Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
3e57ecf6 | 3 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 4 | * All Rights Reserved. |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
239880ef DC |
9 | #include "xfs_format.h" |
10 | #include "xfs_log_format.h" | |
11 | #include "xfs_trans_resv.h" | |
a844f451 | 12 | #include "xfs_bit.h" |
1da177e4 | 13 | #include "xfs_sb.h" |
f5ea1100 | 14 | #include "xfs_mount.h" |
3ab78df2 | 15 | #include "xfs_defer.h" |
2b9ab5ab | 16 | #include "xfs_dir2.h" |
1da177e4 | 17 | #include "xfs_inode.h" |
a844f451 | 18 | #include "xfs_btree.h" |
239880ef | 19 | #include "xfs_trans.h" |
1da177e4 LT |
20 | #include "xfs_alloc.h" |
21 | #include "xfs_bmap.h" | |
68988114 | 22 | #include "xfs_bmap_util.h" |
a4fbe6ab | 23 | #include "xfs_bmap_btree.h" |
1da177e4 | 24 | #include "xfs_rtalloc.h" |
e9e899a2 | 25 | #include "xfs_errortag.h" |
1da177e4 | 26 | #include "xfs_error.h" |
1da177e4 LT |
27 | #include "xfs_quota.h" |
28 | #include "xfs_trans_space.h" | |
29 | #include "xfs_buf_item.h" | |
0b1b213f | 30 | #include "xfs_trace.h" |
a4fbe6ab | 31 | #include "xfs_attr_leaf.h" |
a4fbe6ab | 32 | #include "xfs_filestream.h" |
340785cc | 33 | #include "xfs_rmap.h" |
3fd129b6 | 34 | #include "xfs_ag_resv.h" |
62aab20f | 35 | #include "xfs_refcount.h" |
974ae922 | 36 | #include "xfs_icache.h" |
1da177e4 LT |
37 | |
38 | ||
1da177e4 LT |
39 | kmem_zone_t *xfs_bmap_free_item_zone; |
40 | ||
41 | /* | |
9e5987a7 | 42 | * Miscellaneous helper functions |
1da177e4 LT |
43 | */ |
44 | ||
1da177e4 | 45 | /* |
9e5987a7 DC |
46 | * Compute and fill in the value of the maximum depth of a bmap btree |
47 | * in this filesystem. Done once, during mount. | |
1da177e4 | 48 | */ |
9e5987a7 DC |
49 | void |
50 | xfs_bmap_compute_maxlevels( | |
51 | xfs_mount_t *mp, /* file system mount structure */ | |
52 | int whichfork) /* data or attr fork */ | |
53 | { | |
54 | int level; /* btree level */ | |
55 | uint maxblocks; /* max blocks at this level */ | |
56 | uint maxleafents; /* max leaf entries possible */ | |
57 | int maxrootrecs; /* max records in root block */ | |
58 | int minleafrecs; /* min records in leaf block */ | |
59 | int minnoderecs; /* min records in node block */ | |
60 | int sz; /* root block size */ | |
1da177e4 | 61 | |
9e5987a7 DC |
62 | /* |
63 | * The maximum number of extents in a file, hence the maximum | |
64 | * number of leaf entries, is controlled by the type of di_nextents | |
65 | * (a signed 32-bit number, xfs_extnum_t), or by di_anextents | |
66 | * (a signed 16-bit number, xfs_aextnum_t). | |
67 | * | |
68 | * Note that we can no longer assume that if we are in ATTR1 that | |
69 | * the fork offset of all the inodes will be | |
70 | * (xfs_default_attroffset(ip) >> 3) because we could have mounted | |
71 | * with ATTR2 and then mounted back with ATTR1, keeping the | |
72 | * di_forkoff's fixed but probably at various positions. Therefore, | |
73 | * for both ATTR1 and ATTR2 we have to assume the worst case scenario | |
74 | * of a minimum size available. | |
75 | */ | |
76 | if (whichfork == XFS_DATA_FORK) { | |
77 | maxleafents = MAXEXTNUM; | |
78 | sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS); | |
79 | } else { | |
80 | maxleafents = MAXAEXTNUM; | |
81 | sz = XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
82 | } | |
152d93b7 | 83 | maxrootrecs = xfs_bmdr_maxrecs(sz, 0); |
9e5987a7 DC |
84 | minleafrecs = mp->m_bmap_dmnr[0]; |
85 | minnoderecs = mp->m_bmap_dmnr[1]; | |
86 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
87 | for (level = 1; maxblocks > 1; level++) { | |
88 | if (maxblocks <= maxrootrecs) | |
89 | maxblocks = 1; | |
90 | else | |
91 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
92 | } | |
93 | mp->m_bm_maxlevels[whichfork] = level; | |
94 | } | |
91e11088 | 95 | |
fe033cc8 CH |
96 | STATIC int /* error */ |
97 | xfs_bmbt_lookup_eq( | |
98 | struct xfs_btree_cur *cur, | |
e16cf9b0 | 99 | struct xfs_bmbt_irec *irec, |
fe033cc8 CH |
100 | int *stat) /* success/failure */ |
101 | { | |
e16cf9b0 | 102 | cur->bc_rec.b = *irec; |
fe033cc8 CH |
103 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); |
104 | } | |
105 | ||
106 | STATIC int /* error */ | |
b5cfbc22 | 107 | xfs_bmbt_lookup_first( |
fe033cc8 | 108 | struct xfs_btree_cur *cur, |
fe033cc8 CH |
109 | int *stat) /* success/failure */ |
110 | { | |
b5cfbc22 CH |
111 | cur->bc_rec.b.br_startoff = 0; |
112 | cur->bc_rec.b.br_startblock = 0; | |
113 | cur->bc_rec.b.br_blockcount = 0; | |
fe033cc8 CH |
114 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); |
115 | } | |
116 | ||
278d0ca1 | 117 | /* |
8096b1eb CH |
118 | * Check if the inode needs to be converted to btree format. |
119 | */ | |
120 | static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork) | |
121 | { | |
60b4984f DW |
122 | return whichfork != XFS_COW_FORK && |
123 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
8096b1eb CH |
124 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
125 | XFS_IFORK_MAXEXT(ip, whichfork); | |
126 | } | |
127 | ||
128 | /* | |
129 | * Check if the inode should be converted to extent format. | |
130 | */ | |
131 | static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork) | |
132 | { | |
60b4984f DW |
133 | return whichfork != XFS_COW_FORK && |
134 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE && | |
8096b1eb CH |
135 | XFS_IFORK_NEXTENTS(ip, whichfork) <= |
136 | XFS_IFORK_MAXEXT(ip, whichfork); | |
137 | } | |
138 | ||
139 | /* | |
a67d00a5 | 140 | * Update the record referred to by cur to the value given by irec |
278d0ca1 CH |
141 | * This either works (return 0) or gets an EFSCORRUPTED error. |
142 | */ | |
143 | STATIC int | |
144 | xfs_bmbt_update( | |
145 | struct xfs_btree_cur *cur, | |
a67d00a5 | 146 | struct xfs_bmbt_irec *irec) |
278d0ca1 CH |
147 | { |
148 | union xfs_btree_rec rec; | |
149 | ||
a67d00a5 | 150 | xfs_bmbt_disk_set_all(&rec.bmbt, irec); |
278d0ca1 CH |
151 | return xfs_btree_update(cur, &rec); |
152 | } | |
fe033cc8 | 153 | |
1da177e4 | 154 | /* |
9e5987a7 DC |
155 | * Compute the worst-case number of indirect blocks that will be used |
156 | * for ip's delayed extent of length "len". | |
1da177e4 | 157 | */ |
9e5987a7 DC |
158 | STATIC xfs_filblks_t |
159 | xfs_bmap_worst_indlen( | |
160 | xfs_inode_t *ip, /* incore inode pointer */ | |
161 | xfs_filblks_t len) /* delayed extent length */ | |
1da177e4 | 162 | { |
9e5987a7 DC |
163 | int level; /* btree level number */ |
164 | int maxrecs; /* maximum record count at this level */ | |
165 | xfs_mount_t *mp; /* mount structure */ | |
166 | xfs_filblks_t rval; /* return value */ | |
1da177e4 LT |
167 | |
168 | mp = ip->i_mount; | |
9e5987a7 DC |
169 | maxrecs = mp->m_bmap_dmxr[0]; |
170 | for (level = 0, rval = 0; | |
171 | level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK); | |
172 | level++) { | |
173 | len += maxrecs - 1; | |
174 | do_div(len, maxrecs); | |
175 | rval += len; | |
5e5c943c DW |
176 | if (len == 1) |
177 | return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) - | |
9e5987a7 DC |
178 | level - 1; |
179 | if (level == 0) | |
180 | maxrecs = mp->m_bmap_dmxr[1]; | |
1da177e4 | 181 | } |
9e5987a7 | 182 | return rval; |
1da177e4 LT |
183 | } |
184 | ||
185 | /* | |
9e5987a7 | 186 | * Calculate the default attribute fork offset for newly created inodes. |
1da177e4 | 187 | */ |
9e5987a7 DC |
188 | uint |
189 | xfs_default_attroffset( | |
190 | struct xfs_inode *ip) | |
1da177e4 | 191 | { |
9e5987a7 DC |
192 | struct xfs_mount *mp = ip->i_mount; |
193 | uint offset; | |
1da177e4 | 194 | |
9e5987a7 | 195 | if (mp->m_sb.sb_inodesize == 256) { |
56cea2d0 | 196 | offset = XFS_LITINO(mp, ip->i_d.di_version) - |
9e5987a7 DC |
197 | XFS_BMDR_SPACE_CALC(MINABTPTRS); |
198 | } else { | |
199 | offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); | |
1da177e4 | 200 | } |
9e5987a7 | 201 | |
56cea2d0 | 202 | ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version)); |
9e5987a7 | 203 | return offset; |
1da177e4 LT |
204 | } |
205 | ||
206 | /* | |
9e5987a7 DC |
207 | * Helper routine to reset inode di_forkoff field when switching |
208 | * attribute fork from local to extent format - we reset it where | |
209 | * possible to make space available for inline data fork extents. | |
1e82379b DC |
210 | */ |
211 | STATIC void | |
9e5987a7 | 212 | xfs_bmap_forkoff_reset( |
9e5987a7 DC |
213 | xfs_inode_t *ip, |
214 | int whichfork) | |
1e82379b | 215 | { |
9e5987a7 DC |
216 | if (whichfork == XFS_ATTR_FORK && |
217 | ip->i_d.di_format != XFS_DINODE_FMT_DEV && | |
9e5987a7 DC |
218 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE) { |
219 | uint dfl_forkoff = xfs_default_attroffset(ip) >> 3; | |
220 | ||
221 | if (dfl_forkoff > ip->i_d.di_forkoff) | |
222 | ip->i_d.di_forkoff = dfl_forkoff; | |
223 | } | |
1e82379b DC |
224 | } |
225 | ||
9e5987a7 DC |
226 | #ifdef DEBUG |
227 | STATIC struct xfs_buf * | |
228 | xfs_bmap_get_bp( | |
229 | struct xfs_btree_cur *cur, | |
230 | xfs_fsblock_t bno) | |
231 | { | |
e6631f85 | 232 | struct xfs_log_item *lip; |
9e5987a7 | 233 | int i; |
7574aa92 | 234 | |
9e5987a7 DC |
235 | if (!cur) |
236 | return NULL; | |
237 | ||
238 | for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) { | |
239 | if (!cur->bc_bufs[i]) | |
240 | break; | |
241 | if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno) | |
242 | return cur->bc_bufs[i]; | |
1da177e4 | 243 | } |
7574aa92 | 244 | |
9e5987a7 | 245 | /* Chase down all the log items to see if the bp is there */ |
e6631f85 DC |
246 | list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) { |
247 | struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip; | |
248 | ||
9e5987a7 DC |
249 | if (bip->bli_item.li_type == XFS_LI_BUF && |
250 | XFS_BUF_ADDR(bip->bli_buf) == bno) | |
251 | return bip->bli_buf; | |
252 | } | |
7574aa92 | 253 | |
9e5987a7 DC |
254 | return NULL; |
255 | } | |
0b1b213f | 256 | |
9e5987a7 DC |
257 | STATIC void |
258 | xfs_check_block( | |
259 | struct xfs_btree_block *block, | |
260 | xfs_mount_t *mp, | |
261 | int root, | |
262 | short sz) | |
263 | { | |
264 | int i, j, dmxr; | |
265 | __be64 *pp, *thispa; /* pointer to block address */ | |
266 | xfs_bmbt_key_t *prevp, *keyp; | |
1da177e4 | 267 | |
9e5987a7 | 268 | ASSERT(be16_to_cpu(block->bb_level) > 0); |
ec90c556 | 269 | |
9e5987a7 DC |
270 | prevp = NULL; |
271 | for( i = 1; i <= xfs_btree_get_numrecs(block); i++) { | |
272 | dmxr = mp->m_bmap_dmxr[0]; | |
273 | keyp = XFS_BMBT_KEY_ADDR(mp, block, i); | |
0b1b213f | 274 | |
9e5987a7 DC |
275 | if (prevp) { |
276 | ASSERT(be64_to_cpu(prevp->br_startoff) < | |
277 | be64_to_cpu(keyp->br_startoff)); | |
1da177e4 | 278 | } |
9e5987a7 | 279 | prevp = keyp; |
1da177e4 | 280 | |
1da177e4 | 281 | /* |
9e5987a7 | 282 | * Compare the block numbers to see if there are dups. |
1da177e4 | 283 | */ |
9e5987a7 DC |
284 | if (root) |
285 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz); | |
286 | else | |
287 | pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr); | |
0b1b213f | 288 | |
9e5987a7 DC |
289 | for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) { |
290 | if (root) | |
291 | thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz); | |
292 | else | |
293 | thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr); | |
294 | if (*thispa == *pp) { | |
295 | xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld", | |
296 | __func__, j, i, | |
297 | (unsigned long long)be64_to_cpu(*thispa)); | |
cec57256 | 298 | xfs_err(mp, "%s: ptrs are equal in node\n", |
9e5987a7 | 299 | __func__); |
cec57256 | 300 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
9e5987a7 | 301 | } |
1da177e4 | 302 | } |
9e5987a7 DC |
303 | } |
304 | } | |
1da177e4 | 305 | |
9e5987a7 DC |
306 | /* |
307 | * Check that the extents for the inode ip are in the right order in all | |
e3543819 DC |
308 | * btree leaves. THis becomes prohibitively expensive for large extent count |
309 | * files, so don't bother with inodes that have more than 10,000 extents in | |
310 | * them. The btree record ordering checks will still be done, so for such large | |
311 | * bmapbt constructs that is going to catch most corruptions. | |
9e5987a7 | 312 | */ |
9e5987a7 DC |
313 | STATIC void |
314 | xfs_bmap_check_leaf_extents( | |
315 | xfs_btree_cur_t *cur, /* btree cursor or null */ | |
316 | xfs_inode_t *ip, /* incore inode pointer */ | |
317 | int whichfork) /* data or attr fork */ | |
318 | { | |
319 | struct xfs_btree_block *block; /* current btree block */ | |
320 | xfs_fsblock_t bno; /* block # of "block" */ | |
321 | xfs_buf_t *bp; /* buffer for "block" */ | |
322 | int error; /* error return value */ | |
323 | xfs_extnum_t i=0, j; /* index into the extents list */ | |
3ba738df | 324 | struct xfs_ifork *ifp; /* fork structure */ |
9e5987a7 DC |
325 | int level; /* btree level, for checking */ |
326 | xfs_mount_t *mp; /* file system mount structure */ | |
327 | __be64 *pp; /* pointer to block address */ | |
328 | xfs_bmbt_rec_t *ep; /* pointer to current extent */ | |
329 | xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */ | |
330 | xfs_bmbt_rec_t *nextp; /* pointer to next extent */ | |
331 | int bp_release = 0; | |
332 | ||
333 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) { | |
334 | return; | |
335 | } | |
336 | ||
e3543819 DC |
337 | /* skip large extent count inodes */ |
338 | if (ip->i_d.di_nextents > 10000) | |
339 | return; | |
340 | ||
9e5987a7 DC |
341 | bno = NULLFSBLOCK; |
342 | mp = ip->i_mount; | |
343 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
344 | block = ifp->if_broot; | |
345 | /* | |
346 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
347 | */ | |
348 | level = be16_to_cpu(block->bb_level); | |
349 | ASSERT(level > 0); | |
350 | xfs_check_block(block, mp, 1, ifp->if_broot_bytes); | |
351 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
352 | bno = be64_to_cpu(*pp); | |
353 | ||
d5cf09ba | 354 | ASSERT(bno != NULLFSBLOCK); |
9e5987a7 DC |
355 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
356 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
357 | ||
358 | /* | |
359 | * Go down the tree until leaf level is reached, following the first | |
360 | * pointer (leftmost) at each level. | |
361 | */ | |
362 | while (level-- > 0) { | |
363 | /* See if buf is in cur first */ | |
364 | bp_release = 0; | |
365 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
366 | if (!bp) { | |
367 | bp_release = 1; | |
f5b999c0 | 368 | error = xfs_btree_read_bufl(mp, NULL, bno, &bp, |
9e5987a7 DC |
369 | XFS_BMAP_BTREE_REF, |
370 | &xfs_bmbt_buf_ops); | |
572a4cf0 | 371 | if (error) |
9e5987a7 | 372 | goto error_norelse; |
1da177e4 | 373 | } |
9e5987a7 | 374 | block = XFS_BUF_TO_BLOCK(bp); |
9e5987a7 DC |
375 | if (level == 0) |
376 | break; | |
1da177e4 | 377 | |
1da177e4 | 378 | /* |
9e5987a7 DC |
379 | * Check this block for basic sanity (increasing keys and |
380 | * no duplicate blocks). | |
1da177e4 | 381 | */ |
0b1b213f | 382 | |
9e5987a7 DC |
383 | xfs_check_block(block, mp, 0, 0); |
384 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
385 | bno = be64_to_cpu(*pp); | |
c29aad41 | 386 | XFS_WANT_CORRUPTED_GOTO(mp, |
59f6fec3 | 387 | xfs_verify_fsbno(mp, bno), error0); |
9e5987a7 DC |
388 | if (bp_release) { |
389 | bp_release = 0; | |
390 | xfs_trans_brelse(NULL, bp); | |
1da177e4 | 391 | } |
9e5987a7 | 392 | } |
ec90c556 | 393 | |
9e5987a7 DC |
394 | /* |
395 | * Here with bp and block set to the leftmost leaf node in the tree. | |
396 | */ | |
397 | i = 0; | |
398 | ||
399 | /* | |
400 | * Loop over all leaf nodes checking that all extents are in the right order. | |
401 | */ | |
402 | for (;;) { | |
403 | xfs_fsblock_t nextbno; | |
404 | xfs_extnum_t num_recs; | |
405 | ||
406 | ||
407 | num_recs = xfs_btree_get_numrecs(block); | |
1da177e4 | 408 | |
1da177e4 | 409 | /* |
9e5987a7 | 410 | * Read-ahead the next leaf block, if any. |
1da177e4 | 411 | */ |
8096b1eb | 412 | |
9e5987a7 | 413 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
1da177e4 | 414 | |
1da177e4 | 415 | /* |
9e5987a7 DC |
416 | * Check all the extents to make sure they are OK. |
417 | * If we had a previous block, the last entry should | |
418 | * conform with the first entry in this one. | |
1da177e4 | 419 | */ |
ec90c556 | 420 | |
9e5987a7 DC |
421 | ep = XFS_BMBT_REC_ADDR(mp, block, 1); |
422 | if (i) { | |
423 | ASSERT(xfs_bmbt_disk_get_startoff(&last) + | |
424 | xfs_bmbt_disk_get_blockcount(&last) <= | |
425 | xfs_bmbt_disk_get_startoff(ep)); | |
426 | } | |
427 | for (j = 1; j < num_recs; j++) { | |
428 | nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1); | |
429 | ASSERT(xfs_bmbt_disk_get_startoff(ep) + | |
430 | xfs_bmbt_disk_get_blockcount(ep) <= | |
431 | xfs_bmbt_disk_get_startoff(nextp)); | |
432 | ep = nextp; | |
433 | } | |
1da177e4 | 434 | |
9e5987a7 DC |
435 | last = *ep; |
436 | i += num_recs; | |
437 | if (bp_release) { | |
438 | bp_release = 0; | |
439 | xfs_trans_brelse(NULL, bp); | |
440 | } | |
441 | bno = nextbno; | |
1da177e4 | 442 | /* |
9e5987a7 | 443 | * If we've reached the end, stop. |
1da177e4 | 444 | */ |
9e5987a7 DC |
445 | if (bno == NULLFSBLOCK) |
446 | break; | |
8096b1eb | 447 | |
9e5987a7 DC |
448 | bp_release = 0; |
449 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
450 | if (!bp) { | |
451 | bp_release = 1; | |
f5b999c0 | 452 | error = xfs_btree_read_bufl(mp, NULL, bno, &bp, |
9e5987a7 DC |
453 | XFS_BMAP_BTREE_REF, |
454 | &xfs_bmbt_buf_ops); | |
b9b984d7 | 455 | if (error) |
9e5987a7 | 456 | goto error_norelse; |
1da177e4 | 457 | } |
9e5987a7 | 458 | block = XFS_BUF_TO_BLOCK(bp); |
a5bd606b | 459 | } |
a5fd276b | 460 | |
9e5987a7 | 461 | return; |
a5bd606b | 462 | |
9e5987a7 DC |
463 | error0: |
464 | xfs_warn(mp, "%s: at error0", __func__); | |
465 | if (bp_release) | |
466 | xfs_trans_brelse(NULL, bp); | |
467 | error_norelse: | |
468 | xfs_warn(mp, "%s: BAD after btree leaves for %d extents", | |
469 | __func__, i); | |
cec57256 DW |
470 | xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__); |
471 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
9e5987a7 | 472 | return; |
1da177e4 LT |
473 | } |
474 | ||
9e5987a7 DC |
475 | /* |
476 | * Validate that the bmbt_irecs being returned from bmapi are valid | |
a97f4df7 ZYW |
477 | * given the caller's original parameters. Specifically check the |
478 | * ranges of the returned irecs to ensure that they only extend beyond | |
9e5987a7 DC |
479 | * the given parameters if the XFS_BMAPI_ENTIRE flag was set. |
480 | */ | |
481 | STATIC void | |
482 | xfs_bmap_validate_ret( | |
483 | xfs_fileoff_t bno, | |
484 | xfs_filblks_t len, | |
485 | int flags, | |
486 | xfs_bmbt_irec_t *mval, | |
487 | int nmap, | |
488 | int ret_nmap) | |
489 | { | |
490 | int i; /* index to map values */ | |
a5bd606b | 491 | |
9e5987a7 | 492 | ASSERT(ret_nmap <= nmap); |
a5bd606b | 493 | |
9e5987a7 DC |
494 | for (i = 0; i < ret_nmap; i++) { |
495 | ASSERT(mval[i].br_blockcount > 0); | |
496 | if (!(flags & XFS_BMAPI_ENTIRE)) { | |
497 | ASSERT(mval[i].br_startoff >= bno); | |
498 | ASSERT(mval[i].br_blockcount <= len); | |
499 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount <= | |
500 | bno + len); | |
501 | } else { | |
502 | ASSERT(mval[i].br_startoff < bno + len); | |
503 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount > | |
504 | bno); | |
505 | } | |
506 | ASSERT(i == 0 || | |
507 | mval[i - 1].br_startoff + mval[i - 1].br_blockcount == | |
508 | mval[i].br_startoff); | |
509 | ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK && | |
510 | mval[i].br_startblock != HOLESTARTBLOCK); | |
511 | ASSERT(mval[i].br_state == XFS_EXT_NORM || | |
512 | mval[i].br_state == XFS_EXT_UNWRITTEN); | |
513 | } | |
514 | } | |
7574aa92 | 515 | |
9e5987a7 DC |
516 | #else |
517 | #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0) | |
7bf7a193 | 518 | #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0) |
9e5987a7 | 519 | #endif /* DEBUG */ |
7574aa92 | 520 | |
9e5987a7 DC |
521 | /* |
522 | * bmap free list manipulation functions | |
523 | */ | |
7574aa92 | 524 | |
9e5987a7 DC |
525 | /* |
526 | * Add the extent to the list of extents to be free at transaction end. | |
527 | * The list is maintained sorted (by block number). | |
528 | */ | |
529 | void | |
fcb762f5 | 530 | __xfs_bmap_add_free( |
0f37d178 | 531 | struct xfs_trans *tp, |
340785cc DW |
532 | xfs_fsblock_t bno, |
533 | xfs_filblks_t len, | |
66e3237e | 534 | const struct xfs_owner_info *oinfo, |
fcb762f5 | 535 | bool skip_discard) |
9e5987a7 | 536 | { |
310a75a3 | 537 | struct xfs_extent_free_item *new; /* new element */ |
9e5987a7 | 538 | #ifdef DEBUG |
0f37d178 BF |
539 | struct xfs_mount *mp = tp->t_mountp; |
540 | xfs_agnumber_t agno; | |
541 | xfs_agblock_t agbno; | |
9e5987a7 DC |
542 | |
543 | ASSERT(bno != NULLFSBLOCK); | |
544 | ASSERT(len > 0); | |
545 | ASSERT(len <= MAXEXTLEN); | |
546 | ASSERT(!isnullstartblock(bno)); | |
547 | agno = XFS_FSB_TO_AGNO(mp, bno); | |
548 | agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
549 | ASSERT(agno < mp->m_sb.sb_agcount); | |
550 | ASSERT(agbno < mp->m_sb.sb_agblocks); | |
551 | ASSERT(len < mp->m_sb.sb_agblocks); | |
552 | ASSERT(agbno + len <= mp->m_sb.sb_agblocks); | |
553 | #endif | |
554 | ASSERT(xfs_bmap_free_item_zone != NULL); | |
340785cc | 555 | |
9e5987a7 | 556 | new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP); |
310a75a3 DW |
557 | new->xefi_startblock = bno; |
558 | new->xefi_blockcount = (xfs_extlen_t)len; | |
340785cc DW |
559 | if (oinfo) |
560 | new->xefi_oinfo = *oinfo; | |
561 | else | |
7280feda | 562 | new->xefi_oinfo = XFS_RMAP_OINFO_SKIP_UPDATE; |
fcb762f5 | 563 | new->xefi_skip_discard = skip_discard; |
0f37d178 BF |
564 | trace_xfs_bmap_free_defer(tp->t_mountp, |
565 | XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0, | |
566 | XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len); | |
567 | xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list); | |
9e5987a7 | 568 | } |
0b1b213f | 569 | |
9e5987a7 DC |
570 | /* |
571 | * Inode fork format manipulation functions | |
572 | */ | |
1da177e4 | 573 | |
9e5987a7 | 574 | /* |
b101e334 CH |
575 | * Convert the inode format to extent format if it currently is in btree format, |
576 | * but the extent list is small enough that it fits into the extent format. | |
577 | * | |
578 | * Since the extents are already in-core, all we have to do is give up the space | |
579 | * for the btree root and pitch the leaf block. | |
9e5987a7 DC |
580 | */ |
581 | STATIC int /* error */ | |
582 | xfs_bmap_btree_to_extents( | |
b101e334 CH |
583 | struct xfs_trans *tp, /* transaction pointer */ |
584 | struct xfs_inode *ip, /* incore inode pointer */ | |
585 | struct xfs_btree_cur *cur, /* btree cursor */ | |
9e5987a7 DC |
586 | int *logflagsp, /* inode logging flags */ |
587 | int whichfork) /* data or attr fork */ | |
588 | { | |
b101e334 CH |
589 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
590 | struct xfs_mount *mp = ip->i_mount; | |
591 | struct xfs_btree_block *rblock = ifp->if_broot; | |
9e5987a7 DC |
592 | struct xfs_btree_block *cblock;/* child btree block */ |
593 | xfs_fsblock_t cbno; /* child block number */ | |
594 | xfs_buf_t *cbp; /* child block's buffer */ | |
595 | int error; /* error return value */ | |
9e5987a7 | 596 | __be64 *pp; /* ptr to block address */ |
340785cc | 597 | struct xfs_owner_info oinfo; |
1da177e4 | 598 | |
b101e334 CH |
599 | /* check if we actually need the extent format first: */ |
600 | if (!xfs_bmap_wants_extents(ip, whichfork)) | |
601 | return 0; | |
602 | ||
603 | ASSERT(cur); | |
60b4984f | 604 | ASSERT(whichfork != XFS_COW_FORK); |
9e5987a7 DC |
605 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
606 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
9e5987a7 DC |
607 | ASSERT(be16_to_cpu(rblock->bb_level) == 1); |
608 | ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1); | |
609 | ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1); | |
b101e334 | 610 | |
9e5987a7 DC |
611 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes); |
612 | cbno = be64_to_cpu(*pp); | |
9e5987a7 | 613 | #ifdef DEBUG |
f135761a DW |
614 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, |
615 | xfs_btree_check_lptr(cur, cbno, 1)); | |
9e5987a7 | 616 | #endif |
f5b999c0 | 617 | error = xfs_btree_read_bufl(mp, tp, cbno, &cbp, XFS_BMAP_BTREE_REF, |
9e5987a7 DC |
618 | &xfs_bmbt_buf_ops); |
619 | if (error) | |
620 | return error; | |
621 | cblock = XFS_BUF_TO_BLOCK(cbp); | |
622 | if ((error = xfs_btree_check_block(cur, cblock, 0, cbp))) | |
623 | return error; | |
340785cc | 624 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); |
0f37d178 | 625 | xfs_bmap_add_free(cur->bc_tp, cbno, 1, &oinfo); |
9e5987a7 DC |
626 | ip->i_d.di_nblocks--; |
627 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); | |
628 | xfs_trans_binval(tp, cbp); | |
629 | if (cur->bc_bufs[0] == cbp) | |
630 | cur->bc_bufs[0] = NULL; | |
631 | xfs_iroot_realloc(ip, -1, whichfork); | |
632 | ASSERT(ifp->if_broot == NULL); | |
633 | ASSERT((ifp->if_flags & XFS_IFBROOT) == 0); | |
634 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
b101e334 | 635 | *logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
9e5987a7 DC |
636 | return 0; |
637 | } | |
0b1b213f | 638 | |
9e5987a7 DC |
639 | /* |
640 | * Convert an extents-format file into a btree-format file. | |
641 | * The new file will have a root block (in the inode) and a single child block. | |
642 | */ | |
643 | STATIC int /* error */ | |
644 | xfs_bmap_extents_to_btree( | |
81ba8f3e BF |
645 | struct xfs_trans *tp, /* transaction pointer */ |
646 | struct xfs_inode *ip, /* incore inode pointer */ | |
81ba8f3e | 647 | struct xfs_btree_cur **curp, /* cursor returned to caller */ |
9e5987a7 DC |
648 | int wasdel, /* converting a delayed alloc */ |
649 | int *logflagsp, /* inode logging flags */ | |
650 | int whichfork) /* data or attr fork */ | |
651 | { | |
652 | struct xfs_btree_block *ablock; /* allocated (child) bt block */ | |
81ba8f3e BF |
653 | struct xfs_buf *abp; /* buffer for ablock */ |
654 | struct xfs_alloc_arg args; /* allocation arguments */ | |
655 | struct xfs_bmbt_rec *arp; /* child record pointer */ | |
9e5987a7 | 656 | struct xfs_btree_block *block; /* btree root block */ |
81ba8f3e | 657 | struct xfs_btree_cur *cur; /* bmap btree cursor */ |
9e5987a7 | 658 | int error; /* error return value */ |
81ba8f3e BF |
659 | struct xfs_ifork *ifp; /* inode fork pointer */ |
660 | struct xfs_bmbt_key *kp; /* root block key pointer */ | |
661 | struct xfs_mount *mp; /* mount structure */ | |
9e5987a7 | 662 | xfs_bmbt_ptr_t *pp; /* root block address pointer */ |
b2b1712a | 663 | struct xfs_iext_cursor icur; |
906abed5 | 664 | struct xfs_bmbt_irec rec; |
b2b1712a | 665 | xfs_extnum_t cnt = 0; |
1da177e4 | 666 | |
ee1a47ab | 667 | mp = ip->i_mount; |
60b4984f | 668 | ASSERT(whichfork != XFS_COW_FORK); |
9e5987a7 DC |
669 | ifp = XFS_IFORK_PTR(ip, whichfork); |
670 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS); | |
0b1b213f | 671 | |
9e5987a7 | 672 | /* |
e55ec4dd DC |
673 | * Make space in the inode incore. This needs to be undone if we fail |
674 | * to expand the root. | |
9e5987a7 DC |
675 | */ |
676 | xfs_iroot_realloc(ip, 1, whichfork); | |
677 | ifp->if_flags |= XFS_IFBROOT; | |
ec90c556 | 678 | |
9e5987a7 DC |
679 | /* |
680 | * Fill in the root. | |
681 | */ | |
682 | block = ifp->if_broot; | |
b6f41e44 ES |
683 | xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL, |
684 | XFS_BTNUM_BMAP, 1, 1, ip->i_ino, | |
f88ae46b | 685 | XFS_BTREE_LONG_PTRS); |
9e5987a7 DC |
686 | /* |
687 | * Need a cursor. Can't allocate until bb_level is filled in. | |
688 | */ | |
9e5987a7 | 689 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
9e5987a7 DC |
690 | cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; |
691 | /* | |
692 | * Convert to a btree with two levels, one record in root. | |
693 | */ | |
694 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE); | |
695 | memset(&args, 0, sizeof(args)); | |
696 | args.tp = tp; | |
697 | args.mp = mp; | |
340785cc | 698 | xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork); |
280253d2 | 699 | if (tp->t_firstblock == NULLFSBLOCK) { |
9e5987a7 DC |
700 | args.type = XFS_ALLOCTYPE_START_BNO; |
701 | args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino); | |
1214f1cf | 702 | } else if (tp->t_flags & XFS_TRANS_LOWMODE) { |
9e5987a7 | 703 | args.type = XFS_ALLOCTYPE_START_BNO; |
280253d2 | 704 | args.fsbno = tp->t_firstblock; |
9e5987a7 DC |
705 | } else { |
706 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
280253d2 | 707 | args.fsbno = tp->t_firstblock; |
9e5987a7 DC |
708 | } |
709 | args.minlen = args.maxlen = args.prod = 1; | |
710 | args.wasdel = wasdel; | |
711 | *logflagsp = 0; | |
e55ec4dd DC |
712 | error = xfs_alloc_vextent(&args); |
713 | if (error) | |
714 | goto out_root_realloc; | |
90e2056d | 715 | |
2fcc319d | 716 | if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) { |
01239d77 | 717 | error = -ENOSPC; |
e55ec4dd | 718 | goto out_root_realloc; |
2fcc319d | 719 | } |
e55ec4dd | 720 | |
9e5987a7 DC |
721 | /* |
722 | * Allocation can't fail, the space was reserved. | |
723 | */ | |
280253d2 BF |
724 | ASSERT(tp->t_firstblock == NULLFSBLOCK || |
725 | args.agno >= XFS_FSB_TO_AGNO(mp, tp->t_firstblock)); | |
cf612de7 | 726 | tp->t_firstblock = args.fsbno; |
9e5987a7 DC |
727 | cur->bc_private.b.allocated++; |
728 | ip->i_d.di_nblocks++; | |
729 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L); | |
f5b999c0 | 730 | abp = xfs_btree_get_bufl(mp, tp, args.fsbno); |
01239d77 | 731 | if (!abp) { |
e55ec4dd DC |
732 | error = -EFSCORRUPTED; |
733 | goto out_unreserve_dquot; | |
01239d77 | 734 | } |
e55ec4dd | 735 | |
9e5987a7 DC |
736 | /* |
737 | * Fill in the child block. | |
738 | */ | |
739 | abp->b_ops = &xfs_bmbt_buf_ops; | |
740 | ablock = XFS_BUF_TO_BLOCK(abp); | |
b6f41e44 ES |
741 | xfs_btree_init_block_int(mp, ablock, abp->b_bn, |
742 | XFS_BTNUM_BMAP, 0, 0, ip->i_ino, | |
ee1a47ab CH |
743 | XFS_BTREE_LONG_PTRS); |
744 | ||
b2b1712a | 745 | for_each_xfs_iext(ifp, &icur, &rec) { |
906abed5 CH |
746 | if (isnullstartblock(rec.br_startblock)) |
747 | continue; | |
748 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt); | |
749 | xfs_bmbt_disk_set_all(arp, &rec); | |
750 | cnt++; | |
9e5987a7 DC |
751 | } |
752 | ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork)); | |
753 | xfs_btree_set_numrecs(ablock, cnt); | |
1da177e4 | 754 | |
9e5987a7 DC |
755 | /* |
756 | * Fill in the root key and pointer. | |
757 | */ | |
758 | kp = XFS_BMBT_KEY_ADDR(mp, block, 1); | |
759 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); | |
760 | kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp)); | |
761 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur, | |
762 | be16_to_cpu(block->bb_level))); | |
763 | *pp = cpu_to_be64(args.fsbno); | |
ec90c556 | 764 | |
9e5987a7 DC |
765 | /* |
766 | * Do all this logging at the end so that | |
767 | * the root is at the right level. | |
768 | */ | |
769 | xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS); | |
770 | xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs)); | |
771 | ASSERT(*curp == NULL); | |
772 | *curp = cur; | |
773 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork); | |
774 | return 0; | |
01239d77 | 775 | |
e55ec4dd | 776 | out_unreserve_dquot: |
01239d77 | 777 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); |
e55ec4dd | 778 | out_root_realloc: |
01239d77 SH |
779 | xfs_iroot_realloc(ip, -1, whichfork); |
780 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
e55ec4dd | 781 | ASSERT(ifp->if_broot == NULL); |
01239d77 SH |
782 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); |
783 | ||
784 | return error; | |
9e5987a7 | 785 | } |
ec90c556 | 786 | |
9e5987a7 DC |
787 | /* |
788 | * Convert a local file to an extents file. | |
789 | * This code is out of bounds for data forks of regular files, | |
790 | * since the file data needs to get logged so things will stay consistent. | |
791 | * (The bmap-level manipulations are ok, though). | |
792 | */ | |
f3508bcd DC |
793 | void |
794 | xfs_bmap_local_to_extents_empty( | |
795 | struct xfs_inode *ip, | |
796 | int whichfork) | |
797 | { | |
798 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
799 | ||
60b4984f | 800 | ASSERT(whichfork != XFS_COW_FORK); |
f3508bcd DC |
801 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); |
802 | ASSERT(ifp->if_bytes == 0); | |
803 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0); | |
804 | ||
6a9edd3d | 805 | xfs_bmap_forkoff_reset(ip, whichfork); |
f3508bcd DC |
806 | ifp->if_flags &= ~XFS_IFINLINE; |
807 | ifp->if_flags |= XFS_IFEXTENTS; | |
6bdcf26a CH |
808 | ifp->if_u1.if_root = NULL; |
809 | ifp->if_height = 0; | |
f3508bcd DC |
810 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); |
811 | } | |
812 | ||
813 | ||
9e5987a7 DC |
814 | STATIC int /* error */ |
815 | xfs_bmap_local_to_extents( | |
816 | xfs_trans_t *tp, /* transaction pointer */ | |
817 | xfs_inode_t *ip, /* incore inode pointer */ | |
9e5987a7 DC |
818 | xfs_extlen_t total, /* total blocks needed by transaction */ |
819 | int *logflagsp, /* inode logging flags */ | |
820 | int whichfork, | |
ee1a47ab CH |
821 | void (*init_fn)(struct xfs_trans *tp, |
822 | struct xfs_buf *bp, | |
9e5987a7 DC |
823 | struct xfs_inode *ip, |
824 | struct xfs_ifork *ifp)) | |
825 | { | |
f3508bcd | 826 | int error = 0; |
9e5987a7 | 827 | int flags; /* logging flags returned */ |
3ba738df | 828 | struct xfs_ifork *ifp; /* inode fork pointer */ |
f3508bcd DC |
829 | xfs_alloc_arg_t args; /* allocation arguments */ |
830 | xfs_buf_t *bp; /* buffer for extent block */ | |
50bb44c2 | 831 | struct xfs_bmbt_irec rec; |
b2b1712a | 832 | struct xfs_iext_cursor icur; |
0b1b213f | 833 | |
9e5987a7 DC |
834 | /* |
835 | * We don't want to deal with the case of keeping inode data inline yet. | |
836 | * So sending the data fork of a regular inode is invalid. | |
837 | */ | |
c19b3b05 | 838 | ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK)); |
9e5987a7 DC |
839 | ifp = XFS_IFORK_PTR(ip, whichfork); |
840 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
f3508bcd DC |
841 | |
842 | if (!ifp->if_bytes) { | |
843 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
844 | flags = XFS_ILOG_CORE; | |
845 | goto done; | |
846 | } | |
847 | ||
9e5987a7 DC |
848 | flags = 0; |
849 | error = 0; | |
6bdcf26a | 850 | ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS)) == XFS_IFINLINE); |
f3508bcd DC |
851 | memset(&args, 0, sizeof(args)); |
852 | args.tp = tp; | |
853 | args.mp = ip->i_mount; | |
340785cc | 854 | xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0); |
f3508bcd DC |
855 | /* |
856 | * Allocate a block. We know we need only one, since the | |
857 | * file currently fits in an inode. | |
858 | */ | |
280253d2 | 859 | if (tp->t_firstblock == NULLFSBLOCK) { |
f3508bcd DC |
860 | args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino); |
861 | args.type = XFS_ALLOCTYPE_START_BNO; | |
9e5987a7 | 862 | } else { |
280253d2 | 863 | args.fsbno = tp->t_firstblock; |
f3508bcd | 864 | args.type = XFS_ALLOCTYPE_NEAR_BNO; |
9e5987a7 | 865 | } |
f3508bcd DC |
866 | args.total = total; |
867 | args.minlen = args.maxlen = args.prod = 1; | |
868 | error = xfs_alloc_vextent(&args); | |
869 | if (error) | |
870 | goto done; | |
871 | ||
872 | /* Can't fail, the space was reserved. */ | |
873 | ASSERT(args.fsbno != NULLFSBLOCK); | |
874 | ASSERT(args.len == 1); | |
280253d2 | 875 | tp->t_firstblock = args.fsbno; |
f5b999c0 | 876 | bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno); |
f3508bcd | 877 | |
fe22d552 | 878 | /* |
b7cdc66b | 879 | * Initialize the block, copy the data and log the remote buffer. |
fe22d552 | 880 | * |
b7cdc66b BF |
881 | * The callout is responsible for logging because the remote format |
882 | * might differ from the local format and thus we don't know how much to | |
883 | * log here. Note that init_fn must also set the buffer log item type | |
884 | * correctly. | |
fe22d552 | 885 | */ |
f3508bcd DC |
886 | init_fn(tp, bp, ip, ifp); |
887 | ||
b7cdc66b | 888 | /* account for the change in fork size */ |
f3508bcd DC |
889 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); |
890 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
9e5987a7 | 891 | flags |= XFS_ILOG_CORE; |
f3508bcd | 892 | |
6bdcf26a CH |
893 | ifp->if_u1.if_root = NULL; |
894 | ifp->if_height = 0; | |
895 | ||
50bb44c2 CH |
896 | rec.br_startoff = 0; |
897 | rec.br_startblock = args.fsbno; | |
898 | rec.br_blockcount = 1; | |
899 | rec.br_state = XFS_EXT_NORM; | |
b2b1712a | 900 | xfs_iext_first(ifp, &icur); |
0254c2f2 | 901 | xfs_iext_insert(ip, &icur, &rec, 0); |
50bb44c2 | 902 | |
f3508bcd DC |
903 | XFS_IFORK_NEXT_SET(ip, whichfork, 1); |
904 | ip->i_d.di_nblocks = 1; | |
905 | xfs_trans_mod_dquot_byino(tp, ip, | |
906 | XFS_TRANS_DQ_BCOUNT, 1L); | |
907 | flags |= xfs_ilog_fext(whichfork); | |
908 | ||
9e5987a7 DC |
909 | done: |
910 | *logflagsp = flags; | |
911 | return error; | |
912 | } | |
ec90c556 | 913 | |
9e5987a7 DC |
914 | /* |
915 | * Called from xfs_bmap_add_attrfork to handle btree format files. | |
916 | */ | |
917 | STATIC int /* error */ | |
918 | xfs_bmap_add_attrfork_btree( | |
919 | xfs_trans_t *tp, /* transaction pointer */ | |
920 | xfs_inode_t *ip, /* incore inode pointer */ | |
9e5987a7 DC |
921 | int *flags) /* inode logging flags */ |
922 | { | |
923 | xfs_btree_cur_t *cur; /* btree cursor */ | |
924 | int error; /* error return value */ | |
925 | xfs_mount_t *mp; /* file system mount struct */ | |
926 | int stat; /* newroot status */ | |
ec90c556 | 927 | |
9e5987a7 DC |
928 | mp = ip->i_mount; |
929 | if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip)) | |
930 | *flags |= XFS_ILOG_DBROOT; | |
931 | else { | |
932 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
b5cfbc22 CH |
933 | error = xfs_bmbt_lookup_first(cur, &stat); |
934 | if (error) | |
9e5987a7 DC |
935 | goto error0; |
936 | /* must be at least one entry */ | |
c29aad41 | 937 | XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0); |
9e5987a7 DC |
938 | if ((error = xfs_btree_new_iroot(cur, flags, &stat))) |
939 | goto error0; | |
940 | if (stat == 0) { | |
941 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2451337d | 942 | return -ENOSPC; |
1da177e4 | 943 | } |
9e5987a7 DC |
944 | cur->bc_private.b.allocated = 0; |
945 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
1da177e4 | 946 | } |
9e5987a7 DC |
947 | return 0; |
948 | error0: | |
949 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
950 | return error; | |
951 | } | |
a5bd606b | 952 | |
9e5987a7 DC |
953 | /* |
954 | * Called from xfs_bmap_add_attrfork to handle extents format files. | |
955 | */ | |
956 | STATIC int /* error */ | |
957 | xfs_bmap_add_attrfork_extents( | |
81ba8f3e BF |
958 | struct xfs_trans *tp, /* transaction pointer */ |
959 | struct xfs_inode *ip, /* incore inode pointer */ | |
9e5987a7 DC |
960 | int *flags) /* inode logging flags */ |
961 | { | |
962 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
963 | int error; /* error return value */ | |
a5bd606b | 964 | |
9e5987a7 DC |
965 | if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip)) |
966 | return 0; | |
967 | cur = NULL; | |
280253d2 BF |
968 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags, |
969 | XFS_DATA_FORK); | |
a5bd606b CH |
970 | if (cur) { |
971 | cur->bc_private.b.allocated = 0; | |
0b04b6b8 | 972 | xfs_btree_del_cursor(cur, error); |
a5bd606b | 973 | } |
1da177e4 | 974 | return error; |
1da177e4 LT |
975 | } |
976 | ||
9e5987a7 DC |
977 | /* |
978 | * Called from xfs_bmap_add_attrfork to handle local format files. Each | |
979 | * different data fork content type needs a different callout to do the | |
980 | * conversion. Some are basic and only require special block initialisation | |
981 | * callouts for the data formating, others (directories) are so specialised they | |
982 | * handle everything themselves. | |
983 | * | |
984 | * XXX (dgc): investigate whether directory conversion can use the generic | |
985 | * formatting callout. It should be possible - it's just a very complex | |
ee1a47ab | 986 | * formatter. |
9e5987a7 DC |
987 | */ |
988 | STATIC int /* error */ | |
989 | xfs_bmap_add_attrfork_local( | |
825d75cd BF |
990 | struct xfs_trans *tp, /* transaction pointer */ |
991 | struct xfs_inode *ip, /* incore inode pointer */ | |
9e5987a7 DC |
992 | int *flags) /* inode logging flags */ |
993 | { | |
825d75cd | 994 | struct xfs_da_args dargs; /* args for dir/attr code */ |
7574aa92 | 995 | |
9e5987a7 DC |
996 | if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip)) |
997 | return 0; | |
7574aa92 | 998 | |
c19b3b05 | 999 | if (S_ISDIR(VFS_I(ip)->i_mode)) { |
9e5987a7 | 1000 | memset(&dargs, 0, sizeof(dargs)); |
d6cf1305 | 1001 | dargs.geo = ip->i_mount->m_dir_geo; |
9e5987a7 | 1002 | dargs.dp = ip; |
d6cf1305 | 1003 | dargs.total = dargs.geo->fsbcount; |
9e5987a7 DC |
1004 | dargs.whichfork = XFS_DATA_FORK; |
1005 | dargs.trans = tp; | |
1006 | return xfs_dir2_sf_to_block(&dargs); | |
1da177e4 | 1007 | } |
7574aa92 | 1008 | |
c19b3b05 | 1009 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
280253d2 BF |
1010 | return xfs_bmap_local_to_extents(tp, ip, 1, flags, |
1011 | XFS_DATA_FORK, | |
9e5987a7 | 1012 | xfs_symlink_local_to_remote); |
7574aa92 | 1013 | |
f3508bcd DC |
1014 | /* should only be called for types that support local format data */ |
1015 | ASSERT(0); | |
2451337d | 1016 | return -EFSCORRUPTED; |
9e5987a7 | 1017 | } |
0b1b213f | 1018 | |
2f3cd809 AH |
1019 | /* Set an inode attr fork off based on the format */ |
1020 | int | |
1021 | xfs_bmap_set_attrforkoff( | |
1022 | struct xfs_inode *ip, | |
1023 | int size, | |
1024 | int *version) | |
1025 | { | |
1026 | switch (ip->i_d.di_format) { | |
1027 | case XFS_DINODE_FMT_DEV: | |
1028 | ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; | |
1029 | break; | |
1030 | case XFS_DINODE_FMT_LOCAL: | |
1031 | case XFS_DINODE_FMT_EXTENTS: | |
1032 | case XFS_DINODE_FMT_BTREE: | |
1033 | ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size); | |
1034 | if (!ip->i_d.di_forkoff) | |
1035 | ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3; | |
1036 | else if ((ip->i_mount->m_flags & XFS_MOUNT_ATTR2) && version) | |
1037 | *version = 2; | |
1038 | break; | |
1039 | default: | |
1040 | ASSERT(0); | |
1041 | return -EINVAL; | |
1042 | } | |
1043 | ||
1044 | return 0; | |
1045 | } | |
1046 | ||
9e5987a7 DC |
1047 | /* |
1048 | * Convert inode from non-attributed to attributed. | |
1049 | * Must not be in a transaction, ip must not be locked. | |
1050 | */ | |
1051 | int /* error code */ | |
1052 | xfs_bmap_add_attrfork( | |
1053 | xfs_inode_t *ip, /* incore inode pointer */ | |
1054 | int size, /* space new attribute needs */ | |
1055 | int rsvd) /* xact may use reserved blks */ | |
1056 | { | |
9e5987a7 DC |
1057 | xfs_mount_t *mp; /* mount structure */ |
1058 | xfs_trans_t *tp; /* transaction pointer */ | |
1059 | int blks; /* space reservation */ | |
1060 | int version = 1; /* superblock attr version */ | |
9e5987a7 DC |
1061 | int logflags; /* logging flags */ |
1062 | int error; /* error return value */ | |
0b1b213f | 1063 | |
9e5987a7 | 1064 | ASSERT(XFS_IFORK_Q(ip) == 0); |
1da177e4 | 1065 | |
9e5987a7 DC |
1066 | mp = ip->i_mount; |
1067 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); | |
253f4911 | 1068 | |
9e5987a7 | 1069 | blks = XFS_ADDAFORK_SPACE_RES(mp); |
253f4911 CH |
1070 | |
1071 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0, | |
1072 | rsvd ? XFS_TRANS_RESERVE : 0, &tp); | |
1073 | if (error) | |
9e3908e3 | 1074 | return error; |
253f4911 | 1075 | |
9e5987a7 DC |
1076 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1077 | error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ? | |
1078 | XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES : | |
1079 | XFS_QMOPT_RES_REGBLKS); | |
9e3908e3 MT |
1080 | if (error) |
1081 | goto trans_cancel; | |
9e5987a7 | 1082 | if (XFS_IFORK_Q(ip)) |
9e3908e3 | 1083 | goto trans_cancel; |
0f352f8e DW |
1084 | if (ip->i_d.di_anextents != 0) { |
1085 | error = -EFSCORRUPTED; | |
1086 | goto trans_cancel; | |
1087 | } | |
9e5987a7 | 1088 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) { |
1da177e4 | 1089 | /* |
9e5987a7 | 1090 | * For inodes coming from pre-6.2 filesystems. |
1da177e4 | 1091 | */ |
9e5987a7 DC |
1092 | ASSERT(ip->i_d.di_aformat == 0); |
1093 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
1094 | } | |
ec90c556 | 1095 | |
9e3908e3 | 1096 | xfs_trans_ijoin(tp, ip, 0); |
9e5987a7 | 1097 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
2f3cd809 AH |
1098 | error = xfs_bmap_set_attrforkoff(ip, size, &version); |
1099 | if (error) | |
9e3908e3 | 1100 | goto trans_cancel; |
9e5987a7 DC |
1101 | ASSERT(ip->i_afp == NULL); |
1102 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
1103 | ip->i_afp->if_flags = XFS_IFEXTENTS; | |
1104 | logflags = 0; | |
9e5987a7 DC |
1105 | switch (ip->i_d.di_format) { |
1106 | case XFS_DINODE_FMT_LOCAL: | |
825d75cd | 1107 | error = xfs_bmap_add_attrfork_local(tp, ip, &logflags); |
9e5987a7 DC |
1108 | break; |
1109 | case XFS_DINODE_FMT_EXTENTS: | |
825d75cd | 1110 | error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags); |
9e5987a7 DC |
1111 | break; |
1112 | case XFS_DINODE_FMT_BTREE: | |
825d75cd | 1113 | error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags); |
9e5987a7 DC |
1114 | break; |
1115 | default: | |
1116 | error = 0; | |
1da177e4 LT |
1117 | break; |
1118 | } | |
9e5987a7 DC |
1119 | if (logflags) |
1120 | xfs_trans_log_inode(tp, ip, logflags); | |
1121 | if (error) | |
c8eac49e | 1122 | goto trans_cancel; |
9e5987a7 DC |
1123 | if (!xfs_sb_version_hasattr(&mp->m_sb) || |
1124 | (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) { | |
61e63ecb | 1125 | bool log_sb = false; |
9e5987a7 DC |
1126 | |
1127 | spin_lock(&mp->m_sb_lock); | |
1128 | if (!xfs_sb_version_hasattr(&mp->m_sb)) { | |
1129 | xfs_sb_version_addattr(&mp->m_sb); | |
61e63ecb | 1130 | log_sb = true; |
9e5987a7 DC |
1131 | } |
1132 | if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) { | |
1133 | xfs_sb_version_addattr2(&mp->m_sb); | |
61e63ecb | 1134 | log_sb = true; |
9e5987a7 | 1135 | } |
4d11a402 | 1136 | spin_unlock(&mp->m_sb_lock); |
61e63ecb DC |
1137 | if (log_sb) |
1138 | xfs_log_sb(tp); | |
1da177e4 | 1139 | } |
9e5987a7 | 1140 | |
70393313 | 1141 | error = xfs_trans_commit(tp); |
9e3908e3 MT |
1142 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1143 | return error; | |
1144 | ||
9e3908e3 | 1145 | trans_cancel: |
4906e215 | 1146 | xfs_trans_cancel(tp); |
9e5987a7 | 1147 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
9e5987a7 | 1148 | return error; |
1da177e4 LT |
1149 | } |
1150 | ||
1151 | /* | |
9e5987a7 | 1152 | * Internal and external extent tree search functions. |
1da177e4 | 1153 | */ |
a5bd606b | 1154 | |
9e5987a7 | 1155 | /* |
211e95bb | 1156 | * Read in extents from a btree-format inode. |
9e5987a7 | 1157 | */ |
211e95bb CH |
1158 | int |
1159 | xfs_iread_extents( | |
1160 | struct xfs_trans *tp, | |
1161 | struct xfs_inode *ip, | |
1162 | int whichfork) | |
9e5987a7 | 1163 | { |
211e95bb | 1164 | struct xfs_mount *mp = ip->i_mount; |
e8e0e170 | 1165 | int state = xfs_bmap_fork_to_state(whichfork); |
211e95bb CH |
1166 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1167 | xfs_extnum_t nextents = XFS_IFORK_NEXTENTS(ip, whichfork); | |
1168 | struct xfs_btree_block *block = ifp->if_broot; | |
b2b1712a | 1169 | struct xfs_iext_cursor icur; |
6bdcf26a | 1170 | struct xfs_bmbt_irec new; |
211e95bb CH |
1171 | xfs_fsblock_t bno; |
1172 | struct xfs_buf *bp; | |
1173 | xfs_extnum_t i, j; | |
1174 | int level; | |
1175 | __be64 *pp; | |
1176 | int error; | |
1177 | ||
1178 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
1179 | ||
1180 | if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
1181 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
1182 | return -EFSCORRUPTED; | |
1183 | } | |
1184 | ||
1da177e4 | 1185 | /* |
9e5987a7 | 1186 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. |
1da177e4 | 1187 | */ |
9e5987a7 | 1188 | level = be16_to_cpu(block->bb_level); |
6958d11f BF |
1189 | if (unlikely(level == 0)) { |
1190 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
1191 | return -EFSCORRUPTED; | |
1192 | } | |
9e5987a7 DC |
1193 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); |
1194 | bno = be64_to_cpu(*pp); | |
d5a91bae | 1195 | |
1da177e4 | 1196 | /* |
9e5987a7 DC |
1197 | * Go down the tree until leaf level is reached, following the first |
1198 | * pointer (leftmost) at each level. | |
1da177e4 | 1199 | */ |
9e5987a7 | 1200 | while (level-- > 0) { |
f5b999c0 | 1201 | error = xfs_btree_read_bufl(mp, tp, bno, &bp, |
9e5987a7 DC |
1202 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); |
1203 | if (error) | |
211e95bb | 1204 | goto out; |
9e5987a7 | 1205 | block = XFS_BUF_TO_BLOCK(bp); |
9e5987a7 DC |
1206 | if (level == 0) |
1207 | break; | |
1208 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
1209 | bno = be64_to_cpu(*pp); | |
c29aad41 | 1210 | XFS_WANT_CORRUPTED_GOTO(mp, |
59f6fec3 | 1211 | xfs_verify_fsbno(mp, bno), out_brelse); |
9e5987a7 | 1212 | xfs_trans_brelse(tp, bp); |
1da177e4 | 1213 | } |
211e95bb | 1214 | |
1da177e4 | 1215 | /* |
9e5987a7 | 1216 | * Here with bp and block set to the leftmost leaf node in the tree. |
1da177e4 | 1217 | */ |
9e5987a7 | 1218 | i = 0; |
b2b1712a | 1219 | xfs_iext_first(ifp, &icur); |
211e95bb | 1220 | |
1da177e4 | 1221 | /* |
9e5987a7 | 1222 | * Loop over all leaf nodes. Copy information to the extent records. |
1da177e4 | 1223 | */ |
9e5987a7 DC |
1224 | for (;;) { |
1225 | xfs_bmbt_rec_t *frp; | |
1226 | xfs_fsblock_t nextbno; | |
1227 | xfs_extnum_t num_recs; | |
0b1b213f | 1228 | |
9e5987a7 | 1229 | num_recs = xfs_btree_get_numrecs(block); |
211e95bb | 1230 | if (unlikely(i + num_recs > nextents)) { |
9e5987a7 DC |
1231 | xfs_warn(ip->i_mount, |
1232 | "corrupt dinode %Lu, (btree extents).", | |
1233 | (unsigned long long) ip->i_ino); | |
90a58f95 DW |
1234 | xfs_inode_verifier_error(ip, -EFSCORRUPTED, |
1235 | __func__, block, sizeof(*block), | |
1236 | __this_address); | |
211e95bb CH |
1237 | error = -EFSCORRUPTED; |
1238 | goto out_brelse; | |
1da177e4 | 1239 | } |
1da177e4 | 1240 | /* |
9e5987a7 | 1241 | * Read-ahead the next leaf block, if any. |
1da177e4 | 1242 | */ |
9e5987a7 DC |
1243 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
1244 | if (nextbno != NULLFSBLOCK) | |
1245 | xfs_btree_reada_bufl(mp, nextbno, 1, | |
1246 | &xfs_bmbt_buf_ops); | |
1da177e4 | 1247 | /* |
9e5987a7 | 1248 | * Copy records into the extent records. |
1da177e4 | 1249 | */ |
9e5987a7 | 1250 | frp = XFS_BMBT_REC_ADDR(mp, block, 1); |
6bdcf26a | 1251 | for (j = 0; j < num_recs; j++, frp++, i++) { |
30b0984d DW |
1252 | xfs_failaddr_t fa; |
1253 | ||
dac9c9b1 | 1254 | xfs_bmbt_disk_get_all(frp, &new); |
30b0984d DW |
1255 | fa = xfs_bmap_validate_extent(ip, whichfork, &new); |
1256 | if (fa) { | |
211e95bb | 1257 | error = -EFSCORRUPTED; |
30b0984d DW |
1258 | xfs_inode_verifier_error(ip, error, |
1259 | "xfs_iread_extents(2)", | |
1260 | frp, sizeof(*frp), fa); | |
211e95bb | 1261 | goto out_brelse; |
9e5987a7 | 1262 | } |
0254c2f2 | 1263 | xfs_iext_insert(ip, &icur, &new, state); |
b2b1712a CH |
1264 | trace_xfs_read_extent(ip, &icur, state, _THIS_IP_); |
1265 | xfs_iext_next(ifp, &icur); | |
9e5987a7 DC |
1266 | } |
1267 | xfs_trans_brelse(tp, bp); | |
1268 | bno = nextbno; | |
1da177e4 | 1269 | /* |
9e5987a7 | 1270 | * If we've reached the end, stop. |
1da177e4 | 1271 | */ |
9e5987a7 DC |
1272 | if (bno == NULLFSBLOCK) |
1273 | break; | |
f5b999c0 | 1274 | error = xfs_btree_read_bufl(mp, tp, bno, &bp, |
9e5987a7 DC |
1275 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); |
1276 | if (error) | |
211e95bb | 1277 | goto out; |
9e5987a7 | 1278 | block = XFS_BUF_TO_BLOCK(bp); |
1da177e4 | 1279 | } |
211e95bb CH |
1280 | |
1281 | if (i != XFS_IFORK_NEXTENTS(ip, whichfork)) { | |
1282 | error = -EFSCORRUPTED; | |
1283 | goto out; | |
1284 | } | |
5d829300 | 1285 | ASSERT(i == xfs_iext_count(ifp)); |
211e95bb CH |
1286 | |
1287 | ifp->if_flags |= XFS_IFEXTENTS; | |
9e5987a7 | 1288 | return 0; |
211e95bb CH |
1289 | |
1290 | out_brelse: | |
9e5987a7 | 1291 | xfs_trans_brelse(tp, bp); |
211e95bb CH |
1292 | out: |
1293 | xfs_iext_destroy(ifp); | |
1294 | return error; | |
9e5987a7 | 1295 | } |
a5bd606b | 1296 | |
9e5987a7 | 1297 | /* |
29b3e94a CH |
1298 | * Returns the relative block number of the first unused block(s) in the given |
1299 | * fork with at least "len" logically contiguous blocks free. This is the | |
1300 | * lowest-address hole if the fork has holes, else the first block past the end | |
1301 | * of fork. Return 0 if the fork is currently local (in-inode). | |
9e5987a7 DC |
1302 | */ |
1303 | int /* error */ | |
1304 | xfs_bmap_first_unused( | |
29b3e94a CH |
1305 | struct xfs_trans *tp, /* transaction pointer */ |
1306 | struct xfs_inode *ip, /* incore inode */ | |
1307 | xfs_extlen_t len, /* size of hole to find */ | |
1308 | xfs_fileoff_t *first_unused, /* unused block */ | |
1309 | int whichfork) /* data or attr fork */ | |
9e5987a7 | 1310 | { |
29b3e94a CH |
1311 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1312 | struct xfs_bmbt_irec got; | |
b2b1712a | 1313 | struct xfs_iext_cursor icur; |
29b3e94a CH |
1314 | xfs_fileoff_t lastaddr = 0; |
1315 | xfs_fileoff_t lowest, max; | |
1316 | int error; | |
9e5987a7 DC |
1317 | |
1318 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE || | |
1319 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS || | |
1320 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
29b3e94a | 1321 | |
9e5987a7 DC |
1322 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { |
1323 | *first_unused = 0; | |
1324 | return 0; | |
dd9f438e | 1325 | } |
f2285c14 | 1326 | |
29b3e94a CH |
1327 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
1328 | error = xfs_iread_extents(tp, ip, whichfork); | |
1329 | if (error) | |
1330 | return error; | |
1331 | } | |
f2285c14 | 1332 | |
29b3e94a | 1333 | lowest = max = *first_unused; |
b2b1712a | 1334 | for_each_xfs_iext(ifp, &icur, &got) { |
9e5987a7 DC |
1335 | /* |
1336 | * See if the hole before this extent will work. | |
1337 | */ | |
f2285c14 | 1338 | if (got.br_startoff >= lowest + len && |
29b3e94a CH |
1339 | got.br_startoff - max >= len) |
1340 | break; | |
f2285c14 | 1341 | lastaddr = got.br_startoff + got.br_blockcount; |
9e5987a7 | 1342 | max = XFS_FILEOFF_MAX(lastaddr, lowest); |
dd9f438e | 1343 | } |
29b3e94a | 1344 | |
9e5987a7 DC |
1345 | *first_unused = max; |
1346 | return 0; | |
1347 | } | |
1348 | ||
1349 | /* | |
02bb4873 | 1350 | * Returns the file-relative block number of the last block - 1 before |
9e5987a7 DC |
1351 | * last_block (input value) in the file. |
1352 | * This is not based on i_size, it is based on the extent records. | |
1353 | * Returns 0 for local files, as they do not have extent records. | |
1354 | */ | |
1355 | int /* error */ | |
1356 | xfs_bmap_last_before( | |
86685f7b CH |
1357 | struct xfs_trans *tp, /* transaction pointer */ |
1358 | struct xfs_inode *ip, /* incore inode */ | |
1359 | xfs_fileoff_t *last_block, /* last block */ | |
1360 | int whichfork) /* data or attr fork */ | |
9e5987a7 | 1361 | { |
86685f7b CH |
1362 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1363 | struct xfs_bmbt_irec got; | |
b2b1712a | 1364 | struct xfs_iext_cursor icur; |
86685f7b | 1365 | int error; |
9e5987a7 | 1366 | |
86685f7b CH |
1367 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { |
1368 | case XFS_DINODE_FMT_LOCAL: | |
9e5987a7 DC |
1369 | *last_block = 0; |
1370 | return 0; | |
86685f7b CH |
1371 | case XFS_DINODE_FMT_BTREE: |
1372 | case XFS_DINODE_FMT_EXTENTS: | |
1373 | break; | |
1374 | default: | |
1375 | return -EIO; | |
9e5987a7 | 1376 | } |
86685f7b CH |
1377 | |
1378 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1379 | error = xfs_iread_extents(tp, ip, whichfork); | |
1380 | if (error) | |
1381 | return error; | |
dd9f438e | 1382 | } |
86685f7b | 1383 | |
b2b1712a | 1384 | if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got)) |
dc56015f | 1385 | *last_block = 0; |
9e5987a7 DC |
1386 | return 0; |
1387 | } | |
1388 | ||
68988114 | 1389 | int |
9e5987a7 DC |
1390 | xfs_bmap_last_extent( |
1391 | struct xfs_trans *tp, | |
1392 | struct xfs_inode *ip, | |
1393 | int whichfork, | |
1394 | struct xfs_bmbt_irec *rec, | |
1395 | int *is_empty) | |
1396 | { | |
1397 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
b2b1712a | 1398 | struct xfs_iext_cursor icur; |
9e5987a7 | 1399 | int error; |
9e5987a7 DC |
1400 | |
1401 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1402 | error = xfs_iread_extents(tp, ip, whichfork); | |
1403 | if (error) | |
1404 | return error; | |
dd9f438e NS |
1405 | } |
1406 | ||
b2b1712a CH |
1407 | xfs_iext_last(ifp, &icur); |
1408 | if (!xfs_iext_get_extent(ifp, &icur, rec)) | |
9e5987a7 | 1409 | *is_empty = 1; |
b2b1712a CH |
1410 | else |
1411 | *is_empty = 0; | |
dd9f438e NS |
1412 | return 0; |
1413 | } | |
1414 | ||
9e5987a7 DC |
1415 | /* |
1416 | * Check the last inode extent to determine whether this allocation will result | |
1417 | * in blocks being allocated at the end of the file. When we allocate new data | |
1418 | * blocks at the end of the file which do not start at the previous data block, | |
1419 | * we will try to align the new blocks at stripe unit boundaries. | |
1420 | * | |
6e708bcf | 1421 | * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be |
9e5987a7 DC |
1422 | * at, or past the EOF. |
1423 | */ | |
1424 | STATIC int | |
1425 | xfs_bmap_isaeof( | |
1426 | struct xfs_bmalloca *bma, | |
1427 | int whichfork) | |
1da177e4 | 1428 | { |
9e5987a7 DC |
1429 | struct xfs_bmbt_irec rec; |
1430 | int is_empty; | |
1431 | int error; | |
1da177e4 | 1432 | |
749f24f3 | 1433 | bma->aeof = false; |
9e5987a7 DC |
1434 | error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec, |
1435 | &is_empty); | |
6e708bcf | 1436 | if (error) |
9e5987a7 | 1437 | return error; |
1da177e4 | 1438 | |
6e708bcf | 1439 | if (is_empty) { |
749f24f3 | 1440 | bma->aeof = true; |
6e708bcf DC |
1441 | return 0; |
1442 | } | |
1443 | ||
1da177e4 | 1444 | /* |
9e5987a7 DC |
1445 | * Check if we are allocation or past the last extent, or at least into |
1446 | * the last delayed allocated extent. | |
1da177e4 | 1447 | */ |
9e5987a7 DC |
1448 | bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount || |
1449 | (bma->offset >= rec.br_startoff && | |
1450 | isnullstartblock(rec.br_startblock)); | |
1451 | return 0; | |
1452 | } | |
1da177e4 | 1453 | |
9e5987a7 DC |
1454 | /* |
1455 | * Returns the file-relative block number of the first block past eof in | |
1456 | * the file. This is not based on i_size, it is based on the extent records. | |
1457 | * Returns 0 for local files, as they do not have extent records. | |
1458 | */ | |
1459 | int | |
1460 | xfs_bmap_last_offset( | |
9e5987a7 DC |
1461 | struct xfs_inode *ip, |
1462 | xfs_fileoff_t *last_block, | |
1463 | int whichfork) | |
1464 | { | |
1465 | struct xfs_bmbt_irec rec; | |
1466 | int is_empty; | |
1467 | int error; | |
04e99455 | 1468 | |
9e5987a7 | 1469 | *last_block = 0; |
0892ccd6 | 1470 | |
9e5987a7 DC |
1471 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) |
1472 | return 0; | |
a365bdd5 | 1473 | |
9e5987a7 DC |
1474 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE && |
1475 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
2451337d | 1476 | return -EIO; |
a365bdd5 | 1477 | |
9e5987a7 DC |
1478 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty); |
1479 | if (error || is_empty) | |
a365bdd5 | 1480 | return error; |
9e5987a7 DC |
1481 | |
1482 | *last_block = rec.br_startoff + rec.br_blockcount; | |
a365bdd5 NS |
1483 | return 0; |
1484 | } | |
1485 | ||
9e5987a7 DC |
1486 | /* |
1487 | * Returns whether the selected fork of the inode has exactly one | |
1488 | * block or not. For the data fork we check this matches di_size, | |
1489 | * implying the file's range is 0..bsize-1. | |
1490 | */ | |
1491 | int /* 1=>1 block, 0=>otherwise */ | |
1492 | xfs_bmap_one_block( | |
1493 | xfs_inode_t *ip, /* incore inode */ | |
1494 | int whichfork) /* data or attr fork */ | |
c467c049 | 1495 | { |
3ba738df | 1496 | struct xfs_ifork *ifp; /* inode fork pointer */ |
9e5987a7 DC |
1497 | int rval; /* return value */ |
1498 | xfs_bmbt_irec_t s; /* internal version of extent */ | |
b2b1712a | 1499 | struct xfs_iext_cursor icur; |
c467c049 | 1500 | |
9e5987a7 DC |
1501 | #ifndef DEBUG |
1502 | if (whichfork == XFS_DATA_FORK) | |
1503 | return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize; | |
1504 | #endif /* !DEBUG */ | |
1505 | if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1) | |
1506 | return 0; | |
1507 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
1508 | return 0; | |
1509 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1510 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
b2b1712a CH |
1511 | xfs_iext_first(ifp, &icur); |
1512 | xfs_iext_get_extent(ifp, &icur, &s); | |
9e5987a7 DC |
1513 | rval = s.br_startoff == 0 && s.br_blockcount == 1; |
1514 | if (rval && whichfork == XFS_DATA_FORK) | |
1515 | ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize); | |
1516 | return rval; | |
1517 | } | |
c467c049 | 1518 | |
9e5987a7 DC |
1519 | /* |
1520 | * Extent tree manipulation functions used during allocation. | |
1521 | */ | |
c467c049 | 1522 | |
9e5987a7 DC |
1523 | /* |
1524 | * Convert a delayed allocation to a real allocation. | |
1525 | */ | |
1526 | STATIC int /* error */ | |
1527 | xfs_bmap_add_extent_delay_real( | |
60b4984f DW |
1528 | struct xfs_bmalloca *bma, |
1529 | int whichfork) | |
9e5987a7 DC |
1530 | { |
1531 | struct xfs_bmbt_irec *new = &bma->got; | |
9e5987a7 DC |
1532 | int error; /* error return value */ |
1533 | int i; /* temp state */ | |
3ba738df | 1534 | struct xfs_ifork *ifp; /* inode fork pointer */ |
9e5987a7 DC |
1535 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
1536 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
1537 | /* left is 0, right is 1, prev is 2 */ | |
1538 | int rval=0; /* return value (logging flags) */ | |
060ea65b | 1539 | int state = xfs_bmap_fork_to_state(whichfork); |
9e5987a7 DC |
1540 | xfs_filblks_t da_new; /* new count del alloc blocks used */ |
1541 | xfs_filblks_t da_old; /* old count del alloc blocks used */ | |
1542 | xfs_filblks_t temp=0; /* value for da_new calculations */ | |
9e5987a7 | 1543 | int tmp_rval; /* partial logging flags */ |
c29aad41 | 1544 | struct xfs_mount *mp; |
60b4984f | 1545 | xfs_extnum_t *nextents; |
4dcb8869 | 1546 | struct xfs_bmbt_irec old; |
c467c049 | 1547 | |
f1f96c49 | 1548 | mp = bma->ip->i_mount; |
6d3eb1ec | 1549 | ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
60b4984f DW |
1550 | ASSERT(whichfork != XFS_ATTR_FORK); |
1551 | nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents : | |
1552 | &bma->ip->i_d.di_nextents); | |
c467c049 | 1553 | |
9e5987a7 DC |
1554 | ASSERT(!isnullstartblock(new->br_startblock)); |
1555 | ASSERT(!bma->cur || | |
1556 | (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); | |
c467c049 | 1557 | |
ff6d6af2 | 1558 | XFS_STATS_INC(mp, xs_add_exlist); |
c467c049 | 1559 | |
9e5987a7 DC |
1560 | #define LEFT r[0] |
1561 | #define RIGHT r[1] | |
1562 | #define PREV r[2] | |
c467c049 CH |
1563 | |
1564 | /* | |
9e5987a7 | 1565 | * Set up a bunch of variables to make the tests simpler. |
c467c049 | 1566 | */ |
b2b1712a | 1567 | xfs_iext_get_extent(ifp, &bma->icur, &PREV); |
9e5987a7 | 1568 | new_endoff = new->br_startoff + new->br_blockcount; |
4dcb8869 | 1569 | ASSERT(isnullstartblock(PREV.br_startblock)); |
9e5987a7 DC |
1570 | ASSERT(PREV.br_startoff <= new->br_startoff); |
1571 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
1572 | ||
1573 | da_old = startblockval(PREV.br_startblock); | |
1574 | da_new = 0; | |
1575 | ||
c467c049 | 1576 | /* |
9e5987a7 DC |
1577 | * Set flags determining what part of the previous delayed allocation |
1578 | * extent is being replaced by a real allocation. | |
c467c049 | 1579 | */ |
9e5987a7 DC |
1580 | if (PREV.br_startoff == new->br_startoff) |
1581 | state |= BMAP_LEFT_FILLING; | |
1582 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
1583 | state |= BMAP_RIGHT_FILLING; | |
c467c049 CH |
1584 | |
1585 | /* | |
9e5987a7 DC |
1586 | * Check and set flags if this segment has a left neighbor. |
1587 | * Don't set contiguous if the combined extent would be too large. | |
c467c049 | 1588 | */ |
b2b1712a | 1589 | if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) { |
9e5987a7 | 1590 | state |= BMAP_LEFT_VALID; |
9e5987a7 DC |
1591 | if (isnullstartblock(LEFT.br_startblock)) |
1592 | state |= BMAP_LEFT_DELAY; | |
a365bdd5 | 1593 | } |
a365bdd5 | 1594 | |
9e5987a7 DC |
1595 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && |
1596 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
1597 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
1598 | LEFT.br_state == new->br_state && | |
1599 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
1600 | state |= BMAP_LEFT_CONTIG; | |
a365bdd5 | 1601 | |
1da177e4 | 1602 | /* |
9e5987a7 DC |
1603 | * Check and set flags if this segment has a right neighbor. |
1604 | * Don't set contiguous if the combined extent would be too large. | |
1605 | * Also check for all-three-contiguous being too large. | |
1da177e4 | 1606 | */ |
b2b1712a | 1607 | if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) { |
9e5987a7 | 1608 | state |= BMAP_RIGHT_VALID; |
9e5987a7 DC |
1609 | if (isnullstartblock(RIGHT.br_startblock)) |
1610 | state |= BMAP_RIGHT_DELAY; | |
1da177e4 | 1611 | } |
9e5987a7 DC |
1612 | |
1613 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && | |
1614 | new_endoff == RIGHT.br_startoff && | |
1615 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
1616 | new->br_state == RIGHT.br_state && | |
1617 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
1618 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1619 | BMAP_RIGHT_FILLING)) != | |
1620 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1621 | BMAP_RIGHT_FILLING) || | |
1622 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
1623 | <= MAXEXTLEN)) | |
1624 | state |= BMAP_RIGHT_CONTIG; | |
1625 | ||
1626 | error = 0; | |
1da177e4 | 1627 | /* |
9e5987a7 | 1628 | * Switch out based on the FILLING and CONTIG state bits. |
1da177e4 | 1629 | */ |
9e5987a7 DC |
1630 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | |
1631 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
1632 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1633 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1da177e4 | 1634 | /* |
9e5987a7 DC |
1635 | * Filling in all of a previously delayed allocation extent. |
1636 | * The left and right neighbors are both contiguous with new. | |
1da177e4 | 1637 | */ |
4dcb8869 | 1638 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
9e5987a7 | 1639 | |
c38ccf59 CH |
1640 | xfs_iext_remove(bma->ip, &bma->icur, state); |
1641 | xfs_iext_remove(bma->ip, &bma->icur, state); | |
b2b1712a CH |
1642 | xfs_iext_prev(ifp, &bma->icur); |
1643 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
60b4984f | 1644 | (*nextents)--; |
0d045540 | 1645 | |
9e5987a7 DC |
1646 | if (bma->cur == NULL) |
1647 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1648 | else { | |
1649 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 1650 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
9e5987a7 DC |
1651 | if (error) |
1652 | goto done; | |
c29aad41 | 1653 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1654 | error = xfs_btree_delete(bma->cur, &i); |
1655 | if (error) | |
1656 | goto done; | |
c29aad41 | 1657 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1658 | error = xfs_btree_decrement(bma->cur, 0, &i); |
1659 | if (error) | |
1660 | goto done; | |
c29aad41 | 1661 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 1662 | error = xfs_bmbt_update(bma->cur, &LEFT); |
9e5987a7 DC |
1663 | if (error) |
1664 | goto done; | |
1665 | } | |
1666 | break; | |
1667 | ||
1668 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
a365bdd5 | 1669 | /* |
9e5987a7 DC |
1670 | * Filling in all of a previously delayed allocation extent. |
1671 | * The left neighbor is contiguous, the right is not. | |
a365bdd5 | 1672 | */ |
4dcb8869 | 1673 | old = LEFT; |
4dcb8869 | 1674 | LEFT.br_blockcount += PREV.br_blockcount; |
0d045540 | 1675 | |
c38ccf59 | 1676 | xfs_iext_remove(bma->ip, &bma->icur, state); |
b2b1712a CH |
1677 | xfs_iext_prev(ifp, &bma->icur); |
1678 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
9e5987a7 | 1679 | |
9e5987a7 DC |
1680 | if (bma->cur == NULL) |
1681 | rval = XFS_ILOG_DEXT; | |
1682 | else { | |
1683 | rval = 0; | |
e16cf9b0 | 1684 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
9e5987a7 DC |
1685 | if (error) |
1686 | goto done; | |
c29aad41 | 1687 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 1688 | error = xfs_bmbt_update(bma->cur, &LEFT); |
9e5987a7 DC |
1689 | if (error) |
1690 | goto done; | |
1691 | } | |
1692 | break; | |
1693 | ||
1694 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
0937e0fd | 1695 | /* |
9e5987a7 | 1696 | * Filling in all of a previously delayed allocation extent. |
9230a0b6 DC |
1697 | * The right neighbor is contiguous, the left is not. Take care |
1698 | * with delay -> unwritten extent allocation here because the | |
1699 | * delalloc record we are overwriting is always written. | |
0937e0fd | 1700 | */ |
4dcb8869 CH |
1701 | PREV.br_startblock = new->br_startblock; |
1702 | PREV.br_blockcount += RIGHT.br_blockcount; | |
9230a0b6 | 1703 | PREV.br_state = new->br_state; |
0d045540 | 1704 | |
b2b1712a | 1705 | xfs_iext_next(ifp, &bma->icur); |
c38ccf59 | 1706 | xfs_iext_remove(bma->ip, &bma->icur, state); |
b2b1712a CH |
1707 | xfs_iext_prev(ifp, &bma->icur); |
1708 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); | |
0937e0fd | 1709 | |
9e5987a7 DC |
1710 | if (bma->cur == NULL) |
1711 | rval = XFS_ILOG_DEXT; | |
1712 | else { | |
1713 | rval = 0; | |
e16cf9b0 | 1714 | error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i); |
9e5987a7 DC |
1715 | if (error) |
1716 | goto done; | |
c29aad41 | 1717 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 1718 | error = xfs_bmbt_update(bma->cur, &PREV); |
9e5987a7 DC |
1719 | if (error) |
1720 | goto done; | |
1721 | } | |
1722 | break; | |
1723 | ||
1724 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
a365bdd5 | 1725 | /* |
9e5987a7 DC |
1726 | * Filling in all of a previously delayed allocation extent. |
1727 | * Neither the left nor right neighbors are contiguous with | |
1728 | * the new one. | |
a365bdd5 | 1729 | */ |
4dcb8869 CH |
1730 | PREV.br_startblock = new->br_startblock; |
1731 | PREV.br_state = new->br_state; | |
b2b1712a | 1732 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
a365bdd5 | 1733 | |
60b4984f | 1734 | (*nextents)++; |
9e5987a7 DC |
1735 | if (bma->cur == NULL) |
1736 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1737 | else { | |
1738 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 1739 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
9e5987a7 DC |
1740 | if (error) |
1741 | goto done; | |
c29aad41 | 1742 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
1743 | error = xfs_btree_insert(bma->cur, &i); |
1744 | if (error) | |
1745 | goto done; | |
c29aad41 | 1746 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
1747 | } |
1748 | break; | |
1da177e4 | 1749 | |
9e5987a7 | 1750 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
1da177e4 | 1751 | /* |
9e5987a7 DC |
1752 | * Filling in the first part of a previous delayed allocation. |
1753 | * The left neighbor is contiguous. | |
1da177e4 | 1754 | */ |
4dcb8869 CH |
1755 | old = LEFT; |
1756 | temp = PREV.br_blockcount - new->br_blockcount; | |
1757 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
1758 | startblockval(PREV.br_startblock)); | |
1759 | ||
4dcb8869 | 1760 | LEFT.br_blockcount += new->br_blockcount; |
1da177e4 | 1761 | |
bf99971c | 1762 | PREV.br_blockcount = temp; |
4dcb8869 CH |
1763 | PREV.br_startoff += new->br_blockcount; |
1764 | PREV.br_startblock = nullstartblock(da_new); | |
0d045540 | 1765 | |
b2b1712a CH |
1766 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
1767 | xfs_iext_prev(ifp, &bma->icur); | |
1768 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT); | |
4dcb8869 | 1769 | |
9e5987a7 DC |
1770 | if (bma->cur == NULL) |
1771 | rval = XFS_ILOG_DEXT; | |
1772 | else { | |
1773 | rval = 0; | |
e16cf9b0 | 1774 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
9e5987a7 DC |
1775 | if (error) |
1776 | goto done; | |
c29aad41 | 1777 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 1778 | error = xfs_bmbt_update(bma->cur, &LEFT); |
f3ca8738 | 1779 | if (error) |
1da177e4 | 1780 | goto done; |
1da177e4 | 1781 | } |
9e5987a7 DC |
1782 | break; |
1783 | ||
1784 | case BMAP_LEFT_FILLING: | |
1da177e4 | 1785 | /* |
9e5987a7 DC |
1786 | * Filling in the first part of a previous delayed allocation. |
1787 | * The left neighbor is not contiguous. | |
1da177e4 | 1788 | */ |
b2b1712a | 1789 | xfs_iext_update_extent(bma->ip, state, &bma->icur, new); |
60b4984f | 1790 | (*nextents)++; |
9e5987a7 DC |
1791 | if (bma->cur == NULL) |
1792 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1da177e4 | 1793 | else { |
9e5987a7 | 1794 | rval = XFS_ILOG_CORE; |
e16cf9b0 | 1795 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
9e5987a7 DC |
1796 | if (error) |
1797 | goto done; | |
c29aad41 | 1798 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
1799 | error = xfs_btree_insert(bma->cur, &i); |
1800 | if (error) | |
1da177e4 | 1801 | goto done; |
c29aad41 | 1802 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
1da177e4 | 1803 | } |
233eebb9 | 1804 | |
6d3eb1ec | 1805 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
9e5987a7 | 1806 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
280253d2 | 1807 | &bma->cur, 1, &tmp_rval, whichfork); |
9e5987a7 DC |
1808 | rval |= tmp_rval; |
1809 | if (error) | |
1810 | goto done; | |
1da177e4 | 1811 | } |
4dcb8869 CH |
1812 | |
1813 | temp = PREV.br_blockcount - new->br_blockcount; | |
9e5987a7 DC |
1814 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1815 | startblockval(PREV.br_startblock) - | |
1816 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
4dcb8869 | 1817 | |
4dcb8869 CH |
1818 | PREV.br_startoff = new_endoff; |
1819 | PREV.br_blockcount = temp; | |
1820 | PREV.br_startblock = nullstartblock(da_new); | |
b2b1712a | 1821 | xfs_iext_next(ifp, &bma->icur); |
0254c2f2 | 1822 | xfs_iext_insert(bma->ip, &bma->icur, &PREV, state); |
b2b1712a | 1823 | xfs_iext_prev(ifp, &bma->icur); |
1da177e4 LT |
1824 | break; |
1825 | ||
9e5987a7 | 1826 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
1da177e4 | 1827 | /* |
9e5987a7 DC |
1828 | * Filling in the last part of a previous delayed allocation. |
1829 | * The right neighbor is contiguous with the new allocation. | |
1da177e4 | 1830 | */ |
4dcb8869 | 1831 | old = RIGHT; |
4dcb8869 CH |
1832 | RIGHT.br_startoff = new->br_startoff; |
1833 | RIGHT.br_startblock = new->br_startblock; | |
1834 | RIGHT.br_blockcount += new->br_blockcount; | |
4dcb8869 | 1835 | |
9e5987a7 DC |
1836 | if (bma->cur == NULL) |
1837 | rval = XFS_ILOG_DEXT; | |
1838 | else { | |
1839 | rval = 0; | |
e16cf9b0 | 1840 | error = xfs_bmbt_lookup_eq(bma->cur, &old, &i); |
9e5987a7 DC |
1841 | if (error) |
1842 | goto done; | |
c29aad41 | 1843 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 1844 | error = xfs_bmbt_update(bma->cur, &RIGHT); |
9e5987a7 DC |
1845 | if (error) |
1846 | goto done; | |
1da177e4 | 1847 | } |
9e5987a7 | 1848 | |
4dcb8869 | 1849 | temp = PREV.br_blockcount - new->br_blockcount; |
9e5987a7 DC |
1850 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1851 | startblockval(PREV.br_startblock)); | |
4dcb8869 | 1852 | |
4dcb8869 CH |
1853 | PREV.br_blockcount = temp; |
1854 | PREV.br_startblock = nullstartblock(da_new); | |
9e5987a7 | 1855 | |
b2b1712a CH |
1856 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
1857 | xfs_iext_next(ifp, &bma->icur); | |
1858 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT); | |
1da177e4 LT |
1859 | break; |
1860 | ||
9e5987a7 | 1861 | case BMAP_RIGHT_FILLING: |
1da177e4 | 1862 | /* |
9e5987a7 DC |
1863 | * Filling in the last part of a previous delayed allocation. |
1864 | * The right neighbor is not contiguous. | |
1da177e4 | 1865 | */ |
b2b1712a | 1866 | xfs_iext_update_extent(bma->ip, state, &bma->icur, new); |
60b4984f | 1867 | (*nextents)++; |
9e5987a7 DC |
1868 | if (bma->cur == NULL) |
1869 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1870 | else { | |
1871 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 1872 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
9e5987a7 DC |
1873 | if (error) |
1874 | goto done; | |
c29aad41 | 1875 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
1876 | error = xfs_btree_insert(bma->cur, &i); |
1877 | if (error) | |
1878 | goto done; | |
c29aad41 | 1879 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
1da177e4 | 1880 | } |
9e5987a7 | 1881 | |
6d3eb1ec | 1882 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
9e5987a7 | 1883 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
280253d2 | 1884 | &bma->cur, 1, &tmp_rval, whichfork); |
9e5987a7 DC |
1885 | rval |= tmp_rval; |
1886 | if (error) | |
1887 | goto done; | |
1da177e4 | 1888 | } |
4dcb8869 CH |
1889 | |
1890 | temp = PREV.br_blockcount - new->br_blockcount; | |
9e5987a7 DC |
1891 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1892 | startblockval(PREV.br_startblock) - | |
1893 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
4dcb8869 | 1894 | |
4dcb8869 CH |
1895 | PREV.br_startblock = nullstartblock(da_new); |
1896 | PREV.br_blockcount = temp; | |
0254c2f2 | 1897 | xfs_iext_insert(bma->ip, &bma->icur, &PREV, state); |
b2b1712a | 1898 | xfs_iext_next(ifp, &bma->icur); |
1da177e4 LT |
1899 | break; |
1900 | ||
1901 | case 0: | |
1902 | /* | |
9e5987a7 DC |
1903 | * Filling in the middle part of a previous delayed allocation. |
1904 | * Contiguity is impossible here. | |
1905 | * This case is avoided almost all the time. | |
1906 | * | |
1907 | * We start with a delayed allocation: | |
1908 | * | |
1909 | * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+ | |
1910 | * PREV @ idx | |
1911 | * | |
1912 | * and we are allocating: | |
1913 | * +rrrrrrrrrrrrrrrrr+ | |
1914 | * new | |
1915 | * | |
1916 | * and we set it up for insertion as: | |
1917 | * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+ | |
1918 | * new | |
1919 | * PREV @ idx LEFT RIGHT | |
1920 | * inserted at idx + 1 | |
1da177e4 | 1921 | */ |
4dcb8869 CH |
1922 | old = PREV; |
1923 | ||
1924 | /* LEFT is the new middle */ | |
9e5987a7 | 1925 | LEFT = *new; |
4dcb8869 CH |
1926 | |
1927 | /* RIGHT is the new right */ | |
9e5987a7 | 1928 | RIGHT.br_state = PREV.br_state; |
9e5987a7 | 1929 | RIGHT.br_startoff = new_endoff; |
4dcb8869 CH |
1930 | RIGHT.br_blockcount = |
1931 | PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
1932 | RIGHT.br_startblock = | |
1933 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1934 | RIGHT.br_blockcount)); | |
1935 | ||
1936 | /* truncate PREV */ | |
4dcb8869 CH |
1937 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
1938 | PREV.br_startblock = | |
1939 | nullstartblock(xfs_bmap_worst_indlen(bma->ip, | |
1940 | PREV.br_blockcount)); | |
b2b1712a | 1941 | xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV); |
4dcb8869 | 1942 | |
b2b1712a | 1943 | xfs_iext_next(ifp, &bma->icur); |
0254c2f2 CH |
1944 | xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state); |
1945 | xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state); | |
60b4984f | 1946 | (*nextents)++; |
4dcb8869 | 1947 | |
9e5987a7 DC |
1948 | if (bma->cur == NULL) |
1949 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1950 | else { | |
1951 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 1952 | error = xfs_bmbt_lookup_eq(bma->cur, new, &i); |
9e5987a7 DC |
1953 | if (error) |
1954 | goto done; | |
c29aad41 | 1955 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
1956 | error = xfs_btree_insert(bma->cur, &i); |
1957 | if (error) | |
1958 | goto done; | |
c29aad41 | 1959 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
1da177e4 | 1960 | } |
1da177e4 | 1961 | |
6d3eb1ec | 1962 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
9e5987a7 | 1963 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
280253d2 | 1964 | &bma->cur, 1, &tmp_rval, whichfork); |
9e5987a7 DC |
1965 | rval |= tmp_rval; |
1966 | if (error) | |
1967 | goto done; | |
1968 | } | |
4dcb8869 CH |
1969 | |
1970 | da_new = startblockval(PREV.br_startblock) + | |
1971 | startblockval(RIGHT.br_startblock); | |
9e5987a7 DC |
1972 | break; |
1973 | ||
1974 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1975 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1976 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
1977 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1978 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
1979 | case BMAP_LEFT_CONTIG: | |
1980 | case BMAP_RIGHT_CONTIG: | |
1981 | /* | |
1982 | * These cases are all impossible. | |
1983 | */ | |
1984 | ASSERT(0); | |
1985 | } | |
1986 | ||
95eb308c DW |
1987 | /* add reverse mapping unless caller opted out */ |
1988 | if (!(bma->flags & XFS_BMAPI_NORMAP)) { | |
0f37d178 | 1989 | error = xfs_rmap_map_extent(bma->tp, bma->ip, whichfork, new); |
95eb308c DW |
1990 | if (error) |
1991 | goto done; | |
1992 | } | |
9c194644 | 1993 | |
9e5987a7 | 1994 | /* convert to a btree if necessary */ |
6d3eb1ec | 1995 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
9e5987a7 DC |
1996 | int tmp_logflags; /* partial log flag return val */ |
1997 | ||
1998 | ASSERT(bma->cur == NULL); | |
1999 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, | |
280253d2 BF |
2000 | &bma->cur, da_old > 0, &tmp_logflags, |
2001 | whichfork); | |
9e5987a7 DC |
2002 | bma->logflags |= tmp_logflags; |
2003 | if (error) | |
2004 | goto done; | |
2005 | } | |
2006 | ||
9fe82b8c DW |
2007 | if (da_new != da_old) |
2008 | xfs_mod_delalloc(mp, (int64_t)da_new - da_old); | |
2009 | ||
ca1862b0 CH |
2010 | if (bma->cur) { |
2011 | da_new += bma->cur->bc_private.b.allocated; | |
2012 | bma->cur->bc_private.b.allocated = 0; | |
96540c78 | 2013 | } |
9e5987a7 | 2014 | |
ca1862b0 CH |
2015 | /* adjust for changes in reserved delayed indirect blocks */ |
2016 | if (da_new != da_old) { | |
2017 | ASSERT(state == 0 || da_new < da_old); | |
2018 | error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), | |
2019 | false); | |
2020 | } | |
9e5987a7 | 2021 | |
6d3eb1ec | 2022 | xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork); |
1da177e4 | 2023 | done: |
60b4984f DW |
2024 | if (whichfork != XFS_COW_FORK) |
2025 | bma->logflags |= rval; | |
1da177e4 | 2026 | return error; |
9e5987a7 DC |
2027 | #undef LEFT |
2028 | #undef RIGHT | |
2029 | #undef PREV | |
1da177e4 LT |
2030 | } |
2031 | ||
2032 | /* | |
9e5987a7 | 2033 | * Convert an unwritten allocation to a real allocation or vice versa. |
1da177e4 | 2034 | */ |
26b91c72 | 2035 | int /* error */ |
9e5987a7 DC |
2036 | xfs_bmap_add_extent_unwritten_real( |
2037 | struct xfs_trans *tp, | |
2038 | xfs_inode_t *ip, /* incore inode pointer */ | |
05a630d7 | 2039 | int whichfork, |
b2b1712a | 2040 | struct xfs_iext_cursor *icur, |
9e5987a7 DC |
2041 | xfs_btree_cur_t **curp, /* if *curp is null, not a btree */ |
2042 | xfs_bmbt_irec_t *new, /* new data to add to file extents */ | |
9e5987a7 | 2043 | int *logflagsp) /* inode logging flags */ |
1da177e4 | 2044 | { |
9e5987a7 | 2045 | xfs_btree_cur_t *cur; /* btree cursor */ |
9e5987a7 DC |
2046 | int error; /* error return value */ |
2047 | int i; /* temp state */ | |
3ba738df | 2048 | struct xfs_ifork *ifp; /* inode fork pointer */ |
9e5987a7 | 2049 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
9e5987a7 DC |
2050 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ |
2051 | /* left is 0, right is 1, prev is 2 */ | |
2052 | int rval=0; /* return value (logging flags) */ | |
060ea65b | 2053 | int state = xfs_bmap_fork_to_state(whichfork); |
05a630d7 | 2054 | struct xfs_mount *mp = ip->i_mount; |
79fa6143 | 2055 | struct xfs_bmbt_irec old; |
1da177e4 | 2056 | |
9e5987a7 | 2057 | *logflagsp = 0; |
1da177e4 | 2058 | |
9e5987a7 | 2059 | cur = *curp; |
05a630d7 | 2060 | ifp = XFS_IFORK_PTR(ip, whichfork); |
8096b1eb | 2061 | |
9e5987a7 DC |
2062 | ASSERT(!isnullstartblock(new->br_startblock)); |
2063 | ||
ff6d6af2 | 2064 | XFS_STATS_INC(mp, xs_add_exlist); |
9e5987a7 DC |
2065 | |
2066 | #define LEFT r[0] | |
2067 | #define RIGHT r[1] | |
2068 | #define PREV r[2] | |
7cc95a82 | 2069 | |
1da177e4 | 2070 | /* |
9e5987a7 | 2071 | * Set up a bunch of variables to make the tests simpler. |
1da177e4 | 2072 | */ |
9e5987a7 | 2073 | error = 0; |
b2b1712a | 2074 | xfs_iext_get_extent(ifp, icur, &PREV); |
79fa6143 | 2075 | ASSERT(new->br_state != PREV.br_state); |
9e5987a7 DC |
2076 | new_endoff = new->br_startoff + new->br_blockcount; |
2077 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
2078 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
7cc95a82 | 2079 | |
1da177e4 | 2080 | /* |
9e5987a7 DC |
2081 | * Set flags determining what part of the previous oldext allocation |
2082 | * extent is being replaced by a newext allocation. | |
1da177e4 | 2083 | */ |
9e5987a7 DC |
2084 | if (PREV.br_startoff == new->br_startoff) |
2085 | state |= BMAP_LEFT_FILLING; | |
2086 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
2087 | state |= BMAP_RIGHT_FILLING; | |
2088 | ||
1da177e4 | 2089 | /* |
9e5987a7 DC |
2090 | * Check and set flags if this segment has a left neighbor. |
2091 | * Don't set contiguous if the combined extent would be too large. | |
1da177e4 | 2092 | */ |
b2b1712a | 2093 | if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) { |
9e5987a7 | 2094 | state |= BMAP_LEFT_VALID; |
9e5987a7 DC |
2095 | if (isnullstartblock(LEFT.br_startblock)) |
2096 | state |= BMAP_LEFT_DELAY; | |
1da177e4 | 2097 | } |
9e5987a7 DC |
2098 | |
2099 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2100 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
2101 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
79fa6143 | 2102 | LEFT.br_state == new->br_state && |
9e5987a7 DC |
2103 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) |
2104 | state |= BMAP_LEFT_CONTIG; | |
2105 | ||
1da177e4 | 2106 | /* |
9e5987a7 DC |
2107 | * Check and set flags if this segment has a right neighbor. |
2108 | * Don't set contiguous if the combined extent would be too large. | |
2109 | * Also check for all-three-contiguous being too large. | |
1da177e4 | 2110 | */ |
b2b1712a | 2111 | if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) { |
9e5987a7 | 2112 | state |= BMAP_RIGHT_VALID; |
9e5987a7 DC |
2113 | if (isnullstartblock(RIGHT.br_startblock)) |
2114 | state |= BMAP_RIGHT_DELAY; | |
1da177e4 | 2115 | } |
7cc95a82 | 2116 | |
9e5987a7 DC |
2117 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2118 | new_endoff == RIGHT.br_startoff && | |
2119 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
79fa6143 | 2120 | new->br_state == RIGHT.br_state && |
9e5987a7 DC |
2121 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && |
2122 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2123 | BMAP_RIGHT_FILLING)) != | |
2124 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2125 | BMAP_RIGHT_FILLING) || | |
2126 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
2127 | <= MAXEXTLEN)) | |
2128 | state |= BMAP_RIGHT_CONTIG; | |
136341b4 | 2129 | |
1da177e4 | 2130 | /* |
9e5987a7 | 2131 | * Switch out based on the FILLING and CONTIG state bits. |
1da177e4 | 2132 | */ |
9e5987a7 DC |
2133 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | |
2134 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
2135 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2136 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
2137 | /* | |
2138 | * Setting all of a previous oldext extent to newext. | |
2139 | * The left and right neighbors are both contiguous with new. | |
2140 | */ | |
79fa6143 | 2141 | LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount; |
1a5902c5 | 2142 | |
c38ccf59 CH |
2143 | xfs_iext_remove(ip, icur, state); |
2144 | xfs_iext_remove(ip, icur, state); | |
b2b1712a CH |
2145 | xfs_iext_prev(ifp, icur); |
2146 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
05a630d7 DW |
2147 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2148 | XFS_IFORK_NEXTENTS(ip, whichfork) - 2); | |
9e5987a7 DC |
2149 | if (cur == NULL) |
2150 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2151 | else { | |
2152 | rval = XFS_ILOG_CORE; | |
e16cf9b0 CH |
2153 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2154 | if (error) | |
9e5987a7 | 2155 | goto done; |
c29aad41 | 2156 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2157 | if ((error = xfs_btree_delete(cur, &i))) |
2158 | goto done; | |
c29aad41 | 2159 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2160 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2161 | goto done; | |
c29aad41 | 2162 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2163 | if ((error = xfs_btree_delete(cur, &i))) |
2164 | goto done; | |
c29aad41 | 2165 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2166 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2167 | goto done; | |
c29aad41 | 2168 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2169 | error = xfs_bmbt_update(cur, &LEFT); |
79fa6143 | 2170 | if (error) |
9e5987a7 DC |
2171 | goto done; |
2172 | } | |
2173 | break; | |
1a5902c5 | 2174 | |
9e5987a7 DC |
2175 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: |
2176 | /* | |
2177 | * Setting all of a previous oldext extent to newext. | |
2178 | * The left neighbor is contiguous, the right is not. | |
2179 | */ | |
79fa6143 | 2180 | LEFT.br_blockcount += PREV.br_blockcount; |
1da177e4 | 2181 | |
c38ccf59 | 2182 | xfs_iext_remove(ip, icur, state); |
b2b1712a CH |
2183 | xfs_iext_prev(ifp, icur); |
2184 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
05a630d7 DW |
2185 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2186 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
9e5987a7 DC |
2187 | if (cur == NULL) |
2188 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2189 | else { | |
2190 | rval = XFS_ILOG_CORE; | |
e16cf9b0 CH |
2191 | error = xfs_bmbt_lookup_eq(cur, &PREV, &i); |
2192 | if (error) | |
9e5987a7 | 2193 | goto done; |
c29aad41 | 2194 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2195 | if ((error = xfs_btree_delete(cur, &i))) |
2196 | goto done; | |
c29aad41 | 2197 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2198 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2199 | goto done; | |
c29aad41 | 2200 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2201 | error = xfs_bmbt_update(cur, &LEFT); |
79fa6143 | 2202 | if (error) |
9e5987a7 DC |
2203 | goto done; |
2204 | } | |
2205 | break; | |
1da177e4 | 2206 | |
9e5987a7 | 2207 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
1da177e4 | 2208 | /* |
9e5987a7 DC |
2209 | * Setting all of a previous oldext extent to newext. |
2210 | * The right neighbor is contiguous, the left is not. | |
1da177e4 | 2211 | */ |
79fa6143 CH |
2212 | PREV.br_blockcount += RIGHT.br_blockcount; |
2213 | PREV.br_state = new->br_state; | |
a6818477 | 2214 | |
b2b1712a | 2215 | xfs_iext_next(ifp, icur); |
c38ccf59 | 2216 | xfs_iext_remove(ip, icur, state); |
b2b1712a CH |
2217 | xfs_iext_prev(ifp, icur); |
2218 | xfs_iext_update_extent(ip, state, icur, &PREV); | |
79fa6143 | 2219 | |
05a630d7 DW |
2220 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2221 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
9e5987a7 DC |
2222 | if (cur == NULL) |
2223 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2224 | else { | |
2225 | rval = XFS_ILOG_CORE; | |
e16cf9b0 CH |
2226 | error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i); |
2227 | if (error) | |
9e5987a7 | 2228 | goto done; |
c29aad41 | 2229 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2230 | if ((error = xfs_btree_delete(cur, &i))) |
2231 | goto done; | |
c29aad41 | 2232 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2233 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2234 | goto done; | |
c29aad41 | 2235 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2236 | error = xfs_bmbt_update(cur, &PREV); |
79fa6143 | 2237 | if (error) |
9e5987a7 | 2238 | goto done; |
1da177e4 | 2239 | } |
9e5987a7 | 2240 | break; |
1e82379b | 2241 | |
9e5987a7 DC |
2242 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: |
2243 | /* | |
2244 | * Setting all of a previous oldext extent to newext. | |
2245 | * Neither the left nor right neighbors are contiguous with | |
2246 | * the new one. | |
2247 | */ | |
79fa6143 | 2248 | PREV.br_state = new->br_state; |
b2b1712a | 2249 | xfs_iext_update_extent(ip, state, icur, &PREV); |
1e82379b | 2250 | |
9e5987a7 DC |
2251 | if (cur == NULL) |
2252 | rval = XFS_ILOG_DEXT; | |
2253 | else { | |
2254 | rval = 0; | |
e16cf9b0 CH |
2255 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2256 | if (error) | |
9e5987a7 | 2257 | goto done; |
c29aad41 | 2258 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2259 | error = xfs_bmbt_update(cur, &PREV); |
79fa6143 | 2260 | if (error) |
9e5987a7 DC |
2261 | goto done; |
2262 | } | |
2263 | break; | |
1e82379b | 2264 | |
9e5987a7 DC |
2265 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
2266 | /* | |
2267 | * Setting the first part of a previous oldext extent to newext. | |
2268 | * The left neighbor is contiguous. | |
2269 | */ | |
79fa6143 | 2270 | LEFT.br_blockcount += new->br_blockcount; |
1da177e4 | 2271 | |
79fa6143 | 2272 | old = PREV; |
79fa6143 CH |
2273 | PREV.br_startoff += new->br_blockcount; |
2274 | PREV.br_startblock += new->br_blockcount; | |
2275 | PREV.br_blockcount -= new->br_blockcount; | |
0293ce3a | 2276 | |
b2b1712a CH |
2277 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2278 | xfs_iext_prev(ifp, icur); | |
2279 | xfs_iext_update_extent(ip, state, icur, &LEFT); | |
8867bc9b | 2280 | |
9e5987a7 DC |
2281 | if (cur == NULL) |
2282 | rval = XFS_ILOG_DEXT; | |
2283 | else { | |
2284 | rval = 0; | |
e16cf9b0 | 2285 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
79fa6143 | 2286 | if (error) |
9e5987a7 | 2287 | goto done; |
c29aad41 | 2288 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2289 | error = xfs_bmbt_update(cur, &PREV); |
79fa6143 | 2290 | if (error) |
9e5987a7 | 2291 | goto done; |
79fa6143 CH |
2292 | error = xfs_btree_decrement(cur, 0, &i); |
2293 | if (error) | |
9e5987a7 | 2294 | goto done; |
a67d00a5 | 2295 | error = xfs_bmbt_update(cur, &LEFT); |
9e5987a7 DC |
2296 | if (error) |
2297 | goto done; | |
8867bc9b | 2298 | } |
9e5987a7 | 2299 | break; |
0293ce3a | 2300 | |
9e5987a7 DC |
2301 | case BMAP_LEFT_FILLING: |
2302 | /* | |
2303 | * Setting the first part of a previous oldext extent to newext. | |
2304 | * The left neighbor is not contiguous. | |
2305 | */ | |
79fa6143 | 2306 | old = PREV; |
79fa6143 CH |
2307 | PREV.br_startoff += new->br_blockcount; |
2308 | PREV.br_startblock += new->br_blockcount; | |
2309 | PREV.br_blockcount -= new->br_blockcount; | |
1da177e4 | 2310 | |
b2b1712a | 2311 | xfs_iext_update_extent(ip, state, icur, &PREV); |
0254c2f2 | 2312 | xfs_iext_insert(ip, icur, new, state); |
05a630d7 DW |
2313 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2314 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
9e5987a7 DC |
2315 | if (cur == NULL) |
2316 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2317 | else { | |
2318 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 2319 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
79fa6143 | 2320 | if (error) |
9e5987a7 | 2321 | goto done; |
c29aad41 | 2322 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2323 | error = xfs_bmbt_update(cur, &PREV); |
79fa6143 | 2324 | if (error) |
9e5987a7 DC |
2325 | goto done; |
2326 | cur->bc_rec.b = *new; | |
2327 | if ((error = xfs_btree_insert(cur, &i))) | |
2328 | goto done; | |
c29aad41 | 2329 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2330 | } |
2331 | break; | |
1da177e4 | 2332 | |
9e5987a7 DC |
2333 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2334 | /* | |
2335 | * Setting the last part of a previous oldext extent to newext. | |
2336 | * The right neighbor is contiguous with the new allocation. | |
2337 | */ | |
79fa6143 | 2338 | old = PREV; |
79fa6143 | 2339 | PREV.br_blockcount -= new->br_blockcount; |
1da177e4 | 2340 | |
79fa6143 CH |
2341 | RIGHT.br_startoff = new->br_startoff; |
2342 | RIGHT.br_startblock = new->br_startblock; | |
2343 | RIGHT.br_blockcount += new->br_blockcount; | |
a6818477 | 2344 | |
b2b1712a CH |
2345 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2346 | xfs_iext_next(ifp, icur); | |
2347 | xfs_iext_update_extent(ip, state, icur, &RIGHT); | |
1da177e4 | 2348 | |
9e5987a7 DC |
2349 | if (cur == NULL) |
2350 | rval = XFS_ILOG_DEXT; | |
2351 | else { | |
2352 | rval = 0; | |
e16cf9b0 | 2353 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
79fa6143 | 2354 | if (error) |
9e5987a7 | 2355 | goto done; |
c29aad41 | 2356 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2357 | error = xfs_bmbt_update(cur, &PREV); |
79fa6143 | 2358 | if (error) |
9e5987a7 | 2359 | goto done; |
79fa6143 CH |
2360 | error = xfs_btree_increment(cur, 0, &i); |
2361 | if (error) | |
9e5987a7 | 2362 | goto done; |
a67d00a5 | 2363 | error = xfs_bmbt_update(cur, &RIGHT); |
79fa6143 | 2364 | if (error) |
9e5987a7 DC |
2365 | goto done; |
2366 | } | |
2367 | break; | |
1da177e4 | 2368 | |
9e5987a7 DC |
2369 | case BMAP_RIGHT_FILLING: |
2370 | /* | |
2371 | * Setting the last part of a previous oldext extent to newext. | |
2372 | * The right neighbor is not contiguous. | |
2373 | */ | |
79fa6143 | 2374 | old = PREV; |
79fa6143 | 2375 | PREV.br_blockcount -= new->br_blockcount; |
1da177e4 | 2376 | |
b2b1712a CH |
2377 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2378 | xfs_iext_next(ifp, icur); | |
0254c2f2 | 2379 | xfs_iext_insert(ip, icur, new, state); |
d8cc890d | 2380 | |
05a630d7 DW |
2381 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2382 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
9e5987a7 DC |
2383 | if (cur == NULL) |
2384 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2385 | else { | |
2386 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 2387 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
79fa6143 | 2388 | if (error) |
9e5987a7 | 2389 | goto done; |
c29aad41 | 2390 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2391 | error = xfs_bmbt_update(cur, &PREV); |
79fa6143 | 2392 | if (error) |
9e5987a7 | 2393 | goto done; |
e16cf9b0 CH |
2394 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2395 | if (error) | |
9e5987a7 | 2396 | goto done; |
c29aad41 | 2397 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 DC |
2398 | if ((error = xfs_btree_insert(cur, &i))) |
2399 | goto done; | |
c29aad41 | 2400 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2401 | } |
2402 | break; | |
2403 | ||
2404 | case 0: | |
1da177e4 | 2405 | /* |
9e5987a7 DC |
2406 | * Setting the middle part of a previous oldext extent to |
2407 | * newext. Contiguity is impossible here. | |
2408 | * One extent becomes three extents. | |
1da177e4 | 2409 | */ |
79fa6143 | 2410 | old = PREV; |
79fa6143 | 2411 | PREV.br_blockcount = new->br_startoff - PREV.br_startoff; |
898621d5 | 2412 | |
9e5987a7 DC |
2413 | r[0] = *new; |
2414 | r[1].br_startoff = new_endoff; | |
2415 | r[1].br_blockcount = | |
79fa6143 | 2416 | old.br_startoff + old.br_blockcount - new_endoff; |
9e5987a7 | 2417 | r[1].br_startblock = new->br_startblock + new->br_blockcount; |
79fa6143 | 2418 | r[1].br_state = PREV.br_state; |
898621d5 | 2419 | |
b2b1712a CH |
2420 | xfs_iext_update_extent(ip, state, icur, &PREV); |
2421 | xfs_iext_next(ifp, icur); | |
0254c2f2 CH |
2422 | xfs_iext_insert(ip, icur, &r[1], state); |
2423 | xfs_iext_insert(ip, icur, &r[0], state); | |
9e5987a7 | 2424 | |
05a630d7 DW |
2425 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2426 | XFS_IFORK_NEXTENTS(ip, whichfork) + 2); | |
9e5987a7 DC |
2427 | if (cur == NULL) |
2428 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2429 | else { | |
2430 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 2431 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
79fa6143 | 2432 | if (error) |
9e5987a7 | 2433 | goto done; |
c29aad41 | 2434 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 | 2435 | /* new right extent - oldext */ |
a67d00a5 CH |
2436 | error = xfs_bmbt_update(cur, &r[1]); |
2437 | if (error) | |
9e5987a7 DC |
2438 | goto done; |
2439 | /* new left extent - oldext */ | |
2440 | cur->bc_rec.b = PREV; | |
9e5987a7 DC |
2441 | if ((error = xfs_btree_insert(cur, &i))) |
2442 | goto done; | |
c29aad41 | 2443 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2444 | /* |
2445 | * Reset the cursor to the position of the new extent | |
2446 | * we are about to insert as we can't trust it after | |
2447 | * the previous insert. | |
2448 | */ | |
e16cf9b0 CH |
2449 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
2450 | if (error) | |
9e5987a7 | 2451 | goto done; |
c29aad41 | 2452 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
9e5987a7 | 2453 | /* new middle extent - newext */ |
9e5987a7 DC |
2454 | if ((error = xfs_btree_insert(cur, &i))) |
2455 | goto done; | |
c29aad41 | 2456 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 | 2457 | } |
1da177e4 | 2458 | break; |
9e5987a7 DC |
2459 | |
2460 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2461 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2462 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2463 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2464 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2465 | case BMAP_LEFT_CONTIG: | |
2466 | case BMAP_RIGHT_CONTIG: | |
2467 | /* | |
2468 | * These cases are all impossible. | |
2469 | */ | |
1da177e4 | 2470 | ASSERT(0); |
1da177e4 | 2471 | } |
8096b1eb | 2472 | |
9c194644 | 2473 | /* update reverse mappings */ |
0f37d178 | 2474 | error = xfs_rmap_convert_extent(mp, tp, ip, whichfork, new); |
9c194644 DW |
2475 | if (error) |
2476 | goto done; | |
2477 | ||
9e5987a7 | 2478 | /* convert to a btree if necessary */ |
05a630d7 | 2479 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
9e5987a7 DC |
2480 | int tmp_logflags; /* partial log flag return val */ |
2481 | ||
2482 | ASSERT(cur == NULL); | |
280253d2 BF |
2483 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
2484 | &tmp_logflags, whichfork); | |
9e5987a7 DC |
2485 | *logflagsp |= tmp_logflags; |
2486 | if (error) | |
2487 | goto done; | |
1da177e4 | 2488 | } |
da087bad | 2489 | |
9e5987a7 DC |
2490 | /* clear out the allocated field, done with it now in any case. */ |
2491 | if (cur) { | |
2492 | cur->bc_private.b.allocated = 0; | |
2493 | *curp = cur; | |
1da177e4 | 2494 | } |
8096b1eb | 2495 | |
05a630d7 | 2496 | xfs_bmap_check_leaf_extents(*curp, ip, whichfork); |
9e5987a7 DC |
2497 | done: |
2498 | *logflagsp |= rval; | |
1da177e4 | 2499 | return error; |
9e5987a7 DC |
2500 | #undef LEFT |
2501 | #undef RIGHT | |
2502 | #undef PREV | |
1da177e4 LT |
2503 | } |
2504 | ||
2505 | /* | |
9e5987a7 | 2506 | * Convert a hole to a delayed allocation. |
1da177e4 | 2507 | */ |
9e5987a7 DC |
2508 | STATIC void |
2509 | xfs_bmap_add_extent_hole_delay( | |
2510 | xfs_inode_t *ip, /* incore inode pointer */ | |
be51f811 | 2511 | int whichfork, |
b2b1712a | 2512 | struct xfs_iext_cursor *icur, |
9e5987a7 | 2513 | xfs_bmbt_irec_t *new) /* new data to add to file extents */ |
1da177e4 | 2514 | { |
3ba738df | 2515 | struct xfs_ifork *ifp; /* inode fork pointer */ |
9e5987a7 DC |
2516 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2517 | xfs_filblks_t newlen=0; /* new indirect size */ | |
2518 | xfs_filblks_t oldlen=0; /* old indirect size */ | |
2519 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
060ea65b | 2520 | int state = xfs_bmap_fork_to_state(whichfork); |
3ffc18ec | 2521 | xfs_filblks_t temp; /* temp for indirect calculations */ |
1da177e4 | 2522 | |
be51f811 | 2523 | ifp = XFS_IFORK_PTR(ip, whichfork); |
9e5987a7 | 2524 | ASSERT(isnullstartblock(new->br_startblock)); |
1da177e4 LT |
2525 | |
2526 | /* | |
9e5987a7 | 2527 | * Check and set flags if this segment has a left neighbor |
1da177e4 | 2528 | */ |
b2b1712a | 2529 | if (xfs_iext_peek_prev_extent(ifp, icur, &left)) { |
9e5987a7 | 2530 | state |= BMAP_LEFT_VALID; |
9e5987a7 DC |
2531 | if (isnullstartblock(left.br_startblock)) |
2532 | state |= BMAP_LEFT_DELAY; | |
1da177e4 | 2533 | } |
1da177e4 | 2534 | |
9e5987a7 DC |
2535 | /* |
2536 | * Check and set flags if the current (right) segment exists. | |
2537 | * If it doesn't exist, we're converting the hole at end-of-file. | |
2538 | */ | |
b2b1712a | 2539 | if (xfs_iext_get_extent(ifp, icur, &right)) { |
9e5987a7 | 2540 | state |= BMAP_RIGHT_VALID; |
9e5987a7 DC |
2541 | if (isnullstartblock(right.br_startblock)) |
2542 | state |= BMAP_RIGHT_DELAY; | |
1da177e4 | 2543 | } |
9e5987a7 | 2544 | |
1da177e4 | 2545 | /* |
9e5987a7 DC |
2546 | * Set contiguity flags on the left and right neighbors. |
2547 | * Don't let extents get too large, even if the pieces are contiguous. | |
1da177e4 | 2548 | */ |
9e5987a7 DC |
2549 | if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) && |
2550 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2551 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2552 | state |= BMAP_LEFT_CONTIG; | |
2553 | ||
2554 | if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) && | |
2555 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2556 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2557 | (!(state & BMAP_LEFT_CONTIG) || | |
2558 | (left.br_blockcount + new->br_blockcount + | |
2559 | right.br_blockcount <= MAXEXTLEN))) | |
2560 | state |= BMAP_RIGHT_CONTIG; | |
cc09c0dc DC |
2561 | |
2562 | /* | |
9e5987a7 | 2563 | * Switch out based on the contiguity flags. |
cc09c0dc | 2564 | */ |
9e5987a7 DC |
2565 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { |
2566 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2567 | /* | |
2568 | * New allocation is contiguous with delayed allocations | |
2569 | * on the left and on the right. | |
2570 | * Merge all three into a single extent record. | |
2571 | */ | |
9e5987a7 DC |
2572 | temp = left.br_blockcount + new->br_blockcount + |
2573 | right.br_blockcount; | |
cc09c0dc | 2574 | |
9e5987a7 DC |
2575 | oldlen = startblockval(left.br_startblock) + |
2576 | startblockval(new->br_startblock) + | |
2577 | startblockval(right.br_startblock); | |
0e339ef8 BF |
2578 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2579 | oldlen); | |
3ffc18ec CH |
2580 | left.br_startblock = nullstartblock(newlen); |
2581 | left.br_blockcount = temp; | |
1da177e4 | 2582 | |
c38ccf59 | 2583 | xfs_iext_remove(ip, icur, state); |
b2b1712a CH |
2584 | xfs_iext_prev(ifp, icur); |
2585 | xfs_iext_update_extent(ip, state, icur, &left); | |
9e5987a7 | 2586 | break; |
1da177e4 | 2587 | |
9e5987a7 DC |
2588 | case BMAP_LEFT_CONTIG: |
2589 | /* | |
2590 | * New allocation is contiguous with a delayed allocation | |
2591 | * on the left. | |
2592 | * Merge the new allocation with the left neighbor. | |
2593 | */ | |
9e5987a7 | 2594 | temp = left.br_blockcount + new->br_blockcount; |
1da177e4 | 2595 | |
9e5987a7 DC |
2596 | oldlen = startblockval(left.br_startblock) + |
2597 | startblockval(new->br_startblock); | |
0e339ef8 BF |
2598 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2599 | oldlen); | |
3ffc18ec CH |
2600 | left.br_blockcount = temp; |
2601 | left.br_startblock = nullstartblock(newlen); | |
41d196f4 | 2602 | |
b2b1712a CH |
2603 | xfs_iext_prev(ifp, icur); |
2604 | xfs_iext_update_extent(ip, state, icur, &left); | |
9e5987a7 | 2605 | break; |
1da177e4 | 2606 | |
9e5987a7 DC |
2607 | case BMAP_RIGHT_CONTIG: |
2608 | /* | |
2609 | * New allocation is contiguous with a delayed allocation | |
2610 | * on the right. | |
2611 | * Merge the new allocation with the right neighbor. | |
2612 | */ | |
9e5987a7 DC |
2613 | temp = new->br_blockcount + right.br_blockcount; |
2614 | oldlen = startblockval(new->br_startblock) + | |
2615 | startblockval(right.br_startblock); | |
0e339ef8 BF |
2616 | newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), |
2617 | oldlen); | |
3ffc18ec CH |
2618 | right.br_startoff = new->br_startoff; |
2619 | right.br_startblock = nullstartblock(newlen); | |
2620 | right.br_blockcount = temp; | |
b2b1712a | 2621 | xfs_iext_update_extent(ip, state, icur, &right); |
9e5987a7 DC |
2622 | break; |
2623 | ||
2624 | case 0: | |
2625 | /* | |
2626 | * New allocation is not contiguous with another | |
2627 | * delayed allocation. | |
2628 | * Insert a new entry. | |
2629 | */ | |
2630 | oldlen = newlen = 0; | |
0254c2f2 | 2631 | xfs_iext_insert(ip, icur, new, state); |
9e5987a7 | 2632 | break; |
1da177e4 | 2633 | } |
9e5987a7 DC |
2634 | if (oldlen != newlen) { |
2635 | ASSERT(oldlen > newlen); | |
0d485ada DC |
2636 | xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen), |
2637 | false); | |
1da177e4 | 2638 | /* |
9e5987a7 | 2639 | * Nothing to do for disk quota accounting here. |
1da177e4 | 2640 | */ |
9fe82b8c | 2641 | xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen); |
1da177e4 | 2642 | } |
1da177e4 LT |
2643 | } |
2644 | ||
2645 | /* | |
9e5987a7 | 2646 | * Convert a hole to a real allocation. |
1da177e4 | 2647 | */ |
9e5987a7 DC |
2648 | STATIC int /* error */ |
2649 | xfs_bmap_add_extent_hole_real( | |
6d04558f CH |
2650 | struct xfs_trans *tp, |
2651 | struct xfs_inode *ip, | |
2652 | int whichfork, | |
b2b1712a | 2653 | struct xfs_iext_cursor *icur, |
6d04558f CH |
2654 | struct xfs_btree_cur **curp, |
2655 | struct xfs_bmbt_irec *new, | |
95eb308c DW |
2656 | int *logflagsp, |
2657 | int flags) | |
27a3f8f2 | 2658 | { |
6d04558f CH |
2659 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
2660 | struct xfs_mount *mp = ip->i_mount; | |
2661 | struct xfs_btree_cur *cur = *curp; | |
9e5987a7 DC |
2662 | int error; /* error return value */ |
2663 | int i; /* temp state */ | |
9e5987a7 DC |
2664 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ |
2665 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2666 | int rval=0; /* return value (logging flags) */ | |
060ea65b | 2667 | int state = xfs_bmap_fork_to_state(whichfork); |
1abb9e55 | 2668 | struct xfs_bmbt_irec old; |
27a3f8f2 | 2669 | |
9e5987a7 | 2670 | ASSERT(!isnullstartblock(new->br_startblock)); |
6d04558f | 2671 | ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); |
27a3f8f2 | 2672 | |
ff6d6af2 | 2673 | XFS_STATS_INC(mp, xs_add_exlist); |
27a3f8f2 | 2674 | |
9e5987a7 DC |
2675 | /* |
2676 | * Check and set flags if this segment has a left neighbor. | |
2677 | */ | |
b2b1712a | 2678 | if (xfs_iext_peek_prev_extent(ifp, icur, &left)) { |
9e5987a7 | 2679 | state |= BMAP_LEFT_VALID; |
9e5987a7 DC |
2680 | if (isnullstartblock(left.br_startblock)) |
2681 | state |= BMAP_LEFT_DELAY; | |
2682 | } | |
27a3f8f2 CH |
2683 | |
2684 | /* | |
9e5987a7 DC |
2685 | * Check and set flags if this segment has a current value. |
2686 | * Not true if we're inserting into the "hole" at eof. | |
27a3f8f2 | 2687 | */ |
b2b1712a | 2688 | if (xfs_iext_get_extent(ifp, icur, &right)) { |
9e5987a7 | 2689 | state |= BMAP_RIGHT_VALID; |
9e5987a7 DC |
2690 | if (isnullstartblock(right.br_startblock)) |
2691 | state |= BMAP_RIGHT_DELAY; | |
2692 | } | |
27a3f8f2 | 2693 | |
9e5987a7 DC |
2694 | /* |
2695 | * We're inserting a real allocation between "left" and "right". | |
2696 | * Set the contiguity flags. Don't let extents get too large. | |
2697 | */ | |
2698 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2699 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2700 | left.br_startblock + left.br_blockcount == new->br_startblock && | |
2701 | left.br_state == new->br_state && | |
2702 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2703 | state |= BMAP_LEFT_CONTIG; | |
27a3f8f2 | 2704 | |
9e5987a7 DC |
2705 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2706 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2707 | new->br_startblock + new->br_blockcount == right.br_startblock && | |
2708 | new->br_state == right.br_state && | |
2709 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2710 | (!(state & BMAP_LEFT_CONTIG) || | |
2711 | left.br_blockcount + new->br_blockcount + | |
2712 | right.br_blockcount <= MAXEXTLEN)) | |
2713 | state |= BMAP_RIGHT_CONTIG; | |
27a3f8f2 | 2714 | |
9e5987a7 DC |
2715 | error = 0; |
2716 | /* | |
2717 | * Select which case we're in here, and implement it. | |
2718 | */ | |
2719 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2720 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2721 | /* | |
2722 | * New allocation is contiguous with real allocations on the | |
2723 | * left and on the right. | |
2724 | * Merge all three into a single extent record. | |
2725 | */ | |
1abb9e55 | 2726 | left.br_blockcount += new->br_blockcount + right.br_blockcount; |
27a3f8f2 | 2727 | |
c38ccf59 | 2728 | xfs_iext_remove(ip, icur, state); |
b2b1712a CH |
2729 | xfs_iext_prev(ifp, icur); |
2730 | xfs_iext_update_extent(ip, state, icur, &left); | |
27a3f8f2 | 2731 | |
6d04558f CH |
2732 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2733 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
2734 | if (cur == NULL) { | |
9e5987a7 DC |
2735 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2736 | } else { | |
2737 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 2738 | error = xfs_bmbt_lookup_eq(cur, &right, &i); |
9e5987a7 DC |
2739 | if (error) |
2740 | goto done; | |
c29aad41 | 2741 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
6d04558f | 2742 | error = xfs_btree_delete(cur, &i); |
9e5987a7 DC |
2743 | if (error) |
2744 | goto done; | |
c29aad41 | 2745 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
6d04558f | 2746 | error = xfs_btree_decrement(cur, 0, &i); |
9e5987a7 DC |
2747 | if (error) |
2748 | goto done; | |
c29aad41 | 2749 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2750 | error = xfs_bmbt_update(cur, &left); |
9e5987a7 DC |
2751 | if (error) |
2752 | goto done; | |
2753 | } | |
2754 | break; | |
27a3f8f2 | 2755 | |
9e5987a7 DC |
2756 | case BMAP_LEFT_CONTIG: |
2757 | /* | |
2758 | * New allocation is contiguous with a real allocation | |
2759 | * on the left. | |
2760 | * Merge the new allocation with the left neighbor. | |
2761 | */ | |
1abb9e55 | 2762 | old = left; |
1abb9e55 | 2763 | left.br_blockcount += new->br_blockcount; |
1d2e0089 | 2764 | |
b2b1712a CH |
2765 | xfs_iext_prev(ifp, icur); |
2766 | xfs_iext_update_extent(ip, state, icur, &left); | |
1da177e4 | 2767 | |
6d04558f | 2768 | if (cur == NULL) { |
9e5987a7 DC |
2769 | rval = xfs_ilog_fext(whichfork); |
2770 | } else { | |
2771 | rval = 0; | |
e16cf9b0 | 2772 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
9e5987a7 DC |
2773 | if (error) |
2774 | goto done; | |
c29aad41 | 2775 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2776 | error = xfs_bmbt_update(cur, &left); |
9e5987a7 DC |
2777 | if (error) |
2778 | goto done; | |
2779 | } | |
2780 | break; | |
27a3f8f2 | 2781 | |
9e5987a7 DC |
2782 | case BMAP_RIGHT_CONTIG: |
2783 | /* | |
2784 | * New allocation is contiguous with a real allocation | |
2785 | * on the right. | |
2786 | * Merge the new allocation with the right neighbor. | |
2787 | */ | |
1abb9e55 | 2788 | old = right; |
ca5d8e5b | 2789 | |
1abb9e55 CH |
2790 | right.br_startoff = new->br_startoff; |
2791 | right.br_startblock = new->br_startblock; | |
2792 | right.br_blockcount += new->br_blockcount; | |
b2b1712a | 2793 | xfs_iext_update_extent(ip, state, icur, &right); |
27a3f8f2 | 2794 | |
6d04558f | 2795 | if (cur == NULL) { |
9e5987a7 DC |
2796 | rval = xfs_ilog_fext(whichfork); |
2797 | } else { | |
2798 | rval = 0; | |
e16cf9b0 | 2799 | error = xfs_bmbt_lookup_eq(cur, &old, &i); |
9e5987a7 DC |
2800 | if (error) |
2801 | goto done; | |
c29aad41 | 2802 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
a67d00a5 | 2803 | error = xfs_bmbt_update(cur, &right); |
9e5987a7 DC |
2804 | if (error) |
2805 | goto done; | |
2806 | } | |
2807 | break; | |
2808 | ||
2809 | case 0: | |
2810 | /* | |
2811 | * New allocation is not contiguous with another | |
2812 | * real allocation. | |
2813 | * Insert a new entry. | |
2814 | */ | |
0254c2f2 | 2815 | xfs_iext_insert(ip, icur, new, state); |
6d04558f CH |
2816 | XFS_IFORK_NEXT_SET(ip, whichfork, |
2817 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
2818 | if (cur == NULL) { | |
9e5987a7 DC |
2819 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); |
2820 | } else { | |
2821 | rval = XFS_ILOG_CORE; | |
e16cf9b0 | 2822 | error = xfs_bmbt_lookup_eq(cur, new, &i); |
9e5987a7 DC |
2823 | if (error) |
2824 | goto done; | |
c29aad41 | 2825 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
6d04558f | 2826 | error = xfs_btree_insert(cur, &i); |
9e5987a7 DC |
2827 | if (error) |
2828 | goto done; | |
c29aad41 | 2829 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 DC |
2830 | } |
2831 | break; | |
2832 | } | |
2833 | ||
95eb308c DW |
2834 | /* add reverse mapping unless caller opted out */ |
2835 | if (!(flags & XFS_BMAPI_NORMAP)) { | |
0f37d178 | 2836 | error = xfs_rmap_map_extent(tp, ip, whichfork, new); |
95eb308c DW |
2837 | if (error) |
2838 | goto done; | |
2839 | } | |
9c194644 | 2840 | |
9e5987a7 | 2841 | /* convert to a btree if necessary */ |
6d04558f | 2842 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
9e5987a7 DC |
2843 | int tmp_logflags; /* partial log flag return val */ |
2844 | ||
6d04558f | 2845 | ASSERT(cur == NULL); |
280253d2 BF |
2846 | error = xfs_bmap_extents_to_btree(tp, ip, curp, 0, |
2847 | &tmp_logflags, whichfork); | |
6d04558f CH |
2848 | *logflagsp |= tmp_logflags; |
2849 | cur = *curp; | |
9e5987a7 DC |
2850 | if (error) |
2851 | goto done; | |
2852 | } | |
2853 | ||
2854 | /* clear out the allocated field, done with it now in any case. */ | |
6d04558f CH |
2855 | if (cur) |
2856 | cur->bc_private.b.allocated = 0; | |
9e5987a7 | 2857 | |
6d04558f | 2858 | xfs_bmap_check_leaf_extents(cur, ip, whichfork); |
9e5987a7 | 2859 | done: |
6d04558f | 2860 | *logflagsp |= rval; |
9e5987a7 | 2861 | return error; |
1da177e4 LT |
2862 | } |
2863 | ||
2864 | /* | |
9e5987a7 | 2865 | * Functions used in the extent read, allocate and remove paths |
1da177e4 | 2866 | */ |
1da177e4 | 2867 | |
9e5987a7 DC |
2868 | /* |
2869 | * Adjust the size of the new extent based on di_extsize and rt extsize. | |
2870 | */ | |
68988114 | 2871 | int |
9e5987a7 DC |
2872 | xfs_bmap_extsize_align( |
2873 | xfs_mount_t *mp, | |
2874 | xfs_bmbt_irec_t *gotp, /* next extent pointer */ | |
2875 | xfs_bmbt_irec_t *prevp, /* previous extent pointer */ | |
2876 | xfs_extlen_t extsz, /* align to this extent size */ | |
2877 | int rt, /* is this a realtime inode? */ | |
2878 | int eof, /* is extent at end-of-file? */ | |
2879 | int delay, /* creating delalloc extent? */ | |
2880 | int convert, /* overwriting unwritten extent? */ | |
2881 | xfs_fileoff_t *offp, /* in/out: aligned offset */ | |
2882 | xfs_extlen_t *lenp) /* in/out: aligned length */ | |
4e8938fe | 2883 | { |
9e5987a7 DC |
2884 | xfs_fileoff_t orig_off; /* original offset */ |
2885 | xfs_extlen_t orig_alen; /* original length */ | |
2886 | xfs_fileoff_t orig_end; /* original off+len */ | |
2887 | xfs_fileoff_t nexto; /* next file offset */ | |
2888 | xfs_fileoff_t prevo; /* previous file offset */ | |
2889 | xfs_fileoff_t align_off; /* temp for offset */ | |
2890 | xfs_extlen_t align_alen; /* temp for length */ | |
2891 | xfs_extlen_t temp; /* temp for calculations */ | |
4e8938fe | 2892 | |
9e5987a7 | 2893 | if (convert) |
4e8938fe | 2894 | return 0; |
4e8938fe | 2895 | |
9e5987a7 DC |
2896 | orig_off = align_off = *offp; |
2897 | orig_alen = align_alen = *lenp; | |
2898 | orig_end = orig_off + orig_alen; | |
1da177e4 | 2899 | |
1da177e4 | 2900 | /* |
9e5987a7 DC |
2901 | * If this request overlaps an existing extent, then don't |
2902 | * attempt to perform any additional alignment. | |
1da177e4 | 2903 | */ |
9e5987a7 DC |
2904 | if (!delay && !eof && |
2905 | (orig_off >= gotp->br_startoff) && | |
2906 | (orig_end <= gotp->br_startoff + gotp->br_blockcount)) { | |
2907 | return 0; | |
2908 | } | |
2909 | ||
1da177e4 | 2910 | /* |
9e5987a7 DC |
2911 | * If the file offset is unaligned vs. the extent size |
2912 | * we need to align it. This will be possible unless | |
2913 | * the file was previously written with a kernel that didn't | |
2914 | * perform this alignment, or if a truncate shot us in the | |
2915 | * foot. | |
1da177e4 | 2916 | */ |
0703a8e1 | 2917 | div_u64_rem(orig_off, extsz, &temp); |
9e5987a7 DC |
2918 | if (temp) { |
2919 | align_alen += temp; | |
2920 | align_off -= temp; | |
1da177e4 | 2921 | } |
6dea405e DC |
2922 | |
2923 | /* Same adjustment for the end of the requested area. */ | |
2924 | temp = (align_alen % extsz); | |
2925 | if (temp) | |
2926 | align_alen += extsz - temp; | |
2927 | ||
1da177e4 | 2928 | /* |
6dea405e DC |
2929 | * For large extent hint sizes, the aligned extent might be larger than |
2930 | * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls | |
2931 | * the length back under MAXEXTLEN. The outer allocation loops handle | |
2932 | * short allocation just fine, so it is safe to do this. We only want to | |
2933 | * do it when we are forced to, though, because it means more allocation | |
2934 | * operations are required. | |
1da177e4 | 2935 | */ |
6dea405e DC |
2936 | while (align_alen > MAXEXTLEN) |
2937 | align_alen -= extsz; | |
2938 | ASSERT(align_alen <= MAXEXTLEN); | |
2939 | ||
1da177e4 | 2940 | /* |
9e5987a7 DC |
2941 | * If the previous block overlaps with this proposed allocation |
2942 | * then move the start forward without adjusting the length. | |
1da177e4 | 2943 | */ |
9e5987a7 DC |
2944 | if (prevp->br_startoff != NULLFILEOFF) { |
2945 | if (prevp->br_startblock == HOLESTARTBLOCK) | |
2946 | prevo = prevp->br_startoff; | |
2947 | else | |
2948 | prevo = prevp->br_startoff + prevp->br_blockcount; | |
2949 | } else | |
2950 | prevo = 0; | |
2951 | if (align_off != orig_off && align_off < prevo) | |
2952 | align_off = prevo; | |
2953 | /* | |
2954 | * If the next block overlaps with this proposed allocation | |
2955 | * then move the start back without adjusting the length, | |
2956 | * but not before offset 0. | |
2957 | * This may of course make the start overlap previous block, | |
2958 | * and if we hit the offset 0 limit then the next block | |
2959 | * can still overlap too. | |
2960 | */ | |
2961 | if (!eof && gotp->br_startoff != NULLFILEOFF) { | |
2962 | if ((delay && gotp->br_startblock == HOLESTARTBLOCK) || | |
2963 | (!delay && gotp->br_startblock == DELAYSTARTBLOCK)) | |
2964 | nexto = gotp->br_startoff + gotp->br_blockcount; | |
2965 | else | |
2966 | nexto = gotp->br_startoff; | |
2967 | } else | |
2968 | nexto = NULLFILEOFF; | |
2969 | if (!eof && | |
2970 | align_off + align_alen != orig_end && | |
2971 | align_off + align_alen > nexto) | |
2972 | align_off = nexto > align_alen ? nexto - align_alen : 0; | |
2973 | /* | |
2974 | * If we're now overlapping the next or previous extent that | |
2975 | * means we can't fit an extsz piece in this hole. Just move | |
2976 | * the start forward to the first valid spot and set | |
2977 | * the length so we hit the end. | |
2978 | */ | |
2979 | if (align_off != orig_off && align_off < prevo) | |
2980 | align_off = prevo; | |
2981 | if (align_off + align_alen != orig_end && | |
2982 | align_off + align_alen > nexto && | |
2983 | nexto != NULLFILEOFF) { | |
2984 | ASSERT(nexto > prevo); | |
2985 | align_alen = nexto - align_off; | |
2986 | } | |
1da177e4 | 2987 | |
9e5987a7 DC |
2988 | /* |
2989 | * If realtime, and the result isn't a multiple of the realtime | |
2990 | * extent size we need to remove blocks until it is. | |
2991 | */ | |
2992 | if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) { | |
1da177e4 | 2993 | /* |
9e5987a7 DC |
2994 | * We're not covering the original request, or |
2995 | * we won't be able to once we fix the length. | |
1da177e4 | 2996 | */ |
9e5987a7 DC |
2997 | if (orig_off < align_off || |
2998 | orig_end > align_off + align_alen || | |
2999 | align_alen - temp < orig_alen) | |
2451337d | 3000 | return -EINVAL; |
1da177e4 | 3001 | /* |
9e5987a7 | 3002 | * Try to fix it by moving the start up. |
1da177e4 | 3003 | */ |
9e5987a7 DC |
3004 | if (align_off + temp <= orig_off) { |
3005 | align_alen -= temp; | |
3006 | align_off += temp; | |
1da177e4 | 3007 | } |
9e5987a7 DC |
3008 | /* |
3009 | * Try to fix it by moving the end in. | |
3010 | */ | |
3011 | else if (align_off + align_alen - temp >= orig_end) | |
3012 | align_alen -= temp; | |
3013 | /* | |
3014 | * Set the start to the minimum then trim the length. | |
3015 | */ | |
3016 | else { | |
3017 | align_alen -= orig_off - align_off; | |
3018 | align_off = orig_off; | |
3019 | align_alen -= align_alen % mp->m_sb.sb_rextsize; | |
1da177e4 | 3020 | } |
1da177e4 | 3021 | /* |
9e5987a7 | 3022 | * Result doesn't cover the request, fail it. |
1da177e4 | 3023 | */ |
9e5987a7 | 3024 | if (orig_off < align_off || orig_end > align_off + align_alen) |
2451337d | 3025 | return -EINVAL; |
9e5987a7 DC |
3026 | } else { |
3027 | ASSERT(orig_off >= align_off); | |
6dea405e DC |
3028 | /* see MAXEXTLEN handling above */ |
3029 | ASSERT(orig_end <= align_off + align_alen || | |
3030 | align_alen + extsz > MAXEXTLEN); | |
1da177e4 | 3031 | } |
1da177e4 | 3032 | |
0b1b213f | 3033 | #ifdef DEBUG |
9e5987a7 DC |
3034 | if (!eof && gotp->br_startoff != NULLFILEOFF) |
3035 | ASSERT(align_off + align_alen <= gotp->br_startoff); | |
3036 | if (prevp->br_startoff != NULLFILEOFF) | |
3037 | ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount); | |
3038 | #endif | |
1da177e4 | 3039 | |
9e5987a7 DC |
3040 | *lenp = align_alen; |
3041 | *offp = align_off; | |
3042 | return 0; | |
1da177e4 | 3043 | } |
1da177e4 | 3044 | |
9e5987a7 DC |
3045 | #define XFS_ALLOC_GAP_UNITS 4 |
3046 | ||
68988114 | 3047 | void |
9e5987a7 | 3048 | xfs_bmap_adjacent( |
68988114 | 3049 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
1da177e4 | 3050 | { |
9e5987a7 DC |
3051 | xfs_fsblock_t adjust; /* adjustment to block numbers */ |
3052 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ | |
3053 | xfs_mount_t *mp; /* mount point structure */ | |
3054 | int nullfb; /* true if ap->firstblock isn't set */ | |
3055 | int rt; /* true if inode is realtime */ | |
1da177e4 | 3056 | |
9e5987a7 DC |
3057 | #define ISVALID(x,y) \ |
3058 | (rt ? \ | |
3059 | (x) < mp->m_sb.sb_rblocks : \ | |
3060 | XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \ | |
3061 | XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \ | |
3062 | XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks) | |
1da177e4 | 3063 | |
9e5987a7 | 3064 | mp = ap->ip->i_mount; |
94c07b4d | 3065 | nullfb = ap->tp->t_firstblock == NULLFSBLOCK; |
292378ed DC |
3066 | rt = XFS_IS_REALTIME_INODE(ap->ip) && |
3067 | xfs_alloc_is_userdata(ap->datatype); | |
94c07b4d BF |
3068 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, |
3069 | ap->tp->t_firstblock); | |
9e5987a7 DC |
3070 | /* |
3071 | * If allocating at eof, and there's a previous real block, | |
3072 | * try to use its last block as our starting point. | |
3073 | */ | |
3074 | if (ap->eof && ap->prev.br_startoff != NULLFILEOFF && | |
3075 | !isnullstartblock(ap->prev.br_startblock) && | |
3076 | ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount, | |
3077 | ap->prev.br_startblock)) { | |
3078 | ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount; | |
3079 | /* | |
3080 | * Adjust for the gap between prevp and us. | |
3081 | */ | |
3082 | adjust = ap->offset - | |
3083 | (ap->prev.br_startoff + ap->prev.br_blockcount); | |
3084 | if (adjust && | |
3085 | ISVALID(ap->blkno + adjust, ap->prev.br_startblock)) | |
3086 | ap->blkno += adjust; | |
3087 | } | |
3088 | /* | |
3089 | * If not at eof, then compare the two neighbor blocks. | |
3090 | * Figure out whether either one gives us a good starting point, | |
3091 | * and pick the better one. | |
3092 | */ | |
3093 | else if (!ap->eof) { | |
3094 | xfs_fsblock_t gotbno; /* right side block number */ | |
3095 | xfs_fsblock_t gotdiff=0; /* right side difference */ | |
3096 | xfs_fsblock_t prevbno; /* left side block number */ | |
3097 | xfs_fsblock_t prevdiff=0; /* left side difference */ | |
3098 | ||
3099 | /* | |
3100 | * If there's a previous (left) block, select a requested | |
3101 | * start block based on it. | |
3102 | */ | |
3103 | if (ap->prev.br_startoff != NULLFILEOFF && | |
3104 | !isnullstartblock(ap->prev.br_startblock) && | |
3105 | (prevbno = ap->prev.br_startblock + | |
3106 | ap->prev.br_blockcount) && | |
3107 | ISVALID(prevbno, ap->prev.br_startblock)) { | |
3108 | /* | |
3109 | * Calculate gap to end of previous block. | |
3110 | */ | |
3111 | adjust = prevdiff = ap->offset - | |
3112 | (ap->prev.br_startoff + | |
3113 | ap->prev.br_blockcount); | |
3114 | /* | |
3115 | * Figure the startblock based on the previous block's | |
3116 | * end and the gap size. | |
3117 | * Heuristic! | |
3118 | * If the gap is large relative to the piece we're | |
3119 | * allocating, or using it gives us an invalid block | |
3120 | * number, then just use the end of the previous block. | |
3121 | */ | |
3122 | if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3123 | ISVALID(prevbno + prevdiff, | |
3124 | ap->prev.br_startblock)) | |
3125 | prevbno += adjust; | |
3126 | else | |
3127 | prevdiff += adjust; | |
3128 | /* | |
3129 | * If the firstblock forbids it, can't use it, | |
3130 | * must use default. | |
3131 | */ | |
3132 | if (!rt && !nullfb && | |
3133 | XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno) | |
3134 | prevbno = NULLFSBLOCK; | |
1da177e4 | 3135 | } |
9e5987a7 DC |
3136 | /* |
3137 | * No previous block or can't follow it, just default. | |
3138 | */ | |
3139 | else | |
3140 | prevbno = NULLFSBLOCK; | |
3141 | /* | |
3142 | * If there's a following (right) block, select a requested | |
3143 | * start block based on it. | |
3144 | */ | |
3145 | if (!isnullstartblock(ap->got.br_startblock)) { | |
3146 | /* | |
3147 | * Calculate gap to start of next block. | |
3148 | */ | |
3149 | adjust = gotdiff = ap->got.br_startoff - ap->offset; | |
3150 | /* | |
3151 | * Figure the startblock based on the next block's | |
3152 | * start and the gap size. | |
3153 | */ | |
3154 | gotbno = ap->got.br_startblock; | |
3155 | /* | |
3156 | * Heuristic! | |
3157 | * If the gap is large relative to the piece we're | |
3158 | * allocating, or using it gives us an invalid block | |
3159 | * number, then just use the start of the next block | |
3160 | * offset by our length. | |
3161 | */ | |
3162 | if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3163 | ISVALID(gotbno - gotdiff, gotbno)) | |
3164 | gotbno -= adjust; | |
3165 | else if (ISVALID(gotbno - ap->length, gotbno)) { | |
3166 | gotbno -= ap->length; | |
3167 | gotdiff += adjust - ap->length; | |
3168 | } else | |
3169 | gotdiff += adjust; | |
3170 | /* | |
3171 | * If the firstblock forbids it, can't use it, | |
3172 | * must use default. | |
3173 | */ | |
3174 | if (!rt && !nullfb && | |
3175 | XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno) | |
3176 | gotbno = NULLFSBLOCK; | |
3177 | } | |
3178 | /* | |
3179 | * No next block, just default. | |
3180 | */ | |
3181 | else | |
3182 | gotbno = NULLFSBLOCK; | |
3183 | /* | |
3184 | * If both valid, pick the better one, else the only good | |
3185 | * one, else ap->blkno is already set (to 0 or the inode block). | |
3186 | */ | |
3187 | if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) | |
3188 | ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno; | |
3189 | else if (prevbno != NULLFSBLOCK) | |
3190 | ap->blkno = prevbno; | |
3191 | else if (gotbno != NULLFSBLOCK) | |
3192 | ap->blkno = gotbno; | |
1da177e4 | 3193 | } |
9e5987a7 | 3194 | #undef ISVALID |
1da177e4 | 3195 | } |
1da177e4 | 3196 | |
c977eb10 CH |
3197 | static int |
3198 | xfs_bmap_longest_free_extent( | |
3199 | struct xfs_trans *tp, | |
3200 | xfs_agnumber_t ag, | |
3201 | xfs_extlen_t *blen, | |
3202 | int *notinit) | |
3203 | { | |
3204 | struct xfs_mount *mp = tp->t_mountp; | |
3205 | struct xfs_perag *pag; | |
3206 | xfs_extlen_t longest; | |
3207 | int error = 0; | |
3208 | ||
3209 | pag = xfs_perag_get(mp, ag); | |
3210 | if (!pag->pagf_init) { | |
3211 | error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK); | |
3212 | if (error) | |
3213 | goto out; | |
3214 | ||
3215 | if (!pag->pagf_init) { | |
3216 | *notinit = 1; | |
3217 | goto out; | |
3218 | } | |
3219 | } | |
3220 | ||
a1f69417 | 3221 | longest = xfs_alloc_longest_free_extent(pag, |
3fd129b6 DW |
3222 | xfs_alloc_min_freelist(mp, pag), |
3223 | xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE)); | |
c977eb10 CH |
3224 | if (*blen < longest) |
3225 | *blen = longest; | |
3226 | ||
3227 | out: | |
3228 | xfs_perag_put(pag); | |
3229 | return error; | |
3230 | } | |
3231 | ||
3232 | static void | |
3233 | xfs_bmap_select_minlen( | |
3234 | struct xfs_bmalloca *ap, | |
3235 | struct xfs_alloc_arg *args, | |
3236 | xfs_extlen_t *blen, | |
3237 | int notinit) | |
3238 | { | |
3239 | if (notinit || *blen < ap->minlen) { | |
3240 | /* | |
3241 | * Since we did a BUF_TRYLOCK above, it is possible that | |
3242 | * there is space for this request. | |
3243 | */ | |
3244 | args->minlen = ap->minlen; | |
3245 | } else if (*blen < args->maxlen) { | |
3246 | /* | |
3247 | * If the best seen length is less than the request length, | |
3248 | * use the best as the minimum. | |
3249 | */ | |
3250 | args->minlen = *blen; | |
3251 | } else { | |
3252 | /* | |
3253 | * Otherwise we've seen an extent as big as maxlen, use that | |
3254 | * as the minimum. | |
3255 | */ | |
3256 | args->minlen = args->maxlen; | |
3257 | } | |
3258 | } | |
3259 | ||
b64dfe4e | 3260 | STATIC int |
9e5987a7 DC |
3261 | xfs_bmap_btalloc_nullfb( |
3262 | struct xfs_bmalloca *ap, | |
3263 | struct xfs_alloc_arg *args, | |
3264 | xfs_extlen_t *blen) | |
b64dfe4e | 3265 | { |
9e5987a7 | 3266 | struct xfs_mount *mp = ap->ip->i_mount; |
9e5987a7 DC |
3267 | xfs_agnumber_t ag, startag; |
3268 | int notinit = 0; | |
b64dfe4e CH |
3269 | int error; |
3270 | ||
c977eb10 | 3271 | args->type = XFS_ALLOCTYPE_START_BNO; |
9e5987a7 | 3272 | args->total = ap->total; |
b64dfe4e | 3273 | |
9e5987a7 DC |
3274 | startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno); |
3275 | if (startag == NULLAGNUMBER) | |
3276 | startag = ag = 0; | |
b64dfe4e | 3277 | |
9e5987a7 | 3278 | while (*blen < args->maxlen) { |
c977eb10 CH |
3279 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, |
3280 | ¬init); | |
3281 | if (error) | |
3282 | return error; | |
b64dfe4e | 3283 | |
9e5987a7 DC |
3284 | if (++ag == mp->m_sb.sb_agcount) |
3285 | ag = 0; | |
3286 | if (ag == startag) | |
3287 | break; | |
9e5987a7 | 3288 | } |
b64dfe4e | 3289 | |
c977eb10 CH |
3290 | xfs_bmap_select_minlen(ap, args, blen, notinit); |
3291 | return 0; | |
3292 | } | |
3293 | ||
3294 | STATIC int | |
3295 | xfs_bmap_btalloc_filestreams( | |
3296 | struct xfs_bmalloca *ap, | |
3297 | struct xfs_alloc_arg *args, | |
3298 | xfs_extlen_t *blen) | |
3299 | { | |
3300 | struct xfs_mount *mp = ap->ip->i_mount; | |
3301 | xfs_agnumber_t ag; | |
3302 | int notinit = 0; | |
3303 | int error; | |
3304 | ||
3305 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
3306 | args->total = ap->total; | |
3307 | ||
3308 | ag = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3309 | if (ag == NULLAGNUMBER) | |
3310 | ag = 0; | |
3311 | ||
3312 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init); | |
3313 | if (error) | |
3314 | return error; | |
3315 | ||
3316 | if (*blen < args->maxlen) { | |
3317 | error = xfs_filestream_new_ag(ap, &ag); | |
3318 | if (error) | |
3319 | return error; | |
3320 | ||
3321 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, | |
3322 | ¬init); | |
3323 | if (error) | |
3324 | return error; | |
3325 | ||
3326 | } | |
3327 | ||
3328 | xfs_bmap_select_minlen(ap, args, blen, notinit); | |
b64dfe4e CH |
3329 | |
3330 | /* | |
c977eb10 CH |
3331 | * Set the failure fallback case to look in the selected AG as stream |
3332 | * may have moved. | |
b64dfe4e | 3333 | */ |
c977eb10 | 3334 | ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0); |
b64dfe4e | 3335 | return 0; |
b64dfe4e CH |
3336 | } |
3337 | ||
751f3767 DW |
3338 | /* Update all inode and quota accounting for the allocation we just did. */ |
3339 | static void | |
3340 | xfs_bmap_btalloc_accounting( | |
3341 | struct xfs_bmalloca *ap, | |
3342 | struct xfs_alloc_arg *args) | |
3343 | { | |
4b4c1326 DW |
3344 | if (ap->flags & XFS_BMAPI_COWFORK) { |
3345 | /* | |
3346 | * COW fork blocks are in-core only and thus are treated as | |
3347 | * in-core quota reservation (like delalloc blocks) even when | |
3348 | * converted to real blocks. The quota reservation is not | |
3349 | * accounted to disk until blocks are remapped to the data | |
3350 | * fork. So if these blocks were previously delalloc, we | |
3351 | * already have quota reservation and there's nothing to do | |
3352 | * yet. | |
3353 | */ | |
9fe82b8c DW |
3354 | if (ap->wasdel) { |
3355 | xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len); | |
4b4c1326 | 3356 | return; |
9fe82b8c | 3357 | } |
4b4c1326 DW |
3358 | |
3359 | /* | |
3360 | * Otherwise, we've allocated blocks in a hole. The transaction | |
3361 | * has acquired in-core quota reservation for this extent. | |
3362 | * Rather than account these as real blocks, however, we reduce | |
3363 | * the transaction quota reservation based on the allocation. | |
3364 | * This essentially transfers the transaction quota reservation | |
3365 | * to that of a delalloc extent. | |
3366 | */ | |
3367 | ap->ip->i_delayed_blks += args->len; | |
3368 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, XFS_TRANS_DQ_RES_BLKS, | |
3369 | -(long)args->len); | |
3370 | return; | |
3371 | } | |
3372 | ||
3373 | /* data/attr fork only */ | |
3374 | ap->ip->i_d.di_nblocks += args->len; | |
751f3767 | 3375 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); |
9fe82b8c | 3376 | if (ap->wasdel) { |
751f3767 | 3377 | ap->ip->i_delayed_blks -= args->len; |
9fe82b8c DW |
3378 | xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)args->len); |
3379 | } | |
751f3767 DW |
3380 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, |
3381 | ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : XFS_TRANS_DQ_BCOUNT, | |
3382 | args->len); | |
3383 | } | |
3384 | ||
9e5987a7 DC |
3385 | STATIC int |
3386 | xfs_bmap_btalloc( | |
68988114 | 3387 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
4403280a | 3388 | { |
9e5987a7 DC |
3389 | xfs_mount_t *mp; /* mount point structure */ |
3390 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
292378ed | 3391 | xfs_extlen_t align = 0; /* minimum allocation alignment */ |
9e5987a7 DC |
3392 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ |
3393 | xfs_agnumber_t ag; | |
3394 | xfs_alloc_arg_t args; | |
6d8a45ce DW |
3395 | xfs_fileoff_t orig_offset; |
3396 | xfs_extlen_t orig_length; | |
9e5987a7 DC |
3397 | xfs_extlen_t blen; |
3398 | xfs_extlen_t nextminlen = 0; | |
3399 | int nullfb; /* true if ap->firstblock isn't set */ | |
3400 | int isaligned; | |
3401 | int tryagain; | |
3402 | int error; | |
33177f05 | 3403 | int stripe_align; |
4403280a | 3404 | |
9e5987a7 | 3405 | ASSERT(ap->length); |
6d8a45ce DW |
3406 | orig_offset = ap->offset; |
3407 | orig_length = ap->length; | |
4403280a | 3408 | |
9e5987a7 | 3409 | mp = ap->ip->i_mount; |
33177f05 DC |
3410 | |
3411 | /* stripe alignment for allocation is determined by mount parameters */ | |
3412 | stripe_align = 0; | |
3413 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
3414 | stripe_align = mp->m_swidth; | |
3415 | else if (mp->m_dalign) | |
3416 | stripe_align = mp->m_dalign; | |
3417 | ||
f7ca3522 DW |
3418 | if (ap->flags & XFS_BMAPI_COWFORK) |
3419 | align = xfs_get_cowextsz_hint(ap->ip); | |
3420 | else if (xfs_alloc_is_userdata(ap->datatype)) | |
292378ed | 3421 | align = xfs_get_extsz_hint(ap->ip); |
493611eb | 3422 | if (align) { |
9e5987a7 DC |
3423 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, |
3424 | align, 0, ap->eof, 0, ap->conv, | |
3425 | &ap->offset, &ap->length); | |
3426 | ASSERT(!error); | |
3427 | ASSERT(ap->length); | |
4403280a | 3428 | } |
33177f05 DC |
3429 | |
3430 | ||
94c07b4d BF |
3431 | nullfb = ap->tp->t_firstblock == NULLFSBLOCK; |
3432 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, | |
3433 | ap->tp->t_firstblock); | |
9e5987a7 | 3434 | if (nullfb) { |
292378ed DC |
3435 | if (xfs_alloc_is_userdata(ap->datatype) && |
3436 | xfs_inode_is_filestream(ap->ip)) { | |
9e5987a7 DC |
3437 | ag = xfs_filestream_lookup_ag(ap->ip); |
3438 | ag = (ag != NULLAGNUMBER) ? ag : 0; | |
3439 | ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0); | |
3440 | } else { | |
3441 | ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino); | |
3442 | } | |
3443 | } else | |
94c07b4d | 3444 | ap->blkno = ap->tp->t_firstblock; |
4403280a | 3445 | |
9e5987a7 | 3446 | xfs_bmap_adjacent(ap); |
4403280a | 3447 | |
9e5987a7 DC |
3448 | /* |
3449 | * If allowed, use ap->blkno; otherwise must use firstblock since | |
3450 | * it's in the right allocation group. | |
3451 | */ | |
3452 | if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno) | |
3453 | ; | |
3454 | else | |
94c07b4d | 3455 | ap->blkno = ap->tp->t_firstblock; |
9e5987a7 DC |
3456 | /* |
3457 | * Normal allocation, done through xfs_alloc_vextent. | |
3458 | */ | |
3459 | tryagain = isaligned = 0; | |
3460 | memset(&args, 0, sizeof(args)); | |
3461 | args.tp = ap->tp; | |
3462 | args.mp = mp; | |
3463 | args.fsbno = ap->blkno; | |
7280feda | 3464 | args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE; |
4403280a | 3465 | |
9e5987a7 | 3466 | /* Trim the allocation back to the maximum an AG can fit. */ |
9bb54cb5 | 3467 | args.maxlen = min(ap->length, mp->m_ag_max_usable); |
9e5987a7 DC |
3468 | blen = 0; |
3469 | if (nullfb) { | |
c977eb10 CH |
3470 | /* |
3471 | * Search for an allocation group with a single extent large | |
3472 | * enough for the request. If one isn't found, then adjust | |
3473 | * the minimum allocation size to the largest space found. | |
3474 | */ | |
292378ed DC |
3475 | if (xfs_alloc_is_userdata(ap->datatype) && |
3476 | xfs_inode_is_filestream(ap->ip)) | |
c977eb10 CH |
3477 | error = xfs_bmap_btalloc_filestreams(ap, &args, &blen); |
3478 | else | |
3479 | error = xfs_bmap_btalloc_nullfb(ap, &args, &blen); | |
4403280a CH |
3480 | if (error) |
3481 | return error; | |
1214f1cf | 3482 | } else if (ap->tp->t_flags & XFS_TRANS_LOWMODE) { |
9e5987a7 DC |
3483 | if (xfs_inode_is_filestream(ap->ip)) |
3484 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3485 | else | |
3486 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3487 | args.total = args.minlen = ap->minlen; | |
3488 | } else { | |
3489 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3490 | args.total = ap->total; | |
3491 | args.minlen = ap->minlen; | |
4403280a | 3492 | } |
9e5987a7 | 3493 | /* apply extent size hints if obtained earlier */ |
493611eb | 3494 | if (align) { |
9e5987a7 | 3495 | args.prod = align; |
0703a8e1 DC |
3496 | div_u64_rem(ap->offset, args.prod, &args.mod); |
3497 | if (args.mod) | |
3498 | args.mod = args.prod - args.mod; | |
09cbfeaf | 3499 | } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) { |
9e5987a7 DC |
3500 | args.prod = 1; |
3501 | args.mod = 0; | |
3502 | } else { | |
09cbfeaf | 3503 | args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog; |
0703a8e1 DC |
3504 | div_u64_rem(ap->offset, args.prod, &args.mod); |
3505 | if (args.mod) | |
3506 | args.mod = args.prod - args.mod; | |
4403280a | 3507 | } |
7e47a4ef | 3508 | /* |
9e5987a7 DC |
3509 | * If we are not low on available data blocks, and the |
3510 | * underlying logical volume manager is a stripe, and | |
3511 | * the file offset is zero then try to allocate data | |
3512 | * blocks on stripe unit boundary. | |
3513 | * NOTE: ap->aeof is only set if the allocation length | |
3514 | * is >= the stripe unit and the allocation offset is | |
3515 | * at the end of file. | |
7e47a4ef | 3516 | */ |
1214f1cf | 3517 | if (!(ap->tp->t_flags & XFS_TRANS_LOWMODE) && ap->aeof) { |
9e5987a7 | 3518 | if (!ap->offset) { |
33177f05 | 3519 | args.alignment = stripe_align; |
9e5987a7 DC |
3520 | atype = args.type; |
3521 | isaligned = 1; | |
3522 | /* | |
3523 | * Adjust for alignment | |
3524 | */ | |
3525 | if (blen > args.alignment && blen <= args.maxlen) | |
3526 | args.minlen = blen - args.alignment; | |
3527 | args.minalignslop = 0; | |
3528 | } else { | |
3529 | /* | |
3530 | * First try an exact bno allocation. | |
3531 | * If it fails then do a near or start bno | |
3532 | * allocation with alignment turned on. | |
3533 | */ | |
3534 | atype = args.type; | |
3535 | tryagain = 1; | |
3536 | args.type = XFS_ALLOCTYPE_THIS_BNO; | |
3537 | args.alignment = 1; | |
3538 | /* | |
3539 | * Compute the minlen+alignment for the | |
3540 | * next case. Set slop so that the value | |
3541 | * of minlen+alignment+slop doesn't go up | |
3542 | * between the calls. | |
3543 | */ | |
33177f05 DC |
3544 | if (blen > stripe_align && blen <= args.maxlen) |
3545 | nextminlen = blen - stripe_align; | |
9e5987a7 DC |
3546 | else |
3547 | nextminlen = args.minlen; | |
33177f05 | 3548 | if (nextminlen + stripe_align > args.minlen + 1) |
9e5987a7 | 3549 | args.minalignslop = |
33177f05 | 3550 | nextminlen + stripe_align - |
9e5987a7 DC |
3551 | args.minlen - 1; |
3552 | else | |
3553 | args.minalignslop = 0; | |
7e47a4ef DC |
3554 | } |
3555 | } else { | |
9e5987a7 DC |
3556 | args.alignment = 1; |
3557 | args.minalignslop = 0; | |
7e47a4ef | 3558 | } |
9e5987a7 DC |
3559 | args.minleft = ap->minleft; |
3560 | args.wasdel = ap->wasdel; | |
3fd129b6 | 3561 | args.resv = XFS_AG_RESV_NONE; |
292378ed DC |
3562 | args.datatype = ap->datatype; |
3563 | if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) | |
3fbbbea3 DC |
3564 | args.ip = ap->ip; |
3565 | ||
3566 | error = xfs_alloc_vextent(&args); | |
3567 | if (error) | |
9e5987a7 | 3568 | return error; |
3fbbbea3 | 3569 | |
9e5987a7 DC |
3570 | if (tryagain && args.fsbno == NULLFSBLOCK) { |
3571 | /* | |
3572 | * Exact allocation failed. Now try with alignment | |
3573 | * turned on. | |
3574 | */ | |
3575 | args.type = atype; | |
3576 | args.fsbno = ap->blkno; | |
33177f05 | 3577 | args.alignment = stripe_align; |
9e5987a7 DC |
3578 | args.minlen = nextminlen; |
3579 | args.minalignslop = 0; | |
3580 | isaligned = 1; | |
3581 | if ((error = xfs_alloc_vextent(&args))) | |
3582 | return error; | |
7e47a4ef | 3583 | } |
9e5987a7 DC |
3584 | if (isaligned && args.fsbno == NULLFSBLOCK) { |
3585 | /* | |
3586 | * allocation failed, so turn off alignment and | |
3587 | * try again. | |
3588 | */ | |
3589 | args.type = atype; | |
3590 | args.fsbno = ap->blkno; | |
3591 | args.alignment = 0; | |
3592 | if ((error = xfs_alloc_vextent(&args))) | |
7e47a4ef DC |
3593 | return error; |
3594 | } | |
9e5987a7 DC |
3595 | if (args.fsbno == NULLFSBLOCK && nullfb && |
3596 | args.minlen > ap->minlen) { | |
3597 | args.minlen = ap->minlen; | |
3598 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3599 | args.fsbno = ap->blkno; | |
3600 | if ((error = xfs_alloc_vextent(&args))) | |
3601 | return error; | |
7e47a4ef | 3602 | } |
9e5987a7 DC |
3603 | if (args.fsbno == NULLFSBLOCK && nullfb) { |
3604 | args.fsbno = 0; | |
3605 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3606 | args.total = ap->minlen; | |
9e5987a7 DC |
3607 | if ((error = xfs_alloc_vextent(&args))) |
3608 | return error; | |
1214f1cf | 3609 | ap->tp->t_flags |= XFS_TRANS_LOWMODE; |
9e5987a7 DC |
3610 | } |
3611 | if (args.fsbno != NULLFSBLOCK) { | |
3612 | /* | |
3613 | * check the allocation happened at the same or higher AG than | |
3614 | * the first block that was allocated. | |
3615 | */ | |
94c07b4d BF |
3616 | ASSERT(ap->tp->t_firstblock == NULLFSBLOCK || |
3617 | XFS_FSB_TO_AGNO(mp, ap->tp->t_firstblock) <= | |
410d17f6 | 3618 | XFS_FSB_TO_AGNO(mp, args.fsbno)); |
7e47a4ef | 3619 | |
9e5987a7 | 3620 | ap->blkno = args.fsbno; |
94c07b4d BF |
3621 | if (ap->tp->t_firstblock == NULLFSBLOCK) |
3622 | ap->tp->t_firstblock = args.fsbno; | |
410d17f6 | 3623 | ASSERT(nullfb || fb_agno <= args.agno); |
9e5987a7 | 3624 | ap->length = args.len; |
6d8a45ce DW |
3625 | /* |
3626 | * If the extent size hint is active, we tried to round the | |
3627 | * caller's allocation request offset down to extsz and the | |
3628 | * length up to another extsz boundary. If we found a free | |
3629 | * extent we mapped it in starting at this new offset. If the | |
3630 | * newly mapped space isn't long enough to cover any of the | |
3631 | * range of offsets that was originally requested, move the | |
3632 | * mapping up so that we can fill as much of the caller's | |
3633 | * original request as possible. Free space is apparently | |
3634 | * very fragmented so we're unlikely to be able to satisfy the | |
3635 | * hints anyway. | |
3636 | */ | |
3637 | if (ap->length <= orig_length) | |
3638 | ap->offset = orig_offset; | |
3639 | else if (ap->offset + ap->length < orig_offset + orig_length) | |
3640 | ap->offset = orig_offset + orig_length - ap->length; | |
751f3767 | 3641 | xfs_bmap_btalloc_accounting(ap, &args); |
9e5987a7 DC |
3642 | } else { |
3643 | ap->blkno = NULLFSBLOCK; | |
3644 | ap->length = 0; | |
3645 | } | |
3646 | return 0; | |
3647 | } | |
7e47a4ef | 3648 | |
9e5987a7 DC |
3649 | /* |
3650 | * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file. | |
3651 | * It figures out where to ask the underlying allocator to put the new extent. | |
3652 | */ | |
3653 | STATIC int | |
3654 | xfs_bmap_alloc( | |
68988114 | 3655 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
9e5987a7 | 3656 | { |
292378ed DC |
3657 | if (XFS_IS_REALTIME_INODE(ap->ip) && |
3658 | xfs_alloc_is_userdata(ap->datatype)) | |
9e5987a7 DC |
3659 | return xfs_bmap_rtalloc(ap); |
3660 | return xfs_bmap_btalloc(ap); | |
3661 | } | |
a5bd606b | 3662 | |
0a0af28c DW |
3663 | /* Trim extent to fit a logical block range. */ |
3664 | void | |
3665 | xfs_trim_extent( | |
3666 | struct xfs_bmbt_irec *irec, | |
3667 | xfs_fileoff_t bno, | |
3668 | xfs_filblks_t len) | |
3669 | { | |
3670 | xfs_fileoff_t distance; | |
3671 | xfs_fileoff_t end = bno + len; | |
3672 | ||
3673 | if (irec->br_startoff + irec->br_blockcount <= bno || | |
3674 | irec->br_startoff >= end) { | |
3675 | irec->br_blockcount = 0; | |
3676 | return; | |
3677 | } | |
3678 | ||
3679 | if (irec->br_startoff < bno) { | |
3680 | distance = bno - irec->br_startoff; | |
3681 | if (isnullstartblock(irec->br_startblock)) | |
3682 | irec->br_startblock = DELAYSTARTBLOCK; | |
3683 | if (irec->br_startblock != DELAYSTARTBLOCK && | |
3684 | irec->br_startblock != HOLESTARTBLOCK) | |
3685 | irec->br_startblock += distance; | |
3686 | irec->br_startoff += distance; | |
3687 | irec->br_blockcount -= distance; | |
3688 | } | |
3689 | ||
3690 | if (end < irec->br_startoff + irec->br_blockcount) { | |
3691 | distance = irec->br_startoff + irec->br_blockcount - end; | |
3692 | irec->br_blockcount -= distance; | |
3693 | } | |
3694 | } | |
3695 | ||
9e5987a7 DC |
3696 | /* |
3697 | * Trim the returned map to the required bounds | |
3698 | */ | |
3699 | STATIC void | |
3700 | xfs_bmapi_trim_map( | |
3701 | struct xfs_bmbt_irec *mval, | |
3702 | struct xfs_bmbt_irec *got, | |
3703 | xfs_fileoff_t *bno, | |
3704 | xfs_filblks_t len, | |
3705 | xfs_fileoff_t obno, | |
3706 | xfs_fileoff_t end, | |
3707 | int n, | |
3708 | int flags) | |
3709 | { | |
3710 | if ((flags & XFS_BMAPI_ENTIRE) || | |
3711 | got->br_startoff + got->br_blockcount <= obno) { | |
3712 | *mval = *got; | |
3713 | if (isnullstartblock(got->br_startblock)) | |
3714 | mval->br_startblock = DELAYSTARTBLOCK; | |
3715 | return; | |
3716 | } | |
7e47a4ef | 3717 | |
9e5987a7 DC |
3718 | if (obno > *bno) |
3719 | *bno = obno; | |
3720 | ASSERT((*bno >= obno) || (n == 0)); | |
3721 | ASSERT(*bno < end); | |
3722 | mval->br_startoff = *bno; | |
3723 | if (isnullstartblock(got->br_startblock)) | |
3724 | mval->br_startblock = DELAYSTARTBLOCK; | |
3725 | else | |
3726 | mval->br_startblock = got->br_startblock + | |
3727 | (*bno - got->br_startoff); | |
7e47a4ef | 3728 | /* |
9e5987a7 DC |
3729 | * Return the minimum of what we got and what we asked for for |
3730 | * the length. We can use the len variable here because it is | |
3731 | * modified below and we could have been there before coming | |
3732 | * here if the first part of the allocation didn't overlap what | |
3733 | * was asked for. | |
7e47a4ef | 3734 | */ |
9e5987a7 DC |
3735 | mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno, |
3736 | got->br_blockcount - (*bno - got->br_startoff)); | |
3737 | mval->br_state = got->br_state; | |
3738 | ASSERT(mval->br_blockcount <= len); | |
3739 | return; | |
7e47a4ef DC |
3740 | } |
3741 | ||
9e5987a7 DC |
3742 | /* |
3743 | * Update and validate the extent map to return | |
3744 | */ | |
3745 | STATIC void | |
3746 | xfs_bmapi_update_map( | |
3747 | struct xfs_bmbt_irec **map, | |
3748 | xfs_fileoff_t *bno, | |
3749 | xfs_filblks_t *len, | |
3750 | xfs_fileoff_t obno, | |
3751 | xfs_fileoff_t end, | |
3752 | int *n, | |
3753 | int flags) | |
e04426b9 | 3754 | { |
9e5987a7 | 3755 | xfs_bmbt_irec_t *mval = *map; |
e04426b9 | 3756 | |
9e5987a7 DC |
3757 | ASSERT((flags & XFS_BMAPI_ENTIRE) || |
3758 | ((mval->br_startoff + mval->br_blockcount) <= end)); | |
3759 | ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) || | |
3760 | (mval->br_startoff < obno)); | |
e04426b9 | 3761 | |
9e5987a7 DC |
3762 | *bno = mval->br_startoff + mval->br_blockcount; |
3763 | *len = end - *bno; | |
3764 | if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) { | |
3765 | /* update previous map with new information */ | |
3766 | ASSERT(mval->br_startblock == mval[-1].br_startblock); | |
3767 | ASSERT(mval->br_blockcount > mval[-1].br_blockcount); | |
3768 | ASSERT(mval->br_state == mval[-1].br_state); | |
3769 | mval[-1].br_blockcount = mval->br_blockcount; | |
3770 | mval[-1].br_state = mval->br_state; | |
3771 | } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK && | |
3772 | mval[-1].br_startblock != DELAYSTARTBLOCK && | |
3773 | mval[-1].br_startblock != HOLESTARTBLOCK && | |
3774 | mval->br_startblock == mval[-1].br_startblock + | |
3775 | mval[-1].br_blockcount && | |
c3a2f9ff | 3776 | mval[-1].br_state == mval->br_state) { |
9e5987a7 DC |
3777 | ASSERT(mval->br_startoff == |
3778 | mval[-1].br_startoff + mval[-1].br_blockcount); | |
3779 | mval[-1].br_blockcount += mval->br_blockcount; | |
3780 | } else if (*n > 0 && | |
3781 | mval->br_startblock == DELAYSTARTBLOCK && | |
3782 | mval[-1].br_startblock == DELAYSTARTBLOCK && | |
3783 | mval->br_startoff == | |
3784 | mval[-1].br_startoff + mval[-1].br_blockcount) { | |
3785 | mval[-1].br_blockcount += mval->br_blockcount; | |
3786 | mval[-1].br_state = mval->br_state; | |
3787 | } else if (!((*n == 0) && | |
3788 | ((mval->br_startoff + mval->br_blockcount) <= | |
3789 | obno))) { | |
3790 | mval++; | |
3791 | (*n)++; | |
3792 | } | |
3793 | *map = mval; | |
e04426b9 DC |
3794 | } |
3795 | ||
3796 | /* | |
9e5987a7 | 3797 | * Map file blocks to filesystem blocks without allocation. |
e04426b9 DC |
3798 | */ |
3799 | int | |
9e5987a7 DC |
3800 | xfs_bmapi_read( |
3801 | struct xfs_inode *ip, | |
3802 | xfs_fileoff_t bno, | |
b447fe5a | 3803 | xfs_filblks_t len, |
9e5987a7 DC |
3804 | struct xfs_bmbt_irec *mval, |
3805 | int *nmap, | |
c315c90b | 3806 | int flags) |
b447fe5a | 3807 | { |
9e5987a7 DC |
3808 | struct xfs_mount *mp = ip->i_mount; |
3809 | struct xfs_ifork *ifp; | |
3810 | struct xfs_bmbt_irec got; | |
9e5987a7 DC |
3811 | xfs_fileoff_t obno; |
3812 | xfs_fileoff_t end; | |
b2b1712a | 3813 | struct xfs_iext_cursor icur; |
9e5987a7 | 3814 | int error; |
334f3423 | 3815 | bool eof = false; |
9e5987a7 | 3816 | int n = 0; |
3993baeb | 3817 | int whichfork = xfs_bmapi_whichfork(flags); |
b447fe5a | 3818 | |
9e5987a7 DC |
3819 | ASSERT(*nmap >= 1); |
3820 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE| | |
c3a2f9ff | 3821 | XFS_BMAPI_COWFORK))); |
eef334e5 | 3822 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)); |
b447fe5a | 3823 | |
9e5987a7 DC |
3824 | if (unlikely(XFS_TEST_ERROR( |
3825 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
3826 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 3827 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
9e5987a7 | 3828 | XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp); |
2451337d | 3829 | return -EFSCORRUPTED; |
9e5987a7 | 3830 | } |
b447fe5a | 3831 | |
9e5987a7 | 3832 | if (XFS_FORCED_SHUTDOWN(mp)) |
2451337d | 3833 | return -EIO; |
9e5987a7 | 3834 | |
ff6d6af2 | 3835 | XFS_STATS_INC(mp, xs_blk_mapr); |
9e5987a7 DC |
3836 | |
3837 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
8612de3f DW |
3838 | if (!ifp) { |
3839 | /* No CoW fork? Return a hole. */ | |
3840 | if (whichfork == XFS_COW_FORK) { | |
3841 | mval->br_startoff = bno; | |
3842 | mval->br_startblock = HOLESTARTBLOCK; | |
3843 | mval->br_blockcount = len; | |
3844 | mval->br_state = XFS_EXT_NORM; | |
3845 | *nmap = 1; | |
3846 | return 0; | |
3847 | } | |
9e5987a7 | 3848 | |
8612de3f DW |
3849 | /* |
3850 | * A missing attr ifork implies that the inode says we're in | |
3851 | * extents or btree format but failed to pass the inode fork | |
3852 | * verifier while trying to load it. Treat that as a file | |
3853 | * corruption too. | |
3854 | */ | |
3855 | #ifdef DEBUG | |
3856 | xfs_alert(mp, "%s: inode %llu missing fork %d", | |
3857 | __func__, ip->i_ino, whichfork); | |
3858 | #endif /* DEBUG */ | |
3859 | return -EFSCORRUPTED; | |
3993baeb DW |
3860 | } |
3861 | ||
9e5987a7 DC |
3862 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
3863 | error = xfs_iread_extents(NULL, ip, whichfork); | |
3864 | if (error) | |
3865 | return error; | |
b447fe5a | 3866 | } |
b447fe5a | 3867 | |
b2b1712a | 3868 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) |
334f3423 | 3869 | eof = true; |
9e5987a7 DC |
3870 | end = bno + len; |
3871 | obno = bno; | |
b447fe5a | 3872 | |
9e5987a7 DC |
3873 | while (bno < end && n < *nmap) { |
3874 | /* Reading past eof, act as though there's a hole up to end. */ | |
3875 | if (eof) | |
3876 | got.br_startoff = end; | |
3877 | if (got.br_startoff > bno) { | |
3878 | /* Reading in a hole. */ | |
3879 | mval->br_startoff = bno; | |
3880 | mval->br_startblock = HOLESTARTBLOCK; | |
3881 | mval->br_blockcount = | |
3882 | XFS_FILBLKS_MIN(len, got.br_startoff - bno); | |
3883 | mval->br_state = XFS_EXT_NORM; | |
3884 | bno += mval->br_blockcount; | |
3885 | len -= mval->br_blockcount; | |
3886 | mval++; | |
3887 | n++; | |
3888 | continue; | |
3889 | } | |
b447fe5a | 3890 | |
9e5987a7 DC |
3891 | /* set up the extent map to return. */ |
3892 | xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags); | |
3893 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
3894 | ||
3895 | /* If we're done, stop now. */ | |
3896 | if (bno >= end || n >= *nmap) | |
3897 | break; | |
3898 | ||
3899 | /* Else go on to the next record. */ | |
b2b1712a | 3900 | if (!xfs_iext_next_extent(ifp, &icur, &got)) |
334f3423 | 3901 | eof = true; |
9e5987a7 DC |
3902 | } |
3903 | *nmap = n; | |
b447fe5a DC |
3904 | return 0; |
3905 | } | |
3906 | ||
f65e6fad BF |
3907 | /* |
3908 | * Add a delayed allocation extent to an inode. Blocks are reserved from the | |
3909 | * global pool and the extent inserted into the inode in-core extent tree. | |
3910 | * | |
3911 | * On entry, got refers to the first extent beyond the offset of the extent to | |
3912 | * allocate or eof is specified if no such extent exists. On return, got refers | |
3913 | * to the extent record that was inserted to the inode fork. | |
3914 | * | |
3915 | * Note that the allocated extent may have been merged with contiguous extents | |
3916 | * during insertion into the inode fork. Thus, got does not reflect the current | |
3917 | * state of the inode fork on return. If necessary, the caller can use lastx to | |
3918 | * look up the updated record in the inode fork. | |
3919 | */ | |
51446f5b | 3920 | int |
9e5987a7 DC |
3921 | xfs_bmapi_reserve_delalloc( |
3922 | struct xfs_inode *ip, | |
be51f811 | 3923 | int whichfork, |
974ae922 | 3924 | xfs_fileoff_t off, |
9e5987a7 | 3925 | xfs_filblks_t len, |
974ae922 | 3926 | xfs_filblks_t prealloc, |
9e5987a7 | 3927 | struct xfs_bmbt_irec *got, |
b2b1712a | 3928 | struct xfs_iext_cursor *icur, |
9e5987a7 | 3929 | int eof) |
1da177e4 | 3930 | { |
c0dc7828 | 3931 | struct xfs_mount *mp = ip->i_mount; |
be51f811 | 3932 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
9e5987a7 DC |
3933 | xfs_extlen_t alen; |
3934 | xfs_extlen_t indlen; | |
9e5987a7 | 3935 | int error; |
974ae922 | 3936 | xfs_fileoff_t aoff = off; |
c0dc7828 | 3937 | |
974ae922 BF |
3938 | /* |
3939 | * Cap the alloc length. Keep track of prealloc so we know whether to | |
3940 | * tag the inode before we return. | |
3941 | */ | |
3942 | alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN); | |
9e5987a7 DC |
3943 | if (!eof) |
3944 | alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff); | |
974ae922 BF |
3945 | if (prealloc && alen >= len) |
3946 | prealloc = alen - len; | |
1da177e4 | 3947 | |
9e5987a7 | 3948 | /* Figure out the extent size, adjust alen */ |
6ca30729 | 3949 | if (whichfork == XFS_COW_FORK) { |
65c5f419 | 3950 | struct xfs_bmbt_irec prev; |
6ca30729 | 3951 | xfs_extlen_t extsz = xfs_get_cowextsz_hint(ip); |
65c5f419 | 3952 | |
b2b1712a | 3953 | if (!xfs_iext_peek_prev_extent(ifp, icur, &prev)) |
65c5f419 CH |
3954 | prev.br_startoff = NULLFILEOFF; |
3955 | ||
6ca30729 | 3956 | error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof, |
9e5987a7 DC |
3957 | 1, 0, &aoff, &alen); |
3958 | ASSERT(!error); | |
3959 | } | |
3960 | ||
9e5987a7 DC |
3961 | /* |
3962 | * Make a transaction-less quota reservation for delayed allocation | |
3963 | * blocks. This number gets adjusted later. We return if we haven't | |
3964 | * allocated blocks already inside this loop. | |
3965 | */ | |
3966 | error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0, | |
6ca30729 | 3967 | XFS_QMOPT_RES_REGBLKS); |
9e5987a7 DC |
3968 | if (error) |
3969 | return error; | |
3970 | ||
3971 | /* | |
3972 | * Split changing sb for alen and indlen since they could be coming | |
3973 | * from different places. | |
3974 | */ | |
3975 | indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen); | |
3976 | ASSERT(indlen > 0); | |
3977 | ||
6ca30729 | 3978 | error = xfs_mod_fdblocks(mp, -((int64_t)alen), false); |
9e5987a7 DC |
3979 | if (error) |
3980 | goto out_unreserve_quota; | |
3981 | ||
0d485ada | 3982 | error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false); |
9e5987a7 DC |
3983 | if (error) |
3984 | goto out_unreserve_blocks; | |
3985 | ||
3986 | ||
3987 | ip->i_delayed_blks += alen; | |
9fe82b8c | 3988 | xfs_mod_delalloc(ip->i_mount, alen + indlen); |
9e5987a7 DC |
3989 | |
3990 | got->br_startoff = aoff; | |
3991 | got->br_startblock = nullstartblock(indlen); | |
3992 | got->br_blockcount = alen; | |
3993 | got->br_state = XFS_EXT_NORM; | |
9e5987a7 | 3994 | |
b2b1712a | 3995 | xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got); |
9e5987a7 | 3996 | |
974ae922 BF |
3997 | /* |
3998 | * Tag the inode if blocks were preallocated. Note that COW fork | |
3999 | * preallocation can occur at the start or end of the extent, even when | |
4000 | * prealloc == 0, so we must also check the aligned offset and length. | |
4001 | */ | |
4002 | if (whichfork == XFS_DATA_FORK && prealloc) | |
4003 | xfs_inode_set_eofblocks_tag(ip); | |
4004 | if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len)) | |
4005 | xfs_inode_set_cowblocks_tag(ip); | |
4006 | ||
9e5987a7 DC |
4007 | return 0; |
4008 | ||
4009 | out_unreserve_blocks: | |
6ca30729 | 4010 | xfs_mod_fdblocks(mp, alen, false); |
9e5987a7 DC |
4011 | out_unreserve_quota: |
4012 | if (XFS_IS_QUOTA_ON(mp)) | |
6ca30729 SH |
4013 | xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, |
4014 | XFS_QMOPT_RES_REGBLKS); | |
9e5987a7 DC |
4015 | return error; |
4016 | } | |
4017 | ||
cf11da9c DC |
4018 | static int |
4019 | xfs_bmapi_allocate( | |
9e5987a7 DC |
4020 | struct xfs_bmalloca *bma) |
4021 | { | |
4022 | struct xfs_mount *mp = bma->ip->i_mount; | |
60b4984f | 4023 | int whichfork = xfs_bmapi_whichfork(bma->flags); |
9e5987a7 DC |
4024 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4025 | int tmp_logflags = 0; | |
4026 | int error; | |
9e5987a7 DC |
4027 | |
4028 | ASSERT(bma->length > 0); | |
4029 | ||
1da177e4 | 4030 | /* |
9e5987a7 DC |
4031 | * For the wasdelay case, we could also just allocate the stuff asked |
4032 | * for in this bmap call but that wouldn't be as good. | |
1da177e4 | 4033 | */ |
9e5987a7 DC |
4034 | if (bma->wasdel) { |
4035 | bma->length = (xfs_extlen_t)bma->got.br_blockcount; | |
4036 | bma->offset = bma->got.br_startoff; | |
b2b1712a | 4037 | xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev); |
9e5987a7 DC |
4038 | } else { |
4039 | bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN); | |
4040 | if (!bma->eof) | |
4041 | bma->length = XFS_FILBLKS_MIN(bma->length, | |
4042 | bma->got.br_startoff - bma->offset); | |
1da177e4 | 4043 | } |
1da177e4 | 4044 | |
9e5987a7 | 4045 | /* |
292378ed DC |
4046 | * Set the data type being allocated. For the data fork, the first data |
4047 | * in the file is treated differently to all other allocations. For the | |
4048 | * attribute fork, we only need to ensure the allocated range is not on | |
4049 | * the busy list. | |
9e5987a7 DC |
4050 | */ |
4051 | if (!(bma->flags & XFS_BMAPI_METADATA)) { | |
292378ed DC |
4052 | bma->datatype = XFS_ALLOC_NOBUSY; |
4053 | if (whichfork == XFS_DATA_FORK) { | |
4054 | if (bma->offset == 0) | |
4055 | bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA; | |
4056 | else | |
4057 | bma->datatype |= XFS_ALLOC_USERDATA; | |
4058 | } | |
3fbbbea3 | 4059 | if (bma->flags & XFS_BMAPI_ZERO) |
292378ed | 4060 | bma->datatype |= XFS_ALLOC_USERDATA_ZERO; |
9e5987a7 | 4061 | } |
1da177e4 | 4062 | |
9e5987a7 | 4063 | bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1; |
0b1b213f | 4064 | |
9e5987a7 DC |
4065 | /* |
4066 | * Only want to do the alignment at the eof if it is userdata and | |
4067 | * allocation length is larger than a stripe unit. | |
4068 | */ | |
4069 | if (mp->m_dalign && bma->length >= mp->m_dalign && | |
4070 | !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) { | |
4071 | error = xfs_bmap_isaeof(bma, whichfork); | |
4072 | if (error) | |
4073 | return error; | |
1da177e4 | 4074 | } |
8096b1eb | 4075 | |
9e5987a7 DC |
4076 | error = xfs_bmap_alloc(bma); |
4077 | if (error) | |
1da177e4 | 4078 | return error; |
9e5987a7 | 4079 | |
9e5987a7 | 4080 | if (bma->blkno == NULLFSBLOCK) |
1da177e4 | 4081 | return 0; |
cf612de7 | 4082 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) |
9e5987a7 | 4083 | bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork); |
9e5987a7 DC |
4084 | /* |
4085 | * Bump the number of extents we've allocated | |
4086 | * in this call. | |
4087 | */ | |
4088 | bma->nallocs++; | |
4089 | ||
4090 | if (bma->cur) | |
4091 | bma->cur->bc_private.b.flags = | |
4092 | bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; | |
4093 | ||
4094 | bma->got.br_startoff = bma->offset; | |
4095 | bma->got.br_startblock = bma->blkno; | |
4096 | bma->got.br_blockcount = bma->length; | |
4097 | bma->got.br_state = XFS_EXT_NORM; | |
b4e9181e | 4098 | |
1da177e4 | 4099 | /* |
05a630d7 DW |
4100 | * In the data fork, a wasdelay extent has been initialized, so |
4101 | * shouldn't be flagged as unwritten. | |
4102 | * | |
4103 | * For the cow fork, however, we convert delalloc reservations | |
4104 | * (extents allocated for speculative preallocation) to | |
4105 | * allocated unwritten extents, and only convert the unwritten | |
4106 | * extents to real extents when we're about to write the data. | |
1da177e4 | 4107 | */ |
05a630d7 | 4108 | if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) && |
daa79bae | 4109 | (bma->flags & XFS_BMAPI_PREALLOC)) |
9e5987a7 DC |
4110 | bma->got.br_state = XFS_EXT_UNWRITTEN; |
4111 | ||
4112 | if (bma->wasdel) | |
60b4984f | 4113 | error = xfs_bmap_add_extent_delay_real(bma, whichfork); |
9e5987a7 | 4114 | else |
6d04558f | 4115 | error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip, |
b2b1712a | 4116 | whichfork, &bma->icur, &bma->cur, &bma->got, |
92f9da30 | 4117 | &bma->logflags, bma->flags); |
9e5987a7 DC |
4118 | |
4119 | bma->logflags |= tmp_logflags; | |
4120 | if (error) | |
4121 | return error; | |
4122 | ||
4123 | /* | |
4124 | * Update our extent pointer, given that xfs_bmap_add_extent_delay_real | |
4125 | * or xfs_bmap_add_extent_hole_real might have merged it into one of | |
4126 | * the neighbouring ones. | |
4127 | */ | |
b2b1712a | 4128 | xfs_iext_get_extent(ifp, &bma->icur, &bma->got); |
9e5987a7 DC |
4129 | |
4130 | ASSERT(bma->got.br_startoff <= bma->offset); | |
4131 | ASSERT(bma->got.br_startoff + bma->got.br_blockcount >= | |
4132 | bma->offset + bma->length); | |
4133 | ASSERT(bma->got.br_state == XFS_EXT_NORM || | |
4134 | bma->got.br_state == XFS_EXT_UNWRITTEN); | |
4135 | return 0; | |
4136 | } | |
4137 | ||
9e5987a7 DC |
4138 | STATIC int |
4139 | xfs_bmapi_convert_unwritten( | |
4140 | struct xfs_bmalloca *bma, | |
4141 | struct xfs_bmbt_irec *mval, | |
4142 | xfs_filblks_t len, | |
4143 | int flags) | |
4144 | { | |
3993baeb | 4145 | int whichfork = xfs_bmapi_whichfork(flags); |
9e5987a7 DC |
4146 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4147 | int tmp_logflags = 0; | |
4148 | int error; | |
4149 | ||
4150 | /* check if we need to do unwritten->real conversion */ | |
4151 | if (mval->br_state == XFS_EXT_UNWRITTEN && | |
4152 | (flags & XFS_BMAPI_PREALLOC)) | |
4153 | return 0; | |
4154 | ||
4155 | /* check if we need to do real->unwritten conversion */ | |
4156 | if (mval->br_state == XFS_EXT_NORM && | |
4157 | (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) != | |
4158 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) | |
4159 | return 0; | |
4160 | ||
4161 | /* | |
4162 | * Modify (by adding) the state flag, if writing. | |
4163 | */ | |
4164 | ASSERT(mval->br_blockcount <= len); | |
4165 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4166 | bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp, | |
4167 | bma->ip, whichfork); | |
1da177e4 | 4168 | } |
9e5987a7 DC |
4169 | mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN) |
4170 | ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
5575acc7 | 4171 | |
3fbbbea3 DC |
4172 | /* |
4173 | * Before insertion into the bmbt, zero the range being converted | |
4174 | * if required. | |
4175 | */ | |
4176 | if (flags & XFS_BMAPI_ZERO) { | |
4177 | error = xfs_zero_extent(bma->ip, mval->br_startblock, | |
4178 | mval->br_blockcount); | |
4179 | if (error) | |
4180 | return error; | |
4181 | } | |
4182 | ||
05a630d7 | 4183 | error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork, |
92f9da30 | 4184 | &bma->icur, &bma->cur, mval, &tmp_logflags); |
2e588a46 BF |
4185 | /* |
4186 | * Log the inode core unconditionally in the unwritten extent conversion | |
4187 | * path because the conversion might not have done so (e.g., if the | |
4188 | * extent count hasn't changed). We need to make sure the inode is dirty | |
4189 | * in the transaction for the sake of fsync(), even if nothing has | |
4190 | * changed, because fsync() will not force the log for this transaction | |
4191 | * unless it sees the inode pinned. | |
05a630d7 DW |
4192 | * |
4193 | * Note: If we're only converting cow fork extents, there aren't | |
4194 | * any on-disk updates to make, so we don't need to log anything. | |
2e588a46 | 4195 | */ |
05a630d7 DW |
4196 | if (whichfork != XFS_COW_FORK) |
4197 | bma->logflags |= tmp_logflags | XFS_ILOG_CORE; | |
9e5987a7 DC |
4198 | if (error) |
4199 | return error; | |
4200 | ||
4201 | /* | |
4202 | * Update our extent pointer, given that | |
4203 | * xfs_bmap_add_extent_unwritten_real might have merged it into one | |
4204 | * of the neighbouring ones. | |
4205 | */ | |
b2b1712a | 4206 | xfs_iext_get_extent(ifp, &bma->icur, &bma->got); |
9e5987a7 DC |
4207 | |
4208 | /* | |
4209 | * We may have combined previously unwritten space with written space, | |
4210 | * so generate another request. | |
4211 | */ | |
4212 | if (mval->br_blockcount < len) | |
2451337d | 4213 | return -EAGAIN; |
9e5987a7 DC |
4214 | return 0; |
4215 | } | |
4216 | ||
c8b54673 CH |
4217 | static inline xfs_extlen_t |
4218 | xfs_bmapi_minleft( | |
4219 | struct xfs_trans *tp, | |
4220 | struct xfs_inode *ip, | |
4221 | int fork) | |
4222 | { | |
4223 | if (tp && tp->t_firstblock != NULLFSBLOCK) | |
4224 | return 0; | |
4225 | if (XFS_IFORK_FORMAT(ip, fork) != XFS_DINODE_FMT_BTREE) | |
4226 | return 1; | |
4227 | return be16_to_cpu(XFS_IFORK_PTR(ip, fork)->if_broot->bb_level) + 1; | |
4228 | } | |
4229 | ||
4230 | /* | |
4231 | * Log whatever the flags say, even if error. Otherwise we might miss detecting | |
4232 | * a case where the data is changed, there's an error, and it's not logged so we | |
4233 | * don't shutdown when we should. Don't bother logging extents/btree changes if | |
4234 | * we converted to the other format. | |
4235 | */ | |
4236 | static void | |
4237 | xfs_bmapi_finish( | |
4238 | struct xfs_bmalloca *bma, | |
4239 | int whichfork, | |
4240 | int error) | |
4241 | { | |
4242 | if ((bma->logflags & xfs_ilog_fext(whichfork)) && | |
4243 | XFS_IFORK_FORMAT(bma->ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
4244 | bma->logflags &= ~xfs_ilog_fext(whichfork); | |
4245 | else if ((bma->logflags & xfs_ilog_fbroot(whichfork)) && | |
4246 | XFS_IFORK_FORMAT(bma->ip, whichfork) != XFS_DINODE_FMT_BTREE) | |
4247 | bma->logflags &= ~xfs_ilog_fbroot(whichfork); | |
4248 | ||
4249 | if (bma->logflags) | |
4250 | xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags); | |
4251 | if (bma->cur) | |
4252 | xfs_btree_del_cursor(bma->cur, error); | |
4253 | } | |
4254 | ||
9e5987a7 DC |
4255 | /* |
4256 | * Map file blocks to filesystem blocks, and allocate blocks or convert the | |
4257 | * extent state if necessary. Details behaviour is controlled by the flags | |
4258 | * parameter. Only allocates blocks from a single allocation group, to avoid | |
4259 | * locking problems. | |
9e5987a7 DC |
4260 | */ |
4261 | int | |
4262 | xfs_bmapi_write( | |
4263 | struct xfs_trans *tp, /* transaction pointer */ | |
4264 | struct xfs_inode *ip, /* incore inode */ | |
4265 | xfs_fileoff_t bno, /* starting file offs. mapped */ | |
4266 | xfs_filblks_t len, /* length to map in file */ | |
4267 | int flags, /* XFS_BMAPI_... */ | |
9e5987a7 DC |
4268 | xfs_extlen_t total, /* total blocks needed */ |
4269 | struct xfs_bmbt_irec *mval, /* output: map values */ | |
6e702a5d | 4270 | int *nmap) /* i/o: mval size/count */ |
9e5987a7 | 4271 | { |
4b0bce30 DW |
4272 | struct xfs_bmalloca bma = { |
4273 | .tp = tp, | |
4274 | .ip = ip, | |
4275 | .total = total, | |
4276 | }; | |
9e5987a7 DC |
4277 | struct xfs_mount *mp = ip->i_mount; |
4278 | struct xfs_ifork *ifp; | |
9e5987a7 | 4279 | xfs_fileoff_t end; /* end of mapped file region */ |
2d58f6ef | 4280 | bool eof = false; /* after the end of extents */ |
9e5987a7 DC |
4281 | int error; /* error return */ |
4282 | int n; /* current extent index */ | |
4283 | xfs_fileoff_t obno; /* old block number (offset) */ | |
4284 | int whichfork; /* data or attr fork */ | |
9e5987a7 DC |
4285 | |
4286 | #ifdef DEBUG | |
4287 | xfs_fileoff_t orig_bno; /* original block number value */ | |
4288 | int orig_flags; /* original flags arg value */ | |
4289 | xfs_filblks_t orig_len; /* original value of len arg */ | |
4290 | struct xfs_bmbt_irec *orig_mval; /* original value of mval */ | |
4291 | int orig_nmap; /* original value of *nmap */ | |
4292 | ||
4293 | orig_bno = bno; | |
4294 | orig_len = len; | |
4295 | orig_flags = flags; | |
4296 | orig_mval = mval; | |
4297 | orig_nmap = *nmap; | |
4298 | #endif | |
60b4984f | 4299 | whichfork = xfs_bmapi_whichfork(flags); |
9e5987a7 DC |
4300 | |
4301 | ASSERT(*nmap >= 1); | |
4302 | ASSERT(*nmap <= XFS_BMAP_MAX_NMAP); | |
26b91c72 | 4303 | ASSERT(tp != NULL); |
9e5987a7 | 4304 | ASSERT(len > 0); |
f3508bcd | 4305 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL); |
eef334e5 | 4306 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
6ebd5a44 | 4307 | ASSERT(!(flags & XFS_BMAPI_REMAP)); |
9e5987a7 | 4308 | |
3fbbbea3 DC |
4309 | /* zeroing is for currently only for data extents, not metadata */ |
4310 | ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) != | |
4311 | (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)); | |
4312 | /* | |
4313 | * we can allocate unwritten extents or pre-zero allocated blocks, | |
4314 | * but it makes no sense to do both at once. This would result in | |
4315 | * zeroing the unwritten extent twice, but it still being an | |
4316 | * unwritten extent.... | |
4317 | */ | |
4318 | ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) != | |
4319 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)); | |
4320 | ||
9e5987a7 DC |
4321 | if (unlikely(XFS_TEST_ERROR( |
4322 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
f3508bcd | 4323 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), |
9e24cfd0 | 4324 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
9e5987a7 | 4325 | XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp); |
2451337d | 4326 | return -EFSCORRUPTED; |
5575acc7 KD |
4327 | } |
4328 | ||
9e5987a7 | 4329 | if (XFS_FORCED_SHUTDOWN(mp)) |
2451337d | 4330 | return -EIO; |
9e5987a7 DC |
4331 | |
4332 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
4333 | ||
ff6d6af2 | 4334 | XFS_STATS_INC(mp, xs_blk_mapw); |
9e5987a7 | 4335 | |
9e5987a7 DC |
4336 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
4337 | error = xfs_iread_extents(tp, ip, whichfork); | |
4338 | if (error) | |
4339 | goto error0; | |
4340 | } | |
4341 | ||
b2b1712a | 4342 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got)) |
2d58f6ef | 4343 | eof = true; |
b2b1712a | 4344 | if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev)) |
2d58f6ef | 4345 | bma.prev.br_startoff = NULLFILEOFF; |
c8b54673 | 4346 | bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork); |
9e5987a7 | 4347 | |
627209fb BF |
4348 | n = 0; |
4349 | end = bno + len; | |
4350 | obno = bno; | |
9e5987a7 | 4351 | while (bno < end && n < *nmap) { |
d2b3964a CH |
4352 | bool need_alloc = false, wasdelay = false; |
4353 | ||
be78ff0e | 4354 | /* in hole or beyond EOF? */ |
d2b3964a | 4355 | if (eof || bma.got.br_startoff > bno) { |
be78ff0e DW |
4356 | /* |
4357 | * CoW fork conversions should /never/ hit EOF or | |
4358 | * holes. There should always be something for us | |
4359 | * to work on. | |
4360 | */ | |
4361 | ASSERT(!((flags & XFS_BMAPI_CONVERT) && | |
4362 | (flags & XFS_BMAPI_COWFORK))); | |
4363 | ||
d8ae82e3 | 4364 | need_alloc = true; |
6ebd5a44 CH |
4365 | } else if (isnullstartblock(bma.got.br_startblock)) { |
4366 | wasdelay = true; | |
d2b3964a | 4367 | } |
f65306ea | 4368 | |
1da177e4 | 4369 | /* |
9e5987a7 DC |
4370 | * First, deal with the hole before the allocated space |
4371 | * that we found, if any. | |
1da177e4 | 4372 | */ |
26b91c72 | 4373 | if (need_alloc || wasdelay) { |
9e5987a7 DC |
4374 | bma.eof = eof; |
4375 | bma.conv = !!(flags & XFS_BMAPI_CONVERT); | |
4376 | bma.wasdel = wasdelay; | |
4377 | bma.offset = bno; | |
4378 | bma.flags = flags; | |
4379 | ||
1da177e4 | 4380 | /* |
9e5987a7 DC |
4381 | * There's a 32/64 bit type mismatch between the |
4382 | * allocation length request (which can be 64 bits in | |
4383 | * length) and the bma length request, which is | |
4384 | * xfs_extlen_t and therefore 32 bits. Hence we have to | |
4385 | * check for 32-bit overflows and handle them here. | |
1da177e4 | 4386 | */ |
9e5987a7 DC |
4387 | if (len > (xfs_filblks_t)MAXEXTLEN) |
4388 | bma.length = MAXEXTLEN; | |
4389 | else | |
4390 | bma.length = len; | |
4391 | ||
4392 | ASSERT(len > 0); | |
4393 | ASSERT(bma.length > 0); | |
4394 | error = xfs_bmapi_allocate(&bma); | |
1da177e4 LT |
4395 | if (error) |
4396 | goto error0; | |
9e5987a7 DC |
4397 | if (bma.blkno == NULLFSBLOCK) |
4398 | break; | |
174edb0e DW |
4399 | |
4400 | /* | |
4401 | * If this is a CoW allocation, record the data in | |
4402 | * the refcount btree for orphan recovery. | |
4403 | */ | |
4404 | if (whichfork == XFS_COW_FORK) { | |
0f37d178 BF |
4405 | error = xfs_refcount_alloc_cow_extent(tp, |
4406 | bma.blkno, bma.length); | |
174edb0e DW |
4407 | if (error) |
4408 | goto error0; | |
4409 | } | |
1da177e4 | 4410 | } |
9e5987a7 DC |
4411 | |
4412 | /* Deal with the allocated space we found. */ | |
4413 | xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno, | |
4414 | end, n, flags); | |
4415 | ||
4416 | /* Execute unwritten extent conversion if necessary */ | |
4417 | error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags); | |
2451337d | 4418 | if (error == -EAGAIN) |
9e5987a7 DC |
4419 | continue; |
4420 | if (error) | |
4421 | goto error0; | |
4422 | ||
4423 | /* update the extent map to return */ | |
4424 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4425 | ||
4426 | /* | |
4427 | * If we're done, stop now. Stop when we've allocated | |
4428 | * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise | |
4429 | * the transaction may get too big. | |
4430 | */ | |
4431 | if (bno >= end || n >= *nmap || bma.nallocs >= *nmap) | |
4432 | break; | |
4433 | ||
4434 | /* Else go on to the next record. */ | |
4435 | bma.prev = bma.got; | |
b2b1712a | 4436 | if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got)) |
2d58f6ef | 4437 | eof = true; |
9e5987a7 DC |
4438 | } |
4439 | *nmap = n; | |
4440 | ||
b101e334 CH |
4441 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags, |
4442 | whichfork); | |
4443 | if (error) | |
4444 | goto error0; | |
9e5987a7 DC |
4445 | |
4446 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE || | |
4447 | XFS_IFORK_NEXTENTS(ip, whichfork) > | |
4448 | XFS_IFORK_MAXEXT(ip, whichfork)); | |
c8b54673 CH |
4449 | xfs_bmapi_finish(&bma, whichfork, 0); |
4450 | xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval, | |
4451 | orig_nmap, *nmap); | |
4452 | return 0; | |
9e5987a7 | 4453 | error0: |
c8b54673 | 4454 | xfs_bmapi_finish(&bma, whichfork, error); |
9e5987a7 DC |
4455 | return error; |
4456 | } | |
06d10dd9 | 4457 | |
627209fb BF |
4458 | /* |
4459 | * Convert an existing delalloc extent to real blocks based on file offset. This | |
4460 | * attempts to allocate the entire delalloc extent and may require multiple | |
4461 | * invocations to allocate the target offset if a large enough physical extent | |
4462 | * is not available. | |
4463 | */ | |
4464 | int | |
4465 | xfs_bmapi_convert_delalloc( | |
627209fb | 4466 | struct xfs_inode *ip, |
627209fb | 4467 | int whichfork, |
491ce61e CH |
4468 | xfs_fileoff_t offset_fsb, |
4469 | struct xfs_bmbt_irec *imap, | |
4470 | unsigned int *seq) | |
627209fb | 4471 | { |
d8ae82e3 | 4472 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
491ce61e | 4473 | struct xfs_mount *mp = ip->i_mount; |
d8ae82e3 | 4474 | struct xfs_bmalloca bma = { NULL }; |
491ce61e | 4475 | struct xfs_trans *tp; |
627209fb | 4476 | int error; |
627209fb | 4477 | |
491ce61e CH |
4478 | /* |
4479 | * Space for the extent and indirect blocks was reserved when the | |
4480 | * delalloc extent was created so there's no need to do so here. | |
4481 | */ | |
4482 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, | |
4483 | XFS_TRANS_RESERVE, &tp); | |
4484 | if (error) | |
4485 | return error; | |
4486 | ||
4487 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
4488 | xfs_trans_ijoin(tp, ip, 0); | |
4489 | ||
d8ae82e3 CH |
4490 | if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) || |
4491 | bma.got.br_startoff > offset_fsb) { | |
4492 | /* | |
4493 | * No extent found in the range we are trying to convert. This | |
4494 | * should only happen for the COW fork, where another thread | |
4495 | * might have moved the extent to the data fork in the meantime. | |
4496 | */ | |
4497 | WARN_ON_ONCE(whichfork != XFS_COW_FORK); | |
491ce61e CH |
4498 | error = -EAGAIN; |
4499 | goto out_trans_cancel; | |
d8ae82e3 | 4500 | } |
627209fb BF |
4501 | |
4502 | /* | |
d8ae82e3 CH |
4503 | * If we find a real extent here we raced with another thread converting |
4504 | * the extent. Just return the real extent at this offset. | |
627209fb | 4505 | */ |
d8ae82e3 CH |
4506 | if (!isnullstartblock(bma.got.br_startblock)) { |
4507 | *imap = bma.got; | |
491ce61e CH |
4508 | *seq = READ_ONCE(ifp->if_seq); |
4509 | goto out_trans_cancel; | |
d8ae82e3 CH |
4510 | } |
4511 | ||
4512 | bma.tp = tp; | |
4513 | bma.ip = ip; | |
4514 | bma.wasdel = true; | |
4515 | bma.offset = bma.got.br_startoff; | |
4516 | bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount, MAXEXTLEN); | |
4517 | bma.total = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK); | |
4518 | bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork); | |
4519 | if (whichfork == XFS_COW_FORK) | |
4520 | bma.flags = XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC; | |
4521 | ||
4522 | if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev)) | |
4523 | bma.prev.br_startoff = NULLFILEOFF; | |
4524 | ||
4525 | error = xfs_bmapi_allocate(&bma); | |
4526 | if (error) | |
4527 | goto out_finish; | |
4528 | ||
4529 | error = -ENOSPC; | |
4530 | if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK)) | |
4531 | goto out_finish; | |
4532 | error = -EFSCORRUPTED; | |
4533 | if (WARN_ON_ONCE(!bma.got.br_startblock && !XFS_IS_REALTIME_INODE(ip))) | |
4534 | goto out_finish; | |
4535 | ||
125851ac CH |
4536 | XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length)); |
4537 | XFS_STATS_INC(mp, xs_xstrat_quick); | |
4538 | ||
d8ae82e3 CH |
4539 | ASSERT(!isnullstartblock(bma.got.br_startblock)); |
4540 | *imap = bma.got; | |
491ce61e | 4541 | *seq = READ_ONCE(ifp->if_seq); |
d8ae82e3 CH |
4542 | |
4543 | if (whichfork == XFS_COW_FORK) { | |
4544 | error = xfs_refcount_alloc_cow_extent(tp, bma.blkno, | |
4545 | bma.length); | |
4546 | if (error) | |
4547 | goto out_finish; | |
4548 | } | |
4549 | ||
4550 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags, | |
4551 | whichfork); | |
491ce61e CH |
4552 | if (error) |
4553 | goto out_finish; | |
4554 | ||
4555 | xfs_bmapi_finish(&bma, whichfork, 0); | |
4556 | error = xfs_trans_commit(tp); | |
4557 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4558 | return error; | |
4559 | ||
d8ae82e3 CH |
4560 | out_finish: |
4561 | xfs_bmapi_finish(&bma, whichfork, error); | |
491ce61e CH |
4562 | out_trans_cancel: |
4563 | xfs_trans_cancel(tp); | |
4564 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
627209fb BF |
4565 | return error; |
4566 | } | |
4567 | ||
7cf199ba | 4568 | int |
6ebd5a44 CH |
4569 | xfs_bmapi_remap( |
4570 | struct xfs_trans *tp, | |
4571 | struct xfs_inode *ip, | |
4572 | xfs_fileoff_t bno, | |
4573 | xfs_filblks_t len, | |
4574 | xfs_fsblock_t startblock, | |
7cf199ba | 4575 | int flags) |
6ebd5a44 CH |
4576 | { |
4577 | struct xfs_mount *mp = ip->i_mount; | |
7cf199ba | 4578 | struct xfs_ifork *ifp; |
6ebd5a44 | 4579 | struct xfs_btree_cur *cur = NULL; |
6ebd5a44 | 4580 | struct xfs_bmbt_irec got; |
b2b1712a | 4581 | struct xfs_iext_cursor icur; |
7cf199ba | 4582 | int whichfork = xfs_bmapi_whichfork(flags); |
6ebd5a44 CH |
4583 | int logflags = 0, error; |
4584 | ||
7cf199ba | 4585 | ifp = XFS_IFORK_PTR(ip, whichfork); |
6ebd5a44 CH |
4586 | ASSERT(len > 0); |
4587 | ASSERT(len <= (xfs_filblks_t)MAXEXTLEN); | |
4588 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
7644bd98 DW |
4589 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC | |
4590 | XFS_BMAPI_NORMAP))); | |
4591 | ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) != | |
4592 | (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)); | |
6ebd5a44 CH |
4593 | |
4594 | if (unlikely(XFS_TEST_ERROR( | |
7cf199ba DW |
4595 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && |
4596 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 4597 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
6ebd5a44 CH |
4598 | XFS_ERROR_REPORT("xfs_bmapi_remap", XFS_ERRLEVEL_LOW, mp); |
4599 | return -EFSCORRUPTED; | |
4600 | } | |
4601 | ||
4602 | if (XFS_FORCED_SHUTDOWN(mp)) | |
4603 | return -EIO; | |
4604 | ||
4605 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
7cf199ba | 4606 | error = xfs_iread_extents(tp, ip, whichfork); |
6ebd5a44 CH |
4607 | if (error) |
4608 | return error; | |
4609 | } | |
4610 | ||
b2b1712a | 4611 | if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) { |
6ebd5a44 CH |
4612 | /* make sure we only reflink into a hole. */ |
4613 | ASSERT(got.br_startoff > bno); | |
4614 | ASSERT(got.br_startoff - bno >= len); | |
4615 | } | |
4616 | ||
bf8eadba CH |
4617 | ip->i_d.di_nblocks += len; |
4618 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
6ebd5a44 CH |
4619 | |
4620 | if (ifp->if_flags & XFS_IFBROOT) { | |
7cf199ba | 4621 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
6ebd5a44 CH |
4622 | cur->bc_private.b.flags = 0; |
4623 | } | |
4624 | ||
4625 | got.br_startoff = bno; | |
4626 | got.br_startblock = startblock; | |
4627 | got.br_blockcount = len; | |
7644bd98 DW |
4628 | if (flags & XFS_BMAPI_PREALLOC) |
4629 | got.br_state = XFS_EXT_UNWRITTEN; | |
4630 | else | |
4631 | got.br_state = XFS_EXT_NORM; | |
6ebd5a44 | 4632 | |
7cf199ba | 4633 | error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur, |
92f9da30 | 4634 | &cur, &got, &logflags, flags); |
6ebd5a44 CH |
4635 | if (error) |
4636 | goto error0; | |
4637 | ||
b101e334 | 4638 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork); |
6ebd5a44 CH |
4639 | |
4640 | error0: | |
4641 | if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) | |
4642 | logflags &= ~XFS_ILOG_DEXT; | |
4643 | else if (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) | |
4644 | logflags &= ~XFS_ILOG_DBROOT; | |
4645 | ||
4646 | if (logflags) | |
4647 | xfs_trans_log_inode(tp, ip, logflags); | |
0b04b6b8 DW |
4648 | if (cur) |
4649 | xfs_btree_del_cursor(cur, error); | |
6ebd5a44 CH |
4650 | return error; |
4651 | } | |
4652 | ||
a9bd24ac BF |
4653 | /* |
4654 | * When a delalloc extent is split (e.g., due to a hole punch), the original | |
4655 | * indlen reservation must be shared across the two new extents that are left | |
4656 | * behind. | |
4657 | * | |
4658 | * Given the original reservation and the worst case indlen for the two new | |
4659 | * extents (as calculated by xfs_bmap_worst_indlen()), split the original | |
d34999c9 BF |
4660 | * reservation fairly across the two new extents. If necessary, steal available |
4661 | * blocks from a deleted extent to make up a reservation deficiency (e.g., if | |
4662 | * ores == 1). The number of stolen blocks is returned. The availability and | |
4663 | * subsequent accounting of stolen blocks is the responsibility of the caller. | |
a9bd24ac | 4664 | */ |
d34999c9 | 4665 | static xfs_filblks_t |
a9bd24ac BF |
4666 | xfs_bmap_split_indlen( |
4667 | xfs_filblks_t ores, /* original res. */ | |
4668 | xfs_filblks_t *indlen1, /* ext1 worst indlen */ | |
d34999c9 BF |
4669 | xfs_filblks_t *indlen2, /* ext2 worst indlen */ |
4670 | xfs_filblks_t avail) /* stealable blocks */ | |
a9bd24ac BF |
4671 | { |
4672 | xfs_filblks_t len1 = *indlen1; | |
4673 | xfs_filblks_t len2 = *indlen2; | |
4674 | xfs_filblks_t nres = len1 + len2; /* new total res. */ | |
d34999c9 | 4675 | xfs_filblks_t stolen = 0; |
75d65361 | 4676 | xfs_filblks_t resfactor; |
d34999c9 BF |
4677 | |
4678 | /* | |
4679 | * Steal as many blocks as we can to try and satisfy the worst case | |
4680 | * indlen for both new extents. | |
4681 | */ | |
75d65361 BF |
4682 | if (ores < nres && avail) |
4683 | stolen = XFS_FILBLKS_MIN(nres - ores, avail); | |
4684 | ores += stolen; | |
4685 | ||
4686 | /* nothing else to do if we've satisfied the new reservation */ | |
4687 | if (ores >= nres) | |
4688 | return stolen; | |
4689 | ||
4690 | /* | |
4691 | * We can't meet the total required reservation for the two extents. | |
4692 | * Calculate the percent of the overall shortage between both extents | |
4693 | * and apply this percentage to each of the requested indlen values. | |
4694 | * This distributes the shortage fairly and reduces the chances that one | |
4695 | * of the two extents is left with nothing when extents are repeatedly | |
4696 | * split. | |
4697 | */ | |
4698 | resfactor = (ores * 100); | |
4699 | do_div(resfactor, nres); | |
4700 | len1 *= resfactor; | |
4701 | do_div(len1, 100); | |
4702 | len2 *= resfactor; | |
4703 | do_div(len2, 100); | |
4704 | ASSERT(len1 + len2 <= ores); | |
4705 | ASSERT(len1 < *indlen1 && len2 < *indlen2); | |
a9bd24ac BF |
4706 | |
4707 | /* | |
75d65361 BF |
4708 | * Hand out the remainder to each extent. If one of the two reservations |
4709 | * is zero, we want to make sure that one gets a block first. The loop | |
4710 | * below starts with len1, so hand len2 a block right off the bat if it | |
4711 | * is zero. | |
a9bd24ac | 4712 | */ |
75d65361 BF |
4713 | ores -= (len1 + len2); |
4714 | ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores); | |
4715 | if (ores && !len2 && *indlen2) { | |
4716 | len2++; | |
4717 | ores--; | |
4718 | } | |
4719 | while (ores) { | |
4720 | if (len1 < *indlen1) { | |
4721 | len1++; | |
4722 | ores--; | |
a9bd24ac | 4723 | } |
75d65361 | 4724 | if (!ores) |
a9bd24ac | 4725 | break; |
75d65361 BF |
4726 | if (len2 < *indlen2) { |
4727 | len2++; | |
4728 | ores--; | |
a9bd24ac BF |
4729 | } |
4730 | } | |
4731 | ||
4732 | *indlen1 = len1; | |
4733 | *indlen2 = len2; | |
d34999c9 BF |
4734 | |
4735 | return stolen; | |
a9bd24ac BF |
4736 | } |
4737 | ||
fa5c836c CH |
4738 | int |
4739 | xfs_bmap_del_extent_delay( | |
4740 | struct xfs_inode *ip, | |
4741 | int whichfork, | |
b2b1712a | 4742 | struct xfs_iext_cursor *icur, |
fa5c836c CH |
4743 | struct xfs_bmbt_irec *got, |
4744 | struct xfs_bmbt_irec *del) | |
4745 | { | |
4746 | struct xfs_mount *mp = ip->i_mount; | |
4747 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
4748 | struct xfs_bmbt_irec new; | |
4749 | int64_t da_old, da_new, da_diff = 0; | |
4750 | xfs_fileoff_t del_endoff, got_endoff; | |
4751 | xfs_filblks_t got_indlen, new_indlen, stolen; | |
060ea65b CH |
4752 | int state = xfs_bmap_fork_to_state(whichfork); |
4753 | int error = 0; | |
fa5c836c CH |
4754 | bool isrt; |
4755 | ||
4756 | XFS_STATS_INC(mp, xs_del_exlist); | |
4757 | ||
4758 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); | |
4759 | del_endoff = del->br_startoff + del->br_blockcount; | |
4760 | got_endoff = got->br_startoff + got->br_blockcount; | |
4761 | da_old = startblockval(got->br_startblock); | |
4762 | da_new = 0; | |
4763 | ||
fa5c836c CH |
4764 | ASSERT(del->br_blockcount > 0); |
4765 | ASSERT(got->br_startoff <= del->br_startoff); | |
4766 | ASSERT(got_endoff >= del_endoff); | |
4767 | ||
4768 | if (isrt) { | |
4f1adf33 | 4769 | uint64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount); |
fa5c836c CH |
4770 | |
4771 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
4772 | xfs_mod_frextents(mp, rtexts); | |
4773 | } | |
4774 | ||
4775 | /* | |
4776 | * Update the inode delalloc counter now and wait to update the | |
4777 | * sb counters as we might have to borrow some blocks for the | |
4778 | * indirect block accounting. | |
4779 | */ | |
4fd29ec4 DW |
4780 | error = xfs_trans_reserve_quota_nblks(NULL, ip, |
4781 | -((long)del->br_blockcount), 0, | |
fa5c836c | 4782 | isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); |
4fd29ec4 DW |
4783 | if (error) |
4784 | return error; | |
fa5c836c CH |
4785 | ip->i_delayed_blks -= del->br_blockcount; |
4786 | ||
fa5c836c | 4787 | if (got->br_startoff == del->br_startoff) |
0173c689 | 4788 | state |= BMAP_LEFT_FILLING; |
fa5c836c | 4789 | if (got_endoff == del_endoff) |
0173c689 | 4790 | state |= BMAP_RIGHT_FILLING; |
fa5c836c | 4791 | |
0173c689 CH |
4792 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4793 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
fa5c836c CH |
4794 | /* |
4795 | * Matches the whole extent. Delete the entry. | |
4796 | */ | |
c38ccf59 | 4797 | xfs_iext_remove(ip, icur, state); |
b2b1712a | 4798 | xfs_iext_prev(ifp, icur); |
fa5c836c | 4799 | break; |
0173c689 | 4800 | case BMAP_LEFT_FILLING: |
fa5c836c CH |
4801 | /* |
4802 | * Deleting the first part of the extent. | |
4803 | */ | |
fa5c836c CH |
4804 | got->br_startoff = del_endoff; |
4805 | got->br_blockcount -= del->br_blockcount; | |
4806 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4807 | got->br_blockcount), da_old); | |
4808 | got->br_startblock = nullstartblock((int)da_new); | |
b2b1712a | 4809 | xfs_iext_update_extent(ip, state, icur, got); |
fa5c836c | 4810 | break; |
0173c689 | 4811 | case BMAP_RIGHT_FILLING: |
fa5c836c CH |
4812 | /* |
4813 | * Deleting the last part of the extent. | |
4814 | */ | |
fa5c836c CH |
4815 | got->br_blockcount = got->br_blockcount - del->br_blockcount; |
4816 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4817 | got->br_blockcount), da_old); | |
4818 | got->br_startblock = nullstartblock((int)da_new); | |
b2b1712a | 4819 | xfs_iext_update_extent(ip, state, icur, got); |
fa5c836c CH |
4820 | break; |
4821 | case 0: | |
4822 | /* | |
4823 | * Deleting the middle of the extent. | |
4824 | * | |
4825 | * Distribute the original indlen reservation across the two new | |
4826 | * extents. Steal blocks from the deleted extent if necessary. | |
4827 | * Stealing blocks simply fudges the fdblocks accounting below. | |
4828 | * Warn if either of the new indlen reservations is zero as this | |
4829 | * can lead to delalloc problems. | |
4830 | */ | |
fa5c836c CH |
4831 | got->br_blockcount = del->br_startoff - got->br_startoff; |
4832 | got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount); | |
4833 | ||
4834 | new.br_blockcount = got_endoff - del_endoff; | |
4835 | new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
4836 | ||
4837 | WARN_ON_ONCE(!got_indlen || !new_indlen); | |
4838 | stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen, | |
4839 | del->br_blockcount); | |
4840 | ||
4841 | got->br_startblock = nullstartblock((int)got_indlen); | |
fa5c836c CH |
4842 | |
4843 | new.br_startoff = del_endoff; | |
4844 | new.br_state = got->br_state; | |
4845 | new.br_startblock = nullstartblock((int)new_indlen); | |
4846 | ||
b2b1712a CH |
4847 | xfs_iext_update_extent(ip, state, icur, got); |
4848 | xfs_iext_next(ifp, icur); | |
0254c2f2 | 4849 | xfs_iext_insert(ip, icur, &new, state); |
fa5c836c CH |
4850 | |
4851 | da_new = got_indlen + new_indlen - stolen; | |
4852 | del->br_blockcount -= stolen; | |
4853 | break; | |
4854 | } | |
4855 | ||
4856 | ASSERT(da_old >= da_new); | |
4857 | da_diff = da_old - da_new; | |
4858 | if (!isrt) | |
4859 | da_diff += del->br_blockcount; | |
9fe82b8c | 4860 | if (da_diff) { |
fa5c836c | 4861 | xfs_mod_fdblocks(mp, da_diff, false); |
9fe82b8c DW |
4862 | xfs_mod_delalloc(mp, -da_diff); |
4863 | } | |
fa5c836c CH |
4864 | return error; |
4865 | } | |
4866 | ||
4867 | void | |
4868 | xfs_bmap_del_extent_cow( | |
4869 | struct xfs_inode *ip, | |
b2b1712a | 4870 | struct xfs_iext_cursor *icur, |
fa5c836c CH |
4871 | struct xfs_bmbt_irec *got, |
4872 | struct xfs_bmbt_irec *del) | |
4873 | { | |
4874 | struct xfs_mount *mp = ip->i_mount; | |
4875 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
4876 | struct xfs_bmbt_irec new; | |
4877 | xfs_fileoff_t del_endoff, got_endoff; | |
4878 | int state = BMAP_COWFORK; | |
4879 | ||
4880 | XFS_STATS_INC(mp, xs_del_exlist); | |
4881 | ||
4882 | del_endoff = del->br_startoff + del->br_blockcount; | |
4883 | got_endoff = got->br_startoff + got->br_blockcount; | |
4884 | ||
fa5c836c CH |
4885 | ASSERT(del->br_blockcount > 0); |
4886 | ASSERT(got->br_startoff <= del->br_startoff); | |
4887 | ASSERT(got_endoff >= del_endoff); | |
4888 | ASSERT(!isnullstartblock(got->br_startblock)); | |
4889 | ||
4890 | if (got->br_startoff == del->br_startoff) | |
0173c689 | 4891 | state |= BMAP_LEFT_FILLING; |
fa5c836c | 4892 | if (got_endoff == del_endoff) |
0173c689 | 4893 | state |= BMAP_RIGHT_FILLING; |
fa5c836c | 4894 | |
0173c689 CH |
4895 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { |
4896 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
fa5c836c CH |
4897 | /* |
4898 | * Matches the whole extent. Delete the entry. | |
4899 | */ | |
c38ccf59 | 4900 | xfs_iext_remove(ip, icur, state); |
b2b1712a | 4901 | xfs_iext_prev(ifp, icur); |
fa5c836c | 4902 | break; |
0173c689 | 4903 | case BMAP_LEFT_FILLING: |
fa5c836c CH |
4904 | /* |
4905 | * Deleting the first part of the extent. | |
4906 | */ | |
fa5c836c CH |
4907 | got->br_startoff = del_endoff; |
4908 | got->br_blockcount -= del->br_blockcount; | |
4909 | got->br_startblock = del->br_startblock + del->br_blockcount; | |
b2b1712a | 4910 | xfs_iext_update_extent(ip, state, icur, got); |
fa5c836c | 4911 | break; |
0173c689 | 4912 | case BMAP_RIGHT_FILLING: |
fa5c836c CH |
4913 | /* |
4914 | * Deleting the last part of the extent. | |
4915 | */ | |
fa5c836c | 4916 | got->br_blockcount -= del->br_blockcount; |
b2b1712a | 4917 | xfs_iext_update_extent(ip, state, icur, got); |
fa5c836c CH |
4918 | break; |
4919 | case 0: | |
4920 | /* | |
4921 | * Deleting the middle of the extent. | |
4922 | */ | |
fa5c836c | 4923 | got->br_blockcount = del->br_startoff - got->br_startoff; |
fa5c836c CH |
4924 | |
4925 | new.br_startoff = del_endoff; | |
4926 | new.br_blockcount = got_endoff - del_endoff; | |
4927 | new.br_state = got->br_state; | |
4928 | new.br_startblock = del->br_startblock + del->br_blockcount; | |
4929 | ||
b2b1712a CH |
4930 | xfs_iext_update_extent(ip, state, icur, got); |
4931 | xfs_iext_next(ifp, icur); | |
0254c2f2 | 4932 | xfs_iext_insert(ip, icur, &new, state); |
fa5c836c CH |
4933 | break; |
4934 | } | |
4b4c1326 | 4935 | ip->i_delayed_blks -= del->br_blockcount; |
fa5c836c CH |
4936 | } |
4937 | ||
9e5987a7 DC |
4938 | /* |
4939 | * Called by xfs_bmapi to update file extent records and the btree | |
e1d7553f | 4940 | * after removing space. |
9e5987a7 DC |
4941 | */ |
4942 | STATIC int /* error */ | |
e1d7553f | 4943 | xfs_bmap_del_extent_real( |
9e5987a7 DC |
4944 | xfs_inode_t *ip, /* incore inode pointer */ |
4945 | xfs_trans_t *tp, /* current transaction pointer */ | |
b2b1712a | 4946 | struct xfs_iext_cursor *icur, |
9e5987a7 DC |
4947 | xfs_btree_cur_t *cur, /* if null, not a btree */ |
4948 | xfs_bmbt_irec_t *del, /* data to remove from extents */ | |
4949 | int *logflagsp, /* inode logging flags */ | |
4847acf8 DW |
4950 | int whichfork, /* data or attr fork */ |
4951 | int bflags) /* bmapi flags */ | |
9e5987a7 | 4952 | { |
9e5987a7 DC |
4953 | xfs_fsblock_t del_endblock=0; /* first block past del */ |
4954 | xfs_fileoff_t del_endoff; /* first offset past del */ | |
9e5987a7 | 4955 | int do_fx; /* free extent at end of routine */ |
9e5987a7 | 4956 | int error; /* error return value */ |
1b24b633 | 4957 | int flags = 0;/* inode logging flags */ |
48fd52b1 | 4958 | struct xfs_bmbt_irec got; /* current extent entry */ |
9e5987a7 DC |
4959 | xfs_fileoff_t got_endoff; /* first offset past got */ |
4960 | int i; /* temp state */ | |
3ba738df | 4961 | struct xfs_ifork *ifp; /* inode fork pointer */ |
9e5987a7 DC |
4962 | xfs_mount_t *mp; /* mount structure */ |
4963 | xfs_filblks_t nblks; /* quota/sb block count */ | |
4964 | xfs_bmbt_irec_t new; /* new record to be inserted */ | |
4965 | /* REFERENCED */ | |
4966 | uint qfield; /* quota field to update */ | |
060ea65b | 4967 | int state = xfs_bmap_fork_to_state(whichfork); |
48fd52b1 | 4968 | struct xfs_bmbt_irec old; |
9e5987a7 | 4969 | |
ff6d6af2 BD |
4970 | mp = ip->i_mount; |
4971 | XFS_STATS_INC(mp, xs_del_exlist); | |
9e5987a7 | 4972 | |
9e5987a7 | 4973 | ifp = XFS_IFORK_PTR(ip, whichfork); |
9e5987a7 | 4974 | ASSERT(del->br_blockcount > 0); |
b2b1712a | 4975 | xfs_iext_get_extent(ifp, icur, &got); |
9e5987a7 DC |
4976 | ASSERT(got.br_startoff <= del->br_startoff); |
4977 | del_endoff = del->br_startoff + del->br_blockcount; | |
4978 | got_endoff = got.br_startoff + got.br_blockcount; | |
4979 | ASSERT(got_endoff >= del_endoff); | |
e1d7553f | 4980 | ASSERT(!isnullstartblock(got.br_startblock)); |
9e5987a7 DC |
4981 | qfield = 0; |
4982 | error = 0; | |
e1d7553f | 4983 | |
1b24b633 CH |
4984 | /* |
4985 | * If it's the case where the directory code is running with no block | |
4986 | * reservation, and the deleted block is in the middle of its extent, | |
4987 | * and the resulting insert of an extent would cause transformation to | |
4988 | * btree format, then reject it. The calling code will then swap blocks | |
4989 | * around instead. We have to do this now, rather than waiting for the | |
4990 | * conversion to btree format, since the transaction will be dirty then. | |
4991 | */ | |
4992 | if (tp->t_blk_res == 0 && | |
4993 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
4994 | XFS_IFORK_NEXTENTS(ip, whichfork) >= | |
4995 | XFS_IFORK_MAXEXT(ip, whichfork) && | |
4996 | del->br_startoff > got.br_startoff && del_endoff < got_endoff) | |
4997 | return -ENOSPC; | |
4998 | ||
4999 | flags = XFS_ILOG_CORE; | |
e1d7553f CH |
5000 | if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) { |
5001 | xfs_fsblock_t bno; | |
5002 | xfs_filblks_t len; | |
0703a8e1 DC |
5003 | xfs_extlen_t mod; |
5004 | ||
5005 | bno = div_u64_rem(del->br_startblock, mp->m_sb.sb_rextsize, | |
5006 | &mod); | |
5007 | ASSERT(mod == 0); | |
5008 | len = div_u64_rem(del->br_blockcount, mp->m_sb.sb_rextsize, | |
5009 | &mod); | |
5010 | ASSERT(mod == 0); | |
e1d7553f | 5011 | |
e1d7553f CH |
5012 | error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len); |
5013 | if (error) | |
5014 | goto done; | |
9e5987a7 | 5015 | do_fx = 0; |
e1d7553f CH |
5016 | nblks = len * mp->m_sb.sb_rextsize; |
5017 | qfield = XFS_TRANS_DQ_RTBCOUNT; | |
5018 | } else { | |
5019 | do_fx = 1; | |
5020 | nblks = del->br_blockcount; | |
5021 | qfield = XFS_TRANS_DQ_BCOUNT; | |
5022 | } | |
5023 | ||
5024 | del_endblock = del->br_startblock + del->br_blockcount; | |
5025 | if (cur) { | |
e16cf9b0 | 5026 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
e1d7553f CH |
5027 | if (error) |
5028 | goto done; | |
5029 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); | |
1da177e4 | 5030 | } |
340785cc | 5031 | |
491f6f8a CH |
5032 | if (got.br_startoff == del->br_startoff) |
5033 | state |= BMAP_LEFT_FILLING; | |
5034 | if (got_endoff == del_endoff) | |
5035 | state |= BMAP_RIGHT_FILLING; | |
5036 | ||
5037 | switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) { | |
5038 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
9e5987a7 DC |
5039 | /* |
5040 | * Matches the whole extent. Delete the entry. | |
5041 | */ | |
c38ccf59 | 5042 | xfs_iext_remove(ip, icur, state); |
b2b1712a | 5043 | xfs_iext_prev(ifp, icur); |
9e5987a7 DC |
5044 | XFS_IFORK_NEXT_SET(ip, whichfork, |
5045 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5046 | flags |= XFS_ILOG_CORE; | |
5047 | if (!cur) { | |
5048 | flags |= xfs_ilog_fext(whichfork); | |
5049 | break; | |
5050 | } | |
5051 | if ((error = xfs_btree_delete(cur, &i))) | |
5052 | goto done; | |
c29aad41 | 5053 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
9e5987a7 | 5054 | break; |
491f6f8a | 5055 | case BMAP_LEFT_FILLING: |
9e5987a7 DC |
5056 | /* |
5057 | * Deleting the first part of the extent. | |
5058 | */ | |
48fd52b1 CH |
5059 | got.br_startoff = del_endoff; |
5060 | got.br_startblock = del_endblock; | |
5061 | got.br_blockcount -= del->br_blockcount; | |
b2b1712a | 5062 | xfs_iext_update_extent(ip, state, icur, &got); |
9e5987a7 DC |
5063 | if (!cur) { |
5064 | flags |= xfs_ilog_fext(whichfork); | |
5065 | break; | |
5066 | } | |
a67d00a5 | 5067 | error = xfs_bmbt_update(cur, &got); |
48fd52b1 | 5068 | if (error) |
9e5987a7 DC |
5069 | goto done; |
5070 | break; | |
491f6f8a | 5071 | case BMAP_RIGHT_FILLING: |
9e5987a7 DC |
5072 | /* |
5073 | * Deleting the last part of the extent. | |
5074 | */ | |
48fd52b1 | 5075 | got.br_blockcount -= del->br_blockcount; |
b2b1712a | 5076 | xfs_iext_update_extent(ip, state, icur, &got); |
9e5987a7 DC |
5077 | if (!cur) { |
5078 | flags |= xfs_ilog_fext(whichfork); | |
5079 | break; | |
5080 | } | |
a67d00a5 | 5081 | error = xfs_bmbt_update(cur, &got); |
48fd52b1 | 5082 | if (error) |
9e5987a7 DC |
5083 | goto done; |
5084 | break; | |
9e5987a7 DC |
5085 | case 0: |
5086 | /* | |
5087 | * Deleting the middle of the extent. | |
5088 | */ | |
48fd52b1 | 5089 | old = got; |
ca5d8e5b | 5090 | |
48fd52b1 | 5091 | got.br_blockcount = del->br_startoff - got.br_startoff; |
b2b1712a | 5092 | xfs_iext_update_extent(ip, state, icur, &got); |
48fd52b1 | 5093 | |
9e5987a7 | 5094 | new.br_startoff = del_endoff; |
48fd52b1 | 5095 | new.br_blockcount = got_endoff - del_endoff; |
9e5987a7 | 5096 | new.br_state = got.br_state; |
e1d7553f | 5097 | new.br_startblock = del_endblock; |
48fd52b1 | 5098 | |
e1d7553f CH |
5099 | flags |= XFS_ILOG_CORE; |
5100 | if (cur) { | |
a67d00a5 | 5101 | error = xfs_bmbt_update(cur, &got); |
e1d7553f CH |
5102 | if (error) |
5103 | goto done; | |
5104 | error = xfs_btree_increment(cur, 0, &i); | |
5105 | if (error) | |
5106 | goto done; | |
5107 | cur->bc_rec.b = new; | |
5108 | error = xfs_btree_insert(cur, &i); | |
5109 | if (error && error != -ENOSPC) | |
5110 | goto done; | |
5111 | /* | |
5112 | * If get no-space back from btree insert, it tried a | |
5113 | * split, and we have a zero block reservation. Fix up | |
5114 | * our state and return the error. | |
5115 | */ | |
5116 | if (error == -ENOSPC) { | |
9e5987a7 | 5117 | /* |
e1d7553f CH |
5118 | * Reset the cursor, don't trust it after any |
5119 | * insert operation. | |
9e5987a7 | 5120 | */ |
e16cf9b0 | 5121 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
e1d7553f | 5122 | if (error) |
9e5987a7 | 5123 | goto done; |
c29aad41 | 5124 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
e1d7553f CH |
5125 | /* |
5126 | * Update the btree record back | |
5127 | * to the original value. | |
5128 | */ | |
a67d00a5 | 5129 | error = xfs_bmbt_update(cur, &old); |
e1d7553f CH |
5130 | if (error) |
5131 | goto done; | |
5132 | /* | |
5133 | * Reset the extent record back | |
5134 | * to the original value. | |
5135 | */ | |
b2b1712a | 5136 | xfs_iext_update_extent(ip, state, icur, &old); |
e1d7553f CH |
5137 | flags = 0; |
5138 | error = -ENOSPC; | |
5139 | goto done; | |
5140 | } | |
5141 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); | |
5142 | } else | |
5143 | flags |= xfs_ilog_fext(whichfork); | |
5144 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5145 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
b2b1712a | 5146 | xfs_iext_next(ifp, icur); |
0254c2f2 | 5147 | xfs_iext_insert(ip, icur, &new, state); |
9e5987a7 | 5148 | break; |
1da177e4 | 5149 | } |
9c194644 DW |
5150 | |
5151 | /* remove reverse mapping */ | |
0f37d178 | 5152 | error = xfs_rmap_unmap_extent(tp, ip, whichfork, del); |
e1d7553f CH |
5153 | if (error) |
5154 | goto done; | |
9c194644 | 5155 | |
1da177e4 | 5156 | /* |
9e5987a7 | 5157 | * If we need to, add to list of extents to delete. |
1da177e4 | 5158 | */ |
4847acf8 | 5159 | if (do_fx && !(bflags & XFS_BMAPI_REMAP)) { |
62aab20f | 5160 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) { |
0f37d178 | 5161 | error = xfs_refcount_decrease_extent(tp, del); |
62aab20f DW |
5162 | if (error) |
5163 | goto done; | |
fcb762f5 | 5164 | } else { |
0f37d178 | 5165 | __xfs_bmap_add_free(tp, del->br_startblock, |
4e529339 BF |
5166 | del->br_blockcount, NULL, |
5167 | (bflags & XFS_BMAPI_NODISCARD) || | |
5168 | del->br_state == XFS_EXT_UNWRITTEN); | |
fcb762f5 | 5169 | } |
62aab20f DW |
5170 | } |
5171 | ||
1da177e4 | 5172 | /* |
9e5987a7 | 5173 | * Adjust inode # blocks in the file. |
1da177e4 | 5174 | */ |
9e5987a7 DC |
5175 | if (nblks) |
5176 | ip->i_d.di_nblocks -= nblks; | |
1da177e4 | 5177 | /* |
9e5987a7 | 5178 | * Adjust quota data. |
1da177e4 | 5179 | */ |
4847acf8 | 5180 | if (qfield && !(bflags & XFS_BMAPI_REMAP)) |
9e5987a7 DC |
5181 | xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks); |
5182 | ||
9e5987a7 DC |
5183 | done: |
5184 | *logflagsp = flags; | |
1da177e4 LT |
5185 | return error; |
5186 | } | |
5187 | ||
3bacbcd8 | 5188 | /* |
9e5987a7 DC |
5189 | * Unmap (remove) blocks from a file. |
5190 | * If nexts is nonzero then the number of extents to remove is limited to | |
5191 | * that value. If not all extents in the block range can be removed then | |
5192 | * *done is set. | |
3bacbcd8 | 5193 | */ |
9e5987a7 | 5194 | int /* error */ |
4453593b | 5195 | __xfs_bunmapi( |
ccd9d911 | 5196 | struct xfs_trans *tp, /* transaction pointer */ |
9e5987a7 | 5197 | struct xfs_inode *ip, /* incore inode */ |
8280f6ed | 5198 | xfs_fileoff_t start, /* first file offset deleted */ |
4453593b | 5199 | xfs_filblks_t *rlen, /* i/o: amount remaining */ |
9e5987a7 | 5200 | int flags, /* misc flags */ |
2af52842 | 5201 | xfs_extnum_t nexts) /* number of extents max */ |
3bacbcd8 | 5202 | { |
ccd9d911 BF |
5203 | struct xfs_btree_cur *cur; /* bmap btree cursor */ |
5204 | struct xfs_bmbt_irec del; /* extent being deleted */ | |
9e5987a7 DC |
5205 | int error; /* error return value */ |
5206 | xfs_extnum_t extno; /* extent number in list */ | |
ccd9d911 | 5207 | struct xfs_bmbt_irec got; /* current extent record */ |
3ba738df | 5208 | struct xfs_ifork *ifp; /* inode fork pointer */ |
9e5987a7 | 5209 | int isrt; /* freeing in rt area */ |
9e5987a7 DC |
5210 | int logflags; /* transaction logging flags */ |
5211 | xfs_extlen_t mod; /* rt extent offset */ | |
ccd9d911 | 5212 | struct xfs_mount *mp; /* mount structure */ |
9e5987a7 DC |
5213 | int tmp_logflags; /* partial logging flags */ |
5214 | int wasdel; /* was a delayed alloc extent */ | |
5215 | int whichfork; /* data or attribute fork */ | |
5216 | xfs_fsblock_t sum; | |
4453593b | 5217 | xfs_filblks_t len = *rlen; /* length to unmap in file */ |
e1a4e37c | 5218 | xfs_fileoff_t max_len; |
5b094d6d | 5219 | xfs_agnumber_t prev_agno = NULLAGNUMBER, agno; |
8280f6ed | 5220 | xfs_fileoff_t end; |
b2b1712a CH |
5221 | struct xfs_iext_cursor icur; |
5222 | bool done = false; | |
1da177e4 | 5223 | |
8280f6ed | 5224 | trace_xfs_bunmap(ip, start, len, flags, _RET_IP_); |
1da177e4 | 5225 | |
3993baeb DW |
5226 | whichfork = xfs_bmapi_whichfork(flags); |
5227 | ASSERT(whichfork != XFS_COW_FORK); | |
9e5987a7 DC |
5228 | ifp = XFS_IFORK_PTR(ip, whichfork); |
5229 | if (unlikely( | |
5230 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5231 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
5232 | XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW, | |
5233 | ip->i_mount); | |
2451337d | 5234 | return -EFSCORRUPTED; |
1da177e4 | 5235 | } |
9e5987a7 DC |
5236 | mp = ip->i_mount; |
5237 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 5238 | return -EIO; |
1da177e4 | 5239 | |
eef334e5 | 5240 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
9e5987a7 DC |
5241 | ASSERT(len > 0); |
5242 | ASSERT(nexts >= 0); | |
1da177e4 | 5243 | |
e1a4e37c DW |
5244 | /* |
5245 | * Guesstimate how many blocks we can unmap without running the risk of | |
5246 | * blowing out the transaction with a mix of EFIs and reflink | |
5247 | * adjustments. | |
5248 | */ | |
8c57b886 | 5249 | if (tp && xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) |
e1a4e37c DW |
5250 | max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res)); |
5251 | else | |
5252 | max_len = len; | |
5253 | ||
9e5987a7 DC |
5254 | if (!(ifp->if_flags & XFS_IFEXTENTS) && |
5255 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
5256 | return error; | |
5d829300 | 5257 | if (xfs_iext_count(ifp) == 0) { |
4453593b | 5258 | *rlen = 0; |
9e5987a7 DC |
5259 | return 0; |
5260 | } | |
ff6d6af2 | 5261 | XFS_STATS_INC(mp, xs_blk_unmap); |
9e5987a7 | 5262 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); |
dc56015f | 5263 | end = start + len; |
1da177e4 | 5264 | |
b2b1712a | 5265 | if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) { |
dc56015f CH |
5266 | *rlen = 0; |
5267 | return 0; | |
9e5987a7 | 5268 | } |
dc56015f | 5269 | end--; |
7efc7945 | 5270 | |
9e5987a7 DC |
5271 | logflags = 0; |
5272 | if (ifp->if_flags & XFS_IFBROOT) { | |
5273 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
5274 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
9e5987a7 DC |
5275 | cur->bc_private.b.flags = 0; |
5276 | } else | |
5277 | cur = NULL; | |
5278 | ||
5279 | if (isrt) { | |
5280 | /* | |
5281 | * Synchronize by locking the bitmap inode. | |
5282 | */ | |
f4a0660d | 5283 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
9e5987a7 | 5284 | xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
f4a0660d DW |
5285 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
5286 | xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL); | |
9e5987a7 | 5287 | } |
58e20770 | 5288 | |
9e5987a7 | 5289 | extno = 0; |
b2b1712a | 5290 | while (end != (xfs_fileoff_t)-1 && end >= start && |
e1a4e37c | 5291 | (nexts == 0 || extno < nexts) && max_len > 0) { |
9e5987a7 | 5292 | /* |
8280f6ed | 5293 | * Is the found extent after a hole in which end lives? |
9e5987a7 DC |
5294 | * Just back up to the previous extent, if so. |
5295 | */ | |
b2b1712a CH |
5296 | if (got.br_startoff > end && |
5297 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5298 | done = true; | |
5299 | break; | |
9e5987a7 DC |
5300 | } |
5301 | /* | |
5302 | * Is the last block of this extent before the range | |
5303 | * we're supposed to delete? If so, we're done. | |
5304 | */ | |
8280f6ed | 5305 | end = XFS_FILEOFF_MIN(end, |
9e5987a7 | 5306 | got.br_startoff + got.br_blockcount - 1); |
8280f6ed | 5307 | if (end < start) |
9e5987a7 DC |
5308 | break; |
5309 | /* | |
5310 | * Then deal with the (possibly delayed) allocated space | |
5311 | * we found. | |
5312 | */ | |
9e5987a7 DC |
5313 | del = got; |
5314 | wasdel = isnullstartblock(del.br_startblock); | |
5b094d6d CH |
5315 | |
5316 | /* | |
5317 | * Make sure we don't touch multiple AGF headers out of order | |
5318 | * in a single transaction, as that could cause AB-BA deadlocks. | |
5319 | */ | |
5320 | if (!wasdel) { | |
5321 | agno = XFS_FSB_TO_AGNO(mp, del.br_startblock); | |
5322 | if (prev_agno != NULLAGNUMBER && prev_agno > agno) | |
5323 | break; | |
5324 | prev_agno = agno; | |
5325 | } | |
9e5987a7 DC |
5326 | if (got.br_startoff < start) { |
5327 | del.br_startoff = start; | |
5328 | del.br_blockcount -= start - got.br_startoff; | |
5329 | if (!wasdel) | |
5330 | del.br_startblock += start - got.br_startoff; | |
5331 | } | |
8280f6ed CH |
5332 | if (del.br_startoff + del.br_blockcount > end + 1) |
5333 | del.br_blockcount = end + 1 - del.br_startoff; | |
e1a4e37c DW |
5334 | |
5335 | /* How much can we safely unmap? */ | |
5336 | if (max_len < del.br_blockcount) { | |
5337 | del.br_startoff += del.br_blockcount - max_len; | |
5338 | if (!wasdel) | |
5339 | del.br_startblock += del.br_blockcount - max_len; | |
5340 | del.br_blockcount = max_len; | |
5341 | } | |
5342 | ||
0703a8e1 DC |
5343 | if (!isrt) |
5344 | goto delete; | |
5345 | ||
9e5987a7 | 5346 | sum = del.br_startblock + del.br_blockcount; |
0703a8e1 DC |
5347 | div_u64_rem(sum, mp->m_sb.sb_rextsize, &mod); |
5348 | if (mod) { | |
58e20770 | 5349 | /* |
9e5987a7 DC |
5350 | * Realtime extent not lined up at the end. |
5351 | * The extent could have been split into written | |
5352 | * and unwritten pieces, or we could just be | |
5353 | * unmapping part of it. But we can't really | |
5354 | * get rid of part of a realtime extent. | |
58e20770 | 5355 | */ |
daa79bae | 5356 | if (del.br_state == XFS_EXT_UNWRITTEN) { |
9e5987a7 DC |
5357 | /* |
5358 | * This piece is unwritten, or we're not | |
5359 | * using unwritten extents. Skip over it. | |
5360 | */ | |
8280f6ed CH |
5361 | ASSERT(end >= mod); |
5362 | end -= mod > del.br_blockcount ? | |
9e5987a7 | 5363 | del.br_blockcount : mod; |
b2b1712a CH |
5364 | if (end < got.br_startoff && |
5365 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5366 | done = true; | |
5367 | break; | |
9e5987a7 DC |
5368 | } |
5369 | continue; | |
1da177e4 | 5370 | } |
9af25465 | 5371 | /* |
9e5987a7 DC |
5372 | * It's written, turn it unwritten. |
5373 | * This is better than zeroing it. | |
9af25465 | 5374 | */ |
9e5987a7 | 5375 | ASSERT(del.br_state == XFS_EXT_NORM); |
a7e5d03b | 5376 | ASSERT(tp->t_blk_res > 0); |
9e5987a7 DC |
5377 | /* |
5378 | * If this spans a realtime extent boundary, | |
5379 | * chop it back to the start of the one we end at. | |
5380 | */ | |
5381 | if (del.br_blockcount > mod) { | |
5382 | del.br_startoff += del.br_blockcount - mod; | |
5383 | del.br_startblock += del.br_blockcount - mod; | |
5384 | del.br_blockcount = mod; | |
5385 | } | |
5386 | del.br_state = XFS_EXT_UNWRITTEN; | |
5387 | error = xfs_bmap_add_extent_unwritten_real(tp, ip, | |
b2b1712a | 5388 | whichfork, &icur, &cur, &del, |
92f9da30 | 5389 | &logflags); |
9e5987a7 DC |
5390 | if (error) |
5391 | goto error0; | |
5392 | goto nodelete; | |
5393 | } | |
0703a8e1 DC |
5394 | div_u64_rem(del.br_startblock, mp->m_sb.sb_rextsize, &mod); |
5395 | if (mod) { | |
9e5987a7 DC |
5396 | /* |
5397 | * Realtime extent is lined up at the end but not | |
5398 | * at the front. We'll get rid of full extents if | |
5399 | * we can. | |
5400 | */ | |
5401 | mod = mp->m_sb.sb_rextsize - mod; | |
5402 | if (del.br_blockcount > mod) { | |
5403 | del.br_blockcount -= mod; | |
5404 | del.br_startoff += mod; | |
5405 | del.br_startblock += mod; | |
daa79bae CH |
5406 | } else if (del.br_startoff == start && |
5407 | (del.br_state == XFS_EXT_UNWRITTEN || | |
5408 | tp->t_blk_res == 0)) { | |
9e5987a7 DC |
5409 | /* |
5410 | * Can't make it unwritten. There isn't | |
5411 | * a full extent here so just skip it. | |
5412 | */ | |
8280f6ed CH |
5413 | ASSERT(end >= del.br_blockcount); |
5414 | end -= del.br_blockcount; | |
b2b1712a CH |
5415 | if (got.br_startoff > end && |
5416 | !xfs_iext_prev_extent(ifp, &icur, &got)) { | |
5417 | done = true; | |
5418 | break; | |
5419 | } | |
9af25465 | 5420 | continue; |
9e5987a7 | 5421 | } else if (del.br_state == XFS_EXT_UNWRITTEN) { |
7efc7945 CH |
5422 | struct xfs_bmbt_irec prev; |
5423 | ||
9e5987a7 DC |
5424 | /* |
5425 | * This one is already unwritten. | |
5426 | * It must have a written left neighbor. | |
5427 | * Unwrite the killed part of that one and | |
5428 | * try again. | |
5429 | */ | |
b2b1712a CH |
5430 | if (!xfs_iext_prev_extent(ifp, &icur, &prev)) |
5431 | ASSERT(0); | |
9e5987a7 DC |
5432 | ASSERT(prev.br_state == XFS_EXT_NORM); |
5433 | ASSERT(!isnullstartblock(prev.br_startblock)); | |
5434 | ASSERT(del.br_startblock == | |
5435 | prev.br_startblock + prev.br_blockcount); | |
5436 | if (prev.br_startoff < start) { | |
5437 | mod = start - prev.br_startoff; | |
5438 | prev.br_blockcount -= mod; | |
5439 | prev.br_startblock += mod; | |
5440 | prev.br_startoff = start; | |
5441 | } | |
5442 | prev.br_state = XFS_EXT_UNWRITTEN; | |
9e5987a7 | 5443 | error = xfs_bmap_add_extent_unwritten_real(tp, |
b2b1712a | 5444 | ip, whichfork, &icur, &cur, |
92f9da30 | 5445 | &prev, &logflags); |
9e5987a7 DC |
5446 | if (error) |
5447 | goto error0; | |
5448 | goto nodelete; | |
5449 | } else { | |
5450 | ASSERT(del.br_state == XFS_EXT_NORM); | |
5451 | del.br_state = XFS_EXT_UNWRITTEN; | |
5452 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
b2b1712a | 5453 | ip, whichfork, &icur, &cur, |
92f9da30 | 5454 | &del, &logflags); |
9e5987a7 DC |
5455 | if (error) |
5456 | goto error0; | |
5457 | goto nodelete; | |
9af25465 | 5458 | } |
1da177e4 | 5459 | } |
1da177e4 | 5460 | |
0703a8e1 | 5461 | delete: |
b2706a05 | 5462 | if (wasdel) { |
b2b1712a | 5463 | error = xfs_bmap_del_extent_delay(ip, whichfork, &icur, |
e1d7553f CH |
5464 | &got, &del); |
5465 | } else { | |
81ba8f3e BF |
5466 | error = xfs_bmap_del_extent_real(ip, tp, &icur, cur, |
5467 | &del, &tmp_logflags, whichfork, | |
e1d7553f CH |
5468 | flags); |
5469 | logflags |= tmp_logflags; | |
b213d692 | 5470 | } |
b2706a05 | 5471 | |
9e5987a7 DC |
5472 | if (error) |
5473 | goto error0; | |
b2706a05 | 5474 | |
e1a4e37c | 5475 | max_len -= del.br_blockcount; |
8280f6ed | 5476 | end = del.br_startoff - 1; |
9e5987a7 | 5477 | nodelete: |
1da177e4 | 5478 | /* |
9e5987a7 | 5479 | * If not done go on to the next (previous) record. |
1da177e4 | 5480 | */ |
8280f6ed | 5481 | if (end != (xfs_fileoff_t)-1 && end >= start) { |
b2b1712a CH |
5482 | if (!xfs_iext_get_extent(ifp, &icur, &got) || |
5483 | (got.br_startoff > end && | |
5484 | !xfs_iext_prev_extent(ifp, &icur, &got))) { | |
5485 | done = true; | |
5486 | break; | |
1da177e4 | 5487 | } |
9e5987a7 | 5488 | extno++; |
1da177e4 LT |
5489 | } |
5490 | } | |
b2b1712a | 5491 | if (done || end == (xfs_fileoff_t)-1 || end < start) |
4453593b DW |
5492 | *rlen = 0; |
5493 | else | |
8280f6ed | 5494 | *rlen = end - start + 1; |
576039cf | 5495 | |
1da177e4 | 5496 | /* |
9e5987a7 | 5497 | * Convert to a btree if necessary. |
1da177e4 | 5498 | */ |
9e5987a7 DC |
5499 | if (xfs_bmap_needs_btree(ip, whichfork)) { |
5500 | ASSERT(cur == NULL); | |
280253d2 BF |
5501 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
5502 | &tmp_logflags, whichfork); | |
9e5987a7 | 5503 | logflags |= tmp_logflags; |
b101e334 CH |
5504 | } else { |
5505 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, | |
9e5987a7 | 5506 | whichfork); |
9e5987a7 | 5507 | } |
b101e334 | 5508 | |
9e5987a7 DC |
5509 | error0: |
5510 | /* | |
5511 | * Log everything. Do this after conversion, there's no point in | |
5512 | * logging the extent records if we've converted to btree format. | |
5513 | */ | |
5514 | if ((logflags & xfs_ilog_fext(whichfork)) && | |
5515 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
5516 | logflags &= ~xfs_ilog_fext(whichfork); | |
5517 | else if ((logflags & xfs_ilog_fbroot(whichfork)) && | |
5518 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) | |
5519 | logflags &= ~xfs_ilog_fbroot(whichfork); | |
5520 | /* | |
5521 | * Log inode even in the error case, if the transaction | |
5522 | * is dirty we'll need to shut down the filesystem. | |
5523 | */ | |
5524 | if (logflags) | |
5525 | xfs_trans_log_inode(tp, ip, logflags); | |
5526 | if (cur) { | |
cf612de7 | 5527 | if (!error) |
9e5987a7 | 5528 | cur->bc_private.b.allocated = 0; |
0b04b6b8 | 5529 | xfs_btree_del_cursor(cur, error); |
9e5987a7 DC |
5530 | } |
5531 | return error; | |
5532 | } | |
e1d8fb88 | 5533 | |
4453593b DW |
5534 | /* Unmap a range of a file. */ |
5535 | int | |
5536 | xfs_bunmapi( | |
5537 | xfs_trans_t *tp, | |
5538 | struct xfs_inode *ip, | |
5539 | xfs_fileoff_t bno, | |
5540 | xfs_filblks_t len, | |
5541 | int flags, | |
5542 | xfs_extnum_t nexts, | |
4453593b DW |
5543 | int *done) |
5544 | { | |
5545 | int error; | |
5546 | ||
2af52842 | 5547 | error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts); |
4453593b DW |
5548 | *done = (len == 0); |
5549 | return error; | |
5550 | } | |
5551 | ||
ddb19e31 BF |
5552 | /* |
5553 | * Determine whether an extent shift can be accomplished by a merge with the | |
5554 | * extent that precedes the target hole of the shift. | |
5555 | */ | |
5556 | STATIC bool | |
5557 | xfs_bmse_can_merge( | |
5558 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5559 | struct xfs_bmbt_irec *got, /* current extent to shift */ | |
5560 | xfs_fileoff_t shift) /* shift fsb */ | |
5561 | { | |
5562 | xfs_fileoff_t startoff; | |
5563 | ||
5564 | startoff = got->br_startoff - shift; | |
5565 | ||
5566 | /* | |
5567 | * The extent, once shifted, must be adjacent in-file and on-disk with | |
5568 | * the preceding extent. | |
5569 | */ | |
5570 | if ((left->br_startoff + left->br_blockcount != startoff) || | |
5571 | (left->br_startblock + left->br_blockcount != got->br_startblock) || | |
5572 | (left->br_state != got->br_state) || | |
5573 | (left->br_blockcount + got->br_blockcount > MAXEXTLEN)) | |
5574 | return false; | |
5575 | ||
5576 | return true; | |
5577 | } | |
5578 | ||
5579 | /* | |
5580 | * A bmap extent shift adjusts the file offset of an extent to fill a preceding | |
5581 | * hole in the file. If an extent shift would result in the extent being fully | |
5582 | * adjacent to the extent that currently precedes the hole, we can merge with | |
5583 | * the preceding extent rather than do the shift. | |
5584 | * | |
5585 | * This function assumes the caller has verified a shift-by-merge is possible | |
5586 | * with the provided extents via xfs_bmse_can_merge(). | |
5587 | */ | |
5588 | STATIC int | |
5589 | xfs_bmse_merge( | |
0f37d178 | 5590 | struct xfs_trans *tp, |
ddb19e31 BF |
5591 | struct xfs_inode *ip, |
5592 | int whichfork, | |
5593 | xfs_fileoff_t shift, /* shift fsb */ | |
b2b1712a | 5594 | struct xfs_iext_cursor *icur, |
4da6b514 CH |
5595 | struct xfs_bmbt_irec *got, /* extent to shift */ |
5596 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
ddb19e31 | 5597 | struct xfs_btree_cur *cur, |
0f37d178 | 5598 | int *logflags) /* output */ |
ddb19e31 | 5599 | { |
4da6b514 | 5600 | struct xfs_bmbt_irec new; |
ddb19e31 BF |
5601 | xfs_filblks_t blockcount; |
5602 | int error, i; | |
5fb5aeee | 5603 | struct xfs_mount *mp = ip->i_mount; |
ddb19e31 | 5604 | |
4da6b514 | 5605 | blockcount = left->br_blockcount + got->br_blockcount; |
ddb19e31 BF |
5606 | |
5607 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5608 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
4da6b514 | 5609 | ASSERT(xfs_bmse_can_merge(left, got, shift)); |
ddb19e31 | 5610 | |
4da6b514 CH |
5611 | new = *left; |
5612 | new.br_blockcount = blockcount; | |
ddb19e31 BF |
5613 | |
5614 | /* | |
5615 | * Update the on-disk extent count, the btree if necessary and log the | |
5616 | * inode. | |
5617 | */ | |
5618 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5619 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5620 | *logflags |= XFS_ILOG_CORE; | |
5621 | if (!cur) { | |
5622 | *logflags |= XFS_ILOG_DEXT; | |
4da6b514 | 5623 | goto done; |
ddb19e31 BF |
5624 | } |
5625 | ||
5626 | /* lookup and remove the extent to merge */ | |
e16cf9b0 | 5627 | error = xfs_bmbt_lookup_eq(cur, got, &i); |
ddb19e31 | 5628 | if (error) |
4db431f5 | 5629 | return error; |
5fb5aeee | 5630 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
ddb19e31 BF |
5631 | |
5632 | error = xfs_btree_delete(cur, &i); | |
5633 | if (error) | |
4db431f5 | 5634 | return error; |
5fb5aeee | 5635 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
ddb19e31 BF |
5636 | |
5637 | /* lookup and update size of the previous extent */ | |
e16cf9b0 | 5638 | error = xfs_bmbt_lookup_eq(cur, left, &i); |
ddb19e31 | 5639 | if (error) |
4db431f5 | 5640 | return error; |
5fb5aeee | 5641 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
ddb19e31 | 5642 | |
a67d00a5 | 5643 | error = xfs_bmbt_update(cur, &new); |
4da6b514 CH |
5644 | if (error) |
5645 | return error; | |
5646 | ||
5647 | done: | |
c38ccf59 | 5648 | xfs_iext_remove(ip, icur, 0); |
b2b1712a CH |
5649 | xfs_iext_prev(XFS_IFORK_PTR(ip, whichfork), icur); |
5650 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur, | |
5651 | &new); | |
ddb19e31 | 5652 | |
4cc1ee5e | 5653 | /* update reverse mapping. rmap functions merge the rmaps for us */ |
0f37d178 | 5654 | error = xfs_rmap_unmap_extent(tp, ip, whichfork, got); |
4da6b514 CH |
5655 | if (error) |
5656 | return error; | |
4cc1ee5e DW |
5657 | memcpy(&new, got, sizeof(new)); |
5658 | new.br_startoff = left->br_startoff + left->br_blockcount; | |
0f37d178 | 5659 | return xfs_rmap_map_extent(tp, ip, whichfork, &new); |
ddb19e31 BF |
5660 | } |
5661 | ||
bf806280 CH |
5662 | static int |
5663 | xfs_bmap_shift_update_extent( | |
0f37d178 | 5664 | struct xfs_trans *tp, |
bf806280 CH |
5665 | struct xfs_inode *ip, |
5666 | int whichfork, | |
b2b1712a | 5667 | struct xfs_iext_cursor *icur, |
bf806280 CH |
5668 | struct xfs_bmbt_irec *got, |
5669 | struct xfs_btree_cur *cur, | |
5670 | int *logflags, | |
bf806280 | 5671 | xfs_fileoff_t startoff) |
a979bdfe | 5672 | { |
bf806280 | 5673 | struct xfs_mount *mp = ip->i_mount; |
11f75b3b | 5674 | struct xfs_bmbt_irec prev = *got; |
bf806280 | 5675 | int error, i; |
4da6b514 | 5676 | |
a979bdfe | 5677 | *logflags |= XFS_ILOG_CORE; |
4da6b514 | 5678 | |
11f75b3b | 5679 | got->br_startoff = startoff; |
4da6b514 CH |
5680 | |
5681 | if (cur) { | |
11f75b3b | 5682 | error = xfs_bmbt_lookup_eq(cur, &prev, &i); |
4da6b514 CH |
5683 | if (error) |
5684 | return error; | |
5685 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); | |
5686 | ||
11f75b3b | 5687 | error = xfs_bmbt_update(cur, got); |
4da6b514 CH |
5688 | if (error) |
5689 | return error; | |
5690 | } else { | |
a979bdfe | 5691 | *logflags |= XFS_ILOG_DEXT; |
a979bdfe BF |
5692 | } |
5693 | ||
b2b1712a CH |
5694 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur, |
5695 | got); | |
9c194644 | 5696 | |
9c194644 | 5697 | /* update reverse mapping */ |
0f37d178 | 5698 | error = xfs_rmap_unmap_extent(tp, ip, whichfork, &prev); |
9c194644 DW |
5699 | if (error) |
5700 | return error; | |
0f37d178 | 5701 | return xfs_rmap_map_extent(tp, ip, whichfork, got); |
a979bdfe BF |
5702 | } |
5703 | ||
e1d8fb88 | 5704 | int |
ecfea3f0 | 5705 | xfs_bmap_collapse_extents( |
e1d8fb88 NJ |
5706 | struct xfs_trans *tp, |
5707 | struct xfs_inode *ip, | |
a904b1ca | 5708 | xfs_fileoff_t *next_fsb, |
e1d8fb88 | 5709 | xfs_fileoff_t offset_shift_fsb, |
333f950c | 5710 | bool *done) |
e1d8fb88 | 5711 | { |
ecfea3f0 CH |
5712 | int whichfork = XFS_DATA_FORK; |
5713 | struct xfs_mount *mp = ip->i_mount; | |
5714 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5715 | struct xfs_btree_cur *cur = NULL; | |
bf806280 | 5716 | struct xfs_bmbt_irec got, prev; |
b2b1712a | 5717 | struct xfs_iext_cursor icur; |
bf806280 | 5718 | xfs_fileoff_t new_startoff; |
ecfea3f0 CH |
5719 | int error = 0; |
5720 | int logflags = 0; | |
e1d8fb88 NJ |
5721 | |
5722 | if (unlikely(XFS_TEST_ERROR( | |
5723 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5724 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 5725 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
ecfea3f0 | 5726 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); |
2451337d | 5727 | return -EFSCORRUPTED; |
e1d8fb88 NJ |
5728 | } |
5729 | ||
5730 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2451337d | 5731 | return -EIO; |
e1d8fb88 | 5732 | |
ecfea3f0 | 5733 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); |
e1d8fb88 | 5734 | |
e1d8fb88 | 5735 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
e1d8fb88 NJ |
5736 | error = xfs_iread_extents(tp, ip, whichfork); |
5737 | if (error) | |
5738 | return error; | |
5739 | } | |
5740 | ||
ddb19e31 BF |
5741 | if (ifp->if_flags & XFS_IFBROOT) { |
5742 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
ddb19e31 BF |
5743 | cur->bc_private.b.flags = 0; |
5744 | } | |
5745 | ||
b2b1712a | 5746 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) { |
ecfea3f0 CH |
5747 | *done = true; |
5748 | goto del_cursor; | |
5749 | } | |
d41c6172 ES |
5750 | XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock), |
5751 | del_cursor); | |
ecfea3f0 | 5752 | |
bf806280 | 5753 | new_startoff = got.br_startoff - offset_shift_fsb; |
b2b1712a | 5754 | if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) { |
bf806280 CH |
5755 | if (new_startoff < prev.br_startoff + prev.br_blockcount) { |
5756 | error = -EINVAL; | |
5757 | goto del_cursor; | |
5758 | } | |
5759 | ||
bf806280 | 5760 | if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) { |
0f37d178 BF |
5761 | error = xfs_bmse_merge(tp, ip, whichfork, |
5762 | offset_shift_fsb, &icur, &got, &prev, | |
5763 | cur, &logflags); | |
bf806280 CH |
5764 | if (error) |
5765 | goto del_cursor; | |
5766 | goto done; | |
5767 | } | |
5768 | } else { | |
5769 | if (got.br_startoff < offset_shift_fsb) { | |
5770 | error = -EINVAL; | |
5771 | goto del_cursor; | |
5772 | } | |
5773 | } | |
5774 | ||
0f37d178 BF |
5775 | error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got, |
5776 | cur, &logflags, new_startoff); | |
ecfea3f0 CH |
5777 | if (error) |
5778 | goto del_cursor; | |
40591bdb | 5779 | |
42630361 | 5780 | done: |
b2b1712a CH |
5781 | if (!xfs_iext_next_extent(ifp, &icur, &got)) { |
5782 | *done = true; | |
5783 | goto del_cursor; | |
ecfea3f0 | 5784 | } |
ecfea3f0 | 5785 | |
bf806280 | 5786 | *next_fsb = got.br_startoff; |
ecfea3f0 CH |
5787 | del_cursor: |
5788 | if (cur) | |
0b04b6b8 | 5789 | xfs_btree_del_cursor(cur, error); |
ecfea3f0 CH |
5790 | if (logflags) |
5791 | xfs_trans_log_inode(tp, ip, logflags); | |
ecfea3f0 CH |
5792 | return error; |
5793 | } | |
5794 | ||
f62cb48e DW |
5795 | /* Make sure we won't be right-shifting an extent past the maximum bound. */ |
5796 | int | |
5797 | xfs_bmap_can_insert_extents( | |
5798 | struct xfs_inode *ip, | |
5799 | xfs_fileoff_t off, | |
5800 | xfs_fileoff_t shift) | |
5801 | { | |
5802 | struct xfs_bmbt_irec got; | |
5803 | int is_empty; | |
5804 | int error = 0; | |
5805 | ||
5806 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5807 | ||
5808 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
5809 | return -EIO; | |
5810 | ||
5811 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5812 | error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty); | |
5813 | if (!error && !is_empty && got.br_startoff >= off && | |
5814 | ((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff) | |
5815 | error = -EINVAL; | |
5816 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
5817 | ||
5818 | return error; | |
5819 | } | |
5820 | ||
ecfea3f0 CH |
5821 | int |
5822 | xfs_bmap_insert_extents( | |
5823 | struct xfs_trans *tp, | |
5824 | struct xfs_inode *ip, | |
5825 | xfs_fileoff_t *next_fsb, | |
5826 | xfs_fileoff_t offset_shift_fsb, | |
5827 | bool *done, | |
333f950c | 5828 | xfs_fileoff_t stop_fsb) |
ecfea3f0 CH |
5829 | { |
5830 | int whichfork = XFS_DATA_FORK; | |
5831 | struct xfs_mount *mp = ip->i_mount; | |
5832 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
5833 | struct xfs_btree_cur *cur = NULL; | |
5936dc54 | 5834 | struct xfs_bmbt_irec got, next; |
b2b1712a | 5835 | struct xfs_iext_cursor icur; |
bf806280 | 5836 | xfs_fileoff_t new_startoff; |
ecfea3f0 CH |
5837 | int error = 0; |
5838 | int logflags = 0; | |
5839 | ||
5840 | if (unlikely(XFS_TEST_ERROR( | |
5841 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5842 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
5843 | mp, XFS_ERRTAG_BMAPIFORMAT))) { | |
5844 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
5845 | return -EFSCORRUPTED; | |
5846 | } | |
5847 | ||
5848 | if (XFS_FORCED_SHUTDOWN(mp)) | |
5849 | return -EIO; | |
5850 | ||
5851 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL)); | |
5852 | ||
5853 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
5854 | error = xfs_iread_extents(tp, ip, whichfork); | |
5855 | if (error) | |
5856 | return error; | |
5857 | } | |
5858 | ||
5859 | if (ifp->if_flags & XFS_IFBROOT) { | |
5860 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
ecfea3f0 CH |
5861 | cur->bc_private.b.flags = 0; |
5862 | } | |
5863 | ||
a904b1ca | 5864 | if (*next_fsb == NULLFSBLOCK) { |
b2b1712a CH |
5865 | xfs_iext_last(ifp, &icur); |
5866 | if (!xfs_iext_get_extent(ifp, &icur, &got) || | |
5936dc54 | 5867 | stop_fsb > got.br_startoff) { |
ecfea3f0 | 5868 | *done = true; |
a904b1ca NJ |
5869 | goto del_cursor; |
5870 | } | |
05b7c8ab | 5871 | } else { |
b2b1712a | 5872 | if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) { |
ecfea3f0 | 5873 | *done = true; |
05b7c8ab CH |
5874 | goto del_cursor; |
5875 | } | |
a904b1ca | 5876 | } |
d41c6172 ES |
5877 | XFS_WANT_CORRUPTED_GOTO(mp, !isnullstartblock(got.br_startblock), |
5878 | del_cursor); | |
a904b1ca | 5879 | |
5936dc54 | 5880 | if (stop_fsb >= got.br_startoff + got.br_blockcount) { |
ecfea3f0 CH |
5881 | error = -EIO; |
5882 | goto del_cursor; | |
a904b1ca NJ |
5883 | } |
5884 | ||
bf806280 | 5885 | new_startoff = got.br_startoff + offset_shift_fsb; |
b2b1712a | 5886 | if (xfs_iext_peek_next_extent(ifp, &icur, &next)) { |
bf806280 CH |
5887 | if (new_startoff + got.br_blockcount > next.br_startoff) { |
5888 | error = -EINVAL; | |
5889 | goto del_cursor; | |
5890 | } | |
5891 | ||
5892 | /* | |
5893 | * Unlike a left shift (which involves a hole punch), a right | |
5894 | * shift does not modify extent neighbors in any way. We should | |
5895 | * never find mergeable extents in this scenario. Check anyways | |
5896 | * and warn if we encounter two extents that could be one. | |
5897 | */ | |
5898 | if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb)) | |
5899 | WARN_ON_ONCE(1); | |
5900 | } | |
5901 | ||
0f37d178 BF |
5902 | error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got, |
5903 | cur, &logflags, new_startoff); | |
6b18af0d CH |
5904 | if (error) |
5905 | goto del_cursor; | |
5936dc54 | 5906 | |
b2b1712a | 5907 | if (!xfs_iext_prev_extent(ifp, &icur, &got) || |
5936dc54 | 5908 | stop_fsb >= got.br_startoff + got.br_blockcount) { |
ecfea3f0 | 5909 | *done = true; |
6b18af0d | 5910 | goto del_cursor; |
e1d8fb88 NJ |
5911 | } |
5912 | ||
6b18af0d | 5913 | *next_fsb = got.br_startoff; |
e1d8fb88 NJ |
5914 | del_cursor: |
5915 | if (cur) | |
0b04b6b8 | 5916 | xfs_btree_del_cursor(cur, error); |
ca446d88 BF |
5917 | if (logflags) |
5918 | xfs_trans_log_inode(tp, ip, logflags); | |
e1d8fb88 NJ |
5919 | return error; |
5920 | } | |
a904b1ca NJ |
5921 | |
5922 | /* | |
b2b1712a CH |
5923 | * Splits an extent into two extents at split_fsb block such that it is the |
5924 | * first block of the current_ext. @ext is a target extent to be split. | |
5925 | * @split_fsb is a block where the extents is split. If split_fsb lies in a | |
5926 | * hole or the first block of extents, just return 0. | |
a904b1ca NJ |
5927 | */ |
5928 | STATIC int | |
5929 | xfs_bmap_split_extent_at( | |
5930 | struct xfs_trans *tp, | |
5931 | struct xfs_inode *ip, | |
4b77a088 | 5932 | xfs_fileoff_t split_fsb) |
a904b1ca NJ |
5933 | { |
5934 | int whichfork = XFS_DATA_FORK; | |
5935 | struct xfs_btree_cur *cur = NULL; | |
a904b1ca NJ |
5936 | struct xfs_bmbt_irec got; |
5937 | struct xfs_bmbt_irec new; /* split extent */ | |
5938 | struct xfs_mount *mp = ip->i_mount; | |
5939 | struct xfs_ifork *ifp; | |
5940 | xfs_fsblock_t gotblkcnt; /* new block count for got */ | |
b2b1712a | 5941 | struct xfs_iext_cursor icur; |
a904b1ca NJ |
5942 | int error = 0; |
5943 | int logflags = 0; | |
5944 | int i = 0; | |
5945 | ||
5946 | if (unlikely(XFS_TEST_ERROR( | |
5947 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5948 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
9e24cfd0 | 5949 | mp, XFS_ERRTAG_BMAPIFORMAT))) { |
a904b1ca NJ |
5950 | XFS_ERROR_REPORT("xfs_bmap_split_extent_at", |
5951 | XFS_ERRLEVEL_LOW, mp); | |
5952 | return -EFSCORRUPTED; | |
5953 | } | |
5954 | ||
5955 | if (XFS_FORCED_SHUTDOWN(mp)) | |
5956 | return -EIO; | |
5957 | ||
5958 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
5959 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
5960 | /* Read in all the extents */ | |
5961 | error = xfs_iread_extents(tp, ip, whichfork); | |
5962 | if (error) | |
5963 | return error; | |
5964 | } | |
5965 | ||
5966 | /* | |
4c35445b | 5967 | * If there are not extents, or split_fsb lies in a hole we are done. |
a904b1ca | 5968 | */ |
b2b1712a | 5969 | if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) || |
4c35445b | 5970 | got.br_startoff >= split_fsb) |
a904b1ca NJ |
5971 | return 0; |
5972 | ||
5973 | gotblkcnt = split_fsb - got.br_startoff; | |
5974 | new.br_startoff = split_fsb; | |
5975 | new.br_startblock = got.br_startblock + gotblkcnt; | |
5976 | new.br_blockcount = got.br_blockcount - gotblkcnt; | |
5977 | new.br_state = got.br_state; | |
5978 | ||
5979 | if (ifp->if_flags & XFS_IFBROOT) { | |
5980 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
a904b1ca | 5981 | cur->bc_private.b.flags = 0; |
e16cf9b0 | 5982 | error = xfs_bmbt_lookup_eq(cur, &got, &i); |
a904b1ca NJ |
5983 | if (error) |
5984 | goto del_cursor; | |
5985 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
5986 | } | |
5987 | ||
a904b1ca | 5988 | got.br_blockcount = gotblkcnt; |
b2b1712a CH |
5989 | xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur, |
5990 | &got); | |
a904b1ca NJ |
5991 | |
5992 | logflags = XFS_ILOG_CORE; | |
5993 | if (cur) { | |
a67d00a5 | 5994 | error = xfs_bmbt_update(cur, &got); |
a904b1ca NJ |
5995 | if (error) |
5996 | goto del_cursor; | |
5997 | } else | |
5998 | logflags |= XFS_ILOG_DEXT; | |
5999 | ||
6000 | /* Add new extent */ | |
b2b1712a | 6001 | xfs_iext_next(ifp, &icur); |
0254c2f2 | 6002 | xfs_iext_insert(ip, &icur, &new, 0); |
a904b1ca NJ |
6003 | XFS_IFORK_NEXT_SET(ip, whichfork, |
6004 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
6005 | ||
6006 | if (cur) { | |
e16cf9b0 | 6007 | error = xfs_bmbt_lookup_eq(cur, &new, &i); |
a904b1ca NJ |
6008 | if (error) |
6009 | goto del_cursor; | |
6010 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor); | |
a904b1ca NJ |
6011 | error = xfs_btree_insert(cur, &i); |
6012 | if (error) | |
6013 | goto del_cursor; | |
6014 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
6015 | } | |
6016 | ||
6017 | /* | |
6018 | * Convert to a btree if necessary. | |
6019 | */ | |
6020 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
6021 | int tmp_logflags; /* partial log flag return val */ | |
6022 | ||
6023 | ASSERT(cur == NULL); | |
280253d2 BF |
6024 | error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, |
6025 | &tmp_logflags, whichfork); | |
a904b1ca NJ |
6026 | logflags |= tmp_logflags; |
6027 | } | |
6028 | ||
6029 | del_cursor: | |
6030 | if (cur) { | |
6031 | cur->bc_private.b.allocated = 0; | |
0b04b6b8 | 6032 | xfs_btree_del_cursor(cur, error); |
a904b1ca NJ |
6033 | } |
6034 | ||
6035 | if (logflags) | |
6036 | xfs_trans_log_inode(tp, ip, logflags); | |
6037 | return error; | |
6038 | } | |
6039 | ||
6040 | int | |
6041 | xfs_bmap_split_extent( | |
6042 | struct xfs_inode *ip, | |
6043 | xfs_fileoff_t split_fsb) | |
6044 | { | |
6045 | struct xfs_mount *mp = ip->i_mount; | |
6046 | struct xfs_trans *tp; | |
a904b1ca NJ |
6047 | int error; |
6048 | ||
253f4911 CH |
6049 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, |
6050 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); | |
6051 | if (error) | |
a904b1ca | 6052 | return error; |
a904b1ca NJ |
6053 | |
6054 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
6055 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
6056 | ||
4b77a088 | 6057 | error = xfs_bmap_split_extent_at(tp, ip, split_fsb); |
a904b1ca NJ |
6058 | if (error) |
6059 | goto out; | |
6060 | ||
70393313 | 6061 | return xfs_trans_commit(tp); |
a904b1ca NJ |
6062 | |
6063 | out: | |
4906e215 | 6064 | xfs_trans_cancel(tp); |
a904b1ca NJ |
6065 | return error; |
6066 | } | |
9f3afb57 DW |
6067 | |
6068 | /* Deferred mapping is only for real extents in the data fork. */ | |
6069 | static bool | |
6070 | xfs_bmap_is_update_needed( | |
6071 | struct xfs_bmbt_irec *bmap) | |
6072 | { | |
6073 | return bmap->br_startblock != HOLESTARTBLOCK && | |
6074 | bmap->br_startblock != DELAYSTARTBLOCK; | |
6075 | } | |
6076 | ||
6077 | /* Record a bmap intent. */ | |
6078 | static int | |
6079 | __xfs_bmap_add( | |
0f37d178 | 6080 | struct xfs_trans *tp, |
9f3afb57 DW |
6081 | enum xfs_bmap_intent_type type, |
6082 | struct xfs_inode *ip, | |
6083 | int whichfork, | |
6084 | struct xfs_bmbt_irec *bmap) | |
6085 | { | |
9f3afb57 DW |
6086 | struct xfs_bmap_intent *bi; |
6087 | ||
0f37d178 BF |
6088 | trace_xfs_bmap_defer(tp->t_mountp, |
6089 | XFS_FSB_TO_AGNO(tp->t_mountp, bmap->br_startblock), | |
9f3afb57 | 6090 | type, |
0f37d178 | 6091 | XFS_FSB_TO_AGBNO(tp->t_mountp, bmap->br_startblock), |
9f3afb57 DW |
6092 | ip->i_ino, whichfork, |
6093 | bmap->br_startoff, | |
6094 | bmap->br_blockcount, | |
6095 | bmap->br_state); | |
6096 | ||
6097 | bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS); | |
6098 | INIT_LIST_HEAD(&bi->bi_list); | |
6099 | bi->bi_type = type; | |
6100 | bi->bi_owner = ip; | |
6101 | bi->bi_whichfork = whichfork; | |
6102 | bi->bi_bmap = *bmap; | |
6103 | ||
0f37d178 | 6104 | xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list); |
9f3afb57 DW |
6105 | return 0; |
6106 | } | |
6107 | ||
6108 | /* Map an extent into a file. */ | |
6109 | int | |
6110 | xfs_bmap_map_extent( | |
0f37d178 | 6111 | struct xfs_trans *tp, |
9f3afb57 DW |
6112 | struct xfs_inode *ip, |
6113 | struct xfs_bmbt_irec *PREV) | |
6114 | { | |
6115 | if (!xfs_bmap_is_update_needed(PREV)) | |
6116 | return 0; | |
6117 | ||
0f37d178 | 6118 | return __xfs_bmap_add(tp, XFS_BMAP_MAP, ip, XFS_DATA_FORK, PREV); |
9f3afb57 DW |
6119 | } |
6120 | ||
6121 | /* Unmap an extent out of a file. */ | |
6122 | int | |
6123 | xfs_bmap_unmap_extent( | |
0f37d178 | 6124 | struct xfs_trans *tp, |
9f3afb57 DW |
6125 | struct xfs_inode *ip, |
6126 | struct xfs_bmbt_irec *PREV) | |
6127 | { | |
6128 | if (!xfs_bmap_is_update_needed(PREV)) | |
6129 | return 0; | |
6130 | ||
0f37d178 | 6131 | return __xfs_bmap_add(tp, XFS_BMAP_UNMAP, ip, XFS_DATA_FORK, PREV); |
9f3afb57 DW |
6132 | } |
6133 | ||
6134 | /* | |
6135 | * Process one of the deferred bmap operations. We pass back the | |
6136 | * btree cursor to maintain our lock on the bmapbt between calls. | |
6137 | */ | |
6138 | int | |
6139 | xfs_bmap_finish_one( | |
6140 | struct xfs_trans *tp, | |
9f3afb57 DW |
6141 | struct xfs_inode *ip, |
6142 | enum xfs_bmap_intent_type type, | |
6143 | int whichfork, | |
6144 | xfs_fileoff_t startoff, | |
6145 | xfs_fsblock_t startblock, | |
e1a4e37c | 6146 | xfs_filblks_t *blockcount, |
9f3afb57 DW |
6147 | xfs_exntst_t state) |
6148 | { | |
e1a4e37c | 6149 | int error = 0; |
9f3afb57 | 6150 | |
37283797 | 6151 | ASSERT(tp->t_firstblock == NULLFSBLOCK); |
4c1a67bd | 6152 | |
9f3afb57 DW |
6153 | trace_xfs_bmap_deferred(tp->t_mountp, |
6154 | XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type, | |
6155 | XFS_FSB_TO_AGBNO(tp->t_mountp, startblock), | |
e1a4e37c | 6156 | ip->i_ino, whichfork, startoff, *blockcount, state); |
9f3afb57 | 6157 | |
39e07daa | 6158 | if (WARN_ON_ONCE(whichfork != XFS_DATA_FORK)) |
9f3afb57 | 6159 | return -EFSCORRUPTED; |
9f3afb57 DW |
6160 | |
6161 | if (XFS_TEST_ERROR(false, tp->t_mountp, | |
9e24cfd0 | 6162 | XFS_ERRTAG_BMAP_FINISH_ONE)) |
9f3afb57 DW |
6163 | return -EIO; |
6164 | ||
6165 | switch (type) { | |
6166 | case XFS_BMAP_MAP: | |
e1a4e37c | 6167 | error = xfs_bmapi_remap(tp, ip, startoff, *blockcount, |
ff3edf25 | 6168 | startblock, 0); |
e1a4e37c | 6169 | *blockcount = 0; |
9f3afb57 DW |
6170 | break; |
6171 | case XFS_BMAP_UNMAP: | |
e1a4e37c | 6172 | error = __xfs_bunmapi(tp, ip, startoff, blockcount, |
2af52842 | 6173 | XFS_BMAPI_REMAP, 1); |
9f3afb57 DW |
6174 | break; |
6175 | default: | |
6176 | ASSERT(0); | |
6177 | error = -EFSCORRUPTED; | |
6178 | } | |
6179 | ||
6180 | return error; | |
6181 | } | |
30b0984d DW |
6182 | |
6183 | /* Check that an inode's extent does not have invalid flags or bad ranges. */ | |
6184 | xfs_failaddr_t | |
6185 | xfs_bmap_validate_extent( | |
6186 | struct xfs_inode *ip, | |
6187 | int whichfork, | |
6188 | struct xfs_bmbt_irec *irec) | |
6189 | { | |
6190 | struct xfs_mount *mp = ip->i_mount; | |
6191 | xfs_fsblock_t endfsb; | |
6192 | bool isrt; | |
6193 | ||
6194 | isrt = XFS_IS_REALTIME_INODE(ip); | |
6195 | endfsb = irec->br_startblock + irec->br_blockcount - 1; | |
6196 | if (isrt) { | |
6197 | if (!xfs_verify_rtbno(mp, irec->br_startblock)) | |
6198 | return __this_address; | |
6199 | if (!xfs_verify_rtbno(mp, endfsb)) | |
6200 | return __this_address; | |
6201 | } else { | |
6202 | if (!xfs_verify_fsbno(mp, irec->br_startblock)) | |
6203 | return __this_address; | |
6204 | if (!xfs_verify_fsbno(mp, endfsb)) | |
6205 | return __this_address; | |
6206 | if (XFS_FSB_TO_AGNO(mp, irec->br_startblock) != | |
6207 | XFS_FSB_TO_AGNO(mp, endfsb)) | |
6208 | return __this_address; | |
6209 | } | |
daa79bae CH |
6210 | if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK) |
6211 | return __this_address; | |
30b0984d DW |
6212 | return NULL; |
6213 | } |