]>
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" | |
56d3fc2b | 37 | #include "xfs_errortag.h" |
b626fb59 DC |
38 | #include "xfs_trans_space.h" |
39 | #include "xfs_trace.h" | |
40 | #include "xfs_attr_leaf.h" | |
41 | #include "xfs_quota_defs.h" | |
85aec44f | 42 | #include "xfs_rmap.h" |
cf8ce220 | 43 | #include "xfs_ag_resv.h" |
cfe32f0d | 44 | #include "xfs_refcount.h" |
b626fb59 | 45 | |
2bd0ea18 | 46 | |
5e656dbb BN |
47 | kmem_zone_t *xfs_bmap_free_item_zone; |
48 | ||
49 | /* | |
49f693fa | 50 | * Miscellaneous helper functions |
5e656dbb | 51 | */ |
5e656dbb | 52 | |
5e656dbb | 53 | /* |
49f693fa DC |
54 | * Compute and fill in the value of the maximum depth of a bmap btree |
55 | * in this filesystem. Done once, during mount. | |
5e656dbb | 56 | */ |
49f693fa DC |
57 | void |
58 | xfs_bmap_compute_maxlevels( | |
59 | xfs_mount_t *mp, /* file system mount structure */ | |
60 | int whichfork) /* data or attr fork */ | |
61 | { | |
62 | int level; /* btree level */ | |
63 | uint maxblocks; /* max blocks at this level */ | |
64 | uint maxleafents; /* max leaf entries possible */ | |
65 | int maxrootrecs; /* max records in root block */ | |
66 | int minleafrecs; /* min records in leaf block */ | |
67 | int minnoderecs; /* min records in node block */ | |
68 | int sz; /* root block size */ | |
5e656dbb | 69 | |
49f693fa DC |
70 | /* |
71 | * The maximum number of extents in a file, hence the maximum | |
72 | * number of leaf entries, is controlled by the type of di_nextents | |
73 | * (a signed 32-bit number, xfs_extnum_t), or by di_anextents | |
74 | * (a signed 16-bit number, xfs_aextnum_t). | |
75 | * | |
76 | * Note that we can no longer assume that if we are in ATTR1 that | |
77 | * the fork offset of all the inodes will be | |
78 | * (xfs_default_attroffset(ip) >> 3) because we could have mounted | |
79 | * with ATTR2 and then mounted back with ATTR1, keeping the | |
80 | * di_forkoff's fixed but probably at various positions. Therefore, | |
81 | * for both ATTR1 and ATTR2 we have to assume the worst case scenario | |
82 | * of a minimum size available. | |
83 | */ | |
84 | if (whichfork == XFS_DATA_FORK) { | |
85 | maxleafents = MAXEXTNUM; | |
86 | sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS); | |
87 | } else { | |
88 | maxleafents = MAXAEXTNUM; | |
89 | sz = XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
90 | } | |
ff105f75 | 91 | maxrootrecs = xfs_bmdr_maxrecs(sz, 0); |
49f693fa DC |
92 | minleafrecs = mp->m_bmap_dmnr[0]; |
93 | minnoderecs = mp->m_bmap_dmnr[1]; | |
94 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
95 | for (level = 1; maxblocks > 1; level++) { | |
96 | if (maxblocks <= maxrootrecs) | |
97 | maxblocks = 1; | |
98 | else | |
99 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
100 | } | |
101 | mp->m_bm_maxlevels[whichfork] = level; | |
102 | } | |
5e656dbb | 103 | |
b194c7d8 BN |
104 | STATIC int /* error */ |
105 | xfs_bmbt_lookup_eq( | |
106 | struct xfs_btree_cur *cur, | |
70a93110 | 107 | struct xfs_bmbt_irec *irec, |
b194c7d8 BN |
108 | int *stat) /* success/failure */ |
109 | { | |
70a93110 | 110 | cur->bc_rec.b = *irec; |
b194c7d8 BN |
111 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); |
112 | } | |
113 | ||
114 | STATIC int /* error */ | |
4f76f49c | 115 | xfs_bmbt_lookup_first( |
b194c7d8 | 116 | struct xfs_btree_cur *cur, |
b194c7d8 BN |
117 | int *stat) /* success/failure */ |
118 | { | |
4f76f49c CH |
119 | cur->bc_rec.b.br_startoff = 0; |
120 | cur->bc_rec.b.br_startblock = 0; | |
121 | cur->bc_rec.b.br_blockcount = 0; | |
b194c7d8 BN |
122 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); |
123 | } | |
124 | ||
125 | /* | |
a2ceac1f DC |
126 | * Check if the inode needs to be converted to btree format. |
127 | */ | |
128 | static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork) | |
129 | { | |
1277a5e0 DW |
130 | return whichfork != XFS_COW_FORK && |
131 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
a2ceac1f DC |
132 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
133 | XFS_IFORK_MAXEXT(ip, whichfork); | |
134 | } | |
135 | ||
136 | /* | |
137 | * Check if the inode should be converted to extent format. | |
138 | */ | |
139 | static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork) | |
140 | { | |
1277a5e0 DW |
141 | return whichfork != XFS_COW_FORK && |
142 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE && | |
a2ceac1f DC |
143 | XFS_IFORK_NEXTENTS(ip, whichfork) <= |
144 | XFS_IFORK_MAXEXT(ip, whichfork); | |
145 | } | |
146 | ||
147 | /* | |
d0e5f1ff | 148 | * Update the record referred to by cur to the value given by irec |
b194c7d8 BN |
149 | * This either works (return 0) or gets an EFSCORRUPTED error. |
150 | */ | |
151 | STATIC int | |
152 | xfs_bmbt_update( | |
153 | struct xfs_btree_cur *cur, | |
d0e5f1ff | 154 | struct xfs_bmbt_irec *irec) |
b194c7d8 BN |
155 | { |
156 | union xfs_btree_rec rec; | |
157 | ||
d0e5f1ff | 158 | xfs_bmbt_disk_set_all(&rec.bmbt, irec); |
b194c7d8 BN |
159 | return xfs_btree_update(cur, &rec); |
160 | } | |
161 | ||
5e656dbb | 162 | /* |
49f693fa DC |
163 | * Compute the worst-case number of indirect blocks that will be used |
164 | * for ip's delayed extent of length "len". | |
5e656dbb | 165 | */ |
49f693fa DC |
166 | STATIC xfs_filblks_t |
167 | xfs_bmap_worst_indlen( | |
168 | xfs_inode_t *ip, /* incore inode pointer */ | |
169 | xfs_filblks_t len) /* delayed extent length */ | |
57c9fccb | 170 | { |
49f693fa DC |
171 | int level; /* btree level number */ |
172 | int maxrecs; /* maximum record count at this level */ | |
173 | xfs_mount_t *mp; /* mount structure */ | |
174 | xfs_filblks_t rval; /* return value */ | |
57c9fccb NS |
175 | |
176 | mp = ip->i_mount; | |
49f693fa DC |
177 | maxrecs = mp->m_bmap_dmxr[0]; |
178 | for (level = 0, rval = 0; | |
179 | level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK); | |
180 | level++) { | |
181 | len += maxrecs - 1; | |
182 | do_div(len, maxrecs); | |
183 | rval += len; | |
7fbe9b54 DW |
184 | if (len == 1) |
185 | return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) - | |
49f693fa DC |
186 | level - 1; |
187 | if (level == 0) | |
188 | maxrecs = mp->m_bmap_dmxr[1]; | |
57c9fccb | 189 | } |
49f693fa | 190 | return rval; |
57c9fccb NS |
191 | } |
192 | ||
193 | /* | |
49f693fa | 194 | * Calculate the default attribute fork offset for newly created inodes. |
57c9fccb | 195 | */ |
49f693fa DC |
196 | uint |
197 | xfs_default_attroffset( | |
198 | struct xfs_inode *ip) | |
57c9fccb | 199 | { |
49f693fa DC |
200 | struct xfs_mount *mp = ip->i_mount; |
201 | uint offset; | |
57c9fccb | 202 | |
49f693fa DC |
203 | if (mp->m_sb.sb_inodesize == 256) { |
204 | offset = XFS_LITINO(mp, ip->i_d.di_version) - | |
205 | XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
206 | } else { | |
207 | offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); | |
57c9fccb | 208 | } |
49f693fa DC |
209 | |
210 | ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version)); | |
211 | return offset; | |
57c9fccb NS |
212 | } |
213 | ||
214 | /* | |
49f693fa DC |
215 | * Helper routine to reset inode di_forkoff field when switching |
216 | * attribute fork from local to extent format - we reset it where | |
217 | * possible to make space available for inline data fork extents. | |
57c9fccb | 218 | */ |
49f693fa DC |
219 | STATIC void |
220 | xfs_bmap_forkoff_reset( | |
49f693fa DC |
221 | xfs_inode_t *ip, |
222 | int whichfork) | |
57c9fccb | 223 | { |
49f693fa DC |
224 | if (whichfork == XFS_ATTR_FORK && |
225 | ip->i_d.di_format != XFS_DINODE_FMT_DEV && | |
49f693fa DC |
226 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE) { |
227 | uint dfl_forkoff = xfs_default_attroffset(ip) >> 3; | |
57c9fccb | 228 | |
49f693fa DC |
229 | if (dfl_forkoff > ip->i_d.di_forkoff) |
230 | ip->i_d.di_forkoff = dfl_forkoff; | |
231 | } | |
57c9fccb NS |
232 | } |
233 | ||
49f693fa DC |
234 | #ifdef DEBUG |
235 | STATIC struct xfs_buf * | |
236 | xfs_bmap_get_bp( | |
237 | struct xfs_btree_cur *cur, | |
238 | xfs_fsblock_t bno) | |
239 | { | |
2fdd378a | 240 | struct xfs_log_item *lip; |
49f693fa | 241 | int i; |
56b2de80 | 242 | |
49f693fa DC |
243 | if (!cur) |
244 | return NULL; | |
2bd0ea18 | 245 | |
49f693fa DC |
246 | for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) { |
247 | if (!cur->bc_bufs[i]) | |
248 | break; | |
249 | if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno) | |
250 | return cur->bc_bufs[i]; | |
251 | } | |
56b2de80 | 252 | |
49f693fa | 253 | /* Chase down all the log items to see if the bp is there */ |
2fdd378a DC |
254 | list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) { |
255 | struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip; | |
256 | ||
49f693fa DC |
257 | if (bip->bli_item.li_type == XFS_LI_BUF && |
258 | XFS_BUF_ADDR(bip->bli_buf) == bno) | |
259 | return bip->bli_buf; | |
260 | } | |
2bd0ea18 | 261 | |
49f693fa DC |
262 | return NULL; |
263 | } | |
56b2de80 | 264 | |
49f693fa DC |
265 | STATIC void |
266 | xfs_check_block( | |
267 | struct xfs_btree_block *block, | |
268 | xfs_mount_t *mp, | |
269 | int root, | |
270 | short sz) | |
271 | { | |
272 | int i, j, dmxr; | |
273 | __be64 *pp, *thispa; /* pointer to block address */ | |
274 | xfs_bmbt_key_t *prevp, *keyp; | |
2bd0ea18 | 275 | |
49f693fa | 276 | ASSERT(be16_to_cpu(block->bb_level) > 0); |
56b2de80 | 277 | |
49f693fa DC |
278 | prevp = NULL; |
279 | for( i = 1; i <= xfs_btree_get_numrecs(block); i++) { | |
280 | dmxr = mp->m_bmap_dmxr[0]; | |
281 | keyp = XFS_BMBT_KEY_ADDR(mp, block, i); | |
a2ceac1f | 282 | |
49f693fa DC |
283 | if (prevp) { |
284 | ASSERT(be64_to_cpu(prevp->br_startoff) < | |
285 | be64_to_cpu(keyp->br_startoff)); | |
286 | } | |
287 | prevp = keyp; | |
2bd0ea18 | 288 | |
2bd0ea18 | 289 | /* |
49f693fa | 290 | * Compare the block numbers to see if there are dups. |
2bd0ea18 | 291 | */ |
49f693fa DC |
292 | if (root) |
293 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz); | |
294 | else | |
295 | pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr); | |
296 | ||
297 | for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) { | |
298 | if (root) | |
299 | thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz); | |
300 | else | |
301 | thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr); | |
302 | if (*thispa == *pp) { | |
303 | xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld", | |
304 | __func__, j, i, | |
305 | (unsigned long long)be64_to_cpu(*thispa)); | |
4d5e2888 | 306 | xfs_err(mp, "%s: ptrs are equal in node\n", |
49f693fa | 307 | __func__); |
4d5e2888 | 308 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
49f693fa | 309 | } |
2bd0ea18 | 310 | } |
49f693fa DC |
311 | } |
312 | } | |
a2ceac1f | 313 | |
49f693fa DC |
314 | /* |
315 | * Check that the extents for the inode ip are in the right order in all | |
f07ae2a6 DC |
316 | * btree leaves. THis becomes prohibitively expensive for large extent count |
317 | * files, so don't bother with inodes that have more than 10,000 extents in | |
318 | * them. The btree record ordering checks will still be done, so for such large | |
319 | * bmapbt constructs that is going to catch most corruptions. | |
49f693fa | 320 | */ |
49f693fa DC |
321 | STATIC void |
322 | xfs_bmap_check_leaf_extents( | |
323 | xfs_btree_cur_t *cur, /* btree cursor or null */ | |
324 | xfs_inode_t *ip, /* incore inode pointer */ | |
325 | int whichfork) /* data or attr fork */ | |
326 | { | |
327 | struct xfs_btree_block *block; /* current btree block */ | |
328 | xfs_fsblock_t bno; /* block # of "block" */ | |
329 | xfs_buf_t *bp; /* buffer for "block" */ | |
330 | int error; /* error return value */ | |
331 | xfs_extnum_t i=0, j; /* index into the extents list */ | |
332 | xfs_ifork_t *ifp; /* fork structure */ | |
333 | int level; /* btree level, for checking */ | |
334 | xfs_mount_t *mp; /* file system mount structure */ | |
335 | __be64 *pp; /* pointer to block address */ | |
336 | xfs_bmbt_rec_t *ep; /* pointer to current extent */ | |
337 | xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */ | |
338 | xfs_bmbt_rec_t *nextp; /* pointer to next extent */ | |
339 | int bp_release = 0; | |
340 | ||
341 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) { | |
342 | return; | |
343 | } | |
344 | ||
f07ae2a6 DC |
345 | /* skip large extent count inodes */ |
346 | if (ip->i_d.di_nextents > 10000) | |
347 | return; | |
348 | ||
49f693fa DC |
349 | bno = NULLFSBLOCK; |
350 | mp = ip->i_mount; | |
351 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
352 | block = ifp->if_broot; | |
353 | /* | |
354 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
355 | */ | |
356 | level = be16_to_cpu(block->bb_level); | |
357 | ASSERT(level > 0); | |
358 | xfs_check_block(block, mp, 1, ifp->if_broot_bytes); | |
359 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
360 | bno = be64_to_cpu(*pp); | |
361 | ||
5a35bf2c | 362 | ASSERT(bno != NULLFSBLOCK); |
49f693fa DC |
363 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
364 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
365 | ||
366 | /* | |
367 | * Go down the tree until leaf level is reached, following the first | |
368 | * pointer (leftmost) at each level. | |
369 | */ | |
370 | while (level-- > 0) { | |
371 | /* See if buf is in cur first */ | |
372 | bp_release = 0; | |
373 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
374 | if (!bp) { | |
375 | bp_release = 1; | |
376 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
377 | XFS_BMAP_BTREE_REF, | |
378 | &xfs_bmbt_buf_ops); | |
2bd0ea18 | 379 | if (error) |
49f693fa | 380 | goto error_norelse; |
2bd0ea18 | 381 | } |
49f693fa | 382 | block = XFS_BUF_TO_BLOCK(bp); |
49f693fa DC |
383 | if (level == 0) |
384 | break; | |
2bd0ea18 | 385 | |
2bd0ea18 | 386 | /* |
49f693fa DC |
387 | * Check this block for basic sanity (increasing keys and |
388 | * no duplicate blocks). | |
2bd0ea18 | 389 | */ |
49f693fa DC |
390 | |
391 | xfs_check_block(block, mp, 0, 0); | |
392 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
393 | bno = be64_to_cpu(*pp); | |
19ebedcf | 394 | XFS_WANT_CORRUPTED_GOTO(mp, |
ecdc52ff | 395 | xfs_verify_fsbno(mp, bno), error0); |
49f693fa DC |
396 | if (bp_release) { |
397 | bp_release = 0; | |
398 | xfs_trans_brelse(NULL, bp); | |
2bd0ea18 | 399 | } |
49f693fa | 400 | } |
a2ceac1f | 401 | |
49f693fa DC |
402 | /* |
403 | * Here with bp and block set to the leftmost leaf node in the tree. | |
404 | */ | |
405 | i = 0; | |
a2ceac1f | 406 | |
49f693fa DC |
407 | /* |
408 | * Loop over all leaf nodes checking that all extents are in the right order. | |
409 | */ | |
410 | for (;;) { | |
411 | xfs_fsblock_t nextbno; | |
412 | xfs_extnum_t num_recs; | |
413 | ||
414 | ||
415 | num_recs = xfs_btree_get_numrecs(block); | |
2bd0ea18 | 416 | |
2bd0ea18 | 417 | /* |
49f693fa | 418 | * Read-ahead the next leaf block, if any. |
2bd0ea18 | 419 | */ |
a2ceac1f | 420 | |
49f693fa | 421 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
a2ceac1f | 422 | |
49f693fa DC |
423 | /* |
424 | * Check all the extents to make sure they are OK. | |
425 | * If we had a previous block, the last entry should | |
426 | * conform with the first entry in this one. | |
427 | */ | |
2bd0ea18 | 428 | |
49f693fa DC |
429 | ep = XFS_BMBT_REC_ADDR(mp, block, 1); |
430 | if (i) { | |
431 | ASSERT(xfs_bmbt_disk_get_startoff(&last) + | |
432 | xfs_bmbt_disk_get_blockcount(&last) <= | |
433 | xfs_bmbt_disk_get_startoff(ep)); | |
434 | } | |
435 | for (j = 1; j < num_recs; j++) { | |
436 | nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1); | |
437 | ASSERT(xfs_bmbt_disk_get_startoff(ep) + | |
438 | xfs_bmbt_disk_get_blockcount(ep) <= | |
439 | xfs_bmbt_disk_get_startoff(nextp)); | |
440 | ep = nextp; | |
441 | } | |
442 | ||
443 | last = *ep; | |
444 | i += num_recs; | |
445 | if (bp_release) { | |
446 | bp_release = 0; | |
447 | xfs_trans_brelse(NULL, bp); | |
448 | } | |
449 | bno = nextbno; | |
2bd0ea18 | 450 | /* |
49f693fa | 451 | * If we've reached the end, stop. |
2bd0ea18 | 452 | */ |
49f693fa DC |
453 | if (bno == NULLFSBLOCK) |
454 | break; | |
a2ceac1f | 455 | |
49f693fa DC |
456 | bp_release = 0; |
457 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
458 | if (!bp) { | |
459 | bp_release = 1; | |
460 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
461 | XFS_BMAP_BTREE_REF, | |
462 | &xfs_bmbt_buf_ops); | |
a2ceac1f | 463 | if (error) |
49f693fa | 464 | goto error_norelse; |
2bd0ea18 | 465 | } |
49f693fa | 466 | block = XFS_BUF_TO_BLOCK(bp); |
a2ceac1f | 467 | } |
4d4a192c | 468 | |
49f693fa | 469 | return; |
a2ceac1f | 470 | |
49f693fa DC |
471 | error0: |
472 | xfs_warn(mp, "%s: at error0", __func__); | |
473 | if (bp_release) | |
474 | xfs_trans_brelse(NULL, bp); | |
475 | error_norelse: | |
476 | xfs_warn(mp, "%s: BAD after btree leaves for %d extents", | |
477 | __func__, i); | |
4d5e2888 DW |
478 | xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__); |
479 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
49f693fa | 480 | return; |
2bd0ea18 NS |
481 | } |
482 | ||
49f693fa DC |
483 | /* |
484 | * Validate that the bmbt_irecs being returned from bmapi are valid | |
e6d77a21 DC |
485 | * given the caller's original parameters. Specifically check the |
486 | * ranges of the returned irecs to ensure that they only extend beyond | |
49f693fa DC |
487 | * the given parameters if the XFS_BMAPI_ENTIRE flag was set. |
488 | */ | |
489 | STATIC void | |
490 | xfs_bmap_validate_ret( | |
491 | xfs_fileoff_t bno, | |
492 | xfs_filblks_t len, | |
493 | int flags, | |
494 | xfs_bmbt_irec_t *mval, | |
495 | int nmap, | |
496 | int ret_nmap) | |
497 | { | |
498 | int i; /* index to map values */ | |
a2ceac1f | 499 | |
49f693fa | 500 | ASSERT(ret_nmap <= nmap); |
a2ceac1f | 501 | |
49f693fa DC |
502 | for (i = 0; i < ret_nmap; i++) { |
503 | ASSERT(mval[i].br_blockcount > 0); | |
504 | if (!(flags & XFS_BMAPI_ENTIRE)) { | |
505 | ASSERT(mval[i].br_startoff >= bno); | |
506 | ASSERT(mval[i].br_blockcount <= len); | |
507 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount <= | |
508 | bno + len); | |
509 | } else { | |
510 | ASSERT(mval[i].br_startoff < bno + len); | |
511 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount > | |
512 | bno); | |
513 | } | |
514 | ASSERT(i == 0 || | |
515 | mval[i - 1].br_startoff + mval[i - 1].br_blockcount == | |
516 | mval[i].br_startoff); | |
517 | ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK && | |
518 | mval[i].br_startblock != HOLESTARTBLOCK); | |
519 | ASSERT(mval[i].br_state == XFS_EXT_NORM || | |
520 | mval[i].br_state == XFS_EXT_UNWRITTEN); | |
521 | } | |
522 | } | |
56b2de80 | 523 | |
49f693fa DC |
524 | #else |
525 | #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0) | |
9587a34c | 526 | #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0) |
49f693fa | 527 | #endif /* DEBUG */ |
56b2de80 | 528 | |
49f693fa DC |
529 | /* |
530 | * bmap free list manipulation functions | |
531 | */ | |
56b2de80 | 532 | |
49f693fa DC |
533 | /* |
534 | * Add the extent to the list of extents to be free at transaction end. | |
535 | * The list is maintained sorted (by block number). | |
536 | */ | |
537 | void | |
538 | xfs_bmap_add_free( | |
85aec44f DW |
539 | struct xfs_mount *mp, |
540 | struct xfs_defer_ops *dfops, | |
541 | xfs_fsblock_t bno, | |
542 | xfs_filblks_t len, | |
543 | struct xfs_owner_info *oinfo) | |
49f693fa | 544 | { |
6f530e9a | 545 | struct xfs_extent_free_item *new; /* new element */ |
49f693fa DC |
546 | #ifdef DEBUG |
547 | xfs_agnumber_t agno; | |
548 | xfs_agblock_t agbno; | |
56b2de80 | 549 | |
49f693fa DC |
550 | ASSERT(bno != NULLFSBLOCK); |
551 | ASSERT(len > 0); | |
552 | ASSERT(len <= MAXEXTLEN); | |
553 | ASSERT(!isnullstartblock(bno)); | |
554 | agno = XFS_FSB_TO_AGNO(mp, bno); | |
555 | agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
556 | ASSERT(agno < mp->m_sb.sb_agcount); | |
557 | ASSERT(agbno < mp->m_sb.sb_agblocks); | |
558 | ASSERT(len < mp->m_sb.sb_agblocks); | |
559 | ASSERT(agbno + len <= mp->m_sb.sb_agblocks); | |
560 | #endif | |
561 | ASSERT(xfs_bmap_free_item_zone != NULL); | |
85aec44f | 562 | |
49f693fa | 563 | new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP); |
6f530e9a DW |
564 | new->xefi_startblock = bno; |
565 | new->xefi_blockcount = (xfs_extlen_t)len; | |
85aec44f DW |
566 | if (oinfo) |
567 | new->xefi_oinfo = *oinfo; | |
568 | else | |
569 | xfs_rmap_skip_owner_update(&new->xefi_oinfo); | |
a9da40de DW |
570 | trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0, |
571 | XFS_FSB_TO_AGBNO(mp, bno), len); | |
f33cea1a | 572 | xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list); |
49f693fa | 573 | } |
2bd0ea18 | 574 | |
49f693fa DC |
575 | /* |
576 | * Inode fork format manipulation functions | |
577 | */ | |
a2ceac1f | 578 | |
49f693fa DC |
579 | /* |
580 | * Transform a btree format file with only one leaf node, where the | |
581 | * extents list will fit in the inode, into an extents format file. | |
582 | * Since the file extents are already in-core, all we have to do is | |
583 | * give up the space for the btree root and pitch the leaf block. | |
584 | */ | |
585 | STATIC int /* error */ | |
586 | xfs_bmap_btree_to_extents( | |
587 | xfs_trans_t *tp, /* transaction pointer */ | |
588 | xfs_inode_t *ip, /* incore inode pointer */ | |
589 | xfs_btree_cur_t *cur, /* btree cursor */ | |
590 | int *logflagsp, /* inode logging flags */ | |
591 | int whichfork) /* data or attr fork */ | |
592 | { | |
593 | /* REFERENCED */ | |
594 | struct xfs_btree_block *cblock;/* child btree block */ | |
595 | xfs_fsblock_t cbno; /* child block number */ | |
596 | xfs_buf_t *cbp; /* child block's buffer */ | |
597 | int error; /* error return value */ | |
598 | xfs_ifork_t *ifp; /* inode fork data */ | |
599 | xfs_mount_t *mp; /* mount point structure */ | |
600 | __be64 *pp; /* ptr to block address */ | |
601 | struct xfs_btree_block *rblock;/* root btree block */ | |
85aec44f | 602 | struct xfs_owner_info oinfo; |
56b2de80 | 603 | |
49f693fa DC |
604 | mp = ip->i_mount; |
605 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1277a5e0 | 606 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa DC |
607 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
608 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
609 | rblock = ifp->if_broot; | |
610 | ASSERT(be16_to_cpu(rblock->bb_level) == 1); | |
611 | ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1); | |
612 | ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1); | |
613 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes); | |
614 | cbno = be64_to_cpu(*pp); | |
615 | *logflagsp = 0; | |
616 | #ifdef DEBUG | |
d5c546df DW |
617 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, |
618 | xfs_btree_check_lptr(cur, cbno, 1)); | |
49f693fa DC |
619 | #endif |
620 | error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF, | |
621 | &xfs_bmbt_buf_ops); | |
622 | if (error) | |
623 | return error; | |
624 | cblock = XFS_BUF_TO_BLOCK(cbp); | |
625 | if ((error = xfs_btree_check_block(cur, cblock, 0, cbp))) | |
626 | return error; | |
85aec44f DW |
627 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); |
628 | xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo); | |
49f693fa DC |
629 | ip->i_d.di_nblocks--; |
630 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); | |
631 | xfs_trans_binval(tp, cbp); | |
632 | if (cur->bc_bufs[0] == cbp) | |
633 | cur->bc_bufs[0] = NULL; | |
634 | xfs_iroot_realloc(ip, -1, whichfork); | |
635 | ASSERT(ifp->if_broot == NULL); | |
636 | ASSERT((ifp->if_flags & XFS_IFBROOT) == 0); | |
637 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
638 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); | |
639 | return 0; | |
640 | } | |
2bd0ea18 NS |
641 | |
642 | /* | |
49f693fa DC |
643 | * Convert an extents-format file into a btree-format file. |
644 | * The new file will have a root block (in the inode) and a single child block. | |
2bd0ea18 | 645 | */ |
49f693fa DC |
646 | STATIC int /* error */ |
647 | xfs_bmap_extents_to_btree( | |
648 | xfs_trans_t *tp, /* transaction pointer */ | |
649 | xfs_inode_t *ip, /* incore inode pointer */ | |
650 | xfs_fsblock_t *firstblock, /* first-block-allocated */ | |
f33cea1a | 651 | struct xfs_defer_ops *dfops, /* blocks freed in xaction */ |
49f693fa DC |
652 | xfs_btree_cur_t **curp, /* cursor returned to caller */ |
653 | int wasdel, /* converting a delayed alloc */ | |
654 | int *logflagsp, /* inode logging flags */ | |
655 | int whichfork) /* data or attr fork */ | |
2bd0ea18 | 656 | { |
49f693fa DC |
657 | struct xfs_btree_block *ablock; /* allocated (child) bt block */ |
658 | xfs_buf_t *abp; /* buffer for ablock */ | |
659 | xfs_alloc_arg_t args; /* allocation arguments */ | |
660 | xfs_bmbt_rec_t *arp; /* child record pointer */ | |
661 | struct xfs_btree_block *block; /* btree root block */ | |
662 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
49f693fa | 663 | int error; /* error return value */ |
49f693fa DC |
664 | xfs_ifork_t *ifp; /* inode fork pointer */ |
665 | xfs_bmbt_key_t *kp; /* root block key pointer */ | |
666 | xfs_mount_t *mp; /* mount structure */ | |
49f693fa | 667 | xfs_bmbt_ptr_t *pp; /* root block address pointer */ |
9788e059 | 668 | struct xfs_iext_cursor icur; |
62ed7338 | 669 | struct xfs_bmbt_irec rec; |
9788e059 | 670 | xfs_extnum_t cnt = 0; |
2bd0ea18 | 671 | |
5dfa5cd2 | 672 | mp = ip->i_mount; |
1277a5e0 | 673 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa DC |
674 | ifp = XFS_IFORK_PTR(ip, whichfork); |
675 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS); | |
56b2de80 | 676 | |
2bd0ea18 | 677 | /* |
49f693fa | 678 | * Make space in the inode incore. |
2bd0ea18 | 679 | */ |
49f693fa DC |
680 | xfs_iroot_realloc(ip, 1, whichfork); |
681 | ifp->if_flags |= XFS_IFBROOT; | |
56b2de80 | 682 | |
2bd0ea18 | 683 | /* |
49f693fa | 684 | * Fill in the root. |
2bd0ea18 | 685 | */ |
49f693fa | 686 | block = ifp->if_broot; |
e394a4b1 ES |
687 | xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL, |
688 | XFS_BTNUM_BMAP, 1, 1, ip->i_ino, | |
f4241a08 | 689 | XFS_BTREE_LONG_PTRS); |
49f693fa DC |
690 | /* |
691 | * Need a cursor. Can't allocate until bb_level is filled in. | |
692 | */ | |
49f693fa DC |
693 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
694 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 695 | cur->bc_private.b.dfops = dfops; |
49f693fa DC |
696 | cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; |
697 | /* | |
698 | * Convert to a btree with two levels, one record in root. | |
699 | */ | |
700 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE); | |
701 | memset(&args, 0, sizeof(args)); | |
702 | args.tp = tp; | |
703 | args.mp = mp; | |
85aec44f | 704 | xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork); |
49f693fa DC |
705 | args.firstblock = *firstblock; |
706 | if (*firstblock == NULLFSBLOCK) { | |
707 | args.type = XFS_ALLOCTYPE_START_BNO; | |
708 | args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino); | |
f33cea1a | 709 | } else if (dfops->dop_low) { |
49f693fa DC |
710 | args.type = XFS_ALLOCTYPE_START_BNO; |
711 | args.fsbno = *firstblock; | |
712 | } else { | |
713 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
714 | args.fsbno = *firstblock; | |
715 | } | |
716 | args.minlen = args.maxlen = args.prod = 1; | |
717 | args.wasdel = wasdel; | |
718 | *logflagsp = 0; | |
719 | if ((error = xfs_alloc_vextent(&args))) { | |
720 | xfs_iroot_realloc(ip, -1, whichfork); | |
7c4d311d ES |
721 | ASSERT(ifp->if_broot == NULL); |
722 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
49f693fa DC |
723 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
724 | return error; | |
2bd0ea18 | 725 | } |
68d06dce | 726 | |
c1587ecf CH |
727 | if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) { |
728 | xfs_iroot_realloc(ip, -1, whichfork); | |
7c4d311d ES |
729 | ASSERT(ifp->if_broot == NULL); |
730 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
c1587ecf CH |
731 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
732 | return -ENOSPC; | |
733 | } | |
2bd0ea18 | 734 | /* |
49f693fa | 735 | * Allocation can't fail, the space was reserved. |
2bd0ea18 | 736 | */ |
49f693fa | 737 | ASSERT(*firstblock == NULLFSBLOCK || |
6df6a3ab | 738 | args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock)); |
49f693fa DC |
739 | *firstblock = cur->bc_private.b.firstblock = args.fsbno; |
740 | cur->bc_private.b.allocated++; | |
741 | ip->i_d.di_nblocks++; | |
742 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L); | |
743 | abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0); | |
2bd0ea18 | 744 | /* |
49f693fa | 745 | * Fill in the child block. |
2bd0ea18 | 746 | */ |
49f693fa DC |
747 | abp->b_ops = &xfs_bmbt_buf_ops; |
748 | ablock = XFS_BUF_TO_BLOCK(abp); | |
e394a4b1 ES |
749 | xfs_btree_init_block_int(mp, ablock, abp->b_bn, |
750 | XFS_BTNUM_BMAP, 0, 0, ip->i_ino, | |
5dfa5cd2 DC |
751 | XFS_BTREE_LONG_PTRS); |
752 | ||
9788e059 | 753 | for_each_xfs_iext(ifp, &icur, &rec) { |
62ed7338 CH |
754 | if (isnullstartblock(rec.br_startblock)) |
755 | continue; | |
756 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt); | |
757 | xfs_bmbt_disk_set_all(arp, &rec); | |
758 | cnt++; | |
49f693fa DC |
759 | } |
760 | ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork)); | |
761 | xfs_btree_set_numrecs(ablock, cnt); | |
56b2de80 | 762 | |
49f693fa DC |
763 | /* |
764 | * Fill in the root key and pointer. | |
765 | */ | |
766 | kp = XFS_BMBT_KEY_ADDR(mp, block, 1); | |
767 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); | |
768 | kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp)); | |
769 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur, | |
770 | be16_to_cpu(block->bb_level))); | |
771 | *pp = cpu_to_be64(args.fsbno); | |
2bd0ea18 | 772 | |
49f693fa DC |
773 | /* |
774 | * Do all this logging at the end so that | |
775 | * the root is at the right level. | |
776 | */ | |
5dfa5cd2 | 777 | xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS); |
613e6057 | 778 | xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs)); |
49f693fa DC |
779 | ASSERT(*curp == NULL); |
780 | *curp = cur; | |
781 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork); | |
782 | return 0; | |
783 | } | |
a2ceac1f | 784 | |
49f693fa DC |
785 | /* |
786 | * Convert a local file to an extents file. | |
787 | * This code is out of bounds for data forks of regular files, | |
788 | * since the file data needs to get logged so things will stay consistent. | |
789 | * (The bmap-level manipulations are ok, though). | |
790 | */ | |
3f17ed4b DC |
791 | void |
792 | xfs_bmap_local_to_extents_empty( | |
793 | struct xfs_inode *ip, | |
794 | int whichfork) | |
795 | { | |
796 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
797 | ||
1277a5e0 | 798 | ASSERT(whichfork != XFS_COW_FORK); |
3f17ed4b DC |
799 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); |
800 | ASSERT(ifp->if_bytes == 0); | |
801 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0); | |
802 | ||
ff105f75 | 803 | xfs_bmap_forkoff_reset(ip, whichfork); |
3f17ed4b DC |
804 | ifp->if_flags &= ~XFS_IFINLINE; |
805 | ifp->if_flags |= XFS_IFEXTENTS; | |
b37d753d CH |
806 | ifp->if_u1.if_root = NULL; |
807 | ifp->if_height = 0; | |
3f17ed4b DC |
808 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); |
809 | } | |
810 | ||
811 | ||
49f693fa DC |
812 | STATIC int /* error */ |
813 | xfs_bmap_local_to_extents( | |
814 | xfs_trans_t *tp, /* transaction pointer */ | |
815 | xfs_inode_t *ip, /* incore inode pointer */ | |
816 | xfs_fsblock_t *firstblock, /* first block allocated in xaction */ | |
817 | xfs_extlen_t total, /* total blocks needed by transaction */ | |
818 | int *logflagsp, /* inode logging flags */ | |
819 | int whichfork, | |
5dfa5cd2 DC |
820 | void (*init_fn)(struct xfs_trans *tp, |
821 | struct xfs_buf *bp, | |
49f693fa DC |
822 | struct xfs_inode *ip, |
823 | struct xfs_ifork *ifp)) | |
824 | { | |
3f17ed4b | 825 | int error = 0; |
49f693fa DC |
826 | int flags; /* logging flags returned */ |
827 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
3f17ed4b DC |
828 | xfs_alloc_arg_t args; /* allocation arguments */ |
829 | xfs_buf_t *bp; /* buffer for extent block */ | |
05104e43 | 830 | struct xfs_bmbt_irec rec; |
9788e059 | 831 | struct xfs_iext_cursor icur; |
2bd0ea18 | 832 | |
49f693fa DC |
833 | /* |
834 | * We don't want to deal with the case of keeping inode data inline yet. | |
835 | * So sending the data fork of a regular inode is invalid. | |
836 | */ | |
e37bf53c | 837 | ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK)); |
49f693fa DC |
838 | ifp = XFS_IFORK_PTR(ip, whichfork); |
839 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
3f17ed4b DC |
840 | |
841 | if (!ifp->if_bytes) { | |
842 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
843 | flags = XFS_ILOG_CORE; | |
844 | goto done; | |
845 | } | |
846 | ||
49f693fa DC |
847 | flags = 0; |
848 | error = 0; | |
b37d753d | 849 | ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE); |
3f17ed4b DC |
850 | memset(&args, 0, sizeof(args)); |
851 | args.tp = tp; | |
852 | args.mp = ip->i_mount; | |
85aec44f | 853 | xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0); |
3f17ed4b DC |
854 | args.firstblock = *firstblock; |
855 | /* | |
856 | * Allocate a block. We know we need only one, since the | |
857 | * file currently fits in an inode. | |
858 | */ | |
859 | if (*firstblock == NULLFSBLOCK) { | |
860 | args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino); | |
861 | args.type = XFS_ALLOCTYPE_START_BNO; | |
49f693fa | 862 | } else { |
3f17ed4b DC |
863 | args.fsbno = *firstblock; |
864 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
2bd0ea18 | 865 | } |
3f17ed4b DC |
866 | args.total = total; |
867 | args.minlen = args.maxlen = args.prod = 1; | |
868 | error = xfs_alloc_vextent(&args); | |
869 | if (error) | |
870 | goto done; | |
871 | ||
872 | /* Can't fail, the space was reserved. */ | |
873 | ASSERT(args.fsbno != NULLFSBLOCK); | |
874 | ASSERT(args.len == 1); | |
875 | *firstblock = args.fsbno; | |
876 | bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0); | |
877 | ||
19ebedcf | 878 | /* |
f44fbde0 | 879 | * Initialize the block, copy the data and log the remote buffer. |
19ebedcf | 880 | * |
f44fbde0 BF |
881 | * The callout is responsible for logging because the remote format |
882 | * might differ from the local format and thus we don't know how much to | |
883 | * log here. Note that init_fn must also set the buffer log item type | |
884 | * correctly. | |
19ebedcf | 885 | */ |
3f17ed4b DC |
886 | init_fn(tp, bp, ip, ifp); |
887 | ||
f44fbde0 | 888 | /* account for the change in fork size */ |
3f17ed4b DC |
889 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); |
890 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
49f693fa | 891 | flags |= XFS_ILOG_CORE; |
3f17ed4b | 892 | |
b37d753d CH |
893 | ifp->if_u1.if_root = NULL; |
894 | ifp->if_height = 0; | |
895 | ||
05104e43 CH |
896 | rec.br_startoff = 0; |
897 | rec.br_startblock = args.fsbno; | |
898 | rec.br_blockcount = 1; | |
899 | rec.br_state = XFS_EXT_NORM; | |
9788e059 | 900 | xfs_iext_first(ifp, &icur); |
26a75f67 | 901 | xfs_iext_insert(ip, &icur, &rec, 0); |
05104e43 | 902 | |
3f17ed4b DC |
903 | XFS_IFORK_NEXT_SET(ip, whichfork, 1); |
904 | ip->i_d.di_nblocks = 1; | |
905 | xfs_trans_mod_dquot_byino(tp, ip, | |
906 | XFS_TRANS_DQ_BCOUNT, 1L); | |
907 | flags |= xfs_ilog_fext(whichfork); | |
908 | ||
49f693fa DC |
909 | done: |
910 | *logflagsp = flags; | |
911 | return error; | |
2bd0ea18 NS |
912 | } |
913 | ||
914 | /* | |
49f693fa | 915 | * Called from xfs_bmap_add_attrfork to handle btree format files. |
2bd0ea18 | 916 | */ |
49f693fa DC |
917 | STATIC int /* error */ |
918 | xfs_bmap_add_attrfork_btree( | |
919 | xfs_trans_t *tp, /* transaction pointer */ | |
920 | xfs_inode_t *ip, /* incore inode pointer */ | |
921 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 922 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa | 923 | int *flags) /* inode logging flags */ |
2bd0ea18 | 924 | { |
49f693fa DC |
925 | xfs_btree_cur_t *cur; /* btree cursor */ |
926 | int error; /* error return value */ | |
927 | xfs_mount_t *mp; /* file system mount struct */ | |
928 | int stat; /* newroot status */ | |
56b2de80 | 929 | |
49f693fa DC |
930 | mp = ip->i_mount; |
931 | if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip)) | |
932 | *flags |= XFS_ILOG_DBROOT; | |
933 | else { | |
934 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
f33cea1a | 935 | cur->bc_private.b.dfops = dfops; |
49f693fa | 936 | cur->bc_private.b.firstblock = *firstblock; |
4f76f49c CH |
937 | error = xfs_bmbt_lookup_first(cur, &stat); |
938 | if (error) | |
49f693fa DC |
939 | goto error0; |
940 | /* must be at least one entry */ | |
19ebedcf | 941 | XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0); |
49f693fa DC |
942 | if ((error = xfs_btree_new_iroot(cur, flags, &stat))) |
943 | goto error0; | |
944 | if (stat == 0) { | |
945 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
12b53197 | 946 | return -ENOSPC; |
49f693fa DC |
947 | } |
948 | *firstblock = cur->bc_private.b.firstblock; | |
949 | cur->bc_private.b.allocated = 0; | |
950 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2bd0ea18 | 951 | } |
49f693fa DC |
952 | return 0; |
953 | error0: | |
954 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
955 | return error; | |
956 | } | |
56b2de80 | 957 | |
49f693fa DC |
958 | /* |
959 | * Called from xfs_bmap_add_attrfork to handle extents format files. | |
960 | */ | |
961 | STATIC int /* error */ | |
962 | xfs_bmap_add_attrfork_extents( | |
963 | xfs_trans_t *tp, /* transaction pointer */ | |
964 | xfs_inode_t *ip, /* incore inode pointer */ | |
965 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 966 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa DC |
967 | int *flags) /* inode logging flags */ |
968 | { | |
969 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
970 | int error; /* error return value */ | |
971 | ||
972 | if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip)) | |
973 | return 0; | |
974 | cur = NULL; | |
f33cea1a | 975 | error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0, |
49f693fa DC |
976 | flags, XFS_DATA_FORK); |
977 | if (cur) { | |
978 | cur->bc_private.b.allocated = 0; | |
979 | xfs_btree_del_cursor(cur, | |
980 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
2bd0ea18 | 981 | } |
49f693fa DC |
982 | return error; |
983 | } | |
56b2de80 | 984 | |
49f693fa DC |
985 | /* |
986 | * Called from xfs_bmap_add_attrfork to handle local format files. Each | |
987 | * different data fork content type needs a different callout to do the | |
988 | * conversion. Some are basic and only require special block initialisation | |
989 | * callouts for the data formating, others (directories) are so specialised they | |
990 | * handle everything themselves. | |
991 | * | |
992 | * XXX (dgc): investigate whether directory conversion can use the generic | |
993 | * formatting callout. It should be possible - it's just a very complex | |
5dfa5cd2 | 994 | * formatter. |
49f693fa DC |
995 | */ |
996 | STATIC int /* error */ | |
997 | xfs_bmap_add_attrfork_local( | |
998 | xfs_trans_t *tp, /* transaction pointer */ | |
999 | xfs_inode_t *ip, /* incore inode pointer */ | |
1000 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 1001 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa DC |
1002 | int *flags) /* inode logging flags */ |
1003 | { | |
1004 | xfs_da_args_t dargs; /* args for dir/attr code */ | |
56b2de80 | 1005 | |
49f693fa DC |
1006 | if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip)) |
1007 | return 0; | |
a2ceac1f | 1008 | |
e37bf53c | 1009 | if (S_ISDIR(VFS_I(ip)->i_mode)) { |
49f693fa | 1010 | memset(&dargs, 0, sizeof(dargs)); |
ff105f75 | 1011 | dargs.geo = ip->i_mount->m_dir_geo; |
49f693fa DC |
1012 | dargs.dp = ip; |
1013 | dargs.firstblock = firstblock; | |
f33cea1a | 1014 | dargs.dfops = dfops; |
ff105f75 | 1015 | dargs.total = dargs.geo->fsbcount; |
49f693fa DC |
1016 | dargs.whichfork = XFS_DATA_FORK; |
1017 | dargs.trans = tp; | |
1018 | return xfs_dir2_sf_to_block(&dargs); | |
1019 | } | |
2bd0ea18 | 1020 | |
e37bf53c | 1021 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
49f693fa DC |
1022 | return xfs_bmap_local_to_extents(tp, ip, firstblock, 1, |
1023 | flags, XFS_DATA_FORK, | |
1024 | xfs_symlink_local_to_remote); | |
56b2de80 | 1025 | |
3f17ed4b DC |
1026 | /* should only be called for types that support local format data */ |
1027 | ASSERT(0); | |
12b53197 | 1028 | return -EFSCORRUPTED; |
49f693fa | 1029 | } |
2bd0ea18 | 1030 | |
49f693fa DC |
1031 | /* |
1032 | * Convert inode from non-attributed to attributed. | |
1033 | * Must not be in a transaction, ip must not be locked. | |
1034 | */ | |
1035 | int /* error code */ | |
1036 | xfs_bmap_add_attrfork( | |
1037 | xfs_inode_t *ip, /* incore inode pointer */ | |
1038 | int size, /* space new attribute needs */ | |
1039 | int rsvd) /* xact may use reserved blks */ | |
1040 | { | |
1041 | xfs_fsblock_t firstblock; /* 1st block/ag allocated */ | |
f33cea1a | 1042 | struct xfs_defer_ops dfops; /* freed extent records */ |
49f693fa DC |
1043 | xfs_mount_t *mp; /* mount structure */ |
1044 | xfs_trans_t *tp; /* transaction pointer */ | |
1045 | int blks; /* space reservation */ | |
1046 | int version = 1; /* superblock attr version */ | |
49f693fa DC |
1047 | int logflags; /* logging flags */ |
1048 | int error; /* error return value */ | |
56b2de80 | 1049 | |
49f693fa | 1050 | ASSERT(XFS_IFORK_Q(ip) == 0); |
2bd0ea18 | 1051 | |
49f693fa DC |
1052 | mp = ip->i_mount; |
1053 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); | |
9074815c | 1054 | |
49f693fa | 1055 | blks = XFS_ADDAFORK_SPACE_RES(mp); |
9074815c CH |
1056 | |
1057 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0, | |
1058 | rsvd ? XFS_TRANS_RESERVE : 0, &tp); | |
1059 | if (error) | |
ff105f75 | 1060 | return error; |
9074815c | 1061 | |
49f693fa DC |
1062 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1063 | error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ? | |
1064 | XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES : | |
1065 | XFS_QMOPT_RES_REGBLKS); | |
ff105f75 DC |
1066 | if (error) |
1067 | goto trans_cancel; | |
49f693fa | 1068 | if (XFS_IFORK_Q(ip)) |
ff105f75 | 1069 | goto trans_cancel; |
9656ed10 DW |
1070 | if (ip->i_d.di_anextents != 0) { |
1071 | error = -EFSCORRUPTED; | |
1072 | goto trans_cancel; | |
1073 | } | |
49f693fa | 1074 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) { |
2bd0ea18 | 1075 | /* |
49f693fa | 1076 | * For inodes coming from pre-6.2 filesystems. |
2bd0ea18 | 1077 | */ |
49f693fa DC |
1078 | ASSERT(ip->i_d.di_aformat == 0); |
1079 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
5e656dbb | 1080 | } |
a2ceac1f | 1081 | |
ff105f75 | 1082 | xfs_trans_ijoin(tp, ip, 0); |
49f693fa | 1083 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
a2ceac1f | 1084 | |
49f693fa DC |
1085 | switch (ip->i_d.di_format) { |
1086 | case XFS_DINODE_FMT_DEV: | |
1087 | ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; | |
1088 | break; | |
49f693fa DC |
1089 | case XFS_DINODE_FMT_LOCAL: |
1090 | case XFS_DINODE_FMT_EXTENTS: | |
1091 | case XFS_DINODE_FMT_BTREE: | |
1092 | ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size); | |
1093 | if (!ip->i_d.di_forkoff) | |
1094 | ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3; | |
1095 | else if (mp->m_flags & XFS_MOUNT_ATTR2) | |
1096 | version = 2; | |
1097 | break; | |
1098 | default: | |
1099 | ASSERT(0); | |
12b53197 | 1100 | error = -EINVAL; |
ff105f75 | 1101 | goto trans_cancel; |
a2ceac1f DC |
1102 | } |
1103 | ||
49f693fa DC |
1104 | ASSERT(ip->i_afp == NULL); |
1105 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
1106 | ip->i_afp->if_flags = XFS_IFEXTENTS; | |
1107 | logflags = 0; | |
f33cea1a | 1108 | xfs_defer_init(&dfops, &firstblock); |
49f693fa DC |
1109 | switch (ip->i_d.di_format) { |
1110 | case XFS_DINODE_FMT_LOCAL: | |
f33cea1a | 1111 | error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops, |
49f693fa DC |
1112 | &logflags); |
1113 | break; | |
1114 | case XFS_DINODE_FMT_EXTENTS: | |
1115 | error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock, | |
f33cea1a | 1116 | &dfops, &logflags); |
49f693fa DC |
1117 | break; |
1118 | case XFS_DINODE_FMT_BTREE: | |
f33cea1a | 1119 | error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops, |
49f693fa DC |
1120 | &logflags); |
1121 | break; | |
1122 | default: | |
1123 | error = 0; | |
1124 | break; | |
1125 | } | |
1126 | if (logflags) | |
1127 | xfs_trans_log_inode(tp, ip, logflags); | |
1128 | if (error) | |
ff105f75 | 1129 | goto bmap_cancel; |
49f693fa DC |
1130 | if (!xfs_sb_version_hasattr(&mp->m_sb) || |
1131 | (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) { | |
19ebedcf | 1132 | bool log_sb = false; |
a2ceac1f | 1133 | |
49f693fa DC |
1134 | spin_lock(&mp->m_sb_lock); |
1135 | if (!xfs_sb_version_hasattr(&mp->m_sb)) { | |
1136 | xfs_sb_version_addattr(&mp->m_sb); | |
19ebedcf | 1137 | log_sb = true; |
49f693fa DC |
1138 | } |
1139 | if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) { | |
1140 | xfs_sb_version_addattr2(&mp->m_sb); | |
19ebedcf | 1141 | log_sb = true; |
49f693fa | 1142 | } |
19ebedcf DC |
1143 | spin_unlock(&mp->m_sb_lock); |
1144 | if (log_sb) | |
1145 | xfs_log_sb(tp); | |
49f693fa DC |
1146 | } |
1147 | ||
5c33baee | 1148 | error = xfs_defer_finish(&tp, &dfops); |
49f693fa | 1149 | if (error) |
ff105f75 | 1150 | goto bmap_cancel; |
de5a3f46 | 1151 | error = xfs_trans_commit(tp); |
ff105f75 DC |
1152 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1153 | return error; | |
1154 | ||
1155 | bmap_cancel: | |
f33cea1a | 1156 | xfs_defer_cancel(&dfops); |
ff105f75 | 1157 | trans_cancel: |
3d7434fe | 1158 | xfs_trans_cancel(tp); |
49f693fa | 1159 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5e656dbb | 1160 | return error; |
2bd0ea18 NS |
1161 | } |
1162 | ||
399ab595 | 1163 | /* |
49f693fa DC |
1164 | * Internal and external extent tree search functions. |
1165 | */ | |
399ab595 | 1166 | |
49f693fa | 1167 | /* |
6d79c95c | 1168 | * Read in extents from a btree-format inode. |
49f693fa | 1169 | */ |
6d79c95c CH |
1170 | int |
1171 | xfs_iread_extents( | |
1172 | struct xfs_trans *tp, | |
1173 | struct xfs_inode *ip, | |
1174 | int whichfork) | |
49f693fa | 1175 | { |
6d79c95c | 1176 | struct xfs_mount *mp = ip->i_mount; |
322fd804 | 1177 | int state = xfs_bmap_fork_to_state(whichfork); |
6d79c95c CH |
1178 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1179 | xfs_extnum_t nextents = XFS_IFORK_NEXTENTS(ip, whichfork); | |
1180 | struct xfs_btree_block *block = ifp->if_broot; | |
9788e059 | 1181 | struct xfs_iext_cursor icur; |
b37d753d | 1182 | struct xfs_bmbt_irec new; |
6d79c95c CH |
1183 | xfs_fsblock_t bno; |
1184 | struct xfs_buf *bp; | |
1185 | xfs_extnum_t i, j; | |
1186 | int level; | |
1187 | __be64 *pp; | |
1188 | int error; | |
1189 | ||
1190 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
1191 | ||
1192 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1193 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
1194 | return -EFSCORRUPTED; | |
1195 | } | |
1196 | ||
399ab595 | 1197 | /* |
49f693fa | 1198 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. |
399ab595 | 1199 | */ |
49f693fa DC |
1200 | level = be16_to_cpu(block->bb_level); |
1201 | ASSERT(level > 0); | |
1202 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
1203 | bno = be64_to_cpu(*pp); | |
e2376544 | 1204 | |
399ab595 | 1205 | /* |
49f693fa DC |
1206 | * Go down the tree until leaf level is reached, following the first |
1207 | * pointer (leftmost) at each level. | |
399ab595 | 1208 | */ |
49f693fa DC |
1209 | while (level-- > 0) { |
1210 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1211 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1212 | if (error) | |
6d79c95c | 1213 | goto out; |
49f693fa | 1214 | block = XFS_BUF_TO_BLOCK(bp); |
49f693fa DC |
1215 | if (level == 0) |
1216 | break; | |
1217 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
1218 | bno = be64_to_cpu(*pp); | |
19ebedcf | 1219 | XFS_WANT_CORRUPTED_GOTO(mp, |
ecdc52ff | 1220 | xfs_verify_fsbno(mp, bno), out_brelse); |
49f693fa | 1221 | xfs_trans_brelse(tp, bp); |
399ab595 | 1222 | } |
6d79c95c | 1223 | |
399ab595 | 1224 | /* |
49f693fa | 1225 | * Here with bp and block set to the leftmost leaf node in the tree. |
399ab595 | 1226 | */ |
49f693fa | 1227 | i = 0; |
9788e059 | 1228 | xfs_iext_first(ifp, &icur); |
6d79c95c | 1229 | |
399ab595 | 1230 | /* |
49f693fa | 1231 | * Loop over all leaf nodes. Copy information to the extent records. |
399ab595 | 1232 | */ |
49f693fa DC |
1233 | for (;;) { |
1234 | xfs_bmbt_rec_t *frp; | |
1235 | xfs_fsblock_t nextbno; | |
1236 | xfs_extnum_t num_recs; | |
399ab595 | 1237 | |
49f693fa | 1238 | num_recs = xfs_btree_get_numrecs(block); |
6d79c95c CH |
1239 | if (unlikely(i + num_recs > nextents)) { |
1240 | ASSERT(i + num_recs <= nextents); | |
49f693fa DC |
1241 | xfs_warn(ip->i_mount, |
1242 | "corrupt dinode %Lu, (btree extents).", | |
1243 | (unsigned long long) ip->i_ino); | |
b02a2c9e DW |
1244 | xfs_inode_verifier_error(ip, -EFSCORRUPTED, |
1245 | __func__, block, sizeof(*block), | |
1246 | __this_address); | |
6d79c95c CH |
1247 | error = -EFSCORRUPTED; |
1248 | goto out_brelse; | |
399ab595 NS |
1249 | } |
1250 | /* | |
49f693fa | 1251 | * Read-ahead the next leaf block, if any. |
399ab595 | 1252 | */ |
49f693fa DC |
1253 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
1254 | if (nextbno != NULLFSBLOCK) | |
1255 | xfs_btree_reada_bufl(mp, nextbno, 1, | |
1256 | &xfs_bmbt_buf_ops); | |
399ab595 | 1257 | /* |
49f693fa | 1258 | * Copy records into the extent records. |
399ab595 | 1259 | */ |
49f693fa | 1260 | frp = XFS_BMBT_REC_ADDR(mp, block, 1); |
b37d753d | 1261 | for (j = 0; j < num_recs; j++, frp++, i++) { |
0cf6a3a9 DW |
1262 | xfs_failaddr_t fa; |
1263 | ||
080f0c71 | 1264 | xfs_bmbt_disk_get_all(frp, &new); |
0cf6a3a9 DW |
1265 | fa = xfs_bmap_validate_extent(ip, whichfork, &new); |
1266 | if (fa) { | |
6d79c95c | 1267 | error = -EFSCORRUPTED; |
0cf6a3a9 DW |
1268 | xfs_inode_verifier_error(ip, error, |
1269 | "xfs_iread_extents(2)", | |
1270 | frp, sizeof(*frp), fa); | |
6d79c95c | 1271 | goto out_brelse; |
49f693fa | 1272 | } |
26a75f67 | 1273 | xfs_iext_insert(ip, &icur, &new, state); |
9788e059 CH |
1274 | trace_xfs_read_extent(ip, &icur, state, _THIS_IP_); |
1275 | xfs_iext_next(ifp, &icur); | |
399ab595 | 1276 | } |
49f693fa DC |
1277 | xfs_trans_brelse(tp, bp); |
1278 | bno = nextbno; | |
399ab595 | 1279 | /* |
49f693fa | 1280 | * If we've reached the end, stop. |
399ab595 | 1281 | */ |
49f693fa DC |
1282 | if (bno == NULLFSBLOCK) |
1283 | break; | |
1284 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1285 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1286 | if (error) | |
6d79c95c | 1287 | goto out; |
49f693fa | 1288 | block = XFS_BUF_TO_BLOCK(bp); |
399ab595 | 1289 | } |
6d79c95c CH |
1290 | |
1291 | if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) { | |
1292 | error = -EFSCORRUPTED; | |
1293 | goto out; | |
1294 | } | |
d09d4e5f | 1295 | ASSERT(i == xfs_iext_count(ifp)); |
6d79c95c CH |
1296 | |
1297 | ifp->if_flags |= XFS_IFEXTENTS; | |
49f693fa | 1298 | return 0; |
6d79c95c CH |
1299 | |
1300 | out_brelse: | |
49f693fa | 1301 | xfs_trans_brelse(tp, bp); |
6d79c95c CH |
1302 | out: |
1303 | xfs_iext_destroy(ifp); | |
1304 | return error; | |
49f693fa | 1305 | } |
399ab595 | 1306 | |
49f693fa | 1307 | /* |
2d3a6501 CH |
1308 | * Returns the relative block number of the first unused block(s) in the given |
1309 | * fork with at least "len" logically contiguous blocks free. This is the | |
1310 | * lowest-address hole if the fork has holes, else the first block past the end | |
1311 | * of fork. Return 0 if the fork is currently local (in-inode). | |
49f693fa DC |
1312 | */ |
1313 | int /* error */ | |
1314 | xfs_bmap_first_unused( | |
2d3a6501 CH |
1315 | struct xfs_trans *tp, /* transaction pointer */ |
1316 | struct xfs_inode *ip, /* incore inode */ | |
1317 | xfs_extlen_t len, /* size of hole to find */ | |
1318 | xfs_fileoff_t *first_unused, /* unused block */ | |
1319 | int whichfork) /* data or attr fork */ | |
49f693fa | 1320 | { |
2d3a6501 CH |
1321 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1322 | struct xfs_bmbt_irec got; | |
9788e059 | 1323 | struct xfs_iext_cursor icur; |
2d3a6501 CH |
1324 | xfs_fileoff_t lastaddr = 0; |
1325 | xfs_fileoff_t lowest, max; | |
1326 | int error; | |
49f693fa DC |
1327 | |
1328 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE || | |
1329 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS || | |
1330 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
2d3a6501 | 1331 | |
49f693fa DC |
1332 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { |
1333 | *first_unused = 0; | |
1334 | return 0; | |
1335 | } | |
f71f3bc3 | 1336 | |
2d3a6501 CH |
1337 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
1338 | error = xfs_iread_extents(tp, ip, whichfork); | |
1339 | if (error) | |
1340 | return error; | |
1341 | } | |
f71f3bc3 | 1342 | |
2d3a6501 | 1343 | lowest = max = *first_unused; |
9788e059 | 1344 | for_each_xfs_iext(ifp, &icur, &got) { |
2bd0ea18 | 1345 | /* |
49f693fa | 1346 | * See if the hole before this extent will work. |
2bd0ea18 | 1347 | */ |
f71f3bc3 | 1348 | if (got.br_startoff >= lowest + len && |
2d3a6501 CH |
1349 | got.br_startoff - max >= len) |
1350 | break; | |
f71f3bc3 | 1351 | lastaddr = got.br_startoff + got.br_blockcount; |
49f693fa DC |
1352 | max = XFS_FILEOFF_MAX(lastaddr, lowest); |
1353 | } | |
2d3a6501 | 1354 | |
49f693fa DC |
1355 | *first_unused = max; |
1356 | return 0; | |
1357 | } | |
1358 | ||
1359 | /* | |
e6d77a21 | 1360 | * Returns the file-relative block number of the last block - 1 before |
49f693fa DC |
1361 | * last_block (input value) in the file. |
1362 | * This is not based on i_size, it is based on the extent records. | |
1363 | * Returns 0 for local files, as they do not have extent records. | |
1364 | */ | |
1365 | int /* error */ | |
1366 | xfs_bmap_last_before( | |
b0385364 CH |
1367 | struct xfs_trans *tp, /* transaction pointer */ |
1368 | struct xfs_inode *ip, /* incore inode */ | |
1369 | xfs_fileoff_t *last_block, /* last block */ | |
1370 | int whichfork) /* data or attr fork */ | |
49f693fa | 1371 | { |
b0385364 CH |
1372 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1373 | struct xfs_bmbt_irec got; | |
9788e059 | 1374 | struct xfs_iext_cursor icur; |
b0385364 | 1375 | int error; |
49f693fa | 1376 | |
b0385364 CH |
1377 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { |
1378 | case XFS_DINODE_FMT_LOCAL: | |
49f693fa DC |
1379 | *last_block = 0; |
1380 | return 0; | |
b0385364 CH |
1381 | case XFS_DINODE_FMT_BTREE: |
1382 | case XFS_DINODE_FMT_EXTENTS: | |
1383 | break; | |
1384 | default: | |
1385 | return -EIO; | |
49f693fa | 1386 | } |
b0385364 CH |
1387 | |
1388 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1389 | error = xfs_iread_extents(tp, ip, whichfork); | |
1390 | if (error) | |
1391 | return error; | |
2bd0ea18 | 1392 | } |
b0385364 | 1393 | |
9788e059 | 1394 | if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got)) |
d6fbe8fe | 1395 | *last_block = 0; |
49f693fa | 1396 | return 0; |
5e656dbb BN |
1397 | } |
1398 | ||
613e6057 | 1399 | int |
49f693fa DC |
1400 | xfs_bmap_last_extent( |
1401 | struct xfs_trans *tp, | |
1402 | struct xfs_inode *ip, | |
1403 | int whichfork, | |
1404 | struct xfs_bmbt_irec *rec, | |
1405 | int *is_empty) | |
56b2de80 | 1406 | { |
49f693fa | 1407 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
9788e059 | 1408 | struct xfs_iext_cursor icur; |
56b2de80 DC |
1409 | int error; |
1410 | ||
49f693fa DC |
1411 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
1412 | error = xfs_iread_extents(tp, ip, whichfork); | |
1413 | if (error) | |
1414 | return error; | |
1415 | } | |
1416 | ||
9788e059 CH |
1417 | xfs_iext_last(ifp, &icur); |
1418 | if (!xfs_iext_get_extent(ifp, &icur, rec)) | |
49f693fa | 1419 | *is_empty = 1; |
9788e059 CH |
1420 | else |
1421 | *is_empty = 0; | |
49f693fa DC |
1422 | return 0; |
1423 | } | |
1424 | ||
1425 | /* | |
1426 | * Check the last inode extent to determine whether this allocation will result | |
1427 | * in blocks being allocated at the end of the file. When we allocate new data | |
1428 | * blocks at the end of the file which do not start at the previous data block, | |
1429 | * we will try to align the new blocks at stripe unit boundaries. | |
1430 | * | |
ff105f75 | 1431 | * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be |
49f693fa DC |
1432 | * at, or past the EOF. |
1433 | */ | |
1434 | STATIC int | |
1435 | xfs_bmap_isaeof( | |
1436 | struct xfs_bmalloca *bma, | |
1437 | int whichfork) | |
1438 | { | |
1439 | struct xfs_bmbt_irec rec; | |
1440 | int is_empty; | |
1441 | int error; | |
1442 | ||
180d3cd8 | 1443 | bma->aeof = false; |
49f693fa DC |
1444 | error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec, |
1445 | &is_empty); | |
ff105f75 | 1446 | if (error) |
49f693fa | 1447 | return error; |
56b2de80 | 1448 | |
ff105f75 | 1449 | if (is_empty) { |
180d3cd8 | 1450 | bma->aeof = true; |
ff105f75 DC |
1451 | return 0; |
1452 | } | |
1453 | ||
56b2de80 | 1454 | /* |
49f693fa DC |
1455 | * Check if we are allocation or past the last extent, or at least into |
1456 | * the last delayed allocated extent. | |
56b2de80 | 1457 | */ |
49f693fa DC |
1458 | bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount || |
1459 | (bma->offset >= rec.br_startoff && | |
1460 | isnullstartblock(rec.br_startblock)); | |
1461 | return 0; | |
1462 | } | |
56b2de80 | 1463 | |
49f693fa DC |
1464 | /* |
1465 | * Returns the file-relative block number of the first block past eof in | |
1466 | * the file. This is not based on i_size, it is based on the extent records. | |
1467 | * Returns 0 for local files, as they do not have extent records. | |
1468 | */ | |
1469 | int | |
1470 | xfs_bmap_last_offset( | |
49f693fa DC |
1471 | struct xfs_inode *ip, |
1472 | xfs_fileoff_t *last_block, | |
1473 | int whichfork) | |
1474 | { | |
1475 | struct xfs_bmbt_irec rec; | |
1476 | int is_empty; | |
1477 | int error; | |
1478 | ||
1479 | *last_block = 0; | |
1480 | ||
1481 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) | |
1482 | return 0; | |
1483 | ||
1484 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE && | |
1485 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
12b53197 | 1486 | return -EIO; |
49f693fa DC |
1487 | |
1488 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty); | |
1489 | if (error || is_empty) | |
1490 | return error; | |
1491 | ||
1492 | *last_block = rec.br_startoff + rec.br_blockcount; | |
1493 | return 0; | |
1494 | } | |
1495 | ||
1496 | /* | |
1497 | * Returns whether the selected fork of the inode has exactly one | |
1498 | * block or not. For the data fork we check this matches di_size, | |
1499 | * implying the file's range is 0..bsize-1. | |
1500 | */ | |
1501 | int /* 1=>1 block, 0=>otherwise */ | |
1502 | xfs_bmap_one_block( | |
1503 | xfs_inode_t *ip, /* incore inode */ | |
1504 | int whichfork) /* data or attr fork */ | |
1505 | { | |
49f693fa DC |
1506 | xfs_ifork_t *ifp; /* inode fork pointer */ |
1507 | int rval; /* return value */ | |
1508 | xfs_bmbt_irec_t s; /* internal version of extent */ | |
9788e059 | 1509 | struct xfs_iext_cursor icur; |
49f693fa DC |
1510 | |
1511 | #ifndef DEBUG | |
1512 | if (whichfork == XFS_DATA_FORK) | |
1513 | return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize; | |
1514 | #endif /* !DEBUG */ | |
1515 | if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1) | |
1516 | return 0; | |
1517 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
1518 | return 0; | |
1519 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1520 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
9788e059 CH |
1521 | xfs_iext_first(ifp, &icur); |
1522 | xfs_iext_get_extent(ifp, &icur, &s); | |
49f693fa DC |
1523 | rval = s.br_startoff == 0 && s.br_blockcount == 1; |
1524 | if (rval && whichfork == XFS_DATA_FORK) | |
1525 | ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize); | |
1526 | return rval; | |
1527 | } | |
1528 | ||
1529 | /* | |
1530 | * Extent tree manipulation functions used during allocation. | |
1531 | */ | |
1532 | ||
1533 | /* | |
1534 | * Convert a delayed allocation to a real allocation. | |
1535 | */ | |
1536 | STATIC int /* error */ | |
1537 | xfs_bmap_add_extent_delay_real( | |
1277a5e0 DW |
1538 | struct xfs_bmalloca *bma, |
1539 | int whichfork) | |
49f693fa DC |
1540 | { |
1541 | struct xfs_bmbt_irec *new = &bma->got; | |
49f693fa DC |
1542 | int error; /* error return value */ |
1543 | int i; /* temp state */ | |
1544 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1545 | xfs_fileoff_t new_endoff; /* end offset of new entry */ | |
1546 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
1547 | /* left is 0, right is 1, prev is 2 */ | |
1548 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 1549 | int state = xfs_bmap_fork_to_state(whichfork); |
49f693fa DC |
1550 | xfs_filblks_t da_new; /* new count del alloc blocks used */ |
1551 | xfs_filblks_t da_old; /* old count del alloc blocks used */ | |
1552 | xfs_filblks_t temp=0; /* value for da_new calculations */ | |
49f693fa | 1553 | int tmp_rval; /* partial logging flags */ |
19ebedcf | 1554 | struct xfs_mount *mp; |
1277a5e0 | 1555 | xfs_extnum_t *nextents; |
19884717 | 1556 | struct xfs_bmbt_irec old; |
49f693fa | 1557 | |
65ca3804 | 1558 | mp = bma->ip->i_mount; |
36e8786d | 1559 | ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
1277a5e0 DW |
1560 | ASSERT(whichfork != XFS_ATTR_FORK); |
1561 | nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents : | |
1562 | &bma->ip->i_d.di_nextents); | |
56b2de80 | 1563 | |
49f693fa DC |
1564 | ASSERT(!isnullstartblock(new->br_startblock)); |
1565 | ASSERT(!bma->cur || | |
1566 | (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); | |
56b2de80 | 1567 | |
79896434 | 1568 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
1569 | |
1570 | #define LEFT r[0] | |
1571 | #define RIGHT r[1] | |
1572 | #define PREV r[2] | |
56b2de80 DC |
1573 | |
1574 | /* | |
49f693fa | 1575 | * Set up a bunch of variables to make the tests simpler. |
56b2de80 | 1576 | */ |
9788e059 | 1577 | xfs_iext_get_extent(ifp, &bma->icur, &PREV); |
49f693fa | 1578 | new_endoff = new->br_startoff + new->br_blockcount; |
19884717 | 1579 | ASSERT(isnullstartblock(PREV.br_startblock)); |
49f693fa DC |
1580 | ASSERT(PREV.br_startoff <= new->br_startoff); |
1581 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
1582 | ||
1583 | da_old = startblockval(PREV.br_startblock); | |
1584 | da_new = 0; | |
1585 | ||
56b2de80 | 1586 | /* |
49f693fa DC |
1587 | * Set flags determining what part of the previous delayed allocation |
1588 | * extent is being replaced by a real allocation. | |
56b2de80 | 1589 | */ |
49f693fa DC |
1590 | if (PREV.br_startoff == new->br_startoff) |
1591 | state |= BMAP_LEFT_FILLING; | |
1592 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
1593 | state |= BMAP_RIGHT_FILLING; | |
1594 | ||
56b2de80 | 1595 | /* |
49f693fa DC |
1596 | * Check and set flags if this segment has a left neighbor. |
1597 | * Don't set contiguous if the combined extent would be too large. | |
56b2de80 | 1598 | */ |
9788e059 | 1599 | if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) { |
49f693fa | 1600 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
1601 | if (isnullstartblock(LEFT.br_startblock)) |
1602 | state |= BMAP_LEFT_DELAY; | |
1603 | } | |
1604 | ||
1605 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
1606 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
1607 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
1608 | LEFT.br_state == new->br_state && | |
1609 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
1610 | state |= BMAP_LEFT_CONTIG; | |
56b2de80 DC |
1611 | |
1612 | /* | |
49f693fa DC |
1613 | * Check and set flags if this segment has a right neighbor. |
1614 | * Don't set contiguous if the combined extent would be too large. | |
1615 | * Also check for all-three-contiguous being too large. | |
56b2de80 | 1616 | */ |
9788e059 | 1617 | if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) { |
49f693fa | 1618 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
1619 | if (isnullstartblock(RIGHT.br_startblock)) |
1620 | state |= BMAP_RIGHT_DELAY; | |
1621 | } | |
56b2de80 | 1622 | |
49f693fa DC |
1623 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
1624 | new_endoff == RIGHT.br_startoff && | |
1625 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
1626 | new->br_state == RIGHT.br_state && | |
1627 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
1628 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1629 | BMAP_RIGHT_FILLING)) != | |
1630 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1631 | BMAP_RIGHT_FILLING) || | |
1632 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
1633 | <= MAXEXTLEN)) | |
1634 | state |= BMAP_RIGHT_CONTIG; | |
5e656dbb | 1635 | |
49f693fa DC |
1636 | error = 0; |
1637 | /* | |
1638 | * Switch out based on the FILLING and CONTIG state bits. | |
1639 | */ | |
1640 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1641 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
1642 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1643 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1644 | /* | |
1645 | * Filling in all of a previously delayed allocation extent. | |
1646 | * The left and right neighbors are both contiguous with new. | |
1647 | */ | |
19884717 | 1648 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
a2ceac1f | 1649 | |
cf455a62 CH |
1650 | xfs_iext_remove(bma->ip, &bma->icur, state); |
1651 | xfs_iext_remove(bma->ip, &bma->icur, state); | |
9788e059 CH |
1652 | xfs_iext_prev(ifp, &bma->icur); |
1653 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
1277a5e0 | 1654 | (*nextents)--; |
12ebbfe1 | 1655 | |
49f693fa DC |
1656 | if (bma->cur == NULL) |
1657 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1658 | else { | |
1659 | rval = XFS_ILOG_CORE; | |
70a93110 | 1660 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
49f693fa DC |
1661 | if (error) |
1662 | goto done; | |
19ebedcf | 1663 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1664 | error = xfs_btree_delete(bma->cur, &i); |
1665 | if (error) | |
1666 | goto done; | |
19ebedcf | 1667 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1668 | error = xfs_btree_decrement(bma->cur, 0, &i); |
1669 | if (error) | |
1670 | goto done; | |
19ebedcf | 1671 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1672 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1673 | if (error) |
1674 | goto done; | |
5e656dbb | 1675 | } |
49f693fa DC |
1676 | break; |
1677 | ||
1678 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1679 | /* | |
1680 | * Filling in all of a previously delayed allocation extent. | |
1681 | * The left neighbor is contiguous, the right is not. | |
1682 | */ | |
19884717 | 1683 | old = LEFT; |
19884717 | 1684 | LEFT.br_blockcount += PREV.br_blockcount; |
12ebbfe1 | 1685 | |
cf455a62 | 1686 | xfs_iext_remove(bma->ip, &bma->icur, state); |
9788e059 CH |
1687 | xfs_iext_prev(ifp, &bma->icur); |
1688 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
49f693fa | 1689 | |
49f693fa DC |
1690 | if (bma->cur == NULL) |
1691 | rval = XFS_ILOG_DEXT; | |
1692 | else { | |
1693 | rval = 0; | |
70a93110 | 1694 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1695 | if (error) |
1696 | goto done; | |
19ebedcf | 1697 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1698 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1699 | if (error) |
1700 | goto done; | |
1701 | } | |
1702 | break; | |
1703 | ||
1704 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1705 | /* | |
1706 | * Filling in all of a previously delayed allocation extent. | |
1707 | * The right neighbor is contiguous, the left is not. | |
1708 | */ | |
19884717 CH |
1709 | PREV.br_startblock = new->br_startblock; |
1710 | PREV.br_blockcount += RIGHT.br_blockcount; | |
12ebbfe1 | 1711 | |
9788e059 | 1712 | xfs_iext_next(ifp, &bma->icur); |
cf455a62 | 1713 | xfs_iext_remove(bma->ip, &bma->icur, state); |
9788e059 CH |
1714 | xfs_iext_prev(ifp, &bma->icur); |
1715 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); | |
49f693fa | 1716 | |
49f693fa DC |
1717 | if (bma->cur == NULL) |
1718 | rval = XFS_ILOG_DEXT; | |
1719 | else { | |
1720 | rval = 0; | |
70a93110 | 1721 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
49f693fa DC |
1722 | if (error) |
1723 | goto done; | |
19ebedcf | 1724 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1725 | error = xfs_bmbt_update(bma->cur, &PREV); |
49f693fa DC |
1726 | if (error) |
1727 | goto done; | |
1728 | } | |
1729 | break; | |
1730 | ||
1731 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
1732 | /* | |
1733 | * Filling in all of a previously delayed allocation extent. | |
1734 | * Neither the left nor right neighbors are contiguous with | |
1735 | * the new one. | |
1736 | */ | |
19884717 CH |
1737 | PREV.br_startblock = new->br_startblock; |
1738 | PREV.br_state = new->br_state; | |
9788e059 | 1739 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
5e656dbb | 1740 | |
1277a5e0 | 1741 | (*nextents)++; |
49f693fa DC |
1742 | if (bma->cur == NULL) |
1743 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1744 | else { | |
1745 | rval = XFS_ILOG_CORE; | |
70a93110 | 1746 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1747 | if (error) |
1748 | goto done; | |
19ebedcf | 1749 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1750 | error = xfs_btree_insert(bma->cur, &i); |
1751 | if (error) | |
1752 | goto done; | |
19ebedcf | 1753 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1754 | } |
1755 | break; | |
5e656dbb | 1756 | |
49f693fa DC |
1757 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
1758 | /* | |
1759 | * Filling in the first part of a previous delayed allocation. | |
1760 | * The left neighbor is contiguous. | |
1761 | */ | |
19884717 CH |
1762 | old = LEFT; |
1763 | temp = PREV.br_blockcount - new->br_blockcount; | |
1764 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
1765 | startblockval(PREV.br_startblock)); | |
1766 | ||
19884717 | 1767 | LEFT.br_blockcount += new->br_blockcount; |
a2ceac1f | 1768 | |
bc5dec54 | 1769 | PREV.br_blockcount = temp; |
19884717 CH |
1770 | PREV.br_startoff += new->br_blockcount; |
1771 | PREV.br_startblock = nullstartblock(da_new); | |
12ebbfe1 | 1772 | |
9788e059 CH |
1773 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
1774 | xfs_iext_prev(ifp, &bma->icur); | |
1775 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
19884717 | 1776 | |
49f693fa DC |
1777 | if (bma->cur == NULL) |
1778 | rval = XFS_ILOG_DEXT; | |
1779 | else { | |
1780 | rval = 0; | |
70a93110 | 1781 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1782 | if (error) |
1783 | goto done; | |
19ebedcf | 1784 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1785 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1786 | if (error) |
1787 | goto done; | |
2bd0ea18 | 1788 | } |
49f693fa DC |
1789 | break; |
1790 | ||
1791 | case BMAP_LEFT_FILLING: | |
2bd0ea18 | 1792 | /* |
49f693fa DC |
1793 | * Filling in the first part of a previous delayed allocation. |
1794 | * The left neighbor is not contiguous. | |
5000d01d | 1795 | */ |
9788e059 | 1796 | xfs_iext_update_extent(bma->ip, state, &bma->icur, new); |
1277a5e0 | 1797 | (*nextents)++; |
49f693fa DC |
1798 | if (bma->cur == NULL) |
1799 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1800 | else { | |
1801 | rval = XFS_ILOG_CORE; | |
70a93110 | 1802 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1803 | if (error) |
1804 | goto done; | |
19ebedcf | 1805 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1806 | error = xfs_btree_insert(bma->cur, &i); |
1807 | if (error) | |
1808 | goto done; | |
19ebedcf | 1809 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1810 | } |
1811 | ||
36e8786d | 1812 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1813 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 1814 | bma->firstblock, bma->dfops, |
36e8786d | 1815 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1816 | rval |= tmp_rval; |
1817 | if (error) | |
1818 | goto done; | |
1819 | } | |
19884717 CH |
1820 | |
1821 | temp = PREV.br_blockcount - new->br_blockcount; | |
49f693fa DC |
1822 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1823 | startblockval(PREV.br_startblock) - | |
1824 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
19884717 | 1825 | |
19884717 CH |
1826 | PREV.br_startoff = new_endoff; |
1827 | PREV.br_blockcount = temp; | |
1828 | PREV.br_startblock = nullstartblock(da_new); | |
9788e059 | 1829 | xfs_iext_next(ifp, &bma->icur); |
26a75f67 | 1830 | xfs_iext_insert(bma->ip, &bma->icur, &PREV, state); |
9788e059 | 1831 | xfs_iext_prev(ifp, &bma->icur); |
49f693fa DC |
1832 | break; |
1833 | ||
1834 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
5e656dbb | 1835 | /* |
49f693fa DC |
1836 | * Filling in the last part of a previous delayed allocation. |
1837 | * The right neighbor is contiguous with the new allocation. | |
5e656dbb | 1838 | */ |
19884717 | 1839 | old = RIGHT; |
19884717 CH |
1840 | RIGHT.br_startoff = new->br_startoff; |
1841 | RIGHT.br_startblock = new->br_startblock; | |
1842 | RIGHT.br_blockcount += new->br_blockcount; | |
19884717 | 1843 | |
49f693fa DC |
1844 | if (bma->cur == NULL) |
1845 | rval = XFS_ILOG_DEXT; | |
1846 | else { | |
1847 | rval = 0; | |
70a93110 | 1848 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1849 | if (error) |
1850 | goto done; | |
19ebedcf | 1851 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1852 | error = xfs_bmbt_update(bma->cur, &RIGHT); |
49f693fa DC |
1853 | if (error) |
1854 | goto done; | |
1855 | } | |
1856 | ||
19884717 | 1857 | temp = PREV.br_blockcount - new->br_blockcount; |
49f693fa DC |
1858 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1859 | startblockval(PREV.br_startblock)); | |
19884717 | 1860 | |
19884717 CH |
1861 | PREV.br_blockcount = temp; |
1862 | PREV.br_startblock = nullstartblock(da_new); | |
49f693fa | 1863 | |
9788e059 CH |
1864 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
1865 | xfs_iext_next(ifp, &bma->icur); | |
1866 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT); | |
49f693fa DC |
1867 | break; |
1868 | ||
1869 | case BMAP_RIGHT_FILLING: | |
a2ceac1f | 1870 | /* |
49f693fa DC |
1871 | * Filling in the last part of a previous delayed allocation. |
1872 | * The right neighbor is not contiguous. | |
a2ceac1f | 1873 | */ |
9788e059 | 1874 | xfs_iext_update_extent(bma->ip, state, &bma->icur, new); |
1277a5e0 | 1875 | (*nextents)++; |
49f693fa DC |
1876 | if (bma->cur == NULL) |
1877 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1878 | else { | |
1879 | rval = XFS_ILOG_CORE; | |
70a93110 | 1880 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1881 | if (error) |
1882 | goto done; | |
19ebedcf | 1883 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1884 | error = xfs_btree_insert(bma->cur, &i); |
1885 | if (error) | |
1886 | goto done; | |
19ebedcf | 1887 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 1888 | } |
a2ceac1f | 1889 | |
36e8786d | 1890 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1891 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 1892 | bma->firstblock, bma->dfops, &bma->cur, 1, |
36e8786d | 1893 | &tmp_rval, whichfork); |
49f693fa DC |
1894 | rval |= tmp_rval; |
1895 | if (error) | |
1896 | goto done; | |
1897 | } | |
19884717 CH |
1898 | |
1899 | temp = PREV.br_blockcount - new->br_blockcount; | |
49f693fa DC |
1900 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1901 | startblockval(PREV.br_startblock) - | |
1902 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
19884717 | 1903 | |
19884717 CH |
1904 | PREV.br_startblock = nullstartblock(da_new); |
1905 | PREV.br_blockcount = temp; | |
26a75f67 | 1906 | xfs_iext_insert(bma->ip, &bma->icur, &PREV, state); |
9788e059 | 1907 | xfs_iext_next(ifp, &bma->icur); |
49f693fa DC |
1908 | break; |
1909 | ||
1910 | case 0: | |
5e656dbb | 1911 | /* |
49f693fa DC |
1912 | * Filling in the middle part of a previous delayed allocation. |
1913 | * Contiguity is impossible here. | |
1914 | * This case is avoided almost all the time. | |
1915 | * | |
1916 | * We start with a delayed allocation: | |
1917 | * | |
1918 | * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+ | |
1919 | * PREV @ idx | |
1920 | * | |
1921 | * and we are allocating: | |
1922 | * +rrrrrrrrrrrrrrrrr+ | |
1923 | * new | |
1924 | * | |
1925 | * and we set it up for insertion as: | |
1926 | * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+ | |
1927 | * new | |
1928 | * PREV @ idx LEFT RIGHT | |
1929 | * inserted at idx + 1 | |
5e656dbb | 1930 | */ |
19884717 CH |
1931 | old = PREV; |
1932 | ||
1933 | /* LEFT is the new middle */ | |
49f693fa | 1934 | LEFT = *new; |
19884717 CH |
1935 | |
1936 | /* RIGHT is the new right */ | |
49f693fa | 1937 | RIGHT.br_state = PREV.br_state; |
49f693fa | 1938 | RIGHT.br_startoff = new_endoff; |
19884717 CH |
1939 | RIGHT.br_blockcount = |
1940 | PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
1941 | RIGHT.br_startblock = | |
1942 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1943 | RIGHT.br_blockcount)); | |
1944 | ||
1945 | /* truncate PREV */ | |
19884717 CH |
1946 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
1947 | PREV.br_startblock = | |
1948 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1949 | PREV.br_blockcount)); | |
9788e059 | 1950 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
19884717 | 1951 | |
9788e059 | 1952 | xfs_iext_next(ifp, &bma->icur); |
26a75f67 CH |
1953 | xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state); |
1954 | xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state); | |
1277a5e0 | 1955 | (*nextents)++; |
19884717 | 1956 | |
49f693fa DC |
1957 | if (bma->cur == NULL) |
1958 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1959 | else { | |
1960 | rval = XFS_ILOG_CORE; | |
70a93110 | 1961 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1962 | if (error) |
1963 | goto done; | |
19ebedcf | 1964 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1965 | error = xfs_btree_insert(bma->cur, &i); |
1966 | if (error) | |
1967 | goto done; | |
19ebedcf | 1968 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 1969 | } |
5e656dbb | 1970 | |
36e8786d | 1971 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1972 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 1973 | bma->firstblock, bma->dfops, &bma->cur, |
36e8786d | 1974 | 1, &tmp_rval, whichfork); |
49f693fa DC |
1975 | rval |= tmp_rval; |
1976 | if (error) | |
1977 | goto done; | |
1978 | } | |
19884717 CH |
1979 | |
1980 | da_new = startblockval(PREV.br_startblock) + | |
1981 | startblockval(RIGHT.br_startblock); | |
49f693fa DC |
1982 | break; |
1983 | ||
1984 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1985 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1986 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
1987 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1988 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1989 | case BMAP_LEFT_CONTIG: | |
1990 | case BMAP_RIGHT_CONTIG: | |
1991 | /* | |
1992 | * These cases are all impossible. | |
1993 | */ | |
1994 | ASSERT(0); | |
1995 | } | |
1996 | ||
d7f80320 DW |
1997 | /* add reverse mapping */ |
1998 | error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new); | |
1999 | if (error) | |
2000 | goto done; | |
2001 | ||
49f693fa | 2002 | /* convert to a btree if necessary */ |
36e8786d | 2003 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa DC |
2004 | int tmp_logflags; /* partial log flag return val */ |
2005 | ||
2006 | ASSERT(bma->cur == NULL); | |
2007 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, | |
f33cea1a | 2008 | bma->firstblock, bma->dfops, &bma->cur, |
36e8786d | 2009 | da_old > 0, &tmp_logflags, whichfork); |
49f693fa DC |
2010 | bma->logflags |= tmp_logflags; |
2011 | if (error) | |
2012 | goto done; | |
2013 | } | |
2014 | ||
5bfca476 CH |
2015 | if (bma->cur) { |
2016 | da_new += bma->cur->bc_private.b.allocated; | |
2017 | bma->cur->bc_private.b.allocated = 0; | |
49f693fa DC |
2018 | } |
2019 | ||
5bfca476 CH |
2020 | /* adjust for changes in reserved delayed indirect blocks */ |
2021 | if (da_new != da_old) { | |
2022 | ASSERT(state == 0 || da_new < da_old); | |
2023 | error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), | |
2024 | false); | |
2025 | } | |
49f693fa | 2026 | |
36e8786d | 2027 | xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork); |
49f693fa | 2028 | done: |
1277a5e0 DW |
2029 | if (whichfork != XFS_COW_FORK) |
2030 | bma->logflags |= rval; | |
49f693fa DC |
2031 | return error; |
2032 | #undef LEFT | |
2033 | #undef RIGHT | |
2034 | #undef PREV | |
2bd0ea18 NS |
2035 | } |
2036 | ||
2037 | /* | |
49f693fa | 2038 | * Convert an unwritten allocation to a real allocation or vice versa. |
2bd0ea18 NS |
2039 | */ |
2040 | STATIC int /* error */ | |
49f693fa DC |
2041 | xfs_bmap_add_extent_unwritten_real( |
2042 | struct xfs_trans *tp, | |
2bd0ea18 | 2043 | xfs_inode_t *ip, /* incore inode pointer */ |
4072e4b4 | 2044 | int whichfork, |
9788e059 | 2045 | struct xfs_iext_cursor *icur, |
49f693fa DC |
2046 | xfs_btree_cur_t **curp, /* if *curp is null, not a btree */ |
2047 | xfs_bmbt_irec_t *new, /* new data to add to file extents */ | |
2048 | xfs_fsblock_t *first, /* pointer to firstblock variable */ | |
f33cea1a | 2049 | struct xfs_defer_ops *dfops, /* list of extents to be freed */ |
49f693fa | 2050 | int *logflagsp) /* inode logging flags */ |
2bd0ea18 | 2051 | { |
49f693fa | 2052 | xfs_btree_cur_t *cur; /* btree cursor */ |
2bd0ea18 | 2053 | int error; /* error return value */ |
2bd0ea18 NS |
2054 | int i; /* temp state */ |
2055 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
49f693fa | 2056 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
49f693fa DC |
2057 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ |
2058 | /* left is 0, right is 1, prev is 2 */ | |
2059 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 2060 | int state = xfs_bmap_fork_to_state(whichfork); |
4072e4b4 | 2061 | struct xfs_mount *mp = ip->i_mount; |
fd19685d | 2062 | struct xfs_bmbt_irec old; |
5000d01d | 2063 | |
49f693fa | 2064 | *logflagsp = 0; |
56b2de80 | 2065 | |
49f693fa | 2066 | cur = *curp; |
4072e4b4 | 2067 | ifp = XFS_IFORK_PTR(ip, whichfork); |
56b2de80 | 2068 | |
49f693fa DC |
2069 | ASSERT(!isnullstartblock(new->br_startblock)); |
2070 | ||
79896434 | 2071 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
2072 | |
2073 | #define LEFT r[0] | |
2074 | #define RIGHT r[1] | |
2075 | #define PREV r[2] | |
2076 | ||
2077 | /* | |
2078 | * Set up a bunch of variables to make the tests simpler. | |
2079 | */ | |
2bd0ea18 | 2080 | error = 0; |
9788e059 | 2081 | xfs_iext_get_extent(ifp, icur, &PREV); |
fd19685d | 2082 | ASSERT(new->br_state != PREV.br_state); |
49f693fa DC |
2083 | new_endoff = new->br_startoff + new->br_blockcount; |
2084 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
2085 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
2086 | ||
2bd0ea18 | 2087 | /* |
49f693fa DC |
2088 | * Set flags determining what part of the previous oldext allocation |
2089 | * extent is being replaced by a newext allocation. | |
2bd0ea18 | 2090 | */ |
49f693fa DC |
2091 | if (PREV.br_startoff == new->br_startoff) |
2092 | state |= BMAP_LEFT_FILLING; | |
2093 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
2094 | state |= BMAP_RIGHT_FILLING; | |
2bd0ea18 | 2095 | |
49f693fa DC |
2096 | /* |
2097 | * Check and set flags if this segment has a left neighbor. | |
2098 | * Don't set contiguous if the combined extent would be too large. | |
2099 | */ | |
9788e059 | 2100 | if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) { |
49f693fa | 2101 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
2102 | if (isnullstartblock(LEFT.br_startblock)) |
2103 | state |= BMAP_LEFT_DELAY; | |
2bd0ea18 | 2104 | } |
49f693fa DC |
2105 | |
2106 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2107 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
2108 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
fd19685d | 2109 | LEFT.br_state == new->br_state && |
49f693fa DC |
2110 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) |
2111 | state |= BMAP_LEFT_CONTIG; | |
2112 | ||
2bd0ea18 | 2113 | /* |
49f693fa DC |
2114 | * Check and set flags if this segment has a right neighbor. |
2115 | * Don't set contiguous if the combined extent would be too large. | |
2116 | * Also check for all-three-contiguous being too large. | |
2bd0ea18 | 2117 | */ |
9788e059 | 2118 | if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) { |
49f693fa | 2119 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2120 | if (isnullstartblock(RIGHT.br_startblock)) |
2121 | state |= BMAP_RIGHT_DELAY; | |
2122 | } | |
a2ceac1f | 2123 | |
49f693fa DC |
2124 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2125 | new_endoff == RIGHT.br_startoff && | |
2126 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
fd19685d | 2127 | new->br_state == RIGHT.br_state && |
49f693fa DC |
2128 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && |
2129 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2130 | BMAP_RIGHT_FILLING)) != | |
2131 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2132 | BMAP_RIGHT_FILLING) || | |
2133 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
2134 | <= MAXEXTLEN)) | |
2135 | state |= BMAP_RIGHT_CONTIG; | |
2bd0ea18 | 2136 | |
49f693fa DC |
2137 | /* |
2138 | * Switch out based on the FILLING and CONTIG state bits. | |
2139 | */ | |
2140 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2141 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
2142 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2143 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
2bd0ea18 | 2144 | /* |
49f693fa DC |
2145 | * Setting all of a previous oldext extent to newext. |
2146 | * The left and right neighbors are both contiguous with new. | |
2bd0ea18 | 2147 | */ |
fd19685d | 2148 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
49f693fa | 2149 | |
cf455a62 CH |
2150 | xfs_iext_remove(ip, icur, state); |
2151 | xfs_iext_remove(ip, icur, state); | |
9788e059 CH |
2152 | xfs_iext_prev(ifp, icur); |
2153 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
4072e4b4 DW |
2154 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2155 | XFS_IFORK_NEXTENTS(ip, whichfork) - 2); | |
49f693fa DC |
2156 | if (cur == NULL) |
2157 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2158 | else { | |
2159 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2160 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2161 | if (error) | |
49f693fa | 2162 | goto done; |
19ebedcf | 2163 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2164 | if ((error = xfs_btree_delete(cur, &i))) |
2165 | goto done; | |
19ebedcf | 2166 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2167 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2168 | goto done; | |
19ebedcf | 2169 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2170 | if ((error = xfs_btree_delete(cur, &i))) |
2171 | goto done; | |
19ebedcf | 2172 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2173 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2174 | goto done; | |
19ebedcf | 2175 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2176 | error = xfs_bmbt_update(cur, &LEFT); |
fd19685d | 2177 | if (error) |
49f693fa | 2178 | goto done; |
2bd0ea18 | 2179 | } |
2bd0ea18 NS |
2180 | break; |
2181 | ||
49f693fa | 2182 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: |
2bd0ea18 | 2183 | /* |
49f693fa DC |
2184 | * Setting all of a previous oldext extent to newext. |
2185 | * The left neighbor is contiguous, the right is not. | |
2bd0ea18 | 2186 | */ |
fd19685d | 2187 | LEFT.br_blockcount += PREV.br_blockcount; |
49f693fa | 2188 | |
cf455a62 | 2189 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2190 | xfs_iext_prev(ifp, icur); |
2191 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
4072e4b4 DW |
2192 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2193 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
49f693fa DC |
2194 | if (cur == NULL) |
2195 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2196 | else { | |
2197 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2198 | error = xfs_bmbt_lookup_eq(cur, &PREV, &i); |
2199 | if (error) | |
49f693fa | 2200 | goto done; |
19ebedcf | 2201 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2202 | if ((error = xfs_btree_delete(cur, &i))) |
2203 | goto done; | |
19ebedcf | 2204 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2205 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2206 | goto done; | |
19ebedcf | 2207 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2208 | error = xfs_bmbt_update(cur, &LEFT); |
fd19685d | 2209 | if (error) |
49f693fa | 2210 | goto done; |
2bd0ea18 | 2211 | } |
2bd0ea18 | 2212 | break; |
5000d01d | 2213 | |
49f693fa | 2214 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2bd0ea18 | 2215 | /* |
49f693fa DC |
2216 | * Setting all of a previous oldext extent to newext. |
2217 | * The right neighbor is contiguous, the left is not. | |
2bd0ea18 | 2218 | */ |
fd19685d CH |
2219 | PREV.br_blockcount += RIGHT.br_blockcount; |
2220 | PREV.br_state = new->br_state; | |
8121dd76 | 2221 | |
9788e059 | 2222 | xfs_iext_next(ifp, icur); |
cf455a62 | 2223 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2224 | xfs_iext_prev(ifp, icur); |
2225 | xfs_iext_update_extent(ip, state, icur, &PREV); | |
fd19685d | 2226 | |
4072e4b4 DW |
2227 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2228 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
49f693fa DC |
2229 | if (cur == NULL) |
2230 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2231 | else { | |
2232 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2233 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2234 | if (error) | |
49f693fa | 2235 | goto done; |
19ebedcf | 2236 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2237 | if ((error = xfs_btree_delete(cur, &i))) |
2238 | goto done; | |
19ebedcf | 2239 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2240 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2241 | goto done; | |
19ebedcf | 2242 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2243 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2244 | if (error) |
49f693fa DC |
2245 | goto done; |
2246 | } | |
2247 | break; | |
2bd0ea18 | 2248 | |
49f693fa DC |
2249 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: |
2250 | /* | |
2251 | * Setting all of a previous oldext extent to newext. | |
2252 | * Neither the left nor right neighbors are contiguous with | |
2253 | * the new one. | |
2254 | */ | |
fd19685d | 2255 | PREV.br_state = new->br_state; |
9788e059 | 2256 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2bd0ea18 | 2257 | |
49f693fa DC |
2258 | if (cur == NULL) |
2259 | rval = XFS_ILOG_DEXT; | |
2260 | else { | |
2261 | rval = 0; | |
70a93110 CH |
2262 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2263 | if (error) | |
49f693fa | 2264 | goto done; |
19ebedcf | 2265 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2266 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2267 | if (error) |
49f693fa DC |
2268 | goto done; |
2269 | } | |
2270 | break; | |
2bd0ea18 | 2271 | |
49f693fa DC |
2272 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
2273 | /* | |
2274 | * Setting the first part of a previous oldext extent to newext. | |
2275 | * The left neighbor is contiguous. | |
2276 | */ | |
fd19685d | 2277 | LEFT.br_blockcount += new->br_blockcount; |
a2ceac1f | 2278 | |
fd19685d | 2279 | old = PREV; |
fd19685d CH |
2280 | PREV.br_startoff += new->br_blockcount; |
2281 | PREV.br_startblock += new->br_blockcount; | |
2282 | PREV.br_blockcount -= new->br_blockcount; | |
b3563c19 | 2283 | |
9788e059 CH |
2284 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2285 | xfs_iext_prev(ifp, icur); | |
2286 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
b3563c19 | 2287 | |
49f693fa DC |
2288 | if (cur == NULL) |
2289 | rval = XFS_ILOG_DEXT; | |
2290 | else { | |
2291 | rval = 0; | |
70a93110 | 2292 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2293 | if (error) |
49f693fa | 2294 | goto done; |
19ebedcf | 2295 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2296 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2297 | if (error) |
49f693fa | 2298 | goto done; |
fd19685d CH |
2299 | error = xfs_btree_decrement(cur, 0, &i); |
2300 | if (error) | |
49f693fa | 2301 | goto done; |
d0e5f1ff | 2302 | error = xfs_bmbt_update(cur, &LEFT); |
49f693fa DC |
2303 | if (error) |
2304 | goto done; | |
2bd0ea18 | 2305 | } |
49f693fa | 2306 | break; |
b3563c19 | 2307 | |
49f693fa DC |
2308 | case BMAP_LEFT_FILLING: |
2309 | /* | |
2310 | * Setting the first part of a previous oldext extent to newext. | |
2311 | * The left neighbor is not contiguous. | |
2312 | */ | |
fd19685d | 2313 | old = PREV; |
fd19685d CH |
2314 | PREV.br_startoff += new->br_blockcount; |
2315 | PREV.br_startblock += new->br_blockcount; | |
2316 | PREV.br_blockcount -= new->br_blockcount; | |
2bd0ea18 | 2317 | |
9788e059 | 2318 | xfs_iext_update_extent(ip, state, icur, &PREV); |
26a75f67 | 2319 | xfs_iext_insert(ip, icur, new, state); |
4072e4b4 DW |
2320 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2321 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
49f693fa DC |
2322 | if (cur == NULL) |
2323 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2324 | else { | |
2325 | rval = XFS_ILOG_CORE; | |
70a93110 | 2326 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2327 | if (error) |
49f693fa | 2328 | goto done; |
19ebedcf | 2329 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2330 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2331 | if (error) |
49f693fa DC |
2332 | goto done; |
2333 | cur->bc_rec.b = *new; | |
2334 | if ((error = xfs_btree_insert(cur, &i))) | |
2335 | goto done; | |
19ebedcf | 2336 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2337 | } |
2338 | break; | |
56b2de80 | 2339 | |
49f693fa DC |
2340 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2341 | /* | |
2342 | * Setting the last part of a previous oldext extent to newext. | |
2343 | * The right neighbor is contiguous with the new allocation. | |
2344 | */ | |
fd19685d | 2345 | old = PREV; |
fd19685d | 2346 | PREV.br_blockcount -= new->br_blockcount; |
56b2de80 | 2347 | |
fd19685d CH |
2348 | RIGHT.br_startoff = new->br_startoff; |
2349 | RIGHT.br_startblock = new->br_startblock; | |
2350 | RIGHT.br_blockcount += new->br_blockcount; | |
8121dd76 | 2351 | |
9788e059 CH |
2352 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2353 | xfs_iext_next(ifp, icur); | |
2354 | xfs_iext_update_extent(ip, state, icur, &RIGHT); | |
56b2de80 | 2355 | |
49f693fa DC |
2356 | if (cur == NULL) |
2357 | rval = XFS_ILOG_DEXT; | |
2358 | else { | |
2359 | rval = 0; | |
70a93110 | 2360 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2361 | if (error) |
49f693fa | 2362 | goto done; |
19ebedcf | 2363 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2364 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2365 | if (error) |
49f693fa | 2366 | goto done; |
fd19685d CH |
2367 | error = xfs_btree_increment(cur, 0, &i); |
2368 | if (error) | |
49f693fa | 2369 | goto done; |
d0e5f1ff | 2370 | error = xfs_bmbt_update(cur, &RIGHT); |
fd19685d | 2371 | if (error) |
49f693fa DC |
2372 | goto done; |
2373 | } | |
2374 | break; | |
ca86e759 | 2375 | |
49f693fa DC |
2376 | case BMAP_RIGHT_FILLING: |
2377 | /* | |
2378 | * Setting the last part of a previous oldext extent to newext. | |
2379 | * The right neighbor is not contiguous. | |
2380 | */ | |
fd19685d | 2381 | old = PREV; |
fd19685d | 2382 | PREV.br_blockcount -= new->br_blockcount; |
2bd0ea18 | 2383 | |
9788e059 CH |
2384 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2385 | xfs_iext_next(ifp, icur); | |
26a75f67 | 2386 | xfs_iext_insert(ip, icur, new, state); |
2bd0ea18 | 2387 | |
4072e4b4 DW |
2388 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2389 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
49f693fa DC |
2390 | if (cur == NULL) |
2391 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2392 | else { | |
2393 | rval = XFS_ILOG_CORE; | |
70a93110 | 2394 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2395 | if (error) |
49f693fa | 2396 | goto done; |
19ebedcf | 2397 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2398 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2399 | if (error) |
49f693fa | 2400 | goto done; |
70a93110 CH |
2401 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2402 | if (error) | |
49f693fa | 2403 | goto done; |
19ebedcf | 2404 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
2405 | if ((error = xfs_btree_insert(cur, &i))) |
2406 | goto done; | |
19ebedcf | 2407 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2408 | } |
2409 | break; | |
2410 | ||
2411 | case 0: | |
2bd0ea18 | 2412 | /* |
49f693fa DC |
2413 | * Setting the middle part of a previous oldext extent to |
2414 | * newext. Contiguity is impossible here. | |
2415 | * One extent becomes three extents. | |
2bd0ea18 | 2416 | */ |
fd19685d | 2417 | old = PREV; |
fd19685d | 2418 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
49f693fa DC |
2419 | |
2420 | r[0] = *new; | |
2421 | r[1].br_startoff = new_endoff; | |
2422 | r[1].br_blockcount = | |
fd19685d | 2423 | old.br_startoff + old.br_blockcount - new_endoff; |
49f693fa | 2424 | r[1].br_startblock = new->br_startblock + new->br_blockcount; |
fd19685d | 2425 | r[1].br_state = PREV.br_state; |
49f693fa | 2426 | |
9788e059 CH |
2427 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2428 | xfs_iext_next(ifp, icur); | |
26a75f67 CH |
2429 | xfs_iext_insert(ip, icur, &r[1], state); |
2430 | xfs_iext_insert(ip, icur, &r[0], state); | |
49f693fa | 2431 | |
4072e4b4 DW |
2432 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2433 | XFS_IFORK_NEXTENTS(ip, whichfork) + 2); | |
49f693fa DC |
2434 | if (cur == NULL) |
2435 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2436 | else { | |
2437 | rval = XFS_ILOG_CORE; | |
70a93110 | 2438 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2439 | if (error) |
49f693fa | 2440 | goto done; |
19ebedcf | 2441 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 2442 | /* new right extent - oldext */ |
d0e5f1ff CH |
2443 | error = xfs_bmbt_update(cur, &r[1]); |
2444 | if (error) | |
49f693fa DC |
2445 | goto done; |
2446 | /* new left extent - oldext */ | |
2447 | cur->bc_rec.b = PREV; | |
49f693fa DC |
2448 | if ((error = xfs_btree_insert(cur, &i))) |
2449 | goto done; | |
19ebedcf | 2450 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2451 | /* |
2452 | * Reset the cursor to the position of the new extent | |
2453 | * we are about to insert as we can't trust it after | |
2454 | * the previous insert. | |
2455 | */ | |
70a93110 CH |
2456 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2457 | if (error) | |
49f693fa | 2458 | goto done; |
19ebedcf | 2459 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa | 2460 | /* new middle extent - newext */ |
49f693fa DC |
2461 | if ((error = xfs_btree_insert(cur, &i))) |
2462 | goto done; | |
19ebedcf | 2463 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
2bd0ea18 | 2464 | } |
49f693fa DC |
2465 | break; |
2466 | ||
2467 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2468 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2469 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2470 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2471 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2472 | case BMAP_LEFT_CONTIG: | |
2473 | case BMAP_RIGHT_CONTIG: | |
5000d01d | 2474 | /* |
49f693fa | 2475 | * These cases are all impossible. |
2bd0ea18 | 2476 | */ |
49f693fa DC |
2477 | ASSERT(0); |
2478 | } | |
2479 | ||
d7f80320 | 2480 | /* update reverse mappings */ |
4072e4b4 | 2481 | error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new); |
d7f80320 DW |
2482 | if (error) |
2483 | goto done; | |
2484 | ||
49f693fa | 2485 | /* convert to a btree if necessary */ |
4072e4b4 | 2486 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
49f693fa DC |
2487 | int tmp_logflags; /* partial log flag return val */ |
2488 | ||
2489 | ASSERT(cur == NULL); | |
f33cea1a | 2490 | error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur, |
4072e4b4 | 2491 | 0, &tmp_logflags, whichfork); |
49f693fa DC |
2492 | *logflagsp |= tmp_logflags; |
2493 | if (error) | |
2494 | goto done; | |
ca86e759 | 2495 | } |
49f693fa DC |
2496 | |
2497 | /* clear out the allocated field, done with it now in any case. */ | |
2498 | if (cur) { | |
2499 | cur->bc_private.b.allocated = 0; | |
2500 | *curp = cur; | |
2501 | } | |
2502 | ||
4072e4b4 | 2503 | xfs_bmap_check_leaf_extents(*curp, ip, whichfork); |
2bd0ea18 | 2504 | done: |
49f693fa | 2505 | *logflagsp |= rval; |
2bd0ea18 | 2506 | return error; |
49f693fa DC |
2507 | #undef LEFT |
2508 | #undef RIGHT | |
2509 | #undef PREV | |
2bd0ea18 NS |
2510 | } |
2511 | ||
5e656dbb | 2512 | /* |
49f693fa | 2513 | * Convert a hole to a delayed allocation. |
5e656dbb | 2514 | */ |
49f693fa DC |
2515 | STATIC void |
2516 | xfs_bmap_add_extent_hole_delay( | |
2517 | xfs_inode_t *ip, /* incore inode pointer */ | |
cc66acaf | 2518 | int whichfork, |
9788e059 | 2519 | struct xfs_iext_cursor *icur, |
49f693fa | 2520 | xfs_bmbt_irec_t *new) /* new data to add to file extents */ |
2bd0ea18 | 2521 | { |
49f693fa DC |
2522 | xfs_ifork_t *ifp; /* inode fork pointer */ |
2523 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ | |
2524 | xfs_filblks_t newlen=0; /* new indirect size */ | |
2525 | xfs_filblks_t oldlen=0; /* old indirect size */ | |
2526 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
7bd43334 | 2527 | int state = xfs_bmap_fork_to_state(whichfork); |
450b3981 | 2528 | xfs_filblks_t temp; /* temp for indirect calculations */ |
49f693fa | 2529 | |
cc66acaf | 2530 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa | 2531 | ASSERT(isnullstartblock(new->br_startblock)); |
2bd0ea18 | 2532 | |
062998e3 | 2533 | /* |
49f693fa | 2534 | * Check and set flags if this segment has a left neighbor |
062998e3 | 2535 | */ |
9788e059 | 2536 | if (xfs_iext_peek_prev_extent(ifp, icur, &left)) { |
49f693fa | 2537 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
2538 | if (isnullstartblock(left.br_startblock)) |
2539 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2540 | } |
49f693fa DC |
2541 | |
2542 | /* | |
2543 | * Check and set flags if the current (right) segment exists. | |
2544 | * If it doesn't exist, we're converting the hole at end-of-file. | |
2545 | */ | |
9788e059 | 2546 | if (xfs_iext_get_extent(ifp, icur, &right)) { |
49f693fa | 2547 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2548 | if (isnullstartblock(right.br_startblock)) |
2549 | state |= BMAP_RIGHT_DELAY; | |
2550 | } | |
2551 | ||
2552 | /* | |
2553 | * Set contiguity flags on the left and right neighbors. | |
2554 | * Don't let extents get too large, even if the pieces are contiguous. | |
2555 | */ | |
2556 | if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) && | |
2557 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2558 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2559 | state |= BMAP_LEFT_CONTIG; | |
2560 | ||
2561 | if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) && | |
2562 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2563 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2564 | (!(state & BMAP_LEFT_CONTIG) || | |
2565 | (left.br_blockcount + new->br_blockcount + | |
2566 | right.br_blockcount <= MAXEXTLEN))) | |
2567 | state |= BMAP_RIGHT_CONTIG; | |
2568 | ||
2569 | /* | |
2570 | * Switch out based on the contiguity flags. | |
2571 | */ | |
2572 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2573 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2574 | /* | |
2575 | * New allocation is contiguous with delayed allocations | |
2576 | * on the left and on the right. | |
2577 | * Merge all three into a single extent record. | |
2578 | */ | |
49f693fa DC |
2579 | temp = left.br_blockcount + new->br_blockcount + |
2580 | right.br_blockcount; | |
2581 | ||
49f693fa DC |
2582 | oldlen = startblockval(left.br_startblock) + |
2583 | startblockval(new->br_startblock) + | |
2584 | startblockval(right.br_startblock); | |
f3b62b32 BF |
2585 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2586 | oldlen); | |
450b3981 CH |
2587 | left.br_startblock = nullstartblock(newlen); |
2588 | left.br_blockcount = temp; | |
49f693fa | 2589 | |
cf455a62 | 2590 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2591 | xfs_iext_prev(ifp, icur); |
2592 | xfs_iext_update_extent(ip, state, icur, &left); | |
49f693fa DC |
2593 | break; |
2594 | ||
2595 | case BMAP_LEFT_CONTIG: | |
2596 | /* | |
2597 | * New allocation is contiguous with a delayed allocation | |
2598 | * on the left. | |
2599 | * Merge the new allocation with the left neighbor. | |
2600 | */ | |
49f693fa DC |
2601 | temp = left.br_blockcount + new->br_blockcount; |
2602 | ||
49f693fa DC |
2603 | oldlen = startblockval(left.br_startblock) + |
2604 | startblockval(new->br_startblock); | |
f3b62b32 BF |
2605 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2606 | oldlen); | |
450b3981 CH |
2607 | left.br_blockcount = temp; |
2608 | left.br_startblock = nullstartblock(newlen); | |
ee2bb3f5 | 2609 | |
9788e059 CH |
2610 | xfs_iext_prev(ifp, icur); |
2611 | xfs_iext_update_extent(ip, state, icur, &left); | |
49f693fa DC |
2612 | break; |
2613 | ||
2614 | case BMAP_RIGHT_CONTIG: | |
2615 | /* | |
2616 | * New allocation is contiguous with a delayed allocation | |
2617 | * on the right. | |
2618 | * Merge the new allocation with the right neighbor. | |
2619 | */ | |
49f693fa DC |
2620 | temp = new->br_blockcount + right.br_blockcount; |
2621 | oldlen = startblockval(new->br_startblock) + | |
2622 | startblockval(right.br_startblock); | |
f3b62b32 BF |
2623 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2624 | oldlen); | |
450b3981 CH |
2625 | right.br_startoff = new->br_startoff; |
2626 | right.br_startblock = nullstartblock(newlen); | |
2627 | right.br_blockcount = temp; | |
9788e059 | 2628 | xfs_iext_update_extent(ip, state, icur, &right); |
49f693fa DC |
2629 | break; |
2630 | ||
2631 | case 0: | |
2632 | /* | |
2633 | * New allocation is not contiguous with another | |
2634 | * delayed allocation. | |
2635 | * Insert a new entry. | |
2636 | */ | |
2637 | oldlen = newlen = 0; | |
26a75f67 | 2638 | xfs_iext_insert(ip, icur, new, state); |
49f693fa DC |
2639 | break; |
2640 | } | |
2641 | if (oldlen != newlen) { | |
2642 | ASSERT(oldlen > newlen); | |
19ebedcf DC |
2643 | xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen), |
2644 | false); | |
49f693fa DC |
2645 | /* |
2646 | * Nothing to do for disk quota accounting here. | |
2647 | */ | |
2bd0ea18 | 2648 | } |
2bd0ea18 NS |
2649 | } |
2650 | ||
2651 | /* | |
49f693fa | 2652 | * Convert a hole to a real allocation. |
2bd0ea18 | 2653 | */ |
49f693fa DC |
2654 | STATIC int /* error */ |
2655 | xfs_bmap_add_extent_hole_real( | |
972432f2 CH |
2656 | struct xfs_trans *tp, |
2657 | struct xfs_inode *ip, | |
2658 | int whichfork, | |
9788e059 | 2659 | struct xfs_iext_cursor *icur, |
972432f2 CH |
2660 | struct xfs_btree_cur **curp, |
2661 | struct xfs_bmbt_irec *new, | |
2662 | xfs_fsblock_t *first, | |
2663 | struct xfs_defer_ops *dfops, | |
2664 | int *logflagsp) | |
5000d01d | 2665 | { |
972432f2 CH |
2666 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
2667 | struct xfs_mount *mp = ip->i_mount; | |
2668 | struct xfs_btree_cur *cur = *curp; | |
49f693fa DC |
2669 | int error; /* error return value */ |
2670 | int i; /* temp state */ | |
49f693fa DC |
2671 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2672 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2673 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 2674 | int state = xfs_bmap_fork_to_state(whichfork); |
3281eb91 | 2675 | struct xfs_bmbt_irec old; |
2bd0ea18 | 2676 | |
49f693fa | 2677 | ASSERT(!isnullstartblock(new->br_startblock)); |
972432f2 | 2678 | ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); |
5e656dbb | 2679 | |
79896434 | 2680 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa | 2681 | |
49f693fa DC |
2682 | /* |
2683 | * Check and set flags if this segment has a left neighbor. | |
2684 | */ | |
9788e059 | 2685 | if (xfs_iext_peek_prev_extent(ifp, icur, &left)) { |
49f693fa | 2686 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
2687 | if (isnullstartblock(left.br_startblock)) |
2688 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2689 | } |
2bd0ea18 | 2690 | |
49f693fa DC |
2691 | /* |
2692 | * Check and set flags if this segment has a current value. | |
2693 | * Not true if we're inserting into the "hole" at eof. | |
2694 | */ | |
9788e059 | 2695 | if (xfs_iext_get_extent(ifp, icur, &right)) { |
49f693fa | 2696 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2697 | if (isnullstartblock(right.br_startblock)) |
2698 | state |= BMAP_RIGHT_DELAY; | |
2bd0ea18 | 2699 | } |
2bd0ea18 | 2700 | |
49f693fa DC |
2701 | /* |
2702 | * We're inserting a real allocation between "left" and "right". | |
2703 | * Set the contiguity flags. Don't let extents get too large. | |
2704 | */ | |
2705 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2706 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2707 | left.br_startblock + left.br_blockcount == new->br_startblock && | |
2708 | left.br_state == new->br_state && | |
2709 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2710 | state |= BMAP_LEFT_CONTIG; | |
57c9fccb | 2711 | |
49f693fa DC |
2712 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2713 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2714 | new->br_startblock + new->br_blockcount == right.br_startblock && | |
2715 | new->br_state == right.br_state && | |
2716 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2717 | (!(state & BMAP_LEFT_CONTIG) || | |
2718 | left.br_blockcount + new->br_blockcount + | |
2719 | right.br_blockcount <= MAXEXTLEN)) | |
2720 | state |= BMAP_RIGHT_CONTIG; | |
ca86e759 | 2721 | |
49f693fa DC |
2722 | error = 0; |
2723 | /* | |
2724 | * Select which case we're in here, and implement it. | |
2725 | */ | |
2726 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2727 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
57c9fccb | 2728 | /* |
49f693fa DC |
2729 | * New allocation is contiguous with real allocations on the |
2730 | * left and on the right. | |
2731 | * Merge all three into a single extent record. | |
57c9fccb | 2732 | */ |
3281eb91 | 2733 | left.br_blockcount += new->br_blockcount + right.br_blockcount; |
56b2de80 | 2734 | |
cf455a62 | 2735 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2736 | xfs_iext_prev(ifp, icur); |
2737 | xfs_iext_update_extent(ip, state, icur, &left); | |
56b2de80 | 2738 | |
972432f2 CH |
2739 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2740 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
2741 | if (cur == NULL) { | |
49f693fa DC |
2742 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2743 | } else { | |
2744 | rval = XFS_ILOG_CORE; | |
70a93110 | 2745 | error = xfs_bmbt_lookup_eq(cur, &right, &i); |
49f693fa DC |
2746 | if (error) |
2747 | goto done; | |
19ebedcf | 2748 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
972432f2 | 2749 | error = xfs_btree_delete(cur, &i); |
49f693fa DC |
2750 | if (error) |
2751 | goto done; | |
19ebedcf | 2752 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
972432f2 | 2753 | error = xfs_btree_decrement(cur, 0, &i); |
49f693fa DC |
2754 | if (error) |
2755 | goto done; | |
19ebedcf | 2756 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2757 | error = xfs_bmbt_update(cur, &left); |
49f693fa DC |
2758 | if (error) |
2759 | goto done; | |
2760 | } | |
57c9fccb | 2761 | break; |
49f693fa DC |
2762 | |
2763 | case BMAP_LEFT_CONTIG: | |
2764 | /* | |
2765 | * New allocation is contiguous with a real allocation | |
2766 | * on the left. | |
2767 | * Merge the new allocation with the left neighbor. | |
2768 | */ | |
3281eb91 | 2769 | old = left; |
3281eb91 | 2770 | left.br_blockcount += new->br_blockcount; |
501dd276 | 2771 | |
9788e059 CH |
2772 | xfs_iext_prev(ifp, icur); |
2773 | xfs_iext_update_extent(ip, state, icur, &left); | |
49f693fa | 2774 | |
972432f2 | 2775 | if (cur == NULL) { |
49f693fa DC |
2776 | rval = xfs_ilog_fext(whichfork); |
2777 | } else { | |
2778 | rval = 0; | |
70a93110 | 2779 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
49f693fa DC |
2780 | if (error) |
2781 | goto done; | |
19ebedcf | 2782 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2783 | error = xfs_bmbt_update(cur, &left); |
49f693fa DC |
2784 | if (error) |
2785 | goto done; | |
2786 | } | |
57c9fccb | 2787 | break; |
49f693fa DC |
2788 | |
2789 | case BMAP_RIGHT_CONTIG: | |
2790 | /* | |
2791 | * New allocation is contiguous with a real allocation | |
2792 | * on the right. | |
2793 | * Merge the new allocation with the right neighbor. | |
2794 | */ | |
3281eb91 | 2795 | old = right; |
df926c07 | 2796 | |
3281eb91 CH |
2797 | right.br_startoff = new->br_startoff; |
2798 | right.br_startblock = new->br_startblock; | |
2799 | right.br_blockcount += new->br_blockcount; | |
9788e059 | 2800 | xfs_iext_update_extent(ip, state, icur, &right); |
49f693fa | 2801 | |
972432f2 | 2802 | if (cur == NULL) { |
49f693fa DC |
2803 | rval = xfs_ilog_fext(whichfork); |
2804 | } else { | |
2805 | rval = 0; | |
70a93110 | 2806 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
49f693fa DC |
2807 | if (error) |
2808 | goto done; | |
19ebedcf | 2809 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2810 | error = xfs_bmbt_update(cur, &right); |
49f693fa DC |
2811 | if (error) |
2812 | goto done; | |
2813 | } | |
2814 | break; | |
2815 | ||
2816 | case 0: | |
2817 | /* | |
2818 | * New allocation is not contiguous with another | |
2819 | * real allocation. | |
2820 | * Insert a new entry. | |
2821 | */ | |
26a75f67 | 2822 | xfs_iext_insert(ip, icur, new, state); |
972432f2 CH |
2823 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2824 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
2825 | if (cur == NULL) { | |
49f693fa DC |
2826 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2827 | } else { | |
2828 | rval = XFS_ILOG_CORE; | |
70a93110 | 2829 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
49f693fa DC |
2830 | if (error) |
2831 | goto done; | |
19ebedcf | 2832 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
972432f2 | 2833 | error = xfs_btree_insert(cur, &i); |
49f693fa DC |
2834 | if (error) |
2835 | goto done; | |
19ebedcf | 2836 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 2837 | } |
57c9fccb | 2838 | break; |
57c9fccb | 2839 | } |
a2ceac1f | 2840 | |
d7f80320 | 2841 | /* add reverse mapping */ |
972432f2 | 2842 | error = xfs_rmap_map_extent(mp, dfops, ip, whichfork, new); |
d7f80320 DW |
2843 | if (error) |
2844 | goto done; | |
2845 | ||
49f693fa | 2846 | /* convert to a btree if necessary */ |
972432f2 | 2847 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
49f693fa | 2848 | int tmp_logflags; /* partial log flag return val */ |
3f853c7a | 2849 | |
972432f2 CH |
2850 | ASSERT(cur == NULL); |
2851 | error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, curp, | |
49f693fa | 2852 | 0, &tmp_logflags, whichfork); |
972432f2 CH |
2853 | *logflagsp |= tmp_logflags; |
2854 | cur = *curp; | |
49f693fa DC |
2855 | if (error) |
2856 | goto done; | |
57c9fccb | 2857 | } |
a2ceac1f | 2858 | |
49f693fa | 2859 | /* clear out the allocated field, done with it now in any case. */ |
972432f2 CH |
2860 | if (cur) |
2861 | cur->bc_private.b.allocated = 0; | |
49f693fa | 2862 | |
972432f2 | 2863 | xfs_bmap_check_leaf_extents(cur, ip, whichfork); |
49f693fa | 2864 | done: |
972432f2 | 2865 | *logflagsp |= rval; |
57c9fccb NS |
2866 | return error; |
2867 | } | |
2868 | ||
2bd0ea18 | 2869 | /* |
49f693fa | 2870 | * Functions used in the extent read, allocate and remove paths |
2bd0ea18 | 2871 | */ |
2bd0ea18 | 2872 | |
5000d01d | 2873 | /* |
49f693fa | 2874 | * Adjust the size of the new extent based on di_extsize and rt extsize. |
2bd0ea18 | 2875 | */ |
613e6057 | 2876 | int |
49f693fa DC |
2877 | xfs_bmap_extsize_align( |
2878 | xfs_mount_t *mp, | |
2879 | xfs_bmbt_irec_t *gotp, /* next extent pointer */ | |
2880 | xfs_bmbt_irec_t *prevp, /* previous extent pointer */ | |
2881 | xfs_extlen_t extsz, /* align to this extent size */ | |
2882 | int rt, /* is this a realtime inode? */ | |
2883 | int eof, /* is extent at end-of-file? */ | |
2884 | int delay, /* creating delalloc extent? */ | |
2885 | int convert, /* overwriting unwritten extent? */ | |
2886 | xfs_fileoff_t *offp, /* in/out: aligned offset */ | |
2887 | xfs_extlen_t *lenp) /* in/out: aligned length */ | |
2bd0ea18 | 2888 | { |
49f693fa DC |
2889 | xfs_fileoff_t orig_off; /* original offset */ |
2890 | xfs_extlen_t orig_alen; /* original length */ | |
2891 | xfs_fileoff_t orig_end; /* original off+len */ | |
2892 | xfs_fileoff_t nexto; /* next file offset */ | |
2893 | xfs_fileoff_t prevo; /* previous file offset */ | |
2894 | xfs_fileoff_t align_off; /* temp for offset */ | |
2895 | xfs_extlen_t align_alen; /* temp for length */ | |
2896 | xfs_extlen_t temp; /* temp for calculations */ | |
2897 | ||
2898 | if (convert) | |
2899 | return 0; | |
2900 | ||
2901 | orig_off = align_off = *offp; | |
2902 | orig_alen = align_alen = *lenp; | |
2903 | orig_end = orig_off + orig_alen; | |
2bd0ea18 NS |
2904 | |
2905 | /* | |
49f693fa DC |
2906 | * If this request overlaps an existing extent, then don't |
2907 | * attempt to perform any additional alignment. | |
2bd0ea18 | 2908 | */ |
49f693fa DC |
2909 | if (!delay && !eof && |
2910 | (orig_off >= gotp->br_startoff) && | |
2911 | (orig_end <= gotp->br_startoff + gotp->br_blockcount)) { | |
2912 | return 0; | |
2bd0ea18 | 2913 | } |
57c9fccb | 2914 | |
49f693fa DC |
2915 | /* |
2916 | * If the file offset is unaligned vs. the extent size | |
2917 | * we need to align it. This will be possible unless | |
2918 | * the file was previously written with a kernel that didn't | |
2919 | * perform this alignment, or if a truncate shot us in the | |
2920 | * foot. | |
2921 | */ | |
2922 | temp = do_mod(orig_off, extsz); | |
2923 | if (temp) { | |
2924 | align_alen += temp; | |
2925 | align_off -= temp; | |
2926 | } | |
7cc23f0c DC |
2927 | |
2928 | /* Same adjustment for the end of the requested area. */ | |
2929 | temp = (align_alen % extsz); | |
2930 | if (temp) | |
2931 | align_alen += extsz - temp; | |
2932 | ||
49f693fa | 2933 | /* |
7cc23f0c DC |
2934 | * For large extent hint sizes, the aligned extent might be larger than |
2935 | * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls | |
2936 | * the length back under MAXEXTLEN. The outer allocation loops handle | |
2937 | * short allocation just fine, so it is safe to do this. We only want to | |
2938 | * do it when we are forced to, though, because it means more allocation | |
2939 | * operations are required. | |
49f693fa | 2940 | */ |
7cc23f0c DC |
2941 | while (align_alen > MAXEXTLEN) |
2942 | align_alen -= extsz; | |
2943 | ASSERT(align_alen <= MAXEXTLEN); | |
2944 | ||
49f693fa DC |
2945 | /* |
2946 | * If the previous block overlaps with this proposed allocation | |
2947 | * then move the start forward without adjusting the length. | |
2948 | */ | |
2949 | if (prevp->br_startoff != NULLFILEOFF) { | |
2950 | if (prevp->br_startblock == HOLESTARTBLOCK) | |
2951 | prevo = prevp->br_startoff; | |
2952 | else | |
2953 | prevo = prevp->br_startoff + prevp->br_blockcount; | |
2954 | } else | |
2955 | prevo = 0; | |
2956 | if (align_off != orig_off && align_off < prevo) | |
2957 | align_off = prevo; | |
2958 | /* | |
2959 | * If the next block overlaps with this proposed allocation | |
2960 | * then move the start back without adjusting the length, | |
2961 | * but not before offset 0. | |
2962 | * This may of course make the start overlap previous block, | |
2963 | * and if we hit the offset 0 limit then the next block | |
2964 | * can still overlap too. | |
2965 | */ | |
2966 | if (!eof && gotp->br_startoff != NULLFILEOFF) { | |
2967 | if ((delay && gotp->br_startblock == HOLESTARTBLOCK) || | |
2968 | (!delay && gotp->br_startblock == DELAYSTARTBLOCK)) | |
2969 | nexto = gotp->br_startoff + gotp->br_blockcount; | |
2970 | else | |
2971 | nexto = gotp->br_startoff; | |
2972 | } else | |
2973 | nexto = NULLFILEOFF; | |
2974 | if (!eof && | |
2975 | align_off + align_alen != orig_end && | |
2976 | align_off + align_alen > nexto) | |
2977 | align_off = nexto > align_alen ? nexto - align_alen : 0; | |
2978 | /* | |
2979 | * If we're now overlapping the next or previous extent that | |
2980 | * means we can't fit an extsz piece in this hole. Just move | |
2981 | * the start forward to the first valid spot and set | |
2982 | * the length so we hit the end. | |
2983 | */ | |
2984 | if (align_off != orig_off && align_off < prevo) | |
2985 | align_off = prevo; | |
2986 | if (align_off + align_alen != orig_end && | |
2987 | align_off + align_alen > nexto && | |
2988 | nexto != NULLFILEOFF) { | |
2989 | ASSERT(nexto > prevo); | |
2990 | align_alen = nexto - align_off; | |
57c9fccb | 2991 | } |
2bd0ea18 | 2992 | |
49f693fa DC |
2993 | /* |
2994 | * If realtime, and the result isn't a multiple of the realtime | |
2995 | * extent size we need to remove blocks until it is. | |
2996 | */ | |
2997 | if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) { | |
2bd0ea18 | 2998 | /* |
49f693fa DC |
2999 | * We're not covering the original request, or |
3000 | * we won't be able to once we fix the length. | |
2bd0ea18 | 3001 | */ |
49f693fa DC |
3002 | if (orig_off < align_off || |
3003 | orig_end > align_off + align_alen || | |
3004 | align_alen - temp < orig_alen) | |
12b53197 | 3005 | return -EINVAL; |
49f693fa DC |
3006 | /* |
3007 | * Try to fix it by moving the start up. | |
3008 | */ | |
3009 | if (align_off + temp <= orig_off) { | |
3010 | align_alen -= temp; | |
3011 | align_off += temp; | |
2bd0ea18 | 3012 | } |
49f693fa DC |
3013 | /* |
3014 | * Try to fix it by moving the end in. | |
3015 | */ | |
3016 | else if (align_off + align_alen - temp >= orig_end) | |
3017 | align_alen -= temp; | |
3018 | /* | |
3019 | * Set the start to the minimum then trim the length. | |
3020 | */ | |
3021 | else { | |
3022 | align_alen -= orig_off - align_off; | |
3023 | align_off = orig_off; | |
3024 | align_alen -= align_alen % mp->m_sb.sb_rextsize; | |
3025 | } | |
3026 | /* | |
3027 | * Result doesn't cover the request, fail it. | |
3028 | */ | |
3029 | if (orig_off < align_off || orig_end > align_off + align_alen) | |
12b53197 | 3030 | return -EINVAL; |
49f693fa DC |
3031 | } else { |
3032 | ASSERT(orig_off >= align_off); | |
7cc23f0c DC |
3033 | /* see MAXEXTLEN handling above */ |
3034 | ASSERT(orig_end <= align_off + align_alen || | |
3035 | align_alen + extsz > MAXEXTLEN); | |
2bd0ea18 | 3036 | } |
49f693fa DC |
3037 | |
3038 | #ifdef DEBUG | |
3039 | if (!eof && gotp->br_startoff != NULLFILEOFF) | |
3040 | ASSERT(align_off + align_alen <= gotp->br_startoff); | |
3041 | if (prevp->br_startoff != NULLFILEOFF) | |
3042 | ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount); | |
3043 | #endif | |
3044 | ||
3045 | *lenp = align_alen; | |
3046 | *offp = align_off; | |
2bd0ea18 NS |
3047 | return 0; |
3048 | } | |
3049 | ||
49f693fa DC |
3050 | #define XFS_ALLOC_GAP_UNITS 4 |
3051 | ||
613e6057 | 3052 | void |
49f693fa | 3053 | xfs_bmap_adjacent( |
613e6057 | 3054 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
2bd0ea18 | 3055 | { |
49f693fa DC |
3056 | xfs_fsblock_t adjust; /* adjustment to block numbers */ |
3057 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ | |
3058 | xfs_mount_t *mp; /* mount point structure */ | |
3059 | int nullfb; /* true if ap->firstblock isn't set */ | |
3060 | int rt; /* true if inode is realtime */ | |
2bd0ea18 | 3061 | |
49f693fa DC |
3062 | #define ISVALID(x,y) \ |
3063 | (rt ? \ | |
3064 | (x) < mp->m_sb.sb_rblocks : \ | |
3065 | XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \ | |
3066 | XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \ | |
3067 | XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks) | |
3068 | ||
3069 | mp = ap->ip->i_mount; | |
3070 | nullfb = *ap->firstblock == NULLFSBLOCK; | |
1fccd5c8 DC |
3071 | rt = XFS_IS_REALTIME_INODE(ap->ip) && |
3072 | xfs_alloc_is_userdata(ap->datatype); | |
49f693fa DC |
3073 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock); |
3074 | /* | |
3075 | * If allocating at eof, and there's a previous real block, | |
3076 | * try to use its last block as our starting point. | |
3077 | */ | |
3078 | if (ap->eof && ap->prev.br_startoff != NULLFILEOFF && | |
3079 | !isnullstartblock(ap->prev.br_startblock) && | |
3080 | ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount, | |
3081 | ap->prev.br_startblock)) { | |
3082 | ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount; | |
3083 | /* | |
3084 | * Adjust for the gap between prevp and us. | |
3085 | */ | |
3086 | adjust = ap->offset - | |
3087 | (ap->prev.br_startoff + ap->prev.br_blockcount); | |
3088 | if (adjust && | |
3089 | ISVALID(ap->blkno + adjust, ap->prev.br_startblock)) | |
3090 | ap->blkno += adjust; | |
2bd0ea18 | 3091 | } |
49f693fa DC |
3092 | /* |
3093 | * If not at eof, then compare the two neighbor blocks. | |
3094 | * Figure out whether either one gives us a good starting point, | |
3095 | * and pick the better one. | |
3096 | */ | |
3097 | else if (!ap->eof) { | |
3098 | xfs_fsblock_t gotbno; /* right side block number */ | |
3099 | xfs_fsblock_t gotdiff=0; /* right side difference */ | |
3100 | xfs_fsblock_t prevbno; /* left side block number */ | |
3101 | xfs_fsblock_t prevdiff=0; /* left side difference */ | |
3102 | ||
3103 | /* | |
3104 | * If there's a previous (left) block, select a requested | |
3105 | * start block based on it. | |
3106 | */ | |
3107 | if (ap->prev.br_startoff != NULLFILEOFF && | |
3108 | !isnullstartblock(ap->prev.br_startblock) && | |
3109 | (prevbno = ap->prev.br_startblock + | |
3110 | ap->prev.br_blockcount) && | |
3111 | ISVALID(prevbno, ap->prev.br_startblock)) { | |
3112 | /* | |
3113 | * Calculate gap to end of previous block. | |
3114 | */ | |
3115 | adjust = prevdiff = ap->offset - | |
3116 | (ap->prev.br_startoff + | |
3117 | ap->prev.br_blockcount); | |
3118 | /* | |
3119 | * Figure the startblock based on the previous block's | |
3120 | * end and the gap size. | |
3121 | * Heuristic! | |
3122 | * If the gap is large relative to the piece we're | |
3123 | * allocating, or using it gives us an invalid block | |
3124 | * number, then just use the end of the previous block. | |
3125 | */ | |
3126 | if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3127 | ISVALID(prevbno + prevdiff, | |
3128 | ap->prev.br_startblock)) | |
3129 | prevbno += adjust; | |
3130 | else | |
3131 | prevdiff += adjust; | |
3132 | /* | |
3133 | * If the firstblock forbids it, can't use it, | |
3134 | * must use default. | |
3135 | */ | |
3136 | if (!rt && !nullfb && | |
3137 | XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno) | |
3138 | prevbno = NULLFSBLOCK; | |
3139 | } | |
3140 | /* | |
3141 | * No previous block or can't follow it, just default. | |
3142 | */ | |
3143 | else | |
3144 | prevbno = NULLFSBLOCK; | |
3145 | /* | |
3146 | * If there's a following (right) block, select a requested | |
3147 | * start block based on it. | |
3148 | */ | |
3149 | if (!isnullstartblock(ap->got.br_startblock)) { | |
3150 | /* | |
3151 | * Calculate gap to start of next block. | |
3152 | */ | |
3153 | adjust = gotdiff = ap->got.br_startoff - ap->offset; | |
3154 | /* | |
3155 | * Figure the startblock based on the next block's | |
3156 | * start and the gap size. | |
3157 | */ | |
3158 | gotbno = ap->got.br_startblock; | |
3159 | /* | |
3160 | * Heuristic! | |
3161 | * If the gap is large relative to the piece we're | |
3162 | * allocating, or using it gives us an invalid block | |
3163 | * number, then just use the start of the next block | |
3164 | * offset by our length. | |
3165 | */ | |
3166 | if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3167 | ISVALID(gotbno - gotdiff, gotbno)) | |
3168 | gotbno -= adjust; | |
3169 | else if (ISVALID(gotbno - ap->length, gotbno)) { | |
3170 | gotbno -= ap->length; | |
3171 | gotdiff += adjust - ap->length; | |
3172 | } else | |
3173 | gotdiff += adjust; | |
3174 | /* | |
3175 | * If the firstblock forbids it, can't use it, | |
3176 | * must use default. | |
3177 | */ | |
3178 | if (!rt && !nullfb && | |
3179 | XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno) | |
3180 | gotbno = NULLFSBLOCK; | |
3181 | } | |
3182 | /* | |
3183 | * No next block, just default. | |
3184 | */ | |
2bd0ea18 | 3185 | else |
49f693fa DC |
3186 | gotbno = NULLFSBLOCK; |
3187 | /* | |
3188 | * If both valid, pick the better one, else the only good | |
3189 | * one, else ap->blkno is already set (to 0 or the inode block). | |
3190 | */ | |
3191 | if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) | |
3192 | ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno; | |
3193 | else if (prevbno != NULLFSBLOCK) | |
3194 | ap->blkno = prevbno; | |
3195 | else if (gotbno != NULLFSBLOCK) | |
3196 | ap->blkno = gotbno; | |
a2ceac1f | 3197 | } |
49f693fa | 3198 | #undef ISVALID |
a2ceac1f DC |
3199 | } |
3200 | ||
ff105f75 DC |
3201 | static int |
3202 | xfs_bmap_longest_free_extent( | |
3203 | struct xfs_trans *tp, | |
3204 | xfs_agnumber_t ag, | |
3205 | xfs_extlen_t *blen, | |
3206 | int *notinit) | |
3207 | { | |
3208 | struct xfs_mount *mp = tp->t_mountp; | |
3209 | struct xfs_perag *pag; | |
3210 | xfs_extlen_t longest; | |
3211 | int error = 0; | |
3212 | ||
3213 | pag = xfs_perag_get(mp, ag); | |
3214 | if (!pag->pagf_init) { | |
3215 | error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK); | |
3216 | if (error) | |
3217 | goto out; | |
3218 | ||
3219 | if (!pag->pagf_init) { | |
3220 | *notinit = 1; | |
3221 | goto out; | |
3222 | } | |
3223 | } | |
3224 | ||
1421de38 | 3225 | longest = xfs_alloc_longest_free_extent(pag, |
cf8ce220 DW |
3226 | xfs_alloc_min_freelist(mp, pag), |
3227 | xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE)); | |
ff105f75 DC |
3228 | if (*blen < longest) |
3229 | *blen = longest; | |
3230 | ||
3231 | out: | |
3232 | xfs_perag_put(pag); | |
3233 | return error; | |
3234 | } | |
3235 | ||
3236 | static void | |
3237 | xfs_bmap_select_minlen( | |
3238 | struct xfs_bmalloca *ap, | |
3239 | struct xfs_alloc_arg *args, | |
3240 | xfs_extlen_t *blen, | |
3241 | int notinit) | |
3242 | { | |
3243 | if (notinit || *blen < ap->minlen) { | |
3244 | /* | |
3245 | * Since we did a BUF_TRYLOCK above, it is possible that | |
3246 | * there is space for this request. | |
3247 | */ | |
3248 | args->minlen = ap->minlen; | |
3249 | } else if (*blen < args->maxlen) { | |
3250 | /* | |
3251 | * If the best seen length is less than the request length, | |
3252 | * use the best as the minimum. | |
3253 | */ | |
3254 | args->minlen = *blen; | |
3255 | } else { | |
3256 | /* | |
3257 | * Otherwise we've seen an extent as big as maxlen, use that | |
3258 | * as the minimum. | |
3259 | */ | |
3260 | args->minlen = args->maxlen; | |
3261 | } | |
3262 | } | |
3263 | ||
a2ceac1f | 3264 | STATIC int |
49f693fa DC |
3265 | xfs_bmap_btalloc_nullfb( |
3266 | struct xfs_bmalloca *ap, | |
3267 | struct xfs_alloc_arg *args, | |
3268 | xfs_extlen_t *blen) | |
a2ceac1f | 3269 | { |
49f693fa | 3270 | struct xfs_mount *mp = ap->ip->i_mount; |
49f693fa DC |
3271 | xfs_agnumber_t ag, startag; |
3272 | int notinit = 0; | |
a2ceac1f DC |
3273 | int error; |
3274 | ||
ff105f75 | 3275 | args->type = XFS_ALLOCTYPE_START_BNO; |
49f693fa | 3276 | args->total = ap->total; |
a2ceac1f | 3277 | |
49f693fa DC |
3278 | startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno); |
3279 | if (startag == NULLAGNUMBER) | |
3280 | startag = ag = 0; | |
a2ceac1f | 3281 | |
49f693fa | 3282 | while (*blen < args->maxlen) { |
ff105f75 DC |
3283 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, |
3284 | ¬init); | |
3285 | if (error) | |
3286 | return error; | |
a2ceac1f | 3287 | |
49f693fa DC |
3288 | if (++ag == mp->m_sb.sb_agcount) |
3289 | ag = 0; | |
3290 | if (ag == startag) | |
3291 | break; | |
49f693fa | 3292 | } |
2bd0ea18 | 3293 | |
ff105f75 DC |
3294 | xfs_bmap_select_minlen(ap, args, blen, notinit); |
3295 | return 0; | |
3296 | } | |
3297 | ||
3298 | STATIC int | |
3299 | xfs_bmap_btalloc_filestreams( | |
3300 | struct xfs_bmalloca *ap, | |
3301 | struct xfs_alloc_arg *args, | |
3302 | xfs_extlen_t *blen) | |
3303 | { | |
3304 | struct xfs_mount *mp = ap->ip->i_mount; | |
3305 | xfs_agnumber_t ag; | |
3306 | int notinit = 0; | |
3307 | int error; | |
3308 | ||
3309 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
3310 | args->total = ap->total; | |
3311 | ||
3312 | ag = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3313 | if (ag == NULLAGNUMBER) | |
3314 | ag = 0; | |
3315 | ||
3316 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init); | |
3317 | if (error) | |
3318 | return error; | |
3319 | ||
3320 | if (*blen < args->maxlen) { | |
3321 | error = xfs_filestream_new_ag(ap, &ag); | |
3322 | if (error) | |
3323 | return error; | |
3324 | ||
3325 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, | |
3326 | ¬init); | |
3327 | if (error) | |
3328 | return error; | |
3329 | ||
3330 | } | |
3331 | ||
3332 | xfs_bmap_select_minlen(ap, args, blen, notinit); | |
2bd0ea18 | 3333 | |
49f693fa | 3334 | /* |
ff105f75 DC |
3335 | * Set the failure fallback case to look in the selected AG as stream |
3336 | * may have moved. | |
49f693fa | 3337 | */ |
ff105f75 | 3338 | ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0); |
49f693fa | 3339 | return 0; |
2bd0ea18 NS |
3340 | } |
3341 | ||
3cb68815 DW |
3342 | /* Update all inode and quota accounting for the allocation we just did. */ |
3343 | static void | |
3344 | xfs_bmap_btalloc_accounting( | |
3345 | struct xfs_bmalloca *ap, | |
3346 | struct xfs_alloc_arg *args) | |
3347 | { | |
d07cc724 DW |
3348 | if (ap->flags & XFS_BMAPI_COWFORK) { |
3349 | /* | |
3350 | * COW fork blocks are in-core only and thus are treated as | |
3351 | * in-core quota reservation (like delalloc blocks) even when | |
3352 | * converted to real blocks. The quota reservation is not | |
3353 | * accounted to disk until blocks are remapped to the data | |
3354 | * fork. So if these blocks were previously delalloc, we | |
3355 | * already have quota reservation and there's nothing to do | |
3356 | * yet. | |
3357 | */ | |
3358 | if (ap->wasdel) | |
3359 | return; | |
3360 | ||
3361 | /* | |
3362 | * Otherwise, we've allocated blocks in a hole. The transaction | |
3363 | * has acquired in-core quota reservation for this extent. | |
3364 | * Rather than account these as real blocks, however, we reduce | |
3365 | * the transaction quota reservation based on the allocation. | |
3366 | * This essentially transfers the transaction quota reservation | |
3367 | * to that of a delalloc extent. | |
3368 | */ | |
3369 | ap->ip->i_delayed_blks += args->len; | |
3370 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS, | |
3371 | -(long)args->len); | |
3372 | return; | |
3373 | } | |
3374 | ||
3375 | /* data/attr fork only */ | |
3376 | ap->ip->i_d.di_nblocks += args->len; | |
3cb68815 DW |
3377 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); |
3378 | if (ap->wasdel) | |
3379 | ap->ip->i_delayed_blks -= args->len; | |
3380 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, | |
3381 | ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT, | |
3382 | args->len); | |
3383 | } | |
3384 | ||
b3563c19 | 3385 | STATIC int |
49f693fa | 3386 | xfs_bmap_btalloc( |
613e6057 | 3387 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
b3563c19 | 3388 | { |
49f693fa DC |
3389 | xfs_mount_t *mp; /* mount point structure */ |
3390 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
1fccd5c8 | 3391 | xfs_extlen_t align = 0; /* minimum allocation alignment */ |
49f693fa DC |
3392 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ |
3393 | xfs_agnumber_t ag; | |
3394 | xfs_alloc_arg_t args; | |
c38464ff DW |
3395 | xfs_fileoff_t orig_offset; |
3396 | xfs_extlen_t orig_length; | |
49f693fa DC |
3397 | xfs_extlen_t blen; |
3398 | xfs_extlen_t nextminlen = 0; | |
3399 | int nullfb; /* true if ap->firstblock isn't set */ | |
3400 | int isaligned; | |
3401 | int tryagain; | |
3402 | int error; | |
ff105f75 | 3403 | int stripe_align; |
b3563c19 | 3404 | |
49f693fa | 3405 | ASSERT(ap->length); |
c38464ff DW |
3406 | orig_offset = ap->offset; |
3407 | orig_length = ap->length; | |
b3563c19 | 3408 | |
49f693fa | 3409 | mp = ap->ip->i_mount; |
ff105f75 DC |
3410 | |
3411 | /* stripe alignment for allocation is determined by mount parameters */ | |
3412 | stripe_align = 0; | |
3413 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
3414 | stripe_align = mp->m_swidth; | |
3415 | else if (mp->m_dalign) | |
3416 | stripe_align = mp->m_dalign; | |
3417 | ||
10460994 DW |
3418 | if (ap->flags & XFS_BMAPI_COWFORK) |
3419 | align = xfs_get_cowextsz_hint(ap->ip); | |
3420 | else if (xfs_alloc_is_userdata(ap->datatype)) | |
1fccd5c8 | 3421 | align = xfs_get_extsz_hint(ap->ip); |
3f5dd6c6 | 3422 | if (align) { |
49f693fa DC |
3423 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, |
3424 | align, 0, ap->eof, 0, ap->conv, | |
3425 | &ap->offset, &ap->length); | |
3426 | ASSERT(!error); | |
3427 | ASSERT(ap->length); | |
3428 | } | |
ff105f75 DC |
3429 | |
3430 | ||
49f693fa DC |
3431 | nullfb = *ap->firstblock == NULLFSBLOCK; |
3432 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock); | |
3433 | if (nullfb) { | |
1fccd5c8 DC |
3434 | if (xfs_alloc_is_userdata(ap->datatype) && |
3435 | xfs_inode_is_filestream(ap->ip)) { | |
49f693fa DC |
3436 | ag = xfs_filestream_lookup_ag(ap->ip); |
3437 | ag = (ag != NULLAGNUMBER) ? ag : 0; | |
3438 | ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0); | |
3439 | } else { | |
3440 | ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino); | |
3441 | } | |
3442 | } else | |
3443 | ap->blkno = *ap->firstblock; | |
3444 | ||
3445 | xfs_bmap_adjacent(ap); | |
2bd0ea18 | 3446 | |
2bd0ea18 | 3447 | /* |
49f693fa DC |
3448 | * If allowed, use ap->blkno; otherwise must use firstblock since |
3449 | * it's in the right allocation group. | |
2bd0ea18 | 3450 | */ |
49f693fa DC |
3451 | if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno) |
3452 | ; | |
3453 | else | |
3454 | ap->blkno = *ap->firstblock; | |
2bd0ea18 | 3455 | /* |
49f693fa | 3456 | * Normal allocation, done through xfs_alloc_vextent. |
2bd0ea18 | 3457 | */ |
49f693fa DC |
3458 | tryagain = isaligned = 0; |
3459 | memset(&args, 0, sizeof(args)); | |
3460 | args.tp = ap->tp; | |
3461 | args.mp = mp; | |
3462 | args.fsbno = ap->blkno; | |
85aec44f | 3463 | xfs_rmap_skip_owner_update(&args.oinfo); |
49f693fa DC |
3464 | |
3465 | /* Trim the allocation back to the maximum an AG can fit. */ | |
b8a8d6e5 | 3466 | args.maxlen = MIN(ap->length, mp->m_ag_max_usable); |
49f693fa DC |
3467 | args.firstblock = *ap->firstblock; |
3468 | blen = 0; | |
3469 | if (nullfb) { | |
ff105f75 DC |
3470 | /* |
3471 | * Search for an allocation group with a single extent large | |
3472 | * enough for the request. If one isn't found, then adjust | |
3473 | * the minimum allocation size to the largest space found. | |
3474 | */ | |
1fccd5c8 DC |
3475 | if (xfs_alloc_is_userdata(ap->datatype) && |
3476 | xfs_inode_is_filestream(ap->ip)) | |
ff105f75 DC |
3477 | error = xfs_bmap_btalloc_filestreams(ap, &args, &blen); |
3478 | else | |
3479 | error = xfs_bmap_btalloc_nullfb(ap, &args, &blen); | |
a2ceac1f | 3480 | if (error) |
2bd0ea18 | 3481 | return error; |
f33cea1a | 3482 | } else if (ap->dfops->dop_low) { |
49f693fa DC |
3483 | if (xfs_inode_is_filestream(ap->ip)) |
3484 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3485 | else | |
3486 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3487 | args.total = args.minlen = ap->minlen; | |
3488 | } else { | |
3489 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3490 | args.total = ap->total; | |
3491 | args.minlen = ap->minlen; | |
3492 | } | |
3493 | /* apply extent size hints if obtained earlier */ | |
3f5dd6c6 | 3494 | if (align) { |
49f693fa DC |
3495 | args.prod = align; |
3496 | if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod))) | |
3497 | args.mod = (xfs_extlen_t)(args.prod - args.mod); | |
b2327e1a | 3498 | } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) { |
49f693fa DC |
3499 | args.prod = 1; |
3500 | args.mod = 0; | |
3501 | } else { | |
b2327e1a | 3502 | args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog; |
49f693fa DC |
3503 | if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod)))) |
3504 | args.mod = (xfs_extlen_t)(args.prod - args.mod); | |
2bd0ea18 NS |
3505 | } |
3506 | /* | |
49f693fa DC |
3507 | * If we are not low on available data blocks, and the |
3508 | * underlying logical volume manager is a stripe, and | |
3509 | * the file offset is zero then try to allocate data | |
3510 | * blocks on stripe unit boundary. | |
3511 | * NOTE: ap->aeof is only set if the allocation length | |
3512 | * is >= the stripe unit and the allocation offset is | |
3513 | * at the end of file. | |
2bd0ea18 | 3514 | */ |
f33cea1a | 3515 | if (!ap->dfops->dop_low && ap->aeof) { |
49f693fa | 3516 | if (!ap->offset) { |
ff105f75 | 3517 | args.alignment = stripe_align; |
49f693fa DC |
3518 | atype = args.type; |
3519 | isaligned = 1; | |
3520 | /* | |
3521 | * Adjust for alignment | |
3522 | */ | |
3523 | if (blen > args.alignment && blen <= args.maxlen) | |
3524 | args.minlen = blen - args.alignment; | |
3525 | args.minalignslop = 0; | |
3526 | } else { | |
3527 | /* | |
3528 | * First try an exact bno allocation. | |
3529 | * If it fails then do a near or start bno | |
3530 | * allocation with alignment turned on. | |
3531 | */ | |
3532 | atype = args.type; | |
3533 | tryagain = 1; | |
3534 | args.type = XFS_ALLOCTYPE_THIS_BNO; | |
3535 | args.alignment = 1; | |
3536 | /* | |
3537 | * Compute the minlen+alignment for the | |
3538 | * next case. Set slop so that the value | |
3539 | * of minlen+alignment+slop doesn't go up | |
3540 | * between the calls. | |
3541 | */ | |
ff105f75 DC |
3542 | if (blen > stripe_align && blen <= args.maxlen) |
3543 | nextminlen = blen - stripe_align; | |
49f693fa DC |
3544 | else |
3545 | nextminlen = args.minlen; | |
ff105f75 | 3546 | if (nextminlen + stripe_align > args.minlen + 1) |
49f693fa | 3547 | args.minalignslop = |
ff105f75 | 3548 | nextminlen + stripe_align - |
49f693fa DC |
3549 | args.minlen - 1; |
3550 | else | |
3551 | args.minalignslop = 0; | |
2bd0ea18 | 3552 | } |
49f693fa DC |
3553 | } else { |
3554 | args.alignment = 1; | |
3555 | args.minalignslop = 0; | |
3556 | } | |
3557 | args.minleft = ap->minleft; | |
3558 | args.wasdel = ap->wasdel; | |
cf8ce220 | 3559 | args.resv = XFS_AG_RESV_NONE; |
1fccd5c8 DC |
3560 | args.datatype = ap->datatype; |
3561 | if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) | |
9542ae13 DC |
3562 | args.ip = ap->ip; |
3563 | ||
3564 | error = xfs_alloc_vextent(&args); | |
3565 | if (error) | |
49f693fa | 3566 | return error; |
9542ae13 | 3567 | |
49f693fa DC |
3568 | if (tryagain && args.fsbno == NULLFSBLOCK) { |
3569 | /* | |
3570 | * Exact allocation failed. Now try with alignment | |
3571 | * turned on. | |
3572 | */ | |
3573 | args.type = atype; | |
3574 | args.fsbno = ap->blkno; | |
ff105f75 | 3575 | args.alignment = stripe_align; |
49f693fa DC |
3576 | args.minlen = nextminlen; |
3577 | args.minalignslop = 0; | |
3578 | isaligned = 1; | |
3579 | if ((error = xfs_alloc_vextent(&args))) | |
3580 | return error; | |
3581 | } | |
3582 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
2bd0ea18 | 3583 | /* |
49f693fa DC |
3584 | * allocation failed, so turn off alignment and |
3585 | * try again. | |
2bd0ea18 | 3586 | */ |
49f693fa DC |
3587 | args.type = atype; |
3588 | args.fsbno = ap->blkno; | |
3589 | args.alignment = 0; | |
3590 | if ((error = xfs_alloc_vextent(&args))) | |
3591 | return error; | |
3592 | } | |
3593 | if (args.fsbno == NULLFSBLOCK && nullfb && | |
3594 | args.minlen > ap->minlen) { | |
3595 | args.minlen = ap->minlen; | |
3596 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3597 | args.fsbno = ap->blkno; | |
3598 | if ((error = xfs_alloc_vextent(&args))) | |
3599 | return error; | |
3600 | } | |
3601 | if (args.fsbno == NULLFSBLOCK && nullfb) { | |
3602 | args.fsbno = 0; | |
3603 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3604 | args.total = ap->minlen; | |
49f693fa DC |
3605 | if ((error = xfs_alloc_vextent(&args))) |
3606 | return error; | |
f33cea1a | 3607 | ap->dfops->dop_low = true; |
49f693fa DC |
3608 | } |
3609 | if (args.fsbno != NULLFSBLOCK) { | |
2bd0ea18 | 3610 | /* |
49f693fa DC |
3611 | * check the allocation happened at the same or higher AG than |
3612 | * the first block that was allocated. | |
2bd0ea18 | 3613 | */ |
49f693fa | 3614 | ASSERT(*ap->firstblock == NULLFSBLOCK || |
6df6a3ab CH |
3615 | XFS_FSB_TO_AGNO(mp, *ap->firstblock) <= |
3616 | XFS_FSB_TO_AGNO(mp, args.fsbno)); | |
49f693fa DC |
3617 | |
3618 | ap->blkno = args.fsbno; | |
3619 | if (*ap->firstblock == NULLFSBLOCK) | |
3620 | *ap->firstblock = args.fsbno; | |
6df6a3ab | 3621 | ASSERT(nullfb || fb_agno <= args.agno); |
49f693fa | 3622 | ap->length = args.len; |
c38464ff DW |
3623 | /* |
3624 | * If the extent size hint is active, we tried to round the | |
3625 | * caller's allocation request offset down to extsz and the | |
3626 | * length up to another extsz boundary. If we found a free | |
3627 | * extent we mapped it in starting at this new offset. If the | |
3628 | * newly mapped space isn't long enough to cover any of the | |
3629 | * range of offsets that was originally requested, move the | |
3630 | * mapping up so that we can fill as much of the caller's | |
3631 | * original request as possible. Free space is apparently | |
3632 | * very fragmented so we're unlikely to be able to satisfy the | |
3633 | * hints anyway. | |
3634 | */ | |
3635 | if (ap->length <= orig_length) | |
3636 | ap->offset = orig_offset; | |
3637 | else if (ap->offset + ap->length < orig_offset + orig_length) | |
3638 | ap->offset = orig_offset + orig_length - ap->length; | |
3cb68815 | 3639 | xfs_bmap_btalloc_accounting(ap, &args); |
49f693fa DC |
3640 | } else { |
3641 | ap->blkno = NULLFSBLOCK; | |
3642 | ap->length = 0; | |
2bd0ea18 | 3643 | } |
2bd0ea18 | 3644 | return 0; |
56b2de80 DC |
3645 | } |
3646 | ||
3647 | /* | |
49f693fa DC |
3648 | * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file. |
3649 | * It figures out where to ask the underlying allocator to put the new extent. | |
56b2de80 | 3650 | */ |
49f693fa DC |
3651 | STATIC int |
3652 | xfs_bmap_alloc( | |
613e6057 | 3653 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
56b2de80 | 3654 | { |
1fccd5c8 DC |
3655 | if (XFS_IS_REALTIME_INODE(ap->ip) && |
3656 | xfs_alloc_is_userdata(ap->datatype)) | |
49f693fa DC |
3657 | return xfs_bmap_rtalloc(ap); |
3658 | return xfs_bmap_btalloc(ap); | |
56b2de80 | 3659 | } |
56b2de80 | 3660 | |
b3fd8db7 DW |
3661 | /* Trim extent to fit a logical block range. */ |
3662 | void | |
3663 | xfs_trim_extent( | |
3664 | struct xfs_bmbt_irec *irec, | |
3665 | xfs_fileoff_t bno, | |
3666 | xfs_filblks_t len) | |
3667 | { | |
3668 | xfs_fileoff_t distance; | |
3669 | xfs_fileoff_t end = bno + len; | |
3670 | ||
3671 | if (irec->br_startoff + irec->br_blockcount <= bno || | |
3672 | irec->br_startoff >= end) { | |
3673 | irec->br_blockcount = 0; | |
3674 | return; | |
3675 | } | |
3676 | ||
3677 | if (irec->br_startoff < bno) { | |
3678 | distance = bno - irec->br_startoff; | |
3679 | if (isnullstartblock(irec->br_startblock)) | |
3680 | irec->br_startblock = DELAYSTARTBLOCK; | |
3681 | if (irec->br_startblock != DELAYSTARTBLOCK && | |
3682 | irec->br_startblock != HOLESTARTBLOCK) | |
3683 | irec->br_startblock += distance; | |
3684 | irec->br_startoff += distance; | |
3685 | irec->br_blockcount -= distance; | |
3686 | } | |
3687 | ||
3688 | if (end < irec->br_startoff + irec->br_blockcount) { | |
3689 | distance = irec->br_startoff + irec->br_blockcount - end; | |
3690 | irec->br_blockcount -= distance; | |
3691 | } | |
3692 | } | |
3693 | ||
ca75eb20 BF |
3694 | /* trim extent to within eof */ |
3695 | void | |
3696 | xfs_trim_extent_eof( | |
3697 | struct xfs_bmbt_irec *irec, | |
3698 | struct xfs_inode *ip) | |
3699 | ||
3700 | { | |
3701 | xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount, | |
3702 | i_size_read(VFS_I(ip)))); | |
3703 | } | |
3704 | ||
2bd0ea18 | 3705 | /* |
a2ceac1f | 3706 | * Trim the returned map to the required bounds |
2bd0ea18 | 3707 | */ |
a2ceac1f DC |
3708 | STATIC void |
3709 | xfs_bmapi_trim_map( | |
3710 | struct xfs_bmbt_irec *mval, | |
3711 | struct xfs_bmbt_irec *got, | |
3712 | xfs_fileoff_t *bno, | |
3713 | xfs_filblks_t len, | |
3714 | xfs_fileoff_t obno, | |
3715 | xfs_fileoff_t end, | |
3716 | int n, | |
3717 | int flags) | |
2bd0ea18 | 3718 | { |
a2ceac1f DC |
3719 | if ((flags & XFS_BMAPI_ENTIRE) || |
3720 | got->br_startoff + got->br_blockcount <= obno) { | |
3721 | *mval = *got; | |
3722 | if (isnullstartblock(got->br_startblock)) | |
3723 | mval->br_startblock = DELAYSTARTBLOCK; | |
3724 | return; | |
63be04eb | 3725 | } |
a2ceac1f DC |
3726 | |
3727 | if (obno > *bno) | |
3728 | *bno = obno; | |
3729 | ASSERT((*bno >= obno) || (n == 0)); | |
3730 | ASSERT(*bno < end); | |
3731 | mval->br_startoff = *bno; | |
3732 | if (isnullstartblock(got->br_startblock)) | |
3733 | mval->br_startblock = DELAYSTARTBLOCK; | |
2bd0ea18 | 3734 | else |
a2ceac1f DC |
3735 | mval->br_startblock = got->br_startblock + |
3736 | (*bno - got->br_startoff); | |
2bd0ea18 | 3737 | /* |
a2ceac1f DC |
3738 | * Return the minimum of what we got and what we asked for for |
3739 | * the length. We can use the len variable here because it is | |
3740 | * modified below and we could have been there before coming | |
3741 | * here if the first part of the allocation didn't overlap what | |
3742 | * was asked for. | |
2bd0ea18 | 3743 | */ |
a2ceac1f DC |
3744 | mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno, |
3745 | got->br_blockcount - (*bno - got->br_startoff)); | |
3746 | mval->br_state = got->br_state; | |
3747 | ASSERT(mval->br_blockcount <= len); | |
3748 | return; | |
3749 | } | |
56b2de80 | 3750 | |
a2ceac1f DC |
3751 | /* |
3752 | * Update and validate the extent map to return | |
3753 | */ | |
3754 | STATIC void | |
3755 | xfs_bmapi_update_map( | |
3756 | struct xfs_bmbt_irec **map, | |
3757 | xfs_fileoff_t *bno, | |
3758 | xfs_filblks_t *len, | |
3759 | xfs_fileoff_t obno, | |
3760 | xfs_fileoff_t end, | |
3761 | int *n, | |
3762 | int flags) | |
3763 | { | |
3764 | xfs_bmbt_irec_t *mval = *map; | |
3765 | ||
3766 | ASSERT((flags & XFS_BMAPI_ENTIRE) || | |
3767 | ((mval->br_startoff + mval->br_blockcount) <= end)); | |
3768 | ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) || | |
3769 | (mval->br_startoff < obno)); | |
3770 | ||
3771 | *bno = mval->br_startoff + mval->br_blockcount; | |
3772 | *len = end - *bno; | |
3773 | if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) { | |
3774 | /* update previous map with new information */ | |
3775 | ASSERT(mval->br_startblock == mval[-1].br_startblock); | |
3776 | ASSERT(mval->br_blockcount > mval[-1].br_blockcount); | |
3777 | ASSERT(mval->br_state == mval[-1].br_state); | |
3778 | mval[-1].br_blockcount = mval->br_blockcount; | |
3779 | mval[-1].br_state = mval->br_state; | |
3780 | } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK && | |
3781 | mval[-1].br_startblock != DELAYSTARTBLOCK && | |
3782 | mval[-1].br_startblock != HOLESTARTBLOCK && | |
3783 | mval->br_startblock == mval[-1].br_startblock + | |
3784 | mval[-1].br_blockcount && | |
3785 | ((flags & XFS_BMAPI_IGSTATE) || | |
3786 | mval[-1].br_state == mval->br_state)) { | |
3787 | ASSERT(mval->br_startoff == | |
3788 | mval[-1].br_startoff + mval[-1].br_blockcount); | |
3789 | mval[-1].br_blockcount += mval->br_blockcount; | |
3790 | } else if (*n > 0 && | |
3791 | mval->br_startblock == DELAYSTARTBLOCK && | |
3792 | mval[-1].br_startblock == DELAYSTARTBLOCK && | |
3793 | mval->br_startoff == | |
3794 | mval[-1].br_startoff + mval[-1].br_blockcount) { | |
3795 | mval[-1].br_blockcount += mval->br_blockcount; | |
3796 | mval[-1].br_state = mval->br_state; | |
3797 | } else if (!((*n == 0) && | |
3798 | ((mval->br_startoff + mval->br_blockcount) <= | |
3799 | obno))) { | |
3800 | mval++; | |
3801 | (*n)++; | |
3802 | } | |
3803 | *map = mval; | |
3804 | } | |
399ab595 | 3805 | |
a2ceac1f DC |
3806 | /* |
3807 | * Map file blocks to filesystem blocks without allocation. | |
3808 | */ | |
3809 | int | |
3810 | xfs_bmapi_read( | |
3811 | struct xfs_inode *ip, | |
3812 | xfs_fileoff_t bno, | |
3813 | xfs_filblks_t len, | |
3814 | struct xfs_bmbt_irec *mval, | |
3815 | int *nmap, | |
3816 | int flags) | |
3817 | { | |
3818 | struct xfs_mount *mp = ip->i_mount; | |
3819 | struct xfs_ifork *ifp; | |
3820 | struct xfs_bmbt_irec got; | |
a2ceac1f DC |
3821 | xfs_fileoff_t obno; |
3822 | xfs_fileoff_t end; | |
9788e059 | 3823 | struct xfs_iext_cursor icur; |
a2ceac1f | 3824 | int error; |
c2d73ed3 | 3825 | bool eof = false; |
a2ceac1f | 3826 | int n = 0; |
cb8a004a | 3827 | int whichfork = xfs_bmapi_whichfork(flags); |
399ab595 | 3828 | |
a2ceac1f DC |
3829 | ASSERT(*nmap >= 1); |
3830 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE| | |
cb8a004a | 3831 | XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK))); |
ff105f75 | 3832 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)); |
062998e3 | 3833 | |
a2ceac1f DC |
3834 | if (unlikely(XFS_TEST_ERROR( |
3835 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
3836 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 3837 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a2ceac1f | 3838 | XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp); |
12b53197 | 3839 | return -EFSCORRUPTED; |
a2ceac1f | 3840 | } |
062998e3 | 3841 | |
a2ceac1f | 3842 | if (XFS_FORCED_SHUTDOWN(mp)) |
12b53197 | 3843 | return -EIO; |
399ab595 | 3844 | |
79896434 | 3845 | XFS_STATS_INC(mp, xs_blk_mapr); |
a2ceac1f DC |
3846 | |
3847 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
3848 | ||
cb8a004a DW |
3849 | /* No CoW fork? Return a hole. */ |
3850 | if (whichfork == XFS_COW_FORK && !ifp) { | |
3851 | mval->br_startoff = bno; | |
3852 | mval->br_startblock = HOLESTARTBLOCK; | |
3853 | mval->br_blockcount = len; | |
3854 | mval->br_state = XFS_EXT_NORM; | |
3855 | *nmap = 1; | |
3856 | return 0; | |
3857 | } | |
3858 | ||
a2ceac1f DC |
3859 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
3860 | error = xfs_iread_extents(NULL, ip, whichfork); | |
3861 | if (error) | |
3862 | return error; | |
3863 | } | |
3864 | ||
9788e059 | 3865 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) |
c2d73ed3 | 3866 | eof = true; |
a2ceac1f DC |
3867 | end = bno + len; |
3868 | obno = bno; | |
3869 | ||
3870 | while (bno < end && n < *nmap) { | |
3871 | /* Reading past eof, act as though there's a hole up to end. */ | |
3872 | if (eof) | |
3873 | got.br_startoff = end; | |
3874 | if (got.br_startoff > bno) { | |
3875 | /* Reading in a hole. */ | |
2bd0ea18 NS |
3876 | mval->br_startoff = bno; |
3877 | mval->br_startblock = HOLESTARTBLOCK; | |
3878 | mval->br_blockcount = | |
3879 | XFS_FILBLKS_MIN(len, got.br_startoff - bno); | |
3880 | mval->br_state = XFS_EXT_NORM; | |
3881 | bno += mval->br_blockcount; | |
3882 | len -= mval->br_blockcount; | |
3883 | mval++; | |
3884 | n++; | |
3885 | continue; | |
3886 | } | |
a2ceac1f DC |
3887 | |
3888 | /* set up the extent map to return. */ | |
3889 | xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags); | |
3890 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
3891 | ||
3892 | /* If we're done, stop now. */ | |
3893 | if (bno >= end || n >= *nmap) | |
3894 | break; | |
3895 | ||
3896 | /* Else go on to the next record. */ | |
9788e059 | 3897 | if (!xfs_iext_next_extent(ifp, &icur, &got)) |
c2d73ed3 | 3898 | eof = true; |
a2ceac1f DC |
3899 | } |
3900 | *nmap = n; | |
3901 | return 0; | |
3902 | } | |
3903 | ||
7a868ee8 BF |
3904 | /* |
3905 | * Add a delayed allocation extent to an inode. Blocks are reserved from the | |
3906 | * global pool and the extent inserted into the inode in-core extent tree. | |
3907 | * | |
3908 | * On entry, got refers to the first extent beyond the offset of the extent to | |
3909 | * allocate or eof is specified if no such extent exists. On return, got refers | |
3910 | * to the extent record that was inserted to the inode fork. | |
3911 | * | |
3912 | * Note that the allocated extent may have been merged with contiguous extents | |
3913 | * during insertion into the inode fork. Thus, got does not reflect the current | |
3914 | * state of the inode fork on return. If necessary, the caller can use lastx to | |
3915 | * look up the updated record in the inode fork. | |
3916 | */ | |
4488e421 | 3917 | int |
a2ceac1f DC |
3918 | xfs_bmapi_reserve_delalloc( |
3919 | struct xfs_inode *ip, | |
cc66acaf | 3920 | int whichfork, |
f7b1a8b1 | 3921 | xfs_fileoff_t off, |
a2ceac1f | 3922 | xfs_filblks_t len, |
f7b1a8b1 | 3923 | xfs_filblks_t prealloc, |
a2ceac1f | 3924 | struct xfs_bmbt_irec *got, |
9788e059 | 3925 | struct xfs_iext_cursor *icur, |
a2ceac1f DC |
3926 | int eof) |
3927 | { | |
3928 | struct xfs_mount *mp = ip->i_mount; | |
cc66acaf | 3929 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
a2ceac1f DC |
3930 | xfs_extlen_t alen; |
3931 | xfs_extlen_t indlen; | |
a2ceac1f | 3932 | int error; |
f7b1a8b1 | 3933 | xfs_fileoff_t aoff = off; |
a2ceac1f | 3934 | |
f7b1a8b1 BF |
3935 | /* |
3936 | * Cap the alloc length. Keep track of prealloc so we know whether to | |
3937 | * tag the inode before we return. | |
3938 | */ | |
3939 | alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN); | |
a2ceac1f DC |
3940 | if (!eof) |
3941 | alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff); | |
f7b1a8b1 BF |
3942 | if (prealloc && alen >= len) |
3943 | prealloc = alen - len; | |
a2ceac1f DC |
3944 | |
3945 | /* Figure out the extent size, adjust alen */ | |
885ba5ce | 3946 | if (whichfork == XFS_COW_FORK) { |
d41d2303 | 3947 | struct xfs_bmbt_irec prev; |
885ba5ce | 3948 | xfs_extlen_t extsz = xfs_get_cowextsz_hint(ip); |
d41d2303 | 3949 | |
9788e059 | 3950 | if (!xfs_iext_peek_prev_extent(ifp, icur, &prev)) |
d41d2303 CH |
3951 | prev.br_startoff = NULLFILEOFF; |
3952 | ||
885ba5ce | 3953 | error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof, |
a2ceac1f DC |
3954 | 1, 0, &aoff, &alen); |
3955 | ASSERT(!error); | |
3956 | } | |
3957 | ||
a2ceac1f DC |
3958 | /* |
3959 | * Make a transaction-less quota reservation for delayed allocation | |
3960 | * blocks. This number gets adjusted later. We return if we haven't | |
3961 | * allocated blocks already inside this loop. | |
3962 | */ | |
3963 | error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0, | |
885ba5ce | 3964 | XFS_QMOPT_RES_REGBLKS); |
a2ceac1f DC |
3965 | if (error) |
3966 | return error; | |
3967 | ||
3968 | /* | |
3969 | * Split changing sb for alen and indlen since they could be coming | |
3970 | * from different places. | |
3971 | */ | |
3972 | indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen); | |
3973 | ASSERT(indlen > 0); | |
3974 | ||
885ba5ce | 3975 | error = xfs_mod_fdblocks(mp, -((int64_t)alen), false); |
a2ceac1f DC |
3976 | if (error) |
3977 | goto out_unreserve_quota; | |
3978 | ||
19ebedcf | 3979 | error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false); |
a2ceac1f DC |
3980 | if (error) |
3981 | goto out_unreserve_blocks; | |
3982 | ||
3983 | ||
3984 | ip->i_delayed_blks += alen; | |
3985 | ||
3986 | got->br_startoff = aoff; | |
3987 | got->br_startblock = nullstartblock(indlen); | |
3988 | got->br_blockcount = alen; | |
3989 | got->br_state = XFS_EXT_NORM; | |
a2ceac1f | 3990 | |
9788e059 | 3991 | xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got); |
a2ceac1f | 3992 | |
f7b1a8b1 BF |
3993 | /* |
3994 | * Tag the inode if blocks were preallocated. Note that COW fork | |
3995 | * preallocation can occur at the start or end of the extent, even when | |
3996 | * prealloc == 0, so we must also check the aligned offset and length. | |
3997 | */ | |
3998 | if (whichfork == XFS_DATA_FORK && prealloc) | |
3999 | xfs_inode_set_eofblocks_tag(ip); | |
4000 | if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len)) | |
4001 | xfs_inode_set_cowblocks_tag(ip); | |
4002 | ||
a2ceac1f DC |
4003 | return 0; |
4004 | ||
4005 | out_unreserve_blocks: | |
885ba5ce | 4006 | xfs_mod_fdblocks(mp, alen, false); |
a2ceac1f DC |
4007 | out_unreserve_quota: |
4008 | if (XFS_IS_QUOTA_ON(mp)) | |
885ba5ce SH |
4009 | xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, |
4010 | XFS_QMOPT_RES_REGBLKS); | |
a2ceac1f DC |
4011 | return error; |
4012 | } | |
4013 | ||
ff105f75 DC |
4014 | static int |
4015 | xfs_bmapi_allocate( | |
a2ceac1f DC |
4016 | struct xfs_bmalloca *bma) |
4017 | { | |
4018 | struct xfs_mount *mp = bma->ip->i_mount; | |
1277a5e0 | 4019 | int whichfork = xfs_bmapi_whichfork(bma->flags); |
a2ceac1f DC |
4020 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4021 | int tmp_logflags = 0; | |
4022 | int error; | |
4023 | ||
4024 | ASSERT(bma->length > 0); | |
4025 | ||
4026 | /* | |
4027 | * For the wasdelay case, we could also just allocate the stuff asked | |
4028 | * for in this bmap call but that wouldn't be as good. | |
4029 | */ | |
4030 | if (bma->wasdel) { | |
4031 | bma->length = (xfs_extlen_t)bma->got.br_blockcount; | |
4032 | bma->offset = bma->got.br_startoff; | |
9788e059 | 4033 | xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev); |
a2ceac1f DC |
4034 | } else { |
4035 | bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN); | |
4036 | if (!bma->eof) | |
4037 | bma->length = XFS_FILBLKS_MIN(bma->length, | |
4038 | bma->got.br_startoff - bma->offset); | |
4039 | } | |
4040 | ||
4041 | /* | |
1fccd5c8 DC |
4042 | * Set the data type being allocated. For the data fork, the first data |
4043 | * in the file is treated differently to all other allocations. For the | |
4044 | * attribute fork, we only need to ensure the allocated range is not on | |
4045 | * the busy list. | |
a2ceac1f DC |
4046 | */ |
4047 | if (!(bma->flags & XFS_BMAPI_METADATA)) { | |
1fccd5c8 DC |
4048 | bma->datatype = XFS_ALLOC_NOBUSY; |
4049 | if (whichfork == XFS_DATA_FORK) { | |
4050 | if (bma->offset == 0) | |
4051 | bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA; | |
4052 | else | |
4053 | bma->datatype |= XFS_ALLOC_USERDATA; | |
4054 | } | |
9542ae13 | 4055 | if (bma->flags & XFS_BMAPI_ZERO) |
1fccd5c8 | 4056 | bma->datatype |= XFS_ALLOC_USERDATA_ZERO; |
a2ceac1f DC |
4057 | } |
4058 | ||
4059 | bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1; | |
4060 | ||
4061 | /* | |
4062 | * Only want to do the alignment at the eof if it is userdata and | |
4063 | * allocation length is larger than a stripe unit. | |
4064 | */ | |
4065 | if (mp->m_dalign && bma->length >= mp->m_dalign && | |
4066 | !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) { | |
4067 | error = xfs_bmap_isaeof(bma, whichfork); | |
4068 | if (error) | |
4069 | return error; | |
4070 | } | |
4071 | ||
a2ceac1f DC |
4072 | error = xfs_bmap_alloc(bma); |
4073 | if (error) | |
4074 | return error; | |
4075 | ||
a2ceac1f DC |
4076 | if (bma->cur) |
4077 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
4078 | if (bma->blkno == NULLFSBLOCK) | |
4079 | return 0; | |
4080 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4081 | bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork); | |
4082 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
f33cea1a | 4083 | bma->cur->bc_private.b.dfops = bma->dfops; |
a2ceac1f DC |
4084 | } |
4085 | /* | |
4086 | * Bump the number of extents we've allocated | |
4087 | * in this call. | |
4088 | */ | |
4089 | bma->nallocs++; | |
4090 | ||
4091 | if (bma->cur) | |
4092 | bma->cur->bc_private.b.flags = | |
4093 | bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; | |
4094 | ||
4095 | bma->got.br_startoff = bma->offset; | |
4096 | bma->got.br_startblock = bma->blkno; | |
4097 | bma->got.br_blockcount = bma->length; | |
4098 | bma->got.br_state = XFS_EXT_NORM; | |
4099 | ||
4100 | /* | |
4072e4b4 DW |
4101 | * In the data fork, a wasdelay extent has been initialized, so |
4102 | * shouldn't be flagged as unwritten. | |
4103 | * | |
4104 | * For the cow fork, however, we convert delalloc reservations | |
4105 | * (extents allocated for speculative preallocation) to | |
4106 | * allocated unwritten extents, and only convert the unwritten | |
4107 | * extents to real extents when we're about to write the data. | |
a2ceac1f | 4108 | */ |
4072e4b4 DW |
4109 | if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) && |
4110 | (bma->flags & XFS_BMAPI_PREALLOC) && | |
a2ceac1f DC |
4111 | xfs_sb_version_hasextflgbit(&mp->m_sb)) |
4112 | bma->got.br_state = XFS_EXT_UNWRITTEN; | |
4113 | ||
4114 | if (bma->wasdel) | |
1277a5e0 | 4115 | error = xfs_bmap_add_extent_delay_real(bma, whichfork); |
a2ceac1f | 4116 | else |
972432f2 | 4117 | error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip, |
9788e059 | 4118 | whichfork, &bma->icur, &bma->cur, &bma->got, |
972432f2 | 4119 | bma->firstblock, bma->dfops, &bma->logflags); |
a2ceac1f DC |
4120 | |
4121 | bma->logflags |= tmp_logflags; | |
4122 | if (error) | |
4123 | return error; | |
4124 | ||
4125 | /* | |
4126 | * Update our extent pointer, given that xfs_bmap_add_extent_delay_real | |
4127 | * or xfs_bmap_add_extent_hole_real might have merged it into one of | |
4128 | * the neighbouring ones. | |
4129 | */ | |
9788e059 | 4130 | xfs_iext_get_extent(ifp, &bma->icur, &bma->got); |
a2ceac1f DC |
4131 | |
4132 | ASSERT(bma->got.br_startoff <= bma->offset); | |
4133 | ASSERT(bma->got.br_startoff + bma->got.br_blockcount >= | |
4134 | bma->offset + bma->length); | |
4135 | ASSERT(bma->got.br_state == XFS_EXT_NORM || | |
4136 | bma->got.br_state == XFS_EXT_UNWRITTEN); | |
4137 | return 0; | |
4138 | } | |
4139 | ||
a2ceac1f DC |
4140 | STATIC int |
4141 | xfs_bmapi_convert_unwritten( | |
4142 | struct xfs_bmalloca *bma, | |
4143 | struct xfs_bmbt_irec *mval, | |
4144 | xfs_filblks_t len, | |
4145 | int flags) | |
4146 | { | |
cb8a004a | 4147 | int whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4148 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4149 | int tmp_logflags = 0; | |
4150 | int error; | |
4151 | ||
4152 | /* check if we need to do unwritten->real conversion */ | |
4153 | if (mval->br_state == XFS_EXT_UNWRITTEN && | |
4154 | (flags & XFS_BMAPI_PREALLOC)) | |
4155 | return 0; | |
4156 | ||
4157 | /* check if we need to do real->unwritten conversion */ | |
4158 | if (mval->br_state == XFS_EXT_NORM && | |
4159 | (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) != | |
4160 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) | |
4161 | return 0; | |
4162 | ||
4163 | /* | |
4164 | * Modify (by adding) the state flag, if writing. | |
4165 | */ | |
4166 | ASSERT(mval->br_blockcount <= len); | |
4167 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4168 | bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp, | |
4169 | bma->ip, whichfork); | |
4170 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
f33cea1a | 4171 | bma->cur->bc_private.b.dfops = bma->dfops; |
a2ceac1f DC |
4172 | } |
4173 | mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN) | |
4174 | ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
4175 | ||
9542ae13 DC |
4176 | /* |
4177 | * Before insertion into the bmbt, zero the range being converted | |
4178 | * if required. | |
4179 | */ | |
4180 | if (flags & XFS_BMAPI_ZERO) { | |
4181 | error = xfs_zero_extent(bma->ip, mval->br_startblock, | |
4182 | mval->br_blockcount); | |
4183 | if (error) | |
4184 | return error; | |
4185 | } | |
4186 | ||
4072e4b4 | 4187 | error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork, |
9788e059 CH |
4188 | &bma->icur, &bma->cur, mval, bma->firstblock, |
4189 | bma->dfops, &tmp_logflags); | |
23fc058d BF |
4190 | /* |
4191 | * Log the inode core unconditionally in the unwritten extent conversion | |
4192 | * path because the conversion might not have done so (e.g., if the | |
4193 | * extent count hasn't changed). We need to make sure the inode is dirty | |
4194 | * in the transaction for the sake of fsync(), even if nothing has | |
4195 | * changed, because fsync() will not force the log for this transaction | |
4196 | * unless it sees the inode pinned. | |
4072e4b4 DW |
4197 | * |
4198 | * Note: If we're only converting cow fork extents, there aren't | |
4199 | * any on-disk updates to make, so we don't need to log anything. | |
23fc058d | 4200 | */ |
4072e4b4 DW |
4201 | if (whichfork != XFS_COW_FORK) |
4202 | bma->logflags |= tmp_logflags | XFS_ILOG_CORE; | |
a2ceac1f DC |
4203 | if (error) |
4204 | return error; | |
4205 | ||
4206 | /* | |
4207 | * Update our extent pointer, given that | |
4208 | * xfs_bmap_add_extent_unwritten_real might have merged it into one | |
4209 | * of the neighbouring ones. | |
4210 | */ | |
9788e059 | 4211 | xfs_iext_get_extent(ifp, &bma->icur, &bma->got); |
a2ceac1f DC |
4212 | |
4213 | /* | |
4214 | * We may have combined previously unwritten space with written space, | |
4215 | * so generate another request. | |
4216 | */ | |
4217 | if (mval->br_blockcount < len) | |
12b53197 | 4218 | return -EAGAIN; |
a2ceac1f DC |
4219 | return 0; |
4220 | } | |
4221 | ||
4222 | /* | |
4223 | * Map file blocks to filesystem blocks, and allocate blocks or convert the | |
4224 | * extent state if necessary. Details behaviour is controlled by the flags | |
4225 | * parameter. Only allocates blocks from a single allocation group, to avoid | |
4226 | * locking problems. | |
4227 | * | |
4228 | * The returned value in "firstblock" from the first call in a transaction | |
4229 | * must be remembered and presented to subsequent calls in "firstblock". | |
4230 | * An upper bound for the number of blocks to be allocated is supplied to | |
4231 | * the first call in "total"; if no allocation group has that many free | |
4232 | * blocks then the call will fail (return NULLFSBLOCK in "firstblock"). | |
4233 | */ | |
4234 | int | |
4235 | xfs_bmapi_write( | |
4236 | struct xfs_trans *tp, /* transaction pointer */ | |
4237 | struct xfs_inode *ip, /* incore inode */ | |
4238 | xfs_fileoff_t bno, /* starting file offs. mapped */ | |
4239 | xfs_filblks_t len, /* length to map in file */ | |
4240 | int flags, /* XFS_BMAPI_... */ | |
4241 | xfs_fsblock_t *firstblock, /* first allocated block | |
4242 | controls a.g. for allocs */ | |
4243 | xfs_extlen_t total, /* total blocks needed */ | |
4244 | struct xfs_bmbt_irec *mval, /* output: map values */ | |
4245 | int *nmap, /* i/o: mval size/count */ | |
f33cea1a | 4246 | struct xfs_defer_ops *dfops) /* i/o: list extents to free */ |
a2ceac1f DC |
4247 | { |
4248 | struct xfs_mount *mp = ip->i_mount; | |
4249 | struct xfs_ifork *ifp; | |
389b3b07 | 4250 | struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */ |
a2ceac1f | 4251 | xfs_fileoff_t end; /* end of mapped file region */ |
82411945 | 4252 | bool eof = false; /* after the end of extents */ |
a2ceac1f DC |
4253 | int error; /* error return */ |
4254 | int n; /* current extent index */ | |
4255 | xfs_fileoff_t obno; /* old block number (offset) */ | |
4256 | int whichfork; /* data or attr fork */ | |
a2ceac1f DC |
4257 | |
4258 | #ifdef DEBUG | |
4259 | xfs_fileoff_t orig_bno; /* original block number value */ | |
4260 | int orig_flags; /* original flags arg value */ | |
4261 | xfs_filblks_t orig_len; /* original value of len arg */ | |
4262 | struct xfs_bmbt_irec *orig_mval; /* original value of mval */ | |
4263 | int orig_nmap; /* original value of *nmap */ | |
4264 | ||
4265 | orig_bno = bno; | |
4266 | orig_len = len; | |
4267 | orig_flags = flags; | |
4268 | orig_mval = mval; | |
4269 | orig_nmap = *nmap; | |
4270 | #endif | |
1277a5e0 | 4271 | whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4272 | |
4273 | ASSERT(*nmap >= 1); | |
4274 | ASSERT(*nmap <= XFS_BMAP_MAX_NMAP); | |
4275 | ASSERT(!(flags & XFS_BMAPI_IGSTATE)); | |
4072e4b4 DW |
4276 | ASSERT(tp != NULL || |
4277 | (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) == | |
4278 | (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)); | |
a2ceac1f | 4279 | ASSERT(len > 0); |
3f17ed4b | 4280 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL); |
ff105f75 | 4281 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b44aa85 | 4282 | ASSERT(!(flags & XFS_BMAPI_REMAP)); |
a2ceac1f | 4283 | |
9542ae13 DC |
4284 | /* zeroing is for currently only for data extents, not metadata */ |
4285 | ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) != | |
4286 | (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)); | |
4287 | /* | |
4288 | * we can allocate unwritten extents or pre-zero allocated blocks, | |
4289 | * but it makes no sense to do both at once. This would result in | |
4290 | * zeroing the unwritten extent twice, but it still being an | |
4291 | * unwritten extent.... | |
4292 | */ | |
4293 | ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) != | |
4294 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)); | |
4295 | ||
a2ceac1f DC |
4296 | if (unlikely(XFS_TEST_ERROR( |
4297 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
3f17ed4b | 4298 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), |
e2a190dd | 4299 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a2ceac1f | 4300 | XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp); |
12b53197 | 4301 | return -EFSCORRUPTED; |
a2ceac1f DC |
4302 | } |
4303 | ||
4304 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 4305 | return -EIO; |
a2ceac1f DC |
4306 | |
4307 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
4308 | ||
79896434 | 4309 | XFS_STATS_INC(mp, xs_blk_mapw); |
a2ceac1f | 4310 | |
a2ceac1f DC |
4311 | if (*firstblock == NULLFSBLOCK) { |
4312 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE) | |
4313 | bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1; | |
4314 | else | |
4315 | bma.minleft = 1; | |
4316 | } else { | |
4317 | bma.minleft = 0; | |
4318 | } | |
4319 | ||
4320 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
4321 | error = xfs_iread_extents(tp, ip, whichfork); | |
4322 | if (error) | |
4323 | goto error0; | |
4324 | } | |
4325 | ||
a2ceac1f DC |
4326 | n = 0; |
4327 | end = bno + len; | |
4328 | obno = bno; | |
4329 | ||
9788e059 | 4330 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got)) |
82411945 | 4331 | eof = true; |
9788e059 | 4332 | if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev)) |
82411945 | 4333 | bma.prev.br_startoff = NULLFILEOFF; |
a2ceac1f DC |
4334 | bma.tp = tp; |
4335 | bma.ip = ip; | |
4336 | bma.total = total; | |
1fccd5c8 | 4337 | bma.datatype = 0; |
f33cea1a | 4338 | bma.dfops = dfops; |
a2ceac1f DC |
4339 | bma.firstblock = firstblock; |
4340 | ||
4341 | while (bno < end && n < *nmap) { | |
7075a23f CH |
4342 | bool need_alloc = false, wasdelay = false; |
4343 | ||
faaad1df | 4344 | /* in hole or beyond EOF? */ |
7075a23f | 4345 | if (eof || bma.got.br_startoff > bno) { |
faaad1df DW |
4346 | /* |
4347 | * CoW fork conversions should /never/ hit EOF or | |
4348 | * holes. There should always be something for us | |
4349 | * to work on. | |
4350 | */ | |
4351 | ASSERT(!((flags & XFS_BMAPI_CONVERT) && | |
4352 | (flags & XFS_BMAPI_COWFORK))); | |
4353 | ||
7075a23f CH |
4354 | if (flags & XFS_BMAPI_DELALLOC) { |
4355 | /* | |
4356 | * For the COW fork we can reasonably get a | |
4357 | * request for converting an extent that races | |
4358 | * with other threads already having converted | |
4359 | * part of it, as there converting COW to | |
4360 | * regular blocks is not protected using the | |
4361 | * IOLOCK. | |
4362 | */ | |
4363 | ASSERT(flags & XFS_BMAPI_COWFORK); | |
4364 | if (!(flags & XFS_BMAPI_COWFORK)) { | |
4365 | error = -EIO; | |
4366 | goto error0; | |
4367 | } | |
4368 | ||
4369 | if (eof || bno >= end) | |
4370 | break; | |
4371 | } else { | |
4372 | need_alloc = true; | |
4373 | } | |
0b44aa85 CH |
4374 | } else if (isnullstartblock(bma.got.br_startblock)) { |
4375 | wasdelay = true; | |
7075a23f | 4376 | } |
34621a47 | 4377 | |
2bd0ea18 | 4378 | /* |
a2ceac1f DC |
4379 | * First, deal with the hole before the allocated space |
4380 | * that we found, if any. | |
2bd0ea18 | 4381 | */ |
8e87c884 CH |
4382 | if ((need_alloc || wasdelay) && |
4383 | !(flags & XFS_BMAPI_CONVERT_ONLY)) { | |
a2ceac1f DC |
4384 | bma.eof = eof; |
4385 | bma.conv = !!(flags & XFS_BMAPI_CONVERT); | |
4386 | bma.wasdel = wasdelay; | |
4387 | bma.offset = bno; | |
4388 | bma.flags = flags; | |
4389 | ||
2bd0ea18 | 4390 | /* |
a2ceac1f DC |
4391 | * There's a 32/64 bit type mismatch between the |
4392 | * allocation length request (which can be 64 bits in | |
4393 | * length) and the bma length request, which is | |
4394 | * xfs_extlen_t and therefore 32 bits. Hence we have to | |
4395 | * check for 32-bit overflows and handle them here. | |
2bd0ea18 | 4396 | */ |
a2ceac1f DC |
4397 | if (len > (xfs_filblks_t)MAXEXTLEN) |
4398 | bma.length = MAXEXTLEN; | |
4399 | else | |
4400 | bma.length = len; | |
4401 | ||
4402 | ASSERT(len > 0); | |
4403 | ASSERT(bma.length > 0); | |
4404 | error = xfs_bmapi_allocate(&bma); | |
2bd0ea18 NS |
4405 | if (error) |
4406 | goto error0; | |
a2ceac1f DC |
4407 | if (bma.blkno == NULLFSBLOCK) |
4408 | break; | |
10e65503 DW |
4409 | |
4410 | /* | |
4411 | * If this is a CoW allocation, record the data in | |
4412 | * the refcount btree for orphan recovery. | |
4413 | */ | |
4414 | if (whichfork == XFS_COW_FORK) { | |
4415 | error = xfs_refcount_alloc_cow_extent(mp, dfops, | |
4416 | bma.blkno, bma.length); | |
4417 | if (error) | |
4418 | goto error0; | |
4419 | } | |
2bd0ea18 NS |
4420 | } |
4421 | ||
a2ceac1f DC |
4422 | /* Deal with the allocated space we found. */ |
4423 | xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno, | |
4424 | end, n, flags); | |
4425 | ||
4426 | /* Execute unwritten extent conversion if necessary */ | |
4427 | error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags); | |
12b53197 | 4428 | if (error == -EAGAIN) |
a2ceac1f DC |
4429 | continue; |
4430 | if (error) | |
4431 | goto error0; | |
4432 | ||
4433 | /* update the extent map to return */ | |
4434 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4435 | ||
2bd0ea18 NS |
4436 | /* |
4437 | * If we're done, stop now. Stop when we've allocated | |
4438 | * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise | |
4439 | * the transaction may get too big. | |
4440 | */ | |
a2ceac1f | 4441 | if (bno >= end || n >= *nmap || bma.nallocs >= *nmap) |
2bd0ea18 | 4442 | break; |
a2ceac1f DC |
4443 | |
4444 | /* Else go on to the next record. */ | |
4445 | bma.prev = bma.got; | |
9788e059 | 4446 | if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got)) |
82411945 | 4447 | eof = true; |
2bd0ea18 | 4448 | } |
2bd0ea18 | 4449 | *nmap = n; |
a2ceac1f | 4450 | |
2bd0ea18 NS |
4451 | /* |
4452 | * Transform from btree to extents, give it cur. | |
4453 | */ | |
a2ceac1f DC |
4454 | if (xfs_bmap_wants_extents(ip, whichfork)) { |
4455 | int tmp_logflags = 0; | |
4456 | ||
4457 | ASSERT(bma.cur); | |
4458 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, | |
4ca431fc | 4459 | &tmp_logflags, whichfork); |
a2ceac1f | 4460 | bma.logflags |= tmp_logflags; |
2bd0ea18 NS |
4461 | if (error) |
4462 | goto error0; | |
4463 | } | |
a2ceac1f | 4464 | |
2bd0ea18 | 4465 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE || |
a2ceac1f DC |
4466 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
4467 | XFS_IFORK_MAXEXT(ip, whichfork)); | |
2bd0ea18 | 4468 | error = 0; |
2bd0ea18 NS |
4469 | error0: |
4470 | /* | |
4471 | * Log everything. Do this after conversion, there's no point in | |
5e656dbb | 4472 | * logging the extent records if we've converted to btree format. |
2bd0ea18 | 4473 | */ |
a2ceac1f | 4474 | if ((bma.logflags & xfs_ilog_fext(whichfork)) && |
2bd0ea18 | 4475 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) |
a2ceac1f DC |
4476 | bma.logflags &= ~xfs_ilog_fext(whichfork); |
4477 | else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) && | |
2bd0ea18 | 4478 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) |
a2ceac1f | 4479 | bma.logflags &= ~xfs_ilog_fbroot(whichfork); |
2bd0ea18 NS |
4480 | /* |
4481 | * Log whatever the flags say, even if error. Otherwise we might miss | |
4482 | * detecting a case where the data is changed, there's an error, | |
4483 | * and it's not logged so we don't shutdown when we should. | |
4484 | */ | |
a2ceac1f DC |
4485 | if (bma.logflags) |
4486 | xfs_trans_log_inode(tp, ip, bma.logflags); | |
4487 | ||
4488 | if (bma.cur) { | |
2bd0ea18 NS |
4489 | if (!error) { |
4490 | ASSERT(*firstblock == NULLFSBLOCK || | |
6df6a3ab | 4491 | XFS_FSB_TO_AGNO(mp, *firstblock) <= |
eae766ca | 4492 | XFS_FSB_TO_AGNO(mp, |
6df6a3ab | 4493 | bma.cur->bc_private.b.firstblock)); |
a2ceac1f | 4494 | *firstblock = bma.cur->bc_private.b.firstblock; |
2bd0ea18 | 4495 | } |
a2ceac1f | 4496 | xfs_btree_del_cursor(bma.cur, |
2bd0ea18 NS |
4497 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
4498 | } | |
4499 | if (!error) | |
4500 | xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval, | |
4501 | orig_nmap, *nmap); | |
4502 | return error; | |
4503 | } | |
4504 | ||
0b44aa85 CH |
4505 | static int |
4506 | xfs_bmapi_remap( | |
4507 | struct xfs_trans *tp, | |
4508 | struct xfs_inode *ip, | |
4509 | xfs_fileoff_t bno, | |
4510 | xfs_filblks_t len, | |
4511 | xfs_fsblock_t startblock, | |
4512 | struct xfs_defer_ops *dfops) | |
4513 | { | |
4514 | struct xfs_mount *mp = ip->i_mount; | |
4515 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
4516 | struct xfs_btree_cur *cur = NULL; | |
4517 | xfs_fsblock_t firstblock = NULLFSBLOCK; | |
4518 | struct xfs_bmbt_irec got; | |
9788e059 | 4519 | struct xfs_iext_cursor icur; |
0b44aa85 CH |
4520 | int logflags = 0, error; |
4521 | ||
4522 | ASSERT(len > 0); | |
4523 | ASSERT(len <= (xfs_filblks_t)MAXEXTLEN); | |
4524 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
4525 | ||
4526 | if (unlikely(XFS_TEST_ERROR( | |
4527 | (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS && | |
4528 | XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 4529 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
0b44aa85 CH |
4530 | XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp); |
4531 | return -EFSCORRUPTED; | |
4532 | } | |
4533 | ||
4534 | if (XFS_FORCED_SHUTDOWN(mp)) | |
4535 | return -EIO; | |
4536 | ||
4537 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
4538 | error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK); | |
4539 | if (error) | |
4540 | return error; | |
4541 | } | |
4542 | ||
9788e059 | 4543 | if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) { |
0b44aa85 CH |
4544 | /* make sure we only reflink into a hole. */ |
4545 | ASSERT(got.br_startoff > bno); | |
4546 | ASSERT(got.br_startoff - bno >= len); | |
4547 | } | |
4548 | ||
05422db6 CH |
4549 | ip->i_d.di_nblocks += len; |
4550 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
0b44aa85 CH |
4551 | |
4552 | if (ifp->if_flags & XFS_IFBROOT) { | |
4553 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
4554 | cur->bc_private.b.firstblock = firstblock; | |
4555 | cur->bc_private.b.dfops = dfops; | |
4556 | cur->bc_private.b.flags = 0; | |
4557 | } | |
4558 | ||
4559 | got.br_startoff = bno; | |
4560 | got.br_startblock = startblock; | |
4561 | got.br_blockcount = len; | |
4562 | got.br_state = XFS_EXT_NORM; | |
4563 | ||
9788e059 CH |
4564 | error = xfs_bmap_add_extent_hole_real(tp, ip, XFS_DATA_FORK, &icur, |
4565 | &cur, &got, &firstblock, dfops, &logflags); | |
0b44aa85 CH |
4566 | if (error) |
4567 | goto error0; | |
4568 | ||
4569 | if (xfs_bmap_wants_extents(ip, XFS_DATA_FORK)) { | |
4570 | int tmp_logflags = 0; | |
4571 | ||
4572 | error = xfs_bmap_btree_to_extents(tp, ip, cur, | |
4573 | &tmp_logflags, XFS_DATA_FORK); | |
4574 | logflags |= tmp_logflags; | |
4575 | } | |
4576 | ||
4577 | error0: | |
4578 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) | |
4579 | logflags &= ~XFS_ILOG_DEXT; | |
4580 | else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) | |
4581 | logflags &= ~XFS_ILOG_DBROOT; | |
4582 | ||
4583 | if (logflags) | |
4584 | xfs_trans_log_inode(tp, ip, logflags); | |
4585 | if (cur) { | |
4586 | xfs_btree_del_cursor(cur, | |
4587 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
4588 | } | |
4589 | return error; | |
4590 | } | |
4591 | ||
01d1b786 BF |
4592 | /* |
4593 | * When a delalloc extent is split (e.g., due to a hole punch), the original | |
4594 | * indlen reservation must be shared across the two new extents that are left | |
4595 | * behind. | |
4596 | * | |
4597 | * Given the original reservation and the worst case indlen for the two new | |
4598 | * extents (as calculated by xfs_bmap_worst_indlen()), split the original | |
731ccdf9 BF |
4599 | * reservation fairly across the two new extents. If necessary, steal available |
4600 | * blocks from a deleted extent to make up a reservation deficiency (e.g., if | |
4601 | * ores == 1). The number of stolen blocks is returned. The availability and | |
4602 | * subsequent accounting of stolen blocks is the responsibility of the caller. | |
01d1b786 | 4603 | */ |
731ccdf9 | 4604 | static xfs_filblks_t |
01d1b786 BF |
4605 | xfs_bmap_split_indlen( |
4606 | xfs_filblks_t ores, /* original res. */ | |
4607 | xfs_filblks_t *indlen1, /* ext1 worst indlen */ | |
731ccdf9 BF |
4608 | xfs_filblks_t *indlen2, /* ext2 worst indlen */ |
4609 | xfs_filblks_t avail) /* stealable blocks */ | |
01d1b786 BF |
4610 | { |
4611 | xfs_filblks_t len1 = *indlen1; | |
4612 | xfs_filblks_t len2 = *indlen2; | |
4613 | xfs_filblks_t nres = len1 + len2; /* new total res. */ | |
731ccdf9 | 4614 | xfs_filblks_t stolen = 0; |
775762e4 | 4615 | xfs_filblks_t resfactor; |
731ccdf9 BF |
4616 | |
4617 | /* | |
4618 | * Steal as many blocks as we can to try and satisfy the worst case | |
4619 | * indlen for both new extents. | |
4620 | */ | |
775762e4 BF |
4621 | if (ores < nres && avail) |
4622 | stolen = XFS_FILBLKS_MIN(nres - ores, avail); | |
4623 | ores += stolen; | |
4624 | ||
4625 | /* nothing else to do if we've satisfied the new reservation */ | |
4626 | if (ores >= nres) | |
4627 | return stolen; | |
4628 | ||
4629 | /* | |
4630 | * We can't meet the total required reservation for the two extents. | |
4631 | * Calculate the percent of the overall shortage between both extents | |
4632 | * and apply this percentage to each of the requested indlen values. | |
4633 | * This distributes the shortage fairly and reduces the chances that one | |
4634 | * of the two extents is left with nothing when extents are repeatedly | |
4635 | * split. | |
4636 | */ | |
4637 | resfactor = (ores * 100); | |
4638 | do_div(resfactor, nres); | |
4639 | len1 *= resfactor; | |
4640 | do_div(len1, 100); | |
4641 | len2 *= resfactor; | |
4642 | do_div(len2, 100); | |
4643 | ASSERT(len1 + len2 <= ores); | |
4644 | ASSERT(len1 < *indlen1 && len2 < *indlen2); | |
01d1b786 BF |
4645 | |
4646 | /* | |
775762e4 BF |
4647 | * Hand out the remainder to each extent. If one of the two reservations |
4648 | * is zero, we want to make sure that one gets a block first. The loop | |
4649 | * below starts with len1, so hand len2 a block right off the bat if it | |
4650 | * is zero. | |
01d1b786 | 4651 | */ |
775762e4 BF |
4652 | ores -= (len1 + len2); |
4653 | ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores); | |
4654 | if (ores && !len2 && *indlen2) { | |
4655 | len2++; | |
4656 | ores--; | |
4657 | } | |
4658 | while (ores) { | |
4659 | if (len1 < *indlen1) { | |
4660 | len1++; | |
4661 | ores--; | |
01d1b786 | 4662 | } |
775762e4 | 4663 | if (!ores) |
01d1b786 | 4664 | break; |
775762e4 BF |
4665 | if (len2 < *indlen2) { |
4666 | len2++; | |
4667 | ores--; | |
01d1b786 BF |
4668 | } |
4669 | } | |
4670 | ||
4671 | *indlen1 = len1; | |
4672 | *indlen2 = len2; | |
731ccdf9 BF |
4673 | |
4674 | return stolen; | |
01d1b786 BF |
4675 | } |
4676 | ||
ece930fa CH |
4677 | int |
4678 | xfs_bmap_del_extent_delay( | |
4679 | struct xfs_inode *ip, | |
4680 | int whichfork, | |
9788e059 | 4681 | struct xfs_iext_cursor *icur, |
ece930fa CH |
4682 | struct xfs_bmbt_irec *got, |
4683 | struct xfs_bmbt_irec *del) | |
4684 | { | |
4685 | struct xfs_mount *mp = ip->i_mount; | |
4686 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
4687 | struct xfs_bmbt_irec new; | |
4688 | int64_t da_old, da_new, da_diff = 0; | |
4689 | xfs_fileoff_t del_endoff, got_endoff; | |
4690 | xfs_filblks_t got_indlen, new_indlen, stolen; | |
7bd43334 CH |
4691 | int state = xfs_bmap_fork_to_state(whichfork); |
4692 | int error = 0; | |
ece930fa CH |
4693 | bool isrt; |
4694 | ||
4695 | XFS_STATS_INC(mp, xs_del_exlist); | |
4696 | ||
4697 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); | |
4698 | del_endoff = del->br_startoff + del->br_blockcount; | |
4699 | got_endoff = got->br_startoff + got->br_blockcount; | |
4700 | da_old = startblockval(got->br_startblock); | |
4701 | da_new = 0; | |
4702 | ||
ece930fa CH |
4703 | ASSERT(del->br_blockcount > 0); |
4704 | ASSERT(got->br_startoff <= del->br_startoff); | |
4705 | ASSERT(got_endoff >= del_endoff); | |
4706 | ||
4707 | if (isrt) { | |
f1e0bd8d | 4708 | uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount); |
ece930fa CH |
4709 | |
4710 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
4711 | xfs_mod_frextents(mp, rtexts); | |
4712 | } | |
4713 | ||
4714 | /* | |
4715 | * Update the inode delalloc counter now and wait to update the | |
4716 | * sb counters as we might have to borrow some blocks for the | |
4717 | * indirect block accounting. | |
4718 | */ | |
4719 | error = xfs_trans_reserve_quota_nblks(NULL, ip, | |
4720 | -((long)del->br_blockcount), 0, | |
4721 | isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
6921638c DW |
4722 | if (error) |
4723 | return error; | |
ece930fa CH |
4724 | ip->i_delayed_blks -= del->br_blockcount; |
4725 | ||
ece930fa | 4726 | if (got->br_startoff == del->br_startoff) |
d0a03e5a | 4727 | state |= BMAP_LEFT_FILLING; |
ece930fa | 4728 | if (got_endoff == del_endoff) |
d0a03e5a | 4729 | state |= BMAP_RIGHT_FILLING; |
ece930fa | 4730 | |
d0a03e5a CH |
4731 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4732 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
ece930fa CH |
4733 | /* |
4734 | * Matches the whole extent. Delete the entry. | |
4735 | */ | |
cf455a62 | 4736 | xfs_iext_remove(ip, icur, state); |
9788e059 | 4737 | xfs_iext_prev(ifp, icur); |
ece930fa | 4738 | break; |
d0a03e5a | 4739 | case BMAP_LEFT_FILLING: |
ece930fa CH |
4740 | /* |
4741 | * Deleting the first part of the extent. | |
4742 | */ | |
ece930fa CH |
4743 | got->br_startoff = del_endoff; |
4744 | got->br_blockcount -= del->br_blockcount; | |
4745 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4746 | got->br_blockcount), da_old); | |
4747 | got->br_startblock = nullstartblock((int)da_new); | |
9788e059 | 4748 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa | 4749 | break; |
d0a03e5a | 4750 | case BMAP_RIGHT_FILLING: |
ece930fa CH |
4751 | /* |
4752 | * Deleting the last part of the extent. | |
4753 | */ | |
ece930fa CH |
4754 | got->br_blockcount = got->br_blockcount - del->br_blockcount; |
4755 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4756 | got->br_blockcount), da_old); | |
4757 | got->br_startblock = nullstartblock((int)da_new); | |
9788e059 | 4758 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa CH |
4759 | break; |
4760 | case 0: | |
4761 | /* | |
4762 | * Deleting the middle of the extent. | |
4763 | * | |
4764 | * Distribute the original indlen reservation across the two new | |
4765 | * extents. Steal blocks from the deleted extent if necessary. | |
4766 | * Stealing blocks simply fudges the fdblocks accounting below. | |
4767 | * Warn if either of the new indlen reservations is zero as this | |
4768 | * can lead to delalloc problems. | |
4769 | */ | |
ece930fa CH |
4770 | got->br_blockcount = del->br_startoff - got->br_startoff; |
4771 | got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount); | |
4772 | ||
4773 | new.br_blockcount = got_endoff - del_endoff; | |
4774 | new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
4775 | ||
4776 | WARN_ON_ONCE(!got_indlen || !new_indlen); | |
4777 | stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen, | |
4778 | del->br_blockcount); | |
4779 | ||
4780 | got->br_startblock = nullstartblock((int)got_indlen); | |
ece930fa CH |
4781 | |
4782 | new.br_startoff = del_endoff; | |
4783 | new.br_state = got->br_state; | |
4784 | new.br_startblock = nullstartblock((int)new_indlen); | |
4785 | ||
9788e059 CH |
4786 | xfs_iext_update_extent(ip, state, icur, got); |
4787 | xfs_iext_next(ifp, icur); | |
26a75f67 | 4788 | xfs_iext_insert(ip, icur, &new, state); |
ece930fa CH |
4789 | |
4790 | da_new = got_indlen + new_indlen - stolen; | |
4791 | del->br_blockcount -= stolen; | |
4792 | break; | |
4793 | } | |
4794 | ||
4795 | ASSERT(da_old >= da_new); | |
4796 | da_diff = da_old - da_new; | |
4797 | if (!isrt) | |
4798 | da_diff += del->br_blockcount; | |
4799 | if (da_diff) | |
4800 | xfs_mod_fdblocks(mp, da_diff, false); | |
4801 | return error; | |
4802 | } | |
4803 | ||
4804 | void | |
4805 | xfs_bmap_del_extent_cow( | |
4806 | struct xfs_inode *ip, | |
9788e059 | 4807 | struct xfs_iext_cursor *icur, |
ece930fa CH |
4808 | struct xfs_bmbt_irec *got, |
4809 | struct xfs_bmbt_irec *del) | |
4810 | { | |
4811 | struct xfs_mount *mp = ip->i_mount; | |
4812 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
4813 | struct xfs_bmbt_irec new; | |
4814 | xfs_fileoff_t del_endoff, got_endoff; | |
4815 | int state = BMAP_COWFORK; | |
4816 | ||
4817 | XFS_STATS_INC(mp, xs_del_exlist); | |
4818 | ||
4819 | del_endoff = del->br_startoff + del->br_blockcount; | |
4820 | got_endoff = got->br_startoff + got->br_blockcount; | |
4821 | ||
ece930fa CH |
4822 | ASSERT(del->br_blockcount > 0); |
4823 | ASSERT(got->br_startoff <= del->br_startoff); | |
4824 | ASSERT(got_endoff >= del_endoff); | |
4825 | ASSERT(!isnullstartblock(got->br_startblock)); | |
4826 | ||
4827 | if (got->br_startoff == del->br_startoff) | |
d0a03e5a | 4828 | state |= BMAP_LEFT_FILLING; |
ece930fa | 4829 | if (got_endoff == del_endoff) |
d0a03e5a | 4830 | state |= BMAP_RIGHT_FILLING; |
ece930fa | 4831 | |
d0a03e5a CH |
4832 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4833 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
ece930fa CH |
4834 | /* |
4835 | * Matches the whole extent. Delete the entry. | |
4836 | */ | |
cf455a62 | 4837 | xfs_iext_remove(ip, icur, state); |
9788e059 | 4838 | xfs_iext_prev(ifp, icur); |
ece930fa | 4839 | break; |
d0a03e5a | 4840 | case BMAP_LEFT_FILLING: |
ece930fa CH |
4841 | /* |
4842 | * Deleting the first part of the extent. | |
4843 | */ | |
ece930fa CH |
4844 | got->br_startoff = del_endoff; |
4845 | got->br_blockcount -= del->br_blockcount; | |
4846 | got->br_startblock = del->br_startblock + del->br_blockcount; | |
9788e059 | 4847 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa | 4848 | break; |
d0a03e5a | 4849 | case BMAP_RIGHT_FILLING: |
ece930fa CH |
4850 | /* |
4851 | * Deleting the last part of the extent. | |
4852 | */ | |
ece930fa | 4853 | got->br_blockcount -= del->br_blockcount; |
9788e059 | 4854 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa CH |
4855 | break; |
4856 | case 0: | |
4857 | /* | |
4858 | * Deleting the middle of the extent. | |
4859 | */ | |
ece930fa | 4860 | got->br_blockcount = del->br_startoff - got->br_startoff; |
ece930fa CH |
4861 | |
4862 | new.br_startoff = del_endoff; | |
4863 | new.br_blockcount = got_endoff - del_endoff; | |
4864 | new.br_state = got->br_state; | |
4865 | new.br_startblock = del->br_startblock + del->br_blockcount; | |
4866 | ||
9788e059 CH |
4867 | xfs_iext_update_extent(ip, state, icur, got); |
4868 | xfs_iext_next(ifp, icur); | |
26a75f67 | 4869 | xfs_iext_insert(ip, icur, &new, state); |
ece930fa CH |
4870 | break; |
4871 | } | |
d07cc724 | 4872 | ip->i_delayed_blks -= del->br_blockcount; |
ece930fa CH |
4873 | } |
4874 | ||
2bd0ea18 | 4875 | /* |
49f693fa | 4876 | * Called by xfs_bmapi to update file extent records and the btree |
ad68fd19 | 4877 | * after removing space. |
2bd0ea18 | 4878 | */ |
49f693fa | 4879 | STATIC int /* error */ |
ad68fd19 | 4880 | xfs_bmap_del_extent_real( |
49f693fa DC |
4881 | xfs_inode_t *ip, /* incore inode pointer */ |
4882 | xfs_trans_t *tp, /* current transaction pointer */ | |
9788e059 | 4883 | struct xfs_iext_cursor *icur, |
f33cea1a | 4884 | struct xfs_defer_ops *dfops, /* list of extents to be freed */ |
49f693fa DC |
4885 | xfs_btree_cur_t *cur, /* if null, not a btree */ |
4886 | xfs_bmbt_irec_t *del, /* data to remove from extents */ | |
4887 | int *logflagsp, /* inode logging flags */ | |
36b16da8 DW |
4888 | int whichfork, /* data or attr fork */ |
4889 | int bflags) /* bmapi flags */ | |
2bd0ea18 | 4890 | { |
49f693fa DC |
4891 | xfs_fsblock_t del_endblock=0; /* first block past del */ |
4892 | xfs_fileoff_t del_endoff; /* first offset past del */ | |
49f693fa | 4893 | int do_fx; /* free extent at end of routine */ |
49f693fa | 4894 | int error; /* error return value */ |
bd92a38b | 4895 | int flags = 0;/* inode logging flags */ |
70bf7533 | 4896 | struct xfs_bmbt_irec got; /* current extent entry */ |
49f693fa DC |
4897 | xfs_fileoff_t got_endoff; /* first offset past got */ |
4898 | int i; /* temp state */ | |
4899 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
4900 | xfs_mount_t *mp; /* mount structure */ | |
4901 | xfs_filblks_t nblks; /* quota/sb block count */ | |
4902 | xfs_bmbt_irec_t new; /* new record to be inserted */ | |
4903 | /* REFERENCED */ | |
4904 | uint qfield; /* quota field to update */ | |
7bd43334 | 4905 | int state = xfs_bmap_fork_to_state(whichfork); |
70bf7533 | 4906 | struct xfs_bmbt_irec old; |
a2ceac1f | 4907 | |
79896434 BD |
4908 | mp = ip->i_mount; |
4909 | XFS_STATS_INC(mp, xs_del_exlist); | |
a2ceac1f | 4910 | |
49f693fa | 4911 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa | 4912 | ASSERT(del->br_blockcount > 0); |
9788e059 | 4913 | xfs_iext_get_extent(ifp, icur, &got); |
49f693fa DC |
4914 | ASSERT(got.br_startoff <= del->br_startoff); |
4915 | del_endoff = del->br_startoff + del->br_blockcount; | |
4916 | got_endoff = got.br_startoff + got.br_blockcount; | |
4917 | ASSERT(got_endoff >= del_endoff); | |
ad68fd19 | 4918 | ASSERT(!isnullstartblock(got.br_startblock)); |
49f693fa DC |
4919 | qfield = 0; |
4920 | error = 0; | |
ad68fd19 | 4921 | |
bd92a38b CH |
4922 | /* |
4923 | * If it's the case where the directory code is running with no block | |
4924 | * reservation, and the deleted block is in the middle of its extent, | |
4925 | * and the resulting insert of an extent would cause transformation to | |
4926 | * btree format, then reject it. The calling code will then swap blocks | |
4927 | * around instead. We have to do this now, rather than waiting for the | |
4928 | * conversion to btree format, since the transaction will be dirty then. | |
4929 | */ | |
4930 | if (tp->t_blk_res == 0 && | |
4931 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
4932 | XFS_IFORK_NEXTENTS(ip, whichfork) >= | |
4933 | XFS_IFORK_MAXEXT(ip, whichfork) && | |
4934 | del->br_startoff > got.br_startoff && del_endoff < got_endoff) | |
4935 | return -ENOSPC; | |
4936 | ||
4937 | flags = XFS_ILOG_CORE; | |
ad68fd19 CH |
4938 | if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) { |
4939 | xfs_fsblock_t bno; | |
4940 | xfs_filblks_t len; | |
4941 | ||
4942 | ASSERT(do_mod(del->br_blockcount, mp->m_sb.sb_rextsize) == 0); | |
4943 | ASSERT(do_mod(del->br_startblock, mp->m_sb.sb_rextsize) == 0); | |
4944 | bno = del->br_startblock; | |
4945 | len = del->br_blockcount; | |
4946 | do_div(bno, mp->m_sb.sb_rextsize); | |
4947 | do_div(len, mp->m_sb.sb_rextsize); | |
4948 | error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len); | |
4949 | if (error) | |
4950 | goto done; | |
49f693fa | 4951 | do_fx = 0; |
ad68fd19 CH |
4952 | nblks = len * mp->m_sb.sb_rextsize; |
4953 | qfield = XFS_TRANS_DQ_RTBCOUNT; | |
4954 | } else { | |
4955 | do_fx = 1; | |
4956 | nblks = del->br_blockcount; | |
4957 | qfield = XFS_TRANS_DQ_BCOUNT; | |
4958 | } | |
4959 | ||
4960 | del_endblock = del->br_startblock + del->br_blockcount; | |
4961 | if (cur) { | |
70a93110 | 4962 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
ad68fd19 CH |
4963 | if (error) |
4964 | goto done; | |
4965 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); | |
49f693fa | 4966 | } |
85aec44f | 4967 | |
b039ac79 CH |
4968 | if (got.br_startoff == del->br_startoff) |
4969 | state |= BMAP_LEFT_FILLING; | |
4970 | if (got_endoff == del_endoff) | |
4971 | state |= BMAP_RIGHT_FILLING; | |
4972 | ||
4973 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { | |
4974 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
49f693fa DC |
4975 | /* |
4976 | * Matches the whole extent. Delete the entry. | |
4977 | */ | |
cf455a62 | 4978 | xfs_iext_remove(ip, icur, state); |
9788e059 | 4979 | xfs_iext_prev(ifp, icur); |
49f693fa DC |
4980 | XFS_IFORK_NEXT_SET(ip, whichfork, |
4981 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
4982 | flags |= XFS_ILOG_CORE; | |
4983 | if (!cur) { | |
4984 | flags |= xfs_ilog_fext(whichfork); | |
4985 | break; | |
2bd0ea18 | 4986 | } |
49f693fa DC |
4987 | if ((error = xfs_btree_delete(cur, &i))) |
4988 | goto done; | |
19ebedcf | 4989 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 4990 | break; |
b039ac79 | 4991 | case BMAP_LEFT_FILLING: |
2bd0ea18 | 4992 | /* |
49f693fa | 4993 | * Deleting the first part of the extent. |
2bd0ea18 | 4994 | */ |
70bf7533 CH |
4995 | got.br_startoff = del_endoff; |
4996 | got.br_startblock = del_endblock; | |
4997 | got.br_blockcount -= del->br_blockcount; | |
9788e059 | 4998 | xfs_iext_update_extent(ip, state, icur, &got); |
49f693fa DC |
4999 | if (!cur) { |
5000 | flags |= xfs_ilog_fext(whichfork); | |
5001 | break; | |
5002 | } | |
d0e5f1ff | 5003 | error = xfs_bmbt_update(cur, &got); |
70bf7533 | 5004 | if (error) |
49f693fa DC |
5005 | goto done; |
5006 | break; | |
b039ac79 | 5007 | case BMAP_RIGHT_FILLING: |
2bd0ea18 | 5008 | /* |
49f693fa | 5009 | * Deleting the last part of the extent. |
2bd0ea18 | 5010 | */ |
70bf7533 | 5011 | got.br_blockcount -= del->br_blockcount; |
9788e059 | 5012 | xfs_iext_update_extent(ip, state, icur, &got); |
49f693fa DC |
5013 | if (!cur) { |
5014 | flags |= xfs_ilog_fext(whichfork); | |
5015 | break; | |
5016 | } | |
d0e5f1ff | 5017 | error = xfs_bmbt_update(cur, &got); |
70bf7533 | 5018 | if (error) |
49f693fa DC |
5019 | goto done; |
5020 | break; | |
49f693fa DC |
5021 | case 0: |
5022 | /* | |
5023 | * Deleting the middle of the extent. | |
5024 | */ | |
70bf7533 | 5025 | old = got; |
df926c07 | 5026 | |
70bf7533 | 5027 | got.br_blockcount = del->br_startoff - got.br_startoff; |
9788e059 | 5028 | xfs_iext_update_extent(ip, state, icur, &got); |
70bf7533 | 5029 | |
49f693fa | 5030 | new.br_startoff = del_endoff; |
70bf7533 | 5031 | new.br_blockcount = got_endoff - del_endoff; |
49f693fa | 5032 | new.br_state = got.br_state; |
ad68fd19 | 5033 | new.br_startblock = del_endblock; |
70bf7533 | 5034 | |
ad68fd19 CH |
5035 | flags |= XFS_ILOG_CORE; |
5036 | if (cur) { | |
d0e5f1ff | 5037 | error = xfs_bmbt_update(cur, &got); |
ad68fd19 CH |
5038 | if (error) |
5039 | goto done; | |
5040 | error = xfs_btree_increment(cur, 0, &i); | |
5041 | if (error) | |
5042 | goto done; | |
5043 | cur->bc_rec.b = new; | |
5044 | error = xfs_btree_insert(cur, &i); | |
5045 | if (error && error != -ENOSPC) | |
5046 | goto done; | |
5047 | /* | |
5048 | * If get no-space back from btree insert, it tried a | |
5049 | * split, and we have a zero block reservation. Fix up | |
5050 | * our state and return the error. | |
5051 | */ | |
5052 | if (error == -ENOSPC) { | |
49f693fa | 5053 | /* |
ad68fd19 CH |
5054 | * Reset the cursor, don't trust it after any |
5055 | * insert operation. | |
49f693fa | 5056 | */ |
70a93110 | 5057 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
ad68fd19 | 5058 | if (error) |
49f693fa | 5059 | goto done; |
19ebedcf | 5060 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
ad68fd19 CH |
5061 | /* |
5062 | * Update the btree record back | |
5063 | * to the original value. | |
5064 | */ | |
d0e5f1ff | 5065 | error = xfs_bmbt_update(cur, &old); |
ad68fd19 CH |
5066 | if (error) |
5067 | goto done; | |
5068 | /* | |
5069 | * Reset the extent record back | |
5070 | * to the original value. | |
5071 | */ | |
9788e059 | 5072 | xfs_iext_update_extent(ip, state, icur, &old); |
ad68fd19 CH |
5073 | flags = 0; |
5074 | error = -ENOSPC; | |
5075 | goto done; | |
5076 | } | |
5077 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); | |
5078 | } else | |
5079 | flags |= xfs_ilog_fext(whichfork); | |
5080 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5081 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
9788e059 | 5082 | xfs_iext_next(ifp, icur); |
26a75f67 | 5083 | xfs_iext_insert(ip, icur, &new, state); |
49f693fa | 5084 | break; |
2bd0ea18 | 5085 | } |
d7f80320 DW |
5086 | |
5087 | /* remove reverse mapping */ | |
ad68fd19 CH |
5088 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del); |
5089 | if (error) | |
5090 | goto done; | |
d7f80320 | 5091 | |
2bd0ea18 | 5092 | /* |
49f693fa | 5093 | * If we need to, add to list of extents to delete. |
2bd0ea18 | 5094 | */ |
36b16da8 | 5095 | if (do_fx && !(bflags & XFS_BMAPI_REMAP)) { |
cfe32f0d DW |
5096 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) { |
5097 | error = xfs_refcount_decrease_extent(mp, dfops, del); | |
5098 | if (error) | |
5099 | goto done; | |
5100 | } else | |
5101 | xfs_bmap_add_free(mp, dfops, del->br_startblock, | |
5102 | del->br_blockcount, NULL); | |
5103 | } | |
5104 | ||
2bd0ea18 | 5105 | /* |
49f693fa | 5106 | * Adjust inode # blocks in the file. |
2bd0ea18 | 5107 | */ |
49f693fa DC |
5108 | if (nblks) |
5109 | ip->i_d.di_nblocks -= nblks; | |
2bd0ea18 | 5110 | /* |
49f693fa | 5111 | * Adjust quota data. |
2bd0ea18 | 5112 | */ |
36b16da8 | 5113 | if (qfield && !(bflags & XFS_BMAPI_REMAP)) |
49f693fa DC |
5114 | xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks); |
5115 | ||
49f693fa DC |
5116 | done: |
5117 | *logflagsp = flags; | |
2bd0ea18 NS |
5118 | return error; |
5119 | } | |
5120 | ||
5121 | /* | |
49f693fa DC |
5122 | * Unmap (remove) blocks from a file. |
5123 | * If nexts is nonzero then the number of extents to remove is limited to | |
5124 | * that value. If not all extents in the block range can be removed then | |
5125 | * *done is set. | |
2bd0ea18 | 5126 | */ |
49f693fa | 5127 | int /* error */ |
3d36acda | 5128 | __xfs_bunmapi( |
49f693fa DC |
5129 | xfs_trans_t *tp, /* transaction pointer */ |
5130 | struct xfs_inode *ip, /* incore inode */ | |
675b5a20 | 5131 | xfs_fileoff_t start, /* first file offset deleted */ |
3d36acda | 5132 | xfs_filblks_t *rlen, /* i/o: amount remaining */ |
49f693fa DC |
5133 | int flags, /* misc flags */ |
5134 | xfs_extnum_t nexts, /* number of extents max */ | |
5135 | xfs_fsblock_t *firstblock, /* first allocated block | |
5136 | controls a.g. for allocs */ | |
3d36acda | 5137 | struct xfs_defer_ops *dfops) /* i/o: deferred updates */ |
2bd0ea18 | 5138 | { |
49f693fa DC |
5139 | xfs_btree_cur_t *cur; /* bmap btree cursor */ |
5140 | xfs_bmbt_irec_t del; /* extent being deleted */ | |
49f693fa DC |
5141 | int error; /* error return value */ |
5142 | xfs_extnum_t extno; /* extent number in list */ | |
5143 | xfs_bmbt_irec_t got; /* current extent record */ | |
5144 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
5145 | int isrt; /* freeing in rt area */ | |
49f693fa DC |
5146 | int logflags; /* transaction logging flags */ |
5147 | xfs_extlen_t mod; /* rt extent offset */ | |
5148 | xfs_mount_t *mp; /* mount structure */ | |
49f693fa DC |
5149 | int tmp_logflags; /* partial logging flags */ |
5150 | int wasdel; /* was a delayed alloc extent */ | |
5151 | int whichfork; /* data or attribute fork */ | |
5152 | xfs_fsblock_t sum; | |
3d36acda | 5153 | xfs_filblks_t len = *rlen; /* length to unmap in file */ |
594956fa | 5154 | xfs_fileoff_t max_len; |
15a8bccc | 5155 | xfs_agnumber_t prev_agno = NULLAGNUMBER, agno; |
675b5a20 | 5156 | xfs_fileoff_t end; |
9788e059 CH |
5157 | struct xfs_iext_cursor icur; |
5158 | bool done = false; | |
2bd0ea18 | 5159 | |
675b5a20 | 5160 | trace_xfs_bunmap(ip, start, len, flags, _RET_IP_); |
a2ceac1f | 5161 | |
cb8a004a DW |
5162 | whichfork = xfs_bmapi_whichfork(flags); |
5163 | ASSERT(whichfork != XFS_COW_FORK); | |
a2ceac1f | 5164 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa DC |
5165 | if (unlikely( |
5166 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5167 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
5168 | XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW, | |
5169 | ip->i_mount); | |
12b53197 | 5170 | return -EFSCORRUPTED; |
49f693fa DC |
5171 | } |
5172 | mp = ip->i_mount; | |
5173 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 5174 | return -EIO; |
56b2de80 | 5175 | |
ff105f75 | 5176 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
49f693fa DC |
5177 | ASSERT(len > 0); |
5178 | ASSERT(nexts >= 0); | |
56b2de80 | 5179 | |
594956fa DW |
5180 | /* |
5181 | * Guesstimate how many blocks we can unmap without running the risk of | |
5182 | * blowing out the transaction with a mix of EFIs and reflink | |
5183 | * adjustments. | |
5184 | */ | |
db2cf1d2 | 5185 | if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) |
594956fa DW |
5186 | max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res)); |
5187 | else | |
5188 | max_len = len; | |
5189 | ||
49f693fa DC |
5190 | if (!(ifp->if_flags & XFS_IFEXTENTS) && |
5191 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
5192 | return error; | |
d09d4e5f | 5193 | if (xfs_iext_count(ifp) == 0) { |
3d36acda | 5194 | *rlen = 0; |
49f693fa | 5195 | return 0; |
56b2de80 | 5196 | } |
79896434 | 5197 | XFS_STATS_INC(mp, xs_blk_unmap); |
49f693fa | 5198 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); |
d6fbe8fe | 5199 | end = start + len; |
a2ceac1f | 5200 | |
9788e059 | 5201 | if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) { |
d6fbe8fe CH |
5202 | *rlen = 0; |
5203 | return 0; | |
49f693fa | 5204 | } |
d6fbe8fe | 5205 | end--; |
246eb90a | 5206 | |
49f693fa DC |
5207 | logflags = 0; |
5208 | if (ifp->if_flags & XFS_IFBROOT) { | |
5209 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
5210 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5211 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 5212 | cur->bc_private.b.dfops = dfops; |
49f693fa DC |
5213 | cur->bc_private.b.flags = 0; |
5214 | } else | |
5215 | cur = NULL; | |
a2ceac1f | 5216 | |
49f693fa | 5217 | if (isrt) { |
a2ceac1f | 5218 | /* |
49f693fa | 5219 | * Synchronize by locking the bitmap inode. |
a2ceac1f | 5220 | */ |
a62ed6d3 | 5221 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
49f693fa | 5222 | xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
a62ed6d3 DW |
5223 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
5224 | xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL); | |
49f693fa | 5225 | } |
a2ceac1f | 5226 | |
49f693fa | 5227 | extno = 0; |
9788e059 | 5228 | while (end != (xfs_fileoff_t)-1 && end >= start && |
594956fa | 5229 | (nexts == 0 || extno < nexts) && max_len > 0) { |
a2ceac1f | 5230 | /* |
675b5a20 | 5231 | * Is the found extent after a hole in which end lives? |
49f693fa | 5232 | * Just back up to the previous extent, if so. |
a2ceac1f | 5233 | */ |
9788e059 CH |
5234 | if (got.br_startoff > end && |
5235 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5236 | done = true; | |
5237 | break; | |
a2ceac1f | 5238 | } |
49f693fa DC |
5239 | /* |
5240 | * Is the last block of this extent before the range | |
5241 | * we're supposed to delete? If so, we're done. | |
5242 | */ | |
675b5a20 | 5243 | end = XFS_FILEOFF_MIN(end, |
49f693fa | 5244 | got.br_startoff + got.br_blockcount - 1); |
675b5a20 | 5245 | if (end < start) |
49f693fa DC |
5246 | break; |
5247 | /* | |
5248 | * Then deal with the (possibly delayed) allocated space | |
5249 | * we found. | |
5250 | */ | |
49f693fa DC |
5251 | del = got; |
5252 | wasdel = isnullstartblock(del.br_startblock); | |
15a8bccc CH |
5253 | |
5254 | /* | |
5255 | * Make sure we don't touch multiple AGF headers out of order | |
5256 | * in a single transaction, as that could cause AB-BA deadlocks. | |
5257 | */ | |
5258 | if (!wasdel) { | |
5259 | agno = XFS_FSB_TO_AGNO(mp, del.br_startblock); | |
5260 | if (prev_agno != NULLAGNUMBER && prev_agno > agno) | |
5261 | break; | |
5262 | prev_agno = agno; | |
5263 | } | |
49f693fa DC |
5264 | if (got.br_startoff < start) { |
5265 | del.br_startoff = start; | |
5266 | del.br_blockcount -= start - got.br_startoff; | |
5267 | if (!wasdel) | |
5268 | del.br_startblock += start - got.br_startoff; | |
5269 | } | |
675b5a20 CH |
5270 | if (del.br_startoff + del.br_blockcount > end + 1) |
5271 | del.br_blockcount = end + 1 - del.br_startoff; | |
594956fa DW |
5272 | |
5273 | /* How much can we safely unmap? */ | |
5274 | if (max_len < del.br_blockcount) { | |
5275 | del.br_startoff += del.br_blockcount - max_len; | |
5276 | if (!wasdel) | |
5277 | del.br_startblock += del.br_blockcount - max_len; | |
5278 | del.br_blockcount = max_len; | |
5279 | } | |
5280 | ||
49f693fa DC |
5281 | sum = del.br_startblock + del.br_blockcount; |
5282 | if (isrt && | |
5283 | (mod = do_mod(sum, mp->m_sb.sb_rextsize))) { | |
5284 | /* | |
5285 | * Realtime extent not lined up at the end. | |
5286 | * The extent could have been split into written | |
5287 | * and unwritten pieces, or we could just be | |
5288 | * unmapping part of it. But we can't really | |
5289 | * get rid of part of a realtime extent. | |
5290 | */ | |
5291 | if (del.br_state == XFS_EXT_UNWRITTEN || | |
5292 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { | |
5293 | /* | |
5294 | * This piece is unwritten, or we're not | |
5295 | * using unwritten extents. Skip over it. | |
5296 | */ | |
675b5a20 CH |
5297 | ASSERT(end >= mod); |
5298 | end -= mod > del.br_blockcount ? | |
49f693fa | 5299 | del.br_blockcount : mod; |
9788e059 CH |
5300 | if (end < got.br_startoff && |
5301 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5302 | done = true; | |
5303 | break; | |
49f693fa DC |
5304 | } |
5305 | continue; | |
5306 | } | |
5307 | /* | |
5308 | * It's written, turn it unwritten. | |
5309 | * This is better than zeroing it. | |
5310 | */ | |
5311 | ASSERT(del.br_state == XFS_EXT_NORM); | |
0268fdc3 | 5312 | ASSERT(tp->t_blk_res > 0); |
49f693fa DC |
5313 | /* |
5314 | * If this spans a realtime extent boundary, | |
5315 | * chop it back to the start of the one we end at. | |
5316 | */ | |
5317 | if (del.br_blockcount > mod) { | |
5318 | del.br_startoff += del.br_blockcount - mod; | |
5319 | del.br_startblock += del.br_blockcount - mod; | |
5320 | del.br_blockcount = mod; | |
5321 | } | |
5322 | del.br_state = XFS_EXT_UNWRITTEN; | |
5323 | error = xfs_bmap_add_extent_unwritten_real(tp, ip, | |
9788e059 | 5324 | whichfork, &icur, &cur, &del, |
4072e4b4 | 5325 | firstblock, dfops, &logflags); |
49f693fa DC |
5326 | if (error) |
5327 | goto error0; | |
5328 | goto nodelete; | |
a2ceac1f | 5329 | } |
49f693fa DC |
5330 | if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) { |
5331 | /* | |
5332 | * Realtime extent is lined up at the end but not | |
5333 | * at the front. We'll get rid of full extents if | |
5334 | * we can. | |
5335 | */ | |
5336 | mod = mp->m_sb.sb_rextsize - mod; | |
5337 | if (del.br_blockcount > mod) { | |
5338 | del.br_blockcount -= mod; | |
5339 | del.br_startoff += mod; | |
5340 | del.br_startblock += mod; | |
5341 | } else if ((del.br_startoff == start && | |
5342 | (del.br_state == XFS_EXT_UNWRITTEN || | |
0268fdc3 | 5343 | tp->t_blk_res == 0)) || |
49f693fa DC |
5344 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
5345 | /* | |
5346 | * Can't make it unwritten. There isn't | |
5347 | * a full extent here so just skip it. | |
5348 | */ | |
675b5a20 CH |
5349 | ASSERT(end >= del.br_blockcount); |
5350 | end -= del.br_blockcount; | |
9788e059 CH |
5351 | if (got.br_startoff > end && |
5352 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5353 | done = true; | |
5354 | break; | |
5355 | } | |
49f693fa DC |
5356 | continue; |
5357 | } else if (del.br_state == XFS_EXT_UNWRITTEN) { | |
246eb90a CH |
5358 | struct xfs_bmbt_irec prev; |
5359 | ||
49f693fa DC |
5360 | /* |
5361 | * This one is already unwritten. | |
5362 | * It must have a written left neighbor. | |
5363 | * Unwrite the killed part of that one and | |
5364 | * try again. | |
5365 | */ | |
9788e059 CH |
5366 | if (!xfs_iext_prev_extent(ifp, &icur, &prev)) |
5367 | ASSERT(0); | |
49f693fa DC |
5368 | ASSERT(prev.br_state == XFS_EXT_NORM); |
5369 | ASSERT(!isnullstartblock(prev.br_startblock)); | |
5370 | ASSERT(del.br_startblock == | |
5371 | prev.br_startblock + prev.br_blockcount); | |
5372 | if (prev.br_startoff < start) { | |
5373 | mod = start - prev.br_startoff; | |
5374 | prev.br_blockcount -= mod; | |
5375 | prev.br_startblock += mod; | |
5376 | prev.br_startoff = start; | |
5377 | } | |
5378 | prev.br_state = XFS_EXT_UNWRITTEN; | |
49f693fa | 5379 | error = xfs_bmap_add_extent_unwritten_real(tp, |
9788e059 | 5380 | ip, whichfork, &icur, &cur, |
4072e4b4 DW |
5381 | &prev, firstblock, dfops, |
5382 | &logflags); | |
49f693fa DC |
5383 | if (error) |
5384 | goto error0; | |
5385 | goto nodelete; | |
5386 | } else { | |
5387 | ASSERT(del.br_state == XFS_EXT_NORM); | |
5388 | del.br_state = XFS_EXT_UNWRITTEN; | |
5389 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
9788e059 | 5390 | ip, whichfork, &icur, &cur, |
4072e4b4 DW |
5391 | &del, firstblock, dfops, |
5392 | &logflags); | |
49f693fa DC |
5393 | if (error) |
5394 | goto error0; | |
5395 | goto nodelete; | |
5396 | } | |
5397 | } | |
a2ceac1f | 5398 | |
8359e0b9 | 5399 | if (wasdel) { |
9788e059 | 5400 | error = xfs_bmap_del_extent_delay(ip, whichfork, &icur, |
ad68fd19 CH |
5401 | &got, &del); |
5402 | } else { | |
9788e059 | 5403 | error = xfs_bmap_del_extent_real(ip, tp, &icur, dfops, |
ad68fd19 CH |
5404 | cur, &del, &tmp_logflags, whichfork, |
5405 | flags); | |
5406 | logflags |= tmp_logflags; | |
e9fa15aa | 5407 | } |
8359e0b9 | 5408 | |
49f693fa DC |
5409 | if (error) |
5410 | goto error0; | |
8359e0b9 | 5411 | |
594956fa | 5412 | max_len -= del.br_blockcount; |
675b5a20 | 5413 | end = del.br_startoff - 1; |
49f693fa | 5414 | nodelete: |
a2ceac1f | 5415 | /* |
49f693fa | 5416 | * If not done go on to the next (previous) record. |
a2ceac1f | 5417 | */ |
675b5a20 | 5418 | if (end != (xfs_fileoff_t)-1 && end >= start) { |
9788e059 CH |
5419 | if (!xfs_iext_get_extent(ifp, &icur, &got) || |
5420 | (got.br_startoff > end && | |
5421 | !xfs_iext_prev_extent(ifp, &icur, &got))) { | |
5422 | done = true; | |
5423 | break; | |
49f693fa DC |
5424 | } |
5425 | extno++; | |
a2ceac1f | 5426 | } |
a2ceac1f | 5427 | } |
9788e059 | 5428 | if (done || end == (xfs_fileoff_t)-1 || end < start) |
3d36acda DW |
5429 | *rlen = 0; |
5430 | else | |
675b5a20 | 5431 | *rlen = end - start + 1; |
56b2de80 | 5432 | |
49f693fa DC |
5433 | /* |
5434 | * Convert to a btree if necessary. | |
5435 | */ | |
5436 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
5437 | ASSERT(cur == NULL); | |
f33cea1a | 5438 | error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, |
49f693fa DC |
5439 | &cur, 0, &tmp_logflags, whichfork); |
5440 | logflags |= tmp_logflags; | |
5441 | if (error) | |
5442 | goto error0; | |
56b2de80 | 5443 | } |
56b2de80 | 5444 | /* |
49f693fa | 5445 | * transform from btree to extents, give it cur |
56b2de80 | 5446 | */ |
49f693fa DC |
5447 | else if (xfs_bmap_wants_extents(ip, whichfork)) { |
5448 | ASSERT(cur != NULL); | |
5449 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags, | |
5450 | whichfork); | |
5451 | logflags |= tmp_logflags; | |
5452 | if (error) | |
5453 | goto error0; | |
56b2de80 | 5454 | } |
49f693fa DC |
5455 | /* |
5456 | * transform from extents to local? | |
5457 | */ | |
5458 | error = 0; | |
5459 | error0: | |
5460 | /* | |
5461 | * Log everything. Do this after conversion, there's no point in | |
5462 | * logging the extent records if we've converted to btree format. | |
5463 | */ | |
5464 | if ((logflags & xfs_ilog_fext(whichfork)) && | |
5465 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
5466 | logflags &= ~xfs_ilog_fext(whichfork); | |
5467 | else if ((logflags & xfs_ilog_fbroot(whichfork)) && | |
5468 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) | |
5469 | logflags &= ~xfs_ilog_fbroot(whichfork); | |
5470 | /* | |
5471 | * Log inode even in the error case, if the transaction | |
5472 | * is dirty we'll need to shut down the filesystem. | |
5473 | */ | |
5474 | if (logflags) | |
5475 | xfs_trans_log_inode(tp, ip, logflags); | |
5476 | if (cur) { | |
5477 | if (!error) { | |
5478 | *firstblock = cur->bc_private.b.firstblock; | |
5479 | cur->bc_private.b.allocated = 0; | |
56b2de80 | 5480 | } |
49f693fa DC |
5481 | xfs_btree_del_cursor(cur, |
5482 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
56b2de80 | 5483 | } |
49f693fa | 5484 | return error; |
a2ceac1f | 5485 | } |
ff105f75 | 5486 | |
3d36acda DW |
5487 | /* Unmap a range of a file. */ |
5488 | int | |
5489 | xfs_bunmapi( | |
5490 | xfs_trans_t *tp, | |
5491 | struct xfs_inode *ip, | |
5492 | xfs_fileoff_t bno, | |
5493 | xfs_filblks_t len, | |
5494 | int flags, | |
5495 | xfs_extnum_t nexts, | |
5496 | xfs_fsblock_t *firstblock, | |
5497 | struct xfs_defer_ops *dfops, | |
5498 | int *done) | |
5499 | { | |
5500 | int error; | |
5501 | ||
5502 | error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock, | |
5503 | dfops); | |
5504 | *done = (len == 0); | |
5505 | return error; | |
5506 | } | |
5507 | ||
5a35bf2c DC |
5508 | /* |
5509 | * Determine whether an extent shift can be accomplished by a merge with the | |
5510 | * extent that precedes the target hole of the shift. | |
5511 | */ | |
5512 | STATIC bool | |
5513 | xfs_bmse_can_merge( | |
5514 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5515 | struct xfs_bmbt_irec *got, /* current extent to shift */ | |
5516 | xfs_fileoff_t shift) /* shift fsb */ | |
5517 | { | |
5518 | xfs_fileoff_t startoff; | |
5519 | ||
5520 | startoff = got->br_startoff - shift; | |
5521 | ||
5522 | /* | |
5523 | * The extent, once shifted, must be adjacent in-file and on-disk with | |
5524 | * the preceding extent. | |
5525 | */ | |
5526 | if ((left->br_startoff + left->br_blockcount != startoff) || | |
5527 | (left->br_startblock + left->br_blockcount != got->br_startblock) || | |
5528 | (left->br_state != got->br_state) || | |
5529 | (left->br_blockcount + got->br_blockcount > MAXEXTLEN)) | |
5530 | return false; | |
5531 | ||
5532 | return true; | |
5533 | } | |
5534 | ||
5535 | /* | |
5536 | * A bmap extent shift adjusts the file offset of an extent to fill a preceding | |
5537 | * hole in the file. If an extent shift would result in the extent being fully | |
5538 | * adjacent to the extent that currently precedes the hole, we can merge with | |
5539 | * the preceding extent rather than do the shift. | |
5540 | * | |
5541 | * This function assumes the caller has verified a shift-by-merge is possible | |
5542 | * with the provided extents via xfs_bmse_can_merge(). | |
5543 | */ | |
5544 | STATIC int | |
5545 | xfs_bmse_merge( | |
5546 | struct xfs_inode *ip, | |
5547 | int whichfork, | |
5548 | xfs_fileoff_t shift, /* shift fsb */ | |
9788e059 | 5549 | struct xfs_iext_cursor *icur, |
2f082827 CH |
5550 | struct xfs_bmbt_irec *got, /* extent to shift */ |
5551 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5a35bf2c | 5552 | struct xfs_btree_cur *cur, |
2f082827 CH |
5553 | int *logflags, /* output */ |
5554 | struct xfs_defer_ops *dfops) | |
5a35bf2c | 5555 | { |
2f082827 | 5556 | struct xfs_bmbt_irec new; |
5a35bf2c DC |
5557 | xfs_filblks_t blockcount; |
5558 | int error, i; | |
19ebedcf | 5559 | struct xfs_mount *mp = ip->i_mount; |
5a35bf2c | 5560 | |
2f082827 | 5561 | blockcount = left->br_blockcount + got->br_blockcount; |
5a35bf2c DC |
5562 | |
5563 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5564 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
2f082827 | 5565 | ASSERT(xfs_bmse_can_merge(left, got, shift)); |
5a35bf2c | 5566 | |
2f082827 CH |
5567 | new = *left; |
5568 | new.br_blockcount = blockcount; | |
5a35bf2c DC |
5569 | |
5570 | /* | |
5571 | * Update the on-disk extent count, the btree if necessary and log the | |
5572 | * inode. | |
5573 | */ | |
5574 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5575 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5576 | *logflags |= XFS_ILOG_CORE; | |
5577 | if (!cur) { | |
5578 | *logflags |= XFS_ILOG_DEXT; | |
2f082827 | 5579 | goto done; |
5a35bf2c DC |
5580 | } |
5581 | ||
5582 | /* lookup and remove the extent to merge */ | |
70a93110 | 5583 | error = xfs_bmbt_lookup_eq(cur, got, &i); |
5a35bf2c DC |
5584 | if (error) |
5585 | return error; | |
19ebedcf | 5586 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5587 | |
5588 | error = xfs_btree_delete(cur, &i); | |
5589 | if (error) | |
5590 | return error; | |
19ebedcf | 5591 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5592 | |
5593 | /* lookup and update size of the previous extent */ | |
70a93110 | 5594 | error = xfs_bmbt_lookup_eq(cur, left, &i); |
5a35bf2c DC |
5595 | if (error) |
5596 | return error; | |
19ebedcf | 5597 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c | 5598 | |
d0e5f1ff | 5599 | error = xfs_bmbt_update(cur, &new); |
2f082827 CH |
5600 | if (error) |
5601 | return error; | |
5602 | ||
5603 | done: | |
cf455a62 | 5604 | xfs_iext_remove(ip, icur, 0); |
9788e059 CH |
5605 | xfs_iext_prev(XFS_IFORK_PTR(ip, whichfork), icur); |
5606 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur, | |
5607 | &new); | |
5a35bf2c | 5608 | |
479284b7 | 5609 | /* update reverse mapping. rmap functions merge the rmaps for us */ |
2f082827 CH |
5610 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, got); |
5611 | if (error) | |
5612 | return error; | |
479284b7 DW |
5613 | memcpy(&new, got, sizeof(new)); |
5614 | new.br_startoff = left->br_startoff + left->br_blockcount; | |
2f082827 | 5615 | return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &new); |
5a35bf2c DC |
5616 | } |
5617 | ||
364937e2 CH |
5618 | static int |
5619 | xfs_bmap_shift_update_extent( | |
5620 | struct xfs_inode *ip, | |
5621 | int whichfork, | |
9788e059 | 5622 | struct xfs_iext_cursor *icur, |
364937e2 CH |
5623 | struct xfs_bmbt_irec *got, |
5624 | struct xfs_btree_cur *cur, | |
5625 | int *logflags, | |
5626 | struct xfs_defer_ops *dfops, | |
5627 | xfs_fileoff_t startoff) | |
5a35bf2c | 5628 | { |
364937e2 | 5629 | struct xfs_mount *mp = ip->i_mount; |
f36ccac2 | 5630 | struct xfs_bmbt_irec prev = *got; |
364937e2 | 5631 | int error, i; |
2f082827 | 5632 | |
5a35bf2c | 5633 | *logflags |= XFS_ILOG_CORE; |
2f082827 | 5634 | |
f36ccac2 | 5635 | got->br_startoff = startoff; |
2f082827 CH |
5636 | |
5637 | if (cur) { | |
f36ccac2 | 5638 | error = xfs_bmbt_lookup_eq(cur, &prev, &i); |
2f082827 CH |
5639 | if (error) |
5640 | return error; | |
5641 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); | |
5642 | ||
f36ccac2 | 5643 | error = xfs_bmbt_update(cur, got); |
2f082827 CH |
5644 | if (error) |
5645 | return error; | |
5646 | } else { | |
5a35bf2c | 5647 | *logflags |= XFS_ILOG_DEXT; |
5a35bf2c DC |
5648 | } |
5649 | ||
9788e059 CH |
5650 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur, |
5651 | got); | |
d7f80320 | 5652 | |
d7f80320 | 5653 | /* update reverse mapping */ |
f36ccac2 | 5654 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &prev); |
d7f80320 DW |
5655 | if (error) |
5656 | return error; | |
f36ccac2 | 5657 | return xfs_rmap_map_extent(mp, dfops, ip, whichfork, got); |
5a35bf2c DC |
5658 | } |
5659 | ||
ff105f75 | 5660 | int |
a9f5d25e | 5661 | xfs_bmap_collapse_extents( |
ff105f75 DC |
5662 | struct xfs_trans *tp, |
5663 | struct xfs_inode *ip, | |
19ebedcf | 5664 | xfs_fileoff_t *next_fsb, |
ff105f75 | 5665 | xfs_fileoff_t offset_shift_fsb, |
a9f5d25e | 5666 | bool *done, |
ff105f75 | 5667 | xfs_fsblock_t *firstblock, |
a9f5d25e | 5668 | struct xfs_defer_ops *dfops) |
ff105f75 | 5669 | { |
a9f5d25e CH |
5670 | int whichfork = XFS_DATA_FORK; |
5671 | struct xfs_mount *mp = ip->i_mount; | |
5672 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5673 | struct xfs_btree_cur *cur = NULL; | |
364937e2 | 5674 | struct xfs_bmbt_irec got, prev; |
9788e059 | 5675 | struct xfs_iext_cursor icur; |
364937e2 | 5676 | xfs_fileoff_t new_startoff; |
a9f5d25e CH |
5677 | int error = 0; |
5678 | int logflags = 0; | |
ff105f75 DC |
5679 | |
5680 | if (unlikely(XFS_TEST_ERROR( | |
5681 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5682 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 5683 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a9f5d25e | 5684 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); |
12b53197 | 5685 | return -EFSCORRUPTED; |
ff105f75 DC |
5686 | } |
5687 | ||
5688 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 5689 | return -EIO; |
ff105f75 | 5690 | |
a9f5d25e | 5691 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); |
ff105f75 | 5692 | |
ff105f75 | 5693 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
ff105f75 DC |
5694 | error = xfs_iread_extents(tp, ip, whichfork); |
5695 | if (error) | |
5696 | return error; | |
5697 | } | |
5698 | ||
ff105f75 DC |
5699 | if (ifp->if_flags & XFS_IFBROOT) { |
5700 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5701 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 5702 | cur->bc_private.b.dfops = dfops; |
ff105f75 | 5703 | cur->bc_private.b.flags = 0; |
5a35bf2c DC |
5704 | } |
5705 | ||
9788e059 | 5706 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) { |
a9f5d25e CH |
5707 | *done = true; |
5708 | goto del_cursor; | |
5709 | } | |
5c40ca1f ES |
5710 | XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock), |
5711 | del_cursor); | |
a9f5d25e | 5712 | |
364937e2 | 5713 | new_startoff = got.br_startoff - offset_shift_fsb; |
9788e059 | 5714 | if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) { |
364937e2 CH |
5715 | if (new_startoff < prev.br_startoff + prev.br_blockcount) { |
5716 | error = -EINVAL; | |
5717 | goto del_cursor; | |
5718 | } | |
5719 | ||
364937e2 CH |
5720 | if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) { |
5721 | error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb, | |
9788e059 CH |
5722 | &icur, &got, &prev, cur, &logflags, |
5723 | dfops); | |
364937e2 CH |
5724 | if (error) |
5725 | goto del_cursor; | |
5726 | goto done; | |
5727 | } | |
5728 | } else { | |
5729 | if (got.br_startoff < offset_shift_fsb) { | |
5730 | error = -EINVAL; | |
5731 | goto del_cursor; | |
5732 | } | |
5733 | } | |
5734 | ||
9788e059 CH |
5735 | error = xfs_bmap_shift_update_extent(ip, whichfork, &icur, &got, cur, |
5736 | &logflags, dfops, new_startoff); | |
a9f5d25e CH |
5737 | if (error) |
5738 | goto del_cursor; | |
bac20498 | 5739 | |
c162e319 | 5740 | done: |
9788e059 CH |
5741 | if (!xfs_iext_next_extent(ifp, &icur, &got)) { |
5742 | *done = true; | |
5743 | goto del_cursor; | |
a9f5d25e | 5744 | } |
a9f5d25e | 5745 | |
364937e2 | 5746 | *next_fsb = got.br_startoff; |
a9f5d25e CH |
5747 | del_cursor: |
5748 | if (cur) | |
5749 | xfs_btree_del_cursor(cur, | |
5750 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
a9f5d25e CH |
5751 | if (logflags) |
5752 | xfs_trans_log_inode(tp, ip, logflags); | |
a9f5d25e CH |
5753 | return error; |
5754 | } | |
5755 | ||
5756 | int | |
5757 | xfs_bmap_insert_extents( | |
5758 | struct xfs_trans *tp, | |
5759 | struct xfs_inode *ip, | |
5760 | xfs_fileoff_t *next_fsb, | |
5761 | xfs_fileoff_t offset_shift_fsb, | |
5762 | bool *done, | |
5763 | xfs_fileoff_t stop_fsb, | |
5764 | xfs_fsblock_t *firstblock, | |
5765 | struct xfs_defer_ops *dfops) | |
5766 | { | |
5767 | int whichfork = XFS_DATA_FORK; | |
5768 | struct xfs_mount *mp = ip->i_mount; | |
5769 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5770 | struct xfs_btree_cur *cur = NULL; | |
bd80a804 | 5771 | struct xfs_bmbt_irec got, next; |
9788e059 | 5772 | struct xfs_iext_cursor icur; |
364937e2 | 5773 | xfs_fileoff_t new_startoff; |
a9f5d25e CH |
5774 | int error = 0; |
5775 | int logflags = 0; | |
5776 | ||
5777 | if (unlikely(XFS_TEST_ERROR( | |
5778 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5779 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
5780 | mp, XFS_ERRTAG_BMAPIFORMAT))) { | |
5781 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
5782 | return -EFSCORRUPTED; | |
5783 | } | |
5784 | ||
5785 | if (XFS_FORCED_SHUTDOWN(mp)) | |
5786 | return -EIO; | |
5787 | ||
5788 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); | |
5789 | ||
5790 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
5791 | error = xfs_iread_extents(tp, ip, whichfork); | |
5792 | if (error) | |
5793 | return error; | |
5794 | } | |
5795 | ||
5796 | if (ifp->if_flags & XFS_IFBROOT) { | |
5797 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5798 | cur->bc_private.b.firstblock = *firstblock; | |
5799 | cur->bc_private.b.dfops = dfops; | |
5800 | cur->bc_private.b.flags = 0; | |
5801 | } | |
5802 | ||
19ebedcf | 5803 | if (*next_fsb == NULLFSBLOCK) { |
9788e059 CH |
5804 | xfs_iext_last(ifp, &icur); |
5805 | if (!xfs_iext_get_extent(ifp, &icur, &got) || | |
bd80a804 | 5806 | stop_fsb > got.br_startoff) { |
a9f5d25e | 5807 | *done = true; |
19ebedcf DC |
5808 | goto del_cursor; |
5809 | } | |
b6ad780b | 5810 | } else { |
9788e059 | 5811 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) { |
a9f5d25e | 5812 | *done = true; |
b6ad780b CH |
5813 | goto del_cursor; |
5814 | } | |
19ebedcf | 5815 | } |
5c40ca1f ES |
5816 | XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock), |
5817 | del_cursor); | |
19ebedcf | 5818 | |
bd80a804 | 5819 | if (stop_fsb >= got.br_startoff + got.br_blockcount) { |
a9f5d25e CH |
5820 | error = -EIO; |
5821 | goto del_cursor; | |
19ebedcf DC |
5822 | } |
5823 | ||
364937e2 | 5824 | new_startoff = got.br_startoff + offset_shift_fsb; |
9788e059 | 5825 | if (xfs_iext_peek_next_extent(ifp, &icur, &next)) { |
364937e2 CH |
5826 | if (new_startoff + got.br_blockcount > next.br_startoff) { |
5827 | error = -EINVAL; | |
5828 | goto del_cursor; | |
5829 | } | |
5830 | ||
5831 | /* | |
5832 | * Unlike a left shift (which involves a hole punch), a right | |
5833 | * shift does not modify extent neighbors in any way. We should | |
5834 | * never find mergeable extents in this scenario. Check anyways | |
5835 | * and warn if we encounter two extents that could be one. | |
5836 | */ | |
5837 | if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb)) | |
5838 | WARN_ON_ONCE(1); | |
5839 | } | |
5840 | ||
9788e059 CH |
5841 | error = xfs_bmap_shift_update_extent(ip, whichfork, &icur, &got, cur, |
5842 | &logflags, dfops, new_startoff); | |
6835c363 CH |
5843 | if (error) |
5844 | goto del_cursor; | |
bd80a804 | 5845 | |
9788e059 | 5846 | if (!xfs_iext_prev_extent(ifp, &icur, &got) || |
bd80a804 | 5847 | stop_fsb >= got.br_startoff + got.br_blockcount) { |
a9f5d25e | 5848 | *done = true; |
6835c363 | 5849 | goto del_cursor; |
ff105f75 DC |
5850 | } |
5851 | ||
6835c363 | 5852 | *next_fsb = got.br_startoff; |
ff105f75 DC |
5853 | del_cursor: |
5854 | if (cur) | |
5855 | xfs_btree_del_cursor(cur, | |
5856 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
5a35bf2c DC |
5857 | if (logflags) |
5858 | xfs_trans_log_inode(tp, ip, logflags); | |
ff105f75 DC |
5859 | return error; |
5860 | } | |
19ebedcf DC |
5861 | |
5862 | /* | |
9788e059 CH |
5863 | * Splits an extent into two extents at split_fsb block such that it is the |
5864 | * first block of the current_ext. @ext is a target extent to be split. | |
5865 | * @split_fsb is a block where the extents is split. If split_fsb lies in a | |
5866 | * hole or the first block of extents, just return 0. | |
19ebedcf DC |
5867 | */ |
5868 | STATIC int | |
5869 | xfs_bmap_split_extent_at( | |
5870 | struct xfs_trans *tp, | |
5871 | struct xfs_inode *ip, | |
5872 | xfs_fileoff_t split_fsb, | |
5873 | xfs_fsblock_t *firstfsb, | |
f33cea1a | 5874 | struct xfs_defer_ops *dfops) |
19ebedcf DC |
5875 | { |
5876 | int whichfork = XFS_DATA_FORK; | |
5877 | struct xfs_btree_cur *cur = NULL; | |
19ebedcf DC |
5878 | struct xfs_bmbt_irec got; |
5879 | struct xfs_bmbt_irec new; /* split extent */ | |
5880 | struct xfs_mount *mp = ip->i_mount; | |
5881 | struct xfs_ifork *ifp; | |
5882 | xfs_fsblock_t gotblkcnt; /* new block count for got */ | |
9788e059 | 5883 | struct xfs_iext_cursor icur; |
19ebedcf DC |
5884 | int error = 0; |
5885 | int logflags = 0; | |
5886 | int i = 0; | |
5887 | ||
5888 | if (unlikely(XFS_TEST_ERROR( | |
5889 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5890 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 5891 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
19ebedcf DC |
5892 | XFS_ERROR_REPORT("xfs_bmap_split_extent_at", |
5893 | XFS_ERRLEVEL_LOW, mp); | |
5894 | return -EFSCORRUPTED; | |
5895 | } | |
5896 | ||
5897 | if (XFS_FORCED_SHUTDOWN(mp)) | |
5898 | return -EIO; | |
5899 | ||
5900 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
5901 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
5902 | /* Read in all the extents */ | |
5903 | error = xfs_iread_extents(tp, ip, whichfork); | |
5904 | if (error) | |
5905 | return error; | |
5906 | } | |
5907 | ||
5908 | /* | |
455044a6 | 5909 | * If there are not extents, or split_fsb lies in a hole we are done. |
19ebedcf | 5910 | */ |
9788e059 | 5911 | if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) || |
455044a6 | 5912 | got.br_startoff >= split_fsb) |
19ebedcf DC |
5913 | return 0; |
5914 | ||
5915 | gotblkcnt = split_fsb - got.br_startoff; | |
5916 | new.br_startoff = split_fsb; | |
5917 | new.br_startblock = got.br_startblock + gotblkcnt; | |
5918 | new.br_blockcount = got.br_blockcount - gotblkcnt; | |
5919 | new.br_state = got.br_state; | |
5920 | ||
5921 | if (ifp->if_flags & XFS_IFBROOT) { | |
5922 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5923 | cur->bc_private.b.firstblock = *firstfsb; | |
f33cea1a | 5924 | cur->bc_private.b.dfops = dfops; |
19ebedcf | 5925 | cur->bc_private.b.flags = 0; |
70a93110 | 5926 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
19ebedcf DC |
5927 | if (error) |
5928 | goto del_cursor; | |
5929 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
5930 | } | |
5931 | ||
19ebedcf | 5932 | got.br_blockcount = gotblkcnt; |
9788e059 CH |
5933 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur, |
5934 | &got); | |
19ebedcf DC |
5935 | |
5936 | logflags = XFS_ILOG_CORE; | |
5937 | if (cur) { | |
d0e5f1ff | 5938 | error = xfs_bmbt_update(cur, &got); |
19ebedcf DC |
5939 | if (error) |
5940 | goto del_cursor; | |
5941 | } else | |
5942 | logflags |= XFS_ILOG_DEXT; | |
5943 | ||
5944 | /* Add new extent */ | |
9788e059 | 5945 | xfs_iext_next(ifp, &icur); |
26a75f67 | 5946 | xfs_iext_insert(ip, &icur, &new, 0); |
19ebedcf DC |
5947 | XFS_IFORK_NEXT_SET(ip, whichfork, |
5948 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
5949 | ||
5950 | if (cur) { | |
70a93110 | 5951 | error = xfs_bmbt_lookup_eq(cur, &new, &i); |
19ebedcf DC |
5952 | if (error) |
5953 | goto del_cursor; | |
5954 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor); | |
19ebedcf DC |
5955 | error = xfs_btree_insert(cur, &i); |
5956 | if (error) | |
5957 | goto del_cursor; | |
5958 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
5959 | } | |
5960 | ||
5961 | /* | |
5962 | * Convert to a btree if necessary. | |
5963 | */ | |
5964 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
5965 | int tmp_logflags; /* partial log flag return val */ | |
5966 | ||
5967 | ASSERT(cur == NULL); | |
f33cea1a | 5968 | error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops, |
19ebedcf DC |
5969 | &cur, 0, &tmp_logflags, whichfork); |
5970 | logflags |= tmp_logflags; | |
5971 | } | |
5972 | ||
5973 | del_cursor: | |
5974 | if (cur) { | |
5975 | cur->bc_private.b.allocated = 0; | |
5976 | xfs_btree_del_cursor(cur, | |
5977 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
5978 | } | |
5979 | ||
5980 | if (logflags) | |
5981 | xfs_trans_log_inode(tp, ip, logflags); | |
5982 | return error; | |
5983 | } | |
5984 | ||
5985 | int | |
5986 | xfs_bmap_split_extent( | |
5987 | struct xfs_inode *ip, | |
5988 | xfs_fileoff_t split_fsb) | |
5989 | { | |
5990 | struct xfs_mount *mp = ip->i_mount; | |
5991 | struct xfs_trans *tp; | |
f33cea1a | 5992 | struct xfs_defer_ops dfops; |
19ebedcf | 5993 | xfs_fsblock_t firstfsb; |
19ebedcf DC |
5994 | int error; |
5995 | ||
9074815c CH |
5996 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, |
5997 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); | |
5998 | if (error) | |
19ebedcf | 5999 | return error; |
19ebedcf DC |
6000 | |
6001 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
6002 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
6003 | ||
f33cea1a | 6004 | xfs_defer_init(&dfops, &firstfsb); |
19ebedcf DC |
6005 | |
6006 | error = xfs_bmap_split_extent_at(tp, ip, split_fsb, | |
f33cea1a | 6007 | &firstfsb, &dfops); |
19ebedcf DC |
6008 | if (error) |
6009 | goto out; | |
6010 | ||
5c33baee | 6011 | error = xfs_defer_finish(&tp, &dfops); |
19ebedcf DC |
6012 | if (error) |
6013 | goto out; | |
6014 | ||
de5a3f46 | 6015 | return xfs_trans_commit(tp); |
19ebedcf DC |
6016 | |
6017 | out: | |
f33cea1a | 6018 | xfs_defer_cancel(&dfops); |
3d7434fe | 6019 | xfs_trans_cancel(tp); |
19ebedcf DC |
6020 | return error; |
6021 | } | |
aeb88300 DW |
6022 | |
6023 | /* Deferred mapping is only for real extents in the data fork. */ | |
6024 | static bool | |
6025 | xfs_bmap_is_update_needed( | |
6026 | struct xfs_bmbt_irec *bmap) | |
6027 | { | |
6028 | return bmap->br_startblock != HOLESTARTBLOCK && | |
6029 | bmap->br_startblock != DELAYSTARTBLOCK; | |
6030 | } | |
6031 | ||
6032 | /* Record a bmap intent. */ | |
6033 | static int | |
6034 | __xfs_bmap_add( | |
6035 | struct xfs_mount *mp, | |
6036 | struct xfs_defer_ops *dfops, | |
6037 | enum xfs_bmap_intent_type type, | |
6038 | struct xfs_inode *ip, | |
6039 | int whichfork, | |
6040 | struct xfs_bmbt_irec *bmap) | |
6041 | { | |
6042 | int error; | |
6043 | struct xfs_bmap_intent *bi; | |
6044 | ||
6045 | trace_xfs_bmap_defer(mp, | |
6046 | XFS_FSB_TO_AGNO(mp, bmap->br_startblock), | |
6047 | type, | |
6048 | XFS_FSB_TO_AGBNO(mp, bmap->br_startblock), | |
6049 | ip->i_ino, whichfork, | |
6050 | bmap->br_startoff, | |
6051 | bmap->br_blockcount, | |
6052 | bmap->br_state); | |
6053 | ||
6054 | bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS); | |
6055 | INIT_LIST_HEAD(&bi->bi_list); | |
6056 | bi->bi_type = type; | |
6057 | bi->bi_owner = ip; | |
6058 | bi->bi_whichfork = whichfork; | |
6059 | bi->bi_bmap = *bmap; | |
6060 | ||
277d3c3a | 6061 | error = xfs_defer_ijoin(dfops, bi->bi_owner); |
aeb88300 DW |
6062 | if (error) { |
6063 | kmem_free(bi); | |
6064 | return error; | |
6065 | } | |
6066 | ||
6067 | xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list); | |
6068 | return 0; | |
6069 | } | |
6070 | ||
6071 | /* Map an extent into a file. */ | |
6072 | int | |
6073 | xfs_bmap_map_extent( | |
6074 | struct xfs_mount *mp, | |
6075 | struct xfs_defer_ops *dfops, | |
6076 | struct xfs_inode *ip, | |
6077 | struct xfs_bmbt_irec *PREV) | |
6078 | { | |
6079 | if (!xfs_bmap_is_update_needed(PREV)) | |
6080 | return 0; | |
6081 | ||
6082 | return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip, | |
6083 | XFS_DATA_FORK, PREV); | |
6084 | } | |
6085 | ||
6086 | /* Unmap an extent out of a file. */ | |
6087 | int | |
6088 | xfs_bmap_unmap_extent( | |
6089 | struct xfs_mount *mp, | |
6090 | struct xfs_defer_ops *dfops, | |
6091 | struct xfs_inode *ip, | |
6092 | struct xfs_bmbt_irec *PREV) | |
6093 | { | |
6094 | if (!xfs_bmap_is_update_needed(PREV)) | |
6095 | return 0; | |
6096 | ||
6097 | return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip, | |
6098 | XFS_DATA_FORK, PREV); | |
6099 | } | |
6100 | ||
6101 | /* | |
6102 | * Process one of the deferred bmap operations. We pass back the | |
6103 | * btree cursor to maintain our lock on the bmapbt between calls. | |
6104 | */ | |
6105 | int | |
6106 | xfs_bmap_finish_one( | |
6107 | struct xfs_trans *tp, | |
6108 | struct xfs_defer_ops *dfops, | |
6109 | struct xfs_inode *ip, | |
6110 | enum xfs_bmap_intent_type type, | |
6111 | int whichfork, | |
6112 | xfs_fileoff_t startoff, | |
6113 | xfs_fsblock_t startblock, | |
594956fa | 6114 | xfs_filblks_t *blockcount, |
aeb88300 DW |
6115 | xfs_exntst_t state) |
6116 | { | |
594956fa DW |
6117 | xfs_fsblock_t firstfsb; |
6118 | int error = 0; | |
aeb88300 | 6119 | |
66d19ae1 DW |
6120 | /* |
6121 | * firstfsb is tied to the transaction lifetime and is used to | |
6122 | * ensure correct AG locking order and schedule work item | |
6123 | * continuations. XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us | |
6124 | * to only making one bmap call per transaction, so it should | |
6125 | * be safe to have it as a local variable here. | |
6126 | */ | |
6127 | firstfsb = NULLFSBLOCK; | |
6128 | ||
aeb88300 DW |
6129 | trace_xfs_bmap_deferred(tp->t_mountp, |
6130 | XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type, | |
6131 | XFS_FSB_TO_AGBNO(tp->t_mountp, startblock), | |
594956fa | 6132 | ip->i_ino, whichfork, startoff, *blockcount, state); |
aeb88300 | 6133 | |
95855a23 | 6134 | if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK)) |
aeb88300 | 6135 | return -EFSCORRUPTED; |
aeb88300 DW |
6136 | |
6137 | if (XFS_TEST_ERROR(false, tp->t_mountp, | |
e2a190dd | 6138 | XFS_ERRTAG_BMAP_FINISH_ONE)) |
aeb88300 DW |
6139 | return -EIO; |
6140 | ||
6141 | switch (type) { | |
6142 | case XFS_BMAP_MAP: | |
594956fa | 6143 | error = xfs_bmapi_remap(tp, ip, startoff, *blockcount, |
0b44aa85 | 6144 | startblock, dfops); |
594956fa | 6145 | *blockcount = 0; |
aeb88300 DW |
6146 | break; |
6147 | case XFS_BMAP_UNMAP: | |
594956fa DW |
6148 | error = __xfs_bunmapi(tp, ip, startoff, blockcount, |
6149 | XFS_BMAPI_REMAP, 1, &firstfsb, dfops); | |
aeb88300 DW |
6150 | break; |
6151 | default: | |
6152 | ASSERT(0); | |
6153 | error = -EFSCORRUPTED; | |
6154 | } | |
6155 | ||
6156 | return error; | |
6157 | } | |
0cf6a3a9 DW |
6158 | |
6159 | /* Check that an inode's extent does not have invalid flags or bad ranges. */ | |
6160 | xfs_failaddr_t | |
6161 | xfs_bmap_validate_extent( | |
6162 | struct xfs_inode *ip, | |
6163 | int whichfork, | |
6164 | struct xfs_bmbt_irec *irec) | |
6165 | { | |
6166 | struct xfs_mount *mp = ip->i_mount; | |
6167 | xfs_fsblock_t endfsb; | |
6168 | bool isrt; | |
6169 | ||
6170 | isrt = XFS_IS_REALTIME_INODE(ip); | |
6171 | endfsb = irec->br_startblock + irec->br_blockcount - 1; | |
6172 | if (isrt) { | |
6173 | if (!xfs_verify_rtbno(mp, irec->br_startblock)) | |
6174 | return __this_address; | |
6175 | if (!xfs_verify_rtbno(mp, endfsb)) | |
6176 | return __this_address; | |
6177 | } else { | |
6178 | if (!xfs_verify_fsbno(mp, irec->br_startblock)) | |
6179 | return __this_address; | |
6180 | if (!xfs_verify_fsbno(mp, endfsb)) | |
6181 | return __this_address; | |
6182 | if (XFS_FSB_TO_AGNO(mp, irec->br_startblock) != | |
6183 | XFS_FSB_TO_AGNO(mp, endfsb)) | |
6184 | return __this_address; | |
6185 | } | |
6186 | if (irec->br_state != XFS_EXT_NORM) { | |
6187 | if (whichfork != XFS_DATA_FORK) | |
6188 | return __this_address; | |
6189 | if (!xfs_sb_version_hasextflgbit(&mp->m_sb)) | |
6190 | return __this_address; | |
6191 | } | |
6192 | return NULL; | |
6193 | } |