]>
Commit | Line | Data |
---|---|---|
37b3b4d6 | 1 | // SPDX-License-Identifier: GPL-2.0 |
2bd0ea18 | 2 | /* |
5e656dbb | 3 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
da23017d | 4 | * All Rights Reserved. |
2bd0ea18 | 5 | */ |
9c799827 | 6 | #include "libxfs_priv.h" |
b626fb59 DC |
7 | #include "xfs_fs.h" |
8 | #include "xfs_shared.h" | |
9 | #include "xfs_format.h" | |
10 | #include "xfs_log_format.h" | |
11 | #include "xfs_trans_resv.h" | |
12 | #include "xfs_bit.h" | |
13 | #include "xfs_sb.h" | |
14 | #include "xfs_mount.h" | |
f944d3d0 | 15 | #include "xfs_defer.h" |
b626fb59 DC |
16 | #include "xfs_da_format.h" |
17 | #include "xfs_da_btree.h" | |
18 | #include "xfs_dir2.h" | |
19 | #include "xfs_inode.h" | |
20 | #include "xfs_btree.h" | |
21 | #include "xfs_trans.h" | |
22 | #include "xfs_alloc.h" | |
23 | #include "xfs_bmap.h" | |
24 | #include "xfs_bmap_btree.h" | |
56d3fc2b | 25 | #include "xfs_errortag.h" |
b626fb59 DC |
26 | #include "xfs_trans_space.h" |
27 | #include "xfs_trace.h" | |
28 | #include "xfs_attr_leaf.h" | |
29 | #include "xfs_quota_defs.h" | |
85aec44f | 30 | #include "xfs_rmap.h" |
cf8ce220 | 31 | #include "xfs_ag_resv.h" |
cfe32f0d | 32 | #include "xfs_refcount.h" |
b626fb59 | 33 | |
2bd0ea18 | 34 | |
5e656dbb BN |
35 | kmem_zone_t *xfs_bmap_free_item_zone; |
36 | ||
37 | /* | |
49f693fa | 38 | * Miscellaneous helper functions |
5e656dbb | 39 | */ |
5e656dbb | 40 | |
5e656dbb | 41 | /* |
49f693fa DC |
42 | * Compute and fill in the value of the maximum depth of a bmap btree |
43 | * in this filesystem. Done once, during mount. | |
5e656dbb | 44 | */ |
49f693fa DC |
45 | void |
46 | xfs_bmap_compute_maxlevels( | |
47 | xfs_mount_t *mp, /* file system mount structure */ | |
48 | int whichfork) /* data or attr fork */ | |
49 | { | |
50 | int level; /* btree level */ | |
51 | uint maxblocks; /* max blocks at this level */ | |
52 | uint maxleafents; /* max leaf entries possible */ | |
53 | int maxrootrecs; /* max records in root block */ | |
54 | int minleafrecs; /* min records in leaf block */ | |
55 | int minnoderecs; /* min records in node block */ | |
56 | int sz; /* root block size */ | |
5e656dbb | 57 | |
49f693fa DC |
58 | /* |
59 | * The maximum number of extents in a file, hence the maximum | |
60 | * number of leaf entries, is controlled by the type of di_nextents | |
61 | * (a signed 32-bit number, xfs_extnum_t), or by di_anextents | |
62 | * (a signed 16-bit number, xfs_aextnum_t). | |
63 | * | |
64 | * Note that we can no longer assume that if we are in ATTR1 that | |
65 | * the fork offset of all the inodes will be | |
66 | * (xfs_default_attroffset(ip) >> 3) because we could have mounted | |
67 | * with ATTR2 and then mounted back with ATTR1, keeping the | |
68 | * di_forkoff's fixed but probably at various positions. Therefore, | |
69 | * for both ATTR1 and ATTR2 we have to assume the worst case scenario | |
70 | * of a minimum size available. | |
71 | */ | |
72 | if (whichfork == XFS_DATA_FORK) { | |
73 | maxleafents = MAXEXTNUM; | |
74 | sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS); | |
75 | } else { | |
76 | maxleafents = MAXAEXTNUM; | |
77 | sz = XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
78 | } | |
ff105f75 | 79 | maxrootrecs = xfs_bmdr_maxrecs(sz, 0); |
49f693fa DC |
80 | minleafrecs = mp->m_bmap_dmnr[0]; |
81 | minnoderecs = mp->m_bmap_dmnr[1]; | |
82 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
83 | for (level = 1; maxblocks > 1; level++) { | |
84 | if (maxblocks <= maxrootrecs) | |
85 | maxblocks = 1; | |
86 | else | |
87 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
88 | } | |
89 | mp->m_bm_maxlevels[whichfork] = level; | |
90 | } | |
5e656dbb | 91 | |
b194c7d8 BN |
92 | STATIC int /* error */ |
93 | xfs_bmbt_lookup_eq( | |
94 | struct xfs_btree_cur *cur, | |
70a93110 | 95 | struct xfs_bmbt_irec *irec, |
b194c7d8 BN |
96 | int *stat) /* success/failure */ |
97 | { | |
70a93110 | 98 | cur->bc_rec.b = *irec; |
b194c7d8 BN |
99 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); |
100 | } | |
101 | ||
102 | STATIC int /* error */ | |
4f76f49c | 103 | xfs_bmbt_lookup_first( |
b194c7d8 | 104 | struct xfs_btree_cur *cur, |
b194c7d8 BN |
105 | int *stat) /* success/failure */ |
106 | { | |
4f76f49c CH |
107 | cur->bc_rec.b.br_startoff = 0; |
108 | cur->bc_rec.b.br_startblock = 0; | |
109 | cur->bc_rec.b.br_blockcount = 0; | |
b194c7d8 BN |
110 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); |
111 | } | |
112 | ||
113 | /* | |
a2ceac1f DC |
114 | * Check if the inode needs to be converted to btree format. |
115 | */ | |
116 | static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork) | |
117 | { | |
1277a5e0 DW |
118 | return whichfork != XFS_COW_FORK && |
119 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
a2ceac1f DC |
120 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
121 | XFS_IFORK_MAXEXT(ip, whichfork); | |
122 | } | |
123 | ||
124 | /* | |
125 | * Check if the inode should be converted to extent format. | |
126 | */ | |
127 | static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork) | |
128 | { | |
1277a5e0 DW |
129 | return whichfork != XFS_COW_FORK && |
130 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE && | |
a2ceac1f DC |
131 | XFS_IFORK_NEXTENTS(ip, whichfork) <= |
132 | XFS_IFORK_MAXEXT(ip, whichfork); | |
133 | } | |
134 | ||
135 | /* | |
d0e5f1ff | 136 | * Update the record referred to by cur to the value given by irec |
b194c7d8 BN |
137 | * This either works (return 0) or gets an EFSCORRUPTED error. |
138 | */ | |
139 | STATIC int | |
140 | xfs_bmbt_update( | |
141 | struct xfs_btree_cur *cur, | |
d0e5f1ff | 142 | struct xfs_bmbt_irec *irec) |
b194c7d8 BN |
143 | { |
144 | union xfs_btree_rec rec; | |
145 | ||
d0e5f1ff | 146 | xfs_bmbt_disk_set_all(&rec.bmbt, irec); |
b194c7d8 BN |
147 | return xfs_btree_update(cur, &rec); |
148 | } | |
149 | ||
5e656dbb | 150 | /* |
49f693fa DC |
151 | * Compute the worst-case number of indirect blocks that will be used |
152 | * for ip's delayed extent of length "len". | |
5e656dbb | 153 | */ |
49f693fa DC |
154 | STATIC xfs_filblks_t |
155 | xfs_bmap_worst_indlen( | |
156 | xfs_inode_t *ip, /* incore inode pointer */ | |
157 | xfs_filblks_t len) /* delayed extent length */ | |
57c9fccb | 158 | { |
49f693fa DC |
159 | int level; /* btree level number */ |
160 | int maxrecs; /* maximum record count at this level */ | |
161 | xfs_mount_t *mp; /* mount structure */ | |
162 | xfs_filblks_t rval; /* return value */ | |
57c9fccb NS |
163 | |
164 | mp = ip->i_mount; | |
49f693fa DC |
165 | maxrecs = mp->m_bmap_dmxr[0]; |
166 | for (level = 0, rval = 0; | |
167 | level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK); | |
168 | level++) { | |
169 | len += maxrecs - 1; | |
170 | do_div(len, maxrecs); | |
171 | rval += len; | |
7fbe9b54 DW |
172 | if (len == 1) |
173 | return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) - | |
49f693fa DC |
174 | level - 1; |
175 | if (level == 0) | |
176 | maxrecs = mp->m_bmap_dmxr[1]; | |
57c9fccb | 177 | } |
49f693fa | 178 | return rval; |
57c9fccb NS |
179 | } |
180 | ||
181 | /* | |
49f693fa | 182 | * Calculate the default attribute fork offset for newly created inodes. |
57c9fccb | 183 | */ |
49f693fa DC |
184 | uint |
185 | xfs_default_attroffset( | |
186 | struct xfs_inode *ip) | |
57c9fccb | 187 | { |
49f693fa DC |
188 | struct xfs_mount *mp = ip->i_mount; |
189 | uint offset; | |
57c9fccb | 190 | |
49f693fa DC |
191 | if (mp->m_sb.sb_inodesize == 256) { |
192 | offset = XFS_LITINO(mp, ip->i_d.di_version) - | |
193 | XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
194 | } else { | |
195 | offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); | |
57c9fccb | 196 | } |
49f693fa DC |
197 | |
198 | ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version)); | |
199 | return offset; | |
57c9fccb NS |
200 | } |
201 | ||
202 | /* | |
49f693fa DC |
203 | * Helper routine to reset inode di_forkoff field when switching |
204 | * attribute fork from local to extent format - we reset it where | |
205 | * possible to make space available for inline data fork extents. | |
57c9fccb | 206 | */ |
49f693fa DC |
207 | STATIC void |
208 | xfs_bmap_forkoff_reset( | |
49f693fa DC |
209 | xfs_inode_t *ip, |
210 | int whichfork) | |
57c9fccb | 211 | { |
49f693fa DC |
212 | if (whichfork == XFS_ATTR_FORK && |
213 | ip->i_d.di_format != XFS_DINODE_FMT_DEV && | |
49f693fa DC |
214 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE) { |
215 | uint dfl_forkoff = xfs_default_attroffset(ip) >> 3; | |
57c9fccb | 216 | |
49f693fa DC |
217 | if (dfl_forkoff > ip->i_d.di_forkoff) |
218 | ip->i_d.di_forkoff = dfl_forkoff; | |
219 | } | |
57c9fccb NS |
220 | } |
221 | ||
49f693fa DC |
222 | #ifdef DEBUG |
223 | STATIC struct xfs_buf * | |
224 | xfs_bmap_get_bp( | |
225 | struct xfs_btree_cur *cur, | |
226 | xfs_fsblock_t bno) | |
227 | { | |
2fdd378a | 228 | struct xfs_log_item *lip; |
49f693fa | 229 | int i; |
56b2de80 | 230 | |
49f693fa DC |
231 | if (!cur) |
232 | return NULL; | |
2bd0ea18 | 233 | |
49f693fa DC |
234 | for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) { |
235 | if (!cur->bc_bufs[i]) | |
236 | break; | |
237 | if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno) | |
238 | return cur->bc_bufs[i]; | |
239 | } | |
56b2de80 | 240 | |
49f693fa | 241 | /* Chase down all the log items to see if the bp is there */ |
2fdd378a DC |
242 | list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) { |
243 | struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip; | |
244 | ||
49f693fa DC |
245 | if (bip->bli_item.li_type == XFS_LI_BUF && |
246 | XFS_BUF_ADDR(bip->bli_buf) == bno) | |
247 | return bip->bli_buf; | |
248 | } | |
2bd0ea18 | 249 | |
49f693fa DC |
250 | return NULL; |
251 | } | |
56b2de80 | 252 | |
49f693fa DC |
253 | STATIC void |
254 | xfs_check_block( | |
255 | struct xfs_btree_block *block, | |
256 | xfs_mount_t *mp, | |
257 | int root, | |
258 | short sz) | |
259 | { | |
260 | int i, j, dmxr; | |
261 | __be64 *pp, *thispa; /* pointer to block address */ | |
262 | xfs_bmbt_key_t *prevp, *keyp; | |
2bd0ea18 | 263 | |
49f693fa | 264 | ASSERT(be16_to_cpu(block->bb_level) > 0); |
56b2de80 | 265 | |
49f693fa DC |
266 | prevp = NULL; |
267 | for( i = 1; i <= xfs_btree_get_numrecs(block); i++) { | |
268 | dmxr = mp->m_bmap_dmxr[0]; | |
269 | keyp = XFS_BMBT_KEY_ADDR(mp, block, i); | |
a2ceac1f | 270 | |
49f693fa DC |
271 | if (prevp) { |
272 | ASSERT(be64_to_cpu(prevp->br_startoff) < | |
273 | be64_to_cpu(keyp->br_startoff)); | |
274 | } | |
275 | prevp = keyp; | |
2bd0ea18 | 276 | |
2bd0ea18 | 277 | /* |
49f693fa | 278 | * Compare the block numbers to see if there are dups. |
2bd0ea18 | 279 | */ |
49f693fa DC |
280 | if (root) |
281 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz); | |
282 | else | |
283 | pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr); | |
284 | ||
285 | for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) { | |
286 | if (root) | |
287 | thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz); | |
288 | else | |
289 | thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr); | |
290 | if (*thispa == *pp) { | |
291 | xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld", | |
292 | __func__, j, i, | |
293 | (unsigned long long)be64_to_cpu(*thispa)); | |
4d5e2888 | 294 | xfs_err(mp, "%s: ptrs are equal in node\n", |
49f693fa | 295 | __func__); |
4d5e2888 | 296 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
49f693fa | 297 | } |
2bd0ea18 | 298 | } |
49f693fa DC |
299 | } |
300 | } | |
a2ceac1f | 301 | |
49f693fa DC |
302 | /* |
303 | * Check that the extents for the inode ip are in the right order in all | |
f07ae2a6 DC |
304 | * btree leaves. THis becomes prohibitively expensive for large extent count |
305 | * files, so don't bother with inodes that have more than 10,000 extents in | |
306 | * them. The btree record ordering checks will still be done, so for such large | |
307 | * bmapbt constructs that is going to catch most corruptions. | |
49f693fa | 308 | */ |
49f693fa DC |
309 | STATIC void |
310 | xfs_bmap_check_leaf_extents( | |
311 | xfs_btree_cur_t *cur, /* btree cursor or null */ | |
312 | xfs_inode_t *ip, /* incore inode pointer */ | |
313 | int whichfork) /* data or attr fork */ | |
314 | { | |
315 | struct xfs_btree_block *block; /* current btree block */ | |
316 | xfs_fsblock_t bno; /* block # of "block" */ | |
317 | xfs_buf_t *bp; /* buffer for "block" */ | |
318 | int error; /* error return value */ | |
319 | xfs_extnum_t i=0, j; /* index into the extents list */ | |
e07055b8 | 320 | struct xfs_ifork *ifp; /* fork structure */ |
49f693fa DC |
321 | int level; /* btree level, for checking */ |
322 | xfs_mount_t *mp; /* file system mount structure */ | |
323 | __be64 *pp; /* pointer to block address */ | |
324 | xfs_bmbt_rec_t *ep; /* pointer to current extent */ | |
325 | xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */ | |
326 | xfs_bmbt_rec_t *nextp; /* pointer to next extent */ | |
327 | int bp_release = 0; | |
328 | ||
329 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) { | |
330 | return; | |
331 | } | |
332 | ||
f07ae2a6 DC |
333 | /* skip large extent count inodes */ |
334 | if (ip->i_d.di_nextents > 10000) | |
335 | return; | |
336 | ||
49f693fa DC |
337 | bno = NULLFSBLOCK; |
338 | mp = ip->i_mount; | |
339 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
340 | block = ifp->if_broot; | |
341 | /* | |
342 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
343 | */ | |
344 | level = be16_to_cpu(block->bb_level); | |
345 | ASSERT(level > 0); | |
346 | xfs_check_block(block, mp, 1, ifp->if_broot_bytes); | |
347 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
348 | bno = be64_to_cpu(*pp); | |
349 | ||
5a35bf2c | 350 | ASSERT(bno != NULLFSBLOCK); |
49f693fa DC |
351 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
352 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
353 | ||
354 | /* | |
355 | * Go down the tree until leaf level is reached, following the first | |
356 | * pointer (leftmost) at each level. | |
357 | */ | |
358 | while (level-- > 0) { | |
359 | /* See if buf is in cur first */ | |
360 | bp_release = 0; | |
361 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
362 | if (!bp) { | |
363 | bp_release = 1; | |
364 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
365 | XFS_BMAP_BTREE_REF, | |
366 | &xfs_bmbt_buf_ops); | |
2bd0ea18 | 367 | if (error) |
49f693fa | 368 | goto error_norelse; |
2bd0ea18 | 369 | } |
49f693fa | 370 | block = XFS_BUF_TO_BLOCK(bp); |
49f693fa DC |
371 | if (level == 0) |
372 | break; | |
2bd0ea18 | 373 | |
2bd0ea18 | 374 | /* |
49f693fa DC |
375 | * Check this block for basic sanity (increasing keys and |
376 | * no duplicate blocks). | |
2bd0ea18 | 377 | */ |
49f693fa DC |
378 | |
379 | xfs_check_block(block, mp, 0, 0); | |
380 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
381 | bno = be64_to_cpu(*pp); | |
19ebedcf | 382 | XFS_WANT_CORRUPTED_GOTO(mp, |
ecdc52ff | 383 | xfs_verify_fsbno(mp, bno), error0); |
49f693fa DC |
384 | if (bp_release) { |
385 | bp_release = 0; | |
386 | xfs_trans_brelse(NULL, bp); | |
2bd0ea18 | 387 | } |
49f693fa | 388 | } |
a2ceac1f | 389 | |
49f693fa DC |
390 | /* |
391 | * Here with bp and block set to the leftmost leaf node in the tree. | |
392 | */ | |
393 | i = 0; | |
a2ceac1f | 394 | |
49f693fa DC |
395 | /* |
396 | * Loop over all leaf nodes checking that all extents are in the right order. | |
397 | */ | |
398 | for (;;) { | |
399 | xfs_fsblock_t nextbno; | |
400 | xfs_extnum_t num_recs; | |
401 | ||
402 | ||
403 | num_recs = xfs_btree_get_numrecs(block); | |
2bd0ea18 | 404 | |
2bd0ea18 | 405 | /* |
49f693fa | 406 | * Read-ahead the next leaf block, if any. |
2bd0ea18 | 407 | */ |
a2ceac1f | 408 | |
49f693fa | 409 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
a2ceac1f | 410 | |
49f693fa DC |
411 | /* |
412 | * Check all the extents to make sure they are OK. | |
413 | * If we had a previous block, the last entry should | |
414 | * conform with the first entry in this one. | |
415 | */ | |
2bd0ea18 | 416 | |
49f693fa DC |
417 | ep = XFS_BMBT_REC_ADDR(mp, block, 1); |
418 | if (i) { | |
419 | ASSERT(xfs_bmbt_disk_get_startoff(&last) + | |
420 | xfs_bmbt_disk_get_blockcount(&last) <= | |
421 | xfs_bmbt_disk_get_startoff(ep)); | |
422 | } | |
423 | for (j = 1; j < num_recs; j++) { | |
424 | nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1); | |
425 | ASSERT(xfs_bmbt_disk_get_startoff(ep) + | |
426 | xfs_bmbt_disk_get_blockcount(ep) <= | |
427 | xfs_bmbt_disk_get_startoff(nextp)); | |
428 | ep = nextp; | |
429 | } | |
430 | ||
431 | last = *ep; | |
432 | i += num_recs; | |
433 | if (bp_release) { | |
434 | bp_release = 0; | |
435 | xfs_trans_brelse(NULL, bp); | |
436 | } | |
437 | bno = nextbno; | |
2bd0ea18 | 438 | /* |
49f693fa | 439 | * If we've reached the end, stop. |
2bd0ea18 | 440 | */ |
49f693fa DC |
441 | if (bno == NULLFSBLOCK) |
442 | break; | |
a2ceac1f | 443 | |
49f693fa DC |
444 | bp_release = 0; |
445 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
446 | if (!bp) { | |
447 | bp_release = 1; | |
448 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
449 | XFS_BMAP_BTREE_REF, | |
450 | &xfs_bmbt_buf_ops); | |
a2ceac1f | 451 | if (error) |
49f693fa | 452 | goto error_norelse; |
2bd0ea18 | 453 | } |
49f693fa | 454 | block = XFS_BUF_TO_BLOCK(bp); |
a2ceac1f | 455 | } |
4d4a192c | 456 | |
49f693fa | 457 | return; |
a2ceac1f | 458 | |
49f693fa DC |
459 | error0: |
460 | xfs_warn(mp, "%s: at error0", __func__); | |
461 | if (bp_release) | |
462 | xfs_trans_brelse(NULL, bp); | |
463 | error_norelse: | |
464 | xfs_warn(mp, "%s: BAD after btree leaves for %d extents", | |
465 | __func__, i); | |
4d5e2888 DW |
466 | xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__); |
467 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
49f693fa | 468 | return; |
2bd0ea18 NS |
469 | } |
470 | ||
49f693fa DC |
471 | /* |
472 | * Validate that the bmbt_irecs being returned from bmapi are valid | |
e6d77a21 DC |
473 | * given the caller's original parameters. Specifically check the |
474 | * ranges of the returned irecs to ensure that they only extend beyond | |
49f693fa DC |
475 | * the given parameters if the XFS_BMAPI_ENTIRE flag was set. |
476 | */ | |
477 | STATIC void | |
478 | xfs_bmap_validate_ret( | |
479 | xfs_fileoff_t bno, | |
480 | xfs_filblks_t len, | |
481 | int flags, | |
482 | xfs_bmbt_irec_t *mval, | |
483 | int nmap, | |
484 | int ret_nmap) | |
485 | { | |
486 | int i; /* index to map values */ | |
a2ceac1f | 487 | |
49f693fa | 488 | ASSERT(ret_nmap <= nmap); |
a2ceac1f | 489 | |
49f693fa DC |
490 | for (i = 0; i < ret_nmap; i++) { |
491 | ASSERT(mval[i].br_blockcount > 0); | |
492 | if (!(flags & XFS_BMAPI_ENTIRE)) { | |
493 | ASSERT(mval[i].br_startoff >= bno); | |
494 | ASSERT(mval[i].br_blockcount <= len); | |
495 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount <= | |
496 | bno + len); | |
497 | } else { | |
498 | ASSERT(mval[i].br_startoff < bno + len); | |
499 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount > | |
500 | bno); | |
501 | } | |
502 | ASSERT(i == 0 || | |
503 | mval[i - 1].br_startoff + mval[i - 1].br_blockcount == | |
504 | mval[i].br_startoff); | |
505 | ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK && | |
506 | mval[i].br_startblock != HOLESTARTBLOCK); | |
507 | ASSERT(mval[i].br_state == XFS_EXT_NORM || | |
508 | mval[i].br_state == XFS_EXT_UNWRITTEN); | |
509 | } | |
510 | } | |
56b2de80 | 511 | |
49f693fa DC |
512 | #else |
513 | #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0) | |
9587a34c | 514 | #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0) |
49f693fa | 515 | #endif /* DEBUG */ |
56b2de80 | 516 | |
49f693fa DC |
517 | /* |
518 | * bmap free list manipulation functions | |
519 | */ | |
56b2de80 | 520 | |
49f693fa DC |
521 | /* |
522 | * Add the extent to the list of extents to be free at transaction end. | |
523 | * The list is maintained sorted (by block number). | |
524 | */ | |
525 | void | |
3a13f959 | 526 | __xfs_bmap_add_free( |
21375e5d | 527 | struct xfs_trans *tp, |
85aec44f DW |
528 | xfs_fsblock_t bno, |
529 | xfs_filblks_t len, | |
5837e73b | 530 | const struct xfs_owner_info *oinfo, |
3a13f959 | 531 | bool skip_discard) |
49f693fa | 532 | { |
6f530e9a | 533 | struct xfs_extent_free_item *new; /* new element */ |
49f693fa | 534 | #ifdef DEBUG |
21375e5d BF |
535 | struct xfs_mount *mp = tp->t_mountp; |
536 | xfs_agnumber_t agno; | |
537 | xfs_agblock_t agbno; | |
56b2de80 | 538 | |
49f693fa DC |
539 | ASSERT(bno != NULLFSBLOCK); |
540 | ASSERT(len > 0); | |
541 | ASSERT(len <= MAXEXTLEN); | |
542 | ASSERT(!isnullstartblock(bno)); | |
543 | agno = XFS_FSB_TO_AGNO(mp, bno); | |
544 | agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
545 | ASSERT(agno < mp->m_sb.sb_agcount); | |
546 | ASSERT(agbno < mp->m_sb.sb_agblocks); | |
547 | ASSERT(len < mp->m_sb.sb_agblocks); | |
548 | ASSERT(agbno + len <= mp->m_sb.sb_agblocks); | |
549 | #endif | |
550 | ASSERT(xfs_bmap_free_item_zone != NULL); | |
85aec44f | 551 | |
49f693fa | 552 | new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP); |
6f530e9a DW |
553 | new->xefi_startblock = bno; |
554 | new->xefi_blockcount = (xfs_extlen_t)len; | |
85aec44f DW |
555 | if (oinfo) |
556 | new->xefi_oinfo = *oinfo; | |
557 | else | |
007347e3 | 558 | new->xefi_oinfo = XFS_RMAP_OINFO_SKIP_UPDATE; |
3a13f959 | 559 | new->xefi_skip_discard = skip_discard; |
21375e5d BF |
560 | trace_xfs_bmap_free_defer(tp->t_mountp, |
561 | XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0, | |
562 | XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len); | |
563 | xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list); | |
49f693fa | 564 | } |
2bd0ea18 | 565 | |
49f693fa DC |
566 | /* |
567 | * Inode fork format manipulation functions | |
568 | */ | |
a2ceac1f | 569 | |
49f693fa DC |
570 | /* |
571 | * Transform a btree format file with only one leaf node, where the | |
572 | * extents list will fit in the inode, into an extents format file. | |
573 | * Since the file extents are already in-core, all we have to do is | |
574 | * give up the space for the btree root and pitch the leaf block. | |
575 | */ | |
576 | STATIC int /* error */ | |
577 | xfs_bmap_btree_to_extents( | |
578 | xfs_trans_t *tp, /* transaction pointer */ | |
579 | xfs_inode_t *ip, /* incore inode pointer */ | |
580 | xfs_btree_cur_t *cur, /* btree cursor */ | |
581 | int *logflagsp, /* inode logging flags */ | |
582 | int whichfork) /* data or attr fork */ | |
583 | { | |
584 | /* REFERENCED */ | |
585 | struct xfs_btree_block *cblock;/* child btree block */ | |
586 | xfs_fsblock_t cbno; /* child block number */ | |
587 | xfs_buf_t *cbp; /* child block's buffer */ | |
588 | int error; /* error return value */ | |
e07055b8 | 589 | struct xfs_ifork *ifp; /* inode fork data */ |
49f693fa DC |
590 | xfs_mount_t *mp; /* mount point structure */ |
591 | __be64 *pp; /* ptr to block address */ | |
592 | struct xfs_btree_block *rblock;/* root btree block */ | |
85aec44f | 593 | struct xfs_owner_info oinfo; |
56b2de80 | 594 | |
49f693fa DC |
595 | mp = ip->i_mount; |
596 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1277a5e0 | 597 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa DC |
598 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
599 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
600 | rblock = ifp->if_broot; | |
601 | ASSERT(be16_to_cpu(rblock->bb_level) == 1); | |
602 | ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1); | |
603 | ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1); | |
604 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes); | |
605 | cbno = be64_to_cpu(*pp); | |
606 | *logflagsp = 0; | |
607 | #ifdef DEBUG | |
d5c546df DW |
608 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, |
609 | xfs_btree_check_lptr(cur, cbno, 1)); | |
49f693fa DC |
610 | #endif |
611 | error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF, | |
612 | &xfs_bmbt_buf_ops); | |
613 | if (error) | |
614 | return error; | |
615 | cblock = XFS_BUF_TO_BLOCK(cbp); | |
616 | if ((error = xfs_btree_check_block(cur, cblock, 0, cbp))) | |
617 | return error; | |
85aec44f | 618 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); |
21375e5d | 619 | xfs_bmap_add_free(cur->bc_tp, cbno, 1, &oinfo); |
49f693fa DC |
620 | ip->i_d.di_nblocks--; |
621 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); | |
622 | xfs_trans_binval(tp, cbp); | |
623 | if (cur->bc_bufs[0] == cbp) | |
624 | cur->bc_bufs[0] = NULL; | |
625 | xfs_iroot_realloc(ip, -1, whichfork); | |
626 | ASSERT(ifp->if_broot == NULL); | |
627 | ASSERT((ifp->if_flags & XFS_IFBROOT) == 0); | |
628 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
629 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); | |
630 | return 0; | |
631 | } | |
2bd0ea18 NS |
632 | |
633 | /* | |
49f693fa DC |
634 | * Convert an extents-format file into a btree-format file. |
635 | * The new file will have a root block (in the inode) and a single child block. | |
2bd0ea18 | 636 | */ |
49f693fa DC |
637 | STATIC int /* error */ |
638 | xfs_bmap_extents_to_btree( | |
52f6ed9d BF |
639 | struct xfs_trans *tp, /* transaction pointer */ |
640 | struct xfs_inode *ip, /* incore inode pointer */ | |
52f6ed9d | 641 | struct xfs_btree_cur **curp, /* cursor returned to caller */ |
49f693fa DC |
642 | int wasdel, /* converting a delayed alloc */ |
643 | int *logflagsp, /* inode logging flags */ | |
644 | int whichfork) /* data or attr fork */ | |
2bd0ea18 | 645 | { |
49f693fa | 646 | struct xfs_btree_block *ablock; /* allocated (child) bt block */ |
52f6ed9d BF |
647 | struct xfs_buf *abp; /* buffer for ablock */ |
648 | struct xfs_alloc_arg args; /* allocation arguments */ | |
649 | struct xfs_bmbt_rec *arp; /* child record pointer */ | |
49f693fa | 650 | struct xfs_btree_block *block; /* btree root block */ |
52f6ed9d | 651 | struct xfs_btree_cur *cur; /* bmap btree cursor */ |
49f693fa | 652 | int error; /* error return value */ |
52f6ed9d BF |
653 | struct xfs_ifork *ifp; /* inode fork pointer */ |
654 | struct xfs_bmbt_key *kp; /* root block key pointer */ | |
655 | struct xfs_mount *mp; /* mount structure */ | |
49f693fa | 656 | xfs_bmbt_ptr_t *pp; /* root block address pointer */ |
9788e059 | 657 | struct xfs_iext_cursor icur; |
62ed7338 | 658 | struct xfs_bmbt_irec rec; |
9788e059 | 659 | xfs_extnum_t cnt = 0; |
2bd0ea18 | 660 | |
5dfa5cd2 | 661 | mp = ip->i_mount; |
1277a5e0 | 662 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa DC |
663 | ifp = XFS_IFORK_PTR(ip, whichfork); |
664 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS); | |
56b2de80 | 665 | |
2bd0ea18 | 666 | /* |
d7c2f6cb DC |
667 | * Make space in the inode incore. This needs to be undone if we fail |
668 | * to expand the root. | |
2bd0ea18 | 669 | */ |
49f693fa DC |
670 | xfs_iroot_realloc(ip, 1, whichfork); |
671 | ifp->if_flags |= XFS_IFBROOT; | |
56b2de80 | 672 | |
2bd0ea18 | 673 | /* |
49f693fa | 674 | * Fill in the root. |
2bd0ea18 | 675 | */ |
49f693fa | 676 | block = ifp->if_broot; |
e394a4b1 ES |
677 | xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL, |
678 | XFS_BTNUM_BMAP, 1, 1, ip->i_ino, | |
f4241a08 | 679 | XFS_BTREE_LONG_PTRS); |
49f693fa DC |
680 | /* |
681 | * Need a cursor. Can't allocate until bb_level is filled in. | |
682 | */ | |
49f693fa | 683 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
49f693fa DC |
684 | cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; |
685 | /* | |
686 | * Convert to a btree with two levels, one record in root. | |
687 | */ | |
688 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE); | |
689 | memset(&args, 0, sizeof(args)); | |
690 | args.tp = tp; | |
691 | args.mp = mp; | |
85aec44f | 692 | xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork); |
f7253505 | 693 | if (tp->t_firstblock == NULLFSBLOCK) { |
49f693fa DC |
694 | args.type = XFS_ALLOCTYPE_START_BNO; |
695 | args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino); | |
565e96c6 | 696 | } else if (tp->t_flags & XFS_TRANS_LOWMODE) { |
49f693fa | 697 | args.type = XFS_ALLOCTYPE_START_BNO; |
f7253505 | 698 | args.fsbno = tp->t_firstblock; |
49f693fa DC |
699 | } else { |
700 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
f7253505 | 701 | args.fsbno = tp->t_firstblock; |
49f693fa DC |
702 | } |
703 | args.minlen = args.maxlen = args.prod = 1; | |
704 | args.wasdel = wasdel; | |
705 | *logflagsp = 0; | |
d7c2f6cb DC |
706 | error = xfs_alloc_vextent(&args); |
707 | if (error) | |
708 | goto out_root_realloc; | |
68d06dce | 709 | |
c1587ecf | 710 | if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) { |
b5c71bcd | 711 | error = -ENOSPC; |
d7c2f6cb | 712 | goto out_root_realloc; |
c1587ecf | 713 | } |
d7c2f6cb | 714 | |
2bd0ea18 | 715 | /* |
49f693fa | 716 | * Allocation can't fail, the space was reserved. |
2bd0ea18 | 717 | */ |
f7253505 BF |
718 | ASSERT(tp->t_firstblock == NULLFSBLOCK || |
719 | args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock)); | |
f44a348b | 720 | tp->t_firstblock = args.fsbno; |
49f693fa DC |
721 | cur->bc_private.b.allocated++; |
722 | ip->i_d.di_nblocks++; | |
723 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L); | |
724 | abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0); | |
b5c71bcd | 725 | if (!abp) { |
d7c2f6cb DC |
726 | error = -EFSCORRUPTED; |
727 | goto out_unreserve_dquot; | |
b5c71bcd | 728 | } |
d7c2f6cb | 729 | |
2bd0ea18 | 730 | /* |
49f693fa | 731 | * Fill in the child block. |
2bd0ea18 | 732 | */ |
49f693fa DC |
733 | abp->b_ops = &xfs_bmbt_buf_ops; |
734 | ablock = XFS_BUF_TO_BLOCK(abp); | |
e394a4b1 ES |
735 | xfs_btree_init_block_int(mp, ablock, abp->b_bn, |
736 | XFS_BTNUM_BMAP, 0, 0, ip->i_ino, | |
5dfa5cd2 DC |
737 | XFS_BTREE_LONG_PTRS); |
738 | ||
9788e059 | 739 | for_each_xfs_iext(ifp, &icur, &rec) { |
62ed7338 CH |
740 | if (isnullstartblock(rec.br_startblock)) |
741 | continue; | |
742 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt); | |
743 | xfs_bmbt_disk_set_all(arp, &rec); | |
744 | cnt++; | |
49f693fa DC |
745 | } |
746 | ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork)); | |
747 | xfs_btree_set_numrecs(ablock, cnt); | |
56b2de80 | 748 | |
49f693fa DC |
749 | /* |
750 | * Fill in the root key and pointer. | |
751 | */ | |
752 | kp = XFS_BMBT_KEY_ADDR(mp, block, 1); | |
753 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); | |
754 | kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp)); | |
755 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur, | |
756 | be16_to_cpu(block->bb_level))); | |
757 | *pp = cpu_to_be64(args.fsbno); | |
2bd0ea18 | 758 | |
49f693fa DC |
759 | /* |
760 | * Do all this logging at the end so that | |
761 | * the root is at the right level. | |
762 | */ | |
5dfa5cd2 | 763 | xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS); |
613e6057 | 764 | xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs)); |
49f693fa DC |
765 | ASSERT(*curp == NULL); |
766 | *curp = cur; | |
767 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork); | |
768 | return 0; | |
b5c71bcd | 769 | |
d7c2f6cb | 770 | out_unreserve_dquot: |
b5c71bcd | 771 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); |
d7c2f6cb | 772 | out_root_realloc: |
b5c71bcd SH |
773 | xfs_iroot_realloc(ip, -1, whichfork); |
774 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
d7c2f6cb | 775 | ASSERT(ifp->if_broot == NULL); |
b5c71bcd SH |
776 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
777 | ||
778 | return error; | |
49f693fa | 779 | } |
a2ceac1f | 780 | |
49f693fa DC |
781 | /* |
782 | * Convert a local file to an extents file. | |
783 | * This code is out of bounds for data forks of regular files, | |
784 | * since the file data needs to get logged so things will stay consistent. | |
785 | * (The bmap-level manipulations are ok, though). | |
786 | */ | |
3f17ed4b DC |
787 | void |
788 | xfs_bmap_local_to_extents_empty( | |
789 | struct xfs_inode *ip, | |
790 | int whichfork) | |
791 | { | |
792 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
793 | ||
1277a5e0 | 794 | ASSERT(whichfork != XFS_COW_FORK); |
3f17ed4b DC |
795 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); |
796 | ASSERT(ifp->if_bytes == 0); | |
797 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0); | |
798 | ||
ff105f75 | 799 | xfs_bmap_forkoff_reset(ip, whichfork); |
3f17ed4b DC |
800 | ifp->if_flags &= ~XFS_IFINLINE; |
801 | ifp->if_flags |= XFS_IFEXTENTS; | |
b37d753d CH |
802 | ifp->if_u1.if_root = NULL; |
803 | ifp->if_height = 0; | |
3f17ed4b DC |
804 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); |
805 | } | |
806 | ||
807 | ||
49f693fa DC |
808 | STATIC int /* error */ |
809 | xfs_bmap_local_to_extents( | |
810 | xfs_trans_t *tp, /* transaction pointer */ | |
811 | xfs_inode_t *ip, /* incore inode pointer */ | |
49f693fa DC |
812 | xfs_extlen_t total, /* total blocks needed by transaction */ |
813 | int *logflagsp, /* inode logging flags */ | |
814 | int whichfork, | |
5dfa5cd2 DC |
815 | void (*init_fn)(struct xfs_trans *tp, |
816 | struct xfs_buf *bp, | |
49f693fa DC |
817 | struct xfs_inode *ip, |
818 | struct xfs_ifork *ifp)) | |
819 | { | |
3f17ed4b | 820 | int error = 0; |
49f693fa | 821 | int flags; /* logging flags returned */ |
e07055b8 | 822 | struct xfs_ifork *ifp; /* inode fork pointer */ |
3f17ed4b DC |
823 | xfs_alloc_arg_t args; /* allocation arguments */ |
824 | xfs_buf_t *bp; /* buffer for extent block */ | |
05104e43 | 825 | struct xfs_bmbt_irec rec; |
9788e059 | 826 | struct xfs_iext_cursor icur; |
2bd0ea18 | 827 | |
49f693fa DC |
828 | /* |
829 | * We don't want to deal with the case of keeping inode data inline yet. | |
830 | * So sending the data fork of a regular inode is invalid. | |
831 | */ | |
e37bf53c | 832 | ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK)); |
49f693fa DC |
833 | ifp = XFS_IFORK_PTR(ip, whichfork); |
834 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
3f17ed4b DC |
835 | |
836 | if (!ifp->if_bytes) { | |
837 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
838 | flags = XFS_ILOG_CORE; | |
839 | goto done; | |
840 | } | |
841 | ||
49f693fa DC |
842 | flags = 0; |
843 | error = 0; | |
b37d753d | 844 | ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE); |
3f17ed4b DC |
845 | memset(&args, 0, sizeof(args)); |
846 | args.tp = tp; | |
847 | args.mp = ip->i_mount; | |
85aec44f | 848 | xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0); |
3f17ed4b DC |
849 | /* |
850 | * Allocate a block. We know we need only one, since the | |
851 | * file currently fits in an inode. | |
852 | */ | |
f7253505 | 853 | if (tp->t_firstblock == NULLFSBLOCK) { |
3f17ed4b DC |
854 | args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino); |
855 | args.type = XFS_ALLOCTYPE_START_BNO; | |
49f693fa | 856 | } else { |
f7253505 | 857 | args.fsbno = tp->t_firstblock; |
3f17ed4b | 858 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
2bd0ea18 | 859 | } |
3f17ed4b DC |
860 | args.total = total; |
861 | args.minlen = args.maxlen = args.prod = 1; | |
862 | error = xfs_alloc_vextent(&args); | |
863 | if (error) | |
864 | goto done; | |
865 | ||
866 | /* Can't fail, the space was reserved. */ | |
867 | ASSERT(args.fsbno != NULLFSBLOCK); | |
868 | ASSERT(args.len == 1); | |
f7253505 | 869 | tp->t_firstblock = args.fsbno; |
3f17ed4b DC |
870 | bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0); |
871 | ||
19ebedcf | 872 | /* |
f44fbde0 | 873 | * Initialize the block, copy the data and log the remote buffer. |
19ebedcf | 874 | * |
f44fbde0 BF |
875 | * The callout is responsible for logging because the remote format |
876 | * might differ from the local format and thus we don't know how much to | |
877 | * log here. Note that init_fn must also set the buffer log item type | |
878 | * correctly. | |
19ebedcf | 879 | */ |
3f17ed4b DC |
880 | init_fn(tp, bp, ip, ifp); |
881 | ||
f44fbde0 | 882 | /* account for the change in fork size */ |
3f17ed4b DC |
883 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); |
884 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
49f693fa | 885 | flags |= XFS_ILOG_CORE; |
3f17ed4b | 886 | |
b37d753d CH |
887 | ifp->if_u1.if_root = NULL; |
888 | ifp->if_height = 0; | |
889 | ||
05104e43 CH |
890 | rec.br_startoff = 0; |
891 | rec.br_startblock = args.fsbno; | |
892 | rec.br_blockcount = 1; | |
893 | rec.br_state = XFS_EXT_NORM; | |
9788e059 | 894 | xfs_iext_first(ifp, &icur); |
26a75f67 | 895 | xfs_iext_insert(ip, &icur, &rec, 0); |
05104e43 | 896 | |
3f17ed4b DC |
897 | XFS_IFORK_NEXT_SET(ip, whichfork, 1); |
898 | ip->i_d.di_nblocks = 1; | |
899 | xfs_trans_mod_dquot_byino(tp, ip, | |
900 | XFS_TRANS_DQ_BCOUNT, 1L); | |
901 | flags |= xfs_ilog_fext(whichfork); | |
902 | ||
49f693fa DC |
903 | done: |
904 | *logflagsp = flags; | |
905 | return error; | |
2bd0ea18 NS |
906 | } |
907 | ||
908 | /* | |
49f693fa | 909 | * Called from xfs_bmap_add_attrfork to handle btree format files. |
2bd0ea18 | 910 | */ |
49f693fa DC |
911 | STATIC int /* error */ |
912 | xfs_bmap_add_attrfork_btree( | |
913 | xfs_trans_t *tp, /* transaction pointer */ | |
914 | xfs_inode_t *ip, /* incore inode pointer */ | |
49f693fa | 915 | int *flags) /* inode logging flags */ |
2bd0ea18 | 916 | { |
49f693fa DC |
917 | xfs_btree_cur_t *cur; /* btree cursor */ |
918 | int error; /* error return value */ | |
919 | xfs_mount_t *mp; /* file system mount struct */ | |
920 | int stat; /* newroot status */ | |
56b2de80 | 921 | |
49f693fa DC |
922 | mp = ip->i_mount; |
923 | if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip)) | |
924 | *flags |= XFS_ILOG_DBROOT; | |
925 | else { | |
926 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
4f76f49c CH |
927 | error = xfs_bmbt_lookup_first(cur, &stat); |
928 | if (error) | |
49f693fa DC |
929 | goto error0; |
930 | /* must be at least one entry */ | |
19ebedcf | 931 | XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0); |
49f693fa DC |
932 | if ((error = xfs_btree_new_iroot(cur, flags, &stat))) |
933 | goto error0; | |
934 | if (stat == 0) { | |
935 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
12b53197 | 936 | return -ENOSPC; |
49f693fa | 937 | } |
49f693fa DC |
938 | cur->bc_private.b.allocated = 0; |
939 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2bd0ea18 | 940 | } |
49f693fa DC |
941 | return 0; |
942 | error0: | |
943 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
944 | return error; | |
945 | } | |
56b2de80 | 946 | |
49f693fa DC |
947 | /* |
948 | * Called from xfs_bmap_add_attrfork to handle extents format files. | |
949 | */ | |
950 | STATIC int /* error */ | |
951 | xfs_bmap_add_attrfork_extents( | |
52f6ed9d BF |
952 | struct xfs_trans *tp, /* transaction pointer */ |
953 | struct xfs_inode *ip, /* incore inode pointer */ | |
49f693fa DC |
954 | int *flags) /* inode logging flags */ |
955 | { | |
956 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
957 | int error; /* error return value */ | |
958 | ||
959 | if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip)) | |
960 | return 0; | |
961 | cur = NULL; | |
f7253505 BF |
962 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags, |
963 | XFS_DATA_FORK); | |
49f693fa DC |
964 | if (cur) { |
965 | cur->bc_private.b.allocated = 0; | |
660265b7 | 966 | xfs_btree_del_cursor(cur, error); |
2bd0ea18 | 967 | } |
49f693fa DC |
968 | return error; |
969 | } | |
56b2de80 | 970 | |
49f693fa DC |
971 | /* |
972 | * Called from xfs_bmap_add_attrfork to handle local format files. Each | |
973 | * different data fork content type needs a different callout to do the | |
974 | * conversion. Some are basic and only require special block initialisation | |
975 | * callouts for the data formating, others (directories) are so specialised they | |
976 | * handle everything themselves. | |
977 | * | |
978 | * XXX (dgc): investigate whether directory conversion can use the generic | |
979 | * formatting callout. It should be possible - it's just a very complex | |
5dfa5cd2 | 980 | * formatter. |
49f693fa DC |
981 | */ |
982 | STATIC int /* error */ | |
983 | xfs_bmap_add_attrfork_local( | |
d9313ca4 BF |
984 | struct xfs_trans *tp, /* transaction pointer */ |
985 | struct xfs_inode *ip, /* incore inode pointer */ | |
49f693fa DC |
986 | int *flags) /* inode logging flags */ |
987 | { | |
d9313ca4 | 988 | struct xfs_da_args dargs; /* args for dir/attr code */ |
56b2de80 | 989 | |
49f693fa DC |
990 | if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip)) |
991 | return 0; | |
a2ceac1f | 992 | |
e37bf53c | 993 | if (S_ISDIR(VFS_I(ip)->i_mode)) { |
49f693fa | 994 | memset(&dargs, 0, sizeof(dargs)); |
ff105f75 | 995 | dargs.geo = ip->i_mount->m_dir_geo; |
49f693fa | 996 | dargs.dp = ip; |
ff105f75 | 997 | dargs.total = dargs.geo->fsbcount; |
49f693fa DC |
998 | dargs.whichfork = XFS_DATA_FORK; |
999 | dargs.trans = tp; | |
1000 | return xfs_dir2_sf_to_block(&dargs); | |
1001 | } | |
2bd0ea18 | 1002 | |
e37bf53c | 1003 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
f7253505 BF |
1004 | return xfs_bmap_local_to_extents(tp, ip, 1, flags, |
1005 | XFS_DATA_FORK, | |
49f693fa | 1006 | xfs_symlink_local_to_remote); |
56b2de80 | 1007 | |
3f17ed4b DC |
1008 | /* should only be called for types that support local format data */ |
1009 | ASSERT(0); | |
12b53197 | 1010 | return -EFSCORRUPTED; |
49f693fa | 1011 | } |
2bd0ea18 | 1012 | |
932b563e AH |
1013 | /* Set an inode attr fork off based on the format */ |
1014 | int | |
1015 | xfs_bmap_set_attrforkoff( | |
1016 | struct xfs_inode *ip, | |
1017 | int size, | |
1018 | int *version) | |
1019 | { | |
1020 | switch (ip->i_d.di_format) { | |
1021 | case XFS_DINODE_FMT_DEV: | |
1022 | ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; | |
1023 | break; | |
1024 | case XFS_DINODE_FMT_LOCAL: | |
1025 | case XFS_DINODE_FMT_EXTENTS: | |
1026 | case XFS_DINODE_FMT_BTREE: | |
1027 | ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size); | |
1028 | if (!ip->i_d.di_forkoff) | |
1029 | ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3; | |
1030 | else if ((ip->i_mount->m_flags & XFS_MOUNT_ATTR2) && version) | |
1031 | *version = 2; | |
1032 | break; | |
1033 | default: | |
1034 | ASSERT(0); | |
1035 | return -EINVAL; | |
1036 | } | |
1037 | ||
1038 | return 0; | |
1039 | } | |
1040 | ||
49f693fa DC |
1041 | /* |
1042 | * Convert inode from non-attributed to attributed. | |
1043 | * Must not be in a transaction, ip must not be locked. | |
1044 | */ | |
1045 | int /* error code */ | |
1046 | xfs_bmap_add_attrfork( | |
1047 | xfs_inode_t *ip, /* incore inode pointer */ | |
1048 | int size, /* space new attribute needs */ | |
1049 | int rsvd) /* xact may use reserved blks */ | |
1050 | { | |
49f693fa DC |
1051 | xfs_mount_t *mp; /* mount structure */ |
1052 | xfs_trans_t *tp; /* transaction pointer */ | |
1053 | int blks; /* space reservation */ | |
1054 | int version = 1; /* superblock attr version */ | |
49f693fa DC |
1055 | int logflags; /* logging flags */ |
1056 | int error; /* error return value */ | |
56b2de80 | 1057 | |
49f693fa | 1058 | ASSERT(XFS_IFORK_Q(ip) == 0); |
2bd0ea18 | 1059 | |
49f693fa DC |
1060 | mp = ip->i_mount; |
1061 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); | |
9074815c | 1062 | |
49f693fa | 1063 | blks = XFS_ADDAFORK_SPACE_RES(mp); |
9074815c CH |
1064 | |
1065 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0, | |
1066 | rsvd ? XFS_TRANS_RESERVE : 0, &tp); | |
1067 | if (error) | |
ff105f75 | 1068 | return error; |
9074815c | 1069 | |
49f693fa DC |
1070 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1071 | error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ? | |
1072 | XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES : | |
1073 | XFS_QMOPT_RES_REGBLKS); | |
ff105f75 DC |
1074 | if (error) |
1075 | goto trans_cancel; | |
49f693fa | 1076 | if (XFS_IFORK_Q(ip)) |
ff105f75 | 1077 | goto trans_cancel; |
9656ed10 DW |
1078 | if (ip->i_d.di_anextents != 0) { |
1079 | error = -EFSCORRUPTED; | |
1080 | goto trans_cancel; | |
1081 | } | |
49f693fa | 1082 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) { |
2bd0ea18 | 1083 | /* |
49f693fa | 1084 | * For inodes coming from pre-6.2 filesystems. |
2bd0ea18 | 1085 | */ |
49f693fa DC |
1086 | ASSERT(ip->i_d.di_aformat == 0); |
1087 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
5e656dbb | 1088 | } |
a2ceac1f | 1089 | |
ff105f75 | 1090 | xfs_trans_ijoin(tp, ip, 0); |
49f693fa | 1091 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
932b563e AH |
1092 | error = xfs_bmap_set_attrforkoff(ip, size, &version); |
1093 | if (error) | |
ff105f75 | 1094 | goto trans_cancel; |
49f693fa DC |
1095 | ASSERT(ip->i_afp == NULL); |
1096 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
1097 | ip->i_afp->if_flags = XFS_IFEXTENTS; | |
1098 | logflags = 0; | |
49f693fa DC |
1099 | switch (ip->i_d.di_format) { |
1100 | case XFS_DINODE_FMT_LOCAL: | |
d9313ca4 | 1101 | error = xfs_bmap_add_attrfork_local(tp, ip, &logflags); |
49f693fa DC |
1102 | break; |
1103 | case XFS_DINODE_FMT_EXTENTS: | |
d9313ca4 | 1104 | error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags); |
49f693fa DC |
1105 | break; |
1106 | case XFS_DINODE_FMT_BTREE: | |
d9313ca4 | 1107 | error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags); |
49f693fa DC |
1108 | break; |
1109 | default: | |
1110 | error = 0; | |
1111 | break; | |
1112 | } | |
1113 | if (logflags) | |
1114 | xfs_trans_log_inode(tp, ip, logflags); | |
1115 | if (error) | |
29ae66eb | 1116 | goto trans_cancel; |
49f693fa DC |
1117 | if (!xfs_sb_version_hasattr(&mp->m_sb) || |
1118 | (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) { | |
19ebedcf | 1119 | bool log_sb = false; |
a2ceac1f | 1120 | |
49f693fa DC |
1121 | spin_lock(&mp->m_sb_lock); |
1122 | if (!xfs_sb_version_hasattr(&mp->m_sb)) { | |
1123 | xfs_sb_version_addattr(&mp->m_sb); | |
19ebedcf | 1124 | log_sb = true; |
49f693fa DC |
1125 | } |
1126 | if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) { | |
1127 | xfs_sb_version_addattr2(&mp->m_sb); | |
19ebedcf | 1128 | log_sb = true; |
49f693fa | 1129 | } |
19ebedcf DC |
1130 | spin_unlock(&mp->m_sb_lock); |
1131 | if (log_sb) | |
1132 | xfs_log_sb(tp); | |
49f693fa DC |
1133 | } |
1134 | ||
de5a3f46 | 1135 | error = xfs_trans_commit(tp); |
ff105f75 DC |
1136 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1137 | return error; | |
1138 | ||
ff105f75 | 1139 | trans_cancel: |
3d7434fe | 1140 | xfs_trans_cancel(tp); |
49f693fa | 1141 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5e656dbb | 1142 | return error; |
2bd0ea18 NS |
1143 | } |
1144 | ||
399ab595 | 1145 | /* |
49f693fa DC |
1146 | * Internal and external extent tree search functions. |
1147 | */ | |
399ab595 | 1148 | |
49f693fa | 1149 | /* |
6d79c95c | 1150 | * Read in extents from a btree-format inode. |
49f693fa | 1151 | */ |
6d79c95c CH |
1152 | int |
1153 | xfs_iread_extents( | |
1154 | struct xfs_trans *tp, | |
1155 | struct xfs_inode *ip, | |
1156 | int whichfork) | |
49f693fa | 1157 | { |
6d79c95c | 1158 | struct xfs_mount *mp = ip->i_mount; |
322fd804 | 1159 | int state = xfs_bmap_fork_to_state(whichfork); |
6d79c95c CH |
1160 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1161 | xfs_extnum_t nextents = XFS_IFORK_NEXTENTS(ip, whichfork); | |
1162 | struct xfs_btree_block *block = ifp->if_broot; | |
9788e059 | 1163 | struct xfs_iext_cursor icur; |
b37d753d | 1164 | struct xfs_bmbt_irec new; |
6d79c95c CH |
1165 | xfs_fsblock_t bno; |
1166 | struct xfs_buf *bp; | |
1167 | xfs_extnum_t i, j; | |
1168 | int level; | |
1169 | __be64 *pp; | |
1170 | int error; | |
1171 | ||
1172 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
1173 | ||
1174 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1175 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
1176 | return -EFSCORRUPTED; | |
1177 | } | |
1178 | ||
399ab595 | 1179 | /* |
49f693fa | 1180 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. |
399ab595 | 1181 | */ |
49f693fa DC |
1182 | level = be16_to_cpu(block->bb_level); |
1183 | ASSERT(level > 0); | |
1184 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
1185 | bno = be64_to_cpu(*pp); | |
e2376544 | 1186 | |
399ab595 | 1187 | /* |
49f693fa DC |
1188 | * Go down the tree until leaf level is reached, following the first |
1189 | * pointer (leftmost) at each level. | |
399ab595 | 1190 | */ |
49f693fa DC |
1191 | while (level-- > 0) { |
1192 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1193 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1194 | if (error) | |
6d79c95c | 1195 | goto out; |
49f693fa | 1196 | block = XFS_BUF_TO_BLOCK(bp); |
49f693fa DC |
1197 | if (level == 0) |
1198 | break; | |
1199 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
1200 | bno = be64_to_cpu(*pp); | |
19ebedcf | 1201 | XFS_WANT_CORRUPTED_GOTO(mp, |
ecdc52ff | 1202 | xfs_verify_fsbno(mp, bno), out_brelse); |
49f693fa | 1203 | xfs_trans_brelse(tp, bp); |
399ab595 | 1204 | } |
6d79c95c | 1205 | |
399ab595 | 1206 | /* |
49f693fa | 1207 | * Here with bp and block set to the leftmost leaf node in the tree. |
399ab595 | 1208 | */ |
49f693fa | 1209 | i = 0; |
9788e059 | 1210 | xfs_iext_first(ifp, &icur); |
6d79c95c | 1211 | |
399ab595 | 1212 | /* |
49f693fa | 1213 | * Loop over all leaf nodes. Copy information to the extent records. |
399ab595 | 1214 | */ |
49f693fa DC |
1215 | for (;;) { |
1216 | xfs_bmbt_rec_t *frp; | |
1217 | xfs_fsblock_t nextbno; | |
1218 | xfs_extnum_t num_recs; | |
399ab595 | 1219 | |
49f693fa | 1220 | num_recs = xfs_btree_get_numrecs(block); |
6d79c95c | 1221 | if (unlikely(i + num_recs > nextents)) { |
49f693fa DC |
1222 | xfs_warn(ip->i_mount, |
1223 | "corrupt dinode %Lu, (btree extents).", | |
1224 | (unsigned long long) ip->i_ino); | |
b02a2c9e DW |
1225 | xfs_inode_verifier_error(ip, -EFSCORRUPTED, |
1226 | __func__, block, sizeof(*block), | |
1227 | __this_address); | |
6d79c95c CH |
1228 | error = -EFSCORRUPTED; |
1229 | goto out_brelse; | |
399ab595 NS |
1230 | } |
1231 | /* | |
49f693fa | 1232 | * Read-ahead the next leaf block, if any. |
399ab595 | 1233 | */ |
49f693fa DC |
1234 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
1235 | if (nextbno != NULLFSBLOCK) | |
1236 | xfs_btree_reada_bufl(mp, nextbno, 1, | |
1237 | &xfs_bmbt_buf_ops); | |
399ab595 | 1238 | /* |
49f693fa | 1239 | * Copy records into the extent records. |
399ab595 | 1240 | */ |
49f693fa | 1241 | frp = XFS_BMBT_REC_ADDR(mp, block, 1); |
b37d753d | 1242 | for (j = 0; j < num_recs; j++, frp++, i++) { |
0cf6a3a9 DW |
1243 | xfs_failaddr_t fa; |
1244 | ||
080f0c71 | 1245 | xfs_bmbt_disk_get_all(frp, &new); |
0cf6a3a9 DW |
1246 | fa = xfs_bmap_validate_extent(ip, whichfork, &new); |
1247 | if (fa) { | |
6d79c95c | 1248 | error = -EFSCORRUPTED; |
0cf6a3a9 DW |
1249 | xfs_inode_verifier_error(ip, error, |
1250 | "xfs_iread_extents(2)", | |
1251 | frp, sizeof(*frp), fa); | |
6d79c95c | 1252 | goto out_brelse; |
49f693fa | 1253 | } |
26a75f67 | 1254 | xfs_iext_insert(ip, &icur, &new, state); |
9788e059 CH |
1255 | trace_xfs_read_extent(ip, &icur, state, _THIS_IP_); |
1256 | xfs_iext_next(ifp, &icur); | |
399ab595 | 1257 | } |
49f693fa DC |
1258 | xfs_trans_brelse(tp, bp); |
1259 | bno = nextbno; | |
399ab595 | 1260 | /* |
49f693fa | 1261 | * If we've reached the end, stop. |
399ab595 | 1262 | */ |
49f693fa DC |
1263 | if (bno == NULLFSBLOCK) |
1264 | break; | |
1265 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1266 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1267 | if (error) | |
6d79c95c | 1268 | goto out; |
49f693fa | 1269 | block = XFS_BUF_TO_BLOCK(bp); |
399ab595 | 1270 | } |
6d79c95c CH |
1271 | |
1272 | if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) { | |
1273 | error = -EFSCORRUPTED; | |
1274 | goto out; | |
1275 | } | |
d09d4e5f | 1276 | ASSERT(i == xfs_iext_count(ifp)); |
6d79c95c CH |
1277 | |
1278 | ifp->if_flags |= XFS_IFEXTENTS; | |
49f693fa | 1279 | return 0; |
6d79c95c CH |
1280 | |
1281 | out_brelse: | |
49f693fa | 1282 | xfs_trans_brelse(tp, bp); |
6d79c95c CH |
1283 | out: |
1284 | xfs_iext_destroy(ifp); | |
1285 | return error; | |
49f693fa | 1286 | } |
399ab595 | 1287 | |
49f693fa | 1288 | /* |
2d3a6501 CH |
1289 | * Returns the relative block number of the first unused block(s) in the given |
1290 | * fork with at least "len" logically contiguous blocks free. This is the | |
1291 | * lowest-address hole if the fork has holes, else the first block past the end | |
1292 | * of fork. Return 0 if the fork is currently local (in-inode). | |
49f693fa DC |
1293 | */ |
1294 | int /* error */ | |
1295 | xfs_bmap_first_unused( | |
2d3a6501 CH |
1296 | struct xfs_trans *tp, /* transaction pointer */ |
1297 | struct xfs_inode *ip, /* incore inode */ | |
1298 | xfs_extlen_t len, /* size of hole to find */ | |
1299 | xfs_fileoff_t *first_unused, /* unused block */ | |
1300 | int whichfork) /* data or attr fork */ | |
49f693fa | 1301 | { |
2d3a6501 CH |
1302 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1303 | struct xfs_bmbt_irec got; | |
9788e059 | 1304 | struct xfs_iext_cursor icur; |
2d3a6501 CH |
1305 | xfs_fileoff_t lastaddr = 0; |
1306 | xfs_fileoff_t lowest, max; | |
1307 | int error; | |
49f693fa DC |
1308 | |
1309 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE || | |
1310 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS || | |
1311 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
2d3a6501 | 1312 | |
49f693fa DC |
1313 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { |
1314 | *first_unused = 0; | |
1315 | return 0; | |
1316 | } | |
f71f3bc3 | 1317 | |
2d3a6501 CH |
1318 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
1319 | error = xfs_iread_extents(tp, ip, whichfork); | |
1320 | if (error) | |
1321 | return error; | |
1322 | } | |
f71f3bc3 | 1323 | |
2d3a6501 | 1324 | lowest = max = *first_unused; |
9788e059 | 1325 | for_each_xfs_iext(ifp, &icur, &got) { |
2bd0ea18 | 1326 | /* |
49f693fa | 1327 | * See if the hole before this extent will work. |
2bd0ea18 | 1328 | */ |
f71f3bc3 | 1329 | if (got.br_startoff >= lowest + len && |
2d3a6501 CH |
1330 | got.br_startoff - max >= len) |
1331 | break; | |
f71f3bc3 | 1332 | lastaddr = got.br_startoff + got.br_blockcount; |
49f693fa DC |
1333 | max = XFS_FILEOFF_MAX(lastaddr, lowest); |
1334 | } | |
2d3a6501 | 1335 | |
49f693fa DC |
1336 | *first_unused = max; |
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | /* | |
e6d77a21 | 1341 | * Returns the file-relative block number of the last block - 1 before |
49f693fa DC |
1342 | * last_block (input value) in the file. |
1343 | * This is not based on i_size, it is based on the extent records. | |
1344 | * Returns 0 for local files, as they do not have extent records. | |
1345 | */ | |
1346 | int /* error */ | |
1347 | xfs_bmap_last_before( | |
b0385364 CH |
1348 | struct xfs_trans *tp, /* transaction pointer */ |
1349 | struct xfs_inode *ip, /* incore inode */ | |
1350 | xfs_fileoff_t *last_block, /* last block */ | |
1351 | int whichfork) /* data or attr fork */ | |
49f693fa | 1352 | { |
b0385364 CH |
1353 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1354 | struct xfs_bmbt_irec got; | |
9788e059 | 1355 | struct xfs_iext_cursor icur; |
b0385364 | 1356 | int error; |
49f693fa | 1357 | |
b0385364 CH |
1358 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { |
1359 | case XFS_DINODE_FMT_LOCAL: | |
49f693fa DC |
1360 | *last_block = 0; |
1361 | return 0; | |
b0385364 CH |
1362 | case XFS_DINODE_FMT_BTREE: |
1363 | case XFS_DINODE_FMT_EXTENTS: | |
1364 | break; | |
1365 | default: | |
1366 | return -EIO; | |
49f693fa | 1367 | } |
b0385364 CH |
1368 | |
1369 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1370 | error = xfs_iread_extents(tp, ip, whichfork); | |
1371 | if (error) | |
1372 | return error; | |
2bd0ea18 | 1373 | } |
b0385364 | 1374 | |
9788e059 | 1375 | if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got)) |
d6fbe8fe | 1376 | *last_block = 0; |
49f693fa | 1377 | return 0; |
5e656dbb BN |
1378 | } |
1379 | ||
613e6057 | 1380 | int |
49f693fa DC |
1381 | xfs_bmap_last_extent( |
1382 | struct xfs_trans *tp, | |
1383 | struct xfs_inode *ip, | |
1384 | int whichfork, | |
1385 | struct xfs_bmbt_irec *rec, | |
1386 | int *is_empty) | |
56b2de80 | 1387 | { |
49f693fa | 1388 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
9788e059 | 1389 | struct xfs_iext_cursor icur; |
56b2de80 DC |
1390 | int error; |
1391 | ||
49f693fa DC |
1392 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
1393 | error = xfs_iread_extents(tp, ip, whichfork); | |
1394 | if (error) | |
1395 | return error; | |
1396 | } | |
1397 | ||
9788e059 CH |
1398 | xfs_iext_last(ifp, &icur); |
1399 | if (!xfs_iext_get_extent(ifp, &icur, rec)) | |
49f693fa | 1400 | *is_empty = 1; |
9788e059 CH |
1401 | else |
1402 | *is_empty = 0; | |
49f693fa DC |
1403 | return 0; |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * Check the last inode extent to determine whether this allocation will result | |
1408 | * in blocks being allocated at the end of the file. When we allocate new data | |
1409 | * blocks at the end of the file which do not start at the previous data block, | |
1410 | * we will try to align the new blocks at stripe unit boundaries. | |
1411 | * | |
ff105f75 | 1412 | * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be |
49f693fa DC |
1413 | * at, or past the EOF. |
1414 | */ | |
1415 | STATIC int | |
1416 | xfs_bmap_isaeof( | |
1417 | struct xfs_bmalloca *bma, | |
1418 | int whichfork) | |
1419 | { | |
1420 | struct xfs_bmbt_irec rec; | |
1421 | int is_empty; | |
1422 | int error; | |
1423 | ||
180d3cd8 | 1424 | bma->aeof = false; |
49f693fa DC |
1425 | error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec, |
1426 | &is_empty); | |
ff105f75 | 1427 | if (error) |
49f693fa | 1428 | return error; |
56b2de80 | 1429 | |
ff105f75 | 1430 | if (is_empty) { |
180d3cd8 | 1431 | bma->aeof = true; |
ff105f75 DC |
1432 | return 0; |
1433 | } | |
1434 | ||
56b2de80 | 1435 | /* |
49f693fa DC |
1436 | * Check if we are allocation or past the last extent, or at least into |
1437 | * the last delayed allocated extent. | |
56b2de80 | 1438 | */ |
49f693fa DC |
1439 | bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount || |
1440 | (bma->offset >= rec.br_startoff && | |
1441 | isnullstartblock(rec.br_startblock)); | |
1442 | return 0; | |
1443 | } | |
56b2de80 | 1444 | |
49f693fa DC |
1445 | /* |
1446 | * Returns the file-relative block number of the first block past eof in | |
1447 | * the file. This is not based on i_size, it is based on the extent records. | |
1448 | * Returns 0 for local files, as they do not have extent records. | |
1449 | */ | |
1450 | int | |
1451 | xfs_bmap_last_offset( | |
49f693fa DC |
1452 | struct xfs_inode *ip, |
1453 | xfs_fileoff_t *last_block, | |
1454 | int whichfork) | |
1455 | { | |
1456 | struct xfs_bmbt_irec rec; | |
1457 | int is_empty; | |
1458 | int error; | |
1459 | ||
1460 | *last_block = 0; | |
1461 | ||
1462 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) | |
1463 | return 0; | |
1464 | ||
1465 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE && | |
1466 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
12b53197 | 1467 | return -EIO; |
49f693fa DC |
1468 | |
1469 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty); | |
1470 | if (error || is_empty) | |
1471 | return error; | |
1472 | ||
1473 | *last_block = rec.br_startoff + rec.br_blockcount; | |
1474 | return 0; | |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * Returns whether the selected fork of the inode has exactly one | |
1479 | * block or not. For the data fork we check this matches di_size, | |
1480 | * implying the file's range is 0..bsize-1. | |
1481 | */ | |
1482 | int /* 1=>1 block, 0=>otherwise */ | |
1483 | xfs_bmap_one_block( | |
1484 | xfs_inode_t *ip, /* incore inode */ | |
1485 | int whichfork) /* data or attr fork */ | |
1486 | { | |
e07055b8 | 1487 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa DC |
1488 | int rval; /* return value */ |
1489 | xfs_bmbt_irec_t s; /* internal version of extent */ | |
9788e059 | 1490 | struct xfs_iext_cursor icur; |
49f693fa DC |
1491 | |
1492 | #ifndef DEBUG | |
1493 | if (whichfork == XFS_DATA_FORK) | |
1494 | return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize; | |
1495 | #endif /* !DEBUG */ | |
1496 | if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1) | |
1497 | return 0; | |
1498 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
1499 | return 0; | |
1500 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1501 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
9788e059 CH |
1502 | xfs_iext_first(ifp, &icur); |
1503 | xfs_iext_get_extent(ifp, &icur, &s); | |
49f693fa DC |
1504 | rval = s.br_startoff == 0 && s.br_blockcount == 1; |
1505 | if (rval && whichfork == XFS_DATA_FORK) | |
1506 | ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize); | |
1507 | return rval; | |
1508 | } | |
1509 | ||
1510 | /* | |
1511 | * Extent tree manipulation functions used during allocation. | |
1512 | */ | |
1513 | ||
1514 | /* | |
1515 | * Convert a delayed allocation to a real allocation. | |
1516 | */ | |
1517 | STATIC int /* error */ | |
1518 | xfs_bmap_add_extent_delay_real( | |
1277a5e0 DW |
1519 | struct xfs_bmalloca *bma, |
1520 | int whichfork) | |
49f693fa DC |
1521 | { |
1522 | struct xfs_bmbt_irec *new = &bma->got; | |
49f693fa DC |
1523 | int error; /* error return value */ |
1524 | int i; /* temp state */ | |
e07055b8 | 1525 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa DC |
1526 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
1527 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
1528 | /* left is 0, right is 1, prev is 2 */ | |
1529 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 1530 | int state = xfs_bmap_fork_to_state(whichfork); |
49f693fa DC |
1531 | xfs_filblks_t da_new; /* new count del alloc blocks used */ |
1532 | xfs_filblks_t da_old; /* old count del alloc blocks used */ | |
1533 | xfs_filblks_t temp=0; /* value for da_new calculations */ | |
49f693fa | 1534 | int tmp_rval; /* partial logging flags */ |
19ebedcf | 1535 | struct xfs_mount *mp; |
1277a5e0 | 1536 | xfs_extnum_t *nextents; |
19884717 | 1537 | struct xfs_bmbt_irec old; |
49f693fa | 1538 | |
65ca3804 | 1539 | mp = bma->ip->i_mount; |
36e8786d | 1540 | ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
1277a5e0 DW |
1541 | ASSERT(whichfork != XFS_ATTR_FORK); |
1542 | nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents : | |
1543 | &bma->ip->i_d.di_nextents); | |
56b2de80 | 1544 | |
49f693fa DC |
1545 | ASSERT(!isnullstartblock(new->br_startblock)); |
1546 | ASSERT(!bma->cur || | |
1547 | (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); | |
56b2de80 | 1548 | |
79896434 | 1549 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
1550 | |
1551 | #define LEFT r[0] | |
1552 | #define RIGHT r[1] | |
1553 | #define PREV r[2] | |
56b2de80 DC |
1554 | |
1555 | /* | |
49f693fa | 1556 | * Set up a bunch of variables to make the tests simpler. |
56b2de80 | 1557 | */ |
9788e059 | 1558 | xfs_iext_get_extent(ifp, &bma->icur, &PREV); |
49f693fa | 1559 | new_endoff = new->br_startoff + new->br_blockcount; |
19884717 | 1560 | ASSERT(isnullstartblock(PREV.br_startblock)); |
49f693fa DC |
1561 | ASSERT(PREV.br_startoff <= new->br_startoff); |
1562 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
1563 | ||
1564 | da_old = startblockval(PREV.br_startblock); | |
1565 | da_new = 0; | |
1566 | ||
56b2de80 | 1567 | /* |
49f693fa DC |
1568 | * Set flags determining what part of the previous delayed allocation |
1569 | * extent is being replaced by a real allocation. | |
56b2de80 | 1570 | */ |
49f693fa DC |
1571 | if (PREV.br_startoff == new->br_startoff) |
1572 | state |= BMAP_LEFT_FILLING; | |
1573 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
1574 | state |= BMAP_RIGHT_FILLING; | |
1575 | ||
56b2de80 | 1576 | /* |
49f693fa DC |
1577 | * Check and set flags if this segment has a left neighbor. |
1578 | * Don't set contiguous if the combined extent would be too large. | |
56b2de80 | 1579 | */ |
9788e059 | 1580 | if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) { |
49f693fa | 1581 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
1582 | if (isnullstartblock(LEFT.br_startblock)) |
1583 | state |= BMAP_LEFT_DELAY; | |
1584 | } | |
1585 | ||
1586 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
1587 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
1588 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
1589 | LEFT.br_state == new->br_state && | |
1590 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
1591 | state |= BMAP_LEFT_CONTIG; | |
56b2de80 DC |
1592 | |
1593 | /* | |
49f693fa DC |
1594 | * Check and set flags if this segment has a right neighbor. |
1595 | * Don't set contiguous if the combined extent would be too large. | |
1596 | * Also check for all-three-contiguous being too large. | |
56b2de80 | 1597 | */ |
9788e059 | 1598 | if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) { |
49f693fa | 1599 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
1600 | if (isnullstartblock(RIGHT.br_startblock)) |
1601 | state |= BMAP_RIGHT_DELAY; | |
1602 | } | |
56b2de80 | 1603 | |
49f693fa DC |
1604 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
1605 | new_endoff == RIGHT.br_startoff && | |
1606 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
1607 | new->br_state == RIGHT.br_state && | |
1608 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
1609 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1610 | BMAP_RIGHT_FILLING)) != | |
1611 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1612 | BMAP_RIGHT_FILLING) || | |
1613 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
1614 | <= MAXEXTLEN)) | |
1615 | state |= BMAP_RIGHT_CONTIG; | |
5e656dbb | 1616 | |
49f693fa DC |
1617 | error = 0; |
1618 | /* | |
1619 | * Switch out based on the FILLING and CONTIG state bits. | |
1620 | */ | |
1621 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1622 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
1623 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1624 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1625 | /* | |
1626 | * Filling in all of a previously delayed allocation extent. | |
1627 | * The left and right neighbors are both contiguous with new. | |
1628 | */ | |
19884717 | 1629 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
a2ceac1f | 1630 | |
cf455a62 CH |
1631 | xfs_iext_remove(bma->ip, &bma->icur, state); |
1632 | xfs_iext_remove(bma->ip, &bma->icur, state); | |
9788e059 CH |
1633 | xfs_iext_prev(ifp, &bma->icur); |
1634 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
1277a5e0 | 1635 | (*nextents)--; |
12ebbfe1 | 1636 | |
49f693fa DC |
1637 | if (bma->cur == NULL) |
1638 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1639 | else { | |
1640 | rval = XFS_ILOG_CORE; | |
70a93110 | 1641 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
49f693fa DC |
1642 | if (error) |
1643 | goto done; | |
19ebedcf | 1644 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1645 | error = xfs_btree_delete(bma->cur, &i); |
1646 | if (error) | |
1647 | goto done; | |
19ebedcf | 1648 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1649 | error = xfs_btree_decrement(bma->cur, 0, &i); |
1650 | if (error) | |
1651 | goto done; | |
19ebedcf | 1652 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1653 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1654 | if (error) |
1655 | goto done; | |
5e656dbb | 1656 | } |
49f693fa DC |
1657 | break; |
1658 | ||
1659 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1660 | /* | |
1661 | * Filling in all of a previously delayed allocation extent. | |
1662 | * The left neighbor is contiguous, the right is not. | |
1663 | */ | |
19884717 | 1664 | old = LEFT; |
19884717 | 1665 | LEFT.br_blockcount += PREV.br_blockcount; |
12ebbfe1 | 1666 | |
cf455a62 | 1667 | xfs_iext_remove(bma->ip, &bma->icur, state); |
9788e059 CH |
1668 | xfs_iext_prev(ifp, &bma->icur); |
1669 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
49f693fa | 1670 | |
49f693fa DC |
1671 | if (bma->cur == NULL) |
1672 | rval = XFS_ILOG_DEXT; | |
1673 | else { | |
1674 | rval = 0; | |
70a93110 | 1675 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1676 | if (error) |
1677 | goto done; | |
19ebedcf | 1678 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1679 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1680 | if (error) |
1681 | goto done; | |
1682 | } | |
1683 | break; | |
1684 | ||
1685 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1686 | /* | |
1687 | * Filling in all of a previously delayed allocation extent. | |
3833299f DC |
1688 | * The right neighbor is contiguous, the left is not. Take care |
1689 | * with delay -> unwritten extent allocation here because the | |
1690 | * delalloc record we are overwriting is always written. | |
49f693fa | 1691 | */ |
19884717 CH |
1692 | PREV.br_startblock = new->br_startblock; |
1693 | PREV.br_blockcount += RIGHT.br_blockcount; | |
3833299f | 1694 | PREV.br_state = new->br_state; |
12ebbfe1 | 1695 | |
9788e059 | 1696 | xfs_iext_next(ifp, &bma->icur); |
cf455a62 | 1697 | xfs_iext_remove(bma->ip, &bma->icur, state); |
9788e059 CH |
1698 | xfs_iext_prev(ifp, &bma->icur); |
1699 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); | |
49f693fa | 1700 | |
49f693fa DC |
1701 | if (bma->cur == NULL) |
1702 | rval = XFS_ILOG_DEXT; | |
1703 | else { | |
1704 | rval = 0; | |
70a93110 | 1705 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
49f693fa DC |
1706 | if (error) |
1707 | goto done; | |
19ebedcf | 1708 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1709 | error = xfs_bmbt_update(bma->cur, &PREV); |
49f693fa DC |
1710 | if (error) |
1711 | goto done; | |
1712 | } | |
1713 | break; | |
1714 | ||
1715 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
1716 | /* | |
1717 | * Filling in all of a previously delayed allocation extent. | |
1718 | * Neither the left nor right neighbors are contiguous with | |
1719 | * the new one. | |
1720 | */ | |
19884717 CH |
1721 | PREV.br_startblock = new->br_startblock; |
1722 | PREV.br_state = new->br_state; | |
9788e059 | 1723 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
5e656dbb | 1724 | |
1277a5e0 | 1725 | (*nextents)++; |
49f693fa DC |
1726 | if (bma->cur == NULL) |
1727 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1728 | else { | |
1729 | rval = XFS_ILOG_CORE; | |
70a93110 | 1730 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1731 | if (error) |
1732 | goto done; | |
19ebedcf | 1733 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1734 | error = xfs_btree_insert(bma->cur, &i); |
1735 | if (error) | |
1736 | goto done; | |
19ebedcf | 1737 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1738 | } |
1739 | break; | |
5e656dbb | 1740 | |
49f693fa DC |
1741 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
1742 | /* | |
1743 | * Filling in the first part of a previous delayed allocation. | |
1744 | * The left neighbor is contiguous. | |
1745 | */ | |
19884717 CH |
1746 | old = LEFT; |
1747 | temp = PREV.br_blockcount - new->br_blockcount; | |
1748 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
1749 | startblockval(PREV.br_startblock)); | |
1750 | ||
19884717 | 1751 | LEFT.br_blockcount += new->br_blockcount; |
a2ceac1f | 1752 | |
bc5dec54 | 1753 | PREV.br_blockcount = temp; |
19884717 CH |
1754 | PREV.br_startoff += new->br_blockcount; |
1755 | PREV.br_startblock = nullstartblock(da_new); | |
12ebbfe1 | 1756 | |
9788e059 CH |
1757 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
1758 | xfs_iext_prev(ifp, &bma->icur); | |
1759 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
19884717 | 1760 | |
49f693fa DC |
1761 | if (bma->cur == NULL) |
1762 | rval = XFS_ILOG_DEXT; | |
1763 | else { | |
1764 | rval = 0; | |
70a93110 | 1765 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1766 | if (error) |
1767 | goto done; | |
19ebedcf | 1768 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1769 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1770 | if (error) |
1771 | goto done; | |
2bd0ea18 | 1772 | } |
49f693fa DC |
1773 | break; |
1774 | ||
1775 | case BMAP_LEFT_FILLING: | |
2bd0ea18 | 1776 | /* |
49f693fa DC |
1777 | * Filling in the first part of a previous delayed allocation. |
1778 | * The left neighbor is not contiguous. | |
5000d01d | 1779 | */ |
9788e059 | 1780 | xfs_iext_update_extent(bma->ip, state, &bma->icur, new); |
1277a5e0 | 1781 | (*nextents)++; |
49f693fa DC |
1782 | if (bma->cur == NULL) |
1783 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1784 | else { | |
1785 | rval = XFS_ILOG_CORE; | |
70a93110 | 1786 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1787 | if (error) |
1788 | goto done; | |
19ebedcf | 1789 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1790 | error = xfs_btree_insert(bma->cur, &i); |
1791 | if (error) | |
1792 | goto done; | |
19ebedcf | 1793 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1794 | } |
1795 | ||
36e8786d | 1796 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1797 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f7253505 | 1798 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1799 | rval |= tmp_rval; |
1800 | if (error) | |
1801 | goto done; | |
1802 | } | |
19884717 CH |
1803 | |
1804 | temp = PREV.br_blockcount - new->br_blockcount; | |
49f693fa DC |
1805 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1806 | startblockval(PREV.br_startblock) - | |
1807 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
19884717 | 1808 | |
19884717 CH |
1809 | PREV.br_startoff = new_endoff; |
1810 | PREV.br_blockcount = temp; | |
1811 | PREV.br_startblock = nullstartblock(da_new); | |
9788e059 | 1812 | xfs_iext_next(ifp, &bma->icur); |
26a75f67 | 1813 | xfs_iext_insert(bma->ip, &bma->icur, &PREV, state); |
9788e059 | 1814 | xfs_iext_prev(ifp, &bma->icur); |
49f693fa DC |
1815 | break; |
1816 | ||
1817 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
5e656dbb | 1818 | /* |
49f693fa DC |
1819 | * Filling in the last part of a previous delayed allocation. |
1820 | * The right neighbor is contiguous with the new allocation. | |
5e656dbb | 1821 | */ |
19884717 | 1822 | old = RIGHT; |
19884717 CH |
1823 | RIGHT.br_startoff = new->br_startoff; |
1824 | RIGHT.br_startblock = new->br_startblock; | |
1825 | RIGHT.br_blockcount += new->br_blockcount; | |
19884717 | 1826 | |
49f693fa DC |
1827 | if (bma->cur == NULL) |
1828 | rval = XFS_ILOG_DEXT; | |
1829 | else { | |
1830 | rval = 0; | |
70a93110 | 1831 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1832 | if (error) |
1833 | goto done; | |
19ebedcf | 1834 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 1835 | error = xfs_bmbt_update(bma->cur, &RIGHT); |
49f693fa DC |
1836 | if (error) |
1837 | goto done; | |
1838 | } | |
1839 | ||
19884717 | 1840 | temp = PREV.br_blockcount - new->br_blockcount; |
49f693fa DC |
1841 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1842 | startblockval(PREV.br_startblock)); | |
19884717 | 1843 | |
19884717 CH |
1844 | PREV.br_blockcount = temp; |
1845 | PREV.br_startblock = nullstartblock(da_new); | |
49f693fa | 1846 | |
9788e059 CH |
1847 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
1848 | xfs_iext_next(ifp, &bma->icur); | |
1849 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT); | |
49f693fa DC |
1850 | break; |
1851 | ||
1852 | case BMAP_RIGHT_FILLING: | |
a2ceac1f | 1853 | /* |
49f693fa DC |
1854 | * Filling in the last part of a previous delayed allocation. |
1855 | * The right neighbor is not contiguous. | |
a2ceac1f | 1856 | */ |
9788e059 | 1857 | xfs_iext_update_extent(bma->ip, state, &bma->icur, new); |
1277a5e0 | 1858 | (*nextents)++; |
49f693fa DC |
1859 | if (bma->cur == NULL) |
1860 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1861 | else { | |
1862 | rval = XFS_ILOG_CORE; | |
70a93110 | 1863 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1864 | if (error) |
1865 | goto done; | |
19ebedcf | 1866 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1867 | error = xfs_btree_insert(bma->cur, &i); |
1868 | if (error) | |
1869 | goto done; | |
19ebedcf | 1870 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 1871 | } |
a2ceac1f | 1872 | |
36e8786d | 1873 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1874 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f7253505 | 1875 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1876 | rval |= tmp_rval; |
1877 | if (error) | |
1878 | goto done; | |
1879 | } | |
19884717 CH |
1880 | |
1881 | temp = PREV.br_blockcount - new->br_blockcount; | |
49f693fa DC |
1882 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1883 | startblockval(PREV.br_startblock) - | |
1884 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
19884717 | 1885 | |
19884717 CH |
1886 | PREV.br_startblock = nullstartblock(da_new); |
1887 | PREV.br_blockcount = temp; | |
26a75f67 | 1888 | xfs_iext_insert(bma->ip, &bma->icur, &PREV, state); |
9788e059 | 1889 | xfs_iext_next(ifp, &bma->icur); |
49f693fa DC |
1890 | break; |
1891 | ||
1892 | case 0: | |
5e656dbb | 1893 | /* |
49f693fa DC |
1894 | * Filling in the middle part of a previous delayed allocation. |
1895 | * Contiguity is impossible here. | |
1896 | * This case is avoided almost all the time. | |
1897 | * | |
1898 | * We start with a delayed allocation: | |
1899 | * | |
1900 | * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+ | |
1901 | * PREV @ idx | |
1902 | * | |
1903 | * and we are allocating: | |
1904 | * +rrrrrrrrrrrrrrrrr+ | |
1905 | * new | |
1906 | * | |
1907 | * and we set it up for insertion as: | |
1908 | * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+ | |
1909 | * new | |
1910 | * PREV @ idx LEFT RIGHT | |
1911 | * inserted at idx + 1 | |
5e656dbb | 1912 | */ |
19884717 CH |
1913 | old = PREV; |
1914 | ||
1915 | /* LEFT is the new middle */ | |
49f693fa | 1916 | LEFT = *new; |
19884717 CH |
1917 | |
1918 | /* RIGHT is the new right */ | |
49f693fa | 1919 | RIGHT.br_state = PREV.br_state; |
49f693fa | 1920 | RIGHT.br_startoff = new_endoff; |
19884717 CH |
1921 | RIGHT.br_blockcount = |
1922 | PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
1923 | RIGHT.br_startblock = | |
1924 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1925 | RIGHT.br_blockcount)); | |
1926 | ||
1927 | /* truncate PREV */ | |
19884717 CH |
1928 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
1929 | PREV.br_startblock = | |
1930 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1931 | PREV.br_blockcount)); | |
9788e059 | 1932 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
19884717 | 1933 | |
9788e059 | 1934 | xfs_iext_next(ifp, &bma->icur); |
26a75f67 CH |
1935 | xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state); |
1936 | xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state); | |
1277a5e0 | 1937 | (*nextents)++; |
19884717 | 1938 | |
49f693fa DC |
1939 | if (bma->cur == NULL) |
1940 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1941 | else { | |
1942 | rval = XFS_ILOG_CORE; | |
70a93110 | 1943 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1944 | if (error) |
1945 | goto done; | |
19ebedcf | 1946 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1947 | error = xfs_btree_insert(bma->cur, &i); |
1948 | if (error) | |
1949 | goto done; | |
19ebedcf | 1950 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 1951 | } |
5e656dbb | 1952 | |
36e8786d | 1953 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1954 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f7253505 | 1955 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1956 | rval |= tmp_rval; |
1957 | if (error) | |
1958 | goto done; | |
1959 | } | |
19884717 CH |
1960 | |
1961 | da_new = startblockval(PREV.br_startblock) + | |
1962 | startblockval(RIGHT.br_startblock); | |
49f693fa DC |
1963 | break; |
1964 | ||
1965 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1966 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1967 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
1968 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1969 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1970 | case BMAP_LEFT_CONTIG: | |
1971 | case BMAP_RIGHT_CONTIG: | |
1972 | /* | |
1973 | * These cases are all impossible. | |
1974 | */ | |
1975 | ASSERT(0); | |
1976 | } | |
1977 | ||
5f847f1e DW |
1978 | /* add reverse mapping unless caller opted out */ |
1979 | if (!(bma->flags & XFS_BMAPI_NORMAP)) { | |
21375e5d | 1980 | error = xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new); |
5f847f1e DW |
1981 | if (error) |
1982 | goto done; | |
1983 | } | |
d7f80320 | 1984 | |
49f693fa | 1985 | /* convert to a btree if necessary */ |
36e8786d | 1986 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa DC |
1987 | int tmp_logflags; /* partial log flag return val */ |
1988 | ||
1989 | ASSERT(bma->cur == NULL); | |
1990 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, | |
f7253505 BF |
1991 | &bma->cur, da_old > 0, &tmp_logflags, |
1992 | whichfork); | |
49f693fa DC |
1993 | bma->logflags |= tmp_logflags; |
1994 | if (error) | |
1995 | goto done; | |
1996 | } | |
1997 | ||
5bfca476 CH |
1998 | if (bma->cur) { |
1999 | da_new += bma->cur->bc_private.b.allocated; | |
2000 | bma->cur->bc_private.b.allocated = 0; | |
49f693fa DC |
2001 | } |
2002 | ||
5bfca476 CH |
2003 | /* adjust for changes in reserved delayed indirect blocks */ |
2004 | if (da_new != da_old) { | |
2005 | ASSERT(state == 0 || da_new < da_old); | |
2006 | error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), | |
2007 | false); | |
2008 | } | |
49f693fa | 2009 | |
36e8786d | 2010 | xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork); |
49f693fa | 2011 | done: |
1277a5e0 DW |
2012 | if (whichfork != XFS_COW_FORK) |
2013 | bma->logflags |= rval; | |
49f693fa DC |
2014 | return error; |
2015 | #undef LEFT | |
2016 | #undef RIGHT | |
2017 | #undef PREV | |
2bd0ea18 NS |
2018 | } |
2019 | ||
2020 | /* | |
49f693fa | 2021 | * Convert an unwritten allocation to a real allocation or vice versa. |
2bd0ea18 NS |
2022 | */ |
2023 | STATIC int /* error */ | |
49f693fa DC |
2024 | xfs_bmap_add_extent_unwritten_real( |
2025 | struct xfs_trans *tp, | |
2bd0ea18 | 2026 | xfs_inode_t *ip, /* incore inode pointer */ |
4072e4b4 | 2027 | int whichfork, |
9788e059 | 2028 | struct xfs_iext_cursor *icur, |
49f693fa DC |
2029 | xfs_btree_cur_t **curp, /* if *curp is null, not a btree */ |
2030 | xfs_bmbt_irec_t *new, /* new data to add to file extents */ | |
49f693fa | 2031 | int *logflagsp) /* inode logging flags */ |
2bd0ea18 | 2032 | { |
49f693fa | 2033 | xfs_btree_cur_t *cur; /* btree cursor */ |
2bd0ea18 | 2034 | int error; /* error return value */ |
2bd0ea18 | 2035 | int i; /* temp state */ |
e07055b8 | 2036 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa | 2037 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
49f693fa DC |
2038 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ |
2039 | /* left is 0, right is 1, prev is 2 */ | |
2040 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 2041 | int state = xfs_bmap_fork_to_state(whichfork); |
4072e4b4 | 2042 | struct xfs_mount *mp = ip->i_mount; |
fd19685d | 2043 | struct xfs_bmbt_irec old; |
5000d01d | 2044 | |
49f693fa | 2045 | *logflagsp = 0; |
56b2de80 | 2046 | |
49f693fa | 2047 | cur = *curp; |
4072e4b4 | 2048 | ifp = XFS_IFORK_PTR(ip, whichfork); |
56b2de80 | 2049 | |
49f693fa DC |
2050 | ASSERT(!isnullstartblock(new->br_startblock)); |
2051 | ||
79896434 | 2052 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
2053 | |
2054 | #define LEFT r[0] | |
2055 | #define RIGHT r[1] | |
2056 | #define PREV r[2] | |
2057 | ||
2058 | /* | |
2059 | * Set up a bunch of variables to make the tests simpler. | |
2060 | */ | |
2bd0ea18 | 2061 | error = 0; |
9788e059 | 2062 | xfs_iext_get_extent(ifp, icur, &PREV); |
fd19685d | 2063 | ASSERT(new->br_state != PREV.br_state); |
49f693fa DC |
2064 | new_endoff = new->br_startoff + new->br_blockcount; |
2065 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
2066 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
2067 | ||
2bd0ea18 | 2068 | /* |
49f693fa DC |
2069 | * Set flags determining what part of the previous oldext allocation |
2070 | * extent is being replaced by a newext allocation. | |
2bd0ea18 | 2071 | */ |
49f693fa DC |
2072 | if (PREV.br_startoff == new->br_startoff) |
2073 | state |= BMAP_LEFT_FILLING; | |
2074 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
2075 | state |= BMAP_RIGHT_FILLING; | |
2bd0ea18 | 2076 | |
49f693fa DC |
2077 | /* |
2078 | * Check and set flags if this segment has a left neighbor. | |
2079 | * Don't set contiguous if the combined extent would be too large. | |
2080 | */ | |
9788e059 | 2081 | if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) { |
49f693fa | 2082 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
2083 | if (isnullstartblock(LEFT.br_startblock)) |
2084 | state |= BMAP_LEFT_DELAY; | |
2bd0ea18 | 2085 | } |
49f693fa DC |
2086 | |
2087 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2088 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
2089 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
fd19685d | 2090 | LEFT.br_state == new->br_state && |
49f693fa DC |
2091 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) |
2092 | state |= BMAP_LEFT_CONTIG; | |
2093 | ||
2bd0ea18 | 2094 | /* |
49f693fa DC |
2095 | * Check and set flags if this segment has a right neighbor. |
2096 | * Don't set contiguous if the combined extent would be too large. | |
2097 | * Also check for all-three-contiguous being too large. | |
2bd0ea18 | 2098 | */ |
9788e059 | 2099 | if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) { |
49f693fa | 2100 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2101 | if (isnullstartblock(RIGHT.br_startblock)) |
2102 | state |= BMAP_RIGHT_DELAY; | |
2103 | } | |
a2ceac1f | 2104 | |
49f693fa DC |
2105 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2106 | new_endoff == RIGHT.br_startoff && | |
2107 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
fd19685d | 2108 | new->br_state == RIGHT.br_state && |
49f693fa DC |
2109 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && |
2110 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2111 | BMAP_RIGHT_FILLING)) != | |
2112 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2113 | BMAP_RIGHT_FILLING) || | |
2114 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
2115 | <= MAXEXTLEN)) | |
2116 | state |= BMAP_RIGHT_CONTIG; | |
2bd0ea18 | 2117 | |
49f693fa DC |
2118 | /* |
2119 | * Switch out based on the FILLING and CONTIG state bits. | |
2120 | */ | |
2121 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2122 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
2123 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2124 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
2bd0ea18 | 2125 | /* |
49f693fa DC |
2126 | * Setting all of a previous oldext extent to newext. |
2127 | * The left and right neighbors are both contiguous with new. | |
2bd0ea18 | 2128 | */ |
fd19685d | 2129 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
49f693fa | 2130 | |
cf455a62 CH |
2131 | xfs_iext_remove(ip, icur, state); |
2132 | xfs_iext_remove(ip, icur, state); | |
9788e059 CH |
2133 | xfs_iext_prev(ifp, icur); |
2134 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
4072e4b4 DW |
2135 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2136 | XFS_IFORK_NEXTENTS(ip, whichfork) - 2); | |
49f693fa DC |
2137 | if (cur == NULL) |
2138 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2139 | else { | |
2140 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2141 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2142 | if (error) | |
49f693fa | 2143 | goto done; |
19ebedcf | 2144 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2145 | if ((error = xfs_btree_delete(cur, &i))) |
2146 | goto done; | |
19ebedcf | 2147 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2148 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2149 | goto done; | |
19ebedcf | 2150 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2151 | if ((error = xfs_btree_delete(cur, &i))) |
2152 | goto done; | |
19ebedcf | 2153 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2154 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2155 | goto done; | |
19ebedcf | 2156 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2157 | error = xfs_bmbt_update(cur, &LEFT); |
fd19685d | 2158 | if (error) |
49f693fa | 2159 | goto done; |
2bd0ea18 | 2160 | } |
2bd0ea18 NS |
2161 | break; |
2162 | ||
49f693fa | 2163 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: |
2bd0ea18 | 2164 | /* |
49f693fa DC |
2165 | * Setting all of a previous oldext extent to newext. |
2166 | * The left neighbor is contiguous, the right is not. | |
2bd0ea18 | 2167 | */ |
fd19685d | 2168 | LEFT.br_blockcount += PREV.br_blockcount; |
49f693fa | 2169 | |
cf455a62 | 2170 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2171 | xfs_iext_prev(ifp, icur); |
2172 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
4072e4b4 DW |
2173 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2174 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
49f693fa DC |
2175 | if (cur == NULL) |
2176 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2177 | else { | |
2178 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2179 | error = xfs_bmbt_lookup_eq(cur, &PREV, &i); |
2180 | if (error) | |
49f693fa | 2181 | goto done; |
19ebedcf | 2182 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2183 | if ((error = xfs_btree_delete(cur, &i))) |
2184 | goto done; | |
19ebedcf | 2185 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2186 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2187 | goto done; | |
19ebedcf | 2188 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2189 | error = xfs_bmbt_update(cur, &LEFT); |
fd19685d | 2190 | if (error) |
49f693fa | 2191 | goto done; |
2bd0ea18 | 2192 | } |
2bd0ea18 | 2193 | break; |
5000d01d | 2194 | |
49f693fa | 2195 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2bd0ea18 | 2196 | /* |
49f693fa DC |
2197 | * Setting all of a previous oldext extent to newext. |
2198 | * The right neighbor is contiguous, the left is not. | |
2bd0ea18 | 2199 | */ |
fd19685d CH |
2200 | PREV.br_blockcount += RIGHT.br_blockcount; |
2201 | PREV.br_state = new->br_state; | |
8121dd76 | 2202 | |
9788e059 | 2203 | xfs_iext_next(ifp, icur); |
cf455a62 | 2204 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2205 | xfs_iext_prev(ifp, icur); |
2206 | xfs_iext_update_extent(ip, state, icur, &PREV); | |
fd19685d | 2207 | |
4072e4b4 DW |
2208 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2209 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
49f693fa DC |
2210 | if (cur == NULL) |
2211 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2212 | else { | |
2213 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2214 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2215 | if (error) | |
49f693fa | 2216 | goto done; |
19ebedcf | 2217 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2218 | if ((error = xfs_btree_delete(cur, &i))) |
2219 | goto done; | |
19ebedcf | 2220 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2221 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2222 | goto done; | |
19ebedcf | 2223 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2224 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2225 | if (error) |
49f693fa DC |
2226 | goto done; |
2227 | } | |
2228 | break; | |
2bd0ea18 | 2229 | |
49f693fa DC |
2230 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: |
2231 | /* | |
2232 | * Setting all of a previous oldext extent to newext. | |
2233 | * Neither the left nor right neighbors are contiguous with | |
2234 | * the new one. | |
2235 | */ | |
fd19685d | 2236 | PREV.br_state = new->br_state; |
9788e059 | 2237 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2bd0ea18 | 2238 | |
49f693fa DC |
2239 | if (cur == NULL) |
2240 | rval = XFS_ILOG_DEXT; | |
2241 | else { | |
2242 | rval = 0; | |
70a93110 CH |
2243 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2244 | if (error) | |
49f693fa | 2245 | goto done; |
19ebedcf | 2246 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2247 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2248 | if (error) |
49f693fa DC |
2249 | goto done; |
2250 | } | |
2251 | break; | |
2bd0ea18 | 2252 | |
49f693fa DC |
2253 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
2254 | /* | |
2255 | * Setting the first part of a previous oldext extent to newext. | |
2256 | * The left neighbor is contiguous. | |
2257 | */ | |
fd19685d | 2258 | LEFT.br_blockcount += new->br_blockcount; |
a2ceac1f | 2259 | |
fd19685d | 2260 | old = PREV; |
fd19685d CH |
2261 | PREV.br_startoff += new->br_blockcount; |
2262 | PREV.br_startblock += new->br_blockcount; | |
2263 | PREV.br_blockcount -= new->br_blockcount; | |
b3563c19 | 2264 | |
9788e059 CH |
2265 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2266 | xfs_iext_prev(ifp, icur); | |
2267 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
b3563c19 | 2268 | |
49f693fa DC |
2269 | if (cur == NULL) |
2270 | rval = XFS_ILOG_DEXT; | |
2271 | else { | |
2272 | rval = 0; | |
70a93110 | 2273 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2274 | if (error) |
49f693fa | 2275 | goto done; |
19ebedcf | 2276 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2277 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2278 | if (error) |
49f693fa | 2279 | goto done; |
fd19685d CH |
2280 | error = xfs_btree_decrement(cur, 0, &i); |
2281 | if (error) | |
49f693fa | 2282 | goto done; |
d0e5f1ff | 2283 | error = xfs_bmbt_update(cur, &LEFT); |
49f693fa DC |
2284 | if (error) |
2285 | goto done; | |
2bd0ea18 | 2286 | } |
49f693fa | 2287 | break; |
b3563c19 | 2288 | |
49f693fa DC |
2289 | case BMAP_LEFT_FILLING: |
2290 | /* | |
2291 | * Setting the first part of a previous oldext extent to newext. | |
2292 | * The left neighbor is not contiguous. | |
2293 | */ | |
fd19685d | 2294 | old = PREV; |
fd19685d CH |
2295 | PREV.br_startoff += new->br_blockcount; |
2296 | PREV.br_startblock += new->br_blockcount; | |
2297 | PREV.br_blockcount -= new->br_blockcount; | |
2bd0ea18 | 2298 | |
9788e059 | 2299 | xfs_iext_update_extent(ip, state, icur, &PREV); |
26a75f67 | 2300 | xfs_iext_insert(ip, icur, new, state); |
4072e4b4 DW |
2301 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2302 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
49f693fa DC |
2303 | if (cur == NULL) |
2304 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2305 | else { | |
2306 | rval = XFS_ILOG_CORE; | |
70a93110 | 2307 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2308 | if (error) |
49f693fa | 2309 | goto done; |
19ebedcf | 2310 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2311 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2312 | if (error) |
49f693fa DC |
2313 | goto done; |
2314 | cur->bc_rec.b = *new; | |
2315 | if ((error = xfs_btree_insert(cur, &i))) | |
2316 | goto done; | |
19ebedcf | 2317 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2318 | } |
2319 | break; | |
56b2de80 | 2320 | |
49f693fa DC |
2321 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2322 | /* | |
2323 | * Setting the last part of a previous oldext extent to newext. | |
2324 | * The right neighbor is contiguous with the new allocation. | |
2325 | */ | |
fd19685d | 2326 | old = PREV; |
fd19685d | 2327 | PREV.br_blockcount -= new->br_blockcount; |
56b2de80 | 2328 | |
fd19685d CH |
2329 | RIGHT.br_startoff = new->br_startoff; |
2330 | RIGHT.br_startblock = new->br_startblock; | |
2331 | RIGHT.br_blockcount += new->br_blockcount; | |
8121dd76 | 2332 | |
9788e059 CH |
2333 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2334 | xfs_iext_next(ifp, icur); | |
2335 | xfs_iext_update_extent(ip, state, icur, &RIGHT); | |
56b2de80 | 2336 | |
49f693fa DC |
2337 | if (cur == NULL) |
2338 | rval = XFS_ILOG_DEXT; | |
2339 | else { | |
2340 | rval = 0; | |
70a93110 | 2341 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2342 | if (error) |
49f693fa | 2343 | goto done; |
19ebedcf | 2344 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2345 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2346 | if (error) |
49f693fa | 2347 | goto done; |
fd19685d CH |
2348 | error = xfs_btree_increment(cur, 0, &i); |
2349 | if (error) | |
49f693fa | 2350 | goto done; |
d0e5f1ff | 2351 | error = xfs_bmbt_update(cur, &RIGHT); |
fd19685d | 2352 | if (error) |
49f693fa DC |
2353 | goto done; |
2354 | } | |
2355 | break; | |
ca86e759 | 2356 | |
49f693fa DC |
2357 | case BMAP_RIGHT_FILLING: |
2358 | /* | |
2359 | * Setting the last part of a previous oldext extent to newext. | |
2360 | * The right neighbor is not contiguous. | |
2361 | */ | |
fd19685d | 2362 | old = PREV; |
fd19685d | 2363 | PREV.br_blockcount -= new->br_blockcount; |
2bd0ea18 | 2364 | |
9788e059 CH |
2365 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2366 | xfs_iext_next(ifp, icur); | |
26a75f67 | 2367 | xfs_iext_insert(ip, icur, new, state); |
2bd0ea18 | 2368 | |
4072e4b4 DW |
2369 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2370 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
49f693fa DC |
2371 | if (cur == NULL) |
2372 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2373 | else { | |
2374 | rval = XFS_ILOG_CORE; | |
70a93110 | 2375 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2376 | if (error) |
49f693fa | 2377 | goto done; |
19ebedcf | 2378 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2379 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2380 | if (error) |
49f693fa | 2381 | goto done; |
70a93110 CH |
2382 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2383 | if (error) | |
49f693fa | 2384 | goto done; |
19ebedcf | 2385 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
2386 | if ((error = xfs_btree_insert(cur, &i))) |
2387 | goto done; | |
19ebedcf | 2388 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2389 | } |
2390 | break; | |
2391 | ||
2392 | case 0: | |
2bd0ea18 | 2393 | /* |
49f693fa DC |
2394 | * Setting the middle part of a previous oldext extent to |
2395 | * newext. Contiguity is impossible here. | |
2396 | * One extent becomes three extents. | |
2bd0ea18 | 2397 | */ |
fd19685d | 2398 | old = PREV; |
fd19685d | 2399 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
49f693fa DC |
2400 | |
2401 | r[0] = *new; | |
2402 | r[1].br_startoff = new_endoff; | |
2403 | r[1].br_blockcount = | |
fd19685d | 2404 | old.br_startoff + old.br_blockcount - new_endoff; |
49f693fa | 2405 | r[1].br_startblock = new->br_startblock + new->br_blockcount; |
fd19685d | 2406 | r[1].br_state = PREV.br_state; |
49f693fa | 2407 | |
9788e059 CH |
2408 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2409 | xfs_iext_next(ifp, icur); | |
26a75f67 CH |
2410 | xfs_iext_insert(ip, icur, &r[1], state); |
2411 | xfs_iext_insert(ip, icur, &r[0], state); | |
49f693fa | 2412 | |
4072e4b4 DW |
2413 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2414 | XFS_IFORK_NEXTENTS(ip, whichfork) + 2); | |
49f693fa DC |
2415 | if (cur == NULL) |
2416 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2417 | else { | |
2418 | rval = XFS_ILOG_CORE; | |
70a93110 | 2419 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2420 | if (error) |
49f693fa | 2421 | goto done; |
19ebedcf | 2422 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 2423 | /* new right extent - oldext */ |
d0e5f1ff CH |
2424 | error = xfs_bmbt_update(cur, &r[1]); |
2425 | if (error) | |
49f693fa DC |
2426 | goto done; |
2427 | /* new left extent - oldext */ | |
2428 | cur->bc_rec.b = PREV; | |
49f693fa DC |
2429 | if ((error = xfs_btree_insert(cur, &i))) |
2430 | goto done; | |
19ebedcf | 2431 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2432 | /* |
2433 | * Reset the cursor to the position of the new extent | |
2434 | * we are about to insert as we can't trust it after | |
2435 | * the previous insert. | |
2436 | */ | |
70a93110 CH |
2437 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2438 | if (error) | |
49f693fa | 2439 | goto done; |
19ebedcf | 2440 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa | 2441 | /* new middle extent - newext */ |
49f693fa DC |
2442 | if ((error = xfs_btree_insert(cur, &i))) |
2443 | goto done; | |
19ebedcf | 2444 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
2bd0ea18 | 2445 | } |
49f693fa DC |
2446 | break; |
2447 | ||
2448 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2449 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2450 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2451 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2452 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2453 | case BMAP_LEFT_CONTIG: | |
2454 | case BMAP_RIGHT_CONTIG: | |
5000d01d | 2455 | /* |
49f693fa | 2456 | * These cases are all impossible. |
2bd0ea18 | 2457 | */ |
49f693fa DC |
2458 | ASSERT(0); |
2459 | } | |
2460 | ||
d7f80320 | 2461 | /* update reverse mappings */ |
21375e5d | 2462 | error = xfs_rmap_convert_extent(mp, tp, ip, whichfork, new); |
d7f80320 DW |
2463 | if (error) |
2464 | goto done; | |
2465 | ||
49f693fa | 2466 | /* convert to a btree if necessary */ |
4072e4b4 | 2467 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
49f693fa DC |
2468 | int tmp_logflags; /* partial log flag return val */ |
2469 | ||
2470 | ASSERT(cur == NULL); | |
f7253505 BF |
2471 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
2472 | &tmp_logflags, whichfork); | |
49f693fa DC |
2473 | *logflagsp |= tmp_logflags; |
2474 | if (error) | |
2475 | goto done; | |
ca86e759 | 2476 | } |
49f693fa DC |
2477 | |
2478 | /* clear out the allocated field, done with it now in any case. */ | |
2479 | if (cur) { | |
2480 | cur->bc_private.b.allocated = 0; | |
2481 | *curp = cur; | |
2482 | } | |
2483 | ||
4072e4b4 | 2484 | xfs_bmap_check_leaf_extents(*curp, ip, whichfork); |
2bd0ea18 | 2485 | done: |
49f693fa | 2486 | *logflagsp |= rval; |
2bd0ea18 | 2487 | return error; |
49f693fa DC |
2488 | #undef LEFT |
2489 | #undef RIGHT | |
2490 | #undef PREV | |
2bd0ea18 NS |
2491 | } |
2492 | ||
5e656dbb | 2493 | /* |
49f693fa | 2494 | * Convert a hole to a delayed allocation. |
5e656dbb | 2495 | */ |
49f693fa DC |
2496 | STATIC void |
2497 | xfs_bmap_add_extent_hole_delay( | |
2498 | xfs_inode_t *ip, /* incore inode pointer */ | |
cc66acaf | 2499 | int whichfork, |
9788e059 | 2500 | struct xfs_iext_cursor *icur, |
49f693fa | 2501 | xfs_bmbt_irec_t *new) /* new data to add to file extents */ |
2bd0ea18 | 2502 | { |
e07055b8 | 2503 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa DC |
2504 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2505 | xfs_filblks_t newlen=0; /* new indirect size */ | |
2506 | xfs_filblks_t oldlen=0; /* old indirect size */ | |
2507 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
7bd43334 | 2508 | int state = xfs_bmap_fork_to_state(whichfork); |
450b3981 | 2509 | xfs_filblks_t temp; /* temp for indirect calculations */ |
49f693fa | 2510 | |
cc66acaf | 2511 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa | 2512 | ASSERT(isnullstartblock(new->br_startblock)); |
2bd0ea18 | 2513 | |
062998e3 | 2514 | /* |
49f693fa | 2515 | * Check and set flags if this segment has a left neighbor |
062998e3 | 2516 | */ |
9788e059 | 2517 | if (xfs_iext_peek_prev_extent(ifp, icur, &left)) { |
49f693fa | 2518 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
2519 | if (isnullstartblock(left.br_startblock)) |
2520 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2521 | } |
49f693fa DC |
2522 | |
2523 | /* | |
2524 | * Check and set flags if the current (right) segment exists. | |
2525 | * If it doesn't exist, we're converting the hole at end-of-file. | |
2526 | */ | |
9788e059 | 2527 | if (xfs_iext_get_extent(ifp, icur, &right)) { |
49f693fa | 2528 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2529 | if (isnullstartblock(right.br_startblock)) |
2530 | state |= BMAP_RIGHT_DELAY; | |
2531 | } | |
2532 | ||
2533 | /* | |
2534 | * Set contiguity flags on the left and right neighbors. | |
2535 | * Don't let extents get too large, even if the pieces are contiguous. | |
2536 | */ | |
2537 | if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) && | |
2538 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2539 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2540 | state |= BMAP_LEFT_CONTIG; | |
2541 | ||
2542 | if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) && | |
2543 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2544 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2545 | (!(state & BMAP_LEFT_CONTIG) || | |
2546 | (left.br_blockcount + new->br_blockcount + | |
2547 | right.br_blockcount <= MAXEXTLEN))) | |
2548 | state |= BMAP_RIGHT_CONTIG; | |
2549 | ||
2550 | /* | |
2551 | * Switch out based on the contiguity flags. | |
2552 | */ | |
2553 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2554 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2555 | /* | |
2556 | * New allocation is contiguous with delayed allocations | |
2557 | * on the left and on the right. | |
2558 | * Merge all three into a single extent record. | |
2559 | */ | |
49f693fa DC |
2560 | temp = left.br_blockcount + new->br_blockcount + |
2561 | right.br_blockcount; | |
2562 | ||
49f693fa DC |
2563 | oldlen = startblockval(left.br_startblock) + |
2564 | startblockval(new->br_startblock) + | |
2565 | startblockval(right.br_startblock); | |
f3b62b32 BF |
2566 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2567 | oldlen); | |
450b3981 CH |
2568 | left.br_startblock = nullstartblock(newlen); |
2569 | left.br_blockcount = temp; | |
49f693fa | 2570 | |
cf455a62 | 2571 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2572 | xfs_iext_prev(ifp, icur); |
2573 | xfs_iext_update_extent(ip, state, icur, &left); | |
49f693fa DC |
2574 | break; |
2575 | ||
2576 | case BMAP_LEFT_CONTIG: | |
2577 | /* | |
2578 | * New allocation is contiguous with a delayed allocation | |
2579 | * on the left. | |
2580 | * Merge the new allocation with the left neighbor. | |
2581 | */ | |
49f693fa DC |
2582 | temp = left.br_blockcount + new->br_blockcount; |
2583 | ||
49f693fa DC |
2584 | oldlen = startblockval(left.br_startblock) + |
2585 | startblockval(new->br_startblock); | |
f3b62b32 BF |
2586 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2587 | oldlen); | |
450b3981 CH |
2588 | left.br_blockcount = temp; |
2589 | left.br_startblock = nullstartblock(newlen); | |
ee2bb3f5 | 2590 | |
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_RIGHT_CONTIG: | |
2596 | /* | |
2597 | * New allocation is contiguous with a delayed allocation | |
2598 | * on the right. | |
2599 | * Merge the new allocation with the right neighbor. | |
2600 | */ | |
49f693fa DC |
2601 | temp = new->br_blockcount + right.br_blockcount; |
2602 | oldlen = startblockval(new->br_startblock) + | |
2603 | startblockval(right.br_startblock); | |
f3b62b32 BF |
2604 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2605 | oldlen); | |
450b3981 CH |
2606 | right.br_startoff = new->br_startoff; |
2607 | right.br_startblock = nullstartblock(newlen); | |
2608 | right.br_blockcount = temp; | |
9788e059 | 2609 | xfs_iext_update_extent(ip, state, icur, &right); |
49f693fa DC |
2610 | break; |
2611 | ||
2612 | case 0: | |
2613 | /* | |
2614 | * New allocation is not contiguous with another | |
2615 | * delayed allocation. | |
2616 | * Insert a new entry. | |
2617 | */ | |
2618 | oldlen = newlen = 0; | |
26a75f67 | 2619 | xfs_iext_insert(ip, icur, new, state); |
49f693fa DC |
2620 | break; |
2621 | } | |
2622 | if (oldlen != newlen) { | |
2623 | ASSERT(oldlen > newlen); | |
19ebedcf DC |
2624 | xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen), |
2625 | false); | |
49f693fa DC |
2626 | /* |
2627 | * Nothing to do for disk quota accounting here. | |
2628 | */ | |
2bd0ea18 | 2629 | } |
2bd0ea18 NS |
2630 | } |
2631 | ||
2632 | /* | |
49f693fa | 2633 | * Convert a hole to a real allocation. |
2bd0ea18 | 2634 | */ |
49f693fa DC |
2635 | STATIC int /* error */ |
2636 | xfs_bmap_add_extent_hole_real( | |
972432f2 CH |
2637 | struct xfs_trans *tp, |
2638 | struct xfs_inode *ip, | |
2639 | int whichfork, | |
9788e059 | 2640 | struct xfs_iext_cursor *icur, |
972432f2 CH |
2641 | struct xfs_btree_cur **curp, |
2642 | struct xfs_bmbt_irec *new, | |
5f847f1e DW |
2643 | int *logflagsp, |
2644 | int flags) | |
5000d01d | 2645 | { |
972432f2 CH |
2646 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
2647 | struct xfs_mount *mp = ip->i_mount; | |
2648 | struct xfs_btree_cur *cur = *curp; | |
49f693fa DC |
2649 | int error; /* error return value */ |
2650 | int i; /* temp state */ | |
49f693fa DC |
2651 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2652 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2653 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 2654 | int state = xfs_bmap_fork_to_state(whichfork); |
3281eb91 | 2655 | struct xfs_bmbt_irec old; |
2bd0ea18 | 2656 | |
49f693fa | 2657 | ASSERT(!isnullstartblock(new->br_startblock)); |
972432f2 | 2658 | ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); |
5e656dbb | 2659 | |
79896434 | 2660 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa | 2661 | |
49f693fa DC |
2662 | /* |
2663 | * Check and set flags if this segment has a left neighbor. | |
2664 | */ | |
9788e059 | 2665 | if (xfs_iext_peek_prev_extent(ifp, icur, &left)) { |
49f693fa | 2666 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
2667 | if (isnullstartblock(left.br_startblock)) |
2668 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2669 | } |
2bd0ea18 | 2670 | |
49f693fa DC |
2671 | /* |
2672 | * Check and set flags if this segment has a current value. | |
2673 | * Not true if we're inserting into the "hole" at eof. | |
2674 | */ | |
9788e059 | 2675 | if (xfs_iext_get_extent(ifp, icur, &right)) { |
49f693fa | 2676 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2677 | if (isnullstartblock(right.br_startblock)) |
2678 | state |= BMAP_RIGHT_DELAY; | |
2bd0ea18 | 2679 | } |
2bd0ea18 | 2680 | |
49f693fa DC |
2681 | /* |
2682 | * We're inserting a real allocation between "left" and "right". | |
2683 | * Set the contiguity flags. Don't let extents get too large. | |
2684 | */ | |
2685 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2686 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2687 | left.br_startblock + left.br_blockcount == new->br_startblock && | |
2688 | left.br_state == new->br_state && | |
2689 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2690 | state |= BMAP_LEFT_CONTIG; | |
57c9fccb | 2691 | |
49f693fa DC |
2692 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2693 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2694 | new->br_startblock + new->br_blockcount == right.br_startblock && | |
2695 | new->br_state == right.br_state && | |
2696 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2697 | (!(state & BMAP_LEFT_CONTIG) || | |
2698 | left.br_blockcount + new->br_blockcount + | |
2699 | right.br_blockcount <= MAXEXTLEN)) | |
2700 | state |= BMAP_RIGHT_CONTIG; | |
ca86e759 | 2701 | |
49f693fa DC |
2702 | error = 0; |
2703 | /* | |
2704 | * Select which case we're in here, and implement it. | |
2705 | */ | |
2706 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2707 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
57c9fccb | 2708 | /* |
49f693fa DC |
2709 | * New allocation is contiguous with real allocations on the |
2710 | * left and on the right. | |
2711 | * Merge all three into a single extent record. | |
57c9fccb | 2712 | */ |
3281eb91 | 2713 | left.br_blockcount += new->br_blockcount + right.br_blockcount; |
56b2de80 | 2714 | |
cf455a62 | 2715 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2716 | xfs_iext_prev(ifp, icur); |
2717 | xfs_iext_update_extent(ip, state, icur, &left); | |
56b2de80 | 2718 | |
972432f2 CH |
2719 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2720 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
2721 | if (cur == NULL) { | |
49f693fa DC |
2722 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2723 | } else { | |
2724 | rval = XFS_ILOG_CORE; | |
70a93110 | 2725 | error = xfs_bmbt_lookup_eq(cur, &right, &i); |
49f693fa DC |
2726 | if (error) |
2727 | goto done; | |
19ebedcf | 2728 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
972432f2 | 2729 | error = xfs_btree_delete(cur, &i); |
49f693fa DC |
2730 | if (error) |
2731 | goto done; | |
19ebedcf | 2732 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
972432f2 | 2733 | error = xfs_btree_decrement(cur, 0, &i); |
49f693fa DC |
2734 | if (error) |
2735 | goto done; | |
19ebedcf | 2736 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2737 | error = xfs_bmbt_update(cur, &left); |
49f693fa DC |
2738 | if (error) |
2739 | goto done; | |
2740 | } | |
57c9fccb | 2741 | break; |
49f693fa DC |
2742 | |
2743 | case BMAP_LEFT_CONTIG: | |
2744 | /* | |
2745 | * New allocation is contiguous with a real allocation | |
2746 | * on the left. | |
2747 | * Merge the new allocation with the left neighbor. | |
2748 | */ | |
3281eb91 | 2749 | old = left; |
3281eb91 | 2750 | left.br_blockcount += new->br_blockcount; |
501dd276 | 2751 | |
9788e059 CH |
2752 | xfs_iext_prev(ifp, icur); |
2753 | xfs_iext_update_extent(ip, state, icur, &left); | |
49f693fa | 2754 | |
972432f2 | 2755 | if (cur == NULL) { |
49f693fa DC |
2756 | rval = xfs_ilog_fext(whichfork); |
2757 | } else { | |
2758 | rval = 0; | |
70a93110 | 2759 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
49f693fa DC |
2760 | if (error) |
2761 | goto done; | |
19ebedcf | 2762 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2763 | error = xfs_bmbt_update(cur, &left); |
49f693fa DC |
2764 | if (error) |
2765 | goto done; | |
2766 | } | |
57c9fccb | 2767 | break; |
49f693fa DC |
2768 | |
2769 | case BMAP_RIGHT_CONTIG: | |
2770 | /* | |
2771 | * New allocation is contiguous with a real allocation | |
2772 | * on the right. | |
2773 | * Merge the new allocation with the right neighbor. | |
2774 | */ | |
3281eb91 | 2775 | old = right; |
df926c07 | 2776 | |
3281eb91 CH |
2777 | right.br_startoff = new->br_startoff; |
2778 | right.br_startblock = new->br_startblock; | |
2779 | right.br_blockcount += new->br_blockcount; | |
9788e059 | 2780 | xfs_iext_update_extent(ip, state, icur, &right); |
49f693fa | 2781 | |
972432f2 | 2782 | if (cur == NULL) { |
49f693fa DC |
2783 | rval = xfs_ilog_fext(whichfork); |
2784 | } else { | |
2785 | rval = 0; | |
70a93110 | 2786 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
49f693fa DC |
2787 | if (error) |
2788 | goto done; | |
19ebedcf | 2789 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
d0e5f1ff | 2790 | error = xfs_bmbt_update(cur, &right); |
49f693fa DC |
2791 | if (error) |
2792 | goto done; | |
2793 | } | |
2794 | break; | |
2795 | ||
2796 | case 0: | |
2797 | /* | |
2798 | * New allocation is not contiguous with another | |
2799 | * real allocation. | |
2800 | * Insert a new entry. | |
2801 | */ | |
26a75f67 | 2802 | xfs_iext_insert(ip, icur, new, state); |
972432f2 CH |
2803 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2804 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
2805 | if (cur == NULL) { | |
49f693fa DC |
2806 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2807 | } else { | |
2808 | rval = XFS_ILOG_CORE; | |
70a93110 | 2809 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
49f693fa DC |
2810 | if (error) |
2811 | goto done; | |
19ebedcf | 2812 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
972432f2 | 2813 | error = xfs_btree_insert(cur, &i); |
49f693fa DC |
2814 | if (error) |
2815 | goto done; | |
19ebedcf | 2816 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 2817 | } |
57c9fccb | 2818 | break; |
57c9fccb | 2819 | } |
a2ceac1f | 2820 | |
5f847f1e DW |
2821 | /* add reverse mapping unless caller opted out */ |
2822 | if (!(flags & XFS_BMAPI_NORMAP)) { | |
21375e5d | 2823 | error = xfs_rmap_map_extent(tp, ip, whichfork, new); |
5f847f1e DW |
2824 | if (error) |
2825 | goto done; | |
2826 | } | |
d7f80320 | 2827 | |
49f693fa | 2828 | /* convert to a btree if necessary */ |
972432f2 | 2829 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
49f693fa | 2830 | int tmp_logflags; /* partial log flag return val */ |
3f853c7a | 2831 | |
972432f2 | 2832 | ASSERT(cur == NULL); |
f7253505 BF |
2833 | error = xfs_bmap_extents_to_btree(tp, ip, curp, 0, |
2834 | &tmp_logflags, whichfork); | |
972432f2 CH |
2835 | *logflagsp |= tmp_logflags; |
2836 | cur = *curp; | |
49f693fa DC |
2837 | if (error) |
2838 | goto done; | |
57c9fccb | 2839 | } |
a2ceac1f | 2840 | |
49f693fa | 2841 | /* clear out the allocated field, done with it now in any case. */ |
972432f2 CH |
2842 | if (cur) |
2843 | cur->bc_private.b.allocated = 0; | |
49f693fa | 2844 | |
972432f2 | 2845 | xfs_bmap_check_leaf_extents(cur, ip, whichfork); |
49f693fa | 2846 | done: |
972432f2 | 2847 | *logflagsp |= rval; |
57c9fccb NS |
2848 | return error; |
2849 | } | |
2850 | ||
2bd0ea18 | 2851 | /* |
49f693fa | 2852 | * Functions used in the extent read, allocate and remove paths |
2bd0ea18 | 2853 | */ |
2bd0ea18 | 2854 | |
5000d01d | 2855 | /* |
49f693fa | 2856 | * Adjust the size of the new extent based on di_extsize and rt extsize. |
2bd0ea18 | 2857 | */ |
613e6057 | 2858 | int |
49f693fa DC |
2859 | xfs_bmap_extsize_align( |
2860 | xfs_mount_t *mp, | |
2861 | xfs_bmbt_irec_t *gotp, /* next extent pointer */ | |
2862 | xfs_bmbt_irec_t *prevp, /* previous extent pointer */ | |
2863 | xfs_extlen_t extsz, /* align to this extent size */ | |
2864 | int rt, /* is this a realtime inode? */ | |
2865 | int eof, /* is extent at end-of-file? */ | |
2866 | int delay, /* creating delalloc extent? */ | |
2867 | int convert, /* overwriting unwritten extent? */ | |
2868 | xfs_fileoff_t *offp, /* in/out: aligned offset */ | |
2869 | xfs_extlen_t *lenp) /* in/out: aligned length */ | |
2bd0ea18 | 2870 | { |
49f693fa DC |
2871 | xfs_fileoff_t orig_off; /* original offset */ |
2872 | xfs_extlen_t orig_alen; /* original length */ | |
2873 | xfs_fileoff_t orig_end; /* original off+len */ | |
2874 | xfs_fileoff_t nexto; /* next file offset */ | |
2875 | xfs_fileoff_t prevo; /* previous file offset */ | |
2876 | xfs_fileoff_t align_off; /* temp for offset */ | |
2877 | xfs_extlen_t align_alen; /* temp for length */ | |
2878 | xfs_extlen_t temp; /* temp for calculations */ | |
2879 | ||
2880 | if (convert) | |
2881 | return 0; | |
2882 | ||
2883 | orig_off = align_off = *offp; | |
2884 | orig_alen = align_alen = *lenp; | |
2885 | orig_end = orig_off + orig_alen; | |
2bd0ea18 NS |
2886 | |
2887 | /* | |
49f693fa DC |
2888 | * If this request overlaps an existing extent, then don't |
2889 | * attempt to perform any additional alignment. | |
2bd0ea18 | 2890 | */ |
49f693fa DC |
2891 | if (!delay && !eof && |
2892 | (orig_off >= gotp->br_startoff) && | |
2893 | (orig_end <= gotp->br_startoff + gotp->br_blockcount)) { | |
2894 | return 0; | |
2bd0ea18 | 2895 | } |
57c9fccb | 2896 | |
49f693fa DC |
2897 | /* |
2898 | * If the file offset is unaligned vs. the extent size | |
2899 | * we need to align it. This will be possible unless | |
2900 | * the file was previously written with a kernel that didn't | |
2901 | * perform this alignment, or if a truncate shot us in the | |
2902 | * foot. | |
2903 | */ | |
5a595099 | 2904 | div_u64_rem(orig_off, extsz, &temp); |
49f693fa DC |
2905 | if (temp) { |
2906 | align_alen += temp; | |
2907 | align_off -= temp; | |
2908 | } | |
7cc23f0c DC |
2909 | |
2910 | /* Same adjustment for the end of the requested area. */ | |
2911 | temp = (align_alen % extsz); | |
2912 | if (temp) | |
2913 | align_alen += extsz - temp; | |
2914 | ||
49f693fa | 2915 | /* |
7cc23f0c DC |
2916 | * For large extent hint sizes, the aligned extent might be larger than |
2917 | * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls | |
2918 | * the length back under MAXEXTLEN. The outer allocation loops handle | |
2919 | * short allocation just fine, so it is safe to do this. We only want to | |
2920 | * do it when we are forced to, though, because it means more allocation | |
2921 | * operations are required. | |
49f693fa | 2922 | */ |
7cc23f0c DC |
2923 | while (align_alen > MAXEXTLEN) |
2924 | align_alen -= extsz; | |
2925 | ASSERT(align_alen <= MAXEXTLEN); | |
2926 | ||
49f693fa DC |
2927 | /* |
2928 | * If the previous block overlaps with this proposed allocation | |
2929 | * then move the start forward without adjusting the length. | |
2930 | */ | |
2931 | if (prevp->br_startoff != NULLFILEOFF) { | |
2932 | if (prevp->br_startblock == HOLESTARTBLOCK) | |
2933 | prevo = prevp->br_startoff; | |
2934 | else | |
2935 | prevo = prevp->br_startoff + prevp->br_blockcount; | |
2936 | } else | |
2937 | prevo = 0; | |
2938 | if (align_off != orig_off && align_off < prevo) | |
2939 | align_off = prevo; | |
2940 | /* | |
2941 | * If the next block overlaps with this proposed allocation | |
2942 | * then move the start back without adjusting the length, | |
2943 | * but not before offset 0. | |
2944 | * This may of course make the start overlap previous block, | |
2945 | * and if we hit the offset 0 limit then the next block | |
2946 | * can still overlap too. | |
2947 | */ | |
2948 | if (!eof && gotp->br_startoff != NULLFILEOFF) { | |
2949 | if ((delay && gotp->br_startblock == HOLESTARTBLOCK) || | |
2950 | (!delay && gotp->br_startblock == DELAYSTARTBLOCK)) | |
2951 | nexto = gotp->br_startoff + gotp->br_blockcount; | |
2952 | else | |
2953 | nexto = gotp->br_startoff; | |
2954 | } else | |
2955 | nexto = NULLFILEOFF; | |
2956 | if (!eof && | |
2957 | align_off + align_alen != orig_end && | |
2958 | align_off + align_alen > nexto) | |
2959 | align_off = nexto > align_alen ? nexto - align_alen : 0; | |
2960 | /* | |
2961 | * If we're now overlapping the next or previous extent that | |
2962 | * means we can't fit an extsz piece in this hole. Just move | |
2963 | * the start forward to the first valid spot and set | |
2964 | * the length so we hit the end. | |
2965 | */ | |
2966 | if (align_off != orig_off && align_off < prevo) | |
2967 | align_off = prevo; | |
2968 | if (align_off + align_alen != orig_end && | |
2969 | align_off + align_alen > nexto && | |
2970 | nexto != NULLFILEOFF) { | |
2971 | ASSERT(nexto > prevo); | |
2972 | align_alen = nexto - align_off; | |
57c9fccb | 2973 | } |
2bd0ea18 | 2974 | |
49f693fa DC |
2975 | /* |
2976 | * If realtime, and the result isn't a multiple of the realtime | |
2977 | * extent size we need to remove blocks until it is. | |
2978 | */ | |
2979 | if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) { | |
2bd0ea18 | 2980 | /* |
49f693fa DC |
2981 | * We're not covering the original request, or |
2982 | * we won't be able to once we fix the length. | |
2bd0ea18 | 2983 | */ |
49f693fa DC |
2984 | if (orig_off < align_off || |
2985 | orig_end > align_off + align_alen || | |
2986 | align_alen - temp < orig_alen) | |
12b53197 | 2987 | return -EINVAL; |
49f693fa DC |
2988 | /* |
2989 | * Try to fix it by moving the start up. | |
2990 | */ | |
2991 | if (align_off + temp <= orig_off) { | |
2992 | align_alen -= temp; | |
2993 | align_off += temp; | |
2bd0ea18 | 2994 | } |
49f693fa DC |
2995 | /* |
2996 | * Try to fix it by moving the end in. | |
2997 | */ | |
2998 | else if (align_off + align_alen - temp >= orig_end) | |
2999 | align_alen -= temp; | |
3000 | /* | |
3001 | * Set the start to the minimum then trim the length. | |
3002 | */ | |
3003 | else { | |
3004 | align_alen -= orig_off - align_off; | |
3005 | align_off = orig_off; | |
3006 | align_alen -= align_alen % mp->m_sb.sb_rextsize; | |
3007 | } | |
3008 | /* | |
3009 | * Result doesn't cover the request, fail it. | |
3010 | */ | |
3011 | if (orig_off < align_off || orig_end > align_off + align_alen) | |
12b53197 | 3012 | return -EINVAL; |
49f693fa DC |
3013 | } else { |
3014 | ASSERT(orig_off >= align_off); | |
7cc23f0c DC |
3015 | /* see MAXEXTLEN handling above */ |
3016 | ASSERT(orig_end <= align_off + align_alen || | |
3017 | align_alen + extsz > MAXEXTLEN); | |
2bd0ea18 | 3018 | } |
49f693fa DC |
3019 | |
3020 | #ifdef DEBUG | |
3021 | if (!eof && gotp->br_startoff != NULLFILEOFF) | |
3022 | ASSERT(align_off + align_alen <= gotp->br_startoff); | |
3023 | if (prevp->br_startoff != NULLFILEOFF) | |
3024 | ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount); | |
3025 | #endif | |
3026 | ||
3027 | *lenp = align_alen; | |
3028 | *offp = align_off; | |
2bd0ea18 NS |
3029 | return 0; |
3030 | } | |
3031 | ||
49f693fa DC |
3032 | #define XFS_ALLOC_GAP_UNITS 4 |
3033 | ||
613e6057 | 3034 | void |
49f693fa | 3035 | xfs_bmap_adjacent( |
613e6057 | 3036 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
2bd0ea18 | 3037 | { |
49f693fa DC |
3038 | xfs_fsblock_t adjust; /* adjustment to block numbers */ |
3039 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ | |
3040 | xfs_mount_t *mp; /* mount point structure */ | |
3041 | int nullfb; /* true if ap->firstblock isn't set */ | |
3042 | int rt; /* true if inode is realtime */ | |
2bd0ea18 | 3043 | |
49f693fa DC |
3044 | #define ISVALID(x,y) \ |
3045 | (rt ? \ | |
3046 | (x) < mp->m_sb.sb_rblocks : \ | |
3047 | XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \ | |
3048 | XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \ | |
3049 | XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks) | |
3050 | ||
3051 | mp = ap->ip->i_mount; | |
80dc7f2f | 3052 | nullfb = ap->tp->t_firstblock == NULLFSBLOCK; |
1fccd5c8 DC |
3053 | rt = XFS_IS_REALTIME_INODE(ap->ip) && |
3054 | xfs_alloc_is_userdata(ap->datatype); | |
80dc7f2f BF |
3055 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, |
3056 | ap->tp->t_firstblock); | |
49f693fa DC |
3057 | /* |
3058 | * If allocating at eof, and there's a previous real block, | |
3059 | * try to use its last block as our starting point. | |
3060 | */ | |
3061 | if (ap->eof && ap->prev.br_startoff != NULLFILEOFF && | |
3062 | !isnullstartblock(ap->prev.br_startblock) && | |
3063 | ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount, | |
3064 | ap->prev.br_startblock)) { | |
3065 | ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount; | |
3066 | /* | |
3067 | * Adjust for the gap between prevp and us. | |
3068 | */ | |
3069 | adjust = ap->offset - | |
3070 | (ap->prev.br_startoff + ap->prev.br_blockcount); | |
3071 | if (adjust && | |
3072 | ISVALID(ap->blkno + adjust, ap->prev.br_startblock)) | |
3073 | ap->blkno += adjust; | |
2bd0ea18 | 3074 | } |
49f693fa DC |
3075 | /* |
3076 | * If not at eof, then compare the two neighbor blocks. | |
3077 | * Figure out whether either one gives us a good starting point, | |
3078 | * and pick the better one. | |
3079 | */ | |
3080 | else if (!ap->eof) { | |
3081 | xfs_fsblock_t gotbno; /* right side block number */ | |
3082 | xfs_fsblock_t gotdiff=0; /* right side difference */ | |
3083 | xfs_fsblock_t prevbno; /* left side block number */ | |
3084 | xfs_fsblock_t prevdiff=0; /* left side difference */ | |
3085 | ||
3086 | /* | |
3087 | * If there's a previous (left) block, select a requested | |
3088 | * start block based on it. | |
3089 | */ | |
3090 | if (ap->prev.br_startoff != NULLFILEOFF && | |
3091 | !isnullstartblock(ap->prev.br_startblock) && | |
3092 | (prevbno = ap->prev.br_startblock + | |
3093 | ap->prev.br_blockcount) && | |
3094 | ISVALID(prevbno, ap->prev.br_startblock)) { | |
3095 | /* | |
3096 | * Calculate gap to end of previous block. | |
3097 | */ | |
3098 | adjust = prevdiff = ap->offset - | |
3099 | (ap->prev.br_startoff + | |
3100 | ap->prev.br_blockcount); | |
3101 | /* | |
3102 | * Figure the startblock based on the previous block's | |
3103 | * end and the gap size. | |
3104 | * Heuristic! | |
3105 | * If the gap is large relative to the piece we're | |
3106 | * allocating, or using it gives us an invalid block | |
3107 | * number, then just use the end of the previous block. | |
3108 | */ | |
3109 | if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3110 | ISVALID(prevbno + prevdiff, | |
3111 | ap->prev.br_startblock)) | |
3112 | prevbno += adjust; | |
3113 | else | |
3114 | prevdiff += adjust; | |
3115 | /* | |
3116 | * If the firstblock forbids it, can't use it, | |
3117 | * must use default. | |
3118 | */ | |
3119 | if (!rt && !nullfb && | |
3120 | XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno) | |
3121 | prevbno = NULLFSBLOCK; | |
3122 | } | |
3123 | /* | |
3124 | * No previous block or can't follow it, just default. | |
3125 | */ | |
3126 | else | |
3127 | prevbno = NULLFSBLOCK; | |
3128 | /* | |
3129 | * If there's a following (right) block, select a requested | |
3130 | * start block based on it. | |
3131 | */ | |
3132 | if (!isnullstartblock(ap->got.br_startblock)) { | |
3133 | /* | |
3134 | * Calculate gap to start of next block. | |
3135 | */ | |
3136 | adjust = gotdiff = ap->got.br_startoff - ap->offset; | |
3137 | /* | |
3138 | * Figure the startblock based on the next block's | |
3139 | * start and the gap size. | |
3140 | */ | |
3141 | gotbno = ap->got.br_startblock; | |
3142 | /* | |
3143 | * Heuristic! | |
3144 | * If the gap is large relative to the piece we're | |
3145 | * allocating, or using it gives us an invalid block | |
3146 | * number, then just use the start of the next block | |
3147 | * offset by our length. | |
3148 | */ | |
3149 | if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3150 | ISVALID(gotbno - gotdiff, gotbno)) | |
3151 | gotbno -= adjust; | |
3152 | else if (ISVALID(gotbno - ap->length, gotbno)) { | |
3153 | gotbno -= ap->length; | |
3154 | gotdiff += adjust - ap->length; | |
3155 | } else | |
3156 | gotdiff += adjust; | |
3157 | /* | |
3158 | * If the firstblock forbids it, can't use it, | |
3159 | * must use default. | |
3160 | */ | |
3161 | if (!rt && !nullfb && | |
3162 | XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno) | |
3163 | gotbno = NULLFSBLOCK; | |
3164 | } | |
3165 | /* | |
3166 | * No next block, just default. | |
3167 | */ | |
2bd0ea18 | 3168 | else |
49f693fa DC |
3169 | gotbno = NULLFSBLOCK; |
3170 | /* | |
3171 | * If both valid, pick the better one, else the only good | |
3172 | * one, else ap->blkno is already set (to 0 or the inode block). | |
3173 | */ | |
3174 | if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) | |
3175 | ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno; | |
3176 | else if (prevbno != NULLFSBLOCK) | |
3177 | ap->blkno = prevbno; | |
3178 | else if (gotbno != NULLFSBLOCK) | |
3179 | ap->blkno = gotbno; | |
a2ceac1f | 3180 | } |
49f693fa | 3181 | #undef ISVALID |
a2ceac1f DC |
3182 | } |
3183 | ||
ff105f75 DC |
3184 | static int |
3185 | xfs_bmap_longest_free_extent( | |
3186 | struct xfs_trans *tp, | |
3187 | xfs_agnumber_t ag, | |
3188 | xfs_extlen_t *blen, | |
3189 | int *notinit) | |
3190 | { | |
3191 | struct xfs_mount *mp = tp->t_mountp; | |
3192 | struct xfs_perag *pag; | |
3193 | xfs_extlen_t longest; | |
3194 | int error = 0; | |
3195 | ||
3196 | pag = xfs_perag_get(mp, ag); | |
3197 | if (!pag->pagf_init) { | |
3198 | error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK); | |
3199 | if (error) | |
3200 | goto out; | |
3201 | ||
3202 | if (!pag->pagf_init) { | |
3203 | *notinit = 1; | |
3204 | goto out; | |
3205 | } | |
3206 | } | |
3207 | ||
1421de38 | 3208 | longest = xfs_alloc_longest_free_extent(pag, |
cf8ce220 DW |
3209 | xfs_alloc_min_freelist(mp, pag), |
3210 | xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE)); | |
ff105f75 DC |
3211 | if (*blen < longest) |
3212 | *blen = longest; | |
3213 | ||
3214 | out: | |
3215 | xfs_perag_put(pag); | |
3216 | return error; | |
3217 | } | |
3218 | ||
3219 | static void | |
3220 | xfs_bmap_select_minlen( | |
3221 | struct xfs_bmalloca *ap, | |
3222 | struct xfs_alloc_arg *args, | |
3223 | xfs_extlen_t *blen, | |
3224 | int notinit) | |
3225 | { | |
3226 | if (notinit || *blen < ap->minlen) { | |
3227 | /* | |
3228 | * Since we did a BUF_TRYLOCK above, it is possible that | |
3229 | * there is space for this request. | |
3230 | */ | |
3231 | args->minlen = ap->minlen; | |
3232 | } else if (*blen < args->maxlen) { | |
3233 | /* | |
3234 | * If the best seen length is less than the request length, | |
3235 | * use the best as the minimum. | |
3236 | */ | |
3237 | args->minlen = *blen; | |
3238 | } else { | |
3239 | /* | |
3240 | * Otherwise we've seen an extent as big as maxlen, use that | |
3241 | * as the minimum. | |
3242 | */ | |
3243 | args->minlen = args->maxlen; | |
3244 | } | |
3245 | } | |
3246 | ||
a2ceac1f | 3247 | STATIC int |
49f693fa DC |
3248 | xfs_bmap_btalloc_nullfb( |
3249 | struct xfs_bmalloca *ap, | |
3250 | struct xfs_alloc_arg *args, | |
3251 | xfs_extlen_t *blen) | |
a2ceac1f | 3252 | { |
49f693fa | 3253 | struct xfs_mount *mp = ap->ip->i_mount; |
49f693fa DC |
3254 | xfs_agnumber_t ag, startag; |
3255 | int notinit = 0; | |
a2ceac1f DC |
3256 | int error; |
3257 | ||
ff105f75 | 3258 | args->type = XFS_ALLOCTYPE_START_BNO; |
49f693fa | 3259 | args->total = ap->total; |
a2ceac1f | 3260 | |
49f693fa DC |
3261 | startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno); |
3262 | if (startag == NULLAGNUMBER) | |
3263 | startag = ag = 0; | |
a2ceac1f | 3264 | |
49f693fa | 3265 | while (*blen < args->maxlen) { |
ff105f75 DC |
3266 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, |
3267 | ¬init); | |
3268 | if (error) | |
3269 | return error; | |
a2ceac1f | 3270 | |
49f693fa DC |
3271 | if (++ag == mp->m_sb.sb_agcount) |
3272 | ag = 0; | |
3273 | if (ag == startag) | |
3274 | break; | |
49f693fa | 3275 | } |
2bd0ea18 | 3276 | |
ff105f75 DC |
3277 | xfs_bmap_select_minlen(ap, args, blen, notinit); |
3278 | return 0; | |
3279 | } | |
3280 | ||
3281 | STATIC int | |
3282 | xfs_bmap_btalloc_filestreams( | |
3283 | struct xfs_bmalloca *ap, | |
3284 | struct xfs_alloc_arg *args, | |
3285 | xfs_extlen_t *blen) | |
3286 | { | |
3287 | struct xfs_mount *mp = ap->ip->i_mount; | |
3288 | xfs_agnumber_t ag; | |
3289 | int notinit = 0; | |
3290 | int error; | |
3291 | ||
3292 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
3293 | args->total = ap->total; | |
3294 | ||
3295 | ag = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3296 | if (ag == NULLAGNUMBER) | |
3297 | ag = 0; | |
3298 | ||
3299 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init); | |
3300 | if (error) | |
3301 | return error; | |
3302 | ||
3303 | if (*blen < args->maxlen) { | |
3304 | error = xfs_filestream_new_ag(ap, &ag); | |
3305 | if (error) | |
3306 | return error; | |
3307 | ||
3308 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, | |
3309 | ¬init); | |
3310 | if (error) | |
3311 | return error; | |
3312 | ||
3313 | } | |
3314 | ||
3315 | xfs_bmap_select_minlen(ap, args, blen, notinit); | |
2bd0ea18 | 3316 | |
49f693fa | 3317 | /* |
ff105f75 DC |
3318 | * Set the failure fallback case to look in the selected AG as stream |
3319 | * may have moved. | |
49f693fa | 3320 | */ |
ff105f75 | 3321 | ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0); |
49f693fa | 3322 | return 0; |
2bd0ea18 NS |
3323 | } |
3324 | ||
3cb68815 DW |
3325 | /* Update all inode and quota accounting for the allocation we just did. */ |
3326 | static void | |
3327 | xfs_bmap_btalloc_accounting( | |
3328 | struct xfs_bmalloca *ap, | |
3329 | struct xfs_alloc_arg *args) | |
3330 | { | |
d07cc724 DW |
3331 | if (ap->flags & XFS_BMAPI_COWFORK) { |
3332 | /* | |
3333 | * COW fork blocks are in-core only and thus are treated as | |
3334 | * in-core quota reservation (like delalloc blocks) even when | |
3335 | * converted to real blocks. The quota reservation is not | |
3336 | * accounted to disk until blocks are remapped to the data | |
3337 | * fork. So if these blocks were previously delalloc, we | |
3338 | * already have quota reservation and there's nothing to do | |
3339 | * yet. | |
3340 | */ | |
3341 | if (ap->wasdel) | |
3342 | return; | |
3343 | ||
3344 | /* | |
3345 | * Otherwise, we've allocated blocks in a hole. The transaction | |
3346 | * has acquired in-core quota reservation for this extent. | |
3347 | * Rather than account these as real blocks, however, we reduce | |
3348 | * the transaction quota reservation based on the allocation. | |
3349 | * This essentially transfers the transaction quota reservation | |
3350 | * to that of a delalloc extent. | |
3351 | */ | |
3352 | ap->ip->i_delayed_blks += args->len; | |
3353 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS, | |
3354 | -(long)args->len); | |
3355 | return; | |
3356 | } | |
3357 | ||
3358 | /* data/attr fork only */ | |
3359 | ap->ip->i_d.di_nblocks += args->len; | |
3cb68815 DW |
3360 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); |
3361 | if (ap->wasdel) | |
3362 | ap->ip->i_delayed_blks -= args->len; | |
3363 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, | |
3364 | ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT, | |
3365 | args->len); | |
3366 | } | |
3367 | ||
b3563c19 | 3368 | STATIC int |
49f693fa | 3369 | xfs_bmap_btalloc( |
613e6057 | 3370 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
b3563c19 | 3371 | { |
49f693fa DC |
3372 | xfs_mount_t *mp; /* mount point structure */ |
3373 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
1fccd5c8 | 3374 | xfs_extlen_t align = 0; /* minimum allocation alignment */ |
49f693fa DC |
3375 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ |
3376 | xfs_agnumber_t ag; | |
3377 | xfs_alloc_arg_t args; | |
c38464ff DW |
3378 | xfs_fileoff_t orig_offset; |
3379 | xfs_extlen_t orig_length; | |
49f693fa DC |
3380 | xfs_extlen_t blen; |
3381 | xfs_extlen_t nextminlen = 0; | |
3382 | int nullfb; /* true if ap->firstblock isn't set */ | |
3383 | int isaligned; | |
3384 | int tryagain; | |
3385 | int error; | |
ff105f75 | 3386 | int stripe_align; |
b3563c19 | 3387 | |
49f693fa | 3388 | ASSERT(ap->length); |
c38464ff DW |
3389 | orig_offset = ap->offset; |
3390 | orig_length = ap->length; | |
b3563c19 | 3391 | |
49f693fa | 3392 | mp = ap->ip->i_mount; |
ff105f75 DC |
3393 | |
3394 | /* stripe alignment for allocation is determined by mount parameters */ | |
3395 | stripe_align = 0; | |
3396 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
3397 | stripe_align = mp->m_swidth; | |
3398 | else if (mp->m_dalign) | |
3399 | stripe_align = mp->m_dalign; | |
3400 | ||
10460994 DW |
3401 | if (ap->flags & XFS_BMAPI_COWFORK) |
3402 | align = xfs_get_cowextsz_hint(ap->ip); | |
3403 | else if (xfs_alloc_is_userdata(ap->datatype)) | |
1fccd5c8 | 3404 | align = xfs_get_extsz_hint(ap->ip); |
3f5dd6c6 | 3405 | if (align) { |
49f693fa DC |
3406 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, |
3407 | align, 0, ap->eof, 0, ap->conv, | |
3408 | &ap->offset, &ap->length); | |
3409 | ASSERT(!error); | |
3410 | ASSERT(ap->length); | |
3411 | } | |
ff105f75 DC |
3412 | |
3413 | ||
80dc7f2f BF |
3414 | nullfb = ap->tp->t_firstblock == NULLFSBLOCK; |
3415 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, | |
3416 | ap->tp->t_firstblock); | |
49f693fa | 3417 | if (nullfb) { |
1fccd5c8 DC |
3418 | if (xfs_alloc_is_userdata(ap->datatype) && |
3419 | xfs_inode_is_filestream(ap->ip)) { | |
49f693fa DC |
3420 | ag = xfs_filestream_lookup_ag(ap->ip); |
3421 | ag = (ag != NULLAGNUMBER) ? ag : 0; | |
3422 | ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0); | |
3423 | } else { | |
3424 | ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino); | |
3425 | } | |
3426 | } else | |
80dc7f2f | 3427 | ap->blkno = ap->tp->t_firstblock; |
49f693fa DC |
3428 | |
3429 | xfs_bmap_adjacent(ap); | |
2bd0ea18 | 3430 | |
2bd0ea18 | 3431 | /* |
49f693fa DC |
3432 | * If allowed, use ap->blkno; otherwise must use firstblock since |
3433 | * it's in the right allocation group. | |
2bd0ea18 | 3434 | */ |
49f693fa DC |
3435 | if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno) |
3436 | ; | |
3437 | else | |
80dc7f2f | 3438 | ap->blkno = ap->tp->t_firstblock; |
2bd0ea18 | 3439 | /* |
49f693fa | 3440 | * Normal allocation, done through xfs_alloc_vextent. |
2bd0ea18 | 3441 | */ |
49f693fa DC |
3442 | tryagain = isaligned = 0; |
3443 | memset(&args, 0, sizeof(args)); | |
3444 | args.tp = ap->tp; | |
3445 | args.mp = mp; | |
3446 | args.fsbno = ap->blkno; | |
007347e3 | 3447 | args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE; |
49f693fa DC |
3448 | |
3449 | /* Trim the allocation back to the maximum an AG can fit. */ | |
d7ff12b8 | 3450 | args.maxlen = min(ap->length, mp->m_ag_max_usable); |
49f693fa DC |
3451 | blen = 0; |
3452 | if (nullfb) { | |
ff105f75 DC |
3453 | /* |
3454 | * Search for an allocation group with a single extent large | |
3455 | * enough for the request. If one isn't found, then adjust | |
3456 | * the minimum allocation size to the largest space found. | |
3457 | */ | |
1fccd5c8 DC |
3458 | if (xfs_alloc_is_userdata(ap->datatype) && |
3459 | xfs_inode_is_filestream(ap->ip)) | |
ff105f75 DC |
3460 | error = xfs_bmap_btalloc_filestreams(ap, &args, &blen); |
3461 | else | |
3462 | error = xfs_bmap_btalloc_nullfb(ap, &args, &blen); | |
a2ceac1f | 3463 | if (error) |
2bd0ea18 | 3464 | return error; |
565e96c6 | 3465 | } else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) { |
49f693fa DC |
3466 | if (xfs_inode_is_filestream(ap->ip)) |
3467 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3468 | else | |
3469 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3470 | args.total = args.minlen = ap->minlen; | |
3471 | } else { | |
3472 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3473 | args.total = ap->total; | |
3474 | args.minlen = ap->minlen; | |
3475 | } | |
3476 | /* apply extent size hints if obtained earlier */ | |
3f5dd6c6 | 3477 | if (align) { |
49f693fa | 3478 | args.prod = align; |
5a595099 DC |
3479 | div_u64_rem(ap->offset, args.prod, &args.mod); |
3480 | if (args.mod) | |
3481 | args.mod = args.prod - args.mod; | |
b2327e1a | 3482 | } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) { |
49f693fa DC |
3483 | args.prod = 1; |
3484 | args.mod = 0; | |
3485 | } else { | |
b2327e1a | 3486 | args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog; |
5a595099 DC |
3487 | div_u64_rem(ap->offset, args.prod, &args.mod); |
3488 | if (args.mod) | |
3489 | args.mod = args.prod - args.mod; | |
2bd0ea18 NS |
3490 | } |
3491 | /* | |
49f693fa DC |
3492 | * If we are not low on available data blocks, and the |
3493 | * underlying logical volume manager is a stripe, and | |
3494 | * the file offset is zero then try to allocate data | |
3495 | * blocks on stripe unit boundary. | |
3496 | * NOTE: ap->aeof is only set if the allocation length | |
3497 | * is >= the stripe unit and the allocation offset is | |
3498 | * at the end of file. | |
2bd0ea18 | 3499 | */ |
565e96c6 | 3500 | if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) { |
49f693fa | 3501 | if (!ap->offset) { |
ff105f75 | 3502 | args.alignment = stripe_align; |
49f693fa DC |
3503 | atype = args.type; |
3504 | isaligned = 1; | |
3505 | /* | |
3506 | * Adjust for alignment | |
3507 | */ | |
3508 | if (blen > args.alignment && blen <= args.maxlen) | |
3509 | args.minlen = blen - args.alignment; | |
3510 | args.minalignslop = 0; | |
3511 | } else { | |
3512 | /* | |
3513 | * First try an exact bno allocation. | |
3514 | * If it fails then do a near or start bno | |
3515 | * allocation with alignment turned on. | |
3516 | */ | |
3517 | atype = args.type; | |
3518 | tryagain = 1; | |
3519 | args.type = XFS_ALLOCTYPE_THIS_BNO; | |
3520 | args.alignment = 1; | |
3521 | /* | |
3522 | * Compute the minlen+alignment for the | |
3523 | * next case. Set slop so that the value | |
3524 | * of minlen+alignment+slop doesn't go up | |
3525 | * between the calls. | |
3526 | */ | |
ff105f75 DC |
3527 | if (blen > stripe_align && blen <= args.maxlen) |
3528 | nextminlen = blen - stripe_align; | |
49f693fa DC |
3529 | else |
3530 | nextminlen = args.minlen; | |
ff105f75 | 3531 | if (nextminlen + stripe_align > args.minlen + 1) |
49f693fa | 3532 | args.minalignslop = |
ff105f75 | 3533 | nextminlen + stripe_align - |
49f693fa DC |
3534 | args.minlen - 1; |
3535 | else | |
3536 | args.minalignslop = 0; | |
2bd0ea18 | 3537 | } |
49f693fa DC |
3538 | } else { |
3539 | args.alignment = 1; | |
3540 | args.minalignslop = 0; | |
3541 | } | |
3542 | args.minleft = ap->minleft; | |
3543 | args.wasdel = ap->wasdel; | |
cf8ce220 | 3544 | args.resv = XFS_AG_RESV_NONE; |
1fccd5c8 DC |
3545 | args.datatype = ap->datatype; |
3546 | if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) | |
9542ae13 DC |
3547 | args.ip = ap->ip; |
3548 | ||
3549 | error = xfs_alloc_vextent(&args); | |
3550 | if (error) | |
49f693fa | 3551 | return error; |
9542ae13 | 3552 | |
49f693fa DC |
3553 | if (tryagain && args.fsbno == NULLFSBLOCK) { |
3554 | /* | |
3555 | * Exact allocation failed. Now try with alignment | |
3556 | * turned on. | |
3557 | */ | |
3558 | args.type = atype; | |
3559 | args.fsbno = ap->blkno; | |
ff105f75 | 3560 | args.alignment = stripe_align; |
49f693fa DC |
3561 | args.minlen = nextminlen; |
3562 | args.minalignslop = 0; | |
3563 | isaligned = 1; | |
3564 | if ((error = xfs_alloc_vextent(&args))) | |
3565 | return error; | |
3566 | } | |
3567 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
2bd0ea18 | 3568 | /* |
49f693fa DC |
3569 | * allocation failed, so turn off alignment and |
3570 | * try again. | |
2bd0ea18 | 3571 | */ |
49f693fa DC |
3572 | args.type = atype; |
3573 | args.fsbno = ap->blkno; | |
3574 | args.alignment = 0; | |
3575 | if ((error = xfs_alloc_vextent(&args))) | |
3576 | return error; | |
3577 | } | |
3578 | if (args.fsbno == NULLFSBLOCK && nullfb && | |
3579 | args.minlen > ap->minlen) { | |
3580 | args.minlen = ap->minlen; | |
3581 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3582 | args.fsbno = ap->blkno; | |
3583 | if ((error = xfs_alloc_vextent(&args))) | |
3584 | return error; | |
3585 | } | |
3586 | if (args.fsbno == NULLFSBLOCK && nullfb) { | |
3587 | args.fsbno = 0; | |
3588 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3589 | args.total = ap->minlen; | |
49f693fa DC |
3590 | if ((error = xfs_alloc_vextent(&args))) |
3591 | return error; | |
565e96c6 | 3592 | ap->tp->t_flags |= XFS_TRANS_LOWMODE; |
49f693fa DC |
3593 | } |
3594 | if (args.fsbno != NULLFSBLOCK) { | |
2bd0ea18 | 3595 | /* |
49f693fa DC |
3596 | * check the allocation happened at the same or higher AG than |
3597 | * the first block that was allocated. | |
2bd0ea18 | 3598 | */ |
80dc7f2f BF |
3599 | ASSERT(ap->tp->t_firstblock == NULLFSBLOCK || |
3600 | XFS_FSB_TO_AGNO(mp, ap->tp->t_firstblock) <= | |
6df6a3ab | 3601 | XFS_FSB_TO_AGNO(mp, args.fsbno)); |
49f693fa DC |
3602 | |
3603 | ap->blkno = args.fsbno; | |
80dc7f2f BF |
3604 | if (ap->tp->t_firstblock == NULLFSBLOCK) |
3605 | ap->tp->t_firstblock = args.fsbno; | |
6df6a3ab | 3606 | ASSERT(nullfb || fb_agno <= args.agno); |
49f693fa | 3607 | ap->length = args.len; |
c38464ff DW |
3608 | /* |
3609 | * If the extent size hint is active, we tried to round the | |
3610 | * caller's allocation request offset down to extsz and the | |
3611 | * length up to another extsz boundary. If we found a free | |
3612 | * extent we mapped it in starting at this new offset. If the | |
3613 | * newly mapped space isn't long enough to cover any of the | |
3614 | * range of offsets that was originally requested, move the | |
3615 | * mapping up so that we can fill as much of the caller's | |
3616 | * original request as possible. Free space is apparently | |
3617 | * very fragmented so we're unlikely to be able to satisfy the | |
3618 | * hints anyway. | |
3619 | */ | |
3620 | if (ap->length <= orig_length) | |
3621 | ap->offset = orig_offset; | |
3622 | else if (ap->offset + ap->length < orig_offset + orig_length) | |
3623 | ap->offset = orig_offset + orig_length - ap->length; | |
3cb68815 | 3624 | xfs_bmap_btalloc_accounting(ap, &args); |
49f693fa DC |
3625 | } else { |
3626 | ap->blkno = NULLFSBLOCK; | |
3627 | ap->length = 0; | |
2bd0ea18 | 3628 | } |
2bd0ea18 | 3629 | return 0; |
56b2de80 DC |
3630 | } |
3631 | ||
3632 | /* | |
49f693fa DC |
3633 | * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file. |
3634 | * It figures out where to ask the underlying allocator to put the new extent. | |
56b2de80 | 3635 | */ |
49f693fa DC |
3636 | STATIC int |
3637 | xfs_bmap_alloc( | |
613e6057 | 3638 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
56b2de80 | 3639 | { |
1fccd5c8 DC |
3640 | if (XFS_IS_REALTIME_INODE(ap->ip) && |
3641 | xfs_alloc_is_userdata(ap->datatype)) | |
49f693fa DC |
3642 | return xfs_bmap_rtalloc(ap); |
3643 | return xfs_bmap_btalloc(ap); | |
56b2de80 | 3644 | } |
56b2de80 | 3645 | |
b3fd8db7 DW |
3646 | /* Trim extent to fit a logical block range. */ |
3647 | void | |
3648 | xfs_trim_extent( | |
3649 | struct xfs_bmbt_irec *irec, | |
3650 | xfs_fileoff_t bno, | |
3651 | xfs_filblks_t len) | |
3652 | { | |
3653 | xfs_fileoff_t distance; | |
3654 | xfs_fileoff_t end = bno + len; | |
3655 | ||
3656 | if (irec->br_startoff + irec->br_blockcount <= bno || | |
3657 | irec->br_startoff >= end) { | |
3658 | irec->br_blockcount = 0; | |
3659 | return; | |
3660 | } | |
3661 | ||
3662 | if (irec->br_startoff < bno) { | |
3663 | distance = bno - irec->br_startoff; | |
3664 | if (isnullstartblock(irec->br_startblock)) | |
3665 | irec->br_startblock = DELAYSTARTBLOCK; | |
3666 | if (irec->br_startblock != DELAYSTARTBLOCK && | |
3667 | irec->br_startblock != HOLESTARTBLOCK) | |
3668 | irec->br_startblock += distance; | |
3669 | irec->br_startoff += distance; | |
3670 | irec->br_blockcount -= distance; | |
3671 | } | |
3672 | ||
3673 | if (end < irec->br_startoff + irec->br_blockcount) { | |
3674 | distance = irec->br_startoff + irec->br_blockcount - end; | |
3675 | irec->br_blockcount -= distance; | |
3676 | } | |
3677 | } | |
3678 | ||
ca75eb20 BF |
3679 | /* trim extent to within eof */ |
3680 | void | |
3681 | xfs_trim_extent_eof( | |
3682 | struct xfs_bmbt_irec *irec, | |
3683 | struct xfs_inode *ip) | |
3684 | ||
3685 | { | |
3686 | xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount, | |
3687 | i_size_read(VFS_I(ip)))); | |
3688 | } | |
3689 | ||
2bd0ea18 | 3690 | /* |
a2ceac1f | 3691 | * Trim the returned map to the required bounds |
2bd0ea18 | 3692 | */ |
a2ceac1f DC |
3693 | STATIC void |
3694 | xfs_bmapi_trim_map( | |
3695 | struct xfs_bmbt_irec *mval, | |
3696 | struct xfs_bmbt_irec *got, | |
3697 | xfs_fileoff_t *bno, | |
3698 | xfs_filblks_t len, | |
3699 | xfs_fileoff_t obno, | |
3700 | xfs_fileoff_t end, | |
3701 | int n, | |
3702 | int flags) | |
2bd0ea18 | 3703 | { |
a2ceac1f DC |
3704 | if ((flags & XFS_BMAPI_ENTIRE) || |
3705 | got->br_startoff + got->br_blockcount <= obno) { | |
3706 | *mval = *got; | |
3707 | if (isnullstartblock(got->br_startblock)) | |
3708 | mval->br_startblock = DELAYSTARTBLOCK; | |
3709 | return; | |
63be04eb | 3710 | } |
a2ceac1f DC |
3711 | |
3712 | if (obno > *bno) | |
3713 | *bno = obno; | |
3714 | ASSERT((*bno >= obno) || (n == 0)); | |
3715 | ASSERT(*bno < end); | |
3716 | mval->br_startoff = *bno; | |
3717 | if (isnullstartblock(got->br_startblock)) | |
3718 | mval->br_startblock = DELAYSTARTBLOCK; | |
2bd0ea18 | 3719 | else |
a2ceac1f DC |
3720 | mval->br_startblock = got->br_startblock + |
3721 | (*bno - got->br_startoff); | |
2bd0ea18 | 3722 | /* |
a2ceac1f DC |
3723 | * Return the minimum of what we got and what we asked for for |
3724 | * the length. We can use the len variable here because it is | |
3725 | * modified below and we could have been there before coming | |
3726 | * here if the first part of the allocation didn't overlap what | |
3727 | * was asked for. | |
2bd0ea18 | 3728 | */ |
a2ceac1f DC |
3729 | mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno, |
3730 | got->br_blockcount - (*bno - got->br_startoff)); | |
3731 | mval->br_state = got->br_state; | |
3732 | ASSERT(mval->br_blockcount <= len); | |
3733 | return; | |
3734 | } | |
56b2de80 | 3735 | |
a2ceac1f DC |
3736 | /* |
3737 | * Update and validate the extent map to return | |
3738 | */ | |
3739 | STATIC void | |
3740 | xfs_bmapi_update_map( | |
3741 | struct xfs_bmbt_irec **map, | |
3742 | xfs_fileoff_t *bno, | |
3743 | xfs_filblks_t *len, | |
3744 | xfs_fileoff_t obno, | |
3745 | xfs_fileoff_t end, | |
3746 | int *n, | |
3747 | int flags) | |
3748 | { | |
3749 | xfs_bmbt_irec_t *mval = *map; | |
3750 | ||
3751 | ASSERT((flags & XFS_BMAPI_ENTIRE) || | |
3752 | ((mval->br_startoff + mval->br_blockcount) <= end)); | |
3753 | ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) || | |
3754 | (mval->br_startoff < obno)); | |
3755 | ||
3756 | *bno = mval->br_startoff + mval->br_blockcount; | |
3757 | *len = end - *bno; | |
3758 | if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) { | |
3759 | /* update previous map with new information */ | |
3760 | ASSERT(mval->br_startblock == mval[-1].br_startblock); | |
3761 | ASSERT(mval->br_blockcount > mval[-1].br_blockcount); | |
3762 | ASSERT(mval->br_state == mval[-1].br_state); | |
3763 | mval[-1].br_blockcount = mval->br_blockcount; | |
3764 | mval[-1].br_state = mval->br_state; | |
3765 | } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK && | |
3766 | mval[-1].br_startblock != DELAYSTARTBLOCK && | |
3767 | mval[-1].br_startblock != HOLESTARTBLOCK && | |
3768 | mval->br_startblock == mval[-1].br_startblock + | |
3769 | mval[-1].br_blockcount && | |
8197bceb | 3770 | mval[-1].br_state == mval->br_state) { |
a2ceac1f DC |
3771 | ASSERT(mval->br_startoff == |
3772 | mval[-1].br_startoff + mval[-1].br_blockcount); | |
3773 | mval[-1].br_blockcount += mval->br_blockcount; | |
3774 | } else if (*n > 0 && | |
3775 | mval->br_startblock == DELAYSTARTBLOCK && | |
3776 | mval[-1].br_startblock == DELAYSTARTBLOCK && | |
3777 | mval->br_startoff == | |
3778 | mval[-1].br_startoff + mval[-1].br_blockcount) { | |
3779 | mval[-1].br_blockcount += mval->br_blockcount; | |
3780 | mval[-1].br_state = mval->br_state; | |
3781 | } else if (!((*n == 0) && | |
3782 | ((mval->br_startoff + mval->br_blockcount) <= | |
3783 | obno))) { | |
3784 | mval++; | |
3785 | (*n)++; | |
3786 | } | |
3787 | *map = mval; | |
3788 | } | |
399ab595 | 3789 | |
a2ceac1f DC |
3790 | /* |
3791 | * Map file blocks to filesystem blocks without allocation. | |
3792 | */ | |
3793 | int | |
3794 | xfs_bmapi_read( | |
3795 | struct xfs_inode *ip, | |
3796 | xfs_fileoff_t bno, | |
3797 | xfs_filblks_t len, | |
3798 | struct xfs_bmbt_irec *mval, | |
3799 | int *nmap, | |
3800 | int flags) | |
3801 | { | |
3802 | struct xfs_mount *mp = ip->i_mount; | |
3803 | struct xfs_ifork *ifp; | |
3804 | struct xfs_bmbt_irec got; | |
a2ceac1f DC |
3805 | xfs_fileoff_t obno; |
3806 | xfs_fileoff_t end; | |
9788e059 | 3807 | struct xfs_iext_cursor icur; |
a2ceac1f | 3808 | int error; |
c2d73ed3 | 3809 | bool eof = false; |
a2ceac1f | 3810 | int n = 0; |
cb8a004a | 3811 | int whichfork = xfs_bmapi_whichfork(flags); |
399ab595 | 3812 | |
a2ceac1f DC |
3813 | ASSERT(*nmap >= 1); |
3814 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE| | |
8197bceb | 3815 | XFS_BMAPI_COWFORK))); |
ff105f75 | 3816 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)); |
062998e3 | 3817 | |
a2ceac1f DC |
3818 | if (unlikely(XFS_TEST_ERROR( |
3819 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
3820 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 3821 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a2ceac1f | 3822 | XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp); |
12b53197 | 3823 | return -EFSCORRUPTED; |
a2ceac1f | 3824 | } |
062998e3 | 3825 | |
a2ceac1f | 3826 | if (XFS_FORCED_SHUTDOWN(mp)) |
12b53197 | 3827 | return -EIO; |
399ab595 | 3828 | |
79896434 | 3829 | XFS_STATS_INC(mp, xs_blk_mapr); |
a2ceac1f DC |
3830 | |
3831 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
3832 | ||
cb8a004a DW |
3833 | /* No CoW fork? Return a hole. */ |
3834 | if (whichfork == XFS_COW_FORK && !ifp) { | |
3835 | mval->br_startoff = bno; | |
3836 | mval->br_startblock = HOLESTARTBLOCK; | |
3837 | mval->br_blockcount = len; | |
3838 | mval->br_state = XFS_EXT_NORM; | |
3839 | *nmap = 1; | |
3840 | return 0; | |
3841 | } | |
3842 | ||
a2ceac1f DC |
3843 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
3844 | error = xfs_iread_extents(NULL, ip, whichfork); | |
3845 | if (error) | |
3846 | return error; | |
3847 | } | |
3848 | ||
9788e059 | 3849 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) |
c2d73ed3 | 3850 | eof = true; |
a2ceac1f DC |
3851 | end = bno + len; |
3852 | obno = bno; | |
3853 | ||
3854 | while (bno < end && n < *nmap) { | |
3855 | /* Reading past eof, act as though there's a hole up to end. */ | |
3856 | if (eof) | |
3857 | got.br_startoff = end; | |
3858 | if (got.br_startoff > bno) { | |
3859 | /* Reading in a hole. */ | |
2bd0ea18 NS |
3860 | mval->br_startoff = bno; |
3861 | mval->br_startblock = HOLESTARTBLOCK; | |
3862 | mval->br_blockcount = | |
3863 | XFS_FILBLKS_MIN(len, got.br_startoff - bno); | |
3864 | mval->br_state = XFS_EXT_NORM; | |
3865 | bno += mval->br_blockcount; | |
3866 | len -= mval->br_blockcount; | |
3867 | mval++; | |
3868 | n++; | |
3869 | continue; | |
3870 | } | |
a2ceac1f DC |
3871 | |
3872 | /* set up the extent map to return. */ | |
3873 | xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags); | |
3874 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
3875 | ||
3876 | /* If we're done, stop now. */ | |
3877 | if (bno >= end || n >= *nmap) | |
3878 | break; | |
3879 | ||
3880 | /* Else go on to the next record. */ | |
9788e059 | 3881 | if (!xfs_iext_next_extent(ifp, &icur, &got)) |
c2d73ed3 | 3882 | eof = true; |
a2ceac1f DC |
3883 | } |
3884 | *nmap = n; | |
3885 | return 0; | |
3886 | } | |
3887 | ||
7a868ee8 BF |
3888 | /* |
3889 | * Add a delayed allocation extent to an inode. Blocks are reserved from the | |
3890 | * global pool and the extent inserted into the inode in-core extent tree. | |
3891 | * | |
3892 | * On entry, got refers to the first extent beyond the offset of the extent to | |
3893 | * allocate or eof is specified if no such extent exists. On return, got refers | |
3894 | * to the extent record that was inserted to the inode fork. | |
3895 | * | |
3896 | * Note that the allocated extent may have been merged with contiguous extents | |
3897 | * during insertion into the inode fork. Thus, got does not reflect the current | |
3898 | * state of the inode fork on return. If necessary, the caller can use lastx to | |
3899 | * look up the updated record in the inode fork. | |
3900 | */ | |
4488e421 | 3901 | int |
a2ceac1f DC |
3902 | xfs_bmapi_reserve_delalloc( |
3903 | struct xfs_inode *ip, | |
cc66acaf | 3904 | int whichfork, |
f7b1a8b1 | 3905 | xfs_fileoff_t off, |
a2ceac1f | 3906 | xfs_filblks_t len, |
f7b1a8b1 | 3907 | xfs_filblks_t prealloc, |
a2ceac1f | 3908 | struct xfs_bmbt_irec *got, |
9788e059 | 3909 | struct xfs_iext_cursor *icur, |
a2ceac1f DC |
3910 | int eof) |
3911 | { | |
3912 | struct xfs_mount *mp = ip->i_mount; | |
cc66acaf | 3913 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
a2ceac1f DC |
3914 | xfs_extlen_t alen; |
3915 | xfs_extlen_t indlen; | |
a2ceac1f | 3916 | int error; |
f7b1a8b1 | 3917 | xfs_fileoff_t aoff = off; |
a2ceac1f | 3918 | |
f7b1a8b1 BF |
3919 | /* |
3920 | * Cap the alloc length. Keep track of prealloc so we know whether to | |
3921 | * tag the inode before we return. | |
3922 | */ | |
3923 | alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN); | |
a2ceac1f DC |
3924 | if (!eof) |
3925 | alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff); | |
f7b1a8b1 BF |
3926 | if (prealloc && alen >= len) |
3927 | prealloc = alen - len; | |
a2ceac1f DC |
3928 | |
3929 | /* Figure out the extent size, adjust alen */ | |
885ba5ce | 3930 | if (whichfork == XFS_COW_FORK) { |
d41d2303 | 3931 | struct xfs_bmbt_irec prev; |
885ba5ce | 3932 | xfs_extlen_t extsz = xfs_get_cowextsz_hint(ip); |
d41d2303 | 3933 | |
9788e059 | 3934 | if (!xfs_iext_peek_prev_extent(ifp, icur, &prev)) |
d41d2303 CH |
3935 | prev.br_startoff = NULLFILEOFF; |
3936 | ||
885ba5ce | 3937 | error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof, |
a2ceac1f DC |
3938 | 1, 0, &aoff, &alen); |
3939 | ASSERT(!error); | |
3940 | } | |
3941 | ||
a2ceac1f DC |
3942 | /* |
3943 | * Make a transaction-less quota reservation for delayed allocation | |
3944 | * blocks. This number gets adjusted later. We return if we haven't | |
3945 | * allocated blocks already inside this loop. | |
3946 | */ | |
3947 | error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0, | |
885ba5ce | 3948 | XFS_QMOPT_RES_REGBLKS); |
a2ceac1f DC |
3949 | if (error) |
3950 | return error; | |
3951 | ||
3952 | /* | |
3953 | * Split changing sb for alen and indlen since they could be coming | |
3954 | * from different places. | |
3955 | */ | |
3956 | indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen); | |
3957 | ASSERT(indlen > 0); | |
3958 | ||
885ba5ce | 3959 | error = xfs_mod_fdblocks(mp, -((int64_t)alen), false); |
a2ceac1f DC |
3960 | if (error) |
3961 | goto out_unreserve_quota; | |
3962 | ||
19ebedcf | 3963 | error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false); |
a2ceac1f DC |
3964 | if (error) |
3965 | goto out_unreserve_blocks; | |
3966 | ||
3967 | ||
3968 | ip->i_delayed_blks += alen; | |
3969 | ||
3970 | got->br_startoff = aoff; | |
3971 | got->br_startblock = nullstartblock(indlen); | |
3972 | got->br_blockcount = alen; | |
3973 | got->br_state = XFS_EXT_NORM; | |
a2ceac1f | 3974 | |
9788e059 | 3975 | xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got); |
a2ceac1f | 3976 | |
f7b1a8b1 BF |
3977 | /* |
3978 | * Tag the inode if blocks were preallocated. Note that COW fork | |
3979 | * preallocation can occur at the start or end of the extent, even when | |
3980 | * prealloc == 0, so we must also check the aligned offset and length. | |
3981 | */ | |
3982 | if (whichfork == XFS_DATA_FORK && prealloc) | |
3983 | xfs_inode_set_eofblocks_tag(ip); | |
3984 | if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len)) | |
3985 | xfs_inode_set_cowblocks_tag(ip); | |
3986 | ||
a2ceac1f DC |
3987 | return 0; |
3988 | ||
3989 | out_unreserve_blocks: | |
885ba5ce | 3990 | xfs_mod_fdblocks(mp, alen, false); |
a2ceac1f DC |
3991 | out_unreserve_quota: |
3992 | if (XFS_IS_QUOTA_ON(mp)) | |
885ba5ce SH |
3993 | xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, |
3994 | XFS_QMOPT_RES_REGBLKS); | |
a2ceac1f DC |
3995 | return error; |
3996 | } | |
3997 | ||
ff105f75 DC |
3998 | static int |
3999 | xfs_bmapi_allocate( | |
a2ceac1f DC |
4000 | struct xfs_bmalloca *bma) |
4001 | { | |
4002 | struct xfs_mount *mp = bma->ip->i_mount; | |
1277a5e0 | 4003 | int whichfork = xfs_bmapi_whichfork(bma->flags); |
a2ceac1f DC |
4004 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4005 | int tmp_logflags = 0; | |
4006 | int error; | |
4007 | ||
4008 | ASSERT(bma->length > 0); | |
4009 | ||
4010 | /* | |
4011 | * For the wasdelay case, we could also just allocate the stuff asked | |
4012 | * for in this bmap call but that wouldn't be as good. | |
4013 | */ | |
4014 | if (bma->wasdel) { | |
4015 | bma->length = (xfs_extlen_t)bma->got.br_blockcount; | |
4016 | bma->offset = bma->got.br_startoff; | |
9788e059 | 4017 | xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev); |
a2ceac1f DC |
4018 | } else { |
4019 | bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN); | |
4020 | if (!bma->eof) | |
4021 | bma->length = XFS_FILBLKS_MIN(bma->length, | |
4022 | bma->got.br_startoff - bma->offset); | |
4023 | } | |
4024 | ||
4025 | /* | |
1fccd5c8 DC |
4026 | * Set the data type being allocated. For the data fork, the first data |
4027 | * in the file is treated differently to all other allocations. For the | |
4028 | * attribute fork, we only need to ensure the allocated range is not on | |
4029 | * the busy list. | |
a2ceac1f DC |
4030 | */ |
4031 | if (!(bma->flags & XFS_BMAPI_METADATA)) { | |
1fccd5c8 DC |
4032 | bma->datatype = XFS_ALLOC_NOBUSY; |
4033 | if (whichfork == XFS_DATA_FORK) { | |
4034 | if (bma->offset == 0) | |
4035 | bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA; | |
4036 | else | |
4037 | bma->datatype |= XFS_ALLOC_USERDATA; | |
4038 | } | |
9542ae13 | 4039 | if (bma->flags & XFS_BMAPI_ZERO) |
1fccd5c8 | 4040 | bma->datatype |= XFS_ALLOC_USERDATA_ZERO; |
a2ceac1f DC |
4041 | } |
4042 | ||
4043 | bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1; | |
4044 | ||
4045 | /* | |
4046 | * Only want to do the alignment at the eof if it is userdata and | |
4047 | * allocation length is larger than a stripe unit. | |
4048 | */ | |
4049 | if (mp->m_dalign && bma->length >= mp->m_dalign && | |
4050 | !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) { | |
4051 | error = xfs_bmap_isaeof(bma, whichfork); | |
4052 | if (error) | |
4053 | return error; | |
4054 | } | |
4055 | ||
a2ceac1f DC |
4056 | error = xfs_bmap_alloc(bma); |
4057 | if (error) | |
4058 | return error; | |
4059 | ||
a2ceac1f DC |
4060 | if (bma->blkno == NULLFSBLOCK) |
4061 | return 0; | |
f44a348b | 4062 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) |
a2ceac1f | 4063 | bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork); |
a2ceac1f DC |
4064 | /* |
4065 | * Bump the number of extents we've allocated | |
4066 | * in this call. | |
4067 | */ | |
4068 | bma->nallocs++; | |
4069 | ||
4070 | if (bma->cur) | |
4071 | bma->cur->bc_private.b.flags = | |
4072 | bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; | |
4073 | ||
4074 | bma->got.br_startoff = bma->offset; | |
4075 | bma->got.br_startblock = bma->blkno; | |
4076 | bma->got.br_blockcount = bma->length; | |
4077 | bma->got.br_state = XFS_EXT_NORM; | |
4078 | ||
4079 | /* | |
4072e4b4 DW |
4080 | * In the data fork, a wasdelay extent has been initialized, so |
4081 | * shouldn't be flagged as unwritten. | |
4082 | * | |
4083 | * For the cow fork, however, we convert delalloc reservations | |
4084 | * (extents allocated for speculative preallocation) to | |
4085 | * allocated unwritten extents, and only convert the unwritten | |
4086 | * extents to real extents when we're about to write the data. | |
a2ceac1f | 4087 | */ |
4072e4b4 | 4088 | if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) && |
b7fdeb4d | 4089 | (bma->flags & XFS_BMAPI_PREALLOC)) |
a2ceac1f DC |
4090 | bma->got.br_state = XFS_EXT_UNWRITTEN; |
4091 | ||
4092 | if (bma->wasdel) | |
1277a5e0 | 4093 | error = xfs_bmap_add_extent_delay_real(bma, whichfork); |
a2ceac1f | 4094 | else |
972432f2 | 4095 | error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip, |
9788e059 | 4096 | whichfork, &bma->icur, &bma->cur, &bma->got, |
64e8b4a7 | 4097 | &bma->logflags, bma->flags); |
a2ceac1f DC |
4098 | |
4099 | bma->logflags |= tmp_logflags; | |
4100 | if (error) | |
4101 | return error; | |
4102 | ||
4103 | /* | |
4104 | * Update our extent pointer, given that xfs_bmap_add_extent_delay_real | |
4105 | * or xfs_bmap_add_extent_hole_real might have merged it into one of | |
4106 | * the neighbouring ones. | |
4107 | */ | |
9788e059 | 4108 | xfs_iext_get_extent(ifp, &bma->icur, &bma->got); |
a2ceac1f DC |
4109 | |
4110 | ASSERT(bma->got.br_startoff <= bma->offset); | |
4111 | ASSERT(bma->got.br_startoff + bma->got.br_blockcount >= | |
4112 | bma->offset + bma->length); | |
4113 | ASSERT(bma->got.br_state == XFS_EXT_NORM || | |
4114 | bma->got.br_state == XFS_EXT_UNWRITTEN); | |
4115 | return 0; | |
4116 | } | |
4117 | ||
a2ceac1f DC |
4118 | STATIC int |
4119 | xfs_bmapi_convert_unwritten( | |
4120 | struct xfs_bmalloca *bma, | |
4121 | struct xfs_bmbt_irec *mval, | |
4122 | xfs_filblks_t len, | |
4123 | int flags) | |
4124 | { | |
cb8a004a | 4125 | int whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4126 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4127 | int tmp_logflags = 0; | |
4128 | int error; | |
4129 | ||
4130 | /* check if we need to do unwritten->real conversion */ | |
4131 | if (mval->br_state == XFS_EXT_UNWRITTEN && | |
4132 | (flags & XFS_BMAPI_PREALLOC)) | |
4133 | return 0; | |
4134 | ||
4135 | /* check if we need to do real->unwritten conversion */ | |
4136 | if (mval->br_state == XFS_EXT_NORM && | |
4137 | (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) != | |
4138 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) | |
4139 | return 0; | |
4140 | ||
4141 | /* | |
4142 | * Modify (by adding) the state flag, if writing. | |
4143 | */ | |
4144 | ASSERT(mval->br_blockcount <= len); | |
4145 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4146 | bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp, | |
4147 | bma->ip, whichfork); | |
a2ceac1f DC |
4148 | } |
4149 | mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN) | |
4150 | ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
4151 | ||
9542ae13 DC |
4152 | /* |
4153 | * Before insertion into the bmbt, zero the range being converted | |
4154 | * if required. | |
4155 | */ | |
4156 | if (flags & XFS_BMAPI_ZERO) { | |
4157 | error = xfs_zero_extent(bma->ip, mval->br_startblock, | |
4158 | mval->br_blockcount); | |
4159 | if (error) | |
4160 | return error; | |
4161 | } | |
4162 | ||
4072e4b4 | 4163 | error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork, |
64e8b4a7 | 4164 | &bma->icur, &bma->cur, mval, &tmp_logflags); |
23fc058d BF |
4165 | /* |
4166 | * Log the inode core unconditionally in the unwritten extent conversion | |
4167 | * path because the conversion might not have done so (e.g., if the | |
4168 | * extent count hasn't changed). We need to make sure the inode is dirty | |
4169 | * in the transaction for the sake of fsync(), even if nothing has | |
4170 | * changed, because fsync() will not force the log for this transaction | |
4171 | * unless it sees the inode pinned. | |
4072e4b4 DW |
4172 | * |
4173 | * Note: If we're only converting cow fork extents, there aren't | |
4174 | * any on-disk updates to make, so we don't need to log anything. | |
23fc058d | 4175 | */ |
4072e4b4 DW |
4176 | if (whichfork != XFS_COW_FORK) |
4177 | bma->logflags |= tmp_logflags | XFS_ILOG_CORE; | |
a2ceac1f DC |
4178 | if (error) |
4179 | return error; | |
4180 | ||
4181 | /* | |
4182 | * Update our extent pointer, given that | |
4183 | * xfs_bmap_add_extent_unwritten_real might have merged it into one | |
4184 | * of the neighbouring ones. | |
4185 | */ | |
9788e059 | 4186 | xfs_iext_get_extent(ifp, &bma->icur, &bma->got); |
a2ceac1f DC |
4187 | |
4188 | /* | |
4189 | * We may have combined previously unwritten space with written space, | |
4190 | * so generate another request. | |
4191 | */ | |
4192 | if (mval->br_blockcount < len) | |
12b53197 | 4193 | return -EAGAIN; |
a2ceac1f DC |
4194 | return 0; |
4195 | } | |
4196 | ||
4197 | /* | |
4198 | * Map file blocks to filesystem blocks, and allocate blocks or convert the | |
4199 | * extent state if necessary. Details behaviour is controlled by the flags | |
4200 | * parameter. Only allocates blocks from a single allocation group, to avoid | |
4201 | * locking problems. | |
a2ceac1f DC |
4202 | */ |
4203 | int | |
4204 | xfs_bmapi_write( | |
4205 | struct xfs_trans *tp, /* transaction pointer */ | |
4206 | struct xfs_inode *ip, /* incore inode */ | |
4207 | xfs_fileoff_t bno, /* starting file offs. mapped */ | |
4208 | xfs_filblks_t len, /* length to map in file */ | |
4209 | int flags, /* XFS_BMAPI_... */ | |
a2ceac1f DC |
4210 | xfs_extlen_t total, /* total blocks needed */ |
4211 | struct xfs_bmbt_irec *mval, /* output: map values */ | |
cbdfb3ab | 4212 | int *nmap) /* i/o: mval size/count */ |
a2ceac1f DC |
4213 | { |
4214 | struct xfs_mount *mp = ip->i_mount; | |
4215 | struct xfs_ifork *ifp; | |
389b3b07 | 4216 | struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */ |
a2ceac1f | 4217 | xfs_fileoff_t end; /* end of mapped file region */ |
82411945 | 4218 | bool eof = false; /* after the end of extents */ |
a2ceac1f DC |
4219 | int error; /* error return */ |
4220 | int n; /* current extent index */ | |
4221 | xfs_fileoff_t obno; /* old block number (offset) */ | |
4222 | int whichfork; /* data or attr fork */ | |
a2ceac1f DC |
4223 | |
4224 | #ifdef DEBUG | |
4225 | xfs_fileoff_t orig_bno; /* original block number value */ | |
4226 | int orig_flags; /* original flags arg value */ | |
4227 | xfs_filblks_t orig_len; /* original value of len arg */ | |
4228 | struct xfs_bmbt_irec *orig_mval; /* original value of mval */ | |
4229 | int orig_nmap; /* original value of *nmap */ | |
4230 | ||
4231 | orig_bno = bno; | |
4232 | orig_len = len; | |
4233 | orig_flags = flags; | |
4234 | orig_mval = mval; | |
4235 | orig_nmap = *nmap; | |
4236 | #endif | |
1277a5e0 | 4237 | whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4238 | |
4239 | ASSERT(*nmap >= 1); | |
4240 | ASSERT(*nmap <= XFS_BMAP_MAX_NMAP); | |
4072e4b4 DW |
4241 | ASSERT(tp != NULL || |
4242 | (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) == | |
4243 | (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)); | |
a2ceac1f | 4244 | ASSERT(len > 0); |
3f17ed4b | 4245 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL); |
ff105f75 | 4246 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b44aa85 | 4247 | ASSERT(!(flags & XFS_BMAPI_REMAP)); |
a2ceac1f | 4248 | |
9542ae13 DC |
4249 | /* zeroing is for currently only for data extents, not metadata */ |
4250 | ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) != | |
4251 | (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)); | |
4252 | /* | |
4253 | * we can allocate unwritten extents or pre-zero allocated blocks, | |
4254 | * but it makes no sense to do both at once. This would result in | |
4255 | * zeroing the unwritten extent twice, but it still being an | |
4256 | * unwritten extent.... | |
4257 | */ | |
4258 | ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) != | |
4259 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)); | |
4260 | ||
a2ceac1f DC |
4261 | if (unlikely(XFS_TEST_ERROR( |
4262 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
3f17ed4b | 4263 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), |
e2a190dd | 4264 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a2ceac1f | 4265 | XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp); |
12b53197 | 4266 | return -EFSCORRUPTED; |
a2ceac1f DC |
4267 | } |
4268 | ||
4269 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 4270 | return -EIO; |
a2ceac1f DC |
4271 | |
4272 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
4273 | ||
79896434 | 4274 | XFS_STATS_INC(mp, xs_blk_mapw); |
a2ceac1f | 4275 | |
b39f9fcb | 4276 | if (!tp || tp->t_firstblock == NULLFSBLOCK) { |
a2ceac1f DC |
4277 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE) |
4278 | bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1; | |
4279 | else | |
4280 | bma.minleft = 1; | |
4281 | } else { | |
4282 | bma.minleft = 0; | |
4283 | } | |
4284 | ||
4285 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
4286 | error = xfs_iread_extents(tp, ip, whichfork); | |
4287 | if (error) | |
4288 | goto error0; | |
4289 | } | |
4290 | ||
a2ceac1f DC |
4291 | n = 0; |
4292 | end = bno + len; | |
4293 | obno = bno; | |
4294 | ||
9788e059 | 4295 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got)) |
82411945 | 4296 | eof = true; |
9788e059 | 4297 | if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev)) |
82411945 | 4298 | bma.prev.br_startoff = NULLFILEOFF; |
a2ceac1f DC |
4299 | bma.tp = tp; |
4300 | bma.ip = ip; | |
4301 | bma.total = total; | |
1fccd5c8 | 4302 | bma.datatype = 0; |
a2ceac1f DC |
4303 | |
4304 | while (bno < end && n < *nmap) { | |
7075a23f CH |
4305 | bool need_alloc = false, wasdelay = false; |
4306 | ||
faaad1df | 4307 | /* in hole or beyond EOF? */ |
7075a23f | 4308 | if (eof || bma.got.br_startoff > bno) { |
faaad1df DW |
4309 | /* |
4310 | * CoW fork conversions should /never/ hit EOF or | |
4311 | * holes. There should always be something for us | |
4312 | * to work on. | |
4313 | */ | |
4314 | ASSERT(!((flags & XFS_BMAPI_CONVERT) && | |
4315 | (flags & XFS_BMAPI_COWFORK))); | |
4316 | ||
7075a23f CH |
4317 | if (flags & XFS_BMAPI_DELALLOC) { |
4318 | /* | |
4319 | * For the COW fork we can reasonably get a | |
4320 | * request for converting an extent that races | |
4321 | * with other threads already having converted | |
4322 | * part of it, as there converting COW to | |
4323 | * regular blocks is not protected using the | |
4324 | * IOLOCK. | |
4325 | */ | |
4326 | ASSERT(flags & XFS_BMAPI_COWFORK); | |
4327 | if (!(flags & XFS_BMAPI_COWFORK)) { | |
4328 | error = -EIO; | |
4329 | goto error0; | |
4330 | } | |
4331 | ||
4332 | if (eof || bno >= end) | |
4333 | break; | |
4334 | } else { | |
4335 | need_alloc = true; | |
4336 | } | |
0b44aa85 CH |
4337 | } else if (isnullstartblock(bma.got.br_startblock)) { |
4338 | wasdelay = true; | |
7075a23f | 4339 | } |
34621a47 | 4340 | |
2bd0ea18 | 4341 | /* |
a2ceac1f DC |
4342 | * First, deal with the hole before the allocated space |
4343 | * that we found, if any. | |
2bd0ea18 | 4344 | */ |
8e87c884 CH |
4345 | if ((need_alloc || wasdelay) && |
4346 | !(flags & XFS_BMAPI_CONVERT_ONLY)) { | |
a2ceac1f DC |
4347 | bma.eof = eof; |
4348 | bma.conv = !!(flags & XFS_BMAPI_CONVERT); | |
4349 | bma.wasdel = wasdelay; | |
4350 | bma.offset = bno; | |
4351 | bma.flags = flags; | |
4352 | ||
2bd0ea18 | 4353 | /* |
a2ceac1f DC |
4354 | * There's a 32/64 bit type mismatch between the |
4355 | * allocation length request (which can be 64 bits in | |
4356 | * length) and the bma length request, which is | |
4357 | * xfs_extlen_t and therefore 32 bits. Hence we have to | |
4358 | * check for 32-bit overflows and handle them here. | |
2bd0ea18 | 4359 | */ |
a2ceac1f DC |
4360 | if (len > (xfs_filblks_t)MAXEXTLEN) |
4361 | bma.length = MAXEXTLEN; | |
4362 | else | |
4363 | bma.length = len; | |
4364 | ||
4365 | ASSERT(len > 0); | |
4366 | ASSERT(bma.length > 0); | |
4367 | error = xfs_bmapi_allocate(&bma); | |
2bd0ea18 NS |
4368 | if (error) |
4369 | goto error0; | |
a2ceac1f DC |
4370 | if (bma.blkno == NULLFSBLOCK) |
4371 | break; | |
10e65503 DW |
4372 | |
4373 | /* | |
4374 | * If this is a CoW allocation, record the data in | |
4375 | * the refcount btree for orphan recovery. | |
4376 | */ | |
4377 | if (whichfork == XFS_COW_FORK) { | |
21375e5d BF |
4378 | error = xfs_refcount_alloc_cow_extent(tp, |
4379 | bma.blkno, bma.length); | |
10e65503 DW |
4380 | if (error) |
4381 | goto error0; | |
4382 | } | |
2bd0ea18 NS |
4383 | } |
4384 | ||
a2ceac1f DC |
4385 | /* Deal with the allocated space we found. */ |
4386 | xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno, | |
4387 | end, n, flags); | |
4388 | ||
4389 | /* Execute unwritten extent conversion if necessary */ | |
4390 | error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags); | |
12b53197 | 4391 | if (error == -EAGAIN) |
a2ceac1f DC |
4392 | continue; |
4393 | if (error) | |
4394 | goto error0; | |
4395 | ||
4396 | /* update the extent map to return */ | |
4397 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4398 | ||
2bd0ea18 NS |
4399 | /* |
4400 | * If we're done, stop now. Stop when we've allocated | |
4401 | * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise | |
4402 | * the transaction may get too big. | |
4403 | */ | |
a2ceac1f | 4404 | if (bno >= end || n >= *nmap || bma.nallocs >= *nmap) |
2bd0ea18 | 4405 | break; |
a2ceac1f DC |
4406 | |
4407 | /* Else go on to the next record. */ | |
4408 | bma.prev = bma.got; | |
9788e059 | 4409 | if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got)) |
82411945 | 4410 | eof = true; |
2bd0ea18 | 4411 | } |
2bd0ea18 | 4412 | *nmap = n; |
a2ceac1f | 4413 | |
2bd0ea18 NS |
4414 | /* |
4415 | * Transform from btree to extents, give it cur. | |
4416 | */ | |
a2ceac1f DC |
4417 | if (xfs_bmap_wants_extents(ip, whichfork)) { |
4418 | int tmp_logflags = 0; | |
4419 | ||
4420 | ASSERT(bma.cur); | |
4421 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, | |
4ca431fc | 4422 | &tmp_logflags, whichfork); |
a2ceac1f | 4423 | bma.logflags |= tmp_logflags; |
2bd0ea18 NS |
4424 | if (error) |
4425 | goto error0; | |
4426 | } | |
a2ceac1f | 4427 | |
2bd0ea18 | 4428 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE || |
a2ceac1f DC |
4429 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
4430 | XFS_IFORK_MAXEXT(ip, whichfork)); | |
2bd0ea18 | 4431 | error = 0; |
2bd0ea18 NS |
4432 | error0: |
4433 | /* | |
4434 | * Log everything. Do this after conversion, there's no point in | |
5e656dbb | 4435 | * logging the extent records if we've converted to btree format. |
2bd0ea18 | 4436 | */ |
a2ceac1f | 4437 | if ((bma.logflags & xfs_ilog_fext(whichfork)) && |
2bd0ea18 | 4438 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) |
a2ceac1f DC |
4439 | bma.logflags &= ~xfs_ilog_fext(whichfork); |
4440 | else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) && | |
2bd0ea18 | 4441 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) |
a2ceac1f | 4442 | bma.logflags &= ~xfs_ilog_fbroot(whichfork); |
2bd0ea18 NS |
4443 | /* |
4444 | * Log whatever the flags say, even if error. Otherwise we might miss | |
4445 | * detecting a case where the data is changed, there's an error, | |
4446 | * and it's not logged so we don't shutdown when we should. | |
4447 | */ | |
a2ceac1f DC |
4448 | if (bma.logflags) |
4449 | xfs_trans_log_inode(tp, ip, bma.logflags); | |
4450 | ||
4451 | if (bma.cur) { | |
660265b7 | 4452 | xfs_btree_del_cursor(bma.cur, error); |
2bd0ea18 NS |
4453 | } |
4454 | if (!error) | |
4455 | xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval, | |
4456 | orig_nmap, *nmap); | |
4457 | return error; | |
4458 | } | |
4459 | ||
26d6a481 | 4460 | int |
0b44aa85 CH |
4461 | xfs_bmapi_remap( |
4462 | struct xfs_trans *tp, | |
4463 | struct xfs_inode *ip, | |
4464 | xfs_fileoff_t bno, | |
4465 | xfs_filblks_t len, | |
4466 | xfs_fsblock_t startblock, | |
26d6a481 | 4467 | int flags) |
0b44aa85 CH |
4468 | { |
4469 | struct xfs_mount *mp = ip->i_mount; | |
26d6a481 | 4470 | struct xfs_ifork *ifp; |
0b44aa85 | 4471 | struct xfs_btree_cur *cur = NULL; |
0b44aa85 | 4472 | struct xfs_bmbt_irec got; |
9788e059 | 4473 | struct xfs_iext_cursor icur; |
26d6a481 | 4474 | int whichfork = xfs_bmapi_whichfork(flags); |
0b44aa85 CH |
4475 | int logflags = 0, error; |
4476 | ||
26d6a481 | 4477 | ifp = XFS_IFORK_PTR(ip, whichfork); |
0b44aa85 CH |
4478 | ASSERT(len > 0); |
4479 | ASSERT(len <= (xfs_filblks_t)MAXEXTLEN); | |
4480 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
36cfb334 DW |
4481 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC | |
4482 | XFS_BMAPI_NORMAP))); | |
4483 | ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) != | |
4484 | (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)); | |
0b44aa85 CH |
4485 | |
4486 | if (unlikely(XFS_TEST_ERROR( | |
26d6a481 DW |
4487 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && |
4488 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 4489 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
0b44aa85 CH |
4490 | XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp); |
4491 | return -EFSCORRUPTED; | |
4492 | } | |
4493 | ||
4494 | if (XFS_FORCED_SHUTDOWN(mp)) | |
4495 | return -EIO; | |
4496 | ||
4497 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
26d6a481 | 4498 | error = xfs_iread_extents(tp, ip, whichfork); |
0b44aa85 CH |
4499 | if (error) |
4500 | return error; | |
4501 | } | |
4502 | ||
9788e059 | 4503 | if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) { |
0b44aa85 CH |
4504 | /* make sure we only reflink into a hole. */ |
4505 | ASSERT(got.br_startoff > bno); | |
4506 | ASSERT(got.br_startoff - bno >= len); | |
4507 | } | |
4508 | ||
05422db6 CH |
4509 | ip->i_d.di_nblocks += len; |
4510 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
0b44aa85 CH |
4511 | |
4512 | if (ifp->if_flags & XFS_IFBROOT) { | |
26d6a481 | 4513 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
0b44aa85 CH |
4514 | cur->bc_private.b.flags = 0; |
4515 | } | |
4516 | ||
4517 | got.br_startoff = bno; | |
4518 | got.br_startblock = startblock; | |
4519 | got.br_blockcount = len; | |
36cfb334 DW |
4520 | if (flags & XFS_BMAPI_PREALLOC) |
4521 | got.br_state = XFS_EXT_UNWRITTEN; | |
4522 | else | |
4523 | got.br_state = XFS_EXT_NORM; | |
0b44aa85 | 4524 | |
26d6a481 | 4525 | error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur, |
64e8b4a7 | 4526 | &cur, &got, &logflags, flags); |
0b44aa85 CH |
4527 | if (error) |
4528 | goto error0; | |
4529 | ||
26d6a481 | 4530 | if (xfs_bmap_wants_extents(ip, whichfork)) { |
0b44aa85 CH |
4531 | int tmp_logflags = 0; |
4532 | ||
4533 | error = xfs_bmap_btree_to_extents(tp, ip, cur, | |
26d6a481 | 4534 | &tmp_logflags, whichfork); |
0b44aa85 CH |
4535 | logflags |= tmp_logflags; |
4536 | } | |
4537 | ||
4538 | error0: | |
4539 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) | |
4540 | logflags &= ~XFS_ILOG_DEXT; | |
4541 | else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) | |
4542 | logflags &= ~XFS_ILOG_DBROOT; | |
4543 | ||
4544 | if (logflags) | |
4545 | xfs_trans_log_inode(tp, ip, logflags); | |
660265b7 DW |
4546 | if (cur) |
4547 | xfs_btree_del_cursor(cur, error); | |
0b44aa85 CH |
4548 | return error; |
4549 | } | |
4550 | ||
01d1b786 BF |
4551 | /* |
4552 | * When a delalloc extent is split (e.g., due to a hole punch), the original | |
4553 | * indlen reservation must be shared across the two new extents that are left | |
4554 | * behind. | |
4555 | * | |
4556 | * Given the original reservation and the worst case indlen for the two new | |
4557 | * extents (as calculated by xfs_bmap_worst_indlen()), split the original | |
731ccdf9 BF |
4558 | * reservation fairly across the two new extents. If necessary, steal available |
4559 | * blocks from a deleted extent to make up a reservation deficiency (e.g., if | |
4560 | * ores == 1). The number of stolen blocks is returned. The availability and | |
4561 | * subsequent accounting of stolen blocks is the responsibility of the caller. | |
01d1b786 | 4562 | */ |
731ccdf9 | 4563 | static xfs_filblks_t |
01d1b786 BF |
4564 | xfs_bmap_split_indlen( |
4565 | xfs_filblks_t ores, /* original res. */ | |
4566 | xfs_filblks_t *indlen1, /* ext1 worst indlen */ | |
731ccdf9 BF |
4567 | xfs_filblks_t *indlen2, /* ext2 worst indlen */ |
4568 | xfs_filblks_t avail) /* stealable blocks */ | |
01d1b786 BF |
4569 | { |
4570 | xfs_filblks_t len1 = *indlen1; | |
4571 | xfs_filblks_t len2 = *indlen2; | |
4572 | xfs_filblks_t nres = len1 + len2; /* new total res. */ | |
731ccdf9 | 4573 | xfs_filblks_t stolen = 0; |
775762e4 | 4574 | xfs_filblks_t resfactor; |
731ccdf9 BF |
4575 | |
4576 | /* | |
4577 | * Steal as many blocks as we can to try and satisfy the worst case | |
4578 | * indlen for both new extents. | |
4579 | */ | |
775762e4 BF |
4580 | if (ores < nres && avail) |
4581 | stolen = XFS_FILBLKS_MIN(nres - ores, avail); | |
4582 | ores += stolen; | |
4583 | ||
4584 | /* nothing else to do if we've satisfied the new reservation */ | |
4585 | if (ores >= nres) | |
4586 | return stolen; | |
4587 | ||
4588 | /* | |
4589 | * We can't meet the total required reservation for the two extents. | |
4590 | * Calculate the percent of the overall shortage between both extents | |
4591 | * and apply this percentage to each of the requested indlen values. | |
4592 | * This distributes the shortage fairly and reduces the chances that one | |
4593 | * of the two extents is left with nothing when extents are repeatedly | |
4594 | * split. | |
4595 | */ | |
4596 | resfactor = (ores * 100); | |
4597 | do_div(resfactor, nres); | |
4598 | len1 *= resfactor; | |
4599 | do_div(len1, 100); | |
4600 | len2 *= resfactor; | |
4601 | do_div(len2, 100); | |
4602 | ASSERT(len1 + len2 <= ores); | |
4603 | ASSERT(len1 < *indlen1 && len2 < *indlen2); | |
01d1b786 BF |
4604 | |
4605 | /* | |
775762e4 BF |
4606 | * Hand out the remainder to each extent. If one of the two reservations |
4607 | * is zero, we want to make sure that one gets a block first. The loop | |
4608 | * below starts with len1, so hand len2 a block right off the bat if it | |
4609 | * is zero. | |
01d1b786 | 4610 | */ |
775762e4 BF |
4611 | ores -= (len1 + len2); |
4612 | ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores); | |
4613 | if (ores && !len2 && *indlen2) { | |
4614 | len2++; | |
4615 | ores--; | |
4616 | } | |
4617 | while (ores) { | |
4618 | if (len1 < *indlen1) { | |
4619 | len1++; | |
4620 | ores--; | |
01d1b786 | 4621 | } |
775762e4 | 4622 | if (!ores) |
01d1b786 | 4623 | break; |
775762e4 BF |
4624 | if (len2 < *indlen2) { |
4625 | len2++; | |
4626 | ores--; | |
01d1b786 BF |
4627 | } |
4628 | } | |
4629 | ||
4630 | *indlen1 = len1; | |
4631 | *indlen2 = len2; | |
731ccdf9 BF |
4632 | |
4633 | return stolen; | |
01d1b786 BF |
4634 | } |
4635 | ||
ece930fa CH |
4636 | int |
4637 | xfs_bmap_del_extent_delay( | |
4638 | struct xfs_inode *ip, | |
4639 | int whichfork, | |
9788e059 | 4640 | struct xfs_iext_cursor *icur, |
ece930fa CH |
4641 | struct xfs_bmbt_irec *got, |
4642 | struct xfs_bmbt_irec *del) | |
4643 | { | |
4644 | struct xfs_mount *mp = ip->i_mount; | |
4645 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
4646 | struct xfs_bmbt_irec new; | |
4647 | int64_t da_old, da_new, da_diff = 0; | |
4648 | xfs_fileoff_t del_endoff, got_endoff; | |
4649 | xfs_filblks_t got_indlen, new_indlen, stolen; | |
7bd43334 CH |
4650 | int state = xfs_bmap_fork_to_state(whichfork); |
4651 | int error = 0; | |
ece930fa CH |
4652 | bool isrt; |
4653 | ||
4654 | XFS_STATS_INC(mp, xs_del_exlist); | |
4655 | ||
4656 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); | |
4657 | del_endoff = del->br_startoff + del->br_blockcount; | |
4658 | got_endoff = got->br_startoff + got->br_blockcount; | |
4659 | da_old = startblockval(got->br_startblock); | |
4660 | da_new = 0; | |
4661 | ||
ece930fa CH |
4662 | ASSERT(del->br_blockcount > 0); |
4663 | ASSERT(got->br_startoff <= del->br_startoff); | |
4664 | ASSERT(got_endoff >= del_endoff); | |
4665 | ||
4666 | if (isrt) { | |
f1e0bd8d | 4667 | uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount); |
ece930fa CH |
4668 | |
4669 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
4670 | xfs_mod_frextents(mp, rtexts); | |
4671 | } | |
4672 | ||
4673 | /* | |
4674 | * Update the inode delalloc counter now and wait to update the | |
4675 | * sb counters as we might have to borrow some blocks for the | |
4676 | * indirect block accounting. | |
4677 | */ | |
4678 | error = xfs_trans_reserve_quota_nblks(NULL, ip, | |
4679 | -((long)del->br_blockcount), 0, | |
4680 | isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
6921638c DW |
4681 | if (error) |
4682 | return error; | |
ece930fa CH |
4683 | ip->i_delayed_blks -= del->br_blockcount; |
4684 | ||
ece930fa | 4685 | if (got->br_startoff == del->br_startoff) |
d0a03e5a | 4686 | state |= BMAP_LEFT_FILLING; |
ece930fa | 4687 | if (got_endoff == del_endoff) |
d0a03e5a | 4688 | state |= BMAP_RIGHT_FILLING; |
ece930fa | 4689 | |
d0a03e5a CH |
4690 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4691 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
ece930fa CH |
4692 | /* |
4693 | * Matches the whole extent. Delete the entry. | |
4694 | */ | |
cf455a62 | 4695 | xfs_iext_remove(ip, icur, state); |
9788e059 | 4696 | xfs_iext_prev(ifp, icur); |
ece930fa | 4697 | break; |
d0a03e5a | 4698 | case BMAP_LEFT_FILLING: |
ece930fa CH |
4699 | /* |
4700 | * Deleting the first part of the extent. | |
4701 | */ | |
ece930fa CH |
4702 | got->br_startoff = del_endoff; |
4703 | got->br_blockcount -= del->br_blockcount; | |
4704 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4705 | got->br_blockcount), da_old); | |
4706 | got->br_startblock = nullstartblock((int)da_new); | |
9788e059 | 4707 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa | 4708 | break; |
d0a03e5a | 4709 | case BMAP_RIGHT_FILLING: |
ece930fa CH |
4710 | /* |
4711 | * Deleting the last part of the extent. | |
4712 | */ | |
ece930fa CH |
4713 | got->br_blockcount = got->br_blockcount - del->br_blockcount; |
4714 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4715 | got->br_blockcount), da_old); | |
4716 | got->br_startblock = nullstartblock((int)da_new); | |
9788e059 | 4717 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa CH |
4718 | break; |
4719 | case 0: | |
4720 | /* | |
4721 | * Deleting the middle of the extent. | |
4722 | * | |
4723 | * Distribute the original indlen reservation across the two new | |
4724 | * extents. Steal blocks from the deleted extent if necessary. | |
4725 | * Stealing blocks simply fudges the fdblocks accounting below. | |
4726 | * Warn if either of the new indlen reservations is zero as this | |
4727 | * can lead to delalloc problems. | |
4728 | */ | |
ece930fa CH |
4729 | got->br_blockcount = del->br_startoff - got->br_startoff; |
4730 | got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount); | |
4731 | ||
4732 | new.br_blockcount = got_endoff - del_endoff; | |
4733 | new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
4734 | ||
4735 | WARN_ON_ONCE(!got_indlen || !new_indlen); | |
4736 | stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen, | |
4737 | del->br_blockcount); | |
4738 | ||
4739 | got->br_startblock = nullstartblock((int)got_indlen); | |
ece930fa CH |
4740 | |
4741 | new.br_startoff = del_endoff; | |
4742 | new.br_state = got->br_state; | |
4743 | new.br_startblock = nullstartblock((int)new_indlen); | |
4744 | ||
9788e059 CH |
4745 | xfs_iext_update_extent(ip, state, icur, got); |
4746 | xfs_iext_next(ifp, icur); | |
26a75f67 | 4747 | xfs_iext_insert(ip, icur, &new, state); |
ece930fa CH |
4748 | |
4749 | da_new = got_indlen + new_indlen - stolen; | |
4750 | del->br_blockcount -= stolen; | |
4751 | break; | |
4752 | } | |
4753 | ||
4754 | ASSERT(da_old >= da_new); | |
4755 | da_diff = da_old - da_new; | |
4756 | if (!isrt) | |
4757 | da_diff += del->br_blockcount; | |
4758 | if (da_diff) | |
4759 | xfs_mod_fdblocks(mp, da_diff, false); | |
4760 | return error; | |
4761 | } | |
4762 | ||
4763 | void | |
4764 | xfs_bmap_del_extent_cow( | |
4765 | struct xfs_inode *ip, | |
9788e059 | 4766 | struct xfs_iext_cursor *icur, |
ece930fa CH |
4767 | struct xfs_bmbt_irec *got, |
4768 | struct xfs_bmbt_irec *del) | |
4769 | { | |
4770 | struct xfs_mount *mp = ip->i_mount; | |
4771 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
4772 | struct xfs_bmbt_irec new; | |
4773 | xfs_fileoff_t del_endoff, got_endoff; | |
4774 | int state = BMAP_COWFORK; | |
4775 | ||
4776 | XFS_STATS_INC(mp, xs_del_exlist); | |
4777 | ||
4778 | del_endoff = del->br_startoff + del->br_blockcount; | |
4779 | got_endoff = got->br_startoff + got->br_blockcount; | |
4780 | ||
ece930fa CH |
4781 | ASSERT(del->br_blockcount > 0); |
4782 | ASSERT(got->br_startoff <= del->br_startoff); | |
4783 | ASSERT(got_endoff >= del_endoff); | |
4784 | ASSERT(!isnullstartblock(got->br_startblock)); | |
4785 | ||
4786 | if (got->br_startoff == del->br_startoff) | |
d0a03e5a | 4787 | state |= BMAP_LEFT_FILLING; |
ece930fa | 4788 | if (got_endoff == del_endoff) |
d0a03e5a | 4789 | state |= BMAP_RIGHT_FILLING; |
ece930fa | 4790 | |
d0a03e5a CH |
4791 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4792 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
ece930fa CH |
4793 | /* |
4794 | * Matches the whole extent. Delete the entry. | |
4795 | */ | |
cf455a62 | 4796 | xfs_iext_remove(ip, icur, state); |
9788e059 | 4797 | xfs_iext_prev(ifp, icur); |
ece930fa | 4798 | break; |
d0a03e5a | 4799 | case BMAP_LEFT_FILLING: |
ece930fa CH |
4800 | /* |
4801 | * Deleting the first part of the extent. | |
4802 | */ | |
ece930fa CH |
4803 | got->br_startoff = del_endoff; |
4804 | got->br_blockcount -= del->br_blockcount; | |
4805 | got->br_startblock = del->br_startblock + del->br_blockcount; | |
9788e059 | 4806 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa | 4807 | break; |
d0a03e5a | 4808 | case BMAP_RIGHT_FILLING: |
ece930fa CH |
4809 | /* |
4810 | * Deleting the last part of the extent. | |
4811 | */ | |
ece930fa | 4812 | got->br_blockcount -= del->br_blockcount; |
9788e059 | 4813 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa CH |
4814 | break; |
4815 | case 0: | |
4816 | /* | |
4817 | * Deleting the middle of the extent. | |
4818 | */ | |
ece930fa | 4819 | got->br_blockcount = del->br_startoff - got->br_startoff; |
ece930fa CH |
4820 | |
4821 | new.br_startoff = del_endoff; | |
4822 | new.br_blockcount = got_endoff - del_endoff; | |
4823 | new.br_state = got->br_state; | |
4824 | new.br_startblock = del->br_startblock + del->br_blockcount; | |
4825 | ||
9788e059 CH |
4826 | xfs_iext_update_extent(ip, state, icur, got); |
4827 | xfs_iext_next(ifp, icur); | |
26a75f67 | 4828 | xfs_iext_insert(ip, icur, &new, state); |
ece930fa CH |
4829 | break; |
4830 | } | |
d07cc724 | 4831 | ip->i_delayed_blks -= del->br_blockcount; |
ece930fa CH |
4832 | } |
4833 | ||
2bd0ea18 | 4834 | /* |
49f693fa | 4835 | * Called by xfs_bmapi to update file extent records and the btree |
ad68fd19 | 4836 | * after removing space. |
2bd0ea18 | 4837 | */ |
49f693fa | 4838 | STATIC int /* error */ |
ad68fd19 | 4839 | xfs_bmap_del_extent_real( |
49f693fa DC |
4840 | xfs_inode_t *ip, /* incore inode pointer */ |
4841 | xfs_trans_t *tp, /* current transaction pointer */ | |
9788e059 | 4842 | struct xfs_iext_cursor *icur, |
49f693fa DC |
4843 | xfs_btree_cur_t *cur, /* if null, not a btree */ |
4844 | xfs_bmbt_irec_t *del, /* data to remove from extents */ | |
4845 | int *logflagsp, /* inode logging flags */ | |
36b16da8 DW |
4846 | int whichfork, /* data or attr fork */ |
4847 | int bflags) /* bmapi flags */ | |
2bd0ea18 | 4848 | { |
49f693fa DC |
4849 | xfs_fsblock_t del_endblock=0; /* first block past del */ |
4850 | xfs_fileoff_t del_endoff; /* first offset past del */ | |
49f693fa | 4851 | int do_fx; /* free extent at end of routine */ |
49f693fa | 4852 | int error; /* error return value */ |
bd92a38b | 4853 | int flags = 0;/* inode logging flags */ |
70bf7533 | 4854 | struct xfs_bmbt_irec got; /* current extent entry */ |
49f693fa DC |
4855 | xfs_fileoff_t got_endoff; /* first offset past got */ |
4856 | int i; /* temp state */ | |
e07055b8 | 4857 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa DC |
4858 | xfs_mount_t *mp; /* mount structure */ |
4859 | xfs_filblks_t nblks; /* quota/sb block count */ | |
4860 | xfs_bmbt_irec_t new; /* new record to be inserted */ | |
4861 | /* REFERENCED */ | |
4862 | uint qfield; /* quota field to update */ | |
7bd43334 | 4863 | int state = xfs_bmap_fork_to_state(whichfork); |
70bf7533 | 4864 | struct xfs_bmbt_irec old; |
a2ceac1f | 4865 | |
79896434 BD |
4866 | mp = ip->i_mount; |
4867 | XFS_STATS_INC(mp, xs_del_exlist); | |
a2ceac1f | 4868 | |
49f693fa | 4869 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa | 4870 | ASSERT(del->br_blockcount > 0); |
9788e059 | 4871 | xfs_iext_get_extent(ifp, icur, &got); |
49f693fa DC |
4872 | ASSERT(got.br_startoff <= del->br_startoff); |
4873 | del_endoff = del->br_startoff + del->br_blockcount; | |
4874 | got_endoff = got.br_startoff + got.br_blockcount; | |
4875 | ASSERT(got_endoff >= del_endoff); | |
ad68fd19 | 4876 | ASSERT(!isnullstartblock(got.br_startblock)); |
49f693fa DC |
4877 | qfield = 0; |
4878 | error = 0; | |
ad68fd19 | 4879 | |
bd92a38b CH |
4880 | /* |
4881 | * If it's the case where the directory code is running with no block | |
4882 | * reservation, and the deleted block is in the middle of its extent, | |
4883 | * and the resulting insert of an extent would cause transformation to | |
4884 | * btree format, then reject it. The calling code will then swap blocks | |
4885 | * around instead. We have to do this now, rather than waiting for the | |
4886 | * conversion to btree format, since the transaction will be dirty then. | |
4887 | */ | |
4888 | if (tp->t_blk_res == 0 && | |
4889 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
4890 | XFS_IFORK_NEXTENTS(ip, whichfork) >= | |
4891 | XFS_IFORK_MAXEXT(ip, whichfork) && | |
4892 | del->br_startoff > got.br_startoff && del_endoff < got_endoff) | |
4893 | return -ENOSPC; | |
4894 | ||
4895 | flags = XFS_ILOG_CORE; | |
ad68fd19 CH |
4896 | if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) { |
4897 | xfs_fsblock_t bno; | |
4898 | xfs_filblks_t len; | |
5a595099 DC |
4899 | xfs_extlen_t mod; |
4900 | ||
4901 | bno = div_u64_rem(del->br_startblock, mp->m_sb.sb_rextsize, | |
4902 | &mod); | |
4903 | ASSERT(mod == 0); | |
4904 | len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize, | |
4905 | &mod); | |
4906 | ASSERT(mod == 0); | |
ad68fd19 | 4907 | |
ad68fd19 CH |
4908 | error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len); |
4909 | if (error) | |
4910 | goto done; | |
49f693fa | 4911 | do_fx = 0; |
ad68fd19 CH |
4912 | nblks = len * mp->m_sb.sb_rextsize; |
4913 | qfield = XFS_TRANS_DQ_RTBCOUNT; | |
4914 | } else { | |
4915 | do_fx = 1; | |
4916 | nblks = del->br_blockcount; | |
4917 | qfield = XFS_TRANS_DQ_BCOUNT; | |
4918 | } | |
4919 | ||
4920 | del_endblock = del->br_startblock + del->br_blockcount; | |
4921 | if (cur) { | |
70a93110 | 4922 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
ad68fd19 CH |
4923 | if (error) |
4924 | goto done; | |
4925 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); | |
49f693fa | 4926 | } |
85aec44f | 4927 | |
b039ac79 CH |
4928 | if (got.br_startoff == del->br_startoff) |
4929 | state |= BMAP_LEFT_FILLING; | |
4930 | if (got_endoff == del_endoff) | |
4931 | state |= BMAP_RIGHT_FILLING; | |
4932 | ||
4933 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { | |
4934 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
49f693fa DC |
4935 | /* |
4936 | * Matches the whole extent. Delete the entry. | |
4937 | */ | |
cf455a62 | 4938 | xfs_iext_remove(ip, icur, state); |
9788e059 | 4939 | xfs_iext_prev(ifp, icur); |
49f693fa DC |
4940 | XFS_IFORK_NEXT_SET(ip, whichfork, |
4941 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
4942 | flags |= XFS_ILOG_CORE; | |
4943 | if (!cur) { | |
4944 | flags |= xfs_ilog_fext(whichfork); | |
4945 | break; | |
2bd0ea18 | 4946 | } |
49f693fa DC |
4947 | if ((error = xfs_btree_delete(cur, &i))) |
4948 | goto done; | |
19ebedcf | 4949 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 4950 | break; |
b039ac79 | 4951 | case BMAP_LEFT_FILLING: |
2bd0ea18 | 4952 | /* |
49f693fa | 4953 | * Deleting the first part of the extent. |
2bd0ea18 | 4954 | */ |
70bf7533 CH |
4955 | got.br_startoff = del_endoff; |
4956 | got.br_startblock = del_endblock; | |
4957 | got.br_blockcount -= del->br_blockcount; | |
9788e059 | 4958 | xfs_iext_update_extent(ip, state, icur, &got); |
49f693fa DC |
4959 | if (!cur) { |
4960 | flags |= xfs_ilog_fext(whichfork); | |
4961 | break; | |
4962 | } | |
d0e5f1ff | 4963 | error = xfs_bmbt_update(cur, &got); |
70bf7533 | 4964 | if (error) |
49f693fa DC |
4965 | goto done; |
4966 | break; | |
b039ac79 | 4967 | case BMAP_RIGHT_FILLING: |
2bd0ea18 | 4968 | /* |
49f693fa | 4969 | * Deleting the last part of the extent. |
2bd0ea18 | 4970 | */ |
70bf7533 | 4971 | got.br_blockcount -= del->br_blockcount; |
9788e059 | 4972 | xfs_iext_update_extent(ip, state, icur, &got); |
49f693fa DC |
4973 | if (!cur) { |
4974 | flags |= xfs_ilog_fext(whichfork); | |
4975 | break; | |
4976 | } | |
d0e5f1ff | 4977 | error = xfs_bmbt_update(cur, &got); |
70bf7533 | 4978 | if (error) |
49f693fa DC |
4979 | goto done; |
4980 | break; | |
49f693fa DC |
4981 | case 0: |
4982 | /* | |
4983 | * Deleting the middle of the extent. | |
4984 | */ | |
70bf7533 | 4985 | old = got; |
df926c07 | 4986 | |
70bf7533 | 4987 | got.br_blockcount = del->br_startoff - got.br_startoff; |
9788e059 | 4988 | xfs_iext_update_extent(ip, state, icur, &got); |
70bf7533 | 4989 | |
49f693fa | 4990 | new.br_startoff = del_endoff; |
70bf7533 | 4991 | new.br_blockcount = got_endoff - del_endoff; |
49f693fa | 4992 | new.br_state = got.br_state; |
ad68fd19 | 4993 | new.br_startblock = del_endblock; |
70bf7533 | 4994 | |
ad68fd19 CH |
4995 | flags |= XFS_ILOG_CORE; |
4996 | if (cur) { | |
d0e5f1ff | 4997 | error = xfs_bmbt_update(cur, &got); |
ad68fd19 CH |
4998 | if (error) |
4999 | goto done; | |
5000 | error = xfs_btree_increment(cur, 0, &i); | |
5001 | if (error) | |
5002 | goto done; | |
5003 | cur->bc_rec.b = new; | |
5004 | error = xfs_btree_insert(cur, &i); | |
5005 | if (error && error != -ENOSPC) | |
5006 | goto done; | |
5007 | /* | |
5008 | * If get no-space back from btree insert, it tried a | |
5009 | * split, and we have a zero block reservation. Fix up | |
5010 | * our state and return the error. | |
5011 | */ | |
5012 | if (error == -ENOSPC) { | |
49f693fa | 5013 | /* |
ad68fd19 CH |
5014 | * Reset the cursor, don't trust it after any |
5015 | * insert operation. | |
49f693fa | 5016 | */ |
70a93110 | 5017 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
ad68fd19 | 5018 | if (error) |
49f693fa | 5019 | goto done; |
19ebedcf | 5020 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
ad68fd19 CH |
5021 | /* |
5022 | * Update the btree record back | |
5023 | * to the original value. | |
5024 | */ | |
d0e5f1ff | 5025 | error = xfs_bmbt_update(cur, &old); |
ad68fd19 CH |
5026 | if (error) |
5027 | goto done; | |
5028 | /* | |
5029 | * Reset the extent record back | |
5030 | * to the original value. | |
5031 | */ | |
9788e059 | 5032 | xfs_iext_update_extent(ip, state, icur, &old); |
ad68fd19 CH |
5033 | flags = 0; |
5034 | error = -ENOSPC; | |
5035 | goto done; | |
5036 | } | |
5037 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); | |
5038 | } else | |
5039 | flags |= xfs_ilog_fext(whichfork); | |
5040 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5041 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
9788e059 | 5042 | xfs_iext_next(ifp, icur); |
26a75f67 | 5043 | xfs_iext_insert(ip, icur, &new, state); |
49f693fa | 5044 | break; |
2bd0ea18 | 5045 | } |
d7f80320 DW |
5046 | |
5047 | /* remove reverse mapping */ | |
21375e5d | 5048 | error = xfs_rmap_unmap_extent(tp, ip, whichfork, del); |
ad68fd19 CH |
5049 | if (error) |
5050 | goto done; | |
d7f80320 | 5051 | |
2bd0ea18 | 5052 | /* |
49f693fa | 5053 | * If we need to, add to list of extents to delete. |
2bd0ea18 | 5054 | */ |
36b16da8 | 5055 | if (do_fx && !(bflags & XFS_BMAPI_REMAP)) { |
cfe32f0d | 5056 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) { |
21375e5d | 5057 | error = xfs_refcount_decrease_extent(tp, del); |
cfe32f0d DW |
5058 | if (error) |
5059 | goto done; | |
3a13f959 | 5060 | } else { |
21375e5d | 5061 | __xfs_bmap_add_free(tp, del->br_startblock, |
cc5af22a BF |
5062 | del->br_blockcount, NULL, |
5063 | (bflags & XFS_BMAPI_NODISCARD) || | |
5064 | del->br_state == XFS_EXT_UNWRITTEN); | |
3a13f959 | 5065 | } |
cfe32f0d DW |
5066 | } |
5067 | ||
2bd0ea18 | 5068 | /* |
49f693fa | 5069 | * Adjust inode # blocks in the file. |
2bd0ea18 | 5070 | */ |
49f693fa DC |
5071 | if (nblks) |
5072 | ip->i_d.di_nblocks -= nblks; | |
2bd0ea18 | 5073 | /* |
49f693fa | 5074 | * Adjust quota data. |
2bd0ea18 | 5075 | */ |
36b16da8 | 5076 | if (qfield && !(bflags & XFS_BMAPI_REMAP)) |
49f693fa DC |
5077 | xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks); |
5078 | ||
49f693fa DC |
5079 | done: |
5080 | *logflagsp = flags; | |
2bd0ea18 NS |
5081 | return error; |
5082 | } | |
5083 | ||
5084 | /* | |
49f693fa DC |
5085 | * Unmap (remove) blocks from a file. |
5086 | * If nexts is nonzero then the number of extents to remove is limited to | |
5087 | * that value. If not all extents in the block range can be removed then | |
5088 | * *done is set. | |
2bd0ea18 | 5089 | */ |
49f693fa | 5090 | int /* error */ |
3d36acda | 5091 | __xfs_bunmapi( |
160ed12f | 5092 | struct xfs_trans *tp, /* transaction pointer */ |
49f693fa | 5093 | struct xfs_inode *ip, /* incore inode */ |
675b5a20 | 5094 | xfs_fileoff_t start, /* first file offset deleted */ |
3d36acda | 5095 | xfs_filblks_t *rlen, /* i/o: amount remaining */ |
49f693fa | 5096 | int flags, /* misc flags */ |
d3c5f3dd | 5097 | xfs_extnum_t nexts) /* number of extents max */ |
2bd0ea18 | 5098 | { |
160ed12f BF |
5099 | struct xfs_btree_cur *cur; /* bmap btree cursor */ |
5100 | struct xfs_bmbt_irec del; /* extent being deleted */ | |
49f693fa DC |
5101 | int error; /* error return value */ |
5102 | xfs_extnum_t extno; /* extent number in list */ | |
160ed12f | 5103 | struct xfs_bmbt_irec got; /* current extent record */ |
e07055b8 | 5104 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa | 5105 | int isrt; /* freeing in rt area */ |
49f693fa DC |
5106 | int logflags; /* transaction logging flags */ |
5107 | xfs_extlen_t mod; /* rt extent offset */ | |
160ed12f | 5108 | struct xfs_mount *mp; /* mount structure */ |
49f693fa DC |
5109 | int tmp_logflags; /* partial logging flags */ |
5110 | int wasdel; /* was a delayed alloc extent */ | |
5111 | int whichfork; /* data or attribute fork */ | |
5112 | xfs_fsblock_t sum; | |
3d36acda | 5113 | xfs_filblks_t len = *rlen; /* length to unmap in file */ |
594956fa | 5114 | xfs_fileoff_t max_len; |
15a8bccc | 5115 | xfs_agnumber_t prev_agno = NULLAGNUMBER, agno; |
675b5a20 | 5116 | xfs_fileoff_t end; |
9788e059 CH |
5117 | struct xfs_iext_cursor icur; |
5118 | bool done = false; | |
2bd0ea18 | 5119 | |
675b5a20 | 5120 | trace_xfs_bunmap(ip, start, len, flags, _RET_IP_); |
a2ceac1f | 5121 | |
cb8a004a DW |
5122 | whichfork = xfs_bmapi_whichfork(flags); |
5123 | ASSERT(whichfork != XFS_COW_FORK); | |
a2ceac1f | 5124 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa DC |
5125 | if (unlikely( |
5126 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5127 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
5128 | XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW, | |
5129 | ip->i_mount); | |
12b53197 | 5130 | return -EFSCORRUPTED; |
49f693fa DC |
5131 | } |
5132 | mp = ip->i_mount; | |
5133 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 5134 | return -EIO; |
56b2de80 | 5135 | |
ff105f75 | 5136 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
49f693fa DC |
5137 | ASSERT(len > 0); |
5138 | ASSERT(nexts >= 0); | |
56b2de80 | 5139 | |
594956fa DW |
5140 | /* |
5141 | * Guesstimate how many blocks we can unmap without running the risk of | |
5142 | * blowing out the transaction with a mix of EFIs and reflink | |
5143 | * adjustments. | |
5144 | */ | |
db2cf1d2 | 5145 | if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) |
594956fa DW |
5146 | max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res)); |
5147 | else | |
5148 | max_len = len; | |
5149 | ||
49f693fa DC |
5150 | if (!(ifp->if_flags & XFS_IFEXTENTS) && |
5151 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
5152 | return error; | |
d09d4e5f | 5153 | if (xfs_iext_count(ifp) == 0) { |
3d36acda | 5154 | *rlen = 0; |
49f693fa | 5155 | return 0; |
56b2de80 | 5156 | } |
79896434 | 5157 | XFS_STATS_INC(mp, xs_blk_unmap); |
49f693fa | 5158 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); |
d6fbe8fe | 5159 | end = start + len; |
a2ceac1f | 5160 | |
9788e059 | 5161 | if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) { |
d6fbe8fe CH |
5162 | *rlen = 0; |
5163 | return 0; | |
49f693fa | 5164 | } |
d6fbe8fe | 5165 | end--; |
246eb90a | 5166 | |
49f693fa DC |
5167 | logflags = 0; |
5168 | if (ifp->if_flags & XFS_IFBROOT) { | |
5169 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
5170 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
49f693fa DC |
5171 | cur->bc_private.b.flags = 0; |
5172 | } else | |
5173 | cur = NULL; | |
a2ceac1f | 5174 | |
49f693fa | 5175 | if (isrt) { |
a2ceac1f | 5176 | /* |
49f693fa | 5177 | * Synchronize by locking the bitmap inode. |
a2ceac1f | 5178 | */ |
a62ed6d3 | 5179 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
49f693fa | 5180 | xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
a62ed6d3 DW |
5181 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
5182 | xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL); | |
49f693fa | 5183 | } |
a2ceac1f | 5184 | |
49f693fa | 5185 | extno = 0; |
9788e059 | 5186 | while (end != (xfs_fileoff_t)-1 && end >= start && |
594956fa | 5187 | (nexts == 0 || extno < nexts) && max_len > 0) { |
a2ceac1f | 5188 | /* |
675b5a20 | 5189 | * Is the found extent after a hole in which end lives? |
49f693fa | 5190 | * Just back up to the previous extent, if so. |
a2ceac1f | 5191 | */ |
9788e059 CH |
5192 | if (got.br_startoff > end && |
5193 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5194 | done = true; | |
5195 | break; | |
a2ceac1f | 5196 | } |
49f693fa DC |
5197 | /* |
5198 | * Is the last block of this extent before the range | |
5199 | * we're supposed to delete? If so, we're done. | |
5200 | */ | |
675b5a20 | 5201 | end = XFS_FILEOFF_MIN(end, |
49f693fa | 5202 | got.br_startoff + got.br_blockcount - 1); |
675b5a20 | 5203 | if (end < start) |
49f693fa DC |
5204 | break; |
5205 | /* | |
5206 | * Then deal with the (possibly delayed) allocated space | |
5207 | * we found. | |
5208 | */ | |
49f693fa DC |
5209 | del = got; |
5210 | wasdel = isnullstartblock(del.br_startblock); | |
15a8bccc CH |
5211 | |
5212 | /* | |
5213 | * Make sure we don't touch multiple AGF headers out of order | |
5214 | * in a single transaction, as that could cause AB-BA deadlocks. | |
5215 | */ | |
5216 | if (!wasdel) { | |
5217 | agno = XFS_FSB_TO_AGNO(mp, del.br_startblock); | |
5218 | if (prev_agno != NULLAGNUMBER && prev_agno > agno) | |
5219 | break; | |
5220 | prev_agno = agno; | |
5221 | } | |
49f693fa DC |
5222 | if (got.br_startoff < start) { |
5223 | del.br_startoff = start; | |
5224 | del.br_blockcount -= start - got.br_startoff; | |
5225 | if (!wasdel) | |
5226 | del.br_startblock += start - got.br_startoff; | |
5227 | } | |
675b5a20 CH |
5228 | if (del.br_startoff + del.br_blockcount > end + 1) |
5229 | del.br_blockcount = end + 1 - del.br_startoff; | |
594956fa DW |
5230 | |
5231 | /* How much can we safely unmap? */ | |
5232 | if (max_len < del.br_blockcount) { | |
5233 | del.br_startoff += del.br_blockcount - max_len; | |
5234 | if (!wasdel) | |
5235 | del.br_startblock += del.br_blockcount - max_len; | |
5236 | del.br_blockcount = max_len; | |
5237 | } | |
5238 | ||
5a595099 DC |
5239 | if (!isrt) |
5240 | goto delete; | |
5241 | ||
49f693fa | 5242 | sum = del.br_startblock + del.br_blockcount; |
5a595099 DC |
5243 | div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod); |
5244 | if (mod) { | |
49f693fa DC |
5245 | /* |
5246 | * Realtime extent not lined up at the end. | |
5247 | * The extent could have been split into written | |
5248 | * and unwritten pieces, or we could just be | |
5249 | * unmapping part of it. But we can't really | |
5250 | * get rid of part of a realtime extent. | |
5251 | */ | |
b7fdeb4d | 5252 | if (del.br_state == XFS_EXT_UNWRITTEN) { |
49f693fa DC |
5253 | /* |
5254 | * This piece is unwritten, or we're not | |
5255 | * using unwritten extents. Skip over it. | |
5256 | */ | |
675b5a20 CH |
5257 | ASSERT(end >= mod); |
5258 | end -= mod > del.br_blockcount ? | |
49f693fa | 5259 | del.br_blockcount : mod; |
9788e059 CH |
5260 | if (end < got.br_startoff && |
5261 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5262 | done = true; | |
5263 | break; | |
49f693fa DC |
5264 | } |
5265 | continue; | |
5266 | } | |
5267 | /* | |
5268 | * It's written, turn it unwritten. | |
5269 | * This is better than zeroing it. | |
5270 | */ | |
5271 | ASSERT(del.br_state == XFS_EXT_NORM); | |
0268fdc3 | 5272 | ASSERT(tp->t_blk_res > 0); |
49f693fa DC |
5273 | /* |
5274 | * If this spans a realtime extent boundary, | |
5275 | * chop it back to the start of the one we end at. | |
5276 | */ | |
5277 | if (del.br_blockcount > mod) { | |
5278 | del.br_startoff += del.br_blockcount - mod; | |
5279 | del.br_startblock += del.br_blockcount - mod; | |
5280 | del.br_blockcount = mod; | |
5281 | } | |
5282 | del.br_state = XFS_EXT_UNWRITTEN; | |
5283 | error = xfs_bmap_add_extent_unwritten_real(tp, ip, | |
9788e059 | 5284 | whichfork, &icur, &cur, &del, |
64e8b4a7 | 5285 | &logflags); |
49f693fa DC |
5286 | if (error) |
5287 | goto error0; | |
5288 | goto nodelete; | |
a2ceac1f | 5289 | } |
5a595099 DC |
5290 | div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod); |
5291 | if (mod) { | |
49f693fa DC |
5292 | /* |
5293 | * Realtime extent is lined up at the end but not | |
5294 | * at the front. We'll get rid of full extents if | |
5295 | * we can. | |
5296 | */ | |
5297 | mod = mp->m_sb.sb_rextsize - mod; | |
5298 | if (del.br_blockcount > mod) { | |
5299 | del.br_blockcount -= mod; | |
5300 | del.br_startoff += mod; | |
5301 | del.br_startblock += mod; | |
b7fdeb4d CH |
5302 | } else if (del.br_startoff == start && |
5303 | (del.br_state == XFS_EXT_UNWRITTEN || | |
5304 | tp->t_blk_res == 0)) { | |
49f693fa DC |
5305 | /* |
5306 | * Can't make it unwritten. There isn't | |
5307 | * a full extent here so just skip it. | |
5308 | */ | |
675b5a20 CH |
5309 | ASSERT(end >= del.br_blockcount); |
5310 | end -= del.br_blockcount; | |
9788e059 CH |
5311 | if (got.br_startoff > end && |
5312 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5313 | done = true; | |
5314 | break; | |
5315 | } | |
49f693fa DC |
5316 | continue; |
5317 | } else if (del.br_state == XFS_EXT_UNWRITTEN) { | |
246eb90a CH |
5318 | struct xfs_bmbt_irec prev; |
5319 | ||
49f693fa DC |
5320 | /* |
5321 | * This one is already unwritten. | |
5322 | * It must have a written left neighbor. | |
5323 | * Unwrite the killed part of that one and | |
5324 | * try again. | |
5325 | */ | |
9788e059 CH |
5326 | if (!xfs_iext_prev_extent(ifp, &icur, &prev)) |
5327 | ASSERT(0); | |
49f693fa DC |
5328 | ASSERT(prev.br_state == XFS_EXT_NORM); |
5329 | ASSERT(!isnullstartblock(prev.br_startblock)); | |
5330 | ASSERT(del.br_startblock == | |
5331 | prev.br_startblock + prev.br_blockcount); | |
5332 | if (prev.br_startoff < start) { | |
5333 | mod = start - prev.br_startoff; | |
5334 | prev.br_blockcount -= mod; | |
5335 | prev.br_startblock += mod; | |
5336 | prev.br_startoff = start; | |
5337 | } | |
5338 | prev.br_state = XFS_EXT_UNWRITTEN; | |
49f693fa | 5339 | error = xfs_bmap_add_extent_unwritten_real(tp, |
9788e059 | 5340 | ip, whichfork, &icur, &cur, |
64e8b4a7 | 5341 | &prev, &logflags); |
49f693fa DC |
5342 | if (error) |
5343 | goto error0; | |
5344 | goto nodelete; | |
5345 | } else { | |
5346 | ASSERT(del.br_state == XFS_EXT_NORM); | |
5347 | del.br_state = XFS_EXT_UNWRITTEN; | |
5348 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
9788e059 | 5349 | ip, whichfork, &icur, &cur, |
64e8b4a7 | 5350 | &del, &logflags); |
49f693fa DC |
5351 | if (error) |
5352 | goto error0; | |
5353 | goto nodelete; | |
5354 | } | |
5355 | } | |
a2ceac1f | 5356 | |
5a595099 | 5357 | delete: |
8359e0b9 | 5358 | if (wasdel) { |
9788e059 | 5359 | error = xfs_bmap_del_extent_delay(ip, whichfork, &icur, |
ad68fd19 CH |
5360 | &got, &del); |
5361 | } else { | |
52f6ed9d BF |
5362 | error = xfs_bmap_del_extent_real(ip, tp, &icur, cur, |
5363 | &del, &tmp_logflags, whichfork, | |
ad68fd19 CH |
5364 | flags); |
5365 | logflags |= tmp_logflags; | |
e9fa15aa | 5366 | } |
8359e0b9 | 5367 | |
49f693fa DC |
5368 | if (error) |
5369 | goto error0; | |
8359e0b9 | 5370 | |
594956fa | 5371 | max_len -= del.br_blockcount; |
675b5a20 | 5372 | end = del.br_startoff - 1; |
49f693fa | 5373 | nodelete: |
a2ceac1f | 5374 | /* |
49f693fa | 5375 | * If not done go on to the next (previous) record. |
a2ceac1f | 5376 | */ |
675b5a20 | 5377 | if (end != (xfs_fileoff_t)-1 && end >= start) { |
9788e059 CH |
5378 | if (!xfs_iext_get_extent(ifp, &icur, &got) || |
5379 | (got.br_startoff > end && | |
5380 | !xfs_iext_prev_extent(ifp, &icur, &got))) { | |
5381 | done = true; | |
5382 | break; | |
49f693fa DC |
5383 | } |
5384 | extno++; | |
a2ceac1f | 5385 | } |
a2ceac1f | 5386 | } |
9788e059 | 5387 | if (done || end == (xfs_fileoff_t)-1 || end < start) |
3d36acda DW |
5388 | *rlen = 0; |
5389 | else | |
675b5a20 | 5390 | *rlen = end - start + 1; |
56b2de80 | 5391 | |
49f693fa DC |
5392 | /* |
5393 | * Convert to a btree if necessary. | |
5394 | */ | |
5395 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
5396 | ASSERT(cur == NULL); | |
f7253505 BF |
5397 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
5398 | &tmp_logflags, whichfork); | |
49f693fa DC |
5399 | logflags |= tmp_logflags; |
5400 | if (error) | |
5401 | goto error0; | |
56b2de80 | 5402 | } |
56b2de80 | 5403 | /* |
49f693fa | 5404 | * transform from btree to extents, give it cur |
56b2de80 | 5405 | */ |
49f693fa DC |
5406 | else if (xfs_bmap_wants_extents(ip, whichfork)) { |
5407 | ASSERT(cur != NULL); | |
5408 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags, | |
5409 | whichfork); | |
5410 | logflags |= tmp_logflags; | |
5411 | if (error) | |
5412 | goto error0; | |
56b2de80 | 5413 | } |
49f693fa DC |
5414 | /* |
5415 | * transform from extents to local? | |
5416 | */ | |
5417 | error = 0; | |
5418 | error0: | |
5419 | /* | |
5420 | * Log everything. Do this after conversion, there's no point in | |
5421 | * logging the extent records if we've converted to btree format. | |
5422 | */ | |
5423 | if ((logflags & xfs_ilog_fext(whichfork)) && | |
5424 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
5425 | logflags &= ~xfs_ilog_fext(whichfork); | |
5426 | else if ((logflags & xfs_ilog_fbroot(whichfork)) && | |
5427 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) | |
5428 | logflags &= ~xfs_ilog_fbroot(whichfork); | |
5429 | /* | |
5430 | * Log inode even in the error case, if the transaction | |
5431 | * is dirty we'll need to shut down the filesystem. | |
5432 | */ | |
5433 | if (logflags) | |
5434 | xfs_trans_log_inode(tp, ip, logflags); | |
5435 | if (cur) { | |
f44a348b | 5436 | if (!error) |
49f693fa | 5437 | cur->bc_private.b.allocated = 0; |
660265b7 | 5438 | xfs_btree_del_cursor(cur, error); |
56b2de80 | 5439 | } |
49f693fa | 5440 | return error; |
a2ceac1f | 5441 | } |
ff105f75 | 5442 | |
3d36acda DW |
5443 | /* Unmap a range of a file. */ |
5444 | int | |
5445 | xfs_bunmapi( | |
5446 | xfs_trans_t *tp, | |
5447 | struct xfs_inode *ip, | |
5448 | xfs_fileoff_t bno, | |
5449 | xfs_filblks_t len, | |
5450 | int flags, | |
5451 | xfs_extnum_t nexts, | |
3d36acda DW |
5452 | int *done) |
5453 | { | |
5454 | int error; | |
5455 | ||
d3c5f3dd | 5456 | error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts); |
3d36acda DW |
5457 | *done = (len == 0); |
5458 | return error; | |
5459 | } | |
5460 | ||
5a35bf2c DC |
5461 | /* |
5462 | * Determine whether an extent shift can be accomplished by a merge with the | |
5463 | * extent that precedes the target hole of the shift. | |
5464 | */ | |
5465 | STATIC bool | |
5466 | xfs_bmse_can_merge( | |
5467 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5468 | struct xfs_bmbt_irec *got, /* current extent to shift */ | |
5469 | xfs_fileoff_t shift) /* shift fsb */ | |
5470 | { | |
5471 | xfs_fileoff_t startoff; | |
5472 | ||
5473 | startoff = got->br_startoff - shift; | |
5474 | ||
5475 | /* | |
5476 | * The extent, once shifted, must be adjacent in-file and on-disk with | |
5477 | * the preceding extent. | |
5478 | */ | |
5479 | if ((left->br_startoff + left->br_blockcount != startoff) || | |
5480 | (left->br_startblock + left->br_blockcount != got->br_startblock) || | |
5481 | (left->br_state != got->br_state) || | |
5482 | (left->br_blockcount + got->br_blockcount > MAXEXTLEN)) | |
5483 | return false; | |
5484 | ||
5485 | return true; | |
5486 | } | |
5487 | ||
5488 | /* | |
5489 | * A bmap extent shift adjusts the file offset of an extent to fill a preceding | |
5490 | * hole in the file. If an extent shift would result in the extent being fully | |
5491 | * adjacent to the extent that currently precedes the hole, we can merge with | |
5492 | * the preceding extent rather than do the shift. | |
5493 | * | |
5494 | * This function assumes the caller has verified a shift-by-merge is possible | |
5495 | * with the provided extents via xfs_bmse_can_merge(). | |
5496 | */ | |
5497 | STATIC int | |
5498 | xfs_bmse_merge( | |
21375e5d | 5499 | struct xfs_trans *tp, |
5a35bf2c DC |
5500 | struct xfs_inode *ip, |
5501 | int whichfork, | |
5502 | xfs_fileoff_t shift, /* shift fsb */ | |
9788e059 | 5503 | struct xfs_iext_cursor *icur, |
2f082827 CH |
5504 | struct xfs_bmbt_irec *got, /* extent to shift */ |
5505 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5a35bf2c | 5506 | struct xfs_btree_cur *cur, |
21375e5d | 5507 | int *logflags) /* output */ |
5a35bf2c | 5508 | { |
2f082827 | 5509 | struct xfs_bmbt_irec new; |
5a35bf2c DC |
5510 | xfs_filblks_t blockcount; |
5511 | int error, i; | |
19ebedcf | 5512 | struct xfs_mount *mp = ip->i_mount; |
5a35bf2c | 5513 | |
2f082827 | 5514 | blockcount = left->br_blockcount + got->br_blockcount; |
5a35bf2c DC |
5515 | |
5516 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5517 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
2f082827 | 5518 | ASSERT(xfs_bmse_can_merge(left, got, shift)); |
5a35bf2c | 5519 | |
2f082827 CH |
5520 | new = *left; |
5521 | new.br_blockcount = blockcount; | |
5a35bf2c DC |
5522 | |
5523 | /* | |
5524 | * Update the on-disk extent count, the btree if necessary and log the | |
5525 | * inode. | |
5526 | */ | |
5527 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5528 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5529 | *logflags |= XFS_ILOG_CORE; | |
5530 | if (!cur) { | |
5531 | *logflags |= XFS_ILOG_DEXT; | |
2f082827 | 5532 | goto done; |
5a35bf2c DC |
5533 | } |
5534 | ||
5535 | /* lookup and remove the extent to merge */ | |
70a93110 | 5536 | error = xfs_bmbt_lookup_eq(cur, got, &i); |
5a35bf2c DC |
5537 | if (error) |
5538 | return error; | |
19ebedcf | 5539 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5540 | |
5541 | error = xfs_btree_delete(cur, &i); | |
5542 | if (error) | |
5543 | return error; | |
19ebedcf | 5544 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5545 | |
5546 | /* lookup and update size of the previous extent */ | |
70a93110 | 5547 | error = xfs_bmbt_lookup_eq(cur, left, &i); |
5a35bf2c DC |
5548 | if (error) |
5549 | return error; | |
19ebedcf | 5550 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c | 5551 | |
d0e5f1ff | 5552 | error = xfs_bmbt_update(cur, &new); |
2f082827 CH |
5553 | if (error) |
5554 | return error; | |
5555 | ||
5556 | done: | |
cf455a62 | 5557 | xfs_iext_remove(ip, icur, 0); |
9788e059 CH |
5558 | xfs_iext_prev(XFS_IFORK_PTR(ip, whichfork), icur); |
5559 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur, | |
5560 | &new); | |
5a35bf2c | 5561 | |
479284b7 | 5562 | /* update reverse mapping. rmap functions merge the rmaps for us */ |
21375e5d | 5563 | error = xfs_rmap_unmap_extent(tp, ip, whichfork, got); |
2f082827 CH |
5564 | if (error) |
5565 | return error; | |
479284b7 DW |
5566 | memcpy(&new, got, sizeof(new)); |
5567 | new.br_startoff = left->br_startoff + left->br_blockcount; | |
21375e5d | 5568 | return xfs_rmap_map_extent(tp, ip, whichfork, &new); |
5a35bf2c DC |
5569 | } |
5570 | ||
364937e2 CH |
5571 | static int |
5572 | xfs_bmap_shift_update_extent( | |
21375e5d | 5573 | struct xfs_trans *tp, |
364937e2 CH |
5574 | struct xfs_inode *ip, |
5575 | int whichfork, | |
9788e059 | 5576 | struct xfs_iext_cursor *icur, |
364937e2 CH |
5577 | struct xfs_bmbt_irec *got, |
5578 | struct xfs_btree_cur *cur, | |
5579 | int *logflags, | |
364937e2 | 5580 | xfs_fileoff_t startoff) |
5a35bf2c | 5581 | { |
364937e2 | 5582 | struct xfs_mount *mp = ip->i_mount; |
f36ccac2 | 5583 | struct xfs_bmbt_irec prev = *got; |
364937e2 | 5584 | int error, i; |
2f082827 | 5585 | |
5a35bf2c | 5586 | *logflags |= XFS_ILOG_CORE; |
2f082827 | 5587 | |
f36ccac2 | 5588 | got->br_startoff = startoff; |
2f082827 CH |
5589 | |
5590 | if (cur) { | |
f36ccac2 | 5591 | error = xfs_bmbt_lookup_eq(cur, &prev, &i); |
2f082827 CH |
5592 | if (error) |
5593 | return error; | |
5594 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); | |
5595 | ||
f36ccac2 | 5596 | error = xfs_bmbt_update(cur, got); |
2f082827 CH |
5597 | if (error) |
5598 | return error; | |
5599 | } else { | |
5a35bf2c | 5600 | *logflags |= XFS_ILOG_DEXT; |
5a35bf2c DC |
5601 | } |
5602 | ||
9788e059 CH |
5603 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur, |
5604 | got); | |
d7f80320 | 5605 | |
d7f80320 | 5606 | /* update reverse mapping */ |
21375e5d | 5607 | error = xfs_rmap_unmap_extent(tp, ip, whichfork, &prev); |
d7f80320 DW |
5608 | if (error) |
5609 | return error; | |
21375e5d | 5610 | return xfs_rmap_map_extent(tp, ip, whichfork, got); |
5a35bf2c DC |
5611 | } |
5612 | ||
ff105f75 | 5613 | int |
a9f5d25e | 5614 | xfs_bmap_collapse_extents( |
ff105f75 DC |
5615 | struct xfs_trans *tp, |
5616 | struct xfs_inode *ip, | |
19ebedcf | 5617 | xfs_fileoff_t *next_fsb, |
ff105f75 | 5618 | xfs_fileoff_t offset_shift_fsb, |
a397f27d | 5619 | bool *done) |
ff105f75 | 5620 | { |
a9f5d25e CH |
5621 | int whichfork = XFS_DATA_FORK; |
5622 | struct xfs_mount *mp = ip->i_mount; | |
5623 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5624 | struct xfs_btree_cur *cur = NULL; | |
364937e2 | 5625 | struct xfs_bmbt_irec got, prev; |
9788e059 | 5626 | struct xfs_iext_cursor icur; |
364937e2 | 5627 | xfs_fileoff_t new_startoff; |
a9f5d25e CH |
5628 | int error = 0; |
5629 | int logflags = 0; | |
ff105f75 DC |
5630 | |
5631 | if (unlikely(XFS_TEST_ERROR( | |
5632 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5633 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 5634 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a9f5d25e | 5635 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); |
12b53197 | 5636 | return -EFSCORRUPTED; |
ff105f75 DC |
5637 | } |
5638 | ||
5639 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 5640 | return -EIO; |
ff105f75 | 5641 | |
a9f5d25e | 5642 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); |
ff105f75 | 5643 | |
ff105f75 | 5644 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
ff105f75 DC |
5645 | error = xfs_iread_extents(tp, ip, whichfork); |
5646 | if (error) | |
5647 | return error; | |
5648 | } | |
5649 | ||
ff105f75 DC |
5650 | if (ifp->if_flags & XFS_IFBROOT) { |
5651 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
ff105f75 | 5652 | cur->bc_private.b.flags = 0; |
5a35bf2c DC |
5653 | } |
5654 | ||
9788e059 | 5655 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) { |
a9f5d25e CH |
5656 | *done = true; |
5657 | goto del_cursor; | |
5658 | } | |
5c40ca1f ES |
5659 | XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock), |
5660 | del_cursor); | |
a9f5d25e | 5661 | |
364937e2 | 5662 | new_startoff = got.br_startoff - offset_shift_fsb; |
9788e059 | 5663 | if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) { |
364937e2 CH |
5664 | if (new_startoff < prev.br_startoff + prev.br_blockcount) { |
5665 | error = -EINVAL; | |
5666 | goto del_cursor; | |
5667 | } | |
5668 | ||
364937e2 | 5669 | if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) { |
21375e5d BF |
5670 | error = xfs_bmse_merge(tp, ip, whichfork, |
5671 | offset_shift_fsb, &icur, &got, &prev, | |
5672 | cur, &logflags); | |
364937e2 CH |
5673 | if (error) |
5674 | goto del_cursor; | |
5675 | goto done; | |
5676 | } | |
5677 | } else { | |
5678 | if (got.br_startoff < offset_shift_fsb) { | |
5679 | error = -EINVAL; | |
5680 | goto del_cursor; | |
5681 | } | |
5682 | } | |
5683 | ||
21375e5d BF |
5684 | error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got, |
5685 | cur, &logflags, new_startoff); | |
a9f5d25e CH |
5686 | if (error) |
5687 | goto del_cursor; | |
bac20498 | 5688 | |
c162e319 | 5689 | done: |
9788e059 CH |
5690 | if (!xfs_iext_next_extent(ifp, &icur, &got)) { |
5691 | *done = true; | |
5692 | goto del_cursor; | |
a9f5d25e | 5693 | } |
a9f5d25e | 5694 | |
364937e2 | 5695 | *next_fsb = got.br_startoff; |
a9f5d25e CH |
5696 | del_cursor: |
5697 | if (cur) | |
660265b7 | 5698 | xfs_btree_del_cursor(cur, error); |
a9f5d25e CH |
5699 | if (logflags) |
5700 | xfs_trans_log_inode(tp, ip, logflags); | |
a9f5d25e CH |
5701 | return error; |
5702 | } | |
5703 | ||
5b501597 DW |
5704 | /* Make sure we won't be right-shifting an extent past the maximum bound. */ |
5705 | int | |
5706 | xfs_bmap_can_insert_extents( | |
5707 | struct xfs_inode *ip, | |
5708 | xfs_fileoff_t off, | |
5709 | xfs_fileoff_t shift) | |
5710 | { | |
5711 | struct xfs_bmbt_irec got; | |
5712 | int is_empty; | |
5713 | int error = 0; | |
5714 | ||
5715 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5716 | ||
5717 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
5718 | return -EIO; | |
5719 | ||
5720 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5721 | error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty); | |
5722 | if (!error && !is_empty && got.br_startoff >= off && | |
5723 | ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff) | |
5724 | error = -EINVAL; | |
5725 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
5726 | ||
5727 | return error; | |
5728 | } | |
5729 | ||
a9f5d25e CH |
5730 | int |
5731 | xfs_bmap_insert_extents( | |
5732 | struct xfs_trans *tp, | |
5733 | struct xfs_inode *ip, | |
5734 | xfs_fileoff_t *next_fsb, | |
5735 | xfs_fileoff_t offset_shift_fsb, | |
5736 | bool *done, | |
a397f27d | 5737 | xfs_fileoff_t stop_fsb) |
a9f5d25e CH |
5738 | { |
5739 | int whichfork = XFS_DATA_FORK; | |
5740 | struct xfs_mount *mp = ip->i_mount; | |
5741 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5742 | struct xfs_btree_cur *cur = NULL; | |
bd80a804 | 5743 | struct xfs_bmbt_irec got, next; |
9788e059 | 5744 | struct xfs_iext_cursor icur; |
364937e2 | 5745 | xfs_fileoff_t new_startoff; |
a9f5d25e CH |
5746 | int error = 0; |
5747 | int logflags = 0; | |
5748 | ||
5749 | if (unlikely(XFS_TEST_ERROR( | |
5750 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5751 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
5752 | mp, XFS_ERRTAG_BMAPIFORMAT))) { | |
5753 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
5754 | return -EFSCORRUPTED; | |
5755 | } | |
5756 | ||
5757 | if (XFS_FORCED_SHUTDOWN(mp)) | |
5758 | return -EIO; | |
5759 | ||
5760 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); | |
5761 | ||
5762 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
5763 | error = xfs_iread_extents(tp, ip, whichfork); | |
5764 | if (error) | |
5765 | return error; | |
5766 | } | |
5767 | ||
5768 | if (ifp->if_flags & XFS_IFBROOT) { | |
5769 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
a9f5d25e CH |
5770 | cur->bc_private.b.flags = 0; |
5771 | } | |
5772 | ||
19ebedcf | 5773 | if (*next_fsb == NULLFSBLOCK) { |
9788e059 CH |
5774 | xfs_iext_last(ifp, &icur); |
5775 | if (!xfs_iext_get_extent(ifp, &icur, &got) || | |
bd80a804 | 5776 | stop_fsb > got.br_startoff) { |
a9f5d25e | 5777 | *done = true; |
19ebedcf DC |
5778 | goto del_cursor; |
5779 | } | |
b6ad780b | 5780 | } else { |
9788e059 | 5781 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) { |
a9f5d25e | 5782 | *done = true; |
b6ad780b CH |
5783 | goto del_cursor; |
5784 | } | |
19ebedcf | 5785 | } |
5c40ca1f ES |
5786 | XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock), |
5787 | del_cursor); | |
19ebedcf | 5788 | |
bd80a804 | 5789 | if (stop_fsb >= got.br_startoff + got.br_blockcount) { |
a9f5d25e CH |
5790 | error = -EIO; |
5791 | goto del_cursor; | |
19ebedcf DC |
5792 | } |
5793 | ||
364937e2 | 5794 | new_startoff = got.br_startoff + offset_shift_fsb; |
9788e059 | 5795 | if (xfs_iext_peek_next_extent(ifp, &icur, &next)) { |
364937e2 CH |
5796 | if (new_startoff + got.br_blockcount > next.br_startoff) { |
5797 | error = -EINVAL; | |
5798 | goto del_cursor; | |
5799 | } | |
5800 | ||
5801 | /* | |
5802 | * Unlike a left shift (which involves a hole punch), a right | |
5803 | * shift does not modify extent neighbors in any way. We should | |
5804 | * never find mergeable extents in this scenario. Check anyways | |
5805 | * and warn if we encounter two extents that could be one. | |
5806 | */ | |
5807 | if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb)) | |
5808 | WARN_ON_ONCE(1); | |
5809 | } | |
5810 | ||
21375e5d BF |
5811 | error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got, |
5812 | cur, &logflags, new_startoff); | |
6835c363 CH |
5813 | if (error) |
5814 | goto del_cursor; | |
bd80a804 | 5815 | |
9788e059 | 5816 | if (!xfs_iext_prev_extent(ifp, &icur, &got) || |
bd80a804 | 5817 | stop_fsb >= got.br_startoff + got.br_blockcount) { |
a9f5d25e | 5818 | *done = true; |
6835c363 | 5819 | goto del_cursor; |
ff105f75 DC |
5820 | } |
5821 | ||
6835c363 | 5822 | *next_fsb = got.br_startoff; |
ff105f75 DC |
5823 | del_cursor: |
5824 | if (cur) | |
660265b7 | 5825 | xfs_btree_del_cursor(cur, error); |
5a35bf2c DC |
5826 | if (logflags) |
5827 | xfs_trans_log_inode(tp, ip, logflags); | |
ff105f75 DC |
5828 | return error; |
5829 | } | |
19ebedcf DC |
5830 | |
5831 | /* | |
9788e059 CH |
5832 | * Splits an extent into two extents at split_fsb block such that it is the |
5833 | * first block of the current_ext. @ext is a target extent to be split. | |
5834 | * @split_fsb is a block where the extents is split. If split_fsb lies in a | |
5835 | * hole or the first block of extents, just return 0. | |
19ebedcf DC |
5836 | */ |
5837 | STATIC int | |
5838 | xfs_bmap_split_extent_at( | |
5839 | struct xfs_trans *tp, | |
5840 | struct xfs_inode *ip, | |
312350cd | 5841 | xfs_fileoff_t split_fsb) |
19ebedcf DC |
5842 | { |
5843 | int whichfork = XFS_DATA_FORK; | |
5844 | struct xfs_btree_cur *cur = NULL; | |
19ebedcf DC |
5845 | struct xfs_bmbt_irec got; |
5846 | struct xfs_bmbt_irec new; /* split extent */ | |
5847 | struct xfs_mount *mp = ip->i_mount; | |
5848 | struct xfs_ifork *ifp; | |
5849 | xfs_fsblock_t gotblkcnt; /* new block count for got */ | |
9788e059 | 5850 | struct xfs_iext_cursor icur; |
19ebedcf DC |
5851 | int error = 0; |
5852 | int logflags = 0; | |
5853 | int i = 0; | |
5854 | ||
5855 | if (unlikely(XFS_TEST_ERROR( | |
5856 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5857 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
e2a190dd | 5858 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
19ebedcf DC |
5859 | XFS_ERROR_REPORT("xfs_bmap_split_extent_at", |
5860 | XFS_ERRLEVEL_LOW, mp); | |
5861 | return -EFSCORRUPTED; | |
5862 | } | |
5863 | ||
5864 | if (XFS_FORCED_SHUTDOWN(mp)) | |
5865 | return -EIO; | |
5866 | ||
5867 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
5868 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
5869 | /* Read in all the extents */ | |
5870 | error = xfs_iread_extents(tp, ip, whichfork); | |
5871 | if (error) | |
5872 | return error; | |
5873 | } | |
5874 | ||
5875 | /* | |
455044a6 | 5876 | * If there are not extents, or split_fsb lies in a hole we are done. |
19ebedcf | 5877 | */ |
9788e059 | 5878 | if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) || |
455044a6 | 5879 | got.br_startoff >= split_fsb) |
19ebedcf DC |
5880 | return 0; |
5881 | ||
5882 | gotblkcnt = split_fsb - got.br_startoff; | |
5883 | new.br_startoff = split_fsb; | |
5884 | new.br_startblock = got.br_startblock + gotblkcnt; | |
5885 | new.br_blockcount = got.br_blockcount - gotblkcnt; | |
5886 | new.br_state = got.br_state; | |
5887 | ||
5888 | if (ifp->if_flags & XFS_IFBROOT) { | |
5889 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
19ebedcf | 5890 | cur->bc_private.b.flags = 0; |
70a93110 | 5891 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
19ebedcf DC |
5892 | if (error) |
5893 | goto del_cursor; | |
5894 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
5895 | } | |
5896 | ||
19ebedcf | 5897 | got.br_blockcount = gotblkcnt; |
9788e059 CH |
5898 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur, |
5899 | &got); | |
19ebedcf DC |
5900 | |
5901 | logflags = XFS_ILOG_CORE; | |
5902 | if (cur) { | |
d0e5f1ff | 5903 | error = xfs_bmbt_update(cur, &got); |
19ebedcf DC |
5904 | if (error) |
5905 | goto del_cursor; | |
5906 | } else | |
5907 | logflags |= XFS_ILOG_DEXT; | |
5908 | ||
5909 | /* Add new extent */ | |
9788e059 | 5910 | xfs_iext_next(ifp, &icur); |
26a75f67 | 5911 | xfs_iext_insert(ip, &icur, &new, 0); |
19ebedcf DC |
5912 | XFS_IFORK_NEXT_SET(ip, whichfork, |
5913 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
5914 | ||
5915 | if (cur) { | |
70a93110 | 5916 | error = xfs_bmbt_lookup_eq(cur, &new, &i); |
19ebedcf DC |
5917 | if (error) |
5918 | goto del_cursor; | |
5919 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor); | |
19ebedcf DC |
5920 | error = xfs_btree_insert(cur, &i); |
5921 | if (error) | |
5922 | goto del_cursor; | |
5923 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
5924 | } | |
5925 | ||
5926 | /* | |
5927 | * Convert to a btree if necessary. | |
5928 | */ | |
5929 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
5930 | int tmp_logflags; /* partial log flag return val */ | |
5931 | ||
5932 | ASSERT(cur == NULL); | |
f7253505 BF |
5933 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
5934 | &tmp_logflags, whichfork); | |
19ebedcf DC |
5935 | logflags |= tmp_logflags; |
5936 | } | |
5937 | ||
5938 | del_cursor: | |
5939 | if (cur) { | |
5940 | cur->bc_private.b.allocated = 0; | |
660265b7 | 5941 | xfs_btree_del_cursor(cur, error); |
19ebedcf DC |
5942 | } |
5943 | ||
5944 | if (logflags) | |
5945 | xfs_trans_log_inode(tp, ip, logflags); | |
5946 | return error; | |
5947 | } | |
5948 | ||
5949 | int | |
5950 | xfs_bmap_split_extent( | |
5951 | struct xfs_inode *ip, | |
5952 | xfs_fileoff_t split_fsb) | |
5953 | { | |
5954 | struct xfs_mount *mp = ip->i_mount; | |
5955 | struct xfs_trans *tp; | |
19ebedcf DC |
5956 | int error; |
5957 | ||
9074815c CH |
5958 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, |
5959 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); | |
5960 | if (error) | |
19ebedcf | 5961 | return error; |
19ebedcf DC |
5962 | |
5963 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5964 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
5965 | ||
312350cd | 5966 | error = xfs_bmap_split_extent_at(tp, ip, split_fsb); |
19ebedcf DC |
5967 | if (error) |
5968 | goto out; | |
5969 | ||
de5a3f46 | 5970 | return xfs_trans_commit(tp); |
19ebedcf DC |
5971 | |
5972 | out: | |
3d7434fe | 5973 | xfs_trans_cancel(tp); |
19ebedcf DC |
5974 | return error; |
5975 | } | |
aeb88300 DW |
5976 | |
5977 | /* Deferred mapping is only for real extents in the data fork. */ | |
5978 | static bool | |
5979 | xfs_bmap_is_update_needed( | |
5980 | struct xfs_bmbt_irec *bmap) | |
5981 | { | |
5982 | return bmap->br_startblock != HOLESTARTBLOCK && | |
5983 | bmap->br_startblock != DELAYSTARTBLOCK; | |
5984 | } | |
5985 | ||
5986 | /* Record a bmap intent. */ | |
5987 | static int | |
5988 | __xfs_bmap_add( | |
21375e5d | 5989 | struct xfs_trans *tp, |
aeb88300 DW |
5990 | enum xfs_bmap_intent_type type, |
5991 | struct xfs_inode *ip, | |
5992 | int whichfork, | |
5993 | struct xfs_bmbt_irec *bmap) | |
5994 | { | |
aeb88300 DW |
5995 | struct xfs_bmap_intent *bi; |
5996 | ||
21375e5d BF |
5997 | trace_xfs_bmap_defer(tp->t_mountp, |
5998 | XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock), | |
aeb88300 | 5999 | type, |
21375e5d | 6000 | XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock), |
aeb88300 DW |
6001 | ip->i_ino, whichfork, |
6002 | bmap->br_startoff, | |
6003 | bmap->br_blockcount, | |
6004 | bmap->br_state); | |
6005 | ||
6006 | bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS); | |
6007 | INIT_LIST_HEAD(&bi->bi_list); | |
6008 | bi->bi_type = type; | |
6009 | bi->bi_owner = ip; | |
6010 | bi->bi_whichfork = whichfork; | |
6011 | bi->bi_bmap = *bmap; | |
6012 | ||
21375e5d | 6013 | xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list); |
aeb88300 DW |
6014 | return 0; |
6015 | } | |
6016 | ||
6017 | /* Map an extent into a file. */ | |
6018 | int | |
6019 | xfs_bmap_map_extent( | |
21375e5d | 6020 | struct xfs_trans *tp, |
aeb88300 DW |
6021 | struct xfs_inode *ip, |
6022 | struct xfs_bmbt_irec *PREV) | |
6023 | { | |
6024 | if (!xfs_bmap_is_update_needed(PREV)) | |
6025 | return 0; | |
6026 | ||
21375e5d | 6027 | return __xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV); |
aeb88300 DW |
6028 | } |
6029 | ||
6030 | /* Unmap an extent out of a file. */ | |
6031 | int | |
6032 | xfs_bmap_unmap_extent( | |
21375e5d | 6033 | struct xfs_trans *tp, |
aeb88300 DW |
6034 | struct xfs_inode *ip, |
6035 | struct xfs_bmbt_irec *PREV) | |
6036 | { | |
6037 | if (!xfs_bmap_is_update_needed(PREV)) | |
6038 | return 0; | |
6039 | ||
21375e5d | 6040 | return __xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV); |
aeb88300 DW |
6041 | } |
6042 | ||
6043 | /* | |
6044 | * Process one of the deferred bmap operations. We pass back the | |
6045 | * btree cursor to maintain our lock on the bmapbt between calls. | |
6046 | */ | |
6047 | int | |
6048 | xfs_bmap_finish_one( | |
6049 | struct xfs_trans *tp, | |
aeb88300 DW |
6050 | struct xfs_inode *ip, |
6051 | enum xfs_bmap_intent_type type, | |
6052 | int whichfork, | |
6053 | xfs_fileoff_t startoff, | |
6054 | xfs_fsblock_t startblock, | |
594956fa | 6055 | xfs_filblks_t *blockcount, |
aeb88300 DW |
6056 | xfs_exntst_t state) |
6057 | { | |
594956fa | 6058 | int error = 0; |
aeb88300 | 6059 | |
35f57661 | 6060 | ASSERT(tp->t_firstblock == NULLFSBLOCK); |
66d19ae1 | 6061 | |
aeb88300 DW |
6062 | trace_xfs_bmap_deferred(tp->t_mountp, |
6063 | XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type, | |
6064 | XFS_FSB_TO_AGBNO(tp->t_mountp, startblock), | |
594956fa | 6065 | ip->i_ino, whichfork, startoff, *blockcount, state); |
aeb88300 | 6066 | |
95855a23 | 6067 | if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK)) |
aeb88300 | 6068 | return -EFSCORRUPTED; |
aeb88300 DW |
6069 | |
6070 | if (XFS_TEST_ERROR(false, tp->t_mountp, | |
e2a190dd | 6071 | XFS_ERRTAG_BMAP_FINISH_ONE)) |
aeb88300 DW |
6072 | return -EIO; |
6073 | ||
6074 | switch (type) { | |
6075 | case XFS_BMAP_MAP: | |
594956fa | 6076 | error = xfs_bmapi_remap(tp, ip, startoff, *blockcount, |
b5d7ff9a | 6077 | startblock, 0); |
594956fa | 6078 | *blockcount = 0; |
aeb88300 DW |
6079 | break; |
6080 | case XFS_BMAP_UNMAP: | |
594956fa | 6081 | error = __xfs_bunmapi(tp, ip, startoff, blockcount, |
d3c5f3dd | 6082 | XFS_BMAPI_REMAP, 1); |
aeb88300 DW |
6083 | break; |
6084 | default: | |
6085 | ASSERT(0); | |
6086 | error = -EFSCORRUPTED; | |
6087 | } | |
6088 | ||
6089 | return error; | |
6090 | } | |
0cf6a3a9 DW |
6091 | |
6092 | /* Check that an inode's extent does not have invalid flags or bad ranges. */ | |
6093 | xfs_failaddr_t | |
6094 | xfs_bmap_validate_extent( | |
6095 | struct xfs_inode *ip, | |
6096 | int whichfork, | |
6097 | struct xfs_bmbt_irec *irec) | |
6098 | { | |
6099 | struct xfs_mount *mp = ip->i_mount; | |
6100 | xfs_fsblock_t endfsb; | |
6101 | bool isrt; | |
6102 | ||
6103 | isrt = XFS_IS_REALTIME_INODE(ip); | |
6104 | endfsb = irec->br_startblock + irec->br_blockcount - 1; | |
6105 | if (isrt) { | |
6106 | if (!xfs_verify_rtbno(mp, irec->br_startblock)) | |
6107 | return __this_address; | |
6108 | if (!xfs_verify_rtbno(mp, endfsb)) | |
6109 | return __this_address; | |
6110 | } else { | |
6111 | if (!xfs_verify_fsbno(mp, irec->br_startblock)) | |
6112 | return __this_address; | |
6113 | if (!xfs_verify_fsbno(mp, endfsb)) | |
6114 | return __this_address; | |
6115 | if (XFS_FSB_TO_AGNO(mp, irec->br_startblock) != | |
6116 | XFS_FSB_TO_AGNO(mp, endfsb)) | |
6117 | return __this_address; | |
6118 | } | |
b7fdeb4d CH |
6119 | if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK) |
6120 | return __this_address; | |
0cf6a3a9 DW |
6121 | return NULL; |
6122 | } |