]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
5e656dbb | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
da23017d | 3 | * All Rights Reserved. |
5000d01d | 4 | * |
da23017d 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 | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d 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. | |
5000d01d | 13 | * |
da23017d 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 | |
2bd0ea18 | 17 | */ |
9c799827 | 18 | #include "libxfs_priv.h" |
b626fb59 DC |
19 | #include "xfs_fs.h" |
20 | #include "xfs_shared.h" | |
21 | #include "xfs_format.h" | |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
24 | #include "xfs_bit.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_mount.h" | |
f944d3d0 | 27 | #include "xfs_defer.h" |
b626fb59 DC |
28 | #include "xfs_da_format.h" |
29 | #include "xfs_da_btree.h" | |
30 | #include "xfs_dir2.h" | |
31 | #include "xfs_inode.h" | |
32 | #include "xfs_btree.h" | |
33 | #include "xfs_trans.h" | |
34 | #include "xfs_alloc.h" | |
35 | #include "xfs_bmap.h" | |
36 | #include "xfs_bmap_btree.h" | |
37 | #include "xfs_trans_space.h" | |
38 | #include "xfs_trace.h" | |
39 | #include "xfs_attr_leaf.h" | |
40 | #include "xfs_quota_defs.h" | |
85aec44f | 41 | #include "xfs_rmap.h" |
cf8ce220 | 42 | #include "xfs_ag_resv.h" |
cfe32f0d | 43 | #include "xfs_refcount.h" |
b626fb59 | 44 | |
2bd0ea18 | 45 | |
5e656dbb BN |
46 | kmem_zone_t *xfs_bmap_free_item_zone; |
47 | ||
48 | /* | |
49f693fa | 49 | * Miscellaneous helper functions |
5e656dbb | 50 | */ |
5e656dbb | 51 | |
5e656dbb | 52 | /* |
49f693fa DC |
53 | * Compute and fill in the value of the maximum depth of a bmap btree |
54 | * in this filesystem. Done once, during mount. | |
5e656dbb | 55 | */ |
49f693fa DC |
56 | void |
57 | xfs_bmap_compute_maxlevels( | |
58 | xfs_mount_t *mp, /* file system mount structure */ | |
59 | int whichfork) /* data or attr fork */ | |
60 | { | |
61 | int level; /* btree level */ | |
62 | uint maxblocks; /* max blocks at this level */ | |
63 | uint maxleafents; /* max leaf entries possible */ | |
64 | int maxrootrecs; /* max records in root block */ | |
65 | int minleafrecs; /* min records in leaf block */ | |
66 | int minnoderecs; /* min records in node block */ | |
67 | int sz; /* root block size */ | |
5e656dbb | 68 | |
49f693fa DC |
69 | /* |
70 | * The maximum number of extents in a file, hence the maximum | |
71 | * number of leaf entries, is controlled by the type of di_nextents | |
72 | * (a signed 32-bit number, xfs_extnum_t), or by di_anextents | |
73 | * (a signed 16-bit number, xfs_aextnum_t). | |
74 | * | |
75 | * Note that we can no longer assume that if we are in ATTR1 that | |
76 | * the fork offset of all the inodes will be | |
77 | * (xfs_default_attroffset(ip) >> 3) because we could have mounted | |
78 | * with ATTR2 and then mounted back with ATTR1, keeping the | |
79 | * di_forkoff's fixed but probably at various positions. Therefore, | |
80 | * for both ATTR1 and ATTR2 we have to assume the worst case scenario | |
81 | * of a minimum size available. | |
82 | */ | |
83 | if (whichfork == XFS_DATA_FORK) { | |
84 | maxleafents = MAXEXTNUM; | |
85 | sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS); | |
86 | } else { | |
87 | maxleafents = MAXAEXTNUM; | |
88 | sz = XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
89 | } | |
ff105f75 | 90 | maxrootrecs = xfs_bmdr_maxrecs(sz, 0); |
49f693fa DC |
91 | minleafrecs = mp->m_bmap_dmnr[0]; |
92 | minnoderecs = mp->m_bmap_dmnr[1]; | |
93 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
94 | for (level = 1; maxblocks > 1; level++) { | |
95 | if (maxblocks <= maxrootrecs) | |
96 | maxblocks = 1; | |
97 | else | |
98 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
99 | } | |
100 | mp->m_bm_maxlevels[whichfork] = level; | |
101 | } | |
5e656dbb | 102 | |
b194c7d8 BN |
103 | STATIC int /* error */ |
104 | xfs_bmbt_lookup_eq( | |
105 | struct xfs_btree_cur *cur, | |
70a93110 | 106 | struct xfs_bmbt_irec *irec, |
b194c7d8 BN |
107 | int *stat) /* success/failure */ |
108 | { | |
70a93110 | 109 | cur->bc_rec.b = *irec; |
b194c7d8 BN |
110 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); |
111 | } | |
112 | ||
113 | STATIC int /* error */ | |
4f76f49c | 114 | xfs_bmbt_lookup_first( |
b194c7d8 | 115 | struct xfs_btree_cur *cur, |
b194c7d8 BN |
116 | int *stat) /* success/failure */ |
117 | { | |
4f76f49c CH |
118 | cur->bc_rec.b.br_startoff = 0; |
119 | cur->bc_rec.b.br_startblock = 0; | |
120 | cur->bc_rec.b.br_blockcount = 0; | |
b194c7d8 BN |
121 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); |
122 | } | |
123 | ||
124 | /* | |
a2ceac1f DC |
125 | * Check if the inode needs to be converted to btree format. |
126 | */ | |
127 | static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork) | |
128 | { | |
1277a5e0 DW |
129 | return whichfork != XFS_COW_FORK && |
130 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
a2ceac1f DC |
131 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
132 | XFS_IFORK_MAXEXT(ip, whichfork); | |
133 | } | |
134 | ||
135 | /* | |
136 | * Check if the inode should be converted to extent format. | |
137 | */ | |
138 | static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork) | |
139 | { | |
1277a5e0 DW |
140 | return whichfork != XFS_COW_FORK && |
141 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE && | |
a2ceac1f DC |
142 | XFS_IFORK_NEXTENTS(ip, whichfork) <= |
143 | XFS_IFORK_MAXEXT(ip, whichfork); | |
144 | } | |
145 | ||
146 | /* | |
d0e5f1ff | 147 | * Update the record referred to by cur to the value given by irec |
b194c7d8 BN |
148 | * This either works (return 0) or gets an EFSCORRUPTED error. |
149 | */ | |
150 | STATIC int | |
151 | xfs_bmbt_update( | |
152 | struct xfs_btree_cur *cur, | |
d0e5f1ff | 153 | struct xfs_bmbt_irec *irec) |
b194c7d8 BN |
154 | { |
155 | union xfs_btree_rec rec; | |
156 | ||
d0e5f1ff | 157 | xfs_bmbt_disk_set_all(&rec.bmbt, irec); |
b194c7d8 BN |
158 | return xfs_btree_update(cur, &rec); |
159 | } | |
160 | ||
5e656dbb | 161 | /* |
49f693fa DC |
162 | * Compute the worst-case number of indirect blocks that will be used |
163 | * for ip's delayed extent of length "len". | |
5e656dbb | 164 | */ |
49f693fa DC |
165 | STATIC xfs_filblks_t |
166 | xfs_bmap_worst_indlen( | |
167 | xfs_inode_t *ip, /* incore inode pointer */ | |
168 | xfs_filblks_t len) /* delayed extent length */ | |
57c9fccb | 169 | { |
49f693fa DC |
170 | int level; /* btree level number */ |
171 | int maxrecs; /* maximum record count at this level */ | |
172 | xfs_mount_t *mp; /* mount structure */ | |
173 | xfs_filblks_t rval; /* return value */ | |
57c9fccb NS |
174 | |
175 | mp = ip->i_mount; | |
49f693fa DC |
176 | maxrecs = mp->m_bmap_dmxr[0]; |
177 | for (level = 0, rval = 0; | |
178 | level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK); | |
179 | level++) { | |
180 | len += maxrecs - 1; | |
181 | do_div(len, maxrecs); | |
182 | rval += len; | |
7fbe9b54 DW |
183 | if (len == 1) |
184 | return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) - | |
49f693fa DC |
185 | level - 1; |
186 | if (level == 0) | |
187 | maxrecs = mp->m_bmap_dmxr[1]; | |
57c9fccb | 188 | } |
49f693fa | 189 | return rval; |
57c9fccb NS |
190 | } |
191 | ||
192 | /* | |
49f693fa | 193 | * Calculate the default attribute fork offset for newly created inodes. |
57c9fccb | 194 | */ |
49f693fa DC |
195 | uint |
196 | xfs_default_attroffset( | |
197 | struct xfs_inode *ip) | |
57c9fccb | 198 | { |
49f693fa DC |
199 | struct xfs_mount *mp = ip->i_mount; |
200 | uint offset; | |
57c9fccb | 201 | |
49f693fa DC |
202 | if (mp->m_sb.sb_inodesize == 256) { |
203 | offset = XFS_LITINO(mp, ip->i_d.di_version) - | |
204 | XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
205 | } else { | |
206 | offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); | |
57c9fccb | 207 | } |
49f693fa DC |
208 | |
209 | ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version)); | |
210 | return offset; | |
57c9fccb NS |
211 | } |
212 | ||
213 | /* | |
49f693fa DC |
214 | * Helper routine to reset inode di_forkoff field when switching |
215 | * attribute fork from local to extent format - we reset it where | |
216 | * possible to make space available for inline data fork extents. | |
57c9fccb | 217 | */ |
49f693fa DC |
218 | STATIC void |
219 | xfs_bmap_forkoff_reset( | |
49f693fa DC |
220 | xfs_inode_t *ip, |
221 | int whichfork) | |
57c9fccb | 222 | { |
49f693fa DC |
223 | if (whichfork == XFS_ATTR_FORK && |
224 | ip->i_d.di_format != XFS_DINODE_FMT_DEV && | |
49f693fa DC |
225 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE) { |
226 | uint dfl_forkoff = xfs_default_attroffset(ip) >> 3; | |
57c9fccb | 227 | |
49f693fa DC |
228 | if (dfl_forkoff > ip->i_d.di_forkoff) |
229 | ip->i_d.di_forkoff = dfl_forkoff; | |
230 | } | |
57c9fccb NS |
231 | } |
232 | ||
49f693fa DC |
233 | #ifdef DEBUG |
234 | STATIC struct xfs_buf * | |
235 | xfs_bmap_get_bp( | |
236 | struct xfs_btree_cur *cur, | |
237 | xfs_fsblock_t bno) | |
238 | { | |
239 | struct xfs_log_item_desc *lidp; | |
240 | int i; | |
56b2de80 | 241 | |
49f693fa DC |
242 | if (!cur) |
243 | return NULL; | |
2bd0ea18 | 244 | |
49f693fa DC |
245 | for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) { |
246 | if (!cur->bc_bufs[i]) | |
247 | break; | |
248 | if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno) | |
249 | return cur->bc_bufs[i]; | |
250 | } | |
56b2de80 | 251 | |
49f693fa DC |
252 | /* Chase down all the log items to see if the bp is there */ |
253 | list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) { | |
254 | struct xfs_buf_log_item *bip; | |
255 | bip = (struct xfs_buf_log_item *)lidp->lid_item; | |
256 | if (bip->bli_item.li_type == XFS_LI_BUF && | |
257 | XFS_BUF_ADDR(bip->bli_buf) == bno) | |
258 | return bip->bli_buf; | |
259 | } | |
2bd0ea18 | 260 | |
49f693fa DC |
261 | return NULL; |
262 | } | |
56b2de80 | 263 | |
49f693fa DC |
264 | STATIC void |
265 | xfs_check_block( | |
266 | struct xfs_btree_block *block, | |
267 | xfs_mount_t *mp, | |
268 | int root, | |
269 | short sz) | |
270 | { | |
271 | int i, j, dmxr; | |
272 | __be64 *pp, *thispa; /* pointer to block address */ | |
273 | xfs_bmbt_key_t *prevp, *keyp; | |
2bd0ea18 | 274 | |
49f693fa | 275 | ASSERT(be16_to_cpu(block->bb_level) > 0); |
56b2de80 | 276 | |
49f693fa DC |
277 | prevp = NULL; |
278 | for( i = 1; i <= xfs_btree_get_numrecs(block); i++) { | |
279 | dmxr = mp->m_bmap_dmxr[0]; | |
280 | keyp = XFS_BMBT_KEY_ADDR(mp, block, i); | |
a2ceac1f | 281 | |
49f693fa DC |
282 | if (prevp) { |
283 | ASSERT(be64_to_cpu(prevp->br_startoff) < | |
284 | be64_to_cpu(keyp->br_startoff)); | |
285 | } | |
286 | prevp = keyp; | |
2bd0ea18 | 287 | |
2bd0ea18 | 288 | /* |
49f693fa | 289 | * Compare the block numbers to see if there are dups. |
2bd0ea18 | 290 | */ |
49f693fa DC |
291 | if (root) |
292 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz); | |
293 | else | |
294 | pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr); | |
295 | ||
296 | for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) { | |
297 | if (root) | |
298 | thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz); | |
299 | else | |
300 | thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr); | |
301 | if (*thispa == *pp) { | |
302 | xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld", | |
303 | __func__, j, i, | |
304 | (unsigned long long)be64_to_cpu(*thispa)); | |
305 | panic("%s: ptrs are equal in node\n", | |
306 | __func__); | |
307 | } | |
2bd0ea18 | 308 | } |
49f693fa DC |
309 | } |
310 | } | |
a2ceac1f | 311 | |
49f693fa DC |
312 | /* |
313 | * Check that the extents for the inode ip are in the right order in all | |
f07ae2a6 DC |
314 | * btree leaves. THis becomes prohibitively expensive for large extent count |
315 | * files, so don't bother with inodes that have more than 10,000 extents in | |
316 | * them. The btree record ordering checks will still be done, so for such large | |
317 | * bmapbt constructs that is going to catch most corruptions. | |
49f693fa | 318 | */ |
49f693fa DC |
319 | STATIC void |
320 | xfs_bmap_check_leaf_extents( | |
321 | xfs_btree_cur_t *cur, /* btree cursor or null */ | |
322 | xfs_inode_t *ip, /* incore inode pointer */ | |
323 | int whichfork) /* data or attr fork */ | |
324 | { | |
325 | struct xfs_btree_block *block; /* current btree block */ | |
326 | xfs_fsblock_t bno; /* block # of "block" */ | |
327 | xfs_buf_t *bp; /* buffer for "block" */ | |
328 | int error; /* error return value */ | |
329 | xfs_extnum_t i=0, j; /* index into the extents list */ | |
330 | xfs_ifork_t *ifp; /* fork structure */ | |
331 | int level; /* btree level, for checking */ | |
332 | xfs_mount_t *mp; /* file system mount structure */ | |
333 | __be64 *pp; /* pointer to block address */ | |
334 | xfs_bmbt_rec_t *ep; /* pointer to current extent */ | |
335 | xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */ | |
336 | xfs_bmbt_rec_t *nextp; /* pointer to next extent */ | |
337 | int bp_release = 0; | |
338 | ||
339 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) { | |
340 | return; | |
341 | } | |
342 | ||
f07ae2a6 DC |
343 | /* skip large extent count inodes */ |
344 | if (ip->i_d.di_nextents > 10000) | |
345 | return; | |
346 | ||
49f693fa DC |
347 | bno = NULLFSBLOCK; |
348 | mp = ip->i_mount; | |
349 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
350 | block = ifp->if_broot; | |
351 | /* | |
352 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
353 | */ | |
354 | level = be16_to_cpu(block->bb_level); | |
355 | ASSERT(level > 0); | |
356 | xfs_check_block(block, mp, 1, ifp->if_broot_bytes); | |
357 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
358 | bno = be64_to_cpu(*pp); | |
359 | ||
5a35bf2c | 360 | ASSERT(bno != NULLFSBLOCK); |
49f693fa DC |
361 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
362 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
363 | ||
364 | /* | |
365 | * Go down the tree until leaf level is reached, following the first | |
366 | * pointer (leftmost) at each level. | |
367 | */ | |
368 | while (level-- > 0) { | |
369 | /* See if buf is in cur first */ | |
370 | bp_release = 0; | |
371 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
372 | if (!bp) { | |
373 | bp_release = 1; | |
374 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
375 | XFS_BMAP_BTREE_REF, | |
376 | &xfs_bmbt_buf_ops); | |
2bd0ea18 | 377 | if (error) |
49f693fa | 378 | goto error_norelse; |
2bd0ea18 | 379 | } |
49f693fa | 380 | block = XFS_BUF_TO_BLOCK(bp); |
49f693fa DC |
381 | if (level == 0) |
382 | break; | |
2bd0ea18 | 383 | |
2bd0ea18 | 384 | /* |
49f693fa DC |
385 | * Check this block for basic sanity (increasing keys and |
386 | * no duplicate blocks). | |
2bd0ea18 | 387 | */ |
49f693fa DC |
388 | |
389 | xfs_check_block(block, mp, 0, 0); | |
390 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
391 | bno = be64_to_cpu(*pp); | |
19ebedcf DC |
392 | XFS_WANT_CORRUPTED_GOTO(mp, |
393 | XFS_FSB_SANITY_CHECK(mp, bno), error0); | |
49f693fa DC |
394 | if (bp_release) { |
395 | bp_release = 0; | |
396 | xfs_trans_brelse(NULL, bp); | |
2bd0ea18 | 397 | } |
49f693fa | 398 | } |
a2ceac1f | 399 | |
49f693fa DC |
400 | /* |
401 | * Here with bp and block set to the leftmost leaf node in the tree. | |
402 | */ | |
403 | i = 0; | |
a2ceac1f | 404 | |
49f693fa DC |
405 | /* |
406 | * Loop over all leaf nodes checking that all extents are in the right order. | |
407 | */ | |
408 | for (;;) { | |
409 | xfs_fsblock_t nextbno; | |
410 | xfs_extnum_t num_recs; | |
411 | ||
412 | ||
413 | num_recs = xfs_btree_get_numrecs(block); | |
2bd0ea18 | 414 | |
2bd0ea18 | 415 | /* |
49f693fa | 416 | * Read-ahead the next leaf block, if any. |
2bd0ea18 | 417 | */ |
a2ceac1f | 418 | |
49f693fa | 419 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
a2ceac1f | 420 | |
49f693fa DC |
421 | /* |
422 | * Check all the extents to make sure they are OK. | |
423 | * If we had a previous block, the last entry should | |
424 | * conform with the first entry in this one. | |
425 | */ | |
2bd0ea18 | 426 | |
49f693fa DC |
427 | ep = XFS_BMBT_REC_ADDR(mp, block, 1); |
428 | if (i) { | |
429 | ASSERT(xfs_bmbt_disk_get_startoff(&last) + | |
430 | xfs_bmbt_disk_get_blockcount(&last) <= | |
431 | xfs_bmbt_disk_get_startoff(ep)); | |
432 | } | |
433 | for (j = 1; j < num_recs; j++) { | |
434 | nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1); | |
435 | ASSERT(xfs_bmbt_disk_get_startoff(ep) + | |
436 | xfs_bmbt_disk_get_blockcount(ep) <= | |
437 | xfs_bmbt_disk_get_startoff(nextp)); | |
438 | ep = nextp; | |
439 | } | |
440 | ||
441 | last = *ep; | |
442 | i += num_recs; | |
443 | if (bp_release) { | |
444 | bp_release = 0; | |
445 | xfs_trans_brelse(NULL, bp); | |
446 | } | |
447 | bno = nextbno; | |
2bd0ea18 | 448 | /* |
49f693fa | 449 | * If we've reached the end, stop. |
2bd0ea18 | 450 | */ |
49f693fa DC |
451 | if (bno == NULLFSBLOCK) |
452 | break; | |
a2ceac1f | 453 | |
49f693fa DC |
454 | bp_release = 0; |
455 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
456 | if (!bp) { | |
457 | bp_release = 1; | |
458 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
459 | XFS_BMAP_BTREE_REF, | |
460 | &xfs_bmbt_buf_ops); | |
a2ceac1f | 461 | if (error) |
49f693fa | 462 | goto error_norelse; |
2bd0ea18 | 463 | } |
49f693fa | 464 | block = XFS_BUF_TO_BLOCK(bp); |
a2ceac1f | 465 | } |
4d4a192c | 466 | |
49f693fa | 467 | return; |
a2ceac1f | 468 | |
49f693fa DC |
469 | error0: |
470 | xfs_warn(mp, "%s: at error0", __func__); | |
471 | if (bp_release) | |
472 | xfs_trans_brelse(NULL, bp); | |
473 | error_norelse: | |
474 | xfs_warn(mp, "%s: BAD after btree leaves for %d extents", | |
475 | __func__, i); | |
476 | panic("%s: CORRUPTED BTREE OR SOMETHING", __func__); | |
477 | return; | |
2bd0ea18 NS |
478 | } |
479 | ||
49f693fa DC |
480 | /* |
481 | * Validate that the bmbt_irecs being returned from bmapi are valid | |
e6d77a21 DC |
482 | * given the caller's original parameters. Specifically check the |
483 | * ranges of the returned irecs to ensure that they only extend beyond | |
49f693fa DC |
484 | * the given parameters if the XFS_BMAPI_ENTIRE flag was set. |
485 | */ | |
486 | STATIC void | |
487 | xfs_bmap_validate_ret( | |
488 | xfs_fileoff_t bno, | |
489 | xfs_filblks_t len, | |
490 | int flags, | |
491 | xfs_bmbt_irec_t *mval, | |
492 | int nmap, | |
493 | int ret_nmap) | |
494 | { | |
495 | int i; /* index to map values */ | |
a2ceac1f | 496 | |
49f693fa | 497 | ASSERT(ret_nmap <= nmap); |
a2ceac1f | 498 | |
49f693fa DC |
499 | for (i = 0; i < ret_nmap; i++) { |
500 | ASSERT(mval[i].br_blockcount > 0); | |
501 | if (!(flags & XFS_BMAPI_ENTIRE)) { | |
502 | ASSERT(mval[i].br_startoff >= bno); | |
503 | ASSERT(mval[i].br_blockcount <= len); | |
504 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount <= | |
505 | bno + len); | |
506 | } else { | |
507 | ASSERT(mval[i].br_startoff < bno + len); | |
508 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount > | |
509 | bno); | |
510 | } | |
511 | ASSERT(i == 0 || | |
512 | mval[i - 1].br_startoff + mval[i - 1].br_blockcount == | |
513 | mval[i].br_startoff); | |
514 | ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK && | |
515 | mval[i].br_startblock != HOLESTARTBLOCK); | |
516 | ASSERT(mval[i].br_state == XFS_EXT_NORM || | |
517 | mval[i].br_state == XFS_EXT_UNWRITTEN); | |
518 | } | |
519 | } | |
56b2de80 | 520 | |
49f693fa DC |
521 | #else |
522 | #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0) | |
9587a34c | 523 | #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0) |
49f693fa | 524 | #endif /* DEBUG */ |
56b2de80 | 525 | |
49f693fa DC |
526 | /* |
527 | * bmap free list manipulation functions | |
528 | */ | |
56b2de80 | 529 | |
49f693fa DC |
530 | /* |
531 | * Add the extent to the list of extents to be free at transaction end. | |
532 | * The list is maintained sorted (by block number). | |
533 | */ | |
534 | void | |
535 | xfs_bmap_add_free( | |
85aec44f DW |
536 | struct xfs_mount *mp, |
537 | struct xfs_defer_ops *dfops, | |
538 | xfs_fsblock_t bno, | |
539 | xfs_filblks_t len, | |
540 | struct xfs_owner_info *oinfo) | |
49f693fa | 541 | { |
6f530e9a | 542 | struct xfs_extent_free_item *new; /* new element */ |
49f693fa DC |
543 | #ifdef DEBUG |
544 | xfs_agnumber_t agno; | |
545 | xfs_agblock_t agbno; | |
56b2de80 | 546 | |
49f693fa DC |
547 | ASSERT(bno != NULLFSBLOCK); |
548 | ASSERT(len > 0); | |
549 | ASSERT(len <= MAXEXTLEN); | |
550 | ASSERT(!isnullstartblock(bno)); | |
551 | agno = XFS_FSB_TO_AGNO(mp, bno); | |
552 | agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
553 | ASSERT(agno < mp->m_sb.sb_agcount); | |
554 | ASSERT(agbno < mp->m_sb.sb_agblocks); | |
555 | ASSERT(len < mp->m_sb.sb_agblocks); | |
556 | ASSERT(agbno + len <= mp->m_sb.sb_agblocks); | |
557 | #endif | |
558 | ASSERT(xfs_bmap_free_item_zone != NULL); | |
85aec44f | 559 | |
49f693fa | 560 | new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP); |
6f530e9a DW |
561 | new->xefi_startblock = bno; |
562 | new->xefi_blockcount = (xfs_extlen_t)len; | |
85aec44f DW |
563 | if (oinfo) |
564 | new->xefi_oinfo = *oinfo; | |
565 | else | |
566 | xfs_rmap_skip_owner_update(&new->xefi_oinfo); | |
a9da40de DW |
567 | trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0, |
568 | XFS_FSB_TO_AGBNO(mp, bno), len); | |
f33cea1a | 569 | xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list); |
49f693fa | 570 | } |
2bd0ea18 | 571 | |
49f693fa DC |
572 | /* |
573 | * Inode fork format manipulation functions | |
574 | */ | |
a2ceac1f | 575 | |
49f693fa DC |
576 | /* |
577 | * Transform a btree format file with only one leaf node, where the | |
578 | * extents list will fit in the inode, into an extents format file. | |
579 | * Since the file extents are already in-core, all we have to do is | |
580 | * give up the space for the btree root and pitch the leaf block. | |
581 | */ | |
582 | STATIC int /* error */ | |
583 | xfs_bmap_btree_to_extents( | |
584 | xfs_trans_t *tp, /* transaction pointer */ | |
585 | xfs_inode_t *ip, /* incore inode pointer */ | |
586 | xfs_btree_cur_t *cur, /* btree cursor */ | |
587 | int *logflagsp, /* inode logging flags */ | |
588 | int whichfork) /* data or attr fork */ | |
589 | { | |
590 | /* REFERENCED */ | |
591 | struct xfs_btree_block *cblock;/* child btree block */ | |
592 | xfs_fsblock_t cbno; /* child block number */ | |
593 | xfs_buf_t *cbp; /* child block's buffer */ | |
594 | int error; /* error return value */ | |
595 | xfs_ifork_t *ifp; /* inode fork data */ | |
596 | xfs_mount_t *mp; /* mount point structure */ | |
597 | __be64 *pp; /* ptr to block address */ | |
598 | struct xfs_btree_block *rblock;/* root btree block */ | |
85aec44f | 599 | struct xfs_owner_info oinfo; |
56b2de80 | 600 | |
49f693fa DC |
601 | mp = ip->i_mount; |
602 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1277a5e0 | 603 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa DC |
604 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
605 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
606 | rblock = ifp->if_broot; | |
607 | ASSERT(be16_to_cpu(rblock->bb_level) == 1); | |
608 | ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1); | |
609 | ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1); | |
610 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes); | |
611 | cbno = be64_to_cpu(*pp); | |
612 | *logflagsp = 0; | |
613 | #ifdef DEBUG | |
d5c546df DW |
614 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, |
615 | xfs_btree_check_lptr(cur, cbno, 1)); | |
49f693fa DC |
616 | #endif |
617 | error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF, | |
618 | &xfs_bmbt_buf_ops); | |
619 | if (error) | |
620 | return error; | |
621 | cblock = XFS_BUF_TO_BLOCK(cbp); | |
622 | if ((error = xfs_btree_check_block(cur, cblock, 0, cbp))) | |
623 | return error; | |
85aec44f DW |
624 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); |
625 | xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo); | |
49f693fa DC |
626 | ip->i_d.di_nblocks--; |
627 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); | |
628 | xfs_trans_binval(tp, cbp); | |
629 | if (cur->bc_bufs[0] == cbp) | |
630 | cur->bc_bufs[0] = NULL; | |
631 | xfs_iroot_realloc(ip, -1, whichfork); | |
632 | ASSERT(ifp->if_broot == NULL); | |
633 | ASSERT((ifp->if_flags & XFS_IFBROOT) == 0); | |
634 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
635 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); | |
636 | return 0; | |
637 | } | |
2bd0ea18 NS |
638 | |
639 | /* | |
49f693fa DC |
640 | * Convert an extents-format file into a btree-format file. |
641 | * The new file will have a root block (in the inode) and a single child block. | |
2bd0ea18 | 642 | */ |
49f693fa DC |
643 | STATIC int /* error */ |
644 | xfs_bmap_extents_to_btree( | |
645 | xfs_trans_t *tp, /* transaction pointer */ | |
646 | xfs_inode_t *ip, /* incore inode pointer */ | |
647 | xfs_fsblock_t *firstblock, /* first-block-allocated */ | |
f33cea1a | 648 | struct xfs_defer_ops *dfops, /* blocks freed in xaction */ |
49f693fa DC |
649 | xfs_btree_cur_t **curp, /* cursor returned to caller */ |
650 | int wasdel, /* converting a delayed alloc */ | |
651 | int *logflagsp, /* inode logging flags */ | |
652 | int whichfork) /* data or attr fork */ | |
2bd0ea18 | 653 | { |
49f693fa DC |
654 | struct xfs_btree_block *ablock; /* allocated (child) bt block */ |
655 | xfs_buf_t *abp; /* buffer for ablock */ | |
656 | xfs_alloc_arg_t args; /* allocation arguments */ | |
657 | xfs_bmbt_rec_t *arp; /* child record pointer */ | |
658 | struct xfs_btree_block *block; /* btree root block */ | |
659 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
660 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ | |
661 | int error; /* error return value */ | |
662 | xfs_extnum_t i, cnt; /* extent record index */ | |
663 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
664 | xfs_bmbt_key_t *kp; /* root block key pointer */ | |
665 | xfs_mount_t *mp; /* mount structure */ | |
666 | xfs_extnum_t nextents; /* number of file extents */ | |
667 | xfs_bmbt_ptr_t *pp; /* root block address pointer */ | |
2bd0ea18 | 668 | |
5dfa5cd2 | 669 | mp = ip->i_mount; |
1277a5e0 | 670 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa DC |
671 | ifp = XFS_IFORK_PTR(ip, whichfork); |
672 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS); | |
56b2de80 | 673 | |
2bd0ea18 | 674 | /* |
49f693fa | 675 | * Make space in the inode incore. |
2bd0ea18 | 676 | */ |
49f693fa DC |
677 | xfs_iroot_realloc(ip, 1, whichfork); |
678 | ifp->if_flags |= XFS_IFBROOT; | |
56b2de80 | 679 | |
2bd0ea18 | 680 | /* |
49f693fa | 681 | * Fill in the root. |
2bd0ea18 | 682 | */ |
49f693fa | 683 | block = ifp->if_broot; |
e394a4b1 ES |
684 | xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL, |
685 | XFS_BTNUM_BMAP, 1, 1, ip->i_ino, | |
f4241a08 | 686 | XFS_BTREE_LONG_PTRS); |
49f693fa DC |
687 | /* |
688 | * Need a cursor. Can't allocate until bb_level is filled in. | |
689 | */ | |
49f693fa DC |
690 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
691 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 692 | cur->bc_private.b.dfops = dfops; |
49f693fa DC |
693 | cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; |
694 | /* | |
695 | * Convert to a btree with two levels, one record in root. | |
696 | */ | |
697 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE); | |
698 | memset(&args, 0, sizeof(args)); | |
699 | args.tp = tp; | |
700 | args.mp = mp; | |
85aec44f | 701 | xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork); |
49f693fa DC |
702 | args.firstblock = *firstblock; |
703 | if (*firstblock == NULLFSBLOCK) { | |
704 | args.type = XFS_ALLOCTYPE_START_BNO; | |
705 | args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino); | |
f33cea1a | 706 | } else if (dfops->dop_low) { |
49f693fa DC |
707 | args.type = XFS_ALLOCTYPE_START_BNO; |
708 | args.fsbno = *firstblock; | |
709 | } else { | |
710 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
711 | args.fsbno = *firstblock; | |
712 | } | |
713 | args.minlen = args.maxlen = args.prod = 1; | |
714 | args.wasdel = wasdel; | |
715 | *logflagsp = 0; | |
716 | if ((error = xfs_alloc_vextent(&args))) { | |
717 | xfs_iroot_realloc(ip, -1, whichfork); | |
718 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
719 | return error; | |
2bd0ea18 | 720 | } |
68d06dce | 721 | |
c1587ecf CH |
722 | if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) { |
723 | xfs_iroot_realloc(ip, -1, whichfork); | |
724 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
725 | return -ENOSPC; | |
726 | } | |
2bd0ea18 | 727 | /* |
49f693fa | 728 | * Allocation can't fail, the space was reserved. |
2bd0ea18 | 729 | */ |
49f693fa | 730 | ASSERT(*firstblock == NULLFSBLOCK || |
6df6a3ab | 731 | args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock)); |
49f693fa DC |
732 | *firstblock = cur->bc_private.b.firstblock = args.fsbno; |
733 | cur->bc_private.b.allocated++; | |
734 | ip->i_d.di_nblocks++; | |
735 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L); | |
736 | abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0); | |
2bd0ea18 | 737 | /* |
49f693fa | 738 | * Fill in the child block. |
2bd0ea18 | 739 | */ |
49f693fa DC |
740 | abp->b_ops = &xfs_bmbt_buf_ops; |
741 | ablock = XFS_BUF_TO_BLOCK(abp); | |
e394a4b1 ES |
742 | xfs_btree_init_block_int(mp, ablock, abp->b_bn, |
743 | XFS_BTNUM_BMAP, 0, 0, ip->i_ino, | |
5dfa5cd2 DC |
744 | XFS_BTREE_LONG_PTRS); |
745 | ||
49f693fa | 746 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); |
d09d4e5f | 747 | nextents = xfs_iext_count(ifp); |
49f693fa DC |
748 | for (cnt = i = 0; i < nextents; i++) { |
749 | ep = xfs_iext_get_ext(ifp, i); | |
750 | if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) { | |
751 | arp->l0 = cpu_to_be64(ep->l0); | |
752 | arp->l1 = cpu_to_be64(ep->l1); | |
753 | arp++; cnt++; | |
754 | } | |
755 | } | |
756 | ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork)); | |
757 | xfs_btree_set_numrecs(ablock, cnt); | |
56b2de80 | 758 | |
49f693fa DC |
759 | /* |
760 | * Fill in the root key and pointer. | |
761 | */ | |
762 | kp = XFS_BMBT_KEY_ADDR(mp, block, 1); | |
763 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); | |
764 | kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp)); | |
765 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur, | |
766 | be16_to_cpu(block->bb_level))); | |
767 | *pp = cpu_to_be64(args.fsbno); | |
2bd0ea18 | 768 | |
49f693fa DC |
769 | /* |
770 | * Do all this logging at the end so that | |
771 | * the root is at the right level. | |
772 | */ | |
5dfa5cd2 | 773 | xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS); |
613e6057 | 774 | xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs)); |
49f693fa DC |
775 | ASSERT(*curp == NULL); |
776 | *curp = cur; | |
777 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork); | |
778 | return 0; | |
779 | } | |
a2ceac1f | 780 | |
49f693fa DC |
781 | /* |
782 | * Convert a local file to an extents file. | |
783 | * This code is out of bounds for data forks of regular files, | |
784 | * since the file data needs to get logged so things will stay consistent. | |
785 | * (The bmap-level manipulations are ok, though). | |
786 | */ | |
3f17ed4b DC |
787 | void |
788 | xfs_bmap_local_to_extents_empty( | |
789 | struct xfs_inode *ip, | |
790 | int whichfork) | |
791 | { | |
792 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
793 | ||
1277a5e0 | 794 | ASSERT(whichfork != XFS_COW_FORK); |
3f17ed4b DC |
795 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); |
796 | ASSERT(ifp->if_bytes == 0); | |
797 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0); | |
798 | ||
ff105f75 | 799 | xfs_bmap_forkoff_reset(ip, whichfork); |
3f17ed4b DC |
800 | ifp->if_flags &= ~XFS_IFINLINE; |
801 | ifp->if_flags |= XFS_IFEXTENTS; | |
802 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
803 | } | |
804 | ||
805 | ||
49f693fa DC |
806 | STATIC int /* error */ |
807 | xfs_bmap_local_to_extents( | |
808 | xfs_trans_t *tp, /* transaction pointer */ | |
809 | xfs_inode_t *ip, /* incore inode pointer */ | |
810 | xfs_fsblock_t *firstblock, /* first block allocated in xaction */ | |
811 | xfs_extlen_t total, /* total blocks needed by transaction */ | |
812 | int *logflagsp, /* inode logging flags */ | |
813 | int whichfork, | |
5dfa5cd2 DC |
814 | void (*init_fn)(struct xfs_trans *tp, |
815 | struct xfs_buf *bp, | |
49f693fa DC |
816 | struct xfs_inode *ip, |
817 | struct xfs_ifork *ifp)) | |
818 | { | |
3f17ed4b | 819 | int error = 0; |
49f693fa DC |
820 | int flags; /* logging flags returned */ |
821 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
3f17ed4b DC |
822 | xfs_alloc_arg_t args; /* allocation arguments */ |
823 | xfs_buf_t *bp; /* buffer for extent block */ | |
05104e43 | 824 | struct xfs_bmbt_irec rec; |
2bd0ea18 | 825 | |
49f693fa DC |
826 | /* |
827 | * We don't want to deal with the case of keeping inode data inline yet. | |
828 | * So sending the data fork of a regular inode is invalid. | |
829 | */ | |
e37bf53c | 830 | ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK)); |
49f693fa DC |
831 | ifp = XFS_IFORK_PTR(ip, whichfork); |
832 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
3f17ed4b DC |
833 | |
834 | if (!ifp->if_bytes) { | |
835 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
836 | flags = XFS_ILOG_CORE; | |
837 | goto done; | |
838 | } | |
839 | ||
49f693fa DC |
840 | flags = 0; |
841 | error = 0; | |
3f17ed4b DC |
842 | ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == |
843 | XFS_IFINLINE); | |
844 | memset(&args, 0, sizeof(args)); | |
845 | args.tp = tp; | |
846 | args.mp = ip->i_mount; | |
85aec44f | 847 | xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0); |
3f17ed4b DC |
848 | args.firstblock = *firstblock; |
849 | /* | |
850 | * Allocate a block. We know we need only one, since the | |
851 | * file currently fits in an inode. | |
852 | */ | |
853 | if (*firstblock == NULLFSBLOCK) { | |
854 | args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino); | |
855 | args.type = XFS_ALLOCTYPE_START_BNO; | |
49f693fa | 856 | } else { |
3f17ed4b DC |
857 | args.fsbno = *firstblock; |
858 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
2bd0ea18 | 859 | } |
3f17ed4b DC |
860 | args.total = total; |
861 | args.minlen = args.maxlen = args.prod = 1; | |
862 | error = xfs_alloc_vextent(&args); | |
863 | if (error) | |
864 | goto done; | |
865 | ||
866 | /* Can't fail, the space was reserved. */ | |
867 | ASSERT(args.fsbno != NULLFSBLOCK); | |
868 | ASSERT(args.len == 1); | |
869 | *firstblock = args.fsbno; | |
870 | bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0); | |
871 | ||
19ebedcf | 872 | /* |
f44fbde0 | 873 | * Initialize the block, copy the data and log the remote buffer. |
19ebedcf | 874 | * |
f44fbde0 BF |
875 | * The callout is responsible for logging because the remote format |
876 | * might differ from the local format and thus we don't know how much to | |
877 | * log here. Note that init_fn must also set the buffer log item type | |
878 | * correctly. | |
19ebedcf | 879 | */ |
3f17ed4b DC |
880 | init_fn(tp, bp, ip, ifp); |
881 | ||
f44fbde0 | 882 | /* account for the change in fork size */ |
3f17ed4b DC |
883 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); |
884 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
49f693fa | 885 | flags |= XFS_ILOG_CORE; |
3f17ed4b | 886 | |
05104e43 CH |
887 | rec.br_startoff = 0; |
888 | rec.br_startblock = args.fsbno; | |
889 | rec.br_blockcount = 1; | |
890 | rec.br_state = XFS_EXT_NORM; | |
891 | xfs_iext_insert(ip, 0, 1, &rec, 0); | |
892 | ||
3f17ed4b DC |
893 | XFS_IFORK_NEXT_SET(ip, whichfork, 1); |
894 | ip->i_d.di_nblocks = 1; | |
895 | xfs_trans_mod_dquot_byino(tp, ip, | |
896 | XFS_TRANS_DQ_BCOUNT, 1L); | |
897 | flags |= xfs_ilog_fext(whichfork); | |
898 | ||
49f693fa DC |
899 | done: |
900 | *logflagsp = flags; | |
901 | return error; | |
2bd0ea18 NS |
902 | } |
903 | ||
904 | /* | |
49f693fa | 905 | * Called from xfs_bmap_add_attrfork to handle btree format files. |
2bd0ea18 | 906 | */ |
49f693fa DC |
907 | STATIC int /* error */ |
908 | xfs_bmap_add_attrfork_btree( | |
909 | xfs_trans_t *tp, /* transaction pointer */ | |
910 | xfs_inode_t *ip, /* incore inode pointer */ | |
911 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 912 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa | 913 | int *flags) /* inode logging flags */ |
2bd0ea18 | 914 | { |
49f693fa DC |
915 | xfs_btree_cur_t *cur; /* btree cursor */ |
916 | int error; /* error return value */ | |
917 | xfs_mount_t *mp; /* file system mount struct */ | |
918 | int stat; /* newroot status */ | |
56b2de80 | 919 | |
49f693fa DC |
920 | mp = ip->i_mount; |
921 | if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip)) | |
922 | *flags |= XFS_ILOG_DBROOT; | |
923 | else { | |
924 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
f33cea1a | 925 | cur->bc_private.b.dfops = dfops; |
49f693fa | 926 | cur->bc_private.b.firstblock = *firstblock; |
4f76f49c CH |
927 | error = xfs_bmbt_lookup_first(cur, &stat); |
928 | if (error) | |
49f693fa DC |
929 | goto error0; |
930 | /* must be at least one entry */ | |
19ebedcf | 931 | XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0); |
49f693fa DC |
932 | if ((error = xfs_btree_new_iroot(cur, flags, &stat))) |
933 | goto error0; | |
934 | if (stat == 0) { | |
935 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
12b53197 | 936 | return -ENOSPC; |
49f693fa DC |
937 | } |
938 | *firstblock = cur->bc_private.b.firstblock; | |
939 | cur->bc_private.b.allocated = 0; | |
940 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2bd0ea18 | 941 | } |
49f693fa DC |
942 | return 0; |
943 | error0: | |
944 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
945 | return error; | |
946 | } | |
56b2de80 | 947 | |
49f693fa DC |
948 | /* |
949 | * Called from xfs_bmap_add_attrfork to handle extents format files. | |
950 | */ | |
951 | STATIC int /* error */ | |
952 | xfs_bmap_add_attrfork_extents( | |
953 | xfs_trans_t *tp, /* transaction pointer */ | |
954 | xfs_inode_t *ip, /* incore inode pointer */ | |
955 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 956 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa DC |
957 | int *flags) /* inode logging flags */ |
958 | { | |
959 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
960 | int error; /* error return value */ | |
961 | ||
962 | if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip)) | |
963 | return 0; | |
964 | cur = NULL; | |
f33cea1a | 965 | error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0, |
49f693fa DC |
966 | flags, XFS_DATA_FORK); |
967 | if (cur) { | |
968 | cur->bc_private.b.allocated = 0; | |
969 | xfs_btree_del_cursor(cur, | |
970 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
2bd0ea18 | 971 | } |
49f693fa DC |
972 | return error; |
973 | } | |
56b2de80 | 974 | |
49f693fa DC |
975 | /* |
976 | * Called from xfs_bmap_add_attrfork to handle local format files. Each | |
977 | * different data fork content type needs a different callout to do the | |
978 | * conversion. Some are basic and only require special block initialisation | |
979 | * callouts for the data formating, others (directories) are so specialised they | |
980 | * handle everything themselves. | |
981 | * | |
982 | * XXX (dgc): investigate whether directory conversion can use the generic | |
983 | * formatting callout. It should be possible - it's just a very complex | |
5dfa5cd2 | 984 | * formatter. |
49f693fa DC |
985 | */ |
986 | STATIC int /* error */ | |
987 | xfs_bmap_add_attrfork_local( | |
988 | xfs_trans_t *tp, /* transaction pointer */ | |
989 | xfs_inode_t *ip, /* incore inode pointer */ | |
990 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 991 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa DC |
992 | int *flags) /* inode logging flags */ |
993 | { | |
994 | xfs_da_args_t dargs; /* args for dir/attr code */ | |
56b2de80 | 995 | |
49f693fa DC |
996 | if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip)) |
997 | return 0; | |
a2ceac1f | 998 | |
e37bf53c | 999 | if (S_ISDIR(VFS_I(ip)->i_mode)) { |
49f693fa | 1000 | memset(&dargs, 0, sizeof(dargs)); |
ff105f75 | 1001 | dargs.geo = ip->i_mount->m_dir_geo; |
49f693fa DC |
1002 | dargs.dp = ip; |
1003 | dargs.firstblock = firstblock; | |
f33cea1a | 1004 | dargs.dfops = dfops; |
ff105f75 | 1005 | dargs.total = dargs.geo->fsbcount; |
49f693fa DC |
1006 | dargs.whichfork = XFS_DATA_FORK; |
1007 | dargs.trans = tp; | |
1008 | return xfs_dir2_sf_to_block(&dargs); | |
1009 | } | |
2bd0ea18 | 1010 | |
e37bf53c | 1011 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
49f693fa DC |
1012 | return xfs_bmap_local_to_extents(tp, ip, firstblock, 1, |
1013 | flags, XFS_DATA_FORK, | |
1014 | xfs_symlink_local_to_remote); | |
56b2de80 | 1015 | |
3f17ed4b DC |
1016 | /* should only be called for types that support local format data */ |
1017 | ASSERT(0); | |
12b53197 | 1018 | return -EFSCORRUPTED; |
49f693fa | 1019 | } |
2bd0ea18 | 1020 | |
49f693fa DC |
1021 | /* |
1022 | * Convert inode from non-attributed to attributed. | |
1023 | * Must not be in a transaction, ip must not be locked. | |
1024 | */ | |
1025 | int /* error code */ | |
1026 | xfs_bmap_add_attrfork( | |
1027 | xfs_inode_t *ip, /* incore inode pointer */ | |
1028 | int size, /* space new attribute needs */ | |
1029 | int rsvd) /* xact may use reserved blks */ | |
1030 | { | |
1031 | xfs_fsblock_t firstblock; /* 1st block/ag allocated */ | |
f33cea1a | 1032 | struct xfs_defer_ops dfops; /* freed extent records */ |
49f693fa DC |
1033 | xfs_mount_t *mp; /* mount structure */ |
1034 | xfs_trans_t *tp; /* transaction pointer */ | |
1035 | int blks; /* space reservation */ | |
1036 | int version = 1; /* superblock attr version */ | |
49f693fa DC |
1037 | int logflags; /* logging flags */ |
1038 | int error; /* error return value */ | |
56b2de80 | 1039 | |
49f693fa | 1040 | ASSERT(XFS_IFORK_Q(ip) == 0); |
2bd0ea18 | 1041 | |
49f693fa DC |
1042 | mp = ip->i_mount; |
1043 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); | |
9074815c | 1044 | |
49f693fa | 1045 | blks = XFS_ADDAFORK_SPACE_RES(mp); |
9074815c CH |
1046 | |
1047 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0, | |
1048 | rsvd ? XFS_TRANS_RESERVE : 0, &tp); | |
1049 | if (error) | |
ff105f75 | 1050 | return error; |
9074815c | 1051 | |
49f693fa DC |
1052 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1053 | error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ? | |
1054 | XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES : | |
1055 | XFS_QMOPT_RES_REGBLKS); | |
ff105f75 DC |
1056 | if (error) |
1057 | goto trans_cancel; | |
49f693fa | 1058 | if (XFS_IFORK_Q(ip)) |
ff105f75 | 1059 | goto trans_cancel; |
9656ed10 DW |
1060 | if (ip->i_d.di_anextents != 0) { |
1061 | error = -EFSCORRUPTED; | |
1062 | goto trans_cancel; | |
1063 | } | |
49f693fa | 1064 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) { |
2bd0ea18 | 1065 | /* |
49f693fa | 1066 | * For inodes coming from pre-6.2 filesystems. |
2bd0ea18 | 1067 | */ |
49f693fa DC |
1068 | ASSERT(ip->i_d.di_aformat == 0); |
1069 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
5e656dbb | 1070 | } |
a2ceac1f | 1071 | |
ff105f75 | 1072 | xfs_trans_ijoin(tp, ip, 0); |
49f693fa | 1073 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
a2ceac1f | 1074 | |
49f693fa DC |
1075 | switch (ip->i_d.di_format) { |
1076 | case XFS_DINODE_FMT_DEV: | |
1077 | ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; | |
1078 | break; | |
49f693fa DC |
1079 | case XFS_DINODE_FMT_LOCAL: |
1080 | case XFS_DINODE_FMT_EXTENTS: | |
1081 | case XFS_DINODE_FMT_BTREE: | |
1082 | ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size); | |
1083 | if (!ip->i_d.di_forkoff) | |
1084 | ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3; | |
1085 | else if (mp->m_flags & XFS_MOUNT_ATTR2) | |
1086 | version = 2; | |
1087 | break; | |
1088 | default: | |
1089 | ASSERT(0); | |
12b53197 | 1090 | error = -EINVAL; |
ff105f75 | 1091 | goto trans_cancel; |
a2ceac1f DC |
1092 | } |
1093 | ||
49f693fa DC |
1094 | ASSERT(ip->i_afp == NULL); |
1095 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
1096 | ip->i_afp->if_flags = XFS_IFEXTENTS; | |
1097 | logflags = 0; | |
f33cea1a | 1098 | xfs_defer_init(&dfops, &firstblock); |
49f693fa DC |
1099 | switch (ip->i_d.di_format) { |
1100 | case XFS_DINODE_FMT_LOCAL: | |
f33cea1a | 1101 | error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops, |
49f693fa DC |
1102 | &logflags); |
1103 | break; | |
1104 | case XFS_DINODE_FMT_EXTENTS: | |
1105 | error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock, | |
f33cea1a | 1106 | &dfops, &logflags); |
49f693fa DC |
1107 | break; |
1108 | case XFS_DINODE_FMT_BTREE: | |
f33cea1a | 1109 | error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops, |
49f693fa DC |
1110 | &logflags); |
1111 | break; | |
1112 | default: | |
1113 | error = 0; | |
1114 | break; | |
1115 | } | |
1116 | if (logflags) | |
1117 | xfs_trans_log_inode(tp, ip, logflags); | |
1118 | if (error) | |
ff105f75 | 1119 | goto bmap_cancel; |
49f693fa DC |
1120 | if (!xfs_sb_version_hasattr(&mp->m_sb) || |
1121 | (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) { | |
19ebedcf | 1122 | bool log_sb = false; |
a2ceac1f | 1123 | |
49f693fa DC |
1124 | spin_lock(&mp->m_sb_lock); |
1125 | if (!xfs_sb_version_hasattr(&mp->m_sb)) { | |
1126 | xfs_sb_version_addattr(&mp->m_sb); | |
19ebedcf | 1127 | log_sb = true; |
49f693fa DC |
1128 | } |
1129 | if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) { | |
1130 | xfs_sb_version_addattr2(&mp->m_sb); | |
19ebedcf | 1131 | log_sb = true; |
49f693fa | 1132 | } |
19ebedcf DC |
1133 | spin_unlock(&mp->m_sb_lock); |
1134 | if (log_sb) | |
1135 | xfs_log_sb(tp); | |
49f693fa DC |
1136 | } |
1137 | ||
5c33baee | 1138 | error = xfs_defer_finish(&tp, &dfops); |
49f693fa | 1139 | if (error) |
ff105f75 | 1140 | goto bmap_cancel; |
de5a3f46 | 1141 | error = xfs_trans_commit(tp); |
ff105f75 DC |
1142 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1143 | return error; | |
1144 | ||
1145 | bmap_cancel: | |
f33cea1a | 1146 | xfs_defer_cancel(&dfops); |
ff105f75 | 1147 | trans_cancel: |
3d7434fe | 1148 | xfs_trans_cancel(tp); |
49f693fa | 1149 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5e656dbb | 1150 | return error; |
2bd0ea18 NS |
1151 | } |
1152 | ||
399ab595 | 1153 | /* |
49f693fa DC |
1154 | * Internal and external extent tree search functions. |
1155 | */ | |
399ab595 | 1156 | |
49f693fa DC |
1157 | /* |
1158 | * Read in the extents to if_extents. | |
1159 | * All inode fields are set up by caller, we just traverse the btree | |
1160 | * and copy the records in. If the file system cannot contain unwritten | |
1161 | * extents, the records are checked for no "state" flags. | |
1162 | */ | |
1163 | int /* error */ | |
1164 | xfs_bmap_read_extents( | |
1165 | xfs_trans_t *tp, /* transaction pointer */ | |
1166 | xfs_inode_t *ip, /* incore inode */ | |
1167 | int whichfork) /* data or attr fork */ | |
1168 | { | |
1169 | struct xfs_btree_block *block; /* current btree block */ | |
1170 | xfs_fsblock_t bno; /* block # of "block" */ | |
1171 | xfs_buf_t *bp; /* buffer for "block" */ | |
1172 | int error; /* error return value */ | |
49f693fa DC |
1173 | xfs_extnum_t i, j; /* index into the extents list */ |
1174 | xfs_ifork_t *ifp; /* fork structure */ | |
1175 | int level; /* btree level, for checking */ | |
1176 | xfs_mount_t *mp; /* file system mount structure */ | |
1177 | __be64 *pp; /* pointer to block address */ | |
1178 | /* REFERENCED */ | |
1179 | xfs_extnum_t room; /* number of entries there's room for */ | |
322fd804 | 1180 | int state = xfs_bmap_fork_to_state(whichfork); |
399ab595 | 1181 | |
49f693fa DC |
1182 | mp = ip->i_mount; |
1183 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
49f693fa | 1184 | block = ifp->if_broot; |
399ab595 | 1185 | /* |
49f693fa | 1186 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. |
399ab595 | 1187 | */ |
49f693fa DC |
1188 | level = be16_to_cpu(block->bb_level); |
1189 | ASSERT(level > 0); | |
1190 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
1191 | bno = be64_to_cpu(*pp); | |
e2376544 | 1192 | |
399ab595 | 1193 | /* |
49f693fa DC |
1194 | * Go down the tree until leaf level is reached, following the first |
1195 | * pointer (leftmost) at each level. | |
399ab595 | 1196 | */ |
49f693fa DC |
1197 | while (level-- > 0) { |
1198 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1199 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1200 | if (error) | |
1201 | return error; | |
1202 | block = XFS_BUF_TO_BLOCK(bp); | |
49f693fa DC |
1203 | if (level == 0) |
1204 | break; | |
1205 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
1206 | bno = be64_to_cpu(*pp); | |
19ebedcf DC |
1207 | XFS_WANT_CORRUPTED_GOTO(mp, |
1208 | XFS_FSB_SANITY_CHECK(mp, bno), error0); | |
49f693fa | 1209 | xfs_trans_brelse(tp, bp); |
399ab595 NS |
1210 | } |
1211 | /* | |
49f693fa | 1212 | * Here with bp and block set to the leftmost leaf node in the tree. |
399ab595 | 1213 | */ |
d09d4e5f | 1214 | room = xfs_iext_count(ifp); |
49f693fa | 1215 | i = 0; |
399ab595 | 1216 | /* |
49f693fa | 1217 | * Loop over all leaf nodes. Copy information to the extent records. |
399ab595 | 1218 | */ |
49f693fa DC |
1219 | for (;;) { |
1220 | xfs_bmbt_rec_t *frp; | |
1221 | xfs_fsblock_t nextbno; | |
1222 | xfs_extnum_t num_recs; | |
399ab595 | 1223 | |
49f693fa DC |
1224 | num_recs = xfs_btree_get_numrecs(block); |
1225 | if (unlikely(i + num_recs > room)) { | |
1226 | ASSERT(i + num_recs <= room); | |
1227 | xfs_warn(ip->i_mount, | |
1228 | "corrupt dinode %Lu, (btree extents).", | |
1229 | (unsigned long long) ip->i_ino); | |
1230 | XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)", | |
1231 | XFS_ERRLEVEL_LOW, ip->i_mount, block); | |
1232 | goto error0; | |
399ab595 NS |
1233 | } |
1234 | /* | |
49f693fa | 1235 | * Read-ahead the next leaf block, if any. |
399ab595 | 1236 | */ |
49f693fa DC |
1237 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
1238 | if (nextbno != NULLFSBLOCK) | |
1239 | xfs_btree_reada_bufl(mp, nextbno, 1, | |
1240 | &xfs_bmbt_buf_ops); | |
399ab595 | 1241 | /* |
49f693fa | 1242 | * Copy records into the extent records. |
399ab595 | 1243 | */ |
49f693fa | 1244 | frp = XFS_BMBT_REC_ADDR(mp, block, 1); |
49f693fa DC |
1245 | for (j = 0; j < num_recs; j++, i++, frp++) { |
1246 | xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i); | |
1247 | trp->l0 = be64_to_cpu(frp->l0); | |
1248 | trp->l1 = be64_to_cpu(frp->l1); | |
1f2a4478 | 1249 | if (!xfs_bmbt_validate_extent(mp, whichfork, trp)) { |
49f693fa | 1250 | XFS_ERROR_REPORT("xfs_bmap_read_extents(2)", |
1f2a4478 | 1251 | XFS_ERRLEVEL_LOW, mp); |
49f693fa DC |
1252 | goto error0; |
1253 | } | |
322fd804 | 1254 | trace_xfs_read_extent(ip, i, state, _THIS_IP_); |
399ab595 | 1255 | } |
49f693fa DC |
1256 | xfs_trans_brelse(tp, bp); |
1257 | bno = nextbno; | |
399ab595 | 1258 | /* |
49f693fa | 1259 | * If we've reached the end, stop. |
399ab595 | 1260 | */ |
49f693fa DC |
1261 | if (bno == NULLFSBLOCK) |
1262 | break; | |
1263 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1264 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1265 | if (error) | |
1266 | return error; | |
1267 | block = XFS_BUF_TO_BLOCK(bp); | |
399ab595 | 1268 | } |
43a48f63 DW |
1269 | if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) |
1270 | return -EFSCORRUPTED; | |
d09d4e5f | 1271 | ASSERT(i == xfs_iext_count(ifp)); |
49f693fa DC |
1272 | return 0; |
1273 | error0: | |
1274 | xfs_trans_brelse(tp, bp); | |
12b53197 | 1275 | return -EFSCORRUPTED; |
49f693fa | 1276 | } |
399ab595 | 1277 | |
49f693fa DC |
1278 | /* |
1279 | * Returns the file-relative block number of the first unused block(s) | |
1280 | * in the file with at least "len" logically contiguous blocks free. | |
1281 | * This is the lowest-address hole if the file has holes, else the first block | |
1282 | * past the end of file. | |
1283 | * Return 0 if the file is currently local (in-inode). | |
1284 | */ | |
1285 | int /* error */ | |
1286 | xfs_bmap_first_unused( | |
1287 | xfs_trans_t *tp, /* transaction pointer */ | |
1288 | xfs_inode_t *ip, /* incore inode */ | |
1289 | xfs_extlen_t len, /* size of hole to find */ | |
1290 | xfs_fileoff_t *first_unused, /* unused block */ | |
1291 | int whichfork) /* data or attr fork */ | |
1292 | { | |
1293 | int error; /* error return value */ | |
1294 | int idx; /* extent record index */ | |
1295 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1296 | xfs_fileoff_t lastaddr; /* last block number seen */ | |
1297 | xfs_fileoff_t lowest; /* lowest useful block */ | |
1298 | xfs_fileoff_t max; /* starting useful block */ | |
49f693fa DC |
1299 | xfs_extnum_t nextents; /* number of extent entries */ |
1300 | ||
1301 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE || | |
1302 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS || | |
1303 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
1304 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
1305 | *first_unused = 0; | |
1306 | return 0; | |
1307 | } | |
1308 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1309 | if (!(ifp->if_flags & XFS_IFEXTENTS) && | |
1310 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
1311 | return error; | |
1312 | lowest = *first_unused; | |
d09d4e5f | 1313 | nextents = xfs_iext_count(ifp); |
49f693fa | 1314 | for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) { |
f71f3bc3 CH |
1315 | struct xfs_bmbt_irec got; |
1316 | ||
1317 | xfs_iext_get_extent(ifp, idx, &got); | |
1318 | ||
2bd0ea18 | 1319 | /* |
49f693fa | 1320 | * See if the hole before this extent will work. |
2bd0ea18 | 1321 | */ |
f71f3bc3 CH |
1322 | if (got.br_startoff >= lowest + len && |
1323 | got.br_startoff - max >= len) { | |
49f693fa DC |
1324 | *first_unused = max; |
1325 | return 0; | |
2bd0ea18 | 1326 | } |
f71f3bc3 | 1327 | lastaddr = got.br_startoff + got.br_blockcount; |
49f693fa DC |
1328 | max = XFS_FILEOFF_MAX(lastaddr, lowest); |
1329 | } | |
1330 | *first_unused = max; | |
1331 | return 0; | |
1332 | } | |
1333 | ||
1334 | /* | |
e6d77a21 | 1335 | * Returns the file-relative block number of the last block - 1 before |
49f693fa DC |
1336 | * last_block (input value) in the file. |
1337 | * This is not based on i_size, it is based on the extent records. | |
1338 | * Returns 0 for local files, as they do not have extent records. | |
1339 | */ | |
1340 | int /* error */ | |
1341 | xfs_bmap_last_before( | |
b0385364 CH |
1342 | struct xfs_trans *tp, /* transaction pointer */ |
1343 | struct xfs_inode *ip, /* incore inode */ | |
1344 | xfs_fileoff_t *last_block, /* last block */ | |
1345 | int whichfork) /* data or attr fork */ | |
49f693fa | 1346 | { |
b0385364 CH |
1347 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1348 | struct xfs_bmbt_irec got; | |
1349 | xfs_extnum_t idx; | |
1350 | int error; | |
49f693fa | 1351 | |
b0385364 CH |
1352 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { |
1353 | case XFS_DINODE_FMT_LOCAL: | |
49f693fa DC |
1354 | *last_block = 0; |
1355 | return 0; | |
b0385364 CH |
1356 | case XFS_DINODE_FMT_BTREE: |
1357 | case XFS_DINODE_FMT_EXTENTS: | |
1358 | break; | |
1359 | default: | |
1360 | return -EIO; | |
49f693fa | 1361 | } |
b0385364 CH |
1362 | |
1363 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1364 | error = xfs_iread_extents(tp, ip, whichfork); | |
1365 | if (error) | |
1366 | return error; | |
2bd0ea18 | 1367 | } |
b0385364 CH |
1368 | |
1369 | if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) { | |
1370 | if (got.br_startoff <= *last_block - 1) | |
1371 | return 0; | |
1372 | } | |
1373 | ||
1374 | if (xfs_iext_get_extent(ifp, idx - 1, &got)) { | |
1375 | *last_block = got.br_startoff + got.br_blockcount; | |
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | *last_block = 0; | |
49f693fa | 1380 | return 0; |
5e656dbb BN |
1381 | } |
1382 | ||
613e6057 | 1383 | int |
49f693fa DC |
1384 | xfs_bmap_last_extent( |
1385 | struct xfs_trans *tp, | |
1386 | struct xfs_inode *ip, | |
1387 | int whichfork, | |
1388 | struct xfs_bmbt_irec *rec, | |
1389 | int *is_empty) | |
56b2de80 | 1390 | { |
49f693fa | 1391 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
56b2de80 | 1392 | int error; |
49f693fa | 1393 | int nextents; |
56b2de80 | 1394 | |
49f693fa DC |
1395 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
1396 | error = xfs_iread_extents(tp, ip, whichfork); | |
1397 | if (error) | |
1398 | return error; | |
1399 | } | |
1400 | ||
d09d4e5f | 1401 | nextents = xfs_iext_count(ifp); |
49f693fa DC |
1402 | if (nextents == 0) { |
1403 | *is_empty = 1; | |
1404 | return 0; | |
1405 | } | |
1406 | ||
99a663c1 | 1407 | xfs_iext_get_extent(ifp, nextents - 1, rec); |
49f693fa DC |
1408 | *is_empty = 0; |
1409 | return 0; | |
1410 | } | |
1411 | ||
1412 | /* | |
1413 | * Check the last inode extent to determine whether this allocation will result | |
1414 | * in blocks being allocated at the end of the file. When we allocate new data | |
1415 | * blocks at the end of the file which do not start at the previous data block, | |
1416 | * we will try to align the new blocks at stripe unit boundaries. | |
1417 | * | |
ff105f75 | 1418 | * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be |
49f693fa DC |
1419 | * at, or past the EOF. |
1420 | */ | |
1421 | STATIC int | |
1422 | xfs_bmap_isaeof( | |
1423 | struct xfs_bmalloca *bma, | |
1424 | int whichfork) | |
1425 | { | |
1426 | struct xfs_bmbt_irec rec; | |
1427 | int is_empty; | |
1428 | int error; | |
1429 | ||
180d3cd8 | 1430 | bma->aeof = false; |
49f693fa DC |
1431 | error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec, |
1432 | &is_empty); | |
ff105f75 | 1433 | if (error) |
49f693fa | 1434 | return error; |
56b2de80 | 1435 | |
ff105f75 | 1436 | if (is_empty) { |
180d3cd8 | 1437 | bma->aeof = true; |
ff105f75 DC |
1438 | return 0; |
1439 | } | |
1440 | ||
56b2de80 | 1441 | /* |
49f693fa DC |
1442 | * Check if we are allocation or past the last extent, or at least into |
1443 | * the last delayed allocated extent. | |
56b2de80 | 1444 | */ |
49f693fa DC |
1445 | bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount || |
1446 | (bma->offset >= rec.br_startoff && | |
1447 | isnullstartblock(rec.br_startblock)); | |
1448 | return 0; | |
1449 | } | |
56b2de80 | 1450 | |
49f693fa DC |
1451 | /* |
1452 | * Returns the file-relative block number of the first block past eof in | |
1453 | * the file. This is not based on i_size, it is based on the extent records. | |
1454 | * Returns 0 for local files, as they do not have extent records. | |
1455 | */ | |
1456 | int | |
1457 | xfs_bmap_last_offset( | |
49f693fa DC |
1458 | struct xfs_inode *ip, |
1459 | xfs_fileoff_t *last_block, | |
1460 | int whichfork) | |
1461 | { | |
1462 | struct xfs_bmbt_irec rec; | |
1463 | int is_empty; | |
1464 | int error; | |
1465 | ||
1466 | *last_block = 0; | |
1467 | ||
1468 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) | |
1469 | return 0; | |
1470 | ||
1471 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE && | |
1472 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
12b53197 | 1473 | return -EIO; |
49f693fa DC |
1474 | |
1475 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty); | |
1476 | if (error || is_empty) | |
1477 | return error; | |
1478 | ||
1479 | *last_block = rec.br_startoff + rec.br_blockcount; | |
1480 | return 0; | |
1481 | } | |
1482 | ||
1483 | /* | |
1484 | * Returns whether the selected fork of the inode has exactly one | |
1485 | * block or not. For the data fork we check this matches di_size, | |
1486 | * implying the file's range is 0..bsize-1. | |
1487 | */ | |
1488 | int /* 1=>1 block, 0=>otherwise */ | |
1489 | xfs_bmap_one_block( | |
1490 | xfs_inode_t *ip, /* incore inode */ | |
1491 | int whichfork) /* data or attr fork */ | |
1492 | { | |
49f693fa DC |
1493 | xfs_ifork_t *ifp; /* inode fork pointer */ |
1494 | int rval; /* return value */ | |
1495 | xfs_bmbt_irec_t s; /* internal version of extent */ | |
1496 | ||
1497 | #ifndef DEBUG | |
1498 | if (whichfork == XFS_DATA_FORK) | |
1499 | return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize; | |
1500 | #endif /* !DEBUG */ | |
1501 | if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1) | |
1502 | return 0; | |
1503 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
1504 | return 0; | |
1505 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1506 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
99a663c1 | 1507 | xfs_iext_get_extent(ifp, 0, &s); |
49f693fa DC |
1508 | rval = s.br_startoff == 0 && s.br_blockcount == 1; |
1509 | if (rval && whichfork == XFS_DATA_FORK) | |
1510 | ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize); | |
1511 | return rval; | |
1512 | } | |
1513 | ||
1514 | /* | |
1515 | * Extent tree manipulation functions used during allocation. | |
1516 | */ | |
1517 | ||
1518 | /* | |
1519 | * Convert a delayed allocation to a real allocation. | |
1520 | */ | |
1521 | STATIC int /* error */ | |
1522 | xfs_bmap_add_extent_delay_real( | |
1277a5e0 DW |
1523 | struct xfs_bmalloca *bma, |
1524 | int whichfork) | |
49f693fa DC |
1525 | { |
1526 | struct xfs_bmbt_irec *new = &bma->got; | |
49f693fa DC |
1527 | int error; /* error return value */ |
1528 | int i; /* temp state */ | |
1529 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1530 | xfs_fileoff_t new_endoff; /* end offset of new entry */ | |
1531 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
1532 | /* left is 0, right is 1, prev is 2 */ | |
1533 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 1534 | int state = xfs_bmap_fork_to_state(whichfork); |
49f693fa DC |
1535 | xfs_filblks_t da_new; /* new count del alloc blocks used */ |
1536 | xfs_filblks_t da_old; /* old count del alloc blocks used */ | |
1537 | xfs_filblks_t temp=0; /* value for da_new calculations */ | |
49f693fa | 1538 | int tmp_rval; /* partial logging flags */ |
19ebedcf | 1539 | struct xfs_mount *mp; |
1277a5e0 | 1540 | xfs_extnum_t *nextents; |
19884717 | 1541 | struct xfs_bmbt_irec old; |
49f693fa | 1542 | |
65ca3804 | 1543 | mp = bma->ip->i_mount; |
36e8786d | 1544 | ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
1277a5e0 DW |
1545 | ASSERT(whichfork != XFS_ATTR_FORK); |
1546 | nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents : | |
1547 | &bma->ip->i_d.di_nextents); | |
56b2de80 | 1548 | |
49f693fa | 1549 | ASSERT(bma->idx >= 0); |
d09d4e5f | 1550 | ASSERT(bma->idx <= xfs_iext_count(ifp)); |
49f693fa DC |
1551 | ASSERT(!isnullstartblock(new->br_startblock)); |
1552 | ASSERT(!bma->cur || | |
1553 | (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); | |
56b2de80 | 1554 | |
79896434 | 1555 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
1556 | |
1557 | #define LEFT r[0] | |
1558 | #define RIGHT r[1] | |
1559 | #define PREV r[2] | |
56b2de80 DC |
1560 | |
1561 | /* | |
49f693fa | 1562 | * Set up a bunch of variables to make the tests simpler. |
56b2de80 | 1563 | */ |
19884717 | 1564 | xfs_iext_get_extent(ifp, bma->idx, &PREV); |
49f693fa | 1565 | new_endoff = new->br_startoff + new->br_blockcount; |
19884717 | 1566 | ASSERT(isnullstartblock(PREV.br_startblock)); |
49f693fa DC |
1567 | ASSERT(PREV.br_startoff <= new->br_startoff); |
1568 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
1569 | ||
1570 | da_old = startblockval(PREV.br_startblock); | |
1571 | da_new = 0; | |
1572 | ||
56b2de80 | 1573 | /* |
49f693fa DC |
1574 | * Set flags determining what part of the previous delayed allocation |
1575 | * extent is being replaced by a real allocation. | |
56b2de80 | 1576 | */ |
49f693fa DC |
1577 | if (PREV.br_startoff == new->br_startoff) |
1578 | state |= BMAP_LEFT_FILLING; | |
1579 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
1580 | state |= BMAP_RIGHT_FILLING; | |
1581 | ||
56b2de80 | 1582 | /* |
49f693fa DC |
1583 | * Check and set flags if this segment has a left neighbor. |
1584 | * Don't set contiguous if the combined extent would be too large. | |
56b2de80 | 1585 | */ |
49f693fa DC |
1586 | if (bma->idx > 0) { |
1587 | state |= BMAP_LEFT_VALID; | |
99a663c1 | 1588 | xfs_iext_get_extent(ifp, bma->idx - 1, &LEFT); |
49f693fa DC |
1589 | |
1590 | if (isnullstartblock(LEFT.br_startblock)) | |
1591 | state |= BMAP_LEFT_DELAY; | |
1592 | } | |
1593 | ||
1594 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
1595 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
1596 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
1597 | LEFT.br_state == new->br_state && | |
1598 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
1599 | state |= BMAP_LEFT_CONTIG; | |
56b2de80 DC |
1600 | |
1601 | /* | |
49f693fa DC |
1602 | * Check and set flags if this segment has a right neighbor. |
1603 | * Don't set contiguous if the combined extent would be too large. | |
1604 | * Also check for all-three-contiguous being too large. | |
56b2de80 | 1605 | */ |
d09d4e5f | 1606 | if (bma->idx < xfs_iext_count(ifp) - 1) { |
49f693fa | 1607 | state |= BMAP_RIGHT_VALID; |
99a663c1 | 1608 | xfs_iext_get_extent(ifp, bma->idx + 1, &RIGHT); |
56b2de80 | 1609 | |
49f693fa DC |
1610 | if (isnullstartblock(RIGHT.br_startblock)) |
1611 | state |= BMAP_RIGHT_DELAY; | |
1612 | } | |
56b2de80 | 1613 | |
49f693fa DC |
1614 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
1615 | new_endoff == RIGHT.br_startoff && | |
1616 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
1617 | new->br_state == RIGHT.br_state && | |
1618 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
1619 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1620 | BMAP_RIGHT_FILLING)) != | |
1621 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1622 | BMAP_RIGHT_FILLING) || | |
1623 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
1624 | <= MAXEXTLEN)) | |
1625 | state |= BMAP_RIGHT_CONTIG; | |
5e656dbb | 1626 | |
49f693fa DC |
1627 | error = 0; |
1628 | /* | |
1629 | * Switch out based on the FILLING and CONTIG state bits. | |
1630 | */ | |
1631 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1632 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
1633 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1634 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1635 | /* | |
1636 | * Filling in all of a previously delayed allocation extent. | |
1637 | * The left and right neighbors are both contiguous with new. | |
1638 | */ | |
1639 | bma->idx--; | |
19884717 | 1640 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
df926c07 | 1641 | xfs_iext_update_extent(bma->ip, state, bma->idx, &LEFT); |
a2ceac1f | 1642 | |
49f693fa | 1643 | xfs_iext_remove(bma->ip, bma->idx + 1, 2, state); |
1277a5e0 | 1644 | (*nextents)--; |
49f693fa DC |
1645 | if (bma->cur == NULL) |
1646 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1647 | else { | |
1648 | rval = XFS_ILOG_CORE; | |
70a93110 | 1649 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
49f693fa DC |
1650 | if (error) |
1651 | goto done; | |
19ebedcf | 1652 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1653 | error = xfs_btree_delete(bma->cur, &i); |
1654 | if (error) | |
1655 | goto done; | |
19ebedcf | 1656 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1657 | error = xfs_btree_decrement(bma->cur, 0, &i); |
1658 | if (error) | |
1659 | goto done; | |
19ebedcf | 1660 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1661 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1662 | if (error) |
1663 | goto done; | |
5e656dbb | 1664 | } |
49f693fa DC |
1665 | break; |
1666 | ||
1667 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1668 | /* | |
1669 | * Filling in all of a previously delayed allocation extent. | |
1670 | * The left neighbor is contiguous, the right is not. | |
1671 | */ | |
1672 | bma->idx--; | |
1673 | ||
19884717 | 1674 | old = LEFT; |
19884717 | 1675 | LEFT.br_blockcount += PREV.br_blockcount; |
df926c07 | 1676 | xfs_iext_update_extent(bma->ip, state, bma->idx, &LEFT); |
49f693fa DC |
1677 | |
1678 | xfs_iext_remove(bma->ip, bma->idx + 1, 1, state); | |
1679 | if (bma->cur == NULL) | |
1680 | rval = XFS_ILOG_DEXT; | |
1681 | else { | |
1682 | rval = 0; | |
70a93110 | 1683 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1684 | if (error) |
1685 | goto done; | |
19ebedcf | 1686 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1687 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1688 | if (error) |
1689 | goto done; | |
1690 | } | |
1691 | break; | |
1692 | ||
1693 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1694 | /* | |
1695 | * Filling in all of a previously delayed allocation extent. | |
1696 | * The right neighbor is contiguous, the left is not. | |
1697 | */ | |
19884717 CH |
1698 | PREV.br_startblock = new->br_startblock; |
1699 | PREV.br_blockcount += RIGHT.br_blockcount; | |
df926c07 | 1700 | xfs_iext_update_extent(bma->ip, state, bma->idx, &PREV); |
49f693fa DC |
1701 | |
1702 | xfs_iext_remove(bma->ip, bma->idx + 1, 1, state); | |
1703 | if (bma->cur == NULL) | |
1704 | rval = XFS_ILOG_DEXT; | |
1705 | else { | |
1706 | rval = 0; | |
70a93110 | 1707 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
49f693fa DC |
1708 | if (error) |
1709 | goto done; | |
19ebedcf | 1710 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1711 | error = xfs_bmbt_update(bma->cur, &PREV); |
49f693fa DC |
1712 | if (error) |
1713 | goto done; | |
1714 | } | |
1715 | break; | |
1716 | ||
1717 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
1718 | /* | |
1719 | * Filling in all of a previously delayed allocation extent. | |
1720 | * Neither the left nor right neighbors are contiguous with | |
1721 | * the new one. | |
1722 | */ | |
19884717 CH |
1723 | PREV.br_startblock = new->br_startblock; |
1724 | PREV.br_state = new->br_state; | |
df926c07 | 1725 | xfs_iext_update_extent(bma->ip, state, bma->idx, &PREV); |
5e656dbb | 1726 | |
1277a5e0 | 1727 | (*nextents)++; |
49f693fa DC |
1728 | if (bma->cur == NULL) |
1729 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1730 | else { | |
1731 | rval = XFS_ILOG_CORE; | |
70a93110 | 1732 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1733 | if (error) |
1734 | goto done; | |
19ebedcf | 1735 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1736 | error = xfs_btree_insert(bma->cur, &i); |
1737 | if (error) | |
1738 | goto done; | |
19ebedcf | 1739 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1740 | } |
1741 | break; | |
5e656dbb | 1742 | |
49f693fa DC |
1743 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
1744 | /* | |
1745 | * Filling in the first part of a previous delayed allocation. | |
1746 | * The left neighbor is contiguous. | |
1747 | */ | |
19884717 CH |
1748 | old = LEFT; |
1749 | temp = PREV.br_blockcount - new->br_blockcount; | |
1750 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
1751 | startblockval(PREV.br_startblock)); | |
1752 | ||
19884717 | 1753 | LEFT.br_blockcount += new->br_blockcount; |
df926c07 | 1754 | xfs_iext_update_extent(bma->ip, state, bma->idx - 1, &LEFT); |
a2ceac1f | 1755 | |
19884717 CH |
1756 | PREV.br_blockcount = temp = PREV.br_blockcount - new->br_blockcount; |
1757 | PREV.br_startoff += new->br_blockcount; | |
1758 | PREV.br_startblock = nullstartblock(da_new); | |
df926c07 | 1759 | xfs_iext_update_extent(bma->ip, state, bma->idx, &PREV); |
19884717 | 1760 | |
49f693fa DC |
1761 | if (bma->cur == NULL) |
1762 | rval = XFS_ILOG_DEXT; | |
1763 | else { | |
1764 | rval = 0; | |
70a93110 | 1765 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1766 | if (error) |
1767 | goto done; | |
19ebedcf | 1768 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1769 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1770 | if (error) |
1771 | goto done; | |
2bd0ea18 | 1772 | } |
49f693fa DC |
1773 | |
1774 | bma->idx--; | |
1775 | break; | |
1776 | ||
1777 | case BMAP_LEFT_FILLING: | |
2bd0ea18 | 1778 | /* |
49f693fa DC |
1779 | * Filling in the first part of a previous delayed allocation. |
1780 | * The left neighbor is not contiguous. | |
5000d01d | 1781 | */ |
49f693fa | 1782 | xfs_iext_insert(bma->ip, bma->idx, 1, new, state); |
1277a5e0 | 1783 | (*nextents)++; |
49f693fa DC |
1784 | if (bma->cur == NULL) |
1785 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1786 | else { | |
1787 | rval = XFS_ILOG_CORE; | |
70a93110 | 1788 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1789 | if (error) |
1790 | goto done; | |
19ebedcf | 1791 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1792 | error = xfs_btree_insert(bma->cur, &i); |
1793 | if (error) | |
1794 | goto done; | |
19ebedcf | 1795 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1796 | } |
1797 | ||
36e8786d | 1798 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1799 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 1800 | bma->firstblock, bma->dfops, |
36e8786d | 1801 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1802 | rval |= tmp_rval; |
1803 | if (error) | |
1804 | goto done; | |
1805 | } | |
19884717 CH |
1806 | |
1807 | temp = PREV.br_blockcount - new->br_blockcount; | |
49f693fa DC |
1808 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1809 | startblockval(PREV.br_startblock) - | |
1810 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
19884717 | 1811 | |
19884717 CH |
1812 | PREV.br_startoff = new_endoff; |
1813 | PREV.br_blockcount = temp; | |
1814 | PREV.br_startblock = nullstartblock(da_new); | |
df926c07 | 1815 | xfs_iext_update_extent(bma->ip, state, bma->idx + 1, &PREV); |
49f693fa DC |
1816 | break; |
1817 | ||
1818 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
5e656dbb | 1819 | /* |
49f693fa DC |
1820 | * Filling in the last part of a previous delayed allocation. |
1821 | * The right neighbor is contiguous with the new allocation. | |
5e656dbb | 1822 | */ |
19884717 | 1823 | old = RIGHT; |
19884717 CH |
1824 | RIGHT.br_startoff = new->br_startoff; |
1825 | RIGHT.br_startblock = new->br_startblock; | |
1826 | RIGHT.br_blockcount += new->br_blockcount; | |
df926c07 | 1827 | xfs_iext_update_extent(bma->ip, state, bma->idx + 1, &RIGHT); |
19884717 | 1828 | |
49f693fa DC |
1829 | if (bma->cur == NULL) |
1830 | rval = XFS_ILOG_DEXT; | |
1831 | else { | |
1832 | rval = 0; | |
70a93110 | 1833 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1834 | if (error) |
1835 | goto done; | |
19ebedcf | 1836 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1837 | error = xfs_bmbt_update(bma->cur, &RIGHT); |
49f693fa DC |
1838 | if (error) |
1839 | goto done; | |
1840 | } | |
1841 | ||
19884717 | 1842 | temp = PREV.br_blockcount - new->br_blockcount; |
49f693fa DC |
1843 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1844 | startblockval(PREV.br_startblock)); | |
19884717 | 1845 | |
19884717 CH |
1846 | PREV.br_blockcount = temp; |
1847 | PREV.br_startblock = nullstartblock(da_new); | |
df926c07 | 1848 | xfs_iext_update_extent(bma->ip, state, bma->idx, &PREV); |
49f693fa DC |
1849 | |
1850 | bma->idx++; | |
1851 | break; | |
1852 | ||
1853 | case BMAP_RIGHT_FILLING: | |
a2ceac1f | 1854 | /* |
49f693fa DC |
1855 | * Filling in the last part of a previous delayed allocation. |
1856 | * The right neighbor is not contiguous. | |
a2ceac1f | 1857 | */ |
49f693fa | 1858 | xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state); |
1277a5e0 | 1859 | (*nextents)++; |
49f693fa DC |
1860 | if (bma->cur == NULL) |
1861 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1862 | else { | |
1863 | rval = XFS_ILOG_CORE; | |
70a93110 | 1864 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1865 | if (error) |
1866 | goto done; | |
19ebedcf | 1867 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1868 | error = xfs_btree_insert(bma->cur, &i); |
1869 | if (error) | |
1870 | goto done; | |
19ebedcf | 1871 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 1872 | } |
a2ceac1f | 1873 | |
36e8786d | 1874 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1875 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 1876 | bma->firstblock, bma->dfops, &bma->cur, 1, |
36e8786d | 1877 | &tmp_rval, whichfork); |
49f693fa DC |
1878 | rval |= tmp_rval; |
1879 | if (error) | |
1880 | goto done; | |
1881 | } | |
19884717 CH |
1882 | |
1883 | temp = PREV.br_blockcount - new->br_blockcount; | |
49f693fa DC |
1884 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1885 | startblockval(PREV.br_startblock) - | |
1886 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
19884717 | 1887 | |
19884717 CH |
1888 | PREV.br_startblock = nullstartblock(da_new); |
1889 | PREV.br_blockcount = temp; | |
df926c07 | 1890 | xfs_iext_update_extent(bma->ip, state, bma->idx, &PREV); |
49f693fa DC |
1891 | |
1892 | bma->idx++; | |
1893 | break; | |
1894 | ||
1895 | case 0: | |
5e656dbb | 1896 | /* |
49f693fa DC |
1897 | * Filling in the middle part of a previous delayed allocation. |
1898 | * Contiguity is impossible here. | |
1899 | * This case is avoided almost all the time. | |
1900 | * | |
1901 | * We start with a delayed allocation: | |
1902 | * | |
1903 | * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+ | |
1904 | * PREV @ idx | |
1905 | * | |
1906 | * and we are allocating: | |
1907 | * +rrrrrrrrrrrrrrrrr+ | |
1908 | * new | |
1909 | * | |
1910 | * and we set it up for insertion as: | |
1911 | * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+ | |
1912 | * new | |
1913 | * PREV @ idx LEFT RIGHT | |
1914 | * inserted at idx + 1 | |
5e656dbb | 1915 | */ |
19884717 CH |
1916 | old = PREV; |
1917 | ||
1918 | /* LEFT is the new middle */ | |
49f693fa | 1919 | LEFT = *new; |
19884717 CH |
1920 | |
1921 | /* RIGHT is the new right */ | |
49f693fa | 1922 | RIGHT.br_state = PREV.br_state; |
49f693fa | 1923 | RIGHT.br_startoff = new_endoff; |
19884717 CH |
1924 | RIGHT.br_blockcount = |
1925 | PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
1926 | RIGHT.br_startblock = | |
1927 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1928 | RIGHT.br_blockcount)); | |
1929 | ||
1930 | /* truncate PREV */ | |
19884717 CH |
1931 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
1932 | PREV.br_startblock = | |
1933 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1934 | PREV.br_blockcount)); | |
df926c07 | 1935 | xfs_iext_update_extent(bma->ip, state, bma->idx, &PREV); |
19884717 | 1936 | |
49f693fa DC |
1937 | /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */ |
1938 | xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state); | |
1277a5e0 | 1939 | (*nextents)++; |
19884717 | 1940 | |
49f693fa DC |
1941 | if (bma->cur == NULL) |
1942 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1943 | else { | |
1944 | rval = XFS_ILOG_CORE; | |
70a93110 | 1945 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1946 | if (error) |
1947 | goto done; | |
19ebedcf | 1948 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1949 | error = xfs_btree_insert(bma->cur, &i); |
1950 | if (error) | |
1951 | goto done; | |
19ebedcf | 1952 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 1953 | } |
5e656dbb | 1954 | |
36e8786d | 1955 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1956 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 1957 | bma->firstblock, bma->dfops, &bma->cur, |
36e8786d | 1958 | 1, &tmp_rval, whichfork); |
49f693fa DC |
1959 | rval |= tmp_rval; |
1960 | if (error) | |
1961 | goto done; | |
1962 | } | |
19884717 CH |
1963 | |
1964 | da_new = startblockval(PREV.br_startblock) + | |
1965 | startblockval(RIGHT.br_startblock); | |
49f693fa | 1966 | bma->idx++; |
49f693fa DC |
1967 | break; |
1968 | ||
1969 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1970 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1971 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
1972 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1973 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1974 | case BMAP_LEFT_CONTIG: | |
1975 | case BMAP_RIGHT_CONTIG: | |
1976 | /* | |
1977 | * These cases are all impossible. | |
1978 | */ | |
1979 | ASSERT(0); | |
1980 | } | |
1981 | ||
d7f80320 DW |
1982 | /* add reverse mapping */ |
1983 | error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new); | |
1984 | if (error) | |
1985 | goto done; | |
1986 | ||
49f693fa | 1987 | /* convert to a btree if necessary */ |
36e8786d | 1988 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa DC |
1989 | int tmp_logflags; /* partial log flag return val */ |
1990 | ||
1991 | ASSERT(bma->cur == NULL); | |
1992 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, | |
f33cea1a | 1993 | bma->firstblock, bma->dfops, &bma->cur, |
36e8786d | 1994 | da_old > 0, &tmp_logflags, whichfork); |
49f693fa DC |
1995 | bma->logflags |= tmp_logflags; |
1996 | if (error) | |
1997 | goto done; | |
1998 | } | |
1999 | ||
5bfca476 CH |
2000 | if (bma->cur) { |
2001 | da_new += bma->cur->bc_private.b.allocated; | |
2002 | bma->cur->bc_private.b.allocated = 0; | |
49f693fa DC |
2003 | } |
2004 | ||
5bfca476 CH |
2005 | /* adjust for changes in reserved delayed indirect blocks */ |
2006 | if (da_new != da_old) { | |
2007 | ASSERT(state == 0 || da_new < da_old); | |
2008 | error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), | |
2009 | false); | |
2010 | } | |
49f693fa | 2011 | |
36e8786d | 2012 | xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork); |
49f693fa | 2013 | done: |
1277a5e0 DW |
2014 | if (whichfork != XFS_COW_FORK) |
2015 | bma->logflags |= rval; | |
49f693fa DC |
2016 | return error; |
2017 | #undef LEFT | |
2018 | #undef RIGHT | |
2019 | #undef PREV | |
2bd0ea18 NS |
2020 | } |
2021 | ||
2022 | /* | |
49f693fa | 2023 | * Convert an unwritten allocation to a real allocation or vice versa. |
2bd0ea18 NS |
2024 | */ |
2025 | STATIC int /* error */ | |
49f693fa DC |
2026 | xfs_bmap_add_extent_unwritten_real( |
2027 | struct xfs_trans *tp, | |
2bd0ea18 | 2028 | xfs_inode_t *ip, /* incore inode pointer */ |
4072e4b4 | 2029 | int whichfork, |
49f693fa DC |
2030 | xfs_extnum_t *idx, /* extent number to update/insert */ |
2031 | xfs_btree_cur_t **curp, /* if *curp is null, not a btree */ | |
2032 | xfs_bmbt_irec_t *new, /* new data to add to file extents */ | |
2033 | xfs_fsblock_t *first, /* pointer to firstblock variable */ | |
f33cea1a | 2034 | struct xfs_defer_ops *dfops, /* list of extents to be freed */ |
49f693fa | 2035 | int *logflagsp) /* inode logging flags */ |
2bd0ea18 | 2036 | { |
49f693fa | 2037 | xfs_btree_cur_t *cur; /* btree cursor */ |
2bd0ea18 | 2038 | int error; /* error return value */ |
2bd0ea18 NS |
2039 | int i; /* temp state */ |
2040 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
49f693fa | 2041 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
49f693fa DC |
2042 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ |
2043 | /* left is 0, right is 1, prev is 2 */ | |
2044 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 2045 | int state = xfs_bmap_fork_to_state(whichfork); |
4072e4b4 | 2046 | struct xfs_mount *mp = ip->i_mount; |
fd19685d | 2047 | struct xfs_bmbt_irec old; |
5000d01d | 2048 | |
49f693fa | 2049 | *logflagsp = 0; |
56b2de80 | 2050 | |
49f693fa | 2051 | cur = *curp; |
4072e4b4 | 2052 | ifp = XFS_IFORK_PTR(ip, whichfork); |
56b2de80 | 2053 | |
49f693fa | 2054 | ASSERT(*idx >= 0); |
d09d4e5f | 2055 | ASSERT(*idx <= xfs_iext_count(ifp)); |
49f693fa DC |
2056 | ASSERT(!isnullstartblock(new->br_startblock)); |
2057 | ||
79896434 | 2058 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
2059 | |
2060 | #define LEFT r[0] | |
2061 | #define RIGHT r[1] | |
2062 | #define PREV r[2] | |
2063 | ||
2064 | /* | |
2065 | * Set up a bunch of variables to make the tests simpler. | |
2066 | */ | |
2bd0ea18 | 2067 | error = 0; |
fd19685d CH |
2068 | xfs_iext_get_extent(ifp, *idx, &PREV); |
2069 | ASSERT(new->br_state != PREV.br_state); | |
49f693fa DC |
2070 | new_endoff = new->br_startoff + new->br_blockcount; |
2071 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
2072 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
2073 | ||
2bd0ea18 | 2074 | /* |
49f693fa DC |
2075 | * Set flags determining what part of the previous oldext allocation |
2076 | * extent is being replaced by a newext allocation. | |
2bd0ea18 | 2077 | */ |
49f693fa DC |
2078 | if (PREV.br_startoff == new->br_startoff) |
2079 | state |= BMAP_LEFT_FILLING; | |
2080 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
2081 | state |= BMAP_RIGHT_FILLING; | |
2bd0ea18 | 2082 | |
49f693fa DC |
2083 | /* |
2084 | * Check and set flags if this segment has a left neighbor. | |
2085 | * Don't set contiguous if the combined extent would be too large. | |
2086 | */ | |
2087 | if (*idx > 0) { | |
2088 | state |= BMAP_LEFT_VALID; | |
99a663c1 | 2089 | xfs_iext_get_extent(ifp, *idx - 1, &LEFT); |
49f693fa DC |
2090 | |
2091 | if (isnullstartblock(LEFT.br_startblock)) | |
2092 | state |= BMAP_LEFT_DELAY; | |
2bd0ea18 | 2093 | } |
49f693fa DC |
2094 | |
2095 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2096 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
2097 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
fd19685d | 2098 | LEFT.br_state == new->br_state && |
49f693fa DC |
2099 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) |
2100 | state |= BMAP_LEFT_CONTIG; | |
2101 | ||
2bd0ea18 | 2102 | /* |
49f693fa DC |
2103 | * Check and set flags if this segment has a right neighbor. |
2104 | * Don't set contiguous if the combined extent would be too large. | |
2105 | * Also check for all-three-contiguous being too large. | |
2bd0ea18 | 2106 | */ |
4072e4b4 | 2107 | if (*idx < xfs_iext_count(ifp) - 1) { |
49f693fa | 2108 | state |= BMAP_RIGHT_VALID; |
99a663c1 | 2109 | xfs_iext_get_extent(ifp, *idx + 1, &RIGHT); |
49f693fa DC |
2110 | if (isnullstartblock(RIGHT.br_startblock)) |
2111 | state |= BMAP_RIGHT_DELAY; | |
2112 | } | |
a2ceac1f | 2113 | |
49f693fa DC |
2114 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2115 | new_endoff == RIGHT.br_startoff && | |
2116 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
fd19685d | 2117 | new->br_state == RIGHT.br_state && |
49f693fa DC |
2118 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && |
2119 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2120 | BMAP_RIGHT_FILLING)) != | |
2121 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2122 | BMAP_RIGHT_FILLING) || | |
2123 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
2124 | <= MAXEXTLEN)) | |
2125 | state |= BMAP_RIGHT_CONTIG; | |
2bd0ea18 | 2126 | |
49f693fa DC |
2127 | /* |
2128 | * Switch out based on the FILLING and CONTIG state bits. | |
2129 | */ | |
2130 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2131 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
2132 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2133 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
2bd0ea18 | 2134 | /* |
49f693fa DC |
2135 | * Setting all of a previous oldext extent to newext. |
2136 | * The left and right neighbors are both contiguous with new. | |
2bd0ea18 | 2137 | */ |
49f693fa DC |
2138 | --*idx; |
2139 | ||
fd19685d | 2140 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
df926c07 | 2141 | xfs_iext_update_extent(ip, state, *idx, &LEFT); |
49f693fa DC |
2142 | |
2143 | xfs_iext_remove(ip, *idx + 1, 2, state); | |
4072e4b4 DW |
2144 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2145 | XFS_IFORK_NEXTENTS(ip, whichfork) - 2); | |
49f693fa DC |
2146 | if (cur == NULL) |
2147 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2148 | else { | |
2149 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2150 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2151 | if (error) | |
49f693fa | 2152 | goto done; |
19ebedcf | 2153 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2154 | if ((error = xfs_btree_delete(cur, &i))) |
2155 | goto done; | |
19ebedcf | 2156 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2157 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2158 | goto done; | |
19ebedcf | 2159 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2160 | if ((error = xfs_btree_delete(cur, &i))) |
2161 | goto done; | |
19ebedcf | 2162 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2163 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2164 | goto done; | |
19ebedcf | 2165 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2166 | error = xfs_bmbt_update(cur, &LEFT); |
fd19685d | 2167 | if (error) |
49f693fa | 2168 | goto done; |
2bd0ea18 | 2169 | } |
2bd0ea18 NS |
2170 | break; |
2171 | ||
49f693fa | 2172 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: |
2bd0ea18 | 2173 | /* |
49f693fa DC |
2174 | * Setting all of a previous oldext extent to newext. |
2175 | * The left neighbor is contiguous, the right is not. | |
2bd0ea18 | 2176 | */ |
49f693fa DC |
2177 | --*idx; |
2178 | ||
fd19685d | 2179 | LEFT.br_blockcount += PREV.br_blockcount; |
df926c07 | 2180 | xfs_iext_update_extent(ip, state, *idx, &LEFT); |
49f693fa DC |
2181 | |
2182 | xfs_iext_remove(ip, *idx + 1, 1, state); | |
4072e4b4 DW |
2183 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2184 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
49f693fa DC |
2185 | if (cur == NULL) |
2186 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2187 | else { | |
2188 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2189 | error = xfs_bmbt_lookup_eq(cur, &PREV, &i); |
2190 | if (error) | |
49f693fa | 2191 | goto done; |
19ebedcf | 2192 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2193 | if ((error = xfs_btree_delete(cur, &i))) |
2194 | goto done; | |
19ebedcf | 2195 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2196 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2197 | goto done; | |
19ebedcf | 2198 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2199 | error = xfs_bmbt_update(cur, &LEFT); |
fd19685d | 2200 | if (error) |
49f693fa | 2201 | goto done; |
2bd0ea18 | 2202 | } |
2bd0ea18 | 2203 | break; |
5000d01d | 2204 | |
49f693fa | 2205 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2bd0ea18 | 2206 | /* |
49f693fa DC |
2207 | * Setting all of a previous oldext extent to newext. |
2208 | * The right neighbor is contiguous, the left is not. | |
2bd0ea18 | 2209 | */ |
fd19685d CH |
2210 | PREV.br_blockcount += RIGHT.br_blockcount; |
2211 | PREV.br_state = new->br_state; | |
df926c07 | 2212 | xfs_iext_update_extent(ip, state, *idx, &PREV); |
fd19685d | 2213 | |
49f693fa | 2214 | xfs_iext_remove(ip, *idx + 1, 1, state); |
4072e4b4 DW |
2215 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2216 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
49f693fa DC |
2217 | if (cur == NULL) |
2218 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2219 | else { | |
2220 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2221 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2222 | if (error) | |
49f693fa | 2223 | goto done; |
19ebedcf | 2224 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2225 | if ((error = xfs_btree_delete(cur, &i))) |
2226 | goto done; | |
19ebedcf | 2227 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2228 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2229 | goto done; | |
19ebedcf | 2230 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2231 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2232 | if (error) |
49f693fa DC |
2233 | goto done; |
2234 | } | |
2235 | break; | |
2bd0ea18 | 2236 | |
49f693fa DC |
2237 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: |
2238 | /* | |
2239 | * Setting all of a previous oldext extent to newext. | |
2240 | * Neither the left nor right neighbors are contiguous with | |
2241 | * the new one. | |
2242 | */ | |
fd19685d | 2243 | PREV.br_state = new->br_state; |
df926c07 | 2244 | xfs_iext_update_extent(ip, state, *idx, &PREV); |
2bd0ea18 | 2245 | |
49f693fa DC |
2246 | if (cur == NULL) |
2247 | rval = XFS_ILOG_DEXT; | |
2248 | else { | |
2249 | rval = 0; | |
70a93110 CH |
2250 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2251 | if (error) | |
49f693fa | 2252 | goto done; |
19ebedcf | 2253 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2254 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2255 | if (error) |
49f693fa DC |
2256 | goto done; |
2257 | } | |
2258 | break; | |
2bd0ea18 | 2259 | |
49f693fa DC |
2260 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
2261 | /* | |
2262 | * Setting the first part of a previous oldext extent to newext. | |
2263 | * The left neighbor is contiguous. | |
2264 | */ | |
fd19685d | 2265 | LEFT.br_blockcount += new->br_blockcount; |
df926c07 | 2266 | xfs_iext_update_extent(ip, state, *idx - 1, &LEFT); |
a2ceac1f | 2267 | |
fd19685d | 2268 | old = PREV; |
fd19685d CH |
2269 | PREV.br_startoff += new->br_blockcount; |
2270 | PREV.br_startblock += new->br_blockcount; | |
2271 | PREV.br_blockcount -= new->br_blockcount; | |
df926c07 | 2272 | xfs_iext_update_extent(ip, state, *idx, &PREV); |
b3563c19 | 2273 | |
49f693fa | 2274 | --*idx; |
b3563c19 | 2275 | |
49f693fa DC |
2276 | if (cur == NULL) |
2277 | rval = XFS_ILOG_DEXT; | |
2278 | else { | |
2279 | rval = 0; | |
70a93110 | 2280 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2281 | if (error) |
49f693fa | 2282 | goto done; |
19ebedcf | 2283 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2284 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2285 | if (error) |
49f693fa | 2286 | goto done; |
fd19685d CH |
2287 | error = xfs_btree_decrement(cur, 0, &i); |
2288 | if (error) | |
49f693fa | 2289 | goto done; |
d0e5f1ff | 2290 | error = xfs_bmbt_update(cur, &LEFT); |
49f693fa DC |
2291 | if (error) |
2292 | goto done; | |
2bd0ea18 | 2293 | } |
49f693fa | 2294 | break; |
b3563c19 | 2295 | |
49f693fa DC |
2296 | case BMAP_LEFT_FILLING: |
2297 | /* | |
2298 | * Setting the first part of a previous oldext extent to newext. | |
2299 | * The left neighbor is not contiguous. | |
2300 | */ | |
fd19685d | 2301 | old = PREV; |
fd19685d CH |
2302 | PREV.br_startoff += new->br_blockcount; |
2303 | PREV.br_startblock += new->br_blockcount; | |
2304 | PREV.br_blockcount -= new->br_blockcount; | |
df926c07 | 2305 | xfs_iext_update_extent(ip, state, *idx, &PREV); |
2bd0ea18 | 2306 | |
49f693fa | 2307 | xfs_iext_insert(ip, *idx, 1, new, state); |
4072e4b4 DW |
2308 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2309 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
49f693fa DC |
2310 | if (cur == NULL) |
2311 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2312 | else { | |
2313 | rval = XFS_ILOG_CORE; | |
70a93110 | 2314 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2315 | if (error) |
49f693fa | 2316 | goto done; |
19ebedcf | 2317 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2318 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2319 | if (error) |
49f693fa DC |
2320 | goto done; |
2321 | cur->bc_rec.b = *new; | |
2322 | if ((error = xfs_btree_insert(cur, &i))) | |
2323 | goto done; | |
19ebedcf | 2324 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2325 | } |
2326 | break; | |
56b2de80 | 2327 | |
49f693fa DC |
2328 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2329 | /* | |
2330 | * Setting the last part of a previous oldext extent to newext. | |
2331 | * The right neighbor is contiguous with the new allocation. | |
2332 | */ | |
fd19685d | 2333 | old = PREV; |
fd19685d | 2334 | PREV.br_blockcount -= new->br_blockcount; |
df926c07 | 2335 | xfs_iext_update_extent(ip, state, *idx, &PREV); |
56b2de80 | 2336 | |
49f693fa | 2337 | ++*idx; |
56b2de80 | 2338 | |
fd19685d CH |
2339 | RIGHT.br_startoff = new->br_startoff; |
2340 | RIGHT.br_startblock = new->br_startblock; | |
2341 | RIGHT.br_blockcount += new->br_blockcount; | |
df926c07 | 2342 | xfs_iext_update_extent(ip, state, *idx, &RIGHT); |
56b2de80 | 2343 | |
49f693fa DC |
2344 | if (cur == NULL) |
2345 | rval = XFS_ILOG_DEXT; | |
2346 | else { | |
2347 | rval = 0; | |
70a93110 | 2348 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2349 | if (error) |
49f693fa | 2350 | goto done; |
19ebedcf | 2351 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2352 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2353 | if (error) |
49f693fa | 2354 | goto done; |
fd19685d CH |
2355 | error = xfs_btree_increment(cur, 0, &i); |
2356 | if (error) | |
49f693fa | 2357 | goto done; |
d0e5f1ff | 2358 | error = xfs_bmbt_update(cur, &RIGHT); |
fd19685d | 2359 | if (error) |
49f693fa DC |
2360 | goto done; |
2361 | } | |
2362 | break; | |
ca86e759 | 2363 | |
49f693fa DC |
2364 | case BMAP_RIGHT_FILLING: |
2365 | /* | |
2366 | * Setting the last part of a previous oldext extent to newext. | |
2367 | * The right neighbor is not contiguous. | |
2368 | */ | |
fd19685d | 2369 | old = PREV; |
fd19685d | 2370 | PREV.br_blockcount -= new->br_blockcount; |
df926c07 | 2371 | xfs_iext_update_extent(ip, state, *idx, &PREV); |
2bd0ea18 | 2372 | |
49f693fa DC |
2373 | ++*idx; |
2374 | xfs_iext_insert(ip, *idx, 1, new, state); | |
2bd0ea18 | 2375 | |
4072e4b4 DW |
2376 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2377 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
49f693fa DC |
2378 | if (cur == NULL) |
2379 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2380 | else { | |
2381 | rval = XFS_ILOG_CORE; | |
70a93110 | 2382 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2383 | if (error) |
49f693fa | 2384 | goto done; |
19ebedcf | 2385 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2386 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2387 | if (error) |
49f693fa | 2388 | goto done; |
70a93110 CH |
2389 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2390 | if (error) | |
49f693fa | 2391 | goto done; |
19ebedcf | 2392 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
2393 | if ((error = xfs_btree_insert(cur, &i))) |
2394 | goto done; | |
19ebedcf | 2395 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2396 | } |
2397 | break; | |
2398 | ||
2399 | case 0: | |
2bd0ea18 | 2400 | /* |
49f693fa DC |
2401 | * Setting the middle part of a previous oldext extent to |
2402 | * newext. Contiguity is impossible here. | |
2403 | * One extent becomes three extents. | |
2bd0ea18 | 2404 | */ |
fd19685d | 2405 | old = PREV; |
fd19685d | 2406 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
df926c07 | 2407 | xfs_iext_update_extent(ip, state, *idx, &PREV); |
49f693fa DC |
2408 | |
2409 | r[0] = *new; | |
2410 | r[1].br_startoff = new_endoff; | |
2411 | r[1].br_blockcount = | |
fd19685d | 2412 | old.br_startoff + old.br_blockcount - new_endoff; |
49f693fa | 2413 | r[1].br_startblock = new->br_startblock + new->br_blockcount; |
fd19685d | 2414 | r[1].br_state = PREV.br_state; |
49f693fa DC |
2415 | |
2416 | ++*idx; | |
2417 | xfs_iext_insert(ip, *idx, 2, &r[0], state); | |
2418 | ||
4072e4b4 DW |
2419 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2420 | XFS_IFORK_NEXTENTS(ip, whichfork) + 2); | |
49f693fa DC |
2421 | if (cur == NULL) |
2422 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2423 | else { | |
2424 | rval = XFS_ILOG_CORE; | |
70a93110 | 2425 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2426 | if (error) |
49f693fa | 2427 | goto done; |
19ebedcf | 2428 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 2429 | /* new right extent - oldext */ |
d0e5f1ff CH |
2430 | error = xfs_bmbt_update(cur, &r[1]); |
2431 | if (error) | |
49f693fa DC |
2432 | goto done; |
2433 | /* new left extent - oldext */ | |
2434 | cur->bc_rec.b = PREV; | |
49f693fa DC |
2435 | if ((error = xfs_btree_insert(cur, &i))) |
2436 | goto done; | |
19ebedcf | 2437 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2438 | /* |
2439 | * Reset the cursor to the position of the new extent | |
2440 | * we are about to insert as we can't trust it after | |
2441 | * the previous insert. | |
2442 | */ | |
70a93110 CH |
2443 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2444 | if (error) | |
49f693fa | 2445 | goto done; |
19ebedcf | 2446 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa | 2447 | /* new middle extent - newext */ |
49f693fa DC |
2448 | if ((error = xfs_btree_insert(cur, &i))) |
2449 | goto done; | |
19ebedcf | 2450 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
2bd0ea18 | 2451 | } |
49f693fa DC |
2452 | break; |
2453 | ||
2454 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2455 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2456 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2457 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2458 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2459 | case BMAP_LEFT_CONTIG: | |
2460 | case BMAP_RIGHT_CONTIG: | |
5000d01d | 2461 | /* |
49f693fa | 2462 | * These cases are all impossible. |
2bd0ea18 | 2463 | */ |
49f693fa DC |
2464 | ASSERT(0); |
2465 | } | |
2466 | ||
d7f80320 | 2467 | /* update reverse mappings */ |
4072e4b4 | 2468 | error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new); |
d7f80320 DW |
2469 | if (error) |
2470 | goto done; | |
2471 | ||
49f693fa | 2472 | /* convert to a btree if necessary */ |
4072e4b4 | 2473 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
49f693fa DC |
2474 | int tmp_logflags; /* partial log flag return val */ |
2475 | ||
2476 | ASSERT(cur == NULL); | |
f33cea1a | 2477 | error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur, |
4072e4b4 | 2478 | 0, &tmp_logflags, whichfork); |
49f693fa DC |
2479 | *logflagsp |= tmp_logflags; |
2480 | if (error) | |
2481 | goto done; | |
ca86e759 | 2482 | } |
49f693fa DC |
2483 | |
2484 | /* clear out the allocated field, done with it now in any case. */ | |
2485 | if (cur) { | |
2486 | cur->bc_private.b.allocated = 0; | |
2487 | *curp = cur; | |
2488 | } | |
2489 | ||
4072e4b4 | 2490 | xfs_bmap_check_leaf_extents(*curp, ip, whichfork); |
2bd0ea18 | 2491 | done: |
49f693fa | 2492 | *logflagsp |= rval; |
2bd0ea18 | 2493 | return error; |
49f693fa DC |
2494 | #undef LEFT |
2495 | #undef RIGHT | |
2496 | #undef PREV | |
2bd0ea18 NS |
2497 | } |
2498 | ||
5e656dbb | 2499 | /* |
49f693fa | 2500 | * Convert a hole to a delayed allocation. |
5e656dbb | 2501 | */ |
49f693fa DC |
2502 | STATIC void |
2503 | xfs_bmap_add_extent_hole_delay( | |
2504 | xfs_inode_t *ip, /* incore inode pointer */ | |
cc66acaf | 2505 | int whichfork, |
49f693fa DC |
2506 | xfs_extnum_t *idx, /* extent number to update/insert */ |
2507 | xfs_bmbt_irec_t *new) /* new data to add to file extents */ | |
2bd0ea18 | 2508 | { |
49f693fa DC |
2509 | xfs_ifork_t *ifp; /* inode fork pointer */ |
2510 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ | |
2511 | xfs_filblks_t newlen=0; /* new indirect size */ | |
2512 | xfs_filblks_t oldlen=0; /* old indirect size */ | |
2513 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
7bd43334 | 2514 | int state = xfs_bmap_fork_to_state(whichfork); |
450b3981 | 2515 | xfs_filblks_t temp; /* temp for indirect calculations */ |
49f693fa | 2516 | |
cc66acaf | 2517 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa | 2518 | ASSERT(isnullstartblock(new->br_startblock)); |
2bd0ea18 | 2519 | |
062998e3 | 2520 | /* |
49f693fa | 2521 | * Check and set flags if this segment has a left neighbor |
062998e3 | 2522 | */ |
49f693fa DC |
2523 | if (*idx > 0) { |
2524 | state |= BMAP_LEFT_VALID; | |
99a663c1 | 2525 | xfs_iext_get_extent(ifp, *idx - 1, &left); |
5e656dbb | 2526 | |
49f693fa DC |
2527 | if (isnullstartblock(left.br_startblock)) |
2528 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2529 | } |
49f693fa DC |
2530 | |
2531 | /* | |
2532 | * Check and set flags if the current (right) segment exists. | |
2533 | * If it doesn't exist, we're converting the hole at end-of-file. | |
2534 | */ | |
d09d4e5f | 2535 | if (*idx < xfs_iext_count(ifp)) { |
49f693fa | 2536 | state |= BMAP_RIGHT_VALID; |
99a663c1 | 2537 | xfs_iext_get_extent(ifp, *idx, &right); |
49f693fa DC |
2538 | |
2539 | if (isnullstartblock(right.br_startblock)) | |
2540 | state |= BMAP_RIGHT_DELAY; | |
2541 | } | |
2542 | ||
2543 | /* | |
2544 | * Set contiguity flags on the left and right neighbors. | |
2545 | * Don't let extents get too large, even if the pieces are contiguous. | |
2546 | */ | |
2547 | if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) && | |
2548 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2549 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2550 | state |= BMAP_LEFT_CONTIG; | |
2551 | ||
2552 | if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) && | |
2553 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2554 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2555 | (!(state & BMAP_LEFT_CONTIG) || | |
2556 | (left.br_blockcount + new->br_blockcount + | |
2557 | right.br_blockcount <= MAXEXTLEN))) | |
2558 | state |= BMAP_RIGHT_CONTIG; | |
2559 | ||
2560 | /* | |
2561 | * Switch out based on the contiguity flags. | |
2562 | */ | |
2563 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2564 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2565 | /* | |
2566 | * New allocation is contiguous with delayed allocations | |
2567 | * on the left and on the right. | |
2568 | * Merge all three into a single extent record. | |
2569 | */ | |
2570 | --*idx; | |
2571 | temp = left.br_blockcount + new->br_blockcount + | |
2572 | right.br_blockcount; | |
2573 | ||
49f693fa DC |
2574 | oldlen = startblockval(left.br_startblock) + |
2575 | startblockval(new->br_startblock) + | |
2576 | startblockval(right.br_startblock); | |
f3b62b32 BF |
2577 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2578 | oldlen); | |
450b3981 CH |
2579 | left.br_startblock = nullstartblock(newlen); |
2580 | left.br_blockcount = temp; | |
df926c07 | 2581 | xfs_iext_update_extent(ip, state, *idx, &left); |
49f693fa DC |
2582 | |
2583 | xfs_iext_remove(ip, *idx + 1, 1, state); | |
2584 | break; | |
2585 | ||
2586 | case BMAP_LEFT_CONTIG: | |
2587 | /* | |
2588 | * New allocation is contiguous with a delayed allocation | |
2589 | * on the left. | |
2590 | * Merge the new allocation with the left neighbor. | |
2591 | */ | |
2592 | --*idx; | |
2593 | temp = left.br_blockcount + new->br_blockcount; | |
2594 | ||
49f693fa DC |
2595 | oldlen = startblockval(left.br_startblock) + |
2596 | startblockval(new->br_startblock); | |
f3b62b32 BF |
2597 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2598 | oldlen); | |
450b3981 CH |
2599 | left.br_blockcount = temp; |
2600 | left.br_startblock = nullstartblock(newlen); | |
df926c07 | 2601 | xfs_iext_update_extent(ip, state, *idx, &left); |
49f693fa DC |
2602 | break; |
2603 | ||
2604 | case BMAP_RIGHT_CONTIG: | |
2605 | /* | |
2606 | * New allocation is contiguous with a delayed allocation | |
2607 | * on the right. | |
2608 | * Merge the new allocation with the right neighbor. | |
2609 | */ | |
49f693fa DC |
2610 | temp = new->br_blockcount + right.br_blockcount; |
2611 | oldlen = startblockval(new->br_startblock) + | |
2612 | startblockval(right.br_startblock); | |
f3b62b32 BF |
2613 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2614 | oldlen); | |
450b3981 CH |
2615 | right.br_startoff = new->br_startoff; |
2616 | right.br_startblock = nullstartblock(newlen); | |
2617 | right.br_blockcount = temp; | |
df926c07 | 2618 | xfs_iext_update_extent(ip, state, *idx, &right); |
49f693fa DC |
2619 | break; |
2620 | ||
2621 | case 0: | |
2622 | /* | |
2623 | * New allocation is not contiguous with another | |
2624 | * delayed allocation. | |
2625 | * Insert a new entry. | |
2626 | */ | |
2627 | oldlen = newlen = 0; | |
2628 | xfs_iext_insert(ip, *idx, 1, new, state); | |
2629 | break; | |
2630 | } | |
2631 | if (oldlen != newlen) { | |
2632 | ASSERT(oldlen > newlen); | |
19ebedcf DC |
2633 | xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen), |
2634 | false); | |
49f693fa DC |
2635 | /* |
2636 | * Nothing to do for disk quota accounting here. | |
2637 | */ | |
2bd0ea18 | 2638 | } |
2bd0ea18 NS |
2639 | } |
2640 | ||
2641 | /* | |
49f693fa | 2642 | * Convert a hole to a real allocation. |
2bd0ea18 | 2643 | */ |
49f693fa DC |
2644 | STATIC int /* error */ |
2645 | xfs_bmap_add_extent_hole_real( | |
972432f2 CH |
2646 | struct xfs_trans *tp, |
2647 | struct xfs_inode *ip, | |
2648 | int whichfork, | |
2649 | xfs_extnum_t *idx, | |
2650 | struct xfs_btree_cur **curp, | |
2651 | struct xfs_bmbt_irec *new, | |
2652 | xfs_fsblock_t *first, | |
2653 | struct xfs_defer_ops *dfops, | |
2654 | int *logflagsp) | |
5000d01d | 2655 | { |
972432f2 CH |
2656 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
2657 | struct xfs_mount *mp = ip->i_mount; | |
2658 | struct xfs_btree_cur *cur = *curp; | |
49f693fa DC |
2659 | int error; /* error return value */ |
2660 | int i; /* temp state */ | |
49f693fa DC |
2661 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2662 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2663 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 2664 | int state = xfs_bmap_fork_to_state(whichfork); |
3281eb91 | 2665 | struct xfs_bmbt_irec old; |
2bd0ea18 | 2666 | |
972432f2 CH |
2667 | ASSERT(*idx >= 0); |
2668 | ASSERT(*idx <= xfs_iext_count(ifp)); | |
49f693fa | 2669 | ASSERT(!isnullstartblock(new->br_startblock)); |
972432f2 | 2670 | ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); |
5e656dbb | 2671 | |
79896434 | 2672 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa | 2673 | |
49f693fa DC |
2674 | /* |
2675 | * Check and set flags if this segment has a left neighbor. | |
2676 | */ | |
972432f2 | 2677 | if (*idx > 0) { |
49f693fa | 2678 | state |= BMAP_LEFT_VALID; |
99a663c1 | 2679 | xfs_iext_get_extent(ifp, *idx - 1, &left); |
49f693fa DC |
2680 | if (isnullstartblock(left.br_startblock)) |
2681 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2682 | } |
2bd0ea18 | 2683 | |
49f693fa DC |
2684 | /* |
2685 | * Check and set flags if this segment has a current value. | |
2686 | * Not true if we're inserting into the "hole" at eof. | |
2687 | */ | |
972432f2 | 2688 | if (*idx < xfs_iext_count(ifp)) { |
49f693fa | 2689 | state |= BMAP_RIGHT_VALID; |
99a663c1 | 2690 | xfs_iext_get_extent(ifp, *idx, &right); |
49f693fa DC |
2691 | if (isnullstartblock(right.br_startblock)) |
2692 | state |= BMAP_RIGHT_DELAY; | |
2bd0ea18 | 2693 | } |
2bd0ea18 | 2694 | |
49f693fa DC |
2695 | /* |
2696 | * We're inserting a real allocation between "left" and "right". | |
2697 | * Set the contiguity flags. Don't let extents get too large. | |
2698 | */ | |
2699 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2700 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2701 | left.br_startblock + left.br_blockcount == new->br_startblock && | |
2702 | left.br_state == new->br_state && | |
2703 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2704 | state |= BMAP_LEFT_CONTIG; | |
57c9fccb | 2705 | |
49f693fa DC |
2706 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2707 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2708 | new->br_startblock + new->br_blockcount == right.br_startblock && | |
2709 | new->br_state == right.br_state && | |
2710 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2711 | (!(state & BMAP_LEFT_CONTIG) || | |
2712 | left.br_blockcount + new->br_blockcount + | |
2713 | right.br_blockcount <= MAXEXTLEN)) | |
2714 | state |= BMAP_RIGHT_CONTIG; | |
ca86e759 | 2715 | |
49f693fa DC |
2716 | error = 0; |
2717 | /* | |
2718 | * Select which case we're in here, and implement it. | |
2719 | */ | |
2720 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2721 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
57c9fccb | 2722 | /* |
49f693fa DC |
2723 | * New allocation is contiguous with real allocations on the |
2724 | * left and on the right. | |
2725 | * Merge all three into a single extent record. | |
57c9fccb | 2726 | */ |
972432f2 | 2727 | --*idx; |
3281eb91 | 2728 | left.br_blockcount += new->br_blockcount + right.br_blockcount; |
df926c07 | 2729 | xfs_iext_update_extent(ip, state, *idx, &left); |
56b2de80 | 2730 | |
972432f2 | 2731 | xfs_iext_remove(ip, *idx + 1, 1, state); |
56b2de80 | 2732 | |
972432f2 CH |
2733 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2734 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
2735 | if (cur == NULL) { | |
49f693fa DC |
2736 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2737 | } else { | |
2738 | rval = XFS_ILOG_CORE; | |
70a93110 | 2739 | error = xfs_bmbt_lookup_eq(cur, &right, &i); |
49f693fa DC |
2740 | if (error) |
2741 | goto done; | |
19ebedcf | 2742 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
972432f2 | 2743 | error = xfs_btree_delete(cur, &i); |
49f693fa DC |
2744 | if (error) |
2745 | goto done; | |
19ebedcf | 2746 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
972432f2 | 2747 | error = xfs_btree_decrement(cur, 0, &i); |
49f693fa DC |
2748 | if (error) |
2749 | goto done; | |
19ebedcf | 2750 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2751 | error = xfs_bmbt_update(cur, &left); |
49f693fa DC |
2752 | if (error) |
2753 | goto done; | |
2754 | } | |
57c9fccb | 2755 | break; |
49f693fa DC |
2756 | |
2757 | case BMAP_LEFT_CONTIG: | |
2758 | /* | |
2759 | * New allocation is contiguous with a real allocation | |
2760 | * on the left. | |
2761 | * Merge the new allocation with the left neighbor. | |
2762 | */ | |
972432f2 | 2763 | --*idx; |
3281eb91 | 2764 | old = left; |
df926c07 | 2765 | |
3281eb91 | 2766 | left.br_blockcount += new->br_blockcount; |
df926c07 | 2767 | xfs_iext_update_extent(ip, state, *idx, &left); |
49f693fa | 2768 | |
972432f2 | 2769 | if (cur == NULL) { |
49f693fa DC |
2770 | rval = xfs_ilog_fext(whichfork); |
2771 | } else { | |
2772 | rval = 0; | |
70a93110 | 2773 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
49f693fa DC |
2774 | if (error) |
2775 | goto done; | |
19ebedcf | 2776 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2777 | error = xfs_bmbt_update(cur, &left); |
49f693fa DC |
2778 | if (error) |
2779 | goto done; | |
2780 | } | |
57c9fccb | 2781 | break; |
49f693fa DC |
2782 | |
2783 | case BMAP_RIGHT_CONTIG: | |
2784 | /* | |
2785 | * New allocation is contiguous with a real allocation | |
2786 | * on the right. | |
2787 | * Merge the new allocation with the right neighbor. | |
2788 | */ | |
3281eb91 | 2789 | old = right; |
df926c07 | 2790 | |
3281eb91 CH |
2791 | right.br_startoff = new->br_startoff; |
2792 | right.br_startblock = new->br_startblock; | |
2793 | right.br_blockcount += new->br_blockcount; | |
df926c07 | 2794 | xfs_iext_update_extent(ip, state, *idx, &right); |
49f693fa | 2795 | |
972432f2 | 2796 | if (cur == NULL) { |
49f693fa DC |
2797 | rval = xfs_ilog_fext(whichfork); |
2798 | } else { | |
2799 | rval = 0; | |
70a93110 | 2800 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
49f693fa DC |
2801 | if (error) |
2802 | goto done; | |
19ebedcf | 2803 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2804 | error = xfs_bmbt_update(cur, &right); |
49f693fa DC |
2805 | if (error) |
2806 | goto done; | |
2807 | } | |
2808 | break; | |
2809 | ||
2810 | case 0: | |
2811 | /* | |
2812 | * New allocation is not contiguous with another | |
2813 | * real allocation. | |
2814 | * Insert a new entry. | |
2815 | */ | |
972432f2 CH |
2816 | xfs_iext_insert(ip, *idx, 1, new, state); |
2817 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
2818 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
2819 | if (cur == NULL) { | |
49f693fa DC |
2820 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2821 | } else { | |
2822 | rval = XFS_ILOG_CORE; | |
70a93110 | 2823 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
49f693fa DC |
2824 | if (error) |
2825 | goto done; | |
19ebedcf | 2826 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
972432f2 | 2827 | error = xfs_btree_insert(cur, &i); |
49f693fa DC |
2828 | if (error) |
2829 | goto done; | |
19ebedcf | 2830 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 2831 | } |
57c9fccb | 2832 | break; |
57c9fccb | 2833 | } |
a2ceac1f | 2834 | |
d7f80320 | 2835 | /* add reverse mapping */ |
972432f2 | 2836 | error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new); |
d7f80320 DW |
2837 | if (error) |
2838 | goto done; | |
2839 | ||
49f693fa | 2840 | /* convert to a btree if necessary */ |
972432f2 | 2841 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
49f693fa | 2842 | int tmp_logflags; /* partial log flag return val */ |
3f853c7a | 2843 | |
972432f2 CH |
2844 | ASSERT(cur == NULL); |
2845 | error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp, | |
49f693fa | 2846 | 0, &tmp_logflags, whichfork); |
972432f2 CH |
2847 | *logflagsp |= tmp_logflags; |
2848 | cur = *curp; | |
49f693fa DC |
2849 | if (error) |
2850 | goto done; | |
57c9fccb | 2851 | } |
a2ceac1f | 2852 | |
49f693fa | 2853 | /* clear out the allocated field, done with it now in any case. */ |
972432f2 CH |
2854 | if (cur) |
2855 | cur->bc_private.b.allocated = 0; | |
49f693fa | 2856 | |
972432f2 | 2857 | xfs_bmap_check_leaf_extents(cur, ip, whichfork); |
49f693fa | 2858 | done: |
972432f2 | 2859 | *logflagsp |= rval; |
57c9fccb NS |
2860 | return error; |
2861 | } | |
2862 | ||
2bd0ea18 | 2863 | /* |
49f693fa | 2864 | * Functions used in the extent read, allocate and remove paths |
2bd0ea18 | 2865 | */ |
2bd0ea18 | 2866 | |
5000d01d | 2867 | /* |
49f693fa | 2868 | * Adjust the size of the new extent based on di_extsize and rt extsize. |
2bd0ea18 | 2869 | */ |
613e6057 | 2870 | int |
49f693fa DC |
2871 | xfs_bmap_extsize_align( |
2872 | xfs_mount_t *mp, | |
2873 | xfs_bmbt_irec_t *gotp, /* next extent pointer */ | |
2874 | xfs_bmbt_irec_t *prevp, /* previous extent pointer */ | |
2875 | xfs_extlen_t extsz, /* align to this extent size */ | |
2876 | int rt, /* is this a realtime inode? */ | |
2877 | int eof, /* is extent at end-of-file? */ | |
2878 | int delay, /* creating delalloc extent? */ | |
2879 | int convert, /* overwriting unwritten extent? */ | |
2880 | xfs_fileoff_t *offp, /* in/out: aligned offset */ | |
2881 | xfs_extlen_t *lenp) /* in/out: aligned length */ | |
2bd0ea18 | 2882 | { |
49f693fa DC |
2883 | xfs_fileoff_t orig_off; /* original offset */ |
2884 | xfs_extlen_t orig_alen; /* original length */ | |
2885 | xfs_fileoff_t orig_end; /* original off+len */ | |
2886 | xfs_fileoff_t nexto; /* next file offset */ | |
2887 | xfs_fileoff_t prevo; /* previous file offset */ | |
2888 | xfs_fileoff_t align_off; /* temp for offset */ | |
2889 | xfs_extlen_t align_alen; /* temp for length */ | |
2890 | xfs_extlen_t temp; /* temp for calculations */ | |
2891 | ||
2892 | if (convert) | |
2893 | return 0; | |
2894 | ||
2895 | orig_off = align_off = *offp; | |
2896 | orig_alen = align_alen = *lenp; | |
2897 | orig_end = orig_off + orig_alen; | |
2bd0ea18 NS |
2898 | |
2899 | /* | |
49f693fa DC |
2900 | * If this request overlaps an existing extent, then don't |
2901 | * attempt to perform any additional alignment. | |
2bd0ea18 | 2902 | */ |
49f693fa DC |
2903 | if (!delay && !eof && |
2904 | (orig_off >= gotp->br_startoff) && | |
2905 | (orig_end <= gotp->br_startoff + gotp->br_blockcount)) { | |
2906 | return 0; | |
2bd0ea18 | 2907 | } |
57c9fccb | 2908 | |
49f693fa DC |
2909 | /* |
2910 | * If the file offset is unaligned vs. the extent size | |
2911 | * we need to align it. This will be possible unless | |
2912 | * the file was previously written with a kernel that didn't | |
2913 | * perform this alignment, or if a truncate shot us in the | |
2914 | * foot. | |
2915 | */ | |
2916 | temp = do_mod(orig_off, extsz); | |
2917 | if (temp) { | |
2918 | align_alen += temp; | |
2919 | align_off -= temp; | |
2920 | } | |
7cc23f0c DC |
2921 | |
2922 | /* Same adjustment for the end of the requested area. */ | |
2923 | temp = (align_alen % extsz); | |
2924 | if (temp) | |
2925 | align_alen += extsz - temp; | |
2926 | ||
49f693fa | 2927 | /* |
7cc23f0c DC |
2928 | * For large extent hint sizes, the aligned extent might be larger than |
2929 | * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls | |
2930 | * the length back under MAXEXTLEN. The outer allocation loops handle | |
2931 | * short allocation just fine, so it is safe to do this. We only want to | |
2932 | * do it when we are forced to, though, because it means more allocation | |
2933 | * operations are required. | |
49f693fa | 2934 | */ |
7cc23f0c DC |
2935 | while (align_alen > MAXEXTLEN) |
2936 | align_alen -= extsz; | |
2937 | ASSERT(align_alen <= MAXEXTLEN); | |
2938 | ||
49f693fa DC |
2939 | /* |
2940 | * If the previous block overlaps with this proposed allocation | |
2941 | * then move the start forward without adjusting the length. | |
2942 | */ | |
2943 | if (prevp->br_startoff != NULLFILEOFF) { | |
2944 | if (prevp->br_startblock == HOLESTARTBLOCK) | |
2945 | prevo = prevp->br_startoff; | |
2946 | else | |
2947 | prevo = prevp->br_startoff + prevp->br_blockcount; | |
2948 | } else | |
2949 | prevo = 0; | |
2950 | if (align_off != orig_off && align_off < prevo) | |
2951 | align_off = prevo; | |
2952 | /* | |
2953 | * If the next block overlaps with this proposed allocation | |
2954 | * then move the start back without adjusting the length, | |
2955 | * but not before offset 0. | |
2956 | * This may of course make the start overlap previous block, | |
2957 | * and if we hit the offset 0 limit then the next block | |
2958 | * can still overlap too. | |
2959 | */ | |
2960 | if (!eof && gotp->br_startoff != NULLFILEOFF) { | |
2961 | if ((delay && gotp->br_startblock == HOLESTARTBLOCK) || | |
2962 | (!delay && gotp->br_startblock == DELAYSTARTBLOCK)) | |
2963 | nexto = gotp->br_startoff + gotp->br_blockcount; | |
2964 | else | |
2965 | nexto = gotp->br_startoff; | |
2966 | } else | |
2967 | nexto = NULLFILEOFF; | |
2968 | if (!eof && | |
2969 | align_off + align_alen != orig_end && | |
2970 | align_off + align_alen > nexto) | |
2971 | align_off = nexto > align_alen ? nexto - align_alen : 0; | |
2972 | /* | |
2973 | * If we're now overlapping the next or previous extent that | |
2974 | * means we can't fit an extsz piece in this hole. Just move | |
2975 | * the start forward to the first valid spot and set | |
2976 | * the length so we hit the end. | |
2977 | */ | |
2978 | if (align_off != orig_off && align_off < prevo) | |
2979 | align_off = prevo; | |
2980 | if (align_off + align_alen != orig_end && | |
2981 | align_off + align_alen > nexto && | |
2982 | nexto != NULLFILEOFF) { | |
2983 | ASSERT(nexto > prevo); | |
2984 | align_alen = nexto - align_off; | |
57c9fccb | 2985 | } |
2bd0ea18 | 2986 | |
49f693fa DC |
2987 | /* |
2988 | * If realtime, and the result isn't a multiple of the realtime | |
2989 | * extent size we need to remove blocks until it is. | |
2990 | */ | |
2991 | if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) { | |
2bd0ea18 | 2992 | /* |
49f693fa DC |
2993 | * We're not covering the original request, or |
2994 | * we won't be able to once we fix the length. | |
2bd0ea18 | 2995 | */ |
49f693fa DC |
2996 | if (orig_off < align_off || |
2997 | orig_end > align_off + align_alen || | |
2998 | align_alen - temp < orig_alen) | |
12b53197 | 2999 | return -EINVAL; |
49f693fa DC |
3000 | /* |
3001 | * Try to fix it by moving the start up. | |
3002 | */ | |
3003 | if (align_off + temp <= orig_off) { | |
3004 | align_alen -= temp; | |
3005 | align_off += temp; | |
2bd0ea18 | 3006 | } |
49f693fa DC |
3007 | /* |
3008 | * Try to fix it by moving the end in. | |
3009 | */ | |
3010 | else if (align_off + align_alen - temp >= orig_end) | |
3011 | align_alen -= temp; | |
3012 | /* | |
3013 | * Set the start to the minimum then trim the length. | |
3014 | */ | |
3015 | else { | |
3016 | align_alen -= orig_off - align_off; | |
3017 | align_off = orig_off; | |
3018 | align_alen -= align_alen % mp->m_sb.sb_rextsize; | |
3019 | } | |
3020 | /* | |
3021 | * Result doesn't cover the request, fail it. | |
3022 | */ | |
3023 | if (orig_off < align_off || orig_end > align_off + align_alen) | |
12b53197 | 3024 | return -EINVAL; |
49f693fa DC |
3025 | } else { |
3026 | ASSERT(orig_off >= align_off); | |
7cc23f0c DC |
3027 | /* see MAXEXTLEN handling above */ |
3028 | ASSERT(orig_end <= align_off + align_alen || | |
3029 | align_alen + extsz > MAXEXTLEN); | |
2bd0ea18 | 3030 | } |
49f693fa DC |
3031 | |
3032 | #ifdef DEBUG | |
3033 | if (!eof && gotp->br_startoff != NULLFILEOFF) | |
3034 | ASSERT(align_off + align_alen <= gotp->br_startoff); | |
3035 | if (prevp->br_startoff != NULLFILEOFF) | |
3036 | ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount); | |
3037 | #endif | |
3038 | ||
3039 | *lenp = align_alen; | |
3040 | *offp = align_off; | |
2bd0ea18 NS |
3041 | return 0; |
3042 | } | |
3043 | ||
49f693fa DC |
3044 | #define XFS_ALLOC_GAP_UNITS 4 |
3045 | ||
613e6057 | 3046 | void |
49f693fa | 3047 | xfs_bmap_adjacent( |
613e6057 | 3048 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
2bd0ea18 | 3049 | { |
49f693fa DC |
3050 | xfs_fsblock_t adjust; /* adjustment to block numbers */ |
3051 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ | |
3052 | xfs_mount_t *mp; /* mount point structure */ | |
3053 | int nullfb; /* true if ap->firstblock isn't set */ | |
3054 | int rt; /* true if inode is realtime */ | |
2bd0ea18 | 3055 | |
49f693fa DC |
3056 | #define ISVALID(x,y) \ |
3057 | (rt ? \ | |
3058 | (x) < mp->m_sb.sb_rblocks : \ | |
3059 | XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \ | |
3060 | XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \ | |
3061 | XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks) | |
3062 | ||
3063 | mp = ap->ip->i_mount; | |
3064 | nullfb = *ap->firstblock == NULLFSBLOCK; | |
1fccd5c8 DC |
3065 | rt = XFS_IS_REALTIME_INODE(ap->ip) && |
3066 | xfs_alloc_is_userdata(ap->datatype); | |
49f693fa DC |
3067 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock); |
3068 | /* | |
3069 | * If allocating at eof, and there's a previous real block, | |
3070 | * try to use its last block as our starting point. | |
3071 | */ | |
3072 | if (ap->eof && ap->prev.br_startoff != NULLFILEOFF && | |
3073 | !isnullstartblock(ap->prev.br_startblock) && | |
3074 | ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount, | |
3075 | ap->prev.br_startblock)) { | |
3076 | ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount; | |
3077 | /* | |
3078 | * Adjust for the gap between prevp and us. | |
3079 | */ | |
3080 | adjust = ap->offset - | |
3081 | (ap->prev.br_startoff + ap->prev.br_blockcount); | |
3082 | if (adjust && | |
3083 | ISVALID(ap->blkno + adjust, ap->prev.br_startblock)) | |
3084 | ap->blkno += adjust; | |
2bd0ea18 | 3085 | } |
49f693fa DC |
3086 | /* |
3087 | * If not at eof, then compare the two neighbor blocks. | |
3088 | * Figure out whether either one gives us a good starting point, | |
3089 | * and pick the better one. | |
3090 | */ | |
3091 | else if (!ap->eof) { | |
3092 | xfs_fsblock_t gotbno; /* right side block number */ | |
3093 | xfs_fsblock_t gotdiff=0; /* right side difference */ | |
3094 | xfs_fsblock_t prevbno; /* left side block number */ | |
3095 | xfs_fsblock_t prevdiff=0; /* left side difference */ | |
3096 | ||
3097 | /* | |
3098 | * If there's a previous (left) block, select a requested | |
3099 | * start block based on it. | |
3100 | */ | |
3101 | if (ap->prev.br_startoff != NULLFILEOFF && | |
3102 | !isnullstartblock(ap->prev.br_startblock) && | |
3103 | (prevbno = ap->prev.br_startblock + | |
3104 | ap->prev.br_blockcount) && | |
3105 | ISVALID(prevbno, ap->prev.br_startblock)) { | |
3106 | /* | |
3107 | * Calculate gap to end of previous block. | |
3108 | */ | |
3109 | adjust = prevdiff = ap->offset - | |
3110 | (ap->prev.br_startoff + | |
3111 | ap->prev.br_blockcount); | |
3112 | /* | |
3113 | * Figure the startblock based on the previous block's | |
3114 | * end and the gap size. | |
3115 | * Heuristic! | |
3116 | * If the gap is large relative to the piece we're | |
3117 | * allocating, or using it gives us an invalid block | |
3118 | * number, then just use the end of the previous block. | |
3119 | */ | |
3120 | if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3121 | ISVALID(prevbno + prevdiff, | |
3122 | ap->prev.br_startblock)) | |
3123 | prevbno += adjust; | |
3124 | else | |
3125 | prevdiff += adjust; | |
3126 | /* | |
3127 | * If the firstblock forbids it, can't use it, | |
3128 | * must use default. | |
3129 | */ | |
3130 | if (!rt && !nullfb && | |
3131 | XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno) | |
3132 | prevbno = NULLFSBLOCK; | |
3133 | } | |
3134 | /* | |
3135 | * No previous block or can't follow it, just default. | |
3136 | */ | |
3137 | else | |
3138 | prevbno = NULLFSBLOCK; | |
3139 | /* | |
3140 | * If there's a following (right) block, select a requested | |
3141 | * start block based on it. | |
3142 | */ | |
3143 | if (!isnullstartblock(ap->got.br_startblock)) { | |
3144 | /* | |
3145 | * Calculate gap to start of next block. | |
3146 | */ | |
3147 | adjust = gotdiff = ap->got.br_startoff - ap->offset; | |
3148 | /* | |
3149 | * Figure the startblock based on the next block's | |
3150 | * start and the gap size. | |
3151 | */ | |
3152 | gotbno = ap->got.br_startblock; | |
3153 | /* | |
3154 | * Heuristic! | |
3155 | * If the gap is large relative to the piece we're | |
3156 | * allocating, or using it gives us an invalid block | |
3157 | * number, then just use the start of the next block | |
3158 | * offset by our length. | |
3159 | */ | |
3160 | if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3161 | ISVALID(gotbno - gotdiff, gotbno)) | |
3162 | gotbno -= adjust; | |
3163 | else if (ISVALID(gotbno - ap->length, gotbno)) { | |
3164 | gotbno -= ap->length; | |
3165 | gotdiff += adjust - ap->length; | |
3166 | } else | |
3167 | gotdiff += adjust; | |
3168 | /* | |
3169 | * If the firstblock forbids it, can't use it, | |
3170 | * must use default. | |
3171 | */ | |
3172 | if (!rt && !nullfb && | |
3173 | XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno) | |
3174 | gotbno = NULLFSBLOCK; | |
3175 | } | |
3176 | /* | |
3177 | * No next block, just default. | |
3178 | */ | |
2bd0ea18 | 3179 | else |
49f693fa DC |
3180 | gotbno = NULLFSBLOCK; |
3181 | /* | |
3182 | * If both valid, pick the better one, else the only good | |
3183 | * one, else ap->blkno is already set (to 0 or the inode block). | |
3184 | */ | |
3185 | if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) | |
3186 | ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno; | |
3187 | else if (prevbno != NULLFSBLOCK) | |
3188 | ap->blkno = prevbno; | |
3189 | else if (gotbno != NULLFSBLOCK) | |
3190 | ap->blkno = gotbno; | |
a2ceac1f | 3191 | } |
49f693fa | 3192 | #undef ISVALID |
a2ceac1f DC |
3193 | } |
3194 | ||
ff105f75 DC |
3195 | static int |
3196 | xfs_bmap_longest_free_extent( | |
3197 | struct xfs_trans *tp, | |
3198 | xfs_agnumber_t ag, | |
3199 | xfs_extlen_t *blen, | |
3200 | int *notinit) | |
3201 | { | |
3202 | struct xfs_mount *mp = tp->t_mountp; | |
3203 | struct xfs_perag *pag; | |
3204 | xfs_extlen_t longest; | |
3205 | int error = 0; | |
3206 | ||
3207 | pag = xfs_perag_get(mp, ag); | |
3208 | if (!pag->pagf_init) { | |
3209 | error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK); | |
3210 | if (error) | |
3211 | goto out; | |
3212 | ||
3213 | if (!pag->pagf_init) { | |
3214 | *notinit = 1; | |
3215 | goto out; | |
3216 | } | |
3217 | } | |
3218 | ||
72bda06d | 3219 | longest = xfs_alloc_longest_free_extent(mp, pag, |
cf8ce220 DW |
3220 | xfs_alloc_min_freelist(mp, pag), |
3221 | xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE)); | |
ff105f75 DC |
3222 | if (*blen < longest) |
3223 | *blen = longest; | |
3224 | ||
3225 | out: | |
3226 | xfs_perag_put(pag); | |
3227 | return error; | |
3228 | } | |
3229 | ||
3230 | static void | |
3231 | xfs_bmap_select_minlen( | |
3232 | struct xfs_bmalloca *ap, | |
3233 | struct xfs_alloc_arg *args, | |
3234 | xfs_extlen_t *blen, | |
3235 | int notinit) | |
3236 | { | |
3237 | if (notinit || *blen < ap->minlen) { | |
3238 | /* | |
3239 | * Since we did a BUF_TRYLOCK above, it is possible that | |
3240 | * there is space for this request. | |
3241 | */ | |
3242 | args->minlen = ap->minlen; | |
3243 | } else if (*blen < args->maxlen) { | |
3244 | /* | |
3245 | * If the best seen length is less than the request length, | |
3246 | * use the best as the minimum. | |
3247 | */ | |
3248 | args->minlen = *blen; | |
3249 | } else { | |
3250 | /* | |
3251 | * Otherwise we've seen an extent as big as maxlen, use that | |
3252 | * as the minimum. | |
3253 | */ | |
3254 | args->minlen = args->maxlen; | |
3255 | } | |
3256 | } | |
3257 | ||
a2ceac1f | 3258 | STATIC int |
49f693fa DC |
3259 | xfs_bmap_btalloc_nullfb( |
3260 | struct xfs_bmalloca *ap, | |
3261 | struct xfs_alloc_arg *args, | |
3262 | xfs_extlen_t *blen) | |
a2ceac1f | 3263 | { |
49f693fa | 3264 | struct xfs_mount *mp = ap->ip->i_mount; |
49f693fa DC |
3265 | xfs_agnumber_t ag, startag; |
3266 | int notinit = 0; | |
a2ceac1f DC |
3267 | int error; |
3268 | ||
ff105f75 | 3269 | args->type = XFS_ALLOCTYPE_START_BNO; |
49f693fa | 3270 | args->total = ap->total; |
a2ceac1f | 3271 | |
49f693fa DC |
3272 | startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno); |
3273 | if (startag == NULLAGNUMBER) | |
3274 | startag = ag = 0; | |
a2ceac1f | 3275 | |
49f693fa | 3276 | while (*blen < args->maxlen) { |
ff105f75 DC |
3277 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, |
3278 | ¬init); | |
3279 | if (error) | |
3280 | return error; | |
a2ceac1f | 3281 | |
49f693fa DC |
3282 | if (++ag == mp->m_sb.sb_agcount) |
3283 | ag = 0; | |
3284 | if (ag == startag) | |
3285 | break; | |
49f693fa | 3286 | } |
2bd0ea18 | 3287 | |
ff105f75 DC |
3288 | xfs_bmap_select_minlen(ap, args, blen, notinit); |
3289 | return 0; | |
3290 | } | |
3291 | ||
3292 | STATIC int | |
3293 | xfs_bmap_btalloc_filestreams( | |
3294 | struct xfs_bmalloca *ap, | |
3295 | struct xfs_alloc_arg *args, | |
3296 | xfs_extlen_t *blen) | |
3297 | { | |
3298 | struct xfs_mount *mp = ap->ip->i_mount; | |
3299 | xfs_agnumber_t ag; | |
3300 | int notinit = 0; | |
3301 | int error; | |
3302 | ||
3303 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
3304 | args->total = ap->total; | |
3305 | ||
3306 | ag = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3307 | if (ag == NULLAGNUMBER) | |
3308 | ag = 0; | |
3309 | ||
3310 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init); | |
3311 | if (error) | |
3312 | return error; | |
3313 | ||
3314 | if (*blen < args->maxlen) { | |
3315 | error = xfs_filestream_new_ag(ap, &ag); | |
3316 | if (error) | |
3317 | return error; | |
3318 | ||
3319 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, | |
3320 | ¬init); | |
3321 | if (error) | |
3322 | return error; | |
3323 | ||
3324 | } | |
3325 | ||
3326 | xfs_bmap_select_minlen(ap, args, blen, notinit); | |
2bd0ea18 | 3327 | |
49f693fa | 3328 | /* |
ff105f75 DC |
3329 | * Set the failure fallback case to look in the selected AG as stream |
3330 | * may have moved. | |
49f693fa | 3331 | */ |
ff105f75 | 3332 | ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0); |
49f693fa | 3333 | return 0; |
2bd0ea18 NS |
3334 | } |
3335 | ||
b3563c19 | 3336 | STATIC int |
49f693fa | 3337 | xfs_bmap_btalloc( |
613e6057 | 3338 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
b3563c19 | 3339 | { |
49f693fa DC |
3340 | xfs_mount_t *mp; /* mount point structure */ |
3341 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
1fccd5c8 | 3342 | xfs_extlen_t align = 0; /* minimum allocation alignment */ |
49f693fa DC |
3343 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ |
3344 | xfs_agnumber_t ag; | |
3345 | xfs_alloc_arg_t args; | |
3346 | xfs_extlen_t blen; | |
3347 | xfs_extlen_t nextminlen = 0; | |
3348 | int nullfb; /* true if ap->firstblock isn't set */ | |
3349 | int isaligned; | |
3350 | int tryagain; | |
3351 | int error; | |
ff105f75 | 3352 | int stripe_align; |
b3563c19 | 3353 | |
49f693fa | 3354 | ASSERT(ap->length); |
b3563c19 | 3355 | |
49f693fa | 3356 | mp = ap->ip->i_mount; |
ff105f75 DC |
3357 | |
3358 | /* stripe alignment for allocation is determined by mount parameters */ | |
3359 | stripe_align = 0; | |
3360 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
3361 | stripe_align = mp->m_swidth; | |
3362 | else if (mp->m_dalign) | |
3363 | stripe_align = mp->m_dalign; | |
3364 | ||
10460994 DW |
3365 | if (ap->flags & XFS_BMAPI_COWFORK) |
3366 | align = xfs_get_cowextsz_hint(ap->ip); | |
3367 | else if (xfs_alloc_is_userdata(ap->datatype)) | |
1fccd5c8 | 3368 | align = xfs_get_extsz_hint(ap->ip); |
3f5dd6c6 | 3369 | if (align) { |
49f693fa DC |
3370 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, |
3371 | align, 0, ap->eof, 0, ap->conv, | |
3372 | &ap->offset, &ap->length); | |
3373 | ASSERT(!error); | |
3374 | ASSERT(ap->length); | |
3375 | } | |
ff105f75 DC |
3376 | |
3377 | ||
49f693fa DC |
3378 | nullfb = *ap->firstblock == NULLFSBLOCK; |
3379 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock); | |
3380 | if (nullfb) { | |
1fccd5c8 DC |
3381 | if (xfs_alloc_is_userdata(ap->datatype) && |
3382 | xfs_inode_is_filestream(ap->ip)) { | |
49f693fa DC |
3383 | ag = xfs_filestream_lookup_ag(ap->ip); |
3384 | ag = (ag != NULLAGNUMBER) ? ag : 0; | |
3385 | ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0); | |
3386 | } else { | |
3387 | ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino); | |
3388 | } | |
3389 | } else | |
3390 | ap->blkno = *ap->firstblock; | |
3391 | ||
3392 | xfs_bmap_adjacent(ap); | |
2bd0ea18 | 3393 | |
2bd0ea18 | 3394 | /* |
49f693fa DC |
3395 | * If allowed, use ap->blkno; otherwise must use firstblock since |
3396 | * it's in the right allocation group. | |
2bd0ea18 | 3397 | */ |
49f693fa DC |
3398 | if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno) |
3399 | ; | |
3400 | else | |
3401 | ap->blkno = *ap->firstblock; | |
2bd0ea18 | 3402 | /* |
49f693fa | 3403 | * Normal allocation, done through xfs_alloc_vextent. |
2bd0ea18 | 3404 | */ |
49f693fa DC |
3405 | tryagain = isaligned = 0; |
3406 | memset(&args, 0, sizeof(args)); | |
3407 | args.tp = ap->tp; | |
3408 | args.mp = mp; | |
3409 | args.fsbno = ap->blkno; | |
85aec44f | 3410 | xfs_rmap_skip_owner_update(&args.oinfo); |
49f693fa DC |
3411 | |
3412 | /* Trim the allocation back to the maximum an AG can fit. */ | |
b8a8d6e5 | 3413 | args.maxlen = MIN(ap->length, mp->m_ag_max_usable); |
49f693fa DC |
3414 | args.firstblock = *ap->firstblock; |
3415 | blen = 0; | |
3416 | if (nullfb) { | |
ff105f75 DC |
3417 | /* |
3418 | * Search for an allocation group with a single extent large | |
3419 | * enough for the request. If one isn't found, then adjust | |
3420 | * the minimum allocation size to the largest space found. | |
3421 | */ | |
1fccd5c8 DC |
3422 | if (xfs_alloc_is_userdata(ap->datatype) && |
3423 | xfs_inode_is_filestream(ap->ip)) | |
ff105f75 DC |
3424 | error = xfs_bmap_btalloc_filestreams(ap, &args, &blen); |
3425 | else | |
3426 | error = xfs_bmap_btalloc_nullfb(ap, &args, &blen); | |
a2ceac1f | 3427 | if (error) |
2bd0ea18 | 3428 | return error; |
f33cea1a | 3429 | } else if (ap->dfops->dop_low) { |
49f693fa DC |
3430 | if (xfs_inode_is_filestream(ap->ip)) |
3431 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3432 | else | |
3433 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3434 | args.total = args.minlen = ap->minlen; | |
3435 | } else { | |
3436 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3437 | args.total = ap->total; | |
3438 | args.minlen = ap->minlen; | |
3439 | } | |
3440 | /* apply extent size hints if obtained earlier */ | |
3f5dd6c6 | 3441 | if (align) { |
49f693fa DC |
3442 | args.prod = align; |
3443 | if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod))) | |
3444 | args.mod = (xfs_extlen_t)(args.prod - args.mod); | |
b2327e1a | 3445 | } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) { |
49f693fa DC |
3446 | args.prod = 1; |
3447 | args.mod = 0; | |
3448 | } else { | |
b2327e1a | 3449 | args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog; |
49f693fa DC |
3450 | if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod)))) |
3451 | args.mod = (xfs_extlen_t)(args.prod - args.mod); | |
2bd0ea18 NS |
3452 | } |
3453 | /* | |
49f693fa DC |
3454 | * If we are not low on available data blocks, and the |
3455 | * underlying logical volume manager is a stripe, and | |
3456 | * the file offset is zero then try to allocate data | |
3457 | * blocks on stripe unit boundary. | |
3458 | * NOTE: ap->aeof is only set if the allocation length | |
3459 | * is >= the stripe unit and the allocation offset is | |
3460 | * at the end of file. | |
2bd0ea18 | 3461 | */ |
f33cea1a | 3462 | if (!ap->dfops->dop_low && ap->aeof) { |
49f693fa | 3463 | if (!ap->offset) { |
ff105f75 | 3464 | args.alignment = stripe_align; |
49f693fa DC |
3465 | atype = args.type; |
3466 | isaligned = 1; | |
3467 | /* | |
3468 | * Adjust for alignment | |
3469 | */ | |
3470 | if (blen > args.alignment && blen <= args.maxlen) | |
3471 | args.minlen = blen - args.alignment; | |
3472 | args.minalignslop = 0; | |
3473 | } else { | |
3474 | /* | |
3475 | * First try an exact bno allocation. | |
3476 | * If it fails then do a near or start bno | |
3477 | * allocation with alignment turned on. | |
3478 | */ | |
3479 | atype = args.type; | |
3480 | tryagain = 1; | |
3481 | args.type = XFS_ALLOCTYPE_THIS_BNO; | |
3482 | args.alignment = 1; | |
3483 | /* | |
3484 | * Compute the minlen+alignment for the | |
3485 | * next case. Set slop so that the value | |
3486 | * of minlen+alignment+slop doesn't go up | |
3487 | * between the calls. | |
3488 | */ | |
ff105f75 DC |
3489 | if (blen > stripe_align && blen <= args.maxlen) |
3490 | nextminlen = blen - stripe_align; | |
49f693fa DC |
3491 | else |
3492 | nextminlen = args.minlen; | |
ff105f75 | 3493 | if (nextminlen + stripe_align > args.minlen + 1) |
49f693fa | 3494 | args.minalignslop = |
ff105f75 | 3495 | nextminlen + stripe_align - |
49f693fa DC |
3496 | args.minlen - 1; |
3497 | else | |
3498 | args.minalignslop = 0; | |
2bd0ea18 | 3499 | } |
49f693fa DC |
3500 | } else { |
3501 | args.alignment = 1; | |
3502 | args.minalignslop = 0; | |
3503 | } | |
3504 | args.minleft = ap->minleft; | |
3505 | args.wasdel = ap->wasdel; | |
cf8ce220 | 3506 | args.resv = XFS_AG_RESV_NONE; |
1fccd5c8 DC |
3507 | args.datatype = ap->datatype; |
3508 | if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) | |
9542ae13 DC |
3509 | args.ip = ap->ip; |
3510 | ||
3511 | error = xfs_alloc_vextent(&args); | |
3512 | if (error) | |
49f693fa | 3513 | return error; |
9542ae13 | 3514 | |
49f693fa DC |
3515 | if (tryagain && args.fsbno == NULLFSBLOCK) { |
3516 | /* | |
3517 | * Exact allocation failed. Now try with alignment | |
3518 | * turned on. | |
3519 | */ | |
3520 | args.type = atype; | |
3521 | args.fsbno = ap->blkno; | |
ff105f75 | 3522 | args.alignment = stripe_align; |
49f693fa DC |
3523 | args.minlen = nextminlen; |
3524 | args.minalignslop = 0; | |
3525 | isaligned = 1; | |
3526 | if ((error = xfs_alloc_vextent(&args))) | |
3527 | return error; | |
3528 | } | |
3529 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
2bd0ea18 | 3530 | /* |
49f693fa DC |
3531 | * allocation failed, so turn off alignment and |
3532 | * try again. | |
2bd0ea18 | 3533 | */ |
49f693fa DC |
3534 | args.type = atype; |
3535 | args.fsbno = ap->blkno; | |
3536 | args.alignment = 0; | |
3537 | if ((error = xfs_alloc_vextent(&args))) | |
3538 | return error; | |
3539 | } | |
3540 | if (args.fsbno == NULLFSBLOCK && nullfb && | |
3541 | args.minlen > ap->minlen) { | |
3542 | args.minlen = ap->minlen; | |
3543 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3544 | args.fsbno = ap->blkno; | |
3545 | if ((error = xfs_alloc_vextent(&args))) | |
3546 | return error; | |
3547 | } | |
3548 | if (args.fsbno == NULLFSBLOCK && nullfb) { | |
3549 | args.fsbno = 0; | |
3550 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3551 | args.total = ap->minlen; | |
49f693fa DC |
3552 | if ((error = xfs_alloc_vextent(&args))) |
3553 | return error; | |
f33cea1a | 3554 | ap->dfops->dop_low = true; |
49f693fa DC |
3555 | } |
3556 | if (args.fsbno != NULLFSBLOCK) { | |
2bd0ea18 | 3557 | /* |
49f693fa DC |
3558 | * check the allocation happened at the same or higher AG than |
3559 | * the first block that was allocated. | |
2bd0ea18 | 3560 | */ |
49f693fa | 3561 | ASSERT(*ap->firstblock == NULLFSBLOCK || |
6df6a3ab CH |
3562 | XFS_FSB_TO_AGNO(mp, *ap->firstblock) <= |
3563 | XFS_FSB_TO_AGNO(mp, args.fsbno)); | |
49f693fa DC |
3564 | |
3565 | ap->blkno = args.fsbno; | |
3566 | if (*ap->firstblock == NULLFSBLOCK) | |
3567 | *ap->firstblock = args.fsbno; | |
6df6a3ab | 3568 | ASSERT(nullfb || fb_agno <= args.agno); |
49f693fa | 3569 | ap->length = args.len; |
1277a5e0 DW |
3570 | if (!(ap->flags & XFS_BMAPI_COWFORK)) |
3571 | ap->ip->i_d.di_nblocks += args.len; | |
49f693fa DC |
3572 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); |
3573 | if (ap->wasdel) | |
3574 | ap->ip->i_delayed_blks -= args.len; | |
2bd0ea18 | 3575 | /* |
49f693fa DC |
3576 | * Adjust the disk quota also. This was reserved |
3577 | * earlier. | |
2bd0ea18 | 3578 | */ |
49f693fa DC |
3579 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, |
3580 | ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : | |
3581 | XFS_TRANS_DQ_BCOUNT, | |
3582 | (long) args.len); | |
3583 | } else { | |
3584 | ap->blkno = NULLFSBLOCK; | |
3585 | ap->length = 0; | |
2bd0ea18 | 3586 | } |
2bd0ea18 | 3587 | return 0; |
56b2de80 DC |
3588 | } |
3589 | ||
3590 | /* | |
49f693fa DC |
3591 | * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file. |
3592 | * It figures out where to ask the underlying allocator to put the new extent. | |
56b2de80 | 3593 | */ |
49f693fa DC |
3594 | STATIC int |
3595 | xfs_bmap_alloc( | |
613e6057 | 3596 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
56b2de80 | 3597 | { |
1fccd5c8 DC |
3598 | if (XFS_IS_REALTIME_INODE(ap->ip) && |
3599 | xfs_alloc_is_userdata(ap->datatype)) | |
49f693fa DC |
3600 | return xfs_bmap_rtalloc(ap); |
3601 | return xfs_bmap_btalloc(ap); | |
56b2de80 | 3602 | } |
56b2de80 | 3603 | |
b3fd8db7 DW |
3604 | /* Trim extent to fit a logical block range. */ |
3605 | void | |
3606 | xfs_trim_extent( | |
3607 | struct xfs_bmbt_irec *irec, | |
3608 | xfs_fileoff_t bno, | |
3609 | xfs_filblks_t len) | |
3610 | { | |
3611 | xfs_fileoff_t distance; | |
3612 | xfs_fileoff_t end = bno + len; | |
3613 | ||
3614 | if (irec->br_startoff + irec->br_blockcount <= bno || | |
3615 | irec->br_startoff >= end) { | |
3616 | irec->br_blockcount = 0; | |
3617 | return; | |
3618 | } | |
3619 | ||
3620 | if (irec->br_startoff < bno) { | |
3621 | distance = bno - irec->br_startoff; | |
3622 | if (isnullstartblock(irec->br_startblock)) | |
3623 | irec->br_startblock = DELAYSTARTBLOCK; | |
3624 | if (irec->br_startblock != DELAYSTARTBLOCK && | |
3625 | irec->br_startblock != HOLESTARTBLOCK) | |
3626 | irec->br_startblock += distance; | |
3627 | irec->br_startoff += distance; | |
3628 | irec->br_blockcount -= distance; | |
3629 | } | |
3630 | ||
3631 | if (end < irec->br_startoff + irec->br_blockcount) { | |
3632 | distance = irec->br_startoff + irec->br_blockcount - end; | |
3633 | irec->br_blockcount -= distance; | |
3634 | } | |
3635 | } | |
3636 | ||
ca75eb20 BF |
3637 | /* trim extent to within eof */ |
3638 | void | |
3639 | xfs_trim_extent_eof( | |
3640 | struct xfs_bmbt_irec *irec, | |
3641 | struct xfs_inode *ip) | |
3642 | ||
3643 | { | |
3644 | xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount, | |
3645 | i_size_read(VFS_I(ip)))); | |
3646 | } | |
3647 | ||
2bd0ea18 | 3648 | /* |
a2ceac1f | 3649 | * Trim the returned map to the required bounds |
2bd0ea18 | 3650 | */ |
a2ceac1f DC |
3651 | STATIC void |
3652 | xfs_bmapi_trim_map( | |
3653 | struct xfs_bmbt_irec *mval, | |
3654 | struct xfs_bmbt_irec *got, | |
3655 | xfs_fileoff_t *bno, | |
3656 | xfs_filblks_t len, | |
3657 | xfs_fileoff_t obno, | |
3658 | xfs_fileoff_t end, | |
3659 | int n, | |
3660 | int flags) | |
2bd0ea18 | 3661 | { |
a2ceac1f DC |
3662 | if ((flags & XFS_BMAPI_ENTIRE) || |
3663 | got->br_startoff + got->br_blockcount <= obno) { | |
3664 | *mval = *got; | |
3665 | if (isnullstartblock(got->br_startblock)) | |
3666 | mval->br_startblock = DELAYSTARTBLOCK; | |
3667 | return; | |
63be04eb | 3668 | } |
a2ceac1f DC |
3669 | |
3670 | if (obno > *bno) | |
3671 | *bno = obno; | |
3672 | ASSERT((*bno >= obno) || (n == 0)); | |
3673 | ASSERT(*bno < end); | |
3674 | mval->br_startoff = *bno; | |
3675 | if (isnullstartblock(got->br_startblock)) | |
3676 | mval->br_startblock = DELAYSTARTBLOCK; | |
2bd0ea18 | 3677 | else |
a2ceac1f DC |
3678 | mval->br_startblock = got->br_startblock + |
3679 | (*bno - got->br_startoff); | |
2bd0ea18 | 3680 | /* |
a2ceac1f DC |
3681 | * Return the minimum of what we got and what we asked for for |
3682 | * the length. We can use the len variable here because it is | |
3683 | * modified below and we could have been there before coming | |
3684 | * here if the first part of the allocation didn't overlap what | |
3685 | * was asked for. | |
2bd0ea18 | 3686 | */ |
a2ceac1f DC |
3687 | mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno, |
3688 | got->br_blockcount - (*bno - got->br_startoff)); | |
3689 | mval->br_state = got->br_state; | |
3690 | ASSERT(mval->br_blockcount <= len); | |
3691 | return; | |
3692 | } | |
56b2de80 | 3693 | |
a2ceac1f DC |
3694 | /* |
3695 | * Update and validate the extent map to return | |
3696 | */ | |
3697 | STATIC void | |
3698 | xfs_bmapi_update_map( | |
3699 | struct xfs_bmbt_irec **map, | |
3700 | xfs_fileoff_t *bno, | |
3701 | xfs_filblks_t *len, | |
3702 | xfs_fileoff_t obno, | |
3703 | xfs_fileoff_t end, | |
3704 | int *n, | |
3705 | int flags) | |
3706 | { | |
3707 | xfs_bmbt_irec_t *mval = *map; | |
3708 | ||
3709 | ASSERT((flags & XFS_BMAPI_ENTIRE) || | |
3710 | ((mval->br_startoff + mval->br_blockcount) <= end)); | |
3711 | ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) || | |
3712 | (mval->br_startoff < obno)); | |
3713 | ||
3714 | *bno = mval->br_startoff + mval->br_blockcount; | |
3715 | *len = end - *bno; | |
3716 | if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) { | |
3717 | /* update previous map with new information */ | |
3718 | ASSERT(mval->br_startblock == mval[-1].br_startblock); | |
3719 | ASSERT(mval->br_blockcount > mval[-1].br_blockcount); | |
3720 | ASSERT(mval->br_state == mval[-1].br_state); | |
3721 | mval[-1].br_blockcount = mval->br_blockcount; | |
3722 | mval[-1].br_state = mval->br_state; | |
3723 | } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK && | |
3724 | mval[-1].br_startblock != DELAYSTARTBLOCK && | |
3725 | mval[-1].br_startblock != HOLESTARTBLOCK && | |
3726 | mval->br_startblock == mval[-1].br_startblock + | |
3727 | mval[-1].br_blockcount && | |
3728 | ((flags & XFS_BMAPI_IGSTATE) || | |
3729 | mval[-1].br_state == mval->br_state)) { | |
3730 | ASSERT(mval->br_startoff == | |
3731 | mval[-1].br_startoff + mval[-1].br_blockcount); | |
3732 | mval[-1].br_blockcount += mval->br_blockcount; | |
3733 | } else if (*n > 0 && | |
3734 | mval->br_startblock == DELAYSTARTBLOCK && | |
3735 | mval[-1].br_startblock == DELAYSTARTBLOCK && | |
3736 | mval->br_startoff == | |
3737 | mval[-1].br_startoff + mval[-1].br_blockcount) { | |
3738 | mval[-1].br_blockcount += mval->br_blockcount; | |
3739 | mval[-1].br_state = mval->br_state; | |
3740 | } else if (!((*n == 0) && | |
3741 | ((mval->br_startoff + mval->br_blockcount) <= | |
3742 | obno))) { | |
3743 | mval++; | |
3744 | (*n)++; | |
3745 | } | |
3746 | *map = mval; | |
3747 | } | |
399ab595 | 3748 | |
a2ceac1f DC |
3749 | /* |
3750 | * Map file blocks to filesystem blocks without allocation. | |
3751 | */ | |
3752 | int | |
3753 | xfs_bmapi_read( | |
3754 | struct xfs_inode *ip, | |
3755 | xfs_fileoff_t bno, | |
3756 | xfs_filblks_t len, | |
3757 | struct xfs_bmbt_irec *mval, | |
3758 | int *nmap, | |
3759 | int flags) | |
3760 | { | |
3761 | struct xfs_mount *mp = ip->i_mount; | |
3762 | struct xfs_ifork *ifp; | |
3763 | struct xfs_bmbt_irec got; | |
a2ceac1f DC |
3764 | xfs_fileoff_t obno; |
3765 | xfs_fileoff_t end; | |
c2d73ed3 | 3766 | xfs_extnum_t idx; |
a2ceac1f | 3767 | int error; |
c2d73ed3 | 3768 | bool eof = false; |
a2ceac1f | 3769 | int n = 0; |
cb8a004a | 3770 | int whichfork = xfs_bmapi_whichfork(flags); |
399ab595 | 3771 | |
a2ceac1f DC |
3772 | ASSERT(*nmap >= 1); |
3773 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE| | |
cb8a004a | 3774 | XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK))); |
ff105f75 | 3775 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)); |
062998e3 | 3776 | |
a2ceac1f DC |
3777 | if (unlikely(XFS_TEST_ERROR( |
3778 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
3779 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 3780 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a2ceac1f | 3781 | XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp); |
12b53197 | 3782 | return -EFSCORRUPTED; |
a2ceac1f | 3783 | } |
062998e3 | 3784 | |
a2ceac1f | 3785 | if (XFS_FORCED_SHUTDOWN(mp)) |
12b53197 | 3786 | return -EIO; |
399ab595 | 3787 | |
79896434 | 3788 | XFS_STATS_INC(mp, xs_blk_mapr); |
a2ceac1f DC |
3789 | |
3790 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
3791 | ||
cb8a004a DW |
3792 | /* No CoW fork? Return a hole. */ |
3793 | if (whichfork == XFS_COW_FORK && !ifp) { | |
3794 | mval->br_startoff = bno; | |
3795 | mval->br_startblock = HOLESTARTBLOCK; | |
3796 | mval->br_blockcount = len; | |
3797 | mval->br_state = XFS_EXT_NORM; | |
3798 | *nmap = 1; | |
3799 | return 0; | |
3800 | } | |
3801 | ||
a2ceac1f DC |
3802 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
3803 | error = xfs_iread_extents(NULL, ip, whichfork); | |
3804 | if (error) | |
3805 | return error; | |
3806 | } | |
3807 | ||
c2d73ed3 CH |
3808 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) |
3809 | eof = true; | |
a2ceac1f DC |
3810 | end = bno + len; |
3811 | obno = bno; | |
3812 | ||
3813 | while (bno < end && n < *nmap) { | |
3814 | /* Reading past eof, act as though there's a hole up to end. */ | |
3815 | if (eof) | |
3816 | got.br_startoff = end; | |
3817 | if (got.br_startoff > bno) { | |
3818 | /* Reading in a hole. */ | |
2bd0ea18 NS |
3819 | mval->br_startoff = bno; |
3820 | mval->br_startblock = HOLESTARTBLOCK; | |
3821 | mval->br_blockcount = | |
3822 | XFS_FILBLKS_MIN(len, got.br_startoff - bno); | |
3823 | mval->br_state = XFS_EXT_NORM; | |
3824 | bno += mval->br_blockcount; | |
3825 | len -= mval->br_blockcount; | |
3826 | mval++; | |
3827 | n++; | |
3828 | continue; | |
3829 | } | |
a2ceac1f DC |
3830 | |
3831 | /* set up the extent map to return. */ | |
3832 | xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags); | |
3833 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
3834 | ||
3835 | /* If we're done, stop now. */ | |
3836 | if (bno >= end || n >= *nmap) | |
3837 | break; | |
3838 | ||
3839 | /* Else go on to the next record. */ | |
c2d73ed3 CH |
3840 | if (!xfs_iext_get_extent(ifp, ++idx, &got)) |
3841 | eof = true; | |
a2ceac1f DC |
3842 | } |
3843 | *nmap = n; | |
3844 | return 0; | |
3845 | } | |
3846 | ||
7a868ee8 BF |
3847 | /* |
3848 | * Add a delayed allocation extent to an inode. Blocks are reserved from the | |
3849 | * global pool and the extent inserted into the inode in-core extent tree. | |
3850 | * | |
3851 | * On entry, got refers to the first extent beyond the offset of the extent to | |
3852 | * allocate or eof is specified if no such extent exists. On return, got refers | |
3853 | * to the extent record that was inserted to the inode fork. | |
3854 | * | |
3855 | * Note that the allocated extent may have been merged with contiguous extents | |
3856 | * during insertion into the inode fork. Thus, got does not reflect the current | |
3857 | * state of the inode fork on return. If necessary, the caller can use lastx to | |
3858 | * look up the updated record in the inode fork. | |
3859 | */ | |
4488e421 | 3860 | int |
a2ceac1f DC |
3861 | xfs_bmapi_reserve_delalloc( |
3862 | struct xfs_inode *ip, | |
cc66acaf | 3863 | int whichfork, |
f7b1a8b1 | 3864 | xfs_fileoff_t off, |
a2ceac1f | 3865 | xfs_filblks_t len, |
f7b1a8b1 | 3866 | xfs_filblks_t prealloc, |
a2ceac1f | 3867 | struct xfs_bmbt_irec *got, |
a2ceac1f DC |
3868 | xfs_extnum_t *lastx, |
3869 | int eof) | |
3870 | { | |
3871 | struct xfs_mount *mp = ip->i_mount; | |
cc66acaf | 3872 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
a2ceac1f DC |
3873 | xfs_extlen_t alen; |
3874 | xfs_extlen_t indlen; | |
3875 | char rt = XFS_IS_REALTIME_INODE(ip); | |
3876 | xfs_extlen_t extsz; | |
3877 | int error; | |
f7b1a8b1 | 3878 | xfs_fileoff_t aoff = off; |
a2ceac1f | 3879 | |
f7b1a8b1 BF |
3880 | /* |
3881 | * Cap the alloc length. Keep track of prealloc so we know whether to | |
3882 | * tag the inode before we return. | |
3883 | */ | |
3884 | alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN); | |
a2ceac1f DC |
3885 | if (!eof) |
3886 | alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff); | |
f7b1a8b1 BF |
3887 | if (prealloc && alen >= len) |
3888 | prealloc = alen - len; | |
a2ceac1f DC |
3889 | |
3890 | /* Figure out the extent size, adjust alen */ | |
10460994 DW |
3891 | if (whichfork == XFS_COW_FORK) |
3892 | extsz = xfs_get_cowextsz_hint(ip); | |
3893 | else | |
3894 | extsz = xfs_get_extsz_hint(ip); | |
a2ceac1f | 3895 | if (extsz) { |
d41d2303 CH |
3896 | struct xfs_bmbt_irec prev; |
3897 | ||
3898 | if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev)) | |
3899 | prev.br_startoff = NULLFILEOFF; | |
3900 | ||
3901 | error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof, | |
a2ceac1f DC |
3902 | 1, 0, &aoff, &alen); |
3903 | ASSERT(!error); | |
3904 | } | |
3905 | ||
3906 | if (rt) | |
3907 | extsz = alen / mp->m_sb.sb_rextsize; | |
3908 | ||
3909 | /* | |
3910 | * Make a transaction-less quota reservation for delayed allocation | |
3911 | * blocks. This number gets adjusted later. We return if we haven't | |
3912 | * allocated blocks already inside this loop. | |
3913 | */ | |
3914 | error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0, | |
3915 | rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
3916 | if (error) | |
3917 | return error; | |
3918 | ||
3919 | /* | |
3920 | * Split changing sb for alen and indlen since they could be coming | |
3921 | * from different places. | |
3922 | */ | |
3923 | indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen); | |
3924 | ASSERT(indlen > 0); | |
3925 | ||
3926 | if (rt) { | |
19ebedcf | 3927 | error = xfs_mod_frextents(mp, -((int64_t)extsz)); |
a2ceac1f | 3928 | } else { |
19ebedcf | 3929 | error = xfs_mod_fdblocks(mp, -((int64_t)alen), false); |
a2ceac1f DC |
3930 | } |
3931 | ||
3932 | if (error) | |
3933 | goto out_unreserve_quota; | |
3934 | ||
19ebedcf | 3935 | error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false); |
a2ceac1f DC |
3936 | if (error) |
3937 | goto out_unreserve_blocks; | |
3938 | ||
3939 | ||
3940 | ip->i_delayed_blks += alen; | |
3941 | ||
3942 | got->br_startoff = aoff; | |
3943 | got->br_startblock = nullstartblock(indlen); | |
3944 | got->br_blockcount = alen; | |
3945 | got->br_state = XFS_EXT_NORM; | |
a2ceac1f | 3946 | |
7a868ee8 | 3947 | xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got); |
a2ceac1f | 3948 | |
f7b1a8b1 BF |
3949 | /* |
3950 | * Tag the inode if blocks were preallocated. Note that COW fork | |
3951 | * preallocation can occur at the start or end of the extent, even when | |
3952 | * prealloc == 0, so we must also check the aligned offset and length. | |
3953 | */ | |
3954 | if (whichfork == XFS_DATA_FORK && prealloc) | |
3955 | xfs_inode_set_eofblocks_tag(ip); | |
3956 | if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len)) | |
3957 | xfs_inode_set_cowblocks_tag(ip); | |
3958 | ||
a2ceac1f DC |
3959 | return 0; |
3960 | ||
3961 | out_unreserve_blocks: | |
3962 | if (rt) | |
19ebedcf | 3963 | xfs_mod_frextents(mp, extsz); |
a2ceac1f | 3964 | else |
19ebedcf | 3965 | xfs_mod_fdblocks(mp, alen, false); |
a2ceac1f DC |
3966 | out_unreserve_quota: |
3967 | if (XFS_IS_QUOTA_ON(mp)) | |
3968 | xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ? | |
3969 | XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
3970 | return error; | |
3971 | } | |
3972 | ||
ff105f75 DC |
3973 | static int |
3974 | xfs_bmapi_allocate( | |
a2ceac1f DC |
3975 | struct xfs_bmalloca *bma) |
3976 | { | |
3977 | struct xfs_mount *mp = bma->ip->i_mount; | |
1277a5e0 | 3978 | int whichfork = xfs_bmapi_whichfork(bma->flags); |
a2ceac1f DC |
3979 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
3980 | int tmp_logflags = 0; | |
3981 | int error; | |
3982 | ||
3983 | ASSERT(bma->length > 0); | |
3984 | ||
3985 | /* | |
3986 | * For the wasdelay case, we could also just allocate the stuff asked | |
3987 | * for in this bmap call but that wouldn't be as good. | |
3988 | */ | |
3989 | if (bma->wasdel) { | |
3990 | bma->length = (xfs_extlen_t)bma->got.br_blockcount; | |
3991 | bma->offset = bma->got.br_startoff; | |
99a663c1 CH |
3992 | if (bma->idx) |
3993 | xfs_iext_get_extent(ifp, bma->idx - 1, &bma->prev); | |
a2ceac1f DC |
3994 | } else { |
3995 | bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN); | |
3996 | if (!bma->eof) | |
3997 | bma->length = XFS_FILBLKS_MIN(bma->length, | |
3998 | bma->got.br_startoff - bma->offset); | |
3999 | } | |
4000 | ||
4001 | /* | |
1fccd5c8 DC |
4002 | * Set the data type being allocated. For the data fork, the first data |
4003 | * in the file is treated differently to all other allocations. For the | |
4004 | * attribute fork, we only need to ensure the allocated range is not on | |
4005 | * the busy list. | |
a2ceac1f DC |
4006 | */ |
4007 | if (!(bma->flags & XFS_BMAPI_METADATA)) { | |
1fccd5c8 DC |
4008 | bma->datatype = XFS_ALLOC_NOBUSY; |
4009 | if (whichfork == XFS_DATA_FORK) { | |
4010 | if (bma->offset == 0) | |
4011 | bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA; | |
4012 | else | |
4013 | bma->datatype |= XFS_ALLOC_USERDATA; | |
4014 | } | |
9542ae13 | 4015 | if (bma->flags & XFS_BMAPI_ZERO) |
1fccd5c8 | 4016 | bma->datatype |= XFS_ALLOC_USERDATA_ZERO; |
a2ceac1f DC |
4017 | } |
4018 | ||
4019 | bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1; | |
4020 | ||
4021 | /* | |
4022 | * Only want to do the alignment at the eof if it is userdata and | |
4023 | * allocation length is larger than a stripe unit. | |
4024 | */ | |
4025 | if (mp->m_dalign && bma->length >= mp->m_dalign && | |
4026 | !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) { | |
4027 | error = xfs_bmap_isaeof(bma, whichfork); | |
4028 | if (error) | |
4029 | return error; | |
4030 | } | |
4031 | ||
a2ceac1f DC |
4032 | error = xfs_bmap_alloc(bma); |
4033 | if (error) | |
4034 | return error; | |
4035 | ||
a2ceac1f DC |
4036 | if (bma->cur) |
4037 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
4038 | if (bma->blkno == NULLFSBLOCK) | |
4039 | return 0; | |
4040 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4041 | bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork); | |
4042 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
f33cea1a | 4043 | bma->cur->bc_private.b.dfops = bma->dfops; |
a2ceac1f DC |
4044 | } |
4045 | /* | |
4046 | * Bump the number of extents we've allocated | |
4047 | * in this call. | |
4048 | */ | |
4049 | bma->nallocs++; | |
4050 | ||
4051 | if (bma->cur) | |
4052 | bma->cur->bc_private.b.flags = | |
4053 | bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; | |
4054 | ||
4055 | bma->got.br_startoff = bma->offset; | |
4056 | bma->got.br_startblock = bma->blkno; | |
4057 | bma->got.br_blockcount = bma->length; | |
4058 | bma->got.br_state = XFS_EXT_NORM; | |
4059 | ||
4060 | /* | |
4072e4b4 DW |
4061 | * In the data fork, a wasdelay extent has been initialized, so |
4062 | * shouldn't be flagged as unwritten. | |
4063 | * | |
4064 | * For the cow fork, however, we convert delalloc reservations | |
4065 | * (extents allocated for speculative preallocation) to | |
4066 | * allocated unwritten extents, and only convert the unwritten | |
4067 | * extents to real extents when we're about to write the data. | |
a2ceac1f | 4068 | */ |
4072e4b4 DW |
4069 | if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) && |
4070 | (bma->flags & XFS_BMAPI_PREALLOC) && | |
a2ceac1f DC |
4071 | xfs_sb_version_hasextflgbit(&mp->m_sb)) |
4072 | bma->got.br_state = XFS_EXT_UNWRITTEN; | |
4073 | ||
4074 | if (bma->wasdel) | |
1277a5e0 | 4075 | error = xfs_bmap_add_extent_delay_real(bma, whichfork); |
a2ceac1f | 4076 | else |
972432f2 CH |
4077 | error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip, |
4078 | whichfork, &bma->idx, &bma->cur, &bma->got, | |
4079 | bma->firstblock, bma->dfops, &bma->logflags); | |
a2ceac1f DC |
4080 | |
4081 | bma->logflags |= tmp_logflags; | |
4082 | if (error) | |
4083 | return error; | |
4084 | ||
4085 | /* | |
4086 | * Update our extent pointer, given that xfs_bmap_add_extent_delay_real | |
4087 | * or xfs_bmap_add_extent_hole_real might have merged it into one of | |
4088 | * the neighbouring ones. | |
4089 | */ | |
99a663c1 | 4090 | xfs_iext_get_extent(ifp, bma->idx, &bma->got); |
a2ceac1f DC |
4091 | |
4092 | ASSERT(bma->got.br_startoff <= bma->offset); | |
4093 | ASSERT(bma->got.br_startoff + bma->got.br_blockcount >= | |
4094 | bma->offset + bma->length); | |
4095 | ASSERT(bma->got.br_state == XFS_EXT_NORM || | |
4096 | bma->got.br_state == XFS_EXT_UNWRITTEN); | |
4097 | return 0; | |
4098 | } | |
4099 | ||
a2ceac1f DC |
4100 | STATIC int |
4101 | xfs_bmapi_convert_unwritten( | |
4102 | struct xfs_bmalloca *bma, | |
4103 | struct xfs_bmbt_irec *mval, | |
4104 | xfs_filblks_t len, | |
4105 | int flags) | |
4106 | { | |
cb8a004a | 4107 | int whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4108 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4109 | int tmp_logflags = 0; | |
4110 | int error; | |
4111 | ||
4112 | /* check if we need to do unwritten->real conversion */ | |
4113 | if (mval->br_state == XFS_EXT_UNWRITTEN && | |
4114 | (flags & XFS_BMAPI_PREALLOC)) | |
4115 | return 0; | |
4116 | ||
4117 | /* check if we need to do real->unwritten conversion */ | |
4118 | if (mval->br_state == XFS_EXT_NORM && | |
4119 | (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) != | |
4120 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) | |
4121 | return 0; | |
4122 | ||
4123 | /* | |
4124 | * Modify (by adding) the state flag, if writing. | |
4125 | */ | |
4126 | ASSERT(mval->br_blockcount <= len); | |
4127 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4128 | bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp, | |
4129 | bma->ip, whichfork); | |
4130 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
f33cea1a | 4131 | bma->cur->bc_private.b.dfops = bma->dfops; |
a2ceac1f DC |
4132 | } |
4133 | mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN) | |
4134 | ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
4135 | ||
9542ae13 DC |
4136 | /* |
4137 | * Before insertion into the bmbt, zero the range being converted | |
4138 | * if required. | |
4139 | */ | |
4140 | if (flags & XFS_BMAPI_ZERO) { | |
4141 | error = xfs_zero_extent(bma->ip, mval->br_startblock, | |
4142 | mval->br_blockcount); | |
4143 | if (error) | |
4144 | return error; | |
4145 | } | |
4146 | ||
4072e4b4 DW |
4147 | error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork, |
4148 | &bma->idx, &bma->cur, mval, bma->firstblock, bma->dfops, | |
a2ceac1f | 4149 | &tmp_logflags); |
23fc058d BF |
4150 | /* |
4151 | * Log the inode core unconditionally in the unwritten extent conversion | |
4152 | * path because the conversion might not have done so (e.g., if the | |
4153 | * extent count hasn't changed). We need to make sure the inode is dirty | |
4154 | * in the transaction for the sake of fsync(), even if nothing has | |
4155 | * changed, because fsync() will not force the log for this transaction | |
4156 | * unless it sees the inode pinned. | |
4072e4b4 DW |
4157 | * |
4158 | * Note: If we're only converting cow fork extents, there aren't | |
4159 | * any on-disk updates to make, so we don't need to log anything. | |
23fc058d | 4160 | */ |
4072e4b4 DW |
4161 | if (whichfork != XFS_COW_FORK) |
4162 | bma->logflags |= tmp_logflags | XFS_ILOG_CORE; | |
a2ceac1f DC |
4163 | if (error) |
4164 | return error; | |
4165 | ||
4166 | /* | |
4167 | * Update our extent pointer, given that | |
4168 | * xfs_bmap_add_extent_unwritten_real might have merged it into one | |
4169 | * of the neighbouring ones. | |
4170 | */ | |
99a663c1 | 4171 | xfs_iext_get_extent(ifp, bma->idx, &bma->got); |
a2ceac1f DC |
4172 | |
4173 | /* | |
4174 | * We may have combined previously unwritten space with written space, | |
4175 | * so generate another request. | |
4176 | */ | |
4177 | if (mval->br_blockcount < len) | |
12b53197 | 4178 | return -EAGAIN; |
a2ceac1f DC |
4179 | return 0; |
4180 | } | |
4181 | ||
4182 | /* | |
4183 | * Map file blocks to filesystem blocks, and allocate blocks or convert the | |
4184 | * extent state if necessary. Details behaviour is controlled by the flags | |
4185 | * parameter. Only allocates blocks from a single allocation group, to avoid | |
4186 | * locking problems. | |
4187 | * | |
4188 | * The returned value in "firstblock" from the first call in a transaction | |
4189 | * must be remembered and presented to subsequent calls in "firstblock". | |
4190 | * An upper bound for the number of blocks to be allocated is supplied to | |
4191 | * the first call in "total"; if no allocation group has that many free | |
4192 | * blocks then the call will fail (return NULLFSBLOCK in "firstblock"). | |
4193 | */ | |
4194 | int | |
4195 | xfs_bmapi_write( | |
4196 | struct xfs_trans *tp, /* transaction pointer */ | |
4197 | struct xfs_inode *ip, /* incore inode */ | |
4198 | xfs_fileoff_t bno, /* starting file offs. mapped */ | |
4199 | xfs_filblks_t len, /* length to map in file */ | |
4200 | int flags, /* XFS_BMAPI_... */ | |
4201 | xfs_fsblock_t *firstblock, /* first allocated block | |
4202 | controls a.g. for allocs */ | |
4203 | xfs_extlen_t total, /* total blocks needed */ | |
4204 | struct xfs_bmbt_irec *mval, /* output: map values */ | |
4205 | int *nmap, /* i/o: mval size/count */ | |
f33cea1a | 4206 | struct xfs_defer_ops *dfops) /* i/o: list extents to free */ |
a2ceac1f DC |
4207 | { |
4208 | struct xfs_mount *mp = ip->i_mount; | |
4209 | struct xfs_ifork *ifp; | |
389b3b07 | 4210 | struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */ |
a2ceac1f | 4211 | xfs_fileoff_t end; /* end of mapped file region */ |
82411945 | 4212 | bool eof = false; /* after the end of extents */ |
a2ceac1f DC |
4213 | int error; /* error return */ |
4214 | int n; /* current extent index */ | |
4215 | xfs_fileoff_t obno; /* old block number (offset) */ | |
4216 | int whichfork; /* data or attr fork */ | |
a2ceac1f DC |
4217 | |
4218 | #ifdef DEBUG | |
4219 | xfs_fileoff_t orig_bno; /* original block number value */ | |
4220 | int orig_flags; /* original flags arg value */ | |
4221 | xfs_filblks_t orig_len; /* original value of len arg */ | |
4222 | struct xfs_bmbt_irec *orig_mval; /* original value of mval */ | |
4223 | int orig_nmap; /* original value of *nmap */ | |
4224 | ||
4225 | orig_bno = bno; | |
4226 | orig_len = len; | |
4227 | orig_flags = flags; | |
4228 | orig_mval = mval; | |
4229 | orig_nmap = *nmap; | |
4230 | #endif | |
1277a5e0 | 4231 | whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4232 | |
4233 | ASSERT(*nmap >= 1); | |
4234 | ASSERT(*nmap <= XFS_BMAP_MAX_NMAP); | |
4235 | ASSERT(!(flags & XFS_BMAPI_IGSTATE)); | |
4072e4b4 DW |
4236 | ASSERT(tp != NULL || |
4237 | (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) == | |
4238 | (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)); | |
a2ceac1f | 4239 | ASSERT(len > 0); |
3f17ed4b | 4240 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL); |
ff105f75 | 4241 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b44aa85 | 4242 | ASSERT(!(flags & XFS_BMAPI_REMAP)); |
a2ceac1f | 4243 | |
9542ae13 DC |
4244 | /* zeroing is for currently only for data extents, not metadata */ |
4245 | ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) != | |
4246 | (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)); | |
4247 | /* | |
4248 | * we can allocate unwritten extents or pre-zero allocated blocks, | |
4249 | * but it makes no sense to do both at once. This would result in | |
4250 | * zeroing the unwritten extent twice, but it still being an | |
4251 | * unwritten extent.... | |
4252 | */ | |
4253 | ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) != | |
4254 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)); | |
4255 | ||
a2ceac1f DC |
4256 | if (unlikely(XFS_TEST_ERROR( |
4257 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
3f17ed4b | 4258 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), |
e2a190dd | 4259 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a2ceac1f | 4260 | XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp); |
12b53197 | 4261 | return -EFSCORRUPTED; |
a2ceac1f DC |
4262 | } |
4263 | ||
4264 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 4265 | return -EIO; |
a2ceac1f DC |
4266 | |
4267 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
4268 | ||
79896434 | 4269 | XFS_STATS_INC(mp, xs_blk_mapw); |
a2ceac1f | 4270 | |
a2ceac1f DC |
4271 | if (*firstblock == NULLFSBLOCK) { |
4272 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE) | |
4273 | bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1; | |
4274 | else | |
4275 | bma.minleft = 1; | |
4276 | } else { | |
4277 | bma.minleft = 0; | |
4278 | } | |
4279 | ||
4280 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
4281 | error = xfs_iread_extents(tp, ip, whichfork); | |
4282 | if (error) | |
4283 | goto error0; | |
4284 | } | |
4285 | ||
a2ceac1f DC |
4286 | n = 0; |
4287 | end = bno + len; | |
4288 | obno = bno; | |
4289 | ||
82411945 CH |
4290 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got)) |
4291 | eof = true; | |
4292 | if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev)) | |
4293 | bma.prev.br_startoff = NULLFILEOFF; | |
a2ceac1f DC |
4294 | bma.tp = tp; |
4295 | bma.ip = ip; | |
4296 | bma.total = total; | |
1fccd5c8 | 4297 | bma.datatype = 0; |
f33cea1a | 4298 | bma.dfops = dfops; |
a2ceac1f DC |
4299 | bma.firstblock = firstblock; |
4300 | ||
4301 | while (bno < end && n < *nmap) { | |
7075a23f CH |
4302 | bool need_alloc = false, wasdelay = false; |
4303 | ||
4304 | /* in hole or beyoned EOF? */ | |
4305 | if (eof || bma.got.br_startoff > bno) { | |
4306 | if (flags & XFS_BMAPI_DELALLOC) { | |
4307 | /* | |
4308 | * For the COW fork we can reasonably get a | |
4309 | * request for converting an extent that races | |
4310 | * with other threads already having converted | |
4311 | * part of it, as there converting COW to | |
4312 | * regular blocks is not protected using the | |
4313 | * IOLOCK. | |
4314 | */ | |
4315 | ASSERT(flags & XFS_BMAPI_COWFORK); | |
4316 | if (!(flags & XFS_BMAPI_COWFORK)) { | |
4317 | error = -EIO; | |
4318 | goto error0; | |
4319 | } | |
4320 | ||
4321 | if (eof || bno >= end) | |
4322 | break; | |
4323 | } else { | |
4324 | need_alloc = true; | |
4325 | } | |
0b44aa85 CH |
4326 | } else if (isnullstartblock(bma.got.br_startblock)) { |
4327 | wasdelay = true; | |
7075a23f | 4328 | } |
34621a47 | 4329 | |
2bd0ea18 | 4330 | /* |
a2ceac1f DC |
4331 | * First, deal with the hole before the allocated space |
4332 | * that we found, if any. | |
2bd0ea18 | 4333 | */ |
7075a23f | 4334 | if (need_alloc || wasdelay) { |
a2ceac1f DC |
4335 | bma.eof = eof; |
4336 | bma.conv = !!(flags & XFS_BMAPI_CONVERT); | |
4337 | bma.wasdel = wasdelay; | |
4338 | bma.offset = bno; | |
4339 | bma.flags = flags; | |
4340 | ||
2bd0ea18 | 4341 | /* |
a2ceac1f DC |
4342 | * There's a 32/64 bit type mismatch between the |
4343 | * allocation length request (which can be 64 bits in | |
4344 | * length) and the bma length request, which is | |
4345 | * xfs_extlen_t and therefore 32 bits. Hence we have to | |
4346 | * check for 32-bit overflows and handle them here. | |
2bd0ea18 | 4347 | */ |
a2ceac1f DC |
4348 | if (len > (xfs_filblks_t)MAXEXTLEN) |
4349 | bma.length = MAXEXTLEN; | |
4350 | else | |
4351 | bma.length = len; | |
4352 | ||
4353 | ASSERT(len > 0); | |
4354 | ASSERT(bma.length > 0); | |
4355 | error = xfs_bmapi_allocate(&bma); | |
2bd0ea18 NS |
4356 | if (error) |
4357 | goto error0; | |
a2ceac1f DC |
4358 | if (bma.blkno == NULLFSBLOCK) |
4359 | break; | |
10e65503 DW |
4360 | |
4361 | /* | |
4362 | * If this is a CoW allocation, record the data in | |
4363 | * the refcount btree for orphan recovery. | |
4364 | */ | |
4365 | if (whichfork == XFS_COW_FORK) { | |
4366 | error = xfs_refcount_alloc_cow_extent(mp, dfops, | |
4367 | bma.blkno, bma.length); | |
4368 | if (error) | |
4369 | goto error0; | |
4370 | } | |
2bd0ea18 NS |
4371 | } |
4372 | ||
a2ceac1f DC |
4373 | /* Deal with the allocated space we found. */ |
4374 | xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno, | |
4375 | end, n, flags); | |
4376 | ||
4377 | /* Execute unwritten extent conversion if necessary */ | |
4378 | error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags); | |
12b53197 | 4379 | if (error == -EAGAIN) |
a2ceac1f DC |
4380 | continue; |
4381 | if (error) | |
4382 | goto error0; | |
4383 | ||
4384 | /* update the extent map to return */ | |
4385 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4386 | ||
2bd0ea18 NS |
4387 | /* |
4388 | * If we're done, stop now. Stop when we've allocated | |
4389 | * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise | |
4390 | * the transaction may get too big. | |
4391 | */ | |
a2ceac1f | 4392 | if (bno >= end || n >= *nmap || bma.nallocs >= *nmap) |
2bd0ea18 | 4393 | break; |
a2ceac1f DC |
4394 | |
4395 | /* Else go on to the next record. */ | |
4396 | bma.prev = bma.got; | |
82411945 CH |
4397 | if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got)) |
4398 | eof = true; | |
2bd0ea18 | 4399 | } |
2bd0ea18 | 4400 | *nmap = n; |
a2ceac1f | 4401 | |
2bd0ea18 NS |
4402 | /* |
4403 | * Transform from btree to extents, give it cur. | |
4404 | */ | |
a2ceac1f DC |
4405 | if (xfs_bmap_wants_extents(ip, whichfork)) { |
4406 | int tmp_logflags = 0; | |
4407 | ||
4408 | ASSERT(bma.cur); | |
4409 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, | |
4ca431fc | 4410 | &tmp_logflags, whichfork); |
a2ceac1f | 4411 | bma.logflags |= tmp_logflags; |
2bd0ea18 NS |
4412 | if (error) |
4413 | goto error0; | |
4414 | } | |
a2ceac1f | 4415 | |
2bd0ea18 | 4416 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE || |
a2ceac1f DC |
4417 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
4418 | XFS_IFORK_MAXEXT(ip, whichfork)); | |
2bd0ea18 | 4419 | error = 0; |
2bd0ea18 NS |
4420 | error0: |
4421 | /* | |
4422 | * Log everything. Do this after conversion, there's no point in | |
5e656dbb | 4423 | * logging the extent records if we've converted to btree format. |
2bd0ea18 | 4424 | */ |
a2ceac1f | 4425 | if ((bma.logflags & xfs_ilog_fext(whichfork)) && |
2bd0ea18 | 4426 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) |
a2ceac1f DC |
4427 | bma.logflags &= ~xfs_ilog_fext(whichfork); |
4428 | else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) && | |
2bd0ea18 | 4429 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) |
a2ceac1f | 4430 | bma.logflags &= ~xfs_ilog_fbroot(whichfork); |
2bd0ea18 NS |
4431 | /* |
4432 | * Log whatever the flags say, even if error. Otherwise we might miss | |
4433 | * detecting a case where the data is changed, there's an error, | |
4434 | * and it's not logged so we don't shutdown when we should. | |
4435 | */ | |
a2ceac1f DC |
4436 | if (bma.logflags) |
4437 | xfs_trans_log_inode(tp, ip, bma.logflags); | |
4438 | ||
4439 | if (bma.cur) { | |
2bd0ea18 NS |
4440 | if (!error) { |
4441 | ASSERT(*firstblock == NULLFSBLOCK || | |
6df6a3ab | 4442 | XFS_FSB_TO_AGNO(mp, *firstblock) <= |
eae766ca | 4443 | XFS_FSB_TO_AGNO(mp, |
6df6a3ab | 4444 | bma.cur->bc_private.b.firstblock)); |
a2ceac1f | 4445 | *firstblock = bma.cur->bc_private.b.firstblock; |
2bd0ea18 | 4446 | } |
a2ceac1f | 4447 | xfs_btree_del_cursor(bma.cur, |
2bd0ea18 NS |
4448 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
4449 | } | |
4450 | if (!error) | |
4451 | xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval, | |
4452 | orig_nmap, *nmap); | |
4453 | return error; | |
4454 | } | |
4455 | ||
0b44aa85 CH |
4456 | static int |
4457 | xfs_bmapi_remap( | |
4458 | struct xfs_trans *tp, | |
4459 | struct xfs_inode *ip, | |
4460 | xfs_fileoff_t bno, | |
4461 | xfs_filblks_t len, | |
4462 | xfs_fsblock_t startblock, | |
4463 | struct xfs_defer_ops *dfops) | |
4464 | { | |
4465 | struct xfs_mount *mp = ip->i_mount; | |
4466 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
4467 | struct xfs_btree_cur *cur = NULL; | |
4468 | xfs_fsblock_t firstblock = NULLFSBLOCK; | |
4469 | struct xfs_bmbt_irec got; | |
4470 | xfs_extnum_t idx; | |
4471 | int logflags = 0, error; | |
4472 | ||
4473 | ASSERT(len > 0); | |
4474 | ASSERT(len <= (xfs_filblks_t)MAXEXTLEN); | |
4475 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
4476 | ||
4477 | if (unlikely(XFS_TEST_ERROR( | |
4478 | (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS && | |
4479 | XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 4480 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
0b44aa85 CH |
4481 | XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp); |
4482 | return -EFSCORRUPTED; | |
4483 | } | |
4484 | ||
4485 | if (XFS_FORCED_SHUTDOWN(mp)) | |
4486 | return -EIO; | |
4487 | ||
4488 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
4489 | error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK); | |
4490 | if (error) | |
4491 | return error; | |
4492 | } | |
4493 | ||
4494 | if (xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) { | |
4495 | /* make sure we only reflink into a hole. */ | |
4496 | ASSERT(got.br_startoff > bno); | |
4497 | ASSERT(got.br_startoff - bno >= len); | |
4498 | } | |
4499 | ||
05422db6 CH |
4500 | ip->i_d.di_nblocks += len; |
4501 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
0b44aa85 CH |
4502 | |
4503 | if (ifp->if_flags & XFS_IFBROOT) { | |
4504 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
4505 | cur->bc_private.b.firstblock = firstblock; | |
4506 | cur->bc_private.b.dfops = dfops; | |
4507 | cur->bc_private.b.flags = 0; | |
4508 | } | |
4509 | ||
4510 | got.br_startoff = bno; | |
4511 | got.br_startblock = startblock; | |
4512 | got.br_blockcount = len; | |
4513 | got.br_state = XFS_EXT_NORM; | |
4514 | ||
4515 | error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &idx, &cur, | |
4516 | &got, &firstblock, dfops, &logflags); | |
4517 | if (error) | |
4518 | goto error0; | |
4519 | ||
4520 | if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) { | |
4521 | int tmp_logflags = 0; | |
4522 | ||
4523 | error = xfs_bmap_btree_to_extents(tp, ip, cur, | |
4524 | &tmp_logflags, XFS_DATA_FORK); | |
4525 | logflags |= tmp_logflags; | |
4526 | } | |
4527 | ||
4528 | error0: | |
4529 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) | |
4530 | logflags &= ~XFS_ILOG_DEXT; | |
4531 | else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) | |
4532 | logflags &= ~XFS_ILOG_DBROOT; | |
4533 | ||
4534 | if (logflags) | |
4535 | xfs_trans_log_inode(tp, ip, logflags); | |
4536 | if (cur) { | |
4537 | xfs_btree_del_cursor(cur, | |
4538 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
4539 | } | |
4540 | return error; | |
4541 | } | |
4542 | ||
01d1b786 BF |
4543 | /* |
4544 | * When a delalloc extent is split (e.g., due to a hole punch), the original | |
4545 | * indlen reservation must be shared across the two new extents that are left | |
4546 | * behind. | |
4547 | * | |
4548 | * Given the original reservation and the worst case indlen for the two new | |
4549 | * extents (as calculated by xfs_bmap_worst_indlen()), split the original | |
731ccdf9 BF |
4550 | * reservation fairly across the two new extents. If necessary, steal available |
4551 | * blocks from a deleted extent to make up a reservation deficiency (e.g., if | |
4552 | * ores == 1). The number of stolen blocks is returned. The availability and | |
4553 | * subsequent accounting of stolen blocks is the responsibility of the caller. | |
01d1b786 | 4554 | */ |
731ccdf9 | 4555 | static xfs_filblks_t |
01d1b786 BF |
4556 | xfs_bmap_split_indlen( |
4557 | xfs_filblks_t ores, /* original res. */ | |
4558 | xfs_filblks_t *indlen1, /* ext1 worst indlen */ | |
731ccdf9 BF |
4559 | xfs_filblks_t *indlen2, /* ext2 worst indlen */ |
4560 | xfs_filblks_t avail) /* stealable blocks */ | |
01d1b786 BF |
4561 | { |
4562 | xfs_filblks_t len1 = *indlen1; | |
4563 | xfs_filblks_t len2 = *indlen2; | |
4564 | xfs_filblks_t nres = len1 + len2; /* new total res. */ | |
731ccdf9 | 4565 | xfs_filblks_t stolen = 0; |
775762e4 | 4566 | xfs_filblks_t resfactor; |
731ccdf9 BF |
4567 | |
4568 | /* | |
4569 | * Steal as many blocks as we can to try and satisfy the worst case | |
4570 | * indlen for both new extents. | |
4571 | */ | |
775762e4 BF |
4572 | if (ores < nres && avail) |
4573 | stolen = XFS_FILBLKS_MIN(nres - ores, avail); | |
4574 | ores += stolen; | |
4575 | ||
4576 | /* nothing else to do if we've satisfied the new reservation */ | |
4577 | if (ores >= nres) | |
4578 | return stolen; | |
4579 | ||
4580 | /* | |
4581 | * We can't meet the total required reservation for the two extents. | |
4582 | * Calculate the percent of the overall shortage between both extents | |
4583 | * and apply this percentage to each of the requested indlen values. | |
4584 | * This distributes the shortage fairly and reduces the chances that one | |
4585 | * of the two extents is left with nothing when extents are repeatedly | |
4586 | * split. | |
4587 | */ | |
4588 | resfactor = (ores * 100); | |
4589 | do_div(resfactor, nres); | |
4590 | len1 *= resfactor; | |
4591 | do_div(len1, 100); | |
4592 | len2 *= resfactor; | |
4593 | do_div(len2, 100); | |
4594 | ASSERT(len1 + len2 <= ores); | |
4595 | ASSERT(len1 < *indlen1 && len2 < *indlen2); | |
01d1b786 BF |
4596 | |
4597 | /* | |
775762e4 BF |
4598 | * Hand out the remainder to each extent. If one of the two reservations |
4599 | * is zero, we want to make sure that one gets a block first. The loop | |
4600 | * below starts with len1, so hand len2 a block right off the bat if it | |
4601 | * is zero. | |
01d1b786 | 4602 | */ |
775762e4 BF |
4603 | ores -= (len1 + len2); |
4604 | ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores); | |
4605 | if (ores && !len2 && *indlen2) { | |
4606 | len2++; | |
4607 | ores--; | |
4608 | } | |
4609 | while (ores) { | |
4610 | if (len1 < *indlen1) { | |
4611 | len1++; | |
4612 | ores--; | |
01d1b786 | 4613 | } |
775762e4 | 4614 | if (!ores) |
01d1b786 | 4615 | break; |
775762e4 BF |
4616 | if (len2 < *indlen2) { |
4617 | len2++; | |
4618 | ores--; | |
01d1b786 BF |
4619 | } |
4620 | } | |
4621 | ||
4622 | *indlen1 = len1; | |
4623 | *indlen2 = len2; | |
731ccdf9 BF |
4624 | |
4625 | return stolen; | |
01d1b786 BF |
4626 | } |
4627 | ||
ece930fa CH |
4628 | int |
4629 | xfs_bmap_del_extent_delay( | |
4630 | struct xfs_inode *ip, | |
4631 | int whichfork, | |
4632 | xfs_extnum_t *idx, | |
4633 | struct xfs_bmbt_irec *got, | |
4634 | struct xfs_bmbt_irec *del) | |
4635 | { | |
4636 | struct xfs_mount *mp = ip->i_mount; | |
4637 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
4638 | struct xfs_bmbt_irec new; | |
4639 | int64_t da_old, da_new, da_diff = 0; | |
4640 | xfs_fileoff_t del_endoff, got_endoff; | |
4641 | xfs_filblks_t got_indlen, new_indlen, stolen; | |
7bd43334 CH |
4642 | int state = xfs_bmap_fork_to_state(whichfork); |
4643 | int error = 0; | |
ece930fa CH |
4644 | bool isrt; |
4645 | ||
4646 | XFS_STATS_INC(mp, xs_del_exlist); | |
4647 | ||
4648 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); | |
4649 | del_endoff = del->br_startoff + del->br_blockcount; | |
4650 | got_endoff = got->br_startoff + got->br_blockcount; | |
4651 | da_old = startblockval(got->br_startblock); | |
4652 | da_new = 0; | |
4653 | ||
4654 | ASSERT(*idx >= 0); | |
d09d4e5f | 4655 | ASSERT(*idx <= xfs_iext_count(ifp)); |
ece930fa CH |
4656 | ASSERT(del->br_blockcount > 0); |
4657 | ASSERT(got->br_startoff <= del->br_startoff); | |
4658 | ASSERT(got_endoff >= del_endoff); | |
4659 | ||
4660 | if (isrt) { | |
f1e0bd8d | 4661 | uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount); |
ece930fa CH |
4662 | |
4663 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
4664 | xfs_mod_frextents(mp, rtexts); | |
4665 | } | |
4666 | ||
4667 | /* | |
4668 | * Update the inode delalloc counter now and wait to update the | |
4669 | * sb counters as we might have to borrow some blocks for the | |
4670 | * indirect block accounting. | |
4671 | */ | |
4672 | error = xfs_trans_reserve_quota_nblks(NULL, ip, | |
4673 | -((long)del->br_blockcount), 0, | |
4674 | isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
6921638c DW |
4675 | if (error) |
4676 | return error; | |
ece930fa CH |
4677 | ip->i_delayed_blks -= del->br_blockcount; |
4678 | ||
ece930fa | 4679 | if (got->br_startoff == del->br_startoff) |
d0a03e5a | 4680 | state |= BMAP_LEFT_FILLING; |
ece930fa | 4681 | if (got_endoff == del_endoff) |
d0a03e5a | 4682 | state |= BMAP_RIGHT_FILLING; |
ece930fa | 4683 | |
d0a03e5a CH |
4684 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4685 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
ece930fa CH |
4686 | /* |
4687 | * Matches the whole extent. Delete the entry. | |
4688 | */ | |
4689 | xfs_iext_remove(ip, *idx, 1, state); | |
4690 | --*idx; | |
4691 | break; | |
d0a03e5a | 4692 | case BMAP_LEFT_FILLING: |
ece930fa CH |
4693 | /* |
4694 | * Deleting the first part of the extent. | |
4695 | */ | |
ece930fa CH |
4696 | got->br_startoff = del_endoff; |
4697 | got->br_blockcount -= del->br_blockcount; | |
4698 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4699 | got->br_blockcount), da_old); | |
4700 | got->br_startblock = nullstartblock((int)da_new); | |
df926c07 | 4701 | xfs_iext_update_extent(ip, state, *idx, got); |
ece930fa | 4702 | break; |
d0a03e5a | 4703 | case BMAP_RIGHT_FILLING: |
ece930fa CH |
4704 | /* |
4705 | * Deleting the last part of the extent. | |
4706 | */ | |
ece930fa CH |
4707 | got->br_blockcount = got->br_blockcount - del->br_blockcount; |
4708 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4709 | got->br_blockcount), da_old); | |
4710 | got->br_startblock = nullstartblock((int)da_new); | |
df926c07 | 4711 | xfs_iext_update_extent(ip, state, *idx, got); |
ece930fa CH |
4712 | break; |
4713 | case 0: | |
4714 | /* | |
4715 | * Deleting the middle of the extent. | |
4716 | * | |
4717 | * Distribute the original indlen reservation across the two new | |
4718 | * extents. Steal blocks from the deleted extent if necessary. | |
4719 | * Stealing blocks simply fudges the fdblocks accounting below. | |
4720 | * Warn if either of the new indlen reservations is zero as this | |
4721 | * can lead to delalloc problems. | |
4722 | */ | |
ece930fa CH |
4723 | got->br_blockcount = del->br_startoff - got->br_startoff; |
4724 | got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount); | |
4725 | ||
4726 | new.br_blockcount = got_endoff - del_endoff; | |
4727 | new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
4728 | ||
4729 | WARN_ON_ONCE(!got_indlen || !new_indlen); | |
4730 | stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen, | |
4731 | del->br_blockcount); | |
4732 | ||
4733 | got->br_startblock = nullstartblock((int)got_indlen); | |
df926c07 | 4734 | xfs_iext_update_extent(ip, state, *idx, got); |
ece930fa CH |
4735 | |
4736 | new.br_startoff = del_endoff; | |
4737 | new.br_state = got->br_state; | |
4738 | new.br_startblock = nullstartblock((int)new_indlen); | |
4739 | ||
4740 | ++*idx; | |
4741 | xfs_iext_insert(ip, *idx, 1, &new, state); | |
4742 | ||
4743 | da_new = got_indlen + new_indlen - stolen; | |
4744 | del->br_blockcount -= stolen; | |
4745 | break; | |
4746 | } | |
4747 | ||
4748 | ASSERT(da_old >= da_new); | |
4749 | da_diff = da_old - da_new; | |
4750 | if (!isrt) | |
4751 | da_diff += del->br_blockcount; | |
4752 | if (da_diff) | |
4753 | xfs_mod_fdblocks(mp, da_diff, false); | |
4754 | return error; | |
4755 | } | |
4756 | ||
4757 | void | |
4758 | xfs_bmap_del_extent_cow( | |
4759 | struct xfs_inode *ip, | |
4760 | xfs_extnum_t *idx, | |
4761 | struct xfs_bmbt_irec *got, | |
4762 | struct xfs_bmbt_irec *del) | |
4763 | { | |
4764 | struct xfs_mount *mp = ip->i_mount; | |
4765 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
4766 | struct xfs_bmbt_irec new; | |
4767 | xfs_fileoff_t del_endoff, got_endoff; | |
4768 | int state = BMAP_COWFORK; | |
4769 | ||
4770 | XFS_STATS_INC(mp, xs_del_exlist); | |
4771 | ||
4772 | del_endoff = del->br_startoff + del->br_blockcount; | |
4773 | got_endoff = got->br_startoff + got->br_blockcount; | |
4774 | ||
4775 | ASSERT(*idx >= 0); | |
d09d4e5f | 4776 | ASSERT(*idx <= xfs_iext_count(ifp)); |
ece930fa CH |
4777 | ASSERT(del->br_blockcount > 0); |
4778 | ASSERT(got->br_startoff <= del->br_startoff); | |
4779 | ASSERT(got_endoff >= del_endoff); | |
4780 | ASSERT(!isnullstartblock(got->br_startblock)); | |
4781 | ||
4782 | if (got->br_startoff == del->br_startoff) | |
d0a03e5a | 4783 | state |= BMAP_LEFT_FILLING; |
ece930fa | 4784 | if (got_endoff == del_endoff) |
d0a03e5a | 4785 | state |= BMAP_RIGHT_FILLING; |
ece930fa | 4786 | |
d0a03e5a CH |
4787 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4788 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
ece930fa CH |
4789 | /* |
4790 | * Matches the whole extent. Delete the entry. | |
4791 | */ | |
4792 | xfs_iext_remove(ip, *idx, 1, state); | |
4793 | --*idx; | |
4794 | break; | |
d0a03e5a | 4795 | case BMAP_LEFT_FILLING: |
ece930fa CH |
4796 | /* |
4797 | * Deleting the first part of the extent. | |
4798 | */ | |
ece930fa CH |
4799 | got->br_startoff = del_endoff; |
4800 | got->br_blockcount -= del->br_blockcount; | |
4801 | got->br_startblock = del->br_startblock + del->br_blockcount; | |
df926c07 | 4802 | xfs_iext_update_extent(ip, state, *idx, got); |
ece930fa | 4803 | break; |
d0a03e5a | 4804 | case BMAP_RIGHT_FILLING: |
ece930fa CH |
4805 | /* |
4806 | * Deleting the last part of the extent. | |
4807 | */ | |
ece930fa | 4808 | got->br_blockcount -= del->br_blockcount; |
df926c07 | 4809 | xfs_iext_update_extent(ip, state, *idx, got); |
ece930fa CH |
4810 | break; |
4811 | case 0: | |
4812 | /* | |
4813 | * Deleting the middle of the extent. | |
4814 | */ | |
ece930fa | 4815 | got->br_blockcount = del->br_startoff - got->br_startoff; |
df926c07 | 4816 | xfs_iext_update_extent(ip, state, *idx, got); |
ece930fa CH |
4817 | |
4818 | new.br_startoff = del_endoff; | |
4819 | new.br_blockcount = got_endoff - del_endoff; | |
4820 | new.br_state = got->br_state; | |
4821 | new.br_startblock = del->br_startblock + del->br_blockcount; | |
4822 | ||
4823 | ++*idx; | |
4824 | xfs_iext_insert(ip, *idx, 1, &new, state); | |
4825 | break; | |
4826 | } | |
4827 | } | |
4828 | ||
2bd0ea18 | 4829 | /* |
49f693fa | 4830 | * Called by xfs_bmapi to update file extent records and the btree |
ad68fd19 | 4831 | * after removing space. |
2bd0ea18 | 4832 | */ |
49f693fa | 4833 | STATIC int /* error */ |
ad68fd19 | 4834 | xfs_bmap_del_extent_real( |
49f693fa DC |
4835 | xfs_inode_t *ip, /* incore inode pointer */ |
4836 | xfs_trans_t *tp, /* current transaction pointer */ | |
4837 | xfs_extnum_t *idx, /* extent number to update/delete */ | |
f33cea1a | 4838 | struct xfs_defer_ops *dfops, /* list of extents to be freed */ |
49f693fa DC |
4839 | xfs_btree_cur_t *cur, /* if null, not a btree */ |
4840 | xfs_bmbt_irec_t *del, /* data to remove from extents */ | |
4841 | int *logflagsp, /* inode logging flags */ | |
36b16da8 DW |
4842 | int whichfork, /* data or attr fork */ |
4843 | int bflags) /* bmapi flags */ | |
2bd0ea18 | 4844 | { |
49f693fa DC |
4845 | xfs_fsblock_t del_endblock=0; /* first block past del */ |
4846 | xfs_fileoff_t del_endoff; /* first offset past del */ | |
49f693fa | 4847 | int do_fx; /* free extent at end of routine */ |
49f693fa | 4848 | int error; /* error return value */ |
bd92a38b | 4849 | int flags = 0;/* inode logging flags */ |
70bf7533 | 4850 | struct xfs_bmbt_irec got; /* current extent entry */ |
49f693fa DC |
4851 | xfs_fileoff_t got_endoff; /* first offset past got */ |
4852 | int i; /* temp state */ | |
4853 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
4854 | xfs_mount_t *mp; /* mount structure */ | |
4855 | xfs_filblks_t nblks; /* quota/sb block count */ | |
4856 | xfs_bmbt_irec_t new; /* new record to be inserted */ | |
4857 | /* REFERENCED */ | |
4858 | uint qfield; /* quota field to update */ | |
7bd43334 | 4859 | int state = xfs_bmap_fork_to_state(whichfork); |
70bf7533 | 4860 | struct xfs_bmbt_irec old; |
a2ceac1f | 4861 | |
79896434 BD |
4862 | mp = ip->i_mount; |
4863 | XFS_STATS_INC(mp, xs_del_exlist); | |
a2ceac1f | 4864 | |
49f693fa | 4865 | ifp = XFS_IFORK_PTR(ip, whichfork); |
d09d4e5f | 4866 | ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp))); |
49f693fa | 4867 | ASSERT(del->br_blockcount > 0); |
70bf7533 | 4868 | xfs_iext_get_extent(ifp, *idx, &got); |
49f693fa DC |
4869 | ASSERT(got.br_startoff <= del->br_startoff); |
4870 | del_endoff = del->br_startoff + del->br_blockcount; | |
4871 | got_endoff = got.br_startoff + got.br_blockcount; | |
4872 | ASSERT(got_endoff >= del_endoff); | |
ad68fd19 | 4873 | ASSERT(!isnullstartblock(got.br_startblock)); |
49f693fa DC |
4874 | qfield = 0; |
4875 | error = 0; | |
ad68fd19 | 4876 | |
bd92a38b CH |
4877 | /* |
4878 | * If it's the case where the directory code is running with no block | |
4879 | * reservation, and the deleted block is in the middle of its extent, | |
4880 | * and the resulting insert of an extent would cause transformation to | |
4881 | * btree format, then reject it. The calling code will then swap blocks | |
4882 | * around instead. We have to do this now, rather than waiting for the | |
4883 | * conversion to btree format, since the transaction will be dirty then. | |
4884 | */ | |
4885 | if (tp->t_blk_res == 0 && | |
4886 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
4887 | XFS_IFORK_NEXTENTS(ip, whichfork) >= | |
4888 | XFS_IFORK_MAXEXT(ip, whichfork) && | |
4889 | del->br_startoff > got.br_startoff && del_endoff < got_endoff) | |
4890 | return -ENOSPC; | |
4891 | ||
4892 | flags = XFS_ILOG_CORE; | |
ad68fd19 CH |
4893 | if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) { |
4894 | xfs_fsblock_t bno; | |
4895 | xfs_filblks_t len; | |
4896 | ||
4897 | ASSERT(do_mod(del->br_blockcount, mp->m_sb.sb_rextsize) == 0); | |
4898 | ASSERT(do_mod(del->br_startblock, mp->m_sb.sb_rextsize) == 0); | |
4899 | bno = del->br_startblock; | |
4900 | len = del->br_blockcount; | |
4901 | do_div(bno, mp->m_sb.sb_rextsize); | |
4902 | do_div(len, mp->m_sb.sb_rextsize); | |
4903 | error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len); | |
4904 | if (error) | |
4905 | goto done; | |
49f693fa | 4906 | do_fx = 0; |
ad68fd19 CH |
4907 | nblks = len * mp->m_sb.sb_rextsize; |
4908 | qfield = XFS_TRANS_DQ_RTBCOUNT; | |
4909 | } else { | |
4910 | do_fx = 1; | |
4911 | nblks = del->br_blockcount; | |
4912 | qfield = XFS_TRANS_DQ_BCOUNT; | |
4913 | } | |
4914 | ||
4915 | del_endblock = del->br_startblock + del->br_blockcount; | |
4916 | if (cur) { | |
70a93110 | 4917 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
ad68fd19 CH |
4918 | if (error) |
4919 | goto done; | |
4920 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); | |
49f693fa | 4921 | } |
85aec44f | 4922 | |
b039ac79 CH |
4923 | if (got.br_startoff == del->br_startoff) |
4924 | state |= BMAP_LEFT_FILLING; | |
4925 | if (got_endoff == del_endoff) | |
4926 | state |= BMAP_RIGHT_FILLING; | |
4927 | ||
4928 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { | |
4929 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
49f693fa DC |
4930 | /* |
4931 | * Matches the whole extent. Delete the entry. | |
4932 | */ | |
1a32da85 | 4933 | xfs_iext_remove(ip, *idx, 1, state); |
49f693fa | 4934 | --*idx; |
49f693fa DC |
4935 | |
4936 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
4937 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
4938 | flags |= XFS_ILOG_CORE; | |
4939 | if (!cur) { | |
4940 | flags |= xfs_ilog_fext(whichfork); | |
4941 | break; | |
2bd0ea18 | 4942 | } |
49f693fa DC |
4943 | if ((error = xfs_btree_delete(cur, &i))) |
4944 | goto done; | |
19ebedcf | 4945 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 4946 | break; |
b039ac79 | 4947 | case BMAP_LEFT_FILLING: |
2bd0ea18 | 4948 | /* |
49f693fa | 4949 | * Deleting the first part of the extent. |
2bd0ea18 | 4950 | */ |
70bf7533 CH |
4951 | got.br_startoff = del_endoff; |
4952 | got.br_startblock = del_endblock; | |
4953 | got.br_blockcount -= del->br_blockcount; | |
df926c07 | 4954 | xfs_iext_update_extent(ip, state, *idx, &got); |
49f693fa DC |
4955 | if (!cur) { |
4956 | flags |= xfs_ilog_fext(whichfork); | |
4957 | break; | |
4958 | } | |
d0e5f1ff | 4959 | error = xfs_bmbt_update(cur, &got); |
70bf7533 | 4960 | if (error) |
49f693fa DC |
4961 | goto done; |
4962 | break; | |
b039ac79 | 4963 | case BMAP_RIGHT_FILLING: |
2bd0ea18 | 4964 | /* |
49f693fa | 4965 | * Deleting the last part of the extent. |
2bd0ea18 | 4966 | */ |
70bf7533 | 4967 | got.br_blockcount -= del->br_blockcount; |
df926c07 | 4968 | xfs_iext_update_extent(ip, state, *idx, &got); |
49f693fa DC |
4969 | if (!cur) { |
4970 | flags |= xfs_ilog_fext(whichfork); | |
4971 | break; | |
4972 | } | |
d0e5f1ff | 4973 | error = xfs_bmbt_update(cur, &got); |
70bf7533 | 4974 | if (error) |
49f693fa DC |
4975 | goto done; |
4976 | break; | |
49f693fa DC |
4977 | case 0: |
4978 | /* | |
4979 | * Deleting the middle of the extent. | |
4980 | */ | |
70bf7533 | 4981 | old = got; |
df926c07 | 4982 | |
70bf7533 | 4983 | got.br_blockcount = del->br_startoff - got.br_startoff; |
df926c07 | 4984 | xfs_iext_update_extent(ip, state, *idx, &got); |
70bf7533 | 4985 | |
49f693fa | 4986 | new.br_startoff = del_endoff; |
70bf7533 | 4987 | new.br_blockcount = got_endoff - del_endoff; |
49f693fa | 4988 | new.br_state = got.br_state; |
ad68fd19 | 4989 | new.br_startblock = del_endblock; |
70bf7533 | 4990 | |
ad68fd19 CH |
4991 | flags |= XFS_ILOG_CORE; |
4992 | if (cur) { | |
d0e5f1ff | 4993 | error = xfs_bmbt_update(cur, &got); |
ad68fd19 CH |
4994 | if (error) |
4995 | goto done; | |
4996 | error = xfs_btree_increment(cur, 0, &i); | |
4997 | if (error) | |
4998 | goto done; | |
4999 | cur->bc_rec.b = new; | |
5000 | error = xfs_btree_insert(cur, &i); | |
5001 | if (error && error != -ENOSPC) | |
5002 | goto done; | |
5003 | /* | |
5004 | * If get no-space back from btree insert, it tried a | |
5005 | * split, and we have a zero block reservation. Fix up | |
5006 | * our state and return the error. | |
5007 | */ | |
5008 | if (error == -ENOSPC) { | |
49f693fa | 5009 | /* |
ad68fd19 CH |
5010 | * Reset the cursor, don't trust it after any |
5011 | * insert operation. | |
49f693fa | 5012 | */ |
70a93110 | 5013 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
ad68fd19 | 5014 | if (error) |
49f693fa | 5015 | goto done; |
19ebedcf | 5016 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
ad68fd19 CH |
5017 | /* |
5018 | * Update the btree record back | |
5019 | * to the original value. | |
5020 | */ | |
d0e5f1ff | 5021 | error = xfs_bmbt_update(cur, &old); |
ad68fd19 CH |
5022 | if (error) |
5023 | goto done; | |
5024 | /* | |
5025 | * Reset the extent record back | |
5026 | * to the original value. | |
5027 | */ | |
df926c07 | 5028 | xfs_iext_update_extent(ip, state, *idx, &old); |
ad68fd19 CH |
5029 | flags = 0; |
5030 | error = -ENOSPC; | |
5031 | goto done; | |
5032 | } | |
5033 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); | |
5034 | } else | |
5035 | flags |= xfs_ilog_fext(whichfork); | |
5036 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5037 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
49f693fa DC |
5038 | xfs_iext_insert(ip, *idx + 1, 1, &new, state); |
5039 | ++*idx; | |
5040 | break; | |
2bd0ea18 | 5041 | } |
d7f80320 DW |
5042 | |
5043 | /* remove reverse mapping */ | |
ad68fd19 CH |
5044 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del); |
5045 | if (error) | |
5046 | goto done; | |
d7f80320 | 5047 | |
2bd0ea18 | 5048 | /* |
49f693fa | 5049 | * If we need to, add to list of extents to delete. |
2bd0ea18 | 5050 | */ |
36b16da8 | 5051 | if (do_fx && !(bflags & XFS_BMAPI_REMAP)) { |
cfe32f0d DW |
5052 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) { |
5053 | error = xfs_refcount_decrease_extent(mp, dfops, del); | |
5054 | if (error) | |
5055 | goto done; | |
5056 | } else | |
5057 | xfs_bmap_add_free(mp, dfops, del->br_startblock, | |
5058 | del->br_blockcount, NULL); | |
5059 | } | |
5060 | ||
2bd0ea18 | 5061 | /* |
49f693fa | 5062 | * Adjust inode # blocks in the file. |
2bd0ea18 | 5063 | */ |
49f693fa DC |
5064 | if (nblks) |
5065 | ip->i_d.di_nblocks -= nblks; | |
2bd0ea18 | 5066 | /* |
49f693fa | 5067 | * Adjust quota data. |
2bd0ea18 | 5068 | */ |
36b16da8 | 5069 | if (qfield && !(bflags & XFS_BMAPI_REMAP)) |
49f693fa DC |
5070 | xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks); |
5071 | ||
49f693fa DC |
5072 | done: |
5073 | *logflagsp = flags; | |
2bd0ea18 NS |
5074 | return error; |
5075 | } | |
5076 | ||
5077 | /* | |
49f693fa DC |
5078 | * Unmap (remove) blocks from a file. |
5079 | * If nexts is nonzero then the number of extents to remove is limited to | |
5080 | * that value. If not all extents in the block range can be removed then | |
5081 | * *done is set. | |
2bd0ea18 | 5082 | */ |
49f693fa | 5083 | int /* error */ |
3d36acda | 5084 | __xfs_bunmapi( |
49f693fa DC |
5085 | xfs_trans_t *tp, /* transaction pointer */ |
5086 | struct xfs_inode *ip, /* incore inode */ | |
675b5a20 | 5087 | xfs_fileoff_t start, /* first file offset deleted */ |
3d36acda | 5088 | xfs_filblks_t *rlen, /* i/o: amount remaining */ |
49f693fa DC |
5089 | int flags, /* misc flags */ |
5090 | xfs_extnum_t nexts, /* number of extents max */ | |
5091 | xfs_fsblock_t *firstblock, /* first allocated block | |
5092 | controls a.g. for allocs */ | |
3d36acda | 5093 | struct xfs_defer_ops *dfops) /* i/o: deferred updates */ |
2bd0ea18 | 5094 | { |
49f693fa DC |
5095 | xfs_btree_cur_t *cur; /* bmap btree cursor */ |
5096 | xfs_bmbt_irec_t del; /* extent being deleted */ | |
49f693fa DC |
5097 | int error; /* error return value */ |
5098 | xfs_extnum_t extno; /* extent number in list */ | |
5099 | xfs_bmbt_irec_t got; /* current extent record */ | |
5100 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
5101 | int isrt; /* freeing in rt area */ | |
5102 | xfs_extnum_t lastx; /* last extent index used */ | |
5103 | int logflags; /* transaction logging flags */ | |
5104 | xfs_extlen_t mod; /* rt extent offset */ | |
5105 | xfs_mount_t *mp; /* mount structure */ | |
49f693fa DC |
5106 | int tmp_logflags; /* partial logging flags */ |
5107 | int wasdel; /* was a delayed alloc extent */ | |
5108 | int whichfork; /* data or attribute fork */ | |
5109 | xfs_fsblock_t sum; | |
3d36acda | 5110 | xfs_filblks_t len = *rlen; /* length to unmap in file */ |
594956fa | 5111 | xfs_fileoff_t max_len; |
15a8bccc | 5112 | xfs_agnumber_t prev_agno = NULLAGNUMBER, agno; |
675b5a20 | 5113 | xfs_fileoff_t end; |
2bd0ea18 | 5114 | |
675b5a20 | 5115 | trace_xfs_bunmap(ip, start, len, flags, _RET_IP_); |
a2ceac1f | 5116 | |
cb8a004a DW |
5117 | whichfork = xfs_bmapi_whichfork(flags); |
5118 | ASSERT(whichfork != XFS_COW_FORK); | |
a2ceac1f | 5119 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa DC |
5120 | if (unlikely( |
5121 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5122 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
5123 | XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW, | |
5124 | ip->i_mount); | |
12b53197 | 5125 | return -EFSCORRUPTED; |
49f693fa DC |
5126 | } |
5127 | mp = ip->i_mount; | |
5128 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 5129 | return -EIO; |
56b2de80 | 5130 | |
ff105f75 | 5131 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
49f693fa DC |
5132 | ASSERT(len > 0); |
5133 | ASSERT(nexts >= 0); | |
56b2de80 | 5134 | |
594956fa DW |
5135 | /* |
5136 | * Guesstimate how many blocks we can unmap without running the risk of | |
5137 | * blowing out the transaction with a mix of EFIs and reflink | |
5138 | * adjustments. | |
5139 | */ | |
5140 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) | |
5141 | max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res)); | |
5142 | else | |
5143 | max_len = len; | |
5144 | ||
49f693fa DC |
5145 | if (!(ifp->if_flags & XFS_IFEXTENTS) && |
5146 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
5147 | return error; | |
d09d4e5f | 5148 | if (xfs_iext_count(ifp) == 0) { |
3d36acda | 5149 | *rlen = 0; |
49f693fa | 5150 | return 0; |
56b2de80 | 5151 | } |
79896434 | 5152 | XFS_STATS_INC(mp, xs_blk_unmap); |
49f693fa | 5153 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); |
675b5a20 | 5154 | end = start + len - 1; |
a2ceac1f DC |
5155 | |
5156 | /* | |
49f693fa DC |
5157 | * Check to see if the given block number is past the end of the |
5158 | * file, back up to the last block if so... | |
56b2de80 | 5159 | */ |
675b5a20 | 5160 | if (!xfs_iext_lookup_extent(ip, ifp, end, &lastx, &got)) { |
246eb90a CH |
5161 | ASSERT(lastx > 0); |
5162 | xfs_iext_get_extent(ifp, --lastx, &got); | |
675b5a20 | 5163 | end = got.br_startoff + got.br_blockcount - 1; |
49f693fa | 5164 | } |
246eb90a | 5165 | |
49f693fa DC |
5166 | logflags = 0; |
5167 | if (ifp->if_flags & XFS_IFBROOT) { | |
5168 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
5169 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5170 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 5171 | cur->bc_private.b.dfops = dfops; |
49f693fa DC |
5172 | cur->bc_private.b.flags = 0; |
5173 | } else | |
5174 | cur = NULL; | |
a2ceac1f | 5175 | |
49f693fa | 5176 | if (isrt) { |
a2ceac1f | 5177 | /* |
49f693fa | 5178 | * Synchronize by locking the bitmap inode. |
a2ceac1f | 5179 | */ |
a62ed6d3 | 5180 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
49f693fa | 5181 | xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
a62ed6d3 DW |
5182 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
5183 | xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL); | |
49f693fa | 5184 | } |
a2ceac1f | 5185 | |
49f693fa | 5186 | extno = 0; |
675b5a20 | 5187 | while (end != (xfs_fileoff_t)-1 && end >= start && lastx >= 0 && |
594956fa | 5188 | (nexts == 0 || extno < nexts) && max_len > 0) { |
a2ceac1f | 5189 | /* |
675b5a20 | 5190 | * Is the found extent after a hole in which end lives? |
49f693fa | 5191 | * Just back up to the previous extent, if so. |
a2ceac1f | 5192 | */ |
675b5a20 | 5193 | if (got.br_startoff > end) { |
49f693fa DC |
5194 | if (--lastx < 0) |
5195 | break; | |
246eb90a | 5196 | xfs_iext_get_extent(ifp, lastx, &got); |
a2ceac1f | 5197 | } |
49f693fa DC |
5198 | /* |
5199 | * Is the last block of this extent before the range | |
5200 | * we're supposed to delete? If so, we're done. | |
5201 | */ | |
675b5a20 | 5202 | end = XFS_FILEOFF_MIN(end, |
49f693fa | 5203 | got.br_startoff + got.br_blockcount - 1); |
675b5a20 | 5204 | if (end < start) |
49f693fa DC |
5205 | break; |
5206 | /* | |
5207 | * Then deal with the (possibly delayed) allocated space | |
5208 | * we found. | |
5209 | */ | |
49f693fa DC |
5210 | del = got; |
5211 | wasdel = isnullstartblock(del.br_startblock); | |
15a8bccc CH |
5212 | |
5213 | /* | |
5214 | * Make sure we don't touch multiple AGF headers out of order | |
5215 | * in a single transaction, as that could cause AB-BA deadlocks. | |
5216 | */ | |
5217 | if (!wasdel) { | |
5218 | agno = XFS_FSB_TO_AGNO(mp, del.br_startblock); | |
5219 | if (prev_agno != NULLAGNUMBER && prev_agno > agno) | |
5220 | break; | |
5221 | prev_agno = agno; | |
5222 | } | |
49f693fa DC |
5223 | if (got.br_startoff < start) { |
5224 | del.br_startoff = start; | |
5225 | del.br_blockcount -= start - got.br_startoff; | |
5226 | if (!wasdel) | |
5227 | del.br_startblock += start - got.br_startoff; | |
5228 | } | |
675b5a20 CH |
5229 | if (del.br_startoff + del.br_blockcount > end + 1) |
5230 | del.br_blockcount = end + 1 - del.br_startoff; | |
594956fa DW |
5231 | |
5232 | /* How much can we safely unmap? */ | |
5233 | if (max_len < del.br_blockcount) { | |
5234 | del.br_startoff += del.br_blockcount - max_len; | |
5235 | if (!wasdel) | |
5236 | del.br_startblock += del.br_blockcount - max_len; | |
5237 | del.br_blockcount = max_len; | |
5238 | } | |
5239 | ||
49f693fa DC |
5240 | sum = del.br_startblock + del.br_blockcount; |
5241 | if (isrt && | |
5242 | (mod = do_mod(sum, mp->m_sb.sb_rextsize))) { | |
5243 | /* | |
5244 | * Realtime extent not lined up at the end. | |
5245 | * The extent could have been split into written | |
5246 | * and unwritten pieces, or we could just be | |
5247 | * unmapping part of it. But we can't really | |
5248 | * get rid of part of a realtime extent. | |
5249 | */ | |
5250 | if (del.br_state == XFS_EXT_UNWRITTEN || | |
5251 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { | |
5252 | /* | |
5253 | * This piece is unwritten, or we're not | |
5254 | * using unwritten extents. Skip over it. | |
5255 | */ | |
675b5a20 CH |
5256 | ASSERT(end >= mod); |
5257 | end -= mod > del.br_blockcount ? | |
49f693fa | 5258 | del.br_blockcount : mod; |
675b5a20 | 5259 | if (end < got.br_startoff) { |
49f693fa | 5260 | if (--lastx >= 0) |
99a663c1 CH |
5261 | xfs_iext_get_extent(ifp, lastx, |
5262 | &got); | |
49f693fa DC |
5263 | } |
5264 | continue; | |
5265 | } | |
5266 | /* | |
5267 | * It's written, turn it unwritten. | |
5268 | * This is better than zeroing it. | |
5269 | */ | |
5270 | ASSERT(del.br_state == XFS_EXT_NORM); | |
0268fdc3 | 5271 | ASSERT(tp->t_blk_res > 0); |
49f693fa DC |
5272 | /* |
5273 | * If this spans a realtime extent boundary, | |
5274 | * chop it back to the start of the one we end at. | |
5275 | */ | |
5276 | if (del.br_blockcount > mod) { | |
5277 | del.br_startoff += del.br_blockcount - mod; | |
5278 | del.br_startblock += del.br_blockcount - mod; | |
5279 | del.br_blockcount = mod; | |
5280 | } | |
5281 | del.br_state = XFS_EXT_UNWRITTEN; | |
5282 | error = xfs_bmap_add_extent_unwritten_real(tp, ip, | |
4072e4b4 DW |
5283 | whichfork, &lastx, &cur, &del, |
5284 | firstblock, dfops, &logflags); | |
49f693fa DC |
5285 | if (error) |
5286 | goto error0; | |
5287 | goto nodelete; | |
a2ceac1f | 5288 | } |
49f693fa DC |
5289 | if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) { |
5290 | /* | |
5291 | * Realtime extent is lined up at the end but not | |
5292 | * at the front. We'll get rid of full extents if | |
5293 | * we can. | |
5294 | */ | |
5295 | mod = mp->m_sb.sb_rextsize - mod; | |
5296 | if (del.br_blockcount > mod) { | |
5297 | del.br_blockcount -= mod; | |
5298 | del.br_startoff += mod; | |
5299 | del.br_startblock += mod; | |
5300 | } else if ((del.br_startoff == start && | |
5301 | (del.br_state == XFS_EXT_UNWRITTEN || | |
0268fdc3 | 5302 | tp->t_blk_res == 0)) || |
49f693fa DC |
5303 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
5304 | /* | |
5305 | * Can't make it unwritten. There isn't | |
5306 | * a full extent here so just skip it. | |
5307 | */ | |
675b5a20 CH |
5308 | ASSERT(end >= del.br_blockcount); |
5309 | end -= del.br_blockcount; | |
5310 | if (got.br_startoff > end && --lastx >= 0) | |
246eb90a | 5311 | xfs_iext_get_extent(ifp, lastx, &got); |
49f693fa DC |
5312 | continue; |
5313 | } else if (del.br_state == XFS_EXT_UNWRITTEN) { | |
246eb90a CH |
5314 | struct xfs_bmbt_irec prev; |
5315 | ||
49f693fa DC |
5316 | /* |
5317 | * This one is already unwritten. | |
5318 | * It must have a written left neighbor. | |
5319 | * Unwrite the killed part of that one and | |
5320 | * try again. | |
5321 | */ | |
5322 | ASSERT(lastx > 0); | |
246eb90a | 5323 | xfs_iext_get_extent(ifp, lastx - 1, &prev); |
49f693fa DC |
5324 | ASSERT(prev.br_state == XFS_EXT_NORM); |
5325 | ASSERT(!isnullstartblock(prev.br_startblock)); | |
5326 | ASSERT(del.br_startblock == | |
5327 | prev.br_startblock + prev.br_blockcount); | |
5328 | if (prev.br_startoff < start) { | |
5329 | mod = start - prev.br_startoff; | |
5330 | prev.br_blockcount -= mod; | |
5331 | prev.br_startblock += mod; | |
5332 | prev.br_startoff = start; | |
5333 | } | |
5334 | prev.br_state = XFS_EXT_UNWRITTEN; | |
5335 | lastx--; | |
5336 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
4072e4b4 DW |
5337 | ip, whichfork, &lastx, &cur, |
5338 | &prev, firstblock, dfops, | |
5339 | &logflags); | |
49f693fa DC |
5340 | if (error) |
5341 | goto error0; | |
5342 | goto nodelete; | |
5343 | } else { | |
5344 | ASSERT(del.br_state == XFS_EXT_NORM); | |
5345 | del.br_state = XFS_EXT_UNWRITTEN; | |
5346 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
4072e4b4 DW |
5347 | ip, whichfork, &lastx, &cur, |
5348 | &del, firstblock, dfops, | |
5349 | &logflags); | |
49f693fa DC |
5350 | if (error) |
5351 | goto error0; | |
5352 | goto nodelete; | |
5353 | } | |
5354 | } | |
a2ceac1f | 5355 | |
8359e0b9 | 5356 | if (wasdel) { |
ad68fd19 CH |
5357 | error = xfs_bmap_del_extent_delay(ip, whichfork, &lastx, |
5358 | &got, &del); | |
5359 | } else { | |
ad68fd19 CH |
5360 | error = xfs_bmap_del_extent_real(ip, tp, &lastx, dfops, |
5361 | cur, &del, &tmp_logflags, whichfork, | |
5362 | flags); | |
5363 | logflags |= tmp_logflags; | |
e9fa15aa | 5364 | } |
8359e0b9 | 5365 | |
49f693fa DC |
5366 | if (error) |
5367 | goto error0; | |
8359e0b9 | 5368 | |
594956fa | 5369 | max_len -= del.br_blockcount; |
675b5a20 | 5370 | end = del.br_startoff - 1; |
49f693fa | 5371 | nodelete: |
a2ceac1f | 5372 | /* |
49f693fa | 5373 | * If not done go on to the next (previous) record. |
a2ceac1f | 5374 | */ |
675b5a20 | 5375 | if (end != (xfs_fileoff_t)-1 && end >= start) { |
49f693fa | 5376 | if (lastx >= 0) { |
246eb90a | 5377 | xfs_iext_get_extent(ifp, lastx, &got); |
675b5a20 | 5378 | if (got.br_startoff > end && --lastx >= 0) |
246eb90a | 5379 | xfs_iext_get_extent(ifp, lastx, &got); |
49f693fa DC |
5380 | } |
5381 | extno++; | |
a2ceac1f | 5382 | } |
a2ceac1f | 5383 | } |
675b5a20 | 5384 | if (end == (xfs_fileoff_t)-1 || end < start || lastx < 0) |
3d36acda DW |
5385 | *rlen = 0; |
5386 | else | |
675b5a20 | 5387 | *rlen = end - start + 1; |
56b2de80 | 5388 | |
49f693fa DC |
5389 | /* |
5390 | * Convert to a btree if necessary. | |
5391 | */ | |
5392 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
5393 | ASSERT(cur == NULL); | |
f33cea1a | 5394 | error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, |
49f693fa DC |
5395 | &cur, 0, &tmp_logflags, whichfork); |
5396 | logflags |= tmp_logflags; | |
5397 | if (error) | |
5398 | goto error0; | |
56b2de80 | 5399 | } |
56b2de80 | 5400 | /* |
49f693fa | 5401 | * transform from btree to extents, give it cur |
56b2de80 | 5402 | */ |
49f693fa DC |
5403 | else if (xfs_bmap_wants_extents(ip, whichfork)) { |
5404 | ASSERT(cur != NULL); | |
5405 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags, | |
5406 | whichfork); | |
5407 | logflags |= tmp_logflags; | |
5408 | if (error) | |
5409 | goto error0; | |
56b2de80 | 5410 | } |
49f693fa DC |
5411 | /* |
5412 | * transform from extents to local? | |
5413 | */ | |
5414 | error = 0; | |
5415 | error0: | |
5416 | /* | |
5417 | * Log everything. Do this after conversion, there's no point in | |
5418 | * logging the extent records if we've converted to btree format. | |
5419 | */ | |
5420 | if ((logflags & xfs_ilog_fext(whichfork)) && | |
5421 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
5422 | logflags &= ~xfs_ilog_fext(whichfork); | |
5423 | else if ((logflags & xfs_ilog_fbroot(whichfork)) && | |
5424 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) | |
5425 | logflags &= ~xfs_ilog_fbroot(whichfork); | |
5426 | /* | |
5427 | * Log inode even in the error case, if the transaction | |
5428 | * is dirty we'll need to shut down the filesystem. | |
5429 | */ | |
5430 | if (logflags) | |
5431 | xfs_trans_log_inode(tp, ip, logflags); | |
5432 | if (cur) { | |
5433 | if (!error) { | |
5434 | *firstblock = cur->bc_private.b.firstblock; | |
5435 | cur->bc_private.b.allocated = 0; | |
56b2de80 | 5436 | } |
49f693fa DC |
5437 | xfs_btree_del_cursor(cur, |
5438 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
56b2de80 | 5439 | } |
49f693fa | 5440 | return error; |
a2ceac1f | 5441 | } |
ff105f75 | 5442 | |
3d36acda DW |
5443 | /* Unmap a range of a file. */ |
5444 | int | |
5445 | xfs_bunmapi( | |
5446 | xfs_trans_t *tp, | |
5447 | struct xfs_inode *ip, | |
5448 | xfs_fileoff_t bno, | |
5449 | xfs_filblks_t len, | |
5450 | int flags, | |
5451 | xfs_extnum_t nexts, | |
5452 | xfs_fsblock_t *firstblock, | |
5453 | struct xfs_defer_ops *dfops, | |
5454 | int *done) | |
5455 | { | |
5456 | int error; | |
5457 | ||
5458 | error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock, | |
5459 | dfops); | |
5460 | *done = (len == 0); | |
5461 | return error; | |
5462 | } | |
5463 | ||
5a35bf2c DC |
5464 | /* |
5465 | * Determine whether an extent shift can be accomplished by a merge with the | |
5466 | * extent that precedes the target hole of the shift. | |
5467 | */ | |
5468 | STATIC bool | |
5469 | xfs_bmse_can_merge( | |
5470 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5471 | struct xfs_bmbt_irec *got, /* current extent to shift */ | |
5472 | xfs_fileoff_t shift) /* shift fsb */ | |
5473 | { | |
5474 | xfs_fileoff_t startoff; | |
5475 | ||
5476 | startoff = got->br_startoff - shift; | |
5477 | ||
5478 | /* | |
5479 | * The extent, once shifted, must be adjacent in-file and on-disk with | |
5480 | * the preceding extent. | |
5481 | */ | |
5482 | if ((left->br_startoff + left->br_blockcount != startoff) || | |
5483 | (left->br_startblock + left->br_blockcount != got->br_startblock) || | |
5484 | (left->br_state != got->br_state) || | |
5485 | (left->br_blockcount + got->br_blockcount > MAXEXTLEN)) | |
5486 | return false; | |
5487 | ||
5488 | return true; | |
5489 | } | |
5490 | ||
5491 | /* | |
5492 | * A bmap extent shift adjusts the file offset of an extent to fill a preceding | |
5493 | * hole in the file. If an extent shift would result in the extent being fully | |
5494 | * adjacent to the extent that currently precedes the hole, we can merge with | |
5495 | * the preceding extent rather than do the shift. | |
5496 | * | |
5497 | * This function assumes the caller has verified a shift-by-merge is possible | |
5498 | * with the provided extents via xfs_bmse_can_merge(). | |
5499 | */ | |
5500 | STATIC int | |
5501 | xfs_bmse_merge( | |
5502 | struct xfs_inode *ip, | |
5503 | int whichfork, | |
5504 | xfs_fileoff_t shift, /* shift fsb */ | |
5505 | int current_ext, /* idx of gotp */ | |
2f082827 CH |
5506 | struct xfs_bmbt_irec *got, /* extent to shift */ |
5507 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5a35bf2c | 5508 | struct xfs_btree_cur *cur, |
2f082827 CH |
5509 | int *logflags, /* output */ |
5510 | struct xfs_defer_ops *dfops) | |
5a35bf2c | 5511 | { |
2f082827 | 5512 | struct xfs_bmbt_irec new; |
5a35bf2c DC |
5513 | xfs_filblks_t blockcount; |
5514 | int error, i; | |
19ebedcf | 5515 | struct xfs_mount *mp = ip->i_mount; |
5a35bf2c | 5516 | |
2f082827 | 5517 | blockcount = left->br_blockcount + got->br_blockcount; |
5a35bf2c DC |
5518 | |
5519 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5520 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
2f082827 | 5521 | ASSERT(xfs_bmse_can_merge(left, got, shift)); |
5a35bf2c | 5522 | |
2f082827 CH |
5523 | new = *left; |
5524 | new.br_blockcount = blockcount; | |
5a35bf2c DC |
5525 | |
5526 | /* | |
5527 | * Update the on-disk extent count, the btree if necessary and log the | |
5528 | * inode. | |
5529 | */ | |
5530 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5531 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5532 | *logflags |= XFS_ILOG_CORE; | |
5533 | if (!cur) { | |
5534 | *logflags |= XFS_ILOG_DEXT; | |
2f082827 | 5535 | goto done; |
5a35bf2c DC |
5536 | } |
5537 | ||
5538 | /* lookup and remove the extent to merge */ | |
70a93110 | 5539 | error = xfs_bmbt_lookup_eq(cur, got, &i); |
5a35bf2c DC |
5540 | if (error) |
5541 | return error; | |
19ebedcf | 5542 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5543 | |
5544 | error = xfs_btree_delete(cur, &i); | |
5545 | if (error) | |
5546 | return error; | |
19ebedcf | 5547 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5548 | |
5549 | /* lookup and update size of the previous extent */ | |
70a93110 | 5550 | error = xfs_bmbt_lookup_eq(cur, left, &i); |
5a35bf2c DC |
5551 | if (error) |
5552 | return error; | |
19ebedcf | 5553 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c | 5554 | |
d0e5f1ff | 5555 | error = xfs_bmbt_update(cur, &new); |
2f082827 CH |
5556 | if (error) |
5557 | return error; | |
5558 | ||
5559 | done: | |
df926c07 CH |
5560 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), |
5561 | current_ext - 1, &new); | |
2f082827 | 5562 | xfs_iext_remove(ip, current_ext, 1, 0); |
5a35bf2c | 5563 | |
479284b7 | 5564 | /* update reverse mapping. rmap functions merge the rmaps for us */ |
2f082827 CH |
5565 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got); |
5566 | if (error) | |
5567 | return error; | |
479284b7 DW |
5568 | memcpy(&new, got, sizeof(new)); |
5569 | new.br_startoff = left->br_startoff + left->br_blockcount; | |
2f082827 | 5570 | return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new); |
5a35bf2c DC |
5571 | } |
5572 | ||
364937e2 CH |
5573 | static int |
5574 | xfs_bmap_shift_update_extent( | |
5575 | struct xfs_inode *ip, | |
5576 | int whichfork, | |
5577 | xfs_extnum_t idx, | |
5578 | struct xfs_bmbt_irec *got, | |
5579 | struct xfs_btree_cur *cur, | |
5580 | int *logflags, | |
5581 | struct xfs_defer_ops *dfops, | |
5582 | xfs_fileoff_t startoff) | |
5a35bf2c | 5583 | { |
364937e2 | 5584 | struct xfs_mount *mp = ip->i_mount; |
f36ccac2 | 5585 | struct xfs_bmbt_irec prev = *got; |
364937e2 | 5586 | int error, i; |
2f082827 | 5587 | |
5a35bf2c | 5588 | *logflags |= XFS_ILOG_CORE; |
2f082827 | 5589 | |
f36ccac2 | 5590 | got->br_startoff = startoff; |
2f082827 CH |
5591 | |
5592 | if (cur) { | |
f36ccac2 | 5593 | error = xfs_bmbt_lookup_eq(cur, &prev, &i); |
2f082827 CH |
5594 | if (error) |
5595 | return error; | |
5596 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); | |
5597 | ||
f36ccac2 | 5598 | error = xfs_bmbt_update(cur, got); |
2f082827 CH |
5599 | if (error) |
5600 | return error; | |
5601 | } else { | |
5a35bf2c | 5602 | *logflags |= XFS_ILOG_DEXT; |
5a35bf2c DC |
5603 | } |
5604 | ||
f36ccac2 | 5605 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), idx, got); |
d7f80320 | 5606 | |
d7f80320 | 5607 | /* update reverse mapping */ |
f36ccac2 | 5608 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &prev); |
d7f80320 DW |
5609 | if (error) |
5610 | return error; | |
f36ccac2 | 5611 | return xfs_rmap_map_extent(mp, dfops, ip, whichfork, got); |
5a35bf2c DC |
5612 | } |
5613 | ||
ff105f75 | 5614 | int |
a9f5d25e | 5615 | xfs_bmap_collapse_extents( |
ff105f75 DC |
5616 | struct xfs_trans *tp, |
5617 | struct xfs_inode *ip, | |
19ebedcf | 5618 | xfs_fileoff_t *next_fsb, |
ff105f75 | 5619 | xfs_fileoff_t offset_shift_fsb, |
a9f5d25e | 5620 | bool *done, |
19ebedcf | 5621 | xfs_fileoff_t stop_fsb, |
ff105f75 | 5622 | xfs_fsblock_t *firstblock, |
a9f5d25e | 5623 | struct xfs_defer_ops *dfops) |
ff105f75 | 5624 | { |
a9f5d25e CH |
5625 | int whichfork = XFS_DATA_FORK; |
5626 | struct xfs_mount *mp = ip->i_mount; | |
5627 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5628 | struct xfs_btree_cur *cur = NULL; | |
364937e2 | 5629 | struct xfs_bmbt_irec got, prev; |
a9f5d25e | 5630 | xfs_extnum_t current_ext; |
364937e2 | 5631 | xfs_fileoff_t new_startoff; |
a9f5d25e CH |
5632 | int error = 0; |
5633 | int logflags = 0; | |
ff105f75 DC |
5634 | |
5635 | if (unlikely(XFS_TEST_ERROR( | |
5636 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5637 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 5638 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a9f5d25e | 5639 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); |
12b53197 | 5640 | return -EFSCORRUPTED; |
ff105f75 DC |
5641 | } |
5642 | ||
5643 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 5644 | return -EIO; |
ff105f75 | 5645 | |
a9f5d25e | 5646 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); |
ff105f75 | 5647 | |
ff105f75 | 5648 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
ff105f75 DC |
5649 | error = xfs_iread_extents(tp, ip, whichfork); |
5650 | if (error) | |
5651 | return error; | |
5652 | } | |
5653 | ||
ff105f75 DC |
5654 | if (ifp->if_flags & XFS_IFBROOT) { |
5655 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5656 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 5657 | cur->bc_private.b.dfops = dfops; |
ff105f75 | 5658 | cur->bc_private.b.flags = 0; |
5a35bf2c DC |
5659 | } |
5660 | ||
a9f5d25e CH |
5661 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, ¤t_ext, &got)) { |
5662 | *done = true; | |
5663 | goto del_cursor; | |
5664 | } | |
364937e2 | 5665 | XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock)); |
a9f5d25e | 5666 | |
364937e2 | 5667 | new_startoff = got.br_startoff - offset_shift_fsb; |
bac20498 | 5668 | if (xfs_iext_get_extent(ifp, current_ext - 1, &prev)) { |
364937e2 CH |
5669 | if (new_startoff < prev.br_startoff + prev.br_blockcount) { |
5670 | error = -EINVAL; | |
5671 | goto del_cursor; | |
5672 | } | |
5673 | ||
364937e2 CH |
5674 | if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) { |
5675 | error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb, | |
5676 | current_ext, &got, &prev, cur, | |
5677 | &logflags, dfops); | |
5678 | if (error) | |
5679 | goto del_cursor; | |
bac20498 CH |
5680 | |
5681 | /* update got after merge */ | |
5682 | if (!xfs_iext_get_extent(ifp, current_ext, &got)) { | |
5683 | *done = true; | |
5684 | goto del_cursor; | |
5685 | } | |
364937e2 CH |
5686 | goto done; |
5687 | } | |
5688 | } else { | |
5689 | if (got.br_startoff < offset_shift_fsb) { | |
5690 | error = -EINVAL; | |
5691 | goto del_cursor; | |
5692 | } | |
5693 | } | |
5694 | ||
5695 | error = xfs_bmap_shift_update_extent(ip, whichfork, current_ext, &got, | |
5696 | cur, &logflags, dfops, new_startoff); | |
a9f5d25e CH |
5697 | if (error) |
5698 | goto del_cursor; | |
bac20498 CH |
5699 | |
5700 | if (!xfs_iext_get_extent(ifp, ++current_ext, &got)) { | |
5701 | *done = true; | |
5702 | goto del_cursor; | |
a9f5d25e | 5703 | } |
a9f5d25e | 5704 | |
bac20498 | 5705 | done: |
364937e2 | 5706 | *next_fsb = got.br_startoff; |
a9f5d25e CH |
5707 | del_cursor: |
5708 | if (cur) | |
5709 | xfs_btree_del_cursor(cur, | |
5710 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
a9f5d25e CH |
5711 | if (logflags) |
5712 | xfs_trans_log_inode(tp, ip, logflags); | |
a9f5d25e CH |
5713 | return error; |
5714 | } | |
5715 | ||
5716 | int | |
5717 | xfs_bmap_insert_extents( | |
5718 | struct xfs_trans *tp, | |
5719 | struct xfs_inode *ip, | |
5720 | xfs_fileoff_t *next_fsb, | |
5721 | xfs_fileoff_t offset_shift_fsb, | |
5722 | bool *done, | |
5723 | xfs_fileoff_t stop_fsb, | |
5724 | xfs_fsblock_t *firstblock, | |
5725 | struct xfs_defer_ops *dfops) | |
5726 | { | |
5727 | int whichfork = XFS_DATA_FORK; | |
5728 | struct xfs_mount *mp = ip->i_mount; | |
5729 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5730 | struct xfs_btree_cur *cur = NULL; | |
364937e2 | 5731 | struct xfs_bmbt_irec got, next, s; |
a9f5d25e CH |
5732 | xfs_extnum_t current_ext; |
5733 | xfs_extnum_t total_extents; | |
5734 | xfs_extnum_t stop_extent; | |
364937e2 | 5735 | xfs_fileoff_t new_startoff; |
a9f5d25e CH |
5736 | int error = 0; |
5737 | int logflags = 0; | |
5738 | ||
5739 | if (unlikely(XFS_TEST_ERROR( | |
5740 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5741 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
5742 | mp, XFS_ERRTAG_BMAPIFORMAT))) { | |
5743 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
5744 | return -EFSCORRUPTED; | |
5745 | } | |
5746 | ||
5747 | if (XFS_FORCED_SHUTDOWN(mp)) | |
5748 | return -EIO; | |
5749 | ||
5750 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); | |
5751 | ||
5752 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
5753 | error = xfs_iread_extents(tp, ip, whichfork); | |
5754 | if (error) | |
5755 | return error; | |
5756 | } | |
5757 | ||
5758 | if (ifp->if_flags & XFS_IFBROOT) { | |
5759 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5760 | cur->bc_private.b.firstblock = *firstblock; | |
5761 | cur->bc_private.b.dfops = dfops; | |
5762 | cur->bc_private.b.flags = 0; | |
5763 | } | |
5764 | ||
5765 | /* | |
5766 | * There may be delalloc extents in the data fork before the range we | |
5767 | * are collapsing out, so we cannot use the count of real extents here. | |
5768 | * Instead we have to calculate it from the incore fork. | |
5769 | */ | |
5770 | total_extents = xfs_iext_count(ifp); | |
5771 | if (total_extents == 0) { | |
5772 | *done = true; | |
19ebedcf DC |
5773 | goto del_cursor; |
5774 | } | |
5775 | ||
5776 | /* | |
5777 | * In case of first right shift, we need to initialize next_fsb | |
5778 | */ | |
5779 | if (*next_fsb == NULLFSBLOCK) { | |
b6ad780b | 5780 | current_ext = total_extents - 1; |
2f082827 CH |
5781 | xfs_iext_get_extent(ifp, current_ext, &got); |
5782 | if (stop_fsb > got.br_startoff) { | |
a9f5d25e | 5783 | *done = true; |
19ebedcf DC |
5784 | goto del_cursor; |
5785 | } | |
2f082827 | 5786 | *next_fsb = got.br_startoff; |
b6ad780b CH |
5787 | } else { |
5788 | /* | |
5789 | * Look up the extent index for the fsb where we start shifting. We can | |
5790 | * henceforth iterate with current_ext as extent list changes are locked | |
5791 | * out via ilock. | |
5792 | * | |
2f082827 CH |
5793 | * If next_fsb lies in a hole beyond which there are no extents we are |
5794 | * done. | |
b6ad780b | 5795 | */ |
2f082827 CH |
5796 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, ¤t_ext, |
5797 | &got)) { | |
a9f5d25e | 5798 | *done = true; |
b6ad780b CH |
5799 | goto del_cursor; |
5800 | } | |
19ebedcf | 5801 | } |
364937e2 | 5802 | XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock)); |
19ebedcf DC |
5803 | |
5804 | /* Lookup the extent index at which we have to stop */ | |
a9f5d25e CH |
5805 | xfs_iext_lookup_extent(ip, ifp, stop_fsb, &stop_extent, &s); |
5806 | /* Make stop_extent exclusive of shift range */ | |
5807 | stop_extent--; | |
5808 | if (current_ext <= stop_extent) { | |
5809 | error = -EIO; | |
5810 | goto del_cursor; | |
19ebedcf DC |
5811 | } |
5812 | ||
364937e2 CH |
5813 | new_startoff = got.br_startoff + offset_shift_fsb; |
5814 | if (current_ext < total_extents - 1) { | |
5815 | xfs_iext_get_extent(ifp, current_ext + 1, &next); | |
5816 | if (new_startoff + got.br_blockcount > next.br_startoff) { | |
5817 | error = -EINVAL; | |
5818 | goto del_cursor; | |
5819 | } | |
5820 | ||
5821 | /* | |
5822 | * Unlike a left shift (which involves a hole punch), a right | |
5823 | * shift does not modify extent neighbors in any way. We should | |
5824 | * never find mergeable extents in this scenario. Check anyways | |
5825 | * and warn if we encounter two extents that could be one. | |
5826 | */ | |
5827 | if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb)) | |
5828 | WARN_ON_ONCE(1); | |
5829 | } | |
5830 | ||
5831 | error = xfs_bmap_shift_update_extent(ip, whichfork, current_ext, &got, | |
5832 | cur, &logflags, dfops, new_startoff); | |
6835c363 CH |
5833 | if (error) |
5834 | goto del_cursor; | |
364937e2 | 5835 | if (--current_ext == stop_extent) { |
a9f5d25e | 5836 | *done = true; |
6835c363 | 5837 | goto del_cursor; |
ff105f75 | 5838 | } |
6835c363 | 5839 | xfs_iext_get_extent(ifp, current_ext, &got); |
ff105f75 | 5840 | |
6835c363 | 5841 | *next_fsb = got.br_startoff; |
ff105f75 DC |
5842 | del_cursor: |
5843 | if (cur) | |
5844 | xfs_btree_del_cursor(cur, | |
5845 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
5a35bf2c DC |
5846 | if (logflags) |
5847 | xfs_trans_log_inode(tp, ip, logflags); | |
ff105f75 DC |
5848 | return error; |
5849 | } | |
19ebedcf DC |
5850 | |
5851 | /* | |
5852 | * Splits an extent into two extents at split_fsb block such that it is | |
5853 | * the first block of the current_ext. @current_ext is a target extent | |
5854 | * to be split. @split_fsb is a block where the extents is split. | |
5855 | * If split_fsb lies in a hole or the first block of extents, just return 0. | |
5856 | */ | |
5857 | STATIC int | |
5858 | xfs_bmap_split_extent_at( | |
5859 | struct xfs_trans *tp, | |
5860 | struct xfs_inode *ip, | |
5861 | xfs_fileoff_t split_fsb, | |
5862 | xfs_fsblock_t *firstfsb, | |
f33cea1a | 5863 | struct xfs_defer_ops *dfops) |
19ebedcf DC |
5864 | { |
5865 | int whichfork = XFS_DATA_FORK; | |
5866 | struct xfs_btree_cur *cur = NULL; | |
19ebedcf DC |
5867 | struct xfs_bmbt_irec got; |
5868 | struct xfs_bmbt_irec new; /* split extent */ | |
5869 | struct xfs_mount *mp = ip->i_mount; | |
5870 | struct xfs_ifork *ifp; | |
5871 | xfs_fsblock_t gotblkcnt; /* new block count for got */ | |
5872 | xfs_extnum_t current_ext; | |
5873 | int error = 0; | |
5874 | int logflags = 0; | |
5875 | int i = 0; | |
5876 | ||
5877 | if (unlikely(XFS_TEST_ERROR( | |
5878 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5879 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 5880 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
19ebedcf DC |
5881 | XFS_ERROR_REPORT("xfs_bmap_split_extent_at", |
5882 | XFS_ERRLEVEL_LOW, mp); | |
5883 | return -EFSCORRUPTED; | |
5884 | } | |
5885 | ||
5886 | if (XFS_FORCED_SHUTDOWN(mp)) | |
5887 | return -EIO; | |
5888 | ||
5889 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
5890 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
5891 | /* Read in all the extents */ | |
5892 | error = xfs_iread_extents(tp, ip, whichfork); | |
5893 | if (error) | |
5894 | return error; | |
5895 | } | |
5896 | ||
5897 | /* | |
455044a6 | 5898 | * If there are not extents, or split_fsb lies in a hole we are done. |
19ebedcf | 5899 | */ |
455044a6 CH |
5900 | if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, ¤t_ext, &got) || |
5901 | got.br_startoff >= split_fsb) | |
19ebedcf DC |
5902 | return 0; |
5903 | ||
5904 | gotblkcnt = split_fsb - got.br_startoff; | |
5905 | new.br_startoff = split_fsb; | |
5906 | new.br_startblock = got.br_startblock + gotblkcnt; | |
5907 | new.br_blockcount = got.br_blockcount - gotblkcnt; | |
5908 | new.br_state = got.br_state; | |
5909 | ||
5910 | if (ifp->if_flags & XFS_IFBROOT) { | |
5911 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5912 | cur->bc_private.b.firstblock = *firstfsb; | |
f33cea1a | 5913 | cur->bc_private.b.dfops = dfops; |
19ebedcf | 5914 | cur->bc_private.b.flags = 0; |
70a93110 | 5915 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
19ebedcf DC |
5916 | if (error) |
5917 | goto del_cursor; | |
5918 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
5919 | } | |
5920 | ||
19ebedcf | 5921 | got.br_blockcount = gotblkcnt; |
df926c07 CH |
5922 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), |
5923 | current_ext, &got); | |
19ebedcf DC |
5924 | |
5925 | logflags = XFS_ILOG_CORE; | |
5926 | if (cur) { | |
d0e5f1ff | 5927 | error = xfs_bmbt_update(cur, &got); |
19ebedcf DC |
5928 | if (error) |
5929 | goto del_cursor; | |
5930 | } else | |
5931 | logflags |= XFS_ILOG_DEXT; | |
5932 | ||
5933 | /* Add new extent */ | |
5934 | current_ext++; | |
5935 | xfs_iext_insert(ip, current_ext, 1, &new, 0); | |
5936 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5937 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
5938 | ||
5939 | if (cur) { | |
70a93110 | 5940 | error = xfs_bmbt_lookup_eq(cur, &new, &i); |
19ebedcf DC |
5941 | if (error) |
5942 | goto del_cursor; | |
5943 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor); | |
19ebedcf DC |
5944 | error = xfs_btree_insert(cur, &i); |
5945 | if (error) | |
5946 | goto del_cursor; | |
5947 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
5948 | } | |
5949 | ||
5950 | /* | |
5951 | * Convert to a btree if necessary. | |
5952 | */ | |
5953 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
5954 | int tmp_logflags; /* partial log flag return val */ | |
5955 | ||
5956 | ASSERT(cur == NULL); | |
f33cea1a | 5957 | error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops, |
19ebedcf DC |
5958 | &cur, 0, &tmp_logflags, whichfork); |
5959 | logflags |= tmp_logflags; | |
5960 | } | |
5961 | ||
5962 | del_cursor: | |
5963 | if (cur) { | |
5964 | cur->bc_private.b.allocated = 0; | |
5965 | xfs_btree_del_cursor(cur, | |
5966 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
5967 | } | |
5968 | ||
5969 | if (logflags) | |
5970 | xfs_trans_log_inode(tp, ip, logflags); | |
5971 | return error; | |
5972 | } | |
5973 | ||
5974 | int | |
5975 | xfs_bmap_split_extent( | |
5976 | struct xfs_inode *ip, | |
5977 | xfs_fileoff_t split_fsb) | |
5978 | { | |
5979 | struct xfs_mount *mp = ip->i_mount; | |
5980 | struct xfs_trans *tp; | |
f33cea1a | 5981 | struct xfs_defer_ops dfops; |
19ebedcf | 5982 | xfs_fsblock_t firstfsb; |
19ebedcf DC |
5983 | int error; |
5984 | ||
9074815c CH |
5985 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, |
5986 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); | |
5987 | if (error) | |
19ebedcf | 5988 | return error; |
19ebedcf DC |
5989 | |
5990 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5991 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
5992 | ||
f33cea1a | 5993 | xfs_defer_init(&dfops, &firstfsb); |
19ebedcf DC |
5994 | |
5995 | error = xfs_bmap_split_extent_at(tp, ip, split_fsb, | |
f33cea1a | 5996 | &firstfsb, &dfops); |
19ebedcf DC |
5997 | if (error) |
5998 | goto out; | |
5999 | ||
5c33baee | 6000 | error = xfs_defer_finish(&tp, &dfops); |
19ebedcf DC |
6001 | if (error) |
6002 | goto out; | |
6003 | ||
de5a3f46 | 6004 | return xfs_trans_commit(tp); |
19ebedcf DC |
6005 | |
6006 | out: | |
f33cea1a | 6007 | xfs_defer_cancel(&dfops); |
3d7434fe | 6008 | xfs_trans_cancel(tp); |
19ebedcf DC |
6009 | return error; |
6010 | } | |
aeb88300 DW |
6011 | |
6012 | /* Deferred mapping is only for real extents in the data fork. */ | |
6013 | static bool | |
6014 | xfs_bmap_is_update_needed( | |
6015 | struct xfs_bmbt_irec *bmap) | |
6016 | { | |
6017 | return bmap->br_startblock != HOLESTARTBLOCK && | |
6018 | bmap->br_startblock != DELAYSTARTBLOCK; | |
6019 | } | |
6020 | ||
6021 | /* Record a bmap intent. */ | |
6022 | static int | |
6023 | __xfs_bmap_add( | |
6024 | struct xfs_mount *mp, | |
6025 | struct xfs_defer_ops *dfops, | |
6026 | enum xfs_bmap_intent_type type, | |
6027 | struct xfs_inode *ip, | |
6028 | int whichfork, | |
6029 | struct xfs_bmbt_irec *bmap) | |
6030 | { | |
6031 | int error; | |
6032 | struct xfs_bmap_intent *bi; | |
6033 | ||
6034 | trace_xfs_bmap_defer(mp, | |
6035 | XFS_FSB_TO_AGNO(mp, bmap->br_startblock), | |
6036 | type, | |
6037 | XFS_FSB_TO_AGBNO(mp, bmap->br_startblock), | |
6038 | ip->i_ino, whichfork, | |
6039 | bmap->br_startoff, | |
6040 | bmap->br_blockcount, | |
6041 | bmap->br_state); | |
6042 | ||
6043 | bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS); | |
6044 | INIT_LIST_HEAD(&bi->bi_list); | |
6045 | bi->bi_type = type; | |
6046 | bi->bi_owner = ip; | |
6047 | bi->bi_whichfork = whichfork; | |
6048 | bi->bi_bmap = *bmap; | |
6049 | ||
277d3c3a | 6050 | error = xfs_defer_ijoin(dfops, bi->bi_owner); |
aeb88300 DW |
6051 | if (error) { |
6052 | kmem_free(bi); | |
6053 | return error; | |
6054 | } | |
6055 | ||
6056 | xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list); | |
6057 | return 0; | |
6058 | } | |
6059 | ||
6060 | /* Map an extent into a file. */ | |
6061 | int | |
6062 | xfs_bmap_map_extent( | |
6063 | struct xfs_mount *mp, | |
6064 | struct xfs_defer_ops *dfops, | |
6065 | struct xfs_inode *ip, | |
6066 | struct xfs_bmbt_irec *PREV) | |
6067 | { | |
6068 | if (!xfs_bmap_is_update_needed(PREV)) | |
6069 | return 0; | |
6070 | ||
6071 | return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip, | |
6072 | XFS_DATA_FORK, PREV); | |
6073 | } | |
6074 | ||
6075 | /* Unmap an extent out of a file. */ | |
6076 | int | |
6077 | xfs_bmap_unmap_extent( | |
6078 | struct xfs_mount *mp, | |
6079 | struct xfs_defer_ops *dfops, | |
6080 | struct xfs_inode *ip, | |
6081 | struct xfs_bmbt_irec *PREV) | |
6082 | { | |
6083 | if (!xfs_bmap_is_update_needed(PREV)) | |
6084 | return 0; | |
6085 | ||
6086 | return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip, | |
6087 | XFS_DATA_FORK, PREV); | |
6088 | } | |
6089 | ||
6090 | /* | |
6091 | * Process one of the deferred bmap operations. We pass back the | |
6092 | * btree cursor to maintain our lock on the bmapbt between calls. | |
6093 | */ | |
6094 | int | |
6095 | xfs_bmap_finish_one( | |
6096 | struct xfs_trans *tp, | |
6097 | struct xfs_defer_ops *dfops, | |
6098 | struct xfs_inode *ip, | |
6099 | enum xfs_bmap_intent_type type, | |
6100 | int whichfork, | |
6101 | xfs_fileoff_t startoff, | |
6102 | xfs_fsblock_t startblock, | |
594956fa | 6103 | xfs_filblks_t *blockcount, |
aeb88300 DW |
6104 | xfs_exntst_t state) |
6105 | { | |
594956fa DW |
6106 | xfs_fsblock_t firstfsb; |
6107 | int error = 0; | |
aeb88300 | 6108 | |
66d19ae1 DW |
6109 | /* |
6110 | * firstfsb is tied to the transaction lifetime and is used to | |
6111 | * ensure correct AG locking order and schedule work item | |
6112 | * continuations. XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us | |
6113 | * to only making one bmap call per transaction, so it should | |
6114 | * be safe to have it as a local variable here. | |
6115 | */ | |
6116 | firstfsb = NULLFSBLOCK; | |
6117 | ||
aeb88300 DW |
6118 | trace_xfs_bmap_deferred(tp->t_mountp, |
6119 | XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type, | |
6120 | XFS_FSB_TO_AGBNO(tp->t_mountp, startblock), | |
594956fa | 6121 | ip->i_ino, whichfork, startoff, *blockcount, state); |
aeb88300 | 6122 | |
95855a23 | 6123 | if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK)) |
aeb88300 | 6124 | return -EFSCORRUPTED; |
aeb88300 DW |
6125 | |
6126 | if (XFS_TEST_ERROR(false, tp->t_mountp, | |
e2a190dd | 6127 | XFS_ERRTAG_BMAP_FINISH_ONE)) |
aeb88300 DW |
6128 | return -EIO; |
6129 | ||
6130 | switch (type) { | |
6131 | case XFS_BMAP_MAP: | |
594956fa | 6132 | error = xfs_bmapi_remap(tp, ip, startoff, *blockcount, |
0b44aa85 | 6133 | startblock, dfops); |
594956fa | 6134 | *blockcount = 0; |
aeb88300 DW |
6135 | break; |
6136 | case XFS_BMAP_UNMAP: | |
594956fa DW |
6137 | error = __xfs_bunmapi(tp, ip, startoff, blockcount, |
6138 | XFS_BMAPI_REMAP, 1, &firstfsb, dfops); | |
aeb88300 DW |
6139 | break; |
6140 | default: | |
6141 | ASSERT(0); | |
6142 | error = -EFSCORRUPTED; | |
6143 | } | |
6144 | ||
6145 | return error; | |
6146 | } |