]>
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" |
33479e05 | 48 | #include "xfs_icache.h" |
1da177e4 | 49 | |
1da177e4 LT |
50 | kmem_zone_t *xfs_ifork_zone; |
51 | kmem_zone_t *xfs_inode_zone; | |
1da177e4 LT |
52 | |
53 | /* | |
8f04c47a | 54 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
55 | * freed from a file in a single transaction. |
56 | */ | |
57 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
58 | ||
59 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); | |
60 | STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); | |
61 | STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); | |
62 | STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); | |
63 | ||
2a0ec1d9 DC |
64 | /* |
65 | * helper function to extract extent size hint from inode | |
66 | */ | |
67 | xfs_extlen_t | |
68 | xfs_get_extsz_hint( | |
69 | struct xfs_inode *ip) | |
70 | { | |
71 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
72 | return ip->i_d.di_extsize; | |
73 | if (XFS_IS_REALTIME_INODE(ip)) | |
74 | return ip->i_mount->m_sb.sb_rextsize; | |
75 | return 0; | |
76 | } | |
77 | ||
fa96acad DC |
78 | /* |
79 | * This is a wrapper routine around the xfs_ilock() routine used to centralize | |
80 | * some grungy code. It is used in places that wish to lock the inode solely | |
81 | * for reading the extents. The reason these places can't just call | |
82 | * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the | |
83 | * extents from disk for a file in b-tree format. If the inode is in b-tree | |
84 | * format, then we need to lock the inode exclusively until the extents are read | |
85 | * in. Locking it exclusively all the time would limit our parallelism | |
86 | * unnecessarily, though. What we do instead is check to see if the extents | |
87 | * have been read in yet, and only lock the inode exclusively if they have not. | |
88 | * | |
89 | * The function returns a value which should be given to the corresponding | |
90 | * xfs_iunlock_map_shared(). This value is the mode in which the lock was | |
91 | * actually taken. | |
92 | */ | |
93 | uint | |
94 | xfs_ilock_map_shared( | |
95 | xfs_inode_t *ip) | |
96 | { | |
97 | uint lock_mode; | |
98 | ||
99 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
100 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
101 | lock_mode = XFS_ILOCK_EXCL; | |
102 | } else { | |
103 | lock_mode = XFS_ILOCK_SHARED; | |
104 | } | |
105 | ||
106 | xfs_ilock(ip, lock_mode); | |
107 | ||
108 | return lock_mode; | |
109 | } | |
110 | ||
111 | /* | |
112 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
113 | * All it does is call xfs_iunlock() with the given lock_mode. | |
114 | */ | |
115 | void | |
116 | xfs_iunlock_map_shared( | |
117 | xfs_inode_t *ip, | |
118 | unsigned int lock_mode) | |
119 | { | |
120 | xfs_iunlock(ip, lock_mode); | |
121 | } | |
122 | ||
123 | /* | |
124 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
125 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
126 | * allows either or both of the locks to be obtained. | |
127 | * | |
128 | * The 2 locks should always be ordered so that the IO lock is | |
129 | * obtained first in order to prevent deadlock. | |
130 | * | |
131 | * ip -- the inode being locked | |
132 | * lock_flags -- this parameter indicates the inode's locks | |
133 | * to be locked. It can be: | |
134 | * XFS_IOLOCK_SHARED, | |
135 | * XFS_IOLOCK_EXCL, | |
136 | * XFS_ILOCK_SHARED, | |
137 | * XFS_ILOCK_EXCL, | |
138 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
139 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
140 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
141 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
142 | */ | |
143 | void | |
144 | xfs_ilock( | |
145 | xfs_inode_t *ip, | |
146 | uint lock_flags) | |
147 | { | |
148 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
149 | ||
150 | /* | |
151 | * You can't set both SHARED and EXCL for the same lock, | |
152 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
153 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
154 | */ | |
155 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
156 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
157 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
158 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
159 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
160 | ||
161 | if (lock_flags & XFS_IOLOCK_EXCL) | |
162 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
163 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
164 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
165 | ||
166 | if (lock_flags & XFS_ILOCK_EXCL) | |
167 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
168 | else if (lock_flags & XFS_ILOCK_SHARED) | |
169 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
170 | } | |
171 | ||
172 | /* | |
173 | * This is just like xfs_ilock(), except that the caller | |
174 | * is guaranteed not to sleep. It returns 1 if it gets | |
175 | * the requested locks and 0 otherwise. If the IO lock is | |
176 | * obtained but the inode lock cannot be, then the IO lock | |
177 | * is dropped before returning. | |
178 | * | |
179 | * ip -- the inode being locked | |
180 | * lock_flags -- this parameter indicates the inode's locks to be | |
181 | * to be locked. See the comment for xfs_ilock() for a list | |
182 | * of valid values. | |
183 | */ | |
184 | int | |
185 | xfs_ilock_nowait( | |
186 | xfs_inode_t *ip, | |
187 | uint lock_flags) | |
188 | { | |
189 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
190 | ||
191 | /* | |
192 | * You can't set both SHARED and EXCL for the same lock, | |
193 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
194 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
195 | */ | |
196 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
197 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
198 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
199 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
200 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
201 | ||
202 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
203 | if (!mrtryupdate(&ip->i_iolock)) | |
204 | goto out; | |
205 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
206 | if (!mrtryaccess(&ip->i_iolock)) | |
207 | goto out; | |
208 | } | |
209 | if (lock_flags & XFS_ILOCK_EXCL) { | |
210 | if (!mrtryupdate(&ip->i_lock)) | |
211 | goto out_undo_iolock; | |
212 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
213 | if (!mrtryaccess(&ip->i_lock)) | |
214 | goto out_undo_iolock; | |
215 | } | |
216 | return 1; | |
217 | ||
218 | out_undo_iolock: | |
219 | if (lock_flags & XFS_IOLOCK_EXCL) | |
220 | mrunlock_excl(&ip->i_iolock); | |
221 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
222 | mrunlock_shared(&ip->i_iolock); | |
223 | out: | |
224 | return 0; | |
225 | } | |
226 | ||
227 | /* | |
228 | * xfs_iunlock() is used to drop the inode locks acquired with | |
229 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
230 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
231 | * that we know which locks to drop. | |
232 | * | |
233 | * ip -- the inode being unlocked | |
234 | * lock_flags -- this parameter indicates the inode's locks to be | |
235 | * to be unlocked. See the comment for xfs_ilock() for a list | |
236 | * of valid values for this parameter. | |
237 | * | |
238 | */ | |
239 | void | |
240 | xfs_iunlock( | |
241 | xfs_inode_t *ip, | |
242 | uint lock_flags) | |
243 | { | |
244 | /* | |
245 | * You can't set both SHARED and EXCL for the same lock, | |
246 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
247 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
248 | */ | |
249 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
250 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
251 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
252 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
253 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
254 | ASSERT(lock_flags != 0); | |
255 | ||
256 | if (lock_flags & XFS_IOLOCK_EXCL) | |
257 | mrunlock_excl(&ip->i_iolock); | |
258 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
259 | mrunlock_shared(&ip->i_iolock); | |
260 | ||
261 | if (lock_flags & XFS_ILOCK_EXCL) | |
262 | mrunlock_excl(&ip->i_lock); | |
263 | else if (lock_flags & XFS_ILOCK_SHARED) | |
264 | mrunlock_shared(&ip->i_lock); | |
265 | ||
266 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
267 | } | |
268 | ||
269 | /* | |
270 | * give up write locks. the i/o lock cannot be held nested | |
271 | * if it is being demoted. | |
272 | */ | |
273 | void | |
274 | xfs_ilock_demote( | |
275 | xfs_inode_t *ip, | |
276 | uint lock_flags) | |
277 | { | |
278 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
279 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
280 | ||
281 | if (lock_flags & XFS_ILOCK_EXCL) | |
282 | mrdemote(&ip->i_lock); | |
283 | if (lock_flags & XFS_IOLOCK_EXCL) | |
284 | mrdemote(&ip->i_iolock); | |
285 | ||
286 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
287 | } | |
288 | ||
289 | #ifdef DEBUG | |
290 | int | |
291 | xfs_isilocked( | |
292 | xfs_inode_t *ip, | |
293 | uint lock_flags) | |
294 | { | |
295 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
296 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
297 | return !!ip->i_lock.mr_writer; | |
298 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
299 | } | |
300 | ||
301 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { | |
302 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
303 | return !!ip->i_iolock.mr_writer; | |
304 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
305 | } | |
306 | ||
307 | ASSERT(0); | |
308 | return 0; | |
309 | } | |
310 | #endif | |
311 | ||
312 | void | |
313 | __xfs_iflock( | |
314 | struct xfs_inode *ip) | |
315 | { | |
316 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
317 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
318 | ||
319 | do { | |
320 | prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
321 | if (xfs_isiflocked(ip)) | |
322 | io_schedule(); | |
323 | } while (!xfs_iflock_nowait(ip)); | |
324 | ||
325 | finish_wait(wq, &wait.wait); | |
326 | } | |
327 | ||
1da177e4 LT |
328 | #ifdef DEBUG |
329 | /* | |
330 | * Make sure that the extents in the given memory buffer | |
331 | * are valid. | |
332 | */ | |
333 | STATIC void | |
334 | xfs_validate_extents( | |
4eea22f0 | 335 | xfs_ifork_t *ifp, |
1da177e4 | 336 | int nrecs, |
1da177e4 LT |
337 | xfs_exntfmt_t fmt) |
338 | { | |
339 | xfs_bmbt_irec_t irec; | |
a6f64d4a | 340 | xfs_bmbt_rec_host_t rec; |
1da177e4 LT |
341 | int i; |
342 | ||
343 | for (i = 0; i < nrecs; i++) { | |
a6f64d4a CH |
344 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
345 | rec.l0 = get_unaligned(&ep->l0); | |
346 | rec.l1 = get_unaligned(&ep->l1); | |
347 | xfs_bmbt_get_all(&rec, &irec); | |
1da177e4 LT |
348 | if (fmt == XFS_EXTFMT_NOSTATE) |
349 | ASSERT(irec.br_state == XFS_EXT_NORM); | |
1da177e4 LT |
350 | } |
351 | } | |
352 | #else /* DEBUG */ | |
a6f64d4a | 353 | #define xfs_validate_extents(ifp, nrecs, fmt) |
1da177e4 LT |
354 | #endif /* DEBUG */ |
355 | ||
356 | /* | |
357 | * Check that none of the inode's in the buffer have a next | |
358 | * unlinked field of 0. | |
359 | */ | |
360 | #if defined(DEBUG) | |
361 | void | |
362 | xfs_inobp_check( | |
363 | xfs_mount_t *mp, | |
364 | xfs_buf_t *bp) | |
365 | { | |
366 | int i; | |
367 | int j; | |
368 | xfs_dinode_t *dip; | |
369 | ||
370 | j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; | |
371 | ||
372 | for (i = 0; i < j; i++) { | |
373 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
374 | i * mp->m_sb.sb_inodesize); | |
375 | if (!dip->di_next_unlinked) { | |
53487786 DC |
376 | xfs_alert(mp, |
377 | "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.", | |
1da177e4 LT |
378 | bp); |
379 | ASSERT(dip->di_next_unlinked); | |
380 | } | |
381 | } | |
382 | } | |
383 | #endif | |
384 | ||
612cfbfe | 385 | static void |
af133e86 DC |
386 | xfs_inode_buf_verify( |
387 | struct xfs_buf *bp) | |
388 | { | |
389 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
390 | int i; | |
391 | int ni; | |
392 | ||
393 | /* | |
394 | * Validate the magic number and version of every inode in the buffer | |
395 | */ | |
396 | ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock; | |
397 | for (i = 0; i < ni; i++) { | |
398 | int di_ok; | |
399 | xfs_dinode_t *dip; | |
400 | ||
401 | dip = (struct xfs_dinode *)xfs_buf_offset(bp, | |
402 | (i << mp->m_sb.sb_inodelog)); | |
403 | di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) && | |
404 | XFS_DINODE_GOOD_VERSION(dip->di_version); | |
405 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, | |
406 | XFS_ERRTAG_ITOBP_INOTOBP, | |
407 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
408 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
409 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_HIGH, | |
410 | mp, dip); | |
411 | #ifdef DEBUG | |
412 | xfs_emerg(mp, | |
413 | "bad inode magic/vsn daddr %lld #%d (magic=%x)", | |
414 | (unsigned long long)bp->b_bn, i, | |
415 | be16_to_cpu(dip->di_magic)); | |
416 | ASSERT(0); | |
417 | #endif | |
418 | } | |
419 | } | |
420 | xfs_inobp_check(mp, bp); | |
612cfbfe DC |
421 | } |
422 | ||
423 | static void | |
424 | xfs_inode_buf_write_verify( | |
425 | struct xfs_buf *bp) | |
426 | { | |
427 | xfs_inode_buf_verify(bp); | |
428 | } | |
429 | ||
430 | void | |
431 | xfs_inode_buf_read_verify( | |
432 | struct xfs_buf *bp) | |
433 | { | |
434 | xfs_inode_buf_verify(bp); | |
435 | bp->b_pre_io = xfs_inode_buf_write_verify; | |
af133e86 DC |
436 | bp->b_iodone = NULL; |
437 | xfs_buf_ioend(bp, 0); | |
438 | } | |
439 | ||
4ae29b43 | 440 | /* |
475ee413 CH |
441 | * This routine is called to map an inode to the buffer containing the on-disk |
442 | * version of the inode. It returns a pointer to the buffer containing the | |
443 | * on-disk inode in the bpp parameter, and in the dipp parameter it returns a | |
444 | * pointer to the on-disk inode within that buffer. | |
445 | * | |
446 | * If a non-zero error is returned, then the contents of bpp and dipp are | |
447 | * undefined. | |
4ae29b43 | 448 | */ |
475ee413 | 449 | int |
4ae29b43 | 450 | xfs_imap_to_bp( |
475ee413 CH |
451 | struct xfs_mount *mp, |
452 | struct xfs_trans *tp, | |
453 | struct xfs_imap *imap, | |
af133e86 | 454 | struct xfs_dinode **dipp, |
475ee413 CH |
455 | struct xfs_buf **bpp, |
456 | uint buf_flags, | |
457 | uint iget_flags) | |
4ae29b43 | 458 | { |
475ee413 CH |
459 | struct xfs_buf *bp; |
460 | int error; | |
4ae29b43 | 461 | |
611c9946 | 462 | buf_flags |= XBF_UNMAPPED; |
4ae29b43 | 463 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno, |
af133e86 | 464 | (int)imap->im_len, buf_flags, &bp, |
612cfbfe | 465 | xfs_inode_buf_read_verify); |
4ae29b43 | 466 | if (error) { |
af133e86 | 467 | if (error == EAGAIN) { |
0cadda1c | 468 | ASSERT(buf_flags & XBF_TRYLOCK); |
af133e86 | 469 | return error; |
a3f74ffb | 470 | } |
4ae29b43 | 471 | |
af133e86 DC |
472 | if (error == EFSCORRUPTED && |
473 | (iget_flags & XFS_IGET_UNTRUSTED)) | |
474 | return XFS_ERROR(EINVAL); | |
4ae29b43 | 475 | |
af133e86 DC |
476 | xfs_warn(mp, "%s: xfs_trans_read_buf() returned error %d.", |
477 | __func__, error); | |
478 | return error; | |
4ae29b43 DC |
479 | } |
480 | ||
4ae29b43 | 481 | *bpp = bp; |
475ee413 | 482 | *dipp = (struct xfs_dinode *)xfs_buf_offset(bp, imap->im_boffset); |
4ae29b43 DC |
483 | return 0; |
484 | } | |
485 | ||
1da177e4 LT |
486 | /* |
487 | * Move inode type and inode format specific information from the | |
488 | * on-disk inode to the in-core inode. For fifos, devs, and sockets | |
489 | * this means set if_rdev to the proper value. For files, directories, | |
490 | * and symlinks this means to bring in the in-line data or extent | |
491 | * pointers. For a file in B-tree format, only the root is immediately | |
492 | * brought in-core. The rest will be in-lined in if_extents when it | |
493 | * is first referenced (see xfs_iread_extents()). | |
494 | */ | |
495 | STATIC int | |
496 | xfs_iformat( | |
497 | xfs_inode_t *ip, | |
498 | xfs_dinode_t *dip) | |
499 | { | |
500 | xfs_attr_shortform_t *atp; | |
501 | int size; | |
8096b1eb | 502 | int error = 0; |
1da177e4 | 503 | xfs_fsize_t di_size; |
1da177e4 | 504 | |
81591fe2 CH |
505 | if (unlikely(be32_to_cpu(dip->di_nextents) + |
506 | be16_to_cpu(dip->di_anextents) > | |
507 | be64_to_cpu(dip->di_nblocks))) { | |
65333b4c | 508 | xfs_warn(ip->i_mount, |
3762ec6b | 509 | "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", |
1da177e4 | 510 | (unsigned long long)ip->i_ino, |
81591fe2 CH |
511 | (int)(be32_to_cpu(dip->di_nextents) + |
512 | be16_to_cpu(dip->di_anextents)), | |
1da177e4 | 513 | (unsigned long long) |
81591fe2 | 514 | be64_to_cpu(dip->di_nblocks)); |
1da177e4 LT |
515 | XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, |
516 | ip->i_mount, dip); | |
517 | return XFS_ERROR(EFSCORRUPTED); | |
518 | } | |
519 | ||
81591fe2 | 520 | if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) { |
65333b4c | 521 | xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.", |
1da177e4 | 522 | (unsigned long long)ip->i_ino, |
81591fe2 | 523 | dip->di_forkoff); |
1da177e4 LT |
524 | XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, |
525 | ip->i_mount, dip); | |
526 | return XFS_ERROR(EFSCORRUPTED); | |
527 | } | |
528 | ||
b89d4208 CH |
529 | if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && |
530 | !ip->i_mount->m_rtdev_targp)) { | |
65333b4c | 531 | xfs_warn(ip->i_mount, |
b89d4208 CH |
532 | "corrupt dinode %Lu, has realtime flag set.", |
533 | ip->i_ino); | |
534 | XFS_CORRUPTION_ERROR("xfs_iformat(realtime)", | |
535 | XFS_ERRLEVEL_LOW, ip->i_mount, dip); | |
536 | return XFS_ERROR(EFSCORRUPTED); | |
537 | } | |
538 | ||
1da177e4 LT |
539 | switch (ip->i_d.di_mode & S_IFMT) { |
540 | case S_IFIFO: | |
541 | case S_IFCHR: | |
542 | case S_IFBLK: | |
543 | case S_IFSOCK: | |
81591fe2 | 544 | if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) { |
1da177e4 LT |
545 | XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, |
546 | ip->i_mount, dip); | |
547 | return XFS_ERROR(EFSCORRUPTED); | |
548 | } | |
549 | ip->i_d.di_size = 0; | |
81591fe2 | 550 | ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip); |
1da177e4 LT |
551 | break; |
552 | ||
553 | case S_IFREG: | |
554 | case S_IFLNK: | |
555 | case S_IFDIR: | |
81591fe2 | 556 | switch (dip->di_format) { |
1da177e4 LT |
557 | case XFS_DINODE_FMT_LOCAL: |
558 | /* | |
559 | * no local regular files yet | |
560 | */ | |
abbede1b | 561 | if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) { |
65333b4c DC |
562 | xfs_warn(ip->i_mount, |
563 | "corrupt inode %Lu (local format for regular file).", | |
1da177e4 LT |
564 | (unsigned long long) ip->i_ino); |
565 | XFS_CORRUPTION_ERROR("xfs_iformat(4)", | |
566 | XFS_ERRLEVEL_LOW, | |
567 | ip->i_mount, dip); | |
568 | return XFS_ERROR(EFSCORRUPTED); | |
569 | } | |
570 | ||
81591fe2 | 571 | di_size = be64_to_cpu(dip->di_size); |
1da177e4 | 572 | if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { |
65333b4c DC |
573 | xfs_warn(ip->i_mount, |
574 | "corrupt inode %Lu (bad size %Ld for local inode).", | |
1da177e4 LT |
575 | (unsigned long long) ip->i_ino, |
576 | (long long) di_size); | |
577 | XFS_CORRUPTION_ERROR("xfs_iformat(5)", | |
578 | XFS_ERRLEVEL_LOW, | |
579 | ip->i_mount, dip); | |
580 | return XFS_ERROR(EFSCORRUPTED); | |
581 | } | |
582 | ||
583 | size = (int)di_size; | |
584 | error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); | |
585 | break; | |
586 | case XFS_DINODE_FMT_EXTENTS: | |
587 | error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); | |
588 | break; | |
589 | case XFS_DINODE_FMT_BTREE: | |
590 | error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); | |
591 | break; | |
592 | default: | |
593 | XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, | |
594 | ip->i_mount); | |
595 | return XFS_ERROR(EFSCORRUPTED); | |
596 | } | |
597 | break; | |
598 | ||
599 | default: | |
600 | XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); | |
601 | return XFS_ERROR(EFSCORRUPTED); | |
602 | } | |
603 | if (error) { | |
604 | return error; | |
605 | } | |
606 | if (!XFS_DFORK_Q(dip)) | |
607 | return 0; | |
8096b1eb | 608 | |
1da177e4 | 609 | ASSERT(ip->i_afp == NULL); |
4a7edddc | 610 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS); |
8096b1eb | 611 | |
81591fe2 | 612 | switch (dip->di_aformat) { |
1da177e4 LT |
613 | case XFS_DINODE_FMT_LOCAL: |
614 | atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); | |
3b244aa8 | 615 | size = be16_to_cpu(atp->hdr.totsize); |
2809f76a CH |
616 | |
617 | if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) { | |
65333b4c DC |
618 | xfs_warn(ip->i_mount, |
619 | "corrupt inode %Lu (bad attr fork size %Ld).", | |
2809f76a CH |
620 | (unsigned long long) ip->i_ino, |
621 | (long long) size); | |
622 | XFS_CORRUPTION_ERROR("xfs_iformat(8)", | |
623 | XFS_ERRLEVEL_LOW, | |
624 | ip->i_mount, dip); | |
625 | return XFS_ERROR(EFSCORRUPTED); | |
626 | } | |
627 | ||
1da177e4 LT |
628 | error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); |
629 | break; | |
630 | case XFS_DINODE_FMT_EXTENTS: | |
631 | error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); | |
632 | break; | |
633 | case XFS_DINODE_FMT_BTREE: | |
634 | error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); | |
635 | break; | |
636 | default: | |
637 | error = XFS_ERROR(EFSCORRUPTED); | |
638 | break; | |
639 | } | |
640 | if (error) { | |
641 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
642 | ip->i_afp = NULL; | |
643 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
644 | } | |
645 | return error; | |
646 | } | |
647 | ||
648 | /* | |
649 | * The file is in-lined in the on-disk inode. | |
650 | * If it fits into if_inline_data, then copy | |
651 | * it there, otherwise allocate a buffer for it | |
652 | * and copy the data there. Either way, set | |
653 | * if_data to point at the data. | |
654 | * If we allocate a buffer for the data, make | |
655 | * sure that its size is a multiple of 4 and | |
656 | * record the real size in i_real_bytes. | |
657 | */ | |
658 | STATIC int | |
659 | xfs_iformat_local( | |
660 | xfs_inode_t *ip, | |
661 | xfs_dinode_t *dip, | |
662 | int whichfork, | |
663 | int size) | |
664 | { | |
665 | xfs_ifork_t *ifp; | |
666 | int real_size; | |
667 | ||
668 | /* | |
669 | * If the size is unreasonable, then something | |
670 | * is wrong and we just bail out rather than crash in | |
671 | * kmem_alloc() or memcpy() below. | |
672 | */ | |
673 | if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
65333b4c DC |
674 | xfs_warn(ip->i_mount, |
675 | "corrupt inode %Lu (bad size %d for local fork, size = %d).", | |
1da177e4 LT |
676 | (unsigned long long) ip->i_ino, size, |
677 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); | |
678 | XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, | |
679 | ip->i_mount, dip); | |
680 | return XFS_ERROR(EFSCORRUPTED); | |
681 | } | |
682 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
683 | real_size = 0; | |
684 | if (size == 0) | |
685 | ifp->if_u1.if_data = NULL; | |
686 | else if (size <= sizeof(ifp->if_u2.if_inline_data)) | |
687 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
688 | else { | |
689 | real_size = roundup(size, 4); | |
4a7edddc | 690 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS); |
1da177e4 LT |
691 | } |
692 | ifp->if_bytes = size; | |
693 | ifp->if_real_bytes = real_size; | |
694 | if (size) | |
695 | memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); | |
696 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
697 | ifp->if_flags |= XFS_IFINLINE; | |
698 | return 0; | |
699 | } | |
700 | ||
701 | /* | |
702 | * The file consists of a set of extents all | |
703 | * of which fit into the on-disk inode. | |
704 | * If there are few enough extents to fit into | |
705 | * the if_inline_ext, then copy them there. | |
706 | * Otherwise allocate a buffer for them and copy | |
707 | * them into it. Either way, set if_extents | |
708 | * to point at the extents. | |
709 | */ | |
710 | STATIC int | |
711 | xfs_iformat_extents( | |
712 | xfs_inode_t *ip, | |
713 | xfs_dinode_t *dip, | |
714 | int whichfork) | |
715 | { | |
a6f64d4a | 716 | xfs_bmbt_rec_t *dp; |
1da177e4 LT |
717 | xfs_ifork_t *ifp; |
718 | int nex; | |
1da177e4 LT |
719 | int size; |
720 | int i; | |
721 | ||
722 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
723 | nex = XFS_DFORK_NEXTENTS(dip, whichfork); | |
724 | size = nex * (uint)sizeof(xfs_bmbt_rec_t); | |
725 | ||
726 | /* | |
727 | * If the number of extents is unreasonable, then something | |
728 | * is wrong and we just bail out rather than crash in | |
729 | * kmem_alloc() or memcpy() below. | |
730 | */ | |
731 | if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
65333b4c | 732 | xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).", |
1da177e4 LT |
733 | (unsigned long long) ip->i_ino, nex); |
734 | XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, | |
735 | ip->i_mount, dip); | |
736 | return XFS_ERROR(EFSCORRUPTED); | |
737 | } | |
738 | ||
4eea22f0 | 739 | ifp->if_real_bytes = 0; |
1da177e4 LT |
740 | if (nex == 0) |
741 | ifp->if_u1.if_extents = NULL; | |
742 | else if (nex <= XFS_INLINE_EXTS) | |
743 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
4eea22f0 MK |
744 | else |
745 | xfs_iext_add(ifp, 0, nex); | |
746 | ||
1da177e4 | 747 | ifp->if_bytes = size; |
1da177e4 LT |
748 | if (size) { |
749 | dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); | |
a6f64d4a | 750 | xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip)); |
4eea22f0 | 751 | for (i = 0; i < nex; i++, dp++) { |
a6f64d4a | 752 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
597bca63 HH |
753 | ep->l0 = get_unaligned_be64(&dp->l0); |
754 | ep->l1 = get_unaligned_be64(&dp->l1); | |
1da177e4 | 755 | } |
3a59c94c | 756 | XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork); |
1da177e4 LT |
757 | if (whichfork != XFS_DATA_FORK || |
758 | XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) | |
759 | if (unlikely(xfs_check_nostate_extents( | |
4eea22f0 | 760 | ifp, 0, nex))) { |
1da177e4 LT |
761 | XFS_ERROR_REPORT("xfs_iformat_extents(2)", |
762 | XFS_ERRLEVEL_LOW, | |
763 | ip->i_mount); | |
764 | return XFS_ERROR(EFSCORRUPTED); | |
765 | } | |
766 | } | |
767 | ifp->if_flags |= XFS_IFEXTENTS; | |
768 | return 0; | |
769 | } | |
770 | ||
771 | /* | |
772 | * The file has too many extents to fit into | |
773 | * the inode, so they are in B-tree format. | |
774 | * Allocate a buffer for the root of the B-tree | |
775 | * and copy the root into it. The i_extents | |
776 | * field will remain NULL until all of the | |
777 | * extents are read in (when they are needed). | |
778 | */ | |
779 | STATIC int | |
780 | xfs_iformat_btree( | |
781 | xfs_inode_t *ip, | |
782 | xfs_dinode_t *dip, | |
783 | int whichfork) | |
784 | { | |
785 | xfs_bmdr_block_t *dfp; | |
786 | xfs_ifork_t *ifp; | |
787 | /* REFERENCED */ | |
788 | int nrecs; | |
789 | int size; | |
790 | ||
791 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
792 | dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); | |
793 | size = XFS_BMAP_BROOT_SPACE(dfp); | |
60197e8d | 794 | nrecs = be16_to_cpu(dfp->bb_numrecs); |
1da177e4 LT |
795 | |
796 | /* | |
797 | * blow out if -- fork has less extents than can fit in | |
798 | * fork (fork shouldn't be a btree format), root btree | |
799 | * block has more records than can fit into the fork, | |
800 | * or the number of extents is greater than the number of | |
801 | * blocks. | |
802 | */ | |
8096b1eb CH |
803 | if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= |
804 | XFS_IFORK_MAXEXT(ip, whichfork) || | |
805 | XFS_BMDR_SPACE_CALC(nrecs) > | |
806 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) || | |
807 | XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { | |
65333b4c | 808 | xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).", |
1da177e4 | 809 | (unsigned long long) ip->i_ino); |
65333b4c DC |
810 | XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW, |
811 | ip->i_mount, dip); | |
1da177e4 LT |
812 | return XFS_ERROR(EFSCORRUPTED); |
813 | } | |
814 | ||
815 | ifp->if_broot_bytes = size; | |
4a7edddc | 816 | ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS); |
1da177e4 LT |
817 | ASSERT(ifp->if_broot != NULL); |
818 | /* | |
819 | * Copy and convert from the on-disk structure | |
820 | * to the in-memory structure. | |
821 | */ | |
60197e8d CH |
822 | xfs_bmdr_to_bmbt(ip->i_mount, dfp, |
823 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), | |
824 | ifp->if_broot, size); | |
1da177e4 LT |
825 | ifp->if_flags &= ~XFS_IFEXTENTS; |
826 | ifp->if_flags |= XFS_IFBROOT; | |
827 | ||
828 | return 0; | |
829 | } | |
830 | ||
d96f8f89 | 831 | STATIC void |
347d1c01 CH |
832 | xfs_dinode_from_disk( |
833 | xfs_icdinode_t *to, | |
81591fe2 | 834 | xfs_dinode_t *from) |
1da177e4 | 835 | { |
347d1c01 CH |
836 | to->di_magic = be16_to_cpu(from->di_magic); |
837 | to->di_mode = be16_to_cpu(from->di_mode); | |
838 | to->di_version = from ->di_version; | |
839 | to->di_format = from->di_format; | |
840 | to->di_onlink = be16_to_cpu(from->di_onlink); | |
841 | to->di_uid = be32_to_cpu(from->di_uid); | |
842 | to->di_gid = be32_to_cpu(from->di_gid); | |
843 | to->di_nlink = be32_to_cpu(from->di_nlink); | |
6743099c AM |
844 | to->di_projid_lo = be16_to_cpu(from->di_projid_lo); |
845 | to->di_projid_hi = be16_to_cpu(from->di_projid_hi); | |
347d1c01 CH |
846 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); |
847 | to->di_flushiter = be16_to_cpu(from->di_flushiter); | |
848 | to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec); | |
849 | to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec); | |
850 | to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec); | |
851 | to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec); | |
852 | to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec); | |
853 | to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec); | |
854 | to->di_size = be64_to_cpu(from->di_size); | |
855 | to->di_nblocks = be64_to_cpu(from->di_nblocks); | |
856 | to->di_extsize = be32_to_cpu(from->di_extsize); | |
857 | to->di_nextents = be32_to_cpu(from->di_nextents); | |
858 | to->di_anextents = be16_to_cpu(from->di_anextents); | |
859 | to->di_forkoff = from->di_forkoff; | |
860 | to->di_aformat = from->di_aformat; | |
861 | to->di_dmevmask = be32_to_cpu(from->di_dmevmask); | |
862 | to->di_dmstate = be16_to_cpu(from->di_dmstate); | |
863 | to->di_flags = be16_to_cpu(from->di_flags); | |
864 | to->di_gen = be32_to_cpu(from->di_gen); | |
865 | } | |
866 | ||
867 | void | |
868 | xfs_dinode_to_disk( | |
81591fe2 | 869 | xfs_dinode_t *to, |
347d1c01 CH |
870 | xfs_icdinode_t *from) |
871 | { | |
872 | to->di_magic = cpu_to_be16(from->di_magic); | |
873 | to->di_mode = cpu_to_be16(from->di_mode); | |
874 | to->di_version = from ->di_version; | |
875 | to->di_format = from->di_format; | |
876 | to->di_onlink = cpu_to_be16(from->di_onlink); | |
877 | to->di_uid = cpu_to_be32(from->di_uid); | |
878 | to->di_gid = cpu_to_be32(from->di_gid); | |
879 | to->di_nlink = cpu_to_be32(from->di_nlink); | |
6743099c AM |
880 | to->di_projid_lo = cpu_to_be16(from->di_projid_lo); |
881 | to->di_projid_hi = cpu_to_be16(from->di_projid_hi); | |
347d1c01 CH |
882 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); |
883 | to->di_flushiter = cpu_to_be16(from->di_flushiter); | |
884 | to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec); | |
885 | to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec); | |
886 | to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec); | |
887 | to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec); | |
888 | to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec); | |
889 | to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec); | |
890 | to->di_size = cpu_to_be64(from->di_size); | |
891 | to->di_nblocks = cpu_to_be64(from->di_nblocks); | |
892 | to->di_extsize = cpu_to_be32(from->di_extsize); | |
893 | to->di_nextents = cpu_to_be32(from->di_nextents); | |
894 | to->di_anextents = cpu_to_be16(from->di_anextents); | |
895 | to->di_forkoff = from->di_forkoff; | |
896 | to->di_aformat = from->di_aformat; | |
897 | to->di_dmevmask = cpu_to_be32(from->di_dmevmask); | |
898 | to->di_dmstate = cpu_to_be16(from->di_dmstate); | |
899 | to->di_flags = cpu_to_be16(from->di_flags); | |
900 | to->di_gen = cpu_to_be32(from->di_gen); | |
1da177e4 LT |
901 | } |
902 | ||
903 | STATIC uint | |
904 | _xfs_dic2xflags( | |
1da177e4 LT |
905 | __uint16_t di_flags) |
906 | { | |
907 | uint flags = 0; | |
908 | ||
909 | if (di_flags & XFS_DIFLAG_ANY) { | |
910 | if (di_flags & XFS_DIFLAG_REALTIME) | |
911 | flags |= XFS_XFLAG_REALTIME; | |
912 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
913 | flags |= XFS_XFLAG_PREALLOC; | |
914 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
915 | flags |= XFS_XFLAG_IMMUTABLE; | |
916 | if (di_flags & XFS_DIFLAG_APPEND) | |
917 | flags |= XFS_XFLAG_APPEND; | |
918 | if (di_flags & XFS_DIFLAG_SYNC) | |
919 | flags |= XFS_XFLAG_SYNC; | |
920 | if (di_flags & XFS_DIFLAG_NOATIME) | |
921 | flags |= XFS_XFLAG_NOATIME; | |
922 | if (di_flags & XFS_DIFLAG_NODUMP) | |
923 | flags |= XFS_XFLAG_NODUMP; | |
924 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
925 | flags |= XFS_XFLAG_RTINHERIT; | |
926 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
927 | flags |= XFS_XFLAG_PROJINHERIT; | |
928 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
929 | flags |= XFS_XFLAG_NOSYMLINKS; | |
dd9f438e NS |
930 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
931 | flags |= XFS_XFLAG_EXTSIZE; | |
932 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) | |
933 | flags |= XFS_XFLAG_EXTSZINHERIT; | |
d3446eac BN |
934 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
935 | flags |= XFS_XFLAG_NODEFRAG; | |
2a82b8be DC |
936 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
937 | flags |= XFS_XFLAG_FILESTREAM; | |
1da177e4 LT |
938 | } |
939 | ||
940 | return flags; | |
941 | } | |
942 | ||
943 | uint | |
944 | xfs_ip2xflags( | |
945 | xfs_inode_t *ip) | |
946 | { | |
347d1c01 | 947 | xfs_icdinode_t *dic = &ip->i_d; |
1da177e4 | 948 | |
a916e2bd | 949 | return _xfs_dic2xflags(dic->di_flags) | |
45ba598e | 950 | (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
951 | } |
952 | ||
953 | uint | |
954 | xfs_dic2xflags( | |
45ba598e | 955 | xfs_dinode_t *dip) |
1da177e4 | 956 | { |
81591fe2 | 957 | return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | |
45ba598e | 958 | (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
959 | } |
960 | ||
07c8f675 | 961 | /* |
24f211ba | 962 | * Read the disk inode attributes into the in-core inode structure. |
1da177e4 LT |
963 | */ |
964 | int | |
965 | xfs_iread( | |
966 | xfs_mount_t *mp, | |
967 | xfs_trans_t *tp, | |
24f211ba | 968 | xfs_inode_t *ip, |
24f211ba | 969 | uint iget_flags) |
1da177e4 LT |
970 | { |
971 | xfs_buf_t *bp; | |
972 | xfs_dinode_t *dip; | |
1da177e4 LT |
973 | int error; |
974 | ||
1da177e4 | 975 | /* |
92bfc6e7 | 976 | * Fill in the location information in the in-core inode. |
1da177e4 | 977 | */ |
24f211ba | 978 | error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags); |
76d8b277 | 979 | if (error) |
24f211ba | 980 | return error; |
76d8b277 CH |
981 | |
982 | /* | |
92bfc6e7 | 983 | * Get pointers to the on-disk inode and the buffer containing it. |
76d8b277 | 984 | */ |
475ee413 | 985 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0, iget_flags); |
9ed0451e | 986 | if (error) |
24f211ba | 987 | return error; |
1da177e4 | 988 | |
1da177e4 LT |
989 | /* |
990 | * If we got something that isn't an inode it means someone | |
991 | * (nfs or dmi) has a stale handle. | |
992 | */ | |
69ef921b | 993 | if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) { |
1da177e4 | 994 | #ifdef DEBUG |
53487786 DC |
995 | xfs_alert(mp, |
996 | "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)", | |
997 | __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC); | |
1da177e4 | 998 | #endif /* DEBUG */ |
9ed0451e CH |
999 | error = XFS_ERROR(EINVAL); |
1000 | goto out_brelse; | |
1da177e4 LT |
1001 | } |
1002 | ||
1003 | /* | |
1004 | * If the on-disk inode is already linked to a directory | |
1005 | * entry, copy all of the inode into the in-core inode. | |
1006 | * xfs_iformat() handles copying in the inode format | |
1007 | * specific information. | |
1008 | * Otherwise, just get the truly permanent information. | |
1009 | */ | |
81591fe2 CH |
1010 | if (dip->di_mode) { |
1011 | xfs_dinode_from_disk(&ip->i_d, dip); | |
1da177e4 LT |
1012 | error = xfs_iformat(ip, dip); |
1013 | if (error) { | |
1da177e4 | 1014 | #ifdef DEBUG |
53487786 DC |
1015 | xfs_alert(mp, "%s: xfs_iformat() returned error %d", |
1016 | __func__, error); | |
1da177e4 | 1017 | #endif /* DEBUG */ |
9ed0451e | 1018 | goto out_brelse; |
1da177e4 LT |
1019 | } |
1020 | } else { | |
81591fe2 CH |
1021 | ip->i_d.di_magic = be16_to_cpu(dip->di_magic); |
1022 | ip->i_d.di_version = dip->di_version; | |
1023 | ip->i_d.di_gen = be32_to_cpu(dip->di_gen); | |
1024 | ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter); | |
1da177e4 LT |
1025 | /* |
1026 | * Make sure to pull in the mode here as well in | |
1027 | * case the inode is released without being used. | |
1028 | * This ensures that xfs_inactive() will see that | |
1029 | * the inode is already free and not try to mess | |
1030 | * with the uninitialized part of it. | |
1031 | */ | |
1032 | ip->i_d.di_mode = 0; | |
1da177e4 LT |
1033 | } |
1034 | ||
1da177e4 LT |
1035 | /* |
1036 | * The inode format changed when we moved the link count and | |
1037 | * made it 32 bits long. If this is an old format inode, | |
1038 | * convert it in memory to look like a new one. If it gets | |
1039 | * flushed to disk we will convert back before flushing or | |
1040 | * logging it. We zero out the new projid field and the old link | |
1041 | * count field. We'll handle clearing the pad field (the remains | |
1042 | * of the old uuid field) when we actually convert the inode to | |
1043 | * the new format. We don't change the version number so that we | |
1044 | * can distinguish this from a real new format inode. | |
1045 | */ | |
51ce16d5 | 1046 | if (ip->i_d.di_version == 1) { |
1da177e4 LT |
1047 | ip->i_d.di_nlink = ip->i_d.di_onlink; |
1048 | ip->i_d.di_onlink = 0; | |
6743099c | 1049 | xfs_set_projid(ip, 0); |
1da177e4 LT |
1050 | } |
1051 | ||
1052 | ip->i_delayed_blks = 0; | |
1053 | ||
1054 | /* | |
1055 | * Mark the buffer containing the inode as something to keep | |
1056 | * around for a while. This helps to keep recently accessed | |
1057 | * meta-data in-core longer. | |
1058 | */ | |
821eb21d | 1059 | xfs_buf_set_ref(bp, XFS_INO_REF); |
1da177e4 LT |
1060 | |
1061 | /* | |
1062 | * Use xfs_trans_brelse() to release the buffer containing the | |
1063 | * on-disk inode, because it was acquired with xfs_trans_read_buf() | |
475ee413 | 1064 | * in xfs_imap_to_bp() above. If tp is NULL, this is just a normal |
1da177e4 LT |
1065 | * brelse(). If we're within a transaction, then xfs_trans_brelse() |
1066 | * will only release the buffer if it is not dirty within the | |
1067 | * transaction. It will be OK to release the buffer in this case, | |
1068 | * because inodes on disk are never destroyed and we will be | |
1069 | * locking the new in-core inode before putting it in the hash | |
1070 | * table where other processes can find it. Thus we don't have | |
1071 | * to worry about the inode being changed just because we released | |
1072 | * the buffer. | |
1073 | */ | |
9ed0451e CH |
1074 | out_brelse: |
1075 | xfs_trans_brelse(tp, bp); | |
9ed0451e | 1076 | return error; |
1da177e4 LT |
1077 | } |
1078 | ||
1079 | /* | |
1080 | * Read in extents from a btree-format inode. | |
1081 | * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. | |
1082 | */ | |
1083 | int | |
1084 | xfs_iread_extents( | |
1085 | xfs_trans_t *tp, | |
1086 | xfs_inode_t *ip, | |
1087 | int whichfork) | |
1088 | { | |
1089 | int error; | |
1090 | xfs_ifork_t *ifp; | |
4eea22f0 | 1091 | xfs_extnum_t nextents; |
1da177e4 LT |
1092 | |
1093 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1094 | XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, | |
1095 | ip->i_mount); | |
1096 | return XFS_ERROR(EFSCORRUPTED); | |
1097 | } | |
4eea22f0 | 1098 | nextents = XFS_IFORK_NEXTENTS(ip, whichfork); |
1da177e4 | 1099 | ifp = XFS_IFORK_PTR(ip, whichfork); |
4eea22f0 | 1100 | |
1da177e4 LT |
1101 | /* |
1102 | * We know that the size is valid (it's checked in iformat_btree) | |
1103 | */ | |
4eea22f0 | 1104 | ifp->if_bytes = ifp->if_real_bytes = 0; |
1da177e4 | 1105 | ifp->if_flags |= XFS_IFEXTENTS; |
4eea22f0 | 1106 | xfs_iext_add(ifp, 0, nextents); |
1da177e4 LT |
1107 | error = xfs_bmap_read_extents(tp, ip, whichfork); |
1108 | if (error) { | |
4eea22f0 | 1109 | xfs_iext_destroy(ifp); |
1da177e4 LT |
1110 | ifp->if_flags &= ~XFS_IFEXTENTS; |
1111 | return error; | |
1112 | } | |
a6f64d4a | 1113 | xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
1114 | return 0; |
1115 | } | |
1116 | ||
1117 | /* | |
1118 | * Allocate an inode on disk and return a copy of its in-core version. | |
1119 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
1120 | * appropriately within the inode. The uid and gid for the inode are | |
1121 | * set according to the contents of the given cred structure. | |
1122 | * | |
1123 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
1124 | * has a free inode available, call xfs_iget() to obtain the in-core |
1125 | * version of the allocated inode. Finally, fill in the inode and | |
1126 | * log its initial contents. In this case, ialloc_context would be | |
1127 | * set to NULL. | |
1da177e4 | 1128 | * |
cd856db6 CM |
1129 | * If xfs_dialloc() does not have an available inode, it will replenish |
1130 | * its supply by doing an allocation. Since we can only do one | |
1131 | * allocation within a transaction without deadlocks, we must commit | |
1132 | * the current transaction before returning the inode itself. | |
1133 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
1134 | * The caller should then commit the current transaction, start a new |
1135 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
1136 | * | |
1137 | * To ensure that some other process does not grab the inode that | |
1138 | * was allocated during the first call to xfs_ialloc(), this routine | |
1139 | * also returns the [locked] bp pointing to the head of the freelist | |
1140 | * as ialloc_context. The caller should hold this buffer across | |
1141 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
1142 | * |
1143 | * If we are allocating quota inodes, we do not have a parent inode | |
1144 | * to attach to or associate with (i.e. pip == NULL) because they | |
1145 | * are not linked into the directory structure - they are attached | |
1146 | * directly to the superblock - and so have no parent. | |
1da177e4 LT |
1147 | */ |
1148 | int | |
1149 | xfs_ialloc( | |
1150 | xfs_trans_t *tp, | |
1151 | xfs_inode_t *pip, | |
576b1d67 | 1152 | umode_t mode, |
31b084ae | 1153 | xfs_nlink_t nlink, |
1da177e4 | 1154 | xfs_dev_t rdev, |
6743099c | 1155 | prid_t prid, |
1da177e4 LT |
1156 | int okalloc, |
1157 | xfs_buf_t **ialloc_context, | |
1da177e4 LT |
1158 | xfs_inode_t **ipp) |
1159 | { | |
1160 | xfs_ino_t ino; | |
1161 | xfs_inode_t *ip; | |
1da177e4 LT |
1162 | uint flags; |
1163 | int error; | |
dff35fd4 | 1164 | timespec_t tv; |
bf904248 | 1165 | int filestreams = 0; |
1da177e4 LT |
1166 | |
1167 | /* | |
1168 | * Call the space management code to pick | |
1169 | * the on-disk inode to be allocated. | |
1170 | */ | |
b11f94d5 | 1171 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
08358906 | 1172 | ialloc_context, &ino); |
bf904248 | 1173 | if (error) |
1da177e4 | 1174 | return error; |
08358906 | 1175 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
1176 | *ipp = NULL; |
1177 | return 0; | |
1178 | } | |
1179 | ASSERT(*ialloc_context == NULL); | |
1180 | ||
1181 | /* | |
1182 | * Get the in-core inode with the lock held exclusively. | |
1183 | * This is because we're setting fields here we need | |
1184 | * to prevent others from looking at until we're done. | |
1185 | */ | |
ec3ba85f CH |
1186 | error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE, |
1187 | XFS_ILOCK_EXCL, &ip); | |
bf904248 | 1188 | if (error) |
1da177e4 | 1189 | return error; |
1da177e4 LT |
1190 | ASSERT(ip != NULL); |
1191 | ||
576b1d67 | 1192 | ip->i_d.di_mode = mode; |
1da177e4 LT |
1193 | ip->i_d.di_onlink = 0; |
1194 | ip->i_d.di_nlink = nlink; | |
1195 | ASSERT(ip->i_d.di_nlink == nlink); | |
9e2b2dc4 DH |
1196 | ip->i_d.di_uid = current_fsuid(); |
1197 | ip->i_d.di_gid = current_fsgid(); | |
6743099c | 1198 | xfs_set_projid(ip, prid); |
1da177e4 LT |
1199 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
1200 | ||
1201 | /* | |
1202 | * If the superblock version is up to where we support new format | |
1203 | * inodes and this is currently an old format inode, then change | |
1204 | * the inode version number now. This way we only do the conversion | |
1205 | * here rather than here and in the flush/logging code. | |
1206 | */ | |
62118709 | 1207 | if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) && |
51ce16d5 CH |
1208 | ip->i_d.di_version == 1) { |
1209 | ip->i_d.di_version = 2; | |
1da177e4 LT |
1210 | /* |
1211 | * We've already zeroed the old link count, the projid field, | |
1212 | * and the pad field. | |
1213 | */ | |
1214 | } | |
1215 | ||
1216 | /* | |
1217 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
1218 | */ | |
51ce16d5 | 1219 | if ((prid != 0) && (ip->i_d.di_version == 1)) |
1da177e4 LT |
1220 | xfs_bump_ino_vers2(tp, ip); |
1221 | ||
bd186aa9 | 1222 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 1223 | ip->i_d.di_gid = pip->i_d.di_gid; |
abbede1b | 1224 | if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { |
1da177e4 LT |
1225 | ip->i_d.di_mode |= S_ISGID; |
1226 | } | |
1227 | } | |
1228 | ||
1229 | /* | |
1230 | * If the group ID of the new file does not match the effective group | |
1231 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
1232 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
1233 | */ | |
1234 | if ((irix_sgid_inherit) && | |
1235 | (ip->i_d.di_mode & S_ISGID) && | |
1236 | (!in_group_p((gid_t)ip->i_d.di_gid))) { | |
1237 | ip->i_d.di_mode &= ~S_ISGID; | |
1238 | } | |
1239 | ||
1240 | ip->i_d.di_size = 0; | |
1241 | ip->i_d.di_nextents = 0; | |
1242 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 CH |
1243 | |
1244 | nanotime(&tv); | |
1245 | ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; | |
1246 | ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; | |
1247 | ip->i_d.di_atime = ip->i_d.di_mtime; | |
1248 | ip->i_d.di_ctime = ip->i_d.di_mtime; | |
1249 | ||
1da177e4 LT |
1250 | /* |
1251 | * di_gen will have been taken care of in xfs_iread. | |
1252 | */ | |
1253 | ip->i_d.di_extsize = 0; | |
1254 | ip->i_d.di_dmevmask = 0; | |
1255 | ip->i_d.di_dmstate = 0; | |
1256 | ip->i_d.di_flags = 0; | |
1257 | flags = XFS_ILOG_CORE; | |
1258 | switch (mode & S_IFMT) { | |
1259 | case S_IFIFO: | |
1260 | case S_IFCHR: | |
1261 | case S_IFBLK: | |
1262 | case S_IFSOCK: | |
1263 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1264 | ip->i_df.if_u2.if_rdev = rdev; | |
1265 | ip->i_df.if_flags = 0; | |
1266 | flags |= XFS_ILOG_DEV; | |
1267 | break; | |
1268 | case S_IFREG: | |
bf904248 DC |
1269 | /* |
1270 | * we can't set up filestreams until after the VFS inode | |
1271 | * is set up properly. | |
1272 | */ | |
1273 | if (pip && xfs_inode_is_filestream(pip)) | |
1274 | filestreams = 1; | |
2a82b8be | 1275 | /* fall through */ |
1da177e4 | 1276 | case S_IFDIR: |
b11f94d5 | 1277 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
365ca83d NS |
1278 | uint di_flags = 0; |
1279 | ||
abbede1b | 1280 | if (S_ISDIR(mode)) { |
365ca83d NS |
1281 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
1282 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
1283 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1284 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
1285 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1286 | } | |
abbede1b | 1287 | } else if (S_ISREG(mode)) { |
613d7043 | 1288 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 1289 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
1290 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1291 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1292 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1293 | } | |
1da177e4 LT |
1294 | } |
1295 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
1296 | xfs_inherit_noatime) | |
365ca83d | 1297 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
1298 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
1299 | xfs_inherit_nodump) | |
365ca83d | 1300 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
1301 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
1302 | xfs_inherit_sync) | |
365ca83d | 1303 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
1304 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
1305 | xfs_inherit_nosymlinks) | |
365ca83d NS |
1306 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
1307 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
1308 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
1309 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
1310 | xfs_inherit_nodefrag) | |
1311 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
1312 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
1313 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
365ca83d | 1314 | ip->i_d.di_flags |= di_flags; |
1da177e4 LT |
1315 | } |
1316 | /* FALLTHROUGH */ | |
1317 | case S_IFLNK: | |
1318 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
1319 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
1320 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
1321 | ip->i_df.if_u1.if_extents = NULL; | |
1322 | break; | |
1323 | default: | |
1324 | ASSERT(0); | |
1325 | } | |
1326 | /* | |
1327 | * Attribute fork settings for new inode. | |
1328 | */ | |
1329 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1330 | ip->i_d.di_anextents = 0; | |
1331 | ||
1332 | /* | |
1333 | * Log the new values stuffed into the inode. | |
1334 | */ | |
ddc3415a | 1335 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1336 | xfs_trans_log_inode(tp, ip, flags); |
1337 | ||
b83bd138 | 1338 | /* now that we have an i_mode we can setup inode ops and unlock */ |
41be8bed | 1339 | xfs_setup_inode(ip); |
1da177e4 | 1340 | |
bf904248 DC |
1341 | /* now we have set up the vfs inode we can associate the filestream */ |
1342 | if (filestreams) { | |
1343 | error = xfs_filestream_associate(pip, ip); | |
1344 | if (error < 0) | |
1345 | return -error; | |
1346 | if (!error) | |
1347 | xfs_iflags_set(ip, XFS_IFILESTREAM); | |
1348 | } | |
1349 | ||
1da177e4 LT |
1350 | *ipp = ip; |
1351 | return 0; | |
1352 | } | |
1353 | ||
1da177e4 | 1354 | /* |
8f04c47a CH |
1355 | * Free up the underlying blocks past new_size. The new size must be smaller |
1356 | * than the current size. This routine can be used both for the attribute and | |
1357 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1358 | * |
f6485057 DC |
1359 | * The transaction passed to this routine must have made a permanent log |
1360 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1361 | * given transaction and start new ones, so make sure everything involved in | |
1362 | * the transaction is tidy before calling here. Some transaction will be | |
1363 | * returned to the caller to be committed. The incoming transaction must | |
1364 | * already include the inode, and both inode locks must be held exclusively. | |
1365 | * The inode must also be "held" within the transaction. On return the inode | |
1366 | * will be "held" within the returned transaction. This routine does NOT | |
1367 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1368 | * |
f6485057 DC |
1369 | * If we get an error, we must return with the inode locked and linked into the |
1370 | * current transaction. This keeps things simple for the higher level code, | |
1371 | * because it always knows that the inode is locked and held in the transaction | |
1372 | * that returns to it whether errors occur or not. We don't mark the inode | |
1373 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1374 | */ |
1375 | int | |
8f04c47a CH |
1376 | xfs_itruncate_extents( |
1377 | struct xfs_trans **tpp, | |
1378 | struct xfs_inode *ip, | |
1379 | int whichfork, | |
1380 | xfs_fsize_t new_size) | |
1da177e4 | 1381 | { |
8f04c47a CH |
1382 | struct xfs_mount *mp = ip->i_mount; |
1383 | struct xfs_trans *tp = *tpp; | |
1384 | struct xfs_trans *ntp; | |
1385 | xfs_bmap_free_t free_list; | |
1386 | xfs_fsblock_t first_block; | |
1387 | xfs_fileoff_t first_unmap_block; | |
1388 | xfs_fileoff_t last_block; | |
1389 | xfs_filblks_t unmap_len; | |
1390 | int committed; | |
1391 | int error = 0; | |
1392 | int done = 0; | |
1da177e4 | 1393 | |
0b56185b CH |
1394 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1395 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1396 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1397 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1398 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1399 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1400 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1401 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1402 | |
673e8e59 CH |
1403 | trace_xfs_itruncate_extents_start(ip, new_size); |
1404 | ||
1da177e4 LT |
1405 | /* |
1406 | * Since it is possible for space to become allocated beyond | |
1407 | * the end of the file (in a crash where the space is allocated | |
1408 | * but the inode size is not yet updated), simply remove any | |
1409 | * blocks which show up between the new EOF and the maximum | |
1410 | * possible file size. If the first block to be removed is | |
1411 | * beyond the maximum file size (ie it is the same as last_block), | |
1412 | * then there is nothing to do. | |
1413 | */ | |
8f04c47a | 1414 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1415 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1416 | if (first_unmap_block == last_block) |
1417 | return 0; | |
1418 | ||
1419 | ASSERT(first_unmap_block < last_block); | |
1420 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1421 | while (!done) { |
9d87c319 | 1422 | xfs_bmap_init(&free_list, &first_block); |
8f04c47a | 1423 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 1424 | first_unmap_block, unmap_len, |
8f04c47a | 1425 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 1426 | XFS_ITRUNC_MAX_EXTENTS, |
3e57ecf6 | 1427 | &first_block, &free_list, |
b4e9181e | 1428 | &done); |
8f04c47a CH |
1429 | if (error) |
1430 | goto out_bmap_cancel; | |
1da177e4 LT |
1431 | |
1432 | /* | |
1433 | * Duplicate the transaction that has the permanent | |
1434 | * reservation and commit the old transaction. | |
1435 | */ | |
8f04c47a | 1436 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
898621d5 | 1437 | if (committed) |
ddc3415a | 1438 | xfs_trans_ijoin(tp, ip, 0); |
8f04c47a CH |
1439 | if (error) |
1440 | goto out_bmap_cancel; | |
1da177e4 LT |
1441 | |
1442 | if (committed) { | |
1443 | /* | |
f6485057 | 1444 | * Mark the inode dirty so it will be logged and |
e5720eec | 1445 | * moved forward in the log as part of every commit. |
1da177e4 | 1446 | */ |
8f04c47a | 1447 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1da177e4 | 1448 | } |
f6485057 | 1449 | |
8f04c47a CH |
1450 | ntp = xfs_trans_dup(tp); |
1451 | error = xfs_trans_commit(tp, 0); | |
1452 | tp = ntp; | |
e5720eec | 1453 | |
ddc3415a | 1454 | xfs_trans_ijoin(tp, ip, 0); |
f6485057 | 1455 | |
cc09c0dc | 1456 | if (error) |
8f04c47a CH |
1457 | goto out; |
1458 | ||
cc09c0dc | 1459 | /* |
8f04c47a | 1460 | * Transaction commit worked ok so we can drop the extra ticket |
cc09c0dc DC |
1461 | * reference that we gained in xfs_trans_dup() |
1462 | */ | |
8f04c47a CH |
1463 | xfs_log_ticket_put(tp->t_ticket); |
1464 | error = xfs_trans_reserve(tp, 0, | |
f6485057 DC |
1465 | XFS_ITRUNCATE_LOG_RES(mp), 0, |
1466 | XFS_TRANS_PERM_LOG_RES, | |
1467 | XFS_ITRUNCATE_LOG_COUNT); | |
1468 | if (error) | |
8f04c47a | 1469 | goto out; |
1da177e4 | 1470 | } |
8f04c47a | 1471 | |
673e8e59 CH |
1472 | /* |
1473 | * Always re-log the inode so that our permanent transaction can keep | |
1474 | * on rolling it forward in the log. | |
1475 | */ | |
1476 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1477 | ||
1478 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1479 | ||
8f04c47a CH |
1480 | out: |
1481 | *tpp = tp; | |
1482 | return error; | |
1483 | out_bmap_cancel: | |
1da177e4 | 1484 | /* |
8f04c47a CH |
1485 | * If the bunmapi call encounters an error, return to the caller where |
1486 | * the transaction can be properly aborted. We just need to make sure | |
1487 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1488 | */ |
8f04c47a CH |
1489 | xfs_bmap_cancel(&free_list); |
1490 | goto out; | |
1491 | } | |
1492 | ||
1da177e4 LT |
1493 | /* |
1494 | * This is called when the inode's link count goes to 0. | |
1495 | * We place the on-disk inode on a list in the AGI. It | |
1496 | * will be pulled from this list when the inode is freed. | |
1497 | */ | |
1498 | int | |
1499 | xfs_iunlink( | |
1500 | xfs_trans_t *tp, | |
1501 | xfs_inode_t *ip) | |
1502 | { | |
1503 | xfs_mount_t *mp; | |
1504 | xfs_agi_t *agi; | |
1505 | xfs_dinode_t *dip; | |
1506 | xfs_buf_t *agibp; | |
1507 | xfs_buf_t *ibp; | |
1da177e4 LT |
1508 | xfs_agino_t agino; |
1509 | short bucket_index; | |
1510 | int offset; | |
1511 | int error; | |
1da177e4 LT |
1512 | |
1513 | ASSERT(ip->i_d.di_nlink == 0); | |
1514 | ASSERT(ip->i_d.di_mode != 0); | |
1da177e4 LT |
1515 | |
1516 | mp = tp->t_mountp; | |
1517 | ||
1da177e4 LT |
1518 | /* |
1519 | * Get the agi buffer first. It ensures lock ordering | |
1520 | * on the list. | |
1521 | */ | |
5e1be0fb | 1522 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 1523 | if (error) |
1da177e4 | 1524 | return error; |
1da177e4 | 1525 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1526 | |
1da177e4 LT |
1527 | /* |
1528 | * Get the index into the agi hash table for the | |
1529 | * list this inode will go on. | |
1530 | */ | |
1531 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1532 | ASSERT(agino != 0); | |
1533 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1534 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1535 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1536 | |
69ef921b | 1537 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
1538 | /* |
1539 | * There is already another inode in the bucket we need | |
1540 | * to add ourselves to. Add us at the front of the list. | |
1541 | * Here we put the head pointer into our next pointer, | |
1542 | * and then we fall through to point the head at us. | |
1543 | */ | |
475ee413 CH |
1544 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1545 | 0, 0); | |
c319b58b VA |
1546 | if (error) |
1547 | return error; | |
1548 | ||
69ef921b | 1549 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 1550 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 1551 | offset = ip->i_imap.im_boffset + |
1da177e4 LT |
1552 | offsetof(xfs_dinode_t, di_next_unlinked); |
1553 | xfs_trans_inode_buf(tp, ibp); | |
1554 | xfs_trans_log_buf(tp, ibp, offset, | |
1555 | (offset + sizeof(xfs_agino_t) - 1)); | |
1556 | xfs_inobp_check(mp, ibp); | |
1557 | } | |
1558 | ||
1559 | /* | |
1560 | * Point the bucket head pointer at the inode being inserted. | |
1561 | */ | |
1562 | ASSERT(agino != 0); | |
16259e7d | 1563 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
1564 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1565 | (sizeof(xfs_agino_t) * bucket_index); | |
1566 | xfs_trans_log_buf(tp, agibp, offset, | |
1567 | (offset + sizeof(xfs_agino_t) - 1)); | |
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | /* | |
1572 | * Pull the on-disk inode from the AGI unlinked list. | |
1573 | */ | |
1574 | STATIC int | |
1575 | xfs_iunlink_remove( | |
1576 | xfs_trans_t *tp, | |
1577 | xfs_inode_t *ip) | |
1578 | { | |
1579 | xfs_ino_t next_ino; | |
1580 | xfs_mount_t *mp; | |
1581 | xfs_agi_t *agi; | |
1582 | xfs_dinode_t *dip; | |
1583 | xfs_buf_t *agibp; | |
1584 | xfs_buf_t *ibp; | |
1585 | xfs_agnumber_t agno; | |
1da177e4 LT |
1586 | xfs_agino_t agino; |
1587 | xfs_agino_t next_agino; | |
1588 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 1589 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 1590 | short bucket_index; |
6fdf8ccc | 1591 | int offset, last_offset = 0; |
1da177e4 | 1592 | int error; |
1da177e4 | 1593 | |
1da177e4 | 1594 | mp = tp->t_mountp; |
1da177e4 | 1595 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
1596 | |
1597 | /* | |
1598 | * Get the agi buffer first. It ensures lock ordering | |
1599 | * on the list. | |
1600 | */ | |
5e1be0fb CH |
1601 | error = xfs_read_agi(mp, tp, agno, &agibp); |
1602 | if (error) | |
1da177e4 | 1603 | return error; |
5e1be0fb | 1604 | |
1da177e4 | 1605 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1606 | |
1da177e4 LT |
1607 | /* |
1608 | * Get the index into the agi hash table for the | |
1609 | * list this inode will go on. | |
1610 | */ | |
1611 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1612 | ASSERT(agino != 0); | |
1613 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 1614 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
1615 | ASSERT(agi->agi_unlinked[bucket_index]); |
1616 | ||
16259e7d | 1617 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 1618 | /* |
475ee413 CH |
1619 | * We're at the head of the list. Get the inode's on-disk |
1620 | * buffer to see if there is anyone after us on the list. | |
1621 | * Only modify our next pointer if it is not already NULLAGINO. | |
1622 | * This saves us the overhead of dealing with the buffer when | |
1623 | * there is no need to change it. | |
1da177e4 | 1624 | */ |
475ee413 CH |
1625 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1626 | 0, 0); | |
1da177e4 | 1627 | if (error) { |
475ee413 | 1628 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 1629 | __func__, error); |
1da177e4 LT |
1630 | return error; |
1631 | } | |
347d1c01 | 1632 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
1633 | ASSERT(next_agino != 0); |
1634 | if (next_agino != NULLAGINO) { | |
347d1c01 | 1635 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 1636 | offset = ip->i_imap.im_boffset + |
1da177e4 LT |
1637 | offsetof(xfs_dinode_t, di_next_unlinked); |
1638 | xfs_trans_inode_buf(tp, ibp); | |
1639 | xfs_trans_log_buf(tp, ibp, offset, | |
1640 | (offset + sizeof(xfs_agino_t) - 1)); | |
1641 | xfs_inobp_check(mp, ibp); | |
1642 | } else { | |
1643 | xfs_trans_brelse(tp, ibp); | |
1644 | } | |
1645 | /* | |
1646 | * Point the bucket head pointer at the next inode. | |
1647 | */ | |
1648 | ASSERT(next_agino != 0); | |
1649 | ASSERT(next_agino != agino); | |
16259e7d | 1650 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
1651 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1652 | (sizeof(xfs_agino_t) * bucket_index); | |
1653 | xfs_trans_log_buf(tp, agibp, offset, | |
1654 | (offset + sizeof(xfs_agino_t) - 1)); | |
1655 | } else { | |
1656 | /* | |
1657 | * We need to search the list for the inode being freed. | |
1658 | */ | |
16259e7d | 1659 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
1660 | last_ibp = NULL; |
1661 | while (next_agino != agino) { | |
129dbc9a CH |
1662 | struct xfs_imap imap; |
1663 | ||
1664 | if (last_ibp) | |
1da177e4 | 1665 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
1666 | |
1667 | imap.im_blkno = 0; | |
1da177e4 | 1668 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
1669 | |
1670 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
1671 | if (error) { | |
1672 | xfs_warn(mp, | |
1673 | "%s: xfs_imap returned error %d.", | |
1674 | __func__, error); | |
1675 | return error; | |
1676 | } | |
1677 | ||
1678 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
1679 | &last_ibp, 0, 0); | |
1da177e4 | 1680 | if (error) { |
0b932ccc | 1681 | xfs_warn(mp, |
129dbc9a | 1682 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 1683 | __func__, error); |
1da177e4 LT |
1684 | return error; |
1685 | } | |
129dbc9a CH |
1686 | |
1687 | last_offset = imap.im_boffset; | |
347d1c01 | 1688 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
1689 | ASSERT(next_agino != NULLAGINO); |
1690 | ASSERT(next_agino != 0); | |
1691 | } | |
475ee413 | 1692 | |
1da177e4 | 1693 | /* |
475ee413 CH |
1694 | * Now last_ibp points to the buffer previous to us on the |
1695 | * unlinked list. Pull us from the list. | |
1da177e4 | 1696 | */ |
475ee413 CH |
1697 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1698 | 0, 0); | |
1da177e4 | 1699 | if (error) { |
475ee413 | 1700 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 1701 | __func__, error); |
1da177e4 LT |
1702 | return error; |
1703 | } | |
347d1c01 | 1704 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
1705 | ASSERT(next_agino != 0); |
1706 | ASSERT(next_agino != agino); | |
1707 | if (next_agino != NULLAGINO) { | |
347d1c01 | 1708 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 1709 | offset = ip->i_imap.im_boffset + |
1da177e4 LT |
1710 | offsetof(xfs_dinode_t, di_next_unlinked); |
1711 | xfs_trans_inode_buf(tp, ibp); | |
1712 | xfs_trans_log_buf(tp, ibp, offset, | |
1713 | (offset + sizeof(xfs_agino_t) - 1)); | |
1714 | xfs_inobp_check(mp, ibp); | |
1715 | } else { | |
1716 | xfs_trans_brelse(tp, ibp); | |
1717 | } | |
1718 | /* | |
1719 | * Point the previous inode on the list to the next inode. | |
1720 | */ | |
347d1c01 | 1721 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
1722 | ASSERT(next_agino != 0); |
1723 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
1724 | xfs_trans_inode_buf(tp, last_ibp); | |
1725 | xfs_trans_log_buf(tp, last_ibp, offset, | |
1726 | (offset + sizeof(xfs_agino_t) - 1)); | |
1727 | xfs_inobp_check(mp, last_ibp); | |
1728 | } | |
1729 | return 0; | |
1730 | } | |
1731 | ||
5b3eed75 DC |
1732 | /* |
1733 | * A big issue when freeing the inode cluster is is that we _cannot_ skip any | |
1734 | * inodes that are in memory - they all must be marked stale and attached to | |
1735 | * the cluster buffer. | |
1736 | */ | |
2a30f36d | 1737 | STATIC int |
1da177e4 LT |
1738 | xfs_ifree_cluster( |
1739 | xfs_inode_t *free_ip, | |
1740 | xfs_trans_t *tp, | |
1741 | xfs_ino_t inum) | |
1742 | { | |
1743 | xfs_mount_t *mp = free_ip->i_mount; | |
1744 | int blks_per_cluster; | |
1745 | int nbufs; | |
1746 | int ninodes; | |
5b257b4a | 1747 | int i, j; |
1da177e4 LT |
1748 | xfs_daddr_t blkno; |
1749 | xfs_buf_t *bp; | |
5b257b4a | 1750 | xfs_inode_t *ip; |
1da177e4 LT |
1751 | xfs_inode_log_item_t *iip; |
1752 | xfs_log_item_t *lip; | |
5017e97d | 1753 | struct xfs_perag *pag; |
1da177e4 | 1754 | |
5017e97d | 1755 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
1da177e4 LT |
1756 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { |
1757 | blks_per_cluster = 1; | |
1758 | ninodes = mp->m_sb.sb_inopblock; | |
1759 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
1760 | } else { | |
1761 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
1762 | mp->m_sb.sb_blocksize; | |
1763 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
1764 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
1765 | } | |
1766 | ||
1da177e4 LT |
1767 | for (j = 0; j < nbufs; j++, inum += ninodes) { |
1768 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
1769 | XFS_INO_TO_AGBNO(mp, inum)); | |
1770 | ||
5b257b4a DC |
1771 | /* |
1772 | * We obtain and lock the backing buffer first in the process | |
1773 | * here, as we have to ensure that any dirty inode that we | |
1774 | * can't get the flush lock on is attached to the buffer. | |
1775 | * If we scan the in-memory inodes first, then buffer IO can | |
1776 | * complete before we get a lock on it, and hence we may fail | |
1777 | * to mark all the active inodes on the buffer stale. | |
1778 | */ | |
1779 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
1780 | mp->m_bsize * blks_per_cluster, |
1781 | XBF_UNMAPPED); | |
5b257b4a | 1782 | |
2a30f36d CS |
1783 | if (!bp) |
1784 | return ENOMEM; | |
5b257b4a DC |
1785 | /* |
1786 | * Walk the inodes already attached to the buffer and mark them | |
1787 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
1788 | * in-memory inode walk can't lock them. By marking them all |
1789 | * stale first, we will not attempt to lock them in the loop | |
1790 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 1791 | */ |
adadbeef | 1792 | lip = bp->b_fspriv; |
5b257b4a DC |
1793 | while (lip) { |
1794 | if (lip->li_type == XFS_LI_INODE) { | |
1795 | iip = (xfs_inode_log_item_t *)lip; | |
1796 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 1797 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
1798 | xfs_trans_ail_copy_lsn(mp->m_ail, |
1799 | &iip->ili_flush_lsn, | |
1800 | &iip->ili_item.li_lsn); | |
1801 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a DC |
1802 | } |
1803 | lip = lip->li_bio_list; | |
1804 | } | |
1da177e4 | 1805 | |
5b3eed75 | 1806 | |
1da177e4 | 1807 | /* |
5b257b4a DC |
1808 | * For each inode in memory attempt to add it to the inode |
1809 | * buffer and set it up for being staled on buffer IO | |
1810 | * completion. This is safe as we've locked out tail pushing | |
1811 | * and flushing by locking the buffer. | |
1da177e4 | 1812 | * |
5b257b4a DC |
1813 | * We have already marked every inode that was part of a |
1814 | * transaction stale above, which means there is no point in | |
1815 | * even trying to lock them. | |
1da177e4 | 1816 | */ |
1da177e4 | 1817 | for (i = 0; i < ninodes; i++) { |
5b3eed75 | 1818 | retry: |
1a3e8f3d | 1819 | rcu_read_lock(); |
da353b0d DC |
1820 | ip = radix_tree_lookup(&pag->pag_ici_root, |
1821 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 1822 | |
1a3e8f3d DC |
1823 | /* Inode not in memory, nothing to do */ |
1824 | if (!ip) { | |
1825 | rcu_read_unlock(); | |
1da177e4 LT |
1826 | continue; |
1827 | } | |
1828 | ||
1a3e8f3d DC |
1829 | /* |
1830 | * because this is an RCU protected lookup, we could | |
1831 | * find a recently freed or even reallocated inode | |
1832 | * during the lookup. We need to check under the | |
1833 | * i_flags_lock for a valid inode here. Skip it if it | |
1834 | * is not valid, the wrong inode or stale. | |
1835 | */ | |
1836 | spin_lock(&ip->i_flags_lock); | |
1837 | if (ip->i_ino != inum + i || | |
1838 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
1839 | spin_unlock(&ip->i_flags_lock); | |
1840 | rcu_read_unlock(); | |
1841 | continue; | |
1842 | } | |
1843 | spin_unlock(&ip->i_flags_lock); | |
1844 | ||
5b3eed75 DC |
1845 | /* |
1846 | * Don't try to lock/unlock the current inode, but we | |
1847 | * _cannot_ skip the other inodes that we did not find | |
1848 | * in the list attached to the buffer and are not | |
1849 | * already marked stale. If we can't lock it, back off | |
1850 | * and retry. | |
1851 | */ | |
5b257b4a DC |
1852 | if (ip != free_ip && |
1853 | !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
1a3e8f3d | 1854 | rcu_read_unlock(); |
5b3eed75 DC |
1855 | delay(1); |
1856 | goto retry; | |
1da177e4 | 1857 | } |
1a3e8f3d | 1858 | rcu_read_unlock(); |
1da177e4 | 1859 | |
5b3eed75 | 1860 | xfs_iflock(ip); |
5b257b4a | 1861 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 1862 | |
5b3eed75 DC |
1863 | /* |
1864 | * we don't need to attach clean inodes or those only | |
1865 | * with unlogged changes (which we throw away, anyway). | |
1866 | */ | |
1da177e4 | 1867 | iip = ip->i_itemp; |
5b3eed75 | 1868 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 1869 | ASSERT(ip != free_ip); |
1da177e4 LT |
1870 | xfs_ifunlock(ip); |
1871 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1872 | continue; | |
1873 | } | |
1874 | ||
f5d8d5c4 CH |
1875 | iip->ili_last_fields = iip->ili_fields; |
1876 | iip->ili_fields = 0; | |
1da177e4 | 1877 | iip->ili_logged = 1; |
7b2e2a31 DC |
1878 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
1879 | &iip->ili_item.li_lsn); | |
1da177e4 | 1880 | |
ca30b2a7 CH |
1881 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
1882 | &iip->ili_item); | |
5b257b4a DC |
1883 | |
1884 | if (ip != free_ip) | |
1da177e4 | 1885 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1886 | } |
1887 | ||
5b3eed75 | 1888 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
1889 | xfs_trans_binval(tp, bp); |
1890 | } | |
1891 | ||
5017e97d | 1892 | xfs_perag_put(pag); |
2a30f36d | 1893 | return 0; |
1da177e4 LT |
1894 | } |
1895 | ||
1896 | /* | |
1897 | * This is called to return an inode to the inode free list. | |
1898 | * The inode should already be truncated to 0 length and have | |
1899 | * no pages associated with it. This routine also assumes that | |
1900 | * the inode is already a part of the transaction. | |
1901 | * | |
1902 | * The on-disk copy of the inode will have been added to the list | |
1903 | * of unlinked inodes in the AGI. We need to remove the inode from | |
1904 | * that list atomically with respect to freeing it here. | |
1905 | */ | |
1906 | int | |
1907 | xfs_ifree( | |
1908 | xfs_trans_t *tp, | |
1909 | xfs_inode_t *ip, | |
1910 | xfs_bmap_free_t *flist) | |
1911 | { | |
1912 | int error; | |
1913 | int delete; | |
1914 | xfs_ino_t first_ino; | |
c319b58b VA |
1915 | xfs_dinode_t *dip; |
1916 | xfs_buf_t *ibp; | |
1da177e4 | 1917 | |
579aa9ca | 1918 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
1919 | ASSERT(ip->i_d.di_nlink == 0); |
1920 | ASSERT(ip->i_d.di_nextents == 0); | |
1921 | ASSERT(ip->i_d.di_anextents == 0); | |
ce7ae151 | 1922 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); |
1da177e4 LT |
1923 | ASSERT(ip->i_d.di_nblocks == 0); |
1924 | ||
1925 | /* | |
1926 | * Pull the on-disk inode from the AGI unlinked list. | |
1927 | */ | |
1928 | error = xfs_iunlink_remove(tp, ip); | |
1929 | if (error != 0) { | |
1930 | return error; | |
1931 | } | |
1932 | ||
1933 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
1934 | if (error != 0) { | |
1935 | return error; | |
1936 | } | |
1937 | ip->i_d.di_mode = 0; /* mark incore inode as free */ | |
1938 | ip->i_d.di_flags = 0; | |
1939 | ip->i_d.di_dmevmask = 0; | |
1940 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
1941 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
1942 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1943 | /* | |
1944 | * Bump the generation count so no one will be confused | |
1945 | * by reincarnations of this inode. | |
1946 | */ | |
1947 | ip->i_d.di_gen++; | |
c319b58b | 1948 | |
1da177e4 LT |
1949 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1950 | ||
475ee413 CH |
1951 | error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &dip, &ibp, |
1952 | 0, 0); | |
c319b58b VA |
1953 | if (error) |
1954 | return error; | |
1955 | ||
1956 | /* | |
1957 | * Clear the on-disk di_mode. This is to prevent xfs_bulkstat | |
1958 | * from picking up this inode when it is reclaimed (its incore state | |
1959 | * initialzed but not flushed to disk yet). The in-core di_mode is | |
1960 | * already cleared and a corresponding transaction logged. | |
1961 | * The hack here just synchronizes the in-core to on-disk | |
1962 | * di_mode value in advance before the actual inode sync to disk. | |
1963 | * This is OK because the inode is already unlinked and would never | |
1964 | * change its di_mode again for this inode generation. | |
1965 | * This is a temporary hack that would require a proper fix | |
1966 | * in the future. | |
1967 | */ | |
81591fe2 | 1968 | dip->di_mode = 0; |
c319b58b | 1969 | |
1da177e4 | 1970 | if (delete) { |
2a30f36d | 1971 | error = xfs_ifree_cluster(ip, tp, first_ino); |
1da177e4 LT |
1972 | } |
1973 | ||
2a30f36d | 1974 | return error; |
1da177e4 LT |
1975 | } |
1976 | ||
1977 | /* | |
1978 | * Reallocate the space for if_broot based on the number of records | |
1979 | * being added or deleted as indicated in rec_diff. Move the records | |
1980 | * and pointers in if_broot to fit the new size. When shrinking this | |
1981 | * will eliminate holes between the records and pointers created by | |
1982 | * the caller. When growing this will create holes to be filled in | |
1983 | * by the caller. | |
1984 | * | |
1985 | * The caller must not request to add more records than would fit in | |
1986 | * the on-disk inode root. If the if_broot is currently NULL, then | |
1987 | * if we adding records one will be allocated. The caller must also | |
1988 | * not request that the number of records go below zero, although | |
1989 | * it can go to zero. | |
1990 | * | |
1991 | * ip -- the inode whose if_broot area is changing | |
1992 | * ext_diff -- the change in the number of records, positive or negative, | |
1993 | * requested for the if_broot array. | |
1994 | */ | |
1995 | void | |
1996 | xfs_iroot_realloc( | |
1997 | xfs_inode_t *ip, | |
1998 | int rec_diff, | |
1999 | int whichfork) | |
2000 | { | |
60197e8d | 2001 | struct xfs_mount *mp = ip->i_mount; |
1da177e4 LT |
2002 | int cur_max; |
2003 | xfs_ifork_t *ifp; | |
7cc95a82 | 2004 | struct xfs_btree_block *new_broot; |
1da177e4 LT |
2005 | int new_max; |
2006 | size_t new_size; | |
2007 | char *np; | |
2008 | char *op; | |
2009 | ||
2010 | /* | |
2011 | * Handle the degenerate case quietly. | |
2012 | */ | |
2013 | if (rec_diff == 0) { | |
2014 | return; | |
2015 | } | |
2016 | ||
2017 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2018 | if (rec_diff > 0) { | |
2019 | /* | |
2020 | * If there wasn't any memory allocated before, just | |
2021 | * allocate it now and get out. | |
2022 | */ | |
2023 | if (ifp->if_broot_bytes == 0) { | |
2024 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); | |
4a7edddc | 2025 | ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); |
1da177e4 LT |
2026 | ifp->if_broot_bytes = (int)new_size; |
2027 | return; | |
2028 | } | |
2029 | ||
2030 | /* | |
2031 | * If there is already an existing if_broot, then we need | |
2032 | * to realloc() it and shift the pointers to their new | |
2033 | * location. The records don't change location because | |
2034 | * they are kept butted up against the btree block header. | |
2035 | */ | |
60197e8d | 2036 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
1da177e4 LT |
2037 | new_max = cur_max + rec_diff; |
2038 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
7cc95a82 | 2039 | ifp->if_broot = kmem_realloc(ifp->if_broot, new_size, |
1da177e4 | 2040 | (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ |
4a7edddc | 2041 | KM_SLEEP | KM_NOFS); |
60197e8d CH |
2042 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
2043 | ifp->if_broot_bytes); | |
2044 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, | |
2045 | (int)new_size); | |
1da177e4 LT |
2046 | ifp->if_broot_bytes = (int)new_size; |
2047 | ASSERT(ifp->if_broot_bytes <= | |
2048 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2049 | memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); | |
2050 | return; | |
2051 | } | |
2052 | ||
2053 | /* | |
2054 | * rec_diff is less than 0. In this case, we are shrinking the | |
2055 | * if_broot buffer. It must already exist. If we go to zero | |
2056 | * records, just get rid of the root and clear the status bit. | |
2057 | */ | |
2058 | ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); | |
60197e8d | 2059 | cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); |
1da177e4 LT |
2060 | new_max = cur_max + rec_diff; |
2061 | ASSERT(new_max >= 0); | |
2062 | if (new_max > 0) | |
2063 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2064 | else | |
2065 | new_size = 0; | |
2066 | if (new_size > 0) { | |
4a7edddc | 2067 | new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); |
1da177e4 LT |
2068 | /* |
2069 | * First copy over the btree block header. | |
2070 | */ | |
7cc95a82 | 2071 | memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN); |
1da177e4 LT |
2072 | } else { |
2073 | new_broot = NULL; | |
2074 | ifp->if_flags &= ~XFS_IFBROOT; | |
2075 | } | |
2076 | ||
2077 | /* | |
2078 | * Only copy the records and pointers if there are any. | |
2079 | */ | |
2080 | if (new_max > 0) { | |
2081 | /* | |
2082 | * First copy the records. | |
2083 | */ | |
136341b4 CH |
2084 | op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1); |
2085 | np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1); | |
1da177e4 LT |
2086 | memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); |
2087 | ||
2088 | /* | |
2089 | * Then copy the pointers. | |
2090 | */ | |
60197e8d | 2091 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, |
1da177e4 | 2092 | ifp->if_broot_bytes); |
60197e8d | 2093 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1, |
1da177e4 LT |
2094 | (int)new_size); |
2095 | memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); | |
2096 | } | |
f0e2d93c | 2097 | kmem_free(ifp->if_broot); |
1da177e4 LT |
2098 | ifp->if_broot = new_broot; |
2099 | ifp->if_broot_bytes = (int)new_size; | |
2100 | ASSERT(ifp->if_broot_bytes <= | |
2101 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2102 | return; | |
2103 | } | |
2104 | ||
2105 | ||
1da177e4 LT |
2106 | /* |
2107 | * This is called when the amount of space needed for if_data | |
2108 | * is increased or decreased. The change in size is indicated by | |
2109 | * the number of bytes that need to be added or deleted in the | |
2110 | * byte_diff parameter. | |
2111 | * | |
2112 | * If the amount of space needed has decreased below the size of the | |
2113 | * inline buffer, then switch to using the inline buffer. Otherwise, | |
2114 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer | |
2115 | * to what is needed. | |
2116 | * | |
2117 | * ip -- the inode whose if_data area is changing | |
2118 | * byte_diff -- the change in the number of bytes, positive or negative, | |
2119 | * requested for the if_data array. | |
2120 | */ | |
2121 | void | |
2122 | xfs_idata_realloc( | |
2123 | xfs_inode_t *ip, | |
2124 | int byte_diff, | |
2125 | int whichfork) | |
2126 | { | |
2127 | xfs_ifork_t *ifp; | |
2128 | int new_size; | |
2129 | int real_size; | |
2130 | ||
2131 | if (byte_diff == 0) { | |
2132 | return; | |
2133 | } | |
2134 | ||
2135 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2136 | new_size = (int)ifp->if_bytes + byte_diff; | |
2137 | ASSERT(new_size >= 0); | |
2138 | ||
2139 | if (new_size == 0) { | |
2140 | if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
f0e2d93c | 2141 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2142 | } |
2143 | ifp->if_u1.if_data = NULL; | |
2144 | real_size = 0; | |
2145 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { | |
2146 | /* | |
2147 | * If the valid extents/data can fit in if_inline_ext/data, | |
2148 | * copy them from the malloc'd vector and free it. | |
2149 | */ | |
2150 | if (ifp->if_u1.if_data == NULL) { | |
2151 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2152 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2153 | ASSERT(ifp->if_real_bytes != 0); | |
2154 | memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, | |
2155 | new_size); | |
f0e2d93c | 2156 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2157 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; |
2158 | } | |
2159 | real_size = 0; | |
2160 | } else { | |
2161 | /* | |
2162 | * Stuck with malloc/realloc. | |
2163 | * For inline data, the underlying buffer must be | |
2164 | * a multiple of 4 bytes in size so that it can be | |
2165 | * logged and stay on word boundaries. We enforce | |
2166 | * that here. | |
2167 | */ | |
2168 | real_size = roundup(new_size, 4); | |
2169 | if (ifp->if_u1.if_data == NULL) { | |
2170 | ASSERT(ifp->if_real_bytes == 0); | |
4a7edddc DC |
2171 | ifp->if_u1.if_data = kmem_alloc(real_size, |
2172 | KM_SLEEP | KM_NOFS); | |
1da177e4 LT |
2173 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { |
2174 | /* | |
2175 | * Only do the realloc if the underlying size | |
2176 | * is really changing. | |
2177 | */ | |
2178 | if (ifp->if_real_bytes != real_size) { | |
2179 | ifp->if_u1.if_data = | |
2180 | kmem_realloc(ifp->if_u1.if_data, | |
2181 | real_size, | |
2182 | ifp->if_real_bytes, | |
4a7edddc | 2183 | KM_SLEEP | KM_NOFS); |
1da177e4 LT |
2184 | } |
2185 | } else { | |
2186 | ASSERT(ifp->if_real_bytes == 0); | |
4a7edddc DC |
2187 | ifp->if_u1.if_data = kmem_alloc(real_size, |
2188 | KM_SLEEP | KM_NOFS); | |
1da177e4 LT |
2189 | memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, |
2190 | ifp->if_bytes); | |
2191 | } | |
2192 | } | |
2193 | ifp->if_real_bytes = real_size; | |
2194 | ifp->if_bytes = new_size; | |
2195 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2196 | } | |
2197 | ||
1da177e4 LT |
2198 | void |
2199 | xfs_idestroy_fork( | |
2200 | xfs_inode_t *ip, | |
2201 | int whichfork) | |
2202 | { | |
2203 | xfs_ifork_t *ifp; | |
2204 | ||
2205 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2206 | if (ifp->if_broot != NULL) { | |
f0e2d93c | 2207 | kmem_free(ifp->if_broot); |
1da177e4 LT |
2208 | ifp->if_broot = NULL; |
2209 | } | |
2210 | ||
2211 | /* | |
2212 | * If the format is local, then we can't have an extents | |
2213 | * array so just look for an inline data array. If we're | |
2214 | * not local then we may or may not have an extents list, | |
2215 | * so check and free it up if we do. | |
2216 | */ | |
2217 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
2218 | if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && | |
2219 | (ifp->if_u1.if_data != NULL)) { | |
2220 | ASSERT(ifp->if_real_bytes != 0); | |
f0e2d93c | 2221 | kmem_free(ifp->if_u1.if_data); |
1da177e4 LT |
2222 | ifp->if_u1.if_data = NULL; |
2223 | ifp->if_real_bytes = 0; | |
2224 | } | |
2225 | } else if ((ifp->if_flags & XFS_IFEXTENTS) && | |
0293ce3a MK |
2226 | ((ifp->if_flags & XFS_IFEXTIREC) || |
2227 | ((ifp->if_u1.if_extents != NULL) && | |
2228 | (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { | |
1da177e4 | 2229 | ASSERT(ifp->if_real_bytes != 0); |
4eea22f0 | 2230 | xfs_iext_destroy(ifp); |
1da177e4 LT |
2231 | } |
2232 | ASSERT(ifp->if_u1.if_extents == NULL || | |
2233 | ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); | |
2234 | ASSERT(ifp->if_real_bytes == 0); | |
2235 | if (whichfork == XFS_ATTR_FORK) { | |
2236 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
2237 | ip->i_afp = NULL; | |
2238 | } | |
2239 | } | |
2240 | ||
1da177e4 | 2241 | /* |
60ec6783 CH |
2242 | * This is called to unpin an inode. The caller must have the inode locked |
2243 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2244 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2245 | */ |
60ec6783 | 2246 | static void |
f392e631 | 2247 | xfs_iunpin( |
60ec6783 | 2248 | struct xfs_inode *ip) |
1da177e4 | 2249 | { |
579aa9ca | 2250 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2251 | |
4aaf15d1 DC |
2252 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2253 | ||
a3f74ffb | 2254 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 2255 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 2256 | |
a3f74ffb | 2257 | } |
1da177e4 | 2258 | |
f392e631 CH |
2259 | static void |
2260 | __xfs_iunpin_wait( | |
2261 | struct xfs_inode *ip) | |
2262 | { | |
2263 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2264 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2265 | ||
2266 | xfs_iunpin(ip); | |
2267 | ||
2268 | do { | |
2269 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2270 | if (xfs_ipincount(ip)) | |
2271 | io_schedule(); | |
2272 | } while (xfs_ipincount(ip)); | |
2273 | finish_wait(wq, &wait.wait); | |
2274 | } | |
2275 | ||
777df5af | 2276 | void |
a3f74ffb | 2277 | xfs_iunpin_wait( |
60ec6783 | 2278 | struct xfs_inode *ip) |
a3f74ffb | 2279 | { |
f392e631 CH |
2280 | if (xfs_ipincount(ip)) |
2281 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2282 | } |
2283 | ||
1da177e4 LT |
2284 | /* |
2285 | * xfs_iextents_copy() | |
2286 | * | |
2287 | * This is called to copy the REAL extents (as opposed to the delayed | |
2288 | * allocation extents) from the inode into the given buffer. It | |
2289 | * returns the number of bytes copied into the buffer. | |
2290 | * | |
2291 | * If there are no delayed allocation extents, then we can just | |
2292 | * memcpy() the extents into the buffer. Otherwise, we need to | |
2293 | * examine each extent in turn and skip those which are delayed. | |
2294 | */ | |
2295 | int | |
2296 | xfs_iextents_copy( | |
2297 | xfs_inode_t *ip, | |
a6f64d4a | 2298 | xfs_bmbt_rec_t *dp, |
1da177e4 LT |
2299 | int whichfork) |
2300 | { | |
2301 | int copied; | |
1da177e4 LT |
2302 | int i; |
2303 | xfs_ifork_t *ifp; | |
2304 | int nrecs; | |
2305 | xfs_fsblock_t start_block; | |
2306 | ||
2307 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
579aa9ca | 2308 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 LT |
2309 | ASSERT(ifp->if_bytes > 0); |
2310 | ||
2311 | nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3a59c94c | 2312 | XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork); |
1da177e4 LT |
2313 | ASSERT(nrecs > 0); |
2314 | ||
2315 | /* | |
2316 | * There are some delayed allocation extents in the | |
2317 | * inode, so copy the extents one at a time and skip | |
2318 | * the delayed ones. There must be at least one | |
2319 | * non-delayed extent. | |
2320 | */ | |
1da177e4 LT |
2321 | copied = 0; |
2322 | for (i = 0; i < nrecs; i++) { | |
a6f64d4a | 2323 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
1da177e4 | 2324 | start_block = xfs_bmbt_get_startblock(ep); |
9d87c319 | 2325 | if (isnullstartblock(start_block)) { |
1da177e4 LT |
2326 | /* |
2327 | * It's a delayed allocation extent, so skip it. | |
2328 | */ | |
1da177e4 LT |
2329 | continue; |
2330 | } | |
2331 | ||
2332 | /* Translate to on disk format */ | |
cd8b0a97 CH |
2333 | put_unaligned(cpu_to_be64(ep->l0), &dp->l0); |
2334 | put_unaligned(cpu_to_be64(ep->l1), &dp->l1); | |
a6f64d4a | 2335 | dp++; |
1da177e4 LT |
2336 | copied++; |
2337 | } | |
2338 | ASSERT(copied != 0); | |
a6f64d4a | 2339 | xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
2340 | |
2341 | return (copied * (uint)sizeof(xfs_bmbt_rec_t)); | |
2342 | } | |
2343 | ||
2344 | /* | |
2345 | * Each of the following cases stores data into the same region | |
2346 | * of the on-disk inode, so only one of them can be valid at | |
2347 | * any given time. While it is possible to have conflicting formats | |
2348 | * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is | |
2349 | * in EXTENTS format, this can only happen when the fork has | |
2350 | * changed formats after being modified but before being flushed. | |
2351 | * In these cases, the format always takes precedence, because the | |
2352 | * format indicates the current state of the fork. | |
2353 | */ | |
2354 | /*ARGSUSED*/ | |
e4ac967b | 2355 | STATIC void |
1da177e4 LT |
2356 | xfs_iflush_fork( |
2357 | xfs_inode_t *ip, | |
2358 | xfs_dinode_t *dip, | |
2359 | xfs_inode_log_item_t *iip, | |
2360 | int whichfork, | |
2361 | xfs_buf_t *bp) | |
2362 | { | |
2363 | char *cp; | |
2364 | xfs_ifork_t *ifp; | |
2365 | xfs_mount_t *mp; | |
2366 | #ifdef XFS_TRANS_DEBUG | |
2367 | int first; | |
2368 | #endif | |
2369 | static const short brootflag[2] = | |
2370 | { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; | |
2371 | static const short dataflag[2] = | |
2372 | { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; | |
2373 | static const short extflag[2] = | |
2374 | { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; | |
2375 | ||
e4ac967b DC |
2376 | if (!iip) |
2377 | return; | |
1da177e4 LT |
2378 | ifp = XFS_IFORK_PTR(ip, whichfork); |
2379 | /* | |
2380 | * This can happen if we gave up in iformat in an error path, | |
2381 | * for the attribute fork. | |
2382 | */ | |
e4ac967b | 2383 | if (!ifp) { |
1da177e4 | 2384 | ASSERT(whichfork == XFS_ATTR_FORK); |
e4ac967b | 2385 | return; |
1da177e4 LT |
2386 | } |
2387 | cp = XFS_DFORK_PTR(dip, whichfork); | |
2388 | mp = ip->i_mount; | |
2389 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { | |
2390 | case XFS_DINODE_FMT_LOCAL: | |
f5d8d5c4 | 2391 | if ((iip->ili_fields & dataflag[whichfork]) && |
1da177e4 LT |
2392 | (ifp->if_bytes > 0)) { |
2393 | ASSERT(ifp->if_u1.if_data != NULL); | |
2394 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2395 | memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); | |
2396 | } | |
1da177e4 LT |
2397 | break; |
2398 | ||
2399 | case XFS_DINODE_FMT_EXTENTS: | |
2400 | ASSERT((ifp->if_flags & XFS_IFEXTENTS) || | |
f5d8d5c4 CH |
2401 | !(iip->ili_fields & extflag[whichfork])); |
2402 | if ((iip->ili_fields & extflag[whichfork]) && | |
1da177e4 | 2403 | (ifp->if_bytes > 0)) { |
ab1908a5 | 2404 | ASSERT(xfs_iext_get_ext(ifp, 0)); |
1da177e4 LT |
2405 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); |
2406 | (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, | |
2407 | whichfork); | |
2408 | } | |
2409 | break; | |
2410 | ||
2411 | case XFS_DINODE_FMT_BTREE: | |
f5d8d5c4 | 2412 | if ((iip->ili_fields & brootflag[whichfork]) && |
1da177e4 LT |
2413 | (ifp->if_broot_bytes > 0)) { |
2414 | ASSERT(ifp->if_broot != NULL); | |
2415 | ASSERT(ifp->if_broot_bytes <= | |
2416 | (XFS_IFORK_SIZE(ip, whichfork) + | |
2417 | XFS_BROOT_SIZE_ADJ)); | |
60197e8d | 2418 | xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes, |
1da177e4 LT |
2419 | (xfs_bmdr_block_t *)cp, |
2420 | XFS_DFORK_SIZE(dip, mp, whichfork)); | |
2421 | } | |
2422 | break; | |
2423 | ||
2424 | case XFS_DINODE_FMT_DEV: | |
f5d8d5c4 | 2425 | if (iip->ili_fields & XFS_ILOG_DEV) { |
1da177e4 | 2426 | ASSERT(whichfork == XFS_DATA_FORK); |
81591fe2 | 2427 | xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev); |
1da177e4 LT |
2428 | } |
2429 | break; | |
2430 | ||
2431 | case XFS_DINODE_FMT_UUID: | |
f5d8d5c4 | 2432 | if (iip->ili_fields & XFS_ILOG_UUID) { |
1da177e4 | 2433 | ASSERT(whichfork == XFS_DATA_FORK); |
81591fe2 CH |
2434 | memcpy(XFS_DFORK_DPTR(dip), |
2435 | &ip->i_df.if_u2.if_uuid, | |
2436 | sizeof(uuid_t)); | |
1da177e4 LT |
2437 | } |
2438 | break; | |
2439 | ||
2440 | default: | |
2441 | ASSERT(0); | |
2442 | break; | |
2443 | } | |
1da177e4 LT |
2444 | } |
2445 | ||
bad55843 DC |
2446 | STATIC int |
2447 | xfs_iflush_cluster( | |
2448 | xfs_inode_t *ip, | |
2449 | xfs_buf_t *bp) | |
2450 | { | |
2451 | xfs_mount_t *mp = ip->i_mount; | |
5017e97d | 2452 | struct xfs_perag *pag; |
bad55843 | 2453 | unsigned long first_index, mask; |
c8f5f12e | 2454 | unsigned long inodes_per_cluster; |
bad55843 DC |
2455 | int ilist_size; |
2456 | xfs_inode_t **ilist; | |
2457 | xfs_inode_t *iq; | |
bad55843 DC |
2458 | int nr_found; |
2459 | int clcount = 0; | |
2460 | int bufwasdelwri; | |
2461 | int i; | |
2462 | ||
5017e97d | 2463 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
bad55843 | 2464 | |
c8f5f12e DC |
2465 | inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; |
2466 | ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); | |
49383b0e | 2467 | ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); |
bad55843 | 2468 | if (!ilist) |
44b56e0a | 2469 | goto out_put; |
bad55843 DC |
2470 | |
2471 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); | |
2472 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; | |
1a3e8f3d | 2473 | rcu_read_lock(); |
bad55843 DC |
2474 | /* really need a gang lookup range call here */ |
2475 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, | |
c8f5f12e | 2476 | first_index, inodes_per_cluster); |
bad55843 DC |
2477 | if (nr_found == 0) |
2478 | goto out_free; | |
2479 | ||
2480 | for (i = 0; i < nr_found; i++) { | |
2481 | iq = ilist[i]; | |
2482 | if (iq == ip) | |
2483 | continue; | |
1a3e8f3d DC |
2484 | |
2485 | /* | |
2486 | * because this is an RCU protected lookup, we could find a | |
2487 | * recently freed or even reallocated inode during the lookup. | |
2488 | * We need to check under the i_flags_lock for a valid inode | |
2489 | * here. Skip it if it is not valid or the wrong inode. | |
2490 | */ | |
2491 | spin_lock(&ip->i_flags_lock); | |
2492 | if (!ip->i_ino || | |
2493 | (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { | |
2494 | spin_unlock(&ip->i_flags_lock); | |
2495 | continue; | |
2496 | } | |
2497 | spin_unlock(&ip->i_flags_lock); | |
2498 | ||
bad55843 DC |
2499 | /* |
2500 | * Do an un-protected check to see if the inode is dirty and | |
2501 | * is a candidate for flushing. These checks will be repeated | |
2502 | * later after the appropriate locks are acquired. | |
2503 | */ | |
33540408 | 2504 | if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) |
bad55843 | 2505 | continue; |
bad55843 DC |
2506 | |
2507 | /* | |
2508 | * Try to get locks. If any are unavailable or it is pinned, | |
2509 | * then this inode cannot be flushed and is skipped. | |
2510 | */ | |
2511 | ||
2512 | if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) | |
2513 | continue; | |
2514 | if (!xfs_iflock_nowait(iq)) { | |
2515 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2516 | continue; | |
2517 | } | |
2518 | if (xfs_ipincount(iq)) { | |
2519 | xfs_ifunlock(iq); | |
2520 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2521 | continue; | |
2522 | } | |
2523 | ||
2524 | /* | |
2525 | * arriving here means that this inode can be flushed. First | |
2526 | * re-check that it's dirty before flushing. | |
2527 | */ | |
33540408 DC |
2528 | if (!xfs_inode_clean(iq)) { |
2529 | int error; | |
bad55843 DC |
2530 | error = xfs_iflush_int(iq, bp); |
2531 | if (error) { | |
2532 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2533 | goto cluster_corrupt_out; | |
2534 | } | |
2535 | clcount++; | |
2536 | } else { | |
2537 | xfs_ifunlock(iq); | |
2538 | } | |
2539 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2540 | } | |
2541 | ||
2542 | if (clcount) { | |
2543 | XFS_STATS_INC(xs_icluster_flushcnt); | |
2544 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
2545 | } | |
2546 | ||
2547 | out_free: | |
1a3e8f3d | 2548 | rcu_read_unlock(); |
f0e2d93c | 2549 | kmem_free(ilist); |
44b56e0a DC |
2550 | out_put: |
2551 | xfs_perag_put(pag); | |
bad55843 DC |
2552 | return 0; |
2553 | ||
2554 | ||
2555 | cluster_corrupt_out: | |
2556 | /* | |
2557 | * Corruption detected in the clustering loop. Invalidate the | |
2558 | * inode buffer and shut down the filesystem. | |
2559 | */ | |
1a3e8f3d | 2560 | rcu_read_unlock(); |
bad55843 | 2561 | /* |
43ff2122 | 2562 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
2563 | * brelse can handle it with no problems. If not, shut down the |
2564 | * filesystem before releasing the buffer. | |
2565 | */ | |
43ff2122 | 2566 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
2567 | if (bufwasdelwri) |
2568 | xfs_buf_relse(bp); | |
2569 | ||
2570 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
2571 | ||
2572 | if (!bufwasdelwri) { | |
2573 | /* | |
2574 | * Just like incore_relse: if we have b_iodone functions, | |
2575 | * mark the buffer as an error and call them. Otherwise | |
2576 | * mark it as stale and brelse. | |
2577 | */ | |
cb669ca5 | 2578 | if (bp->b_iodone) { |
bad55843 | 2579 | XFS_BUF_UNDONE(bp); |
c867cb61 | 2580 | xfs_buf_stale(bp); |
5a52c2a5 | 2581 | xfs_buf_ioerror(bp, EIO); |
1a1a3e97 | 2582 | xfs_buf_ioend(bp, 0); |
bad55843 | 2583 | } else { |
c867cb61 | 2584 | xfs_buf_stale(bp); |
bad55843 DC |
2585 | xfs_buf_relse(bp); |
2586 | } | |
2587 | } | |
2588 | ||
2589 | /* | |
2590 | * Unlocks the flush lock | |
2591 | */ | |
04913fdd | 2592 | xfs_iflush_abort(iq, false); |
f0e2d93c | 2593 | kmem_free(ilist); |
44b56e0a | 2594 | xfs_perag_put(pag); |
bad55843 DC |
2595 | return XFS_ERROR(EFSCORRUPTED); |
2596 | } | |
2597 | ||
1da177e4 | 2598 | /* |
4c46819a CH |
2599 | * Flush dirty inode metadata into the backing buffer. |
2600 | * | |
2601 | * The caller must have the inode lock and the inode flush lock held. The | |
2602 | * inode lock will still be held upon return to the caller, and the inode | |
2603 | * flush lock will be released after the inode has reached the disk. | |
2604 | * | |
2605 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
2606 | */ |
2607 | int | |
2608 | xfs_iflush( | |
4c46819a CH |
2609 | struct xfs_inode *ip, |
2610 | struct xfs_buf **bpp) | |
1da177e4 | 2611 | { |
4c46819a CH |
2612 | struct xfs_mount *mp = ip->i_mount; |
2613 | struct xfs_buf *bp; | |
2614 | struct xfs_dinode *dip; | |
1da177e4 | 2615 | int error; |
1da177e4 LT |
2616 | |
2617 | XFS_STATS_INC(xs_iflush_count); | |
2618 | ||
579aa9ca | 2619 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 2620 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 2621 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 2622 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 2623 | |
4c46819a | 2624 | *bpp = NULL; |
1da177e4 | 2625 | |
1da177e4 LT |
2626 | xfs_iunpin_wait(ip); |
2627 | ||
4b6a4688 DC |
2628 | /* |
2629 | * For stale inodes we cannot rely on the backing buffer remaining | |
2630 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 2631 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
2632 | * inodes below. We have to check this after ensuring the inode is |
2633 | * unpinned so that it is safe to reclaim the stale inode after the | |
2634 | * flush call. | |
2635 | */ | |
2636 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
2637 | xfs_ifunlock(ip); | |
2638 | return 0; | |
2639 | } | |
2640 | ||
1da177e4 LT |
2641 | /* |
2642 | * This may have been unpinned because the filesystem is shutting | |
2643 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
2644 | * to disk, because the log record didn't make it to disk. |
2645 | * | |
2646 | * We also have to remove the log item from the AIL in this case, | |
2647 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
2648 | */ |
2649 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
32ce90a4 CH |
2650 | error = XFS_ERROR(EIO); |
2651 | goto abort_out; | |
1da177e4 LT |
2652 | } |
2653 | ||
a3f74ffb DC |
2654 | /* |
2655 | * Get the buffer containing the on-disk inode. | |
2656 | */ | |
475ee413 CH |
2657 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
2658 | 0); | |
a3f74ffb DC |
2659 | if (error || !bp) { |
2660 | xfs_ifunlock(ip); | |
2661 | return error; | |
2662 | } | |
2663 | ||
1da177e4 LT |
2664 | /* |
2665 | * First flush out the inode that xfs_iflush was called with. | |
2666 | */ | |
2667 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 2668 | if (error) |
1da177e4 | 2669 | goto corrupt_out; |
1da177e4 | 2670 | |
a3f74ffb DC |
2671 | /* |
2672 | * If the buffer is pinned then push on the log now so we won't | |
2673 | * get stuck waiting in the write for too long. | |
2674 | */ | |
811e64c7 | 2675 | if (xfs_buf_ispinned(bp)) |
a14a348b | 2676 | xfs_log_force(mp, 0); |
a3f74ffb | 2677 | |
1da177e4 LT |
2678 | /* |
2679 | * inode clustering: | |
2680 | * see if other inodes can be gathered into this write | |
2681 | */ | |
bad55843 DC |
2682 | error = xfs_iflush_cluster(ip, bp); |
2683 | if (error) | |
2684 | goto cluster_corrupt_out; | |
1da177e4 | 2685 | |
4c46819a CH |
2686 | *bpp = bp; |
2687 | return 0; | |
1da177e4 LT |
2688 | |
2689 | corrupt_out: | |
2690 | xfs_buf_relse(bp); | |
7d04a335 | 2691 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 2692 | cluster_corrupt_out: |
32ce90a4 CH |
2693 | error = XFS_ERROR(EFSCORRUPTED); |
2694 | abort_out: | |
1da177e4 LT |
2695 | /* |
2696 | * Unlocks the flush lock | |
2697 | */ | |
04913fdd | 2698 | xfs_iflush_abort(ip, false); |
32ce90a4 | 2699 | return error; |
1da177e4 LT |
2700 | } |
2701 | ||
2702 | ||
2703 | STATIC int | |
2704 | xfs_iflush_int( | |
2705 | xfs_inode_t *ip, | |
2706 | xfs_buf_t *bp) | |
2707 | { | |
2708 | xfs_inode_log_item_t *iip; | |
2709 | xfs_dinode_t *dip; | |
2710 | xfs_mount_t *mp; | |
2711 | #ifdef XFS_TRANS_DEBUG | |
2712 | int first; | |
2713 | #endif | |
1da177e4 | 2714 | |
579aa9ca | 2715 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 2716 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 2717 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 2718 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 LT |
2719 | |
2720 | iip = ip->i_itemp; | |
2721 | mp = ip->i_mount; | |
2722 | ||
1da177e4 | 2723 | /* set *dip = inode's place in the buffer */ |
92bfc6e7 | 2724 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 2725 | |
69ef921b | 2726 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
1da177e4 | 2727 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
6a19d939 DC |
2728 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2729 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
2730 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); | |
1da177e4 LT |
2731 | goto corrupt_out; |
2732 | } | |
2733 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
2734 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
6a19d939 DC |
2735 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2736 | "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
2737 | __func__, ip->i_ino, ip, ip->i_d.di_magic); | |
1da177e4 LT |
2738 | goto corrupt_out; |
2739 | } | |
abbede1b | 2740 | if (S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
2741 | if (XFS_TEST_ERROR( |
2742 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
2743 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
2744 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
6a19d939 DC |
2745 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2746 | "%s: Bad regular inode %Lu, ptr 0x%p", | |
2747 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
2748 | goto corrupt_out; |
2749 | } | |
abbede1b | 2750 | } else if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
2751 | if (XFS_TEST_ERROR( |
2752 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
2753 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
2754 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
2755 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
6a19d939 DC |
2756 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2757 | "%s: Bad directory inode %Lu, ptr 0x%p", | |
2758 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
2759 | goto corrupt_out; |
2760 | } | |
2761 | } | |
2762 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
2763 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
2764 | XFS_RANDOM_IFLUSH_5)) { | |
6a19d939 DC |
2765 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2766 | "%s: detected corrupt incore inode %Lu, " | |
2767 | "total extents = %d, nblocks = %Ld, ptr 0x%p", | |
2768 | __func__, ip->i_ino, | |
1da177e4 | 2769 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 2770 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
2771 | goto corrupt_out; |
2772 | } | |
2773 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
2774 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
6a19d939 DC |
2775 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2776 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
2777 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); | |
1da177e4 LT |
2778 | goto corrupt_out; |
2779 | } | |
2780 | /* | |
2781 | * bump the flush iteration count, used to detect flushes which | |
2782 | * postdate a log record during recovery. | |
2783 | */ | |
2784 | ||
2785 | ip->i_d.di_flushiter++; | |
2786 | ||
2787 | /* | |
2788 | * Copy the dirty parts of the inode into the on-disk | |
2789 | * inode. We always copy out the core of the inode, | |
2790 | * because if the inode is dirty at all the core must | |
2791 | * be. | |
2792 | */ | |
81591fe2 | 2793 | xfs_dinode_to_disk(dip, &ip->i_d); |
1da177e4 LT |
2794 | |
2795 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
2796 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
2797 | ip->i_d.di_flushiter = 0; | |
2798 | ||
2799 | /* | |
2800 | * If this is really an old format inode and the superblock version | |
2801 | * has not been updated to support only new format inodes, then | |
2802 | * convert back to the old inode format. If the superblock version | |
2803 | * has been updated, then make the conversion permanent. | |
2804 | */ | |
51ce16d5 CH |
2805 | ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); |
2806 | if (ip->i_d.di_version == 1) { | |
62118709 | 2807 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
1da177e4 LT |
2808 | /* |
2809 | * Convert it back. | |
2810 | */ | |
2811 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
81591fe2 | 2812 | dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink); |
1da177e4 LT |
2813 | } else { |
2814 | /* | |
2815 | * The superblock version has already been bumped, | |
2816 | * so just make the conversion to the new inode | |
2817 | * format permanent. | |
2818 | */ | |
51ce16d5 CH |
2819 | ip->i_d.di_version = 2; |
2820 | dip->di_version = 2; | |
1da177e4 | 2821 | ip->i_d.di_onlink = 0; |
81591fe2 | 2822 | dip->di_onlink = 0; |
1da177e4 | 2823 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
81591fe2 CH |
2824 | memset(&(dip->di_pad[0]), 0, |
2825 | sizeof(dip->di_pad)); | |
6743099c | 2826 | ASSERT(xfs_get_projid(ip) == 0); |
1da177e4 LT |
2827 | } |
2828 | } | |
2829 | ||
e4ac967b DC |
2830 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); |
2831 | if (XFS_IFORK_Q(ip)) | |
2832 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
1da177e4 LT |
2833 | xfs_inobp_check(mp, bp); |
2834 | ||
2835 | /* | |
f5d8d5c4 CH |
2836 | * We've recorded everything logged in the inode, so we'd like to clear |
2837 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
2838 | * However, we can't stop logging all this information until the data | |
2839 | * we've copied into the disk buffer is written to disk. If we did we | |
2840 | * might overwrite the copy of the inode in the log with all the data | |
2841 | * after re-logging only part of it, and in the face of a crash we | |
2842 | * wouldn't have all the data we need to recover. | |
1da177e4 | 2843 | * |
f5d8d5c4 CH |
2844 | * What we do is move the bits to the ili_last_fields field. When |
2845 | * logging the inode, these bits are moved back to the ili_fields field. | |
2846 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
2847 | * know that the information those bits represent is permanently on | |
2848 | * disk. As long as the flush completes before the inode is logged | |
2849 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 2850 | * |
f5d8d5c4 CH |
2851 | * We can play with the ili_fields bits here, because the inode lock |
2852 | * must be held exclusively in order to set bits there and the flush | |
2853 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
2854 | * done routine can tell whether or not to look in the AIL. Also, store | |
2855 | * the current LSN of the inode so that we can tell whether the item has | |
2856 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
2857 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
2858 | * atomically. | |
1da177e4 | 2859 | */ |
f5d8d5c4 CH |
2860 | if (iip != NULL && iip->ili_fields != 0) { |
2861 | iip->ili_last_fields = iip->ili_fields; | |
2862 | iip->ili_fields = 0; | |
1da177e4 LT |
2863 | iip->ili_logged = 1; |
2864 | ||
7b2e2a31 DC |
2865 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2866 | &iip->ili_item.li_lsn); | |
1da177e4 LT |
2867 | |
2868 | /* | |
2869 | * Attach the function xfs_iflush_done to the inode's | |
2870 | * buffer. This will remove the inode from the AIL | |
2871 | * and unlock the inode's flush lock when the inode is | |
2872 | * completely written to disk. | |
2873 | */ | |
ca30b2a7 | 2874 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); |
1da177e4 | 2875 | |
adadbeef | 2876 | ASSERT(bp->b_fspriv != NULL); |
cb669ca5 | 2877 | ASSERT(bp->b_iodone != NULL); |
1da177e4 LT |
2878 | } else { |
2879 | /* | |
2880 | * We're flushing an inode which is not in the AIL and has | |
8a9c9980 | 2881 | * not been logged. For this case we can immediately drop |
1da177e4 LT |
2882 | * the inode flush lock because we can avoid the whole |
2883 | * AIL state thing. It's OK to drop the flush lock now, | |
2884 | * because we've already locked the buffer and to do anything | |
2885 | * you really need both. | |
2886 | */ | |
2887 | if (iip != NULL) { | |
2888 | ASSERT(iip->ili_logged == 0); | |
2889 | ASSERT(iip->ili_last_fields == 0); | |
2890 | ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0); | |
2891 | } | |
2892 | xfs_ifunlock(ip); | |
2893 | } | |
2894 | ||
2895 | return 0; | |
2896 | ||
2897 | corrupt_out: | |
2898 | return XFS_ERROR(EFSCORRUPTED); | |
2899 | } | |
2900 | ||
4eea22f0 MK |
2901 | /* |
2902 | * Return a pointer to the extent record at file index idx. | |
2903 | */ | |
a6f64d4a | 2904 | xfs_bmbt_rec_host_t * |
4eea22f0 MK |
2905 | xfs_iext_get_ext( |
2906 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
2907 | xfs_extnum_t idx) /* index of target extent */ | |
2908 | { | |
2909 | ASSERT(idx >= 0); | |
87bef181 CH |
2910 | ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)); |
2911 | ||
0293ce3a MK |
2912 | if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { |
2913 | return ifp->if_u1.if_ext_irec->er_extbuf; | |
2914 | } else if (ifp->if_flags & XFS_IFEXTIREC) { | |
2915 | xfs_ext_irec_t *erp; /* irec pointer */ | |
2916 | int erp_idx = 0; /* irec index */ | |
2917 | xfs_extnum_t page_idx = idx; /* ext index in target list */ | |
2918 | ||
2919 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
2920 | return &erp->er_extbuf[page_idx]; | |
2921 | } else if (ifp->if_bytes) { | |
4eea22f0 MK |
2922 | return &ifp->if_u1.if_extents[idx]; |
2923 | } else { | |
2924 | return NULL; | |
2925 | } | |
2926 | } | |
2927 | ||
2928 | /* | |
2929 | * Insert new item(s) into the extent records for incore inode | |
2930 | * fork 'ifp'. 'count' new items are inserted at index 'idx'. | |
2931 | */ | |
2932 | void | |
2933 | xfs_iext_insert( | |
6ef35544 | 2934 | xfs_inode_t *ip, /* incore inode pointer */ |
4eea22f0 MK |
2935 | xfs_extnum_t idx, /* starting index of new items */ |
2936 | xfs_extnum_t count, /* number of inserted items */ | |
6ef35544 CH |
2937 | xfs_bmbt_irec_t *new, /* items to insert */ |
2938 | int state) /* type of extent conversion */ | |
4eea22f0 | 2939 | { |
6ef35544 | 2940 | xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df; |
4eea22f0 MK |
2941 | xfs_extnum_t i; /* extent record index */ |
2942 | ||
0b1b213f CH |
2943 | trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_); |
2944 | ||
4eea22f0 MK |
2945 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
2946 | xfs_iext_add(ifp, idx, count); | |
a6f64d4a CH |
2947 | for (i = idx; i < idx + count; i++, new++) |
2948 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new); | |
4eea22f0 MK |
2949 | } |
2950 | ||
2951 | /* | |
2952 | * This is called when the amount of space required for incore file | |
2953 | * extents needs to be increased. The ext_diff parameter stores the | |
2954 | * number of new extents being added and the idx parameter contains | |
2955 | * the extent index where the new extents will be added. If the new | |
2956 | * extents are being appended, then we just need to (re)allocate and | |
2957 | * initialize the space. Otherwise, if the new extents are being | |
2958 | * inserted into the middle of the existing entries, a bit more work | |
2959 | * is required to make room for the new extents to be inserted. The | |
2960 | * caller is responsible for filling in the new extent entries upon | |
2961 | * return. | |
2962 | */ | |
2963 | void | |
2964 | xfs_iext_add( | |
2965 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
2966 | xfs_extnum_t idx, /* index to begin adding exts */ | |
c41564b5 | 2967 | int ext_diff) /* number of extents to add */ |
4eea22f0 MK |
2968 | { |
2969 | int byte_diff; /* new bytes being added */ | |
2970 | int new_size; /* size of extents after adding */ | |
2971 | xfs_extnum_t nextents; /* number of extents in file */ | |
2972 | ||
2973 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
2974 | ASSERT((idx >= 0) && (idx <= nextents)); | |
2975 | byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); | |
2976 | new_size = ifp->if_bytes + byte_diff; | |
2977 | /* | |
2978 | * If the new number of extents (nextents + ext_diff) | |
2979 | * fits inside the inode, then continue to use the inline | |
2980 | * extent buffer. | |
2981 | */ | |
2982 | if (nextents + ext_diff <= XFS_INLINE_EXTS) { | |
2983 | if (idx < nextents) { | |
2984 | memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], | |
2985 | &ifp->if_u2.if_inline_ext[idx], | |
2986 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
2987 | memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); | |
2988 | } | |
2989 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
2990 | ifp->if_real_bytes = 0; | |
2991 | } | |
2992 | /* | |
2993 | * Otherwise use a linear (direct) extent list. | |
2994 | * If the extents are currently inside the inode, | |
2995 | * xfs_iext_realloc_direct will switch us from | |
2996 | * inline to direct extent allocation mode. | |
2997 | */ | |
0293ce3a | 2998 | else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { |
4eea22f0 MK |
2999 | xfs_iext_realloc_direct(ifp, new_size); |
3000 | if (idx < nextents) { | |
3001 | memmove(&ifp->if_u1.if_extents[idx + ext_diff], | |
3002 | &ifp->if_u1.if_extents[idx], | |
3003 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
3004 | memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); | |
3005 | } | |
3006 | } | |
0293ce3a MK |
3007 | /* Indirection array */ |
3008 | else { | |
3009 | xfs_ext_irec_t *erp; | |
3010 | int erp_idx = 0; | |
3011 | int page_idx = idx; | |
3012 | ||
3013 | ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); | |
3014 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3015 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); | |
3016 | } else { | |
3017 | xfs_iext_irec_init(ifp); | |
3018 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3019 | erp = ifp->if_u1.if_ext_irec; | |
3020 | } | |
3021 | /* Extents fit in target extent page */ | |
3022 | if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { | |
3023 | if (page_idx < erp->er_extcount) { | |
3024 | memmove(&erp->er_extbuf[page_idx + ext_diff], | |
3025 | &erp->er_extbuf[page_idx], | |
3026 | (erp->er_extcount - page_idx) * | |
3027 | sizeof(xfs_bmbt_rec_t)); | |
3028 | memset(&erp->er_extbuf[page_idx], 0, byte_diff); | |
3029 | } | |
3030 | erp->er_extcount += ext_diff; | |
3031 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3032 | } | |
3033 | /* Insert a new extent page */ | |
3034 | else if (erp) { | |
3035 | xfs_iext_add_indirect_multi(ifp, | |
3036 | erp_idx, page_idx, ext_diff); | |
3037 | } | |
3038 | /* | |
3039 | * If extent(s) are being appended to the last page in | |
3040 | * the indirection array and the new extent(s) don't fit | |
3041 | * in the page, then erp is NULL and erp_idx is set to | |
3042 | * the next index needed in the indirection array. | |
3043 | */ | |
3044 | else { | |
3045 | int count = ext_diff; | |
3046 | ||
3047 | while (count) { | |
3048 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3049 | erp->er_extcount = count; | |
3050 | count -= MIN(count, (int)XFS_LINEAR_EXTS); | |
3051 | if (count) { | |
3052 | erp_idx++; | |
3053 | } | |
3054 | } | |
3055 | } | |
3056 | } | |
4eea22f0 MK |
3057 | ifp->if_bytes = new_size; |
3058 | } | |
3059 | ||
0293ce3a MK |
3060 | /* |
3061 | * This is called when incore extents are being added to the indirection | |
3062 | * array and the new extents do not fit in the target extent list. The | |
3063 | * erp_idx parameter contains the irec index for the target extent list | |
3064 | * in the indirection array, and the idx parameter contains the extent | |
3065 | * index within the list. The number of extents being added is stored | |
3066 | * in the count parameter. | |
3067 | * | |
3068 | * |-------| |-------| | |
3069 | * | | | | idx - number of extents before idx | |
3070 | * | idx | | count | | |
3071 | * | | | | count - number of extents being inserted at idx | |
3072 | * |-------| |-------| | |
3073 | * | count | | nex2 | nex2 - number of extents after idx + count | |
3074 | * |-------| |-------| | |
3075 | */ | |
3076 | void | |
3077 | xfs_iext_add_indirect_multi( | |
3078 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3079 | int erp_idx, /* target extent irec index */ | |
3080 | xfs_extnum_t idx, /* index within target list */ | |
3081 | int count) /* new extents being added */ | |
3082 | { | |
3083 | int byte_diff; /* new bytes being added */ | |
3084 | xfs_ext_irec_t *erp; /* pointer to irec entry */ | |
3085 | xfs_extnum_t ext_diff; /* number of extents to add */ | |
3086 | xfs_extnum_t ext_cnt; /* new extents still needed */ | |
3087 | xfs_extnum_t nex2; /* extents after idx + count */ | |
3088 | xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ | |
3089 | int nlists; /* number of irec's (lists) */ | |
3090 | ||
3091 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3092 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3093 | nex2 = erp->er_extcount - idx; | |
3094 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3095 | ||
3096 | /* | |
3097 | * Save second part of target extent list | |
3098 | * (all extents past */ | |
3099 | if (nex2) { | |
3100 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
6785073b | 3101 | nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS); |
0293ce3a MK |
3102 | memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); |
3103 | erp->er_extcount -= nex2; | |
3104 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); | |
3105 | memset(&erp->er_extbuf[idx], 0, byte_diff); | |
3106 | } | |
3107 | ||
3108 | /* | |
3109 | * Add the new extents to the end of the target | |
3110 | * list, then allocate new irec record(s) and | |
3111 | * extent buffer(s) as needed to store the rest | |
3112 | * of the new extents. | |
3113 | */ | |
3114 | ext_cnt = count; | |
3115 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); | |
3116 | if (ext_diff) { | |
3117 | erp->er_extcount += ext_diff; | |
3118 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3119 | ext_cnt -= ext_diff; | |
3120 | } | |
3121 | while (ext_cnt) { | |
3122 | erp_idx++; | |
3123 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3124 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); | |
3125 | erp->er_extcount = ext_diff; | |
3126 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3127 | ext_cnt -= ext_diff; | |
3128 | } | |
3129 | ||
3130 | /* Add nex2 extents back to indirection array */ | |
3131 | if (nex2) { | |
3132 | xfs_extnum_t ext_avail; | |
3133 | int i; | |
3134 | ||
3135 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
3136 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; | |
3137 | i = 0; | |
3138 | /* | |
3139 | * If nex2 extents fit in the current page, append | |
3140 | * nex2_ep after the new extents. | |
3141 | */ | |
3142 | if (nex2 <= ext_avail) { | |
3143 | i = erp->er_extcount; | |
3144 | } | |
3145 | /* | |
3146 | * Otherwise, check if space is available in the | |
3147 | * next page. | |
3148 | */ | |
3149 | else if ((erp_idx < nlists - 1) && | |
3150 | (nex2 <= (ext_avail = XFS_LINEAR_EXTS - | |
3151 | ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { | |
3152 | erp_idx++; | |
3153 | erp++; | |
3154 | /* Create a hole for nex2 extents */ | |
3155 | memmove(&erp->er_extbuf[nex2], erp->er_extbuf, | |
3156 | erp->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
3157 | } | |
3158 | /* | |
3159 | * Final choice, create a new extent page for | |
3160 | * nex2 extents. | |
3161 | */ | |
3162 | else { | |
3163 | erp_idx++; | |
3164 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3165 | } | |
3166 | memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); | |
f0e2d93c | 3167 | kmem_free(nex2_ep); |
0293ce3a MK |
3168 | erp->er_extcount += nex2; |
3169 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); | |
3170 | } | |
3171 | } | |
3172 | ||
4eea22f0 MK |
3173 | /* |
3174 | * This is called when the amount of space required for incore file | |
3175 | * extents needs to be decreased. The ext_diff parameter stores the | |
3176 | * number of extents to be removed and the idx parameter contains | |
3177 | * the extent index where the extents will be removed from. | |
0293ce3a MK |
3178 | * |
3179 | * If the amount of space needed has decreased below the linear | |
3180 | * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous | |
3181 | * extent array. Otherwise, use kmem_realloc() to adjust the | |
3182 | * size to what is needed. | |
4eea22f0 MK |
3183 | */ |
3184 | void | |
3185 | xfs_iext_remove( | |
6ef35544 | 3186 | xfs_inode_t *ip, /* incore inode pointer */ |
4eea22f0 | 3187 | xfs_extnum_t idx, /* index to begin removing exts */ |
6ef35544 CH |
3188 | int ext_diff, /* number of extents to remove */ |
3189 | int state) /* type of extent conversion */ | |
4eea22f0 | 3190 | { |
6ef35544 | 3191 | xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df; |
4eea22f0 MK |
3192 | xfs_extnum_t nextents; /* number of extents in file */ |
3193 | int new_size; /* size of extents after removal */ | |
3194 | ||
0b1b213f CH |
3195 | trace_xfs_iext_remove(ip, idx, state, _RET_IP_); |
3196 | ||
4eea22f0 MK |
3197 | ASSERT(ext_diff > 0); |
3198 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3199 | new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); | |
3200 | ||
3201 | if (new_size == 0) { | |
3202 | xfs_iext_destroy(ifp); | |
0293ce3a MK |
3203 | } else if (ifp->if_flags & XFS_IFEXTIREC) { |
3204 | xfs_iext_remove_indirect(ifp, idx, ext_diff); | |
4eea22f0 MK |
3205 | } else if (ifp->if_real_bytes) { |
3206 | xfs_iext_remove_direct(ifp, idx, ext_diff); | |
3207 | } else { | |
3208 | xfs_iext_remove_inline(ifp, idx, ext_diff); | |
3209 | } | |
3210 | ifp->if_bytes = new_size; | |
3211 | } | |
3212 | ||
3213 | /* | |
3214 | * This removes ext_diff extents from the inline buffer, beginning | |
3215 | * at extent index idx. | |
3216 | */ | |
3217 | void | |
3218 | xfs_iext_remove_inline( | |
3219 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3220 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3221 | int ext_diff) /* number of extents to remove */ | |
3222 | { | |
3223 | int nextents; /* number of extents in file */ | |
3224 | ||
0293ce3a | 3225 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
3226 | ASSERT(idx < XFS_INLINE_EXTS); |
3227 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3228 | ASSERT(((nextents - ext_diff) > 0) && | |
3229 | (nextents - ext_diff) < XFS_INLINE_EXTS); | |
3230 | ||
3231 | if (idx + ext_diff < nextents) { | |
3232 | memmove(&ifp->if_u2.if_inline_ext[idx], | |
3233 | &ifp->if_u2.if_inline_ext[idx + ext_diff], | |
3234 | (nextents - (idx + ext_diff)) * | |
3235 | sizeof(xfs_bmbt_rec_t)); | |
3236 | memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], | |
3237 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3238 | } else { | |
3239 | memset(&ifp->if_u2.if_inline_ext[idx], 0, | |
3240 | ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3241 | } | |
3242 | } | |
3243 | ||
3244 | /* | |
3245 | * This removes ext_diff extents from a linear (direct) extent list, | |
3246 | * beginning at extent index idx. If the extents are being removed | |
3247 | * from the end of the list (ie. truncate) then we just need to re- | |
3248 | * allocate the list to remove the extra space. Otherwise, if the | |
3249 | * extents are being removed from the middle of the existing extent | |
3250 | * entries, then we first need to move the extent records beginning | |
3251 | * at idx + ext_diff up in the list to overwrite the records being | |
3252 | * removed, then remove the extra space via kmem_realloc. | |
3253 | */ | |
3254 | void | |
3255 | xfs_iext_remove_direct( | |
3256 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3257 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3258 | int ext_diff) /* number of extents to remove */ | |
3259 | { | |
3260 | xfs_extnum_t nextents; /* number of extents in file */ | |
3261 | int new_size; /* size of extents after removal */ | |
3262 | ||
0293ce3a | 3263 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
3264 | new_size = ifp->if_bytes - |
3265 | (ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3266 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3267 | ||
3268 | if (new_size == 0) { | |
3269 | xfs_iext_destroy(ifp); | |
3270 | return; | |
3271 | } | |
3272 | /* Move extents up in the list (if needed) */ | |
3273 | if (idx + ext_diff < nextents) { | |
3274 | memmove(&ifp->if_u1.if_extents[idx], | |
3275 | &ifp->if_u1.if_extents[idx + ext_diff], | |
3276 | (nextents - (idx + ext_diff)) * | |
3277 | sizeof(xfs_bmbt_rec_t)); | |
3278 | } | |
3279 | memset(&ifp->if_u1.if_extents[nextents - ext_diff], | |
3280 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3281 | /* | |
3282 | * Reallocate the direct extent list. If the extents | |
3283 | * will fit inside the inode then xfs_iext_realloc_direct | |
3284 | * will switch from direct to inline extent allocation | |
3285 | * mode for us. | |
3286 | */ | |
3287 | xfs_iext_realloc_direct(ifp, new_size); | |
3288 | ifp->if_bytes = new_size; | |
3289 | } | |
3290 | ||
0293ce3a MK |
3291 | /* |
3292 | * This is called when incore extents are being removed from the | |
3293 | * indirection array and the extents being removed span multiple extent | |
3294 | * buffers. The idx parameter contains the file extent index where we | |
3295 | * want to begin removing extents, and the count parameter contains | |
3296 | * how many extents need to be removed. | |
3297 | * | |
3298 | * |-------| |-------| | |
3299 | * | nex1 | | | nex1 - number of extents before idx | |
3300 | * |-------| | count | | |
3301 | * | | | | count - number of extents being removed at idx | |
3302 | * | count | |-------| | |
3303 | * | | | nex2 | nex2 - number of extents after idx + count | |
3304 | * |-------| |-------| | |
3305 | */ | |
3306 | void | |
3307 | xfs_iext_remove_indirect( | |
3308 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3309 | xfs_extnum_t idx, /* index to begin removing extents */ | |
3310 | int count) /* number of extents to remove */ | |
3311 | { | |
3312 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3313 | int erp_idx = 0; /* indirection array index */ | |
3314 | xfs_extnum_t ext_cnt; /* extents left to remove */ | |
3315 | xfs_extnum_t ext_diff; /* extents to remove in current list */ | |
3316 | xfs_extnum_t nex1; /* number of extents before idx */ | |
3317 | xfs_extnum_t nex2; /* extents after idx + count */ | |
0293ce3a MK |
3318 | int page_idx = idx; /* index in target extent list */ |
3319 | ||
3320 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3321 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
3322 | ASSERT(erp != NULL); | |
0293ce3a MK |
3323 | nex1 = page_idx; |
3324 | ext_cnt = count; | |
3325 | while (ext_cnt) { | |
3326 | nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); | |
3327 | ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); | |
3328 | /* | |
3329 | * Check for deletion of entire list; | |
3330 | * xfs_iext_irec_remove() updates extent offsets. | |
3331 | */ | |
3332 | if (ext_diff == erp->er_extcount) { | |
3333 | xfs_iext_irec_remove(ifp, erp_idx); | |
3334 | ext_cnt -= ext_diff; | |
3335 | nex1 = 0; | |
3336 | if (ext_cnt) { | |
3337 | ASSERT(erp_idx < ifp->if_real_bytes / | |
3338 | XFS_IEXT_BUFSZ); | |
3339 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3340 | nex1 = 0; | |
3341 | continue; | |
3342 | } else { | |
3343 | break; | |
3344 | } | |
3345 | } | |
3346 | /* Move extents up (if needed) */ | |
3347 | if (nex2) { | |
3348 | memmove(&erp->er_extbuf[nex1], | |
3349 | &erp->er_extbuf[nex1 + ext_diff], | |
3350 | nex2 * sizeof(xfs_bmbt_rec_t)); | |
3351 | } | |
3352 | /* Zero out rest of page */ | |
3353 | memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - | |
3354 | ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); | |
3355 | /* Update remaining counters */ | |
3356 | erp->er_extcount -= ext_diff; | |
3357 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); | |
3358 | ext_cnt -= ext_diff; | |
3359 | nex1 = 0; | |
3360 | erp_idx++; | |
3361 | erp++; | |
3362 | } | |
3363 | ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); | |
3364 | xfs_iext_irec_compact(ifp); | |
3365 | } | |
3366 | ||
4eea22f0 MK |
3367 | /* |
3368 | * Create, destroy, or resize a linear (direct) block of extents. | |
3369 | */ | |
3370 | void | |
3371 | xfs_iext_realloc_direct( | |
3372 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3373 | int new_size) /* new size of extents */ | |
3374 | { | |
3375 | int rnew_size; /* real new size of extents */ | |
3376 | ||
3377 | rnew_size = new_size; | |
3378 | ||
0293ce3a MK |
3379 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || |
3380 | ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && | |
3381 | (new_size != ifp->if_real_bytes))); | |
3382 | ||
4eea22f0 MK |
3383 | /* Free extent records */ |
3384 | if (new_size == 0) { | |
3385 | xfs_iext_destroy(ifp); | |
3386 | } | |
3387 | /* Resize direct extent list and zero any new bytes */ | |
3388 | else if (ifp->if_real_bytes) { | |
3389 | /* Check if extents will fit inside the inode */ | |
3390 | if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { | |
3391 | xfs_iext_direct_to_inline(ifp, new_size / | |
3392 | (uint)sizeof(xfs_bmbt_rec_t)); | |
3393 | ifp->if_bytes = new_size; | |
3394 | return; | |
3395 | } | |
16a087d8 | 3396 | if (!is_power_of_2(new_size)){ |
40ebd81d | 3397 | rnew_size = roundup_pow_of_two(new_size); |
4eea22f0 MK |
3398 | } |
3399 | if (rnew_size != ifp->if_real_bytes) { | |
a6f64d4a | 3400 | ifp->if_u1.if_extents = |
4eea22f0 MK |
3401 | kmem_realloc(ifp->if_u1.if_extents, |
3402 | rnew_size, | |
6785073b | 3403 | ifp->if_real_bytes, KM_NOFS); |
4eea22f0 MK |
3404 | } |
3405 | if (rnew_size > ifp->if_real_bytes) { | |
3406 | memset(&ifp->if_u1.if_extents[ifp->if_bytes / | |
3407 | (uint)sizeof(xfs_bmbt_rec_t)], 0, | |
3408 | rnew_size - ifp->if_real_bytes); | |
3409 | } | |
3410 | } | |
3411 | /* | |
3412 | * Switch from the inline extent buffer to a direct | |
3413 | * extent list. Be sure to include the inline extent | |
3414 | * bytes in new_size. | |
3415 | */ | |
3416 | else { | |
3417 | new_size += ifp->if_bytes; | |
16a087d8 | 3418 | if (!is_power_of_2(new_size)) { |
40ebd81d | 3419 | rnew_size = roundup_pow_of_two(new_size); |
4eea22f0 MK |
3420 | } |
3421 | xfs_iext_inline_to_direct(ifp, rnew_size); | |
3422 | } | |
3423 | ifp->if_real_bytes = rnew_size; | |
3424 | ifp->if_bytes = new_size; | |
3425 | } | |
3426 | ||
3427 | /* | |
3428 | * Switch from linear (direct) extent records to inline buffer. | |
3429 | */ | |
3430 | void | |
3431 | xfs_iext_direct_to_inline( | |
3432 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3433 | xfs_extnum_t nextents) /* number of extents in file */ | |
3434 | { | |
3435 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
3436 | ASSERT(nextents <= XFS_INLINE_EXTS); | |
3437 | /* | |
3438 | * The inline buffer was zeroed when we switched | |
3439 | * from inline to direct extent allocation mode, | |
3440 | * so we don't need to clear it here. | |
3441 | */ | |
3442 | memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, | |
3443 | nextents * sizeof(xfs_bmbt_rec_t)); | |
f0e2d93c | 3444 | kmem_free(ifp->if_u1.if_extents); |
4eea22f0 MK |
3445 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; |
3446 | ifp->if_real_bytes = 0; | |
3447 | } | |
3448 | ||
3449 | /* | |
3450 | * Switch from inline buffer to linear (direct) extent records. | |
3451 | * new_size should already be rounded up to the next power of 2 | |
3452 | * by the caller (when appropriate), so use new_size as it is. | |
3453 | * However, since new_size may be rounded up, we can't update | |
3454 | * if_bytes here. It is the caller's responsibility to update | |
3455 | * if_bytes upon return. | |
3456 | */ | |
3457 | void | |
3458 | xfs_iext_inline_to_direct( | |
3459 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3460 | int new_size) /* number of extents in file */ | |
3461 | { | |
6785073b | 3462 | ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS); |
4eea22f0 MK |
3463 | memset(ifp->if_u1.if_extents, 0, new_size); |
3464 | if (ifp->if_bytes) { | |
3465 | memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, | |
3466 | ifp->if_bytes); | |
3467 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
3468 | sizeof(xfs_bmbt_rec_t)); | |
3469 | } | |
3470 | ifp->if_real_bytes = new_size; | |
3471 | } | |
3472 | ||
0293ce3a MK |
3473 | /* |
3474 | * Resize an extent indirection array to new_size bytes. | |
3475 | */ | |
d96f8f89 | 3476 | STATIC void |
0293ce3a MK |
3477 | xfs_iext_realloc_indirect( |
3478 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3479 | int new_size) /* new indirection array size */ | |
3480 | { | |
3481 | int nlists; /* number of irec's (ex lists) */ | |
3482 | int size; /* current indirection array size */ | |
3483 | ||
3484 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3485 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3486 | size = nlists * sizeof(xfs_ext_irec_t); | |
3487 | ASSERT(ifp->if_real_bytes); | |
3488 | ASSERT((new_size >= 0) && (new_size != size)); | |
3489 | if (new_size == 0) { | |
3490 | xfs_iext_destroy(ifp); | |
3491 | } else { | |
3492 | ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) | |
3493 | kmem_realloc(ifp->if_u1.if_ext_irec, | |
6785073b | 3494 | new_size, size, KM_NOFS); |
0293ce3a MK |
3495 | } |
3496 | } | |
3497 | ||
3498 | /* | |
3499 | * Switch from indirection array to linear (direct) extent allocations. | |
3500 | */ | |
d96f8f89 | 3501 | STATIC void |
0293ce3a MK |
3502 | xfs_iext_indirect_to_direct( |
3503 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3504 | { | |
a6f64d4a | 3505 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ |
0293ce3a MK |
3506 | xfs_extnum_t nextents; /* number of extents in file */ |
3507 | int size; /* size of file extents */ | |
3508 | ||
3509 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3510 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3511 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
3512 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
3513 | ||
71a8c87f | 3514 | xfs_iext_irec_compact_pages(ifp); |
0293ce3a MK |
3515 | ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); |
3516 | ||
3517 | ep = ifp->if_u1.if_ext_irec->er_extbuf; | |
f0e2d93c | 3518 | kmem_free(ifp->if_u1.if_ext_irec); |
0293ce3a MK |
3519 | ifp->if_flags &= ~XFS_IFEXTIREC; |
3520 | ifp->if_u1.if_extents = ep; | |
3521 | ifp->if_bytes = size; | |
3522 | if (nextents < XFS_LINEAR_EXTS) { | |
3523 | xfs_iext_realloc_direct(ifp, size); | |
3524 | } | |
3525 | } | |
3526 | ||
4eea22f0 MK |
3527 | /* |
3528 | * Free incore file extents. | |
3529 | */ | |
3530 | void | |
3531 | xfs_iext_destroy( | |
3532 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3533 | { | |
0293ce3a MK |
3534 | if (ifp->if_flags & XFS_IFEXTIREC) { |
3535 | int erp_idx; | |
3536 | int nlists; | |
3537 | ||
3538 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3539 | for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { | |
3540 | xfs_iext_irec_remove(ifp, erp_idx); | |
3541 | } | |
3542 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
3543 | } else if (ifp->if_real_bytes) { | |
f0e2d93c | 3544 | kmem_free(ifp->if_u1.if_extents); |
4eea22f0 MK |
3545 | } else if (ifp->if_bytes) { |
3546 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
3547 | sizeof(xfs_bmbt_rec_t)); | |
3548 | } | |
3549 | ifp->if_u1.if_extents = NULL; | |
3550 | ifp->if_real_bytes = 0; | |
3551 | ifp->if_bytes = 0; | |
3552 | } | |
0293ce3a | 3553 | |
8867bc9b MK |
3554 | /* |
3555 | * Return a pointer to the extent record for file system block bno. | |
3556 | */ | |
a6f64d4a | 3557 | xfs_bmbt_rec_host_t * /* pointer to found extent record */ |
8867bc9b MK |
3558 | xfs_iext_bno_to_ext( |
3559 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3560 | xfs_fileoff_t bno, /* block number to search for */ | |
3561 | xfs_extnum_t *idxp) /* index of target extent */ | |
3562 | { | |
a6f64d4a | 3563 | xfs_bmbt_rec_host_t *base; /* pointer to first extent */ |
8867bc9b | 3564 | xfs_filblks_t blockcount = 0; /* number of blocks in extent */ |
a6f64d4a | 3565 | xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */ |
8867bc9b | 3566 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ |
c41564b5 | 3567 | int high; /* upper boundary in search */ |
8867bc9b | 3568 | xfs_extnum_t idx = 0; /* index of target extent */ |
c41564b5 | 3569 | int low; /* lower boundary in search */ |
8867bc9b MK |
3570 | xfs_extnum_t nextents; /* number of file extents */ |
3571 | xfs_fileoff_t startoff = 0; /* start offset of extent */ | |
3572 | ||
3573 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3574 | if (nextents == 0) { | |
3575 | *idxp = 0; | |
3576 | return NULL; | |
3577 | } | |
3578 | low = 0; | |
3579 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3580 | /* Find target extent list */ | |
3581 | int erp_idx = 0; | |
3582 | erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); | |
3583 | base = erp->er_extbuf; | |
3584 | high = erp->er_extcount - 1; | |
3585 | } else { | |
3586 | base = ifp->if_u1.if_extents; | |
3587 | high = nextents - 1; | |
3588 | } | |
3589 | /* Binary search extent records */ | |
3590 | while (low <= high) { | |
3591 | idx = (low + high) >> 1; | |
3592 | ep = base + idx; | |
3593 | startoff = xfs_bmbt_get_startoff(ep); | |
3594 | blockcount = xfs_bmbt_get_blockcount(ep); | |
3595 | if (bno < startoff) { | |
3596 | high = idx - 1; | |
3597 | } else if (bno >= startoff + blockcount) { | |
3598 | low = idx + 1; | |
3599 | } else { | |
3600 | /* Convert back to file-based extent index */ | |
3601 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3602 | idx += erp->er_extoff; | |
3603 | } | |
3604 | *idxp = idx; | |
3605 | return ep; | |
3606 | } | |
3607 | } | |
3608 | /* Convert back to file-based extent index */ | |
3609 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3610 | idx += erp->er_extoff; | |
3611 | } | |
3612 | if (bno >= startoff + blockcount) { | |
3613 | if (++idx == nextents) { | |
3614 | ep = NULL; | |
3615 | } else { | |
3616 | ep = xfs_iext_get_ext(ifp, idx); | |
3617 | } | |
3618 | } | |
3619 | *idxp = idx; | |
3620 | return ep; | |
3621 | } | |
3622 | ||
0293ce3a MK |
3623 | /* |
3624 | * Return a pointer to the indirection array entry containing the | |
3625 | * extent record for filesystem block bno. Store the index of the | |
3626 | * target irec in *erp_idxp. | |
3627 | */ | |
8867bc9b | 3628 | xfs_ext_irec_t * /* pointer to found extent record */ |
0293ce3a MK |
3629 | xfs_iext_bno_to_irec( |
3630 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3631 | xfs_fileoff_t bno, /* block number to search for */ | |
3632 | int *erp_idxp) /* irec index of target ext list */ | |
3633 | { | |
3634 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ | |
3635 | xfs_ext_irec_t *erp_next; /* next indirection array entry */ | |
8867bc9b | 3636 | int erp_idx; /* indirection array index */ |
0293ce3a MK |
3637 | int nlists; /* number of extent irec's (lists) */ |
3638 | int high; /* binary search upper limit */ | |
3639 | int low; /* binary search lower limit */ | |
3640 | ||
3641 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3642 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3643 | erp_idx = 0; | |
3644 | low = 0; | |
3645 | high = nlists - 1; | |
3646 | while (low <= high) { | |
3647 | erp_idx = (low + high) >> 1; | |
3648 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3649 | erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; | |
3650 | if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { | |
3651 | high = erp_idx - 1; | |
3652 | } else if (erp_next && bno >= | |
3653 | xfs_bmbt_get_startoff(erp_next->er_extbuf)) { | |
3654 | low = erp_idx + 1; | |
3655 | } else { | |
3656 | break; | |
3657 | } | |
3658 | } | |
3659 | *erp_idxp = erp_idx; | |
3660 | return erp; | |
3661 | } | |
3662 | ||
3663 | /* | |
3664 | * Return a pointer to the indirection array entry containing the | |
3665 | * extent record at file extent index *idxp. Store the index of the | |
3666 | * target irec in *erp_idxp and store the page index of the target | |
3667 | * extent record in *idxp. | |
3668 | */ | |
3669 | xfs_ext_irec_t * | |
3670 | xfs_iext_idx_to_irec( | |
3671 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3672 | xfs_extnum_t *idxp, /* extent index (file -> page) */ | |
3673 | int *erp_idxp, /* pointer to target irec */ | |
3674 | int realloc) /* new bytes were just added */ | |
3675 | { | |
3676 | xfs_ext_irec_t *prev; /* pointer to previous irec */ | |
3677 | xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ | |
3678 | int erp_idx; /* indirection array index */ | |
3679 | int nlists; /* number of irec's (ex lists) */ | |
3680 | int high; /* binary search upper limit */ | |
3681 | int low; /* binary search lower limit */ | |
3682 | xfs_extnum_t page_idx = *idxp; /* extent index in target list */ | |
3683 | ||
3684 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
87bef181 CH |
3685 | ASSERT(page_idx >= 0); |
3686 | ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t)); | |
3687 | ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc); | |
3688 | ||
0293ce3a MK |
3689 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; |
3690 | erp_idx = 0; | |
3691 | low = 0; | |
3692 | high = nlists - 1; | |
3693 | ||
3694 | /* Binary search extent irec's */ | |
3695 | while (low <= high) { | |
3696 | erp_idx = (low + high) >> 1; | |
3697 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3698 | prev = erp_idx > 0 ? erp - 1 : NULL; | |
3699 | if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && | |
3700 | realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { | |
3701 | high = erp_idx - 1; | |
3702 | } else if (page_idx > erp->er_extoff + erp->er_extcount || | |
3703 | (page_idx == erp->er_extoff + erp->er_extcount && | |
3704 | !realloc)) { | |
3705 | low = erp_idx + 1; | |
3706 | } else if (page_idx == erp->er_extoff + erp->er_extcount && | |
3707 | erp->er_extcount == XFS_LINEAR_EXTS) { | |
3708 | ASSERT(realloc); | |
3709 | page_idx = 0; | |
3710 | erp_idx++; | |
3711 | erp = erp_idx < nlists ? erp + 1 : NULL; | |
3712 | break; | |
3713 | } else { | |
3714 | page_idx -= erp->er_extoff; | |
3715 | break; | |
3716 | } | |
3717 | } | |
3718 | *idxp = page_idx; | |
3719 | *erp_idxp = erp_idx; | |
3720 | return(erp); | |
3721 | } | |
3722 | ||
3723 | /* | |
3724 | * Allocate and initialize an indirection array once the space needed | |
3725 | * for incore extents increases above XFS_IEXT_BUFSZ. | |
3726 | */ | |
3727 | void | |
3728 | xfs_iext_irec_init( | |
3729 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3730 | { | |
3731 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3732 | xfs_extnum_t nextents; /* number of extents in file */ | |
3733 | ||
3734 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); | |
3735 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3736 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
3737 | ||
6785073b | 3738 | erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS); |
0293ce3a MK |
3739 | |
3740 | if (nextents == 0) { | |
6785073b | 3741 | ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); |
0293ce3a MK |
3742 | } else if (!ifp->if_real_bytes) { |
3743 | xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); | |
3744 | } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { | |
3745 | xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); | |
3746 | } | |
3747 | erp->er_extbuf = ifp->if_u1.if_extents; | |
3748 | erp->er_extcount = nextents; | |
3749 | erp->er_extoff = 0; | |
3750 | ||
3751 | ifp->if_flags |= XFS_IFEXTIREC; | |
3752 | ifp->if_real_bytes = XFS_IEXT_BUFSZ; | |
3753 | ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); | |
3754 | ifp->if_u1.if_ext_irec = erp; | |
3755 | ||
3756 | return; | |
3757 | } | |
3758 | ||
3759 | /* | |
3760 | * Allocate and initialize a new entry in the indirection array. | |
3761 | */ | |
3762 | xfs_ext_irec_t * | |
3763 | xfs_iext_irec_new( | |
3764 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3765 | int erp_idx) /* index for new irec */ | |
3766 | { | |
3767 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3768 | int i; /* loop counter */ | |
3769 | int nlists; /* number of irec's (ex lists) */ | |
3770 | ||
3771 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3772 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3773 | ||
3774 | /* Resize indirection array */ | |
3775 | xfs_iext_realloc_indirect(ifp, ++nlists * | |
3776 | sizeof(xfs_ext_irec_t)); | |
3777 | /* | |
3778 | * Move records down in the array so the | |
3779 | * new page can use erp_idx. | |
3780 | */ | |
3781 | erp = ifp->if_u1.if_ext_irec; | |
3782 | for (i = nlists - 1; i > erp_idx; i--) { | |
3783 | memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); | |
3784 | } | |
3785 | ASSERT(i == erp_idx); | |
3786 | ||
3787 | /* Initialize new extent record */ | |
3788 | erp = ifp->if_u1.if_ext_irec; | |
6785073b | 3789 | erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); |
0293ce3a MK |
3790 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; |
3791 | memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); | |
3792 | erp[erp_idx].er_extcount = 0; | |
3793 | erp[erp_idx].er_extoff = erp_idx > 0 ? | |
3794 | erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; | |
3795 | return (&erp[erp_idx]); | |
3796 | } | |
3797 | ||
3798 | /* | |
3799 | * Remove a record from the indirection array. | |
3800 | */ | |
3801 | void | |
3802 | xfs_iext_irec_remove( | |
3803 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3804 | int erp_idx) /* irec index to remove */ | |
3805 | { | |
3806 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
3807 | int i; /* loop counter */ | |
3808 | int nlists; /* number of irec's (ex lists) */ | |
3809 | ||
3810 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3811 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3812 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3813 | if (erp->er_extbuf) { | |
3814 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, | |
3815 | -erp->er_extcount); | |
f0e2d93c | 3816 | kmem_free(erp->er_extbuf); |
0293ce3a MK |
3817 | } |
3818 | /* Compact extent records */ | |
3819 | erp = ifp->if_u1.if_ext_irec; | |
3820 | for (i = erp_idx; i < nlists - 1; i++) { | |
3821 | memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); | |
3822 | } | |
3823 | /* | |
3824 | * Manually free the last extent record from the indirection | |
3825 | * array. A call to xfs_iext_realloc_indirect() with a size | |
3826 | * of zero would result in a call to xfs_iext_destroy() which | |
3827 | * would in turn call this function again, creating a nasty | |
3828 | * infinite loop. | |
3829 | */ | |
3830 | if (--nlists) { | |
3831 | xfs_iext_realloc_indirect(ifp, | |
3832 | nlists * sizeof(xfs_ext_irec_t)); | |
3833 | } else { | |
f0e2d93c | 3834 | kmem_free(ifp->if_u1.if_ext_irec); |
0293ce3a MK |
3835 | } |
3836 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; | |
3837 | } | |
3838 | ||
3839 | /* | |
3840 | * This is called to clean up large amounts of unused memory allocated | |
3841 | * by the indirection array. Before compacting anything though, verify | |
3842 | * that the indirection array is still needed and switch back to the | |
3843 | * linear extent list (or even the inline buffer) if possible. The | |
3844 | * compaction policy is as follows: | |
3845 | * | |
3846 | * Full Compaction: Extents fit into a single page (or inline buffer) | |
71a8c87f | 3847 | * Partial Compaction: Extents occupy less than 50% of allocated space |
0293ce3a MK |
3848 | * No Compaction: Extents occupy at least 50% of allocated space |
3849 | */ | |
3850 | void | |
3851 | xfs_iext_irec_compact( | |
3852 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3853 | { | |
3854 | xfs_extnum_t nextents; /* number of extents in file */ | |
3855 | int nlists; /* number of irec's (ex lists) */ | |
3856 | ||
3857 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3858 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3859 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3860 | ||
3861 | if (nextents == 0) { | |
3862 | xfs_iext_destroy(ifp); | |
3863 | } else if (nextents <= XFS_INLINE_EXTS) { | |
3864 | xfs_iext_indirect_to_direct(ifp); | |
3865 | xfs_iext_direct_to_inline(ifp, nextents); | |
3866 | } else if (nextents <= XFS_LINEAR_EXTS) { | |
3867 | xfs_iext_indirect_to_direct(ifp); | |
0293ce3a MK |
3868 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { |
3869 | xfs_iext_irec_compact_pages(ifp); | |
3870 | } | |
3871 | } | |
3872 | ||
3873 | /* | |
3874 | * Combine extents from neighboring extent pages. | |
3875 | */ | |
3876 | void | |
3877 | xfs_iext_irec_compact_pages( | |
3878 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
3879 | { | |
3880 | xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ | |
3881 | int erp_idx = 0; /* indirection array index */ | |
3882 | int nlists; /* number of irec's (ex lists) */ | |
3883 | ||
3884 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3885 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3886 | while (erp_idx < nlists - 1) { | |
3887 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3888 | erp_next = erp + 1; | |
3889 | if (erp_next->er_extcount <= | |
3890 | (XFS_LINEAR_EXTS - erp->er_extcount)) { | |
71a8c87f | 3891 | memcpy(&erp->er_extbuf[erp->er_extcount], |
0293ce3a MK |
3892 | erp_next->er_extbuf, erp_next->er_extcount * |
3893 | sizeof(xfs_bmbt_rec_t)); | |
3894 | erp->er_extcount += erp_next->er_extcount; | |
3895 | /* | |
3896 | * Free page before removing extent record | |
3897 | * so er_extoffs don't get modified in | |
3898 | * xfs_iext_irec_remove. | |
3899 | */ | |
f0e2d93c | 3900 | kmem_free(erp_next->er_extbuf); |
0293ce3a MK |
3901 | erp_next->er_extbuf = NULL; |
3902 | xfs_iext_irec_remove(ifp, erp_idx + 1); | |
3903 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3904 | } else { | |
3905 | erp_idx++; | |
3906 | } | |
3907 | } | |
3908 | } | |
3909 | ||
0293ce3a MK |
3910 | /* |
3911 | * This is called to update the er_extoff field in the indirection | |
3912 | * array when extents have been added or removed from one of the | |
3913 | * extent lists. erp_idx contains the irec index to begin updating | |
3914 | * at and ext_diff contains the number of extents that were added | |
3915 | * or removed. | |
3916 | */ | |
3917 | void | |
3918 | xfs_iext_irec_update_extoffs( | |
3919 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3920 | int erp_idx, /* irec index to update */ | |
3921 | int ext_diff) /* number of new extents */ | |
3922 | { | |
3923 | int i; /* loop counter */ | |
3924 | int nlists; /* number of irec's (ex lists */ | |
3925 | ||
3926 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3927 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3928 | for (i = erp_idx; i < nlists; i++) { | |
3929 | ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; | |
3930 | } | |
3931 | } | |
72b53efa BF |
3932 | |
3933 | /* | |
3934 | * Test whether it is appropriate to check an inode for and free post EOF | |
3935 | * blocks. The 'force' parameter determines whether we should also consider | |
3936 | * regular files that are marked preallocated or append-only. | |
3937 | */ | |
3938 | bool | |
3939 | xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) | |
3940 | { | |
3941 | /* prealloc/delalloc exists only on regular files */ | |
3942 | if (!S_ISREG(ip->i_d.di_mode)) | |
3943 | return false; | |
3944 | ||
3945 | /* | |
3946 | * Zero sized files with no cached pages and delalloc blocks will not | |
3947 | * have speculative prealloc/delalloc blocks to remove. | |
3948 | */ | |
3949 | if (VFS_I(ip)->i_size == 0 && | |
3950 | VN_CACHED(VFS_I(ip)) == 0 && | |
3951 | ip->i_delayed_blks == 0) | |
3952 | return false; | |
3953 | ||
3954 | /* If we haven't read in the extent list, then don't do it now. */ | |
3955 | if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) | |
3956 | return false; | |
3957 | ||
3958 | /* | |
3959 | * Do not free real preallocated or append-only files unless the file | |
3960 | * has delalloc blocks and we are forced to remove them. | |
3961 | */ | |
3962 | if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) | |
3963 | if (!force || ip->i_delayed_blks == 0) | |
3964 | return false; | |
3965 | ||
3966 | return true; | |
3967 | } | |
3968 |