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