]>
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_dir2.h" |
17 | #include "xfs_inode.h" | |
18 | #include "xfs_btree.h" | |
19 | #include "xfs_trans.h" | |
20 | #include "xfs_alloc.h" | |
21 | #include "xfs_bmap.h" | |
22 | #include "xfs_bmap_btree.h" | |
2cf10e4c | 23 | #include "xfs_errortag.h" |
b626fb59 DC |
24 | #include "xfs_trans_space.h" |
25 | #include "xfs_trace.h" | |
26 | #include "xfs_attr_leaf.h" | |
27 | #include "xfs_quota_defs.h" | |
85aec44f | 28 | #include "xfs_rmap.h" |
f93d2173 | 29 | #include "xfs_ag.h" |
cf8ce220 | 30 | #include "xfs_ag_resv.h" |
cfe32f0d | 31 | #include "xfs_refcount.h" |
b626fb59 | 32 | |
1577541c | 33 | struct kmem_cache *xfs_bmap_intent_cache; |
5e656dbb BN |
34 | |
35 | /* | |
49f693fa | 36 | * Miscellaneous helper functions |
5e656dbb | 37 | */ |
5e656dbb | 38 | |
5e656dbb | 39 | /* |
49f693fa DC |
40 | * Compute and fill in the value of the maximum depth of a bmap btree |
41 | * in this filesystem. Done once, during mount. | |
5e656dbb | 42 | */ |
49f693fa DC |
43 | void |
44 | xfs_bmap_compute_maxlevels( | |
45 | xfs_mount_t *mp, /* file system mount structure */ | |
46 | int whichfork) /* data or attr fork */ | |
47 | { | |
48 | int level; /* btree level */ | |
49 | uint maxblocks; /* max blocks at this level */ | |
50 | uint maxleafents; /* max leaf entries possible */ | |
51 | int maxrootrecs; /* max records in root block */ | |
52 | int minleafrecs; /* min records in leaf block */ | |
53 | int minnoderecs; /* min records in node block */ | |
54 | int sz; /* root block size */ | |
5e656dbb | 55 | |
49f693fa | 56 | /* |
87c472b7 CH |
57 | * The maximum number of extents in a file, hence the maximum number of |
58 | * leaf entries, is controlled by the size of the on-disk extent count, | |
59 | * either a signed 32-bit number for the data fork, or a signed 16-bit | |
60 | * number for the attr fork. | |
49f693fa | 61 | * |
073f2424 CH |
62 | * Note that we can no longer assume that if we are in ATTR1 that the |
63 | * fork offset of all the inodes will be | |
64 | * (xfs_default_attroffset(ip) >> 3) because we could have mounted with | |
65 | * ATTR2 and then mounted back with ATTR1, keeping the i_forkoff's fixed | |
66 | * but probably at various positions. Therefore, for both ATTR1 and | |
67 | * ATTR2 we have to assume the worst case scenario of a minimum size | |
68 | * available. | |
49f693fa DC |
69 | */ |
70 | if (whichfork == XFS_DATA_FORK) { | |
71 | maxleafents = MAXEXTNUM; | |
72 | sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS); | |
73 | } else { | |
74 | maxleafents = MAXAEXTNUM; | |
75 | sz = XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
76 | } | |
ff105f75 | 77 | maxrootrecs = xfs_bmdr_maxrecs(sz, 0); |
49f693fa DC |
78 | minleafrecs = mp->m_bmap_dmnr[0]; |
79 | minnoderecs = mp->m_bmap_dmnr[1]; | |
80 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
81 | for (level = 1; maxblocks > 1; level++) { | |
82 | if (maxblocks <= maxrootrecs) | |
83 | maxblocks = 1; | |
84 | else | |
85 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
86 | } | |
87 | mp->m_bm_maxlevels[whichfork] = level; | |
441815c7 | 88 | ASSERT(mp->m_bm_maxlevels[whichfork] <= xfs_bmbt_maxlevels_ondisk()); |
49f693fa | 89 | } |
5e656dbb | 90 | |
4acdeb81 DC |
91 | unsigned int |
92 | xfs_bmap_compute_attr_offset( | |
93 | struct xfs_mount *mp) | |
94 | { | |
95 | if (mp->m_sb.sb_inodesize == 256) | |
96 | return XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
97 | return XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); | |
98 | } | |
99 | ||
b194c7d8 BN |
100 | STATIC int /* error */ |
101 | xfs_bmbt_lookup_eq( | |
102 | struct xfs_btree_cur *cur, | |
70a93110 | 103 | struct xfs_bmbt_irec *irec, |
b194c7d8 BN |
104 | int *stat) /* success/failure */ |
105 | { | |
70a93110 | 106 | cur->bc_rec.b = *irec; |
b194c7d8 BN |
107 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); |
108 | } | |
109 | ||
110 | STATIC int /* error */ | |
4f76f49c | 111 | xfs_bmbt_lookup_first( |
b194c7d8 | 112 | struct xfs_btree_cur *cur, |
b194c7d8 BN |
113 | int *stat) /* success/failure */ |
114 | { | |
4f76f49c CH |
115 | cur->bc_rec.b.br_startoff = 0; |
116 | cur->bc_rec.b.br_startblock = 0; | |
117 | cur->bc_rec.b.br_blockcount = 0; | |
b194c7d8 BN |
118 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); |
119 | } | |
120 | ||
121 | /* | |
a2ceac1f DC |
122 | * Check if the inode needs to be converted to btree format. |
123 | */ | |
124 | static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork) | |
125 | { | |
87c472b7 CH |
126 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
127 | ||
1277a5e0 | 128 | return whichfork != XFS_COW_FORK && |
d967a68d | 129 | ifp->if_format == XFS_DINODE_FMT_EXTENTS && |
87c472b7 | 130 | ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork); |
a2ceac1f DC |
131 | } |
132 | ||
133 | /* | |
134 | * Check if the inode should be converted to extent format. | |
135 | */ | |
136 | static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork) | |
137 | { | |
87c472b7 CH |
138 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
139 | ||
1277a5e0 | 140 | return whichfork != XFS_COW_FORK && |
d967a68d | 141 | ifp->if_format == XFS_DINODE_FMT_BTREE && |
87c472b7 | 142 | ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork); |
a2ceac1f DC |
143 | } |
144 | ||
145 | /* | |
d0e5f1ff | 146 | * Update the record referred to by cur to the value given by irec |
b194c7d8 BN |
147 | * This either works (return 0) or gets an EFSCORRUPTED error. |
148 | */ | |
149 | STATIC int | |
150 | xfs_bmbt_update( | |
151 | struct xfs_btree_cur *cur, | |
d0e5f1ff | 152 | struct xfs_bmbt_irec *irec) |
b194c7d8 BN |
153 | { |
154 | union xfs_btree_rec rec; | |
155 | ||
d0e5f1ff | 156 | xfs_bmbt_disk_set_all(&rec.bmbt, irec); |
b194c7d8 BN |
157 | return xfs_btree_update(cur, &rec); |
158 | } | |
159 | ||
5e656dbb | 160 | /* |
49f693fa DC |
161 | * Compute the worst-case number of indirect blocks that will be used |
162 | * for ip's delayed extent of length "len". | |
5e656dbb | 163 | */ |
49f693fa DC |
164 | STATIC xfs_filblks_t |
165 | xfs_bmap_worst_indlen( | |
166 | xfs_inode_t *ip, /* incore inode pointer */ | |
167 | xfs_filblks_t len) /* delayed extent length */ | |
57c9fccb | 168 | { |
49f693fa DC |
169 | int level; /* btree level number */ |
170 | int maxrecs; /* maximum record count at this level */ | |
171 | xfs_mount_t *mp; /* mount structure */ | |
172 | xfs_filblks_t rval; /* return value */ | |
57c9fccb NS |
173 | |
174 | mp = ip->i_mount; | |
49f693fa DC |
175 | maxrecs = mp->m_bmap_dmxr[0]; |
176 | for (level = 0, rval = 0; | |
177 | level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK); | |
178 | level++) { | |
179 | len += maxrecs - 1; | |
180 | do_div(len, maxrecs); | |
181 | rval += len; | |
7fbe9b54 DW |
182 | if (len == 1) |
183 | return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) - | |
49f693fa DC |
184 | level - 1; |
185 | if (level == 0) | |
186 | maxrecs = mp->m_bmap_dmxr[1]; | |
57c9fccb | 187 | } |
49f693fa | 188 | return rval; |
57c9fccb NS |
189 | } |
190 | ||
191 | /* | |
49f693fa | 192 | * Calculate the default attribute fork offset for newly created inodes. |
57c9fccb | 193 | */ |
49f693fa DC |
194 | uint |
195 | xfs_default_attroffset( | |
196 | struct xfs_inode *ip) | |
57c9fccb | 197 | { |
c06c9d5a DC |
198 | if (ip->i_df.if_format == XFS_DINODE_FMT_DEV) |
199 | return roundup(sizeof(xfs_dev_t), 8); | |
4acdeb81 | 200 | return M_IGEO(ip->i_mount)->attr_fork_offset; |
57c9fccb NS |
201 | } |
202 | ||
203 | /* | |
073f2424 CH |
204 | * Helper routine to reset inode i_forkoff field when switching attribute fork |
205 | * from local to extent format - we reset it where possible to make space | |
206 | * available for inline data fork extents. | |
57c9fccb | 207 | */ |
49f693fa DC |
208 | STATIC void |
209 | xfs_bmap_forkoff_reset( | |
49f693fa DC |
210 | xfs_inode_t *ip, |
211 | int whichfork) | |
57c9fccb | 212 | { |
49f693fa | 213 | if (whichfork == XFS_ATTR_FORK && |
d967a68d CH |
214 | ip->i_df.if_format != XFS_DINODE_FMT_DEV && |
215 | ip->i_df.if_format != XFS_DINODE_FMT_BTREE) { | |
49f693fa | 216 | uint dfl_forkoff = xfs_default_attroffset(ip) >> 3; |
57c9fccb | 217 | |
073f2424 CH |
218 | if (dfl_forkoff > ip->i_forkoff) |
219 | ip->i_forkoff = dfl_forkoff; | |
49f693fa | 220 | } |
57c9fccb NS |
221 | } |
222 | ||
49f693fa DC |
223 | #ifdef DEBUG |
224 | STATIC struct xfs_buf * | |
225 | xfs_bmap_get_bp( | |
226 | struct xfs_btree_cur *cur, | |
227 | xfs_fsblock_t bno) | |
228 | { | |
2fdd378a | 229 | struct xfs_log_item *lip; |
49f693fa | 230 | int i; |
56b2de80 | 231 | |
49f693fa DC |
232 | if (!cur) |
233 | return NULL; | |
2bd0ea18 | 234 | |
100a1b52 | 235 | for (i = 0; i < cur->bc_maxlevels; i++) { |
5df9b067 | 236 | if (!cur->bc_levels[i].bp) |
49f693fa | 237 | break; |
5df9b067 DW |
238 | if (xfs_buf_daddr(cur->bc_levels[i].bp) == bno) |
239 | return cur->bc_levels[i].bp; | |
49f693fa | 240 | } |
56b2de80 | 241 | |
49f693fa | 242 | /* Chase down all the log items to see if the bp is there */ |
2fdd378a DC |
243 | list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) { |
244 | struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip; | |
245 | ||
49f693fa | 246 | if (bip->bli_item.li_type == XFS_LI_BUF && |
d4aaa66b | 247 | xfs_buf_daddr(bip->bli_buf) == bno) |
49f693fa DC |
248 | return bip->bli_buf; |
249 | } | |
2bd0ea18 | 250 | |
49f693fa DC |
251 | return NULL; |
252 | } | |
56b2de80 | 253 | |
49f693fa DC |
254 | STATIC void |
255 | xfs_check_block( | |
256 | struct xfs_btree_block *block, | |
257 | xfs_mount_t *mp, | |
258 | int root, | |
259 | short sz) | |
260 | { | |
261 | int i, j, dmxr; | |
262 | __be64 *pp, *thispa; /* pointer to block address */ | |
263 | xfs_bmbt_key_t *prevp, *keyp; | |
2bd0ea18 | 264 | |
49f693fa | 265 | ASSERT(be16_to_cpu(block->bb_level) > 0); |
56b2de80 | 266 | |
49f693fa DC |
267 | prevp = NULL; |
268 | for( i = 1; i <= xfs_btree_get_numrecs(block); i++) { | |
269 | dmxr = mp->m_bmap_dmxr[0]; | |
270 | keyp = XFS_BMBT_KEY_ADDR(mp, block, i); | |
a2ceac1f | 271 | |
49f693fa DC |
272 | if (prevp) { |
273 | ASSERT(be64_to_cpu(prevp->br_startoff) < | |
274 | be64_to_cpu(keyp->br_startoff)); | |
275 | } | |
276 | prevp = keyp; | |
2bd0ea18 | 277 | |
2bd0ea18 | 278 | /* |
49f693fa | 279 | * Compare the block numbers to see if there are dups. |
2bd0ea18 | 280 | */ |
49f693fa DC |
281 | if (root) |
282 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz); | |
283 | else | |
284 | pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr); | |
285 | ||
286 | for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) { | |
287 | if (root) | |
288 | thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz); | |
289 | else | |
290 | thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr); | |
291 | if (*thispa == *pp) { | |
292 | xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld", | |
293 | __func__, j, i, | |
294 | (unsigned long long)be64_to_cpu(*thispa)); | |
4d5e2888 | 295 | xfs_err(mp, "%s: ptrs are equal in node\n", |
49f693fa | 296 | __func__); |
4d5e2888 | 297 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
49f693fa | 298 | } |
2bd0ea18 | 299 | } |
49f693fa DC |
300 | } |
301 | } | |
a2ceac1f | 302 | |
49f693fa DC |
303 | /* |
304 | * Check that the extents for the inode ip are in the right order in all | |
f07ae2a6 DC |
305 | * btree leaves. THis becomes prohibitively expensive for large extent count |
306 | * files, so don't bother with inodes that have more than 10,000 extents in | |
307 | * them. The btree record ordering checks will still be done, so for such large | |
308 | * bmapbt constructs that is going to catch most corruptions. | |
49f693fa | 309 | */ |
49f693fa DC |
310 | STATIC void |
311 | xfs_bmap_check_leaf_extents( | |
ec924d04 | 312 | struct xfs_btree_cur *cur, /* btree cursor or null */ |
49f693fa DC |
313 | xfs_inode_t *ip, /* incore inode pointer */ |
314 | int whichfork) /* data or attr fork */ | |
315 | { | |
d967a68d CH |
316 | struct xfs_mount *mp = ip->i_mount; |
317 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
49f693fa DC |
318 | struct xfs_btree_block *block; /* current btree block */ |
319 | xfs_fsblock_t bno; /* block # of "block" */ | |
167137fe | 320 | struct xfs_buf *bp; /* buffer for "block" */ |
49f693fa DC |
321 | int error; /* error return value */ |
322 | xfs_extnum_t i=0, j; /* index into the extents list */ | |
49f693fa | 323 | int level; /* btree level, for checking */ |
49f693fa DC |
324 | __be64 *pp; /* pointer to block address */ |
325 | xfs_bmbt_rec_t *ep; /* pointer to current extent */ | |
326 | xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */ | |
327 | xfs_bmbt_rec_t *nextp; /* pointer to next extent */ | |
328 | int bp_release = 0; | |
329 | ||
d967a68d | 330 | if (ifp->if_format != XFS_DINODE_FMT_BTREE) |
49f693fa | 331 | return; |
49f693fa | 332 | |
f07ae2a6 | 333 | /* skip large extent count inodes */ |
87c472b7 | 334 | if (ip->i_df.if_nextents > 10000) |
f07ae2a6 DC |
335 | return; |
336 | ||
49f693fa | 337 | bno = NULLFSBLOCK; |
49f693fa DC |
338 | block = ifp->if_broot; |
339 | /* | |
340 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
341 | */ | |
342 | level = be16_to_cpu(block->bb_level); | |
343 | ASSERT(level > 0); | |
344 | xfs_check_block(block, mp, 1, ifp->if_broot_bytes); | |
345 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
346 | bno = be64_to_cpu(*pp); | |
347 | ||
5a35bf2c | 348 | ASSERT(bno != NULLFSBLOCK); |
49f693fa DC |
349 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
350 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
351 | ||
352 | /* | |
353 | * Go down the tree until leaf level is reached, following the first | |
354 | * pointer (leftmost) at each level. | |
355 | */ | |
356 | while (level-- > 0) { | |
357 | /* See if buf is in cur first */ | |
358 | bp_release = 0; | |
359 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
360 | if (!bp) { | |
361 | bp_release = 1; | |
4aa01a59 | 362 | error = xfs_btree_read_bufl(mp, NULL, bno, &bp, |
49f693fa DC |
363 | XFS_BMAP_BTREE_REF, |
364 | &xfs_bmbt_buf_ops); | |
2bd0ea18 | 365 | if (error) |
49f693fa | 366 | goto error_norelse; |
2bd0ea18 | 367 | } |
49f693fa | 368 | block = XFS_BUF_TO_BLOCK(bp); |
49f693fa DC |
369 | if (level == 0) |
370 | break; | |
2bd0ea18 | 371 | |
2bd0ea18 | 372 | /* |
49f693fa DC |
373 | * Check this block for basic sanity (increasing keys and |
374 | * no duplicate blocks). | |
2bd0ea18 | 375 | */ |
49f693fa DC |
376 | |
377 | xfs_check_block(block, mp, 0, 0); | |
378 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
379 | bno = be64_to_cpu(*pp); | |
fbb4fa7f DW |
380 | if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) { |
381 | error = -EFSCORRUPTED; | |
382 | goto error0; | |
383 | } | |
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; | |
4aa01a59 | 448 | error = xfs_btree_read_bufl(mp, NULL, bno, &bp, |
49f693fa DC |
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 | * Inode fork format manipulation functions | |
519 | */ | |
a2ceac1f | 520 | |
49f693fa | 521 | /* |
939ebc1a CH |
522 | * Convert the inode format to extent format if it currently is in btree format, |
523 | * but the extent list is small enough that it fits into the extent format. | |
524 | * | |
525 | * Since the extents are already in-core, all we have to do is give up the space | |
526 | * for the btree root and pitch the leaf block. | |
49f693fa DC |
527 | */ |
528 | STATIC int /* error */ | |
529 | xfs_bmap_btree_to_extents( | |
939ebc1a CH |
530 | struct xfs_trans *tp, /* transaction pointer */ |
531 | struct xfs_inode *ip, /* incore inode pointer */ | |
532 | struct xfs_btree_cur *cur, /* btree cursor */ | |
49f693fa DC |
533 | int *logflagsp, /* inode logging flags */ |
534 | int whichfork) /* data or attr fork */ | |
535 | { | |
939ebc1a CH |
536 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
537 | struct xfs_mount *mp = ip->i_mount; | |
538 | struct xfs_btree_block *rblock = ifp->if_broot; | |
49f693fa DC |
539 | struct xfs_btree_block *cblock;/* child btree block */ |
540 | xfs_fsblock_t cbno; /* child block number */ | |
167137fe | 541 | struct xfs_buf *cbp; /* child block's buffer */ |
49f693fa | 542 | int error; /* error return value */ |
49f693fa | 543 | __be64 *pp; /* ptr to block address */ |
85aec44f | 544 | struct xfs_owner_info oinfo; |
56b2de80 | 545 | |
939ebc1a CH |
546 | /* check if we actually need the extent format first: */ |
547 | if (!xfs_bmap_wants_extents(ip, whichfork)) | |
548 | return 0; | |
549 | ||
550 | ASSERT(cur); | |
1277a5e0 | 551 | ASSERT(whichfork != XFS_COW_FORK); |
d967a68d | 552 | ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE); |
49f693fa DC |
553 | ASSERT(be16_to_cpu(rblock->bb_level) == 1); |
554 | ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1); | |
555 | ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1); | |
939ebc1a | 556 | |
49f693fa DC |
557 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes); |
558 | cbno = be64_to_cpu(*pp); | |
49f693fa | 559 | #ifdef DEBUG |
fbb4fa7f DW |
560 | if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_btree_check_lptr(cur, cbno, 1))) |
561 | return -EFSCORRUPTED; | |
49f693fa | 562 | #endif |
4aa01a59 | 563 | error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF, |
49f693fa DC |
564 | &xfs_bmbt_buf_ops); |
565 | if (error) | |
566 | return error; | |
567 | cblock = XFS_BUF_TO_BLOCK(cbp); | |
568 | if ((error = xfs_btree_check_block(cur, cblock, 0, cbp))) | |
569 | return error; | |
85aec44f | 570 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); |
7d84b02d | 571 | xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo); |
aa00f286 | 572 | ip->i_nblocks--; |
49f693fa DC |
573 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); |
574 | xfs_trans_binval(tp, cbp); | |
5df9b067 DW |
575 | if (cur->bc_levels[0].bp == cbp) |
576 | cur->bc_levels[0].bp = NULL; | |
49f693fa DC |
577 | xfs_iroot_realloc(ip, -1, whichfork); |
578 | ASSERT(ifp->if_broot == NULL); | |
d967a68d | 579 | ifp->if_format = XFS_DINODE_FMT_EXTENTS; |
939ebc1a | 580 | *logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
49f693fa DC |
581 | return 0; |
582 | } | |
2bd0ea18 NS |
583 | |
584 | /* | |
49f693fa DC |
585 | * Convert an extents-format file into a btree-format file. |
586 | * The new file will have a root block (in the inode) and a single child block. | |
2bd0ea18 | 587 | */ |
49f693fa DC |
588 | STATIC int /* error */ |
589 | xfs_bmap_extents_to_btree( | |
52f6ed9d BF |
590 | struct xfs_trans *tp, /* transaction pointer */ |
591 | struct xfs_inode *ip, /* incore inode pointer */ | |
52f6ed9d | 592 | struct xfs_btree_cur **curp, /* cursor returned to caller */ |
49f693fa DC |
593 | int wasdel, /* converting a delayed alloc */ |
594 | int *logflagsp, /* inode logging flags */ | |
595 | int whichfork) /* data or attr fork */ | |
2bd0ea18 | 596 | { |
49f693fa | 597 | struct xfs_btree_block *ablock; /* allocated (child) bt block */ |
52f6ed9d BF |
598 | struct xfs_buf *abp; /* buffer for ablock */ |
599 | struct xfs_alloc_arg args; /* allocation arguments */ | |
600 | struct xfs_bmbt_rec *arp; /* child record pointer */ | |
49f693fa | 601 | struct xfs_btree_block *block; /* btree root block */ |
52f6ed9d | 602 | struct xfs_btree_cur *cur; /* bmap btree cursor */ |
49f693fa | 603 | int error; /* error return value */ |
52f6ed9d BF |
604 | struct xfs_ifork *ifp; /* inode fork pointer */ |
605 | struct xfs_bmbt_key *kp; /* root block key pointer */ | |
606 | struct xfs_mount *mp; /* mount structure */ | |
49f693fa | 607 | xfs_bmbt_ptr_t *pp; /* root block address pointer */ |
9788e059 | 608 | struct xfs_iext_cursor icur; |
62ed7338 | 609 | struct xfs_bmbt_irec rec; |
9788e059 | 610 | xfs_extnum_t cnt = 0; |
2bd0ea18 | 611 | |
5dfa5cd2 | 612 | mp = ip->i_mount; |
1277a5e0 | 613 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa | 614 | ifp = XFS_IFORK_PTR(ip, whichfork); |
d967a68d | 615 | ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS); |
56b2de80 | 616 | |
2bd0ea18 | 617 | /* |
d7c2f6cb DC |
618 | * Make space in the inode incore. This needs to be undone if we fail |
619 | * to expand the root. | |
2bd0ea18 | 620 | */ |
49f693fa | 621 | xfs_iroot_realloc(ip, 1, whichfork); |
56b2de80 | 622 | |
2bd0ea18 | 623 | /* |
49f693fa | 624 | * Fill in the root. |
2bd0ea18 | 625 | */ |
49f693fa | 626 | block = ifp->if_broot; |
e394a4b1 ES |
627 | xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL, |
628 | XFS_BTNUM_BMAP, 1, 1, ip->i_ino, | |
f4241a08 | 629 | XFS_BTREE_LONG_PTRS); |
49f693fa DC |
630 | /* |
631 | * Need a cursor. Can't allocate until bb_level is filled in. | |
632 | */ | |
49f693fa | 633 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
bbf943f8 | 634 | cur->bc_ino.flags = wasdel ? XFS_BTCUR_BMBT_WASDEL : 0; |
49f693fa DC |
635 | /* |
636 | * Convert to a btree with two levels, one record in root. | |
637 | */ | |
d967a68d | 638 | ifp->if_format = XFS_DINODE_FMT_BTREE; |
49f693fa DC |
639 | memset(&args, 0, sizeof(args)); |
640 | args.tp = tp; | |
641 | args.mp = mp; | |
85aec44f | 642 | xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork); |
f7253505 | 643 | if (tp->t_firstblock == NULLFSBLOCK) { |
49f693fa DC |
644 | args.type = XFS_ALLOCTYPE_START_BNO; |
645 | args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino); | |
565e96c6 | 646 | } else if (tp->t_flags & XFS_TRANS_LOWMODE) { |
49f693fa | 647 | args.type = XFS_ALLOCTYPE_START_BNO; |
f7253505 | 648 | args.fsbno = tp->t_firstblock; |
49f693fa DC |
649 | } else { |
650 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
f7253505 | 651 | args.fsbno = tp->t_firstblock; |
49f693fa DC |
652 | } |
653 | args.minlen = args.maxlen = args.prod = 1; | |
654 | args.wasdel = wasdel; | |
655 | *logflagsp = 0; | |
d7c2f6cb DC |
656 | error = xfs_alloc_vextent(&args); |
657 | if (error) | |
658 | goto out_root_realloc; | |
68d06dce | 659 | |
c1587ecf | 660 | if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) { |
b5c71bcd | 661 | error = -ENOSPC; |
d7c2f6cb | 662 | goto out_root_realloc; |
c1587ecf | 663 | } |
d7c2f6cb | 664 | |
2bd0ea18 | 665 | /* |
49f693fa | 666 | * Allocation can't fail, the space was reserved. |
2bd0ea18 | 667 | */ |
f7253505 BF |
668 | ASSERT(tp->t_firstblock == NULLFSBLOCK || |
669 | args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock)); | |
f44a348b | 670 | tp->t_firstblock = args.fsbno; |
116c6a88 | 671 | cur->bc_ino.allocated++; |
aa00f286 | 672 | ip->i_nblocks++; |
49f693fa | 673 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L); |
78fcd346 DW |
674 | error = xfs_trans_get_buf(tp, mp->m_ddev_targp, |
675 | XFS_FSB_TO_DADDR(mp, args.fsbno), | |
676 | mp->m_bsize, 0, &abp); | |
677 | if (error) | |
d7c2f6cb | 678 | goto out_unreserve_dquot; |
d7c2f6cb | 679 | |
2bd0ea18 | 680 | /* |
49f693fa | 681 | * Fill in the child block. |
2bd0ea18 | 682 | */ |
49f693fa DC |
683 | abp->b_ops = &xfs_bmbt_buf_ops; |
684 | ablock = XFS_BUF_TO_BLOCK(abp); | |
f1208396 | 685 | xfs_btree_init_block_int(mp, ablock, xfs_buf_daddr(abp), |
e394a4b1 | 686 | XFS_BTNUM_BMAP, 0, 0, ip->i_ino, |
5dfa5cd2 DC |
687 | XFS_BTREE_LONG_PTRS); |
688 | ||
9788e059 | 689 | for_each_xfs_iext(ifp, &icur, &rec) { |
62ed7338 CH |
690 | if (isnullstartblock(rec.br_startblock)) |
691 | continue; | |
692 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt); | |
693 | xfs_bmbt_disk_set_all(arp, &rec); | |
694 | cnt++; | |
49f693fa | 695 | } |
87c472b7 | 696 | ASSERT(cnt == ifp->if_nextents); |
49f693fa | 697 | xfs_btree_set_numrecs(ablock, cnt); |
56b2de80 | 698 | |
49f693fa DC |
699 | /* |
700 | * Fill in the root key and pointer. | |
701 | */ | |
702 | kp = XFS_BMBT_KEY_ADDR(mp, block, 1); | |
703 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); | |
704 | kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp)); | |
705 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur, | |
706 | be16_to_cpu(block->bb_level))); | |
707 | *pp = cpu_to_be64(args.fsbno); | |
2bd0ea18 | 708 | |
49f693fa DC |
709 | /* |
710 | * Do all this logging at the end so that | |
711 | * the root is at the right level. | |
712 | */ | |
5dfa5cd2 | 713 | xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS); |
613e6057 | 714 | xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs)); |
49f693fa DC |
715 | ASSERT(*curp == NULL); |
716 | *curp = cur; | |
717 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork); | |
718 | return 0; | |
b5c71bcd | 719 | |
d7c2f6cb | 720 | out_unreserve_dquot: |
b5c71bcd | 721 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); |
d7c2f6cb | 722 | out_root_realloc: |
b5c71bcd | 723 | xfs_iroot_realloc(ip, -1, whichfork); |
d967a68d | 724 | ifp->if_format = XFS_DINODE_FMT_EXTENTS; |
d7c2f6cb | 725 | ASSERT(ifp->if_broot == NULL); |
b5c71bcd SH |
726 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
727 | ||
728 | return error; | |
49f693fa | 729 | } |
a2ceac1f | 730 | |
49f693fa DC |
731 | /* |
732 | * Convert a local file to an extents file. | |
733 | * This code is out of bounds for data forks of regular files, | |
734 | * since the file data needs to get logged so things will stay consistent. | |
735 | * (The bmap-level manipulations are ok, though). | |
736 | */ | |
3f17ed4b DC |
737 | void |
738 | xfs_bmap_local_to_extents_empty( | |
feee8e52 | 739 | struct xfs_trans *tp, |
3f17ed4b DC |
740 | struct xfs_inode *ip, |
741 | int whichfork) | |
742 | { | |
743 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
744 | ||
1277a5e0 | 745 | ASSERT(whichfork != XFS_COW_FORK); |
d967a68d | 746 | ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL); |
3f17ed4b | 747 | ASSERT(ifp->if_bytes == 0); |
87c472b7 | 748 | ASSERT(ifp->if_nextents == 0); |
3f17ed4b | 749 | |
ff105f75 | 750 | xfs_bmap_forkoff_reset(ip, whichfork); |
b37d753d CH |
751 | ifp->if_u1.if_root = NULL; |
752 | ifp->if_height = 0; | |
d967a68d | 753 | ifp->if_format = XFS_DINODE_FMT_EXTENTS; |
feee8e52 | 754 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
3f17ed4b DC |
755 | } |
756 | ||
757 | ||
49f693fa DC |
758 | STATIC int /* error */ |
759 | xfs_bmap_local_to_extents( | |
760 | xfs_trans_t *tp, /* transaction pointer */ | |
761 | xfs_inode_t *ip, /* incore inode pointer */ | |
49f693fa DC |
762 | xfs_extlen_t total, /* total blocks needed by transaction */ |
763 | int *logflagsp, /* inode logging flags */ | |
764 | int whichfork, | |
5dfa5cd2 DC |
765 | void (*init_fn)(struct xfs_trans *tp, |
766 | struct xfs_buf *bp, | |
49f693fa DC |
767 | struct xfs_inode *ip, |
768 | struct xfs_ifork *ifp)) | |
769 | { | |
3f17ed4b | 770 | int error = 0; |
49f693fa | 771 | int flags; /* logging flags returned */ |
e07055b8 | 772 | struct xfs_ifork *ifp; /* inode fork pointer */ |
3f17ed4b | 773 | xfs_alloc_arg_t args; /* allocation arguments */ |
167137fe | 774 | struct xfs_buf *bp; /* buffer for extent block */ |
05104e43 | 775 | struct xfs_bmbt_irec rec; |
9788e059 | 776 | struct xfs_iext_cursor icur; |
2bd0ea18 | 777 | |
49f693fa DC |
778 | /* |
779 | * We don't want to deal with the case of keeping inode data inline yet. | |
780 | * So sending the data fork of a regular inode is invalid. | |
781 | */ | |
e37bf53c | 782 | ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK)); |
49f693fa | 783 | ifp = XFS_IFORK_PTR(ip, whichfork); |
d967a68d | 784 | ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL); |
3f17ed4b DC |
785 | |
786 | if (!ifp->if_bytes) { | |
feee8e52 | 787 | xfs_bmap_local_to_extents_empty(tp, ip, whichfork); |
3f17ed4b DC |
788 | flags = XFS_ILOG_CORE; |
789 | goto done; | |
790 | } | |
791 | ||
49f693fa DC |
792 | flags = 0; |
793 | error = 0; | |
3f17ed4b DC |
794 | memset(&args, 0, sizeof(args)); |
795 | args.tp = tp; | |
796 | args.mp = ip->i_mount; | |
85aec44f | 797 | xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0); |
3f17ed4b DC |
798 | /* |
799 | * Allocate a block. We know we need only one, since the | |
800 | * file currently fits in an inode. | |
801 | */ | |
f7253505 | 802 | if (tp->t_firstblock == NULLFSBLOCK) { |
3f17ed4b DC |
803 | args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino); |
804 | args.type = XFS_ALLOCTYPE_START_BNO; | |
49f693fa | 805 | } else { |
f7253505 | 806 | args.fsbno = tp->t_firstblock; |
3f17ed4b | 807 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
2bd0ea18 | 808 | } |
3f17ed4b DC |
809 | args.total = total; |
810 | args.minlen = args.maxlen = args.prod = 1; | |
811 | error = xfs_alloc_vextent(&args); | |
812 | if (error) | |
813 | goto done; | |
814 | ||
815 | /* Can't fail, the space was reserved. */ | |
816 | ASSERT(args.fsbno != NULLFSBLOCK); | |
817 | ASSERT(args.len == 1); | |
f7253505 | 818 | tp->t_firstblock = args.fsbno; |
78fcd346 DW |
819 | error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp, |
820 | XFS_FSB_TO_DADDR(args.mp, args.fsbno), | |
821 | args.mp->m_bsize, 0, &bp); | |
822 | if (error) | |
823 | goto done; | |
3f17ed4b | 824 | |
19ebedcf | 825 | /* |
f44fbde0 | 826 | * Initialize the block, copy the data and log the remote buffer. |
19ebedcf | 827 | * |
f44fbde0 BF |
828 | * The callout is responsible for logging because the remote format |
829 | * might differ from the local format and thus we don't know how much to | |
830 | * log here. Note that init_fn must also set the buffer log item type | |
831 | * correctly. | |
19ebedcf | 832 | */ |
3f17ed4b DC |
833 | init_fn(tp, bp, ip, ifp); |
834 | ||
f44fbde0 | 835 | /* account for the change in fork size */ |
3f17ed4b | 836 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); |
feee8e52 | 837 | xfs_bmap_local_to_extents_empty(tp, ip, whichfork); |
49f693fa | 838 | flags |= XFS_ILOG_CORE; |
3f17ed4b | 839 | |
b37d753d CH |
840 | ifp->if_u1.if_root = NULL; |
841 | ifp->if_height = 0; | |
842 | ||
05104e43 CH |
843 | rec.br_startoff = 0; |
844 | rec.br_startblock = args.fsbno; | |
845 | rec.br_blockcount = 1; | |
846 | rec.br_state = XFS_EXT_NORM; | |
9788e059 | 847 | xfs_iext_first(ifp, &icur); |
26a75f67 | 848 | xfs_iext_insert(ip, &icur, &rec, 0); |
05104e43 | 849 | |
87c472b7 | 850 | ifp->if_nextents = 1; |
aa00f286 | 851 | ip->i_nblocks = 1; |
3f17ed4b DC |
852 | xfs_trans_mod_dquot_byino(tp, ip, |
853 | XFS_TRANS_DQ_BCOUNT, 1L); | |
854 | flags |= xfs_ilog_fext(whichfork); | |
855 | ||
49f693fa DC |
856 | done: |
857 | *logflagsp = flags; | |
858 | return error; | |
2bd0ea18 NS |
859 | } |
860 | ||
861 | /* | |
49f693fa | 862 | * Called from xfs_bmap_add_attrfork to handle btree format files. |
2bd0ea18 | 863 | */ |
49f693fa DC |
864 | STATIC int /* error */ |
865 | xfs_bmap_add_attrfork_btree( | |
866 | xfs_trans_t *tp, /* transaction pointer */ | |
867 | xfs_inode_t *ip, /* incore inode pointer */ | |
49f693fa | 868 | int *flags) /* inode logging flags */ |
2bd0ea18 | 869 | { |
e0144e40 | 870 | struct xfs_btree_block *block = ip->i_df.if_broot; |
ec924d04 | 871 | struct xfs_btree_cur *cur; /* btree cursor */ |
49f693fa DC |
872 | int error; /* error return value */ |
873 | xfs_mount_t *mp; /* file system mount struct */ | |
874 | int stat; /* newroot status */ | |
56b2de80 | 875 | |
49f693fa | 876 | mp = ip->i_mount; |
e0144e40 CB |
877 | |
878 | if (XFS_BMAP_BMDR_SPACE(block) <= XFS_IFORK_DSIZE(ip)) | |
49f693fa DC |
879 | *flags |= XFS_ILOG_DBROOT; |
880 | else { | |
881 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
4f76f49c CH |
882 | error = xfs_bmbt_lookup_first(cur, &stat); |
883 | if (error) | |
49f693fa DC |
884 | goto error0; |
885 | /* must be at least one entry */ | |
fbb4fa7f DW |
886 | if (XFS_IS_CORRUPT(mp, stat != 1)) { |
887 | error = -EFSCORRUPTED; | |
888 | goto error0; | |
889 | } | |
49f693fa DC |
890 | if ((error = xfs_btree_new_iroot(cur, flags, &stat))) |
891 | goto error0; | |
892 | if (stat == 0) { | |
893 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
12b53197 | 894 | return -ENOSPC; |
49f693fa | 895 | } |
116c6a88 | 896 | cur->bc_ino.allocated = 0; |
49f693fa | 897 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); |
2bd0ea18 | 898 | } |
49f693fa DC |
899 | return 0; |
900 | error0: | |
901 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
902 | return error; | |
903 | } | |
56b2de80 | 904 | |
49f693fa DC |
905 | /* |
906 | * Called from xfs_bmap_add_attrfork to handle extents format files. | |
907 | */ | |
908 | STATIC int /* error */ | |
909 | xfs_bmap_add_attrfork_extents( | |
52f6ed9d BF |
910 | struct xfs_trans *tp, /* transaction pointer */ |
911 | struct xfs_inode *ip, /* incore inode pointer */ | |
49f693fa DC |
912 | int *flags) /* inode logging flags */ |
913 | { | |
ec924d04 | 914 | struct xfs_btree_cur *cur; /* bmap btree cursor */ |
49f693fa DC |
915 | int error; /* error return value */ |
916 | ||
87c472b7 CH |
917 | if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <= |
918 | XFS_IFORK_DSIZE(ip)) | |
49f693fa DC |
919 | return 0; |
920 | cur = NULL; | |
f7253505 BF |
921 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags, |
922 | XFS_DATA_FORK); | |
49f693fa | 923 | if (cur) { |
116c6a88 | 924 | cur->bc_ino.allocated = 0; |
660265b7 | 925 | xfs_btree_del_cursor(cur, error); |
2bd0ea18 | 926 | } |
49f693fa DC |
927 | return error; |
928 | } | |
56b2de80 | 929 | |
49f693fa DC |
930 | /* |
931 | * Called from xfs_bmap_add_attrfork to handle local format files. Each | |
932 | * different data fork content type needs a different callout to do the | |
933 | * conversion. Some are basic and only require special block initialisation | |
934 | * callouts for the data formating, others (directories) are so specialised they | |
935 | * handle everything themselves. | |
936 | * | |
937 | * XXX (dgc): investigate whether directory conversion can use the generic | |
938 | * formatting callout. It should be possible - it's just a very complex | |
5dfa5cd2 | 939 | * formatter. |
49f693fa DC |
940 | */ |
941 | STATIC int /* error */ | |
942 | xfs_bmap_add_attrfork_local( | |
d9313ca4 BF |
943 | struct xfs_trans *tp, /* transaction pointer */ |
944 | struct xfs_inode *ip, /* incore inode pointer */ | |
49f693fa DC |
945 | int *flags) /* inode logging flags */ |
946 | { | |
d9313ca4 | 947 | struct xfs_da_args dargs; /* args for dir/attr code */ |
56b2de80 | 948 | |
49f693fa DC |
949 | if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip)) |
950 | return 0; | |
a2ceac1f | 951 | |
e37bf53c | 952 | if (S_ISDIR(VFS_I(ip)->i_mode)) { |
49f693fa | 953 | memset(&dargs, 0, sizeof(dargs)); |
ff105f75 | 954 | dargs.geo = ip->i_mount->m_dir_geo; |
49f693fa | 955 | dargs.dp = ip; |
ff105f75 | 956 | dargs.total = dargs.geo->fsbcount; |
49f693fa DC |
957 | dargs.whichfork = XFS_DATA_FORK; |
958 | dargs.trans = tp; | |
959 | return xfs_dir2_sf_to_block(&dargs); | |
960 | } | |
2bd0ea18 | 961 | |
e37bf53c | 962 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
f7253505 BF |
963 | return xfs_bmap_local_to_extents(tp, ip, 1, flags, |
964 | XFS_DATA_FORK, | |
49f693fa | 965 | xfs_symlink_local_to_remote); |
56b2de80 | 966 | |
3f17ed4b DC |
967 | /* should only be called for types that support local format data */ |
968 | ASSERT(0); | |
12b53197 | 969 | return -EFSCORRUPTED; |
49f693fa | 970 | } |
2bd0ea18 | 971 | |
652a2133 DC |
972 | /* |
973 | * Set an inode attr fork offset based on the format of the data fork. | |
974 | */ | |
8a3be25d | 975 | static int |
932b563e AH |
976 | xfs_bmap_set_attrforkoff( |
977 | struct xfs_inode *ip, | |
978 | int size, | |
979 | int *version) | |
980 | { | |
c06c9d5a DC |
981 | int default_size = xfs_default_attroffset(ip) >> 3; |
982 | ||
d967a68d | 983 | switch (ip->i_df.if_format) { |
932b563e | 984 | case XFS_DINODE_FMT_DEV: |
c06c9d5a | 985 | ip->i_forkoff = default_size; |
932b563e AH |
986 | break; |
987 | case XFS_DINODE_FMT_LOCAL: | |
988 | case XFS_DINODE_FMT_EXTENTS: | |
989 | case XFS_DINODE_FMT_BTREE: | |
073f2424 CH |
990 | ip->i_forkoff = xfs_attr_shortform_bytesfit(ip, size); |
991 | if (!ip->i_forkoff) | |
c06c9d5a | 992 | ip->i_forkoff = default_size; |
914e2a04 | 993 | else if (xfs_has_attr2(ip->i_mount) && version) |
932b563e AH |
994 | *version = 2; |
995 | break; | |
996 | default: | |
997 | ASSERT(0); | |
998 | return -EINVAL; | |
999 | } | |
1000 | ||
1001 | return 0; | |
1002 | } | |
1003 | ||
49f693fa DC |
1004 | /* |
1005 | * Convert inode from non-attributed to attributed. | |
1006 | * Must not be in a transaction, ip must not be locked. | |
1007 | */ | |
1008 | int /* error code */ | |
1009 | xfs_bmap_add_attrfork( | |
1010 | xfs_inode_t *ip, /* incore inode pointer */ | |
1011 | int size, /* space new attribute needs */ | |
1012 | int rsvd) /* xact may use reserved blks */ | |
1013 | { | |
49f693fa DC |
1014 | xfs_mount_t *mp; /* mount structure */ |
1015 | xfs_trans_t *tp; /* transaction pointer */ | |
1016 | int blks; /* space reservation */ | |
1017 | int version = 1; /* superblock attr version */ | |
49f693fa DC |
1018 | int logflags; /* logging flags */ |
1019 | int error; /* error return value */ | |
56b2de80 | 1020 | |
49f693fa | 1021 | ASSERT(XFS_IFORK_Q(ip) == 0); |
2bd0ea18 | 1022 | |
49f693fa DC |
1023 | mp = ip->i_mount; |
1024 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); | |
9074815c | 1025 | |
49f693fa | 1026 | blks = XFS_ADDAFORK_SPACE_RES(mp); |
9074815c | 1027 | |
36bd1bdd | 1028 | error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_addafork, blks, 0, |
d2b662c2 | 1029 | rsvd, &tp); |
9074815c | 1030 | if (error) |
ff105f75 | 1031 | return error; |
49f693fa | 1032 | if (XFS_IFORK_Q(ip)) |
ff105f75 | 1033 | goto trans_cancel; |
a2ceac1f | 1034 | |
49f693fa | 1035 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
932b563e AH |
1036 | error = xfs_bmap_set_attrforkoff(ip, size, &version); |
1037 | if (error) | |
ff105f75 | 1038 | goto trans_cancel; |
49f693fa | 1039 | ASSERT(ip->i_afp == NULL); |
17e074de | 1040 | |
652a2133 | 1041 | ip->i_afp = xfs_ifork_alloc(XFS_DINODE_FMT_EXTENTS, 0); |
49f693fa | 1042 | logflags = 0; |
d967a68d | 1043 | switch (ip->i_df.if_format) { |
49f693fa | 1044 | case XFS_DINODE_FMT_LOCAL: |
d9313ca4 | 1045 | error = xfs_bmap_add_attrfork_local(tp, ip, &logflags); |
49f693fa DC |
1046 | break; |
1047 | case XFS_DINODE_FMT_EXTENTS: | |
d9313ca4 | 1048 | error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags); |
49f693fa DC |
1049 | break; |
1050 | case XFS_DINODE_FMT_BTREE: | |
d9313ca4 | 1051 | error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags); |
49f693fa DC |
1052 | break; |
1053 | default: | |
1054 | error = 0; | |
1055 | break; | |
1056 | } | |
1057 | if (logflags) | |
1058 | xfs_trans_log_inode(tp, ip, logflags); | |
1059 | if (error) | |
29ae66eb | 1060 | goto trans_cancel; |
b16a427a DC |
1061 | if (!xfs_has_attr(mp) || |
1062 | (!xfs_has_attr2(mp) && version == 2)) { | |
19ebedcf | 1063 | bool log_sb = false; |
a2ceac1f | 1064 | |
49f693fa | 1065 | spin_lock(&mp->m_sb_lock); |
b16a427a DC |
1066 | if (!xfs_has_attr(mp)) { |
1067 | xfs_add_attr(mp); | |
19ebedcf | 1068 | log_sb = true; |
49f693fa | 1069 | } |
b16a427a DC |
1070 | if (!xfs_has_attr2(mp) && version == 2) { |
1071 | xfs_add_attr2(mp); | |
19ebedcf | 1072 | log_sb = true; |
49f693fa | 1073 | } |
19ebedcf DC |
1074 | spin_unlock(&mp->m_sb_lock); |
1075 | if (log_sb) | |
1076 | xfs_log_sb(tp); | |
49f693fa DC |
1077 | } |
1078 | ||
de5a3f46 | 1079 | error = xfs_trans_commit(tp); |
ff105f75 DC |
1080 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1081 | return error; | |
1082 | ||
ff105f75 | 1083 | trans_cancel: |
3d7434fe | 1084 | xfs_trans_cancel(tp); |
49f693fa | 1085 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5e656dbb | 1086 | return error; |
2bd0ea18 NS |
1087 | } |
1088 | ||
399ab595 | 1089 | /* |
49f693fa DC |
1090 | * Internal and external extent tree search functions. |
1091 | */ | |
399ab595 | 1092 | |
9a7ae5a1 DW |
1093 | struct xfs_iread_state { |
1094 | struct xfs_iext_cursor icur; | |
1095 | xfs_extnum_t loaded; | |
1096 | }; | |
1097 | ||
1098 | /* Stuff every bmbt record from this block into the incore extent map. */ | |
1099 | static int | |
1100 | xfs_iread_bmbt_block( | |
1101 | struct xfs_btree_cur *cur, | |
1102 | int level, | |
1103 | void *priv) | |
1104 | { | |
1105 | struct xfs_iread_state *ir = priv; | |
1106 | struct xfs_mount *mp = cur->bc_mp; | |
116c6a88 | 1107 | struct xfs_inode *ip = cur->bc_ino.ip; |
9a7ae5a1 DW |
1108 | struct xfs_btree_block *block; |
1109 | struct xfs_buf *bp; | |
1110 | struct xfs_bmbt_rec *frp; | |
1111 | xfs_extnum_t num_recs; | |
1112 | xfs_extnum_t j; | |
116c6a88 | 1113 | int whichfork = cur->bc_ino.whichfork; |
87c472b7 | 1114 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
9a7ae5a1 DW |
1115 | |
1116 | block = xfs_btree_get_block(cur, level, &bp); | |
1117 | ||
1118 | /* Abort if we find more records than nextents. */ | |
1119 | num_recs = xfs_btree_get_numrecs(block); | |
87c472b7 | 1120 | if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) { |
9a7ae5a1 DW |
1121 | xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).", |
1122 | (unsigned long long)ip->i_ino); | |
1123 | xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block, | |
1124 | sizeof(*block), __this_address); | |
1125 | return -EFSCORRUPTED; | |
1126 | } | |
1127 | ||
1128 | /* Copy records into the incore cache. */ | |
1129 | frp = XFS_BMBT_REC_ADDR(mp, block, 1); | |
1130 | for (j = 0; j < num_recs; j++, frp++, ir->loaded++) { | |
1131 | struct xfs_bmbt_irec new; | |
1132 | xfs_failaddr_t fa; | |
1133 | ||
1134 | xfs_bmbt_disk_get_all(frp, &new); | |
1135 | fa = xfs_bmap_validate_extent(ip, whichfork, &new); | |
1136 | if (fa) { | |
1137 | xfs_inode_verifier_error(ip, -EFSCORRUPTED, | |
1138 | "xfs_iread_extents(2)", frp, | |
1139 | sizeof(*frp), fa); | |
1140 | return -EFSCORRUPTED; | |
1141 | } | |
1142 | xfs_iext_insert(ip, &ir->icur, &new, | |
1143 | xfs_bmap_fork_to_state(whichfork)); | |
1144 | trace_xfs_read_extent(ip, &ir->icur, | |
1145 | xfs_bmap_fork_to_state(whichfork), _THIS_IP_); | |
87c472b7 | 1146 | xfs_iext_next(ifp, &ir->icur); |
9a7ae5a1 DW |
1147 | } |
1148 | ||
1149 | return 0; | |
1150 | } | |
1151 | ||
49f693fa | 1152 | /* |
6d79c95c | 1153 | * Read in extents from a btree-format inode. |
49f693fa | 1154 | */ |
6d79c95c CH |
1155 | int |
1156 | xfs_iread_extents( | |
1157 | struct xfs_trans *tp, | |
1158 | struct xfs_inode *ip, | |
1159 | int whichfork) | |
49f693fa | 1160 | { |
9a7ae5a1 | 1161 | struct xfs_iread_state ir; |
6d79c95c | 1162 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
9a7ae5a1 DW |
1163 | struct xfs_mount *mp = ip->i_mount; |
1164 | struct xfs_btree_cur *cur; | |
6d79c95c CH |
1165 | int error; |
1166 | ||
30de9f1c | 1167 | if (!xfs_need_iread_extents(ifp)) |
e00c57e7 CH |
1168 | return 0; |
1169 | ||
6d79c95c CH |
1170 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1171 | ||
9a7ae5a1 DW |
1172 | ir.loaded = 0; |
1173 | xfs_iext_first(ifp, &ir.icur); | |
1174 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
1175 | error = xfs_btree_visit_blocks(cur, xfs_iread_bmbt_block, | |
1176 | XFS_BTREE_VISIT_RECORDS, &ir); | |
1177 | xfs_btree_del_cursor(cur, error); | |
1178 | if (error) | |
1179 | goto out; | |
6d79c95c | 1180 | |
87c472b7 | 1181 | if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) { |
6d79c95c CH |
1182 | error = -EFSCORRUPTED; |
1183 | goto out; | |
1184 | } | |
9a7ae5a1 | 1185 | ASSERT(ir.loaded == xfs_iext_count(ifp)); |
49f693fa | 1186 | return 0; |
6d79c95c CH |
1187 | out: |
1188 | xfs_iext_destroy(ifp); | |
1189 | return error; | |
49f693fa | 1190 | } |
399ab595 | 1191 | |
49f693fa | 1192 | /* |
2d3a6501 CH |
1193 | * Returns the relative block number of the first unused block(s) in the given |
1194 | * fork with at least "len" logically contiguous blocks free. This is the | |
1195 | * lowest-address hole if the fork has holes, else the first block past the end | |
1196 | * of fork. Return 0 if the fork is currently local (in-inode). | |
49f693fa DC |
1197 | */ |
1198 | int /* error */ | |
1199 | xfs_bmap_first_unused( | |
2d3a6501 CH |
1200 | struct xfs_trans *tp, /* transaction pointer */ |
1201 | struct xfs_inode *ip, /* incore inode */ | |
1202 | xfs_extlen_t len, /* size of hole to find */ | |
1203 | xfs_fileoff_t *first_unused, /* unused block */ | |
1204 | int whichfork) /* data or attr fork */ | |
49f693fa | 1205 | { |
2d3a6501 CH |
1206 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1207 | struct xfs_bmbt_irec got; | |
9788e059 | 1208 | struct xfs_iext_cursor icur; |
2d3a6501 CH |
1209 | xfs_fileoff_t lastaddr = 0; |
1210 | xfs_fileoff_t lowest, max; | |
1211 | int error; | |
49f693fa | 1212 | |
d967a68d | 1213 | if (ifp->if_format == XFS_DINODE_FMT_LOCAL) { |
49f693fa DC |
1214 | *first_unused = 0; |
1215 | return 0; | |
1216 | } | |
f71f3bc3 | 1217 | |
d967a68d CH |
1218 | ASSERT(xfs_ifork_has_extents(ifp)); |
1219 | ||
e00c57e7 CH |
1220 | error = xfs_iread_extents(tp, ip, whichfork); |
1221 | if (error) | |
1222 | return error; | |
f71f3bc3 | 1223 | |
2d3a6501 | 1224 | lowest = max = *first_unused; |
9788e059 | 1225 | for_each_xfs_iext(ifp, &icur, &got) { |
2bd0ea18 | 1226 | /* |
49f693fa | 1227 | * See if the hole before this extent will work. |
2bd0ea18 | 1228 | */ |
f71f3bc3 | 1229 | if (got.br_startoff >= lowest + len && |
2d3a6501 CH |
1230 | got.br_startoff - max >= len) |
1231 | break; | |
f71f3bc3 | 1232 | lastaddr = got.br_startoff + got.br_blockcount; |
49f693fa DC |
1233 | max = XFS_FILEOFF_MAX(lastaddr, lowest); |
1234 | } | |
2d3a6501 | 1235 | |
49f693fa DC |
1236 | *first_unused = max; |
1237 | return 0; | |
1238 | } | |
1239 | ||
1240 | /* | |
e6d77a21 | 1241 | * Returns the file-relative block number of the last block - 1 before |
49f693fa DC |
1242 | * last_block (input value) in the file. |
1243 | * This is not based on i_size, it is based on the extent records. | |
1244 | * Returns 0 for local files, as they do not have extent records. | |
1245 | */ | |
1246 | int /* error */ | |
1247 | xfs_bmap_last_before( | |
b0385364 CH |
1248 | struct xfs_trans *tp, /* transaction pointer */ |
1249 | struct xfs_inode *ip, /* incore inode */ | |
1250 | xfs_fileoff_t *last_block, /* last block */ | |
1251 | int whichfork) /* data or attr fork */ | |
49f693fa | 1252 | { |
b0385364 CH |
1253 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1254 | struct xfs_bmbt_irec got; | |
9788e059 | 1255 | struct xfs_iext_cursor icur; |
b0385364 | 1256 | int error; |
49f693fa | 1257 | |
d967a68d | 1258 | switch (ifp->if_format) { |
b0385364 | 1259 | case XFS_DINODE_FMT_LOCAL: |
49f693fa DC |
1260 | *last_block = 0; |
1261 | return 0; | |
b0385364 CH |
1262 | case XFS_DINODE_FMT_BTREE: |
1263 | case XFS_DINODE_FMT_EXTENTS: | |
1264 | break; | |
1265 | default: | |
a0264b73 | 1266 | ASSERT(0); |
88afc9cc | 1267 | return -EFSCORRUPTED; |
49f693fa | 1268 | } |
b0385364 | 1269 | |
e00c57e7 CH |
1270 | error = xfs_iread_extents(tp, ip, whichfork); |
1271 | if (error) | |
1272 | return error; | |
b0385364 | 1273 | |
9788e059 | 1274 | if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got)) |
d6fbe8fe | 1275 | *last_block = 0; |
49f693fa | 1276 | return 0; |
5e656dbb BN |
1277 | } |
1278 | ||
613e6057 | 1279 | int |
49f693fa DC |
1280 | xfs_bmap_last_extent( |
1281 | struct xfs_trans *tp, | |
1282 | struct xfs_inode *ip, | |
1283 | int whichfork, | |
1284 | struct xfs_bmbt_irec *rec, | |
1285 | int *is_empty) | |
56b2de80 | 1286 | { |
49f693fa | 1287 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
9788e059 | 1288 | struct xfs_iext_cursor icur; |
56b2de80 DC |
1289 | int error; |
1290 | ||
e00c57e7 CH |
1291 | error = xfs_iread_extents(tp, ip, whichfork); |
1292 | if (error) | |
1293 | return error; | |
49f693fa | 1294 | |
9788e059 CH |
1295 | xfs_iext_last(ifp, &icur); |
1296 | if (!xfs_iext_get_extent(ifp, &icur, rec)) | |
49f693fa | 1297 | *is_empty = 1; |
9788e059 CH |
1298 | else |
1299 | *is_empty = 0; | |
49f693fa DC |
1300 | return 0; |
1301 | } | |
1302 | ||
1303 | /* | |
1304 | * Check the last inode extent to determine whether this allocation will result | |
1305 | * in blocks being allocated at the end of the file. When we allocate new data | |
1306 | * blocks at the end of the file which do not start at the previous data block, | |
1307 | * we will try to align the new blocks at stripe unit boundaries. | |
1308 | * | |
ff105f75 | 1309 | * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be |
49f693fa DC |
1310 | * at, or past the EOF. |
1311 | */ | |
1312 | STATIC int | |
1313 | xfs_bmap_isaeof( | |
1314 | struct xfs_bmalloca *bma, | |
1315 | int whichfork) | |
1316 | { | |
1317 | struct xfs_bmbt_irec rec; | |
1318 | int is_empty; | |
1319 | int error; | |
1320 | ||
180d3cd8 | 1321 | bma->aeof = false; |
49f693fa DC |
1322 | error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec, |
1323 | &is_empty); | |
ff105f75 | 1324 | if (error) |
49f693fa | 1325 | return error; |
56b2de80 | 1326 | |
ff105f75 | 1327 | if (is_empty) { |
180d3cd8 | 1328 | bma->aeof = true; |
ff105f75 DC |
1329 | return 0; |
1330 | } | |
1331 | ||
56b2de80 | 1332 | /* |
49f693fa DC |
1333 | * Check if we are allocation or past the last extent, or at least into |
1334 | * the last delayed allocated extent. | |
56b2de80 | 1335 | */ |
49f693fa DC |
1336 | bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount || |
1337 | (bma->offset >= rec.br_startoff && | |
1338 | isnullstartblock(rec.br_startblock)); | |
1339 | return 0; | |
1340 | } | |
56b2de80 | 1341 | |
49f693fa DC |
1342 | /* |
1343 | * Returns the file-relative block number of the first block past eof in | |
1344 | * the file. This is not based on i_size, it is based on the extent records. | |
1345 | * Returns 0 for local files, as they do not have extent records. | |
1346 | */ | |
1347 | int | |
1348 | xfs_bmap_last_offset( | |
49f693fa DC |
1349 | struct xfs_inode *ip, |
1350 | xfs_fileoff_t *last_block, | |
1351 | int whichfork) | |
1352 | { | |
d967a68d | 1353 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa DC |
1354 | struct xfs_bmbt_irec rec; |
1355 | int is_empty; | |
1356 | int error; | |
1357 | ||
1358 | *last_block = 0; | |
1359 | ||
d967a68d | 1360 | if (ifp->if_format == XFS_DINODE_FMT_LOCAL) |
49f693fa DC |
1361 | return 0; |
1362 | ||
d967a68d | 1363 | if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp))) |
88afc9cc | 1364 | return -EFSCORRUPTED; |
49f693fa DC |
1365 | |
1366 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty); | |
1367 | if (error || is_empty) | |
1368 | return error; | |
1369 | ||
1370 | *last_block = rec.br_startoff + rec.br_blockcount; | |
1371 | return 0; | |
1372 | } | |
1373 | ||
49f693fa DC |
1374 | /* |
1375 | * Extent tree manipulation functions used during allocation. | |
1376 | */ | |
1377 | ||
1378 | /* | |
1379 | * Convert a delayed allocation to a real allocation. | |
1380 | */ | |
1381 | STATIC int /* error */ | |
1382 | xfs_bmap_add_extent_delay_real( | |
1277a5e0 DW |
1383 | struct xfs_bmalloca *bma, |
1384 | int whichfork) | |
49f693fa | 1385 | { |
87c472b7 CH |
1386 | struct xfs_mount *mp = bma->ip->i_mount; |
1387 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); | |
49f693fa | 1388 | struct xfs_bmbt_irec *new = &bma->got; |
49f693fa DC |
1389 | int error; /* error return value */ |
1390 | int i; /* temp state */ | |
49f693fa DC |
1391 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
1392 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
1393 | /* left is 0, right is 1, prev is 2 */ | |
1394 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 1395 | int state = xfs_bmap_fork_to_state(whichfork); |
49f693fa DC |
1396 | xfs_filblks_t da_new; /* new count del alloc blocks used */ |
1397 | xfs_filblks_t da_old; /* old count del alloc blocks used */ | |
1398 | xfs_filblks_t temp=0; /* value for da_new calculations */ | |
49f693fa | 1399 | int tmp_rval; /* partial logging flags */ |
19884717 | 1400 | struct xfs_bmbt_irec old; |
49f693fa | 1401 | |
1277a5e0 | 1402 | ASSERT(whichfork != XFS_ATTR_FORK); |
49f693fa DC |
1403 | ASSERT(!isnullstartblock(new->br_startblock)); |
1404 | ASSERT(!bma->cur || | |
bbf943f8 | 1405 | (bma->cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL)); |
56b2de80 | 1406 | |
79896434 | 1407 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
1408 | |
1409 | #define LEFT r[0] | |
1410 | #define RIGHT r[1] | |
1411 | #define PREV r[2] | |
56b2de80 DC |
1412 | |
1413 | /* | |
49f693fa | 1414 | * Set up a bunch of variables to make the tests simpler. |
56b2de80 | 1415 | */ |
9788e059 | 1416 | xfs_iext_get_extent(ifp, &bma->icur, &PREV); |
49f693fa | 1417 | new_endoff = new->br_startoff + new->br_blockcount; |
19884717 | 1418 | ASSERT(isnullstartblock(PREV.br_startblock)); |
49f693fa DC |
1419 | ASSERT(PREV.br_startoff <= new->br_startoff); |
1420 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
1421 | ||
1422 | da_old = startblockval(PREV.br_startblock); | |
1423 | da_new = 0; | |
1424 | ||
56b2de80 | 1425 | /* |
49f693fa DC |
1426 | * Set flags determining what part of the previous delayed allocation |
1427 | * extent is being replaced by a real allocation. | |
56b2de80 | 1428 | */ |
49f693fa DC |
1429 | if (PREV.br_startoff == new->br_startoff) |
1430 | state |= BMAP_LEFT_FILLING; | |
1431 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
1432 | state |= BMAP_RIGHT_FILLING; | |
1433 | ||
56b2de80 | 1434 | /* |
49f693fa DC |
1435 | * Check and set flags if this segment has a left neighbor. |
1436 | * Don't set contiguous if the combined extent would be too large. | |
56b2de80 | 1437 | */ |
9788e059 | 1438 | if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) { |
49f693fa | 1439 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
1440 | if (isnullstartblock(LEFT.br_startblock)) |
1441 | state |= BMAP_LEFT_DELAY; | |
1442 | } | |
1443 | ||
1444 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
1445 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
1446 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
1447 | LEFT.br_state == new->br_state && | |
1448 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
1449 | state |= BMAP_LEFT_CONTIG; | |
56b2de80 DC |
1450 | |
1451 | /* | |
49f693fa DC |
1452 | * Check and set flags if this segment has a right neighbor. |
1453 | * Don't set contiguous if the combined extent would be too large. | |
1454 | * Also check for all-three-contiguous being too large. | |
56b2de80 | 1455 | */ |
9788e059 | 1456 | if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) { |
49f693fa | 1457 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
1458 | if (isnullstartblock(RIGHT.br_startblock)) |
1459 | state |= BMAP_RIGHT_DELAY; | |
1460 | } | |
56b2de80 | 1461 | |
49f693fa DC |
1462 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
1463 | new_endoff == RIGHT.br_startoff && | |
1464 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
1465 | new->br_state == RIGHT.br_state && | |
1466 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
1467 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1468 | BMAP_RIGHT_FILLING)) != | |
1469 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1470 | BMAP_RIGHT_FILLING) || | |
1471 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
1472 | <= MAXEXTLEN)) | |
1473 | state |= BMAP_RIGHT_CONTIG; | |
5e656dbb | 1474 | |
49f693fa DC |
1475 | error = 0; |
1476 | /* | |
1477 | * Switch out based on the FILLING and CONTIG state bits. | |
1478 | */ | |
1479 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1480 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
1481 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1482 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1483 | /* | |
1484 | * Filling in all of a previously delayed allocation extent. | |
1485 | * The left and right neighbors are both contiguous with new. | |
1486 | */ | |
19884717 | 1487 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
a2ceac1f | 1488 | |
cf455a62 CH |
1489 | xfs_iext_remove(bma->ip, &bma->icur, state); |
1490 | xfs_iext_remove(bma->ip, &bma->icur, state); | |
9788e059 CH |
1491 | xfs_iext_prev(ifp, &bma->icur); |
1492 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
87c472b7 | 1493 | ifp->if_nextents--; |
12ebbfe1 | 1494 | |
49f693fa DC |
1495 | if (bma->cur == NULL) |
1496 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1497 | else { | |
1498 | rval = XFS_ILOG_CORE; | |
70a93110 | 1499 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
49f693fa DC |
1500 | if (error) |
1501 | goto done; | |
fbb4fa7f DW |
1502 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1503 | error = -EFSCORRUPTED; | |
1504 | goto done; | |
1505 | } | |
49f693fa DC |
1506 | error = xfs_btree_delete(bma->cur, &i); |
1507 | if (error) | |
1508 | goto done; | |
fbb4fa7f DW |
1509 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1510 | error = -EFSCORRUPTED; | |
1511 | goto done; | |
1512 | } | |
49f693fa DC |
1513 | error = xfs_btree_decrement(bma->cur, 0, &i); |
1514 | if (error) | |
1515 | goto done; | |
fbb4fa7f DW |
1516 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1517 | error = -EFSCORRUPTED; | |
1518 | goto done; | |
1519 | } | |
d0e5f1ff | 1520 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1521 | if (error) |
1522 | goto done; | |
5e656dbb | 1523 | } |
49f693fa DC |
1524 | break; |
1525 | ||
1526 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1527 | /* | |
1528 | * Filling in all of a previously delayed allocation extent. | |
1529 | * The left neighbor is contiguous, the right is not. | |
1530 | */ | |
19884717 | 1531 | old = LEFT; |
19884717 | 1532 | LEFT.br_blockcount += PREV.br_blockcount; |
12ebbfe1 | 1533 | |
cf455a62 | 1534 | xfs_iext_remove(bma->ip, &bma->icur, state); |
9788e059 CH |
1535 | xfs_iext_prev(ifp, &bma->icur); |
1536 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
49f693fa | 1537 | |
49f693fa DC |
1538 | if (bma->cur == NULL) |
1539 | rval = XFS_ILOG_DEXT; | |
1540 | else { | |
1541 | rval = 0; | |
70a93110 | 1542 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1543 | if (error) |
1544 | goto done; | |
fbb4fa7f DW |
1545 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1546 | error = -EFSCORRUPTED; | |
1547 | goto done; | |
1548 | } | |
d0e5f1ff | 1549 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1550 | if (error) |
1551 | goto done; | |
1552 | } | |
1553 | break; | |
1554 | ||
1555 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1556 | /* | |
1557 | * Filling in all of a previously delayed allocation extent. | |
3833299f DC |
1558 | * The right neighbor is contiguous, the left is not. Take care |
1559 | * with delay -> unwritten extent allocation here because the | |
1560 | * delalloc record we are overwriting is always written. | |
49f693fa | 1561 | */ |
19884717 CH |
1562 | PREV.br_startblock = new->br_startblock; |
1563 | PREV.br_blockcount += RIGHT.br_blockcount; | |
3833299f | 1564 | PREV.br_state = new->br_state; |
12ebbfe1 | 1565 | |
9788e059 | 1566 | xfs_iext_next(ifp, &bma->icur); |
cf455a62 | 1567 | xfs_iext_remove(bma->ip, &bma->icur, state); |
9788e059 CH |
1568 | xfs_iext_prev(ifp, &bma->icur); |
1569 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); | |
49f693fa | 1570 | |
49f693fa DC |
1571 | if (bma->cur == NULL) |
1572 | rval = XFS_ILOG_DEXT; | |
1573 | else { | |
1574 | rval = 0; | |
70a93110 | 1575 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
49f693fa DC |
1576 | if (error) |
1577 | goto done; | |
fbb4fa7f DW |
1578 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1579 | error = -EFSCORRUPTED; | |
1580 | goto done; | |
1581 | } | |
d0e5f1ff | 1582 | error = xfs_bmbt_update(bma->cur, &PREV); |
49f693fa DC |
1583 | if (error) |
1584 | goto done; | |
1585 | } | |
1586 | break; | |
1587 | ||
1588 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
1589 | /* | |
1590 | * Filling in all of a previously delayed allocation extent. | |
1591 | * Neither the left nor right neighbors are contiguous with | |
1592 | * the new one. | |
1593 | */ | |
19884717 CH |
1594 | PREV.br_startblock = new->br_startblock; |
1595 | PREV.br_state = new->br_state; | |
9788e059 | 1596 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
87c472b7 | 1597 | ifp->if_nextents++; |
5e656dbb | 1598 | |
49f693fa DC |
1599 | if (bma->cur == NULL) |
1600 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1601 | else { | |
1602 | rval = XFS_ILOG_CORE; | |
70a93110 | 1603 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1604 | if (error) |
1605 | goto done; | |
fbb4fa7f DW |
1606 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
1607 | error = -EFSCORRUPTED; | |
1608 | goto done; | |
1609 | } | |
49f693fa DC |
1610 | error = xfs_btree_insert(bma->cur, &i); |
1611 | if (error) | |
1612 | goto done; | |
fbb4fa7f DW |
1613 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1614 | error = -EFSCORRUPTED; | |
1615 | goto done; | |
1616 | } | |
49f693fa DC |
1617 | } |
1618 | break; | |
5e656dbb | 1619 | |
49f693fa DC |
1620 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
1621 | /* | |
1622 | * Filling in the first part of a previous delayed allocation. | |
1623 | * The left neighbor is contiguous. | |
1624 | */ | |
19884717 CH |
1625 | old = LEFT; |
1626 | temp = PREV.br_blockcount - new->br_blockcount; | |
1627 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
1628 | startblockval(PREV.br_startblock)); | |
1629 | ||
19884717 | 1630 | LEFT.br_blockcount += new->br_blockcount; |
a2ceac1f | 1631 | |
bc5dec54 | 1632 | PREV.br_blockcount = temp; |
19884717 CH |
1633 | PREV.br_startoff += new->br_blockcount; |
1634 | PREV.br_startblock = nullstartblock(da_new); | |
12ebbfe1 | 1635 | |
9788e059 CH |
1636 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
1637 | xfs_iext_prev(ifp, &bma->icur); | |
1638 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
19884717 | 1639 | |
49f693fa DC |
1640 | if (bma->cur == NULL) |
1641 | rval = XFS_ILOG_DEXT; | |
1642 | else { | |
1643 | rval = 0; | |
70a93110 | 1644 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1645 | if (error) |
1646 | goto done; | |
fbb4fa7f DW |
1647 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1648 | error = -EFSCORRUPTED; | |
1649 | goto done; | |
1650 | } | |
d0e5f1ff | 1651 | error = xfs_bmbt_update(bma->cur, &LEFT); |
49f693fa DC |
1652 | if (error) |
1653 | goto done; | |
2bd0ea18 | 1654 | } |
49f693fa DC |
1655 | break; |
1656 | ||
1657 | case BMAP_LEFT_FILLING: | |
2bd0ea18 | 1658 | /* |
49f693fa DC |
1659 | * Filling in the first part of a previous delayed allocation. |
1660 | * The left neighbor is not contiguous. | |
5000d01d | 1661 | */ |
9788e059 | 1662 | xfs_iext_update_extent(bma->ip, state, &bma->icur, new); |
87c472b7 CH |
1663 | ifp->if_nextents++; |
1664 | ||
49f693fa DC |
1665 | if (bma->cur == NULL) |
1666 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1667 | else { | |
1668 | rval = XFS_ILOG_CORE; | |
70a93110 | 1669 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1670 | if (error) |
1671 | goto done; | |
fbb4fa7f DW |
1672 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
1673 | error = -EFSCORRUPTED; | |
1674 | goto done; | |
1675 | } | |
49f693fa DC |
1676 | error = xfs_btree_insert(bma->cur, &i); |
1677 | if (error) | |
1678 | goto done; | |
fbb4fa7f DW |
1679 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1680 | error = -EFSCORRUPTED; | |
1681 | goto done; | |
1682 | } | |
49f693fa DC |
1683 | } |
1684 | ||
36e8786d | 1685 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1686 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f7253505 | 1687 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1688 | rval |= tmp_rval; |
1689 | if (error) | |
1690 | goto done; | |
1691 | } | |
19884717 CH |
1692 | |
1693 | temp = PREV.br_blockcount - new->br_blockcount; | |
49f693fa DC |
1694 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1695 | startblockval(PREV.br_startblock) - | |
116c6a88 | 1696 | (bma->cur ? bma->cur->bc_ino.allocated : 0)); |
19884717 | 1697 | |
19884717 CH |
1698 | PREV.br_startoff = new_endoff; |
1699 | PREV.br_blockcount = temp; | |
1700 | PREV.br_startblock = nullstartblock(da_new); | |
9788e059 | 1701 | xfs_iext_next(ifp, &bma->icur); |
26a75f67 | 1702 | xfs_iext_insert(bma->ip, &bma->icur, &PREV, state); |
9788e059 | 1703 | xfs_iext_prev(ifp, &bma->icur); |
49f693fa DC |
1704 | break; |
1705 | ||
1706 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
5e656dbb | 1707 | /* |
49f693fa DC |
1708 | * Filling in the last part of a previous delayed allocation. |
1709 | * The right neighbor is contiguous with the new allocation. | |
5e656dbb | 1710 | */ |
19884717 | 1711 | old = RIGHT; |
19884717 CH |
1712 | RIGHT.br_startoff = new->br_startoff; |
1713 | RIGHT.br_startblock = new->br_startblock; | |
1714 | RIGHT.br_blockcount += new->br_blockcount; | |
19884717 | 1715 | |
49f693fa DC |
1716 | if (bma->cur == NULL) |
1717 | rval = XFS_ILOG_DEXT; | |
1718 | else { | |
1719 | rval = 0; | |
70a93110 | 1720 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
49f693fa DC |
1721 | if (error) |
1722 | goto done; | |
fbb4fa7f DW |
1723 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1724 | error = -EFSCORRUPTED; | |
1725 | goto done; | |
1726 | } | |
d0e5f1ff | 1727 | error = xfs_bmbt_update(bma->cur, &RIGHT); |
49f693fa DC |
1728 | if (error) |
1729 | goto done; | |
1730 | } | |
1731 | ||
19884717 | 1732 | temp = PREV.br_blockcount - new->br_blockcount; |
49f693fa DC |
1733 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1734 | startblockval(PREV.br_startblock)); | |
19884717 | 1735 | |
19884717 CH |
1736 | PREV.br_blockcount = temp; |
1737 | PREV.br_startblock = nullstartblock(da_new); | |
49f693fa | 1738 | |
9788e059 CH |
1739 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
1740 | xfs_iext_next(ifp, &bma->icur); | |
1741 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT); | |
49f693fa DC |
1742 | break; |
1743 | ||
1744 | case BMAP_RIGHT_FILLING: | |
a2ceac1f | 1745 | /* |
49f693fa DC |
1746 | * Filling in the last part of a previous delayed allocation. |
1747 | * The right neighbor is not contiguous. | |
a2ceac1f | 1748 | */ |
9788e059 | 1749 | xfs_iext_update_extent(bma->ip, state, &bma->icur, new); |
87c472b7 CH |
1750 | ifp->if_nextents++; |
1751 | ||
49f693fa DC |
1752 | if (bma->cur == NULL) |
1753 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1754 | else { | |
1755 | rval = XFS_ILOG_CORE; | |
70a93110 | 1756 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1757 | if (error) |
1758 | goto done; | |
fbb4fa7f DW |
1759 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
1760 | error = -EFSCORRUPTED; | |
1761 | goto done; | |
1762 | } | |
49f693fa DC |
1763 | error = xfs_btree_insert(bma->cur, &i); |
1764 | if (error) | |
1765 | goto done; | |
fbb4fa7f DW |
1766 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1767 | error = -EFSCORRUPTED; | |
1768 | goto done; | |
1769 | } | |
49f693fa | 1770 | } |
a2ceac1f | 1771 | |
36e8786d | 1772 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1773 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f7253505 | 1774 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1775 | rval |= tmp_rval; |
1776 | if (error) | |
1777 | goto done; | |
1778 | } | |
19884717 CH |
1779 | |
1780 | temp = PREV.br_blockcount - new->br_blockcount; | |
49f693fa DC |
1781 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1782 | startblockval(PREV.br_startblock) - | |
116c6a88 | 1783 | (bma->cur ? bma->cur->bc_ino.allocated : 0)); |
19884717 | 1784 | |
19884717 CH |
1785 | PREV.br_startblock = nullstartblock(da_new); |
1786 | PREV.br_blockcount = temp; | |
26a75f67 | 1787 | xfs_iext_insert(bma->ip, &bma->icur, &PREV, state); |
9788e059 | 1788 | xfs_iext_next(ifp, &bma->icur); |
49f693fa DC |
1789 | break; |
1790 | ||
1791 | case 0: | |
5e656dbb | 1792 | /* |
49f693fa DC |
1793 | * Filling in the middle part of a previous delayed allocation. |
1794 | * Contiguity is impossible here. | |
1795 | * This case is avoided almost all the time. | |
1796 | * | |
1797 | * We start with a delayed allocation: | |
1798 | * | |
1799 | * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+ | |
1800 | * PREV @ idx | |
1801 | * | |
1802 | * and we are allocating: | |
1803 | * +rrrrrrrrrrrrrrrrr+ | |
1804 | * new | |
1805 | * | |
1806 | * and we set it up for insertion as: | |
1807 | * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+ | |
1808 | * new | |
1809 | * PREV @ idx LEFT RIGHT | |
1810 | * inserted at idx + 1 | |
5e656dbb | 1811 | */ |
19884717 CH |
1812 | old = PREV; |
1813 | ||
1814 | /* LEFT is the new middle */ | |
49f693fa | 1815 | LEFT = *new; |
19884717 CH |
1816 | |
1817 | /* RIGHT is the new right */ | |
49f693fa | 1818 | RIGHT.br_state = PREV.br_state; |
49f693fa | 1819 | RIGHT.br_startoff = new_endoff; |
19884717 CH |
1820 | RIGHT.br_blockcount = |
1821 | PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
1822 | RIGHT.br_startblock = | |
1823 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1824 | RIGHT.br_blockcount)); | |
1825 | ||
1826 | /* truncate PREV */ | |
19884717 CH |
1827 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
1828 | PREV.br_startblock = | |
1829 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1830 | PREV.br_blockcount)); | |
9788e059 | 1831 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
19884717 | 1832 | |
9788e059 | 1833 | xfs_iext_next(ifp, &bma->icur); |
26a75f67 CH |
1834 | xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state); |
1835 | xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state); | |
87c472b7 | 1836 | ifp->if_nextents++; |
19884717 | 1837 | |
49f693fa DC |
1838 | if (bma->cur == NULL) |
1839 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1840 | else { | |
1841 | rval = XFS_ILOG_CORE; | |
70a93110 | 1842 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
49f693fa DC |
1843 | if (error) |
1844 | goto done; | |
fbb4fa7f DW |
1845 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
1846 | error = -EFSCORRUPTED; | |
1847 | goto done; | |
1848 | } | |
49f693fa DC |
1849 | error = xfs_btree_insert(bma->cur, &i); |
1850 | if (error) | |
1851 | goto done; | |
fbb4fa7f DW |
1852 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1853 | error = -EFSCORRUPTED; | |
1854 | goto done; | |
1855 | } | |
49f693fa | 1856 | } |
5e656dbb | 1857 | |
36e8786d | 1858 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1859 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f7253505 | 1860 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1861 | rval |= tmp_rval; |
1862 | if (error) | |
1863 | goto done; | |
1864 | } | |
19884717 CH |
1865 | |
1866 | da_new = startblockval(PREV.br_startblock) + | |
1867 | startblockval(RIGHT.br_startblock); | |
49f693fa DC |
1868 | break; |
1869 | ||
1870 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1871 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1872 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
1873 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1874 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1875 | case BMAP_LEFT_CONTIG: | |
1876 | case BMAP_RIGHT_CONTIG: | |
1877 | /* | |
1878 | * These cases are all impossible. | |
1879 | */ | |
1880 | ASSERT(0); | |
1881 | } | |
1882 | ||
5f847f1e | 1883 | /* add reverse mapping unless caller opted out */ |
46d29bb9 DW |
1884 | if (!(bma->flags & XFS_BMAPI_NORMAP)) |
1885 | xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new); | |
d7f80320 | 1886 | |
49f693fa | 1887 | /* convert to a btree if necessary */ |
36e8786d | 1888 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa DC |
1889 | int tmp_logflags; /* partial log flag return val */ |
1890 | ||
1891 | ASSERT(bma->cur == NULL); | |
1892 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, | |
f7253505 BF |
1893 | &bma->cur, da_old > 0, &tmp_logflags, |
1894 | whichfork); | |
49f693fa DC |
1895 | bma->logflags |= tmp_logflags; |
1896 | if (error) | |
1897 | goto done; | |
1898 | } | |
1899 | ||
f73690fe DW |
1900 | if (da_new != da_old) |
1901 | xfs_mod_delalloc(mp, (int64_t)da_new - da_old); | |
1902 | ||
5bfca476 | 1903 | if (bma->cur) { |
116c6a88 DC |
1904 | da_new += bma->cur->bc_ino.allocated; |
1905 | bma->cur->bc_ino.allocated = 0; | |
49f693fa DC |
1906 | } |
1907 | ||
5bfca476 CH |
1908 | /* adjust for changes in reserved delayed indirect blocks */ |
1909 | if (da_new != da_old) { | |
1910 | ASSERT(state == 0 || da_new < da_old); | |
1911 | error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), | |
1912 | false); | |
1913 | } | |
49f693fa | 1914 | |
36e8786d | 1915 | xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork); |
49f693fa | 1916 | done: |
1277a5e0 DW |
1917 | if (whichfork != XFS_COW_FORK) |
1918 | bma->logflags |= rval; | |
49f693fa DC |
1919 | return error; |
1920 | #undef LEFT | |
1921 | #undef RIGHT | |
1922 | #undef PREV | |
2bd0ea18 NS |
1923 | } |
1924 | ||
1925 | /* | |
49f693fa | 1926 | * Convert an unwritten allocation to a real allocation or vice versa. |
2bd0ea18 | 1927 | */ |
9fa4db19 | 1928 | int /* error */ |
49f693fa DC |
1929 | xfs_bmap_add_extent_unwritten_real( |
1930 | struct xfs_trans *tp, | |
2bd0ea18 | 1931 | xfs_inode_t *ip, /* incore inode pointer */ |
4072e4b4 | 1932 | int whichfork, |
9788e059 | 1933 | struct xfs_iext_cursor *icur, |
ec924d04 | 1934 | struct xfs_btree_cur **curp, /* if *curp is null, not a btree */ |
49f693fa | 1935 | xfs_bmbt_irec_t *new, /* new data to add to file extents */ |
49f693fa | 1936 | int *logflagsp) /* inode logging flags */ |
2bd0ea18 | 1937 | { |
ec924d04 | 1938 | struct xfs_btree_cur *cur; /* btree cursor */ |
2bd0ea18 | 1939 | int error; /* error return value */ |
2bd0ea18 | 1940 | int i; /* temp state */ |
e07055b8 | 1941 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa | 1942 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
49f693fa DC |
1943 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ |
1944 | /* left is 0, right is 1, prev is 2 */ | |
1945 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 1946 | int state = xfs_bmap_fork_to_state(whichfork); |
4072e4b4 | 1947 | struct xfs_mount *mp = ip->i_mount; |
fd19685d | 1948 | struct xfs_bmbt_irec old; |
5000d01d | 1949 | |
49f693fa | 1950 | *logflagsp = 0; |
56b2de80 | 1951 | |
49f693fa | 1952 | cur = *curp; |
4072e4b4 | 1953 | ifp = XFS_IFORK_PTR(ip, whichfork); |
56b2de80 | 1954 | |
49f693fa DC |
1955 | ASSERT(!isnullstartblock(new->br_startblock)); |
1956 | ||
79896434 | 1957 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
1958 | |
1959 | #define LEFT r[0] | |
1960 | #define RIGHT r[1] | |
1961 | #define PREV r[2] | |
1962 | ||
1963 | /* | |
1964 | * Set up a bunch of variables to make the tests simpler. | |
1965 | */ | |
2bd0ea18 | 1966 | error = 0; |
9788e059 | 1967 | xfs_iext_get_extent(ifp, icur, &PREV); |
fd19685d | 1968 | ASSERT(new->br_state != PREV.br_state); |
49f693fa DC |
1969 | new_endoff = new->br_startoff + new->br_blockcount; |
1970 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
1971 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
1972 | ||
2bd0ea18 | 1973 | /* |
49f693fa DC |
1974 | * Set flags determining what part of the previous oldext allocation |
1975 | * extent is being replaced by a newext allocation. | |
2bd0ea18 | 1976 | */ |
49f693fa DC |
1977 | if (PREV.br_startoff == new->br_startoff) |
1978 | state |= BMAP_LEFT_FILLING; | |
1979 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
1980 | state |= BMAP_RIGHT_FILLING; | |
2bd0ea18 | 1981 | |
49f693fa DC |
1982 | /* |
1983 | * Check and set flags if this segment has a left neighbor. | |
1984 | * Don't set contiguous if the combined extent would be too large. | |
1985 | */ | |
9788e059 | 1986 | if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) { |
49f693fa | 1987 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
1988 | if (isnullstartblock(LEFT.br_startblock)) |
1989 | state |= BMAP_LEFT_DELAY; | |
2bd0ea18 | 1990 | } |
49f693fa DC |
1991 | |
1992 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
1993 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
1994 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
fd19685d | 1995 | LEFT.br_state == new->br_state && |
49f693fa DC |
1996 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) |
1997 | state |= BMAP_LEFT_CONTIG; | |
1998 | ||
2bd0ea18 | 1999 | /* |
49f693fa DC |
2000 | * Check and set flags if this segment has a right neighbor. |
2001 | * Don't set contiguous if the combined extent would be too large. | |
2002 | * Also check for all-three-contiguous being too large. | |
2bd0ea18 | 2003 | */ |
9788e059 | 2004 | if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) { |
49f693fa | 2005 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2006 | if (isnullstartblock(RIGHT.br_startblock)) |
2007 | state |= BMAP_RIGHT_DELAY; | |
2008 | } | |
a2ceac1f | 2009 | |
49f693fa DC |
2010 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2011 | new_endoff == RIGHT.br_startoff && | |
2012 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
fd19685d | 2013 | new->br_state == RIGHT.br_state && |
49f693fa DC |
2014 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && |
2015 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2016 | BMAP_RIGHT_FILLING)) != | |
2017 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2018 | BMAP_RIGHT_FILLING) || | |
2019 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
2020 | <= MAXEXTLEN)) | |
2021 | state |= BMAP_RIGHT_CONTIG; | |
2bd0ea18 | 2022 | |
49f693fa DC |
2023 | /* |
2024 | * Switch out based on the FILLING and CONTIG state bits. | |
2025 | */ | |
2026 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2027 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
2028 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2029 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
2bd0ea18 | 2030 | /* |
49f693fa DC |
2031 | * Setting all of a previous oldext extent to newext. |
2032 | * The left and right neighbors are both contiguous with new. | |
2bd0ea18 | 2033 | */ |
fd19685d | 2034 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
49f693fa | 2035 | |
cf455a62 CH |
2036 | xfs_iext_remove(ip, icur, state); |
2037 | xfs_iext_remove(ip, icur, state); | |
9788e059 CH |
2038 | xfs_iext_prev(ifp, icur); |
2039 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
87c472b7 | 2040 | ifp->if_nextents -= 2; |
49f693fa DC |
2041 | if (cur == NULL) |
2042 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2043 | else { | |
2044 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2045 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2046 | if (error) | |
49f693fa | 2047 | goto done; |
fbb4fa7f DW |
2048 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2049 | error = -EFSCORRUPTED; | |
2050 | goto done; | |
2051 | } | |
49f693fa DC |
2052 | if ((error = xfs_btree_delete(cur, &i))) |
2053 | goto done; | |
fbb4fa7f DW |
2054 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2055 | error = -EFSCORRUPTED; | |
2056 | goto done; | |
2057 | } | |
49f693fa DC |
2058 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2059 | goto done; | |
fbb4fa7f DW |
2060 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2061 | error = -EFSCORRUPTED; | |
2062 | goto done; | |
2063 | } | |
49f693fa DC |
2064 | if ((error = xfs_btree_delete(cur, &i))) |
2065 | goto done; | |
fbb4fa7f DW |
2066 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2067 | error = -EFSCORRUPTED; | |
2068 | goto done; | |
2069 | } | |
49f693fa DC |
2070 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2071 | goto done; | |
fbb4fa7f DW |
2072 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2073 | error = -EFSCORRUPTED; | |
2074 | goto done; | |
2075 | } | |
d0e5f1ff | 2076 | error = xfs_bmbt_update(cur, &LEFT); |
fd19685d | 2077 | if (error) |
49f693fa | 2078 | goto done; |
2bd0ea18 | 2079 | } |
2bd0ea18 NS |
2080 | break; |
2081 | ||
49f693fa | 2082 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: |
2bd0ea18 | 2083 | /* |
49f693fa DC |
2084 | * Setting all of a previous oldext extent to newext. |
2085 | * The left neighbor is contiguous, the right is not. | |
2bd0ea18 | 2086 | */ |
fd19685d | 2087 | LEFT.br_blockcount += PREV.br_blockcount; |
49f693fa | 2088 | |
cf455a62 | 2089 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2090 | xfs_iext_prev(ifp, icur); |
2091 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
87c472b7 | 2092 | ifp->if_nextents--; |
49f693fa DC |
2093 | if (cur == NULL) |
2094 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2095 | else { | |
2096 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2097 | error = xfs_bmbt_lookup_eq(cur, &PREV, &i); |
2098 | if (error) | |
49f693fa | 2099 | goto done; |
fbb4fa7f DW |
2100 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2101 | error = -EFSCORRUPTED; | |
2102 | goto done; | |
2103 | } | |
49f693fa DC |
2104 | if ((error = xfs_btree_delete(cur, &i))) |
2105 | goto done; | |
fbb4fa7f DW |
2106 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2107 | error = -EFSCORRUPTED; | |
2108 | goto done; | |
2109 | } | |
49f693fa DC |
2110 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2111 | goto done; | |
fbb4fa7f DW |
2112 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2113 | error = -EFSCORRUPTED; | |
2114 | goto done; | |
2115 | } | |
d0e5f1ff | 2116 | error = xfs_bmbt_update(cur, &LEFT); |
fd19685d | 2117 | if (error) |
49f693fa | 2118 | goto done; |
2bd0ea18 | 2119 | } |
2bd0ea18 | 2120 | break; |
5000d01d | 2121 | |
49f693fa | 2122 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2bd0ea18 | 2123 | /* |
49f693fa DC |
2124 | * Setting all of a previous oldext extent to newext. |
2125 | * The right neighbor is contiguous, the left is not. | |
2bd0ea18 | 2126 | */ |
fd19685d CH |
2127 | PREV.br_blockcount += RIGHT.br_blockcount; |
2128 | PREV.br_state = new->br_state; | |
8121dd76 | 2129 | |
9788e059 | 2130 | xfs_iext_next(ifp, icur); |
cf455a62 | 2131 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2132 | xfs_iext_prev(ifp, icur); |
2133 | xfs_iext_update_extent(ip, state, icur, &PREV); | |
87c472b7 | 2134 | ifp->if_nextents--; |
fd19685d | 2135 | |
49f693fa DC |
2136 | if (cur == NULL) |
2137 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2138 | else { | |
2139 | rval = XFS_ILOG_CORE; | |
70a93110 CH |
2140 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2141 | if (error) | |
49f693fa | 2142 | goto done; |
fbb4fa7f DW |
2143 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2144 | error = -EFSCORRUPTED; | |
2145 | goto done; | |
2146 | } | |
49f693fa DC |
2147 | if ((error = xfs_btree_delete(cur, &i))) |
2148 | goto done; | |
fbb4fa7f DW |
2149 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2150 | error = -EFSCORRUPTED; | |
2151 | goto done; | |
2152 | } | |
49f693fa DC |
2153 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2154 | goto done; | |
fbb4fa7f DW |
2155 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2156 | error = -EFSCORRUPTED; | |
2157 | goto done; | |
2158 | } | |
d0e5f1ff | 2159 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2160 | if (error) |
49f693fa DC |
2161 | goto done; |
2162 | } | |
2163 | break; | |
2bd0ea18 | 2164 | |
49f693fa DC |
2165 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: |
2166 | /* | |
2167 | * Setting all of a previous oldext extent to newext. | |
2168 | * Neither the left nor right neighbors are contiguous with | |
2169 | * the new one. | |
2170 | */ | |
fd19685d | 2171 | PREV.br_state = new->br_state; |
9788e059 | 2172 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2bd0ea18 | 2173 | |
49f693fa DC |
2174 | if (cur == NULL) |
2175 | rval = XFS_ILOG_DEXT; | |
2176 | else { | |
2177 | rval = 0; | |
70a93110 CH |
2178 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2179 | if (error) | |
49f693fa | 2180 | goto done; |
fbb4fa7f DW |
2181 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2182 | error = -EFSCORRUPTED; | |
2183 | goto done; | |
2184 | } | |
d0e5f1ff | 2185 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2186 | if (error) |
49f693fa DC |
2187 | goto done; |
2188 | } | |
2189 | break; | |
2bd0ea18 | 2190 | |
49f693fa DC |
2191 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
2192 | /* | |
2193 | * Setting the first part of a previous oldext extent to newext. | |
2194 | * The left neighbor is contiguous. | |
2195 | */ | |
fd19685d | 2196 | LEFT.br_blockcount += new->br_blockcount; |
a2ceac1f | 2197 | |
fd19685d | 2198 | old = PREV; |
fd19685d CH |
2199 | PREV.br_startoff += new->br_blockcount; |
2200 | PREV.br_startblock += new->br_blockcount; | |
2201 | PREV.br_blockcount -= new->br_blockcount; | |
b3563c19 | 2202 | |
9788e059 CH |
2203 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2204 | xfs_iext_prev(ifp, icur); | |
2205 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
b3563c19 | 2206 | |
49f693fa DC |
2207 | if (cur == NULL) |
2208 | rval = XFS_ILOG_DEXT; | |
2209 | else { | |
2210 | rval = 0; | |
70a93110 | 2211 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2212 | if (error) |
49f693fa | 2213 | goto done; |
fbb4fa7f DW |
2214 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2215 | error = -EFSCORRUPTED; | |
2216 | goto done; | |
2217 | } | |
d0e5f1ff | 2218 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2219 | if (error) |
49f693fa | 2220 | goto done; |
fd19685d CH |
2221 | error = xfs_btree_decrement(cur, 0, &i); |
2222 | if (error) | |
49f693fa | 2223 | goto done; |
d0e5f1ff | 2224 | error = xfs_bmbt_update(cur, &LEFT); |
49f693fa DC |
2225 | if (error) |
2226 | goto done; | |
2bd0ea18 | 2227 | } |
49f693fa | 2228 | break; |
b3563c19 | 2229 | |
49f693fa DC |
2230 | case BMAP_LEFT_FILLING: |
2231 | /* | |
2232 | * Setting the first part of a previous oldext extent to newext. | |
2233 | * The left neighbor is not contiguous. | |
2234 | */ | |
fd19685d | 2235 | old = PREV; |
fd19685d CH |
2236 | PREV.br_startoff += new->br_blockcount; |
2237 | PREV.br_startblock += new->br_blockcount; | |
2238 | PREV.br_blockcount -= new->br_blockcount; | |
2bd0ea18 | 2239 | |
9788e059 | 2240 | xfs_iext_update_extent(ip, state, icur, &PREV); |
26a75f67 | 2241 | xfs_iext_insert(ip, icur, new, state); |
87c472b7 CH |
2242 | ifp->if_nextents++; |
2243 | ||
49f693fa DC |
2244 | if (cur == NULL) |
2245 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2246 | else { | |
2247 | rval = XFS_ILOG_CORE; | |
70a93110 | 2248 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2249 | if (error) |
49f693fa | 2250 | goto done; |
fbb4fa7f DW |
2251 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2252 | error = -EFSCORRUPTED; | |
2253 | goto done; | |
2254 | } | |
d0e5f1ff | 2255 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2256 | if (error) |
49f693fa DC |
2257 | goto done; |
2258 | cur->bc_rec.b = *new; | |
2259 | if ((error = xfs_btree_insert(cur, &i))) | |
2260 | goto done; | |
fbb4fa7f DW |
2261 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2262 | error = -EFSCORRUPTED; | |
2263 | goto done; | |
2264 | } | |
49f693fa DC |
2265 | } |
2266 | break; | |
56b2de80 | 2267 | |
49f693fa DC |
2268 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2269 | /* | |
2270 | * Setting the last part of a previous oldext extent to newext. | |
2271 | * The right neighbor is contiguous with the new allocation. | |
2272 | */ | |
fd19685d | 2273 | old = PREV; |
fd19685d | 2274 | PREV.br_blockcount -= new->br_blockcount; |
56b2de80 | 2275 | |
fd19685d CH |
2276 | RIGHT.br_startoff = new->br_startoff; |
2277 | RIGHT.br_startblock = new->br_startblock; | |
2278 | RIGHT.br_blockcount += new->br_blockcount; | |
8121dd76 | 2279 | |
9788e059 CH |
2280 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2281 | xfs_iext_next(ifp, icur); | |
2282 | xfs_iext_update_extent(ip, state, icur, &RIGHT); | |
56b2de80 | 2283 | |
49f693fa DC |
2284 | if (cur == NULL) |
2285 | rval = XFS_ILOG_DEXT; | |
2286 | else { | |
2287 | rval = 0; | |
70a93110 | 2288 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2289 | if (error) |
49f693fa | 2290 | goto done; |
fbb4fa7f DW |
2291 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2292 | error = -EFSCORRUPTED; | |
2293 | goto done; | |
2294 | } | |
d0e5f1ff | 2295 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2296 | if (error) |
49f693fa | 2297 | goto done; |
fd19685d CH |
2298 | error = xfs_btree_increment(cur, 0, &i); |
2299 | if (error) | |
49f693fa | 2300 | goto done; |
d0e5f1ff | 2301 | error = xfs_bmbt_update(cur, &RIGHT); |
fd19685d | 2302 | if (error) |
49f693fa DC |
2303 | goto done; |
2304 | } | |
2305 | break; | |
ca86e759 | 2306 | |
49f693fa DC |
2307 | case BMAP_RIGHT_FILLING: |
2308 | /* | |
2309 | * Setting the last part of a previous oldext extent to newext. | |
2310 | * The right neighbor is not contiguous. | |
2311 | */ | |
fd19685d | 2312 | old = PREV; |
fd19685d | 2313 | PREV.br_blockcount -= new->br_blockcount; |
2bd0ea18 | 2314 | |
9788e059 CH |
2315 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2316 | xfs_iext_next(ifp, icur); | |
26a75f67 | 2317 | xfs_iext_insert(ip, icur, new, state); |
87c472b7 | 2318 | ifp->if_nextents++; |
2bd0ea18 | 2319 | |
49f693fa DC |
2320 | if (cur == NULL) |
2321 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2322 | else { | |
2323 | rval = XFS_ILOG_CORE; | |
70a93110 | 2324 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2325 | if (error) |
49f693fa | 2326 | goto done; |
fbb4fa7f DW |
2327 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2328 | error = -EFSCORRUPTED; | |
2329 | goto done; | |
2330 | } | |
d0e5f1ff | 2331 | error = xfs_bmbt_update(cur, &PREV); |
fd19685d | 2332 | if (error) |
49f693fa | 2333 | goto done; |
70a93110 CH |
2334 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2335 | if (error) | |
49f693fa | 2336 | goto done; |
fbb4fa7f DW |
2337 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
2338 | error = -EFSCORRUPTED; | |
2339 | goto done; | |
2340 | } | |
49f693fa DC |
2341 | if ((error = xfs_btree_insert(cur, &i))) |
2342 | goto done; | |
fbb4fa7f DW |
2343 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2344 | error = -EFSCORRUPTED; | |
2345 | goto done; | |
2346 | } | |
49f693fa DC |
2347 | } |
2348 | break; | |
2349 | ||
2350 | case 0: | |
2bd0ea18 | 2351 | /* |
49f693fa DC |
2352 | * Setting the middle part of a previous oldext extent to |
2353 | * newext. Contiguity is impossible here. | |
2354 | * One extent becomes three extents. | |
2bd0ea18 | 2355 | */ |
fd19685d | 2356 | old = PREV; |
fd19685d | 2357 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
49f693fa DC |
2358 | |
2359 | r[0] = *new; | |
2360 | r[1].br_startoff = new_endoff; | |
2361 | r[1].br_blockcount = | |
fd19685d | 2362 | old.br_startoff + old.br_blockcount - new_endoff; |
49f693fa | 2363 | r[1].br_startblock = new->br_startblock + new->br_blockcount; |
fd19685d | 2364 | r[1].br_state = PREV.br_state; |
49f693fa | 2365 | |
9788e059 CH |
2366 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2367 | xfs_iext_next(ifp, icur); | |
26a75f67 CH |
2368 | xfs_iext_insert(ip, icur, &r[1], state); |
2369 | xfs_iext_insert(ip, icur, &r[0], state); | |
87c472b7 | 2370 | ifp->if_nextents += 2; |
49f693fa | 2371 | |
49f693fa DC |
2372 | if (cur == NULL) |
2373 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2374 | else { | |
2375 | rval = XFS_ILOG_CORE; | |
70a93110 | 2376 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
fd19685d | 2377 | if (error) |
49f693fa | 2378 | goto done; |
fbb4fa7f DW |
2379 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2380 | error = -EFSCORRUPTED; | |
2381 | goto done; | |
2382 | } | |
49f693fa | 2383 | /* new right extent - oldext */ |
d0e5f1ff CH |
2384 | error = xfs_bmbt_update(cur, &r[1]); |
2385 | if (error) | |
49f693fa DC |
2386 | goto done; |
2387 | /* new left extent - oldext */ | |
2388 | cur->bc_rec.b = PREV; | |
49f693fa DC |
2389 | if ((error = xfs_btree_insert(cur, &i))) |
2390 | goto done; | |
fbb4fa7f DW |
2391 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2392 | error = -EFSCORRUPTED; | |
2393 | goto done; | |
2394 | } | |
49f693fa DC |
2395 | /* |
2396 | * Reset the cursor to the position of the new extent | |
2397 | * we are about to insert as we can't trust it after | |
2398 | * the previous insert. | |
2399 | */ | |
70a93110 CH |
2400 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2401 | if (error) | |
49f693fa | 2402 | goto done; |
fbb4fa7f DW |
2403 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
2404 | error = -EFSCORRUPTED; | |
2405 | goto done; | |
2406 | } | |
49f693fa | 2407 | /* new middle extent - newext */ |
49f693fa DC |
2408 | if ((error = xfs_btree_insert(cur, &i))) |
2409 | goto done; | |
fbb4fa7f DW |
2410 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2411 | error = -EFSCORRUPTED; | |
2412 | goto done; | |
2413 | } | |
2bd0ea18 | 2414 | } |
49f693fa DC |
2415 | break; |
2416 | ||
2417 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2418 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2419 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2420 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2421 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2422 | case BMAP_LEFT_CONTIG: | |
2423 | case BMAP_RIGHT_CONTIG: | |
5000d01d | 2424 | /* |
49f693fa | 2425 | * These cases are all impossible. |
2bd0ea18 | 2426 | */ |
49f693fa DC |
2427 | ASSERT(0); |
2428 | } | |
2429 | ||
d7f80320 | 2430 | /* update reverse mappings */ |
46d29bb9 | 2431 | xfs_rmap_convert_extent(mp, tp, ip, whichfork, new); |
d7f80320 | 2432 | |
49f693fa | 2433 | /* convert to a btree if necessary */ |
4072e4b4 | 2434 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
49f693fa DC |
2435 | int tmp_logflags; /* partial log flag return val */ |
2436 | ||
2437 | ASSERT(cur == NULL); | |
f7253505 BF |
2438 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
2439 | &tmp_logflags, whichfork); | |
49f693fa DC |
2440 | *logflagsp |= tmp_logflags; |
2441 | if (error) | |
2442 | goto done; | |
ca86e759 | 2443 | } |
49f693fa DC |
2444 | |
2445 | /* clear out the allocated field, done with it now in any case. */ | |
2446 | if (cur) { | |
116c6a88 | 2447 | cur->bc_ino.allocated = 0; |
49f693fa DC |
2448 | *curp = cur; |
2449 | } | |
2450 | ||
4072e4b4 | 2451 | xfs_bmap_check_leaf_extents(*curp, ip, whichfork); |
2bd0ea18 | 2452 | done: |
49f693fa | 2453 | *logflagsp |= rval; |
2bd0ea18 | 2454 | return error; |
49f693fa DC |
2455 | #undef LEFT |
2456 | #undef RIGHT | |
2457 | #undef PREV | |
2bd0ea18 NS |
2458 | } |
2459 | ||
5e656dbb | 2460 | /* |
49f693fa | 2461 | * Convert a hole to a delayed allocation. |
5e656dbb | 2462 | */ |
49f693fa DC |
2463 | STATIC void |
2464 | xfs_bmap_add_extent_hole_delay( | |
2465 | xfs_inode_t *ip, /* incore inode pointer */ | |
cc66acaf | 2466 | int whichfork, |
9788e059 | 2467 | struct xfs_iext_cursor *icur, |
49f693fa | 2468 | xfs_bmbt_irec_t *new) /* new data to add to file extents */ |
2bd0ea18 | 2469 | { |
e07055b8 | 2470 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa DC |
2471 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2472 | xfs_filblks_t newlen=0; /* new indirect size */ | |
2473 | xfs_filblks_t oldlen=0; /* old indirect size */ | |
2474 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
7bd43334 | 2475 | int state = xfs_bmap_fork_to_state(whichfork); |
450b3981 | 2476 | xfs_filblks_t temp; /* temp for indirect calculations */ |
49f693fa | 2477 | |
cc66acaf | 2478 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa | 2479 | ASSERT(isnullstartblock(new->br_startblock)); |
2bd0ea18 | 2480 | |
062998e3 | 2481 | /* |
49f693fa | 2482 | * Check and set flags if this segment has a left neighbor |
062998e3 | 2483 | */ |
9788e059 | 2484 | if (xfs_iext_peek_prev_extent(ifp, icur, &left)) { |
49f693fa | 2485 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
2486 | if (isnullstartblock(left.br_startblock)) |
2487 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2488 | } |
49f693fa DC |
2489 | |
2490 | /* | |
2491 | * Check and set flags if the current (right) segment exists. | |
2492 | * If it doesn't exist, we're converting the hole at end-of-file. | |
2493 | */ | |
9788e059 | 2494 | if (xfs_iext_get_extent(ifp, icur, &right)) { |
49f693fa | 2495 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2496 | if (isnullstartblock(right.br_startblock)) |
2497 | state |= BMAP_RIGHT_DELAY; | |
2498 | } | |
2499 | ||
2500 | /* | |
2501 | * Set contiguity flags on the left and right neighbors. | |
2502 | * Don't let extents get too large, even if the pieces are contiguous. | |
2503 | */ | |
2504 | if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) && | |
2505 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2506 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2507 | state |= BMAP_LEFT_CONTIG; | |
2508 | ||
2509 | if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) && | |
2510 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2511 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2512 | (!(state & BMAP_LEFT_CONTIG) || | |
2513 | (left.br_blockcount + new->br_blockcount + | |
2514 | right.br_blockcount <= MAXEXTLEN))) | |
2515 | state |= BMAP_RIGHT_CONTIG; | |
2516 | ||
2517 | /* | |
2518 | * Switch out based on the contiguity flags. | |
2519 | */ | |
2520 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2521 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2522 | /* | |
2523 | * New allocation is contiguous with delayed allocations | |
2524 | * on the left and on the right. | |
2525 | * Merge all three into a single extent record. | |
2526 | */ | |
49f693fa DC |
2527 | temp = left.br_blockcount + new->br_blockcount + |
2528 | right.br_blockcount; | |
2529 | ||
49f693fa DC |
2530 | oldlen = startblockval(left.br_startblock) + |
2531 | startblockval(new->br_startblock) + | |
2532 | startblockval(right.br_startblock); | |
f3b62b32 BF |
2533 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2534 | oldlen); | |
450b3981 CH |
2535 | left.br_startblock = nullstartblock(newlen); |
2536 | left.br_blockcount = temp; | |
49f693fa | 2537 | |
cf455a62 | 2538 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2539 | xfs_iext_prev(ifp, icur); |
2540 | xfs_iext_update_extent(ip, state, icur, &left); | |
49f693fa DC |
2541 | break; |
2542 | ||
2543 | case BMAP_LEFT_CONTIG: | |
2544 | /* | |
2545 | * New allocation is contiguous with a delayed allocation | |
2546 | * on the left. | |
2547 | * Merge the new allocation with the left neighbor. | |
2548 | */ | |
49f693fa DC |
2549 | temp = left.br_blockcount + new->br_blockcount; |
2550 | ||
49f693fa DC |
2551 | oldlen = startblockval(left.br_startblock) + |
2552 | startblockval(new->br_startblock); | |
f3b62b32 BF |
2553 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2554 | oldlen); | |
450b3981 CH |
2555 | left.br_blockcount = temp; |
2556 | left.br_startblock = nullstartblock(newlen); | |
ee2bb3f5 | 2557 | |
9788e059 CH |
2558 | xfs_iext_prev(ifp, icur); |
2559 | xfs_iext_update_extent(ip, state, icur, &left); | |
49f693fa DC |
2560 | break; |
2561 | ||
2562 | case BMAP_RIGHT_CONTIG: | |
2563 | /* | |
2564 | * New allocation is contiguous with a delayed allocation | |
2565 | * on the right. | |
2566 | * Merge the new allocation with the right neighbor. | |
2567 | */ | |
49f693fa DC |
2568 | temp = new->br_blockcount + right.br_blockcount; |
2569 | oldlen = startblockval(new->br_startblock) + | |
2570 | startblockval(right.br_startblock); | |
f3b62b32 BF |
2571 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2572 | oldlen); | |
450b3981 CH |
2573 | right.br_startoff = new->br_startoff; |
2574 | right.br_startblock = nullstartblock(newlen); | |
2575 | right.br_blockcount = temp; | |
9788e059 | 2576 | xfs_iext_update_extent(ip, state, icur, &right); |
49f693fa DC |
2577 | break; |
2578 | ||
2579 | case 0: | |
2580 | /* | |
2581 | * New allocation is not contiguous with another | |
2582 | * delayed allocation. | |
2583 | * Insert a new entry. | |
2584 | */ | |
2585 | oldlen = newlen = 0; | |
26a75f67 | 2586 | xfs_iext_insert(ip, icur, new, state); |
49f693fa DC |
2587 | break; |
2588 | } | |
2589 | if (oldlen != newlen) { | |
2590 | ASSERT(oldlen > newlen); | |
19ebedcf DC |
2591 | xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen), |
2592 | false); | |
49f693fa DC |
2593 | /* |
2594 | * Nothing to do for disk quota accounting here. | |
2595 | */ | |
f73690fe | 2596 | xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen); |
2bd0ea18 | 2597 | } |
2bd0ea18 NS |
2598 | } |
2599 | ||
2600 | /* | |
49f693fa | 2601 | * Convert a hole to a real allocation. |
2bd0ea18 | 2602 | */ |
49f693fa DC |
2603 | STATIC int /* error */ |
2604 | xfs_bmap_add_extent_hole_real( | |
972432f2 CH |
2605 | struct xfs_trans *tp, |
2606 | struct xfs_inode *ip, | |
2607 | int whichfork, | |
9788e059 | 2608 | struct xfs_iext_cursor *icur, |
972432f2 CH |
2609 | struct xfs_btree_cur **curp, |
2610 | struct xfs_bmbt_irec *new, | |
5f847f1e DW |
2611 | int *logflagsp, |
2612 | int flags) | |
5000d01d | 2613 | { |
972432f2 CH |
2614 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
2615 | struct xfs_mount *mp = ip->i_mount; | |
2616 | struct xfs_btree_cur *cur = *curp; | |
49f693fa DC |
2617 | int error; /* error return value */ |
2618 | int i; /* temp state */ | |
49f693fa DC |
2619 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2620 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2621 | int rval=0; /* return value (logging flags) */ | |
7bd43334 | 2622 | int state = xfs_bmap_fork_to_state(whichfork); |
3281eb91 | 2623 | struct xfs_bmbt_irec old; |
2bd0ea18 | 2624 | |
49f693fa | 2625 | ASSERT(!isnullstartblock(new->br_startblock)); |
bbf943f8 | 2626 | ASSERT(!cur || !(cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL)); |
5e656dbb | 2627 | |
79896434 | 2628 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa | 2629 | |
49f693fa DC |
2630 | /* |
2631 | * Check and set flags if this segment has a left neighbor. | |
2632 | */ | |
9788e059 | 2633 | if (xfs_iext_peek_prev_extent(ifp, icur, &left)) { |
49f693fa | 2634 | state |= BMAP_LEFT_VALID; |
49f693fa DC |
2635 | if (isnullstartblock(left.br_startblock)) |
2636 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2637 | } |
2bd0ea18 | 2638 | |
49f693fa DC |
2639 | /* |
2640 | * Check and set flags if this segment has a current value. | |
2641 | * Not true if we're inserting into the "hole" at eof. | |
2642 | */ | |
9788e059 | 2643 | if (xfs_iext_get_extent(ifp, icur, &right)) { |
49f693fa | 2644 | state |= BMAP_RIGHT_VALID; |
49f693fa DC |
2645 | if (isnullstartblock(right.br_startblock)) |
2646 | state |= BMAP_RIGHT_DELAY; | |
2bd0ea18 | 2647 | } |
2bd0ea18 | 2648 | |
49f693fa DC |
2649 | /* |
2650 | * We're inserting a real allocation between "left" and "right". | |
2651 | * Set the contiguity flags. Don't let extents get too large. | |
2652 | */ | |
2653 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2654 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2655 | left.br_startblock + left.br_blockcount == new->br_startblock && | |
2656 | left.br_state == new->br_state && | |
2657 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2658 | state |= BMAP_LEFT_CONTIG; | |
57c9fccb | 2659 | |
49f693fa DC |
2660 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2661 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2662 | new->br_startblock + new->br_blockcount == right.br_startblock && | |
2663 | new->br_state == right.br_state && | |
2664 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2665 | (!(state & BMAP_LEFT_CONTIG) || | |
2666 | left.br_blockcount + new->br_blockcount + | |
2667 | right.br_blockcount <= MAXEXTLEN)) | |
2668 | state |= BMAP_RIGHT_CONTIG; | |
ca86e759 | 2669 | |
49f693fa DC |
2670 | error = 0; |
2671 | /* | |
2672 | * Select which case we're in here, and implement it. | |
2673 | */ | |
2674 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2675 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
57c9fccb | 2676 | /* |
49f693fa DC |
2677 | * New allocation is contiguous with real allocations on the |
2678 | * left and on the right. | |
2679 | * Merge all three into a single extent record. | |
57c9fccb | 2680 | */ |
3281eb91 | 2681 | left.br_blockcount += new->br_blockcount + right.br_blockcount; |
56b2de80 | 2682 | |
cf455a62 | 2683 | xfs_iext_remove(ip, icur, state); |
9788e059 CH |
2684 | xfs_iext_prev(ifp, icur); |
2685 | xfs_iext_update_extent(ip, state, icur, &left); | |
87c472b7 | 2686 | ifp->if_nextents--; |
56b2de80 | 2687 | |
972432f2 | 2688 | if (cur == NULL) { |
49f693fa DC |
2689 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2690 | } else { | |
2691 | rval = XFS_ILOG_CORE; | |
70a93110 | 2692 | error = xfs_bmbt_lookup_eq(cur, &right, &i); |
49f693fa DC |
2693 | if (error) |
2694 | goto done; | |
fbb4fa7f DW |
2695 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2696 | error = -EFSCORRUPTED; | |
2697 | goto done; | |
2698 | } | |
972432f2 | 2699 | error = xfs_btree_delete(cur, &i); |
49f693fa DC |
2700 | if (error) |
2701 | goto done; | |
fbb4fa7f DW |
2702 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2703 | error = -EFSCORRUPTED; | |
2704 | goto done; | |
2705 | } | |
972432f2 | 2706 | error = xfs_btree_decrement(cur, 0, &i); |
49f693fa DC |
2707 | if (error) |
2708 | goto done; | |
fbb4fa7f DW |
2709 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2710 | error = -EFSCORRUPTED; | |
2711 | goto done; | |
2712 | } | |
d0e5f1ff | 2713 | error = xfs_bmbt_update(cur, &left); |
49f693fa DC |
2714 | if (error) |
2715 | goto done; | |
2716 | } | |
57c9fccb | 2717 | break; |
49f693fa DC |
2718 | |
2719 | case BMAP_LEFT_CONTIG: | |
2720 | /* | |
2721 | * New allocation is contiguous with a real allocation | |
2722 | * on the left. | |
2723 | * Merge the new allocation with the left neighbor. | |
2724 | */ | |
3281eb91 | 2725 | old = left; |
3281eb91 | 2726 | left.br_blockcount += new->br_blockcount; |
501dd276 | 2727 | |
9788e059 CH |
2728 | xfs_iext_prev(ifp, icur); |
2729 | xfs_iext_update_extent(ip, state, icur, &left); | |
49f693fa | 2730 | |
972432f2 | 2731 | if (cur == NULL) { |
49f693fa DC |
2732 | rval = xfs_ilog_fext(whichfork); |
2733 | } else { | |
2734 | rval = 0; | |
70a93110 | 2735 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
49f693fa DC |
2736 | if (error) |
2737 | goto done; | |
fbb4fa7f DW |
2738 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2739 | error = -EFSCORRUPTED; | |
2740 | goto done; | |
2741 | } | |
d0e5f1ff | 2742 | error = xfs_bmbt_update(cur, &left); |
49f693fa DC |
2743 | if (error) |
2744 | goto done; | |
2745 | } | |
57c9fccb | 2746 | break; |
49f693fa DC |
2747 | |
2748 | case BMAP_RIGHT_CONTIG: | |
2749 | /* | |
2750 | * New allocation is contiguous with a real allocation | |
2751 | * on the right. | |
2752 | * Merge the new allocation with the right neighbor. | |
2753 | */ | |
3281eb91 | 2754 | old = right; |
df926c07 | 2755 | |
3281eb91 CH |
2756 | right.br_startoff = new->br_startoff; |
2757 | right.br_startblock = new->br_startblock; | |
2758 | right.br_blockcount += new->br_blockcount; | |
9788e059 | 2759 | xfs_iext_update_extent(ip, state, icur, &right); |
49f693fa | 2760 | |
972432f2 | 2761 | if (cur == NULL) { |
49f693fa DC |
2762 | rval = xfs_ilog_fext(whichfork); |
2763 | } else { | |
2764 | rval = 0; | |
70a93110 | 2765 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
49f693fa DC |
2766 | if (error) |
2767 | goto done; | |
fbb4fa7f DW |
2768 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2769 | error = -EFSCORRUPTED; | |
2770 | goto done; | |
2771 | } | |
d0e5f1ff | 2772 | error = xfs_bmbt_update(cur, &right); |
49f693fa DC |
2773 | if (error) |
2774 | goto done; | |
2775 | } | |
2776 | break; | |
2777 | ||
2778 | case 0: | |
2779 | /* | |
2780 | * New allocation is not contiguous with another | |
2781 | * real allocation. | |
2782 | * Insert a new entry. | |
2783 | */ | |
26a75f67 | 2784 | xfs_iext_insert(ip, icur, new, state); |
87c472b7 CH |
2785 | ifp->if_nextents++; |
2786 | ||
972432f2 | 2787 | if (cur == NULL) { |
49f693fa DC |
2788 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2789 | } else { | |
2790 | rval = XFS_ILOG_CORE; | |
70a93110 | 2791 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
49f693fa DC |
2792 | if (error) |
2793 | goto done; | |
fbb4fa7f DW |
2794 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
2795 | error = -EFSCORRUPTED; | |
2796 | goto done; | |
2797 | } | |
972432f2 | 2798 | error = xfs_btree_insert(cur, &i); |
49f693fa DC |
2799 | if (error) |
2800 | goto done; | |
fbb4fa7f DW |
2801 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2802 | error = -EFSCORRUPTED; | |
2803 | goto done; | |
2804 | } | |
49f693fa | 2805 | } |
57c9fccb | 2806 | break; |
57c9fccb | 2807 | } |
a2ceac1f | 2808 | |
5f847f1e | 2809 | /* add reverse mapping unless caller opted out */ |
46d29bb9 DW |
2810 | if (!(flags & XFS_BMAPI_NORMAP)) |
2811 | xfs_rmap_map_extent(tp, ip, whichfork, new); | |
d7f80320 | 2812 | |
49f693fa | 2813 | /* convert to a btree if necessary */ |
972432f2 | 2814 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
49f693fa | 2815 | int tmp_logflags; /* partial log flag return val */ |
3f853c7a | 2816 | |
972432f2 | 2817 | ASSERT(cur == NULL); |
f7253505 BF |
2818 | error = xfs_bmap_extents_to_btree(tp, ip, curp, 0, |
2819 | &tmp_logflags, whichfork); | |
972432f2 CH |
2820 | *logflagsp |= tmp_logflags; |
2821 | cur = *curp; | |
49f693fa DC |
2822 | if (error) |
2823 | goto done; | |
57c9fccb | 2824 | } |
a2ceac1f | 2825 | |
49f693fa | 2826 | /* clear out the allocated field, done with it now in any case. */ |
972432f2 | 2827 | if (cur) |
116c6a88 | 2828 | cur->bc_ino.allocated = 0; |
49f693fa | 2829 | |
972432f2 | 2830 | xfs_bmap_check_leaf_extents(cur, ip, whichfork); |
49f693fa | 2831 | done: |
972432f2 | 2832 | *logflagsp |= rval; |
57c9fccb NS |
2833 | return error; |
2834 | } | |
2835 | ||
2bd0ea18 | 2836 | /* |
49f693fa | 2837 | * Functions used in the extent read, allocate and remove paths |
2bd0ea18 | 2838 | */ |
2bd0ea18 | 2839 | |
5000d01d | 2840 | /* |
fd2f92c8 | 2841 | * Adjust the size of the new extent based on i_extsize and rt extsize. |
2bd0ea18 | 2842 | */ |
613e6057 | 2843 | int |
49f693fa DC |
2844 | xfs_bmap_extsize_align( |
2845 | xfs_mount_t *mp, | |
2846 | xfs_bmbt_irec_t *gotp, /* next extent pointer */ | |
2847 | xfs_bmbt_irec_t *prevp, /* previous extent pointer */ | |
2848 | xfs_extlen_t extsz, /* align to this extent size */ | |
2849 | int rt, /* is this a realtime inode? */ | |
2850 | int eof, /* is extent at end-of-file? */ | |
2851 | int delay, /* creating delalloc extent? */ | |
2852 | int convert, /* overwriting unwritten extent? */ | |
2853 | xfs_fileoff_t *offp, /* in/out: aligned offset */ | |
2854 | xfs_extlen_t *lenp) /* in/out: aligned length */ | |
2bd0ea18 | 2855 | { |
49f693fa DC |
2856 | xfs_fileoff_t orig_off; /* original offset */ |
2857 | xfs_extlen_t orig_alen; /* original length */ | |
2858 | xfs_fileoff_t orig_end; /* original off+len */ | |
2859 | xfs_fileoff_t nexto; /* next file offset */ | |
2860 | xfs_fileoff_t prevo; /* previous file offset */ | |
2861 | xfs_fileoff_t align_off; /* temp for offset */ | |
2862 | xfs_extlen_t align_alen; /* temp for length */ | |
2863 | xfs_extlen_t temp; /* temp for calculations */ | |
2864 | ||
2865 | if (convert) | |
2866 | return 0; | |
2867 | ||
2868 | orig_off = align_off = *offp; | |
2869 | orig_alen = align_alen = *lenp; | |
2870 | orig_end = orig_off + orig_alen; | |
2bd0ea18 NS |
2871 | |
2872 | /* | |
49f693fa DC |
2873 | * If this request overlaps an existing extent, then don't |
2874 | * attempt to perform any additional alignment. | |
2bd0ea18 | 2875 | */ |
49f693fa DC |
2876 | if (!delay && !eof && |
2877 | (orig_off >= gotp->br_startoff) && | |
2878 | (orig_end <= gotp->br_startoff + gotp->br_blockcount)) { | |
2879 | return 0; | |
2bd0ea18 | 2880 | } |
57c9fccb | 2881 | |
49f693fa DC |
2882 | /* |
2883 | * If the file offset is unaligned vs. the extent size | |
2884 | * we need to align it. This will be possible unless | |
2885 | * the file was previously written with a kernel that didn't | |
2886 | * perform this alignment, or if a truncate shot us in the | |
2887 | * foot. | |
2888 | */ | |
5a595099 | 2889 | div_u64_rem(orig_off, extsz, &temp); |
49f693fa DC |
2890 | if (temp) { |
2891 | align_alen += temp; | |
2892 | align_off -= temp; | |
2893 | } | |
7cc23f0c DC |
2894 | |
2895 | /* Same adjustment for the end of the requested area. */ | |
2896 | temp = (align_alen % extsz); | |
2897 | if (temp) | |
2898 | align_alen += extsz - temp; | |
2899 | ||
49f693fa | 2900 | /* |
7cc23f0c DC |
2901 | * For large extent hint sizes, the aligned extent might be larger than |
2902 | * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls | |
2903 | * the length back under MAXEXTLEN. The outer allocation loops handle | |
2904 | * short allocation just fine, so it is safe to do this. We only want to | |
2905 | * do it when we are forced to, though, because it means more allocation | |
2906 | * operations are required. | |
49f693fa | 2907 | */ |
7cc23f0c DC |
2908 | while (align_alen > MAXEXTLEN) |
2909 | align_alen -= extsz; | |
2910 | ASSERT(align_alen <= MAXEXTLEN); | |
2911 | ||
49f693fa DC |
2912 | /* |
2913 | * If the previous block overlaps with this proposed allocation | |
2914 | * then move the start forward without adjusting the length. | |
2915 | */ | |
2916 | if (prevp->br_startoff != NULLFILEOFF) { | |
2917 | if (prevp->br_startblock == HOLESTARTBLOCK) | |
2918 | prevo = prevp->br_startoff; | |
2919 | else | |
2920 | prevo = prevp->br_startoff + prevp->br_blockcount; | |
2921 | } else | |
2922 | prevo = 0; | |
2923 | if (align_off != orig_off && align_off < prevo) | |
2924 | align_off = prevo; | |
2925 | /* | |
2926 | * If the next block overlaps with this proposed allocation | |
2927 | * then move the start back without adjusting the length, | |
2928 | * but not before offset 0. | |
2929 | * This may of course make the start overlap previous block, | |
2930 | * and if we hit the offset 0 limit then the next block | |
2931 | * can still overlap too. | |
2932 | */ | |
2933 | if (!eof && gotp->br_startoff != NULLFILEOFF) { | |
2934 | if ((delay && gotp->br_startblock == HOLESTARTBLOCK) || | |
2935 | (!delay && gotp->br_startblock == DELAYSTARTBLOCK)) | |
2936 | nexto = gotp->br_startoff + gotp->br_blockcount; | |
2937 | else | |
2938 | nexto = gotp->br_startoff; | |
2939 | } else | |
2940 | nexto = NULLFILEOFF; | |
2941 | if (!eof && | |
2942 | align_off + align_alen != orig_end && | |
2943 | align_off + align_alen > nexto) | |
2944 | align_off = nexto > align_alen ? nexto - align_alen : 0; | |
2945 | /* | |
2946 | * If we're now overlapping the next or previous extent that | |
2947 | * means we can't fit an extsz piece in this hole. Just move | |
2948 | * the start forward to the first valid spot and set | |
2949 | * the length so we hit the end. | |
2950 | */ | |
2951 | if (align_off != orig_off && align_off < prevo) | |
2952 | align_off = prevo; | |
2953 | if (align_off + align_alen != orig_end && | |
2954 | align_off + align_alen > nexto && | |
2955 | nexto != NULLFILEOFF) { | |
2956 | ASSERT(nexto > prevo); | |
2957 | align_alen = nexto - align_off; | |
57c9fccb | 2958 | } |
2bd0ea18 | 2959 | |
49f693fa DC |
2960 | /* |
2961 | * If realtime, and the result isn't a multiple of the realtime | |
2962 | * extent size we need to remove blocks until it is. | |
2963 | */ | |
2964 | if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) { | |
2bd0ea18 | 2965 | /* |
49f693fa DC |
2966 | * We're not covering the original request, or |
2967 | * we won't be able to once we fix the length. | |
2bd0ea18 | 2968 | */ |
49f693fa DC |
2969 | if (orig_off < align_off || |
2970 | orig_end > align_off + align_alen || | |
2971 | align_alen - temp < orig_alen) | |
12b53197 | 2972 | return -EINVAL; |
49f693fa DC |
2973 | /* |
2974 | * Try to fix it by moving the start up. | |
2975 | */ | |
2976 | if (align_off + temp <= orig_off) { | |
2977 | align_alen -= temp; | |
2978 | align_off += temp; | |
2bd0ea18 | 2979 | } |
49f693fa DC |
2980 | /* |
2981 | * Try to fix it by moving the end in. | |
2982 | */ | |
2983 | else if (align_off + align_alen - temp >= orig_end) | |
2984 | align_alen -= temp; | |
2985 | /* | |
2986 | * Set the start to the minimum then trim the length. | |
2987 | */ | |
2988 | else { | |
2989 | align_alen -= orig_off - align_off; | |
2990 | align_off = orig_off; | |
2991 | align_alen -= align_alen % mp->m_sb.sb_rextsize; | |
2992 | } | |
2993 | /* | |
2994 | * Result doesn't cover the request, fail it. | |
2995 | */ | |
2996 | if (orig_off < align_off || orig_end > align_off + align_alen) | |
12b53197 | 2997 | return -EINVAL; |
49f693fa DC |
2998 | } else { |
2999 | ASSERT(orig_off >= align_off); | |
7cc23f0c DC |
3000 | /* see MAXEXTLEN handling above */ |
3001 | ASSERT(orig_end <= align_off + align_alen || | |
3002 | align_alen + extsz > MAXEXTLEN); | |
2bd0ea18 | 3003 | } |
49f693fa DC |
3004 | |
3005 | #ifdef DEBUG | |
3006 | if (!eof && gotp->br_startoff != NULLFILEOFF) | |
3007 | ASSERT(align_off + align_alen <= gotp->br_startoff); | |
3008 | if (prevp->br_startoff != NULLFILEOFF) | |
3009 | ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount); | |
3010 | #endif | |
3011 | ||
3012 | *lenp = align_alen; | |
3013 | *offp = align_off; | |
2bd0ea18 NS |
3014 | return 0; |
3015 | } | |
3016 | ||
49f693fa DC |
3017 | #define XFS_ALLOC_GAP_UNITS 4 |
3018 | ||
613e6057 | 3019 | void |
49f693fa | 3020 | xfs_bmap_adjacent( |
613e6057 | 3021 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
2bd0ea18 | 3022 | { |
49f693fa DC |
3023 | xfs_fsblock_t adjust; /* adjustment to block numbers */ |
3024 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ | |
3025 | xfs_mount_t *mp; /* mount point structure */ | |
3026 | int nullfb; /* true if ap->firstblock isn't set */ | |
3027 | int rt; /* true if inode is realtime */ | |
2bd0ea18 | 3028 | |
49f693fa DC |
3029 | #define ISVALID(x,y) \ |
3030 | (rt ? \ | |
3031 | (x) < mp->m_sb.sb_rblocks : \ | |
3032 | XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \ | |
3033 | XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \ | |
3034 | XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks) | |
3035 | ||
3036 | mp = ap->ip->i_mount; | |
80dc7f2f | 3037 | nullfb = ap->tp->t_firstblock == NULLFSBLOCK; |
1fccd5c8 | 3038 | rt = XFS_IS_REALTIME_INODE(ap->ip) && |
a85522b6 | 3039 | (ap->datatype & XFS_ALLOC_USERDATA); |
80dc7f2f BF |
3040 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, |
3041 | ap->tp->t_firstblock); | |
49f693fa DC |
3042 | /* |
3043 | * If allocating at eof, and there's a previous real block, | |
3044 | * try to use its last block as our starting point. | |
3045 | */ | |
3046 | if (ap->eof && ap->prev.br_startoff != NULLFILEOFF && | |
3047 | !isnullstartblock(ap->prev.br_startblock) && | |
3048 | ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount, | |
3049 | ap->prev.br_startblock)) { | |
3050 | ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount; | |
3051 | /* | |
3052 | * Adjust for the gap between prevp and us. | |
3053 | */ | |
3054 | adjust = ap->offset - | |
3055 | (ap->prev.br_startoff + ap->prev.br_blockcount); | |
3056 | if (adjust && | |
3057 | ISVALID(ap->blkno + adjust, ap->prev.br_startblock)) | |
3058 | ap->blkno += adjust; | |
2bd0ea18 | 3059 | } |
49f693fa DC |
3060 | /* |
3061 | * If not at eof, then compare the two neighbor blocks. | |
3062 | * Figure out whether either one gives us a good starting point, | |
3063 | * and pick the better one. | |
3064 | */ | |
3065 | else if (!ap->eof) { | |
3066 | xfs_fsblock_t gotbno; /* right side block number */ | |
3067 | xfs_fsblock_t gotdiff=0; /* right side difference */ | |
3068 | xfs_fsblock_t prevbno; /* left side block number */ | |
3069 | xfs_fsblock_t prevdiff=0; /* left side difference */ | |
3070 | ||
3071 | /* | |
3072 | * If there's a previous (left) block, select a requested | |
3073 | * start block based on it. | |
3074 | */ | |
3075 | if (ap->prev.br_startoff != NULLFILEOFF && | |
3076 | !isnullstartblock(ap->prev.br_startblock) && | |
3077 | (prevbno = ap->prev.br_startblock + | |
3078 | ap->prev.br_blockcount) && | |
3079 | ISVALID(prevbno, ap->prev.br_startblock)) { | |
3080 | /* | |
3081 | * Calculate gap to end of previous block. | |
3082 | */ | |
3083 | adjust = prevdiff = ap->offset - | |
3084 | (ap->prev.br_startoff + | |
3085 | ap->prev.br_blockcount); | |
3086 | /* | |
3087 | * Figure the startblock based on the previous block's | |
3088 | * end and the gap size. | |
3089 | * Heuristic! | |
3090 | * If the gap is large relative to the piece we're | |
3091 | * allocating, or using it gives us an invalid block | |
3092 | * number, then just use the end of the previous block. | |
3093 | */ | |
3094 | if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3095 | ISVALID(prevbno + prevdiff, | |
3096 | ap->prev.br_startblock)) | |
3097 | prevbno += adjust; | |
3098 | else | |
3099 | prevdiff += adjust; | |
3100 | /* | |
3101 | * If the firstblock forbids it, can't use it, | |
3102 | * must use default. | |
3103 | */ | |
3104 | if (!rt && !nullfb && | |
3105 | XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno) | |
3106 | prevbno = NULLFSBLOCK; | |
3107 | } | |
3108 | /* | |
3109 | * No previous block or can't follow it, just default. | |
3110 | */ | |
3111 | else | |
3112 | prevbno = NULLFSBLOCK; | |
3113 | /* | |
3114 | * If there's a following (right) block, select a requested | |
3115 | * start block based on it. | |
3116 | */ | |
3117 | if (!isnullstartblock(ap->got.br_startblock)) { | |
3118 | /* | |
3119 | * Calculate gap to start of next block. | |
3120 | */ | |
3121 | adjust = gotdiff = ap->got.br_startoff - ap->offset; | |
3122 | /* | |
3123 | * Figure the startblock based on the next block's | |
3124 | * start and the gap size. | |
3125 | */ | |
3126 | gotbno = ap->got.br_startblock; | |
3127 | /* | |
3128 | * Heuristic! | |
3129 | * If the gap is large relative to the piece we're | |
3130 | * allocating, or using it gives us an invalid block | |
3131 | * number, then just use the start of the next block | |
3132 | * offset by our length. | |
3133 | */ | |
3134 | if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3135 | ISVALID(gotbno - gotdiff, gotbno)) | |
3136 | gotbno -= adjust; | |
3137 | else if (ISVALID(gotbno - ap->length, gotbno)) { | |
3138 | gotbno -= ap->length; | |
3139 | gotdiff += adjust - ap->length; | |
3140 | } else | |
3141 | gotdiff += adjust; | |
3142 | /* | |
3143 | * If the firstblock forbids it, can't use it, | |
3144 | * must use default. | |
3145 | */ | |
3146 | if (!rt && !nullfb && | |
3147 | XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno) | |
3148 | gotbno = NULLFSBLOCK; | |
3149 | } | |
3150 | /* | |
3151 | * No next block, just default. | |
3152 | */ | |
2bd0ea18 | 3153 | else |
49f693fa DC |
3154 | gotbno = NULLFSBLOCK; |
3155 | /* | |
3156 | * If both valid, pick the better one, else the only good | |
3157 | * one, else ap->blkno is already set (to 0 or the inode block). | |
3158 | */ | |
3159 | if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) | |
3160 | ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno; | |
3161 | else if (prevbno != NULLFSBLOCK) | |
3162 | ap->blkno = prevbno; | |
3163 | else if (gotbno != NULLFSBLOCK) | |
3164 | ap->blkno = gotbno; | |
a2ceac1f | 3165 | } |
49f693fa | 3166 | #undef ISVALID |
a2ceac1f DC |
3167 | } |
3168 | ||
ff105f75 DC |
3169 | static int |
3170 | xfs_bmap_longest_free_extent( | |
3171 | struct xfs_trans *tp, | |
3172 | xfs_agnumber_t ag, | |
3173 | xfs_extlen_t *blen, | |
3174 | int *notinit) | |
3175 | { | |
3176 | struct xfs_mount *mp = tp->t_mountp; | |
3177 | struct xfs_perag *pag; | |
3178 | xfs_extlen_t longest; | |
3179 | int error = 0; | |
3180 | ||
3181 | pag = xfs_perag_get(mp, ag); | |
3182 | if (!pag->pagf_init) { | |
3183 | error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK); | |
475f184c DW |
3184 | if (error) { |
3185 | /* Couldn't lock the AGF, so skip this AG. */ | |
3186 | if (error == -EAGAIN) { | |
3187 | *notinit = 1; | |
3188 | error = 0; | |
3189 | } | |
ff105f75 DC |
3190 | goto out; |
3191 | } | |
3192 | } | |
3193 | ||
1421de38 | 3194 | longest = xfs_alloc_longest_free_extent(pag, |
cf8ce220 DW |
3195 | xfs_alloc_min_freelist(mp, pag), |
3196 | xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE)); | |
ff105f75 DC |
3197 | if (*blen < longest) |
3198 | *blen = longest; | |
3199 | ||
3200 | out: | |
3201 | xfs_perag_put(pag); | |
3202 | return error; | |
3203 | } | |
3204 | ||
3205 | static void | |
3206 | xfs_bmap_select_minlen( | |
3207 | struct xfs_bmalloca *ap, | |
3208 | struct xfs_alloc_arg *args, | |
3209 | xfs_extlen_t *blen, | |
3210 | int notinit) | |
3211 | { | |
3212 | if (notinit || *blen < ap->minlen) { | |
3213 | /* | |
3214 | * Since we did a BUF_TRYLOCK above, it is possible that | |
3215 | * there is space for this request. | |
3216 | */ | |
3217 | args->minlen = ap->minlen; | |
3218 | } else if (*blen < args->maxlen) { | |
3219 | /* | |
3220 | * If the best seen length is less than the request length, | |
3221 | * use the best as the minimum. | |
3222 | */ | |
3223 | args->minlen = *blen; | |
3224 | } else { | |
3225 | /* | |
3226 | * Otherwise we've seen an extent as big as maxlen, use that | |
3227 | * as the minimum. | |
3228 | */ | |
3229 | args->minlen = args->maxlen; | |
3230 | } | |
3231 | } | |
3232 | ||
a2ceac1f | 3233 | STATIC int |
49f693fa DC |
3234 | xfs_bmap_btalloc_nullfb( |
3235 | struct xfs_bmalloca *ap, | |
3236 | struct xfs_alloc_arg *args, | |
3237 | xfs_extlen_t *blen) | |
a2ceac1f | 3238 | { |
49f693fa | 3239 | struct xfs_mount *mp = ap->ip->i_mount; |
49f693fa DC |
3240 | xfs_agnumber_t ag, startag; |
3241 | int notinit = 0; | |
a2ceac1f DC |
3242 | int error; |
3243 | ||
ff105f75 | 3244 | args->type = XFS_ALLOCTYPE_START_BNO; |
49f693fa | 3245 | args->total = ap->total; |
a2ceac1f | 3246 | |
49f693fa DC |
3247 | startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno); |
3248 | if (startag == NULLAGNUMBER) | |
3249 | startag = ag = 0; | |
a2ceac1f | 3250 | |
49f693fa | 3251 | while (*blen < args->maxlen) { |
ff105f75 DC |
3252 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, |
3253 | ¬init); | |
3254 | if (error) | |
3255 | return error; | |
a2ceac1f | 3256 | |
49f693fa DC |
3257 | if (++ag == mp->m_sb.sb_agcount) |
3258 | ag = 0; | |
3259 | if (ag == startag) | |
3260 | break; | |
49f693fa | 3261 | } |
2bd0ea18 | 3262 | |
ff105f75 DC |
3263 | xfs_bmap_select_minlen(ap, args, blen, notinit); |
3264 | return 0; | |
3265 | } | |
3266 | ||
3267 | STATIC int | |
3268 | xfs_bmap_btalloc_filestreams( | |
3269 | struct xfs_bmalloca *ap, | |
3270 | struct xfs_alloc_arg *args, | |
3271 | xfs_extlen_t *blen) | |
3272 | { | |
3273 | struct xfs_mount *mp = ap->ip->i_mount; | |
3274 | xfs_agnumber_t ag; | |
3275 | int notinit = 0; | |
3276 | int error; | |
3277 | ||
3278 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
3279 | args->total = ap->total; | |
3280 | ||
3281 | ag = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3282 | if (ag == NULLAGNUMBER) | |
3283 | ag = 0; | |
3284 | ||
3285 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init); | |
3286 | if (error) | |
3287 | return error; | |
3288 | ||
3289 | if (*blen < args->maxlen) { | |
3290 | error = xfs_filestream_new_ag(ap, &ag); | |
3291 | if (error) | |
3292 | return error; | |
3293 | ||
3294 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, | |
3295 | ¬init); | |
3296 | if (error) | |
3297 | return error; | |
3298 | ||
3299 | } | |
3300 | ||
3301 | xfs_bmap_select_minlen(ap, args, blen, notinit); | |
2bd0ea18 | 3302 | |
49f693fa | 3303 | /* |
ff105f75 DC |
3304 | * Set the failure fallback case to look in the selected AG as stream |
3305 | * may have moved. | |
49f693fa | 3306 | */ |
ff105f75 | 3307 | ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0); |
49f693fa | 3308 | return 0; |
2bd0ea18 NS |
3309 | } |
3310 | ||
3cb68815 DW |
3311 | /* Update all inode and quota accounting for the allocation we just did. */ |
3312 | static void | |
3313 | xfs_bmap_btalloc_accounting( | |
3314 | struct xfs_bmalloca *ap, | |
3315 | struct xfs_alloc_arg *args) | |
3316 | { | |
d07cc724 DW |
3317 | if (ap->flags & XFS_BMAPI_COWFORK) { |
3318 | /* | |
3319 | * COW fork blocks are in-core only and thus are treated as | |
3320 | * in-core quota reservation (like delalloc blocks) even when | |
3321 | * converted to real blocks. The quota reservation is not | |
3322 | * accounted to disk until blocks are remapped to the data | |
3323 | * fork. So if these blocks were previously delalloc, we | |
3324 | * already have quota reservation and there's nothing to do | |
3325 | * yet. | |
3326 | */ | |
f73690fe DW |
3327 | if (ap->wasdel) { |
3328 | xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len); | |
d07cc724 | 3329 | return; |
f73690fe | 3330 | } |
d07cc724 DW |
3331 | |
3332 | /* | |
3333 | * Otherwise, we've allocated blocks in a hole. The transaction | |
3334 | * has acquired in-core quota reservation for this extent. | |
3335 | * Rather than account these as real blocks, however, we reduce | |
3336 | * the transaction quota reservation based on the allocation. | |
3337 | * This essentially transfers the transaction quota reservation | |
3338 | * to that of a delalloc extent. | |
3339 | */ | |
3340 | ap->ip->i_delayed_blks += args->len; | |
3341 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS, | |
3342 | -(long)args->len); | |
3343 | return; | |
3344 | } | |
3345 | ||
3346 | /* data/attr fork only */ | |
aa00f286 | 3347 | ap->ip->i_nblocks += args->len; |
3cb68815 | 3348 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); |
f73690fe | 3349 | if (ap->wasdel) { |
3cb68815 | 3350 | ap->ip->i_delayed_blks -= args->len; |
f73690fe DW |
3351 | xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len); |
3352 | } | |
3cb68815 DW |
3353 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, |
3354 | ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT, | |
3355 | args->len); | |
3356 | } | |
3357 | ||
3f08f006 CB |
3358 | static int |
3359 | xfs_bmap_compute_alignments( | |
3360 | struct xfs_bmalloca *ap, | |
3361 | struct xfs_alloc_arg *args) | |
3362 | { | |
3363 | struct xfs_mount *mp = args->mp; | |
3364 | xfs_extlen_t align = 0; /* minimum allocation alignment */ | |
3365 | int stripe_align = 0; | |
3f08f006 CB |
3366 | |
3367 | /* stripe alignment for allocation is determined by mount parameters */ | |
914e2a04 | 3368 | if (mp->m_swidth && xfs_has_swalloc(mp)) |
3f08f006 CB |
3369 | stripe_align = mp->m_swidth; |
3370 | else if (mp->m_dalign) | |
3371 | stripe_align = mp->m_dalign; | |
3372 | ||
3373 | if (ap->flags & XFS_BMAPI_COWFORK) | |
3374 | align = xfs_get_cowextsz_hint(ap->ip); | |
3375 | else if (ap->datatype & XFS_ALLOC_USERDATA) | |
3376 | align = xfs_get_extsz_hint(ap->ip); | |
3377 | if (align) { | |
3af6ab0a CB |
3378 | if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0, |
3379 | ap->eof, 0, ap->conv, &ap->offset, | |
3380 | &ap->length)) | |
3381 | ASSERT(0); | |
3f08f006 CB |
3382 | ASSERT(ap->length); |
3383 | } | |
3384 | ||
3385 | /* apply extent size hints if obtained earlier */ | |
3386 | if (align) { | |
3387 | args->prod = align; | |
3388 | div_u64_rem(ap->offset, args->prod, &args->mod); | |
3389 | if (args->mod) | |
3390 | args->mod = args->prod - args->mod; | |
3391 | } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) { | |
3392 | args->prod = 1; | |
3393 | args->mod = 0; | |
3394 | } else { | |
3395 | args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog; | |
3396 | div_u64_rem(ap->offset, args->prod, &args->mod); | |
3397 | if (args->mod) | |
3398 | args->mod = args->prod - args->mod; | |
3399 | } | |
3400 | ||
3401 | return stripe_align; | |
3402 | } | |
3403 | ||
fc177ab0 CB |
3404 | static void |
3405 | xfs_bmap_process_allocated_extent( | |
3406 | struct xfs_bmalloca *ap, | |
3407 | struct xfs_alloc_arg *args, | |
3408 | xfs_fileoff_t orig_offset, | |
3409 | xfs_extlen_t orig_length) | |
3410 | { | |
3411 | int nullfb; | |
3412 | ||
3413 | nullfb = ap->tp->t_firstblock == NULLFSBLOCK; | |
3414 | ||
3415 | /* | |
3416 | * check the allocation happened at the same or higher AG than | |
3417 | * the first block that was allocated. | |
3418 | */ | |
3419 | ASSERT(nullfb || | |
3420 | XFS_FSB_TO_AGNO(args->mp, ap->tp->t_firstblock) <= | |
3421 | XFS_FSB_TO_AGNO(args->mp, args->fsbno)); | |
3422 | ||
3423 | ap->blkno = args->fsbno; | |
3424 | if (nullfb) | |
3425 | ap->tp->t_firstblock = args->fsbno; | |
3426 | ap->length = args->len; | |
3427 | /* | |
3428 | * If the extent size hint is active, we tried to round the | |
3429 | * caller's allocation request offset down to extsz and the | |
3430 | * length up to another extsz boundary. If we found a free | |
3431 | * extent we mapped it in starting at this new offset. If the | |
3432 | * newly mapped space isn't long enough to cover any of the | |
3433 | * range of offsets that was originally requested, move the | |
3434 | * mapping up so that we can fill as much of the caller's | |
3435 | * original request as possible. Free space is apparently | |
3436 | * very fragmented so we're unlikely to be able to satisfy the | |
3437 | * hints anyway. | |
3438 | */ | |
3439 | if (ap->length <= orig_length) | |
3440 | ap->offset = orig_offset; | |
3441 | else if (ap->offset + ap->length < orig_offset + orig_length) | |
3442 | ap->offset = orig_offset + orig_length - ap->length; | |
3443 | xfs_bmap_btalloc_accounting(ap, args); | |
3444 | } | |
3445 | ||
3006cea4 CB |
3446 | #ifdef DEBUG |
3447 | static int | |
3448 | xfs_bmap_exact_minlen_extent_alloc( | |
3449 | struct xfs_bmalloca *ap) | |
b3563c19 | 3450 | { |
3006cea4 CB |
3451 | struct xfs_mount *mp = ap->ip->i_mount; |
3452 | struct xfs_alloc_arg args = { .tp = ap->tp, .mp = mp }; | |
3453 | xfs_fileoff_t orig_offset; | |
3454 | xfs_extlen_t orig_length; | |
3455 | int error; | |
b3563c19 | 3456 | |
49f693fa | 3457 | ASSERT(ap->length); |
3006cea4 CB |
3458 | |
3459 | if (ap->minlen != 1) { | |
3460 | ap->blkno = NULLFSBLOCK; | |
3461 | ap->length = 0; | |
3462 | return 0; | |
3463 | } | |
3464 | ||
c38464ff DW |
3465 | orig_offset = ap->offset; |
3466 | orig_length = ap->length; | |
b3563c19 | 3467 | |
3006cea4 | 3468 | args.alloc_minlen_only = 1; |
ff105f75 | 3469 | |
3006cea4 CB |
3470 | xfs_bmap_compute_alignments(ap, &args); |
3471 | ||
3472 | if (ap->tp->t_firstblock == NULLFSBLOCK) { | |
3473 | /* | |
3474 | * Unlike the longest extent available in an AG, we don't track | |
3475 | * the length of an AG's shortest extent. | |
3476 | * XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and | |
3477 | * hence we can afford to start traversing from the 0th AG since | |
3478 | * we need not be concerned about a drop in performance in | |
3479 | * "debug only" code paths. | |
3480 | */ | |
3481 | ap->blkno = XFS_AGB_TO_FSB(mp, 0, 0); | |
3482 | } else { | |
3483 | ap->blkno = ap->tp->t_firstblock; | |
3484 | } | |
3485 | ||
3486 | args.fsbno = ap->blkno; | |
3487 | args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE; | |
3488 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
839b0c6d CB |
3489 | args.minlen = args.maxlen = ap->minlen; |
3490 | args.total = ap->total; | |
3006cea4 CB |
3491 | |
3492 | args.alignment = 1; | |
3493 | args.minalignslop = 0; | |
3494 | ||
3495 | args.minleft = ap->minleft; | |
3496 | args.wasdel = ap->wasdel; | |
3497 | args.resv = XFS_AG_RESV_NONE; | |
3498 | args.datatype = ap->datatype; | |
3499 | ||
3500 | error = xfs_alloc_vextent(&args); | |
3501 | if (error) | |
3502 | return error; | |
3503 | ||
3504 | if (args.fsbno != NULLFSBLOCK) { | |
3505 | xfs_bmap_process_allocated_extent(ap, &args, orig_offset, | |
3506 | orig_length); | |
3507 | } else { | |
3508 | ap->blkno = NULLFSBLOCK; | |
3509 | ap->length = 0; | |
3510 | } | |
3511 | ||
3512 | return 0; | |
3513 | } | |
3514 | #else | |
3515 | ||
3516 | #define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED) | |
3517 | ||
3518 | #endif | |
3519 | ||
3520 | STATIC int | |
3521 | xfs_bmap_btalloc( | |
3522 | struct xfs_bmalloca *ap) | |
3523 | { | |
3524 | struct xfs_mount *mp = ap->ip->i_mount; | |
3525 | struct xfs_alloc_arg args = { .tp = ap->tp, .mp = mp }; | |
3526 | xfs_alloctype_t atype = 0; | |
3527 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ | |
3528 | xfs_agnumber_t ag; | |
3529 | xfs_fileoff_t orig_offset; | |
3530 | xfs_extlen_t orig_length; | |
3531 | xfs_extlen_t blen; | |
3532 | xfs_extlen_t nextminlen = 0; | |
3533 | int nullfb; /* true if ap->firstblock isn't set */ | |
3534 | int isaligned; | |
3535 | int tryagain; | |
3536 | int error; | |
3537 | int stripe_align; | |
3538 | ||
3539 | ASSERT(ap->length); | |
3540 | orig_offset = ap->offset; | |
3541 | orig_length = ap->length; | |
ff105f75 | 3542 | |
3f08f006 | 3543 | stripe_align = xfs_bmap_compute_alignments(ap, &args); |
ff105f75 | 3544 | |
80dc7f2f BF |
3545 | nullfb = ap->tp->t_firstblock == NULLFSBLOCK; |
3546 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, | |
3547 | ap->tp->t_firstblock); | |
49f693fa | 3548 | if (nullfb) { |
a85522b6 | 3549 | if ((ap->datatype & XFS_ALLOC_USERDATA) && |
1fccd5c8 | 3550 | xfs_inode_is_filestream(ap->ip)) { |
49f693fa DC |
3551 | ag = xfs_filestream_lookup_ag(ap->ip); |
3552 | ag = (ag != NULLAGNUMBER) ? ag : 0; | |
3553 | ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0); | |
3554 | } else { | |
3555 | ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino); | |
3556 | } | |
3557 | } else | |
80dc7f2f | 3558 | ap->blkno = ap->tp->t_firstblock; |
49f693fa DC |
3559 | |
3560 | xfs_bmap_adjacent(ap); | |
2bd0ea18 | 3561 | |
2bd0ea18 | 3562 | /* |
49f693fa DC |
3563 | * If allowed, use ap->blkno; otherwise must use firstblock since |
3564 | * it's in the right allocation group. | |
2bd0ea18 | 3565 | */ |
49f693fa DC |
3566 | if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno) |
3567 | ; | |
3568 | else | |
80dc7f2f | 3569 | ap->blkno = ap->tp->t_firstblock; |
2bd0ea18 | 3570 | /* |
49f693fa | 3571 | * Normal allocation, done through xfs_alloc_vextent. |
2bd0ea18 | 3572 | */ |
49f693fa | 3573 | tryagain = isaligned = 0; |
49f693fa | 3574 | args.fsbno = ap->blkno; |
007347e3 | 3575 | args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE; |
49f693fa DC |
3576 | |
3577 | /* Trim the allocation back to the maximum an AG can fit. */ | |
d7ff12b8 | 3578 | args.maxlen = min(ap->length, mp->m_ag_max_usable); |
49f693fa DC |
3579 | blen = 0; |
3580 | if (nullfb) { | |
ff105f75 DC |
3581 | /* |
3582 | * Search for an allocation group with a single extent large | |
3583 | * enough for the request. If one isn't found, then adjust | |
3584 | * the minimum allocation size to the largest space found. | |
3585 | */ | |
a85522b6 | 3586 | if ((ap->datatype & XFS_ALLOC_USERDATA) && |
1fccd5c8 | 3587 | xfs_inode_is_filestream(ap->ip)) |
ff105f75 DC |
3588 | error = xfs_bmap_btalloc_filestreams(ap, &args, &blen); |
3589 | else | |
3590 | error = xfs_bmap_btalloc_nullfb(ap, &args, &blen); | |
a2ceac1f | 3591 | if (error) |
2bd0ea18 | 3592 | return error; |
565e96c6 | 3593 | } else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) { |
49f693fa DC |
3594 | if (xfs_inode_is_filestream(ap->ip)) |
3595 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3596 | else | |
3597 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3598 | args.total = args.minlen = ap->minlen; | |
3599 | } else { | |
3600 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3601 | args.total = ap->total; | |
3602 | args.minlen = ap->minlen; | |
3603 | } | |
3f08f006 | 3604 | |
2bd0ea18 | 3605 | /* |
14058d94 DC |
3606 | * If we are not low on available data blocks, and the underlying |
3607 | * logical volume manager is a stripe, and the file offset is zero then | |
3608 | * try to allocate data blocks on stripe unit boundary. NOTE: ap->aeof | |
3609 | * is only set if the allocation length is >= the stripe unit and the | |
3610 | * allocation offset is at the end of file. | |
2bd0ea18 | 3611 | */ |
565e96c6 | 3612 | if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) { |
49f693fa | 3613 | if (!ap->offset) { |
ff105f75 | 3614 | args.alignment = stripe_align; |
49f693fa DC |
3615 | atype = args.type; |
3616 | isaligned = 1; | |
3617 | /* | |
14058d94 DC |
3618 | * Adjust minlen to try and preserve alignment if we |
3619 | * can't guarantee an aligned maxlen extent. | |
49f693fa | 3620 | */ |
14058d94 DC |
3621 | if (blen > args.alignment && |
3622 | blen <= args.maxlen + args.alignment) | |
49f693fa DC |
3623 | args.minlen = blen - args.alignment; |
3624 | args.minalignslop = 0; | |
3625 | } else { | |
3626 | /* | |
3627 | * First try an exact bno allocation. | |
3628 | * If it fails then do a near or start bno | |
3629 | * allocation with alignment turned on. | |
3630 | */ | |
3631 | atype = args.type; | |
3632 | tryagain = 1; | |
3633 | args.type = XFS_ALLOCTYPE_THIS_BNO; | |
3634 | args.alignment = 1; | |
3635 | /* | |
3636 | * Compute the minlen+alignment for the | |
3637 | * next case. Set slop so that the value | |
3638 | * of minlen+alignment+slop doesn't go up | |
3639 | * between the calls. | |
3640 | */ | |
ff105f75 DC |
3641 | if (blen > stripe_align && blen <= args.maxlen) |
3642 | nextminlen = blen - stripe_align; | |
49f693fa DC |
3643 | else |
3644 | nextminlen = args.minlen; | |
ff105f75 | 3645 | if (nextminlen + stripe_align > args.minlen + 1) |
49f693fa | 3646 | args.minalignslop = |
ff105f75 | 3647 | nextminlen + stripe_align - |
49f693fa DC |
3648 | args.minlen - 1; |
3649 | else | |
3650 | args.minalignslop = 0; | |
2bd0ea18 | 3651 | } |
49f693fa DC |
3652 | } else { |
3653 | args.alignment = 1; | |
3654 | args.minalignslop = 0; | |
3655 | } | |
3656 | args.minleft = ap->minleft; | |
3657 | args.wasdel = ap->wasdel; | |
cf8ce220 | 3658 | args.resv = XFS_AG_RESV_NONE; |
1fccd5c8 | 3659 | args.datatype = ap->datatype; |
9542ae13 DC |
3660 | |
3661 | error = xfs_alloc_vextent(&args); | |
3662 | if (error) | |
49f693fa | 3663 | return error; |
9542ae13 | 3664 | |
49f693fa DC |
3665 | if (tryagain && args.fsbno == NULLFSBLOCK) { |
3666 | /* | |
3667 | * Exact allocation failed. Now try with alignment | |
3668 | * turned on. | |
3669 | */ | |
3670 | args.type = atype; | |
3671 | args.fsbno = ap->blkno; | |
ff105f75 | 3672 | args.alignment = stripe_align; |
49f693fa DC |
3673 | args.minlen = nextminlen; |
3674 | args.minalignslop = 0; | |
3675 | isaligned = 1; | |
3676 | if ((error = xfs_alloc_vextent(&args))) | |
3677 | return error; | |
3678 | } | |
3679 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
2bd0ea18 | 3680 | /* |
49f693fa DC |
3681 | * allocation failed, so turn off alignment and |
3682 | * try again. | |
2bd0ea18 | 3683 | */ |
49f693fa DC |
3684 | args.type = atype; |
3685 | args.fsbno = ap->blkno; | |
3686 | args.alignment = 0; | |
3687 | if ((error = xfs_alloc_vextent(&args))) | |
3688 | return error; | |
3689 | } | |
3690 | if (args.fsbno == NULLFSBLOCK && nullfb && | |
3691 | args.minlen > ap->minlen) { | |
3692 | args.minlen = ap->minlen; | |
3693 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3694 | args.fsbno = ap->blkno; | |
3695 | if ((error = xfs_alloc_vextent(&args))) | |
3696 | return error; | |
3697 | } | |
3698 | if (args.fsbno == NULLFSBLOCK && nullfb) { | |
3699 | args.fsbno = 0; | |
3700 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3701 | args.total = ap->minlen; | |
49f693fa DC |
3702 | if ((error = xfs_alloc_vextent(&args))) |
3703 | return error; | |
565e96c6 | 3704 | ap->tp->t_flags |= XFS_TRANS_LOWMODE; |
49f693fa | 3705 | } |
fc177ab0 | 3706 | |
49f693fa | 3707 | if (args.fsbno != NULLFSBLOCK) { |
fc177ab0 CB |
3708 | xfs_bmap_process_allocated_extent(ap, &args, orig_offset, |
3709 | orig_length); | |
49f693fa DC |
3710 | } else { |
3711 | ap->blkno = NULLFSBLOCK; | |
3712 | ap->length = 0; | |
2bd0ea18 | 3713 | } |
2bd0ea18 | 3714 | return 0; |
56b2de80 DC |
3715 | } |
3716 | ||
b3fd8db7 DW |
3717 | /* Trim extent to fit a logical block range. */ |
3718 | void | |
3719 | xfs_trim_extent( | |
3720 | struct xfs_bmbt_irec *irec, | |
3721 | xfs_fileoff_t bno, | |
3722 | xfs_filblks_t len) | |
3723 | { | |
3724 | xfs_fileoff_t distance; | |
3725 | xfs_fileoff_t end = bno + len; | |
3726 | ||
3727 | if (irec->br_startoff + irec->br_blockcount <= bno || | |
3728 | irec->br_startoff >= end) { | |
3729 | irec->br_blockcount = 0; | |
3730 | return; | |
3731 | } | |
3732 | ||
3733 | if (irec->br_startoff < bno) { | |
3734 | distance = bno - irec->br_startoff; | |
3735 | if (isnullstartblock(irec->br_startblock)) | |
3736 | irec->br_startblock = DELAYSTARTBLOCK; | |
3737 | if (irec->br_startblock != DELAYSTARTBLOCK && | |
3738 | irec->br_startblock != HOLESTARTBLOCK) | |
3739 | irec->br_startblock += distance; | |
3740 | irec->br_startoff += distance; | |
3741 | irec->br_blockcount -= distance; | |
3742 | } | |
3743 | ||
3744 | if (end < irec->br_startoff + irec->br_blockcount) { | |
3745 | distance = irec->br_startoff + irec->br_blockcount - end; | |
3746 | irec->br_blockcount -= distance; | |
3747 | } | |
3748 | } | |
3749 | ||
2bd0ea18 | 3750 | /* |
a2ceac1f | 3751 | * Trim the returned map to the required bounds |
2bd0ea18 | 3752 | */ |
a2ceac1f DC |
3753 | STATIC void |
3754 | xfs_bmapi_trim_map( | |
3755 | struct xfs_bmbt_irec *mval, | |
3756 | struct xfs_bmbt_irec *got, | |
3757 | xfs_fileoff_t *bno, | |
3758 | xfs_filblks_t len, | |
3759 | xfs_fileoff_t obno, | |
3760 | xfs_fileoff_t end, | |
3761 | int n, | |
3762 | int flags) | |
2bd0ea18 | 3763 | { |
a2ceac1f DC |
3764 | if ((flags & XFS_BMAPI_ENTIRE) || |
3765 | got->br_startoff + got->br_blockcount <= obno) { | |
3766 | *mval = *got; | |
3767 | if (isnullstartblock(got->br_startblock)) | |
3768 | mval->br_startblock = DELAYSTARTBLOCK; | |
3769 | return; | |
63be04eb | 3770 | } |
a2ceac1f DC |
3771 | |
3772 | if (obno > *bno) | |
3773 | *bno = obno; | |
3774 | ASSERT((*bno >= obno) || (n == 0)); | |
3775 | ASSERT(*bno < end); | |
3776 | mval->br_startoff = *bno; | |
3777 | if (isnullstartblock(got->br_startblock)) | |
3778 | mval->br_startblock = DELAYSTARTBLOCK; | |
2bd0ea18 | 3779 | else |
a2ceac1f DC |
3780 | mval->br_startblock = got->br_startblock + |
3781 | (*bno - got->br_startoff); | |
2bd0ea18 | 3782 | /* |
a2ceac1f DC |
3783 | * Return the minimum of what we got and what we asked for for |
3784 | * the length. We can use the len variable here because it is | |
3785 | * modified below and we could have been there before coming | |
3786 | * here if the first part of the allocation didn't overlap what | |
3787 | * was asked for. | |
2bd0ea18 | 3788 | */ |
a2ceac1f DC |
3789 | mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno, |
3790 | got->br_blockcount - (*bno - got->br_startoff)); | |
3791 | mval->br_state = got->br_state; | |
3792 | ASSERT(mval->br_blockcount <= len); | |
3793 | return; | |
3794 | } | |
56b2de80 | 3795 | |
a2ceac1f DC |
3796 | /* |
3797 | * Update and validate the extent map to return | |
3798 | */ | |
3799 | STATIC void | |
3800 | xfs_bmapi_update_map( | |
3801 | struct xfs_bmbt_irec **map, | |
3802 | xfs_fileoff_t *bno, | |
3803 | xfs_filblks_t *len, | |
3804 | xfs_fileoff_t obno, | |
3805 | xfs_fileoff_t end, | |
3806 | int *n, | |
3807 | int flags) | |
3808 | { | |
3809 | xfs_bmbt_irec_t *mval = *map; | |
3810 | ||
3811 | ASSERT((flags & XFS_BMAPI_ENTIRE) || | |
3812 | ((mval->br_startoff + mval->br_blockcount) <= end)); | |
3813 | ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) || | |
3814 | (mval->br_startoff < obno)); | |
3815 | ||
3816 | *bno = mval->br_startoff + mval->br_blockcount; | |
3817 | *len = end - *bno; | |
3818 | if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) { | |
3819 | /* update previous map with new information */ | |
3820 | ASSERT(mval->br_startblock == mval[-1].br_startblock); | |
3821 | ASSERT(mval->br_blockcount > mval[-1].br_blockcount); | |
3822 | ASSERT(mval->br_state == mval[-1].br_state); | |
3823 | mval[-1].br_blockcount = mval->br_blockcount; | |
3824 | mval[-1].br_state = mval->br_state; | |
3825 | } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK && | |
3826 | mval[-1].br_startblock != DELAYSTARTBLOCK && | |
3827 | mval[-1].br_startblock != HOLESTARTBLOCK && | |
3828 | mval->br_startblock == mval[-1].br_startblock + | |
3829 | mval[-1].br_blockcount && | |
8197bceb | 3830 | mval[-1].br_state == mval->br_state) { |
a2ceac1f DC |
3831 | ASSERT(mval->br_startoff == |
3832 | mval[-1].br_startoff + mval[-1].br_blockcount); | |
3833 | mval[-1].br_blockcount += mval->br_blockcount; | |
3834 | } else if (*n > 0 && | |
3835 | mval->br_startblock == DELAYSTARTBLOCK && | |
3836 | mval[-1].br_startblock == DELAYSTARTBLOCK && | |
3837 | mval->br_startoff == | |
3838 | mval[-1].br_startoff + mval[-1].br_blockcount) { | |
3839 | mval[-1].br_blockcount += mval->br_blockcount; | |
3840 | mval[-1].br_state = mval->br_state; | |
3841 | } else if (!((*n == 0) && | |
3842 | ((mval->br_startoff + mval->br_blockcount) <= | |
3843 | obno))) { | |
3844 | mval++; | |
3845 | (*n)++; | |
3846 | } | |
3847 | *map = mval; | |
3848 | } | |
399ab595 | 3849 | |
a2ceac1f DC |
3850 | /* |
3851 | * Map file blocks to filesystem blocks without allocation. | |
3852 | */ | |
3853 | int | |
3854 | xfs_bmapi_read( | |
3855 | struct xfs_inode *ip, | |
3856 | xfs_fileoff_t bno, | |
3857 | xfs_filblks_t len, | |
3858 | struct xfs_bmbt_irec *mval, | |
3859 | int *nmap, | |
3860 | int flags) | |
3861 | { | |
3862 | struct xfs_mount *mp = ip->i_mount; | |
212be827 CH |
3863 | int whichfork = xfs_bmapi_whichfork(flags); |
3864 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
a2ceac1f | 3865 | struct xfs_bmbt_irec got; |
a2ceac1f DC |
3866 | xfs_fileoff_t obno; |
3867 | xfs_fileoff_t end; | |
9788e059 | 3868 | struct xfs_iext_cursor icur; |
a2ceac1f | 3869 | int error; |
c2d73ed3 | 3870 | bool eof = false; |
a2ceac1f | 3871 | int n = 0; |
399ab595 | 3872 | |
a2ceac1f | 3873 | ASSERT(*nmap >= 1); |
00773e64 | 3874 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE))); |
ff105f75 | 3875 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)); |
062998e3 | 3876 | |
212be827 CH |
3877 | if (WARN_ON_ONCE(!ifp)) |
3878 | return -EFSCORRUPTED; | |
3879 | ||
d967a68d CH |
3880 | if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) || |
3881 | XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) | |
12b53197 | 3882 | return -EFSCORRUPTED; |
062998e3 | 3883 | |
93adb06a | 3884 | if (xfs_is_shutdown(mp)) |
12b53197 | 3885 | return -EIO; |
399ab595 | 3886 | |
79896434 | 3887 | XFS_STATS_INC(mp, xs_blk_mapr); |
a2ceac1f | 3888 | |
e00c57e7 CH |
3889 | error = xfs_iread_extents(NULL, ip, whichfork); |
3890 | if (error) | |
3891 | return error; | |
a2ceac1f | 3892 | |
9788e059 | 3893 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) |
c2d73ed3 | 3894 | eof = true; |
a2ceac1f DC |
3895 | end = bno + len; |
3896 | obno = bno; | |
3897 | ||
3898 | while (bno < end && n < *nmap) { | |
3899 | /* Reading past eof, act as though there's a hole up to end. */ | |
3900 | if (eof) | |
3901 | got.br_startoff = end; | |
3902 | if (got.br_startoff > bno) { | |
3903 | /* Reading in a hole. */ | |
2bd0ea18 NS |
3904 | mval->br_startoff = bno; |
3905 | mval->br_startblock = HOLESTARTBLOCK; | |
3906 | mval->br_blockcount = | |
3907 | XFS_FILBLKS_MIN(len, got.br_startoff - bno); | |
3908 | mval->br_state = XFS_EXT_NORM; | |
3909 | bno += mval->br_blockcount; | |
3910 | len -= mval->br_blockcount; | |
3911 | mval++; | |
3912 | n++; | |
3913 | continue; | |
3914 | } | |
a2ceac1f DC |
3915 | |
3916 | /* set up the extent map to return. */ | |
3917 | xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags); | |
3918 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
3919 | ||
3920 | /* If we're done, stop now. */ | |
3921 | if (bno >= end || n >= *nmap) | |
3922 | break; | |
3923 | ||
3924 | /* Else go on to the next record. */ | |
9788e059 | 3925 | if (!xfs_iext_next_extent(ifp, &icur, &got)) |
c2d73ed3 | 3926 | eof = true; |
a2ceac1f DC |
3927 | } |
3928 | *nmap = n; | |
3929 | return 0; | |
3930 | } | |
3931 | ||
7a868ee8 BF |
3932 | /* |
3933 | * Add a delayed allocation extent to an inode. Blocks are reserved from the | |
3934 | * global pool and the extent inserted into the inode in-core extent tree. | |
3935 | * | |
3936 | * On entry, got refers to the first extent beyond the offset of the extent to | |
3937 | * allocate or eof is specified if no such extent exists. On return, got refers | |
3938 | * to the extent record that was inserted to the inode fork. | |
3939 | * | |
3940 | * Note that the allocated extent may have been merged with contiguous extents | |
3941 | * during insertion into the inode fork. Thus, got does not reflect the current | |
3942 | * state of the inode fork on return. If necessary, the caller can use lastx to | |
3943 | * look up the updated record in the inode fork. | |
3944 | */ | |
4488e421 | 3945 | int |
a2ceac1f DC |
3946 | xfs_bmapi_reserve_delalloc( |
3947 | struct xfs_inode *ip, | |
cc66acaf | 3948 | int whichfork, |
f7b1a8b1 | 3949 | xfs_fileoff_t off, |
a2ceac1f | 3950 | xfs_filblks_t len, |
f7b1a8b1 | 3951 | xfs_filblks_t prealloc, |
a2ceac1f | 3952 | struct xfs_bmbt_irec *got, |
9788e059 | 3953 | struct xfs_iext_cursor *icur, |
a2ceac1f DC |
3954 | int eof) |
3955 | { | |
3956 | struct xfs_mount *mp = ip->i_mount; | |
cc66acaf | 3957 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
a2ceac1f DC |
3958 | xfs_extlen_t alen; |
3959 | xfs_extlen_t indlen; | |
a2ceac1f | 3960 | int error; |
f7b1a8b1 | 3961 | xfs_fileoff_t aoff = off; |
a2ceac1f | 3962 | |
f7b1a8b1 BF |
3963 | /* |
3964 | * Cap the alloc length. Keep track of prealloc so we know whether to | |
3965 | * tag the inode before we return. | |
3966 | */ | |
3967 | alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN); | |
a2ceac1f DC |
3968 | if (!eof) |
3969 | alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff); | |
f7b1a8b1 BF |
3970 | if (prealloc && alen >= len) |
3971 | prealloc = alen - len; | |
a2ceac1f DC |
3972 | |
3973 | /* Figure out the extent size, adjust alen */ | |
885ba5ce | 3974 | if (whichfork == XFS_COW_FORK) { |
d41d2303 | 3975 | struct xfs_bmbt_irec prev; |
885ba5ce | 3976 | xfs_extlen_t extsz = xfs_get_cowextsz_hint(ip); |
d41d2303 | 3977 | |
9788e059 | 3978 | if (!xfs_iext_peek_prev_extent(ifp, icur, &prev)) |
d41d2303 CH |
3979 | prev.br_startoff = NULLFILEOFF; |
3980 | ||
885ba5ce | 3981 | error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof, |
a2ceac1f DC |
3982 | 1, 0, &aoff, &alen); |
3983 | ASSERT(!error); | |
3984 | } | |
3985 | ||
a2ceac1f DC |
3986 | /* |
3987 | * Make a transaction-less quota reservation for delayed allocation | |
3988 | * blocks. This number gets adjusted later. We return if we haven't | |
3989 | * allocated blocks already inside this loop. | |
3990 | */ | |
9fcc3af9 | 3991 | error = xfs_quota_reserve_blkres(ip, alen); |
a2ceac1f DC |
3992 | if (error) |
3993 | return error; | |
3994 | ||
3995 | /* | |
3996 | * Split changing sb for alen and indlen since they could be coming | |
3997 | * from different places. | |
3998 | */ | |
3999 | indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen); | |
4000 | ASSERT(indlen > 0); | |
4001 | ||
885ba5ce | 4002 | error = xfs_mod_fdblocks(mp, -((int64_t)alen), false); |
a2ceac1f DC |
4003 | if (error) |
4004 | goto out_unreserve_quota; | |
4005 | ||
19ebedcf | 4006 | error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false); |
a2ceac1f DC |
4007 | if (error) |
4008 | goto out_unreserve_blocks; | |
4009 | ||
4010 | ||
4011 | ip->i_delayed_blks += alen; | |
f73690fe | 4012 | xfs_mod_delalloc(ip->i_mount, alen + indlen); |
a2ceac1f DC |
4013 | |
4014 | got->br_startoff = aoff; | |
4015 | got->br_startblock = nullstartblock(indlen); | |
4016 | got->br_blockcount = alen; | |
4017 | got->br_state = XFS_EXT_NORM; | |
a2ceac1f | 4018 | |
9788e059 | 4019 | xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got); |
a2ceac1f | 4020 | |
f7b1a8b1 BF |
4021 | /* |
4022 | * Tag the inode if blocks were preallocated. Note that COW fork | |
4023 | * preallocation can occur at the start or end of the extent, even when | |
4024 | * prealloc == 0, so we must also check the aligned offset and length. | |
4025 | */ | |
4026 | if (whichfork == XFS_DATA_FORK && prealloc) | |
4027 | xfs_inode_set_eofblocks_tag(ip); | |
4028 | if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len)) | |
4029 | xfs_inode_set_cowblocks_tag(ip); | |
4030 | ||
a2ceac1f DC |
4031 | return 0; |
4032 | ||
4033 | out_unreserve_blocks: | |
885ba5ce | 4034 | xfs_mod_fdblocks(mp, alen, false); |
a2ceac1f DC |
4035 | out_unreserve_quota: |
4036 | if (XFS_IS_QUOTA_ON(mp)) | |
9fcc3af9 | 4037 | xfs_quota_unreserve_blkres(ip, alen); |
a2ceac1f DC |
4038 | return error; |
4039 | } | |
4040 | ||
2ac7663a CH |
4041 | static int |
4042 | xfs_bmap_alloc_userdata( | |
4043 | struct xfs_bmalloca *bma) | |
4044 | { | |
4045 | struct xfs_mount *mp = bma->ip->i_mount; | |
4046 | int whichfork = xfs_bmapi_whichfork(bma->flags); | |
4047 | int error; | |
4048 | ||
4049 | /* | |
4050 | * Set the data type being allocated. For the data fork, the first data | |
4051 | * in the file is treated differently to all other allocations. For the | |
4052 | * attribute fork, we only need to ensure the allocated range is not on | |
4053 | * the busy list. | |
4054 | */ | |
4055 | bma->datatype = XFS_ALLOC_NOBUSY; | |
2ac7663a | 4056 | if (whichfork == XFS_DATA_FORK) { |
a85522b6 | 4057 | bma->datatype |= XFS_ALLOC_USERDATA; |
2ac7663a CH |
4058 | if (bma->offset == 0) |
4059 | bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA; | |
2ac7663a CH |
4060 | |
4061 | if (mp->m_dalign && bma->length >= mp->m_dalign) { | |
4062 | error = xfs_bmap_isaeof(bma, whichfork); | |
4063 | if (error) | |
4064 | return error; | |
4065 | } | |
4066 | ||
4067 | if (XFS_IS_REALTIME_INODE(bma->ip)) | |
4068 | return xfs_bmap_rtalloc(bma); | |
4069 | } | |
4070 | ||
3006cea4 CB |
4071 | if (unlikely(XFS_TEST_ERROR(false, mp, |
4072 | XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT))) | |
4073 | return xfs_bmap_exact_minlen_extent_alloc(bma); | |
4074 | ||
2ac7663a CH |
4075 | return xfs_bmap_btalloc(bma); |
4076 | } | |
4077 | ||
ff105f75 DC |
4078 | static int |
4079 | xfs_bmapi_allocate( | |
a2ceac1f DC |
4080 | struct xfs_bmalloca *bma) |
4081 | { | |
4082 | struct xfs_mount *mp = bma->ip->i_mount; | |
1277a5e0 | 4083 | int whichfork = xfs_bmapi_whichfork(bma->flags); |
a2ceac1f DC |
4084 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4085 | int tmp_logflags = 0; | |
4086 | int error; | |
4087 | ||
4088 | ASSERT(bma->length > 0); | |
4089 | ||
4090 | /* | |
4091 | * For the wasdelay case, we could also just allocate the stuff asked | |
4092 | * for in this bmap call but that wouldn't be as good. | |
4093 | */ | |
4094 | if (bma->wasdel) { | |
4095 | bma->length = (xfs_extlen_t)bma->got.br_blockcount; | |
4096 | bma->offset = bma->got.br_startoff; | |
530bc0fc DW |
4097 | if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev)) |
4098 | bma->prev.br_startoff = NULLFILEOFF; | |
a2ceac1f DC |
4099 | } else { |
4100 | bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN); | |
4101 | if (!bma->eof) | |
4102 | bma->length = XFS_FILBLKS_MIN(bma->length, | |
4103 | bma->got.br_startoff - bma->offset); | |
4104 | } | |
4105 | ||
2ac7663a CH |
4106 | if (bma->flags & XFS_BMAPI_CONTIG) |
4107 | bma->minlen = bma->length; | |
4108 | else | |
4109 | bma->minlen = 1; | |
a2ceac1f | 4110 | |
3006cea4 CB |
4111 | if (bma->flags & XFS_BMAPI_METADATA) { |
4112 | if (unlikely(XFS_TEST_ERROR(false, mp, | |
4113 | XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT))) | |
4114 | error = xfs_bmap_exact_minlen_extent_alloc(bma); | |
4115 | else | |
4116 | error = xfs_bmap_btalloc(bma); | |
4117 | } else { | |
2ac7663a | 4118 | error = xfs_bmap_alloc_userdata(bma); |
3006cea4 | 4119 | } |
2ac7663a | 4120 | if (error || bma->blkno == NULLFSBLOCK) |
a2ceac1f DC |
4121 | return error; |
4122 | ||
c3a24cde CH |
4123 | if (bma->flags & XFS_BMAPI_ZERO) { |
4124 | error = xfs_zero_extent(bma->ip, bma->blkno, bma->length); | |
4125 | if (error) | |
4126 | return error; | |
4127 | } | |
4128 | ||
84094546 | 4129 | if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) |
a2ceac1f | 4130 | bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork); |
a2ceac1f DC |
4131 | /* |
4132 | * Bump the number of extents we've allocated | |
4133 | * in this call. | |
4134 | */ | |
4135 | bma->nallocs++; | |
4136 | ||
4137 | if (bma->cur) | |
116c6a88 | 4138 | bma->cur->bc_ino.flags = |
bbf943f8 | 4139 | bma->wasdel ? XFS_BTCUR_BMBT_WASDEL : 0; |
a2ceac1f DC |
4140 | |
4141 | bma->got.br_startoff = bma->offset; | |
4142 | bma->got.br_startblock = bma->blkno; | |
4143 | bma->got.br_blockcount = bma->length; | |
4144 | bma->got.br_state = XFS_EXT_NORM; | |
4145 | ||
edc9bb69 | 4146 | if (bma->flags & XFS_BMAPI_PREALLOC) |
a2ceac1f DC |
4147 | bma->got.br_state = XFS_EXT_UNWRITTEN; |
4148 | ||
4149 | if (bma->wasdel) | |
1277a5e0 | 4150 | error = xfs_bmap_add_extent_delay_real(bma, whichfork); |
a2ceac1f | 4151 | else |
972432f2 | 4152 | error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip, |
9788e059 | 4153 | whichfork, &bma->icur, &bma->cur, &bma->got, |
64e8b4a7 | 4154 | &bma->logflags, bma->flags); |
a2ceac1f DC |
4155 | |
4156 | bma->logflags |= tmp_logflags; | |
4157 | if (error) | |
4158 | return error; | |
4159 | ||
4160 | /* | |
4161 | * Update our extent pointer, given that xfs_bmap_add_extent_delay_real | |
4162 | * or xfs_bmap_add_extent_hole_real might have merged it into one of | |
4163 | * the neighbouring ones. | |
4164 | */ | |
9788e059 | 4165 | xfs_iext_get_extent(ifp, &bma->icur, &bma->got); |
a2ceac1f DC |
4166 | |
4167 | ASSERT(bma->got.br_startoff <= bma->offset); | |
4168 | ASSERT(bma->got.br_startoff + bma->got.br_blockcount >= | |
4169 | bma->offset + bma->length); | |
4170 | ASSERT(bma->got.br_state == XFS_EXT_NORM || | |
4171 | bma->got.br_state == XFS_EXT_UNWRITTEN); | |
4172 | return 0; | |
4173 | } | |
4174 | ||
a2ceac1f DC |
4175 | STATIC int |
4176 | xfs_bmapi_convert_unwritten( | |
4177 | struct xfs_bmalloca *bma, | |
4178 | struct xfs_bmbt_irec *mval, | |
4179 | xfs_filblks_t len, | |
4180 | int flags) | |
4181 | { | |
cb8a004a | 4182 | int whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4183 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4184 | int tmp_logflags = 0; | |
4185 | int error; | |
4186 | ||
4187 | /* check if we need to do unwritten->real conversion */ | |
4188 | if (mval->br_state == XFS_EXT_UNWRITTEN && | |
4189 | (flags & XFS_BMAPI_PREALLOC)) | |
4190 | return 0; | |
4191 | ||
4192 | /* check if we need to do real->unwritten conversion */ | |
4193 | if (mval->br_state == XFS_EXT_NORM && | |
4194 | (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) != | |
4195 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) | |
4196 | return 0; | |
4197 | ||
4198 | /* | |
4199 | * Modify (by adding) the state flag, if writing. | |
4200 | */ | |
4201 | ASSERT(mval->br_blockcount <= len); | |
84094546 | 4202 | if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) { |
a2ceac1f DC |
4203 | bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp, |
4204 | bma->ip, whichfork); | |
a2ceac1f DC |
4205 | } |
4206 | mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN) | |
4207 | ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
4208 | ||
9542ae13 DC |
4209 | /* |
4210 | * Before insertion into the bmbt, zero the range being converted | |
4211 | * if required. | |
4212 | */ | |
4213 | if (flags & XFS_BMAPI_ZERO) { | |
4214 | error = xfs_zero_extent(bma->ip, mval->br_startblock, | |
4215 | mval->br_blockcount); | |
4216 | if (error) | |
4217 | return error; | |
4218 | } | |
4219 | ||
4072e4b4 | 4220 | error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork, |
64e8b4a7 | 4221 | &bma->icur, &bma->cur, mval, &tmp_logflags); |
23fc058d BF |
4222 | /* |
4223 | * Log the inode core unconditionally in the unwritten extent conversion | |
4224 | * path because the conversion might not have done so (e.g., if the | |
4225 | * extent count hasn't changed). We need to make sure the inode is dirty | |
4226 | * in the transaction for the sake of fsync(), even if nothing has | |
4227 | * changed, because fsync() will not force the log for this transaction | |
4228 | * unless it sees the inode pinned. | |
4072e4b4 DW |
4229 | * |
4230 | * Note: If we're only converting cow fork extents, there aren't | |
4231 | * any on-disk updates to make, so we don't need to log anything. | |
23fc058d | 4232 | */ |
4072e4b4 DW |
4233 | if (whichfork != XFS_COW_FORK) |
4234 | bma->logflags |= tmp_logflags | XFS_ILOG_CORE; | |
a2ceac1f DC |
4235 | if (error) |
4236 | return error; | |
4237 | ||
4238 | /* | |
4239 | * Update our extent pointer, given that | |
4240 | * xfs_bmap_add_extent_unwritten_real might have merged it into one | |
4241 | * of the neighbouring ones. | |
4242 | */ | |
9788e059 | 4243 | xfs_iext_get_extent(ifp, &bma->icur, &bma->got); |
a2ceac1f DC |
4244 | |
4245 | /* | |
4246 | * We may have combined previously unwritten space with written space, | |
4247 | * so generate another request. | |
4248 | */ | |
4249 | if (mval->br_blockcount < len) | |
12b53197 | 4250 | return -EAGAIN; |
a2ceac1f DC |
4251 | return 0; |
4252 | } | |
4253 | ||
ee622798 CH |
4254 | static inline xfs_extlen_t |
4255 | xfs_bmapi_minleft( | |
4256 | struct xfs_trans *tp, | |
4257 | struct xfs_inode *ip, | |
4258 | int fork) | |
4259 | { | |
d967a68d CH |
4260 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, fork); |
4261 | ||
ee622798 CH |
4262 | if (tp && tp->t_firstblock != NULLFSBLOCK) |
4263 | return 0; | |
d967a68d | 4264 | if (ifp->if_format != XFS_DINODE_FMT_BTREE) |
ee622798 | 4265 | return 1; |
d967a68d | 4266 | return be16_to_cpu(ifp->if_broot->bb_level) + 1; |
ee622798 CH |
4267 | } |
4268 | ||
4269 | /* | |
4270 | * Log whatever the flags say, even if error. Otherwise we might miss detecting | |
4271 | * a case where the data is changed, there's an error, and it's not logged so we | |
4272 | * don't shutdown when we should. Don't bother logging extents/btree changes if | |
4273 | * we converted to the other format. | |
4274 | */ | |
4275 | static void | |
4276 | xfs_bmapi_finish( | |
4277 | struct xfs_bmalloca *bma, | |
4278 | int whichfork, | |
4279 | int error) | |
4280 | { | |
d967a68d CH |
4281 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4282 | ||
ee622798 | 4283 | if ((bma->logflags & xfs_ilog_fext(whichfork)) && |
d967a68d | 4284 | ifp->if_format != XFS_DINODE_FMT_EXTENTS) |
ee622798 CH |
4285 | bma->logflags &= ~xfs_ilog_fext(whichfork); |
4286 | else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) && | |
d967a68d | 4287 | ifp->if_format != XFS_DINODE_FMT_BTREE) |
ee622798 CH |
4288 | bma->logflags &= ~xfs_ilog_fbroot(whichfork); |
4289 | ||
4290 | if (bma->logflags) | |
4291 | xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags); | |
4292 | if (bma->cur) | |
4293 | xfs_btree_del_cursor(bma->cur, error); | |
4294 | } | |
4295 | ||
a2ceac1f DC |
4296 | /* |
4297 | * Map file blocks to filesystem blocks, and allocate blocks or convert the | |
4298 | * extent state if necessary. Details behaviour is controlled by the flags | |
4299 | * parameter. Only allocates blocks from a single allocation group, to avoid | |
4300 | * locking problems. | |
a2ceac1f DC |
4301 | */ |
4302 | int | |
4303 | xfs_bmapi_write( | |
4304 | struct xfs_trans *tp, /* transaction pointer */ | |
4305 | struct xfs_inode *ip, /* incore inode */ | |
4306 | xfs_fileoff_t bno, /* starting file offs. mapped */ | |
4307 | xfs_filblks_t len, /* length to map in file */ | |
4308 | int flags, /* XFS_BMAPI_... */ | |
a2ceac1f DC |
4309 | xfs_extlen_t total, /* total blocks needed */ |
4310 | struct xfs_bmbt_irec *mval, /* output: map values */ | |
cbdfb3ab | 4311 | int *nmap) /* i/o: mval size/count */ |
a2ceac1f | 4312 | { |
93ad4d9c DW |
4313 | struct xfs_bmalloca bma = { |
4314 | .tp = tp, | |
4315 | .ip = ip, | |
4316 | .total = total, | |
4317 | }; | |
a2ceac1f | 4318 | struct xfs_mount *mp = ip->i_mount; |
d967a68d CH |
4319 | int whichfork = xfs_bmapi_whichfork(flags); |
4320 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
a2ceac1f | 4321 | xfs_fileoff_t end; /* end of mapped file region */ |
82411945 | 4322 | bool eof = false; /* after the end of extents */ |
a2ceac1f DC |
4323 | int error; /* error return */ |
4324 | int n; /* current extent index */ | |
4325 | xfs_fileoff_t obno; /* old block number (offset) */ | |
a2ceac1f DC |
4326 | |
4327 | #ifdef DEBUG | |
4328 | xfs_fileoff_t orig_bno; /* original block number value */ | |
4329 | int orig_flags; /* original flags arg value */ | |
4330 | xfs_filblks_t orig_len; /* original value of len arg */ | |
4331 | struct xfs_bmbt_irec *orig_mval; /* original value of mval */ | |
4332 | int orig_nmap; /* original value of *nmap */ | |
4333 | ||
4334 | orig_bno = bno; | |
4335 | orig_len = len; | |
4336 | orig_flags = flags; | |
4337 | orig_mval = mval; | |
4338 | orig_nmap = *nmap; | |
4339 | #endif | |
4340 | ||
4341 | ASSERT(*nmap >= 1); | |
4342 | ASSERT(*nmap <= XFS_BMAP_MAX_NMAP); | |
9fa4db19 | 4343 | ASSERT(tp != NULL); |
a2ceac1f | 4344 | ASSERT(len > 0); |
d967a68d | 4345 | ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL); |
ff105f75 | 4346 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
0b44aa85 | 4347 | ASSERT(!(flags & XFS_BMAPI_REMAP)); |
a2ceac1f | 4348 | |
9542ae13 DC |
4349 | /* zeroing is for currently only for data extents, not metadata */ |
4350 | ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) != | |
4351 | (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)); | |
4352 | /* | |
4353 | * we can allocate unwritten extents or pre-zero allocated blocks, | |
4354 | * but it makes no sense to do both at once. This would result in | |
4355 | * zeroing the unwritten extent twice, but it still being an | |
4356 | * unwritten extent.... | |
4357 | */ | |
4358 | ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) != | |
4359 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)); | |
4360 | ||
d967a68d | 4361 | if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) || |
bc73da84 | 4362 | XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) { |
12b53197 | 4363 | return -EFSCORRUPTED; |
a2ceac1f DC |
4364 | } |
4365 | ||
93adb06a | 4366 | if (xfs_is_shutdown(mp)) |
12b53197 | 4367 | return -EIO; |
a2ceac1f | 4368 | |
79896434 | 4369 | XFS_STATS_INC(mp, xs_blk_mapw); |
a2ceac1f | 4370 | |
e00c57e7 CH |
4371 | error = xfs_iread_extents(tp, ip, whichfork); |
4372 | if (error) | |
4373 | goto error0; | |
a2ceac1f | 4374 | |
9788e059 | 4375 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got)) |
82411945 | 4376 | eof = true; |
9788e059 | 4377 | if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev)) |
82411945 | 4378 | bma.prev.br_startoff = NULLFILEOFF; |
ee622798 | 4379 | bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork); |
a2ceac1f | 4380 | |
badaa597 BF |
4381 | n = 0; |
4382 | end = bno + len; | |
4383 | obno = bno; | |
a2ceac1f | 4384 | while (bno < end && n < *nmap) { |
7075a23f CH |
4385 | bool need_alloc = false, wasdelay = false; |
4386 | ||
faaad1df | 4387 | /* in hole or beyond EOF? */ |
7075a23f | 4388 | if (eof || bma.got.br_startoff > bno) { |
faaad1df DW |
4389 | /* |
4390 | * CoW fork conversions should /never/ hit EOF or | |
4391 | * holes. There should always be something for us | |
4392 | * to work on. | |
4393 | */ | |
4394 | ASSERT(!((flags & XFS_BMAPI_CONVERT) && | |
4395 | (flags & XFS_BMAPI_COWFORK))); | |
4396 | ||
330a35fe | 4397 | need_alloc = true; |
0b44aa85 CH |
4398 | } else if (isnullstartblock(bma.got.br_startblock)) { |
4399 | wasdelay = true; | |
7075a23f | 4400 | } |
34621a47 | 4401 | |
2bd0ea18 | 4402 | /* |
a2ceac1f DC |
4403 | * First, deal with the hole before the allocated space |
4404 | * that we found, if any. | |
2bd0ea18 | 4405 | */ |
9fa4db19 | 4406 | if (need_alloc || wasdelay) { |
a2ceac1f DC |
4407 | bma.eof = eof; |
4408 | bma.conv = !!(flags & XFS_BMAPI_CONVERT); | |
4409 | bma.wasdel = wasdelay; | |
4410 | bma.offset = bno; | |
4411 | bma.flags = flags; | |
4412 | ||
2bd0ea18 | 4413 | /* |
a2ceac1f DC |
4414 | * There's a 32/64 bit type mismatch between the |
4415 | * allocation length request (which can be 64 bits in | |
4416 | * length) and the bma length request, which is | |
4417 | * xfs_extlen_t and therefore 32 bits. Hence we have to | |
4418 | * check for 32-bit overflows and handle them here. | |
2bd0ea18 | 4419 | */ |
a2ceac1f DC |
4420 | if (len > (xfs_filblks_t)MAXEXTLEN) |
4421 | bma.length = MAXEXTLEN; | |
4422 | else | |
4423 | bma.length = len; | |
4424 | ||
4425 | ASSERT(len > 0); | |
4426 | ASSERT(bma.length > 0); | |
4427 | error = xfs_bmapi_allocate(&bma); | |
2bd0ea18 NS |
4428 | if (error) |
4429 | goto error0; | |
a2ceac1f DC |
4430 | if (bma.blkno == NULLFSBLOCK) |
4431 | break; | |
10e65503 DW |
4432 | |
4433 | /* | |
4434 | * If this is a CoW allocation, record the data in | |
4435 | * the refcount btree for orphan recovery. | |
4436 | */ | |
5965a482 DW |
4437 | if (whichfork == XFS_COW_FORK) |
4438 | xfs_refcount_alloc_cow_extent(tp, bma.blkno, | |
4439 | bma.length); | |
2bd0ea18 NS |
4440 | } |
4441 | ||
a2ceac1f DC |
4442 | /* Deal with the allocated space we found. */ |
4443 | xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno, | |
4444 | end, n, flags); | |
4445 | ||
4446 | /* Execute unwritten extent conversion if necessary */ | |
4447 | error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags); | |
12b53197 | 4448 | if (error == -EAGAIN) |
a2ceac1f DC |
4449 | continue; |
4450 | if (error) | |
4451 | goto error0; | |
4452 | ||
4453 | /* update the extent map to return */ | |
4454 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4455 | ||
2bd0ea18 NS |
4456 | /* |
4457 | * If we're done, stop now. Stop when we've allocated | |
4458 | * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise | |
4459 | * the transaction may get too big. | |
4460 | */ | |
a2ceac1f | 4461 | if (bno >= end || n >= *nmap || bma.nallocs >= *nmap) |
2bd0ea18 | 4462 | break; |
a2ceac1f DC |
4463 | |
4464 | /* Else go on to the next record. */ | |
4465 | bma.prev = bma.got; | |
9788e059 | 4466 | if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got)) |
82411945 | 4467 | eof = true; |
2bd0ea18 | 4468 | } |
2bd0ea18 | 4469 | *nmap = n; |
a2ceac1f | 4470 | |
939ebc1a CH |
4471 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags, |
4472 | whichfork); | |
4473 | if (error) | |
4474 | goto error0; | |
a2ceac1f | 4475 | |
d967a68d | 4476 | ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE || |
87c472b7 | 4477 | ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork)); |
ee622798 CH |
4478 | xfs_bmapi_finish(&bma, whichfork, 0); |
4479 | xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval, | |
4480 | orig_nmap, *nmap); | |
4481 | return 0; | |
2bd0ea18 | 4482 | error0: |
ee622798 | 4483 | xfs_bmapi_finish(&bma, whichfork, error); |
2bd0ea18 NS |
4484 | return error; |
4485 | } | |
4486 | ||
badaa597 BF |
4487 | /* |
4488 | * Convert an existing delalloc extent to real blocks based on file offset. This | |
4489 | * attempts to allocate the entire delalloc extent and may require multiple | |
4490 | * invocations to allocate the target offset if a large enough physical extent | |
4491 | * is not available. | |
4492 | */ | |
4493 | int | |
4494 | xfs_bmapi_convert_delalloc( | |
badaa597 | 4495 | struct xfs_inode *ip, |
badaa597 | 4496 | int whichfork, |
75f533e6 CH |
4497 | xfs_off_t offset, |
4498 | struct iomap *iomap, | |
c784c9d2 | 4499 | unsigned int *seq) |
badaa597 | 4500 | { |
330a35fe | 4501 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
c784c9d2 | 4502 | struct xfs_mount *mp = ip->i_mount; |
75f533e6 | 4503 | xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); |
330a35fe | 4504 | struct xfs_bmalloca bma = { NULL }; |
75f533e6 | 4505 | uint16_t flags = 0; |
c784c9d2 | 4506 | struct xfs_trans *tp; |
badaa597 | 4507 | int error; |
badaa597 | 4508 | |
75f533e6 CH |
4509 | if (whichfork == XFS_COW_FORK) |
4510 | flags |= IOMAP_F_SHARED; | |
4511 | ||
c784c9d2 CH |
4512 | /* |
4513 | * Space for the extent and indirect blocks was reserved when the | |
4514 | * delalloc extent was created so there's no need to do so here. | |
4515 | */ | |
4516 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, | |
4517 | XFS_TRANS_RESERVE, &tp); | |
4518 | if (error) | |
4519 | return error; | |
4520 | ||
4521 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
75cb5a60 CB |
4522 | |
4523 | error = xfs_iext_count_may_overflow(ip, whichfork, | |
4524 | XFS_IEXT_ADD_NOSPLIT_CNT); | |
4525 | if (error) | |
4526 | goto out_trans_cancel; | |
4527 | ||
c784c9d2 CH |
4528 | xfs_trans_ijoin(tp, ip, 0); |
4529 | ||
330a35fe CH |
4530 | if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) || |
4531 | bma.got.br_startoff > offset_fsb) { | |
4532 | /* | |
4533 | * No extent found in the range we are trying to convert. This | |
4534 | * should only happen for the COW fork, where another thread | |
4535 | * might have moved the extent to the data fork in the meantime. | |
4536 | */ | |
4537 | WARN_ON_ONCE(whichfork != XFS_COW_FORK); | |
c784c9d2 CH |
4538 | error = -EAGAIN; |
4539 | goto out_trans_cancel; | |
330a35fe | 4540 | } |
badaa597 BF |
4541 | |
4542 | /* | |
330a35fe CH |
4543 | * If we find a real extent here we raced with another thread converting |
4544 | * the extent. Just return the real extent at this offset. | |
badaa597 | 4545 | */ |
330a35fe | 4546 | if (!isnullstartblock(bma.got.br_startblock)) { |
8b4c5cd1 | 4547 | xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags); |
c784c9d2 CH |
4548 | *seq = READ_ONCE(ifp->if_seq); |
4549 | goto out_trans_cancel; | |
330a35fe CH |
4550 | } |
4551 | ||
4552 | bma.tp = tp; | |
4553 | bma.ip = ip; | |
4554 | bma.wasdel = true; | |
4555 | bma.offset = bma.got.br_startoff; | |
4556 | bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount, MAXEXTLEN); | |
330a35fe | 4557 | bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork); |
edc9bb69 DW |
4558 | |
4559 | /* | |
4560 | * When we're converting the delalloc reservations backing dirty pages | |
4561 | * in the page cache, we must be careful about how we create the new | |
4562 | * extents: | |
4563 | * | |
4564 | * New CoW fork extents are created unwritten, turned into real extents | |
4565 | * when we're about to write the data to disk, and mapped into the data | |
4566 | * fork after the write finishes. End of story. | |
4567 | * | |
4568 | * New data fork extents must be mapped in as unwritten and converted | |
4569 | * to real extents after the write succeeds to avoid exposing stale | |
4570 | * disk contents if we crash. | |
4571 | */ | |
4572 | bma.flags = XFS_BMAPI_PREALLOC; | |
330a35fe | 4573 | if (whichfork == XFS_COW_FORK) |
edc9bb69 | 4574 | bma.flags |= XFS_BMAPI_COWFORK; |
330a35fe CH |
4575 | |
4576 | if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev)) | |
4577 | bma.prev.br_startoff = NULLFILEOFF; | |
4578 | ||
4579 | error = xfs_bmapi_allocate(&bma); | |
4580 | if (error) | |
4581 | goto out_finish; | |
4582 | ||
4583 | error = -ENOSPC; | |
4584 | if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK)) | |
4585 | goto out_finish; | |
4586 | error = -EFSCORRUPTED; | |
14790ed0 | 4587 | if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock))) |
330a35fe CH |
4588 | goto out_finish; |
4589 | ||
44e165df CH |
4590 | XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length)); |
4591 | XFS_STATS_INC(mp, xs_xstrat_quick); | |
4592 | ||
330a35fe | 4593 | ASSERT(!isnullstartblock(bma.got.br_startblock)); |
8b4c5cd1 | 4594 | xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags); |
c784c9d2 | 4595 | *seq = READ_ONCE(ifp->if_seq); |
330a35fe | 4596 | |
5965a482 DW |
4597 | if (whichfork == XFS_COW_FORK) |
4598 | xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length); | |
330a35fe CH |
4599 | |
4600 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags, | |
4601 | whichfork); | |
c784c9d2 CH |
4602 | if (error) |
4603 | goto out_finish; | |
4604 | ||
4605 | xfs_bmapi_finish(&bma, whichfork, 0); | |
4606 | error = xfs_trans_commit(tp); | |
4607 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4608 | return error; | |
4609 | ||
330a35fe CH |
4610 | out_finish: |
4611 | xfs_bmapi_finish(&bma, whichfork, error); | |
c784c9d2 CH |
4612 | out_trans_cancel: |
4613 | xfs_trans_cancel(tp); | |
4614 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
badaa597 BF |
4615 | return error; |
4616 | } | |
4617 | ||
26d6a481 | 4618 | int |
0b44aa85 CH |
4619 | xfs_bmapi_remap( |
4620 | struct xfs_trans *tp, | |
4621 | struct xfs_inode *ip, | |
4622 | xfs_fileoff_t bno, | |
4623 | xfs_filblks_t len, | |
4624 | xfs_fsblock_t startblock, | |
26d6a481 | 4625 | int flags) |
0b44aa85 CH |
4626 | { |
4627 | struct xfs_mount *mp = ip->i_mount; | |
26d6a481 | 4628 | struct xfs_ifork *ifp; |
0b44aa85 | 4629 | struct xfs_btree_cur *cur = NULL; |
0b44aa85 | 4630 | struct xfs_bmbt_irec got; |
9788e059 | 4631 | struct xfs_iext_cursor icur; |
26d6a481 | 4632 | int whichfork = xfs_bmapi_whichfork(flags); |
0b44aa85 CH |
4633 | int logflags = 0, error; |
4634 | ||
26d6a481 | 4635 | ifp = XFS_IFORK_PTR(ip, whichfork); |
0b44aa85 CH |
4636 | ASSERT(len > 0); |
4637 | ASSERT(len <= (xfs_filblks_t)MAXEXTLEN); | |
4638 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
36cfb334 DW |
4639 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC | |
4640 | XFS_BMAPI_NORMAP))); | |
4641 | ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) != | |
4642 | (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)); | |
0b44aa85 | 4643 | |
d967a68d | 4644 | if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) || |
bc73da84 | 4645 | XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) { |
0b44aa85 CH |
4646 | return -EFSCORRUPTED; |
4647 | } | |
4648 | ||
93adb06a | 4649 | if (xfs_is_shutdown(mp)) |
0b44aa85 CH |
4650 | return -EIO; |
4651 | ||
e00c57e7 CH |
4652 | error = xfs_iread_extents(tp, ip, whichfork); |
4653 | if (error) | |
4654 | return error; | |
0b44aa85 | 4655 | |
9788e059 | 4656 | if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) { |
0b44aa85 CH |
4657 | /* make sure we only reflink into a hole. */ |
4658 | ASSERT(got.br_startoff > bno); | |
4659 | ASSERT(got.br_startoff - bno >= len); | |
4660 | } | |
4661 | ||
aa00f286 | 4662 | ip->i_nblocks += len; |
05422db6 | 4663 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
0b44aa85 | 4664 | |
84094546 | 4665 | if (ifp->if_format == XFS_DINODE_FMT_BTREE) { |
26d6a481 | 4666 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
116c6a88 | 4667 | cur->bc_ino.flags = 0; |
0b44aa85 CH |
4668 | } |
4669 | ||
4670 | got.br_startoff = bno; | |
4671 | got.br_startblock = startblock; | |
4672 | got.br_blockcount = len; | |
36cfb334 DW |
4673 | if (flags & XFS_BMAPI_PREALLOC) |
4674 | got.br_state = XFS_EXT_UNWRITTEN; | |
4675 | else | |
4676 | got.br_state = XFS_EXT_NORM; | |
0b44aa85 | 4677 | |
26d6a481 | 4678 | error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur, |
64e8b4a7 | 4679 | &cur, &got, &logflags, flags); |
0b44aa85 CH |
4680 | if (error) |
4681 | goto error0; | |
4682 | ||
939ebc1a | 4683 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork); |
0b44aa85 CH |
4684 | |
4685 | error0: | |
d967a68d | 4686 | if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS) |
0b44aa85 | 4687 | logflags &= ~XFS_ILOG_DEXT; |
d967a68d | 4688 | else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE) |
0b44aa85 CH |
4689 | logflags &= ~XFS_ILOG_DBROOT; |
4690 | ||
4691 | if (logflags) | |
4692 | xfs_trans_log_inode(tp, ip, logflags); | |
660265b7 DW |
4693 | if (cur) |
4694 | xfs_btree_del_cursor(cur, error); | |
0b44aa85 CH |
4695 | return error; |
4696 | } | |
4697 | ||
01d1b786 BF |
4698 | /* |
4699 | * When a delalloc extent is split (e.g., due to a hole punch), the original | |
4700 | * indlen reservation must be shared across the two new extents that are left | |
4701 | * behind. | |
4702 | * | |
4703 | * Given the original reservation and the worst case indlen for the two new | |
4704 | * extents (as calculated by xfs_bmap_worst_indlen()), split the original | |
731ccdf9 BF |
4705 | * reservation fairly across the two new extents. If necessary, steal available |
4706 | * blocks from a deleted extent to make up a reservation deficiency (e.g., if | |
4707 | * ores == 1). The number of stolen blocks is returned. The availability and | |
4708 | * subsequent accounting of stolen blocks is the responsibility of the caller. | |
01d1b786 | 4709 | */ |
731ccdf9 | 4710 | static xfs_filblks_t |
01d1b786 BF |
4711 | xfs_bmap_split_indlen( |
4712 | xfs_filblks_t ores, /* original res. */ | |
4713 | xfs_filblks_t *indlen1, /* ext1 worst indlen */ | |
731ccdf9 BF |
4714 | xfs_filblks_t *indlen2, /* ext2 worst indlen */ |
4715 | xfs_filblks_t avail) /* stealable blocks */ | |
01d1b786 BF |
4716 | { |
4717 | xfs_filblks_t len1 = *indlen1; | |
4718 | xfs_filblks_t len2 = *indlen2; | |
4719 | xfs_filblks_t nres = len1 + len2; /* new total res. */ | |
731ccdf9 | 4720 | xfs_filblks_t stolen = 0; |
775762e4 | 4721 | xfs_filblks_t resfactor; |
731ccdf9 BF |
4722 | |
4723 | /* | |
4724 | * Steal as many blocks as we can to try and satisfy the worst case | |
4725 | * indlen for both new extents. | |
4726 | */ | |
775762e4 BF |
4727 | if (ores < nres && avail) |
4728 | stolen = XFS_FILBLKS_MIN(nres - ores, avail); | |
4729 | ores += stolen; | |
4730 | ||
4731 | /* nothing else to do if we've satisfied the new reservation */ | |
4732 | if (ores >= nres) | |
4733 | return stolen; | |
4734 | ||
4735 | /* | |
4736 | * We can't meet the total required reservation for the two extents. | |
4737 | * Calculate the percent of the overall shortage between both extents | |
4738 | * and apply this percentage to each of the requested indlen values. | |
4739 | * This distributes the shortage fairly and reduces the chances that one | |
4740 | * of the two extents is left with nothing when extents are repeatedly | |
4741 | * split. | |
4742 | */ | |
4743 | resfactor = (ores * 100); | |
4744 | do_div(resfactor, nres); | |
4745 | len1 *= resfactor; | |
4746 | do_div(len1, 100); | |
4747 | len2 *= resfactor; | |
4748 | do_div(len2, 100); | |
4749 | ASSERT(len1 + len2 <= ores); | |
4750 | ASSERT(len1 < *indlen1 && len2 < *indlen2); | |
01d1b786 BF |
4751 | |
4752 | /* | |
775762e4 BF |
4753 | * Hand out the remainder to each extent. If one of the two reservations |
4754 | * is zero, we want to make sure that one gets a block first. The loop | |
4755 | * below starts with len1, so hand len2 a block right off the bat if it | |
4756 | * is zero. | |
01d1b786 | 4757 | */ |
775762e4 BF |
4758 | ores -= (len1 + len2); |
4759 | ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores); | |
4760 | if (ores && !len2 && *indlen2) { | |
4761 | len2++; | |
4762 | ores--; | |
4763 | } | |
4764 | while (ores) { | |
4765 | if (len1 < *indlen1) { | |
4766 | len1++; | |
4767 | ores--; | |
01d1b786 | 4768 | } |
775762e4 | 4769 | if (!ores) |
01d1b786 | 4770 | break; |
775762e4 BF |
4771 | if (len2 < *indlen2) { |
4772 | len2++; | |
4773 | ores--; | |
01d1b786 BF |
4774 | } |
4775 | } | |
4776 | ||
4777 | *indlen1 = len1; | |
4778 | *indlen2 = len2; | |
731ccdf9 BF |
4779 | |
4780 | return stolen; | |
01d1b786 BF |
4781 | } |
4782 | ||
ece930fa CH |
4783 | int |
4784 | xfs_bmap_del_extent_delay( | |
4785 | struct xfs_inode *ip, | |
4786 | int whichfork, | |
9788e059 | 4787 | struct xfs_iext_cursor *icur, |
ece930fa CH |
4788 | struct xfs_bmbt_irec *got, |
4789 | struct xfs_bmbt_irec *del) | |
4790 | { | |
4791 | struct xfs_mount *mp = ip->i_mount; | |
4792 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
4793 | struct xfs_bmbt_irec new; | |
4794 | int64_t da_old, da_new, da_diff = 0; | |
4795 | xfs_fileoff_t del_endoff, got_endoff; | |
4796 | xfs_filblks_t got_indlen, new_indlen, stolen; | |
7bd43334 CH |
4797 | int state = xfs_bmap_fork_to_state(whichfork); |
4798 | int error = 0; | |
ece930fa CH |
4799 | bool isrt; |
4800 | ||
4801 | XFS_STATS_INC(mp, xs_del_exlist); | |
4802 | ||
4803 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); | |
4804 | del_endoff = del->br_startoff + del->br_blockcount; | |
4805 | got_endoff = got->br_startoff + got->br_blockcount; | |
4806 | da_old = startblockval(got->br_startblock); | |
4807 | da_new = 0; | |
4808 | ||
ece930fa CH |
4809 | ASSERT(del->br_blockcount > 0); |
4810 | ASSERT(got->br_startoff <= del->br_startoff); | |
4811 | ASSERT(got_endoff >= del_endoff); | |
4812 | ||
4813 | if (isrt) { | |
f1e0bd8d | 4814 | uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount); |
ece930fa CH |
4815 | |
4816 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
4817 | xfs_mod_frextents(mp, rtexts); | |
4818 | } | |
4819 | ||
4820 | /* | |
4821 | * Update the inode delalloc counter now and wait to update the | |
4822 | * sb counters as we might have to borrow some blocks for the | |
4823 | * indirect block accounting. | |
4824 | */ | |
9fcc3af9 DW |
4825 | ASSERT(!isrt); |
4826 | error = xfs_quota_unreserve_blkres(ip, del->br_blockcount); | |
6921638c DW |
4827 | if (error) |
4828 | return error; | |
ece930fa CH |
4829 | ip->i_delayed_blks -= del->br_blockcount; |
4830 | ||
ece930fa | 4831 | if (got->br_startoff == del->br_startoff) |
d0a03e5a | 4832 | state |= BMAP_LEFT_FILLING; |
ece930fa | 4833 | if (got_endoff == del_endoff) |
d0a03e5a | 4834 | state |= BMAP_RIGHT_FILLING; |
ece930fa | 4835 | |
d0a03e5a CH |
4836 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4837 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
ece930fa CH |
4838 | /* |
4839 | * Matches the whole extent. Delete the entry. | |
4840 | */ | |
cf455a62 | 4841 | xfs_iext_remove(ip, icur, state); |
9788e059 | 4842 | xfs_iext_prev(ifp, icur); |
ece930fa | 4843 | break; |
d0a03e5a | 4844 | case BMAP_LEFT_FILLING: |
ece930fa CH |
4845 | /* |
4846 | * Deleting the first part of the extent. | |
4847 | */ | |
ece930fa CH |
4848 | got->br_startoff = del_endoff; |
4849 | got->br_blockcount -= del->br_blockcount; | |
4850 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4851 | got->br_blockcount), da_old); | |
4852 | got->br_startblock = nullstartblock((int)da_new); | |
9788e059 | 4853 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa | 4854 | break; |
d0a03e5a | 4855 | case BMAP_RIGHT_FILLING: |
ece930fa CH |
4856 | /* |
4857 | * Deleting the last part of the extent. | |
4858 | */ | |
ece930fa CH |
4859 | got->br_blockcount = got->br_blockcount - del->br_blockcount; |
4860 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4861 | got->br_blockcount), da_old); | |
4862 | got->br_startblock = nullstartblock((int)da_new); | |
9788e059 | 4863 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa CH |
4864 | break; |
4865 | case 0: | |
4866 | /* | |
4867 | * Deleting the middle of the extent. | |
4868 | * | |
4869 | * Distribute the original indlen reservation across the two new | |
4870 | * extents. Steal blocks from the deleted extent if necessary. | |
4871 | * Stealing blocks simply fudges the fdblocks accounting below. | |
4872 | * Warn if either of the new indlen reservations is zero as this | |
4873 | * can lead to delalloc problems. | |
4874 | */ | |
ece930fa CH |
4875 | got->br_blockcount = del->br_startoff - got->br_startoff; |
4876 | got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount); | |
4877 | ||
4878 | new.br_blockcount = got_endoff - del_endoff; | |
4879 | new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
4880 | ||
4881 | WARN_ON_ONCE(!got_indlen || !new_indlen); | |
4882 | stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen, | |
4883 | del->br_blockcount); | |
4884 | ||
4885 | got->br_startblock = nullstartblock((int)got_indlen); | |
ece930fa CH |
4886 | |
4887 | new.br_startoff = del_endoff; | |
4888 | new.br_state = got->br_state; | |
4889 | new.br_startblock = nullstartblock((int)new_indlen); | |
4890 | ||
9788e059 CH |
4891 | xfs_iext_update_extent(ip, state, icur, got); |
4892 | xfs_iext_next(ifp, icur); | |
26a75f67 | 4893 | xfs_iext_insert(ip, icur, &new, state); |
ece930fa CH |
4894 | |
4895 | da_new = got_indlen + new_indlen - stolen; | |
4896 | del->br_blockcount -= stolen; | |
4897 | break; | |
4898 | } | |
4899 | ||
4900 | ASSERT(da_old >= da_new); | |
4901 | da_diff = da_old - da_new; | |
4902 | if (!isrt) | |
4903 | da_diff += del->br_blockcount; | |
f73690fe | 4904 | if (da_diff) { |
ece930fa | 4905 | xfs_mod_fdblocks(mp, da_diff, false); |
f73690fe DW |
4906 | xfs_mod_delalloc(mp, -da_diff); |
4907 | } | |
ece930fa CH |
4908 | return error; |
4909 | } | |
4910 | ||
4911 | void | |
4912 | xfs_bmap_del_extent_cow( | |
4913 | struct xfs_inode *ip, | |
9788e059 | 4914 | struct xfs_iext_cursor *icur, |
ece930fa CH |
4915 | struct xfs_bmbt_irec *got, |
4916 | struct xfs_bmbt_irec *del) | |
4917 | { | |
4918 | struct xfs_mount *mp = ip->i_mount; | |
4919 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
4920 | struct xfs_bmbt_irec new; | |
4921 | xfs_fileoff_t del_endoff, got_endoff; | |
4922 | int state = BMAP_COWFORK; | |
4923 | ||
4924 | XFS_STATS_INC(mp, xs_del_exlist); | |
4925 | ||
4926 | del_endoff = del->br_startoff + del->br_blockcount; | |
4927 | got_endoff = got->br_startoff + got->br_blockcount; | |
4928 | ||
ece930fa CH |
4929 | ASSERT(del->br_blockcount > 0); |
4930 | ASSERT(got->br_startoff <= del->br_startoff); | |
4931 | ASSERT(got_endoff >= del_endoff); | |
4932 | ASSERT(!isnullstartblock(got->br_startblock)); | |
4933 | ||
4934 | if (got->br_startoff == del->br_startoff) | |
d0a03e5a | 4935 | state |= BMAP_LEFT_FILLING; |
ece930fa | 4936 | if (got_endoff == del_endoff) |
d0a03e5a | 4937 | state |= BMAP_RIGHT_FILLING; |
ece930fa | 4938 | |
d0a03e5a CH |
4939 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4940 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
ece930fa CH |
4941 | /* |
4942 | * Matches the whole extent. Delete the entry. | |
4943 | */ | |
cf455a62 | 4944 | xfs_iext_remove(ip, icur, state); |
9788e059 | 4945 | xfs_iext_prev(ifp, icur); |
ece930fa | 4946 | break; |
d0a03e5a | 4947 | case BMAP_LEFT_FILLING: |
ece930fa CH |
4948 | /* |
4949 | * Deleting the first part of the extent. | |
4950 | */ | |
ece930fa CH |
4951 | got->br_startoff = del_endoff; |
4952 | got->br_blockcount -= del->br_blockcount; | |
4953 | got->br_startblock = del->br_startblock + del->br_blockcount; | |
9788e059 | 4954 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa | 4955 | break; |
d0a03e5a | 4956 | case BMAP_RIGHT_FILLING: |
ece930fa CH |
4957 | /* |
4958 | * Deleting the last part of the extent. | |
4959 | */ | |
ece930fa | 4960 | got->br_blockcount -= del->br_blockcount; |
9788e059 | 4961 | xfs_iext_update_extent(ip, state, icur, got); |
ece930fa CH |
4962 | break; |
4963 | case 0: | |
4964 | /* | |
4965 | * Deleting the middle of the extent. | |
4966 | */ | |
ece930fa | 4967 | got->br_blockcount = del->br_startoff - got->br_startoff; |
ece930fa CH |
4968 | |
4969 | new.br_startoff = del_endoff; | |
4970 | new.br_blockcount = got_endoff - del_endoff; | |
4971 | new.br_state = got->br_state; | |
4972 | new.br_startblock = del->br_startblock + del->br_blockcount; | |
4973 | ||
9788e059 CH |
4974 | xfs_iext_update_extent(ip, state, icur, got); |
4975 | xfs_iext_next(ifp, icur); | |
26a75f67 | 4976 | xfs_iext_insert(ip, icur, &new, state); |
ece930fa CH |
4977 | break; |
4978 | } | |
d07cc724 | 4979 | ip->i_delayed_blks -= del->br_blockcount; |
ece930fa CH |
4980 | } |
4981 | ||
2bd0ea18 | 4982 | /* |
49f693fa | 4983 | * Called by xfs_bmapi to update file extent records and the btree |
ad68fd19 | 4984 | * after removing space. |
2bd0ea18 | 4985 | */ |
49f693fa | 4986 | STATIC int /* error */ |
ad68fd19 | 4987 | xfs_bmap_del_extent_real( |
49f693fa DC |
4988 | xfs_inode_t *ip, /* incore inode pointer */ |
4989 | xfs_trans_t *tp, /* current transaction pointer */ | |
9788e059 | 4990 | struct xfs_iext_cursor *icur, |
ec924d04 | 4991 | struct xfs_btree_cur *cur, /* if null, not a btree */ |
49f693fa DC |
4992 | xfs_bmbt_irec_t *del, /* data to remove from extents */ |
4993 | int *logflagsp, /* inode logging flags */ | |
36b16da8 DW |
4994 | int whichfork, /* data or attr fork */ |
4995 | int bflags) /* bmapi flags */ | |
2bd0ea18 | 4996 | { |
49f693fa DC |
4997 | xfs_fsblock_t del_endblock=0; /* first block past del */ |
4998 | xfs_fileoff_t del_endoff; /* first offset past del */ | |
49f693fa | 4999 | int do_fx; /* free extent at end of routine */ |
49f693fa | 5000 | int error; /* error return value */ |
bd92a38b | 5001 | int flags = 0;/* inode logging flags */ |
70bf7533 | 5002 | struct xfs_bmbt_irec got; /* current extent entry */ |
49f693fa DC |
5003 | xfs_fileoff_t got_endoff; /* first offset past got */ |
5004 | int i; /* temp state */ | |
e07055b8 | 5005 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa DC |
5006 | xfs_mount_t *mp; /* mount structure */ |
5007 | xfs_filblks_t nblks; /* quota/sb block count */ | |
5008 | xfs_bmbt_irec_t new; /* new record to be inserted */ | |
5009 | /* REFERENCED */ | |
5010 | uint qfield; /* quota field to update */ | |
7bd43334 | 5011 | int state = xfs_bmap_fork_to_state(whichfork); |
70bf7533 | 5012 | struct xfs_bmbt_irec old; |
a2ceac1f | 5013 | |
79896434 BD |
5014 | mp = ip->i_mount; |
5015 | XFS_STATS_INC(mp, xs_del_exlist); | |
a2ceac1f | 5016 | |
49f693fa | 5017 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa | 5018 | ASSERT(del->br_blockcount > 0); |
9788e059 | 5019 | xfs_iext_get_extent(ifp, icur, &got); |
49f693fa DC |
5020 | ASSERT(got.br_startoff <= del->br_startoff); |
5021 | del_endoff = del->br_startoff + del->br_blockcount; | |
5022 | got_endoff = got.br_startoff + got.br_blockcount; | |
5023 | ASSERT(got_endoff >= del_endoff); | |
ad68fd19 | 5024 | ASSERT(!isnullstartblock(got.br_startblock)); |
49f693fa DC |
5025 | qfield = 0; |
5026 | error = 0; | |
ad68fd19 | 5027 | |
bd92a38b CH |
5028 | /* |
5029 | * If it's the case where the directory code is running with no block | |
5030 | * reservation, and the deleted block is in the middle of its extent, | |
5031 | * and the resulting insert of an extent would cause transformation to | |
5032 | * btree format, then reject it. The calling code will then swap blocks | |
5033 | * around instead. We have to do this now, rather than waiting for the | |
5034 | * conversion to btree format, since the transaction will be dirty then. | |
5035 | */ | |
5036 | if (tp->t_blk_res == 0 && | |
d967a68d | 5037 | ifp->if_format == XFS_DINODE_FMT_EXTENTS && |
87c472b7 | 5038 | ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) && |
bd92a38b CH |
5039 | del->br_startoff > got.br_startoff && del_endoff < got_endoff) |
5040 | return -ENOSPC; | |
5041 | ||
5042 | flags = XFS_ILOG_CORE; | |
ad68fd19 | 5043 | if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) { |
ad68fd19 | 5044 | xfs_filblks_t len; |
5a595099 DC |
5045 | xfs_extlen_t mod; |
5046 | ||
5a595099 DC |
5047 | len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize, |
5048 | &mod); | |
5049 | ASSERT(mod == 0); | |
ad68fd19 | 5050 | |
d5fe08a6 DW |
5051 | if (!(bflags & XFS_BMAPI_REMAP)) { |
5052 | xfs_fsblock_t bno; | |
5053 | ||
5054 | bno = div_u64_rem(del->br_startblock, | |
5055 | mp->m_sb.sb_rextsize, &mod); | |
5056 | ASSERT(mod == 0); | |
5057 | ||
5058 | error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len); | |
5059 | if (error) | |
5060 | goto done; | |
5061 | } | |
5062 | ||
49f693fa | 5063 | do_fx = 0; |
ad68fd19 CH |
5064 | nblks = len * mp->m_sb.sb_rextsize; |
5065 | qfield = XFS_TRANS_DQ_RTBCOUNT; | |
5066 | } else { | |
5067 | do_fx = 1; | |
5068 | nblks = del->br_blockcount; | |
5069 | qfield = XFS_TRANS_DQ_BCOUNT; | |
5070 | } | |
5071 | ||
5072 | del_endblock = del->br_startblock + del->br_blockcount; | |
5073 | if (cur) { | |
70a93110 | 5074 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
ad68fd19 CH |
5075 | if (error) |
5076 | goto done; | |
fbb4fa7f DW |
5077 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
5078 | error = -EFSCORRUPTED; | |
5079 | goto done; | |
5080 | } | |
49f693fa | 5081 | } |
85aec44f | 5082 | |
b039ac79 CH |
5083 | if (got.br_startoff == del->br_startoff) |
5084 | state |= BMAP_LEFT_FILLING; | |
5085 | if (got_endoff == del_endoff) | |
5086 | state |= BMAP_RIGHT_FILLING; | |
5087 | ||
5088 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { | |
5089 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
49f693fa DC |
5090 | /* |
5091 | * Matches the whole extent. Delete the entry. | |
5092 | */ | |
cf455a62 | 5093 | xfs_iext_remove(ip, icur, state); |
9788e059 | 5094 | xfs_iext_prev(ifp, icur); |
87c472b7 CH |
5095 | ifp->if_nextents--; |
5096 | ||
49f693fa DC |
5097 | flags |= XFS_ILOG_CORE; |
5098 | if (!cur) { | |
5099 | flags |= xfs_ilog_fext(whichfork); | |
5100 | break; | |
2bd0ea18 | 5101 | } |
49f693fa DC |
5102 | if ((error = xfs_btree_delete(cur, &i))) |
5103 | goto done; | |
fbb4fa7f DW |
5104 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
5105 | error = -EFSCORRUPTED; | |
5106 | goto done; | |
5107 | } | |
49f693fa | 5108 | break; |
b039ac79 | 5109 | case BMAP_LEFT_FILLING: |
2bd0ea18 | 5110 | /* |
49f693fa | 5111 | * Deleting the first part of the extent. |
2bd0ea18 | 5112 | */ |
70bf7533 CH |
5113 | got.br_startoff = del_endoff; |
5114 | got.br_startblock = del_endblock; | |
5115 | got.br_blockcount -= del->br_blockcount; | |
9788e059 | 5116 | xfs_iext_update_extent(ip, state, icur, &got); |
49f693fa DC |
5117 | if (!cur) { |
5118 | flags |= xfs_ilog_fext(whichfork); | |
5119 | break; | |
5120 | } | |
d0e5f1ff | 5121 | error = xfs_bmbt_update(cur, &got); |
70bf7533 | 5122 | if (error) |
49f693fa DC |
5123 | goto done; |
5124 | break; | |
b039ac79 | 5125 | case BMAP_RIGHT_FILLING: |
2bd0ea18 | 5126 | /* |
49f693fa | 5127 | * Deleting the last part of the extent. |
2bd0ea18 | 5128 | */ |
70bf7533 | 5129 | got.br_blockcount -= del->br_blockcount; |
9788e059 | 5130 | xfs_iext_update_extent(ip, state, icur, &got); |
49f693fa DC |
5131 | if (!cur) { |
5132 | flags |= xfs_ilog_fext(whichfork); | |
5133 | break; | |
5134 | } | |
d0e5f1ff | 5135 | error = xfs_bmbt_update(cur, &got); |
70bf7533 | 5136 | if (error) |
49f693fa DC |
5137 | goto done; |
5138 | break; | |
49f693fa DC |
5139 | case 0: |
5140 | /* | |
5141 | * Deleting the middle of the extent. | |
5142 | */ | |
7428ec5b CB |
5143 | |
5144 | /* | |
5145 | * For directories, -ENOSPC is returned since a directory entry | |
5146 | * remove operation must not fail due to low extent count | |
5147 | * availability. -ENOSPC will be handled by higher layers of XFS | |
5148 | * by letting the corresponding empty Data/Free blocks to linger | |
5149 | * until a future remove operation. Dabtree blocks would be | |
5150 | * swapped with the last block in the leaf space and then the | |
5151 | * new last block will be unmapped. | |
45140de8 CB |
5152 | * |
5153 | * The above logic also applies to the source directory entry of | |
5154 | * a rename operation. | |
7428ec5b CB |
5155 | */ |
5156 | error = xfs_iext_count_may_overflow(ip, whichfork, 1); | |
5157 | if (error) { | |
5158 | ASSERT(S_ISDIR(VFS_I(ip)->i_mode) && | |
5159 | whichfork == XFS_DATA_FORK); | |
5160 | error = -ENOSPC; | |
5161 | goto done; | |
5162 | } | |
5163 | ||
70bf7533 | 5164 | old = got; |
df926c07 | 5165 | |
70bf7533 | 5166 | got.br_blockcount = del->br_startoff - got.br_startoff; |
9788e059 | 5167 | xfs_iext_update_extent(ip, state, icur, &got); |
70bf7533 | 5168 | |
49f693fa | 5169 | new.br_startoff = del_endoff; |
70bf7533 | 5170 | new.br_blockcount = got_endoff - del_endoff; |
49f693fa | 5171 | new.br_state = got.br_state; |
ad68fd19 | 5172 | new.br_startblock = del_endblock; |
70bf7533 | 5173 | |
ad68fd19 CH |
5174 | flags |= XFS_ILOG_CORE; |
5175 | if (cur) { | |
d0e5f1ff | 5176 | error = xfs_bmbt_update(cur, &got); |
ad68fd19 CH |
5177 | if (error) |
5178 | goto done; | |
5179 | error = xfs_btree_increment(cur, 0, &i); | |
5180 | if (error) | |
5181 | goto done; | |
5182 | cur->bc_rec.b = new; | |
5183 | error = xfs_btree_insert(cur, &i); | |
5184 | if (error && error != -ENOSPC) | |
5185 | goto done; | |
5186 | /* | |
5187 | * If get no-space back from btree insert, it tried a | |
5188 | * split, and we have a zero block reservation. Fix up | |
5189 | * our state and return the error. | |
5190 | */ | |
5191 | if (error == -ENOSPC) { | |
49f693fa | 5192 | /* |
ad68fd19 CH |
5193 | * Reset the cursor, don't trust it after any |
5194 | * insert operation. | |
49f693fa | 5195 | */ |
70a93110 | 5196 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
ad68fd19 | 5197 | if (error) |
49f693fa | 5198 | goto done; |
fbb4fa7f DW |
5199 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
5200 | error = -EFSCORRUPTED; | |
5201 | goto done; | |
5202 | } | |
ad68fd19 CH |
5203 | /* |
5204 | * Update the btree record back | |
5205 | * to the original value. | |
5206 | */ | |
d0e5f1ff | 5207 | error = xfs_bmbt_update(cur, &old); |
ad68fd19 CH |
5208 | if (error) |
5209 | goto done; | |
5210 | /* | |
5211 | * Reset the extent record back | |
5212 | * to the original value. | |
5213 | */ | |
9788e059 | 5214 | xfs_iext_update_extent(ip, state, icur, &old); |
ad68fd19 CH |
5215 | flags = 0; |
5216 | error = -ENOSPC; | |
5217 | goto done; | |
5218 | } | |
fbb4fa7f DW |
5219 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
5220 | error = -EFSCORRUPTED; | |
5221 | goto done; | |
5222 | } | |
ad68fd19 CH |
5223 | } else |
5224 | flags |= xfs_ilog_fext(whichfork); | |
87c472b7 CH |
5225 | |
5226 | ifp->if_nextents++; | |
9788e059 | 5227 | xfs_iext_next(ifp, icur); |
26a75f67 | 5228 | xfs_iext_insert(ip, icur, &new, state); |
49f693fa | 5229 | break; |
2bd0ea18 | 5230 | } |
d7f80320 DW |
5231 | |
5232 | /* remove reverse mapping */ | |
46d29bb9 | 5233 | xfs_rmap_unmap_extent(tp, ip, whichfork, del); |
d7f80320 | 5234 | |
2bd0ea18 | 5235 | /* |
49f693fa | 5236 | * If we need to, add to list of extents to delete. |
2bd0ea18 | 5237 | */ |
36b16da8 | 5238 | if (do_fx && !(bflags & XFS_BMAPI_REMAP)) { |
cfe32f0d | 5239 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) { |
5965a482 | 5240 | xfs_refcount_decrease_extent(tp, del); |
3a13f959 | 5241 | } else { |
7d84b02d | 5242 | __xfs_free_extent_later(tp, del->br_startblock, |
cc5af22a BF |
5243 | del->br_blockcount, NULL, |
5244 | (bflags & XFS_BMAPI_NODISCARD) || | |
5245 | del->br_state == XFS_EXT_UNWRITTEN); | |
3a13f959 | 5246 | } |
cfe32f0d DW |
5247 | } |
5248 | ||
2bd0ea18 | 5249 | /* |
49f693fa | 5250 | * Adjust inode # blocks in the file. |
2bd0ea18 | 5251 | */ |
49f693fa | 5252 | if (nblks) |
aa00f286 | 5253 | ip->i_nblocks -= nblks; |
2bd0ea18 | 5254 | /* |
49f693fa | 5255 | * Adjust quota data. |
2bd0ea18 | 5256 | */ |
36b16da8 | 5257 | if (qfield && !(bflags & XFS_BMAPI_REMAP)) |
49f693fa DC |
5258 | xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks); |
5259 | ||
49f693fa DC |
5260 | done: |
5261 | *logflagsp = flags; | |
2bd0ea18 NS |
5262 | return error; |
5263 | } | |
5264 | ||
5265 | /* | |
49f693fa DC |
5266 | * Unmap (remove) blocks from a file. |
5267 | * If nexts is nonzero then the number of extents to remove is limited to | |
5268 | * that value. If not all extents in the block range can be removed then | |
5269 | * *done is set. | |
2bd0ea18 | 5270 | */ |
49f693fa | 5271 | int /* error */ |
3d36acda | 5272 | __xfs_bunmapi( |
160ed12f | 5273 | struct xfs_trans *tp, /* transaction pointer */ |
49f693fa | 5274 | struct xfs_inode *ip, /* incore inode */ |
675b5a20 | 5275 | xfs_fileoff_t start, /* first file offset deleted */ |
3d36acda | 5276 | xfs_filblks_t *rlen, /* i/o: amount remaining */ |
49f693fa | 5277 | int flags, /* misc flags */ |
d3c5f3dd | 5278 | xfs_extnum_t nexts) /* number of extents max */ |
2bd0ea18 | 5279 | { |
160ed12f BF |
5280 | struct xfs_btree_cur *cur; /* bmap btree cursor */ |
5281 | struct xfs_bmbt_irec del; /* extent being deleted */ | |
49f693fa DC |
5282 | int error; /* error return value */ |
5283 | xfs_extnum_t extno; /* extent number in list */ | |
160ed12f | 5284 | struct xfs_bmbt_irec got; /* current extent record */ |
e07055b8 | 5285 | struct xfs_ifork *ifp; /* inode fork pointer */ |
49f693fa | 5286 | int isrt; /* freeing in rt area */ |
49f693fa DC |
5287 | int logflags; /* transaction logging flags */ |
5288 | xfs_extlen_t mod; /* rt extent offset */ | |
bc73da84 | 5289 | struct xfs_mount *mp = ip->i_mount; |
49f693fa DC |
5290 | int tmp_logflags; /* partial logging flags */ |
5291 | int wasdel; /* was a delayed alloc extent */ | |
5292 | int whichfork; /* data or attribute fork */ | |
5293 | xfs_fsblock_t sum; | |
3d36acda | 5294 | xfs_filblks_t len = *rlen; /* length to unmap in file */ |
594956fa | 5295 | xfs_fileoff_t max_len; |
675b5a20 | 5296 | xfs_fileoff_t end; |
9788e059 CH |
5297 | struct xfs_iext_cursor icur; |
5298 | bool done = false; | |
2bd0ea18 | 5299 | |
675b5a20 | 5300 | trace_xfs_bunmap(ip, start, len, flags, _RET_IP_); |
a2ceac1f | 5301 | |
cb8a004a DW |
5302 | whichfork = xfs_bmapi_whichfork(flags); |
5303 | ASSERT(whichfork != XFS_COW_FORK); | |
a2ceac1f | 5304 | ifp = XFS_IFORK_PTR(ip, whichfork); |
d967a68d | 5305 | if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp))) |
12b53197 | 5306 | return -EFSCORRUPTED; |
93adb06a | 5307 | if (xfs_is_shutdown(mp)) |
12b53197 | 5308 | return -EIO; |
56b2de80 | 5309 | |
ff105f75 | 5310 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
49f693fa DC |
5311 | ASSERT(len > 0); |
5312 | ASSERT(nexts >= 0); | |
56b2de80 | 5313 | |
594956fa DW |
5314 | /* |
5315 | * Guesstimate how many blocks we can unmap without running the risk of | |
5316 | * blowing out the transaction with a mix of EFIs and reflink | |
5317 | * adjustments. | |
5318 | */ | |
db2cf1d2 | 5319 | if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) |
594956fa DW |
5320 | max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res)); |
5321 | else | |
5322 | max_len = len; | |
5323 | ||
e00c57e7 CH |
5324 | error = xfs_iread_extents(tp, ip, whichfork); |
5325 | if (error) | |
49f693fa | 5326 | return error; |
e00c57e7 | 5327 | |
d09d4e5f | 5328 | if (xfs_iext_count(ifp) == 0) { |
3d36acda | 5329 | *rlen = 0; |
49f693fa | 5330 | return 0; |
56b2de80 | 5331 | } |
79896434 | 5332 | XFS_STATS_INC(mp, xs_blk_unmap); |
49f693fa | 5333 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); |
d6fbe8fe | 5334 | end = start + len; |
a2ceac1f | 5335 | |
9788e059 | 5336 | if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) { |
d6fbe8fe CH |
5337 | *rlen = 0; |
5338 | return 0; | |
49f693fa | 5339 | } |
d6fbe8fe | 5340 | end--; |
246eb90a | 5341 | |
49f693fa | 5342 | logflags = 0; |
84094546 | 5343 | if (ifp->if_format == XFS_DINODE_FMT_BTREE) { |
d967a68d | 5344 | ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE); |
49f693fa | 5345 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
116c6a88 | 5346 | cur->bc_ino.flags = 0; |
49f693fa DC |
5347 | } else |
5348 | cur = NULL; | |
a2ceac1f | 5349 | |
49f693fa | 5350 | if (isrt) { |
a2ceac1f | 5351 | /* |
49f693fa | 5352 | * Synchronize by locking the bitmap inode. |
a2ceac1f | 5353 | */ |
a62ed6d3 | 5354 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
49f693fa | 5355 | xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
a62ed6d3 DW |
5356 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
5357 | xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL); | |
49f693fa | 5358 | } |
a2ceac1f | 5359 | |
49f693fa | 5360 | extno = 0; |
9788e059 | 5361 | while (end != (xfs_fileoff_t)-1 && end >= start && |
594956fa | 5362 | (nexts == 0 || extno < nexts) && max_len > 0) { |
a2ceac1f | 5363 | /* |
675b5a20 | 5364 | * Is the found extent after a hole in which end lives? |
49f693fa | 5365 | * Just back up to the previous extent, if so. |
a2ceac1f | 5366 | */ |
9788e059 CH |
5367 | if (got.br_startoff > end && |
5368 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5369 | done = true; | |
5370 | break; | |
a2ceac1f | 5371 | } |
49f693fa DC |
5372 | /* |
5373 | * Is the last block of this extent before the range | |
5374 | * we're supposed to delete? If so, we're done. | |
5375 | */ | |
675b5a20 | 5376 | end = XFS_FILEOFF_MIN(end, |
49f693fa | 5377 | got.br_startoff + got.br_blockcount - 1); |
675b5a20 | 5378 | if (end < start) |
49f693fa DC |
5379 | break; |
5380 | /* | |
5381 | * Then deal with the (possibly delayed) allocated space | |
5382 | * we found. | |
5383 | */ | |
49f693fa DC |
5384 | del = got; |
5385 | wasdel = isnullstartblock(del.br_startblock); | |
15a8bccc | 5386 | |
49f693fa DC |
5387 | if (got.br_startoff < start) { |
5388 | del.br_startoff = start; | |
5389 | del.br_blockcount -= start - got.br_startoff; | |
5390 | if (!wasdel) | |
5391 | del.br_startblock += start - got.br_startoff; | |
5392 | } | |
675b5a20 CH |
5393 | if (del.br_startoff + del.br_blockcount > end + 1) |
5394 | del.br_blockcount = end + 1 - del.br_startoff; | |
594956fa DW |
5395 | |
5396 | /* How much can we safely unmap? */ | |
5397 | if (max_len < del.br_blockcount) { | |
5398 | del.br_startoff += del.br_blockcount - max_len; | |
5399 | if (!wasdel) | |
5400 | del.br_startblock += del.br_blockcount - max_len; | |
5401 | del.br_blockcount = max_len; | |
5402 | } | |
5403 | ||
5a595099 DC |
5404 | if (!isrt) |
5405 | goto delete; | |
5406 | ||
49f693fa | 5407 | sum = del.br_startblock + del.br_blockcount; |
5a595099 DC |
5408 | div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod); |
5409 | if (mod) { | |
49f693fa DC |
5410 | /* |
5411 | * Realtime extent not lined up at the end. | |
5412 | * The extent could have been split into written | |
5413 | * and unwritten pieces, or we could just be | |
5414 | * unmapping part of it. But we can't really | |
5415 | * get rid of part of a realtime extent. | |
5416 | */ | |
b7fdeb4d | 5417 | if (del.br_state == XFS_EXT_UNWRITTEN) { |
49f693fa DC |
5418 | /* |
5419 | * This piece is unwritten, or we're not | |
5420 | * using unwritten extents. Skip over it. | |
5421 | */ | |
675b5a20 CH |
5422 | ASSERT(end >= mod); |
5423 | end -= mod > del.br_blockcount ? | |
49f693fa | 5424 | del.br_blockcount : mod; |
9788e059 CH |
5425 | if (end < got.br_startoff && |
5426 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5427 | done = true; | |
5428 | break; | |
49f693fa DC |
5429 | } |
5430 | continue; | |
5431 | } | |
5432 | /* | |
5433 | * It's written, turn it unwritten. | |
5434 | * This is better than zeroing it. | |
5435 | */ | |
5436 | ASSERT(del.br_state == XFS_EXT_NORM); | |
0268fdc3 | 5437 | ASSERT(tp->t_blk_res > 0); |
49f693fa DC |
5438 | /* |
5439 | * If this spans a realtime extent boundary, | |
5440 | * chop it back to the start of the one we end at. | |
5441 | */ | |
5442 | if (del.br_blockcount > mod) { | |
5443 | del.br_startoff += del.br_blockcount - mod; | |
5444 | del.br_startblock += del.br_blockcount - mod; | |
5445 | del.br_blockcount = mod; | |
5446 | } | |
5447 | del.br_state = XFS_EXT_UNWRITTEN; | |
5448 | error = xfs_bmap_add_extent_unwritten_real(tp, ip, | |
9788e059 | 5449 | whichfork, &icur, &cur, &del, |
64e8b4a7 | 5450 | &logflags); |
49f693fa DC |
5451 | if (error) |
5452 | goto error0; | |
5453 | goto nodelete; | |
a2ceac1f | 5454 | } |
5a595099 DC |
5455 | div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod); |
5456 | if (mod) { | |
f8446e51 OS |
5457 | xfs_extlen_t off = mp->m_sb.sb_rextsize - mod; |
5458 | ||
49f693fa DC |
5459 | /* |
5460 | * Realtime extent is lined up at the end but not | |
5461 | * at the front. We'll get rid of full extents if | |
5462 | * we can. | |
5463 | */ | |
f8446e51 OS |
5464 | if (del.br_blockcount > off) { |
5465 | del.br_blockcount -= off; | |
5466 | del.br_startoff += off; | |
5467 | del.br_startblock += off; | |
b7fdeb4d CH |
5468 | } else if (del.br_startoff == start && |
5469 | (del.br_state == XFS_EXT_UNWRITTEN || | |
5470 | tp->t_blk_res == 0)) { | |
49f693fa DC |
5471 | /* |
5472 | * Can't make it unwritten. There isn't | |
5473 | * a full extent here so just skip it. | |
5474 | */ | |
675b5a20 CH |
5475 | ASSERT(end >= del.br_blockcount); |
5476 | end -= del.br_blockcount; | |
9788e059 CH |
5477 | if (got.br_startoff > end && |
5478 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5479 | done = true; | |
5480 | break; | |
5481 | } | |
49f693fa DC |
5482 | continue; |
5483 | } else if (del.br_state == XFS_EXT_UNWRITTEN) { | |
246eb90a | 5484 | struct xfs_bmbt_irec prev; |
f8446e51 | 5485 | xfs_fileoff_t unwrite_start; |
246eb90a | 5486 | |
49f693fa DC |
5487 | /* |
5488 | * This one is already unwritten. | |
5489 | * It must have a written left neighbor. | |
5490 | * Unwrite the killed part of that one and | |
5491 | * try again. | |
5492 | */ | |
9788e059 CH |
5493 | if (!xfs_iext_prev_extent(ifp, &icur, &prev)) |
5494 | ASSERT(0); | |
49f693fa DC |
5495 | ASSERT(prev.br_state == XFS_EXT_NORM); |
5496 | ASSERT(!isnullstartblock(prev.br_startblock)); | |
5497 | ASSERT(del.br_startblock == | |
5498 | prev.br_startblock + prev.br_blockcount); | |
f8446e51 OS |
5499 | unwrite_start = max3(start, |
5500 | del.br_startoff - mod, | |
5501 | prev.br_startoff); | |
5502 | mod = unwrite_start - prev.br_startoff; | |
5503 | prev.br_startoff = unwrite_start; | |
5504 | prev.br_startblock += mod; | |
5505 | prev.br_blockcount -= mod; | |
49f693fa | 5506 | prev.br_state = XFS_EXT_UNWRITTEN; |
49f693fa | 5507 | error = xfs_bmap_add_extent_unwritten_real(tp, |
9788e059 | 5508 | ip, whichfork, &icur, &cur, |
64e8b4a7 | 5509 | &prev, &logflags); |
49f693fa DC |
5510 | if (error) |
5511 | goto error0; | |
5512 | goto nodelete; | |
5513 | } else { | |
5514 | ASSERT(del.br_state == XFS_EXT_NORM); | |
5515 | del.br_state = XFS_EXT_UNWRITTEN; | |
5516 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
9788e059 | 5517 | ip, whichfork, &icur, &cur, |
64e8b4a7 | 5518 | &del, &logflags); |
49f693fa DC |
5519 | if (error) |
5520 | goto error0; | |
5521 | goto nodelete; | |
5522 | } | |
5523 | } | |
a2ceac1f | 5524 | |
5a595099 | 5525 | delete: |
8359e0b9 | 5526 | if (wasdel) { |
9788e059 | 5527 | error = xfs_bmap_del_extent_delay(ip, whichfork, &icur, |
ad68fd19 CH |
5528 | &got, &del); |
5529 | } else { | |
52f6ed9d BF |
5530 | error = xfs_bmap_del_extent_real(ip, tp, &icur, cur, |
5531 | &del, &tmp_logflags, whichfork, | |
ad68fd19 CH |
5532 | flags); |
5533 | logflags |= tmp_logflags; | |
e9fa15aa | 5534 | } |
8359e0b9 | 5535 | |
49f693fa DC |
5536 | if (error) |
5537 | goto error0; | |
8359e0b9 | 5538 | |
594956fa | 5539 | max_len -= del.br_blockcount; |
675b5a20 | 5540 | end = del.br_startoff - 1; |
49f693fa | 5541 | nodelete: |
a2ceac1f | 5542 | /* |
49f693fa | 5543 | * If not done go on to the next (previous) record. |
a2ceac1f | 5544 | */ |
675b5a20 | 5545 | if (end != (xfs_fileoff_t)-1 && end >= start) { |
9788e059 CH |
5546 | if (!xfs_iext_get_extent(ifp, &icur, &got) || |
5547 | (got.br_startoff > end && | |
5548 | !xfs_iext_prev_extent(ifp, &icur, &got))) { | |
5549 | done = true; | |
5550 | break; | |
49f693fa DC |
5551 | } |
5552 | extno++; | |
a2ceac1f | 5553 | } |
a2ceac1f | 5554 | } |
9788e059 | 5555 | if (done || end == (xfs_fileoff_t)-1 || end < start) |
3d36acda DW |
5556 | *rlen = 0; |
5557 | else | |
675b5a20 | 5558 | *rlen = end - start + 1; |
56b2de80 | 5559 | |
49f693fa DC |
5560 | /* |
5561 | * Convert to a btree if necessary. | |
5562 | */ | |
5563 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
5564 | ASSERT(cur == NULL); | |
f7253505 BF |
5565 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
5566 | &tmp_logflags, whichfork); | |
49f693fa | 5567 | logflags |= tmp_logflags; |
939ebc1a CH |
5568 | } else { |
5569 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, | |
49f693fa | 5570 | whichfork); |
56b2de80 | 5571 | } |
939ebc1a | 5572 | |
49f693fa DC |
5573 | error0: |
5574 | /* | |
5575 | * Log everything. Do this after conversion, there's no point in | |
5576 | * logging the extent records if we've converted to btree format. | |
5577 | */ | |
5578 | if ((logflags & xfs_ilog_fext(whichfork)) && | |
d967a68d | 5579 | ifp->if_format != XFS_DINODE_FMT_EXTENTS) |
49f693fa DC |
5580 | logflags &= ~xfs_ilog_fext(whichfork); |
5581 | else if ((logflags & xfs_ilog_fbroot(whichfork)) && | |
d967a68d | 5582 | ifp->if_format != XFS_DINODE_FMT_BTREE) |
49f693fa DC |
5583 | logflags &= ~xfs_ilog_fbroot(whichfork); |
5584 | /* | |
5585 | * Log inode even in the error case, if the transaction | |
5586 | * is dirty we'll need to shut down the filesystem. | |
5587 | */ | |
5588 | if (logflags) | |
5589 | xfs_trans_log_inode(tp, ip, logflags); | |
5590 | if (cur) { | |
f44a348b | 5591 | if (!error) |
116c6a88 | 5592 | cur->bc_ino.allocated = 0; |
660265b7 | 5593 | xfs_btree_del_cursor(cur, error); |
56b2de80 | 5594 | } |
49f693fa | 5595 | return error; |
a2ceac1f | 5596 | } |
ff105f75 | 5597 | |
3d36acda DW |
5598 | /* Unmap a range of a file. */ |
5599 | int | |
5600 | xfs_bunmapi( | |
5601 | xfs_trans_t *tp, | |
5602 | struct xfs_inode *ip, | |
5603 | xfs_fileoff_t bno, | |
5604 | xfs_filblks_t len, | |
5605 | int flags, | |
5606 | xfs_extnum_t nexts, | |
3d36acda DW |
5607 | int *done) |
5608 | { | |
5609 | int error; | |
5610 | ||
d3c5f3dd | 5611 | error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts); |
3d36acda DW |
5612 | *done = (len == 0); |
5613 | return error; | |
5614 | } | |
5615 | ||
5a35bf2c DC |
5616 | /* |
5617 | * Determine whether an extent shift can be accomplished by a merge with the | |
5618 | * extent that precedes the target hole of the shift. | |
5619 | */ | |
5620 | STATIC bool | |
5621 | xfs_bmse_can_merge( | |
5622 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5623 | struct xfs_bmbt_irec *got, /* current extent to shift */ | |
5624 | xfs_fileoff_t shift) /* shift fsb */ | |
5625 | { | |
5626 | xfs_fileoff_t startoff; | |
5627 | ||
5628 | startoff = got->br_startoff - shift; | |
5629 | ||
5630 | /* | |
5631 | * The extent, once shifted, must be adjacent in-file and on-disk with | |
5632 | * the preceding extent. | |
5633 | */ | |
5634 | if ((left->br_startoff + left->br_blockcount != startoff) || | |
5635 | (left->br_startblock + left->br_blockcount != got->br_startblock) || | |
5636 | (left->br_state != got->br_state) || | |
5637 | (left->br_blockcount + got->br_blockcount > MAXEXTLEN)) | |
5638 | return false; | |
5639 | ||
5640 | return true; | |
5641 | } | |
5642 | ||
5643 | /* | |
5644 | * A bmap extent shift adjusts the file offset of an extent to fill a preceding | |
5645 | * hole in the file. If an extent shift would result in the extent being fully | |
5646 | * adjacent to the extent that currently precedes the hole, we can merge with | |
5647 | * the preceding extent rather than do the shift. | |
5648 | * | |
5649 | * This function assumes the caller has verified a shift-by-merge is possible | |
5650 | * with the provided extents via xfs_bmse_can_merge(). | |
5651 | */ | |
5652 | STATIC int | |
5653 | xfs_bmse_merge( | |
21375e5d | 5654 | struct xfs_trans *tp, |
5a35bf2c DC |
5655 | struct xfs_inode *ip, |
5656 | int whichfork, | |
5657 | xfs_fileoff_t shift, /* shift fsb */ | |
9788e059 | 5658 | struct xfs_iext_cursor *icur, |
2f082827 CH |
5659 | struct xfs_bmbt_irec *got, /* extent to shift */ |
5660 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5a35bf2c | 5661 | struct xfs_btree_cur *cur, |
21375e5d | 5662 | int *logflags) /* output */ |
5a35bf2c | 5663 | { |
87c472b7 | 5664 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
2f082827 | 5665 | struct xfs_bmbt_irec new; |
5a35bf2c DC |
5666 | xfs_filblks_t blockcount; |
5667 | int error, i; | |
19ebedcf | 5668 | struct xfs_mount *mp = ip->i_mount; |
5a35bf2c | 5669 | |
2f082827 | 5670 | blockcount = left->br_blockcount + got->br_blockcount; |
5a35bf2c DC |
5671 | |
5672 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5673 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
2f082827 | 5674 | ASSERT(xfs_bmse_can_merge(left, got, shift)); |
5a35bf2c | 5675 | |
2f082827 CH |
5676 | new = *left; |
5677 | new.br_blockcount = blockcount; | |
5a35bf2c DC |
5678 | |
5679 | /* | |
5680 | * Update the on-disk extent count, the btree if necessary and log the | |
5681 | * inode. | |
5682 | */ | |
87c472b7 | 5683 | ifp->if_nextents--; |
5a35bf2c DC |
5684 | *logflags |= XFS_ILOG_CORE; |
5685 | if (!cur) { | |
5686 | *logflags |= XFS_ILOG_DEXT; | |
2f082827 | 5687 | goto done; |
5a35bf2c DC |
5688 | } |
5689 | ||
5690 | /* lookup and remove the extent to merge */ | |
70a93110 | 5691 | error = xfs_bmbt_lookup_eq(cur, got, &i); |
5a35bf2c DC |
5692 | if (error) |
5693 | return error; | |
fbb4fa7f DW |
5694 | if (XFS_IS_CORRUPT(mp, i != 1)) |
5695 | return -EFSCORRUPTED; | |
5a35bf2c DC |
5696 | |
5697 | error = xfs_btree_delete(cur, &i); | |
5698 | if (error) | |
5699 | return error; | |
fbb4fa7f DW |
5700 | if (XFS_IS_CORRUPT(mp, i != 1)) |
5701 | return -EFSCORRUPTED; | |
5a35bf2c DC |
5702 | |
5703 | /* lookup and update size of the previous extent */ | |
70a93110 | 5704 | error = xfs_bmbt_lookup_eq(cur, left, &i); |
5a35bf2c DC |
5705 | if (error) |
5706 | return error; | |
fbb4fa7f DW |
5707 | if (XFS_IS_CORRUPT(mp, i != 1)) |
5708 | return -EFSCORRUPTED; | |
5a35bf2c | 5709 | |
d0e5f1ff | 5710 | error = xfs_bmbt_update(cur, &new); |
2f082827 CH |
5711 | if (error) |
5712 | return error; | |
5713 | ||
e102336b BF |
5714 | /* change to extent format if required after extent removal */ |
5715 | error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork); | |
5716 | if (error) | |
5717 | return error; | |
5718 | ||
2f082827 | 5719 | done: |
cf455a62 | 5720 | xfs_iext_remove(ip, icur, 0); |
87c472b7 | 5721 | xfs_iext_prev(ifp, icur); |
9788e059 CH |
5722 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur, |
5723 | &new); | |
5a35bf2c | 5724 | |
479284b7 | 5725 | /* update reverse mapping. rmap functions merge the rmaps for us */ |
46d29bb9 | 5726 | xfs_rmap_unmap_extent(tp, ip, whichfork, got); |
479284b7 DW |
5727 | memcpy(&new, got, sizeof(new)); |
5728 | new.br_startoff = left->br_startoff + left->br_blockcount; | |
46d29bb9 DW |
5729 | xfs_rmap_map_extent(tp, ip, whichfork, &new); |
5730 | return 0; | |
5a35bf2c DC |
5731 | } |
5732 | ||
364937e2 CH |
5733 | static int |
5734 | xfs_bmap_shift_update_extent( | |
21375e5d | 5735 | struct xfs_trans *tp, |
364937e2 CH |
5736 | struct xfs_inode *ip, |
5737 | int whichfork, | |
9788e059 | 5738 | struct xfs_iext_cursor *icur, |
364937e2 CH |
5739 | struct xfs_bmbt_irec *got, |
5740 | struct xfs_btree_cur *cur, | |
5741 | int *logflags, | |
364937e2 | 5742 | xfs_fileoff_t startoff) |
5a35bf2c | 5743 | { |
364937e2 | 5744 | struct xfs_mount *mp = ip->i_mount; |
f36ccac2 | 5745 | struct xfs_bmbt_irec prev = *got; |
364937e2 | 5746 | int error, i; |
2f082827 | 5747 | |
5a35bf2c | 5748 | *logflags |= XFS_ILOG_CORE; |
2f082827 | 5749 | |
f36ccac2 | 5750 | got->br_startoff = startoff; |
2f082827 CH |
5751 | |
5752 | if (cur) { | |
f36ccac2 | 5753 | error = xfs_bmbt_lookup_eq(cur, &prev, &i); |
2f082827 CH |
5754 | if (error) |
5755 | return error; | |
fbb4fa7f DW |
5756 | if (XFS_IS_CORRUPT(mp, i != 1)) |
5757 | return -EFSCORRUPTED; | |
2f082827 | 5758 | |
f36ccac2 | 5759 | error = xfs_bmbt_update(cur, got); |
2f082827 CH |
5760 | if (error) |
5761 | return error; | |
5762 | } else { | |
5a35bf2c | 5763 | *logflags |= XFS_ILOG_DEXT; |
5a35bf2c DC |
5764 | } |
5765 | ||
9788e059 CH |
5766 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur, |
5767 | got); | |
d7f80320 | 5768 | |
d7f80320 | 5769 | /* update reverse mapping */ |
46d29bb9 DW |
5770 | xfs_rmap_unmap_extent(tp, ip, whichfork, &prev); |
5771 | xfs_rmap_map_extent(tp, ip, whichfork, got); | |
5772 | return 0; | |
5a35bf2c DC |
5773 | } |
5774 | ||
ff105f75 | 5775 | int |
a9f5d25e | 5776 | xfs_bmap_collapse_extents( |
ff105f75 DC |
5777 | struct xfs_trans *tp, |
5778 | struct xfs_inode *ip, | |
19ebedcf | 5779 | xfs_fileoff_t *next_fsb, |
ff105f75 | 5780 | xfs_fileoff_t offset_shift_fsb, |
a397f27d | 5781 | bool *done) |
ff105f75 | 5782 | { |
a9f5d25e CH |
5783 | int whichfork = XFS_DATA_FORK; |
5784 | struct xfs_mount *mp = ip->i_mount; | |
5785 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5786 | struct xfs_btree_cur *cur = NULL; | |
364937e2 | 5787 | struct xfs_bmbt_irec got, prev; |
9788e059 | 5788 | struct xfs_iext_cursor icur; |
364937e2 | 5789 | xfs_fileoff_t new_startoff; |
a9f5d25e CH |
5790 | int error = 0; |
5791 | int logflags = 0; | |
ff105f75 | 5792 | |
d967a68d | 5793 | if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) || |
bc73da84 | 5794 | XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) { |
12b53197 | 5795 | return -EFSCORRUPTED; |
ff105f75 DC |
5796 | } |
5797 | ||
93adb06a | 5798 | if (xfs_is_shutdown(mp)) |
12b53197 | 5799 | return -EIO; |
ff105f75 | 5800 | |
a9f5d25e | 5801 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); |
ff105f75 | 5802 | |
e00c57e7 CH |
5803 | error = xfs_iread_extents(tp, ip, whichfork); |
5804 | if (error) | |
5805 | return error; | |
ff105f75 | 5806 | |
84094546 | 5807 | if (ifp->if_format == XFS_DINODE_FMT_BTREE) { |
ff105f75 | 5808 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
116c6a88 | 5809 | cur->bc_ino.flags = 0; |
5a35bf2c DC |
5810 | } |
5811 | ||
9788e059 | 5812 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) { |
a9f5d25e CH |
5813 | *done = true; |
5814 | goto del_cursor; | |
5815 | } | |
fbb4fa7f DW |
5816 | if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) { |
5817 | error = -EFSCORRUPTED; | |
5818 | goto del_cursor; | |
5819 | } | |
a9f5d25e | 5820 | |
364937e2 | 5821 | new_startoff = got.br_startoff - offset_shift_fsb; |
9788e059 | 5822 | if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) { |
364937e2 CH |
5823 | if (new_startoff < prev.br_startoff + prev.br_blockcount) { |
5824 | error = -EINVAL; | |
5825 | goto del_cursor; | |
5826 | } | |
5827 | ||
364937e2 | 5828 | if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) { |
21375e5d BF |
5829 | error = xfs_bmse_merge(tp, ip, whichfork, |
5830 | offset_shift_fsb, &icur, &got, &prev, | |
5831 | cur, &logflags); | |
364937e2 CH |
5832 | if (error) |
5833 | goto del_cursor; | |
5834 | goto done; | |
5835 | } | |
5836 | } else { | |
5837 | if (got.br_startoff < offset_shift_fsb) { | |
5838 | error = -EINVAL; | |
5839 | goto del_cursor; | |
5840 | } | |
5841 | } | |
5842 | ||
21375e5d BF |
5843 | error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got, |
5844 | cur, &logflags, new_startoff); | |
a9f5d25e CH |
5845 | if (error) |
5846 | goto del_cursor; | |
bac20498 | 5847 | |
c162e319 | 5848 | done: |
9788e059 CH |
5849 | if (!xfs_iext_next_extent(ifp, &icur, &got)) { |
5850 | *done = true; | |
5851 | goto del_cursor; | |
a9f5d25e | 5852 | } |
a9f5d25e | 5853 | |
364937e2 | 5854 | *next_fsb = got.br_startoff; |
a9f5d25e CH |
5855 | del_cursor: |
5856 | if (cur) | |
660265b7 | 5857 | xfs_btree_del_cursor(cur, error); |
a9f5d25e CH |
5858 | if (logflags) |
5859 | xfs_trans_log_inode(tp, ip, logflags); | |
a9f5d25e CH |
5860 | return error; |
5861 | } | |
5862 | ||
5b501597 DW |
5863 | /* Make sure we won't be right-shifting an extent past the maximum bound. */ |
5864 | int | |
5865 | xfs_bmap_can_insert_extents( | |
5866 | struct xfs_inode *ip, | |
5867 | xfs_fileoff_t off, | |
5868 | xfs_fileoff_t shift) | |
5869 | { | |
5870 | struct xfs_bmbt_irec got; | |
5871 | int is_empty; | |
5872 | int error = 0; | |
5873 | ||
5874 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5875 | ||
93adb06a | 5876 | if (xfs_is_shutdown(ip->i_mount)) |
5b501597 DW |
5877 | return -EIO; |
5878 | ||
5879 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5880 | error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty); | |
5881 | if (!error && !is_empty && got.br_startoff >= off && | |
5882 | ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff) | |
5883 | error = -EINVAL; | |
5884 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
5885 | ||
5886 | return error; | |
5887 | } | |
5888 | ||
a9f5d25e CH |
5889 | int |
5890 | xfs_bmap_insert_extents( | |
5891 | struct xfs_trans *tp, | |
5892 | struct xfs_inode *ip, | |
5893 | xfs_fileoff_t *next_fsb, | |
5894 | xfs_fileoff_t offset_shift_fsb, | |
5895 | bool *done, | |
a397f27d | 5896 | xfs_fileoff_t stop_fsb) |
a9f5d25e CH |
5897 | { |
5898 | int whichfork = XFS_DATA_FORK; | |
5899 | struct xfs_mount *mp = ip->i_mount; | |
5900 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5901 | struct xfs_btree_cur *cur = NULL; | |
bd80a804 | 5902 | struct xfs_bmbt_irec got, next; |
9788e059 | 5903 | struct xfs_iext_cursor icur; |
364937e2 | 5904 | xfs_fileoff_t new_startoff; |
a9f5d25e CH |
5905 | int error = 0; |
5906 | int logflags = 0; | |
5907 | ||
d967a68d | 5908 | if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) || |
bc73da84 | 5909 | XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) { |
a9f5d25e CH |
5910 | return -EFSCORRUPTED; |
5911 | } | |
5912 | ||
93adb06a | 5913 | if (xfs_is_shutdown(mp)) |
a9f5d25e CH |
5914 | return -EIO; |
5915 | ||
5916 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); | |
5917 | ||
e00c57e7 CH |
5918 | error = xfs_iread_extents(tp, ip, whichfork); |
5919 | if (error) | |
5920 | return error; | |
a9f5d25e | 5921 | |
84094546 | 5922 | if (ifp->if_format == XFS_DINODE_FMT_BTREE) { |
a9f5d25e | 5923 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
116c6a88 | 5924 | cur->bc_ino.flags = 0; |
a9f5d25e CH |
5925 | } |
5926 | ||
19ebedcf | 5927 | if (*next_fsb == NULLFSBLOCK) { |
9788e059 CH |
5928 | xfs_iext_last(ifp, &icur); |
5929 | if (!xfs_iext_get_extent(ifp, &icur, &got) || | |
bd80a804 | 5930 | stop_fsb > got.br_startoff) { |
a9f5d25e | 5931 | *done = true; |
19ebedcf DC |
5932 | goto del_cursor; |
5933 | } | |
b6ad780b | 5934 | } else { |
9788e059 | 5935 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) { |
a9f5d25e | 5936 | *done = true; |
b6ad780b CH |
5937 | goto del_cursor; |
5938 | } | |
19ebedcf | 5939 | } |
fbb4fa7f DW |
5940 | if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) { |
5941 | error = -EFSCORRUPTED; | |
5942 | goto del_cursor; | |
5943 | } | |
19ebedcf | 5944 | |
e4d719e1 | 5945 | if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) { |
88afc9cc | 5946 | error = -EFSCORRUPTED; |
a9f5d25e | 5947 | goto del_cursor; |
19ebedcf DC |
5948 | } |
5949 | ||
364937e2 | 5950 | new_startoff = got.br_startoff + offset_shift_fsb; |
9788e059 | 5951 | if (xfs_iext_peek_next_extent(ifp, &icur, &next)) { |
364937e2 CH |
5952 | if (new_startoff + got.br_blockcount > next.br_startoff) { |
5953 | error = -EINVAL; | |
5954 | goto del_cursor; | |
5955 | } | |
5956 | ||
5957 | /* | |
5958 | * Unlike a left shift (which involves a hole punch), a right | |
5959 | * shift does not modify extent neighbors in any way. We should | |
5960 | * never find mergeable extents in this scenario. Check anyways | |
5961 | * and warn if we encounter two extents that could be one. | |
5962 | */ | |
5963 | if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb)) | |
5964 | WARN_ON_ONCE(1); | |
5965 | } | |
5966 | ||
21375e5d BF |
5967 | error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got, |
5968 | cur, &logflags, new_startoff); | |
6835c363 CH |
5969 | if (error) |
5970 | goto del_cursor; | |
bd80a804 | 5971 | |
9788e059 | 5972 | if (!xfs_iext_prev_extent(ifp, &icur, &got) || |
bd80a804 | 5973 | stop_fsb >= got.br_startoff + got.br_blockcount) { |
a9f5d25e | 5974 | *done = true; |
6835c363 | 5975 | goto del_cursor; |
ff105f75 DC |
5976 | } |
5977 | ||
6835c363 | 5978 | *next_fsb = got.br_startoff; |
ff105f75 DC |
5979 | del_cursor: |
5980 | if (cur) | |
660265b7 | 5981 | xfs_btree_del_cursor(cur, error); |
5a35bf2c DC |
5982 | if (logflags) |
5983 | xfs_trans_log_inode(tp, ip, logflags); | |
ff105f75 DC |
5984 | return error; |
5985 | } | |
19ebedcf DC |
5986 | |
5987 | /* | |
9788e059 CH |
5988 | * Splits an extent into two extents at split_fsb block such that it is the |
5989 | * first block of the current_ext. @ext is a target extent to be split. | |
5990 | * @split_fsb is a block where the extents is split. If split_fsb lies in a | |
5991 | * hole or the first block of extents, just return 0. | |
19ebedcf | 5992 | */ |
905a3c99 BF |
5993 | int |
5994 | xfs_bmap_split_extent( | |
19ebedcf DC |
5995 | struct xfs_trans *tp, |
5996 | struct xfs_inode *ip, | |
312350cd | 5997 | xfs_fileoff_t split_fsb) |
19ebedcf DC |
5998 | { |
5999 | int whichfork = XFS_DATA_FORK; | |
d967a68d | 6000 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
19ebedcf | 6001 | struct xfs_btree_cur *cur = NULL; |
19ebedcf DC |
6002 | struct xfs_bmbt_irec got; |
6003 | struct xfs_bmbt_irec new; /* split extent */ | |
6004 | struct xfs_mount *mp = ip->i_mount; | |
19ebedcf | 6005 | xfs_fsblock_t gotblkcnt; /* new block count for got */ |
9788e059 | 6006 | struct xfs_iext_cursor icur; |
19ebedcf DC |
6007 | int error = 0; |
6008 | int logflags = 0; | |
6009 | int i = 0; | |
6010 | ||
d967a68d | 6011 | if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) || |
bc73da84 | 6012 | XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) { |
19ebedcf DC |
6013 | return -EFSCORRUPTED; |
6014 | } | |
6015 | ||
93adb06a | 6016 | if (xfs_is_shutdown(mp)) |
19ebedcf DC |
6017 | return -EIO; |
6018 | ||
e00c57e7 CH |
6019 | /* Read in all the extents */ |
6020 | error = xfs_iread_extents(tp, ip, whichfork); | |
6021 | if (error) | |
6022 | return error; | |
19ebedcf DC |
6023 | |
6024 | /* | |
455044a6 | 6025 | * If there are not extents, or split_fsb lies in a hole we are done. |
19ebedcf | 6026 | */ |
9788e059 | 6027 | if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) || |
455044a6 | 6028 | got.br_startoff >= split_fsb) |
19ebedcf DC |
6029 | return 0; |
6030 | ||
6031 | gotblkcnt = split_fsb - got.br_startoff; | |
6032 | new.br_startoff = split_fsb; | |
6033 | new.br_startblock = got.br_startblock + gotblkcnt; | |
6034 | new.br_blockcount = got.br_blockcount - gotblkcnt; | |
6035 | new.br_state = got.br_state; | |
6036 | ||
84094546 | 6037 | if (ifp->if_format == XFS_DINODE_FMT_BTREE) { |
19ebedcf | 6038 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
116c6a88 | 6039 | cur->bc_ino.flags = 0; |
70a93110 | 6040 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
19ebedcf DC |
6041 | if (error) |
6042 | goto del_cursor; | |
fbb4fa7f DW |
6043 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
6044 | error = -EFSCORRUPTED; | |
6045 | goto del_cursor; | |
6046 | } | |
19ebedcf DC |
6047 | } |
6048 | ||
19ebedcf | 6049 | got.br_blockcount = gotblkcnt; |
9788e059 CH |
6050 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur, |
6051 | &got); | |
19ebedcf DC |
6052 | |
6053 | logflags = XFS_ILOG_CORE; | |
6054 | if (cur) { | |
d0e5f1ff | 6055 | error = xfs_bmbt_update(cur, &got); |
19ebedcf DC |
6056 | if (error) |
6057 | goto del_cursor; | |
6058 | } else | |
6059 | logflags |= XFS_ILOG_DEXT; | |
6060 | ||
6061 | /* Add new extent */ | |
9788e059 | 6062 | xfs_iext_next(ifp, &icur); |
26a75f67 | 6063 | xfs_iext_insert(ip, &icur, &new, 0); |
87c472b7 | 6064 | ifp->if_nextents++; |
19ebedcf DC |
6065 | |
6066 | if (cur) { | |
70a93110 | 6067 | error = xfs_bmbt_lookup_eq(cur, &new, &i); |
19ebedcf DC |
6068 | if (error) |
6069 | goto del_cursor; | |
fbb4fa7f DW |
6070 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
6071 | error = -EFSCORRUPTED; | |
6072 | goto del_cursor; | |
6073 | } | |
19ebedcf DC |
6074 | error = xfs_btree_insert(cur, &i); |
6075 | if (error) | |
6076 | goto del_cursor; | |
fbb4fa7f DW |
6077 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
6078 | error = -EFSCORRUPTED; | |
6079 | goto del_cursor; | |
6080 | } | |
19ebedcf DC |
6081 | } |
6082 | ||
6083 | /* | |
6084 | * Convert to a btree if necessary. | |
6085 | */ | |
6086 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
6087 | int tmp_logflags; /* partial log flag return val */ | |
6088 | ||
6089 | ASSERT(cur == NULL); | |
f7253505 BF |
6090 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
6091 | &tmp_logflags, whichfork); | |
19ebedcf DC |
6092 | logflags |= tmp_logflags; |
6093 | } | |
6094 | ||
6095 | del_cursor: | |
6096 | if (cur) { | |
116c6a88 | 6097 | cur->bc_ino.allocated = 0; |
660265b7 | 6098 | xfs_btree_del_cursor(cur, error); |
19ebedcf DC |
6099 | } |
6100 | ||
6101 | if (logflags) | |
6102 | xfs_trans_log_inode(tp, ip, logflags); | |
6103 | return error; | |
6104 | } | |
6105 | ||
aeb88300 DW |
6106 | /* Deferred mapping is only for real extents in the data fork. */ |
6107 | static bool | |
6108 | xfs_bmap_is_update_needed( | |
6109 | struct xfs_bmbt_irec *bmap) | |
6110 | { | |
6111 | return bmap->br_startblock != HOLESTARTBLOCK && | |
6112 | bmap->br_startblock != DELAYSTARTBLOCK; | |
6113 | } | |
6114 | ||
6115 | /* Record a bmap intent. */ | |
6116 | static int | |
6117 | __xfs_bmap_add( | |
21375e5d | 6118 | struct xfs_trans *tp, |
aeb88300 DW |
6119 | enum xfs_bmap_intent_type type, |
6120 | struct xfs_inode *ip, | |
6121 | int whichfork, | |
6122 | struct xfs_bmbt_irec *bmap) | |
6123 | { | |
aeb88300 DW |
6124 | struct xfs_bmap_intent *bi; |
6125 | ||
21375e5d BF |
6126 | trace_xfs_bmap_defer(tp->t_mountp, |
6127 | XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock), | |
aeb88300 | 6128 | type, |
21375e5d | 6129 | XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock), |
aeb88300 DW |
6130 | ip->i_ino, whichfork, |
6131 | bmap->br_startoff, | |
6132 | bmap->br_blockcount, | |
6133 | bmap->br_state); | |
6134 | ||
1577541c | 6135 | bi = kmem_cache_alloc(xfs_bmap_intent_cache, GFP_NOFS | __GFP_NOFAIL); |
aeb88300 DW |
6136 | INIT_LIST_HEAD(&bi->bi_list); |
6137 | bi->bi_type = type; | |
6138 | bi->bi_owner = ip; | |
6139 | bi->bi_whichfork = whichfork; | |
6140 | bi->bi_bmap = *bmap; | |
6141 | ||
21375e5d | 6142 | xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list); |
aeb88300 DW |
6143 | return 0; |
6144 | } | |
6145 | ||
6146 | /* Map an extent into a file. */ | |
60a802cc | 6147 | void |
aeb88300 | 6148 | xfs_bmap_map_extent( |
21375e5d | 6149 | struct xfs_trans *tp, |
aeb88300 DW |
6150 | struct xfs_inode *ip, |
6151 | struct xfs_bmbt_irec *PREV) | |
6152 | { | |
6153 | if (!xfs_bmap_is_update_needed(PREV)) | |
60a802cc | 6154 | return; |
aeb88300 | 6155 | |
60a802cc | 6156 | __xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV); |
aeb88300 DW |
6157 | } |
6158 | ||
6159 | /* Unmap an extent out of a file. */ | |
60a802cc | 6160 | void |
aeb88300 | 6161 | xfs_bmap_unmap_extent( |
21375e5d | 6162 | struct xfs_trans *tp, |
aeb88300 DW |
6163 | struct xfs_inode *ip, |
6164 | struct xfs_bmbt_irec *PREV) | |
6165 | { | |
6166 | if (!xfs_bmap_is_update_needed(PREV)) | |
60a802cc | 6167 | return; |
aeb88300 | 6168 | |
60a802cc | 6169 | __xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV); |
aeb88300 DW |
6170 | } |
6171 | ||
6172 | /* | |
6173 | * Process one of the deferred bmap operations. We pass back the | |
6174 | * btree cursor to maintain our lock on the bmapbt between calls. | |
6175 | */ | |
6176 | int | |
6177 | xfs_bmap_finish_one( | |
6178 | struct xfs_trans *tp, | |
aeb88300 DW |
6179 | struct xfs_inode *ip, |
6180 | enum xfs_bmap_intent_type type, | |
6181 | int whichfork, | |
6182 | xfs_fileoff_t startoff, | |
6183 | xfs_fsblock_t startblock, | |
594956fa | 6184 | xfs_filblks_t *blockcount, |
aeb88300 DW |
6185 | xfs_exntst_t state) |
6186 | { | |
594956fa | 6187 | int error = 0; |
aeb88300 | 6188 | |
35f57661 | 6189 | ASSERT(tp->t_firstblock == NULLFSBLOCK); |
66d19ae1 | 6190 | |
aeb88300 DW |
6191 | trace_xfs_bmap_deferred(tp->t_mountp, |
6192 | XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type, | |
6193 | XFS_FSB_TO_AGBNO(tp->t_mountp, startblock), | |
594956fa | 6194 | ip->i_ino, whichfork, startoff, *blockcount, state); |
aeb88300 | 6195 | |
95855a23 | 6196 | if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK)) |
aeb88300 | 6197 | return -EFSCORRUPTED; |
aeb88300 DW |
6198 | |
6199 | if (XFS_TEST_ERROR(false, tp->t_mountp, | |
e2a190dd | 6200 | XFS_ERRTAG_BMAP_FINISH_ONE)) |
aeb88300 DW |
6201 | return -EIO; |
6202 | ||
6203 | switch (type) { | |
6204 | case XFS_BMAP_MAP: | |
594956fa | 6205 | error = xfs_bmapi_remap(tp, ip, startoff, *blockcount, |
b5d7ff9a | 6206 | startblock, 0); |
594956fa | 6207 | *blockcount = 0; |
aeb88300 DW |
6208 | break; |
6209 | case XFS_BMAP_UNMAP: | |
594956fa | 6210 | error = __xfs_bunmapi(tp, ip, startoff, blockcount, |
d3c5f3dd | 6211 | XFS_BMAPI_REMAP, 1); |
aeb88300 DW |
6212 | break; |
6213 | default: | |
6214 | ASSERT(0); | |
6215 | error = -EFSCORRUPTED; | |
6216 | } | |
6217 | ||
6218 | return error; | |
6219 | } | |
0cf6a3a9 DW |
6220 | |
6221 | /* Check that an inode's extent does not have invalid flags or bad ranges. */ | |
6222 | xfs_failaddr_t | |
6223 | xfs_bmap_validate_extent( | |
6224 | struct xfs_inode *ip, | |
6225 | int whichfork, | |
6226 | struct xfs_bmbt_irec *irec) | |
6227 | { | |
6228 | struct xfs_mount *mp = ip->i_mount; | |
0cf6a3a9 | 6229 | |
7626c690 | 6230 | if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount)) |
92415d01 DW |
6231 | return __this_address; |
6232 | ||
23e306ae DW |
6233 | if (XFS_IS_REALTIME_INODE(ip) && whichfork == XFS_DATA_FORK) { |
6234 | if (!xfs_verify_rtext(mp, irec->br_startblock, | |
6235 | irec->br_blockcount)) | |
0cf6a3a9 DW |
6236 | return __this_address; |
6237 | } else { | |
f28f7e4a DW |
6238 | if (!xfs_verify_fsbext(mp, irec->br_startblock, |
6239 | irec->br_blockcount)) | |
0cf6a3a9 DW |
6240 | return __this_address; |
6241 | } | |
b7fdeb4d CH |
6242 | if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK) |
6243 | return __this_address; | |
0cf6a3a9 DW |
6244 | return NULL; |
6245 | } | |
1577541c DW |
6246 | |
6247 | int __init | |
6248 | xfs_bmap_intent_init_cache(void) | |
6249 | { | |
6250 | xfs_bmap_intent_cache = kmem_cache_create("xfs_bmap_intent", | |
6251 | sizeof(struct xfs_bmap_intent), | |
6252 | 0, 0, NULL); | |
6253 | ||
6254 | return xfs_bmap_intent_cache != NULL ? 0 : -ENOMEM; | |
6255 | } | |
6256 | ||
6257 | void | |
6258 | xfs_bmap_intent_destroy_cache(void) | |
6259 | { | |
6260 | kmem_cache_destroy(xfs_bmap_intent_cache); | |
6261 | xfs_bmap_intent_cache = NULL; | |
6262 | } |