Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 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 | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 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. | |
1da177e4 | 13 | * |
7b718769 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 | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_format.h" |
239880ef | 21 | #include "xfs_log_format.h" |
70a9883c | 22 | #include "xfs_shared.h" |
239880ef | 23 | #include "xfs_trans_resv.h" |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
3ab78df2 | 27 | #include "xfs_defer.h" |
a844f451 | 28 | #include "xfs_inode.h" |
1da177e4 | 29 | #include "xfs_btree.h" |
673930c3 | 30 | #include "xfs_rmap.h" |
a4fbe6ab | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_alloc.h" |
efc27b52 | 33 | #include "xfs_extent_busy.h" |
1da177e4 | 34 | #include "xfs_error.h" |
4e0e6040 | 35 | #include "xfs_cksum.h" |
0b1b213f | 36 | #include "xfs_trace.h" |
239880ef | 37 | #include "xfs_trans.h" |
4e0e6040 | 38 | #include "xfs_buf_item.h" |
239880ef | 39 | #include "xfs_log.h" |
1da177e4 | 40 | |
c999a223 | 41 | struct workqueue_struct *xfs_alloc_wq; |
1da177e4 LT |
42 | |
43 | #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b))) | |
44 | ||
45 | #define XFSA_FIXUP_BNO_OK 1 | |
46 | #define XFSA_FIXUP_CNT_OK 2 | |
47 | ||
1da177e4 LT |
48 | STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *); |
49 | STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *); | |
50 | STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *); | |
51 | STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *, | |
e26f0501 | 52 | xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *); |
1da177e4 | 53 | |
8018026e DW |
54 | xfs_extlen_t |
55 | xfs_prealloc_blocks( | |
56 | struct xfs_mount *mp) | |
57 | { | |
58 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
59 | return XFS_RMAP_BLOCK(mp) + 1; | |
60 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) | |
61 | return XFS_FIBT_BLOCK(mp) + 1; | |
62 | return XFS_IBT_BLOCK(mp) + 1; | |
63 | } | |
64 | ||
52548852 DW |
65 | /* |
66 | * In order to avoid ENOSPC-related deadlock caused by out-of-order locking of | |
67 | * AGF buffer (PV 947395), we place constraints on the relationship among | |
68 | * actual allocations for data blocks, freelist blocks, and potential file data | |
69 | * bmap btree blocks. However, these restrictions may result in no actual space | |
70 | * allocated for a delayed extent, for example, a data block in a certain AG is | |
71 | * allocated but there is no additional block for the additional bmap btree | |
72 | * block due to a split of the bmap btree of the file. The result of this may | |
73 | * lead to an infinite loop when the file gets flushed to disk and all delayed | |
74 | * extents need to be actually allocated. To get around this, we explicitly set | |
75 | * aside a few blocks which will not be reserved in delayed allocation. | |
76 | * | |
77 | * When rmap is disabled, we need to reserve 4 fsbs _per AG_ for the freelist | |
78 | * and 4 more to handle a potential split of the file's bmap btree. | |
79 | * | |
80 | * When rmap is enabled, we must also be able to handle two rmap btree inserts | |
81 | * to record both the file data extent and a new bmbt block. The bmbt block | |
82 | * might not be in the same AG as the file data extent. In the worst case | |
83 | * the bmap btree splits multiple levels and all the new blocks come from | |
84 | * different AGs, so set aside enough to handle rmap btree splits in all AGs. | |
85 | */ | |
86 | unsigned int | |
87 | xfs_alloc_set_aside( | |
88 | struct xfs_mount *mp) | |
89 | { | |
90 | unsigned int blocks; | |
91 | ||
92 | blocks = 4 + (mp->m_sb.sb_agcount * XFS_ALLOC_AGFL_RESERVE); | |
93 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
94 | blocks += mp->m_sb.sb_agcount * mp->m_rmap_maxlevels; | |
95 | return blocks; | |
96 | } | |
97 | ||
98 | /* | |
99 | * When deciding how much space to allocate out of an AG, we limit the | |
100 | * allocation maximum size to the size the AG. However, we cannot use all the | |
101 | * blocks in the AG - some are permanently used by metadata. These | |
102 | * blocks are generally: | |
103 | * - the AG superblock, AGF, AGI and AGFL | |
104 | * - the AGF (bno and cnt) and AGI btree root blocks, and optionally | |
105 | * the AGI free inode and rmap btree root blocks. | |
106 | * - blocks on the AGFL according to xfs_alloc_set_aside() limits | |
107 | * - the rmapbt root block | |
108 | * | |
109 | * The AG headers are sector sized, so the amount of space they take up is | |
110 | * dependent on filesystem geometry. The others are all single blocks. | |
111 | */ | |
112 | unsigned int | |
113 | xfs_alloc_ag_max_usable( | |
114 | struct xfs_mount *mp) | |
115 | { | |
116 | unsigned int blocks; | |
117 | ||
118 | blocks = XFS_BB_TO_FSB(mp, XFS_FSS_TO_BB(mp, 4)); /* ag headers */ | |
119 | blocks += XFS_ALLOC_AGFL_RESERVE; | |
120 | blocks += 3; /* AGF, AGI btree root blocks */ | |
121 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) | |
122 | blocks++; /* finobt root block */ | |
123 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
124 | blocks++; /* rmap root block */ | |
125 | ||
126 | return mp->m_sb.sb_agblocks - blocks; | |
127 | } | |
128 | ||
fe033cc8 CH |
129 | /* |
130 | * Lookup the record equal to [bno, len] in the btree given by cur. | |
131 | */ | |
132 | STATIC int /* error */ | |
133 | xfs_alloc_lookup_eq( | |
134 | struct xfs_btree_cur *cur, /* btree cursor */ | |
135 | xfs_agblock_t bno, /* starting block of extent */ | |
136 | xfs_extlen_t len, /* length of extent */ | |
137 | int *stat) /* success/failure */ | |
138 | { | |
139 | cur->bc_rec.a.ar_startblock = bno; | |
140 | cur->bc_rec.a.ar_blockcount = len; | |
141 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); | |
142 | } | |
143 | ||
144 | /* | |
145 | * Lookup the first record greater than or equal to [bno, len] | |
146 | * in the btree given by cur. | |
147 | */ | |
a66d6363 | 148 | int /* error */ |
fe033cc8 CH |
149 | xfs_alloc_lookup_ge( |
150 | struct xfs_btree_cur *cur, /* btree cursor */ | |
151 | xfs_agblock_t bno, /* starting block of extent */ | |
152 | xfs_extlen_t len, /* length of extent */ | |
153 | int *stat) /* success/failure */ | |
154 | { | |
155 | cur->bc_rec.a.ar_startblock = bno; | |
156 | cur->bc_rec.a.ar_blockcount = len; | |
157 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); | |
158 | } | |
159 | ||
160 | /* | |
161 | * Lookup the first record less than or equal to [bno, len] | |
162 | * in the btree given by cur. | |
163 | */ | |
0d5a75e9 | 164 | static int /* error */ |
fe033cc8 CH |
165 | xfs_alloc_lookup_le( |
166 | struct xfs_btree_cur *cur, /* btree cursor */ | |
167 | xfs_agblock_t bno, /* starting block of extent */ | |
168 | xfs_extlen_t len, /* length of extent */ | |
169 | int *stat) /* success/failure */ | |
170 | { | |
171 | cur->bc_rec.a.ar_startblock = bno; | |
172 | cur->bc_rec.a.ar_blockcount = len; | |
173 | return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat); | |
174 | } | |
175 | ||
278d0ca1 CH |
176 | /* |
177 | * Update the record referred to by cur to the value given | |
178 | * by [bno, len]. | |
179 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
180 | */ | |
181 | STATIC int /* error */ | |
182 | xfs_alloc_update( | |
183 | struct xfs_btree_cur *cur, /* btree cursor */ | |
184 | xfs_agblock_t bno, /* starting block of extent */ | |
185 | xfs_extlen_t len) /* length of extent */ | |
186 | { | |
187 | union xfs_btree_rec rec; | |
188 | ||
189 | rec.alloc.ar_startblock = cpu_to_be32(bno); | |
190 | rec.alloc.ar_blockcount = cpu_to_be32(len); | |
191 | return xfs_btree_update(cur, &rec); | |
192 | } | |
fe033cc8 | 193 | |
8cc938fe CH |
194 | /* |
195 | * Get the data from the pointed-to record. | |
196 | */ | |
a46db608 | 197 | int /* error */ |
8cc938fe CH |
198 | xfs_alloc_get_rec( |
199 | struct xfs_btree_cur *cur, /* btree cursor */ | |
200 | xfs_agblock_t *bno, /* output: starting block of extent */ | |
201 | xfs_extlen_t *len, /* output: length of extent */ | |
202 | int *stat) /* output: success/failure */ | |
203 | { | |
204 | union xfs_btree_rec *rec; | |
205 | int error; | |
206 | ||
207 | error = xfs_btree_get_rec(cur, &rec, stat); | |
208 | if (!error && *stat == 1) { | |
209 | *bno = be32_to_cpu(rec->alloc.ar_startblock); | |
210 | *len = be32_to_cpu(rec->alloc.ar_blockcount); | |
211 | } | |
212 | return error; | |
213 | } | |
214 | ||
1da177e4 LT |
215 | /* |
216 | * Compute aligned version of the found extent. | |
217 | * Takes alignment and min length into account. | |
218 | */ | |
12375c82 | 219 | STATIC void |
1da177e4 | 220 | xfs_alloc_compute_aligned( |
86fa8af6 | 221 | xfs_alloc_arg_t *args, /* allocation argument structure */ |
1da177e4 LT |
222 | xfs_agblock_t foundbno, /* starting block in found extent */ |
223 | xfs_extlen_t foundlen, /* length in found extent */ | |
1da177e4 LT |
224 | xfs_agblock_t *resbno, /* result block number */ |
225 | xfs_extlen_t *reslen) /* result length */ | |
226 | { | |
227 | xfs_agblock_t bno; | |
1da177e4 | 228 | xfs_extlen_t len; |
bfe46d4e | 229 | xfs_extlen_t diff; |
1da177e4 | 230 | |
e26f0501 | 231 | /* Trim busy sections out of found extent */ |
4ecbfe63 | 232 | xfs_extent_busy_trim(args, foundbno, foundlen, &bno, &len); |
e26f0501 | 233 | |
bfe46d4e BF |
234 | /* |
235 | * If we have a largish extent that happens to start before min_agbno, | |
236 | * see if we can shift it into range... | |
237 | */ | |
238 | if (bno < args->min_agbno && bno + len > args->min_agbno) { | |
239 | diff = args->min_agbno - bno; | |
240 | if (len > diff) { | |
241 | bno += diff; | |
242 | len -= diff; | |
243 | } | |
244 | } | |
245 | ||
e26f0501 CH |
246 | if (args->alignment > 1 && len >= args->minlen) { |
247 | xfs_agblock_t aligned_bno = roundup(bno, args->alignment); | |
bfe46d4e BF |
248 | |
249 | diff = aligned_bno - bno; | |
e26f0501 CH |
250 | |
251 | *resbno = aligned_bno; | |
252 | *reslen = diff >= len ? 0 : len - diff; | |
1da177e4 | 253 | } else { |
e26f0501 CH |
254 | *resbno = bno; |
255 | *reslen = len; | |
1da177e4 | 256 | } |
1da177e4 LT |
257 | } |
258 | ||
259 | /* | |
260 | * Compute best start block and diff for "near" allocations. | |
261 | * freelen >= wantlen already checked by caller. | |
262 | */ | |
263 | STATIC xfs_extlen_t /* difference value (absolute) */ | |
264 | xfs_alloc_compute_diff( | |
265 | xfs_agblock_t wantbno, /* target starting block */ | |
266 | xfs_extlen_t wantlen, /* target length */ | |
267 | xfs_extlen_t alignment, /* target alignment */ | |
211d022c | 268 | char userdata, /* are we allocating data? */ |
1da177e4 LT |
269 | xfs_agblock_t freebno, /* freespace's starting block */ |
270 | xfs_extlen_t freelen, /* freespace's length */ | |
271 | xfs_agblock_t *newbnop) /* result: best start block from free */ | |
272 | { | |
273 | xfs_agblock_t freeend; /* end of freespace extent */ | |
274 | xfs_agblock_t newbno1; /* return block number */ | |
275 | xfs_agblock_t newbno2; /* other new block number */ | |
276 | xfs_extlen_t newlen1=0; /* length with newbno1 */ | |
277 | xfs_extlen_t newlen2=0; /* length with newbno2 */ | |
278 | xfs_agblock_t wantend; /* end of target extent */ | |
279 | ||
280 | ASSERT(freelen >= wantlen); | |
281 | freeend = freebno + freelen; | |
282 | wantend = wantbno + wantlen; | |
211d022c JK |
283 | /* |
284 | * We want to allocate from the start of a free extent if it is past | |
285 | * the desired block or if we are allocating user data and the free | |
286 | * extent is before desired block. The second case is there to allow | |
287 | * for contiguous allocation from the remaining free space if the file | |
288 | * grows in the short term. | |
289 | */ | |
290 | if (freebno >= wantbno || (userdata && freeend < wantend)) { | |
1da177e4 LT |
291 | if ((newbno1 = roundup(freebno, alignment)) >= freeend) |
292 | newbno1 = NULLAGBLOCK; | |
293 | } else if (freeend >= wantend && alignment > 1) { | |
294 | newbno1 = roundup(wantbno, alignment); | |
295 | newbno2 = newbno1 - alignment; | |
296 | if (newbno1 >= freeend) | |
297 | newbno1 = NULLAGBLOCK; | |
298 | else | |
299 | newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1); | |
300 | if (newbno2 < freebno) | |
301 | newbno2 = NULLAGBLOCK; | |
302 | else | |
303 | newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2); | |
304 | if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) { | |
305 | if (newlen1 < newlen2 || | |
306 | (newlen1 == newlen2 && | |
307 | XFS_ABSDIFF(newbno1, wantbno) > | |
308 | XFS_ABSDIFF(newbno2, wantbno))) | |
309 | newbno1 = newbno2; | |
310 | } else if (newbno2 != NULLAGBLOCK) | |
311 | newbno1 = newbno2; | |
312 | } else if (freeend >= wantend) { | |
313 | newbno1 = wantbno; | |
314 | } else if (alignment > 1) { | |
315 | newbno1 = roundup(freeend - wantlen, alignment); | |
316 | if (newbno1 > freeend - wantlen && | |
317 | newbno1 - alignment >= freebno) | |
318 | newbno1 -= alignment; | |
319 | else if (newbno1 >= freeend) | |
320 | newbno1 = NULLAGBLOCK; | |
321 | } else | |
322 | newbno1 = freeend - wantlen; | |
323 | *newbnop = newbno1; | |
324 | return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno); | |
325 | } | |
326 | ||
327 | /* | |
328 | * Fix up the length, based on mod and prod. | |
329 | * len should be k * prod + mod for some k. | |
330 | * If len is too small it is returned unchanged. | |
331 | * If len hits maxlen it is left alone. | |
332 | */ | |
333 | STATIC void | |
334 | xfs_alloc_fix_len( | |
335 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
336 | { | |
337 | xfs_extlen_t k; | |
338 | xfs_extlen_t rlen; | |
339 | ||
340 | ASSERT(args->mod < args->prod); | |
341 | rlen = args->len; | |
342 | ASSERT(rlen >= args->minlen); | |
343 | ASSERT(rlen <= args->maxlen); | |
344 | if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen || | |
345 | (args->mod == 0 && rlen < args->prod)) | |
346 | return; | |
347 | k = rlen % args->prod; | |
348 | if (k == args->mod) | |
349 | return; | |
30265117 JK |
350 | if (k > args->mod) |
351 | rlen = rlen - (k - args->mod); | |
352 | else | |
353 | rlen = rlen - args->prod + (args->mod - k); | |
3790a8cd | 354 | /* casts to (int) catch length underflows */ |
30265117 JK |
355 | if ((int)rlen < (int)args->minlen) |
356 | return; | |
357 | ASSERT(rlen >= args->minlen && rlen <= args->maxlen); | |
358 | ASSERT(rlen % args->prod == args->mod); | |
1da177e4 LT |
359 | args->len = rlen; |
360 | } | |
361 | ||
362 | /* | |
363 | * Fix up length if there is too little space left in the a.g. | |
364 | * Return 1 if ok, 0 if too little, should give up. | |
365 | */ | |
366 | STATIC int | |
367 | xfs_alloc_fix_minleft( | |
368 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
369 | { | |
370 | xfs_agf_t *agf; /* a.g. freelist header */ | |
371 | int diff; /* free space difference */ | |
372 | ||
373 | if (args->minleft == 0) | |
374 | return 1; | |
375 | agf = XFS_BUF_TO_AGF(args->agbp); | |
16259e7d | 376 | diff = be32_to_cpu(agf->agf_freeblks) |
1da177e4 LT |
377 | - args->len - args->minleft; |
378 | if (diff >= 0) | |
379 | return 1; | |
380 | args->len += diff; /* shrink the allocated space */ | |
3790a8cd DC |
381 | /* casts to (int) catch length underflows */ |
382 | if ((int)args->len >= (int)args->minlen) | |
1da177e4 LT |
383 | return 1; |
384 | args->agbno = NULLAGBLOCK; | |
385 | return 0; | |
386 | } | |
387 | ||
388 | /* | |
389 | * Update the two btrees, logically removing from freespace the extent | |
390 | * starting at rbno, rlen blocks. The extent is contained within the | |
391 | * actual (current) free extent fbno for flen blocks. | |
392 | * Flags are passed in indicating whether the cursors are set to the | |
393 | * relevant records. | |
394 | */ | |
395 | STATIC int /* error code */ | |
396 | xfs_alloc_fixup_trees( | |
397 | xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */ | |
398 | xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */ | |
399 | xfs_agblock_t fbno, /* starting block of free extent */ | |
400 | xfs_extlen_t flen, /* length of free extent */ | |
401 | xfs_agblock_t rbno, /* starting block of returned extent */ | |
402 | xfs_extlen_t rlen, /* length of returned extent */ | |
403 | int flags) /* flags, XFSA_FIXUP_... */ | |
404 | { | |
405 | int error; /* error code */ | |
406 | int i; /* operation results */ | |
407 | xfs_agblock_t nfbno1; /* first new free startblock */ | |
408 | xfs_agblock_t nfbno2; /* second new free startblock */ | |
409 | xfs_extlen_t nflen1=0; /* first new free length */ | |
410 | xfs_extlen_t nflen2=0; /* second new free length */ | |
5fb5aeee ES |
411 | struct xfs_mount *mp; |
412 | ||
413 | mp = cnt_cur->bc_mp; | |
1da177e4 LT |
414 | |
415 | /* | |
416 | * Look up the record in the by-size tree if necessary. | |
417 | */ | |
418 | if (flags & XFSA_FIXUP_CNT_OK) { | |
419 | #ifdef DEBUG | |
420 | if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i))) | |
421 | return error; | |
5fb5aeee | 422 | XFS_WANT_CORRUPTED_RETURN(mp, |
1da177e4 LT |
423 | i == 1 && nfbno1 == fbno && nflen1 == flen); |
424 | #endif | |
425 | } else { | |
426 | if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i))) | |
427 | return error; | |
5fb5aeee | 428 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
429 | } |
430 | /* | |
431 | * Look up the record in the by-block tree if necessary. | |
432 | */ | |
433 | if (flags & XFSA_FIXUP_BNO_OK) { | |
434 | #ifdef DEBUG | |
435 | if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i))) | |
436 | return error; | |
5fb5aeee | 437 | XFS_WANT_CORRUPTED_RETURN(mp, |
1da177e4 LT |
438 | i == 1 && nfbno1 == fbno && nflen1 == flen); |
439 | #endif | |
440 | } else { | |
441 | if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i))) | |
442 | return error; | |
5fb5aeee | 443 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 | 444 | } |
7cc95a82 | 445 | |
1da177e4 | 446 | #ifdef DEBUG |
7cc95a82 CH |
447 | if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) { |
448 | struct xfs_btree_block *bnoblock; | |
449 | struct xfs_btree_block *cntblock; | |
450 | ||
451 | bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]); | |
452 | cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]); | |
1da177e4 | 453 | |
5fb5aeee | 454 | XFS_WANT_CORRUPTED_RETURN(mp, |
7cc95a82 | 455 | bnoblock->bb_numrecs == cntblock->bb_numrecs); |
1da177e4 LT |
456 | } |
457 | #endif | |
7cc95a82 | 458 | |
1da177e4 LT |
459 | /* |
460 | * Deal with all four cases: the allocated record is contained | |
461 | * within the freespace record, so we can have new freespace | |
462 | * at either (or both) end, or no freespace remaining. | |
463 | */ | |
464 | if (rbno == fbno && rlen == flen) | |
465 | nfbno1 = nfbno2 = NULLAGBLOCK; | |
466 | else if (rbno == fbno) { | |
467 | nfbno1 = rbno + rlen; | |
468 | nflen1 = flen - rlen; | |
469 | nfbno2 = NULLAGBLOCK; | |
470 | } else if (rbno + rlen == fbno + flen) { | |
471 | nfbno1 = fbno; | |
472 | nflen1 = flen - rlen; | |
473 | nfbno2 = NULLAGBLOCK; | |
474 | } else { | |
475 | nfbno1 = fbno; | |
476 | nflen1 = rbno - fbno; | |
477 | nfbno2 = rbno + rlen; | |
478 | nflen2 = (fbno + flen) - nfbno2; | |
479 | } | |
480 | /* | |
481 | * Delete the entry from the by-size btree. | |
482 | */ | |
91cca5df | 483 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 484 | return error; |
5fb5aeee | 485 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
486 | /* |
487 | * Add new by-size btree entry(s). | |
488 | */ | |
489 | if (nfbno1 != NULLAGBLOCK) { | |
490 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i))) | |
491 | return error; | |
5fb5aeee | 492 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
4b22a571 | 493 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 494 | return error; |
5fb5aeee | 495 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
496 | } |
497 | if (nfbno2 != NULLAGBLOCK) { | |
498 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i))) | |
499 | return error; | |
5fb5aeee | 500 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
4b22a571 | 501 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 502 | return error; |
5fb5aeee | 503 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
504 | } |
505 | /* | |
506 | * Fix up the by-block btree entry(s). | |
507 | */ | |
508 | if (nfbno1 == NULLAGBLOCK) { | |
509 | /* | |
510 | * No remaining freespace, just delete the by-block tree entry. | |
511 | */ | |
91cca5df | 512 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 | 513 | return error; |
5fb5aeee | 514 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
515 | } else { |
516 | /* | |
517 | * Update the by-block entry to start later|be shorter. | |
518 | */ | |
519 | if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1))) | |
520 | return error; | |
521 | } | |
522 | if (nfbno2 != NULLAGBLOCK) { | |
523 | /* | |
524 | * 2 resulting free entries, need to add one. | |
525 | */ | |
526 | if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i))) | |
527 | return error; | |
5fb5aeee | 528 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
4b22a571 | 529 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 | 530 | return error; |
5fb5aeee | 531 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
532 | } |
533 | return 0; | |
534 | } | |
535 | ||
77c95bba | 536 | static bool |
612cfbfe | 537 | xfs_agfl_verify( |
bb80c6d7 DC |
538 | struct xfs_buf *bp) |
539 | { | |
bb80c6d7 DC |
540 | struct xfs_mount *mp = bp->b_target->bt_mount; |
541 | struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp); | |
bb80c6d7 DC |
542 | int i; |
543 | ||
ce748eaa | 544 | if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid)) |
77c95bba CH |
545 | return false; |
546 | if (be32_to_cpu(agfl->agfl_magicnum) != XFS_AGFL_MAGIC) | |
547 | return false; | |
548 | /* | |
549 | * during growfs operations, the perag is not fully initialised, | |
550 | * so we can't use it for any useful checking. growfs ensures we can't | |
551 | * use it by using uncached buffers that don't have the perag attached | |
552 | * so we can detect and avoid this problem. | |
553 | */ | |
554 | if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno) | |
555 | return false; | |
556 | ||
bb80c6d7 | 557 | for (i = 0; i < XFS_AGFL_SIZE(mp); i++) { |
77c95bba | 558 | if (be32_to_cpu(agfl->agfl_bno[i]) != NULLAGBLOCK && |
bb80c6d7 | 559 | be32_to_cpu(agfl->agfl_bno[i]) >= mp->m_sb.sb_agblocks) |
77c95bba | 560 | return false; |
bb80c6d7 | 561 | } |
a45086e2 BF |
562 | |
563 | return xfs_log_check_lsn(mp, | |
564 | be64_to_cpu(XFS_BUF_TO_AGFL(bp)->agfl_lsn)); | |
77c95bba CH |
565 | } |
566 | ||
567 | static void | |
568 | xfs_agfl_read_verify( | |
569 | struct xfs_buf *bp) | |
570 | { | |
571 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
77c95bba CH |
572 | |
573 | /* | |
574 | * There is no verification of non-crc AGFLs because mkfs does not | |
575 | * initialise the AGFL to zero or NULL. Hence the only valid part of the | |
576 | * AGFL is what the AGF says is active. We can't get to the AGF, so we | |
577 | * can't verify just those entries are valid. | |
578 | */ | |
579 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
580 | return; | |
581 | ||
ce5028cf | 582 | if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF)) |
2451337d | 583 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf | 584 | else if (!xfs_agfl_verify(bp)) |
2451337d | 585 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
586 | |
587 | if (bp->b_error) | |
588 | xfs_verifier_error(bp); | |
612cfbfe DC |
589 | } |
590 | ||
1813dd64 | 591 | static void |
612cfbfe DC |
592 | xfs_agfl_write_verify( |
593 | struct xfs_buf *bp) | |
594 | { | |
77c95bba CH |
595 | struct xfs_mount *mp = bp->b_target->bt_mount; |
596 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
612cfbfe | 597 | |
77c95bba CH |
598 | /* no verification of non-crc AGFLs */ |
599 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
600 | return; | |
601 | ||
602 | if (!xfs_agfl_verify(bp)) { | |
2451337d | 603 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 604 | xfs_verifier_error(bp); |
77c95bba CH |
605 | return; |
606 | } | |
607 | ||
608 | if (bip) | |
609 | XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
610 | ||
f1dbcd7e | 611 | xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF); |
bb80c6d7 DC |
612 | } |
613 | ||
1813dd64 | 614 | const struct xfs_buf_ops xfs_agfl_buf_ops = { |
233135b7 | 615 | .name = "xfs_agfl", |
1813dd64 DC |
616 | .verify_read = xfs_agfl_read_verify, |
617 | .verify_write = xfs_agfl_write_verify, | |
618 | }; | |
619 | ||
1da177e4 LT |
620 | /* |
621 | * Read in the allocation group free block array. | |
622 | */ | |
623 | STATIC int /* error */ | |
624 | xfs_alloc_read_agfl( | |
625 | xfs_mount_t *mp, /* mount point structure */ | |
626 | xfs_trans_t *tp, /* transaction pointer */ | |
627 | xfs_agnumber_t agno, /* allocation group number */ | |
628 | xfs_buf_t **bpp) /* buffer for the ag free block array */ | |
629 | { | |
630 | xfs_buf_t *bp; /* return value */ | |
631 | int error; | |
632 | ||
633 | ASSERT(agno != NULLAGNUMBER); | |
634 | error = xfs_trans_read_buf( | |
635 | mp, tp, mp->m_ddev_targp, | |
636 | XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)), | |
1813dd64 | 637 | XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops); |
1da177e4 LT |
638 | if (error) |
639 | return error; | |
38f23232 | 640 | xfs_buf_set_ref(bp, XFS_AGFL_REF); |
1da177e4 LT |
641 | *bpp = bp; |
642 | return 0; | |
643 | } | |
644 | ||
ecb6928f CH |
645 | STATIC int |
646 | xfs_alloc_update_counters( | |
647 | struct xfs_trans *tp, | |
648 | struct xfs_perag *pag, | |
649 | struct xfs_buf *agbp, | |
650 | long len) | |
651 | { | |
652 | struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); | |
653 | ||
654 | pag->pagf_freeblks += len; | |
655 | be32_add_cpu(&agf->agf_freeblks, len); | |
656 | ||
657 | xfs_trans_agblocks_delta(tp, len); | |
658 | if (unlikely(be32_to_cpu(agf->agf_freeblks) > | |
659 | be32_to_cpu(agf->agf_length))) | |
2451337d | 660 | return -EFSCORRUPTED; |
ecb6928f CH |
661 | |
662 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS); | |
663 | return 0; | |
664 | } | |
665 | ||
1da177e4 LT |
666 | /* |
667 | * Allocation group level functions. | |
668 | */ | |
669 | ||
670 | /* | |
671 | * Allocate a variable extent in the allocation group agno. | |
672 | * Type and bno are used to determine where in the allocation group the | |
673 | * extent will start. | |
674 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
675 | * and of the form k * prod + mod unless there's nothing that large. | |
676 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
677 | */ | |
678 | STATIC int /* error */ | |
679 | xfs_alloc_ag_vextent( | |
680 | xfs_alloc_arg_t *args) /* argument structure for allocation */ | |
681 | { | |
682 | int error=0; | |
1da177e4 LT |
683 | |
684 | ASSERT(args->minlen > 0); | |
685 | ASSERT(args->maxlen > 0); | |
686 | ASSERT(args->minlen <= args->maxlen); | |
687 | ASSERT(args->mod < args->prod); | |
688 | ASSERT(args->alignment > 0); | |
689 | /* | |
690 | * Branch to correct routine based on the type. | |
691 | */ | |
692 | args->wasfromfl = 0; | |
693 | switch (args->type) { | |
694 | case XFS_ALLOCTYPE_THIS_AG: | |
695 | error = xfs_alloc_ag_vextent_size(args); | |
696 | break; | |
697 | case XFS_ALLOCTYPE_NEAR_BNO: | |
698 | error = xfs_alloc_ag_vextent_near(args); | |
699 | break; | |
700 | case XFS_ALLOCTYPE_THIS_BNO: | |
701 | error = xfs_alloc_ag_vextent_exact(args); | |
702 | break; | |
703 | default: | |
704 | ASSERT(0); | |
705 | /* NOTREACHED */ | |
706 | } | |
ecb6928f CH |
707 | |
708 | if (error || args->agbno == NULLAGBLOCK) | |
1da177e4 | 709 | return error; |
ecb6928f CH |
710 | |
711 | ASSERT(args->len >= args->minlen); | |
712 | ASSERT(args->len <= args->maxlen); | |
713 | ASSERT(!args->wasfromfl || !args->isfl); | |
714 | ASSERT(args->agbno % args->alignment == 0); | |
715 | ||
673930c3 DW |
716 | /* if not file data, insert new block into the reverse map btree */ |
717 | if (args->oinfo.oi_owner != XFS_RMAP_OWN_UNKNOWN) { | |
718 | error = xfs_rmap_alloc(args->tp, args->agbp, args->agno, | |
719 | args->agbno, args->len, &args->oinfo); | |
720 | if (error) | |
721 | return error; | |
722 | } | |
723 | ||
ecb6928f CH |
724 | if (!args->wasfromfl) { |
725 | error = xfs_alloc_update_counters(args->tp, args->pag, | |
726 | args->agbp, | |
727 | -((long)(args->len))); | |
728 | if (error) | |
729 | return error; | |
730 | ||
4ecbfe63 | 731 | ASSERT(!xfs_extent_busy_search(args->mp, args->agno, |
e26f0501 | 732 | args->agbno, args->len)); |
1da177e4 | 733 | } |
ecb6928f CH |
734 | |
735 | if (!args->isfl) { | |
736 | xfs_trans_mod_sb(args->tp, args->wasdel ? | |
737 | XFS_TRANS_SB_RES_FDBLOCKS : | |
738 | XFS_TRANS_SB_FDBLOCKS, | |
739 | -((long)(args->len))); | |
740 | } | |
741 | ||
ff6d6af2 BD |
742 | XFS_STATS_INC(args->mp, xs_allocx); |
743 | XFS_STATS_ADD(args->mp, xs_allocb, args->len); | |
ecb6928f | 744 | return error; |
1da177e4 LT |
745 | } |
746 | ||
747 | /* | |
748 | * Allocate a variable extent at exactly agno/bno. | |
749 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
750 | * and of the form k * prod + mod unless there's nothing that large. | |
751 | * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it. | |
752 | */ | |
753 | STATIC int /* error */ | |
754 | xfs_alloc_ag_vextent_exact( | |
755 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
756 | { | |
757 | xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */ | |
758 | xfs_btree_cur_t *cnt_cur;/* by count btree cursor */ | |
1da177e4 LT |
759 | int error; |
760 | xfs_agblock_t fbno; /* start block of found extent */ | |
1da177e4 | 761 | xfs_extlen_t flen; /* length of found extent */ |
e26f0501 CH |
762 | xfs_agblock_t tbno; /* start block of trimmed extent */ |
763 | xfs_extlen_t tlen; /* length of trimmed extent */ | |
764 | xfs_agblock_t tend; /* end block of trimmed extent */ | |
1da177e4 | 765 | int i; /* success/failure of operation */ |
1da177e4 LT |
766 | |
767 | ASSERT(args->alignment == 1); | |
9f9baab3 | 768 | |
1da177e4 LT |
769 | /* |
770 | * Allocate/initialize a cursor for the by-number freespace btree. | |
771 | */ | |
561f7d17 | 772 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
9f9baab3 CH |
773 | args->agno, XFS_BTNUM_BNO); |
774 | ||
1da177e4 LT |
775 | /* |
776 | * Lookup bno and minlen in the btree (minlen is irrelevant, really). | |
777 | * Look for the closest free block <= bno, it must contain bno | |
778 | * if any free block does. | |
779 | */ | |
9f9baab3 CH |
780 | error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i); |
781 | if (error) | |
1da177e4 | 782 | goto error0; |
9f9baab3 CH |
783 | if (!i) |
784 | goto not_found; | |
785 | ||
1da177e4 LT |
786 | /* |
787 | * Grab the freespace record. | |
788 | */ | |
9f9baab3 CH |
789 | error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i); |
790 | if (error) | |
1da177e4 | 791 | goto error0; |
c29aad41 | 792 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 | 793 | ASSERT(fbno <= args->agbno); |
9f9baab3 | 794 | |
1da177e4 | 795 | /* |
e26f0501 | 796 | * Check for overlapping busy extents. |
1da177e4 | 797 | */ |
4ecbfe63 | 798 | xfs_extent_busy_trim(args, fbno, flen, &tbno, &tlen); |
e26f0501 CH |
799 | |
800 | /* | |
801 | * Give up if the start of the extent is busy, or the freespace isn't | |
802 | * long enough for the minimum request. | |
803 | */ | |
804 | if (tbno > args->agbno) | |
805 | goto not_found; | |
806 | if (tlen < args->minlen) | |
807 | goto not_found; | |
808 | tend = tbno + tlen; | |
809 | if (tend < args->agbno + args->minlen) | |
9f9baab3 CH |
810 | goto not_found; |
811 | ||
1da177e4 LT |
812 | /* |
813 | * End of extent will be smaller of the freespace end and the | |
814 | * maximal requested end. | |
9f9baab3 | 815 | * |
1da177e4 LT |
816 | * Fix the length according to mod and prod if given. |
817 | */ | |
81463b1c CS |
818 | args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen) |
819 | - args->agbno; | |
1da177e4 | 820 | xfs_alloc_fix_len(args); |
9f9baab3 CH |
821 | if (!xfs_alloc_fix_minleft(args)) |
822 | goto not_found; | |
823 | ||
81463b1c | 824 | ASSERT(args->agbno + args->len <= tend); |
9f9baab3 | 825 | |
1da177e4 | 826 | /* |
81463b1c | 827 | * We are allocating agbno for args->len |
1da177e4 LT |
828 | * Allocate/initialize a cursor for the by-size btree. |
829 | */ | |
561f7d17 CH |
830 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
831 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 832 | ASSERT(args->agbno + args->len <= |
16259e7d | 833 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
9f9baab3 CH |
834 | error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno, |
835 | args->len, XFSA_FIXUP_BNO_OK); | |
836 | if (error) { | |
1da177e4 LT |
837 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); |
838 | goto error0; | |
839 | } | |
9f9baab3 | 840 | |
1da177e4 LT |
841 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); |
842 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 843 | |
1da177e4 | 844 | args->wasfromfl = 0; |
9f9baab3 CH |
845 | trace_xfs_alloc_exact_done(args); |
846 | return 0; | |
847 | ||
848 | not_found: | |
849 | /* Didn't find it, return null. */ | |
850 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
851 | args->agbno = NULLAGBLOCK; | |
852 | trace_xfs_alloc_exact_notfound(args); | |
1da177e4 LT |
853 | return 0; |
854 | ||
855 | error0: | |
856 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
0b1b213f | 857 | trace_xfs_alloc_exact_error(args); |
1da177e4 LT |
858 | return error; |
859 | } | |
860 | ||
489a150f CH |
861 | /* |
862 | * Search the btree in a given direction via the search cursor and compare | |
863 | * the records found against the good extent we've already found. | |
864 | */ | |
865 | STATIC int | |
866 | xfs_alloc_find_best_extent( | |
867 | struct xfs_alloc_arg *args, /* allocation argument structure */ | |
868 | struct xfs_btree_cur **gcur, /* good cursor */ | |
869 | struct xfs_btree_cur **scur, /* searching cursor */ | |
870 | xfs_agblock_t gdiff, /* difference for search comparison */ | |
871 | xfs_agblock_t *sbno, /* extent found by search */ | |
e26f0501 CH |
872 | xfs_extlen_t *slen, /* extent length */ |
873 | xfs_agblock_t *sbnoa, /* aligned extent found by search */ | |
874 | xfs_extlen_t *slena, /* aligned extent length */ | |
489a150f CH |
875 | int dir) /* 0 = search right, 1 = search left */ |
876 | { | |
489a150f CH |
877 | xfs_agblock_t new; |
878 | xfs_agblock_t sdiff; | |
879 | int error; | |
880 | int i; | |
881 | ||
882 | /* The good extent is perfect, no need to search. */ | |
883 | if (!gdiff) | |
884 | goto out_use_good; | |
885 | ||
886 | /* | |
887 | * Look until we find a better one, run out of space or run off the end. | |
888 | */ | |
889 | do { | |
890 | error = xfs_alloc_get_rec(*scur, sbno, slen, &i); | |
891 | if (error) | |
892 | goto error0; | |
c29aad41 | 893 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
e26f0501 | 894 | xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena); |
489a150f CH |
895 | |
896 | /* | |
897 | * The good extent is closer than this one. | |
898 | */ | |
899 | if (!dir) { | |
bfe46d4e BF |
900 | if (*sbnoa > args->max_agbno) |
901 | goto out_use_good; | |
e26f0501 | 902 | if (*sbnoa >= args->agbno + gdiff) |
489a150f CH |
903 | goto out_use_good; |
904 | } else { | |
bfe46d4e BF |
905 | if (*sbnoa < args->min_agbno) |
906 | goto out_use_good; | |
e26f0501 | 907 | if (*sbnoa <= args->agbno - gdiff) |
489a150f CH |
908 | goto out_use_good; |
909 | } | |
910 | ||
911 | /* | |
912 | * Same distance, compare length and pick the best. | |
913 | */ | |
914 | if (*slena >= args->minlen) { | |
915 | args->len = XFS_EXTLEN_MIN(*slena, args->maxlen); | |
916 | xfs_alloc_fix_len(args); | |
917 | ||
918 | sdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
211d022c JK |
919 | args->alignment, |
920 | args->userdata, *sbnoa, | |
e26f0501 | 921 | *slena, &new); |
489a150f CH |
922 | |
923 | /* | |
924 | * Choose closer size and invalidate other cursor. | |
925 | */ | |
926 | if (sdiff < gdiff) | |
927 | goto out_use_search; | |
928 | goto out_use_good; | |
929 | } | |
930 | ||
931 | if (!dir) | |
932 | error = xfs_btree_increment(*scur, 0, &i); | |
933 | else | |
934 | error = xfs_btree_decrement(*scur, 0, &i); | |
935 | if (error) | |
936 | goto error0; | |
937 | } while (i); | |
938 | ||
939 | out_use_good: | |
940 | xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR); | |
941 | *scur = NULL; | |
942 | return 0; | |
943 | ||
944 | out_use_search: | |
945 | xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR); | |
946 | *gcur = NULL; | |
947 | return 0; | |
948 | ||
949 | error0: | |
950 | /* caller invalidates cursors */ | |
951 | return error; | |
952 | } | |
953 | ||
1da177e4 LT |
954 | /* |
955 | * Allocate a variable extent near bno in the allocation group agno. | |
956 | * Extent's length (returned in len) will be between minlen and maxlen, | |
957 | * and of the form k * prod + mod unless there's nothing that large. | |
958 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
959 | */ | |
960 | STATIC int /* error */ | |
961 | xfs_alloc_ag_vextent_near( | |
962 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
963 | { | |
964 | xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */ | |
965 | xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */ | |
966 | xfs_btree_cur_t *cnt_cur; /* cursor for count btree */ | |
1da177e4 LT |
967 | xfs_agblock_t gtbno; /* start bno of right side entry */ |
968 | xfs_agblock_t gtbnoa; /* aligned ... */ | |
969 | xfs_extlen_t gtdiff; /* difference to right side entry */ | |
970 | xfs_extlen_t gtlen; /* length of right side entry */ | |
e26f0501 | 971 | xfs_extlen_t gtlena; /* aligned ... */ |
1da177e4 LT |
972 | xfs_agblock_t gtnew; /* useful start bno of right side */ |
973 | int error; /* error code */ | |
974 | int i; /* result code, temporary */ | |
975 | int j; /* result code, temporary */ | |
976 | xfs_agblock_t ltbno; /* start bno of left side entry */ | |
977 | xfs_agblock_t ltbnoa; /* aligned ... */ | |
978 | xfs_extlen_t ltdiff; /* difference to left side entry */ | |
1da177e4 | 979 | xfs_extlen_t ltlen; /* length of left side entry */ |
e26f0501 | 980 | xfs_extlen_t ltlena; /* aligned ... */ |
1da177e4 LT |
981 | xfs_agblock_t ltnew; /* useful start bno of left side */ |
982 | xfs_extlen_t rlen; /* length of returned extent */ | |
e26f0501 | 983 | int forced = 0; |
63d20d6e | 984 | #ifdef DEBUG |
1da177e4 LT |
985 | /* |
986 | * Randomly don't execute the first algorithm. | |
987 | */ | |
988 | int dofirst; /* set to do first algorithm */ | |
989 | ||
ecb3403d | 990 | dofirst = prandom_u32() & 1; |
1da177e4 | 991 | #endif |
e26f0501 | 992 | |
bfe46d4e BF |
993 | /* handle unitialized agbno range so caller doesn't have to */ |
994 | if (!args->min_agbno && !args->max_agbno) | |
995 | args->max_agbno = args->mp->m_sb.sb_agblocks - 1; | |
996 | ASSERT(args->min_agbno <= args->max_agbno); | |
997 | ||
998 | /* clamp agbno to the range if it's outside */ | |
999 | if (args->agbno < args->min_agbno) | |
1000 | args->agbno = args->min_agbno; | |
1001 | if (args->agbno > args->max_agbno) | |
1002 | args->agbno = args->max_agbno; | |
1003 | ||
e26f0501 CH |
1004 | restart: |
1005 | bno_cur_lt = NULL; | |
1006 | bno_cur_gt = NULL; | |
1007 | ltlen = 0; | |
1008 | gtlena = 0; | |
1009 | ltlena = 0; | |
1010 | ||
1da177e4 LT |
1011 | /* |
1012 | * Get a cursor for the by-size btree. | |
1013 | */ | |
561f7d17 CH |
1014 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1015 | args->agno, XFS_BTNUM_CNT); | |
e26f0501 | 1016 | |
1da177e4 LT |
1017 | /* |
1018 | * See if there are any free extents as big as maxlen. | |
1019 | */ | |
1020 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i))) | |
1021 | goto error0; | |
1022 | /* | |
1023 | * If none, then pick up the last entry in the tree unless the | |
1024 | * tree is empty. | |
1025 | */ | |
1026 | if (!i) { | |
1027 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno, | |
1028 | <len, &i))) | |
1029 | goto error0; | |
1030 | if (i == 0 || ltlen == 0) { | |
1031 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
e26f0501 | 1032 | trace_xfs_alloc_near_noentry(args); |
1da177e4 LT |
1033 | return 0; |
1034 | } | |
1035 | ASSERT(i == 1); | |
1036 | } | |
1037 | args->wasfromfl = 0; | |
e26f0501 | 1038 | |
1da177e4 LT |
1039 | /* |
1040 | * First algorithm. | |
1041 | * If the requested extent is large wrt the freespaces available | |
1042 | * in this a.g., then the cursor will be pointing to a btree entry | |
1043 | * near the right edge of the tree. If it's in the last btree leaf | |
1044 | * block, then we just examine all the entries in that block | |
1045 | * that are big enough, and pick the best one. | |
1046 | * This is written as a while loop so we can break out of it, | |
1047 | * but we never loop back to the top. | |
1048 | */ | |
1049 | while (xfs_btree_islastblock(cnt_cur, 0)) { | |
1050 | xfs_extlen_t bdiff; | |
1051 | int besti=0; | |
1052 | xfs_extlen_t blen=0; | |
1053 | xfs_agblock_t bnew=0; | |
1054 | ||
63d20d6e DC |
1055 | #ifdef DEBUG |
1056 | if (dofirst) | |
1da177e4 LT |
1057 | break; |
1058 | #endif | |
1059 | /* | |
1060 | * Start from the entry that lookup found, sequence through | |
1061 | * all larger free blocks. If we're actually pointing at a | |
1062 | * record smaller than maxlen, go to the start of this block, | |
1063 | * and skip all those smaller than minlen. | |
1064 | */ | |
1065 | if (ltlen || args->alignment > 1) { | |
1066 | cnt_cur->bc_ptrs[0] = 1; | |
1067 | do { | |
1068 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, | |
1069 | <len, &i))) | |
1070 | goto error0; | |
c29aad41 | 1071 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 LT |
1072 | if (ltlen >= args->minlen) |
1073 | break; | |
637aa50f | 1074 | if ((error = xfs_btree_increment(cnt_cur, 0, &i))) |
1da177e4 LT |
1075 | goto error0; |
1076 | } while (i); | |
1077 | ASSERT(ltlen >= args->minlen); | |
1078 | if (!i) | |
1079 | break; | |
1080 | } | |
1081 | i = cnt_cur->bc_ptrs[0]; | |
1082 | for (j = 1, blen = 0, bdiff = 0; | |
1083 | !error && j && (blen < args->maxlen || bdiff > 0); | |
637aa50f | 1084 | error = xfs_btree_increment(cnt_cur, 0, &j)) { |
1da177e4 LT |
1085 | /* |
1086 | * For each entry, decide if it's better than | |
1087 | * the previous best entry. | |
1088 | */ | |
1089 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
1090 | goto error0; | |
c29aad41 | 1091 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
86fa8af6 CH |
1092 | xfs_alloc_compute_aligned(args, ltbno, ltlen, |
1093 | <bnoa, <lena); | |
e6430037 | 1094 | if (ltlena < args->minlen) |
1da177e4 | 1095 | continue; |
bfe46d4e BF |
1096 | if (ltbnoa < args->min_agbno || ltbnoa > args->max_agbno) |
1097 | continue; | |
1da177e4 LT |
1098 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); |
1099 | xfs_alloc_fix_len(args); | |
1100 | ASSERT(args->len >= args->minlen); | |
1101 | if (args->len < blen) | |
1102 | continue; | |
1103 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
211d022c JK |
1104 | args->alignment, args->userdata, ltbnoa, |
1105 | ltlena, <new); | |
1da177e4 LT |
1106 | if (ltnew != NULLAGBLOCK && |
1107 | (args->len > blen || ltdiff < bdiff)) { | |
1108 | bdiff = ltdiff; | |
1109 | bnew = ltnew; | |
1110 | blen = args->len; | |
1111 | besti = cnt_cur->bc_ptrs[0]; | |
1112 | } | |
1113 | } | |
1114 | /* | |
1115 | * It didn't work. We COULD be in a case where | |
1116 | * there's a good record somewhere, so try again. | |
1117 | */ | |
1118 | if (blen == 0) | |
1119 | break; | |
1120 | /* | |
1121 | * Point at the best entry, and retrieve it again. | |
1122 | */ | |
1123 | cnt_cur->bc_ptrs[0] = besti; | |
1124 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
1125 | goto error0; | |
c29aad41 | 1126 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
73523a2e | 1127 | ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
1128 | args->len = blen; |
1129 | if (!xfs_alloc_fix_minleft(args)) { | |
1130 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 1131 | trace_xfs_alloc_near_nominleft(args); |
1da177e4 LT |
1132 | return 0; |
1133 | } | |
1134 | blen = args->len; | |
1135 | /* | |
1136 | * We are allocating starting at bnew for blen blocks. | |
1137 | */ | |
1138 | args->agbno = bnew; | |
1139 | ASSERT(bnew >= ltbno); | |
73523a2e | 1140 | ASSERT(bnew + blen <= ltbno + ltlen); |
1da177e4 LT |
1141 | /* |
1142 | * Set up a cursor for the by-bno tree. | |
1143 | */ | |
561f7d17 CH |
1144 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, |
1145 | args->agbp, args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1146 | /* |
1147 | * Fix up the btree entries. | |
1148 | */ | |
1149 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, | |
1150 | ltlen, bnew, blen, XFSA_FIXUP_CNT_OK))) | |
1151 | goto error0; | |
1152 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1153 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
0b1b213f CH |
1154 | |
1155 | trace_xfs_alloc_near_first(args); | |
1da177e4 LT |
1156 | return 0; |
1157 | } | |
1158 | /* | |
1159 | * Second algorithm. | |
1160 | * Search in the by-bno tree to the left and to the right | |
1161 | * simultaneously, until in each case we find a space big enough, | |
1162 | * or run into the edge of the tree. When we run into the edge, | |
1163 | * we deallocate that cursor. | |
1164 | * If both searches succeed, we compare the two spaces and pick | |
1165 | * the better one. | |
1166 | * With alignment, it's possible for both to fail; the upper | |
1167 | * level algorithm that picks allocation groups for allocations | |
1168 | * is not supposed to do this. | |
1169 | */ | |
1170 | /* | |
1171 | * Allocate and initialize the cursor for the leftward search. | |
1172 | */ | |
561f7d17 CH |
1173 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1174 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1175 | /* |
1176 | * Lookup <= bno to find the leftward search's starting point. | |
1177 | */ | |
1178 | if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i))) | |
1179 | goto error0; | |
1180 | if (!i) { | |
1181 | /* | |
1182 | * Didn't find anything; use this cursor for the rightward | |
1183 | * search. | |
1184 | */ | |
1185 | bno_cur_gt = bno_cur_lt; | |
1186 | bno_cur_lt = NULL; | |
1187 | } | |
1188 | /* | |
1189 | * Found something. Duplicate the cursor for the rightward search. | |
1190 | */ | |
1191 | else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt))) | |
1192 | goto error0; | |
1193 | /* | |
1194 | * Increment the cursor, so we will point at the entry just right | |
1195 | * of the leftward entry if any, or to the leftmost entry. | |
1196 | */ | |
637aa50f | 1197 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
1198 | goto error0; |
1199 | if (!i) { | |
1200 | /* | |
1201 | * It failed, there are no rightward entries. | |
1202 | */ | |
1203 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR); | |
1204 | bno_cur_gt = NULL; | |
1205 | } | |
1206 | /* | |
1207 | * Loop going left with the leftward cursor, right with the | |
1208 | * rightward cursor, until either both directions give up or | |
1209 | * we find an entry at least as big as minlen. | |
1210 | */ | |
1211 | do { | |
1212 | if (bno_cur_lt) { | |
1213 | if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i))) | |
1214 | goto error0; | |
c29aad41 | 1215 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
86fa8af6 CH |
1216 | xfs_alloc_compute_aligned(args, ltbno, ltlen, |
1217 | <bnoa, <lena); | |
bfe46d4e | 1218 | if (ltlena >= args->minlen && ltbnoa >= args->min_agbno) |
1da177e4 | 1219 | break; |
8df4da4a | 1220 | if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i))) |
1da177e4 | 1221 | goto error0; |
bfe46d4e | 1222 | if (!i || ltbnoa < args->min_agbno) { |
1da177e4 LT |
1223 | xfs_btree_del_cursor(bno_cur_lt, |
1224 | XFS_BTREE_NOERROR); | |
1225 | bno_cur_lt = NULL; | |
1226 | } | |
1227 | } | |
1228 | if (bno_cur_gt) { | |
1229 | if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i))) | |
1230 | goto error0; | |
c29aad41 | 1231 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
86fa8af6 CH |
1232 | xfs_alloc_compute_aligned(args, gtbno, gtlen, |
1233 | >bnoa, >lena); | |
bfe46d4e | 1234 | if (gtlena >= args->minlen && gtbnoa <= args->max_agbno) |
1da177e4 | 1235 | break; |
637aa50f | 1236 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 | 1237 | goto error0; |
bfe46d4e | 1238 | if (!i || gtbnoa > args->max_agbno) { |
1da177e4 LT |
1239 | xfs_btree_del_cursor(bno_cur_gt, |
1240 | XFS_BTREE_NOERROR); | |
1241 | bno_cur_gt = NULL; | |
1242 | } | |
1243 | } | |
1244 | } while (bno_cur_lt || bno_cur_gt); | |
489a150f | 1245 | |
1da177e4 LT |
1246 | /* |
1247 | * Got both cursors still active, need to find better entry. | |
1248 | */ | |
1249 | if (bno_cur_lt && bno_cur_gt) { | |
1da177e4 LT |
1250 | if (ltlena >= args->minlen) { |
1251 | /* | |
489a150f | 1252 | * Left side is good, look for a right side entry. |
1da177e4 LT |
1253 | */ |
1254 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1255 | xfs_alloc_fix_len(args); | |
489a150f | 1256 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, |
211d022c JK |
1257 | args->alignment, args->userdata, ltbnoa, |
1258 | ltlena, <new); | |
489a150f CH |
1259 | |
1260 | error = xfs_alloc_find_best_extent(args, | |
1261 | &bno_cur_lt, &bno_cur_gt, | |
e26f0501 CH |
1262 | ltdiff, >bno, >len, |
1263 | >bnoa, >lena, | |
489a150f CH |
1264 | 0 /* search right */); |
1265 | } else { | |
1266 | ASSERT(gtlena >= args->minlen); | |
1267 | ||
1da177e4 | 1268 | /* |
489a150f | 1269 | * Right side is good, look for a left side entry. |
1da177e4 LT |
1270 | */ |
1271 | args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen); | |
1272 | xfs_alloc_fix_len(args); | |
489a150f | 1273 | gtdiff = xfs_alloc_compute_diff(args->agbno, args->len, |
211d022c JK |
1274 | args->alignment, args->userdata, gtbnoa, |
1275 | gtlena, >new); | |
489a150f CH |
1276 | |
1277 | error = xfs_alloc_find_best_extent(args, | |
1278 | &bno_cur_gt, &bno_cur_lt, | |
e26f0501 CH |
1279 | gtdiff, <bno, <len, |
1280 | <bnoa, <lena, | |
489a150f | 1281 | 1 /* search left */); |
1da177e4 | 1282 | } |
489a150f CH |
1283 | |
1284 | if (error) | |
1285 | goto error0; | |
1da177e4 | 1286 | } |
489a150f | 1287 | |
1da177e4 LT |
1288 | /* |
1289 | * If we couldn't get anything, give up. | |
1290 | */ | |
1291 | if (bno_cur_lt == NULL && bno_cur_gt == NULL) { | |
e3a746f5 DC |
1292 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1293 | ||
e26f0501 CH |
1294 | if (!forced++) { |
1295 | trace_xfs_alloc_near_busy(args); | |
1296 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1297 | goto restart; | |
1298 | } | |
0b1b213f | 1299 | trace_xfs_alloc_size_neither(args); |
1da177e4 LT |
1300 | args->agbno = NULLAGBLOCK; |
1301 | return 0; | |
1302 | } | |
489a150f | 1303 | |
1da177e4 LT |
1304 | /* |
1305 | * At this point we have selected a freespace entry, either to the | |
1306 | * left or to the right. If it's on the right, copy all the | |
1307 | * useful variables to the "left" set so we only have one | |
1308 | * copy of this code. | |
1309 | */ | |
1310 | if (bno_cur_gt) { | |
1311 | bno_cur_lt = bno_cur_gt; | |
1312 | bno_cur_gt = NULL; | |
1313 | ltbno = gtbno; | |
1314 | ltbnoa = gtbnoa; | |
1315 | ltlen = gtlen; | |
1316 | ltlena = gtlena; | |
1317 | j = 1; | |
1318 | } else | |
1319 | j = 0; | |
489a150f | 1320 | |
1da177e4 LT |
1321 | /* |
1322 | * Fix up the length and compute the useful address. | |
1323 | */ | |
1da177e4 LT |
1324 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); |
1325 | xfs_alloc_fix_len(args); | |
1326 | if (!xfs_alloc_fix_minleft(args)) { | |
0b1b213f | 1327 | trace_xfs_alloc_near_nominleft(args); |
1da177e4 LT |
1328 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); |
1329 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1330 | return 0; | |
1331 | } | |
1332 | rlen = args->len; | |
e26f0501 | 1333 | (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, |
211d022c | 1334 | args->userdata, ltbnoa, ltlena, <new); |
1da177e4 | 1335 | ASSERT(ltnew >= ltbno); |
e26f0501 | 1336 | ASSERT(ltnew + rlen <= ltbnoa + ltlena); |
16259e7d | 1337 | ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
bfe46d4e | 1338 | ASSERT(ltnew >= args->min_agbno && ltnew <= args->max_agbno); |
1da177e4 | 1339 | args->agbno = ltnew; |
e26f0501 | 1340 | |
1da177e4 LT |
1341 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen, |
1342 | ltnew, rlen, XFSA_FIXUP_BNO_OK))) | |
1343 | goto error0; | |
0b1b213f CH |
1344 | |
1345 | if (j) | |
1346 | trace_xfs_alloc_near_greater(args); | |
1347 | else | |
1348 | trace_xfs_alloc_near_lesser(args); | |
1349 | ||
1da177e4 LT |
1350 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1351 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1352 | return 0; | |
1353 | ||
1354 | error0: | |
0b1b213f | 1355 | trace_xfs_alloc_near_error(args); |
1da177e4 LT |
1356 | if (cnt_cur != NULL) |
1357 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1358 | if (bno_cur_lt != NULL) | |
1359 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR); | |
1360 | if (bno_cur_gt != NULL) | |
1361 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR); | |
1362 | return error; | |
1363 | } | |
1364 | ||
1365 | /* | |
1366 | * Allocate a variable extent anywhere in the allocation group agno. | |
1367 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1368 | * and of the form k * prod + mod unless there's nothing that large. | |
1369 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1370 | */ | |
1371 | STATIC int /* error */ | |
1372 | xfs_alloc_ag_vextent_size( | |
1373 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
1374 | { | |
1375 | xfs_btree_cur_t *bno_cur; /* cursor for bno btree */ | |
1376 | xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */ | |
1377 | int error; /* error result */ | |
1378 | xfs_agblock_t fbno; /* start of found freespace */ | |
1379 | xfs_extlen_t flen; /* length of found freespace */ | |
1da177e4 LT |
1380 | int i; /* temp status variable */ |
1381 | xfs_agblock_t rbno; /* returned block number */ | |
1382 | xfs_extlen_t rlen; /* length of returned extent */ | |
e26f0501 | 1383 | int forced = 0; |
1da177e4 | 1384 | |
e26f0501 | 1385 | restart: |
1da177e4 LT |
1386 | /* |
1387 | * Allocate and initialize a cursor for the by-size btree. | |
1388 | */ | |
561f7d17 CH |
1389 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1390 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 1391 | bno_cur = NULL; |
e26f0501 | 1392 | |
1da177e4 LT |
1393 | /* |
1394 | * Look for an entry >= maxlen+alignment-1 blocks. | |
1395 | */ | |
1396 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, | |
1397 | args->maxlen + args->alignment - 1, &i))) | |
1398 | goto error0; | |
e26f0501 | 1399 | |
1da177e4 | 1400 | /* |
e26f0501 CH |
1401 | * If none or we have busy extents that we cannot allocate from, then |
1402 | * we have to settle for a smaller extent. In the case that there are | |
1403 | * no large extents, this will return the last entry in the tree unless | |
1404 | * the tree is empty. In the case that there are only busy large | |
1405 | * extents, this will return the largest small extent unless there | |
1406 | * are no smaller extents available. | |
1da177e4 | 1407 | */ |
e26f0501 CH |
1408 | if (!i || forced > 1) { |
1409 | error = xfs_alloc_ag_vextent_small(args, cnt_cur, | |
1410 | &fbno, &flen, &i); | |
1411 | if (error) | |
1da177e4 LT |
1412 | goto error0; |
1413 | if (i == 0 || flen == 0) { | |
1414 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 1415 | trace_xfs_alloc_size_noentry(args); |
1da177e4 LT |
1416 | return 0; |
1417 | } | |
1418 | ASSERT(i == 1); | |
e26f0501 CH |
1419 | xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen); |
1420 | } else { | |
1421 | /* | |
1422 | * Search for a non-busy extent that is large enough. | |
1423 | * If we are at low space, don't check, or if we fall of | |
1424 | * the end of the btree, turn off the busy check and | |
1425 | * restart. | |
1426 | */ | |
1427 | for (;;) { | |
1428 | error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i); | |
1429 | if (error) | |
1430 | goto error0; | |
c29aad41 | 1431 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
e26f0501 CH |
1432 | |
1433 | xfs_alloc_compute_aligned(args, fbno, flen, | |
1434 | &rbno, &rlen); | |
1435 | ||
1436 | if (rlen >= args->maxlen) | |
1437 | break; | |
1438 | ||
1439 | error = xfs_btree_increment(cnt_cur, 0, &i); | |
1440 | if (error) | |
1441 | goto error0; | |
1442 | if (i == 0) { | |
1443 | /* | |
1444 | * Our only valid extents must have been busy. | |
1445 | * Make it unbusy by forcing the log out and | |
1446 | * retrying. If we've been here before, forcing | |
1447 | * the log isn't making the extents available, | |
1448 | * which means they have probably been freed in | |
1449 | * this transaction. In that case, we have to | |
1450 | * give up on them and we'll attempt a minlen | |
1451 | * allocation the next time around. | |
1452 | */ | |
1453 | xfs_btree_del_cursor(cnt_cur, | |
1454 | XFS_BTREE_NOERROR); | |
1455 | trace_xfs_alloc_size_busy(args); | |
1456 | if (!forced++) | |
1457 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1458 | goto restart; | |
1459 | } | |
1460 | } | |
1da177e4 | 1461 | } |
e26f0501 | 1462 | |
1da177e4 LT |
1463 | /* |
1464 | * In the first case above, we got the last entry in the | |
1465 | * by-size btree. Now we check to see if the space hits maxlen | |
1466 | * once aligned; if not, we search left for something better. | |
1467 | * This can't happen in the second case above. | |
1468 | */ | |
1da177e4 | 1469 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
c29aad41 | 1470 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen == 0 || |
1da177e4 LT |
1471 | (rlen <= flen && rbno + rlen <= fbno + flen), error0); |
1472 | if (rlen < args->maxlen) { | |
1473 | xfs_agblock_t bestfbno; | |
1474 | xfs_extlen_t bestflen; | |
1475 | xfs_agblock_t bestrbno; | |
1476 | xfs_extlen_t bestrlen; | |
1477 | ||
1478 | bestrlen = rlen; | |
1479 | bestrbno = rbno; | |
1480 | bestflen = flen; | |
1481 | bestfbno = fbno; | |
1482 | for (;;) { | |
8df4da4a | 1483 | if ((error = xfs_btree_decrement(cnt_cur, 0, &i))) |
1da177e4 LT |
1484 | goto error0; |
1485 | if (i == 0) | |
1486 | break; | |
1487 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, | |
1488 | &i))) | |
1489 | goto error0; | |
c29aad41 | 1490 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 LT |
1491 | if (flen < bestrlen) |
1492 | break; | |
86fa8af6 CH |
1493 | xfs_alloc_compute_aligned(args, fbno, flen, |
1494 | &rbno, &rlen); | |
1da177e4 | 1495 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
c29aad41 | 1496 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen == 0 || |
1da177e4 LT |
1497 | (rlen <= flen && rbno + rlen <= fbno + flen), |
1498 | error0); | |
1499 | if (rlen > bestrlen) { | |
1500 | bestrlen = rlen; | |
1501 | bestrbno = rbno; | |
1502 | bestflen = flen; | |
1503 | bestfbno = fbno; | |
1504 | if (rlen == args->maxlen) | |
1505 | break; | |
1506 | } | |
1507 | } | |
1508 | if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen, | |
1509 | &i))) | |
1510 | goto error0; | |
c29aad41 | 1511 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 LT |
1512 | rlen = bestrlen; |
1513 | rbno = bestrbno; | |
1514 | flen = bestflen; | |
1515 | fbno = bestfbno; | |
1516 | } | |
1517 | args->wasfromfl = 0; | |
1518 | /* | |
1519 | * Fix up the length. | |
1520 | */ | |
1521 | args->len = rlen; | |
e26f0501 CH |
1522 | if (rlen < args->minlen) { |
1523 | if (!forced++) { | |
1524 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1525 | trace_xfs_alloc_size_busy(args); | |
1526 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1527 | goto restart; | |
1528 | } | |
1529 | goto out_nominleft; | |
1da177e4 | 1530 | } |
e26f0501 CH |
1531 | xfs_alloc_fix_len(args); |
1532 | ||
1533 | if (!xfs_alloc_fix_minleft(args)) | |
1534 | goto out_nominleft; | |
1da177e4 | 1535 | rlen = args->len; |
c29aad41 | 1536 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen <= flen, error0); |
1da177e4 LT |
1537 | /* |
1538 | * Allocate and initialize a cursor for the by-block tree. | |
1539 | */ | |
561f7d17 CH |
1540 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1541 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1542 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
1543 | rbno, rlen, XFSA_FIXUP_CNT_OK))) | |
1544 | goto error0; | |
1545 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1546 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1547 | cnt_cur = bno_cur = NULL; | |
1548 | args->len = rlen; | |
1549 | args->agbno = rbno; | |
c29aad41 | 1550 | XFS_WANT_CORRUPTED_GOTO(args->mp, |
1da177e4 | 1551 | args->agbno + args->len <= |
16259e7d | 1552 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 | 1553 | error0); |
0b1b213f | 1554 | trace_xfs_alloc_size_done(args); |
1da177e4 LT |
1555 | return 0; |
1556 | ||
1557 | error0: | |
0b1b213f | 1558 | trace_xfs_alloc_size_error(args); |
1da177e4 LT |
1559 | if (cnt_cur) |
1560 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1561 | if (bno_cur) | |
1562 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1563 | return error; | |
e26f0501 CH |
1564 | |
1565 | out_nominleft: | |
1566 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1567 | trace_xfs_alloc_size_nominleft(args); | |
1568 | args->agbno = NULLAGBLOCK; | |
1569 | return 0; | |
1da177e4 LT |
1570 | } |
1571 | ||
1572 | /* | |
1573 | * Deal with the case where only small freespaces remain. | |
1574 | * Either return the contents of the last freespace record, | |
1575 | * or allocate space from the freelist if there is nothing in the tree. | |
1576 | */ | |
1577 | STATIC int /* error */ | |
1578 | xfs_alloc_ag_vextent_small( | |
1579 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1580 | xfs_btree_cur_t *ccur, /* by-size cursor */ | |
1581 | xfs_agblock_t *fbnop, /* result block number */ | |
1582 | xfs_extlen_t *flenp, /* result length */ | |
1583 | int *stat) /* status: 0-freelist, 1-normal/none */ | |
1584 | { | |
1585 | int error; | |
1586 | xfs_agblock_t fbno; | |
1587 | xfs_extlen_t flen; | |
1da177e4 LT |
1588 | int i; |
1589 | ||
8df4da4a | 1590 | if ((error = xfs_btree_decrement(ccur, 0, &i))) |
1da177e4 LT |
1591 | goto error0; |
1592 | if (i) { | |
1593 | if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i))) | |
1594 | goto error0; | |
c29aad41 | 1595 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 LT |
1596 | } |
1597 | /* | |
1598 | * Nothing in the btree, try the freelist. Make sure | |
1599 | * to respect minleft even when pulling from the | |
1600 | * freelist. | |
1601 | */ | |
1602 | else if (args->minlen == 1 && args->alignment == 1 && !args->isfl && | |
16259e7d CH |
1603 | (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount) |
1604 | > args->minleft)) { | |
92821e2b DC |
1605 | error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0); |
1606 | if (error) | |
1da177e4 LT |
1607 | goto error0; |
1608 | if (fbno != NULLAGBLOCK) { | |
4ecbfe63 | 1609 | xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1, |
97d3ac75 CH |
1610 | args->userdata); |
1611 | ||
1da177e4 LT |
1612 | if (args->userdata) { |
1613 | xfs_buf_t *bp; | |
1614 | ||
1615 | bp = xfs_btree_get_bufs(args->mp, args->tp, | |
1616 | args->agno, fbno, 0); | |
1617 | xfs_trans_binval(args->tp, bp); | |
1618 | } | |
1619 | args->len = 1; | |
1620 | args->agbno = fbno; | |
c29aad41 | 1621 | XFS_WANT_CORRUPTED_GOTO(args->mp, |
1da177e4 | 1622 | args->agbno + args->len <= |
16259e7d | 1623 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1624 | error0); |
1625 | args->wasfromfl = 1; | |
0b1b213f | 1626 | trace_xfs_alloc_small_freelist(args); |
1da177e4 LT |
1627 | *stat = 0; |
1628 | return 0; | |
1629 | } | |
1630 | /* | |
1631 | * Nothing in the freelist. | |
1632 | */ | |
1633 | else | |
1634 | flen = 0; | |
1635 | } | |
1636 | /* | |
1637 | * Can't allocate from the freelist for some reason. | |
1638 | */ | |
d432c80e NS |
1639 | else { |
1640 | fbno = NULLAGBLOCK; | |
1da177e4 | 1641 | flen = 0; |
d432c80e | 1642 | } |
1da177e4 LT |
1643 | /* |
1644 | * Can't do the allocation, give up. | |
1645 | */ | |
1646 | if (flen < args->minlen) { | |
1647 | args->agbno = NULLAGBLOCK; | |
0b1b213f | 1648 | trace_xfs_alloc_small_notenough(args); |
1da177e4 LT |
1649 | flen = 0; |
1650 | } | |
1651 | *fbnop = fbno; | |
1652 | *flenp = flen; | |
1653 | *stat = 1; | |
0b1b213f | 1654 | trace_xfs_alloc_small_done(args); |
1da177e4 LT |
1655 | return 0; |
1656 | ||
1657 | error0: | |
0b1b213f | 1658 | trace_xfs_alloc_small_error(args); |
1da177e4 LT |
1659 | return error; |
1660 | } | |
1661 | ||
1662 | /* | |
1663 | * Free the extent starting at agno/bno for length. | |
1664 | */ | |
340785cc | 1665 | STATIC int |
1da177e4 | 1666 | xfs_free_ag_extent( |
340785cc DW |
1667 | xfs_trans_t *tp, |
1668 | xfs_buf_t *agbp, | |
1669 | xfs_agnumber_t agno, | |
1670 | xfs_agblock_t bno, | |
1671 | xfs_extlen_t len, | |
1672 | struct xfs_owner_info *oinfo, | |
1673 | int isfl) | |
1da177e4 LT |
1674 | { |
1675 | xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */ | |
1676 | xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */ | |
1677 | int error; /* error return value */ | |
1da177e4 LT |
1678 | xfs_agblock_t gtbno; /* start of right neighbor block */ |
1679 | xfs_extlen_t gtlen; /* length of right neighbor block */ | |
1680 | int haveleft; /* have a left neighbor block */ | |
1681 | int haveright; /* have a right neighbor block */ | |
1682 | int i; /* temp, result code */ | |
1683 | xfs_agblock_t ltbno; /* start of left neighbor block */ | |
1684 | xfs_extlen_t ltlen; /* length of left neighbor block */ | |
1685 | xfs_mount_t *mp; /* mount point struct for filesystem */ | |
1686 | xfs_agblock_t nbno; /* new starting block of freespace */ | |
1687 | xfs_extlen_t nlen; /* new length of freespace */ | |
ecb6928f | 1688 | xfs_perag_t *pag; /* per allocation group data */ |
1da177e4 | 1689 | |
673930c3 | 1690 | bno_cur = cnt_cur = NULL; |
1da177e4 | 1691 | mp = tp->t_mountp; |
673930c3 DW |
1692 | |
1693 | if (oinfo->oi_owner != XFS_RMAP_OWN_UNKNOWN) { | |
1694 | error = xfs_rmap_free(tp, agbp, agno, bno, len, oinfo); | |
1695 | if (error) | |
1696 | goto error0; | |
1697 | } | |
1698 | ||
1da177e4 LT |
1699 | /* |
1700 | * Allocate and initialize a cursor for the by-block btree. | |
1701 | */ | |
561f7d17 | 1702 | bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO); |
1da177e4 LT |
1703 | /* |
1704 | * Look for a neighboring block on the left (lower block numbers) | |
1705 | * that is contiguous with this space. | |
1706 | */ | |
1707 | if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft))) | |
1708 | goto error0; | |
1709 | if (haveleft) { | |
1710 | /* | |
1711 | * There is a block to our left. | |
1712 | */ | |
1713 | if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i))) | |
1714 | goto error0; | |
c29aad41 | 1715 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1716 | /* |
1717 | * It's not contiguous, though. | |
1718 | */ | |
1719 | if (ltbno + ltlen < bno) | |
1720 | haveleft = 0; | |
1721 | else { | |
1722 | /* | |
1723 | * If this failure happens the request to free this | |
1724 | * space was invalid, it's (partly) already free. | |
1725 | * Very bad. | |
1726 | */ | |
c29aad41 ES |
1727 | XFS_WANT_CORRUPTED_GOTO(mp, |
1728 | ltbno + ltlen <= bno, error0); | |
1da177e4 LT |
1729 | } |
1730 | } | |
1731 | /* | |
1732 | * Look for a neighboring block on the right (higher block numbers) | |
1733 | * that is contiguous with this space. | |
1734 | */ | |
637aa50f | 1735 | if ((error = xfs_btree_increment(bno_cur, 0, &haveright))) |
1da177e4 LT |
1736 | goto error0; |
1737 | if (haveright) { | |
1738 | /* | |
1739 | * There is a block to our right. | |
1740 | */ | |
1741 | if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i))) | |
1742 | goto error0; | |
c29aad41 | 1743 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1744 | /* |
1745 | * It's not contiguous, though. | |
1746 | */ | |
1747 | if (bno + len < gtbno) | |
1748 | haveright = 0; | |
1749 | else { | |
1750 | /* | |
1751 | * If this failure happens the request to free this | |
1752 | * space was invalid, it's (partly) already free. | |
1753 | * Very bad. | |
1754 | */ | |
c29aad41 | 1755 | XFS_WANT_CORRUPTED_GOTO(mp, gtbno >= bno + len, error0); |
1da177e4 LT |
1756 | } |
1757 | } | |
1758 | /* | |
1759 | * Now allocate and initialize a cursor for the by-size tree. | |
1760 | */ | |
561f7d17 | 1761 | cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT); |
1da177e4 LT |
1762 | /* |
1763 | * Have both left and right contiguous neighbors. | |
1764 | * Merge all three into a single free block. | |
1765 | */ | |
1766 | if (haveleft && haveright) { | |
1767 | /* | |
1768 | * Delete the old by-size entry on the left. | |
1769 | */ | |
1770 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1771 | goto error0; | |
c29aad41 | 1772 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
91cca5df | 1773 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1774 | goto error0; |
c29aad41 | 1775 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1776 | /* |
1777 | * Delete the old by-size entry on the right. | |
1778 | */ | |
1779 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1780 | goto error0; | |
c29aad41 | 1781 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
91cca5df | 1782 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1783 | goto error0; |
c29aad41 | 1784 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1785 | /* |
1786 | * Delete the old by-block entry for the right block. | |
1787 | */ | |
91cca5df | 1788 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 | 1789 | goto error0; |
c29aad41 | 1790 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1791 | /* |
1792 | * Move the by-block cursor back to the left neighbor. | |
1793 | */ | |
8df4da4a | 1794 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 | 1795 | goto error0; |
c29aad41 | 1796 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1797 | #ifdef DEBUG |
1798 | /* | |
1799 | * Check that this is the right record: delete didn't | |
1800 | * mangle the cursor. | |
1801 | */ | |
1802 | { | |
1803 | xfs_agblock_t xxbno; | |
1804 | xfs_extlen_t xxlen; | |
1805 | ||
1806 | if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen, | |
1807 | &i))) | |
1808 | goto error0; | |
c29aad41 | 1809 | XFS_WANT_CORRUPTED_GOTO(mp, |
1da177e4 LT |
1810 | i == 1 && xxbno == ltbno && xxlen == ltlen, |
1811 | error0); | |
1812 | } | |
1813 | #endif | |
1814 | /* | |
1815 | * Update remaining by-block entry to the new, joined block. | |
1816 | */ | |
1817 | nbno = ltbno; | |
1818 | nlen = len + ltlen + gtlen; | |
1819 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1820 | goto error0; | |
1821 | } | |
1822 | /* | |
1823 | * Have only a left contiguous neighbor. | |
1824 | * Merge it together with the new freespace. | |
1825 | */ | |
1826 | else if (haveleft) { | |
1827 | /* | |
1828 | * Delete the old by-size entry on the left. | |
1829 | */ | |
1830 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1831 | goto error0; | |
c29aad41 | 1832 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
91cca5df | 1833 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1834 | goto error0; |
c29aad41 | 1835 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1836 | /* |
1837 | * Back up the by-block cursor to the left neighbor, and | |
1838 | * update its length. | |
1839 | */ | |
8df4da4a | 1840 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 | 1841 | goto error0; |
c29aad41 | 1842 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1843 | nbno = ltbno; |
1844 | nlen = len + ltlen; | |
1845 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1846 | goto error0; | |
1847 | } | |
1848 | /* | |
1849 | * Have only a right contiguous neighbor. | |
1850 | * Merge it together with the new freespace. | |
1851 | */ | |
1852 | else if (haveright) { | |
1853 | /* | |
1854 | * Delete the old by-size entry on the right. | |
1855 | */ | |
1856 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1857 | goto error0; | |
c29aad41 | 1858 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
91cca5df | 1859 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1860 | goto error0; |
c29aad41 | 1861 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1862 | /* |
1863 | * Update the starting block and length of the right | |
1864 | * neighbor in the by-block tree. | |
1865 | */ | |
1866 | nbno = bno; | |
1867 | nlen = len + gtlen; | |
1868 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1869 | goto error0; | |
1870 | } | |
1871 | /* | |
1872 | * No contiguous neighbors. | |
1873 | * Insert the new freespace into the by-block tree. | |
1874 | */ | |
1875 | else { | |
1876 | nbno = bno; | |
1877 | nlen = len; | |
4b22a571 | 1878 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 | 1879 | goto error0; |
c29aad41 | 1880 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1881 | } |
1882 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1883 | bno_cur = NULL; | |
1884 | /* | |
1885 | * In all cases we need to insert the new freespace in the by-size tree. | |
1886 | */ | |
1887 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i))) | |
1888 | goto error0; | |
c29aad41 | 1889 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, error0); |
4b22a571 | 1890 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 1891 | goto error0; |
c29aad41 | 1892 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1893 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1894 | cnt_cur = NULL; | |
ecb6928f | 1895 | |
1da177e4 LT |
1896 | /* |
1897 | * Update the freespace totals in the ag and superblock. | |
1898 | */ | |
ecb6928f CH |
1899 | pag = xfs_perag_get(mp, agno); |
1900 | error = xfs_alloc_update_counters(tp, pag, agbp, len); | |
1901 | xfs_perag_put(pag); | |
1902 | if (error) | |
1903 | goto error0; | |
1904 | ||
1905 | if (!isfl) | |
1906 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len); | |
ff6d6af2 BD |
1907 | XFS_STATS_INC(mp, xs_freex); |
1908 | XFS_STATS_ADD(mp, xs_freeb, len); | |
0b1b213f CH |
1909 | |
1910 | trace_xfs_free_extent(mp, agno, bno, len, isfl, haveleft, haveright); | |
1da177e4 | 1911 | |
1da177e4 LT |
1912 | return 0; |
1913 | ||
1914 | error0: | |
0b1b213f | 1915 | trace_xfs_free_extent(mp, agno, bno, len, isfl, -1, -1); |
1da177e4 LT |
1916 | if (bno_cur) |
1917 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1918 | if (cnt_cur) | |
1919 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1920 | return error; | |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * Visible (exported) allocation/free functions. | |
1925 | * Some of these are used just by xfs_alloc_btree.c and this file. | |
1926 | */ | |
1927 | ||
1928 | /* | |
1929 | * Compute and fill in value of m_ag_maxlevels. | |
1930 | */ | |
1931 | void | |
1932 | xfs_alloc_compute_maxlevels( | |
1933 | xfs_mount_t *mp) /* file system mount structure */ | |
1934 | { | |
19b54ee6 DW |
1935 | mp->m_ag_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_alloc_mnr, |
1936 | (mp->m_sb.sb_agblocks + 1) / 2); | |
1da177e4 LT |
1937 | } |
1938 | ||
6cc87645 DC |
1939 | /* |
1940 | * Find the length of the longest extent in an AG. | |
1941 | */ | |
1942 | xfs_extlen_t | |
1943 | xfs_alloc_longest_free_extent( | |
1944 | struct xfs_mount *mp, | |
50adbcb4 DC |
1945 | struct xfs_perag *pag, |
1946 | xfs_extlen_t need) | |
6cc87645 | 1947 | { |
50adbcb4 | 1948 | xfs_extlen_t delta = 0; |
6cc87645 | 1949 | |
6cc87645 DC |
1950 | if (need > pag->pagf_flcount) |
1951 | delta = need - pag->pagf_flcount; | |
1952 | ||
1953 | if (pag->pagf_longest > delta) | |
1954 | return pag->pagf_longest - delta; | |
1955 | return pag->pagf_flcount > 0 || pag->pagf_longest > 0; | |
1956 | } | |
1957 | ||
496817b4 DC |
1958 | unsigned int |
1959 | xfs_alloc_min_freelist( | |
1960 | struct xfs_mount *mp, | |
1961 | struct xfs_perag *pag) | |
1962 | { | |
1963 | unsigned int min_free; | |
1964 | ||
1965 | /* space needed by-bno freespace btree */ | |
1966 | min_free = min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_BNOi] + 1, | |
1967 | mp->m_ag_maxlevels); | |
1968 | /* space needed by-size freespace btree */ | |
1969 | min_free += min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_CNTi] + 1, | |
1970 | mp->m_ag_maxlevels); | |
52548852 DW |
1971 | /* space needed reverse mapping used space btree */ |
1972 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
1973 | min_free += min_t(unsigned int, | |
1974 | pag->pagf_levels[XFS_BTNUM_RMAPi] + 1, | |
1975 | mp->m_rmap_maxlevels); | |
496817b4 DC |
1976 | |
1977 | return min_free; | |
1978 | } | |
1979 | ||
72d55285 DC |
1980 | /* |
1981 | * Check if the operation we are fixing up the freelist for should go ahead or | |
1982 | * not. If we are freeing blocks, we always allow it, otherwise the allocation | |
1983 | * is dependent on whether the size and shape of free space available will | |
1984 | * permit the requested allocation to take place. | |
1985 | */ | |
1986 | static bool | |
1987 | xfs_alloc_space_available( | |
1988 | struct xfs_alloc_arg *args, | |
1989 | xfs_extlen_t min_free, | |
1990 | int flags) | |
1991 | { | |
1992 | struct xfs_perag *pag = args->pag; | |
1993 | xfs_extlen_t longest; | |
1994 | int available; | |
1995 | ||
1996 | if (flags & XFS_ALLOC_FLAG_FREEING) | |
1997 | return true; | |
1998 | ||
1999 | /* do we have enough contiguous free space for the allocation? */ | |
2000 | longest = xfs_alloc_longest_free_extent(args->mp, pag, min_free); | |
2001 | if ((args->minlen + args->alignment + args->minalignslop - 1) > longest) | |
2002 | return false; | |
2003 | ||
2004 | /* do have enough free space remaining for the allocation? */ | |
2005 | available = (int)(pag->pagf_freeblks + pag->pagf_flcount - | |
2006 | min_free - args->total); | |
2007 | if (available < (int)args->minleft) | |
2008 | return false; | |
2009 | ||
2010 | return true; | |
2011 | } | |
2012 | ||
1da177e4 LT |
2013 | /* |
2014 | * Decide whether to use this allocation group for this allocation. | |
2015 | * If so, fix up the btree freelist's size. | |
2016 | */ | |
2e9101da | 2017 | int /* error */ |
1da177e4 | 2018 | xfs_alloc_fix_freelist( |
396503fc DC |
2019 | struct xfs_alloc_arg *args, /* allocation argument structure */ |
2020 | int flags) /* XFS_ALLOC_FLAG_... */ | |
1da177e4 | 2021 | { |
396503fc DC |
2022 | struct xfs_mount *mp = args->mp; |
2023 | struct xfs_perag *pag = args->pag; | |
2024 | struct xfs_trans *tp = args->tp; | |
2025 | struct xfs_buf *agbp = NULL; | |
2026 | struct xfs_buf *agflbp = NULL; | |
2027 | struct xfs_alloc_arg targs; /* local allocation arguments */ | |
2028 | xfs_agblock_t bno; /* freelist block */ | |
2029 | xfs_extlen_t need; /* total blocks needed in freelist */ | |
c184f855 | 2030 | int error = 0; |
396503fc | 2031 | |
1da177e4 | 2032 | if (!pag->pagf_init) { |
396503fc DC |
2033 | error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp); |
2034 | if (error) | |
2035 | goto out_no_agbp; | |
1da177e4 | 2036 | if (!pag->pagf_init) { |
0e1edbd9 NS |
2037 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
2038 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
396503fc | 2039 | goto out_agbp_relse; |
1da177e4 | 2040 | } |
396503fc | 2041 | } |
1da177e4 | 2042 | |
0e1edbd9 | 2043 | /* |
396503fc DC |
2044 | * If this is a metadata preferred pag and we are user data then try |
2045 | * somewhere else if we are not being asked to try harder at this | |
2046 | * point | |
1da177e4 | 2047 | */ |
0e1edbd9 NS |
2048 | if (pag->pagf_metadata && args->userdata && |
2049 | (flags & XFS_ALLOC_FLAG_TRYLOCK)) { | |
2050 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
396503fc | 2051 | goto out_agbp_relse; |
1da177e4 LT |
2052 | } |
2053 | ||
496817b4 | 2054 | need = xfs_alloc_min_freelist(mp, pag); |
396503fc DC |
2055 | if (!xfs_alloc_space_available(args, need, flags)) |
2056 | goto out_agbp_relse; | |
0e1edbd9 | 2057 | |
1da177e4 LT |
2058 | /* |
2059 | * Get the a.g. freespace buffer. | |
2060 | * Can fail if we're not blocking on locks, and it's held. | |
2061 | */ | |
396503fc DC |
2062 | if (!agbp) { |
2063 | error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp); | |
2064 | if (error) | |
2065 | goto out_no_agbp; | |
2066 | if (!agbp) { | |
0e1edbd9 NS |
2067 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
2068 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
396503fc | 2069 | goto out_no_agbp; |
0e1edbd9 | 2070 | } |
1da177e4 | 2071 | } |
50adbcb4 | 2072 | |
50adbcb4 | 2073 | /* If there isn't enough total space or single-extent, reject it. */ |
496817b4 | 2074 | need = xfs_alloc_min_freelist(mp, pag); |
396503fc DC |
2075 | if (!xfs_alloc_space_available(args, need, flags)) |
2076 | goto out_agbp_relse; | |
72d55285 | 2077 | |
1da177e4 LT |
2078 | /* |
2079 | * Make the freelist shorter if it's too long. | |
50adbcb4 | 2080 | * |
396503fc DC |
2081 | * Note that from this point onwards, we will always release the agf and |
2082 | * agfl buffers on error. This handles the case where we error out and | |
2083 | * the buffers are clean or may not have been joined to the transaction | |
2084 | * and hence need to be released manually. If they have been joined to | |
2085 | * the transaction, then xfs_trans_brelse() will handle them | |
2086 | * appropriately based on the recursion count and dirty state of the | |
2087 | * buffer. | |
2088 | * | |
50adbcb4 DC |
2089 | * XXX (dgc): When we have lots of free space, does this buy us |
2090 | * anything other than extra overhead when we need to put more blocks | |
2091 | * back on the free list? Maybe we should only do this when space is | |
2092 | * getting low or the AGFL is more than half full? | |
04f13060 DW |
2093 | * |
2094 | * The NOSHRINK flag prevents the AGFL from being shrunk if it's too | |
2095 | * big; the NORMAP flag prevents AGFL expand/shrink operations from | |
2096 | * updating the rmapbt. Both flags are used in xfs_repair while we're | |
2097 | * rebuilding the rmapbt, and neither are used by the kernel. They're | |
2098 | * both required to ensure that rmaps are correctly recorded for the | |
2099 | * regenerated AGFL, bnobt, and cntbt. See repair/phase5.c and | |
2100 | * repair/rmap.c in xfsprogs for details. | |
1da177e4 | 2101 | */ |
04f13060 DW |
2102 | memset(&targs, 0, sizeof(targs)); |
2103 | if (flags & XFS_ALLOC_FLAG_NORMAP) | |
2104 | xfs_rmap_skip_owner_update(&targs.oinfo); | |
2105 | else | |
2106 | xfs_rmap_ag_owner(&targs.oinfo, XFS_RMAP_OWN_AG); | |
2107 | while (!(flags & XFS_ALLOC_FLAG_NOSHRINK) && pag->pagf_flcount > need) { | |
50adbcb4 | 2108 | struct xfs_buf *bp; |
1da177e4 | 2109 | |
92821e2b DC |
2110 | error = xfs_alloc_get_freelist(tp, agbp, &bno, 0); |
2111 | if (error) | |
396503fc | 2112 | goto out_agbp_relse; |
340785cc DW |
2113 | error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, |
2114 | &targs.oinfo, 1); | |
50adbcb4 | 2115 | if (error) |
396503fc | 2116 | goto out_agbp_relse; |
1da177e4 LT |
2117 | bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0); |
2118 | xfs_trans_binval(tp, bp); | |
2119 | } | |
50adbcb4 | 2120 | |
1da177e4 LT |
2121 | targs.tp = tp; |
2122 | targs.mp = mp; | |
2123 | targs.agbp = agbp; | |
2124 | targs.agno = args->agno; | |
1da177e4 LT |
2125 | targs.alignment = targs.minlen = targs.prod = targs.isfl = 1; |
2126 | targs.type = XFS_ALLOCTYPE_THIS_AG; | |
2127 | targs.pag = pag; | |
50adbcb4 DC |
2128 | error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp); |
2129 | if (error) | |
396503fc | 2130 | goto out_agbp_relse; |
50adbcb4 DC |
2131 | |
2132 | /* Make the freelist longer if it's too short. */ | |
2133 | while (pag->pagf_flcount < need) { | |
1da177e4 | 2134 | targs.agbno = 0; |
50adbcb4 DC |
2135 | targs.maxlen = need - pag->pagf_flcount; |
2136 | ||
2137 | /* Allocate as many blocks as possible at once. */ | |
2138 | error = xfs_alloc_ag_vextent(&targs); | |
396503fc DC |
2139 | if (error) |
2140 | goto out_agflbp_relse; | |
2141 | ||
1da177e4 LT |
2142 | /* |
2143 | * Stop if we run out. Won't happen if callers are obeying | |
2144 | * the restrictions correctly. Can happen for free calls | |
2145 | * on a completely full ag. | |
2146 | */ | |
d210a28c | 2147 | if (targs.agbno == NULLAGBLOCK) { |
0e1edbd9 NS |
2148 | if (flags & XFS_ALLOC_FLAG_FREEING) |
2149 | break; | |
396503fc | 2150 | goto out_agflbp_relse; |
d210a28c | 2151 | } |
1da177e4 LT |
2152 | /* |
2153 | * Put each allocated block on the list. | |
2154 | */ | |
2155 | for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) { | |
92821e2b DC |
2156 | error = xfs_alloc_put_freelist(tp, agbp, |
2157 | agflbp, bno, 0); | |
2158 | if (error) | |
396503fc | 2159 | goto out_agflbp_relse; |
1da177e4 LT |
2160 | } |
2161 | } | |
e63a3690 | 2162 | xfs_trans_brelse(tp, agflbp); |
1da177e4 LT |
2163 | args->agbp = agbp; |
2164 | return 0; | |
396503fc DC |
2165 | |
2166 | out_agflbp_relse: | |
2167 | xfs_trans_brelse(tp, agflbp); | |
2168 | out_agbp_relse: | |
2169 | if (agbp) | |
2170 | xfs_trans_brelse(tp, agbp); | |
2171 | out_no_agbp: | |
2172 | args->agbp = NULL; | |
2173 | return error; | |
1da177e4 LT |
2174 | } |
2175 | ||
2176 | /* | |
2177 | * Get a block from the freelist. | |
2178 | * Returns with the buffer for the block gotten. | |
2179 | */ | |
2180 | int /* error */ | |
2181 | xfs_alloc_get_freelist( | |
2182 | xfs_trans_t *tp, /* transaction pointer */ | |
2183 | xfs_buf_t *agbp, /* buffer containing the agf structure */ | |
92821e2b DC |
2184 | xfs_agblock_t *bnop, /* block address retrieved from freelist */ |
2185 | int btreeblk) /* destination is a AGF btree */ | |
1da177e4 LT |
2186 | { |
2187 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
1da177e4 LT |
2188 | xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */ |
2189 | xfs_agblock_t bno; /* block number returned */ | |
77c95bba | 2190 | __be32 *agfl_bno; |
1da177e4 | 2191 | int error; |
92821e2b | 2192 | int logflags; |
77c95bba | 2193 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
2194 | xfs_perag_t *pag; /* per allocation group data */ |
2195 | ||
1da177e4 LT |
2196 | /* |
2197 | * Freelist is empty, give up. | |
2198 | */ | |
77c95bba | 2199 | agf = XFS_BUF_TO_AGF(agbp); |
1da177e4 LT |
2200 | if (!agf->agf_flcount) { |
2201 | *bnop = NULLAGBLOCK; | |
2202 | return 0; | |
2203 | } | |
2204 | /* | |
2205 | * Read the array of free blocks. | |
2206 | */ | |
77c95bba CH |
2207 | error = xfs_alloc_read_agfl(mp, tp, be32_to_cpu(agf->agf_seqno), |
2208 | &agflbp); | |
2209 | if (error) | |
1da177e4 | 2210 | return error; |
77c95bba CH |
2211 | |
2212 | ||
1da177e4 LT |
2213 | /* |
2214 | * Get the block number and update the data structures. | |
2215 | */ | |
77c95bba CH |
2216 | agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp); |
2217 | bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]); | |
413d57c9 | 2218 | be32_add_cpu(&agf->agf_flfirst, 1); |
1da177e4 | 2219 | xfs_trans_brelse(tp, agflbp); |
16259e7d | 2220 | if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2221 | agf->agf_flfirst = 0; |
a862e0fd DC |
2222 | |
2223 | pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); | |
413d57c9 | 2224 | be32_add_cpu(&agf->agf_flcount, -1); |
1da177e4 LT |
2225 | xfs_trans_agflist_delta(tp, -1); |
2226 | pag->pagf_flcount--; | |
a862e0fd | 2227 | xfs_perag_put(pag); |
92821e2b DC |
2228 | |
2229 | logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; | |
2230 | if (btreeblk) { | |
413d57c9 | 2231 | be32_add_cpu(&agf->agf_btreeblks, 1); |
92821e2b DC |
2232 | pag->pagf_btreeblks++; |
2233 | logflags |= XFS_AGF_BTREEBLKS; | |
2234 | } | |
2235 | ||
92821e2b | 2236 | xfs_alloc_log_agf(tp, agbp, logflags); |
1da177e4 LT |
2237 | *bnop = bno; |
2238 | ||
1da177e4 LT |
2239 | return 0; |
2240 | } | |
2241 | ||
2242 | /* | |
2243 | * Log the given fields from the agf structure. | |
2244 | */ | |
2245 | void | |
2246 | xfs_alloc_log_agf( | |
2247 | xfs_trans_t *tp, /* transaction pointer */ | |
2248 | xfs_buf_t *bp, /* buffer for a.g. freelist header */ | |
2249 | int fields) /* mask of fields to be logged (XFS_AGF_...) */ | |
2250 | { | |
2251 | int first; /* first byte offset */ | |
2252 | int last; /* last byte offset */ | |
2253 | static const short offsets[] = { | |
2254 | offsetof(xfs_agf_t, agf_magicnum), | |
2255 | offsetof(xfs_agf_t, agf_versionnum), | |
2256 | offsetof(xfs_agf_t, agf_seqno), | |
2257 | offsetof(xfs_agf_t, agf_length), | |
2258 | offsetof(xfs_agf_t, agf_roots[0]), | |
2259 | offsetof(xfs_agf_t, agf_levels[0]), | |
2260 | offsetof(xfs_agf_t, agf_flfirst), | |
2261 | offsetof(xfs_agf_t, agf_fllast), | |
2262 | offsetof(xfs_agf_t, agf_flcount), | |
2263 | offsetof(xfs_agf_t, agf_freeblks), | |
2264 | offsetof(xfs_agf_t, agf_longest), | |
92821e2b | 2265 | offsetof(xfs_agf_t, agf_btreeblks), |
4e0e6040 | 2266 | offsetof(xfs_agf_t, agf_uuid), |
f32866fd | 2267 | offsetof(xfs_agf_t, agf_rmap_blocks), |
1da177e4 LT |
2268 | sizeof(xfs_agf_t) |
2269 | }; | |
2270 | ||
0b1b213f CH |
2271 | trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_); |
2272 | ||
61fe135c | 2273 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF); |
4e0e6040 | 2274 | |
1da177e4 LT |
2275 | xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last); |
2276 | xfs_trans_log_buf(tp, bp, (uint)first, (uint)last); | |
2277 | } | |
2278 | ||
2279 | /* | |
2280 | * Interface for inode allocation to force the pag data to be initialized. | |
2281 | */ | |
2282 | int /* error */ | |
2283 | xfs_alloc_pagf_init( | |
2284 | xfs_mount_t *mp, /* file system mount structure */ | |
2285 | xfs_trans_t *tp, /* transaction pointer */ | |
2286 | xfs_agnumber_t agno, /* allocation group number */ | |
2287 | int flags) /* XFS_ALLOC_FLAGS_... */ | |
2288 | { | |
2289 | xfs_buf_t *bp; | |
2290 | int error; | |
2291 | ||
2292 | if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp))) | |
2293 | return error; | |
2294 | if (bp) | |
2295 | xfs_trans_brelse(tp, bp); | |
2296 | return 0; | |
2297 | } | |
2298 | ||
2299 | /* | |
2300 | * Put the block on the freelist for the allocation group. | |
2301 | */ | |
2302 | int /* error */ | |
2303 | xfs_alloc_put_freelist( | |
2304 | xfs_trans_t *tp, /* transaction pointer */ | |
2305 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
2306 | xfs_buf_t *agflbp,/* buffer for a.g. free block array */ | |
92821e2b DC |
2307 | xfs_agblock_t bno, /* block being freed */ |
2308 | int btreeblk) /* block came from a AGF btree */ | |
1da177e4 LT |
2309 | { |
2310 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
e2101005 | 2311 | __be32 *blockp;/* pointer to array entry */ |
1da177e4 | 2312 | int error; |
92821e2b | 2313 | int logflags; |
1da177e4 LT |
2314 | xfs_mount_t *mp; /* mount structure */ |
2315 | xfs_perag_t *pag; /* per allocation group data */ | |
77c95bba CH |
2316 | __be32 *agfl_bno; |
2317 | int startoff; | |
1da177e4 LT |
2318 | |
2319 | agf = XFS_BUF_TO_AGF(agbp); | |
2320 | mp = tp->t_mountp; | |
2321 | ||
2322 | if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2323 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 | 2324 | return error; |
413d57c9 | 2325 | be32_add_cpu(&agf->agf_fllast, 1); |
16259e7d | 2326 | if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2327 | agf->agf_fllast = 0; |
a862e0fd DC |
2328 | |
2329 | pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); | |
413d57c9 | 2330 | be32_add_cpu(&agf->agf_flcount, 1); |
1da177e4 LT |
2331 | xfs_trans_agflist_delta(tp, 1); |
2332 | pag->pagf_flcount++; | |
92821e2b DC |
2333 | |
2334 | logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT; | |
2335 | if (btreeblk) { | |
413d57c9 | 2336 | be32_add_cpu(&agf->agf_btreeblks, -1); |
92821e2b DC |
2337 | pag->pagf_btreeblks--; |
2338 | logflags |= XFS_AGF_BTREEBLKS; | |
2339 | } | |
a862e0fd | 2340 | xfs_perag_put(pag); |
92821e2b | 2341 | |
92821e2b DC |
2342 | xfs_alloc_log_agf(tp, agbp, logflags); |
2343 | ||
16259e7d | 2344 | ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)); |
77c95bba CH |
2345 | |
2346 | agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp); | |
2347 | blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)]; | |
e2101005 | 2348 | *blockp = cpu_to_be32(bno); |
77c95bba CH |
2349 | startoff = (char *)blockp - (char *)agflbp->b_addr; |
2350 | ||
92821e2b | 2351 | xfs_alloc_log_agf(tp, agbp, logflags); |
77c95bba | 2352 | |
61fe135c | 2353 | xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF); |
77c95bba CH |
2354 | xfs_trans_log_buf(tp, agflbp, startoff, |
2355 | startoff + sizeof(xfs_agblock_t) - 1); | |
1da177e4 LT |
2356 | return 0; |
2357 | } | |
2358 | ||
4e0e6040 | 2359 | static bool |
612cfbfe | 2360 | xfs_agf_verify( |
4e0e6040 | 2361 | struct xfs_mount *mp, |
5d5f527d DC |
2362 | struct xfs_buf *bp) |
2363 | { | |
4e0e6040 | 2364 | struct xfs_agf *agf = XFS_BUF_TO_AGF(bp); |
5d5f527d | 2365 | |
a45086e2 BF |
2366 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2367 | if (!uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid)) | |
4e0e6040 | 2368 | return false; |
a45086e2 BF |
2369 | if (!xfs_log_check_lsn(mp, |
2370 | be64_to_cpu(XFS_BUF_TO_AGF(bp)->agf_lsn))) | |
2371 | return false; | |
2372 | } | |
5d5f527d | 2373 | |
4e0e6040 DC |
2374 | if (!(agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) && |
2375 | XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) && | |
2376 | be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) && | |
2377 | be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) && | |
2378 | be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) && | |
2379 | be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp))) | |
2380 | return false; | |
5d5f527d | 2381 | |
e1b05723 ES |
2382 | if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) > XFS_BTREE_MAXLEVELS || |
2383 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) > XFS_BTREE_MAXLEVELS) | |
2384 | return false; | |
2385 | ||
b8704944 DW |
2386 | if (xfs_sb_version_hasrmapbt(&mp->m_sb) && |
2387 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) > XFS_BTREE_MAXLEVELS) | |
2388 | return false; | |
2389 | ||
5d5f527d DC |
2390 | /* |
2391 | * during growfs operations, the perag is not fully initialised, | |
2392 | * so we can't use it for any useful checking. growfs ensures we can't | |
2393 | * use it by using uncached buffers that don't have the perag attached | |
2394 | * so we can detect and avoid this problem. | |
2395 | */ | |
4e0e6040 DC |
2396 | if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno) |
2397 | return false; | |
5d5f527d | 2398 | |
4e0e6040 DC |
2399 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
2400 | be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length)) | |
2401 | return false; | |
2402 | ||
2403 | return true;; | |
5d5f527d | 2404 | |
612cfbfe DC |
2405 | } |
2406 | ||
1813dd64 DC |
2407 | static void |
2408 | xfs_agf_read_verify( | |
612cfbfe DC |
2409 | struct xfs_buf *bp) |
2410 | { | |
4e0e6040 | 2411 | struct xfs_mount *mp = bp->b_target->bt_mount; |
4e0e6040 | 2412 | |
ce5028cf ES |
2413 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2414 | !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF)) | |
2451337d | 2415 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf ES |
2416 | else if (XFS_TEST_ERROR(!xfs_agf_verify(mp, bp), mp, |
2417 | XFS_ERRTAG_ALLOC_READ_AGF, | |
2418 | XFS_RANDOM_ALLOC_READ_AGF)) | |
2451337d | 2419 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
2420 | |
2421 | if (bp->b_error) | |
2422 | xfs_verifier_error(bp); | |
612cfbfe | 2423 | } |
5d5f527d | 2424 | |
b0f539de | 2425 | static void |
1813dd64 | 2426 | xfs_agf_write_verify( |
612cfbfe DC |
2427 | struct xfs_buf *bp) |
2428 | { | |
4e0e6040 DC |
2429 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2430 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2431 | ||
2432 | if (!xfs_agf_verify(mp, bp)) { | |
2451337d | 2433 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 2434 | xfs_verifier_error(bp); |
4e0e6040 DC |
2435 | return; |
2436 | } | |
2437 | ||
2438 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2439 | return; | |
2440 | ||
2441 | if (bip) | |
2442 | XFS_BUF_TO_AGF(bp)->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
2443 | ||
f1dbcd7e | 2444 | xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF); |
5d5f527d DC |
2445 | } |
2446 | ||
1813dd64 | 2447 | const struct xfs_buf_ops xfs_agf_buf_ops = { |
233135b7 | 2448 | .name = "xfs_agf", |
1813dd64 DC |
2449 | .verify_read = xfs_agf_read_verify, |
2450 | .verify_write = xfs_agf_write_verify, | |
2451 | }; | |
2452 | ||
1da177e4 LT |
2453 | /* |
2454 | * Read in the allocation group header (free/alloc section). | |
2455 | */ | |
2456 | int /* error */ | |
4805621a FCH |
2457 | xfs_read_agf( |
2458 | struct xfs_mount *mp, /* mount point structure */ | |
2459 | struct xfs_trans *tp, /* transaction pointer */ | |
2460 | xfs_agnumber_t agno, /* allocation group number */ | |
2461 | int flags, /* XFS_BUF_ */ | |
2462 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
1da177e4 | 2463 | { |
1da177e4 LT |
2464 | int error; |
2465 | ||
d123031a DC |
2466 | trace_xfs_read_agf(mp, agno); |
2467 | ||
1da177e4 LT |
2468 | ASSERT(agno != NULLAGNUMBER); |
2469 | error = xfs_trans_read_buf( | |
2470 | mp, tp, mp->m_ddev_targp, | |
2471 | XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)), | |
1813dd64 | 2472 | XFS_FSS_TO_BB(mp, 1), flags, bpp, &xfs_agf_buf_ops); |
1da177e4 LT |
2473 | if (error) |
2474 | return error; | |
4805621a | 2475 | if (!*bpp) |
1da177e4 | 2476 | return 0; |
4805621a | 2477 | |
5a52c2a5 | 2478 | ASSERT(!(*bpp)->b_error); |
38f23232 | 2479 | xfs_buf_set_ref(*bpp, XFS_AGF_REF); |
4805621a FCH |
2480 | return 0; |
2481 | } | |
2482 | ||
2483 | /* | |
2484 | * Read in the allocation group header (free/alloc section). | |
2485 | */ | |
2486 | int /* error */ | |
2487 | xfs_alloc_read_agf( | |
2488 | struct xfs_mount *mp, /* mount point structure */ | |
2489 | struct xfs_trans *tp, /* transaction pointer */ | |
2490 | xfs_agnumber_t agno, /* allocation group number */ | |
2491 | int flags, /* XFS_ALLOC_FLAG_... */ | |
2492 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
2493 | { | |
2494 | struct xfs_agf *agf; /* ag freelist header */ | |
2495 | struct xfs_perag *pag; /* per allocation group data */ | |
2496 | int error; | |
2497 | ||
d123031a | 2498 | trace_xfs_alloc_read_agf(mp, agno); |
4805621a | 2499 | |
d123031a | 2500 | ASSERT(agno != NULLAGNUMBER); |
4805621a | 2501 | error = xfs_read_agf(mp, tp, agno, |
0cadda1c | 2502 | (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0, |
4805621a FCH |
2503 | bpp); |
2504 | if (error) | |
2505 | return error; | |
2506 | if (!*bpp) | |
2507 | return 0; | |
5a52c2a5 | 2508 | ASSERT(!(*bpp)->b_error); |
4805621a FCH |
2509 | |
2510 | agf = XFS_BUF_TO_AGF(*bpp); | |
a862e0fd | 2511 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2512 | if (!pag->pagf_init) { |
16259e7d | 2513 | pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); |
92821e2b | 2514 | pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); |
16259e7d CH |
2515 | pag->pagf_flcount = be32_to_cpu(agf->agf_flcount); |
2516 | pag->pagf_longest = be32_to_cpu(agf->agf_longest); | |
1da177e4 | 2517 | pag->pagf_levels[XFS_BTNUM_BNOi] = |
16259e7d | 2518 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); |
1da177e4 | 2519 | pag->pagf_levels[XFS_BTNUM_CNTi] = |
16259e7d | 2520 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); |
b8704944 DW |
2521 | pag->pagf_levels[XFS_BTNUM_RMAPi] = |
2522 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]); | |
007c61c6 | 2523 | spin_lock_init(&pag->pagb_lock); |
e57336ff | 2524 | pag->pagb_count = 0; |
ed3b4d6c | 2525 | pag->pagb_tree = RB_ROOT; |
1da177e4 LT |
2526 | pag->pagf_init = 1; |
2527 | } | |
2528 | #ifdef DEBUG | |
2529 | else if (!XFS_FORCED_SHUTDOWN(mp)) { | |
16259e7d | 2530 | ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks)); |
89b28393 | 2531 | ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks)); |
16259e7d CH |
2532 | ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount)); |
2533 | ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest)); | |
1da177e4 | 2534 | ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] == |
16259e7d | 2535 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi])); |
1da177e4 | 2536 | ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] == |
16259e7d | 2537 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); |
1da177e4 LT |
2538 | } |
2539 | #endif | |
a862e0fd | 2540 | xfs_perag_put(pag); |
1da177e4 LT |
2541 | return 0; |
2542 | } | |
2543 | ||
2544 | /* | |
2545 | * Allocate an extent (variable-size). | |
2546 | * Depending on the allocation type, we either look in a single allocation | |
2547 | * group or loop over the allocation groups to find the result. | |
2548 | */ | |
2549 | int /* error */ | |
e04426b9 | 2550 | xfs_alloc_vextent( |
1da177e4 LT |
2551 | xfs_alloc_arg_t *args) /* allocation argument structure */ |
2552 | { | |
2553 | xfs_agblock_t agsize; /* allocation group size */ | |
2554 | int error; | |
2555 | int flags; /* XFS_ALLOC_FLAG_... locking flags */ | |
1da177e4 LT |
2556 | xfs_extlen_t minleft;/* minimum left value, temp copy */ |
2557 | xfs_mount_t *mp; /* mount structure pointer */ | |
2558 | xfs_agnumber_t sagno; /* starting allocation group number */ | |
2559 | xfs_alloctype_t type; /* input allocation type */ | |
2560 | int bump_rotor = 0; | |
2561 | int no_min = 0; | |
2562 | xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */ | |
2563 | ||
2564 | mp = args->mp; | |
2565 | type = args->otype = args->type; | |
2566 | args->agbno = NULLAGBLOCK; | |
2567 | /* | |
2568 | * Just fix this up, for the case where the last a.g. is shorter | |
2569 | * (or there's only one a.g.) and the caller couldn't easily figure | |
2570 | * that out (xfs_bmap_alloc). | |
2571 | */ | |
2572 | agsize = mp->m_sb.sb_agblocks; | |
2573 | if (args->maxlen > agsize) | |
2574 | args->maxlen = agsize; | |
2575 | if (args->alignment == 0) | |
2576 | args->alignment = 1; | |
2577 | ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount); | |
2578 | ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize); | |
2579 | ASSERT(args->minlen <= args->maxlen); | |
2580 | ASSERT(args->minlen <= agsize); | |
2581 | ASSERT(args->mod < args->prod); | |
2582 | if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount || | |
2583 | XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize || | |
2584 | args->minlen > args->maxlen || args->minlen > agsize || | |
2585 | args->mod >= args->prod) { | |
2586 | args->fsbno = NULLFSBLOCK; | |
0b1b213f | 2587 | trace_xfs_alloc_vextent_badargs(args); |
1da177e4 LT |
2588 | return 0; |
2589 | } | |
2590 | minleft = args->minleft; | |
2591 | ||
2592 | switch (type) { | |
2593 | case XFS_ALLOCTYPE_THIS_AG: | |
2594 | case XFS_ALLOCTYPE_NEAR_BNO: | |
2595 | case XFS_ALLOCTYPE_THIS_BNO: | |
2596 | /* | |
2597 | * These three force us into a single a.g. | |
2598 | */ | |
2599 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
a862e0fd | 2600 | args->pag = xfs_perag_get(mp, args->agno); |
1da177e4 LT |
2601 | args->minleft = 0; |
2602 | error = xfs_alloc_fix_freelist(args, 0); | |
2603 | args->minleft = minleft; | |
2604 | if (error) { | |
0b1b213f | 2605 | trace_xfs_alloc_vextent_nofix(args); |
1da177e4 LT |
2606 | goto error0; |
2607 | } | |
2608 | if (!args->agbp) { | |
0b1b213f | 2609 | trace_xfs_alloc_vextent_noagbp(args); |
1da177e4 LT |
2610 | break; |
2611 | } | |
2612 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2613 | if ((error = xfs_alloc_ag_vextent(args))) | |
2614 | goto error0; | |
1da177e4 LT |
2615 | break; |
2616 | case XFS_ALLOCTYPE_START_BNO: | |
2617 | /* | |
2618 | * Try near allocation first, then anywhere-in-ag after | |
2619 | * the first a.g. fails. | |
2620 | */ | |
3fbbbea3 | 2621 | if ((args->userdata & XFS_ALLOC_INITIAL_USER_DATA) && |
1da177e4 LT |
2622 | (mp->m_flags & XFS_MOUNT_32BITINODES)) { |
2623 | args->fsbno = XFS_AGB_TO_FSB(mp, | |
2624 | ((mp->m_agfrotor / rotorstep) % | |
2625 | mp->m_sb.sb_agcount), 0); | |
2626 | bump_rotor = 1; | |
2627 | } | |
2628 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2629 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2630 | /* FALLTHROUGH */ | |
2631 | case XFS_ALLOCTYPE_ANY_AG: | |
2632 | case XFS_ALLOCTYPE_START_AG: | |
2633 | case XFS_ALLOCTYPE_FIRST_AG: | |
2634 | /* | |
2635 | * Rotate through the allocation groups looking for a winner. | |
2636 | */ | |
2637 | if (type == XFS_ALLOCTYPE_ANY_AG) { | |
2638 | /* | |
2639 | * Start with the last place we left off. | |
2640 | */ | |
2641 | args->agno = sagno = (mp->m_agfrotor / rotorstep) % | |
2642 | mp->m_sb.sb_agcount; | |
2643 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2644 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2645 | } else if (type == XFS_ALLOCTYPE_FIRST_AG) { | |
2646 | /* | |
2647 | * Start with allocation group given by bno. | |
2648 | */ | |
2649 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2650 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2651 | sagno = 0; | |
2652 | flags = 0; | |
2653 | } else { | |
2654 | if (type == XFS_ALLOCTYPE_START_AG) | |
2655 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2656 | /* | |
2657 | * Start with the given allocation group. | |
2658 | */ | |
2659 | args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2660 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2661 | } | |
2662 | /* | |
2663 | * Loop over allocation groups twice; first time with | |
2664 | * trylock set, second time without. | |
2665 | */ | |
1da177e4 | 2666 | for (;;) { |
a862e0fd | 2667 | args->pag = xfs_perag_get(mp, args->agno); |
1da177e4 LT |
2668 | if (no_min) args->minleft = 0; |
2669 | error = xfs_alloc_fix_freelist(args, flags); | |
2670 | args->minleft = minleft; | |
2671 | if (error) { | |
0b1b213f | 2672 | trace_xfs_alloc_vextent_nofix(args); |
1da177e4 LT |
2673 | goto error0; |
2674 | } | |
2675 | /* | |
2676 | * If we get a buffer back then the allocation will fly. | |
2677 | */ | |
2678 | if (args->agbp) { | |
2679 | if ((error = xfs_alloc_ag_vextent(args))) | |
2680 | goto error0; | |
2681 | break; | |
2682 | } | |
0b1b213f CH |
2683 | |
2684 | trace_xfs_alloc_vextent_loopfailed(args); | |
2685 | ||
1da177e4 LT |
2686 | /* |
2687 | * Didn't work, figure out the next iteration. | |
2688 | */ | |
2689 | if (args->agno == sagno && | |
2690 | type == XFS_ALLOCTYPE_START_BNO) | |
2691 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
d210a28c YL |
2692 | /* |
2693 | * For the first allocation, we can try any AG to get | |
2694 | * space. However, if we already have allocated a | |
2695 | * block, we don't want to try AGs whose number is below | |
2696 | * sagno. Otherwise, we may end up with out-of-order | |
2697 | * locking of AGF, which might cause deadlock. | |
2698 | */ | |
2699 | if (++(args->agno) == mp->m_sb.sb_agcount) { | |
2700 | if (args->firstblock != NULLFSBLOCK) | |
2701 | args->agno = sagno; | |
2702 | else | |
2703 | args->agno = 0; | |
2704 | } | |
1da177e4 LT |
2705 | /* |
2706 | * Reached the starting a.g., must either be done | |
2707 | * or switch to non-trylock mode. | |
2708 | */ | |
2709 | if (args->agno == sagno) { | |
2710 | if (no_min == 1) { | |
2711 | args->agbno = NULLAGBLOCK; | |
0b1b213f | 2712 | trace_xfs_alloc_vextent_allfailed(args); |
1da177e4 LT |
2713 | break; |
2714 | } | |
2715 | if (flags == 0) { | |
2716 | no_min = 1; | |
2717 | } else { | |
2718 | flags = 0; | |
2719 | if (type == XFS_ALLOCTYPE_START_BNO) { | |
2720 | args->agbno = XFS_FSB_TO_AGBNO(mp, | |
2721 | args->fsbno); | |
2722 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2723 | } | |
2724 | } | |
2725 | } | |
a862e0fd | 2726 | xfs_perag_put(args->pag); |
1da177e4 | 2727 | } |
1da177e4 LT |
2728 | if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) { |
2729 | if (args->agno == sagno) | |
2730 | mp->m_agfrotor = (mp->m_agfrotor + 1) % | |
2731 | (mp->m_sb.sb_agcount * rotorstep); | |
2732 | else | |
2733 | mp->m_agfrotor = (args->agno * rotorstep + 1) % | |
2734 | (mp->m_sb.sb_agcount * rotorstep); | |
2735 | } | |
2736 | break; | |
2737 | default: | |
2738 | ASSERT(0); | |
2739 | /* NOTREACHED */ | |
2740 | } | |
2741 | if (args->agbno == NULLAGBLOCK) | |
2742 | args->fsbno = NULLFSBLOCK; | |
2743 | else { | |
2744 | args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno); | |
2745 | #ifdef DEBUG | |
2746 | ASSERT(args->len >= args->minlen); | |
2747 | ASSERT(args->len <= args->maxlen); | |
2748 | ASSERT(args->agbno % args->alignment == 0); | |
2749 | XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), | |
2750 | args->len); | |
2751 | #endif | |
3fbbbea3 DC |
2752 | |
2753 | /* Zero the extent if we were asked to do so */ | |
2754 | if (args->userdata & XFS_ALLOC_USERDATA_ZERO) { | |
2755 | error = xfs_zero_extent(args->ip, args->fsbno, args->len); | |
2756 | if (error) | |
2757 | goto error0; | |
2758 | } | |
2759 | ||
1da177e4 | 2760 | } |
a862e0fd | 2761 | xfs_perag_put(args->pag); |
1da177e4 LT |
2762 | return 0; |
2763 | error0: | |
a862e0fd | 2764 | xfs_perag_put(args->pag); |
1da177e4 LT |
2765 | return error; |
2766 | } | |
2767 | ||
4d89e20b DC |
2768 | /* Ensure that the freelist is at full capacity. */ |
2769 | int | |
2770 | xfs_free_extent_fix_freelist( | |
2771 | struct xfs_trans *tp, | |
2772 | xfs_agnumber_t agno, | |
2773 | struct xfs_buf **agbp) | |
1da177e4 | 2774 | { |
4d89e20b DC |
2775 | struct xfs_alloc_arg args; |
2776 | int error; | |
1da177e4 | 2777 | |
4d89e20b | 2778 | memset(&args, 0, sizeof(struct xfs_alloc_arg)); |
1da177e4 LT |
2779 | args.tp = tp; |
2780 | args.mp = tp->t_mountp; | |
4d89e20b | 2781 | args.agno = agno; |
be65b18a DC |
2782 | |
2783 | /* | |
2784 | * validate that the block number is legal - the enables us to detect | |
2785 | * and handle a silent filesystem corruption rather than crashing. | |
2786 | */ | |
be65b18a | 2787 | if (args.agno >= args.mp->m_sb.sb_agcount) |
2451337d | 2788 | return -EFSCORRUPTED; |
be65b18a | 2789 | |
a862e0fd | 2790 | args.pag = xfs_perag_get(args.mp, args.agno); |
be65b18a DC |
2791 | ASSERT(args.pag); |
2792 | ||
2793 | error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING); | |
2794 | if (error) | |
4d89e20b DC |
2795 | goto out; |
2796 | ||
2797 | *agbp = args.agbp; | |
2798 | out: | |
2799 | xfs_perag_put(args.pag); | |
2800 | return error; | |
2801 | } | |
2802 | ||
2803 | /* | |
2804 | * Free an extent. | |
2805 | * Just break up the extent address and hand off to xfs_free_ag_extent | |
2806 | * after fixing up the freelist. | |
2807 | */ | |
2808 | int /* error */ | |
2809 | xfs_free_extent( | |
2810 | struct xfs_trans *tp, /* transaction pointer */ | |
2811 | xfs_fsblock_t bno, /* starting block number of extent */ | |
340785cc DW |
2812 | xfs_extlen_t len, /* length of extent */ |
2813 | struct xfs_owner_info *oinfo) /* extent owner */ | |
4d89e20b DC |
2814 | { |
2815 | struct xfs_mount *mp = tp->t_mountp; | |
2816 | struct xfs_buf *agbp; | |
2817 | xfs_agnumber_t agno = XFS_FSB_TO_AGNO(mp, bno); | |
2818 | xfs_agblock_t agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
2819 | int error; | |
2820 | ||
2821 | ASSERT(len != 0); | |
2822 | ||
ba9e7802 DW |
2823 | if (XFS_TEST_ERROR(false, mp, |
2824 | XFS_ERRTAG_FREE_EXTENT, | |
2825 | XFS_RANDOM_FREE_EXTENT)) | |
2826 | return -EIO; | |
2827 | ||
4d89e20b DC |
2828 | error = xfs_free_extent_fix_freelist(tp, agno, &agbp); |
2829 | if (error) | |
2830 | return error; | |
2831 | ||
2832 | XFS_WANT_CORRUPTED_GOTO(mp, agbno < mp->m_sb.sb_agblocks, err); | |
be65b18a DC |
2833 | |
2834 | /* validate the extent size is legal now we have the agf locked */ | |
4d89e20b DC |
2835 | XFS_WANT_CORRUPTED_GOTO(mp, |
2836 | agbno + len <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_length), | |
2837 | err); | |
be65b18a | 2838 | |
340785cc | 2839 | error = xfs_free_ag_extent(tp, agbp, agno, agbno, len, oinfo, 0); |
4d89e20b DC |
2840 | if (error) |
2841 | goto err; | |
2842 | ||
2843 | xfs_extent_busy_insert(tp, agno, agbno, len, 0); | |
2844 | return 0; | |
2845 | ||
2846 | err: | |
2847 | xfs_trans_brelse(tp, agbp); | |
1da177e4 LT |
2848 | return error; |
2849 | } |