if (error) { |
a3f74ffb | 412 | if (error != EAGAIN) { |
0b932ccc DC |
413 | xfs_warn(mp, |
414 | "%s: xfs_trans_read_buf() returned error %d.", | |
415 | __func__, error); | |
a3f74ffb | 416 | } else { |
0cadda1c | 417 | ASSERT(buf_flags & XBF_TRYLOCK); |
a3f74ffb | 418 | } |
4ae29b43 DC |
419 | return error; |
420 | } | |
421 | ||
422 | /* | |
423 | * Validate the magic number and version of every inode in the buffer | |
424 | * (if DEBUG kernel) or the first inode in the buffer, otherwise. | |
425 | */ | |
426 | #ifdef DEBUG | |
427 | ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog; | |
428 | #else /* usual case */ | |
429 | ni = 1; | |
430 | #endif | |
431 | ||
432 | for (i = 0; i < ni; i++) { | |
433 | int di_ok; | |
434 | xfs_dinode_t *dip; | |
435 | ||
436 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
437 | (i << mp->m_sb.sb_inodelog)); | |
69ef921b | 438 | di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) && |
81591fe2 | 439 | XFS_DINODE_GOOD_VERSION(dip->di_version); |
4ae29b43 DC |
440 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, |
441 | XFS_ERRTAG_ITOBP_INOTOBP, | |
442 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
1920779e | 443 | if (iget_flags & XFS_IGET_UNTRUSTED) { |
4ae29b43 DC |
444 | xfs_trans_brelse(tp, bp); |
445 | return XFS_ERROR(EINVAL); | |
446 | } | |
475ee413 CH |
447 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_HIGH, |
448 | mp, dip); | |
4ae29b43 | 449 | #ifdef DEBUG |
0b932ccc DC |
450 | xfs_emerg(mp, |
451 | "bad inode magic/vsn daddr %lld #%d (magic=%x)", | |
4ae29b43 | 452 | (unsigned long long)imap->im_blkno, i, |
81591fe2 | 453 | be16_to_cpu(dip->di_magic)); |
0b932ccc | 454 | ASSERT(0); |
4ae29b43 DC |
455 | #endif |
456 | xfs_trans_brelse(tp, bp); | |
457 | return XFS_ERROR(EFSCORRUPTED); | |
458 | } | |
459 | } | |
460 | ||
461 | xfs_inobp_check(mp, bp); | |
475ee413 | 462 | |
4ae29b43 | 463 | *bpp = bp; |
475ee413 | 464 | *dipp = (struct xfs_dinode *)xfs_buf_offset(bp, imap->im_boffset); |
4ae29b43 DC |
465 | return 0; |
466 | } | |
467 | ||
1da177e4 LT |
468 | /* |
469 | * Move inode type and inode format specific information from the | |
470 | * on-disk inode to the in-core inode. For fifos, devs, and sockets | |
471 | * this means set if_rdev to the proper value. For files, directories, | |
472 | * and symlinks this means to bring in the in-line data or extent | |
473 | * pointers. For a file in B-tree format, only the root is immediately | |
474 | * brought in-core. The rest will be in-lined in if_extents when it | |
475 | * is first referenced (see xfs_iread_extents()). | |
476 | */ | |
477 | STATIC int | |
478 | xfs_iformat( | |
479 | xfs_inode_t *ip, | |
480 | xfs_dinode_t *dip) | |
481 | { | |
482 | xfs_attr_shortform_t *atp; | |
483 | int size; | |
8096b1eb | 484 | int error = 0; |
1da177e4 | 485 | xfs_fsize_t di_size; |
1da177e4 | 486 | |
81591fe2 CH |
487 | if (unlikely(be32_to_cpu(dip->di_nextents) + |
488 | be16_to_cpu(dip->di_anextents) > | |
489 | be64_to_cpu(dip->di_nblocks))) { | |
65333b4c | 490 | xfs_warn(ip->i_mount, |
3762ec6b | 491 | "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", |
1da177e4 | 492 | (unsigned long long)ip->i_ino, |
81591fe2 CH |
493 | (int)(be32_to_cpu(dip->di_nextents) + |
494 | be16_to_cpu(dip->di_anextents)), | |
1da177e4 | 495 | (unsigned long long) |
81591fe2 | 496 | be64_to_cpu(dip->di_nblocks)); |
1da177e4 LT |
497 | XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, |
498 | ip->i_mount, dip); | |
499 | return XFS_ERROR(EFSCORRUPTED); | |
500 | } | |
501 | ||
81591fe2 | 502 | if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) { |
65333b4c | 503 | xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.", |
1da177e4 | 504 | (unsigned long long)ip->i_ino, |
81591fe2 | 505 | dip->di_forkoff); |
1da177e4 LT |
506 | XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, |
507 | ip->i_mount, dip); | |
508 | return XFS_ERROR(EFSCORRUPTED); | |
509 | } | |
510 | ||
b89d4208 CH |
511 | if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && |
512 | !ip->i_mount->m_rtdev_targp)) { | |
65333b4c | 513 | xfs_warn(ip->i_mount, |
b89d4208 CH |
514 | "corrupt dinode %Lu, has realtime flag set.", |
515 | ip->i_ino); | |
516 | XFS_CORRUPTION_ERROR("xfs_iformat(realtime)", | |
517 | XFS_ERRLEVEL_LOW, ip->i_mount, dip); | |
518 | return XFS_ERROR(EFSCORRUPTED); | |
519 | } | |
520 | ||
1da177e4 LT |
521 | switch (ip->i_d.di_mode & S_IFMT) { |
522 | case S_IFIFO: | |
523 | case S_IFCHR: | |
524 | case S_IFBLK: | |
525 | case S_IFSOCK: | |
81591fe2 | 526 | if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) { |
1da177e4 LT |
527 | XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, |
528 | ip->i_mount, dip); | |
529 | return XFS_ERROR(EFSCORRUPTED); | |
530 | } | |
531 | ip->i_d.di_size = 0; | |
81591fe2 | 532 | ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip); |
1da177e4 LT |
533 | break; |
534 | ||
535 | case S_IFREG: | |
536 | case S_IFLNK: | |
537 | case S_IFDIR: | |
81591fe2 | 538 | switch (dip->di_format) { |
1da177e4 LT |
539 | case XFS_DINODE_FMT_LOCAL: |
540 | /* | |
541 | * no local regular files yet | |
542 | */ | |
abbede1b | 543 | if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) { |
65333b4c DC |
544 | xfs_warn(ip->i_mount, |
545 | "corrupt inode %Lu (local format for regular file).", | |
1da177e4 LT |
546 | (unsigned long long) ip->i_ino); |
547 | XFS_CORRUPTION_ERROR("xfs_iformat(4)", | |
548 | XFS_ERRLEVEL_LOW, | |
549 | ip->i_mount, dip); | |
550 | return XFS_ERROR(EFSCORRUPTED); | |
551 | } | |
552 | ||
81591fe2 | 553 | di_size = be64_to_cpu(dip->di_size); |
1da177e4 | 554 | if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { |
65333b4c DC |
555 | xfs_warn(ip->i_mount, |
556 | "corrupt inode %Lu (bad size %Ld for local inode).", | |
1da177e4 LT |
557 | (unsigned long long) ip->i_ino, |
558 | (long long) di_size); | |
559 | XFS_CORRUPTION_ERROR("xfs_iformat(5)", | |
560 | XFS_ERRLEVEL_LOW, | |
561 | ip->i_mount, dip); | |
562 | return XFS_ERROR(EFSCORRUPTED); | |
563 | } | |
564 | ||
565 | size = (int)di_size; | |
566 | error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); | |
567 | break; | |
568 | case XFS_DINODE_FMT_EXTENTS: | |
569 | error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); | |
570 | break; | |
571 | case XFS_DINODE_FMT_BTREE: | |
572 | error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); | |
573 | break; | |
574 | default: | |
575 | XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, | |
576 | ip->i_mount); | |
577 | return XFS_ERROR(EFSCORRUPTED); | |
578 | } | |
579 | break; | |
580 | ||
581 | default: | |
582 | XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); | |
583 | return XFS_ERROR(EFSCORRUPTED); | |
584 | } | |
585 | if (error) { | |
586 | return error; | |
587 | } | |
588 | if (!XFS_DFORK_Q(dip)) | |
589 | return 0; | |
8096b1eb | 590 | |
1da177e4 | 591 | ASSERT(ip->i_afp == NULL); |
4a7edddc | 592 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS); |
8096b1eb | 593 | |
81591fe2 | 594 | switch (dip->di_aformat) { |
1da177e4 LT |
595 | case XFS_DINODE_FMT_LOCAL: |
596 | atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); | |
3b244aa8 | 597 | size = be16_to_cpu(atp->hdr.totsize); |
2809f76a CH |
598 | |
599 | if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) { | |
65333b4c DC |
600 | xfs_warn(ip->i_mount, |
601 | "corrupt inode %Lu (bad attr fork size %Ld).", | |
2809f76a CH |
602 | (unsigned long long) ip->i_ino, |
603 | (long long) size); | |
604 | XFS_CORRUPTION_ERROR("xfs_iformat(8)", | |
605 | XFS_ERRLEVEL_LOW, | |
606 | ip->i_mount, dip); | |
607 | return XFS_ERROR(EFSCORRUPTED); | |
608 | } | |
609 | ||
1da177e4 LT |
610 | error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); |
611 | break; | |
612 | case XFS_DINODE_FMT_EXTENTS: | |
613 | error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); | |
614 | break; | |
615 | case XFS_DINODE_FMT_BTREE: | |
616 | error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); | |
617 | break; | |
618 | default: | |
619 | error = XFS_ERROR(EFSCORRUPTED); | |
620 | break; | |
621 | } | |
622 | if (error) { | |
623 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
624 | ip->i_afp = NULL; | |
625 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
626 | } | |
627 | return error; | |
628 | } | |
629 | ||
630 | /* | |
631 | * The file is in-lined in the on-disk inode. | |
632 | * If it fits into if_inline_data, then copy | |
633 | * it there, otherwise allocate a buffer for it | |
634 | * and copy the data there. Either way, set | |
635 | * if_data to point at the data. | |
636 | * If we allocate a buffer for the data, make | |
637 | * sure that its size is a multiple of 4 and | |
638 | * record the real size in i_real_bytes. | |
639 | */ | |
640 | STATIC int | |
641 | xfs_iformat_local( | |
642 | xfs_inode_t *ip, | |
643 | xfs_dinode_t *dip, | |
644 | int whichfork, | |
645 | int size) | |
646 | { | |
647 | xfs_ifork_t *ifp; | |
648 | int real_size; | |
649 | ||
650 | /* | |
651 | * If the size is unreasonable, then something | |
652 | * is wrong and we just bail out rather than crash in | |
653 | * kmem_alloc() or memcpy() below. | |
654 | */ | |
655 | if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
65333b4c DC |
656 | xfs_warn(ip->i_mount, |
657 | "corrupt inode %Lu (bad size %d for local fork, size = %d).", | |
1da177e4 LT |
658 | (unsigned long long) ip->i_ino, size, |
659 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); | |
660 | XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, | |
661 | ip->i_mount, dip); | |
662 | return XFS_ERROR(EFSCORRUPTED); | |
663 | } | |
664 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
665 | real_size = 0; | |
666 | if (size == 0) | |
667 | ifp->if_u1.if_data = NULL; | |
668 | else if (size <= sizeof(ifp->if_u2.if_inline_data)) | |
669 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
670 | else { | |
671 | real_size = roundup(size, 4); | |
4a7edddc | 672 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS); |
1da177e4 LT |
673 | } |
674 | ifp->if_bytes = size; | |
675 | ifp->if_real_bytes = real_size; | |
676 | if (size) | |
677 | memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); | |
678 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
679 | ifp->if_flags |= XFS_IFINLINE; | |
680 | return 0; | |
681 | } | |
682 | ||
683 | /* | |
684 | * The file consists of a set of extents all | |
685 | * of which fit into the on-disk inode. | |
686 | * If there are few enough extents to fit into | |
687 | * the if_inline_ext, then copy them there. | |
688 | * Otherwise allocate a buffer for them and copy | |
689 | * them into it. Either way, set if_extents | |
690 | * to point at the extents. | |
691 | */ | |
692 | STATIC int | |
693 | xfs_iformat_extents( | |
694 | xfs_inode_t *ip, | |
695 | xfs_dinode_t *dip, | |
696 | int whichfork) | |
697 | { | |
a6f64d4a | 698 | xfs_bmbt_rec_t *dp; |
1da177e4 LT |
699 | xfs_ifork_t *ifp; |
700 | int nex; | |
1da177e4 LT |
701 | int size; |
702 | int i; | |
703 | ||
704 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
705 | nex = XFS_DFORK_NEXTENTS(dip, whichfork); | |
706 | size = nex * (uint)sizeof(xfs_bmbt_rec_t); | |
707 | ||
708 | /* | |
709 | * If the number of extents is unreasonable, then something | |
710 | * is wrong and we just bail out rather than crash in | |
711 | * kmem_alloc() or memcpy() below. | |
712 | */ | |
713 | if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
65333b4c | 714 | xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).", |
1da177e4 LT |
715 | (unsigned long long) ip->i_ino, nex); |
716 | XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, | |
717 | ip->i_mount, dip); | |
718 | return XFS_ERROR(EFSCORRUPTED); | |
719 | } | |
720 | ||
4eea22f0 | 721 | ifp->if_real_bytes = 0; |
1da177e4 LT |
722 | if (nex == 0) |
723 | ifp->if_u1.if_extents = NULL; | |
724 | else if (nex <= XFS_INLINE_EXTS) | |
725 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
4eea22f0 MK |
726 | else |
727 | xfs_iext_add(ifp, 0, nex); | |
728 | ||
1da177e4 | 729 | ifp->if_bytes = size; |
1da177e4 LT |
730 | if (size) { |
731 | dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); | |
a6f64d4a | 732 | xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip)); |
4eea22f0 | 733 | for (i = 0; i < nex; i++, dp++) { |
a6f64d4a | 734 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
597bca63 HH |
735 | ep->l0 = get_unaligned_be64(&dp->l0); |
736 | ep->l1 = get_unaligned_be64(&dp->l1); | |
1da177e4 | 737 | } |
3a59c94c | 738 | XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork); |
1da177e4 LT |
739 | if (whichfork != XFS_DATA_FORK || |
740 | XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) | |
741 | if (unlikely(xfs_check_nostate_extents( | |
4eea22f0 | 742 | ifp, 0, nex))) { |
1da177e4 LT |
743 | XFS_ERROR_REPORT("xfs_iformat_extents(2)", |
744 | XFS_ERRLEVEL_LOW, | |
745 | ip->i_mount); | |
746 | return XFS_ERROR(EFSCORRUPTED); | |
747 | } | |
748 | } | |
749 | ifp->if_flags |= XFS_IFEXTENTS; | |
750 | return 0; | |
751 | } | |
752 | ||
753 | /* | |
754 | * The file has too many extents to fit into | |
755 | * the inode, so they are in B-tree format. | |
756 | * Allocate a buffer for the root of the B-tree | |
757 | * and copy the root into it. The i_extents | |
758 | * field will remain NULL until all of the | |
759 | * extents are read in (when they are needed). | |
760 | */ | |
761 | STATIC int | |
762 | xfs_iformat_btree( | |
763 | xfs_inode_t *ip, | |
764 | xfs_dinode_t *dip, | |
765 | int whichfork) | |
766 | { | |
767 | xfs_bmdr_block_t *dfp; | |
768 | xfs_ifork_t *ifp; | |
769 | /* REFERENCED */ | |
770 | int nrecs; | |
771 | int size; | |
772 | ||
773 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
774 | dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); | |
775 | size = XFS_BMAP_BROOT_SPACE(dfp); | |
60197e8d | 776 | nrecs = be16_to_cpu(dfp->bb_numrecs); |
1da177e4 LT |
777 | |
778 | /* | |
779 | * blow out if -- fork has less extents than can fit in | |
780 | * fork (fork shouldn't be a btree format), root btree | |
781 | * block has more records than can fit into the fork, | |
782 | * or the number of extents is greater than the number of | |
783 | * blocks. | |
784 | */ | |
8096b1eb CH |
785 | if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= |
786 | XFS_IFORK_MAXEXT(ip, whichfork) || | |
787 | XFS_BMDR_SPACE_CALC(nrecs) > | |
788 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) || | |
789 | XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { | |
65333b4c | 790 | xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).", |
1da177e4 | 791 | (unsigned long long) ip->i_ino); |
65333b4c DC |
792 | XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW, |
793 | ip->i_mount, dip); | |
1da177e4 LT |
794 | return XFS_ERROR(EFSCORRUPTED); |
795 | } | |
796 | ||
797 | ifp->if_broot_bytes = size; | |
4a7edddc | 798 | ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS); |
1da177e4 LT |
799 | ASSERT(ifp->if_broot != NULL); |
800 | /* | |
801 | * Copy and convert from the on-disk structure | |
802 | * to the in-memory structure. | |
803 | */ | |
60197e8d CH |
804 | xfs_bmdr_to_bmbt(ip->i_mount, dfp, |
805 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), | |
806 | ifp->if_broot, size); | |
1da177e4 LT |
807 | ifp->if_flags &= ~XFS_IFEXTENTS; |
808 | ifp->if_flags |= XFS_IFBROOT; | |
809 | ||
810 | return 0; | |
811 | } | |
812 | ||
d96f8f89 | 813 | STATIC void |
347d1c01 CH |
814 | xfs_dinode_from_disk( |
815 | xfs_icdinode_t *to, | |
81591fe2 | 816 | xfs_dinode_t *from) |
1da177e4 | 817 | { |
347d1c01 CH |
818 | to->di_magic = be16_to_cpu(from->di_magic); |
819 | to->di_mode = be16_to_cpu(from->di_mode); | |
820 | to->di_version = from ->di_version; | |
821 | to->di_format = from->di_format; | |
822 | to->di_onlink = be16_to_cpu(from->di_onlink); | |
823 | to->di_uid = be32_to_cpu(from->di_uid); | |
824 | to->di_gid = be32_to_cpu(from->di_gid); | |
825 | to->di_nlink = be32_to_cpu(from->di_nlink); | |
6743099c AM |
826 | to->di_projid_lo = be16_to_cpu(from->di_projid_lo); |
827 | to->di_projid_hi = be16_to_cpu(from->di_projid_hi); | |
347d1c01 CH |
828 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); |
829 | to->di_flushiter = be16_to_cpu(from->di_flushiter); | |
830 | to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec); | |
831 | to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec); | |
832 | to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec); | |
833 | to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec); | |
834 | to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec); | |
835 | to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec); | |
836 | to->di_size = be64_to_cpu(from->di_size); | |
837 | to->di_nblocks = be64_to_cpu(from->di_nblocks); | |
838 | to->di_extsize = be32_to_cpu(from->di_extsize); | |
839 | to->di_nextents = be32_to_cpu(from->di_nextents); | |
840 | to->di_anextents = be16_to_cpu(from->di_anextents); | |
841 | to->di_forkoff = from->di_forkoff; | |
842 | to->di_aformat = from->di_aformat; | |
843 | to->di_dmevmask = be32_to_cpu(from->di_dmevmask); | |
844 | to->di_dmstate = be16_to_cpu(from->di_dmstate); | |
845 | to->di_flags = be16_to_cpu(from->di_flags); | |
846 | to->di_gen = be32_to_cpu(from->di_gen); | |
847 | } | |
848 | ||
849 | void | |
850 | xfs_dinode_to_disk( | |
81591fe2 | 851 | xfs_dinode_t *to, |
347d1c01 CH |
852 | xfs_icdinode_t *from) |
853 | { | |
854 | to->di_magic = cpu_to_be16(from->di_magic); | |
855 | to->di_mode = cpu_to_be16(from->di_mode); | |
856 | to->di_version = from ->di_version; | |
857 | to->di_format = from->di_format; | |
858 | to->di_onlink = cpu_to_be16(from->di_onlink); | |
859 | to->di_uid = cpu_to_be32(from->di_uid); | |
860 | to->di_gid = cpu_to_be32(from->di_gid); | |
861 | to->di_nlink = cpu_to_be32(from->di_nlink); | |
6743099c AM |
862 | to->di_projid_lo = cpu_to_be16(from->di_projid_lo); |
863 | to->di_projid_hi = cpu_to_be16(from->di_projid_hi); | |
347d1c01 CH |
864 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); |
865 | to->di_flushiter = cpu_to_be16(from->di_flushiter); | |
866 | to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec); | |
867 | to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec); | |
868 | to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec); | |
869 | to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec); | |
870 | to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec); | |
871 | to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec); | |
872 | to->di_size = cpu_to_be64(from->di_size); | |
873 | to->di_nblocks = cpu_to_be64(from->di_nblocks); | |
874 | to->di_extsize = cpu_to_be32(from->di_extsize); | |
875 | to->di_nextents = cpu_to_be32(from->di_nextents); | |
876 | to->di_anextents = cpu_to_be16(from->di_anextents); | |
877 | to->di_forkoff = from->di_forkoff; | |
878 | to->di_aformat = from->di_aformat; | |
879 | to->di_dmevmask = cpu_to_be32(from->di_dmevmask); | |
880 | to->di_dmstate = cpu_to_be16(from->di_dmstate); | |
881 | to->di_flags = cpu_to_be16(from->di_flags); | |
882 | to->di_gen = cpu_to_be32(from->di_gen); | |
1da177e4 LT |
883 | } |
884 | ||
885 | STATIC uint | |
886 | _xfs_dic2xflags( | |
1da177e4 LT |
887 | __uint16_t di_flags) |
888 | { | |
889 | uint flags = 0; | |
890 | ||
891 | if (di_flags & XFS_DIFLAG_ANY) { | |
892 | if (di_flags & XFS_DIFLAG_REALTIME) | |
893 | flags |= XFS_XFLAG_REALTIME; | |
894 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
895 | flags |= XFS_XFLAG_PREALLOC; | |
896 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
897 | flags |= XFS_XFLAG_IMMUTABLE; | |
898 | if (di_flags & XFS_DIFLAG_APPEND) | |
899 | flags |= XFS_XFLAG_APPEND; | |
900 | if (di_flags & XFS_DIFLAG_SYNC) | |
901 | flags |= XFS_XFLAG_SYNC; | |
902 | if (di_flags & XFS_DIFLAG_NOATIME) | |
903 | flags |= XFS_XFLAG_NOATIME; | |
904 | if (di_flags & XFS_DIFLAG_NODUMP) | |
905 | flags |= XFS_XFLAG_NODUMP; | |
906 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
907 | flags |= XFS_XFLAG_RTINHERIT; | |
908 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
909 | flags |= XFS_XFLAG_PROJINHERIT; | |
910 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
911 | flags |= XFS_XFLAG_NOSYMLINKS; | |
dd9f438e NS |
912 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
913 | flags |= XFS_XFLAG_EXTSIZE; | |
914 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) | |
915 | flags |= XFS_XFLAG_EXTSZINHERIT; | |
d3446eac BN |
916 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
917 | flags |= XFS_XFLAG_NODEFRAG; | |
2a82b8be DC |
918 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
919 | flags |= XFS_XFLAG_FILESTREAM; | |
1da177e4 LT |
920 | } |
921 | ||
922 | return flags; | |
923 | } | |
924 | ||
925 | uint | |
926 | xfs_ip2xflags( | |
927 | xfs_inode_t *ip) | |
928 | { | |
347d1c01 | 929 | xfs_icdinode_t *dic = &ip->i_d; |
1da177e4 | 930 | |
a916e2bd | 931 | return _xfs_dic2xflags(dic->di_flags) | |
45ba598e | 932 | (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
933 | } |
934 | ||
935 | uint | |
936 | xfs_dic2xflags( | |
45ba598e | 937 | xfs_dinode_t *dip) |
1da177e4 | 938 | { |
81591fe2 | 939 | return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | |
45ba598e | 940 | (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
941 | } |
942 | ||
07c8f675 | 943 | /* |
24f211ba | 944 | * Read the disk inode attributes into the in-core inode structure. |
1da177e4 LT |
945 | */ |
946 | int | |
947 | xfs_iread( | |
948 | xfs_mount_t *mp, | |
949 | xfs_trans_t *tp, | |
24f211ba | 950 | xfs_inode_t *ip, |
24f211ba | 951 | uint iget_flags) |
1da177e4 LT |
952 | { |
953 | xfs_buf_t *bp; | |
954 | xfs_dinode_t *dip; | |
1da177e4 LT |
955 | int error; |
956 | ||
1da177e4 | 957 | /* |
92bfc6e7 | 958 | * Fill in the location information in the in-core inode. |
1da177e4 | 959 | */ |
24f211ba | 960 | error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags); |
76d8b277 | 961 | if (error) |
24f211ba | 962 | return error; |
76d8b277 CH |
963 | |
964 | /* | |
92bfc6e7 | 965 | * Get pointers to the on-disk inode and the buffer containing it. |
76d8b277 | 966 | */ |
475ee413 | 967 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0, iget_flags); |
9ed0451e | 968 | if (error) |
24f211ba | 969 | return error; |
1da177e4 | 970 | |
1da177e4 LT |
971 | /* |
972 | * If we got something that isn't an inode it means someone | |
973 | * (nfs or dmi) has a stale handle. | |
974 | */ | |
69ef921b | 975 | if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) { |
1da177e4 | 976 | #ifdef DEBUG |
53487786 DC |
977 | xfs_alert(mp, |
978 | "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)", | |
979 | __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC); | |
1da177e4 | 980 | #endif /* DEBUG */ |
9ed0451e CH |
981 | error = XFS_ERROR(EINVAL); |
982 | goto out_brelse; | |
1da177e4 LT |
983 | } |
984 | ||
985 | /* | |
986 | * If the on-disk inode is already linked to a directory | |
987 | * entry, copy all of the inode into the in-core inode. | |
988 | * xfs_iformat() handles copying in the inode format | |
989 | * specific information. | |
990 | * Otherwise, just get the truly permanent information. | |
991 | */ | |
81591fe2 CH |
992 | if (dip->di_mode) { |
993 | xfs_dinode_from_disk(&ip->i_d, dip); | |
1da177e4 LT |
994 | error = xfs_iformat(ip, dip); |
995 | if (error) { | |
1da177e4 | 996 | #ifdef DEBUG |
53487786 DC |
997 | xfs_alert(mp, "%s: xfs_iformat() returned error %d", |
998 | __func__, error); | |
1da177e4 | 999 | #endif /* DEBUG */ |
9ed0451e | 1000 | goto out_brelse; |
1da177e4 LT |
1001 | } |
1002 | } else { | |
81591fe2 CH |
1003 | ip->i_d.di_magic = be16_to_cpu(dip->di_magic); |
1004 | ip->i_d.di_version = dip->di_version; | |
1005 | ip->i_d.di_gen = be32_to_cpu(dip->di_gen); | |
1006 | ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter); | |
1da177e4 LT |
1007 | /* |
1008 | * Make sure to pull in the mode here as well in | |
1009 | * case the inode is released without being used. | |
1010 | * This ensures that xfs_inactive() will see that | |
1011 | * the inode is already free and not try to mess | |
1012 | * with the uninitialized part of it. | |
1013 | */ | |
1014 | ip->i_d.di_mode = 0; | |
1da177e4 LT |
1015 | } |
1016 | ||
1da177e4 LT |
1017 | /* |
1018 | * The inode format changed when we moved the link count and | |
1019 | * made it 32 bits long. If this is an old format inode, | |
1020 | * convert it in memory to look like a new one. If it gets | |
1021 | * flushed to disk we will convert back before flushing or | |
1022 | * logging it. We zero out the new projid field and the old link | |
1023 | * count field. We'll handle clearing the pad field (the remains | |
1024 | * of the old uuid field) when we actually convert the inode to | |
1025 | * the new format. We don't change the version number so that we | |
1026 | * can distinguish this from a real new format inode. | |
1027 | */ | |
51ce16d5 | 1028 | if (ip->i_d.di_version == 1) { |
1da177e4 LT |
1029 | ip->i_d.di_nlink = ip->i_d.di_onlink; |
1030 | ip->i_d.di_onlink = 0; | |
6743099c | 1031 | xfs_set_projid(ip, 0); |
1da177e4 LT |
1032 | } |
1033 | ||
1034 | ip->i_delayed_blks = 0; | |
1035 | ||
1036 | /* | |
1037 | * Mark the buffer containing the inode as something to keep | |
1038 | * around for a while. This helps to keep recently accessed | |
1039 | * meta-data in-core longer. | |
1040 | */ | |
821eb21d | 1041 | xfs_buf_set_ref(bp, XFS_INO_REF); |
1da177e4 LT |
1042 | |
1043 | /* | |
1044 | * Use xfs_trans_brelse() to release the buffer containing the | |
1045 | * on-disk inode, because it was acquired with xfs_trans_read_buf() | |
475ee413 | 1046 | * in xfs_imap_to_bp() above. If tp is NULL, this is just a normal |
1da177e4 LT |
1047 | * brelse(). If we're within a transaction, then xfs_trans_brelse() |
1048 | * will only release the buffer if it is not dirty within the | |
1049 | * transaction. It will be OK to release the buffer in this case, | |
1050 | * because inodes on disk are never destroyed and we will be | |
1051 | * locking the new in-core inode before putting it in the hash | |
1052 | * table where other processes can find it. Thus we don't have | |
1053 | * to worry about the inode being changed just because we released | |
1054 | * the buffer. | |
1055 | */ | |
9ed0451e CH |
1056 | out_brelse: |
1057 | xfs_trans_brelse(tp, bp); | |
9ed0451e | 1058 | return error; |
1da177e4 LT |
1059 | } |
1060 | ||
1061 | /* | |
1062 | * Read in extents from a btree-format inode. | |
1063 | * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. | |
1064 | */ | |
1065 | int | |
1066 | xfs_iread_extents( | |
1067 | xfs_trans_t *tp, | |
1068 | xfs_inode_t *ip, | |
1069 | int whichfork) | |
1070 | { | |
1071 | int error; | |
1072 | xfs_ifork_t *ifp; | |
4eea22f0 | 1073 | xfs_extnum_t nextents; |
1da177e4 LT |
1074 | |
1075 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1076 | XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, | |
1077 | ip->i_mount); | |
1078 | return XFS_ERROR(EFSCORRUPTED); | |
1079 | } | |
4eea22f0 | 1080 | nextents = XFS_IFORK_NEXTENTS(ip, whichfork); |
1da177e4 | 1081 | ifp = XFS_IFORK_PTR(ip, whichfork); |
4eea22f0 | 1082 | |
1da177e4 LT |
1083 | /* |
1084 | * We know that the size is valid (it's checked in iformat_btree) | |
1085 | */ | |
4eea22f0 | 1086 | ifp->if_bytes = ifp->if_real_bytes = 0; |
1da177e4 | 1087 | ifp->if_flags |= XFS_IFEXTENTS; |
4eea22f0 | 1088 | xfs_iext_add(ifp, 0, nextents); |
1da177e4 LT |
1089 | error = xfs_bmap_read_extents(tp, ip, whichfork); |
1090 | if (error) { | |
4eea22f0 | 1091 | xfs_iext_destroy(ifp); |
1da177e4 LT |
1092 | ifp->if_flags &= ~XFS_IFEXTENTS; |
1093 | return error; | |
1094 | } | |
a6f64d4a | 1095 | xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
1096 | return 0; |
1097 | } | |
1098 | ||
1099 | /* | |
1100 | * Allocate an inode on disk and return a copy of its in-core version. | |
1101 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
1102 | * appropriately within the inode. The uid and gid for the inode are | |
1103 | * set according to the contents of the given cred structure. | |
1104 | * | |
1105 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
1106 | * has a free inode available, call xfs_iget() | |
1107 | * to obtain the in-core version of the allocated inode. Finally, | |
1108 | * fill in the inode and log its initial contents. In this case, | |
1109 | * ialloc_context would be set to NULL and call_again set to false. | |
1110 | * | |
1111 | * If xfs_dialloc() does not have an available inode, | |
1112 | * it will replenish its supply by doing an allocation. Since we can | |
1113 | * only do one allocation within a transaction without deadlocks, we | |
1114 | * must commit the current transaction before returning the inode itself. | |
1115 | * In this case, therefore, we will set call_again to true and return. | |
1116 | * The caller should then commit the current transaction, start a new | |
1117 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
1118 | * | |
1119 | * To ensure that some other process does not grab the inode that | |
1120 | * was allocated during the first call to xfs_ialloc(), this routine | |
1121 | * also returns the [locked] bp pointing to the head of the freelist | |
1122 | * as ialloc_context. The caller should hold this buffer across | |
1123 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
1124 | * |
1125 | * If we are allocating quota inodes, we do not have a parent inode | |
1126 | * to attach to or associate with (i.e. pip == NULL) because they | |
1127 | * are not linked into the directory structure - they are attached | |
1128 | * directly to the superblock - and so have no parent. | |
1da177e4 LT |
1129 | */ |
1130 | int | |
1131 | xfs_ialloc( | |
1132 | xfs_trans_t *tp, | |
1133 | xfs_inode_t *pip, | |
576b1d67 | 1134 | umode_t mode, |
31b084ae | 1135 | xfs_nlink_t nlink, |
1da177e4 | 1136 | xfs_dev_t rdev, |
6743099c | 1137 | prid_t prid, |
1da177e4 LT |
1138 | int okalloc, |
1139 | xfs_buf_t **ialloc_context, | |
1da177e4 LT |
1140 | xfs_inode_t **ipp) |
1141 | { | |
1142 | xfs_ino_t ino; | |
1143 | xfs_inode_t *ip; | |
1da177e4 LT |
1144 | uint flags; |
1145 | int error; | |
dff35fd4 | 1146 | timespec_t tv; |
bf904248 | 1147 | int filestreams = 0; |
1da177e4 LT |
1148 | |
1149 | /* | |
1150 | * Call the space management code to pick | |
1151 | * the on-disk inode to be allocated. | |
1152 | */ | |
b11f94d5 | 1153 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
08358906 | 1154 | ialloc_context, &ino); |
bf904248 | 1155 | if (error) |
1da177e4 | 1156 | return error; |
08358906 | 1157 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
1158 | *ipp = NULL; |
1159 | return 0; | |
1160 | } | |
1161 | ASSERT(*ialloc_context == NULL); | |
1162 | ||
1163 | /* | |
1164 | * Get the in-core inode with the lock held exclusively. | |
1165 | * This is because we're setting fields here we need | |
1166 | * to prevent others from looking at until we're done. | |
1167 | */ | |
ec3ba85f CH |
1168 | error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE, |
1169 | XFS_ILOCK_EXCL, &ip); | |
bf904248 | 1170 | if (error) |
1da177e4 | 1171 | return error; |
1da177e4 LT |
1172 | ASSERT(ip != NULL); |
1173 | ||
576b1d67 | 1174 | ip->i_d.di_mode = mode; |
1da177e4 LT |
1175 | ip->i_d.di_onlink = 0; |
1176 | ip->i_d.di_nlink = nlink; | |
1177 | ASSERT(ip->i_d.di_nlink == nlink); | |
9e2b2dc4 DH |
1178 | ip->i_d.di_uid = current_fsuid(); |
1179 | ip->i_d.di_gid = current_fsgid(); | |
6743099c | 1180 | xfs_set_projid(ip, prid); |
1da177e4 LT |
1181 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
1182 | ||
1183 | /* | |
1184 | * If the superblock version is up to where we support new format | |
1185 | * inodes and this is currently an old format inode, then change | |
1186 | * the inode version number now. This way we only do the conversion | |
1187 | * here rather than here and in the flush/logging code. | |
1188 | */ | |
62118709 | 1189 | if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) && |
51ce16d5 CH |
1190 | ip->i_d.di_version == 1) { |
1191 | ip->i_d.di_version = 2; | |
1da177e4 LT |
1192 | /* |
1193 | * We've already zeroed the old link count, the projid field, | |
1194 | * and the pad field. | |
1195 | */ | |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
1200 | */ | |
51ce16d5 | 1201 | if ((prid != 0) && (ip->i_d.di_version == 1)) |
1da177e4 LT |
1202 | xfs_bump_ino_vers2(tp, ip); |
1203 | ||
bd186aa9 | 1204 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 1205 | ip->i_d.di_gid = pip->i_d.di_gid; |
abbede1b | 1206 | if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { |
1da177e4 LT |
1207 | ip->i_d.di_mode |= S_ISGID; |
1208 | } | |
1209 | } | |
1210 | ||
1211 | /* | |
1212 | * If the group ID of the new file does not match the effective group | |
1213 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
1214 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
1215 | */ | |
1216 | if ((irix_sgid_inherit) && | |
1217 | (ip->i_d.di_mode & S_ISGID) && | |
1218 | (!in_group_p((gid_t)ip->i_d.di_gid))) { | |
1219 | ip->i_d.di_mode &= ~S_ISGID; | |
1220 | } | |
1221 | ||
1222 | ip->i_d.di_size = 0; | |
1223 | ip->i_d.di_nextents = 0; | |
1224 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 CH |
1225 | |
1226 | nanotime(&tv); | |
1227 | ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; | |
1228 | ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; | |
1229 | ip->i_d.di_atime = ip->i_d.di_mtime; | |
1230 | ip->i_d.di_ctime = ip->i_d.di_mtime; | |
1231 | ||
1da177e4 LT |
1232 | /* |
1233 | * di_gen will have been taken care of in xfs_iread. | |
1234 | */ | |
1235 | ip->i_d.di_extsize = 0; | |
1236 | ip->i_d.di_dmevmask = 0; | |
1237 | ip->i_d.di_dmstate = 0; | |
1238 | ip->i_d.di_flags = 0; | |
1239 | flags = XFS_ILOG_CORE; | |
1240 | switch (mode & S_IFMT) { | |
1241 | case S_IFIFO: | |
1242 | case S_IFCHR: | |
1243 | case S_IFBLK: | |
1244 | case S_IFSOCK: | |
1245 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1246 | ip->i_df.if_u2.if_rdev = rdev; | |
1247 | ip->i_df.if_flags = 0; | |
1248 | flags |= XFS_ILOG_DEV; | |
1249 | break; | |
1250 | case S_IFREG: | |
bf904248 DC |
1251 | /* |
1252 | * we can't set up filestreams until after the VFS inode | |
1253 | * is set up properly. | |
1254 | */ | |
1255 | if (pip && xfs_inode_is_filestream(pip)) | |
1256 | filestreams = 1; | |
2a82b8be | 1257 | /* fall through */ |
1da177e4 | 1258 | case S_IFDIR: |
b11f94d5 | 1259 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
365ca83d NS |
1260 | uint di_flags = 0; |
1261 | ||
abbede1b | 1262 | if (S_ISDIR(mode)) { |
365ca83d NS |
1263 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
1264 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
1265 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1266 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
1267 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1268 | } | |
abbede1b | 1269 | } else if (S_ISREG(mode)) { |
613d7043 | 1270 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 1271 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
1272 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1273 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1274 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1275 | } | |
1da177e4 LT |
1276 | } |
1277 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
1278 | xfs_inherit_noatime) | |
365ca83d | 1279 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
1280 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
1281 | xfs_inherit_nodump) | |
365ca83d | 1282 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
1283 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
1284 | xfs_inherit_sync) | |
365ca83d | 1285 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
1286 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
1287 | xfs_inherit_nosymlinks) | |
365ca83d NS |
1288 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
1289 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
1290 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
1291 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
1292 | xfs_inherit_nodefrag) | |
1293 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
1294 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
1295 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
365ca83d | 1296 | ip->i_d.di_flags |= di_flags; |
1da177e4 LT |
1297 | } |
1298 | /* FALLTHROUGH */ | |
1299 | case S_IFLNK: | |
1300 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
1301 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
1302 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
1303 | ip->i_df.if_u1.if_extents = NULL; | |
1304 | break; | |
1305 | default: | |
1306 | ASSERT(0); | |
1307 | } | |
1308 | /* | |
1309 | * Attribute fork settings for new inode. | |
1310 | */ | |
1311 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1312 | ip->i_d.di_anextents = 0; | |
1313 | ||
1314 | /* | |
1315 | * Log the new values stuffed into the inode. | |
1316 | */ | |
ddc3415a | 1317 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1318 | xfs_trans_log_inode(tp, ip, flags); |
1319 | ||
b83bd138 | 1320 | /* now that we have an i_mode we can setup inode ops and unlock */ |
41be8bed | 1321 | xfs_setup_inode(ip); |
1da177e4 | 1322 | |
bf904248 DC |
1323 | /* now we have set up the vfs inode we can associate the filestream */ |
1324 | if (filestreams) { | |
1325 | error = xfs_filestream_associate(pip, ip); | |
1326 | if (error < 0) | |
1327 | return -error; | |
1328 | if (!error) | |
1329 | xfs_iflags_set(ip, XFS_IFILESTREAM); | |
1330 | } | |
1331 | ||
1da177e4 LT |
1332 | *ipp = ip; |
1333 | return 0; | |
1334 | } | |
1335 | ||
1da177e4 | 1336 | /* |
8f04c47a CH |
1337 | * Free up the underlying blocks past new_size. The new size must be smaller |
1338 | * than the current size. This routine can be used both for the attribute and | |
1339 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1340 | * |
f6485057 DC |
1341 | * The transaction passed to this routine must have made a permanent log |
1342 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1343 | * given transaction and start new ones, so make sure everything involved in | |
1344 | * the transaction is tidy before calling here. Some transaction will be | |
1345 | * returned to the caller to be committed. The incoming transaction must | |
1346 | * already include the inode, and both inode locks must be held exclusively. | |
1347 | * The inode must also be "held" within the transaction. On return the inode | |
1348 | * will be "held" within the returned transaction. This routine does NOT | |
1349 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1350 | * |
f6485057 DC |
1351 | * If we get an error, we must return with the inode locked and linked into the |
1352 | * current transaction. This keeps things simple for the higher level code, | |
1353 | * because it always knows that the inode is locked and held in the transaction | |
1354 | * that returns to it whether errors occur or not. We don't mark the inode | |
1355 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1356 | */ |
1357 | int | |
8f04c47a CH |
1358 | xfs_itruncate_extents( |
1359 | struct xfs_trans **tpp, | |
1360 | struct xfs_inode *ip, | |
1361 | int whichfork, | |
1362 | xfs_fsize_t new_size) | |
1da177e4 | 1363 | { |
8f04c47a CH |
1364 | struct xfs_mount *mp = ip->i_mount; |
1365 | struct xfs_trans *tp = *tpp; | |
1366 | struct xfs_trans *ntp; | |
1367 | xfs_bmap_free_t free_list; | |
1368 | xfs_fsblock_t first_block; | |
1369 | xfs_fileoff_t first_unmap_block; | |
1370 | xfs_fileoff_t last_block; | |
1371 | xfs_filblks_t unmap_len; | |
1372 | int committed; | |
1373 | int error = 0; | |
1374 | int done = 0; | |
1da177e4 | 1375 | |
0b56185b CH |
1376 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1377 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1378 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1379 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1380 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1381 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1382 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1383 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1384 | |
673e8e59 CH |
1385 | trace_xfs_itruncate_extents_start(ip, new_size); |
1386 | ||
1da177e4 LT |
1387 | /* |
1388 | * Since it is possible for space to become allocated beyond | |
1389 | * the end of the file (in a crash where the space is allocated | |
1390 | * but the inode size is not yet updated), simply remove any | |
1391 | * blocks which show up between the new EOF and the maximum | |
1392 | * possible file size. If the first block to be removed is | |
1393 | * beyond the maximum file size (ie it is the same as last_block), | |
1394 | * then there is nothing to do. | |
1395 | */ | |
8f04c47a | 1396 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1397 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1398 | if (first_unmap_block == last_block) |
1399 | return 0; | |
1400 | ||
1401 | ASSERT(first_unmap_block < last_block); | |
1402 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1403 | while (!done) { |
9d87c319 | 1404 | xfs_bmap_init(&free_list, &first_block); |
8f04c47a | 1405 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 1406 | first_unmap_block, unmap_len, |
8f04c47a | 1407 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 1408 | XFS_ITRUNC_MAX_EXTENTS, |
3e57ecf6 | 1409 | &first_block, &free_list, |
b4e9181e | 1410 | &done); |
8f04c47a CH |
1411 | if (error) |
1412 | goto out_bmap_cancel; | |
1da177e4 LT |
1413 | |
1414 | /* | |
1415 | * Duplicate the transaction that has the permanent | |
1416 | * reservation and commit the old transaction. | |
1417 | */ | |
8f04c47a | 1418 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
898621d5 | 1419 | if (committed) |
ddc3415a | 1420 | xfs_trans_ijoin(tp, ip, 0); |
8f04c47a CH |
1421 | if (error) |
1422 | goto out_bmap_cancel; | |
1da177e4 LT |
1423 | |
1424 | if (committed) { | |
1425 | /* | |
f6485057 | 1426 | * Mark the inode dirty so it will be logged and |
e5720eec | 1427 | * moved forward in the log as part of every commit. |
1da177e4 | 1428 | */ |
8f04c47a | 1429 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1da177e4 | 1430 | } |
f6485057 | 1431 | |
8f04c47a CH |
1432 | ntp = xfs_trans_dup(tp); |
1433 | error = xfs_trans_commit(tp, 0); | |
1434 | tp = ntp; | |
e5720eec | 1435 | |
ddc3415a | 1436 | xfs_trans_ijoin(tp, ip, 0); |
f6485057 | 1437 | |
cc09c0dc | 1438 | if (error) |
8f04c47a CH |
1439 | goto out; |
1440 | ||
cc09c0dc | 1441 | /* |
8f04c47a | 1442 | * Transaction commit worked ok so we can drop the extra ticket |
cc09c0dc DC |
1443 | * reference that we gained in xfs_trans_dup() |
1444 | */ | |
8f04c47a CH |
1445 | xfs_log_ticket_put(tp->t_ticket); |
1446 | error = xfs_trans_reserve(tp, 0, | |
f6485057 DC |
1447 | XFS_ITRUNCATE_LOG_RES(mp), 0, |
1448 | XFS_TRANS_PERM_LOG_RES, | |
1449 | XFS_ITRUNCATE_LOG_COUNT); | |
1450 | if (error) | |
8f04c47a | 1451 | goto out; |
1da177e4 | 1452 | } |
8f04c47a | 1453 | |
673e8e59 CH |
1454 | /* |
1455 | * Always re-log the inode so that our permanent transaction can keep | |
1456 | * on rolling it forward in the log. | |
1457 | */ | |
1458 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1459 | ||
1460 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1461 | ||
8f04c47a CH |
1462 | out: |
1463 | *tpp = tp; | |
1464 | return error; | |
1465 | out_bmap_cancel: | |
1da177e4 | 1466 | /* |
8f04c47a CH |
1467 | * If the bunmapi call encounters an error, return to the caller where |
1468 | * the transaction can be properly aborted. We just need to make sure | |
1469 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1470 | */ |
8f04c47a CH |
1471 | xfs_bmap_cancel(&free_list); |
1472 | goto out; | |
1473 | } | |
1474 | ||
1da177e4 LT |
1475 | /* |
1476 | * This is called when the inode's link count goes to 0. | |
1477 | * We place the on-disk inode on a list in the AGI. It | |
1478 | * will be pulled from this list when the inode is freed. | |
1479 | */ | |
1480 | int | |
1481 | xfs_iunlink( | |
1482 | xfs_trans_t *tp, | |
1483 | xfs_inode_t *ip) | |
1484 | { | |
1485 | xfs_mount_t *mp; | |
1486 | xfs_agi_t *agi; | |
1487 | xfs_dinode_t *dip; | |
1488 | xfs_buf_t *agibp; | |
1489 | xfs_buf_t *ibp; | |
1da177e4 LT |
1490 | xfs_agino_t agino; |
1491 | short bucket_index; | |
1492 | int offset; | |
1493 | int error; | |
1da177e4 LT |
1494 | |
1495 | ASSERT(ip->i_d.di_nlink == 0); | |
1496 | ASSERT(ip->i_d.di_mode != 0); | |
1da177e4 LT |
1497 | |
1498 | mp = tp->t_mountp; | |
1499 | ||
1da177e4 LT |
1500 | /* |
1501 | * Get the agi buffer first. It ensures lock ordering | |
1502 | * on the list. | |
1503 | */ | |
5e1be0fb | 1504 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 1505 | if (error) |
1da177e4 | 1506 | return error; |
1da177e4 | 1507 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1508 | |
1da177e4 LT |
1509 | /* |
1510 | * Get the index into the agi hash table for the | |
1511 | * list this inode will go on. | |
1512 | */ | |
1513 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1514 | ASSERT(agino != 0); | |
1515 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1516 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1517 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1518 | |
69ef921b | 1519 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
1520 | /* |
1521 | * There is already another inode in the bucket we need | |
1522 | * to add ourselves to. Add us at the front of the list. | |
1523 | * Here we put the head pointer into our next pointer, | |
1524 | * and then we fall through to point the head at us. | |
1525 | */ | |
475ee413 CH |
1526 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1527 | 0, 0); | |
c319b58b VA |
1528 | if (error) |
1529 | return error; | |
1530 | ||
69ef921b | 1531 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 1532 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 1533 | offset = ip->i_imap.im_boffset + |
1da177e4 LT |
1534 | offsetof(xfs_dinode_t, di_next_unlinked); |
1535 | xfs_trans_inode_buf(tp, ibp); | |
1536 | xfs_trans_log_buf(tp, ibp, offset, | |
1537 | (offset + sizeof(xfs_agino_t) - 1)); | |
1538 | xfs_inobp_check(mp, ibp); | |
1539 | } | |
1540 | ||
1541 | /* | |
1542 | * Point the bucket head pointer at the inode being inserted. | |
1543 | */ | |
1544 | ASSERT(agino != 0); | |
16259e7d | 1545 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
1546 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1547 | (sizeof(xfs_agino_t) * bucket_index); | |
1548 | xfs_trans_log_buf(tp, agibp, offset, | |
1549 | (offset + sizeof(xfs_agino_t) - 1)); | |
1550 | return 0; | |
1551 | } | |
1552 | ||
1553 | /* | |
1554 | * Pull the on-disk inode from the AGI unlinked list. | |
1555 | */ | |
1556 | STATIC int | |
1557 | xfs_iunlink_remove( | |
1558 | xfs_trans_t *tp, | |
1559 | xfs_inode_t *ip) | |
1560 | { | |
1561 | xfs_ino_t next_ino; | |
1562 | xfs_mount_t *mp; | |
1563 | xfs_agi_t *agi; | |
1564 | xfs_dinode_t *dip; | |
1565 | xfs_buf_t *agibp; | |
1566 | xfs_buf_t *ibp; | |
1567 | xfs_agnumber_t agno; | |
1da177e4 LT |
1568 | xfs_agino_t agino; |
1569 | xfs_agino_t next_agino; | |
1570 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 1571 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 1572 | short bucket_index; |
6fdf8ccc | 1573 | int offset, last_offset = 0; |
1da177e4 | 1574 | int error; |
1da177e4 | 1575 | |
1da177e4 | 1576 | mp = tp->t_mountp; |
1da177e4 | 1577 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
1578 | |
1579 | /* | |
1580 | * Get the agi buffer first. It ensures lock ordering | |
1581 | * on the list. | |
1582 | */ | |
5e1be0fb CH |
1583 | error = xfs_read_agi(mp, tp, agno, &agibp); |
1584 | if (error) | |
1da177e4 | 1585 | return error; |
5e1be0fb | 1586 | |
1da177e4 | 1587 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1588 | |
1da177e4 LT |
1589 | /* |
1590 | * Get the index into the agi hash table for the | |
1591 | * list this inode will go on. | |
1592 | */ | |
1593 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1594 | ASSERT(agino != 0); | |
1595 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 1596 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
1597 | ASSERT(agi->agi_unlinked[bucket_index]); |
1598 | ||
16259e7d | 1599 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 1600 | /* |
475ee413 CH |
1601 | * We're at the head of the list. Get the inode's on-disk |
1602 | * buffer to see if there is anyone after us on the list. | |
1603 | * Only modify our next pointer if it is not already NULLAGINO. | |
1604 | * This saves us the overhead of dealing with the buffer when | |
1605 | * there is no need to change it. | |
1da177e4 | 1606 | */ |
475ee413 CH |
1607 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1608 | 0, 0); | |
1da177e4 | 1609 | if (error) { |
475ee413 | 1610 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 1611 | __func__, error); |
1da177e4 LT |
1612 | return error; |
1613 | } | |
347d1c01 | 1614 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
1615 | ASSERT(next_agino != 0); |
1616 | if (next_agino != NULLAGINO) { | |
347d1c01 | 1617 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 1618 | offset = ip->i_imap.im_boffset + |
1da177e4 LT |
1619 | offsetof(xfs_dinode_t, di_next_unlinked); |
1620 | xfs_trans_inode_buf(tp, ibp); | |
1621 | xfs_trans_log_buf(tp, ibp, offset, | |
1622 | (offset + sizeof(xfs_agino_t) - 1)); | |
1623 | xfs_inobp_check(mp, ibp); | |
1624 | } else { | |
1625 | xfs_trans_brelse(tp, ibp); | |
1626 | } | |
1627 | /* | |
1628 | * Point the bucket head pointer at the next inode. | |
1629 | */ | |
1630 | ASSERT(next_agino != 0); | |
1631 | ASSERT(next_agino != agino); | |
16259e7d | 1632 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
1633 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1634 | (sizeof(xfs_agino_t) * bucket_index); | |
1635 | xfs_trans_log_buf(tp, agibp, offset, | |
1636 | (offset + sizeof(xfs_agino_t) - 1)); | |
1637 | } else { | |
1638 | /* | |
1639 | * We need to search the list for the inode being freed. | |
1640 | */ | |
16259e7d | 1641 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
1642 | last_ibp = NULL; |
1643 | while (next_agino != agino) { | |
129dbc9a CH |
1644 | struct xfs_imap imap; |
1645 | ||
1646 | if (last_ibp) | |
1da177e4 | 1647 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
1648 | |
1649 | imap.im_blkno = 0; | |
1da177e4 | 1650 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
1651 | |
1652 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
1653 | if (error) { | |
1654 | xfs_warn(mp, | |
1655 | "%s: xfs_imap returned error %d.", | |
1656 | __func__, error); | |
1657 | return error; | |
1658 | } | |
1659 | ||
1660 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
1661 | &last_ibp, 0, 0); | |
1da177e4 | 1662 | if (error) { |
0b932ccc | 1663 | xfs_warn(mp, |
129dbc9a | 1664 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 1665 | __func__, error); |
1da177e4 LT |
1666 | return error; |
1667 | } | |
129dbc9a CH |
1668 | |
1669 | last_offset = imap.im_boffset; | |
347d1c01 | 1670 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
1671 | ASSERT(next_agino != NULLAGINO); |
1672 | ASSERT(next_agino != 0); | |
1673 | } | |
475ee413 | 1674 | |
1da177e4 | 1675 | /* |
475ee413 CH |
1676 | * Now last_ibp points to the buffer previous to us on the |
1677 | * unlinked list. Pull us from the list. | |
1da177e4 | 1678 | */ |
475ee413 CH |
1679 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1680 | 0, 0); | |
1da177e4 | 1681 | if (error) { |
475ee413 | 1682 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 1683 | __func__, error); |
1da177e4 LT |
1684 | return error; |
1685 | } | |
347d1c01 | 1686 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
1687 | ASSERT(next_agino != 0); |
1688 | ASSERT(next_agino != agino); | |
1689 | if (next_agino != NULLAGINO) { | |
347d1c01 | 1690 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 1691 | offset = ip->i_imap.im_boffset + |
1da177e4 LT |
1692 | offsetof(xfs_dinode_t, di_next_unlinked); |
1693 | xfs_trans_inode_buf(tp, ibp); | |
1694 | xfs_trans_log_buf(tp, ibp, offset, | |
1695 | (offset + sizeof(xfs_agino_t) - 1)); | |
1696 | xfs_inobp_check(mp, ibp); | |
1697 | } else { | |
1698 | xfs_trans_brelse(tp, ibp); | |
1699 | } | |
1700 | /* | |
1701 | * Point the previous inode on the list to the next inode. | |
1702 | */ | |
347d1c01 | 1703 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
1704 | ASSERT(next_agino != 0); |
1705 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
1706 | xfs_trans_inode_buf(tp, last_ibp); | |
1707 | xfs_trans_log_buf(tp, last_ibp, offset, | |
1708 | (offset + sizeof(xfs_agino_t) - 1)); | |
1709 | xfs_inobp_check(mp, last_ibp); | |
1710 | } | |
1711 | return 0; | |
1712 | } | |
1713 | ||
5b3eed75 DC |
1714 | /* |
1715 | * A big issue when freeing the inode cluster is is that we _cannot_ skip any | |
1716 | * inodes that are in memory - they all must be marked stale and attached to | |
1717 | * the cluster buffer. | |
1718 | */ | |
2a30f36d | 1719 | STATIC int |
1da177e4 LT |
1720 | xfs_ifree_cluster( |
1721 | xfs_inode_t *free_ip, | |
1722 | xfs_trans_t *tp, | |
1723 | xfs_ino_t inum) | |
1724 | { | |
1725 | xfs_mount_t *mp = free_ip->i_mount; | |
1726 | int blks_per_cluster; | |
1727 | int nbufs; | |
1728 | int ninodes; | |
5b257b4a | 1729 | int i, j; |
1da177e4 LT |
1730 | xfs_daddr_t blkno; |
1731 | xfs_buf_t *bp; | |
5b257b4a | 1732 | xfs_inode_t *ip; |
1da177e4 LT |
1733 | xfs_inode_log_item_t *iip; |
1734 | xfs_log_item_t *lip; | |
5017e97d | 1735 | struct xfs_perag *pag; |
1da177e4 | 1736 | |
5017e97d | 1737 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
1da177e4 LT |
1738 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { |
1739 | blks_per_cluster = 1; | |
1740 | ninodes = mp->m_sb.sb_inopblock; | |
1741 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
1742 | } else { | |
1743 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
1744 | mp->m_sb.sb_blocksize; | |
1745 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
1746 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
1747 | } | |
1748 | ||
1da177e4 LT |
1749 | for (j = 0; j < nbufs; j++, inum += ninodes) { |
1750 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
1751 | XFS_INO_TO_AGBNO(mp, inum)); | |
1752 | ||
5b257b4a DC |
1753 | /* |
1754 | * We obtain and lock the backing buffer first in the process | |
1755 | * here, as we have to ensure that any dirty inode that we | |
1756 | * can't get the flush lock on is attached to the buffer. | |
1757 | * If we scan the in-memory inodes first, then buffer IO can | |
1758 | * complete before we get a lock on it, and hence we may fail | |
1759 | * to mark all the active inodes on the buffer stale. | |
1760 | */ | |
1761 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
a8acad70 | 1762 | mp->m_bsize * blks_per_cluster, 0); |
5b257b4a | 1763 | |
2a30f36d CS |
1764 | if (!bp) |
1765 | return ENOMEM; | |
5b257b4a DC |
1766 | /* |
1767 | * Walk the inodes already attached to the buffer and mark them | |
1768 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
1769 | * in-memory inode walk can't lock them. By marking them all |
1770 | * stale first, we will not attempt to lock them in the loop | |
1771 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 1772 | */ |
adadbeef | 1773 | lip = bp->b_fspriv; |
5b257b4a DC |
1774 | while (lip) { |
1775 | if (lip->li_type == XFS_LI_INODE) { | |
1776 | iip = (xfs_inode_log_item_t *)lip; | |
1777 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 1778 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
1779 | xfs_trans_ail_copy_lsn(mp->m_ail, |
1780 | &iip->ili_flush_lsn, | |
1781 | &iip->ili_item.li_lsn); | |
1782 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a DC |
1783 | } |
1784 | lip = lip->li_bio_list; | |
1785 | } | |
1da177e4 | 1786 | |
5b3eed75 | 1787 | |
1da177e4 | 1788 | /* |
5b257b4a DC |
1789 | * For each inode in memory attempt to add it to the inode |
1790 | * buffer and set it up for being staled on buffer IO | |
1791 | * completion. This is safe as we've locked out tail pushing | |
1792 | * and flushing by locking the buffer. | |
1da177e4 | 1793 | * |
5b257b4a DC |
1794 | * We have already marked every inode that was part of a |
1795 | * transaction stale above, which means there is no point in | |
1796 | * even trying to lock them. | |
1da177e4 | 1797 | */ |
1da177e4 | 1798 | for (i = 0; i < ninodes; i++) { |
5b3eed75 | 1799 | retry: |
1a3e8f3d | 1800 | rcu_read_lock(); |
da353b0d DC |
1801 | ip = radix_tree_lookup(&pag->pag_ici_root, |
1802 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 1803 | |
1a3e8f3d DC |
1804 | /* Inode not in memory, nothing to do */ |
1805 | if (!ip) { | |
1806 | rcu_read_unlock(); | |
1da177e4 LT |
1807 | continue; |
1808 | } | |
1809 | ||
1a3e8f3d DC |
1810 | /* |
1811 | * because this is an RCU protected lookup, we could | |
1812 | * find a recently freed or even reallocated inode | |
1813 | * during the lookup. We need to check under the | |
1814 | * i_flags_lock for a valid inode here. Skip it if it | |
1815 | * is not valid, the wrong inode or stale. | |
1816 | */ | |
1817 | spin_lock(&ip->i_flags_lock); | |
1818 | if (ip->i_ino != inum + i || | |
1819 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
1820 | spin_unlock(&ip->i_flags_lock); | |
1821 | rcu_read_unlock(); | |
1822 | continue; | |
1823 | } | |
1824 | spin_unlock(&ip->i_flags_lock); | |
1825 | ||
5b3eed75 DC |
1826 | /* |
1827 | * Don't try to lock/unlock the current inode, but we | |
1828 | * _cannot_ skip the other inodes that we did not find | |
1829 | * in the list attached to the buffer and are not | |
1830 | * already marked stale. If we can't lock it, back off | |
1831 | * and retry. | |
1832 | */ | |
5b257b4a DC |
1833 | if (ip != free_ip && |
1834 | !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
1a3e8f3d | 1835 | rcu_read_unlock(); |
5b3eed75 DC |
1836 | delay(1); |
1837 | goto retry; | |
1da177e4 | 1838 | } |
1a3e8f3d | 1839 | rcu_read_unlock(); |
1da177e4 | 1840 | |
5b3eed75 | 1841 | xfs_iflock(ip); |
5b257b4a | 1842 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 1843 | |
5b3eed75 DC |
1844 | /* |
1845 | * we don't need to attach clean inodes or those only | |
1846 | * with unlogged changes (which we throw away, anyway). | |
1847 | */ | |
1da177e4 | 1848 | iip = ip->i_itemp; |
5b3eed75 | 1849 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 1850 | ASSERT(ip != free_ip); |
1da177e4 LT |
1851 | xfs_ifunlock(ip); |
1852 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1853 | continue; | |
1854 | } | |
1855 | ||
f5d8d5c4 CH |
1856 | iip->ili_last_fields = iip->ili_fields; |
1857 | iip->ili_fields = 0; | |
1da177e4 | 1858 | iip->ili_logged = 1; |
7b2e2a31 DC |
1859 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
1860 | &iip->ili_item.li_lsn); | |
1da177e4 | 1861 | |
ca30b2a7 CH |
1862 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
1863 | &iip->ili_item); | |
5b257b4a DC |
1864 | |
1865 | if (ip != free_ip) | |
1da177e4 | 1866 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1867 | } |
1868 | ||
5b3eed75 | 1869 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
1870 | xfs_trans_binval(tp, bp); |
1871 | } | |
1872 | ||
5017e97d | 1873 | xfs_perag_put(pag); |
2a30f36d | 1874 | return 0; |
1da177e4 LT |
1875 | } |
1876 | ||
1877 | /* | |
1878 | * This is called to return an inode to the inode free list. | |
1879 | * The inode should already be truncated to 0 length and have | |
1880 | * no pages associated with it. This routine also assumes that | |
1881 | * the inode is already a part of the transaction. | |
1882 | * | |
1883 | * The on-disk copy of the inode will have been added to the list | |
1884 | * of unlinked inodes in the AGI. We need to remove the inode from | |
1885 | * that list atomically with respect to freeing it here. | |
1886 | */ | |
1887 | int | |
1888 | xfs_ifree( | |
1889 | xfs_trans_t *tp, | |
1890 | xfs_inode_t *ip, | |
1891 | xfs_bmap_free_t *flist) | |
1892 | { | |
1893 | int error; | |
1894 | int delete; | |
1895 | xfs_ino_t first_ino; | |
c319b58b VA |
1896 | xfs_dinode_t *dip; |
1897 | xfs_buf_t *ibp; | |
1da177e4 | 1898 | |
579aa9ca | 1899 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
1900 | ASSERT(ip->i_d.di_nlink == 0); |
1901 | ASSERT(ip->i_d.di_nextents == 0); | |
1902 | ASSERT(ip->i_d.di_anextents == 0); | |
ce7ae151 | 1903 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); |
1da177e4 LT |
1904 | ASSERT(ip->i_d.di_nblocks == 0); |
1905 | ||
1906 | /* | |
1907 | * Pull the on-disk inode from the AGI unlinked list. | |
1908 | */ | |
1909 | error = xfs_iunlink_remove(tp, ip); | |
1910 | if (error != 0) { | |
1911 | return error; | |
1912 | } | |
1913 | ||
1914 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
1915 | if (error != 0) { | |
1916 | return error; | |
1917 | } | |
1918 | ip->i_d.di_mode = 0; /* mark incore inode as free */ | |
1919 | ip->i_d.di_flags = 0; | |
1920 | ip->i_d.di_dmevmask = 0; | |
1921 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
1922 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
1923 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1924 | /* | |
1925 | * Bump the generation count so no one will be confused | |
1926 | * by reincarnations of this inode. | |
1927 | */ | |
1928 | ip->i_d.di_gen++; | |
c319b58b | 1929 | |
1da177e4 LT |
1930 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1931 | ||
475ee413 CH |
1932 | error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &dip, &ibp, |
1933 | 0, 0); | |
c319b58b VA |
1934 | if (error) |
1935 | return error; | |
1936 | ||
1937 | /* | |
1938 | * Clear the on-disk di_mode. This is to prevent xfs_bulkstat | |
1939 | * from picking up this inode when it is reclaimed (its incore state | |
1940 | * initialzed but not flushed to disk yet). The in-core di_mode is | |
1941 | * already cleared and a corresponding transaction logged. | |
1942 | * The hack here just synchronizes the in-core to on-disk | |
1943 | * di_mode value in advance before the actual inode sync to disk. | |
1944 | * This is OK because the inode is already unlinked and would never | |
1945 | * change its di_mode again for this inode generation. | |
1946 | * This is a temporary hack that would require a proper fix | |
1947 | * in the future. | |
1948 | */ | |
81591fe2 | 1949 | dip->di_mode = 0; |
c319b58b | 1950 | |
1da177e4 | 1951 | if (delete) { |
2a30f36d | 1952 | error = xfs_ifree_cluster(ip, tp, first_ino); |
1da177e4 LT |
1953 | } |
1954 | ||
2a30f36d | 1955 | return error; |
1da177e4 LT |
1956 | } |
1957 | ||
1958 | /* | |
1959 | * Reallocate the space for if_broot based on the number of records | |
1960 | * being added or deleted as indicated in rec_diff. Move the records | |
1961 | * and pointers in if_broot to fit the new size. When shrinking this | |
1962 | * will eliminate holes between the records and pointers created by | |
1963 | * the caller. When growing this will create holes to be filled in | |
1964 | * by the caller. | |
1965 | * | |
1966 | * The caller must not request to add more records than would fit in | |
1967 | * the on-disk inode root. If the if_broot is currently NULL, then | |
1968 | * if we adding records one will be allocated. The caller must also | |
1969 | * not request that the number of records go below zero, although | |
1970 | * it can go to zero. | |
1971 | * | |
1972 | * ip -- the inode whose if_broot area is changing | |
1973 | * ext_diff -- the change in the number of records, positive or negative, | |
1974 | * requested for the if_broot array. | |
1975 | */ | |
1976 | void | |
1977 | xfs_iroot_realloc( | |
1978 | xfs_inode_t *ip, | |
1979 | int rec_diff, | |
1980 | int whichfork) | |
1981 | { | |
60197e8d | 1982 | struct xfs_mount *mp = ip->i_mount; |
1da177e4 LT |
1983 | int cur_max; |
1984 | xfs_ifork_t *ifp; | |
7cc95a82 | 1985 | struct xfs_btree_block *new_broot; |
1da177e4 LT |
1986 | int new_max; |
1987 | size_t new_size; | |
1988 | char *np; | |
1989 | char *op; | |
1990 | ||
1991 | /* | |
1992 | * Handle the degenerate case quietly. | |
1993 | */ | |
1994 | if (rec_diff == 0) { | |
1995 | return; | |
1996 | } | |
1997 | ||
1998 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1999 | if (rec_diff > 0) { | |
2000 | /* | |
2001 | * If there wasn't any memory allocated before, just | |
2002 | * allocate it now and get out. | |
2003 | */ | |
2004 | if (ifp->if_broot_bytes == 0) { | |
2005 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); | |
4a7edddc | 2006 | ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); |
1da177e4 LT |
2007 | ifp->if_broot_bytes = (int)new_size; |
2008 | return; | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * If there is already an existing if_broot, then we need | |
2013 | * to realloc() it and shift the pointers to their new | |
2014 | * location. The records don't change location because | |
2015 | * they are kept butted up against the btree block header. | |
2016 | */ | |
60197e8d | 2017 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
1da177e4 LT |
2018 | new_max = cur_max + rec_diff; |
2019 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
7cc95a82 | 2020 | ifp->if_broot = kmem_realloc(ifp->if_broot, new_size, |
1da177e4 | 2021 | (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ |
4a7edddc | 2022 | KM_SLEEP | KM_NOFS); |
60197e8d CH |
2023 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
2024 | ifp->if_broot_bytes); | |
2025 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, | |
2026 | (int)new_size); | |
1da177e4 LT |
2027 | ifp->if_broot_bytes = (int)new_size; |
2028 | ASSERT(ifp->if_broot_bytes <= | |
2029 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2030 | memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); | |
2031 | return; | |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * rec_diff is less than 0. In this case, we are shrinking the | |
2036 | * if_broot buffer. It must already exist. If we go to zero | |
2037 | * records, just get rid of the root and clear the status bit. | |
2038 | */ | |
2039 | ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); | |
60197e8d | 2040 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
1da177e4 LT |
2041 | new_max = cur_max + rec_diff; |
2042 | ASSERT(new_max >= 0); | |
2043 | if (new_max > 0) | |
2044 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2045 | else | |
2046 | new_size = 0; | |
2047 | if (new_size > 0) { | |
4a7edddc | 2048 | new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); |
1da177e4 LT |
2049 | /* |
2050 | * First copy over the btree block header. | |
2051 | */ | |
7cc95a82 | 2052 | memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN); |
1da177e4 LT |
2053 | } else { |
2054 | new_broot = NULL; | |
2055 | ifp->if_flags &= ~XFS_IFBROOT; | |
2056 | } | |
2057 | ||
2058 | /* | |
2059 | * Only copy the records and pointers if there are any. | |
2060 | */ | |
2061 | if (new_max > 0) { | |
2062 | /* | |
2063 | * First copy the records. | |
2064 | */ | |
136341b4 CH |
2065 | op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1); |
2066 | np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1); | |
1da177e4 LT |
2067 | memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); |
2068 | ||
2069 | /* | |
2070 | * Then copy the pointers. | |
2071 | */ | |
60197e8d | 2072 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
1da177e4 | 2073 | ifp->if_broot_bytes); |
60197e8d | 2074 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1, |
1da177e4 LT |
2075 | (int)new_size); |
2076 | memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); | |
2077 | } | |
f0e2d93c | 2078 | kmem_free(ifp->if_broot); |
1da177e4 LT |
2079 | ifp->if_broot = new_broot; |
2080 | ifp->if_broot_bytes = (int)new_size; | |
2081 | ASSERT(ifp->if_broot_bytes <= | |
2082 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2083 | return; | |
2084 | } | |
2085 | ||
2086 | ||
1da177e4 LT |
2087 | /* |
2088 | * This is called when the amount of space needed for if_data | |
2089 | * is increased or decreased. The change in size is indicated by | |
2090 | * the number of bytes that need to be added or deleted in the | |
2091 | * byte_diff parameter. | |
2092 | * | |
2093 | * If the amount of space needed has decreased below the size of the | |
2094 | * inline buffer, then switch to using the inline buffer. Otherwise, | |
2095 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer | |
2096 | * to what is needed. | |
2097 | * | |
2098 | * ip -- the inode whose if_data area is changing | |
2099 | * byte_diff -- the change in the number of bytes, positive or negative, | |
2100 | * requested for the if_data array. | |
2101 | */ | |
2102 | void | |
2103 | xfs_idata_realloc( | |
2104 | xfs_inode_t *ip, | |
2105 | int byte_diff, | |
2106 | int whichfork) | |
2107 | { | |
2108 | xfs_ifork_t *ifp; | |
2109 | int new_size; | |
2110 | int real_size; | |
2111 | ||
2112 | if (byte_diff == 0) { | |
2113 | return; | |
2114 | } | |
2115 | ||
2116 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2117 | new_size = (int)ifp->if_bytes + byte_diff; | |
2118 | ASSERT(new_size >= 0); | |
2119 | ||
2120 | if (new_size == 0) { | |
2121 | if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
f0e2d93c | 2122 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2123 | } |
2124 | ifp->if_u1.if_data = NULL; | |
2125 | real_size = 0; | |
2126 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { | |
2127 | /* | |
2128 | * If the valid extents/data can fit in if_inline_ext/data, | |
2129 | * copy them from the malloc'd vector and free it. | |
2130 | */ | |
2131 | if (ifp->if_u1.if_data == NULL) { | |
2132 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2133 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2134 | ASSERT(ifp->if_real_bytes != 0); | |
2135 | memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, | |
2136 | new_size); | |
f0e2d93c | 2137 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2138 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; |
2139 | } | |
2140 | real_size = 0; | |
2141 | } else { | |
2142 | /* | |
2143 | * Stuck with malloc/realloc. | |
2144 | * For inline data, the underlying buffer must be | |
2145 | * a multiple of 4 bytes in size so that it can be | |
2146 | * logged and stay on word boundaries. We enforce | |
2147 | * that here. | |
2148 | */ | |
2149 | real_size = roundup(new_size, 4); | |
2150 | if (ifp->if_u1.if_data == NULL) { | |
2151 | ASSERT(ifp->if_real_bytes == 0); | |
4a7edddc DC |
2152 | ifp->if_u1.if_data = kmem_alloc(real_size, |
2153 | KM_SLEEP | KM_NOFS); | |
1da177e4 LT |
2154 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { |
2155 | /* | |
2156 | * Only do the realloc if the underlying size | |
2157 | * is really changing. | |
2158 | */ | |
2159 | if (ifp->if_real_bytes != real_size) { | |
2160 | ifp->if_u1.if_data = | |
2161 | kmem_realloc(ifp->if_u1.if_data, | |
2162 | real_size, | |
2163 | ifp->if_real_bytes, | |
4a7edddc | 2164 | KM_SLEEP | KM_NOFS); |
1da177e4 LT |
2165 | } |
2166 | } else { | |
2167 | ASSERT(ifp->if_real_bytes == 0); | |
4a7edddc DC |
2168 | ifp->if_u1.if_data = kmem_alloc(real_size, |
2169 | KM_SLEEP | KM_NOFS); | |
1da177e4 LT |
2170 | memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, |
2171 | ifp->if_bytes); | |
2172 | } | |
2173 | } | |
2174 | ifp->if_real_bytes = real_size; | |
2175 | ifp->if_bytes = new_size; | |
2176 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2177 | } | |
2178 | ||
1da177e4 LT |
2179 | void |
2180 | xfs_idestroy_fork( | |
2181 | xfs_inode_t *ip, | |
2182 | int whichfork) | |
2183 | { | |
2184 | xfs_ifork_t *ifp; | |
2185 | ||
2186 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2187 | if (ifp->if_broot != NULL) { | |
f0e2d93c | 2188 | kmem_free(ifp->if_broot); |
1da177e4 LT |
2189 | ifp->if_broot = NULL; |
2190 | } | |
2191 | ||
2192 | /* | |
2193 | * If the format is local, then we can't have an extents | |
2194 | * array so just look for an inline data array. If we're | |
2195 | * not local then we may or may not have an extents list, | |
2196 | * so check and free it up if we do. | |
2197 | */ | |
2198 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
2199 | if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && | |
2200 | (ifp->if_u1.if_data != NULL)) { | |
2201 | ASSERT(ifp->if_real_bytes != 0); | |
f0e2d93c | 2202 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2203 | ifp->if_u1.if_data = NULL; |
2204 | ifp->if_real_bytes = 0; | |
2205 | } | |
2206 | } else if ((ifp->if_flags & XFS_IFEXTENTS) && | |
0293ce3a MK |
2207 | ((ifp->if_flags & XFS_IFEXTIREC) || |
2208 | ((ifp->if_u1.if_extents != NULL) && | |
2209 | (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { | |
1da177e4 | 2210 | ASSERT(ifp->if_real_bytes != 0); |
4eea22f0 | 2211 | xfs_iext_destroy(ifp); |
1da177e4 LT |
2212 | } |
2213 | ASSERT(ifp->if_u1.if_extents == NULL || | |
2214 | ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); | |
2215 | ASSERT(ifp->if_real_bytes == 0); | |
2216 | if (whichfork == XFS_ATTR_FORK) { | |
2217 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
2218 | ip->i_afp = NULL; | |
2219 | } | |
2220 | } | |
2221 | ||
1da177e4 | 2222 | /* |
60ec6783 CH |
2223 | * This is called to unpin an inode. The caller must have the inode locked |
2224 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2225 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2226 | */ |
60ec6783 | 2227 | static void |
f392e631 | 2228 | xfs_iunpin( |
60ec6783 | 2229 | struct xfs_inode *ip) |
1da177e4 | 2230 | { |
579aa9ca | 2231 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2232 | |
4aaf15d1 DC |
2233 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2234 | ||
a3f74ffb | 2235 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 2236 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 2237 | |
a3f74ffb | 2238 | } |
1da177e4 | 2239 | |
f392e631 CH |
2240 | static void |
2241 | __xfs_iunpin_wait( | |
2242 | struct xfs_inode *ip) | |
2243 | { | |
2244 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2245 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2246 | ||
2247 | xfs_iunpin(ip); | |
2248 | ||
2249 | do { | |
2250 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2251 | if (xfs_ipincount(ip)) | |
2252 | io_schedule(); | |
2253 | } while (xfs_ipincount(ip)); | |
2254 | finish_wait(wq, &wait.wait); | |
2255 | } | |
2256 | ||
777df5af | 2257 | void |
a3f74ffb | 2258 | xfs_iunpin_wait( |
60ec6783 | 2259 | struct xfs_inode *ip) |
a3f74ffb | 2260 | { |
f392e631 CH |
2261 | if (xfs_ipincount(ip)) |
2262 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2263 | } |
2264 | ||
1da177e4 LT |
2265 | /* |
2266 | * xfs_iextents_copy() | |
2267 | * | |
2268 | * This is called to copy the REAL extents (as opposed to the delayed | |
2269 | * allocation extents) from the inode into the given buffer. It | |
2270 | * returns the number of bytes copied into the buffer. | |
2271 | * | |
2272 | * If there are no delayed allocation extents, then we can just | |
2273 | * memcpy() the extents into the buffer. Otherwise, we need to | |
2274 | * examine each extent in turn and skip those which are delayed. | |
2275 | */ | |
2276 | int | |
2277 | xfs_iextents_copy( | |
2278 | xfs_inode_t *ip, | |
a6f64d4a | 2279 | xfs_bmbt_rec_t *dp, |
1da177e4 LT |
2280 | int whichfork) |
2281 | { | |
2282 | int copied; | |
1da177e4 LT |
2283 | int i; |
2284 | xfs_ifork_t *ifp; | |
2285 | int nrecs; | |
2286 | xfs_fsblock_t start_block; | |
2287 | ||
2288 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
579aa9ca | 2289 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 LT |
2290 | ASSERT(ifp->if_bytes > 0); |
2291 | ||
2292 | nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3a59c94c | 2293 | XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork); |
1da177e4 LT |
2294 | ASSERT(nrecs > 0); |
2295 | ||
2296 | /* | |
2297 | * There are some delayed allocation extents in the | |
2298 | * inode, so copy the extents one at a time and skip | |
2299 | * the delayed ones. There must be at least one | |
2300 | * non-delayed extent. | |
2301 | */ | |
1da177e4 LT |
2302 | copied = 0; |
2303 | for (i = 0; i < nrecs; i++) { | |
a6f64d4a | 2304 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
1da177e4 | 2305 | start_block = xfs_bmbt_get_startblock(ep); |
9d87c319 | 2306 | if (isnullstartblock(start_block)) { |
1da177e4 LT |
2307 | /* |
2308 | * It's a delayed allocation extent, so skip it. | |
2309 | */ | |
1da177e4 LT |
2310 | continue; |
2311 | } | |
2312 | ||
2313 | /* Translate to on disk format */ | |
cd8b0a97 CH |
2314 | put_unaligned(cpu_to_be64(ep->l0), &dp->l0); |
2315 | put_unaligned(cpu_to_be64(ep->l1), &dp->l1); | |
a6f64d4a | 2316 | dp++; |
1da177e4 LT |
2317 | copied++; |
2318 | } | |
2319 | ASSERT(copied != 0); | |
a6f64d4a | 2320 | xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
2321 | |
2322 | return (copied * (uint)sizeof(xfs_bmbt_rec_t)); | |
2323 | } | |
2324 | ||
2325 | /* | |
2326 | * Each of the following cases stores data into the same region | |
2327 | * of the on-disk inode, so only one of them can be valid at | |
2328 | * any given time. While it is possible to have conflicting formats | |
2329 | * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is | |
2330 | * in EXTENTS format, this can only happen when the fork has | |
2331 | * changed formats after being modified but before being flushed. | |
2332 | * In these cases, the format always takes precedence, because the | |
2333 | * format indicates the current state of the fork. | |
2334 | */ | |
2335 | /*ARGSUSED*/ | |
e4ac967b | 2336 | STATIC void |
1da177e4 LT |
2337 | xfs_iflush_fork( |
2338 | xfs_inode_t *ip, | |
2339 | xfs_dinode_t *dip, | |
2340 | xfs_inode_log_item_t *iip, | |
2341 | int whichfork, | |
2342 | xfs_buf_t *bp) | |
2343 | { | |
2344 | char *cp; | |
2345 | xfs_ifork_t *ifp; | |
2346 | xfs_mount_t *mp; | |
2347 | #ifdef XFS_TRANS_DEBUG | |
2348 | int first; | |
2349 | #endif | |
2350 | static const short brootflag[2] = | |
2351 | { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; | |
2352 | static const short dataflag[2] = | |
2353 | { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; | |
2354 | static const short extflag[2] = | |
2355 | { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; | |
2356 | ||
e4ac967b DC |
2357 | if (!iip) |
2358 | return; | |
1da177e4 LT |
2359 | ifp = XFS_IFORK_PTR(ip, whichfork); |
2360 | /* | |
2361 | * This can happen if we gave up in iformat in an error path, | |
2362 | * for the attribute fork. | |
2363 | */ | |
e4ac967b | 2364 | if (!ifp) { |
1da177e4 | 2365 | ASSERT(whichfork == XFS_ATTR_FORK); |
e4ac967b | 2366 | return; |
1da177e4 LT |
2367 | } |
2368 | cp = XFS_DFORK_PTR(dip, whichfork); | |
2369 | mp = ip->i_mount; | |
2370 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { | |
2371 | case XFS_DINODE_FMT_LOCAL: | |
f5d8d5c4 | 2372 | if ((iip->ili_fields & dataflag[whichfork]) && |
1da177e4 LT |
2373 | (ifp->if_bytes > 0)) { |
2374 | ASSERT(ifp->if_u1.if_data != NULL); | |
2375 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2376 | memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); | |
2377 | } | |
1da177e4 LT |
2378 | break; |
2379 | ||
2380 | case XFS_DINODE_FMT_EXTENTS: | |
2381 | ASSERT((ifp->if_flags & XFS_IFEXTENTS) || | |
f5d8d5c4 CH |
2382 | !(iip->ili_fields & extflag[whichfork])); |
2383 | if ((iip->ili_fields & extflag[whichfork]) && | |
1da177e4 | 2384 | (ifp->if_bytes > 0)) { |
ab1908a5 | 2385 | ASSERT(xfs_iext_get_ext(ifp, 0)); |
1da177e4 LT |
2386 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); |
2387 | (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, | |
2388 | whichfork); | |
2389 | } | |
2390 | break; | |
2391 | ||
2392 | case XFS_DINODE_FMT_BTREE: | |
f5d8d5c4 | 2393 | if ((iip->ili_fields & brootflag[whichfork]) && |
1da177e4 LT |
2394 | (ifp->if_broot_bytes > 0)) { |
2395 | ASSERT(ifp->if_broot != NULL); | |
2396 | ASSERT(ifp->if_broot_bytes <= | |
2397 | (XFS_IFORK_SIZE(ip, whichfork) + | |
2398 | XFS_BROOT_SIZE_ADJ)); | |
60197e8d | 2399 | xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes, |
1da177e4 LT |
2400 | (xfs_bmdr_block_t *)cp, |
2401 | XFS_DFORK_SIZE(dip, mp, whichfork)); | |
2402 | } | |
2403 | break; | |
2404 | ||
2405 | case XFS_DINODE_FMT_DEV: | |
f5d8d5c4 | 2406 | if (iip->ili_fields & XFS_ILOG_DEV) { |
1da177e4 | 2407 | ASSERT(whichfork == XFS_DATA_FORK); |
81591fe2 | 2408 | xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev); |
1da177e4 LT |
2409 | } |
2410 | break; | |
2411 | ||
2412 | case XFS_DINODE_FMT_UUID: | |
f5d8d5c4 | 2413 | if (iip->ili_fields & XFS_ILOG_UUID) { |
1da177e4 | 2414 | ASSERT(whichfork == XFS_DATA_FORK); |
81591fe2 CH |
2415 | memcpy(XFS_DFORK_DPTR(dip), |
2416 | &ip->i_df.if_u2.if_uuid, | |
2417 | sizeof(uuid_t)); | |
1da177e4 LT |
2418 | } |
2419 | break; | |
2420 | ||
2421 | default: | |
2422 | ASSERT(0); | |
2423 | break; | |
2424 | } | |
1da177e4 LT |
2425 | } |
2426 | ||
bad55843 DC |
2427 | STATIC int |
2428 | xfs_iflush_cluster( | |
2429 | xfs_inode_t *ip, | |
2430 | xfs_buf_t *bp) | |
2431 | { | |
2432 | xfs_mount_t *mp = ip->i_mount; | |
5017e97d | 2433 | struct xfs_perag *pag; |
bad55843 | 2434 | unsigned long first_index, mask; |
c8f5f12e | 2435 | unsigned long inodes_per_cluster; |
bad55843 DC |
2436 | int ilist_size; |
2437 | xfs_inode_t **ilist; | |
2438 | xfs_inode_t *iq; | |
bad55843 DC |
2439 | int nr_found; |
2440 | int clcount = 0; | |
2441 | int bufwasdelwri; | |
2442 | int i; | |
2443 | ||
5017e97d | 2444 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
bad55843 | 2445 | |
c8f5f12e DC |
2446 | inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; |
2447 | ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); | |
49383b0e | 2448 | ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); |
bad55843 | 2449 | if (!ilist) |
44b56e0a | 2450 | goto out_put; |
bad55843 DC |
2451 | |
2452 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); | |
2453 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; | |
1a3e8f3d | 2454 | rcu_read_lock(); |
bad55843 DC |
2455 | /* really need a gang lookup range call here */ |
2456 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, | |
c8f5f12e | 2457 | first_index, inodes_per_cluster); |
bad55843 DC |
2458 | if (nr_found == 0) |
2459 | goto out_free; | |
2460 | ||
2461 | for (i = 0; i < nr_found; i++) { | |
2462 | iq = ilist[i]; | |
2463 | if (iq == ip) | |
2464 | continue; | |
1a3e8f3d DC |
2465 | |
2466 | /* | |
2467 | * because this is an RCU protected lookup, we could find a | |
2468 | * recently freed or even reallocated inode during the lookup. | |
2469 | * We need to check under the i_flags_lock for a valid inode | |
2470 | * here. Skip it if it is not valid or the wrong inode. | |
2471 | */ | |
2472 | spin_lock(&ip->i_flags_lock); | |
2473 | if (!ip->i_ino || | |
2474 | (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { | |
2475 | spin_unlock(&ip->i_flags_lock); | |
2476 | continue; | |
2477 | } | |
2478 | spin_unlock(&ip->i_flags_lock); | |
2479 | ||
bad55843 DC |
2480 | /* |
2481 | * Do an un-protected check to see if the inode is dirty and | |
2482 | * is a candidate for flushing. These checks will be repeated | |
2483 | * later after the appropriate locks are acquired. | |
2484 | */ | |
33540408 | 2485 | if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) |
bad55843 | 2486 | continue; |
bad55843 DC |
2487 | |
2488 | /* | |
2489 | * Try to get locks. If any are unavailable or it is pinned, | |
2490 | * then this inode cannot be flushed and is skipped. | |
2491 | */ | |
2492 | ||
2493 | if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) | |
2494 | continue; | |
2495 | if (!xfs_iflock_nowait(iq)) { | |
2496 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2497 | continue; | |
2498 | } | |
2499 | if (xfs_ipincount(iq)) { | |
2500 | xfs_ifunlock(iq); | |
2501 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2502 | continue; | |
2503 | } | |
2504 | ||
2505 | /* | |
2506 | * arriving here means that this inode can be flushed. First | |
2507 | * re-check that it's dirty before flushing. | |
2508 | */ | |
33540408 DC |
2509 | if (!xfs_inode_clean(iq)) { |
2510 | int error; | |
bad55843 DC |
2511 | error = xfs_iflush_int(iq, bp); |
2512 | if (error) { | |
2513 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2514 | goto cluster_corrupt_out; | |
2515 | } | |
2516 | clcount++; | |
2517 | } else { | |
2518 | xfs_ifunlock(iq); | |
2519 | } | |
2520 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2521 | } | |
2522 | ||
2523 | if (clcount) { | |
2524 | XFS_STATS_INC(xs_icluster_flushcnt); | |
2525 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
2526 | } | |
2527 | ||
2528 | out_free: | |
1a3e8f3d | 2529 | rcu_read_unlock(); |
f0e2d93c | 2530 | kmem_free(ilist); |
44b56e0a DC |
2531 | out_put: |
2532 | xfs_perag_put(pag); | |
bad55843 DC |
2533 | return 0; |
2534 | ||
2535 | ||
2536 | cluster_corrupt_out: | |
2537 | /* | |
2538 | * Corruption detected in the clustering loop. Invalidate the | |
2539 | * inode buffer and shut down the filesystem. | |
2540 | */ | |
1a3e8f3d | 2541 | rcu_read_unlock(); |
bad55843 | 2542 | /* |
43ff2122 | 2543 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
2544 | * brelse can handle it with no problems. If not, shut down the |
2545 | * filesystem before releasing the buffer. | |
2546 | */ | |
43ff2122 | 2547 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
2548 | if (bufwasdelwri) |
2549 | xfs_buf_relse(bp); | |
2550 | ||
2551 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
2552 | ||
2553 | if (!bufwasdelwri) { | |
2554 | /* | |
2555 | * Just like incore_relse: if we have b_iodone functions, | |
2556 | * mark the buffer as an error and call them. Otherwise | |
2557 | * mark it as stale and brelse. | |
2558 | */ | |
cb669ca5 | 2559 | if (bp->b_iodone) { |
bad55843 | 2560 | XFS_BUF_UNDONE(bp); |
c867cb61 | 2561 | xfs_buf_stale(bp); |
5a52c2a5 | 2562 | xfs_buf_ioerror(bp, EIO); |
1a1a3e97 | 2563 | xfs_buf_ioend(bp, 0); |
bad55843 | 2564 | } else { |
c867cb61 | 2565 | xfs_buf_stale(bp); |
bad55843 DC |
2566 | xfs_buf_relse(bp); |
2567 | } | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * Unlocks the flush lock | |
2572 | */ | |
04913fdd | 2573 | xfs_iflush_abort(iq, false); |
f0e2d93c | 2574 | kmem_free(ilist); |
44b56e0a | 2575 | xfs_perag_put(pag); |
bad55843 DC |
2576 | return XFS_ERROR(EFSCORRUPTED); |
2577 | } | |
2578 | ||
1da177e4 | 2579 | /* |
4c46819a CH |
2580 | * Flush dirty inode metadata into the backing buffer. |
2581 | * | |
2582 | * The caller must have the inode lock and the inode flush lock held. The | |
2583 | * inode lock will still be held upon return to the caller, and the inode | |
2584 | * flush lock will be released after the inode has reached the disk. | |
2585 | * | |
2586 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
2587 | */ |
2588 | int | |
2589 | xfs_iflush( | |
4c46819a CH |
2590 | struct xfs_inode *ip, |
2591 | struct xfs_buf **bpp) | |
1da177e4 | 2592 | { |
4c46819a CH |
2593 | struct xfs_mount *mp = ip->i_mount; |
2594 | struct xfs_buf *bp; | |
2595 | struct xfs_dinode *dip; | |
1da177e4 | 2596 | int error; |
1da177e4 LT |
2597 | |
2598 | XFS_STATS_INC(xs_iflush_count); | |
2599 | ||
579aa9ca | 2600 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 2601 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 2602 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 2603 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 2604 | |
4c46819a | 2605 | *bpp = NULL; |
1da177e4 | 2606 | |
1da177e4 LT |
2607 | xfs_iunpin_wait(ip); |
2608 | ||
4b6a4688 DC |
2609 | /* |
2610 | * For stale inodes we cannot rely on the backing buffer remaining | |
2611 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 2612 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
2613 | * inodes below. We have to check this after ensuring the inode is |
2614 | * unpinned so that it is safe to reclaim the stale inode after the | |
2615 | * flush call. | |
2616 | */ | |
2617 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
2618 | xfs_ifunlock(ip); | |
2619 | return 0; | |
2620 | } | |
2621 | ||
1da177e4 LT |
2622 | /* |
2623 | * This may have been unpinned because the filesystem is shutting | |
2624 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
2625 | * to disk, because the log record didn't make it to disk. |
2626 | * | |
2627 | * We also have to remove the log item from the AIL in this case, | |
2628 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
2629 | */ |
2630 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
32ce90a4 CH |
2631 | error = XFS_ERROR(EIO); |
2632 | goto abort_out; | |
1da177e4 LT |
2633 | } |
2634 | ||
a3f74ffb DC |
2635 | /* |
2636 | * Get the buffer containing the on-disk inode. | |
2637 | */ | |
475ee413 CH |
2638 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
2639 | 0); | |
a3f74ffb DC |
2640 | if (error || !bp) { |
2641 | xfs_ifunlock(ip); | |
2642 | return error; | |
2643 | } | |
2644 | ||
1da177e4 LT |
2645 | /* |
2646 | * First flush out the inode that xfs_iflush was called with. | |
2647 | */ | |
2648 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 2649 | if (error) |
1da177e4 | 2650 | goto corrupt_out; |
1da177e4 | 2651 | |
a3f74ffb DC |
2652 | /* |
2653 | * If the buffer is pinned then push on the log now so we won't | |
2654 | * get stuck waiting in the write for too long. | |
2655 | */ | |
811e64c7 | 2656 | if (xfs_buf_ispinned(bp)) |
a14a348b | 2657 | xfs_log_force(mp, 0); |
a3f74ffb | 2658 | |
1da177e4 LT |
2659 | /* |
2660 | * inode clustering: | |
2661 | * see if other inodes can be gathered into this write | |
2662 | */ | |
bad55843 DC |
2663 | error = xfs_iflush_cluster(ip, bp); |
2664 | if (error) | |
2665 | goto cluster_corrupt_out; | |
1da177e4 | 2666 | |
4c46819a CH |
2667 | *bpp = bp; |
2668 | return 0; | |
1da177e4 LT |
2669 | |
2670 | corrupt_out: | |
2671 | xfs_buf_relse(bp); | |
7d04a335 | 2672 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 2673 | cluster_corrupt_out: |
32ce90a4 CH |
2674 | error = XFS_ERROR(EFSCORRUPTED); |
2675 | abort_out: | |
1da177e4 LT |
2676 | /* |
2677 | * Unlocks the flush lock | |
2678 | */ | |
04913fdd | 2679 | xfs_iflush_abort(ip, false); |
32ce90a4 | 2680 | return error; |
1da177e4 LT |
2681 | } |
2682 | ||
2683 | ||
2684 | STATIC int | |
2685 | xfs_iflush_int( | |
2686 | xfs_inode_t *ip, | |
2687 | xfs_buf_t *bp) | |
2688 | { | |
2689 | xfs_inode_log_item_t *iip; | |
2690 | xfs_dinode_t *dip; | |
2691 | xfs_mount_t *mp; | |
2692 | #ifdef XFS_TRANS_DEBUG | |
2693 | int first; | |
2694 | #endif | |
1da177e4 | 2695 | |
579aa9ca | 2696 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 2697 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 2698 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 2699 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 LT |
2700 | |
2701 | iip = ip->i_itemp; | |
2702 | mp = ip->i_mount; | |
2703 | ||
1da177e4 | 2704 | /* set *dip = inode's place in the buffer */ |
92bfc6e7 | 2705 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 2706 | |
69ef921b | 2707 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
1da177e4 | 2708 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
6a19d939 DC |
2709 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2710 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
2711 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); | |
1da177e4 LT |
2712 | goto corrupt_out; |
2713 | } | |
2714 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
2715 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
6a19d939 DC |
2716 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2717 | "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
2718 | __func__, ip->i_ino, ip, ip->i_d.di_magic); | |
1da177e4 LT |
2719 | goto corrupt_out; |
2720 | } | |
abbede1b | 2721 | if (S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
2722 | if (XFS_TEST_ERROR( |
2723 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
2724 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
2725 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
6a19d939 DC |
2726 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2727 | "%s: Bad regular inode %Lu, ptr 0x%p", | |
2728 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
2729 | goto corrupt_out; |
2730 | } | |
abbede1b | 2731 | } else if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
2732 | if (XFS_TEST_ERROR( |
2733 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
2734 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
2735 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
2736 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
6a19d939 DC |
2737 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2738 | "%s: Bad directory inode %Lu, ptr 0x%p", | |
2739 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
2740 | goto corrupt_out; |
2741 | } | |
2742 | } | |
2743 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
2744 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
2745 | XFS_RANDOM_IFLUSH_5)) { | |
6a19d939 DC |
2746 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2747 | "%s: detected corrupt incore inode %Lu, " | |
2748 | "total extents = %d, nblocks = %Ld, ptr 0x%p", | |
2749 | __func__, ip->i_ino, | |
1da177e4 | 2750 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 2751 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
2752 | goto corrupt_out; |
2753 | } | |
2754 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
2755 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
6a19d939 DC |
2756 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2757 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
2758 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); | |
1da177e4 LT |
2759 | goto corrupt_out; |
2760 | } | |
2761 | /* | |
2762 | * bump the flush iteration count, used to detect flushes which | |
2763 | * postdate a log record during recovery. | |
2764 | */ | |
2765 | ||
2766 | ip->i_d.di_flushiter++; | |
2767 | ||
2768 | /* | |
2769 | * Copy the dirty parts of the inode into the on-disk | |
2770 | * inode. We always copy out the core of the inode, | |
2771 | * because if the inode is dirty at all the core must | |
2772 | * be. | |
2773 | */ | |
81591fe2 | 2774 | xfs_dinode_to_disk(dip, &ip->i_d); |
1da177e4 LT |
2775 | |
2776 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
2777 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
2778 | ip->i_d.di_flushiter = 0; | |
2779 | ||
2780 | /* | |
2781 | * If this is really an old format inode and the superblock version | |
2782 | * has not been updated to support only new format inodes, then | |
2783 | * convert back to the old inode format. If the superblock version | |
2784 | * has been updated, then make the conversion permanent. | |
2785 | */ | |
51ce16d5 CH |
2786 | ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); |
2787 | if (ip->i_d.di_version == 1) { | |
62118709 | 2788 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
1da177e4 LT |
2789 | /* |
2790 | * Convert it back. | |
2791 | */ | |
2792 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
81591fe2 | 2793 | dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink); |
1da177e4 LT |
2794 | } else { |
2795 | /* | |
2796 | * The superblock version has already been bumped, | |
2797 | * so just make the conversion to the new inode | |
2798 | * format permanent. | |
2799 | */ | |
51ce16d5 CH |
2800 | ip->i_d.di_version = 2; |
2801 | dip->di_version = 2; | |
1da177e4 | 2802 | ip->i_d.di_onlink = 0; |
81591fe2 | 2803 | dip->di_onlink = 0; |
1da177e4 | 2804 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
81591fe2 CH |
2805 | memset(&(dip->di_pad[0]), 0, |
2806 | sizeof(dip->di_pad)); | |
6743099c | 2807 | ASSERT(xfs_get_projid(ip) == 0); |
1da177e4 LT |
2808 | } |
2809 | } | |
2810 | ||
e4ac967b DC |
2811 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); |
2812 | if (XFS_IFORK_Q(ip)) | |
2813 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
1da177e4 LT |
2814 | xfs_inobp_check(mp, bp); |
2815 | ||
2816 | /* | |
f5d8d5c4 CH |
2817 | * We've recorded everything logged in the inode, so we'd like to clear |
2818 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
2819 | * However, we can't stop logging all this information until the data | |
2820 | * we've copied into the disk buffer is written to disk. If we did we | |
2821 | * might overwrite the copy of the inode in the log with all the data | |
2822 | * after re-logging only part of it, and in the face of a crash we | |
2823 | * wouldn't have all the data we need to recover. | |
1da177e4 | 2824 | * |
f5d8d5c4 CH |
2825 | * What we do is move the bits to the ili_last_fields field. When |
2826 | * logging the inode, these bits are moved back to the ili_fields field. | |
2827 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
2828 | * know that the information those bits represent is permanently on | |
2829 | * disk. As long as the flush completes before the inode is logged | |
2830 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 2831 | * |
f5d8d5c4 CH |
2832 | * We can play with the ili_fields bits here, because the inode lock |
2833 | * must be held exclusively in order to set bits there and the flush | |
2834 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
2835 | * done routine can tell whether or not to look in the AIL. Also, store | |
2836 | * the current LSN of the inode so that we can tell whether the item has | |
2837 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
2838 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
2839 | * atomically. | |
1da177e4 | 2840 | */ |
f5d8d5c4 CH |
2841 | if (iip != NULL && iip->ili_fields != 0) { |
2842 | iip->ili_last_fields = iip->ili_fields; | |
2843 | iip->ili_fields = 0; | |
1da177e4 LT |
2844 | iip->ili_logged = 1; |
2845 | ||
7b2e2a31 DC |
2846 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2847 | &iip->ili_item.li_lsn); | |
1da177e4 LT |
2848 | |
2849 | /* | |
2850 | * Attach the function xfs_iflush_done to the inode's | |
2851 | * buffer. This will remove the inode from the AIL | |
2852 | * and unlock the inode's flush lock when the inode is | |
2853 | * completely written to disk. | |
2854 | */ | |
ca30b2a7 | 2855 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); |
1da177e4 | 2856 | |
adadbeef | 2857 | ASSERT(bp->b_fspriv != NULL); |
cb669ca5 | 2858 | ASSERT(bp->b_iodone != NULL); |
1da177e4 LT |
2859 | } else { |
2860 | /* | |
2861 | * We're flushing an inode which is not in the AIL and has | |
8a9c9980 | 2862 | * not been logged. For this case we can immediately drop |
1da177e4 LT |
2863 | * the inode flush lock because we can avoid the whole |
2864 | * AIL state thing. It's OK to drop the flush lock now, | |
2865 | * because we've already locked the buffer and to do anything | |
2866 | * you really need both. | |
2867 | */ | |
2868 | if (iip != NULL) { | |
2869 | ASSERT(iip->ili_logged == 0); | |
2870 | ASSERT(iip->ili_last_fields == 0); | |
2871 | ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0); | |
2872 | } | |
2873 | xfs_ifunlock(ip); | |
2874 | } | |
2875 | ||
2876 | return 0; | |
2877 | ||
2878 | corrupt_out: | |
2879 | return XFS_ERROR(EFSCORRUPTED); | |
2880 | } | |
2881 | ||
4eea22f0 MK |
2882 | /* |
2883 | * Return a pointer to the extent record at file index idx. | |
2884 | */ | |
a6f64d4a | 2885 | xfs_bmbt_rec_host_t * |
4eea22f0 MK |
2886 | xfs_iext_get_ext( |
2887 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
2888 | xfs_extnum_t idx) /* index of target extent */ | |
2889 | { | |
2890 | ASSERT(idx >= 0); | |
87bef181 CH |
2891 | ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)); |
2892 | ||
0293ce3a MK |
2893 | if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { |
2894 | return ifp->if_u1.if_ext_irec->er_extbuf; | |
2895 | } else if (ifp->if_flags & XFS_IFEXTIREC) { | |
2896 | xfs_ext_irec_t *erp; /* irec pointer */ | |
2897 | int erp_idx = 0; /* irec index */ | |
2898 | xfs_extnum_t page_idx = idx; /* ext index in target list */ | |
2899 | ||
2900 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
2901 | return &erp->er_extbuf[page_idx]; | |
2902 | } else if (ifp->if_bytes) { | |
4eea22f0 MK |
2903 | return &ifp->if_u1.if_extents[idx]; |
2904 | } else { | |
2905 | return NULL; | |
2906 | } | |
2907 | } | |
2908 | ||
2909 | /* | |
2910 | * Insert new item(s) into the extent records for incore inode | |
2911 | * fork 'ifp'. 'count' new items are inserted at index 'idx'. | |
2912 | */ | |
2913 | void | |
2914 | xfs_iext_insert( | |
6ef35544 | 2915 | xfs_inode_t *ip, /* incore inode pointer */ |
4eea22f0 MK |
2916 | xfs_extnum_t idx, /* starting index of new items */ |
2917 | xfs_extnum_t count, /* number of inserted items */ | |
6ef35544 CH |
2918 | xfs_bmbt_irec_t *new, /* items to insert */ |
2919 | int state) /* type of extent conversion */ | |
4eea22f0 | 2920 | { |
6ef35544 | 2921 | xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df; |
4eea22f0 MK |
2922 | xfs_extnum_t i; /* extent record index */ |
2923 | ||
0b1b213f CH |
2924 | trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_); |
2925 | ||
4eea22f0 MK |
2926 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
2927 | xfs_iext_add(ifp, idx, count); | |
a6f64d4a CH |
2928 | for (i = idx; i < idx + count; i++, new++) |
2929 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new); | |
4eea22f0 MK |
2930 | } |
2931 | ||
2932 | /* | |
2933 | * This is called when the amount of space required for incore file | |
2934 | * extents needs to be increased. The ext_diff parameter stores the | |
2935 | * number of new extents being added and the idx parameter contains | |
2936 | * the extent index where the new extents will be added. If the new | |
2937 | * extents are being appended, then we just need to (re)allocate and | |
2938 | * initialize the space. Otherwise, if the new extents are being | |
2939 | * inserted into the middle of the existing entries, a bit more work | |
2940 | * is required to make room for the new extents to be inserted. The | |
2941 | * caller is responsible for filling in the new extent entries upon | |
2942 | * return. | |
2943 | */ | |
2944 | void | |
2945 | xfs_iext_add( | |
2946 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
2947 | xfs_extnum_t idx, /* index to begin adding exts */ | |
c41564b5 | 2948 | int ext_diff) /* number of extents to add */ |
4eea22f0 MK |
2949 | { |
2950 | int byte_diff; /* new bytes being added */ | |
2951 | int new_size; /* size of extents after adding */ | |
2952 | xfs_extnum_t nextents; /* number of extents in file */ | |
2953 | ||
2954 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
2955 | ASSERT((idx >= 0) && (idx <= nextents)); | |
2956 | byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); | |
2957 | new_size = ifp->if_bytes + byte_diff; | |
2958 | /* | |
2959 | * If the new number of extents (nextents + ext_diff) | |
2960 | * fits inside the inode, then continue to use the inline | |
2961 | * extent buffer. | |
2962 | */ | |
2963 | if (nextents + ext_diff <= XFS_INLINE_EXTS) { | |
2964 | if (idx < nextents) { | |
2965 | memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], | |
2966 | &ifp->if_u2.if_inline_ext[idx], | |
2967 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
2968 | memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); | |
2969 | } | |
2970 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
2971 | ifp->if_real_bytes = 0; | |
2972 | } | |
2973 | /* | |
2974 | * Otherwise use a linear (direct) extent list. | |
2975 | * If the extents are currently inside the inode, | |
2976 | * xfs_iext_realloc_direct will switch us from | |
2977 | * inline to direct extent allocation mode. | |
2978 | */ | |
0293ce3a | 2979 | else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { |
4eea22f0 MK |
2980 | xfs_iext_realloc_direct(ifp, new_size); |
2981 | if (idx < nextents) { | |
2982 | memmove(&ifp->if_u1.if_extents[idx + ext_diff], | |
2983 | &ifp->if_u1.if_extents[idx], | |
2984 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
2985 | memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); | |
2986 | } | |
2987 | } | |
0293ce3a MK |
2988 | /* Indirection array */ |
2989 | else { | |
2990 | xfs_ext_irec_t *erp; | |
2991 | int erp_idx = 0; | |
2992 | int page_idx = idx; | |
2993 | ||
2994 | ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); | |
2995 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
2996 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); | |
2997 | } else { | |
2998 | xfs_iext_irec_init(ifp); | |
2999 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3000 | erp = ifp->if_u1.if_ext_irec; | |
3001 | } | |
3002 | /* Extents fit in target extent page */ | |
3003 | if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { | |
3004 | if (page_idx < erp->er_extcount) { | |
3005 | memmove(&erp->er_extbuf[page_idx + ext_diff], | |
3006 | &erp->er_extbuf[page_idx], | |
3007 | (erp->er_extcount - page_idx) * | |
3008 | sizeof(xfs_bmbt_rec_t)); | |
3009 | memset(&erp->er_extbuf[page_idx], 0, byte_diff); | |
3010 | } | |
3011 | erp->er_extcount += ext_diff; | |
3012 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3013 | } | |
3014 | /* Insert a new extent page */ | |
3015 | else if (erp) { | |
3016 | xfs_iext_add_indirect_multi(ifp, | |
3017 | erp_idx, page_idx, ext_diff); | |
3018 | } | |
3019 | /* | |
3020 | * If extent(s) are being appended to the last page in | |
3021 | * the indirection array and the new extent(s) don't fit | |
3022 | * in the page, then erp is NULL and erp_idx is set to | |
3023 | * the next index needed in the indirection array. | |
3024 | */ | |
3025 | else { | |
3026 | int count = ext_diff; | |
3027 | ||
3028 | while (count) { | |
3029 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3030 | erp->er_extcount = count; | |
3031 | count -= MIN(count, (int)XFS_LINEAR_EXTS); | |
3032 | if (count) { | |
3033 | erp_idx++; | |
3034 | } | |
3035 | } | |
3036 | } | |
3037 | } | |
4eea22f0 MK |
3038 | ifp->if_bytes = new_size; |
3039 | } | |
3040 | ||
0293ce3a MK |
3041 | /* |
3042 | * This is called when incore extents are being added to the indirection | |
3043 | * array and the new extents do not fit in the target extent list. The | |
3044 | * erp_idx parameter contains the irec index for the target extent list | |
3045 | * in the indirection array, and the idx parameter contains the extent | |
3046 | * index within the list. The number of extents being added is stored | |
3047 | * in the count parameter. | |
3048 | * | |
3049 | * |-------| |-------| | |
3050 | * | | | | idx - number of extents before idx | |
3051 | * | idx | | count | | |
3052 | * | | | | count - number of extents being inserted at idx | |
3053 | * |-------| |-------| | |
3054 | * | count | | nex2 | nex2 - number of extents after idx + count | |
3055 | * |-------| |-------| | |
3056 | */ | |
3057 | void | |
3058 | xfs_iext_add_indirect_multi( | |
3059 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3060 | int erp_idx, /* target extent irec index */ | |
3061 | xfs_extnum_t idx, /* index within target list */ | |
3062 | int count) /* new extents being added */ | |
3063 | { | |
3064 | int byte_diff; /* new bytes being added */ | |
3065 | xfs_ext_irec_t *erp; /* pointer to irec entry */ | |
3066 | xfs_extnum_t ext_diff; /* number of extents to add */ | |
3067 | xfs_extnum_t ext_cnt; /* new extents still needed */ | |
3068 | xfs_extnum_t nex2; /* extents after idx + count */ | |
3069 | xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ | |
3070 | int nlists; /* number of irec's (lists) */ | |
3071 | ||
3072 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3073 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3074 | nex2 = erp->er_extcount - idx; | |
3075 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3076 | ||
3077 | /* | |
3078 | * Save second part of target extent list | |
3079 | * (all extents past */ | |
3080 | if (nex2) { | |
3081 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
6785073b | 3082 | nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS); |
0293ce3a MK |
3083 | memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); |
3084 | erp->er_extcount -= nex2; | |
3085 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); | |
3086 | memset(&erp->er_extbuf[idx], 0, byte_diff); | |
3087 | } | |
3088 | ||
3089 | /* | |
3090 | * Add the new extents to the end of the target | |
3091 | * list, then allocate new irec record(s) and | |
3092 | * extent buffer(s) as needed to store the rest | |
3093 | * of the new extents. | |
3094 | */ | |
3095 | ext_cnt = count; | |
3096 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); | |
3097 | if (ext_diff) { | |
3098 | erp->er_extcount += ext_diff; | |
3099 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3100 | ext_cnt -= ext_diff; | |
3101 | } | |
3102 | while (ext_cnt) { | |
3103 | erp_idx++; | |
3104 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3105 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); | |
3106 | erp->er_extcount = ext_diff; | |
3107 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3108 | ext_cnt -= ext_diff; | |
3109 | } | |
3110 | ||
3111 | /* Add nex2 extents back to indirection array */ | |
3112 | if (nex2) { | |
3113 | xfs_extnum_t ext_avail; | |
3114 | int i; | |
3115 | ||
3116 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
3117 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; | |
3118 | i = 0; | |
3119 | /* | |
3120 | * If nex2 extents fit in the current page, append | |
3121 | * nex2_ep after the new extents. | |
3122 | */ | |
3123 | if (nex2 <= ext_avail) { | |
3124 | i = erp->er_extcount; | |
3125 | } | |
3126 | /* | |
3127 | * Otherwise, check if space is available in the | |
3128 | * next page. | |
3129 | */ | |
3130 | else if ((erp_idx < nlists - 1) && | |
3131 | (nex2 <= (ext_avail = XFS_LINEAR_EXTS - | |
3132 | ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { | |
3133 | erp_idx++; | |
3134 | erp++; | |
3135 | /* Create a hole for nex2 extents */ | |
3136 | memmove(&erp->er_extbuf[nex2], erp->er_extbuf, | |
3137 | erp->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
3138 | } | |
3139 | /* | |
3140 | * Final choice, create a new extent page for | |
3141 | * nex2 extents. | |
3142 | */ | |
3143 | else { | |
3144 | erp_idx++; | |
3145 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3146 | } | |
3147 | memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); | |
f0e2d93c | 3148 | kmem_free(nex2_ep); |
0293ce3a MK |
3149 | erp->er_extcount += nex2; |
3150 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); | |
3151 | } | |
3152 | } | |
3153 | ||
4eea22f0 MK |
3154 | /* |
3155 | * This is called when the amount of space required for incore file | |
3156 | * extents needs to be decreased. The ext_diff parameter stores the | |
3157 | * number of extents to be removed and the idx parameter contains | |
3158 | * the extent index where the extents will be removed from. | |
0293ce3a MK |
3159 | * |
3160 | * If the amount of space needed has decreased below the linear | |
3161 | * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous | |
3162 | * extent array. Otherwise, use kmem_realloc() to adjust the | |
3163 | * size to what is needed. | |
4eea22f0 MK |
3164 | */ |
3165 | void | |
3166 | xfs_iext_remove( | |
6ef35544 | 3167 | xfs_inode_t *ip, /* incore inode pointer */ |
4eea22f0 | 3168 | xfs_extnum_t idx, /* index to begin removing exts */ |
6ef35544 CH |
3169 | int ext_diff, /* number of extents to remove */ |
3170 | int state) /* type of extent conversion */ | |
4eea22f0 | 3171 | { |
6ef35544 | 3172 | xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df; |
4eea22f0 MK |
3173 | xfs_extnum_t nextents; /* number of extents in file */ |
3174 | int new_size; /* size of extents after removal */ | |
3175 | ||
0b1b213f CH |
3176 | trace_xfs_iext_remove(ip, idx, state, _RET_IP_); |
3177 | ||
4eea22f0 MK |
3178 | ASSERT(ext_diff > 0); |
3179 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3180 | new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); | |
3181 | ||
3182 | if (new_size == 0) { | |
3183 | xfs_iext_destroy(ifp); | |
0293ce3a MK |
3184 | } else if (ifp->if_flags & XFS_IFEXTIREC) { |
3185 | xfs_iext_remove_indirect(ifp, idx, ext_diff); | |
4eea22f0 MK |
3186 | } else if (ifp->if_real_bytes) { |
3187 | xfs_iext_remove_direct(ifp, idx, ext_diff); | |
3188 | } else { | |
3189 | xfs_iext_remove_inline(ifp, idx, ext_diff); | |
3190 | } | |
3191 | ifp->if_bytes = new_size; | |
3192 | } | |
3193 | ||
3194 | /* | |
3195 | * This removes ext_diff extents from the inline buffer, beginning | |
3196 | * at extent index idx. | |
3197 | */ | |
3198 | void | |
3199 | xfs_iext_remove_inline( | |
3200 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3201 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3202 | int ext_diff) /* number of extents to remove */ | |
3203 | { | |
3204 | int nextents; /* number of extents in file */ | |
3205 | ||
0293ce3a | 3206 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
3207 | ASSERT(idx < XFS_INLINE_EXTS); |
3208 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3209 | ASSERT(((nextents - ext_diff) > 0) && | |
3210 | (nextents - ext_diff) < XFS_INLINE_EXTS); | |
3211 | ||
3212 | if (idx + ext_diff < nextents) { | |
3213 | memmove(&ifp->if_u2.if_inline_ext[idx], | |
3214 | &ifp->if_u2.if_inline_ext[idx + ext_diff], | |
3215 | (nextents - (idx + ext_diff)) * | |
3216 | sizeof(xfs_bmbt_rec_t)); | |
3217 | memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], | |
3218 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3219 | } else { | |
3220 | memset(&ifp->if_u2.if_inline_ext[idx], 0, | |
3221 | ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3222 | } | |
3223 | } | |
3224 | ||
3225 | /* | |
3226 | * This removes ext_diff extents from a linear (direct) extent list, | |
3227 | * beginning at extent index idx. If the extents are being removed | |
3228 | * from the end of the list (ie. truncate) then we just need to re- | |
3229 | * allocate the list to remove the extra space. Otherwise, if the | |
3230 | * extents are being removed from the middle of the existing extent | |
3231 | * entries, then we first need to move the extent records beginning | |
3232 | * at idx + ext_diff up in the list to overwrite the records being | |
3233 | * removed, then remove the extra space via kmem_realloc. | |
3234 | */ | |
3235 | void | |
3236 | xfs_iext_remove_direct( | |
3237 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3238 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3239 | int ext_diff) /* number of extents to remove */ | |
3240 | { | |
3241 | xfs_extnum_t nextents; /* number of extents in file */ | |
3242 | int new_size; /* size of extents after removal */ | |
3243 | ||
0293ce3a | 3244 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
3245 | new_size = ifp->if_bytes - |
3246 | (ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3247 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3248 | ||
3249 | if (new_size == 0) { | |
3250 | xfs_iext_destroy(ifp); | |
3251 | return; | |
3252 | } | |
3253 | /* Move extents up in the list (if needed) */ | |
3254 | if (idx + ext_diff < nextents) { | |
3255 | memmove(&ifp->if_u1.if_extents[idx], | |
3256 | &ifp->if_u1.if_extents[idx + ext_diff], | |
3257 | (nextents - (idx + ext_diff)) * | |
3258 | sizeof(xfs_bmbt_rec_t)); | |
3259 | } | |
3260 | memset(&ifp->if_u1.if_extents[nextents - ext_diff], | |
3261 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3262 | /* | |
3263 | * Reallocate the direct extent list. If the extents | |
3264 | * will fit inside the inode then xfs_iext_realloc_direct | |
3265 | * will switch from direct to inline extent allocation | |
3266 | * mode for us. | |
3267 | */ | |
3268 | xfs_iext_realloc_direct(ifp, new_size); | |
3269 | ifp->if_bytes = new_size; | |
3270 | } | |
3271 | ||
0293ce3a MK |
3272 | /* |
3273 | * This is called when incore extents are being removed from the | |
3274 | * indirection array and the extents being removed span multiple extent | |
3275 | * buffers. The idx parameter contains the file extent index where we | |
3276 | * want to begin removing extents, and the count parameter contains | |
3277 | * how many extents need to be removed. | |
3278 | * | |
3279 | * |-------| |-------| | |
3280 | * | nex1 | | | nex1 - number of extents before idx | |
3281 | * |-------| | count | | |
3282 | * | | | | count - number of extents being removed at idx | |
3283 | * | count | |-------| | |
3284 | * | | | nex2 | nex2 - number of extents after idx + count | |
3285 | * |-------| |-------| | |
3286 | */ | |
3287 | void | |
3288 | xfs_iext_remove_indirect( | |
3289 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3290 | xfs_extnum_t idx, /* index to begin removing extents */ | |
3291 | int count) /* number of extents to remove */ | |
3292 | { | |
3293 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3294 | int erp_idx = 0; /* indirection array index */ | |
3295 | xfs_extnum_t ext_cnt; /* extents left to remove */ | |
3296 | xfs_extnum_t ext_diff; /* extents to remove in current list */ | |
3297 | xfs_extnum_t nex1; /* number of extents before idx */ | |
3298 | xfs_extnum_t nex2; /* extents after idx + count */ | |
0293ce3a MK |
3299 | int page_idx = idx; /* index in target extent list */ |
3300 | ||
3301 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3302 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
3303 | ASSERT(erp != NULL); | |
0293ce3a MK |
3304 | nex1 = page_idx; |
3305 | ext_cnt = count; | |
3306 | while (ext_cnt) { | |
3307 | nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); | |
3308 | ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); | |
3309 | /* | |
3310 | * Check for deletion of entire list; | |
3311 | * xfs_iext_irec_remove() updates extent offsets. | |
3312 | */ | |
3313 | if (ext_diff == erp->er_extcount) { | |
3314 | xfs_iext_irec_remove(ifp, erp_idx); | |
3315 | ext_cnt -= ext_diff; | |
3316 | nex1 = 0; | |
3317 | if (ext_cnt) { | |
3318 | ASSERT(erp_idx < ifp->if_real_bytes / | |
3319 | XFS_IEXT_BUFSZ); | |
3320 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3321 | nex1 = 0; | |
3322 | continue; | |
3323 | } else { | |
3324 | break; | |
3325 | } | |
3326 | } | |
3327 | /* Move extents up (if needed) */ | |
3328 | if (nex2) { | |
3329 | memmove(&erp->er_extbuf[nex1], | |
3330 | &erp->er_extbuf[nex1 + ext_diff], | |
3331 | nex2 * sizeof(xfs_bmbt_rec_t)); | |
3332 | } | |
3333 | /* Zero out rest of page */ | |
3334 | memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - | |
3335 | ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); | |
3336 | /* Update remaining counters */ | |
3337 | erp->er_extcount -= ext_diff; | |
3338 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); | |
3339 | ext_cnt -= ext_diff; | |
3340 | nex1 = 0; | |
3341 | erp_idx++; | |
3342 | erp++; | |
3343 | } | |
3344 | ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); | |
3345 | xfs_iext_irec_compact(ifp); | |
3346 | } | |
3347 | ||
4eea22f0 MK |
3348 | /* |
3349 | * Create, destroy, or resize a linear (direct) block of extents. | |
3350 | */ | |
3351 | void | |
3352 | xfs_iext_realloc_direct( | |
3353 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3354 | int new_size) /* new size of extents */ | |
3355 | { | |
3356 | int rnew_size; /* real new size of extents */ | |
3357 | ||
3358 | rnew_size = new_size; | |
3359 | ||
0293ce3a MK |
3360 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || |
3361 | ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && | |
3362 | (new_size != ifp->if_real_bytes))); | |
3363 | ||
4eea22f0 MK |
3364 | /* Free extent records */ |
3365 | if (new_size == 0) { | |
3366 | xfs_iext_destroy(ifp); | |
3367 | } | |
3368 | /* Resize direct extent list and zero any new bytes */ | |
3369 | else if (ifp->if_real_bytes) { | |
3370 | /* Check if extents will fit inside the inode */ | |
3371 | if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { | |
3372 | xfs_iext_direct_to_inline(ifp, new_size / | |
3373 | (uint)sizeof(xfs_bmbt_rec_t)); | |
3374 | ifp->if_bytes = new_size; | |
3375 | return; | |
3376 | } | |
16a087d8 | 3377 | if (!is_power_of_2(new_size)){ |
40ebd81d | 3378 | rnew_size = roundup_pow_of_two(new_size); |
4eea22f0 MK |
3379 | } |
3380 | if (rnew_size != ifp->if_real_bytes) { | |
a6f64d4a | 3381 | ifp->if_u1.if_extents = |
4eea22f0 MK |
3382 | kmem_realloc(ifp->if_u1.if_extents, |
3383 | rnew_size, | |
6785073b | 3384 | ifp->if_real_bytes, KM_NOFS); |
4eea22f0 MK |
3385 | } |
3386 | if (rnew_size > ifp->if_real_bytes) { | |
3387 | memset(&ifp->if_u1.if_extents[ifp->if_bytes / | |
3388 | (uint)sizeof(xfs_bmbt_rec_t)], 0, | |
3389 | rnew_size - ifp->if_real_bytes); | |
3390 | } | |
3391 | } | |
3392 | /* | |
3393 | * Switch from the inline extent buffer to a direct | |
3394 | * extent list. Be sure to include the inline extent | |
3395 | * bytes in new_size. | |
3396 | */ | |
3397 | else { | |
3398 | new_size += ifp->if_bytes; | |
16a087d8 | 3399 | if (!is_power_of_2(new_size)) { |
40ebd81d | 3400 | rnew_size = roundup_pow_of_two(new_size); |
4eea22f0 MK |
3401 | } |
3402 | xfs_iext_inline_to_direct(ifp, rnew_size); | |
3403 | } | |
3404 | ifp->if_real_bytes = rnew_size; | |
3405 | ifp->if_bytes = new_size; | |
3406 | } | |
3407 | ||
3408 | /* | |
3409 | * Switch from linear (direct) extent records to inline buffer. | |
3410 | */ | |
3411 | void | |
3412 | xfs_iext_direct_to_inline( | |
3413 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3414 | xfs_extnum_t nextents) /* number of extents in file */ | |
3415 | { | |
3416 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
3417 | ASSERT(nextents <= XFS_INLINE_EXTS); | |
3418 | /* | |
3419 | * The inline buffer was zeroed when we switched | |
3420 | * from inline to direct extent allocation mode, | |
3421 | * so we don't need to clear it here. | |
3422 | */ | |
3423 | memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, | |
3424 | nextents * sizeof(xfs_bmbt_rec_t)); | |
f0e2d93c | 3425 | kmem_free(ifp->if_u1.if_extents); |
4eea22f0 MK |
3426 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; |
3427 | ifp->if_real_bytes = 0; | |
3428 | } | |
3429 | ||
3430 | /* | |
3431 | * Switch from inline buffer to linear (direct) extent records. | |
3432 | * new_size should already be rounded up to the next power of 2 | |
3433 | * by the caller (when appropriate), so use new_size as it is. | |
3434 | * However, since new_size may be rounded up, we can't update | |
3435 | * if_bytes here. It is the caller's responsibility to update | |
3436 | * if_bytes upon return. | |
3437 | */ | |
3438 | void | |
3439 | xfs_iext_inline_to_direct( | |
3440 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3441 | int new_size) /* number of extents in file */ | |
3442 | { | |
6785073b | 3443 | ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS); |
4eea22f0 MK |
3444 | memset(ifp->if_u1.if_extents, 0, new_size); |
3445 | if (ifp->if_bytes) { | |
3446 | memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, | |
3447 | ifp->if_bytes); | |
3448 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
3449 | sizeof(xfs_bmbt_rec_t)); | |
3450 | } | |
3451 | ifp->if_real_bytes = new_size; | |
3452 | } | |
3453 | ||
0293ce3a MK |
3454 | /* |
3455 | * Resize an extent indirection array to new_size bytes. | |
3456 | */ | |
d96f8f89 | 3457 | STATIC void |
0293ce3a MK |
3458 | xfs_iext_realloc_indirect( |
3459 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3460 | int new_size) /* new indirection array size */ | |
3461 | { | |
3462 | int nlists; /* number of irec's (ex lists) */ | |
3463 | int size; /* current indirection array size */ | |
3464 | ||
3465 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3466 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3467 | size = nlists * sizeof(xfs_ext_irec_t); | |
3468 | ASSERT(ifp->if_real_bytes); | |
3469 | ASSERT((new_size >= 0) && (new_size != size)); | |
3470 | if (new_size == 0) { | |
3471 | xfs_iext_destroy(ifp); | |
3472 | } else { | |
3473 | ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) | |
3474 | kmem_realloc(ifp->if_u1.if_ext_irec, | |
6785073b | 3475 | new_size, size, KM_NOFS); |
0293ce3a MK |
3476 | } |
3477 | } | |
3478 | ||
3479 | /* | |
3480 | * Switch from indirection array to linear (direct) extent allocations. | |
3481 | */ | |
d96f8f89 | 3482 | STATIC void |
0293ce3a MK |
3483 | xfs_iext_indirect_to_direct( |
3484 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3485 | { | |
a6f64d4a | 3486 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ |
0293ce3a MK |
3487 | xfs_extnum_t nextents; /* number of extents in file */ |
3488 | int size; /* size of file extents */ | |
3489 | ||
3490 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3491 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3492 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
3493 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
3494 | ||
71a8c87f | 3495 | xfs_iext_irec_compact_pages(ifp); |
0293ce3a MK |
3496 | ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); |
3497 | ||
3498 | ep = ifp->if_u1.if_ext_irec->er_extbuf; | |
f0e2d93c | 3499 | kmem_free(ifp->if_u1.if_ext_irec); |
0293ce3a MK |
3500 | ifp->if_flags &= ~XFS_IFEXTIREC; |
3501 | ifp->if_u1.if_extents = ep; | |
3502 | ifp->if_bytes = size; | |
3503 | if (nextents < XFS_LINEAR_EXTS) { | |
3504 | xfs_iext_realloc_direct(ifp, size); | |
3505 | } | |
3506 | } | |
3507 | ||
4eea22f0 MK |
3508 | /* |
3509 | * Free incore file extents. | |
3510 | */ | |
3511 | void | |
3512 | xfs_iext_destroy( | |
3513 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3514 | { | |
0293ce3a MK |
3515 | if (ifp->if_flags & XFS_IFEXTIREC) { |
3516 | int erp_idx; | |
3517 | int nlists; | |
3518 | ||
3519 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3520 | for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { | |
3521 | xfs_iext_irec_remove(ifp, erp_idx); | |
3522 | } | |
3523 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
3524 | } else if (ifp->if_real_bytes) { | |
f0e2d93c | 3525 | kmem_free(ifp->if_u1.if_extents); |
4eea22f0 MK |
3526 | } else if (ifp->if_bytes) { |
3527 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
3528 | sizeof(xfs_bmbt_rec_t)); | |
3529 | } | |
3530 | ifp->if_u1.if_extents = NULL; | |
3531 | ifp->if_real_bytes = 0; | |
3532 | ifp->if_bytes = 0; | |
3533 | } | |
0293ce3a | 3534 | |
8867bc9b MK |
3535 | /* |
3536 | * Return a pointer to the extent record for file system block bno. | |
3537 | */ | |
a6f64d4a | 3538 | xfs_bmbt_rec_host_t * /* pointer to found extent record */ |
8867bc9b MK |
3539 | xfs_iext_bno_to_ext( |
3540 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3541 | xfs_fileoff_t bno, /* block number to search for */ | |
3542 | xfs_extnum_t *idxp) /* index of target extent */ | |
3543 | { | |
a6f64d4a | 3544 | xfs_bmbt_rec_host_t *base; /* pointer to first extent */ |
8867bc9b | 3545 | xfs_filblks_t blockcount = 0; /* number of blocks in extent */ |
a6f64d4a | 3546 | xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */ |
8867bc9b | 3547 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ |
c41564b5 | 3548 | int high; /* upper boundary in search */ |
8867bc9b | 3549 | xfs_extnum_t idx = 0; /* index of target extent */ |
c41564b5 | 3550 | int low; /* lower boundary in search */ |
8867bc9b MK |
3551 | xfs_extnum_t nextents; /* number of file extents */ |
3552 | xfs_fileoff_t startoff = 0; /* start offset of extent */ | |
3553 | ||
3554 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3555 | if (nextents == 0) { | |
3556 | *idxp = 0; | |
3557 | return NULL; | |
3558 | } | |
3559 | low = 0; | |
3560 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3561 | /* Find target extent list */ | |
3562 | int erp_idx = 0; | |
3563 | erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); | |
3564 | base = erp->er_extbuf; | |
3565 | high = erp->er_extcount - 1; | |
3566 | } else { | |
3567 | base = ifp->if_u1.if_extents; | |
3568 | high = nextents - 1; | |
3569 | } | |
3570 | /* Binary search extent records */ | |
3571 | while (low <= high) { | |
3572 | idx = (low + high) >> 1; | |
3573 | ep = base + idx; | |
3574 | startoff = xfs_bmbt_get_startoff(ep); | |
3575 | blockcount = xfs_bmbt_get_blockcount(ep); | |
3576 | if (bno < startoff) { | |
3577 | high = idx - 1; | |
3578 | } else if (bno >= startoff + blockcount) { | |
3579 | low = idx + 1; | |
3580 | } else { | |
3581 | /* Convert back to file-based extent index */ | |
3582 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3583 | idx += erp->er_extoff; | |
3584 | } | |
3585 | *idxp = idx; | |
3586 | return ep; | |
3587 | } | |
3588 | } | |
3589 | /* Convert back to file-based extent index */ | |
3590 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3591 | idx += erp->er_extoff; | |
3592 | } | |
3593 | if (bno >= startoff + blockcount) { | |
3594 | if (++idx == nextents) { | |
3595 | ep = NULL; | |
3596 | } else { | |
3597 | ep = xfs_iext_get_ext(ifp, idx); | |
3598 | } | |
3599 | } | |
3600 | *idxp = idx; | |
3601 | return ep; | |
3602 | } | |
3603 | ||
0293ce3a MK |
3604 | /* |
3605 | * Return a pointer to the indirection array entry containing the | |
3606 | * extent record for filesystem block bno. Store the index of the | |
3607 | * target irec in *erp_idxp. | |
3608 | */ | |
8867bc9b | 3609 | xfs_ext_irec_t * /* pointer to found extent record */ |
0293ce3a MK |
3610 | xfs_iext_bno_to_irec( |
3611 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3612 | xfs_fileoff_t bno, /* block number to search for */ | |
3613 | int *erp_idxp) /* irec index of target ext list */ | |
3614 | { | |
3615 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ | |
3616 | xfs_ext_irec_t *erp_next; /* next indirection array entry */ | |
8867bc9b | 3617 | int erp_idx; /* indirection array index */ |
0293ce3a MK |
3618 | int nlists; /* number of extent irec's (lists) */ |
3619 | int high; /* binary search upper limit */ | |
3620 | int low; /* binary search lower limit */ | |
3621 | ||
3622 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3623 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3624 | erp_idx = 0; | |
3625 | low = 0; | |
3626 | high = nlists - 1; | |
3627 | while (low <= high) { | |
3628 | erp_idx = (low + high) >> 1; | |
3629 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3630 | erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; | |
3631 | if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { | |
3632 | high = erp_idx - 1; | |
3633 | } else if (erp_next && bno >= | |
3634 | xfs_bmbt_get_startoff(erp_next->er_extbuf)) { | |
3635 | low = erp_idx + 1; | |
3636 | } else { | |
3637 | break; | |
3638 | } | |
3639 | } | |
3640 | *erp_idxp = erp_idx; | |
3641 | return erp; | |
3642 | } | |
3643 | ||
3644 | /* | |
3645 | * Return a pointer to the indirection array entry containing the | |
3646 | * extent record at file extent index *idxp. Store the index of the | |
3647 | * target irec in *erp_idxp and store the page index of the target | |
3648 | * extent record in *idxp. | |
3649 | */ | |
3650 | xfs_ext_irec_t * | |
3651 | xfs_iext_idx_to_irec( | |
3652 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3653 | xfs_extnum_t *idxp, /* extent index (file -> page) */ | |
3654 | int *erp_idxp, /* pointer to target irec */ | |
3655 | int realloc) /* new bytes were just added */ | |
3656 | { | |
3657 | xfs_ext_irec_t *prev; /* pointer to previous irec */ | |
3658 | xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ | |
3659 | int erp_idx; /* indirection array index */ | |
3660 | int nlists; /* number of irec's (ex lists) */ | |
3661 | int high; /* binary search upper limit */ | |
3662 | int low; /* binary search lower limit */ | |
3663 | xfs_extnum_t page_idx = *idxp; /* extent index in target list */ | |
3664 | ||
3665 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
87bef181 CH |
3666 | ASSERT(page_idx >= 0); |
3667 | ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t)); | |
3668 | ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc); | |
3669 | ||
0293ce3a MK |
3670 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
3671 | erp_idx = 0; | |
3672 | low = 0; | |
3673 | high = nlists - 1; | |
3674 | ||
3675 | /* Binary search extent irec's */ | |
3676 | while (low <= high) { | |
3677 | erp_idx = (low + high) >> 1; | |
3678 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3679 | prev = erp_idx > 0 ? erp - 1 : NULL; | |
3680 | if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && | |
3681 | realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { | |
3682 | high = erp_idx - 1; | |
3683 | } else if (page_idx > erp->er_extoff + erp->er_extcount || | |
3684 | (page_idx == erp->er_extoff + erp->er_extcount && | |
3685 | !realloc)) { | |
3686 | low = erp_idx + 1; | |
3687 | } else if (page_idx == erp->er_extoff + erp->er_extcount && | |
3688 | erp->er_extcount == XFS_LINEAR_EXTS) { | |
3689 | ASSERT(realloc); | |
3690 | page_idx = 0; | |
3691 | erp_idx++; | |
3692 | erp = erp_idx < nlists ? erp + 1 : NULL; | |
3693 | break; | |
3694 | } else { | |
3695 | page_idx -= erp->er_extoff; | |
3696 | break; | |
3697 | } | |
3698 | } | |
3699 | *idxp = page_idx; | |
3700 | *erp_idxp = erp_idx; | |
3701 | return(erp); | |
3702 | } | |
3703 | ||
3704 | /* | |
3705 | * Allocate and initialize an indirection array once the space needed | |
3706 | * for incore extents increases above XFS_IEXT_BUFSZ. | |
3707 | */ | |
3708 | void | |
3709 | xfs_iext_irec_init( | |
3710 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3711 | { | |
3712 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3713 | xfs_extnum_t nextents; /* number of extents in file */ | |
3714 | ||
3715 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); | |
3716 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3717 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
3718 | ||
6785073b | 3719 | erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS); |
0293ce3a MK |
3720 | |
3721 | if (nextents == 0) { | |
6785073b | 3722 | ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); |
0293ce3a MK |
3723 | } else if (!ifp->if_real_bytes) { |
3724 | xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); | |
3725 | } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { | |
3726 | xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); | |
3727 | } | |
3728 | erp->er_extbuf = ifp->if_u1.if_extents; | |
3729 | erp->er_extcount = nextents; | |
3730 | erp->er_extoff = 0; | |
3731 | ||
3732 | ifp->if_flags |= XFS_IFEXTIREC; | |
3733 | ifp->if_real_bytes = XFS_IEXT_BUFSZ; | |
3734 | ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); | |
3735 | ifp->if_u1.if_ext_irec = erp; | |
3736 | ||
3737 | return; | |
3738 | } | |
3739 | ||
3740 | /* | |
3741 | * Allocate and initialize a new entry in the indirection array. | |
3742 | */ | |
3743 | xfs_ext_irec_t * | |
3744 | xfs_iext_irec_new( | |
3745 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3746 | int erp_idx) /* index for new irec */ | |
3747 | { | |
3748 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3749 | int i; /* loop counter */ | |
3750 | int nlists; /* number of irec's (ex lists) */ | |
3751 | ||
3752 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3753 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3754 | ||
3755 | /* Resize indirection array */ | |
3756 | xfs_iext_realloc_indirect(ifp, ++nlists * | |
3757 | sizeof(xfs_ext_irec_t)); | |
3758 | /* | |
3759 | * Move records down in the array so the | |
3760 | * new page can use erp_idx. | |
3761 | */ | |
3762 | erp = ifp->if_u1.if_ext_irec; | |
3763 | for (i = nlists - 1; i > erp_idx; i--) { | |
3764 | memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); | |
3765 | } | |
3766 | ASSERT(i == erp_idx); | |
3767 | ||
3768 | /* Initialize new extent record */ | |
3769 | erp = ifp->if_u1.if_ext_irec; | |
6785073b | 3770 | erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); |
0293ce3a MK |
3771 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; |
3772 | memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); | |
3773 | erp[erp_idx].er_extcount = 0; | |
3774 | erp[erp_idx].er_extoff = erp_idx > 0 ? | |
3775 | erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; | |
3776 | return (&erp[erp_idx]); | |
3777 | } | |
3778 | ||
3779 | /* | |
3780 | * Remove a record from the indirection array. | |
3781 | */ | |
3782 | void | |
3783 | xfs_iext_irec_remove( | |
3784 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3785 | int erp_idx) /* irec index to remove */ | |
3786 | { | |
3787 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3788 | int i; /* loop counter */ | |
3789 | int nlists; /* number of irec's (ex lists) */ | |
3790 | ||
3791 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3792 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3793 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3794 | if (erp->er_extbuf) { | |
3795 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, | |
3796 | -erp->er_extcount); | |
f0e2d93c | 3797 | kmem_free(erp->er_extbuf); |
0293ce3a MK |
3798 | } |
3799 | /* Compact extent records */ | |
3800 | erp = ifp->if_u1.if_ext_irec; | |
3801 | for (i = erp_idx; i < nlists - 1; i++) { | |
3802 | memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); | |
3803 | } | |
3804 | /* | |
3805 | * Manually free the last extent record from the indirection | |
3806 | * array. A call to xfs_iext_realloc_indirect() with a size | |
3807 | * of zero would result in a call to xfs_iext_destroy() which | |
3808 | * would in turn call this function again, creating a nasty | |
3809 | * infinite loop. | |
3810 | */ | |
3811 | if (--nlists) { | |
3812 | xfs_iext_realloc_indirect(ifp, | |
3813 | nlists * sizeof(xfs_ext_irec_t)); | |
3814 | } else { | |
f0e2d93c | 3815 | kmem_free(ifp->if_u1.if_ext_irec); |
0293ce3a MK |
3816 | } |
3817 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; | |
3818 | } | |
3819 | ||
3820 | /* | |
3821 | * This is called to clean up large amounts of unused memory allocated | |
3822 | * by the indirection array. Before compacting anything though, verify | |
3823 | * that the indirection array is still needed and switch back to the | |
3824 | * linear extent list (or even the inline buffer) if possible. The | |
3825 | * compaction policy is as follows: | |
3826 | * | |
3827 | * Full Compaction: Extents fit into a single page (or inline buffer) | |
71a8c87f | 3828 | * Partial Compaction: Extents occupy less than 50% of allocated space |
0293ce3a MK |
3829 | * No Compaction: Extents occupy at least 50% of allocated space |
3830 | */ | |
3831 | void | |
3832 | xfs_iext_irec_compact( | |
3833 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3834 | { | |
3835 | xfs_extnum_t nextents; /* number of extents in file */ | |
3836 | int nlists; /* number of irec's (ex lists) */ | |
3837 | ||
3838 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3839 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3840 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3841 | ||
3842 | if (nextents == 0) { | |
3843 | xfs_iext_destroy(ifp); | |
3844 | } else if (nextents <= XFS_INLINE_EXTS) { | |
3845 | xfs_iext_indirect_to_direct(ifp); | |
3846 | xfs_iext_direct_to_inline(ifp, nextents); | |
3847 | } else if (nextents <= XFS_LINEAR_EXTS) { | |
3848 | xfs_iext_indirect_to_direct(ifp); | |
0293ce3a MK |
3849 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { |
3850 | xfs_iext_irec_compact_pages(ifp); | |
3851 | } | |
3852 | } | |
3853 | ||
3854 | /* | |
3855 | * Combine extents from neighboring extent pages. | |
3856 | */ | |
3857 | void | |
3858 | xfs_iext_irec_compact_pages( | |
3859 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3860 | { | |
3861 | xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ | |
3862 | int erp_idx = 0; /* indirection array index */ | |
3863 | int nlists; /* number of irec's (ex lists) */ | |
3864 | ||
3865 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3866 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3867 | while (erp_idx < nlists - 1) { | |
3868 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3869 | erp_next = erp + 1; | |
3870 | if (erp_next->er_extcount <= | |
3871 | (XFS_LINEAR_EXTS - erp->er_extcount)) { | |
71a8c87f | 3872 | memcpy(&erp->er_extbuf[erp->er_extcount], |
0293ce3a MK |
3873 | erp_next->er_extbuf, erp_next->er_extcount * |
3874 | sizeof(xfs_bmbt_rec_t)); | |
3875 | erp->er_extcount += erp_next->er_extcount; | |
3876 | /* | |
3877 | * Free page before removing extent record | |
3878 | * so er_extoffs don't get modified in | |
3879 | * xfs_iext_irec_remove. | |
3880 | */ | |
f0e2d93c | 3881 | kmem_free(erp_next->er_extbuf); |
0293ce3a MK |
3882 | erp_next->er_extbuf = NULL; |
3883 | xfs_iext_irec_remove(ifp, erp_idx + 1); | |
3884 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3885 | } else { | |
3886 | erp_idx++; | |
3887 | } | |
3888 | } | |
3889 | } | |
3890 | ||
0293ce3a MK |
3891 | /* |
3892 | * This is called to update the er_extoff field in the indirection | |
3893 | * array when extents have been added or removed from one of the | |
3894 | * extent lists. erp_idx contains the irec index to begin updating | |
3895 | * at and ext_diff contains the number of extents that were added | |
3896 | * or removed. | |
3897 | */ | |
3898 | void | |
3899 | xfs_iext_irec_update_extoffs( | |
3900 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3901 | int erp_idx, /* irec index to update */ | |
3902 | int ext_diff) /* number of new extents */ | |
3903 | { | |
3904 | int i; /* loop counter */ | |
3905 | int nlists; /* number of irec's (ex lists */ | |
3906 | ||
3907 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3908 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3909 | for (i = erp_idx; i < nlists; i++) { | |
3910 | ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; | |
3911 | } | |
3912 | } |