]>
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" | |
1da177e4 | 50 | #include "xfs_acl.h" |
2a82b8be | 51 | #include "xfs_filestream.h" |
739bfb2a | 52 | #include "xfs_vnodeops.h" |
1da177e4 | 53 | |
1da177e4 LT |
54 | kmem_zone_t *xfs_ifork_zone; |
55 | kmem_zone_t *xfs_inode_zone; | |
da353b0d | 56 | kmem_zone_t *xfs_icluster_zone; |
1da177e4 LT |
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 | ||
1da177e4 LT |
69 | #ifdef DEBUG |
70 | /* | |
71 | * Make sure that the extents in the given memory buffer | |
72 | * are valid. | |
73 | */ | |
74 | STATIC void | |
75 | xfs_validate_extents( | |
4eea22f0 | 76 | xfs_ifork_t *ifp, |
1da177e4 | 77 | int nrecs, |
1da177e4 LT |
78 | xfs_exntfmt_t fmt) |
79 | { | |
80 | xfs_bmbt_irec_t irec; | |
a6f64d4a | 81 | xfs_bmbt_rec_host_t rec; |
1da177e4 LT |
82 | int i; |
83 | ||
84 | for (i = 0; i < nrecs; i++) { | |
a6f64d4a CH |
85 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
86 | rec.l0 = get_unaligned(&ep->l0); | |
87 | rec.l1 = get_unaligned(&ep->l1); | |
88 | xfs_bmbt_get_all(&rec, &irec); | |
1da177e4 LT |
89 | if (fmt == XFS_EXTFMT_NOSTATE) |
90 | ASSERT(irec.br_state == XFS_EXT_NORM); | |
1da177e4 LT |
91 | } |
92 | } | |
93 | #else /* DEBUG */ | |
a6f64d4a | 94 | #define xfs_validate_extents(ifp, nrecs, fmt) |
1da177e4 LT |
95 | #endif /* DEBUG */ |
96 | ||
97 | /* | |
98 | * Check that none of the inode's in the buffer have a next | |
99 | * unlinked field of 0. | |
100 | */ | |
101 | #if defined(DEBUG) | |
102 | void | |
103 | xfs_inobp_check( | |
104 | xfs_mount_t *mp, | |
105 | xfs_buf_t *bp) | |
106 | { | |
107 | int i; | |
108 | int j; | |
109 | xfs_dinode_t *dip; | |
110 | ||
111 | j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; | |
112 | ||
113 | for (i = 0; i < j; i++) { | |
114 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
115 | i * mp->m_sb.sb_inodesize); | |
116 | if (!dip->di_next_unlinked) { | |
117 | xfs_fs_cmn_err(CE_ALERT, mp, | |
118 | "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.", | |
119 | bp); | |
120 | ASSERT(dip->di_next_unlinked); | |
121 | } | |
122 | } | |
123 | } | |
124 | #endif | |
125 | ||
1da177e4 LT |
126 | /* |
127 | * This routine is called to map an inode number within a file | |
128 | * system to the buffer containing the on-disk version of the | |
129 | * inode. It returns a pointer to the buffer containing the | |
130 | * on-disk inode in the bpp parameter, and in the dip parameter | |
131 | * it returns a pointer to the on-disk inode within that buffer. | |
132 | * | |
133 | * If a non-zero error is returned, then the contents of bpp and | |
134 | * dipp are undefined. | |
135 | * | |
136 | * Use xfs_imap() to determine the size and location of the | |
137 | * buffer to read from disk. | |
138 | */ | |
ba0f32d4 | 139 | STATIC int |
1da177e4 LT |
140 | xfs_inotobp( |
141 | xfs_mount_t *mp, | |
142 | xfs_trans_t *tp, | |
143 | xfs_ino_t ino, | |
144 | xfs_dinode_t **dipp, | |
145 | xfs_buf_t **bpp, | |
146 | int *offset) | |
147 | { | |
148 | int di_ok; | |
149 | xfs_imap_t imap; | |
150 | xfs_buf_t *bp; | |
151 | int error; | |
152 | xfs_dinode_t *dip; | |
153 | ||
154 | /* | |
c41564b5 | 155 | * Call the space management code to find the location of the |
1da177e4 LT |
156 | * inode on disk. |
157 | */ | |
158 | imap.im_blkno = 0; | |
159 | error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP); | |
160 | if (error != 0) { | |
161 | cmn_err(CE_WARN, | |
162 | "xfs_inotobp: xfs_imap() returned an " | |
163 | "error %d on %s. Returning error.", error, mp->m_fsname); | |
164 | return error; | |
165 | } | |
166 | ||
167 | /* | |
168 | * If the inode number maps to a block outside the bounds of the | |
169 | * file system then return NULL rather than calling read_buf | |
170 | * and panicing when we get an error from the driver. | |
171 | */ | |
172 | if ((imap.im_blkno + imap.im_len) > | |
173 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
174 | cmn_err(CE_WARN, | |
da1650a5 | 175 | "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds " |
1da177e4 | 176 | "of the file system %s. Returning EINVAL.", |
da1650a5 CH |
177 | (unsigned long long)imap.im_blkno, |
178 | imap.im_len, mp->m_fsname); | |
1da177e4 LT |
179 | return XFS_ERROR(EINVAL); |
180 | } | |
181 | ||
182 | /* | |
183 | * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will | |
184 | * default to just a read_buf() call. | |
185 | */ | |
186 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno, | |
187 | (int)imap.im_len, XFS_BUF_LOCK, &bp); | |
188 | ||
189 | if (error) { | |
190 | cmn_err(CE_WARN, | |
191 | "xfs_inotobp: xfs_trans_read_buf() returned an " | |
192 | "error %d on %s. Returning error.", error, mp->m_fsname); | |
193 | return error; | |
194 | } | |
195 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0); | |
196 | di_ok = | |
347d1c01 CH |
197 | be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC && |
198 | XFS_DINODE_GOOD_VERSION(dip->di_core.di_version); | |
1da177e4 LT |
199 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP, |
200 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
201 | XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip); | |
202 | xfs_trans_brelse(tp, bp); | |
203 | cmn_err(CE_WARN, | |
204 | "xfs_inotobp: XFS_TEST_ERROR() returned an " | |
205 | "error on %s. Returning EFSCORRUPTED.", mp->m_fsname); | |
206 | return XFS_ERROR(EFSCORRUPTED); | |
207 | } | |
208 | ||
209 | xfs_inobp_check(mp, bp); | |
210 | ||
211 | /* | |
212 | * Set *dipp to point to the on-disk inode in the buffer. | |
213 | */ | |
214 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); | |
215 | *bpp = bp; | |
216 | *offset = imap.im_boffset; | |
217 | return 0; | |
218 | } | |
219 | ||
220 | ||
221 | /* | |
222 | * This routine is called to map an inode to the buffer containing | |
223 | * the on-disk version of the inode. It returns a pointer to the | |
224 | * buffer containing the on-disk inode in the bpp parameter, and in | |
225 | * the dip parameter it returns a pointer to the on-disk inode within | |
226 | * that buffer. | |
227 | * | |
228 | * If a non-zero error is returned, then the contents of bpp and | |
229 | * dipp are undefined. | |
230 | * | |
231 | * If the inode is new and has not yet been initialized, use xfs_imap() | |
232 | * to determine the size and location of the buffer to read from disk. | |
233 | * If the inode has already been mapped to its buffer and read in once, | |
234 | * then use the mapping information stored in the inode rather than | |
235 | * calling xfs_imap(). This allows us to avoid the overhead of looking | |
236 | * at the inode btree for small block file systems (see xfs_dilocate()). | |
237 | * We can tell whether the inode has been mapped in before by comparing | |
238 | * its disk block address to 0. Only uninitialized inodes will have | |
239 | * 0 for the disk block address. | |
240 | */ | |
241 | int | |
242 | xfs_itobp( | |
243 | xfs_mount_t *mp, | |
244 | xfs_trans_t *tp, | |
245 | xfs_inode_t *ip, | |
246 | xfs_dinode_t **dipp, | |
247 | xfs_buf_t **bpp, | |
b12dd342 NS |
248 | xfs_daddr_t bno, |
249 | uint imap_flags) | |
1da177e4 | 250 | { |
4d1a2ed3 | 251 | xfs_imap_t imap; |
1da177e4 LT |
252 | xfs_buf_t *bp; |
253 | int error; | |
1da177e4 LT |
254 | int i; |
255 | int ni; | |
1da177e4 LT |
256 | |
257 | if (ip->i_blkno == (xfs_daddr_t)0) { | |
258 | /* | |
259 | * Call the space management code to find the location of the | |
260 | * inode on disk. | |
261 | */ | |
262 | imap.im_blkno = bno; | |
b12dd342 NS |
263 | if ((error = xfs_imap(mp, tp, ip->i_ino, &imap, |
264 | XFS_IMAP_LOOKUP | imap_flags))) | |
1da177e4 | 265 | return error; |
1da177e4 LT |
266 | |
267 | /* | |
268 | * If the inode number maps to a block outside the bounds | |
269 | * of the file system then return NULL rather than calling | |
270 | * read_buf and panicing when we get an error from the | |
271 | * driver. | |
272 | */ | |
273 | if ((imap.im_blkno + imap.im_len) > | |
274 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) { | |
275 | #ifdef DEBUG | |
276 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: " | |
277 | "(imap.im_blkno (0x%llx) " | |
278 | "+ imap.im_len (0x%llx)) > " | |
279 | " XFS_FSB_TO_BB(mp, " | |
280 | "mp->m_sb.sb_dblocks) (0x%llx)", | |
281 | (unsigned long long) imap.im_blkno, | |
282 | (unsigned long long) imap.im_len, | |
283 | XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)); | |
284 | #endif /* DEBUG */ | |
285 | return XFS_ERROR(EINVAL); | |
286 | } | |
287 | ||
288 | /* | |
289 | * Fill in the fields in the inode that will be used to | |
290 | * map the inode to its buffer from now on. | |
291 | */ | |
292 | ip->i_blkno = imap.im_blkno; | |
293 | ip->i_len = imap.im_len; | |
294 | ip->i_boffset = imap.im_boffset; | |
295 | } else { | |
296 | /* | |
297 | * We've already mapped the inode once, so just use the | |
298 | * mapping that we saved the first time. | |
299 | */ | |
300 | imap.im_blkno = ip->i_blkno; | |
301 | imap.im_len = ip->i_len; | |
302 | imap.im_boffset = ip->i_boffset; | |
303 | } | |
304 | ASSERT(bno == 0 || bno == imap.im_blkno); | |
305 | ||
306 | /* | |
307 | * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will | |
308 | * default to just a read_buf() call. | |
309 | */ | |
310 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno, | |
311 | (int)imap.im_len, XFS_BUF_LOCK, &bp); | |
1da177e4 LT |
312 | if (error) { |
313 | #ifdef DEBUG | |
314 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: " | |
315 | "xfs_trans_read_buf() returned error %d, " | |
316 | "imap.im_blkno 0x%llx, imap.im_len 0x%llx", | |
317 | error, (unsigned long long) imap.im_blkno, | |
318 | (unsigned long long) imap.im_len); | |
319 | #endif /* DEBUG */ | |
320 | return error; | |
321 | } | |
4d1a2ed3 | 322 | |
1da177e4 LT |
323 | /* |
324 | * Validate the magic number and version of every inode in the buffer | |
325 | * (if DEBUG kernel) or the first inode in the buffer, otherwise. | |
4d1a2ed3 | 326 | * No validation is done here in userspace (xfs_repair). |
1da177e4 | 327 | */ |
4d1a2ed3 NS |
328 | #if !defined(__KERNEL__) |
329 | ni = 0; | |
330 | #elif defined(DEBUG) | |
41ff715a | 331 | ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog; |
4d1a2ed3 | 332 | #else /* usual case */ |
41ff715a | 333 | ni = 1; |
1da177e4 | 334 | #endif |
4d1a2ed3 | 335 | |
1da177e4 LT |
336 | for (i = 0; i < ni; i++) { |
337 | int di_ok; | |
338 | xfs_dinode_t *dip; | |
339 | ||
340 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, | |
341 | (i << mp->m_sb.sb_inodelog)); | |
347d1c01 CH |
342 | di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC && |
343 | XFS_DINODE_GOOD_VERSION(dip->di_core.di_version); | |
41ff715a NS |
344 | if (unlikely(XFS_TEST_ERROR(!di_ok, mp, |
345 | XFS_ERRTAG_ITOBP_INOTOBP, | |
346 | XFS_RANDOM_ITOBP_INOTOBP))) { | |
347 | if (imap_flags & XFS_IMAP_BULKSTAT) { | |
348 | xfs_trans_brelse(tp, bp); | |
349 | return XFS_ERROR(EINVAL); | |
350 | } | |
1da177e4 | 351 | #ifdef DEBUG |
41ff715a | 352 | cmn_err(CE_ALERT, |
4d1a2ed3 NS |
353 | "Device %s - bad inode magic/vsn " |
354 | "daddr %lld #%d (magic=%x)", | |
b6574520 | 355 | XFS_BUFTARG_NAME(mp->m_ddev_targp), |
1da177e4 | 356 | (unsigned long long)imap.im_blkno, i, |
347d1c01 | 357 | be16_to_cpu(dip->di_core.di_magic)); |
1da177e4 LT |
358 | #endif |
359 | XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH, | |
360 | mp, dip); | |
361 | xfs_trans_brelse(tp, bp); | |
362 | return XFS_ERROR(EFSCORRUPTED); | |
363 | } | |
364 | } | |
1da177e4 LT |
365 | |
366 | xfs_inobp_check(mp, bp); | |
367 | ||
368 | /* | |
369 | * Mark the buffer as an inode buffer now that it looks good | |
370 | */ | |
371 | XFS_BUF_SET_VTYPE(bp, B_FS_INO); | |
372 | ||
373 | /* | |
374 | * Set *dipp to point to the on-disk inode in the buffer. | |
375 | */ | |
376 | *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset); | |
377 | *bpp = bp; | |
378 | return 0; | |
379 | } | |
380 | ||
381 | /* | |
382 | * Move inode type and inode format specific information from the | |
383 | * on-disk inode to the in-core inode. For fifos, devs, and sockets | |
384 | * this means set if_rdev to the proper value. For files, directories, | |
385 | * and symlinks this means to bring in the in-line data or extent | |
386 | * pointers. For a file in B-tree format, only the root is immediately | |
387 | * brought in-core. The rest will be in-lined in if_extents when it | |
388 | * is first referenced (see xfs_iread_extents()). | |
389 | */ | |
390 | STATIC int | |
391 | xfs_iformat( | |
392 | xfs_inode_t *ip, | |
393 | xfs_dinode_t *dip) | |
394 | { | |
395 | xfs_attr_shortform_t *atp; | |
396 | int size; | |
397 | int error; | |
398 | xfs_fsize_t di_size; | |
399 | ip->i_df.if_ext_max = | |
400 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
401 | error = 0; | |
402 | ||
347d1c01 CH |
403 | if (unlikely(be32_to_cpu(dip->di_core.di_nextents) + |
404 | be16_to_cpu(dip->di_core.di_anextents) > | |
405 | be64_to_cpu(dip->di_core.di_nblocks))) { | |
3762ec6b NS |
406 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
407 | "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", | |
1da177e4 | 408 | (unsigned long long)ip->i_ino, |
347d1c01 CH |
409 | (int)(be32_to_cpu(dip->di_core.di_nextents) + |
410 | be16_to_cpu(dip->di_core.di_anextents)), | |
1da177e4 | 411 | (unsigned long long) |
347d1c01 | 412 | be64_to_cpu(dip->di_core.di_nblocks)); |
1da177e4 LT |
413 | XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, |
414 | ip->i_mount, dip); | |
415 | return XFS_ERROR(EFSCORRUPTED); | |
416 | } | |
417 | ||
347d1c01 | 418 | if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) { |
3762ec6b NS |
419 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
420 | "corrupt dinode %Lu, forkoff = 0x%x.", | |
1da177e4 | 421 | (unsigned long long)ip->i_ino, |
347d1c01 | 422 | dip->di_core.di_forkoff); |
1da177e4 LT |
423 | XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, |
424 | ip->i_mount, dip); | |
425 | return XFS_ERROR(EFSCORRUPTED); | |
426 | } | |
427 | ||
428 | switch (ip->i_d.di_mode & S_IFMT) { | |
429 | case S_IFIFO: | |
430 | case S_IFCHR: | |
431 | case S_IFBLK: | |
432 | case S_IFSOCK: | |
347d1c01 | 433 | if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) { |
1da177e4 LT |
434 | XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, |
435 | ip->i_mount, dip); | |
436 | return XFS_ERROR(EFSCORRUPTED); | |
437 | } | |
438 | ip->i_d.di_size = 0; | |
ba87ea69 | 439 | ip->i_size = 0; |
347d1c01 | 440 | ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev); |
1da177e4 LT |
441 | break; |
442 | ||
443 | case S_IFREG: | |
444 | case S_IFLNK: | |
445 | case S_IFDIR: | |
347d1c01 | 446 | switch (dip->di_core.di_format) { |
1da177e4 LT |
447 | case XFS_DINODE_FMT_LOCAL: |
448 | /* | |
449 | * no local regular files yet | |
450 | */ | |
347d1c01 | 451 | if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) { |
3762ec6b NS |
452 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
453 | "corrupt inode %Lu " | |
454 | "(local format for regular file).", | |
1da177e4 LT |
455 | (unsigned long long) ip->i_ino); |
456 | XFS_CORRUPTION_ERROR("xfs_iformat(4)", | |
457 | XFS_ERRLEVEL_LOW, | |
458 | ip->i_mount, dip); | |
459 | return XFS_ERROR(EFSCORRUPTED); | |
460 | } | |
461 | ||
347d1c01 | 462 | di_size = be64_to_cpu(dip->di_core.di_size); |
1da177e4 | 463 | if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { |
3762ec6b NS |
464 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
465 | "corrupt inode %Lu " | |
466 | "(bad size %Ld for local inode).", | |
1da177e4 LT |
467 | (unsigned long long) ip->i_ino, |
468 | (long long) di_size); | |
469 | XFS_CORRUPTION_ERROR("xfs_iformat(5)", | |
470 | XFS_ERRLEVEL_LOW, | |
471 | ip->i_mount, dip); | |
472 | return XFS_ERROR(EFSCORRUPTED); | |
473 | } | |
474 | ||
475 | size = (int)di_size; | |
476 | error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); | |
477 | break; | |
478 | case XFS_DINODE_FMT_EXTENTS: | |
479 | error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); | |
480 | break; | |
481 | case XFS_DINODE_FMT_BTREE: | |
482 | error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); | |
483 | break; | |
484 | default: | |
485 | XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, | |
486 | ip->i_mount); | |
487 | return XFS_ERROR(EFSCORRUPTED); | |
488 | } | |
489 | break; | |
490 | ||
491 | default: | |
492 | XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); | |
493 | return XFS_ERROR(EFSCORRUPTED); | |
494 | } | |
495 | if (error) { | |
496 | return error; | |
497 | } | |
498 | if (!XFS_DFORK_Q(dip)) | |
499 | return 0; | |
500 | ASSERT(ip->i_afp == NULL); | |
501 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
502 | ip->i_afp->if_ext_max = | |
503 | XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
347d1c01 | 504 | switch (dip->di_core.di_aformat) { |
1da177e4 LT |
505 | case XFS_DINODE_FMT_LOCAL: |
506 | atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); | |
3b244aa8 | 507 | size = be16_to_cpu(atp->hdr.totsize); |
1da177e4 LT |
508 | error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); |
509 | break; | |
510 | case XFS_DINODE_FMT_EXTENTS: | |
511 | error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); | |
512 | break; | |
513 | case XFS_DINODE_FMT_BTREE: | |
514 | error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); | |
515 | break; | |
516 | default: | |
517 | error = XFS_ERROR(EFSCORRUPTED); | |
518 | break; | |
519 | } | |
520 | if (error) { | |
521 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
522 | ip->i_afp = NULL; | |
523 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
524 | } | |
525 | return error; | |
526 | } | |
527 | ||
528 | /* | |
529 | * The file is in-lined in the on-disk inode. | |
530 | * If it fits into if_inline_data, then copy | |
531 | * it there, otherwise allocate a buffer for it | |
532 | * and copy the data there. Either way, set | |
533 | * if_data to point at the data. | |
534 | * If we allocate a buffer for the data, make | |
535 | * sure that its size is a multiple of 4 and | |
536 | * record the real size in i_real_bytes. | |
537 | */ | |
538 | STATIC int | |
539 | xfs_iformat_local( | |
540 | xfs_inode_t *ip, | |
541 | xfs_dinode_t *dip, | |
542 | int whichfork, | |
543 | int size) | |
544 | { | |
545 | xfs_ifork_t *ifp; | |
546 | int real_size; | |
547 | ||
548 | /* | |
549 | * If the size is unreasonable, then something | |
550 | * is wrong and we just bail out rather than crash in | |
551 | * kmem_alloc() or memcpy() below. | |
552 | */ | |
553 | if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
3762ec6b NS |
554 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
555 | "corrupt inode %Lu " | |
556 | "(bad size %d for local fork, size = %d).", | |
1da177e4 LT |
557 | (unsigned long long) ip->i_ino, size, |
558 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); | |
559 | XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, | |
560 | ip->i_mount, dip); | |
561 | return XFS_ERROR(EFSCORRUPTED); | |
562 | } | |
563 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
564 | real_size = 0; | |
565 | if (size == 0) | |
566 | ifp->if_u1.if_data = NULL; | |
567 | else if (size <= sizeof(ifp->if_u2.if_inline_data)) | |
568 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
569 | else { | |
570 | real_size = roundup(size, 4); | |
571 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
572 | } | |
573 | ifp->if_bytes = size; | |
574 | ifp->if_real_bytes = real_size; | |
575 | if (size) | |
576 | memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); | |
577 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
578 | ifp->if_flags |= XFS_IFINLINE; | |
579 | return 0; | |
580 | } | |
581 | ||
582 | /* | |
583 | * The file consists of a set of extents all | |
584 | * of which fit into the on-disk inode. | |
585 | * If there are few enough extents to fit into | |
586 | * the if_inline_ext, then copy them there. | |
587 | * Otherwise allocate a buffer for them and copy | |
588 | * them into it. Either way, set if_extents | |
589 | * to point at the extents. | |
590 | */ | |
591 | STATIC int | |
592 | xfs_iformat_extents( | |
593 | xfs_inode_t *ip, | |
594 | xfs_dinode_t *dip, | |
595 | int whichfork) | |
596 | { | |
a6f64d4a | 597 | xfs_bmbt_rec_t *dp; |
1da177e4 LT |
598 | xfs_ifork_t *ifp; |
599 | int nex; | |
1da177e4 LT |
600 | int size; |
601 | int i; | |
602 | ||
603 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
604 | nex = XFS_DFORK_NEXTENTS(dip, whichfork); | |
605 | size = nex * (uint)sizeof(xfs_bmbt_rec_t); | |
606 | ||
607 | /* | |
608 | * If the number of extents is unreasonable, then something | |
609 | * is wrong and we just bail out rather than crash in | |
610 | * kmem_alloc() or memcpy() below. | |
611 | */ | |
612 | if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { | |
3762ec6b NS |
613 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
614 | "corrupt inode %Lu ((a)extents = %d).", | |
1da177e4 LT |
615 | (unsigned long long) ip->i_ino, nex); |
616 | XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, | |
617 | ip->i_mount, dip); | |
618 | return XFS_ERROR(EFSCORRUPTED); | |
619 | } | |
620 | ||
4eea22f0 | 621 | ifp->if_real_bytes = 0; |
1da177e4 LT |
622 | if (nex == 0) |
623 | ifp->if_u1.if_extents = NULL; | |
624 | else if (nex <= XFS_INLINE_EXTS) | |
625 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
4eea22f0 MK |
626 | else |
627 | xfs_iext_add(ifp, 0, nex); | |
628 | ||
1da177e4 | 629 | ifp->if_bytes = size; |
1da177e4 LT |
630 | if (size) { |
631 | dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); | |
a6f64d4a | 632 | xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip)); |
4eea22f0 | 633 | for (i = 0; i < nex; i++, dp++) { |
a6f64d4a | 634 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
cd8b0a97 CH |
635 | ep->l0 = be64_to_cpu(get_unaligned(&dp->l0)); |
636 | ep->l1 = be64_to_cpu(get_unaligned(&dp->l1)); | |
1da177e4 | 637 | } |
3a59c94c | 638 | XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork); |
1da177e4 LT |
639 | if (whichfork != XFS_DATA_FORK || |
640 | XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) | |
641 | if (unlikely(xfs_check_nostate_extents( | |
4eea22f0 | 642 | ifp, 0, nex))) { |
1da177e4 LT |
643 | XFS_ERROR_REPORT("xfs_iformat_extents(2)", |
644 | XFS_ERRLEVEL_LOW, | |
645 | ip->i_mount); | |
646 | return XFS_ERROR(EFSCORRUPTED); | |
647 | } | |
648 | } | |
649 | ifp->if_flags |= XFS_IFEXTENTS; | |
650 | return 0; | |
651 | } | |
652 | ||
653 | /* | |
654 | * The file has too many extents to fit into | |
655 | * the inode, so they are in B-tree format. | |
656 | * Allocate a buffer for the root of the B-tree | |
657 | * and copy the root into it. The i_extents | |
658 | * field will remain NULL until all of the | |
659 | * extents are read in (when they are needed). | |
660 | */ | |
661 | STATIC int | |
662 | xfs_iformat_btree( | |
663 | xfs_inode_t *ip, | |
664 | xfs_dinode_t *dip, | |
665 | int whichfork) | |
666 | { | |
667 | xfs_bmdr_block_t *dfp; | |
668 | xfs_ifork_t *ifp; | |
669 | /* REFERENCED */ | |
670 | int nrecs; | |
671 | int size; | |
672 | ||
673 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
674 | dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); | |
675 | size = XFS_BMAP_BROOT_SPACE(dfp); | |
676 | nrecs = XFS_BMAP_BROOT_NUMRECS(dfp); | |
677 | ||
678 | /* | |
679 | * blow out if -- fork has less extents than can fit in | |
680 | * fork (fork shouldn't be a btree format), root btree | |
681 | * block has more records than can fit into the fork, | |
682 | * or the number of extents is greater than the number of | |
683 | * blocks. | |
684 | */ | |
685 | if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max | |
686 | || XFS_BMDR_SPACE_CALC(nrecs) > | |
687 | XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) | |
688 | || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { | |
3762ec6b NS |
689 | xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount, |
690 | "corrupt inode %Lu (btree).", | |
1da177e4 LT |
691 | (unsigned long long) ip->i_ino); |
692 | XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW, | |
693 | ip->i_mount); | |
694 | return XFS_ERROR(EFSCORRUPTED); | |
695 | } | |
696 | ||
697 | ifp->if_broot_bytes = size; | |
698 | ifp->if_broot = kmem_alloc(size, KM_SLEEP); | |
699 | ASSERT(ifp->if_broot != NULL); | |
700 | /* | |
701 | * Copy and convert from the on-disk structure | |
702 | * to the in-memory structure. | |
703 | */ | |
704 | xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), | |
705 | ifp->if_broot, size); | |
706 | ifp->if_flags &= ~XFS_IFEXTENTS; | |
707 | ifp->if_flags |= XFS_IFBROOT; | |
708 | ||
709 | return 0; | |
710 | } | |
711 | ||
1da177e4 | 712 | void |
347d1c01 CH |
713 | xfs_dinode_from_disk( |
714 | xfs_icdinode_t *to, | |
715 | xfs_dinode_core_t *from) | |
1da177e4 | 716 | { |
347d1c01 CH |
717 | to->di_magic = be16_to_cpu(from->di_magic); |
718 | to->di_mode = be16_to_cpu(from->di_mode); | |
719 | to->di_version = from ->di_version; | |
720 | to->di_format = from->di_format; | |
721 | to->di_onlink = be16_to_cpu(from->di_onlink); | |
722 | to->di_uid = be32_to_cpu(from->di_uid); | |
723 | to->di_gid = be32_to_cpu(from->di_gid); | |
724 | to->di_nlink = be32_to_cpu(from->di_nlink); | |
725 | to->di_projid = be16_to_cpu(from->di_projid); | |
726 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); | |
727 | to->di_flushiter = be16_to_cpu(from->di_flushiter); | |
728 | to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec); | |
729 | to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec); | |
730 | to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec); | |
731 | to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec); | |
732 | to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec); | |
733 | to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec); | |
734 | to->di_size = be64_to_cpu(from->di_size); | |
735 | to->di_nblocks = be64_to_cpu(from->di_nblocks); | |
736 | to->di_extsize = be32_to_cpu(from->di_extsize); | |
737 | to->di_nextents = be32_to_cpu(from->di_nextents); | |
738 | to->di_anextents = be16_to_cpu(from->di_anextents); | |
739 | to->di_forkoff = from->di_forkoff; | |
740 | to->di_aformat = from->di_aformat; | |
741 | to->di_dmevmask = be32_to_cpu(from->di_dmevmask); | |
742 | to->di_dmstate = be16_to_cpu(from->di_dmstate); | |
743 | to->di_flags = be16_to_cpu(from->di_flags); | |
744 | to->di_gen = be32_to_cpu(from->di_gen); | |
745 | } | |
746 | ||
747 | void | |
748 | xfs_dinode_to_disk( | |
749 | xfs_dinode_core_t *to, | |
750 | xfs_icdinode_t *from) | |
751 | { | |
752 | to->di_magic = cpu_to_be16(from->di_magic); | |
753 | to->di_mode = cpu_to_be16(from->di_mode); | |
754 | to->di_version = from ->di_version; | |
755 | to->di_format = from->di_format; | |
756 | to->di_onlink = cpu_to_be16(from->di_onlink); | |
757 | to->di_uid = cpu_to_be32(from->di_uid); | |
758 | to->di_gid = cpu_to_be32(from->di_gid); | |
759 | to->di_nlink = cpu_to_be32(from->di_nlink); | |
760 | to->di_projid = cpu_to_be16(from->di_projid); | |
761 | memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); | |
762 | to->di_flushiter = cpu_to_be16(from->di_flushiter); | |
763 | to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec); | |
764 | to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec); | |
765 | to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec); | |
766 | to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec); | |
767 | to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec); | |
768 | to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec); | |
769 | to->di_size = cpu_to_be64(from->di_size); | |
770 | to->di_nblocks = cpu_to_be64(from->di_nblocks); | |
771 | to->di_extsize = cpu_to_be32(from->di_extsize); | |
772 | to->di_nextents = cpu_to_be32(from->di_nextents); | |
773 | to->di_anextents = cpu_to_be16(from->di_anextents); | |
774 | to->di_forkoff = from->di_forkoff; | |
775 | to->di_aformat = from->di_aformat; | |
776 | to->di_dmevmask = cpu_to_be32(from->di_dmevmask); | |
777 | to->di_dmstate = cpu_to_be16(from->di_dmstate); | |
778 | to->di_flags = cpu_to_be16(from->di_flags); | |
779 | to->di_gen = cpu_to_be32(from->di_gen); | |
1da177e4 LT |
780 | } |
781 | ||
782 | STATIC uint | |
783 | _xfs_dic2xflags( | |
1da177e4 LT |
784 | __uint16_t di_flags) |
785 | { | |
786 | uint flags = 0; | |
787 | ||
788 | if (di_flags & XFS_DIFLAG_ANY) { | |
789 | if (di_flags & XFS_DIFLAG_REALTIME) | |
790 | flags |= XFS_XFLAG_REALTIME; | |
791 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
792 | flags |= XFS_XFLAG_PREALLOC; | |
793 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
794 | flags |= XFS_XFLAG_IMMUTABLE; | |
795 | if (di_flags & XFS_DIFLAG_APPEND) | |
796 | flags |= XFS_XFLAG_APPEND; | |
797 | if (di_flags & XFS_DIFLAG_SYNC) | |
798 | flags |= XFS_XFLAG_SYNC; | |
799 | if (di_flags & XFS_DIFLAG_NOATIME) | |
800 | flags |= XFS_XFLAG_NOATIME; | |
801 | if (di_flags & XFS_DIFLAG_NODUMP) | |
802 | flags |= XFS_XFLAG_NODUMP; | |
803 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
804 | flags |= XFS_XFLAG_RTINHERIT; | |
805 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
806 | flags |= XFS_XFLAG_PROJINHERIT; | |
807 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
808 | flags |= XFS_XFLAG_NOSYMLINKS; | |
dd9f438e NS |
809 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
810 | flags |= XFS_XFLAG_EXTSIZE; | |
811 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) | |
812 | flags |= XFS_XFLAG_EXTSZINHERIT; | |
d3446eac BN |
813 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
814 | flags |= XFS_XFLAG_NODEFRAG; | |
2a82b8be DC |
815 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
816 | flags |= XFS_XFLAG_FILESTREAM; | |
1da177e4 LT |
817 | } |
818 | ||
819 | return flags; | |
820 | } | |
821 | ||
822 | uint | |
823 | xfs_ip2xflags( | |
824 | xfs_inode_t *ip) | |
825 | { | |
347d1c01 | 826 | xfs_icdinode_t *dic = &ip->i_d; |
1da177e4 | 827 | |
a916e2bd NS |
828 | return _xfs_dic2xflags(dic->di_flags) | |
829 | (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0); | |
1da177e4 LT |
830 | } |
831 | ||
832 | uint | |
833 | xfs_dic2xflags( | |
834 | xfs_dinode_core_t *dic) | |
835 | { | |
347d1c01 | 836 | return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) | |
a916e2bd | 837 | (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
838 | } |
839 | ||
840 | /* | |
841 | * Given a mount structure and an inode number, return a pointer | |
c41564b5 | 842 | * to a newly allocated in-core inode corresponding to the given |
1da177e4 LT |
843 | * inode number. |
844 | * | |
845 | * Initialize the inode's attributes and extent pointers if it | |
846 | * already has them (it will not if the inode has no links). | |
847 | */ | |
848 | int | |
849 | xfs_iread( | |
850 | xfs_mount_t *mp, | |
851 | xfs_trans_t *tp, | |
852 | xfs_ino_t ino, | |
853 | xfs_inode_t **ipp, | |
745b1f47 NS |
854 | xfs_daddr_t bno, |
855 | uint imap_flags) | |
1da177e4 LT |
856 | { |
857 | xfs_buf_t *bp; | |
858 | xfs_dinode_t *dip; | |
859 | xfs_inode_t *ip; | |
860 | int error; | |
861 | ||
862 | ASSERT(xfs_inode_zone != NULL); | |
863 | ||
864 | ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP); | |
865 | ip->i_ino = ino; | |
866 | ip->i_mount = mp; | |
b677c210 | 867 | atomic_set(&ip->i_iocount, 0); |
f273ab84 | 868 | spin_lock_init(&ip->i_flags_lock); |
1da177e4 LT |
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 | */ | |
745b1f47 | 877 | error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags); |
b12dd342 | 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 | */ | |
cf441eeb LM |
887 | #ifdef XFS_INODE_TRACE |
888 | ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_SLEEP); | |
1543d79c | 889 | #endif |
1da177e4 LT |
890 | #ifdef XFS_BMAP_TRACE |
891 | ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP); | |
892 | #endif | |
893 | #ifdef XFS_BMBT_TRACE | |
894 | ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP); | |
895 | #endif | |
896 | #ifdef XFS_RW_TRACE | |
897 | ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP); | |
898 | #endif | |
899 | #ifdef XFS_ILOCK_TRACE | |
900 | ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP); | |
901 | #endif | |
902 | #ifdef XFS_DIR2_TRACE | |
903 | ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP); | |
904 | #endif | |
905 | ||
906 | /* | |
907 | * If we got something that isn't an inode it means someone | |
908 | * (nfs or dmi) has a stale handle. | |
909 | */ | |
347d1c01 | 910 | if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) { |
1da177e4 LT |
911 | kmem_zone_free(xfs_inode_zone, ip); |
912 | xfs_trans_brelse(tp, bp); | |
913 | #ifdef DEBUG | |
914 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " | |
915 | "dip->di_core.di_magic (0x%x) != " | |
916 | "XFS_DINODE_MAGIC (0x%x)", | |
347d1c01 | 917 | be16_to_cpu(dip->di_core.di_magic), |
1da177e4 LT |
918 | XFS_DINODE_MAGIC); |
919 | #endif /* DEBUG */ | |
920 | return XFS_ERROR(EINVAL); | |
921 | } | |
922 | ||
923 | /* | |
924 | * If the on-disk inode is already linked to a directory | |
925 | * entry, copy all of the inode into the in-core inode. | |
926 | * xfs_iformat() handles copying in the inode format | |
927 | * specific information. | |
928 | * Otherwise, just get the truly permanent information. | |
929 | */ | |
930 | if (dip->di_core.di_mode) { | |
347d1c01 | 931 | xfs_dinode_from_disk(&ip->i_d, &dip->di_core); |
1da177e4 LT |
932 | error = xfs_iformat(ip, dip); |
933 | if (error) { | |
934 | kmem_zone_free(xfs_inode_zone, ip); | |
935 | xfs_trans_brelse(tp, bp); | |
936 | #ifdef DEBUG | |
937 | xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: " | |
938 | "xfs_iformat() returned error %d", | |
939 | error); | |
940 | #endif /* DEBUG */ | |
941 | return error; | |
942 | } | |
943 | } else { | |
347d1c01 CH |
944 | ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic); |
945 | ip->i_d.di_version = dip->di_core.di_version; | |
946 | ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen); | |
947 | ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter); | |
1da177e4 LT |
948 | /* |
949 | * Make sure to pull in the mode here as well in | |
950 | * case the inode is released without being used. | |
951 | * This ensures that xfs_inactive() will see that | |
952 | * the inode is already free and not try to mess | |
953 | * with the uninitialized part of it. | |
954 | */ | |
955 | ip->i_d.di_mode = 0; | |
956 | /* | |
957 | * Initialize the per-fork minima and maxima for a new | |
958 | * inode here. xfs_iformat will do it for old inodes. | |
959 | */ | |
960 | ip->i_df.if_ext_max = | |
961 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
962 | } | |
963 | ||
964 | INIT_LIST_HEAD(&ip->i_reclaim); | |
965 | ||
966 | /* | |
967 | * The inode format changed when we moved the link count and | |
968 | * made it 32 bits long. If this is an old format inode, | |
969 | * convert it in memory to look like a new one. If it gets | |
970 | * flushed to disk we will convert back before flushing or | |
971 | * logging it. We zero out the new projid field and the old link | |
972 | * count field. We'll handle clearing the pad field (the remains | |
973 | * of the old uuid field) when we actually convert the inode to | |
974 | * the new format. We don't change the version number so that we | |
975 | * can distinguish this from a real new format inode. | |
976 | */ | |
977 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
978 | ip->i_d.di_nlink = ip->i_d.di_onlink; | |
979 | ip->i_d.di_onlink = 0; | |
980 | ip->i_d.di_projid = 0; | |
981 | } | |
982 | ||
983 | ip->i_delayed_blks = 0; | |
ba87ea69 | 984 | ip->i_size = ip->i_d.di_size; |
1da177e4 LT |
985 | |
986 | /* | |
987 | * Mark the buffer containing the inode as something to keep | |
988 | * around for a while. This helps to keep recently accessed | |
989 | * meta-data in-core longer. | |
990 | */ | |
991 | XFS_BUF_SET_REF(bp, XFS_INO_REF); | |
992 | ||
993 | /* | |
994 | * Use xfs_trans_brelse() to release the buffer containing the | |
995 | * on-disk inode, because it was acquired with xfs_trans_read_buf() | |
996 | * in xfs_itobp() above. If tp is NULL, this is just a normal | |
997 | * brelse(). If we're within a transaction, then xfs_trans_brelse() | |
998 | * will only release the buffer if it is not dirty within the | |
999 | * transaction. It will be OK to release the buffer in this case, | |
1000 | * because inodes on disk are never destroyed and we will be | |
1001 | * locking the new in-core inode before putting it in the hash | |
1002 | * table where other processes can find it. Thus we don't have | |
1003 | * to worry about the inode being changed just because we released | |
1004 | * the buffer. | |
1005 | */ | |
1006 | xfs_trans_brelse(tp, bp); | |
1007 | *ipp = ip; | |
1008 | return 0; | |
1009 | } | |
1010 | ||
1011 | /* | |
1012 | * Read in extents from a btree-format inode. | |
1013 | * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. | |
1014 | */ | |
1015 | int | |
1016 | xfs_iread_extents( | |
1017 | xfs_trans_t *tp, | |
1018 | xfs_inode_t *ip, | |
1019 | int whichfork) | |
1020 | { | |
1021 | int error; | |
1022 | xfs_ifork_t *ifp; | |
4eea22f0 | 1023 | xfs_extnum_t nextents; |
1da177e4 LT |
1024 | size_t size; |
1025 | ||
1026 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1027 | XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, | |
1028 | ip->i_mount); | |
1029 | return XFS_ERROR(EFSCORRUPTED); | |
1030 | } | |
4eea22f0 MK |
1031 | nextents = XFS_IFORK_NEXTENTS(ip, whichfork); |
1032 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
1da177e4 | 1033 | ifp = XFS_IFORK_PTR(ip, whichfork); |
4eea22f0 | 1034 | |
1da177e4 LT |
1035 | /* |
1036 | * We know that the size is valid (it's checked in iformat_btree) | |
1037 | */ | |
1da177e4 | 1038 | ifp->if_lastex = NULLEXTNUM; |
4eea22f0 | 1039 | ifp->if_bytes = ifp->if_real_bytes = 0; |
1da177e4 | 1040 | ifp->if_flags |= XFS_IFEXTENTS; |
4eea22f0 | 1041 | xfs_iext_add(ifp, 0, nextents); |
1da177e4 LT |
1042 | error = xfs_bmap_read_extents(tp, ip, whichfork); |
1043 | if (error) { | |
4eea22f0 | 1044 | xfs_iext_destroy(ifp); |
1da177e4 LT |
1045 | ifp->if_flags &= ~XFS_IFEXTENTS; |
1046 | return error; | |
1047 | } | |
a6f64d4a | 1048 | xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
1049 | return 0; |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * Allocate an inode on disk and return a copy of its in-core version. | |
1054 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
1055 | * appropriately within the inode. The uid and gid for the inode are | |
1056 | * set according to the contents of the given cred structure. | |
1057 | * | |
1058 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
1059 | * has a free inode available, call xfs_iget() | |
1060 | * to obtain the in-core version of the allocated inode. Finally, | |
1061 | * fill in the inode and log its initial contents. In this case, | |
1062 | * ialloc_context would be set to NULL and call_again set to false. | |
1063 | * | |
1064 | * If xfs_dialloc() does not have an available inode, | |
1065 | * it will replenish its supply by doing an allocation. Since we can | |
1066 | * only do one allocation within a transaction without deadlocks, we | |
1067 | * must commit the current transaction before returning the inode itself. | |
1068 | * In this case, therefore, we will set call_again to true and return. | |
1069 | * The caller should then commit the current transaction, start a new | |
1070 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
1071 | * | |
1072 | * To ensure that some other process does not grab the inode that | |
1073 | * was allocated during the first call to xfs_ialloc(), this routine | |
1074 | * also returns the [locked] bp pointing to the head of the freelist | |
1075 | * as ialloc_context. The caller should hold this buffer across | |
1076 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
1077 | * |
1078 | * If we are allocating quota inodes, we do not have a parent inode | |
1079 | * to attach to or associate with (i.e. pip == NULL) because they | |
1080 | * are not linked into the directory structure - they are attached | |
1081 | * directly to the superblock - and so have no parent. | |
1da177e4 LT |
1082 | */ |
1083 | int | |
1084 | xfs_ialloc( | |
1085 | xfs_trans_t *tp, | |
1086 | xfs_inode_t *pip, | |
1087 | mode_t mode, | |
31b084ae | 1088 | xfs_nlink_t nlink, |
1da177e4 LT |
1089 | xfs_dev_t rdev, |
1090 | cred_t *cr, | |
1091 | xfs_prid_t prid, | |
1092 | int okalloc, | |
1093 | xfs_buf_t **ialloc_context, | |
1094 | boolean_t *call_again, | |
1095 | xfs_inode_t **ipp) | |
1096 | { | |
1097 | xfs_ino_t ino; | |
1098 | xfs_inode_t *ip; | |
67fcaa73 | 1099 | bhv_vnode_t *vp; |
1da177e4 LT |
1100 | uint flags; |
1101 | int error; | |
1102 | ||
1103 | /* | |
1104 | * Call the space management code to pick | |
1105 | * the on-disk inode to be allocated. | |
1106 | */ | |
b11f94d5 | 1107 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
1da177e4 LT |
1108 | ialloc_context, call_again, &ino); |
1109 | if (error != 0) { | |
1110 | return error; | |
1111 | } | |
1112 | if (*call_again || ino == NULLFSINO) { | |
1113 | *ipp = NULL; | |
1114 | return 0; | |
1115 | } | |
1116 | ASSERT(*ialloc_context == NULL); | |
1117 | ||
1118 | /* | |
1119 | * Get the in-core inode with the lock held exclusively. | |
1120 | * This is because we're setting fields here we need | |
1121 | * to prevent others from looking at until we're done. | |
1122 | */ | |
1123 | error = xfs_trans_iget(tp->t_mountp, tp, ino, | |
745b1f47 | 1124 | XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip); |
1da177e4 LT |
1125 | if (error != 0) { |
1126 | return error; | |
1127 | } | |
1128 | ASSERT(ip != NULL); | |
1129 | ||
1130 | vp = XFS_ITOV(ip); | |
1da177e4 LT |
1131 | ip->i_d.di_mode = (__uint16_t)mode; |
1132 | ip->i_d.di_onlink = 0; | |
1133 | ip->i_d.di_nlink = nlink; | |
1134 | ASSERT(ip->i_d.di_nlink == nlink); | |
1135 | ip->i_d.di_uid = current_fsuid(cr); | |
1136 | ip->i_d.di_gid = current_fsgid(cr); | |
1137 | ip->i_d.di_projid = prid; | |
1138 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
1139 | ||
1140 | /* | |
1141 | * If the superblock version is up to where we support new format | |
1142 | * inodes and this is currently an old format inode, then change | |
1143 | * the inode version number now. This way we only do the conversion | |
1144 | * here rather than here and in the flush/logging code. | |
1145 | */ | |
1146 | if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) && | |
1147 | ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
1148 | ip->i_d.di_version = XFS_DINODE_VERSION_2; | |
1149 | /* | |
1150 | * We've already zeroed the old link count, the projid field, | |
1151 | * and the pad field. | |
1152 | */ | |
1153 | } | |
1154 | ||
1155 | /* | |
1156 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
1157 | */ | |
2a82b8be | 1158 | if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1)) |
1da177e4 LT |
1159 | xfs_bump_ino_vers2(tp, ip); |
1160 | ||
bd186aa9 | 1161 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 LT |
1162 | ip->i_d.di_gid = pip->i_d.di_gid; |
1163 | if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) { | |
1164 | ip->i_d.di_mode |= S_ISGID; | |
1165 | } | |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * If the group ID of the new file does not match the effective group | |
1170 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
1171 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
1172 | */ | |
1173 | if ((irix_sgid_inherit) && | |
1174 | (ip->i_d.di_mode & S_ISGID) && | |
1175 | (!in_group_p((gid_t)ip->i_d.di_gid))) { | |
1176 | ip->i_d.di_mode &= ~S_ISGID; | |
1177 | } | |
1178 | ||
1179 | ip->i_d.di_size = 0; | |
ba87ea69 | 1180 | ip->i_size = 0; |
1da177e4 LT |
1181 | ip->i_d.di_nextents = 0; |
1182 | ASSERT(ip->i_d.di_nblocks == 0); | |
1183 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD); | |
1184 | /* | |
1185 | * di_gen will have been taken care of in xfs_iread. | |
1186 | */ | |
1187 | ip->i_d.di_extsize = 0; | |
1188 | ip->i_d.di_dmevmask = 0; | |
1189 | ip->i_d.di_dmstate = 0; | |
1190 | ip->i_d.di_flags = 0; | |
1191 | flags = XFS_ILOG_CORE; | |
1192 | switch (mode & S_IFMT) { | |
1193 | case S_IFIFO: | |
1194 | case S_IFCHR: | |
1195 | case S_IFBLK: | |
1196 | case S_IFSOCK: | |
1197 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
1198 | ip->i_df.if_u2.if_rdev = rdev; | |
1199 | ip->i_df.if_flags = 0; | |
1200 | flags |= XFS_ILOG_DEV; | |
1201 | break; | |
1202 | case S_IFREG: | |
b11f94d5 | 1203 | if (pip && xfs_inode_is_filestream(pip)) { |
2a82b8be DC |
1204 | error = xfs_filestream_associate(pip, ip); |
1205 | if (error < 0) | |
1206 | return -error; | |
1207 | if (!error) | |
1208 | xfs_iflags_set(ip, XFS_IFILESTREAM); | |
1209 | } | |
1210 | /* fall through */ | |
1da177e4 | 1211 | case S_IFDIR: |
b11f94d5 | 1212 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
365ca83d NS |
1213 | uint di_flags = 0; |
1214 | ||
1215 | if ((mode & S_IFMT) == S_IFDIR) { | |
1216 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) | |
1217 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
1218 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1219 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
1220 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1221 | } | |
1222 | } else if ((mode & S_IFMT) == S_IFREG) { | |
365ca83d NS |
1223 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) { |
1224 | di_flags |= XFS_DIFLAG_REALTIME; | |
1da177e4 LT |
1225 | ip->i_iocore.io_flags |= XFS_IOCORE_RT; |
1226 | } | |
dd9f438e NS |
1227 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
1228 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
1229 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
1230 | } | |
1da177e4 LT |
1231 | } |
1232 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
1233 | xfs_inherit_noatime) | |
365ca83d | 1234 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
1235 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
1236 | xfs_inherit_nodump) | |
365ca83d | 1237 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
1238 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
1239 | xfs_inherit_sync) | |
365ca83d | 1240 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
1241 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
1242 | xfs_inherit_nosymlinks) | |
365ca83d NS |
1243 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
1244 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
1245 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
1246 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
1247 | xfs_inherit_nodefrag) | |
1248 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
1249 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
1250 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
365ca83d | 1251 | ip->i_d.di_flags |= di_flags; |
1da177e4 LT |
1252 | } |
1253 | /* FALLTHROUGH */ | |
1254 | case S_IFLNK: | |
1255 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
1256 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
1257 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
1258 | ip->i_df.if_u1.if_extents = NULL; | |
1259 | break; | |
1260 | default: | |
1261 | ASSERT(0); | |
1262 | } | |
1263 | /* | |
1264 | * Attribute fork settings for new inode. | |
1265 | */ | |
1266 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1267 | ip->i_d.di_anextents = 0; | |
1268 | ||
1269 | /* | |
1270 | * Log the new values stuffed into the inode. | |
1271 | */ | |
1272 | xfs_trans_log_inode(tp, ip, flags); | |
1273 | ||
b83bd138 | 1274 | /* now that we have an i_mode we can setup inode ops and unlock */ |
745f6919 | 1275 | xfs_initialize_vnode(tp->t_mountp, vp, ip); |
1da177e4 LT |
1276 | |
1277 | *ipp = ip; | |
1278 | return 0; | |
1279 | } | |
1280 | ||
1281 | /* | |
1282 | * Check to make sure that there are no blocks allocated to the | |
1283 | * file beyond the size of the file. We don't check this for | |
1284 | * files with fixed size extents or real time extents, but we | |
1285 | * at least do it for regular files. | |
1286 | */ | |
1287 | #ifdef DEBUG | |
1288 | void | |
1289 | xfs_isize_check( | |
1290 | xfs_mount_t *mp, | |
1291 | xfs_inode_t *ip, | |
1292 | xfs_fsize_t isize) | |
1293 | { | |
1294 | xfs_fileoff_t map_first; | |
1295 | int nimaps; | |
1296 | xfs_bmbt_irec_t imaps[2]; | |
1297 | ||
1298 | if ((ip->i_d.di_mode & S_IFMT) != S_IFREG) | |
1299 | return; | |
1300 | ||
dd9f438e | 1301 | if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE)) |
1da177e4 LT |
1302 | return; |
1303 | ||
1304 | nimaps = 2; | |
1305 | map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
1306 | /* | |
1307 | * The filesystem could be shutting down, so bmapi may return | |
1308 | * an error. | |
1309 | */ | |
1310 | if (xfs_bmapi(NULL, ip, map_first, | |
1311 | (XFS_B_TO_FSB(mp, | |
1312 | (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) - | |
1313 | map_first), | |
1314 | XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps, | |
3e57ecf6 | 1315 | NULL, NULL)) |
1da177e4 LT |
1316 | return; |
1317 | ASSERT(nimaps == 1); | |
1318 | ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK); | |
1319 | } | |
1320 | #endif /* DEBUG */ | |
1321 | ||
1322 | /* | |
1323 | * Calculate the last possible buffered byte in a file. This must | |
1324 | * include data that was buffered beyond the EOF by the write code. | |
1325 | * This also needs to deal with overflowing the xfs_fsize_t type | |
1326 | * which can happen for sizes near the limit. | |
1327 | * | |
1328 | * We also need to take into account any blocks beyond the EOF. It | |
1329 | * may be the case that they were buffered by a write which failed. | |
1330 | * In that case the pages will still be in memory, but the inode size | |
1331 | * will never have been updated. | |
1332 | */ | |
1333 | xfs_fsize_t | |
1334 | xfs_file_last_byte( | |
1335 | xfs_inode_t *ip) | |
1336 | { | |
1337 | xfs_mount_t *mp; | |
1338 | xfs_fsize_t last_byte; | |
1339 | xfs_fileoff_t last_block; | |
1340 | xfs_fileoff_t size_last_block; | |
1341 | int error; | |
1342 | ||
1343 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS)); | |
1344 | ||
1345 | mp = ip->i_mount; | |
1346 | /* | |
1347 | * Only check for blocks beyond the EOF if the extents have | |
1348 | * been read in. This eliminates the need for the inode lock, | |
1349 | * and it also saves us from looking when it really isn't | |
1350 | * necessary. | |
1351 | */ | |
1352 | if (ip->i_df.if_flags & XFS_IFEXTENTS) { | |
1353 | error = xfs_bmap_last_offset(NULL, ip, &last_block, | |
1354 | XFS_DATA_FORK); | |
1355 | if (error) { | |
1356 | last_block = 0; | |
1357 | } | |
1358 | } else { | |
1359 | last_block = 0; | |
1360 | } | |
ba87ea69 | 1361 | size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size); |
1da177e4 LT |
1362 | last_block = XFS_FILEOFF_MAX(last_block, size_last_block); |
1363 | ||
1364 | last_byte = XFS_FSB_TO_B(mp, last_block); | |
1365 | if (last_byte < 0) { | |
1366 | return XFS_MAXIOFFSET(mp); | |
1367 | } | |
1368 | last_byte += (1 << mp->m_writeio_log); | |
1369 | if (last_byte < 0) { | |
1370 | return XFS_MAXIOFFSET(mp); | |
1371 | } | |
1372 | return last_byte; | |
1373 | } | |
1374 | ||
1375 | #if defined(XFS_RW_TRACE) | |
1376 | STATIC void | |
1377 | xfs_itrunc_trace( | |
1378 | int tag, | |
1379 | xfs_inode_t *ip, | |
1380 | int flag, | |
1381 | xfs_fsize_t new_size, | |
1382 | xfs_off_t toss_start, | |
1383 | xfs_off_t toss_finish) | |
1384 | { | |
1385 | if (ip->i_rwtrace == NULL) { | |
1386 | return; | |
1387 | } | |
1388 | ||
1389 | ktrace_enter(ip->i_rwtrace, | |
1390 | (void*)((long)tag), | |
1391 | (void*)ip, | |
1392 | (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff), | |
1393 | (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff), | |
1394 | (void*)((long)flag), | |
1395 | (void*)(unsigned long)((new_size >> 32) & 0xffffffff), | |
1396 | (void*)(unsigned long)(new_size & 0xffffffff), | |
1397 | (void*)(unsigned long)((toss_start >> 32) & 0xffffffff), | |
1398 | (void*)(unsigned long)(toss_start & 0xffffffff), | |
1399 | (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff), | |
1400 | (void*)(unsigned long)(toss_finish & 0xffffffff), | |
1401 | (void*)(unsigned long)current_cpu(), | |
f1fdc848 YL |
1402 | (void*)(unsigned long)current_pid(), |
1403 | (void*)NULL, | |
1404 | (void*)NULL, | |
1405 | (void*)NULL); | |
1da177e4 LT |
1406 | } |
1407 | #else | |
1408 | #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish) | |
1409 | #endif | |
1410 | ||
1411 | /* | |
1412 | * Start the truncation of the file to new_size. The new size | |
1413 | * must be smaller than the current size. This routine will | |
1414 | * clear the buffer and page caches of file data in the removed | |
1415 | * range, and xfs_itruncate_finish() will remove the underlying | |
1416 | * disk blocks. | |
1417 | * | |
1418 | * The inode must have its I/O lock locked EXCLUSIVELY, and it | |
1419 | * must NOT have the inode lock held at all. This is because we're | |
1420 | * calling into the buffer/page cache code and we can't hold the | |
1421 | * inode lock when we do so. | |
1422 | * | |
38e2299a DC |
1423 | * We need to wait for any direct I/Os in flight to complete before we |
1424 | * proceed with the truncate. This is needed to prevent the extents | |
1425 | * being read or written by the direct I/Os from being removed while the | |
1426 | * I/O is in flight as there is no other method of synchronising | |
1427 | * direct I/O with the truncate operation. Also, because we hold | |
1428 | * the IOLOCK in exclusive mode, we prevent new direct I/Os from being | |
1429 | * started until the truncate completes and drops the lock. Essentially, | |
1430 | * the vn_iowait() call forms an I/O barrier that provides strict ordering | |
1431 | * between direct I/Os and the truncate operation. | |
1432 | * | |
1da177e4 LT |
1433 | * The flags parameter can have either the value XFS_ITRUNC_DEFINITE |
1434 | * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used | |
1435 | * in the case that the caller is locking things out of order and | |
1436 | * may not be able to call xfs_itruncate_finish() with the inode lock | |
1437 | * held without dropping the I/O lock. If the caller must drop the | |
1438 | * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start() | |
1439 | * must be called again with all the same restrictions as the initial | |
1440 | * call. | |
1441 | */ | |
d3cf2094 | 1442 | int |
1da177e4 LT |
1443 | xfs_itruncate_start( |
1444 | xfs_inode_t *ip, | |
1445 | uint flags, | |
1446 | xfs_fsize_t new_size) | |
1447 | { | |
1448 | xfs_fsize_t last_byte; | |
1449 | xfs_off_t toss_start; | |
1450 | xfs_mount_t *mp; | |
67fcaa73 | 1451 | bhv_vnode_t *vp; |
d3cf2094 | 1452 | int error = 0; |
1da177e4 LT |
1453 | |
1454 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0); | |
ba87ea69 | 1455 | ASSERT((new_size == 0) || (new_size <= ip->i_size)); |
1da177e4 LT |
1456 | ASSERT((flags == XFS_ITRUNC_DEFINITE) || |
1457 | (flags == XFS_ITRUNC_MAYBE)); | |
1458 | ||
1459 | mp = ip->i_mount; | |
1460 | vp = XFS_ITOV(ip); | |
9fa8046f | 1461 | |
c734c79b LM |
1462 | /* wait for the completion of any pending DIOs */ |
1463 | if (new_size < ip->i_size) | |
1464 | vn_iowait(ip); | |
1465 | ||
1da177e4 | 1466 | /* |
67fcaa73 | 1467 | * Call toss_pages or flushinval_pages to get rid of pages |
1da177e4 | 1468 | * overlapping the region being removed. We have to use |
67fcaa73 | 1469 | * the less efficient flushinval_pages in the case that the |
1da177e4 LT |
1470 | * caller may not be able to finish the truncate without |
1471 | * dropping the inode's I/O lock. Make sure | |
1472 | * to catch any pages brought in by buffers overlapping | |
1473 | * the EOF by searching out beyond the isize by our | |
1474 | * block size. We round new_size up to a block boundary | |
1475 | * so that we don't toss things on the same block as | |
1476 | * new_size but before it. | |
1477 | * | |
67fcaa73 | 1478 | * Before calling toss_page or flushinval_pages, make sure to |
1da177e4 LT |
1479 | * call remapf() over the same region if the file is mapped. |
1480 | * This frees up mapped file references to the pages in the | |
67fcaa73 | 1481 | * given range and for the flushinval_pages case it ensures |
1da177e4 LT |
1482 | * that we get the latest mapped changes flushed out. |
1483 | */ | |
1484 | toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); | |
1485 | toss_start = XFS_FSB_TO_B(mp, toss_start); | |
1486 | if (toss_start < 0) { | |
1487 | /* | |
1488 | * The place to start tossing is beyond our maximum | |
1489 | * file size, so there is no way that the data extended | |
1490 | * out there. | |
1491 | */ | |
d3cf2094 | 1492 | return 0; |
1da177e4 LT |
1493 | } |
1494 | last_byte = xfs_file_last_byte(ip); | |
1495 | xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start, | |
1496 | last_byte); | |
1497 | if (last_byte > toss_start) { | |
1498 | if (flags & XFS_ITRUNC_DEFINITE) { | |
739bfb2a CH |
1499 | xfs_tosspages(ip, toss_start, |
1500 | -1, FI_REMAPF_LOCKED); | |
1da177e4 | 1501 | } else { |
739bfb2a CH |
1502 | error = xfs_flushinval_pages(ip, toss_start, |
1503 | -1, FI_REMAPF_LOCKED); | |
1da177e4 LT |
1504 | } |
1505 | } | |
1506 | ||
1507 | #ifdef DEBUG | |
1508 | if (new_size == 0) { | |
1509 | ASSERT(VN_CACHED(vp) == 0); | |
1510 | } | |
1511 | #endif | |
d3cf2094 | 1512 | return error; |
1da177e4 LT |
1513 | } |
1514 | ||
1515 | /* | |
1516 | * Shrink the file to the given new_size. The new | |
1517 | * size must be smaller than the current size. | |
1518 | * This will free up the underlying blocks | |
1519 | * in the removed range after a call to xfs_itruncate_start() | |
1520 | * or xfs_atruncate_start(). | |
1521 | * | |
1522 | * The transaction passed to this routine must have made | |
1523 | * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES. | |
1524 | * This routine may commit the given transaction and | |
1525 | * start new ones, so make sure everything involved in | |
1526 | * the transaction is tidy before calling here. | |
1527 | * Some transaction will be returned to the caller to be | |
1528 | * committed. The incoming transaction must already include | |
1529 | * the inode, and both inode locks must be held exclusively. | |
1530 | * The inode must also be "held" within the transaction. On | |
1531 | * return the inode will be "held" within the returned transaction. | |
1532 | * This routine does NOT require any disk space to be reserved | |
1533 | * for it within the transaction. | |
1534 | * | |
1535 | * The fork parameter must be either xfs_attr_fork or xfs_data_fork, | |
1536 | * and it indicates the fork which is to be truncated. For the | |
1537 | * attribute fork we only support truncation to size 0. | |
1538 | * | |
1539 | * We use the sync parameter to indicate whether or not the first | |
1540 | * transaction we perform might have to be synchronous. For the attr fork, | |
1541 | * it needs to be so if the unlink of the inode is not yet known to be | |
1542 | * permanent in the log. This keeps us from freeing and reusing the | |
1543 | * blocks of the attribute fork before the unlink of the inode becomes | |
1544 | * permanent. | |
1545 | * | |
1546 | * For the data fork, we normally have to run synchronously if we're | |
1547 | * being called out of the inactive path or we're being called | |
1548 | * out of the create path where we're truncating an existing file. | |
1549 | * Either way, the truncate needs to be sync so blocks don't reappear | |
1550 | * in the file with altered data in case of a crash. wsync filesystems | |
1551 | * can run the first case async because anything that shrinks the inode | |
1552 | * has to run sync so by the time we're called here from inactive, the | |
1553 | * inode size is permanently set to 0. | |
1554 | * | |
1555 | * Calls from the truncate path always need to be sync unless we're | |
1556 | * in a wsync filesystem and the file has already been unlinked. | |
1557 | * | |
1558 | * The caller is responsible for correctly setting the sync parameter. | |
1559 | * It gets too hard for us to guess here which path we're being called | |
1560 | * out of just based on inode state. | |
1561 | */ | |
1562 | int | |
1563 | xfs_itruncate_finish( | |
1564 | xfs_trans_t **tp, | |
1565 | xfs_inode_t *ip, | |
1566 | xfs_fsize_t new_size, | |
1567 | int fork, | |
1568 | int sync) | |
1569 | { | |
1570 | xfs_fsblock_t first_block; | |
1571 | xfs_fileoff_t first_unmap_block; | |
1572 | xfs_fileoff_t last_block; | |
1573 | xfs_filblks_t unmap_len=0; | |
1574 | xfs_mount_t *mp; | |
1575 | xfs_trans_t *ntp; | |
1576 | int done; | |
1577 | int committed; | |
1578 | xfs_bmap_free_t free_list; | |
1579 | int error; | |
1580 | ||
1581 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0); | |
1582 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0); | |
ba87ea69 | 1583 | ASSERT((new_size == 0) || (new_size <= ip->i_size)); |
1da177e4 LT |
1584 | ASSERT(*tp != NULL); |
1585 | ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES); | |
1586 | ASSERT(ip->i_transp == *tp); | |
1587 | ASSERT(ip->i_itemp != NULL); | |
1588 | ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD); | |
1589 | ||
1590 | ||
1591 | ntp = *tp; | |
1592 | mp = (ntp)->t_mountp; | |
1593 | ASSERT(! XFS_NOT_DQATTACHED(mp, ip)); | |
1594 | ||
1595 | /* | |
1596 | * We only support truncating the entire attribute fork. | |
1597 | */ | |
1598 | if (fork == XFS_ATTR_FORK) { | |
1599 | new_size = 0LL; | |
1600 | } | |
1601 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); | |
1602 | xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0); | |
1603 | /* | |
1604 | * The first thing we do is set the size to new_size permanently | |
1605 | * on disk. This way we don't have to worry about anyone ever | |
1606 | * being able to look at the data being freed even in the face | |
1607 | * of a crash. What we're getting around here is the case where | |
1608 | * we free a block, it is allocated to another file, it is written | |
1609 | * to, and then we crash. If the new data gets written to the | |
1610 | * file but the log buffers containing the free and reallocation | |
1611 | * don't, then we'd end up with garbage in the blocks being freed. | |
1612 | * As long as we make the new_size permanent before actually | |
1613 | * freeing any blocks it doesn't matter if they get writtten to. | |
1614 | * | |
1615 | * The callers must signal into us whether or not the size | |
1616 | * setting here must be synchronous. There are a few cases | |
1617 | * where it doesn't have to be synchronous. Those cases | |
1618 | * occur if the file is unlinked and we know the unlink is | |
1619 | * permanent or if the blocks being truncated are guaranteed | |
1620 | * to be beyond the inode eof (regardless of the link count) | |
1621 | * and the eof value is permanent. Both of these cases occur | |
1622 | * only on wsync-mounted filesystems. In those cases, we're | |
1623 | * guaranteed that no user will ever see the data in the blocks | |
1624 | * that are being truncated so the truncate can run async. | |
1625 | * In the free beyond eof case, the file may wind up with | |
1626 | * more blocks allocated to it than it needs if we crash | |
1627 | * and that won't get fixed until the next time the file | |
1628 | * is re-opened and closed but that's ok as that shouldn't | |
1629 | * be too many blocks. | |
1630 | * | |
1631 | * However, we can't just make all wsync xactions run async | |
1632 | * because there's one call out of the create path that needs | |
1633 | * to run sync where it's truncating an existing file to size | |
1634 | * 0 whose size is > 0. | |
1635 | * | |
1636 | * It's probably possible to come up with a test in this | |
1637 | * routine that would correctly distinguish all the above | |
1638 | * cases from the values of the function parameters and the | |
1639 | * inode state but for sanity's sake, I've decided to let the | |
1640 | * layers above just tell us. It's simpler to correctly figure | |
1641 | * out in the layer above exactly under what conditions we | |
1642 | * can run async and I think it's easier for others read and | |
1643 | * follow the logic in case something has to be changed. | |
1644 | * cscope is your friend -- rcc. | |
1645 | * | |
1646 | * The attribute fork is much simpler. | |
1647 | * | |
1648 | * For the attribute fork we allow the caller to tell us whether | |
1649 | * the unlink of the inode that led to this call is yet permanent | |
1650 | * in the on disk log. If it is not and we will be freeing extents | |
1651 | * in this inode then we make the first transaction synchronous | |
1652 | * to make sure that the unlink is permanent by the time we free | |
1653 | * the blocks. | |
1654 | */ | |
1655 | if (fork == XFS_DATA_FORK) { | |
1656 | if (ip->i_d.di_nextents > 0) { | |
ba87ea69 LM |
1657 | /* |
1658 | * If we are not changing the file size then do | |
1659 | * not update the on-disk file size - we may be | |
1660 | * called from xfs_inactive_free_eofblocks(). If we | |
1661 | * update the on-disk file size and then the system | |
1662 | * crashes before the contents of the file are | |
1663 | * flushed to disk then the files may be full of | |
1664 | * holes (ie NULL files bug). | |
1665 | */ | |
1666 | if (ip->i_size != new_size) { | |
1667 | ip->i_d.di_size = new_size; | |
1668 | ip->i_size = new_size; | |
1669 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1670 | } | |
1da177e4 LT |
1671 | } |
1672 | } else if (sync) { | |
1673 | ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC)); | |
1674 | if (ip->i_d.di_anextents > 0) | |
1675 | xfs_trans_set_sync(ntp); | |
1676 | } | |
1677 | ASSERT(fork == XFS_DATA_FORK || | |
1678 | (fork == XFS_ATTR_FORK && | |
1679 | ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) || | |
1680 | (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC))))); | |
1681 | ||
1682 | /* | |
1683 | * Since it is possible for space to become allocated beyond | |
1684 | * the end of the file (in a crash where the space is allocated | |
1685 | * but the inode size is not yet updated), simply remove any | |
1686 | * blocks which show up between the new EOF and the maximum | |
1687 | * possible file size. If the first block to be removed is | |
1688 | * beyond the maximum file size (ie it is the same as last_block), | |
1689 | * then there is nothing to do. | |
1690 | */ | |
1691 | last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp)); | |
1692 | ASSERT(first_unmap_block <= last_block); | |
1693 | done = 0; | |
1694 | if (last_block == first_unmap_block) { | |
1695 | done = 1; | |
1696 | } else { | |
1697 | unmap_len = last_block - first_unmap_block + 1; | |
1698 | } | |
1699 | while (!done) { | |
1700 | /* | |
1701 | * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi() | |
1702 | * will tell us whether it freed the entire range or | |
1703 | * not. If this is a synchronous mount (wsync), | |
1704 | * then we can tell bunmapi to keep all the | |
1705 | * transactions asynchronous since the unlink | |
1706 | * transaction that made this inode inactive has | |
1707 | * already hit the disk. There's no danger of | |
1708 | * the freed blocks being reused, there being a | |
1709 | * crash, and the reused blocks suddenly reappearing | |
1710 | * in this file with garbage in them once recovery | |
1711 | * runs. | |
1712 | */ | |
1713 | XFS_BMAP_INIT(&free_list, &first_block); | |
541d7d3c | 1714 | error = xfs_bunmapi(ntp, ip, |
3e57ecf6 | 1715 | first_unmap_block, unmap_len, |
1da177e4 LT |
1716 | XFS_BMAPI_AFLAG(fork) | |
1717 | (sync ? 0 : XFS_BMAPI_ASYNC), | |
1718 | XFS_ITRUNC_MAX_EXTENTS, | |
3e57ecf6 OW |
1719 | &first_block, &free_list, |
1720 | NULL, &done); | |
1da177e4 LT |
1721 | if (error) { |
1722 | /* | |
1723 | * If the bunmapi call encounters an error, | |
1724 | * return to the caller where the transaction | |
1725 | * can be properly aborted. We just need to | |
1726 | * make sure we're not holding any resources | |
1727 | * that we were not when we came in. | |
1728 | */ | |
1729 | xfs_bmap_cancel(&free_list); | |
1730 | return error; | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * Duplicate the transaction that has the permanent | |
1735 | * reservation and commit the old transaction. | |
1736 | */ | |
f7c99b6f | 1737 | error = xfs_bmap_finish(tp, &free_list, &committed); |
1da177e4 LT |
1738 | ntp = *tp; |
1739 | if (error) { | |
1740 | /* | |
1741 | * If the bmap finish call encounters an error, | |
1742 | * return to the caller where the transaction | |
1743 | * can be properly aborted. We just need to | |
1744 | * make sure we're not holding any resources | |
1745 | * that we were not when we came in. | |
1746 | * | |
1747 | * Aborting from this point might lose some | |
1748 | * blocks in the file system, but oh well. | |
1749 | */ | |
1750 | xfs_bmap_cancel(&free_list); | |
1751 | if (committed) { | |
1752 | /* | |
1753 | * If the passed in transaction committed | |
1754 | * in xfs_bmap_finish(), then we want to | |
1755 | * add the inode to this one before returning. | |
1756 | * This keeps things simple for the higher | |
1757 | * level code, because it always knows that | |
1758 | * the inode is locked and held in the | |
1759 | * transaction that returns to it whether | |
1760 | * errors occur or not. We don't mark the | |
1761 | * inode dirty so that this transaction can | |
1762 | * be easily aborted if possible. | |
1763 | */ | |
1764 | xfs_trans_ijoin(ntp, ip, | |
1765 | XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1766 | xfs_trans_ihold(ntp, ip); | |
1767 | } | |
1768 | return error; | |
1769 | } | |
1770 | ||
1771 | if (committed) { | |
1772 | /* | |
1773 | * The first xact was committed, | |
1774 | * so add the inode to the new one. | |
1775 | * Mark it dirty so it will be logged | |
1776 | * and moved forward in the log as | |
1777 | * part of every commit. | |
1778 | */ | |
1779 | xfs_trans_ijoin(ntp, ip, | |
1780 | XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1781 | xfs_trans_ihold(ntp, ip); | |
1782 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1783 | } | |
1784 | ntp = xfs_trans_dup(ntp); | |
1c72bf90 | 1785 | (void) xfs_trans_commit(*tp, 0); |
1da177e4 LT |
1786 | *tp = ntp; |
1787 | error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, | |
1788 | XFS_TRANS_PERM_LOG_RES, | |
1789 | XFS_ITRUNCATE_LOG_COUNT); | |
1790 | /* | |
1791 | * Add the inode being truncated to the next chained | |
1792 | * transaction. | |
1793 | */ | |
1794 | xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); | |
1795 | xfs_trans_ihold(ntp, ip); | |
1796 | if (error) | |
1797 | return (error); | |
1798 | } | |
1799 | /* | |
1800 | * Only update the size in the case of the data fork, but | |
1801 | * always re-log the inode so that our permanent transaction | |
1802 | * can keep on rolling it forward in the log. | |
1803 | */ | |
1804 | if (fork == XFS_DATA_FORK) { | |
1805 | xfs_isize_check(mp, ip, new_size); | |
ba87ea69 LM |
1806 | /* |
1807 | * If we are not changing the file size then do | |
1808 | * not update the on-disk file size - we may be | |
1809 | * called from xfs_inactive_free_eofblocks(). If we | |
1810 | * update the on-disk file size and then the system | |
1811 | * crashes before the contents of the file are | |
1812 | * flushed to disk then the files may be full of | |
1813 | * holes (ie NULL files bug). | |
1814 | */ | |
1815 | if (ip->i_size != new_size) { | |
1816 | ip->i_d.di_size = new_size; | |
1817 | ip->i_size = new_size; | |
1818 | } | |
1da177e4 LT |
1819 | } |
1820 | xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE); | |
1821 | ASSERT((new_size != 0) || | |
1822 | (fork == XFS_ATTR_FORK) || | |
1823 | (ip->i_delayed_blks == 0)); | |
1824 | ASSERT((new_size != 0) || | |
1825 | (fork == XFS_ATTR_FORK) || | |
1826 | (ip->i_d.di_nextents == 0)); | |
1827 | xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0); | |
1828 | return 0; | |
1829 | } | |
1830 | ||
1831 | ||
1832 | /* | |
1833 | * xfs_igrow_start | |
1834 | * | |
1835 | * Do the first part of growing a file: zero any data in the last | |
1836 | * block that is beyond the old EOF. We need to do this before | |
1837 | * the inode is joined to the transaction to modify the i_size. | |
1838 | * That way we can drop the inode lock and call into the buffer | |
1839 | * cache to get the buffer mapping the EOF. | |
1840 | */ | |
1841 | int | |
1842 | xfs_igrow_start( | |
1843 | xfs_inode_t *ip, | |
1844 | xfs_fsize_t new_size, | |
1845 | cred_t *credp) | |
1846 | { | |
1da177e4 LT |
1847 | ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0); |
1848 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0); | |
ba87ea69 | 1849 | ASSERT(new_size > ip->i_size); |
1da177e4 | 1850 | |
1da177e4 LT |
1851 | /* |
1852 | * Zero any pages that may have been created by | |
1853 | * xfs_write_file() beyond the end of the file | |
1854 | * and any blocks between the old and new file sizes. | |
1855 | */ | |
541d7d3c | 1856 | return xfs_zero_eof(ip, new_size, ip->i_size); |
1da177e4 LT |
1857 | } |
1858 | ||
1859 | /* | |
1860 | * xfs_igrow_finish | |
1861 | * | |
1862 | * This routine is called to extend the size of a file. | |
1863 | * The inode must have both the iolock and the ilock locked | |
1864 | * for update and it must be a part of the current transaction. | |
1865 | * The xfs_igrow_start() function must have been called previously. | |
1866 | * If the change_flag is not zero, the inode change timestamp will | |
1867 | * be updated. | |
1868 | */ | |
1869 | void | |
1870 | xfs_igrow_finish( | |
1871 | xfs_trans_t *tp, | |
1872 | xfs_inode_t *ip, | |
1873 | xfs_fsize_t new_size, | |
1874 | int change_flag) | |
1875 | { | |
1876 | ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0); | |
1877 | ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0); | |
1878 | ASSERT(ip->i_transp == tp); | |
ba87ea69 | 1879 | ASSERT(new_size > ip->i_size); |
1da177e4 LT |
1880 | |
1881 | /* | |
1882 | * Update the file size. Update the inode change timestamp | |
1883 | * if change_flag set. | |
1884 | */ | |
1885 | ip->i_d.di_size = new_size; | |
ba87ea69 | 1886 | ip->i_size = new_size; |
1da177e4 LT |
1887 | if (change_flag) |
1888 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG); | |
1889 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1890 | ||
1891 | } | |
1892 | ||
1893 | ||
1894 | /* | |
1895 | * This is called when the inode's link count goes to 0. | |
1896 | * We place the on-disk inode on a list in the AGI. It | |
1897 | * will be pulled from this list when the inode is freed. | |
1898 | */ | |
1899 | int | |
1900 | xfs_iunlink( | |
1901 | xfs_trans_t *tp, | |
1902 | xfs_inode_t *ip) | |
1903 | { | |
1904 | xfs_mount_t *mp; | |
1905 | xfs_agi_t *agi; | |
1906 | xfs_dinode_t *dip; | |
1907 | xfs_buf_t *agibp; | |
1908 | xfs_buf_t *ibp; | |
1909 | xfs_agnumber_t agno; | |
1910 | xfs_daddr_t agdaddr; | |
1911 | xfs_agino_t agino; | |
1912 | short bucket_index; | |
1913 | int offset; | |
1914 | int error; | |
1915 | int agi_ok; | |
1916 | ||
1917 | ASSERT(ip->i_d.di_nlink == 0); | |
1918 | ASSERT(ip->i_d.di_mode != 0); | |
1919 | ASSERT(ip->i_transp == tp); | |
1920 | ||
1921 | mp = tp->t_mountp; | |
1922 | ||
1923 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); | |
1924 | agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
1925 | ||
1926 | /* | |
1927 | * Get the agi buffer first. It ensures lock ordering | |
1928 | * on the list. | |
1929 | */ | |
1930 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, | |
1931 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
859d7182 | 1932 | if (error) |
1da177e4 | 1933 | return error; |
859d7182 | 1934 | |
1da177e4 LT |
1935 | /* |
1936 | * Validate the magic number of the agi block. | |
1937 | */ | |
1938 | agi = XFS_BUF_TO_AGI(agibp); | |
1939 | agi_ok = | |
16259e7d CH |
1940 | be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC && |
1941 | XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)); | |
1da177e4 LT |
1942 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK, |
1943 | XFS_RANDOM_IUNLINK))) { | |
1944 | XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi); | |
1945 | xfs_trans_brelse(tp, agibp); | |
1946 | return XFS_ERROR(EFSCORRUPTED); | |
1947 | } | |
1948 | /* | |
1949 | * Get the index into the agi hash table for the | |
1950 | * list this inode will go on. | |
1951 | */ | |
1952 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1953 | ASSERT(agino != 0); | |
1954 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1955 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1956 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1957 | |
859d7182 VA |
1958 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); |
1959 | if (error) | |
1960 | return error; | |
1961 | ||
1962 | /* | |
1963 | * Clear the on-disk di_nlink. This is to prevent xfs_bulkstat | |
1964 | * from picking up this inode when it is reclaimed (its incore state | |
1965 | * initialzed but not flushed to disk yet). The in-core di_nlink is | |
1966 | * already cleared in xfs_droplink() and a corresponding transaction | |
1967 | * logged. The hack here just synchronizes the in-core to on-disk | |
1968 | * di_nlink value in advance before the actual inode sync to disk. | |
1969 | * This is OK because the inode is already unlinked and would never | |
1970 | * change its di_nlink again for this inode generation. | |
1971 | * This is a temporary hack that would require a proper fix | |
1972 | * in the future. | |
1973 | */ | |
1974 | dip->di_core.di_nlink = 0; | |
1975 | ||
16259e7d | 1976 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) { |
1da177e4 LT |
1977 | /* |
1978 | * There is already another inode in the bucket we need | |
1979 | * to add ourselves to. Add us at the front of the list. | |
1980 | * Here we put the head pointer into our next pointer, | |
1981 | * and then we fall through to point the head at us. | |
1982 | */ | |
347d1c01 | 1983 | ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO); |
1da177e4 LT |
1984 | /* both on-disk, don't endian flip twice */ |
1985 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; | |
1986 | offset = ip->i_boffset + | |
1987 | offsetof(xfs_dinode_t, di_next_unlinked); | |
1988 | xfs_trans_inode_buf(tp, ibp); | |
1989 | xfs_trans_log_buf(tp, ibp, offset, | |
1990 | (offset + sizeof(xfs_agino_t) - 1)); | |
1991 | xfs_inobp_check(mp, ibp); | |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * Point the bucket head pointer at the inode being inserted. | |
1996 | */ | |
1997 | ASSERT(agino != 0); | |
16259e7d | 1998 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
1999 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2000 | (sizeof(xfs_agino_t) * bucket_index); | |
2001 | xfs_trans_log_buf(tp, agibp, offset, | |
2002 | (offset + sizeof(xfs_agino_t) - 1)); | |
2003 | return 0; | |
2004 | } | |
2005 | ||
2006 | /* | |
2007 | * Pull the on-disk inode from the AGI unlinked list. | |
2008 | */ | |
2009 | STATIC int | |
2010 | xfs_iunlink_remove( | |
2011 | xfs_trans_t *tp, | |
2012 | xfs_inode_t *ip) | |
2013 | { | |
2014 | xfs_ino_t next_ino; | |
2015 | xfs_mount_t *mp; | |
2016 | xfs_agi_t *agi; | |
2017 | xfs_dinode_t *dip; | |
2018 | xfs_buf_t *agibp; | |
2019 | xfs_buf_t *ibp; | |
2020 | xfs_agnumber_t agno; | |
2021 | xfs_daddr_t agdaddr; | |
2022 | xfs_agino_t agino; | |
2023 | xfs_agino_t next_agino; | |
2024 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 2025 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 2026 | short bucket_index; |
6fdf8ccc | 2027 | int offset, last_offset = 0; |
1da177e4 LT |
2028 | int error; |
2029 | int agi_ok; | |
2030 | ||
2031 | /* | |
2032 | * First pull the on-disk inode from the AGI unlinked list. | |
2033 | */ | |
2034 | mp = tp->t_mountp; | |
2035 | ||
2036 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); | |
2037 | agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)); | |
2038 | ||
2039 | /* | |
2040 | * Get the agi buffer first. It ensures lock ordering | |
2041 | * on the list. | |
2042 | */ | |
2043 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr, | |
2044 | XFS_FSS_TO_BB(mp, 1), 0, &agibp); | |
2045 | if (error) { | |
2046 | cmn_err(CE_WARN, | |
2047 | "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.", | |
2048 | error, mp->m_fsname); | |
2049 | return error; | |
2050 | } | |
2051 | /* | |
2052 | * Validate the magic number of the agi block. | |
2053 | */ | |
2054 | agi = XFS_BUF_TO_AGI(agibp); | |
2055 | agi_ok = | |
16259e7d CH |
2056 | be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC && |
2057 | XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)); | |
1da177e4 LT |
2058 | if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE, |
2059 | XFS_RANDOM_IUNLINK_REMOVE))) { | |
2060 | XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW, | |
2061 | mp, agi); | |
2062 | xfs_trans_brelse(tp, agibp); | |
2063 | cmn_err(CE_WARN, | |
2064 | "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.", | |
2065 | mp->m_fsname); | |
2066 | return XFS_ERROR(EFSCORRUPTED); | |
2067 | } | |
2068 | /* | |
2069 | * Get the index into the agi hash table for the | |
2070 | * list this inode will go on. | |
2071 | */ | |
2072 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
2073 | ASSERT(agino != 0); | |
2074 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
16259e7d | 2075 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO); |
1da177e4 LT |
2076 | ASSERT(agi->agi_unlinked[bucket_index]); |
2077 | ||
16259e7d | 2078 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 LT |
2079 | /* |
2080 | * We're at the head of the list. Get the inode's | |
2081 | * on-disk buffer to see if there is anyone after us | |
2082 | * on the list. Only modify our next pointer if it | |
2083 | * is not already NULLAGINO. This saves us the overhead | |
2084 | * of dealing with the buffer when there is no need to | |
2085 | * change it. | |
2086 | */ | |
b12dd342 | 2087 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); |
1da177e4 LT |
2088 | if (error) { |
2089 | cmn_err(CE_WARN, | |
2090 | "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", | |
2091 | error, mp->m_fsname); | |
2092 | return error; | |
2093 | } | |
347d1c01 | 2094 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2095 | ASSERT(next_agino != 0); |
2096 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2097 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
2098 | offset = ip->i_boffset + |
2099 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2100 | xfs_trans_inode_buf(tp, ibp); | |
2101 | xfs_trans_log_buf(tp, ibp, offset, | |
2102 | (offset + sizeof(xfs_agino_t) - 1)); | |
2103 | xfs_inobp_check(mp, ibp); | |
2104 | } else { | |
2105 | xfs_trans_brelse(tp, ibp); | |
2106 | } | |
2107 | /* | |
2108 | * Point the bucket head pointer at the next inode. | |
2109 | */ | |
2110 | ASSERT(next_agino != 0); | |
2111 | ASSERT(next_agino != agino); | |
16259e7d | 2112 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2113 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2114 | (sizeof(xfs_agino_t) * bucket_index); | |
2115 | xfs_trans_log_buf(tp, agibp, offset, | |
2116 | (offset + sizeof(xfs_agino_t) - 1)); | |
2117 | } else { | |
2118 | /* | |
2119 | * We need to search the list for the inode being freed. | |
2120 | */ | |
16259e7d | 2121 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2122 | last_ibp = NULL; |
2123 | while (next_agino != agino) { | |
2124 | /* | |
2125 | * If the last inode wasn't the one pointing to | |
2126 | * us, then release its buffer since we're not | |
2127 | * going to do anything with it. | |
2128 | */ | |
2129 | if (last_ibp != NULL) { | |
2130 | xfs_trans_brelse(tp, last_ibp); | |
2131 | } | |
2132 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); | |
2133 | error = xfs_inotobp(mp, tp, next_ino, &last_dip, | |
2134 | &last_ibp, &last_offset); | |
2135 | if (error) { | |
2136 | cmn_err(CE_WARN, | |
2137 | "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.", | |
2138 | error, mp->m_fsname); | |
2139 | return error; | |
2140 | } | |
347d1c01 | 2141 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
2142 | ASSERT(next_agino != NULLAGINO); |
2143 | ASSERT(next_agino != 0); | |
2144 | } | |
2145 | /* | |
2146 | * Now last_ibp points to the buffer previous to us on | |
2147 | * the unlinked list. Pull us from the list. | |
2148 | */ | |
b12dd342 | 2149 | error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0); |
1da177e4 LT |
2150 | if (error) { |
2151 | cmn_err(CE_WARN, | |
2152 | "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.", | |
2153 | error, mp->m_fsname); | |
2154 | return error; | |
2155 | } | |
347d1c01 | 2156 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2157 | ASSERT(next_agino != 0); |
2158 | ASSERT(next_agino != agino); | |
2159 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2160 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
1da177e4 LT |
2161 | offset = ip->i_boffset + |
2162 | offsetof(xfs_dinode_t, di_next_unlinked); | |
2163 | xfs_trans_inode_buf(tp, ibp); | |
2164 | xfs_trans_log_buf(tp, ibp, offset, | |
2165 | (offset + sizeof(xfs_agino_t) - 1)); | |
2166 | xfs_inobp_check(mp, ibp); | |
2167 | } else { | |
2168 | xfs_trans_brelse(tp, ibp); | |
2169 | } | |
2170 | /* | |
2171 | * Point the previous inode on the list to the next inode. | |
2172 | */ | |
347d1c01 | 2173 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
2174 | ASSERT(next_agino != 0); |
2175 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
2176 | xfs_trans_inode_buf(tp, last_ibp); | |
2177 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2178 | (offset + sizeof(xfs_agino_t) - 1)); | |
2179 | xfs_inobp_check(mp, last_ibp); | |
2180 | } | |
2181 | return 0; | |
2182 | } | |
2183 | ||
7989cb8e | 2184 | STATIC_INLINE int xfs_inode_clean(xfs_inode_t *ip) |
1da177e4 LT |
2185 | { |
2186 | return (((ip->i_itemp == NULL) || | |
2187 | !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) && | |
2188 | (ip->i_update_core == 0)); | |
2189 | } | |
2190 | ||
ba0f32d4 | 2191 | STATIC void |
1da177e4 LT |
2192 | xfs_ifree_cluster( |
2193 | xfs_inode_t *free_ip, | |
2194 | xfs_trans_t *tp, | |
2195 | xfs_ino_t inum) | |
2196 | { | |
2197 | xfs_mount_t *mp = free_ip->i_mount; | |
2198 | int blks_per_cluster; | |
2199 | int nbufs; | |
2200 | int ninodes; | |
2201 | int i, j, found, pre_flushed; | |
2202 | xfs_daddr_t blkno; | |
2203 | xfs_buf_t *bp; | |
1da177e4 LT |
2204 | xfs_inode_t *ip, **ip_found; |
2205 | xfs_inode_log_item_t *iip; | |
2206 | xfs_log_item_t *lip; | |
da353b0d | 2207 | xfs_perag_t *pag = xfs_get_perag(mp, inum); |
1da177e4 LT |
2208 | |
2209 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { | |
2210 | blks_per_cluster = 1; | |
2211 | ninodes = mp->m_sb.sb_inopblock; | |
2212 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
2213 | } else { | |
2214 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
2215 | mp->m_sb.sb_blocksize; | |
2216 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
2217 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
2218 | } | |
2219 | ||
2220 | ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS); | |
2221 | ||
2222 | for (j = 0; j < nbufs; j++, inum += ninodes) { | |
2223 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
2224 | XFS_INO_TO_AGBNO(mp, inum)); | |
2225 | ||
2226 | ||
2227 | /* | |
2228 | * Look for each inode in memory and attempt to lock it, | |
2229 | * we can be racing with flush and tail pushing here. | |
2230 | * any inode we get the locks on, add to an array of | |
2231 | * inode items to process later. | |
2232 | * | |
2233 | * The get the buffer lock, we could beat a flush | |
2234 | * or tail pushing thread to the lock here, in which | |
2235 | * case they will go looking for the inode buffer | |
2236 | * and fail, we need some other form of interlock | |
2237 | * here. | |
2238 | */ | |
2239 | found = 0; | |
2240 | for (i = 0; i < ninodes; i++) { | |
da353b0d DC |
2241 | read_lock(&pag->pag_ici_lock); |
2242 | ip = radix_tree_lookup(&pag->pag_ici_root, | |
2243 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 LT |
2244 | |
2245 | /* Inode not in memory or we found it already, | |
2246 | * nothing to do | |
2247 | */ | |
7a18c386 | 2248 | if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) { |
da353b0d | 2249 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
2250 | continue; |
2251 | } | |
2252 | ||
2253 | if (xfs_inode_clean(ip)) { | |
da353b0d | 2254 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
2255 | continue; |
2256 | } | |
2257 | ||
2258 | /* If we can get the locks then add it to the | |
2259 | * list, otherwise by the time we get the bp lock | |
2260 | * below it will already be attached to the | |
2261 | * inode buffer. | |
2262 | */ | |
2263 | ||
2264 | /* This inode will already be locked - by us, lets | |
2265 | * keep it that way. | |
2266 | */ | |
2267 | ||
2268 | if (ip == free_ip) { | |
2269 | if (xfs_iflock_nowait(ip)) { | |
7a18c386 | 2270 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 LT |
2271 | if (xfs_inode_clean(ip)) { |
2272 | xfs_ifunlock(ip); | |
2273 | } else { | |
2274 | ip_found[found++] = ip; | |
2275 | } | |
2276 | } | |
da353b0d | 2277 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
2278 | continue; |
2279 | } | |
2280 | ||
2281 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
2282 | if (xfs_iflock_nowait(ip)) { | |
7a18c386 | 2283 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 LT |
2284 | |
2285 | if (xfs_inode_clean(ip)) { | |
2286 | xfs_ifunlock(ip); | |
2287 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2288 | } else { | |
2289 | ip_found[found++] = ip; | |
2290 | } | |
2291 | } else { | |
2292 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2293 | } | |
2294 | } | |
da353b0d | 2295 | read_unlock(&pag->pag_ici_lock); |
1da177e4 LT |
2296 | } |
2297 | ||
2298 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
2299 | mp->m_bsize * blks_per_cluster, | |
2300 | XFS_BUF_LOCK); | |
2301 | ||
2302 | pre_flushed = 0; | |
2303 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); | |
2304 | while (lip) { | |
2305 | if (lip->li_type == XFS_LI_INODE) { | |
2306 | iip = (xfs_inode_log_item_t *)lip; | |
2307 | ASSERT(iip->ili_logged == 1); | |
2308 | lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done; | |
287f3dad | 2309 | spin_lock(&mp->m_ail_lock); |
1da177e4 | 2310 | iip->ili_flush_lsn = iip->ili_item.li_lsn; |
287f3dad | 2311 | spin_unlock(&mp->m_ail_lock); |
e5ffd2bb | 2312 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); |
1da177e4 LT |
2313 | pre_flushed++; |
2314 | } | |
2315 | lip = lip->li_bio_list; | |
2316 | } | |
2317 | ||
2318 | for (i = 0; i < found; i++) { | |
2319 | ip = ip_found[i]; | |
2320 | iip = ip->i_itemp; | |
2321 | ||
2322 | if (!iip) { | |
2323 | ip->i_update_core = 0; | |
2324 | xfs_ifunlock(ip); | |
2325 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2326 | continue; | |
2327 | } | |
2328 | ||
2329 | iip->ili_last_fields = iip->ili_format.ilf_fields; | |
2330 | iip->ili_format.ilf_fields = 0; | |
2331 | iip->ili_logged = 1; | |
287f3dad | 2332 | spin_lock(&mp->m_ail_lock); |
1da177e4 | 2333 | iip->ili_flush_lsn = iip->ili_item.li_lsn; |
287f3dad | 2334 | spin_unlock(&mp->m_ail_lock); |
1da177e4 LT |
2335 | |
2336 | xfs_buf_attach_iodone(bp, | |
2337 | (void(*)(xfs_buf_t*,xfs_log_item_t*)) | |
2338 | xfs_istale_done, (xfs_log_item_t *)iip); | |
2339 | if (ip != free_ip) { | |
2340 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2341 | } | |
2342 | } | |
2343 | ||
2344 | if (found || pre_flushed) | |
2345 | xfs_trans_stale_inode_buf(tp, bp); | |
2346 | xfs_trans_binval(tp, bp); | |
2347 | } | |
2348 | ||
2349 | kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *)); | |
da353b0d | 2350 | xfs_put_perag(mp, pag); |
1da177e4 LT |
2351 | } |
2352 | ||
2353 | /* | |
2354 | * This is called to return an inode to the inode free list. | |
2355 | * The inode should already be truncated to 0 length and have | |
2356 | * no pages associated with it. This routine also assumes that | |
2357 | * the inode is already a part of the transaction. | |
2358 | * | |
2359 | * The on-disk copy of the inode will have been added to the list | |
2360 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2361 | * that list atomically with respect to freeing it here. | |
2362 | */ | |
2363 | int | |
2364 | xfs_ifree( | |
2365 | xfs_trans_t *tp, | |
2366 | xfs_inode_t *ip, | |
2367 | xfs_bmap_free_t *flist) | |
2368 | { | |
2369 | int error; | |
2370 | int delete; | |
2371 | xfs_ino_t first_ino; | |
2372 | ||
2373 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); | |
2374 | ASSERT(ip->i_transp == tp); | |
2375 | ASSERT(ip->i_d.di_nlink == 0); | |
2376 | ASSERT(ip->i_d.di_nextents == 0); | |
2377 | ASSERT(ip->i_d.di_anextents == 0); | |
ba87ea69 | 2378 | ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) || |
1da177e4 LT |
2379 | ((ip->i_d.di_mode & S_IFMT) != S_IFREG)); |
2380 | ASSERT(ip->i_d.di_nblocks == 0); | |
2381 | ||
2382 | /* | |
2383 | * Pull the on-disk inode from the AGI unlinked list. | |
2384 | */ | |
2385 | error = xfs_iunlink_remove(tp, ip); | |
2386 | if (error != 0) { | |
2387 | return error; | |
2388 | } | |
2389 | ||
2390 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
2391 | if (error != 0) { | |
2392 | return error; | |
2393 | } | |
2394 | ip->i_d.di_mode = 0; /* mark incore inode as free */ | |
2395 | ip->i_d.di_flags = 0; | |
2396 | ip->i_d.di_dmevmask = 0; | |
2397 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
2398 | ip->i_df.if_ext_max = | |
2399 | XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t); | |
2400 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
2401 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2402 | /* | |
2403 | * Bump the generation count so no one will be confused | |
2404 | * by reincarnations of this inode. | |
2405 | */ | |
2406 | ip->i_d.di_gen++; | |
2407 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2408 | ||
2409 | if (delete) { | |
2410 | xfs_ifree_cluster(ip, tp, first_ino); | |
2411 | } | |
2412 | ||
2413 | return 0; | |
2414 | } | |
2415 | ||
2416 | /* | |
2417 | * Reallocate the space for if_broot based on the number of records | |
2418 | * being added or deleted as indicated in rec_diff. Move the records | |
2419 | * and pointers in if_broot to fit the new size. When shrinking this | |
2420 | * will eliminate holes between the records and pointers created by | |
2421 | * the caller. When growing this will create holes to be filled in | |
2422 | * by the caller. | |
2423 | * | |
2424 | * The caller must not request to add more records than would fit in | |
2425 | * the on-disk inode root. If the if_broot is currently NULL, then | |
2426 | * if we adding records one will be allocated. The caller must also | |
2427 | * not request that the number of records go below zero, although | |
2428 | * it can go to zero. | |
2429 | * | |
2430 | * ip -- the inode whose if_broot area is changing | |
2431 | * ext_diff -- the change in the number of records, positive or negative, | |
2432 | * requested for the if_broot array. | |
2433 | */ | |
2434 | void | |
2435 | xfs_iroot_realloc( | |
2436 | xfs_inode_t *ip, | |
2437 | int rec_diff, | |
2438 | int whichfork) | |
2439 | { | |
2440 | int cur_max; | |
2441 | xfs_ifork_t *ifp; | |
2442 | xfs_bmbt_block_t *new_broot; | |
2443 | int new_max; | |
2444 | size_t new_size; | |
2445 | char *np; | |
2446 | char *op; | |
2447 | ||
2448 | /* | |
2449 | * Handle the degenerate case quietly. | |
2450 | */ | |
2451 | if (rec_diff == 0) { | |
2452 | return; | |
2453 | } | |
2454 | ||
2455 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2456 | if (rec_diff > 0) { | |
2457 | /* | |
2458 | * If there wasn't any memory allocated before, just | |
2459 | * allocate it now and get out. | |
2460 | */ | |
2461 | if (ifp->if_broot_bytes == 0) { | |
2462 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); | |
2463 | ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size, | |
2464 | KM_SLEEP); | |
2465 | ifp->if_broot_bytes = (int)new_size; | |
2466 | return; | |
2467 | } | |
2468 | ||
2469 | /* | |
2470 | * If there is already an existing if_broot, then we need | |
2471 | * to realloc() it and shift the pointers to their new | |
2472 | * location. The records don't change location because | |
2473 | * they are kept butted up against the btree block header. | |
2474 | */ | |
2475 | cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes); | |
2476 | new_max = cur_max + rec_diff; | |
2477 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2478 | ifp->if_broot = (xfs_bmbt_block_t *) | |
2479 | kmem_realloc(ifp->if_broot, | |
2480 | new_size, | |
2481 | (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ | |
2482 | KM_SLEEP); | |
2483 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2484 | ifp->if_broot_bytes); | |
2485 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2486 | (int)new_size); | |
2487 | ifp->if_broot_bytes = (int)new_size; | |
2488 | ASSERT(ifp->if_broot_bytes <= | |
2489 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2490 | memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); | |
2491 | return; | |
2492 | } | |
2493 | ||
2494 | /* | |
2495 | * rec_diff is less than 0. In this case, we are shrinking the | |
2496 | * if_broot buffer. It must already exist. If we go to zero | |
2497 | * records, just get rid of the root and clear the status bit. | |
2498 | */ | |
2499 | ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); | |
2500 | cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes); | |
2501 | new_max = cur_max + rec_diff; | |
2502 | ASSERT(new_max >= 0); | |
2503 | if (new_max > 0) | |
2504 | new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); | |
2505 | else | |
2506 | new_size = 0; | |
2507 | if (new_size > 0) { | |
2508 | new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP); | |
2509 | /* | |
2510 | * First copy over the btree block header. | |
2511 | */ | |
2512 | memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t)); | |
2513 | } else { | |
2514 | new_broot = NULL; | |
2515 | ifp->if_flags &= ~XFS_IFBROOT; | |
2516 | } | |
2517 | ||
2518 | /* | |
2519 | * Only copy the records and pointers if there are any. | |
2520 | */ | |
2521 | if (new_max > 0) { | |
2522 | /* | |
2523 | * First copy the records. | |
2524 | */ | |
2525 | op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1, | |
2526 | ifp->if_broot_bytes); | |
2527 | np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1, | |
2528 | (int)new_size); | |
2529 | memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); | |
2530 | ||
2531 | /* | |
2532 | * Then copy the pointers. | |
2533 | */ | |
2534 | op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1, | |
2535 | ifp->if_broot_bytes); | |
2536 | np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1, | |
2537 | (int)new_size); | |
2538 | memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); | |
2539 | } | |
2540 | kmem_free(ifp->if_broot, ifp->if_broot_bytes); | |
2541 | ifp->if_broot = new_broot; | |
2542 | ifp->if_broot_bytes = (int)new_size; | |
2543 | ASSERT(ifp->if_broot_bytes <= | |
2544 | XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); | |
2545 | return; | |
2546 | } | |
2547 | ||
2548 | ||
1da177e4 LT |
2549 | /* |
2550 | * This is called when the amount of space needed for if_data | |
2551 | * is increased or decreased. The change in size is indicated by | |
2552 | * the number of bytes that need to be added or deleted in the | |
2553 | * byte_diff parameter. | |
2554 | * | |
2555 | * If the amount of space needed has decreased below the size of the | |
2556 | * inline buffer, then switch to using the inline buffer. Otherwise, | |
2557 | * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer | |
2558 | * to what is needed. | |
2559 | * | |
2560 | * ip -- the inode whose if_data area is changing | |
2561 | * byte_diff -- the change in the number of bytes, positive or negative, | |
2562 | * requested for the if_data array. | |
2563 | */ | |
2564 | void | |
2565 | xfs_idata_realloc( | |
2566 | xfs_inode_t *ip, | |
2567 | int byte_diff, | |
2568 | int whichfork) | |
2569 | { | |
2570 | xfs_ifork_t *ifp; | |
2571 | int new_size; | |
2572 | int real_size; | |
2573 | ||
2574 | if (byte_diff == 0) { | |
2575 | return; | |
2576 | } | |
2577 | ||
2578 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2579 | new_size = (int)ifp->if_bytes + byte_diff; | |
2580 | ASSERT(new_size >= 0); | |
2581 | ||
2582 | if (new_size == 0) { | |
2583 | if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2584 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2585 | } | |
2586 | ifp->if_u1.if_data = NULL; | |
2587 | real_size = 0; | |
2588 | } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { | |
2589 | /* | |
2590 | * If the valid extents/data can fit in if_inline_ext/data, | |
2591 | * copy them from the malloc'd vector and free it. | |
2592 | */ | |
2593 | if (ifp->if_u1.if_data == NULL) { | |
2594 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2595 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2596 | ASSERT(ifp->if_real_bytes != 0); | |
2597 | memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, | |
2598 | new_size); | |
2599 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2600 | ifp->if_u1.if_data = ifp->if_u2.if_inline_data; | |
2601 | } | |
2602 | real_size = 0; | |
2603 | } else { | |
2604 | /* | |
2605 | * Stuck with malloc/realloc. | |
2606 | * For inline data, the underlying buffer must be | |
2607 | * a multiple of 4 bytes in size so that it can be | |
2608 | * logged and stay on word boundaries. We enforce | |
2609 | * that here. | |
2610 | */ | |
2611 | real_size = roundup(new_size, 4); | |
2612 | if (ifp->if_u1.if_data == NULL) { | |
2613 | ASSERT(ifp->if_real_bytes == 0); | |
2614 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
2615 | } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { | |
2616 | /* | |
2617 | * Only do the realloc if the underlying size | |
2618 | * is really changing. | |
2619 | */ | |
2620 | if (ifp->if_real_bytes != real_size) { | |
2621 | ifp->if_u1.if_data = | |
2622 | kmem_realloc(ifp->if_u1.if_data, | |
2623 | real_size, | |
2624 | ifp->if_real_bytes, | |
2625 | KM_SLEEP); | |
2626 | } | |
2627 | } else { | |
2628 | ASSERT(ifp->if_real_bytes == 0); | |
2629 | ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP); | |
2630 | memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, | |
2631 | ifp->if_bytes); | |
2632 | } | |
2633 | } | |
2634 | ifp->if_real_bytes = real_size; | |
2635 | ifp->if_bytes = new_size; | |
2636 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2637 | } | |
2638 | ||
2639 | ||
2640 | ||
2641 | ||
2642 | /* | |
2643 | * Map inode to disk block and offset. | |
2644 | * | |
2645 | * mp -- the mount point structure for the current file system | |
2646 | * tp -- the current transaction | |
2647 | * ino -- the inode number of the inode to be located | |
2648 | * imap -- this structure is filled in with the information necessary | |
2649 | * to retrieve the given inode from disk | |
2650 | * flags -- flags to pass to xfs_dilocate indicating whether or not | |
2651 | * lookups in the inode btree were OK or not | |
2652 | */ | |
2653 | int | |
2654 | xfs_imap( | |
2655 | xfs_mount_t *mp, | |
2656 | xfs_trans_t *tp, | |
2657 | xfs_ino_t ino, | |
2658 | xfs_imap_t *imap, | |
2659 | uint flags) | |
2660 | { | |
2661 | xfs_fsblock_t fsbno; | |
2662 | int len; | |
2663 | int off; | |
2664 | int error; | |
2665 | ||
2666 | fsbno = imap->im_blkno ? | |
2667 | XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK; | |
2668 | error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags); | |
2669 | if (error != 0) { | |
2670 | return error; | |
2671 | } | |
2672 | imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno); | |
2673 | imap->im_len = XFS_FSB_TO_BB(mp, len); | |
2674 | imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno); | |
2675 | imap->im_ioffset = (ushort)off; | |
2676 | imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog); | |
2677 | return 0; | |
2678 | } | |
2679 | ||
2680 | void | |
2681 | xfs_idestroy_fork( | |
2682 | xfs_inode_t *ip, | |
2683 | int whichfork) | |
2684 | { | |
2685 | xfs_ifork_t *ifp; | |
2686 | ||
2687 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2688 | if (ifp->if_broot != NULL) { | |
2689 | kmem_free(ifp->if_broot, ifp->if_broot_bytes); | |
2690 | ifp->if_broot = NULL; | |
2691 | } | |
2692 | ||
2693 | /* | |
2694 | * If the format is local, then we can't have an extents | |
2695 | * array so just look for an inline data array. If we're | |
2696 | * not local then we may or may not have an extents list, | |
2697 | * so check and free it up if we do. | |
2698 | */ | |
2699 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
2700 | if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && | |
2701 | (ifp->if_u1.if_data != NULL)) { | |
2702 | ASSERT(ifp->if_real_bytes != 0); | |
2703 | kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes); | |
2704 | ifp->if_u1.if_data = NULL; | |
2705 | ifp->if_real_bytes = 0; | |
2706 | } | |
2707 | } else if ((ifp->if_flags & XFS_IFEXTENTS) && | |
0293ce3a MK |
2708 | ((ifp->if_flags & XFS_IFEXTIREC) || |
2709 | ((ifp->if_u1.if_extents != NULL) && | |
2710 | (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { | |
1da177e4 | 2711 | ASSERT(ifp->if_real_bytes != 0); |
4eea22f0 | 2712 | xfs_iext_destroy(ifp); |
1da177e4 LT |
2713 | } |
2714 | ASSERT(ifp->if_u1.if_extents == NULL || | |
2715 | ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); | |
2716 | ASSERT(ifp->if_real_bytes == 0); | |
2717 | if (whichfork == XFS_ATTR_FORK) { | |
2718 | kmem_zone_free(xfs_ifork_zone, ip->i_afp); | |
2719 | ip->i_afp = NULL; | |
2720 | } | |
2721 | } | |
2722 | ||
2723 | /* | |
2724 | * This is called free all the memory associated with an inode. | |
2725 | * It must free the inode itself and any buffers allocated for | |
2726 | * if_extents/if_data and if_broot. It must also free the lock | |
2727 | * associated with the inode. | |
2728 | */ | |
2729 | void | |
2730 | xfs_idestroy( | |
2731 | xfs_inode_t *ip) | |
2732 | { | |
1da177e4 LT |
2733 | switch (ip->i_d.di_mode & S_IFMT) { |
2734 | case S_IFREG: | |
2735 | case S_IFDIR: | |
2736 | case S_IFLNK: | |
2737 | xfs_idestroy_fork(ip, XFS_DATA_FORK); | |
2738 | break; | |
2739 | } | |
2740 | if (ip->i_afp) | |
2741 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
2742 | mrfree(&ip->i_lock); | |
2743 | mrfree(&ip->i_iolock); | |
2744 | freesema(&ip->i_flock); | |
1543d79c | 2745 | |
cf441eeb | 2746 | #ifdef XFS_INODE_TRACE |
1543d79c CH |
2747 | ktrace_free(ip->i_trace); |
2748 | #endif | |
1da177e4 LT |
2749 | #ifdef XFS_BMAP_TRACE |
2750 | ktrace_free(ip->i_xtrace); | |
2751 | #endif | |
2752 | #ifdef XFS_BMBT_TRACE | |
2753 | ktrace_free(ip->i_btrace); | |
2754 | #endif | |
2755 | #ifdef XFS_RW_TRACE | |
2756 | ktrace_free(ip->i_rwtrace); | |
2757 | #endif | |
2758 | #ifdef XFS_ILOCK_TRACE | |
2759 | ktrace_free(ip->i_lock_trace); | |
2760 | #endif | |
2761 | #ifdef XFS_DIR2_TRACE | |
2762 | ktrace_free(ip->i_dir_trace); | |
2763 | #endif | |
2764 | if (ip->i_itemp) { | |
f74eaf59 DC |
2765 | /* |
2766 | * Only if we are shutting down the fs will we see an | |
2767 | * inode still in the AIL. If it is there, we should remove | |
2768 | * it to prevent a use-after-free from occurring. | |
2769 | */ | |
2770 | xfs_mount_t *mp = ip->i_mount; | |
2771 | xfs_log_item_t *lip = &ip->i_itemp->ili_item; | |
f74eaf59 DC |
2772 | |
2773 | ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) || | |
2774 | XFS_FORCED_SHUTDOWN(ip->i_mount)); | |
2775 | if (lip->li_flags & XFS_LI_IN_AIL) { | |
287f3dad | 2776 | spin_lock(&mp->m_ail_lock); |
f74eaf59 | 2777 | if (lip->li_flags & XFS_LI_IN_AIL) |
287f3dad | 2778 | xfs_trans_delete_ail(mp, lip); |
f74eaf59 | 2779 | else |
287f3dad | 2780 | spin_unlock(&mp->m_ail_lock); |
f74eaf59 | 2781 | } |
1da177e4 LT |
2782 | xfs_inode_item_destroy(ip); |
2783 | } | |
2784 | kmem_zone_free(xfs_inode_zone, ip); | |
2785 | } | |
2786 | ||
2787 | ||
2788 | /* | |
2789 | * Increment the pin count of the given buffer. | |
2790 | * This value is protected by ipinlock spinlock in the mount structure. | |
2791 | */ | |
2792 | void | |
2793 | xfs_ipin( | |
2794 | xfs_inode_t *ip) | |
2795 | { | |
2796 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); | |
2797 | ||
2798 | atomic_inc(&ip->i_pincount); | |
2799 | } | |
2800 | ||
2801 | /* | |
2802 | * Decrement the pin count of the given inode, and wake up | |
2803 | * anyone in xfs_iwait_unpin() if the count goes to 0. The | |
c41564b5 | 2804 | * inode must have been previously pinned with a call to xfs_ipin(). |
1da177e4 LT |
2805 | */ |
2806 | void | |
2807 | xfs_iunpin( | |
2808 | xfs_inode_t *ip) | |
2809 | { | |
2810 | ASSERT(atomic_read(&ip->i_pincount) > 0); | |
2811 | ||
4c60658e DC |
2812 | if (atomic_dec_and_lock(&ip->i_pincount, &ip->i_flags_lock)) { |
2813 | ||
58829e49 | 2814 | /* |
4c60658e DC |
2815 | * If the inode is currently being reclaimed, the link between |
2816 | * the bhv_vnode and the xfs_inode will be broken after the | |
2817 | * XFS_IRECLAIM* flag is set. Hence, if these flags are not | |
2818 | * set, then we can move forward and mark the linux inode dirty | |
2819 | * knowing that it is still valid as it won't freed until after | |
2820 | * the bhv_vnode<->xfs_inode link is broken in xfs_reclaim. The | |
2821 | * i_flags_lock is used to synchronise the setting of the | |
2822 | * XFS_IRECLAIM* flags and the breaking of the link, and so we | |
2823 | * can execute atomically w.r.t to reclaim by holding this lock | |
2824 | * here. | |
58829e49 | 2825 | * |
4c60658e DC |
2826 | * However, we still need to issue the unpin wakeup call as the |
2827 | * inode reclaim may be blocked waiting for the inode to become | |
2828 | * unpinned. | |
58829e49 | 2829 | */ |
f273ab84 | 2830 | |
7a18c386 | 2831 | if (!__xfs_iflags_test(ip, XFS_IRECLAIM|XFS_IRECLAIMABLE)) { |
67fcaa73 | 2832 | bhv_vnode_t *vp = XFS_ITOV_NULL(ip); |
4c60658e DC |
2833 | struct inode *inode = NULL; |
2834 | ||
2835 | BUG_ON(vp == NULL); | |
2836 | inode = vn_to_inode(vp); | |
2837 | BUG_ON(inode->i_state & I_CLEAR); | |
1da177e4 | 2838 | |
58829e49 | 2839 | /* make sync come back and flush this inode */ |
4c60658e DC |
2840 | if (!(inode->i_state & (I_NEW|I_FREEING))) |
2841 | mark_inode_dirty_sync(inode); | |
1da177e4 | 2842 | } |
f273ab84 | 2843 | spin_unlock(&ip->i_flags_lock); |
1da177e4 LT |
2844 | wake_up(&ip->i_ipin_wait); |
2845 | } | |
2846 | } | |
2847 | ||
2848 | /* | |
2849 | * This is called to wait for the given inode to be unpinned. | |
2850 | * It will sleep until this happens. The caller must have the | |
2851 | * inode locked in at least shared mode so that the buffer cannot | |
2852 | * be subsequently pinned once someone is waiting for it to be | |
2853 | * unpinned. | |
2854 | */ | |
ba0f32d4 | 2855 | STATIC void |
1da177e4 LT |
2856 | xfs_iunpin_wait( |
2857 | xfs_inode_t *ip) | |
2858 | { | |
2859 | xfs_inode_log_item_t *iip; | |
2860 | xfs_lsn_t lsn; | |
2861 | ||
2862 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS)); | |
2863 | ||
2864 | if (atomic_read(&ip->i_pincount) == 0) { | |
2865 | return; | |
2866 | } | |
2867 | ||
2868 | iip = ip->i_itemp; | |
2869 | if (iip && iip->ili_last_lsn) { | |
2870 | lsn = iip->ili_last_lsn; | |
2871 | } else { | |
2872 | lsn = (xfs_lsn_t)0; | |
2873 | } | |
2874 | ||
2875 | /* | |
2876 | * Give the log a push so we don't wait here too long. | |
2877 | */ | |
2878 | xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE); | |
2879 | ||
2880 | wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0)); | |
2881 | } | |
2882 | ||
2883 | ||
2884 | /* | |
2885 | * xfs_iextents_copy() | |
2886 | * | |
2887 | * This is called to copy the REAL extents (as opposed to the delayed | |
2888 | * allocation extents) from the inode into the given buffer. It | |
2889 | * returns the number of bytes copied into the buffer. | |
2890 | * | |
2891 | * If there are no delayed allocation extents, then we can just | |
2892 | * memcpy() the extents into the buffer. Otherwise, we need to | |
2893 | * examine each extent in turn and skip those which are delayed. | |
2894 | */ | |
2895 | int | |
2896 | xfs_iextents_copy( | |
2897 | xfs_inode_t *ip, | |
a6f64d4a | 2898 | xfs_bmbt_rec_t *dp, |
1da177e4 LT |
2899 | int whichfork) |
2900 | { | |
2901 | int copied; | |
1da177e4 LT |
2902 | int i; |
2903 | xfs_ifork_t *ifp; | |
2904 | int nrecs; | |
2905 | xfs_fsblock_t start_block; | |
2906 | ||
2907 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2908 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
2909 | ASSERT(ifp->if_bytes > 0); | |
2910 | ||
2911 | nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3a59c94c | 2912 | XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork); |
1da177e4 LT |
2913 | ASSERT(nrecs > 0); |
2914 | ||
2915 | /* | |
2916 | * There are some delayed allocation extents in the | |
2917 | * inode, so copy the extents one at a time and skip | |
2918 | * the delayed ones. There must be at least one | |
2919 | * non-delayed extent. | |
2920 | */ | |
1da177e4 LT |
2921 | copied = 0; |
2922 | for (i = 0; i < nrecs; i++) { | |
a6f64d4a | 2923 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); |
1da177e4 LT |
2924 | start_block = xfs_bmbt_get_startblock(ep); |
2925 | if (ISNULLSTARTBLOCK(start_block)) { | |
2926 | /* | |
2927 | * It's a delayed allocation extent, so skip it. | |
2928 | */ | |
1da177e4 LT |
2929 | continue; |
2930 | } | |
2931 | ||
2932 | /* Translate to on disk format */ | |
cd8b0a97 CH |
2933 | put_unaligned(cpu_to_be64(ep->l0), &dp->l0); |
2934 | put_unaligned(cpu_to_be64(ep->l1), &dp->l1); | |
a6f64d4a | 2935 | dp++; |
1da177e4 LT |
2936 | copied++; |
2937 | } | |
2938 | ASSERT(copied != 0); | |
a6f64d4a | 2939 | xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip)); |
1da177e4 LT |
2940 | |
2941 | return (copied * (uint)sizeof(xfs_bmbt_rec_t)); | |
2942 | } | |
2943 | ||
2944 | /* | |
2945 | * Each of the following cases stores data into the same region | |
2946 | * of the on-disk inode, so only one of them can be valid at | |
2947 | * any given time. While it is possible to have conflicting formats | |
2948 | * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is | |
2949 | * in EXTENTS format, this can only happen when the fork has | |
2950 | * changed formats after being modified but before being flushed. | |
2951 | * In these cases, the format always takes precedence, because the | |
2952 | * format indicates the current state of the fork. | |
2953 | */ | |
2954 | /*ARGSUSED*/ | |
2955 | STATIC int | |
2956 | xfs_iflush_fork( | |
2957 | xfs_inode_t *ip, | |
2958 | xfs_dinode_t *dip, | |
2959 | xfs_inode_log_item_t *iip, | |
2960 | int whichfork, | |
2961 | xfs_buf_t *bp) | |
2962 | { | |
2963 | char *cp; | |
2964 | xfs_ifork_t *ifp; | |
2965 | xfs_mount_t *mp; | |
2966 | #ifdef XFS_TRANS_DEBUG | |
2967 | int first; | |
2968 | #endif | |
2969 | static const short brootflag[2] = | |
2970 | { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; | |
2971 | static const short dataflag[2] = | |
2972 | { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; | |
2973 | static const short extflag[2] = | |
2974 | { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; | |
2975 | ||
2976 | if (iip == NULL) | |
2977 | return 0; | |
2978 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
2979 | /* | |
2980 | * This can happen if we gave up in iformat in an error path, | |
2981 | * for the attribute fork. | |
2982 | */ | |
2983 | if (ifp == NULL) { | |
2984 | ASSERT(whichfork == XFS_ATTR_FORK); | |
2985 | return 0; | |
2986 | } | |
2987 | cp = XFS_DFORK_PTR(dip, whichfork); | |
2988 | mp = ip->i_mount; | |
2989 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { | |
2990 | case XFS_DINODE_FMT_LOCAL: | |
2991 | if ((iip->ili_format.ilf_fields & dataflag[whichfork]) && | |
2992 | (ifp->if_bytes > 0)) { | |
2993 | ASSERT(ifp->if_u1.if_data != NULL); | |
2994 | ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); | |
2995 | memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); | |
2996 | } | |
1da177e4 LT |
2997 | break; |
2998 | ||
2999 | case XFS_DINODE_FMT_EXTENTS: | |
3000 | ASSERT((ifp->if_flags & XFS_IFEXTENTS) || | |
3001 | !(iip->ili_format.ilf_fields & extflag[whichfork])); | |
4eea22f0 MK |
3002 | ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) || |
3003 | (ifp->if_bytes == 0)); | |
3004 | ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) || | |
3005 | (ifp->if_bytes > 0)); | |
1da177e4 LT |
3006 | if ((iip->ili_format.ilf_fields & extflag[whichfork]) && |
3007 | (ifp->if_bytes > 0)) { | |
3008 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); | |
3009 | (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, | |
3010 | whichfork); | |
3011 | } | |
3012 | break; | |
3013 | ||
3014 | case XFS_DINODE_FMT_BTREE: | |
3015 | if ((iip->ili_format.ilf_fields & brootflag[whichfork]) && | |
3016 | (ifp->if_broot_bytes > 0)) { | |
3017 | ASSERT(ifp->if_broot != NULL); | |
3018 | ASSERT(ifp->if_broot_bytes <= | |
3019 | (XFS_IFORK_SIZE(ip, whichfork) + | |
3020 | XFS_BROOT_SIZE_ADJ)); | |
3021 | xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes, | |
3022 | (xfs_bmdr_block_t *)cp, | |
3023 | XFS_DFORK_SIZE(dip, mp, whichfork)); | |
3024 | } | |
3025 | break; | |
3026 | ||
3027 | case XFS_DINODE_FMT_DEV: | |
3028 | if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) { | |
3029 | ASSERT(whichfork == XFS_DATA_FORK); | |
347d1c01 | 3030 | dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev); |
1da177e4 LT |
3031 | } |
3032 | break; | |
3033 | ||
3034 | case XFS_DINODE_FMT_UUID: | |
3035 | if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) { | |
3036 | ASSERT(whichfork == XFS_DATA_FORK); | |
3037 | memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid, | |
3038 | sizeof(uuid_t)); | |
3039 | } | |
3040 | break; | |
3041 | ||
3042 | default: | |
3043 | ASSERT(0); | |
3044 | break; | |
3045 | } | |
3046 | ||
3047 | return 0; | |
3048 | } | |
3049 | ||
3050 | /* | |
3051 | * xfs_iflush() will write a modified inode's changes out to the | |
3052 | * inode's on disk home. The caller must have the inode lock held | |
3053 | * in at least shared mode and the inode flush semaphore must be | |
3054 | * held as well. The inode lock will still be held upon return from | |
3055 | * the call and the caller is free to unlock it. | |
3056 | * The inode flush lock will be unlocked when the inode reaches the disk. | |
3057 | * The flags indicate how the inode's buffer should be written out. | |
3058 | */ | |
3059 | int | |
3060 | xfs_iflush( | |
3061 | xfs_inode_t *ip, | |
3062 | uint flags) | |
3063 | { | |
3064 | xfs_inode_log_item_t *iip; | |
3065 | xfs_buf_t *bp; | |
3066 | xfs_dinode_t *dip; | |
3067 | xfs_mount_t *mp; | |
3068 | int error; | |
3069 | /* REFERENCED */ | |
1da177e4 LT |
3070 | xfs_inode_t *iq; |
3071 | int clcount; /* count of inodes clustered */ | |
3072 | int bufwasdelwri; | |
da353b0d | 3073 | struct hlist_node *entry; |
1da177e4 | 3074 | enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) }; |
1da177e4 LT |
3075 | |
3076 | XFS_STATS_INC(xs_iflush_count); | |
3077 | ||
3078 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
0d8fee32 | 3079 | ASSERT(issemalocked(&(ip->i_flock))); |
1da177e4 LT |
3080 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
3081 | ip->i_d.di_nextents > ip->i_df.if_ext_max); | |
3082 | ||
3083 | iip = ip->i_itemp; | |
3084 | mp = ip->i_mount; | |
3085 | ||
3086 | /* | |
3087 | * If the inode isn't dirty, then just release the inode | |
3088 | * flush lock and do nothing. | |
3089 | */ | |
3090 | if ((ip->i_update_core == 0) && | |
3091 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3092 | ASSERT((iip != NULL) ? | |
3093 | !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1); | |
3094 | xfs_ifunlock(ip); | |
3095 | return 0; | |
3096 | } | |
3097 | ||
3098 | /* | |
3099 | * We can't flush the inode until it is unpinned, so | |
3100 | * wait for it. We know noone new can pin it, because | |
3101 | * we are holding the inode lock shared and you need | |
3102 | * to hold it exclusively to pin the inode. | |
3103 | */ | |
3104 | xfs_iunpin_wait(ip); | |
3105 | ||
3106 | /* | |
3107 | * This may have been unpinned because the filesystem is shutting | |
3108 | * down forcibly. If that's the case we must not write this inode | |
3109 | * to disk, because the log record didn't make it to disk! | |
3110 | */ | |
3111 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
3112 | ip->i_update_core = 0; | |
3113 | if (iip) | |
3114 | iip->ili_format.ilf_fields = 0; | |
3115 | xfs_ifunlock(ip); | |
3116 | return XFS_ERROR(EIO); | |
3117 | } | |
3118 | ||
3119 | /* | |
3120 | * Get the buffer containing the on-disk inode. | |
3121 | */ | |
b12dd342 NS |
3122 | error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0); |
3123 | if (error) { | |
1da177e4 LT |
3124 | xfs_ifunlock(ip); |
3125 | return error; | |
3126 | } | |
3127 | ||
3128 | /* | |
3129 | * Decide how buffer will be flushed out. This is done before | |
3130 | * the call to xfs_iflush_int because this field is zeroed by it. | |
3131 | */ | |
3132 | if (iip != NULL && iip->ili_format.ilf_fields != 0) { | |
3133 | /* | |
3134 | * Flush out the inode buffer according to the directions | |
3135 | * of the caller. In the cases where the caller has given | |
3136 | * us a choice choose the non-delwri case. This is because | |
3137 | * the inode is in the AIL and we need to get it out soon. | |
3138 | */ | |
3139 | switch (flags) { | |
3140 | case XFS_IFLUSH_SYNC: | |
3141 | case XFS_IFLUSH_DELWRI_ELSE_SYNC: | |
3142 | flags = 0; | |
3143 | break; | |
3144 | case XFS_IFLUSH_ASYNC: | |
3145 | case XFS_IFLUSH_DELWRI_ELSE_ASYNC: | |
3146 | flags = INT_ASYNC; | |
3147 | break; | |
3148 | case XFS_IFLUSH_DELWRI: | |
3149 | flags = INT_DELWRI; | |
3150 | break; | |
3151 | default: | |
3152 | ASSERT(0); | |
3153 | flags = 0; | |
3154 | break; | |
3155 | } | |
3156 | } else { | |
3157 | switch (flags) { | |
3158 | case XFS_IFLUSH_DELWRI_ELSE_SYNC: | |
3159 | case XFS_IFLUSH_DELWRI_ELSE_ASYNC: | |
3160 | case XFS_IFLUSH_DELWRI: | |
3161 | flags = INT_DELWRI; | |
3162 | break; | |
3163 | case XFS_IFLUSH_ASYNC: | |
3164 | flags = INT_ASYNC; | |
3165 | break; | |
3166 | case XFS_IFLUSH_SYNC: | |
3167 | flags = 0; | |
3168 | break; | |
3169 | default: | |
3170 | ASSERT(0); | |
3171 | flags = 0; | |
3172 | break; | |
3173 | } | |
3174 | } | |
3175 | ||
3176 | /* | |
3177 | * First flush out the inode that xfs_iflush was called with. | |
3178 | */ | |
3179 | error = xfs_iflush_int(ip, bp); | |
3180 | if (error) { | |
3181 | goto corrupt_out; | |
3182 | } | |
3183 | ||
3184 | /* | |
3185 | * inode clustering: | |
3186 | * see if other inodes can be gathered into this write | |
3187 | */ | |
da353b0d DC |
3188 | spin_lock(&ip->i_cluster->icl_lock); |
3189 | ip->i_cluster->icl_buf = bp; | |
1da177e4 LT |
3190 | |
3191 | clcount = 0; | |
da353b0d DC |
3192 | hlist_for_each_entry(iq, entry, &ip->i_cluster->icl_inodes, i_cnode) { |
3193 | if (iq == ip) | |
3194 | continue; | |
3195 | ||
1da177e4 LT |
3196 | /* |
3197 | * Do an un-protected check to see if the inode is dirty and | |
3198 | * is a candidate for flushing. These checks will be repeated | |
3199 | * later after the appropriate locks are acquired. | |
3200 | */ | |
3201 | iip = iq->i_itemp; | |
3202 | if ((iq->i_update_core == 0) && | |
3203 | ((iip == NULL) || | |
3204 | !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) && | |
3205 | xfs_ipincount(iq) == 0) { | |
3206 | continue; | |
3207 | } | |
3208 | ||
3209 | /* | |
3210 | * Try to get locks. If any are unavailable, | |
3211 | * then this inode cannot be flushed and is skipped. | |
3212 | */ | |
3213 | ||
3214 | /* get inode locks (just i_lock) */ | |
3215 | if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) { | |
3216 | /* get inode flush lock */ | |
3217 | if (xfs_iflock_nowait(iq)) { | |
3218 | /* check if pinned */ | |
3219 | if (xfs_ipincount(iq) == 0) { | |
3220 | /* arriving here means that | |
3221 | * this inode can be flushed. | |
3222 | * first re-check that it's | |
3223 | * dirty | |
3224 | */ | |
3225 | iip = iq->i_itemp; | |
3226 | if ((iq->i_update_core != 0)|| | |
3227 | ((iip != NULL) && | |
3228 | (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3229 | clcount++; | |
3230 | error = xfs_iflush_int(iq, bp); | |
3231 | if (error) { | |
3232 | xfs_iunlock(iq, | |
3233 | XFS_ILOCK_SHARED); | |
3234 | goto cluster_corrupt_out; | |
3235 | } | |
3236 | } else { | |
3237 | xfs_ifunlock(iq); | |
3238 | } | |
3239 | } else { | |
3240 | xfs_ifunlock(iq); | |
3241 | } | |
3242 | } | |
3243 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
3244 | } | |
3245 | } | |
da353b0d | 3246 | spin_unlock(&ip->i_cluster->icl_lock); |
1da177e4 LT |
3247 | |
3248 | if (clcount) { | |
3249 | XFS_STATS_INC(xs_icluster_flushcnt); | |
3250 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
3251 | } | |
3252 | ||
3253 | /* | |
3254 | * If the buffer is pinned then push on the log so we won't | |
3255 | * get stuck waiting in the write for too long. | |
3256 | */ | |
3257 | if (XFS_BUF_ISPINNED(bp)){ | |
3258 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
3259 | } | |
3260 | ||
3261 | if (flags & INT_DELWRI) { | |
3262 | xfs_bdwrite(mp, bp); | |
3263 | } else if (flags & INT_ASYNC) { | |
3264 | xfs_bawrite(mp, bp); | |
3265 | } else { | |
3266 | error = xfs_bwrite(mp, bp); | |
3267 | } | |
3268 | return error; | |
3269 | ||
3270 | corrupt_out: | |
3271 | xfs_buf_relse(bp); | |
7d04a335 | 3272 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 LT |
3273 | xfs_iflush_abort(ip); |
3274 | /* | |
3275 | * Unlocks the flush lock | |
3276 | */ | |
3277 | return XFS_ERROR(EFSCORRUPTED); | |
3278 | ||
3279 | cluster_corrupt_out: | |
3280 | /* Corruption detected in the clustering loop. Invalidate the | |
3281 | * inode buffer and shut down the filesystem. | |
3282 | */ | |
da353b0d | 3283 | spin_unlock(&ip->i_cluster->icl_lock); |
1da177e4 LT |
3284 | |
3285 | /* | |
3286 | * Clean up the buffer. If it was B_DELWRI, just release it -- | |
3287 | * brelse can handle it with no problems. If not, shut down the | |
3288 | * filesystem before releasing the buffer. | |
3289 | */ | |
3290 | if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) { | |
3291 | xfs_buf_relse(bp); | |
3292 | } | |
3293 | ||
7d04a335 | 3294 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 LT |
3295 | |
3296 | if(!bufwasdelwri) { | |
3297 | /* | |
3298 | * Just like incore_relse: if we have b_iodone functions, | |
3299 | * mark the buffer as an error and call them. Otherwise | |
3300 | * mark it as stale and brelse. | |
3301 | */ | |
3302 | if (XFS_BUF_IODONE_FUNC(bp)) { | |
3303 | XFS_BUF_CLR_BDSTRAT_FUNC(bp); | |
3304 | XFS_BUF_UNDONE(bp); | |
3305 | XFS_BUF_STALE(bp); | |
3306 | XFS_BUF_SHUT(bp); | |
3307 | XFS_BUF_ERROR(bp,EIO); | |
3308 | xfs_biodone(bp); | |
3309 | } else { | |
3310 | XFS_BUF_STALE(bp); | |
3311 | xfs_buf_relse(bp); | |
3312 | } | |
3313 | } | |
3314 | ||
3315 | xfs_iflush_abort(iq); | |
3316 | /* | |
3317 | * Unlocks the flush lock | |
3318 | */ | |
3319 | return XFS_ERROR(EFSCORRUPTED); | |
3320 | } | |
3321 | ||
3322 | ||
3323 | STATIC int | |
3324 | xfs_iflush_int( | |
3325 | xfs_inode_t *ip, | |
3326 | xfs_buf_t *bp) | |
3327 | { | |
3328 | xfs_inode_log_item_t *iip; | |
3329 | xfs_dinode_t *dip; | |
3330 | xfs_mount_t *mp; | |
3331 | #ifdef XFS_TRANS_DEBUG | |
3332 | int first; | |
3333 | #endif | |
1da177e4 LT |
3334 | |
3335 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS)); | |
0d8fee32 | 3336 | ASSERT(issemalocked(&(ip->i_flock))); |
1da177e4 LT |
3337 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
3338 | ip->i_d.di_nextents > ip->i_df.if_ext_max); | |
3339 | ||
3340 | iip = ip->i_itemp; | |
3341 | mp = ip->i_mount; | |
3342 | ||
3343 | ||
3344 | /* | |
3345 | * If the inode isn't dirty, then just release the inode | |
3346 | * flush lock and do nothing. | |
3347 | */ | |
3348 | if ((ip->i_update_core == 0) && | |
3349 | ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) { | |
3350 | xfs_ifunlock(ip); | |
3351 | return 0; | |
3352 | } | |
3353 | ||
3354 | /* set *dip = inode's place in the buffer */ | |
3355 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset); | |
3356 | ||
3357 | /* | |
3358 | * Clear i_update_core before copying out the data. | |
3359 | * This is for coordination with our timestamp updates | |
3360 | * that don't hold the inode lock. They will always | |
3361 | * update the timestamps BEFORE setting i_update_core, | |
3362 | * so if we clear i_update_core after they set it we | |
3363 | * are guaranteed to see their updates to the timestamps. | |
3364 | * I believe that this depends on strongly ordered memory | |
3365 | * semantics, but we have that. We use the SYNCHRONIZE | |
3366 | * macro to make sure that the compiler does not reorder | |
3367 | * the i_update_core access below the data copy below. | |
3368 | */ | |
3369 | ip->i_update_core = 0; | |
3370 | SYNCHRONIZE(); | |
3371 | ||
42fe2b1f CH |
3372 | /* |
3373 | * Make sure to get the latest atime from the Linux inode. | |
3374 | */ | |
3375 | xfs_synchronize_atime(ip); | |
3376 | ||
347d1c01 | 3377 | if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC, |
1da177e4 LT |
3378 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
3379 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3380 | "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
347d1c01 | 3381 | ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip); |
1da177e4 LT |
3382 | goto corrupt_out; |
3383 | } | |
3384 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
3385 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
3386 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3387 | "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
3388 | ip->i_ino, ip, ip->i_d.di_magic); | |
3389 | goto corrupt_out; | |
3390 | } | |
3391 | if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) { | |
3392 | if (XFS_TEST_ERROR( | |
3393 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3394 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
3395 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
3396 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3397 | "xfs_iflush: Bad regular inode %Lu, ptr 0x%p", | |
3398 | ip->i_ino, ip); | |
3399 | goto corrupt_out; | |
3400 | } | |
3401 | } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) { | |
3402 | if (XFS_TEST_ERROR( | |
3403 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3404 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3405 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
3406 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
3407 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3408 | "xfs_iflush: Bad directory inode %Lu, ptr 0x%p", | |
3409 | ip->i_ino, ip); | |
3410 | goto corrupt_out; | |
3411 | } | |
3412 | } | |
3413 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
3414 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
3415 | XFS_RANDOM_IFLUSH_5)) { | |
3416 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3417 | "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p", | |
3418 | ip->i_ino, | |
3419 | ip->i_d.di_nextents + ip->i_d.di_anextents, | |
3420 | ip->i_d.di_nblocks, | |
3421 | ip); | |
3422 | goto corrupt_out; | |
3423 | } | |
3424 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
3425 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
3426 | xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp, | |
3427 | "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
3428 | ip->i_ino, ip->i_d.di_forkoff, ip); | |
3429 | goto corrupt_out; | |
3430 | } | |
3431 | /* | |
3432 | * bump the flush iteration count, used to detect flushes which | |
3433 | * postdate a log record during recovery. | |
3434 | */ | |
3435 | ||
3436 | ip->i_d.di_flushiter++; | |
3437 | ||
3438 | /* | |
3439 | * Copy the dirty parts of the inode into the on-disk | |
3440 | * inode. We always copy out the core of the inode, | |
3441 | * because if the inode is dirty at all the core must | |
3442 | * be. | |
3443 | */ | |
347d1c01 | 3444 | xfs_dinode_to_disk(&dip->di_core, &ip->i_d); |
1da177e4 LT |
3445 | |
3446 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3447 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3448 | ip->i_d.di_flushiter = 0; | |
3449 | ||
3450 | /* | |
3451 | * If this is really an old format inode and the superblock version | |
3452 | * has not been updated to support only new format inodes, then | |
3453 | * convert back to the old inode format. If the superblock version | |
3454 | * has been updated, then make the conversion permanent. | |
3455 | */ | |
3456 | ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 || | |
3457 | XFS_SB_VERSION_HASNLINK(&mp->m_sb)); | |
3458 | if (ip->i_d.di_version == XFS_DINODE_VERSION_1) { | |
3459 | if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) { | |
3460 | /* | |
3461 | * Convert it back. | |
3462 | */ | |
3463 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
347d1c01 | 3464 | dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink); |
1da177e4 LT |
3465 | } else { |
3466 | /* | |
3467 | * The superblock version has already been bumped, | |
3468 | * so just make the conversion to the new inode | |
3469 | * format permanent. | |
3470 | */ | |
3471 | ip->i_d.di_version = XFS_DINODE_VERSION_2; | |
347d1c01 | 3472 | dip->di_core.di_version = XFS_DINODE_VERSION_2; |
1da177e4 LT |
3473 | ip->i_d.di_onlink = 0; |
3474 | dip->di_core.di_onlink = 0; | |
3475 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
3476 | memset(&(dip->di_core.di_pad[0]), 0, | |
3477 | sizeof(dip->di_core.di_pad)); | |
3478 | ASSERT(ip->i_d.di_projid == 0); | |
3479 | } | |
3480 | } | |
3481 | ||
3482 | if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) { | |
3483 | goto corrupt_out; | |
3484 | } | |
3485 | ||
3486 | if (XFS_IFORK_Q(ip)) { | |
3487 | /* | |
3488 | * The only error from xfs_iflush_fork is on the data fork. | |
3489 | */ | |
3490 | (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
3491 | } | |
3492 | xfs_inobp_check(mp, bp); | |
3493 | ||
3494 | /* | |
3495 | * We've recorded everything logged in the inode, so we'd | |
3496 | * like to clear the ilf_fields bits so we don't log and | |
3497 | * flush things unnecessarily. However, we can't stop | |
3498 | * logging all this information until the data we've copied | |
3499 | * into the disk buffer is written to disk. If we did we might | |
3500 | * overwrite the copy of the inode in the log with all the | |
3501 | * data after re-logging only part of it, and in the face of | |
3502 | * a crash we wouldn't have all the data we need to recover. | |
3503 | * | |
3504 | * What we do is move the bits to the ili_last_fields field. | |
3505 | * When logging the inode, these bits are moved back to the | |
3506 | * ilf_fields field. In the xfs_iflush_done() routine we | |
3507 | * clear ili_last_fields, since we know that the information | |
3508 | * those bits represent is permanently on disk. As long as | |
3509 | * the flush completes before the inode is logged again, then | |
3510 | * both ilf_fields and ili_last_fields will be cleared. | |
3511 | * | |
3512 | * We can play with the ilf_fields bits here, because the inode | |
3513 | * lock must be held exclusively in order to set bits there | |
3514 | * and the flush lock protects the ili_last_fields bits. | |
3515 | * Set ili_logged so the flush done | |
3516 | * routine can tell whether or not to look in the AIL. | |
3517 | * Also, store the current LSN of the inode so that we can tell | |
3518 | * whether the item has moved in the AIL from xfs_iflush_done(). | |
3519 | * In order to read the lsn we need the AIL lock, because | |
3520 | * it is a 64 bit value that cannot be read atomically. | |
3521 | */ | |
3522 | if (iip != NULL && iip->ili_format.ilf_fields != 0) { | |
3523 | iip->ili_last_fields = iip->ili_format.ilf_fields; | |
3524 | iip->ili_format.ilf_fields = 0; | |
3525 | iip->ili_logged = 1; | |
3526 | ||
3527 | ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */ | |
287f3dad | 3528 | spin_lock(&mp->m_ail_lock); |
1da177e4 | 3529 | iip->ili_flush_lsn = iip->ili_item.li_lsn; |
287f3dad | 3530 | spin_unlock(&mp->m_ail_lock); |
1da177e4 LT |
3531 | |
3532 | /* | |
3533 | * Attach the function xfs_iflush_done to the inode's | |
3534 | * buffer. This will remove the inode from the AIL | |
3535 | * and unlock the inode's flush lock when the inode is | |
3536 | * completely written to disk. | |
3537 | */ | |
3538 | xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*)) | |
3539 | xfs_iflush_done, (xfs_log_item_t *)iip); | |
3540 | ||
3541 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); | |
3542 | ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL); | |
3543 | } else { | |
3544 | /* | |
3545 | * We're flushing an inode which is not in the AIL and has | |
3546 | * not been logged but has i_update_core set. For this | |
3547 | * case we can use a B_DELWRI flush and immediately drop | |
3548 | * the inode flush lock because we can avoid the whole | |
3549 | * AIL state thing. It's OK to drop the flush lock now, | |
3550 | * because we've already locked the buffer and to do anything | |
3551 | * you really need both. | |
3552 | */ | |
3553 | if (iip != NULL) { | |
3554 | ASSERT(iip->ili_logged == 0); | |
3555 | ASSERT(iip->ili_last_fields == 0); | |
3556 | ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0); | |
3557 | } | |
3558 | xfs_ifunlock(ip); | |
3559 | } | |
3560 | ||
3561 | return 0; | |
3562 | ||
3563 | corrupt_out: | |
3564 | return XFS_ERROR(EFSCORRUPTED); | |
3565 | } | |
3566 | ||
3567 | ||
3568 | /* | |
efa80278 | 3569 | * Flush all inactive inodes in mp. |
1da177e4 | 3570 | */ |
efa80278 | 3571 | void |
1da177e4 | 3572 | xfs_iflush_all( |
efa80278 | 3573 | xfs_mount_t *mp) |
1da177e4 | 3574 | { |
1da177e4 | 3575 | xfs_inode_t *ip; |
67fcaa73 | 3576 | bhv_vnode_t *vp; |
1da177e4 | 3577 | |
efa80278 CH |
3578 | again: |
3579 | XFS_MOUNT_ILOCK(mp); | |
3580 | ip = mp->m_inodes; | |
3581 | if (ip == NULL) | |
3582 | goto out; | |
1da177e4 | 3583 | |
efa80278 CH |
3584 | do { |
3585 | /* Make sure we skip markers inserted by sync */ | |
3586 | if (ip->i_mount == NULL) { | |
3587 | ip = ip->i_mnext; | |
3588 | continue; | |
3589 | } | |
1da177e4 | 3590 | |
efa80278 CH |
3591 | vp = XFS_ITOV_NULL(ip); |
3592 | if (!vp) { | |
1da177e4 | 3593 | XFS_MOUNT_IUNLOCK(mp); |
efa80278 CH |
3594 | xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC); |
3595 | goto again; | |
3596 | } | |
1da177e4 | 3597 | |
efa80278 | 3598 | ASSERT(vn_count(vp) == 0); |
1da177e4 | 3599 | |
efa80278 CH |
3600 | ip = ip->i_mnext; |
3601 | } while (ip != mp->m_inodes); | |
3602 | out: | |
1da177e4 | 3603 | XFS_MOUNT_IUNLOCK(mp); |
1da177e4 LT |
3604 | } |
3605 | ||
1da177e4 LT |
3606 | /* |
3607 | * xfs_iaccess: check accessibility of inode for mode. | |
3608 | */ | |
3609 | int | |
3610 | xfs_iaccess( | |
3611 | xfs_inode_t *ip, | |
3612 | mode_t mode, | |
3613 | cred_t *cr) | |
3614 | { | |
3615 | int error; | |
3616 | mode_t orgmode = mode; | |
ec86dc02 | 3617 | struct inode *inode = vn_to_inode(XFS_ITOV(ip)); |
1da177e4 LT |
3618 | |
3619 | if (mode & S_IWUSR) { | |
3620 | umode_t imode = inode->i_mode; | |
3621 | ||
3622 | if (IS_RDONLY(inode) && | |
3623 | (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode))) | |
3624 | return XFS_ERROR(EROFS); | |
3625 | ||
3626 | if (IS_IMMUTABLE(inode)) | |
3627 | return XFS_ERROR(EACCES); | |
3628 | } | |
3629 | ||
3630 | /* | |
3631 | * If there's an Access Control List it's used instead of | |
3632 | * the mode bits. | |
3633 | */ | |
3634 | if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1) | |
3635 | return error ? XFS_ERROR(error) : 0; | |
3636 | ||
3637 | if (current_fsuid(cr) != ip->i_d.di_uid) { | |
3638 | mode >>= 3; | |
3639 | if (!in_group_p((gid_t)ip->i_d.di_gid)) | |
3640 | mode >>= 3; | |
3641 | } | |
3642 | ||
3643 | /* | |
3644 | * If the DACs are ok we don't need any capability check. | |
3645 | */ | |
3646 | if ((ip->i_d.di_mode & mode) == mode) | |
3647 | return 0; | |
3648 | /* | |
3649 | * Read/write DACs are always overridable. | |
3650 | * Executable DACs are overridable if at least one exec bit is set. | |
3651 | */ | |
3652 | if (!(orgmode & S_IXUSR) || | |
3653 | (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) | |
3654 | if (capable_cred(cr, CAP_DAC_OVERRIDE)) | |
3655 | return 0; | |
3656 | ||
3657 | if ((orgmode == S_IRUSR) || | |
3658 | (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) { | |
3659 | if (capable_cred(cr, CAP_DAC_READ_SEARCH)) | |
3660 | return 0; | |
3661 | #ifdef NOISE | |
3662 | cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode); | |
3663 | #endif /* NOISE */ | |
3664 | return XFS_ERROR(EACCES); | |
3665 | } | |
3666 | return XFS_ERROR(EACCES); | |
3667 | } | |
3668 | ||
3669 | /* | |
3670 | * xfs_iroundup: round up argument to next power of two | |
3671 | */ | |
3672 | uint | |
3673 | xfs_iroundup( | |
3674 | uint v) | |
3675 | { | |
3676 | int i; | |
3677 | uint m; | |
3678 | ||
3679 | if ((v & (v - 1)) == 0) | |
3680 | return v; | |
3681 | ASSERT((v & 0x80000000) == 0); | |
3682 | if ((v & (v + 1)) == 0) | |
3683 | return v + 1; | |
3684 | for (i = 0, m = 1; i < 31; i++, m <<= 1) { | |
3685 | if (v & m) | |
3686 | continue; | |
3687 | v |= m; | |
3688 | if ((v & (v + 1)) == 0) | |
3689 | return v + 1; | |
3690 | } | |
3691 | ASSERT(0); | |
3692 | return( 0 ); | |
3693 | } | |
3694 | ||
1da177e4 LT |
3695 | #ifdef XFS_ILOCK_TRACE |
3696 | ktrace_t *xfs_ilock_trace_buf; | |
3697 | ||
3698 | void | |
3699 | xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra) | |
3700 | { | |
3701 | ktrace_enter(ip->i_lock_trace, | |
3702 | (void *)ip, | |
3703 | (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */ | |
3704 | (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */ | |
3705 | (void *)ra, /* caller of ilock */ | |
3706 | (void *)(unsigned long)current_cpu(), | |
3707 | (void *)(unsigned long)current_pid(), | |
3708 | NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL); | |
3709 | } | |
3710 | #endif | |
4eea22f0 MK |
3711 | |
3712 | /* | |
3713 | * Return a pointer to the extent record at file index idx. | |
3714 | */ | |
a6f64d4a | 3715 | xfs_bmbt_rec_host_t * |
4eea22f0 MK |
3716 | xfs_iext_get_ext( |
3717 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3718 | xfs_extnum_t idx) /* index of target extent */ | |
3719 | { | |
3720 | ASSERT(idx >= 0); | |
0293ce3a MK |
3721 | if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { |
3722 | return ifp->if_u1.if_ext_irec->er_extbuf; | |
3723 | } else if (ifp->if_flags & XFS_IFEXTIREC) { | |
3724 | xfs_ext_irec_t *erp; /* irec pointer */ | |
3725 | int erp_idx = 0; /* irec index */ | |
3726 | xfs_extnum_t page_idx = idx; /* ext index in target list */ | |
3727 | ||
3728 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
3729 | return &erp->er_extbuf[page_idx]; | |
3730 | } else if (ifp->if_bytes) { | |
4eea22f0 MK |
3731 | return &ifp->if_u1.if_extents[idx]; |
3732 | } else { | |
3733 | return NULL; | |
3734 | } | |
3735 | } | |
3736 | ||
3737 | /* | |
3738 | * Insert new item(s) into the extent records for incore inode | |
3739 | * fork 'ifp'. 'count' new items are inserted at index 'idx'. | |
3740 | */ | |
3741 | void | |
3742 | xfs_iext_insert( | |
3743 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3744 | xfs_extnum_t idx, /* starting index of new items */ | |
3745 | xfs_extnum_t count, /* number of inserted items */ | |
3746 | xfs_bmbt_irec_t *new) /* items to insert */ | |
3747 | { | |
4eea22f0 MK |
3748 | xfs_extnum_t i; /* extent record index */ |
3749 | ||
3750 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
3751 | xfs_iext_add(ifp, idx, count); | |
a6f64d4a CH |
3752 | for (i = idx; i < idx + count; i++, new++) |
3753 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new); | |
4eea22f0 MK |
3754 | } |
3755 | ||
3756 | /* | |
3757 | * This is called when the amount of space required for incore file | |
3758 | * extents needs to be increased. The ext_diff parameter stores the | |
3759 | * number of new extents being added and the idx parameter contains | |
3760 | * the extent index where the new extents will be added. If the new | |
3761 | * extents are being appended, then we just need to (re)allocate and | |
3762 | * initialize the space. Otherwise, if the new extents are being | |
3763 | * inserted into the middle of the existing entries, a bit more work | |
3764 | * is required to make room for the new extents to be inserted. The | |
3765 | * caller is responsible for filling in the new extent entries upon | |
3766 | * return. | |
3767 | */ | |
3768 | void | |
3769 | xfs_iext_add( | |
3770 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3771 | xfs_extnum_t idx, /* index to begin adding exts */ | |
c41564b5 | 3772 | int ext_diff) /* number of extents to add */ |
4eea22f0 MK |
3773 | { |
3774 | int byte_diff; /* new bytes being added */ | |
3775 | int new_size; /* size of extents after adding */ | |
3776 | xfs_extnum_t nextents; /* number of extents in file */ | |
3777 | ||
3778 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
3779 | ASSERT((idx >= 0) && (idx <= nextents)); | |
3780 | byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); | |
3781 | new_size = ifp->if_bytes + byte_diff; | |
3782 | /* | |
3783 | * If the new number of extents (nextents + ext_diff) | |
3784 | * fits inside the inode, then continue to use the inline | |
3785 | * extent buffer. | |
3786 | */ | |
3787 | if (nextents + ext_diff <= XFS_INLINE_EXTS) { | |
3788 | if (idx < nextents) { | |
3789 | memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], | |
3790 | &ifp->if_u2.if_inline_ext[idx], | |
3791 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
3792 | memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); | |
3793 | } | |
3794 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; | |
3795 | ifp->if_real_bytes = 0; | |
0293ce3a | 3796 | ifp->if_lastex = nextents + ext_diff; |
4eea22f0 MK |
3797 | } |
3798 | /* | |
3799 | * Otherwise use a linear (direct) extent list. | |
3800 | * If the extents are currently inside the inode, | |
3801 | * xfs_iext_realloc_direct will switch us from | |
3802 | * inline to direct extent allocation mode. | |
3803 | */ | |
0293ce3a | 3804 | else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { |
4eea22f0 MK |
3805 | xfs_iext_realloc_direct(ifp, new_size); |
3806 | if (idx < nextents) { | |
3807 | memmove(&ifp->if_u1.if_extents[idx + ext_diff], | |
3808 | &ifp->if_u1.if_extents[idx], | |
3809 | (nextents - idx) * sizeof(xfs_bmbt_rec_t)); | |
3810 | memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); | |
3811 | } | |
3812 | } | |
0293ce3a MK |
3813 | /* Indirection array */ |
3814 | else { | |
3815 | xfs_ext_irec_t *erp; | |
3816 | int erp_idx = 0; | |
3817 | int page_idx = idx; | |
3818 | ||
3819 | ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); | |
3820 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
3821 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); | |
3822 | } else { | |
3823 | xfs_iext_irec_init(ifp); | |
3824 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3825 | erp = ifp->if_u1.if_ext_irec; | |
3826 | } | |
3827 | /* Extents fit in target extent page */ | |
3828 | if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { | |
3829 | if (page_idx < erp->er_extcount) { | |
3830 | memmove(&erp->er_extbuf[page_idx + ext_diff], | |
3831 | &erp->er_extbuf[page_idx], | |
3832 | (erp->er_extcount - page_idx) * | |
3833 | sizeof(xfs_bmbt_rec_t)); | |
3834 | memset(&erp->er_extbuf[page_idx], 0, byte_diff); | |
3835 | } | |
3836 | erp->er_extcount += ext_diff; | |
3837 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3838 | } | |
3839 | /* Insert a new extent page */ | |
3840 | else if (erp) { | |
3841 | xfs_iext_add_indirect_multi(ifp, | |
3842 | erp_idx, page_idx, ext_diff); | |
3843 | } | |
3844 | /* | |
3845 | * If extent(s) are being appended to the last page in | |
3846 | * the indirection array and the new extent(s) don't fit | |
3847 | * in the page, then erp is NULL and erp_idx is set to | |
3848 | * the next index needed in the indirection array. | |
3849 | */ | |
3850 | else { | |
3851 | int count = ext_diff; | |
3852 | ||
3853 | while (count) { | |
3854 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3855 | erp->er_extcount = count; | |
3856 | count -= MIN(count, (int)XFS_LINEAR_EXTS); | |
3857 | if (count) { | |
3858 | erp_idx++; | |
3859 | } | |
3860 | } | |
3861 | } | |
3862 | } | |
4eea22f0 MK |
3863 | ifp->if_bytes = new_size; |
3864 | } | |
3865 | ||
0293ce3a MK |
3866 | /* |
3867 | * This is called when incore extents are being added to the indirection | |
3868 | * array and the new extents do not fit in the target extent list. The | |
3869 | * erp_idx parameter contains the irec index for the target extent list | |
3870 | * in the indirection array, and the idx parameter contains the extent | |
3871 | * index within the list. The number of extents being added is stored | |
3872 | * in the count parameter. | |
3873 | * | |
3874 | * |-------| |-------| | |
3875 | * | | | | idx - number of extents before idx | |
3876 | * | idx | | count | | |
3877 | * | | | | count - number of extents being inserted at idx | |
3878 | * |-------| |-------| | |
3879 | * | count | | nex2 | nex2 - number of extents after idx + count | |
3880 | * |-------| |-------| | |
3881 | */ | |
3882 | void | |
3883 | xfs_iext_add_indirect_multi( | |
3884 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3885 | int erp_idx, /* target extent irec index */ | |
3886 | xfs_extnum_t idx, /* index within target list */ | |
3887 | int count) /* new extents being added */ | |
3888 | { | |
3889 | int byte_diff; /* new bytes being added */ | |
3890 | xfs_ext_irec_t *erp; /* pointer to irec entry */ | |
3891 | xfs_extnum_t ext_diff; /* number of extents to add */ | |
3892 | xfs_extnum_t ext_cnt; /* new extents still needed */ | |
3893 | xfs_extnum_t nex2; /* extents after idx + count */ | |
3894 | xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ | |
3895 | int nlists; /* number of irec's (lists) */ | |
3896 | ||
3897 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
3898 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
3899 | nex2 = erp->er_extcount - idx; | |
3900 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
3901 | ||
3902 | /* | |
3903 | * Save second part of target extent list | |
3904 | * (all extents past */ | |
3905 | if (nex2) { | |
3906 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
3907 | nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP); | |
3908 | memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); | |
3909 | erp->er_extcount -= nex2; | |
3910 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); | |
3911 | memset(&erp->er_extbuf[idx], 0, byte_diff); | |
3912 | } | |
3913 | ||
3914 | /* | |
3915 | * Add the new extents to the end of the target | |
3916 | * list, then allocate new irec record(s) and | |
3917 | * extent buffer(s) as needed to store the rest | |
3918 | * of the new extents. | |
3919 | */ | |
3920 | ext_cnt = count; | |
3921 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); | |
3922 | if (ext_diff) { | |
3923 | erp->er_extcount += ext_diff; | |
3924 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3925 | ext_cnt -= ext_diff; | |
3926 | } | |
3927 | while (ext_cnt) { | |
3928 | erp_idx++; | |
3929 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3930 | ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); | |
3931 | erp->er_extcount = ext_diff; | |
3932 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); | |
3933 | ext_cnt -= ext_diff; | |
3934 | } | |
3935 | ||
3936 | /* Add nex2 extents back to indirection array */ | |
3937 | if (nex2) { | |
3938 | xfs_extnum_t ext_avail; | |
3939 | int i; | |
3940 | ||
3941 | byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); | |
3942 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; | |
3943 | i = 0; | |
3944 | /* | |
3945 | * If nex2 extents fit in the current page, append | |
3946 | * nex2_ep after the new extents. | |
3947 | */ | |
3948 | if (nex2 <= ext_avail) { | |
3949 | i = erp->er_extcount; | |
3950 | } | |
3951 | /* | |
3952 | * Otherwise, check if space is available in the | |
3953 | * next page. | |
3954 | */ | |
3955 | else if ((erp_idx < nlists - 1) && | |
3956 | (nex2 <= (ext_avail = XFS_LINEAR_EXTS - | |
3957 | ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { | |
3958 | erp_idx++; | |
3959 | erp++; | |
3960 | /* Create a hole for nex2 extents */ | |
3961 | memmove(&erp->er_extbuf[nex2], erp->er_extbuf, | |
3962 | erp->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
3963 | } | |
3964 | /* | |
3965 | * Final choice, create a new extent page for | |
3966 | * nex2 extents. | |
3967 | */ | |
3968 | else { | |
3969 | erp_idx++; | |
3970 | erp = xfs_iext_irec_new(ifp, erp_idx); | |
3971 | } | |
3972 | memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); | |
3973 | kmem_free(nex2_ep, byte_diff); | |
3974 | erp->er_extcount += nex2; | |
3975 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); | |
3976 | } | |
3977 | } | |
3978 | ||
4eea22f0 MK |
3979 | /* |
3980 | * This is called when the amount of space required for incore file | |
3981 | * extents needs to be decreased. The ext_diff parameter stores the | |
3982 | * number of extents to be removed and the idx parameter contains | |
3983 | * the extent index where the extents will be removed from. | |
0293ce3a MK |
3984 | * |
3985 | * If the amount of space needed has decreased below the linear | |
3986 | * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous | |
3987 | * extent array. Otherwise, use kmem_realloc() to adjust the | |
3988 | * size to what is needed. | |
4eea22f0 MK |
3989 | */ |
3990 | void | |
3991 | xfs_iext_remove( | |
3992 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
3993 | xfs_extnum_t idx, /* index to begin removing exts */ | |
3994 | int ext_diff) /* number of extents to remove */ | |
3995 | { | |
3996 | xfs_extnum_t nextents; /* number of extents in file */ | |
3997 | int new_size; /* size of extents after removal */ | |
3998 | ||
3999 | ASSERT(ext_diff > 0); | |
4000 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4001 | new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); | |
4002 | ||
4003 | if (new_size == 0) { | |
4004 | xfs_iext_destroy(ifp); | |
0293ce3a MK |
4005 | } else if (ifp->if_flags & XFS_IFEXTIREC) { |
4006 | xfs_iext_remove_indirect(ifp, idx, ext_diff); | |
4eea22f0 MK |
4007 | } else if (ifp->if_real_bytes) { |
4008 | xfs_iext_remove_direct(ifp, idx, ext_diff); | |
4009 | } else { | |
4010 | xfs_iext_remove_inline(ifp, idx, ext_diff); | |
4011 | } | |
4012 | ifp->if_bytes = new_size; | |
4013 | } | |
4014 | ||
4015 | /* | |
4016 | * This removes ext_diff extents from the inline buffer, beginning | |
4017 | * at extent index idx. | |
4018 | */ | |
4019 | void | |
4020 | xfs_iext_remove_inline( | |
4021 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4022 | xfs_extnum_t idx, /* index to begin removing exts */ | |
4023 | int ext_diff) /* number of extents to remove */ | |
4024 | { | |
4025 | int nextents; /* number of extents in file */ | |
4026 | ||
0293ce3a | 4027 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
4028 | ASSERT(idx < XFS_INLINE_EXTS); |
4029 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4030 | ASSERT(((nextents - ext_diff) > 0) && | |
4031 | (nextents - ext_diff) < XFS_INLINE_EXTS); | |
4032 | ||
4033 | if (idx + ext_diff < nextents) { | |
4034 | memmove(&ifp->if_u2.if_inline_ext[idx], | |
4035 | &ifp->if_u2.if_inline_ext[idx + ext_diff], | |
4036 | (nextents - (idx + ext_diff)) * | |
4037 | sizeof(xfs_bmbt_rec_t)); | |
4038 | memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], | |
4039 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
4040 | } else { | |
4041 | memset(&ifp->if_u2.if_inline_ext[idx], 0, | |
4042 | ext_diff * sizeof(xfs_bmbt_rec_t)); | |
4043 | } | |
4044 | } | |
4045 | ||
4046 | /* | |
4047 | * This removes ext_diff extents from a linear (direct) extent list, | |
4048 | * beginning at extent index idx. If the extents are being removed | |
4049 | * from the end of the list (ie. truncate) then we just need to re- | |
4050 | * allocate the list to remove the extra space. Otherwise, if the | |
4051 | * extents are being removed from the middle of the existing extent | |
4052 | * entries, then we first need to move the extent records beginning | |
4053 | * at idx + ext_diff up in the list to overwrite the records being | |
4054 | * removed, then remove the extra space via kmem_realloc. | |
4055 | */ | |
4056 | void | |
4057 | xfs_iext_remove_direct( | |
4058 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4059 | xfs_extnum_t idx, /* index to begin removing exts */ | |
4060 | int ext_diff) /* number of extents to remove */ | |
4061 | { | |
4062 | xfs_extnum_t nextents; /* number of extents in file */ | |
4063 | int new_size; /* size of extents after removal */ | |
4064 | ||
0293ce3a | 4065 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); |
4eea22f0 MK |
4066 | new_size = ifp->if_bytes - |
4067 | (ext_diff * sizeof(xfs_bmbt_rec_t)); | |
4068 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4069 | ||
4070 | if (new_size == 0) { | |
4071 | xfs_iext_destroy(ifp); | |
4072 | return; | |
4073 | } | |
4074 | /* Move extents up in the list (if needed) */ | |
4075 | if (idx + ext_diff < nextents) { | |
4076 | memmove(&ifp->if_u1.if_extents[idx], | |
4077 | &ifp->if_u1.if_extents[idx + ext_diff], | |
4078 | (nextents - (idx + ext_diff)) * | |
4079 | sizeof(xfs_bmbt_rec_t)); | |
4080 | } | |
4081 | memset(&ifp->if_u1.if_extents[nextents - ext_diff], | |
4082 | 0, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
4083 | /* | |
4084 | * Reallocate the direct extent list. If the extents | |
4085 | * will fit inside the inode then xfs_iext_realloc_direct | |
4086 | * will switch from direct to inline extent allocation | |
4087 | * mode for us. | |
4088 | */ | |
4089 | xfs_iext_realloc_direct(ifp, new_size); | |
4090 | ifp->if_bytes = new_size; | |
4091 | } | |
4092 | ||
0293ce3a MK |
4093 | /* |
4094 | * This is called when incore extents are being removed from the | |
4095 | * indirection array and the extents being removed span multiple extent | |
4096 | * buffers. The idx parameter contains the file extent index where we | |
4097 | * want to begin removing extents, and the count parameter contains | |
4098 | * how many extents need to be removed. | |
4099 | * | |
4100 | * |-------| |-------| | |
4101 | * | nex1 | | | nex1 - number of extents before idx | |
4102 | * |-------| | count | | |
4103 | * | | | | count - number of extents being removed at idx | |
4104 | * | count | |-------| | |
4105 | * | | | nex2 | nex2 - number of extents after idx + count | |
4106 | * |-------| |-------| | |
4107 | */ | |
4108 | void | |
4109 | xfs_iext_remove_indirect( | |
4110 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4111 | xfs_extnum_t idx, /* index to begin removing extents */ | |
4112 | int count) /* number of extents to remove */ | |
4113 | { | |
4114 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4115 | int erp_idx = 0; /* indirection array index */ | |
4116 | xfs_extnum_t ext_cnt; /* extents left to remove */ | |
4117 | xfs_extnum_t ext_diff; /* extents to remove in current list */ | |
4118 | xfs_extnum_t nex1; /* number of extents before idx */ | |
4119 | xfs_extnum_t nex2; /* extents after idx + count */ | |
c41564b5 | 4120 | int nlists; /* entries in indirection array */ |
0293ce3a MK |
4121 | int page_idx = idx; /* index in target extent list */ |
4122 | ||
4123 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4124 | erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); | |
4125 | ASSERT(erp != NULL); | |
4126 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4127 | nex1 = page_idx; | |
4128 | ext_cnt = count; | |
4129 | while (ext_cnt) { | |
4130 | nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); | |
4131 | ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); | |
4132 | /* | |
4133 | * Check for deletion of entire list; | |
4134 | * xfs_iext_irec_remove() updates extent offsets. | |
4135 | */ | |
4136 | if (ext_diff == erp->er_extcount) { | |
4137 | xfs_iext_irec_remove(ifp, erp_idx); | |
4138 | ext_cnt -= ext_diff; | |
4139 | nex1 = 0; | |
4140 | if (ext_cnt) { | |
4141 | ASSERT(erp_idx < ifp->if_real_bytes / | |
4142 | XFS_IEXT_BUFSZ); | |
4143 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4144 | nex1 = 0; | |
4145 | continue; | |
4146 | } else { | |
4147 | break; | |
4148 | } | |
4149 | } | |
4150 | /* Move extents up (if needed) */ | |
4151 | if (nex2) { | |
4152 | memmove(&erp->er_extbuf[nex1], | |
4153 | &erp->er_extbuf[nex1 + ext_diff], | |
4154 | nex2 * sizeof(xfs_bmbt_rec_t)); | |
4155 | } | |
4156 | /* Zero out rest of page */ | |
4157 | memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - | |
4158 | ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); | |
4159 | /* Update remaining counters */ | |
4160 | erp->er_extcount -= ext_diff; | |
4161 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); | |
4162 | ext_cnt -= ext_diff; | |
4163 | nex1 = 0; | |
4164 | erp_idx++; | |
4165 | erp++; | |
4166 | } | |
4167 | ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); | |
4168 | xfs_iext_irec_compact(ifp); | |
4169 | } | |
4170 | ||
4eea22f0 MK |
4171 | /* |
4172 | * Create, destroy, or resize a linear (direct) block of extents. | |
4173 | */ | |
4174 | void | |
4175 | xfs_iext_realloc_direct( | |
4176 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4177 | int new_size) /* new size of extents */ | |
4178 | { | |
4179 | int rnew_size; /* real new size of extents */ | |
4180 | ||
4181 | rnew_size = new_size; | |
4182 | ||
0293ce3a MK |
4183 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || |
4184 | ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && | |
4185 | (new_size != ifp->if_real_bytes))); | |
4186 | ||
4eea22f0 MK |
4187 | /* Free extent records */ |
4188 | if (new_size == 0) { | |
4189 | xfs_iext_destroy(ifp); | |
4190 | } | |
4191 | /* Resize direct extent list and zero any new bytes */ | |
4192 | else if (ifp->if_real_bytes) { | |
4193 | /* Check if extents will fit inside the inode */ | |
4194 | if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { | |
4195 | xfs_iext_direct_to_inline(ifp, new_size / | |
4196 | (uint)sizeof(xfs_bmbt_rec_t)); | |
4197 | ifp->if_bytes = new_size; | |
4198 | return; | |
4199 | } | |
16a087d8 | 4200 | if (!is_power_of_2(new_size)){ |
4eea22f0 MK |
4201 | rnew_size = xfs_iroundup(new_size); |
4202 | } | |
4203 | if (rnew_size != ifp->if_real_bytes) { | |
a6f64d4a | 4204 | ifp->if_u1.if_extents = |
4eea22f0 MK |
4205 | kmem_realloc(ifp->if_u1.if_extents, |
4206 | rnew_size, | |
4207 | ifp->if_real_bytes, | |
4208 | KM_SLEEP); | |
4209 | } | |
4210 | if (rnew_size > ifp->if_real_bytes) { | |
4211 | memset(&ifp->if_u1.if_extents[ifp->if_bytes / | |
4212 | (uint)sizeof(xfs_bmbt_rec_t)], 0, | |
4213 | rnew_size - ifp->if_real_bytes); | |
4214 | } | |
4215 | } | |
4216 | /* | |
4217 | * Switch from the inline extent buffer to a direct | |
4218 | * extent list. Be sure to include the inline extent | |
4219 | * bytes in new_size. | |
4220 | */ | |
4221 | else { | |
4222 | new_size += ifp->if_bytes; | |
16a087d8 | 4223 | if (!is_power_of_2(new_size)) { |
4eea22f0 MK |
4224 | rnew_size = xfs_iroundup(new_size); |
4225 | } | |
4226 | xfs_iext_inline_to_direct(ifp, rnew_size); | |
4227 | } | |
4228 | ifp->if_real_bytes = rnew_size; | |
4229 | ifp->if_bytes = new_size; | |
4230 | } | |
4231 | ||
4232 | /* | |
4233 | * Switch from linear (direct) extent records to inline buffer. | |
4234 | */ | |
4235 | void | |
4236 | xfs_iext_direct_to_inline( | |
4237 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4238 | xfs_extnum_t nextents) /* number of extents in file */ | |
4239 | { | |
4240 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
4241 | ASSERT(nextents <= XFS_INLINE_EXTS); | |
4242 | /* | |
4243 | * The inline buffer was zeroed when we switched | |
4244 | * from inline to direct extent allocation mode, | |
4245 | * so we don't need to clear it here. | |
4246 | */ | |
4247 | memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, | |
4248 | nextents * sizeof(xfs_bmbt_rec_t)); | |
fe6c1e72 | 4249 | kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes); |
4eea22f0 MK |
4250 | ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; |
4251 | ifp->if_real_bytes = 0; | |
4252 | } | |
4253 | ||
4254 | /* | |
4255 | * Switch from inline buffer to linear (direct) extent records. | |
4256 | * new_size should already be rounded up to the next power of 2 | |
4257 | * by the caller (when appropriate), so use new_size as it is. | |
4258 | * However, since new_size may be rounded up, we can't update | |
4259 | * if_bytes here. It is the caller's responsibility to update | |
4260 | * if_bytes upon return. | |
4261 | */ | |
4262 | void | |
4263 | xfs_iext_inline_to_direct( | |
4264 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4265 | int new_size) /* number of extents in file */ | |
4266 | { | |
a6f64d4a | 4267 | ifp->if_u1.if_extents = kmem_alloc(new_size, KM_SLEEP); |
4eea22f0 MK |
4268 | memset(ifp->if_u1.if_extents, 0, new_size); |
4269 | if (ifp->if_bytes) { | |
4270 | memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, | |
4271 | ifp->if_bytes); | |
4272 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
4273 | sizeof(xfs_bmbt_rec_t)); | |
4274 | } | |
4275 | ifp->if_real_bytes = new_size; | |
4276 | } | |
4277 | ||
0293ce3a MK |
4278 | /* |
4279 | * Resize an extent indirection array to new_size bytes. | |
4280 | */ | |
4281 | void | |
4282 | xfs_iext_realloc_indirect( | |
4283 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4284 | int new_size) /* new indirection array size */ | |
4285 | { | |
4286 | int nlists; /* number of irec's (ex lists) */ | |
4287 | int size; /* current indirection array size */ | |
4288 | ||
4289 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4290 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4291 | size = nlists * sizeof(xfs_ext_irec_t); | |
4292 | ASSERT(ifp->if_real_bytes); | |
4293 | ASSERT((new_size >= 0) && (new_size != size)); | |
4294 | if (new_size == 0) { | |
4295 | xfs_iext_destroy(ifp); | |
4296 | } else { | |
4297 | ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) | |
4298 | kmem_realloc(ifp->if_u1.if_ext_irec, | |
4299 | new_size, size, KM_SLEEP); | |
4300 | } | |
4301 | } | |
4302 | ||
4303 | /* | |
4304 | * Switch from indirection array to linear (direct) extent allocations. | |
4305 | */ | |
4306 | void | |
4307 | xfs_iext_indirect_to_direct( | |
4308 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4309 | { | |
a6f64d4a | 4310 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ |
0293ce3a MK |
4311 | xfs_extnum_t nextents; /* number of extents in file */ |
4312 | int size; /* size of file extents */ | |
4313 | ||
4314 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4315 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4316 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
4317 | size = nextents * sizeof(xfs_bmbt_rec_t); | |
4318 | ||
4319 | xfs_iext_irec_compact_full(ifp); | |
4320 | ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); | |
4321 | ||
4322 | ep = ifp->if_u1.if_ext_irec->er_extbuf; | |
4323 | kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t)); | |
4324 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
4325 | ifp->if_u1.if_extents = ep; | |
4326 | ifp->if_bytes = size; | |
4327 | if (nextents < XFS_LINEAR_EXTS) { | |
4328 | xfs_iext_realloc_direct(ifp, size); | |
4329 | } | |
4330 | } | |
4331 | ||
4eea22f0 MK |
4332 | /* |
4333 | * Free incore file extents. | |
4334 | */ | |
4335 | void | |
4336 | xfs_iext_destroy( | |
4337 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4338 | { | |
0293ce3a MK |
4339 | if (ifp->if_flags & XFS_IFEXTIREC) { |
4340 | int erp_idx; | |
4341 | int nlists; | |
4342 | ||
4343 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4344 | for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { | |
4345 | xfs_iext_irec_remove(ifp, erp_idx); | |
4346 | } | |
4347 | ifp->if_flags &= ~XFS_IFEXTIREC; | |
4348 | } else if (ifp->if_real_bytes) { | |
4eea22f0 MK |
4349 | kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes); |
4350 | } else if (ifp->if_bytes) { | |
4351 | memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * | |
4352 | sizeof(xfs_bmbt_rec_t)); | |
4353 | } | |
4354 | ifp->if_u1.if_extents = NULL; | |
4355 | ifp->if_real_bytes = 0; | |
4356 | ifp->if_bytes = 0; | |
4357 | } | |
0293ce3a | 4358 | |
8867bc9b MK |
4359 | /* |
4360 | * Return a pointer to the extent record for file system block bno. | |
4361 | */ | |
a6f64d4a | 4362 | xfs_bmbt_rec_host_t * /* pointer to found extent record */ |
8867bc9b MK |
4363 | xfs_iext_bno_to_ext( |
4364 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4365 | xfs_fileoff_t bno, /* block number to search for */ | |
4366 | xfs_extnum_t *idxp) /* index of target extent */ | |
4367 | { | |
a6f64d4a | 4368 | xfs_bmbt_rec_host_t *base; /* pointer to first extent */ |
8867bc9b | 4369 | xfs_filblks_t blockcount = 0; /* number of blocks in extent */ |
a6f64d4a | 4370 | xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */ |
8867bc9b | 4371 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ |
c41564b5 | 4372 | int high; /* upper boundary in search */ |
8867bc9b | 4373 | xfs_extnum_t idx = 0; /* index of target extent */ |
c41564b5 | 4374 | int low; /* lower boundary in search */ |
8867bc9b MK |
4375 | xfs_extnum_t nextents; /* number of file extents */ |
4376 | xfs_fileoff_t startoff = 0; /* start offset of extent */ | |
4377 | ||
4378 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4379 | if (nextents == 0) { | |
4380 | *idxp = 0; | |
4381 | return NULL; | |
4382 | } | |
4383 | low = 0; | |
4384 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4385 | /* Find target extent list */ | |
4386 | int erp_idx = 0; | |
4387 | erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); | |
4388 | base = erp->er_extbuf; | |
4389 | high = erp->er_extcount - 1; | |
4390 | } else { | |
4391 | base = ifp->if_u1.if_extents; | |
4392 | high = nextents - 1; | |
4393 | } | |
4394 | /* Binary search extent records */ | |
4395 | while (low <= high) { | |
4396 | idx = (low + high) >> 1; | |
4397 | ep = base + idx; | |
4398 | startoff = xfs_bmbt_get_startoff(ep); | |
4399 | blockcount = xfs_bmbt_get_blockcount(ep); | |
4400 | if (bno < startoff) { | |
4401 | high = idx - 1; | |
4402 | } else if (bno >= startoff + blockcount) { | |
4403 | low = idx + 1; | |
4404 | } else { | |
4405 | /* Convert back to file-based extent index */ | |
4406 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4407 | idx += erp->er_extoff; | |
4408 | } | |
4409 | *idxp = idx; | |
4410 | return ep; | |
4411 | } | |
4412 | } | |
4413 | /* Convert back to file-based extent index */ | |
4414 | if (ifp->if_flags & XFS_IFEXTIREC) { | |
4415 | idx += erp->er_extoff; | |
4416 | } | |
4417 | if (bno >= startoff + blockcount) { | |
4418 | if (++idx == nextents) { | |
4419 | ep = NULL; | |
4420 | } else { | |
4421 | ep = xfs_iext_get_ext(ifp, idx); | |
4422 | } | |
4423 | } | |
4424 | *idxp = idx; | |
4425 | return ep; | |
4426 | } | |
4427 | ||
0293ce3a MK |
4428 | /* |
4429 | * Return a pointer to the indirection array entry containing the | |
4430 | * extent record for filesystem block bno. Store the index of the | |
4431 | * target irec in *erp_idxp. | |
4432 | */ | |
8867bc9b | 4433 | xfs_ext_irec_t * /* pointer to found extent record */ |
0293ce3a MK |
4434 | xfs_iext_bno_to_irec( |
4435 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4436 | xfs_fileoff_t bno, /* block number to search for */ | |
4437 | int *erp_idxp) /* irec index of target ext list */ | |
4438 | { | |
4439 | xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ | |
4440 | xfs_ext_irec_t *erp_next; /* next indirection array entry */ | |
8867bc9b | 4441 | int erp_idx; /* indirection array index */ |
0293ce3a MK |
4442 | int nlists; /* number of extent irec's (lists) */ |
4443 | int high; /* binary search upper limit */ | |
4444 | int low; /* binary search lower limit */ | |
4445 | ||
4446 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4447 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4448 | erp_idx = 0; | |
4449 | low = 0; | |
4450 | high = nlists - 1; | |
4451 | while (low <= high) { | |
4452 | erp_idx = (low + high) >> 1; | |
4453 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4454 | erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; | |
4455 | if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { | |
4456 | high = erp_idx - 1; | |
4457 | } else if (erp_next && bno >= | |
4458 | xfs_bmbt_get_startoff(erp_next->er_extbuf)) { | |
4459 | low = erp_idx + 1; | |
4460 | } else { | |
4461 | break; | |
4462 | } | |
4463 | } | |
4464 | *erp_idxp = erp_idx; | |
4465 | return erp; | |
4466 | } | |
4467 | ||
4468 | /* | |
4469 | * Return a pointer to the indirection array entry containing the | |
4470 | * extent record at file extent index *idxp. Store the index of the | |
4471 | * target irec in *erp_idxp and store the page index of the target | |
4472 | * extent record in *idxp. | |
4473 | */ | |
4474 | xfs_ext_irec_t * | |
4475 | xfs_iext_idx_to_irec( | |
4476 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4477 | xfs_extnum_t *idxp, /* extent index (file -> page) */ | |
4478 | int *erp_idxp, /* pointer to target irec */ | |
4479 | int realloc) /* new bytes were just added */ | |
4480 | { | |
4481 | xfs_ext_irec_t *prev; /* pointer to previous irec */ | |
4482 | xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ | |
4483 | int erp_idx; /* indirection array index */ | |
4484 | int nlists; /* number of irec's (ex lists) */ | |
4485 | int high; /* binary search upper limit */ | |
4486 | int low; /* binary search lower limit */ | |
4487 | xfs_extnum_t page_idx = *idxp; /* extent index in target list */ | |
4488 | ||
4489 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4490 | ASSERT(page_idx >= 0 && page_idx <= | |
4491 | ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)); | |
4492 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4493 | erp_idx = 0; | |
4494 | low = 0; | |
4495 | high = nlists - 1; | |
4496 | ||
4497 | /* Binary search extent irec's */ | |
4498 | while (low <= high) { | |
4499 | erp_idx = (low + high) >> 1; | |
4500 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4501 | prev = erp_idx > 0 ? erp - 1 : NULL; | |
4502 | if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && | |
4503 | realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { | |
4504 | high = erp_idx - 1; | |
4505 | } else if (page_idx > erp->er_extoff + erp->er_extcount || | |
4506 | (page_idx == erp->er_extoff + erp->er_extcount && | |
4507 | !realloc)) { | |
4508 | low = erp_idx + 1; | |
4509 | } else if (page_idx == erp->er_extoff + erp->er_extcount && | |
4510 | erp->er_extcount == XFS_LINEAR_EXTS) { | |
4511 | ASSERT(realloc); | |
4512 | page_idx = 0; | |
4513 | erp_idx++; | |
4514 | erp = erp_idx < nlists ? erp + 1 : NULL; | |
4515 | break; | |
4516 | } else { | |
4517 | page_idx -= erp->er_extoff; | |
4518 | break; | |
4519 | } | |
4520 | } | |
4521 | *idxp = page_idx; | |
4522 | *erp_idxp = erp_idx; | |
4523 | return(erp); | |
4524 | } | |
4525 | ||
4526 | /* | |
4527 | * Allocate and initialize an indirection array once the space needed | |
4528 | * for incore extents increases above XFS_IEXT_BUFSZ. | |
4529 | */ | |
4530 | void | |
4531 | xfs_iext_irec_init( | |
4532 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4533 | { | |
4534 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4535 | xfs_extnum_t nextents; /* number of extents in file */ | |
4536 | ||
4537 | ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); | |
4538 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4539 | ASSERT(nextents <= XFS_LINEAR_EXTS); | |
4540 | ||
4541 | erp = (xfs_ext_irec_t *) | |
4542 | kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP); | |
4543 | ||
4544 | if (nextents == 0) { | |
a6f64d4a | 4545 | ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP); |
0293ce3a MK |
4546 | } else if (!ifp->if_real_bytes) { |
4547 | xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); | |
4548 | } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { | |
4549 | xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); | |
4550 | } | |
4551 | erp->er_extbuf = ifp->if_u1.if_extents; | |
4552 | erp->er_extcount = nextents; | |
4553 | erp->er_extoff = 0; | |
4554 | ||
4555 | ifp->if_flags |= XFS_IFEXTIREC; | |
4556 | ifp->if_real_bytes = XFS_IEXT_BUFSZ; | |
4557 | ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); | |
4558 | ifp->if_u1.if_ext_irec = erp; | |
4559 | ||
4560 | return; | |
4561 | } | |
4562 | ||
4563 | /* | |
4564 | * Allocate and initialize a new entry in the indirection array. | |
4565 | */ | |
4566 | xfs_ext_irec_t * | |
4567 | xfs_iext_irec_new( | |
4568 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4569 | int erp_idx) /* index for new irec */ | |
4570 | { | |
4571 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4572 | int i; /* loop counter */ | |
4573 | int nlists; /* number of irec's (ex lists) */ | |
4574 | ||
4575 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4576 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4577 | ||
4578 | /* Resize indirection array */ | |
4579 | xfs_iext_realloc_indirect(ifp, ++nlists * | |
4580 | sizeof(xfs_ext_irec_t)); | |
4581 | /* | |
4582 | * Move records down in the array so the | |
4583 | * new page can use erp_idx. | |
4584 | */ | |
4585 | erp = ifp->if_u1.if_ext_irec; | |
4586 | for (i = nlists - 1; i > erp_idx; i--) { | |
4587 | memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); | |
4588 | } | |
4589 | ASSERT(i == erp_idx); | |
4590 | ||
4591 | /* Initialize new extent record */ | |
4592 | erp = ifp->if_u1.if_ext_irec; | |
a6f64d4a | 4593 | erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP); |
0293ce3a MK |
4594 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; |
4595 | memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); | |
4596 | erp[erp_idx].er_extcount = 0; | |
4597 | erp[erp_idx].er_extoff = erp_idx > 0 ? | |
4598 | erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; | |
4599 | return (&erp[erp_idx]); | |
4600 | } | |
4601 | ||
4602 | /* | |
4603 | * Remove a record from the indirection array. | |
4604 | */ | |
4605 | void | |
4606 | xfs_iext_irec_remove( | |
4607 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4608 | int erp_idx) /* irec index to remove */ | |
4609 | { | |
4610 | xfs_ext_irec_t *erp; /* indirection array pointer */ | |
4611 | int i; /* loop counter */ | |
4612 | int nlists; /* number of irec's (ex lists) */ | |
4613 | ||
4614 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4615 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4616 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4617 | if (erp->er_extbuf) { | |
4618 | xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, | |
4619 | -erp->er_extcount); | |
4620 | kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ); | |
4621 | } | |
4622 | /* Compact extent records */ | |
4623 | erp = ifp->if_u1.if_ext_irec; | |
4624 | for (i = erp_idx; i < nlists - 1; i++) { | |
4625 | memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); | |
4626 | } | |
4627 | /* | |
4628 | * Manually free the last extent record from the indirection | |
4629 | * array. A call to xfs_iext_realloc_indirect() with a size | |
4630 | * of zero would result in a call to xfs_iext_destroy() which | |
4631 | * would in turn call this function again, creating a nasty | |
4632 | * infinite loop. | |
4633 | */ | |
4634 | if (--nlists) { | |
4635 | xfs_iext_realloc_indirect(ifp, | |
4636 | nlists * sizeof(xfs_ext_irec_t)); | |
4637 | } else { | |
4638 | kmem_free(ifp->if_u1.if_ext_irec, | |
4639 | sizeof(xfs_ext_irec_t)); | |
4640 | } | |
4641 | ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; | |
4642 | } | |
4643 | ||
4644 | /* | |
4645 | * This is called to clean up large amounts of unused memory allocated | |
4646 | * by the indirection array. Before compacting anything though, verify | |
4647 | * that the indirection array is still needed and switch back to the | |
4648 | * linear extent list (or even the inline buffer) if possible. The | |
4649 | * compaction policy is as follows: | |
4650 | * | |
4651 | * Full Compaction: Extents fit into a single page (or inline buffer) | |
4652 | * Full Compaction: Extents occupy less than 10% of allocated space | |
4653 | * Partial Compaction: Extents occupy > 10% and < 50% of allocated space | |
4654 | * No Compaction: Extents occupy at least 50% of allocated space | |
4655 | */ | |
4656 | void | |
4657 | xfs_iext_irec_compact( | |
4658 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4659 | { | |
4660 | xfs_extnum_t nextents; /* number of extents in file */ | |
4661 | int nlists; /* number of irec's (ex lists) */ | |
4662 | ||
4663 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4664 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4665 | nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); | |
4666 | ||
4667 | if (nextents == 0) { | |
4668 | xfs_iext_destroy(ifp); | |
4669 | } else if (nextents <= XFS_INLINE_EXTS) { | |
4670 | xfs_iext_indirect_to_direct(ifp); | |
4671 | xfs_iext_direct_to_inline(ifp, nextents); | |
4672 | } else if (nextents <= XFS_LINEAR_EXTS) { | |
4673 | xfs_iext_indirect_to_direct(ifp); | |
4674 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) { | |
4675 | xfs_iext_irec_compact_full(ifp); | |
4676 | } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { | |
4677 | xfs_iext_irec_compact_pages(ifp); | |
4678 | } | |
4679 | } | |
4680 | ||
4681 | /* | |
4682 | * Combine extents from neighboring extent pages. | |
4683 | */ | |
4684 | void | |
4685 | xfs_iext_irec_compact_pages( | |
4686 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4687 | { | |
4688 | xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ | |
4689 | int erp_idx = 0; /* indirection array index */ | |
4690 | int nlists; /* number of irec's (ex lists) */ | |
4691 | ||
4692 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4693 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4694 | while (erp_idx < nlists - 1) { | |
4695 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4696 | erp_next = erp + 1; | |
4697 | if (erp_next->er_extcount <= | |
4698 | (XFS_LINEAR_EXTS - erp->er_extcount)) { | |
4699 | memmove(&erp->er_extbuf[erp->er_extcount], | |
4700 | erp_next->er_extbuf, erp_next->er_extcount * | |
4701 | sizeof(xfs_bmbt_rec_t)); | |
4702 | erp->er_extcount += erp_next->er_extcount; | |
4703 | /* | |
4704 | * Free page before removing extent record | |
4705 | * so er_extoffs don't get modified in | |
4706 | * xfs_iext_irec_remove. | |
4707 | */ | |
4708 | kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ); | |
4709 | erp_next->er_extbuf = NULL; | |
4710 | xfs_iext_irec_remove(ifp, erp_idx + 1); | |
4711 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4712 | } else { | |
4713 | erp_idx++; | |
4714 | } | |
4715 | } | |
4716 | } | |
4717 | ||
4718 | /* | |
4719 | * Fully compact the extent records managed by the indirection array. | |
4720 | */ | |
4721 | void | |
4722 | xfs_iext_irec_compact_full( | |
4723 | xfs_ifork_t *ifp) /* inode fork pointer */ | |
4724 | { | |
a6f64d4a | 4725 | xfs_bmbt_rec_host_t *ep, *ep_next; /* extent record pointers */ |
0293ce3a MK |
4726 | xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */ |
4727 | int erp_idx = 0; /* extent irec index */ | |
4728 | int ext_avail; /* empty entries in ex list */ | |
4729 | int ext_diff; /* number of exts to add */ | |
4730 | int nlists; /* number of irec's (ex lists) */ | |
4731 | ||
4732 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4733 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4734 | erp = ifp->if_u1.if_ext_irec; | |
4735 | ep = &erp->er_extbuf[erp->er_extcount]; | |
4736 | erp_next = erp + 1; | |
4737 | ep_next = erp_next->er_extbuf; | |
4738 | while (erp_idx < nlists - 1) { | |
4739 | ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; | |
4740 | ext_diff = MIN(ext_avail, erp_next->er_extcount); | |
4741 | memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t)); | |
4742 | erp->er_extcount += ext_diff; | |
4743 | erp_next->er_extcount -= ext_diff; | |
4744 | /* Remove next page */ | |
4745 | if (erp_next->er_extcount == 0) { | |
4746 | /* | |
4747 | * Free page before removing extent record | |
4748 | * so er_extoffs don't get modified in | |
4749 | * xfs_iext_irec_remove. | |
4750 | */ | |
4751 | kmem_free(erp_next->er_extbuf, | |
4752 | erp_next->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
4753 | erp_next->er_extbuf = NULL; | |
4754 | xfs_iext_irec_remove(ifp, erp_idx + 1); | |
4755 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4756 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4757 | /* Update next page */ | |
4758 | } else { | |
4759 | /* Move rest of page up to become next new page */ | |
4760 | memmove(erp_next->er_extbuf, ep_next, | |
4761 | erp_next->er_extcount * sizeof(xfs_bmbt_rec_t)); | |
4762 | ep_next = erp_next->er_extbuf; | |
4763 | memset(&ep_next[erp_next->er_extcount], 0, | |
4764 | (XFS_LINEAR_EXTS - erp_next->er_extcount) * | |
4765 | sizeof(xfs_bmbt_rec_t)); | |
4766 | } | |
4767 | if (erp->er_extcount == XFS_LINEAR_EXTS) { | |
4768 | erp_idx++; | |
4769 | if (erp_idx < nlists) | |
4770 | erp = &ifp->if_u1.if_ext_irec[erp_idx]; | |
4771 | else | |
4772 | break; | |
4773 | } | |
4774 | ep = &erp->er_extbuf[erp->er_extcount]; | |
4775 | erp_next = erp + 1; | |
4776 | ep_next = erp_next->er_extbuf; | |
4777 | } | |
4778 | } | |
4779 | ||
4780 | /* | |
4781 | * This is called to update the er_extoff field in the indirection | |
4782 | * array when extents have been added or removed from one of the | |
4783 | * extent lists. erp_idx contains the irec index to begin updating | |
4784 | * at and ext_diff contains the number of extents that were added | |
4785 | * or removed. | |
4786 | */ | |
4787 | void | |
4788 | xfs_iext_irec_update_extoffs( | |
4789 | xfs_ifork_t *ifp, /* inode fork pointer */ | |
4790 | int erp_idx, /* irec index to update */ | |
4791 | int ext_diff) /* number of new extents */ | |
4792 | { | |
4793 | int i; /* loop counter */ | |
4794 | int nlists; /* number of irec's (ex lists */ | |
4795 | ||
4796 | ASSERT(ifp->if_flags & XFS_IFEXTIREC); | |
4797 | nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; | |
4798 | for (i = erp_idx; i < nlists; i++) { | |
4799 | ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; | |
4800 | } | |
4801 | } |