]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
5e656dbb | 2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
da23017d | 3 | * All Rights Reserved. |
5000d01d | 4 | * |
da23017d NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
5000d01d | 13 | * |
da23017d NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
2bd0ea18 | 17 | */ |
9c799827 | 18 | #include "libxfs_priv.h" |
b626fb59 DC |
19 | #include "xfs_fs.h" |
20 | #include "xfs_format.h" | |
21 | #include "xfs_log_format.h" | |
22 | #include "xfs_shared.h" | |
23 | #include "xfs_trans_resv.h" | |
24 | #include "xfs_bit.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_mount.h" | |
f944d3d0 | 27 | #include "xfs_defer.h" |
b626fb59 DC |
28 | #include "xfs_inode.h" |
29 | #include "xfs_btree.h" | |
631ac87a | 30 | #include "xfs_rmap.h" |
b626fb59 DC |
31 | #include "xfs_alloc_btree.h" |
32 | #include "xfs_alloc.h" | |
33 | #include "xfs_cksum.h" | |
34 | #include "xfs_trace.h" | |
35 | #include "xfs_trans.h" | |
cf8ce220 | 36 | #include "xfs_ag_resv.h" |
2bd0ea18 | 37 | |
ff105f75 DC |
38 | struct workqueue_struct *xfs_alloc_wq; |
39 | ||
2bd0ea18 | 40 | #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b))) |
5e656dbb BN |
41 | |
42 | #define XFSA_FIXUP_BNO_OK 1 | |
43 | #define XFSA_FIXUP_CNT_OK 2 | |
44 | ||
5e656dbb BN |
45 | STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *); |
46 | STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *); | |
47 | STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *); | |
48 | STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *, | |
a2ceac1f | 49 | xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *); |
2bd0ea18 | 50 | |
2a96beb9 DW |
51 | unsigned int |
52 | xfs_refc_block( | |
53 | struct xfs_mount *mp) | |
54 | { | |
55 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
56 | return XFS_RMAP_BLOCK(mp) + 1; | |
57 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) | |
58 | return XFS_FIBT_BLOCK(mp) + 1; | |
59 | return XFS_IBT_BLOCK(mp) + 1; | |
60 | } | |
61 | ||
ef5340cd DW |
62 | xfs_extlen_t |
63 | xfs_prealloc_blocks( | |
64 | struct xfs_mount *mp) | |
65 | { | |
2a96beb9 DW |
66 | if (xfs_sb_version_hasreflink(&mp->m_sb)) |
67 | return xfs_refc_block(mp) + 1; | |
ef5340cd DW |
68 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) |
69 | return XFS_RMAP_BLOCK(mp) + 1; | |
70 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) | |
71 | return XFS_FIBT_BLOCK(mp) + 1; | |
72 | return XFS_IBT_BLOCK(mp) + 1; | |
73 | } | |
74 | ||
b8a8d6e5 DW |
75 | /* |
76 | * In order to avoid ENOSPC-related deadlock caused by out-of-order locking of | |
77 | * AGF buffer (PV 947395), we place constraints on the relationship among | |
78 | * actual allocations for data blocks, freelist blocks, and potential file data | |
79 | * bmap btree blocks. However, these restrictions may result in no actual space | |
80 | * allocated for a delayed extent, for example, a data block in a certain AG is | |
81 | * allocated but there is no additional block for the additional bmap btree | |
82 | * block due to a split of the bmap btree of the file. The result of this may | |
83 | * lead to an infinite loop when the file gets flushed to disk and all delayed | |
84 | * extents need to be actually allocated. To get around this, we explicitly set | |
85 | * aside a few blocks which will not be reserved in delayed allocation. | |
86 | * | |
cf8ce220 DW |
87 | * We need to reserve 4 fsbs _per AG_ for the freelist and 4 more to handle a |
88 | * potential split of the file's bmap btree. | |
b8a8d6e5 DW |
89 | */ |
90 | unsigned int | |
91 | xfs_alloc_set_aside( | |
92 | struct xfs_mount *mp) | |
93 | { | |
8eeb15ea | 94 | return mp->m_sb.sb_agcount * (XFS_ALLOC_AGFL_RESERVE + 4); |
b8a8d6e5 DW |
95 | } |
96 | ||
97 | /* | |
98 | * When deciding how much space to allocate out of an AG, we limit the | |
99 | * allocation maximum size to the size the AG. However, we cannot use all the | |
100 | * blocks in the AG - some are permanently used by metadata. These | |
101 | * blocks are generally: | |
102 | * - the AG superblock, AGF, AGI and AGFL | |
103 | * - the AGF (bno and cnt) and AGI btree root blocks, and optionally | |
104 | * the AGI free inode and rmap btree root blocks. | |
105 | * - blocks on the AGFL according to xfs_alloc_set_aside() limits | |
106 | * - the rmapbt root block | |
107 | * | |
108 | * The AG headers are sector sized, so the amount of space they take up is | |
109 | * dependent on filesystem geometry. The others are all single blocks. | |
110 | */ | |
111 | unsigned int | |
112 | xfs_alloc_ag_max_usable( | |
113 | struct xfs_mount *mp) | |
114 | { | |
115 | unsigned int blocks; | |
116 | ||
117 | blocks = XFS_BB_TO_FSB(mp, XFS_FSS_TO_BB(mp, 4)); /* ag headers */ | |
118 | blocks += XFS_ALLOC_AGFL_RESERVE; | |
119 | blocks += 3; /* AGF, AGI btree root blocks */ | |
120 | if (xfs_sb_version_hasfinobt(&mp->m_sb)) | |
121 | blocks++; /* finobt root block */ | |
122 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
123 | blocks++; /* rmap root block */ | |
868c70e3 DW |
124 | if (xfs_sb_version_hasreflink(&mp->m_sb)) |
125 | blocks++; /* refcount root block */ | |
b8a8d6e5 DW |
126 | |
127 | return mp->m_sb.sb_agblocks - blocks; | |
128 | } | |
129 | ||
b194c7d8 BN |
130 | /* |
131 | * Lookup the record equal to [bno, len] in the btree given by cur. | |
132 | */ | |
133 | STATIC int /* error */ | |
134 | xfs_alloc_lookup_eq( | |
135 | struct xfs_btree_cur *cur, /* btree cursor */ | |
136 | xfs_agblock_t bno, /* starting block of extent */ | |
137 | xfs_extlen_t len, /* length of extent */ | |
138 | int *stat) /* success/failure */ | |
139 | { | |
140 | cur->bc_rec.a.ar_startblock = bno; | |
141 | cur->bc_rec.a.ar_blockcount = len; | |
142 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); | |
143 | } | |
144 | ||
145 | /* | |
146 | * Lookup the first record greater than or equal to [bno, len] | |
147 | * in the btree given by cur. | |
148 | */ | |
a2ceac1f | 149 | int /* error */ |
b194c7d8 BN |
150 | xfs_alloc_lookup_ge( |
151 | struct xfs_btree_cur *cur, /* btree cursor */ | |
152 | xfs_agblock_t bno, /* starting block of extent */ | |
153 | xfs_extlen_t len, /* length of extent */ | |
154 | int *stat) /* success/failure */ | |
155 | { | |
156 | cur->bc_rec.a.ar_startblock = bno; | |
157 | cur->bc_rec.a.ar_blockcount = len; | |
158 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); | |
159 | } | |
160 | ||
161 | /* | |
162 | * Lookup the first record less than or equal to [bno, len] | |
163 | * in the btree given by cur. | |
164 | */ | |
2d066e16 | 165 | static int /* error */ |
b194c7d8 BN |
166 | xfs_alloc_lookup_le( |
167 | struct xfs_btree_cur *cur, /* btree cursor */ | |
168 | xfs_agblock_t bno, /* starting block of extent */ | |
169 | xfs_extlen_t len, /* length of extent */ | |
170 | int *stat) /* success/failure */ | |
171 | { | |
172 | cur->bc_rec.a.ar_startblock = bno; | |
173 | cur->bc_rec.a.ar_blockcount = len; | |
174 | return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat); | |
175 | } | |
176 | ||
177 | /* | |
178 | * Update the record referred to by cur to the value given | |
179 | * by [bno, len]. | |
180 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
181 | */ | |
182 | STATIC int /* error */ | |
183 | xfs_alloc_update( | |
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 | { | |
188 | union xfs_btree_rec rec; | |
189 | ||
190 | rec.alloc.ar_startblock = cpu_to_be32(bno); | |
191 | rec.alloc.ar_blockcount = cpu_to_be32(len); | |
192 | return xfs_btree_update(cur, &rec); | |
193 | } | |
194 | ||
195 | /* | |
196 | * Get the data from the pointed-to record. | |
197 | */ | |
a2ceac1f | 198 | int /* error */ |
b194c7d8 BN |
199 | xfs_alloc_get_rec( |
200 | struct xfs_btree_cur *cur, /* btree cursor */ | |
201 | xfs_agblock_t *bno, /* output: starting block of extent */ | |
202 | xfs_extlen_t *len, /* output: length of extent */ | |
203 | int *stat) /* output: success/failure */ | |
204 | { | |
205 | union xfs_btree_rec *rec; | |
206 | int error; | |
207 | ||
208 | error = xfs_btree_get_rec(cur, &rec, stat); | |
209 | if (!error && *stat == 1) { | |
210 | *bno = be32_to_cpu(rec->alloc.ar_startblock); | |
211 | *len = be32_to_cpu(rec->alloc.ar_blockcount); | |
212 | } | |
213 | return error; | |
214 | } | |
215 | ||
2bd0ea18 NS |
216 | /* |
217 | * Compute aligned version of the found extent. | |
218 | * Takes alignment and min length into account. | |
219 | */ | |
5e656dbb | 220 | STATIC void |
2bd0ea18 | 221 | xfs_alloc_compute_aligned( |
a2ceac1f | 222 | xfs_alloc_arg_t *args, /* allocation argument structure */ |
2bd0ea18 NS |
223 | xfs_agblock_t foundbno, /* starting block in found extent */ |
224 | xfs_extlen_t foundlen, /* length in found extent */ | |
2bd0ea18 NS |
225 | xfs_agblock_t *resbno, /* result block number */ |
226 | xfs_extlen_t *reslen) /* result length */ | |
227 | { | |
228 | xfs_agblock_t bno; | |
2bd0ea18 | 229 | xfs_extlen_t len; |
ff3263dd | 230 | xfs_extlen_t diff; |
2bd0ea18 | 231 | |
a2ceac1f DC |
232 | /* Trim busy sections out of found extent */ |
233 | xfs_extent_busy_trim(args, foundbno, foundlen, &bno, &len); | |
234 | ||
ff3263dd BF |
235 | /* |
236 | * If we have a largish extent that happens to start before min_agbno, | |
237 | * see if we can shift it into range... | |
238 | */ | |
239 | if (bno < args->min_agbno && bno + len > args->min_agbno) { | |
240 | diff = args->min_agbno - bno; | |
241 | if (len > diff) { | |
242 | bno += diff; | |
243 | len -= diff; | |
244 | } | |
245 | } | |
246 | ||
a2ceac1f DC |
247 | if (args->alignment > 1 && len >= args->minlen) { |
248 | xfs_agblock_t aligned_bno = roundup(bno, args->alignment); | |
ff3263dd BF |
249 | |
250 | diff = aligned_bno - bno; | |
a2ceac1f DC |
251 | |
252 | *resbno = aligned_bno; | |
253 | *reslen = diff >= len ? 0 : len - diff; | |
2bd0ea18 | 254 | } else { |
a2ceac1f DC |
255 | *resbno = bno; |
256 | *reslen = len; | |
2bd0ea18 | 257 | } |
2bd0ea18 NS |
258 | } |
259 | ||
260 | /* | |
261 | * Compute best start block and diff for "near" allocations. | |
262 | * freelen >= wantlen already checked by caller. | |
263 | */ | |
264 | STATIC xfs_extlen_t /* difference value (absolute) */ | |
265 | xfs_alloc_compute_diff( | |
266 | xfs_agblock_t wantbno, /* target starting block */ | |
267 | xfs_extlen_t wantlen, /* target length */ | |
268 | xfs_extlen_t alignment, /* target alignment */ | |
1fccd5c8 | 269 | int datatype, /* are we allocating data? */ |
2bd0ea18 NS |
270 | xfs_agblock_t freebno, /* freespace's starting block */ |
271 | xfs_extlen_t freelen, /* freespace's length */ | |
272 | xfs_agblock_t *newbnop) /* result: best start block from free */ | |
273 | { | |
274 | xfs_agblock_t freeend; /* end of freespace extent */ | |
275 | xfs_agblock_t newbno1; /* return block number */ | |
276 | xfs_agblock_t newbno2; /* other new block number */ | |
0e266570 NS |
277 | xfs_extlen_t newlen1=0; /* length with newbno1 */ |
278 | xfs_extlen_t newlen2=0; /* length with newbno2 */ | |
2bd0ea18 | 279 | xfs_agblock_t wantend; /* end of target extent */ |
1fccd5c8 | 280 | bool userdata = xfs_alloc_is_userdata(datatype); |
2bd0ea18 NS |
281 | |
282 | ASSERT(freelen >= wantlen); | |
283 | freeend = freebno + freelen; | |
284 | wantend = wantbno + wantlen; | |
84a62eea DC |
285 | /* |
286 | * We want to allocate from the start of a free extent if it is past | |
287 | * the desired block or if we are allocating user data and the free | |
288 | * extent is before desired block. The second case is there to allow | |
289 | * for contiguous allocation from the remaining free space if the file | |
290 | * grows in the short term. | |
291 | */ | |
292 | if (freebno >= wantbno || (userdata && freeend < wantend)) { | |
2bd0ea18 NS |
293 | if ((newbno1 = roundup(freebno, alignment)) >= freeend) |
294 | newbno1 = NULLAGBLOCK; | |
295 | } else if (freeend >= wantend && alignment > 1) { | |
296 | newbno1 = roundup(wantbno, alignment); | |
297 | newbno2 = newbno1 - alignment; | |
298 | if (newbno1 >= freeend) | |
299 | newbno1 = NULLAGBLOCK; | |
300 | else | |
301 | newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1); | |
302 | if (newbno2 < freebno) | |
303 | newbno2 = NULLAGBLOCK; | |
304 | else | |
305 | newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2); | |
306 | if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) { | |
307 | if (newlen1 < newlen2 || | |
308 | (newlen1 == newlen2 && | |
309 | XFS_ABSDIFF(newbno1, wantbno) > | |
310 | XFS_ABSDIFF(newbno2, wantbno))) | |
311 | newbno1 = newbno2; | |
312 | } else if (newbno2 != NULLAGBLOCK) | |
313 | newbno1 = newbno2; | |
314 | } else if (freeend >= wantend) { | |
315 | newbno1 = wantbno; | |
316 | } else if (alignment > 1) { | |
317 | newbno1 = roundup(freeend - wantlen, alignment); | |
318 | if (newbno1 > freeend - wantlen && | |
319 | newbno1 - alignment >= freebno) | |
320 | newbno1 -= alignment; | |
321 | else if (newbno1 >= freeend) | |
322 | newbno1 = NULLAGBLOCK; | |
323 | } else | |
324 | newbno1 = freeend - wantlen; | |
325 | *newbnop = newbno1; | |
326 | return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno); | |
327 | } | |
328 | ||
329 | /* | |
330 | * Fix up the length, based on mod and prod. | |
331 | * len should be k * prod + mod for some k. | |
332 | * If len is too small it is returned unchanged. | |
333 | * If len hits maxlen it is left alone. | |
334 | */ | |
335 | STATIC void | |
336 | xfs_alloc_fix_len( | |
dfc130f3 | 337 | xfs_alloc_arg_t *args) /* allocation argument structure */ |
2bd0ea18 NS |
338 | { |
339 | xfs_extlen_t k; | |
340 | xfs_extlen_t rlen; | |
341 | ||
342 | ASSERT(args->mod < args->prod); | |
343 | rlen = args->len; | |
344 | ASSERT(rlen >= args->minlen); | |
345 | ASSERT(rlen <= args->maxlen); | |
346 | if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen || | |
347 | (args->mod == 0 && rlen < args->prod)) | |
348 | return; | |
349 | k = rlen % args->prod; | |
350 | if (k == args->mod) | |
351 | return; | |
ff105f75 DC |
352 | if (k > args->mod) |
353 | rlen = rlen - (k - args->mod); | |
354 | else | |
355 | rlen = rlen - args->prod + (args->mod - k); | |
19ebedcf | 356 | /* casts to (int) catch length underflows */ |
ff105f75 DC |
357 | if ((int)rlen < (int)args->minlen) |
358 | return; | |
359 | ASSERT(rlen >= args->minlen && rlen <= args->maxlen); | |
360 | ASSERT(rlen % args->prod == args->mod); | |
2c003dc2 CH |
361 | ASSERT(args->pag->pagf_freeblks + args->pag->pagf_flcount >= |
362 | rlen + args->minleft); | |
2bd0ea18 NS |
363 | args->len = rlen; |
364 | } | |
365 | ||
2bd0ea18 NS |
366 | /* |
367 | * Update the two btrees, logically removing from freespace the extent | |
368 | * starting at rbno, rlen blocks. The extent is contained within the | |
369 | * actual (current) free extent fbno for flen blocks. | |
370 | * Flags are passed in indicating whether the cursors are set to the | |
371 | * relevant records. | |
372 | */ | |
373 | STATIC int /* error code */ | |
374 | xfs_alloc_fixup_trees( | |
dfc130f3 RC |
375 | xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */ |
376 | xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */ | |
2bd0ea18 NS |
377 | xfs_agblock_t fbno, /* starting block of free extent */ |
378 | xfs_extlen_t flen, /* length of free extent */ | |
379 | xfs_agblock_t rbno, /* starting block of returned extent */ | |
380 | xfs_extlen_t rlen, /* length of returned extent */ | |
381 | int flags) /* flags, XFSA_FIXUP_... */ | |
382 | { | |
383 | int error; /* error code */ | |
384 | int i; /* operation results */ | |
385 | xfs_agblock_t nfbno1; /* first new free startblock */ | |
386 | xfs_agblock_t nfbno2; /* second new free startblock */ | |
0e266570 NS |
387 | xfs_extlen_t nflen1=0; /* first new free length */ |
388 | xfs_extlen_t nflen2=0; /* second new free length */ | |
19ebedcf DC |
389 | struct xfs_mount *mp; |
390 | ||
391 | mp = cnt_cur->bc_mp; | |
2bd0ea18 NS |
392 | |
393 | /* | |
394 | * Look up the record in the by-size tree if necessary. | |
395 | */ | |
396 | if (flags & XFSA_FIXUP_CNT_OK) { | |
397 | #ifdef DEBUG | |
0e266570 | 398 | if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i))) |
2bd0ea18 | 399 | return error; |
19ebedcf | 400 | XFS_WANT_CORRUPTED_RETURN(mp, |
2bd0ea18 NS |
401 | i == 1 && nfbno1 == fbno && nflen1 == flen); |
402 | #endif | |
403 | } else { | |
0e266570 | 404 | if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i))) |
2bd0ea18 | 405 | return error; |
19ebedcf | 406 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
2bd0ea18 NS |
407 | } |
408 | /* | |
409 | * Look up the record in the by-block tree if necessary. | |
410 | */ | |
411 | if (flags & XFSA_FIXUP_BNO_OK) { | |
412 | #ifdef DEBUG | |
0e266570 | 413 | if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i))) |
2bd0ea18 | 414 | return error; |
19ebedcf | 415 | XFS_WANT_CORRUPTED_RETURN(mp, |
2bd0ea18 NS |
416 | i == 1 && nfbno1 == fbno && nflen1 == flen); |
417 | #endif | |
418 | } else { | |
0e266570 | 419 | if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i))) |
2bd0ea18 | 420 | return error; |
19ebedcf | 421 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
2bd0ea18 | 422 | } |
b3563c19 | 423 | |
2bd0ea18 | 424 | #ifdef DEBUG |
b3563c19 BN |
425 | if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) { |
426 | struct xfs_btree_block *bnoblock; | |
427 | struct xfs_btree_block *cntblock; | |
428 | ||
429 | bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]); | |
430 | cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]); | |
2bd0ea18 | 431 | |
19ebedcf | 432 | XFS_WANT_CORRUPTED_RETURN(mp, |
b3563c19 | 433 | bnoblock->bb_numrecs == cntblock->bb_numrecs); |
2bd0ea18 NS |
434 | } |
435 | #endif | |
b3563c19 | 436 | |
2bd0ea18 NS |
437 | /* |
438 | * Deal with all four cases: the allocated record is contained | |
439 | * within the freespace record, so we can have new freespace | |
440 | * at either (or both) end, or no freespace remaining. | |
441 | */ | |
442 | if (rbno == fbno && rlen == flen) | |
443 | nfbno1 = nfbno2 = NULLAGBLOCK; | |
444 | else if (rbno == fbno) { | |
445 | nfbno1 = rbno + rlen; | |
446 | nflen1 = flen - rlen; | |
447 | nfbno2 = NULLAGBLOCK; | |
448 | } else if (rbno + rlen == fbno + flen) { | |
449 | nfbno1 = fbno; | |
450 | nflen1 = flen - rlen; | |
451 | nfbno2 = NULLAGBLOCK; | |
452 | } else { | |
453 | nfbno1 = fbno; | |
454 | nflen1 = rbno - fbno; | |
455 | nfbno2 = rbno + rlen; | |
456 | nflen2 = (fbno + flen) - nfbno2; | |
457 | } | |
458 | /* | |
459 | * Delete the entry from the by-size btree. | |
460 | */ | |
b194c7d8 | 461 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
2bd0ea18 | 462 | return error; |
19ebedcf | 463 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
2bd0ea18 NS |
464 | /* |
465 | * Add new by-size btree entry(s). | |
466 | */ | |
467 | if (nfbno1 != NULLAGBLOCK) { | |
0e266570 | 468 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i))) |
2bd0ea18 | 469 | return error; |
19ebedcf | 470 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
b194c7d8 | 471 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
2bd0ea18 | 472 | return error; |
19ebedcf | 473 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
2bd0ea18 NS |
474 | } |
475 | if (nfbno2 != NULLAGBLOCK) { | |
0e266570 | 476 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i))) |
2bd0ea18 | 477 | return error; |
19ebedcf | 478 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
b194c7d8 | 479 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
2bd0ea18 | 480 | return error; |
19ebedcf | 481 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
2bd0ea18 NS |
482 | } |
483 | /* | |
484 | * Fix up the by-block btree entry(s). | |
485 | */ | |
486 | if (nfbno1 == NULLAGBLOCK) { | |
487 | /* | |
488 | * No remaining freespace, just delete the by-block tree entry. | |
489 | */ | |
b194c7d8 | 490 | if ((error = xfs_btree_delete(bno_cur, &i))) |
2bd0ea18 | 491 | return error; |
19ebedcf | 492 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
2bd0ea18 NS |
493 | } else { |
494 | /* | |
495 | * Update the by-block entry to start later|be shorter. | |
496 | */ | |
0e266570 | 497 | if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1))) |
2bd0ea18 NS |
498 | return error; |
499 | } | |
500 | if (nfbno2 != NULLAGBLOCK) { | |
501 | /* | |
502 | * 2 resulting free entries, need to add one. | |
503 | */ | |
0e266570 | 504 | if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i))) |
2bd0ea18 | 505 | return error; |
19ebedcf | 506 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
b194c7d8 | 507 | if ((error = xfs_btree_insert(bno_cur, &i))) |
2bd0ea18 | 508 | return error; |
19ebedcf | 509 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
2bd0ea18 NS |
510 | } |
511 | return 0; | |
512 | } | |
513 | ||
dd5b876e | 514 | static bool |
a2ceac1f DC |
515 | xfs_agfl_verify( |
516 | struct xfs_buf *bp) | |
517 | { | |
a2ceac1f DC |
518 | struct xfs_mount *mp = bp->b_target->bt_mount; |
519 | struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp); | |
a2ceac1f DC |
520 | int i; |
521 | ||
9c4e12fb | 522 | if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid)) |
dd5b876e DC |
523 | return false; |
524 | if (be32_to_cpu(agfl->agfl_magicnum) != XFS_AGFL_MAGIC) | |
525 | return false; | |
526 | /* | |
527 | * during growfs operations, the perag is not fully initialised, | |
528 | * so we can't use it for any useful checking. growfs ensures we can't | |
529 | * use it by using uncached buffers that don't have the perag attached | |
530 | * so we can detect and avoid this problem. | |
531 | */ | |
532 | if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno) | |
533 | return false; | |
534 | ||
a2ceac1f | 535 | for (i = 0; i < XFS_AGFL_SIZE(mp); i++) { |
dd5b876e | 536 | if (be32_to_cpu(agfl->agfl_bno[i]) != NULLAGBLOCK && |
a2ceac1f | 537 | be32_to_cpu(agfl->agfl_bno[i]) >= mp->m_sb.sb_agblocks) |
dd5b876e | 538 | return false; |
a2ceac1f | 539 | } |
a65d8d29 BF |
540 | |
541 | return xfs_log_check_lsn(mp, | |
542 | be64_to_cpu(XFS_BUF_TO_AGFL(bp)->agfl_lsn)); | |
dd5b876e DC |
543 | } |
544 | ||
545 | static void | |
546 | xfs_agfl_read_verify( | |
547 | struct xfs_buf *bp) | |
548 | { | |
549 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
dd5b876e DC |
550 | |
551 | /* | |
552 | * There is no verification of non-crc AGFLs because mkfs does not | |
553 | * initialise the AGFL to zero or NULL. Hence the only valid part of the | |
554 | * AGFL is what the AGF says is active. We can't get to the AGF, so we | |
555 | * can't verify just those entries are valid. | |
556 | */ | |
557 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
558 | return; | |
559 | ||
45922933 | 560 | if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF)) |
12b53197 | 561 | xfs_buf_ioerror(bp, -EFSBADCRC); |
45922933 | 562 | else if (!xfs_agfl_verify(bp)) |
12b53197 | 563 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 DC |
564 | |
565 | if (bp->b_error) | |
566 | xfs_verifier_error(bp); | |
a2ceac1f DC |
567 | } |
568 | ||
569 | static void | |
570 | xfs_agfl_write_verify( | |
571 | struct xfs_buf *bp) | |
572 | { | |
dd5b876e DC |
573 | struct xfs_mount *mp = bp->b_target->bt_mount; |
574 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
a2ceac1f | 575 | |
dd5b876e DC |
576 | /* no verification of non-crc AGFLs */ |
577 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
578 | return; | |
579 | ||
580 | if (!xfs_agfl_verify(bp)) { | |
12b53197 | 581 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 | 582 | xfs_verifier_error(bp); |
dd5b876e DC |
583 | return; |
584 | } | |
585 | ||
586 | if (bip) | |
587 | XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
588 | ||
43b5aeed | 589 | xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF); |
a2ceac1f DC |
590 | } |
591 | ||
592 | const struct xfs_buf_ops xfs_agfl_buf_ops = { | |
a3fac935 | 593 | .name = "xfs_agfl", |
a2ceac1f DC |
594 | .verify_read = xfs_agfl_read_verify, |
595 | .verify_write = xfs_agfl_write_verify, | |
596 | }; | |
597 | ||
2bd0ea18 NS |
598 | /* |
599 | * Read in the allocation group free block array. | |
600 | */ | |
601 | STATIC int /* error */ | |
602 | xfs_alloc_read_agfl( | |
603 | xfs_mount_t *mp, /* mount point structure */ | |
604 | xfs_trans_t *tp, /* transaction pointer */ | |
605 | xfs_agnumber_t agno, /* allocation group number */ | |
606 | xfs_buf_t **bpp) /* buffer for the ag free block array */ | |
607 | { | |
608 | xfs_buf_t *bp; /* return value */ | |
2bd0ea18 NS |
609 | int error; |
610 | ||
611 | ASSERT(agno != NULLAGNUMBER); | |
9440d84d NS |
612 | error = xfs_trans_read_buf( |
613 | mp, tp, mp->m_ddev_targp, | |
614 | XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)), | |
a2ceac1f | 615 | XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops); |
9440d84d | 616 | if (error) |
2bd0ea18 | 617 | return error; |
a2ceac1f | 618 | xfs_buf_set_ref(bp, XFS_AGFL_REF); |
2bd0ea18 NS |
619 | *bpp = bp; |
620 | return 0; | |
621 | } | |
622 | ||
a2ceac1f DC |
623 | STATIC int |
624 | xfs_alloc_update_counters( | |
625 | struct xfs_trans *tp, | |
626 | struct xfs_perag *pag, | |
627 | struct xfs_buf *agbp, | |
628 | long len) | |
629 | { | |
630 | struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); | |
631 | ||
632 | pag->pagf_freeblks += len; | |
633 | be32_add_cpu(&agf->agf_freeblks, len); | |
634 | ||
635 | xfs_trans_agblocks_delta(tp, len); | |
636 | if (unlikely(be32_to_cpu(agf->agf_freeblks) > | |
637 | be32_to_cpu(agf->agf_length))) | |
12b53197 | 638 | return -EFSCORRUPTED; |
a2ceac1f DC |
639 | |
640 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS); | |
641 | return 0; | |
642 | } | |
643 | ||
2bd0ea18 NS |
644 | /* |
645 | * Allocation group level functions. | |
646 | */ | |
647 | ||
648 | /* | |
649 | * Allocate a variable extent in the allocation group agno. | |
650 | * Type and bno are used to determine where in the allocation group the | |
651 | * extent will start. | |
652 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
653 | * and of the form k * prod + mod unless there's nothing that large. | |
654 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
655 | */ | |
656 | STATIC int /* error */ | |
657 | xfs_alloc_ag_vextent( | |
dfc130f3 | 658 | xfs_alloc_arg_t *args) /* argument structure for allocation */ |
2bd0ea18 | 659 | { |
0e266570 | 660 | int error=0; |
2bd0ea18 NS |
661 | |
662 | ASSERT(args->minlen > 0); | |
663 | ASSERT(args->maxlen > 0); | |
664 | ASSERT(args->minlen <= args->maxlen); | |
665 | ASSERT(args->mod < args->prod); | |
666 | ASSERT(args->alignment > 0); | |
cf8ce220 | 667 | |
2bd0ea18 NS |
668 | /* |
669 | * Branch to correct routine based on the type. | |
670 | */ | |
671 | args->wasfromfl = 0; | |
672 | switch (args->type) { | |
673 | case XFS_ALLOCTYPE_THIS_AG: | |
674 | error = xfs_alloc_ag_vextent_size(args); | |
675 | break; | |
676 | case XFS_ALLOCTYPE_NEAR_BNO: | |
677 | error = xfs_alloc_ag_vextent_near(args); | |
678 | break; | |
679 | case XFS_ALLOCTYPE_THIS_BNO: | |
680 | error = xfs_alloc_ag_vextent_exact(args); | |
681 | break; | |
682 | default: | |
683 | ASSERT(0); | |
684 | /* NOTREACHED */ | |
685 | } | |
a2ceac1f DC |
686 | |
687 | if (error || args->agbno == NULLAGBLOCK) | |
2bd0ea18 | 688 | return error; |
2bd0ea18 | 689 | |
a2ceac1f DC |
690 | ASSERT(args->len >= args->minlen); |
691 | ASSERT(args->len <= args->maxlen); | |
cf8ce220 | 692 | ASSERT(!args->wasfromfl || args->resv != XFS_AG_RESV_AGFL); |
a2ceac1f DC |
693 | ASSERT(args->agbno % args->alignment == 0); |
694 | ||
631ac87a DW |
695 | /* if not file data, insert new block into the reverse map btree */ |
696 | if (args->oinfo.oi_owner != XFS_RMAP_OWN_UNKNOWN) { | |
697 | error = xfs_rmap_alloc(args->tp, args->agbp, args->agno, | |
698 | args->agbno, args->len, &args->oinfo); | |
699 | if (error) | |
700 | return error; | |
701 | } | |
702 | ||
a2ceac1f DC |
703 | if (!args->wasfromfl) { |
704 | error = xfs_alloc_update_counters(args->tp, args->pag, | |
705 | args->agbp, | |
706 | -((long)(args->len))); | |
707 | if (error) | |
708 | return error; | |
709 | ||
710 | ASSERT(!xfs_extent_busy_search(args->mp, args->agno, | |
711 | args->agbno, args->len)); | |
2bd0ea18 | 712 | } |
a2ceac1f | 713 | |
cf8ce220 | 714 | xfs_ag_resv_alloc_extent(args->pag, args->resv, args); |
a2ceac1f | 715 | |
79896434 BD |
716 | XFS_STATS_INC(args->mp, xs_allocx); |
717 | XFS_STATS_ADD(args->mp, xs_allocb, args->len); | |
a2ceac1f | 718 | return error; |
2bd0ea18 NS |
719 | } |
720 | ||
721 | /* | |
722 | * Allocate a variable extent at exactly agno/bno. | |
723 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
724 | * and of the form k * prod + mod unless there's nothing that large. | |
725 | * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it. | |
726 | */ | |
727 | STATIC int /* error */ | |
728 | xfs_alloc_ag_vextent_exact( | |
dfc130f3 | 729 | xfs_alloc_arg_t *args) /* allocation argument structure */ |
2bd0ea18 | 730 | { |
dfc130f3 RC |
731 | xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */ |
732 | xfs_btree_cur_t *cnt_cur;/* by count btree cursor */ | |
2bd0ea18 NS |
733 | int error; |
734 | xfs_agblock_t fbno; /* start block of found extent */ | |
2bd0ea18 | 735 | xfs_extlen_t flen; /* length of found extent */ |
a2ceac1f DC |
736 | xfs_agblock_t tbno; /* start block of trimmed extent */ |
737 | xfs_extlen_t tlen; /* length of trimmed extent */ | |
738 | xfs_agblock_t tend; /* end block of trimmed extent */ | |
2bd0ea18 | 739 | int i; /* success/failure of operation */ |
2bd0ea18 NS |
740 | |
741 | ASSERT(args->alignment == 1); | |
a2ceac1f | 742 | |
2bd0ea18 NS |
743 | /* |
744 | * Allocate/initialize a cursor for the by-number freespace btree. | |
745 | */ | |
b194c7d8 | 746 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
56b2de80 DC |
747 | args->agno, XFS_BTNUM_BNO); |
748 | ||
2bd0ea18 NS |
749 | /* |
750 | * Lookup bno and minlen in the btree (minlen is irrelevant, really). | |
751 | * Look for the closest free block <= bno, it must contain bno | |
752 | * if any free block does. | |
753 | */ | |
56b2de80 DC |
754 | error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i); |
755 | if (error) | |
2bd0ea18 | 756 | goto error0; |
56b2de80 DC |
757 | if (!i) |
758 | goto not_found; | |
759 | ||
2bd0ea18 NS |
760 | /* |
761 | * Grab the freespace record. | |
762 | */ | |
56b2de80 DC |
763 | error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i); |
764 | if (error) | |
2bd0ea18 | 765 | goto error0; |
19ebedcf | 766 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
2bd0ea18 | 767 | ASSERT(fbno <= args->agbno); |
56b2de80 | 768 | |
5000d01d | 769 | /* |
a2ceac1f DC |
770 | * Check for overlapping busy extents. |
771 | */ | |
772 | xfs_extent_busy_trim(args, fbno, flen, &tbno, &tlen); | |
773 | ||
774 | /* | |
775 | * Give up if the start of the extent is busy, or the freespace isn't | |
776 | * long enough for the minimum request. | |
2bd0ea18 | 777 | */ |
a2ceac1f DC |
778 | if (tbno > args->agbno) |
779 | goto not_found; | |
780 | if (tlen < args->minlen) | |
781 | goto not_found; | |
782 | tend = tbno + tlen; | |
783 | if (tend < args->agbno + args->minlen) | |
56b2de80 DC |
784 | goto not_found; |
785 | ||
2bd0ea18 NS |
786 | /* |
787 | * End of extent will be smaller of the freespace end and the | |
788 | * maximal requested end. | |
56b2de80 | 789 | * |
2bd0ea18 NS |
790 | * Fix the length according to mod and prod if given. |
791 | */ | |
a2ceac1f DC |
792 | args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen) |
793 | - args->agbno; | |
2bd0ea18 | 794 | xfs_alloc_fix_len(args); |
a2ceac1f | 795 | ASSERT(args->agbno + args->len <= tend); |
56b2de80 | 796 | |
2bd0ea18 | 797 | /* |
a2ceac1f | 798 | * We are allocating agbno for args->len |
2bd0ea18 NS |
799 | * Allocate/initialize a cursor for the by-size btree. |
800 | */ | |
b194c7d8 BN |
801 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
802 | args->agno, XFS_BTNUM_CNT); | |
2bd0ea18 | 803 | ASSERT(args->agbno + args->len <= |
6e3140c7 | 804 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
56b2de80 DC |
805 | error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno, |
806 | args->len, XFSA_FIXUP_BNO_OK); | |
807 | if (error) { | |
2bd0ea18 NS |
808 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); |
809 | goto error0; | |
810 | } | |
a2ceac1f | 811 | |
2bd0ea18 NS |
812 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); |
813 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
a2ceac1f | 814 | |
2bd0ea18 | 815 | args->wasfromfl = 0; |
56b2de80 DC |
816 | trace_xfs_alloc_exact_done(args); |
817 | return 0; | |
818 | ||
819 | not_found: | |
820 | /* Didn't find it, return null. */ | |
821 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
822 | args->agbno = NULLAGBLOCK; | |
823 | trace_xfs_alloc_exact_notfound(args); | |
2bd0ea18 NS |
824 | return 0; |
825 | ||
826 | error0: | |
827 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
56b2de80 DC |
828 | trace_xfs_alloc_exact_error(args); |
829 | return error; | |
830 | } | |
831 | ||
832 | /* | |
833 | * Search the btree in a given direction via the search cursor and compare | |
834 | * the records found against the good extent we've already found. | |
835 | */ | |
836 | STATIC int | |
837 | xfs_alloc_find_best_extent( | |
838 | struct xfs_alloc_arg *args, /* allocation argument structure */ | |
839 | struct xfs_btree_cur **gcur, /* good cursor */ | |
840 | struct xfs_btree_cur **scur, /* searching cursor */ | |
841 | xfs_agblock_t gdiff, /* difference for search comparison */ | |
842 | xfs_agblock_t *sbno, /* extent found by search */ | |
a2ceac1f DC |
843 | xfs_extlen_t *slen, /* extent length */ |
844 | xfs_agblock_t *sbnoa, /* aligned extent found by search */ | |
845 | xfs_extlen_t *slena, /* aligned extent length */ | |
56b2de80 DC |
846 | int dir) /* 0 = search right, 1 = search left */ |
847 | { | |
56b2de80 DC |
848 | xfs_agblock_t new; |
849 | xfs_agblock_t sdiff; | |
850 | int error; | |
851 | int i; | |
852 | ||
853 | /* The good extent is perfect, no need to search. */ | |
854 | if (!gdiff) | |
855 | goto out_use_good; | |
856 | ||
857 | /* | |
858 | * Look until we find a better one, run out of space or run off the end. | |
859 | */ | |
860 | do { | |
861 | error = xfs_alloc_get_rec(*scur, sbno, slen, &i); | |
862 | if (error) | |
863 | goto error0; | |
19ebedcf | 864 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
a2ceac1f | 865 | xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena); |
56b2de80 DC |
866 | |
867 | /* | |
868 | * The good extent is closer than this one. | |
869 | */ | |
870 | if (!dir) { | |
ff3263dd BF |
871 | if (*sbnoa > args->max_agbno) |
872 | goto out_use_good; | |
a2ceac1f | 873 | if (*sbnoa >= args->agbno + gdiff) |
56b2de80 DC |
874 | goto out_use_good; |
875 | } else { | |
ff3263dd BF |
876 | if (*sbnoa < args->min_agbno) |
877 | goto out_use_good; | |
a2ceac1f | 878 | if (*sbnoa <= args->agbno - gdiff) |
56b2de80 DC |
879 | goto out_use_good; |
880 | } | |
881 | ||
882 | /* | |
883 | * Same distance, compare length and pick the best. | |
884 | */ | |
885 | if (*slena >= args->minlen) { | |
886 | args->len = XFS_EXTLEN_MIN(*slena, args->maxlen); | |
887 | xfs_alloc_fix_len(args); | |
888 | ||
889 | sdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
84a62eea | 890 | args->alignment, |
1fccd5c8 | 891 | args->datatype, *sbnoa, |
a2ceac1f | 892 | *slena, &new); |
56b2de80 DC |
893 | |
894 | /* | |
895 | * Choose closer size and invalidate other cursor. | |
896 | */ | |
897 | if (sdiff < gdiff) | |
898 | goto out_use_search; | |
899 | goto out_use_good; | |
900 | } | |
901 | ||
902 | if (!dir) | |
903 | error = xfs_btree_increment(*scur, 0, &i); | |
904 | else | |
905 | error = xfs_btree_decrement(*scur, 0, &i); | |
906 | if (error) | |
907 | goto error0; | |
908 | } while (i); | |
909 | ||
910 | out_use_good: | |
911 | xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR); | |
912 | *scur = NULL; | |
913 | return 0; | |
914 | ||
915 | out_use_search: | |
916 | xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR); | |
917 | *gcur = NULL; | |
918 | return 0; | |
919 | ||
920 | error0: | |
921 | /* caller invalidates cursors */ | |
2bd0ea18 NS |
922 | return error; |
923 | } | |
924 | ||
925 | /* | |
926 | * Allocate a variable extent near bno in the allocation group agno. | |
927 | * Extent's length (returned in len) will be between minlen and maxlen, | |
928 | * and of the form k * prod + mod unless there's nothing that large. | |
929 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
930 | */ | |
931 | STATIC int /* error */ | |
932 | xfs_alloc_ag_vextent_near( | |
dfc130f3 | 933 | xfs_alloc_arg_t *args) /* allocation argument structure */ |
2bd0ea18 | 934 | { |
dfc130f3 RC |
935 | xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */ |
936 | xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */ | |
937 | xfs_btree_cur_t *cnt_cur; /* cursor for count btree */ | |
2bd0ea18 NS |
938 | xfs_agblock_t gtbno; /* start bno of right side entry */ |
939 | xfs_agblock_t gtbnoa; /* aligned ... */ | |
940 | xfs_extlen_t gtdiff; /* difference to right side entry */ | |
941 | xfs_extlen_t gtlen; /* length of right side entry */ | |
a2ceac1f | 942 | xfs_extlen_t gtlena; /* aligned ... */ |
2bd0ea18 NS |
943 | xfs_agblock_t gtnew; /* useful start bno of right side */ |
944 | int error; /* error code */ | |
945 | int i; /* result code, temporary */ | |
946 | int j; /* result code, temporary */ | |
947 | xfs_agblock_t ltbno; /* start bno of left side entry */ | |
948 | xfs_agblock_t ltbnoa; /* aligned ... */ | |
949 | xfs_extlen_t ltdiff; /* difference to left side entry */ | |
2bd0ea18 | 950 | xfs_extlen_t ltlen; /* length of left side entry */ |
a2ceac1f | 951 | xfs_extlen_t ltlena; /* aligned ... */ |
2bd0ea18 NS |
952 | xfs_agblock_t ltnew; /* useful start bno of left side */ |
953 | xfs_extlen_t rlen; /* length of returned extent */ | |
a2ceac1f | 954 | int forced = 0; |
6beba453 | 955 | #ifdef DEBUG |
2bd0ea18 NS |
956 | /* |
957 | * Randomly don't execute the first algorithm. | |
958 | */ | |
2bd0ea18 | 959 | int dofirst; /* set to do first algorithm */ |
2bd0ea18 | 960 | |
49f693fa | 961 | dofirst = prandom_u32() & 1; |
2bd0ea18 | 962 | #endif |
a2ceac1f | 963 | |
ff3263dd BF |
964 | /* handle unitialized agbno range so caller doesn't have to */ |
965 | if (!args->min_agbno && !args->max_agbno) | |
966 | args->max_agbno = args->mp->m_sb.sb_agblocks - 1; | |
967 | ASSERT(args->min_agbno <= args->max_agbno); | |
968 | ||
969 | /* clamp agbno to the range if it's outside */ | |
970 | if (args->agbno < args->min_agbno) | |
971 | args->agbno = args->min_agbno; | |
972 | if (args->agbno > args->max_agbno) | |
973 | args->agbno = args->max_agbno; | |
974 | ||
a2ceac1f DC |
975 | restart: |
976 | bno_cur_lt = NULL; | |
977 | bno_cur_gt = NULL; | |
978 | ltlen = 0; | |
979 | gtlena = 0; | |
980 | ltlena = 0; | |
981 | ||
2bd0ea18 NS |
982 | /* |
983 | * Get a cursor for the by-size btree. | |
984 | */ | |
b194c7d8 BN |
985 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
986 | args->agno, XFS_BTNUM_CNT); | |
a2ceac1f | 987 | |
2bd0ea18 NS |
988 | /* |
989 | * See if there are any free extents as big as maxlen. | |
990 | */ | |
0e266570 | 991 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i))) |
2bd0ea18 NS |
992 | goto error0; |
993 | /* | |
994 | * If none, then pick up the last entry in the tree unless the | |
995 | * tree is empty. | |
5000d01d | 996 | */ |
2bd0ea18 | 997 | if (!i) { |
0e266570 NS |
998 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno, |
999 | <len, &i))) | |
2bd0ea18 NS |
1000 | goto error0; |
1001 | if (i == 0 || ltlen == 0) { | |
1002 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
a2ceac1f | 1003 | trace_xfs_alloc_near_noentry(args); |
2bd0ea18 NS |
1004 | return 0; |
1005 | } | |
1006 | ASSERT(i == 1); | |
1007 | } | |
1008 | args->wasfromfl = 0; | |
a2ceac1f | 1009 | |
5000d01d | 1010 | /* |
2bd0ea18 NS |
1011 | * First algorithm. |
1012 | * If the requested extent is large wrt the freespaces available | |
1013 | * in this a.g., then the cursor will be pointing to a btree entry | |
1014 | * near the right edge of the tree. If it's in the last btree leaf | |
1015 | * block, then we just examine all the entries in that block | |
1016 | * that are big enough, and pick the best one. | |
1017 | * This is written as a while loop so we can break out of it, | |
1018 | * but we never loop back to the top. | |
1019 | */ | |
1020 | while (xfs_btree_islastblock(cnt_cur, 0)) { | |
1021 | xfs_extlen_t bdiff; | |
0e266570 NS |
1022 | int besti=0; |
1023 | xfs_extlen_t blen=0; | |
1024 | xfs_agblock_t bnew=0; | |
2bd0ea18 | 1025 | |
6beba453 DC |
1026 | #ifdef DEBUG |
1027 | if (dofirst) | |
2bd0ea18 NS |
1028 | break; |
1029 | #endif | |
1030 | /* | |
1031 | * Start from the entry that lookup found, sequence through | |
1032 | * all larger free blocks. If we're actually pointing at a | |
1033 | * record smaller than maxlen, go to the start of this block, | |
1034 | * and skip all those smaller than minlen. | |
1035 | */ | |
1036 | if (ltlen || args->alignment > 1) { | |
1037 | cnt_cur->bc_ptrs[0] = 1; | |
1038 | do { | |
0e266570 NS |
1039 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, |
1040 | <len, &i))) | |
2bd0ea18 | 1041 | goto error0; |
19ebedcf | 1042 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
2bd0ea18 NS |
1043 | if (ltlen >= args->minlen) |
1044 | break; | |
b194c7d8 | 1045 | if ((error = xfs_btree_increment(cnt_cur, 0, &i))) |
2bd0ea18 NS |
1046 | goto error0; |
1047 | } while (i); | |
1048 | ASSERT(ltlen >= args->minlen); | |
1049 | if (!i) | |
1050 | break; | |
1051 | } | |
1052 | i = cnt_cur->bc_ptrs[0]; | |
1053 | for (j = 1, blen = 0, bdiff = 0; | |
1054 | !error && j && (blen < args->maxlen || bdiff > 0); | |
b194c7d8 | 1055 | error = xfs_btree_increment(cnt_cur, 0, &j)) { |
2bd0ea18 NS |
1056 | /* |
1057 | * For each entry, decide if it's better than | |
1058 | * the previous best entry. | |
1059 | */ | |
0e266570 | 1060 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) |
2bd0ea18 | 1061 | goto error0; |
19ebedcf | 1062 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
a2ceac1f DC |
1063 | xfs_alloc_compute_aligned(args, ltbno, ltlen, |
1064 | <bnoa, <lena); | |
5e656dbb | 1065 | if (ltlena < args->minlen) |
2bd0ea18 | 1066 | continue; |
ff3263dd BF |
1067 | if (ltbnoa < args->min_agbno || ltbnoa > args->max_agbno) |
1068 | continue; | |
2bd0ea18 NS |
1069 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); |
1070 | xfs_alloc_fix_len(args); | |
1071 | ASSERT(args->len >= args->minlen); | |
1072 | if (args->len < blen) | |
1073 | continue; | |
1074 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
1fccd5c8 | 1075 | args->alignment, args->datatype, ltbnoa, |
84a62eea | 1076 | ltlena, <new); |
2bd0ea18 NS |
1077 | if (ltnew != NULLAGBLOCK && |
1078 | (args->len > blen || ltdiff < bdiff)) { | |
1079 | bdiff = ltdiff; | |
1080 | bnew = ltnew; | |
1081 | blen = args->len; | |
1082 | besti = cnt_cur->bc_ptrs[0]; | |
1083 | } | |
1084 | } | |
1085 | /* | |
1086 | * It didn't work. We COULD be in a case where | |
1087 | * there's a good record somewhere, so try again. | |
1088 | */ | |
1089 | if (blen == 0) | |
1090 | break; | |
1091 | /* | |
1092 | * Point at the best entry, and retrieve it again. | |
1093 | */ | |
1094 | cnt_cur->bc_ptrs[0] = besti; | |
0e266570 | 1095 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) |
2bd0ea18 | 1096 | goto error0; |
19ebedcf | 1097 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
56b2de80 | 1098 | ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
2bd0ea18 | 1099 | args->len = blen; |
2c003dc2 | 1100 | |
2bd0ea18 NS |
1101 | /* |
1102 | * We are allocating starting at bnew for blen blocks. | |
1103 | */ | |
1104 | args->agbno = bnew; | |
1105 | ASSERT(bnew >= ltbno); | |
56b2de80 | 1106 | ASSERT(bnew + blen <= ltbno + ltlen); |
2bd0ea18 NS |
1107 | /* |
1108 | * Set up a cursor for the by-bno tree. | |
1109 | */ | |
b194c7d8 BN |
1110 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, |
1111 | args->agbp, args->agno, XFS_BTNUM_BNO); | |
2bd0ea18 NS |
1112 | /* |
1113 | * Fix up the btree entries. | |
1114 | */ | |
0e266570 NS |
1115 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, |
1116 | ltlen, bnew, blen, XFSA_FIXUP_CNT_OK))) | |
2bd0ea18 NS |
1117 | goto error0; |
1118 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1119 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
56b2de80 DC |
1120 | |
1121 | trace_xfs_alloc_near_first(args); | |
2bd0ea18 NS |
1122 | return 0; |
1123 | } | |
1124 | /* | |
1125 | * Second algorithm. | |
1126 | * Search in the by-bno tree to the left and to the right | |
1127 | * simultaneously, until in each case we find a space big enough, | |
1128 | * or run into the edge of the tree. When we run into the edge, | |
1129 | * we deallocate that cursor. | |
1130 | * If both searches succeed, we compare the two spaces and pick | |
1131 | * the better one. | |
1132 | * With alignment, it's possible for both to fail; the upper | |
1133 | * level algorithm that picks allocation groups for allocations | |
1134 | * is not supposed to do this. | |
1135 | */ | |
1136 | /* | |
1137 | * Allocate and initialize the cursor for the leftward search. | |
1138 | */ | |
b194c7d8 BN |
1139 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1140 | args->agno, XFS_BTNUM_BNO); | |
2bd0ea18 NS |
1141 | /* |
1142 | * Lookup <= bno to find the leftward search's starting point. | |
1143 | */ | |
0e266570 | 1144 | if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i))) |
2bd0ea18 NS |
1145 | goto error0; |
1146 | if (!i) { | |
1147 | /* | |
1148 | * Didn't find anything; use this cursor for the rightward | |
1149 | * search. | |
1150 | */ | |
1151 | bno_cur_gt = bno_cur_lt; | |
062998e3 | 1152 | bno_cur_lt = NULL; |
2bd0ea18 NS |
1153 | } |
1154 | /* | |
1155 | * Found something. Duplicate the cursor for the rightward search. | |
1156 | */ | |
0e266570 | 1157 | else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt))) |
2bd0ea18 NS |
1158 | goto error0; |
1159 | /* | |
1160 | * Increment the cursor, so we will point at the entry just right | |
1161 | * of the leftward entry if any, or to the leftmost entry. | |
1162 | */ | |
b194c7d8 | 1163 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
2bd0ea18 NS |
1164 | goto error0; |
1165 | if (!i) { | |
1166 | /* | |
1167 | * It failed, there are no rightward entries. | |
1168 | */ | |
1169 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR); | |
1170 | bno_cur_gt = NULL; | |
1171 | } | |
1172 | /* | |
1173 | * Loop going left with the leftward cursor, right with the | |
1174 | * rightward cursor, until either both directions give up or | |
1175 | * we find an entry at least as big as minlen. | |
1176 | */ | |
1177 | do { | |
1178 | if (bno_cur_lt) { | |
0e266570 | 1179 | if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i))) |
2bd0ea18 | 1180 | goto error0; |
19ebedcf | 1181 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
a2ceac1f DC |
1182 | xfs_alloc_compute_aligned(args, ltbno, ltlen, |
1183 | <bnoa, <lena); | |
ff3263dd | 1184 | if (ltlena >= args->minlen && ltbnoa >= args->min_agbno) |
2bd0ea18 | 1185 | break; |
b194c7d8 | 1186 | if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i))) |
2bd0ea18 | 1187 | goto error0; |
ff3263dd | 1188 | if (!i || ltbnoa < args->min_agbno) { |
2bd0ea18 NS |
1189 | xfs_btree_del_cursor(bno_cur_lt, |
1190 | XFS_BTREE_NOERROR); | |
1191 | bno_cur_lt = NULL; | |
1192 | } | |
1193 | } | |
1194 | if (bno_cur_gt) { | |
0e266570 | 1195 | if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i))) |
2bd0ea18 | 1196 | goto error0; |
19ebedcf | 1197 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
a2ceac1f DC |
1198 | xfs_alloc_compute_aligned(args, gtbno, gtlen, |
1199 | >bnoa, >lena); | |
ff3263dd | 1200 | if (gtlena >= args->minlen && gtbnoa <= args->max_agbno) |
2bd0ea18 | 1201 | break; |
b194c7d8 | 1202 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
2bd0ea18 | 1203 | goto error0; |
ff3263dd | 1204 | if (!i || gtbnoa > args->max_agbno) { |
2bd0ea18 NS |
1205 | xfs_btree_del_cursor(bno_cur_gt, |
1206 | XFS_BTREE_NOERROR); | |
1207 | bno_cur_gt = NULL; | |
1208 | } | |
1209 | } | |
1210 | } while (bno_cur_lt || bno_cur_gt); | |
56b2de80 | 1211 | |
2bd0ea18 NS |
1212 | /* |
1213 | * Got both cursors still active, need to find better entry. | |
1214 | */ | |
1215 | if (bno_cur_lt && bno_cur_gt) { | |
2bd0ea18 NS |
1216 | if (ltlena >= args->minlen) { |
1217 | /* | |
56b2de80 | 1218 | * Left side is good, look for a right side entry. |
2bd0ea18 NS |
1219 | */ |
1220 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1221 | xfs_alloc_fix_len(args); | |
56b2de80 | 1222 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, |
1fccd5c8 | 1223 | args->alignment, args->datatype, ltbnoa, |
84a62eea | 1224 | ltlena, <new); |
56b2de80 DC |
1225 | |
1226 | error = xfs_alloc_find_best_extent(args, | |
1227 | &bno_cur_lt, &bno_cur_gt, | |
a2ceac1f DC |
1228 | ltdiff, >bno, >len, |
1229 | >bnoa, >lena, | |
56b2de80 DC |
1230 | 0 /* search right */); |
1231 | } else { | |
1232 | ASSERT(gtlena >= args->minlen); | |
1233 | ||
2bd0ea18 | 1234 | /* |
56b2de80 | 1235 | * Right side is good, look for a left side entry. |
2bd0ea18 NS |
1236 | */ |
1237 | args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen); | |
1238 | xfs_alloc_fix_len(args); | |
56b2de80 | 1239 | gtdiff = xfs_alloc_compute_diff(args->agbno, args->len, |
1fccd5c8 | 1240 | args->alignment, args->datatype, gtbnoa, |
84a62eea | 1241 | gtlena, >new); |
56b2de80 DC |
1242 | |
1243 | error = xfs_alloc_find_best_extent(args, | |
1244 | &bno_cur_gt, &bno_cur_lt, | |
a2ceac1f DC |
1245 | gtdiff, <bno, <len, |
1246 | <bnoa, <lena, | |
56b2de80 | 1247 | 1 /* search left */); |
2bd0ea18 | 1248 | } |
56b2de80 DC |
1249 | |
1250 | if (error) | |
1251 | goto error0; | |
2bd0ea18 | 1252 | } |
56b2de80 | 1253 | |
2bd0ea18 NS |
1254 | /* |
1255 | * If we couldn't get anything, give up. | |
1256 | */ | |
1257 | if (bno_cur_lt == NULL && bno_cur_gt == NULL) { | |
a2ceac1f DC |
1258 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1259 | ||
1260 | if (!forced++) { | |
1261 | trace_xfs_alloc_near_busy(args); | |
1262 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1263 | goto restart; | |
1264 | } | |
56b2de80 | 1265 | trace_xfs_alloc_size_neither(args); |
2bd0ea18 NS |
1266 | args->agbno = NULLAGBLOCK; |
1267 | return 0; | |
1268 | } | |
56b2de80 | 1269 | |
2bd0ea18 NS |
1270 | /* |
1271 | * At this point we have selected a freespace entry, either to the | |
1272 | * left or to the right. If it's on the right, copy all the | |
1273 | * useful variables to the "left" set so we only have one | |
1274 | * copy of this code. | |
1275 | */ | |
1276 | if (bno_cur_gt) { | |
1277 | bno_cur_lt = bno_cur_gt; | |
1278 | bno_cur_gt = NULL; | |
1279 | ltbno = gtbno; | |
1280 | ltbnoa = gtbnoa; | |
1281 | ltlen = gtlen; | |
1282 | ltlena = gtlena; | |
1283 | j = 1; | |
1284 | } else | |
1285 | j = 0; | |
56b2de80 | 1286 | |
2bd0ea18 NS |
1287 | /* |
1288 | * Fix up the length and compute the useful address. | |
1289 | */ | |
2bd0ea18 NS |
1290 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); |
1291 | xfs_alloc_fix_len(args); | |
2bd0ea18 | 1292 | rlen = args->len; |
a2ceac1f | 1293 | (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, |
1fccd5c8 | 1294 | args->datatype, ltbnoa, ltlena, <new); |
2bd0ea18 | 1295 | ASSERT(ltnew >= ltbno); |
a2ceac1f | 1296 | ASSERT(ltnew + rlen <= ltbnoa + ltlena); |
6e3140c7 | 1297 | ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
ff3263dd | 1298 | ASSERT(ltnew >= args->min_agbno && ltnew <= args->max_agbno); |
2bd0ea18 | 1299 | args->agbno = ltnew; |
a2ceac1f | 1300 | |
0e266570 NS |
1301 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen, |
1302 | ltnew, rlen, XFSA_FIXUP_BNO_OK))) | |
2bd0ea18 | 1303 | goto error0; |
56b2de80 DC |
1304 | |
1305 | if (j) | |
1306 | trace_xfs_alloc_near_greater(args); | |
1307 | else | |
1308 | trace_xfs_alloc_near_lesser(args); | |
1309 | ||
2bd0ea18 NS |
1310 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1311 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1312 | return 0; | |
1313 | ||
1314 | error0: | |
56b2de80 | 1315 | trace_xfs_alloc_near_error(args); |
2bd0ea18 NS |
1316 | if (cnt_cur != NULL) |
1317 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1318 | if (bno_cur_lt != NULL) | |
1319 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR); | |
1320 | if (bno_cur_gt != NULL) | |
1321 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR); | |
1322 | return error; | |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * Allocate a variable extent anywhere in the allocation group agno. | |
1327 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1328 | * and of the form k * prod + mod unless there's nothing that large. | |
1329 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1330 | */ | |
1331 | STATIC int /* error */ | |
1332 | xfs_alloc_ag_vextent_size( | |
dfc130f3 | 1333 | xfs_alloc_arg_t *args) /* allocation argument structure */ |
2bd0ea18 | 1334 | { |
dfc130f3 RC |
1335 | xfs_btree_cur_t *bno_cur; /* cursor for bno btree */ |
1336 | xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */ | |
2bd0ea18 NS |
1337 | int error; /* error result */ |
1338 | xfs_agblock_t fbno; /* start of found freespace */ | |
1339 | xfs_extlen_t flen; /* length of found freespace */ | |
2bd0ea18 NS |
1340 | int i; /* temp status variable */ |
1341 | xfs_agblock_t rbno; /* returned block number */ | |
1342 | xfs_extlen_t rlen; /* length of returned extent */ | |
a2ceac1f | 1343 | int forced = 0; |
2bd0ea18 | 1344 | |
a2ceac1f | 1345 | restart: |
2bd0ea18 NS |
1346 | /* |
1347 | * Allocate and initialize a cursor for the by-size btree. | |
1348 | */ | |
b194c7d8 BN |
1349 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1350 | args->agno, XFS_BTNUM_CNT); | |
2bd0ea18 | 1351 | bno_cur = NULL; |
a2ceac1f | 1352 | |
2bd0ea18 NS |
1353 | /* |
1354 | * Look for an entry >= maxlen+alignment-1 blocks. | |
1355 | */ | |
0e266570 NS |
1356 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, |
1357 | args->maxlen + args->alignment - 1, &i))) | |
2bd0ea18 | 1358 | goto error0; |
a2ceac1f | 1359 | |
2bd0ea18 | 1360 | /* |
a2ceac1f DC |
1361 | * If none or we have busy extents that we cannot allocate from, then |
1362 | * we have to settle for a smaller extent. In the case that there are | |
1363 | * no large extents, this will return the last entry in the tree unless | |
1364 | * the tree is empty. In the case that there are only busy large | |
1365 | * extents, this will return the largest small extent unless there | |
1366 | * are no smaller extents available. | |
5000d01d | 1367 | */ |
a2ceac1f DC |
1368 | if (!i || forced > 1) { |
1369 | error = xfs_alloc_ag_vextent_small(args, cnt_cur, | |
1370 | &fbno, &flen, &i); | |
1371 | if (error) | |
2bd0ea18 NS |
1372 | goto error0; |
1373 | if (i == 0 || flen == 0) { | |
1374 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
56b2de80 | 1375 | trace_xfs_alloc_size_noentry(args); |
2bd0ea18 NS |
1376 | return 0; |
1377 | } | |
1378 | ASSERT(i == 1); | |
a2ceac1f DC |
1379 | xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen); |
1380 | } else { | |
1381 | /* | |
1382 | * Search for a non-busy extent that is large enough. | |
1383 | * If we are at low space, don't check, or if we fall of | |
1384 | * the end of the btree, turn off the busy check and | |
1385 | * restart. | |
1386 | */ | |
1387 | for (;;) { | |
1388 | error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i); | |
1389 | if (error) | |
1390 | goto error0; | |
19ebedcf | 1391 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
a2ceac1f DC |
1392 | |
1393 | xfs_alloc_compute_aligned(args, fbno, flen, | |
1394 | &rbno, &rlen); | |
1395 | ||
1396 | if (rlen >= args->maxlen) | |
1397 | break; | |
1398 | ||
1399 | error = xfs_btree_increment(cnt_cur, 0, &i); | |
1400 | if (error) | |
1401 | goto error0; | |
1402 | if (i == 0) { | |
1403 | /* | |
1404 | * Our only valid extents must have been busy. | |
1405 | * Make it unbusy by forcing the log out and | |
1406 | * retrying. If we've been here before, forcing | |
1407 | * the log isn't making the extents available, | |
1408 | * which means they have probably been freed in | |
1409 | * this transaction. In that case, we have to | |
1410 | * give up on them and we'll attempt a minlen | |
1411 | * allocation the next time around. | |
1412 | */ | |
1413 | xfs_btree_del_cursor(cnt_cur, | |
1414 | XFS_BTREE_NOERROR); | |
1415 | trace_xfs_alloc_size_busy(args); | |
1416 | if (!forced++) | |
1417 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1418 | goto restart; | |
1419 | } | |
1420 | } | |
2bd0ea18 | 1421 | } |
a2ceac1f | 1422 | |
2bd0ea18 NS |
1423 | /* |
1424 | * In the first case above, we got the last entry in the | |
1425 | * by-size btree. Now we check to see if the space hits maxlen | |
1426 | * once aligned; if not, we search left for something better. | |
1427 | * This can't happen in the second case above. | |
1428 | */ | |
2bd0ea18 | 1429 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
19ebedcf | 1430 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen == 0 || |
2bd0ea18 NS |
1431 | (rlen <= flen && rbno + rlen <= fbno + flen), error0); |
1432 | if (rlen < args->maxlen) { | |
1433 | xfs_agblock_t bestfbno; | |
1434 | xfs_extlen_t bestflen; | |
1435 | xfs_agblock_t bestrbno; | |
1436 | xfs_extlen_t bestrlen; | |
1437 | ||
1438 | bestrlen = rlen; | |
1439 | bestrbno = rbno; | |
1440 | bestflen = flen; | |
1441 | bestfbno = fbno; | |
1442 | for (;;) { | |
b194c7d8 | 1443 | if ((error = xfs_btree_decrement(cnt_cur, 0, &i))) |
2bd0ea18 NS |
1444 | goto error0; |
1445 | if (i == 0) | |
1446 | break; | |
0e266570 NS |
1447 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, |
1448 | &i))) | |
2bd0ea18 | 1449 | goto error0; |
19ebedcf | 1450 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
2bd0ea18 NS |
1451 | if (flen < bestrlen) |
1452 | break; | |
a2ceac1f DC |
1453 | xfs_alloc_compute_aligned(args, fbno, flen, |
1454 | &rbno, &rlen); | |
2bd0ea18 | 1455 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
19ebedcf | 1456 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen == 0 || |
2bd0ea18 NS |
1457 | (rlen <= flen && rbno + rlen <= fbno + flen), |
1458 | error0); | |
1459 | if (rlen > bestrlen) { | |
1460 | bestrlen = rlen; | |
1461 | bestrbno = rbno; | |
1462 | bestflen = flen; | |
1463 | bestfbno = fbno; | |
1464 | if (rlen == args->maxlen) | |
1465 | break; | |
1466 | } | |
5000d01d | 1467 | } |
0e266570 NS |
1468 | if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen, |
1469 | &i))) | |
2bd0ea18 | 1470 | goto error0; |
19ebedcf | 1471 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
2bd0ea18 NS |
1472 | rlen = bestrlen; |
1473 | rbno = bestrbno; | |
1474 | flen = bestflen; | |
1475 | fbno = bestfbno; | |
1476 | } | |
1477 | args->wasfromfl = 0; | |
1478 | /* | |
1479 | * Fix up the length. | |
1480 | */ | |
1481 | args->len = rlen; | |
a2ceac1f DC |
1482 | if (rlen < args->minlen) { |
1483 | if (!forced++) { | |
1484 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1485 | trace_xfs_alloc_size_busy(args); | |
1486 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1487 | goto restart; | |
1488 | } | |
1489 | goto out_nominleft; | |
2bd0ea18 | 1490 | } |
a2ceac1f DC |
1491 | xfs_alloc_fix_len(args); |
1492 | ||
2bd0ea18 | 1493 | rlen = args->len; |
19ebedcf | 1494 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen <= flen, error0); |
2bd0ea18 NS |
1495 | /* |
1496 | * Allocate and initialize a cursor for the by-block tree. | |
1497 | */ | |
b194c7d8 BN |
1498 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1499 | args->agno, XFS_BTNUM_BNO); | |
0e266570 NS |
1500 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
1501 | rbno, rlen, XFSA_FIXUP_CNT_OK))) | |
2bd0ea18 NS |
1502 | goto error0; |
1503 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1504 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1505 | cnt_cur = bno_cur = NULL; | |
1506 | args->len = rlen; | |
1507 | args->agbno = rbno; | |
19ebedcf | 1508 | XFS_WANT_CORRUPTED_GOTO(args->mp, |
2bd0ea18 | 1509 | args->agbno + args->len <= |
6e3140c7 | 1510 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
2bd0ea18 | 1511 | error0); |
56b2de80 | 1512 | trace_xfs_alloc_size_done(args); |
2bd0ea18 NS |
1513 | return 0; |
1514 | ||
1515 | error0: | |
56b2de80 | 1516 | trace_xfs_alloc_size_error(args); |
2bd0ea18 NS |
1517 | if (cnt_cur) |
1518 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1519 | if (bno_cur) | |
1520 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1521 | return error; | |
a2ceac1f DC |
1522 | |
1523 | out_nominleft: | |
1524 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1525 | trace_xfs_alloc_size_nominleft(args); | |
1526 | args->agbno = NULLAGBLOCK; | |
1527 | return 0; | |
2bd0ea18 NS |
1528 | } |
1529 | ||
1530 | /* | |
1531 | * Deal with the case where only small freespaces remain. | |
1532 | * Either return the contents of the last freespace record, | |
1533 | * or allocate space from the freelist if there is nothing in the tree. | |
1534 | */ | |
1535 | STATIC int /* error */ | |
1536 | xfs_alloc_ag_vextent_small( | |
dfc130f3 RC |
1537 | xfs_alloc_arg_t *args, /* allocation argument structure */ |
1538 | xfs_btree_cur_t *ccur, /* by-size cursor */ | |
1539 | xfs_agblock_t *fbnop, /* result block number */ | |
1540 | xfs_extlen_t *flenp, /* result length */ | |
2bd0ea18 NS |
1541 | int *stat) /* status: 0-freelist, 1-normal/none */ |
1542 | { | |
59b86360 DW |
1543 | struct xfs_owner_info oinfo; |
1544 | struct xfs_perag *pag; | |
2bd0ea18 NS |
1545 | int error; |
1546 | xfs_agblock_t fbno; | |
1547 | xfs_extlen_t flen; | |
2bd0ea18 NS |
1548 | int i; |
1549 | ||
b194c7d8 | 1550 | if ((error = xfs_btree_decrement(ccur, 0, &i))) |
2bd0ea18 NS |
1551 | goto error0; |
1552 | if (i) { | |
0e266570 | 1553 | if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i))) |
2bd0ea18 | 1554 | goto error0; |
19ebedcf | 1555 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
2bd0ea18 NS |
1556 | } |
1557 | /* | |
1558 | * Nothing in the btree, try the freelist. Make sure | |
1559 | * to respect minleft even when pulling from the | |
1560 | * freelist. | |
1561 | */ | |
cf8ce220 DW |
1562 | else if (args->minlen == 1 && args->alignment == 1 && |
1563 | args->resv != XFS_AG_RESV_AGFL && | |
6e3140c7 NS |
1564 | (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount) |
1565 | > args->minleft)) { | |
cdded3d8 DC |
1566 | error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0); |
1567 | if (error) | |
2bd0ea18 NS |
1568 | goto error0; |
1569 | if (fbno != NULLAGBLOCK) { | |
a2ceac1f | 1570 | xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1, |
1fccd5c8 | 1571 | xfs_alloc_allow_busy_reuse(args->datatype)); |
a2ceac1f | 1572 | |
1fccd5c8 | 1573 | if (xfs_alloc_is_userdata(args->datatype)) { |
2bd0ea18 NS |
1574 | xfs_buf_t *bp; |
1575 | ||
1576 | bp = xfs_btree_get_bufs(args->mp, args->tp, | |
1577 | args->agno, fbno, 0); | |
1578 | xfs_trans_binval(args->tp, bp); | |
2bd0ea18 NS |
1579 | } |
1580 | args->len = 1; | |
1581 | args->agbno = fbno; | |
19ebedcf | 1582 | XFS_WANT_CORRUPTED_GOTO(args->mp, |
2bd0ea18 | 1583 | args->agbno + args->len <= |
6e3140c7 | 1584 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
2bd0ea18 NS |
1585 | error0); |
1586 | args->wasfromfl = 1; | |
56b2de80 | 1587 | trace_xfs_alloc_small_freelist(args); |
59b86360 DW |
1588 | |
1589 | /* | |
1590 | * If we're feeding an AGFL block to something that | |
1591 | * doesn't live in the free space, we need to clear | |
cf8ce220 DW |
1592 | * out the OWN_AG rmap and add the block back to |
1593 | * the AGFL per-AG reservation. | |
59b86360 DW |
1594 | */ |
1595 | xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_AG); | |
1596 | error = xfs_rmap_free(args->tp, args->agbp, args->agno, | |
1597 | fbno, 1, &oinfo); | |
1598 | if (error) | |
1599 | goto error0; | |
1600 | pag = xfs_perag_get(args->mp, args->agno); | |
cf8ce220 DW |
1601 | xfs_ag_resv_free_extent(pag, XFS_AG_RESV_AGFL, |
1602 | args->tp, 1); | |
59b86360 DW |
1603 | xfs_perag_put(pag); |
1604 | ||
2bd0ea18 NS |
1605 | *stat = 0; |
1606 | return 0; | |
1607 | } | |
1608 | /* | |
1609 | * Nothing in the freelist. | |
1610 | */ | |
1611 | else | |
1612 | flen = 0; | |
1613 | } | |
1614 | /* | |
1615 | * Can't allocate from the freelist for some reason. | |
1616 | */ | |
5e656dbb BN |
1617 | else { |
1618 | fbno = NULLAGBLOCK; | |
2bd0ea18 | 1619 | flen = 0; |
5e656dbb | 1620 | } |
2bd0ea18 NS |
1621 | /* |
1622 | * Can't do the allocation, give up. | |
1623 | */ | |
1624 | if (flen < args->minlen) { | |
1625 | args->agbno = NULLAGBLOCK; | |
56b2de80 | 1626 | trace_xfs_alloc_small_notenough(args); |
2bd0ea18 NS |
1627 | flen = 0; |
1628 | } | |
1629 | *fbnop = fbno; | |
1630 | *flenp = flen; | |
1631 | *stat = 1; | |
56b2de80 | 1632 | trace_xfs_alloc_small_done(args); |
2bd0ea18 NS |
1633 | return 0; |
1634 | ||
1635 | error0: | |
56b2de80 | 1636 | trace_xfs_alloc_small_error(args); |
2bd0ea18 NS |
1637 | return error; |
1638 | } | |
1639 | ||
1640 | /* | |
1641 | * Free the extent starting at agno/bno for length. | |
1642 | */ | |
85aec44f | 1643 | STATIC int |
2bd0ea18 | 1644 | xfs_free_ag_extent( |
85aec44f DW |
1645 | xfs_trans_t *tp, |
1646 | xfs_buf_t *agbp, | |
1647 | xfs_agnumber_t agno, | |
1648 | xfs_agblock_t bno, | |
1649 | xfs_extlen_t len, | |
1650 | struct xfs_owner_info *oinfo, | |
cf8ce220 | 1651 | enum xfs_ag_resv_type type) |
2bd0ea18 | 1652 | { |
dfc130f3 RC |
1653 | xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */ |
1654 | xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */ | |
2bd0ea18 | 1655 | int error; /* error return value */ |
2bd0ea18 NS |
1656 | xfs_agblock_t gtbno; /* start of right neighbor block */ |
1657 | xfs_extlen_t gtlen; /* length of right neighbor block */ | |
1658 | int haveleft; /* have a left neighbor block */ | |
1659 | int haveright; /* have a right neighbor block */ | |
1660 | int i; /* temp, result code */ | |
1661 | xfs_agblock_t ltbno; /* start of left neighbor block */ | |
1662 | xfs_extlen_t ltlen; /* length of left neighbor block */ | |
1663 | xfs_mount_t *mp; /* mount point struct for filesystem */ | |
1664 | xfs_agblock_t nbno; /* new starting block of freespace */ | |
1665 | xfs_extlen_t nlen; /* new length of freespace */ | |
a2ceac1f | 1666 | xfs_perag_t *pag; /* per allocation group data */ |
2bd0ea18 | 1667 | |
631ac87a | 1668 | bno_cur = cnt_cur = NULL; |
2bd0ea18 | 1669 | mp = tp->t_mountp; |
631ac87a DW |
1670 | |
1671 | if (oinfo->oi_owner != XFS_RMAP_OWN_UNKNOWN) { | |
1672 | error = xfs_rmap_free(tp, agbp, agno, bno, len, oinfo); | |
1673 | if (error) | |
1674 | goto error0; | |
1675 | } | |
1676 | ||
5000d01d | 1677 | /* |
2bd0ea18 NS |
1678 | * Allocate and initialize a cursor for the by-block btree. |
1679 | */ | |
b194c7d8 | 1680 | bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO); |
5000d01d | 1681 | /* |
2bd0ea18 NS |
1682 | * Look for a neighboring block on the left (lower block numbers) |
1683 | * that is contiguous with this space. | |
1684 | */ | |
0e266570 | 1685 | if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft))) |
2bd0ea18 NS |
1686 | goto error0; |
1687 | if (haveleft) { | |
1688 | /* | |
1689 | * There is a block to our left. | |
1690 | */ | |
0e266570 | 1691 | if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i))) |
2bd0ea18 | 1692 | goto error0; |
19ebedcf | 1693 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1694 | /* |
1695 | * It's not contiguous, though. | |
1696 | */ | |
1697 | if (ltbno + ltlen < bno) | |
1698 | haveleft = 0; | |
1699 | else { | |
1700 | /* | |
1701 | * If this failure happens the request to free this | |
1702 | * space was invalid, it's (partly) already free. | |
1703 | * Very bad. | |
1704 | */ | |
19ebedcf DC |
1705 | XFS_WANT_CORRUPTED_GOTO(mp, |
1706 | ltbno + ltlen <= bno, error0); | |
2bd0ea18 NS |
1707 | } |
1708 | } | |
5000d01d | 1709 | /* |
2bd0ea18 NS |
1710 | * Look for a neighboring block on the right (higher block numbers) |
1711 | * that is contiguous with this space. | |
1712 | */ | |
b194c7d8 | 1713 | if ((error = xfs_btree_increment(bno_cur, 0, &haveright))) |
2bd0ea18 NS |
1714 | goto error0; |
1715 | if (haveright) { | |
1716 | /* | |
1717 | * There is a block to our right. | |
1718 | */ | |
0e266570 | 1719 | if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i))) |
2bd0ea18 | 1720 | goto error0; |
19ebedcf | 1721 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1722 | /* |
1723 | * It's not contiguous, though. | |
1724 | */ | |
1725 | if (bno + len < gtbno) | |
1726 | haveright = 0; | |
1727 | else { | |
1728 | /* | |
1729 | * If this failure happens the request to free this | |
1730 | * space was invalid, it's (partly) already free. | |
1731 | * Very bad. | |
1732 | */ | |
19ebedcf | 1733 | XFS_WANT_CORRUPTED_GOTO(mp, gtbno >= bno + len, error0); |
2bd0ea18 NS |
1734 | } |
1735 | } | |
1736 | /* | |
1737 | * Now allocate and initialize a cursor for the by-size tree. | |
1738 | */ | |
b194c7d8 | 1739 | cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT); |
2bd0ea18 NS |
1740 | /* |
1741 | * Have both left and right contiguous neighbors. | |
1742 | * Merge all three into a single free block. | |
1743 | */ | |
1744 | if (haveleft && haveright) { | |
1745 | /* | |
1746 | * Delete the old by-size entry on the left. | |
1747 | */ | |
0e266570 | 1748 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) |
2bd0ea18 | 1749 | goto error0; |
19ebedcf | 1750 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
b194c7d8 | 1751 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
2bd0ea18 | 1752 | goto error0; |
19ebedcf | 1753 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1754 | /* |
1755 | * Delete the old by-size entry on the right. | |
1756 | */ | |
0e266570 | 1757 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) |
2bd0ea18 | 1758 | goto error0; |
19ebedcf | 1759 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
b194c7d8 | 1760 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
2bd0ea18 | 1761 | goto error0; |
19ebedcf | 1762 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1763 | /* |
1764 | * Delete the old by-block entry for the right block. | |
1765 | */ | |
b194c7d8 | 1766 | if ((error = xfs_btree_delete(bno_cur, &i))) |
2bd0ea18 | 1767 | goto error0; |
19ebedcf | 1768 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1769 | /* |
1770 | * Move the by-block cursor back to the left neighbor. | |
1771 | */ | |
b194c7d8 | 1772 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
2bd0ea18 | 1773 | goto error0; |
19ebedcf | 1774 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1775 | #ifdef DEBUG |
1776 | /* | |
1777 | * Check that this is the right record: delete didn't | |
1778 | * mangle the cursor. | |
1779 | */ | |
1780 | { | |
1781 | xfs_agblock_t xxbno; | |
1782 | xfs_extlen_t xxlen; | |
1783 | ||
0e266570 NS |
1784 | if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen, |
1785 | &i))) | |
2bd0ea18 | 1786 | goto error0; |
19ebedcf | 1787 | XFS_WANT_CORRUPTED_GOTO(mp, |
2bd0ea18 NS |
1788 | i == 1 && xxbno == ltbno && xxlen == ltlen, |
1789 | error0); | |
1790 | } | |
1791 | #endif | |
1792 | /* | |
1793 | * Update remaining by-block entry to the new, joined block. | |
1794 | */ | |
1795 | nbno = ltbno; | |
1796 | nlen = len + ltlen + gtlen; | |
0e266570 | 1797 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) |
2bd0ea18 NS |
1798 | goto error0; |
1799 | } | |
1800 | /* | |
1801 | * Have only a left contiguous neighbor. | |
1802 | * Merge it together with the new freespace. | |
1803 | */ | |
1804 | else if (haveleft) { | |
1805 | /* | |
1806 | * Delete the old by-size entry on the left. | |
1807 | */ | |
0e266570 | 1808 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) |
2bd0ea18 | 1809 | goto error0; |
19ebedcf | 1810 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
b194c7d8 | 1811 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
2bd0ea18 | 1812 | goto error0; |
19ebedcf | 1813 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1814 | /* |
1815 | * Back up the by-block cursor to the left neighbor, and | |
1816 | * update its length. | |
1817 | */ | |
b194c7d8 | 1818 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
2bd0ea18 | 1819 | goto error0; |
19ebedcf | 1820 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1821 | nbno = ltbno; |
1822 | nlen = len + ltlen; | |
0e266570 | 1823 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) |
2bd0ea18 NS |
1824 | goto error0; |
1825 | } | |
1826 | /* | |
1827 | * Have only a right contiguous neighbor. | |
1828 | * Merge it together with the new freespace. | |
1829 | */ | |
1830 | else if (haveright) { | |
1831 | /* | |
1832 | * Delete the old by-size entry on the right. | |
1833 | */ | |
0e266570 | 1834 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) |
2bd0ea18 | 1835 | goto error0; |
19ebedcf | 1836 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
b194c7d8 | 1837 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
2bd0ea18 | 1838 | goto error0; |
19ebedcf | 1839 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 | 1840 | /* |
5000d01d | 1841 | * Update the starting block and length of the right |
2bd0ea18 NS |
1842 | * neighbor in the by-block tree. |
1843 | */ | |
1844 | nbno = bno; | |
1845 | nlen = len + gtlen; | |
0e266570 | 1846 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) |
2bd0ea18 NS |
1847 | goto error0; |
1848 | } | |
1849 | /* | |
1850 | * No contiguous neighbors. | |
1851 | * Insert the new freespace into the by-block tree. | |
1852 | */ | |
1853 | else { | |
1854 | nbno = bno; | |
1855 | nlen = len; | |
b194c7d8 | 1856 | if ((error = xfs_btree_insert(bno_cur, &i))) |
2bd0ea18 | 1857 | goto error0; |
19ebedcf | 1858 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1859 | } |
1860 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1861 | bno_cur = NULL; | |
1862 | /* | |
1863 | * In all cases we need to insert the new freespace in the by-size tree. | |
1864 | */ | |
0e266570 | 1865 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i))) |
2bd0ea18 | 1866 | goto error0; |
19ebedcf | 1867 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, error0); |
b194c7d8 | 1868 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
2bd0ea18 | 1869 | goto error0; |
19ebedcf | 1870 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
2bd0ea18 NS |
1871 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1872 | cnt_cur = NULL; | |
a2ceac1f | 1873 | |
2bd0ea18 NS |
1874 | /* |
1875 | * Update the freespace totals in the ag and superblock. | |
1876 | */ | |
a2ceac1f DC |
1877 | pag = xfs_perag_get(mp, agno); |
1878 | error = xfs_alloc_update_counters(tp, pag, agbp, len); | |
cf8ce220 | 1879 | xfs_ag_resv_free_extent(pag, type, tp, len); |
a2ceac1f DC |
1880 | xfs_perag_put(pag); |
1881 | if (error) | |
1882 | goto error0; | |
1883 | ||
79896434 BD |
1884 | XFS_STATS_INC(mp, xs_freex); |
1885 | XFS_STATS_ADD(mp, xs_freeb, len); | |
56b2de80 | 1886 | |
cf8ce220 DW |
1887 | trace_xfs_free_extent(mp, agno, bno, len, type == XFS_AG_RESV_AGFL, |
1888 | haveleft, haveright); | |
3e535bba | 1889 | |
2bd0ea18 NS |
1890 | return 0; |
1891 | ||
1892 | error0: | |
cf8ce220 DW |
1893 | trace_xfs_free_extent(mp, agno, bno, len, type == XFS_AG_RESV_AGFL, |
1894 | -1, -1); | |
2bd0ea18 NS |
1895 | if (bno_cur) |
1896 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1897 | if (cnt_cur) | |
1898 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1899 | return error; | |
1900 | } | |
1901 | ||
5000d01d | 1902 | /* |
2bd0ea18 NS |
1903 | * Visible (exported) allocation/free functions. |
1904 | * Some of these are used just by xfs_alloc_btree.c and this file. | |
1905 | */ | |
1906 | ||
1907 | /* | |
1908 | * Compute and fill in value of m_ag_maxlevels. | |
1909 | */ | |
1910 | void | |
1911 | xfs_alloc_compute_maxlevels( | |
1912 | xfs_mount_t *mp) /* file system mount structure */ | |
1913 | { | |
730e2a19 DW |
1914 | mp->m_ag_maxlevels = xfs_btree_compute_maxlevels(mp, mp->m_alloc_mnr, |
1915 | (mp->m_sb.sb_agblocks + 1) / 2); | |
2bd0ea18 NS |
1916 | } |
1917 | ||
56b2de80 | 1918 | /* |
cf8ce220 DW |
1919 | * Find the length of the longest extent in an AG. The 'need' parameter |
1920 | * specifies how much space we're going to need for the AGFL and the | |
1921 | * 'reserved' parameter tells us how many blocks in this AG are reserved for | |
1922 | * other callers. | |
56b2de80 DC |
1923 | */ |
1924 | xfs_extlen_t | |
1925 | xfs_alloc_longest_free_extent( | |
1926 | struct xfs_mount *mp, | |
72bda06d | 1927 | struct xfs_perag *pag, |
cf8ce220 DW |
1928 | xfs_extlen_t need, |
1929 | xfs_extlen_t reserved) | |
56b2de80 | 1930 | { |
72bda06d | 1931 | xfs_extlen_t delta = 0; |
56b2de80 | 1932 | |
cf8ce220 DW |
1933 | /* |
1934 | * If the AGFL needs a recharge, we'll have to subtract that from the | |
1935 | * longest extent. | |
1936 | */ | |
56b2de80 DC |
1937 | if (need > pag->pagf_flcount) |
1938 | delta = need - pag->pagf_flcount; | |
1939 | ||
cf8ce220 DW |
1940 | /* |
1941 | * If we cannot maintain others' reservations with space from the | |
1942 | * not-longest freesp extents, we'll have to subtract /that/ from | |
1943 | * the longest extent too. | |
1944 | */ | |
1945 | if (pag->pagf_freeblks - pag->pagf_longest < reserved) | |
1946 | delta += reserved - (pag->pagf_freeblks - pag->pagf_longest); | |
1947 | ||
1948 | /* | |
1949 | * If the longest extent is long enough to satisfy all the | |
1950 | * reservations and AGFL rules in place, we can return this extent. | |
1951 | */ | |
56b2de80 DC |
1952 | if (pag->pagf_longest > delta) |
1953 | return pag->pagf_longest - delta; | |
cf8ce220 DW |
1954 | |
1955 | /* Otherwise, let the caller try for 1 block if there's space. */ | |
56b2de80 DC |
1956 | return pag->pagf_flcount > 0 || pag->pagf_longest > 0; |
1957 | } | |
1958 | ||
de046644 DC |
1959 | unsigned int |
1960 | xfs_alloc_min_freelist( | |
1961 | struct xfs_mount *mp, | |
1962 | struct xfs_perag *pag) | |
1963 | { | |
1964 | unsigned int min_free; | |
1965 | ||
1966 | /* space needed by-bno freespace btree */ | |
1967 | min_free = min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_BNOi] + 1, | |
1968 | mp->m_ag_maxlevels); | |
1969 | /* space needed by-size freespace btree */ | |
1970 | min_free += min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_CNTi] + 1, | |
1971 | mp->m_ag_maxlevels); | |
b8a8d6e5 DW |
1972 | /* space needed reverse mapping used space btree */ |
1973 | if (xfs_sb_version_hasrmapbt(&mp->m_sb)) | |
1974 | min_free += min_t(unsigned int, | |
1975 | pag->pagf_levels[XFS_BTNUM_RMAPi] + 1, | |
1976 | mp->m_rmap_maxlevels); | |
de046644 DC |
1977 | |
1978 | return min_free; | |
1979 | } | |
1980 | ||
5515b7c1 DC |
1981 | /* |
1982 | * Check if the operation we are fixing up the freelist for should go ahead or | |
1983 | * not. If we are freeing blocks, we always allow it, otherwise the allocation | |
1984 | * is dependent on whether the size and shape of free space available will | |
1985 | * permit the requested allocation to take place. | |
1986 | */ | |
1987 | static bool | |
1988 | xfs_alloc_space_available( | |
1989 | struct xfs_alloc_arg *args, | |
1990 | xfs_extlen_t min_free, | |
1991 | int flags) | |
1992 | { | |
1993 | struct xfs_perag *pag = args->pag; | |
3fe4a6dd | 1994 | xfs_extlen_t alloc_len, longest; |
cf8ce220 | 1995 | xfs_extlen_t reservation; /* blocks that are still reserved */ |
5515b7c1 DC |
1996 | int available; |
1997 | ||
1998 | if (flags & XFS_ALLOC_FLAG_FREEING) | |
1999 | return true; | |
2000 | ||
cf8ce220 DW |
2001 | reservation = xfs_ag_resv_needed(pag, args->resv); |
2002 | ||
5515b7c1 | 2003 | /* do we have enough contiguous free space for the allocation? */ |
3fe4a6dd | 2004 | alloc_len = args->minlen + (args->alignment - 1) + args->minalignslop; |
cf8ce220 DW |
2005 | longest = xfs_alloc_longest_free_extent(args->mp, pag, min_free, |
2006 | reservation); | |
3fe4a6dd | 2007 | if (longest < alloc_len) |
5515b7c1 DC |
2008 | return false; |
2009 | ||
cf8ce220 | 2010 | /* do we have enough free space remaining for the allocation? */ |
5515b7c1 | 2011 | available = (int)(pag->pagf_freeblks + pag->pagf_flcount - |
2c003dc2 | 2012 | reservation - min_free - args->minleft); |
3fe4a6dd | 2013 | if (available < (int)max(args->total, alloc_len)) |
5515b7c1 DC |
2014 | return false; |
2015 | ||
2c003dc2 CH |
2016 | /* |
2017 | * Clamp maxlen to the amount of free space available for the actual | |
2018 | * extent allocation. | |
2019 | */ | |
2020 | if (available < (int)args->maxlen && !(flags & XFS_ALLOC_FLAG_CHECK)) { | |
2021 | args->maxlen = available; | |
2022 | ASSERT(args->maxlen > 0); | |
2023 | ASSERT(args->maxlen >= args->minlen); | |
2024 | } | |
2025 | ||
5515b7c1 DC |
2026 | return true; |
2027 | } | |
2028 | ||
2bd0ea18 NS |
2029 | /* |
2030 | * Decide whether to use this allocation group for this allocation. | |
2031 | * If so, fix up the btree freelist's size. | |
2bd0ea18 | 2032 | */ |
ff105f75 | 2033 | int /* error */ |
2bd0ea18 | 2034 | xfs_alloc_fix_freelist( |
c98e644e DC |
2035 | struct xfs_alloc_arg *args, /* allocation argument structure */ |
2036 | int flags) /* XFS_ALLOC_FLAG_... */ | |
2bd0ea18 | 2037 | { |
c98e644e DC |
2038 | struct xfs_mount *mp = args->mp; |
2039 | struct xfs_perag *pag = args->pag; | |
2040 | struct xfs_trans *tp = args->tp; | |
2041 | struct xfs_buf *agbp = NULL; | |
2042 | struct xfs_buf *agflbp = NULL; | |
2043 | struct xfs_alloc_arg targs; /* local allocation arguments */ | |
2044 | xfs_agblock_t bno; /* freelist block */ | |
2045 | xfs_extlen_t need; /* total blocks needed in freelist */ | |
fcdd428c | 2046 | int error = 0; |
c98e644e | 2047 | |
2bd0ea18 | 2048 | if (!pag->pagf_init) { |
c98e644e DC |
2049 | error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp); |
2050 | if (error) | |
2051 | goto out_no_agbp; | |
2bd0ea18 | 2052 | if (!pag->pagf_init) { |
5e656dbb BN |
2053 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
2054 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
c98e644e | 2055 | goto out_agbp_relse; |
2bd0ea18 | 2056 | } |
c98e644e | 2057 | } |
34317449 | 2058 | |
5e656dbb | 2059 | /* |
c98e644e DC |
2060 | * If this is a metadata preferred pag and we are user data then try |
2061 | * somewhere else if we are not being asked to try harder at this | |
2062 | * point | |
34317449 | 2063 | */ |
1fccd5c8 | 2064 | if (pag->pagf_metadata && xfs_alloc_is_userdata(args->datatype) && |
5e656dbb BN |
2065 | (flags & XFS_ALLOC_FLAG_TRYLOCK)) { |
2066 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
c98e644e | 2067 | goto out_agbp_relse; |
34317449 NS |
2068 | } |
2069 | ||
de046644 | 2070 | need = xfs_alloc_min_freelist(mp, pag); |
2c003dc2 CH |
2071 | if (!xfs_alloc_space_available(args, need, flags | |
2072 | XFS_ALLOC_FLAG_CHECK)) | |
c98e644e | 2073 | goto out_agbp_relse; |
5e656dbb | 2074 | |
2bd0ea18 NS |
2075 | /* |
2076 | * Get the a.g. freespace buffer. | |
2077 | * Can fail if we're not blocking on locks, and it's held. | |
2078 | */ | |
c98e644e DC |
2079 | if (!agbp) { |
2080 | error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp); | |
2081 | if (error) | |
2082 | goto out_no_agbp; | |
2083 | if (!agbp) { | |
5e656dbb BN |
2084 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
2085 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
c98e644e | 2086 | goto out_no_agbp; |
2bd0ea18 NS |
2087 | } |
2088 | } | |
72bda06d | 2089 | |
72bda06d | 2090 | /* If there isn't enough total space or single-extent, reject it. */ |
de046644 | 2091 | need = xfs_alloc_min_freelist(mp, pag); |
c98e644e DC |
2092 | if (!xfs_alloc_space_available(args, need, flags)) |
2093 | goto out_agbp_relse; | |
5515b7c1 | 2094 | |
2bd0ea18 NS |
2095 | /* |
2096 | * Make the freelist shorter if it's too long. | |
72bda06d | 2097 | * |
c98e644e DC |
2098 | * Note that from this point onwards, we will always release the agf and |
2099 | * agfl buffers on error. This handles the case where we error out and | |
2100 | * the buffers are clean or may not have been joined to the transaction | |
2101 | * and hence need to be released manually. If they have been joined to | |
2102 | * the transaction, then xfs_trans_brelse() will handle them | |
2103 | * appropriately based on the recursion count and dirty state of the | |
2104 | * buffer. | |
2105 | * | |
72bda06d DC |
2106 | * XXX (dgc): When we have lots of free space, does this buy us |
2107 | * anything other than extra overhead when we need to put more blocks | |
2108 | * back on the free list? Maybe we should only do this when space is | |
2109 | * getting low or the AGFL is more than half full? | |
e365af6f DW |
2110 | * |
2111 | * The NOSHRINK flag prevents the AGFL from being shrunk if it's too | |
2112 | * big; the NORMAP flag prevents AGFL expand/shrink operations from | |
2113 | * updating the rmapbt. Both flags are used in xfs_repair while we're | |
2114 | * rebuilding the rmapbt, and neither are used by the kernel. They're | |
2115 | * both required to ensure that rmaps are correctly recorded for the | |
2116 | * regenerated AGFL, bnobt, and cntbt. See repair/phase5.c and | |
2117 | * repair/rmap.c in xfsprogs for details. | |
2bd0ea18 | 2118 | */ |
e365af6f DW |
2119 | memset(&targs, 0, sizeof(targs)); |
2120 | if (flags & XFS_ALLOC_FLAG_NORMAP) | |
2121 | xfs_rmap_skip_owner_update(&targs.oinfo); | |
2122 | else | |
2123 | xfs_rmap_ag_owner(&targs.oinfo, XFS_RMAP_OWN_AG); | |
2124 | while (!(flags & XFS_ALLOC_FLAG_NOSHRINK) && pag->pagf_flcount > need) { | |
72bda06d | 2125 | struct xfs_buf *bp; |
2bd0ea18 | 2126 | |
5e656dbb BN |
2127 | error = xfs_alloc_get_freelist(tp, agbp, &bno, 0); |
2128 | if (error) | |
c98e644e | 2129 | goto out_agbp_relse; |
85aec44f | 2130 | error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, |
cf8ce220 | 2131 | &targs.oinfo, XFS_AG_RESV_AGFL); |
72bda06d | 2132 | if (error) |
c98e644e | 2133 | goto out_agbp_relse; |
2bd0ea18 NS |
2134 | bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0); |
2135 | xfs_trans_binval(tp, bp); | |
2bd0ea18 | 2136 | } |
72bda06d | 2137 | |
2bd0ea18 NS |
2138 | targs.tp = tp; |
2139 | targs.mp = mp; | |
2140 | targs.agbp = agbp; | |
2141 | targs.agno = args->agno; | |
cf8ce220 | 2142 | targs.alignment = targs.minlen = targs.prod = 1; |
2bd0ea18 NS |
2143 | targs.type = XFS_ALLOCTYPE_THIS_AG; |
2144 | targs.pag = pag; | |
72bda06d DC |
2145 | error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp); |
2146 | if (error) | |
c98e644e | 2147 | goto out_agbp_relse; |
72bda06d DC |
2148 | |
2149 | /* Make the freelist longer if it's too short. */ | |
2150 | while (pag->pagf_flcount < need) { | |
2bd0ea18 | 2151 | targs.agbno = 0; |
72bda06d | 2152 | targs.maxlen = need - pag->pagf_flcount; |
cf8ce220 | 2153 | targs.resv = XFS_AG_RESV_AGFL; |
72bda06d DC |
2154 | |
2155 | /* Allocate as many blocks as possible at once. */ | |
2156 | error = xfs_alloc_ag_vextent(&targs); | |
c98e644e DC |
2157 | if (error) |
2158 | goto out_agflbp_relse; | |
2159 | ||
2bd0ea18 | 2160 | /* |
dfc130f3 RC |
2161 | * Stop if we run out. Won't happen if callers are obeying |
2162 | * the restrictions correctly. Can happen for free calls | |
2bd0ea18 NS |
2163 | * on a completely full ag. |
2164 | */ | |
5e656dbb BN |
2165 | if (targs.agbno == NULLAGBLOCK) { |
2166 | if (flags & XFS_ALLOC_FLAG_FREEING) | |
2167 | break; | |
c98e644e | 2168 | goto out_agflbp_relse; |
5e656dbb | 2169 | } |
2bd0ea18 NS |
2170 | /* |
2171 | * Put each allocated block on the list. | |
2172 | */ | |
2173 | for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) { | |
5e656dbb BN |
2174 | error = xfs_alloc_put_freelist(tp, agbp, |
2175 | agflbp, bno, 0); | |
2176 | if (error) | |
c98e644e | 2177 | goto out_agflbp_relse; |
2bd0ea18 NS |
2178 | } |
2179 | } | |
cb4deb22 | 2180 | xfs_trans_brelse(tp, agflbp); |
2bd0ea18 NS |
2181 | args->agbp = agbp; |
2182 | return 0; | |
c98e644e DC |
2183 | |
2184 | out_agflbp_relse: | |
2185 | xfs_trans_brelse(tp, agflbp); | |
2186 | out_agbp_relse: | |
2187 | if (agbp) | |
2188 | xfs_trans_brelse(tp, agbp); | |
2189 | out_no_agbp: | |
2190 | args->agbp = NULL; | |
2191 | return error; | |
2bd0ea18 NS |
2192 | } |
2193 | ||
2194 | /* | |
2195 | * Get a block from the freelist. | |
2196 | * Returns with the buffer for the block gotten. | |
2197 | */ | |
2198 | int /* error */ | |
2199 | xfs_alloc_get_freelist( | |
2200 | xfs_trans_t *tp, /* transaction pointer */ | |
2201 | xfs_buf_t *agbp, /* buffer containing the agf structure */ | |
cdded3d8 DC |
2202 | xfs_agblock_t *bnop, /* block address retrieved from freelist */ |
2203 | int btreeblk) /* destination is a AGF btree */ | |
2bd0ea18 NS |
2204 | { |
2205 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
2bd0ea18 NS |
2206 | xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */ |
2207 | xfs_agblock_t bno; /* block number returned */ | |
dd5b876e | 2208 | __be32 *agfl_bno; |
2bd0ea18 | 2209 | int error; |
cdded3d8 | 2210 | int logflags; |
dd5b876e | 2211 | xfs_mount_t *mp = tp->t_mountp; |
2bd0ea18 NS |
2212 | xfs_perag_t *pag; /* per allocation group data */ |
2213 | ||
2bd0ea18 NS |
2214 | /* |
2215 | * Freelist is empty, give up. | |
2216 | */ | |
dd5b876e | 2217 | agf = XFS_BUF_TO_AGF(agbp); |
46eca962 | 2218 | if (!agf->agf_flcount) { |
2bd0ea18 NS |
2219 | *bnop = NULLAGBLOCK; |
2220 | return 0; | |
2221 | } | |
2222 | /* | |
2223 | * Read the array of free blocks. | |
2224 | */ | |
dd5b876e DC |
2225 | error = xfs_alloc_read_agfl(mp, tp, be32_to_cpu(agf->agf_seqno), |
2226 | &agflbp); | |
2227 | if (error) | |
2bd0ea18 | 2228 | return error; |
dd5b876e DC |
2229 | |
2230 | ||
2bd0ea18 NS |
2231 | /* |
2232 | * Get the block number and update the data structures. | |
2233 | */ | |
dd5b876e DC |
2234 | agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp); |
2235 | bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]); | |
5e656dbb | 2236 | be32_add_cpu(&agf->agf_flfirst, 1); |
2bd0ea18 | 2237 | xfs_trans_brelse(tp, agflbp); |
6e3140c7 | 2238 | if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp)) |
46eca962 | 2239 | agf->agf_flfirst = 0; |
56b2de80 DC |
2240 | |
2241 | pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); | |
5e656dbb | 2242 | be32_add_cpu(&agf->agf_flcount, -1); |
2bd0ea18 NS |
2243 | xfs_trans_agflist_delta(tp, -1); |
2244 | pag->pagf_flcount--; | |
56b2de80 | 2245 | xfs_perag_put(pag); |
cdded3d8 DC |
2246 | |
2247 | logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; | |
2248 | if (btreeblk) { | |
5e656dbb | 2249 | be32_add_cpu(&agf->agf_btreeblks, 1); |
cdded3d8 DC |
2250 | pag->pagf_btreeblks++; |
2251 | logflags |= XFS_AGF_BTREEBLKS; | |
2252 | } | |
2253 | ||
cdded3d8 | 2254 | xfs_alloc_log_agf(tp, agbp, logflags); |
2bd0ea18 | 2255 | *bnop = bno; |
3e535bba | 2256 | |
2bd0ea18 NS |
2257 | return 0; |
2258 | } | |
2259 | ||
2260 | /* | |
2261 | * Log the given fields from the agf structure. | |
2262 | */ | |
2263 | void | |
2264 | xfs_alloc_log_agf( | |
2265 | xfs_trans_t *tp, /* transaction pointer */ | |
2266 | xfs_buf_t *bp, /* buffer for a.g. freelist header */ | |
dfc130f3 | 2267 | int fields) /* mask of fields to be logged (XFS_AGF_...) */ |
2bd0ea18 NS |
2268 | { |
2269 | int first; /* first byte offset */ | |
2270 | int last; /* last byte offset */ | |
2271 | static const short offsets[] = { | |
2272 | offsetof(xfs_agf_t, agf_magicnum), | |
2273 | offsetof(xfs_agf_t, agf_versionnum), | |
2274 | offsetof(xfs_agf_t, agf_seqno), | |
2275 | offsetof(xfs_agf_t, agf_length), | |
2276 | offsetof(xfs_agf_t, agf_roots[0]), | |
2277 | offsetof(xfs_agf_t, agf_levels[0]), | |
2278 | offsetof(xfs_agf_t, agf_flfirst), | |
2279 | offsetof(xfs_agf_t, agf_fllast), | |
2280 | offsetof(xfs_agf_t, agf_flcount), | |
2281 | offsetof(xfs_agf_t, agf_freeblks), | |
2282 | offsetof(xfs_agf_t, agf_longest), | |
cdded3d8 | 2283 | offsetof(xfs_agf_t, agf_btreeblks), |
dd5b876e | 2284 | offsetof(xfs_agf_t, agf_uuid), |
8511b71a | 2285 | offsetof(xfs_agf_t, agf_rmap_blocks), |
bc859611 DW |
2286 | offsetof(xfs_agf_t, agf_refcount_blocks), |
2287 | offsetof(xfs_agf_t, agf_refcount_root), | |
2288 | offsetof(xfs_agf_t, agf_refcount_level), | |
8511b71a DW |
2289 | /* needed so that we don't log the whole rest of the structure: */ |
2290 | offsetof(xfs_agf_t, agf_spare64), | |
2bd0ea18 NS |
2291 | sizeof(xfs_agf_t) |
2292 | }; | |
2293 | ||
56b2de80 DC |
2294 | trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_); |
2295 | ||
bdc16ee5 | 2296 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF); |
dd5b876e | 2297 | |
2bd0ea18 NS |
2298 | xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last); |
2299 | xfs_trans_log_buf(tp, bp, (uint)first, (uint)last); | |
2300 | } | |
2301 | ||
2302 | /* | |
2303 | * Interface for inode allocation to force the pag data to be initialized. | |
2304 | */ | |
2305 | int /* error */ | |
2306 | xfs_alloc_pagf_init( | |
2307 | xfs_mount_t *mp, /* file system mount structure */ | |
2308 | xfs_trans_t *tp, /* transaction pointer */ | |
2309 | xfs_agnumber_t agno, /* allocation group number */ | |
2310 | int flags) /* XFS_ALLOC_FLAGS_... */ | |
2311 | { | |
7a3bffe4 | 2312 | xfs_buf_t *bp; |
2bd0ea18 NS |
2313 | int error; |
2314 | ||
0e266570 | 2315 | if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp))) |
2bd0ea18 NS |
2316 | return error; |
2317 | if (bp) | |
2318 | xfs_trans_brelse(tp, bp); | |
2319 | return 0; | |
2320 | } | |
2321 | ||
2322 | /* | |
2323 | * Put the block on the freelist for the allocation group. | |
2324 | */ | |
2325 | int /* error */ | |
2326 | xfs_alloc_put_freelist( | |
2327 | xfs_trans_t *tp, /* transaction pointer */ | |
2328 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
2329 | xfs_buf_t *agflbp,/* buffer for a.g. free block array */ | |
cdded3d8 DC |
2330 | xfs_agblock_t bno, /* block being freed */ |
2331 | int btreeblk) /* block came from a AGF btree */ | |
2bd0ea18 NS |
2332 | { |
2333 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
5e656dbb | 2334 | __be32 *blockp;/* pointer to array entry */ |
2bd0ea18 | 2335 | int error; |
cdded3d8 | 2336 | int logflags; |
2bd0ea18 NS |
2337 | xfs_mount_t *mp; /* mount structure */ |
2338 | xfs_perag_t *pag; /* per allocation group data */ | |
dd5b876e DC |
2339 | __be32 *agfl_bno; |
2340 | int startoff; | |
2bd0ea18 NS |
2341 | |
2342 | agf = XFS_BUF_TO_AGF(agbp); | |
2343 | mp = tp->t_mountp; | |
2344 | ||
2345 | if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp, | |
6e3140c7 | 2346 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
2bd0ea18 | 2347 | return error; |
5e656dbb | 2348 | be32_add_cpu(&agf->agf_fllast, 1); |
6e3140c7 | 2349 | if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp)) |
46eca962 | 2350 | agf->agf_fllast = 0; |
56b2de80 DC |
2351 | |
2352 | pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); | |
5e656dbb | 2353 | be32_add_cpu(&agf->agf_flcount, 1); |
2bd0ea18 NS |
2354 | xfs_trans_agflist_delta(tp, 1); |
2355 | pag->pagf_flcount++; | |
cdded3d8 DC |
2356 | |
2357 | logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT; | |
2358 | if (btreeblk) { | |
5e656dbb | 2359 | be32_add_cpu(&agf->agf_btreeblks, -1); |
cdded3d8 DC |
2360 | pag->pagf_btreeblks--; |
2361 | logflags |= XFS_AGF_BTREEBLKS; | |
2362 | } | |
56b2de80 | 2363 | xfs_perag_put(pag); |
cdded3d8 | 2364 | |
5e656dbb BN |
2365 | xfs_alloc_log_agf(tp, agbp, logflags); |
2366 | ||
6e3140c7 | 2367 | ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)); |
dd5b876e DC |
2368 | |
2369 | agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp); | |
2370 | blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)]; | |
5e656dbb | 2371 | *blockp = cpu_to_be32(bno); |
dd5b876e DC |
2372 | startoff = (char *)blockp - (char *)agflbp->b_addr; |
2373 | ||
cdded3d8 | 2374 | xfs_alloc_log_agf(tp, agbp, logflags); |
dd5b876e | 2375 | |
bdc16ee5 | 2376 | xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF); |
dd5b876e DC |
2377 | xfs_trans_log_buf(tp, agflbp, startoff, |
2378 | startoff + sizeof(xfs_agblock_t) - 1); | |
2bd0ea18 NS |
2379 | return 0; |
2380 | } | |
2381 | ||
dd5b876e | 2382 | static bool |
a2ceac1f | 2383 | xfs_agf_verify( |
dd5b876e | 2384 | struct xfs_mount *mp, |
a2ceac1f DC |
2385 | struct xfs_buf *bp) |
2386 | { | |
dd5b876e | 2387 | struct xfs_agf *agf = XFS_BUF_TO_AGF(bp); |
a2ceac1f | 2388 | |
a65d8d29 BF |
2389 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2390 | if (!uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid)) | |
dd5b876e | 2391 | return false; |
a65d8d29 BF |
2392 | if (!xfs_log_check_lsn(mp, |
2393 | be64_to_cpu(XFS_BUF_TO_AGF(bp)->agf_lsn))) | |
2394 | return false; | |
2395 | } | |
a2ceac1f | 2396 | |
dd5b876e DC |
2397 | if (!(agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) && |
2398 | XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) && | |
2399 | be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) && | |
2400 | be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) && | |
2401 | be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) && | |
2402 | be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp))) | |
2403 | return false; | |
a2ceac1f | 2404 | |
00795aae DW |
2405 | if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) < 1 || |
2406 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) < 1 || | |
2407 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) > XFS_BTREE_MAXLEVELS || | |
5a35bf2c DC |
2408 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) > XFS_BTREE_MAXLEVELS) |
2409 | return false; | |
2410 | ||
e37838e5 | 2411 | if (xfs_sb_version_hasrmapbt(&mp->m_sb) && |
00795aae DW |
2412 | (be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) < 1 || |
2413 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]) > XFS_BTREE_MAXLEVELS)) | |
e37838e5 DW |
2414 | return false; |
2415 | ||
a2ceac1f DC |
2416 | /* |
2417 | * during growfs operations, the perag is not fully initialised, | |
2418 | * so we can't use it for any useful checking. growfs ensures we can't | |
2419 | * use it by using uncached buffers that don't have the perag attached | |
2420 | * so we can detect and avoid this problem. | |
2421 | */ | |
dd5b876e DC |
2422 | if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno) |
2423 | return false; | |
a2ceac1f | 2424 | |
dd5b876e DC |
2425 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
2426 | be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length)) | |
2427 | return false; | |
2428 | ||
88ce0792 | 2429 | if (xfs_sb_version_hasreflink(&mp->m_sb) && |
00795aae DW |
2430 | (be32_to_cpu(agf->agf_refcount_level) < 1 || |
2431 | be32_to_cpu(agf->agf_refcount_level) > XFS_BTREE_MAXLEVELS)) | |
88ce0792 DW |
2432 | return false; |
2433 | ||
dd5b876e | 2434 | return true;; |
a2ceac1f | 2435 | |
a2ceac1f DC |
2436 | } |
2437 | ||
2438 | static void | |
2439 | xfs_agf_read_verify( | |
2440 | struct xfs_buf *bp) | |
2441 | { | |
dd5b876e | 2442 | struct xfs_mount *mp = bp->b_target->bt_mount; |
dd5b876e | 2443 | |
45922933 DC |
2444 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2445 | !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF)) | |
12b53197 | 2446 | xfs_buf_ioerror(bp, -EFSBADCRC); |
45922933 DC |
2447 | else if (XFS_TEST_ERROR(!xfs_agf_verify(mp, bp), mp, |
2448 | XFS_ERRTAG_ALLOC_READ_AGF, | |
2449 | XFS_RANDOM_ALLOC_READ_AGF)) | |
12b53197 | 2450 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 DC |
2451 | |
2452 | if (bp->b_error) | |
2453 | xfs_verifier_error(bp); | |
a2ceac1f DC |
2454 | } |
2455 | ||
2456 | static void | |
2457 | xfs_agf_write_verify( | |
2458 | struct xfs_buf *bp) | |
2459 | { | |
dd5b876e DC |
2460 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2461 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2462 | ||
2463 | if (!xfs_agf_verify(mp, bp)) { | |
12b53197 | 2464 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
45922933 | 2465 | xfs_verifier_error(bp); |
dd5b876e DC |
2466 | return; |
2467 | } | |
2468 | ||
2469 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2470 | return; | |
2471 | ||
2472 | if (bip) | |
2473 | XFS_BUF_TO_AGF(bp)->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
2474 | ||
43b5aeed | 2475 | xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF); |
a2ceac1f DC |
2476 | } |
2477 | ||
2478 | const struct xfs_buf_ops xfs_agf_buf_ops = { | |
a3fac935 | 2479 | .name = "xfs_agf", |
a2ceac1f DC |
2480 | .verify_read = xfs_agf_read_verify, |
2481 | .verify_write = xfs_agf_write_verify, | |
2482 | }; | |
2483 | ||
2bd0ea18 NS |
2484 | /* |
2485 | * Read in the allocation group header (free/alloc section). | |
2486 | */ | |
2487 | int /* error */ | |
56b2de80 DC |
2488 | xfs_read_agf( |
2489 | struct xfs_mount *mp, /* mount point structure */ | |
2490 | struct xfs_trans *tp, /* transaction pointer */ | |
2491 | xfs_agnumber_t agno, /* allocation group number */ | |
2492 | int flags, /* XFS_BUF_ */ | |
2493 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
2bd0ea18 | 2494 | { |
9440d84d | 2495 | int error; |
2bd0ea18 | 2496 | |
ff105f75 DC |
2497 | trace_xfs_read_agf(mp, agno); |
2498 | ||
2bd0ea18 | 2499 | ASSERT(agno != NULLAGNUMBER); |
9440d84d NS |
2500 | error = xfs_trans_read_buf( |
2501 | mp, tp, mp->m_ddev_targp, | |
2502 | XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)), | |
a2ceac1f | 2503 | XFS_FSS_TO_BB(mp, 1), flags, bpp, &xfs_agf_buf_ops); |
9440d84d | 2504 | if (error) |
2bd0ea18 | 2505 | return error; |
56b2de80 | 2506 | if (!*bpp) |
2bd0ea18 | 2507 | return 0; |
56b2de80 | 2508 | |
a2ceac1f DC |
2509 | ASSERT(!(*bpp)->b_error); |
2510 | xfs_buf_set_ref(*bpp, XFS_AGF_REF); | |
56b2de80 DC |
2511 | return 0; |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | * Read in the allocation group header (free/alloc section). | |
2516 | */ | |
2517 | int /* error */ | |
2518 | xfs_alloc_read_agf( | |
2519 | struct xfs_mount *mp, /* mount point structure */ | |
2520 | struct xfs_trans *tp, /* transaction pointer */ | |
2521 | xfs_agnumber_t agno, /* allocation group number */ | |
2522 | int flags, /* XFS_ALLOC_FLAG_... */ | |
2523 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
2524 | { | |
2525 | struct xfs_agf *agf; /* ag freelist header */ | |
2526 | struct xfs_perag *pag; /* per allocation group data */ | |
2527 | int error; | |
2528 | ||
ff105f75 | 2529 | trace_xfs_alloc_read_agf(mp, agno); |
56b2de80 | 2530 | |
ff105f75 | 2531 | ASSERT(agno != NULLAGNUMBER); |
56b2de80 DC |
2532 | error = xfs_read_agf(mp, tp, agno, |
2533 | (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0, | |
2534 | bpp); | |
2535 | if (error) | |
2536 | return error; | |
2537 | if (!*bpp) | |
2538 | return 0; | |
a2ceac1f | 2539 | ASSERT(!(*bpp)->b_error); |
56b2de80 DC |
2540 | |
2541 | agf = XFS_BUF_TO_AGF(*bpp); | |
2542 | pag = xfs_perag_get(mp, agno); | |
2bd0ea18 | 2543 | if (!pag->pagf_init) { |
6e3140c7 | 2544 | pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); |
cdded3d8 | 2545 | pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); |
6e3140c7 NS |
2546 | pag->pagf_flcount = be32_to_cpu(agf->agf_flcount); |
2547 | pag->pagf_longest = be32_to_cpu(agf->agf_longest); | |
2bd0ea18 | 2548 | pag->pagf_levels[XFS_BTNUM_BNOi] = |
6e3140c7 | 2549 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); |
2bd0ea18 | 2550 | pag->pagf_levels[XFS_BTNUM_CNTi] = |
6e3140c7 | 2551 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); |
e37838e5 DW |
2552 | pag->pagf_levels[XFS_BTNUM_RMAPi] = |
2553 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAPi]); | |
88ce0792 | 2554 | pag->pagf_refcount_level = be32_to_cpu(agf->agf_refcount_level); |
5e656dbb | 2555 | spin_lock_init(&pag->pagb_lock); |
56b2de80 | 2556 | pag->pagb_count = 0; |
ff105f75 DC |
2557 | /* XXX: pagb_tree doesn't exist in userspace */ |
2558 | //pag->pagb_tree = RB_ROOT; | |
2bd0ea18 NS |
2559 | pag->pagf_init = 1; |
2560 | } | |
2561 | #ifdef DEBUG | |
2562 | else if (!XFS_FORCED_SHUTDOWN(mp)) { | |
6e3140c7 | 2563 | ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks)); |
cdded3d8 | 2564 | ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks)); |
6e3140c7 NS |
2565 | ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount)); |
2566 | ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest)); | |
2bd0ea18 | 2567 | ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] == |
6e3140c7 | 2568 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi])); |
2bd0ea18 | 2569 | ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] == |
6e3140c7 | 2570 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); |
2bd0ea18 NS |
2571 | } |
2572 | #endif | |
56b2de80 | 2573 | xfs_perag_put(pag); |
2bd0ea18 NS |
2574 | return 0; |
2575 | } | |
2576 | ||
2577 | /* | |
2578 | * Allocate an extent (variable-size). | |
2579 | * Depending on the allocation type, we either look in a single allocation | |
2580 | * group or loop over the allocation groups to find the result. | |
2581 | */ | |
2582 | int /* error */ | |
2583 | xfs_alloc_vextent( | |
dfc130f3 | 2584 | xfs_alloc_arg_t *args) /* allocation argument structure */ |
2bd0ea18 | 2585 | { |
dfc130f3 | 2586 | xfs_agblock_t agsize; /* allocation group size */ |
2bd0ea18 NS |
2587 | int error; |
2588 | int flags; /* XFS_ALLOC_FLAG_... locking flags */ | |
2bd0ea18 NS |
2589 | xfs_mount_t *mp; /* mount structure pointer */ |
2590 | xfs_agnumber_t sagno; /* starting allocation group number */ | |
dfc130f3 | 2591 | xfs_alloctype_t type; /* input allocation type */ |
34317449 | 2592 | int bump_rotor = 0; |
46eca962 | 2593 | xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */ |
2bd0ea18 NS |
2594 | |
2595 | mp = args->mp; | |
2596 | type = args->otype = args->type; | |
2597 | args->agbno = NULLAGBLOCK; | |
2598 | /* | |
2599 | * Just fix this up, for the case where the last a.g. is shorter | |
2600 | * (or there's only one a.g.) and the caller couldn't easily figure | |
2601 | * that out (xfs_bmap_alloc). | |
2602 | */ | |
2603 | agsize = mp->m_sb.sb_agblocks; | |
2604 | if (args->maxlen > agsize) | |
2605 | args->maxlen = agsize; | |
2606 | if (args->alignment == 0) | |
2607 | args->alignment = 1; | |
2608 | ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount); | |
2609 | ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize); | |
2610 | ASSERT(args->minlen <= args->maxlen); | |
2611 | ASSERT(args->minlen <= agsize); | |
2612 | ASSERT(args->mod < args->prod); | |
2613 | if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount || | |
2614 | XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize || | |
2615 | args->minlen > args->maxlen || args->minlen > agsize || | |
2616 | args->mod >= args->prod) { | |
2617 | args->fsbno = NULLFSBLOCK; | |
56b2de80 | 2618 | trace_xfs_alloc_vextent_badargs(args); |
2bd0ea18 NS |
2619 | return 0; |
2620 | } | |
9baa549b | 2621 | |
2bd0ea18 NS |
2622 | switch (type) { |
2623 | case XFS_ALLOCTYPE_THIS_AG: | |
2624 | case XFS_ALLOCTYPE_NEAR_BNO: | |
2625 | case XFS_ALLOCTYPE_THIS_BNO: | |
2626 | /* | |
2627 | * These three force us into a single a.g. | |
2628 | */ | |
2629 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
56b2de80 | 2630 | args->pag = xfs_perag_get(mp, args->agno); |
2bd0ea18 | 2631 | error = xfs_alloc_fix_freelist(args, 0); |
2bd0ea18 | 2632 | if (error) { |
56b2de80 | 2633 | trace_xfs_alloc_vextent_nofix(args); |
2bd0ea18 NS |
2634 | goto error0; |
2635 | } | |
2636 | if (!args->agbp) { | |
56b2de80 | 2637 | trace_xfs_alloc_vextent_noagbp(args); |
2bd0ea18 NS |
2638 | break; |
2639 | } | |
2640 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
0e266570 | 2641 | if ((error = xfs_alloc_ag_vextent(args))) |
2bd0ea18 | 2642 | goto error0; |
2bd0ea18 NS |
2643 | break; |
2644 | case XFS_ALLOCTYPE_START_BNO: | |
2645 | /* | |
2646 | * Try near allocation first, then anywhere-in-ag after | |
2647 | * the first a.g. fails. | |
2648 | */ | |
1fccd5c8 | 2649 | if ((args->datatype & XFS_ALLOC_INITIAL_USER_DATA) && |
34317449 | 2650 | (mp->m_flags & XFS_MOUNT_32BITINODES)) { |
46eca962 NS |
2651 | args->fsbno = XFS_AGB_TO_FSB(mp, |
2652 | ((mp->m_agfrotor / rotorstep) % | |
2653 | mp->m_sb.sb_agcount), 0); | |
34317449 NS |
2654 | bump_rotor = 1; |
2655 | } | |
2bd0ea18 NS |
2656 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); |
2657 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2658 | /* FALLTHROUGH */ | |
2659 | case XFS_ALLOCTYPE_ANY_AG: | |
2660 | case XFS_ALLOCTYPE_START_AG: | |
2661 | case XFS_ALLOCTYPE_FIRST_AG: | |
2662 | /* | |
2663 | * Rotate through the allocation groups looking for a winner. | |
2664 | */ | |
2665 | if (type == XFS_ALLOCTYPE_ANY_AG) { | |
2666 | /* | |
2667 | * Start with the last place we left off. | |
2668 | */ | |
46eca962 NS |
2669 | args->agno = sagno = (mp->m_agfrotor / rotorstep) % |
2670 | mp->m_sb.sb_agcount; | |
2bd0ea18 NS |
2671 | args->type = XFS_ALLOCTYPE_THIS_AG; |
2672 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2673 | } else if (type == XFS_ALLOCTYPE_FIRST_AG) { | |
2674 | /* | |
2675 | * Start with allocation group given by bno. | |
2676 | */ | |
2677 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2678 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2679 | sagno = 0; | |
2680 | flags = 0; | |
2681 | } else { | |
2682 | if (type == XFS_ALLOCTYPE_START_AG) | |
2683 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2684 | /* | |
2685 | * Start with the given allocation group. | |
2686 | */ | |
2687 | args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2688 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2689 | } | |
2690 | /* | |
2691 | * Loop over allocation groups twice; first time with | |
2692 | * trylock set, second time without. | |
2693 | */ | |
2694 | for (;;) { | |
56b2de80 | 2695 | args->pag = xfs_perag_get(mp, args->agno); |
9baa549b | 2696 | error = xfs_alloc_fix_freelist(args, flags); |
9baa549b | 2697 | if (error) { |
56b2de80 | 2698 | trace_xfs_alloc_vextent_nofix(args); |
2bd0ea18 NS |
2699 | goto error0; |
2700 | } | |
2701 | /* | |
2702 | * If we get a buffer back then the allocation will fly. | |
2703 | */ | |
2704 | if (args->agbp) { | |
0e266570 | 2705 | if ((error = xfs_alloc_ag_vextent(args))) |
2bd0ea18 | 2706 | goto error0; |
2bd0ea18 NS |
2707 | break; |
2708 | } | |
56b2de80 DC |
2709 | |
2710 | trace_xfs_alloc_vextent_loopfailed(args); | |
2711 | ||
2bd0ea18 NS |
2712 | /* |
2713 | * Didn't work, figure out the next iteration. | |
2714 | */ | |
2715 | if (args->agno == sagno && | |
2716 | type == XFS_ALLOCTYPE_START_BNO) | |
2717 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
5e656dbb BN |
2718 | /* |
2719 | * For the first allocation, we can try any AG to get | |
2720 | * space. However, if we already have allocated a | |
2721 | * block, we don't want to try AGs whose number is below | |
2722 | * sagno. Otherwise, we may end up with out-of-order | |
2723 | * locking of AGF, which might cause deadlock. | |
2724 | */ | |
2725 | if (++(args->agno) == mp->m_sb.sb_agcount) { | |
2726 | if (args->firstblock != NULLFSBLOCK) | |
2727 | args->agno = sagno; | |
2728 | else | |
2729 | args->agno = 0; | |
2730 | } | |
5000d01d | 2731 | /* |
2bd0ea18 NS |
2732 | * Reached the starting a.g., must either be done |
2733 | * or switch to non-trylock mode. | |
2734 | */ | |
2735 | if (args->agno == sagno) { | |
a3b4a951 | 2736 | if (flags == 0) { |
2bd0ea18 | 2737 | args->agbno = NULLAGBLOCK; |
56b2de80 | 2738 | trace_xfs_alloc_vextent_allfailed(args); |
2bd0ea18 NS |
2739 | break; |
2740 | } | |
a3b4a951 CH |
2741 | |
2742 | flags = 0; | |
2743 | if (type == XFS_ALLOCTYPE_START_BNO) { | |
2744 | args->agbno = XFS_FSB_TO_AGBNO(mp, | |
2745 | args->fsbno); | |
2746 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2bd0ea18 NS |
2747 | } |
2748 | } | |
56b2de80 | 2749 | xfs_perag_put(args->pag); |
2bd0ea18 | 2750 | } |
46eca962 NS |
2751 | if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) { |
2752 | if (args->agno == sagno) | |
2753 | mp->m_agfrotor = (mp->m_agfrotor + 1) % | |
2754 | (mp->m_sb.sb_agcount * rotorstep); | |
2755 | else | |
2756 | mp->m_agfrotor = (args->agno * rotorstep + 1) % | |
2757 | (mp->m_sb.sb_agcount * rotorstep); | |
2758 | } | |
2bd0ea18 NS |
2759 | break; |
2760 | default: | |
2761 | ASSERT(0); | |
2762 | /* NOTREACHED */ | |
2763 | } | |
2764 | if (args->agbno == NULLAGBLOCK) | |
2765 | args->fsbno = NULLFSBLOCK; | |
2766 | else { | |
2767 | args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno); | |
2768 | #ifdef DEBUG | |
2769 | ASSERT(args->len >= args->minlen); | |
2770 | ASSERT(args->len <= args->maxlen); | |
2771 | ASSERT(args->agbno % args->alignment == 0); | |
2772 | XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), | |
2773 | args->len); | |
2774 | #endif | |
9542ae13 DC |
2775 | |
2776 | /* Zero the extent if we were asked to do so */ | |
1fccd5c8 | 2777 | if (args->datatype & XFS_ALLOC_USERDATA_ZERO) { |
9542ae13 DC |
2778 | error = xfs_zero_extent(args->ip, args->fsbno, args->len); |
2779 | if (error) | |
2780 | goto error0; | |
2781 | } | |
2782 | ||
2bd0ea18 | 2783 | } |
56b2de80 | 2784 | xfs_perag_put(args->pag); |
2bd0ea18 NS |
2785 | return 0; |
2786 | error0: | |
56b2de80 | 2787 | xfs_perag_put(args->pag); |
2bd0ea18 NS |
2788 | return error; |
2789 | } | |
2790 | ||
2a6da3b8 DC |
2791 | /* Ensure that the freelist is at full capacity. */ |
2792 | int | |
2793 | xfs_free_extent_fix_freelist( | |
2794 | struct xfs_trans *tp, | |
2795 | xfs_agnumber_t agno, | |
2796 | struct xfs_buf **agbp) | |
2bd0ea18 | 2797 | { |
2a6da3b8 DC |
2798 | struct xfs_alloc_arg args; |
2799 | int error; | |
2bd0ea18 | 2800 | |
2a6da3b8 | 2801 | memset(&args, 0, sizeof(struct xfs_alloc_arg)); |
2bd0ea18 NS |
2802 | args.tp = tp; |
2803 | args.mp = tp->t_mountp; | |
2a6da3b8 | 2804 | args.agno = agno; |
a2ceac1f DC |
2805 | |
2806 | /* | |
2807 | * validate that the block number is legal - the enables us to detect | |
2808 | * and handle a silent filesystem corruption rather than crashing. | |
2809 | */ | |
a2ceac1f | 2810 | if (args.agno >= args.mp->m_sb.sb_agcount) |
12b53197 | 2811 | return -EFSCORRUPTED; |
a2ceac1f | 2812 | |
56b2de80 | 2813 | args.pag = xfs_perag_get(args.mp, args.agno); |
a2ceac1f DC |
2814 | ASSERT(args.pag); |
2815 | ||
2816 | error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING); | |
2817 | if (error) | |
2a6da3b8 DC |
2818 | goto out; |
2819 | ||
2820 | *agbp = args.agbp; | |
2821 | out: | |
2822 | xfs_perag_put(args.pag); | |
2823 | return error; | |
2824 | } | |
2825 | ||
2826 | /* | |
2827 | * Free an extent. | |
2828 | * Just break up the extent address and hand off to xfs_free_ag_extent | |
2829 | * after fixing up the freelist. | |
2830 | */ | |
2831 | int /* error */ | |
2832 | xfs_free_extent( | |
2833 | struct xfs_trans *tp, /* transaction pointer */ | |
2834 | xfs_fsblock_t bno, /* starting block number of extent */ | |
85aec44f | 2835 | xfs_extlen_t len, /* length of extent */ |
cf8ce220 DW |
2836 | struct xfs_owner_info *oinfo, /* extent owner */ |
2837 | enum xfs_ag_resv_type type) /* block reservation type */ | |
2a6da3b8 DC |
2838 | { |
2839 | struct xfs_mount *mp = tp->t_mountp; | |
2840 | struct xfs_buf *agbp; | |
2841 | xfs_agnumber_t agno = XFS_FSB_TO_AGNO(mp, bno); | |
2842 | xfs_agblock_t agbno = XFS_FSB_TO_AGBNO(mp, bno); | |
2843 | int error; | |
2844 | ||
2845 | ASSERT(len != 0); | |
cf8ce220 | 2846 | ASSERT(type != XFS_AG_RESV_AGFL); |
2a6da3b8 | 2847 | |
a9da40de DW |
2848 | if (XFS_TEST_ERROR(false, mp, |
2849 | XFS_ERRTAG_FREE_EXTENT, | |
2850 | XFS_RANDOM_FREE_EXTENT)) | |
2851 | return -EIO; | |
2852 | ||
2a6da3b8 DC |
2853 | error = xfs_free_extent_fix_freelist(tp, agno, &agbp); |
2854 | if (error) | |
2855 | return error; | |
2856 | ||
2857 | XFS_WANT_CORRUPTED_GOTO(mp, agbno < mp->m_sb.sb_agblocks, err); | |
a2ceac1f DC |
2858 | |
2859 | /* validate the extent size is legal now we have the agf locked */ | |
2a6da3b8 DC |
2860 | XFS_WANT_CORRUPTED_GOTO(mp, |
2861 | agbno + len <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_length), | |
2862 | err); | |
a2ceac1f | 2863 | |
cf8ce220 | 2864 | error = xfs_free_ag_extent(tp, agbp, agno, agbno, len, oinfo, type); |
2a6da3b8 DC |
2865 | if (error) |
2866 | goto err; | |
2867 | ||
2868 | xfs_extent_busy_insert(tp, agno, agbno, len, 0); | |
2869 | return 0; | |
2870 | ||
2871 | err: | |
2872 | xfs_trans_brelse(tp, agbp); | |
2bd0ea18 NS |
2873 | return error; |
2874 | } |