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 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 NS |
23 | #include "xfs_inum.h" |
24 | #include "xfs_imap.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 LT |
29 | #include "xfs_dir2.h" |
30 | #include "xfs_dmapi.h" | |
31 | #include "xfs_mount.h" | |
1da177e4 | 32 | #include "xfs_bmap_btree.h" |
a844f451 | 33 | #include "xfs_alloc_btree.h" |
1da177e4 | 34 | #include "xfs_ialloc_btree.h" |
1da177e4 | 35 | #include "xfs_dir2_sf.h" |
a844f451 | 36 | #include "xfs_attr_sf.h" |
1da177e4 | 37 | #include "xfs_dinode.h" |
1da177e4 | 38 | #include "xfs_inode.h" |
1da177e4 | 39 | #include "xfs_buf_item.h" |
a844f451 NS |
40 | #include "xfs_inode_item.h" |
41 | #include "xfs_btree.h" | |
42 | #include "xfs_alloc.h" | |
43 | #include "xfs_ialloc.h" | |
44 | #include "xfs_bmap.h" | |
1da177e4 LT |
45 | #include "xfs_rw.h" |
46 | #include "xfs_error.h" | |
1da177e4 LT |
47 | #include "xfs_utils.h" |
48 | #include "xfs_dir2_trace.h" | |
49 | #include "xfs_quota.h" | |
50 | #include "xfs_mac.h" | |
51 | #include "xfs_acl.h" | |
52 | ||
53 | ||
54 | kmem_zone_t *xfs_ifork_zone; | |
55 | kmem_zone_t *xfs_inode_zone; | |
56 | kmem_zone_t *xfs_chashlist_zone; | |
57 | ||
58 | /* | |
59 | * Used in xfs_itruncate(). This is the maximum number of extents | |
60 | * freed from a file in a single transaction. | |
61 | */ | |
62 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
63 | ||
64 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); | |
65 | STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); | |
66 | STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); | |
67 | STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); | |
68 | ||
69 | ||
70 | #ifdef DEBUG | |
71 | /* | |
72 | * Make sure that the extents in the given memory buffer | |
73 | * are valid. | |
74 | */ | |
75 | STATIC void | |
76 | xfs_validate_extents( | |
4eea22f0 | 77 | xfs_ifork_t *ifp, |
1da177e4 LT |
78 | int nrecs, |
79 | int disk, | |
80 | xfs_exntfmt_t fmt) | |
81 | { | |
4eea22f0 | 82 | xfs_bmbt_rec_t *ep; |
1da177e4 LT |
83 | xfs_bmbt_irec_t irec; |
84 | xfs_bmbt_rec_t rec; | |
85 | int i; | |
86 | ||
87 | for (i = 0; i < nrecs; i++) { | |
4eea22f0 | 88 | ep = xfs_iext_get_ext(ifp, i); |
1da177e4 LT |
89 | rec.l0 = get_unaligned((__uint64_t*)&ep->l0); |
90 | rec.l1 = get_unaligned((__uint64_t*)&ep->l1); | |
91 | if (disk) | |
92 | xfs_bmbt_disk_get_all(&rec, &irec); | |
93 | else | |
94 | xfs_bmbt_get_all(&rec, &irec); | |
95 | if (fmt == XFS_EXTFMT_NOSTATE) | |
96 | ASSERT(irec.br_state == XFS_EXT_NORM); | |
1da177e4 LT |
97 | } |
98 | } | |
99 | #else /* DEBUG */ | |
4eea22f0 | 100 | #define xfs_validate_extents(ifp, nrecs, disk, fmt) |
1da177e4 LT |
101 | #endif /* DEBUG */ |
102 | ||
103 | /* | |
104 | * Check that none of the inode's in the buffer have a next | |
105 | * unlinked field of 0. | |
106 | */ | |
107 | #if defined(DEBUG) | |
108 | void | |
109 | xfs_inobp_check( | |
110 | xfs_mount_t *mp, | |
111 | xfs_buf_t *bp) | |
112 | { | |
113 | int i; | |
114 | int j; | |
115 | xfs_dinode_t *dip; | |
116 | ||
117 | j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; | |
118 | ||
119 | for (i = 0; i < j; i++) { | |
120 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
121 | i * mp->m_sb.sb_inodesize); | |
122 | if (!dip->di_next_unlinked) { | |
123 | xfs_fs_cmn_err(CE_ALERT, mp, | |
124 | "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.", | |
125 | bp); | |
126 | ASSERT(dip->di_next_unlinked); | |
127 | } | |
128 | } | |
129 | } | |
130 | #endif | |
131 | ||
1da177e4 LT |
132 | /* |
133 | * This routine is called to map an inode number within a file | |
134 | * system to the buffer containing the on-disk version of the | |
135 | * inode. It returns a pointer to the buffer containing the | |
136 | * on-disk inode in the bpp parameter, and in the dip parameter | |
137 | * it returns a pointer to the on-disk inode within that buffer. | |
138 | * | |
139 | * If a non-zero error is returned, then the contents of bpp and | |
140 | * dipp are undefined. | |
141 | * | |
142 | * Use xfs_imap() to determine the size and location of the | |
143 | * buffer to read from disk. | |
144 | */ | |
ba0f32d4 | 145 | STATIC int |
1da177e4 LT |
146 | xfs_inotobp( |
147 | xfs_mount_t *mp, | |
148 | xfs_trans_t *tp, | |
149 | xfs_ino_t ino, | |
150 | xfs_dinode_t **dipp, | |
151 | xfs_buf_t **bpp, | |
152 | int *offset) | |
153 | { | |
154 | int di_ok; | |
155 | xfs_imap_t imap; | |
156 | xfs_buf_t *bp; | |
157 | int error; | |
158 | xfs_dinode_t *dip; | |
159 | ||
160 | /* | |
c41564b5 | 161 | * Call the space management code to find the location of the |
1da177e4 LT |
162 | * inode on disk. |
163 | */ | |
164 | imap.im_blkno = 0; | |
165 | error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP); | |
166 | if (error != 0) { | |
167 | cmn_err(CE_WARN, | |
168 | "xfs_inotobp: xfs_imap() returned an " | |
169 | "error %d on %s. Returning error.", error, mp->m_fsname); | |
170 | return error; | |
171 | } | |
172 | ||
173 | /* | |
174 | * If the inode number maps to a block outside the bounds of the | |
175 | * file system then return NULL rather than calling read_buf | |
176 | * and panicing when we get an error from the driver. | |
177 | */ | |
178 | if ((imap.im_blkno + imap.im_len) > | |
179 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
180 | cmn_err(CE_WARN, | |
da1650a5 | 181 | "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds " |
1da177e4 | 182 | "of the file system %s. Returning EINVAL.", |
da1650a5 CH |
183 | (unsigned long long)imap.im_blkno, |
184 | imap.im_len, mp->m_fsname); | |
1da177e4 LT |
185 | return XFS_ERROR(EINVAL); |
186 | } | |
187 | ||
188 | /* | |
189 | * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will | |
190 | * default to just a read_buf() call. | |
191 | */ | |
192 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno, | |
193 | (int)imap.im_len, XFS_BUF_LOCK, &bp); | |
194 | ||
195 | if (error) { | |
196 | cmn_err(CE_WARN, | |
197 | "xfs_inotobp: xfs_trans_read_buf() returned an " | |
198 | "error %d on %s. Returning error.", error, mp->m_fsname); | |
199 | return error; | |
200 | } | |
201 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0); | |
202 | di_ok = | |
203 | INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC && | |
204 | XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT)); | |
205 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP, | |
206 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
207 | XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip); | |
208 | xfs_trans_brelse(tp, bp); | |
209 | cmn_err(CE_WARN, | |
210 | "xfs_inotobp: XFS_TEST_ERROR() returned an " | |
211 | "error on %s. Returning EFSCORRUPTED.", mp->m_fsname); | |
212 | return XFS_ERROR(EFSCORRUPTED); | |
213 | } | |
214 | ||
215 | xfs_inobp_check(mp, bp); | |
216 | ||
217 | /* | |
218 | * Set *dipp to point to the on-disk inode in the buffer. | |
219 | */ | |
220 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); | |
221 | *bpp = bp; | |
222 | *offset = imap.im_boffset; | |
223 | return 0; | |
224 | } | |
225 | ||
226 | ||
227 | /* | |
228 | * This routine is called to map an inode to the buffer containing | |
229 | * the on-disk version of the inode. It returns a pointer to the | |
230 | * buffer containing the on-disk inode in the bpp parameter, and in | |
231 | * the dip parameter it returns a pointer to the on-disk inode within | |
232 | * that buffer. | |
233 | * | |
234 | * If a non-zero error is returned, then the contents of bpp and | |
235 | * dipp are undefined. | |
236 | * | |
237 | * If the inode is new and has not yet been initialized, use xfs_imap() | |
238 | * to determine the size and location of the buffer to read from disk. | |
239 | * If the inode has already been mapped to its buffer and read in once, | |
240 | * then use the mapping information stored in the inode rather than | |
241 | * calling xfs_imap(). This allows us to avoid the overhead of looking | |
242 | * at the inode btree for small block file systems (see xfs_dilocate()). | |
243 | * We can tell whether the inode has been mapped in before by comparing | |
244 | * its disk block address to 0. Only uninitialized inodes will have | |
245 | * 0 for the disk block address. | |
246 | */ | |
247 | int | |
248 | xfs_itobp( | |
249 | xfs_mount_t *mp, | |
250 | xfs_trans_t *tp, | |
251 | xfs_inode_t *ip, | |
252 | xfs_dinode_t **dipp, | |
253 | xfs_buf_t **bpp, | |
b12dd342 NS |
254 | xfs_daddr_t bno, |
255 | uint imap_flags) | |
1da177e4 | 256 | { |
4d1a2ed3 | 257 | xfs_imap_t imap; |
1da177e4 LT |
258 | xfs_buf_t *bp; |
259 | int error; | |
1da177e4 LT |
260 | int i; |
261 | int ni; | |
1da177e4 LT |
262 | |
263 | if (ip->i_blkno == (xfs_daddr_t)0) { | |
264 | /* | |
265 | * Call the space management code to find the location of the | |
266 | * inode on disk. | |
267 | */ | |
268 | imap.im_blkno = bno; | |
b12dd342 NS |
269 | if ((error = xfs_imap(mp, tp, ip->i_ino, &imap, |
270 | XFS_IMAP_LOOKUP | imap_flags))) | |
1da177e4 | 271 | return error; |
1da177e4 LT |
272 | |
273 | /* | |
274 | * If the inode number maps to a block outside the bounds | |
275 | * of the file system then return NULL rather than calling | |
276 | * read_buf and panicing when we get an error from the | |
277 | * driver. | |
278 | */ | |
279 | if ((imap.im_blkno + imap.im_len) > | |
280 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
281 | #ifdef DEBUG | |
282 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: " | |
283 | "(imap.im_blkno (0x%llx) " | |
284 | "+ imap.im_len (0x%llx)) > " | |
285 | " XFS_FSB_TO_BB(mp, " | |
286 | "mp->m_sb.sb_dblocks) (0x%llx)", | |
287 | (unsigned long long) imap.im_blkno, | |
288 | (unsigned long long) imap.im_len, | |
289 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
290 | #endif /* DEBUG */ | |
291 | return XFS_ERROR(EINVAL); | |
292 | } | |
293 | ||
294 | /* | |
295 | * Fill in the fields in the inode that will be used to | |
296 | * map the inode to its buffer from now on. | |
297 | */ | |
298 | ip->i_blkno = imap.im_blkno; | |
299 | ip->i_len = imap.im_len; | |
300 | ip->i_boffset = imap.im_boffset; | |
301 | } else { | |
302 | /* | |
303 | * We've already mapped the inode once, so just use the | |
304 | * mapping that we saved the first time. | |
305 | */ | |
306 | imap.im_blkno = ip->i_blkno; | |
307 | imap.im_len = ip->i_len; | |
308 | imap.im_boffset = ip->i_boffset; | |
309 | } | |
310 | ASSERT(bno == 0 || bno == imap.im_blkno); | |
311 | ||
312 | /* | |
313 | * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will | |
314 | * default to just a read_buf() call. | |
315 | */ | |
316 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno, | |
317 | (int)imap.im_len, XFS_BUF_LOCK, &bp); | |
1da177e4 LT |
318 | if (error) { |
319 | #ifdef DEBUG | |
320 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: " | |
321 | "xfs_trans_read_buf() returned error %d, " | |
322 | "imap.im_blkno 0x%llx, imap.im_len 0x%llx", | |
323 | error, (unsigned long long) imap.im_blkno, | |
324 | (unsigned long long) imap.im_len); | |
325 | #endif /* DEBUG */ | |
326 | return error; | |
327 | } | |
4d1a2ed3 | 328 | |
1da177e4 LT |
329 | /* |
330 | * Validate the magic number and version of every inode in the buffer | |
331 | * (if DEBUG kernel) or the first inode in the buffer, otherwise. | |
4d1a2ed3 | 332 | * No validation is done here in userspace (xfs_repair). |
1da177e4 | 333 | */ |
4d1a2ed3 NS |
334 | #if !defined(__KERNEL__) |
335 | ni = 0; | |
336 | #elif defined(DEBUG) | |
41ff715a | 337 | ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog; |
4d1a2ed3 | 338 | #else /* usual case */ |
41ff715a | 339 | ni = 1; |
1da177e4 | 340 | #endif |
4d1a2ed3 | 341 | |
1da177e4 LT |
342 | for (i = 0; i < ni; i++) { |
343 | int di_ok; | |
344 | xfs_dinode_t *dip; | |
345 | ||
346 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
347 | (i << mp->m_sb.sb_inodelog)); | |
348 | di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC && | |
349 | XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT)); | |
41ff715a NS |
350 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, |
351 | XFS_ERRTAG_ITOBP_INOTOBP, | |
352 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
353 | if (imap_flags & XFS_IMAP_BULKSTAT) { | |
354 | xfs_trans_brelse(tp, bp); | |
355 | return XFS_ERROR(EINVAL); | |
356 | } | |
1da177e4 | 357 | #ifdef DEBUG |
41ff715a | 358 | cmn_err(CE_ALERT, |
4d1a2ed3 NS |
359 | "Device %s - bad inode magic/vsn " |
360 | "daddr %lld #%d (magic=%x)", | |
b6574520 | 361 | XFS_BUFTARG_NAME(mp->m_ddev_targp), |
1da177e4 LT |
362 | (unsigned long long)imap.im_blkno, i, |
363 | INT_GET(dip->di_core.di_magic, ARCH_CONVERT)); | |
364 | #endif | |
365 | XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH, | |
366 | mp, dip); | |
367 | xfs_trans_brelse(tp, bp); | |
368 | return XFS_ERROR(EFSCORRUPTED); | |
369 | } | |
370 | } | |
1da177e4 LT |
371 | |
372 | xfs_inobp_check(mp, bp); | |
373 | ||
374 | /* | |
375 | * Mark the buffer as an inode buffer now that it looks good | |
376 | */ | |
377 | XFS_BUF_SET_VTYPE(bp, B_FS_INO); | |
378 | ||
379 | /* | |
380 | * Set *dipp to point to the on-disk inode in the buffer. | |
381 | */ | |
382 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); | |
383 | *bpp = bp; | |
384 | return 0; | |
385 | } | |
386 | ||
387 | /* | |
388 | * Move inode type and inode format specific information from the | |
389 | * on-disk inode to the in-core inode. For fifos, devs, and sockets | |
390 | * this means set if_rdev to the proper value. For files, directories, | |
391 | * and symlinks this means to bring in the in-line data or extent | |
392 | * pointers. For a file in B-tree format, only the root is immediately | |
393 | * brought in-core. The rest will be in-lined in if_extents when it | |
394 | * is first referenced (see xfs_iread_extents()). | |
395 | */ | |
396 | STATIC int | |
397 | xfs_iformat( | |
398 | xfs_inode_t *ip, | |
399 | xfs_dinode_t *dip) | |
400 | { | |
401 | xfs_attr_shortform_t *atp; | |
402 | int size; | |
403 | int error; | |
404 | xfs_fsize_t di_size; | |
405 | ip->i_df.if_ext_max = | |
406 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
407 | error = 0; | |
408 | ||
409 | if (unlikely( | |
410 | INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) + | |
411 | INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) > | |
412 | INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) { | |
3762ec6b NS |
413 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
414 | "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", | |
1da177e4 LT |
415 | (unsigned long long)ip->i_ino, |
416 | (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) | |
417 | + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)), | |
418 | (unsigned long long) | |
419 | INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT)); | |
420 | XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, | |
421 | ip->i_mount, dip); | |
422 | return XFS_ERROR(EFSCORRUPTED); | |
423 | } | |
424 | ||
425 | if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) { | |
3762ec6b NS |
426 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
427 | "corrupt dinode %Lu, forkoff = 0x%x.", | |
1da177e4 LT |
428 | (unsigned long long)ip->i_ino, |
429 | (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT))); | |
430 | XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, | |
431 | ip->i_mount, dip); | |
432 | return XFS_ERROR(EFSCORRUPTED); | |
433 | } | |
434 | ||
435 | switch (ip->i_d.di_mode & S_IFMT) { | |
436 | case S_IFIFO: | |
437 | case S_IFCHR: | |
438 | case S_IFBLK: | |
439 | case S_IFSOCK: | |
440 | if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) { | |
441 | XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, | |
442 | ip->i_mount, dip); | |
443 | return XFS_ERROR(EFSCORRUPTED); | |
444 | } | |
445 | ip->i_d.di_size = 0; | |
446 | ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT); | |
447 | break; | |
448 | ||
449 | case S_IFREG: | |
450 | case S_IFLNK: | |
451 | case S_IFDIR: | |
452 | switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) { | |
453 | case XFS_DINODE_FMT_LOCAL: | |
454 | /* | |
455 | * no local regular files yet | |
456 | */ | |
457 | if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) { | |
3762ec6b NS |
458 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
459 | "corrupt inode %Lu " | |
460 | "(local format for regular file).", | |
1da177e4 LT |
461 | (unsigned long long) ip->i_ino); |
462 | XFS_CORRUPTION_ERROR("xfs_iformat(4)", | |
463 | XFS_ERRLEVEL_LOW, | |
464 | ip->i_mount, dip); | |
465 | return XFS_ERROR(EFSCORRUPTED); | |
466 | } | |
467 | ||
468 | di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT); | |
469 | if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { | |
3762ec6b NS |
470 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
471 | "corrupt inode %Lu " | |
472 | "(bad size %Ld for local inode).", | |
1da177e4 LT |
473 | (unsigned long long) ip->i_ino, |
474 | (long long) di_size); | |
475 | XFS_CORRUPTION_ERROR("xfs_iformat(5)", | |
476 | XFS_ERRLEVEL_LOW, | |
477 | ip->i_mount, dip); | |
478 | return XFS_ERROR(EFSCORRUPTED); | |
479 | } | |
480 | ||
481 | size = (int)di_size; | |
482 | error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); | |
483 | break; | |
484 | case XFS_DINODE_FMT_EXTENTS: | |
485 | error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); | |
486 | break; | |
487 | case XFS_DINODE_FMT_BTREE: | |
488 | error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); | |
489 | break; | |
490 | default: | |
491 | XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, | |
492 | ip->i_mount); | |
493 | return XFS_ERROR(EFSCORRUPTED); | |
494 | } | |
495 | break; | |
496 | ||
497 | default: | |
498 | XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); | |
499 | return XFS_ERROR(EFSCORRUPTED); | |
500 | } | |
501 | if (error) { | |
502 | return error; | |
503 | } | |
504 | if (!XFS_DFORK_Q(dip)) | |
505 | return 0; | |
506 | ASSERT(ip->i_afp == NULL); | |
507 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
508 | ip->i_afp->if_ext_max = | |
509 | XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
510 | switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) { | |
511 | case XFS_DINODE_FMT_LOCAL: | |
512 | atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); | |
3b244aa8 | 513 | size = be16_to_cpu(atp->hdr.totsize); |
1da177e4 LT |
514 | error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); |
515 | break; | |
516 | case XFS_DINODE_FMT_EXTENTS: | |
517 | error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); | |
518 | break; | |
519 | case XFS_DINODE_FMT_BTREE: | |
520 | error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); | |
521 | break; | |
522 | default: | |
523 | error = XFS_ERROR(EFSCORRUPTED); | |
524 | break; | |
525 | } | |
526 | if (error) { | |
527 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
528 | ip->i_afp = NULL; | |
529 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
530 | } | |
531 | return error; | |
532 | } | |
533 | ||
534 | /* | |
535 | * The file is in-lined in the on-disk inode. | |
536 | * If it fits into if_inline_data, then copy | |
537 | * it there, otherwise allocate a buffer for it | |
538 | * and copy the data there. Either way, set | |
539 | * if_data to point at the data. | |
540 | * If we allocate a buffer for the data, make | |
541 | * sure that its size is a multiple of 4 and | |
542 | * record the real size in i_real_bytes. | |
543 | */ | |
544 | STATIC int | |
545 | xfs_iformat_local( | |
546 | xfs_inode_t *ip, | |
547 | xfs_dinode_t *dip, | |
548 | int whichfork, | |
549 | int size) | |
550 | { | |
551 | xfs_ifork_t *ifp; | |
552 | int real_size; | |
553 | ||
554 | /* | |
555 | * If the size is unreasonable, then something | |
556 | * is wrong and we just bail out rather than crash in | |
557 | * kmem_alloc() or memcpy() below. | |
558 | */ | |
559 | if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
3762ec6b NS |
560 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
561 | "corrupt inode %Lu " | |
562 | "(bad size %d for local fork, size = %d).", | |
1da177e4 LT |
563 | (unsigned long long) ip->i_ino, size, |
564 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); | |
565 | XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, | |
566 | ip->i_mount, dip); | |
567 | return XFS_ERROR(EFSCORRUPTED); | |
568 | } | |
569 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
570 | real_size = 0; | |
571 | if (size == 0) | |
572 | ifp->if_u1.if_data = NULL; | |
573 | else if (size <= sizeof(ifp->if_u2.if_inline_data)) | |
574 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
575 | else { | |
576 | real_size = roundup(size, 4); | |
577 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
578 | } | |
579 | ifp->if_bytes = size; | |
580 | ifp->if_real_bytes = real_size; | |
581 | if (size) | |
582 | memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); | |
583 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
584 | ifp->if_flags |= XFS_IFINLINE; | |
585 | return 0; | |
586 | } | |
587 | ||
588 | /* | |
589 | * The file consists of a set of extents all | |
590 | * of which fit into the on-disk inode. | |
591 | * If there are few enough extents to fit into | |
592 | * the if_inline_ext, then copy them there. | |
593 | * Otherwise allocate a buffer for them and copy | |
594 | * them into it. Either way, set if_extents | |
595 | * to point at the extents. | |
596 | */ | |
597 | STATIC int | |
598 | xfs_iformat_extents( | |
599 | xfs_inode_t *ip, | |
600 | xfs_dinode_t *dip, | |
601 | int whichfork) | |
602 | { | |
603 | xfs_bmbt_rec_t *ep, *dp; | |
604 | xfs_ifork_t *ifp; | |
605 | int nex; | |
1da177e4 LT |
606 | int size; |
607 | int i; | |
608 | ||
609 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
610 | nex = XFS_DFORK_NEXTENTS(dip, whichfork); | |
611 | size = nex * (uint)sizeof(xfs_bmbt_rec_t); | |
612 | ||
613 | /* | |
614 | * If the number of extents is unreasonable, then something | |
615 | * is wrong and we just bail out rather than crash in | |
616 | * kmem_alloc() or memcpy() below. | |
617 | */ | |
618 | if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
3762ec6b NS |
619 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
620 | "corrupt inode %Lu ((a)extents = %d).", | |
1da177e4 LT |
621 | (unsigned long long) ip->i_ino, nex); |
622 | XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, | |
623 | ip->i_mount, dip); | |
624 | return XFS_ERROR(EFSCORRUPTED); | |
625 | } | |
626 | ||
4eea22f0 | 627 | ifp->if_real_bytes = 0; |
1da177e4 LT |
628 | if (nex == 0) |
629 | ifp->if_u1.if_extents = NULL; | |
630 | else if (nex <= XFS_INLINE_EXTS) | |
631 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
4eea22f0 MK |
632 | else |
633 | xfs_iext_add(ifp, 0, nex); | |
634 | ||
1da177e4 | 635 | ifp->if_bytes = size; |
1da177e4 LT |
636 | if (size) { |
637 | dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); | |
4eea22f0 MK |
638 | xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip)); |
639 | for (i = 0; i < nex; i++, dp++) { | |
640 | ep = xfs_iext_get_ext(ifp, i); | |
1da177e4 LT |
641 | ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0), |
642 | ARCH_CONVERT); | |
643 | ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1), | |
644 | ARCH_CONVERT); | |
645 | } | |
646 | xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex, | |
647 | whichfork); | |
648 | if (whichfork != XFS_DATA_FORK || | |
649 | XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) | |
650 | if (unlikely(xfs_check_nostate_extents( | |
4eea22f0 | 651 | ifp, 0, nex))) { |
1da177e4 LT |
652 | XFS_ERROR_REPORT("xfs_iformat_extents(2)", |
653 | XFS_ERRLEVEL_LOW, | |
654 | ip->i_mount); | |
655 | return XFS_ERROR(EFSCORRUPTED); | |
656 | } | |
657 | } | |
658 | ifp->if_flags |= XFS_IFEXTENTS; | |
659 | return 0; | |
660 | } | |
661 | ||
662 | /* | |
663 | * The file has too many extents to fit into | |
664 | * the inode, so they are in B-tree format. | |
665 | * Allocate a buffer for the root of the B-tree | |
666 | * and copy the root into it. The i_extents | |
667 | * field will remain NULL until all of the | |
668 | * extents are read in (when they are needed). | |
669 | */ | |
670 | STATIC int | |
671 | xfs_iformat_btree( | |
672 | xfs_inode_t *ip, | |
673 | xfs_dinode_t *dip, | |
674 | int whichfork) | |
675 | { | |
676 | xfs_bmdr_block_t *dfp; | |
677 | xfs_ifork_t *ifp; | |
678 | /* REFERENCED */ | |
679 | int nrecs; | |
680 | int size; | |
681 | ||
682 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
683 | dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); | |
684 | size = XFS_BMAP_BROOT_SPACE(dfp); | |
685 | nrecs = XFS_BMAP_BROOT_NUMRECS(dfp); | |
686 | ||
687 | /* | |
688 | * blow out if -- fork has less extents than can fit in | |
689 | * fork (fork shouldn't be a btree format), root btree | |
690 | * block has more records than can fit into the fork, | |
691 | * or the number of extents is greater than the number of | |
692 | * blocks. | |
693 | */ | |
694 | if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max | |
695 | || XFS_BMDR_SPACE_CALC(nrecs) > | |
696 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) | |
697 | || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { | |
3762ec6b NS |
698 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
699 | "corrupt inode %Lu (btree).", | |
1da177e4 LT |
700 | (unsigned long long) ip->i_ino); |
701 | XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW, | |
702 | ip->i_mount); | |
703 | return XFS_ERROR(EFSCORRUPTED); | |
704 | } | |
705 | ||
706 | ifp->if_broot_bytes = size; | |
707 | ifp->if_broot = kmem_alloc(size, KM_SLEEP); | |
708 | ASSERT(ifp->if_broot != NULL); | |
709 | /* | |
710 | * Copy and convert from the on-disk structure | |
711 | * to the in-memory structure. | |
712 | */ | |
713 | xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), | |
714 | ifp->if_broot, size); | |
715 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
716 | ifp->if_flags |= XFS_IFBROOT; | |
717 | ||
718 | return 0; | |
719 | } | |
720 | ||
721 | /* | |
722 | * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk | |
723 | * and native format | |
724 | * | |
725 | * buf = on-disk representation | |
726 | * dip = native representation | |
727 | * dir = direction - +ve -> disk to native | |
728 | * -ve -> native to disk | |
729 | */ | |
730 | void | |
731 | xfs_xlate_dinode_core( | |
732 | xfs_caddr_t buf, | |
733 | xfs_dinode_core_t *dip, | |
734 | int dir) | |
735 | { | |
736 | xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf; | |
737 | xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip; | |
738 | xfs_arch_t arch = ARCH_CONVERT; | |
739 | ||
740 | ASSERT(dir); | |
741 | ||
742 | INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch); | |
743 | INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch); | |
744 | INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch); | |
745 | INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch); | |
746 | INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch); | |
747 | INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch); | |
748 | INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch); | |
749 | INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch); | |
750 | INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch); | |
751 | ||
752 | if (dir > 0) { | |
753 | memcpy(mem_core->di_pad, buf_core->di_pad, | |
754 | sizeof(buf_core->di_pad)); | |
755 | } else { | |
756 | memcpy(buf_core->di_pad, mem_core->di_pad, | |
757 | sizeof(buf_core->di_pad)); | |
758 | } | |
759 | ||
760 | INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch); | |
761 | ||
762 | INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec, | |
763 | dir, arch); | |
764 | INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec, | |
765 | dir, arch); | |
766 | INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec, | |
767 | dir, arch); | |
768 | INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec, | |
769 | dir, arch); | |
770 | INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec, | |
771 | dir, arch); | |
772 | INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec, | |
773 | dir, arch); | |
774 | INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch); | |
775 | INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch); | |
776 | INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch); | |
777 | INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch); | |
778 | INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch); | |
779 | INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch); | |
780 | INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch); | |
781 | INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch); | |
782 | INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch); | |
783 | INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch); | |
784 | INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch); | |
785 | } | |
786 | ||
787 | STATIC uint | |
788 | _xfs_dic2xflags( | |
1da177e4 LT |
789 | __uint16_t di_flags) |
790 | { | |
791 | uint flags = 0; | |
792 | ||
793 | if (di_flags & XFS_DIFLAG_ANY) { | |
794 | if (di_flags & XFS_DIFLAG_REALTIME) | |
795 | flags |= XFS_XFLAG_REALTIME; | |
796 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
797 | flags |= XFS_XFLAG_PREALLOC; | |
798 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
799 | flags |= XFS_XFLAG_IMMUTABLE; | |
800 | if (di_flags & XFS_DIFLAG_APPEND) | |
801 | flags |= XFS_XFLAG_APPEND; | |
802 | if (di_flags & XFS_DIFLAG_SYNC) | |
803 | flags |= XFS_XFLAG_SYNC; | |
804 | if (di_flags & XFS_DIFLAG_NOATIME) | |
805 | flags |= XFS_XFLAG_NOATIME; | |
806 | if (di_flags & XFS_DIFLAG_NODUMP) | |
807 | flags |= XFS_XFLAG_NODUMP; | |
808 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
809 | flags |= XFS_XFLAG_RTINHERIT; | |
810 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
811 | flags |= XFS_XFLAG_PROJINHERIT; | |
812 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
813 | flags |= XFS_XFLAG_NOSYMLINKS; | |
dd9f438e NS |
814 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
815 | flags |= XFS_XFLAG_EXTSIZE; | |
816 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) | |
817 | flags |= XFS_XFLAG_EXTSZINHERIT; | |
d3446eac BN |
818 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
819 | flags |= XFS_XFLAG_NODEFRAG; | |
1da177e4 LT |
820 | } |
821 | ||
822 | return flags; | |
823 | } | |
824 | ||
825 | uint | |
826 | xfs_ip2xflags( | |
827 | xfs_inode_t *ip) | |
828 | { | |
829 | xfs_dinode_core_t *dic = &ip->i_d; | |
830 | ||
a916e2bd NS |
831 | return _xfs_dic2xflags(dic->di_flags) | |
832 | (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0); | |
1da177e4 LT |
833 | } |
834 | ||
835 | uint | |
836 | xfs_dic2xflags( | |
837 | xfs_dinode_core_t *dic) | |
838 | { | |
a916e2bd NS |
839 | return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) | |
840 | (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0); | |
1da177e4 LT |
841 | } |
842 | ||
843 | /* | |
844 | * Given a mount structure and an inode number, return a pointer | |
c41564b5 | 845 | * to a newly allocated in-core inode corresponding to the given |
1da177e4 LT |
846 | * inode number. |
847 | * | |
848 | * Initialize the inode's attributes and extent pointers if it | |
849 | * already has them (it will not if the inode has no links). | |
850 | */ | |
851 | int | |
852 | xfs_iread( | |
853 | xfs_mount_t *mp, | |
854 | xfs_trans_t *tp, | |
855 | xfs_ino_t ino, | |
856 | xfs_inode_t **ipp, | |
857 | xfs_daddr_t bno) | |
858 | { | |
859 | xfs_buf_t *bp; | |
860 | xfs_dinode_t *dip; | |
861 | xfs_inode_t *ip; | |
862 | int error; | |
863 | ||
864 | ASSERT(xfs_inode_zone != NULL); | |
865 | ||
866 | ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP); | |
867 | ip->i_ino = ino; | |
868 | ip->i_mount = mp; | |
869 | ||
870 | /* | |
871 | * Get pointer's to the on-disk inode and the buffer containing it. | |
872 | * If the inode number refers to a block outside the file system | |
873 | * then xfs_itobp() will return NULL. In this case we should | |
874 | * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will | |
875 | * know that this is a new incore inode. | |
876 | */ | |
b12dd342 NS |
877 | error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, 0); |
878 | if (error) { | |
1da177e4 LT |
879 | kmem_zone_free(xfs_inode_zone, ip); |
880 | return error; | |
881 | } | |
882 | ||
883 | /* | |
884 | * Initialize inode's trace buffers. | |
885 | * Do this before xfs_iformat in case it adds entries. | |
886 | */ | |
887 | #ifdef XFS_BMAP_TRACE | |
888 | ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP); | |
889 | #endif | |
890 | #ifdef XFS_BMBT_TRACE | |
891 | ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP); | |
892 | #endif | |
893 | #ifdef XFS_RW_TRACE | |
894 | ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP); | |
895 | #endif | |
896 | #ifdef XFS_ILOCK_TRACE | |
897 | ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP); | |
898 | #endif | |
899 | #ifdef XFS_DIR2_TRACE | |
900 | ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP); | |
901 | #endif | |
902 | ||
903 | /* | |
904 | * If we got something that isn't an inode it means someone | |
905 | * (nfs or dmi) has a stale handle. | |
906 | */ | |
907 | if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) { | |
908 | kmem_zone_free(xfs_inode_zone, ip); | |
909 | xfs_trans_brelse(tp, bp); | |
910 | #ifdef DEBUG | |
911 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " | |
912 | "dip->di_core.di_magic (0x%x) != " | |
913 | "XFS_DINODE_MAGIC (0x%x)", | |
914 | INT_GET(dip->di_core.di_magic, ARCH_CONVERT), | |
915 | XFS_DINODE_MAGIC); | |
916 | #endif /* DEBUG */ | |
917 | return XFS_ERROR(EINVAL); | |
918 | } | |
919 | ||
920 | /* | |
921 | * If the on-disk inode is already linked to a directory | |
922 | * entry, copy all of the inode into the in-core inode. | |
923 | * xfs_iformat() handles copying in the inode format | |
924 | * specific information. | |
925 | * Otherwise, just get the truly permanent information. | |
926 | */ | |
927 | if (dip->di_core.di_mode) { | |
928 | xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core, | |
929 | &(ip->i_d), 1); | |
930 | error = xfs_iformat(ip, dip); | |
931 | if (error) { | |
932 | kmem_zone_free(xfs_inode_zone, ip); | |
933 | xfs_trans_brelse(tp, bp); | |
934 | #ifdef DEBUG | |
935 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " | |
936 | "xfs_iformat() returned error %d", | |
937 | error); | |
938 | #endif /* DEBUG */ | |
939 | return error; | |
940 | } | |
941 | } else { | |
942 | ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT); | |
943 | ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT); | |
944 | ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT); | |
945 | ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT); | |
946 | /* | |
947 | * Make sure to pull in the mode here as well in | |
948 | * case the inode is released without being used. | |
949 | * This ensures that xfs_inactive() will see that | |
950 | * the inode is already free and not try to mess | |
951 | * with the uninitialized part of it. | |
952 | */ | |
953 | ip->i_d.di_mode = 0; | |
954 | /* | |
955 | * Initialize the per-fork minima and maxima for a new | |
956 | * inode here. xfs_iformat will do it for old inodes. | |
957 | */ | |
958 | ip->i_df.if_ext_max = | |
959 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
960 | } | |
961 | ||
962 | INIT_LIST_HEAD(&ip->i_reclaim); | |
963 | ||
964 | /* | |
965 | * The inode format changed when we moved the link count and | |
966 | * made it 32 bits long. If this is an old format inode, | |
967 | * convert it in memory to look like a new one. If it gets | |
968 | * flushed to disk we will convert back before flushing or | |
969 | * logging it. We zero out the new projid field and the old link | |
970 | * count field. We'll handle clearing the pad field (the remains | |
971 | * of the old uuid field) when we actually convert the inode to | |
972 | * the new format. We don't change the version number so that we | |
973 | * can distinguish this from a real new format inode. | |
974 | */ | |
975 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
976 | ip->i_d.di_nlink = ip->i_d.di_onlink; | |
977 | ip->i_d.di_onlink = 0; | |
978 | ip->i_d.di_projid = 0; | |
979 | } | |
980 | ||
981 | ip->i_delayed_blks = 0; | |
982 | ||
983 | /* | |
984 | * Mark the buffer containing the inode as something to keep | |
985 | * around for a while. This helps to keep recently accessed | |
986 | * meta-data in-core longer. | |
987 | */ | |
988 | XFS_BUF_SET_REF(bp, XFS_INO_REF); | |
989 | ||
990 | /* | |
991 | * Use xfs_trans_brelse() to release the buffer containing the | |
992 | * on-disk inode, because it was acquired with xfs_trans_read_buf() | |
993 | * in xfs_itobp() above. If tp is NULL, this is just a normal | |
994 | * brelse(). If we're within a transaction, then xfs_trans_brelse() | |
995 | * will only release the buffer if it is not dirty within the | |
996 | * transaction. It will be OK to release the buffer in this case, | |
997 | * because inodes on disk are never destroyed and we will be | |
998 | * locking the new in-core inode before putting it in the hash | |
999 | * table where other processes can find it. Thus we don't have | |
1000 | * to worry about the inode being changed just because we released | |
1001 | * the buffer. | |
1002 | */ | |
1003 | xfs_trans_brelse(tp, bp); | |
1004 | *ipp = ip; | |
1005 | return 0; | |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * Read in extents from a btree-format inode. | |
1010 | * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. | |
1011 | */ | |
1012 | int | |
1013 | xfs_iread_extents( | |
1014 | xfs_trans_t *tp, | |
1015 | xfs_inode_t *ip, | |
1016 | int whichfork) | |
1017 | { | |
1018 | int error; | |
1019 | xfs_ifork_t *ifp; | |
4eea22f0 | 1020 | xfs_extnum_t nextents; |
1da177e4 LT |
1021 | size_t size; |
1022 | ||
1023 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1024 | XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, | |
1025 | ip->i_mount); | |
1026 | return XFS_ERROR(EFSCORRUPTED); | |
1027 | } | |
4eea22f0 MK |
1028 | nextents = XFS_IFORK_NEXTENTS(ip, whichfork); |
1029 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
1da177e4 | 1030 | ifp = XFS_IFORK_PTR(ip, whichfork); |
4eea22f0 | 1031 | |
1da177e4 LT |
1032 | /* |
1033 | * We know that the size is valid (it's checked in iformat_btree) | |
1034 | */ | |
1da177e4 | 1035 | ifp->if_lastex = NULLEXTNUM; |
4eea22f0 | 1036 | ifp->if_bytes = ifp->if_real_bytes = 0; |
1da177e4 | 1037 | ifp->if_flags |= XFS_IFEXTENTS; |
4eea22f0 | 1038 | xfs_iext_add(ifp, 0, nextents); |
1da177e4 LT |
1039 | error = xfs_bmap_read_extents(tp, ip, whichfork); |
1040 | if (error) { | |
4eea22f0 | 1041 | xfs_iext_destroy(ifp); |
1da177e4 LT |
1042 | ifp->if_flags &= ~XFS_IFEXTENTS; |
1043 | return error; | |
1044 | } | |
4eea22f0 | 1045 | xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
1046 | return 0; |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Allocate an inode on disk and return a copy of its in-core version. | |
1051 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
1052 | * appropriately within the inode. The uid and gid for the inode are | |
1053 | * set according to the contents of the given cred structure. | |
1054 | * | |
1055 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
1056 | * has a free inode available, call xfs_iget() | |
1057 | * to obtain the in-core version of the allocated inode. Finally, | |
1058 | * fill in the inode and log its initial contents. In this case, | |
1059 | * ialloc_context would be set to NULL and call_again set to false. | |
1060 | * | |
1061 | * If xfs_dialloc() does not have an available inode, | |
1062 | * it will replenish its supply by doing an allocation. Since we can | |
1063 | * only do one allocation within a transaction without deadlocks, we | |
1064 | * must commit the current transaction before returning the inode itself. | |
1065 | * In this case, therefore, we will set call_again to true and return. | |
1066 | * The caller should then commit the current transaction, start a new | |
1067 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
1068 | * | |
1069 | * To ensure that some other process does not grab the inode that | |
1070 | * was allocated during the first call to xfs_ialloc(), this routine | |
1071 | * also returns the [locked] bp pointing to the head of the freelist | |
1072 | * as ialloc_context. The caller should hold this buffer across | |
1073 | * the commit and pass it back into this routine on the second call. | |
1074 | */ | |
1075 | int | |
1076 | xfs_ialloc( | |
1077 | xfs_trans_t *tp, | |
1078 | xfs_inode_t *pip, | |
1079 | mode_t mode, | |
31b084ae | 1080 | xfs_nlink_t nlink, |
1da177e4 LT |
1081 | xfs_dev_t rdev, |
1082 | cred_t *cr, | |
1083 | xfs_prid_t prid, | |
1084 | int okalloc, | |
1085 | xfs_buf_t **ialloc_context, | |
1086 | boolean_t *call_again, | |
1087 | xfs_inode_t **ipp) | |
1088 | { | |
1089 | xfs_ino_t ino; | |
1090 | xfs_inode_t *ip; | |
67fcaa73 | 1091 | bhv_vnode_t *vp; |
1da177e4 LT |
1092 | uint flags; |
1093 | int error; | |
1094 | ||
1095 | /* | |
1096 | * Call the space management code to pick | |
1097 | * the on-disk inode to be allocated. | |
1098 | */ | |
1099 | error = xfs_dialloc(tp, pip->i_ino, mode, okalloc, | |
1100 | ialloc_context, call_again, &ino); | |
1101 | if (error != 0) { | |
1102 | return error; | |
1103 | } | |
1104 | if (*call_again || ino == NULLFSINO) { | |
1105 | *ipp = NULL; | |
1106 | return 0; | |
1107 | } | |
1108 | ASSERT(*ialloc_context == NULL); | |
1109 | ||
1110 | /* | |
1111 | * Get the in-core inode with the lock held exclusively. | |
1112 | * This is because we're setting fields here we need | |
1113 | * to prevent others from looking at until we're done. | |
1114 | */ | |
1115 | error = xfs_trans_iget(tp->t_mountp, tp, ino, | |
1116 | IGET_CREATE, XFS_ILOCK_EXCL, &ip); | |
1117 | if (error != 0) { | |
1118 | return error; | |
1119 | } | |
1120 | ASSERT(ip != NULL); | |
1121 | ||
1122 | vp = XFS_ITOV(ip); | |
1da177e4 LT |
1123 | ip->i_d.di_mode = (__uint16_t)mode; |
1124 | ip->i_d.di_onlink = 0; | |
1125 | ip->i_d.di_nlink = nlink; | |
1126 | ASSERT(ip->i_d.di_nlink == nlink); | |
1127 | ip->i_d.di_uid = current_fsuid(cr); | |
1128 | ip->i_d.di_gid = current_fsgid(cr); | |
1129 | ip->i_d.di_projid = prid; | |
1130 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
1131 | ||
1132 | /* | |
1133 | * If the superblock version is up to where we support new format | |
1134 | * inodes and this is currently an old format inode, then change | |
1135 | * the inode version number now. This way we only do the conversion | |
1136 | * here rather than here and in the flush/logging code. | |
1137 | */ | |
1138 | if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) && | |
1139 | ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
1140 | ip->i_d.di_version = XFS_DINODE_VERSION_2; | |
1141 | /* | |
1142 | * We've already zeroed the old link count, the projid field, | |
1143 | * and the pad field. | |
1144 | */ | |
1145 | } | |
1146 | ||
1147 | /* | |
1148 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
1149 | */ | |
1150 | if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1)) | |
1151 | xfs_bump_ino_vers2(tp, ip); | |
1152 | ||
1153 | if (XFS_INHERIT_GID(pip, vp->v_vfsp)) { | |
1154 | ip->i_d.di_gid = pip->i_d.di_gid; | |
1155 | if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) { | |
1156 | ip->i_d.di_mode |= S_ISGID; | |
1157 | } | |
1158 | } | |
1159 | ||
1160 | /* | |
1161 | * If the group ID of the new file does not match the effective group | |
1162 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
1163 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
1164 | */ | |
1165 | if ((irix_sgid_inherit) && | |
1166 | (ip->i_d.di_mode & S_ISGID) && | |
1167 | (!in_group_p((gid_t)ip->i_d.di_gid))) { | |
1168 | ip->i_d.di_mode &= ~S_ISGID; | |
1169 | } | |
1170 | ||
1171 | ip->i_d.di_size = 0; | |
1172 | ip->i_d.di_nextents = 0; | |
1173 | ASSERT(ip->i_d.di_nblocks == 0); | |
1174 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD); | |
1175 | /* | |
1176 | * di_gen will have been taken care of in xfs_iread. | |
1177 | */ | |
1178 | ip->i_d.di_extsize = 0; | |
1179 | ip->i_d.di_dmevmask = 0; | |
1180 | ip->i_d.di_dmstate = 0; | |
1181 | ip->i_d.di_flags = 0; | |
1182 | flags = XFS_ILOG_CORE; | |
1183 | switch (mode & S_IFMT) { | |
1184 | case S_IFIFO: | |
1185 | case S_IFCHR: | |
1186 | case S_IFBLK: | |
1187 | case S_IFSOCK: | |
1188 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1189 | ip->i_df.if_u2.if_rdev = rdev; | |
1190 | ip->i_df.if_flags = 0; | |
1191 | flags |= XFS_ILOG_DEV; | |
1192 | break; | |
1193 | case S_IFREG: | |
1194 | case S_IFDIR: | |
1195 | if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) { | |
365ca83d NS |
1196 | uint di_flags = 0; |
1197 | ||
1198 | if ((mode & S_IFMT) == S_IFDIR) { | |
1199 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) | |
1200 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
1201 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1202 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
1203 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1204 | } | |
1205 | } else if ((mode & S_IFMT) == S_IFREG) { | |
365ca83d NS |
1206 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) { |
1207 | di_flags |= XFS_DIFLAG_REALTIME; | |
1da177e4 LT |
1208 | ip->i_iocore.io_flags |= XFS_IOCORE_RT; |
1209 | } | |
dd9f438e NS |
1210 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1211 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1212 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1213 | } | |
1da177e4 LT |
1214 | } |
1215 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
1216 | xfs_inherit_noatime) | |
365ca83d | 1217 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
1218 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
1219 | xfs_inherit_nodump) | |
365ca83d | 1220 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
1221 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
1222 | xfs_inherit_sync) | |
365ca83d | 1223 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
1224 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
1225 | xfs_inherit_nosymlinks) | |
365ca83d NS |
1226 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
1227 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
1228 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
1229 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
1230 | xfs_inherit_nodefrag) | |
1231 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
365ca83d | 1232 | ip->i_d.di_flags |= di_flags; |
1da177e4 LT |
1233 | } |
1234 | /* FALLTHROUGH */ | |
1235 | case S_IFLNK: | |
1236 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
1237 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
1238 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
1239 | ip->i_df.if_u1.if_extents = NULL; | |
1240 | break; | |
1241 | default: | |
1242 | ASSERT(0); | |
1243 | } | |
1244 | /* | |
1245 | * Attribute fork settings for new inode. | |
1246 | */ | |
1247 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1248 | ip->i_d.di_anextents = 0; | |
1249 | ||
1250 | /* | |
1251 | * Log the new values stuffed into the inode. | |
1252 | */ | |
1253 | xfs_trans_log_inode(tp, ip, flags); | |
1254 | ||
b83bd138 NS |
1255 | /* now that we have an i_mode we can setup inode ops and unlock */ |
1256 | bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1); | |
1da177e4 LT |
1257 | |
1258 | *ipp = ip; | |
1259 | return 0; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * Check to make sure that there are no blocks allocated to the | |
1264 | * file beyond the size of the file. We don't check this for | |
1265 | * files with fixed size extents or real time extents, but we | |
1266 | * at least do it for regular files. | |
1267 | */ | |
1268 | #ifdef DEBUG | |
1269 | void | |
1270 | xfs_isize_check( | |
1271 | xfs_mount_t *mp, | |
1272 | xfs_inode_t *ip, | |
1273 | xfs_fsize_t isize) | |
1274 | { | |
1275 | xfs_fileoff_t map_first; | |
1276 | int nimaps; | |
1277 | xfs_bmbt_irec_t imaps[2]; | |
1278 | ||
1279 | if ((ip->i_d.di_mode & S_IFMT) != S_IFREG) | |
1280 | return; | |
1281 | ||
dd9f438e | 1282 | if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE)) |
1da177e4 LT |
1283 | return; |
1284 | ||
1285 | nimaps = 2; | |
1286 | map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
1287 | /* | |
1288 | * The filesystem could be shutting down, so bmapi may return | |
1289 | * an error. | |
1290 | */ | |
1291 | if (xfs_bmapi(NULL, ip, map_first, | |
1292 | (XFS_B_TO_FSB(mp, | |
1293 | (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) - | |
1294 | map_first), | |
1295 | XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps, | |
3e57ecf6 | 1296 | NULL, NULL)) |
1da177e4 LT |
1297 | return; |
1298 | ASSERT(nimaps == 1); | |
1299 | ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK); | |
1300 | } | |
1301 | #endif /* DEBUG */ | |
1302 | ||
1303 | /* | |
1304 | * Calculate the last possible buffered byte in a file. This must | |
1305 | * include data that was buffered beyond the EOF by the write code. | |
1306 | * This also needs to deal with overflowing the xfs_fsize_t type | |
1307 | * which can happen for sizes near the limit. | |
1308 | * | |
1309 | * We also need to take into account any blocks beyond the EOF. It | |
1310 | * may be the case that they were buffered by a write which failed. | |
1311 | * In that case the pages will still be in memory, but the inode size | |
1312 | * will never have been updated. | |
1313 | */ | |
1314 | xfs_fsize_t | |
1315 | xfs_file_last_byte( | |
1316 | xfs_inode_t *ip) | |
1317 | { | |
1318 | xfs_mount_t *mp; | |
1319 | xfs_fsize_t last_byte; | |
1320 | xfs_fileoff_t last_block; | |
1321 | xfs_fileoff_t size_last_block; | |
1322 | int error; | |
1323 | ||
1324 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS)); | |
1325 | ||
1326 | mp = ip->i_mount; | |
1327 | /* | |
1328 | * Only check for blocks beyond the EOF if the extents have | |
1329 | * been read in. This eliminates the need for the inode lock, | |
1330 | * and it also saves us from looking when it really isn't | |
1331 | * necessary. | |
1332 | */ | |
1333 | if (ip->i_df.if_flags & XFS_IFEXTENTS) { | |
1334 | error = xfs_bmap_last_offset(NULL, ip, &last_block, | |
1335 | XFS_DATA_FORK); | |
1336 | if (error) { | |
1337 | last_block = 0; | |
1338 | } | |
1339 | } else { | |
1340 | last_block = 0; | |
1341 | } | |
1342 | size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size); | |
1343 | last_block = XFS_FILEOFF_MAX(last_block, size_last_block); | |
1344 | ||
1345 | last_byte = XFS_FSB_TO_B(mp, last_block); | |
1346 | if (last_byte < 0) { | |
1347 | return XFS_MAXIOFFSET(mp); | |
1348 | } | |
1349 | last_byte += (1 << mp->m_writeio_log); | |
1350 | if (last_byte < 0) { | |
1351 | return XFS_MAXIOFFSET(mp); | |
1352 | } | |
1353 | return last_byte; | |
1354 | } | |
1355 | ||
1356 | #if defined(XFS_RW_TRACE) | |
1357 | STATIC void | |
1358 | xfs_itrunc_trace( | |
1359 | int tag, | |
1360 | xfs_inode_t *ip, | |
1361 | int flag, | |
1362 | xfs_fsize_t new_size, | |
1363 | xfs_off_t toss_start, | |
1364 | xfs_off_t toss_finish) | |
1365 | { | |
1366 | if (ip->i_rwtrace == NULL) { | |
1367 | return; | |
1368 | } | |
1369 | ||
1370 | ktrace_enter(ip->i_rwtrace, | |
1371 | (void*)((long)tag), | |
1372 | (void*)ip, | |
1373 | (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff), | |
1374 | (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff), | |
1375 | (void*)((long)flag), | |
1376 | (void*)(unsigned long)((new_size >> 32) & 0xffffffff), | |
1377 | (void*)(unsigned long)(new_size & 0xffffffff), | |
1378 | (void*)(unsigned long)((toss_start >> 32) & 0xffffffff), | |
1379 | (void*)(unsigned long)(toss_start & 0xffffffff), | |
1380 | (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff), | |
1381 | (void*)(unsigned long)(toss_finish & 0xffffffff), | |
1382 | (void*)(unsigned long)current_cpu(), | |
f1fdc848 YL |
1383 | (void*)(unsigned long)current_pid(), |
1384 | (void*)NULL, | |
1385 | (void*)NULL, | |
1386 | (void*)NULL); | |
1da177e4 LT |
1387 | } |
1388 | #else | |
1389 | #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish) | |
1390 | #endif | |
1391 | ||
1392 | /* | |
1393 | * Start the truncation of the file to new_size. The new size | |
1394 | * must be smaller than the current size. This routine will | |
1395 | * clear the buffer and page caches of file data in the removed | |
1396 | * range, and xfs_itruncate_finish() will remove the underlying | |
1397 | * disk blocks. | |
1398 | * | |
1399 | * The inode must have its I/O lock locked EXCLUSIVELY, and it | |
1400 | * must NOT have the inode lock held at all. This is because we're | |
1401 | * calling into the buffer/page cache code and we can't hold the | |
1402 | * inode lock when we do so. | |
1403 | * | |
38e2299a DC |
1404 | * We need to wait for any direct I/Os in flight to complete before we |
1405 | * proceed with the truncate. This is needed to prevent the extents | |
1406 | * being read or written by the direct I/Os from being removed while the | |
1407 | * I/O is in flight as there is no other method of synchronising | |
1408 | * direct I/O with the truncate operation. Also, because we hold | |
1409 | * the IOLOCK in exclusive mode, we prevent new direct I/Os from being | |
1410 | * started until the truncate completes and drops the lock. Essentially, | |
1411 | * the vn_iowait() call forms an I/O barrier that provides strict ordering | |
1412 | * between direct I/Os and the truncate operation. | |
1413 | * | |
1da177e4 LT |
1414 | * The flags parameter can have either the value XFS_ITRUNC_DEFINITE |
1415 | * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used | |
1416 | * in the case that the caller is locking things out of order and | |
1417 | * may not be able to call xfs_itruncate_finish() with the inode lock | |
1418 | * held without dropping the I/O lock. If the caller must drop the | |
1419 | * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start() | |
1420 | * must be called again with all the same restrictions as the initial | |
1421 | * call. | |
1422 | */ | |
1423 | void | |
1424 | xfs_itruncate_start( | |
1425 | xfs_inode_t *ip, | |
1426 | uint flags, | |
1427 | xfs_fsize_t new_size) | |
1428 | { | |
1429 | xfs_fsize_t last_byte; | |
1430 | xfs_off_t toss_start; | |
1431 | xfs_mount_t *mp; | |
67fcaa73 | 1432 | bhv_vnode_t *vp; |
1da177e4 LT |
1433 | |
1434 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0); | |
1435 | ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size)); | |
1436 | ASSERT((flags == XFS_ITRUNC_DEFINITE) || | |
1437 | (flags == XFS_ITRUNC_MAYBE)); | |
1438 | ||
1439 | mp = ip->i_mount; | |
1440 | vp = XFS_ITOV(ip); | |
9fa8046f YL |
1441 | |
1442 | vn_iowait(vp); /* wait for the completion of any pending DIOs */ | |
1443 | ||
1da177e4 | 1444 | /* |
67fcaa73 | 1445 | * Call toss_pages or flushinval_pages to get rid of pages |
1da177e4 | 1446 | * overlapping the region being removed. We have to use |
67fcaa73 | 1447 | * the less efficient flushinval_pages in the case that the |
1da177e4 LT |
1448 | * caller may not be able to finish the truncate without |
1449 | * dropping the inode's I/O lock. Make sure | |
1450 | * to catch any pages brought in by buffers overlapping | |
1451 | * the EOF by searching out beyond the isize by our | |
1452 | * block size. We round new_size up to a block boundary | |
1453 | * so that we don't toss things on the same block as | |
1454 | * new_size but before it. | |
1455 | * | |
67fcaa73 | 1456 | * Before calling toss_page or flushinval_pages, make sure to |
1da177e4 LT |
1457 | * call remapf() over the same region if the file is mapped. |
1458 | * This frees up mapped file references to the pages in the | |
67fcaa73 | 1459 | * given range and for the flushinval_pages case it ensures |
1da177e4 LT |
1460 | * that we get the latest mapped changes flushed out. |
1461 | */ | |
1462 | toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); | |
1463 | toss_start = XFS_FSB_TO_B(mp, toss_start); | |
1464 | if (toss_start < 0) { | |
1465 | /* | |
1466 | * The place to start tossing is beyond our maximum | |
1467 | * file size, so there is no way that the data extended | |
1468 | * out there. | |
1469 | */ | |
1470 | return; | |
1471 | } | |
1472 | last_byte = xfs_file_last_byte(ip); | |
1473 | xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start, | |
1474 | last_byte); | |
1475 | if (last_byte > toss_start) { | |
1476 | if (flags & XFS_ITRUNC_DEFINITE) { | |
67fcaa73 | 1477 | bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED); |
1da177e4 | 1478 | } else { |
67fcaa73 | 1479 | bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED); |
1da177e4 LT |
1480 | } |
1481 | } | |
1482 | ||
1483 | #ifdef DEBUG | |
1484 | if (new_size == 0) { | |
1485 | ASSERT(VN_CACHED(vp) == 0); | |
1486 | } | |
1487 | #endif | |
1488 | } | |
1489 | ||
1490 | /* | |
1491 | * Shrink the file to the given new_size. The new | |
1492 | * size must be smaller than the current size. | |
1493 | * This will free up the underlying blocks | |
1494 | * in the removed range after a call to xfs_itruncate_start() | |
1495 | * or xfs_atruncate_start(). | |
1496 | * | |
1497 | * The transaction passed to this routine must have made | |
1498 | * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES. | |
1499 | * This routine may commit the given transaction and | |
1500 | * start new ones, so make sure everything involved in | |
1501 | * the transaction is tidy before calling here. | |
1502 | * Some transaction will be returned to the caller to be | |
1503 | * committed. The incoming transaction must already include | |
1504 | * the inode, and both inode locks must be held exclusively. | |
1505 | * The inode must also be "held" within the transaction. On | |
1506 | * return the inode will be "held" within the returned transaction. | |
1507 | * This routine does NOT require any disk space to be reserved | |
1508 | * for it within the transaction. | |
1509 | * | |
1510 | * The fork parameter must be either xfs_attr_fork or xfs_data_fork, | |
1511 | * and it indicates the fork which is to be truncated. For the | |
1512 | * attribute fork we only support truncation to size 0. | |
1513 | * | |
1514 | * We use the sync parameter to indicate whether or not the first | |
1515 | * transaction we perform might have to be synchronous. For the attr fork, | |
1516 | * it needs to be so if the unlink of the inode is not yet known to be | |
1517 | * permanent in the log. This keeps us from freeing and reusing the | |
1518 | * blocks of the attribute fork before the unlink of the inode becomes | |
1519 | * permanent. | |
1520 | * | |
1521 | * For the data fork, we normally have to run synchronously if we're | |
1522 | * being called out of the inactive path or we're being called | |
1523 | * out of the create path where we're truncating an existing file. | |
1524 | * Either way, the truncate needs to be sync so blocks don't reappear | |
1525 | * in the file with altered data in case of a crash. wsync filesystems | |
1526 | * can run the first case async because anything that shrinks the inode | |
1527 | * has to run sync so by the time we're called here from inactive, the | |
1528 | * inode size is permanently set to 0. | |
1529 | * | |
1530 | * Calls from the truncate path always need to be sync unless we're | |
1531 | * in a wsync filesystem and the file has already been unlinked. | |
1532 | * | |
1533 | * The caller is responsible for correctly setting the sync parameter. | |
1534 | * It gets too hard for us to guess here which path we're being called | |
1535 | * out of just based on inode state. | |
1536 | */ | |
1537 | int | |
1538 | xfs_itruncate_finish( | |
1539 | xfs_trans_t **tp, | |
1540 | xfs_inode_t *ip, | |
1541 | xfs_fsize_t new_size, | |
1542 | int fork, | |
1543 | int sync) | |
1544 | { | |
1545 | xfs_fsblock_t first_block; | |
1546 | xfs_fileoff_t first_unmap_block; | |
1547 | xfs_fileoff_t last_block; | |
1548 | xfs_filblks_t unmap_len=0; | |
1549 | xfs_mount_t *mp; | |
1550 | xfs_trans_t *ntp; | |
1551 | int done; | |
1552 | int committed; | |
1553 | xfs_bmap_free_t free_list; | |
1554 | int error; | |
1555 | ||
1556 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0); | |
1557 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0); | |
1558 | ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size)); | |
1559 | ASSERT(*tp != NULL); | |
1560 | ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1561 | ASSERT(ip->i_transp == *tp); | |
1562 | ASSERT(ip->i_itemp != NULL); | |
1563 | ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD); | |
1564 | ||
1565 | ||
1566 | ntp = *tp; | |
1567 | mp = (ntp)->t_mountp; | |
1568 | ASSERT(! XFS_NOT_DQATTACHED(mp, ip)); | |
1569 | ||
1570 | /* | |
1571 | * We only support truncating the entire attribute fork. | |
1572 | */ | |
1573 | if (fork == XFS_ATTR_FORK) { | |
1574 | new_size = 0LL; | |
1575 | } | |
1576 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); | |
1577 | xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0); | |
1578 | /* | |
1579 | * The first thing we do is set the size to new_size permanently | |
1580 | * on disk. This way we don't have to worry about anyone ever | |
1581 | * being able to look at the data being freed even in the face | |
1582 | * of a crash. What we're getting around here is the case where | |
1583 | * we free a block, it is allocated to another file, it is written | |
1584 | * to, and then we crash. If the new data gets written to the | |
1585 | * file but the log buffers containing the free and reallocation | |
1586 | * don't, then we'd end up with garbage in the blocks being freed. | |
1587 | * As long as we make the new_size permanent before actually | |
1588 | * freeing any blocks it doesn't matter if they get writtten to. | |
1589 | * | |
1590 | * The callers must signal into us whether or not the size | |
1591 | * setting here must be synchronous. There are a few cases | |
1592 | * where it doesn't have to be synchronous. Those cases | |
1593 | * occur if the file is unlinked and we know the unlink is | |
1594 | * permanent or if the blocks being truncated are guaranteed | |
1595 | * to be beyond the inode eof (regardless of the link count) | |
1596 | * and the eof value is permanent. Both of these cases occur | |
1597 | * only on wsync-mounted filesystems. In those cases, we're | |
1598 | * guaranteed that no user will ever see the data in the blocks | |
1599 | * that are being truncated so the truncate can run async. | |
1600 | * In the free beyond eof case, the file may wind up with | |
1601 | * more blocks allocated to it than it needs if we crash | |
1602 | * and that won't get fixed until the next time the file | |
1603 | * is re-opened and closed but that's ok as that shouldn't | |
1604 | * be too many blocks. | |
1605 | * | |
1606 | * However, we can't just make all wsync xactions run async | |
1607 | * because there's one call out of the create path that needs | |
1608 | * to run sync where it's truncating an existing file to size | |
1609 | * 0 whose size is > 0. | |
1610 | * | |
1611 | * It's probably possible to come up with a test in this | |
1612 | * routine that would correctly distinguish all the above | |
1613 | * cases from the values of the function parameters and the | |
1614 | * inode state but for sanity's sake, I've decided to let the | |
1615 | * layers above just tell us. It's simpler to correctly figure | |
1616 | * out in the layer above exactly under what conditions we | |
1617 | * can run async and I think it's easier for others read and | |
1618 | * follow the logic in case something has to be changed. | |
1619 | * cscope is your friend -- rcc. | |
1620 | * | |
1621 | * The attribute fork is much simpler. | |
1622 | * | |
1623 | * For the attribute fork we allow the caller to tell us whether | |
1624 | * the unlink of the inode that led to this call is yet permanent | |
1625 | * in the on disk log. If it is not and we will be freeing extents | |
1626 | * in this inode then we make the first transaction synchronous | |
1627 | * to make sure that the unlink is permanent by the time we free | |
1628 | * the blocks. | |
1629 | */ | |
1630 | if (fork == XFS_DATA_FORK) { | |
1631 | if (ip->i_d.di_nextents > 0) { | |
1632 | ip->i_d.di_size = new_size; | |
1633 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1634 | } | |
1635 | } else if (sync) { | |
1636 | ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC)); | |
1637 | if (ip->i_d.di_anextents > 0) | |
1638 | xfs_trans_set_sync(ntp); | |
1639 | } | |
1640 | ASSERT(fork == XFS_DATA_FORK || | |
1641 | (fork == XFS_ATTR_FORK && | |
1642 | ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) || | |
1643 | (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC))))); | |
1644 | ||
1645 | /* | |
1646 | * Since it is possible for space to become allocated beyond | |
1647 | * the end of the file (in a crash where the space is allocated | |
1648 | * but the inode size is not yet updated), simply remove any | |
1649 | * blocks which show up between the new EOF and the maximum | |
1650 | * possible file size. If the first block to be removed is | |
1651 | * beyond the maximum file size (ie it is the same as last_block), | |
1652 | * then there is nothing to do. | |
1653 | */ | |
1654 | last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); | |
1655 | ASSERT(first_unmap_block <= last_block); | |
1656 | done = 0; | |
1657 | if (last_block == first_unmap_block) { | |
1658 | done = 1; | |
1659 | } else { | |
1660 | unmap_len = last_block - first_unmap_block + 1; | |
1661 | } | |
1662 | while (!done) { | |
1663 | /* | |
1664 | * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi() | |
1665 | * will tell us whether it freed the entire range or | |
1666 | * not. If this is a synchronous mount (wsync), | |
1667 | * then we can tell bunmapi to keep all the | |
1668 | * transactions asynchronous since the unlink | |
1669 | * transaction that made this inode inactive has | |
1670 | * already hit the disk. There's no danger of | |
1671 | * the freed blocks being reused, there being a | |
1672 | * crash, and the reused blocks suddenly reappearing | |
1673 | * in this file with garbage in them once recovery | |
1674 | * runs. | |
1675 | */ | |
1676 | XFS_BMAP_INIT(&free_list, &first_block); | |
3e57ecf6 OW |
1677 | error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore, |
1678 | first_unmap_block, unmap_len, | |
1da177e4 LT |
1679 | XFS_BMAPI_AFLAG(fork) | |
1680 | (sync ? 0 : XFS_BMAPI_ASYNC), | |
1681 | XFS_ITRUNC_MAX_EXTENTS, | |
3e57ecf6 OW |
1682 | &first_block, &free_list, |
1683 | NULL, &done); | |
1da177e4 LT |
1684 | if (error) { |
1685 | /* | |
1686 | * If the bunmapi call encounters an error, | |
1687 | * return to the caller where the transaction | |
1688 | * can be properly aborted. We just need to | |
1689 | * make sure we're not holding any resources | |
1690 | * that we were not when we came in. | |
1691 | */ | |
1692 | xfs_bmap_cancel(&free_list); | |
1693 | return error; | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * Duplicate the transaction that has the permanent | |
1698 | * reservation and commit the old transaction. | |
1699 | */ | |
1700 | error = xfs_bmap_finish(tp, &free_list, first_block, | |
1701 | &committed); | |
1702 | ntp = *tp; | |
1703 | if (error) { | |
1704 | /* | |
1705 | * If the bmap finish call encounters an error, | |
1706 | * return to the caller where the transaction | |
1707 | * can be properly aborted. We just need to | |
1708 | * make sure we're not holding any resources | |
1709 | * that we were not when we came in. | |
1710 | * | |
1711 | * Aborting from this point might lose some | |
1712 | * blocks in the file system, but oh well. | |
1713 | */ | |
1714 | xfs_bmap_cancel(&free_list); | |
1715 | if (committed) { | |
1716 | /* | |
1717 | * If the passed in transaction committed | |
1718 | * in xfs_bmap_finish(), then we want to | |
1719 | * add the inode to this one before returning. | |
1720 | * This keeps things simple for the higher | |
1721 | * level code, because it always knows that | |
1722 | * the inode is locked and held in the | |
1723 | * transaction that returns to it whether | |
1724 | * errors occur or not. We don't mark the | |
1725 | * inode dirty so that this transaction can | |
1726 | * be easily aborted if possible. | |
1727 | */ | |
1728 | xfs_trans_ijoin(ntp, ip, | |
1729 | XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1730 | xfs_trans_ihold(ntp, ip); | |
1731 | } | |
1732 | return error; | |
1733 | } | |
1734 | ||
1735 | if (committed) { | |
1736 | /* | |
1737 | * The first xact was committed, | |
1738 | * so add the inode to the new one. | |
1739 | * Mark it dirty so it will be logged | |
1740 | * and moved forward in the log as | |
1741 | * part of every commit. | |
1742 | */ | |
1743 | xfs_trans_ijoin(ntp, ip, | |
1744 | XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1745 | xfs_trans_ihold(ntp, ip); | |
1746 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1747 | } | |
1748 | ntp = xfs_trans_dup(ntp); | |
1749 | (void) xfs_trans_commit(*tp, 0, NULL); | |
1750 | *tp = ntp; | |
1751 | error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1752 | XFS_TRANS_PERM_LOG_RES, | |
1753 | XFS_ITRUNCATE_LOG_COUNT); | |
1754 | /* | |
1755 | * Add the inode being truncated to the next chained | |
1756 | * transaction. | |
1757 | */ | |
1758 | xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1759 | xfs_trans_ihold(ntp, ip); | |
1760 | if (error) | |
1761 | return (error); | |
1762 | } | |
1763 | /* | |
1764 | * Only update the size in the case of the data fork, but | |
1765 | * always re-log the inode so that our permanent transaction | |
1766 | * can keep on rolling it forward in the log. | |
1767 | */ | |
1768 | if (fork == XFS_DATA_FORK) { | |
1769 | xfs_isize_check(mp, ip, new_size); | |
1770 | ip->i_d.di_size = new_size; | |
1771 | } | |
1772 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1773 | ASSERT((new_size != 0) || | |
1774 | (fork == XFS_ATTR_FORK) || | |
1775 | (ip->i_delayed_blks == 0)); | |
1776 | ASSERT((new_size != 0) || | |
1777 | (fork == XFS_ATTR_FORK) || | |
1778 | (ip->i_d.di_nextents == 0)); | |
1779 | xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0); | |
1780 | return 0; | |
1781 | } | |
1782 | ||
1783 | ||
1784 | /* | |
1785 | * xfs_igrow_start | |
1786 | * | |
1787 | * Do the first part of growing a file: zero any data in the last | |
1788 | * block that is beyond the old EOF. We need to do this before | |
1789 | * the inode is joined to the transaction to modify the i_size. | |
1790 | * That way we can drop the inode lock and call into the buffer | |
1791 | * cache to get the buffer mapping the EOF. | |
1792 | */ | |
1793 | int | |
1794 | xfs_igrow_start( | |
1795 | xfs_inode_t *ip, | |
1796 | xfs_fsize_t new_size, | |
1797 | cred_t *credp) | |
1798 | { | |
1da177e4 LT |
1799 | int error; |
1800 | ||
1801 | ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0); | |
1802 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0); | |
1803 | ASSERT(new_size > ip->i_d.di_size); | |
1804 | ||
1da177e4 LT |
1805 | /* |
1806 | * Zero any pages that may have been created by | |
1807 | * xfs_write_file() beyond the end of the file | |
1808 | * and any blocks between the old and new file sizes. | |
1809 | */ | |
24ee8088 ES |
1810 | error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size, |
1811 | ip->i_d.di_size, new_size); | |
1da177e4 LT |
1812 | return error; |
1813 | } | |
1814 | ||
1815 | /* | |
1816 | * xfs_igrow_finish | |
1817 | * | |
1818 | * This routine is called to extend the size of a file. | |
1819 | * The inode must have both the iolock and the ilock locked | |
1820 | * for update and it must be a part of the current transaction. | |
1821 | * The xfs_igrow_start() function must have been called previously. | |
1822 | * If the change_flag is not zero, the inode change timestamp will | |
1823 | * be updated. | |
1824 | */ | |
1825 | void | |
1826 | xfs_igrow_finish( | |
1827 | xfs_trans_t *tp, | |
1828 | xfs_inode_t *ip, | |
1829 | xfs_fsize_t new_size, | |
1830 | int change_flag) | |
1831 | { | |
1832 | ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0); | |
1833 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0); | |
1834 | ASSERT(ip->i_transp == tp); | |
1835 | ASSERT(new_size > ip->i_d.di_size); | |
1836 | ||
1837 | /* | |
1838 | * Update the file size. Update the inode change timestamp | |
1839 | * if change_flag set. | |
1840 | */ | |
1841 | ip->i_d.di_size = new_size; | |
1842 | if (change_flag) | |
1843 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG); | |
1844 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1845 | ||
1846 | } | |
1847 | ||
1848 | ||
1849 | /* | |
1850 | * This is called when the inode's link count goes to 0. | |
1851 | * We place the on-disk inode on a list in the AGI. It | |
1852 | * will be pulled from this list when the inode is freed. | |
1853 | */ | |
1854 | int | |
1855 | xfs_iunlink( | |
1856 | xfs_trans_t *tp, | |
1857 | xfs_inode_t *ip) | |
1858 | { | |
1859 | xfs_mount_t *mp; | |
1860 | xfs_agi_t *agi; | |
1861 | xfs_dinode_t *dip; | |
1862 | xfs_buf_t *agibp; | |
1863 | xfs_buf_t *ibp; | |
1864 | xfs_agnumber_t agno; | |
1865 | xfs_daddr_t agdaddr; | |
1866 | xfs_agino_t agino; | |
1867 | short bucket_index; | |
1868 | int offset; | |
1869 | int error; | |
1870 | int agi_ok; | |
1871 | ||
1872 | ASSERT(ip->i_d.di_nlink == 0); | |
1873 | ASSERT(ip->i_d.di_mode != 0); | |
1874 | ASSERT(ip->i_transp == tp); | |
1875 | ||
1876 | mp = tp->t_mountp; | |
1877 | ||
1878 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); | |
1879 | agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
1880 | ||
1881 | /* | |
1882 | * Get the agi buffer first. It ensures lock ordering | |
1883 | * on the list. | |
1884 | */ | |
1885 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, | |
1886 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
1887 | if (error) { | |
1888 | return error; | |
1889 | } | |
1890 | /* | |
1891 | * Validate the magic number of the agi block. | |
1892 | */ | |
1893 | agi = XFS_BUF_TO_AGI(agibp); | |
1894 | agi_ok = | |
16259e7d CH |
1895 | be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC && |
1896 | XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)); | |
1da177e4 LT |
1897 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK, |
1898 | XFS_RANDOM_IUNLINK))) { | |
1899 | XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi); | |
1900 | xfs_trans_brelse(tp, agibp); | |
1901 | return XFS_ERROR(EFSCORRUPTED); | |
1902 | } | |
1903 | /* | |
1904 | * Get the index into the agi hash table for the | |
1905 | * list this inode will go on. | |
1906 | */ | |
1907 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1908 | ASSERT(agino != 0); | |
1909 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1910 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1911 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1912 | |
16259e7d | 1913 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) { |
1da177e4 LT |
1914 | /* |
1915 | * There is already another inode in the bucket we need | |
1916 | * to add ourselves to. Add us at the front of the list. | |
1917 | * Here we put the head pointer into our next pointer, | |
1918 | * and then we fall through to point the head at us. | |
1919 | */ | |
b12dd342 | 1920 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); |
1da177e4 LT |
1921 | if (error) { |
1922 | return error; | |
1923 | } | |
1924 | ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO); | |
1925 | ASSERT(dip->di_next_unlinked); | |
1926 | /* both on-disk, don't endian flip twice */ | |
1927 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; | |
1928 | offset = ip->i_boffset + | |
1929 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1930 | xfs_trans_inode_buf(tp, ibp); | |
1931 | xfs_trans_log_buf(tp, ibp, offset, | |
1932 | (offset + sizeof(xfs_agino_t) - 1)); | |
1933 | xfs_inobp_check(mp, ibp); | |
1934 | } | |
1935 | ||
1936 | /* | |
1937 | * Point the bucket head pointer at the inode being inserted. | |
1938 | */ | |
1939 | ASSERT(agino != 0); | |
16259e7d | 1940 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
1941 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1942 | (sizeof(xfs_agino_t) * bucket_index); | |
1943 | xfs_trans_log_buf(tp, agibp, offset, | |
1944 | (offset + sizeof(xfs_agino_t) - 1)); | |
1945 | return 0; | |
1946 | } | |
1947 | ||
1948 | /* | |
1949 | * Pull the on-disk inode from the AGI unlinked list. | |
1950 | */ | |
1951 | STATIC int | |
1952 | xfs_iunlink_remove( | |
1953 | xfs_trans_t *tp, | |
1954 | xfs_inode_t *ip) | |
1955 | { | |
1956 | xfs_ino_t next_ino; | |
1957 | xfs_mount_t *mp; | |
1958 | xfs_agi_t *agi; | |
1959 | xfs_dinode_t *dip; | |
1960 | xfs_buf_t *agibp; | |
1961 | xfs_buf_t *ibp; | |
1962 | xfs_agnumber_t agno; | |
1963 | xfs_daddr_t agdaddr; | |
1964 | xfs_agino_t agino; | |
1965 | xfs_agino_t next_agino; | |
1966 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 1967 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 1968 | short bucket_index; |
6fdf8ccc | 1969 | int offset, last_offset = 0; |
1da177e4 LT |
1970 | int error; |
1971 | int agi_ok; | |
1972 | ||
1973 | /* | |
1974 | * First pull the on-disk inode from the AGI unlinked list. | |
1975 | */ | |
1976 | mp = tp->t_mountp; | |
1977 | ||
1978 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); | |
1979 | agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
1980 | ||
1981 | /* | |
1982 | * Get the agi buffer first. It ensures lock ordering | |
1983 | * on the list. | |
1984 | */ | |
1985 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, | |
1986 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
1987 | if (error) { | |
1988 | cmn_err(CE_WARN, | |
1989 | "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.", | |
1990 | error, mp->m_fsname); | |
1991 | return error; | |
1992 | } | |
1993 | /* | |
1994 | * Validate the magic number of the agi block. | |
1995 | */ | |
1996 | agi = XFS_BUF_TO_AGI(agibp); | |
1997 | agi_ok = | |
16259e7d CH |
1998 | be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC && |
1999 | XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)); | |
1da177e4 LT |
2000 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE, |
2001 | XFS_RANDOM_IUNLINK_REMOVE))) { | |
2002 | XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW, | |
2003 | mp, agi); | |
2004 | xfs_trans_brelse(tp, agibp); | |
2005 | cmn_err(CE_WARN, | |
2006 | "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.", | |
2007 | mp->m_fsname); | |
2008 | return XFS_ERROR(EFSCORRUPTED); | |
2009 | } | |
2010 | /* | |
2011 | * Get the index into the agi hash table for the | |
2012 | * list this inode will go on. | |
2013 | */ | |
2014 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
2015 | ASSERT(agino != 0); | |
2016 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
16259e7d | 2017 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO); |
1da177e4 LT |
2018 | ASSERT(agi->agi_unlinked[bucket_index]); |
2019 | ||
16259e7d | 2020 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 LT |
2021 | /* |
2022 | * We're at the head of the list. Get the inode's | |
2023 | * on-disk buffer to see if there is anyone after us | |
2024 | * on the list. Only modify our next pointer if it | |
2025 | * is not already NULLAGINO. This saves us the overhead | |
2026 | * of dealing with the buffer when there is no need to | |
2027 | * change it. | |
2028 | */ | |
b12dd342 | 2029 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); |
1da177e4 LT |
2030 | if (error) { |
2031 | cmn_err(CE_WARN, | |
2032 | "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", | |
2033 | error, mp->m_fsname); | |
2034 | return error; | |
2035 | } | |
2036 | next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT); | |
2037 | ASSERT(next_agino != 0); | |
2038 | if (next_agino != NULLAGINO) { | |
2039 | INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO); | |
2040 | offset = ip->i_boffset + | |
2041 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2042 | xfs_trans_inode_buf(tp, ibp); | |
2043 | xfs_trans_log_buf(tp, ibp, offset, | |
2044 | (offset + sizeof(xfs_agino_t) - 1)); | |
2045 | xfs_inobp_check(mp, ibp); | |
2046 | } else { | |
2047 | xfs_trans_brelse(tp, ibp); | |
2048 | } | |
2049 | /* | |
2050 | * Point the bucket head pointer at the next inode. | |
2051 | */ | |
2052 | ASSERT(next_agino != 0); | |
2053 | ASSERT(next_agino != agino); | |
16259e7d | 2054 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2055 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2056 | (sizeof(xfs_agino_t) * bucket_index); | |
2057 | xfs_trans_log_buf(tp, agibp, offset, | |
2058 | (offset + sizeof(xfs_agino_t) - 1)); | |
2059 | } else { | |
2060 | /* | |
2061 | * We need to search the list for the inode being freed. | |
2062 | */ | |
16259e7d | 2063 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2064 | last_ibp = NULL; |
2065 | while (next_agino != agino) { | |
2066 | /* | |
2067 | * If the last inode wasn't the one pointing to | |
2068 | * us, then release its buffer since we're not | |
2069 | * going to do anything with it. | |
2070 | */ | |
2071 | if (last_ibp != NULL) { | |
2072 | xfs_trans_brelse(tp, last_ibp); | |
2073 | } | |
2074 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); | |
2075 | error = xfs_inotobp(mp, tp, next_ino, &last_dip, | |
2076 | &last_ibp, &last_offset); | |
2077 | if (error) { | |
2078 | cmn_err(CE_WARN, | |
2079 | "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.", | |
2080 | error, mp->m_fsname); | |
2081 | return error; | |
2082 | } | |
2083 | next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT); | |
2084 | ASSERT(next_agino != NULLAGINO); | |
2085 | ASSERT(next_agino != 0); | |
2086 | } | |
2087 | /* | |
2088 | * Now last_ibp points to the buffer previous to us on | |
2089 | * the unlinked list. Pull us from the list. | |
2090 | */ | |
b12dd342 | 2091 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); |
1da177e4 LT |
2092 | if (error) { |
2093 | cmn_err(CE_WARN, | |
2094 | "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", | |
2095 | error, mp->m_fsname); | |
2096 | return error; | |
2097 | } | |
2098 | next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT); | |
2099 | ASSERT(next_agino != 0); | |
2100 | ASSERT(next_agino != agino); | |
2101 | if (next_agino != NULLAGINO) { | |
2102 | INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO); | |
2103 | offset = ip->i_boffset + | |
2104 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2105 | xfs_trans_inode_buf(tp, ibp); | |
2106 | xfs_trans_log_buf(tp, ibp, offset, | |
2107 | (offset + sizeof(xfs_agino_t) - 1)); | |
2108 | xfs_inobp_check(mp, ibp); | |
2109 | } else { | |
2110 | xfs_trans_brelse(tp, ibp); | |
2111 | } | |
2112 | /* | |
2113 | * Point the previous inode on the list to the next inode. | |
2114 | */ | |
2115 | INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino); | |
2116 | ASSERT(next_agino != 0); | |
2117 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
2118 | xfs_trans_inode_buf(tp, last_ibp); | |
2119 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2120 | (offset + sizeof(xfs_agino_t) - 1)); | |
2121 | xfs_inobp_check(mp, last_ibp); | |
2122 | } | |
2123 | return 0; | |
2124 | } | |
2125 | ||
2126 | static __inline__ int xfs_inode_clean(xfs_inode_t *ip) | |
2127 | { | |
2128 | return (((ip->i_itemp == NULL) || | |
2129 | !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) && | |
2130 | (ip->i_update_core == 0)); | |
2131 | } | |
2132 | ||
ba0f32d4 | 2133 | STATIC void |
1da177e4 LT |
2134 | xfs_ifree_cluster( |
2135 | xfs_inode_t *free_ip, | |
2136 | xfs_trans_t *tp, | |
2137 | xfs_ino_t inum) | |
2138 | { | |
2139 | xfs_mount_t *mp = free_ip->i_mount; | |
2140 | int blks_per_cluster; | |
2141 | int nbufs; | |
2142 | int ninodes; | |
2143 | int i, j, found, pre_flushed; | |
2144 | xfs_daddr_t blkno; | |
2145 | xfs_buf_t *bp; | |
2146 | xfs_ihash_t *ih; | |
2147 | xfs_inode_t *ip, **ip_found; | |
2148 | xfs_inode_log_item_t *iip; | |
2149 | xfs_log_item_t *lip; | |
2150 | SPLDECL(s); | |
2151 | ||
2152 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { | |
2153 | blks_per_cluster = 1; | |
2154 | ninodes = mp->m_sb.sb_inopblock; | |
2155 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
2156 | } else { | |
2157 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
2158 | mp->m_sb.sb_blocksize; | |
2159 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
2160 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
2161 | } | |
2162 | ||
2163 | ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS); | |
2164 | ||
2165 | for (j = 0; j < nbufs; j++, inum += ninodes) { | |
2166 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
2167 | XFS_INO_TO_AGBNO(mp, inum)); | |
2168 | ||
2169 | ||
2170 | /* | |
2171 | * Look for each inode in memory and attempt to lock it, | |
2172 | * we can be racing with flush and tail pushing here. | |
2173 | * any inode we get the locks on, add to an array of | |
2174 | * inode items to process later. | |
2175 | * | |
2176 | * The get the buffer lock, we could beat a flush | |
2177 | * or tail pushing thread to the lock here, in which | |
2178 | * case they will go looking for the inode buffer | |
2179 | * and fail, we need some other form of interlock | |
2180 | * here. | |
2181 | */ | |
2182 | found = 0; | |
2183 | for (i = 0; i < ninodes; i++) { | |
2184 | ih = XFS_IHASH(mp, inum + i); | |
2185 | read_lock(&ih->ih_lock); | |
2186 | for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) { | |
2187 | if (ip->i_ino == inum + i) | |
2188 | break; | |
2189 | } | |
2190 | ||
2191 | /* Inode not in memory or we found it already, | |
2192 | * nothing to do | |
2193 | */ | |
2194 | if (!ip || (ip->i_flags & XFS_ISTALE)) { | |
2195 | read_unlock(&ih->ih_lock); | |
2196 | continue; | |
2197 | } | |
2198 | ||
2199 | if (xfs_inode_clean(ip)) { | |
2200 | read_unlock(&ih->ih_lock); | |
2201 | continue; | |
2202 | } | |
2203 | ||
2204 | /* If we can get the locks then add it to the | |
2205 | * list, otherwise by the time we get the bp lock | |
2206 | * below it will already be attached to the | |
2207 | * inode buffer. | |
2208 | */ | |
2209 | ||
2210 | /* This inode will already be locked - by us, lets | |
2211 | * keep it that way. | |
2212 | */ | |
2213 | ||
2214 | if (ip == free_ip) { | |
2215 | if (xfs_iflock_nowait(ip)) { | |
2216 | ip->i_flags |= XFS_ISTALE; | |
2217 | ||
2218 | if (xfs_inode_clean(ip)) { | |
2219 | xfs_ifunlock(ip); | |
2220 | } else { | |
2221 | ip_found[found++] = ip; | |
2222 | } | |
2223 | } | |
2224 | read_unlock(&ih->ih_lock); | |
2225 | continue; | |
2226 | } | |
2227 | ||
2228 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2229 | if (xfs_iflock_nowait(ip)) { | |
2230 | ip->i_flags |= XFS_ISTALE; | |
2231 | ||
2232 | if (xfs_inode_clean(ip)) { | |
2233 | xfs_ifunlock(ip); | |
2234 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2235 | } else { | |
2236 | ip_found[found++] = ip; | |
2237 | } | |
2238 | } else { | |
2239 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2240 | } | |
2241 | } | |
2242 | ||
2243 | read_unlock(&ih->ih_lock); | |
2244 | } | |
2245 | ||
2246 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
2247 | mp->m_bsize * blks_per_cluster, | |
2248 | XFS_BUF_LOCK); | |
2249 | ||
2250 | pre_flushed = 0; | |
2251 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
2252 | while (lip) { | |
2253 | if (lip->li_type == XFS_LI_INODE) { | |
2254 | iip = (xfs_inode_log_item_t *)lip; | |
2255 | ASSERT(iip->ili_logged == 1); | |
2256 | lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done; | |
2257 | AIL_LOCK(mp,s); | |
2258 | iip->ili_flush_lsn = iip->ili_item.li_lsn; | |
2259 | AIL_UNLOCK(mp, s); | |
2260 | iip->ili_inode->i_flags |= XFS_ISTALE; | |
2261 | pre_flushed++; | |
2262 | } | |
2263 | lip = lip->li_bio_list; | |
2264 | } | |
2265 | ||
2266 | for (i = 0; i < found; i++) { | |
2267 | ip = ip_found[i]; | |
2268 | iip = ip->i_itemp; | |
2269 | ||
2270 | if (!iip) { | |
2271 | ip->i_update_core = 0; | |
2272 | xfs_ifunlock(ip); | |
2273 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2274 | continue; | |
2275 | } | |
2276 | ||
2277 | iip->ili_last_fields = iip->ili_format.ilf_fields; | |
2278 | iip->ili_format.ilf_fields = 0; | |
2279 | iip->ili_logged = 1; | |
2280 | AIL_LOCK(mp,s); | |
2281 | iip->ili_flush_lsn = iip->ili_item.li_lsn; | |
2282 | AIL_UNLOCK(mp, s); | |
2283 | ||
2284 | xfs_buf_attach_iodone(bp, | |
2285 | (void(*)(xfs_buf_t*,xfs_log_item_t*)) | |
2286 | xfs_istale_done, (xfs_log_item_t *)iip); | |
2287 | if (ip != free_ip) { | |
2288 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2289 | } | |
2290 | } | |
2291 | ||
2292 | if (found || pre_flushed) | |
2293 | xfs_trans_stale_inode_buf(tp, bp); | |
2294 | xfs_trans_binval(tp, bp); | |
2295 | } | |
2296 | ||
2297 | kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *)); | |
2298 | } | |
2299 | ||
2300 | /* | |
2301 | * This is called to return an inode to the inode free list. | |
2302 | * The inode should already be truncated to 0 length and have | |
2303 | * no pages associated with it. This routine also assumes that | |
2304 | * the inode is already a part of the transaction. | |
2305 | * | |
2306 | * The on-disk copy of the inode will have been added to the list | |
2307 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2308 | * that list atomically with respect to freeing it here. | |
2309 | */ | |
2310 | int | |
2311 | xfs_ifree( | |
2312 | xfs_trans_t *tp, | |
2313 | xfs_inode_t *ip, | |
2314 | xfs_bmap_free_t *flist) | |
2315 | { | |
2316 | int error; | |
2317 | int delete; | |
2318 | xfs_ino_t first_ino; | |
2319 | ||
2320 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); | |
2321 | ASSERT(ip->i_transp == tp); | |
2322 | ASSERT(ip->i_d.di_nlink == 0); | |
2323 | ASSERT(ip->i_d.di_nextents == 0); | |
2324 | ASSERT(ip->i_d.di_anextents == 0); | |
2325 | ASSERT((ip->i_d.di_size == 0) || | |
2326 | ((ip->i_d.di_mode & S_IFMT) != S_IFREG)); | |
2327 | ASSERT(ip->i_d.di_nblocks == 0); | |
2328 | ||
2329 | /* | |
2330 | * Pull the on-disk inode from the AGI unlinked list. | |
2331 | */ | |
2332 | error = xfs_iunlink_remove(tp, ip); | |
2333 | if (error != 0) { | |
2334 | return error; | |
2335 | } | |
2336 | ||
2337 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
2338 | if (error != 0) { | |
2339 | return error; | |
2340 | } | |
2341 | ip->i_d.di_mode = 0; /* mark incore inode as free */ | |
2342 | ip->i_d.di_flags = 0; | |
2343 | ip->i_d.di_dmevmask = 0; | |
2344 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
2345 | ip->i_df.if_ext_max = | |
2346 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
2347 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
2348 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2349 | /* | |
2350 | * Bump the generation count so no one will be confused | |
2351 | * by reincarnations of this inode. | |
2352 | */ | |
2353 | ip->i_d.di_gen++; | |
2354 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2355 | ||
2356 | if (delete) { | |
2357 | xfs_ifree_cluster(ip, tp, first_ino); | |
2358 | } | |
2359 | ||
2360 | return 0; | |
2361 | } | |
2362 | ||
2363 | /* | |
2364 | * Reallocate the space for if_broot based on the number of records | |
2365 | * being added or deleted as indicated in rec_diff. Move the records | |
2366 | * and pointers in if_broot to fit the new size. When shrinking this | |
2367 | * will eliminate holes between the records and pointers created by | |
2368 | * the caller. When growing this will create holes to be filled in | |
2369 | * by the caller. | |
2370 | * | |
2371 | * The caller must not request to add more records than would fit in | |
2372 | * the on-disk inode root. If the if_broot is currently NULL, then | |
2373 | * if we adding records one will be allocated. The caller must also | |
2374 | * not request that the number of records go below zero, although | |
2375 | * it can go to zero. | |
2376 | * | |
2377 | * ip -- the inode whose if_broot area is changing | |
2378 | * ext_diff -- the change in the number of records, positive or negative, | |
2379 | * requested for the if_broot array. | |
2380 | */ | |
2381 | void | |
2382 | xfs_iroot_realloc( | |
2383 | xfs_inode_t *ip, | |
2384 | int rec_diff, | |
2385 | int whichfork) | |
2386 | { | |
2387 | int cur_max; | |
2388 | xfs_ifork_t *ifp; | |
2389 | xfs_bmbt_block_t *new_broot; | |
2390 | int new_max; | |
2391 | size_t new_size; | |
2392 | char *np; | |
2393 | char *op; | |
2394 | ||
2395 | /* | |
2396 | * Handle the degenerate case quietly. | |
2397 | */ | |
2398 | if (rec_diff == 0) { | |
2399 | return; | |
2400 | } | |
2401 | ||
2402 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2403 | if (rec_diff > 0) { | |
2404 | /* | |
2405 | * If there wasn't any memory allocated before, just | |
2406 | * allocate it now and get out. | |
2407 | */ | |
2408 | if (ifp->if_broot_bytes == 0) { | |
2409 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); | |
2410 | ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size, | |
2411 | KM_SLEEP); | |
2412 | ifp->if_broot_bytes = (int)new_size; | |
2413 | return; | |
2414 | } | |
2415 | ||
2416 | /* | |
2417 | * If there is already an existing if_broot, then we need | |
2418 | * to realloc() it and shift the pointers to their new | |
2419 | * location. The records don't change location because | |
2420 | * they are kept butted up against the btree block header. | |
2421 | */ | |
2422 | cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes); | |
2423 | new_max = cur_max + rec_diff; | |
2424 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2425 | ifp->if_broot = (xfs_bmbt_block_t *) | |
2426 | kmem_realloc(ifp->if_broot, | |
2427 | new_size, | |
2428 | (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ | |
2429 | KM_SLEEP); | |
2430 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2431 | ifp->if_broot_bytes); | |
2432 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2433 | (int)new_size); | |
2434 | ifp->if_broot_bytes = (int)new_size; | |
2435 | ASSERT(ifp->if_broot_bytes <= | |
2436 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2437 | memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); | |
2438 | return; | |
2439 | } | |
2440 | ||
2441 | /* | |
2442 | * rec_diff is less than 0. In this case, we are shrinking the | |
2443 | * if_broot buffer. It must already exist. If we go to zero | |
2444 | * records, just get rid of the root and clear the status bit. | |
2445 | */ | |
2446 | ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); | |
2447 | cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes); | |
2448 | new_max = cur_max + rec_diff; | |
2449 | ASSERT(new_max >= 0); | |
2450 | if (new_max > 0) | |
2451 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2452 | else | |
2453 | new_size = 0; | |
2454 | if (new_size > 0) { | |
2455 | new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP); | |
2456 | /* | |
2457 | * First copy over the btree block header. | |
2458 | */ | |
2459 | memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t)); | |
2460 | } else { | |
2461 | new_broot = NULL; | |
2462 | ifp->if_flags &= ~XFS_IFBROOT; | |
2463 | } | |
2464 | ||
2465 | /* | |
2466 | * Only copy the records and pointers if there are any. | |
2467 | */ | |
2468 | if (new_max > 0) { | |
2469 | /* | |
2470 | * First copy the records. | |
2471 | */ | |
2472 | op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1, | |
2473 | ifp->if_broot_bytes); | |
2474 | np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1, | |
2475 | (int)new_size); | |
2476 | memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); | |
2477 | ||
2478 | /* | |
2479 | * Then copy the pointers. | |
2480 | */ | |
2481 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2482 | ifp->if_broot_bytes); | |
2483 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1, | |
2484 | (int)new_size); | |
2485 | memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); | |
2486 | } | |
2487 | kmem_free(ifp->if_broot, ifp->if_broot_bytes); | |
2488 | ifp->if_broot = new_broot; | |
2489 | ifp->if_broot_bytes = (int)new_size; | |
2490 | ASSERT(ifp->if_broot_bytes <= | |
2491 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2492 | return; | |
2493 | } | |
2494 | ||
2495 | ||
1da177e4 LT |
2496 | /* |
2497 | * This is called when the amount of space needed for if_data | |
2498 | * is increased or decreased. The change in size is indicated by | |
2499 | * the number of bytes that need to be added or deleted in the | |
2500 | * byte_diff parameter. | |
2501 | * | |
2502 | * If the amount of space needed has decreased below the size of the | |
2503 | * inline buffer, then switch to using the inline buffer. Otherwise, | |
2504 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer | |
2505 | * to what is needed. | |
2506 | * | |
2507 | * ip -- the inode whose if_data area is changing | |
2508 | * byte_diff -- the change in the number of bytes, positive or negative, | |
2509 | * requested for the if_data array. | |
2510 | */ | |
2511 | void | |
2512 | xfs_idata_realloc( | |
2513 | xfs_inode_t *ip, | |
2514 | int byte_diff, | |
2515 | int whichfork) | |
2516 | { | |
2517 | xfs_ifork_t *ifp; | |
2518 | int new_size; | |
2519 | int real_size; | |
2520 | ||
2521 | if (byte_diff == 0) { | |
2522 | return; | |
2523 | } | |
2524 | ||
2525 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2526 | new_size = (int)ifp->if_bytes + byte_diff; | |
2527 | ASSERT(new_size >= 0); | |
2528 | ||
2529 | if (new_size == 0) { | |
2530 | if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2531 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2532 | } | |
2533 | ifp->if_u1.if_data = NULL; | |
2534 | real_size = 0; | |
2535 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { | |
2536 | /* | |
2537 | * If the valid extents/data can fit in if_inline_ext/data, | |
2538 | * copy them from the malloc'd vector and free it. | |
2539 | */ | |
2540 | if (ifp->if_u1.if_data == NULL) { | |
2541 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2542 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2543 | ASSERT(ifp->if_real_bytes != 0); | |
2544 | memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, | |
2545 | new_size); | |
2546 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2547 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2548 | } | |
2549 | real_size = 0; | |
2550 | } else { | |
2551 | /* | |
2552 | * Stuck with malloc/realloc. | |
2553 | * For inline data, the underlying buffer must be | |
2554 | * a multiple of 4 bytes in size so that it can be | |
2555 | * logged and stay on word boundaries. We enforce | |
2556 | * that here. | |
2557 | */ | |
2558 | real_size = roundup(new_size, 4); | |
2559 | if (ifp->if_u1.if_data == NULL) { | |
2560 | ASSERT(ifp->if_real_bytes == 0); | |
2561 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
2562 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2563 | /* | |
2564 | * Only do the realloc if the underlying size | |
2565 | * is really changing. | |
2566 | */ | |
2567 | if (ifp->if_real_bytes != real_size) { | |
2568 | ifp->if_u1.if_data = | |
2569 | kmem_realloc(ifp->if_u1.if_data, | |
2570 | real_size, | |
2571 | ifp->if_real_bytes, | |
2572 | KM_SLEEP); | |
2573 | } | |
2574 | } else { | |
2575 | ASSERT(ifp->if_real_bytes == 0); | |
2576 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
2577 | memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, | |
2578 | ifp->if_bytes); | |
2579 | } | |
2580 | } | |
2581 | ifp->if_real_bytes = real_size; | |
2582 | ifp->if_bytes = new_size; | |
2583 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2584 | } | |
2585 | ||
2586 | ||
2587 | ||
2588 | ||
2589 | /* | |
2590 | * Map inode to disk block and offset. | |
2591 | * | |
2592 | * mp -- the mount point structure for the current file system | |
2593 | * tp -- the current transaction | |
2594 | * ino -- the inode number of the inode to be located | |
2595 | * imap -- this structure is filled in with the information necessary | |
2596 | * to retrieve the given inode from disk | |
2597 | * flags -- flags to pass to xfs_dilocate indicating whether or not | |
2598 | * lookups in the inode btree were OK or not | |
2599 | */ | |
2600 | int | |
2601 | xfs_imap( | |
2602 | xfs_mount_t *mp, | |
2603 | xfs_trans_t *tp, | |
2604 | xfs_ino_t ino, | |
2605 | xfs_imap_t *imap, | |
2606 | uint flags) | |
2607 | { | |
2608 | xfs_fsblock_t fsbno; | |
2609 | int len; | |
2610 | int off; | |
2611 | int error; | |
2612 | ||
2613 | fsbno = imap->im_blkno ? | |
2614 | XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK; | |
2615 | error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags); | |
2616 | if (error != 0) { | |
2617 | return error; | |
2618 | } | |
2619 | imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno); | |
2620 | imap->im_len = XFS_FSB_TO_BB(mp, len); | |
2621 | imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno); | |
2622 | imap->im_ioffset = (ushort)off; | |
2623 | imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog); | |
2624 | return 0; | |
2625 | } | |
2626 | ||
2627 | void | |
2628 | xfs_idestroy_fork( | |
2629 | xfs_inode_t *ip, | |
2630 | int whichfork) | |
2631 | { | |
2632 | xfs_ifork_t *ifp; | |
2633 | ||
2634 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2635 | if (ifp->if_broot != NULL) { | |
2636 | kmem_free(ifp->if_broot, ifp->if_broot_bytes); | |
2637 | ifp->if_broot = NULL; | |
2638 | } | |
2639 | ||
2640 | /* | |
2641 | * If the format is local, then we can't have an extents | |
2642 | * array so just look for an inline data array. If we're | |
2643 | * not local then we may or may not have an extents list, | |
2644 | * so check and free it up if we do. | |
2645 | */ | |
2646 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
2647 | if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && | |
2648 | (ifp->if_u1.if_data != NULL)) { | |
2649 | ASSERT(ifp->if_real_bytes != 0); | |
2650 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2651 | ifp->if_u1.if_data = NULL; | |
2652 | ifp->if_real_bytes = 0; | |
2653 | } | |
2654 | } else if ((ifp->if_flags & XFS_IFEXTENTS) && | |
0293ce3a MK |
2655 | ((ifp->if_flags & XFS_IFEXTIREC) || |
2656 | ((ifp->if_u1.if_extents != NULL) && | |
2657 | (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { | |
1da177e4 | 2658 | ASSERT(ifp->if_real_bytes != 0); |
4eea22f0 | 2659 | xfs_iext_destroy(ifp); |
1da177e4 LT |
2660 | } |
2661 | ASSERT(ifp->if_u1.if_extents == NULL || | |
2662 | ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); | |
2663 | ASSERT(ifp->if_real_bytes == 0); | |
2664 | if (whichfork == XFS_ATTR_FORK) { | |
2665 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
2666 | ip->i_afp = NULL; | |
2667 | } | |
2668 | } | |
2669 | ||
2670 | /* | |
2671 | * This is called free all the memory associated with an inode. | |
2672 | * It must free the inode itself and any buffers allocated for | |
2673 | * if_extents/if_data and if_broot. It must also free the lock | |
2674 | * associated with the inode. | |
2675 | */ | |
2676 | void | |
2677 | xfs_idestroy( | |
2678 | xfs_inode_t *ip) | |
2679 | { | |
2680 | ||
2681 | switch (ip->i_d.di_mode & S_IFMT) { | |
2682 | case S_IFREG: | |
2683 | case S_IFDIR: | |
2684 | case S_IFLNK: | |
2685 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
2686 | break; | |
2687 | } | |
2688 | if (ip->i_afp) | |
2689 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
2690 | mrfree(&ip->i_lock); | |
2691 | mrfree(&ip->i_iolock); | |
2692 | freesema(&ip->i_flock); | |
2693 | #ifdef XFS_BMAP_TRACE | |
2694 | ktrace_free(ip->i_xtrace); | |
2695 | #endif | |
2696 | #ifdef XFS_BMBT_TRACE | |
2697 | ktrace_free(ip->i_btrace); | |
2698 | #endif | |
2699 | #ifdef XFS_RW_TRACE | |
2700 | ktrace_free(ip->i_rwtrace); | |
2701 | #endif | |
2702 | #ifdef XFS_ILOCK_TRACE | |
2703 | ktrace_free(ip->i_lock_trace); | |
2704 | #endif | |
2705 | #ifdef XFS_DIR2_TRACE | |
2706 | ktrace_free(ip->i_dir_trace); | |
2707 | #endif | |
2708 | if (ip->i_itemp) { | |
2709 | /* XXXdpd should be able to assert this but shutdown | |
2710 | * is leaving the AIL behind. */ | |
2711 | ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) || | |
2712 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
2713 | xfs_inode_item_destroy(ip); | |
2714 | } | |
2715 | kmem_zone_free(xfs_inode_zone, ip); | |
2716 | } | |
2717 | ||
2718 | ||
2719 | /* | |
2720 | * Increment the pin count of the given buffer. | |
2721 | * This value is protected by ipinlock spinlock in the mount structure. | |
2722 | */ | |
2723 | void | |
2724 | xfs_ipin( | |
2725 | xfs_inode_t *ip) | |
2726 | { | |
2727 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); | |
2728 | ||
2729 | atomic_inc(&ip->i_pincount); | |
2730 | } | |
2731 | ||
2732 | /* | |
2733 | * Decrement the pin count of the given inode, and wake up | |
2734 | * anyone in xfs_iwait_unpin() if the count goes to 0. The | |
c41564b5 | 2735 | * inode must have been previously pinned with a call to xfs_ipin(). |
1da177e4 LT |
2736 | */ |
2737 | void | |
2738 | xfs_iunpin( | |
2739 | xfs_inode_t *ip) | |
2740 | { | |
2741 | ASSERT(atomic_read(&ip->i_pincount) > 0); | |
2742 | ||
2743 | if (atomic_dec_and_test(&ip->i_pincount)) { | |
58829e49 DC |
2744 | /* |
2745 | * If the inode is currently being reclaimed, the | |
2746 | * linux inode _and_ the xfs vnode may have been | |
2747 | * freed so we cannot reference either of them safely. | |
2748 | * Hence we should not try to do anything to them | |
2749 | * if the xfs inode is currently in the reclaim | |
2750 | * path. | |
2751 | * | |
2752 | * However, we still need to issue the unpin wakeup | |
2753 | * call as the inode reclaim may be blocked waiting for | |
2754 | * the inode to become unpinned. | |
2755 | */ | |
2756 | if (!(ip->i_flags & (XFS_IRECLAIM|XFS_IRECLAIMABLE))) { | |
67fcaa73 | 2757 | bhv_vnode_t *vp = XFS_ITOV_NULL(ip); |
1da177e4 | 2758 | |
58829e49 DC |
2759 | /* make sync come back and flush this inode */ |
2760 | if (vp) { | |
2761 | struct inode *inode = vn_to_inode(vp); | |
1da177e4 | 2762 | |
71425087 DC |
2763 | if (!(inode->i_state & |
2764 | (I_NEW|I_FREEING|I_CLEAR))) | |
58829e49 DC |
2765 | mark_inode_dirty_sync(inode); |
2766 | } | |
1da177e4 | 2767 | } |
1da177e4 LT |
2768 | wake_up(&ip->i_ipin_wait); |
2769 | } | |
2770 | } | |
2771 | ||
2772 | /* | |
2773 | * This is called to wait for the given inode to be unpinned. | |
2774 | * It will sleep until this happens. The caller must have the | |
2775 | * inode locked in at least shared mode so that the buffer cannot | |
2776 | * be subsequently pinned once someone is waiting for it to be | |
2777 | * unpinned. | |
2778 | */ | |
ba0f32d4 | 2779 | STATIC void |
1da177e4 LT |
2780 | xfs_iunpin_wait( |
2781 | xfs_inode_t *ip) | |
2782 | { | |
2783 | xfs_inode_log_item_t *iip; | |
2784 | xfs_lsn_t lsn; | |
2785 | ||
2786 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS)); | |
2787 | ||
2788 | if (atomic_read(&ip->i_pincount) == 0) { | |
2789 | return; | |
2790 | } | |
2791 | ||
2792 | iip = ip->i_itemp; | |
2793 | if (iip && iip->ili_last_lsn) { | |
2794 | lsn = iip->ili_last_lsn; | |
2795 | } else { | |
2796 | lsn = (xfs_lsn_t)0; | |
2797 | } | |
2798 | ||
2799 | /* | |
2800 | * Give the log a push so we don't wait here too long. | |
2801 | */ | |
2802 | xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE); | |
2803 | ||
2804 | wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0)); | |
2805 | } | |
2806 | ||
2807 | ||
2808 | /* | |
2809 | * xfs_iextents_copy() | |
2810 | * | |
2811 | * This is called to copy the REAL extents (as opposed to the delayed | |
2812 | * allocation extents) from the inode into the given buffer. It | |
2813 | * returns the number of bytes copied into the buffer. | |
2814 | * | |
2815 | * If there are no delayed allocation extents, then we can just | |
2816 | * memcpy() the extents into the buffer. Otherwise, we need to | |
2817 | * examine each extent in turn and skip those which are delayed. | |
2818 | */ | |
2819 | int | |
2820 | xfs_iextents_copy( | |
2821 | xfs_inode_t *ip, | |
2822 | xfs_bmbt_rec_t *buffer, | |
2823 | int whichfork) | |
2824 | { | |
2825 | int copied; | |
2826 | xfs_bmbt_rec_t *dest_ep; | |
2827 | xfs_bmbt_rec_t *ep; | |
2828 | #ifdef XFS_BMAP_TRACE | |
2829 | static char fname[] = "xfs_iextents_copy"; | |
2830 | #endif | |
2831 | int i; | |
2832 | xfs_ifork_t *ifp; | |
2833 | int nrecs; | |
2834 | xfs_fsblock_t start_block; | |
2835 | ||
2836 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2837 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
2838 | ASSERT(ifp->if_bytes > 0); | |
2839 | ||
2840 | nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
2841 | xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork); | |
2842 | ASSERT(nrecs > 0); | |
2843 | ||
2844 | /* | |
2845 | * There are some delayed allocation extents in the | |
2846 | * inode, so copy the extents one at a time and skip | |
2847 | * the delayed ones. There must be at least one | |
2848 | * non-delayed extent. | |
2849 | */ | |
1da177e4 LT |
2850 | dest_ep = buffer; |
2851 | copied = 0; | |
2852 | for (i = 0; i < nrecs; i++) { | |
4eea22f0 | 2853 | ep = xfs_iext_get_ext(ifp, i); |
1da177e4 LT |
2854 | start_block = xfs_bmbt_get_startblock(ep); |
2855 | if (ISNULLSTARTBLOCK(start_block)) { | |
2856 | /* | |
2857 | * It's a delayed allocation extent, so skip it. | |
2858 | */ | |
1da177e4 LT |
2859 | continue; |
2860 | } | |
2861 | ||
2862 | /* Translate to on disk format */ | |
2863 | put_unaligned(INT_GET(ep->l0, ARCH_CONVERT), | |
2864 | (__uint64_t*)&dest_ep->l0); | |
2865 | put_unaligned(INT_GET(ep->l1, ARCH_CONVERT), | |
2866 | (__uint64_t*)&dest_ep->l1); | |
2867 | dest_ep++; | |
1da177e4 LT |
2868 | copied++; |
2869 | } | |
2870 | ASSERT(copied != 0); | |
4eea22f0 | 2871 | xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
2872 | |
2873 | return (copied * (uint)sizeof(xfs_bmbt_rec_t)); | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * Each of the following cases stores data into the same region | |
2878 | * of the on-disk inode, so only one of them can be valid at | |
2879 | * any given time. While it is possible to have conflicting formats | |
2880 | * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is | |
2881 | * in EXTENTS format, this can only happen when the fork has | |
2882 | * changed formats after being modified but before being flushed. | |
2883 | * In these cases, the format always takes precedence, because the | |
2884 | * format indicates the current state of the fork. | |
2885 | */ | |
2886 | /*ARGSUSED*/ | |
2887 | STATIC int | |
2888 | xfs_iflush_fork( | |
2889 | xfs_inode_t *ip, | |
2890 | xfs_dinode_t *dip, | |
2891 | xfs_inode_log_item_t *iip, | |
2892 | int whichfork, | |
2893 | xfs_buf_t *bp) | |
2894 | { | |
2895 | char *cp; | |
2896 | xfs_ifork_t *ifp; | |
2897 | xfs_mount_t *mp; | |
2898 | #ifdef XFS_TRANS_DEBUG | |
2899 | int first; | |
2900 | #endif | |
2901 | static const short brootflag[2] = | |
2902 | { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; | |
2903 | static const short dataflag[2] = | |
2904 | { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; | |
2905 | static const short extflag[2] = | |
2906 | { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; | |
2907 | ||
2908 | if (iip == NULL) | |
2909 | return 0; | |
2910 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2911 | /* | |
2912 | * This can happen if we gave up in iformat in an error path, | |
2913 | * for the attribute fork. | |
2914 | */ | |
2915 | if (ifp == NULL) { | |
2916 | ASSERT(whichfork == XFS_ATTR_FORK); | |
2917 | return 0; | |
2918 | } | |
2919 | cp = XFS_DFORK_PTR(dip, whichfork); | |
2920 | mp = ip->i_mount; | |
2921 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { | |
2922 | case XFS_DINODE_FMT_LOCAL: | |
2923 | if ((iip->ili_format.ilf_fields & dataflag[whichfork]) && | |
2924 | (ifp->if_bytes > 0)) { | |
2925 | ASSERT(ifp->if_u1.if_data != NULL); | |
2926 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2927 | memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); | |
2928 | } | |
1da177e4 LT |
2929 | break; |
2930 | ||
2931 | case XFS_DINODE_FMT_EXTENTS: | |
2932 | ASSERT((ifp->if_flags & XFS_IFEXTENTS) || | |
2933 | !(iip->ili_format.ilf_fields & extflag[whichfork])); | |
4eea22f0 MK |
2934 | ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) || |
2935 | (ifp->if_bytes == 0)); | |
2936 | ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) || | |
2937 | (ifp->if_bytes > 0)); | |
1da177e4 LT |
2938 | if ((iip->ili_format.ilf_fields & extflag[whichfork]) && |
2939 | (ifp->if_bytes > 0)) { | |
2940 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); | |
2941 | (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, | |
2942 | whichfork); | |
2943 | } | |
2944 | break; | |
2945 | ||
2946 | case XFS_DINODE_FMT_BTREE: | |
2947 | if ((iip->ili_format.ilf_fields & brootflag[whichfork]) && | |
2948 | (ifp->if_broot_bytes > 0)) { | |
2949 | ASSERT(ifp->if_broot != NULL); | |
2950 | ASSERT(ifp->if_broot_bytes <= | |
2951 | (XFS_IFORK_SIZE(ip, whichfork) + | |
2952 | XFS_BROOT_SIZE_ADJ)); | |
2953 | xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes, | |
2954 | (xfs_bmdr_block_t *)cp, | |
2955 | XFS_DFORK_SIZE(dip, mp, whichfork)); | |
2956 | } | |
2957 | break; | |
2958 | ||
2959 | case XFS_DINODE_FMT_DEV: | |
2960 | if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) { | |
2961 | ASSERT(whichfork == XFS_DATA_FORK); | |
2962 | INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev); | |
2963 | } | |
2964 | break; | |
2965 | ||
2966 | case XFS_DINODE_FMT_UUID: | |
2967 | if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) { | |
2968 | ASSERT(whichfork == XFS_DATA_FORK); | |
2969 | memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid, | |
2970 | sizeof(uuid_t)); | |
2971 | } | |
2972 | break; | |
2973 | ||
2974 | default: | |
2975 | ASSERT(0); | |
2976 | break; | |
2977 | } | |
2978 | ||
2979 | return 0; | |
2980 | } | |
2981 | ||
2982 | /* | |
2983 | * xfs_iflush() will write a modified inode's changes out to the | |
2984 | * inode's on disk home. The caller must have the inode lock held | |
2985 | * in at least shared mode and the inode flush semaphore must be | |
2986 | * held as well. The inode lock will still be held upon return from | |
2987 | * the call and the caller is free to unlock it. | |
2988 | * The inode flush lock will be unlocked when the inode reaches the disk. | |
2989 | * The flags indicate how the inode's buffer should be written out. | |
2990 | */ | |
2991 | int | |
2992 | xfs_iflush( | |
2993 | xfs_inode_t *ip, | |
2994 | uint flags) | |
2995 | { | |
2996 | xfs_inode_log_item_t *iip; | |
2997 | xfs_buf_t *bp; | |
2998 | xfs_dinode_t *dip; | |
2999 | xfs_mount_t *mp; | |
3000 | int error; | |
3001 | /* REFERENCED */ | |
3002 | xfs_chash_t *ch; | |
3003 | xfs_inode_t *iq; | |
3004 | int clcount; /* count of inodes clustered */ | |
3005 | int bufwasdelwri; | |
3006 | enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) }; | |
3007 | SPLDECL(s); | |
3008 | ||
3009 | XFS_STATS_INC(xs_iflush_count); | |
3010 | ||
3011 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
0d8fee32 | 3012 | ASSERT(issemalocked(&(ip->i_flock))); |
1da177e4 LT |
3013 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
3014 | ip->i_d.di_nextents > ip->i_df.if_ext_max); | |
3015 | ||
3016 | iip = ip->i_itemp; | |
3017 | mp = ip->i_mount; | |
3018 | ||
3019 | /* | |
3020 | * If the inode isn't dirty, then just release the inode | |
3021 | * flush lock and do nothing. | |
3022 | */ | |
3023 | if ((ip->i_update_core == 0) && | |
3024 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3025 | ASSERT((iip != NULL) ? | |
3026 | !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1); | |
3027 | xfs_ifunlock(ip); | |
3028 | return 0; | |
3029 | } | |
3030 | ||
3031 | /* | |
3032 | * We can't flush the inode until it is unpinned, so | |
3033 | * wait for it. We know noone new can pin it, because | |
3034 | * we are holding the inode lock shared and you need | |
3035 | * to hold it exclusively to pin the inode. | |
3036 | */ | |
3037 | xfs_iunpin_wait(ip); | |
3038 | ||
3039 | /* | |
3040 | * This may have been unpinned because the filesystem is shutting | |
3041 | * down forcibly. If that's the case we must not write this inode | |
3042 | * to disk, because the log record didn't make it to disk! | |
3043 | */ | |
3044 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
3045 | ip->i_update_core = 0; | |
3046 | if (iip) | |
3047 | iip->ili_format.ilf_fields = 0; | |
3048 | xfs_ifunlock(ip); | |
3049 | return XFS_ERROR(EIO); | |
3050 | } | |
3051 | ||
3052 | /* | |
3053 | * Get the buffer containing the on-disk inode. | |
3054 | */ | |
b12dd342 NS |
3055 | error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0); |
3056 | if (error) { | |
1da177e4 LT |
3057 | xfs_ifunlock(ip); |
3058 | return error; | |
3059 | } | |
3060 | ||
3061 | /* | |
3062 | * Decide how buffer will be flushed out. This is done before | |
3063 | * the call to xfs_iflush_int because this field is zeroed by it. | |
3064 | */ | |
3065 | if (iip != NULL && iip->ili_format.ilf_fields != 0) { | |
3066 | /* | |
3067 | * Flush out the inode buffer according to the directions | |
3068 | * of the caller. In the cases where the caller has given | |
3069 | * us a choice choose the non-delwri case. This is because | |
3070 | * the inode is in the AIL and we need to get it out soon. | |
3071 | */ | |
3072 | switch (flags) { | |
3073 | case XFS_IFLUSH_SYNC: | |
3074 | case XFS_IFLUSH_DELWRI_ELSE_SYNC: | |
3075 | flags = 0; | |
3076 | break; | |
3077 | case XFS_IFLUSH_ASYNC: | |
3078 | case XFS_IFLUSH_DELWRI_ELSE_ASYNC: | |
3079 | flags = INT_ASYNC; | |
3080 | break; | |
3081 | case XFS_IFLUSH_DELWRI: | |
3082 | flags = INT_DELWRI; | |
3083 | break; | |
3084 | default: | |
3085 | ASSERT(0); | |
3086 | flags = 0; | |
3087 | break; | |
3088 | } | |
3089 | } else { | |
3090 | switch (flags) { | |
3091 | case XFS_IFLUSH_DELWRI_ELSE_SYNC: | |
3092 | case XFS_IFLUSH_DELWRI_ELSE_ASYNC: | |
3093 | case XFS_IFLUSH_DELWRI: | |
3094 | flags = INT_DELWRI; | |
3095 | break; | |
3096 | case XFS_IFLUSH_ASYNC: | |
3097 | flags = INT_ASYNC; | |
3098 | break; | |
3099 | case XFS_IFLUSH_SYNC: | |
3100 | flags = 0; | |
3101 | break; | |
3102 | default: | |
3103 | ASSERT(0); | |
3104 | flags = 0; | |
3105 | break; | |
3106 | } | |
3107 | } | |
3108 | ||
3109 | /* | |
3110 | * First flush out the inode that xfs_iflush was called with. | |
3111 | */ | |
3112 | error = xfs_iflush_int(ip, bp); | |
3113 | if (error) { | |
3114 | goto corrupt_out; | |
3115 | } | |
3116 | ||
3117 | /* | |
3118 | * inode clustering: | |
3119 | * see if other inodes can be gathered into this write | |
3120 | */ | |
3121 | ||
3122 | ip->i_chash->chl_buf = bp; | |
3123 | ||
3124 | ch = XFS_CHASH(mp, ip->i_blkno); | |
3125 | s = mutex_spinlock(&ch->ch_lock); | |
3126 | ||
3127 | clcount = 0; | |
3128 | for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) { | |
3129 | /* | |
3130 | * Do an un-protected check to see if the inode is dirty and | |
3131 | * is a candidate for flushing. These checks will be repeated | |
3132 | * later after the appropriate locks are acquired. | |
3133 | */ | |
3134 | iip = iq->i_itemp; | |
3135 | if ((iq->i_update_core == 0) && | |
3136 | ((iip == NULL) || | |
3137 | !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) && | |
3138 | xfs_ipincount(iq) == 0) { | |
3139 | continue; | |
3140 | } | |
3141 | ||
3142 | /* | |
3143 | * Try to get locks. If any are unavailable, | |
3144 | * then this inode cannot be flushed and is skipped. | |
3145 | */ | |
3146 | ||
3147 | /* get inode locks (just i_lock) */ | |
3148 | if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) { | |
3149 | /* get inode flush lock */ | |
3150 | if (xfs_iflock_nowait(iq)) { | |
3151 | /* check if pinned */ | |
3152 | if (xfs_ipincount(iq) == 0) { | |
3153 | /* arriving here means that | |
3154 | * this inode can be flushed. | |
3155 | * first re-check that it's | |
3156 | * dirty | |
3157 | */ | |
3158 | iip = iq->i_itemp; | |
3159 | if ((iq->i_update_core != 0)|| | |
3160 | ((iip != NULL) && | |
3161 | (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3162 | clcount++; | |
3163 | error = xfs_iflush_int(iq, bp); | |
3164 | if (error) { | |
3165 | xfs_iunlock(iq, | |
3166 | XFS_ILOCK_SHARED); | |
3167 | goto cluster_corrupt_out; | |
3168 | } | |
3169 | } else { | |
3170 | xfs_ifunlock(iq); | |
3171 | } | |
3172 | } else { | |
3173 | xfs_ifunlock(iq); | |
3174 | } | |
3175 | } | |
3176 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3177 | } | |
3178 | } | |
3179 | mutex_spinunlock(&ch->ch_lock, s); | |
3180 | ||
3181 | if (clcount) { | |
3182 | XFS_STATS_INC(xs_icluster_flushcnt); | |
3183 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
3184 | } | |
3185 | ||
3186 | /* | |
3187 | * If the buffer is pinned then push on the log so we won't | |
3188 | * get stuck waiting in the write for too long. | |
3189 | */ | |
3190 | if (XFS_BUF_ISPINNED(bp)){ | |
3191 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
3192 | } | |
3193 | ||
3194 | if (flags & INT_DELWRI) { | |
3195 | xfs_bdwrite(mp, bp); | |
3196 | } else if (flags & INT_ASYNC) { | |
3197 | xfs_bawrite(mp, bp); | |
3198 | } else { | |
3199 | error = xfs_bwrite(mp, bp); | |
3200 | } | |
3201 | return error; | |
3202 | ||
3203 | corrupt_out: | |
3204 | xfs_buf_relse(bp); | |
7d04a335 | 3205 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 LT |
3206 | xfs_iflush_abort(ip); |
3207 | /* | |
3208 | * Unlocks the flush lock | |
3209 | */ | |
3210 | return XFS_ERROR(EFSCORRUPTED); | |
3211 | ||
3212 | cluster_corrupt_out: | |
3213 | /* Corruption detected in the clustering loop. Invalidate the | |
3214 | * inode buffer and shut down the filesystem. | |
3215 | */ | |
3216 | mutex_spinunlock(&ch->ch_lock, s); | |
3217 | ||
3218 | /* | |
3219 | * Clean up the buffer. If it was B_DELWRI, just release it -- | |
3220 | * brelse can handle it with no problems. If not, shut down the | |
3221 | * filesystem before releasing the buffer. | |
3222 | */ | |
3223 | if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) { | |
3224 | xfs_buf_relse(bp); | |
3225 | } | |
3226 | ||
7d04a335 | 3227 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 LT |
3228 | |
3229 | if(!bufwasdelwri) { | |
3230 | /* | |
3231 | * Just like incore_relse: if we have b_iodone functions, | |
3232 | * mark the buffer as an error and call them. Otherwise | |
3233 | * mark it as stale and brelse. | |
3234 | */ | |
3235 | if (XFS_BUF_IODONE_FUNC(bp)) { | |
3236 | XFS_BUF_CLR_BDSTRAT_FUNC(bp); | |
3237 | XFS_BUF_UNDONE(bp); | |
3238 | XFS_BUF_STALE(bp); | |
3239 | XFS_BUF_SHUT(bp); | |
3240 | XFS_BUF_ERROR(bp,EIO); | |
3241 | xfs_biodone(bp); | |
3242 | } else { | |
3243 | XFS_BUF_STALE(bp); | |
3244 | xfs_buf_relse(bp); | |
3245 | } | |
3246 | } | |
3247 | ||
3248 | xfs_iflush_abort(iq); | |
3249 | /* | |
3250 | * Unlocks the flush lock | |
3251 | */ | |
3252 | return XFS_ERROR(EFSCORRUPTED); | |
3253 | } | |
3254 | ||
3255 | ||
3256 | STATIC int | |
3257 | xfs_iflush_int( | |
3258 | xfs_inode_t *ip, | |
3259 | xfs_buf_t *bp) | |
3260 | { | |
3261 | xfs_inode_log_item_t *iip; | |
3262 | xfs_dinode_t *dip; | |
3263 | xfs_mount_t *mp; | |
3264 | #ifdef XFS_TRANS_DEBUG | |
3265 | int first; | |
3266 | #endif | |
3267 | SPLDECL(s); | |
3268 | ||
3269 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
0d8fee32 | 3270 | ASSERT(issemalocked(&(ip->i_flock))); |
1da177e4 LT |
3271 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
3272 | ip->i_d.di_nextents > ip->i_df.if_ext_max); | |
3273 | ||
3274 | iip = ip->i_itemp; | |
3275 | mp = ip->i_mount; | |
3276 | ||
3277 | ||
3278 | /* | |
3279 | * If the inode isn't dirty, then just release the inode | |
3280 | * flush lock and do nothing. | |
3281 | */ | |
3282 | if ((ip->i_update_core == 0) && | |
3283 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3284 | xfs_ifunlock(ip); | |
3285 | return 0; | |
3286 | } | |
3287 | ||
3288 | /* set *dip = inode's place in the buffer */ | |
3289 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset); | |
3290 | ||
3291 | /* | |
3292 | * Clear i_update_core before copying out the data. | |
3293 | * This is for coordination with our timestamp updates | |
3294 | * that don't hold the inode lock. They will always | |
3295 | * update the timestamps BEFORE setting i_update_core, | |
3296 | * so if we clear i_update_core after they set it we | |
3297 | * are guaranteed to see their updates to the timestamps. | |
3298 | * I believe that this depends on strongly ordered memory | |
3299 | * semantics, but we have that. We use the SYNCHRONIZE | |
3300 | * macro to make sure that the compiler does not reorder | |
3301 | * the i_update_core access below the data copy below. | |
3302 | */ | |
3303 | ip->i_update_core = 0; | |
3304 | SYNCHRONIZE(); | |
3305 | ||
42fe2b1f CH |
3306 | /* |
3307 | * Make sure to get the latest atime from the Linux inode. | |
3308 | */ | |
3309 | xfs_synchronize_atime(ip); | |
3310 | ||
1da177e4 LT |
3311 | if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC, |
3312 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { | |
3313 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3314 | "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
3315 | ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip); | |
3316 | goto corrupt_out; | |
3317 | } | |
3318 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
3319 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
3320 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3321 | "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
3322 | ip->i_ino, ip, ip->i_d.di_magic); | |
3323 | goto corrupt_out; | |
3324 | } | |
3325 | if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
3326 | if (XFS_TEST_ERROR( | |
3327 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3328 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
3329 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
3330 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3331 | "xfs_iflush: Bad regular inode %Lu, ptr 0x%p", | |
3332 | ip->i_ino, ip); | |
3333 | goto corrupt_out; | |
3334 | } | |
3335 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { | |
3336 | if (XFS_TEST_ERROR( | |
3337 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3338 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3339 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
3340 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
3341 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3342 | "xfs_iflush: Bad directory inode %Lu, ptr 0x%p", | |
3343 | ip->i_ino, ip); | |
3344 | goto corrupt_out; | |
3345 | } | |
3346 | } | |
3347 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
3348 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
3349 | XFS_RANDOM_IFLUSH_5)) { | |
3350 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3351 | "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p", | |
3352 | ip->i_ino, | |
3353 | ip->i_d.di_nextents + ip->i_d.di_anextents, | |
3354 | ip->i_d.di_nblocks, | |
3355 | ip); | |
3356 | goto corrupt_out; | |
3357 | } | |
3358 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
3359 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
3360 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3361 | "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
3362 | ip->i_ino, ip->i_d.di_forkoff, ip); | |
3363 | goto corrupt_out; | |
3364 | } | |
3365 | /* | |
3366 | * bump the flush iteration count, used to detect flushes which | |
3367 | * postdate a log record during recovery. | |
3368 | */ | |
3369 | ||
3370 | ip->i_d.di_flushiter++; | |
3371 | ||
3372 | /* | |
3373 | * Copy the dirty parts of the inode into the on-disk | |
3374 | * inode. We always copy out the core of the inode, | |
3375 | * because if the inode is dirty at all the core must | |
3376 | * be. | |
3377 | */ | |
3378 | xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1); | |
3379 | ||
3380 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3381 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3382 | ip->i_d.di_flushiter = 0; | |
3383 | ||
3384 | /* | |
3385 | * If this is really an old format inode and the superblock version | |
3386 | * has not been updated to support only new format inodes, then | |
3387 | * convert back to the old inode format. If the superblock version | |
3388 | * has been updated, then make the conversion permanent. | |
3389 | */ | |
3390 | ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 || | |
3391 | XFS_SB_VERSION_HASNLINK(&mp->m_sb)); | |
3392 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
3393 | if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) { | |
3394 | /* | |
3395 | * Convert it back. | |
3396 | */ | |
3397 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
3398 | INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink); | |
3399 | } else { | |
3400 | /* | |
3401 | * The superblock version has already been bumped, | |
3402 | * so just make the conversion to the new inode | |
3403 | * format permanent. | |
3404 | */ | |
3405 | ip->i_d.di_version = XFS_DINODE_VERSION_2; | |
3406 | INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2); | |
3407 | ip->i_d.di_onlink = 0; | |
3408 | dip->di_core.di_onlink = 0; | |
3409 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
3410 | memset(&(dip->di_core.di_pad[0]), 0, | |
3411 | sizeof(dip->di_core.di_pad)); | |
3412 | ASSERT(ip->i_d.di_projid == 0); | |
3413 | } | |
3414 | } | |
3415 | ||
3416 | if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) { | |
3417 | goto corrupt_out; | |
3418 | } | |
3419 | ||
3420 | if (XFS_IFORK_Q(ip)) { | |
3421 | /* | |
3422 | * The only error from xfs_iflush_fork is on the data fork. | |
3423 | */ | |
3424 | (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
3425 | } | |
3426 | xfs_inobp_check(mp, bp); | |
3427 | ||
3428 | /* | |
3429 | * We've recorded everything logged in the inode, so we'd | |
3430 | * like to clear the ilf_fields bits so we don't log and | |
3431 | * flush things unnecessarily. However, we can't stop | |
3432 | * logging all this information until the data we've copied | |
3433 | * into the disk buffer is written to disk. If we did we might | |
3434 | * overwrite the copy of the inode in the log with all the | |
3435 | * data after re-logging only part of it, and in the face of | |
3436 | * a crash we wouldn't have all the data we need to recover. | |
3437 | * | |
3438 | * What we do is move the bits to the ili_last_fields field. | |
3439 | * When logging the inode, these bits are moved back to the | |
3440 | * ilf_fields field. In the xfs_iflush_done() routine we | |
3441 | * clear ili_last_fields, since we know that the information | |
3442 | * those bits represent is permanently on disk. As long as | |
3443 | * the flush completes before the inode is logged again, then | |
3444 | * both ilf_fields and ili_last_fields will be cleared. | |
3445 | * | |
3446 | * We can play with the ilf_fields bits here, because the inode | |
3447 | * lock must be held exclusively in order to set bits there | |
3448 | * and the flush lock protects the ili_last_fields bits. | |
3449 | * Set ili_logged so the flush done | |
3450 | * routine can tell whether or not to look in the AIL. | |
3451 | * Also, store the current LSN of the inode so that we can tell | |
3452 | * whether the item has moved in the AIL from xfs_iflush_done(). | |
3453 | * In order to read the lsn we need the AIL lock, because | |
3454 | * it is a 64 bit value that cannot be read atomically. | |
3455 | */ | |
3456 | if (iip != NULL && iip->ili_format.ilf_fields != 0) { | |
3457 | iip->ili_last_fields = iip->ili_format.ilf_fields; | |
3458 | iip->ili_format.ilf_fields = 0; | |
3459 | iip->ili_logged = 1; | |
3460 | ||
3461 | ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */ | |
3462 | AIL_LOCK(mp,s); | |
3463 | iip->ili_flush_lsn = iip->ili_item.li_lsn; | |
3464 | AIL_UNLOCK(mp, s); | |
3465 | ||
3466 | /* | |
3467 | * Attach the function xfs_iflush_done to the inode's | |
3468 | * buffer. This will remove the inode from the AIL | |
3469 | * and unlock the inode's flush lock when the inode is | |
3470 | * completely written to disk. | |
3471 | */ | |
3472 | xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*)) | |
3473 | xfs_iflush_done, (xfs_log_item_t *)iip); | |
3474 | ||
3475 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
3476 | ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL); | |
3477 | } else { | |
3478 | /* | |
3479 | * We're flushing an inode which is not in the AIL and has | |
3480 | * not been logged but has i_update_core set. For this | |
3481 | * case we can use a B_DELWRI flush and immediately drop | |
3482 | * the inode flush lock because we can avoid the whole | |
3483 | * AIL state thing. It's OK to drop the flush lock now, | |
3484 | * because we've already locked the buffer and to do anything | |
3485 | * you really need both. | |
3486 | */ | |
3487 | if (iip != NULL) { | |
3488 | ASSERT(iip->ili_logged == 0); | |
3489 | ASSERT(iip->ili_last_fields == 0); | |
3490 | ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0); | |
3491 | } | |
3492 | xfs_ifunlock(ip); | |
3493 | } | |
3494 | ||
3495 | return 0; | |
3496 | ||
3497 | corrupt_out: | |
3498 | return XFS_ERROR(EFSCORRUPTED); | |
3499 | } | |
3500 | ||
3501 | ||
3502 | /* | |
efa80278 | 3503 | * Flush all inactive inodes in mp. |
1da177e4 | 3504 | */ |
efa80278 | 3505 | void |
1da177e4 | 3506 | xfs_iflush_all( |
efa80278 | 3507 | xfs_mount_t *mp) |
1da177e4 | 3508 | { |
1da177e4 | 3509 | xfs_inode_t *ip; |
67fcaa73 | 3510 | bhv_vnode_t *vp; |
1da177e4 | 3511 | |
efa80278 CH |
3512 | again: |
3513 | XFS_MOUNT_ILOCK(mp); | |
3514 | ip = mp->m_inodes; | |
3515 | if (ip == NULL) | |
3516 | goto out; | |
1da177e4 | 3517 | |
efa80278 CH |
3518 | do { |
3519 | /* Make sure we skip markers inserted by sync */ | |
3520 | if (ip->i_mount == NULL) { | |
3521 | ip = ip->i_mnext; | |
3522 | continue; | |
3523 | } | |
1da177e4 | 3524 | |
efa80278 CH |
3525 | vp = XFS_ITOV_NULL(ip); |
3526 | if (!vp) { | |
1da177e4 | 3527 | XFS_MOUNT_IUNLOCK(mp); |
efa80278 CH |
3528 | xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC); |
3529 | goto again; | |
3530 | } | |
1da177e4 | 3531 | |
efa80278 | 3532 | ASSERT(vn_count(vp) == 0); |
1da177e4 | 3533 | |
efa80278 CH |
3534 | ip = ip->i_mnext; |
3535 | } while (ip != mp->m_inodes); | |
3536 | out: | |
1da177e4 | 3537 | XFS_MOUNT_IUNLOCK(mp); |
1da177e4 LT |
3538 | } |
3539 | ||
1da177e4 LT |
3540 | /* |
3541 | * xfs_iaccess: check accessibility of inode for mode. | |
3542 | */ | |
3543 | int | |
3544 | xfs_iaccess( | |
3545 | xfs_inode_t *ip, | |
3546 | mode_t mode, | |
3547 | cred_t *cr) | |
3548 | { | |
3549 | int error; | |
3550 | mode_t orgmode = mode; | |
ec86dc02 | 3551 | struct inode *inode = vn_to_inode(XFS_ITOV(ip)); |
1da177e4 LT |
3552 | |
3553 | if (mode & S_IWUSR) { | |
3554 | umode_t imode = inode->i_mode; | |
3555 | ||
3556 | if (IS_RDONLY(inode) && | |
3557 | (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode))) | |
3558 | return XFS_ERROR(EROFS); | |
3559 | ||
3560 | if (IS_IMMUTABLE(inode)) | |
3561 | return XFS_ERROR(EACCES); | |
3562 | } | |
3563 | ||
3564 | /* | |
3565 | * If there's an Access Control List it's used instead of | |
3566 | * the mode bits. | |
3567 | */ | |
3568 | if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1) | |
3569 | return error ? XFS_ERROR(error) : 0; | |
3570 | ||
3571 | if (current_fsuid(cr) != ip->i_d.di_uid) { | |
3572 | mode >>= 3; | |
3573 | if (!in_group_p((gid_t)ip->i_d.di_gid)) | |
3574 | mode >>= 3; | |
3575 | } | |
3576 | ||
3577 | /* | |
3578 | * If the DACs are ok we don't need any capability check. | |
3579 | */ | |
3580 | if ((ip->i_d.di_mode & mode) == mode) | |
3581 | return 0; | |
3582 | /* | |
3583 | * Read/write DACs are always overridable. | |
3584 | * Executable DACs are overridable if at least one exec bit is set. | |
3585 | */ | |
3586 | if (!(orgmode & S_IXUSR) || | |
3587 | (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) | |
3588 | if (capable_cred(cr, CAP_DAC_OVERRIDE)) | |
3589 | return 0; | |
3590 | ||
3591 | if ((orgmode == S_IRUSR) || | |
3592 | (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) { | |
3593 | if (capable_cred(cr, CAP_DAC_READ_SEARCH)) | |
3594 | return 0; | |
3595 | #ifdef NOISE | |
3596 | cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode); | |
3597 | #endif /* NOISE */ | |
3598 | return XFS_ERROR(EACCES); | |
3599 | } | |
3600 | return XFS_ERROR(EACCES); | |
3601 | } | |
3602 | ||
3603 | /* | |
3604 | * xfs_iroundup: round up argument to next power of two | |
3605 | */ | |
3606 | uint | |
3607 | xfs_iroundup( | |
3608 | uint v) | |
3609 | { | |
3610 | int i; | |
3611 | uint m; | |
3612 | ||
3613 | if ((v & (v - 1)) == 0) | |
3614 | return v; | |
3615 | ASSERT((v & 0x80000000) == 0); | |
3616 | if ((v & (v + 1)) == 0) | |
3617 | return v + 1; | |
3618 | for (i = 0, m = 1; i < 31; i++, m <<= 1) { | |
3619 | if (v & m) | |
3620 | continue; | |
3621 | v |= m; | |
3622 | if ((v & (v + 1)) == 0) | |
3623 | return v + 1; | |
3624 | } | |
3625 | ASSERT(0); | |
3626 | return( 0 ); | |
3627 | } | |
3628 | ||
1da177e4 LT |
3629 | #ifdef XFS_ILOCK_TRACE |
3630 | ktrace_t *xfs_ilock_trace_buf; | |
3631 | ||
3632 | void | |
3633 | xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra) | |
3634 | { | |
3635 | ktrace_enter(ip->i_lock_trace, | |
3636 | (void *)ip, | |
3637 | (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */ | |
3638 | (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */ | |
3639 | (void *)ra, /* caller of ilock */ | |
3640 | (void *)(unsigned long)current_cpu(), | |
3641 | (void *)(unsigned long)current_pid(), | |
3642 | NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL); | |
3643 | } | |
3644 | #endif | |
4eea22f0 MK |
3645 | |
3646 | /* | |
3647 | * Return a pointer to the extent record at file index idx. | |
3648 | */ | |
3649 | xfs_bmbt_rec_t * | |
3650 | xfs_iext_get_ext( | |
3651 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3652 | xfs_extnum_t idx) /* index of target extent */ | |
3653 | { | |
3654 | ASSERT(idx >= 0); | |
0293ce3a MK |
3655 | if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { |
3656 | return ifp->if_u1.if_ext_irec->er_extbuf; | |
3657 | } else if (ifp->if_flags & XFS_IFEXTIREC) { | |
3658 | xfs_ext_irec_t *erp; /* irec pointer */ | |
3659 | int erp_idx = 0; /* irec index */ | |
3660 | xfs_extnum_t page_idx = idx; /* ext index in target list */ | |
3661 | ||
3662 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
3663 | return &erp->er_extbuf[page_idx]; | |
3664 | } else if (ifp->if_bytes) { | |
4eea22f0 MK |
3665 | return &ifp->if_u1.if_extents[idx]; |
3666 | } else { | |
3667 | return NULL; | |
3668 | } | |
3669 | } | |
3670 | ||
3671 | /* | |
3672 | * Insert new item(s) into the extent records for incore inode | |
3673 | * fork 'ifp'. 'count' new items are inserted at index 'idx'. | |
3674 | */ | |
3675 | void | |
3676 | xfs_iext_insert( | |
3677 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3678 | xfs_extnum_t idx, /* starting index of new items */ | |
3679 | xfs_extnum_t count, /* number of inserted items */ | |
3680 | xfs_bmbt_irec_t *new) /* items to insert */ | |
3681 | { | |
3682 | xfs_bmbt_rec_t *ep; /* extent record pointer */ | |
3683 | xfs_extnum_t i; /* extent record index */ | |
3684 | ||
3685 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
3686 | xfs_iext_add(ifp, idx, count); | |
3687 | for (i = idx; i < idx + count; i++, new++) { | |
3688 | ep = xfs_iext_get_ext(ifp, i); | |
3689 | xfs_bmbt_set_all(ep, new); | |
3690 | } | |
3691 | } | |
3692 | ||
3693 | /* | |
3694 | * This is called when the amount of space required for incore file | |
3695 | * extents needs to be increased. The ext_diff parameter stores the | |
3696 | * number of new extents being added and the idx parameter contains | |
3697 | * the extent index where the new extents will be added. If the new | |
3698 | * extents are being appended, then we just need to (re)allocate and | |
3699 | * initialize the space. Otherwise, if the new extents are being | |
3700 | * inserted into the middle of the existing entries, a bit more work | |
3701 | * is required to make room for the new extents to be inserted. The | |
3702 | * caller is responsible for filling in the new extent entries upon | |
3703 | * return. | |
3704 | */ | |
3705 | void | |
3706 | xfs_iext_add( | |
3707 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3708 | xfs_extnum_t idx, /* index to begin adding exts */ | |
c41564b5 | 3709 | int ext_diff) /* number of extents to add */ |
4eea22f0 MK |
3710 | { |
3711 | int byte_diff; /* new bytes being added */ | |
3712 | int new_size; /* size of extents after adding */ | |
3713 | xfs_extnum_t nextents; /* number of extents in file */ | |
3714 | ||
3715 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3716 | ASSERT((idx >= 0) && (idx <= nextents)); | |
3717 | byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); | |
3718 | new_size = ifp->if_bytes + byte_diff; | |
3719 | /* | |
3720 | * If the new number of extents (nextents + ext_diff) | |
3721 | * fits inside the inode, then continue to use the inline | |
3722 | * extent buffer. | |
3723 | */ | |
3724 | if (nextents + ext_diff <= XFS_INLINE_EXTS) { | |
3725 | if (idx < nextents) { | |
3726 | memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], | |
3727 | &ifp->if_u2.if_inline_ext[idx], | |
3728 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
3729 | memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); | |
3730 | } | |
3731 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
3732 | ifp->if_real_bytes = 0; | |
0293ce3a | 3733 | ifp->if_lastex = nextents + ext_diff; |
4eea22f0 MK |
3734 | } |
3735 | /* | |
3736 | * Otherwise use a linear (direct) extent list. | |
3737 | * If the extents are currently inside the inode, | |
3738 | * xfs_iext_realloc_direct will switch us from | |
3739 | * inline to direct extent allocation mode. | |
3740 | */ | |
0293ce3a | 3741 | else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { |
4eea22f0 MK |
3742 | xfs_iext_realloc_direct(ifp, new_size); |
3743 | if (idx < nextents) { | |
3744 | memmove(&ifp->if_u1.if_extents[idx + ext_diff], | |
3745 | &ifp->if_u1.if_extents[idx], | |
3746 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
3747 | memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); | |
3748 | } | |
3749 | } | |
0293ce3a MK |
3750 | /* Indirection array */ |
3751 | else { | |
3752 | xfs_ext_irec_t *erp; | |
3753 | int erp_idx = 0; | |
3754 | int page_idx = idx; | |
3755 | ||
3756 | ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); | |
3757 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3758 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); | |
3759 | } else { | |
3760 | xfs_iext_irec_init(ifp); | |
3761 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3762 | erp = ifp->if_u1.if_ext_irec; | |
3763 | } | |
3764 | /* Extents fit in target extent page */ | |
3765 | if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { | |
3766 | if (page_idx < erp->er_extcount) { | |
3767 | memmove(&erp->er_extbuf[page_idx + ext_diff], | |
3768 | &erp->er_extbuf[page_idx], | |
3769 | (erp->er_extcount - page_idx) * | |
3770 | sizeof(xfs_bmbt_rec_t)); | |
3771 | memset(&erp->er_extbuf[page_idx], 0, byte_diff); | |
3772 | } | |
3773 | erp->er_extcount += ext_diff; | |
3774 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3775 | } | |
3776 | /* Insert a new extent page */ | |
3777 | else if (erp) { | |
3778 | xfs_iext_add_indirect_multi(ifp, | |
3779 | erp_idx, page_idx, ext_diff); | |
3780 | } | |
3781 | /* | |
3782 | * If extent(s) are being appended to the last page in | |
3783 | * the indirection array and the new extent(s) don't fit | |
3784 | * in the page, then erp is NULL and erp_idx is set to | |
3785 | * the next index needed in the indirection array. | |
3786 | */ | |
3787 | else { | |
3788 | int count = ext_diff; | |
3789 | ||
3790 | while (count) { | |
3791 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3792 | erp->er_extcount = count; | |
3793 | count -= MIN(count, (int)XFS_LINEAR_EXTS); | |
3794 | if (count) { | |
3795 | erp_idx++; | |
3796 | } | |
3797 | } | |
3798 | } | |
3799 | } | |
4eea22f0 MK |
3800 | ifp->if_bytes = new_size; |
3801 | } | |
3802 | ||
0293ce3a MK |
3803 | /* |
3804 | * This is called when incore extents are being added to the indirection | |
3805 | * array and the new extents do not fit in the target extent list. The | |
3806 | * erp_idx parameter contains the irec index for the target extent list | |
3807 | * in the indirection array, and the idx parameter contains the extent | |
3808 | * index within the list. The number of extents being added is stored | |
3809 | * in the count parameter. | |
3810 | * | |
3811 | * |-------| |-------| | |
3812 | * | | | | idx - number of extents before idx | |
3813 | * | idx | | count | | |
3814 | * | | | | count - number of extents being inserted at idx | |
3815 | * |-------| |-------| | |
3816 | * | count | | nex2 | nex2 - number of extents after idx + count | |
3817 | * |-------| |-------| | |
3818 | */ | |
3819 | void | |
3820 | xfs_iext_add_indirect_multi( | |
3821 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3822 | int erp_idx, /* target extent irec index */ | |
3823 | xfs_extnum_t idx, /* index within target list */ | |
3824 | int count) /* new extents being added */ | |
3825 | { | |
3826 | int byte_diff; /* new bytes being added */ | |
3827 | xfs_ext_irec_t *erp; /* pointer to irec entry */ | |
3828 | xfs_extnum_t ext_diff; /* number of extents to add */ | |
3829 | xfs_extnum_t ext_cnt; /* new extents still needed */ | |
3830 | xfs_extnum_t nex2; /* extents after idx + count */ | |
3831 | xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ | |
3832 | int nlists; /* number of irec's (lists) */ | |
3833 | ||
3834 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3835 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3836 | nex2 = erp->er_extcount - idx; | |
3837 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3838 | ||
3839 | /* | |
3840 | * Save second part of target extent list | |
3841 | * (all extents past */ | |
3842 | if (nex2) { | |
3843 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
3844 | nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP); | |
3845 | memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); | |
3846 | erp->er_extcount -= nex2; | |
3847 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); | |
3848 | memset(&erp->er_extbuf[idx], 0, byte_diff); | |
3849 | } | |
3850 | ||
3851 | /* | |
3852 | * Add the new extents to the end of the target | |
3853 | * list, then allocate new irec record(s) and | |
3854 | * extent buffer(s) as needed to store the rest | |
3855 | * of the new extents. | |
3856 | */ | |
3857 | ext_cnt = count; | |
3858 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); | |
3859 | if (ext_diff) { | |
3860 | erp->er_extcount += ext_diff; | |
3861 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3862 | ext_cnt -= ext_diff; | |
3863 | } | |
3864 | while (ext_cnt) { | |
3865 | erp_idx++; | |
3866 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3867 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); | |
3868 | erp->er_extcount = ext_diff; | |
3869 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3870 | ext_cnt -= ext_diff; | |
3871 | } | |
3872 | ||
3873 | /* Add nex2 extents back to indirection array */ | |
3874 | if (nex2) { | |
3875 | xfs_extnum_t ext_avail; | |
3876 | int i; | |
3877 | ||
3878 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
3879 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; | |
3880 | i = 0; | |
3881 | /* | |
3882 | * If nex2 extents fit in the current page, append | |
3883 | * nex2_ep after the new extents. | |
3884 | */ | |
3885 | if (nex2 <= ext_avail) { | |
3886 | i = erp->er_extcount; | |
3887 | } | |
3888 | /* | |
3889 | * Otherwise, check if space is available in the | |
3890 | * next page. | |
3891 | */ | |
3892 | else if ((erp_idx < nlists - 1) && | |
3893 | (nex2 <= (ext_avail = XFS_LINEAR_EXTS - | |
3894 | ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { | |
3895 | erp_idx++; | |
3896 | erp++; | |
3897 | /* Create a hole for nex2 extents */ | |
3898 | memmove(&erp->er_extbuf[nex2], erp->er_extbuf, | |
3899 | erp->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
3900 | } | |
3901 | /* | |
3902 | * Final choice, create a new extent page for | |
3903 | * nex2 extents. | |
3904 | */ | |
3905 | else { | |
3906 | erp_idx++; | |
3907 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3908 | } | |
3909 | memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); | |
3910 | kmem_free(nex2_ep, byte_diff); | |
3911 | erp->er_extcount += nex2; | |
3912 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); | |
3913 | } | |
3914 | } | |
3915 | ||
4eea22f0 MK |
3916 | /* |
3917 | * This is called when the amount of space required for incore file | |
3918 | * extents needs to be decreased. The ext_diff parameter stores the | |
3919 | * number of extents to be removed and the idx parameter contains | |
3920 | * the extent index where the extents will be removed from. | |
0293ce3a MK |
3921 | * |
3922 | * If the amount of space needed has decreased below the linear | |
3923 | * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous | |
3924 | * extent array. Otherwise, use kmem_realloc() to adjust the | |
3925 | * size to what is needed. | |
4eea22f0 MK |
3926 | */ |
3927 | void | |
3928 | xfs_iext_remove( | |
3929 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3930 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3931 | int ext_diff) /* number of extents to remove */ | |
3932 | { | |
3933 | xfs_extnum_t nextents; /* number of extents in file */ | |
3934 | int new_size; /* size of extents after removal */ | |
3935 | ||
3936 | ASSERT(ext_diff > 0); | |
3937 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3938 | new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); | |
3939 | ||
3940 | if (new_size == 0) { | |
3941 | xfs_iext_destroy(ifp); | |
0293ce3a MK |
3942 | } else if (ifp->if_flags & XFS_IFEXTIREC) { |
3943 | xfs_iext_remove_indirect(ifp, idx, ext_diff); | |
4eea22f0 MK |
3944 | } else if (ifp->if_real_bytes) { |
3945 | xfs_iext_remove_direct(ifp, idx, ext_diff); | |
3946 | } else { | |
3947 | xfs_iext_remove_inline(ifp, idx, ext_diff); | |
3948 | } | |
3949 | ifp->if_bytes = new_size; | |
3950 | } | |
3951 | ||
3952 | /* | |
3953 | * This removes ext_diff extents from the inline buffer, beginning | |
3954 | * at extent index idx. | |
3955 | */ | |
3956 | void | |
3957 | xfs_iext_remove_inline( | |
3958 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3959 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3960 | int ext_diff) /* number of extents to remove */ | |
3961 | { | |
3962 | int nextents; /* number of extents in file */ | |
3963 | ||
0293ce3a | 3964 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
3965 | ASSERT(idx < XFS_INLINE_EXTS); |
3966 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3967 | ASSERT(((nextents - ext_diff) > 0) && | |
3968 | (nextents - ext_diff) < XFS_INLINE_EXTS); | |
3969 | ||
3970 | if (idx + ext_diff < nextents) { | |
3971 | memmove(&ifp->if_u2.if_inline_ext[idx], | |
3972 | &ifp->if_u2.if_inline_ext[idx + ext_diff], | |
3973 | (nextents - (idx + ext_diff)) * | |
3974 | sizeof(xfs_bmbt_rec_t)); | |
3975 | memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], | |
3976 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3977 | } else { | |
3978 | memset(&ifp->if_u2.if_inline_ext[idx], 0, | |
3979 | ext_diff * sizeof(xfs_bmbt_rec_t)); | |
3980 | } | |
3981 | } | |
3982 | ||
3983 | /* | |
3984 | * This removes ext_diff extents from a linear (direct) extent list, | |
3985 | * beginning at extent index idx. If the extents are being removed | |
3986 | * from the end of the list (ie. truncate) then we just need to re- | |
3987 | * allocate the list to remove the extra space. Otherwise, if the | |
3988 | * extents are being removed from the middle of the existing extent | |
3989 | * entries, then we first need to move the extent records beginning | |
3990 | * at idx + ext_diff up in the list to overwrite the records being | |
3991 | * removed, then remove the extra space via kmem_realloc. | |
3992 | */ | |
3993 | void | |
3994 | xfs_iext_remove_direct( | |
3995 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3996 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3997 | int ext_diff) /* number of extents to remove */ | |
3998 | { | |
3999 | xfs_extnum_t nextents; /* number of extents in file */ | |
4000 | int new_size; /* size of extents after removal */ | |
4001 | ||
0293ce3a | 4002 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
4003 | new_size = ifp->if_bytes - |
4004 | (ext_diff * sizeof(xfs_bmbt_rec_t)); | |
4005 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4006 | ||
4007 | if (new_size == 0) { | |
4008 | xfs_iext_destroy(ifp); | |
4009 | return; | |
4010 | } | |
4011 | /* Move extents up in the list (if needed) */ | |
4012 | if (idx + ext_diff < nextents) { | |
4013 | memmove(&ifp->if_u1.if_extents[idx], | |
4014 | &ifp->if_u1.if_extents[idx + ext_diff], | |
4015 | (nextents - (idx + ext_diff)) * | |
4016 | sizeof(xfs_bmbt_rec_t)); | |
4017 | } | |
4018 | memset(&ifp->if_u1.if_extents[nextents - ext_diff], | |
4019 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
4020 | /* | |
4021 | * Reallocate the direct extent list. If the extents | |
4022 | * will fit inside the inode then xfs_iext_realloc_direct | |
4023 | * will switch from direct to inline extent allocation | |
4024 | * mode for us. | |
4025 | */ | |
4026 | xfs_iext_realloc_direct(ifp, new_size); | |
4027 | ifp->if_bytes = new_size; | |
4028 | } | |
4029 | ||
0293ce3a MK |
4030 | /* |
4031 | * This is called when incore extents are being removed from the | |
4032 | * indirection array and the extents being removed span multiple extent | |
4033 | * buffers. The idx parameter contains the file extent index where we | |
4034 | * want to begin removing extents, and the count parameter contains | |
4035 | * how many extents need to be removed. | |
4036 | * | |
4037 | * |-------| |-------| | |
4038 | * | nex1 | | | nex1 - number of extents before idx | |
4039 | * |-------| | count | | |
4040 | * | | | | count - number of extents being removed at idx | |
4041 | * | count | |-------| | |
4042 | * | | | nex2 | nex2 - number of extents after idx + count | |
4043 | * |-------| |-------| | |
4044 | */ | |
4045 | void | |
4046 | xfs_iext_remove_indirect( | |
4047 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4048 | xfs_extnum_t idx, /* index to begin removing extents */ | |
4049 | int count) /* number of extents to remove */ | |
4050 | { | |
4051 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4052 | int erp_idx = 0; /* indirection array index */ | |
4053 | xfs_extnum_t ext_cnt; /* extents left to remove */ | |
4054 | xfs_extnum_t ext_diff; /* extents to remove in current list */ | |
4055 | xfs_extnum_t nex1; /* number of extents before idx */ | |
4056 | xfs_extnum_t nex2; /* extents after idx + count */ | |
c41564b5 | 4057 | int nlists; /* entries in indirection array */ |
0293ce3a MK |
4058 | int page_idx = idx; /* index in target extent list */ |
4059 | ||
4060 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4061 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
4062 | ASSERT(erp != NULL); | |
4063 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4064 | nex1 = page_idx; | |
4065 | ext_cnt = count; | |
4066 | while (ext_cnt) { | |
4067 | nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); | |
4068 | ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); | |
4069 | /* | |
4070 | * Check for deletion of entire list; | |
4071 | * xfs_iext_irec_remove() updates extent offsets. | |
4072 | */ | |
4073 | if (ext_diff == erp->er_extcount) { | |
4074 | xfs_iext_irec_remove(ifp, erp_idx); | |
4075 | ext_cnt -= ext_diff; | |
4076 | nex1 = 0; | |
4077 | if (ext_cnt) { | |
4078 | ASSERT(erp_idx < ifp->if_real_bytes / | |
4079 | XFS_IEXT_BUFSZ); | |
4080 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4081 | nex1 = 0; | |
4082 | continue; | |
4083 | } else { | |
4084 | break; | |
4085 | } | |
4086 | } | |
4087 | /* Move extents up (if needed) */ | |
4088 | if (nex2) { | |
4089 | memmove(&erp->er_extbuf[nex1], | |
4090 | &erp->er_extbuf[nex1 + ext_diff], | |
4091 | nex2 * sizeof(xfs_bmbt_rec_t)); | |
4092 | } | |
4093 | /* Zero out rest of page */ | |
4094 | memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - | |
4095 | ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); | |
4096 | /* Update remaining counters */ | |
4097 | erp->er_extcount -= ext_diff; | |
4098 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); | |
4099 | ext_cnt -= ext_diff; | |
4100 | nex1 = 0; | |
4101 | erp_idx++; | |
4102 | erp++; | |
4103 | } | |
4104 | ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); | |
4105 | xfs_iext_irec_compact(ifp); | |
4106 | } | |
4107 | ||
4eea22f0 MK |
4108 | /* |
4109 | * Create, destroy, or resize a linear (direct) block of extents. | |
4110 | */ | |
4111 | void | |
4112 | xfs_iext_realloc_direct( | |
4113 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4114 | int new_size) /* new size of extents */ | |
4115 | { | |
4116 | int rnew_size; /* real new size of extents */ | |
4117 | ||
4118 | rnew_size = new_size; | |
4119 | ||
0293ce3a MK |
4120 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || |
4121 | ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && | |
4122 | (new_size != ifp->if_real_bytes))); | |
4123 | ||
4eea22f0 MK |
4124 | /* Free extent records */ |
4125 | if (new_size == 0) { | |
4126 | xfs_iext_destroy(ifp); | |
4127 | } | |
4128 | /* Resize direct extent list and zero any new bytes */ | |
4129 | else if (ifp->if_real_bytes) { | |
4130 | /* Check if extents will fit inside the inode */ | |
4131 | if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { | |
4132 | xfs_iext_direct_to_inline(ifp, new_size / | |
4133 | (uint)sizeof(xfs_bmbt_rec_t)); | |
4134 | ifp->if_bytes = new_size; | |
4135 | return; | |
4136 | } | |
4137 | if ((new_size & (new_size - 1)) != 0) { | |
4138 | rnew_size = xfs_iroundup(new_size); | |
4139 | } | |
4140 | if (rnew_size != ifp->if_real_bytes) { | |
4141 | ifp->if_u1.if_extents = (xfs_bmbt_rec_t *) | |
4142 | kmem_realloc(ifp->if_u1.if_extents, | |
4143 | rnew_size, | |
4144 | ifp->if_real_bytes, | |
4145 | KM_SLEEP); | |
4146 | } | |
4147 | if (rnew_size > ifp->if_real_bytes) { | |
4148 | memset(&ifp->if_u1.if_extents[ifp->if_bytes / | |
4149 | (uint)sizeof(xfs_bmbt_rec_t)], 0, | |
4150 | rnew_size - ifp->if_real_bytes); | |
4151 | } | |
4152 | } | |
4153 | /* | |
4154 | * Switch from the inline extent buffer to a direct | |
4155 | * extent list. Be sure to include the inline extent | |
4156 | * bytes in new_size. | |
4157 | */ | |
4158 | else { | |
4159 | new_size += ifp->if_bytes; | |
4160 | if ((new_size & (new_size - 1)) != 0) { | |
4161 | rnew_size = xfs_iroundup(new_size); | |
4162 | } | |
4163 | xfs_iext_inline_to_direct(ifp, rnew_size); | |
4164 | } | |
4165 | ifp->if_real_bytes = rnew_size; | |
4166 | ifp->if_bytes = new_size; | |
4167 | } | |
4168 | ||
4169 | /* | |
4170 | * Switch from linear (direct) extent records to inline buffer. | |
4171 | */ | |
4172 | void | |
4173 | xfs_iext_direct_to_inline( | |
4174 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4175 | xfs_extnum_t nextents) /* number of extents in file */ | |
4176 | { | |
4177 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
4178 | ASSERT(nextents <= XFS_INLINE_EXTS); | |
4179 | /* | |
4180 | * The inline buffer was zeroed when we switched | |
4181 | * from inline to direct extent allocation mode, | |
4182 | * so we don't need to clear it here. | |
4183 | */ | |
4184 | memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, | |
4185 | nextents * sizeof(xfs_bmbt_rec_t)); | |
fe6c1e72 | 4186 | kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes); |
4eea22f0 MK |
4187 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; |
4188 | ifp->if_real_bytes = 0; | |
4189 | } | |
4190 | ||
4191 | /* | |
4192 | * Switch from inline buffer to linear (direct) extent records. | |
4193 | * new_size should already be rounded up to the next power of 2 | |
4194 | * by the caller (when appropriate), so use new_size as it is. | |
4195 | * However, since new_size may be rounded up, we can't update | |
4196 | * if_bytes here. It is the caller's responsibility to update | |
4197 | * if_bytes upon return. | |
4198 | */ | |
4199 | void | |
4200 | xfs_iext_inline_to_direct( | |
4201 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4202 | int new_size) /* number of extents in file */ | |
4203 | { | |
4204 | ifp->if_u1.if_extents = (xfs_bmbt_rec_t *) | |
4205 | kmem_alloc(new_size, KM_SLEEP); | |
4206 | memset(ifp->if_u1.if_extents, 0, new_size); | |
4207 | if (ifp->if_bytes) { | |
4208 | memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, | |
4209 | ifp->if_bytes); | |
4210 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
4211 | sizeof(xfs_bmbt_rec_t)); | |
4212 | } | |
4213 | ifp->if_real_bytes = new_size; | |
4214 | } | |
4215 | ||
0293ce3a MK |
4216 | /* |
4217 | * Resize an extent indirection array to new_size bytes. | |
4218 | */ | |
4219 | void | |
4220 | xfs_iext_realloc_indirect( | |
4221 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4222 | int new_size) /* new indirection array size */ | |
4223 | { | |
4224 | int nlists; /* number of irec's (ex lists) */ | |
4225 | int size; /* current indirection array size */ | |
4226 | ||
4227 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4228 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4229 | size = nlists * sizeof(xfs_ext_irec_t); | |
4230 | ASSERT(ifp->if_real_bytes); | |
4231 | ASSERT((new_size >= 0) && (new_size != size)); | |
4232 | if (new_size == 0) { | |
4233 | xfs_iext_destroy(ifp); | |
4234 | } else { | |
4235 | ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) | |
4236 | kmem_realloc(ifp->if_u1.if_ext_irec, | |
4237 | new_size, size, KM_SLEEP); | |
4238 | } | |
4239 | } | |
4240 | ||
4241 | /* | |
4242 | * Switch from indirection array to linear (direct) extent allocations. | |
4243 | */ | |
4244 | void | |
4245 | xfs_iext_indirect_to_direct( | |
4246 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4247 | { | |
4248 | xfs_bmbt_rec_t *ep; /* extent record pointer */ | |
4249 | xfs_extnum_t nextents; /* number of extents in file */ | |
4250 | int size; /* size of file extents */ | |
4251 | ||
4252 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4253 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4254 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
4255 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
4256 | ||
4257 | xfs_iext_irec_compact_full(ifp); | |
4258 | ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); | |
4259 | ||
4260 | ep = ifp->if_u1.if_ext_irec->er_extbuf; | |
4261 | kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t)); | |
4262 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
4263 | ifp->if_u1.if_extents = ep; | |
4264 | ifp->if_bytes = size; | |
4265 | if (nextents < XFS_LINEAR_EXTS) { | |
4266 | xfs_iext_realloc_direct(ifp, size); | |
4267 | } | |
4268 | } | |
4269 | ||
4eea22f0 MK |
4270 | /* |
4271 | * Free incore file extents. | |
4272 | */ | |
4273 | void | |
4274 | xfs_iext_destroy( | |
4275 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4276 | { | |
0293ce3a MK |
4277 | if (ifp->if_flags & XFS_IFEXTIREC) { |
4278 | int erp_idx; | |
4279 | int nlists; | |
4280 | ||
4281 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4282 | for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { | |
4283 | xfs_iext_irec_remove(ifp, erp_idx); | |
4284 | } | |
4285 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
4286 | } else if (ifp->if_real_bytes) { | |
4eea22f0 MK |
4287 | kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes); |
4288 | } else if (ifp->if_bytes) { | |
4289 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
4290 | sizeof(xfs_bmbt_rec_t)); | |
4291 | } | |
4292 | ifp->if_u1.if_extents = NULL; | |
4293 | ifp->if_real_bytes = 0; | |
4294 | ifp->if_bytes = 0; | |
4295 | } | |
0293ce3a | 4296 | |
8867bc9b MK |
4297 | /* |
4298 | * Return a pointer to the extent record for file system block bno. | |
4299 | */ | |
4300 | xfs_bmbt_rec_t * /* pointer to found extent record */ | |
4301 | xfs_iext_bno_to_ext( | |
4302 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4303 | xfs_fileoff_t bno, /* block number to search for */ | |
4304 | xfs_extnum_t *idxp) /* index of target extent */ | |
4305 | { | |
4306 | xfs_bmbt_rec_t *base; /* pointer to first extent */ | |
4307 | xfs_filblks_t blockcount = 0; /* number of blocks in extent */ | |
4308 | xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */ | |
4309 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ | |
c41564b5 | 4310 | int high; /* upper boundary in search */ |
8867bc9b | 4311 | xfs_extnum_t idx = 0; /* index of target extent */ |
c41564b5 | 4312 | int low; /* lower boundary in search */ |
8867bc9b MK |
4313 | xfs_extnum_t nextents; /* number of file extents */ |
4314 | xfs_fileoff_t startoff = 0; /* start offset of extent */ | |
4315 | ||
4316 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4317 | if (nextents == 0) { | |
4318 | *idxp = 0; | |
4319 | return NULL; | |
4320 | } | |
4321 | low = 0; | |
4322 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4323 | /* Find target extent list */ | |
4324 | int erp_idx = 0; | |
4325 | erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); | |
4326 | base = erp->er_extbuf; | |
4327 | high = erp->er_extcount - 1; | |
4328 | } else { | |
4329 | base = ifp->if_u1.if_extents; | |
4330 | high = nextents - 1; | |
4331 | } | |
4332 | /* Binary search extent records */ | |
4333 | while (low <= high) { | |
4334 | idx = (low + high) >> 1; | |
4335 | ep = base + idx; | |
4336 | startoff = xfs_bmbt_get_startoff(ep); | |
4337 | blockcount = xfs_bmbt_get_blockcount(ep); | |
4338 | if (bno < startoff) { | |
4339 | high = idx - 1; | |
4340 | } else if (bno >= startoff + blockcount) { | |
4341 | low = idx + 1; | |
4342 | } else { | |
4343 | /* Convert back to file-based extent index */ | |
4344 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4345 | idx += erp->er_extoff; | |
4346 | } | |
4347 | *idxp = idx; | |
4348 | return ep; | |
4349 | } | |
4350 | } | |
4351 | /* Convert back to file-based extent index */ | |
4352 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4353 | idx += erp->er_extoff; | |
4354 | } | |
4355 | if (bno >= startoff + blockcount) { | |
4356 | if (++idx == nextents) { | |
4357 | ep = NULL; | |
4358 | } else { | |
4359 | ep = xfs_iext_get_ext(ifp, idx); | |
4360 | } | |
4361 | } | |
4362 | *idxp = idx; | |
4363 | return ep; | |
4364 | } | |
4365 | ||
0293ce3a MK |
4366 | /* |
4367 | * Return a pointer to the indirection array entry containing the | |
4368 | * extent record for filesystem block bno. Store the index of the | |
4369 | * target irec in *erp_idxp. | |
4370 | */ | |
8867bc9b | 4371 | xfs_ext_irec_t * /* pointer to found extent record */ |
0293ce3a MK |
4372 | xfs_iext_bno_to_irec( |
4373 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4374 | xfs_fileoff_t bno, /* block number to search for */ | |
4375 | int *erp_idxp) /* irec index of target ext list */ | |
4376 | { | |
4377 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ | |
4378 | xfs_ext_irec_t *erp_next; /* next indirection array entry */ | |
8867bc9b | 4379 | int erp_idx; /* indirection array index */ |
0293ce3a MK |
4380 | int nlists; /* number of extent irec's (lists) */ |
4381 | int high; /* binary search upper limit */ | |
4382 | int low; /* binary search lower limit */ | |
4383 | ||
4384 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4385 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4386 | erp_idx = 0; | |
4387 | low = 0; | |
4388 | high = nlists - 1; | |
4389 | while (low <= high) { | |
4390 | erp_idx = (low + high) >> 1; | |
4391 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4392 | erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; | |
4393 | if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { | |
4394 | high = erp_idx - 1; | |
4395 | } else if (erp_next && bno >= | |
4396 | xfs_bmbt_get_startoff(erp_next->er_extbuf)) { | |
4397 | low = erp_idx + 1; | |
4398 | } else { | |
4399 | break; | |
4400 | } | |
4401 | } | |
4402 | *erp_idxp = erp_idx; | |
4403 | return erp; | |
4404 | } | |
4405 | ||
4406 | /* | |
4407 | * Return a pointer to the indirection array entry containing the | |
4408 | * extent record at file extent index *idxp. Store the index of the | |
4409 | * target irec in *erp_idxp and store the page index of the target | |
4410 | * extent record in *idxp. | |
4411 | */ | |
4412 | xfs_ext_irec_t * | |
4413 | xfs_iext_idx_to_irec( | |
4414 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4415 | xfs_extnum_t *idxp, /* extent index (file -> page) */ | |
4416 | int *erp_idxp, /* pointer to target irec */ | |
4417 | int realloc) /* new bytes were just added */ | |
4418 | { | |
4419 | xfs_ext_irec_t *prev; /* pointer to previous irec */ | |
4420 | xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ | |
4421 | int erp_idx; /* indirection array index */ | |
4422 | int nlists; /* number of irec's (ex lists) */ | |
4423 | int high; /* binary search upper limit */ | |
4424 | int low; /* binary search lower limit */ | |
4425 | xfs_extnum_t page_idx = *idxp; /* extent index in target list */ | |
4426 | ||
4427 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4428 | ASSERT(page_idx >= 0 && page_idx <= | |
4429 | ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)); | |
4430 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4431 | erp_idx = 0; | |
4432 | low = 0; | |
4433 | high = nlists - 1; | |
4434 | ||
4435 | /* Binary search extent irec's */ | |
4436 | while (low <= high) { | |
4437 | erp_idx = (low + high) >> 1; | |
4438 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4439 | prev = erp_idx > 0 ? erp - 1 : NULL; | |
4440 | if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && | |
4441 | realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { | |
4442 | high = erp_idx - 1; | |
4443 | } else if (page_idx > erp->er_extoff + erp->er_extcount || | |
4444 | (page_idx == erp->er_extoff + erp->er_extcount && | |
4445 | !realloc)) { | |
4446 | low = erp_idx + 1; | |
4447 | } else if (page_idx == erp->er_extoff + erp->er_extcount && | |
4448 | erp->er_extcount == XFS_LINEAR_EXTS) { | |
4449 | ASSERT(realloc); | |
4450 | page_idx = 0; | |
4451 | erp_idx++; | |
4452 | erp = erp_idx < nlists ? erp + 1 : NULL; | |
4453 | break; | |
4454 | } else { | |
4455 | page_idx -= erp->er_extoff; | |
4456 | break; | |
4457 | } | |
4458 | } | |
4459 | *idxp = page_idx; | |
4460 | *erp_idxp = erp_idx; | |
4461 | return(erp); | |
4462 | } | |
4463 | ||
4464 | /* | |
4465 | * Allocate and initialize an indirection array once the space needed | |
4466 | * for incore extents increases above XFS_IEXT_BUFSZ. | |
4467 | */ | |
4468 | void | |
4469 | xfs_iext_irec_init( | |
4470 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4471 | { | |
4472 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4473 | xfs_extnum_t nextents; /* number of extents in file */ | |
4474 | ||
4475 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); | |
4476 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4477 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
4478 | ||
4479 | erp = (xfs_ext_irec_t *) | |
4480 | kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP); | |
4481 | ||
4482 | if (nextents == 0) { | |
4483 | ifp->if_u1.if_extents = (xfs_bmbt_rec_t *) | |
4484 | kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP); | |
4485 | } else if (!ifp->if_real_bytes) { | |
4486 | xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); | |
4487 | } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { | |
4488 | xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); | |
4489 | } | |
4490 | erp->er_extbuf = ifp->if_u1.if_extents; | |
4491 | erp->er_extcount = nextents; | |
4492 | erp->er_extoff = 0; | |
4493 | ||
4494 | ifp->if_flags |= XFS_IFEXTIREC; | |
4495 | ifp->if_real_bytes = XFS_IEXT_BUFSZ; | |
4496 | ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); | |
4497 | ifp->if_u1.if_ext_irec = erp; | |
4498 | ||
4499 | return; | |
4500 | } | |
4501 | ||
4502 | /* | |
4503 | * Allocate and initialize a new entry in the indirection array. | |
4504 | */ | |
4505 | xfs_ext_irec_t * | |
4506 | xfs_iext_irec_new( | |
4507 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4508 | int erp_idx) /* index for new irec */ | |
4509 | { | |
4510 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4511 | int i; /* loop counter */ | |
4512 | int nlists; /* number of irec's (ex lists) */ | |
4513 | ||
4514 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4515 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4516 | ||
4517 | /* Resize indirection array */ | |
4518 | xfs_iext_realloc_indirect(ifp, ++nlists * | |
4519 | sizeof(xfs_ext_irec_t)); | |
4520 | /* | |
4521 | * Move records down in the array so the | |
4522 | * new page can use erp_idx. | |
4523 | */ | |
4524 | erp = ifp->if_u1.if_ext_irec; | |
4525 | for (i = nlists - 1; i > erp_idx; i--) { | |
4526 | memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); | |
4527 | } | |
4528 | ASSERT(i == erp_idx); | |
4529 | ||
4530 | /* Initialize new extent record */ | |
4531 | erp = ifp->if_u1.if_ext_irec; | |
4532 | erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *) | |
4533 | kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP); | |
4534 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; | |
4535 | memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); | |
4536 | erp[erp_idx].er_extcount = 0; | |
4537 | erp[erp_idx].er_extoff = erp_idx > 0 ? | |
4538 | erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; | |
4539 | return (&erp[erp_idx]); | |
4540 | } | |
4541 | ||
4542 | /* | |
4543 | * Remove a record from the indirection array. | |
4544 | */ | |
4545 | void | |
4546 | xfs_iext_irec_remove( | |
4547 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4548 | int erp_idx) /* irec index to remove */ | |
4549 | { | |
4550 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4551 | int i; /* loop counter */ | |
4552 | int nlists; /* number of irec's (ex lists) */ | |
4553 | ||
4554 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4555 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4556 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4557 | if (erp->er_extbuf) { | |
4558 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, | |
4559 | -erp->er_extcount); | |
4560 | kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ); | |
4561 | } | |
4562 | /* Compact extent records */ | |
4563 | erp = ifp->if_u1.if_ext_irec; | |
4564 | for (i = erp_idx; i < nlists - 1; i++) { | |
4565 | memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); | |
4566 | } | |
4567 | /* | |
4568 | * Manually free the last extent record from the indirection | |
4569 | * array. A call to xfs_iext_realloc_indirect() with a size | |
4570 | * of zero would result in a call to xfs_iext_destroy() which | |
4571 | * would in turn call this function again, creating a nasty | |
4572 | * infinite loop. | |
4573 | */ | |
4574 | if (--nlists) { | |
4575 | xfs_iext_realloc_indirect(ifp, | |
4576 | nlists * sizeof(xfs_ext_irec_t)); | |
4577 | } else { | |
4578 | kmem_free(ifp->if_u1.if_ext_irec, | |
4579 | sizeof(xfs_ext_irec_t)); | |
4580 | } | |
4581 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; | |
4582 | } | |
4583 | ||
4584 | /* | |
4585 | * This is called to clean up large amounts of unused memory allocated | |
4586 | * by the indirection array. Before compacting anything though, verify | |
4587 | * that the indirection array is still needed and switch back to the | |
4588 | * linear extent list (or even the inline buffer) if possible. The | |
4589 | * compaction policy is as follows: | |
4590 | * | |
4591 | * Full Compaction: Extents fit into a single page (or inline buffer) | |
4592 | * Full Compaction: Extents occupy less than 10% of allocated space | |
4593 | * Partial Compaction: Extents occupy > 10% and < 50% of allocated space | |
4594 | * No Compaction: Extents occupy at least 50% of allocated space | |
4595 | */ | |
4596 | void | |
4597 | xfs_iext_irec_compact( | |
4598 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4599 | { | |
4600 | xfs_extnum_t nextents; /* number of extents in file */ | |
4601 | int nlists; /* number of irec's (ex lists) */ | |
4602 | ||
4603 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4604 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4605 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4606 | ||
4607 | if (nextents == 0) { | |
4608 | xfs_iext_destroy(ifp); | |
4609 | } else if (nextents <= XFS_INLINE_EXTS) { | |
4610 | xfs_iext_indirect_to_direct(ifp); | |
4611 | xfs_iext_direct_to_inline(ifp, nextents); | |
4612 | } else if (nextents <= XFS_LINEAR_EXTS) { | |
4613 | xfs_iext_indirect_to_direct(ifp); | |
4614 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) { | |
4615 | xfs_iext_irec_compact_full(ifp); | |
4616 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { | |
4617 | xfs_iext_irec_compact_pages(ifp); | |
4618 | } | |
4619 | } | |
4620 | ||
4621 | /* | |
4622 | * Combine extents from neighboring extent pages. | |
4623 | */ | |
4624 | void | |
4625 | xfs_iext_irec_compact_pages( | |
4626 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4627 | { | |
4628 | xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ | |
4629 | int erp_idx = 0; /* indirection array index */ | |
4630 | int nlists; /* number of irec's (ex lists) */ | |
4631 | ||
4632 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4633 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4634 | while (erp_idx < nlists - 1) { | |
4635 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4636 | erp_next = erp + 1; | |
4637 | if (erp_next->er_extcount <= | |
4638 | (XFS_LINEAR_EXTS - erp->er_extcount)) { | |
4639 | memmove(&erp->er_extbuf[erp->er_extcount], | |
4640 | erp_next->er_extbuf, erp_next->er_extcount * | |
4641 | sizeof(xfs_bmbt_rec_t)); | |
4642 | erp->er_extcount += erp_next->er_extcount; | |
4643 | /* | |
4644 | * Free page before removing extent record | |
4645 | * so er_extoffs don't get modified in | |
4646 | * xfs_iext_irec_remove. | |
4647 | */ | |
4648 | kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ); | |
4649 | erp_next->er_extbuf = NULL; | |
4650 | xfs_iext_irec_remove(ifp, erp_idx + 1); | |
4651 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4652 | } else { | |
4653 | erp_idx++; | |
4654 | } | |
4655 | } | |
4656 | } | |
4657 | ||
4658 | /* | |
4659 | * Fully compact the extent records managed by the indirection array. | |
4660 | */ | |
4661 | void | |
4662 | xfs_iext_irec_compact_full( | |
4663 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4664 | { | |
4665 | xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */ | |
4666 | xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */ | |
4667 | int erp_idx = 0; /* extent irec index */ | |
4668 | int ext_avail; /* empty entries in ex list */ | |
4669 | int ext_diff; /* number of exts to add */ | |
4670 | int nlists; /* number of irec's (ex lists) */ | |
4671 | ||
4672 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4673 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4674 | erp = ifp->if_u1.if_ext_irec; | |
4675 | ep = &erp->er_extbuf[erp->er_extcount]; | |
4676 | erp_next = erp + 1; | |
4677 | ep_next = erp_next->er_extbuf; | |
4678 | while (erp_idx < nlists - 1) { | |
4679 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; | |
4680 | ext_diff = MIN(ext_avail, erp_next->er_extcount); | |
4681 | memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
4682 | erp->er_extcount += ext_diff; | |
4683 | erp_next->er_extcount -= ext_diff; | |
4684 | /* Remove next page */ | |
4685 | if (erp_next->er_extcount == 0) { | |
4686 | /* | |
4687 | * Free page before removing extent record | |
4688 | * so er_extoffs don't get modified in | |
4689 | * xfs_iext_irec_remove. | |
4690 | */ | |
4691 | kmem_free(erp_next->er_extbuf, | |
4692 | erp_next->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
4693 | erp_next->er_extbuf = NULL; | |
4694 | xfs_iext_irec_remove(ifp, erp_idx + 1); | |
4695 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4696 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4697 | /* Update next page */ | |
4698 | } else { | |
4699 | /* Move rest of page up to become next new page */ | |
4700 | memmove(erp_next->er_extbuf, ep_next, | |
4701 | erp_next->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
4702 | ep_next = erp_next->er_extbuf; | |
4703 | memset(&ep_next[erp_next->er_extcount], 0, | |
4704 | (XFS_LINEAR_EXTS - erp_next->er_extcount) * | |
4705 | sizeof(xfs_bmbt_rec_t)); | |
4706 | } | |
4707 | if (erp->er_extcount == XFS_LINEAR_EXTS) { | |
4708 | erp_idx++; | |
4709 | if (erp_idx < nlists) | |
4710 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4711 | else | |
4712 | break; | |
4713 | } | |
4714 | ep = &erp->er_extbuf[erp->er_extcount]; | |
4715 | erp_next = erp + 1; | |
4716 | ep_next = erp_next->er_extbuf; | |
4717 | } | |
4718 | } | |
4719 | ||
4720 | /* | |
4721 | * This is called to update the er_extoff field in the indirection | |
4722 | * array when extents have been added or removed from one of the | |
4723 | * extent lists. erp_idx contains the irec index to begin updating | |
4724 | * at and ext_diff contains the number of extents that were added | |
4725 | * or removed. | |
4726 | */ | |
4727 | void | |
4728 | xfs_iext_irec_update_extoffs( | |
4729 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4730 | int erp_idx, /* irec index to update */ | |
4731 | int ext_diff) /* number of new extents */ | |
4732 | { | |
4733 | int i; /* loop counter */ | |
4734 | int nlists; /* number of irec's (ex lists */ | |
4735 | ||
4736 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4737 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4738 | for (i = erp_idx; i < nlists; i++) { | |
4739 | ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; | |
4740 | } | |
4741 | } |