]>
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" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
f5ea1100 | 26 | #include "xfs_mount.h" |
3ab78df2 | 27 | #include "xfs_defer.h" |
57062787 | 28 | #include "xfs_da_format.h" |
a844f451 | 29 | #include "xfs_da_btree.h" |
2b9ab5ab | 30 | #include "xfs_dir2.h" |
1da177e4 | 31 | #include "xfs_inode.h" |
a844f451 | 32 | #include "xfs_btree.h" |
239880ef | 33 | #include "xfs_trans.h" |
a844f451 | 34 | #include "xfs_inode_item.h" |
1da177e4 LT |
35 | #include "xfs_extfree_item.h" |
36 | #include "xfs_alloc.h" | |
37 | #include "xfs_bmap.h" | |
68988114 | 38 | #include "xfs_bmap_util.h" |
a4fbe6ab | 39 | #include "xfs_bmap_btree.h" |
1da177e4 LT |
40 | #include "xfs_rtalloc.h" |
41 | #include "xfs_error.h" | |
1da177e4 LT |
42 | #include "xfs_quota.h" |
43 | #include "xfs_trans_space.h" | |
44 | #include "xfs_buf_item.h" | |
0b1b213f | 45 | #include "xfs_trace.h" |
19de7351 | 46 | #include "xfs_symlink.h" |
a4fbe6ab | 47 | #include "xfs_attr_leaf.h" |
a4fbe6ab | 48 | #include "xfs_filestream.h" |
340785cc | 49 | #include "xfs_rmap.h" |
3fd129b6 | 50 | #include "xfs_ag_resv.h" |
62aab20f | 51 | #include "xfs_refcount.h" |
fd26a880 | 52 | #include "xfs_rmap_btree.h" |
974ae922 | 53 | #include "xfs_icache.h" |
1da177e4 LT |
54 | |
55 | ||
1da177e4 LT |
56 | kmem_zone_t *xfs_bmap_free_item_zone; |
57 | ||
58 | /* | |
9e5987a7 | 59 | * Miscellaneous helper functions |
1da177e4 LT |
60 | */ |
61 | ||
1da177e4 | 62 | /* |
9e5987a7 DC |
63 | * Compute and fill in the value of the maximum depth of a bmap btree |
64 | * in this filesystem. Done once, during mount. | |
1da177e4 | 65 | */ |
9e5987a7 DC |
66 | void |
67 | xfs_bmap_compute_maxlevels( | |
68 | xfs_mount_t *mp, /* file system mount structure */ | |
69 | int whichfork) /* data or attr fork */ | |
70 | { | |
71 | int level; /* btree level */ | |
72 | uint maxblocks; /* max blocks at this level */ | |
73 | uint maxleafents; /* max leaf entries possible */ | |
74 | int maxrootrecs; /* max records in root block */ | |
75 | int minleafrecs; /* min records in leaf block */ | |
76 | int minnoderecs; /* min records in node block */ | |
77 | int sz; /* root block size */ | |
1da177e4 | 78 | |
9e5987a7 DC |
79 | /* |
80 | * The maximum number of extents in a file, hence the maximum | |
81 | * number of leaf entries, is controlled by the type of di_nextents | |
82 | * (a signed 32-bit number, xfs_extnum_t), or by di_anextents | |
83 | * (a signed 16-bit number, xfs_aextnum_t). | |
84 | * | |
85 | * Note that we can no longer assume that if we are in ATTR1 that | |
86 | * the fork offset of all the inodes will be | |
87 | * (xfs_default_attroffset(ip) >> 3) because we could have mounted | |
88 | * with ATTR2 and then mounted back with ATTR1, keeping the | |
89 | * di_forkoff's fixed but probably at various positions. Therefore, | |
90 | * for both ATTR1 and ATTR2 we have to assume the worst case scenario | |
91 | * of a minimum size available. | |
92 | */ | |
93 | if (whichfork == XFS_DATA_FORK) { | |
94 | maxleafents = MAXEXTNUM; | |
95 | sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS); | |
96 | } else { | |
97 | maxleafents = MAXAEXTNUM; | |
98 | sz = XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
99 | } | |
152d93b7 | 100 | maxrootrecs = xfs_bmdr_maxrecs(sz, 0); |
9e5987a7 DC |
101 | minleafrecs = mp->m_bmap_dmnr[0]; |
102 | minnoderecs = mp->m_bmap_dmnr[1]; | |
103 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
104 | for (level = 1; maxblocks > 1; level++) { | |
105 | if (maxblocks <= maxrootrecs) | |
106 | maxblocks = 1; | |
107 | else | |
108 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
109 | } | |
110 | mp->m_bm_maxlevels[whichfork] = level; | |
111 | } | |
91e11088 | 112 | |
fe033cc8 CH |
113 | STATIC int /* error */ |
114 | xfs_bmbt_lookup_eq( | |
115 | struct xfs_btree_cur *cur, | |
116 | xfs_fileoff_t off, | |
117 | xfs_fsblock_t bno, | |
118 | xfs_filblks_t len, | |
119 | int *stat) /* success/failure */ | |
120 | { | |
121 | cur->bc_rec.b.br_startoff = off; | |
122 | cur->bc_rec.b.br_startblock = bno; | |
123 | cur->bc_rec.b.br_blockcount = len; | |
124 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); | |
125 | } | |
126 | ||
127 | STATIC int /* error */ | |
128 | xfs_bmbt_lookup_ge( | |
129 | struct xfs_btree_cur *cur, | |
130 | xfs_fileoff_t off, | |
131 | xfs_fsblock_t bno, | |
132 | xfs_filblks_t len, | |
133 | int *stat) /* success/failure */ | |
134 | { | |
135 | cur->bc_rec.b.br_startoff = off; | |
136 | cur->bc_rec.b.br_startblock = bno; | |
137 | cur->bc_rec.b.br_blockcount = len; | |
138 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); | |
139 | } | |
140 | ||
278d0ca1 | 141 | /* |
8096b1eb CH |
142 | * Check if the inode needs to be converted to btree format. |
143 | */ | |
144 | static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork) | |
145 | { | |
60b4984f DW |
146 | return whichfork != XFS_COW_FORK && |
147 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
8096b1eb CH |
148 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
149 | XFS_IFORK_MAXEXT(ip, whichfork); | |
150 | } | |
151 | ||
152 | /* | |
153 | * Check if the inode should be converted to extent format. | |
154 | */ | |
155 | static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork) | |
156 | { | |
60b4984f DW |
157 | return whichfork != XFS_COW_FORK && |
158 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE && | |
8096b1eb CH |
159 | XFS_IFORK_NEXTENTS(ip, whichfork) <= |
160 | XFS_IFORK_MAXEXT(ip, whichfork); | |
161 | } | |
162 | ||
163 | /* | |
164 | * Update the record referred to by cur to the value given | |
278d0ca1 CH |
165 | * by [off, bno, len, state]. |
166 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
167 | */ | |
168 | STATIC int | |
169 | xfs_bmbt_update( | |
170 | struct xfs_btree_cur *cur, | |
171 | xfs_fileoff_t off, | |
172 | xfs_fsblock_t bno, | |
173 | xfs_filblks_t len, | |
174 | xfs_exntst_t state) | |
175 | { | |
176 | union xfs_btree_rec rec; | |
177 | ||
178 | xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state); | |
179 | return xfs_btree_update(cur, &rec); | |
180 | } | |
fe033cc8 | 181 | |
1da177e4 | 182 | /* |
9e5987a7 DC |
183 | * Compute the worst-case number of indirect blocks that will be used |
184 | * for ip's delayed extent of length "len". | |
1da177e4 | 185 | */ |
9e5987a7 DC |
186 | STATIC xfs_filblks_t |
187 | xfs_bmap_worst_indlen( | |
188 | xfs_inode_t *ip, /* incore inode pointer */ | |
189 | xfs_filblks_t len) /* delayed extent length */ | |
1da177e4 | 190 | { |
9e5987a7 DC |
191 | int level; /* btree level number */ |
192 | int maxrecs; /* maximum record count at this level */ | |
193 | xfs_mount_t *mp; /* mount structure */ | |
194 | xfs_filblks_t rval; /* return value */ | |
fd26a880 | 195 | xfs_filblks_t orig_len; |
1da177e4 LT |
196 | |
197 | mp = ip->i_mount; | |
fd26a880 DW |
198 | |
199 | /* Calculate the worst-case size of the bmbt. */ | |
200 | orig_len = len; | |
9e5987a7 DC |
201 | maxrecs = mp->m_bmap_dmxr[0]; |
202 | for (level = 0, rval = 0; | |
203 | level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK); | |
204 | level++) { | |
205 | len += maxrecs - 1; | |
206 | do_div(len, maxrecs); | |
207 | rval += len; | |
fd26a880 DW |
208 | if (len == 1) { |
209 | rval += XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) - | |
9e5987a7 | 210 | level - 1; |
fd26a880 DW |
211 | break; |
212 | } | |
9e5987a7 DC |
213 | if (level == 0) |
214 | maxrecs = mp->m_bmap_dmxr[1]; | |
1da177e4 | 215 | } |
fd26a880 DW |
216 | |
217 | /* Calculate the worst-case size of the rmapbt. */ | |
218 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
219 | rval += 1 + xfs_rmapbt_calc_size(mp, orig_len) + | |
220 | mp->m_rmap_maxlevels; | |
221 | ||
9e5987a7 | 222 | return rval; |
1da177e4 LT |
223 | } |
224 | ||
225 | /* | |
9e5987a7 | 226 | * Calculate the default attribute fork offset for newly created inodes. |
1da177e4 | 227 | */ |
9e5987a7 DC |
228 | uint |
229 | xfs_default_attroffset( | |
230 | struct xfs_inode *ip) | |
1da177e4 | 231 | { |
9e5987a7 DC |
232 | struct xfs_mount *mp = ip->i_mount; |
233 | uint offset; | |
1da177e4 | 234 | |
9e5987a7 | 235 | if (mp->m_sb.sb_inodesize == 256) { |
56cea2d0 | 236 | offset = XFS_LITINO(mp, ip->i_d.di_version) - |
9e5987a7 DC |
237 | XFS_BMDR_SPACE_CALC(MINABTPTRS); |
238 | } else { | |
239 | offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); | |
1da177e4 | 240 | } |
9e5987a7 | 241 | |
56cea2d0 | 242 | ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version)); |
9e5987a7 | 243 | return offset; |
1da177e4 LT |
244 | } |
245 | ||
246 | /* | |
9e5987a7 DC |
247 | * Helper routine to reset inode di_forkoff field when switching |
248 | * attribute fork from local to extent format - we reset it where | |
249 | * possible to make space available for inline data fork extents. | |
1e82379b DC |
250 | */ |
251 | STATIC void | |
9e5987a7 | 252 | xfs_bmap_forkoff_reset( |
9e5987a7 DC |
253 | xfs_inode_t *ip, |
254 | int whichfork) | |
1e82379b | 255 | { |
9e5987a7 DC |
256 | if (whichfork == XFS_ATTR_FORK && |
257 | ip->i_d.di_format != XFS_DINODE_FMT_DEV && | |
258 | ip->i_d.di_format != XFS_DINODE_FMT_UUID && | |
259 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE) { | |
260 | uint dfl_forkoff = xfs_default_attroffset(ip) >> 3; | |
261 | ||
262 | if (dfl_forkoff > ip->i_d.di_forkoff) | |
263 | ip->i_d.di_forkoff = dfl_forkoff; | |
264 | } | |
1e82379b DC |
265 | } |
266 | ||
9e5987a7 DC |
267 | #ifdef DEBUG |
268 | STATIC struct xfs_buf * | |
269 | xfs_bmap_get_bp( | |
270 | struct xfs_btree_cur *cur, | |
271 | xfs_fsblock_t bno) | |
272 | { | |
273 | struct xfs_log_item_desc *lidp; | |
274 | int i; | |
7574aa92 | 275 | |
9e5987a7 DC |
276 | if (!cur) |
277 | return NULL; | |
278 | ||
279 | for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) { | |
280 | if (!cur->bc_bufs[i]) | |
281 | break; | |
282 | if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno) | |
283 | return cur->bc_bufs[i]; | |
1da177e4 | 284 | } |
7574aa92 | 285 | |
9e5987a7 DC |
286 | /* Chase down all the log items to see if the bp is there */ |
287 | list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) { | |
288 | struct xfs_buf_log_item *bip; | |
289 | bip = (struct xfs_buf_log_item *)lidp->lid_item; | |
290 | if (bip->bli_item.li_type == XFS_LI_BUF && | |
291 | XFS_BUF_ADDR(bip->bli_buf) == bno) | |
292 | return bip->bli_buf; | |
293 | } | |
7574aa92 | 294 | |
9e5987a7 DC |
295 | return NULL; |
296 | } | |
0b1b213f | 297 | |
9e5987a7 DC |
298 | STATIC void |
299 | xfs_check_block( | |
300 | struct xfs_btree_block *block, | |
301 | xfs_mount_t *mp, | |
302 | int root, | |
303 | short sz) | |
304 | { | |
305 | int i, j, dmxr; | |
306 | __be64 *pp, *thispa; /* pointer to block address */ | |
307 | xfs_bmbt_key_t *prevp, *keyp; | |
1da177e4 | 308 | |
9e5987a7 | 309 | ASSERT(be16_to_cpu(block->bb_level) > 0); |
ec90c556 | 310 | |
9e5987a7 DC |
311 | prevp = NULL; |
312 | for( i = 1; i <= xfs_btree_get_numrecs(block); i++) { | |
313 | dmxr = mp->m_bmap_dmxr[0]; | |
314 | keyp = XFS_BMBT_KEY_ADDR(mp, block, i); | |
0b1b213f | 315 | |
9e5987a7 DC |
316 | if (prevp) { |
317 | ASSERT(be64_to_cpu(prevp->br_startoff) < | |
318 | be64_to_cpu(keyp->br_startoff)); | |
1da177e4 | 319 | } |
9e5987a7 | 320 | prevp = keyp; |
1da177e4 | 321 | |
1da177e4 | 322 | /* |
9e5987a7 | 323 | * Compare the block numbers to see if there are dups. |
1da177e4 | 324 | */ |
9e5987a7 DC |
325 | if (root) |
326 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz); | |
327 | else | |
328 | pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr); | |
0b1b213f | 329 | |
9e5987a7 DC |
330 | for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) { |
331 | if (root) | |
332 | thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz); | |
333 | else | |
334 | thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr); | |
335 | if (*thispa == *pp) { | |
336 | xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld", | |
337 | __func__, j, i, | |
338 | (unsigned long long)be64_to_cpu(*thispa)); | |
339 | panic("%s: ptrs are equal in node\n", | |
340 | __func__); | |
341 | } | |
1da177e4 | 342 | } |
9e5987a7 DC |
343 | } |
344 | } | |
1da177e4 | 345 | |
9e5987a7 DC |
346 | /* |
347 | * Check that the extents for the inode ip are in the right order in all | |
e3543819 DC |
348 | * btree leaves. THis becomes prohibitively expensive for large extent count |
349 | * files, so don't bother with inodes that have more than 10,000 extents in | |
350 | * them. The btree record ordering checks will still be done, so for such large | |
351 | * bmapbt constructs that is going to catch most corruptions. | |
9e5987a7 | 352 | */ |
9e5987a7 DC |
353 | STATIC void |
354 | xfs_bmap_check_leaf_extents( | |
355 | xfs_btree_cur_t *cur, /* btree cursor or null */ | |
356 | xfs_inode_t *ip, /* incore inode pointer */ | |
357 | int whichfork) /* data or attr fork */ | |
358 | { | |
359 | struct xfs_btree_block *block; /* current btree block */ | |
360 | xfs_fsblock_t bno; /* block # of "block" */ | |
361 | xfs_buf_t *bp; /* buffer for "block" */ | |
362 | int error; /* error return value */ | |
363 | xfs_extnum_t i=0, j; /* index into the extents list */ | |
364 | xfs_ifork_t *ifp; /* fork structure */ | |
365 | int level; /* btree level, for checking */ | |
366 | xfs_mount_t *mp; /* file system mount structure */ | |
367 | __be64 *pp; /* pointer to block address */ | |
368 | xfs_bmbt_rec_t *ep; /* pointer to current extent */ | |
369 | xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */ | |
370 | xfs_bmbt_rec_t *nextp; /* pointer to next extent */ | |
371 | int bp_release = 0; | |
372 | ||
373 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) { | |
374 | return; | |
375 | } | |
376 | ||
e3543819 DC |
377 | /* skip large extent count inodes */ |
378 | if (ip->i_d.di_nextents > 10000) | |
379 | return; | |
380 | ||
9e5987a7 DC |
381 | bno = NULLFSBLOCK; |
382 | mp = ip->i_mount; | |
383 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
384 | block = ifp->if_broot; | |
385 | /* | |
386 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
387 | */ | |
388 | level = be16_to_cpu(block->bb_level); | |
389 | ASSERT(level > 0); | |
390 | xfs_check_block(block, mp, 1, ifp->if_broot_bytes); | |
391 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
392 | bno = be64_to_cpu(*pp); | |
393 | ||
d5cf09ba | 394 | ASSERT(bno != NULLFSBLOCK); |
9e5987a7 DC |
395 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
396 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
397 | ||
398 | /* | |
399 | * Go down the tree until leaf level is reached, following the first | |
400 | * pointer (leftmost) at each level. | |
401 | */ | |
402 | while (level-- > 0) { | |
403 | /* See if buf is in cur first */ | |
404 | bp_release = 0; | |
405 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
406 | if (!bp) { | |
407 | bp_release = 1; | |
408 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
409 | XFS_BMAP_BTREE_REF, | |
410 | &xfs_bmbt_buf_ops); | |
572a4cf0 | 411 | if (error) |
9e5987a7 | 412 | goto error_norelse; |
1da177e4 | 413 | } |
9e5987a7 | 414 | block = XFS_BUF_TO_BLOCK(bp); |
9e5987a7 DC |
415 | if (level == 0) |
416 | break; | |
1da177e4 | 417 | |
1da177e4 | 418 | /* |
9e5987a7 DC |
419 | * Check this block for basic sanity (increasing keys and |
420 | * no duplicate blocks). | |
1da177e4 | 421 | */ |
0b1b213f | 422 | |
9e5987a7 DC |
423 | xfs_check_block(block, mp, 0, 0); |
424 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
425 | bno = be64_to_cpu(*pp); | |
c29aad41 ES |
426 | XFS_WANT_CORRUPTED_GOTO(mp, |
427 | XFS_FSB_SANITY_CHECK(mp, bno), error0); | |
9e5987a7 DC |
428 | if (bp_release) { |
429 | bp_release = 0; | |
430 | xfs_trans_brelse(NULL, bp); | |
1da177e4 | 431 | } |
9e5987a7 | 432 | } |
ec90c556 | 433 | |
9e5987a7 DC |
434 | /* |
435 | * Here with bp and block set to the leftmost leaf node in the tree. | |
436 | */ | |
437 | i = 0; | |
438 | ||
439 | /* | |
440 | * Loop over all leaf nodes checking that all extents are in the right order. | |
441 | */ | |
442 | for (;;) { | |
443 | xfs_fsblock_t nextbno; | |
444 | xfs_extnum_t num_recs; | |
445 | ||
446 | ||
447 | num_recs = xfs_btree_get_numrecs(block); | |
1da177e4 | 448 | |
1da177e4 | 449 | /* |
9e5987a7 | 450 | * Read-ahead the next leaf block, if any. |
1da177e4 | 451 | */ |
8096b1eb | 452 | |
9e5987a7 | 453 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
1da177e4 | 454 | |
1da177e4 | 455 | /* |
9e5987a7 DC |
456 | * Check all the extents to make sure they are OK. |
457 | * If we had a previous block, the last entry should | |
458 | * conform with the first entry in this one. | |
1da177e4 | 459 | */ |
ec90c556 | 460 | |
9e5987a7 DC |
461 | ep = XFS_BMBT_REC_ADDR(mp, block, 1); |
462 | if (i) { | |
463 | ASSERT(xfs_bmbt_disk_get_startoff(&last) + | |
464 | xfs_bmbt_disk_get_blockcount(&last) <= | |
465 | xfs_bmbt_disk_get_startoff(ep)); | |
466 | } | |
467 | for (j = 1; j < num_recs; j++) { | |
468 | nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1); | |
469 | ASSERT(xfs_bmbt_disk_get_startoff(ep) + | |
470 | xfs_bmbt_disk_get_blockcount(ep) <= | |
471 | xfs_bmbt_disk_get_startoff(nextp)); | |
472 | ep = nextp; | |
473 | } | |
1da177e4 | 474 | |
9e5987a7 DC |
475 | last = *ep; |
476 | i += num_recs; | |
477 | if (bp_release) { | |
478 | bp_release = 0; | |
479 | xfs_trans_brelse(NULL, bp); | |
480 | } | |
481 | bno = nextbno; | |
1da177e4 | 482 | /* |
9e5987a7 | 483 | * If we've reached the end, stop. |
1da177e4 | 484 | */ |
9e5987a7 DC |
485 | if (bno == NULLFSBLOCK) |
486 | break; | |
8096b1eb | 487 | |
9e5987a7 DC |
488 | bp_release = 0; |
489 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
490 | if (!bp) { | |
491 | bp_release = 1; | |
492 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
493 | XFS_BMAP_BTREE_REF, | |
494 | &xfs_bmbt_buf_ops); | |
b9b984d7 | 495 | if (error) |
9e5987a7 | 496 | goto error_norelse; |
1da177e4 | 497 | } |
9e5987a7 | 498 | block = XFS_BUF_TO_BLOCK(bp); |
a5bd606b | 499 | } |
a5fd276b | 500 | |
9e5987a7 | 501 | return; |
a5bd606b | 502 | |
9e5987a7 DC |
503 | error0: |
504 | xfs_warn(mp, "%s: at error0", __func__); | |
505 | if (bp_release) | |
506 | xfs_trans_brelse(NULL, bp); | |
507 | error_norelse: | |
508 | xfs_warn(mp, "%s: BAD after btree leaves for %d extents", | |
509 | __func__, i); | |
510 | panic("%s: CORRUPTED BTREE OR SOMETHING", __func__); | |
511 | return; | |
1da177e4 LT |
512 | } |
513 | ||
514 | /* | |
9e5987a7 | 515 | * Add bmap trace insert entries for all the contents of the extent records. |
1da177e4 | 516 | */ |
9e5987a7 DC |
517 | void |
518 | xfs_bmap_trace_exlist( | |
519 | xfs_inode_t *ip, /* incore inode pointer */ | |
520 | xfs_extnum_t cnt, /* count of entries in the list */ | |
c44a1f22 | 521 | int whichfork, /* data or attr or cow fork */ |
9e5987a7 | 522 | unsigned long caller_ip) |
1da177e4 | 523 | { |
9e5987a7 DC |
524 | xfs_extnum_t idx; /* extent record index */ |
525 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
526 | int state = 0; | |
a5bd606b | 527 | |
9e5987a7 DC |
528 | if (whichfork == XFS_ATTR_FORK) |
529 | state |= BMAP_ATTRFORK; | |
c44a1f22 ES |
530 | else if (whichfork == XFS_COW_FORK) |
531 | state |= BMAP_COWFORK; | |
a5bd606b | 532 | |
9e5987a7 | 533 | ifp = XFS_IFORK_PTR(ip, whichfork); |
5d829300 | 534 | ASSERT(cnt == xfs_iext_count(ifp)); |
9e5987a7 | 535 | for (idx = 0; idx < cnt; idx++) |
7710517f | 536 | trace_xfs_extlist(ip, idx, state, caller_ip); |
9e5987a7 | 537 | } |
a5bd606b | 538 | |
9e5987a7 DC |
539 | /* |
540 | * Validate that the bmbt_irecs being returned from bmapi are valid | |
a97f4df7 ZYW |
541 | * given the caller's original parameters. Specifically check the |
542 | * ranges of the returned irecs to ensure that they only extend beyond | |
9e5987a7 DC |
543 | * the given parameters if the XFS_BMAPI_ENTIRE flag was set. |
544 | */ | |
545 | STATIC void | |
546 | xfs_bmap_validate_ret( | |
547 | xfs_fileoff_t bno, | |
548 | xfs_filblks_t len, | |
549 | int flags, | |
550 | xfs_bmbt_irec_t *mval, | |
551 | int nmap, | |
552 | int ret_nmap) | |
553 | { | |
554 | int i; /* index to map values */ | |
a5bd606b | 555 | |
9e5987a7 | 556 | ASSERT(ret_nmap <= nmap); |
a5bd606b | 557 | |
9e5987a7 DC |
558 | for (i = 0; i < ret_nmap; i++) { |
559 | ASSERT(mval[i].br_blockcount > 0); | |
560 | if (!(flags & XFS_BMAPI_ENTIRE)) { | |
561 | ASSERT(mval[i].br_startoff >= bno); | |
562 | ASSERT(mval[i].br_blockcount <= len); | |
563 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount <= | |
564 | bno + len); | |
565 | } else { | |
566 | ASSERT(mval[i].br_startoff < bno + len); | |
567 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount > | |
568 | bno); | |
569 | } | |
570 | ASSERT(i == 0 || | |
571 | mval[i - 1].br_startoff + mval[i - 1].br_blockcount == | |
572 | mval[i].br_startoff); | |
573 | ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK && | |
574 | mval[i].br_startblock != HOLESTARTBLOCK); | |
575 | ASSERT(mval[i].br_state == XFS_EXT_NORM || | |
576 | mval[i].br_state == XFS_EXT_UNWRITTEN); | |
577 | } | |
578 | } | |
7574aa92 | 579 | |
9e5987a7 DC |
580 | #else |
581 | #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0) | |
8ae947ef | 582 | #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0) |
9e5987a7 | 583 | #endif /* DEBUG */ |
7574aa92 | 584 | |
9e5987a7 DC |
585 | /* |
586 | * bmap free list manipulation functions | |
587 | */ | |
7574aa92 | 588 | |
9e5987a7 DC |
589 | /* |
590 | * Add the extent to the list of extents to be free at transaction end. | |
591 | * The list is maintained sorted (by block number). | |
592 | */ | |
593 | void | |
594 | xfs_bmap_add_free( | |
340785cc DW |
595 | struct xfs_mount *mp, |
596 | struct xfs_defer_ops *dfops, | |
597 | xfs_fsblock_t bno, | |
598 | xfs_filblks_t len, | |
599 | struct xfs_owner_info *oinfo) | |
9e5987a7 | 600 | { |
310a75a3 | 601 | struct xfs_extent_free_item *new; /* new element */ |
9e5987a7 DC |
602 | #ifdef DEBUG |
603 | xfs_agnumber_t agno; | |
604 | xfs_agblock_t agbno; | |
605 | ||
606 | ASSERT(bno != NULLFSBLOCK); | |
607 | ASSERT(len > 0); | |
608 | ASSERT(len <= MAXEXTLEN); | |
609 | ASSERT(!isnullstartblock(bno)); | |
610 | agno = XFS_FSB_TO_AGNO(mp, bno); | |
611 | agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
612 | ASSERT(agno < mp->m_sb.sb_agcount); | |
613 | ASSERT(agbno < mp->m_sb.sb_agblocks); | |
614 | ASSERT(len < mp->m_sb.sb_agblocks); | |
615 | ASSERT(agbno + len <= mp->m_sb.sb_agblocks); | |
616 | #endif | |
617 | ASSERT(xfs_bmap_free_item_zone != NULL); | |
340785cc | 618 | |
9e5987a7 | 619 | new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP); |
310a75a3 DW |
620 | new->xefi_startblock = bno; |
621 | new->xefi_blockcount = (xfs_extlen_t)len; | |
340785cc DW |
622 | if (oinfo) |
623 | new->xefi_oinfo = *oinfo; | |
624 | else | |
625 | xfs_rmap_skip_owner_update(&new->xefi_oinfo); | |
ba9e7802 DW |
626 | trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0, |
627 | XFS_FSB_TO_AGBNO(mp, bno), len); | |
2c3234d1 | 628 | xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list); |
9e5987a7 | 629 | } |
0b1b213f | 630 | |
9e5987a7 DC |
631 | /* |
632 | * Inode fork format manipulation functions | |
633 | */ | |
1da177e4 | 634 | |
9e5987a7 DC |
635 | /* |
636 | * Transform a btree format file with only one leaf node, where the | |
637 | * extents list will fit in the inode, into an extents format file. | |
638 | * Since the file extents are already in-core, all we have to do is | |
639 | * give up the space for the btree root and pitch the leaf block. | |
640 | */ | |
641 | STATIC int /* error */ | |
642 | xfs_bmap_btree_to_extents( | |
643 | xfs_trans_t *tp, /* transaction pointer */ | |
644 | xfs_inode_t *ip, /* incore inode pointer */ | |
645 | xfs_btree_cur_t *cur, /* btree cursor */ | |
646 | int *logflagsp, /* inode logging flags */ | |
647 | int whichfork) /* data or attr fork */ | |
648 | { | |
649 | /* REFERENCED */ | |
650 | struct xfs_btree_block *cblock;/* child btree block */ | |
651 | xfs_fsblock_t cbno; /* child block number */ | |
652 | xfs_buf_t *cbp; /* child block's buffer */ | |
653 | int error; /* error return value */ | |
654 | xfs_ifork_t *ifp; /* inode fork data */ | |
655 | xfs_mount_t *mp; /* mount point structure */ | |
656 | __be64 *pp; /* ptr to block address */ | |
657 | struct xfs_btree_block *rblock;/* root btree block */ | |
340785cc | 658 | struct xfs_owner_info oinfo; |
1da177e4 | 659 | |
9e5987a7 DC |
660 | mp = ip->i_mount; |
661 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
60b4984f | 662 | ASSERT(whichfork != XFS_COW_FORK); |
9e5987a7 DC |
663 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
664 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
665 | rblock = ifp->if_broot; | |
666 | ASSERT(be16_to_cpu(rblock->bb_level) == 1); | |
667 | ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1); | |
668 | ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1); | |
669 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes); | |
670 | cbno = be64_to_cpu(*pp); | |
671 | *logflagsp = 0; | |
672 | #ifdef DEBUG | |
673 | if ((error = xfs_btree_check_lptr(cur, cbno, 1))) | |
674 | return error; | |
675 | #endif | |
676 | error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF, | |
677 | &xfs_bmbt_buf_ops); | |
678 | if (error) | |
679 | return error; | |
680 | cblock = XFS_BUF_TO_BLOCK(cbp); | |
681 | if ((error = xfs_btree_check_block(cur, cblock, 0, cbp))) | |
682 | return error; | |
340785cc DW |
683 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); |
684 | xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo); | |
9e5987a7 DC |
685 | ip->i_d.di_nblocks--; |
686 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); | |
687 | xfs_trans_binval(tp, cbp); | |
688 | if (cur->bc_bufs[0] == cbp) | |
689 | cur->bc_bufs[0] = NULL; | |
690 | xfs_iroot_realloc(ip, -1, whichfork); | |
691 | ASSERT(ifp->if_broot == NULL); | |
692 | ASSERT((ifp->if_flags & XFS_IFBROOT) == 0); | |
693 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
694 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); | |
695 | return 0; | |
696 | } | |
0b1b213f | 697 | |
9e5987a7 DC |
698 | /* |
699 | * Convert an extents-format file into a btree-format file. | |
700 | * The new file will have a root block (in the inode) and a single child block. | |
701 | */ | |
702 | STATIC int /* error */ | |
703 | xfs_bmap_extents_to_btree( | |
704 | xfs_trans_t *tp, /* transaction pointer */ | |
705 | xfs_inode_t *ip, /* incore inode pointer */ | |
706 | xfs_fsblock_t *firstblock, /* first-block-allocated */ | |
2c3234d1 | 707 | struct xfs_defer_ops *dfops, /* blocks freed in xaction */ |
9e5987a7 DC |
708 | xfs_btree_cur_t **curp, /* cursor returned to caller */ |
709 | int wasdel, /* converting a delayed alloc */ | |
710 | int *logflagsp, /* inode logging flags */ | |
711 | int whichfork) /* data or attr fork */ | |
712 | { | |
713 | struct xfs_btree_block *ablock; /* allocated (child) bt block */ | |
714 | xfs_buf_t *abp; /* buffer for ablock */ | |
715 | xfs_alloc_arg_t args; /* allocation arguments */ | |
716 | xfs_bmbt_rec_t *arp; /* child record pointer */ | |
717 | struct xfs_btree_block *block; /* btree root block */ | |
718 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
719 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ | |
720 | int error; /* error return value */ | |
721 | xfs_extnum_t i, cnt; /* extent record index */ | |
722 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
723 | xfs_bmbt_key_t *kp; /* root block key pointer */ | |
724 | xfs_mount_t *mp; /* mount structure */ | |
725 | xfs_extnum_t nextents; /* number of file extents */ | |
726 | xfs_bmbt_ptr_t *pp; /* root block address pointer */ | |
1da177e4 | 727 | |
ee1a47ab | 728 | mp = ip->i_mount; |
60b4984f | 729 | ASSERT(whichfork != XFS_COW_FORK); |
9e5987a7 DC |
730 | ifp = XFS_IFORK_PTR(ip, whichfork); |
731 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS); | |
0b1b213f | 732 | |
9e5987a7 DC |
733 | /* |
734 | * Make space in the inode incore. | |
735 | */ | |
736 | xfs_iroot_realloc(ip, 1, whichfork); | |
737 | ifp->if_flags |= XFS_IFBROOT; | |
ec90c556 | 738 | |
9e5987a7 DC |
739 | /* |
740 | * Fill in the root. | |
741 | */ | |
742 | block = ifp->if_broot; | |
b6f41e44 ES |
743 | xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL, |
744 | XFS_BTNUM_BMAP, 1, 1, ip->i_ino, | |
f88ae46b | 745 | XFS_BTREE_LONG_PTRS); |
9e5987a7 DC |
746 | /* |
747 | * Need a cursor. Can't allocate until bb_level is filled in. | |
748 | */ | |
9e5987a7 DC |
749 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
750 | cur->bc_private.b.firstblock = *firstblock; | |
2c3234d1 | 751 | cur->bc_private.b.dfops = dfops; |
9e5987a7 DC |
752 | cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; |
753 | /* | |
754 | * Convert to a btree with two levels, one record in root. | |
755 | */ | |
756 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE); | |
757 | memset(&args, 0, sizeof(args)); | |
758 | args.tp = tp; | |
759 | args.mp = mp; | |
340785cc | 760 | xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork); |
9e5987a7 DC |
761 | args.firstblock = *firstblock; |
762 | if (*firstblock == NULLFSBLOCK) { | |
763 | args.type = XFS_ALLOCTYPE_START_BNO; | |
764 | args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino); | |
2c3234d1 | 765 | } else if (dfops->dop_low) { |
9e5987a7 DC |
766 | args.type = XFS_ALLOCTYPE_START_BNO; |
767 | args.fsbno = *firstblock; | |
768 | } else { | |
769 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
770 | args.fsbno = *firstblock; | |
771 | } | |
772 | args.minlen = args.maxlen = args.prod = 1; | |
773 | args.wasdel = wasdel; | |
774 | *logflagsp = 0; | |
775 | if ((error = xfs_alloc_vextent(&args))) { | |
776 | xfs_iroot_realloc(ip, -1, whichfork); | |
777 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
778 | return error; | |
779 | } | |
90e2056d | 780 | |
2fcc319d CH |
781 | if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) { |
782 | xfs_iroot_realloc(ip, -1, whichfork); | |
783 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
784 | return -ENOSPC; | |
785 | } | |
9e5987a7 DC |
786 | /* |
787 | * Allocation can't fail, the space was reserved. | |
788 | */ | |
9e5987a7 | 789 | ASSERT(*firstblock == NULLFSBLOCK || |
410d17f6 | 790 | args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock)); |
9e5987a7 DC |
791 | *firstblock = cur->bc_private.b.firstblock = args.fsbno; |
792 | cur->bc_private.b.allocated++; | |
793 | ip->i_d.di_nblocks++; | |
794 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L); | |
795 | abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0); | |
796 | /* | |
797 | * Fill in the child block. | |
798 | */ | |
799 | abp->b_ops = &xfs_bmbt_buf_ops; | |
800 | ablock = XFS_BUF_TO_BLOCK(abp); | |
b6f41e44 ES |
801 | xfs_btree_init_block_int(mp, ablock, abp->b_bn, |
802 | XFS_BTNUM_BMAP, 0, 0, ip->i_ino, | |
ee1a47ab CH |
803 | XFS_BTREE_LONG_PTRS); |
804 | ||
9e5987a7 | 805 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); |
5d829300 | 806 | nextents = xfs_iext_count(ifp); |
9e5987a7 DC |
807 | for (cnt = i = 0; i < nextents; i++) { |
808 | ep = xfs_iext_get_ext(ifp, i); | |
809 | if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) { | |
810 | arp->l0 = cpu_to_be64(ep->l0); | |
811 | arp->l1 = cpu_to_be64(ep->l1); | |
812 | arp++; cnt++; | |
1da177e4 | 813 | } |
9e5987a7 DC |
814 | } |
815 | ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork)); | |
816 | xfs_btree_set_numrecs(ablock, cnt); | |
1da177e4 | 817 | |
9e5987a7 DC |
818 | /* |
819 | * Fill in the root key and pointer. | |
820 | */ | |
821 | kp = XFS_BMBT_KEY_ADDR(mp, block, 1); | |
822 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); | |
823 | kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp)); | |
824 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur, | |
825 | be16_to_cpu(block->bb_level))); | |
826 | *pp = cpu_to_be64(args.fsbno); | |
ec90c556 | 827 | |
9e5987a7 DC |
828 | /* |
829 | * Do all this logging at the end so that | |
830 | * the root is at the right level. | |
831 | */ | |
832 | xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS); | |
833 | xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs)); | |
834 | ASSERT(*curp == NULL); | |
835 | *curp = cur; | |
836 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork); | |
837 | return 0; | |
838 | } | |
ec90c556 | 839 | |
9e5987a7 DC |
840 | /* |
841 | * Convert a local file to an extents file. | |
842 | * This code is out of bounds for data forks of regular files, | |
843 | * since the file data needs to get logged so things will stay consistent. | |
844 | * (The bmap-level manipulations are ok, though). | |
845 | */ | |
f3508bcd DC |
846 | void |
847 | xfs_bmap_local_to_extents_empty( | |
848 | struct xfs_inode *ip, | |
849 | int whichfork) | |
850 | { | |
851 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
852 | ||
60b4984f | 853 | ASSERT(whichfork != XFS_COW_FORK); |
f3508bcd DC |
854 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); |
855 | ASSERT(ifp->if_bytes == 0); | |
856 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0); | |
857 | ||
6a9edd3d | 858 | xfs_bmap_forkoff_reset(ip, whichfork); |
f3508bcd DC |
859 | ifp->if_flags &= ~XFS_IFINLINE; |
860 | ifp->if_flags |= XFS_IFEXTENTS; | |
861 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
862 | } | |
863 | ||
864 | ||
9e5987a7 DC |
865 | STATIC int /* error */ |
866 | xfs_bmap_local_to_extents( | |
867 | xfs_trans_t *tp, /* transaction pointer */ | |
868 | xfs_inode_t *ip, /* incore inode pointer */ | |
869 | xfs_fsblock_t *firstblock, /* first block allocated in xaction */ | |
870 | xfs_extlen_t total, /* total blocks needed by transaction */ | |
871 | int *logflagsp, /* inode logging flags */ | |
872 | int whichfork, | |
ee1a47ab CH |
873 | void (*init_fn)(struct xfs_trans *tp, |
874 | struct xfs_buf *bp, | |
9e5987a7 DC |
875 | struct xfs_inode *ip, |
876 | struct xfs_ifork *ifp)) | |
877 | { | |
f3508bcd | 878 | int error = 0; |
9e5987a7 DC |
879 | int flags; /* logging flags returned */ |
880 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
f3508bcd DC |
881 | xfs_alloc_arg_t args; /* allocation arguments */ |
882 | xfs_buf_t *bp; /* buffer for extent block */ | |
883 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ | |
0b1b213f | 884 | |
9e5987a7 DC |
885 | /* |
886 | * We don't want to deal with the case of keeping inode data inline yet. | |
887 | * So sending the data fork of a regular inode is invalid. | |
888 | */ | |
c19b3b05 | 889 | ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK)); |
9e5987a7 DC |
890 | ifp = XFS_IFORK_PTR(ip, whichfork); |
891 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
f3508bcd DC |
892 | |
893 | if (!ifp->if_bytes) { | |
894 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
895 | flags = XFS_ILOG_CORE; | |
896 | goto done; | |
897 | } | |
898 | ||
9e5987a7 DC |
899 | flags = 0; |
900 | error = 0; | |
f3508bcd DC |
901 | ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == |
902 | XFS_IFINLINE); | |
903 | memset(&args, 0, sizeof(args)); | |
904 | args.tp = tp; | |
905 | args.mp = ip->i_mount; | |
340785cc | 906 | xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0); |
f3508bcd DC |
907 | args.firstblock = *firstblock; |
908 | /* | |
909 | * Allocate a block. We know we need only one, since the | |
910 | * file currently fits in an inode. | |
911 | */ | |
912 | if (*firstblock == NULLFSBLOCK) { | |
913 | args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino); | |
914 | args.type = XFS_ALLOCTYPE_START_BNO; | |
9e5987a7 | 915 | } else { |
f3508bcd DC |
916 | args.fsbno = *firstblock; |
917 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
9e5987a7 | 918 | } |
f3508bcd DC |
919 | args.total = total; |
920 | args.minlen = args.maxlen = args.prod = 1; | |
921 | error = xfs_alloc_vextent(&args); | |
922 | if (error) | |
923 | goto done; | |
924 | ||
925 | /* Can't fail, the space was reserved. */ | |
926 | ASSERT(args.fsbno != NULLFSBLOCK); | |
927 | ASSERT(args.len == 1); | |
928 | *firstblock = args.fsbno; | |
929 | bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0); | |
930 | ||
fe22d552 | 931 | /* |
b7cdc66b | 932 | * Initialize the block, copy the data and log the remote buffer. |
fe22d552 | 933 | * |
b7cdc66b BF |
934 | * The callout is responsible for logging because the remote format |
935 | * might differ from the local format and thus we don't know how much to | |
936 | * log here. Note that init_fn must also set the buffer log item type | |
937 | * correctly. | |
fe22d552 | 938 | */ |
f3508bcd DC |
939 | init_fn(tp, bp, ip, ifp); |
940 | ||
b7cdc66b | 941 | /* account for the change in fork size */ |
f3508bcd DC |
942 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); |
943 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
9e5987a7 | 944 | flags |= XFS_ILOG_CORE; |
f3508bcd DC |
945 | |
946 | xfs_iext_add(ifp, 0, 1); | |
947 | ep = xfs_iext_get_ext(ifp, 0); | |
948 | xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM); | |
949 | trace_xfs_bmap_post_update(ip, 0, | |
950 | whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0, | |
951 | _THIS_IP_); | |
952 | XFS_IFORK_NEXT_SET(ip, whichfork, 1); | |
953 | ip->i_d.di_nblocks = 1; | |
954 | xfs_trans_mod_dquot_byino(tp, ip, | |
955 | XFS_TRANS_DQ_BCOUNT, 1L); | |
956 | flags |= xfs_ilog_fext(whichfork); | |
957 | ||
9e5987a7 DC |
958 | done: |
959 | *logflagsp = flags; | |
960 | return error; | |
961 | } | |
ec90c556 | 962 | |
9e5987a7 DC |
963 | /* |
964 | * Called from xfs_bmap_add_attrfork to handle btree format files. | |
965 | */ | |
966 | STATIC int /* error */ | |
967 | xfs_bmap_add_attrfork_btree( | |
968 | xfs_trans_t *tp, /* transaction pointer */ | |
969 | xfs_inode_t *ip, /* incore inode pointer */ | |
970 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
2c3234d1 | 971 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
9e5987a7 DC |
972 | int *flags) /* inode logging flags */ |
973 | { | |
974 | xfs_btree_cur_t *cur; /* btree cursor */ | |
975 | int error; /* error return value */ | |
976 | xfs_mount_t *mp; /* file system mount struct */ | |
977 | int stat; /* newroot status */ | |
ec90c556 | 978 | |
9e5987a7 DC |
979 | mp = ip->i_mount; |
980 | if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip)) | |
981 | *flags |= XFS_ILOG_DBROOT; | |
982 | else { | |
983 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
2c3234d1 | 984 | cur->bc_private.b.dfops = dfops; |
9e5987a7 DC |
985 | cur->bc_private.b.firstblock = *firstblock; |
986 | if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat))) | |
987 | goto error0; | |
988 | /* must be at least one entry */ | |
c29aad41 | 989 | XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0); |
9e5987a7 DC |
990 | if ((error = xfs_btree_new_iroot(cur, flags, &stat))) |
991 | goto error0; | |
992 | if (stat == 0) { | |
993 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2451337d | 994 | return -ENOSPC; |
1da177e4 | 995 | } |
9e5987a7 DC |
996 | *firstblock = cur->bc_private.b.firstblock; |
997 | cur->bc_private.b.allocated = 0; | |
998 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1da177e4 | 999 | } |
9e5987a7 DC |
1000 | return 0; |
1001 | error0: | |
1002 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1003 | return error; | |
1004 | } | |
a5bd606b | 1005 | |
9e5987a7 DC |
1006 | /* |
1007 | * Called from xfs_bmap_add_attrfork to handle extents format files. | |
1008 | */ | |
1009 | STATIC int /* error */ | |
1010 | xfs_bmap_add_attrfork_extents( | |
1011 | xfs_trans_t *tp, /* transaction pointer */ | |
1012 | xfs_inode_t *ip, /* incore inode pointer */ | |
1013 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
2c3234d1 | 1014 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
9e5987a7 DC |
1015 | int *flags) /* inode logging flags */ |
1016 | { | |
1017 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
1018 | int error; /* error return value */ | |
a5bd606b | 1019 | |
9e5987a7 DC |
1020 | if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip)) |
1021 | return 0; | |
1022 | cur = NULL; | |
2c3234d1 | 1023 | error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0, |
9e5987a7 | 1024 | flags, XFS_DATA_FORK); |
a5bd606b CH |
1025 | if (cur) { |
1026 | cur->bc_private.b.allocated = 0; | |
9e5987a7 DC |
1027 | xfs_btree_del_cursor(cur, |
1028 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
a5bd606b | 1029 | } |
1da177e4 | 1030 | return error; |
1da177e4 LT |
1031 | } |
1032 | ||
9e5987a7 DC |
1033 | /* |
1034 | * Called from xfs_bmap_add_attrfork to handle local format files. Each | |
1035 | * different data fork content type needs a different callout to do the | |
1036 | * conversion. Some are basic and only require special block initialisation | |
1037 | * callouts for the data formating, others (directories) are so specialised they | |
1038 | * handle everything themselves. | |
1039 | * | |
1040 | * XXX (dgc): investigate whether directory conversion can use the generic | |
1041 | * formatting callout. It should be possible - it's just a very complex | |
ee1a47ab | 1042 | * formatter. |
9e5987a7 DC |
1043 | */ |
1044 | STATIC int /* error */ | |
1045 | xfs_bmap_add_attrfork_local( | |
1046 | xfs_trans_t *tp, /* transaction pointer */ | |
1047 | xfs_inode_t *ip, /* incore inode pointer */ | |
1048 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
2c3234d1 | 1049 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
9e5987a7 DC |
1050 | int *flags) /* inode logging flags */ |
1051 | { | |
1052 | xfs_da_args_t dargs; /* args for dir/attr code */ | |
7574aa92 | 1053 | |
9e5987a7 DC |
1054 | if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip)) |
1055 | return 0; | |
7574aa92 | 1056 | |
c19b3b05 | 1057 | if (S_ISDIR(VFS_I(ip)->i_mode)) { |
9e5987a7 | 1058 | memset(&dargs, 0, sizeof(dargs)); |
d6cf1305 | 1059 | dargs.geo = ip->i_mount->m_dir_geo; |
9e5987a7 DC |
1060 | dargs.dp = ip; |
1061 | dargs.firstblock = firstblock; | |
2c3234d1 | 1062 | dargs.dfops = dfops; |
d6cf1305 | 1063 | dargs.total = dargs.geo->fsbcount; |
9e5987a7 DC |
1064 | dargs.whichfork = XFS_DATA_FORK; |
1065 | dargs.trans = tp; | |
1066 | return xfs_dir2_sf_to_block(&dargs); | |
1da177e4 | 1067 | } |
7574aa92 | 1068 | |
c19b3b05 | 1069 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
9e5987a7 DC |
1070 | return xfs_bmap_local_to_extents(tp, ip, firstblock, 1, |
1071 | flags, XFS_DATA_FORK, | |
1072 | xfs_symlink_local_to_remote); | |
7574aa92 | 1073 | |
f3508bcd DC |
1074 | /* should only be called for types that support local format data */ |
1075 | ASSERT(0); | |
2451337d | 1076 | return -EFSCORRUPTED; |
9e5987a7 | 1077 | } |
0b1b213f | 1078 | |
9e5987a7 DC |
1079 | /* |
1080 | * Convert inode from non-attributed to attributed. | |
1081 | * Must not be in a transaction, ip must not be locked. | |
1082 | */ | |
1083 | int /* error code */ | |
1084 | xfs_bmap_add_attrfork( | |
1085 | xfs_inode_t *ip, /* incore inode pointer */ | |
1086 | int size, /* space new attribute needs */ | |
1087 | int rsvd) /* xact may use reserved blks */ | |
1088 | { | |
1089 | xfs_fsblock_t firstblock; /* 1st block/ag allocated */ | |
2c3234d1 | 1090 | struct xfs_defer_ops dfops; /* freed extent records */ |
9e5987a7 DC |
1091 | xfs_mount_t *mp; /* mount structure */ |
1092 | xfs_trans_t *tp; /* transaction pointer */ | |
1093 | int blks; /* space reservation */ | |
1094 | int version = 1; /* superblock attr version */ | |
9e5987a7 DC |
1095 | int logflags; /* logging flags */ |
1096 | int error; /* error return value */ | |
0b1b213f | 1097 | |
9e5987a7 | 1098 | ASSERT(XFS_IFORK_Q(ip) == 0); |
1da177e4 | 1099 | |
9e5987a7 DC |
1100 | mp = ip->i_mount; |
1101 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); | |
253f4911 | 1102 | |
9e5987a7 | 1103 | blks = XFS_ADDAFORK_SPACE_RES(mp); |
253f4911 CH |
1104 | |
1105 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0, | |
1106 | rsvd ? XFS_TRANS_RESERVE : 0, &tp); | |
1107 | if (error) | |
9e3908e3 | 1108 | return error; |
253f4911 | 1109 | |
9e5987a7 DC |
1110 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1111 | error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ? | |
1112 | XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES : | |
1113 | XFS_QMOPT_RES_REGBLKS); | |
9e3908e3 MT |
1114 | if (error) |
1115 | goto trans_cancel; | |
9e5987a7 | 1116 | if (XFS_IFORK_Q(ip)) |
9e3908e3 | 1117 | goto trans_cancel; |
0f352f8e DW |
1118 | if (ip->i_d.di_anextents != 0) { |
1119 | error = -EFSCORRUPTED; | |
1120 | goto trans_cancel; | |
1121 | } | |
9e5987a7 | 1122 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) { |
1da177e4 | 1123 | /* |
9e5987a7 | 1124 | * For inodes coming from pre-6.2 filesystems. |
1da177e4 | 1125 | */ |
9e5987a7 DC |
1126 | ASSERT(ip->i_d.di_aformat == 0); |
1127 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1128 | } | |
ec90c556 | 1129 | |
9e3908e3 | 1130 | xfs_trans_ijoin(tp, ip, 0); |
9e5987a7 | 1131 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1da177e4 | 1132 | |
9e5987a7 DC |
1133 | switch (ip->i_d.di_format) { |
1134 | case XFS_DINODE_FMT_DEV: | |
1135 | ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; | |
1136 | break; | |
1137 | case XFS_DINODE_FMT_UUID: | |
1138 | ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3; | |
1139 | break; | |
1140 | case XFS_DINODE_FMT_LOCAL: | |
1141 | case XFS_DINODE_FMT_EXTENTS: | |
1142 | case XFS_DINODE_FMT_BTREE: | |
1143 | ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size); | |
1144 | if (!ip->i_d.di_forkoff) | |
1145 | ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3; | |
1146 | else if (mp->m_flags & XFS_MOUNT_ATTR2) | |
1147 | version = 2; | |
1da177e4 | 1148 | break; |
9e5987a7 DC |
1149 | default: |
1150 | ASSERT(0); | |
2451337d | 1151 | error = -EINVAL; |
9e3908e3 | 1152 | goto trans_cancel; |
9e5987a7 | 1153 | } |
1da177e4 | 1154 | |
9e5987a7 DC |
1155 | ASSERT(ip->i_afp == NULL); |
1156 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
1157 | ip->i_afp->if_flags = XFS_IFEXTENTS; | |
1158 | logflags = 0; | |
2c3234d1 | 1159 | xfs_defer_init(&dfops, &firstblock); |
9e5987a7 DC |
1160 | switch (ip->i_d.di_format) { |
1161 | case XFS_DINODE_FMT_LOCAL: | |
2c3234d1 | 1162 | error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops, |
9e5987a7 DC |
1163 | &logflags); |
1164 | break; | |
1165 | case XFS_DINODE_FMT_EXTENTS: | |
1166 | error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock, | |
2c3234d1 | 1167 | &dfops, &logflags); |
9e5987a7 DC |
1168 | break; |
1169 | case XFS_DINODE_FMT_BTREE: | |
2c3234d1 | 1170 | error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops, |
9e5987a7 DC |
1171 | &logflags); |
1172 | break; | |
1173 | default: | |
1174 | error = 0; | |
1da177e4 LT |
1175 | break; |
1176 | } | |
9e5987a7 DC |
1177 | if (logflags) |
1178 | xfs_trans_log_inode(tp, ip, logflags); | |
1179 | if (error) | |
9e3908e3 | 1180 | goto bmap_cancel; |
9e5987a7 DC |
1181 | if (!xfs_sb_version_hasattr(&mp->m_sb) || |
1182 | (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) { | |
61e63ecb | 1183 | bool log_sb = false; |
9e5987a7 DC |
1184 | |
1185 | spin_lock(&mp->m_sb_lock); | |
1186 | if (!xfs_sb_version_hasattr(&mp->m_sb)) { | |
1187 | xfs_sb_version_addattr(&mp->m_sb); | |
61e63ecb | 1188 | log_sb = true; |
9e5987a7 DC |
1189 | } |
1190 | if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) { | |
1191 | xfs_sb_version_addattr2(&mp->m_sb); | |
61e63ecb | 1192 | log_sb = true; |
9e5987a7 | 1193 | } |
4d11a402 | 1194 | spin_unlock(&mp->m_sb_lock); |
61e63ecb DC |
1195 | if (log_sb) |
1196 | xfs_log_sb(tp); | |
1da177e4 | 1197 | } |
9e5987a7 | 1198 | |
2c3234d1 | 1199 | error = xfs_defer_finish(&tp, &dfops, NULL); |
9e5987a7 | 1200 | if (error) |
9e3908e3 | 1201 | goto bmap_cancel; |
70393313 | 1202 | error = xfs_trans_commit(tp); |
9e3908e3 MT |
1203 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1204 | return error; | |
1205 | ||
1206 | bmap_cancel: | |
2c3234d1 | 1207 | xfs_defer_cancel(&dfops); |
9e3908e3 | 1208 | trans_cancel: |
4906e215 | 1209 | xfs_trans_cancel(tp); |
9e5987a7 | 1210 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
9e5987a7 | 1211 | return error; |
1da177e4 LT |
1212 | } |
1213 | ||
1214 | /* | |
9e5987a7 | 1215 | * Internal and external extent tree search functions. |
1da177e4 | 1216 | */ |
a5bd606b | 1217 | |
9e5987a7 DC |
1218 | /* |
1219 | * Read in the extents to if_extents. | |
1220 | * All inode fields are set up by caller, we just traverse the btree | |
1221 | * and copy the records in. If the file system cannot contain unwritten | |
1222 | * extents, the records are checked for no "state" flags. | |
1223 | */ | |
1224 | int /* error */ | |
1225 | xfs_bmap_read_extents( | |
1226 | xfs_trans_t *tp, /* transaction pointer */ | |
1227 | xfs_inode_t *ip, /* incore inode */ | |
1228 | int whichfork) /* data or attr fork */ | |
1229 | { | |
1230 | struct xfs_btree_block *block; /* current btree block */ | |
1231 | xfs_fsblock_t bno; /* block # of "block" */ | |
1232 | xfs_buf_t *bp; /* buffer for "block" */ | |
1233 | int error; /* error return value */ | |
9e5987a7 DC |
1234 | xfs_extnum_t i, j; /* index into the extents list */ |
1235 | xfs_ifork_t *ifp; /* fork structure */ | |
1236 | int level; /* btree level, for checking */ | |
1237 | xfs_mount_t *mp; /* file system mount structure */ | |
1238 | __be64 *pp; /* pointer to block address */ | |
1239 | /* REFERENCED */ | |
1240 | xfs_extnum_t room; /* number of entries there's room for */ | |
6ef35544 | 1241 | |
9e5987a7 DC |
1242 | mp = ip->i_mount; |
1243 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
9e5987a7 | 1244 | block = ifp->if_broot; |
1da177e4 | 1245 | /* |
9e5987a7 | 1246 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. |
1da177e4 | 1247 | */ |
9e5987a7 DC |
1248 | level = be16_to_cpu(block->bb_level); |
1249 | ASSERT(level > 0); | |
1250 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
1251 | bno = be64_to_cpu(*pp); | |
d5a91bae | 1252 | |
1da177e4 | 1253 | /* |
9e5987a7 DC |
1254 | * Go down the tree until leaf level is reached, following the first |
1255 | * pointer (leftmost) at each level. | |
1da177e4 | 1256 | */ |
9e5987a7 DC |
1257 | while (level-- > 0) { |
1258 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1259 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1260 | if (error) | |
1261 | return error; | |
1262 | block = XFS_BUF_TO_BLOCK(bp); | |
9e5987a7 DC |
1263 | if (level == 0) |
1264 | break; | |
1265 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
1266 | bno = be64_to_cpu(*pp); | |
c29aad41 ES |
1267 | XFS_WANT_CORRUPTED_GOTO(mp, |
1268 | XFS_FSB_SANITY_CHECK(mp, bno), error0); | |
9e5987a7 | 1269 | xfs_trans_brelse(tp, bp); |
1da177e4 LT |
1270 | } |
1271 | /* | |
9e5987a7 | 1272 | * Here with bp and block set to the leftmost leaf node in the tree. |
1da177e4 | 1273 | */ |
5d829300 | 1274 | room = xfs_iext_count(ifp); |
9e5987a7 | 1275 | i = 0; |
1da177e4 | 1276 | /* |
9e5987a7 | 1277 | * Loop over all leaf nodes. Copy information to the extent records. |
1da177e4 | 1278 | */ |
9e5987a7 DC |
1279 | for (;;) { |
1280 | xfs_bmbt_rec_t *frp; | |
1281 | xfs_fsblock_t nextbno; | |
1282 | xfs_extnum_t num_recs; | |
0b1b213f | 1283 | |
9e5987a7 DC |
1284 | num_recs = xfs_btree_get_numrecs(block); |
1285 | if (unlikely(i + num_recs > room)) { | |
1286 | ASSERT(i + num_recs <= room); | |
1287 | xfs_warn(ip->i_mount, | |
1288 | "corrupt dinode %Lu, (btree extents).", | |
1289 | (unsigned long long) ip->i_ino); | |
1290 | XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)", | |
1291 | XFS_ERRLEVEL_LOW, ip->i_mount, block); | |
1292 | goto error0; | |
1da177e4 | 1293 | } |
1da177e4 | 1294 | /* |
9e5987a7 | 1295 | * Read-ahead the next leaf block, if any. |
1da177e4 | 1296 | */ |
9e5987a7 DC |
1297 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
1298 | if (nextbno != NULLFSBLOCK) | |
1299 | xfs_btree_reada_bufl(mp, nextbno, 1, | |
1300 | &xfs_bmbt_buf_ops); | |
1da177e4 | 1301 | /* |
9e5987a7 | 1302 | * Copy records into the extent records. |
1da177e4 | 1303 | */ |
9e5987a7 | 1304 | frp = XFS_BMBT_REC_ADDR(mp, block, 1); |
9e5987a7 DC |
1305 | for (j = 0; j < num_recs; j++, i++, frp++) { |
1306 | xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i); | |
1307 | trp->l0 = be64_to_cpu(frp->l0); | |
1308 | trp->l1 = be64_to_cpu(frp->l1); | |
0c1d9e4a | 1309 | if (!xfs_bmbt_validate_extent(mp, whichfork, trp)) { |
9e5987a7 | 1310 | XFS_ERROR_REPORT("xfs_bmap_read_extents(2)", |
0c1d9e4a | 1311 | XFS_ERRLEVEL_LOW, mp); |
9e5987a7 DC |
1312 | goto error0; |
1313 | } | |
1314 | } | |
1315 | xfs_trans_brelse(tp, bp); | |
1316 | bno = nextbno; | |
1da177e4 | 1317 | /* |
9e5987a7 | 1318 | * If we've reached the end, stop. |
1da177e4 | 1319 | */ |
9e5987a7 DC |
1320 | if (bno == NULLFSBLOCK) |
1321 | break; | |
1322 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1323 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1324 | if (error) | |
1325 | return error; | |
1326 | block = XFS_BUF_TO_BLOCK(bp); | |
1da177e4 | 1327 | } |
356a3225 DW |
1328 | if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) |
1329 | return -EFSCORRUPTED; | |
5d829300 | 1330 | ASSERT(i == xfs_iext_count(ifp)); |
9e5987a7 DC |
1331 | XFS_BMAP_TRACE_EXLIST(ip, i, whichfork); |
1332 | return 0; | |
1333 | error0: | |
1334 | xfs_trans_brelse(tp, bp); | |
2451337d | 1335 | return -EFSCORRUPTED; |
9e5987a7 | 1336 | } |
a5bd606b | 1337 | |
9e5987a7 DC |
1338 | /* |
1339 | * Returns the file-relative block number of the first unused block(s) | |
1340 | * in the file with at least "len" logically contiguous blocks free. | |
1341 | * This is the lowest-address hole if the file has holes, else the first block | |
1342 | * past the end of file. | |
1343 | * Return 0 if the file is currently local (in-inode). | |
1344 | */ | |
1345 | int /* error */ | |
1346 | xfs_bmap_first_unused( | |
1347 | xfs_trans_t *tp, /* transaction pointer */ | |
1348 | xfs_inode_t *ip, /* incore inode */ | |
1349 | xfs_extlen_t len, /* size of hole to find */ | |
1350 | xfs_fileoff_t *first_unused, /* unused block */ | |
1351 | int whichfork) /* data or attr fork */ | |
1352 | { | |
1353 | int error; /* error return value */ | |
1354 | int idx; /* extent record index */ | |
1355 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1356 | xfs_fileoff_t lastaddr; /* last block number seen */ | |
1357 | xfs_fileoff_t lowest; /* lowest useful block */ | |
1358 | xfs_fileoff_t max; /* starting useful block */ | |
1359 | xfs_fileoff_t off; /* offset for this block */ | |
1360 | xfs_extnum_t nextents; /* number of extent entries */ | |
1361 | ||
1362 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE || | |
1363 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS || | |
1364 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
1365 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
1366 | *first_unused = 0; | |
1367 | return 0; | |
dd9f438e | 1368 | } |
9e5987a7 DC |
1369 | ifp = XFS_IFORK_PTR(ip, whichfork); |
1370 | if (!(ifp->if_flags & XFS_IFEXTENTS) && | |
1371 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
1372 | return error; | |
1373 | lowest = *first_unused; | |
5d829300 | 1374 | nextents = xfs_iext_count(ifp); |
9e5987a7 DC |
1375 | for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) { |
1376 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx); | |
1377 | off = xfs_bmbt_get_startoff(ep); | |
1378 | /* | |
1379 | * See if the hole before this extent will work. | |
1380 | */ | |
1381 | if (off >= lowest + len && off - max >= len) { | |
1382 | *first_unused = max; | |
1383 | return 0; | |
1384 | } | |
1385 | lastaddr = off + xfs_bmbt_get_blockcount(ep); | |
1386 | max = XFS_FILEOFF_MAX(lastaddr, lowest); | |
dd9f438e | 1387 | } |
9e5987a7 DC |
1388 | *first_unused = max; |
1389 | return 0; | |
1390 | } | |
1391 | ||
1392 | /* | |
02bb4873 | 1393 | * Returns the file-relative block number of the last block - 1 before |
9e5987a7 DC |
1394 | * last_block (input value) in the file. |
1395 | * This is not based on i_size, it is based on the extent records. | |
1396 | * Returns 0 for local files, as they do not have extent records. | |
1397 | */ | |
1398 | int /* error */ | |
1399 | xfs_bmap_last_before( | |
86685f7b CH |
1400 | struct xfs_trans *tp, /* transaction pointer */ |
1401 | struct xfs_inode *ip, /* incore inode */ | |
1402 | xfs_fileoff_t *last_block, /* last block */ | |
1403 | int whichfork) /* data or attr fork */ | |
9e5987a7 | 1404 | { |
86685f7b CH |
1405 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1406 | struct xfs_bmbt_irec got; | |
1407 | xfs_extnum_t idx; | |
1408 | int error; | |
9e5987a7 | 1409 | |
86685f7b CH |
1410 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { |
1411 | case XFS_DINODE_FMT_LOCAL: | |
9e5987a7 DC |
1412 | *last_block = 0; |
1413 | return 0; | |
86685f7b CH |
1414 | case XFS_DINODE_FMT_BTREE: |
1415 | case XFS_DINODE_FMT_EXTENTS: | |
1416 | break; | |
1417 | default: | |
1418 | return -EIO; | |
9e5987a7 | 1419 | } |
86685f7b CH |
1420 | |
1421 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1422 | error = xfs_iread_extents(tp, ip, whichfork); | |
1423 | if (error) | |
1424 | return error; | |
dd9f438e | 1425 | } |
86685f7b CH |
1426 | |
1427 | if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) { | |
1428 | if (got.br_startoff <= *last_block - 1) | |
1429 | return 0; | |
1430 | } | |
1431 | ||
1432 | if (xfs_iext_get_extent(ifp, idx - 1, &got)) { | |
1433 | *last_block = got.br_startoff + got.br_blockcount; | |
1434 | return 0; | |
1435 | } | |
1436 | ||
1437 | *last_block = 0; | |
9e5987a7 DC |
1438 | return 0; |
1439 | } | |
1440 | ||
68988114 | 1441 | int |
9e5987a7 DC |
1442 | xfs_bmap_last_extent( |
1443 | struct xfs_trans *tp, | |
1444 | struct xfs_inode *ip, | |
1445 | int whichfork, | |
1446 | struct xfs_bmbt_irec *rec, | |
1447 | int *is_empty) | |
1448 | { | |
1449 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
1450 | int error; | |
1451 | int nextents; | |
1452 | ||
1453 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1454 | error = xfs_iread_extents(tp, ip, whichfork); | |
1455 | if (error) | |
1456 | return error; | |
dd9f438e NS |
1457 | } |
1458 | ||
5d829300 | 1459 | nextents = xfs_iext_count(ifp); |
9e5987a7 DC |
1460 | if (nextents == 0) { |
1461 | *is_empty = 1; | |
1462 | return 0; | |
1463 | } | |
dd9f438e | 1464 | |
9e5987a7 DC |
1465 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec); |
1466 | *is_empty = 0; | |
dd9f438e NS |
1467 | return 0; |
1468 | } | |
1469 | ||
9e5987a7 DC |
1470 | /* |
1471 | * Check the last inode extent to determine whether this allocation will result | |
1472 | * in blocks being allocated at the end of the file. When we allocate new data | |
1473 | * blocks at the end of the file which do not start at the previous data block, | |
1474 | * we will try to align the new blocks at stripe unit boundaries. | |
1475 | * | |
6e708bcf | 1476 | * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be |
9e5987a7 DC |
1477 | * at, or past the EOF. |
1478 | */ | |
1479 | STATIC int | |
1480 | xfs_bmap_isaeof( | |
1481 | struct xfs_bmalloca *bma, | |
1482 | int whichfork) | |
1da177e4 | 1483 | { |
9e5987a7 DC |
1484 | struct xfs_bmbt_irec rec; |
1485 | int is_empty; | |
1486 | int error; | |
1da177e4 | 1487 | |
9e5987a7 DC |
1488 | bma->aeof = 0; |
1489 | error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec, | |
1490 | &is_empty); | |
6e708bcf | 1491 | if (error) |
9e5987a7 | 1492 | return error; |
1da177e4 | 1493 | |
6e708bcf DC |
1494 | if (is_empty) { |
1495 | bma->aeof = 1; | |
1496 | return 0; | |
1497 | } | |
1498 | ||
1da177e4 | 1499 | /* |
9e5987a7 DC |
1500 | * Check if we are allocation or past the last extent, or at least into |
1501 | * the last delayed allocated extent. | |
1da177e4 | 1502 | */ |
9e5987a7 DC |
1503 | bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount || |
1504 | (bma->offset >= rec.br_startoff && | |
1505 | isnullstartblock(rec.br_startblock)); | |
1506 | return 0; | |
1507 | } | |
1da177e4 | 1508 | |
9e5987a7 DC |
1509 | /* |
1510 | * Returns the file-relative block number of the first block past eof in | |
1511 | * the file. This is not based on i_size, it is based on the extent records. | |
1512 | * Returns 0 for local files, as they do not have extent records. | |
1513 | */ | |
1514 | int | |
1515 | xfs_bmap_last_offset( | |
9e5987a7 DC |
1516 | struct xfs_inode *ip, |
1517 | xfs_fileoff_t *last_block, | |
1518 | int whichfork) | |
1519 | { | |
1520 | struct xfs_bmbt_irec rec; | |
1521 | int is_empty; | |
1522 | int error; | |
04e99455 | 1523 | |
9e5987a7 | 1524 | *last_block = 0; |
0892ccd6 | 1525 | |
9e5987a7 DC |
1526 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) |
1527 | return 0; | |
a365bdd5 | 1528 | |
9e5987a7 DC |
1529 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE && |
1530 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
2451337d | 1531 | return -EIO; |
a365bdd5 | 1532 | |
9e5987a7 DC |
1533 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty); |
1534 | if (error || is_empty) | |
a365bdd5 | 1535 | return error; |
9e5987a7 DC |
1536 | |
1537 | *last_block = rec.br_startoff + rec.br_blockcount; | |
a365bdd5 NS |
1538 | return 0; |
1539 | } | |
1540 | ||
9e5987a7 DC |
1541 | /* |
1542 | * Returns whether the selected fork of the inode has exactly one | |
1543 | * block or not. For the data fork we check this matches di_size, | |
1544 | * implying the file's range is 0..bsize-1. | |
1545 | */ | |
1546 | int /* 1=>1 block, 0=>otherwise */ | |
1547 | xfs_bmap_one_block( | |
1548 | xfs_inode_t *ip, /* incore inode */ | |
1549 | int whichfork) /* data or attr fork */ | |
c467c049 | 1550 | { |
9e5987a7 DC |
1551 | xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */ |
1552 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1553 | int rval; /* return value */ | |
1554 | xfs_bmbt_irec_t s; /* internal version of extent */ | |
c467c049 | 1555 | |
9e5987a7 DC |
1556 | #ifndef DEBUG |
1557 | if (whichfork == XFS_DATA_FORK) | |
1558 | return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize; | |
1559 | #endif /* !DEBUG */ | |
1560 | if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1) | |
1561 | return 0; | |
1562 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
1563 | return 0; | |
1564 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1565 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
1566 | ep = xfs_iext_get_ext(ifp, 0); | |
1567 | xfs_bmbt_get_all(ep, &s); | |
1568 | rval = s.br_startoff == 0 && s.br_blockcount == 1; | |
1569 | if (rval && whichfork == XFS_DATA_FORK) | |
1570 | ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize); | |
1571 | return rval; | |
1572 | } | |
c467c049 | 1573 | |
9e5987a7 DC |
1574 | /* |
1575 | * Extent tree manipulation functions used during allocation. | |
1576 | */ | |
c467c049 | 1577 | |
9e5987a7 DC |
1578 | /* |
1579 | * Convert a delayed allocation to a real allocation. | |
1580 | */ | |
1581 | STATIC int /* error */ | |
1582 | xfs_bmap_add_extent_delay_real( | |
60b4984f DW |
1583 | struct xfs_bmalloca *bma, |
1584 | int whichfork) | |
9e5987a7 DC |
1585 | { |
1586 | struct xfs_bmbt_irec *new = &bma->got; | |
1587 | int diff; /* temp value */ | |
1588 | xfs_bmbt_rec_host_t *ep; /* extent entry for idx */ | |
1589 | int error; /* error return value */ | |
1590 | int i; /* temp state */ | |
1591 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1592 | xfs_fileoff_t new_endoff; /* end offset of new entry */ | |
1593 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
1594 | /* left is 0, right is 1, prev is 2 */ | |
1595 | int rval=0; /* return value (logging flags) */ | |
1596 | int state = 0;/* state bits, accessed thru macros */ | |
1597 | xfs_filblks_t da_new; /* new count del alloc blocks used */ | |
1598 | xfs_filblks_t da_old; /* old count del alloc blocks used */ | |
1599 | xfs_filblks_t temp=0; /* value for da_new calculations */ | |
1600 | xfs_filblks_t temp2=0;/* value for da_new calculations */ | |
1601 | int tmp_rval; /* partial logging flags */ | |
c29aad41 | 1602 | struct xfs_mount *mp; |
60b4984f | 1603 | xfs_extnum_t *nextents; |
c467c049 | 1604 | |
f1f96c49 | 1605 | mp = bma->ip->i_mount; |
6d3eb1ec | 1606 | ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
60b4984f DW |
1607 | ASSERT(whichfork != XFS_ATTR_FORK); |
1608 | nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents : | |
1609 | &bma->ip->i_d.di_nextents); | |
c467c049 | 1610 | |
9e5987a7 | 1611 | ASSERT(bma->idx >= 0); |
5d829300 | 1612 | ASSERT(bma->idx <= xfs_iext_count(ifp)); |
9e5987a7 DC |
1613 | ASSERT(!isnullstartblock(new->br_startblock)); |
1614 | ASSERT(!bma->cur || | |
1615 | (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); | |
c467c049 | 1616 | |
ff6d6af2 | 1617 | XFS_STATS_INC(mp, xs_add_exlist); |
c467c049 | 1618 | |
9e5987a7 DC |
1619 | #define LEFT r[0] |
1620 | #define RIGHT r[1] | |
1621 | #define PREV r[2] | |
c467c049 | 1622 | |
60b4984f DW |
1623 | if (whichfork == XFS_COW_FORK) |
1624 | state |= BMAP_COWFORK; | |
1625 | ||
c467c049 | 1626 | /* |
9e5987a7 | 1627 | * Set up a bunch of variables to make the tests simpler. |
c467c049 | 1628 | */ |
9e5987a7 DC |
1629 | ep = xfs_iext_get_ext(ifp, bma->idx); |
1630 | xfs_bmbt_get_all(ep, &PREV); | |
1631 | new_endoff = new->br_startoff + new->br_blockcount; | |
1632 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
1633 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
1634 | ||
1635 | da_old = startblockval(PREV.br_startblock); | |
1636 | da_new = 0; | |
1637 | ||
c467c049 | 1638 | /* |
9e5987a7 DC |
1639 | * Set flags determining what part of the previous delayed allocation |
1640 | * extent is being replaced by a real allocation. | |
c467c049 | 1641 | */ |
9e5987a7 DC |
1642 | if (PREV.br_startoff == new->br_startoff) |
1643 | state |= BMAP_LEFT_FILLING; | |
1644 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
1645 | state |= BMAP_RIGHT_FILLING; | |
c467c049 CH |
1646 | |
1647 | /* | |
9e5987a7 DC |
1648 | * Check and set flags if this segment has a left neighbor. |
1649 | * Don't set contiguous if the combined extent would be too large. | |
c467c049 | 1650 | */ |
9e5987a7 DC |
1651 | if (bma->idx > 0) { |
1652 | state |= BMAP_LEFT_VALID; | |
1653 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT); | |
a99ebf43 | 1654 | |
9e5987a7 DC |
1655 | if (isnullstartblock(LEFT.br_startblock)) |
1656 | state |= BMAP_LEFT_DELAY; | |
a365bdd5 | 1657 | } |
a365bdd5 | 1658 | |
9e5987a7 DC |
1659 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && |
1660 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
1661 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
1662 | LEFT.br_state == new->br_state && | |
1663 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
1664 | state |= BMAP_LEFT_CONTIG; | |
a365bdd5 | 1665 | |
1da177e4 | 1666 | /* |
9e5987a7 DC |
1667 | * Check and set flags if this segment has a right neighbor. |
1668 | * Don't set contiguous if the combined extent would be too large. | |
1669 | * Also check for all-three-contiguous being too large. | |
1da177e4 | 1670 | */ |
5d829300 | 1671 | if (bma->idx < xfs_iext_count(ifp) - 1) { |
9e5987a7 DC |
1672 | state |= BMAP_RIGHT_VALID; |
1673 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT); | |
14b064ce | 1674 | |
9e5987a7 DC |
1675 | if (isnullstartblock(RIGHT.br_startblock)) |
1676 | state |= BMAP_RIGHT_DELAY; | |
1da177e4 | 1677 | } |
9e5987a7 DC |
1678 | |
1679 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && | |
1680 | new_endoff == RIGHT.br_startoff && | |
1681 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
1682 | new->br_state == RIGHT.br_state && | |
1683 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
1684 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1685 | BMAP_RIGHT_FILLING)) != | |
1686 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1687 | BMAP_RIGHT_FILLING) || | |
1688 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
1689 | <= MAXEXTLEN)) | |
1690 | state |= BMAP_RIGHT_CONTIG; | |
1691 | ||
1692 | error = 0; | |
1da177e4 | 1693 | /* |
9e5987a7 | 1694 | * Switch out based on the FILLING and CONTIG state bits. |
1da177e4 | 1695 | */ |
9e5987a7 DC |
1696 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | |
1697 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
1698 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1699 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1da177e4 | 1700 | /* |
9e5987a7 DC |
1701 | * Filling in all of a previously delayed allocation extent. |
1702 | * The left and right neighbors are both contiguous with new. | |
1da177e4 | 1703 | */ |
9e5987a7 DC |
1704 | bma->idx--; |
1705 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1706 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx), | |
1707 | LEFT.br_blockcount + PREV.br_blockcount + | |
1708 | RIGHT.br_blockcount); | |
1709 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1710 | ||
1711 | xfs_iext_remove(bma->ip, bma->idx + 1, 2, state); | |
60b4984f | 1712 | (*nextents)--; |
9e5987a7 DC |
1713 | if (bma->cur == NULL) |
1714 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1715 | else { | |
1716 | rval = XFS_ILOG_CORE; | |
1717 | error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff, | |
1718 | RIGHT.br_startblock, | |
1719 | RIGHT.br_blockcount, &i); | |
1720 | if (error) | |
1721 | goto done; | |
c29aad41 | 1722 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1723 | error = xfs_btree_delete(bma->cur, &i); |
1724 | if (error) | |
1725 | goto done; | |
c29aad41 | 1726 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1727 | error = xfs_btree_decrement(bma->cur, 0, &i); |
1728 | if (error) | |
1729 | goto done; | |
c29aad41 | 1730 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1731 | error = xfs_bmbt_update(bma->cur, LEFT.br_startoff, |
1732 | LEFT.br_startblock, | |
1733 | LEFT.br_blockcount + | |
1734 | PREV.br_blockcount + | |
1735 | RIGHT.br_blockcount, LEFT.br_state); | |
1736 | if (error) | |
1737 | goto done; | |
1738 | } | |
1739 | break; | |
1740 | ||
1741 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
a365bdd5 | 1742 | /* |
9e5987a7 DC |
1743 | * Filling in all of a previously delayed allocation extent. |
1744 | * The left neighbor is contiguous, the right is not. | |
a365bdd5 | 1745 | */ |
9e5987a7 DC |
1746 | bma->idx--; |
1747 | ||
1748 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1749 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx), | |
1750 | LEFT.br_blockcount + PREV.br_blockcount); | |
1751 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1752 | ||
1753 | xfs_iext_remove(bma->ip, bma->idx + 1, 1, state); | |
1754 | if (bma->cur == NULL) | |
1755 | rval = XFS_ILOG_DEXT; | |
1756 | else { | |
1757 | rval = 0; | |
1758 | error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff, | |
1759 | LEFT.br_startblock, LEFT.br_blockcount, | |
1760 | &i); | |
1761 | if (error) | |
1762 | goto done; | |
c29aad41 | 1763 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1764 | error = xfs_bmbt_update(bma->cur, LEFT.br_startoff, |
1765 | LEFT.br_startblock, | |
1766 | LEFT.br_blockcount + | |
1767 | PREV.br_blockcount, LEFT.br_state); | |
1768 | if (error) | |
1769 | goto done; | |
1770 | } | |
1771 | break; | |
1772 | ||
1773 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
0937e0fd | 1774 | /* |
9e5987a7 DC |
1775 | * Filling in all of a previously delayed allocation extent. |
1776 | * The right neighbor is contiguous, the left is not. | |
0937e0fd | 1777 | */ |
9e5987a7 DC |
1778 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); |
1779 | xfs_bmbt_set_startblock(ep, new->br_startblock); | |
1780 | xfs_bmbt_set_blockcount(ep, | |
1781 | PREV.br_blockcount + RIGHT.br_blockcount); | |
1782 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
0937e0fd | 1783 | |
9e5987a7 DC |
1784 | xfs_iext_remove(bma->ip, bma->idx + 1, 1, state); |
1785 | if (bma->cur == NULL) | |
1786 | rval = XFS_ILOG_DEXT; | |
1787 | else { | |
1788 | rval = 0; | |
1789 | error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff, | |
1790 | RIGHT.br_startblock, | |
1791 | RIGHT.br_blockcount, &i); | |
1792 | if (error) | |
1793 | goto done; | |
c29aad41 | 1794 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1795 | error = xfs_bmbt_update(bma->cur, PREV.br_startoff, |
1796 | new->br_startblock, | |
1797 | PREV.br_blockcount + | |
1798 | RIGHT.br_blockcount, PREV.br_state); | |
1799 | if (error) | |
1800 | goto done; | |
1801 | } | |
1802 | break; | |
1803 | ||
1804 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
a365bdd5 | 1805 | /* |
9e5987a7 DC |
1806 | * Filling in all of a previously delayed allocation extent. |
1807 | * Neither the left nor right neighbors are contiguous with | |
1808 | * the new one. | |
a365bdd5 | 1809 | */ |
9e5987a7 DC |
1810 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); |
1811 | xfs_bmbt_set_startblock(ep, new->br_startblock); | |
05a630d7 | 1812 | xfs_bmbt_set_state(ep, new->br_state); |
9e5987a7 | 1813 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); |
a365bdd5 | 1814 | |
60b4984f | 1815 | (*nextents)++; |
9e5987a7 DC |
1816 | if (bma->cur == NULL) |
1817 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1818 | else { | |
1819 | rval = XFS_ILOG_CORE; | |
1820 | error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff, | |
1821 | new->br_startblock, new->br_blockcount, | |
1822 | &i); | |
1823 | if (error) | |
1824 | goto done; | |
c29aad41 | 1825 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
1826 | bma->cur->bc_rec.b.br_state = XFS_EXT_NORM; |
1827 | error = xfs_btree_insert(bma->cur, &i); | |
1828 | if (error) | |
1829 | goto done; | |
c29aad41 | 1830 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1831 | } |
1832 | break; | |
1da177e4 | 1833 | |
9e5987a7 | 1834 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
1da177e4 | 1835 | /* |
9e5987a7 DC |
1836 | * Filling in the first part of a previous delayed allocation. |
1837 | * The left neighbor is contiguous. | |
1da177e4 | 1838 | */ |
9e5987a7 DC |
1839 | trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_); |
1840 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1), | |
1841 | LEFT.br_blockcount + new->br_blockcount); | |
1842 | xfs_bmbt_set_startoff(ep, | |
1843 | PREV.br_startoff + new->br_blockcount); | |
1844 | trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_); | |
1da177e4 | 1845 | |
9e5987a7 DC |
1846 | temp = PREV.br_blockcount - new->br_blockcount; |
1847 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1848 | xfs_bmbt_set_blockcount(ep, temp); | |
1849 | if (bma->cur == NULL) | |
1850 | rval = XFS_ILOG_DEXT; | |
1851 | else { | |
1852 | rval = 0; | |
1853 | error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff, | |
1854 | LEFT.br_startblock, LEFT.br_blockcount, | |
1855 | &i); | |
1856 | if (error) | |
1857 | goto done; | |
c29aad41 | 1858 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1859 | error = xfs_bmbt_update(bma->cur, LEFT.br_startoff, |
1860 | LEFT.br_startblock, | |
1861 | LEFT.br_blockcount + | |
1862 | new->br_blockcount, | |
1863 | LEFT.br_state); | |
f3ca8738 | 1864 | if (error) |
1da177e4 | 1865 | goto done; |
1da177e4 | 1866 | } |
9e5987a7 DC |
1867 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1868 | startblockval(PREV.br_startblock)); | |
1869 | xfs_bmbt_set_startblock(ep, nullstartblock(da_new)); | |
1870 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1871 | ||
1872 | bma->idx--; | |
1873 | break; | |
1874 | ||
1875 | case BMAP_LEFT_FILLING: | |
1da177e4 | 1876 | /* |
9e5987a7 DC |
1877 | * Filling in the first part of a previous delayed allocation. |
1878 | * The left neighbor is not contiguous. | |
1da177e4 | 1879 | */ |
9e5987a7 DC |
1880 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); |
1881 | xfs_bmbt_set_startoff(ep, new_endoff); | |
1882 | temp = PREV.br_blockcount - new->br_blockcount; | |
1883 | xfs_bmbt_set_blockcount(ep, temp); | |
1884 | xfs_iext_insert(bma->ip, bma->idx, 1, new, state); | |
60b4984f | 1885 | (*nextents)++; |
9e5987a7 DC |
1886 | if (bma->cur == NULL) |
1887 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1da177e4 | 1888 | else { |
9e5987a7 DC |
1889 | rval = XFS_ILOG_CORE; |
1890 | error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff, | |
1891 | new->br_startblock, new->br_blockcount, | |
1892 | &i); | |
1893 | if (error) | |
1894 | goto done; | |
c29aad41 | 1895 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
1896 | bma->cur->bc_rec.b.br_state = XFS_EXT_NORM; |
1897 | error = xfs_btree_insert(bma->cur, &i); | |
1898 | if (error) | |
1da177e4 | 1899 | goto done; |
c29aad41 | 1900 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
1da177e4 | 1901 | } |
233eebb9 | 1902 | |
6d3eb1ec | 1903 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
9e5987a7 | 1904 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
2c3234d1 | 1905 | bma->firstblock, bma->dfops, |
6d3eb1ec | 1906 | &bma->cur, 1, &tmp_rval, whichfork); |
9e5987a7 DC |
1907 | rval |= tmp_rval; |
1908 | if (error) | |
1909 | goto done; | |
1da177e4 | 1910 | } |
9e5987a7 DC |
1911 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1912 | startblockval(PREV.br_startblock) - | |
1913 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
1914 | ep = xfs_iext_get_ext(ifp, bma->idx + 1); | |
1915 | xfs_bmbt_set_startblock(ep, nullstartblock(da_new)); | |
1916 | trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_); | |
1da177e4 LT |
1917 | break; |
1918 | ||
9e5987a7 | 1919 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
1da177e4 | 1920 | /* |
9e5987a7 DC |
1921 | * Filling in the last part of a previous delayed allocation. |
1922 | * The right neighbor is contiguous with the new allocation. | |
1da177e4 | 1923 | */ |
9e5987a7 DC |
1924 | temp = PREV.br_blockcount - new->br_blockcount; |
1925 | trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_); | |
1da177e4 | 1926 | xfs_bmbt_set_blockcount(ep, temp); |
9e5987a7 DC |
1927 | xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1), |
1928 | new->br_startoff, new->br_startblock, | |
1929 | new->br_blockcount + RIGHT.br_blockcount, | |
1930 | RIGHT.br_state); | |
1931 | trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_); | |
1932 | if (bma->cur == NULL) | |
1933 | rval = XFS_ILOG_DEXT; | |
1934 | else { | |
1935 | rval = 0; | |
1936 | error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff, | |
1937 | RIGHT.br_startblock, | |
1938 | RIGHT.br_blockcount, &i); | |
1939 | if (error) | |
1940 | goto done; | |
c29aad41 | 1941 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1942 | error = xfs_bmbt_update(bma->cur, new->br_startoff, |
1943 | new->br_startblock, | |
1944 | new->br_blockcount + | |
1945 | RIGHT.br_blockcount, | |
1946 | RIGHT.br_state); | |
1947 | if (error) | |
1948 | goto done; | |
1da177e4 | 1949 | } |
9e5987a7 DC |
1950 | |
1951 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
1952 | startblockval(PREV.br_startblock)); | |
1953 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1954 | xfs_bmbt_set_startblock(ep, nullstartblock(da_new)); | |
1955 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1956 | ||
1957 | bma->idx++; | |
1da177e4 LT |
1958 | break; |
1959 | ||
9e5987a7 | 1960 | case BMAP_RIGHT_FILLING: |
1da177e4 | 1961 | /* |
9e5987a7 DC |
1962 | * Filling in the last part of a previous delayed allocation. |
1963 | * The right neighbor is not contiguous. | |
1da177e4 | 1964 | */ |
9e5987a7 DC |
1965 | temp = PREV.br_blockcount - new->br_blockcount; |
1966 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1da177e4 | 1967 | xfs_bmbt_set_blockcount(ep, temp); |
9e5987a7 | 1968 | xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state); |
60b4984f | 1969 | (*nextents)++; |
9e5987a7 DC |
1970 | if (bma->cur == NULL) |
1971 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1972 | else { | |
1973 | rval = XFS_ILOG_CORE; | |
1974 | error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff, | |
1975 | new->br_startblock, new->br_blockcount, | |
1976 | &i); | |
1977 | if (error) | |
1978 | goto done; | |
c29aad41 | 1979 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
1980 | bma->cur->bc_rec.b.br_state = XFS_EXT_NORM; |
1981 | error = xfs_btree_insert(bma->cur, &i); | |
1982 | if (error) | |
1983 | goto done; | |
c29aad41 | 1984 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
1da177e4 | 1985 | } |
9e5987a7 | 1986 | |
6d3eb1ec | 1987 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
9e5987a7 | 1988 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
2c3234d1 | 1989 | bma->firstblock, bma->dfops, &bma->cur, 1, |
6d3eb1ec | 1990 | &tmp_rval, whichfork); |
9e5987a7 DC |
1991 | rval |= tmp_rval; |
1992 | if (error) | |
1993 | goto done; | |
1da177e4 | 1994 | } |
9e5987a7 DC |
1995 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1996 | startblockval(PREV.br_startblock) - | |
1997 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
1998 | ep = xfs_iext_get_ext(ifp, bma->idx); | |
1999 | xfs_bmbt_set_startblock(ep, nullstartblock(da_new)); | |
2000 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
2001 | ||
2002 | bma->idx++; | |
1da177e4 LT |
2003 | break; |
2004 | ||
2005 | case 0: | |
2006 | /* | |
9e5987a7 DC |
2007 | * Filling in the middle part of a previous delayed allocation. |
2008 | * Contiguity is impossible here. | |
2009 | * This case is avoided almost all the time. | |
2010 | * | |
2011 | * We start with a delayed allocation: | |
2012 | * | |
2013 | * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+ | |
2014 | * PREV @ idx | |
2015 | * | |
2016 | * and we are allocating: | |
2017 | * +rrrrrrrrrrrrrrrrr+ | |
2018 | * new | |
2019 | * | |
2020 | * and we set it up for insertion as: | |
2021 | * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+ | |
2022 | * new | |
2023 | * PREV @ idx LEFT RIGHT | |
2024 | * inserted at idx + 1 | |
1da177e4 | 2025 | */ |
9e5987a7 DC |
2026 | temp = new->br_startoff - PREV.br_startoff; |
2027 | temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
2028 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_); | |
2029 | xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */ | |
2030 | LEFT = *new; | |
2031 | RIGHT.br_state = PREV.br_state; | |
2032 | RIGHT.br_startblock = nullstartblock( | |
2033 | (int)xfs_bmap_worst_indlen(bma->ip, temp2)); | |
2034 | RIGHT.br_startoff = new_endoff; | |
2035 | RIGHT.br_blockcount = temp2; | |
2036 | /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */ | |
2037 | xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state); | |
60b4984f | 2038 | (*nextents)++; |
9e5987a7 DC |
2039 | if (bma->cur == NULL) |
2040 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2041 | else { | |
2042 | rval = XFS_ILOG_CORE; | |
2043 | error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff, | |
2044 | new->br_startblock, new->br_blockcount, | |
2045 | &i); | |
2046 | if (error) | |
2047 | goto done; | |
c29aad41 | 2048 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
2049 | bma->cur->bc_rec.b.br_state = XFS_EXT_NORM; |
2050 | error = xfs_btree_insert(bma->cur, &i); | |
2051 | if (error) | |
2052 | goto done; | |
c29aad41 | 2053 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
1da177e4 | 2054 | } |
1da177e4 | 2055 | |
6d3eb1ec | 2056 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
9e5987a7 | 2057 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
2c3234d1 | 2058 | bma->firstblock, bma->dfops, &bma->cur, |
6d3eb1ec | 2059 | 1, &tmp_rval, whichfork); |
9e5987a7 DC |
2060 | rval |= tmp_rval; |
2061 | if (error) | |
2062 | goto done; | |
2063 | } | |
2064 | temp = xfs_bmap_worst_indlen(bma->ip, temp); | |
2065 | temp2 = xfs_bmap_worst_indlen(bma->ip, temp2); | |
0daaecac BF |
2066 | diff = (int)(temp + temp2 - |
2067 | (startblockval(PREV.br_startblock) - | |
2068 | (bma->cur ? | |
2069 | bma->cur->bc_private.b.allocated : 0))); | |
9e5987a7 | 2070 | if (diff > 0) { |
0d485ada DC |
2071 | error = xfs_mod_fdblocks(bma->ip->i_mount, |
2072 | -((int64_t)diff), false); | |
9e5987a7 DC |
2073 | ASSERT(!error); |
2074 | if (error) | |
2075 | goto done; | |
2076 | } | |
2077 | ||
2078 | ep = xfs_iext_get_ext(ifp, bma->idx); | |
2079 | xfs_bmbt_set_startblock(ep, nullstartblock((int)temp)); | |
2080 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
2081 | trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_); | |
2082 | xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2), | |
2083 | nullstartblock((int)temp2)); | |
2084 | trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_); | |
2085 | ||
2086 | bma->idx++; | |
2087 | da_new = temp + temp2; | |
2088 | break; | |
2089 | ||
2090 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2091 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2092 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2093 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2094 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2095 | case BMAP_LEFT_CONTIG: | |
2096 | case BMAP_RIGHT_CONTIG: | |
2097 | /* | |
2098 | * These cases are all impossible. | |
2099 | */ | |
2100 | ASSERT(0); | |
2101 | } | |
2102 | ||
9c194644 DW |
2103 | /* add reverse mapping */ |
2104 | error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new); | |
2105 | if (error) | |
2106 | goto done; | |
2107 | ||
9e5987a7 | 2108 | /* convert to a btree if necessary */ |
6d3eb1ec | 2109 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
9e5987a7 DC |
2110 | int tmp_logflags; /* partial log flag return val */ |
2111 | ||
2112 | ASSERT(bma->cur == NULL); | |
2113 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, | |
2c3234d1 | 2114 | bma->firstblock, bma->dfops, &bma->cur, |
6d3eb1ec | 2115 | da_old > 0, &tmp_logflags, whichfork); |
9e5987a7 DC |
2116 | bma->logflags |= tmp_logflags; |
2117 | if (error) | |
2118 | goto done; | |
2119 | } | |
2120 | ||
2121 | /* adjust for changes in reserved delayed indirect blocks */ | |
2122 | if (da_old || da_new) { | |
2123 | temp = da_new; | |
2124 | if (bma->cur) | |
2125 | temp += bma->cur->bc_private.b.allocated; | |
9e5987a7 | 2126 | if (temp < da_old) |
0d485ada DC |
2127 | xfs_mod_fdblocks(bma->ip->i_mount, |
2128 | (int64_t)(da_old - temp), false); | |
96540c78 | 2129 | } |
9e5987a7 DC |
2130 | |
2131 | /* clear out the allocated field, done with it now in any case. */ | |
2132 | if (bma->cur) | |
2133 | bma->cur->bc_private.b.allocated = 0; | |
2134 | ||
6d3eb1ec | 2135 | xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork); |
1da177e4 | 2136 | done: |
60b4984f DW |
2137 | if (whichfork != XFS_COW_FORK) |
2138 | bma->logflags |= rval; | |
1da177e4 | 2139 | return error; |
9e5987a7 DC |
2140 | #undef LEFT |
2141 | #undef RIGHT | |
2142 | #undef PREV | |
1da177e4 LT |
2143 | } |
2144 | ||
2145 | /* | |
9e5987a7 | 2146 | * Convert an unwritten allocation to a real allocation or vice versa. |
1da177e4 | 2147 | */ |
9e5987a7 DC |
2148 | STATIC int /* error */ |
2149 | xfs_bmap_add_extent_unwritten_real( | |
2150 | struct xfs_trans *tp, | |
2151 | xfs_inode_t *ip, /* incore inode pointer */ | |
05a630d7 | 2152 | int whichfork, |
9e5987a7 DC |
2153 | xfs_extnum_t *idx, /* extent number to update/insert */ |
2154 | xfs_btree_cur_t **curp, /* if *curp is null, not a btree */ | |
2155 | xfs_bmbt_irec_t *new, /* new data to add to file extents */ | |
2156 | xfs_fsblock_t *first, /* pointer to firstblock variable */ | |
2c3234d1 | 2157 | struct xfs_defer_ops *dfops, /* list of extents to be freed */ |
9e5987a7 | 2158 | int *logflagsp) /* inode logging flags */ |
1da177e4 | 2159 | { |
9e5987a7 DC |
2160 | xfs_btree_cur_t *cur; /* btree cursor */ |
2161 | xfs_bmbt_rec_host_t *ep; /* extent entry for idx */ | |
2162 | int error; /* error return value */ | |
2163 | int i; /* temp state */ | |
2164 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
2165 | xfs_fileoff_t new_endoff; /* end offset of new entry */ | |
2166 | xfs_exntst_t newext; /* new extent state */ | |
2167 | xfs_exntst_t oldext; /* old extent state */ | |
2168 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
2169 | /* left is 0, right is 1, prev is 2 */ | |
2170 | int rval=0; /* return value (logging flags) */ | |
2171 | int state = 0;/* state bits, accessed thru macros */ | |
05a630d7 | 2172 | struct xfs_mount *mp = ip->i_mount; |
1da177e4 | 2173 | |
9e5987a7 | 2174 | *logflagsp = 0; |
1da177e4 | 2175 | |
9e5987a7 | 2176 | cur = *curp; |
05a630d7 DW |
2177 | ifp = XFS_IFORK_PTR(ip, whichfork); |
2178 | if (whichfork == XFS_COW_FORK) | |
2179 | state |= BMAP_COWFORK; | |
8096b1eb | 2180 | |
9e5987a7 | 2181 | ASSERT(*idx >= 0); |
5d829300 | 2182 | ASSERT(*idx <= xfs_iext_count(ifp)); |
9e5987a7 DC |
2183 | ASSERT(!isnullstartblock(new->br_startblock)); |
2184 | ||
ff6d6af2 | 2185 | XFS_STATS_INC(mp, xs_add_exlist); |
9e5987a7 DC |
2186 | |
2187 | #define LEFT r[0] | |
2188 | #define RIGHT r[1] | |
2189 | #define PREV r[2] | |
7cc95a82 | 2190 | |
1da177e4 | 2191 | /* |
9e5987a7 | 2192 | * Set up a bunch of variables to make the tests simpler. |
1da177e4 | 2193 | */ |
9e5987a7 DC |
2194 | error = 0; |
2195 | ep = xfs_iext_get_ext(ifp, *idx); | |
2196 | xfs_bmbt_get_all(ep, &PREV); | |
2197 | newext = new->br_state; | |
2198 | oldext = (newext == XFS_EXT_UNWRITTEN) ? | |
2199 | XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
2200 | ASSERT(PREV.br_state == oldext); | |
2201 | new_endoff = new->br_startoff + new->br_blockcount; | |
2202 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
2203 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
7cc95a82 | 2204 | |
1da177e4 | 2205 | /* |
9e5987a7 DC |
2206 | * Set flags determining what part of the previous oldext allocation |
2207 | * extent is being replaced by a newext allocation. | |
1da177e4 | 2208 | */ |
9e5987a7 DC |
2209 | if (PREV.br_startoff == new->br_startoff) |
2210 | state |= BMAP_LEFT_FILLING; | |
2211 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
2212 | state |= BMAP_RIGHT_FILLING; | |
2213 | ||
1da177e4 | 2214 | /* |
9e5987a7 DC |
2215 | * Check and set flags if this segment has a left neighbor. |
2216 | * Don't set contiguous if the combined extent would be too large. | |
1da177e4 | 2217 | */ |
9e5987a7 DC |
2218 | if (*idx > 0) { |
2219 | state |= BMAP_LEFT_VALID; | |
2220 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT); | |
2221 | ||
2222 | if (isnullstartblock(LEFT.br_startblock)) | |
2223 | state |= BMAP_LEFT_DELAY; | |
1da177e4 | 2224 | } |
9e5987a7 DC |
2225 | |
2226 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2227 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
2228 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
2229 | LEFT.br_state == newext && | |
2230 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2231 | state |= BMAP_LEFT_CONTIG; | |
2232 | ||
1da177e4 | 2233 | /* |
9e5987a7 DC |
2234 | * Check and set flags if this segment has a right neighbor. |
2235 | * Don't set contiguous if the combined extent would be too large. | |
2236 | * Also check for all-three-contiguous being too large. | |
1da177e4 | 2237 | */ |
05a630d7 | 2238 | if (*idx < xfs_iext_count(ifp) - 1) { |
9e5987a7 DC |
2239 | state |= BMAP_RIGHT_VALID; |
2240 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT); | |
2241 | if (isnullstartblock(RIGHT.br_startblock)) | |
2242 | state |= BMAP_RIGHT_DELAY; | |
1da177e4 | 2243 | } |
7cc95a82 | 2244 | |
9e5987a7 DC |
2245 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2246 | new_endoff == RIGHT.br_startoff && | |
2247 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
2248 | newext == RIGHT.br_state && | |
2249 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
2250 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2251 | BMAP_RIGHT_FILLING)) != | |
2252 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2253 | BMAP_RIGHT_FILLING) || | |
2254 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
2255 | <= MAXEXTLEN)) | |
2256 | state |= BMAP_RIGHT_CONTIG; | |
136341b4 | 2257 | |
1da177e4 | 2258 | /* |
9e5987a7 | 2259 | * Switch out based on the FILLING and CONTIG state bits. |
1da177e4 | 2260 | */ |
9e5987a7 DC |
2261 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | |
2262 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
2263 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2264 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
2265 | /* | |
2266 | * Setting all of a previous oldext extent to newext. | |
2267 | * The left and right neighbors are both contiguous with new. | |
2268 | */ | |
2269 | --*idx; | |
1a5902c5 | 2270 | |
9e5987a7 DC |
2271 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2272 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), | |
2273 | LEFT.br_blockcount + PREV.br_blockcount + | |
2274 | RIGHT.br_blockcount); | |
2275 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
1a5902c5 | 2276 | |
9e5987a7 | 2277 | xfs_iext_remove(ip, *idx + 1, 2, state); |
05a630d7 DW |
2278 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2279 | XFS_IFORK_NEXTENTS(ip, whichfork) - 2); | |
9e5987a7 DC |
2280 | if (cur == NULL) |
2281 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2282 | else { | |
2283 | rval = XFS_ILOG_CORE; | |
2284 | if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff, | |
2285 | RIGHT.br_startblock, | |
2286 | RIGHT.br_blockcount, &i))) | |
2287 | goto done; | |
c29aad41 | 2288 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2289 | if ((error = xfs_btree_delete(cur, &i))) |
2290 | goto done; | |
c29aad41 | 2291 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2292 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2293 | goto done; | |
c29aad41 | 2294 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2295 | if ((error = xfs_btree_delete(cur, &i))) |
2296 | goto done; | |
c29aad41 | 2297 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2298 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2299 | goto done; | |
c29aad41 | 2300 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2301 | if ((error = xfs_bmbt_update(cur, LEFT.br_startoff, |
2302 | LEFT.br_startblock, | |
2303 | LEFT.br_blockcount + PREV.br_blockcount + | |
2304 | RIGHT.br_blockcount, LEFT.br_state))) | |
2305 | goto done; | |
2306 | } | |
2307 | break; | |
1a5902c5 | 2308 | |
9e5987a7 DC |
2309 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: |
2310 | /* | |
2311 | * Setting all of a previous oldext extent to newext. | |
2312 | * The left neighbor is contiguous, the right is not. | |
2313 | */ | |
2314 | --*idx; | |
d8cc890d | 2315 | |
9e5987a7 DC |
2316 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2317 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), | |
2318 | LEFT.br_blockcount + PREV.br_blockcount); | |
2319 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
1da177e4 | 2320 | |
9e5987a7 | 2321 | xfs_iext_remove(ip, *idx + 1, 1, state); |
05a630d7 DW |
2322 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2323 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
9e5987a7 DC |
2324 | if (cur == NULL) |
2325 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2326 | else { | |
2327 | rval = XFS_ILOG_CORE; | |
2328 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2329 | PREV.br_startblock, PREV.br_blockcount, | |
2330 | &i))) | |
2331 | goto done; | |
c29aad41 | 2332 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2333 | if ((error = xfs_btree_delete(cur, &i))) |
2334 | goto done; | |
c29aad41 | 2335 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2336 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2337 | goto done; | |
c29aad41 | 2338 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2339 | if ((error = xfs_bmbt_update(cur, LEFT.br_startoff, |
2340 | LEFT.br_startblock, | |
2341 | LEFT.br_blockcount + PREV.br_blockcount, | |
2342 | LEFT.br_state))) | |
2343 | goto done; | |
2344 | } | |
2345 | break; | |
1da177e4 | 2346 | |
9e5987a7 | 2347 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
1da177e4 | 2348 | /* |
9e5987a7 DC |
2349 | * Setting all of a previous oldext extent to newext. |
2350 | * The right neighbor is contiguous, the left is not. | |
1da177e4 | 2351 | */ |
9e5987a7 DC |
2352 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2353 | xfs_bmbt_set_blockcount(ep, | |
2354 | PREV.br_blockcount + RIGHT.br_blockcount); | |
2355 | xfs_bmbt_set_state(ep, newext); | |
2356 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2357 | xfs_iext_remove(ip, *idx + 1, 1, state); | |
05a630d7 DW |
2358 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2359 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
9e5987a7 DC |
2360 | if (cur == NULL) |
2361 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2362 | else { | |
2363 | rval = XFS_ILOG_CORE; | |
2364 | if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff, | |
2365 | RIGHT.br_startblock, | |
2366 | RIGHT.br_blockcount, &i))) | |
2367 | goto done; | |
c29aad41 | 2368 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2369 | if ((error = xfs_btree_delete(cur, &i))) |
2370 | goto done; | |
c29aad41 | 2371 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2372 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2373 | goto done; | |
c29aad41 | 2374 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2375 | if ((error = xfs_bmbt_update(cur, new->br_startoff, |
2376 | new->br_startblock, | |
2377 | new->br_blockcount + RIGHT.br_blockcount, | |
2378 | newext))) | |
2379 | goto done; | |
1da177e4 | 2380 | } |
9e5987a7 | 2381 | break; |
1e82379b | 2382 | |
9e5987a7 DC |
2383 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: |
2384 | /* | |
2385 | * Setting all of a previous oldext extent to newext. | |
2386 | * Neither the left nor right neighbors are contiguous with | |
2387 | * the new one. | |
2388 | */ | |
2389 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2390 | xfs_bmbt_set_state(ep, newext); | |
2391 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
1e82379b | 2392 | |
9e5987a7 DC |
2393 | if (cur == NULL) |
2394 | rval = XFS_ILOG_DEXT; | |
2395 | else { | |
2396 | rval = 0; | |
2397 | if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff, | |
2398 | new->br_startblock, new->br_blockcount, | |
2399 | &i))) | |
2400 | goto done; | |
c29aad41 | 2401 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2402 | if ((error = xfs_bmbt_update(cur, new->br_startoff, |
2403 | new->br_startblock, new->br_blockcount, | |
2404 | newext))) | |
2405 | goto done; | |
2406 | } | |
2407 | break; | |
1e82379b | 2408 | |
9e5987a7 DC |
2409 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
2410 | /* | |
2411 | * Setting the first part of a previous oldext extent to newext. | |
2412 | * The left neighbor is contiguous. | |
2413 | */ | |
2414 | trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_); | |
2415 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1), | |
2416 | LEFT.br_blockcount + new->br_blockcount); | |
2417 | xfs_bmbt_set_startoff(ep, | |
2418 | PREV.br_startoff + new->br_blockcount); | |
2419 | trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_); | |
1da177e4 | 2420 | |
9e5987a7 DC |
2421 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2422 | xfs_bmbt_set_startblock(ep, | |
2423 | new->br_startblock + new->br_blockcount); | |
2424 | xfs_bmbt_set_blockcount(ep, | |
2425 | PREV.br_blockcount - new->br_blockcount); | |
2426 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
0293ce3a | 2427 | |
9e5987a7 | 2428 | --*idx; |
8867bc9b | 2429 | |
9e5987a7 DC |
2430 | if (cur == NULL) |
2431 | rval = XFS_ILOG_DEXT; | |
2432 | else { | |
2433 | rval = 0; | |
2434 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2435 | PREV.br_startblock, PREV.br_blockcount, | |
2436 | &i))) | |
2437 | goto done; | |
c29aad41 | 2438 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2439 | if ((error = xfs_bmbt_update(cur, |
2440 | PREV.br_startoff + new->br_blockcount, | |
2441 | PREV.br_startblock + new->br_blockcount, | |
2442 | PREV.br_blockcount - new->br_blockcount, | |
2443 | oldext))) | |
2444 | goto done; | |
2445 | if ((error = xfs_btree_decrement(cur, 0, &i))) | |
2446 | goto done; | |
2447 | error = xfs_bmbt_update(cur, LEFT.br_startoff, | |
2448 | LEFT.br_startblock, | |
2449 | LEFT.br_blockcount + new->br_blockcount, | |
2450 | LEFT.br_state); | |
2451 | if (error) | |
2452 | goto done; | |
8867bc9b | 2453 | } |
9e5987a7 | 2454 | break; |
0293ce3a | 2455 | |
9e5987a7 DC |
2456 | case BMAP_LEFT_FILLING: |
2457 | /* | |
2458 | * Setting the first part of a previous oldext extent to newext. | |
2459 | * The left neighbor is not contiguous. | |
2460 | */ | |
2461 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2462 | ASSERT(ep && xfs_bmbt_get_state(ep) == oldext); | |
2463 | xfs_bmbt_set_startoff(ep, new_endoff); | |
2464 | xfs_bmbt_set_blockcount(ep, | |
2465 | PREV.br_blockcount - new->br_blockcount); | |
2466 | xfs_bmbt_set_startblock(ep, | |
2467 | new->br_startblock + new->br_blockcount); | |
2468 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
1da177e4 | 2469 | |
9e5987a7 | 2470 | xfs_iext_insert(ip, *idx, 1, new, state); |
05a630d7 DW |
2471 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2472 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
9e5987a7 DC |
2473 | if (cur == NULL) |
2474 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2475 | else { | |
2476 | rval = XFS_ILOG_CORE; | |
2477 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2478 | PREV.br_startblock, PREV.br_blockcount, | |
2479 | &i))) | |
2480 | goto done; | |
c29aad41 | 2481 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2482 | if ((error = xfs_bmbt_update(cur, |
2483 | PREV.br_startoff + new->br_blockcount, | |
2484 | PREV.br_startblock + new->br_blockcount, | |
2485 | PREV.br_blockcount - new->br_blockcount, | |
2486 | oldext))) | |
2487 | goto done; | |
2488 | cur->bc_rec.b = *new; | |
2489 | if ((error = xfs_btree_insert(cur, &i))) | |
2490 | goto done; | |
c29aad41 | 2491 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2492 | } |
2493 | break; | |
1da177e4 | 2494 | |
9e5987a7 DC |
2495 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2496 | /* | |
2497 | * Setting the last part of a previous oldext extent to newext. | |
2498 | * The right neighbor is contiguous with the new allocation. | |
2499 | */ | |
2500 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2501 | xfs_bmbt_set_blockcount(ep, | |
2502 | PREV.br_blockcount - new->br_blockcount); | |
2503 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
0293ce3a | 2504 | |
9e5987a7 | 2505 | ++*idx; |
1da177e4 | 2506 | |
9e5987a7 DC |
2507 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2508 | xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx), | |
2509 | new->br_startoff, new->br_startblock, | |
2510 | new->br_blockcount + RIGHT.br_blockcount, newext); | |
2511 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
1da177e4 | 2512 | |
9e5987a7 DC |
2513 | if (cur == NULL) |
2514 | rval = XFS_ILOG_DEXT; | |
2515 | else { | |
2516 | rval = 0; | |
2517 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2518 | PREV.br_startblock, | |
2519 | PREV.br_blockcount, &i))) | |
2520 | goto done; | |
c29aad41 | 2521 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2522 | if ((error = xfs_bmbt_update(cur, PREV.br_startoff, |
2523 | PREV.br_startblock, | |
2524 | PREV.br_blockcount - new->br_blockcount, | |
2525 | oldext))) | |
2526 | goto done; | |
2527 | if ((error = xfs_btree_increment(cur, 0, &i))) | |
2528 | goto done; | |
2529 | if ((error = xfs_bmbt_update(cur, new->br_startoff, | |
2530 | new->br_startblock, | |
2531 | new->br_blockcount + RIGHT.br_blockcount, | |
2532 | newext))) | |
2533 | goto done; | |
2534 | } | |
2535 | break; | |
1da177e4 | 2536 | |
9e5987a7 DC |
2537 | case BMAP_RIGHT_FILLING: |
2538 | /* | |
2539 | * Setting the last part of a previous oldext extent to newext. | |
2540 | * The right neighbor is not contiguous. | |
2541 | */ | |
2542 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2543 | xfs_bmbt_set_blockcount(ep, | |
2544 | PREV.br_blockcount - new->br_blockcount); | |
2545 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
1da177e4 | 2546 | |
9e5987a7 DC |
2547 | ++*idx; |
2548 | xfs_iext_insert(ip, *idx, 1, new, state); | |
d8cc890d | 2549 | |
05a630d7 DW |
2550 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2551 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
9e5987a7 DC |
2552 | if (cur == NULL) |
2553 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2554 | else { | |
2555 | rval = XFS_ILOG_CORE; | |
2556 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2557 | PREV.br_startblock, PREV.br_blockcount, | |
2558 | &i))) | |
2559 | goto done; | |
c29aad41 | 2560 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2561 | if ((error = xfs_bmbt_update(cur, PREV.br_startoff, |
2562 | PREV.br_startblock, | |
2563 | PREV.br_blockcount - new->br_blockcount, | |
2564 | oldext))) | |
2565 | goto done; | |
2566 | if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff, | |
2567 | new->br_startblock, new->br_blockcount, | |
2568 | &i))) | |
2569 | goto done; | |
c29aad41 | 2570 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
2571 | cur->bc_rec.b.br_state = XFS_EXT_NORM; |
2572 | if ((error = xfs_btree_insert(cur, &i))) | |
2573 | goto done; | |
c29aad41 | 2574 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2575 | } |
2576 | break; | |
2577 | ||
2578 | case 0: | |
1da177e4 | 2579 | /* |
9e5987a7 DC |
2580 | * Setting the middle part of a previous oldext extent to |
2581 | * newext. Contiguity is impossible here. | |
2582 | * One extent becomes three extents. | |
1da177e4 | 2583 | */ |
9e5987a7 DC |
2584 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2585 | xfs_bmbt_set_blockcount(ep, | |
2586 | new->br_startoff - PREV.br_startoff); | |
2587 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
898621d5 | 2588 | |
9e5987a7 DC |
2589 | r[0] = *new; |
2590 | r[1].br_startoff = new_endoff; | |
2591 | r[1].br_blockcount = | |
2592 | PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
2593 | r[1].br_startblock = new->br_startblock + new->br_blockcount; | |
2594 | r[1].br_state = oldext; | |
898621d5 | 2595 | |
9e5987a7 DC |
2596 | ++*idx; |
2597 | xfs_iext_insert(ip, *idx, 2, &r[0], state); | |
2598 | ||
05a630d7 DW |
2599 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2600 | XFS_IFORK_NEXTENTS(ip, whichfork) + 2); | |
9e5987a7 DC |
2601 | if (cur == NULL) |
2602 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2603 | else { | |
2604 | rval = XFS_ILOG_CORE; | |
2605 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2606 | PREV.br_startblock, PREV.br_blockcount, | |
2607 | &i))) | |
2608 | goto done; | |
c29aad41 | 2609 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2610 | /* new right extent - oldext */ |
2611 | if ((error = xfs_bmbt_update(cur, r[1].br_startoff, | |
2612 | r[1].br_startblock, r[1].br_blockcount, | |
2613 | r[1].br_state))) | |
2614 | goto done; | |
2615 | /* new left extent - oldext */ | |
2616 | cur->bc_rec.b = PREV; | |
2617 | cur->bc_rec.b.br_blockcount = | |
2618 | new->br_startoff - PREV.br_startoff; | |
2619 | if ((error = xfs_btree_insert(cur, &i))) | |
2620 | goto done; | |
c29aad41 | 2621 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2622 | /* |
2623 | * Reset the cursor to the position of the new extent | |
2624 | * we are about to insert as we can't trust it after | |
2625 | * the previous insert. | |
2626 | */ | |
2627 | if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff, | |
2628 | new->br_startblock, new->br_blockcount, | |
2629 | &i))) | |
2630 | goto done; | |
c29aad41 | 2631 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
2632 | /* new middle extent - newext */ |
2633 | cur->bc_rec.b.br_state = new->br_state; | |
2634 | if ((error = xfs_btree_insert(cur, &i))) | |
2635 | goto done; | |
c29aad41 | 2636 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 | 2637 | } |
1da177e4 | 2638 | break; |
9e5987a7 DC |
2639 | |
2640 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2641 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2642 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2643 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2644 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2645 | case BMAP_LEFT_CONTIG: | |
2646 | case BMAP_RIGHT_CONTIG: | |
2647 | /* | |
2648 | * These cases are all impossible. | |
2649 | */ | |
1da177e4 | 2650 | ASSERT(0); |
1da177e4 | 2651 | } |
8096b1eb | 2652 | |
9c194644 | 2653 | /* update reverse mappings */ |
05a630d7 | 2654 | error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new); |
9c194644 DW |
2655 | if (error) |
2656 | goto done; | |
2657 | ||
9e5987a7 | 2658 | /* convert to a btree if necessary */ |
05a630d7 | 2659 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
9e5987a7 DC |
2660 | int tmp_logflags; /* partial log flag return val */ |
2661 | ||
2662 | ASSERT(cur == NULL); | |
2c3234d1 | 2663 | error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur, |
05a630d7 | 2664 | 0, &tmp_logflags, whichfork); |
9e5987a7 DC |
2665 | *logflagsp |= tmp_logflags; |
2666 | if (error) | |
2667 | goto done; | |
1da177e4 | 2668 | } |
da087bad | 2669 | |
9e5987a7 DC |
2670 | /* clear out the allocated field, done with it now in any case. */ |
2671 | if (cur) { | |
2672 | cur->bc_private.b.allocated = 0; | |
2673 | *curp = cur; | |
1da177e4 | 2674 | } |
8096b1eb | 2675 | |
05a630d7 | 2676 | xfs_bmap_check_leaf_extents(*curp, ip, whichfork); |
9e5987a7 DC |
2677 | done: |
2678 | *logflagsp |= rval; | |
1da177e4 | 2679 | return error; |
9e5987a7 DC |
2680 | #undef LEFT |
2681 | #undef RIGHT | |
2682 | #undef PREV | |
1da177e4 LT |
2683 | } |
2684 | ||
2685 | /* | |
9e5987a7 | 2686 | * Convert a hole to a delayed allocation. |
1da177e4 | 2687 | */ |
9e5987a7 DC |
2688 | STATIC void |
2689 | xfs_bmap_add_extent_hole_delay( | |
2690 | xfs_inode_t *ip, /* incore inode pointer */ | |
be51f811 | 2691 | int whichfork, |
9e5987a7 DC |
2692 | xfs_extnum_t *idx, /* extent number to update/insert */ |
2693 | xfs_bmbt_irec_t *new) /* new data to add to file extents */ | |
1da177e4 | 2694 | { |
9e5987a7 DC |
2695 | xfs_ifork_t *ifp; /* inode fork pointer */ |
2696 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ | |
2697 | xfs_filblks_t newlen=0; /* new indirect size */ | |
2698 | xfs_filblks_t oldlen=0; /* old indirect size */ | |
2699 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2700 | int state; /* state bits, accessed thru macros */ | |
2701 | xfs_filblks_t temp=0; /* temp for indirect calculations */ | |
1da177e4 | 2702 | |
be51f811 | 2703 | ifp = XFS_IFORK_PTR(ip, whichfork); |
9e5987a7 | 2704 | state = 0; |
be51f811 DW |
2705 | if (whichfork == XFS_COW_FORK) |
2706 | state |= BMAP_COWFORK; | |
9e5987a7 | 2707 | ASSERT(isnullstartblock(new->br_startblock)); |
1da177e4 LT |
2708 | |
2709 | /* | |
9e5987a7 | 2710 | * Check and set flags if this segment has a left neighbor |
1da177e4 | 2711 | */ |
9e5987a7 DC |
2712 | if (*idx > 0) { |
2713 | state |= BMAP_LEFT_VALID; | |
2714 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left); | |
2715 | ||
2716 | if (isnullstartblock(left.br_startblock)) | |
2717 | state |= BMAP_LEFT_DELAY; | |
1da177e4 | 2718 | } |
1da177e4 | 2719 | |
9e5987a7 DC |
2720 | /* |
2721 | * Check and set flags if the current (right) segment exists. | |
2722 | * If it doesn't exist, we're converting the hole at end-of-file. | |
2723 | */ | |
5d829300 | 2724 | if (*idx < xfs_iext_count(ifp)) { |
9e5987a7 DC |
2725 | state |= BMAP_RIGHT_VALID; |
2726 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right); | |
1da177e4 | 2727 | |
9e5987a7 DC |
2728 | if (isnullstartblock(right.br_startblock)) |
2729 | state |= BMAP_RIGHT_DELAY; | |
1da177e4 | 2730 | } |
9e5987a7 | 2731 | |
1da177e4 | 2732 | /* |
9e5987a7 DC |
2733 | * Set contiguity flags on the left and right neighbors. |
2734 | * Don't let extents get too large, even if the pieces are contiguous. | |
1da177e4 | 2735 | */ |
9e5987a7 DC |
2736 | if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) && |
2737 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2738 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2739 | state |= BMAP_LEFT_CONTIG; | |
2740 | ||
2741 | if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) && | |
2742 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2743 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2744 | (!(state & BMAP_LEFT_CONTIG) || | |
2745 | (left.br_blockcount + new->br_blockcount + | |
2746 | right.br_blockcount <= MAXEXTLEN))) | |
2747 | state |= BMAP_RIGHT_CONTIG; | |
cc09c0dc DC |
2748 | |
2749 | /* | |
9e5987a7 | 2750 | * Switch out based on the contiguity flags. |
cc09c0dc | 2751 | */ |
9e5987a7 DC |
2752 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { |
2753 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2754 | /* | |
2755 | * New allocation is contiguous with delayed allocations | |
2756 | * on the left and on the right. | |
2757 | * Merge all three into a single extent record. | |
2758 | */ | |
2759 | --*idx; | |
2760 | temp = left.br_blockcount + new->br_blockcount + | |
2761 | right.br_blockcount; | |
cc09c0dc | 2762 | |
9e5987a7 DC |
2763 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2764 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp); | |
2765 | oldlen = startblockval(left.br_startblock) + | |
2766 | startblockval(new->br_startblock) + | |
2767 | startblockval(right.br_startblock); | |
0e339ef8 BF |
2768 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2769 | oldlen); | |
9e5987a7 DC |
2770 | xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx), |
2771 | nullstartblock((int)newlen)); | |
2772 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
1da177e4 | 2773 | |
9e5987a7 DC |
2774 | xfs_iext_remove(ip, *idx + 1, 1, state); |
2775 | break; | |
1da177e4 | 2776 | |
9e5987a7 DC |
2777 | case BMAP_LEFT_CONTIG: |
2778 | /* | |
2779 | * New allocation is contiguous with a delayed allocation | |
2780 | * on the left. | |
2781 | * Merge the new allocation with the left neighbor. | |
2782 | */ | |
2783 | --*idx; | |
2784 | temp = left.br_blockcount + new->br_blockcount; | |
1da177e4 | 2785 | |
9e5987a7 DC |
2786 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2787 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp); | |
2788 | oldlen = startblockval(left.br_startblock) + | |
2789 | startblockval(new->br_startblock); | |
0e339ef8 BF |
2790 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2791 | oldlen); | |
9e5987a7 DC |
2792 | xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx), |
2793 | nullstartblock((int)newlen)); | |
2794 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2795 | break; | |
1da177e4 | 2796 | |
9e5987a7 DC |
2797 | case BMAP_RIGHT_CONTIG: |
2798 | /* | |
2799 | * New allocation is contiguous with a delayed allocation | |
2800 | * on the right. | |
2801 | * Merge the new allocation with the right neighbor. | |
2802 | */ | |
2803 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2804 | temp = new->br_blockcount + right.br_blockcount; | |
2805 | oldlen = startblockval(new->br_startblock) + | |
2806 | startblockval(right.br_startblock); | |
0e339ef8 BF |
2807 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2808 | oldlen); | |
9e5987a7 DC |
2809 | xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx), |
2810 | new->br_startoff, | |
2811 | nullstartblock((int)newlen), temp, right.br_state); | |
2812 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2813 | break; | |
2814 | ||
2815 | case 0: | |
2816 | /* | |
2817 | * New allocation is not contiguous with another | |
2818 | * delayed allocation. | |
2819 | * Insert a new entry. | |
2820 | */ | |
2821 | oldlen = newlen = 0; | |
2822 | xfs_iext_insert(ip, *idx, 1, new, state); | |
2823 | break; | |
1da177e4 | 2824 | } |
9e5987a7 DC |
2825 | if (oldlen != newlen) { |
2826 | ASSERT(oldlen > newlen); | |
0d485ada DC |
2827 | xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen), |
2828 | false); | |
1da177e4 | 2829 | /* |
9e5987a7 | 2830 | * Nothing to do for disk quota accounting here. |
1da177e4 | 2831 | */ |
1da177e4 | 2832 | } |
1da177e4 LT |
2833 | } |
2834 | ||
2835 | /* | |
9e5987a7 | 2836 | * Convert a hole to a real allocation. |
1da177e4 | 2837 | */ |
9e5987a7 DC |
2838 | STATIC int /* error */ |
2839 | xfs_bmap_add_extent_hole_real( | |
6d04558f CH |
2840 | struct xfs_trans *tp, |
2841 | struct xfs_inode *ip, | |
2842 | int whichfork, | |
2843 | xfs_extnum_t *idx, | |
2844 | struct xfs_btree_cur **curp, | |
2845 | struct xfs_bmbt_irec *new, | |
2846 | xfs_fsblock_t *first, | |
2847 | struct xfs_defer_ops *dfops, | |
2848 | int *logflagsp) | |
27a3f8f2 | 2849 | { |
6d04558f CH |
2850 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
2851 | struct xfs_mount *mp = ip->i_mount; | |
2852 | struct xfs_btree_cur *cur = *curp; | |
9e5987a7 DC |
2853 | int error; /* error return value */ |
2854 | int i; /* temp state */ | |
9e5987a7 DC |
2855 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2856 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2857 | int rval=0; /* return value (logging flags) */ | |
2858 | int state; /* state bits, accessed thru macros */ | |
27a3f8f2 | 2859 | |
6d04558f CH |
2860 | ASSERT(*idx >= 0); |
2861 | ASSERT(*idx <= xfs_iext_count(ifp)); | |
9e5987a7 | 2862 | ASSERT(!isnullstartblock(new->br_startblock)); |
6d04558f | 2863 | ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); |
27a3f8f2 | 2864 | |
ff6d6af2 | 2865 | XFS_STATS_INC(mp, xs_add_exlist); |
27a3f8f2 | 2866 | |
9e5987a7 DC |
2867 | state = 0; |
2868 | if (whichfork == XFS_ATTR_FORK) | |
2869 | state |= BMAP_ATTRFORK; | |
a14234c7 CH |
2870 | if (whichfork == XFS_COW_FORK) |
2871 | state |= BMAP_COWFORK; | |
27a3f8f2 | 2872 | |
9e5987a7 DC |
2873 | /* |
2874 | * Check and set flags if this segment has a left neighbor. | |
2875 | */ | |
6d04558f | 2876 | if (*idx > 0) { |
9e5987a7 | 2877 | state |= BMAP_LEFT_VALID; |
6d04558f | 2878 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left); |
9e5987a7 DC |
2879 | if (isnullstartblock(left.br_startblock)) |
2880 | state |= BMAP_LEFT_DELAY; | |
2881 | } | |
27a3f8f2 CH |
2882 | |
2883 | /* | |
9e5987a7 DC |
2884 | * Check and set flags if this segment has a current value. |
2885 | * Not true if we're inserting into the "hole" at eof. | |
27a3f8f2 | 2886 | */ |
6d04558f | 2887 | if (*idx < xfs_iext_count(ifp)) { |
9e5987a7 | 2888 | state |= BMAP_RIGHT_VALID; |
6d04558f | 2889 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right); |
9e5987a7 DC |
2890 | if (isnullstartblock(right.br_startblock)) |
2891 | state |= BMAP_RIGHT_DELAY; | |
2892 | } | |
27a3f8f2 | 2893 | |
9e5987a7 DC |
2894 | /* |
2895 | * We're inserting a real allocation between "left" and "right". | |
2896 | * Set the contiguity flags. Don't let extents get too large. | |
2897 | */ | |
2898 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2899 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2900 | left.br_startblock + left.br_blockcount == new->br_startblock && | |
2901 | left.br_state == new->br_state && | |
2902 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2903 | state |= BMAP_LEFT_CONTIG; | |
27a3f8f2 | 2904 | |
9e5987a7 DC |
2905 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2906 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2907 | new->br_startblock + new->br_blockcount == right.br_startblock && | |
2908 | new->br_state == right.br_state && | |
2909 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2910 | (!(state & BMAP_LEFT_CONTIG) || | |
2911 | left.br_blockcount + new->br_blockcount + | |
2912 | right.br_blockcount <= MAXEXTLEN)) | |
2913 | state |= BMAP_RIGHT_CONTIG; | |
27a3f8f2 | 2914 | |
9e5987a7 DC |
2915 | error = 0; |
2916 | /* | |
2917 | * Select which case we're in here, and implement it. | |
2918 | */ | |
2919 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2920 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2921 | /* | |
2922 | * New allocation is contiguous with real allocations on the | |
2923 | * left and on the right. | |
2924 | * Merge all three into a single extent record. | |
2925 | */ | |
6d04558f CH |
2926 | --*idx; |
2927 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2928 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), | |
9e5987a7 DC |
2929 | left.br_blockcount + new->br_blockcount + |
2930 | right.br_blockcount); | |
6d04558f | 2931 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); |
27a3f8f2 | 2932 | |
6d04558f | 2933 | xfs_iext_remove(ip, *idx + 1, 1, state); |
27a3f8f2 | 2934 | |
6d04558f CH |
2935 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2936 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
2937 | if (cur == NULL) { | |
9e5987a7 DC |
2938 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2939 | } else { | |
2940 | rval = XFS_ILOG_CORE; | |
6d04558f | 2941 | error = xfs_bmbt_lookup_eq(cur, right.br_startoff, |
9e5987a7 DC |
2942 | right.br_startblock, right.br_blockcount, |
2943 | &i); | |
2944 | if (error) | |
2945 | goto done; | |
c29aad41 | 2946 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
6d04558f | 2947 | error = xfs_btree_delete(cur, &i); |
9e5987a7 DC |
2948 | if (error) |
2949 | goto done; | |
c29aad41 | 2950 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
6d04558f | 2951 | error = xfs_btree_decrement(cur, 0, &i); |
9e5987a7 DC |
2952 | if (error) |
2953 | goto done; | |
c29aad41 | 2954 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
6d04558f | 2955 | error = xfs_bmbt_update(cur, left.br_startoff, |
9e5987a7 DC |
2956 | left.br_startblock, |
2957 | left.br_blockcount + | |
2958 | new->br_blockcount + | |
2959 | right.br_blockcount, | |
2960 | left.br_state); | |
2961 | if (error) | |
2962 | goto done; | |
2963 | } | |
2964 | break; | |
27a3f8f2 | 2965 | |
9e5987a7 DC |
2966 | case BMAP_LEFT_CONTIG: |
2967 | /* | |
2968 | * New allocation is contiguous with a real allocation | |
2969 | * on the left. | |
2970 | * Merge the new allocation with the left neighbor. | |
2971 | */ | |
6d04558f CH |
2972 | --*idx; |
2973 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2974 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), | |
9e5987a7 | 2975 | left.br_blockcount + new->br_blockcount); |
6d04558f | 2976 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); |
1da177e4 | 2977 | |
6d04558f | 2978 | if (cur == NULL) { |
9e5987a7 DC |
2979 | rval = xfs_ilog_fext(whichfork); |
2980 | } else { | |
2981 | rval = 0; | |
6d04558f | 2982 | error = xfs_bmbt_lookup_eq(cur, left.br_startoff, |
9e5987a7 DC |
2983 | left.br_startblock, left.br_blockcount, |
2984 | &i); | |
2985 | if (error) | |
2986 | goto done; | |
c29aad41 | 2987 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
6d04558f | 2988 | error = xfs_bmbt_update(cur, left.br_startoff, |
9e5987a7 DC |
2989 | left.br_startblock, |
2990 | left.br_blockcount + | |
2991 | new->br_blockcount, | |
2992 | left.br_state); | |
2993 | if (error) | |
2994 | goto done; | |
2995 | } | |
2996 | break; | |
27a3f8f2 | 2997 | |
9e5987a7 DC |
2998 | case BMAP_RIGHT_CONTIG: |
2999 | /* | |
3000 | * New allocation is contiguous with a real allocation | |
3001 | * on the right. | |
3002 | * Merge the new allocation with the right neighbor. | |
3003 | */ | |
6d04558f CH |
3004 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
3005 | xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx), | |
9e5987a7 DC |
3006 | new->br_startoff, new->br_startblock, |
3007 | new->br_blockcount + right.br_blockcount, | |
3008 | right.br_state); | |
6d04558f | 3009 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); |
27a3f8f2 | 3010 | |
6d04558f | 3011 | if (cur == NULL) { |
9e5987a7 DC |
3012 | rval = xfs_ilog_fext(whichfork); |
3013 | } else { | |
3014 | rval = 0; | |
6d04558f | 3015 | error = xfs_bmbt_lookup_eq(cur, |
9e5987a7 DC |
3016 | right.br_startoff, |
3017 | right.br_startblock, | |
3018 | right.br_blockcount, &i); | |
3019 | if (error) | |
3020 | goto done; | |
c29aad41 | 3021 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
6d04558f | 3022 | error = xfs_bmbt_update(cur, new->br_startoff, |
9e5987a7 DC |
3023 | new->br_startblock, |
3024 | new->br_blockcount + | |
3025 | right.br_blockcount, | |
3026 | right.br_state); | |
3027 | if (error) | |
3028 | goto done; | |
3029 | } | |
3030 | break; | |
3031 | ||
3032 | case 0: | |
3033 | /* | |
3034 | * New allocation is not contiguous with another | |
3035 | * real allocation. | |
3036 | * Insert a new entry. | |
3037 | */ | |
6d04558f CH |
3038 | xfs_iext_insert(ip, *idx, 1, new, state); |
3039 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
3040 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
3041 | if (cur == NULL) { | |
9e5987a7 DC |
3042 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
3043 | } else { | |
3044 | rval = XFS_ILOG_CORE; | |
6d04558f | 3045 | error = xfs_bmbt_lookup_eq(cur, |
9e5987a7 DC |
3046 | new->br_startoff, |
3047 | new->br_startblock, | |
3048 | new->br_blockcount, &i); | |
3049 | if (error) | |
3050 | goto done; | |
c29aad41 | 3051 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
6d04558f CH |
3052 | cur->bc_rec.b.br_state = new->br_state; |
3053 | error = xfs_btree_insert(cur, &i); | |
9e5987a7 DC |
3054 | if (error) |
3055 | goto done; | |
c29aad41 | 3056 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
3057 | } |
3058 | break; | |
3059 | } | |
3060 | ||
9c194644 | 3061 | /* add reverse mapping */ |
6d04558f | 3062 | error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new); |
9c194644 DW |
3063 | if (error) |
3064 | goto done; | |
3065 | ||
9e5987a7 | 3066 | /* convert to a btree if necessary */ |
6d04558f | 3067 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
9e5987a7 DC |
3068 | int tmp_logflags; /* partial log flag return val */ |
3069 | ||
6d04558f CH |
3070 | ASSERT(cur == NULL); |
3071 | error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp, | |
9e5987a7 | 3072 | 0, &tmp_logflags, whichfork); |
6d04558f CH |
3073 | *logflagsp |= tmp_logflags; |
3074 | cur = *curp; | |
9e5987a7 DC |
3075 | if (error) |
3076 | goto done; | |
3077 | } | |
3078 | ||
3079 | /* clear out the allocated field, done with it now in any case. */ | |
6d04558f CH |
3080 | if (cur) |
3081 | cur->bc_private.b.allocated = 0; | |
9e5987a7 | 3082 | |
6d04558f | 3083 | xfs_bmap_check_leaf_extents(cur, ip, whichfork); |
9e5987a7 | 3084 | done: |
6d04558f | 3085 | *logflagsp |= rval; |
9e5987a7 | 3086 | return error; |
1da177e4 LT |
3087 | } |
3088 | ||
3089 | /* | |
9e5987a7 | 3090 | * Functions used in the extent read, allocate and remove paths |
1da177e4 | 3091 | */ |
1da177e4 | 3092 | |
9e5987a7 DC |
3093 | /* |
3094 | * Adjust the size of the new extent based on di_extsize and rt extsize. | |
3095 | */ | |
68988114 | 3096 | int |
9e5987a7 DC |
3097 | xfs_bmap_extsize_align( |
3098 | xfs_mount_t *mp, | |
3099 | xfs_bmbt_irec_t *gotp, /* next extent pointer */ | |
3100 | xfs_bmbt_irec_t *prevp, /* previous extent pointer */ | |
3101 | xfs_extlen_t extsz, /* align to this extent size */ | |
3102 | int rt, /* is this a realtime inode? */ | |
3103 | int eof, /* is extent at end-of-file? */ | |
3104 | int delay, /* creating delalloc extent? */ | |
3105 | int convert, /* overwriting unwritten extent? */ | |
3106 | xfs_fileoff_t *offp, /* in/out: aligned offset */ | |
3107 | xfs_extlen_t *lenp) /* in/out: aligned length */ | |
4e8938fe | 3108 | { |
9e5987a7 DC |
3109 | xfs_fileoff_t orig_off; /* original offset */ |
3110 | xfs_extlen_t orig_alen; /* original length */ | |
3111 | xfs_fileoff_t orig_end; /* original off+len */ | |
3112 | xfs_fileoff_t nexto; /* next file offset */ | |
3113 | xfs_fileoff_t prevo; /* previous file offset */ | |
3114 | xfs_fileoff_t align_off; /* temp for offset */ | |
3115 | xfs_extlen_t align_alen; /* temp for length */ | |
3116 | xfs_extlen_t temp; /* temp for calculations */ | |
4e8938fe | 3117 | |
9e5987a7 | 3118 | if (convert) |
4e8938fe | 3119 | return 0; |
4e8938fe | 3120 | |
9e5987a7 DC |
3121 | orig_off = align_off = *offp; |
3122 | orig_alen = align_alen = *lenp; | |
3123 | orig_end = orig_off + orig_alen; | |
1da177e4 | 3124 | |
1da177e4 | 3125 | /* |
9e5987a7 DC |
3126 | * If this request overlaps an existing extent, then don't |
3127 | * attempt to perform any additional alignment. | |
1da177e4 | 3128 | */ |
9e5987a7 DC |
3129 | if (!delay && !eof && |
3130 | (orig_off >= gotp->br_startoff) && | |
3131 | (orig_end <= gotp->br_startoff + gotp->br_blockcount)) { | |
3132 | return 0; | |
3133 | } | |
3134 | ||
1da177e4 | 3135 | /* |
9e5987a7 DC |
3136 | * If the file offset is unaligned vs. the extent size |
3137 | * we need to align it. This will be possible unless | |
3138 | * the file was previously written with a kernel that didn't | |
3139 | * perform this alignment, or if a truncate shot us in the | |
3140 | * foot. | |
1da177e4 | 3141 | */ |
9e5987a7 DC |
3142 | temp = do_mod(orig_off, extsz); |
3143 | if (temp) { | |
3144 | align_alen += temp; | |
3145 | align_off -= temp; | |
1da177e4 | 3146 | } |
6dea405e DC |
3147 | |
3148 | /* Same adjustment for the end of the requested area. */ | |
3149 | temp = (align_alen % extsz); | |
3150 | if (temp) | |
3151 | align_alen += extsz - temp; | |
3152 | ||
1da177e4 | 3153 | /* |
6dea405e DC |
3154 | * For large extent hint sizes, the aligned extent might be larger than |
3155 | * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls | |
3156 | * the length back under MAXEXTLEN. The outer allocation loops handle | |
3157 | * short allocation just fine, so it is safe to do this. We only want to | |
3158 | * do it when we are forced to, though, because it means more allocation | |
3159 | * operations are required. | |
1da177e4 | 3160 | */ |
6dea405e DC |
3161 | while (align_alen > MAXEXTLEN) |
3162 | align_alen -= extsz; | |
3163 | ASSERT(align_alen <= MAXEXTLEN); | |
3164 | ||
1da177e4 | 3165 | /* |
9e5987a7 DC |
3166 | * If the previous block overlaps with this proposed allocation |
3167 | * then move the start forward without adjusting the length. | |
1da177e4 | 3168 | */ |
9e5987a7 DC |
3169 | if (prevp->br_startoff != NULLFILEOFF) { |
3170 | if (prevp->br_startblock == HOLESTARTBLOCK) | |
3171 | prevo = prevp->br_startoff; | |
3172 | else | |
3173 | prevo = prevp->br_startoff + prevp->br_blockcount; | |
3174 | } else | |
3175 | prevo = 0; | |
3176 | if (align_off != orig_off && align_off < prevo) | |
3177 | align_off = prevo; | |
3178 | /* | |
3179 | * If the next block overlaps with this proposed allocation | |
3180 | * then move the start back without adjusting the length, | |
3181 | * but not before offset 0. | |
3182 | * This may of course make the start overlap previous block, | |
3183 | * and if we hit the offset 0 limit then the next block | |
3184 | * can still overlap too. | |
3185 | */ | |
3186 | if (!eof && gotp->br_startoff != NULLFILEOFF) { | |
3187 | if ((delay && gotp->br_startblock == HOLESTARTBLOCK) || | |
3188 | (!delay && gotp->br_startblock == DELAYSTARTBLOCK)) | |
3189 | nexto = gotp->br_startoff + gotp->br_blockcount; | |
3190 | else | |
3191 | nexto = gotp->br_startoff; | |
3192 | } else | |
3193 | nexto = NULLFILEOFF; | |
3194 | if (!eof && | |
3195 | align_off + align_alen != orig_end && | |
3196 | align_off + align_alen > nexto) | |
3197 | align_off = nexto > align_alen ? nexto - align_alen : 0; | |
3198 | /* | |
3199 | * If we're now overlapping the next or previous extent that | |
3200 | * means we can't fit an extsz piece in this hole. Just move | |
3201 | * the start forward to the first valid spot and set | |
3202 | * the length so we hit the end. | |
3203 | */ | |
3204 | if (align_off != orig_off && align_off < prevo) | |
3205 | align_off = prevo; | |
3206 | if (align_off + align_alen != orig_end && | |
3207 | align_off + align_alen > nexto && | |
3208 | nexto != NULLFILEOFF) { | |
3209 | ASSERT(nexto > prevo); | |
3210 | align_alen = nexto - align_off; | |
3211 | } | |
1da177e4 | 3212 | |
9e5987a7 DC |
3213 | /* |
3214 | * If realtime, and the result isn't a multiple of the realtime | |
3215 | * extent size we need to remove blocks until it is. | |
3216 | */ | |
3217 | if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) { | |
1da177e4 | 3218 | /* |
9e5987a7 DC |
3219 | * We're not covering the original request, or |
3220 | * we won't be able to once we fix the length. | |
1da177e4 | 3221 | */ |
9e5987a7 DC |
3222 | if (orig_off < align_off || |
3223 | orig_end > align_off + align_alen || | |
3224 | align_alen - temp < orig_alen) | |
2451337d | 3225 | return -EINVAL; |
1da177e4 | 3226 | /* |
9e5987a7 | 3227 | * Try to fix it by moving the start up. |
1da177e4 | 3228 | */ |
9e5987a7 DC |
3229 | if (align_off + temp <= orig_off) { |
3230 | align_alen -= temp; | |
3231 | align_off += temp; | |
1da177e4 | 3232 | } |
9e5987a7 DC |
3233 | /* |
3234 | * Try to fix it by moving the end in. | |
3235 | */ | |
3236 | else if (align_off + align_alen - temp >= orig_end) | |
3237 | align_alen -= temp; | |
3238 | /* | |
3239 | * Set the start to the minimum then trim the length. | |
3240 | */ | |
3241 | else { | |
3242 | align_alen -= orig_off - align_off; | |
3243 | align_off = orig_off; | |
3244 | align_alen -= align_alen % mp->m_sb.sb_rextsize; | |
1da177e4 | 3245 | } |
1da177e4 | 3246 | /* |
9e5987a7 | 3247 | * Result doesn't cover the request, fail it. |
1da177e4 | 3248 | */ |
9e5987a7 | 3249 | if (orig_off < align_off || orig_end > align_off + align_alen) |
2451337d | 3250 | return -EINVAL; |
9e5987a7 DC |
3251 | } else { |
3252 | ASSERT(orig_off >= align_off); | |
6dea405e DC |
3253 | /* see MAXEXTLEN handling above */ |
3254 | ASSERT(orig_end <= align_off + align_alen || | |
3255 | align_alen + extsz > MAXEXTLEN); | |
1da177e4 | 3256 | } |
1da177e4 | 3257 | |
0b1b213f | 3258 | #ifdef DEBUG |
9e5987a7 DC |
3259 | if (!eof && gotp->br_startoff != NULLFILEOFF) |
3260 | ASSERT(align_off + align_alen <= gotp->br_startoff); | |
3261 | if (prevp->br_startoff != NULLFILEOFF) | |
3262 | ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount); | |
3263 | #endif | |
1da177e4 | 3264 | |
9e5987a7 DC |
3265 | *lenp = align_alen; |
3266 | *offp = align_off; | |
3267 | return 0; | |
1da177e4 | 3268 | } |
1da177e4 | 3269 | |
9e5987a7 DC |
3270 | #define XFS_ALLOC_GAP_UNITS 4 |
3271 | ||
68988114 | 3272 | void |
9e5987a7 | 3273 | xfs_bmap_adjacent( |
68988114 | 3274 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
1da177e4 | 3275 | { |
9e5987a7 DC |
3276 | xfs_fsblock_t adjust; /* adjustment to block numbers */ |
3277 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ | |
3278 | xfs_mount_t *mp; /* mount point structure */ | |
3279 | int nullfb; /* true if ap->firstblock isn't set */ | |
3280 | int rt; /* true if inode is realtime */ | |
1da177e4 | 3281 | |
9e5987a7 DC |
3282 | #define ISVALID(x,y) \ |
3283 | (rt ? \ | |
3284 | (x) < mp->m_sb.sb_rblocks : \ | |
3285 | XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \ | |
3286 | XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \ | |
3287 | XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks) | |
1da177e4 | 3288 | |
9e5987a7 DC |
3289 | mp = ap->ip->i_mount; |
3290 | nullfb = *ap->firstblock == NULLFSBLOCK; | |
292378ed DC |
3291 | rt = XFS_IS_REALTIME_INODE(ap->ip) && |
3292 | xfs_alloc_is_userdata(ap->datatype); | |
9e5987a7 DC |
3293 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock); |
3294 | /* | |
3295 | * If allocating at eof, and there's a previous real block, | |
3296 | * try to use its last block as our starting point. | |
3297 | */ | |
3298 | if (ap->eof && ap->prev.br_startoff != NULLFILEOFF && | |
3299 | !isnullstartblock(ap->prev.br_startblock) && | |
3300 | ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount, | |
3301 | ap->prev.br_startblock)) { | |
3302 | ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount; | |
3303 | /* | |
3304 | * Adjust for the gap between prevp and us. | |
3305 | */ | |
3306 | adjust = ap->offset - | |
3307 | (ap->prev.br_startoff + ap->prev.br_blockcount); | |
3308 | if (adjust && | |
3309 | ISVALID(ap->blkno + adjust, ap->prev.br_startblock)) | |
3310 | ap->blkno += adjust; | |
3311 | } | |
3312 | /* | |
3313 | * If not at eof, then compare the two neighbor blocks. | |
3314 | * Figure out whether either one gives us a good starting point, | |
3315 | * and pick the better one. | |
3316 | */ | |
3317 | else if (!ap->eof) { | |
3318 | xfs_fsblock_t gotbno; /* right side block number */ | |
3319 | xfs_fsblock_t gotdiff=0; /* right side difference */ | |
3320 | xfs_fsblock_t prevbno; /* left side block number */ | |
3321 | xfs_fsblock_t prevdiff=0; /* left side difference */ | |
3322 | ||
3323 | /* | |
3324 | * If there's a previous (left) block, select a requested | |
3325 | * start block based on it. | |
3326 | */ | |
3327 | if (ap->prev.br_startoff != NULLFILEOFF && | |
3328 | !isnullstartblock(ap->prev.br_startblock) && | |
3329 | (prevbno = ap->prev.br_startblock + | |
3330 | ap->prev.br_blockcount) && | |
3331 | ISVALID(prevbno, ap->prev.br_startblock)) { | |
3332 | /* | |
3333 | * Calculate gap to end of previous block. | |
3334 | */ | |
3335 | adjust = prevdiff = ap->offset - | |
3336 | (ap->prev.br_startoff + | |
3337 | ap->prev.br_blockcount); | |
3338 | /* | |
3339 | * Figure the startblock based on the previous block's | |
3340 | * end and the gap size. | |
3341 | * Heuristic! | |
3342 | * If the gap is large relative to the piece we're | |
3343 | * allocating, or using it gives us an invalid block | |
3344 | * number, then just use the end of the previous block. | |
3345 | */ | |
3346 | if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3347 | ISVALID(prevbno + prevdiff, | |
3348 | ap->prev.br_startblock)) | |
3349 | prevbno += adjust; | |
3350 | else | |
3351 | prevdiff += adjust; | |
3352 | /* | |
3353 | * If the firstblock forbids it, can't use it, | |
3354 | * must use default. | |
3355 | */ | |
3356 | if (!rt && !nullfb && | |
3357 | XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno) | |
3358 | prevbno = NULLFSBLOCK; | |
1da177e4 | 3359 | } |
9e5987a7 DC |
3360 | /* |
3361 | * No previous block or can't follow it, just default. | |
3362 | */ | |
3363 | else | |
3364 | prevbno = NULLFSBLOCK; | |
3365 | /* | |
3366 | * If there's a following (right) block, select a requested | |
3367 | * start block based on it. | |
3368 | */ | |
3369 | if (!isnullstartblock(ap->got.br_startblock)) { | |
3370 | /* | |
3371 | * Calculate gap to start of next block. | |
3372 | */ | |
3373 | adjust = gotdiff = ap->got.br_startoff - ap->offset; | |
3374 | /* | |
3375 | * Figure the startblock based on the next block's | |
3376 | * start and the gap size. | |
3377 | */ | |
3378 | gotbno = ap->got.br_startblock; | |
3379 | /* | |
3380 | * Heuristic! | |
3381 | * If the gap is large relative to the piece we're | |
3382 | * allocating, or using it gives us an invalid block | |
3383 | * number, then just use the start of the next block | |
3384 | * offset by our length. | |
3385 | */ | |
3386 | if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3387 | ISVALID(gotbno - gotdiff, gotbno)) | |
3388 | gotbno -= adjust; | |
3389 | else if (ISVALID(gotbno - ap->length, gotbno)) { | |
3390 | gotbno -= ap->length; | |
3391 | gotdiff += adjust - ap->length; | |
3392 | } else | |
3393 | gotdiff += adjust; | |
3394 | /* | |
3395 | * If the firstblock forbids it, can't use it, | |
3396 | * must use default. | |
3397 | */ | |
3398 | if (!rt && !nullfb && | |
3399 | XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno) | |
3400 | gotbno = NULLFSBLOCK; | |
3401 | } | |
3402 | /* | |
3403 | * No next block, just default. | |
3404 | */ | |
3405 | else | |
3406 | gotbno = NULLFSBLOCK; | |
3407 | /* | |
3408 | * If both valid, pick the better one, else the only good | |
3409 | * one, else ap->blkno is already set (to 0 or the inode block). | |
3410 | */ | |
3411 | if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) | |
3412 | ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno; | |
3413 | else if (prevbno != NULLFSBLOCK) | |
3414 | ap->blkno = prevbno; | |
3415 | else if (gotbno != NULLFSBLOCK) | |
3416 | ap->blkno = gotbno; | |
1da177e4 | 3417 | } |
9e5987a7 | 3418 | #undef ISVALID |
1da177e4 | 3419 | } |
1da177e4 | 3420 | |
c977eb10 CH |
3421 | static int |
3422 | xfs_bmap_longest_free_extent( | |
3423 | struct xfs_trans *tp, | |
3424 | xfs_agnumber_t ag, | |
3425 | xfs_extlen_t *blen, | |
3426 | int *notinit) | |
3427 | { | |
3428 | struct xfs_mount *mp = tp->t_mountp; | |
3429 | struct xfs_perag *pag; | |
3430 | xfs_extlen_t longest; | |
3431 | int error = 0; | |
3432 | ||
3433 | pag = xfs_perag_get(mp, ag); | |
3434 | if (!pag->pagf_init) { | |
3435 | error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK); | |
3436 | if (error) | |
3437 | goto out; | |
3438 | ||
3439 | if (!pag->pagf_init) { | |
3440 | *notinit = 1; | |
3441 | goto out; | |
3442 | } | |
3443 | } | |
3444 | ||
50adbcb4 | 3445 | longest = xfs_alloc_longest_free_extent(mp, pag, |
3fd129b6 DW |
3446 | xfs_alloc_min_freelist(mp, pag), |
3447 | xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE)); | |
c977eb10 CH |
3448 | if (*blen < longest) |
3449 | *blen = longest; | |
3450 | ||
3451 | out: | |
3452 | xfs_perag_put(pag); | |
3453 | return error; | |
3454 | } | |
3455 | ||
3456 | static void | |
3457 | xfs_bmap_select_minlen( | |
3458 | struct xfs_bmalloca *ap, | |
3459 | struct xfs_alloc_arg *args, | |
3460 | xfs_extlen_t *blen, | |
3461 | int notinit) | |
3462 | { | |
3463 | if (notinit || *blen < ap->minlen) { | |
3464 | /* | |
3465 | * Since we did a BUF_TRYLOCK above, it is possible that | |
3466 | * there is space for this request. | |
3467 | */ | |
3468 | args->minlen = ap->minlen; | |
3469 | } else if (*blen < args->maxlen) { | |
3470 | /* | |
3471 | * If the best seen length is less than the request length, | |
3472 | * use the best as the minimum. | |
3473 | */ | |
3474 | args->minlen = *blen; | |
3475 | } else { | |
3476 | /* | |
3477 | * Otherwise we've seen an extent as big as maxlen, use that | |
3478 | * as the minimum. | |
3479 | */ | |
3480 | args->minlen = args->maxlen; | |
3481 | } | |
3482 | } | |
3483 | ||
b64dfe4e | 3484 | STATIC int |
9e5987a7 DC |
3485 | xfs_bmap_btalloc_nullfb( |
3486 | struct xfs_bmalloca *ap, | |
3487 | struct xfs_alloc_arg *args, | |
3488 | xfs_extlen_t *blen) | |
b64dfe4e | 3489 | { |
9e5987a7 | 3490 | struct xfs_mount *mp = ap->ip->i_mount; |
9e5987a7 DC |
3491 | xfs_agnumber_t ag, startag; |
3492 | int notinit = 0; | |
b64dfe4e CH |
3493 | int error; |
3494 | ||
c977eb10 | 3495 | args->type = XFS_ALLOCTYPE_START_BNO; |
9e5987a7 | 3496 | args->total = ap->total; |
b64dfe4e | 3497 | |
9e5987a7 DC |
3498 | startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno); |
3499 | if (startag == NULLAGNUMBER) | |
3500 | startag = ag = 0; | |
b64dfe4e | 3501 | |
9e5987a7 | 3502 | while (*blen < args->maxlen) { |
c977eb10 CH |
3503 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, |
3504 | ¬init); | |
3505 | if (error) | |
3506 | return error; | |
b64dfe4e | 3507 | |
9e5987a7 DC |
3508 | if (++ag == mp->m_sb.sb_agcount) |
3509 | ag = 0; | |
3510 | if (ag == startag) | |
3511 | break; | |
9e5987a7 | 3512 | } |
b64dfe4e | 3513 | |
c977eb10 CH |
3514 | xfs_bmap_select_minlen(ap, args, blen, notinit); |
3515 | return 0; | |
3516 | } | |
3517 | ||
3518 | STATIC int | |
3519 | xfs_bmap_btalloc_filestreams( | |
3520 | struct xfs_bmalloca *ap, | |
3521 | struct xfs_alloc_arg *args, | |
3522 | xfs_extlen_t *blen) | |
3523 | { | |
3524 | struct xfs_mount *mp = ap->ip->i_mount; | |
3525 | xfs_agnumber_t ag; | |
3526 | int notinit = 0; | |
3527 | int error; | |
3528 | ||
3529 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
3530 | args->total = ap->total; | |
3531 | ||
3532 | ag = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3533 | if (ag == NULLAGNUMBER) | |
3534 | ag = 0; | |
3535 | ||
3536 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init); | |
3537 | if (error) | |
3538 | return error; | |
3539 | ||
3540 | if (*blen < args->maxlen) { | |
3541 | error = xfs_filestream_new_ag(ap, &ag); | |
3542 | if (error) | |
3543 | return error; | |
3544 | ||
3545 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, | |
3546 | ¬init); | |
3547 | if (error) | |
3548 | return error; | |
3549 | ||
3550 | } | |
3551 | ||
3552 | xfs_bmap_select_minlen(ap, args, blen, notinit); | |
b64dfe4e CH |
3553 | |
3554 | /* | |
c977eb10 CH |
3555 | * Set the failure fallback case to look in the selected AG as stream |
3556 | * may have moved. | |
b64dfe4e | 3557 | */ |
c977eb10 | 3558 | ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0); |
b64dfe4e | 3559 | return 0; |
b64dfe4e CH |
3560 | } |
3561 | ||
9e5987a7 DC |
3562 | STATIC int |
3563 | xfs_bmap_btalloc( | |
68988114 | 3564 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
4403280a | 3565 | { |
9e5987a7 DC |
3566 | xfs_mount_t *mp; /* mount point structure */ |
3567 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
292378ed | 3568 | xfs_extlen_t align = 0; /* minimum allocation alignment */ |
9e5987a7 DC |
3569 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ |
3570 | xfs_agnumber_t ag; | |
3571 | xfs_alloc_arg_t args; | |
3572 | xfs_extlen_t blen; | |
3573 | xfs_extlen_t nextminlen = 0; | |
3574 | int nullfb; /* true if ap->firstblock isn't set */ | |
3575 | int isaligned; | |
3576 | int tryagain; | |
3577 | int error; | |
33177f05 | 3578 | int stripe_align; |
4403280a | 3579 | |
9e5987a7 | 3580 | ASSERT(ap->length); |
4403280a | 3581 | |
9e5987a7 | 3582 | mp = ap->ip->i_mount; |
33177f05 DC |
3583 | |
3584 | /* stripe alignment for allocation is determined by mount parameters */ | |
3585 | stripe_align = 0; | |
3586 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
3587 | stripe_align = mp->m_swidth; | |
3588 | else if (mp->m_dalign) | |
3589 | stripe_align = mp->m_dalign; | |
3590 | ||
f7ca3522 DW |
3591 | if (ap->flags & XFS_BMAPI_COWFORK) |
3592 | align = xfs_get_cowextsz_hint(ap->ip); | |
3593 | else if (xfs_alloc_is_userdata(ap->datatype)) | |
292378ed | 3594 | align = xfs_get_extsz_hint(ap->ip); |
493611eb | 3595 | if (align) { |
9e5987a7 DC |
3596 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, |
3597 | align, 0, ap->eof, 0, ap->conv, | |
3598 | &ap->offset, &ap->length); | |
3599 | ASSERT(!error); | |
3600 | ASSERT(ap->length); | |
4403280a | 3601 | } |
33177f05 DC |
3602 | |
3603 | ||
9e5987a7 DC |
3604 | nullfb = *ap->firstblock == NULLFSBLOCK; |
3605 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock); | |
3606 | if (nullfb) { | |
292378ed DC |
3607 | if (xfs_alloc_is_userdata(ap->datatype) && |
3608 | xfs_inode_is_filestream(ap->ip)) { | |
9e5987a7 DC |
3609 | ag = xfs_filestream_lookup_ag(ap->ip); |
3610 | ag = (ag != NULLAGNUMBER) ? ag : 0; | |
3611 | ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0); | |
3612 | } else { | |
3613 | ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino); | |
3614 | } | |
3615 | } else | |
3616 | ap->blkno = *ap->firstblock; | |
4403280a | 3617 | |
9e5987a7 | 3618 | xfs_bmap_adjacent(ap); |
4403280a | 3619 | |
9e5987a7 DC |
3620 | /* |
3621 | * If allowed, use ap->blkno; otherwise must use firstblock since | |
3622 | * it's in the right allocation group. | |
3623 | */ | |
3624 | if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno) | |
3625 | ; | |
3626 | else | |
3627 | ap->blkno = *ap->firstblock; | |
3628 | /* | |
3629 | * Normal allocation, done through xfs_alloc_vextent. | |
3630 | */ | |
3631 | tryagain = isaligned = 0; | |
3632 | memset(&args, 0, sizeof(args)); | |
3633 | args.tp = ap->tp; | |
3634 | args.mp = mp; | |
3635 | args.fsbno = ap->blkno; | |
340785cc | 3636 | xfs_rmap_skip_owner_update(&args.oinfo); |
4403280a | 3637 | |
9e5987a7 | 3638 | /* Trim the allocation back to the maximum an AG can fit. */ |
52548852 | 3639 | args.maxlen = MIN(ap->length, mp->m_ag_max_usable); |
9e5987a7 DC |
3640 | args.firstblock = *ap->firstblock; |
3641 | blen = 0; | |
3642 | if (nullfb) { | |
c977eb10 CH |
3643 | /* |
3644 | * Search for an allocation group with a single extent large | |
3645 | * enough for the request. If one isn't found, then adjust | |
3646 | * the minimum allocation size to the largest space found. | |
3647 | */ | |
292378ed DC |
3648 | if (xfs_alloc_is_userdata(ap->datatype) && |
3649 | xfs_inode_is_filestream(ap->ip)) | |
c977eb10 CH |
3650 | error = xfs_bmap_btalloc_filestreams(ap, &args, &blen); |
3651 | else | |
3652 | error = xfs_bmap_btalloc_nullfb(ap, &args, &blen); | |
4403280a CH |
3653 | if (error) |
3654 | return error; | |
2c3234d1 | 3655 | } else if (ap->dfops->dop_low) { |
9e5987a7 DC |
3656 | if (xfs_inode_is_filestream(ap->ip)) |
3657 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3658 | else | |
3659 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3660 | args.total = args.minlen = ap->minlen; | |
3661 | } else { | |
3662 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3663 | args.total = ap->total; | |
3664 | args.minlen = ap->minlen; | |
4403280a | 3665 | } |
9e5987a7 | 3666 | /* apply extent size hints if obtained earlier */ |
493611eb | 3667 | if (align) { |
9e5987a7 DC |
3668 | args.prod = align; |
3669 | if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod))) | |
3670 | args.mod = (xfs_extlen_t)(args.prod - args.mod); | |
09cbfeaf | 3671 | } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) { |
9e5987a7 DC |
3672 | args.prod = 1; |
3673 | args.mod = 0; | |
3674 | } else { | |
09cbfeaf | 3675 | args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog; |
9e5987a7 DC |
3676 | if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod)))) |
3677 | args.mod = (xfs_extlen_t)(args.prod - args.mod); | |
4403280a | 3678 | } |
7e47a4ef | 3679 | /* |
9e5987a7 DC |
3680 | * If we are not low on available data blocks, and the |
3681 | * underlying logical volume manager is a stripe, and | |
3682 | * the file offset is zero then try to allocate data | |
3683 | * blocks on stripe unit boundary. | |
3684 | * NOTE: ap->aeof is only set if the allocation length | |
3685 | * is >= the stripe unit and the allocation offset is | |
3686 | * at the end of file. | |
7e47a4ef | 3687 | */ |
2c3234d1 | 3688 | if (!ap->dfops->dop_low && ap->aeof) { |
9e5987a7 | 3689 | if (!ap->offset) { |
33177f05 | 3690 | args.alignment = stripe_align; |
9e5987a7 DC |
3691 | atype = args.type; |
3692 | isaligned = 1; | |
3693 | /* | |
3694 | * Adjust for alignment | |
3695 | */ | |
3696 | if (blen > args.alignment && blen <= args.maxlen) | |
3697 | args.minlen = blen - args.alignment; | |
3698 | args.minalignslop = 0; | |
3699 | } else { | |
3700 | /* | |
3701 | * First try an exact bno allocation. | |
3702 | * If it fails then do a near or start bno | |
3703 | * allocation with alignment turned on. | |
3704 | */ | |
3705 | atype = args.type; | |
3706 | tryagain = 1; | |
3707 | args.type = XFS_ALLOCTYPE_THIS_BNO; | |
3708 | args.alignment = 1; | |
3709 | /* | |
3710 | * Compute the minlen+alignment for the | |
3711 | * next case. Set slop so that the value | |
3712 | * of minlen+alignment+slop doesn't go up | |
3713 | * between the calls. | |
3714 | */ | |
33177f05 DC |
3715 | if (blen > stripe_align && blen <= args.maxlen) |
3716 | nextminlen = blen - stripe_align; | |
9e5987a7 DC |
3717 | else |
3718 | nextminlen = args.minlen; | |
33177f05 | 3719 | if (nextminlen + stripe_align > args.minlen + 1) |
9e5987a7 | 3720 | args.minalignslop = |
33177f05 | 3721 | nextminlen + stripe_align - |
9e5987a7 DC |
3722 | args.minlen - 1; |
3723 | else | |
3724 | args.minalignslop = 0; | |
7e47a4ef DC |
3725 | } |
3726 | } else { | |
9e5987a7 DC |
3727 | args.alignment = 1; |
3728 | args.minalignslop = 0; | |
7e47a4ef | 3729 | } |
9e5987a7 DC |
3730 | args.minleft = ap->minleft; |
3731 | args.wasdel = ap->wasdel; | |
3fd129b6 | 3732 | args.resv = XFS_AG_RESV_NONE; |
292378ed DC |
3733 | args.datatype = ap->datatype; |
3734 | if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) | |
3fbbbea3 DC |
3735 | args.ip = ap->ip; |
3736 | ||
3737 | error = xfs_alloc_vextent(&args); | |
3738 | if (error) | |
9e5987a7 | 3739 | return error; |
3fbbbea3 | 3740 | |
9e5987a7 DC |
3741 | if (tryagain && args.fsbno == NULLFSBLOCK) { |
3742 | /* | |
3743 | * Exact allocation failed. Now try with alignment | |
3744 | * turned on. | |
3745 | */ | |
3746 | args.type = atype; | |
3747 | args.fsbno = ap->blkno; | |
33177f05 | 3748 | args.alignment = stripe_align; |
9e5987a7 DC |
3749 | args.minlen = nextminlen; |
3750 | args.minalignslop = 0; | |
3751 | isaligned = 1; | |
3752 | if ((error = xfs_alloc_vextent(&args))) | |
3753 | return error; | |
7e47a4ef | 3754 | } |
9e5987a7 DC |
3755 | if (isaligned && args.fsbno == NULLFSBLOCK) { |
3756 | /* | |
3757 | * allocation failed, so turn off alignment and | |
3758 | * try again. | |
3759 | */ | |
3760 | args.type = atype; | |
3761 | args.fsbno = ap->blkno; | |
3762 | args.alignment = 0; | |
3763 | if ((error = xfs_alloc_vextent(&args))) | |
7e47a4ef DC |
3764 | return error; |
3765 | } | |
9e5987a7 DC |
3766 | if (args.fsbno == NULLFSBLOCK && nullfb && |
3767 | args.minlen > ap->minlen) { | |
3768 | args.minlen = ap->minlen; | |
3769 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3770 | args.fsbno = ap->blkno; | |
3771 | if ((error = xfs_alloc_vextent(&args))) | |
3772 | return error; | |
7e47a4ef | 3773 | } |
9e5987a7 DC |
3774 | if (args.fsbno == NULLFSBLOCK && nullfb) { |
3775 | args.fsbno = 0; | |
3776 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3777 | args.total = ap->minlen; | |
9e5987a7 DC |
3778 | if ((error = xfs_alloc_vextent(&args))) |
3779 | return error; | |
2c3234d1 | 3780 | ap->dfops->dop_low = true; |
9e5987a7 DC |
3781 | } |
3782 | if (args.fsbno != NULLFSBLOCK) { | |
3783 | /* | |
3784 | * check the allocation happened at the same or higher AG than | |
3785 | * the first block that was allocated. | |
3786 | */ | |
3787 | ASSERT(*ap->firstblock == NULLFSBLOCK || | |
410d17f6 CH |
3788 | XFS_FSB_TO_AGNO(mp, *ap->firstblock) <= |
3789 | XFS_FSB_TO_AGNO(mp, args.fsbno)); | |
7e47a4ef | 3790 | |
9e5987a7 DC |
3791 | ap->blkno = args.fsbno; |
3792 | if (*ap->firstblock == NULLFSBLOCK) | |
3793 | *ap->firstblock = args.fsbno; | |
410d17f6 | 3794 | ASSERT(nullfb || fb_agno <= args.agno); |
9e5987a7 | 3795 | ap->length = args.len; |
60b4984f DW |
3796 | if (!(ap->flags & XFS_BMAPI_COWFORK)) |
3797 | ap->ip->i_d.di_nblocks += args.len; | |
9e5987a7 DC |
3798 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); |
3799 | if (ap->wasdel) | |
3800 | ap->ip->i_delayed_blks -= args.len; | |
3801 | /* | |
3802 | * Adjust the disk quota also. This was reserved | |
3803 | * earlier. | |
3804 | */ | |
3805 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, | |
3806 | ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : | |
3807 | XFS_TRANS_DQ_BCOUNT, | |
3808 | (long) args.len); | |
3809 | } else { | |
3810 | ap->blkno = NULLFSBLOCK; | |
3811 | ap->length = 0; | |
3812 | } | |
3813 | return 0; | |
3814 | } | |
7e47a4ef | 3815 | |
9e5987a7 DC |
3816 | /* |
3817 | * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file. | |
3818 | * It figures out where to ask the underlying allocator to put the new extent. | |
3819 | */ | |
3820 | STATIC int | |
3821 | xfs_bmap_alloc( | |
68988114 | 3822 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
9e5987a7 | 3823 | { |
292378ed DC |
3824 | if (XFS_IS_REALTIME_INODE(ap->ip) && |
3825 | xfs_alloc_is_userdata(ap->datatype)) | |
9e5987a7 DC |
3826 | return xfs_bmap_rtalloc(ap); |
3827 | return xfs_bmap_btalloc(ap); | |
3828 | } | |
a5bd606b | 3829 | |
0a0af28c DW |
3830 | /* Trim extent to fit a logical block range. */ |
3831 | void | |
3832 | xfs_trim_extent( | |
3833 | struct xfs_bmbt_irec *irec, | |
3834 | xfs_fileoff_t bno, | |
3835 | xfs_filblks_t len) | |
3836 | { | |
3837 | xfs_fileoff_t distance; | |
3838 | xfs_fileoff_t end = bno + len; | |
3839 | ||
3840 | if (irec->br_startoff + irec->br_blockcount <= bno || | |
3841 | irec->br_startoff >= end) { | |
3842 | irec->br_blockcount = 0; | |
3843 | return; | |
3844 | } | |
3845 | ||
3846 | if (irec->br_startoff < bno) { | |
3847 | distance = bno - irec->br_startoff; | |
3848 | if (isnullstartblock(irec->br_startblock)) | |
3849 | irec->br_startblock = DELAYSTARTBLOCK; | |
3850 | if (irec->br_startblock != DELAYSTARTBLOCK && | |
3851 | irec->br_startblock != HOLESTARTBLOCK) | |
3852 | irec->br_startblock += distance; | |
3853 | irec->br_startoff += distance; | |
3854 | irec->br_blockcount -= distance; | |
3855 | } | |
3856 | ||
3857 | if (end < irec->br_startoff + irec->br_blockcount) { | |
3858 | distance = irec->br_startoff + irec->br_blockcount - end; | |
3859 | irec->br_blockcount -= distance; | |
3860 | } | |
3861 | } | |
3862 | ||
879ca375 BF |
3863 | /* trim extent to within eof */ |
3864 | void | |
3865 | xfs_trim_extent_eof( | |
3866 | struct xfs_bmbt_irec *irec, | |
3867 | struct xfs_inode *ip) | |
3868 | ||
3869 | { | |
3870 | xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount, | |
3871 | i_size_read(VFS_I(ip)))); | |
3872 | } | |
3873 | ||
9e5987a7 DC |
3874 | /* |
3875 | * Trim the returned map to the required bounds | |
3876 | */ | |
3877 | STATIC void | |
3878 | xfs_bmapi_trim_map( | |
3879 | struct xfs_bmbt_irec *mval, | |
3880 | struct xfs_bmbt_irec *got, | |
3881 | xfs_fileoff_t *bno, | |
3882 | xfs_filblks_t len, | |
3883 | xfs_fileoff_t obno, | |
3884 | xfs_fileoff_t end, | |
3885 | int n, | |
3886 | int flags) | |
3887 | { | |
3888 | if ((flags & XFS_BMAPI_ENTIRE) || | |
3889 | got->br_startoff + got->br_blockcount <= obno) { | |
3890 | *mval = *got; | |
3891 | if (isnullstartblock(got->br_startblock)) | |
3892 | mval->br_startblock = DELAYSTARTBLOCK; | |
3893 | return; | |
3894 | } | |
7e47a4ef | 3895 | |
9e5987a7 DC |
3896 | if (obno > *bno) |
3897 | *bno = obno; | |
3898 | ASSERT((*bno >= obno) || (n == 0)); | |
3899 | ASSERT(*bno < end); | |
3900 | mval->br_startoff = *bno; | |
3901 | if (isnullstartblock(got->br_startblock)) | |
3902 | mval->br_startblock = DELAYSTARTBLOCK; | |
3903 | else | |
3904 | mval->br_startblock = got->br_startblock + | |
3905 | (*bno - got->br_startoff); | |
7e47a4ef | 3906 | /* |
9e5987a7 DC |
3907 | * Return the minimum of what we got and what we asked for for |
3908 | * the length. We can use the len variable here because it is | |
3909 | * modified below and we could have been there before coming | |
3910 | * here if the first part of the allocation didn't overlap what | |
3911 | * was asked for. | |
7e47a4ef | 3912 | */ |
9e5987a7 DC |
3913 | mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno, |
3914 | got->br_blockcount - (*bno - got->br_startoff)); | |
3915 | mval->br_state = got->br_state; | |
3916 | ASSERT(mval->br_blockcount <= len); | |
3917 | return; | |
7e47a4ef DC |
3918 | } |
3919 | ||
9e5987a7 DC |
3920 | /* |
3921 | * Update and validate the extent map to return | |
3922 | */ | |
3923 | STATIC void | |
3924 | xfs_bmapi_update_map( | |
3925 | struct xfs_bmbt_irec **map, | |
3926 | xfs_fileoff_t *bno, | |
3927 | xfs_filblks_t *len, | |
3928 | xfs_fileoff_t obno, | |
3929 | xfs_fileoff_t end, | |
3930 | int *n, | |
3931 | int flags) | |
e04426b9 | 3932 | { |
9e5987a7 | 3933 | xfs_bmbt_irec_t *mval = *map; |
e04426b9 | 3934 | |
9e5987a7 DC |
3935 | ASSERT((flags & XFS_BMAPI_ENTIRE) || |
3936 | ((mval->br_startoff + mval->br_blockcount) <= end)); | |
3937 | ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) || | |
3938 | (mval->br_startoff < obno)); | |
e04426b9 | 3939 | |
9e5987a7 DC |
3940 | *bno = mval->br_startoff + mval->br_blockcount; |
3941 | *len = end - *bno; | |
3942 | if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) { | |
3943 | /* update previous map with new information */ | |
3944 | ASSERT(mval->br_startblock == mval[-1].br_startblock); | |
3945 | ASSERT(mval->br_blockcount > mval[-1].br_blockcount); | |
3946 | ASSERT(mval->br_state == mval[-1].br_state); | |
3947 | mval[-1].br_blockcount = mval->br_blockcount; | |
3948 | mval[-1].br_state = mval->br_state; | |
3949 | } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK && | |
3950 | mval[-1].br_startblock != DELAYSTARTBLOCK && | |
3951 | mval[-1].br_startblock != HOLESTARTBLOCK && | |
3952 | mval->br_startblock == mval[-1].br_startblock + | |
3953 | mval[-1].br_blockcount && | |
3954 | ((flags & XFS_BMAPI_IGSTATE) || | |
3955 | mval[-1].br_state == mval->br_state)) { | |
3956 | ASSERT(mval->br_startoff == | |
3957 | mval[-1].br_startoff + mval[-1].br_blockcount); | |
3958 | mval[-1].br_blockcount += mval->br_blockcount; | |
3959 | } else if (*n > 0 && | |
3960 | mval->br_startblock == DELAYSTARTBLOCK && | |
3961 | mval[-1].br_startblock == DELAYSTARTBLOCK && | |
3962 | mval->br_startoff == | |
3963 | mval[-1].br_startoff + mval[-1].br_blockcount) { | |
3964 | mval[-1].br_blockcount += mval->br_blockcount; | |
3965 | mval[-1].br_state = mval->br_state; | |
3966 | } else if (!((*n == 0) && | |
3967 | ((mval->br_startoff + mval->br_blockcount) <= | |
3968 | obno))) { | |
3969 | mval++; | |
3970 | (*n)++; | |
3971 | } | |
3972 | *map = mval; | |
e04426b9 DC |
3973 | } |
3974 | ||
3975 | /* | |
9e5987a7 | 3976 | * Map file blocks to filesystem blocks without allocation. |
e04426b9 DC |
3977 | */ |
3978 | int | |
9e5987a7 DC |
3979 | xfs_bmapi_read( |
3980 | struct xfs_inode *ip, | |
3981 | xfs_fileoff_t bno, | |
b447fe5a | 3982 | xfs_filblks_t len, |
9e5987a7 DC |
3983 | struct xfs_bmbt_irec *mval, |
3984 | int *nmap, | |
c315c90b | 3985 | int flags) |
b447fe5a | 3986 | { |
9e5987a7 DC |
3987 | struct xfs_mount *mp = ip->i_mount; |
3988 | struct xfs_ifork *ifp; | |
3989 | struct xfs_bmbt_irec got; | |
9e5987a7 DC |
3990 | xfs_fileoff_t obno; |
3991 | xfs_fileoff_t end; | |
334f3423 | 3992 | xfs_extnum_t idx; |
9e5987a7 | 3993 | int error; |
334f3423 | 3994 | bool eof = false; |
9e5987a7 | 3995 | int n = 0; |
3993baeb | 3996 | int whichfork = xfs_bmapi_whichfork(flags); |
b447fe5a | 3997 | |
9e5987a7 DC |
3998 | ASSERT(*nmap >= 1); |
3999 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE| | |
3993baeb | 4000 | XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK))); |
eef334e5 | 4001 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)); |
b447fe5a | 4002 | |
9e5987a7 DC |
4003 | if (unlikely(XFS_TEST_ERROR( |
4004 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
4005 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 4006 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
9e5987a7 | 4007 | XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp); |
2451337d | 4008 | return -EFSCORRUPTED; |
9e5987a7 | 4009 | } |
b447fe5a | 4010 | |
9e5987a7 | 4011 | if (XFS_FORCED_SHUTDOWN(mp)) |
2451337d | 4012 | return -EIO; |
9e5987a7 | 4013 | |
ff6d6af2 | 4014 | XFS_STATS_INC(mp, xs_blk_mapr); |
9e5987a7 DC |
4015 | |
4016 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
4017 | ||
3993baeb DW |
4018 | /* No CoW fork? Return a hole. */ |
4019 | if (whichfork == XFS_COW_FORK && !ifp) { | |
4020 | mval->br_startoff = bno; | |
4021 | mval->br_startblock = HOLESTARTBLOCK; | |
4022 | mval->br_blockcount = len; | |
4023 | mval->br_state = XFS_EXT_NORM; | |
4024 | *nmap = 1; | |
4025 | return 0; | |
4026 | } | |
4027 | ||
9e5987a7 DC |
4028 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
4029 | error = xfs_iread_extents(NULL, ip, whichfork); | |
4030 | if (error) | |
4031 | return error; | |
b447fe5a | 4032 | } |
b447fe5a | 4033 | |
334f3423 CH |
4034 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) |
4035 | eof = true; | |
9e5987a7 DC |
4036 | end = bno + len; |
4037 | obno = bno; | |
b447fe5a | 4038 | |
9e5987a7 DC |
4039 | while (bno < end && n < *nmap) { |
4040 | /* Reading past eof, act as though there's a hole up to end. */ | |
4041 | if (eof) | |
4042 | got.br_startoff = end; | |
4043 | if (got.br_startoff > bno) { | |
4044 | /* Reading in a hole. */ | |
4045 | mval->br_startoff = bno; | |
4046 | mval->br_startblock = HOLESTARTBLOCK; | |
4047 | mval->br_blockcount = | |
4048 | XFS_FILBLKS_MIN(len, got.br_startoff - bno); | |
4049 | mval->br_state = XFS_EXT_NORM; | |
4050 | bno += mval->br_blockcount; | |
4051 | len -= mval->br_blockcount; | |
4052 | mval++; | |
4053 | n++; | |
4054 | continue; | |
4055 | } | |
b447fe5a | 4056 | |
9e5987a7 DC |
4057 | /* set up the extent map to return. */ |
4058 | xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags); | |
4059 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4060 | ||
4061 | /* If we're done, stop now. */ | |
4062 | if (bno >= end || n >= *nmap) | |
4063 | break; | |
4064 | ||
4065 | /* Else go on to the next record. */ | |
334f3423 CH |
4066 | if (!xfs_iext_get_extent(ifp, ++idx, &got)) |
4067 | eof = true; | |
9e5987a7 DC |
4068 | } |
4069 | *nmap = n; | |
b447fe5a DC |
4070 | return 0; |
4071 | } | |
4072 | ||
f65e6fad BF |
4073 | /* |
4074 | * Add a delayed allocation extent to an inode. Blocks are reserved from the | |
4075 | * global pool and the extent inserted into the inode in-core extent tree. | |
4076 | * | |
4077 | * On entry, got refers to the first extent beyond the offset of the extent to | |
4078 | * allocate or eof is specified if no such extent exists. On return, got refers | |
4079 | * to the extent record that was inserted to the inode fork. | |
4080 | * | |
4081 | * Note that the allocated extent may have been merged with contiguous extents | |
4082 | * during insertion into the inode fork. Thus, got does not reflect the current | |
4083 | * state of the inode fork on return. If necessary, the caller can use lastx to | |
4084 | * look up the updated record in the inode fork. | |
4085 | */ | |
51446f5b | 4086 | int |
9e5987a7 DC |
4087 | xfs_bmapi_reserve_delalloc( |
4088 | struct xfs_inode *ip, | |
be51f811 | 4089 | int whichfork, |
974ae922 | 4090 | xfs_fileoff_t off, |
9e5987a7 | 4091 | xfs_filblks_t len, |
974ae922 | 4092 | xfs_filblks_t prealloc, |
9e5987a7 | 4093 | struct xfs_bmbt_irec *got, |
9e5987a7 DC |
4094 | xfs_extnum_t *lastx, |
4095 | int eof) | |
1da177e4 | 4096 | { |
c0dc7828 | 4097 | struct xfs_mount *mp = ip->i_mount; |
be51f811 | 4098 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
9e5987a7 DC |
4099 | xfs_extlen_t alen; |
4100 | xfs_extlen_t indlen; | |
4101 | char rt = XFS_IS_REALTIME_INODE(ip); | |
4102 | xfs_extlen_t extsz; | |
4103 | int error; | |
974ae922 | 4104 | xfs_fileoff_t aoff = off; |
c0dc7828 | 4105 | |
974ae922 BF |
4106 | /* |
4107 | * Cap the alloc length. Keep track of prealloc so we know whether to | |
4108 | * tag the inode before we return. | |
4109 | */ | |
4110 | alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN); | |
9e5987a7 DC |
4111 | if (!eof) |
4112 | alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff); | |
974ae922 BF |
4113 | if (prealloc && alen >= len) |
4114 | prealloc = alen - len; | |
1da177e4 | 4115 | |
9e5987a7 | 4116 | /* Figure out the extent size, adjust alen */ |
f7ca3522 DW |
4117 | if (whichfork == XFS_COW_FORK) |
4118 | extsz = xfs_get_cowextsz_hint(ip); | |
4119 | else | |
4120 | extsz = xfs_get_extsz_hint(ip); | |
9e5987a7 | 4121 | if (extsz) { |
65c5f419 CH |
4122 | struct xfs_bmbt_irec prev; |
4123 | ||
4124 | if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev)) | |
4125 | prev.br_startoff = NULLFILEOFF; | |
4126 | ||
4127 | error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof, | |
9e5987a7 DC |
4128 | 1, 0, &aoff, &alen); |
4129 | ASSERT(!error); | |
4130 | } | |
4131 | ||
4132 | if (rt) | |
4133 | extsz = alen / mp->m_sb.sb_rextsize; | |
4134 | ||
4135 | /* | |
4136 | * Make a transaction-less quota reservation for delayed allocation | |
4137 | * blocks. This number gets adjusted later. We return if we haven't | |
4138 | * allocated blocks already inside this loop. | |
4139 | */ | |
4140 | error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0, | |
4141 | rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
4142 | if (error) | |
4143 | return error; | |
4144 | ||
4145 | /* | |
4146 | * Split changing sb for alen and indlen since they could be coming | |
4147 | * from different places. | |
4148 | */ | |
4149 | indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen); | |
4150 | ASSERT(indlen > 0); | |
4151 | ||
4152 | if (rt) { | |
bab98bbe | 4153 | error = xfs_mod_frextents(mp, -((int64_t)extsz)); |
9e5987a7 | 4154 | } else { |
0d485ada | 4155 | error = xfs_mod_fdblocks(mp, -((int64_t)alen), false); |
9e5987a7 DC |
4156 | } |
4157 | ||
4158 | if (error) | |
4159 | goto out_unreserve_quota; | |
4160 | ||
0d485ada | 4161 | error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false); |
9e5987a7 DC |
4162 | if (error) |
4163 | goto out_unreserve_blocks; | |
4164 | ||
4165 | ||
4166 | ip->i_delayed_blks += alen; | |
4167 | ||
4168 | got->br_startoff = aoff; | |
4169 | got->br_startblock = nullstartblock(indlen); | |
4170 | got->br_blockcount = alen; | |
4171 | got->br_state = XFS_EXT_NORM; | |
9e5987a7 | 4172 | |
f65e6fad | 4173 | xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got); |
9e5987a7 | 4174 | |
974ae922 BF |
4175 | /* |
4176 | * Tag the inode if blocks were preallocated. Note that COW fork | |
4177 | * preallocation can occur at the start or end of the extent, even when | |
4178 | * prealloc == 0, so we must also check the aligned offset and length. | |
4179 | */ | |
4180 | if (whichfork == XFS_DATA_FORK && prealloc) | |
4181 | xfs_inode_set_eofblocks_tag(ip); | |
4182 | if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len)) | |
4183 | xfs_inode_set_cowblocks_tag(ip); | |
4184 | ||
9e5987a7 DC |
4185 | return 0; |
4186 | ||
4187 | out_unreserve_blocks: | |
4188 | if (rt) | |
bab98bbe | 4189 | xfs_mod_frextents(mp, extsz); |
9e5987a7 | 4190 | else |
0d485ada | 4191 | xfs_mod_fdblocks(mp, alen, false); |
9e5987a7 DC |
4192 | out_unreserve_quota: |
4193 | if (XFS_IS_QUOTA_ON(mp)) | |
4194 | xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ? | |
4195 | XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
4196 | return error; | |
4197 | } | |
4198 | ||
cf11da9c DC |
4199 | static int |
4200 | xfs_bmapi_allocate( | |
9e5987a7 DC |
4201 | struct xfs_bmalloca *bma) |
4202 | { | |
4203 | struct xfs_mount *mp = bma->ip->i_mount; | |
60b4984f | 4204 | int whichfork = xfs_bmapi_whichfork(bma->flags); |
9e5987a7 DC |
4205 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4206 | int tmp_logflags = 0; | |
4207 | int error; | |
9e5987a7 DC |
4208 | |
4209 | ASSERT(bma->length > 0); | |
4210 | ||
1da177e4 | 4211 | /* |
9e5987a7 DC |
4212 | * For the wasdelay case, we could also just allocate the stuff asked |
4213 | * for in this bmap call but that wouldn't be as good. | |
1da177e4 | 4214 | */ |
9e5987a7 DC |
4215 | if (bma->wasdel) { |
4216 | bma->length = (xfs_extlen_t)bma->got.br_blockcount; | |
4217 | bma->offset = bma->got.br_startoff; | |
0e8d630b | 4218 | if (bma->idx) { |
9e5987a7 DC |
4219 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), |
4220 | &bma->prev); | |
1da177e4 | 4221 | } |
9e5987a7 DC |
4222 | } else { |
4223 | bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN); | |
4224 | if (!bma->eof) | |
4225 | bma->length = XFS_FILBLKS_MIN(bma->length, | |
4226 | bma->got.br_startoff - bma->offset); | |
1da177e4 | 4227 | } |
1da177e4 | 4228 | |
9e5987a7 | 4229 | /* |
292378ed DC |
4230 | * Set the data type being allocated. For the data fork, the first data |
4231 | * in the file is treated differently to all other allocations. For the | |
4232 | * attribute fork, we only need to ensure the allocated range is not on | |
4233 | * the busy list. | |
9e5987a7 DC |
4234 | */ |
4235 | if (!(bma->flags & XFS_BMAPI_METADATA)) { | |
292378ed DC |
4236 | bma->datatype = XFS_ALLOC_NOBUSY; |
4237 | if (whichfork == XFS_DATA_FORK) { | |
4238 | if (bma->offset == 0) | |
4239 | bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA; | |
4240 | else | |
4241 | bma->datatype |= XFS_ALLOC_USERDATA; | |
4242 | } | |
3fbbbea3 | 4243 | if (bma->flags & XFS_BMAPI_ZERO) |
292378ed | 4244 | bma->datatype |= XFS_ALLOC_USERDATA_ZERO; |
9e5987a7 | 4245 | } |
1da177e4 | 4246 | |
9e5987a7 | 4247 | bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1; |
0b1b213f | 4248 | |
9e5987a7 DC |
4249 | /* |
4250 | * Only want to do the alignment at the eof if it is userdata and | |
4251 | * allocation length is larger than a stripe unit. | |
4252 | */ | |
4253 | if (mp->m_dalign && bma->length >= mp->m_dalign && | |
4254 | !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) { | |
4255 | error = xfs_bmap_isaeof(bma, whichfork); | |
4256 | if (error) | |
4257 | return error; | |
1da177e4 | 4258 | } |
8096b1eb | 4259 | |
9e5987a7 DC |
4260 | error = xfs_bmap_alloc(bma); |
4261 | if (error) | |
1da177e4 | 4262 | return error; |
9e5987a7 | 4263 | |
9e5987a7 DC |
4264 | if (bma->cur) |
4265 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
4266 | if (bma->blkno == NULLFSBLOCK) | |
1da177e4 | 4267 | return 0; |
9e5987a7 DC |
4268 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { |
4269 | bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork); | |
4270 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
2c3234d1 | 4271 | bma->cur->bc_private.b.dfops = bma->dfops; |
1da177e4 | 4272 | } |
9e5987a7 DC |
4273 | /* |
4274 | * Bump the number of extents we've allocated | |
4275 | * in this call. | |
4276 | */ | |
4277 | bma->nallocs++; | |
4278 | ||
4279 | if (bma->cur) | |
4280 | bma->cur->bc_private.b.flags = | |
4281 | bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; | |
4282 | ||
4283 | bma->got.br_startoff = bma->offset; | |
4284 | bma->got.br_startblock = bma->blkno; | |
4285 | bma->got.br_blockcount = bma->length; | |
4286 | bma->got.br_state = XFS_EXT_NORM; | |
b4e9181e | 4287 | |
1da177e4 | 4288 | /* |
05a630d7 DW |
4289 | * In the data fork, a wasdelay extent has been initialized, so |
4290 | * shouldn't be flagged as unwritten. | |
4291 | * | |
4292 | * For the cow fork, however, we convert delalloc reservations | |
4293 | * (extents allocated for speculative preallocation) to | |
4294 | * allocated unwritten extents, and only convert the unwritten | |
4295 | * extents to real extents when we're about to write the data. | |
1da177e4 | 4296 | */ |
05a630d7 DW |
4297 | if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) && |
4298 | (bma->flags & XFS_BMAPI_PREALLOC) && | |
9e5987a7 DC |
4299 | xfs_sb_version_hasextflgbit(&mp->m_sb)) |
4300 | bma->got.br_state = XFS_EXT_UNWRITTEN; | |
4301 | ||
4302 | if (bma->wasdel) | |
60b4984f | 4303 | error = xfs_bmap_add_extent_delay_real(bma, whichfork); |
9e5987a7 | 4304 | else |
6d04558f CH |
4305 | error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip, |
4306 | whichfork, &bma->idx, &bma->cur, &bma->got, | |
4307 | bma->firstblock, bma->dfops, &bma->logflags); | |
9e5987a7 DC |
4308 | |
4309 | bma->logflags |= tmp_logflags; | |
4310 | if (error) | |
4311 | return error; | |
4312 | ||
4313 | /* | |
4314 | * Update our extent pointer, given that xfs_bmap_add_extent_delay_real | |
4315 | * or xfs_bmap_add_extent_hole_real might have merged it into one of | |
4316 | * the neighbouring ones. | |
4317 | */ | |
4318 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got); | |
4319 | ||
4320 | ASSERT(bma->got.br_startoff <= bma->offset); | |
4321 | ASSERT(bma->got.br_startoff + bma->got.br_blockcount >= | |
4322 | bma->offset + bma->length); | |
4323 | ASSERT(bma->got.br_state == XFS_EXT_NORM || | |
4324 | bma->got.br_state == XFS_EXT_UNWRITTEN); | |
4325 | return 0; | |
4326 | } | |
4327 | ||
9e5987a7 DC |
4328 | STATIC int |
4329 | xfs_bmapi_convert_unwritten( | |
4330 | struct xfs_bmalloca *bma, | |
4331 | struct xfs_bmbt_irec *mval, | |
4332 | xfs_filblks_t len, | |
4333 | int flags) | |
4334 | { | |
3993baeb | 4335 | int whichfork = xfs_bmapi_whichfork(flags); |
9e5987a7 DC |
4336 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4337 | int tmp_logflags = 0; | |
4338 | int error; | |
4339 | ||
4340 | /* check if we need to do unwritten->real conversion */ | |
4341 | if (mval->br_state == XFS_EXT_UNWRITTEN && | |
4342 | (flags & XFS_BMAPI_PREALLOC)) | |
4343 | return 0; | |
4344 | ||
4345 | /* check if we need to do real->unwritten conversion */ | |
4346 | if (mval->br_state == XFS_EXT_NORM && | |
4347 | (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) != | |
4348 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) | |
4349 | return 0; | |
4350 | ||
4351 | /* | |
4352 | * Modify (by adding) the state flag, if writing. | |
4353 | */ | |
4354 | ASSERT(mval->br_blockcount <= len); | |
4355 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4356 | bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp, | |
4357 | bma->ip, whichfork); | |
4358 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
2c3234d1 | 4359 | bma->cur->bc_private.b.dfops = bma->dfops; |
1da177e4 | 4360 | } |
9e5987a7 DC |
4361 | mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN) |
4362 | ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
5575acc7 | 4363 | |
3fbbbea3 DC |
4364 | /* |
4365 | * Before insertion into the bmbt, zero the range being converted | |
4366 | * if required. | |
4367 | */ | |
4368 | if (flags & XFS_BMAPI_ZERO) { | |
4369 | error = xfs_zero_extent(bma->ip, mval->br_startblock, | |
4370 | mval->br_blockcount); | |
4371 | if (error) | |
4372 | return error; | |
4373 | } | |
4374 | ||
05a630d7 DW |
4375 | error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork, |
4376 | &bma->idx, &bma->cur, mval, bma->firstblock, bma->dfops, | |
9e5987a7 | 4377 | &tmp_logflags); |
2e588a46 BF |
4378 | /* |
4379 | * Log the inode core unconditionally in the unwritten extent conversion | |
4380 | * path because the conversion might not have done so (e.g., if the | |
4381 | * extent count hasn't changed). We need to make sure the inode is dirty | |
4382 | * in the transaction for the sake of fsync(), even if nothing has | |
4383 | * changed, because fsync() will not force the log for this transaction | |
4384 | * unless it sees the inode pinned. | |
05a630d7 DW |
4385 | * |
4386 | * Note: If we're only converting cow fork extents, there aren't | |
4387 | * any on-disk updates to make, so we don't need to log anything. | |
2e588a46 | 4388 | */ |
05a630d7 DW |
4389 | if (whichfork != XFS_COW_FORK) |
4390 | bma->logflags |= tmp_logflags | XFS_ILOG_CORE; | |
9e5987a7 DC |
4391 | if (error) |
4392 | return error; | |
4393 | ||
4394 | /* | |
4395 | * Update our extent pointer, given that | |
4396 | * xfs_bmap_add_extent_unwritten_real might have merged it into one | |
4397 | * of the neighbouring ones. | |
4398 | */ | |
4399 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got); | |
4400 | ||
4401 | /* | |
4402 | * We may have combined previously unwritten space with written space, | |
4403 | * so generate another request. | |
4404 | */ | |
4405 | if (mval->br_blockcount < len) | |
2451337d | 4406 | return -EAGAIN; |
9e5987a7 DC |
4407 | return 0; |
4408 | } | |
4409 | ||
4410 | /* | |
4411 | * Map file blocks to filesystem blocks, and allocate blocks or convert the | |
4412 | * extent state if necessary. Details behaviour is controlled by the flags | |
4413 | * parameter. Only allocates blocks from a single allocation group, to avoid | |
4414 | * locking problems. | |
4415 | * | |
4416 | * The returned value in "firstblock" from the first call in a transaction | |
4417 | * must be remembered and presented to subsequent calls in "firstblock". | |
4418 | * An upper bound for the number of blocks to be allocated is supplied to | |
4419 | * the first call in "total"; if no allocation group has that many free | |
4420 | * blocks then the call will fail (return NULLFSBLOCK in "firstblock"). | |
4421 | */ | |
4422 | int | |
4423 | xfs_bmapi_write( | |
4424 | struct xfs_trans *tp, /* transaction pointer */ | |
4425 | struct xfs_inode *ip, /* incore inode */ | |
4426 | xfs_fileoff_t bno, /* starting file offs. mapped */ | |
4427 | xfs_filblks_t len, /* length to map in file */ | |
4428 | int flags, /* XFS_BMAPI_... */ | |
4429 | xfs_fsblock_t *firstblock, /* first allocated block | |
4430 | controls a.g. for allocs */ | |
4431 | xfs_extlen_t total, /* total blocks needed */ | |
4432 | struct xfs_bmbt_irec *mval, /* output: map values */ | |
4433 | int *nmap, /* i/o: mval size/count */ | |
2c3234d1 | 4434 | struct xfs_defer_ops *dfops) /* i/o: list extents to free */ |
9e5987a7 DC |
4435 | { |
4436 | struct xfs_mount *mp = ip->i_mount; | |
4437 | struct xfs_ifork *ifp; | |
a30b0367 | 4438 | struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */ |
9e5987a7 | 4439 | xfs_fileoff_t end; /* end of mapped file region */ |
2d58f6ef | 4440 | bool eof = false; /* after the end of extents */ |
9e5987a7 DC |
4441 | int error; /* error return */ |
4442 | int n; /* current extent index */ | |
4443 | xfs_fileoff_t obno; /* old block number (offset) */ | |
4444 | int whichfork; /* data or attr fork */ | |
9e5987a7 DC |
4445 | |
4446 | #ifdef DEBUG | |
4447 | xfs_fileoff_t orig_bno; /* original block number value */ | |
4448 | int orig_flags; /* original flags arg value */ | |
4449 | xfs_filblks_t orig_len; /* original value of len arg */ | |
4450 | struct xfs_bmbt_irec *orig_mval; /* original value of mval */ | |
4451 | int orig_nmap; /* original value of *nmap */ | |
4452 | ||
4453 | orig_bno = bno; | |
4454 | orig_len = len; | |
4455 | orig_flags = flags; | |
4456 | orig_mval = mval; | |
4457 | orig_nmap = *nmap; | |
4458 | #endif | |
60b4984f | 4459 | whichfork = xfs_bmapi_whichfork(flags); |
9e5987a7 DC |
4460 | |
4461 | ASSERT(*nmap >= 1); | |
4462 | ASSERT(*nmap <= XFS_BMAP_MAX_NMAP); | |
4463 | ASSERT(!(flags & XFS_BMAPI_IGSTATE)); | |
05a630d7 DW |
4464 | ASSERT(tp != NULL || |
4465 | (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) == | |
4466 | (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)); | |
9e5987a7 | 4467 | ASSERT(len > 0); |
f3508bcd | 4468 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL); |
eef334e5 | 4469 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
6ebd5a44 | 4470 | ASSERT(!(flags & XFS_BMAPI_REMAP)); |
9e5987a7 | 4471 | |
3fbbbea3 DC |
4472 | /* zeroing is for currently only for data extents, not metadata */ |
4473 | ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) != | |
4474 | (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)); | |
4475 | /* | |
4476 | * we can allocate unwritten extents or pre-zero allocated blocks, | |
4477 | * but it makes no sense to do both at once. This would result in | |
4478 | * zeroing the unwritten extent twice, but it still being an | |
4479 | * unwritten extent.... | |
4480 | */ | |
4481 | ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) != | |
4482 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)); | |
4483 | ||
9e5987a7 DC |
4484 | if (unlikely(XFS_TEST_ERROR( |
4485 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
f3508bcd | 4486 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), |
9e24cfd0 | 4487 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
9e5987a7 | 4488 | XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp); |
2451337d | 4489 | return -EFSCORRUPTED; |
5575acc7 KD |
4490 | } |
4491 | ||
9e5987a7 | 4492 | if (XFS_FORCED_SHUTDOWN(mp)) |
2451337d | 4493 | return -EIO; |
9e5987a7 DC |
4494 | |
4495 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
4496 | ||
ff6d6af2 | 4497 | XFS_STATS_INC(mp, xs_blk_mapw); |
9e5987a7 | 4498 | |
9e5987a7 DC |
4499 | if (*firstblock == NULLFSBLOCK) { |
4500 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE) | |
4501 | bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1; | |
4502 | else | |
4503 | bma.minleft = 1; | |
4504 | } else { | |
4505 | bma.minleft = 0; | |
4506 | } | |
4507 | ||
4508 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
4509 | error = xfs_iread_extents(tp, ip, whichfork); | |
4510 | if (error) | |
4511 | goto error0; | |
4512 | } | |
4513 | ||
9e5987a7 DC |
4514 | n = 0; |
4515 | end = bno + len; | |
4516 | obno = bno; | |
4517 | ||
2d58f6ef CH |
4518 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got)) |
4519 | eof = true; | |
4520 | if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev)) | |
4521 | bma.prev.br_startoff = NULLFILEOFF; | |
9e5987a7 DC |
4522 | bma.tp = tp; |
4523 | bma.ip = ip; | |
4524 | bma.total = total; | |
292378ed | 4525 | bma.datatype = 0; |
2c3234d1 | 4526 | bma.dfops = dfops; |
9e5987a7 DC |
4527 | bma.firstblock = firstblock; |
4528 | ||
9e5987a7 | 4529 | while (bno < end && n < *nmap) { |
d2b3964a CH |
4530 | bool need_alloc = false, wasdelay = false; |
4531 | ||
4532 | /* in hole or beyoned EOF? */ | |
4533 | if (eof || bma.got.br_startoff > bno) { | |
4534 | if (flags & XFS_BMAPI_DELALLOC) { | |
4535 | /* | |
4536 | * For the COW fork we can reasonably get a | |
4537 | * request for converting an extent that races | |
4538 | * with other threads already having converted | |
4539 | * part of it, as there converting COW to | |
4540 | * regular blocks is not protected using the | |
4541 | * IOLOCK. | |
4542 | */ | |
4543 | ASSERT(flags & XFS_BMAPI_COWFORK); | |
4544 | if (!(flags & XFS_BMAPI_COWFORK)) { | |
4545 | error = -EIO; | |
4546 | goto error0; | |
4547 | } | |
4548 | ||
4549 | if (eof || bno >= end) | |
4550 | break; | |
4551 | } else { | |
4552 | need_alloc = true; | |
4553 | } | |
6ebd5a44 CH |
4554 | } else if (isnullstartblock(bma.got.br_startblock)) { |
4555 | wasdelay = true; | |
d2b3964a | 4556 | } |
f65306ea | 4557 | |
1da177e4 | 4558 | /* |
9e5987a7 DC |
4559 | * First, deal with the hole before the allocated space |
4560 | * that we found, if any. | |
1da177e4 | 4561 | */ |
d2b3964a | 4562 | if (need_alloc || wasdelay) { |
9e5987a7 DC |
4563 | bma.eof = eof; |
4564 | bma.conv = !!(flags & XFS_BMAPI_CONVERT); | |
4565 | bma.wasdel = wasdelay; | |
4566 | bma.offset = bno; | |
4567 | bma.flags = flags; | |
4568 | ||
1da177e4 | 4569 | /* |
9e5987a7 DC |
4570 | * There's a 32/64 bit type mismatch between the |
4571 | * allocation length request (which can be 64 bits in | |
4572 | * length) and the bma length request, which is | |
4573 | * xfs_extlen_t and therefore 32 bits. Hence we have to | |
4574 | * check for 32-bit overflows and handle them here. | |
1da177e4 | 4575 | */ |
9e5987a7 DC |
4576 | if (len > (xfs_filblks_t)MAXEXTLEN) |
4577 | bma.length = MAXEXTLEN; | |
4578 | else | |
4579 | bma.length = len; | |
4580 | ||
4581 | ASSERT(len > 0); | |
4582 | ASSERT(bma.length > 0); | |
4583 | error = xfs_bmapi_allocate(&bma); | |
1da177e4 LT |
4584 | if (error) |
4585 | goto error0; | |
9e5987a7 DC |
4586 | if (bma.blkno == NULLFSBLOCK) |
4587 | break; | |
174edb0e DW |
4588 | |
4589 | /* | |
4590 | * If this is a CoW allocation, record the data in | |
4591 | * the refcount btree for orphan recovery. | |
4592 | */ | |
4593 | if (whichfork == XFS_COW_FORK) { | |
4594 | error = xfs_refcount_alloc_cow_extent(mp, dfops, | |
4595 | bma.blkno, bma.length); | |
4596 | if (error) | |
4597 | goto error0; | |
4598 | } | |
1da177e4 | 4599 | } |
9e5987a7 DC |
4600 | |
4601 | /* Deal with the allocated space we found. */ | |
4602 | xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno, | |
4603 | end, n, flags); | |
4604 | ||
4605 | /* Execute unwritten extent conversion if necessary */ | |
4606 | error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags); | |
2451337d | 4607 | if (error == -EAGAIN) |
9e5987a7 DC |
4608 | continue; |
4609 | if (error) | |
4610 | goto error0; | |
4611 | ||
4612 | /* update the extent map to return */ | |
4613 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4614 | ||
4615 | /* | |
4616 | * If we're done, stop now. Stop when we've allocated | |
4617 | * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise | |
4618 | * the transaction may get too big. | |
4619 | */ | |
4620 | if (bno >= end || n >= *nmap || bma.nallocs >= *nmap) | |
4621 | break; | |
4622 | ||
4623 | /* Else go on to the next record. */ | |
4624 | bma.prev = bma.got; | |
2d58f6ef CH |
4625 | if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got)) |
4626 | eof = true; | |
9e5987a7 DC |
4627 | } |
4628 | *nmap = n; | |
4629 | ||
4630 | /* | |
4631 | * Transform from btree to extents, give it cur. | |
4632 | */ | |
4633 | if (xfs_bmap_wants_extents(ip, whichfork)) { | |
4634 | int tmp_logflags = 0; | |
4635 | ||
4636 | ASSERT(bma.cur); | |
4637 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, | |
4638 | &tmp_logflags, whichfork); | |
4639 | bma.logflags |= tmp_logflags; | |
4640 | if (error) | |
4641 | goto error0; | |
4642 | } | |
4643 | ||
4644 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE || | |
4645 | XFS_IFORK_NEXTENTS(ip, whichfork) > | |
4646 | XFS_IFORK_MAXEXT(ip, whichfork)); | |
4647 | error = 0; | |
4648 | error0: | |
4649 | /* | |
4650 | * Log everything. Do this after conversion, there's no point in | |
4651 | * logging the extent records if we've converted to btree format. | |
4652 | */ | |
4653 | if ((bma.logflags & xfs_ilog_fext(whichfork)) && | |
4654 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
4655 | bma.logflags &= ~xfs_ilog_fext(whichfork); | |
4656 | else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) && | |
4657 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) | |
4658 | bma.logflags &= ~xfs_ilog_fbroot(whichfork); | |
4659 | /* | |
4660 | * Log whatever the flags say, even if error. Otherwise we might miss | |
4661 | * detecting a case where the data is changed, there's an error, | |
4662 | * and it's not logged so we don't shutdown when we should. | |
4663 | */ | |
4664 | if (bma.logflags) | |
4665 | xfs_trans_log_inode(tp, ip, bma.logflags); | |
4666 | ||
4667 | if (bma.cur) { | |
4668 | if (!error) { | |
4669 | ASSERT(*firstblock == NULLFSBLOCK || | |
410d17f6 | 4670 | XFS_FSB_TO_AGNO(mp, *firstblock) <= |
9e5987a7 | 4671 | XFS_FSB_TO_AGNO(mp, |
410d17f6 | 4672 | bma.cur->bc_private.b.firstblock)); |
9e5987a7 | 4673 | *firstblock = bma.cur->bc_private.b.firstblock; |
1da177e4 | 4674 | } |
9e5987a7 DC |
4675 | xfs_btree_del_cursor(bma.cur, |
4676 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
4677 | } | |
4678 | if (!error) | |
4679 | xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval, | |
4680 | orig_nmap, *nmap); | |
4681 | return error; | |
4682 | } | |
06d10dd9 | 4683 | |
6ebd5a44 CH |
4684 | static int |
4685 | xfs_bmapi_remap( | |
4686 | struct xfs_trans *tp, | |
4687 | struct xfs_inode *ip, | |
4688 | xfs_fileoff_t bno, | |
4689 | xfs_filblks_t len, | |
4690 | xfs_fsblock_t startblock, | |
4691 | struct xfs_defer_ops *dfops) | |
4692 | { | |
4693 | struct xfs_mount *mp = ip->i_mount; | |
4694 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
4695 | struct xfs_btree_cur *cur = NULL; | |
4696 | xfs_fsblock_t firstblock = NULLFSBLOCK; | |
4697 | struct xfs_bmbt_irec got; | |
4698 | xfs_extnum_t idx; | |
4699 | int logflags = 0, error; | |
4700 | ||
4701 | ASSERT(len > 0); | |
4702 | ASSERT(len <= (xfs_filblks_t)MAXEXTLEN); | |
4703 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
4704 | ||
4705 | if (unlikely(XFS_TEST_ERROR( | |
4706 | (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS && | |
4707 | XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 4708 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
6ebd5a44 CH |
4709 | XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp); |
4710 | return -EFSCORRUPTED; | |
4711 | } | |
4712 | ||
4713 | if (XFS_FORCED_SHUTDOWN(mp)) | |
4714 | return -EIO; | |
4715 | ||
4716 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
4717 | error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK); | |
4718 | if (error) | |
4719 | return error; | |
4720 | } | |
4721 | ||
4722 | if (xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) { | |
4723 | /* make sure we only reflink into a hole. */ | |
4724 | ASSERT(got.br_startoff > bno); | |
4725 | ASSERT(got.br_startoff - bno >= len); | |
4726 | } | |
4727 | ||
bf8eadba CH |
4728 | ip->i_d.di_nblocks += len; |
4729 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
6ebd5a44 CH |
4730 | |
4731 | if (ifp->if_flags & XFS_IFBROOT) { | |
4732 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
4733 | cur->bc_private.b.firstblock = firstblock; | |
4734 | cur->bc_private.b.dfops = dfops; | |
4735 | cur->bc_private.b.flags = 0; | |
4736 | } | |
4737 | ||
4738 | got.br_startoff = bno; | |
4739 | got.br_startblock = startblock; | |
4740 | got.br_blockcount = len; | |
4741 | got.br_state = XFS_EXT_NORM; | |
4742 | ||
4743 | error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &idx, &cur, | |
4744 | &got, &firstblock, dfops, &logflags); | |
4745 | if (error) | |
4746 | goto error0; | |
4747 | ||
4748 | if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) { | |
4749 | int tmp_logflags = 0; | |
4750 | ||
4751 | error = xfs_bmap_btree_to_extents(tp, ip, cur, | |
4752 | &tmp_logflags, XFS_DATA_FORK); | |
4753 | logflags |= tmp_logflags; | |
4754 | } | |
4755 | ||
4756 | error0: | |
4757 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) | |
4758 | logflags &= ~XFS_ILOG_DEXT; | |
4759 | else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) | |
4760 | logflags &= ~XFS_ILOG_DBROOT; | |
4761 | ||
4762 | if (logflags) | |
4763 | xfs_trans_log_inode(tp, ip, logflags); | |
4764 | if (cur) { | |
4765 | xfs_btree_del_cursor(cur, | |
4766 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
4767 | } | |
4768 | return error; | |
4769 | } | |
4770 | ||
a9bd24ac BF |
4771 | /* |
4772 | * When a delalloc extent is split (e.g., due to a hole punch), the original | |
4773 | * indlen reservation must be shared across the two new extents that are left | |
4774 | * behind. | |
4775 | * | |
4776 | * Given the original reservation and the worst case indlen for the two new | |
4777 | * extents (as calculated by xfs_bmap_worst_indlen()), split the original | |
d34999c9 BF |
4778 | * reservation fairly across the two new extents. If necessary, steal available |
4779 | * blocks from a deleted extent to make up a reservation deficiency (e.g., if | |
4780 | * ores == 1). The number of stolen blocks is returned. The availability and | |
4781 | * subsequent accounting of stolen blocks is the responsibility of the caller. | |
a9bd24ac | 4782 | */ |
d34999c9 | 4783 | static xfs_filblks_t |
a9bd24ac BF |
4784 | xfs_bmap_split_indlen( |
4785 | xfs_filblks_t ores, /* original res. */ | |
4786 | xfs_filblks_t *indlen1, /* ext1 worst indlen */ | |
d34999c9 BF |
4787 | xfs_filblks_t *indlen2, /* ext2 worst indlen */ |
4788 | xfs_filblks_t avail) /* stealable blocks */ | |
a9bd24ac BF |
4789 | { |
4790 | xfs_filblks_t len1 = *indlen1; | |
4791 | xfs_filblks_t len2 = *indlen2; | |
4792 | xfs_filblks_t nres = len1 + len2; /* new total res. */ | |
d34999c9 | 4793 | xfs_filblks_t stolen = 0; |
75d65361 | 4794 | xfs_filblks_t resfactor; |
d34999c9 BF |
4795 | |
4796 | /* | |
4797 | * Steal as many blocks as we can to try and satisfy the worst case | |
4798 | * indlen for both new extents. | |
4799 | */ | |
75d65361 BF |
4800 | if (ores < nres && avail) |
4801 | stolen = XFS_FILBLKS_MIN(nres - ores, avail); | |
4802 | ores += stolen; | |
4803 | ||
4804 | /* nothing else to do if we've satisfied the new reservation */ | |
4805 | if (ores >= nres) | |
4806 | return stolen; | |
4807 | ||
4808 | /* | |
4809 | * We can't meet the total required reservation for the two extents. | |
4810 | * Calculate the percent of the overall shortage between both extents | |
4811 | * and apply this percentage to each of the requested indlen values. | |
4812 | * This distributes the shortage fairly and reduces the chances that one | |
4813 | * of the two extents is left with nothing when extents are repeatedly | |
4814 | * split. | |
4815 | */ | |
4816 | resfactor = (ores * 100); | |
4817 | do_div(resfactor, nres); | |
4818 | len1 *= resfactor; | |
4819 | do_div(len1, 100); | |
4820 | len2 *= resfactor; | |
4821 | do_div(len2, 100); | |
4822 | ASSERT(len1 + len2 <= ores); | |
4823 | ASSERT(len1 < *indlen1 && len2 < *indlen2); | |
a9bd24ac BF |
4824 | |
4825 | /* | |
75d65361 BF |
4826 | * Hand out the remainder to each extent. If one of the two reservations |
4827 | * is zero, we want to make sure that one gets a block first. The loop | |
4828 | * below starts with len1, so hand len2 a block right off the bat if it | |
4829 | * is zero. | |
a9bd24ac | 4830 | */ |
75d65361 BF |
4831 | ores -= (len1 + len2); |
4832 | ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores); | |
4833 | if (ores && !len2 && *indlen2) { | |
4834 | len2++; | |
4835 | ores--; | |
4836 | } | |
4837 | while (ores) { | |
4838 | if (len1 < *indlen1) { | |
4839 | len1++; | |
4840 | ores--; | |
a9bd24ac | 4841 | } |
75d65361 | 4842 | if (!ores) |
a9bd24ac | 4843 | break; |
75d65361 BF |
4844 | if (len2 < *indlen2) { |
4845 | len2++; | |
4846 | ores--; | |
a9bd24ac BF |
4847 | } |
4848 | } | |
4849 | ||
4850 | *indlen1 = len1; | |
4851 | *indlen2 = len2; | |
d34999c9 BF |
4852 | |
4853 | return stolen; | |
a9bd24ac BF |
4854 | } |
4855 | ||
fa5c836c CH |
4856 | int |
4857 | xfs_bmap_del_extent_delay( | |
4858 | struct xfs_inode *ip, | |
4859 | int whichfork, | |
4860 | xfs_extnum_t *idx, | |
4861 | struct xfs_bmbt_irec *got, | |
4862 | struct xfs_bmbt_irec *del) | |
4863 | { | |
4864 | struct xfs_mount *mp = ip->i_mount; | |
4865 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
4866 | struct xfs_bmbt_irec new; | |
4867 | int64_t da_old, da_new, da_diff = 0; | |
4868 | xfs_fileoff_t del_endoff, got_endoff; | |
4869 | xfs_filblks_t got_indlen, new_indlen, stolen; | |
4870 | int error = 0, state = 0; | |
4871 | bool isrt; | |
4872 | ||
4873 | XFS_STATS_INC(mp, xs_del_exlist); | |
4874 | ||
4875 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); | |
4876 | del_endoff = del->br_startoff + del->br_blockcount; | |
4877 | got_endoff = got->br_startoff + got->br_blockcount; | |
4878 | da_old = startblockval(got->br_startblock); | |
4879 | da_new = 0; | |
4880 | ||
4881 | ASSERT(*idx >= 0); | |
5d829300 | 4882 | ASSERT(*idx <= xfs_iext_count(ifp)); |
fa5c836c CH |
4883 | ASSERT(del->br_blockcount > 0); |
4884 | ASSERT(got->br_startoff <= del->br_startoff); | |
4885 | ASSERT(got_endoff >= del_endoff); | |
4886 | ||
4887 | if (isrt) { | |
4f1adf33 | 4888 | uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount); |
fa5c836c CH |
4889 | |
4890 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
4891 | xfs_mod_frextents(mp, rtexts); | |
4892 | } | |
4893 | ||
4894 | /* | |
4895 | * Update the inode delalloc counter now and wait to update the | |
4896 | * sb counters as we might have to borrow some blocks for the | |
4897 | * indirect block accounting. | |
4898 | */ | |
4fd29ec4 DW |
4899 | error = xfs_trans_reserve_quota_nblks(NULL, ip, |
4900 | -((long)del->br_blockcount), 0, | |
fa5c836c | 4901 | isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); |
4fd29ec4 DW |
4902 | if (error) |
4903 | return error; | |
fa5c836c CH |
4904 | ip->i_delayed_blks -= del->br_blockcount; |
4905 | ||
4906 | if (whichfork == XFS_COW_FORK) | |
4907 | state |= BMAP_COWFORK; | |
4908 | ||
4909 | if (got->br_startoff == del->br_startoff) | |
4910 | state |= BMAP_LEFT_CONTIG; | |
4911 | if (got_endoff == del_endoff) | |
4912 | state |= BMAP_RIGHT_CONTIG; | |
4913 | ||
4914 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
4915 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
4916 | /* | |
4917 | * Matches the whole extent. Delete the entry. | |
4918 | */ | |
4919 | xfs_iext_remove(ip, *idx, 1, state); | |
4920 | --*idx; | |
4921 | break; | |
4922 | case BMAP_LEFT_CONTIG: | |
4923 | /* | |
4924 | * Deleting the first part of the extent. | |
4925 | */ | |
4926 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4927 | got->br_startoff = del_endoff; | |
4928 | got->br_blockcount -= del->br_blockcount; | |
4929 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4930 | got->br_blockcount), da_old); | |
4931 | got->br_startblock = nullstartblock((int)da_new); | |
4932 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4933 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
4934 | break; | |
4935 | case BMAP_RIGHT_CONTIG: | |
4936 | /* | |
4937 | * Deleting the last part of the extent. | |
4938 | */ | |
4939 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4940 | got->br_blockcount = got->br_blockcount - del->br_blockcount; | |
4941 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4942 | got->br_blockcount), da_old); | |
4943 | got->br_startblock = nullstartblock((int)da_new); | |
4944 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4945 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
4946 | break; | |
4947 | case 0: | |
4948 | /* | |
4949 | * Deleting the middle of the extent. | |
4950 | * | |
4951 | * Distribute the original indlen reservation across the two new | |
4952 | * extents. Steal blocks from the deleted extent if necessary. | |
4953 | * Stealing blocks simply fudges the fdblocks accounting below. | |
4954 | * Warn if either of the new indlen reservations is zero as this | |
4955 | * can lead to delalloc problems. | |
4956 | */ | |
4957 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4958 | ||
4959 | got->br_blockcount = del->br_startoff - got->br_startoff; | |
4960 | got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount); | |
4961 | ||
4962 | new.br_blockcount = got_endoff - del_endoff; | |
4963 | new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
4964 | ||
4965 | WARN_ON_ONCE(!got_indlen || !new_indlen); | |
4966 | stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen, | |
4967 | del->br_blockcount); | |
4968 | ||
4969 | got->br_startblock = nullstartblock((int)got_indlen); | |
4970 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4971 | trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_); | |
4972 | ||
4973 | new.br_startoff = del_endoff; | |
4974 | new.br_state = got->br_state; | |
4975 | new.br_startblock = nullstartblock((int)new_indlen); | |
4976 | ||
4977 | ++*idx; | |
4978 | xfs_iext_insert(ip, *idx, 1, &new, state); | |
4979 | ||
4980 | da_new = got_indlen + new_indlen - stolen; | |
4981 | del->br_blockcount -= stolen; | |
4982 | break; | |
4983 | } | |
4984 | ||
4985 | ASSERT(da_old >= da_new); | |
4986 | da_diff = da_old - da_new; | |
4987 | if (!isrt) | |
4988 | da_diff += del->br_blockcount; | |
4989 | if (da_diff) | |
4990 | xfs_mod_fdblocks(mp, da_diff, false); | |
4991 | return error; | |
4992 | } | |
4993 | ||
4994 | void | |
4995 | xfs_bmap_del_extent_cow( | |
4996 | struct xfs_inode *ip, | |
4997 | xfs_extnum_t *idx, | |
4998 | struct xfs_bmbt_irec *got, | |
4999 | struct xfs_bmbt_irec *del) | |
5000 | { | |
5001 | struct xfs_mount *mp = ip->i_mount; | |
5002 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
5003 | struct xfs_bmbt_irec new; | |
5004 | xfs_fileoff_t del_endoff, got_endoff; | |
5005 | int state = BMAP_COWFORK; | |
5006 | ||
5007 | XFS_STATS_INC(mp, xs_del_exlist); | |
5008 | ||
5009 | del_endoff = del->br_startoff + del->br_blockcount; | |
5010 | got_endoff = got->br_startoff + got->br_blockcount; | |
5011 | ||
5012 | ASSERT(*idx >= 0); | |
5d829300 | 5013 | ASSERT(*idx <= xfs_iext_count(ifp)); |
fa5c836c CH |
5014 | ASSERT(del->br_blockcount > 0); |
5015 | ASSERT(got->br_startoff <= del->br_startoff); | |
5016 | ASSERT(got_endoff >= del_endoff); | |
5017 | ASSERT(!isnullstartblock(got->br_startblock)); | |
5018 | ||
5019 | if (got->br_startoff == del->br_startoff) | |
5020 | state |= BMAP_LEFT_CONTIG; | |
5021 | if (got_endoff == del_endoff) | |
5022 | state |= BMAP_RIGHT_CONTIG; | |
5023 | ||
5024 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
5025 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
5026 | /* | |
5027 | * Matches the whole extent. Delete the entry. | |
5028 | */ | |
5029 | xfs_iext_remove(ip, *idx, 1, state); | |
5030 | --*idx; | |
5031 | break; | |
5032 | case BMAP_LEFT_CONTIG: | |
5033 | /* | |
5034 | * Deleting the first part of the extent. | |
5035 | */ | |
5036 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
5037 | got->br_startoff = del_endoff; | |
5038 | got->br_blockcount -= del->br_blockcount; | |
5039 | got->br_startblock = del->br_startblock + del->br_blockcount; | |
5040 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
5041 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5042 | break; | |
5043 | case BMAP_RIGHT_CONTIG: | |
5044 | /* | |
5045 | * Deleting the last part of the extent. | |
5046 | */ | |
5047 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
5048 | got->br_blockcount -= del->br_blockcount; | |
5049 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
5050 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5051 | break; | |
5052 | case 0: | |
5053 | /* | |
5054 | * Deleting the middle of the extent. | |
5055 | */ | |
5056 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
5057 | got->br_blockcount = del->br_startoff - got->br_startoff; | |
5058 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
5059 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5060 | ||
5061 | new.br_startoff = del_endoff; | |
5062 | new.br_blockcount = got_endoff - del_endoff; | |
5063 | new.br_state = got->br_state; | |
5064 | new.br_startblock = del->br_startblock + del->br_blockcount; | |
5065 | ||
5066 | ++*idx; | |
5067 | xfs_iext_insert(ip, *idx, 1, &new, state); | |
5068 | break; | |
5069 | } | |
5070 | } | |
5071 | ||
9e5987a7 DC |
5072 | /* |
5073 | * Called by xfs_bmapi to update file extent records and the btree | |
5074 | * after removing space (or undoing a delayed allocation). | |
5075 | */ | |
5076 | STATIC int /* error */ | |
5077 | xfs_bmap_del_extent( | |
5078 | xfs_inode_t *ip, /* incore inode pointer */ | |
5079 | xfs_trans_t *tp, /* current transaction pointer */ | |
5080 | xfs_extnum_t *idx, /* extent number to update/delete */ | |
2c3234d1 | 5081 | struct xfs_defer_ops *dfops, /* list of extents to be freed */ |
9e5987a7 DC |
5082 | xfs_btree_cur_t *cur, /* if null, not a btree */ |
5083 | xfs_bmbt_irec_t *del, /* data to remove from extents */ | |
5084 | int *logflagsp, /* inode logging flags */ | |
4847acf8 DW |
5085 | int whichfork, /* data or attr fork */ |
5086 | int bflags) /* bmapi flags */ | |
9e5987a7 DC |
5087 | { |
5088 | xfs_filblks_t da_new; /* new delay-alloc indirect blocks */ | |
5089 | xfs_filblks_t da_old; /* old delay-alloc indirect blocks */ | |
5090 | xfs_fsblock_t del_endblock=0; /* first block past del */ | |
5091 | xfs_fileoff_t del_endoff; /* first offset past del */ | |
5092 | int delay; /* current block is delayed allocated */ | |
5093 | int do_fx; /* free extent at end of routine */ | |
5094 | xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */ | |
5095 | int error; /* error return value */ | |
5096 | int flags; /* inode logging flags */ | |
5097 | xfs_bmbt_irec_t got; /* current extent entry */ | |
5098 | xfs_fileoff_t got_endoff; /* first offset past got */ | |
5099 | int i; /* temp state */ | |
5100 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
5101 | xfs_mount_t *mp; /* mount structure */ | |
5102 | xfs_filblks_t nblks; /* quota/sb block count */ | |
5103 | xfs_bmbt_irec_t new; /* new record to be inserted */ | |
5104 | /* REFERENCED */ | |
5105 | uint qfield; /* quota field to update */ | |
5106 | xfs_filblks_t temp; /* for indirect length calculations */ | |
5107 | xfs_filblks_t temp2; /* for indirect length calculations */ | |
5108 | int state = 0; | |
5109 | ||
ff6d6af2 BD |
5110 | mp = ip->i_mount; |
5111 | XFS_STATS_INC(mp, xs_del_exlist); | |
9e5987a7 DC |
5112 | |
5113 | if (whichfork == XFS_ATTR_FORK) | |
5114 | state |= BMAP_ATTRFORK; | |
3993baeb DW |
5115 | else if (whichfork == XFS_COW_FORK) |
5116 | state |= BMAP_COWFORK; | |
9e5987a7 | 5117 | |
9e5987a7 | 5118 | ifp = XFS_IFORK_PTR(ip, whichfork); |
5d829300 | 5119 | ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp))); |
9e5987a7 DC |
5120 | ASSERT(del->br_blockcount > 0); |
5121 | ep = xfs_iext_get_ext(ifp, *idx); | |
5122 | xfs_bmbt_get_all(ep, &got); | |
5123 | ASSERT(got.br_startoff <= del->br_startoff); | |
5124 | del_endoff = del->br_startoff + del->br_blockcount; | |
5125 | got_endoff = got.br_startoff + got.br_blockcount; | |
5126 | ASSERT(got_endoff >= del_endoff); | |
5127 | delay = isnullstartblock(got.br_startblock); | |
5128 | ASSERT(isnullstartblock(del->br_startblock) == delay); | |
5129 | flags = 0; | |
5130 | qfield = 0; | |
5131 | error = 0; | |
5132 | /* | |
5133 | * If deleting a real allocation, must free up the disk space. | |
5134 | */ | |
5135 | if (!delay) { | |
5136 | flags = XFS_ILOG_CORE; | |
5137 | /* | |
5138 | * Realtime allocation. Free it and record di_nblocks update. | |
5139 | */ | |
5140 | if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) { | |
5141 | xfs_fsblock_t bno; | |
5142 | xfs_filblks_t len; | |
5143 | ||
5144 | ASSERT(do_mod(del->br_blockcount, | |
5145 | mp->m_sb.sb_rextsize) == 0); | |
5146 | ASSERT(do_mod(del->br_startblock, | |
5147 | mp->m_sb.sb_rextsize) == 0); | |
5148 | bno = del->br_startblock; | |
5149 | len = del->br_blockcount; | |
5150 | do_div(bno, mp->m_sb.sb_rextsize); | |
5151 | do_div(len, mp->m_sb.sb_rextsize); | |
5152 | error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len); | |
5153 | if (error) | |
5154 | goto done; | |
5155 | do_fx = 0; | |
5156 | nblks = len * mp->m_sb.sb_rextsize; | |
5157 | qfield = XFS_TRANS_DQ_RTBCOUNT; | |
5158 | } | |
1da177e4 | 5159 | /* |
9e5987a7 | 5160 | * Ordinary allocation. |
1da177e4 | 5161 | */ |
9e5987a7 DC |
5162 | else { |
5163 | do_fx = 1; | |
5164 | nblks = del->br_blockcount; | |
5165 | qfield = XFS_TRANS_DQ_BCOUNT; | |
1da177e4 | 5166 | } |
1da177e4 | 5167 | /* |
9e5987a7 | 5168 | * Set up del_endblock and cur for later. |
1da177e4 | 5169 | */ |
9e5987a7 DC |
5170 | del_endblock = del->br_startblock + del->br_blockcount; |
5171 | if (cur) { | |
5172 | if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff, | |
5173 | got.br_startblock, got.br_blockcount, | |
5174 | &i))) | |
5175 | goto done; | |
c29aad41 | 5176 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
1da177e4 | 5177 | } |
9e5987a7 DC |
5178 | da_old = da_new = 0; |
5179 | } else { | |
5180 | da_old = startblockval(got.br_startblock); | |
5181 | da_new = 0; | |
5182 | nblks = 0; | |
5183 | do_fx = 0; | |
1da177e4 | 5184 | } |
340785cc | 5185 | |
1da177e4 | 5186 | /* |
9e5987a7 DC |
5187 | * Set flag value to use in switch statement. |
5188 | * Left-contig is 2, right-contig is 1. | |
1da177e4 | 5189 | */ |
9e5987a7 DC |
5190 | switch (((got.br_startoff == del->br_startoff) << 1) | |
5191 | (got_endoff == del_endoff)) { | |
5192 | case 3: | |
5193 | /* | |
5194 | * Matches the whole extent. Delete the entry. | |
5195 | */ | |
4862cfe8 | 5196 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
9e5987a7 DC |
5197 | xfs_iext_remove(ip, *idx, 1, |
5198 | whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0); | |
5199 | --*idx; | |
5200 | if (delay) | |
5201 | break; | |
5202 | ||
5203 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5204 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5205 | flags |= XFS_ILOG_CORE; | |
5206 | if (!cur) { | |
5207 | flags |= xfs_ilog_fext(whichfork); | |
5208 | break; | |
5209 | } | |
5210 | if ((error = xfs_btree_delete(cur, &i))) | |
5211 | goto done; | |
c29aad41 | 5212 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
5213 | break; |
5214 | ||
5215 | case 2: | |
5216 | /* | |
5217 | * Deleting the first part of the extent. | |
5218 | */ | |
5219 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
5220 | xfs_bmbt_set_startoff(ep, del_endoff); | |
5221 | temp = got.br_blockcount - del->br_blockcount; | |
5222 | xfs_bmbt_set_blockcount(ep, temp); | |
5223 | if (delay) { | |
5224 | temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), | |
5225 | da_old); | |
5226 | xfs_bmbt_set_startblock(ep, nullstartblock((int)temp)); | |
5227 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5228 | da_new = temp; | |
5229 | break; | |
5230 | } | |
5231 | xfs_bmbt_set_startblock(ep, del_endblock); | |
5232 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5233 | if (!cur) { | |
5234 | flags |= xfs_ilog_fext(whichfork); | |
5235 | break; | |
5236 | } | |
5237 | if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock, | |
5238 | got.br_blockcount - del->br_blockcount, | |
5239 | got.br_state))) | |
5240 | goto done; | |
5241 | break; | |
5242 | ||
5243 | case 1: | |
5244 | /* | |
5245 | * Deleting the last part of the extent. | |
5246 | */ | |
5247 | temp = got.br_blockcount - del->br_blockcount; | |
5248 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
5249 | xfs_bmbt_set_blockcount(ep, temp); | |
5250 | if (delay) { | |
5251 | temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), | |
5252 | da_old); | |
5253 | xfs_bmbt_set_startblock(ep, nullstartblock((int)temp)); | |
5254 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5255 | da_new = temp; | |
5256 | break; | |
5257 | } | |
5258 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5259 | if (!cur) { | |
5260 | flags |= xfs_ilog_fext(whichfork); | |
5261 | break; | |
5262 | } | |
5263 | if ((error = xfs_bmbt_update(cur, got.br_startoff, | |
5264 | got.br_startblock, | |
5265 | got.br_blockcount - del->br_blockcount, | |
5266 | got.br_state))) | |
5267 | goto done; | |
5268 | break; | |
5269 | ||
5270 | case 0: | |
5271 | /* | |
5272 | * Deleting the middle of the extent. | |
5273 | */ | |
5274 | temp = del->br_startoff - got.br_startoff; | |
5275 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
5276 | xfs_bmbt_set_blockcount(ep, temp); | |
5277 | new.br_startoff = del_endoff; | |
5278 | temp2 = got_endoff - del_endoff; | |
5279 | new.br_blockcount = temp2; | |
5280 | new.br_state = got.br_state; | |
5281 | if (!delay) { | |
5282 | new.br_startblock = del_endblock; | |
5283 | flags |= XFS_ILOG_CORE; | |
5284 | if (cur) { | |
5285 | if ((error = xfs_bmbt_update(cur, | |
5286 | got.br_startoff, | |
5287 | got.br_startblock, temp, | |
5288 | got.br_state))) | |
5289 | goto done; | |
5290 | if ((error = xfs_btree_increment(cur, 0, &i))) | |
5291 | goto done; | |
5292 | cur->bc_rec.b = new; | |
5293 | error = xfs_btree_insert(cur, &i); | |
2451337d | 5294 | if (error && error != -ENOSPC) |
9e5987a7 DC |
5295 | goto done; |
5296 | /* | |
5297 | * If get no-space back from btree insert, | |
5298 | * it tried a split, and we have a zero | |
5299 | * block reservation. | |
5300 | * Fix up our state and return the error. | |
5301 | */ | |
2451337d | 5302 | if (error == -ENOSPC) { |
9e5987a7 DC |
5303 | /* |
5304 | * Reset the cursor, don't trust | |
5305 | * it after any insert operation. | |
5306 | */ | |
5307 | if ((error = xfs_bmbt_lookup_eq(cur, | |
5308 | got.br_startoff, | |
5309 | got.br_startblock, | |
5310 | temp, &i))) | |
5311 | goto done; | |
c29aad41 ES |
5312 | XFS_WANT_CORRUPTED_GOTO(mp, |
5313 | i == 1, done); | |
9e5987a7 DC |
5314 | /* |
5315 | * Update the btree record back | |
5316 | * to the original value. | |
5317 | */ | |
5318 | if ((error = xfs_bmbt_update(cur, | |
5319 | got.br_startoff, | |
5320 | got.br_startblock, | |
5321 | got.br_blockcount, | |
5322 | got.br_state))) | |
5323 | goto done; | |
5324 | /* | |
5325 | * Reset the extent record back | |
5326 | * to the original value. | |
5327 | */ | |
5328 | xfs_bmbt_set_blockcount(ep, | |
5329 | got.br_blockcount); | |
5330 | flags = 0; | |
2451337d | 5331 | error = -ENOSPC; |
9e5987a7 DC |
5332 | goto done; |
5333 | } | |
c29aad41 | 5334 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
5335 | } else |
5336 | flags |= xfs_ilog_fext(whichfork); | |
5337 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5338 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
5339 | } else { | |
d34999c9 | 5340 | xfs_filblks_t stolen; |
9e5987a7 | 5341 | ASSERT(whichfork == XFS_DATA_FORK); |
a9bd24ac BF |
5342 | |
5343 | /* | |
5344 | * Distribute the original indlen reservation across the | |
d34999c9 BF |
5345 | * two new extents. Steal blocks from the deleted extent |
5346 | * if necessary. Stealing blocks simply fudges the | |
5347 | * fdblocks accounting in xfs_bunmapi(). | |
a9bd24ac BF |
5348 | */ |
5349 | temp = xfs_bmap_worst_indlen(ip, got.br_blockcount); | |
5350 | temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
d34999c9 BF |
5351 | stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2, |
5352 | del->br_blockcount); | |
5353 | da_new = temp + temp2 - stolen; | |
5354 | del->br_blockcount -= stolen; | |
a9bd24ac BF |
5355 | |
5356 | /* | |
d34999c9 BF |
5357 | * Set the reservation for each extent. Warn if either |
5358 | * is zero as this can lead to delalloc problems. | |
a9bd24ac | 5359 | */ |
d34999c9 | 5360 | WARN_ON_ONCE(!temp || !temp2); |
9e5987a7 | 5361 | xfs_bmbt_set_startblock(ep, nullstartblock((int)temp)); |
9e5987a7 | 5362 | new.br_startblock = nullstartblock((int)temp2); |
9e5987a7 DC |
5363 | } |
5364 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5365 | xfs_iext_insert(ip, *idx + 1, 1, &new, state); | |
5366 | ++*idx; | |
5367 | break; | |
1da177e4 | 5368 | } |
9c194644 DW |
5369 | |
5370 | /* remove reverse mapping */ | |
5371 | if (!delay) { | |
5372 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del); | |
5373 | if (error) | |
5374 | goto done; | |
5375 | } | |
5376 | ||
1da177e4 | 5377 | /* |
9e5987a7 | 5378 | * If we need to, add to list of extents to delete. |
1da177e4 | 5379 | */ |
4847acf8 | 5380 | if (do_fx && !(bflags & XFS_BMAPI_REMAP)) { |
62aab20f DW |
5381 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) { |
5382 | error = xfs_refcount_decrease_extent(mp, dfops, del); | |
5383 | if (error) | |
5384 | goto done; | |
5385 | } else | |
5386 | xfs_bmap_add_free(mp, dfops, del->br_startblock, | |
5387 | del->br_blockcount, NULL); | |
5388 | } | |
5389 | ||
1da177e4 | 5390 | /* |
9e5987a7 | 5391 | * Adjust inode # blocks in the file. |
1da177e4 | 5392 | */ |
9e5987a7 DC |
5393 | if (nblks) |
5394 | ip->i_d.di_nblocks -= nblks; | |
1da177e4 | 5395 | /* |
9e5987a7 | 5396 | * Adjust quota data. |
1da177e4 | 5397 | */ |
4847acf8 | 5398 | if (qfield && !(bflags & XFS_BMAPI_REMAP)) |
9e5987a7 DC |
5399 | xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks); |
5400 | ||
5401 | /* | |
5402 | * Account for change in delayed indirect blocks. | |
5403 | * Nothing to do for disk quota accounting here. | |
5404 | */ | |
5405 | ASSERT(da_old >= da_new); | |
0d485ada DC |
5406 | if (da_old > da_new) |
5407 | xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false); | |
9e5987a7 DC |
5408 | done: |
5409 | *logflagsp = flags; | |
1da177e4 LT |
5410 | return error; |
5411 | } | |
5412 | ||
3bacbcd8 | 5413 | /* |
9e5987a7 DC |
5414 | * Unmap (remove) blocks from a file. |
5415 | * If nexts is nonzero then the number of extents to remove is limited to | |
5416 | * that value. If not all extents in the block range can be removed then | |
5417 | * *done is set. | |
3bacbcd8 | 5418 | */ |
9e5987a7 | 5419 | int /* error */ |
4453593b | 5420 | __xfs_bunmapi( |
9e5987a7 DC |
5421 | xfs_trans_t *tp, /* transaction pointer */ |
5422 | struct xfs_inode *ip, /* incore inode */ | |
5423 | xfs_fileoff_t bno, /* starting offset to unmap */ | |
4453593b | 5424 | xfs_filblks_t *rlen, /* i/o: amount remaining */ |
9e5987a7 DC |
5425 | int flags, /* misc flags */ |
5426 | xfs_extnum_t nexts, /* number of extents max */ | |
5427 | xfs_fsblock_t *firstblock, /* first allocated block | |
5428 | controls a.g. for allocs */ | |
4453593b | 5429 | struct xfs_defer_ops *dfops) /* i/o: deferred updates */ |
3bacbcd8 | 5430 | { |
9e5987a7 DC |
5431 | xfs_btree_cur_t *cur; /* bmap btree cursor */ |
5432 | xfs_bmbt_irec_t del; /* extent being deleted */ | |
9e5987a7 DC |
5433 | int error; /* error return value */ |
5434 | xfs_extnum_t extno; /* extent number in list */ | |
5435 | xfs_bmbt_irec_t got; /* current extent record */ | |
5af317c9 | 5436 | xfs_ifork_t *ifp; /* inode fork pointer */ |
9e5987a7 DC |
5437 | int isrt; /* freeing in rt area */ |
5438 | xfs_extnum_t lastx; /* last extent index used */ | |
5439 | int logflags; /* transaction logging flags */ | |
5440 | xfs_extlen_t mod; /* rt extent offset */ | |
5441 | xfs_mount_t *mp; /* mount structure */ | |
9e5987a7 DC |
5442 | xfs_fileoff_t start; /* first file offset deleted */ |
5443 | int tmp_logflags; /* partial logging flags */ | |
5444 | int wasdel; /* was a delayed alloc extent */ | |
5445 | int whichfork; /* data or attribute fork */ | |
5446 | xfs_fsblock_t sum; | |
4453593b | 5447 | xfs_filblks_t len = *rlen; /* length to unmap in file */ |
e1a4e37c | 5448 | xfs_fileoff_t max_len; |
5b094d6d | 5449 | xfs_agnumber_t prev_agno = NULLAGNUMBER, agno; |
1da177e4 | 5450 | |
9e5987a7 | 5451 | trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_); |
1da177e4 | 5452 | |
3993baeb DW |
5453 | whichfork = xfs_bmapi_whichfork(flags); |
5454 | ASSERT(whichfork != XFS_COW_FORK); | |
9e5987a7 DC |
5455 | ifp = XFS_IFORK_PTR(ip, whichfork); |
5456 | if (unlikely( | |
5457 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5458 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
5459 | XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW, | |
5460 | ip->i_mount); | |
2451337d | 5461 | return -EFSCORRUPTED; |
1da177e4 | 5462 | } |
9e5987a7 DC |
5463 | mp = ip->i_mount; |
5464 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 5465 | return -EIO; |
1da177e4 | 5466 | |
eef334e5 | 5467 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
9e5987a7 DC |
5468 | ASSERT(len > 0); |
5469 | ASSERT(nexts >= 0); | |
1da177e4 | 5470 | |
e1a4e37c DW |
5471 | /* |
5472 | * Guesstimate how many blocks we can unmap without running the risk of | |
5473 | * blowing out the transaction with a mix of EFIs and reflink | |
5474 | * adjustments. | |
5475 | */ | |
5476 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) | |
5477 | max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res)); | |
5478 | else | |
5479 | max_len = len; | |
5480 | ||
9e5987a7 DC |
5481 | if (!(ifp->if_flags & XFS_IFEXTENTS) && |
5482 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
5483 | return error; | |
5d829300 | 5484 | if (xfs_iext_count(ifp) == 0) { |
4453593b | 5485 | *rlen = 0; |
9e5987a7 DC |
5486 | return 0; |
5487 | } | |
ff6d6af2 | 5488 | XFS_STATS_INC(mp, xs_blk_unmap); |
9e5987a7 DC |
5489 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); |
5490 | start = bno; | |
5491 | bno = start + len - 1; | |
1da177e4 | 5492 | |
9e5987a7 DC |
5493 | /* |
5494 | * Check to see if the given block number is past the end of the | |
5495 | * file, back up to the last block if so... | |
5496 | */ | |
7efc7945 CH |
5497 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &lastx, &got)) { |
5498 | ASSERT(lastx > 0); | |
5499 | xfs_iext_get_extent(ifp, --lastx, &got); | |
9e5987a7 DC |
5500 | bno = got.br_startoff + got.br_blockcount - 1; |
5501 | } | |
7efc7945 | 5502 | |
9e5987a7 DC |
5503 | logflags = 0; |
5504 | if (ifp->if_flags & XFS_IFBROOT) { | |
5505 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
5506 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5507 | cur->bc_private.b.firstblock = *firstblock; | |
2c3234d1 | 5508 | cur->bc_private.b.dfops = dfops; |
9e5987a7 DC |
5509 | cur->bc_private.b.flags = 0; |
5510 | } else | |
5511 | cur = NULL; | |
5512 | ||
5513 | if (isrt) { | |
5514 | /* | |
5515 | * Synchronize by locking the bitmap inode. | |
5516 | */ | |
f4a0660d | 5517 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
9e5987a7 | 5518 | xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
f4a0660d DW |
5519 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
5520 | xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL); | |
9e5987a7 | 5521 | } |
58e20770 | 5522 | |
9e5987a7 DC |
5523 | extno = 0; |
5524 | while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 && | |
e1a4e37c | 5525 | (nexts == 0 || extno < nexts) && max_len > 0) { |
9e5987a7 DC |
5526 | /* |
5527 | * Is the found extent after a hole in which bno lives? | |
5528 | * Just back up to the previous extent, if so. | |
5529 | */ | |
5530 | if (got.br_startoff > bno) { | |
5531 | if (--lastx < 0) | |
5532 | break; | |
7efc7945 | 5533 | xfs_iext_get_extent(ifp, lastx, &got); |
9e5987a7 DC |
5534 | } |
5535 | /* | |
5536 | * Is the last block of this extent before the range | |
5537 | * we're supposed to delete? If so, we're done. | |
5538 | */ | |
5539 | bno = XFS_FILEOFF_MIN(bno, | |
5540 | got.br_startoff + got.br_blockcount - 1); | |
5541 | if (bno < start) | |
5542 | break; | |
5543 | /* | |
5544 | * Then deal with the (possibly delayed) allocated space | |
5545 | * we found. | |
5546 | */ | |
9e5987a7 DC |
5547 | del = got; |
5548 | wasdel = isnullstartblock(del.br_startblock); | |
5b094d6d CH |
5549 | |
5550 | /* | |
5551 | * Make sure we don't touch multiple AGF headers out of order | |
5552 | * in a single transaction, as that could cause AB-BA deadlocks. | |
5553 | */ | |
5554 | if (!wasdel) { | |
5555 | agno = XFS_FSB_TO_AGNO(mp, del.br_startblock); | |
5556 | if (prev_agno != NULLAGNUMBER && prev_agno > agno) | |
5557 | break; | |
5558 | prev_agno = agno; | |
5559 | } | |
9e5987a7 DC |
5560 | if (got.br_startoff < start) { |
5561 | del.br_startoff = start; | |
5562 | del.br_blockcount -= start - got.br_startoff; | |
5563 | if (!wasdel) | |
5564 | del.br_startblock += start - got.br_startoff; | |
5565 | } | |
5566 | if (del.br_startoff + del.br_blockcount > bno + 1) | |
5567 | del.br_blockcount = bno + 1 - del.br_startoff; | |
e1a4e37c DW |
5568 | |
5569 | /* How much can we safely unmap? */ | |
5570 | if (max_len < del.br_blockcount) { | |
5571 | del.br_startoff += del.br_blockcount - max_len; | |
5572 | if (!wasdel) | |
5573 | del.br_startblock += del.br_blockcount - max_len; | |
5574 | del.br_blockcount = max_len; | |
5575 | } | |
5576 | ||
9e5987a7 DC |
5577 | sum = del.br_startblock + del.br_blockcount; |
5578 | if (isrt && | |
5579 | (mod = do_mod(sum, mp->m_sb.sb_rextsize))) { | |
58e20770 | 5580 | /* |
9e5987a7 DC |
5581 | * Realtime extent not lined up at the end. |
5582 | * The extent could have been split into written | |
5583 | * and unwritten pieces, or we could just be | |
5584 | * unmapping part of it. But we can't really | |
5585 | * get rid of part of a realtime extent. | |
58e20770 | 5586 | */ |
9e5987a7 DC |
5587 | if (del.br_state == XFS_EXT_UNWRITTEN || |
5588 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { | |
5589 | /* | |
5590 | * This piece is unwritten, or we're not | |
5591 | * using unwritten extents. Skip over it. | |
5592 | */ | |
5593 | ASSERT(bno >= mod); | |
5594 | bno -= mod > del.br_blockcount ? | |
5595 | del.br_blockcount : mod; | |
5596 | if (bno < got.br_startoff) { | |
5597 | if (--lastx >= 0) | |
5598 | xfs_bmbt_get_all(xfs_iext_get_ext( | |
5599 | ifp, lastx), &got); | |
5600 | } | |
5601 | continue; | |
1da177e4 | 5602 | } |
9af25465 | 5603 | /* |
9e5987a7 DC |
5604 | * It's written, turn it unwritten. |
5605 | * This is better than zeroing it. | |
9af25465 | 5606 | */ |
9e5987a7 | 5607 | ASSERT(del.br_state == XFS_EXT_NORM); |
a7e5d03b | 5608 | ASSERT(tp->t_blk_res > 0); |
9e5987a7 DC |
5609 | /* |
5610 | * If this spans a realtime extent boundary, | |
5611 | * chop it back to the start of the one we end at. | |
5612 | */ | |
5613 | if (del.br_blockcount > mod) { | |
5614 | del.br_startoff += del.br_blockcount - mod; | |
5615 | del.br_startblock += del.br_blockcount - mod; | |
5616 | del.br_blockcount = mod; | |
5617 | } | |
5618 | del.br_state = XFS_EXT_UNWRITTEN; | |
5619 | error = xfs_bmap_add_extent_unwritten_real(tp, ip, | |
05a630d7 DW |
5620 | whichfork, &lastx, &cur, &del, |
5621 | firstblock, dfops, &logflags); | |
9e5987a7 DC |
5622 | if (error) |
5623 | goto error0; | |
5624 | goto nodelete; | |
5625 | } | |
5626 | if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) { | |
5627 | /* | |
5628 | * Realtime extent is lined up at the end but not | |
5629 | * at the front. We'll get rid of full extents if | |
5630 | * we can. | |
5631 | */ | |
5632 | mod = mp->m_sb.sb_rextsize - mod; | |
5633 | if (del.br_blockcount > mod) { | |
5634 | del.br_blockcount -= mod; | |
5635 | del.br_startoff += mod; | |
5636 | del.br_startblock += mod; | |
5637 | } else if ((del.br_startoff == start && | |
5638 | (del.br_state == XFS_EXT_UNWRITTEN || | |
a7e5d03b | 5639 | tp->t_blk_res == 0)) || |
9e5987a7 DC |
5640 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
5641 | /* | |
5642 | * Can't make it unwritten. There isn't | |
5643 | * a full extent here so just skip it. | |
5644 | */ | |
5645 | ASSERT(bno >= del.br_blockcount); | |
5646 | bno -= del.br_blockcount; | |
7efc7945 CH |
5647 | if (got.br_startoff > bno && --lastx >= 0) |
5648 | xfs_iext_get_extent(ifp, lastx, &got); | |
9af25465 | 5649 | continue; |
9e5987a7 | 5650 | } else if (del.br_state == XFS_EXT_UNWRITTEN) { |
7efc7945 CH |
5651 | struct xfs_bmbt_irec prev; |
5652 | ||
9e5987a7 DC |
5653 | /* |
5654 | * This one is already unwritten. | |
5655 | * It must have a written left neighbor. | |
5656 | * Unwrite the killed part of that one and | |
5657 | * try again. | |
5658 | */ | |
5659 | ASSERT(lastx > 0); | |
7efc7945 | 5660 | xfs_iext_get_extent(ifp, lastx - 1, &prev); |
9e5987a7 DC |
5661 | ASSERT(prev.br_state == XFS_EXT_NORM); |
5662 | ASSERT(!isnullstartblock(prev.br_startblock)); | |
5663 | ASSERT(del.br_startblock == | |
5664 | prev.br_startblock + prev.br_blockcount); | |
5665 | if (prev.br_startoff < start) { | |
5666 | mod = start - prev.br_startoff; | |
5667 | prev.br_blockcount -= mod; | |
5668 | prev.br_startblock += mod; | |
5669 | prev.br_startoff = start; | |
5670 | } | |
5671 | prev.br_state = XFS_EXT_UNWRITTEN; | |
5672 | lastx--; | |
5673 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
05a630d7 DW |
5674 | ip, whichfork, &lastx, &cur, |
5675 | &prev, firstblock, dfops, | |
5676 | &logflags); | |
9e5987a7 DC |
5677 | if (error) |
5678 | goto error0; | |
5679 | goto nodelete; | |
5680 | } else { | |
5681 | ASSERT(del.br_state == XFS_EXT_NORM); | |
5682 | del.br_state = XFS_EXT_UNWRITTEN; | |
5683 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
05a630d7 DW |
5684 | ip, whichfork, &lastx, &cur, |
5685 | &del, firstblock, dfops, | |
5686 | &logflags); | |
9e5987a7 DC |
5687 | if (error) |
5688 | goto error0; | |
5689 | goto nodelete; | |
9af25465 | 5690 | } |
1da177e4 | 5691 | } |
1da177e4 | 5692 | |
9e5987a7 DC |
5693 | /* |
5694 | * If it's the case where the directory code is running | |
5695 | * with no block reservation, and the deleted block is in | |
5696 | * the middle of its extent, and the resulting insert | |
5697 | * of an extent would cause transformation to btree format, | |
5698 | * then reject it. The calling code will then swap | |
5699 | * blocks around instead. | |
5700 | * We have to do this now, rather than waiting for the | |
5701 | * conversion to btree format, since the transaction | |
5702 | * will be dirty. | |
5703 | */ | |
a7e5d03b | 5704 | if (!wasdel && tp->t_blk_res == 0 && |
9e5987a7 DC |
5705 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && |
5706 | XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */ | |
5707 | XFS_IFORK_MAXEXT(ip, whichfork) && | |
5708 | del.br_startoff > got.br_startoff && | |
5709 | del.br_startoff + del.br_blockcount < | |
5710 | got.br_startoff + got.br_blockcount) { | |
2451337d | 5711 | error = -ENOSPC; |
9e5987a7 | 5712 | goto error0; |
1da177e4 | 5713 | } |
b2706a05 BF |
5714 | |
5715 | /* | |
5716 | * Unreserve quota and update realtime free space, if | |
5717 | * appropriate. If delayed allocation, update the inode delalloc | |
5718 | * counter now and wait to update the sb counters as | |
5719 | * xfs_bmap_del_extent() might need to borrow some blocks. | |
5720 | */ | |
5721 | if (wasdel) { | |
5722 | ASSERT(startblockval(del.br_startblock) > 0); | |
5723 | if (isrt) { | |
5724 | xfs_filblks_t rtexts; | |
5725 | ||
5726 | rtexts = XFS_FSB_TO_B(mp, del.br_blockcount); | |
5727 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
5728 | xfs_mod_frextents(mp, (int64_t)rtexts); | |
5729 | (void)xfs_trans_reserve_quota_nblks(NULL, | |
5730 | ip, -((long)del.br_blockcount), 0, | |
5731 | XFS_QMOPT_RES_RTBLKS); | |
5732 | } else { | |
5733 | (void)xfs_trans_reserve_quota_nblks(NULL, | |
5734 | ip, -((long)del.br_blockcount), 0, | |
5735 | XFS_QMOPT_RES_REGBLKS); | |
5736 | } | |
5737 | ip->i_delayed_blks -= del.br_blockcount; | |
5738 | if (cur) | |
5739 | cur->bc_private.b.flags |= | |
5740 | XFS_BTCUR_BPRV_WASDEL; | |
5741 | } else if (cur) | |
5742 | cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL; | |
5743 | ||
2c3234d1 | 5744 | error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del, |
4847acf8 | 5745 | &tmp_logflags, whichfork, flags); |
9e5987a7 DC |
5746 | logflags |= tmp_logflags; |
5747 | if (error) | |
5748 | goto error0; | |
b2706a05 BF |
5749 | |
5750 | if (!isrt && wasdel) | |
5751 | xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false); | |
5752 | ||
e1a4e37c | 5753 | max_len -= del.br_blockcount; |
9e5987a7 DC |
5754 | bno = del.br_startoff - 1; |
5755 | nodelete: | |
1da177e4 | 5756 | /* |
9e5987a7 | 5757 | * If not done go on to the next (previous) record. |
1da177e4 | 5758 | */ |
9e5987a7 DC |
5759 | if (bno != (xfs_fileoff_t)-1 && bno >= start) { |
5760 | if (lastx >= 0) { | |
7efc7945 CH |
5761 | xfs_iext_get_extent(ifp, lastx, &got); |
5762 | if (got.br_startoff > bno && --lastx >= 0) | |
5763 | xfs_iext_get_extent(ifp, lastx, &got); | |
1da177e4 | 5764 | } |
9e5987a7 | 5765 | extno++; |
1da177e4 LT |
5766 | } |
5767 | } | |
4453593b DW |
5768 | if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0) |
5769 | *rlen = 0; | |
5770 | else | |
5771 | *rlen = bno - start + 1; | |
576039cf | 5772 | |
1da177e4 | 5773 | /* |
9e5987a7 | 5774 | * Convert to a btree if necessary. |
1da177e4 | 5775 | */ |
9e5987a7 DC |
5776 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
5777 | ASSERT(cur == NULL); | |
2c3234d1 | 5778 | error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, |
9e5987a7 DC |
5779 | &cur, 0, &tmp_logflags, whichfork); |
5780 | logflags |= tmp_logflags; | |
5781 | if (error) | |
5782 | goto error0; | |
1da177e4 | 5783 | } |
1da177e4 | 5784 | /* |
9e5987a7 | 5785 | * transform from btree to extents, give it cur |
1da177e4 | 5786 | */ |
9e5987a7 DC |
5787 | else if (xfs_bmap_wants_extents(ip, whichfork)) { |
5788 | ASSERT(cur != NULL); | |
5789 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags, | |
5790 | whichfork); | |
5791 | logflags |= tmp_logflags; | |
5792 | if (error) | |
5793 | goto error0; | |
5794 | } | |
1da177e4 | 5795 | /* |
9e5987a7 | 5796 | * transform from extents to local? |
1da177e4 | 5797 | */ |
9e5987a7 DC |
5798 | error = 0; |
5799 | error0: | |
5800 | /* | |
5801 | * Log everything. Do this after conversion, there's no point in | |
5802 | * logging the extent records if we've converted to btree format. | |
5803 | */ | |
5804 | if ((logflags & xfs_ilog_fext(whichfork)) && | |
5805 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
5806 | logflags &= ~xfs_ilog_fext(whichfork); | |
5807 | else if ((logflags & xfs_ilog_fbroot(whichfork)) && | |
5808 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) | |
5809 | logflags &= ~xfs_ilog_fbroot(whichfork); | |
5810 | /* | |
5811 | * Log inode even in the error case, if the transaction | |
5812 | * is dirty we'll need to shut down the filesystem. | |
5813 | */ | |
5814 | if (logflags) | |
5815 | xfs_trans_log_inode(tp, ip, logflags); | |
5816 | if (cur) { | |
5817 | if (!error) { | |
5818 | *firstblock = cur->bc_private.b.firstblock; | |
5819 | cur->bc_private.b.allocated = 0; | |
5820 | } | |
5821 | xfs_btree_del_cursor(cur, | |
5822 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
5823 | } | |
5824 | return error; | |
5825 | } | |
e1d8fb88 | 5826 | |
4453593b DW |
5827 | /* Unmap a range of a file. */ |
5828 | int | |
5829 | xfs_bunmapi( | |
5830 | xfs_trans_t *tp, | |
5831 | struct xfs_inode *ip, | |
5832 | xfs_fileoff_t bno, | |
5833 | xfs_filblks_t len, | |
5834 | int flags, | |
5835 | xfs_extnum_t nexts, | |
5836 | xfs_fsblock_t *firstblock, | |
5837 | struct xfs_defer_ops *dfops, | |
5838 | int *done) | |
5839 | { | |
5840 | int error; | |
5841 | ||
5842 | error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock, | |
5843 | dfops); | |
5844 | *done = (len == 0); | |
5845 | return error; | |
5846 | } | |
5847 | ||
ddb19e31 BF |
5848 | /* |
5849 | * Determine whether an extent shift can be accomplished by a merge with the | |
5850 | * extent that precedes the target hole of the shift. | |
5851 | */ | |
5852 | STATIC bool | |
5853 | xfs_bmse_can_merge( | |
5854 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5855 | struct xfs_bmbt_irec *got, /* current extent to shift */ | |
5856 | xfs_fileoff_t shift) /* shift fsb */ | |
5857 | { | |
5858 | xfs_fileoff_t startoff; | |
5859 | ||
5860 | startoff = got->br_startoff - shift; | |
5861 | ||
5862 | /* | |
5863 | * The extent, once shifted, must be adjacent in-file and on-disk with | |
5864 | * the preceding extent. | |
5865 | */ | |
5866 | if ((left->br_startoff + left->br_blockcount != startoff) || | |
5867 | (left->br_startblock + left->br_blockcount != got->br_startblock) || | |
5868 | (left->br_state != got->br_state) || | |
5869 | (left->br_blockcount + got->br_blockcount > MAXEXTLEN)) | |
5870 | return false; | |
5871 | ||
5872 | return true; | |
5873 | } | |
5874 | ||
5875 | /* | |
5876 | * A bmap extent shift adjusts the file offset of an extent to fill a preceding | |
5877 | * hole in the file. If an extent shift would result in the extent being fully | |
5878 | * adjacent to the extent that currently precedes the hole, we can merge with | |
5879 | * the preceding extent rather than do the shift. | |
5880 | * | |
5881 | * This function assumes the caller has verified a shift-by-merge is possible | |
5882 | * with the provided extents via xfs_bmse_can_merge(). | |
5883 | */ | |
5884 | STATIC int | |
5885 | xfs_bmse_merge( | |
5886 | struct xfs_inode *ip, | |
5887 | int whichfork, | |
5888 | xfs_fileoff_t shift, /* shift fsb */ | |
5889 | int current_ext, /* idx of gotp */ | |
5890 | struct xfs_bmbt_rec_host *gotp, /* extent to shift */ | |
5891 | struct xfs_bmbt_rec_host *leftp, /* preceding extent */ | |
5892 | struct xfs_btree_cur *cur, | |
5893 | int *logflags) /* output */ | |
5894 | { | |
ddb19e31 BF |
5895 | struct xfs_bmbt_irec got; |
5896 | struct xfs_bmbt_irec left; | |
5897 | xfs_filblks_t blockcount; | |
5898 | int error, i; | |
5fb5aeee | 5899 | struct xfs_mount *mp = ip->i_mount; |
ddb19e31 | 5900 | |
ddb19e31 BF |
5901 | xfs_bmbt_get_all(gotp, &got); |
5902 | xfs_bmbt_get_all(leftp, &left); | |
5903 | blockcount = left.br_blockcount + got.br_blockcount; | |
5904 | ||
5905 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5906 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
5907 | ASSERT(xfs_bmse_can_merge(&left, &got, shift)); | |
5908 | ||
5909 | /* | |
5910 | * Merge the in-core extents. Note that the host record pointers and | |
5911 | * current_ext index are invalid once the extent has been removed via | |
5912 | * xfs_iext_remove(). | |
5913 | */ | |
5914 | xfs_bmbt_set_blockcount(leftp, blockcount); | |
5915 | xfs_iext_remove(ip, current_ext, 1, 0); | |
5916 | ||
5917 | /* | |
5918 | * Update the on-disk extent count, the btree if necessary and log the | |
5919 | * inode. | |
5920 | */ | |
5921 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5922 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5923 | *logflags |= XFS_ILOG_CORE; | |
5924 | if (!cur) { | |
5925 | *logflags |= XFS_ILOG_DEXT; | |
5926 | return 0; | |
5927 | } | |
5928 | ||
5929 | /* lookup and remove the extent to merge */ | |
5930 | error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock, | |
5931 | got.br_blockcount, &i); | |
5932 | if (error) | |
4db431f5 | 5933 | return error; |
5fb5aeee | 5934 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
ddb19e31 BF |
5935 | |
5936 | error = xfs_btree_delete(cur, &i); | |
5937 | if (error) | |
4db431f5 | 5938 | return error; |
5fb5aeee | 5939 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
ddb19e31 BF |
5940 | |
5941 | /* lookup and update size of the previous extent */ | |
5942 | error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock, | |
5943 | left.br_blockcount, &i); | |
5944 | if (error) | |
4db431f5 | 5945 | return error; |
5fb5aeee | 5946 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
ddb19e31 BF |
5947 | |
5948 | left.br_blockcount = blockcount; | |
5949 | ||
4db431f5 DC |
5950 | return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock, |
5951 | left.br_blockcount, left.br_state); | |
ddb19e31 BF |
5952 | } |
5953 | ||
a979bdfe BF |
5954 | /* |
5955 | * Shift a single extent. | |
5956 | */ | |
5957 | STATIC int | |
5958 | xfs_bmse_shift_one( | |
5959 | struct xfs_inode *ip, | |
5960 | int whichfork, | |
5961 | xfs_fileoff_t offset_shift_fsb, | |
5962 | int *current_ext, | |
5963 | struct xfs_bmbt_rec_host *gotp, | |
5964 | struct xfs_btree_cur *cur, | |
a904b1ca | 5965 | int *logflags, |
9c194644 DW |
5966 | enum shift_direction direction, |
5967 | struct xfs_defer_ops *dfops) | |
a979bdfe BF |
5968 | { |
5969 | struct xfs_ifork *ifp; | |
5fb5aeee | 5970 | struct xfs_mount *mp; |
a979bdfe | 5971 | xfs_fileoff_t startoff; |
a904b1ca | 5972 | struct xfs_bmbt_rec_host *adj_irecp; |
a979bdfe | 5973 | struct xfs_bmbt_irec got; |
a904b1ca | 5974 | struct xfs_bmbt_irec adj_irec; |
a979bdfe BF |
5975 | int error; |
5976 | int i; | |
a904b1ca | 5977 | int total_extents; |
a979bdfe | 5978 | |
5fb5aeee | 5979 | mp = ip->i_mount; |
a979bdfe | 5980 | ifp = XFS_IFORK_PTR(ip, whichfork); |
5d829300 | 5981 | total_extents = xfs_iext_count(ifp); |
a979bdfe BF |
5982 | |
5983 | xfs_bmbt_get_all(gotp, &got); | |
a979bdfe | 5984 | |
f71721d0 | 5985 | /* delalloc extents should be prevented by caller */ |
5fb5aeee | 5986 | XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock)); |
f71721d0 | 5987 | |
a904b1ca NJ |
5988 | if (direction == SHIFT_LEFT) { |
5989 | startoff = got.br_startoff - offset_shift_fsb; | |
5990 | ||
5991 | /* | |
5992 | * Check for merge if we've got an extent to the left, | |
5993 | * otherwise make sure there's enough room at the start | |
5994 | * of the file for the shift. | |
5995 | */ | |
5996 | if (!*current_ext) { | |
5997 | if (got.br_startoff < offset_shift_fsb) | |
5998 | return -EINVAL; | |
5999 | goto update_current_ext; | |
6000 | } | |
6001 | /* | |
6002 | * grab the left extent and check for a large | |
6003 | * enough hole. | |
6004 | */ | |
6005 | adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1); | |
6006 | xfs_bmbt_get_all(adj_irecp, &adj_irec); | |
a979bdfe | 6007 | |
a904b1ca NJ |
6008 | if (startoff < |
6009 | adj_irec.br_startoff + adj_irec.br_blockcount) | |
b11bd671 | 6010 | return -EINVAL; |
a979bdfe | 6011 | |
b11bd671 | 6012 | /* check whether to merge the extent or shift it down */ |
a904b1ca NJ |
6013 | if (xfs_bmse_can_merge(&adj_irec, &got, |
6014 | offset_shift_fsb)) { | |
9c194644 DW |
6015 | error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb, |
6016 | *current_ext, gotp, adj_irecp, | |
6017 | cur, logflags); | |
6018 | if (error) | |
6019 | return error; | |
6020 | adj_irec = got; | |
6021 | goto update_rmap; | |
b11bd671 | 6022 | } |
a904b1ca NJ |
6023 | } else { |
6024 | startoff = got.br_startoff + offset_shift_fsb; | |
6025 | /* nothing to move if this is the last extent */ | |
6026 | if (*current_ext >= (total_extents - 1)) | |
6027 | goto update_current_ext; | |
6028 | /* | |
6029 | * If this is not the last extent in the file, make sure there | |
6030 | * is enough room between current extent and next extent for | |
6031 | * accommodating the shift. | |
6032 | */ | |
6033 | adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1); | |
6034 | xfs_bmbt_get_all(adj_irecp, &adj_irec); | |
6035 | if (startoff + got.br_blockcount > adj_irec.br_startoff) | |
6036 | return -EINVAL; | |
6037 | /* | |
6038 | * Unlike a left shift (which involves a hole punch), | |
6039 | * a right shift does not modify extent neighbors | |
6040 | * in any way. We should never find mergeable extents | |
6041 | * in this scenario. Check anyways and warn if we | |
6042 | * encounter two extents that could be one. | |
6043 | */ | |
6044 | if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb)) | |
6045 | WARN_ON_ONCE(1); | |
6046 | } | |
a979bdfe BF |
6047 | /* |
6048 | * Increment the extent index for the next iteration, update the start | |
6049 | * offset of the in-core extent and update the btree if applicable. | |
6050 | */ | |
a904b1ca NJ |
6051 | update_current_ext: |
6052 | if (direction == SHIFT_LEFT) | |
6053 | (*current_ext)++; | |
6054 | else | |
6055 | (*current_ext)--; | |
a979bdfe BF |
6056 | xfs_bmbt_set_startoff(gotp, startoff); |
6057 | *logflags |= XFS_ILOG_CORE; | |
9c194644 | 6058 | adj_irec = got; |
a979bdfe BF |
6059 | if (!cur) { |
6060 | *logflags |= XFS_ILOG_DEXT; | |
9c194644 | 6061 | goto update_rmap; |
a979bdfe BF |
6062 | } |
6063 | ||
6064 | error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock, | |
6065 | got.br_blockcount, &i); | |
6066 | if (error) | |
6067 | return error; | |
5fb5aeee | 6068 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
a979bdfe BF |
6069 | |
6070 | got.br_startoff = startoff; | |
9c194644 DW |
6071 | error = xfs_bmbt_update(cur, got.br_startoff, got.br_startblock, |
6072 | got.br_blockcount, got.br_state); | |
6073 | if (error) | |
6074 | return error; | |
6075 | ||
6076 | update_rmap: | |
6077 | /* update reverse mapping */ | |
6078 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &adj_irec); | |
6079 | if (error) | |
6080 | return error; | |
6081 | adj_irec.br_startoff = startoff; | |
6082 | return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &adj_irec); | |
a979bdfe BF |
6083 | } |
6084 | ||
e1d8fb88 | 6085 | /* |
a904b1ca | 6086 | * Shift extent records to the left/right to cover/create a hole. |
e1d8fb88 | 6087 | * |
2c845f5a | 6088 | * The maximum number of extents to be shifted in a single operation is |
a904b1ca | 6089 | * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the |
2c845f5a BF |
6090 | * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb |
6091 | * is the length by which each extent is shifted. If there is no hole to shift | |
6092 | * the extents into, this will be considered invalid operation and we abort | |
6093 | * immediately. | |
e1d8fb88 NJ |
6094 | */ |
6095 | int | |
6096 | xfs_bmap_shift_extents( | |
6097 | struct xfs_trans *tp, | |
6098 | struct xfs_inode *ip, | |
a904b1ca | 6099 | xfs_fileoff_t *next_fsb, |
e1d8fb88 | 6100 | xfs_fileoff_t offset_shift_fsb, |
2c845f5a | 6101 | int *done, |
a904b1ca | 6102 | xfs_fileoff_t stop_fsb, |
e1d8fb88 | 6103 | xfs_fsblock_t *firstblock, |
2c3234d1 | 6104 | struct xfs_defer_ops *dfops, |
a904b1ca | 6105 | enum shift_direction direction, |
e1d8fb88 NJ |
6106 | int num_exts) |
6107 | { | |
ca446d88 | 6108 | struct xfs_btree_cur *cur = NULL; |
e1d8fb88 NJ |
6109 | struct xfs_bmbt_rec_host *gotp; |
6110 | struct xfs_bmbt_irec got; | |
e1d8fb88 NJ |
6111 | struct xfs_mount *mp = ip->i_mount; |
6112 | struct xfs_ifork *ifp; | |
6113 | xfs_extnum_t nexts = 0; | |
2c845f5a | 6114 | xfs_extnum_t current_ext; |
a904b1ca NJ |
6115 | xfs_extnum_t total_extents; |
6116 | xfs_extnum_t stop_extent; | |
e1d8fb88 | 6117 | int error = 0; |
e1d8fb88 | 6118 | int whichfork = XFS_DATA_FORK; |
ca446d88 | 6119 | int logflags = 0; |
e1d8fb88 NJ |
6120 | |
6121 | if (unlikely(XFS_TEST_ERROR( | |
6122 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
6123 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 6124 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
e1d8fb88 NJ |
6125 | XFS_ERROR_REPORT("xfs_bmap_shift_extents", |
6126 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 6127 | return -EFSCORRUPTED; |
e1d8fb88 NJ |
6128 | } |
6129 | ||
6130 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 6131 | return -EIO; |
e1d8fb88 | 6132 | |
2c845f5a BF |
6133 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
6134 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
a904b1ca NJ |
6135 | ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT); |
6136 | ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT); | |
e1d8fb88 NJ |
6137 | |
6138 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
e1d8fb88 NJ |
6139 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
6140 | /* Read in all the extents */ | |
6141 | error = xfs_iread_extents(tp, ip, whichfork); | |
6142 | if (error) | |
6143 | return error; | |
6144 | } | |
6145 | ||
ddb19e31 BF |
6146 | if (ifp->if_flags & XFS_IFBROOT) { |
6147 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
6148 | cur->bc_private.b.firstblock = *firstblock; | |
2c3234d1 | 6149 | cur->bc_private.b.dfops = dfops; |
ddb19e31 BF |
6150 | cur->bc_private.b.flags = 0; |
6151 | } | |
6152 | ||
a904b1ca NJ |
6153 | /* |
6154 | * There may be delalloc extents in the data fork before the range we | |
6155 | * are collapsing out, so we cannot use the count of real extents here. | |
6156 | * Instead we have to calculate it from the incore fork. | |
6157 | */ | |
5d829300 | 6158 | total_extents = xfs_iext_count(ifp); |
a904b1ca NJ |
6159 | if (total_extents == 0) { |
6160 | *done = 1; | |
6161 | goto del_cursor; | |
6162 | } | |
6163 | ||
6164 | /* | |
6165 | * In case of first right shift, we need to initialize next_fsb | |
6166 | */ | |
6167 | if (*next_fsb == NULLFSBLOCK) { | |
6168 | gotp = xfs_iext_get_ext(ifp, total_extents - 1); | |
6169 | xfs_bmbt_get_all(gotp, &got); | |
6170 | *next_fsb = got.br_startoff; | |
6171 | if (stop_fsb > *next_fsb) { | |
6172 | *done = 1; | |
6173 | goto del_cursor; | |
6174 | } | |
6175 | } | |
6176 | ||
6177 | /* Lookup the extent index at which we have to stop */ | |
6178 | if (direction == SHIFT_RIGHT) { | |
6179 | gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent); | |
6180 | /* Make stop_extent exclusive of shift range */ | |
6181 | stop_extent--; | |
6182 | } else | |
6183 | stop_extent = total_extents; | |
6184 | ||
e1d8fb88 | 6185 | /* |
2c845f5a BF |
6186 | * Look up the extent index for the fsb where we start shifting. We can |
6187 | * henceforth iterate with current_ext as extent list changes are locked | |
6188 | * out via ilock. | |
6189 | * | |
6190 | * gotp can be null in 2 cases: 1) if there are no extents or 2) | |
a904b1ca | 6191 | * *next_fsb lies in a hole beyond which there are no extents. Either |
2c845f5a | 6192 | * way, we are done. |
e1d8fb88 | 6193 | */ |
a904b1ca | 6194 | gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, ¤t_ext); |
2c845f5a BF |
6195 | if (!gotp) { |
6196 | *done = 1; | |
ddb19e31 | 6197 | goto del_cursor; |
e1d8fb88 | 6198 | } |
e1d8fb88 | 6199 | |
a904b1ca NJ |
6200 | /* some sanity checking before we finally start shifting extents */ |
6201 | if ((direction == SHIFT_LEFT && current_ext >= stop_extent) || | |
6202 | (direction == SHIFT_RIGHT && current_ext <= stop_extent)) { | |
6203 | error = -EIO; | |
6204 | goto del_cursor; | |
6205 | } | |
6206 | ||
6207 | while (nexts++ < num_exts) { | |
a979bdfe | 6208 | error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb, |
a904b1ca | 6209 | ¤t_ext, gotp, cur, &logflags, |
9c194644 | 6210 | direction, dfops); |
a979bdfe BF |
6211 | if (error) |
6212 | goto del_cursor; | |
a904b1ca NJ |
6213 | /* |
6214 | * If there was an extent merge during the shift, the extent | |
6215 | * count can change. Update the total and grade the next record. | |
6216 | */ | |
6217 | if (direction == SHIFT_LEFT) { | |
5d829300 | 6218 | total_extents = xfs_iext_count(ifp); |
a904b1ca NJ |
6219 | stop_extent = total_extents; |
6220 | } | |
e1d8fb88 | 6221 | |
a904b1ca NJ |
6222 | if (current_ext == stop_extent) { |
6223 | *done = 1; | |
6224 | *next_fsb = NULLFSBLOCK; | |
ddb19e31 | 6225 | break; |
a904b1ca | 6226 | } |
ddb19e31 | 6227 | gotp = xfs_iext_get_ext(ifp, current_ext); |
e1d8fb88 NJ |
6228 | } |
6229 | ||
a904b1ca | 6230 | if (!*done) { |
a979bdfe | 6231 | xfs_bmbt_get_all(gotp, &got); |
2c845f5a | 6232 | *next_fsb = got.br_startoff; |
a979bdfe | 6233 | } |
e1d8fb88 NJ |
6234 | |
6235 | del_cursor: | |
6236 | if (cur) | |
6237 | xfs_btree_del_cursor(cur, | |
6238 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
6239 | ||
ca446d88 BF |
6240 | if (logflags) |
6241 | xfs_trans_log_inode(tp, ip, logflags); | |
2c845f5a | 6242 | |
e1d8fb88 NJ |
6243 | return error; |
6244 | } | |
a904b1ca NJ |
6245 | |
6246 | /* | |
6247 | * Splits an extent into two extents at split_fsb block such that it is | |
6248 | * the first block of the current_ext. @current_ext is a target extent | |
6249 | * to be split. @split_fsb is a block where the extents is split. | |
6250 | * If split_fsb lies in a hole or the first block of extents, just return 0. | |
6251 | */ | |
6252 | STATIC int | |
6253 | xfs_bmap_split_extent_at( | |
6254 | struct xfs_trans *tp, | |
6255 | struct xfs_inode *ip, | |
6256 | xfs_fileoff_t split_fsb, | |
6257 | xfs_fsblock_t *firstfsb, | |
2c3234d1 | 6258 | struct xfs_defer_ops *dfops) |
a904b1ca NJ |
6259 | { |
6260 | int whichfork = XFS_DATA_FORK; | |
6261 | struct xfs_btree_cur *cur = NULL; | |
6262 | struct xfs_bmbt_rec_host *gotp; | |
6263 | struct xfs_bmbt_irec got; | |
6264 | struct xfs_bmbt_irec new; /* split extent */ | |
6265 | struct xfs_mount *mp = ip->i_mount; | |
6266 | struct xfs_ifork *ifp; | |
6267 | xfs_fsblock_t gotblkcnt; /* new block count for got */ | |
6268 | xfs_extnum_t current_ext; | |
6269 | int error = 0; | |
6270 | int logflags = 0; | |
6271 | int i = 0; | |
6272 | ||
6273 | if (unlikely(XFS_TEST_ERROR( | |
6274 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
6275 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 6276 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a904b1ca NJ |
6277 | XFS_ERROR_REPORT("xfs_bmap_split_extent_at", |
6278 | XFS_ERRLEVEL_LOW, mp); | |
6279 | return -EFSCORRUPTED; | |
6280 | } | |
6281 | ||
6282 | if (XFS_FORCED_SHUTDOWN(mp)) | |
6283 | return -EIO; | |
6284 | ||
6285 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
6286 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
6287 | /* Read in all the extents */ | |
6288 | error = xfs_iread_extents(tp, ip, whichfork); | |
6289 | if (error) | |
6290 | return error; | |
6291 | } | |
6292 | ||
6293 | /* | |
6294 | * gotp can be null in 2 cases: 1) if there are no extents | |
6295 | * or 2) split_fsb lies in a hole beyond which there are | |
6296 | * no extents. Either way, we are done. | |
6297 | */ | |
6298 | gotp = xfs_iext_bno_to_ext(ifp, split_fsb, ¤t_ext); | |
6299 | if (!gotp) | |
6300 | return 0; | |
6301 | ||
6302 | xfs_bmbt_get_all(gotp, &got); | |
6303 | ||
6304 | /* | |
6305 | * Check split_fsb lies in a hole or the start boundary offset | |
6306 | * of the extent. | |
6307 | */ | |
6308 | if (got.br_startoff >= split_fsb) | |
6309 | return 0; | |
6310 | ||
6311 | gotblkcnt = split_fsb - got.br_startoff; | |
6312 | new.br_startoff = split_fsb; | |
6313 | new.br_startblock = got.br_startblock + gotblkcnt; | |
6314 | new.br_blockcount = got.br_blockcount - gotblkcnt; | |
6315 | new.br_state = got.br_state; | |
6316 | ||
6317 | if (ifp->if_flags & XFS_IFBROOT) { | |
6318 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
6319 | cur->bc_private.b.firstblock = *firstfsb; | |
2c3234d1 | 6320 | cur->bc_private.b.dfops = dfops; |
a904b1ca NJ |
6321 | cur->bc_private.b.flags = 0; |
6322 | error = xfs_bmbt_lookup_eq(cur, got.br_startoff, | |
6323 | got.br_startblock, | |
6324 | got.br_blockcount, | |
6325 | &i); | |
6326 | if (error) | |
6327 | goto del_cursor; | |
6328 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
6329 | } | |
6330 | ||
6331 | xfs_bmbt_set_blockcount(gotp, gotblkcnt); | |
6332 | got.br_blockcount = gotblkcnt; | |
6333 | ||
6334 | logflags = XFS_ILOG_CORE; | |
6335 | if (cur) { | |
6336 | error = xfs_bmbt_update(cur, got.br_startoff, | |
6337 | got.br_startblock, | |
6338 | got.br_blockcount, | |
6339 | got.br_state); | |
6340 | if (error) | |
6341 | goto del_cursor; | |
6342 | } else | |
6343 | logflags |= XFS_ILOG_DEXT; | |
6344 | ||
6345 | /* Add new extent */ | |
6346 | current_ext++; | |
6347 | xfs_iext_insert(ip, current_ext, 1, &new, 0); | |
6348 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
6349 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
6350 | ||
6351 | if (cur) { | |
6352 | error = xfs_bmbt_lookup_eq(cur, new.br_startoff, | |
6353 | new.br_startblock, new.br_blockcount, | |
6354 | &i); | |
6355 | if (error) | |
6356 | goto del_cursor; | |
6357 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor); | |
6358 | cur->bc_rec.b.br_state = new.br_state; | |
6359 | ||
6360 | error = xfs_btree_insert(cur, &i); | |
6361 | if (error) | |
6362 | goto del_cursor; | |
6363 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
6364 | } | |
6365 | ||
6366 | /* | |
6367 | * Convert to a btree if necessary. | |
6368 | */ | |
6369 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
6370 | int tmp_logflags; /* partial log flag return val */ | |
6371 | ||
6372 | ASSERT(cur == NULL); | |
2c3234d1 | 6373 | error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops, |
a904b1ca NJ |
6374 | &cur, 0, &tmp_logflags, whichfork); |
6375 | logflags |= tmp_logflags; | |
6376 | } | |
6377 | ||
6378 | del_cursor: | |
6379 | if (cur) { | |
6380 | cur->bc_private.b.allocated = 0; | |
6381 | xfs_btree_del_cursor(cur, | |
6382 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
6383 | } | |
6384 | ||
6385 | if (logflags) | |
6386 | xfs_trans_log_inode(tp, ip, logflags); | |
6387 | return error; | |
6388 | } | |
6389 | ||
6390 | int | |
6391 | xfs_bmap_split_extent( | |
6392 | struct xfs_inode *ip, | |
6393 | xfs_fileoff_t split_fsb) | |
6394 | { | |
6395 | struct xfs_mount *mp = ip->i_mount; | |
6396 | struct xfs_trans *tp; | |
2c3234d1 | 6397 | struct xfs_defer_ops dfops; |
a904b1ca | 6398 | xfs_fsblock_t firstfsb; |
a904b1ca NJ |
6399 | int error; |
6400 | ||
253f4911 CH |
6401 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, |
6402 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); | |
6403 | if (error) | |
a904b1ca | 6404 | return error; |
a904b1ca NJ |
6405 | |
6406 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
6407 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
6408 | ||
2c3234d1 | 6409 | xfs_defer_init(&dfops, &firstfsb); |
a904b1ca NJ |
6410 | |
6411 | error = xfs_bmap_split_extent_at(tp, ip, split_fsb, | |
2c3234d1 | 6412 | &firstfsb, &dfops); |
a904b1ca NJ |
6413 | if (error) |
6414 | goto out; | |
6415 | ||
2c3234d1 | 6416 | error = xfs_defer_finish(&tp, &dfops, NULL); |
a904b1ca NJ |
6417 | if (error) |
6418 | goto out; | |
6419 | ||
70393313 | 6420 | return xfs_trans_commit(tp); |
a904b1ca NJ |
6421 | |
6422 | out: | |
2c3234d1 | 6423 | xfs_defer_cancel(&dfops); |
4906e215 | 6424 | xfs_trans_cancel(tp); |
a904b1ca NJ |
6425 | return error; |
6426 | } | |
9f3afb57 DW |
6427 | |
6428 | /* Deferred mapping is only for real extents in the data fork. */ | |
6429 | static bool | |
6430 | xfs_bmap_is_update_needed( | |
6431 | struct xfs_bmbt_irec *bmap) | |
6432 | { | |
6433 | return bmap->br_startblock != HOLESTARTBLOCK && | |
6434 | bmap->br_startblock != DELAYSTARTBLOCK; | |
6435 | } | |
6436 | ||
6437 | /* Record a bmap intent. */ | |
6438 | static int | |
6439 | __xfs_bmap_add( | |
6440 | struct xfs_mount *mp, | |
6441 | struct xfs_defer_ops *dfops, | |
6442 | enum xfs_bmap_intent_type type, | |
6443 | struct xfs_inode *ip, | |
6444 | int whichfork, | |
6445 | struct xfs_bmbt_irec *bmap) | |
6446 | { | |
6447 | int error; | |
6448 | struct xfs_bmap_intent *bi; | |
6449 | ||
6450 | trace_xfs_bmap_defer(mp, | |
6451 | XFS_FSB_TO_AGNO(mp, bmap->br_startblock), | |
6452 | type, | |
6453 | XFS_FSB_TO_AGBNO(mp, bmap->br_startblock), | |
6454 | ip->i_ino, whichfork, | |
6455 | bmap->br_startoff, | |
6456 | bmap->br_blockcount, | |
6457 | bmap->br_state); | |
6458 | ||
6459 | bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS); | |
6460 | INIT_LIST_HEAD(&bi->bi_list); | |
6461 | bi->bi_type = type; | |
6462 | bi->bi_owner = ip; | |
6463 | bi->bi_whichfork = whichfork; | |
6464 | bi->bi_bmap = *bmap; | |
6465 | ||
6466 | error = xfs_defer_join(dfops, bi->bi_owner); | |
6467 | if (error) { | |
6468 | kmem_free(bi); | |
6469 | return error; | |
6470 | } | |
6471 | ||
6472 | xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list); | |
6473 | return 0; | |
6474 | } | |
6475 | ||
6476 | /* Map an extent into a file. */ | |
6477 | int | |
6478 | xfs_bmap_map_extent( | |
6479 | struct xfs_mount *mp, | |
6480 | struct xfs_defer_ops *dfops, | |
6481 | struct xfs_inode *ip, | |
6482 | struct xfs_bmbt_irec *PREV) | |
6483 | { | |
6484 | if (!xfs_bmap_is_update_needed(PREV)) | |
6485 | return 0; | |
6486 | ||
6487 | return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip, | |
6488 | XFS_DATA_FORK, PREV); | |
6489 | } | |
6490 | ||
6491 | /* Unmap an extent out of a file. */ | |
6492 | int | |
6493 | xfs_bmap_unmap_extent( | |
6494 | struct xfs_mount *mp, | |
6495 | struct xfs_defer_ops *dfops, | |
6496 | struct xfs_inode *ip, | |
6497 | struct xfs_bmbt_irec *PREV) | |
6498 | { | |
6499 | if (!xfs_bmap_is_update_needed(PREV)) | |
6500 | return 0; | |
6501 | ||
6502 | return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip, | |
6503 | XFS_DATA_FORK, PREV); | |
6504 | } | |
6505 | ||
6506 | /* | |
6507 | * Process one of the deferred bmap operations. We pass back the | |
6508 | * btree cursor to maintain our lock on the bmapbt between calls. | |
6509 | */ | |
6510 | int | |
6511 | xfs_bmap_finish_one( | |
6512 | struct xfs_trans *tp, | |
6513 | struct xfs_defer_ops *dfops, | |
6514 | struct xfs_inode *ip, | |
6515 | enum xfs_bmap_intent_type type, | |
6516 | int whichfork, | |
6517 | xfs_fileoff_t startoff, | |
6518 | xfs_fsblock_t startblock, | |
e1a4e37c | 6519 | xfs_filblks_t *blockcount, |
9f3afb57 DW |
6520 | xfs_exntst_t state) |
6521 | { | |
e1a4e37c DW |
6522 | xfs_fsblock_t firstfsb; |
6523 | int error = 0; | |
9f3afb57 | 6524 | |
4c1a67bd DW |
6525 | /* |
6526 | * firstfsb is tied to the transaction lifetime and is used to | |
6527 | * ensure correct AG locking order and schedule work item | |
6528 | * continuations. XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us | |
6529 | * to only making one bmap call per transaction, so it should | |
6530 | * be safe to have it as a local variable here. | |
6531 | */ | |
6532 | firstfsb = NULLFSBLOCK; | |
6533 | ||
9f3afb57 DW |
6534 | trace_xfs_bmap_deferred(tp->t_mountp, |
6535 | XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type, | |
6536 | XFS_FSB_TO_AGBNO(tp->t_mountp, startblock), | |
e1a4e37c | 6537 | ip->i_ino, whichfork, startoff, *blockcount, state); |
9f3afb57 | 6538 | |
39e07daa | 6539 | if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK)) |
9f3afb57 | 6540 | return -EFSCORRUPTED; |
9f3afb57 DW |
6541 | |
6542 | if (XFS_TEST_ERROR(false, tp->t_mountp, | |
9e24cfd0 | 6543 | XFS_ERRTAG_BMAP_FINISH_ONE)) |
9f3afb57 DW |
6544 | return -EIO; |
6545 | ||
6546 | switch (type) { | |
6547 | case XFS_BMAP_MAP: | |
e1a4e37c | 6548 | error = xfs_bmapi_remap(tp, ip, startoff, *blockcount, |
6ebd5a44 | 6549 | startblock, dfops); |
e1a4e37c | 6550 | *blockcount = 0; |
9f3afb57 DW |
6551 | break; |
6552 | case XFS_BMAP_UNMAP: | |
e1a4e37c DW |
6553 | error = __xfs_bunmapi(tp, ip, startoff, blockcount, |
6554 | XFS_BMAPI_REMAP, 1, &firstfsb, dfops); | |
9f3afb57 DW |
6555 | break; |
6556 | default: | |
6557 | ASSERT(0); | |
6558 | error = -EFSCORRUPTED; | |
6559 | } | |
6560 | ||
6561 | return error; | |
6562 | } |