]>
Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_format.h" |
239880ef | 9 | #include "xfs_log_format.h" |
70a9883c | 10 | #include "xfs_shared.h" |
239880ef | 11 | #include "xfs_trans_resv.h" |
a844f451 | 12 | #include "xfs_bit.h" |
1da177e4 | 13 | #include "xfs_mount.h" |
3ab78df2 | 14 | #include "xfs_defer.h" |
1da177e4 | 15 | #include "xfs_btree.h" |
673930c3 | 16 | #include "xfs_rmap.h" |
a4fbe6ab | 17 | #include "xfs_alloc_btree.h" |
1da177e4 | 18 | #include "xfs_alloc.h" |
efc27b52 | 19 | #include "xfs_extent_busy.h" |
e9e899a2 | 20 | #include "xfs_errortag.h" |
1da177e4 | 21 | #include "xfs_error.h" |
0b1b213f | 22 | #include "xfs_trace.h" |
239880ef | 23 | #include "xfs_trans.h" |
4e0e6040 | 24 | #include "xfs_buf_item.h" |
239880ef | 25 | #include "xfs_log.h" |
9bbafc71 | 26 | #include "xfs_ag.h" |
3fd129b6 | 27 | #include "xfs_ag_resv.h" |
f8f2835a BF |
28 | #include "xfs_bmap.h" |
29 | ||
c201d9ca | 30 | struct kmem_cache *xfs_extfree_item_cache; |
1da177e4 | 31 | |
c999a223 | 32 | struct workqueue_struct *xfs_alloc_wq; |
1da177e4 LT |
33 | |
34 | #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b))) | |
35 | ||
36 | #define XFSA_FIXUP_BNO_OK 1 | |
37 | #define XFSA_FIXUP_CNT_OK 2 | |
38 | ||
1da177e4 LT |
39 | STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *); |
40 | STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *); | |
41 | STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *); | |
1da177e4 | 42 | |
a78ee256 DC |
43 | /* |
44 | * Size of the AGFL. For CRC-enabled filesystes we steal a couple of slots in | |
45 | * the beginning of the block for a proper header with the location information | |
46 | * and CRC. | |
47 | */ | |
48 | unsigned int | |
49 | xfs_agfl_size( | |
50 | struct xfs_mount *mp) | |
51 | { | |
52 | unsigned int size = mp->m_sb.sb_sectsize; | |
53 | ||
38c26bfd | 54 | if (xfs_has_crc(mp)) |
a78ee256 DC |
55 | size -= sizeof(struct xfs_agfl); |
56 | ||
57 | return size / sizeof(xfs_agblock_t); | |
58 | } | |
59 | ||
af30dfa1 DW |
60 | unsigned int |
61 | xfs_refc_block( | |
62 | struct xfs_mount *mp) | |
63 | { | |
38c26bfd | 64 | if (xfs_has_rmapbt(mp)) |
af30dfa1 | 65 | return XFS_RMAP_BLOCK(mp) + 1; |
38c26bfd | 66 | if (xfs_has_finobt(mp)) |
af30dfa1 DW |
67 | return XFS_FIBT_BLOCK(mp) + 1; |
68 | return XFS_IBT_BLOCK(mp) + 1; | |
69 | } | |
70 | ||
8018026e DW |
71 | xfs_extlen_t |
72 | xfs_prealloc_blocks( | |
73 | struct xfs_mount *mp) | |
74 | { | |
38c26bfd | 75 | if (xfs_has_reflink(mp)) |
af30dfa1 | 76 | return xfs_refc_block(mp) + 1; |
38c26bfd | 77 | if (xfs_has_rmapbt(mp)) |
8018026e | 78 | return XFS_RMAP_BLOCK(mp) + 1; |
38c26bfd | 79 | if (xfs_has_finobt(mp)) |
8018026e DW |
80 | return XFS_FIBT_BLOCK(mp) + 1; |
81 | return XFS_IBT_BLOCK(mp) + 1; | |
82 | } | |
83 | ||
52548852 DW |
84 | /* |
85 | * In order to avoid ENOSPC-related deadlock caused by out-of-order locking of | |
86 | * AGF buffer (PV 947395), we place constraints on the relationship among | |
87 | * actual allocations for data blocks, freelist blocks, and potential file data | |
88 | * bmap btree blocks. However, these restrictions may result in no actual space | |
89 | * allocated for a delayed extent, for example, a data block in a certain AG is | |
90 | * allocated but there is no additional block for the additional bmap btree | |
91 | * block due to a split of the bmap btree of the file. The result of this may | |
92 | * lead to an infinite loop when the file gets flushed to disk and all delayed | |
93 | * extents need to be actually allocated. To get around this, we explicitly set | |
94 | * aside a few blocks which will not be reserved in delayed allocation. | |
95 | * | |
3fd129b6 DW |
96 | * We need to reserve 4 fsbs _per AG_ for the freelist and 4 more to handle a |
97 | * potential split of the file's bmap btree. | |
52548852 DW |
98 | */ |
99 | unsigned int | |
100 | xfs_alloc_set_aside( | |
101 | struct xfs_mount *mp) | |
102 | { | |
5149fd32 | 103 | return mp->m_sb.sb_agcount * (XFS_ALLOC_AGFL_RESERVE + 4); |
52548852 DW |
104 | } |
105 | ||
106 | /* | |
107 | * When deciding how much space to allocate out of an AG, we limit the | |
108 | * allocation maximum size to the size the AG. However, we cannot use all the | |
109 | * blocks in the AG - some are permanently used by metadata. These | |
110 | * blocks are generally: | |
111 | * - the AG superblock, AGF, AGI and AGFL | |
112 | * - the AGF (bno and cnt) and AGI btree root blocks, and optionally | |
113 | * the AGI free inode and rmap btree root blocks. | |
114 | * - blocks on the AGFL according to xfs_alloc_set_aside() limits | |
115 | * - the rmapbt root block | |
116 | * | |
117 | * The AG headers are sector sized, so the amount of space they take up is | |
118 | * dependent on filesystem geometry. The others are all single blocks. | |
119 | */ | |
120 | unsigned int | |
121 | xfs_alloc_ag_max_usable( | |
122 | struct xfs_mount *mp) | |
123 | { | |
124 | unsigned int blocks; | |
125 | ||
126 | blocks = XFS_BB_TO_FSB(mp, XFS_FSS_TO_BB(mp, 4)); /* ag headers */ | |
127 | blocks += XFS_ALLOC_AGFL_RESERVE; | |
128 | blocks += 3; /* AGF, AGI btree root blocks */ | |
38c26bfd | 129 | if (xfs_has_finobt(mp)) |
52548852 | 130 | blocks++; /* finobt root block */ |
38c26bfd | 131 | if (xfs_has_rmapbt(mp)) |
52548852 | 132 | blocks++; /* rmap root block */ |
38c26bfd | 133 | if (xfs_has_reflink(mp)) |
d0e853f3 | 134 | blocks++; /* refcount root block */ |
52548852 DW |
135 | |
136 | return mp->m_sb.sb_agblocks - blocks; | |
137 | } | |
138 | ||
fe033cc8 CH |
139 | /* |
140 | * Lookup the record equal to [bno, len] in the btree given by cur. | |
141 | */ | |
142 | STATIC int /* error */ | |
143 | xfs_alloc_lookup_eq( | |
144 | struct xfs_btree_cur *cur, /* btree cursor */ | |
145 | xfs_agblock_t bno, /* starting block of extent */ | |
146 | xfs_extlen_t len, /* length of extent */ | |
147 | int *stat) /* success/failure */ | |
148 | { | |
f6b428a4 BF |
149 | int error; |
150 | ||
fe033cc8 CH |
151 | cur->bc_rec.a.ar_startblock = bno; |
152 | cur->bc_rec.a.ar_blockcount = len; | |
f6b428a4 | 153 | error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); |
c4aa10d0 | 154 | cur->bc_ag.abt.active = (*stat == 1); |
f6b428a4 | 155 | return error; |
fe033cc8 CH |
156 | } |
157 | ||
158 | /* | |
159 | * Lookup the first record greater than or equal to [bno, len] | |
160 | * in the btree given by cur. | |
161 | */ | |
a66d6363 | 162 | int /* error */ |
fe033cc8 CH |
163 | xfs_alloc_lookup_ge( |
164 | struct xfs_btree_cur *cur, /* btree cursor */ | |
165 | xfs_agblock_t bno, /* starting block of extent */ | |
166 | xfs_extlen_t len, /* length of extent */ | |
167 | int *stat) /* success/failure */ | |
168 | { | |
f6b428a4 BF |
169 | int error; |
170 | ||
fe033cc8 CH |
171 | cur->bc_rec.a.ar_startblock = bno; |
172 | cur->bc_rec.a.ar_blockcount = len; | |
f6b428a4 | 173 | error = xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); |
c4aa10d0 | 174 | cur->bc_ag.abt.active = (*stat == 1); |
f6b428a4 | 175 | return error; |
fe033cc8 CH |
176 | } |
177 | ||
178 | /* | |
179 | * Lookup the first record less than or equal to [bno, len] | |
180 | * in the btree given by cur. | |
181 | */ | |
ce1d802e | 182 | int /* error */ |
fe033cc8 CH |
183 | xfs_alloc_lookup_le( |
184 | struct xfs_btree_cur *cur, /* btree cursor */ | |
185 | xfs_agblock_t bno, /* starting block of extent */ | |
186 | xfs_extlen_t len, /* length of extent */ | |
187 | int *stat) /* success/failure */ | |
188 | { | |
f6b428a4 | 189 | int error; |
fe033cc8 CH |
190 | cur->bc_rec.a.ar_startblock = bno; |
191 | cur->bc_rec.a.ar_blockcount = len; | |
f6b428a4 | 192 | error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat); |
c4aa10d0 | 193 | cur->bc_ag.abt.active = (*stat == 1); |
f6b428a4 BF |
194 | return error; |
195 | } | |
196 | ||
197 | static inline bool | |
198 | xfs_alloc_cur_active( | |
199 | struct xfs_btree_cur *cur) | |
200 | { | |
c4aa10d0 | 201 | return cur && cur->bc_ag.abt.active; |
fe033cc8 CH |
202 | } |
203 | ||
278d0ca1 CH |
204 | /* |
205 | * Update the record referred to by cur to the value given | |
206 | * by [bno, len]. | |
207 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
208 | */ | |
209 | STATIC int /* error */ | |
210 | xfs_alloc_update( | |
211 | struct xfs_btree_cur *cur, /* btree cursor */ | |
212 | xfs_agblock_t bno, /* starting block of extent */ | |
213 | xfs_extlen_t len) /* length of extent */ | |
214 | { | |
215 | union xfs_btree_rec rec; | |
216 | ||
217 | rec.alloc.ar_startblock = cpu_to_be32(bno); | |
218 | rec.alloc.ar_blockcount = cpu_to_be32(len); | |
219 | return xfs_btree_update(cur, &rec); | |
220 | } | |
fe033cc8 | 221 | |
8cc938fe CH |
222 | /* |
223 | * Get the data from the pointed-to record. | |
224 | */ | |
a46db608 | 225 | int /* error */ |
8cc938fe CH |
226 | xfs_alloc_get_rec( |
227 | struct xfs_btree_cur *cur, /* btree cursor */ | |
228 | xfs_agblock_t *bno, /* output: starting block of extent */ | |
229 | xfs_extlen_t *len, /* output: length of extent */ | |
230 | int *stat) /* output: success/failure */ | |
231 | { | |
9e6c08d4 | 232 | struct xfs_mount *mp = cur->bc_mp; |
50f02fe3 | 233 | xfs_agnumber_t agno = cur->bc_ag.pag->pag_agno; |
8cc938fe CH |
234 | union xfs_btree_rec *rec; |
235 | int error; | |
236 | ||
237 | error = xfs_btree_get_rec(cur, &rec, stat); | |
a37f7b12 DW |
238 | if (error || !(*stat)) |
239 | return error; | |
a37f7b12 DW |
240 | |
241 | *bno = be32_to_cpu(rec->alloc.ar_startblock); | |
242 | *len = be32_to_cpu(rec->alloc.ar_blockcount); | |
243 | ||
efe80327 CM |
244 | if (*len == 0) |
245 | goto out_bad_rec; | |
246 | ||
9e6c08d4 DC |
247 | /* check for valid extent range, including overflow */ |
248 | if (!xfs_verify_agbno(mp, agno, *bno)) | |
249 | goto out_bad_rec; | |
250 | if (*bno > *bno + *len) | |
251 | goto out_bad_rec; | |
252 | if (!xfs_verify_agbno(mp, agno, *bno + *len - 1)) | |
253 | goto out_bad_rec; | |
254 | ||
255 | return 0; | |
256 | ||
257 | out_bad_rec: | |
258 | xfs_warn(mp, | |
259 | "%s Freespace BTree record corruption in AG %d detected!", | |
260 | cur->bc_btnum == XFS_BTNUM_BNO ? "Block" : "Size", agno); | |
261 | xfs_warn(mp, | |
262 | "start block 0x%x block count 0x%x", *bno, *len); | |
263 | return -EFSCORRUPTED; | |
8cc938fe CH |
264 | } |
265 | ||
1da177e4 LT |
266 | /* |
267 | * Compute aligned version of the found extent. | |
268 | * Takes alignment and min length into account. | |
269 | */ | |
ebf55872 | 270 | STATIC bool |
1da177e4 | 271 | xfs_alloc_compute_aligned( |
86fa8af6 | 272 | xfs_alloc_arg_t *args, /* allocation argument structure */ |
1da177e4 LT |
273 | xfs_agblock_t foundbno, /* starting block in found extent */ |
274 | xfs_extlen_t foundlen, /* length in found extent */ | |
1da177e4 | 275 | xfs_agblock_t *resbno, /* result block number */ |
ebf55872 CH |
276 | xfs_extlen_t *reslen, /* result length */ |
277 | unsigned *busy_gen) | |
1da177e4 | 278 | { |
ebf55872 CH |
279 | xfs_agblock_t bno = foundbno; |
280 | xfs_extlen_t len = foundlen; | |
bfe46d4e | 281 | xfs_extlen_t diff; |
ebf55872 | 282 | bool busy; |
1da177e4 | 283 | |
e26f0501 | 284 | /* Trim busy sections out of found extent */ |
ebf55872 | 285 | busy = xfs_extent_busy_trim(args, &bno, &len, busy_gen); |
e26f0501 | 286 | |
bfe46d4e BF |
287 | /* |
288 | * If we have a largish extent that happens to start before min_agbno, | |
289 | * see if we can shift it into range... | |
290 | */ | |
291 | if (bno < args->min_agbno && bno + len > args->min_agbno) { | |
292 | diff = args->min_agbno - bno; | |
293 | if (len > diff) { | |
294 | bno += diff; | |
295 | len -= diff; | |
296 | } | |
297 | } | |
298 | ||
e26f0501 CH |
299 | if (args->alignment > 1 && len >= args->minlen) { |
300 | xfs_agblock_t aligned_bno = roundup(bno, args->alignment); | |
bfe46d4e BF |
301 | |
302 | diff = aligned_bno - bno; | |
e26f0501 CH |
303 | |
304 | *resbno = aligned_bno; | |
305 | *reslen = diff >= len ? 0 : len - diff; | |
1da177e4 | 306 | } else { |
e26f0501 CH |
307 | *resbno = bno; |
308 | *reslen = len; | |
1da177e4 | 309 | } |
ebf55872 CH |
310 | |
311 | return busy; | |
1da177e4 LT |
312 | } |
313 | ||
314 | /* | |
315 | * Compute best start block and diff for "near" allocations. | |
316 | * freelen >= wantlen already checked by caller. | |
317 | */ | |
318 | STATIC xfs_extlen_t /* difference value (absolute) */ | |
319 | xfs_alloc_compute_diff( | |
320 | xfs_agblock_t wantbno, /* target starting block */ | |
321 | xfs_extlen_t wantlen, /* target length */ | |
322 | xfs_extlen_t alignment, /* target alignment */ | |
292378ed | 323 | int datatype, /* are we allocating data? */ |
1da177e4 LT |
324 | xfs_agblock_t freebno, /* freespace's starting block */ |
325 | xfs_extlen_t freelen, /* freespace's length */ | |
326 | xfs_agblock_t *newbnop) /* result: best start block from free */ | |
327 | { | |
328 | xfs_agblock_t freeend; /* end of freespace extent */ | |
329 | xfs_agblock_t newbno1; /* return block number */ | |
330 | xfs_agblock_t newbno2; /* other new block number */ | |
331 | xfs_extlen_t newlen1=0; /* length with newbno1 */ | |
332 | xfs_extlen_t newlen2=0; /* length with newbno2 */ | |
333 | xfs_agblock_t wantend; /* end of target extent */ | |
c34d570d | 334 | bool userdata = datatype & XFS_ALLOC_USERDATA; |
1da177e4 LT |
335 | |
336 | ASSERT(freelen >= wantlen); | |
337 | freeend = freebno + freelen; | |
338 | wantend = wantbno + wantlen; | |
211d022c JK |
339 | /* |
340 | * We want to allocate from the start of a free extent if it is past | |
341 | * the desired block or if we are allocating user data and the free | |
342 | * extent is before desired block. The second case is there to allow | |
343 | * for contiguous allocation from the remaining free space if the file | |
344 | * grows in the short term. | |
345 | */ | |
346 | if (freebno >= wantbno || (userdata && freeend < wantend)) { | |
1da177e4 LT |
347 | if ((newbno1 = roundup(freebno, alignment)) >= freeend) |
348 | newbno1 = NULLAGBLOCK; | |
349 | } else if (freeend >= wantend && alignment > 1) { | |
350 | newbno1 = roundup(wantbno, alignment); | |
351 | newbno2 = newbno1 - alignment; | |
352 | if (newbno1 >= freeend) | |
353 | newbno1 = NULLAGBLOCK; | |
354 | else | |
355 | newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1); | |
356 | if (newbno2 < freebno) | |
357 | newbno2 = NULLAGBLOCK; | |
358 | else | |
359 | newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2); | |
360 | if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) { | |
361 | if (newlen1 < newlen2 || | |
362 | (newlen1 == newlen2 && | |
363 | XFS_ABSDIFF(newbno1, wantbno) > | |
364 | XFS_ABSDIFF(newbno2, wantbno))) | |
365 | newbno1 = newbno2; | |
366 | } else if (newbno2 != NULLAGBLOCK) | |
367 | newbno1 = newbno2; | |
368 | } else if (freeend >= wantend) { | |
369 | newbno1 = wantbno; | |
370 | } else if (alignment > 1) { | |
371 | newbno1 = roundup(freeend - wantlen, alignment); | |
372 | if (newbno1 > freeend - wantlen && | |
373 | newbno1 - alignment >= freebno) | |
374 | newbno1 -= alignment; | |
375 | else if (newbno1 >= freeend) | |
376 | newbno1 = NULLAGBLOCK; | |
377 | } else | |
378 | newbno1 = freeend - wantlen; | |
379 | *newbnop = newbno1; | |
380 | return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno); | |
381 | } | |
382 | ||
383 | /* | |
384 | * Fix up the length, based on mod and prod. | |
385 | * len should be k * prod + mod for some k. | |
386 | * If len is too small it is returned unchanged. | |
387 | * If len hits maxlen it is left alone. | |
388 | */ | |
389 | STATIC void | |
390 | xfs_alloc_fix_len( | |
391 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
392 | { | |
393 | xfs_extlen_t k; | |
394 | xfs_extlen_t rlen; | |
395 | ||
396 | ASSERT(args->mod < args->prod); | |
397 | rlen = args->len; | |
398 | ASSERT(rlen >= args->minlen); | |
399 | ASSERT(rlen <= args->maxlen); | |
400 | if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen || | |
401 | (args->mod == 0 && rlen < args->prod)) | |
402 | return; | |
403 | k = rlen % args->prod; | |
404 | if (k == args->mod) | |
405 | return; | |
30265117 JK |
406 | if (k > args->mod) |
407 | rlen = rlen - (k - args->mod); | |
408 | else | |
409 | rlen = rlen - args->prod + (args->mod - k); | |
3790a8cd | 410 | /* casts to (int) catch length underflows */ |
30265117 JK |
411 | if ((int)rlen < (int)args->minlen) |
412 | return; | |
413 | ASSERT(rlen >= args->minlen && rlen <= args->maxlen); | |
414 | ASSERT(rlen % args->prod == args->mod); | |
54fee133 CH |
415 | ASSERT(args->pag->pagf_freeblks + args->pag->pagf_flcount >= |
416 | rlen + args->minleft); | |
1da177e4 LT |
417 | args->len = rlen; |
418 | } | |
419 | ||
1da177e4 LT |
420 | /* |
421 | * Update the two btrees, logically removing from freespace the extent | |
422 | * starting at rbno, rlen blocks. The extent is contained within the | |
423 | * actual (current) free extent fbno for flen blocks. | |
424 | * Flags are passed in indicating whether the cursors are set to the | |
425 | * relevant records. | |
426 | */ | |
427 | STATIC int /* error code */ | |
428 | xfs_alloc_fixup_trees( | |
ae127f08 DW |
429 | struct xfs_btree_cur *cnt_cur, /* cursor for by-size btree */ |
430 | struct xfs_btree_cur *bno_cur, /* cursor for by-block btree */ | |
1da177e4 LT |
431 | xfs_agblock_t fbno, /* starting block of free extent */ |
432 | xfs_extlen_t flen, /* length of free extent */ | |
433 | xfs_agblock_t rbno, /* starting block of returned extent */ | |
434 | xfs_extlen_t rlen, /* length of returned extent */ | |
435 | int flags) /* flags, XFSA_FIXUP_... */ | |
436 | { | |
437 | int error; /* error code */ | |
438 | int i; /* operation results */ | |
439 | xfs_agblock_t nfbno1; /* first new free startblock */ | |
440 | xfs_agblock_t nfbno2; /* second new free startblock */ | |
441 | xfs_extlen_t nflen1=0; /* first new free length */ | |
442 | xfs_extlen_t nflen2=0; /* second new free length */ | |
5fb5aeee ES |
443 | struct xfs_mount *mp; |
444 | ||
445 | mp = cnt_cur->bc_mp; | |
1da177e4 LT |
446 | |
447 | /* | |
448 | * Look up the record in the by-size tree if necessary. | |
449 | */ | |
450 | if (flags & XFSA_FIXUP_CNT_OK) { | |
451 | #ifdef DEBUG | |
452 | if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i))) | |
453 | return error; | |
f9e03706 DW |
454 | if (XFS_IS_CORRUPT(mp, |
455 | i != 1 || | |
456 | nfbno1 != fbno || | |
457 | nflen1 != flen)) | |
458 | return -EFSCORRUPTED; | |
1da177e4 LT |
459 | #endif |
460 | } else { | |
461 | if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i))) | |
462 | return error; | |
f9e03706 DW |
463 | if (XFS_IS_CORRUPT(mp, i != 1)) |
464 | return -EFSCORRUPTED; | |
1da177e4 LT |
465 | } |
466 | /* | |
467 | * Look up the record in the by-block tree if necessary. | |
468 | */ | |
469 | if (flags & XFSA_FIXUP_BNO_OK) { | |
470 | #ifdef DEBUG | |
471 | if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i))) | |
472 | return error; | |
f9e03706 DW |
473 | if (XFS_IS_CORRUPT(mp, |
474 | i != 1 || | |
475 | nfbno1 != fbno || | |
476 | nflen1 != flen)) | |
477 | return -EFSCORRUPTED; | |
1da177e4 LT |
478 | #endif |
479 | } else { | |
480 | if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i))) | |
481 | return error; | |
f9e03706 DW |
482 | if (XFS_IS_CORRUPT(mp, i != 1)) |
483 | return -EFSCORRUPTED; | |
1da177e4 | 484 | } |
7cc95a82 | 485 | |
1da177e4 | 486 | #ifdef DEBUG |
7cc95a82 CH |
487 | if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) { |
488 | struct xfs_btree_block *bnoblock; | |
489 | struct xfs_btree_block *cntblock; | |
490 | ||
6ca444cf DW |
491 | bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_levels[0].bp); |
492 | cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_levels[0].bp); | |
1da177e4 | 493 | |
f9e03706 DW |
494 | if (XFS_IS_CORRUPT(mp, |
495 | bnoblock->bb_numrecs != | |
496 | cntblock->bb_numrecs)) | |
497 | return -EFSCORRUPTED; | |
1da177e4 LT |
498 | } |
499 | #endif | |
7cc95a82 | 500 | |
1da177e4 LT |
501 | /* |
502 | * Deal with all four cases: the allocated record is contained | |
503 | * within the freespace record, so we can have new freespace | |
504 | * at either (or both) end, or no freespace remaining. | |
505 | */ | |
506 | if (rbno == fbno && rlen == flen) | |
507 | nfbno1 = nfbno2 = NULLAGBLOCK; | |
508 | else if (rbno == fbno) { | |
509 | nfbno1 = rbno + rlen; | |
510 | nflen1 = flen - rlen; | |
511 | nfbno2 = NULLAGBLOCK; | |
512 | } else if (rbno + rlen == fbno + flen) { | |
513 | nfbno1 = fbno; | |
514 | nflen1 = flen - rlen; | |
515 | nfbno2 = NULLAGBLOCK; | |
516 | } else { | |
517 | nfbno1 = fbno; | |
518 | nflen1 = rbno - fbno; | |
519 | nfbno2 = rbno + rlen; | |
520 | nflen2 = (fbno + flen) - nfbno2; | |
521 | } | |
522 | /* | |
523 | * Delete the entry from the by-size btree. | |
524 | */ | |
91cca5df | 525 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 526 | return error; |
f9e03706 DW |
527 | if (XFS_IS_CORRUPT(mp, i != 1)) |
528 | return -EFSCORRUPTED; | |
1da177e4 LT |
529 | /* |
530 | * Add new by-size btree entry(s). | |
531 | */ | |
532 | if (nfbno1 != NULLAGBLOCK) { | |
533 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i))) | |
534 | return error; | |
f9e03706 DW |
535 | if (XFS_IS_CORRUPT(mp, i != 0)) |
536 | return -EFSCORRUPTED; | |
4b22a571 | 537 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 538 | return error; |
f9e03706 DW |
539 | if (XFS_IS_CORRUPT(mp, i != 1)) |
540 | return -EFSCORRUPTED; | |
1da177e4 LT |
541 | } |
542 | if (nfbno2 != NULLAGBLOCK) { | |
543 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i))) | |
544 | return error; | |
f9e03706 DW |
545 | if (XFS_IS_CORRUPT(mp, i != 0)) |
546 | return -EFSCORRUPTED; | |
4b22a571 | 547 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 548 | return error; |
f9e03706 DW |
549 | if (XFS_IS_CORRUPT(mp, i != 1)) |
550 | return -EFSCORRUPTED; | |
1da177e4 LT |
551 | } |
552 | /* | |
553 | * Fix up the by-block btree entry(s). | |
554 | */ | |
555 | if (nfbno1 == NULLAGBLOCK) { | |
556 | /* | |
557 | * No remaining freespace, just delete the by-block tree entry. | |
558 | */ | |
91cca5df | 559 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 | 560 | return error; |
f9e03706 DW |
561 | if (XFS_IS_CORRUPT(mp, i != 1)) |
562 | return -EFSCORRUPTED; | |
1da177e4 LT |
563 | } else { |
564 | /* | |
565 | * Update the by-block entry to start later|be shorter. | |
566 | */ | |
567 | if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1))) | |
568 | return error; | |
569 | } | |
570 | if (nfbno2 != NULLAGBLOCK) { | |
571 | /* | |
572 | * 2 resulting free entries, need to add one. | |
573 | */ | |
574 | if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i))) | |
575 | return error; | |
f9e03706 DW |
576 | if (XFS_IS_CORRUPT(mp, i != 0)) |
577 | return -EFSCORRUPTED; | |
4b22a571 | 578 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 | 579 | return error; |
f9e03706 DW |
580 | if (XFS_IS_CORRUPT(mp, i != 1)) |
581 | return -EFSCORRUPTED; | |
1da177e4 LT |
582 | } |
583 | return 0; | |
584 | } | |
585 | ||
a6a781a5 | 586 | static xfs_failaddr_t |
612cfbfe | 587 | xfs_agfl_verify( |
bb80c6d7 DC |
588 | struct xfs_buf *bp) |
589 | { | |
dbd329f1 | 590 | struct xfs_mount *mp = bp->b_mount; |
bb80c6d7 | 591 | struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp); |
183606d8 | 592 | __be32 *agfl_bno = xfs_buf_to_agfl_bno(bp); |
bb80c6d7 DC |
593 | int i; |
594 | ||
b5572597 DW |
595 | /* |
596 | * There is no verification of non-crc AGFLs because mkfs does not | |
597 | * initialise the AGFL to zero or NULL. Hence the only valid part of the | |
598 | * AGFL is what the AGF says is active. We can't get to the AGF, so we | |
599 | * can't verify just those entries are valid. | |
600 | */ | |
38c26bfd | 601 | if (!xfs_has_crc(mp)) |
b5572597 DW |
602 | return NULL; |
603 | ||
39708c20 | 604 | if (!xfs_verify_magic(bp, agfl->agfl_magicnum)) |
a6a781a5 | 605 | return __this_address; |
39708c20 | 606 | if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid)) |
a6a781a5 | 607 | return __this_address; |
77c95bba CH |
608 | /* |
609 | * during growfs operations, the perag is not fully initialised, | |
610 | * so we can't use it for any useful checking. growfs ensures we can't | |
611 | * use it by using uncached buffers that don't have the perag attached | |
612 | * so we can detect and avoid this problem. | |
613 | */ | |
614 | if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno) | |
a6a781a5 | 615 | return __this_address; |
77c95bba | 616 | |
a78ee256 | 617 | for (i = 0; i < xfs_agfl_size(mp); i++) { |
183606d8 CH |
618 | if (be32_to_cpu(agfl_bno[i]) != NULLAGBLOCK && |
619 | be32_to_cpu(agfl_bno[i]) >= mp->m_sb.sb_agblocks) | |
a6a781a5 | 620 | return __this_address; |
bb80c6d7 | 621 | } |
a45086e2 | 622 | |
a6a781a5 DW |
623 | if (!xfs_log_check_lsn(mp, be64_to_cpu(XFS_BUF_TO_AGFL(bp)->agfl_lsn))) |
624 | return __this_address; | |
625 | return NULL; | |
77c95bba CH |
626 | } |
627 | ||
628 | static void | |
629 | xfs_agfl_read_verify( | |
630 | struct xfs_buf *bp) | |
631 | { | |
dbd329f1 | 632 | struct xfs_mount *mp = bp->b_mount; |
bc1a09b8 | 633 | xfs_failaddr_t fa; |
77c95bba CH |
634 | |
635 | /* | |
636 | * There is no verification of non-crc AGFLs because mkfs does not | |
637 | * initialise the AGFL to zero or NULL. Hence the only valid part of the | |
638 | * AGFL is what the AGF says is active. We can't get to the AGF, so we | |
639 | * can't verify just those entries are valid. | |
640 | */ | |
38c26bfd | 641 | if (!xfs_has_crc(mp)) |
77c95bba CH |
642 | return; |
643 | ||
ce5028cf | 644 | if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF)) |
bc1a09b8 DW |
645 | xfs_verifier_error(bp, -EFSBADCRC, __this_address); |
646 | else { | |
647 | fa = xfs_agfl_verify(bp); | |
648 | if (fa) | |
649 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
650 | } | |
612cfbfe DC |
651 | } |
652 | ||
1813dd64 | 653 | static void |
612cfbfe DC |
654 | xfs_agfl_write_verify( |
655 | struct xfs_buf *bp) | |
656 | { | |
dbd329f1 | 657 | struct xfs_mount *mp = bp->b_mount; |
fb1755a6 | 658 | struct xfs_buf_log_item *bip = bp->b_log_item; |
bc1a09b8 | 659 | xfs_failaddr_t fa; |
612cfbfe | 660 | |
77c95bba | 661 | /* no verification of non-crc AGFLs */ |
38c26bfd | 662 | if (!xfs_has_crc(mp)) |
77c95bba CH |
663 | return; |
664 | ||
bc1a09b8 DW |
665 | fa = xfs_agfl_verify(bp); |
666 | if (fa) { | |
667 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
77c95bba CH |
668 | return; |
669 | } | |
670 | ||
671 | if (bip) | |
672 | XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
673 | ||
f1dbcd7e | 674 | xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF); |
bb80c6d7 DC |
675 | } |
676 | ||
1813dd64 | 677 | const struct xfs_buf_ops xfs_agfl_buf_ops = { |
233135b7 | 678 | .name = "xfs_agfl", |
39708c20 | 679 | .magic = { cpu_to_be32(XFS_AGFL_MAGIC), cpu_to_be32(XFS_AGFL_MAGIC) }, |
1813dd64 DC |
680 | .verify_read = xfs_agfl_read_verify, |
681 | .verify_write = xfs_agfl_write_verify, | |
b5572597 | 682 | .verify_struct = xfs_agfl_verify, |
1813dd64 DC |
683 | }; |
684 | ||
1da177e4 LT |
685 | /* |
686 | * Read in the allocation group free block array. | |
687 | */ | |
26788097 | 688 | int /* error */ |
1da177e4 LT |
689 | xfs_alloc_read_agfl( |
690 | xfs_mount_t *mp, /* mount point structure */ | |
691 | xfs_trans_t *tp, /* transaction pointer */ | |
692 | xfs_agnumber_t agno, /* allocation group number */ | |
e8222613 | 693 | struct xfs_buf **bpp) /* buffer for the ag free block array */ |
1da177e4 | 694 | { |
e8222613 | 695 | struct xfs_buf *bp; /* return value */ |
1da177e4 LT |
696 | int error; |
697 | ||
698 | ASSERT(agno != NULLAGNUMBER); | |
699 | error = xfs_trans_read_buf( | |
700 | mp, tp, mp->m_ddev_targp, | |
701 | XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)), | |
1813dd64 | 702 | XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops); |
1da177e4 LT |
703 | if (error) |
704 | return error; | |
38f23232 | 705 | xfs_buf_set_ref(bp, XFS_AGFL_REF); |
1da177e4 LT |
706 | *bpp = bp; |
707 | return 0; | |
708 | } | |
709 | ||
ecb6928f CH |
710 | STATIC int |
711 | xfs_alloc_update_counters( | |
712 | struct xfs_trans *tp, | |
ecb6928f CH |
713 | struct xfs_buf *agbp, |
714 | long len) | |
715 | { | |
9798f615 | 716 | struct xfs_agf *agf = agbp->b_addr; |
ecb6928f | 717 | |
92a00544 | 718 | agbp->b_pag->pagf_freeblks += len; |
ecb6928f CH |
719 | be32_add_cpu(&agf->agf_freeblks, len); |
720 | ||
ecb6928f | 721 | if (unlikely(be32_to_cpu(agf->agf_freeblks) > |
a5155b87 | 722 | be32_to_cpu(agf->agf_length))) { |
8d57c216 | 723 | xfs_buf_mark_corrupt(agbp); |
2451337d | 724 | return -EFSCORRUPTED; |
a5155b87 | 725 | } |
ecb6928f CH |
726 | |
727 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS); | |
728 | return 0; | |
729 | } | |
730 | ||
1da177e4 | 731 | /* |
f5e7dbea | 732 | * Block allocation algorithm and data structures. |
1da177e4 | 733 | */ |
f5e7dbea BF |
734 | struct xfs_alloc_cur { |
735 | struct xfs_btree_cur *cnt; /* btree cursors */ | |
736 | struct xfs_btree_cur *bnolt; | |
737 | struct xfs_btree_cur *bnogt; | |
dc8e69bd | 738 | xfs_extlen_t cur_len;/* current search length */ |
c62321a2 BF |
739 | xfs_agblock_t rec_bno;/* extent startblock */ |
740 | xfs_extlen_t rec_len;/* extent length */ | |
741 | xfs_agblock_t bno; /* alloc bno */ | |
742 | xfs_extlen_t len; /* alloc len */ | |
743 | xfs_extlen_t diff; /* diff from search bno */ | |
d6d3aff2 BF |
744 | unsigned int busy_gen;/* busy state */ |
745 | bool busy; | |
f5e7dbea BF |
746 | }; |
747 | ||
748 | /* | |
749 | * Set up cursors, etc. in the extent allocation cursor. This function can be | |
750 | * called multiple times to reset an initialized structure without having to | |
751 | * reallocate cursors. | |
752 | */ | |
753 | static int | |
754 | xfs_alloc_cur_setup( | |
755 | struct xfs_alloc_arg *args, | |
756 | struct xfs_alloc_cur *acur) | |
757 | { | |
758 | int error; | |
759 | int i; | |
760 | ||
761 | ASSERT(args->alignment == 1 || args->type != XFS_ALLOCTYPE_THIS_BNO); | |
762 | ||
dc8e69bd | 763 | acur->cur_len = args->maxlen; |
c62321a2 BF |
764 | acur->rec_bno = 0; |
765 | acur->rec_len = 0; | |
766 | acur->bno = 0; | |
767 | acur->len = 0; | |
396bbf3c | 768 | acur->diff = -1; |
d6d3aff2 BF |
769 | acur->busy = false; |
770 | acur->busy_gen = 0; | |
771 | ||
f5e7dbea BF |
772 | /* |
773 | * Perform an initial cntbt lookup to check for availability of maxlen | |
774 | * extents. If this fails, we'll return -ENOSPC to signal the caller to | |
775 | * attempt a small allocation. | |
776 | */ | |
777 | if (!acur->cnt) | |
778 | acur->cnt = xfs_allocbt_init_cursor(args->mp, args->tp, | |
289d38d2 | 779 | args->agbp, args->pag, XFS_BTNUM_CNT); |
f5e7dbea BF |
780 | error = xfs_alloc_lookup_ge(acur->cnt, 0, args->maxlen, &i); |
781 | if (error) | |
782 | return error; | |
783 | ||
784 | /* | |
785 | * Allocate the bnobt left and right search cursors. | |
786 | */ | |
787 | if (!acur->bnolt) | |
788 | acur->bnolt = xfs_allocbt_init_cursor(args->mp, args->tp, | |
289d38d2 | 789 | args->agbp, args->pag, XFS_BTNUM_BNO); |
f5e7dbea BF |
790 | if (!acur->bnogt) |
791 | acur->bnogt = xfs_allocbt_init_cursor(args->mp, args->tp, | |
289d38d2 | 792 | args->agbp, args->pag, XFS_BTNUM_BNO); |
f5e7dbea BF |
793 | return i == 1 ? 0 : -ENOSPC; |
794 | } | |
795 | ||
796 | static void | |
797 | xfs_alloc_cur_close( | |
798 | struct xfs_alloc_cur *acur, | |
799 | bool error) | |
800 | { | |
801 | int cur_error = XFS_BTREE_NOERROR; | |
802 | ||
803 | if (error) | |
804 | cur_error = XFS_BTREE_ERROR; | |
805 | ||
806 | if (acur->cnt) | |
807 | xfs_btree_del_cursor(acur->cnt, cur_error); | |
808 | if (acur->bnolt) | |
809 | xfs_btree_del_cursor(acur->bnolt, cur_error); | |
810 | if (acur->bnogt) | |
811 | xfs_btree_del_cursor(acur->bnogt, cur_error); | |
812 | acur->cnt = acur->bnolt = acur->bnogt = NULL; | |
813 | } | |
1da177e4 | 814 | |
396bbf3c BF |
815 | /* |
816 | * Check an extent for allocation and track the best available candidate in the | |
817 | * allocation structure. The cursor is deactivated if it has entered an out of | |
818 | * range state based on allocation arguments. Optionally return the extent | |
819 | * extent geometry and allocation status if requested by the caller. | |
820 | */ | |
821 | static int | |
822 | xfs_alloc_cur_check( | |
823 | struct xfs_alloc_arg *args, | |
824 | struct xfs_alloc_cur *acur, | |
825 | struct xfs_btree_cur *cur, | |
826 | int *new) | |
827 | { | |
828 | int error, i; | |
829 | xfs_agblock_t bno, bnoa, bnew; | |
830 | xfs_extlen_t len, lena, diff = -1; | |
831 | bool busy; | |
832 | unsigned busy_gen = 0; | |
833 | bool deactivate = false; | |
fec0afda | 834 | bool isbnobt = cur->bc_btnum == XFS_BTNUM_BNO; |
396bbf3c BF |
835 | |
836 | *new = 0; | |
837 | ||
838 | error = xfs_alloc_get_rec(cur, &bno, &len, &i); | |
839 | if (error) | |
840 | return error; | |
f9e03706 DW |
841 | if (XFS_IS_CORRUPT(args->mp, i != 1)) |
842 | return -EFSCORRUPTED; | |
396bbf3c BF |
843 | |
844 | /* | |
845 | * Check minlen and deactivate a cntbt cursor if out of acceptable size | |
846 | * range (i.e., walking backwards looking for a minlen extent). | |
847 | */ | |
848 | if (len < args->minlen) { | |
fec0afda | 849 | deactivate = !isbnobt; |
396bbf3c BF |
850 | goto out; |
851 | } | |
852 | ||
853 | busy = xfs_alloc_compute_aligned(args, bno, len, &bnoa, &lena, | |
854 | &busy_gen); | |
855 | acur->busy |= busy; | |
856 | if (busy) | |
857 | acur->busy_gen = busy_gen; | |
858 | /* deactivate a bnobt cursor outside of locality range */ | |
fec0afda BF |
859 | if (bnoa < args->min_agbno || bnoa > args->max_agbno) { |
860 | deactivate = isbnobt; | |
396bbf3c | 861 | goto out; |
fec0afda | 862 | } |
396bbf3c BF |
863 | if (lena < args->minlen) |
864 | goto out; | |
865 | ||
866 | args->len = XFS_EXTLEN_MIN(lena, args->maxlen); | |
867 | xfs_alloc_fix_len(args); | |
868 | ASSERT(args->len >= args->minlen); | |
869 | if (args->len < acur->len) | |
870 | goto out; | |
871 | ||
872 | /* | |
873 | * We have an aligned record that satisfies minlen and beats or matches | |
874 | * the candidate extent size. Compare locality for near allocation mode. | |
875 | */ | |
876 | ASSERT(args->type == XFS_ALLOCTYPE_NEAR_BNO); | |
877 | diff = xfs_alloc_compute_diff(args->agbno, args->len, | |
878 | args->alignment, args->datatype, | |
879 | bnoa, lena, &bnew); | |
880 | if (bnew == NULLAGBLOCK) | |
881 | goto out; | |
fec0afda BF |
882 | |
883 | /* | |
884 | * Deactivate a bnobt cursor with worse locality than the current best. | |
885 | */ | |
886 | if (diff > acur->diff) { | |
887 | deactivate = isbnobt; | |
396bbf3c | 888 | goto out; |
fec0afda | 889 | } |
396bbf3c BF |
890 | |
891 | ASSERT(args->len > acur->len || | |
892 | (args->len == acur->len && diff <= acur->diff)); | |
893 | acur->rec_bno = bno; | |
894 | acur->rec_len = len; | |
895 | acur->bno = bnew; | |
896 | acur->len = args->len; | |
897 | acur->diff = diff; | |
898 | *new = 1; | |
899 | ||
78d7aabd BF |
900 | /* |
901 | * We're done if we found a perfect allocation. This only deactivates | |
902 | * the current cursor, but this is just an optimization to terminate a | |
903 | * cntbt search that otherwise runs to the edge of the tree. | |
904 | */ | |
905 | if (acur->diff == 0 && acur->len == args->maxlen) | |
906 | deactivate = true; | |
396bbf3c BF |
907 | out: |
908 | if (deactivate) | |
c4aa10d0 | 909 | cur->bc_ag.abt.active = false; |
396bbf3c BF |
910 | trace_xfs_alloc_cur_check(args->mp, cur->bc_btnum, bno, len, diff, |
911 | *new); | |
912 | return 0; | |
913 | } | |
914 | ||
d2968825 BF |
915 | /* |
916 | * Complete an allocation of a candidate extent. Remove the extent from both | |
917 | * trees and update the args structure. | |
918 | */ | |
919 | STATIC int | |
920 | xfs_alloc_cur_finish( | |
921 | struct xfs_alloc_arg *args, | |
922 | struct xfs_alloc_cur *acur) | |
923 | { | |
9798f615 | 924 | struct xfs_agf __maybe_unused *agf = args->agbp->b_addr; |
d2968825 BF |
925 | int error; |
926 | ||
927 | ASSERT(acur->cnt && acur->bnolt); | |
928 | ASSERT(acur->bno >= acur->rec_bno); | |
929 | ASSERT(acur->bno + acur->len <= acur->rec_bno + acur->rec_len); | |
9798f615 | 930 | ASSERT(acur->rec_bno + acur->rec_len <= be32_to_cpu(agf->agf_length)); |
d2968825 BF |
931 | |
932 | error = xfs_alloc_fixup_trees(acur->cnt, acur->bnolt, acur->rec_bno, | |
933 | acur->rec_len, acur->bno, acur->len, 0); | |
934 | if (error) | |
935 | return error; | |
936 | ||
937 | args->agbno = acur->bno; | |
938 | args->len = acur->len; | |
939 | args->wasfromfl = 0; | |
940 | ||
941 | trace_xfs_alloc_cur(args); | |
942 | return 0; | |
943 | } | |
944 | ||
dc8e69bd BF |
945 | /* |
946 | * Locality allocation lookup algorithm. This expects a cntbt cursor and uses | |
947 | * bno optimized lookup to search for extents with ideal size and locality. | |
948 | */ | |
949 | STATIC int | |
950 | xfs_alloc_cntbt_iter( | |
951 | struct xfs_alloc_arg *args, | |
952 | struct xfs_alloc_cur *acur) | |
953 | { | |
954 | struct xfs_btree_cur *cur = acur->cnt; | |
955 | xfs_agblock_t bno; | |
956 | xfs_extlen_t len, cur_len; | |
957 | int error; | |
958 | int i; | |
959 | ||
960 | if (!xfs_alloc_cur_active(cur)) | |
961 | return 0; | |
962 | ||
963 | /* locality optimized lookup */ | |
964 | cur_len = acur->cur_len; | |
965 | error = xfs_alloc_lookup_ge(cur, args->agbno, cur_len, &i); | |
966 | if (error) | |
967 | return error; | |
968 | if (i == 0) | |
969 | return 0; | |
970 | error = xfs_alloc_get_rec(cur, &bno, &len, &i); | |
971 | if (error) | |
972 | return error; | |
973 | ||
974 | /* check the current record and update search length from it */ | |
975 | error = xfs_alloc_cur_check(args, acur, cur, &i); | |
976 | if (error) | |
977 | return error; | |
978 | ASSERT(len >= acur->cur_len); | |
979 | acur->cur_len = len; | |
980 | ||
981 | /* | |
982 | * We looked up the first record >= [agbno, len] above. The agbno is a | |
983 | * secondary key and so the current record may lie just before or after | |
984 | * agbno. If it is past agbno, check the previous record too so long as | |
985 | * the length matches as it may be closer. Don't check a smaller record | |
986 | * because that could deactivate our cursor. | |
987 | */ | |
988 | if (bno > args->agbno) { | |
989 | error = xfs_btree_decrement(cur, 0, &i); | |
990 | if (!error && i) { | |
991 | error = xfs_alloc_get_rec(cur, &bno, &len, &i); | |
992 | if (!error && i && len == acur->cur_len) | |
993 | error = xfs_alloc_cur_check(args, acur, cur, | |
994 | &i); | |
995 | } | |
996 | if (error) | |
997 | return error; | |
998 | } | |
999 | ||
1000 | /* | |
1001 | * Increment the search key until we find at least one allocation | |
1002 | * candidate or if the extent we found was larger. Otherwise, double the | |
1003 | * search key to optimize the search. Efficiency is more important here | |
1004 | * than absolute best locality. | |
1005 | */ | |
1006 | cur_len <<= 1; | |
1007 | if (!acur->len || acur->cur_len >= cur_len) | |
1008 | acur->cur_len++; | |
1009 | else | |
1010 | acur->cur_len = cur_len; | |
1011 | ||
1012 | return error; | |
1013 | } | |
1014 | ||
c63cdd4f BF |
1015 | /* |
1016 | * Deal with the case where only small freespaces remain. Either return the | |
1017 | * contents of the last freespace record, or allocate space from the freelist if | |
1018 | * there is nothing in the tree. | |
1019 | */ | |
1020 | STATIC int /* error */ | |
1021 | xfs_alloc_ag_vextent_small( | |
1022 | struct xfs_alloc_arg *args, /* allocation argument structure */ | |
1023 | struct xfs_btree_cur *ccur, /* optional by-size cursor */ | |
1024 | xfs_agblock_t *fbnop, /* result block number */ | |
1025 | xfs_extlen_t *flenp, /* result length */ | |
1026 | int *stat) /* status: 0-freelist, 1-normal/none */ | |
1027 | { | |
9798f615 | 1028 | struct xfs_agf *agf = args->agbp->b_addr; |
c63cdd4f BF |
1029 | int error = 0; |
1030 | xfs_agblock_t fbno = NULLAGBLOCK; | |
1031 | xfs_extlen_t flen = 0; | |
6691cd92 | 1032 | int i = 0; |
c63cdd4f | 1033 | |
6691cd92 BF |
1034 | /* |
1035 | * If a cntbt cursor is provided, try to allocate the largest record in | |
1036 | * the tree. Try the AGFL if the cntbt is empty, otherwise fail the | |
1037 | * allocation. Make sure to respect minleft even when pulling from the | |
1038 | * freelist. | |
1039 | */ | |
1040 | if (ccur) | |
1041 | error = xfs_btree_decrement(ccur, 0, &i); | |
c63cdd4f BF |
1042 | if (error) |
1043 | goto error; | |
1044 | if (i) { | |
1045 | error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i); | |
1046 | if (error) | |
1047 | goto error; | |
f9e03706 DW |
1048 | if (XFS_IS_CORRUPT(args->mp, i != 1)) { |
1049 | error = -EFSCORRUPTED; | |
1050 | goto error; | |
1051 | } | |
c63cdd4f BF |
1052 | goto out; |
1053 | } | |
1054 | ||
1055 | if (args->minlen != 1 || args->alignment != 1 || | |
1056 | args->resv == XFS_AG_RESV_AGFL || | |
9798f615 | 1057 | be32_to_cpu(agf->agf_flcount) <= args->minleft) |
c63cdd4f BF |
1058 | goto out; |
1059 | ||
1060 | error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0); | |
1061 | if (error) | |
1062 | goto error; | |
1063 | if (fbno == NULLAGBLOCK) | |
1064 | goto out; | |
1065 | ||
45d06621 | 1066 | xfs_extent_busy_reuse(args->mp, args->pag, fbno, 1, |
c34d570d | 1067 | (args->datatype & XFS_ALLOC_NOBUSY)); |
c63cdd4f | 1068 | |
c34d570d | 1069 | if (args->datatype & XFS_ALLOC_USERDATA) { |
c63cdd4f BF |
1070 | struct xfs_buf *bp; |
1071 | ||
ee647f85 DW |
1072 | error = xfs_trans_get_buf(args->tp, args->mp->m_ddev_targp, |
1073 | XFS_AGB_TO_DADDR(args->mp, args->agno, fbno), | |
1074 | args->mp->m_bsize, 0, &bp); | |
1075 | if (error) | |
c63cdd4f | 1076 | goto error; |
c63cdd4f BF |
1077 | xfs_trans_binval(args->tp, bp); |
1078 | } | |
7e36a3a6 BF |
1079 | *fbnop = args->agbno = fbno; |
1080 | *flenp = args->len = 1; | |
9798f615 | 1081 | if (XFS_IS_CORRUPT(args->mp, fbno >= be32_to_cpu(agf->agf_length))) { |
f9e03706 DW |
1082 | error = -EFSCORRUPTED; |
1083 | goto error; | |
1084 | } | |
c63cdd4f BF |
1085 | args->wasfromfl = 1; |
1086 | trace_xfs_alloc_small_freelist(args); | |
1087 | ||
1088 | /* | |
1089 | * If we're feeding an AGFL block to something that doesn't live in the | |
1090 | * free space, we need to clear out the OWN_AG rmap. | |
1091 | */ | |
fa9c3c19 | 1092 | error = xfs_rmap_free(args->tp, args->agbp, args->pag, fbno, 1, |
c63cdd4f BF |
1093 | &XFS_RMAP_OINFO_AG); |
1094 | if (error) | |
1095 | goto error; | |
1096 | ||
1097 | *stat = 0; | |
1098 | return 0; | |
1099 | ||
1100 | out: | |
1101 | /* | |
1102 | * Can't do the allocation, give up. | |
1103 | */ | |
1104 | if (flen < args->minlen) { | |
1105 | args->agbno = NULLAGBLOCK; | |
1106 | trace_xfs_alloc_small_notenough(args); | |
1107 | flen = 0; | |
1108 | } | |
1109 | *fbnop = fbno; | |
1110 | *flenp = flen; | |
1111 | *stat = 1; | |
1112 | trace_xfs_alloc_small_done(args); | |
1113 | return 0; | |
1114 | ||
1115 | error: | |
1116 | trace_xfs_alloc_small_error(args); | |
1117 | return error; | |
1118 | } | |
1119 | ||
1da177e4 LT |
1120 | /* |
1121 | * Allocate a variable extent in the allocation group agno. | |
1122 | * Type and bno are used to determine where in the allocation group the | |
1123 | * extent will start. | |
1124 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
1125 | * and of the form k * prod + mod unless there's nothing that large. | |
1126 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1127 | */ | |
1128 | STATIC int /* error */ | |
1129 | xfs_alloc_ag_vextent( | |
1130 | xfs_alloc_arg_t *args) /* argument structure for allocation */ | |
1131 | { | |
1132 | int error=0; | |
1da177e4 LT |
1133 | |
1134 | ASSERT(args->minlen > 0); | |
1135 | ASSERT(args->maxlen > 0); | |
1136 | ASSERT(args->minlen <= args->maxlen); | |
1137 | ASSERT(args->mod < args->prod); | |
1138 | ASSERT(args->alignment > 0); | |
3fd129b6 | 1139 | |
1da177e4 LT |
1140 | /* |
1141 | * Branch to correct routine based on the type. | |
1142 | */ | |
1143 | args->wasfromfl = 0; | |
1144 | switch (args->type) { | |
1145 | case XFS_ALLOCTYPE_THIS_AG: | |
1146 | error = xfs_alloc_ag_vextent_size(args); | |
1147 | break; | |
1148 | case XFS_ALLOCTYPE_NEAR_BNO: | |
1149 | error = xfs_alloc_ag_vextent_near(args); | |
1150 | break; | |
1151 | case XFS_ALLOCTYPE_THIS_BNO: | |
1152 | error = xfs_alloc_ag_vextent_exact(args); | |
1153 | break; | |
1154 | default: | |
1155 | ASSERT(0); | |
1156 | /* NOTREACHED */ | |
1157 | } | |
ecb6928f CH |
1158 | |
1159 | if (error || args->agbno == NULLAGBLOCK) | |
1da177e4 | 1160 | return error; |
ecb6928f CH |
1161 | |
1162 | ASSERT(args->len >= args->minlen); | |
1163 | ASSERT(args->len <= args->maxlen); | |
0ab32086 | 1164 | ASSERT(!args->wasfromfl || args->resv != XFS_AG_RESV_AGFL); |
ecb6928f CH |
1165 | ASSERT(args->agbno % args->alignment == 0); |
1166 | ||
673930c3 | 1167 | /* if not file data, insert new block into the reverse map btree */ |
33df3a9c | 1168 | if (!xfs_rmap_should_skip_owner_update(&args->oinfo)) { |
fa9c3c19 | 1169 | error = xfs_rmap_alloc(args->tp, args->agbp, args->pag, |
673930c3 DW |
1170 | args->agbno, args->len, &args->oinfo); |
1171 | if (error) | |
1172 | return error; | |
1173 | } | |
1174 | ||
ecb6928f | 1175 | if (!args->wasfromfl) { |
92a00544 | 1176 | error = xfs_alloc_update_counters(args->tp, args->agbp, |
ecb6928f CH |
1177 | -((long)(args->len))); |
1178 | if (error) | |
1179 | return error; | |
1180 | ||
45d06621 | 1181 | ASSERT(!xfs_extent_busy_search(args->mp, args->pag, |
e26f0501 | 1182 | args->agbno, args->len)); |
1da177e4 | 1183 | } |
ecb6928f | 1184 | |
3fd129b6 | 1185 | xfs_ag_resv_alloc_extent(args->pag, args->resv, args); |
ecb6928f | 1186 | |
ff6d6af2 BD |
1187 | XFS_STATS_INC(args->mp, xs_allocx); |
1188 | XFS_STATS_ADD(args->mp, xs_allocb, args->len); | |
ecb6928f | 1189 | return error; |
1da177e4 LT |
1190 | } |
1191 | ||
1192 | /* | |
1193 | * Allocate a variable extent at exactly agno/bno. | |
1194 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
1195 | * and of the form k * prod + mod unless there's nothing that large. | |
1196 | * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it. | |
1197 | */ | |
1198 | STATIC int /* error */ | |
1199 | xfs_alloc_ag_vextent_exact( | |
1200 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
1201 | { | |
9798f615 | 1202 | struct xfs_agf __maybe_unused *agf = args->agbp->b_addr; |
ae127f08 DW |
1203 | struct xfs_btree_cur *bno_cur;/* by block-number btree cursor */ |
1204 | struct xfs_btree_cur *cnt_cur;/* by count btree cursor */ | |
1da177e4 LT |
1205 | int error; |
1206 | xfs_agblock_t fbno; /* start block of found extent */ | |
1da177e4 | 1207 | xfs_extlen_t flen; /* length of found extent */ |
ebf55872 CH |
1208 | xfs_agblock_t tbno; /* start block of busy extent */ |
1209 | xfs_extlen_t tlen; /* length of busy extent */ | |
1210 | xfs_agblock_t tend; /* end block of busy extent */ | |
1da177e4 | 1211 | int i; /* success/failure of operation */ |
ebf55872 | 1212 | unsigned busy_gen; |
1da177e4 LT |
1213 | |
1214 | ASSERT(args->alignment == 1); | |
9f9baab3 | 1215 | |
1da177e4 LT |
1216 | /* |
1217 | * Allocate/initialize a cursor for the by-number freespace btree. | |
1218 | */ | |
561f7d17 | 1219 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
289d38d2 | 1220 | args->pag, XFS_BTNUM_BNO); |
9f9baab3 | 1221 | |
1da177e4 LT |
1222 | /* |
1223 | * Lookup bno and minlen in the btree (minlen is irrelevant, really). | |
1224 | * Look for the closest free block <= bno, it must contain bno | |
1225 | * if any free block does. | |
1226 | */ | |
9f9baab3 CH |
1227 | error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i); |
1228 | if (error) | |
1da177e4 | 1229 | goto error0; |
9f9baab3 CH |
1230 | if (!i) |
1231 | goto not_found; | |
1232 | ||
1da177e4 LT |
1233 | /* |
1234 | * Grab the freespace record. | |
1235 | */ | |
9f9baab3 CH |
1236 | error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i); |
1237 | if (error) | |
1da177e4 | 1238 | goto error0; |
f9e03706 DW |
1239 | if (XFS_IS_CORRUPT(args->mp, i != 1)) { |
1240 | error = -EFSCORRUPTED; | |
1241 | goto error0; | |
1242 | } | |
1da177e4 | 1243 | ASSERT(fbno <= args->agbno); |
9f9baab3 | 1244 | |
1da177e4 | 1245 | /* |
e26f0501 | 1246 | * Check for overlapping busy extents. |
1da177e4 | 1247 | */ |
ebf55872 CH |
1248 | tbno = fbno; |
1249 | tlen = flen; | |
1250 | xfs_extent_busy_trim(args, &tbno, &tlen, &busy_gen); | |
e26f0501 CH |
1251 | |
1252 | /* | |
1253 | * Give up if the start of the extent is busy, or the freespace isn't | |
1254 | * long enough for the minimum request. | |
1255 | */ | |
1256 | if (tbno > args->agbno) | |
1257 | goto not_found; | |
1258 | if (tlen < args->minlen) | |
1259 | goto not_found; | |
1260 | tend = tbno + tlen; | |
1261 | if (tend < args->agbno + args->minlen) | |
9f9baab3 CH |
1262 | goto not_found; |
1263 | ||
1da177e4 LT |
1264 | /* |
1265 | * End of extent will be smaller of the freespace end and the | |
1266 | * maximal requested end. | |
9f9baab3 | 1267 | * |
1da177e4 LT |
1268 | * Fix the length according to mod and prod if given. |
1269 | */ | |
81463b1c CS |
1270 | args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen) |
1271 | - args->agbno; | |
1da177e4 | 1272 | xfs_alloc_fix_len(args); |
81463b1c | 1273 | ASSERT(args->agbno + args->len <= tend); |
9f9baab3 | 1274 | |
1da177e4 | 1275 | /* |
81463b1c | 1276 | * We are allocating agbno for args->len |
1da177e4 LT |
1277 | * Allocate/initialize a cursor for the by-size btree. |
1278 | */ | |
561f7d17 | 1279 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
289d38d2 | 1280 | args->pag, XFS_BTNUM_CNT); |
9798f615 | 1281 | ASSERT(args->agbno + args->len <= be32_to_cpu(agf->agf_length)); |
9f9baab3 CH |
1282 | error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno, |
1283 | args->len, XFSA_FIXUP_BNO_OK); | |
1284 | if (error) { | |
1da177e4 LT |
1285 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); |
1286 | goto error0; | |
1287 | } | |
9f9baab3 | 1288 | |
1da177e4 LT |
1289 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); |
1290 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 1291 | |
1da177e4 | 1292 | args->wasfromfl = 0; |
9f9baab3 CH |
1293 | trace_xfs_alloc_exact_done(args); |
1294 | return 0; | |
1295 | ||
1296 | not_found: | |
1297 | /* Didn't find it, return null. */ | |
1298 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1299 | args->agbno = NULLAGBLOCK; | |
1300 | trace_xfs_alloc_exact_notfound(args); | |
1da177e4 LT |
1301 | return 0; |
1302 | ||
1303 | error0: | |
1304 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
0b1b213f | 1305 | trace_xfs_alloc_exact_error(args); |
1da177e4 LT |
1306 | return error; |
1307 | } | |
1308 | ||
489a150f | 1309 | /* |
78d7aabd BF |
1310 | * Search a given number of btree records in a given direction. Check each |
1311 | * record against the good extent we've already found. | |
489a150f CH |
1312 | */ |
1313 | STATIC int | |
78d7aabd | 1314 | xfs_alloc_walk_iter( |
fec0afda BF |
1315 | struct xfs_alloc_arg *args, |
1316 | struct xfs_alloc_cur *acur, | |
1317 | struct xfs_btree_cur *cur, | |
78d7aabd BF |
1318 | bool increment, |
1319 | bool find_one, /* quit on first candidate */ | |
1320 | int count, /* rec count (-1 for infinite) */ | |
1321 | int *stat) | |
489a150f | 1322 | { |
489a150f CH |
1323 | int error; |
1324 | int i; | |
489a150f | 1325 | |
78d7aabd BF |
1326 | *stat = 0; |
1327 | ||
489a150f | 1328 | /* |
fec0afda BF |
1329 | * Search so long as the cursor is active or we find a better extent. |
1330 | * The cursor is deactivated if it extends beyond the range of the | |
1331 | * current allocation candidate. | |
489a150f | 1332 | */ |
78d7aabd | 1333 | while (xfs_alloc_cur_active(cur) && count) { |
fec0afda | 1334 | error = xfs_alloc_cur_check(args, acur, cur, &i); |
489a150f | 1335 | if (error) |
fec0afda | 1336 | return error; |
78d7aabd BF |
1337 | if (i == 1) { |
1338 | *stat = 1; | |
1339 | if (find_one) | |
1340 | break; | |
1341 | } | |
fec0afda BF |
1342 | if (!xfs_alloc_cur_active(cur)) |
1343 | break; | |
489a150f | 1344 | |
fec0afda BF |
1345 | if (increment) |
1346 | error = xfs_btree_increment(cur, 0, &i); | |
489a150f | 1347 | else |
fec0afda | 1348 | error = xfs_btree_decrement(cur, 0, &i); |
489a150f | 1349 | if (error) |
fec0afda BF |
1350 | return error; |
1351 | if (i == 0) | |
c4aa10d0 | 1352 | cur->bc_ag.abt.active = false; |
78d7aabd BF |
1353 | |
1354 | if (count > 0) | |
1355 | count--; | |
fec0afda | 1356 | } |
489a150f | 1357 | |
489a150f | 1358 | return 0; |
489a150f CH |
1359 | } |
1360 | ||
0e26d5ca | 1361 | /* |
dc8e69bd BF |
1362 | * Search the by-bno and by-size btrees in parallel in search of an extent with |
1363 | * ideal locality based on the NEAR mode ->agbno locality hint. | |
0e26d5ca BF |
1364 | */ |
1365 | STATIC int | |
dc8e69bd | 1366 | xfs_alloc_ag_vextent_locality( |
0e26d5ca BF |
1367 | struct xfs_alloc_arg *args, |
1368 | struct xfs_alloc_cur *acur, | |
1369 | int *stat) | |
1370 | { | |
1371 | struct xfs_btree_cur *fbcur = NULL; | |
1372 | int error; | |
1373 | int i; | |
1374 | bool fbinc; | |
1375 | ||
1376 | ASSERT(acur->len == 0); | |
1377 | ASSERT(args->type == XFS_ALLOCTYPE_NEAR_BNO); | |
1378 | ||
1379 | *stat = 0; | |
1380 | ||
dc8e69bd BF |
1381 | error = xfs_alloc_lookup_ge(acur->cnt, args->agbno, acur->cur_len, &i); |
1382 | if (error) | |
1383 | return error; | |
0e26d5ca BF |
1384 | error = xfs_alloc_lookup_le(acur->bnolt, args->agbno, 0, &i); |
1385 | if (error) | |
1386 | return error; | |
1387 | error = xfs_alloc_lookup_ge(acur->bnogt, args->agbno, 0, &i); | |
1388 | if (error) | |
1389 | return error; | |
1390 | ||
1391 | /* | |
dc8e69bd BF |
1392 | * Search the bnobt and cntbt in parallel. Search the bnobt left and |
1393 | * right and lookup the closest extent to the locality hint for each | |
1394 | * extent size key in the cntbt. The entire search terminates | |
1395 | * immediately on a bnobt hit because that means we've found best case | |
1396 | * locality. Otherwise the search continues until the cntbt cursor runs | |
1397 | * off the end of the tree. If no allocation candidate is found at this | |
1398 | * point, give up on locality, walk backwards from the end of the cntbt | |
1399 | * and take the first available extent. | |
1400 | * | |
1401 | * The parallel tree searches balance each other out to provide fairly | |
1402 | * consistent performance for various situations. The bnobt search can | |
1403 | * have pathological behavior in the worst case scenario of larger | |
1404 | * allocation requests and fragmented free space. On the other hand, the | |
1405 | * bnobt is able to satisfy most smaller allocation requests much more | |
1406 | * quickly than the cntbt. The cntbt search can sift through fragmented | |
1407 | * free space and sets of free extents for larger allocation requests | |
1408 | * more quickly than the bnobt. Since the locality hint is just a hint | |
1409 | * and we don't want to scan the entire bnobt for perfect locality, the | |
1410 | * cntbt search essentially bounds the bnobt search such that we can | |
1411 | * find good enough locality at reasonable performance in most cases. | |
0e26d5ca BF |
1412 | */ |
1413 | while (xfs_alloc_cur_active(acur->bnolt) || | |
dc8e69bd BF |
1414 | xfs_alloc_cur_active(acur->bnogt) || |
1415 | xfs_alloc_cur_active(acur->cnt)) { | |
1416 | ||
1417 | trace_xfs_alloc_cur_lookup(args); | |
1418 | ||
1419 | /* | |
1420 | * Search the bnobt left and right. In the case of a hit, finish | |
1421 | * the search in the opposite direction and we're done. | |
1422 | */ | |
0e26d5ca BF |
1423 | error = xfs_alloc_walk_iter(args, acur, acur->bnolt, false, |
1424 | true, 1, &i); | |
1425 | if (error) | |
1426 | return error; | |
1427 | if (i == 1) { | |
1428 | trace_xfs_alloc_cur_left(args); | |
1429 | fbcur = acur->bnogt; | |
1430 | fbinc = true; | |
1431 | break; | |
1432 | } | |
0e26d5ca BF |
1433 | error = xfs_alloc_walk_iter(args, acur, acur->bnogt, true, true, |
1434 | 1, &i); | |
1435 | if (error) | |
1436 | return error; | |
1437 | if (i == 1) { | |
1438 | trace_xfs_alloc_cur_right(args); | |
1439 | fbcur = acur->bnolt; | |
1440 | fbinc = false; | |
1441 | break; | |
1442 | } | |
dc8e69bd BF |
1443 | |
1444 | /* | |
1445 | * Check the extent with best locality based on the current | |
1446 | * extent size search key and keep track of the best candidate. | |
1447 | */ | |
1448 | error = xfs_alloc_cntbt_iter(args, acur); | |
1449 | if (error) | |
1450 | return error; | |
1451 | if (!xfs_alloc_cur_active(acur->cnt)) { | |
1452 | trace_xfs_alloc_cur_lookup_done(args); | |
1453 | break; | |
1454 | } | |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * If we failed to find anything due to busy extents, return empty | |
1459 | * handed so the caller can flush and retry. If no busy extents were | |
1460 | * found, walk backwards from the end of the cntbt as a last resort. | |
1461 | */ | |
1462 | if (!xfs_alloc_cur_active(acur->cnt) && !acur->len && !acur->busy) { | |
1463 | error = xfs_btree_decrement(acur->cnt, 0, &i); | |
1464 | if (error) | |
1465 | return error; | |
1466 | if (i) { | |
c4aa10d0 | 1467 | acur->cnt->bc_ag.abt.active = true; |
dc8e69bd BF |
1468 | fbcur = acur->cnt; |
1469 | fbinc = false; | |
1470 | } | |
0e26d5ca BF |
1471 | } |
1472 | ||
dc8e69bd BF |
1473 | /* |
1474 | * Search in the opposite direction for a better entry in the case of | |
1475 | * a bnobt hit or walk backwards from the end of the cntbt. | |
1476 | */ | |
0e26d5ca BF |
1477 | if (fbcur) { |
1478 | error = xfs_alloc_walk_iter(args, acur, fbcur, fbinc, true, -1, | |
1479 | &i); | |
1480 | if (error) | |
1481 | return error; | |
1482 | } | |
1483 | ||
1484 | if (acur->len) | |
1485 | *stat = 1; | |
1486 | ||
1487 | return 0; | |
1488 | } | |
1489 | ||
5113f8ec DW |
1490 | /* Check the last block of the cnt btree for allocations. */ |
1491 | static int | |
1492 | xfs_alloc_ag_vextent_lastblock( | |
1493 | struct xfs_alloc_arg *args, | |
1494 | struct xfs_alloc_cur *acur, | |
1495 | xfs_agblock_t *bno, | |
1496 | xfs_extlen_t *len, | |
1497 | bool *allocated) | |
1498 | { | |
1499 | int error; | |
1500 | int i; | |
1501 | ||
1502 | #ifdef DEBUG | |
1503 | /* Randomly don't execute the first algorithm. */ | |
1504 | if (prandom_u32() & 1) | |
1505 | return 0; | |
1506 | #endif | |
1507 | ||
1508 | /* | |
1509 | * Start from the entry that lookup found, sequence through all larger | |
1510 | * free blocks. If we're actually pointing at a record smaller than | |
1511 | * maxlen, go to the start of this block, and skip all those smaller | |
1512 | * than minlen. | |
1513 | */ | |
77ca1eed | 1514 | if (*len || args->alignment > 1) { |
6ca444cf | 1515 | acur->cnt->bc_levels[0].ptr = 1; |
5113f8ec DW |
1516 | do { |
1517 | error = xfs_alloc_get_rec(acur->cnt, bno, len, &i); | |
1518 | if (error) | |
1519 | return error; | |
f9e03706 DW |
1520 | if (XFS_IS_CORRUPT(args->mp, i != 1)) |
1521 | return -EFSCORRUPTED; | |
5113f8ec DW |
1522 | if (*len >= args->minlen) |
1523 | break; | |
1524 | error = xfs_btree_increment(acur->cnt, 0, &i); | |
1525 | if (error) | |
1526 | return error; | |
1527 | } while (i); | |
1528 | ASSERT(*len >= args->minlen); | |
1529 | if (!i) | |
1530 | return 0; | |
1531 | } | |
1532 | ||
1533 | error = xfs_alloc_walk_iter(args, acur, acur->cnt, true, false, -1, &i); | |
1534 | if (error) | |
1535 | return error; | |
1536 | ||
1537 | /* | |
1538 | * It didn't work. We COULD be in a case where there's a good record | |
1539 | * somewhere, so try again. | |
1540 | */ | |
1541 | if (acur->len == 0) | |
1542 | return 0; | |
1543 | ||
1544 | trace_xfs_alloc_near_first(args); | |
1545 | *allocated = true; | |
1546 | return 0; | |
1547 | } | |
1548 | ||
1da177e4 LT |
1549 | /* |
1550 | * Allocate a variable extent near bno in the allocation group agno. | |
1551 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1552 | * and of the form k * prod + mod unless there's nothing that large. | |
1553 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1554 | */ | |
f5e7dbea | 1555 | STATIC int |
1da177e4 | 1556 | xfs_alloc_ag_vextent_near( |
f5e7dbea | 1557 | struct xfs_alloc_arg *args) |
1da177e4 | 1558 | { |
f5e7dbea | 1559 | struct xfs_alloc_cur acur = {}; |
fec0afda BF |
1560 | int error; /* error code */ |
1561 | int i; /* result code, temporary */ | |
fec0afda BF |
1562 | xfs_agblock_t bno; |
1563 | xfs_extlen_t len; | |
e26f0501 | 1564 | |
cf085a1b | 1565 | /* handle uninitialized agbno range so caller doesn't have to */ |
bfe46d4e BF |
1566 | if (!args->min_agbno && !args->max_agbno) |
1567 | args->max_agbno = args->mp->m_sb.sb_agblocks - 1; | |
1568 | ASSERT(args->min_agbno <= args->max_agbno); | |
1569 | ||
1570 | /* clamp agbno to the range if it's outside */ | |
1571 | if (args->agbno < args->min_agbno) | |
1572 | args->agbno = args->min_agbno; | |
1573 | if (args->agbno > args->max_agbno) | |
1574 | args->agbno = args->max_agbno; | |
1575 | ||
e26f0501 | 1576 | restart: |
fec0afda | 1577 | len = 0; |
e26f0501 | 1578 | |
1da177e4 | 1579 | /* |
f5e7dbea BF |
1580 | * Set up cursors and see if there are any free extents as big as |
1581 | * maxlen. If not, pick the last entry in the tree unless the tree is | |
1582 | * empty. | |
1da177e4 | 1583 | */ |
f5e7dbea BF |
1584 | error = xfs_alloc_cur_setup(args, &acur); |
1585 | if (error == -ENOSPC) { | |
fec0afda BF |
1586 | error = xfs_alloc_ag_vextent_small(args, acur.cnt, &bno, |
1587 | &len, &i); | |
f5e7dbea BF |
1588 | if (error) |
1589 | goto out; | |
fec0afda | 1590 | if (i == 0 || len == 0) { |
e26f0501 | 1591 | trace_xfs_alloc_near_noentry(args); |
f5e7dbea | 1592 | goto out; |
1da177e4 LT |
1593 | } |
1594 | ASSERT(i == 1); | |
f5e7dbea BF |
1595 | } else if (error) { |
1596 | goto out; | |
1da177e4 | 1597 | } |
e26f0501 | 1598 | |
1da177e4 LT |
1599 | /* |
1600 | * First algorithm. | |
1601 | * If the requested extent is large wrt the freespaces available | |
1602 | * in this a.g., then the cursor will be pointing to a btree entry | |
1603 | * near the right edge of the tree. If it's in the last btree leaf | |
1604 | * block, then we just examine all the entries in that block | |
1605 | * that are big enough, and pick the best one. | |
1da177e4 | 1606 | */ |
5113f8ec DW |
1607 | if (xfs_btree_islastblock(acur.cnt, 0)) { |
1608 | bool allocated = false; | |
78d7aabd | 1609 | |
5113f8ec DW |
1610 | error = xfs_alloc_ag_vextent_lastblock(args, &acur, &bno, &len, |
1611 | &allocated); | |
78d7aabd BF |
1612 | if (error) |
1613 | goto out; | |
5113f8ec DW |
1614 | if (allocated) |
1615 | goto alloc_finish; | |
1da177e4 | 1616 | } |
f5e7dbea | 1617 | |
1da177e4 | 1618 | /* |
dc8e69bd BF |
1619 | * Second algorithm. Combined cntbt and bnobt search to find ideal |
1620 | * locality. | |
1da177e4 | 1621 | */ |
dc8e69bd | 1622 | error = xfs_alloc_ag_vextent_locality(args, &acur, &i); |
f5e7dbea BF |
1623 | if (error) |
1624 | goto out; | |
1625 | ||
1da177e4 LT |
1626 | /* |
1627 | * If we couldn't get anything, give up. | |
1628 | */ | |
fec0afda | 1629 | if (!acur.len) { |
d6d3aff2 | 1630 | if (acur.busy) { |
e26f0501 | 1631 | trace_xfs_alloc_near_busy(args); |
d6d3aff2 BF |
1632 | xfs_extent_busy_flush(args->mp, args->pag, |
1633 | acur.busy_gen); | |
e26f0501 CH |
1634 | goto restart; |
1635 | } | |
0b1b213f | 1636 | trace_xfs_alloc_size_neither(args); |
1da177e4 | 1637 | args->agbno = NULLAGBLOCK; |
f5e7dbea | 1638 | goto out; |
1da177e4 | 1639 | } |
489a150f | 1640 | |
5113f8ec | 1641 | alloc_finish: |
d2968825 BF |
1642 | /* fix up btrees on a successful allocation */ |
1643 | error = xfs_alloc_cur_finish(args, &acur); | |
0b1b213f | 1644 | |
f5e7dbea BF |
1645 | out: |
1646 | xfs_alloc_cur_close(&acur, error); | |
1da177e4 LT |
1647 | return error; |
1648 | } | |
1649 | ||
1650 | /* | |
1651 | * Allocate a variable extent anywhere in the allocation group agno. | |
1652 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1653 | * and of the form k * prod + mod unless there's nothing that large. | |
1654 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1655 | */ | |
1656 | STATIC int /* error */ | |
1657 | xfs_alloc_ag_vextent_size( | |
1658 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
1659 | { | |
9798f615 | 1660 | struct xfs_agf *agf = args->agbp->b_addr; |
ae127f08 DW |
1661 | struct xfs_btree_cur *bno_cur; /* cursor for bno btree */ |
1662 | struct xfs_btree_cur *cnt_cur; /* cursor for cnt btree */ | |
1da177e4 LT |
1663 | int error; /* error result */ |
1664 | xfs_agblock_t fbno; /* start of found freespace */ | |
1665 | xfs_extlen_t flen; /* length of found freespace */ | |
1da177e4 LT |
1666 | int i; /* temp status variable */ |
1667 | xfs_agblock_t rbno; /* returned block number */ | |
1668 | xfs_extlen_t rlen; /* length of returned extent */ | |
ebf55872 CH |
1669 | bool busy; |
1670 | unsigned busy_gen; | |
1da177e4 | 1671 | |
e26f0501 | 1672 | restart: |
1da177e4 LT |
1673 | /* |
1674 | * Allocate and initialize a cursor for the by-size btree. | |
1675 | */ | |
561f7d17 | 1676 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
289d38d2 | 1677 | args->pag, XFS_BTNUM_CNT); |
1da177e4 | 1678 | bno_cur = NULL; |
e26f0501 | 1679 | |
1da177e4 LT |
1680 | /* |
1681 | * Look for an entry >= maxlen+alignment-1 blocks. | |
1682 | */ | |
1683 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, | |
1684 | args->maxlen + args->alignment - 1, &i))) | |
1685 | goto error0; | |
e26f0501 | 1686 | |
1da177e4 | 1687 | /* |
ebf55872 CH |
1688 | * If none then we have to settle for a smaller extent. In the case that |
1689 | * there are no large extents, this will return the last entry in the | |
1690 | * tree unless the tree is empty. In the case that there are only busy | |
1691 | * large extents, this will return the largest small extent unless there | |
e26f0501 | 1692 | * are no smaller extents available. |
1da177e4 | 1693 | */ |
ebf55872 | 1694 | if (!i) { |
e26f0501 CH |
1695 | error = xfs_alloc_ag_vextent_small(args, cnt_cur, |
1696 | &fbno, &flen, &i); | |
1697 | if (error) | |
1da177e4 LT |
1698 | goto error0; |
1699 | if (i == 0 || flen == 0) { | |
1700 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 1701 | trace_xfs_alloc_size_noentry(args); |
1da177e4 LT |
1702 | return 0; |
1703 | } | |
1704 | ASSERT(i == 1); | |
ebf55872 CH |
1705 | busy = xfs_alloc_compute_aligned(args, fbno, flen, &rbno, |
1706 | &rlen, &busy_gen); | |
e26f0501 CH |
1707 | } else { |
1708 | /* | |
1709 | * Search for a non-busy extent that is large enough. | |
e26f0501 CH |
1710 | */ |
1711 | for (;;) { | |
1712 | error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i); | |
1713 | if (error) | |
1714 | goto error0; | |
f9e03706 DW |
1715 | if (XFS_IS_CORRUPT(args->mp, i != 1)) { |
1716 | error = -EFSCORRUPTED; | |
1717 | goto error0; | |
1718 | } | |
e26f0501 | 1719 | |
ebf55872 CH |
1720 | busy = xfs_alloc_compute_aligned(args, fbno, flen, |
1721 | &rbno, &rlen, &busy_gen); | |
e26f0501 CH |
1722 | |
1723 | if (rlen >= args->maxlen) | |
1724 | break; | |
1725 | ||
1726 | error = xfs_btree_increment(cnt_cur, 0, &i); | |
1727 | if (error) | |
1728 | goto error0; | |
1729 | if (i == 0) { | |
1730 | /* | |
1731 | * Our only valid extents must have been busy. | |
1732 | * Make it unbusy by forcing the log out and | |
ebf55872 | 1733 | * retrying. |
e26f0501 CH |
1734 | */ |
1735 | xfs_btree_del_cursor(cnt_cur, | |
1736 | XFS_BTREE_NOERROR); | |
1737 | trace_xfs_alloc_size_busy(args); | |
ebf55872 CH |
1738 | xfs_extent_busy_flush(args->mp, |
1739 | args->pag, busy_gen); | |
e26f0501 CH |
1740 | goto restart; |
1741 | } | |
1742 | } | |
1da177e4 | 1743 | } |
e26f0501 | 1744 | |
1da177e4 LT |
1745 | /* |
1746 | * In the first case above, we got the last entry in the | |
1747 | * by-size btree. Now we check to see if the space hits maxlen | |
1748 | * once aligned; if not, we search left for something better. | |
1749 | * This can't happen in the second case above. | |
1750 | */ | |
1da177e4 | 1751 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
f9e03706 DW |
1752 | if (XFS_IS_CORRUPT(args->mp, |
1753 | rlen != 0 && | |
1754 | (rlen > flen || | |
1755 | rbno + rlen > fbno + flen))) { | |
1756 | error = -EFSCORRUPTED; | |
1757 | goto error0; | |
1758 | } | |
1da177e4 LT |
1759 | if (rlen < args->maxlen) { |
1760 | xfs_agblock_t bestfbno; | |
1761 | xfs_extlen_t bestflen; | |
1762 | xfs_agblock_t bestrbno; | |
1763 | xfs_extlen_t bestrlen; | |
1764 | ||
1765 | bestrlen = rlen; | |
1766 | bestrbno = rbno; | |
1767 | bestflen = flen; | |
1768 | bestfbno = fbno; | |
1769 | for (;;) { | |
8df4da4a | 1770 | if ((error = xfs_btree_decrement(cnt_cur, 0, &i))) |
1da177e4 LT |
1771 | goto error0; |
1772 | if (i == 0) | |
1773 | break; | |
1774 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, | |
1775 | &i))) | |
1776 | goto error0; | |
f9e03706 DW |
1777 | if (XFS_IS_CORRUPT(args->mp, i != 1)) { |
1778 | error = -EFSCORRUPTED; | |
1779 | goto error0; | |
1780 | } | |
1da177e4 LT |
1781 | if (flen < bestrlen) |
1782 | break; | |
ebf55872 CH |
1783 | busy = xfs_alloc_compute_aligned(args, fbno, flen, |
1784 | &rbno, &rlen, &busy_gen); | |
1da177e4 | 1785 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
f9e03706 DW |
1786 | if (XFS_IS_CORRUPT(args->mp, |
1787 | rlen != 0 && | |
1788 | (rlen > flen || | |
1789 | rbno + rlen > fbno + flen))) { | |
1790 | error = -EFSCORRUPTED; | |
1791 | goto error0; | |
1792 | } | |
1da177e4 LT |
1793 | if (rlen > bestrlen) { |
1794 | bestrlen = rlen; | |
1795 | bestrbno = rbno; | |
1796 | bestflen = flen; | |
1797 | bestfbno = fbno; | |
1798 | if (rlen == args->maxlen) | |
1799 | break; | |
1800 | } | |
1801 | } | |
1802 | if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen, | |
1803 | &i))) | |
1804 | goto error0; | |
f9e03706 DW |
1805 | if (XFS_IS_CORRUPT(args->mp, i != 1)) { |
1806 | error = -EFSCORRUPTED; | |
1807 | goto error0; | |
1808 | } | |
1da177e4 LT |
1809 | rlen = bestrlen; |
1810 | rbno = bestrbno; | |
1811 | flen = bestflen; | |
1812 | fbno = bestfbno; | |
1813 | } | |
1814 | args->wasfromfl = 0; | |
1815 | /* | |
1816 | * Fix up the length. | |
1817 | */ | |
1818 | args->len = rlen; | |
e26f0501 | 1819 | if (rlen < args->minlen) { |
ebf55872 | 1820 | if (busy) { |
e26f0501 CH |
1821 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1822 | trace_xfs_alloc_size_busy(args); | |
ebf55872 | 1823 | xfs_extent_busy_flush(args->mp, args->pag, busy_gen); |
e26f0501 CH |
1824 | goto restart; |
1825 | } | |
1826 | goto out_nominleft; | |
1da177e4 | 1827 | } |
e26f0501 CH |
1828 | xfs_alloc_fix_len(args); |
1829 | ||
1da177e4 | 1830 | rlen = args->len; |
f9e03706 DW |
1831 | if (XFS_IS_CORRUPT(args->mp, rlen > flen)) { |
1832 | error = -EFSCORRUPTED; | |
1833 | goto error0; | |
1834 | } | |
1da177e4 LT |
1835 | /* |
1836 | * Allocate and initialize a cursor for the by-block tree. | |
1837 | */ | |
561f7d17 | 1838 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
289d38d2 | 1839 | args->pag, XFS_BTNUM_BNO); |
1da177e4 LT |
1840 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
1841 | rbno, rlen, XFSA_FIXUP_CNT_OK))) | |
1842 | goto error0; | |
1843 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1844 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1845 | cnt_cur = bno_cur = NULL; | |
1846 | args->len = rlen; | |
1847 | args->agbno = rbno; | |
f9e03706 DW |
1848 | if (XFS_IS_CORRUPT(args->mp, |
1849 | args->agbno + args->len > | |
9798f615 | 1850 | be32_to_cpu(agf->agf_length))) { |
f9e03706 DW |
1851 | error = -EFSCORRUPTED; |
1852 | goto error0; | |
1853 | } | |
0b1b213f | 1854 | trace_xfs_alloc_size_done(args); |
1da177e4 LT |
1855 | return 0; |
1856 | ||
1857 | error0: | |
0b1b213f | 1858 | trace_xfs_alloc_size_error(args); |
1da177e4 LT |
1859 | if (cnt_cur) |
1860 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1861 | if (bno_cur) | |
1862 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1863 | return error; | |
e26f0501 CH |
1864 | |
1865 | out_nominleft: | |
1866 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1867 | trace_xfs_alloc_size_nominleft(args); | |
1868 | args->agbno = NULLAGBLOCK; | |
1869 | return 0; | |
1da177e4 LT |
1870 | } |
1871 | ||
1da177e4 LT |
1872 | /* |
1873 | * Free the extent starting at agno/bno for length. | |
1874 | */ | |
340785cc | 1875 | STATIC int |
1da177e4 | 1876 | xfs_free_ag_extent( |
66e3237e DW |
1877 | struct xfs_trans *tp, |
1878 | struct xfs_buf *agbp, | |
1879 | xfs_agnumber_t agno, | |
1880 | xfs_agblock_t bno, | |
1881 | xfs_extlen_t len, | |
1882 | const struct xfs_owner_info *oinfo, | |
1883 | enum xfs_ag_resv_type type) | |
1da177e4 | 1884 | { |
66e3237e | 1885 | struct xfs_mount *mp; |
66e3237e DW |
1886 | struct xfs_btree_cur *bno_cur; |
1887 | struct xfs_btree_cur *cnt_cur; | |
1888 | xfs_agblock_t gtbno; /* start of right neighbor */ | |
1889 | xfs_extlen_t gtlen; /* length of right neighbor */ | |
1890 | xfs_agblock_t ltbno; /* start of left neighbor */ | |
1891 | xfs_extlen_t ltlen; /* length of left neighbor */ | |
1892 | xfs_agblock_t nbno; /* new starting block of freesp */ | |
1893 | xfs_extlen_t nlen; /* new length of freespace */ | |
1894 | int haveleft; /* have a left neighbor */ | |
1895 | int haveright; /* have a right neighbor */ | |
1896 | int i; | |
1897 | int error; | |
fa9c3c19 | 1898 | struct xfs_perag *pag = agbp->b_pag; |
1da177e4 | 1899 | |
673930c3 | 1900 | bno_cur = cnt_cur = NULL; |
1da177e4 | 1901 | mp = tp->t_mountp; |
673930c3 | 1902 | |
33df3a9c | 1903 | if (!xfs_rmap_should_skip_owner_update(oinfo)) { |
fa9c3c19 | 1904 | error = xfs_rmap_free(tp, agbp, pag, bno, len, oinfo); |
673930c3 DW |
1905 | if (error) |
1906 | goto error0; | |
1907 | } | |
1908 | ||
1da177e4 LT |
1909 | /* |
1910 | * Allocate and initialize a cursor for the by-block btree. | |
1911 | */ | |
289d38d2 | 1912 | bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_BNO); |
1da177e4 LT |
1913 | /* |
1914 | * Look for a neighboring block on the left (lower block numbers) | |
1915 | * that is contiguous with this space. | |
1916 | */ | |
1917 | if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft))) | |
1918 | goto error0; | |
1919 | if (haveleft) { | |
1920 | /* | |
1921 | * There is a block to our left. | |
1922 | */ | |
1923 | if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i))) | |
1924 | goto error0; | |
f9e03706 DW |
1925 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1926 | error = -EFSCORRUPTED; | |
1927 | goto error0; | |
1928 | } | |
1da177e4 LT |
1929 | /* |
1930 | * It's not contiguous, though. | |
1931 | */ | |
1932 | if (ltbno + ltlen < bno) | |
1933 | haveleft = 0; | |
1934 | else { | |
1935 | /* | |
1936 | * If this failure happens the request to free this | |
1937 | * space was invalid, it's (partly) already free. | |
1938 | * Very bad. | |
1939 | */ | |
f9e03706 DW |
1940 | if (XFS_IS_CORRUPT(mp, ltbno + ltlen > bno)) { |
1941 | error = -EFSCORRUPTED; | |
1942 | goto error0; | |
1943 | } | |
1da177e4 LT |
1944 | } |
1945 | } | |
1946 | /* | |
1947 | * Look for a neighboring block on the right (higher block numbers) | |
1948 | * that is contiguous with this space. | |
1949 | */ | |
637aa50f | 1950 | if ((error = xfs_btree_increment(bno_cur, 0, &haveright))) |
1da177e4 LT |
1951 | goto error0; |
1952 | if (haveright) { | |
1953 | /* | |
1954 | * There is a block to our right. | |
1955 | */ | |
1956 | if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i))) | |
1957 | goto error0; | |
f9e03706 DW |
1958 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1959 | error = -EFSCORRUPTED; | |
1960 | goto error0; | |
1961 | } | |
1da177e4 LT |
1962 | /* |
1963 | * It's not contiguous, though. | |
1964 | */ | |
1965 | if (bno + len < gtbno) | |
1966 | haveright = 0; | |
1967 | else { | |
1968 | /* | |
1969 | * If this failure happens the request to free this | |
1970 | * space was invalid, it's (partly) already free. | |
1971 | * Very bad. | |
1972 | */ | |
f9e03706 DW |
1973 | if (XFS_IS_CORRUPT(mp, bno + len > gtbno)) { |
1974 | error = -EFSCORRUPTED; | |
1975 | goto error0; | |
1976 | } | |
1da177e4 LT |
1977 | } |
1978 | } | |
1979 | /* | |
1980 | * Now allocate and initialize a cursor for the by-size tree. | |
1981 | */ | |
289d38d2 | 1982 | cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, pag, XFS_BTNUM_CNT); |
1da177e4 LT |
1983 | /* |
1984 | * Have both left and right contiguous neighbors. | |
1985 | * Merge all three into a single free block. | |
1986 | */ | |
1987 | if (haveleft && haveright) { | |
1988 | /* | |
1989 | * Delete the old by-size entry on the left. | |
1990 | */ | |
1991 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1992 | goto error0; | |
f9e03706 DW |
1993 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
1994 | error = -EFSCORRUPTED; | |
1995 | goto error0; | |
1996 | } | |
91cca5df | 1997 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1998 | goto error0; |
f9e03706 DW |
1999 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2000 | error = -EFSCORRUPTED; | |
2001 | goto error0; | |
2002 | } | |
1da177e4 LT |
2003 | /* |
2004 | * Delete the old by-size entry on the right. | |
2005 | */ | |
2006 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
2007 | goto error0; | |
f9e03706 DW |
2008 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2009 | error = -EFSCORRUPTED; | |
2010 | goto error0; | |
2011 | } | |
91cca5df | 2012 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 2013 | goto error0; |
f9e03706 DW |
2014 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2015 | error = -EFSCORRUPTED; | |
2016 | goto error0; | |
2017 | } | |
1da177e4 LT |
2018 | /* |
2019 | * Delete the old by-block entry for the right block. | |
2020 | */ | |
91cca5df | 2021 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 | 2022 | goto error0; |
f9e03706 DW |
2023 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2024 | error = -EFSCORRUPTED; | |
2025 | goto error0; | |
2026 | } | |
1da177e4 LT |
2027 | /* |
2028 | * Move the by-block cursor back to the left neighbor. | |
2029 | */ | |
8df4da4a | 2030 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 | 2031 | goto error0; |
f9e03706 DW |
2032 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2033 | error = -EFSCORRUPTED; | |
2034 | goto error0; | |
2035 | } | |
1da177e4 LT |
2036 | #ifdef DEBUG |
2037 | /* | |
2038 | * Check that this is the right record: delete didn't | |
2039 | * mangle the cursor. | |
2040 | */ | |
2041 | { | |
2042 | xfs_agblock_t xxbno; | |
2043 | xfs_extlen_t xxlen; | |
2044 | ||
2045 | if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen, | |
2046 | &i))) | |
2047 | goto error0; | |
f9e03706 DW |
2048 | if (XFS_IS_CORRUPT(mp, |
2049 | i != 1 || | |
2050 | xxbno != ltbno || | |
2051 | xxlen != ltlen)) { | |
2052 | error = -EFSCORRUPTED; | |
2053 | goto error0; | |
2054 | } | |
1da177e4 LT |
2055 | } |
2056 | #endif | |
2057 | /* | |
2058 | * Update remaining by-block entry to the new, joined block. | |
2059 | */ | |
2060 | nbno = ltbno; | |
2061 | nlen = len + ltlen + gtlen; | |
2062 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
2063 | goto error0; | |
2064 | } | |
2065 | /* | |
2066 | * Have only a left contiguous neighbor. | |
2067 | * Merge it together with the new freespace. | |
2068 | */ | |
2069 | else if (haveleft) { | |
2070 | /* | |
2071 | * Delete the old by-size entry on the left. | |
2072 | */ | |
2073 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
2074 | goto error0; | |
f9e03706 DW |
2075 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2076 | error = -EFSCORRUPTED; | |
2077 | goto error0; | |
2078 | } | |
91cca5df | 2079 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 2080 | goto error0; |
f9e03706 DW |
2081 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2082 | error = -EFSCORRUPTED; | |
2083 | goto error0; | |
2084 | } | |
1da177e4 LT |
2085 | /* |
2086 | * Back up the by-block cursor to the left neighbor, and | |
2087 | * update its length. | |
2088 | */ | |
8df4da4a | 2089 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 | 2090 | goto error0; |
f9e03706 DW |
2091 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2092 | error = -EFSCORRUPTED; | |
2093 | goto error0; | |
2094 | } | |
1da177e4 LT |
2095 | nbno = ltbno; |
2096 | nlen = len + ltlen; | |
2097 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
2098 | goto error0; | |
2099 | } | |
2100 | /* | |
2101 | * Have only a right contiguous neighbor. | |
2102 | * Merge it together with the new freespace. | |
2103 | */ | |
2104 | else if (haveright) { | |
2105 | /* | |
2106 | * Delete the old by-size entry on the right. | |
2107 | */ | |
2108 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
2109 | goto error0; | |
f9e03706 DW |
2110 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2111 | error = -EFSCORRUPTED; | |
2112 | goto error0; | |
2113 | } | |
91cca5df | 2114 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 2115 | goto error0; |
f9e03706 DW |
2116 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2117 | error = -EFSCORRUPTED; | |
2118 | goto error0; | |
2119 | } | |
1da177e4 LT |
2120 | /* |
2121 | * Update the starting block and length of the right | |
2122 | * neighbor in the by-block tree. | |
2123 | */ | |
2124 | nbno = bno; | |
2125 | nlen = len + gtlen; | |
2126 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
2127 | goto error0; | |
2128 | } | |
2129 | /* | |
2130 | * No contiguous neighbors. | |
2131 | * Insert the new freespace into the by-block tree. | |
2132 | */ | |
2133 | else { | |
2134 | nbno = bno; | |
2135 | nlen = len; | |
4b22a571 | 2136 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 | 2137 | goto error0; |
f9e03706 DW |
2138 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2139 | error = -EFSCORRUPTED; | |
2140 | goto error0; | |
2141 | } | |
1da177e4 LT |
2142 | } |
2143 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
2144 | bno_cur = NULL; | |
2145 | /* | |
2146 | * In all cases we need to insert the new freespace in the by-size tree. | |
2147 | */ | |
2148 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i))) | |
2149 | goto error0; | |
f9e03706 DW |
2150 | if (XFS_IS_CORRUPT(mp, i != 0)) { |
2151 | error = -EFSCORRUPTED; | |
2152 | goto error0; | |
2153 | } | |
4b22a571 | 2154 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 2155 | goto error0; |
f9e03706 DW |
2156 | if (XFS_IS_CORRUPT(mp, i != 1)) { |
2157 | error = -EFSCORRUPTED; | |
2158 | goto error0; | |
2159 | } | |
1da177e4 LT |
2160 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
2161 | cnt_cur = NULL; | |
ecb6928f | 2162 | |
1da177e4 LT |
2163 | /* |
2164 | * Update the freespace totals in the ag and superblock. | |
2165 | */ | |
92a00544 GX |
2166 | error = xfs_alloc_update_counters(tp, agbp, len); |
2167 | xfs_ag_resv_free_extent(agbp->b_pag, type, tp, len); | |
ecb6928f CH |
2168 | if (error) |
2169 | goto error0; | |
2170 | ||
ff6d6af2 BD |
2171 | XFS_STATS_INC(mp, xs_freex); |
2172 | XFS_STATS_ADD(mp, xs_freeb, len); | |
0b1b213f | 2173 | |
21592863 | 2174 | trace_xfs_free_extent(mp, agno, bno, len, type, haveleft, haveright); |
1da177e4 | 2175 | |
1da177e4 LT |
2176 | return 0; |
2177 | ||
2178 | error0: | |
21592863 | 2179 | trace_xfs_free_extent(mp, agno, bno, len, type, -1, -1); |
1da177e4 LT |
2180 | if (bno_cur) |
2181 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
2182 | if (cnt_cur) | |
2183 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
2184 | return error; | |
2185 | } | |
2186 | ||
2187 | /* | |
2188 | * Visible (exported) allocation/free functions. | |
2189 | * Some of these are used just by xfs_alloc_btree.c and this file. | |
2190 | */ | |
2191 | ||
2192 | /* | |
7cb3efb4 | 2193 | * Compute and fill in value of m_alloc_maxlevels. |
1da177e4 LT |
2194 | */ |
2195 | void | |
2196 | xfs_alloc_compute_maxlevels( | |
2197 | xfs_mount_t *mp) /* file system mount structure */ | |
2198 | { | |
7cb3efb4 | 2199 | mp->m_alloc_maxlevels = xfs_btree_compute_maxlevels(mp->m_alloc_mnr, |
19b54ee6 | 2200 | (mp->m_sb.sb_agblocks + 1) / 2); |
0ed5f735 | 2201 | ASSERT(mp->m_alloc_maxlevels <= xfs_allocbt_maxlevels_ondisk()); |
1da177e4 LT |
2202 | } |
2203 | ||
6cc87645 | 2204 | /* |
3fd129b6 DW |
2205 | * Find the length of the longest extent in an AG. The 'need' parameter |
2206 | * specifies how much space we're going to need for the AGFL and the | |
2207 | * 'reserved' parameter tells us how many blocks in this AG are reserved for | |
2208 | * other callers. | |
6cc87645 DC |
2209 | */ |
2210 | xfs_extlen_t | |
2211 | xfs_alloc_longest_free_extent( | |
50adbcb4 | 2212 | struct xfs_perag *pag, |
3fd129b6 DW |
2213 | xfs_extlen_t need, |
2214 | xfs_extlen_t reserved) | |
6cc87645 | 2215 | { |
50adbcb4 | 2216 | xfs_extlen_t delta = 0; |
6cc87645 | 2217 | |
3fd129b6 DW |
2218 | /* |
2219 | * If the AGFL needs a recharge, we'll have to subtract that from the | |
2220 | * longest extent. | |
2221 | */ | |
6cc87645 DC |
2222 | if (need > pag->pagf_flcount) |
2223 | delta = need - pag->pagf_flcount; | |
2224 | ||
3fd129b6 DW |
2225 | /* |
2226 | * If we cannot maintain others' reservations with space from the | |
2227 | * not-longest freesp extents, we'll have to subtract /that/ from | |
2228 | * the longest extent too. | |
2229 | */ | |
2230 | if (pag->pagf_freeblks - pag->pagf_longest < reserved) | |
2231 | delta += reserved - (pag->pagf_freeblks - pag->pagf_longest); | |
2232 | ||
2233 | /* | |
2234 | * If the longest extent is long enough to satisfy all the | |
2235 | * reservations and AGFL rules in place, we can return this extent. | |
2236 | */ | |
6cc87645 | 2237 | if (pag->pagf_longest > delta) |
1c743574 DC |
2238 | return min_t(xfs_extlen_t, pag->pag_mount->m_ag_max_usable, |
2239 | pag->pagf_longest - delta); | |
3fd129b6 DW |
2240 | |
2241 | /* Otherwise, let the caller try for 1 block if there's space. */ | |
6cc87645 DC |
2242 | return pag->pagf_flcount > 0 || pag->pagf_longest > 0; |
2243 | } | |
2244 | ||
1cac233c DW |
2245 | /* |
2246 | * Compute the minimum length of the AGFL in the given AG. If @pag is NULL, | |
2247 | * return the largest possible minimum length. | |
2248 | */ | |
496817b4 DC |
2249 | unsigned int |
2250 | xfs_alloc_min_freelist( | |
2251 | struct xfs_mount *mp, | |
2252 | struct xfs_perag *pag) | |
2253 | { | |
1cac233c DW |
2254 | /* AG btrees have at least 1 level. */ |
2255 | static const uint8_t fake_levels[XFS_BTNUM_AGF] = {1, 1, 1}; | |
2256 | const uint8_t *levels = pag ? pag->pagf_levels : fake_levels; | |
496817b4 DC |
2257 | unsigned int min_free; |
2258 | ||
7cb3efb4 | 2259 | ASSERT(mp->m_alloc_maxlevels > 0); |
1cac233c | 2260 | |
496817b4 | 2261 | /* space needed by-bno freespace btree */ |
1cac233c | 2262 | min_free = min_t(unsigned int, levels[XFS_BTNUM_BNOi] + 1, |
7cb3efb4 | 2263 | mp->m_alloc_maxlevels); |
496817b4 | 2264 | /* space needed by-size freespace btree */ |
1cac233c | 2265 | min_free += min_t(unsigned int, levels[XFS_BTNUM_CNTi] + 1, |
7cb3efb4 | 2266 | mp->m_alloc_maxlevels); |
52548852 | 2267 | /* space needed reverse mapping used space btree */ |
ebd9027d | 2268 | if (xfs_has_rmapbt(mp)) |
1cac233c DW |
2269 | min_free += min_t(unsigned int, levels[XFS_BTNUM_RMAPi] + 1, |
2270 | mp->m_rmap_maxlevels); | |
496817b4 DC |
2271 | |
2272 | return min_free; | |
2273 | } | |
2274 | ||
72d55285 DC |
2275 | /* |
2276 | * Check if the operation we are fixing up the freelist for should go ahead or | |
2277 | * not. If we are freeing blocks, we always allow it, otherwise the allocation | |
2278 | * is dependent on whether the size and shape of free space available will | |
2279 | * permit the requested allocation to take place. | |
2280 | */ | |
2281 | static bool | |
2282 | xfs_alloc_space_available( | |
2283 | struct xfs_alloc_arg *args, | |
2284 | xfs_extlen_t min_free, | |
2285 | int flags) | |
2286 | { | |
2287 | struct xfs_perag *pag = args->pag; | |
12ef8301 | 2288 | xfs_extlen_t alloc_len, longest; |
3fd129b6 | 2289 | xfs_extlen_t reservation; /* blocks that are still reserved */ |
72d55285 | 2290 | int available; |
1ca89fbc | 2291 | xfs_extlen_t agflcount; |
72d55285 DC |
2292 | |
2293 | if (flags & XFS_ALLOC_FLAG_FREEING) | |
2294 | return true; | |
2295 | ||
3fd129b6 DW |
2296 | reservation = xfs_ag_resv_needed(pag, args->resv); |
2297 | ||
72d55285 | 2298 | /* do we have enough contiguous free space for the allocation? */ |
12ef8301 | 2299 | alloc_len = args->minlen + (args->alignment - 1) + args->minalignslop; |
a1f69417 | 2300 | longest = xfs_alloc_longest_free_extent(pag, min_free, reservation); |
12ef8301 | 2301 | if (longest < alloc_len) |
72d55285 DC |
2302 | return false; |
2303 | ||
1ca89fbc BF |
2304 | /* |
2305 | * Do we have enough free space remaining for the allocation? Don't | |
2306 | * account extra agfl blocks because we are about to defer free them, | |
2307 | * making them unavailable until the current transaction commits. | |
2308 | */ | |
2309 | agflcount = min_t(xfs_extlen_t, pag->pagf_flcount, min_free); | |
2310 | available = (int)(pag->pagf_freeblks + agflcount - | |
54fee133 | 2311 | reservation - min_free - args->minleft); |
12ef8301 | 2312 | if (available < (int)max(args->total, alloc_len)) |
72d55285 DC |
2313 | return false; |
2314 | ||
54fee133 CH |
2315 | /* |
2316 | * Clamp maxlen to the amount of free space available for the actual | |
2317 | * extent allocation. | |
2318 | */ | |
2319 | if (available < (int)args->maxlen && !(flags & XFS_ALLOC_FLAG_CHECK)) { | |
2320 | args->maxlen = available; | |
2321 | ASSERT(args->maxlen > 0); | |
2322 | ASSERT(args->maxlen >= args->minlen); | |
2323 | } | |
2324 | ||
72d55285 DC |
2325 | return true; |
2326 | } | |
2327 | ||
4223f659 BF |
2328 | int |
2329 | xfs_free_agfl_block( | |
2330 | struct xfs_trans *tp, | |
2331 | xfs_agnumber_t agno, | |
2332 | xfs_agblock_t agbno, | |
2333 | struct xfs_buf *agbp, | |
2334 | struct xfs_owner_info *oinfo) | |
2335 | { | |
2336 | int error; | |
2337 | struct xfs_buf *bp; | |
2338 | ||
2339 | error = xfs_free_ag_extent(tp, agbp, agno, agbno, 1, oinfo, | |
2340 | XFS_AG_RESV_AGFL); | |
2341 | if (error) | |
2342 | return error; | |
2343 | ||
ee647f85 DW |
2344 | error = xfs_trans_get_buf(tp, tp->t_mountp->m_ddev_targp, |
2345 | XFS_AGB_TO_DADDR(tp->t_mountp, agno, agbno), | |
2346 | tp->t_mountp->m_bsize, 0, &bp); | |
2347 | if (error) | |
2348 | return error; | |
4223f659 BF |
2349 | xfs_trans_binval(tp, bp); |
2350 | ||
2351 | return 0; | |
2352 | } | |
2353 | ||
a27ba260 BF |
2354 | /* |
2355 | * Check the agfl fields of the agf for inconsistency or corruption. The purpose | |
2356 | * is to detect an agfl header padding mismatch between current and early v5 | |
2357 | * kernels. This problem manifests as a 1-slot size difference between the | |
2358 | * on-disk flcount and the active [first, last] range of a wrapped agfl. This | |
2359 | * may also catch variants of agfl count corruption unrelated to padding. Either | |
2360 | * way, we'll reset the agfl and warn the user. | |
2361 | * | |
2362 | * Return true if a reset is required before the agfl can be used, false | |
2363 | * otherwise. | |
2364 | */ | |
2365 | static bool | |
2366 | xfs_agfl_needs_reset( | |
2367 | struct xfs_mount *mp, | |
2368 | struct xfs_agf *agf) | |
2369 | { | |
2370 | uint32_t f = be32_to_cpu(agf->agf_flfirst); | |
2371 | uint32_t l = be32_to_cpu(agf->agf_fllast); | |
2372 | uint32_t c = be32_to_cpu(agf->agf_flcount); | |
2373 | int agfl_size = xfs_agfl_size(mp); | |
2374 | int active; | |
2375 | ||
2376 | /* no agfl header on v4 supers */ | |
38c26bfd | 2377 | if (!xfs_has_crc(mp)) |
a27ba260 BF |
2378 | return false; |
2379 | ||
2380 | /* | |
2381 | * The agf read verifier catches severe corruption of these fields. | |
2382 | * Repeat some sanity checks to cover a packed -> unpacked mismatch if | |
2383 | * the verifier allows it. | |
2384 | */ | |
2385 | if (f >= agfl_size || l >= agfl_size) | |
2386 | return true; | |
2387 | if (c > agfl_size) | |
2388 | return true; | |
2389 | ||
2390 | /* | |
2391 | * Check consistency between the on-disk count and the active range. An | |
2392 | * agfl padding mismatch manifests as an inconsistent flcount. | |
2393 | */ | |
2394 | if (c && l >= f) | |
2395 | active = l - f + 1; | |
2396 | else if (c) | |
2397 | active = agfl_size - f + l + 1; | |
2398 | else | |
2399 | active = 0; | |
2400 | ||
2401 | return active != c; | |
2402 | } | |
2403 | ||
2404 | /* | |
2405 | * Reset the agfl to an empty state. Ignore/drop any existing blocks since the | |
2406 | * agfl content cannot be trusted. Warn the user that a repair is required to | |
2407 | * recover leaked blocks. | |
2408 | * | |
2409 | * The purpose of this mechanism is to handle filesystems affected by the agfl | |
2410 | * header padding mismatch problem. A reset keeps the filesystem online with a | |
2411 | * relatively minor free space accounting inconsistency rather than suffer the | |
2412 | * inevitable crash from use of an invalid agfl block. | |
2413 | */ | |
2414 | static void | |
2415 | xfs_agfl_reset( | |
2416 | struct xfs_trans *tp, | |
2417 | struct xfs_buf *agbp, | |
2418 | struct xfs_perag *pag) | |
2419 | { | |
2420 | struct xfs_mount *mp = tp->t_mountp; | |
9798f615 | 2421 | struct xfs_agf *agf = agbp->b_addr; |
a27ba260 BF |
2422 | |
2423 | ASSERT(pag->pagf_agflreset); | |
2424 | trace_xfs_agfl_reset(mp, agf, 0, _RET_IP_); | |
2425 | ||
2426 | xfs_warn(mp, | |
2427 | "WARNING: Reset corrupted AGFL on AG %u. %d blocks leaked. " | |
2428 | "Please unmount and run xfs_repair.", | |
2429 | pag->pag_agno, pag->pagf_flcount); | |
2430 | ||
2431 | agf->agf_flfirst = 0; | |
2432 | agf->agf_fllast = cpu_to_be32(xfs_agfl_size(mp) - 1); | |
2433 | agf->agf_flcount = 0; | |
2434 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLFIRST | XFS_AGF_FLLAST | | |
2435 | XFS_AGF_FLCOUNT); | |
2436 | ||
2437 | pag->pagf_flcount = 0; | |
2438 | pag->pagf_agflreset = false; | |
2439 | } | |
2440 | ||
f8f2835a BF |
2441 | /* |
2442 | * Defer an AGFL block free. This is effectively equivalent to | |
c201d9ca | 2443 | * xfs_free_extent_later() with some special handling particular to AGFL blocks. |
f8f2835a BF |
2444 | * |
2445 | * Deferring AGFL frees helps prevent log reservation overruns due to too many | |
2446 | * allocation operations in a transaction. AGFL frees are prone to this problem | |
2447 | * because for one they are always freed one at a time. Further, an immediate | |
2448 | * AGFL block free can cause a btree join and require another block free before | |
2449 | * the real allocation can proceed. Deferring the free disconnects freeing up | |
2450 | * the AGFL slot from freeing the block. | |
2451 | */ | |
2452 | STATIC void | |
2453 | xfs_defer_agfl_block( | |
0f37d178 | 2454 | struct xfs_trans *tp, |
f8f2835a BF |
2455 | xfs_agnumber_t agno, |
2456 | xfs_fsblock_t agbno, | |
2457 | struct xfs_owner_info *oinfo) | |
2458 | { | |
0f37d178 | 2459 | struct xfs_mount *mp = tp->t_mountp; |
f8f2835a BF |
2460 | struct xfs_extent_free_item *new; /* new element */ |
2461 | ||
c201d9ca | 2462 | ASSERT(xfs_extfree_item_cache != NULL); |
f8f2835a BF |
2463 | ASSERT(oinfo != NULL); |
2464 | ||
b3b5ff41 | 2465 | new = kmem_cache_zalloc(xfs_extfree_item_cache, |
3050bd0b | 2466 | GFP_KERNEL | __GFP_NOFAIL); |
f8f2835a BF |
2467 | new->xefi_startblock = XFS_AGB_TO_FSB(mp, agno, agbno); |
2468 | new->xefi_blockcount = 1; | |
b3b5ff41 | 2469 | new->xefi_owner = oinfo->oi_owner; |
f8f2835a BF |
2470 | |
2471 | trace_xfs_agfl_free_defer(mp, agno, 0, agbno, 1); | |
2472 | ||
0f37d178 | 2473 | xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_AGFL_FREE, &new->xefi_list); |
f8f2835a BF |
2474 | } |
2475 | ||
c201d9ca DW |
2476 | /* |
2477 | * Add the extent to the list of extents to be free at transaction end. | |
2478 | * The list is maintained sorted (by block number). | |
2479 | */ | |
2480 | void | |
2481 | __xfs_free_extent_later( | |
2482 | struct xfs_trans *tp, | |
2483 | xfs_fsblock_t bno, | |
2484 | xfs_filblks_t len, | |
2485 | const struct xfs_owner_info *oinfo, | |
2486 | bool skip_discard) | |
2487 | { | |
2488 | struct xfs_extent_free_item *new; /* new element */ | |
2489 | #ifdef DEBUG | |
2490 | struct xfs_mount *mp = tp->t_mountp; | |
2491 | xfs_agnumber_t agno; | |
2492 | xfs_agblock_t agbno; | |
2493 | ||
2494 | ASSERT(bno != NULLFSBLOCK); | |
2495 | ASSERT(len > 0); | |
2496 | ASSERT(len <= MAXEXTLEN); | |
2497 | ASSERT(!isnullstartblock(bno)); | |
2498 | agno = XFS_FSB_TO_AGNO(mp, bno); | |
2499 | agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
2500 | ASSERT(agno < mp->m_sb.sb_agcount); | |
2501 | ASSERT(agbno < mp->m_sb.sb_agblocks); | |
2502 | ASSERT(len < mp->m_sb.sb_agblocks); | |
2503 | ASSERT(agbno + len <= mp->m_sb.sb_agblocks); | |
2504 | #endif | |
2505 | ASSERT(xfs_extfree_item_cache != NULL); | |
2506 | ||
b3b5ff41 | 2507 | new = kmem_cache_zalloc(xfs_extfree_item_cache, |
c201d9ca DW |
2508 | GFP_KERNEL | __GFP_NOFAIL); |
2509 | new->xefi_startblock = bno; | |
2510 | new->xefi_blockcount = (xfs_extlen_t)len; | |
b3b5ff41 DW |
2511 | if (skip_discard) |
2512 | new->xefi_flags |= XFS_EFI_SKIP_DISCARD; | |
2513 | if (oinfo) { | |
2514 | ASSERT(oinfo->oi_offset == 0); | |
2515 | ||
2516 | if (oinfo->oi_flags & XFS_OWNER_INFO_ATTR_FORK) | |
2517 | new->xefi_flags |= XFS_EFI_ATTR_FORK; | |
2518 | if (oinfo->oi_flags & XFS_OWNER_INFO_BMBT_BLOCK) | |
2519 | new->xefi_flags |= XFS_EFI_BMBT_BLOCK; | |
2520 | new->xefi_owner = oinfo->oi_owner; | |
2521 | } else { | |
2522 | new->xefi_owner = XFS_RMAP_OWN_NULL; | |
2523 | } | |
c201d9ca DW |
2524 | trace_xfs_bmap_free_defer(tp->t_mountp, |
2525 | XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0, | |
2526 | XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len); | |
2527 | xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list); | |
2528 | } | |
2529 | ||
30151967 CB |
2530 | #ifdef DEBUG |
2531 | /* | |
2532 | * Check if an AGF has a free extent record whose length is equal to | |
2533 | * args->minlen. | |
2534 | */ | |
2535 | STATIC int | |
2536 | xfs_exact_minlen_extent_available( | |
2537 | struct xfs_alloc_arg *args, | |
2538 | struct xfs_buf *agbp, | |
2539 | int *stat) | |
2540 | { | |
2541 | struct xfs_btree_cur *cnt_cur; | |
2542 | xfs_agblock_t fbno; | |
2543 | xfs_extlen_t flen; | |
2544 | int error = 0; | |
2545 | ||
2546 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, agbp, | |
289d38d2 | 2547 | args->pag, XFS_BTNUM_CNT); |
30151967 CB |
2548 | error = xfs_alloc_lookup_ge(cnt_cur, 0, args->minlen, stat); |
2549 | if (error) | |
2550 | goto out; | |
2551 | ||
2552 | if (*stat == 0) { | |
2553 | error = -EFSCORRUPTED; | |
2554 | goto out; | |
2555 | } | |
2556 | ||
2557 | error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, stat); | |
2558 | if (error) | |
2559 | goto out; | |
2560 | ||
2561 | if (*stat == 1 && flen != args->minlen) | |
2562 | *stat = 0; | |
2563 | ||
2564 | out: | |
2565 | xfs_btree_del_cursor(cnt_cur, error); | |
2566 | ||
2567 | return error; | |
2568 | } | |
2569 | #endif | |
2570 | ||
1da177e4 LT |
2571 | /* |
2572 | * Decide whether to use this allocation group for this allocation. | |
2573 | * If so, fix up the btree freelist's size. | |
2574 | */ | |
2e9101da | 2575 | int /* error */ |
1da177e4 | 2576 | xfs_alloc_fix_freelist( |
396503fc DC |
2577 | struct xfs_alloc_arg *args, /* allocation argument structure */ |
2578 | int flags) /* XFS_ALLOC_FLAG_... */ | |
1da177e4 | 2579 | { |
396503fc DC |
2580 | struct xfs_mount *mp = args->mp; |
2581 | struct xfs_perag *pag = args->pag; | |
2582 | struct xfs_trans *tp = args->tp; | |
2583 | struct xfs_buf *agbp = NULL; | |
2584 | struct xfs_buf *agflbp = NULL; | |
2585 | struct xfs_alloc_arg targs; /* local allocation arguments */ | |
2586 | xfs_agblock_t bno; /* freelist block */ | |
2587 | xfs_extlen_t need; /* total blocks needed in freelist */ | |
c184f855 | 2588 | int error = 0; |
396503fc | 2589 | |
362f5e74 BF |
2590 | /* deferred ops (AGFL block frees) require permanent transactions */ |
2591 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
2592 | ||
1da177e4 | 2593 | if (!pag->pagf_init) { |
396503fc | 2594 | error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp); |
f48e2df8 DW |
2595 | if (error) { |
2596 | /* Couldn't lock the AGF so skip this AG. */ | |
2597 | if (error == -EAGAIN) | |
2598 | error = 0; | |
396503fc | 2599 | goto out_no_agbp; |
1da177e4 | 2600 | } |
396503fc | 2601 | } |
1da177e4 | 2602 | |
0e1edbd9 | 2603 | /* |
396503fc DC |
2604 | * If this is a metadata preferred pag and we are user data then try |
2605 | * somewhere else if we are not being asked to try harder at this | |
2606 | * point | |
1da177e4 | 2607 | */ |
c34d570d | 2608 | if (pag->pagf_metadata && (args->datatype & XFS_ALLOC_USERDATA) && |
0e1edbd9 NS |
2609 | (flags & XFS_ALLOC_FLAG_TRYLOCK)) { |
2610 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
396503fc | 2611 | goto out_agbp_relse; |
1da177e4 LT |
2612 | } |
2613 | ||
496817b4 | 2614 | need = xfs_alloc_min_freelist(mp, pag); |
54fee133 CH |
2615 | if (!xfs_alloc_space_available(args, need, flags | |
2616 | XFS_ALLOC_FLAG_CHECK)) | |
396503fc | 2617 | goto out_agbp_relse; |
0e1edbd9 | 2618 | |
1da177e4 LT |
2619 | /* |
2620 | * Get the a.g. freespace buffer. | |
2621 | * Can fail if we're not blocking on locks, and it's held. | |
2622 | */ | |
396503fc DC |
2623 | if (!agbp) { |
2624 | error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp); | |
f48e2df8 DW |
2625 | if (error) { |
2626 | /* Couldn't lock the AGF so skip this AG. */ | |
2627 | if (error == -EAGAIN) | |
2628 | error = 0; | |
396503fc | 2629 | goto out_no_agbp; |
0e1edbd9 | 2630 | } |
1da177e4 | 2631 | } |
50adbcb4 | 2632 | |
a27ba260 BF |
2633 | /* reset a padding mismatched agfl before final free space check */ |
2634 | if (pag->pagf_agflreset) | |
2635 | xfs_agfl_reset(tp, agbp, pag); | |
2636 | ||
50adbcb4 | 2637 | /* If there isn't enough total space or single-extent, reject it. */ |
496817b4 | 2638 | need = xfs_alloc_min_freelist(mp, pag); |
396503fc DC |
2639 | if (!xfs_alloc_space_available(args, need, flags)) |
2640 | goto out_agbp_relse; | |
72d55285 | 2641 | |
30151967 CB |
2642 | #ifdef DEBUG |
2643 | if (args->alloc_minlen_only) { | |
2644 | int stat; | |
2645 | ||
2646 | error = xfs_exact_minlen_extent_available(args, agbp, &stat); | |
2647 | if (error || !stat) | |
2648 | goto out_agbp_relse; | |
2649 | } | |
2650 | #endif | |
1da177e4 LT |
2651 | /* |
2652 | * Make the freelist shorter if it's too long. | |
50adbcb4 | 2653 | * |
396503fc DC |
2654 | * Note that from this point onwards, we will always release the agf and |
2655 | * agfl buffers on error. This handles the case where we error out and | |
2656 | * the buffers are clean or may not have been joined to the transaction | |
2657 | * and hence need to be released manually. If they have been joined to | |
2658 | * the transaction, then xfs_trans_brelse() will handle them | |
2659 | * appropriately based on the recursion count and dirty state of the | |
2660 | * buffer. | |
2661 | * | |
50adbcb4 DC |
2662 | * XXX (dgc): When we have lots of free space, does this buy us |
2663 | * anything other than extra overhead when we need to put more blocks | |
2664 | * back on the free list? Maybe we should only do this when space is | |
2665 | * getting low or the AGFL is more than half full? | |
04f13060 DW |
2666 | * |
2667 | * The NOSHRINK flag prevents the AGFL from being shrunk if it's too | |
2668 | * big; the NORMAP flag prevents AGFL expand/shrink operations from | |
2669 | * updating the rmapbt. Both flags are used in xfs_repair while we're | |
2670 | * rebuilding the rmapbt, and neither are used by the kernel. They're | |
2671 | * both required to ensure that rmaps are correctly recorded for the | |
2672 | * regenerated AGFL, bnobt, and cntbt. See repair/phase5.c and | |
2673 | * repair/rmap.c in xfsprogs for details. | |
1da177e4 | 2674 | */ |
04f13060 | 2675 | memset(&targs, 0, sizeof(targs)); |
7280feda | 2676 | /* struct copy below */ |
04f13060 | 2677 | if (flags & XFS_ALLOC_FLAG_NORMAP) |
7280feda | 2678 | targs.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE; |
04f13060 | 2679 | else |
7280feda | 2680 | targs.oinfo = XFS_RMAP_OINFO_AG; |
04f13060 | 2681 | while (!(flags & XFS_ALLOC_FLAG_NOSHRINK) && pag->pagf_flcount > need) { |
92821e2b DC |
2682 | error = xfs_alloc_get_freelist(tp, agbp, &bno, 0); |
2683 | if (error) | |
396503fc | 2684 | goto out_agbp_relse; |
4223f659 | 2685 | |
c03edc9e BF |
2686 | /* defer agfl frees */ |
2687 | xfs_defer_agfl_block(tp, args->agno, bno, &targs.oinfo); | |
1da177e4 | 2688 | } |
50adbcb4 | 2689 | |
1da177e4 LT |
2690 | targs.tp = tp; |
2691 | targs.mp = mp; | |
2692 | targs.agbp = agbp; | |
2693 | targs.agno = args->agno; | |
3fd129b6 | 2694 | targs.alignment = targs.minlen = targs.prod = 1; |
1da177e4 LT |
2695 | targs.type = XFS_ALLOCTYPE_THIS_AG; |
2696 | targs.pag = pag; | |
50adbcb4 DC |
2697 | error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp); |
2698 | if (error) | |
396503fc | 2699 | goto out_agbp_relse; |
50adbcb4 DC |
2700 | |
2701 | /* Make the freelist longer if it's too short. */ | |
2702 | while (pag->pagf_flcount < need) { | |
1da177e4 | 2703 | targs.agbno = 0; |
50adbcb4 | 2704 | targs.maxlen = need - pag->pagf_flcount; |
0ab32086 | 2705 | targs.resv = XFS_AG_RESV_AGFL; |
50adbcb4 DC |
2706 | |
2707 | /* Allocate as many blocks as possible at once. */ | |
2708 | error = xfs_alloc_ag_vextent(&targs); | |
396503fc DC |
2709 | if (error) |
2710 | goto out_agflbp_relse; | |
2711 | ||
1da177e4 LT |
2712 | /* |
2713 | * Stop if we run out. Won't happen if callers are obeying | |
2714 | * the restrictions correctly. Can happen for free calls | |
2715 | * on a completely full ag. | |
2716 | */ | |
d210a28c | 2717 | if (targs.agbno == NULLAGBLOCK) { |
0e1edbd9 NS |
2718 | if (flags & XFS_ALLOC_FLAG_FREEING) |
2719 | break; | |
396503fc | 2720 | goto out_agflbp_relse; |
d210a28c | 2721 | } |
1da177e4 LT |
2722 | /* |
2723 | * Put each allocated block on the list. | |
2724 | */ | |
2725 | for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) { | |
92821e2b DC |
2726 | error = xfs_alloc_put_freelist(tp, agbp, |
2727 | agflbp, bno, 0); | |
2728 | if (error) | |
396503fc | 2729 | goto out_agflbp_relse; |
1da177e4 LT |
2730 | } |
2731 | } | |
e63a3690 | 2732 | xfs_trans_brelse(tp, agflbp); |
1da177e4 LT |
2733 | args->agbp = agbp; |
2734 | return 0; | |
396503fc DC |
2735 | |
2736 | out_agflbp_relse: | |
2737 | xfs_trans_brelse(tp, agflbp); | |
2738 | out_agbp_relse: | |
2739 | if (agbp) | |
2740 | xfs_trans_brelse(tp, agbp); | |
2741 | out_no_agbp: | |
2742 | args->agbp = NULL; | |
2743 | return error; | |
1da177e4 LT |
2744 | } |
2745 | ||
2746 | /* | |
2747 | * Get a block from the freelist. | |
2748 | * Returns with the buffer for the block gotten. | |
2749 | */ | |
50920116 | 2750 | int |
1da177e4 | 2751 | xfs_alloc_get_freelist( |
50920116 DC |
2752 | struct xfs_trans *tp, |
2753 | struct xfs_buf *agbp, | |
2754 | xfs_agblock_t *bnop, | |
2755 | int btreeblk) | |
1da177e4 | 2756 | { |
50920116 DC |
2757 | struct xfs_agf *agf = agbp->b_addr; |
2758 | struct xfs_buf *agflbp; | |
2759 | xfs_agblock_t bno; | |
2760 | __be32 *agfl_bno; | |
2761 | int error; | |
2762 | int logflags; | |
2763 | struct xfs_mount *mp = tp->t_mountp; | |
2764 | struct xfs_perag *pag; | |
1da177e4 | 2765 | |
1da177e4 LT |
2766 | /* |
2767 | * Freelist is empty, give up. | |
2768 | */ | |
2769 | if (!agf->agf_flcount) { | |
2770 | *bnop = NULLAGBLOCK; | |
2771 | return 0; | |
2772 | } | |
2773 | /* | |
2774 | * Read the array of free blocks. | |
2775 | */ | |
77c95bba CH |
2776 | error = xfs_alloc_read_agfl(mp, tp, be32_to_cpu(agf->agf_seqno), |
2777 | &agflbp); | |
2778 | if (error) | |
1da177e4 | 2779 | return error; |
77c95bba CH |
2780 | |
2781 | ||
1da177e4 LT |
2782 | /* |
2783 | * Get the block number and update the data structures. | |
2784 | */ | |
183606d8 | 2785 | agfl_bno = xfs_buf_to_agfl_bno(agflbp); |
77c95bba | 2786 | bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]); |
413d57c9 | 2787 | be32_add_cpu(&agf->agf_flfirst, 1); |
1da177e4 | 2788 | xfs_trans_brelse(tp, agflbp); |
a78ee256 | 2789 | if (be32_to_cpu(agf->agf_flfirst) == xfs_agfl_size(mp)) |
1da177e4 | 2790 | agf->agf_flfirst = 0; |
a862e0fd | 2791 | |
92a00544 | 2792 | pag = agbp->b_pag; |
a27ba260 | 2793 | ASSERT(!pag->pagf_agflreset); |
413d57c9 | 2794 | be32_add_cpu(&agf->agf_flcount, -1); |
1da177e4 | 2795 | pag->pagf_flcount--; |
92821e2b DC |
2796 | |
2797 | logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; | |
2798 | if (btreeblk) { | |
413d57c9 | 2799 | be32_add_cpu(&agf->agf_btreeblks, 1); |
92821e2b DC |
2800 | pag->pagf_btreeblks++; |
2801 | logflags |= XFS_AGF_BTREEBLKS; | |
2802 | } | |
2803 | ||
92821e2b | 2804 | xfs_alloc_log_agf(tp, agbp, logflags); |
1da177e4 LT |
2805 | *bnop = bno; |
2806 | ||
1da177e4 LT |
2807 | return 0; |
2808 | } | |
2809 | ||
2810 | /* | |
2811 | * Log the given fields from the agf structure. | |
2812 | */ | |
2813 | void | |
2814 | xfs_alloc_log_agf( | |
2815 | xfs_trans_t *tp, /* transaction pointer */ | |
e8222613 | 2816 | struct xfs_buf *bp, /* buffer for a.g. freelist header */ |
1da177e4 LT |
2817 | int fields) /* mask of fields to be logged (XFS_AGF_...) */ |
2818 | { | |
2819 | int first; /* first byte offset */ | |
2820 | int last; /* last byte offset */ | |
2821 | static const short offsets[] = { | |
2822 | offsetof(xfs_agf_t, agf_magicnum), | |
2823 | offsetof(xfs_agf_t, agf_versionnum), | |
2824 | offsetof(xfs_agf_t, agf_seqno), | |
2825 | offsetof(xfs_agf_t, agf_length), | |
2826 | offsetof(xfs_agf_t, agf_roots[0]), | |
2827 | offsetof(xfs_agf_t, agf_levels[0]), | |
2828 | offsetof(xfs_agf_t, agf_flfirst), | |
2829 | offsetof(xfs_agf_t, agf_fllast), | |
2830 | offsetof(xfs_agf_t, agf_flcount), | |
2831 | offsetof(xfs_agf_t, agf_freeblks), | |
2832 | offsetof(xfs_agf_t, agf_longest), | |
92821e2b | 2833 | offsetof(xfs_agf_t, agf_btreeblks), |
4e0e6040 | 2834 | offsetof(xfs_agf_t, agf_uuid), |
f32866fd | 2835 | offsetof(xfs_agf_t, agf_rmap_blocks), |
bdf28630 DW |
2836 | offsetof(xfs_agf_t, agf_refcount_blocks), |
2837 | offsetof(xfs_agf_t, agf_refcount_root), | |
2838 | offsetof(xfs_agf_t, agf_refcount_level), | |
da1f039d DW |
2839 | /* needed so that we don't log the whole rest of the structure: */ |
2840 | offsetof(xfs_agf_t, agf_spare64), | |
1da177e4 LT |
2841 | sizeof(xfs_agf_t) |
2842 | }; | |
2843 | ||
9798f615 | 2844 | trace_xfs_agf(tp->t_mountp, bp->b_addr, fields, _RET_IP_); |
0b1b213f | 2845 | |
61fe135c | 2846 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF); |
4e0e6040 | 2847 | |
1da177e4 LT |
2848 | xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last); |
2849 | xfs_trans_log_buf(tp, bp, (uint)first, (uint)last); | |
2850 | } | |
2851 | ||
2852 | /* | |
2853 | * Interface for inode allocation to force the pag data to be initialized. | |
2854 | */ | |
2855 | int /* error */ | |
2856 | xfs_alloc_pagf_init( | |
2857 | xfs_mount_t *mp, /* file system mount structure */ | |
2858 | xfs_trans_t *tp, /* transaction pointer */ | |
2859 | xfs_agnumber_t agno, /* allocation group number */ | |
2860 | int flags) /* XFS_ALLOC_FLAGS_... */ | |
2861 | { | |
e8222613 | 2862 | struct xfs_buf *bp; |
1da177e4 LT |
2863 | int error; |
2864 | ||
f48e2df8 DW |
2865 | error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp); |
2866 | if (!error) | |
1da177e4 | 2867 | xfs_trans_brelse(tp, bp); |
f48e2df8 | 2868 | return error; |
1da177e4 LT |
2869 | } |
2870 | ||
2871 | /* | |
2872 | * Put the block on the freelist for the allocation group. | |
2873 | */ | |
50920116 | 2874 | int |
1da177e4 | 2875 | xfs_alloc_put_freelist( |
50920116 DC |
2876 | struct xfs_trans *tp, |
2877 | struct xfs_buf *agbp, | |
2878 | struct xfs_buf *agflbp, | |
2879 | xfs_agblock_t bno, | |
2880 | int btreeblk) | |
1da177e4 | 2881 | { |
9798f615 CH |
2882 | struct xfs_mount *mp = tp->t_mountp; |
2883 | struct xfs_agf *agf = agbp->b_addr; | |
50920116 DC |
2884 | struct xfs_perag *pag; |
2885 | __be32 *blockp; | |
1da177e4 | 2886 | int error; |
92821e2b | 2887 | int logflags; |
77c95bba CH |
2888 | __be32 *agfl_bno; |
2889 | int startoff; | |
1da177e4 | 2890 | |
1da177e4 | 2891 | if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp, |
16259e7d | 2892 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 | 2893 | return error; |
413d57c9 | 2894 | be32_add_cpu(&agf->agf_fllast, 1); |
a78ee256 | 2895 | if (be32_to_cpu(agf->agf_fllast) == xfs_agfl_size(mp)) |
1da177e4 | 2896 | agf->agf_fllast = 0; |
a862e0fd | 2897 | |
92a00544 | 2898 | pag = agbp->b_pag; |
a27ba260 | 2899 | ASSERT(!pag->pagf_agflreset); |
413d57c9 | 2900 | be32_add_cpu(&agf->agf_flcount, 1); |
1da177e4 | 2901 | pag->pagf_flcount++; |
92821e2b DC |
2902 | |
2903 | logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT; | |
2904 | if (btreeblk) { | |
413d57c9 | 2905 | be32_add_cpu(&agf->agf_btreeblks, -1); |
92821e2b DC |
2906 | pag->pagf_btreeblks--; |
2907 | logflags |= XFS_AGF_BTREEBLKS; | |
2908 | } | |
2909 | ||
92821e2b DC |
2910 | xfs_alloc_log_agf(tp, agbp, logflags); |
2911 | ||
a78ee256 | 2912 | ASSERT(be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp)); |
77c95bba | 2913 | |
183606d8 | 2914 | agfl_bno = xfs_buf_to_agfl_bno(agflbp); |
77c95bba | 2915 | blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)]; |
e2101005 | 2916 | *blockp = cpu_to_be32(bno); |
77c95bba CH |
2917 | startoff = (char *)blockp - (char *)agflbp->b_addr; |
2918 | ||
92821e2b | 2919 | xfs_alloc_log_agf(tp, agbp, logflags); |
77c95bba | 2920 | |
61fe135c | 2921 | xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF); |
77c95bba CH |
2922 | xfs_trans_log_buf(tp, agflbp, startoff, |
2923 | startoff + sizeof(xfs_agblock_t) - 1); | |
1da177e4 LT |
2924 | return 0; |
2925 | } | |
2926 | ||
a6a781a5 | 2927 | static xfs_failaddr_t |
612cfbfe | 2928 | xfs_agf_verify( |
b5572597 DW |
2929 | struct xfs_buf *bp) |
2930 | { | |
dbd329f1 | 2931 | struct xfs_mount *mp = bp->b_mount; |
9798f615 | 2932 | struct xfs_agf *agf = bp->b_addr; |
5d5f527d | 2933 | |
38c26bfd | 2934 | if (xfs_has_crc(mp)) { |
a45086e2 | 2935 | if (!uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid)) |
a6a781a5 | 2936 | return __this_address; |
9798f615 | 2937 | if (!xfs_log_check_lsn(mp, be64_to_cpu(agf->agf_lsn))) |
a6a781a5 | 2938 | return __this_address; |
a45086e2 | 2939 | } |
5d5f527d | 2940 | |
39708c20 BF |
2941 | if (!xfs_verify_magic(bp, agf->agf_magicnum)) |
2942 | return __this_address; | |
2943 | ||
2944 | if (!(XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) && | |
4e0e6040 | 2945 | be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) && |
a78ee256 DC |
2946 | be32_to_cpu(agf->agf_flfirst) < xfs_agfl_size(mp) && |
2947 | be32_to_cpu(agf->agf_fllast) < xfs_agfl_size(mp) && | |
2948 | be32_to_cpu(agf->agf_flcount) <= xfs_agfl_size(mp))) | |
a6a781a5 | 2949 | return __this_address; |
5d5f527d | 2950 | |
d0c7feaf ZB |
2951 | if (be32_to_cpu(agf->agf_length) > mp->m_sb.sb_dblocks) |
2952 | return __this_address; | |
2953 | ||
2954 | if (be32_to_cpu(agf->agf_freeblks) < be32_to_cpu(agf->agf_longest) || | |
2955 | be32_to_cpu(agf->agf_freeblks) > be32_to_cpu(agf->agf_length)) | |
2956 | return __this_address; | |
2957 | ||
d2a047f3 DW |
2958 | if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) < 1 || |
2959 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) < 1 || | |
7cb3efb4 DW |
2960 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) > |
2961 | mp->m_alloc_maxlevels || | |
2962 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) > | |
2963 | mp->m_alloc_maxlevels) | |
a6a781a5 | 2964 | return __this_address; |
e1b05723 | 2965 | |
38c26bfd | 2966 | if (xfs_has_rmapbt(mp) && |
d2a047f3 | 2967 | (be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) < 1 || |
7cb3efb4 DW |
2968 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) > |
2969 | mp->m_rmap_maxlevels)) | |
a6a781a5 | 2970 | return __this_address; |
b8704944 | 2971 | |
ebd9027d | 2972 | if (xfs_has_rmapbt(mp) && |
d0c7feaf ZB |
2973 | be32_to_cpu(agf->agf_rmap_blocks) > be32_to_cpu(agf->agf_length)) |
2974 | return __this_address; | |
2975 | ||
5d5f527d DC |
2976 | /* |
2977 | * during growfs operations, the perag is not fully initialised, | |
2978 | * so we can't use it for any useful checking. growfs ensures we can't | |
2979 | * use it by using uncached buffers that don't have the perag attached | |
2980 | * so we can detect and avoid this problem. | |
2981 | */ | |
4e0e6040 | 2982 | if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno) |
a6a781a5 | 2983 | return __this_address; |
5d5f527d | 2984 | |
ebd9027d | 2985 | if (xfs_has_lazysbcount(mp) && |
4e0e6040 | 2986 | be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length)) |
a6a781a5 | 2987 | return __this_address; |
4e0e6040 | 2988 | |
ebd9027d | 2989 | if (xfs_has_reflink(mp) && |
d0c7feaf ZB |
2990 | be32_to_cpu(agf->agf_refcount_blocks) > |
2991 | be32_to_cpu(agf->agf_length)) | |
2992 | return __this_address; | |
2993 | ||
ebd9027d | 2994 | if (xfs_has_reflink(mp) && |
d2a047f3 | 2995 | (be32_to_cpu(agf->agf_refcount_level) < 1 || |
973975b7 | 2996 | be32_to_cpu(agf->agf_refcount_level) > mp->m_refc_maxlevels)) |
a6a781a5 | 2997 | return __this_address; |
46eeb521 | 2998 | |
a6a781a5 | 2999 | return NULL; |
5d5f527d | 3000 | |
612cfbfe DC |
3001 | } |
3002 | ||
1813dd64 DC |
3003 | static void |
3004 | xfs_agf_read_verify( | |
612cfbfe DC |
3005 | struct xfs_buf *bp) |
3006 | { | |
dbd329f1 | 3007 | struct xfs_mount *mp = bp->b_mount; |
bc1a09b8 | 3008 | xfs_failaddr_t fa; |
4e0e6040 | 3009 | |
38c26bfd | 3010 | if (xfs_has_crc(mp) && |
ce5028cf | 3011 | !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF)) |
bc1a09b8 DW |
3012 | xfs_verifier_error(bp, -EFSBADCRC, __this_address); |
3013 | else { | |
b5572597 | 3014 | fa = xfs_agf_verify(bp); |
bc1a09b8 DW |
3015 | if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_ALLOC_READ_AGF)) |
3016 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
3017 | } | |
612cfbfe | 3018 | } |
5d5f527d | 3019 | |
b0f539de | 3020 | static void |
1813dd64 | 3021 | xfs_agf_write_verify( |
612cfbfe DC |
3022 | struct xfs_buf *bp) |
3023 | { | |
dbd329f1 | 3024 | struct xfs_mount *mp = bp->b_mount; |
fb1755a6 | 3025 | struct xfs_buf_log_item *bip = bp->b_log_item; |
9798f615 | 3026 | struct xfs_agf *agf = bp->b_addr; |
bc1a09b8 | 3027 | xfs_failaddr_t fa; |
4e0e6040 | 3028 | |
b5572597 | 3029 | fa = xfs_agf_verify(bp); |
bc1a09b8 DW |
3030 | if (fa) { |
3031 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
4e0e6040 DC |
3032 | return; |
3033 | } | |
3034 | ||
38c26bfd | 3035 | if (!xfs_has_crc(mp)) |
4e0e6040 DC |
3036 | return; |
3037 | ||
3038 | if (bip) | |
9798f615 | 3039 | agf->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn); |
4e0e6040 | 3040 | |
f1dbcd7e | 3041 | xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF); |
5d5f527d DC |
3042 | } |
3043 | ||
1813dd64 | 3044 | const struct xfs_buf_ops xfs_agf_buf_ops = { |
233135b7 | 3045 | .name = "xfs_agf", |
39708c20 | 3046 | .magic = { cpu_to_be32(XFS_AGF_MAGIC), cpu_to_be32(XFS_AGF_MAGIC) }, |
1813dd64 DC |
3047 | .verify_read = xfs_agf_read_verify, |
3048 | .verify_write = xfs_agf_write_verify, | |
b5572597 | 3049 | .verify_struct = xfs_agf_verify, |
1813dd64 DC |
3050 | }; |
3051 | ||
1da177e4 LT |
3052 | /* |
3053 | * Read in the allocation group header (free/alloc section). | |
3054 | */ | |
3055 | int /* error */ | |
4805621a CH |
3056 | xfs_read_agf( |
3057 | struct xfs_mount *mp, /* mount point structure */ | |
3058 | struct xfs_trans *tp, /* transaction pointer */ | |
3059 | xfs_agnumber_t agno, /* allocation group number */ | |
3060 | int flags, /* XFS_BUF_ */ | |
3061 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
1da177e4 | 3062 | { |
1da177e4 LT |
3063 | int error; |
3064 | ||
d123031a DC |
3065 | trace_xfs_read_agf(mp, agno); |
3066 | ||
1da177e4 | 3067 | ASSERT(agno != NULLAGNUMBER); |
4ed8e27b | 3068 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, |
1da177e4 | 3069 | XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)), |
1813dd64 | 3070 | XFS_FSS_TO_BB(mp, 1), flags, bpp, &xfs_agf_buf_ops); |
1da177e4 LT |
3071 | if (error) |
3072 | return error; | |
4805621a | 3073 | |
5a52c2a5 | 3074 | ASSERT(!(*bpp)->b_error); |
38f23232 | 3075 | xfs_buf_set_ref(*bpp, XFS_AGF_REF); |
4805621a CH |
3076 | return 0; |
3077 | } | |
3078 | ||
3079 | /* | |
3080 | * Read in the allocation group header (free/alloc section). | |
3081 | */ | |
3082 | int /* error */ | |
3083 | xfs_alloc_read_agf( | |
3084 | struct xfs_mount *mp, /* mount point structure */ | |
3085 | struct xfs_trans *tp, /* transaction pointer */ | |
3086 | xfs_agnumber_t agno, /* allocation group number */ | |
3087 | int flags, /* XFS_ALLOC_FLAG_... */ | |
3088 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
3089 | { | |
3090 | struct xfs_agf *agf; /* ag freelist header */ | |
3091 | struct xfs_perag *pag; /* per allocation group data */ | |
3092 | int error; | |
16eaab83 | 3093 | int allocbt_blks; |
4805621a | 3094 | |
d123031a | 3095 | trace_xfs_alloc_read_agf(mp, agno); |
4805621a | 3096 | |
f48e2df8 DW |
3097 | /* We don't support trylock when freeing. */ |
3098 | ASSERT((flags & (XFS_ALLOC_FLAG_FREEING | XFS_ALLOC_FLAG_TRYLOCK)) != | |
3099 | (XFS_ALLOC_FLAG_FREEING | XFS_ALLOC_FLAG_TRYLOCK)); | |
d123031a | 3100 | ASSERT(agno != NULLAGNUMBER); |
4805621a | 3101 | error = xfs_read_agf(mp, tp, agno, |
0cadda1c | 3102 | (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0, |
4805621a CH |
3103 | bpp); |
3104 | if (error) | |
3105 | return error; | |
5a52c2a5 | 3106 | ASSERT(!(*bpp)->b_error); |
4805621a | 3107 | |
9798f615 | 3108 | agf = (*bpp)->b_addr; |
92a00544 | 3109 | pag = (*bpp)->b_pag; |
1da177e4 | 3110 | if (!pag->pagf_init) { |
16259e7d | 3111 | pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); |
92821e2b | 3112 | pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); |
16259e7d CH |
3113 | pag->pagf_flcount = be32_to_cpu(agf->agf_flcount); |
3114 | pag->pagf_longest = be32_to_cpu(agf->agf_longest); | |
1da177e4 | 3115 | pag->pagf_levels[XFS_BTNUM_BNOi] = |
16259e7d | 3116 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); |
1da177e4 | 3117 | pag->pagf_levels[XFS_BTNUM_CNTi] = |
16259e7d | 3118 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); |
b8704944 DW |
3119 | pag->pagf_levels[XFS_BTNUM_RMAPi] = |
3120 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]); | |
46eeb521 | 3121 | pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level); |
1da177e4 | 3122 | pag->pagf_init = 1; |
a27ba260 | 3123 | pag->pagf_agflreset = xfs_agfl_needs_reset(mp, agf); |
16eaab83 BF |
3124 | |
3125 | /* | |
3126 | * Update the in-core allocbt counter. Filter out the rmapbt | |
3127 | * subset of the btreeblks counter because the rmapbt is managed | |
3128 | * by perag reservation. Subtract one for the rmapbt root block | |
3129 | * because the rmap counter includes it while the btreeblks | |
3130 | * counter only tracks non-root blocks. | |
3131 | */ | |
3132 | allocbt_blks = pag->pagf_btreeblks; | |
ebd9027d | 3133 | if (xfs_has_rmapbt(mp)) |
16eaab83 BF |
3134 | allocbt_blks -= be32_to_cpu(agf->agf_rmap_blocks) - 1; |
3135 | if (allocbt_blks > 0) | |
3136 | atomic64_add(allocbt_blks, &mp->m_allocbt_blks); | |
1da177e4 LT |
3137 | } |
3138 | #ifdef DEBUG | |
75c8c50f | 3139 | else if (!xfs_is_shutdown(mp)) { |
16259e7d | 3140 | ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks)); |
89b28393 | 3141 | ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks)); |
16259e7d CH |
3142 | ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount)); |
3143 | ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest)); | |
1da177e4 | 3144 | ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] == |
16259e7d | 3145 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi])); |
1da177e4 | 3146 | ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] == |
16259e7d | 3147 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); |
1da177e4 LT |
3148 | } |
3149 | #endif | |
1da177e4 LT |
3150 | return 0; |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * Allocate an extent (variable-size). | |
3155 | * Depending on the allocation type, we either look in a single allocation | |
3156 | * group or loop over the allocation groups to find the result. | |
3157 | */ | |
3158 | int /* error */ | |
e04426b9 | 3159 | xfs_alloc_vextent( |
64396ff2 | 3160 | struct xfs_alloc_arg *args) /* allocation argument structure */ |
1da177e4 | 3161 | { |
64396ff2 BF |
3162 | xfs_agblock_t agsize; /* allocation group size */ |
3163 | int error; | |
3164 | int flags; /* XFS_ALLOC_FLAG_... locking flags */ | |
3165 | struct xfs_mount *mp; /* mount structure pointer */ | |
3166 | xfs_agnumber_t sagno; /* starting allocation group number */ | |
3167 | xfs_alloctype_t type; /* input allocation type */ | |
3168 | int bump_rotor = 0; | |
3169 | xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */ | |
1da177e4 LT |
3170 | |
3171 | mp = args->mp; | |
3172 | type = args->otype = args->type; | |
3173 | args->agbno = NULLAGBLOCK; | |
3174 | /* | |
3175 | * Just fix this up, for the case where the last a.g. is shorter | |
3176 | * (or there's only one a.g.) and the caller couldn't easily figure | |
3177 | * that out (xfs_bmap_alloc). | |
3178 | */ | |
3179 | agsize = mp->m_sb.sb_agblocks; | |
3180 | if (args->maxlen > agsize) | |
3181 | args->maxlen = agsize; | |
3182 | if (args->alignment == 0) | |
3183 | args->alignment = 1; | |
3184 | ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount); | |
3185 | ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize); | |
3186 | ASSERT(args->minlen <= args->maxlen); | |
3187 | ASSERT(args->minlen <= agsize); | |
3188 | ASSERT(args->mod < args->prod); | |
3189 | if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount || | |
3190 | XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize || | |
3191 | args->minlen > args->maxlen || args->minlen > agsize || | |
3192 | args->mod >= args->prod) { | |
3193 | args->fsbno = NULLFSBLOCK; | |
0b1b213f | 3194 | trace_xfs_alloc_vextent_badargs(args); |
1da177e4 LT |
3195 | return 0; |
3196 | } | |
1da177e4 LT |
3197 | |
3198 | switch (type) { | |
3199 | case XFS_ALLOCTYPE_THIS_AG: | |
3200 | case XFS_ALLOCTYPE_NEAR_BNO: | |
3201 | case XFS_ALLOCTYPE_THIS_BNO: | |
3202 | /* | |
3203 | * These three force us into a single a.g. | |
3204 | */ | |
3205 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
a862e0fd | 3206 | args->pag = xfs_perag_get(mp, args->agno); |
1da177e4 | 3207 | error = xfs_alloc_fix_freelist(args, 0); |
1da177e4 | 3208 | if (error) { |
0b1b213f | 3209 | trace_xfs_alloc_vextent_nofix(args); |
1da177e4 LT |
3210 | goto error0; |
3211 | } | |
3212 | if (!args->agbp) { | |
0b1b213f | 3213 | trace_xfs_alloc_vextent_noagbp(args); |
1da177e4 LT |
3214 | break; |
3215 | } | |
3216 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
3217 | if ((error = xfs_alloc_ag_vextent(args))) | |
3218 | goto error0; | |
1da177e4 LT |
3219 | break; |
3220 | case XFS_ALLOCTYPE_START_BNO: | |
3221 | /* | |
3222 | * Try near allocation first, then anywhere-in-ag after | |
3223 | * the first a.g. fails. | |
3224 | */ | |
292378ed | 3225 | if ((args->datatype & XFS_ALLOC_INITIAL_USER_DATA) && |
2e973b2c | 3226 | xfs_is_inode32(mp)) { |
1da177e4 LT |
3227 | args->fsbno = XFS_AGB_TO_FSB(mp, |
3228 | ((mp->m_agfrotor / rotorstep) % | |
3229 | mp->m_sb.sb_agcount), 0); | |
3230 | bump_rotor = 1; | |
3231 | } | |
3232 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
3233 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
53004ee7 | 3234 | fallthrough; |
1da177e4 LT |
3235 | case XFS_ALLOCTYPE_FIRST_AG: |
3236 | /* | |
3237 | * Rotate through the allocation groups looking for a winner. | |
3238 | */ | |
8d242e93 | 3239 | if (type == XFS_ALLOCTYPE_FIRST_AG) { |
1da177e4 LT |
3240 | /* |
3241 | * Start with allocation group given by bno. | |
3242 | */ | |
3243 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3244 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
3245 | sagno = 0; | |
3246 | flags = 0; | |
3247 | } else { | |
1da177e4 LT |
3248 | /* |
3249 | * Start with the given allocation group. | |
3250 | */ | |
3251 | args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
3252 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
3253 | } | |
3254 | /* | |
3255 | * Loop over allocation groups twice; first time with | |
3256 | * trylock set, second time without. | |
3257 | */ | |
1da177e4 | 3258 | for (;;) { |
a862e0fd | 3259 | args->pag = xfs_perag_get(mp, args->agno); |
1da177e4 | 3260 | error = xfs_alloc_fix_freelist(args, flags); |
1da177e4 | 3261 | if (error) { |
0b1b213f | 3262 | trace_xfs_alloc_vextent_nofix(args); |
1da177e4 LT |
3263 | goto error0; |
3264 | } | |
3265 | /* | |
3266 | * If we get a buffer back then the allocation will fly. | |
3267 | */ | |
3268 | if (args->agbp) { | |
3269 | if ((error = xfs_alloc_ag_vextent(args))) | |
3270 | goto error0; | |
3271 | break; | |
3272 | } | |
0b1b213f CH |
3273 | |
3274 | trace_xfs_alloc_vextent_loopfailed(args); | |
3275 | ||
1da177e4 LT |
3276 | /* |
3277 | * Didn't work, figure out the next iteration. | |
3278 | */ | |
3279 | if (args->agno == sagno && | |
3280 | type == XFS_ALLOCTYPE_START_BNO) | |
3281 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
d210a28c YL |
3282 | /* |
3283 | * For the first allocation, we can try any AG to get | |
3284 | * space. However, if we already have allocated a | |
3285 | * block, we don't want to try AGs whose number is below | |
3286 | * sagno. Otherwise, we may end up with out-of-order | |
3287 | * locking of AGF, which might cause deadlock. | |
3288 | */ | |
3289 | if (++(args->agno) == mp->m_sb.sb_agcount) { | |
64396ff2 | 3290 | if (args->tp->t_firstblock != NULLFSBLOCK) |
d210a28c YL |
3291 | args->agno = sagno; |
3292 | else | |
3293 | args->agno = 0; | |
3294 | } | |
1da177e4 LT |
3295 | /* |
3296 | * Reached the starting a.g., must either be done | |
3297 | * or switch to non-trylock mode. | |
3298 | */ | |
3299 | if (args->agno == sagno) { | |
255c5162 | 3300 | if (flags == 0) { |
1da177e4 | 3301 | args->agbno = NULLAGBLOCK; |
0b1b213f | 3302 | trace_xfs_alloc_vextent_allfailed(args); |
1da177e4 LT |
3303 | break; |
3304 | } | |
255c5162 CH |
3305 | |
3306 | flags = 0; | |
3307 | if (type == XFS_ALLOCTYPE_START_BNO) { | |
3308 | args->agbno = XFS_FSB_TO_AGBNO(mp, | |
3309 | args->fsbno); | |
3310 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
1da177e4 LT |
3311 | } |
3312 | } | |
a862e0fd | 3313 | xfs_perag_put(args->pag); |
1da177e4 | 3314 | } |
8d242e93 | 3315 | if (bump_rotor) { |
1da177e4 LT |
3316 | if (args->agno == sagno) |
3317 | mp->m_agfrotor = (mp->m_agfrotor + 1) % | |
3318 | (mp->m_sb.sb_agcount * rotorstep); | |
3319 | else | |
3320 | mp->m_agfrotor = (args->agno * rotorstep + 1) % | |
3321 | (mp->m_sb.sb_agcount * rotorstep); | |
3322 | } | |
3323 | break; | |
3324 | default: | |
3325 | ASSERT(0); | |
3326 | /* NOTREACHED */ | |
3327 | } | |
3328 | if (args->agbno == NULLAGBLOCK) | |
3329 | args->fsbno = NULLFSBLOCK; | |
3330 | else { | |
3331 | args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno); | |
3332 | #ifdef DEBUG | |
3333 | ASSERT(args->len >= args->minlen); | |
3334 | ASSERT(args->len <= args->maxlen); | |
3335 | ASSERT(args->agbno % args->alignment == 0); | |
3336 | XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), | |
3337 | args->len); | |
3338 | #endif | |
3fbbbea3 | 3339 | |
1da177e4 | 3340 | } |
a862e0fd | 3341 | xfs_perag_put(args->pag); |
1da177e4 LT |
3342 | return 0; |
3343 | error0: | |
a862e0fd | 3344 | xfs_perag_put(args->pag); |
1da177e4 LT |
3345 | return error; |
3346 | } | |
3347 | ||
4d89e20b DC |
3348 | /* Ensure that the freelist is at full capacity. */ |
3349 | int | |
3350 | xfs_free_extent_fix_freelist( | |
3351 | struct xfs_trans *tp, | |
45d06621 | 3352 | struct xfs_perag *pag, |
4d89e20b | 3353 | struct xfs_buf **agbp) |
1da177e4 | 3354 | { |
4d89e20b DC |
3355 | struct xfs_alloc_arg args; |
3356 | int error; | |
1da177e4 | 3357 | |
4d89e20b | 3358 | memset(&args, 0, sizeof(struct xfs_alloc_arg)); |
1da177e4 LT |
3359 | args.tp = tp; |
3360 | args.mp = tp->t_mountp; | |
45d06621 DC |
3361 | args.agno = pag->pag_agno; |
3362 | args.pag = pag; | |
be65b18a DC |
3363 | |
3364 | /* | |
3365 | * validate that the block number is legal - the enables us to detect | |
3366 | * and handle a silent filesystem corruption rather than crashing. | |
3367 | */ | |
be65b18a | 3368 | if (args.agno >= args.mp->m_sb.sb_agcount) |
2451337d | 3369 | return -EFSCORRUPTED; |
be65b18a | 3370 | |
be65b18a DC |
3371 | error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING); |
3372 | if (error) | |
45d06621 | 3373 | return error; |
4d89e20b DC |
3374 | |
3375 | *agbp = args.agbp; | |
45d06621 | 3376 | return 0; |
4d89e20b DC |
3377 | } |
3378 | ||
3379 | /* | |
3380 | * Free an extent. | |
3381 | * Just break up the extent address and hand off to xfs_free_ag_extent | |
3382 | * after fixing up the freelist. | |
3383 | */ | |
66e3237e | 3384 | int |
fcb762f5 | 3385 | __xfs_free_extent( |
66e3237e DW |
3386 | struct xfs_trans *tp, |
3387 | xfs_fsblock_t bno, | |
3388 | xfs_extlen_t len, | |
3389 | const struct xfs_owner_info *oinfo, | |
3390 | enum xfs_ag_resv_type type, | |
3391 | bool skip_discard) | |
4d89e20b | 3392 | { |
66e3237e DW |
3393 | struct xfs_mount *mp = tp->t_mountp; |
3394 | struct xfs_buf *agbp; | |
3395 | xfs_agnumber_t agno = XFS_FSB_TO_AGNO(mp, bno); | |
3396 | xfs_agblock_t agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
9798f615 | 3397 | struct xfs_agf *agf; |
66e3237e DW |
3398 | int error; |
3399 | unsigned int busy_flags = 0; | |
45d06621 | 3400 | struct xfs_perag *pag; |
4d89e20b DC |
3401 | |
3402 | ASSERT(len != 0); | |
0ab32086 | 3403 | ASSERT(type != XFS_AG_RESV_AGFL); |
4d89e20b | 3404 | |
ba9e7802 | 3405 | if (XFS_TEST_ERROR(false, mp, |
9e24cfd0 | 3406 | XFS_ERRTAG_FREE_EXTENT)) |
ba9e7802 DW |
3407 | return -EIO; |
3408 | ||
45d06621 DC |
3409 | pag = xfs_perag_get(mp, agno); |
3410 | error = xfs_free_extent_fix_freelist(tp, pag, &agbp); | |
4d89e20b | 3411 | if (error) |
45d06621 | 3412 | goto err; |
9798f615 | 3413 | agf = agbp->b_addr; |
4d89e20b | 3414 | |
f9e03706 DW |
3415 | if (XFS_IS_CORRUPT(mp, agbno >= mp->m_sb.sb_agblocks)) { |
3416 | error = -EFSCORRUPTED; | |
45d06621 | 3417 | goto err_release; |
f9e03706 | 3418 | } |
be65b18a DC |
3419 | |
3420 | /* validate the extent size is legal now we have the agf locked */ | |
9798f615 | 3421 | if (XFS_IS_CORRUPT(mp, agbno + len > be32_to_cpu(agf->agf_length))) { |
f9e03706 | 3422 | error = -EFSCORRUPTED; |
45d06621 | 3423 | goto err_release; |
f9e03706 | 3424 | } |
be65b18a | 3425 | |
3fd129b6 | 3426 | error = xfs_free_ag_extent(tp, agbp, agno, agbno, len, oinfo, type); |
4d89e20b | 3427 | if (error) |
45d06621 | 3428 | goto err_release; |
4d89e20b | 3429 | |
fcb762f5 BF |
3430 | if (skip_discard) |
3431 | busy_flags |= XFS_EXTENT_BUSY_SKIP_DISCARD; | |
45d06621 DC |
3432 | xfs_extent_busy_insert(tp, pag, agbno, len, busy_flags); |
3433 | xfs_perag_put(pag); | |
4d89e20b DC |
3434 | return 0; |
3435 | ||
45d06621 | 3436 | err_release: |
4d89e20b | 3437 | xfs_trans_brelse(tp, agbp); |
45d06621 DC |
3438 | err: |
3439 | xfs_perag_put(pag); | |
1da177e4 LT |
3440 | return error; |
3441 | } | |
2d520bfa DW |
3442 | |
3443 | struct xfs_alloc_query_range_info { | |
3444 | xfs_alloc_query_range_fn fn; | |
3445 | void *priv; | |
3446 | }; | |
3447 | ||
3448 | /* Format btree record and pass to our callback. */ | |
3449 | STATIC int | |
3450 | xfs_alloc_query_range_helper( | |
3451 | struct xfs_btree_cur *cur, | |
159eb69d | 3452 | const union xfs_btree_rec *rec, |
2d520bfa DW |
3453 | void *priv) |
3454 | { | |
3455 | struct xfs_alloc_query_range_info *query = priv; | |
3456 | struct xfs_alloc_rec_incore irec; | |
3457 | ||
3458 | irec.ar_startblock = be32_to_cpu(rec->alloc.ar_startblock); | |
3459 | irec.ar_blockcount = be32_to_cpu(rec->alloc.ar_blockcount); | |
3460 | return query->fn(cur, &irec, query->priv); | |
3461 | } | |
3462 | ||
3463 | /* Find all free space within a given range of blocks. */ | |
3464 | int | |
3465 | xfs_alloc_query_range( | |
3466 | struct xfs_btree_cur *cur, | |
04dcb474 DW |
3467 | const struct xfs_alloc_rec_incore *low_rec, |
3468 | const struct xfs_alloc_rec_incore *high_rec, | |
2d520bfa DW |
3469 | xfs_alloc_query_range_fn fn, |
3470 | void *priv) | |
3471 | { | |
3472 | union xfs_btree_irec low_brec; | |
3473 | union xfs_btree_irec high_brec; | |
3474 | struct xfs_alloc_query_range_info query; | |
3475 | ||
3476 | ASSERT(cur->bc_btnum == XFS_BTNUM_BNO); | |
3477 | low_brec.a = *low_rec; | |
3478 | high_brec.a = *high_rec; | |
3479 | query.priv = priv; | |
3480 | query.fn = fn; | |
3481 | return xfs_btree_query_range(cur, &low_brec, &high_brec, | |
3482 | xfs_alloc_query_range_helper, &query); | |
3483 | } | |
e9a2599a DW |
3484 | |
3485 | /* Find all free space records. */ | |
3486 | int | |
3487 | xfs_alloc_query_all( | |
3488 | struct xfs_btree_cur *cur, | |
3489 | xfs_alloc_query_range_fn fn, | |
3490 | void *priv) | |
3491 | { | |
3492 | struct xfs_alloc_query_range_info query; | |
3493 | ||
3494 | ASSERT(cur->bc_btnum == XFS_BTNUM_BNO); | |
3495 | query.priv = priv; | |
3496 | query.fn = fn; | |
3497 | return xfs_btree_query_all(cur, xfs_alloc_query_range_helper, &query); | |
3498 | } | |
21ec5416 | 3499 | |
ce1d802e DW |
3500 | /* Is there a record covering a given extent? */ |
3501 | int | |
3502 | xfs_alloc_has_record( | |
3503 | struct xfs_btree_cur *cur, | |
3504 | xfs_agblock_t bno, | |
3505 | xfs_extlen_t len, | |
3506 | bool *exists) | |
3507 | { | |
3508 | union xfs_btree_irec low; | |
3509 | union xfs_btree_irec high; | |
3510 | ||
3511 | memset(&low, 0, sizeof(low)); | |
3512 | low.a.ar_startblock = bno; | |
3513 | memset(&high, 0xFF, sizeof(high)); | |
3514 | high.a.ar_startblock = bno + len - 1; | |
3515 | ||
3516 | return xfs_btree_has_record(cur, &low, &high, exists); | |
3517 | } | |
9f3a080e DW |
3518 | |
3519 | /* | |
3520 | * Walk all the blocks in the AGFL. The @walk_fn can return any negative | |
5bb46e3e | 3521 | * error code or XFS_ITER_*. |
9f3a080e DW |
3522 | */ |
3523 | int | |
3524 | xfs_agfl_walk( | |
3525 | struct xfs_mount *mp, | |
3526 | struct xfs_agf *agf, | |
3527 | struct xfs_buf *agflbp, | |
3528 | xfs_agfl_walk_fn walk_fn, | |
3529 | void *priv) | |
3530 | { | |
3531 | __be32 *agfl_bno; | |
3532 | unsigned int i; | |
3533 | int error; | |
3534 | ||
183606d8 | 3535 | agfl_bno = xfs_buf_to_agfl_bno(agflbp); |
9f3a080e DW |
3536 | i = be32_to_cpu(agf->agf_flfirst); |
3537 | ||
3538 | /* Nothing to walk in an empty AGFL. */ | |
3539 | if (agf->agf_flcount == cpu_to_be32(0)) | |
3540 | return 0; | |
3541 | ||
3542 | /* Otherwise, walk from first to last, wrapping as needed. */ | |
3543 | for (;;) { | |
3544 | error = walk_fn(mp, be32_to_cpu(agfl_bno[i]), priv); | |
3545 | if (error) | |
3546 | return error; | |
3547 | if (i == be32_to_cpu(agf->agf_fllast)) | |
3548 | break; | |
3549 | if (++i == xfs_agfl_size(mp)) | |
3550 | i = 0; | |
3551 | } | |
3552 | ||
3553 | return 0; | |
3554 | } | |
c201d9ca DW |
3555 | |
3556 | int __init | |
3557 | xfs_extfree_intent_init_cache(void) | |
3558 | { | |
3559 | xfs_extfree_item_cache = kmem_cache_create("xfs_extfree_intent", | |
3560 | sizeof(struct xfs_extent_free_item), | |
3561 | 0, 0, NULL); | |
3562 | ||
3563 | return xfs_extfree_item_cache != NULL ? 0 : -ENOMEM; | |
3564 | } | |
3565 | ||
3566 | void | |
3567 | xfs_extfree_intent_destroy_cache(void) | |
3568 | { | |
3569 | kmem_cache_destroy(xfs_extfree_item_cache); | |
3570 | xfs_extfree_item_cache = NULL; | |
3571 | } |