]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
5e656dbb | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
da23017d | 3 | * All Rights Reserved. |
5000d01d | 4 | * |
da23017d NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
5000d01d | 13 | * |
da23017d NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
2bd0ea18 | 17 | */ |
9c799827 | 18 | #include "libxfs_priv.h" |
b626fb59 DC |
19 | #include "xfs_fs.h" |
20 | #include "xfs_shared.h" | |
21 | #include "xfs_format.h" | |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
24 | #include "xfs_bit.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_mount.h" | |
f944d3d0 | 27 | #include "xfs_defer.h" |
b626fb59 DC |
28 | #include "xfs_da_format.h" |
29 | #include "xfs_da_btree.h" | |
30 | #include "xfs_dir2.h" | |
31 | #include "xfs_inode.h" | |
32 | #include "xfs_btree.h" | |
33 | #include "xfs_trans.h" | |
34 | #include "xfs_alloc.h" | |
35 | #include "xfs_bmap.h" | |
36 | #include "xfs_bmap_btree.h" | |
37 | #include "xfs_trans_space.h" | |
38 | #include "xfs_trace.h" | |
39 | #include "xfs_attr_leaf.h" | |
40 | #include "xfs_quota_defs.h" | |
85aec44f | 41 | #include "xfs_rmap.h" |
cf8ce220 | 42 | #include "xfs_ag_resv.h" |
cfe32f0d | 43 | #include "xfs_refcount.h" |
f0d92f04 | 44 | #include "xfs_rmap_btree.h" |
b626fb59 | 45 | |
2bd0ea18 | 46 | |
5e656dbb BN |
47 | kmem_zone_t *xfs_bmap_free_item_zone; |
48 | ||
49 | /* | |
49f693fa | 50 | * Miscellaneous helper functions |
5e656dbb | 51 | */ |
5e656dbb | 52 | |
5e656dbb | 53 | /* |
49f693fa DC |
54 | * Compute and fill in the value of the maximum depth of a bmap btree |
55 | * in this filesystem. Done once, during mount. | |
5e656dbb | 56 | */ |
49f693fa DC |
57 | void |
58 | xfs_bmap_compute_maxlevels( | |
59 | xfs_mount_t *mp, /* file system mount structure */ | |
60 | int whichfork) /* data or attr fork */ | |
61 | { | |
62 | int level; /* btree level */ | |
63 | uint maxblocks; /* max blocks at this level */ | |
64 | uint maxleafents; /* max leaf entries possible */ | |
65 | int maxrootrecs; /* max records in root block */ | |
66 | int minleafrecs; /* min records in leaf block */ | |
67 | int minnoderecs; /* min records in node block */ | |
68 | int sz; /* root block size */ | |
5e656dbb | 69 | |
49f693fa DC |
70 | /* |
71 | * The maximum number of extents in a file, hence the maximum | |
72 | * number of leaf entries, is controlled by the type of di_nextents | |
73 | * (a signed 32-bit number, xfs_extnum_t), or by di_anextents | |
74 | * (a signed 16-bit number, xfs_aextnum_t). | |
75 | * | |
76 | * Note that we can no longer assume that if we are in ATTR1 that | |
77 | * the fork offset of all the inodes will be | |
78 | * (xfs_default_attroffset(ip) >> 3) because we could have mounted | |
79 | * with ATTR2 and then mounted back with ATTR1, keeping the | |
80 | * di_forkoff's fixed but probably at various positions. Therefore, | |
81 | * for both ATTR1 and ATTR2 we have to assume the worst case scenario | |
82 | * of a minimum size available. | |
83 | */ | |
84 | if (whichfork == XFS_DATA_FORK) { | |
85 | maxleafents = MAXEXTNUM; | |
86 | sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS); | |
87 | } else { | |
88 | maxleafents = MAXAEXTNUM; | |
89 | sz = XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
90 | } | |
ff105f75 | 91 | maxrootrecs = xfs_bmdr_maxrecs(sz, 0); |
49f693fa DC |
92 | minleafrecs = mp->m_bmap_dmnr[0]; |
93 | minnoderecs = mp->m_bmap_dmnr[1]; | |
94 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
95 | for (level = 1; maxblocks > 1; level++) { | |
96 | if (maxblocks <= maxrootrecs) | |
97 | maxblocks = 1; | |
98 | else | |
99 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
100 | } | |
101 | mp->m_bm_maxlevels[whichfork] = level; | |
102 | } | |
5e656dbb | 103 | |
b194c7d8 BN |
104 | STATIC int /* error */ |
105 | xfs_bmbt_lookup_eq( | |
106 | struct xfs_btree_cur *cur, | |
107 | xfs_fileoff_t off, | |
108 | xfs_fsblock_t bno, | |
109 | xfs_filblks_t len, | |
110 | int *stat) /* success/failure */ | |
111 | { | |
112 | cur->bc_rec.b.br_startoff = off; | |
113 | cur->bc_rec.b.br_startblock = bno; | |
114 | cur->bc_rec.b.br_blockcount = len; | |
115 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); | |
116 | } | |
117 | ||
118 | STATIC int /* error */ | |
119 | xfs_bmbt_lookup_ge( | |
120 | struct xfs_btree_cur *cur, | |
121 | xfs_fileoff_t off, | |
122 | xfs_fsblock_t bno, | |
123 | xfs_filblks_t len, | |
124 | int *stat) /* success/failure */ | |
125 | { | |
126 | cur->bc_rec.b.br_startoff = off; | |
127 | cur->bc_rec.b.br_startblock = bno; | |
128 | cur->bc_rec.b.br_blockcount = len; | |
129 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); | |
130 | } | |
131 | ||
132 | /* | |
a2ceac1f DC |
133 | * Check if the inode needs to be converted to btree format. |
134 | */ | |
135 | static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork) | |
136 | { | |
1277a5e0 DW |
137 | return whichfork != XFS_COW_FORK && |
138 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && | |
a2ceac1f DC |
139 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
140 | XFS_IFORK_MAXEXT(ip, whichfork); | |
141 | } | |
142 | ||
143 | /* | |
144 | * Check if the inode should be converted to extent format. | |
145 | */ | |
146 | static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork) | |
147 | { | |
1277a5e0 DW |
148 | return whichfork != XFS_COW_FORK && |
149 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE && | |
a2ceac1f DC |
150 | XFS_IFORK_NEXTENTS(ip, whichfork) <= |
151 | XFS_IFORK_MAXEXT(ip, whichfork); | |
152 | } | |
153 | ||
154 | /* | |
155 | * Update the record referred to by cur to the value given | |
b194c7d8 BN |
156 | * by [off, bno, len, state]. |
157 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
158 | */ | |
159 | STATIC int | |
160 | xfs_bmbt_update( | |
161 | struct xfs_btree_cur *cur, | |
162 | xfs_fileoff_t off, | |
163 | xfs_fsblock_t bno, | |
164 | xfs_filblks_t len, | |
165 | xfs_exntst_t state) | |
166 | { | |
167 | union xfs_btree_rec rec; | |
168 | ||
169 | xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state); | |
170 | return xfs_btree_update(cur, &rec); | |
171 | } | |
172 | ||
5e656dbb | 173 | /* |
49f693fa DC |
174 | * Compute the worst-case number of indirect blocks that will be used |
175 | * for ip's delayed extent of length "len". | |
5e656dbb | 176 | */ |
49f693fa DC |
177 | STATIC xfs_filblks_t |
178 | xfs_bmap_worst_indlen( | |
179 | xfs_inode_t *ip, /* incore inode pointer */ | |
180 | xfs_filblks_t len) /* delayed extent length */ | |
57c9fccb | 181 | { |
49f693fa DC |
182 | int level; /* btree level number */ |
183 | int maxrecs; /* maximum record count at this level */ | |
184 | xfs_mount_t *mp; /* mount structure */ | |
185 | xfs_filblks_t rval; /* return value */ | |
f0d92f04 | 186 | xfs_filblks_t orig_len; |
57c9fccb NS |
187 | |
188 | mp = ip->i_mount; | |
f0d92f04 DW |
189 | |
190 | /* Calculate the worst-case size of the bmbt. */ | |
191 | orig_len = len; | |
49f693fa DC |
192 | maxrecs = mp->m_bmap_dmxr[0]; |
193 | for (level = 0, rval = 0; | |
194 | level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK); | |
195 | level++) { | |
196 | len += maxrecs - 1; | |
197 | do_div(len, maxrecs); | |
198 | rval += len; | |
f0d92f04 DW |
199 | if (len == 1) { |
200 | rval += XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) - | |
49f693fa | 201 | level - 1; |
f0d92f04 DW |
202 | break; |
203 | } | |
49f693fa DC |
204 | if (level == 0) |
205 | maxrecs = mp->m_bmap_dmxr[1]; | |
57c9fccb | 206 | } |
f0d92f04 DW |
207 | |
208 | /* Calculate the worst-case size of the rmapbt. */ | |
209 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
210 | rval += 1 + xfs_rmapbt_calc_size(mp, orig_len) + | |
211 | mp->m_rmap_maxlevels; | |
212 | ||
49f693fa | 213 | return rval; |
57c9fccb NS |
214 | } |
215 | ||
216 | /* | |
49f693fa | 217 | * Calculate the default attribute fork offset for newly created inodes. |
57c9fccb | 218 | */ |
49f693fa DC |
219 | uint |
220 | xfs_default_attroffset( | |
221 | struct xfs_inode *ip) | |
57c9fccb | 222 | { |
49f693fa DC |
223 | struct xfs_mount *mp = ip->i_mount; |
224 | uint offset; | |
57c9fccb | 225 | |
49f693fa DC |
226 | if (mp->m_sb.sb_inodesize == 256) { |
227 | offset = XFS_LITINO(mp, ip->i_d.di_version) - | |
228 | XFS_BMDR_SPACE_CALC(MINABTPTRS); | |
229 | } else { | |
230 | offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS); | |
57c9fccb | 231 | } |
49f693fa DC |
232 | |
233 | ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version)); | |
234 | return offset; | |
57c9fccb NS |
235 | } |
236 | ||
237 | /* | |
49f693fa DC |
238 | * Helper routine to reset inode di_forkoff field when switching |
239 | * attribute fork from local to extent format - we reset it where | |
240 | * possible to make space available for inline data fork extents. | |
57c9fccb | 241 | */ |
49f693fa DC |
242 | STATIC void |
243 | xfs_bmap_forkoff_reset( | |
49f693fa DC |
244 | xfs_inode_t *ip, |
245 | int whichfork) | |
57c9fccb | 246 | { |
49f693fa DC |
247 | if (whichfork == XFS_ATTR_FORK && |
248 | ip->i_d.di_format != XFS_DINODE_FMT_DEV && | |
249 | ip->i_d.di_format != XFS_DINODE_FMT_UUID && | |
250 | ip->i_d.di_format != XFS_DINODE_FMT_BTREE) { | |
251 | uint dfl_forkoff = xfs_default_attroffset(ip) >> 3; | |
57c9fccb | 252 | |
49f693fa DC |
253 | if (dfl_forkoff > ip->i_d.di_forkoff) |
254 | ip->i_d.di_forkoff = dfl_forkoff; | |
255 | } | |
57c9fccb NS |
256 | } |
257 | ||
49f693fa DC |
258 | #ifdef DEBUG |
259 | STATIC struct xfs_buf * | |
260 | xfs_bmap_get_bp( | |
261 | struct xfs_btree_cur *cur, | |
262 | xfs_fsblock_t bno) | |
263 | { | |
264 | struct xfs_log_item_desc *lidp; | |
265 | int i; | |
56b2de80 | 266 | |
49f693fa DC |
267 | if (!cur) |
268 | return NULL; | |
2bd0ea18 | 269 | |
49f693fa DC |
270 | for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) { |
271 | if (!cur->bc_bufs[i]) | |
272 | break; | |
273 | if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno) | |
274 | return cur->bc_bufs[i]; | |
275 | } | |
56b2de80 | 276 | |
49f693fa DC |
277 | /* Chase down all the log items to see if the bp is there */ |
278 | list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) { | |
279 | struct xfs_buf_log_item *bip; | |
280 | bip = (struct xfs_buf_log_item *)lidp->lid_item; | |
281 | if (bip->bli_item.li_type == XFS_LI_BUF && | |
282 | XFS_BUF_ADDR(bip->bli_buf) == bno) | |
283 | return bip->bli_buf; | |
284 | } | |
2bd0ea18 | 285 | |
49f693fa DC |
286 | return NULL; |
287 | } | |
56b2de80 | 288 | |
49f693fa DC |
289 | STATIC void |
290 | xfs_check_block( | |
291 | struct xfs_btree_block *block, | |
292 | xfs_mount_t *mp, | |
293 | int root, | |
294 | short sz) | |
295 | { | |
296 | int i, j, dmxr; | |
297 | __be64 *pp, *thispa; /* pointer to block address */ | |
298 | xfs_bmbt_key_t *prevp, *keyp; | |
2bd0ea18 | 299 | |
49f693fa | 300 | ASSERT(be16_to_cpu(block->bb_level) > 0); |
56b2de80 | 301 | |
49f693fa DC |
302 | prevp = NULL; |
303 | for( i = 1; i <= xfs_btree_get_numrecs(block); i++) { | |
304 | dmxr = mp->m_bmap_dmxr[0]; | |
305 | keyp = XFS_BMBT_KEY_ADDR(mp, block, i); | |
a2ceac1f | 306 | |
49f693fa DC |
307 | if (prevp) { |
308 | ASSERT(be64_to_cpu(prevp->br_startoff) < | |
309 | be64_to_cpu(keyp->br_startoff)); | |
310 | } | |
311 | prevp = keyp; | |
2bd0ea18 | 312 | |
2bd0ea18 | 313 | /* |
49f693fa | 314 | * Compare the block numbers to see if there are dups. |
2bd0ea18 | 315 | */ |
49f693fa DC |
316 | if (root) |
317 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz); | |
318 | else | |
319 | pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr); | |
320 | ||
321 | for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) { | |
322 | if (root) | |
323 | thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz); | |
324 | else | |
325 | thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr); | |
326 | if (*thispa == *pp) { | |
327 | xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld", | |
328 | __func__, j, i, | |
329 | (unsigned long long)be64_to_cpu(*thispa)); | |
330 | panic("%s: ptrs are equal in node\n", | |
331 | __func__); | |
332 | } | |
2bd0ea18 | 333 | } |
49f693fa DC |
334 | } |
335 | } | |
a2ceac1f | 336 | |
49f693fa DC |
337 | /* |
338 | * Check that the extents for the inode ip are in the right order in all | |
f07ae2a6 DC |
339 | * btree leaves. THis becomes prohibitively expensive for large extent count |
340 | * files, so don't bother with inodes that have more than 10,000 extents in | |
341 | * them. The btree record ordering checks will still be done, so for such large | |
342 | * bmapbt constructs that is going to catch most corruptions. | |
49f693fa | 343 | */ |
49f693fa DC |
344 | STATIC void |
345 | xfs_bmap_check_leaf_extents( | |
346 | xfs_btree_cur_t *cur, /* btree cursor or null */ | |
347 | xfs_inode_t *ip, /* incore inode pointer */ | |
348 | int whichfork) /* data or attr fork */ | |
349 | { | |
350 | struct xfs_btree_block *block; /* current btree block */ | |
351 | xfs_fsblock_t bno; /* block # of "block" */ | |
352 | xfs_buf_t *bp; /* buffer for "block" */ | |
353 | int error; /* error return value */ | |
354 | xfs_extnum_t i=0, j; /* index into the extents list */ | |
355 | xfs_ifork_t *ifp; /* fork structure */ | |
356 | int level; /* btree level, for checking */ | |
357 | xfs_mount_t *mp; /* file system mount structure */ | |
358 | __be64 *pp; /* pointer to block address */ | |
359 | xfs_bmbt_rec_t *ep; /* pointer to current extent */ | |
360 | xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */ | |
361 | xfs_bmbt_rec_t *nextp; /* pointer to next extent */ | |
362 | int bp_release = 0; | |
363 | ||
364 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) { | |
365 | return; | |
366 | } | |
367 | ||
f07ae2a6 DC |
368 | /* skip large extent count inodes */ |
369 | if (ip->i_d.di_nextents > 10000) | |
370 | return; | |
371 | ||
49f693fa DC |
372 | bno = NULLFSBLOCK; |
373 | mp = ip->i_mount; | |
374 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
375 | block = ifp->if_broot; | |
376 | /* | |
377 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. | |
378 | */ | |
379 | level = be16_to_cpu(block->bb_level); | |
380 | ASSERT(level > 0); | |
381 | xfs_check_block(block, mp, 1, ifp->if_broot_bytes); | |
382 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
383 | bno = be64_to_cpu(*pp); | |
384 | ||
5a35bf2c | 385 | ASSERT(bno != NULLFSBLOCK); |
49f693fa DC |
386 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
387 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
388 | ||
389 | /* | |
390 | * Go down the tree until leaf level is reached, following the first | |
391 | * pointer (leftmost) at each level. | |
392 | */ | |
393 | while (level-- > 0) { | |
394 | /* See if buf is in cur first */ | |
395 | bp_release = 0; | |
396 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
397 | if (!bp) { | |
398 | bp_release = 1; | |
399 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
400 | XFS_BMAP_BTREE_REF, | |
401 | &xfs_bmbt_buf_ops); | |
2bd0ea18 | 402 | if (error) |
49f693fa | 403 | goto error_norelse; |
2bd0ea18 | 404 | } |
49f693fa | 405 | block = XFS_BUF_TO_BLOCK(bp); |
49f693fa DC |
406 | if (level == 0) |
407 | break; | |
2bd0ea18 | 408 | |
2bd0ea18 | 409 | /* |
49f693fa DC |
410 | * Check this block for basic sanity (increasing keys and |
411 | * no duplicate blocks). | |
2bd0ea18 | 412 | */ |
49f693fa DC |
413 | |
414 | xfs_check_block(block, mp, 0, 0); | |
415 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
416 | bno = be64_to_cpu(*pp); | |
19ebedcf DC |
417 | XFS_WANT_CORRUPTED_GOTO(mp, |
418 | XFS_FSB_SANITY_CHECK(mp, bno), error0); | |
49f693fa DC |
419 | if (bp_release) { |
420 | bp_release = 0; | |
421 | xfs_trans_brelse(NULL, bp); | |
2bd0ea18 | 422 | } |
49f693fa | 423 | } |
a2ceac1f | 424 | |
49f693fa DC |
425 | /* |
426 | * Here with bp and block set to the leftmost leaf node in the tree. | |
427 | */ | |
428 | i = 0; | |
a2ceac1f | 429 | |
49f693fa DC |
430 | /* |
431 | * Loop over all leaf nodes checking that all extents are in the right order. | |
432 | */ | |
433 | for (;;) { | |
434 | xfs_fsblock_t nextbno; | |
435 | xfs_extnum_t num_recs; | |
436 | ||
437 | ||
438 | num_recs = xfs_btree_get_numrecs(block); | |
2bd0ea18 | 439 | |
2bd0ea18 | 440 | /* |
49f693fa | 441 | * Read-ahead the next leaf block, if any. |
2bd0ea18 | 442 | */ |
a2ceac1f | 443 | |
49f693fa | 444 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
a2ceac1f | 445 | |
49f693fa DC |
446 | /* |
447 | * Check all the extents to make sure they are OK. | |
448 | * If we had a previous block, the last entry should | |
449 | * conform with the first entry in this one. | |
450 | */ | |
2bd0ea18 | 451 | |
49f693fa DC |
452 | ep = XFS_BMBT_REC_ADDR(mp, block, 1); |
453 | if (i) { | |
454 | ASSERT(xfs_bmbt_disk_get_startoff(&last) + | |
455 | xfs_bmbt_disk_get_blockcount(&last) <= | |
456 | xfs_bmbt_disk_get_startoff(ep)); | |
457 | } | |
458 | for (j = 1; j < num_recs; j++) { | |
459 | nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1); | |
460 | ASSERT(xfs_bmbt_disk_get_startoff(ep) + | |
461 | xfs_bmbt_disk_get_blockcount(ep) <= | |
462 | xfs_bmbt_disk_get_startoff(nextp)); | |
463 | ep = nextp; | |
464 | } | |
465 | ||
466 | last = *ep; | |
467 | i += num_recs; | |
468 | if (bp_release) { | |
469 | bp_release = 0; | |
470 | xfs_trans_brelse(NULL, bp); | |
471 | } | |
472 | bno = nextbno; | |
2bd0ea18 | 473 | /* |
49f693fa | 474 | * If we've reached the end, stop. |
2bd0ea18 | 475 | */ |
49f693fa DC |
476 | if (bno == NULLFSBLOCK) |
477 | break; | |
a2ceac1f | 478 | |
49f693fa DC |
479 | bp_release = 0; |
480 | bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno)); | |
481 | if (!bp) { | |
482 | bp_release = 1; | |
483 | error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp, | |
484 | XFS_BMAP_BTREE_REF, | |
485 | &xfs_bmbt_buf_ops); | |
a2ceac1f | 486 | if (error) |
49f693fa | 487 | goto error_norelse; |
2bd0ea18 | 488 | } |
49f693fa | 489 | block = XFS_BUF_TO_BLOCK(bp); |
a2ceac1f | 490 | } |
4d4a192c | 491 | |
49f693fa | 492 | return; |
a2ceac1f | 493 | |
49f693fa DC |
494 | error0: |
495 | xfs_warn(mp, "%s: at error0", __func__); | |
496 | if (bp_release) | |
497 | xfs_trans_brelse(NULL, bp); | |
498 | error_norelse: | |
499 | xfs_warn(mp, "%s: BAD after btree leaves for %d extents", | |
500 | __func__, i); | |
501 | panic("%s: CORRUPTED BTREE OR SOMETHING", __func__); | |
502 | return; | |
2bd0ea18 NS |
503 | } |
504 | ||
505 | /* | |
49f693fa | 506 | * Add bmap trace insert entries for all the contents of the extent records. |
2bd0ea18 | 507 | */ |
49f693fa DC |
508 | void |
509 | xfs_bmap_trace_exlist( | |
510 | xfs_inode_t *ip, /* incore inode pointer */ | |
511 | xfs_extnum_t cnt, /* count of entries in the list */ | |
512 | int whichfork, /* data or attr fork */ | |
513 | unsigned long caller_ip) | |
2bd0ea18 | 514 | { |
49f693fa DC |
515 | xfs_extnum_t idx; /* extent record index */ |
516 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
517 | int state = 0; | |
a2ceac1f | 518 | |
49f693fa DC |
519 | if (whichfork == XFS_ATTR_FORK) |
520 | state |= BMAP_ATTRFORK; | |
a2ceac1f | 521 | |
49f693fa | 522 | ifp = XFS_IFORK_PTR(ip, whichfork); |
d09d4e5f | 523 | ASSERT(cnt == xfs_iext_count(ifp)); |
49f693fa | 524 | for (idx = 0; idx < cnt; idx++) |
b12f5cba | 525 | trace_xfs_extlist(ip, idx, state, caller_ip); |
49f693fa | 526 | } |
a2ceac1f | 527 | |
49f693fa DC |
528 | /* |
529 | * Validate that the bmbt_irecs being returned from bmapi are valid | |
e6d77a21 DC |
530 | * given the caller's original parameters. Specifically check the |
531 | * ranges of the returned irecs to ensure that they only extend beyond | |
49f693fa DC |
532 | * the given parameters if the XFS_BMAPI_ENTIRE flag was set. |
533 | */ | |
534 | STATIC void | |
535 | xfs_bmap_validate_ret( | |
536 | xfs_fileoff_t bno, | |
537 | xfs_filblks_t len, | |
538 | int flags, | |
539 | xfs_bmbt_irec_t *mval, | |
540 | int nmap, | |
541 | int ret_nmap) | |
542 | { | |
543 | int i; /* index to map values */ | |
a2ceac1f | 544 | |
49f693fa | 545 | ASSERT(ret_nmap <= nmap); |
a2ceac1f | 546 | |
49f693fa DC |
547 | for (i = 0; i < ret_nmap; i++) { |
548 | ASSERT(mval[i].br_blockcount > 0); | |
549 | if (!(flags & XFS_BMAPI_ENTIRE)) { | |
550 | ASSERT(mval[i].br_startoff >= bno); | |
551 | ASSERT(mval[i].br_blockcount <= len); | |
552 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount <= | |
553 | bno + len); | |
554 | } else { | |
555 | ASSERT(mval[i].br_startoff < bno + len); | |
556 | ASSERT(mval[i].br_startoff + mval[i].br_blockcount > | |
557 | bno); | |
558 | } | |
559 | ASSERT(i == 0 || | |
560 | mval[i - 1].br_startoff + mval[i - 1].br_blockcount == | |
561 | mval[i].br_startoff); | |
562 | ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK && | |
563 | mval[i].br_startblock != HOLESTARTBLOCK); | |
564 | ASSERT(mval[i].br_state == XFS_EXT_NORM || | |
565 | mval[i].br_state == XFS_EXT_UNWRITTEN); | |
566 | } | |
567 | } | |
56b2de80 | 568 | |
49f693fa DC |
569 | #else |
570 | #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0) | |
571 | #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) | |
572 | #endif /* DEBUG */ | |
56b2de80 | 573 | |
49f693fa DC |
574 | /* |
575 | * bmap free list manipulation functions | |
576 | */ | |
56b2de80 | 577 | |
49f693fa DC |
578 | /* |
579 | * Add the extent to the list of extents to be free at transaction end. | |
580 | * The list is maintained sorted (by block number). | |
581 | */ | |
582 | void | |
583 | xfs_bmap_add_free( | |
85aec44f DW |
584 | struct xfs_mount *mp, |
585 | struct xfs_defer_ops *dfops, | |
586 | xfs_fsblock_t bno, | |
587 | xfs_filblks_t len, | |
588 | struct xfs_owner_info *oinfo) | |
49f693fa | 589 | { |
6f530e9a | 590 | struct xfs_extent_free_item *new; /* new element */ |
49f693fa DC |
591 | #ifdef DEBUG |
592 | xfs_agnumber_t agno; | |
593 | xfs_agblock_t agbno; | |
56b2de80 | 594 | |
49f693fa DC |
595 | ASSERT(bno != NULLFSBLOCK); |
596 | ASSERT(len > 0); | |
597 | ASSERT(len <= MAXEXTLEN); | |
598 | ASSERT(!isnullstartblock(bno)); | |
599 | agno = XFS_FSB_TO_AGNO(mp, bno); | |
600 | agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
601 | ASSERT(agno < mp->m_sb.sb_agcount); | |
602 | ASSERT(agbno < mp->m_sb.sb_agblocks); | |
603 | ASSERT(len < mp->m_sb.sb_agblocks); | |
604 | ASSERT(agbno + len <= mp->m_sb.sb_agblocks); | |
605 | #endif | |
606 | ASSERT(xfs_bmap_free_item_zone != NULL); | |
85aec44f | 607 | |
49f693fa | 608 | new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP); |
6f530e9a DW |
609 | new->xefi_startblock = bno; |
610 | new->xefi_blockcount = (xfs_extlen_t)len; | |
85aec44f DW |
611 | if (oinfo) |
612 | new->xefi_oinfo = *oinfo; | |
613 | else | |
614 | xfs_rmap_skip_owner_update(&new->xefi_oinfo); | |
a9da40de DW |
615 | trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0, |
616 | XFS_FSB_TO_AGBNO(mp, bno), len); | |
f33cea1a | 617 | xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list); |
49f693fa | 618 | } |
2bd0ea18 | 619 | |
49f693fa DC |
620 | /* |
621 | * Inode fork format manipulation functions | |
622 | */ | |
a2ceac1f | 623 | |
49f693fa DC |
624 | /* |
625 | * Transform a btree format file with only one leaf node, where the | |
626 | * extents list will fit in the inode, into an extents format file. | |
627 | * Since the file extents are already in-core, all we have to do is | |
628 | * give up the space for the btree root and pitch the leaf block. | |
629 | */ | |
630 | STATIC int /* error */ | |
631 | xfs_bmap_btree_to_extents( | |
632 | xfs_trans_t *tp, /* transaction pointer */ | |
633 | xfs_inode_t *ip, /* incore inode pointer */ | |
634 | xfs_btree_cur_t *cur, /* btree cursor */ | |
635 | int *logflagsp, /* inode logging flags */ | |
636 | int whichfork) /* data or attr fork */ | |
637 | { | |
638 | /* REFERENCED */ | |
639 | struct xfs_btree_block *cblock;/* child btree block */ | |
640 | xfs_fsblock_t cbno; /* child block number */ | |
641 | xfs_buf_t *cbp; /* child block's buffer */ | |
642 | int error; /* error return value */ | |
643 | xfs_ifork_t *ifp; /* inode fork data */ | |
644 | xfs_mount_t *mp; /* mount point structure */ | |
645 | __be64 *pp; /* ptr to block address */ | |
646 | struct xfs_btree_block *rblock;/* root btree block */ | |
85aec44f | 647 | struct xfs_owner_info oinfo; |
56b2de80 | 648 | |
49f693fa DC |
649 | mp = ip->i_mount; |
650 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1277a5e0 | 651 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa DC |
652 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); |
653 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
654 | rblock = ifp->if_broot; | |
655 | ASSERT(be16_to_cpu(rblock->bb_level) == 1); | |
656 | ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1); | |
657 | ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1); | |
658 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes); | |
659 | cbno = be64_to_cpu(*pp); | |
660 | *logflagsp = 0; | |
661 | #ifdef DEBUG | |
662 | if ((error = xfs_btree_check_lptr(cur, cbno, 1))) | |
663 | return error; | |
664 | #endif | |
665 | error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF, | |
666 | &xfs_bmbt_buf_ops); | |
667 | if (error) | |
668 | return error; | |
669 | cblock = XFS_BUF_TO_BLOCK(cbp); | |
670 | if ((error = xfs_btree_check_block(cur, cblock, 0, cbp))) | |
671 | return error; | |
85aec44f DW |
672 | xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork); |
673 | xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo); | |
49f693fa DC |
674 | ip->i_d.di_nblocks--; |
675 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L); | |
676 | xfs_trans_binval(tp, cbp); | |
677 | if (cur->bc_bufs[0] == cbp) | |
678 | cur->bc_bufs[0] = NULL; | |
679 | xfs_iroot_realloc(ip, -1, whichfork); | |
680 | ASSERT(ifp->if_broot == NULL); | |
681 | ASSERT((ifp->if_flags & XFS_IFBROOT) == 0); | |
682 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
683 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); | |
684 | return 0; | |
685 | } | |
2bd0ea18 NS |
686 | |
687 | /* | |
49f693fa DC |
688 | * Convert an extents-format file into a btree-format file. |
689 | * The new file will have a root block (in the inode) and a single child block. | |
2bd0ea18 | 690 | */ |
49f693fa DC |
691 | STATIC int /* error */ |
692 | xfs_bmap_extents_to_btree( | |
693 | xfs_trans_t *tp, /* transaction pointer */ | |
694 | xfs_inode_t *ip, /* incore inode pointer */ | |
695 | xfs_fsblock_t *firstblock, /* first-block-allocated */ | |
f33cea1a | 696 | struct xfs_defer_ops *dfops, /* blocks freed in xaction */ |
49f693fa DC |
697 | xfs_btree_cur_t **curp, /* cursor returned to caller */ |
698 | int wasdel, /* converting a delayed alloc */ | |
699 | int *logflagsp, /* inode logging flags */ | |
700 | int whichfork) /* data or attr fork */ | |
2bd0ea18 | 701 | { |
49f693fa DC |
702 | struct xfs_btree_block *ablock; /* allocated (child) bt block */ |
703 | xfs_buf_t *abp; /* buffer for ablock */ | |
704 | xfs_alloc_arg_t args; /* allocation arguments */ | |
705 | xfs_bmbt_rec_t *arp; /* child record pointer */ | |
706 | struct xfs_btree_block *block; /* btree root block */ | |
707 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
708 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ | |
709 | int error; /* error return value */ | |
710 | xfs_extnum_t i, cnt; /* extent record index */ | |
711 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
712 | xfs_bmbt_key_t *kp; /* root block key pointer */ | |
713 | xfs_mount_t *mp; /* mount structure */ | |
714 | xfs_extnum_t nextents; /* number of file extents */ | |
715 | xfs_bmbt_ptr_t *pp; /* root block address pointer */ | |
2bd0ea18 | 716 | |
5dfa5cd2 | 717 | mp = ip->i_mount; |
1277a5e0 | 718 | ASSERT(whichfork != XFS_COW_FORK); |
49f693fa DC |
719 | ifp = XFS_IFORK_PTR(ip, whichfork); |
720 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS); | |
56b2de80 | 721 | |
2bd0ea18 | 722 | /* |
49f693fa | 723 | * Make space in the inode incore. |
2bd0ea18 | 724 | */ |
49f693fa DC |
725 | xfs_iroot_realloc(ip, 1, whichfork); |
726 | ifp->if_flags |= XFS_IFBROOT; | |
56b2de80 | 727 | |
2bd0ea18 | 728 | /* |
49f693fa | 729 | * Fill in the root. |
2bd0ea18 | 730 | */ |
49f693fa | 731 | block = ifp->if_broot; |
5dfa5cd2 DC |
732 | if (xfs_sb_version_hascrc(&mp->m_sb)) |
733 | xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL, | |
734 | XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino, | |
735 | XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS); | |
736 | else | |
737 | xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL, | |
738 | XFS_BMAP_MAGIC, 1, 1, ip->i_ino, | |
739 | XFS_BTREE_LONG_PTRS); | |
56b2de80 | 740 | |
49f693fa DC |
741 | /* |
742 | * Need a cursor. Can't allocate until bb_level is filled in. | |
743 | */ | |
49f693fa DC |
744 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); |
745 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 746 | cur->bc_private.b.dfops = dfops; |
49f693fa DC |
747 | cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; |
748 | /* | |
749 | * Convert to a btree with two levels, one record in root. | |
750 | */ | |
751 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE); | |
752 | memset(&args, 0, sizeof(args)); | |
753 | args.tp = tp; | |
754 | args.mp = mp; | |
85aec44f | 755 | xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork); |
49f693fa DC |
756 | args.firstblock = *firstblock; |
757 | if (*firstblock == NULLFSBLOCK) { | |
758 | args.type = XFS_ALLOCTYPE_START_BNO; | |
759 | args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino); | |
f33cea1a | 760 | } else if (dfops->dop_low) { |
68d06dce | 761 | try_another_ag: |
49f693fa DC |
762 | args.type = XFS_ALLOCTYPE_START_BNO; |
763 | args.fsbno = *firstblock; | |
764 | } else { | |
765 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
766 | args.fsbno = *firstblock; | |
767 | } | |
768 | args.minlen = args.maxlen = args.prod = 1; | |
769 | args.wasdel = wasdel; | |
770 | *logflagsp = 0; | |
771 | if ((error = xfs_alloc_vextent(&args))) { | |
772 | xfs_iroot_realloc(ip, -1, whichfork); | |
773 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
774 | return error; | |
2bd0ea18 | 775 | } |
68d06dce DW |
776 | |
777 | /* | |
778 | * During a CoW operation, the allocation and bmbt updates occur in | |
779 | * different transactions. The mapping code tries to put new bmbt | |
780 | * blocks near extents being mapped, but the only way to guarantee this | |
781 | * is if the alloc and the mapping happen in a single transaction that | |
782 | * has a block reservation. That isn't the case here, so if we run out | |
783 | * of space we'll try again with another AG. | |
784 | */ | |
785 | if (xfs_sb_version_hasreflink(&cur->bc_mp->m_sb) && | |
786 | args.fsbno == NULLFSBLOCK && | |
787 | args.type == XFS_ALLOCTYPE_NEAR_BNO) { | |
788 | dfops->dop_low = true; | |
789 | goto try_another_ag; | |
790 | } | |
2bd0ea18 | 791 | /* |
49f693fa | 792 | * Allocation can't fail, the space was reserved. |
2bd0ea18 | 793 | */ |
49f693fa DC |
794 | ASSERT(args.fsbno != NULLFSBLOCK); |
795 | ASSERT(*firstblock == NULLFSBLOCK || | |
796 | args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) || | |
f33cea1a | 797 | (dfops->dop_low && |
49f693fa DC |
798 | args.agno > XFS_FSB_TO_AGNO(mp, *firstblock))); |
799 | *firstblock = cur->bc_private.b.firstblock = args.fsbno; | |
800 | cur->bc_private.b.allocated++; | |
801 | ip->i_d.di_nblocks++; | |
802 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L); | |
803 | abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0); | |
2bd0ea18 | 804 | /* |
49f693fa | 805 | * Fill in the child block. |
2bd0ea18 | 806 | */ |
49f693fa DC |
807 | abp->b_ops = &xfs_bmbt_buf_ops; |
808 | ablock = XFS_BUF_TO_BLOCK(abp); | |
5dfa5cd2 DC |
809 | if (xfs_sb_version_hascrc(&mp->m_sb)) |
810 | xfs_btree_init_block_int(mp, ablock, abp->b_bn, | |
811 | XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino, | |
812 | XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS); | |
813 | else | |
814 | xfs_btree_init_block_int(mp, ablock, abp->b_bn, | |
815 | XFS_BMAP_MAGIC, 0, 0, ip->i_ino, | |
816 | XFS_BTREE_LONG_PTRS); | |
817 | ||
49f693fa | 818 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); |
d09d4e5f | 819 | nextents = xfs_iext_count(ifp); |
49f693fa DC |
820 | for (cnt = i = 0; i < nextents; i++) { |
821 | ep = xfs_iext_get_ext(ifp, i); | |
822 | if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) { | |
823 | arp->l0 = cpu_to_be64(ep->l0); | |
824 | arp->l1 = cpu_to_be64(ep->l1); | |
825 | arp++; cnt++; | |
826 | } | |
827 | } | |
828 | ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork)); | |
829 | xfs_btree_set_numrecs(ablock, cnt); | |
56b2de80 | 830 | |
49f693fa DC |
831 | /* |
832 | * Fill in the root key and pointer. | |
833 | */ | |
834 | kp = XFS_BMBT_KEY_ADDR(mp, block, 1); | |
835 | arp = XFS_BMBT_REC_ADDR(mp, ablock, 1); | |
836 | kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp)); | |
837 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur, | |
838 | be16_to_cpu(block->bb_level))); | |
839 | *pp = cpu_to_be64(args.fsbno); | |
2bd0ea18 | 840 | |
49f693fa DC |
841 | /* |
842 | * Do all this logging at the end so that | |
843 | * the root is at the right level. | |
844 | */ | |
5dfa5cd2 | 845 | xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS); |
613e6057 | 846 | xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs)); |
49f693fa DC |
847 | ASSERT(*curp == NULL); |
848 | *curp = cur; | |
849 | *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork); | |
850 | return 0; | |
851 | } | |
a2ceac1f | 852 | |
49f693fa DC |
853 | /* |
854 | * Convert a local file to an extents file. | |
855 | * This code is out of bounds for data forks of regular files, | |
856 | * since the file data needs to get logged so things will stay consistent. | |
857 | * (The bmap-level manipulations are ok, though). | |
858 | */ | |
3f17ed4b DC |
859 | void |
860 | xfs_bmap_local_to_extents_empty( | |
861 | struct xfs_inode *ip, | |
862 | int whichfork) | |
863 | { | |
864 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
865 | ||
1277a5e0 | 866 | ASSERT(whichfork != XFS_COW_FORK); |
3f17ed4b DC |
867 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); |
868 | ASSERT(ifp->if_bytes == 0); | |
869 | ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0); | |
870 | ||
ff105f75 | 871 | xfs_bmap_forkoff_reset(ip, whichfork); |
3f17ed4b DC |
872 | ifp->if_flags &= ~XFS_IFINLINE; |
873 | ifp->if_flags |= XFS_IFEXTENTS; | |
874 | XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS); | |
875 | } | |
876 | ||
877 | ||
49f693fa DC |
878 | STATIC int /* error */ |
879 | xfs_bmap_local_to_extents( | |
880 | xfs_trans_t *tp, /* transaction pointer */ | |
881 | xfs_inode_t *ip, /* incore inode pointer */ | |
882 | xfs_fsblock_t *firstblock, /* first block allocated in xaction */ | |
883 | xfs_extlen_t total, /* total blocks needed by transaction */ | |
884 | int *logflagsp, /* inode logging flags */ | |
885 | int whichfork, | |
5dfa5cd2 DC |
886 | void (*init_fn)(struct xfs_trans *tp, |
887 | struct xfs_buf *bp, | |
49f693fa DC |
888 | struct xfs_inode *ip, |
889 | struct xfs_ifork *ifp)) | |
890 | { | |
3f17ed4b | 891 | int error = 0; |
49f693fa DC |
892 | int flags; /* logging flags returned */ |
893 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
3f17ed4b DC |
894 | xfs_alloc_arg_t args; /* allocation arguments */ |
895 | xfs_buf_t *bp; /* buffer for extent block */ | |
896 | xfs_bmbt_rec_host_t *ep; /* extent record pointer */ | |
2bd0ea18 | 897 | |
49f693fa DC |
898 | /* |
899 | * We don't want to deal with the case of keeping inode data inline yet. | |
900 | * So sending the data fork of a regular inode is invalid. | |
901 | */ | |
e37bf53c | 902 | ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK)); |
49f693fa DC |
903 | ifp = XFS_IFORK_PTR(ip, whichfork); |
904 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
3f17ed4b DC |
905 | |
906 | if (!ifp->if_bytes) { | |
907 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
908 | flags = XFS_ILOG_CORE; | |
909 | goto done; | |
910 | } | |
911 | ||
49f693fa DC |
912 | flags = 0; |
913 | error = 0; | |
3f17ed4b DC |
914 | ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == |
915 | XFS_IFINLINE); | |
916 | memset(&args, 0, sizeof(args)); | |
917 | args.tp = tp; | |
918 | args.mp = ip->i_mount; | |
85aec44f | 919 | xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0); |
3f17ed4b DC |
920 | args.firstblock = *firstblock; |
921 | /* | |
922 | * Allocate a block. We know we need only one, since the | |
923 | * file currently fits in an inode. | |
924 | */ | |
925 | if (*firstblock == NULLFSBLOCK) { | |
68d06dce | 926 | try_another_ag: |
3f17ed4b DC |
927 | args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino); |
928 | args.type = XFS_ALLOCTYPE_START_BNO; | |
49f693fa | 929 | } else { |
3f17ed4b DC |
930 | args.fsbno = *firstblock; |
931 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
2bd0ea18 | 932 | } |
3f17ed4b DC |
933 | args.total = total; |
934 | args.minlen = args.maxlen = args.prod = 1; | |
935 | error = xfs_alloc_vextent(&args); | |
936 | if (error) | |
937 | goto done; | |
938 | ||
68d06dce DW |
939 | /* |
940 | * During a CoW operation, the allocation and bmbt updates occur in | |
941 | * different transactions. The mapping code tries to put new bmbt | |
942 | * blocks near extents being mapped, but the only way to guarantee this | |
943 | * is if the alloc and the mapping happen in a single transaction that | |
944 | * has a block reservation. That isn't the case here, so if we run out | |
945 | * of space we'll try again with another AG. | |
946 | */ | |
947 | if (xfs_sb_version_hasreflink(&ip->i_mount->m_sb) && | |
948 | args.fsbno == NULLFSBLOCK && | |
949 | args.type == XFS_ALLOCTYPE_NEAR_BNO) { | |
950 | goto try_another_ag; | |
951 | } | |
3f17ed4b DC |
952 | /* Can't fail, the space was reserved. */ |
953 | ASSERT(args.fsbno != NULLFSBLOCK); | |
954 | ASSERT(args.len == 1); | |
955 | *firstblock = args.fsbno; | |
956 | bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0); | |
957 | ||
19ebedcf | 958 | /* |
f44fbde0 | 959 | * Initialize the block, copy the data and log the remote buffer. |
19ebedcf | 960 | * |
f44fbde0 BF |
961 | * The callout is responsible for logging because the remote format |
962 | * might differ from the local format and thus we don't know how much to | |
963 | * log here. Note that init_fn must also set the buffer log item type | |
964 | * correctly. | |
19ebedcf | 965 | */ |
3f17ed4b DC |
966 | init_fn(tp, bp, ip, ifp); |
967 | ||
f44fbde0 | 968 | /* account for the change in fork size */ |
3f17ed4b DC |
969 | xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); |
970 | xfs_bmap_local_to_extents_empty(ip, whichfork); | |
49f693fa | 971 | flags |= XFS_ILOG_CORE; |
3f17ed4b DC |
972 | |
973 | xfs_iext_add(ifp, 0, 1); | |
974 | ep = xfs_iext_get_ext(ifp, 0); | |
975 | xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM); | |
976 | trace_xfs_bmap_post_update(ip, 0, | |
977 | whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0, | |
978 | _THIS_IP_); | |
979 | XFS_IFORK_NEXT_SET(ip, whichfork, 1); | |
980 | ip->i_d.di_nblocks = 1; | |
981 | xfs_trans_mod_dquot_byino(tp, ip, | |
982 | XFS_TRANS_DQ_BCOUNT, 1L); | |
983 | flags |= xfs_ilog_fext(whichfork); | |
984 | ||
49f693fa DC |
985 | done: |
986 | *logflagsp = flags; | |
987 | return error; | |
2bd0ea18 NS |
988 | } |
989 | ||
990 | /* | |
49f693fa | 991 | * Called from xfs_bmap_add_attrfork to handle btree format files. |
2bd0ea18 | 992 | */ |
49f693fa DC |
993 | STATIC int /* error */ |
994 | xfs_bmap_add_attrfork_btree( | |
995 | xfs_trans_t *tp, /* transaction pointer */ | |
996 | xfs_inode_t *ip, /* incore inode pointer */ | |
997 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 998 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa | 999 | int *flags) /* inode logging flags */ |
2bd0ea18 | 1000 | { |
49f693fa DC |
1001 | xfs_btree_cur_t *cur; /* btree cursor */ |
1002 | int error; /* error return value */ | |
1003 | xfs_mount_t *mp; /* file system mount struct */ | |
1004 | int stat; /* newroot status */ | |
56b2de80 | 1005 | |
49f693fa DC |
1006 | mp = ip->i_mount; |
1007 | if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip)) | |
1008 | *flags |= XFS_ILOG_DBROOT; | |
1009 | else { | |
1010 | cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK); | |
f33cea1a | 1011 | cur->bc_private.b.dfops = dfops; |
49f693fa DC |
1012 | cur->bc_private.b.firstblock = *firstblock; |
1013 | if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat))) | |
1014 | goto error0; | |
1015 | /* must be at least one entry */ | |
19ebedcf | 1016 | XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0); |
49f693fa DC |
1017 | if ((error = xfs_btree_new_iroot(cur, flags, &stat))) |
1018 | goto error0; | |
1019 | if (stat == 0) { | |
1020 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
12b53197 | 1021 | return -ENOSPC; |
49f693fa DC |
1022 | } |
1023 | *firstblock = cur->bc_private.b.firstblock; | |
1024 | cur->bc_private.b.allocated = 0; | |
1025 | xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); | |
2bd0ea18 | 1026 | } |
49f693fa DC |
1027 | return 0; |
1028 | error0: | |
1029 | xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); | |
1030 | return error; | |
1031 | } | |
56b2de80 | 1032 | |
49f693fa DC |
1033 | /* |
1034 | * Called from xfs_bmap_add_attrfork to handle extents format files. | |
1035 | */ | |
1036 | STATIC int /* error */ | |
1037 | xfs_bmap_add_attrfork_extents( | |
1038 | xfs_trans_t *tp, /* transaction pointer */ | |
1039 | xfs_inode_t *ip, /* incore inode pointer */ | |
1040 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 1041 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa DC |
1042 | int *flags) /* inode logging flags */ |
1043 | { | |
1044 | xfs_btree_cur_t *cur; /* bmap btree cursor */ | |
1045 | int error; /* error return value */ | |
1046 | ||
1047 | if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip)) | |
1048 | return 0; | |
1049 | cur = NULL; | |
f33cea1a | 1050 | error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0, |
49f693fa DC |
1051 | flags, XFS_DATA_FORK); |
1052 | if (cur) { | |
1053 | cur->bc_private.b.allocated = 0; | |
1054 | xfs_btree_del_cursor(cur, | |
1055 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
2bd0ea18 | 1056 | } |
49f693fa DC |
1057 | return error; |
1058 | } | |
56b2de80 | 1059 | |
49f693fa DC |
1060 | /* |
1061 | * Called from xfs_bmap_add_attrfork to handle local format files. Each | |
1062 | * different data fork content type needs a different callout to do the | |
1063 | * conversion. Some are basic and only require special block initialisation | |
1064 | * callouts for the data formating, others (directories) are so specialised they | |
1065 | * handle everything themselves. | |
1066 | * | |
1067 | * XXX (dgc): investigate whether directory conversion can use the generic | |
1068 | * formatting callout. It should be possible - it's just a very complex | |
5dfa5cd2 | 1069 | * formatter. |
49f693fa DC |
1070 | */ |
1071 | STATIC int /* error */ | |
1072 | xfs_bmap_add_attrfork_local( | |
1073 | xfs_trans_t *tp, /* transaction pointer */ | |
1074 | xfs_inode_t *ip, /* incore inode pointer */ | |
1075 | xfs_fsblock_t *firstblock, /* first block allocated */ | |
f33cea1a | 1076 | struct xfs_defer_ops *dfops, /* blocks to free at commit */ |
49f693fa DC |
1077 | int *flags) /* inode logging flags */ |
1078 | { | |
1079 | xfs_da_args_t dargs; /* args for dir/attr code */ | |
56b2de80 | 1080 | |
49f693fa DC |
1081 | if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip)) |
1082 | return 0; | |
a2ceac1f | 1083 | |
e37bf53c | 1084 | if (S_ISDIR(VFS_I(ip)->i_mode)) { |
49f693fa | 1085 | memset(&dargs, 0, sizeof(dargs)); |
ff105f75 | 1086 | dargs.geo = ip->i_mount->m_dir_geo; |
49f693fa DC |
1087 | dargs.dp = ip; |
1088 | dargs.firstblock = firstblock; | |
f33cea1a | 1089 | dargs.dfops = dfops; |
ff105f75 | 1090 | dargs.total = dargs.geo->fsbcount; |
49f693fa DC |
1091 | dargs.whichfork = XFS_DATA_FORK; |
1092 | dargs.trans = tp; | |
1093 | return xfs_dir2_sf_to_block(&dargs); | |
1094 | } | |
2bd0ea18 | 1095 | |
e37bf53c | 1096 | if (S_ISLNK(VFS_I(ip)->i_mode)) |
49f693fa DC |
1097 | return xfs_bmap_local_to_extents(tp, ip, firstblock, 1, |
1098 | flags, XFS_DATA_FORK, | |
1099 | xfs_symlink_local_to_remote); | |
56b2de80 | 1100 | |
3f17ed4b DC |
1101 | /* should only be called for types that support local format data */ |
1102 | ASSERT(0); | |
12b53197 | 1103 | return -EFSCORRUPTED; |
49f693fa | 1104 | } |
2bd0ea18 | 1105 | |
49f693fa DC |
1106 | /* |
1107 | * Convert inode from non-attributed to attributed. | |
1108 | * Must not be in a transaction, ip must not be locked. | |
1109 | */ | |
1110 | int /* error code */ | |
1111 | xfs_bmap_add_attrfork( | |
1112 | xfs_inode_t *ip, /* incore inode pointer */ | |
1113 | int size, /* space new attribute needs */ | |
1114 | int rsvd) /* xact may use reserved blks */ | |
1115 | { | |
1116 | xfs_fsblock_t firstblock; /* 1st block/ag allocated */ | |
f33cea1a | 1117 | struct xfs_defer_ops dfops; /* freed extent records */ |
49f693fa DC |
1118 | xfs_mount_t *mp; /* mount structure */ |
1119 | xfs_trans_t *tp; /* transaction pointer */ | |
1120 | int blks; /* space reservation */ | |
1121 | int version = 1; /* superblock attr version */ | |
49f693fa DC |
1122 | int logflags; /* logging flags */ |
1123 | int error; /* error return value */ | |
56b2de80 | 1124 | |
49f693fa | 1125 | ASSERT(XFS_IFORK_Q(ip) == 0); |
2bd0ea18 | 1126 | |
49f693fa DC |
1127 | mp = ip->i_mount; |
1128 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); | |
9074815c | 1129 | |
49f693fa | 1130 | blks = XFS_ADDAFORK_SPACE_RES(mp); |
9074815c CH |
1131 | |
1132 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0, | |
1133 | rsvd ? XFS_TRANS_RESERVE : 0, &tp); | |
1134 | if (error) | |
ff105f75 | 1135 | return error; |
9074815c | 1136 | |
49f693fa DC |
1137 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1138 | error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ? | |
1139 | XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES : | |
1140 | XFS_QMOPT_RES_REGBLKS); | |
ff105f75 DC |
1141 | if (error) |
1142 | goto trans_cancel; | |
49f693fa | 1143 | if (XFS_IFORK_Q(ip)) |
ff105f75 | 1144 | goto trans_cancel; |
49f693fa | 1145 | if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) { |
2bd0ea18 | 1146 | /* |
49f693fa | 1147 | * For inodes coming from pre-6.2 filesystems. |
2bd0ea18 | 1148 | */ |
49f693fa DC |
1149 | ASSERT(ip->i_d.di_aformat == 0); |
1150 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
5e656dbb | 1151 | } |
49f693fa | 1152 | ASSERT(ip->i_d.di_anextents == 0); |
a2ceac1f | 1153 | |
ff105f75 | 1154 | xfs_trans_ijoin(tp, ip, 0); |
49f693fa | 1155 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
a2ceac1f | 1156 | |
49f693fa DC |
1157 | switch (ip->i_d.di_format) { |
1158 | case XFS_DINODE_FMT_DEV: | |
1159 | ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3; | |
1160 | break; | |
1161 | case XFS_DINODE_FMT_UUID: | |
1162 | ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3; | |
1163 | break; | |
1164 | case XFS_DINODE_FMT_LOCAL: | |
1165 | case XFS_DINODE_FMT_EXTENTS: | |
1166 | case XFS_DINODE_FMT_BTREE: | |
1167 | ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size); | |
1168 | if (!ip->i_d.di_forkoff) | |
1169 | ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3; | |
1170 | else if (mp->m_flags & XFS_MOUNT_ATTR2) | |
1171 | version = 2; | |
1172 | break; | |
1173 | default: | |
1174 | ASSERT(0); | |
12b53197 | 1175 | error = -EINVAL; |
ff105f75 | 1176 | goto trans_cancel; |
a2ceac1f DC |
1177 | } |
1178 | ||
49f693fa DC |
1179 | ASSERT(ip->i_afp == NULL); |
1180 | ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP); | |
1181 | ip->i_afp->if_flags = XFS_IFEXTENTS; | |
1182 | logflags = 0; | |
f33cea1a | 1183 | xfs_defer_init(&dfops, &firstblock); |
49f693fa DC |
1184 | switch (ip->i_d.di_format) { |
1185 | case XFS_DINODE_FMT_LOCAL: | |
f33cea1a | 1186 | error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops, |
49f693fa DC |
1187 | &logflags); |
1188 | break; | |
1189 | case XFS_DINODE_FMT_EXTENTS: | |
1190 | error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock, | |
f33cea1a | 1191 | &dfops, &logflags); |
49f693fa DC |
1192 | break; |
1193 | case XFS_DINODE_FMT_BTREE: | |
f33cea1a | 1194 | error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops, |
49f693fa DC |
1195 | &logflags); |
1196 | break; | |
1197 | default: | |
1198 | error = 0; | |
1199 | break; | |
1200 | } | |
1201 | if (logflags) | |
1202 | xfs_trans_log_inode(tp, ip, logflags); | |
1203 | if (error) | |
ff105f75 | 1204 | goto bmap_cancel; |
49f693fa DC |
1205 | if (!xfs_sb_version_hasattr(&mp->m_sb) || |
1206 | (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) { | |
19ebedcf | 1207 | bool log_sb = false; |
a2ceac1f | 1208 | |
49f693fa DC |
1209 | spin_lock(&mp->m_sb_lock); |
1210 | if (!xfs_sb_version_hasattr(&mp->m_sb)) { | |
1211 | xfs_sb_version_addattr(&mp->m_sb); | |
19ebedcf | 1212 | log_sb = true; |
49f693fa DC |
1213 | } |
1214 | if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) { | |
1215 | xfs_sb_version_addattr2(&mp->m_sb); | |
19ebedcf | 1216 | log_sb = true; |
49f693fa | 1217 | } |
19ebedcf DC |
1218 | spin_unlock(&mp->m_sb_lock); |
1219 | if (log_sb) | |
1220 | xfs_log_sb(tp); | |
49f693fa DC |
1221 | } |
1222 | ||
f33cea1a | 1223 | error = xfs_defer_finish(&tp, &dfops, NULL); |
49f693fa | 1224 | if (error) |
ff105f75 | 1225 | goto bmap_cancel; |
de5a3f46 | 1226 | error = xfs_trans_commit(tp); |
ff105f75 DC |
1227 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1228 | return error; | |
1229 | ||
1230 | bmap_cancel: | |
f33cea1a | 1231 | xfs_defer_cancel(&dfops); |
ff105f75 | 1232 | trans_cancel: |
3d7434fe | 1233 | xfs_trans_cancel(tp); |
49f693fa | 1234 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5e656dbb | 1235 | return error; |
2bd0ea18 NS |
1236 | } |
1237 | ||
399ab595 | 1238 | /* |
49f693fa DC |
1239 | * Internal and external extent tree search functions. |
1240 | */ | |
399ab595 | 1241 | |
49f693fa DC |
1242 | /* |
1243 | * Read in the extents to if_extents. | |
1244 | * All inode fields are set up by caller, we just traverse the btree | |
1245 | * and copy the records in. If the file system cannot contain unwritten | |
1246 | * extents, the records are checked for no "state" flags. | |
1247 | */ | |
1248 | int /* error */ | |
1249 | xfs_bmap_read_extents( | |
1250 | xfs_trans_t *tp, /* transaction pointer */ | |
1251 | xfs_inode_t *ip, /* incore inode */ | |
1252 | int whichfork) /* data or attr fork */ | |
1253 | { | |
1254 | struct xfs_btree_block *block; /* current btree block */ | |
1255 | xfs_fsblock_t bno; /* block # of "block" */ | |
1256 | xfs_buf_t *bp; /* buffer for "block" */ | |
1257 | int error; /* error return value */ | |
1258 | xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */ | |
1259 | xfs_extnum_t i, j; /* index into the extents list */ | |
1260 | xfs_ifork_t *ifp; /* fork structure */ | |
1261 | int level; /* btree level, for checking */ | |
1262 | xfs_mount_t *mp; /* file system mount structure */ | |
1263 | __be64 *pp; /* pointer to block address */ | |
1264 | /* REFERENCED */ | |
1265 | xfs_extnum_t room; /* number of entries there's room for */ | |
399ab595 | 1266 | |
49f693fa DC |
1267 | bno = NULLFSBLOCK; |
1268 | mp = ip->i_mount; | |
1269 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1270 | exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE : | |
1271 | XFS_EXTFMT_INODE(ip); | |
1272 | block = ifp->if_broot; | |
399ab595 | 1273 | /* |
49f693fa | 1274 | * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. |
399ab595 | 1275 | */ |
49f693fa DC |
1276 | level = be16_to_cpu(block->bb_level); |
1277 | ASSERT(level > 0); | |
1278 | pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); | |
1279 | bno = be64_to_cpu(*pp); | |
5a35bf2c | 1280 | ASSERT(bno != NULLFSBLOCK); |
49f693fa DC |
1281 | ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); |
1282 | ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); | |
399ab595 | 1283 | /* |
49f693fa DC |
1284 | * Go down the tree until leaf level is reached, following the first |
1285 | * pointer (leftmost) at each level. | |
399ab595 | 1286 | */ |
49f693fa DC |
1287 | while (level-- > 0) { |
1288 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1289 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1290 | if (error) | |
1291 | return error; | |
1292 | block = XFS_BUF_TO_BLOCK(bp); | |
49f693fa DC |
1293 | if (level == 0) |
1294 | break; | |
1295 | pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); | |
1296 | bno = be64_to_cpu(*pp); | |
19ebedcf DC |
1297 | XFS_WANT_CORRUPTED_GOTO(mp, |
1298 | XFS_FSB_SANITY_CHECK(mp, bno), error0); | |
49f693fa | 1299 | xfs_trans_brelse(tp, bp); |
399ab595 NS |
1300 | } |
1301 | /* | |
49f693fa | 1302 | * Here with bp and block set to the leftmost leaf node in the tree. |
399ab595 | 1303 | */ |
d09d4e5f | 1304 | room = xfs_iext_count(ifp); |
49f693fa | 1305 | i = 0; |
399ab595 | 1306 | /* |
49f693fa | 1307 | * Loop over all leaf nodes. Copy information to the extent records. |
399ab595 | 1308 | */ |
49f693fa DC |
1309 | for (;;) { |
1310 | xfs_bmbt_rec_t *frp; | |
1311 | xfs_fsblock_t nextbno; | |
1312 | xfs_extnum_t num_recs; | |
1313 | xfs_extnum_t start; | |
399ab595 | 1314 | |
49f693fa DC |
1315 | num_recs = xfs_btree_get_numrecs(block); |
1316 | if (unlikely(i + num_recs > room)) { | |
1317 | ASSERT(i + num_recs <= room); | |
1318 | xfs_warn(ip->i_mount, | |
1319 | "corrupt dinode %Lu, (btree extents).", | |
1320 | (unsigned long long) ip->i_ino); | |
1321 | XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)", | |
1322 | XFS_ERRLEVEL_LOW, ip->i_mount, block); | |
1323 | goto error0; | |
399ab595 NS |
1324 | } |
1325 | /* | |
49f693fa | 1326 | * Read-ahead the next leaf block, if any. |
399ab595 | 1327 | */ |
49f693fa DC |
1328 | nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); |
1329 | if (nextbno != NULLFSBLOCK) | |
1330 | xfs_btree_reada_bufl(mp, nextbno, 1, | |
1331 | &xfs_bmbt_buf_ops); | |
399ab595 | 1332 | /* |
49f693fa | 1333 | * Copy records into the extent records. |
399ab595 | 1334 | */ |
49f693fa DC |
1335 | frp = XFS_BMBT_REC_ADDR(mp, block, 1); |
1336 | start = i; | |
1337 | for (j = 0; j < num_recs; j++, i++, frp++) { | |
1338 | xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i); | |
1339 | trp->l0 = be64_to_cpu(frp->l0); | |
1340 | trp->l1 = be64_to_cpu(frp->l1); | |
1341 | } | |
1342 | if (exntf == XFS_EXTFMT_NOSTATE) { | |
1343 | /* | |
1344 | * Check all attribute bmap btree records and | |
1345 | * any "older" data bmap btree records for a | |
1346 | * set bit in the "extent flag" position. | |
1347 | */ | |
1348 | if (unlikely(xfs_check_nostate_extents(ifp, | |
1349 | start, num_recs))) { | |
1350 | XFS_ERROR_REPORT("xfs_bmap_read_extents(2)", | |
1351 | XFS_ERRLEVEL_LOW, | |
1352 | ip->i_mount); | |
1353 | goto error0; | |
1354 | } | |
399ab595 | 1355 | } |
49f693fa DC |
1356 | xfs_trans_brelse(tp, bp); |
1357 | bno = nextbno; | |
399ab595 | 1358 | /* |
49f693fa | 1359 | * If we've reached the end, stop. |
399ab595 | 1360 | */ |
49f693fa DC |
1361 | if (bno == NULLFSBLOCK) |
1362 | break; | |
1363 | error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, | |
1364 | XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops); | |
1365 | if (error) | |
1366 | return error; | |
1367 | block = XFS_BUF_TO_BLOCK(bp); | |
399ab595 | 1368 | } |
d09d4e5f | 1369 | ASSERT(i == xfs_iext_count(ifp)); |
49f693fa DC |
1370 | ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork)); |
1371 | XFS_BMAP_TRACE_EXLIST(ip, i, whichfork); | |
1372 | return 0; | |
1373 | error0: | |
1374 | xfs_trans_brelse(tp, bp); | |
12b53197 | 1375 | return -EFSCORRUPTED; |
49f693fa | 1376 | } |
399ab595 | 1377 | |
49f693fa DC |
1378 | /* |
1379 | * Returns the file-relative block number of the first unused block(s) | |
1380 | * in the file with at least "len" logically contiguous blocks free. | |
1381 | * This is the lowest-address hole if the file has holes, else the first block | |
1382 | * past the end of file. | |
1383 | * Return 0 if the file is currently local (in-inode). | |
1384 | */ | |
1385 | int /* error */ | |
1386 | xfs_bmap_first_unused( | |
1387 | xfs_trans_t *tp, /* transaction pointer */ | |
1388 | xfs_inode_t *ip, /* incore inode */ | |
1389 | xfs_extlen_t len, /* size of hole to find */ | |
1390 | xfs_fileoff_t *first_unused, /* unused block */ | |
1391 | int whichfork) /* data or attr fork */ | |
1392 | { | |
1393 | int error; /* error return value */ | |
1394 | int idx; /* extent record index */ | |
1395 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1396 | xfs_fileoff_t lastaddr; /* last block number seen */ | |
1397 | xfs_fileoff_t lowest; /* lowest useful block */ | |
1398 | xfs_fileoff_t max; /* starting useful block */ | |
1399 | xfs_fileoff_t off; /* offset for this block */ | |
1400 | xfs_extnum_t nextents; /* number of extent entries */ | |
1401 | ||
1402 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE || | |
1403 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS || | |
1404 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL); | |
1405 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { | |
1406 | *first_unused = 0; | |
1407 | return 0; | |
1408 | } | |
1409 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1410 | if (!(ifp->if_flags & XFS_IFEXTENTS) && | |
1411 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
1412 | return error; | |
1413 | lowest = *first_unused; | |
d09d4e5f | 1414 | nextents = xfs_iext_count(ifp); |
49f693fa DC |
1415 | for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) { |
1416 | xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx); | |
1417 | off = xfs_bmbt_get_startoff(ep); | |
2bd0ea18 | 1418 | /* |
49f693fa | 1419 | * See if the hole before this extent will work. |
2bd0ea18 | 1420 | */ |
49f693fa DC |
1421 | if (off >= lowest + len && off - max >= len) { |
1422 | *first_unused = max; | |
1423 | return 0; | |
2bd0ea18 | 1424 | } |
49f693fa DC |
1425 | lastaddr = off + xfs_bmbt_get_blockcount(ep); |
1426 | max = XFS_FILEOFF_MAX(lastaddr, lowest); | |
1427 | } | |
1428 | *first_unused = max; | |
1429 | return 0; | |
1430 | } | |
1431 | ||
1432 | /* | |
e6d77a21 | 1433 | * Returns the file-relative block number of the last block - 1 before |
49f693fa DC |
1434 | * last_block (input value) in the file. |
1435 | * This is not based on i_size, it is based on the extent records. | |
1436 | * Returns 0 for local files, as they do not have extent records. | |
1437 | */ | |
1438 | int /* error */ | |
1439 | xfs_bmap_last_before( | |
b0385364 CH |
1440 | struct xfs_trans *tp, /* transaction pointer */ |
1441 | struct xfs_inode *ip, /* incore inode */ | |
1442 | xfs_fileoff_t *last_block, /* last block */ | |
1443 | int whichfork) /* data or attr fork */ | |
49f693fa | 1444 | { |
b0385364 CH |
1445 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
1446 | struct xfs_bmbt_irec got; | |
1447 | xfs_extnum_t idx; | |
1448 | int error; | |
49f693fa | 1449 | |
b0385364 CH |
1450 | switch (XFS_IFORK_FORMAT(ip, whichfork)) { |
1451 | case XFS_DINODE_FMT_LOCAL: | |
49f693fa DC |
1452 | *last_block = 0; |
1453 | return 0; | |
b0385364 CH |
1454 | case XFS_DINODE_FMT_BTREE: |
1455 | case XFS_DINODE_FMT_EXTENTS: | |
1456 | break; | |
1457 | default: | |
1458 | return -EIO; | |
49f693fa | 1459 | } |
b0385364 CH |
1460 | |
1461 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
1462 | error = xfs_iread_extents(tp, ip, whichfork); | |
1463 | if (error) | |
1464 | return error; | |
2bd0ea18 | 1465 | } |
b0385364 CH |
1466 | |
1467 | if (xfs_iext_lookup_extent(ip, ifp, *last_block - 1, &idx, &got)) { | |
1468 | if (got.br_startoff <= *last_block - 1) | |
1469 | return 0; | |
1470 | } | |
1471 | ||
1472 | if (xfs_iext_get_extent(ifp, idx - 1, &got)) { | |
1473 | *last_block = got.br_startoff + got.br_blockcount; | |
1474 | return 0; | |
1475 | } | |
1476 | ||
1477 | *last_block = 0; | |
49f693fa | 1478 | return 0; |
5e656dbb BN |
1479 | } |
1480 | ||
613e6057 | 1481 | int |
49f693fa DC |
1482 | xfs_bmap_last_extent( |
1483 | struct xfs_trans *tp, | |
1484 | struct xfs_inode *ip, | |
1485 | int whichfork, | |
1486 | struct xfs_bmbt_irec *rec, | |
1487 | int *is_empty) | |
56b2de80 | 1488 | { |
49f693fa | 1489 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
56b2de80 | 1490 | int error; |
49f693fa | 1491 | int nextents; |
56b2de80 | 1492 | |
49f693fa DC |
1493 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
1494 | error = xfs_iread_extents(tp, ip, whichfork); | |
1495 | if (error) | |
1496 | return error; | |
1497 | } | |
1498 | ||
d09d4e5f | 1499 | nextents = xfs_iext_count(ifp); |
49f693fa DC |
1500 | if (nextents == 0) { |
1501 | *is_empty = 1; | |
1502 | return 0; | |
1503 | } | |
1504 | ||
1505 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec); | |
1506 | *is_empty = 0; | |
1507 | return 0; | |
1508 | } | |
1509 | ||
1510 | /* | |
1511 | * Check the last inode extent to determine whether this allocation will result | |
1512 | * in blocks being allocated at the end of the file. When we allocate new data | |
1513 | * blocks at the end of the file which do not start at the previous data block, | |
1514 | * we will try to align the new blocks at stripe unit boundaries. | |
1515 | * | |
ff105f75 | 1516 | * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be |
49f693fa DC |
1517 | * at, or past the EOF. |
1518 | */ | |
1519 | STATIC int | |
1520 | xfs_bmap_isaeof( | |
1521 | struct xfs_bmalloca *bma, | |
1522 | int whichfork) | |
1523 | { | |
1524 | struct xfs_bmbt_irec rec; | |
1525 | int is_empty; | |
1526 | int error; | |
1527 | ||
1528 | bma->aeof = 0; | |
1529 | error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec, | |
1530 | &is_empty); | |
ff105f75 | 1531 | if (error) |
49f693fa | 1532 | return error; |
56b2de80 | 1533 | |
ff105f75 DC |
1534 | if (is_empty) { |
1535 | bma->aeof = 1; | |
1536 | return 0; | |
1537 | } | |
1538 | ||
56b2de80 | 1539 | /* |
49f693fa DC |
1540 | * Check if we are allocation or past the last extent, or at least into |
1541 | * the last delayed allocated extent. | |
56b2de80 | 1542 | */ |
49f693fa DC |
1543 | bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount || |
1544 | (bma->offset >= rec.br_startoff && | |
1545 | isnullstartblock(rec.br_startblock)); | |
1546 | return 0; | |
1547 | } | |
56b2de80 | 1548 | |
49f693fa DC |
1549 | /* |
1550 | * Returns the file-relative block number of the first block past eof in | |
1551 | * the file. This is not based on i_size, it is based on the extent records. | |
1552 | * Returns 0 for local files, as they do not have extent records. | |
1553 | */ | |
1554 | int | |
1555 | xfs_bmap_last_offset( | |
49f693fa DC |
1556 | struct xfs_inode *ip, |
1557 | xfs_fileoff_t *last_block, | |
1558 | int whichfork) | |
1559 | { | |
1560 | struct xfs_bmbt_irec rec; | |
1561 | int is_empty; | |
1562 | int error; | |
1563 | ||
1564 | *last_block = 0; | |
1565 | ||
1566 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) | |
1567 | return 0; | |
1568 | ||
1569 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE && | |
1570 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
12b53197 | 1571 | return -EIO; |
49f693fa DC |
1572 | |
1573 | error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty); | |
1574 | if (error || is_empty) | |
1575 | return error; | |
1576 | ||
1577 | *last_block = rec.br_startoff + rec.br_blockcount; | |
1578 | return 0; | |
1579 | } | |
1580 | ||
1581 | /* | |
1582 | * Returns whether the selected fork of the inode has exactly one | |
1583 | * block or not. For the data fork we check this matches di_size, | |
1584 | * implying the file's range is 0..bsize-1. | |
1585 | */ | |
1586 | int /* 1=>1 block, 0=>otherwise */ | |
1587 | xfs_bmap_one_block( | |
1588 | xfs_inode_t *ip, /* incore inode */ | |
1589 | int whichfork) /* data or attr fork */ | |
1590 | { | |
1591 | xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */ | |
1592 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1593 | int rval; /* return value */ | |
1594 | xfs_bmbt_irec_t s; /* internal version of extent */ | |
1595 | ||
1596 | #ifndef DEBUG | |
1597 | if (whichfork == XFS_DATA_FORK) | |
1598 | return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize; | |
1599 | #endif /* !DEBUG */ | |
1600 | if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1) | |
1601 | return 0; | |
1602 | if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
1603 | return 0; | |
1604 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
1605 | ASSERT(ifp->if_flags & XFS_IFEXTENTS); | |
1606 | ep = xfs_iext_get_ext(ifp, 0); | |
1607 | xfs_bmbt_get_all(ep, &s); | |
1608 | rval = s.br_startoff == 0 && s.br_blockcount == 1; | |
1609 | if (rval && whichfork == XFS_DATA_FORK) | |
1610 | ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize); | |
1611 | return rval; | |
1612 | } | |
1613 | ||
1614 | /* | |
1615 | * Extent tree manipulation functions used during allocation. | |
1616 | */ | |
1617 | ||
1618 | /* | |
1619 | * Convert a delayed allocation to a real allocation. | |
1620 | */ | |
1621 | STATIC int /* error */ | |
1622 | xfs_bmap_add_extent_delay_real( | |
1277a5e0 DW |
1623 | struct xfs_bmalloca *bma, |
1624 | int whichfork) | |
49f693fa DC |
1625 | { |
1626 | struct xfs_bmbt_irec *new = &bma->got; | |
1627 | int diff; /* temp value */ | |
1628 | xfs_bmbt_rec_host_t *ep; /* extent entry for idx */ | |
1629 | int error; /* error return value */ | |
1630 | int i; /* temp state */ | |
1631 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
1632 | xfs_fileoff_t new_endoff; /* end offset of new entry */ | |
1633 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
1634 | /* left is 0, right is 1, prev is 2 */ | |
1635 | int rval=0; /* return value (logging flags) */ | |
1636 | int state = 0;/* state bits, accessed thru macros */ | |
1637 | xfs_filblks_t da_new; /* new count del alloc blocks used */ | |
1638 | xfs_filblks_t da_old; /* old count del alloc blocks used */ | |
1639 | xfs_filblks_t temp=0; /* value for da_new calculations */ | |
1640 | xfs_filblks_t temp2=0;/* value for da_new calculations */ | |
1641 | int tmp_rval; /* partial logging flags */ | |
19ebedcf | 1642 | struct xfs_mount *mp; |
1277a5e0 | 1643 | xfs_extnum_t *nextents; |
49f693fa | 1644 | |
65ca3804 | 1645 | mp = bma->ip->i_mount; |
36e8786d | 1646 | ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
1277a5e0 DW |
1647 | ASSERT(whichfork != XFS_ATTR_FORK); |
1648 | nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents : | |
1649 | &bma->ip->i_d.di_nextents); | |
56b2de80 | 1650 | |
49f693fa | 1651 | ASSERT(bma->idx >= 0); |
d09d4e5f | 1652 | ASSERT(bma->idx <= xfs_iext_count(ifp)); |
49f693fa DC |
1653 | ASSERT(!isnullstartblock(new->br_startblock)); |
1654 | ASSERT(!bma->cur || | |
1655 | (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); | |
56b2de80 | 1656 | |
79896434 | 1657 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
1658 | |
1659 | #define LEFT r[0] | |
1660 | #define RIGHT r[1] | |
1661 | #define PREV r[2] | |
56b2de80 | 1662 | |
1277a5e0 DW |
1663 | if (whichfork == XFS_COW_FORK) |
1664 | state |= BMAP_COWFORK; | |
1665 | ||
56b2de80 | 1666 | /* |
49f693fa | 1667 | * Set up a bunch of variables to make the tests simpler. |
56b2de80 | 1668 | */ |
49f693fa DC |
1669 | ep = xfs_iext_get_ext(ifp, bma->idx); |
1670 | xfs_bmbt_get_all(ep, &PREV); | |
1671 | new_endoff = new->br_startoff + new->br_blockcount; | |
1672 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
1673 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
1674 | ||
1675 | da_old = startblockval(PREV.br_startblock); | |
1676 | da_new = 0; | |
1677 | ||
56b2de80 | 1678 | /* |
49f693fa DC |
1679 | * Set flags determining what part of the previous delayed allocation |
1680 | * extent is being replaced by a real allocation. | |
56b2de80 | 1681 | */ |
49f693fa DC |
1682 | if (PREV.br_startoff == new->br_startoff) |
1683 | state |= BMAP_LEFT_FILLING; | |
1684 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
1685 | state |= BMAP_RIGHT_FILLING; | |
1686 | ||
56b2de80 | 1687 | /* |
49f693fa DC |
1688 | * Check and set flags if this segment has a left neighbor. |
1689 | * Don't set contiguous if the combined extent would be too large. | |
56b2de80 | 1690 | */ |
49f693fa DC |
1691 | if (bma->idx > 0) { |
1692 | state |= BMAP_LEFT_VALID; | |
1693 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT); | |
1694 | ||
1695 | if (isnullstartblock(LEFT.br_startblock)) | |
1696 | state |= BMAP_LEFT_DELAY; | |
1697 | } | |
1698 | ||
1699 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
1700 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
1701 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
1702 | LEFT.br_state == new->br_state && | |
1703 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
1704 | state |= BMAP_LEFT_CONTIG; | |
56b2de80 DC |
1705 | |
1706 | /* | |
49f693fa DC |
1707 | * Check and set flags if this segment has a right neighbor. |
1708 | * Don't set contiguous if the combined extent would be too large. | |
1709 | * Also check for all-three-contiguous being too large. | |
56b2de80 | 1710 | */ |
d09d4e5f | 1711 | if (bma->idx < xfs_iext_count(ifp) - 1) { |
49f693fa DC |
1712 | state |= BMAP_RIGHT_VALID; |
1713 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT); | |
56b2de80 | 1714 | |
49f693fa DC |
1715 | if (isnullstartblock(RIGHT.br_startblock)) |
1716 | state |= BMAP_RIGHT_DELAY; | |
1717 | } | |
56b2de80 | 1718 | |
49f693fa DC |
1719 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
1720 | new_endoff == RIGHT.br_startoff && | |
1721 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
1722 | new->br_state == RIGHT.br_state && | |
1723 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
1724 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1725 | BMAP_RIGHT_FILLING)) != | |
1726 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
1727 | BMAP_RIGHT_FILLING) || | |
1728 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
1729 | <= MAXEXTLEN)) | |
1730 | state |= BMAP_RIGHT_CONTIG; | |
5e656dbb | 1731 | |
49f693fa DC |
1732 | error = 0; |
1733 | /* | |
1734 | * Switch out based on the FILLING and CONTIG state bits. | |
1735 | */ | |
1736 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1737 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
1738 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
1739 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1740 | /* | |
1741 | * Filling in all of a previously delayed allocation extent. | |
1742 | * The left and right neighbors are both contiguous with new. | |
1743 | */ | |
1744 | bma->idx--; | |
1745 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1746 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx), | |
1747 | LEFT.br_blockcount + PREV.br_blockcount + | |
1748 | RIGHT.br_blockcount); | |
1749 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
a2ceac1f | 1750 | |
49f693fa | 1751 | xfs_iext_remove(bma->ip, bma->idx + 1, 2, state); |
1277a5e0 | 1752 | (*nextents)--; |
49f693fa DC |
1753 | if (bma->cur == NULL) |
1754 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1755 | else { | |
1756 | rval = XFS_ILOG_CORE; | |
1757 | error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff, | |
1758 | RIGHT.br_startblock, | |
1759 | RIGHT.br_blockcount, &i); | |
1760 | if (error) | |
1761 | goto done; | |
19ebedcf | 1762 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1763 | error = xfs_btree_delete(bma->cur, &i); |
1764 | if (error) | |
1765 | goto done; | |
19ebedcf | 1766 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1767 | error = xfs_btree_decrement(bma->cur, 0, &i); |
1768 | if (error) | |
1769 | goto done; | |
19ebedcf | 1770 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1771 | error = xfs_bmbt_update(bma->cur, LEFT.br_startoff, |
1772 | LEFT.br_startblock, | |
1773 | LEFT.br_blockcount + | |
1774 | PREV.br_blockcount + | |
1775 | RIGHT.br_blockcount, LEFT.br_state); | |
1776 | if (error) | |
1777 | goto done; | |
5e656dbb | 1778 | } |
49f693fa DC |
1779 | break; |
1780 | ||
1781 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
1782 | /* | |
1783 | * Filling in all of a previously delayed allocation extent. | |
1784 | * The left neighbor is contiguous, the right is not. | |
1785 | */ | |
1786 | bma->idx--; | |
1787 | ||
1788 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1789 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx), | |
1790 | LEFT.br_blockcount + PREV.br_blockcount); | |
1791 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1792 | ||
1793 | xfs_iext_remove(bma->ip, bma->idx + 1, 1, state); | |
1794 | if (bma->cur == NULL) | |
1795 | rval = XFS_ILOG_DEXT; | |
1796 | else { | |
1797 | rval = 0; | |
1798 | error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff, | |
1799 | LEFT.br_startblock, LEFT.br_blockcount, | |
1800 | &i); | |
1801 | if (error) | |
1802 | goto done; | |
19ebedcf | 1803 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1804 | error = xfs_bmbt_update(bma->cur, LEFT.br_startoff, |
1805 | LEFT.br_startblock, | |
1806 | LEFT.br_blockcount + | |
1807 | PREV.br_blockcount, LEFT.br_state); | |
1808 | if (error) | |
1809 | goto done; | |
1810 | } | |
1811 | break; | |
1812 | ||
1813 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
1814 | /* | |
1815 | * Filling in all of a previously delayed allocation extent. | |
1816 | * The right neighbor is contiguous, the left is not. | |
1817 | */ | |
1818 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1819 | xfs_bmbt_set_startblock(ep, new->br_startblock); | |
1820 | xfs_bmbt_set_blockcount(ep, | |
1821 | PREV.br_blockcount + RIGHT.br_blockcount); | |
1822 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1823 | ||
1824 | xfs_iext_remove(bma->ip, bma->idx + 1, 1, state); | |
1825 | if (bma->cur == NULL) | |
1826 | rval = XFS_ILOG_DEXT; | |
1827 | else { | |
1828 | rval = 0; | |
1829 | error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff, | |
1830 | RIGHT.br_startblock, | |
1831 | RIGHT.br_blockcount, &i); | |
1832 | if (error) | |
1833 | goto done; | |
19ebedcf | 1834 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1835 | error = xfs_bmbt_update(bma->cur, PREV.br_startoff, |
1836 | new->br_startblock, | |
1837 | PREV.br_blockcount + | |
1838 | RIGHT.br_blockcount, PREV.br_state); | |
1839 | if (error) | |
1840 | goto done; | |
1841 | } | |
1842 | break; | |
1843 | ||
1844 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: | |
1845 | /* | |
1846 | * Filling in all of a previously delayed allocation extent. | |
1847 | * Neither the left nor right neighbors are contiguous with | |
1848 | * the new one. | |
1849 | */ | |
1850 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1851 | xfs_bmbt_set_startblock(ep, new->br_startblock); | |
1852 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
5e656dbb | 1853 | |
1277a5e0 | 1854 | (*nextents)++; |
49f693fa DC |
1855 | if (bma->cur == NULL) |
1856 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1857 | else { | |
1858 | rval = XFS_ILOG_CORE; | |
1859 | error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff, | |
1860 | new->br_startblock, new->br_blockcount, | |
1861 | &i); | |
1862 | if (error) | |
1863 | goto done; | |
19ebedcf | 1864 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1865 | bma->cur->bc_rec.b.br_state = XFS_EXT_NORM; |
1866 | error = xfs_btree_insert(bma->cur, &i); | |
1867 | if (error) | |
1868 | goto done; | |
19ebedcf | 1869 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1870 | } |
1871 | break; | |
5e656dbb | 1872 | |
49f693fa DC |
1873 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
1874 | /* | |
1875 | * Filling in the first part of a previous delayed allocation. | |
1876 | * The left neighbor is contiguous. | |
1877 | */ | |
1878 | trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_); | |
1879 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1), | |
1880 | LEFT.br_blockcount + new->br_blockcount); | |
1881 | xfs_bmbt_set_startoff(ep, | |
1882 | PREV.br_startoff + new->br_blockcount); | |
1883 | trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_); | |
a2ceac1f | 1884 | |
49f693fa DC |
1885 | temp = PREV.br_blockcount - new->br_blockcount; |
1886 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1887 | xfs_bmbt_set_blockcount(ep, temp); | |
1888 | if (bma->cur == NULL) | |
1889 | rval = XFS_ILOG_DEXT; | |
1890 | else { | |
1891 | rval = 0; | |
1892 | error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff, | |
1893 | LEFT.br_startblock, LEFT.br_blockcount, | |
1894 | &i); | |
1895 | if (error) | |
1896 | goto done; | |
19ebedcf | 1897 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1898 | error = xfs_bmbt_update(bma->cur, LEFT.br_startoff, |
1899 | LEFT.br_startblock, | |
1900 | LEFT.br_blockcount + | |
1901 | new->br_blockcount, | |
1902 | LEFT.br_state); | |
1903 | if (error) | |
1904 | goto done; | |
2bd0ea18 | 1905 | } |
49f693fa DC |
1906 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), |
1907 | startblockval(PREV.br_startblock)); | |
1908 | xfs_bmbt_set_startblock(ep, nullstartblock(da_new)); | |
1909 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1910 | ||
1911 | bma->idx--; | |
1912 | break; | |
1913 | ||
1914 | case BMAP_LEFT_FILLING: | |
2bd0ea18 | 1915 | /* |
49f693fa DC |
1916 | * Filling in the first part of a previous delayed allocation. |
1917 | * The left neighbor is not contiguous. | |
5000d01d | 1918 | */ |
49f693fa DC |
1919 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); |
1920 | xfs_bmbt_set_startoff(ep, new_endoff); | |
1921 | temp = PREV.br_blockcount - new->br_blockcount; | |
1922 | xfs_bmbt_set_blockcount(ep, temp); | |
1923 | xfs_iext_insert(bma->ip, bma->idx, 1, new, state); | |
1277a5e0 | 1924 | (*nextents)++; |
49f693fa DC |
1925 | if (bma->cur == NULL) |
1926 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
1927 | else { | |
1928 | rval = XFS_ILOG_CORE; | |
1929 | error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff, | |
1930 | new->br_startblock, new->br_blockcount, | |
1931 | &i); | |
1932 | if (error) | |
1933 | goto done; | |
19ebedcf | 1934 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
1935 | bma->cur->bc_rec.b.br_state = XFS_EXT_NORM; |
1936 | error = xfs_btree_insert(bma->cur, &i); | |
1937 | if (error) | |
1938 | goto done; | |
19ebedcf | 1939 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1940 | } |
1941 | ||
36e8786d | 1942 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 1943 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 1944 | bma->firstblock, bma->dfops, |
36e8786d | 1945 | &bma->cur, 1, &tmp_rval, whichfork); |
49f693fa DC |
1946 | rval |= tmp_rval; |
1947 | if (error) | |
1948 | goto done; | |
1949 | } | |
1950 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
1951 | startblockval(PREV.br_startblock) - | |
1952 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
1953 | ep = xfs_iext_get_ext(ifp, bma->idx + 1); | |
1954 | xfs_bmbt_set_startblock(ep, nullstartblock(da_new)); | |
1955 | trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_); | |
1956 | break; | |
1957 | ||
1958 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
5e656dbb | 1959 | /* |
49f693fa DC |
1960 | * Filling in the last part of a previous delayed allocation. |
1961 | * The right neighbor is contiguous with the new allocation. | |
5e656dbb | 1962 | */ |
49f693fa DC |
1963 | temp = PREV.br_blockcount - new->br_blockcount; |
1964 | trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_); | |
1965 | xfs_bmbt_set_blockcount(ep, temp); | |
1966 | xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1), | |
1967 | new->br_startoff, new->br_startblock, | |
1968 | new->br_blockcount + RIGHT.br_blockcount, | |
1969 | RIGHT.br_state); | |
1970 | trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_); | |
1971 | if (bma->cur == NULL) | |
1972 | rval = XFS_ILOG_DEXT; | |
1973 | else { | |
1974 | rval = 0; | |
1975 | error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff, | |
1976 | RIGHT.br_startblock, | |
1977 | RIGHT.br_blockcount, &i); | |
1978 | if (error) | |
1979 | goto done; | |
19ebedcf | 1980 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
1981 | error = xfs_bmbt_update(bma->cur, new->br_startoff, |
1982 | new->br_startblock, | |
1983 | new->br_blockcount + | |
1984 | RIGHT.br_blockcount, | |
1985 | RIGHT.br_state); | |
1986 | if (error) | |
1987 | goto done; | |
1988 | } | |
1989 | ||
1990 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
1991 | startblockval(PREV.br_startblock)); | |
1992 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1993 | xfs_bmbt_set_startblock(ep, nullstartblock(da_new)); | |
1994 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
1995 | ||
1996 | bma->idx++; | |
1997 | break; | |
1998 | ||
1999 | case BMAP_RIGHT_FILLING: | |
a2ceac1f | 2000 | /* |
49f693fa DC |
2001 | * Filling in the last part of a previous delayed allocation. |
2002 | * The right neighbor is not contiguous. | |
a2ceac1f | 2003 | */ |
49f693fa DC |
2004 | temp = PREV.br_blockcount - new->br_blockcount; |
2005 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
2006 | xfs_bmbt_set_blockcount(ep, temp); | |
2007 | xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state); | |
1277a5e0 | 2008 | (*nextents)++; |
49f693fa DC |
2009 | if (bma->cur == NULL) |
2010 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2011 | else { | |
2012 | rval = XFS_ILOG_CORE; | |
2013 | error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff, | |
2014 | new->br_startblock, new->br_blockcount, | |
2015 | &i); | |
2016 | if (error) | |
2017 | goto done; | |
19ebedcf | 2018 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
2019 | bma->cur->bc_rec.b.br_state = XFS_EXT_NORM; |
2020 | error = xfs_btree_insert(bma->cur, &i); | |
2021 | if (error) | |
2022 | goto done; | |
19ebedcf | 2023 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 2024 | } |
a2ceac1f | 2025 | |
36e8786d | 2026 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 2027 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 2028 | bma->firstblock, bma->dfops, &bma->cur, 1, |
36e8786d | 2029 | &tmp_rval, whichfork); |
49f693fa DC |
2030 | rval |= tmp_rval; |
2031 | if (error) | |
2032 | goto done; | |
2033 | } | |
2034 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp), | |
2035 | startblockval(PREV.br_startblock) - | |
2036 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
2037 | ep = xfs_iext_get_ext(ifp, bma->idx); | |
2038 | xfs_bmbt_set_startblock(ep, nullstartblock(da_new)); | |
2039 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
2040 | ||
2041 | bma->idx++; | |
2042 | break; | |
2043 | ||
2044 | case 0: | |
5e656dbb | 2045 | /* |
49f693fa DC |
2046 | * Filling in the middle part of a previous delayed allocation. |
2047 | * Contiguity is impossible here. | |
2048 | * This case is avoided almost all the time. | |
2049 | * | |
2050 | * We start with a delayed allocation: | |
2051 | * | |
2052 | * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+ | |
2053 | * PREV @ idx | |
2054 | * | |
2055 | * and we are allocating: | |
2056 | * +rrrrrrrrrrrrrrrrr+ | |
2057 | * new | |
2058 | * | |
2059 | * and we set it up for insertion as: | |
2060 | * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+ | |
2061 | * new | |
2062 | * PREV @ idx LEFT RIGHT | |
2063 | * inserted at idx + 1 | |
5e656dbb | 2064 | */ |
49f693fa DC |
2065 | temp = new->br_startoff - PREV.br_startoff; |
2066 | temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
2067 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_); | |
2068 | xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */ | |
2069 | LEFT = *new; | |
2070 | RIGHT.br_state = PREV.br_state; | |
2071 | RIGHT.br_startblock = nullstartblock( | |
2072 | (int)xfs_bmap_worst_indlen(bma->ip, temp2)); | |
2073 | RIGHT.br_startoff = new_endoff; | |
2074 | RIGHT.br_blockcount = temp2; | |
2075 | /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */ | |
2076 | xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state); | |
1277a5e0 | 2077 | (*nextents)++; |
49f693fa DC |
2078 | if (bma->cur == NULL) |
2079 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2080 | else { | |
2081 | rval = XFS_ILOG_CORE; | |
2082 | error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff, | |
2083 | new->br_startblock, new->br_blockcount, | |
2084 | &i); | |
2085 | if (error) | |
2086 | goto done; | |
19ebedcf | 2087 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
2088 | bma->cur->bc_rec.b.br_state = XFS_EXT_NORM; |
2089 | error = xfs_btree_insert(bma->cur, &i); | |
2090 | if (error) | |
2091 | goto done; | |
19ebedcf | 2092 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 2093 | } |
5e656dbb | 2094 | |
36e8786d | 2095 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa | 2096 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, |
f33cea1a | 2097 | bma->firstblock, bma->dfops, &bma->cur, |
36e8786d | 2098 | 1, &tmp_rval, whichfork); |
49f693fa DC |
2099 | rval |= tmp_rval; |
2100 | if (error) | |
2101 | goto done; | |
2102 | } | |
2103 | temp = xfs_bmap_worst_indlen(bma->ip, temp); | |
2104 | temp2 = xfs_bmap_worst_indlen(bma->ip, temp2); | |
2105 | diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) - | |
2106 | (bma->cur ? bma->cur->bc_private.b.allocated : 0)); | |
2107 | if (diff > 0) { | |
19ebedcf DC |
2108 | error = xfs_mod_fdblocks(bma->ip->i_mount, |
2109 | -((int64_t)diff), false); | |
49f693fa DC |
2110 | ASSERT(!error); |
2111 | if (error) | |
2112 | goto done; | |
2113 | } | |
2bd0ea18 | 2114 | |
49f693fa DC |
2115 | ep = xfs_iext_get_ext(ifp, bma->idx); |
2116 | xfs_bmbt_set_startblock(ep, nullstartblock((int)temp)); | |
2117 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
2118 | trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_); | |
2119 | xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2), | |
2120 | nullstartblock((int)temp2)); | |
2121 | trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_); | |
2bd0ea18 | 2122 | |
49f693fa DC |
2123 | bma->idx++; |
2124 | da_new = temp + temp2; | |
2125 | break; | |
2126 | ||
2127 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2128 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2129 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2130 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2131 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2132 | case BMAP_LEFT_CONTIG: | |
2133 | case BMAP_RIGHT_CONTIG: | |
2134 | /* | |
2135 | * These cases are all impossible. | |
2136 | */ | |
2137 | ASSERT(0); | |
2138 | } | |
2139 | ||
d7f80320 DW |
2140 | /* add reverse mapping */ |
2141 | error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new); | |
2142 | if (error) | |
2143 | goto done; | |
2144 | ||
49f693fa | 2145 | /* convert to a btree if necessary */ |
36e8786d | 2146 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { |
49f693fa DC |
2147 | int tmp_logflags; /* partial log flag return val */ |
2148 | ||
2149 | ASSERT(bma->cur == NULL); | |
2150 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, | |
f33cea1a | 2151 | bma->firstblock, bma->dfops, &bma->cur, |
36e8786d | 2152 | da_old > 0, &tmp_logflags, whichfork); |
49f693fa DC |
2153 | bma->logflags |= tmp_logflags; |
2154 | if (error) | |
2155 | goto done; | |
2156 | } | |
2157 | ||
2158 | /* adjust for changes in reserved delayed indirect blocks */ | |
2159 | if (da_old || da_new) { | |
2160 | temp = da_new; | |
2161 | if (bma->cur) | |
2162 | temp += bma->cur->bc_private.b.allocated; | |
2163 | ASSERT(temp <= da_old); | |
2164 | if (temp < da_old) | |
19ebedcf DC |
2165 | xfs_mod_fdblocks(bma->ip->i_mount, |
2166 | (int64_t)(da_old - temp), false); | |
49f693fa DC |
2167 | } |
2168 | ||
2169 | /* clear out the allocated field, done with it now in any case. */ | |
2170 | if (bma->cur) | |
2171 | bma->cur->bc_private.b.allocated = 0; | |
2172 | ||
36e8786d | 2173 | xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork); |
49f693fa | 2174 | done: |
1277a5e0 DW |
2175 | if (whichfork != XFS_COW_FORK) |
2176 | bma->logflags |= rval; | |
49f693fa DC |
2177 | return error; |
2178 | #undef LEFT | |
2179 | #undef RIGHT | |
2180 | #undef PREV | |
2bd0ea18 NS |
2181 | } |
2182 | ||
2183 | /* | |
49f693fa | 2184 | * Convert an unwritten allocation to a real allocation or vice versa. |
2bd0ea18 NS |
2185 | */ |
2186 | STATIC int /* error */ | |
49f693fa DC |
2187 | xfs_bmap_add_extent_unwritten_real( |
2188 | struct xfs_trans *tp, | |
2bd0ea18 | 2189 | xfs_inode_t *ip, /* incore inode pointer */ |
49f693fa DC |
2190 | xfs_extnum_t *idx, /* extent number to update/insert */ |
2191 | xfs_btree_cur_t **curp, /* if *curp is null, not a btree */ | |
2192 | xfs_bmbt_irec_t *new, /* new data to add to file extents */ | |
2193 | xfs_fsblock_t *first, /* pointer to firstblock variable */ | |
f33cea1a | 2194 | struct xfs_defer_ops *dfops, /* list of extents to be freed */ |
49f693fa | 2195 | int *logflagsp) /* inode logging flags */ |
2bd0ea18 | 2196 | { |
49f693fa DC |
2197 | xfs_btree_cur_t *cur; /* btree cursor */ |
2198 | xfs_bmbt_rec_host_t *ep; /* extent entry for idx */ | |
2bd0ea18 | 2199 | int error; /* error return value */ |
2bd0ea18 NS |
2200 | int i; /* temp state */ |
2201 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
49f693fa DC |
2202 | xfs_fileoff_t new_endoff; /* end offset of new entry */ |
2203 | xfs_exntst_t newext; /* new extent state */ | |
2204 | xfs_exntst_t oldext; /* old extent state */ | |
2205 | xfs_bmbt_irec_t r[3]; /* neighbor extent entries */ | |
2206 | /* left is 0, right is 1, prev is 2 */ | |
2207 | int rval=0; /* return value (logging flags) */ | |
2208 | int state = 0;/* state bits, accessed thru macros */ | |
19ebedcf | 2209 | struct xfs_mount *mp = tp->t_mountp; |
5000d01d | 2210 | |
49f693fa | 2211 | *logflagsp = 0; |
56b2de80 | 2212 | |
49f693fa DC |
2213 | cur = *curp; |
2214 | ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
56b2de80 | 2215 | |
49f693fa | 2216 | ASSERT(*idx >= 0); |
d09d4e5f | 2217 | ASSERT(*idx <= xfs_iext_count(ifp)); |
49f693fa DC |
2218 | ASSERT(!isnullstartblock(new->br_startblock)); |
2219 | ||
79896434 | 2220 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
2221 | |
2222 | #define LEFT r[0] | |
2223 | #define RIGHT r[1] | |
2224 | #define PREV r[2] | |
2225 | ||
2226 | /* | |
2227 | * Set up a bunch of variables to make the tests simpler. | |
2228 | */ | |
2bd0ea18 | 2229 | error = 0; |
49f693fa DC |
2230 | ep = xfs_iext_get_ext(ifp, *idx); |
2231 | xfs_bmbt_get_all(ep, &PREV); | |
2232 | newext = new->br_state; | |
2233 | oldext = (newext == XFS_EXT_UNWRITTEN) ? | |
2234 | XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
2235 | ASSERT(PREV.br_state == oldext); | |
2236 | new_endoff = new->br_startoff + new->br_blockcount; | |
2237 | ASSERT(PREV.br_startoff <= new->br_startoff); | |
2238 | ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff); | |
2239 | ||
2bd0ea18 | 2240 | /* |
49f693fa DC |
2241 | * Set flags determining what part of the previous oldext allocation |
2242 | * extent is being replaced by a newext allocation. | |
2bd0ea18 | 2243 | */ |
49f693fa DC |
2244 | if (PREV.br_startoff == new->br_startoff) |
2245 | state |= BMAP_LEFT_FILLING; | |
2246 | if (PREV.br_startoff + PREV.br_blockcount == new_endoff) | |
2247 | state |= BMAP_RIGHT_FILLING; | |
2bd0ea18 | 2248 | |
49f693fa DC |
2249 | /* |
2250 | * Check and set flags if this segment has a left neighbor. | |
2251 | * Don't set contiguous if the combined extent would be too large. | |
2252 | */ | |
2253 | if (*idx > 0) { | |
2254 | state |= BMAP_LEFT_VALID; | |
2255 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT); | |
2256 | ||
2257 | if (isnullstartblock(LEFT.br_startblock)) | |
2258 | state |= BMAP_LEFT_DELAY; | |
2bd0ea18 | 2259 | } |
49f693fa DC |
2260 | |
2261 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2262 | LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff && | |
2263 | LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock && | |
2264 | LEFT.br_state == newext && | |
2265 | LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2266 | state |= BMAP_LEFT_CONTIG; | |
2267 | ||
2bd0ea18 | 2268 | /* |
49f693fa DC |
2269 | * Check and set flags if this segment has a right neighbor. |
2270 | * Don't set contiguous if the combined extent would be too large. | |
2271 | * Also check for all-three-contiguous being too large. | |
2bd0ea18 | 2272 | */ |
d09d4e5f | 2273 | if (*idx < xfs_iext_count(&ip->i_df) - 1) { |
49f693fa DC |
2274 | state |= BMAP_RIGHT_VALID; |
2275 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT); | |
2276 | if (isnullstartblock(RIGHT.br_startblock)) | |
2277 | state |= BMAP_RIGHT_DELAY; | |
2278 | } | |
a2ceac1f | 2279 | |
49f693fa DC |
2280 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2281 | new_endoff == RIGHT.br_startoff && | |
2282 | new->br_startblock + new->br_blockcount == RIGHT.br_startblock && | |
2283 | newext == RIGHT.br_state && | |
2284 | new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN && | |
2285 | ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2286 | BMAP_RIGHT_FILLING)) != | |
2287 | (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING | | |
2288 | BMAP_RIGHT_FILLING) || | |
2289 | LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount | |
2290 | <= MAXEXTLEN)) | |
2291 | state |= BMAP_RIGHT_CONTIG; | |
2bd0ea18 | 2292 | |
49f693fa DC |
2293 | /* |
2294 | * Switch out based on the FILLING and CONTIG state bits. | |
2295 | */ | |
2296 | switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2297 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) { | |
2298 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | | |
2299 | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: | |
2bd0ea18 | 2300 | /* |
49f693fa DC |
2301 | * Setting all of a previous oldext extent to newext. |
2302 | * The left and right neighbors are both contiguous with new. | |
2bd0ea18 | 2303 | */ |
49f693fa DC |
2304 | --*idx; |
2305 | ||
a2ceac1f | 2306 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
49f693fa DC |
2307 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), |
2308 | LEFT.br_blockcount + PREV.br_blockcount + | |
2309 | RIGHT.br_blockcount); | |
a2ceac1f | 2310 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); |
49f693fa DC |
2311 | |
2312 | xfs_iext_remove(ip, *idx + 1, 2, state); | |
2313 | ip->i_d.di_nextents -= 2; | |
2314 | if (cur == NULL) | |
2315 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2316 | else { | |
2317 | rval = XFS_ILOG_CORE; | |
2318 | if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff, | |
2319 | RIGHT.br_startblock, | |
2320 | RIGHT.br_blockcount, &i))) | |
2321 | goto done; | |
19ebedcf | 2322 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2323 | if ((error = xfs_btree_delete(cur, &i))) |
2324 | goto done; | |
19ebedcf | 2325 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2326 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2327 | goto done; | |
19ebedcf | 2328 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2329 | if ((error = xfs_btree_delete(cur, &i))) |
2330 | goto done; | |
19ebedcf | 2331 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2332 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2333 | goto done; | |
19ebedcf | 2334 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2335 | if ((error = xfs_bmbt_update(cur, LEFT.br_startoff, |
2336 | LEFT.br_startblock, | |
2337 | LEFT.br_blockcount + PREV.br_blockcount + | |
2338 | RIGHT.br_blockcount, LEFT.br_state))) | |
2339 | goto done; | |
2bd0ea18 | 2340 | } |
2bd0ea18 NS |
2341 | break; |
2342 | ||
49f693fa | 2343 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: |
2bd0ea18 | 2344 | /* |
49f693fa DC |
2345 | * Setting all of a previous oldext extent to newext. |
2346 | * The left neighbor is contiguous, the right is not. | |
2bd0ea18 | 2347 | */ |
49f693fa DC |
2348 | --*idx; |
2349 | ||
a2ceac1f | 2350 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
49f693fa DC |
2351 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), |
2352 | LEFT.br_blockcount + PREV.br_blockcount); | |
a2ceac1f | 2353 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); |
49f693fa DC |
2354 | |
2355 | xfs_iext_remove(ip, *idx + 1, 1, state); | |
2356 | ip->i_d.di_nextents--; | |
2357 | if (cur == NULL) | |
2358 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2359 | else { | |
2360 | rval = XFS_ILOG_CORE; | |
2361 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2362 | PREV.br_startblock, PREV.br_blockcount, | |
2363 | &i))) | |
2364 | goto done; | |
19ebedcf | 2365 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2366 | if ((error = xfs_btree_delete(cur, &i))) |
2367 | goto done; | |
19ebedcf | 2368 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2369 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2370 | goto done; | |
19ebedcf | 2371 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2372 | if ((error = xfs_bmbt_update(cur, LEFT.br_startoff, |
2373 | LEFT.br_startblock, | |
2374 | LEFT.br_blockcount + PREV.br_blockcount, | |
2375 | LEFT.br_state))) | |
2376 | goto done; | |
2bd0ea18 | 2377 | } |
2bd0ea18 | 2378 | break; |
5000d01d | 2379 | |
49f693fa | 2380 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2bd0ea18 | 2381 | /* |
49f693fa DC |
2382 | * Setting all of a previous oldext extent to newext. |
2383 | * The right neighbor is contiguous, the left is not. | |
2bd0ea18 | 2384 | */ |
a2ceac1f | 2385 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
49f693fa DC |
2386 | xfs_bmbt_set_blockcount(ep, |
2387 | PREV.br_blockcount + RIGHT.br_blockcount); | |
2388 | xfs_bmbt_set_state(ep, newext); | |
a2ceac1f | 2389 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); |
49f693fa DC |
2390 | xfs_iext_remove(ip, *idx + 1, 1, state); |
2391 | ip->i_d.di_nextents--; | |
2392 | if (cur == NULL) | |
2393 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2394 | else { | |
2395 | rval = XFS_ILOG_CORE; | |
2396 | if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff, | |
2397 | RIGHT.br_startblock, | |
2398 | RIGHT.br_blockcount, &i))) | |
2399 | goto done; | |
19ebedcf | 2400 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2401 | if ((error = xfs_btree_delete(cur, &i))) |
2402 | goto done; | |
19ebedcf | 2403 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2404 | if ((error = xfs_btree_decrement(cur, 0, &i))) |
2405 | goto done; | |
19ebedcf | 2406 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2407 | if ((error = xfs_bmbt_update(cur, new->br_startoff, |
2408 | new->br_startblock, | |
2409 | new->br_blockcount + RIGHT.br_blockcount, | |
2410 | newext))) | |
2411 | goto done; | |
2412 | } | |
2413 | break; | |
2bd0ea18 | 2414 | |
49f693fa DC |
2415 | case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING: |
2416 | /* | |
2417 | * Setting all of a previous oldext extent to newext. | |
2418 | * Neither the left nor right neighbors are contiguous with | |
2419 | * the new one. | |
2420 | */ | |
2421 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2422 | xfs_bmbt_set_state(ep, newext); | |
2423 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2bd0ea18 | 2424 | |
49f693fa DC |
2425 | if (cur == NULL) |
2426 | rval = XFS_ILOG_DEXT; | |
2427 | else { | |
2428 | rval = 0; | |
2429 | if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff, | |
2430 | new->br_startblock, new->br_blockcount, | |
2431 | &i))) | |
2432 | goto done; | |
19ebedcf | 2433 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2434 | if ((error = xfs_bmbt_update(cur, new->br_startoff, |
2435 | new->br_startblock, new->br_blockcount, | |
2436 | newext))) | |
2437 | goto done; | |
2438 | } | |
2439 | break; | |
2bd0ea18 | 2440 | |
49f693fa DC |
2441 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG: |
2442 | /* | |
2443 | * Setting the first part of a previous oldext extent to newext. | |
2444 | * The left neighbor is contiguous. | |
2445 | */ | |
2446 | trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_); | |
2447 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1), | |
2448 | LEFT.br_blockcount + new->br_blockcount); | |
2449 | xfs_bmbt_set_startoff(ep, | |
2450 | PREV.br_startoff + new->br_blockcount); | |
2451 | trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_); | |
a2ceac1f | 2452 | |
49f693fa DC |
2453 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2454 | xfs_bmbt_set_startblock(ep, | |
2455 | new->br_startblock + new->br_blockcount); | |
2456 | xfs_bmbt_set_blockcount(ep, | |
2457 | PREV.br_blockcount - new->br_blockcount); | |
2458 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
b3563c19 | 2459 | |
49f693fa | 2460 | --*idx; |
b3563c19 | 2461 | |
49f693fa DC |
2462 | if (cur == NULL) |
2463 | rval = XFS_ILOG_DEXT; | |
2464 | else { | |
2465 | rval = 0; | |
2466 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2467 | PREV.br_startblock, PREV.br_blockcount, | |
2468 | &i))) | |
2469 | goto done; | |
19ebedcf | 2470 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2471 | if ((error = xfs_bmbt_update(cur, |
2472 | PREV.br_startoff + new->br_blockcount, | |
2473 | PREV.br_startblock + new->br_blockcount, | |
2474 | PREV.br_blockcount - new->br_blockcount, | |
2475 | oldext))) | |
2476 | goto done; | |
2477 | if ((error = xfs_btree_decrement(cur, 0, &i))) | |
2478 | goto done; | |
2479 | error = xfs_bmbt_update(cur, LEFT.br_startoff, | |
2480 | LEFT.br_startblock, | |
2481 | LEFT.br_blockcount + new->br_blockcount, | |
2482 | LEFT.br_state); | |
2483 | if (error) | |
2484 | goto done; | |
2bd0ea18 | 2485 | } |
49f693fa | 2486 | break; |
b3563c19 | 2487 | |
49f693fa DC |
2488 | case BMAP_LEFT_FILLING: |
2489 | /* | |
2490 | * Setting the first part of a previous oldext extent to newext. | |
2491 | * The left neighbor is not contiguous. | |
2492 | */ | |
2493 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2494 | ASSERT(ep && xfs_bmbt_get_state(ep) == oldext); | |
2495 | xfs_bmbt_set_startoff(ep, new_endoff); | |
2496 | xfs_bmbt_set_blockcount(ep, | |
2497 | PREV.br_blockcount - new->br_blockcount); | |
2498 | xfs_bmbt_set_startblock(ep, | |
2499 | new->br_startblock + new->br_blockcount); | |
2500 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2bd0ea18 | 2501 | |
49f693fa DC |
2502 | xfs_iext_insert(ip, *idx, 1, new, state); |
2503 | ip->i_d.di_nextents++; | |
2504 | if (cur == NULL) | |
2505 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2506 | else { | |
2507 | rval = XFS_ILOG_CORE; | |
2508 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2509 | PREV.br_startblock, PREV.br_blockcount, | |
2510 | &i))) | |
2511 | goto done; | |
19ebedcf | 2512 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2513 | if ((error = xfs_bmbt_update(cur, |
2514 | PREV.br_startoff + new->br_blockcount, | |
2515 | PREV.br_startblock + new->br_blockcount, | |
2516 | PREV.br_blockcount - new->br_blockcount, | |
2517 | oldext))) | |
2518 | goto done; | |
2519 | cur->bc_rec.b = *new; | |
2520 | if ((error = xfs_btree_insert(cur, &i))) | |
2521 | goto done; | |
19ebedcf | 2522 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2523 | } |
2524 | break; | |
56b2de80 | 2525 | |
49f693fa DC |
2526 | case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG: |
2527 | /* | |
2528 | * Setting the last part of a previous oldext extent to newext. | |
2529 | * The right neighbor is contiguous with the new allocation. | |
2530 | */ | |
2531 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2532 | xfs_bmbt_set_blockcount(ep, | |
2533 | PREV.br_blockcount - new->br_blockcount); | |
2534 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
56b2de80 | 2535 | |
49f693fa | 2536 | ++*idx; |
56b2de80 | 2537 | |
49f693fa DC |
2538 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2539 | xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx), | |
2540 | new->br_startoff, new->br_startblock, | |
2541 | new->br_blockcount + RIGHT.br_blockcount, newext); | |
2542 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
56b2de80 | 2543 | |
49f693fa DC |
2544 | if (cur == NULL) |
2545 | rval = XFS_ILOG_DEXT; | |
2546 | else { | |
2547 | rval = 0; | |
2548 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2549 | PREV.br_startblock, | |
2550 | PREV.br_blockcount, &i))) | |
2551 | goto done; | |
19ebedcf | 2552 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2553 | if ((error = xfs_bmbt_update(cur, PREV.br_startoff, |
2554 | PREV.br_startblock, | |
2555 | PREV.br_blockcount - new->br_blockcount, | |
2556 | oldext))) | |
2557 | goto done; | |
2558 | if ((error = xfs_btree_increment(cur, 0, &i))) | |
2559 | goto done; | |
2560 | if ((error = xfs_bmbt_update(cur, new->br_startoff, | |
2561 | new->br_startblock, | |
2562 | new->br_blockcount + RIGHT.br_blockcount, | |
2563 | newext))) | |
2564 | goto done; | |
2565 | } | |
2566 | break; | |
ca86e759 | 2567 | |
49f693fa DC |
2568 | case BMAP_RIGHT_FILLING: |
2569 | /* | |
2570 | * Setting the last part of a previous oldext extent to newext. | |
2571 | * The right neighbor is not contiguous. | |
2572 | */ | |
2573 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2574 | xfs_bmbt_set_blockcount(ep, | |
2575 | PREV.br_blockcount - new->br_blockcount); | |
2576 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2bd0ea18 | 2577 | |
49f693fa DC |
2578 | ++*idx; |
2579 | xfs_iext_insert(ip, *idx, 1, new, state); | |
2bd0ea18 | 2580 | |
49f693fa DC |
2581 | ip->i_d.di_nextents++; |
2582 | if (cur == NULL) | |
2583 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2584 | else { | |
2585 | rval = XFS_ILOG_CORE; | |
2586 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2587 | PREV.br_startblock, PREV.br_blockcount, | |
2588 | &i))) | |
2589 | goto done; | |
19ebedcf | 2590 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2591 | if ((error = xfs_bmbt_update(cur, PREV.br_startoff, |
2592 | PREV.br_startblock, | |
2593 | PREV.br_blockcount - new->br_blockcount, | |
2594 | oldext))) | |
2595 | goto done; | |
2596 | if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff, | |
2597 | new->br_startblock, new->br_blockcount, | |
2598 | &i))) | |
2599 | goto done; | |
19ebedcf | 2600 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
2601 | cur->bc_rec.b.br_state = XFS_EXT_NORM; |
2602 | if ((error = xfs_btree_insert(cur, &i))) | |
2603 | goto done; | |
19ebedcf | 2604 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2605 | } |
2606 | break; | |
2607 | ||
2608 | case 0: | |
2bd0ea18 | 2609 | /* |
49f693fa DC |
2610 | * Setting the middle part of a previous oldext extent to |
2611 | * newext. Contiguity is impossible here. | |
2612 | * One extent becomes three extents. | |
2bd0ea18 | 2613 | */ |
49f693fa DC |
2614 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
2615 | xfs_bmbt_set_blockcount(ep, | |
2616 | new->br_startoff - PREV.br_startoff); | |
2617 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2618 | ||
2619 | r[0] = *new; | |
2620 | r[1].br_startoff = new_endoff; | |
2621 | r[1].br_blockcount = | |
2622 | PREV.br_startoff + PREV.br_blockcount - new_endoff; | |
2623 | r[1].br_startblock = new->br_startblock + new->br_blockcount; | |
2624 | r[1].br_state = oldext; | |
2625 | ||
2626 | ++*idx; | |
2627 | xfs_iext_insert(ip, *idx, 2, &r[0], state); | |
2628 | ||
2629 | ip->i_d.di_nextents += 2; | |
2630 | if (cur == NULL) | |
2631 | rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; | |
2632 | else { | |
2633 | rval = XFS_ILOG_CORE; | |
2634 | if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff, | |
2635 | PREV.br_startblock, PREV.br_blockcount, | |
2636 | &i))) | |
2637 | goto done; | |
19ebedcf | 2638 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2639 | /* new right extent - oldext */ |
2640 | if ((error = xfs_bmbt_update(cur, r[1].br_startoff, | |
2641 | r[1].br_startblock, r[1].br_blockcount, | |
2642 | r[1].br_state))) | |
2643 | goto done; | |
2644 | /* new left extent - oldext */ | |
2645 | cur->bc_rec.b = PREV; | |
2646 | cur->bc_rec.b.br_blockcount = | |
2647 | new->br_startoff - PREV.br_startoff; | |
2648 | if ((error = xfs_btree_insert(cur, &i))) | |
2649 | goto done; | |
19ebedcf | 2650 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2651 | /* |
2652 | * Reset the cursor to the position of the new extent | |
2653 | * we are about to insert as we can't trust it after | |
2654 | * the previous insert. | |
2655 | */ | |
2656 | if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff, | |
2657 | new->br_startblock, new->br_blockcount, | |
2658 | &i))) | |
2659 | goto done; | |
19ebedcf | 2660 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
2661 | /* new middle extent - newext */ |
2662 | cur->bc_rec.b.br_state = new->br_state; | |
2663 | if ((error = xfs_btree_insert(cur, &i))) | |
2664 | goto done; | |
19ebedcf | 2665 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
2bd0ea18 | 2666 | } |
49f693fa DC |
2667 | break; |
2668 | ||
2669 | case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2670 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2671 | case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG: | |
2672 | case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG: | |
2673 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2674 | case BMAP_LEFT_CONTIG: | |
2675 | case BMAP_RIGHT_CONTIG: | |
5000d01d | 2676 | /* |
49f693fa | 2677 | * These cases are all impossible. |
2bd0ea18 | 2678 | */ |
49f693fa DC |
2679 | ASSERT(0); |
2680 | } | |
2681 | ||
d7f80320 DW |
2682 | /* update reverse mappings */ |
2683 | error = xfs_rmap_convert_extent(mp, dfops, ip, XFS_DATA_FORK, new); | |
2684 | if (error) | |
2685 | goto done; | |
2686 | ||
49f693fa DC |
2687 | /* convert to a btree if necessary */ |
2688 | if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) { | |
2689 | int tmp_logflags; /* partial log flag return val */ | |
2690 | ||
2691 | ASSERT(cur == NULL); | |
f33cea1a | 2692 | error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur, |
49f693fa DC |
2693 | 0, &tmp_logflags, XFS_DATA_FORK); |
2694 | *logflagsp |= tmp_logflags; | |
2695 | if (error) | |
2696 | goto done; | |
ca86e759 | 2697 | } |
49f693fa DC |
2698 | |
2699 | /* clear out the allocated field, done with it now in any case. */ | |
2700 | if (cur) { | |
2701 | cur->bc_private.b.allocated = 0; | |
2702 | *curp = cur; | |
2703 | } | |
2704 | ||
2705 | xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK); | |
2bd0ea18 | 2706 | done: |
49f693fa | 2707 | *logflagsp |= rval; |
2bd0ea18 | 2708 | return error; |
49f693fa DC |
2709 | #undef LEFT |
2710 | #undef RIGHT | |
2711 | #undef PREV | |
2bd0ea18 NS |
2712 | } |
2713 | ||
5e656dbb | 2714 | /* |
49f693fa | 2715 | * Convert a hole to a delayed allocation. |
5e656dbb | 2716 | */ |
49f693fa DC |
2717 | STATIC void |
2718 | xfs_bmap_add_extent_hole_delay( | |
2719 | xfs_inode_t *ip, /* incore inode pointer */ | |
cc66acaf | 2720 | int whichfork, |
49f693fa DC |
2721 | xfs_extnum_t *idx, /* extent number to update/insert */ |
2722 | xfs_bmbt_irec_t *new) /* new data to add to file extents */ | |
2bd0ea18 | 2723 | { |
49f693fa DC |
2724 | xfs_ifork_t *ifp; /* inode fork pointer */ |
2725 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ | |
2726 | xfs_filblks_t newlen=0; /* new indirect size */ | |
2727 | xfs_filblks_t oldlen=0; /* old indirect size */ | |
2728 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2729 | int state; /* state bits, accessed thru macros */ | |
2730 | xfs_filblks_t temp=0; /* temp for indirect calculations */ | |
2731 | ||
cc66acaf | 2732 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa | 2733 | state = 0; |
cc66acaf DW |
2734 | if (whichfork == XFS_COW_FORK) |
2735 | state |= BMAP_COWFORK; | |
49f693fa | 2736 | ASSERT(isnullstartblock(new->br_startblock)); |
2bd0ea18 | 2737 | |
062998e3 | 2738 | /* |
49f693fa | 2739 | * Check and set flags if this segment has a left neighbor |
062998e3 | 2740 | */ |
49f693fa DC |
2741 | if (*idx > 0) { |
2742 | state |= BMAP_LEFT_VALID; | |
2743 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left); | |
5e656dbb | 2744 | |
49f693fa DC |
2745 | if (isnullstartblock(left.br_startblock)) |
2746 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2747 | } |
49f693fa DC |
2748 | |
2749 | /* | |
2750 | * Check and set flags if the current (right) segment exists. | |
2751 | * If it doesn't exist, we're converting the hole at end-of-file. | |
2752 | */ | |
d09d4e5f | 2753 | if (*idx < xfs_iext_count(ifp)) { |
49f693fa DC |
2754 | state |= BMAP_RIGHT_VALID; |
2755 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right); | |
2756 | ||
2757 | if (isnullstartblock(right.br_startblock)) | |
2758 | state |= BMAP_RIGHT_DELAY; | |
2759 | } | |
2760 | ||
2761 | /* | |
2762 | * Set contiguity flags on the left and right neighbors. | |
2763 | * Don't let extents get too large, even if the pieces are contiguous. | |
2764 | */ | |
2765 | if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) && | |
2766 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2767 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2768 | state |= BMAP_LEFT_CONTIG; | |
2769 | ||
2770 | if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) && | |
2771 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2772 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2773 | (!(state & BMAP_LEFT_CONTIG) || | |
2774 | (left.br_blockcount + new->br_blockcount + | |
2775 | right.br_blockcount <= MAXEXTLEN))) | |
2776 | state |= BMAP_RIGHT_CONTIG; | |
2777 | ||
2778 | /* | |
2779 | * Switch out based on the contiguity flags. | |
2780 | */ | |
2781 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2782 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
2783 | /* | |
2784 | * New allocation is contiguous with delayed allocations | |
2785 | * on the left and on the right. | |
2786 | * Merge all three into a single extent record. | |
2787 | */ | |
2788 | --*idx; | |
2789 | temp = left.br_blockcount + new->br_blockcount + | |
2790 | right.br_blockcount; | |
2791 | ||
2792 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2793 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp); | |
2794 | oldlen = startblockval(left.br_startblock) + | |
2795 | startblockval(new->br_startblock) + | |
2796 | startblockval(right.br_startblock); | |
2797 | newlen = xfs_bmap_worst_indlen(ip, temp); | |
2798 | xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx), | |
2799 | nullstartblock((int)newlen)); | |
2800 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2801 | ||
2802 | xfs_iext_remove(ip, *idx + 1, 1, state); | |
2803 | break; | |
2804 | ||
2805 | case BMAP_LEFT_CONTIG: | |
2806 | /* | |
2807 | * New allocation is contiguous with a delayed allocation | |
2808 | * on the left. | |
2809 | * Merge the new allocation with the left neighbor. | |
2810 | */ | |
2811 | --*idx; | |
2812 | temp = left.br_blockcount + new->br_blockcount; | |
2813 | ||
2814 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2815 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp); | |
2816 | oldlen = startblockval(left.br_startblock) + | |
2817 | startblockval(new->br_startblock); | |
2818 | newlen = xfs_bmap_worst_indlen(ip, temp); | |
2819 | xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx), | |
2820 | nullstartblock((int)newlen)); | |
2821 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2822 | break; | |
2823 | ||
2824 | case BMAP_RIGHT_CONTIG: | |
2825 | /* | |
2826 | * New allocation is contiguous with a delayed allocation | |
2827 | * on the right. | |
2828 | * Merge the new allocation with the right neighbor. | |
2829 | */ | |
2830 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
2831 | temp = new->br_blockcount + right.br_blockcount; | |
2832 | oldlen = startblockval(new->br_startblock) + | |
2833 | startblockval(right.br_startblock); | |
2834 | newlen = xfs_bmap_worst_indlen(ip, temp); | |
2835 | xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx), | |
2836 | new->br_startoff, | |
2837 | nullstartblock((int)newlen), temp, right.br_state); | |
2838 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
2839 | break; | |
2840 | ||
2841 | case 0: | |
2842 | /* | |
2843 | * New allocation is not contiguous with another | |
2844 | * delayed allocation. | |
2845 | * Insert a new entry. | |
2846 | */ | |
2847 | oldlen = newlen = 0; | |
2848 | xfs_iext_insert(ip, *idx, 1, new, state); | |
2849 | break; | |
2850 | } | |
2851 | if (oldlen != newlen) { | |
2852 | ASSERT(oldlen > newlen); | |
19ebedcf DC |
2853 | xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen), |
2854 | false); | |
49f693fa DC |
2855 | /* |
2856 | * Nothing to do for disk quota accounting here. | |
2857 | */ | |
2bd0ea18 | 2858 | } |
2bd0ea18 NS |
2859 | } |
2860 | ||
2861 | /* | |
49f693fa | 2862 | * Convert a hole to a real allocation. |
2bd0ea18 | 2863 | */ |
49f693fa DC |
2864 | STATIC int /* error */ |
2865 | xfs_bmap_add_extent_hole_real( | |
2866 | struct xfs_bmalloca *bma, | |
2867 | int whichfork) | |
5000d01d | 2868 | { |
49f693fa DC |
2869 | struct xfs_bmbt_irec *new = &bma->got; |
2870 | int error; /* error return value */ | |
2871 | int i; /* temp state */ | |
2872 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
2873 | xfs_bmbt_irec_t left; /* left neighbor extent entry */ | |
2874 | xfs_bmbt_irec_t right; /* right neighbor extent entry */ | |
2875 | int rval=0; /* return value (logging flags) */ | |
2876 | int state; /* state bits, accessed thru macros */ | |
19ebedcf | 2877 | struct xfs_mount *mp; |
2bd0ea18 | 2878 | |
65ca3804 | 2879 | mp = bma->ip->i_mount; |
49f693fa | 2880 | ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
2bd0ea18 | 2881 | |
49f693fa | 2882 | ASSERT(bma->idx >= 0); |
d09d4e5f | 2883 | ASSERT(bma->idx <= xfs_iext_count(ifp)); |
49f693fa DC |
2884 | ASSERT(!isnullstartblock(new->br_startblock)); |
2885 | ASSERT(!bma->cur || | |
2886 | !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL)); | |
cb8a004a | 2887 | ASSERT(whichfork != XFS_COW_FORK); |
5e656dbb | 2888 | |
79896434 | 2889 | XFS_STATS_INC(mp, xs_add_exlist); |
49f693fa DC |
2890 | |
2891 | state = 0; | |
2892 | if (whichfork == XFS_ATTR_FORK) | |
2893 | state |= BMAP_ATTRFORK; | |
2894 | ||
2895 | /* | |
2896 | * Check and set flags if this segment has a left neighbor. | |
2897 | */ | |
2898 | if (bma->idx > 0) { | |
2899 | state |= BMAP_LEFT_VALID; | |
2900 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left); | |
2901 | if (isnullstartblock(left.br_startblock)) | |
2902 | state |= BMAP_LEFT_DELAY; | |
5e656dbb | 2903 | } |
2bd0ea18 | 2904 | |
49f693fa DC |
2905 | /* |
2906 | * Check and set flags if this segment has a current value. | |
2907 | * Not true if we're inserting into the "hole" at eof. | |
2908 | */ | |
d09d4e5f | 2909 | if (bma->idx < xfs_iext_count(ifp)) { |
49f693fa DC |
2910 | state |= BMAP_RIGHT_VALID; |
2911 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right); | |
2912 | if (isnullstartblock(right.br_startblock)) | |
2913 | state |= BMAP_RIGHT_DELAY; | |
2bd0ea18 | 2914 | } |
2bd0ea18 | 2915 | |
49f693fa DC |
2916 | /* |
2917 | * We're inserting a real allocation between "left" and "right". | |
2918 | * Set the contiguity flags. Don't let extents get too large. | |
2919 | */ | |
2920 | if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) && | |
2921 | left.br_startoff + left.br_blockcount == new->br_startoff && | |
2922 | left.br_startblock + left.br_blockcount == new->br_startblock && | |
2923 | left.br_state == new->br_state && | |
2924 | left.br_blockcount + new->br_blockcount <= MAXEXTLEN) | |
2925 | state |= BMAP_LEFT_CONTIG; | |
57c9fccb | 2926 | |
49f693fa DC |
2927 | if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) && |
2928 | new->br_startoff + new->br_blockcount == right.br_startoff && | |
2929 | new->br_startblock + new->br_blockcount == right.br_startblock && | |
2930 | new->br_state == right.br_state && | |
2931 | new->br_blockcount + right.br_blockcount <= MAXEXTLEN && | |
2932 | (!(state & BMAP_LEFT_CONTIG) || | |
2933 | left.br_blockcount + new->br_blockcount + | |
2934 | right.br_blockcount <= MAXEXTLEN)) | |
2935 | state |= BMAP_RIGHT_CONTIG; | |
ca86e759 | 2936 | |
49f693fa DC |
2937 | error = 0; |
2938 | /* | |
2939 | * Select which case we're in here, and implement it. | |
2940 | */ | |
2941 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
2942 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
57c9fccb | 2943 | /* |
49f693fa DC |
2944 | * New allocation is contiguous with real allocations on the |
2945 | * left and on the right. | |
2946 | * Merge all three into a single extent record. | |
57c9fccb | 2947 | */ |
49f693fa DC |
2948 | --bma->idx; |
2949 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
2950 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx), | |
2951 | left.br_blockcount + new->br_blockcount + | |
2952 | right.br_blockcount); | |
2953 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
56b2de80 | 2954 | |
49f693fa | 2955 | xfs_iext_remove(bma->ip, bma->idx + 1, 1, state); |
56b2de80 | 2956 | |
49f693fa DC |
2957 | XFS_IFORK_NEXT_SET(bma->ip, whichfork, |
2958 | XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1); | |
2959 | if (bma->cur == NULL) { | |
2960 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); | |
2961 | } else { | |
2962 | rval = XFS_ILOG_CORE; | |
2963 | error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff, | |
2964 | right.br_startblock, right.br_blockcount, | |
2965 | &i); | |
2966 | if (error) | |
2967 | goto done; | |
19ebedcf | 2968 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2969 | error = xfs_btree_delete(bma->cur, &i); |
2970 | if (error) | |
2971 | goto done; | |
19ebedcf | 2972 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2973 | error = xfs_btree_decrement(bma->cur, 0, &i); |
2974 | if (error) | |
2975 | goto done; | |
19ebedcf | 2976 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
2977 | error = xfs_bmbt_update(bma->cur, left.br_startoff, |
2978 | left.br_startblock, | |
2979 | left.br_blockcount + | |
2980 | new->br_blockcount + | |
2981 | right.br_blockcount, | |
2982 | left.br_state); | |
2983 | if (error) | |
2984 | goto done; | |
2985 | } | |
57c9fccb | 2986 | break; |
49f693fa DC |
2987 | |
2988 | case BMAP_LEFT_CONTIG: | |
2989 | /* | |
2990 | * New allocation is contiguous with a real allocation | |
2991 | * on the left. | |
2992 | * Merge the new allocation with the left neighbor. | |
2993 | */ | |
2994 | --bma->idx; | |
2995 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
2996 | xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx), | |
2997 | left.br_blockcount + new->br_blockcount); | |
2998 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
2999 | ||
3000 | if (bma->cur == NULL) { | |
3001 | rval = xfs_ilog_fext(whichfork); | |
3002 | } else { | |
3003 | rval = 0; | |
3004 | error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff, | |
3005 | left.br_startblock, left.br_blockcount, | |
3006 | &i); | |
3007 | if (error) | |
3008 | goto done; | |
19ebedcf | 3009 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
3010 | error = xfs_bmbt_update(bma->cur, left.br_startoff, |
3011 | left.br_startblock, | |
3012 | left.br_blockcount + | |
3013 | new->br_blockcount, | |
3014 | left.br_state); | |
3015 | if (error) | |
3016 | goto done; | |
3017 | } | |
57c9fccb | 3018 | break; |
49f693fa DC |
3019 | |
3020 | case BMAP_RIGHT_CONTIG: | |
3021 | /* | |
3022 | * New allocation is contiguous with a real allocation | |
3023 | * on the right. | |
3024 | * Merge the new allocation with the right neighbor. | |
3025 | */ | |
3026 | trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_); | |
3027 | xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx), | |
3028 | new->br_startoff, new->br_startblock, | |
3029 | new->br_blockcount + right.br_blockcount, | |
3030 | right.br_state); | |
3031 | trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_); | |
3032 | ||
3033 | if (bma->cur == NULL) { | |
3034 | rval = xfs_ilog_fext(whichfork); | |
3035 | } else { | |
3036 | rval = 0; | |
3037 | error = xfs_bmbt_lookup_eq(bma->cur, | |
3038 | right.br_startoff, | |
3039 | right.br_startblock, | |
3040 | right.br_blockcount, &i); | |
3041 | if (error) | |
3042 | goto done; | |
19ebedcf | 3043 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
3044 | error = xfs_bmbt_update(bma->cur, new->br_startoff, |
3045 | new->br_startblock, | |
3046 | new->br_blockcount + | |
3047 | right.br_blockcount, | |
3048 | right.br_state); | |
3049 | if (error) | |
3050 | goto done; | |
3051 | } | |
3052 | break; | |
3053 | ||
3054 | case 0: | |
3055 | /* | |
3056 | * New allocation is not contiguous with another | |
3057 | * real allocation. | |
3058 | * Insert a new entry. | |
3059 | */ | |
3060 | xfs_iext_insert(bma->ip, bma->idx, 1, new, state); | |
3061 | XFS_IFORK_NEXT_SET(bma->ip, whichfork, | |
3062 | XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1); | |
3063 | if (bma->cur == NULL) { | |
3064 | rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork); | |
3065 | } else { | |
3066 | rval = XFS_ILOG_CORE; | |
3067 | error = xfs_bmbt_lookup_eq(bma->cur, | |
3068 | new->br_startoff, | |
3069 | new->br_startblock, | |
3070 | new->br_blockcount, &i); | |
3071 | if (error) | |
3072 | goto done; | |
19ebedcf | 3073 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done); |
49f693fa DC |
3074 | bma->cur->bc_rec.b.br_state = new->br_state; |
3075 | error = xfs_btree_insert(bma->cur, &i); | |
3076 | if (error) | |
3077 | goto done; | |
19ebedcf | 3078 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 3079 | } |
57c9fccb | 3080 | break; |
57c9fccb | 3081 | } |
a2ceac1f | 3082 | |
d7f80320 DW |
3083 | /* add reverse mapping */ |
3084 | error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new); | |
3085 | if (error) | |
3086 | goto done; | |
3087 | ||
49f693fa DC |
3088 | /* convert to a btree if necessary */ |
3089 | if (xfs_bmap_needs_btree(bma->ip, whichfork)) { | |
3090 | int tmp_logflags; /* partial log flag return val */ | |
3f853c7a | 3091 | |
49f693fa DC |
3092 | ASSERT(bma->cur == NULL); |
3093 | error = xfs_bmap_extents_to_btree(bma->tp, bma->ip, | |
f33cea1a | 3094 | bma->firstblock, bma->dfops, &bma->cur, |
49f693fa DC |
3095 | 0, &tmp_logflags, whichfork); |
3096 | bma->logflags |= tmp_logflags; | |
3097 | if (error) | |
3098 | goto done; | |
57c9fccb | 3099 | } |
a2ceac1f | 3100 | |
49f693fa DC |
3101 | /* clear out the allocated field, done with it now in any case. */ |
3102 | if (bma->cur) | |
3103 | bma->cur->bc_private.b.allocated = 0; | |
3104 | ||
3105 | xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork); | |
3106 | done: | |
3107 | bma->logflags |= rval; | |
57c9fccb NS |
3108 | return error; |
3109 | } | |
3110 | ||
2bd0ea18 | 3111 | /* |
49f693fa | 3112 | * Functions used in the extent read, allocate and remove paths |
2bd0ea18 | 3113 | */ |
2bd0ea18 | 3114 | |
5000d01d | 3115 | /* |
49f693fa | 3116 | * Adjust the size of the new extent based on di_extsize and rt extsize. |
2bd0ea18 | 3117 | */ |
613e6057 | 3118 | int |
49f693fa DC |
3119 | xfs_bmap_extsize_align( |
3120 | xfs_mount_t *mp, | |
3121 | xfs_bmbt_irec_t *gotp, /* next extent pointer */ | |
3122 | xfs_bmbt_irec_t *prevp, /* previous extent pointer */ | |
3123 | xfs_extlen_t extsz, /* align to this extent size */ | |
3124 | int rt, /* is this a realtime inode? */ | |
3125 | int eof, /* is extent at end-of-file? */ | |
3126 | int delay, /* creating delalloc extent? */ | |
3127 | int convert, /* overwriting unwritten extent? */ | |
3128 | xfs_fileoff_t *offp, /* in/out: aligned offset */ | |
3129 | xfs_extlen_t *lenp) /* in/out: aligned length */ | |
2bd0ea18 | 3130 | { |
49f693fa DC |
3131 | xfs_fileoff_t orig_off; /* original offset */ |
3132 | xfs_extlen_t orig_alen; /* original length */ | |
3133 | xfs_fileoff_t orig_end; /* original off+len */ | |
3134 | xfs_fileoff_t nexto; /* next file offset */ | |
3135 | xfs_fileoff_t prevo; /* previous file offset */ | |
3136 | xfs_fileoff_t align_off; /* temp for offset */ | |
3137 | xfs_extlen_t align_alen; /* temp for length */ | |
3138 | xfs_extlen_t temp; /* temp for calculations */ | |
3139 | ||
3140 | if (convert) | |
3141 | return 0; | |
3142 | ||
3143 | orig_off = align_off = *offp; | |
3144 | orig_alen = align_alen = *lenp; | |
3145 | orig_end = orig_off + orig_alen; | |
2bd0ea18 NS |
3146 | |
3147 | /* | |
49f693fa DC |
3148 | * If this request overlaps an existing extent, then don't |
3149 | * attempt to perform any additional alignment. | |
2bd0ea18 | 3150 | */ |
49f693fa DC |
3151 | if (!delay && !eof && |
3152 | (orig_off >= gotp->br_startoff) && | |
3153 | (orig_end <= gotp->br_startoff + gotp->br_blockcount)) { | |
3154 | return 0; | |
2bd0ea18 | 3155 | } |
57c9fccb | 3156 | |
49f693fa DC |
3157 | /* |
3158 | * If the file offset is unaligned vs. the extent size | |
3159 | * we need to align it. This will be possible unless | |
3160 | * the file was previously written with a kernel that didn't | |
3161 | * perform this alignment, or if a truncate shot us in the | |
3162 | * foot. | |
3163 | */ | |
3164 | temp = do_mod(orig_off, extsz); | |
3165 | if (temp) { | |
3166 | align_alen += temp; | |
3167 | align_off -= temp; | |
3168 | } | |
7cc23f0c DC |
3169 | |
3170 | /* Same adjustment for the end of the requested area. */ | |
3171 | temp = (align_alen % extsz); | |
3172 | if (temp) | |
3173 | align_alen += extsz - temp; | |
3174 | ||
49f693fa | 3175 | /* |
7cc23f0c DC |
3176 | * For large extent hint sizes, the aligned extent might be larger than |
3177 | * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls | |
3178 | * the length back under MAXEXTLEN. The outer allocation loops handle | |
3179 | * short allocation just fine, so it is safe to do this. We only want to | |
3180 | * do it when we are forced to, though, because it means more allocation | |
3181 | * operations are required. | |
49f693fa | 3182 | */ |
7cc23f0c DC |
3183 | while (align_alen > MAXEXTLEN) |
3184 | align_alen -= extsz; | |
3185 | ASSERT(align_alen <= MAXEXTLEN); | |
3186 | ||
49f693fa DC |
3187 | /* |
3188 | * If the previous block overlaps with this proposed allocation | |
3189 | * then move the start forward without adjusting the length. | |
3190 | */ | |
3191 | if (prevp->br_startoff != NULLFILEOFF) { | |
3192 | if (prevp->br_startblock == HOLESTARTBLOCK) | |
3193 | prevo = prevp->br_startoff; | |
3194 | else | |
3195 | prevo = prevp->br_startoff + prevp->br_blockcount; | |
3196 | } else | |
3197 | prevo = 0; | |
3198 | if (align_off != orig_off && align_off < prevo) | |
3199 | align_off = prevo; | |
3200 | /* | |
3201 | * If the next block overlaps with this proposed allocation | |
3202 | * then move the start back without adjusting the length, | |
3203 | * but not before offset 0. | |
3204 | * This may of course make the start overlap previous block, | |
3205 | * and if we hit the offset 0 limit then the next block | |
3206 | * can still overlap too. | |
3207 | */ | |
3208 | if (!eof && gotp->br_startoff != NULLFILEOFF) { | |
3209 | if ((delay && gotp->br_startblock == HOLESTARTBLOCK) || | |
3210 | (!delay && gotp->br_startblock == DELAYSTARTBLOCK)) | |
3211 | nexto = gotp->br_startoff + gotp->br_blockcount; | |
3212 | else | |
3213 | nexto = gotp->br_startoff; | |
3214 | } else | |
3215 | nexto = NULLFILEOFF; | |
3216 | if (!eof && | |
3217 | align_off + align_alen != orig_end && | |
3218 | align_off + align_alen > nexto) | |
3219 | align_off = nexto > align_alen ? nexto - align_alen : 0; | |
3220 | /* | |
3221 | * If we're now overlapping the next or previous extent that | |
3222 | * means we can't fit an extsz piece in this hole. Just move | |
3223 | * the start forward to the first valid spot and set | |
3224 | * the length so we hit the end. | |
3225 | */ | |
3226 | if (align_off != orig_off && align_off < prevo) | |
3227 | align_off = prevo; | |
3228 | if (align_off + align_alen != orig_end && | |
3229 | align_off + align_alen > nexto && | |
3230 | nexto != NULLFILEOFF) { | |
3231 | ASSERT(nexto > prevo); | |
3232 | align_alen = nexto - align_off; | |
57c9fccb | 3233 | } |
2bd0ea18 | 3234 | |
49f693fa DC |
3235 | /* |
3236 | * If realtime, and the result isn't a multiple of the realtime | |
3237 | * extent size we need to remove blocks until it is. | |
3238 | */ | |
3239 | if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) { | |
2bd0ea18 | 3240 | /* |
49f693fa DC |
3241 | * We're not covering the original request, or |
3242 | * we won't be able to once we fix the length. | |
2bd0ea18 | 3243 | */ |
49f693fa DC |
3244 | if (orig_off < align_off || |
3245 | orig_end > align_off + align_alen || | |
3246 | align_alen - temp < orig_alen) | |
12b53197 | 3247 | return -EINVAL; |
49f693fa DC |
3248 | /* |
3249 | * Try to fix it by moving the start up. | |
3250 | */ | |
3251 | if (align_off + temp <= orig_off) { | |
3252 | align_alen -= temp; | |
3253 | align_off += temp; | |
2bd0ea18 | 3254 | } |
49f693fa DC |
3255 | /* |
3256 | * Try to fix it by moving the end in. | |
3257 | */ | |
3258 | else if (align_off + align_alen - temp >= orig_end) | |
3259 | align_alen -= temp; | |
3260 | /* | |
3261 | * Set the start to the minimum then trim the length. | |
3262 | */ | |
3263 | else { | |
3264 | align_alen -= orig_off - align_off; | |
3265 | align_off = orig_off; | |
3266 | align_alen -= align_alen % mp->m_sb.sb_rextsize; | |
3267 | } | |
3268 | /* | |
3269 | * Result doesn't cover the request, fail it. | |
3270 | */ | |
3271 | if (orig_off < align_off || orig_end > align_off + align_alen) | |
12b53197 | 3272 | return -EINVAL; |
49f693fa DC |
3273 | } else { |
3274 | ASSERT(orig_off >= align_off); | |
7cc23f0c DC |
3275 | /* see MAXEXTLEN handling above */ |
3276 | ASSERT(orig_end <= align_off + align_alen || | |
3277 | align_alen + extsz > MAXEXTLEN); | |
2bd0ea18 | 3278 | } |
49f693fa DC |
3279 | |
3280 | #ifdef DEBUG | |
3281 | if (!eof && gotp->br_startoff != NULLFILEOFF) | |
3282 | ASSERT(align_off + align_alen <= gotp->br_startoff); | |
3283 | if (prevp->br_startoff != NULLFILEOFF) | |
3284 | ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount); | |
3285 | #endif | |
3286 | ||
3287 | *lenp = align_alen; | |
3288 | *offp = align_off; | |
2bd0ea18 NS |
3289 | return 0; |
3290 | } | |
3291 | ||
49f693fa DC |
3292 | #define XFS_ALLOC_GAP_UNITS 4 |
3293 | ||
613e6057 | 3294 | void |
49f693fa | 3295 | xfs_bmap_adjacent( |
613e6057 | 3296 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
2bd0ea18 | 3297 | { |
49f693fa DC |
3298 | xfs_fsblock_t adjust; /* adjustment to block numbers */ |
3299 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ | |
3300 | xfs_mount_t *mp; /* mount point structure */ | |
3301 | int nullfb; /* true if ap->firstblock isn't set */ | |
3302 | int rt; /* true if inode is realtime */ | |
2bd0ea18 | 3303 | |
49f693fa DC |
3304 | #define ISVALID(x,y) \ |
3305 | (rt ? \ | |
3306 | (x) < mp->m_sb.sb_rblocks : \ | |
3307 | XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \ | |
3308 | XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \ | |
3309 | XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks) | |
3310 | ||
3311 | mp = ap->ip->i_mount; | |
3312 | nullfb = *ap->firstblock == NULLFSBLOCK; | |
1fccd5c8 DC |
3313 | rt = XFS_IS_REALTIME_INODE(ap->ip) && |
3314 | xfs_alloc_is_userdata(ap->datatype); | |
49f693fa DC |
3315 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock); |
3316 | /* | |
3317 | * If allocating at eof, and there's a previous real block, | |
3318 | * try to use its last block as our starting point. | |
3319 | */ | |
3320 | if (ap->eof && ap->prev.br_startoff != NULLFILEOFF && | |
3321 | !isnullstartblock(ap->prev.br_startblock) && | |
3322 | ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount, | |
3323 | ap->prev.br_startblock)) { | |
3324 | ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount; | |
3325 | /* | |
3326 | * Adjust for the gap between prevp and us. | |
3327 | */ | |
3328 | adjust = ap->offset - | |
3329 | (ap->prev.br_startoff + ap->prev.br_blockcount); | |
3330 | if (adjust && | |
3331 | ISVALID(ap->blkno + adjust, ap->prev.br_startblock)) | |
3332 | ap->blkno += adjust; | |
2bd0ea18 | 3333 | } |
49f693fa DC |
3334 | /* |
3335 | * If not at eof, then compare the two neighbor blocks. | |
3336 | * Figure out whether either one gives us a good starting point, | |
3337 | * and pick the better one. | |
3338 | */ | |
3339 | else if (!ap->eof) { | |
3340 | xfs_fsblock_t gotbno; /* right side block number */ | |
3341 | xfs_fsblock_t gotdiff=0; /* right side difference */ | |
3342 | xfs_fsblock_t prevbno; /* left side block number */ | |
3343 | xfs_fsblock_t prevdiff=0; /* left side difference */ | |
3344 | ||
3345 | /* | |
3346 | * If there's a previous (left) block, select a requested | |
3347 | * start block based on it. | |
3348 | */ | |
3349 | if (ap->prev.br_startoff != NULLFILEOFF && | |
3350 | !isnullstartblock(ap->prev.br_startblock) && | |
3351 | (prevbno = ap->prev.br_startblock + | |
3352 | ap->prev.br_blockcount) && | |
3353 | ISVALID(prevbno, ap->prev.br_startblock)) { | |
3354 | /* | |
3355 | * Calculate gap to end of previous block. | |
3356 | */ | |
3357 | adjust = prevdiff = ap->offset - | |
3358 | (ap->prev.br_startoff + | |
3359 | ap->prev.br_blockcount); | |
3360 | /* | |
3361 | * Figure the startblock based on the previous block's | |
3362 | * end and the gap size. | |
3363 | * Heuristic! | |
3364 | * If the gap is large relative to the piece we're | |
3365 | * allocating, or using it gives us an invalid block | |
3366 | * number, then just use the end of the previous block. | |
3367 | */ | |
3368 | if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3369 | ISVALID(prevbno + prevdiff, | |
3370 | ap->prev.br_startblock)) | |
3371 | prevbno += adjust; | |
3372 | else | |
3373 | prevdiff += adjust; | |
3374 | /* | |
3375 | * If the firstblock forbids it, can't use it, | |
3376 | * must use default. | |
3377 | */ | |
3378 | if (!rt && !nullfb && | |
3379 | XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno) | |
3380 | prevbno = NULLFSBLOCK; | |
3381 | } | |
3382 | /* | |
3383 | * No previous block or can't follow it, just default. | |
3384 | */ | |
3385 | else | |
3386 | prevbno = NULLFSBLOCK; | |
3387 | /* | |
3388 | * If there's a following (right) block, select a requested | |
3389 | * start block based on it. | |
3390 | */ | |
3391 | if (!isnullstartblock(ap->got.br_startblock)) { | |
3392 | /* | |
3393 | * Calculate gap to start of next block. | |
3394 | */ | |
3395 | adjust = gotdiff = ap->got.br_startoff - ap->offset; | |
3396 | /* | |
3397 | * Figure the startblock based on the next block's | |
3398 | * start and the gap size. | |
3399 | */ | |
3400 | gotbno = ap->got.br_startblock; | |
3401 | /* | |
3402 | * Heuristic! | |
3403 | * If the gap is large relative to the piece we're | |
3404 | * allocating, or using it gives us an invalid block | |
3405 | * number, then just use the start of the next block | |
3406 | * offset by our length. | |
3407 | */ | |
3408 | if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length && | |
3409 | ISVALID(gotbno - gotdiff, gotbno)) | |
3410 | gotbno -= adjust; | |
3411 | else if (ISVALID(gotbno - ap->length, gotbno)) { | |
3412 | gotbno -= ap->length; | |
3413 | gotdiff += adjust - ap->length; | |
3414 | } else | |
3415 | gotdiff += adjust; | |
3416 | /* | |
3417 | * If the firstblock forbids it, can't use it, | |
3418 | * must use default. | |
3419 | */ | |
3420 | if (!rt && !nullfb && | |
3421 | XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno) | |
3422 | gotbno = NULLFSBLOCK; | |
3423 | } | |
3424 | /* | |
3425 | * No next block, just default. | |
3426 | */ | |
2bd0ea18 | 3427 | else |
49f693fa DC |
3428 | gotbno = NULLFSBLOCK; |
3429 | /* | |
3430 | * If both valid, pick the better one, else the only good | |
3431 | * one, else ap->blkno is already set (to 0 or the inode block). | |
3432 | */ | |
3433 | if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) | |
3434 | ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno; | |
3435 | else if (prevbno != NULLFSBLOCK) | |
3436 | ap->blkno = prevbno; | |
3437 | else if (gotbno != NULLFSBLOCK) | |
3438 | ap->blkno = gotbno; | |
a2ceac1f | 3439 | } |
49f693fa | 3440 | #undef ISVALID |
a2ceac1f DC |
3441 | } |
3442 | ||
ff105f75 DC |
3443 | static int |
3444 | xfs_bmap_longest_free_extent( | |
3445 | struct xfs_trans *tp, | |
3446 | xfs_agnumber_t ag, | |
3447 | xfs_extlen_t *blen, | |
3448 | int *notinit) | |
3449 | { | |
3450 | struct xfs_mount *mp = tp->t_mountp; | |
3451 | struct xfs_perag *pag; | |
3452 | xfs_extlen_t longest; | |
3453 | int error = 0; | |
3454 | ||
3455 | pag = xfs_perag_get(mp, ag); | |
3456 | if (!pag->pagf_init) { | |
3457 | error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK); | |
3458 | if (error) | |
3459 | goto out; | |
3460 | ||
3461 | if (!pag->pagf_init) { | |
3462 | *notinit = 1; | |
3463 | goto out; | |
3464 | } | |
3465 | } | |
3466 | ||
72bda06d | 3467 | longest = xfs_alloc_longest_free_extent(mp, pag, |
cf8ce220 DW |
3468 | xfs_alloc_min_freelist(mp, pag), |
3469 | xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE)); | |
ff105f75 DC |
3470 | if (*blen < longest) |
3471 | *blen = longest; | |
3472 | ||
3473 | out: | |
3474 | xfs_perag_put(pag); | |
3475 | return error; | |
3476 | } | |
3477 | ||
3478 | static void | |
3479 | xfs_bmap_select_minlen( | |
3480 | struct xfs_bmalloca *ap, | |
3481 | struct xfs_alloc_arg *args, | |
3482 | xfs_extlen_t *blen, | |
3483 | int notinit) | |
3484 | { | |
3485 | if (notinit || *blen < ap->minlen) { | |
3486 | /* | |
3487 | * Since we did a BUF_TRYLOCK above, it is possible that | |
3488 | * there is space for this request. | |
3489 | */ | |
3490 | args->minlen = ap->minlen; | |
3491 | } else if (*blen < args->maxlen) { | |
3492 | /* | |
3493 | * If the best seen length is less than the request length, | |
3494 | * use the best as the minimum. | |
3495 | */ | |
3496 | args->minlen = *blen; | |
3497 | } else { | |
3498 | /* | |
3499 | * Otherwise we've seen an extent as big as maxlen, use that | |
3500 | * as the minimum. | |
3501 | */ | |
3502 | args->minlen = args->maxlen; | |
3503 | } | |
3504 | } | |
3505 | ||
a2ceac1f | 3506 | STATIC int |
49f693fa DC |
3507 | xfs_bmap_btalloc_nullfb( |
3508 | struct xfs_bmalloca *ap, | |
3509 | struct xfs_alloc_arg *args, | |
3510 | xfs_extlen_t *blen) | |
a2ceac1f | 3511 | { |
49f693fa | 3512 | struct xfs_mount *mp = ap->ip->i_mount; |
49f693fa DC |
3513 | xfs_agnumber_t ag, startag; |
3514 | int notinit = 0; | |
a2ceac1f DC |
3515 | int error; |
3516 | ||
ff105f75 | 3517 | args->type = XFS_ALLOCTYPE_START_BNO; |
49f693fa | 3518 | args->total = ap->total; |
a2ceac1f | 3519 | |
49f693fa DC |
3520 | startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno); |
3521 | if (startag == NULLAGNUMBER) | |
3522 | startag = ag = 0; | |
a2ceac1f | 3523 | |
49f693fa | 3524 | while (*blen < args->maxlen) { |
ff105f75 DC |
3525 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, |
3526 | ¬init); | |
3527 | if (error) | |
3528 | return error; | |
a2ceac1f | 3529 | |
49f693fa DC |
3530 | if (++ag == mp->m_sb.sb_agcount) |
3531 | ag = 0; | |
3532 | if (ag == startag) | |
3533 | break; | |
49f693fa | 3534 | } |
2bd0ea18 | 3535 | |
ff105f75 DC |
3536 | xfs_bmap_select_minlen(ap, args, blen, notinit); |
3537 | return 0; | |
3538 | } | |
3539 | ||
3540 | STATIC int | |
3541 | xfs_bmap_btalloc_filestreams( | |
3542 | struct xfs_bmalloca *ap, | |
3543 | struct xfs_alloc_arg *args, | |
3544 | xfs_extlen_t *blen) | |
3545 | { | |
3546 | struct xfs_mount *mp = ap->ip->i_mount; | |
3547 | xfs_agnumber_t ag; | |
3548 | int notinit = 0; | |
3549 | int error; | |
3550 | ||
3551 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
3552 | args->total = ap->total; | |
3553 | ||
3554 | ag = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3555 | if (ag == NULLAGNUMBER) | |
3556 | ag = 0; | |
3557 | ||
3558 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init); | |
3559 | if (error) | |
3560 | return error; | |
3561 | ||
3562 | if (*blen < args->maxlen) { | |
3563 | error = xfs_filestream_new_ag(ap, &ag); | |
3564 | if (error) | |
3565 | return error; | |
3566 | ||
3567 | error = xfs_bmap_longest_free_extent(args->tp, ag, blen, | |
3568 | ¬init); | |
3569 | if (error) | |
3570 | return error; | |
3571 | ||
3572 | } | |
3573 | ||
3574 | xfs_bmap_select_minlen(ap, args, blen, notinit); | |
2bd0ea18 | 3575 | |
49f693fa | 3576 | /* |
ff105f75 DC |
3577 | * Set the failure fallback case to look in the selected AG as stream |
3578 | * may have moved. | |
49f693fa | 3579 | */ |
ff105f75 | 3580 | ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0); |
49f693fa | 3581 | return 0; |
2bd0ea18 NS |
3582 | } |
3583 | ||
b3563c19 | 3584 | STATIC int |
49f693fa | 3585 | xfs_bmap_btalloc( |
613e6057 | 3586 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
b3563c19 | 3587 | { |
49f693fa DC |
3588 | xfs_mount_t *mp; /* mount point structure */ |
3589 | xfs_alloctype_t atype = 0; /* type for allocation routines */ | |
1fccd5c8 | 3590 | xfs_extlen_t align = 0; /* minimum allocation alignment */ |
49f693fa DC |
3591 | xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */ |
3592 | xfs_agnumber_t ag; | |
3593 | xfs_alloc_arg_t args; | |
3594 | xfs_extlen_t blen; | |
3595 | xfs_extlen_t nextminlen = 0; | |
3596 | int nullfb; /* true if ap->firstblock isn't set */ | |
3597 | int isaligned; | |
3598 | int tryagain; | |
3599 | int error; | |
ff105f75 | 3600 | int stripe_align; |
b3563c19 | 3601 | |
49f693fa | 3602 | ASSERT(ap->length); |
b3563c19 | 3603 | |
49f693fa | 3604 | mp = ap->ip->i_mount; |
ff105f75 DC |
3605 | |
3606 | /* stripe alignment for allocation is determined by mount parameters */ | |
3607 | stripe_align = 0; | |
3608 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
3609 | stripe_align = mp->m_swidth; | |
3610 | else if (mp->m_dalign) | |
3611 | stripe_align = mp->m_dalign; | |
3612 | ||
10460994 DW |
3613 | if (ap->flags & XFS_BMAPI_COWFORK) |
3614 | align = xfs_get_cowextsz_hint(ap->ip); | |
3615 | else if (xfs_alloc_is_userdata(ap->datatype)) | |
1fccd5c8 | 3616 | align = xfs_get_extsz_hint(ap->ip); |
49f693fa DC |
3617 | if (unlikely(align)) { |
3618 | error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, | |
3619 | align, 0, ap->eof, 0, ap->conv, | |
3620 | &ap->offset, &ap->length); | |
3621 | ASSERT(!error); | |
3622 | ASSERT(ap->length); | |
3623 | } | |
ff105f75 DC |
3624 | |
3625 | ||
49f693fa DC |
3626 | nullfb = *ap->firstblock == NULLFSBLOCK; |
3627 | fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock); | |
3628 | if (nullfb) { | |
1fccd5c8 DC |
3629 | if (xfs_alloc_is_userdata(ap->datatype) && |
3630 | xfs_inode_is_filestream(ap->ip)) { | |
49f693fa DC |
3631 | ag = xfs_filestream_lookup_ag(ap->ip); |
3632 | ag = (ag != NULLAGNUMBER) ? ag : 0; | |
3633 | ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0); | |
3634 | } else { | |
3635 | ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino); | |
3636 | } | |
3637 | } else | |
3638 | ap->blkno = *ap->firstblock; | |
3639 | ||
3640 | xfs_bmap_adjacent(ap); | |
2bd0ea18 | 3641 | |
2bd0ea18 | 3642 | /* |
49f693fa DC |
3643 | * If allowed, use ap->blkno; otherwise must use firstblock since |
3644 | * it's in the right allocation group. | |
2bd0ea18 | 3645 | */ |
49f693fa DC |
3646 | if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno) |
3647 | ; | |
3648 | else | |
3649 | ap->blkno = *ap->firstblock; | |
2bd0ea18 | 3650 | /* |
49f693fa | 3651 | * Normal allocation, done through xfs_alloc_vextent. |
2bd0ea18 | 3652 | */ |
49f693fa DC |
3653 | tryagain = isaligned = 0; |
3654 | memset(&args, 0, sizeof(args)); | |
3655 | args.tp = ap->tp; | |
3656 | args.mp = mp; | |
3657 | args.fsbno = ap->blkno; | |
85aec44f | 3658 | xfs_rmap_skip_owner_update(&args.oinfo); |
49f693fa DC |
3659 | |
3660 | /* Trim the allocation back to the maximum an AG can fit. */ | |
b8a8d6e5 | 3661 | args.maxlen = MIN(ap->length, mp->m_ag_max_usable); |
49f693fa DC |
3662 | args.firstblock = *ap->firstblock; |
3663 | blen = 0; | |
3664 | if (nullfb) { | |
ff105f75 DC |
3665 | /* |
3666 | * Search for an allocation group with a single extent large | |
3667 | * enough for the request. If one isn't found, then adjust | |
3668 | * the minimum allocation size to the largest space found. | |
3669 | */ | |
1fccd5c8 DC |
3670 | if (xfs_alloc_is_userdata(ap->datatype) && |
3671 | xfs_inode_is_filestream(ap->ip)) | |
ff105f75 DC |
3672 | error = xfs_bmap_btalloc_filestreams(ap, &args, &blen); |
3673 | else | |
3674 | error = xfs_bmap_btalloc_nullfb(ap, &args, &blen); | |
a2ceac1f | 3675 | if (error) |
2bd0ea18 | 3676 | return error; |
f33cea1a | 3677 | } else if (ap->dfops->dop_low) { |
49f693fa DC |
3678 | if (xfs_inode_is_filestream(ap->ip)) |
3679 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3680 | else | |
3681 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3682 | args.total = args.minlen = ap->minlen; | |
3683 | } else { | |
3684 | args.type = XFS_ALLOCTYPE_NEAR_BNO; | |
3685 | args.total = ap->total; | |
3686 | args.minlen = ap->minlen; | |
3687 | } | |
3688 | /* apply extent size hints if obtained earlier */ | |
3689 | if (unlikely(align)) { | |
3690 | args.prod = align; | |
3691 | if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod))) | |
3692 | args.mod = (xfs_extlen_t)(args.prod - args.mod); | |
b2327e1a | 3693 | } else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) { |
49f693fa DC |
3694 | args.prod = 1; |
3695 | args.mod = 0; | |
3696 | } else { | |
b2327e1a | 3697 | args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog; |
49f693fa DC |
3698 | if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod)))) |
3699 | args.mod = (xfs_extlen_t)(args.prod - args.mod); | |
2bd0ea18 NS |
3700 | } |
3701 | /* | |
49f693fa DC |
3702 | * If we are not low on available data blocks, and the |
3703 | * underlying logical volume manager is a stripe, and | |
3704 | * the file offset is zero then try to allocate data | |
3705 | * blocks on stripe unit boundary. | |
3706 | * NOTE: ap->aeof is only set if the allocation length | |
3707 | * is >= the stripe unit and the allocation offset is | |
3708 | * at the end of file. | |
2bd0ea18 | 3709 | */ |
f33cea1a | 3710 | if (!ap->dfops->dop_low && ap->aeof) { |
49f693fa | 3711 | if (!ap->offset) { |
ff105f75 | 3712 | args.alignment = stripe_align; |
49f693fa DC |
3713 | atype = args.type; |
3714 | isaligned = 1; | |
3715 | /* | |
3716 | * Adjust for alignment | |
3717 | */ | |
3718 | if (blen > args.alignment && blen <= args.maxlen) | |
3719 | args.minlen = blen - args.alignment; | |
3720 | args.minalignslop = 0; | |
3721 | } else { | |
3722 | /* | |
3723 | * First try an exact bno allocation. | |
3724 | * If it fails then do a near or start bno | |
3725 | * allocation with alignment turned on. | |
3726 | */ | |
3727 | atype = args.type; | |
3728 | tryagain = 1; | |
3729 | args.type = XFS_ALLOCTYPE_THIS_BNO; | |
3730 | args.alignment = 1; | |
3731 | /* | |
3732 | * Compute the minlen+alignment for the | |
3733 | * next case. Set slop so that the value | |
3734 | * of minlen+alignment+slop doesn't go up | |
3735 | * between the calls. | |
3736 | */ | |
ff105f75 DC |
3737 | if (blen > stripe_align && blen <= args.maxlen) |
3738 | nextminlen = blen - stripe_align; | |
49f693fa DC |
3739 | else |
3740 | nextminlen = args.minlen; | |
ff105f75 | 3741 | if (nextminlen + stripe_align > args.minlen + 1) |
49f693fa | 3742 | args.minalignslop = |
ff105f75 | 3743 | nextminlen + stripe_align - |
49f693fa DC |
3744 | args.minlen - 1; |
3745 | else | |
3746 | args.minalignslop = 0; | |
2bd0ea18 | 3747 | } |
49f693fa DC |
3748 | } else { |
3749 | args.alignment = 1; | |
3750 | args.minalignslop = 0; | |
3751 | } | |
3752 | args.minleft = ap->minleft; | |
3753 | args.wasdel = ap->wasdel; | |
cf8ce220 | 3754 | args.resv = XFS_AG_RESV_NONE; |
1fccd5c8 DC |
3755 | args.datatype = ap->datatype; |
3756 | if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) | |
9542ae13 DC |
3757 | args.ip = ap->ip; |
3758 | ||
3759 | error = xfs_alloc_vextent(&args); | |
3760 | if (error) | |
49f693fa | 3761 | return error; |
9542ae13 | 3762 | |
49f693fa DC |
3763 | if (tryagain && args.fsbno == NULLFSBLOCK) { |
3764 | /* | |
3765 | * Exact allocation failed. Now try with alignment | |
3766 | * turned on. | |
3767 | */ | |
3768 | args.type = atype; | |
3769 | args.fsbno = ap->blkno; | |
ff105f75 | 3770 | args.alignment = stripe_align; |
49f693fa DC |
3771 | args.minlen = nextminlen; |
3772 | args.minalignslop = 0; | |
3773 | isaligned = 1; | |
3774 | if ((error = xfs_alloc_vextent(&args))) | |
3775 | return error; | |
3776 | } | |
3777 | if (isaligned && args.fsbno == NULLFSBLOCK) { | |
2bd0ea18 | 3778 | /* |
49f693fa DC |
3779 | * allocation failed, so turn off alignment and |
3780 | * try again. | |
2bd0ea18 | 3781 | */ |
49f693fa DC |
3782 | args.type = atype; |
3783 | args.fsbno = ap->blkno; | |
3784 | args.alignment = 0; | |
3785 | if ((error = xfs_alloc_vextent(&args))) | |
3786 | return error; | |
3787 | } | |
3788 | if (args.fsbno == NULLFSBLOCK && nullfb && | |
3789 | args.minlen > ap->minlen) { | |
3790 | args.minlen = ap->minlen; | |
3791 | args.type = XFS_ALLOCTYPE_START_BNO; | |
3792 | args.fsbno = ap->blkno; | |
3793 | if ((error = xfs_alloc_vextent(&args))) | |
3794 | return error; | |
3795 | } | |
3796 | if (args.fsbno == NULLFSBLOCK && nullfb) { | |
3797 | args.fsbno = 0; | |
3798 | args.type = XFS_ALLOCTYPE_FIRST_AG; | |
3799 | args.total = ap->minlen; | |
3800 | args.minleft = 0; | |
3801 | if ((error = xfs_alloc_vextent(&args))) | |
3802 | return error; | |
f33cea1a | 3803 | ap->dfops->dop_low = true; |
49f693fa DC |
3804 | } |
3805 | if (args.fsbno != NULLFSBLOCK) { | |
2bd0ea18 | 3806 | /* |
49f693fa DC |
3807 | * check the allocation happened at the same or higher AG than |
3808 | * the first block that was allocated. | |
2bd0ea18 | 3809 | */ |
49f693fa DC |
3810 | ASSERT(*ap->firstblock == NULLFSBLOCK || |
3811 | XFS_FSB_TO_AGNO(mp, *ap->firstblock) == | |
3812 | XFS_FSB_TO_AGNO(mp, args.fsbno) || | |
f33cea1a | 3813 | (ap->dfops->dop_low && |
49f693fa DC |
3814 | XFS_FSB_TO_AGNO(mp, *ap->firstblock) < |
3815 | XFS_FSB_TO_AGNO(mp, args.fsbno))); | |
3816 | ||
3817 | ap->blkno = args.fsbno; | |
3818 | if (*ap->firstblock == NULLFSBLOCK) | |
3819 | *ap->firstblock = args.fsbno; | |
3820 | ASSERT(nullfb || fb_agno == args.agno || | |
f33cea1a | 3821 | (ap->dfops->dop_low && fb_agno < args.agno)); |
49f693fa | 3822 | ap->length = args.len; |
1277a5e0 DW |
3823 | if (!(ap->flags & XFS_BMAPI_COWFORK)) |
3824 | ap->ip->i_d.di_nblocks += args.len; | |
49f693fa DC |
3825 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); |
3826 | if (ap->wasdel) | |
3827 | ap->ip->i_delayed_blks -= args.len; | |
2bd0ea18 | 3828 | /* |
49f693fa DC |
3829 | * Adjust the disk quota also. This was reserved |
3830 | * earlier. | |
2bd0ea18 | 3831 | */ |
49f693fa DC |
3832 | xfs_trans_mod_dquot_byino(ap->tp, ap->ip, |
3833 | ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT : | |
3834 | XFS_TRANS_DQ_BCOUNT, | |
3835 | (long) args.len); | |
3836 | } else { | |
3837 | ap->blkno = NULLFSBLOCK; | |
3838 | ap->length = 0; | |
2bd0ea18 | 3839 | } |
2bd0ea18 | 3840 | return 0; |
56b2de80 DC |
3841 | } |
3842 | ||
34621a47 DW |
3843 | /* |
3844 | * For a remap operation, just "allocate" an extent at the address that the | |
3845 | * caller passed in, and ensure that the AGFL is the right size. The caller | |
3846 | * will then map the "allocated" extent into the file somewhere. | |
3847 | */ | |
3848 | STATIC int | |
3849 | xfs_bmap_remap_alloc( | |
3850 | struct xfs_bmalloca *ap) | |
3851 | { | |
3852 | struct xfs_trans *tp = ap->tp; | |
3853 | struct xfs_mount *mp = tp->t_mountp; | |
3854 | xfs_agblock_t bno; | |
3855 | struct xfs_alloc_arg args; | |
3856 | int error; | |
3857 | ||
3858 | /* | |
3859 | * validate that the block number is legal - the enables us to detect | |
3860 | * and handle a silent filesystem corruption rather than crashing. | |
3861 | */ | |
3862 | memset(&args, 0, sizeof(struct xfs_alloc_arg)); | |
3863 | args.tp = ap->tp; | |
3864 | args.mp = ap->tp->t_mountp; | |
3865 | bno = *ap->firstblock; | |
3866 | args.agno = XFS_FSB_TO_AGNO(mp, bno); | |
3867 | args.agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
3868 | if (args.agno >= mp->m_sb.sb_agcount || | |
3869 | args.agbno >= mp->m_sb.sb_agblocks) | |
3870 | return -EFSCORRUPTED; | |
3871 | ||
3872 | /* "Allocate" the extent from the range we passed in. */ | |
3873 | trace_xfs_bmap_remap_alloc(ap->ip, *ap->firstblock, ap->length); | |
3874 | ap->blkno = bno; | |
3875 | ap->ip->i_d.di_nblocks += ap->length; | |
3876 | xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); | |
3877 | ||
3878 | /* Fix the freelist, like a real allocator does. */ | |
3879 | args.datatype = ap->datatype; | |
3880 | args.pag = xfs_perag_get(args.mp, args.agno); | |
3881 | ASSERT(args.pag); | |
3882 | ||
3883 | /* | |
3884 | * The freelist fixing code will decline the allocation if | |
3885 | * the size and shape of the free space doesn't allow for | |
3886 | * allocating the extent and updating all the metadata that | |
3887 | * happens during an allocation. We're remapping, not | |
3888 | * allocating, so skip that check by pretending to be freeing. | |
3889 | */ | |
3890 | error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING); | |
34621a47 DW |
3891 | xfs_perag_put(args.pag); |
3892 | if (error) | |
3893 | trace_xfs_bmap_remap_alloc_error(ap->ip, error, _RET_IP_); | |
3894 | return error; | |
3895 | } | |
3896 | ||
56b2de80 | 3897 | /* |
49f693fa DC |
3898 | * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file. |
3899 | * It figures out where to ask the underlying allocator to put the new extent. | |
56b2de80 | 3900 | */ |
49f693fa DC |
3901 | STATIC int |
3902 | xfs_bmap_alloc( | |
613e6057 | 3903 | struct xfs_bmalloca *ap) /* bmap alloc argument struct */ |
56b2de80 | 3904 | { |
34621a47 DW |
3905 | if (ap->flags & XFS_BMAPI_REMAP) |
3906 | return xfs_bmap_remap_alloc(ap); | |
1fccd5c8 DC |
3907 | if (XFS_IS_REALTIME_INODE(ap->ip) && |
3908 | xfs_alloc_is_userdata(ap->datatype)) | |
49f693fa DC |
3909 | return xfs_bmap_rtalloc(ap); |
3910 | return xfs_bmap_btalloc(ap); | |
56b2de80 | 3911 | } |
56b2de80 | 3912 | |
b3fd8db7 DW |
3913 | /* Trim extent to fit a logical block range. */ |
3914 | void | |
3915 | xfs_trim_extent( | |
3916 | struct xfs_bmbt_irec *irec, | |
3917 | xfs_fileoff_t bno, | |
3918 | xfs_filblks_t len) | |
3919 | { | |
3920 | xfs_fileoff_t distance; | |
3921 | xfs_fileoff_t end = bno + len; | |
3922 | ||
3923 | if (irec->br_startoff + irec->br_blockcount <= bno || | |
3924 | irec->br_startoff >= end) { | |
3925 | irec->br_blockcount = 0; | |
3926 | return; | |
3927 | } | |
3928 | ||
3929 | if (irec->br_startoff < bno) { | |
3930 | distance = bno - irec->br_startoff; | |
3931 | if (isnullstartblock(irec->br_startblock)) | |
3932 | irec->br_startblock = DELAYSTARTBLOCK; | |
3933 | if (irec->br_startblock != DELAYSTARTBLOCK && | |
3934 | irec->br_startblock != HOLESTARTBLOCK) | |
3935 | irec->br_startblock += distance; | |
3936 | irec->br_startoff += distance; | |
3937 | irec->br_blockcount -= distance; | |
3938 | } | |
3939 | ||
3940 | if (end < irec->br_startoff + irec->br_blockcount) { | |
3941 | distance = irec->br_startoff + irec->br_blockcount - end; | |
3942 | irec->br_blockcount -= distance; | |
3943 | } | |
3944 | } | |
3945 | ||
2bd0ea18 | 3946 | /* |
a2ceac1f | 3947 | * Trim the returned map to the required bounds |
2bd0ea18 | 3948 | */ |
a2ceac1f DC |
3949 | STATIC void |
3950 | xfs_bmapi_trim_map( | |
3951 | struct xfs_bmbt_irec *mval, | |
3952 | struct xfs_bmbt_irec *got, | |
3953 | xfs_fileoff_t *bno, | |
3954 | xfs_filblks_t len, | |
3955 | xfs_fileoff_t obno, | |
3956 | xfs_fileoff_t end, | |
3957 | int n, | |
3958 | int flags) | |
2bd0ea18 | 3959 | { |
a2ceac1f DC |
3960 | if ((flags & XFS_BMAPI_ENTIRE) || |
3961 | got->br_startoff + got->br_blockcount <= obno) { | |
3962 | *mval = *got; | |
3963 | if (isnullstartblock(got->br_startblock)) | |
3964 | mval->br_startblock = DELAYSTARTBLOCK; | |
3965 | return; | |
63be04eb | 3966 | } |
a2ceac1f DC |
3967 | |
3968 | if (obno > *bno) | |
3969 | *bno = obno; | |
3970 | ASSERT((*bno >= obno) || (n == 0)); | |
3971 | ASSERT(*bno < end); | |
3972 | mval->br_startoff = *bno; | |
3973 | if (isnullstartblock(got->br_startblock)) | |
3974 | mval->br_startblock = DELAYSTARTBLOCK; | |
2bd0ea18 | 3975 | else |
a2ceac1f DC |
3976 | mval->br_startblock = got->br_startblock + |
3977 | (*bno - got->br_startoff); | |
2bd0ea18 | 3978 | /* |
a2ceac1f DC |
3979 | * Return the minimum of what we got and what we asked for for |
3980 | * the length. We can use the len variable here because it is | |
3981 | * modified below and we could have been there before coming | |
3982 | * here if the first part of the allocation didn't overlap what | |
3983 | * was asked for. | |
2bd0ea18 | 3984 | */ |
a2ceac1f DC |
3985 | mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno, |
3986 | got->br_blockcount - (*bno - got->br_startoff)); | |
3987 | mval->br_state = got->br_state; | |
3988 | ASSERT(mval->br_blockcount <= len); | |
3989 | return; | |
3990 | } | |
56b2de80 | 3991 | |
a2ceac1f DC |
3992 | /* |
3993 | * Update and validate the extent map to return | |
3994 | */ | |
3995 | STATIC void | |
3996 | xfs_bmapi_update_map( | |
3997 | struct xfs_bmbt_irec **map, | |
3998 | xfs_fileoff_t *bno, | |
3999 | xfs_filblks_t *len, | |
4000 | xfs_fileoff_t obno, | |
4001 | xfs_fileoff_t end, | |
4002 | int *n, | |
4003 | int flags) | |
4004 | { | |
4005 | xfs_bmbt_irec_t *mval = *map; | |
4006 | ||
4007 | ASSERT((flags & XFS_BMAPI_ENTIRE) || | |
4008 | ((mval->br_startoff + mval->br_blockcount) <= end)); | |
4009 | ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) || | |
4010 | (mval->br_startoff < obno)); | |
4011 | ||
4012 | *bno = mval->br_startoff + mval->br_blockcount; | |
4013 | *len = end - *bno; | |
4014 | if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) { | |
4015 | /* update previous map with new information */ | |
4016 | ASSERT(mval->br_startblock == mval[-1].br_startblock); | |
4017 | ASSERT(mval->br_blockcount > mval[-1].br_blockcount); | |
4018 | ASSERT(mval->br_state == mval[-1].br_state); | |
4019 | mval[-1].br_blockcount = mval->br_blockcount; | |
4020 | mval[-1].br_state = mval->br_state; | |
4021 | } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK && | |
4022 | mval[-1].br_startblock != DELAYSTARTBLOCK && | |
4023 | mval[-1].br_startblock != HOLESTARTBLOCK && | |
4024 | mval->br_startblock == mval[-1].br_startblock + | |
4025 | mval[-1].br_blockcount && | |
4026 | ((flags & XFS_BMAPI_IGSTATE) || | |
4027 | mval[-1].br_state == mval->br_state)) { | |
4028 | ASSERT(mval->br_startoff == | |
4029 | mval[-1].br_startoff + mval[-1].br_blockcount); | |
4030 | mval[-1].br_blockcount += mval->br_blockcount; | |
4031 | } else if (*n > 0 && | |
4032 | mval->br_startblock == DELAYSTARTBLOCK && | |
4033 | mval[-1].br_startblock == DELAYSTARTBLOCK && | |
4034 | mval->br_startoff == | |
4035 | mval[-1].br_startoff + mval[-1].br_blockcount) { | |
4036 | mval[-1].br_blockcount += mval->br_blockcount; | |
4037 | mval[-1].br_state = mval->br_state; | |
4038 | } else if (!((*n == 0) && | |
4039 | ((mval->br_startoff + mval->br_blockcount) <= | |
4040 | obno))) { | |
4041 | mval++; | |
4042 | (*n)++; | |
4043 | } | |
4044 | *map = mval; | |
4045 | } | |
399ab595 | 4046 | |
a2ceac1f DC |
4047 | /* |
4048 | * Map file blocks to filesystem blocks without allocation. | |
4049 | */ | |
4050 | int | |
4051 | xfs_bmapi_read( | |
4052 | struct xfs_inode *ip, | |
4053 | xfs_fileoff_t bno, | |
4054 | xfs_filblks_t len, | |
4055 | struct xfs_bmbt_irec *mval, | |
4056 | int *nmap, | |
4057 | int flags) | |
4058 | { | |
4059 | struct xfs_mount *mp = ip->i_mount; | |
4060 | struct xfs_ifork *ifp; | |
4061 | struct xfs_bmbt_irec got; | |
a2ceac1f DC |
4062 | xfs_fileoff_t obno; |
4063 | xfs_fileoff_t end; | |
c2d73ed3 | 4064 | xfs_extnum_t idx; |
a2ceac1f | 4065 | int error; |
c2d73ed3 | 4066 | bool eof = false; |
a2ceac1f | 4067 | int n = 0; |
cb8a004a | 4068 | int whichfork = xfs_bmapi_whichfork(flags); |
399ab595 | 4069 | |
a2ceac1f DC |
4070 | ASSERT(*nmap >= 1); |
4071 | ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE| | |
cb8a004a | 4072 | XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK))); |
ff105f75 | 4073 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)); |
062998e3 | 4074 | |
a2ceac1f DC |
4075 | if (unlikely(XFS_TEST_ERROR( |
4076 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
4077 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
4078 | mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) { | |
4079 | XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 4080 | return -EFSCORRUPTED; |
a2ceac1f | 4081 | } |
062998e3 | 4082 | |
a2ceac1f | 4083 | if (XFS_FORCED_SHUTDOWN(mp)) |
12b53197 | 4084 | return -EIO; |
399ab595 | 4085 | |
79896434 | 4086 | XFS_STATS_INC(mp, xs_blk_mapr); |
a2ceac1f DC |
4087 | |
4088 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
4089 | ||
cb8a004a DW |
4090 | /* No CoW fork? Return a hole. */ |
4091 | if (whichfork == XFS_COW_FORK && !ifp) { | |
4092 | mval->br_startoff = bno; | |
4093 | mval->br_startblock = HOLESTARTBLOCK; | |
4094 | mval->br_blockcount = len; | |
4095 | mval->br_state = XFS_EXT_NORM; | |
4096 | *nmap = 1; | |
4097 | return 0; | |
4098 | } | |
4099 | ||
a2ceac1f DC |
4100 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
4101 | error = xfs_iread_extents(NULL, ip, whichfork); | |
4102 | if (error) | |
4103 | return error; | |
4104 | } | |
4105 | ||
c2d73ed3 CH |
4106 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &idx, &got)) |
4107 | eof = true; | |
a2ceac1f DC |
4108 | end = bno + len; |
4109 | obno = bno; | |
4110 | ||
4111 | while (bno < end && n < *nmap) { | |
4112 | /* Reading past eof, act as though there's a hole up to end. */ | |
4113 | if (eof) | |
4114 | got.br_startoff = end; | |
4115 | if (got.br_startoff > bno) { | |
4116 | /* Reading in a hole. */ | |
2bd0ea18 NS |
4117 | mval->br_startoff = bno; |
4118 | mval->br_startblock = HOLESTARTBLOCK; | |
4119 | mval->br_blockcount = | |
4120 | XFS_FILBLKS_MIN(len, got.br_startoff - bno); | |
4121 | mval->br_state = XFS_EXT_NORM; | |
4122 | bno += mval->br_blockcount; | |
4123 | len -= mval->br_blockcount; | |
4124 | mval++; | |
4125 | n++; | |
4126 | continue; | |
4127 | } | |
a2ceac1f DC |
4128 | |
4129 | /* set up the extent map to return. */ | |
4130 | xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags); | |
4131 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4132 | ||
4133 | /* If we're done, stop now. */ | |
4134 | if (bno >= end || n >= *nmap) | |
4135 | break; | |
4136 | ||
4137 | /* Else go on to the next record. */ | |
c2d73ed3 CH |
4138 | if (!xfs_iext_get_extent(ifp, ++idx, &got)) |
4139 | eof = true; | |
a2ceac1f DC |
4140 | } |
4141 | *nmap = n; | |
4142 | return 0; | |
4143 | } | |
4144 | ||
4488e421 | 4145 | int |
a2ceac1f DC |
4146 | xfs_bmapi_reserve_delalloc( |
4147 | struct xfs_inode *ip, | |
cc66acaf | 4148 | int whichfork, |
f7b1a8b1 | 4149 | xfs_fileoff_t off, |
a2ceac1f | 4150 | xfs_filblks_t len, |
f7b1a8b1 | 4151 | xfs_filblks_t prealloc, |
a2ceac1f | 4152 | struct xfs_bmbt_irec *got, |
a2ceac1f DC |
4153 | xfs_extnum_t *lastx, |
4154 | int eof) | |
4155 | { | |
4156 | struct xfs_mount *mp = ip->i_mount; | |
cc66acaf | 4157 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); |
a2ceac1f DC |
4158 | xfs_extlen_t alen; |
4159 | xfs_extlen_t indlen; | |
4160 | char rt = XFS_IS_REALTIME_INODE(ip); | |
4161 | xfs_extlen_t extsz; | |
4162 | int error; | |
f7b1a8b1 | 4163 | xfs_fileoff_t aoff = off; |
a2ceac1f | 4164 | |
f7b1a8b1 BF |
4165 | /* |
4166 | * Cap the alloc length. Keep track of prealloc so we know whether to | |
4167 | * tag the inode before we return. | |
4168 | */ | |
4169 | alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN); | |
a2ceac1f DC |
4170 | if (!eof) |
4171 | alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff); | |
f7b1a8b1 BF |
4172 | if (prealloc && alen >= len) |
4173 | prealloc = alen - len; | |
a2ceac1f DC |
4174 | |
4175 | /* Figure out the extent size, adjust alen */ | |
10460994 DW |
4176 | if (whichfork == XFS_COW_FORK) |
4177 | extsz = xfs_get_cowextsz_hint(ip); | |
4178 | else | |
4179 | extsz = xfs_get_extsz_hint(ip); | |
a2ceac1f | 4180 | if (extsz) { |
d41d2303 CH |
4181 | struct xfs_bmbt_irec prev; |
4182 | ||
4183 | if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev)) | |
4184 | prev.br_startoff = NULLFILEOFF; | |
4185 | ||
4186 | error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof, | |
a2ceac1f DC |
4187 | 1, 0, &aoff, &alen); |
4188 | ASSERT(!error); | |
4189 | } | |
4190 | ||
4191 | if (rt) | |
4192 | extsz = alen / mp->m_sb.sb_rextsize; | |
4193 | ||
4194 | /* | |
4195 | * Make a transaction-less quota reservation for delayed allocation | |
4196 | * blocks. This number gets adjusted later. We return if we haven't | |
4197 | * allocated blocks already inside this loop. | |
4198 | */ | |
4199 | error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0, | |
4200 | rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
4201 | if (error) | |
4202 | return error; | |
4203 | ||
4204 | /* | |
4205 | * Split changing sb for alen and indlen since they could be coming | |
4206 | * from different places. | |
4207 | */ | |
4208 | indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen); | |
4209 | ASSERT(indlen > 0); | |
4210 | ||
4211 | if (rt) { | |
19ebedcf | 4212 | error = xfs_mod_frextents(mp, -((int64_t)extsz)); |
a2ceac1f | 4213 | } else { |
19ebedcf | 4214 | error = xfs_mod_fdblocks(mp, -((int64_t)alen), false); |
a2ceac1f DC |
4215 | } |
4216 | ||
4217 | if (error) | |
4218 | goto out_unreserve_quota; | |
4219 | ||
19ebedcf | 4220 | error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false); |
a2ceac1f DC |
4221 | if (error) |
4222 | goto out_unreserve_blocks; | |
4223 | ||
4224 | ||
4225 | ip->i_delayed_blks += alen; | |
4226 | ||
4227 | got->br_startoff = aoff; | |
4228 | got->br_startblock = nullstartblock(indlen); | |
4229 | got->br_blockcount = alen; | |
4230 | got->br_state = XFS_EXT_NORM; | |
cc66acaf | 4231 | xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got); |
a2ceac1f DC |
4232 | |
4233 | /* | |
4234 | * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay | |
4235 | * might have merged it into one of the neighbouring ones. | |
4236 | */ | |
4237 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got); | |
4238 | ||
f7b1a8b1 BF |
4239 | /* |
4240 | * Tag the inode if blocks were preallocated. Note that COW fork | |
4241 | * preallocation can occur at the start or end of the extent, even when | |
4242 | * prealloc == 0, so we must also check the aligned offset and length. | |
4243 | */ | |
4244 | if (whichfork == XFS_DATA_FORK && prealloc) | |
4245 | xfs_inode_set_eofblocks_tag(ip); | |
4246 | if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len)) | |
4247 | xfs_inode_set_cowblocks_tag(ip); | |
4248 | ||
a2ceac1f DC |
4249 | ASSERT(got->br_startoff <= aoff); |
4250 | ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen); | |
4251 | ASSERT(isnullstartblock(got->br_startblock)); | |
4252 | ASSERT(got->br_state == XFS_EXT_NORM); | |
4253 | return 0; | |
4254 | ||
4255 | out_unreserve_blocks: | |
4256 | if (rt) | |
19ebedcf | 4257 | xfs_mod_frextents(mp, extsz); |
a2ceac1f | 4258 | else |
19ebedcf | 4259 | xfs_mod_fdblocks(mp, alen, false); |
a2ceac1f DC |
4260 | out_unreserve_quota: |
4261 | if (XFS_IS_QUOTA_ON(mp)) | |
4262 | xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ? | |
4263 | XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
4264 | return error; | |
4265 | } | |
4266 | ||
ff105f75 DC |
4267 | static int |
4268 | xfs_bmapi_allocate( | |
a2ceac1f DC |
4269 | struct xfs_bmalloca *bma) |
4270 | { | |
4271 | struct xfs_mount *mp = bma->ip->i_mount; | |
1277a5e0 | 4272 | int whichfork = xfs_bmapi_whichfork(bma->flags); |
a2ceac1f DC |
4273 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4274 | int tmp_logflags = 0; | |
4275 | int error; | |
4276 | ||
4277 | ASSERT(bma->length > 0); | |
4278 | ||
4279 | /* | |
4280 | * For the wasdelay case, we could also just allocate the stuff asked | |
4281 | * for in this bmap call but that wouldn't be as good. | |
4282 | */ | |
4283 | if (bma->wasdel) { | |
4284 | bma->length = (xfs_extlen_t)bma->got.br_blockcount; | |
4285 | bma->offset = bma->got.br_startoff; | |
f5f0bea9 | 4286 | if (bma->idx) { |
a2ceac1f DC |
4287 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), |
4288 | &bma->prev); | |
4289 | } | |
4290 | } else { | |
4291 | bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN); | |
4292 | if (!bma->eof) | |
4293 | bma->length = XFS_FILBLKS_MIN(bma->length, | |
4294 | bma->got.br_startoff - bma->offset); | |
4295 | } | |
4296 | ||
4297 | /* | |
1fccd5c8 DC |
4298 | * Set the data type being allocated. For the data fork, the first data |
4299 | * in the file is treated differently to all other allocations. For the | |
4300 | * attribute fork, we only need to ensure the allocated range is not on | |
4301 | * the busy list. | |
a2ceac1f DC |
4302 | */ |
4303 | if (!(bma->flags & XFS_BMAPI_METADATA)) { | |
1fccd5c8 DC |
4304 | bma->datatype = XFS_ALLOC_NOBUSY; |
4305 | if (whichfork == XFS_DATA_FORK) { | |
4306 | if (bma->offset == 0) | |
4307 | bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA; | |
4308 | else | |
4309 | bma->datatype |= XFS_ALLOC_USERDATA; | |
4310 | } | |
9542ae13 | 4311 | if (bma->flags & XFS_BMAPI_ZERO) |
1fccd5c8 | 4312 | bma->datatype |= XFS_ALLOC_USERDATA_ZERO; |
a2ceac1f DC |
4313 | } |
4314 | ||
4315 | bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1; | |
4316 | ||
4317 | /* | |
4318 | * Only want to do the alignment at the eof if it is userdata and | |
4319 | * allocation length is larger than a stripe unit. | |
4320 | */ | |
4321 | if (mp->m_dalign && bma->length >= mp->m_dalign && | |
4322 | !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) { | |
4323 | error = xfs_bmap_isaeof(bma, whichfork); | |
4324 | if (error) | |
4325 | return error; | |
4326 | } | |
4327 | ||
a2ceac1f DC |
4328 | error = xfs_bmap_alloc(bma); |
4329 | if (error) | |
4330 | return error; | |
4331 | ||
f33cea1a | 4332 | if (bma->dfops->dop_low) |
a2ceac1f DC |
4333 | bma->minleft = 0; |
4334 | if (bma->cur) | |
4335 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
4336 | if (bma->blkno == NULLFSBLOCK) | |
4337 | return 0; | |
4338 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4339 | bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork); | |
4340 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
f33cea1a | 4341 | bma->cur->bc_private.b.dfops = bma->dfops; |
a2ceac1f DC |
4342 | } |
4343 | /* | |
4344 | * Bump the number of extents we've allocated | |
4345 | * in this call. | |
4346 | */ | |
4347 | bma->nallocs++; | |
4348 | ||
4349 | if (bma->cur) | |
4350 | bma->cur->bc_private.b.flags = | |
4351 | bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0; | |
4352 | ||
4353 | bma->got.br_startoff = bma->offset; | |
4354 | bma->got.br_startblock = bma->blkno; | |
4355 | bma->got.br_blockcount = bma->length; | |
4356 | bma->got.br_state = XFS_EXT_NORM; | |
4357 | ||
4358 | /* | |
4359 | * A wasdelay extent has been initialized, so shouldn't be flagged | |
4360 | * as unwritten. | |
4361 | */ | |
4362 | if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) && | |
4363 | xfs_sb_version_hasextflgbit(&mp->m_sb)) | |
4364 | bma->got.br_state = XFS_EXT_UNWRITTEN; | |
4365 | ||
4366 | if (bma->wasdel) | |
1277a5e0 | 4367 | error = xfs_bmap_add_extent_delay_real(bma, whichfork); |
a2ceac1f DC |
4368 | else |
4369 | error = xfs_bmap_add_extent_hole_real(bma, whichfork); | |
4370 | ||
4371 | bma->logflags |= tmp_logflags; | |
4372 | if (error) | |
4373 | return error; | |
4374 | ||
4375 | /* | |
4376 | * Update our extent pointer, given that xfs_bmap_add_extent_delay_real | |
4377 | * or xfs_bmap_add_extent_hole_real might have merged it into one of | |
4378 | * the neighbouring ones. | |
4379 | */ | |
4380 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got); | |
4381 | ||
4382 | ASSERT(bma->got.br_startoff <= bma->offset); | |
4383 | ASSERT(bma->got.br_startoff + bma->got.br_blockcount >= | |
4384 | bma->offset + bma->length); | |
4385 | ASSERT(bma->got.br_state == XFS_EXT_NORM || | |
4386 | bma->got.br_state == XFS_EXT_UNWRITTEN); | |
4387 | return 0; | |
4388 | } | |
4389 | ||
a2ceac1f DC |
4390 | STATIC int |
4391 | xfs_bmapi_convert_unwritten( | |
4392 | struct xfs_bmalloca *bma, | |
4393 | struct xfs_bmbt_irec *mval, | |
4394 | xfs_filblks_t len, | |
4395 | int flags) | |
4396 | { | |
cb8a004a | 4397 | int whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4398 | struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork); |
4399 | int tmp_logflags = 0; | |
4400 | int error; | |
4401 | ||
4402 | /* check if we need to do unwritten->real conversion */ | |
4403 | if (mval->br_state == XFS_EXT_UNWRITTEN && | |
4404 | (flags & XFS_BMAPI_PREALLOC)) | |
4405 | return 0; | |
4406 | ||
4407 | /* check if we need to do real->unwritten conversion */ | |
4408 | if (mval->br_state == XFS_EXT_NORM && | |
4409 | (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) != | |
4410 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) | |
4411 | return 0; | |
4412 | ||
cb8a004a DW |
4413 | ASSERT(whichfork != XFS_COW_FORK); |
4414 | ||
a2ceac1f DC |
4415 | /* |
4416 | * Modify (by adding) the state flag, if writing. | |
4417 | */ | |
4418 | ASSERT(mval->br_blockcount <= len); | |
4419 | if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) { | |
4420 | bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp, | |
4421 | bma->ip, whichfork); | |
4422 | bma->cur->bc_private.b.firstblock = *bma->firstblock; | |
f33cea1a | 4423 | bma->cur->bc_private.b.dfops = bma->dfops; |
a2ceac1f DC |
4424 | } |
4425 | mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN) | |
4426 | ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN; | |
4427 | ||
9542ae13 DC |
4428 | /* |
4429 | * Before insertion into the bmbt, zero the range being converted | |
4430 | * if required. | |
4431 | */ | |
4432 | if (flags & XFS_BMAPI_ZERO) { | |
4433 | error = xfs_zero_extent(bma->ip, mval->br_startblock, | |
4434 | mval->br_blockcount); | |
4435 | if (error) | |
4436 | return error; | |
4437 | } | |
4438 | ||
a2ceac1f | 4439 | error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx, |
f33cea1a | 4440 | &bma->cur, mval, bma->firstblock, bma->dfops, |
a2ceac1f | 4441 | &tmp_logflags); |
23fc058d BF |
4442 | /* |
4443 | * Log the inode core unconditionally in the unwritten extent conversion | |
4444 | * path because the conversion might not have done so (e.g., if the | |
4445 | * extent count hasn't changed). We need to make sure the inode is dirty | |
4446 | * in the transaction for the sake of fsync(), even if nothing has | |
4447 | * changed, because fsync() will not force the log for this transaction | |
4448 | * unless it sees the inode pinned. | |
4449 | */ | |
4450 | bma->logflags |= tmp_logflags | XFS_ILOG_CORE; | |
a2ceac1f DC |
4451 | if (error) |
4452 | return error; | |
4453 | ||
4454 | /* | |
4455 | * Update our extent pointer, given that | |
4456 | * xfs_bmap_add_extent_unwritten_real might have merged it into one | |
4457 | * of the neighbouring ones. | |
4458 | */ | |
4459 | xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got); | |
4460 | ||
4461 | /* | |
4462 | * We may have combined previously unwritten space with written space, | |
4463 | * so generate another request. | |
4464 | */ | |
4465 | if (mval->br_blockcount < len) | |
12b53197 | 4466 | return -EAGAIN; |
a2ceac1f DC |
4467 | return 0; |
4468 | } | |
4469 | ||
4470 | /* | |
4471 | * Map file blocks to filesystem blocks, and allocate blocks or convert the | |
4472 | * extent state if necessary. Details behaviour is controlled by the flags | |
4473 | * parameter. Only allocates blocks from a single allocation group, to avoid | |
4474 | * locking problems. | |
4475 | * | |
4476 | * The returned value in "firstblock" from the first call in a transaction | |
4477 | * must be remembered and presented to subsequent calls in "firstblock". | |
4478 | * An upper bound for the number of blocks to be allocated is supplied to | |
4479 | * the first call in "total"; if no allocation group has that many free | |
4480 | * blocks then the call will fail (return NULLFSBLOCK in "firstblock"). | |
4481 | */ | |
4482 | int | |
4483 | xfs_bmapi_write( | |
4484 | struct xfs_trans *tp, /* transaction pointer */ | |
4485 | struct xfs_inode *ip, /* incore inode */ | |
4486 | xfs_fileoff_t bno, /* starting file offs. mapped */ | |
4487 | xfs_filblks_t len, /* length to map in file */ | |
4488 | int flags, /* XFS_BMAPI_... */ | |
4489 | xfs_fsblock_t *firstblock, /* first allocated block | |
4490 | controls a.g. for allocs */ | |
4491 | xfs_extlen_t total, /* total blocks needed */ | |
4492 | struct xfs_bmbt_irec *mval, /* output: map values */ | |
4493 | int *nmap, /* i/o: mval size/count */ | |
f33cea1a | 4494 | struct xfs_defer_ops *dfops) /* i/o: list extents to free */ |
a2ceac1f DC |
4495 | { |
4496 | struct xfs_mount *mp = ip->i_mount; | |
4497 | struct xfs_ifork *ifp; | |
389b3b07 | 4498 | struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */ |
a2ceac1f | 4499 | xfs_fileoff_t end; /* end of mapped file region */ |
82411945 | 4500 | bool eof = false; /* after the end of extents */ |
a2ceac1f DC |
4501 | int error; /* error return */ |
4502 | int n; /* current extent index */ | |
4503 | xfs_fileoff_t obno; /* old block number (offset) */ | |
4504 | int whichfork; /* data or attr fork */ | |
4505 | char inhole; /* current location is hole in file */ | |
4506 | char wasdelay; /* old extent was delayed */ | |
4507 | ||
4508 | #ifdef DEBUG | |
4509 | xfs_fileoff_t orig_bno; /* original block number value */ | |
4510 | int orig_flags; /* original flags arg value */ | |
4511 | xfs_filblks_t orig_len; /* original value of len arg */ | |
4512 | struct xfs_bmbt_irec *orig_mval; /* original value of mval */ | |
4513 | int orig_nmap; /* original value of *nmap */ | |
4514 | ||
4515 | orig_bno = bno; | |
4516 | orig_len = len; | |
4517 | orig_flags = flags; | |
4518 | orig_mval = mval; | |
4519 | orig_nmap = *nmap; | |
4520 | #endif | |
1277a5e0 | 4521 | whichfork = xfs_bmapi_whichfork(flags); |
a2ceac1f DC |
4522 | |
4523 | ASSERT(*nmap >= 1); | |
4524 | ASSERT(*nmap <= XFS_BMAP_MAX_NMAP); | |
4525 | ASSERT(!(flags & XFS_BMAPI_IGSTATE)); | |
4526 | ASSERT(tp != NULL); | |
4527 | ASSERT(len > 0); | |
3f17ed4b | 4528 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL); |
ff105f75 | 4529 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
34621a47 DW |
4530 | ASSERT(!(flags & XFS_BMAPI_REMAP) || whichfork == XFS_DATA_FORK); |
4531 | ASSERT(!(flags & XFS_BMAPI_PREALLOC) || !(flags & XFS_BMAPI_REMAP)); | |
4532 | ASSERT(!(flags & XFS_BMAPI_CONVERT) || !(flags & XFS_BMAPI_REMAP)); | |
1277a5e0 DW |
4533 | ASSERT(!(flags & XFS_BMAPI_PREALLOC) || whichfork != XFS_COW_FORK); |
4534 | ASSERT(!(flags & XFS_BMAPI_CONVERT) || whichfork != XFS_COW_FORK); | |
a2ceac1f | 4535 | |
9542ae13 DC |
4536 | /* zeroing is for currently only for data extents, not metadata */ |
4537 | ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) != | |
4538 | (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)); | |
4539 | /* | |
4540 | * we can allocate unwritten extents or pre-zero allocated blocks, | |
4541 | * but it makes no sense to do both at once. This would result in | |
4542 | * zeroing the unwritten extent twice, but it still being an | |
4543 | * unwritten extent.... | |
4544 | */ | |
4545 | ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) != | |
4546 | (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)); | |
4547 | ||
a2ceac1f DC |
4548 | if (unlikely(XFS_TEST_ERROR( |
4549 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
3f17ed4b | 4550 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), |
a2ceac1f DC |
4551 | mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) { |
4552 | XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 4553 | return -EFSCORRUPTED; |
a2ceac1f DC |
4554 | } |
4555 | ||
4556 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 4557 | return -EIO; |
a2ceac1f DC |
4558 | |
4559 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
4560 | ||
79896434 | 4561 | XFS_STATS_INC(mp, xs_blk_mapw); |
a2ceac1f | 4562 | |
a2ceac1f DC |
4563 | if (*firstblock == NULLFSBLOCK) { |
4564 | if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE) | |
4565 | bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1; | |
4566 | else | |
4567 | bma.minleft = 1; | |
4568 | } else { | |
4569 | bma.minleft = 0; | |
4570 | } | |
4571 | ||
4572 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
4573 | error = xfs_iread_extents(tp, ip, whichfork); | |
4574 | if (error) | |
4575 | goto error0; | |
4576 | } | |
4577 | ||
a2ceac1f DC |
4578 | n = 0; |
4579 | end = bno + len; | |
4580 | obno = bno; | |
4581 | ||
82411945 CH |
4582 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.idx, &bma.got)) |
4583 | eof = true; | |
4584 | if (!xfs_iext_get_extent(ifp, bma.idx - 1, &bma.prev)) | |
4585 | bma.prev.br_startoff = NULLFILEOFF; | |
a2ceac1f DC |
4586 | bma.tp = tp; |
4587 | bma.ip = ip; | |
4588 | bma.total = total; | |
1fccd5c8 | 4589 | bma.datatype = 0; |
f33cea1a | 4590 | bma.dfops = dfops; |
a2ceac1f DC |
4591 | bma.firstblock = firstblock; |
4592 | ||
4593 | while (bno < end && n < *nmap) { | |
4594 | inhole = eof || bma.got.br_startoff > bno; | |
4595 | wasdelay = !inhole && isnullstartblock(bma.got.br_startblock); | |
4596 | ||
34621a47 DW |
4597 | /* |
4598 | * Make sure we only reflink into a hole. | |
4599 | */ | |
4600 | if (flags & XFS_BMAPI_REMAP) | |
4601 | ASSERT(inhole); | |
1277a5e0 DW |
4602 | if (flags & XFS_BMAPI_COWFORK) |
4603 | ASSERT(!inhole); | |
34621a47 | 4604 | |
2bd0ea18 | 4605 | /* |
a2ceac1f DC |
4606 | * First, deal with the hole before the allocated space |
4607 | * that we found, if any. | |
2bd0ea18 | 4608 | */ |
a2ceac1f DC |
4609 | if (inhole || wasdelay) { |
4610 | bma.eof = eof; | |
4611 | bma.conv = !!(flags & XFS_BMAPI_CONVERT); | |
4612 | bma.wasdel = wasdelay; | |
4613 | bma.offset = bno; | |
4614 | bma.flags = flags; | |
4615 | ||
2bd0ea18 | 4616 | /* |
a2ceac1f DC |
4617 | * There's a 32/64 bit type mismatch between the |
4618 | * allocation length request (which can be 64 bits in | |
4619 | * length) and the bma length request, which is | |
4620 | * xfs_extlen_t and therefore 32 bits. Hence we have to | |
4621 | * check for 32-bit overflows and handle them here. | |
2bd0ea18 | 4622 | */ |
a2ceac1f DC |
4623 | if (len > (xfs_filblks_t)MAXEXTLEN) |
4624 | bma.length = MAXEXTLEN; | |
4625 | else | |
4626 | bma.length = len; | |
4627 | ||
4628 | ASSERT(len > 0); | |
4629 | ASSERT(bma.length > 0); | |
4630 | error = xfs_bmapi_allocate(&bma); | |
2bd0ea18 NS |
4631 | if (error) |
4632 | goto error0; | |
a2ceac1f DC |
4633 | if (bma.blkno == NULLFSBLOCK) |
4634 | break; | |
10e65503 DW |
4635 | |
4636 | /* | |
4637 | * If this is a CoW allocation, record the data in | |
4638 | * the refcount btree for orphan recovery. | |
4639 | */ | |
4640 | if (whichfork == XFS_COW_FORK) { | |
4641 | error = xfs_refcount_alloc_cow_extent(mp, dfops, | |
4642 | bma.blkno, bma.length); | |
4643 | if (error) | |
4644 | goto error0; | |
4645 | } | |
2bd0ea18 NS |
4646 | } |
4647 | ||
a2ceac1f DC |
4648 | /* Deal with the allocated space we found. */ |
4649 | xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno, | |
4650 | end, n, flags); | |
4651 | ||
4652 | /* Execute unwritten extent conversion if necessary */ | |
4653 | error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags); | |
12b53197 | 4654 | if (error == -EAGAIN) |
a2ceac1f DC |
4655 | continue; |
4656 | if (error) | |
4657 | goto error0; | |
4658 | ||
4659 | /* update the extent map to return */ | |
4660 | xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags); | |
4661 | ||
2bd0ea18 NS |
4662 | /* |
4663 | * If we're done, stop now. Stop when we've allocated | |
4664 | * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise | |
4665 | * the transaction may get too big. | |
4666 | */ | |
a2ceac1f | 4667 | if (bno >= end || n >= *nmap || bma.nallocs >= *nmap) |
2bd0ea18 | 4668 | break; |
a2ceac1f DC |
4669 | |
4670 | /* Else go on to the next record. */ | |
4671 | bma.prev = bma.got; | |
82411945 CH |
4672 | if (!xfs_iext_get_extent(ifp, ++bma.idx, &bma.got)) |
4673 | eof = true; | |
2bd0ea18 | 4674 | } |
2bd0ea18 | 4675 | *nmap = n; |
a2ceac1f | 4676 | |
2bd0ea18 NS |
4677 | /* |
4678 | * Transform from btree to extents, give it cur. | |
4679 | */ | |
a2ceac1f DC |
4680 | if (xfs_bmap_wants_extents(ip, whichfork)) { |
4681 | int tmp_logflags = 0; | |
4682 | ||
4683 | ASSERT(bma.cur); | |
4684 | error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, | |
4ca431fc | 4685 | &tmp_logflags, whichfork); |
a2ceac1f | 4686 | bma.logflags |= tmp_logflags; |
2bd0ea18 NS |
4687 | if (error) |
4688 | goto error0; | |
4689 | } | |
a2ceac1f | 4690 | |
2bd0ea18 | 4691 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE || |
a2ceac1f DC |
4692 | XFS_IFORK_NEXTENTS(ip, whichfork) > |
4693 | XFS_IFORK_MAXEXT(ip, whichfork)); | |
2bd0ea18 | 4694 | error = 0; |
2bd0ea18 NS |
4695 | error0: |
4696 | /* | |
4697 | * Log everything. Do this after conversion, there's no point in | |
5e656dbb | 4698 | * logging the extent records if we've converted to btree format. |
2bd0ea18 | 4699 | */ |
a2ceac1f | 4700 | if ((bma.logflags & xfs_ilog_fext(whichfork)) && |
2bd0ea18 | 4701 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) |
a2ceac1f DC |
4702 | bma.logflags &= ~xfs_ilog_fext(whichfork); |
4703 | else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) && | |
2bd0ea18 | 4704 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) |
a2ceac1f | 4705 | bma.logflags &= ~xfs_ilog_fbroot(whichfork); |
2bd0ea18 NS |
4706 | /* |
4707 | * Log whatever the flags say, even if error. Otherwise we might miss | |
4708 | * detecting a case where the data is changed, there's an error, | |
4709 | * and it's not logged so we don't shutdown when we should. | |
4710 | */ | |
a2ceac1f DC |
4711 | if (bma.logflags) |
4712 | xfs_trans_log_inode(tp, ip, bma.logflags); | |
4713 | ||
4714 | if (bma.cur) { | |
2bd0ea18 NS |
4715 | if (!error) { |
4716 | ASSERT(*firstblock == NULLFSBLOCK || | |
eae766ca NS |
4717 | XFS_FSB_TO_AGNO(mp, *firstblock) == |
4718 | XFS_FSB_TO_AGNO(mp, | |
a2ceac1f | 4719 | bma.cur->bc_private.b.firstblock) || |
f33cea1a | 4720 | (dfops->dop_low && |
eae766ca NS |
4721 | XFS_FSB_TO_AGNO(mp, *firstblock) < |
4722 | XFS_FSB_TO_AGNO(mp, | |
a2ceac1f DC |
4723 | bma.cur->bc_private.b.firstblock))); |
4724 | *firstblock = bma.cur->bc_private.b.firstblock; | |
2bd0ea18 | 4725 | } |
a2ceac1f | 4726 | xfs_btree_del_cursor(bma.cur, |
2bd0ea18 NS |
4727 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); |
4728 | } | |
4729 | if (!error) | |
4730 | xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval, | |
4731 | orig_nmap, *nmap); | |
4732 | return error; | |
4733 | } | |
4734 | ||
01d1b786 BF |
4735 | /* |
4736 | * When a delalloc extent is split (e.g., due to a hole punch), the original | |
4737 | * indlen reservation must be shared across the two new extents that are left | |
4738 | * behind. | |
4739 | * | |
4740 | * Given the original reservation and the worst case indlen for the two new | |
4741 | * extents (as calculated by xfs_bmap_worst_indlen()), split the original | |
731ccdf9 BF |
4742 | * reservation fairly across the two new extents. If necessary, steal available |
4743 | * blocks from a deleted extent to make up a reservation deficiency (e.g., if | |
4744 | * ores == 1). The number of stolen blocks is returned. The availability and | |
4745 | * subsequent accounting of stolen blocks is the responsibility of the caller. | |
01d1b786 | 4746 | */ |
731ccdf9 | 4747 | static xfs_filblks_t |
01d1b786 BF |
4748 | xfs_bmap_split_indlen( |
4749 | xfs_filblks_t ores, /* original res. */ | |
4750 | xfs_filblks_t *indlen1, /* ext1 worst indlen */ | |
731ccdf9 BF |
4751 | xfs_filblks_t *indlen2, /* ext2 worst indlen */ |
4752 | xfs_filblks_t avail) /* stealable blocks */ | |
01d1b786 BF |
4753 | { |
4754 | xfs_filblks_t len1 = *indlen1; | |
4755 | xfs_filblks_t len2 = *indlen2; | |
4756 | xfs_filblks_t nres = len1 + len2; /* new total res. */ | |
731ccdf9 BF |
4757 | xfs_filblks_t stolen = 0; |
4758 | ||
4759 | /* | |
4760 | * Steal as many blocks as we can to try and satisfy the worst case | |
4761 | * indlen for both new extents. | |
4762 | */ | |
4763 | while (nres > ores && avail) { | |
4764 | nres--; | |
4765 | avail--; | |
4766 | stolen++; | |
4767 | } | |
01d1b786 BF |
4768 | |
4769 | /* | |
731ccdf9 BF |
4770 | * The only blocks available are those reserved for the original |
4771 | * extent and what we can steal from the extent being removed. | |
4772 | * If this still isn't enough to satisfy the combined | |
4773 | * requirements for the two new extents, skim blocks off of each | |
4774 | * of the new reservations until they match what is available. | |
01d1b786 BF |
4775 | */ |
4776 | while (nres > ores) { | |
4777 | if (len1) { | |
4778 | len1--; | |
4779 | nres--; | |
4780 | } | |
4781 | if (nres == ores) | |
4782 | break; | |
4783 | if (len2) { | |
4784 | len2--; | |
4785 | nres--; | |
4786 | } | |
4787 | } | |
4788 | ||
4789 | *indlen1 = len1; | |
4790 | *indlen2 = len2; | |
731ccdf9 BF |
4791 | |
4792 | return stolen; | |
01d1b786 BF |
4793 | } |
4794 | ||
ece930fa CH |
4795 | int |
4796 | xfs_bmap_del_extent_delay( | |
4797 | struct xfs_inode *ip, | |
4798 | int whichfork, | |
4799 | xfs_extnum_t *idx, | |
4800 | struct xfs_bmbt_irec *got, | |
4801 | struct xfs_bmbt_irec *del) | |
4802 | { | |
4803 | struct xfs_mount *mp = ip->i_mount; | |
4804 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
4805 | struct xfs_bmbt_irec new; | |
4806 | int64_t da_old, da_new, da_diff = 0; | |
4807 | xfs_fileoff_t del_endoff, got_endoff; | |
4808 | xfs_filblks_t got_indlen, new_indlen, stolen; | |
4809 | int error = 0, state = 0; | |
4810 | bool isrt; | |
4811 | ||
4812 | XFS_STATS_INC(mp, xs_del_exlist); | |
4813 | ||
4814 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); | |
4815 | del_endoff = del->br_startoff + del->br_blockcount; | |
4816 | got_endoff = got->br_startoff + got->br_blockcount; | |
4817 | da_old = startblockval(got->br_startblock); | |
4818 | da_new = 0; | |
4819 | ||
4820 | ASSERT(*idx >= 0); | |
d09d4e5f | 4821 | ASSERT(*idx <= xfs_iext_count(ifp)); |
ece930fa CH |
4822 | ASSERT(del->br_blockcount > 0); |
4823 | ASSERT(got->br_startoff <= del->br_startoff); | |
4824 | ASSERT(got_endoff >= del_endoff); | |
4825 | ||
4826 | if (isrt) { | |
4827 | int64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount); | |
4828 | ||
4829 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
4830 | xfs_mod_frextents(mp, rtexts); | |
4831 | } | |
4832 | ||
4833 | /* | |
4834 | * Update the inode delalloc counter now and wait to update the | |
4835 | * sb counters as we might have to borrow some blocks for the | |
4836 | * indirect block accounting. | |
4837 | */ | |
4838 | error = xfs_trans_reserve_quota_nblks(NULL, ip, | |
4839 | -((long)del->br_blockcount), 0, | |
4840 | isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS); | |
6921638c DW |
4841 | if (error) |
4842 | return error; | |
ece930fa CH |
4843 | ip->i_delayed_blks -= del->br_blockcount; |
4844 | ||
4845 | if (whichfork == XFS_COW_FORK) | |
4846 | state |= BMAP_COWFORK; | |
4847 | ||
4848 | if (got->br_startoff == del->br_startoff) | |
4849 | state |= BMAP_LEFT_CONTIG; | |
4850 | if (got_endoff == del_endoff) | |
4851 | state |= BMAP_RIGHT_CONTIG; | |
4852 | ||
4853 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
4854 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
4855 | /* | |
4856 | * Matches the whole extent. Delete the entry. | |
4857 | */ | |
4858 | xfs_iext_remove(ip, *idx, 1, state); | |
4859 | --*idx; | |
4860 | break; | |
4861 | case BMAP_LEFT_CONTIG: | |
4862 | /* | |
4863 | * Deleting the first part of the extent. | |
4864 | */ | |
4865 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4866 | got->br_startoff = del_endoff; | |
4867 | got->br_blockcount -= del->br_blockcount; | |
4868 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4869 | got->br_blockcount), da_old); | |
4870 | got->br_startblock = nullstartblock((int)da_new); | |
4871 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4872 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
4873 | break; | |
4874 | case BMAP_RIGHT_CONTIG: | |
4875 | /* | |
4876 | * Deleting the last part of the extent. | |
4877 | */ | |
4878 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4879 | got->br_blockcount = got->br_blockcount - del->br_blockcount; | |
4880 | da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, | |
4881 | got->br_blockcount), da_old); | |
4882 | got->br_startblock = nullstartblock((int)da_new); | |
4883 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4884 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
4885 | break; | |
4886 | case 0: | |
4887 | /* | |
4888 | * Deleting the middle of the extent. | |
4889 | * | |
4890 | * Distribute the original indlen reservation across the two new | |
4891 | * extents. Steal blocks from the deleted extent if necessary. | |
4892 | * Stealing blocks simply fudges the fdblocks accounting below. | |
4893 | * Warn if either of the new indlen reservations is zero as this | |
4894 | * can lead to delalloc problems. | |
4895 | */ | |
4896 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4897 | ||
4898 | got->br_blockcount = del->br_startoff - got->br_startoff; | |
4899 | got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount); | |
4900 | ||
4901 | new.br_blockcount = got_endoff - del_endoff; | |
4902 | new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
4903 | ||
4904 | WARN_ON_ONCE(!got_indlen || !new_indlen); | |
4905 | stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen, | |
4906 | del->br_blockcount); | |
4907 | ||
4908 | got->br_startblock = nullstartblock((int)got_indlen); | |
4909 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4910 | trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_); | |
4911 | ||
4912 | new.br_startoff = del_endoff; | |
4913 | new.br_state = got->br_state; | |
4914 | new.br_startblock = nullstartblock((int)new_indlen); | |
4915 | ||
4916 | ++*idx; | |
4917 | xfs_iext_insert(ip, *idx, 1, &new, state); | |
4918 | ||
4919 | da_new = got_indlen + new_indlen - stolen; | |
4920 | del->br_blockcount -= stolen; | |
4921 | break; | |
4922 | } | |
4923 | ||
4924 | ASSERT(da_old >= da_new); | |
4925 | da_diff = da_old - da_new; | |
4926 | if (!isrt) | |
4927 | da_diff += del->br_blockcount; | |
4928 | if (da_diff) | |
4929 | xfs_mod_fdblocks(mp, da_diff, false); | |
4930 | return error; | |
4931 | } | |
4932 | ||
4933 | void | |
4934 | xfs_bmap_del_extent_cow( | |
4935 | struct xfs_inode *ip, | |
4936 | xfs_extnum_t *idx, | |
4937 | struct xfs_bmbt_irec *got, | |
4938 | struct xfs_bmbt_irec *del) | |
4939 | { | |
4940 | struct xfs_mount *mp = ip->i_mount; | |
4941 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); | |
4942 | struct xfs_bmbt_irec new; | |
4943 | xfs_fileoff_t del_endoff, got_endoff; | |
4944 | int state = BMAP_COWFORK; | |
4945 | ||
4946 | XFS_STATS_INC(mp, xs_del_exlist); | |
4947 | ||
4948 | del_endoff = del->br_startoff + del->br_blockcount; | |
4949 | got_endoff = got->br_startoff + got->br_blockcount; | |
4950 | ||
4951 | ASSERT(*idx >= 0); | |
d09d4e5f | 4952 | ASSERT(*idx <= xfs_iext_count(ifp)); |
ece930fa CH |
4953 | ASSERT(del->br_blockcount > 0); |
4954 | ASSERT(got->br_startoff <= del->br_startoff); | |
4955 | ASSERT(got_endoff >= del_endoff); | |
4956 | ASSERT(!isnullstartblock(got->br_startblock)); | |
4957 | ||
4958 | if (got->br_startoff == del->br_startoff) | |
4959 | state |= BMAP_LEFT_CONTIG; | |
4960 | if (got_endoff == del_endoff) | |
4961 | state |= BMAP_RIGHT_CONTIG; | |
4962 | ||
4963 | switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) { | |
4964 | case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG: | |
4965 | /* | |
4966 | * Matches the whole extent. Delete the entry. | |
4967 | */ | |
4968 | xfs_iext_remove(ip, *idx, 1, state); | |
4969 | --*idx; | |
4970 | break; | |
4971 | case BMAP_LEFT_CONTIG: | |
4972 | /* | |
4973 | * Deleting the first part of the extent. | |
4974 | */ | |
4975 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4976 | got->br_startoff = del_endoff; | |
4977 | got->br_blockcount -= del->br_blockcount; | |
4978 | got->br_startblock = del->br_startblock + del->br_blockcount; | |
4979 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4980 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
4981 | break; | |
4982 | case BMAP_RIGHT_CONTIG: | |
4983 | /* | |
4984 | * Deleting the last part of the extent. | |
4985 | */ | |
4986 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4987 | got->br_blockcount -= del->br_blockcount; | |
4988 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4989 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
4990 | break; | |
4991 | case 0: | |
4992 | /* | |
4993 | * Deleting the middle of the extent. | |
4994 | */ | |
4995 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
4996 | got->br_blockcount = del->br_startoff - got->br_startoff; | |
4997 | xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got); | |
4998 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
4999 | ||
5000 | new.br_startoff = del_endoff; | |
5001 | new.br_blockcount = got_endoff - del_endoff; | |
5002 | new.br_state = got->br_state; | |
5003 | new.br_startblock = del->br_startblock + del->br_blockcount; | |
5004 | ||
5005 | ++*idx; | |
5006 | xfs_iext_insert(ip, *idx, 1, &new, state); | |
5007 | break; | |
5008 | } | |
5009 | } | |
5010 | ||
2bd0ea18 | 5011 | /* |
49f693fa DC |
5012 | * Called by xfs_bmapi to update file extent records and the btree |
5013 | * after removing space (or undoing a delayed allocation). | |
2bd0ea18 | 5014 | */ |
49f693fa DC |
5015 | STATIC int /* error */ |
5016 | xfs_bmap_del_extent( | |
5017 | xfs_inode_t *ip, /* incore inode pointer */ | |
5018 | xfs_trans_t *tp, /* current transaction pointer */ | |
5019 | xfs_extnum_t *idx, /* extent number to update/delete */ | |
f33cea1a | 5020 | struct xfs_defer_ops *dfops, /* list of extents to be freed */ |
49f693fa DC |
5021 | xfs_btree_cur_t *cur, /* if null, not a btree */ |
5022 | xfs_bmbt_irec_t *del, /* data to remove from extents */ | |
5023 | int *logflagsp, /* inode logging flags */ | |
36b16da8 DW |
5024 | int whichfork, /* data or attr fork */ |
5025 | int bflags) /* bmapi flags */ | |
2bd0ea18 | 5026 | { |
49f693fa DC |
5027 | xfs_filblks_t da_new; /* new delay-alloc indirect blocks */ |
5028 | xfs_filblks_t da_old; /* old delay-alloc indirect blocks */ | |
5029 | xfs_fsblock_t del_endblock=0; /* first block past del */ | |
5030 | xfs_fileoff_t del_endoff; /* first offset past del */ | |
5031 | int delay; /* current block is delayed allocated */ | |
5032 | int do_fx; /* free extent at end of routine */ | |
5033 | xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */ | |
5034 | int error; /* error return value */ | |
5035 | int flags; /* inode logging flags */ | |
5036 | xfs_bmbt_irec_t got; /* current extent entry */ | |
5037 | xfs_fileoff_t got_endoff; /* first offset past got */ | |
5038 | int i; /* temp state */ | |
5039 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
5040 | xfs_mount_t *mp; /* mount structure */ | |
5041 | xfs_filblks_t nblks; /* quota/sb block count */ | |
5042 | xfs_bmbt_irec_t new; /* new record to be inserted */ | |
5043 | /* REFERENCED */ | |
5044 | uint qfield; /* quota field to update */ | |
5045 | xfs_filblks_t temp; /* for indirect length calculations */ | |
5046 | xfs_filblks_t temp2; /* for indirect length calculations */ | |
5047 | int state = 0; | |
a2ceac1f | 5048 | |
79896434 BD |
5049 | mp = ip->i_mount; |
5050 | XFS_STATS_INC(mp, xs_del_exlist); | |
a2ceac1f | 5051 | |
49f693fa DC |
5052 | if (whichfork == XFS_ATTR_FORK) |
5053 | state |= BMAP_ATTRFORK; | |
cb8a004a DW |
5054 | else if (whichfork == XFS_COW_FORK) |
5055 | state |= BMAP_COWFORK; | |
56b2de80 | 5056 | |
49f693fa | 5057 | ifp = XFS_IFORK_PTR(ip, whichfork); |
d09d4e5f | 5058 | ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp))); |
49f693fa DC |
5059 | ASSERT(del->br_blockcount > 0); |
5060 | ep = xfs_iext_get_ext(ifp, *idx); | |
5061 | xfs_bmbt_get_all(ep, &got); | |
5062 | ASSERT(got.br_startoff <= del->br_startoff); | |
5063 | del_endoff = del->br_startoff + del->br_blockcount; | |
5064 | got_endoff = got.br_startoff + got.br_blockcount; | |
5065 | ASSERT(got_endoff >= del_endoff); | |
5066 | delay = isnullstartblock(got.br_startblock); | |
5067 | ASSERT(isnullstartblock(del->br_startblock) == delay); | |
5068 | flags = 0; | |
5069 | qfield = 0; | |
5070 | error = 0; | |
2bd0ea18 | 5071 | /* |
49f693fa | 5072 | * If deleting a real allocation, must free up the disk space. |
2bd0ea18 | 5073 | */ |
49f693fa DC |
5074 | if (!delay) { |
5075 | flags = XFS_ILOG_CORE; | |
a2ceac1f | 5076 | /* |
49f693fa | 5077 | * Realtime allocation. Free it and record di_nblocks update. |
a2ceac1f | 5078 | */ |
49f693fa DC |
5079 | if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) { |
5080 | xfs_fsblock_t bno; | |
5081 | xfs_filblks_t len; | |
a2ceac1f | 5082 | |
49f693fa DC |
5083 | ASSERT(do_mod(del->br_blockcount, |
5084 | mp->m_sb.sb_rextsize) == 0); | |
5085 | ASSERT(do_mod(del->br_startblock, | |
5086 | mp->m_sb.sb_rextsize) == 0); | |
5087 | bno = del->br_startblock; | |
5088 | len = del->br_blockcount; | |
5089 | do_div(bno, mp->m_sb.sb_rextsize); | |
5090 | do_div(len, mp->m_sb.sb_rextsize); | |
5091 | error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len); | |
5092 | if (error) | |
5093 | goto done; | |
5094 | do_fx = 0; | |
5095 | nblks = len * mp->m_sb.sb_rextsize; | |
5096 | qfield = XFS_TRANS_DQ_RTBCOUNT; | |
2bd0ea18 NS |
5097 | } |
5098 | /* | |
49f693fa | 5099 | * Ordinary allocation. |
2bd0ea18 | 5100 | */ |
49f693fa DC |
5101 | else { |
5102 | do_fx = 1; | |
5103 | nblks = del->br_blockcount; | |
5104 | qfield = XFS_TRANS_DQ_BCOUNT; | |
5105 | } | |
2bd0ea18 | 5106 | /* |
49f693fa | 5107 | * Set up del_endblock and cur for later. |
2bd0ea18 | 5108 | */ |
49f693fa DC |
5109 | del_endblock = del->br_startblock + del->br_blockcount; |
5110 | if (cur) { | |
5111 | if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff, | |
5112 | got.br_startblock, got.br_blockcount, | |
5113 | &i))) | |
5114 | goto done; | |
19ebedcf | 5115 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
2bd0ea18 | 5116 | } |
49f693fa DC |
5117 | da_old = da_new = 0; |
5118 | } else { | |
5119 | da_old = startblockval(got.br_startblock); | |
5120 | da_new = 0; | |
5121 | nblks = 0; | |
5122 | do_fx = 0; | |
5123 | } | |
85aec44f | 5124 | |
49f693fa DC |
5125 | /* |
5126 | * Set flag value to use in switch statement. | |
5127 | * Left-contig is 2, right-contig is 1. | |
5128 | */ | |
5129 | switch (((got.br_startoff == del->br_startoff) << 1) | | |
5130 | (got_endoff == del_endoff)) { | |
5131 | case 3: | |
5132 | /* | |
5133 | * Matches the whole extent. Delete the entry. | |
5134 | */ | |
d35205ba | 5135 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
49f693fa DC |
5136 | xfs_iext_remove(ip, *idx, 1, |
5137 | whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0); | |
5138 | --*idx; | |
5139 | if (delay) | |
5140 | break; | |
5141 | ||
5142 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5143 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5144 | flags |= XFS_ILOG_CORE; | |
5145 | if (!cur) { | |
5146 | flags |= xfs_ilog_fext(whichfork); | |
5147 | break; | |
2bd0ea18 | 5148 | } |
49f693fa DC |
5149 | if ((error = xfs_btree_delete(cur, &i))) |
5150 | goto done; | |
19ebedcf | 5151 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa | 5152 | break; |
399ab595 | 5153 | |
49f693fa | 5154 | case 2: |
2bd0ea18 | 5155 | /* |
49f693fa | 5156 | * Deleting the first part of the extent. |
2bd0ea18 | 5157 | */ |
49f693fa DC |
5158 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); |
5159 | xfs_bmbt_set_startoff(ep, del_endoff); | |
5160 | temp = got.br_blockcount - del->br_blockcount; | |
5161 | xfs_bmbt_set_blockcount(ep, temp); | |
5162 | if (delay) { | |
5163 | temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), | |
5164 | da_old); | |
5165 | xfs_bmbt_set_startblock(ep, nullstartblock((int)temp)); | |
5166 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5167 | da_new = temp; | |
5168 | break; | |
2bd0ea18 | 5169 | } |
49f693fa DC |
5170 | xfs_bmbt_set_startblock(ep, del_endblock); |
5171 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5172 | if (!cur) { | |
5173 | flags |= xfs_ilog_fext(whichfork); | |
5174 | break; | |
5175 | } | |
5176 | if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock, | |
5177 | got.br_blockcount - del->br_blockcount, | |
5178 | got.br_state))) | |
5179 | goto done; | |
5180 | break; | |
5181 | ||
5182 | case 1: | |
2bd0ea18 | 5183 | /* |
49f693fa | 5184 | * Deleting the last part of the extent. |
2bd0ea18 | 5185 | */ |
49f693fa DC |
5186 | temp = got.br_blockcount - del->br_blockcount; |
5187 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
5188 | xfs_bmbt_set_blockcount(ep, temp); | |
5189 | if (delay) { | |
5190 | temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp), | |
5191 | da_old); | |
5192 | xfs_bmbt_set_startblock(ep, nullstartblock((int)temp)); | |
5193 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5194 | da_new = temp; | |
5195 | break; | |
5196 | } | |
5197 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); | |
5198 | if (!cur) { | |
5199 | flags |= xfs_ilog_fext(whichfork); | |
5200 | break; | |
5201 | } | |
5202 | if ((error = xfs_bmbt_update(cur, got.br_startoff, | |
5203 | got.br_startblock, | |
5204 | got.br_blockcount - del->br_blockcount, | |
5205 | got.br_state))) | |
5206 | goto done; | |
5207 | break; | |
5208 | ||
5209 | case 0: | |
5210 | /* | |
5211 | * Deleting the middle of the extent. | |
5212 | */ | |
5213 | temp = del->br_startoff - got.br_startoff; | |
5214 | trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_); | |
5215 | xfs_bmbt_set_blockcount(ep, temp); | |
5216 | new.br_startoff = del_endoff; | |
5217 | temp2 = got_endoff - del_endoff; | |
5218 | new.br_blockcount = temp2; | |
5219 | new.br_state = got.br_state; | |
5220 | if (!delay) { | |
5221 | new.br_startblock = del_endblock; | |
5222 | flags |= XFS_ILOG_CORE; | |
5223 | if (cur) { | |
5224 | if ((error = xfs_bmbt_update(cur, | |
5225 | got.br_startoff, | |
5226 | got.br_startblock, temp, | |
5227 | got.br_state))) | |
5228 | goto done; | |
5229 | if ((error = xfs_btree_increment(cur, 0, &i))) | |
5230 | goto done; | |
5231 | cur->bc_rec.b = new; | |
5232 | error = xfs_btree_insert(cur, &i); | |
12b53197 | 5233 | if (error && error != -ENOSPC) |
49f693fa DC |
5234 | goto done; |
5235 | /* | |
5236 | * If get no-space back from btree insert, | |
5237 | * it tried a split, and we have a zero | |
5238 | * block reservation. | |
5239 | * Fix up our state and return the error. | |
5240 | */ | |
12b53197 | 5241 | if (error == -ENOSPC) { |
49f693fa DC |
5242 | /* |
5243 | * Reset the cursor, don't trust | |
5244 | * it after any insert operation. | |
5245 | */ | |
5246 | if ((error = xfs_bmbt_lookup_eq(cur, | |
5247 | got.br_startoff, | |
5248 | got.br_startblock, | |
5249 | temp, &i))) | |
5250 | goto done; | |
19ebedcf DC |
5251 | XFS_WANT_CORRUPTED_GOTO(mp, |
5252 | i == 1, done); | |
49f693fa DC |
5253 | /* |
5254 | * Update the btree record back | |
5255 | * to the original value. | |
5256 | */ | |
5257 | if ((error = xfs_bmbt_update(cur, | |
5258 | got.br_startoff, | |
5259 | got.br_startblock, | |
5260 | got.br_blockcount, | |
5261 | got.br_state))) | |
5262 | goto done; | |
5263 | /* | |
5264 | * Reset the extent record back | |
5265 | * to the original value. | |
5266 | */ | |
5267 | xfs_bmbt_set_blockcount(ep, | |
5268 | got.br_blockcount); | |
5269 | flags = 0; | |
12b53197 | 5270 | error = -ENOSPC; |
49f693fa DC |
5271 | goto done; |
5272 | } | |
19ebedcf | 5273 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done); |
49f693fa DC |
5274 | } else |
5275 | flags |= xfs_ilog_fext(whichfork); | |
5276 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5277 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
5278 | } else { | |
731ccdf9 | 5279 | xfs_filblks_t stolen; |
49f693fa | 5280 | ASSERT(whichfork == XFS_DATA_FORK); |
01d1b786 BF |
5281 | |
5282 | /* | |
5283 | * Distribute the original indlen reservation across the | |
731ccdf9 BF |
5284 | * two new extents. Steal blocks from the deleted extent |
5285 | * if necessary. Stealing blocks simply fudges the | |
5286 | * fdblocks accounting in xfs_bunmapi(). | |
01d1b786 BF |
5287 | */ |
5288 | temp = xfs_bmap_worst_indlen(ip, got.br_blockcount); | |
5289 | temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount); | |
731ccdf9 BF |
5290 | stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2, |
5291 | del->br_blockcount); | |
5292 | da_new = temp + temp2 - stolen; | |
5293 | del->br_blockcount -= stolen; | |
01d1b786 BF |
5294 | |
5295 | /* | |
731ccdf9 BF |
5296 | * Set the reservation for each extent. Warn if either |
5297 | * is zero as this can lead to delalloc problems. | |
01d1b786 | 5298 | */ |
731ccdf9 | 5299 | WARN_ON_ONCE(!temp || !temp2); |
49f693fa | 5300 | xfs_bmbt_set_startblock(ep, nullstartblock((int)temp)); |
49f693fa | 5301 | new.br_startblock = nullstartblock((int)temp2); |
2bd0ea18 | 5302 | } |
49f693fa DC |
5303 | trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_); |
5304 | xfs_iext_insert(ip, *idx + 1, 1, &new, state); | |
5305 | ++*idx; | |
5306 | break; | |
2bd0ea18 | 5307 | } |
d7f80320 DW |
5308 | |
5309 | /* remove reverse mapping */ | |
5310 | if (!delay) { | |
5311 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del); | |
5312 | if (error) | |
5313 | goto done; | |
5314 | } | |
5315 | ||
2bd0ea18 | 5316 | /* |
49f693fa | 5317 | * If we need to, add to list of extents to delete. |
2bd0ea18 | 5318 | */ |
36b16da8 | 5319 | if (do_fx && !(bflags & XFS_BMAPI_REMAP)) { |
cfe32f0d DW |
5320 | if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) { |
5321 | error = xfs_refcount_decrease_extent(mp, dfops, del); | |
5322 | if (error) | |
5323 | goto done; | |
5324 | } else | |
5325 | xfs_bmap_add_free(mp, dfops, del->br_startblock, | |
5326 | del->br_blockcount, NULL); | |
5327 | } | |
5328 | ||
2bd0ea18 | 5329 | /* |
49f693fa | 5330 | * Adjust inode # blocks in the file. |
2bd0ea18 | 5331 | */ |
49f693fa DC |
5332 | if (nblks) |
5333 | ip->i_d.di_nblocks -= nblks; | |
2bd0ea18 | 5334 | /* |
49f693fa | 5335 | * Adjust quota data. |
2bd0ea18 | 5336 | */ |
36b16da8 | 5337 | if (qfield && !(bflags & XFS_BMAPI_REMAP)) |
49f693fa DC |
5338 | xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks); |
5339 | ||
2bd0ea18 | 5340 | /* |
49f693fa DC |
5341 | * Account for change in delayed indirect blocks. |
5342 | * Nothing to do for disk quota accounting here. | |
2bd0ea18 | 5343 | */ |
49f693fa | 5344 | ASSERT(da_old >= da_new); |
19ebedcf DC |
5345 | if (da_old > da_new) |
5346 | xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false); | |
49f693fa DC |
5347 | done: |
5348 | *logflagsp = flags; | |
2bd0ea18 NS |
5349 | return error; |
5350 | } | |
5351 | ||
5352 | /* | |
49f693fa DC |
5353 | * Unmap (remove) blocks from a file. |
5354 | * If nexts is nonzero then the number of extents to remove is limited to | |
5355 | * that value. If not all extents in the block range can be removed then | |
5356 | * *done is set. | |
2bd0ea18 | 5357 | */ |
49f693fa | 5358 | int /* error */ |
3d36acda | 5359 | __xfs_bunmapi( |
49f693fa DC |
5360 | xfs_trans_t *tp, /* transaction pointer */ |
5361 | struct xfs_inode *ip, /* incore inode */ | |
5362 | xfs_fileoff_t bno, /* starting offset to unmap */ | |
3d36acda | 5363 | xfs_filblks_t *rlen, /* i/o: amount remaining */ |
49f693fa DC |
5364 | int flags, /* misc flags */ |
5365 | xfs_extnum_t nexts, /* number of extents max */ | |
5366 | xfs_fsblock_t *firstblock, /* first allocated block | |
5367 | controls a.g. for allocs */ | |
3d36acda | 5368 | struct xfs_defer_ops *dfops) /* i/o: deferred updates */ |
2bd0ea18 | 5369 | { |
49f693fa DC |
5370 | xfs_btree_cur_t *cur; /* bmap btree cursor */ |
5371 | xfs_bmbt_irec_t del; /* extent being deleted */ | |
49f693fa DC |
5372 | int error; /* error return value */ |
5373 | xfs_extnum_t extno; /* extent number in list */ | |
5374 | xfs_bmbt_irec_t got; /* current extent record */ | |
5375 | xfs_ifork_t *ifp; /* inode fork pointer */ | |
5376 | int isrt; /* freeing in rt area */ | |
5377 | xfs_extnum_t lastx; /* last extent index used */ | |
5378 | int logflags; /* transaction logging flags */ | |
5379 | xfs_extlen_t mod; /* rt extent offset */ | |
5380 | xfs_mount_t *mp; /* mount structure */ | |
49f693fa DC |
5381 | xfs_fileoff_t start; /* first file offset deleted */ |
5382 | int tmp_logflags; /* partial logging flags */ | |
5383 | int wasdel; /* was a delayed alloc extent */ | |
5384 | int whichfork; /* data or attribute fork */ | |
5385 | xfs_fsblock_t sum; | |
3d36acda | 5386 | xfs_filblks_t len = *rlen; /* length to unmap in file */ |
2bd0ea18 | 5387 | |
49f693fa | 5388 | trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_); |
a2ceac1f | 5389 | |
cb8a004a DW |
5390 | whichfork = xfs_bmapi_whichfork(flags); |
5391 | ASSERT(whichfork != XFS_COW_FORK); | |
a2ceac1f | 5392 | ifp = XFS_IFORK_PTR(ip, whichfork); |
49f693fa DC |
5393 | if (unlikely( |
5394 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
5395 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { | |
5396 | XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW, | |
5397 | ip->i_mount); | |
12b53197 | 5398 | return -EFSCORRUPTED; |
49f693fa DC |
5399 | } |
5400 | mp = ip->i_mount; | |
5401 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 5402 | return -EIO; |
56b2de80 | 5403 | |
ff105f75 | 5404 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
49f693fa DC |
5405 | ASSERT(len > 0); |
5406 | ASSERT(nexts >= 0); | |
56b2de80 | 5407 | |
49f693fa DC |
5408 | if (!(ifp->if_flags & XFS_IFEXTENTS) && |
5409 | (error = xfs_iread_extents(tp, ip, whichfork))) | |
5410 | return error; | |
d09d4e5f | 5411 | if (xfs_iext_count(ifp) == 0) { |
3d36acda | 5412 | *rlen = 0; |
49f693fa | 5413 | return 0; |
56b2de80 | 5414 | } |
79896434 | 5415 | XFS_STATS_INC(mp, xs_blk_unmap); |
49f693fa DC |
5416 | isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip); |
5417 | start = bno; | |
5418 | bno = start + len - 1; | |
a2ceac1f DC |
5419 | |
5420 | /* | |
49f693fa DC |
5421 | * Check to see if the given block number is past the end of the |
5422 | * file, back up to the last block if so... | |
56b2de80 | 5423 | */ |
246eb90a CH |
5424 | if (!xfs_iext_lookup_extent(ip, ifp, bno, &lastx, &got)) { |
5425 | ASSERT(lastx > 0); | |
5426 | xfs_iext_get_extent(ifp, --lastx, &got); | |
49f693fa DC |
5427 | bno = got.br_startoff + got.br_blockcount - 1; |
5428 | } | |
246eb90a | 5429 | |
49f693fa DC |
5430 | logflags = 0; |
5431 | if (ifp->if_flags & XFS_IFBROOT) { | |
5432 | ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE); | |
5433 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
5434 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 5435 | cur->bc_private.b.dfops = dfops; |
49f693fa DC |
5436 | cur->bc_private.b.flags = 0; |
5437 | } else | |
5438 | cur = NULL; | |
a2ceac1f | 5439 | |
49f693fa | 5440 | if (isrt) { |
a2ceac1f | 5441 | /* |
49f693fa | 5442 | * Synchronize by locking the bitmap inode. |
a2ceac1f | 5443 | */ |
a62ed6d3 | 5444 | xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); |
49f693fa | 5445 | xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL); |
a62ed6d3 DW |
5446 | xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); |
5447 | xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL); | |
49f693fa | 5448 | } |
a2ceac1f | 5449 | |
49f693fa DC |
5450 | extno = 0; |
5451 | while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 && | |
5452 | (nexts == 0 || extno < nexts)) { | |
a2ceac1f | 5453 | /* |
49f693fa DC |
5454 | * Is the found extent after a hole in which bno lives? |
5455 | * Just back up to the previous extent, if so. | |
a2ceac1f | 5456 | */ |
49f693fa DC |
5457 | if (got.br_startoff > bno) { |
5458 | if (--lastx < 0) | |
5459 | break; | |
246eb90a | 5460 | xfs_iext_get_extent(ifp, lastx, &got); |
a2ceac1f | 5461 | } |
49f693fa DC |
5462 | /* |
5463 | * Is the last block of this extent before the range | |
5464 | * we're supposed to delete? If so, we're done. | |
5465 | */ | |
5466 | bno = XFS_FILEOFF_MIN(bno, | |
5467 | got.br_startoff + got.br_blockcount - 1); | |
5468 | if (bno < start) | |
5469 | break; | |
5470 | /* | |
5471 | * Then deal with the (possibly delayed) allocated space | |
5472 | * we found. | |
5473 | */ | |
49f693fa DC |
5474 | del = got; |
5475 | wasdel = isnullstartblock(del.br_startblock); | |
5476 | if (got.br_startoff < start) { | |
5477 | del.br_startoff = start; | |
5478 | del.br_blockcount -= start - got.br_startoff; | |
5479 | if (!wasdel) | |
5480 | del.br_startblock += start - got.br_startoff; | |
5481 | } | |
5482 | if (del.br_startoff + del.br_blockcount > bno + 1) | |
5483 | del.br_blockcount = bno + 1 - del.br_startoff; | |
5484 | sum = del.br_startblock + del.br_blockcount; | |
5485 | if (isrt && | |
5486 | (mod = do_mod(sum, mp->m_sb.sb_rextsize))) { | |
5487 | /* | |
5488 | * Realtime extent not lined up at the end. | |
5489 | * The extent could have been split into written | |
5490 | * and unwritten pieces, or we could just be | |
5491 | * unmapping part of it. But we can't really | |
5492 | * get rid of part of a realtime extent. | |
5493 | */ | |
5494 | if (del.br_state == XFS_EXT_UNWRITTEN || | |
5495 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { | |
5496 | /* | |
5497 | * This piece is unwritten, or we're not | |
5498 | * using unwritten extents. Skip over it. | |
5499 | */ | |
5500 | ASSERT(bno >= mod); | |
5501 | bno -= mod > del.br_blockcount ? | |
5502 | del.br_blockcount : mod; | |
5503 | if (bno < got.br_startoff) { | |
5504 | if (--lastx >= 0) | |
5505 | xfs_bmbt_get_all(xfs_iext_get_ext( | |
5506 | ifp, lastx), &got); | |
5507 | } | |
5508 | continue; | |
5509 | } | |
5510 | /* | |
5511 | * It's written, turn it unwritten. | |
5512 | * This is better than zeroing it. | |
5513 | */ | |
5514 | ASSERT(del.br_state == XFS_EXT_NORM); | |
0268fdc3 | 5515 | ASSERT(tp->t_blk_res > 0); |
49f693fa DC |
5516 | /* |
5517 | * If this spans a realtime extent boundary, | |
5518 | * chop it back to the start of the one we end at. | |
5519 | */ | |
5520 | if (del.br_blockcount > mod) { | |
5521 | del.br_startoff += del.br_blockcount - mod; | |
5522 | del.br_startblock += del.br_blockcount - mod; | |
5523 | del.br_blockcount = mod; | |
5524 | } | |
5525 | del.br_state = XFS_EXT_UNWRITTEN; | |
5526 | error = xfs_bmap_add_extent_unwritten_real(tp, ip, | |
f33cea1a | 5527 | &lastx, &cur, &del, firstblock, dfops, |
49f693fa DC |
5528 | &logflags); |
5529 | if (error) | |
5530 | goto error0; | |
5531 | goto nodelete; | |
a2ceac1f | 5532 | } |
49f693fa DC |
5533 | if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) { |
5534 | /* | |
5535 | * Realtime extent is lined up at the end but not | |
5536 | * at the front. We'll get rid of full extents if | |
5537 | * we can. | |
5538 | */ | |
5539 | mod = mp->m_sb.sb_rextsize - mod; | |
5540 | if (del.br_blockcount > mod) { | |
5541 | del.br_blockcount -= mod; | |
5542 | del.br_startoff += mod; | |
5543 | del.br_startblock += mod; | |
5544 | } else if ((del.br_startoff == start && | |
5545 | (del.br_state == XFS_EXT_UNWRITTEN || | |
0268fdc3 | 5546 | tp->t_blk_res == 0)) || |
49f693fa DC |
5547 | !xfs_sb_version_hasextflgbit(&mp->m_sb)) { |
5548 | /* | |
5549 | * Can't make it unwritten. There isn't | |
5550 | * a full extent here so just skip it. | |
5551 | */ | |
5552 | ASSERT(bno >= del.br_blockcount); | |
5553 | bno -= del.br_blockcount; | |
246eb90a CH |
5554 | if (got.br_startoff > bno && --lastx >= 0) |
5555 | xfs_iext_get_extent(ifp, lastx, &got); | |
49f693fa DC |
5556 | continue; |
5557 | } else if (del.br_state == XFS_EXT_UNWRITTEN) { | |
246eb90a CH |
5558 | struct xfs_bmbt_irec prev; |
5559 | ||
49f693fa DC |
5560 | /* |
5561 | * This one is already unwritten. | |
5562 | * It must have a written left neighbor. | |
5563 | * Unwrite the killed part of that one and | |
5564 | * try again. | |
5565 | */ | |
5566 | ASSERT(lastx > 0); | |
246eb90a | 5567 | xfs_iext_get_extent(ifp, lastx - 1, &prev); |
49f693fa DC |
5568 | ASSERT(prev.br_state == XFS_EXT_NORM); |
5569 | ASSERT(!isnullstartblock(prev.br_startblock)); | |
5570 | ASSERT(del.br_startblock == | |
5571 | prev.br_startblock + prev.br_blockcount); | |
5572 | if (prev.br_startoff < start) { | |
5573 | mod = start - prev.br_startoff; | |
5574 | prev.br_blockcount -= mod; | |
5575 | prev.br_startblock += mod; | |
5576 | prev.br_startoff = start; | |
5577 | } | |
5578 | prev.br_state = XFS_EXT_UNWRITTEN; | |
5579 | lastx--; | |
5580 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
5581 | ip, &lastx, &cur, &prev, | |
f33cea1a | 5582 | firstblock, dfops, &logflags); |
49f693fa DC |
5583 | if (error) |
5584 | goto error0; | |
5585 | goto nodelete; | |
5586 | } else { | |
5587 | ASSERT(del.br_state == XFS_EXT_NORM); | |
5588 | del.br_state = XFS_EXT_UNWRITTEN; | |
5589 | error = xfs_bmap_add_extent_unwritten_real(tp, | |
5590 | ip, &lastx, &cur, &del, | |
f33cea1a | 5591 | firstblock, dfops, &logflags); |
49f693fa DC |
5592 | if (error) |
5593 | goto error0; | |
5594 | goto nodelete; | |
5595 | } | |
5596 | } | |
a2ceac1f | 5597 | |
49f693fa DC |
5598 | /* |
5599 | * If it's the case where the directory code is running | |
5600 | * with no block reservation, and the deleted block is in | |
5601 | * the middle of its extent, and the resulting insert | |
5602 | * of an extent would cause transformation to btree format, | |
5603 | * then reject it. The calling code will then swap | |
5604 | * blocks around instead. | |
5605 | * We have to do this now, rather than waiting for the | |
5606 | * conversion to btree format, since the transaction | |
5607 | * will be dirty. | |
5608 | */ | |
0268fdc3 | 5609 | if (!wasdel && tp->t_blk_res == 0 && |
49f693fa DC |
5610 | XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS && |
5611 | XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */ | |
5612 | XFS_IFORK_MAXEXT(ip, whichfork) && | |
5613 | del.br_startoff > got.br_startoff && | |
5614 | del.br_startoff + del.br_blockcount < | |
5615 | got.br_startoff + got.br_blockcount) { | |
12b53197 | 5616 | error = -ENOSPC; |
49f693fa | 5617 | goto error0; |
a2ceac1f | 5618 | } |
8359e0b9 BF |
5619 | |
5620 | /* | |
5621 | * Unreserve quota and update realtime free space, if | |
5622 | * appropriate. If delayed allocation, update the inode delalloc | |
5623 | * counter now and wait to update the sb counters as | |
5624 | * xfs_bmap_del_extent() might need to borrow some blocks. | |
5625 | */ | |
5626 | if (wasdel) { | |
5627 | ASSERT(startblockval(del.br_startblock) > 0); | |
5628 | if (isrt) { | |
5629 | xfs_filblks_t rtexts; | |
5630 | ||
5631 | rtexts = XFS_FSB_TO_B(mp, del.br_blockcount); | |
5632 | do_div(rtexts, mp->m_sb.sb_rextsize); | |
5633 | xfs_mod_frextents(mp, (int64_t)rtexts); | |
5634 | (void)xfs_trans_reserve_quota_nblks(NULL, | |
5635 | ip, -((long)del.br_blockcount), 0, | |
5636 | XFS_QMOPT_RES_RTBLKS); | |
5637 | } else { | |
5638 | (void)xfs_trans_reserve_quota_nblks(NULL, | |
5639 | ip, -((long)del.br_blockcount), 0, | |
5640 | XFS_QMOPT_RES_REGBLKS); | |
5641 | } | |
5642 | ip->i_delayed_blks -= del.br_blockcount; | |
5643 | if (cur) | |
5644 | cur->bc_private.b.flags |= | |
5645 | XFS_BTCUR_BPRV_WASDEL; | |
5646 | } else if (cur) | |
5647 | cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL; | |
5648 | ||
f33cea1a | 5649 | error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del, |
36b16da8 | 5650 | &tmp_logflags, whichfork, flags); |
49f693fa DC |
5651 | logflags |= tmp_logflags; |
5652 | if (error) | |
5653 | goto error0; | |
8359e0b9 BF |
5654 | |
5655 | if (!isrt && wasdel) | |
5656 | xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false); | |
5657 | ||
49f693fa DC |
5658 | bno = del.br_startoff - 1; |
5659 | nodelete: | |
a2ceac1f | 5660 | /* |
49f693fa | 5661 | * If not done go on to the next (previous) record. |
a2ceac1f | 5662 | */ |
49f693fa DC |
5663 | if (bno != (xfs_fileoff_t)-1 && bno >= start) { |
5664 | if (lastx >= 0) { | |
246eb90a CH |
5665 | xfs_iext_get_extent(ifp, lastx, &got); |
5666 | if (got.br_startoff > bno && --lastx >= 0) | |
5667 | xfs_iext_get_extent(ifp, lastx, &got); | |
49f693fa DC |
5668 | } |
5669 | extno++; | |
a2ceac1f | 5670 | } |
a2ceac1f | 5671 | } |
3d36acda DW |
5672 | if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0) |
5673 | *rlen = 0; | |
5674 | else | |
5675 | *rlen = bno - start + 1; | |
56b2de80 | 5676 | |
49f693fa DC |
5677 | /* |
5678 | * Convert to a btree if necessary. | |
5679 | */ | |
5680 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
5681 | ASSERT(cur == NULL); | |
f33cea1a | 5682 | error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, |
49f693fa DC |
5683 | &cur, 0, &tmp_logflags, whichfork); |
5684 | logflags |= tmp_logflags; | |
5685 | if (error) | |
5686 | goto error0; | |
56b2de80 | 5687 | } |
56b2de80 | 5688 | /* |
49f693fa | 5689 | * transform from btree to extents, give it cur |
56b2de80 | 5690 | */ |
49f693fa DC |
5691 | else if (xfs_bmap_wants_extents(ip, whichfork)) { |
5692 | ASSERT(cur != NULL); | |
5693 | error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags, | |
5694 | whichfork); | |
5695 | logflags |= tmp_logflags; | |
5696 | if (error) | |
5697 | goto error0; | |
56b2de80 | 5698 | } |
49f693fa DC |
5699 | /* |
5700 | * transform from extents to local? | |
5701 | */ | |
5702 | error = 0; | |
5703 | error0: | |
5704 | /* | |
5705 | * Log everything. Do this after conversion, there's no point in | |
5706 | * logging the extent records if we've converted to btree format. | |
5707 | */ | |
5708 | if ((logflags & xfs_ilog_fext(whichfork)) && | |
5709 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS) | |
5710 | logflags &= ~xfs_ilog_fext(whichfork); | |
5711 | else if ((logflags & xfs_ilog_fbroot(whichfork)) && | |
5712 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) | |
5713 | logflags &= ~xfs_ilog_fbroot(whichfork); | |
5714 | /* | |
5715 | * Log inode even in the error case, if the transaction | |
5716 | * is dirty we'll need to shut down the filesystem. | |
5717 | */ | |
5718 | if (logflags) | |
5719 | xfs_trans_log_inode(tp, ip, logflags); | |
5720 | if (cur) { | |
5721 | if (!error) { | |
5722 | *firstblock = cur->bc_private.b.firstblock; | |
5723 | cur->bc_private.b.allocated = 0; | |
56b2de80 | 5724 | } |
49f693fa DC |
5725 | xfs_btree_del_cursor(cur, |
5726 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
56b2de80 | 5727 | } |
49f693fa | 5728 | return error; |
a2ceac1f | 5729 | } |
ff105f75 | 5730 | |
3d36acda DW |
5731 | /* Unmap a range of a file. */ |
5732 | int | |
5733 | xfs_bunmapi( | |
5734 | xfs_trans_t *tp, | |
5735 | struct xfs_inode *ip, | |
5736 | xfs_fileoff_t bno, | |
5737 | xfs_filblks_t len, | |
5738 | int flags, | |
5739 | xfs_extnum_t nexts, | |
5740 | xfs_fsblock_t *firstblock, | |
5741 | struct xfs_defer_ops *dfops, | |
5742 | int *done) | |
5743 | { | |
5744 | int error; | |
5745 | ||
5746 | error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock, | |
5747 | dfops); | |
5748 | *done = (len == 0); | |
5749 | return error; | |
5750 | } | |
5751 | ||
5a35bf2c DC |
5752 | /* |
5753 | * Determine whether an extent shift can be accomplished by a merge with the | |
5754 | * extent that precedes the target hole of the shift. | |
5755 | */ | |
5756 | STATIC bool | |
5757 | xfs_bmse_can_merge( | |
5758 | struct xfs_bmbt_irec *left, /* preceding extent */ | |
5759 | struct xfs_bmbt_irec *got, /* current extent to shift */ | |
5760 | xfs_fileoff_t shift) /* shift fsb */ | |
5761 | { | |
5762 | xfs_fileoff_t startoff; | |
5763 | ||
5764 | startoff = got->br_startoff - shift; | |
5765 | ||
5766 | /* | |
5767 | * The extent, once shifted, must be adjacent in-file and on-disk with | |
5768 | * the preceding extent. | |
5769 | */ | |
5770 | if ((left->br_startoff + left->br_blockcount != startoff) || | |
5771 | (left->br_startblock + left->br_blockcount != got->br_startblock) || | |
5772 | (left->br_state != got->br_state) || | |
5773 | (left->br_blockcount + got->br_blockcount > MAXEXTLEN)) | |
5774 | return false; | |
5775 | ||
5776 | return true; | |
5777 | } | |
5778 | ||
5779 | /* | |
5780 | * A bmap extent shift adjusts the file offset of an extent to fill a preceding | |
5781 | * hole in the file. If an extent shift would result in the extent being fully | |
5782 | * adjacent to the extent that currently precedes the hole, we can merge with | |
5783 | * the preceding extent rather than do the shift. | |
5784 | * | |
5785 | * This function assumes the caller has verified a shift-by-merge is possible | |
5786 | * with the provided extents via xfs_bmse_can_merge(). | |
5787 | */ | |
5788 | STATIC int | |
5789 | xfs_bmse_merge( | |
5790 | struct xfs_inode *ip, | |
5791 | int whichfork, | |
5792 | xfs_fileoff_t shift, /* shift fsb */ | |
5793 | int current_ext, /* idx of gotp */ | |
5794 | struct xfs_bmbt_rec_host *gotp, /* extent to shift */ | |
5795 | struct xfs_bmbt_rec_host *leftp, /* preceding extent */ | |
5796 | struct xfs_btree_cur *cur, | |
5797 | int *logflags) /* output */ | |
5798 | { | |
5799 | struct xfs_bmbt_irec got; | |
5800 | struct xfs_bmbt_irec left; | |
5801 | xfs_filblks_t blockcount; | |
5802 | int error, i; | |
19ebedcf | 5803 | struct xfs_mount *mp = ip->i_mount; |
5a35bf2c DC |
5804 | |
5805 | xfs_bmbt_get_all(gotp, &got); | |
5806 | xfs_bmbt_get_all(leftp, &left); | |
5807 | blockcount = left.br_blockcount + got.br_blockcount; | |
5808 | ||
5809 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
5810 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
5811 | ASSERT(xfs_bmse_can_merge(&left, &got, shift)); | |
5812 | ||
5813 | /* | |
5814 | * Merge the in-core extents. Note that the host record pointers and | |
5815 | * current_ext index are invalid once the extent has been removed via | |
5816 | * xfs_iext_remove(). | |
5817 | */ | |
5818 | xfs_bmbt_set_blockcount(leftp, blockcount); | |
5819 | xfs_iext_remove(ip, current_ext, 1, 0); | |
5820 | ||
5821 | /* | |
5822 | * Update the on-disk extent count, the btree if necessary and log the | |
5823 | * inode. | |
5824 | */ | |
5825 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
5826 | XFS_IFORK_NEXTENTS(ip, whichfork) - 1); | |
5827 | *logflags |= XFS_ILOG_CORE; | |
5828 | if (!cur) { | |
5829 | *logflags |= XFS_ILOG_DEXT; | |
5830 | return 0; | |
5831 | } | |
5832 | ||
5833 | /* lookup and remove the extent to merge */ | |
5834 | error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock, | |
5835 | got.br_blockcount, &i); | |
5836 | if (error) | |
5837 | return error; | |
19ebedcf | 5838 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5839 | |
5840 | error = xfs_btree_delete(cur, &i); | |
5841 | if (error) | |
5842 | return error; | |
19ebedcf | 5843 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5844 | |
5845 | /* lookup and update size of the previous extent */ | |
5846 | error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock, | |
5847 | left.br_blockcount, &i); | |
5848 | if (error) | |
5849 | return error; | |
19ebedcf | 5850 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5851 | |
5852 | left.br_blockcount = blockcount; | |
5853 | ||
5854 | return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock, | |
5855 | left.br_blockcount, left.br_state); | |
5856 | } | |
5857 | ||
5858 | /* | |
5859 | * Shift a single extent. | |
5860 | */ | |
5861 | STATIC int | |
5862 | xfs_bmse_shift_one( | |
5863 | struct xfs_inode *ip, | |
5864 | int whichfork, | |
5865 | xfs_fileoff_t offset_shift_fsb, | |
5866 | int *current_ext, | |
5867 | struct xfs_bmbt_rec_host *gotp, | |
5868 | struct xfs_btree_cur *cur, | |
19ebedcf | 5869 | int *logflags, |
d7f80320 DW |
5870 | enum shift_direction direction, |
5871 | struct xfs_defer_ops *dfops) | |
5a35bf2c DC |
5872 | { |
5873 | struct xfs_ifork *ifp; | |
19ebedcf | 5874 | struct xfs_mount *mp; |
5a35bf2c | 5875 | xfs_fileoff_t startoff; |
19ebedcf | 5876 | struct xfs_bmbt_rec_host *adj_irecp; |
5a35bf2c | 5877 | struct xfs_bmbt_irec got; |
19ebedcf | 5878 | struct xfs_bmbt_irec adj_irec; |
5a35bf2c DC |
5879 | int error; |
5880 | int i; | |
19ebedcf | 5881 | int total_extents; |
5a35bf2c | 5882 | |
19ebedcf | 5883 | mp = ip->i_mount; |
5a35bf2c | 5884 | ifp = XFS_IFORK_PTR(ip, whichfork); |
d09d4e5f | 5885 | total_extents = xfs_iext_count(ifp); |
5a35bf2c DC |
5886 | |
5887 | xfs_bmbt_get_all(gotp, &got); | |
5a35bf2c DC |
5888 | |
5889 | /* delalloc extents should be prevented by caller */ | |
19ebedcf | 5890 | XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock)); |
5a35bf2c | 5891 | |
19ebedcf DC |
5892 | if (direction == SHIFT_LEFT) { |
5893 | startoff = got.br_startoff - offset_shift_fsb; | |
5894 | ||
5895 | /* | |
5896 | * Check for merge if we've got an extent to the left, | |
5897 | * otherwise make sure there's enough room at the start | |
5898 | * of the file for the shift. | |
5899 | */ | |
5900 | if (!*current_ext) { | |
5901 | if (got.br_startoff < offset_shift_fsb) | |
5902 | return -EINVAL; | |
5903 | goto update_current_ext; | |
5904 | } | |
5905 | /* | |
5906 | * grab the left extent and check for a large | |
5907 | * enough hole. | |
5908 | */ | |
5909 | adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1); | |
5910 | xfs_bmbt_get_all(adj_irecp, &adj_irec); | |
5a35bf2c | 5911 | |
19ebedcf DC |
5912 | if (startoff < |
5913 | adj_irec.br_startoff + adj_irec.br_blockcount) | |
5a35bf2c DC |
5914 | return -EINVAL; |
5915 | ||
5916 | /* check whether to merge the extent or shift it down */ | |
19ebedcf DC |
5917 | if (xfs_bmse_can_merge(&adj_irec, &got, |
5918 | offset_shift_fsb)) { | |
d7f80320 DW |
5919 | error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb, |
5920 | *current_ext, gotp, adj_irecp, | |
5921 | cur, logflags); | |
5922 | if (error) | |
5923 | return error; | |
5924 | adj_irec = got; | |
5925 | goto update_rmap; | |
5a35bf2c | 5926 | } |
19ebedcf DC |
5927 | } else { |
5928 | startoff = got.br_startoff + offset_shift_fsb; | |
5929 | /* nothing to move if this is the last extent */ | |
5930 | if (*current_ext >= (total_extents - 1)) | |
5931 | goto update_current_ext; | |
5932 | /* | |
5933 | * If this is not the last extent in the file, make sure there | |
5934 | * is enough room between current extent and next extent for | |
5935 | * accommodating the shift. | |
5936 | */ | |
5937 | adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1); | |
5938 | xfs_bmbt_get_all(adj_irecp, &adj_irec); | |
5939 | if (startoff + got.br_blockcount > adj_irec.br_startoff) | |
5940 | return -EINVAL; | |
5941 | /* | |
5942 | * Unlike a left shift (which involves a hole punch), | |
5943 | * a right shift does not modify extent neighbors | |
5944 | * in any way. We should never find mergeable extents | |
5945 | * in this scenario. Check anyways and warn if we | |
5946 | * encounter two extents that could be one. | |
5947 | */ | |
5948 | if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb)) | |
5949 | WARN_ON_ONCE(1); | |
5950 | } | |
5a35bf2c DC |
5951 | /* |
5952 | * Increment the extent index for the next iteration, update the start | |
5953 | * offset of the in-core extent and update the btree if applicable. | |
5954 | */ | |
19ebedcf DC |
5955 | update_current_ext: |
5956 | if (direction == SHIFT_LEFT) | |
5957 | (*current_ext)++; | |
5958 | else | |
5959 | (*current_ext)--; | |
5a35bf2c DC |
5960 | xfs_bmbt_set_startoff(gotp, startoff); |
5961 | *logflags |= XFS_ILOG_CORE; | |
d7f80320 | 5962 | adj_irec = got; |
5a35bf2c DC |
5963 | if (!cur) { |
5964 | *logflags |= XFS_ILOG_DEXT; | |
d7f80320 | 5965 | goto update_rmap; |
5a35bf2c DC |
5966 | } |
5967 | ||
5968 | error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock, | |
5969 | got.br_blockcount, &i); | |
5970 | if (error) | |
5971 | return error; | |
19ebedcf | 5972 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
5a35bf2c DC |
5973 | |
5974 | got.br_startoff = startoff; | |
d7f80320 DW |
5975 | error = xfs_bmbt_update(cur, got.br_startoff, got.br_startblock, |
5976 | got.br_blockcount, got.br_state); | |
5977 | if (error) | |
5978 | return error; | |
5979 | ||
5980 | update_rmap: | |
5981 | /* update reverse mapping */ | |
5982 | error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &adj_irec); | |
5983 | if (error) | |
5984 | return error; | |
5985 | adj_irec.br_startoff = startoff; | |
5986 | return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &adj_irec); | |
5a35bf2c DC |
5987 | } |
5988 | ||
ff105f75 | 5989 | /* |
19ebedcf | 5990 | * Shift extent records to the left/right to cover/create a hole. |
ff105f75 | 5991 | * |
5a35bf2c | 5992 | * The maximum number of extents to be shifted in a single operation is |
19ebedcf | 5993 | * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the |
5a35bf2c DC |
5994 | * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb |
5995 | * is the length by which each extent is shifted. If there is no hole to shift | |
5996 | * the extents into, this will be considered invalid operation and we abort | |
5997 | * immediately. | |
ff105f75 DC |
5998 | */ |
5999 | int | |
6000 | xfs_bmap_shift_extents( | |
6001 | struct xfs_trans *tp, | |
6002 | struct xfs_inode *ip, | |
19ebedcf | 6003 | xfs_fileoff_t *next_fsb, |
ff105f75 | 6004 | xfs_fileoff_t offset_shift_fsb, |
5a35bf2c | 6005 | int *done, |
19ebedcf | 6006 | xfs_fileoff_t stop_fsb, |
ff105f75 | 6007 | xfs_fsblock_t *firstblock, |
f33cea1a | 6008 | struct xfs_defer_ops *dfops, |
19ebedcf | 6009 | enum shift_direction direction, |
ff105f75 DC |
6010 | int num_exts) |
6011 | { | |
5a35bf2c | 6012 | struct xfs_btree_cur *cur = NULL; |
ff105f75 DC |
6013 | struct xfs_bmbt_rec_host *gotp; |
6014 | struct xfs_bmbt_irec got; | |
ff105f75 DC |
6015 | struct xfs_mount *mp = ip->i_mount; |
6016 | struct xfs_ifork *ifp; | |
6017 | xfs_extnum_t nexts = 0; | |
5a35bf2c | 6018 | xfs_extnum_t current_ext; |
19ebedcf DC |
6019 | xfs_extnum_t total_extents; |
6020 | xfs_extnum_t stop_extent; | |
ff105f75 | 6021 | int error = 0; |
ff105f75 | 6022 | int whichfork = XFS_DATA_FORK; |
5a35bf2c | 6023 | int logflags = 0; |
ff105f75 DC |
6024 | |
6025 | if (unlikely(XFS_TEST_ERROR( | |
6026 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
6027 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
6028 | mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) { | |
6029 | XFS_ERROR_REPORT("xfs_bmap_shift_extents", | |
6030 | XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 6031 | return -EFSCORRUPTED; |
ff105f75 DC |
6032 | } |
6033 | ||
6034 | if (XFS_FORCED_SHUTDOWN(mp)) | |
12b53197 | 6035 | return -EIO; |
ff105f75 | 6036 | |
5a35bf2c DC |
6037 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
6038 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
19ebedcf DC |
6039 | ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT); |
6040 | ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT); | |
ff105f75 DC |
6041 | |
6042 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
6043 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
6044 | /* Read in all the extents */ | |
6045 | error = xfs_iread_extents(tp, ip, whichfork); | |
6046 | if (error) | |
6047 | return error; | |
6048 | } | |
6049 | ||
ff105f75 DC |
6050 | if (ifp->if_flags & XFS_IFBROOT) { |
6051 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
6052 | cur->bc_private.b.firstblock = *firstblock; | |
f33cea1a | 6053 | cur->bc_private.b.dfops = dfops; |
ff105f75 | 6054 | cur->bc_private.b.flags = 0; |
5a35bf2c DC |
6055 | } |
6056 | ||
19ebedcf DC |
6057 | /* |
6058 | * There may be delalloc extents in the data fork before the range we | |
6059 | * are collapsing out, so we cannot use the count of real extents here. | |
6060 | * Instead we have to calculate it from the incore fork. | |
6061 | */ | |
d09d4e5f | 6062 | total_extents = xfs_iext_count(ifp); |
19ebedcf DC |
6063 | if (total_extents == 0) { |
6064 | *done = 1; | |
6065 | goto del_cursor; | |
6066 | } | |
6067 | ||
6068 | /* | |
6069 | * In case of first right shift, we need to initialize next_fsb | |
6070 | */ | |
6071 | if (*next_fsb == NULLFSBLOCK) { | |
6072 | gotp = xfs_iext_get_ext(ifp, total_extents - 1); | |
6073 | xfs_bmbt_get_all(gotp, &got); | |
6074 | *next_fsb = got.br_startoff; | |
6075 | if (stop_fsb > *next_fsb) { | |
6076 | *done = 1; | |
6077 | goto del_cursor; | |
6078 | } | |
6079 | } | |
6080 | ||
6081 | /* Lookup the extent index at which we have to stop */ | |
6082 | if (direction == SHIFT_RIGHT) { | |
6083 | gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent); | |
6084 | /* Make stop_extent exclusive of shift range */ | |
6085 | stop_extent--; | |
6086 | } else | |
6087 | stop_extent = total_extents; | |
6088 | ||
5a35bf2c DC |
6089 | /* |
6090 | * Look up the extent index for the fsb where we start shifting. We can | |
6091 | * henceforth iterate with current_ext as extent list changes are locked | |
6092 | * out via ilock. | |
6093 | * | |
6094 | * gotp can be null in 2 cases: 1) if there are no extents or 2) | |
19ebedcf | 6095 | * *next_fsb lies in a hole beyond which there are no extents. Either |
5a35bf2c DC |
6096 | * way, we are done. |
6097 | */ | |
19ebedcf | 6098 | gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, ¤t_ext); |
5a35bf2c DC |
6099 | if (!gotp) { |
6100 | *done = 1; | |
6101 | goto del_cursor; | |
ff105f75 DC |
6102 | } |
6103 | ||
19ebedcf DC |
6104 | /* some sanity checking before we finally start shifting extents */ |
6105 | if ((direction == SHIFT_LEFT && current_ext >= stop_extent) || | |
6106 | (direction == SHIFT_RIGHT && current_ext <= stop_extent)) { | |
6107 | error = -EIO; | |
6108 | goto del_cursor; | |
6109 | } | |
6110 | ||
6111 | while (nexts++ < num_exts) { | |
5a35bf2c | 6112 | error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb, |
19ebedcf | 6113 | ¤t_ext, gotp, cur, &logflags, |
d7f80320 | 6114 | direction, dfops); |
ff105f75 DC |
6115 | if (error) |
6116 | goto del_cursor; | |
19ebedcf DC |
6117 | /* |
6118 | * If there was an extent merge during the shift, the extent | |
6119 | * count can change. Update the total and grade the next record. | |
6120 | */ | |
6121 | if (direction == SHIFT_LEFT) { | |
d09d4e5f | 6122 | total_extents = xfs_iext_count(ifp); |
19ebedcf DC |
6123 | stop_extent = total_extents; |
6124 | } | |
ff105f75 | 6125 | |
19ebedcf DC |
6126 | if (current_ext == stop_extent) { |
6127 | *done = 1; | |
6128 | *next_fsb = NULLFSBLOCK; | |
5a35bf2c | 6129 | break; |
19ebedcf | 6130 | } |
5a35bf2c | 6131 | gotp = xfs_iext_get_ext(ifp, current_ext); |
ff105f75 DC |
6132 | } |
6133 | ||
19ebedcf | 6134 | if (!*done) { |
5a35bf2c DC |
6135 | xfs_bmbt_get_all(gotp, &got); |
6136 | *next_fsb = got.br_startoff; | |
6137 | } | |
ff105f75 DC |
6138 | |
6139 | del_cursor: | |
6140 | if (cur) | |
6141 | xfs_btree_del_cursor(cur, | |
6142 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
6143 | ||
5a35bf2c DC |
6144 | if (logflags) |
6145 | xfs_trans_log_inode(tp, ip, logflags); | |
6146 | ||
ff105f75 DC |
6147 | return error; |
6148 | } | |
19ebedcf DC |
6149 | |
6150 | /* | |
6151 | * Splits an extent into two extents at split_fsb block such that it is | |
6152 | * the first block of the current_ext. @current_ext is a target extent | |
6153 | * to be split. @split_fsb is a block where the extents is split. | |
6154 | * If split_fsb lies in a hole or the first block of extents, just return 0. | |
6155 | */ | |
6156 | STATIC int | |
6157 | xfs_bmap_split_extent_at( | |
6158 | struct xfs_trans *tp, | |
6159 | struct xfs_inode *ip, | |
6160 | xfs_fileoff_t split_fsb, | |
6161 | xfs_fsblock_t *firstfsb, | |
f33cea1a | 6162 | struct xfs_defer_ops *dfops) |
19ebedcf DC |
6163 | { |
6164 | int whichfork = XFS_DATA_FORK; | |
6165 | struct xfs_btree_cur *cur = NULL; | |
6166 | struct xfs_bmbt_rec_host *gotp; | |
6167 | struct xfs_bmbt_irec got; | |
6168 | struct xfs_bmbt_irec new; /* split extent */ | |
6169 | struct xfs_mount *mp = ip->i_mount; | |
6170 | struct xfs_ifork *ifp; | |
6171 | xfs_fsblock_t gotblkcnt; /* new block count for got */ | |
6172 | xfs_extnum_t current_ext; | |
6173 | int error = 0; | |
6174 | int logflags = 0; | |
6175 | int i = 0; | |
6176 | ||
6177 | if (unlikely(XFS_TEST_ERROR( | |
6178 | (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS && | |
6179 | XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE), | |
6180 | mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) { | |
6181 | XFS_ERROR_REPORT("xfs_bmap_split_extent_at", | |
6182 | XFS_ERRLEVEL_LOW, mp); | |
6183 | return -EFSCORRUPTED; | |
6184 | } | |
6185 | ||
6186 | if (XFS_FORCED_SHUTDOWN(mp)) | |
6187 | return -EIO; | |
6188 | ||
6189 | ifp = XFS_IFORK_PTR(ip, whichfork); | |
6190 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { | |
6191 | /* Read in all the extents */ | |
6192 | error = xfs_iread_extents(tp, ip, whichfork); | |
6193 | if (error) | |
6194 | return error; | |
6195 | } | |
6196 | ||
6197 | /* | |
6198 | * gotp can be null in 2 cases: 1) if there are no extents | |
6199 | * or 2) split_fsb lies in a hole beyond which there are | |
6200 | * no extents. Either way, we are done. | |
6201 | */ | |
6202 | gotp = xfs_iext_bno_to_ext(ifp, split_fsb, ¤t_ext); | |
6203 | if (!gotp) | |
6204 | return 0; | |
6205 | ||
6206 | xfs_bmbt_get_all(gotp, &got); | |
6207 | ||
6208 | /* | |
6209 | * Check split_fsb lies in a hole or the start boundary offset | |
6210 | * of the extent. | |
6211 | */ | |
6212 | if (got.br_startoff >= split_fsb) | |
6213 | return 0; | |
6214 | ||
6215 | gotblkcnt = split_fsb - got.br_startoff; | |
6216 | new.br_startoff = split_fsb; | |
6217 | new.br_startblock = got.br_startblock + gotblkcnt; | |
6218 | new.br_blockcount = got.br_blockcount - gotblkcnt; | |
6219 | new.br_state = got.br_state; | |
6220 | ||
6221 | if (ifp->if_flags & XFS_IFBROOT) { | |
6222 | cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); | |
6223 | cur->bc_private.b.firstblock = *firstfsb; | |
f33cea1a | 6224 | cur->bc_private.b.dfops = dfops; |
19ebedcf DC |
6225 | cur->bc_private.b.flags = 0; |
6226 | error = xfs_bmbt_lookup_eq(cur, got.br_startoff, | |
6227 | got.br_startblock, | |
6228 | got.br_blockcount, | |
6229 | &i); | |
6230 | if (error) | |
6231 | goto del_cursor; | |
6232 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
6233 | } | |
6234 | ||
6235 | xfs_bmbt_set_blockcount(gotp, gotblkcnt); | |
6236 | got.br_blockcount = gotblkcnt; | |
6237 | ||
6238 | logflags = XFS_ILOG_CORE; | |
6239 | if (cur) { | |
6240 | error = xfs_bmbt_update(cur, got.br_startoff, | |
6241 | got.br_startblock, | |
6242 | got.br_blockcount, | |
6243 | got.br_state); | |
6244 | if (error) | |
6245 | goto del_cursor; | |
6246 | } else | |
6247 | logflags |= XFS_ILOG_DEXT; | |
6248 | ||
6249 | /* Add new extent */ | |
6250 | current_ext++; | |
6251 | xfs_iext_insert(ip, current_ext, 1, &new, 0); | |
6252 | XFS_IFORK_NEXT_SET(ip, whichfork, | |
6253 | XFS_IFORK_NEXTENTS(ip, whichfork) + 1); | |
6254 | ||
6255 | if (cur) { | |
6256 | error = xfs_bmbt_lookup_eq(cur, new.br_startoff, | |
6257 | new.br_startblock, new.br_blockcount, | |
6258 | &i); | |
6259 | if (error) | |
6260 | goto del_cursor; | |
6261 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor); | |
6262 | cur->bc_rec.b.br_state = new.br_state; | |
6263 | ||
6264 | error = xfs_btree_insert(cur, &i); | |
6265 | if (error) | |
6266 | goto del_cursor; | |
6267 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor); | |
6268 | } | |
6269 | ||
6270 | /* | |
6271 | * Convert to a btree if necessary. | |
6272 | */ | |
6273 | if (xfs_bmap_needs_btree(ip, whichfork)) { | |
6274 | int tmp_logflags; /* partial log flag return val */ | |
6275 | ||
6276 | ASSERT(cur == NULL); | |
f33cea1a | 6277 | error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops, |
19ebedcf DC |
6278 | &cur, 0, &tmp_logflags, whichfork); |
6279 | logflags |= tmp_logflags; | |
6280 | } | |
6281 | ||
6282 | del_cursor: | |
6283 | if (cur) { | |
6284 | cur->bc_private.b.allocated = 0; | |
6285 | xfs_btree_del_cursor(cur, | |
6286 | error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR); | |
6287 | } | |
6288 | ||
6289 | if (logflags) | |
6290 | xfs_trans_log_inode(tp, ip, logflags); | |
6291 | return error; | |
6292 | } | |
6293 | ||
6294 | int | |
6295 | xfs_bmap_split_extent( | |
6296 | struct xfs_inode *ip, | |
6297 | xfs_fileoff_t split_fsb) | |
6298 | { | |
6299 | struct xfs_mount *mp = ip->i_mount; | |
6300 | struct xfs_trans *tp; | |
f33cea1a | 6301 | struct xfs_defer_ops dfops; |
19ebedcf | 6302 | xfs_fsblock_t firstfsb; |
19ebedcf DC |
6303 | int error; |
6304 | ||
9074815c CH |
6305 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, |
6306 | XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); | |
6307 | if (error) | |
19ebedcf | 6308 | return error; |
19ebedcf DC |
6309 | |
6310 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
6311 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
6312 | ||
f33cea1a | 6313 | xfs_defer_init(&dfops, &firstfsb); |
19ebedcf DC |
6314 | |
6315 | error = xfs_bmap_split_extent_at(tp, ip, split_fsb, | |
f33cea1a | 6316 | &firstfsb, &dfops); |
19ebedcf DC |
6317 | if (error) |
6318 | goto out; | |
6319 | ||
f33cea1a | 6320 | error = xfs_defer_finish(&tp, &dfops, NULL); |
19ebedcf DC |
6321 | if (error) |
6322 | goto out; | |
6323 | ||
de5a3f46 | 6324 | return xfs_trans_commit(tp); |
19ebedcf DC |
6325 | |
6326 | out: | |
f33cea1a | 6327 | xfs_defer_cancel(&dfops); |
3d7434fe | 6328 | xfs_trans_cancel(tp); |
19ebedcf DC |
6329 | return error; |
6330 | } | |
aeb88300 DW |
6331 | |
6332 | /* Deferred mapping is only for real extents in the data fork. */ | |
6333 | static bool | |
6334 | xfs_bmap_is_update_needed( | |
6335 | struct xfs_bmbt_irec *bmap) | |
6336 | { | |
6337 | return bmap->br_startblock != HOLESTARTBLOCK && | |
6338 | bmap->br_startblock != DELAYSTARTBLOCK; | |
6339 | } | |
6340 | ||
6341 | /* Record a bmap intent. */ | |
6342 | static int | |
6343 | __xfs_bmap_add( | |
6344 | struct xfs_mount *mp, | |
6345 | struct xfs_defer_ops *dfops, | |
6346 | enum xfs_bmap_intent_type type, | |
6347 | struct xfs_inode *ip, | |
6348 | int whichfork, | |
6349 | struct xfs_bmbt_irec *bmap) | |
6350 | { | |
6351 | int error; | |
6352 | struct xfs_bmap_intent *bi; | |
6353 | ||
6354 | trace_xfs_bmap_defer(mp, | |
6355 | XFS_FSB_TO_AGNO(mp, bmap->br_startblock), | |
6356 | type, | |
6357 | XFS_FSB_TO_AGBNO(mp, bmap->br_startblock), | |
6358 | ip->i_ino, whichfork, | |
6359 | bmap->br_startoff, | |
6360 | bmap->br_blockcount, | |
6361 | bmap->br_state); | |
6362 | ||
6363 | bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS); | |
6364 | INIT_LIST_HEAD(&bi->bi_list); | |
6365 | bi->bi_type = type; | |
6366 | bi->bi_owner = ip; | |
6367 | bi->bi_whichfork = whichfork; | |
6368 | bi->bi_bmap = *bmap; | |
6369 | ||
6370 | error = xfs_defer_join(dfops, bi->bi_owner); | |
6371 | if (error) { | |
6372 | kmem_free(bi); | |
6373 | return error; | |
6374 | } | |
6375 | ||
6376 | xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list); | |
6377 | return 0; | |
6378 | } | |
6379 | ||
6380 | /* Map an extent into a file. */ | |
6381 | int | |
6382 | xfs_bmap_map_extent( | |
6383 | struct xfs_mount *mp, | |
6384 | struct xfs_defer_ops *dfops, | |
6385 | struct xfs_inode *ip, | |
6386 | struct xfs_bmbt_irec *PREV) | |
6387 | { | |
6388 | if (!xfs_bmap_is_update_needed(PREV)) | |
6389 | return 0; | |
6390 | ||
6391 | return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip, | |
6392 | XFS_DATA_FORK, PREV); | |
6393 | } | |
6394 | ||
6395 | /* Unmap an extent out of a file. */ | |
6396 | int | |
6397 | xfs_bmap_unmap_extent( | |
6398 | struct xfs_mount *mp, | |
6399 | struct xfs_defer_ops *dfops, | |
6400 | struct xfs_inode *ip, | |
6401 | struct xfs_bmbt_irec *PREV) | |
6402 | { | |
6403 | if (!xfs_bmap_is_update_needed(PREV)) | |
6404 | return 0; | |
6405 | ||
6406 | return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip, | |
6407 | XFS_DATA_FORK, PREV); | |
6408 | } | |
6409 | ||
6410 | /* | |
6411 | * Process one of the deferred bmap operations. We pass back the | |
6412 | * btree cursor to maintain our lock on the bmapbt between calls. | |
6413 | */ | |
6414 | int | |
6415 | xfs_bmap_finish_one( | |
6416 | struct xfs_trans *tp, | |
6417 | struct xfs_defer_ops *dfops, | |
6418 | struct xfs_inode *ip, | |
6419 | enum xfs_bmap_intent_type type, | |
6420 | int whichfork, | |
6421 | xfs_fileoff_t startoff, | |
6422 | xfs_fsblock_t startblock, | |
6423 | xfs_filblks_t blockcount, | |
6424 | xfs_exntst_t state) | |
6425 | { | |
6426 | struct xfs_bmbt_irec bmap; | |
6427 | int nimaps = 1; | |
6428 | xfs_fsblock_t firstfsb; | |
6429 | int flags = XFS_BMAPI_REMAP; | |
6430 | int done; | |
6431 | int error = 0; | |
6432 | ||
6433 | bmap.br_startblock = startblock; | |
6434 | bmap.br_startoff = startoff; | |
6435 | bmap.br_blockcount = blockcount; | |
6436 | bmap.br_state = state; | |
6437 | ||
6438 | trace_xfs_bmap_deferred(tp->t_mountp, | |
6439 | XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type, | |
6440 | XFS_FSB_TO_AGBNO(tp->t_mountp, startblock), | |
6441 | ip->i_ino, whichfork, startoff, blockcount, state); | |
6442 | ||
6443 | if (whichfork != XFS_DATA_FORK && whichfork != XFS_ATTR_FORK) | |
6444 | return -EFSCORRUPTED; | |
6445 | if (whichfork == XFS_ATTR_FORK) | |
6446 | flags |= XFS_BMAPI_ATTRFORK; | |
6447 | ||
6448 | if (XFS_TEST_ERROR(false, tp->t_mountp, | |
6449 | XFS_ERRTAG_BMAP_FINISH_ONE, | |
6450 | XFS_RANDOM_BMAP_FINISH_ONE)) | |
6451 | return -EIO; | |
6452 | ||
6453 | switch (type) { | |
6454 | case XFS_BMAP_MAP: | |
6455 | firstfsb = bmap.br_startblock; | |
6456 | error = xfs_bmapi_write(tp, ip, bmap.br_startoff, | |
6457 | bmap.br_blockcount, flags, &firstfsb, | |
6458 | bmap.br_blockcount, &bmap, &nimaps, | |
6459 | dfops); | |
6460 | break; | |
6461 | case XFS_BMAP_UNMAP: | |
6462 | error = xfs_bunmapi(tp, ip, bmap.br_startoff, | |
6463 | bmap.br_blockcount, flags, 1, &firstfsb, | |
6464 | dfops, &done); | |
6465 | ASSERT(done); | |
6466 | break; | |
6467 | default: | |
6468 | ASSERT(0); | |
6469 | error = -EFSCORRUPTED; | |
6470 | } | |
6471 | ||
6472 | return error; | |
6473 | } |