Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_format.h" |
239880ef | 21 | #include "xfs_log_format.h" |
70a9883c | 22 | #include "xfs_shared.h" |
239880ef | 23 | #include "xfs_trans_resv.h" |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
a844f451 | 27 | #include "xfs_inode.h" |
1da177e4 | 28 | #include "xfs_btree.h" |
a4fbe6ab | 29 | #include "xfs_alloc_btree.h" |
1da177e4 | 30 | #include "xfs_alloc.h" |
efc27b52 | 31 | #include "xfs_extent_busy.h" |
1da177e4 | 32 | #include "xfs_error.h" |
4e0e6040 | 33 | #include "xfs_cksum.h" |
0b1b213f | 34 | #include "xfs_trace.h" |
239880ef | 35 | #include "xfs_trans.h" |
4e0e6040 | 36 | #include "xfs_buf_item.h" |
239880ef | 37 | #include "xfs_log.h" |
1da177e4 | 38 | |
c999a223 | 39 | struct workqueue_struct *xfs_alloc_wq; |
1da177e4 LT |
40 | |
41 | #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b))) | |
42 | ||
43 | #define XFSA_FIXUP_BNO_OK 1 | |
44 | #define XFSA_FIXUP_CNT_OK 2 | |
45 | ||
1da177e4 LT |
46 | STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *); |
47 | STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *); | |
48 | STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *); | |
49 | STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *, | |
e26f0501 | 50 | xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *); |
1da177e4 | 51 | |
fe033cc8 CH |
52 | /* |
53 | * Lookup the record equal to [bno, len] in the btree given by cur. | |
54 | */ | |
55 | STATIC int /* error */ | |
56 | xfs_alloc_lookup_eq( | |
57 | struct xfs_btree_cur *cur, /* btree cursor */ | |
58 | xfs_agblock_t bno, /* starting block of extent */ | |
59 | xfs_extlen_t len, /* length of extent */ | |
60 | int *stat) /* success/failure */ | |
61 | { | |
62 | cur->bc_rec.a.ar_startblock = bno; | |
63 | cur->bc_rec.a.ar_blockcount = len; | |
64 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); | |
65 | } | |
66 | ||
67 | /* | |
68 | * Lookup the first record greater than or equal to [bno, len] | |
69 | * in the btree given by cur. | |
70 | */ | |
a66d6363 | 71 | int /* error */ |
fe033cc8 CH |
72 | xfs_alloc_lookup_ge( |
73 | struct xfs_btree_cur *cur, /* btree cursor */ | |
74 | xfs_agblock_t bno, /* starting block of extent */ | |
75 | xfs_extlen_t len, /* length of extent */ | |
76 | int *stat) /* success/failure */ | |
77 | { | |
78 | cur->bc_rec.a.ar_startblock = bno; | |
79 | cur->bc_rec.a.ar_blockcount = len; | |
80 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); | |
81 | } | |
82 | ||
83 | /* | |
84 | * Lookup the first record less than or equal to [bno, len] | |
85 | * in the btree given by cur. | |
86 | */ | |
a46db608 | 87 | int /* error */ |
fe033cc8 CH |
88 | xfs_alloc_lookup_le( |
89 | struct xfs_btree_cur *cur, /* btree cursor */ | |
90 | xfs_agblock_t bno, /* starting block of extent */ | |
91 | xfs_extlen_t len, /* length of extent */ | |
92 | int *stat) /* success/failure */ | |
93 | { | |
94 | cur->bc_rec.a.ar_startblock = bno; | |
95 | cur->bc_rec.a.ar_blockcount = len; | |
96 | return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat); | |
97 | } | |
98 | ||
278d0ca1 CH |
99 | /* |
100 | * Update the record referred to by cur to the value given | |
101 | * by [bno, len]. | |
102 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
103 | */ | |
104 | STATIC int /* error */ | |
105 | xfs_alloc_update( | |
106 | struct xfs_btree_cur *cur, /* btree cursor */ | |
107 | xfs_agblock_t bno, /* starting block of extent */ | |
108 | xfs_extlen_t len) /* length of extent */ | |
109 | { | |
110 | union xfs_btree_rec rec; | |
111 | ||
112 | rec.alloc.ar_startblock = cpu_to_be32(bno); | |
113 | rec.alloc.ar_blockcount = cpu_to_be32(len); | |
114 | return xfs_btree_update(cur, &rec); | |
115 | } | |
fe033cc8 | 116 | |
8cc938fe CH |
117 | /* |
118 | * Get the data from the pointed-to record. | |
119 | */ | |
a46db608 | 120 | int /* error */ |
8cc938fe CH |
121 | xfs_alloc_get_rec( |
122 | struct xfs_btree_cur *cur, /* btree cursor */ | |
123 | xfs_agblock_t *bno, /* output: starting block of extent */ | |
124 | xfs_extlen_t *len, /* output: length of extent */ | |
125 | int *stat) /* output: success/failure */ | |
126 | { | |
127 | union xfs_btree_rec *rec; | |
128 | int error; | |
129 | ||
130 | error = xfs_btree_get_rec(cur, &rec, stat); | |
131 | if (!error && *stat == 1) { | |
132 | *bno = be32_to_cpu(rec->alloc.ar_startblock); | |
133 | *len = be32_to_cpu(rec->alloc.ar_blockcount); | |
134 | } | |
135 | return error; | |
136 | } | |
137 | ||
1da177e4 LT |
138 | /* |
139 | * Compute aligned version of the found extent. | |
140 | * Takes alignment and min length into account. | |
141 | */ | |
12375c82 | 142 | STATIC void |
1da177e4 | 143 | xfs_alloc_compute_aligned( |
86fa8af6 | 144 | xfs_alloc_arg_t *args, /* allocation argument structure */ |
1da177e4 LT |
145 | xfs_agblock_t foundbno, /* starting block in found extent */ |
146 | xfs_extlen_t foundlen, /* length in found extent */ | |
1da177e4 LT |
147 | xfs_agblock_t *resbno, /* result block number */ |
148 | xfs_extlen_t *reslen) /* result length */ | |
149 | { | |
150 | xfs_agblock_t bno; | |
1da177e4 | 151 | xfs_extlen_t len; |
bfe46d4e | 152 | xfs_extlen_t diff; |
1da177e4 | 153 | |
e26f0501 | 154 | /* Trim busy sections out of found extent */ |
4ecbfe63 | 155 | xfs_extent_busy_trim(args, foundbno, foundlen, &bno, &len); |
e26f0501 | 156 | |
bfe46d4e BF |
157 | /* |
158 | * If we have a largish extent that happens to start before min_agbno, | |
159 | * see if we can shift it into range... | |
160 | */ | |
161 | if (bno < args->min_agbno && bno + len > args->min_agbno) { | |
162 | diff = args->min_agbno - bno; | |
163 | if (len > diff) { | |
164 | bno += diff; | |
165 | len -= diff; | |
166 | } | |
167 | } | |
168 | ||
e26f0501 CH |
169 | if (args->alignment > 1 && len >= args->minlen) { |
170 | xfs_agblock_t aligned_bno = roundup(bno, args->alignment); | |
bfe46d4e BF |
171 | |
172 | diff = aligned_bno - bno; | |
e26f0501 CH |
173 | |
174 | *resbno = aligned_bno; | |
175 | *reslen = diff >= len ? 0 : len - diff; | |
1da177e4 | 176 | } else { |
e26f0501 CH |
177 | *resbno = bno; |
178 | *reslen = len; | |
1da177e4 | 179 | } |
1da177e4 LT |
180 | } |
181 | ||
182 | /* | |
183 | * Compute best start block and diff for "near" allocations. | |
184 | * freelen >= wantlen already checked by caller. | |
185 | */ | |
186 | STATIC xfs_extlen_t /* difference value (absolute) */ | |
187 | xfs_alloc_compute_diff( | |
188 | xfs_agblock_t wantbno, /* target starting block */ | |
189 | xfs_extlen_t wantlen, /* target length */ | |
190 | xfs_extlen_t alignment, /* target alignment */ | |
211d022c | 191 | char userdata, /* are we allocating data? */ |
1da177e4 LT |
192 | xfs_agblock_t freebno, /* freespace's starting block */ |
193 | xfs_extlen_t freelen, /* freespace's length */ | |
194 | xfs_agblock_t *newbnop) /* result: best start block from free */ | |
195 | { | |
196 | xfs_agblock_t freeend; /* end of freespace extent */ | |
197 | xfs_agblock_t newbno1; /* return block number */ | |
198 | xfs_agblock_t newbno2; /* other new block number */ | |
199 | xfs_extlen_t newlen1=0; /* length with newbno1 */ | |
200 | xfs_extlen_t newlen2=0; /* length with newbno2 */ | |
201 | xfs_agblock_t wantend; /* end of target extent */ | |
202 | ||
203 | ASSERT(freelen >= wantlen); | |
204 | freeend = freebno + freelen; | |
205 | wantend = wantbno + wantlen; | |
211d022c JK |
206 | /* |
207 | * We want to allocate from the start of a free extent if it is past | |
208 | * the desired block or if we are allocating user data and the free | |
209 | * extent is before desired block. The second case is there to allow | |
210 | * for contiguous allocation from the remaining free space if the file | |
211 | * grows in the short term. | |
212 | */ | |
213 | if (freebno >= wantbno || (userdata && freeend < wantend)) { | |
1da177e4 LT |
214 | if ((newbno1 = roundup(freebno, alignment)) >= freeend) |
215 | newbno1 = NULLAGBLOCK; | |
216 | } else if (freeend >= wantend && alignment > 1) { | |
217 | newbno1 = roundup(wantbno, alignment); | |
218 | newbno2 = newbno1 - alignment; | |
219 | if (newbno1 >= freeend) | |
220 | newbno1 = NULLAGBLOCK; | |
221 | else | |
222 | newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1); | |
223 | if (newbno2 < freebno) | |
224 | newbno2 = NULLAGBLOCK; | |
225 | else | |
226 | newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2); | |
227 | if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) { | |
228 | if (newlen1 < newlen2 || | |
229 | (newlen1 == newlen2 && | |
230 | XFS_ABSDIFF(newbno1, wantbno) > | |
231 | XFS_ABSDIFF(newbno2, wantbno))) | |
232 | newbno1 = newbno2; | |
233 | } else if (newbno2 != NULLAGBLOCK) | |
234 | newbno1 = newbno2; | |
235 | } else if (freeend >= wantend) { | |
236 | newbno1 = wantbno; | |
237 | } else if (alignment > 1) { | |
238 | newbno1 = roundup(freeend - wantlen, alignment); | |
239 | if (newbno1 > freeend - wantlen && | |
240 | newbno1 - alignment >= freebno) | |
241 | newbno1 -= alignment; | |
242 | else if (newbno1 >= freeend) | |
243 | newbno1 = NULLAGBLOCK; | |
244 | } else | |
245 | newbno1 = freeend - wantlen; | |
246 | *newbnop = newbno1; | |
247 | return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno); | |
248 | } | |
249 | ||
250 | /* | |
251 | * Fix up the length, based on mod and prod. | |
252 | * len should be k * prod + mod for some k. | |
253 | * If len is too small it is returned unchanged. | |
254 | * If len hits maxlen it is left alone. | |
255 | */ | |
256 | STATIC void | |
257 | xfs_alloc_fix_len( | |
258 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
259 | { | |
260 | xfs_extlen_t k; | |
261 | xfs_extlen_t rlen; | |
262 | ||
263 | ASSERT(args->mod < args->prod); | |
264 | rlen = args->len; | |
265 | ASSERT(rlen >= args->minlen); | |
266 | ASSERT(rlen <= args->maxlen); | |
267 | if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen || | |
268 | (args->mod == 0 && rlen < args->prod)) | |
269 | return; | |
270 | k = rlen % args->prod; | |
271 | if (k == args->mod) | |
272 | return; | |
30265117 JK |
273 | if (k > args->mod) |
274 | rlen = rlen - (k - args->mod); | |
275 | else | |
276 | rlen = rlen - args->prod + (args->mod - k); | |
3790a8cd | 277 | /* casts to (int) catch length underflows */ |
30265117 JK |
278 | if ((int)rlen < (int)args->minlen) |
279 | return; | |
280 | ASSERT(rlen >= args->minlen && rlen <= args->maxlen); | |
281 | ASSERT(rlen % args->prod == args->mod); | |
1da177e4 LT |
282 | args->len = rlen; |
283 | } | |
284 | ||
285 | /* | |
286 | * Fix up length if there is too little space left in the a.g. | |
287 | * Return 1 if ok, 0 if too little, should give up. | |
288 | */ | |
289 | STATIC int | |
290 | xfs_alloc_fix_minleft( | |
291 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
292 | { | |
293 | xfs_agf_t *agf; /* a.g. freelist header */ | |
294 | int diff; /* free space difference */ | |
295 | ||
296 | if (args->minleft == 0) | |
297 | return 1; | |
298 | agf = XFS_BUF_TO_AGF(args->agbp); | |
16259e7d | 299 | diff = be32_to_cpu(agf->agf_freeblks) |
1da177e4 LT |
300 | - args->len - args->minleft; |
301 | if (diff >= 0) | |
302 | return 1; | |
303 | args->len += diff; /* shrink the allocated space */ | |
3790a8cd DC |
304 | /* casts to (int) catch length underflows */ |
305 | if ((int)args->len >= (int)args->minlen) | |
1da177e4 LT |
306 | return 1; |
307 | args->agbno = NULLAGBLOCK; | |
308 | return 0; | |
309 | } | |
310 | ||
311 | /* | |
312 | * Update the two btrees, logically removing from freespace the extent | |
313 | * starting at rbno, rlen blocks. The extent is contained within the | |
314 | * actual (current) free extent fbno for flen blocks. | |
315 | * Flags are passed in indicating whether the cursors are set to the | |
316 | * relevant records. | |
317 | */ | |
318 | STATIC int /* error code */ | |
319 | xfs_alloc_fixup_trees( | |
320 | xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */ | |
321 | xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */ | |
322 | xfs_agblock_t fbno, /* starting block of free extent */ | |
323 | xfs_extlen_t flen, /* length of free extent */ | |
324 | xfs_agblock_t rbno, /* starting block of returned extent */ | |
325 | xfs_extlen_t rlen, /* length of returned extent */ | |
326 | int flags) /* flags, XFSA_FIXUP_... */ | |
327 | { | |
328 | int error; /* error code */ | |
329 | int i; /* operation results */ | |
330 | xfs_agblock_t nfbno1; /* first new free startblock */ | |
331 | xfs_agblock_t nfbno2; /* second new free startblock */ | |
332 | xfs_extlen_t nflen1=0; /* first new free length */ | |
333 | xfs_extlen_t nflen2=0; /* second new free length */ | |
5fb5aeee ES |
334 | struct xfs_mount *mp; |
335 | ||
336 | mp = cnt_cur->bc_mp; | |
1da177e4 LT |
337 | |
338 | /* | |
339 | * Look up the record in the by-size tree if necessary. | |
340 | */ | |
341 | if (flags & XFSA_FIXUP_CNT_OK) { | |
342 | #ifdef DEBUG | |
343 | if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i))) | |
344 | return error; | |
5fb5aeee | 345 | XFS_WANT_CORRUPTED_RETURN(mp, |
1da177e4 LT |
346 | i == 1 && nfbno1 == fbno && nflen1 == flen); |
347 | #endif | |
348 | } else { | |
349 | if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i))) | |
350 | return error; | |
5fb5aeee | 351 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
352 | } |
353 | /* | |
354 | * Look up the record in the by-block tree if necessary. | |
355 | */ | |
356 | if (flags & XFSA_FIXUP_BNO_OK) { | |
357 | #ifdef DEBUG | |
358 | if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i))) | |
359 | return error; | |
5fb5aeee | 360 | XFS_WANT_CORRUPTED_RETURN(mp, |
1da177e4 LT |
361 | i == 1 && nfbno1 == fbno && nflen1 == flen); |
362 | #endif | |
363 | } else { | |
364 | if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i))) | |
365 | return error; | |
5fb5aeee | 366 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 | 367 | } |
7cc95a82 | 368 | |
1da177e4 | 369 | #ifdef DEBUG |
7cc95a82 CH |
370 | if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) { |
371 | struct xfs_btree_block *bnoblock; | |
372 | struct xfs_btree_block *cntblock; | |
373 | ||
374 | bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]); | |
375 | cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]); | |
1da177e4 | 376 | |
5fb5aeee | 377 | XFS_WANT_CORRUPTED_RETURN(mp, |
7cc95a82 | 378 | bnoblock->bb_numrecs == cntblock->bb_numrecs); |
1da177e4 LT |
379 | } |
380 | #endif | |
7cc95a82 | 381 | |
1da177e4 LT |
382 | /* |
383 | * Deal with all four cases: the allocated record is contained | |
384 | * within the freespace record, so we can have new freespace | |
385 | * at either (or both) end, or no freespace remaining. | |
386 | */ | |
387 | if (rbno == fbno && rlen == flen) | |
388 | nfbno1 = nfbno2 = NULLAGBLOCK; | |
389 | else if (rbno == fbno) { | |
390 | nfbno1 = rbno + rlen; | |
391 | nflen1 = flen - rlen; | |
392 | nfbno2 = NULLAGBLOCK; | |
393 | } else if (rbno + rlen == fbno + flen) { | |
394 | nfbno1 = fbno; | |
395 | nflen1 = flen - rlen; | |
396 | nfbno2 = NULLAGBLOCK; | |
397 | } else { | |
398 | nfbno1 = fbno; | |
399 | nflen1 = rbno - fbno; | |
400 | nfbno2 = rbno + rlen; | |
401 | nflen2 = (fbno + flen) - nfbno2; | |
402 | } | |
403 | /* | |
404 | * Delete the entry from the by-size btree. | |
405 | */ | |
91cca5df | 406 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 407 | return error; |
5fb5aeee | 408 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
409 | /* |
410 | * Add new by-size btree entry(s). | |
411 | */ | |
412 | if (nfbno1 != NULLAGBLOCK) { | |
413 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i))) | |
414 | return error; | |
5fb5aeee | 415 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
4b22a571 | 416 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 417 | return error; |
5fb5aeee | 418 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
419 | } |
420 | if (nfbno2 != NULLAGBLOCK) { | |
421 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i))) | |
422 | return error; | |
5fb5aeee | 423 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
4b22a571 | 424 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 425 | return error; |
5fb5aeee | 426 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
427 | } |
428 | /* | |
429 | * Fix up the by-block btree entry(s). | |
430 | */ | |
431 | if (nfbno1 == NULLAGBLOCK) { | |
432 | /* | |
433 | * No remaining freespace, just delete the by-block tree entry. | |
434 | */ | |
91cca5df | 435 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 | 436 | return error; |
5fb5aeee | 437 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
438 | } else { |
439 | /* | |
440 | * Update the by-block entry to start later|be shorter. | |
441 | */ | |
442 | if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1))) | |
443 | return error; | |
444 | } | |
445 | if (nfbno2 != NULLAGBLOCK) { | |
446 | /* | |
447 | * 2 resulting free entries, need to add one. | |
448 | */ | |
449 | if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i))) | |
450 | return error; | |
5fb5aeee | 451 | XFS_WANT_CORRUPTED_RETURN(mp, i == 0); |
4b22a571 | 452 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 | 453 | return error; |
5fb5aeee | 454 | XFS_WANT_CORRUPTED_RETURN(mp, i == 1); |
1da177e4 LT |
455 | } |
456 | return 0; | |
457 | } | |
458 | ||
77c95bba | 459 | static bool |
612cfbfe | 460 | xfs_agfl_verify( |
bb80c6d7 DC |
461 | struct xfs_buf *bp) |
462 | { | |
bb80c6d7 DC |
463 | struct xfs_mount *mp = bp->b_target->bt_mount; |
464 | struct xfs_agfl *agfl = XFS_BUF_TO_AGFL(bp); | |
bb80c6d7 DC |
465 | int i; |
466 | ||
ce748eaa | 467 | if (!uuid_equal(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid)) |
77c95bba CH |
468 | return false; |
469 | if (be32_to_cpu(agfl->agfl_magicnum) != XFS_AGFL_MAGIC) | |
470 | return false; | |
471 | /* | |
472 | * during growfs operations, the perag is not fully initialised, | |
473 | * so we can't use it for any useful checking. growfs ensures we can't | |
474 | * use it by using uncached buffers that don't have the perag attached | |
475 | * so we can detect and avoid this problem. | |
476 | */ | |
477 | if (bp->b_pag && be32_to_cpu(agfl->agfl_seqno) != bp->b_pag->pag_agno) | |
478 | return false; | |
479 | ||
bb80c6d7 | 480 | for (i = 0; i < XFS_AGFL_SIZE(mp); i++) { |
77c95bba | 481 | if (be32_to_cpu(agfl->agfl_bno[i]) != NULLAGBLOCK && |
bb80c6d7 | 482 | be32_to_cpu(agfl->agfl_bno[i]) >= mp->m_sb.sb_agblocks) |
77c95bba | 483 | return false; |
bb80c6d7 | 484 | } |
a45086e2 BF |
485 | |
486 | return xfs_log_check_lsn(mp, | |
487 | be64_to_cpu(XFS_BUF_TO_AGFL(bp)->agfl_lsn)); | |
77c95bba CH |
488 | } |
489 | ||
490 | static void | |
491 | xfs_agfl_read_verify( | |
492 | struct xfs_buf *bp) | |
493 | { | |
494 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
77c95bba CH |
495 | |
496 | /* | |
497 | * There is no verification of non-crc AGFLs because mkfs does not | |
498 | * initialise the AGFL to zero or NULL. Hence the only valid part of the | |
499 | * AGFL is what the AGF says is active. We can't get to the AGF, so we | |
500 | * can't verify just those entries are valid. | |
501 | */ | |
502 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
503 | return; | |
504 | ||
ce5028cf | 505 | if (!xfs_buf_verify_cksum(bp, XFS_AGFL_CRC_OFF)) |
2451337d | 506 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf | 507 | else if (!xfs_agfl_verify(bp)) |
2451337d | 508 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
509 | |
510 | if (bp->b_error) | |
511 | xfs_verifier_error(bp); | |
612cfbfe DC |
512 | } |
513 | ||
1813dd64 | 514 | static void |
612cfbfe DC |
515 | xfs_agfl_write_verify( |
516 | struct xfs_buf *bp) | |
517 | { | |
77c95bba CH |
518 | struct xfs_mount *mp = bp->b_target->bt_mount; |
519 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
612cfbfe | 520 | |
77c95bba CH |
521 | /* no verification of non-crc AGFLs */ |
522 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
523 | return; | |
524 | ||
525 | if (!xfs_agfl_verify(bp)) { | |
2451337d | 526 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 527 | xfs_verifier_error(bp); |
77c95bba CH |
528 | return; |
529 | } | |
530 | ||
531 | if (bip) | |
532 | XFS_BUF_TO_AGFL(bp)->agfl_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
533 | ||
f1dbcd7e | 534 | xfs_buf_update_cksum(bp, XFS_AGFL_CRC_OFF); |
bb80c6d7 DC |
535 | } |
536 | ||
1813dd64 DC |
537 | const struct xfs_buf_ops xfs_agfl_buf_ops = { |
538 | .verify_read = xfs_agfl_read_verify, | |
539 | .verify_write = xfs_agfl_write_verify, | |
540 | }; | |
541 | ||
1da177e4 LT |
542 | /* |
543 | * Read in the allocation group free block array. | |
544 | */ | |
545 | STATIC int /* error */ | |
546 | xfs_alloc_read_agfl( | |
547 | xfs_mount_t *mp, /* mount point structure */ | |
548 | xfs_trans_t *tp, /* transaction pointer */ | |
549 | xfs_agnumber_t agno, /* allocation group number */ | |
550 | xfs_buf_t **bpp) /* buffer for the ag free block array */ | |
551 | { | |
552 | xfs_buf_t *bp; /* return value */ | |
553 | int error; | |
554 | ||
555 | ASSERT(agno != NULLAGNUMBER); | |
556 | error = xfs_trans_read_buf( | |
557 | mp, tp, mp->m_ddev_targp, | |
558 | XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)), | |
1813dd64 | 559 | XFS_FSS_TO_BB(mp, 1), 0, &bp, &xfs_agfl_buf_ops); |
1da177e4 LT |
560 | if (error) |
561 | return error; | |
38f23232 | 562 | xfs_buf_set_ref(bp, XFS_AGFL_REF); |
1da177e4 LT |
563 | *bpp = bp; |
564 | return 0; | |
565 | } | |
566 | ||
ecb6928f CH |
567 | STATIC int |
568 | xfs_alloc_update_counters( | |
569 | struct xfs_trans *tp, | |
570 | struct xfs_perag *pag, | |
571 | struct xfs_buf *agbp, | |
572 | long len) | |
573 | { | |
574 | struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); | |
575 | ||
576 | pag->pagf_freeblks += len; | |
577 | be32_add_cpu(&agf->agf_freeblks, len); | |
578 | ||
579 | xfs_trans_agblocks_delta(tp, len); | |
580 | if (unlikely(be32_to_cpu(agf->agf_freeblks) > | |
581 | be32_to_cpu(agf->agf_length))) | |
2451337d | 582 | return -EFSCORRUPTED; |
ecb6928f CH |
583 | |
584 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS); | |
585 | return 0; | |
586 | } | |
587 | ||
1da177e4 LT |
588 | /* |
589 | * Allocation group level functions. | |
590 | */ | |
591 | ||
592 | /* | |
593 | * Allocate a variable extent in the allocation group agno. | |
594 | * Type and bno are used to determine where in the allocation group the | |
595 | * extent will start. | |
596 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
597 | * and of the form k * prod + mod unless there's nothing that large. | |
598 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
599 | */ | |
600 | STATIC int /* error */ | |
601 | xfs_alloc_ag_vextent( | |
602 | xfs_alloc_arg_t *args) /* argument structure for allocation */ | |
603 | { | |
604 | int error=0; | |
1da177e4 LT |
605 | |
606 | ASSERT(args->minlen > 0); | |
607 | ASSERT(args->maxlen > 0); | |
608 | ASSERT(args->minlen <= args->maxlen); | |
609 | ASSERT(args->mod < args->prod); | |
610 | ASSERT(args->alignment > 0); | |
611 | /* | |
612 | * Branch to correct routine based on the type. | |
613 | */ | |
614 | args->wasfromfl = 0; | |
615 | switch (args->type) { | |
616 | case XFS_ALLOCTYPE_THIS_AG: | |
617 | error = xfs_alloc_ag_vextent_size(args); | |
618 | break; | |
619 | case XFS_ALLOCTYPE_NEAR_BNO: | |
620 | error = xfs_alloc_ag_vextent_near(args); | |
621 | break; | |
622 | case XFS_ALLOCTYPE_THIS_BNO: | |
623 | error = xfs_alloc_ag_vextent_exact(args); | |
624 | break; | |
625 | default: | |
626 | ASSERT(0); | |
627 | /* NOTREACHED */ | |
628 | } | |
ecb6928f CH |
629 | |
630 | if (error || args->agbno == NULLAGBLOCK) | |
1da177e4 | 631 | return error; |
ecb6928f CH |
632 | |
633 | ASSERT(args->len >= args->minlen); | |
634 | ASSERT(args->len <= args->maxlen); | |
635 | ASSERT(!args->wasfromfl || !args->isfl); | |
636 | ASSERT(args->agbno % args->alignment == 0); | |
637 | ||
638 | if (!args->wasfromfl) { | |
639 | error = xfs_alloc_update_counters(args->tp, args->pag, | |
640 | args->agbp, | |
641 | -((long)(args->len))); | |
642 | if (error) | |
643 | return error; | |
644 | ||
4ecbfe63 | 645 | ASSERT(!xfs_extent_busy_search(args->mp, args->agno, |
e26f0501 | 646 | args->agbno, args->len)); |
1da177e4 | 647 | } |
ecb6928f CH |
648 | |
649 | if (!args->isfl) { | |
650 | xfs_trans_mod_sb(args->tp, args->wasdel ? | |
651 | XFS_TRANS_SB_RES_FDBLOCKS : | |
652 | XFS_TRANS_SB_FDBLOCKS, | |
653 | -((long)(args->len))); | |
654 | } | |
655 | ||
ff6d6af2 BD |
656 | XFS_STATS_INC(args->mp, xs_allocx); |
657 | XFS_STATS_ADD(args->mp, xs_allocb, args->len); | |
ecb6928f | 658 | return error; |
1da177e4 LT |
659 | } |
660 | ||
661 | /* | |
662 | * Allocate a variable extent at exactly agno/bno. | |
663 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
664 | * and of the form k * prod + mod unless there's nothing that large. | |
665 | * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it. | |
666 | */ | |
667 | STATIC int /* error */ | |
668 | xfs_alloc_ag_vextent_exact( | |
669 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
670 | { | |
671 | xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */ | |
672 | xfs_btree_cur_t *cnt_cur;/* by count btree cursor */ | |
1da177e4 LT |
673 | int error; |
674 | xfs_agblock_t fbno; /* start block of found extent */ | |
1da177e4 | 675 | xfs_extlen_t flen; /* length of found extent */ |
e26f0501 CH |
676 | xfs_agblock_t tbno; /* start block of trimmed extent */ |
677 | xfs_extlen_t tlen; /* length of trimmed extent */ | |
678 | xfs_agblock_t tend; /* end block of trimmed extent */ | |
1da177e4 | 679 | int i; /* success/failure of operation */ |
1da177e4 LT |
680 | |
681 | ASSERT(args->alignment == 1); | |
9f9baab3 | 682 | |
1da177e4 LT |
683 | /* |
684 | * Allocate/initialize a cursor for the by-number freespace btree. | |
685 | */ | |
561f7d17 | 686 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
9f9baab3 CH |
687 | args->agno, XFS_BTNUM_BNO); |
688 | ||
1da177e4 LT |
689 | /* |
690 | * Lookup bno and minlen in the btree (minlen is irrelevant, really). | |
691 | * Look for the closest free block <= bno, it must contain bno | |
692 | * if any free block does. | |
693 | */ | |
9f9baab3 CH |
694 | error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i); |
695 | if (error) | |
1da177e4 | 696 | goto error0; |
9f9baab3 CH |
697 | if (!i) |
698 | goto not_found; | |
699 | ||
1da177e4 LT |
700 | /* |
701 | * Grab the freespace record. | |
702 | */ | |
9f9baab3 CH |
703 | error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i); |
704 | if (error) | |
1da177e4 | 705 | goto error0; |
c29aad41 | 706 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 | 707 | ASSERT(fbno <= args->agbno); |
9f9baab3 | 708 | |
1da177e4 | 709 | /* |
e26f0501 | 710 | * Check for overlapping busy extents. |
1da177e4 | 711 | */ |
4ecbfe63 | 712 | xfs_extent_busy_trim(args, fbno, flen, &tbno, &tlen); |
e26f0501 CH |
713 | |
714 | /* | |
715 | * Give up if the start of the extent is busy, or the freespace isn't | |
716 | * long enough for the minimum request. | |
717 | */ | |
718 | if (tbno > args->agbno) | |
719 | goto not_found; | |
720 | if (tlen < args->minlen) | |
721 | goto not_found; | |
722 | tend = tbno + tlen; | |
723 | if (tend < args->agbno + args->minlen) | |
9f9baab3 CH |
724 | goto not_found; |
725 | ||
1da177e4 LT |
726 | /* |
727 | * End of extent will be smaller of the freespace end and the | |
728 | * maximal requested end. | |
9f9baab3 | 729 | * |
1da177e4 LT |
730 | * Fix the length according to mod and prod if given. |
731 | */ | |
81463b1c CS |
732 | args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen) |
733 | - args->agbno; | |
1da177e4 | 734 | xfs_alloc_fix_len(args); |
9f9baab3 CH |
735 | if (!xfs_alloc_fix_minleft(args)) |
736 | goto not_found; | |
737 | ||
81463b1c | 738 | ASSERT(args->agbno + args->len <= tend); |
9f9baab3 | 739 | |
1da177e4 | 740 | /* |
81463b1c | 741 | * We are allocating agbno for args->len |
1da177e4 LT |
742 | * Allocate/initialize a cursor for the by-size btree. |
743 | */ | |
561f7d17 CH |
744 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
745 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 746 | ASSERT(args->agbno + args->len <= |
16259e7d | 747 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
9f9baab3 CH |
748 | error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno, |
749 | args->len, XFSA_FIXUP_BNO_OK); | |
750 | if (error) { | |
1da177e4 LT |
751 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); |
752 | goto error0; | |
753 | } | |
9f9baab3 | 754 | |
1da177e4 LT |
755 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); |
756 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 757 | |
1da177e4 | 758 | args->wasfromfl = 0; |
9f9baab3 CH |
759 | trace_xfs_alloc_exact_done(args); |
760 | return 0; | |
761 | ||
762 | not_found: | |
763 | /* Didn't find it, return null. */ | |
764 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
765 | args->agbno = NULLAGBLOCK; | |
766 | trace_xfs_alloc_exact_notfound(args); | |
1da177e4 LT |
767 | return 0; |
768 | ||
769 | error0: | |
770 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
0b1b213f | 771 | trace_xfs_alloc_exact_error(args); |
1da177e4 LT |
772 | return error; |
773 | } | |
774 | ||
489a150f CH |
775 | /* |
776 | * Search the btree in a given direction via the search cursor and compare | |
777 | * the records found against the good extent we've already found. | |
778 | */ | |
779 | STATIC int | |
780 | xfs_alloc_find_best_extent( | |
781 | struct xfs_alloc_arg *args, /* allocation argument structure */ | |
782 | struct xfs_btree_cur **gcur, /* good cursor */ | |
783 | struct xfs_btree_cur **scur, /* searching cursor */ | |
784 | xfs_agblock_t gdiff, /* difference for search comparison */ | |
785 | xfs_agblock_t *sbno, /* extent found by search */ | |
e26f0501 CH |
786 | xfs_extlen_t *slen, /* extent length */ |
787 | xfs_agblock_t *sbnoa, /* aligned extent found by search */ | |
788 | xfs_extlen_t *slena, /* aligned extent length */ | |
489a150f CH |
789 | int dir) /* 0 = search right, 1 = search left */ |
790 | { | |
489a150f CH |
791 | xfs_agblock_t new; |
792 | xfs_agblock_t sdiff; | |
793 | int error; | |
794 | int i; | |
795 | ||
796 | /* The good extent is perfect, no need to search. */ | |
797 | if (!gdiff) | |
798 | goto out_use_good; | |
799 | ||
800 | /* | |
801 | * Look until we find a better one, run out of space or run off the end. | |
802 | */ | |
803 | do { | |
804 | error = xfs_alloc_get_rec(*scur, sbno, slen, &i); | |
805 | if (error) | |
806 | goto error0; | |
c29aad41 | 807 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
e26f0501 | 808 | xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena); |
489a150f CH |
809 | |
810 | /* | |
811 | * The good extent is closer than this one. | |
812 | */ | |
813 | if (!dir) { | |
bfe46d4e BF |
814 | if (*sbnoa > args->max_agbno) |
815 | goto out_use_good; | |
e26f0501 | 816 | if (*sbnoa >= args->agbno + gdiff) |
489a150f CH |
817 | goto out_use_good; |
818 | } else { | |
bfe46d4e BF |
819 | if (*sbnoa < args->min_agbno) |
820 | goto out_use_good; | |
e26f0501 | 821 | if (*sbnoa <= args->agbno - gdiff) |
489a150f CH |
822 | goto out_use_good; |
823 | } | |
824 | ||
825 | /* | |
826 | * Same distance, compare length and pick the best. | |
827 | */ | |
828 | if (*slena >= args->minlen) { | |
829 | args->len = XFS_EXTLEN_MIN(*slena, args->maxlen); | |
830 | xfs_alloc_fix_len(args); | |
831 | ||
832 | sdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
211d022c JK |
833 | args->alignment, |
834 | args->userdata, *sbnoa, | |
e26f0501 | 835 | *slena, &new); |
489a150f CH |
836 | |
837 | /* | |
838 | * Choose closer size and invalidate other cursor. | |
839 | */ | |
840 | if (sdiff < gdiff) | |
841 | goto out_use_search; | |
842 | goto out_use_good; | |
843 | } | |
844 | ||
845 | if (!dir) | |
846 | error = xfs_btree_increment(*scur, 0, &i); | |
847 | else | |
848 | error = xfs_btree_decrement(*scur, 0, &i); | |
849 | if (error) | |
850 | goto error0; | |
851 | } while (i); | |
852 | ||
853 | out_use_good: | |
854 | xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR); | |
855 | *scur = NULL; | |
856 | return 0; | |
857 | ||
858 | out_use_search: | |
859 | xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR); | |
860 | *gcur = NULL; | |
861 | return 0; | |
862 | ||
863 | error0: | |
864 | /* caller invalidates cursors */ | |
865 | return error; | |
866 | } | |
867 | ||
1da177e4 LT |
868 | /* |
869 | * Allocate a variable extent near bno in the allocation group agno. | |
870 | * Extent's length (returned in len) will be between minlen and maxlen, | |
871 | * and of the form k * prod + mod unless there's nothing that large. | |
872 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
873 | */ | |
874 | STATIC int /* error */ | |
875 | xfs_alloc_ag_vextent_near( | |
876 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
877 | { | |
878 | xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */ | |
879 | xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */ | |
880 | xfs_btree_cur_t *cnt_cur; /* cursor for count btree */ | |
1da177e4 LT |
881 | xfs_agblock_t gtbno; /* start bno of right side entry */ |
882 | xfs_agblock_t gtbnoa; /* aligned ... */ | |
883 | xfs_extlen_t gtdiff; /* difference to right side entry */ | |
884 | xfs_extlen_t gtlen; /* length of right side entry */ | |
e26f0501 | 885 | xfs_extlen_t gtlena; /* aligned ... */ |
1da177e4 LT |
886 | xfs_agblock_t gtnew; /* useful start bno of right side */ |
887 | int error; /* error code */ | |
888 | int i; /* result code, temporary */ | |
889 | int j; /* result code, temporary */ | |
890 | xfs_agblock_t ltbno; /* start bno of left side entry */ | |
891 | xfs_agblock_t ltbnoa; /* aligned ... */ | |
892 | xfs_extlen_t ltdiff; /* difference to left side entry */ | |
1da177e4 | 893 | xfs_extlen_t ltlen; /* length of left side entry */ |
e26f0501 | 894 | xfs_extlen_t ltlena; /* aligned ... */ |
1da177e4 LT |
895 | xfs_agblock_t ltnew; /* useful start bno of left side */ |
896 | xfs_extlen_t rlen; /* length of returned extent */ | |
e26f0501 | 897 | int forced = 0; |
63d20d6e | 898 | #ifdef DEBUG |
1da177e4 LT |
899 | /* |
900 | * Randomly don't execute the first algorithm. | |
901 | */ | |
902 | int dofirst; /* set to do first algorithm */ | |
903 | ||
ecb3403d | 904 | dofirst = prandom_u32() & 1; |
1da177e4 | 905 | #endif |
e26f0501 | 906 | |
bfe46d4e BF |
907 | /* handle unitialized agbno range so caller doesn't have to */ |
908 | if (!args->min_agbno && !args->max_agbno) | |
909 | args->max_agbno = args->mp->m_sb.sb_agblocks - 1; | |
910 | ASSERT(args->min_agbno <= args->max_agbno); | |
911 | ||
912 | /* clamp agbno to the range if it's outside */ | |
913 | if (args->agbno < args->min_agbno) | |
914 | args->agbno = args->min_agbno; | |
915 | if (args->agbno > args->max_agbno) | |
916 | args->agbno = args->max_agbno; | |
917 | ||
e26f0501 CH |
918 | restart: |
919 | bno_cur_lt = NULL; | |
920 | bno_cur_gt = NULL; | |
921 | ltlen = 0; | |
922 | gtlena = 0; | |
923 | ltlena = 0; | |
924 | ||
1da177e4 LT |
925 | /* |
926 | * Get a cursor for the by-size btree. | |
927 | */ | |
561f7d17 CH |
928 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
929 | args->agno, XFS_BTNUM_CNT); | |
e26f0501 | 930 | |
1da177e4 LT |
931 | /* |
932 | * See if there are any free extents as big as maxlen. | |
933 | */ | |
934 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i))) | |
935 | goto error0; | |
936 | /* | |
937 | * If none, then pick up the last entry in the tree unless the | |
938 | * tree is empty. | |
939 | */ | |
940 | if (!i) { | |
941 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno, | |
942 | <len, &i))) | |
943 | goto error0; | |
944 | if (i == 0 || ltlen == 0) { | |
945 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
e26f0501 | 946 | trace_xfs_alloc_near_noentry(args); |
1da177e4 LT |
947 | return 0; |
948 | } | |
949 | ASSERT(i == 1); | |
950 | } | |
951 | args->wasfromfl = 0; | |
e26f0501 | 952 | |
1da177e4 LT |
953 | /* |
954 | * First algorithm. | |
955 | * If the requested extent is large wrt the freespaces available | |
956 | * in this a.g., then the cursor will be pointing to a btree entry | |
957 | * near the right edge of the tree. If it's in the last btree leaf | |
958 | * block, then we just examine all the entries in that block | |
959 | * that are big enough, and pick the best one. | |
960 | * This is written as a while loop so we can break out of it, | |
961 | * but we never loop back to the top. | |
962 | */ | |
963 | while (xfs_btree_islastblock(cnt_cur, 0)) { | |
964 | xfs_extlen_t bdiff; | |
965 | int besti=0; | |
966 | xfs_extlen_t blen=0; | |
967 | xfs_agblock_t bnew=0; | |
968 | ||
63d20d6e DC |
969 | #ifdef DEBUG |
970 | if (dofirst) | |
1da177e4 LT |
971 | break; |
972 | #endif | |
973 | /* | |
974 | * Start from the entry that lookup found, sequence through | |
975 | * all larger free blocks. If we're actually pointing at a | |
976 | * record smaller than maxlen, go to the start of this block, | |
977 | * and skip all those smaller than minlen. | |
978 | */ | |
979 | if (ltlen || args->alignment > 1) { | |
980 | cnt_cur->bc_ptrs[0] = 1; | |
981 | do { | |
982 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, | |
983 | <len, &i))) | |
984 | goto error0; | |
c29aad41 | 985 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 LT |
986 | if (ltlen >= args->minlen) |
987 | break; | |
637aa50f | 988 | if ((error = xfs_btree_increment(cnt_cur, 0, &i))) |
1da177e4 LT |
989 | goto error0; |
990 | } while (i); | |
991 | ASSERT(ltlen >= args->minlen); | |
992 | if (!i) | |
993 | break; | |
994 | } | |
995 | i = cnt_cur->bc_ptrs[0]; | |
996 | for (j = 1, blen = 0, bdiff = 0; | |
997 | !error && j && (blen < args->maxlen || bdiff > 0); | |
637aa50f | 998 | error = xfs_btree_increment(cnt_cur, 0, &j)) { |
1da177e4 LT |
999 | /* |
1000 | * For each entry, decide if it's better than | |
1001 | * the previous best entry. | |
1002 | */ | |
1003 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
1004 | goto error0; | |
c29aad41 | 1005 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
86fa8af6 CH |
1006 | xfs_alloc_compute_aligned(args, ltbno, ltlen, |
1007 | <bnoa, <lena); | |
e6430037 | 1008 | if (ltlena < args->minlen) |
1da177e4 | 1009 | continue; |
bfe46d4e BF |
1010 | if (ltbnoa < args->min_agbno || ltbnoa > args->max_agbno) |
1011 | continue; | |
1da177e4 LT |
1012 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); |
1013 | xfs_alloc_fix_len(args); | |
1014 | ASSERT(args->len >= args->minlen); | |
1015 | if (args->len < blen) | |
1016 | continue; | |
1017 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
211d022c JK |
1018 | args->alignment, args->userdata, ltbnoa, |
1019 | ltlena, <new); | |
1da177e4 LT |
1020 | if (ltnew != NULLAGBLOCK && |
1021 | (args->len > blen || ltdiff < bdiff)) { | |
1022 | bdiff = ltdiff; | |
1023 | bnew = ltnew; | |
1024 | blen = args->len; | |
1025 | besti = cnt_cur->bc_ptrs[0]; | |
1026 | } | |
1027 | } | |
1028 | /* | |
1029 | * It didn't work. We COULD be in a case where | |
1030 | * there's a good record somewhere, so try again. | |
1031 | */ | |
1032 | if (blen == 0) | |
1033 | break; | |
1034 | /* | |
1035 | * Point at the best entry, and retrieve it again. | |
1036 | */ | |
1037 | cnt_cur->bc_ptrs[0] = besti; | |
1038 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
1039 | goto error0; | |
c29aad41 | 1040 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
73523a2e | 1041 | ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
1042 | args->len = blen; |
1043 | if (!xfs_alloc_fix_minleft(args)) { | |
1044 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 1045 | trace_xfs_alloc_near_nominleft(args); |
1da177e4 LT |
1046 | return 0; |
1047 | } | |
1048 | blen = args->len; | |
1049 | /* | |
1050 | * We are allocating starting at bnew for blen blocks. | |
1051 | */ | |
1052 | args->agbno = bnew; | |
1053 | ASSERT(bnew >= ltbno); | |
73523a2e | 1054 | ASSERT(bnew + blen <= ltbno + ltlen); |
1da177e4 LT |
1055 | /* |
1056 | * Set up a cursor for the by-bno tree. | |
1057 | */ | |
561f7d17 CH |
1058 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, |
1059 | args->agbp, args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1060 | /* |
1061 | * Fix up the btree entries. | |
1062 | */ | |
1063 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, | |
1064 | ltlen, bnew, blen, XFSA_FIXUP_CNT_OK))) | |
1065 | goto error0; | |
1066 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1067 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
0b1b213f CH |
1068 | |
1069 | trace_xfs_alloc_near_first(args); | |
1da177e4 LT |
1070 | return 0; |
1071 | } | |
1072 | /* | |
1073 | * Second algorithm. | |
1074 | * Search in the by-bno tree to the left and to the right | |
1075 | * simultaneously, until in each case we find a space big enough, | |
1076 | * or run into the edge of the tree. When we run into the edge, | |
1077 | * we deallocate that cursor. | |
1078 | * If both searches succeed, we compare the two spaces and pick | |
1079 | * the better one. | |
1080 | * With alignment, it's possible for both to fail; the upper | |
1081 | * level algorithm that picks allocation groups for allocations | |
1082 | * is not supposed to do this. | |
1083 | */ | |
1084 | /* | |
1085 | * Allocate and initialize the cursor for the leftward search. | |
1086 | */ | |
561f7d17 CH |
1087 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1088 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1089 | /* |
1090 | * Lookup <= bno to find the leftward search's starting point. | |
1091 | */ | |
1092 | if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i))) | |
1093 | goto error0; | |
1094 | if (!i) { | |
1095 | /* | |
1096 | * Didn't find anything; use this cursor for the rightward | |
1097 | * search. | |
1098 | */ | |
1099 | bno_cur_gt = bno_cur_lt; | |
1100 | bno_cur_lt = NULL; | |
1101 | } | |
1102 | /* | |
1103 | * Found something. Duplicate the cursor for the rightward search. | |
1104 | */ | |
1105 | else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt))) | |
1106 | goto error0; | |
1107 | /* | |
1108 | * Increment the cursor, so we will point at the entry just right | |
1109 | * of the leftward entry if any, or to the leftmost entry. | |
1110 | */ | |
637aa50f | 1111 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
1112 | goto error0; |
1113 | if (!i) { | |
1114 | /* | |
1115 | * It failed, there are no rightward entries. | |
1116 | */ | |
1117 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR); | |
1118 | bno_cur_gt = NULL; | |
1119 | } | |
1120 | /* | |
1121 | * Loop going left with the leftward cursor, right with the | |
1122 | * rightward cursor, until either both directions give up or | |
1123 | * we find an entry at least as big as minlen. | |
1124 | */ | |
1125 | do { | |
1126 | if (bno_cur_lt) { | |
1127 | if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i))) | |
1128 | goto error0; | |
c29aad41 | 1129 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
86fa8af6 CH |
1130 | xfs_alloc_compute_aligned(args, ltbno, ltlen, |
1131 | <bnoa, <lena); | |
bfe46d4e | 1132 | if (ltlena >= args->minlen && ltbnoa >= args->min_agbno) |
1da177e4 | 1133 | break; |
8df4da4a | 1134 | if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i))) |
1da177e4 | 1135 | goto error0; |
bfe46d4e | 1136 | if (!i || ltbnoa < args->min_agbno) { |
1da177e4 LT |
1137 | xfs_btree_del_cursor(bno_cur_lt, |
1138 | XFS_BTREE_NOERROR); | |
1139 | bno_cur_lt = NULL; | |
1140 | } | |
1141 | } | |
1142 | if (bno_cur_gt) { | |
1143 | if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i))) | |
1144 | goto error0; | |
c29aad41 | 1145 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
86fa8af6 CH |
1146 | xfs_alloc_compute_aligned(args, gtbno, gtlen, |
1147 | >bnoa, >lena); | |
bfe46d4e | 1148 | if (gtlena >= args->minlen && gtbnoa <= args->max_agbno) |
1da177e4 | 1149 | break; |
637aa50f | 1150 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 | 1151 | goto error0; |
bfe46d4e | 1152 | if (!i || gtbnoa > args->max_agbno) { |
1da177e4 LT |
1153 | xfs_btree_del_cursor(bno_cur_gt, |
1154 | XFS_BTREE_NOERROR); | |
1155 | bno_cur_gt = NULL; | |
1156 | } | |
1157 | } | |
1158 | } while (bno_cur_lt || bno_cur_gt); | |
489a150f | 1159 | |
1da177e4 LT |
1160 | /* |
1161 | * Got both cursors still active, need to find better entry. | |
1162 | */ | |
1163 | if (bno_cur_lt && bno_cur_gt) { | |
1da177e4 LT |
1164 | if (ltlena >= args->minlen) { |
1165 | /* | |
489a150f | 1166 | * Left side is good, look for a right side entry. |
1da177e4 LT |
1167 | */ |
1168 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1169 | xfs_alloc_fix_len(args); | |
489a150f | 1170 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, |
211d022c JK |
1171 | args->alignment, args->userdata, ltbnoa, |
1172 | ltlena, <new); | |
489a150f CH |
1173 | |
1174 | error = xfs_alloc_find_best_extent(args, | |
1175 | &bno_cur_lt, &bno_cur_gt, | |
e26f0501 CH |
1176 | ltdiff, >bno, >len, |
1177 | >bnoa, >lena, | |
489a150f CH |
1178 | 0 /* search right */); |
1179 | } else { | |
1180 | ASSERT(gtlena >= args->minlen); | |
1181 | ||
1da177e4 | 1182 | /* |
489a150f | 1183 | * Right side is good, look for a left side entry. |
1da177e4 LT |
1184 | */ |
1185 | args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen); | |
1186 | xfs_alloc_fix_len(args); | |
489a150f | 1187 | gtdiff = xfs_alloc_compute_diff(args->agbno, args->len, |
211d022c JK |
1188 | args->alignment, args->userdata, gtbnoa, |
1189 | gtlena, >new); | |
489a150f CH |
1190 | |
1191 | error = xfs_alloc_find_best_extent(args, | |
1192 | &bno_cur_gt, &bno_cur_lt, | |
e26f0501 CH |
1193 | gtdiff, <bno, <len, |
1194 | <bnoa, <lena, | |
489a150f | 1195 | 1 /* search left */); |
1da177e4 | 1196 | } |
489a150f CH |
1197 | |
1198 | if (error) | |
1199 | goto error0; | |
1da177e4 | 1200 | } |
489a150f | 1201 | |
1da177e4 LT |
1202 | /* |
1203 | * If we couldn't get anything, give up. | |
1204 | */ | |
1205 | if (bno_cur_lt == NULL && bno_cur_gt == NULL) { | |
e3a746f5 DC |
1206 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1207 | ||
e26f0501 CH |
1208 | if (!forced++) { |
1209 | trace_xfs_alloc_near_busy(args); | |
1210 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1211 | goto restart; | |
1212 | } | |
0b1b213f | 1213 | trace_xfs_alloc_size_neither(args); |
1da177e4 LT |
1214 | args->agbno = NULLAGBLOCK; |
1215 | return 0; | |
1216 | } | |
489a150f | 1217 | |
1da177e4 LT |
1218 | /* |
1219 | * At this point we have selected a freespace entry, either to the | |
1220 | * left or to the right. If it's on the right, copy all the | |
1221 | * useful variables to the "left" set so we only have one | |
1222 | * copy of this code. | |
1223 | */ | |
1224 | if (bno_cur_gt) { | |
1225 | bno_cur_lt = bno_cur_gt; | |
1226 | bno_cur_gt = NULL; | |
1227 | ltbno = gtbno; | |
1228 | ltbnoa = gtbnoa; | |
1229 | ltlen = gtlen; | |
1230 | ltlena = gtlena; | |
1231 | j = 1; | |
1232 | } else | |
1233 | j = 0; | |
489a150f | 1234 | |
1da177e4 LT |
1235 | /* |
1236 | * Fix up the length and compute the useful address. | |
1237 | */ | |
1da177e4 LT |
1238 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); |
1239 | xfs_alloc_fix_len(args); | |
1240 | if (!xfs_alloc_fix_minleft(args)) { | |
0b1b213f | 1241 | trace_xfs_alloc_near_nominleft(args); |
1da177e4 LT |
1242 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); |
1243 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1244 | return 0; | |
1245 | } | |
1246 | rlen = args->len; | |
e26f0501 | 1247 | (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, |
211d022c | 1248 | args->userdata, ltbnoa, ltlena, <new); |
1da177e4 | 1249 | ASSERT(ltnew >= ltbno); |
e26f0501 | 1250 | ASSERT(ltnew + rlen <= ltbnoa + ltlena); |
16259e7d | 1251 | ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
bfe46d4e | 1252 | ASSERT(ltnew >= args->min_agbno && ltnew <= args->max_agbno); |
1da177e4 | 1253 | args->agbno = ltnew; |
e26f0501 | 1254 | |
1da177e4 LT |
1255 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen, |
1256 | ltnew, rlen, XFSA_FIXUP_BNO_OK))) | |
1257 | goto error0; | |
0b1b213f CH |
1258 | |
1259 | if (j) | |
1260 | trace_xfs_alloc_near_greater(args); | |
1261 | else | |
1262 | trace_xfs_alloc_near_lesser(args); | |
1263 | ||
1da177e4 LT |
1264 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1265 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1266 | return 0; | |
1267 | ||
1268 | error0: | |
0b1b213f | 1269 | trace_xfs_alloc_near_error(args); |
1da177e4 LT |
1270 | if (cnt_cur != NULL) |
1271 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1272 | if (bno_cur_lt != NULL) | |
1273 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR); | |
1274 | if (bno_cur_gt != NULL) | |
1275 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR); | |
1276 | return error; | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Allocate a variable extent anywhere in the allocation group agno. | |
1281 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1282 | * and of the form k * prod + mod unless there's nothing that large. | |
1283 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1284 | */ | |
1285 | STATIC int /* error */ | |
1286 | xfs_alloc_ag_vextent_size( | |
1287 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
1288 | { | |
1289 | xfs_btree_cur_t *bno_cur; /* cursor for bno btree */ | |
1290 | xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */ | |
1291 | int error; /* error result */ | |
1292 | xfs_agblock_t fbno; /* start of found freespace */ | |
1293 | xfs_extlen_t flen; /* length of found freespace */ | |
1da177e4 LT |
1294 | int i; /* temp status variable */ |
1295 | xfs_agblock_t rbno; /* returned block number */ | |
1296 | xfs_extlen_t rlen; /* length of returned extent */ | |
e26f0501 | 1297 | int forced = 0; |
1da177e4 | 1298 | |
e26f0501 | 1299 | restart: |
1da177e4 LT |
1300 | /* |
1301 | * Allocate and initialize a cursor for the by-size btree. | |
1302 | */ | |
561f7d17 CH |
1303 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1304 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 1305 | bno_cur = NULL; |
e26f0501 | 1306 | |
1da177e4 LT |
1307 | /* |
1308 | * Look for an entry >= maxlen+alignment-1 blocks. | |
1309 | */ | |
1310 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, | |
1311 | args->maxlen + args->alignment - 1, &i))) | |
1312 | goto error0; | |
e26f0501 | 1313 | |
1da177e4 | 1314 | /* |
e26f0501 CH |
1315 | * If none or we have busy extents that we cannot allocate from, then |
1316 | * we have to settle for a smaller extent. In the case that there are | |
1317 | * no large extents, this will return the last entry in the tree unless | |
1318 | * the tree is empty. In the case that there are only busy large | |
1319 | * extents, this will return the largest small extent unless there | |
1320 | * are no smaller extents available. | |
1da177e4 | 1321 | */ |
e26f0501 CH |
1322 | if (!i || forced > 1) { |
1323 | error = xfs_alloc_ag_vextent_small(args, cnt_cur, | |
1324 | &fbno, &flen, &i); | |
1325 | if (error) | |
1da177e4 LT |
1326 | goto error0; |
1327 | if (i == 0 || flen == 0) { | |
1328 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 1329 | trace_xfs_alloc_size_noentry(args); |
1da177e4 LT |
1330 | return 0; |
1331 | } | |
1332 | ASSERT(i == 1); | |
e26f0501 CH |
1333 | xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen); |
1334 | } else { | |
1335 | /* | |
1336 | * Search for a non-busy extent that is large enough. | |
1337 | * If we are at low space, don't check, or if we fall of | |
1338 | * the end of the btree, turn off the busy check and | |
1339 | * restart. | |
1340 | */ | |
1341 | for (;;) { | |
1342 | error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i); | |
1343 | if (error) | |
1344 | goto error0; | |
c29aad41 | 1345 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
e26f0501 CH |
1346 | |
1347 | xfs_alloc_compute_aligned(args, fbno, flen, | |
1348 | &rbno, &rlen); | |
1349 | ||
1350 | if (rlen >= args->maxlen) | |
1351 | break; | |
1352 | ||
1353 | error = xfs_btree_increment(cnt_cur, 0, &i); | |
1354 | if (error) | |
1355 | goto error0; | |
1356 | if (i == 0) { | |
1357 | /* | |
1358 | * Our only valid extents must have been busy. | |
1359 | * Make it unbusy by forcing the log out and | |
1360 | * retrying. If we've been here before, forcing | |
1361 | * the log isn't making the extents available, | |
1362 | * which means they have probably been freed in | |
1363 | * this transaction. In that case, we have to | |
1364 | * give up on them and we'll attempt a minlen | |
1365 | * allocation the next time around. | |
1366 | */ | |
1367 | xfs_btree_del_cursor(cnt_cur, | |
1368 | XFS_BTREE_NOERROR); | |
1369 | trace_xfs_alloc_size_busy(args); | |
1370 | if (!forced++) | |
1371 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1372 | goto restart; | |
1373 | } | |
1374 | } | |
1da177e4 | 1375 | } |
e26f0501 | 1376 | |
1da177e4 LT |
1377 | /* |
1378 | * In the first case above, we got the last entry in the | |
1379 | * by-size btree. Now we check to see if the space hits maxlen | |
1380 | * once aligned; if not, we search left for something better. | |
1381 | * This can't happen in the second case above. | |
1382 | */ | |
1da177e4 | 1383 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
c29aad41 | 1384 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen == 0 || |
1da177e4 LT |
1385 | (rlen <= flen && rbno + rlen <= fbno + flen), error0); |
1386 | if (rlen < args->maxlen) { | |
1387 | xfs_agblock_t bestfbno; | |
1388 | xfs_extlen_t bestflen; | |
1389 | xfs_agblock_t bestrbno; | |
1390 | xfs_extlen_t bestrlen; | |
1391 | ||
1392 | bestrlen = rlen; | |
1393 | bestrbno = rbno; | |
1394 | bestflen = flen; | |
1395 | bestfbno = fbno; | |
1396 | for (;;) { | |
8df4da4a | 1397 | if ((error = xfs_btree_decrement(cnt_cur, 0, &i))) |
1da177e4 LT |
1398 | goto error0; |
1399 | if (i == 0) | |
1400 | break; | |
1401 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, | |
1402 | &i))) | |
1403 | goto error0; | |
c29aad41 | 1404 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 LT |
1405 | if (flen < bestrlen) |
1406 | break; | |
86fa8af6 CH |
1407 | xfs_alloc_compute_aligned(args, fbno, flen, |
1408 | &rbno, &rlen); | |
1da177e4 | 1409 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
c29aad41 | 1410 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen == 0 || |
1da177e4 LT |
1411 | (rlen <= flen && rbno + rlen <= fbno + flen), |
1412 | error0); | |
1413 | if (rlen > bestrlen) { | |
1414 | bestrlen = rlen; | |
1415 | bestrbno = rbno; | |
1416 | bestflen = flen; | |
1417 | bestfbno = fbno; | |
1418 | if (rlen == args->maxlen) | |
1419 | break; | |
1420 | } | |
1421 | } | |
1422 | if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen, | |
1423 | &i))) | |
1424 | goto error0; | |
c29aad41 | 1425 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 LT |
1426 | rlen = bestrlen; |
1427 | rbno = bestrbno; | |
1428 | flen = bestflen; | |
1429 | fbno = bestfbno; | |
1430 | } | |
1431 | args->wasfromfl = 0; | |
1432 | /* | |
1433 | * Fix up the length. | |
1434 | */ | |
1435 | args->len = rlen; | |
e26f0501 CH |
1436 | if (rlen < args->minlen) { |
1437 | if (!forced++) { | |
1438 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1439 | trace_xfs_alloc_size_busy(args); | |
1440 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1441 | goto restart; | |
1442 | } | |
1443 | goto out_nominleft; | |
1da177e4 | 1444 | } |
e26f0501 CH |
1445 | xfs_alloc_fix_len(args); |
1446 | ||
1447 | if (!xfs_alloc_fix_minleft(args)) | |
1448 | goto out_nominleft; | |
1da177e4 | 1449 | rlen = args->len; |
c29aad41 | 1450 | XFS_WANT_CORRUPTED_GOTO(args->mp, rlen <= flen, error0); |
1da177e4 LT |
1451 | /* |
1452 | * Allocate and initialize a cursor for the by-block tree. | |
1453 | */ | |
561f7d17 CH |
1454 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1455 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1456 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
1457 | rbno, rlen, XFSA_FIXUP_CNT_OK))) | |
1458 | goto error0; | |
1459 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1460 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1461 | cnt_cur = bno_cur = NULL; | |
1462 | args->len = rlen; | |
1463 | args->agbno = rbno; | |
c29aad41 | 1464 | XFS_WANT_CORRUPTED_GOTO(args->mp, |
1da177e4 | 1465 | args->agbno + args->len <= |
16259e7d | 1466 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 | 1467 | error0); |
0b1b213f | 1468 | trace_xfs_alloc_size_done(args); |
1da177e4 LT |
1469 | return 0; |
1470 | ||
1471 | error0: | |
0b1b213f | 1472 | trace_xfs_alloc_size_error(args); |
1da177e4 LT |
1473 | if (cnt_cur) |
1474 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1475 | if (bno_cur) | |
1476 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1477 | return error; | |
e26f0501 CH |
1478 | |
1479 | out_nominleft: | |
1480 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1481 | trace_xfs_alloc_size_nominleft(args); | |
1482 | args->agbno = NULLAGBLOCK; | |
1483 | return 0; | |
1da177e4 LT |
1484 | } |
1485 | ||
1486 | /* | |
1487 | * Deal with the case where only small freespaces remain. | |
1488 | * Either return the contents of the last freespace record, | |
1489 | * or allocate space from the freelist if there is nothing in the tree. | |
1490 | */ | |
1491 | STATIC int /* error */ | |
1492 | xfs_alloc_ag_vextent_small( | |
1493 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1494 | xfs_btree_cur_t *ccur, /* by-size cursor */ | |
1495 | xfs_agblock_t *fbnop, /* result block number */ | |
1496 | xfs_extlen_t *flenp, /* result length */ | |
1497 | int *stat) /* status: 0-freelist, 1-normal/none */ | |
1498 | { | |
1499 | int error; | |
1500 | xfs_agblock_t fbno; | |
1501 | xfs_extlen_t flen; | |
1da177e4 LT |
1502 | int i; |
1503 | ||
8df4da4a | 1504 | if ((error = xfs_btree_decrement(ccur, 0, &i))) |
1da177e4 LT |
1505 | goto error0; |
1506 | if (i) { | |
1507 | if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i))) | |
1508 | goto error0; | |
c29aad41 | 1509 | XFS_WANT_CORRUPTED_GOTO(args->mp, i == 1, error0); |
1da177e4 LT |
1510 | } |
1511 | /* | |
1512 | * Nothing in the btree, try the freelist. Make sure | |
1513 | * to respect minleft even when pulling from the | |
1514 | * freelist. | |
1515 | */ | |
1516 | else if (args->minlen == 1 && args->alignment == 1 && !args->isfl && | |
16259e7d CH |
1517 | (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount) |
1518 | > args->minleft)) { | |
92821e2b DC |
1519 | error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0); |
1520 | if (error) | |
1da177e4 LT |
1521 | goto error0; |
1522 | if (fbno != NULLAGBLOCK) { | |
4ecbfe63 | 1523 | xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1, |
97d3ac75 CH |
1524 | args->userdata); |
1525 | ||
1da177e4 LT |
1526 | if (args->userdata) { |
1527 | xfs_buf_t *bp; | |
1528 | ||
1529 | bp = xfs_btree_get_bufs(args->mp, args->tp, | |
1530 | args->agno, fbno, 0); | |
1531 | xfs_trans_binval(args->tp, bp); | |
1532 | } | |
1533 | args->len = 1; | |
1534 | args->agbno = fbno; | |
c29aad41 | 1535 | XFS_WANT_CORRUPTED_GOTO(args->mp, |
1da177e4 | 1536 | args->agbno + args->len <= |
16259e7d | 1537 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1538 | error0); |
1539 | args->wasfromfl = 1; | |
0b1b213f | 1540 | trace_xfs_alloc_small_freelist(args); |
1da177e4 LT |
1541 | *stat = 0; |
1542 | return 0; | |
1543 | } | |
1544 | /* | |
1545 | * Nothing in the freelist. | |
1546 | */ | |
1547 | else | |
1548 | flen = 0; | |
1549 | } | |
1550 | /* | |
1551 | * Can't allocate from the freelist for some reason. | |
1552 | */ | |
d432c80e NS |
1553 | else { |
1554 | fbno = NULLAGBLOCK; | |
1da177e4 | 1555 | flen = 0; |
d432c80e | 1556 | } |
1da177e4 LT |
1557 | /* |
1558 | * Can't do the allocation, give up. | |
1559 | */ | |
1560 | if (flen < args->minlen) { | |
1561 | args->agbno = NULLAGBLOCK; | |
0b1b213f | 1562 | trace_xfs_alloc_small_notenough(args); |
1da177e4 LT |
1563 | flen = 0; |
1564 | } | |
1565 | *fbnop = fbno; | |
1566 | *flenp = flen; | |
1567 | *stat = 1; | |
0b1b213f | 1568 | trace_xfs_alloc_small_done(args); |
1da177e4 LT |
1569 | return 0; |
1570 | ||
1571 | error0: | |
0b1b213f | 1572 | trace_xfs_alloc_small_error(args); |
1da177e4 LT |
1573 | return error; |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * Free the extent starting at agno/bno for length. | |
1578 | */ | |
1579 | STATIC int /* error */ | |
1580 | xfs_free_ag_extent( | |
1581 | xfs_trans_t *tp, /* transaction pointer */ | |
1582 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
1583 | xfs_agnumber_t agno, /* allocation group number */ | |
1584 | xfs_agblock_t bno, /* starting block number */ | |
1585 | xfs_extlen_t len, /* length of extent */ | |
1586 | int isfl) /* set if is freelist blocks - no sb acctg */ | |
1587 | { | |
1588 | xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */ | |
1589 | xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */ | |
1590 | int error; /* error return value */ | |
1da177e4 LT |
1591 | xfs_agblock_t gtbno; /* start of right neighbor block */ |
1592 | xfs_extlen_t gtlen; /* length of right neighbor block */ | |
1593 | int haveleft; /* have a left neighbor block */ | |
1594 | int haveright; /* have a right neighbor block */ | |
1595 | int i; /* temp, result code */ | |
1596 | xfs_agblock_t ltbno; /* start of left neighbor block */ | |
1597 | xfs_extlen_t ltlen; /* length of left neighbor block */ | |
1598 | xfs_mount_t *mp; /* mount point struct for filesystem */ | |
1599 | xfs_agblock_t nbno; /* new starting block of freespace */ | |
1600 | xfs_extlen_t nlen; /* new length of freespace */ | |
ecb6928f | 1601 | xfs_perag_t *pag; /* per allocation group data */ |
1da177e4 LT |
1602 | |
1603 | mp = tp->t_mountp; | |
1604 | /* | |
1605 | * Allocate and initialize a cursor for the by-block btree. | |
1606 | */ | |
561f7d17 | 1607 | bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO); |
1da177e4 LT |
1608 | cnt_cur = NULL; |
1609 | /* | |
1610 | * Look for a neighboring block on the left (lower block numbers) | |
1611 | * that is contiguous with this space. | |
1612 | */ | |
1613 | if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft))) | |
1614 | goto error0; | |
1615 | if (haveleft) { | |
1616 | /* | |
1617 | * There is a block to our left. | |
1618 | */ | |
1619 | if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i))) | |
1620 | goto error0; | |
c29aad41 | 1621 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1622 | /* |
1623 | * It's not contiguous, though. | |
1624 | */ | |
1625 | if (ltbno + ltlen < bno) | |
1626 | haveleft = 0; | |
1627 | else { | |
1628 | /* | |
1629 | * If this failure happens the request to free this | |
1630 | * space was invalid, it's (partly) already free. | |
1631 | * Very bad. | |
1632 | */ | |
c29aad41 ES |
1633 | XFS_WANT_CORRUPTED_GOTO(mp, |
1634 | ltbno + ltlen <= bno, error0); | |
1da177e4 LT |
1635 | } |
1636 | } | |
1637 | /* | |
1638 | * Look for a neighboring block on the right (higher block numbers) | |
1639 | * that is contiguous with this space. | |
1640 | */ | |
637aa50f | 1641 | if ((error = xfs_btree_increment(bno_cur, 0, &haveright))) |
1da177e4 LT |
1642 | goto error0; |
1643 | if (haveright) { | |
1644 | /* | |
1645 | * There is a block to our right. | |
1646 | */ | |
1647 | if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i))) | |
1648 | goto error0; | |
c29aad41 | 1649 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1650 | /* |
1651 | * It's not contiguous, though. | |
1652 | */ | |
1653 | if (bno + len < gtbno) | |
1654 | haveright = 0; | |
1655 | else { | |
1656 | /* | |
1657 | * If this failure happens the request to free this | |
1658 | * space was invalid, it's (partly) already free. | |
1659 | * Very bad. | |
1660 | */ | |
c29aad41 | 1661 | XFS_WANT_CORRUPTED_GOTO(mp, gtbno >= bno + len, error0); |
1da177e4 LT |
1662 | } |
1663 | } | |
1664 | /* | |
1665 | * Now allocate and initialize a cursor for the by-size tree. | |
1666 | */ | |
561f7d17 | 1667 | cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT); |
1da177e4 LT |
1668 | /* |
1669 | * Have both left and right contiguous neighbors. | |
1670 | * Merge all three into a single free block. | |
1671 | */ | |
1672 | if (haveleft && haveright) { | |
1673 | /* | |
1674 | * Delete the old by-size entry on the left. | |
1675 | */ | |
1676 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1677 | goto error0; | |
c29aad41 | 1678 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
91cca5df | 1679 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1680 | goto error0; |
c29aad41 | 1681 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1682 | /* |
1683 | * Delete the old by-size entry on the right. | |
1684 | */ | |
1685 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1686 | goto error0; | |
c29aad41 | 1687 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
91cca5df | 1688 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1689 | goto error0; |
c29aad41 | 1690 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1691 | /* |
1692 | * Delete the old by-block entry for the right block. | |
1693 | */ | |
91cca5df | 1694 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 | 1695 | goto error0; |
c29aad41 | 1696 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1697 | /* |
1698 | * Move the by-block cursor back to the left neighbor. | |
1699 | */ | |
8df4da4a | 1700 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 | 1701 | goto error0; |
c29aad41 | 1702 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1703 | #ifdef DEBUG |
1704 | /* | |
1705 | * Check that this is the right record: delete didn't | |
1706 | * mangle the cursor. | |
1707 | */ | |
1708 | { | |
1709 | xfs_agblock_t xxbno; | |
1710 | xfs_extlen_t xxlen; | |
1711 | ||
1712 | if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen, | |
1713 | &i))) | |
1714 | goto error0; | |
c29aad41 | 1715 | XFS_WANT_CORRUPTED_GOTO(mp, |
1da177e4 LT |
1716 | i == 1 && xxbno == ltbno && xxlen == ltlen, |
1717 | error0); | |
1718 | } | |
1719 | #endif | |
1720 | /* | |
1721 | * Update remaining by-block entry to the new, joined block. | |
1722 | */ | |
1723 | nbno = ltbno; | |
1724 | nlen = len + ltlen + gtlen; | |
1725 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1726 | goto error0; | |
1727 | } | |
1728 | /* | |
1729 | * Have only a left contiguous neighbor. | |
1730 | * Merge it together with the new freespace. | |
1731 | */ | |
1732 | else if (haveleft) { | |
1733 | /* | |
1734 | * Delete the old by-size entry on the left. | |
1735 | */ | |
1736 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1737 | goto error0; | |
c29aad41 | 1738 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
91cca5df | 1739 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1740 | goto error0; |
c29aad41 | 1741 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1742 | /* |
1743 | * Back up the by-block cursor to the left neighbor, and | |
1744 | * update its length. | |
1745 | */ | |
8df4da4a | 1746 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 | 1747 | goto error0; |
c29aad41 | 1748 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1749 | nbno = ltbno; |
1750 | nlen = len + ltlen; | |
1751 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1752 | goto error0; | |
1753 | } | |
1754 | /* | |
1755 | * Have only a right contiguous neighbor. | |
1756 | * Merge it together with the new freespace. | |
1757 | */ | |
1758 | else if (haveright) { | |
1759 | /* | |
1760 | * Delete the old by-size entry on the right. | |
1761 | */ | |
1762 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1763 | goto error0; | |
c29aad41 | 1764 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
91cca5df | 1765 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 | 1766 | goto error0; |
c29aad41 | 1767 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1768 | /* |
1769 | * Update the starting block and length of the right | |
1770 | * neighbor in the by-block tree. | |
1771 | */ | |
1772 | nbno = bno; | |
1773 | nlen = len + gtlen; | |
1774 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1775 | goto error0; | |
1776 | } | |
1777 | /* | |
1778 | * No contiguous neighbors. | |
1779 | * Insert the new freespace into the by-block tree. | |
1780 | */ | |
1781 | else { | |
1782 | nbno = bno; | |
1783 | nlen = len; | |
4b22a571 | 1784 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 | 1785 | goto error0; |
c29aad41 | 1786 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1787 | } |
1788 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1789 | bno_cur = NULL; | |
1790 | /* | |
1791 | * In all cases we need to insert the new freespace in the by-size tree. | |
1792 | */ | |
1793 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i))) | |
1794 | goto error0; | |
c29aad41 | 1795 | XFS_WANT_CORRUPTED_GOTO(mp, i == 0, error0); |
4b22a571 | 1796 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 | 1797 | goto error0; |
c29aad41 | 1798 | XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0); |
1da177e4 LT |
1799 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1800 | cnt_cur = NULL; | |
ecb6928f | 1801 | |
1da177e4 LT |
1802 | /* |
1803 | * Update the freespace totals in the ag and superblock. | |
1804 | */ | |
ecb6928f CH |
1805 | pag = xfs_perag_get(mp, agno); |
1806 | error = xfs_alloc_update_counters(tp, pag, agbp, len); | |
1807 | xfs_perag_put(pag); | |
1808 | if (error) | |
1809 | goto error0; | |
1810 | ||
1811 | if (!isfl) | |
1812 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len); | |
ff6d6af2 BD |
1813 | XFS_STATS_INC(mp, xs_freex); |
1814 | XFS_STATS_ADD(mp, xs_freeb, len); | |
0b1b213f CH |
1815 | |
1816 | trace_xfs_free_extent(mp, agno, bno, len, isfl, haveleft, haveright); | |
1da177e4 | 1817 | |
1da177e4 LT |
1818 | return 0; |
1819 | ||
1820 | error0: | |
0b1b213f | 1821 | trace_xfs_free_extent(mp, agno, bno, len, isfl, -1, -1); |
1da177e4 LT |
1822 | if (bno_cur) |
1823 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1824 | if (cnt_cur) | |
1825 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1826 | return error; | |
1827 | } | |
1828 | ||
1829 | /* | |
1830 | * Visible (exported) allocation/free functions. | |
1831 | * Some of these are used just by xfs_alloc_btree.c and this file. | |
1832 | */ | |
1833 | ||
1834 | /* | |
1835 | * Compute and fill in value of m_ag_maxlevels. | |
1836 | */ | |
1837 | void | |
1838 | xfs_alloc_compute_maxlevels( | |
1839 | xfs_mount_t *mp) /* file system mount structure */ | |
1840 | { | |
1841 | int level; | |
1842 | uint maxblocks; | |
1843 | uint maxleafents; | |
1844 | int minleafrecs; | |
1845 | int minnoderecs; | |
1846 | ||
1847 | maxleafents = (mp->m_sb.sb_agblocks + 1) / 2; | |
1848 | minleafrecs = mp->m_alloc_mnr[0]; | |
1849 | minnoderecs = mp->m_alloc_mnr[1]; | |
1850 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1851 | for (level = 1; maxblocks > 1; level++) | |
1852 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1853 | mp->m_ag_maxlevels = level; | |
1854 | } | |
1855 | ||
6cc87645 DC |
1856 | /* |
1857 | * Find the length of the longest extent in an AG. | |
1858 | */ | |
1859 | xfs_extlen_t | |
1860 | xfs_alloc_longest_free_extent( | |
1861 | struct xfs_mount *mp, | |
50adbcb4 DC |
1862 | struct xfs_perag *pag, |
1863 | xfs_extlen_t need) | |
6cc87645 | 1864 | { |
50adbcb4 | 1865 | xfs_extlen_t delta = 0; |
6cc87645 | 1866 | |
6cc87645 DC |
1867 | if (need > pag->pagf_flcount) |
1868 | delta = need - pag->pagf_flcount; | |
1869 | ||
1870 | if (pag->pagf_longest > delta) | |
1871 | return pag->pagf_longest - delta; | |
1872 | return pag->pagf_flcount > 0 || pag->pagf_longest > 0; | |
1873 | } | |
1874 | ||
496817b4 DC |
1875 | unsigned int |
1876 | xfs_alloc_min_freelist( | |
1877 | struct xfs_mount *mp, | |
1878 | struct xfs_perag *pag) | |
1879 | { | |
1880 | unsigned int min_free; | |
1881 | ||
1882 | /* space needed by-bno freespace btree */ | |
1883 | min_free = min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_BNOi] + 1, | |
1884 | mp->m_ag_maxlevels); | |
1885 | /* space needed by-size freespace btree */ | |
1886 | min_free += min_t(unsigned int, pag->pagf_levels[XFS_BTNUM_CNTi] + 1, | |
1887 | mp->m_ag_maxlevels); | |
1888 | ||
1889 | return min_free; | |
1890 | } | |
1891 | ||
72d55285 DC |
1892 | /* |
1893 | * Check if the operation we are fixing up the freelist for should go ahead or | |
1894 | * not. If we are freeing blocks, we always allow it, otherwise the allocation | |
1895 | * is dependent on whether the size and shape of free space available will | |
1896 | * permit the requested allocation to take place. | |
1897 | */ | |
1898 | static bool | |
1899 | xfs_alloc_space_available( | |
1900 | struct xfs_alloc_arg *args, | |
1901 | xfs_extlen_t min_free, | |
1902 | int flags) | |
1903 | { | |
1904 | struct xfs_perag *pag = args->pag; | |
1905 | xfs_extlen_t longest; | |
1906 | int available; | |
1907 | ||
1908 | if (flags & XFS_ALLOC_FLAG_FREEING) | |
1909 | return true; | |
1910 | ||
1911 | /* do we have enough contiguous free space for the allocation? */ | |
1912 | longest = xfs_alloc_longest_free_extent(args->mp, pag, min_free); | |
1913 | if ((args->minlen + args->alignment + args->minalignslop - 1) > longest) | |
1914 | return false; | |
1915 | ||
1916 | /* do have enough free space remaining for the allocation? */ | |
1917 | available = (int)(pag->pagf_freeblks + pag->pagf_flcount - | |
1918 | min_free - args->total); | |
1919 | if (available < (int)args->minleft) | |
1920 | return false; | |
1921 | ||
1922 | return true; | |
1923 | } | |
1924 | ||
1da177e4 LT |
1925 | /* |
1926 | * Decide whether to use this allocation group for this allocation. | |
1927 | * If so, fix up the btree freelist's size. | |
1928 | */ | |
1929 | STATIC int /* error */ | |
1930 | xfs_alloc_fix_freelist( | |
396503fc DC |
1931 | struct xfs_alloc_arg *args, /* allocation argument structure */ |
1932 | int flags) /* XFS_ALLOC_FLAG_... */ | |
1da177e4 | 1933 | { |
396503fc DC |
1934 | struct xfs_mount *mp = args->mp; |
1935 | struct xfs_perag *pag = args->pag; | |
1936 | struct xfs_trans *tp = args->tp; | |
1937 | struct xfs_buf *agbp = NULL; | |
1938 | struct xfs_buf *agflbp = NULL; | |
1939 | struct xfs_alloc_arg targs; /* local allocation arguments */ | |
1940 | xfs_agblock_t bno; /* freelist block */ | |
1941 | xfs_extlen_t need; /* total blocks needed in freelist */ | |
c184f855 | 1942 | int error = 0; |
396503fc | 1943 | |
1da177e4 | 1944 | if (!pag->pagf_init) { |
396503fc DC |
1945 | error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp); |
1946 | if (error) | |
1947 | goto out_no_agbp; | |
1da177e4 | 1948 | if (!pag->pagf_init) { |
0e1edbd9 NS |
1949 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1950 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
396503fc | 1951 | goto out_agbp_relse; |
1da177e4 | 1952 | } |
396503fc | 1953 | } |
1da177e4 | 1954 | |
0e1edbd9 | 1955 | /* |
396503fc DC |
1956 | * If this is a metadata preferred pag and we are user data then try |
1957 | * somewhere else if we are not being asked to try harder at this | |
1958 | * point | |
1da177e4 | 1959 | */ |
0e1edbd9 NS |
1960 | if (pag->pagf_metadata && args->userdata && |
1961 | (flags & XFS_ALLOC_FLAG_TRYLOCK)) { | |
1962 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
396503fc | 1963 | goto out_agbp_relse; |
1da177e4 LT |
1964 | } |
1965 | ||
496817b4 | 1966 | need = xfs_alloc_min_freelist(mp, pag); |
396503fc DC |
1967 | if (!xfs_alloc_space_available(args, need, flags)) |
1968 | goto out_agbp_relse; | |
0e1edbd9 | 1969 | |
1da177e4 LT |
1970 | /* |
1971 | * Get the a.g. freespace buffer. | |
1972 | * Can fail if we're not blocking on locks, and it's held. | |
1973 | */ | |
396503fc DC |
1974 | if (!agbp) { |
1975 | error = xfs_alloc_read_agf(mp, tp, args->agno, flags, &agbp); | |
1976 | if (error) | |
1977 | goto out_no_agbp; | |
1978 | if (!agbp) { | |
0e1edbd9 NS |
1979 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1980 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
396503fc | 1981 | goto out_no_agbp; |
0e1edbd9 | 1982 | } |
1da177e4 | 1983 | } |
50adbcb4 | 1984 | |
50adbcb4 | 1985 | /* If there isn't enough total space or single-extent, reject it. */ |
496817b4 | 1986 | need = xfs_alloc_min_freelist(mp, pag); |
396503fc DC |
1987 | if (!xfs_alloc_space_available(args, need, flags)) |
1988 | goto out_agbp_relse; | |
72d55285 | 1989 | |
1da177e4 LT |
1990 | /* |
1991 | * Make the freelist shorter if it's too long. | |
50adbcb4 | 1992 | * |
396503fc DC |
1993 | * Note that from this point onwards, we will always release the agf and |
1994 | * agfl buffers on error. This handles the case where we error out and | |
1995 | * the buffers are clean or may not have been joined to the transaction | |
1996 | * and hence need to be released manually. If they have been joined to | |
1997 | * the transaction, then xfs_trans_brelse() will handle them | |
1998 | * appropriately based on the recursion count and dirty state of the | |
1999 | * buffer. | |
2000 | * | |
50adbcb4 DC |
2001 | * XXX (dgc): When we have lots of free space, does this buy us |
2002 | * anything other than extra overhead when we need to put more blocks | |
2003 | * back on the free list? Maybe we should only do this when space is | |
2004 | * getting low or the AGFL is more than half full? | |
1da177e4 | 2005 | */ |
50adbcb4 DC |
2006 | while (pag->pagf_flcount > need) { |
2007 | struct xfs_buf *bp; | |
1da177e4 | 2008 | |
92821e2b DC |
2009 | error = xfs_alloc_get_freelist(tp, agbp, &bno, 0); |
2010 | if (error) | |
396503fc | 2011 | goto out_agbp_relse; |
50adbcb4 DC |
2012 | error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1); |
2013 | if (error) | |
396503fc | 2014 | goto out_agbp_relse; |
1da177e4 LT |
2015 | bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0); |
2016 | xfs_trans_binval(tp, bp); | |
2017 | } | |
50adbcb4 | 2018 | |
a0041684 | 2019 | memset(&targs, 0, sizeof(targs)); |
1da177e4 LT |
2020 | targs.tp = tp; |
2021 | targs.mp = mp; | |
2022 | targs.agbp = agbp; | |
2023 | targs.agno = args->agno; | |
1da177e4 LT |
2024 | targs.alignment = targs.minlen = targs.prod = targs.isfl = 1; |
2025 | targs.type = XFS_ALLOCTYPE_THIS_AG; | |
2026 | targs.pag = pag; | |
50adbcb4 DC |
2027 | error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp); |
2028 | if (error) | |
396503fc | 2029 | goto out_agbp_relse; |
50adbcb4 DC |
2030 | |
2031 | /* Make the freelist longer if it's too short. */ | |
2032 | while (pag->pagf_flcount < need) { | |
1da177e4 | 2033 | targs.agbno = 0; |
50adbcb4 DC |
2034 | targs.maxlen = need - pag->pagf_flcount; |
2035 | ||
2036 | /* Allocate as many blocks as possible at once. */ | |
2037 | error = xfs_alloc_ag_vextent(&targs); | |
396503fc DC |
2038 | if (error) |
2039 | goto out_agflbp_relse; | |
2040 | ||
1da177e4 LT |
2041 | /* |
2042 | * Stop if we run out. Won't happen if callers are obeying | |
2043 | * the restrictions correctly. Can happen for free calls | |
2044 | * on a completely full ag. | |
2045 | */ | |
d210a28c | 2046 | if (targs.agbno == NULLAGBLOCK) { |
0e1edbd9 NS |
2047 | if (flags & XFS_ALLOC_FLAG_FREEING) |
2048 | break; | |
396503fc | 2049 | goto out_agflbp_relse; |
d210a28c | 2050 | } |
1da177e4 LT |
2051 | /* |
2052 | * Put each allocated block on the list. | |
2053 | */ | |
2054 | for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) { | |
92821e2b DC |
2055 | error = xfs_alloc_put_freelist(tp, agbp, |
2056 | agflbp, bno, 0); | |
2057 | if (error) | |
396503fc | 2058 | goto out_agflbp_relse; |
1da177e4 LT |
2059 | } |
2060 | } | |
e63a3690 | 2061 | xfs_trans_brelse(tp, agflbp); |
1da177e4 LT |
2062 | args->agbp = agbp; |
2063 | return 0; | |
396503fc DC |
2064 | |
2065 | out_agflbp_relse: | |
2066 | xfs_trans_brelse(tp, agflbp); | |
2067 | out_agbp_relse: | |
2068 | if (agbp) | |
2069 | xfs_trans_brelse(tp, agbp); | |
2070 | out_no_agbp: | |
2071 | args->agbp = NULL; | |
2072 | return error; | |
1da177e4 LT |
2073 | } |
2074 | ||
2075 | /* | |
2076 | * Get a block from the freelist. | |
2077 | * Returns with the buffer for the block gotten. | |
2078 | */ | |
2079 | int /* error */ | |
2080 | xfs_alloc_get_freelist( | |
2081 | xfs_trans_t *tp, /* transaction pointer */ | |
2082 | xfs_buf_t *agbp, /* buffer containing the agf structure */ | |
92821e2b DC |
2083 | xfs_agblock_t *bnop, /* block address retrieved from freelist */ |
2084 | int btreeblk) /* destination is a AGF btree */ | |
1da177e4 LT |
2085 | { |
2086 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
1da177e4 LT |
2087 | xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */ |
2088 | xfs_agblock_t bno; /* block number returned */ | |
77c95bba | 2089 | __be32 *agfl_bno; |
1da177e4 | 2090 | int error; |
92821e2b | 2091 | int logflags; |
77c95bba | 2092 | xfs_mount_t *mp = tp->t_mountp; |
1da177e4 LT |
2093 | xfs_perag_t *pag; /* per allocation group data */ |
2094 | ||
1da177e4 LT |
2095 | /* |
2096 | * Freelist is empty, give up. | |
2097 | */ | |
77c95bba | 2098 | agf = XFS_BUF_TO_AGF(agbp); |
1da177e4 LT |
2099 | if (!agf->agf_flcount) { |
2100 | *bnop = NULLAGBLOCK; | |
2101 | return 0; | |
2102 | } | |
2103 | /* | |
2104 | * Read the array of free blocks. | |
2105 | */ | |
77c95bba CH |
2106 | error = xfs_alloc_read_agfl(mp, tp, be32_to_cpu(agf->agf_seqno), |
2107 | &agflbp); | |
2108 | if (error) | |
1da177e4 | 2109 | return error; |
77c95bba CH |
2110 | |
2111 | ||
1da177e4 LT |
2112 | /* |
2113 | * Get the block number and update the data structures. | |
2114 | */ | |
77c95bba CH |
2115 | agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp); |
2116 | bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]); | |
413d57c9 | 2117 | be32_add_cpu(&agf->agf_flfirst, 1); |
1da177e4 | 2118 | xfs_trans_brelse(tp, agflbp); |
16259e7d | 2119 | if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2120 | agf->agf_flfirst = 0; |
a862e0fd DC |
2121 | |
2122 | pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); | |
413d57c9 | 2123 | be32_add_cpu(&agf->agf_flcount, -1); |
1da177e4 LT |
2124 | xfs_trans_agflist_delta(tp, -1); |
2125 | pag->pagf_flcount--; | |
a862e0fd | 2126 | xfs_perag_put(pag); |
92821e2b DC |
2127 | |
2128 | logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; | |
2129 | if (btreeblk) { | |
413d57c9 | 2130 | be32_add_cpu(&agf->agf_btreeblks, 1); |
92821e2b DC |
2131 | pag->pagf_btreeblks++; |
2132 | logflags |= XFS_AGF_BTREEBLKS; | |
2133 | } | |
2134 | ||
92821e2b | 2135 | xfs_alloc_log_agf(tp, agbp, logflags); |
1da177e4 LT |
2136 | *bnop = bno; |
2137 | ||
1da177e4 LT |
2138 | return 0; |
2139 | } | |
2140 | ||
2141 | /* | |
2142 | * Log the given fields from the agf structure. | |
2143 | */ | |
2144 | void | |
2145 | xfs_alloc_log_agf( | |
2146 | xfs_trans_t *tp, /* transaction pointer */ | |
2147 | xfs_buf_t *bp, /* buffer for a.g. freelist header */ | |
2148 | int fields) /* mask of fields to be logged (XFS_AGF_...) */ | |
2149 | { | |
2150 | int first; /* first byte offset */ | |
2151 | int last; /* last byte offset */ | |
2152 | static const short offsets[] = { | |
2153 | offsetof(xfs_agf_t, agf_magicnum), | |
2154 | offsetof(xfs_agf_t, agf_versionnum), | |
2155 | offsetof(xfs_agf_t, agf_seqno), | |
2156 | offsetof(xfs_agf_t, agf_length), | |
2157 | offsetof(xfs_agf_t, agf_roots[0]), | |
2158 | offsetof(xfs_agf_t, agf_levels[0]), | |
2159 | offsetof(xfs_agf_t, agf_flfirst), | |
2160 | offsetof(xfs_agf_t, agf_fllast), | |
2161 | offsetof(xfs_agf_t, agf_flcount), | |
2162 | offsetof(xfs_agf_t, agf_freeblks), | |
2163 | offsetof(xfs_agf_t, agf_longest), | |
92821e2b | 2164 | offsetof(xfs_agf_t, agf_btreeblks), |
4e0e6040 | 2165 | offsetof(xfs_agf_t, agf_uuid), |
1da177e4 LT |
2166 | sizeof(xfs_agf_t) |
2167 | }; | |
2168 | ||
0b1b213f CH |
2169 | trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_); |
2170 | ||
61fe135c | 2171 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGF_BUF); |
4e0e6040 | 2172 | |
1da177e4 LT |
2173 | xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last); |
2174 | xfs_trans_log_buf(tp, bp, (uint)first, (uint)last); | |
2175 | } | |
2176 | ||
2177 | /* | |
2178 | * Interface for inode allocation to force the pag data to be initialized. | |
2179 | */ | |
2180 | int /* error */ | |
2181 | xfs_alloc_pagf_init( | |
2182 | xfs_mount_t *mp, /* file system mount structure */ | |
2183 | xfs_trans_t *tp, /* transaction pointer */ | |
2184 | xfs_agnumber_t agno, /* allocation group number */ | |
2185 | int flags) /* XFS_ALLOC_FLAGS_... */ | |
2186 | { | |
2187 | xfs_buf_t *bp; | |
2188 | int error; | |
2189 | ||
2190 | if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp))) | |
2191 | return error; | |
2192 | if (bp) | |
2193 | xfs_trans_brelse(tp, bp); | |
2194 | return 0; | |
2195 | } | |
2196 | ||
2197 | /* | |
2198 | * Put the block on the freelist for the allocation group. | |
2199 | */ | |
2200 | int /* error */ | |
2201 | xfs_alloc_put_freelist( | |
2202 | xfs_trans_t *tp, /* transaction pointer */ | |
2203 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
2204 | xfs_buf_t *agflbp,/* buffer for a.g. free block array */ | |
92821e2b DC |
2205 | xfs_agblock_t bno, /* block being freed */ |
2206 | int btreeblk) /* block came from a AGF btree */ | |
1da177e4 LT |
2207 | { |
2208 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
e2101005 | 2209 | __be32 *blockp;/* pointer to array entry */ |
1da177e4 | 2210 | int error; |
92821e2b | 2211 | int logflags; |
1da177e4 LT |
2212 | xfs_mount_t *mp; /* mount structure */ |
2213 | xfs_perag_t *pag; /* per allocation group data */ | |
77c95bba CH |
2214 | __be32 *agfl_bno; |
2215 | int startoff; | |
1da177e4 LT |
2216 | |
2217 | agf = XFS_BUF_TO_AGF(agbp); | |
2218 | mp = tp->t_mountp; | |
2219 | ||
2220 | if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2221 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 | 2222 | return error; |
413d57c9 | 2223 | be32_add_cpu(&agf->agf_fllast, 1); |
16259e7d | 2224 | if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2225 | agf->agf_fllast = 0; |
a862e0fd DC |
2226 | |
2227 | pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); | |
413d57c9 | 2228 | be32_add_cpu(&agf->agf_flcount, 1); |
1da177e4 LT |
2229 | xfs_trans_agflist_delta(tp, 1); |
2230 | pag->pagf_flcount++; | |
92821e2b DC |
2231 | |
2232 | logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT; | |
2233 | if (btreeblk) { | |
413d57c9 | 2234 | be32_add_cpu(&agf->agf_btreeblks, -1); |
92821e2b DC |
2235 | pag->pagf_btreeblks--; |
2236 | logflags |= XFS_AGF_BTREEBLKS; | |
2237 | } | |
a862e0fd | 2238 | xfs_perag_put(pag); |
92821e2b | 2239 | |
92821e2b DC |
2240 | xfs_alloc_log_agf(tp, agbp, logflags); |
2241 | ||
16259e7d | 2242 | ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)); |
77c95bba CH |
2243 | |
2244 | agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, agflbp); | |
2245 | blockp = &agfl_bno[be32_to_cpu(agf->agf_fllast)]; | |
e2101005 | 2246 | *blockp = cpu_to_be32(bno); |
77c95bba CH |
2247 | startoff = (char *)blockp - (char *)agflbp->b_addr; |
2248 | ||
92821e2b | 2249 | xfs_alloc_log_agf(tp, agbp, logflags); |
77c95bba | 2250 | |
61fe135c | 2251 | xfs_trans_buf_set_type(tp, agflbp, XFS_BLFT_AGFL_BUF); |
77c95bba CH |
2252 | xfs_trans_log_buf(tp, agflbp, startoff, |
2253 | startoff + sizeof(xfs_agblock_t) - 1); | |
1da177e4 LT |
2254 | return 0; |
2255 | } | |
2256 | ||
4e0e6040 | 2257 | static bool |
612cfbfe | 2258 | xfs_agf_verify( |
4e0e6040 | 2259 | struct xfs_mount *mp, |
5d5f527d DC |
2260 | struct xfs_buf *bp) |
2261 | { | |
4e0e6040 | 2262 | struct xfs_agf *agf = XFS_BUF_TO_AGF(bp); |
5d5f527d | 2263 | |
a45086e2 BF |
2264 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
2265 | if (!uuid_equal(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid)) | |
4e0e6040 | 2266 | return false; |
a45086e2 BF |
2267 | if (!xfs_log_check_lsn(mp, |
2268 | be64_to_cpu(XFS_BUF_TO_AGF(bp)->agf_lsn))) | |
2269 | return false; | |
2270 | } | |
5d5f527d | 2271 | |
4e0e6040 DC |
2272 | if (!(agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) && |
2273 | XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) && | |
2274 | be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) && | |
2275 | be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) && | |
2276 | be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) && | |
2277 | be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp))) | |
2278 | return false; | |
5d5f527d | 2279 | |
e1b05723 ES |
2280 | if (be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]) > XFS_BTREE_MAXLEVELS || |
2281 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]) > XFS_BTREE_MAXLEVELS) | |
2282 | return false; | |
2283 | ||
5d5f527d DC |
2284 | /* |
2285 | * during growfs operations, the perag is not fully initialised, | |
2286 | * so we can't use it for any useful checking. growfs ensures we can't | |
2287 | * use it by using uncached buffers that don't have the perag attached | |
2288 | * so we can detect and avoid this problem. | |
2289 | */ | |
4e0e6040 DC |
2290 | if (bp->b_pag && be32_to_cpu(agf->agf_seqno) != bp->b_pag->pag_agno) |
2291 | return false; | |
5d5f527d | 2292 | |
4e0e6040 DC |
2293 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
2294 | be32_to_cpu(agf->agf_btreeblks) > be32_to_cpu(agf->agf_length)) | |
2295 | return false; | |
2296 | ||
2297 | return true;; | |
5d5f527d | 2298 | |
612cfbfe DC |
2299 | } |
2300 | ||
1813dd64 DC |
2301 | static void |
2302 | xfs_agf_read_verify( | |
612cfbfe DC |
2303 | struct xfs_buf *bp) |
2304 | { | |
4e0e6040 | 2305 | struct xfs_mount *mp = bp->b_target->bt_mount; |
4e0e6040 | 2306 | |
ce5028cf ES |
2307 | if (xfs_sb_version_hascrc(&mp->m_sb) && |
2308 | !xfs_buf_verify_cksum(bp, XFS_AGF_CRC_OFF)) | |
2451337d | 2309 | xfs_buf_ioerror(bp, -EFSBADCRC); |
ce5028cf ES |
2310 | else if (XFS_TEST_ERROR(!xfs_agf_verify(mp, bp), mp, |
2311 | XFS_ERRTAG_ALLOC_READ_AGF, | |
2312 | XFS_RANDOM_ALLOC_READ_AGF)) | |
2451337d | 2313 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf ES |
2314 | |
2315 | if (bp->b_error) | |
2316 | xfs_verifier_error(bp); | |
612cfbfe | 2317 | } |
5d5f527d | 2318 | |
b0f539de | 2319 | static void |
1813dd64 | 2320 | xfs_agf_write_verify( |
612cfbfe DC |
2321 | struct xfs_buf *bp) |
2322 | { | |
4e0e6040 DC |
2323 | struct xfs_mount *mp = bp->b_target->bt_mount; |
2324 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
2325 | ||
2326 | if (!xfs_agf_verify(mp, bp)) { | |
2451337d | 2327 | xfs_buf_ioerror(bp, -EFSCORRUPTED); |
ce5028cf | 2328 | xfs_verifier_error(bp); |
4e0e6040 DC |
2329 | return; |
2330 | } | |
2331 | ||
2332 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
2333 | return; | |
2334 | ||
2335 | if (bip) | |
2336 | XFS_BUF_TO_AGF(bp)->agf_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
2337 | ||
f1dbcd7e | 2338 | xfs_buf_update_cksum(bp, XFS_AGF_CRC_OFF); |
5d5f527d DC |
2339 | } |
2340 | ||
1813dd64 DC |
2341 | const struct xfs_buf_ops xfs_agf_buf_ops = { |
2342 | .verify_read = xfs_agf_read_verify, | |
2343 | .verify_write = xfs_agf_write_verify, | |
2344 | }; | |
2345 | ||
1da177e4 LT |
2346 | /* |
2347 | * Read in the allocation group header (free/alloc section). | |
2348 | */ | |
2349 | int /* error */ | |
4805621a FCH |
2350 | xfs_read_agf( |
2351 | struct xfs_mount *mp, /* mount point structure */ | |
2352 | struct xfs_trans *tp, /* transaction pointer */ | |
2353 | xfs_agnumber_t agno, /* allocation group number */ | |
2354 | int flags, /* XFS_BUF_ */ | |
2355 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
1da177e4 | 2356 | { |
1da177e4 LT |
2357 | int error; |
2358 | ||
d123031a DC |
2359 | trace_xfs_read_agf(mp, agno); |
2360 | ||
1da177e4 LT |
2361 | ASSERT(agno != NULLAGNUMBER); |
2362 | error = xfs_trans_read_buf( | |
2363 | mp, tp, mp->m_ddev_targp, | |
2364 | XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)), | |
1813dd64 | 2365 | XFS_FSS_TO_BB(mp, 1), flags, bpp, &xfs_agf_buf_ops); |
1da177e4 LT |
2366 | if (error) |
2367 | return error; | |
4805621a | 2368 | if (!*bpp) |
1da177e4 | 2369 | return 0; |
4805621a | 2370 | |
5a52c2a5 | 2371 | ASSERT(!(*bpp)->b_error); |
38f23232 | 2372 | xfs_buf_set_ref(*bpp, XFS_AGF_REF); |
4805621a FCH |
2373 | return 0; |
2374 | } | |
2375 | ||
2376 | /* | |
2377 | * Read in the allocation group header (free/alloc section). | |
2378 | */ | |
2379 | int /* error */ | |
2380 | xfs_alloc_read_agf( | |
2381 | struct xfs_mount *mp, /* mount point structure */ | |
2382 | struct xfs_trans *tp, /* transaction pointer */ | |
2383 | xfs_agnumber_t agno, /* allocation group number */ | |
2384 | int flags, /* XFS_ALLOC_FLAG_... */ | |
2385 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
2386 | { | |
2387 | struct xfs_agf *agf; /* ag freelist header */ | |
2388 | struct xfs_perag *pag; /* per allocation group data */ | |
2389 | int error; | |
2390 | ||
d123031a | 2391 | trace_xfs_alloc_read_agf(mp, agno); |
4805621a | 2392 | |
d123031a | 2393 | ASSERT(agno != NULLAGNUMBER); |
4805621a | 2394 | error = xfs_read_agf(mp, tp, agno, |
0cadda1c | 2395 | (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0, |
4805621a FCH |
2396 | bpp); |
2397 | if (error) | |
2398 | return error; | |
2399 | if (!*bpp) | |
2400 | return 0; | |
5a52c2a5 | 2401 | ASSERT(!(*bpp)->b_error); |
4805621a FCH |
2402 | |
2403 | agf = XFS_BUF_TO_AGF(*bpp); | |
a862e0fd | 2404 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2405 | if (!pag->pagf_init) { |
16259e7d | 2406 | pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); |
92821e2b | 2407 | pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); |
16259e7d CH |
2408 | pag->pagf_flcount = be32_to_cpu(agf->agf_flcount); |
2409 | pag->pagf_longest = be32_to_cpu(agf->agf_longest); | |
1da177e4 | 2410 | pag->pagf_levels[XFS_BTNUM_BNOi] = |
16259e7d | 2411 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); |
1da177e4 | 2412 | pag->pagf_levels[XFS_BTNUM_CNTi] = |
16259e7d | 2413 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); |
007c61c6 | 2414 | spin_lock_init(&pag->pagb_lock); |
e57336ff | 2415 | pag->pagb_count = 0; |
ed3b4d6c | 2416 | pag->pagb_tree = RB_ROOT; |
1da177e4 LT |
2417 | pag->pagf_init = 1; |
2418 | } | |
2419 | #ifdef DEBUG | |
2420 | else if (!XFS_FORCED_SHUTDOWN(mp)) { | |
16259e7d | 2421 | ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks)); |
89b28393 | 2422 | ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks)); |
16259e7d CH |
2423 | ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount)); |
2424 | ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest)); | |
1da177e4 | 2425 | ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] == |
16259e7d | 2426 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi])); |
1da177e4 | 2427 | ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] == |
16259e7d | 2428 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); |
1da177e4 LT |
2429 | } |
2430 | #endif | |
a862e0fd | 2431 | xfs_perag_put(pag); |
1da177e4 LT |
2432 | return 0; |
2433 | } | |
2434 | ||
2435 | /* | |
2436 | * Allocate an extent (variable-size). | |
2437 | * Depending on the allocation type, we either look in a single allocation | |
2438 | * group or loop over the allocation groups to find the result. | |
2439 | */ | |
2440 | int /* error */ | |
e04426b9 | 2441 | xfs_alloc_vextent( |
1da177e4 LT |
2442 | xfs_alloc_arg_t *args) /* allocation argument structure */ |
2443 | { | |
2444 | xfs_agblock_t agsize; /* allocation group size */ | |
2445 | int error; | |
2446 | int flags; /* XFS_ALLOC_FLAG_... locking flags */ | |
1da177e4 LT |
2447 | xfs_extlen_t minleft;/* minimum left value, temp copy */ |
2448 | xfs_mount_t *mp; /* mount structure pointer */ | |
2449 | xfs_agnumber_t sagno; /* starting allocation group number */ | |
2450 | xfs_alloctype_t type; /* input allocation type */ | |
2451 | int bump_rotor = 0; | |
2452 | int no_min = 0; | |
2453 | xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */ | |
2454 | ||
2455 | mp = args->mp; | |
2456 | type = args->otype = args->type; | |
2457 | args->agbno = NULLAGBLOCK; | |
2458 | /* | |
2459 | * Just fix this up, for the case where the last a.g. is shorter | |
2460 | * (or there's only one a.g.) and the caller couldn't easily figure | |
2461 | * that out (xfs_bmap_alloc). | |
2462 | */ | |
2463 | agsize = mp->m_sb.sb_agblocks; | |
2464 | if (args->maxlen > agsize) | |
2465 | args->maxlen = agsize; | |
2466 | if (args->alignment == 0) | |
2467 | args->alignment = 1; | |
2468 | ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount); | |
2469 | ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize); | |
2470 | ASSERT(args->minlen <= args->maxlen); | |
2471 | ASSERT(args->minlen <= agsize); | |
2472 | ASSERT(args->mod < args->prod); | |
2473 | if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount || | |
2474 | XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize || | |
2475 | args->minlen > args->maxlen || args->minlen > agsize || | |
2476 | args->mod >= args->prod) { | |
2477 | args->fsbno = NULLFSBLOCK; | |
0b1b213f | 2478 | trace_xfs_alloc_vextent_badargs(args); |
1da177e4 LT |
2479 | return 0; |
2480 | } | |
2481 | minleft = args->minleft; | |
2482 | ||
2483 | switch (type) { | |
2484 | case XFS_ALLOCTYPE_THIS_AG: | |
2485 | case XFS_ALLOCTYPE_NEAR_BNO: | |
2486 | case XFS_ALLOCTYPE_THIS_BNO: | |
2487 | /* | |
2488 | * These three force us into a single a.g. | |
2489 | */ | |
2490 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
a862e0fd | 2491 | args->pag = xfs_perag_get(mp, args->agno); |
1da177e4 LT |
2492 | args->minleft = 0; |
2493 | error = xfs_alloc_fix_freelist(args, 0); | |
2494 | args->minleft = minleft; | |
2495 | if (error) { | |
0b1b213f | 2496 | trace_xfs_alloc_vextent_nofix(args); |
1da177e4 LT |
2497 | goto error0; |
2498 | } | |
2499 | if (!args->agbp) { | |
0b1b213f | 2500 | trace_xfs_alloc_vextent_noagbp(args); |
1da177e4 LT |
2501 | break; |
2502 | } | |
2503 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2504 | if ((error = xfs_alloc_ag_vextent(args))) | |
2505 | goto error0; | |
1da177e4 LT |
2506 | break; |
2507 | case XFS_ALLOCTYPE_START_BNO: | |
2508 | /* | |
2509 | * Try near allocation first, then anywhere-in-ag after | |
2510 | * the first a.g. fails. | |
2511 | */ | |
3fbbbea3 | 2512 | if ((args->userdata & XFS_ALLOC_INITIAL_USER_DATA) && |
1da177e4 LT |
2513 | (mp->m_flags & XFS_MOUNT_32BITINODES)) { |
2514 | args->fsbno = XFS_AGB_TO_FSB(mp, | |
2515 | ((mp->m_agfrotor / rotorstep) % | |
2516 | mp->m_sb.sb_agcount), 0); | |
2517 | bump_rotor = 1; | |
2518 | } | |
2519 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2520 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2521 | /* FALLTHROUGH */ | |
2522 | case XFS_ALLOCTYPE_ANY_AG: | |
2523 | case XFS_ALLOCTYPE_START_AG: | |
2524 | case XFS_ALLOCTYPE_FIRST_AG: | |
2525 | /* | |
2526 | * Rotate through the allocation groups looking for a winner. | |
2527 | */ | |
2528 | if (type == XFS_ALLOCTYPE_ANY_AG) { | |
2529 | /* | |
2530 | * Start with the last place we left off. | |
2531 | */ | |
2532 | args->agno = sagno = (mp->m_agfrotor / rotorstep) % | |
2533 | mp->m_sb.sb_agcount; | |
2534 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2535 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2536 | } else if (type == XFS_ALLOCTYPE_FIRST_AG) { | |
2537 | /* | |
2538 | * Start with allocation group given by bno. | |
2539 | */ | |
2540 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2541 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2542 | sagno = 0; | |
2543 | flags = 0; | |
2544 | } else { | |
2545 | if (type == XFS_ALLOCTYPE_START_AG) | |
2546 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2547 | /* | |
2548 | * Start with the given allocation group. | |
2549 | */ | |
2550 | args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2551 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2552 | } | |
2553 | /* | |
2554 | * Loop over allocation groups twice; first time with | |
2555 | * trylock set, second time without. | |
2556 | */ | |
1da177e4 | 2557 | for (;;) { |
a862e0fd | 2558 | args->pag = xfs_perag_get(mp, args->agno); |
1da177e4 LT |
2559 | if (no_min) args->minleft = 0; |
2560 | error = xfs_alloc_fix_freelist(args, flags); | |
2561 | args->minleft = minleft; | |
2562 | if (error) { | |
0b1b213f | 2563 | trace_xfs_alloc_vextent_nofix(args); |
1da177e4 LT |
2564 | goto error0; |
2565 | } | |
2566 | /* | |
2567 | * If we get a buffer back then the allocation will fly. | |
2568 | */ | |
2569 | if (args->agbp) { | |
2570 | if ((error = xfs_alloc_ag_vextent(args))) | |
2571 | goto error0; | |
2572 | break; | |
2573 | } | |
0b1b213f CH |
2574 | |
2575 | trace_xfs_alloc_vextent_loopfailed(args); | |
2576 | ||
1da177e4 LT |
2577 | /* |
2578 | * Didn't work, figure out the next iteration. | |
2579 | */ | |
2580 | if (args->agno == sagno && | |
2581 | type == XFS_ALLOCTYPE_START_BNO) | |
2582 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
d210a28c YL |
2583 | /* |
2584 | * For the first allocation, we can try any AG to get | |
2585 | * space. However, if we already have allocated a | |
2586 | * block, we don't want to try AGs whose number is below | |
2587 | * sagno. Otherwise, we may end up with out-of-order | |
2588 | * locking of AGF, which might cause deadlock. | |
2589 | */ | |
2590 | if (++(args->agno) == mp->m_sb.sb_agcount) { | |
2591 | if (args->firstblock != NULLFSBLOCK) | |
2592 | args->agno = sagno; | |
2593 | else | |
2594 | args->agno = 0; | |
2595 | } | |
1da177e4 LT |
2596 | /* |
2597 | * Reached the starting a.g., must either be done | |
2598 | * or switch to non-trylock mode. | |
2599 | */ | |
2600 | if (args->agno == sagno) { | |
2601 | if (no_min == 1) { | |
2602 | args->agbno = NULLAGBLOCK; | |
0b1b213f | 2603 | trace_xfs_alloc_vextent_allfailed(args); |
1da177e4 LT |
2604 | break; |
2605 | } | |
2606 | if (flags == 0) { | |
2607 | no_min = 1; | |
2608 | } else { | |
2609 | flags = 0; | |
2610 | if (type == XFS_ALLOCTYPE_START_BNO) { | |
2611 | args->agbno = XFS_FSB_TO_AGBNO(mp, | |
2612 | args->fsbno); | |
2613 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2614 | } | |
2615 | } | |
2616 | } | |
a862e0fd | 2617 | xfs_perag_put(args->pag); |
1da177e4 | 2618 | } |
1da177e4 LT |
2619 | if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) { |
2620 | if (args->agno == sagno) | |
2621 | mp->m_agfrotor = (mp->m_agfrotor + 1) % | |
2622 | (mp->m_sb.sb_agcount * rotorstep); | |
2623 | else | |
2624 | mp->m_agfrotor = (args->agno * rotorstep + 1) % | |
2625 | (mp->m_sb.sb_agcount * rotorstep); | |
2626 | } | |
2627 | break; | |
2628 | default: | |
2629 | ASSERT(0); | |
2630 | /* NOTREACHED */ | |
2631 | } | |
2632 | if (args->agbno == NULLAGBLOCK) | |
2633 | args->fsbno = NULLFSBLOCK; | |
2634 | else { | |
2635 | args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno); | |
2636 | #ifdef DEBUG | |
2637 | ASSERT(args->len >= args->minlen); | |
2638 | ASSERT(args->len <= args->maxlen); | |
2639 | ASSERT(args->agbno % args->alignment == 0); | |
2640 | XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), | |
2641 | args->len); | |
2642 | #endif | |
3fbbbea3 DC |
2643 | |
2644 | /* Zero the extent if we were asked to do so */ | |
2645 | if (args->userdata & XFS_ALLOC_USERDATA_ZERO) { | |
2646 | error = xfs_zero_extent(args->ip, args->fsbno, args->len); | |
2647 | if (error) | |
2648 | goto error0; | |
2649 | } | |
2650 | ||
1da177e4 | 2651 | } |
a862e0fd | 2652 | xfs_perag_put(args->pag); |
1da177e4 LT |
2653 | return 0; |
2654 | error0: | |
a862e0fd | 2655 | xfs_perag_put(args->pag); |
1da177e4 LT |
2656 | return error; |
2657 | } | |
2658 | ||
2659 | /* | |
2660 | * Free an extent. | |
2661 | * Just break up the extent address and hand off to xfs_free_ag_extent | |
2662 | * after fixing up the freelist. | |
2663 | */ | |
2664 | int /* error */ | |
2665 | xfs_free_extent( | |
2666 | xfs_trans_t *tp, /* transaction pointer */ | |
2667 | xfs_fsblock_t bno, /* starting block number of extent */ | |
2668 | xfs_extlen_t len) /* length of extent */ | |
2669 | { | |
0e1edbd9 | 2670 | xfs_alloc_arg_t args; |
1da177e4 LT |
2671 | int error; |
2672 | ||
2673 | ASSERT(len != 0); | |
0e1edbd9 | 2674 | memset(&args, 0, sizeof(xfs_alloc_arg_t)); |
1da177e4 LT |
2675 | args.tp = tp; |
2676 | args.mp = tp->t_mountp; | |
be65b18a DC |
2677 | |
2678 | /* | |
2679 | * validate that the block number is legal - the enables us to detect | |
2680 | * and handle a silent filesystem corruption rather than crashing. | |
2681 | */ | |
1da177e4 | 2682 | args.agno = XFS_FSB_TO_AGNO(args.mp, bno); |
be65b18a | 2683 | if (args.agno >= args.mp->m_sb.sb_agcount) |
2451337d | 2684 | return -EFSCORRUPTED; |
be65b18a | 2685 | |
1da177e4 | 2686 | args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno); |
be65b18a | 2687 | if (args.agbno >= args.mp->m_sb.sb_agblocks) |
2451337d | 2688 | return -EFSCORRUPTED; |
be65b18a | 2689 | |
a862e0fd | 2690 | args.pag = xfs_perag_get(args.mp, args.agno); |
be65b18a DC |
2691 | ASSERT(args.pag); |
2692 | ||
2693 | error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING); | |
2694 | if (error) | |
1da177e4 | 2695 | goto error0; |
be65b18a DC |
2696 | |
2697 | /* validate the extent size is legal now we have the agf locked */ | |
2698 | if (args.agbno + len > | |
2699 | be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)) { | |
2451337d | 2700 | error = -EFSCORRUPTED; |
be65b18a DC |
2701 | goto error0; |
2702 | } | |
2703 | ||
0e1edbd9 | 2704 | error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0); |
a870acd9 | 2705 | if (!error) |
4ecbfe63 | 2706 | xfs_extent_busy_insert(tp, args.agno, args.agbno, len, 0); |
1da177e4 | 2707 | error0: |
a862e0fd | 2708 | xfs_perag_put(args.pag); |
1da177e4 LT |
2709 | return error; |
2710 | } |