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" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 LT |
22 | #include "xfs_log.h" |
23 | #include "xfs_trans.h" | |
24 | #include "xfs_sb.h" | |
25 | #include "xfs_ag.h" | |
1da177e4 | 26 | #include "xfs_mount.h" |
1da177e4 | 27 | #include "xfs_bmap_btree.h" |
a844f451 | 28 | #include "xfs_alloc_btree.h" |
1da177e4 | 29 | #include "xfs_ialloc_btree.h" |
a844f451 NS |
30 | #include "xfs_dinode.h" |
31 | #include "xfs_inode.h" | |
1da177e4 | 32 | #include "xfs_btree.h" |
1da177e4 | 33 | #include "xfs_alloc.h" |
efc27b52 | 34 | #include "xfs_extent_busy.h" |
1da177e4 | 35 | #include "xfs_error.h" |
0b1b213f | 36 | #include "xfs_trace.h" |
1da177e4 | 37 | |
c999a223 | 38 | struct workqueue_struct *xfs_alloc_wq; |
1da177e4 LT |
39 | |
40 | #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b))) | |
41 | ||
42 | #define XFSA_FIXUP_BNO_OK 1 | |
43 | #define XFSA_FIXUP_CNT_OK 2 | |
44 | ||
1da177e4 LT |
45 | STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *); |
46 | STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *); | |
47 | STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *); | |
48 | STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *, | |
e26f0501 | 49 | xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *); |
1da177e4 | 50 | |
fe033cc8 CH |
51 | /* |
52 | * Lookup the record equal to [bno, len] in the btree given by cur. | |
53 | */ | |
54 | STATIC int /* error */ | |
55 | xfs_alloc_lookup_eq( | |
56 | struct xfs_btree_cur *cur, /* btree cursor */ | |
57 | xfs_agblock_t bno, /* starting block of extent */ | |
58 | xfs_extlen_t len, /* length of extent */ | |
59 | int *stat) /* success/failure */ | |
60 | { | |
61 | cur->bc_rec.a.ar_startblock = bno; | |
62 | cur->bc_rec.a.ar_blockcount = len; | |
63 | return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat); | |
64 | } | |
65 | ||
66 | /* | |
67 | * Lookup the first record greater than or equal to [bno, len] | |
68 | * in the btree given by cur. | |
69 | */ | |
a66d6363 | 70 | int /* error */ |
fe033cc8 CH |
71 | xfs_alloc_lookup_ge( |
72 | struct xfs_btree_cur *cur, /* btree cursor */ | |
73 | xfs_agblock_t bno, /* starting block of extent */ | |
74 | xfs_extlen_t len, /* length of extent */ | |
75 | int *stat) /* success/failure */ | |
76 | { | |
77 | cur->bc_rec.a.ar_startblock = bno; | |
78 | cur->bc_rec.a.ar_blockcount = len; | |
79 | return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); | |
80 | } | |
81 | ||
82 | /* | |
83 | * Lookup the first record less than or equal to [bno, len] | |
84 | * in the btree given by cur. | |
85 | */ | |
a46db608 | 86 | int /* error */ |
fe033cc8 CH |
87 | xfs_alloc_lookup_le( |
88 | struct xfs_btree_cur *cur, /* btree cursor */ | |
89 | xfs_agblock_t bno, /* starting block of extent */ | |
90 | xfs_extlen_t len, /* length of extent */ | |
91 | int *stat) /* success/failure */ | |
92 | { | |
93 | cur->bc_rec.a.ar_startblock = bno; | |
94 | cur->bc_rec.a.ar_blockcount = len; | |
95 | return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat); | |
96 | } | |
97 | ||
278d0ca1 CH |
98 | /* |
99 | * Update the record referred to by cur to the value given | |
100 | * by [bno, len]. | |
101 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
102 | */ | |
103 | STATIC int /* error */ | |
104 | xfs_alloc_update( | |
105 | struct xfs_btree_cur *cur, /* btree cursor */ | |
106 | xfs_agblock_t bno, /* starting block of extent */ | |
107 | xfs_extlen_t len) /* length of extent */ | |
108 | { | |
109 | union xfs_btree_rec rec; | |
110 | ||
111 | rec.alloc.ar_startblock = cpu_to_be32(bno); | |
112 | rec.alloc.ar_blockcount = cpu_to_be32(len); | |
113 | return xfs_btree_update(cur, &rec); | |
114 | } | |
fe033cc8 | 115 | |
8cc938fe CH |
116 | /* |
117 | * Get the data from the pointed-to record. | |
118 | */ | |
a46db608 | 119 | int /* error */ |
8cc938fe CH |
120 | xfs_alloc_get_rec( |
121 | struct xfs_btree_cur *cur, /* btree cursor */ | |
122 | xfs_agblock_t *bno, /* output: starting block of extent */ | |
123 | xfs_extlen_t *len, /* output: length of extent */ | |
124 | int *stat) /* output: success/failure */ | |
125 | { | |
126 | union xfs_btree_rec *rec; | |
127 | int error; | |
128 | ||
129 | error = xfs_btree_get_rec(cur, &rec, stat); | |
130 | if (!error && *stat == 1) { | |
131 | *bno = be32_to_cpu(rec->alloc.ar_startblock); | |
132 | *len = be32_to_cpu(rec->alloc.ar_blockcount); | |
133 | } | |
134 | return error; | |
135 | } | |
136 | ||
1da177e4 LT |
137 | /* |
138 | * Compute aligned version of the found extent. | |
139 | * Takes alignment and min length into account. | |
140 | */ | |
12375c82 | 141 | STATIC void |
1da177e4 | 142 | xfs_alloc_compute_aligned( |
86fa8af6 | 143 | xfs_alloc_arg_t *args, /* allocation argument structure */ |
1da177e4 LT |
144 | xfs_agblock_t foundbno, /* starting block in found extent */ |
145 | xfs_extlen_t foundlen, /* length in found extent */ | |
1da177e4 LT |
146 | xfs_agblock_t *resbno, /* result block number */ |
147 | xfs_extlen_t *reslen) /* result length */ | |
148 | { | |
149 | xfs_agblock_t bno; | |
1da177e4 LT |
150 | xfs_extlen_t len; |
151 | ||
e26f0501 | 152 | /* Trim busy sections out of found extent */ |
4ecbfe63 | 153 | xfs_extent_busy_trim(args, foundbno, foundlen, &bno, &len); |
e26f0501 CH |
154 | |
155 | if (args->alignment > 1 && len >= args->minlen) { | |
156 | xfs_agblock_t aligned_bno = roundup(bno, args->alignment); | |
157 | xfs_extlen_t diff = aligned_bno - bno; | |
158 | ||
159 | *resbno = aligned_bno; | |
160 | *reslen = diff >= len ? 0 : len - diff; | |
1da177e4 | 161 | } else { |
e26f0501 CH |
162 | *resbno = bno; |
163 | *reslen = len; | |
1da177e4 | 164 | } |
1da177e4 LT |
165 | } |
166 | ||
167 | /* | |
168 | * Compute best start block and diff for "near" allocations. | |
169 | * freelen >= wantlen already checked by caller. | |
170 | */ | |
171 | STATIC xfs_extlen_t /* difference value (absolute) */ | |
172 | xfs_alloc_compute_diff( | |
173 | xfs_agblock_t wantbno, /* target starting block */ | |
174 | xfs_extlen_t wantlen, /* target length */ | |
175 | xfs_extlen_t alignment, /* target alignment */ | |
176 | xfs_agblock_t freebno, /* freespace's starting block */ | |
177 | xfs_extlen_t freelen, /* freespace's length */ | |
178 | xfs_agblock_t *newbnop) /* result: best start block from free */ | |
179 | { | |
180 | xfs_agblock_t freeend; /* end of freespace extent */ | |
181 | xfs_agblock_t newbno1; /* return block number */ | |
182 | xfs_agblock_t newbno2; /* other new block number */ | |
183 | xfs_extlen_t newlen1=0; /* length with newbno1 */ | |
184 | xfs_extlen_t newlen2=0; /* length with newbno2 */ | |
185 | xfs_agblock_t wantend; /* end of target extent */ | |
186 | ||
187 | ASSERT(freelen >= wantlen); | |
188 | freeend = freebno + freelen; | |
189 | wantend = wantbno + wantlen; | |
190 | if (freebno >= wantbno) { | |
191 | if ((newbno1 = roundup(freebno, alignment)) >= freeend) | |
192 | newbno1 = NULLAGBLOCK; | |
193 | } else if (freeend >= wantend && alignment > 1) { | |
194 | newbno1 = roundup(wantbno, alignment); | |
195 | newbno2 = newbno1 - alignment; | |
196 | if (newbno1 >= freeend) | |
197 | newbno1 = NULLAGBLOCK; | |
198 | else | |
199 | newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1); | |
200 | if (newbno2 < freebno) | |
201 | newbno2 = NULLAGBLOCK; | |
202 | else | |
203 | newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2); | |
204 | if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) { | |
205 | if (newlen1 < newlen2 || | |
206 | (newlen1 == newlen2 && | |
207 | XFS_ABSDIFF(newbno1, wantbno) > | |
208 | XFS_ABSDIFF(newbno2, wantbno))) | |
209 | newbno1 = newbno2; | |
210 | } else if (newbno2 != NULLAGBLOCK) | |
211 | newbno1 = newbno2; | |
212 | } else if (freeend >= wantend) { | |
213 | newbno1 = wantbno; | |
214 | } else if (alignment > 1) { | |
215 | newbno1 = roundup(freeend - wantlen, alignment); | |
216 | if (newbno1 > freeend - wantlen && | |
217 | newbno1 - alignment >= freebno) | |
218 | newbno1 -= alignment; | |
219 | else if (newbno1 >= freeend) | |
220 | newbno1 = NULLAGBLOCK; | |
221 | } else | |
222 | newbno1 = freeend - wantlen; | |
223 | *newbnop = newbno1; | |
224 | return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno); | |
225 | } | |
226 | ||
227 | /* | |
228 | * Fix up the length, based on mod and prod. | |
229 | * len should be k * prod + mod for some k. | |
230 | * If len is too small it is returned unchanged. | |
231 | * If len hits maxlen it is left alone. | |
232 | */ | |
233 | STATIC void | |
234 | xfs_alloc_fix_len( | |
235 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
236 | { | |
237 | xfs_extlen_t k; | |
238 | xfs_extlen_t rlen; | |
239 | ||
240 | ASSERT(args->mod < args->prod); | |
241 | rlen = args->len; | |
242 | ASSERT(rlen >= args->minlen); | |
243 | ASSERT(rlen <= args->maxlen); | |
244 | if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen || | |
245 | (args->mod == 0 && rlen < args->prod)) | |
246 | return; | |
247 | k = rlen % args->prod; | |
248 | if (k == args->mod) | |
249 | return; | |
250 | if (k > args->mod) { | |
251 | if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen) | |
252 | return; | |
253 | } else { | |
254 | if ((int)(rlen = rlen - args->prod - (args->mod - k)) < | |
255 | (int)args->minlen) | |
256 | return; | |
257 | } | |
258 | ASSERT(rlen >= args->minlen); | |
259 | ASSERT(rlen <= args->maxlen); | |
260 | args->len = rlen; | |
261 | } | |
262 | ||
263 | /* | |
264 | * Fix up length if there is too little space left in the a.g. | |
265 | * Return 1 if ok, 0 if too little, should give up. | |
266 | */ | |
267 | STATIC int | |
268 | xfs_alloc_fix_minleft( | |
269 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
270 | { | |
271 | xfs_agf_t *agf; /* a.g. freelist header */ | |
272 | int diff; /* free space difference */ | |
273 | ||
274 | if (args->minleft == 0) | |
275 | return 1; | |
276 | agf = XFS_BUF_TO_AGF(args->agbp); | |
16259e7d | 277 | diff = be32_to_cpu(agf->agf_freeblks) |
1da177e4 LT |
278 | - args->len - args->minleft; |
279 | if (diff >= 0) | |
280 | return 1; | |
281 | args->len += diff; /* shrink the allocated space */ | |
282 | if (args->len >= args->minlen) | |
283 | return 1; | |
284 | args->agbno = NULLAGBLOCK; | |
285 | return 0; | |
286 | } | |
287 | ||
288 | /* | |
289 | * Update the two btrees, logically removing from freespace the extent | |
290 | * starting at rbno, rlen blocks. The extent is contained within the | |
291 | * actual (current) free extent fbno for flen blocks. | |
292 | * Flags are passed in indicating whether the cursors are set to the | |
293 | * relevant records. | |
294 | */ | |
295 | STATIC int /* error code */ | |
296 | xfs_alloc_fixup_trees( | |
297 | xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */ | |
298 | xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */ | |
299 | xfs_agblock_t fbno, /* starting block of free extent */ | |
300 | xfs_extlen_t flen, /* length of free extent */ | |
301 | xfs_agblock_t rbno, /* starting block of returned extent */ | |
302 | xfs_extlen_t rlen, /* length of returned extent */ | |
303 | int flags) /* flags, XFSA_FIXUP_... */ | |
304 | { | |
305 | int error; /* error code */ | |
306 | int i; /* operation results */ | |
307 | xfs_agblock_t nfbno1; /* first new free startblock */ | |
308 | xfs_agblock_t nfbno2; /* second new free startblock */ | |
309 | xfs_extlen_t nflen1=0; /* first new free length */ | |
310 | xfs_extlen_t nflen2=0; /* second new free length */ | |
311 | ||
312 | /* | |
313 | * Look up the record in the by-size tree if necessary. | |
314 | */ | |
315 | if (flags & XFSA_FIXUP_CNT_OK) { | |
316 | #ifdef DEBUG | |
317 | if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i))) | |
318 | return error; | |
319 | XFS_WANT_CORRUPTED_RETURN( | |
320 | i == 1 && nfbno1 == fbno && nflen1 == flen); | |
321 | #endif | |
322 | } else { | |
323 | if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i))) | |
324 | return error; | |
325 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
326 | } | |
327 | /* | |
328 | * Look up the record in the by-block tree if necessary. | |
329 | */ | |
330 | if (flags & XFSA_FIXUP_BNO_OK) { | |
331 | #ifdef DEBUG | |
332 | if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i))) | |
333 | return error; | |
334 | XFS_WANT_CORRUPTED_RETURN( | |
335 | i == 1 && nfbno1 == fbno && nflen1 == flen); | |
336 | #endif | |
337 | } else { | |
338 | if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i))) | |
339 | return error; | |
340 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
341 | } | |
7cc95a82 | 342 | |
1da177e4 | 343 | #ifdef DEBUG |
7cc95a82 CH |
344 | if (bno_cur->bc_nlevels == 1 && cnt_cur->bc_nlevels == 1) { |
345 | struct xfs_btree_block *bnoblock; | |
346 | struct xfs_btree_block *cntblock; | |
347 | ||
348 | bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]); | |
349 | cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]); | |
1da177e4 | 350 | |
7cc95a82 CH |
351 | XFS_WANT_CORRUPTED_RETURN( |
352 | bnoblock->bb_numrecs == cntblock->bb_numrecs); | |
1da177e4 LT |
353 | } |
354 | #endif | |
7cc95a82 | 355 | |
1da177e4 LT |
356 | /* |
357 | * Deal with all four cases: the allocated record is contained | |
358 | * within the freespace record, so we can have new freespace | |
359 | * at either (or both) end, or no freespace remaining. | |
360 | */ | |
361 | if (rbno == fbno && rlen == flen) | |
362 | nfbno1 = nfbno2 = NULLAGBLOCK; | |
363 | else if (rbno == fbno) { | |
364 | nfbno1 = rbno + rlen; | |
365 | nflen1 = flen - rlen; | |
366 | nfbno2 = NULLAGBLOCK; | |
367 | } else if (rbno + rlen == fbno + flen) { | |
368 | nfbno1 = fbno; | |
369 | nflen1 = flen - rlen; | |
370 | nfbno2 = NULLAGBLOCK; | |
371 | } else { | |
372 | nfbno1 = fbno; | |
373 | nflen1 = rbno - fbno; | |
374 | nfbno2 = rbno + rlen; | |
375 | nflen2 = (fbno + flen) - nfbno2; | |
376 | } | |
377 | /* | |
378 | * Delete the entry from the by-size btree. | |
379 | */ | |
91cca5df | 380 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
381 | return error; |
382 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
383 | /* | |
384 | * Add new by-size btree entry(s). | |
385 | */ | |
386 | if (nfbno1 != NULLAGBLOCK) { | |
387 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i))) | |
388 | return error; | |
389 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 390 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
391 | return error; |
392 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
393 | } | |
394 | if (nfbno2 != NULLAGBLOCK) { | |
395 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i))) | |
396 | return error; | |
397 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 398 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
399 | return error; |
400 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
401 | } | |
402 | /* | |
403 | * Fix up the by-block btree entry(s). | |
404 | */ | |
405 | if (nfbno1 == NULLAGBLOCK) { | |
406 | /* | |
407 | * No remaining freespace, just delete the by-block tree entry. | |
408 | */ | |
91cca5df | 409 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 LT |
410 | return error; |
411 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
412 | } else { | |
413 | /* | |
414 | * Update the by-block entry to start later|be shorter. | |
415 | */ | |
416 | if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1))) | |
417 | return error; | |
418 | } | |
419 | if (nfbno2 != NULLAGBLOCK) { | |
420 | /* | |
421 | * 2 resulting free entries, need to add one. | |
422 | */ | |
423 | if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i))) | |
424 | return error; | |
425 | XFS_WANT_CORRUPTED_RETURN(i == 0); | |
4b22a571 | 426 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 LT |
427 | return error; |
428 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
429 | } | |
430 | return 0; | |
431 | } | |
432 | ||
433 | /* | |
434 | * Read in the allocation group free block array. | |
435 | */ | |
436 | STATIC int /* error */ | |
437 | xfs_alloc_read_agfl( | |
438 | xfs_mount_t *mp, /* mount point structure */ | |
439 | xfs_trans_t *tp, /* transaction pointer */ | |
440 | xfs_agnumber_t agno, /* allocation group number */ | |
441 | xfs_buf_t **bpp) /* buffer for the ag free block array */ | |
442 | { | |
443 | xfs_buf_t *bp; /* return value */ | |
444 | int error; | |
445 | ||
446 | ASSERT(agno != NULLAGNUMBER); | |
447 | error = xfs_trans_read_buf( | |
448 | mp, tp, mp->m_ddev_targp, | |
449 | XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)), | |
450 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
451 | if (error) | |
452 | return error; | |
5a52c2a5 | 453 | ASSERT(!xfs_buf_geterror(bp)); |
38f23232 | 454 | xfs_buf_set_ref(bp, XFS_AGFL_REF); |
1da177e4 LT |
455 | *bpp = bp; |
456 | return 0; | |
457 | } | |
458 | ||
ecb6928f CH |
459 | STATIC int |
460 | xfs_alloc_update_counters( | |
461 | struct xfs_trans *tp, | |
462 | struct xfs_perag *pag, | |
463 | struct xfs_buf *agbp, | |
464 | long len) | |
465 | { | |
466 | struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); | |
467 | ||
468 | pag->pagf_freeblks += len; | |
469 | be32_add_cpu(&agf->agf_freeblks, len); | |
470 | ||
471 | xfs_trans_agblocks_delta(tp, len); | |
472 | if (unlikely(be32_to_cpu(agf->agf_freeblks) > | |
473 | be32_to_cpu(agf->agf_length))) | |
474 | return EFSCORRUPTED; | |
475 | ||
476 | xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS); | |
477 | return 0; | |
478 | } | |
479 | ||
1da177e4 LT |
480 | /* |
481 | * Allocation group level functions. | |
482 | */ | |
483 | ||
484 | /* | |
485 | * Allocate a variable extent in the allocation group agno. | |
486 | * Type and bno are used to determine where in the allocation group the | |
487 | * extent will start. | |
488 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
489 | * and of the form k * prod + mod unless there's nothing that large. | |
490 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
491 | */ | |
492 | STATIC int /* error */ | |
493 | xfs_alloc_ag_vextent( | |
494 | xfs_alloc_arg_t *args) /* argument structure for allocation */ | |
495 | { | |
496 | int error=0; | |
1da177e4 LT |
497 | |
498 | ASSERT(args->minlen > 0); | |
499 | ASSERT(args->maxlen > 0); | |
500 | ASSERT(args->minlen <= args->maxlen); | |
501 | ASSERT(args->mod < args->prod); | |
502 | ASSERT(args->alignment > 0); | |
503 | /* | |
504 | * Branch to correct routine based on the type. | |
505 | */ | |
506 | args->wasfromfl = 0; | |
507 | switch (args->type) { | |
508 | case XFS_ALLOCTYPE_THIS_AG: | |
509 | error = xfs_alloc_ag_vextent_size(args); | |
510 | break; | |
511 | case XFS_ALLOCTYPE_NEAR_BNO: | |
512 | error = xfs_alloc_ag_vextent_near(args); | |
513 | break; | |
514 | case XFS_ALLOCTYPE_THIS_BNO: | |
515 | error = xfs_alloc_ag_vextent_exact(args); | |
516 | break; | |
517 | default: | |
518 | ASSERT(0); | |
519 | /* NOTREACHED */ | |
520 | } | |
ecb6928f CH |
521 | |
522 | if (error || args->agbno == NULLAGBLOCK) | |
1da177e4 | 523 | return error; |
ecb6928f CH |
524 | |
525 | ASSERT(args->len >= args->minlen); | |
526 | ASSERT(args->len <= args->maxlen); | |
527 | ASSERT(!args->wasfromfl || !args->isfl); | |
528 | ASSERT(args->agbno % args->alignment == 0); | |
529 | ||
530 | if (!args->wasfromfl) { | |
531 | error = xfs_alloc_update_counters(args->tp, args->pag, | |
532 | args->agbp, | |
533 | -((long)(args->len))); | |
534 | if (error) | |
535 | return error; | |
536 | ||
4ecbfe63 | 537 | ASSERT(!xfs_extent_busy_search(args->mp, args->agno, |
e26f0501 | 538 | args->agbno, args->len)); |
1da177e4 | 539 | } |
ecb6928f CH |
540 | |
541 | if (!args->isfl) { | |
542 | xfs_trans_mod_sb(args->tp, args->wasdel ? | |
543 | XFS_TRANS_SB_RES_FDBLOCKS : | |
544 | XFS_TRANS_SB_FDBLOCKS, | |
545 | -((long)(args->len))); | |
546 | } | |
547 | ||
548 | XFS_STATS_INC(xs_allocx); | |
549 | XFS_STATS_ADD(xs_allocb, args->len); | |
550 | return error; | |
1da177e4 LT |
551 | } |
552 | ||
553 | /* | |
554 | * Allocate a variable extent at exactly agno/bno. | |
555 | * Extent's length (returned in *len) will be between minlen and maxlen, | |
556 | * and of the form k * prod + mod unless there's nothing that large. | |
557 | * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it. | |
558 | */ | |
559 | STATIC int /* error */ | |
560 | xfs_alloc_ag_vextent_exact( | |
561 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
562 | { | |
563 | xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */ | |
564 | xfs_btree_cur_t *cnt_cur;/* by count btree cursor */ | |
1da177e4 LT |
565 | int error; |
566 | xfs_agblock_t fbno; /* start block of found extent */ | |
1da177e4 | 567 | xfs_extlen_t flen; /* length of found extent */ |
e26f0501 CH |
568 | xfs_agblock_t tbno; /* start block of trimmed extent */ |
569 | xfs_extlen_t tlen; /* length of trimmed extent */ | |
570 | xfs_agblock_t tend; /* end block of trimmed extent */ | |
1da177e4 | 571 | int i; /* success/failure of operation */ |
1da177e4 LT |
572 | |
573 | ASSERT(args->alignment == 1); | |
9f9baab3 | 574 | |
1da177e4 LT |
575 | /* |
576 | * Allocate/initialize a cursor for the by-number freespace btree. | |
577 | */ | |
561f7d17 | 578 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
9f9baab3 CH |
579 | args->agno, XFS_BTNUM_BNO); |
580 | ||
1da177e4 LT |
581 | /* |
582 | * Lookup bno and minlen in the btree (minlen is irrelevant, really). | |
583 | * Look for the closest free block <= bno, it must contain bno | |
584 | * if any free block does. | |
585 | */ | |
9f9baab3 CH |
586 | error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i); |
587 | if (error) | |
1da177e4 | 588 | goto error0; |
9f9baab3 CH |
589 | if (!i) |
590 | goto not_found; | |
591 | ||
1da177e4 LT |
592 | /* |
593 | * Grab the freespace record. | |
594 | */ | |
9f9baab3 CH |
595 | error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i); |
596 | if (error) | |
1da177e4 LT |
597 | goto error0; |
598 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
599 | ASSERT(fbno <= args->agbno); | |
9f9baab3 | 600 | |
1da177e4 | 601 | /* |
e26f0501 | 602 | * Check for overlapping busy extents. |
1da177e4 | 603 | */ |
4ecbfe63 | 604 | xfs_extent_busy_trim(args, fbno, flen, &tbno, &tlen); |
e26f0501 CH |
605 | |
606 | /* | |
607 | * Give up if the start of the extent is busy, or the freespace isn't | |
608 | * long enough for the minimum request. | |
609 | */ | |
610 | if (tbno > args->agbno) | |
611 | goto not_found; | |
612 | if (tlen < args->minlen) | |
613 | goto not_found; | |
614 | tend = tbno + tlen; | |
615 | if (tend < args->agbno + args->minlen) | |
9f9baab3 CH |
616 | goto not_found; |
617 | ||
1da177e4 LT |
618 | /* |
619 | * End of extent will be smaller of the freespace end and the | |
620 | * maximal requested end. | |
9f9baab3 | 621 | * |
1da177e4 LT |
622 | * Fix the length according to mod and prod if given. |
623 | */ | |
81463b1c CS |
624 | args->len = XFS_AGBLOCK_MIN(tend, args->agbno + args->maxlen) |
625 | - args->agbno; | |
1da177e4 | 626 | xfs_alloc_fix_len(args); |
9f9baab3 CH |
627 | if (!xfs_alloc_fix_minleft(args)) |
628 | goto not_found; | |
629 | ||
81463b1c | 630 | ASSERT(args->agbno + args->len <= tend); |
9f9baab3 | 631 | |
1da177e4 | 632 | /* |
81463b1c | 633 | * We are allocating agbno for args->len |
1da177e4 LT |
634 | * Allocate/initialize a cursor for the by-size btree. |
635 | */ | |
561f7d17 CH |
636 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
637 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 638 | ASSERT(args->agbno + args->len <= |
16259e7d | 639 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
9f9baab3 CH |
640 | error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, args->agbno, |
641 | args->len, XFSA_FIXUP_BNO_OK); | |
642 | if (error) { | |
1da177e4 LT |
643 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); |
644 | goto error0; | |
645 | } | |
9f9baab3 | 646 | |
1da177e4 LT |
647 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); |
648 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 649 | |
1da177e4 | 650 | args->wasfromfl = 0; |
9f9baab3 CH |
651 | trace_xfs_alloc_exact_done(args); |
652 | return 0; | |
653 | ||
654 | not_found: | |
655 | /* Didn't find it, return null. */ | |
656 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
657 | args->agbno = NULLAGBLOCK; | |
658 | trace_xfs_alloc_exact_notfound(args); | |
1da177e4 LT |
659 | return 0; |
660 | ||
661 | error0: | |
662 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
0b1b213f | 663 | trace_xfs_alloc_exact_error(args); |
1da177e4 LT |
664 | return error; |
665 | } | |
666 | ||
489a150f CH |
667 | /* |
668 | * Search the btree in a given direction via the search cursor and compare | |
669 | * the records found against the good extent we've already found. | |
670 | */ | |
671 | STATIC int | |
672 | xfs_alloc_find_best_extent( | |
673 | struct xfs_alloc_arg *args, /* allocation argument structure */ | |
674 | struct xfs_btree_cur **gcur, /* good cursor */ | |
675 | struct xfs_btree_cur **scur, /* searching cursor */ | |
676 | xfs_agblock_t gdiff, /* difference for search comparison */ | |
677 | xfs_agblock_t *sbno, /* extent found by search */ | |
e26f0501 CH |
678 | xfs_extlen_t *slen, /* extent length */ |
679 | xfs_agblock_t *sbnoa, /* aligned extent found by search */ | |
680 | xfs_extlen_t *slena, /* aligned extent length */ | |
489a150f CH |
681 | int dir) /* 0 = search right, 1 = search left */ |
682 | { | |
489a150f CH |
683 | xfs_agblock_t new; |
684 | xfs_agblock_t sdiff; | |
685 | int error; | |
686 | int i; | |
687 | ||
688 | /* The good extent is perfect, no need to search. */ | |
689 | if (!gdiff) | |
690 | goto out_use_good; | |
691 | ||
692 | /* | |
693 | * Look until we find a better one, run out of space or run off the end. | |
694 | */ | |
695 | do { | |
696 | error = xfs_alloc_get_rec(*scur, sbno, slen, &i); | |
697 | if (error) | |
698 | goto error0; | |
699 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
e26f0501 | 700 | xfs_alloc_compute_aligned(args, *sbno, *slen, sbnoa, slena); |
489a150f CH |
701 | |
702 | /* | |
703 | * The good extent is closer than this one. | |
704 | */ | |
705 | if (!dir) { | |
e26f0501 | 706 | if (*sbnoa >= args->agbno + gdiff) |
489a150f CH |
707 | goto out_use_good; |
708 | } else { | |
e26f0501 | 709 | if (*sbnoa <= args->agbno - gdiff) |
489a150f CH |
710 | goto out_use_good; |
711 | } | |
712 | ||
713 | /* | |
714 | * Same distance, compare length and pick the best. | |
715 | */ | |
716 | if (*slena >= args->minlen) { | |
717 | args->len = XFS_EXTLEN_MIN(*slena, args->maxlen); | |
718 | xfs_alloc_fix_len(args); | |
719 | ||
720 | sdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
e26f0501 CH |
721 | args->alignment, *sbnoa, |
722 | *slena, &new); | |
489a150f CH |
723 | |
724 | /* | |
725 | * Choose closer size and invalidate other cursor. | |
726 | */ | |
727 | if (sdiff < gdiff) | |
728 | goto out_use_search; | |
729 | goto out_use_good; | |
730 | } | |
731 | ||
732 | if (!dir) | |
733 | error = xfs_btree_increment(*scur, 0, &i); | |
734 | else | |
735 | error = xfs_btree_decrement(*scur, 0, &i); | |
736 | if (error) | |
737 | goto error0; | |
738 | } while (i); | |
739 | ||
740 | out_use_good: | |
741 | xfs_btree_del_cursor(*scur, XFS_BTREE_NOERROR); | |
742 | *scur = NULL; | |
743 | return 0; | |
744 | ||
745 | out_use_search: | |
746 | xfs_btree_del_cursor(*gcur, XFS_BTREE_NOERROR); | |
747 | *gcur = NULL; | |
748 | return 0; | |
749 | ||
750 | error0: | |
751 | /* caller invalidates cursors */ | |
752 | return error; | |
753 | } | |
754 | ||
1da177e4 LT |
755 | /* |
756 | * Allocate a variable extent near bno in the allocation group agno. | |
757 | * Extent's length (returned in len) will be between minlen and maxlen, | |
758 | * and of the form k * prod + mod unless there's nothing that large. | |
759 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
760 | */ | |
761 | STATIC int /* error */ | |
762 | xfs_alloc_ag_vextent_near( | |
763 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
764 | { | |
765 | xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */ | |
766 | xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */ | |
767 | xfs_btree_cur_t *cnt_cur; /* cursor for count btree */ | |
1da177e4 LT |
768 | xfs_agblock_t gtbno; /* start bno of right side entry */ |
769 | xfs_agblock_t gtbnoa; /* aligned ... */ | |
770 | xfs_extlen_t gtdiff; /* difference to right side entry */ | |
771 | xfs_extlen_t gtlen; /* length of right side entry */ | |
e26f0501 | 772 | xfs_extlen_t gtlena; /* aligned ... */ |
1da177e4 LT |
773 | xfs_agblock_t gtnew; /* useful start bno of right side */ |
774 | int error; /* error code */ | |
775 | int i; /* result code, temporary */ | |
776 | int j; /* result code, temporary */ | |
777 | xfs_agblock_t ltbno; /* start bno of left side entry */ | |
778 | xfs_agblock_t ltbnoa; /* aligned ... */ | |
779 | xfs_extlen_t ltdiff; /* difference to left side entry */ | |
1da177e4 | 780 | xfs_extlen_t ltlen; /* length of left side entry */ |
e26f0501 | 781 | xfs_extlen_t ltlena; /* aligned ... */ |
1da177e4 LT |
782 | xfs_agblock_t ltnew; /* useful start bno of left side */ |
783 | xfs_extlen_t rlen; /* length of returned extent */ | |
e26f0501 | 784 | int forced = 0; |
1da177e4 LT |
785 | #if defined(DEBUG) && defined(__KERNEL__) |
786 | /* | |
787 | * Randomly don't execute the first algorithm. | |
788 | */ | |
789 | int dofirst; /* set to do first algorithm */ | |
790 | ||
e7a23a9b | 791 | dofirst = random32() & 1; |
1da177e4 | 792 | #endif |
e26f0501 CH |
793 | |
794 | restart: | |
795 | bno_cur_lt = NULL; | |
796 | bno_cur_gt = NULL; | |
797 | ltlen = 0; | |
798 | gtlena = 0; | |
799 | ltlena = 0; | |
800 | ||
1da177e4 LT |
801 | /* |
802 | * Get a cursor for the by-size btree. | |
803 | */ | |
561f7d17 CH |
804 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
805 | args->agno, XFS_BTNUM_CNT); | |
e26f0501 | 806 | |
1da177e4 LT |
807 | /* |
808 | * See if there are any free extents as big as maxlen. | |
809 | */ | |
810 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i))) | |
811 | goto error0; | |
812 | /* | |
813 | * If none, then pick up the last entry in the tree unless the | |
814 | * tree is empty. | |
815 | */ | |
816 | if (!i) { | |
817 | if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, <bno, | |
818 | <len, &i))) | |
819 | goto error0; | |
820 | if (i == 0 || ltlen == 0) { | |
821 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
e26f0501 | 822 | trace_xfs_alloc_near_noentry(args); |
1da177e4 LT |
823 | return 0; |
824 | } | |
825 | ASSERT(i == 1); | |
826 | } | |
827 | args->wasfromfl = 0; | |
e26f0501 | 828 | |
1da177e4 LT |
829 | /* |
830 | * First algorithm. | |
831 | * If the requested extent is large wrt the freespaces available | |
832 | * in this a.g., then the cursor will be pointing to a btree entry | |
833 | * near the right edge of the tree. If it's in the last btree leaf | |
834 | * block, then we just examine all the entries in that block | |
835 | * that are big enough, and pick the best one. | |
836 | * This is written as a while loop so we can break out of it, | |
837 | * but we never loop back to the top. | |
838 | */ | |
839 | while (xfs_btree_islastblock(cnt_cur, 0)) { | |
840 | xfs_extlen_t bdiff; | |
841 | int besti=0; | |
842 | xfs_extlen_t blen=0; | |
843 | xfs_agblock_t bnew=0; | |
844 | ||
845 | #if defined(DEBUG) && defined(__KERNEL__) | |
846 | if (!dofirst) | |
847 | break; | |
848 | #endif | |
849 | /* | |
850 | * Start from the entry that lookup found, sequence through | |
851 | * all larger free blocks. If we're actually pointing at a | |
852 | * record smaller than maxlen, go to the start of this block, | |
853 | * and skip all those smaller than minlen. | |
854 | */ | |
855 | if (ltlen || args->alignment > 1) { | |
856 | cnt_cur->bc_ptrs[0] = 1; | |
857 | do { | |
858 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, | |
859 | <len, &i))) | |
860 | goto error0; | |
861 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
862 | if (ltlen >= args->minlen) | |
863 | break; | |
637aa50f | 864 | if ((error = xfs_btree_increment(cnt_cur, 0, &i))) |
1da177e4 LT |
865 | goto error0; |
866 | } while (i); | |
867 | ASSERT(ltlen >= args->minlen); | |
868 | if (!i) | |
869 | break; | |
870 | } | |
871 | i = cnt_cur->bc_ptrs[0]; | |
872 | for (j = 1, blen = 0, bdiff = 0; | |
873 | !error && j && (blen < args->maxlen || bdiff > 0); | |
637aa50f | 874 | error = xfs_btree_increment(cnt_cur, 0, &j)) { |
1da177e4 LT |
875 | /* |
876 | * For each entry, decide if it's better than | |
877 | * the previous best entry. | |
878 | */ | |
879 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
880 | goto error0; | |
881 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
86fa8af6 CH |
882 | xfs_alloc_compute_aligned(args, ltbno, ltlen, |
883 | <bnoa, <lena); | |
e6430037 | 884 | if (ltlena < args->minlen) |
1da177e4 LT |
885 | continue; |
886 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
887 | xfs_alloc_fix_len(args); | |
888 | ASSERT(args->len >= args->minlen); | |
889 | if (args->len < blen) | |
890 | continue; | |
891 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, | |
e26f0501 | 892 | args->alignment, ltbnoa, ltlena, <new); |
1da177e4 LT |
893 | if (ltnew != NULLAGBLOCK && |
894 | (args->len > blen || ltdiff < bdiff)) { | |
895 | bdiff = ltdiff; | |
896 | bnew = ltnew; | |
897 | blen = args->len; | |
898 | besti = cnt_cur->bc_ptrs[0]; | |
899 | } | |
900 | } | |
901 | /* | |
902 | * It didn't work. We COULD be in a case where | |
903 | * there's a good record somewhere, so try again. | |
904 | */ | |
905 | if (blen == 0) | |
906 | break; | |
907 | /* | |
908 | * Point at the best entry, and retrieve it again. | |
909 | */ | |
910 | cnt_cur->bc_ptrs[0] = besti; | |
911 | if ((error = xfs_alloc_get_rec(cnt_cur, <bno, <len, &i))) | |
912 | goto error0; | |
913 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
73523a2e | 914 | ASSERT(ltbno + ltlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 LT |
915 | args->len = blen; |
916 | if (!xfs_alloc_fix_minleft(args)) { | |
917 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 918 | trace_xfs_alloc_near_nominleft(args); |
1da177e4 LT |
919 | return 0; |
920 | } | |
921 | blen = args->len; | |
922 | /* | |
923 | * We are allocating starting at bnew for blen blocks. | |
924 | */ | |
925 | args->agbno = bnew; | |
926 | ASSERT(bnew >= ltbno); | |
73523a2e | 927 | ASSERT(bnew + blen <= ltbno + ltlen); |
1da177e4 LT |
928 | /* |
929 | * Set up a cursor for the by-bno tree. | |
930 | */ | |
561f7d17 CH |
931 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, |
932 | args->agbp, args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
933 | /* |
934 | * Fix up the btree entries. | |
935 | */ | |
936 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, | |
937 | ltlen, bnew, blen, XFSA_FIXUP_CNT_OK))) | |
938 | goto error0; | |
939 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
940 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
0b1b213f CH |
941 | |
942 | trace_xfs_alloc_near_first(args); | |
1da177e4 LT |
943 | return 0; |
944 | } | |
945 | /* | |
946 | * Second algorithm. | |
947 | * Search in the by-bno tree to the left and to the right | |
948 | * simultaneously, until in each case we find a space big enough, | |
949 | * or run into the edge of the tree. When we run into the edge, | |
950 | * we deallocate that cursor. | |
951 | * If both searches succeed, we compare the two spaces and pick | |
952 | * the better one. | |
953 | * With alignment, it's possible for both to fail; the upper | |
954 | * level algorithm that picks allocation groups for allocations | |
955 | * is not supposed to do this. | |
956 | */ | |
957 | /* | |
958 | * Allocate and initialize the cursor for the leftward search. | |
959 | */ | |
561f7d17 CH |
960 | bno_cur_lt = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
961 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
962 | /* |
963 | * Lookup <= bno to find the leftward search's starting point. | |
964 | */ | |
965 | if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i))) | |
966 | goto error0; | |
967 | if (!i) { | |
968 | /* | |
969 | * Didn't find anything; use this cursor for the rightward | |
970 | * search. | |
971 | */ | |
972 | bno_cur_gt = bno_cur_lt; | |
973 | bno_cur_lt = NULL; | |
974 | } | |
975 | /* | |
976 | * Found something. Duplicate the cursor for the rightward search. | |
977 | */ | |
978 | else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt))) | |
979 | goto error0; | |
980 | /* | |
981 | * Increment the cursor, so we will point at the entry just right | |
982 | * of the leftward entry if any, or to the leftmost entry. | |
983 | */ | |
637aa50f | 984 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
985 | goto error0; |
986 | if (!i) { | |
987 | /* | |
988 | * It failed, there are no rightward entries. | |
989 | */ | |
990 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR); | |
991 | bno_cur_gt = NULL; | |
992 | } | |
993 | /* | |
994 | * Loop going left with the leftward cursor, right with the | |
995 | * rightward cursor, until either both directions give up or | |
996 | * we find an entry at least as big as minlen. | |
997 | */ | |
998 | do { | |
999 | if (bno_cur_lt) { | |
1000 | if ((error = xfs_alloc_get_rec(bno_cur_lt, <bno, <len, &i))) | |
1001 | goto error0; | |
1002 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
86fa8af6 CH |
1003 | xfs_alloc_compute_aligned(args, ltbno, ltlen, |
1004 | <bnoa, <lena); | |
12375c82 | 1005 | if (ltlena >= args->minlen) |
1da177e4 | 1006 | break; |
8df4da4a | 1007 | if ((error = xfs_btree_decrement(bno_cur_lt, 0, &i))) |
1da177e4 LT |
1008 | goto error0; |
1009 | if (!i) { | |
1010 | xfs_btree_del_cursor(bno_cur_lt, | |
1011 | XFS_BTREE_NOERROR); | |
1012 | bno_cur_lt = NULL; | |
1013 | } | |
1014 | } | |
1015 | if (bno_cur_gt) { | |
1016 | if ((error = xfs_alloc_get_rec(bno_cur_gt, >bno, >len, &i))) | |
1017 | goto error0; | |
1018 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
86fa8af6 CH |
1019 | xfs_alloc_compute_aligned(args, gtbno, gtlen, |
1020 | >bnoa, >lena); | |
12375c82 | 1021 | if (gtlena >= args->minlen) |
1da177e4 | 1022 | break; |
637aa50f | 1023 | if ((error = xfs_btree_increment(bno_cur_gt, 0, &i))) |
1da177e4 LT |
1024 | goto error0; |
1025 | if (!i) { | |
1026 | xfs_btree_del_cursor(bno_cur_gt, | |
1027 | XFS_BTREE_NOERROR); | |
1028 | bno_cur_gt = NULL; | |
1029 | } | |
1030 | } | |
1031 | } while (bno_cur_lt || bno_cur_gt); | |
489a150f | 1032 | |
1da177e4 LT |
1033 | /* |
1034 | * Got both cursors still active, need to find better entry. | |
1035 | */ | |
1036 | if (bno_cur_lt && bno_cur_gt) { | |
1da177e4 LT |
1037 | if (ltlena >= args->minlen) { |
1038 | /* | |
489a150f | 1039 | * Left side is good, look for a right side entry. |
1da177e4 LT |
1040 | */ |
1041 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); | |
1042 | xfs_alloc_fix_len(args); | |
489a150f | 1043 | ltdiff = xfs_alloc_compute_diff(args->agbno, args->len, |
e26f0501 | 1044 | args->alignment, ltbnoa, ltlena, <new); |
489a150f CH |
1045 | |
1046 | error = xfs_alloc_find_best_extent(args, | |
1047 | &bno_cur_lt, &bno_cur_gt, | |
e26f0501 CH |
1048 | ltdiff, >bno, >len, |
1049 | >bnoa, >lena, | |
489a150f CH |
1050 | 0 /* search right */); |
1051 | } else { | |
1052 | ASSERT(gtlena >= args->minlen); | |
1053 | ||
1da177e4 | 1054 | /* |
489a150f | 1055 | * Right side is good, look for a left side entry. |
1da177e4 LT |
1056 | */ |
1057 | args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen); | |
1058 | xfs_alloc_fix_len(args); | |
489a150f | 1059 | gtdiff = xfs_alloc_compute_diff(args->agbno, args->len, |
e26f0501 | 1060 | args->alignment, gtbnoa, gtlena, >new); |
489a150f CH |
1061 | |
1062 | error = xfs_alloc_find_best_extent(args, | |
1063 | &bno_cur_gt, &bno_cur_lt, | |
e26f0501 CH |
1064 | gtdiff, <bno, <len, |
1065 | <bnoa, <lena, | |
489a150f | 1066 | 1 /* search left */); |
1da177e4 | 1067 | } |
489a150f CH |
1068 | |
1069 | if (error) | |
1070 | goto error0; | |
1da177e4 | 1071 | } |
489a150f | 1072 | |
1da177e4 LT |
1073 | /* |
1074 | * If we couldn't get anything, give up. | |
1075 | */ | |
1076 | if (bno_cur_lt == NULL && bno_cur_gt == NULL) { | |
e3a746f5 DC |
1077 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1078 | ||
e26f0501 CH |
1079 | if (!forced++) { |
1080 | trace_xfs_alloc_near_busy(args); | |
1081 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1082 | goto restart; | |
1083 | } | |
0b1b213f | 1084 | trace_xfs_alloc_size_neither(args); |
1da177e4 LT |
1085 | args->agbno = NULLAGBLOCK; |
1086 | return 0; | |
1087 | } | |
489a150f | 1088 | |
1da177e4 LT |
1089 | /* |
1090 | * At this point we have selected a freespace entry, either to the | |
1091 | * left or to the right. If it's on the right, copy all the | |
1092 | * useful variables to the "left" set so we only have one | |
1093 | * copy of this code. | |
1094 | */ | |
1095 | if (bno_cur_gt) { | |
1096 | bno_cur_lt = bno_cur_gt; | |
1097 | bno_cur_gt = NULL; | |
1098 | ltbno = gtbno; | |
1099 | ltbnoa = gtbnoa; | |
1100 | ltlen = gtlen; | |
1101 | ltlena = gtlena; | |
1102 | j = 1; | |
1103 | } else | |
1104 | j = 0; | |
489a150f | 1105 | |
1da177e4 LT |
1106 | /* |
1107 | * Fix up the length and compute the useful address. | |
1108 | */ | |
1da177e4 LT |
1109 | args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen); |
1110 | xfs_alloc_fix_len(args); | |
1111 | if (!xfs_alloc_fix_minleft(args)) { | |
0b1b213f | 1112 | trace_xfs_alloc_near_nominleft(args); |
1da177e4 LT |
1113 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); |
1114 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1115 | return 0; | |
1116 | } | |
1117 | rlen = args->len; | |
e26f0501 CH |
1118 | (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, |
1119 | ltbnoa, ltlena, <new); | |
1da177e4 | 1120 | ASSERT(ltnew >= ltbno); |
e26f0501 | 1121 | ASSERT(ltnew + rlen <= ltbnoa + ltlena); |
16259e7d | 1122 | ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length)); |
1da177e4 | 1123 | args->agbno = ltnew; |
e26f0501 | 1124 | |
1da177e4 LT |
1125 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen, |
1126 | ltnew, rlen, XFSA_FIXUP_BNO_OK))) | |
1127 | goto error0; | |
0b1b213f CH |
1128 | |
1129 | if (j) | |
1130 | trace_xfs_alloc_near_greater(args); | |
1131 | else | |
1132 | trace_xfs_alloc_near_lesser(args); | |
1133 | ||
1da177e4 LT |
1134 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); |
1135 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR); | |
1136 | return 0; | |
1137 | ||
1138 | error0: | |
0b1b213f | 1139 | trace_xfs_alloc_near_error(args); |
1da177e4 LT |
1140 | if (cnt_cur != NULL) |
1141 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1142 | if (bno_cur_lt != NULL) | |
1143 | xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR); | |
1144 | if (bno_cur_gt != NULL) | |
1145 | xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR); | |
1146 | return error; | |
1147 | } | |
1148 | ||
1149 | /* | |
1150 | * Allocate a variable extent anywhere in the allocation group agno. | |
1151 | * Extent's length (returned in len) will be between minlen and maxlen, | |
1152 | * and of the form k * prod + mod unless there's nothing that large. | |
1153 | * Return the starting a.g. block, or NULLAGBLOCK if we can't do it. | |
1154 | */ | |
1155 | STATIC int /* error */ | |
1156 | xfs_alloc_ag_vextent_size( | |
1157 | xfs_alloc_arg_t *args) /* allocation argument structure */ | |
1158 | { | |
1159 | xfs_btree_cur_t *bno_cur; /* cursor for bno btree */ | |
1160 | xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */ | |
1161 | int error; /* error result */ | |
1162 | xfs_agblock_t fbno; /* start of found freespace */ | |
1163 | xfs_extlen_t flen; /* length of found freespace */ | |
1da177e4 LT |
1164 | int i; /* temp status variable */ |
1165 | xfs_agblock_t rbno; /* returned block number */ | |
1166 | xfs_extlen_t rlen; /* length of returned extent */ | |
e26f0501 | 1167 | int forced = 0; |
1da177e4 | 1168 | |
e26f0501 | 1169 | restart: |
1da177e4 LT |
1170 | /* |
1171 | * Allocate and initialize a cursor for the by-size btree. | |
1172 | */ | |
561f7d17 CH |
1173 | cnt_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1174 | args->agno, XFS_BTNUM_CNT); | |
1da177e4 | 1175 | bno_cur = NULL; |
e26f0501 | 1176 | |
1da177e4 LT |
1177 | /* |
1178 | * Look for an entry >= maxlen+alignment-1 blocks. | |
1179 | */ | |
1180 | if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, | |
1181 | args->maxlen + args->alignment - 1, &i))) | |
1182 | goto error0; | |
e26f0501 | 1183 | |
1da177e4 | 1184 | /* |
e26f0501 CH |
1185 | * If none or we have busy extents that we cannot allocate from, then |
1186 | * we have to settle for a smaller extent. In the case that there are | |
1187 | * no large extents, this will return the last entry in the tree unless | |
1188 | * the tree is empty. In the case that there are only busy large | |
1189 | * extents, this will return the largest small extent unless there | |
1190 | * are no smaller extents available. | |
1da177e4 | 1191 | */ |
e26f0501 CH |
1192 | if (!i || forced > 1) { |
1193 | error = xfs_alloc_ag_vextent_small(args, cnt_cur, | |
1194 | &fbno, &flen, &i); | |
1195 | if (error) | |
1da177e4 LT |
1196 | goto error0; |
1197 | if (i == 0 || flen == 0) { | |
1198 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
0b1b213f | 1199 | trace_xfs_alloc_size_noentry(args); |
1da177e4 LT |
1200 | return 0; |
1201 | } | |
1202 | ASSERT(i == 1); | |
e26f0501 CH |
1203 | xfs_alloc_compute_aligned(args, fbno, flen, &rbno, &rlen); |
1204 | } else { | |
1205 | /* | |
1206 | * Search for a non-busy extent that is large enough. | |
1207 | * If we are at low space, don't check, or if we fall of | |
1208 | * the end of the btree, turn off the busy check and | |
1209 | * restart. | |
1210 | */ | |
1211 | for (;;) { | |
1212 | error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i); | |
1213 | if (error) | |
1214 | goto error0; | |
1215 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1216 | ||
1217 | xfs_alloc_compute_aligned(args, fbno, flen, | |
1218 | &rbno, &rlen); | |
1219 | ||
1220 | if (rlen >= args->maxlen) | |
1221 | break; | |
1222 | ||
1223 | error = xfs_btree_increment(cnt_cur, 0, &i); | |
1224 | if (error) | |
1225 | goto error0; | |
1226 | if (i == 0) { | |
1227 | /* | |
1228 | * Our only valid extents must have been busy. | |
1229 | * Make it unbusy by forcing the log out and | |
1230 | * retrying. If we've been here before, forcing | |
1231 | * the log isn't making the extents available, | |
1232 | * which means they have probably been freed in | |
1233 | * this transaction. In that case, we have to | |
1234 | * give up on them and we'll attempt a minlen | |
1235 | * allocation the next time around. | |
1236 | */ | |
1237 | xfs_btree_del_cursor(cnt_cur, | |
1238 | XFS_BTREE_NOERROR); | |
1239 | trace_xfs_alloc_size_busy(args); | |
1240 | if (!forced++) | |
1241 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1242 | goto restart; | |
1243 | } | |
1244 | } | |
1da177e4 | 1245 | } |
e26f0501 | 1246 | |
1da177e4 LT |
1247 | /* |
1248 | * In the first case above, we got the last entry in the | |
1249 | * by-size btree. Now we check to see if the space hits maxlen | |
1250 | * once aligned; if not, we search left for something better. | |
1251 | * This can't happen in the second case above. | |
1252 | */ | |
1da177e4 LT |
1253 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
1254 | XFS_WANT_CORRUPTED_GOTO(rlen == 0 || | |
1255 | (rlen <= flen && rbno + rlen <= fbno + flen), error0); | |
1256 | if (rlen < args->maxlen) { | |
1257 | xfs_agblock_t bestfbno; | |
1258 | xfs_extlen_t bestflen; | |
1259 | xfs_agblock_t bestrbno; | |
1260 | xfs_extlen_t bestrlen; | |
1261 | ||
1262 | bestrlen = rlen; | |
1263 | bestrbno = rbno; | |
1264 | bestflen = flen; | |
1265 | bestfbno = fbno; | |
1266 | for (;;) { | |
8df4da4a | 1267 | if ((error = xfs_btree_decrement(cnt_cur, 0, &i))) |
1da177e4 LT |
1268 | goto error0; |
1269 | if (i == 0) | |
1270 | break; | |
1271 | if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, | |
1272 | &i))) | |
1273 | goto error0; | |
1274 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1275 | if (flen < bestrlen) | |
1276 | break; | |
86fa8af6 CH |
1277 | xfs_alloc_compute_aligned(args, fbno, flen, |
1278 | &rbno, &rlen); | |
1da177e4 LT |
1279 | rlen = XFS_EXTLEN_MIN(args->maxlen, rlen); |
1280 | XFS_WANT_CORRUPTED_GOTO(rlen == 0 || | |
1281 | (rlen <= flen && rbno + rlen <= fbno + flen), | |
1282 | error0); | |
1283 | if (rlen > bestrlen) { | |
1284 | bestrlen = rlen; | |
1285 | bestrbno = rbno; | |
1286 | bestflen = flen; | |
1287 | bestfbno = fbno; | |
1288 | if (rlen == args->maxlen) | |
1289 | break; | |
1290 | } | |
1291 | } | |
1292 | if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen, | |
1293 | &i))) | |
1294 | goto error0; | |
1295 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1296 | rlen = bestrlen; | |
1297 | rbno = bestrbno; | |
1298 | flen = bestflen; | |
1299 | fbno = bestfbno; | |
1300 | } | |
1301 | args->wasfromfl = 0; | |
1302 | /* | |
1303 | * Fix up the length. | |
1304 | */ | |
1305 | args->len = rlen; | |
e26f0501 CH |
1306 | if (rlen < args->minlen) { |
1307 | if (!forced++) { | |
1308 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1309 | trace_xfs_alloc_size_busy(args); | |
1310 | xfs_log_force(args->mp, XFS_LOG_SYNC); | |
1311 | goto restart; | |
1312 | } | |
1313 | goto out_nominleft; | |
1da177e4 | 1314 | } |
e26f0501 CH |
1315 | xfs_alloc_fix_len(args); |
1316 | ||
1317 | if (!xfs_alloc_fix_minleft(args)) | |
1318 | goto out_nominleft; | |
1da177e4 LT |
1319 | rlen = args->len; |
1320 | XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0); | |
1321 | /* | |
1322 | * Allocate and initialize a cursor for the by-block tree. | |
1323 | */ | |
561f7d17 CH |
1324 | bno_cur = xfs_allocbt_init_cursor(args->mp, args->tp, args->agbp, |
1325 | args->agno, XFS_BTNUM_BNO); | |
1da177e4 LT |
1326 | if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen, |
1327 | rbno, rlen, XFSA_FIXUP_CNT_OK))) | |
1328 | goto error0; | |
1329 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1330 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1331 | cnt_cur = bno_cur = NULL; | |
1332 | args->len = rlen; | |
1333 | args->agbno = rbno; | |
1334 | XFS_WANT_CORRUPTED_GOTO( | |
1335 | args->agbno + args->len <= | |
16259e7d | 1336 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 | 1337 | error0); |
0b1b213f | 1338 | trace_xfs_alloc_size_done(args); |
1da177e4 LT |
1339 | return 0; |
1340 | ||
1341 | error0: | |
0b1b213f | 1342 | trace_xfs_alloc_size_error(args); |
1da177e4 LT |
1343 | if (cnt_cur) |
1344 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1345 | if (bno_cur) | |
1346 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1347 | return error; | |
e26f0501 CH |
1348 | |
1349 | out_nominleft: | |
1350 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1351 | trace_xfs_alloc_size_nominleft(args); | |
1352 | args->agbno = NULLAGBLOCK; | |
1353 | return 0; | |
1da177e4 LT |
1354 | } |
1355 | ||
1356 | /* | |
1357 | * Deal with the case where only small freespaces remain. | |
1358 | * Either return the contents of the last freespace record, | |
1359 | * or allocate space from the freelist if there is nothing in the tree. | |
1360 | */ | |
1361 | STATIC int /* error */ | |
1362 | xfs_alloc_ag_vextent_small( | |
1363 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1364 | xfs_btree_cur_t *ccur, /* by-size cursor */ | |
1365 | xfs_agblock_t *fbnop, /* result block number */ | |
1366 | xfs_extlen_t *flenp, /* result length */ | |
1367 | int *stat) /* status: 0-freelist, 1-normal/none */ | |
1368 | { | |
1369 | int error; | |
1370 | xfs_agblock_t fbno; | |
1371 | xfs_extlen_t flen; | |
1da177e4 LT |
1372 | int i; |
1373 | ||
8df4da4a | 1374 | if ((error = xfs_btree_decrement(ccur, 0, &i))) |
1da177e4 LT |
1375 | goto error0; |
1376 | if (i) { | |
1377 | if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i))) | |
1378 | goto error0; | |
1379 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1380 | } | |
1381 | /* | |
1382 | * Nothing in the btree, try the freelist. Make sure | |
1383 | * to respect minleft even when pulling from the | |
1384 | * freelist. | |
1385 | */ | |
1386 | else if (args->minlen == 1 && args->alignment == 1 && !args->isfl && | |
16259e7d CH |
1387 | (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount) |
1388 | > args->minleft)) { | |
92821e2b DC |
1389 | error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0); |
1390 | if (error) | |
1da177e4 LT |
1391 | goto error0; |
1392 | if (fbno != NULLAGBLOCK) { | |
4ecbfe63 | 1393 | xfs_extent_busy_reuse(args->mp, args->agno, fbno, 1, |
97d3ac75 CH |
1394 | args->userdata); |
1395 | ||
1da177e4 LT |
1396 | if (args->userdata) { |
1397 | xfs_buf_t *bp; | |
1398 | ||
1399 | bp = xfs_btree_get_bufs(args->mp, args->tp, | |
1400 | args->agno, fbno, 0); | |
1401 | xfs_trans_binval(args->tp, bp); | |
1402 | } | |
1403 | args->len = 1; | |
1404 | args->agbno = fbno; | |
1405 | XFS_WANT_CORRUPTED_GOTO( | |
1406 | args->agbno + args->len <= | |
16259e7d | 1407 | be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length), |
1da177e4 LT |
1408 | error0); |
1409 | args->wasfromfl = 1; | |
0b1b213f | 1410 | trace_xfs_alloc_small_freelist(args); |
1da177e4 LT |
1411 | *stat = 0; |
1412 | return 0; | |
1413 | } | |
1414 | /* | |
1415 | * Nothing in the freelist. | |
1416 | */ | |
1417 | else | |
1418 | flen = 0; | |
1419 | } | |
1420 | /* | |
1421 | * Can't allocate from the freelist for some reason. | |
1422 | */ | |
d432c80e NS |
1423 | else { |
1424 | fbno = NULLAGBLOCK; | |
1da177e4 | 1425 | flen = 0; |
d432c80e | 1426 | } |
1da177e4 LT |
1427 | /* |
1428 | * Can't do the allocation, give up. | |
1429 | */ | |
1430 | if (flen < args->minlen) { | |
1431 | args->agbno = NULLAGBLOCK; | |
0b1b213f | 1432 | trace_xfs_alloc_small_notenough(args); |
1da177e4 LT |
1433 | flen = 0; |
1434 | } | |
1435 | *fbnop = fbno; | |
1436 | *flenp = flen; | |
1437 | *stat = 1; | |
0b1b213f | 1438 | trace_xfs_alloc_small_done(args); |
1da177e4 LT |
1439 | return 0; |
1440 | ||
1441 | error0: | |
0b1b213f | 1442 | trace_xfs_alloc_small_error(args); |
1da177e4 LT |
1443 | return error; |
1444 | } | |
1445 | ||
1446 | /* | |
1447 | * Free the extent starting at agno/bno for length. | |
1448 | */ | |
1449 | STATIC int /* error */ | |
1450 | xfs_free_ag_extent( | |
1451 | xfs_trans_t *tp, /* transaction pointer */ | |
1452 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
1453 | xfs_agnumber_t agno, /* allocation group number */ | |
1454 | xfs_agblock_t bno, /* starting block number */ | |
1455 | xfs_extlen_t len, /* length of extent */ | |
1456 | int isfl) /* set if is freelist blocks - no sb acctg */ | |
1457 | { | |
1458 | xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */ | |
1459 | xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */ | |
1460 | int error; /* error return value */ | |
1da177e4 LT |
1461 | xfs_agblock_t gtbno; /* start of right neighbor block */ |
1462 | xfs_extlen_t gtlen; /* length of right neighbor block */ | |
1463 | int haveleft; /* have a left neighbor block */ | |
1464 | int haveright; /* have a right neighbor block */ | |
1465 | int i; /* temp, result code */ | |
1466 | xfs_agblock_t ltbno; /* start of left neighbor block */ | |
1467 | xfs_extlen_t ltlen; /* length of left neighbor block */ | |
1468 | xfs_mount_t *mp; /* mount point struct for filesystem */ | |
1469 | xfs_agblock_t nbno; /* new starting block of freespace */ | |
1470 | xfs_extlen_t nlen; /* new length of freespace */ | |
ecb6928f | 1471 | xfs_perag_t *pag; /* per allocation group data */ |
1da177e4 LT |
1472 | |
1473 | mp = tp->t_mountp; | |
1474 | /* | |
1475 | * Allocate and initialize a cursor for the by-block btree. | |
1476 | */ | |
561f7d17 | 1477 | bno_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO); |
1da177e4 LT |
1478 | cnt_cur = NULL; |
1479 | /* | |
1480 | * Look for a neighboring block on the left (lower block numbers) | |
1481 | * that is contiguous with this space. | |
1482 | */ | |
1483 | if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft))) | |
1484 | goto error0; | |
1485 | if (haveleft) { | |
1486 | /* | |
1487 | * There is a block to our left. | |
1488 | */ | |
1489 | if ((error = xfs_alloc_get_rec(bno_cur, <bno, <len, &i))) | |
1490 | goto error0; | |
1491 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1492 | /* | |
1493 | * It's not contiguous, though. | |
1494 | */ | |
1495 | if (ltbno + ltlen < bno) | |
1496 | haveleft = 0; | |
1497 | else { | |
1498 | /* | |
1499 | * If this failure happens the request to free this | |
1500 | * space was invalid, it's (partly) already free. | |
1501 | * Very bad. | |
1502 | */ | |
1503 | XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0); | |
1504 | } | |
1505 | } | |
1506 | /* | |
1507 | * Look for a neighboring block on the right (higher block numbers) | |
1508 | * that is contiguous with this space. | |
1509 | */ | |
637aa50f | 1510 | if ((error = xfs_btree_increment(bno_cur, 0, &haveright))) |
1da177e4 LT |
1511 | goto error0; |
1512 | if (haveright) { | |
1513 | /* | |
1514 | * There is a block to our right. | |
1515 | */ | |
1516 | if ((error = xfs_alloc_get_rec(bno_cur, >bno, >len, &i))) | |
1517 | goto error0; | |
1518 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1519 | /* | |
1520 | * It's not contiguous, though. | |
1521 | */ | |
1522 | if (bno + len < gtbno) | |
1523 | haveright = 0; | |
1524 | else { | |
1525 | /* | |
1526 | * If this failure happens the request to free this | |
1527 | * space was invalid, it's (partly) already free. | |
1528 | * Very bad. | |
1529 | */ | |
1530 | XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0); | |
1531 | } | |
1532 | } | |
1533 | /* | |
1534 | * Now allocate and initialize a cursor for the by-size tree. | |
1535 | */ | |
561f7d17 | 1536 | cnt_cur = xfs_allocbt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT); |
1da177e4 LT |
1537 | /* |
1538 | * Have both left and right contiguous neighbors. | |
1539 | * Merge all three into a single free block. | |
1540 | */ | |
1541 | if (haveleft && haveright) { | |
1542 | /* | |
1543 | * Delete the old by-size entry on the left. | |
1544 | */ | |
1545 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1546 | goto error0; | |
1547 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1548 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1549 | goto error0; |
1550 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1551 | /* | |
1552 | * Delete the old by-size entry on the right. | |
1553 | */ | |
1554 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1555 | goto error0; | |
1556 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1557 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1558 | goto error0; |
1559 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1560 | /* | |
1561 | * Delete the old by-block entry for the right block. | |
1562 | */ | |
91cca5df | 1563 | if ((error = xfs_btree_delete(bno_cur, &i))) |
1da177e4 LT |
1564 | goto error0; |
1565 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1566 | /* | |
1567 | * Move the by-block cursor back to the left neighbor. | |
1568 | */ | |
8df4da4a | 1569 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 LT |
1570 | goto error0; |
1571 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1572 | #ifdef DEBUG | |
1573 | /* | |
1574 | * Check that this is the right record: delete didn't | |
1575 | * mangle the cursor. | |
1576 | */ | |
1577 | { | |
1578 | xfs_agblock_t xxbno; | |
1579 | xfs_extlen_t xxlen; | |
1580 | ||
1581 | if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen, | |
1582 | &i))) | |
1583 | goto error0; | |
1584 | XFS_WANT_CORRUPTED_GOTO( | |
1585 | i == 1 && xxbno == ltbno && xxlen == ltlen, | |
1586 | error0); | |
1587 | } | |
1588 | #endif | |
1589 | /* | |
1590 | * Update remaining by-block entry to the new, joined block. | |
1591 | */ | |
1592 | nbno = ltbno; | |
1593 | nlen = len + ltlen + gtlen; | |
1594 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1595 | goto error0; | |
1596 | } | |
1597 | /* | |
1598 | * Have only a left contiguous neighbor. | |
1599 | * Merge it together with the new freespace. | |
1600 | */ | |
1601 | else if (haveleft) { | |
1602 | /* | |
1603 | * Delete the old by-size entry on the left. | |
1604 | */ | |
1605 | if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i))) | |
1606 | goto error0; | |
1607 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1608 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1609 | goto error0; |
1610 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1611 | /* | |
1612 | * Back up the by-block cursor to the left neighbor, and | |
1613 | * update its length. | |
1614 | */ | |
8df4da4a | 1615 | if ((error = xfs_btree_decrement(bno_cur, 0, &i))) |
1da177e4 LT |
1616 | goto error0; |
1617 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1618 | nbno = ltbno; | |
1619 | nlen = len + ltlen; | |
1620 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1621 | goto error0; | |
1622 | } | |
1623 | /* | |
1624 | * Have only a right contiguous neighbor. | |
1625 | * Merge it together with the new freespace. | |
1626 | */ | |
1627 | else if (haveright) { | |
1628 | /* | |
1629 | * Delete the old by-size entry on the right. | |
1630 | */ | |
1631 | if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i))) | |
1632 | goto error0; | |
1633 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
91cca5df | 1634 | if ((error = xfs_btree_delete(cnt_cur, &i))) |
1da177e4 LT |
1635 | goto error0; |
1636 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1637 | /* | |
1638 | * Update the starting block and length of the right | |
1639 | * neighbor in the by-block tree. | |
1640 | */ | |
1641 | nbno = bno; | |
1642 | nlen = len + gtlen; | |
1643 | if ((error = xfs_alloc_update(bno_cur, nbno, nlen))) | |
1644 | goto error0; | |
1645 | } | |
1646 | /* | |
1647 | * No contiguous neighbors. | |
1648 | * Insert the new freespace into the by-block tree. | |
1649 | */ | |
1650 | else { | |
1651 | nbno = bno; | |
1652 | nlen = len; | |
4b22a571 | 1653 | if ((error = xfs_btree_insert(bno_cur, &i))) |
1da177e4 LT |
1654 | goto error0; |
1655 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1656 | } | |
1657 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR); | |
1658 | bno_cur = NULL; | |
1659 | /* | |
1660 | * In all cases we need to insert the new freespace in the by-size tree. | |
1661 | */ | |
1662 | if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i))) | |
1663 | goto error0; | |
1664 | XFS_WANT_CORRUPTED_GOTO(i == 0, error0); | |
4b22a571 | 1665 | if ((error = xfs_btree_insert(cnt_cur, &i))) |
1da177e4 LT |
1666 | goto error0; |
1667 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1668 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR); | |
1669 | cnt_cur = NULL; | |
ecb6928f | 1670 | |
1da177e4 LT |
1671 | /* |
1672 | * Update the freespace totals in the ag and superblock. | |
1673 | */ | |
ecb6928f CH |
1674 | pag = xfs_perag_get(mp, agno); |
1675 | error = xfs_alloc_update_counters(tp, pag, agbp, len); | |
1676 | xfs_perag_put(pag); | |
1677 | if (error) | |
1678 | goto error0; | |
1679 | ||
1680 | if (!isfl) | |
1681 | xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len); | |
1682 | XFS_STATS_INC(xs_freex); | |
1683 | XFS_STATS_ADD(xs_freeb, len); | |
0b1b213f CH |
1684 | |
1685 | trace_xfs_free_extent(mp, agno, bno, len, isfl, haveleft, haveright); | |
1da177e4 | 1686 | |
1da177e4 LT |
1687 | return 0; |
1688 | ||
1689 | error0: | |
0b1b213f | 1690 | trace_xfs_free_extent(mp, agno, bno, len, isfl, -1, -1); |
1da177e4 LT |
1691 | if (bno_cur) |
1692 | xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR); | |
1693 | if (cnt_cur) | |
1694 | xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR); | |
1695 | return error; | |
1696 | } | |
1697 | ||
1698 | /* | |
1699 | * Visible (exported) allocation/free functions. | |
1700 | * Some of these are used just by xfs_alloc_btree.c and this file. | |
1701 | */ | |
1702 | ||
1703 | /* | |
1704 | * Compute and fill in value of m_ag_maxlevels. | |
1705 | */ | |
1706 | void | |
1707 | xfs_alloc_compute_maxlevels( | |
1708 | xfs_mount_t *mp) /* file system mount structure */ | |
1709 | { | |
1710 | int level; | |
1711 | uint maxblocks; | |
1712 | uint maxleafents; | |
1713 | int minleafrecs; | |
1714 | int minnoderecs; | |
1715 | ||
1716 | maxleafents = (mp->m_sb.sb_agblocks + 1) / 2; | |
1717 | minleafrecs = mp->m_alloc_mnr[0]; | |
1718 | minnoderecs = mp->m_alloc_mnr[1]; | |
1719 | maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs; | |
1720 | for (level = 1; maxblocks > 1; level++) | |
1721 | maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs; | |
1722 | mp->m_ag_maxlevels = level; | |
1723 | } | |
1724 | ||
6cc87645 DC |
1725 | /* |
1726 | * Find the length of the longest extent in an AG. | |
1727 | */ | |
1728 | xfs_extlen_t | |
1729 | xfs_alloc_longest_free_extent( | |
1730 | struct xfs_mount *mp, | |
1731 | struct xfs_perag *pag) | |
1732 | { | |
1733 | xfs_extlen_t need, delta = 0; | |
1734 | ||
1735 | need = XFS_MIN_FREELIST_PAG(pag, mp); | |
1736 | if (need > pag->pagf_flcount) | |
1737 | delta = need - pag->pagf_flcount; | |
1738 | ||
1739 | if (pag->pagf_longest > delta) | |
1740 | return pag->pagf_longest - delta; | |
1741 | return pag->pagf_flcount > 0 || pag->pagf_longest > 0; | |
1742 | } | |
1743 | ||
1da177e4 LT |
1744 | /* |
1745 | * Decide whether to use this allocation group for this allocation. | |
1746 | * If so, fix up the btree freelist's size. | |
1747 | */ | |
1748 | STATIC int /* error */ | |
1749 | xfs_alloc_fix_freelist( | |
1750 | xfs_alloc_arg_t *args, /* allocation argument structure */ | |
1751 | int flags) /* XFS_ALLOC_FLAG_... */ | |
1752 | { | |
1753 | xfs_buf_t *agbp; /* agf buffer pointer */ | |
1754 | xfs_agf_t *agf; /* a.g. freespace structure pointer */ | |
1755 | xfs_buf_t *agflbp;/* agfl buffer pointer */ | |
1756 | xfs_agblock_t bno; /* freelist block */ | |
1757 | xfs_extlen_t delta; /* new blocks needed in freelist */ | |
1758 | int error; /* error result code */ | |
1759 | xfs_extlen_t longest;/* longest extent in allocation group */ | |
1760 | xfs_mount_t *mp; /* file system mount point structure */ | |
1761 | xfs_extlen_t need; /* total blocks needed in freelist */ | |
1762 | xfs_perag_t *pag; /* per-ag information structure */ | |
1763 | xfs_alloc_arg_t targs; /* local allocation arguments */ | |
1764 | xfs_trans_t *tp; /* transaction pointer */ | |
1765 | ||
1766 | mp = args->mp; | |
1767 | ||
1768 | pag = args->pag; | |
1769 | tp = args->tp; | |
1770 | if (!pag->pagf_init) { | |
1771 | if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags, | |
1772 | &agbp))) | |
1773 | return error; | |
1774 | if (!pag->pagf_init) { | |
0e1edbd9 NS |
1775 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1776 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1777 | args->agbp = NULL; |
1778 | return 0; | |
1779 | } | |
1780 | } else | |
1781 | agbp = NULL; | |
1782 | ||
0e1edbd9 NS |
1783 | /* |
1784 | * If this is a metadata preferred pag and we are user data | |
1da177e4 LT |
1785 | * then try somewhere else if we are not being asked to |
1786 | * try harder at this point | |
1787 | */ | |
0e1edbd9 NS |
1788 | if (pag->pagf_metadata && args->userdata && |
1789 | (flags & XFS_ALLOC_FLAG_TRYLOCK)) { | |
1790 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1791 | args->agbp = NULL; |
1792 | return 0; | |
1793 | } | |
1794 | ||
0e1edbd9 | 1795 | if (!(flags & XFS_ALLOC_FLAG_FREEING)) { |
0e1edbd9 NS |
1796 | /* |
1797 | * If it looks like there isn't a long enough extent, or enough | |
1798 | * total blocks, reject it. | |
1799 | */ | |
6cc87645 DC |
1800 | need = XFS_MIN_FREELIST_PAG(pag, mp); |
1801 | longest = xfs_alloc_longest_free_extent(mp, pag); | |
0e1edbd9 NS |
1802 | if ((args->minlen + args->alignment + args->minalignslop - 1) > |
1803 | longest || | |
1804 | ((int)(pag->pagf_freeblks + pag->pagf_flcount - | |
1805 | need - args->total) < (int)args->minleft)) { | |
1806 | if (agbp) | |
1807 | xfs_trans_brelse(tp, agbp); | |
1808 | args->agbp = NULL; | |
1809 | return 0; | |
1810 | } | |
1da177e4 | 1811 | } |
0e1edbd9 | 1812 | |
1da177e4 LT |
1813 | /* |
1814 | * Get the a.g. freespace buffer. | |
1815 | * Can fail if we're not blocking on locks, and it's held. | |
1816 | */ | |
1817 | if (agbp == NULL) { | |
1818 | if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags, | |
1819 | &agbp))) | |
1820 | return error; | |
1821 | if (agbp == NULL) { | |
0e1edbd9 NS |
1822 | ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK); |
1823 | ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING)); | |
1da177e4 LT |
1824 | args->agbp = NULL; |
1825 | return 0; | |
1826 | } | |
1827 | } | |
1828 | /* | |
1829 | * Figure out how many blocks we should have in the freelist. | |
1830 | */ | |
1831 | agf = XFS_BUF_TO_AGF(agbp); | |
1832 | need = XFS_MIN_FREELIST(agf, mp); | |
1da177e4 LT |
1833 | /* |
1834 | * If there isn't enough total or single-extent, reject it. | |
1835 | */ | |
0e1edbd9 NS |
1836 | if (!(flags & XFS_ALLOC_FLAG_FREEING)) { |
1837 | delta = need > be32_to_cpu(agf->agf_flcount) ? | |
1838 | (need - be32_to_cpu(agf->agf_flcount)) : 0; | |
1839 | longest = be32_to_cpu(agf->agf_longest); | |
1840 | longest = (longest > delta) ? (longest - delta) : | |
1841 | (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0); | |
1842 | if ((args->minlen + args->alignment + args->minalignslop - 1) > | |
1843 | longest || | |
1844 | ((int)(be32_to_cpu(agf->agf_freeblks) + | |
1845 | be32_to_cpu(agf->agf_flcount) - need - args->total) < | |
1846 | (int)args->minleft)) { | |
1847 | xfs_trans_brelse(tp, agbp); | |
1848 | args->agbp = NULL; | |
1849 | return 0; | |
1850 | } | |
1da177e4 LT |
1851 | } |
1852 | /* | |
1853 | * Make the freelist shorter if it's too long. | |
1854 | */ | |
16259e7d | 1855 | while (be32_to_cpu(agf->agf_flcount) > need) { |
1da177e4 LT |
1856 | xfs_buf_t *bp; |
1857 | ||
92821e2b DC |
1858 | error = xfs_alloc_get_freelist(tp, agbp, &bno, 0); |
1859 | if (error) | |
1da177e4 LT |
1860 | return error; |
1861 | if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1))) | |
1862 | return error; | |
1863 | bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0); | |
1864 | xfs_trans_binval(tp, bp); | |
1865 | } | |
1866 | /* | |
1867 | * Initialize the args structure. | |
1868 | */ | |
1869 | targs.tp = tp; | |
1870 | targs.mp = mp; | |
1871 | targs.agbp = agbp; | |
1872 | targs.agno = args->agno; | |
1873 | targs.mod = targs.minleft = targs.wasdel = targs.userdata = | |
1874 | targs.minalignslop = 0; | |
1875 | targs.alignment = targs.minlen = targs.prod = targs.isfl = 1; | |
1876 | targs.type = XFS_ALLOCTYPE_THIS_AG; | |
1877 | targs.pag = pag; | |
1878 | if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp))) | |
1879 | return error; | |
1880 | /* | |
1881 | * Make the freelist longer if it's too short. | |
1882 | */ | |
16259e7d | 1883 | while (be32_to_cpu(agf->agf_flcount) < need) { |
1da177e4 | 1884 | targs.agbno = 0; |
16259e7d | 1885 | targs.maxlen = need - be32_to_cpu(agf->agf_flcount); |
1da177e4 LT |
1886 | /* |
1887 | * Allocate as many blocks as possible at once. | |
1888 | */ | |
e63a3690 NS |
1889 | if ((error = xfs_alloc_ag_vextent(&targs))) { |
1890 | xfs_trans_brelse(tp, agflbp); | |
1da177e4 | 1891 | return error; |
e63a3690 | 1892 | } |
1da177e4 LT |
1893 | /* |
1894 | * Stop if we run out. Won't happen if callers are obeying | |
1895 | * the restrictions correctly. Can happen for free calls | |
1896 | * on a completely full ag. | |
1897 | */ | |
d210a28c | 1898 | if (targs.agbno == NULLAGBLOCK) { |
0e1edbd9 NS |
1899 | if (flags & XFS_ALLOC_FLAG_FREEING) |
1900 | break; | |
1901 | xfs_trans_brelse(tp, agflbp); | |
1902 | args->agbp = NULL; | |
1903 | return 0; | |
d210a28c | 1904 | } |
1da177e4 LT |
1905 | /* |
1906 | * Put each allocated block on the list. | |
1907 | */ | |
1908 | for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) { | |
92821e2b DC |
1909 | error = xfs_alloc_put_freelist(tp, agbp, |
1910 | agflbp, bno, 0); | |
1911 | if (error) | |
1da177e4 LT |
1912 | return error; |
1913 | } | |
1914 | } | |
e63a3690 | 1915 | xfs_trans_brelse(tp, agflbp); |
1da177e4 LT |
1916 | args->agbp = agbp; |
1917 | return 0; | |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * Get a block from the freelist. | |
1922 | * Returns with the buffer for the block gotten. | |
1923 | */ | |
1924 | int /* error */ | |
1925 | xfs_alloc_get_freelist( | |
1926 | xfs_trans_t *tp, /* transaction pointer */ | |
1927 | xfs_buf_t *agbp, /* buffer containing the agf structure */ | |
92821e2b DC |
1928 | xfs_agblock_t *bnop, /* block address retrieved from freelist */ |
1929 | int btreeblk) /* destination is a AGF btree */ | |
1da177e4 LT |
1930 | { |
1931 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
1932 | xfs_agfl_t *agfl; /* a.g. freelist structure */ | |
1933 | xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */ | |
1934 | xfs_agblock_t bno; /* block number returned */ | |
1935 | int error; | |
92821e2b | 1936 | int logflags; |
1da177e4 LT |
1937 | xfs_mount_t *mp; /* mount structure */ |
1938 | xfs_perag_t *pag; /* per allocation group data */ | |
1939 | ||
1940 | agf = XFS_BUF_TO_AGF(agbp); | |
1941 | /* | |
1942 | * Freelist is empty, give up. | |
1943 | */ | |
1944 | if (!agf->agf_flcount) { | |
1945 | *bnop = NULLAGBLOCK; | |
1946 | return 0; | |
1947 | } | |
1948 | /* | |
1949 | * Read the array of free blocks. | |
1950 | */ | |
1951 | mp = tp->t_mountp; | |
1952 | if ((error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 1953 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 LT |
1954 | return error; |
1955 | agfl = XFS_BUF_TO_AGFL(agflbp); | |
1956 | /* | |
1957 | * Get the block number and update the data structures. | |
1958 | */ | |
e2101005 | 1959 | bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]); |
413d57c9 | 1960 | be32_add_cpu(&agf->agf_flfirst, 1); |
1da177e4 | 1961 | xfs_trans_brelse(tp, agflbp); |
16259e7d | 1962 | if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 1963 | agf->agf_flfirst = 0; |
a862e0fd DC |
1964 | |
1965 | pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); | |
413d57c9 | 1966 | be32_add_cpu(&agf->agf_flcount, -1); |
1da177e4 LT |
1967 | xfs_trans_agflist_delta(tp, -1); |
1968 | pag->pagf_flcount--; | |
a862e0fd | 1969 | xfs_perag_put(pag); |
92821e2b DC |
1970 | |
1971 | logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; | |
1972 | if (btreeblk) { | |
413d57c9 | 1973 | be32_add_cpu(&agf->agf_btreeblks, 1); |
92821e2b DC |
1974 | pag->pagf_btreeblks++; |
1975 | logflags |= XFS_AGF_BTREEBLKS; | |
1976 | } | |
1977 | ||
92821e2b | 1978 | xfs_alloc_log_agf(tp, agbp, logflags); |
1da177e4 LT |
1979 | *bnop = bno; |
1980 | ||
1da177e4 LT |
1981 | return 0; |
1982 | } | |
1983 | ||
1984 | /* | |
1985 | * Log the given fields from the agf structure. | |
1986 | */ | |
1987 | void | |
1988 | xfs_alloc_log_agf( | |
1989 | xfs_trans_t *tp, /* transaction pointer */ | |
1990 | xfs_buf_t *bp, /* buffer for a.g. freelist header */ | |
1991 | int fields) /* mask of fields to be logged (XFS_AGF_...) */ | |
1992 | { | |
1993 | int first; /* first byte offset */ | |
1994 | int last; /* last byte offset */ | |
1995 | static const short offsets[] = { | |
1996 | offsetof(xfs_agf_t, agf_magicnum), | |
1997 | offsetof(xfs_agf_t, agf_versionnum), | |
1998 | offsetof(xfs_agf_t, agf_seqno), | |
1999 | offsetof(xfs_agf_t, agf_length), | |
2000 | offsetof(xfs_agf_t, agf_roots[0]), | |
2001 | offsetof(xfs_agf_t, agf_levels[0]), | |
2002 | offsetof(xfs_agf_t, agf_flfirst), | |
2003 | offsetof(xfs_agf_t, agf_fllast), | |
2004 | offsetof(xfs_agf_t, agf_flcount), | |
2005 | offsetof(xfs_agf_t, agf_freeblks), | |
2006 | offsetof(xfs_agf_t, agf_longest), | |
92821e2b | 2007 | offsetof(xfs_agf_t, agf_btreeblks), |
1da177e4 LT |
2008 | sizeof(xfs_agf_t) |
2009 | }; | |
2010 | ||
0b1b213f CH |
2011 | trace_xfs_agf(tp->t_mountp, XFS_BUF_TO_AGF(bp), fields, _RET_IP_); |
2012 | ||
1da177e4 LT |
2013 | xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last); |
2014 | xfs_trans_log_buf(tp, bp, (uint)first, (uint)last); | |
2015 | } | |
2016 | ||
2017 | /* | |
2018 | * Interface for inode allocation to force the pag data to be initialized. | |
2019 | */ | |
2020 | int /* error */ | |
2021 | xfs_alloc_pagf_init( | |
2022 | xfs_mount_t *mp, /* file system mount structure */ | |
2023 | xfs_trans_t *tp, /* transaction pointer */ | |
2024 | xfs_agnumber_t agno, /* allocation group number */ | |
2025 | int flags) /* XFS_ALLOC_FLAGS_... */ | |
2026 | { | |
2027 | xfs_buf_t *bp; | |
2028 | int error; | |
2029 | ||
2030 | if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp))) | |
2031 | return error; | |
2032 | if (bp) | |
2033 | xfs_trans_brelse(tp, bp); | |
2034 | return 0; | |
2035 | } | |
2036 | ||
2037 | /* | |
2038 | * Put the block on the freelist for the allocation group. | |
2039 | */ | |
2040 | int /* error */ | |
2041 | xfs_alloc_put_freelist( | |
2042 | xfs_trans_t *tp, /* transaction pointer */ | |
2043 | xfs_buf_t *agbp, /* buffer for a.g. freelist header */ | |
2044 | xfs_buf_t *agflbp,/* buffer for a.g. free block array */ | |
92821e2b DC |
2045 | xfs_agblock_t bno, /* block being freed */ |
2046 | int btreeblk) /* block came from a AGF btree */ | |
1da177e4 LT |
2047 | { |
2048 | xfs_agf_t *agf; /* a.g. freespace structure */ | |
2049 | xfs_agfl_t *agfl; /* a.g. free block array */ | |
e2101005 | 2050 | __be32 *blockp;/* pointer to array entry */ |
1da177e4 | 2051 | int error; |
92821e2b | 2052 | int logflags; |
1da177e4 LT |
2053 | xfs_mount_t *mp; /* mount structure */ |
2054 | xfs_perag_t *pag; /* per allocation group data */ | |
2055 | ||
2056 | agf = XFS_BUF_TO_AGF(agbp); | |
2057 | mp = tp->t_mountp; | |
2058 | ||
2059 | if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp, | |
16259e7d | 2060 | be32_to_cpu(agf->agf_seqno), &agflbp))) |
1da177e4 LT |
2061 | return error; |
2062 | agfl = XFS_BUF_TO_AGFL(agflbp); | |
413d57c9 | 2063 | be32_add_cpu(&agf->agf_fllast, 1); |
16259e7d | 2064 | if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp)) |
1da177e4 | 2065 | agf->agf_fllast = 0; |
a862e0fd DC |
2066 | |
2067 | pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); | |
413d57c9 | 2068 | be32_add_cpu(&agf->agf_flcount, 1); |
1da177e4 LT |
2069 | xfs_trans_agflist_delta(tp, 1); |
2070 | pag->pagf_flcount++; | |
92821e2b DC |
2071 | |
2072 | logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT; | |
2073 | if (btreeblk) { | |
413d57c9 | 2074 | be32_add_cpu(&agf->agf_btreeblks, -1); |
92821e2b DC |
2075 | pag->pagf_btreeblks--; |
2076 | logflags |= XFS_AGF_BTREEBLKS; | |
2077 | } | |
a862e0fd | 2078 | xfs_perag_put(pag); |
92821e2b | 2079 | |
92821e2b DC |
2080 | xfs_alloc_log_agf(tp, agbp, logflags); |
2081 | ||
16259e7d CH |
2082 | ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp)); |
2083 | blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)]; | |
e2101005 | 2084 | *blockp = cpu_to_be32(bno); |
92821e2b | 2085 | xfs_alloc_log_agf(tp, agbp, logflags); |
1da177e4 LT |
2086 | xfs_trans_log_buf(tp, agflbp, |
2087 | (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl), | |
2088 | (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl + | |
2089 | sizeof(xfs_agblock_t) - 1)); | |
2090 | return 0; | |
2091 | } | |
2092 | ||
2093 | /* | |
2094 | * Read in the allocation group header (free/alloc section). | |
2095 | */ | |
2096 | int /* error */ | |
4805621a FCH |
2097 | xfs_read_agf( |
2098 | struct xfs_mount *mp, /* mount point structure */ | |
2099 | struct xfs_trans *tp, /* transaction pointer */ | |
2100 | xfs_agnumber_t agno, /* allocation group number */ | |
2101 | int flags, /* XFS_BUF_ */ | |
2102 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
1da177e4 | 2103 | { |
4805621a | 2104 | struct xfs_agf *agf; /* ag freelist header */ |
1da177e4 | 2105 | int agf_ok; /* set if agf is consistent */ |
1da177e4 LT |
2106 | int error; |
2107 | ||
2108 | ASSERT(agno != NULLAGNUMBER); | |
2109 | error = xfs_trans_read_buf( | |
2110 | mp, tp, mp->m_ddev_targp, | |
2111 | XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)), | |
4805621a | 2112 | XFS_FSS_TO_BB(mp, 1), flags, bpp); |
1da177e4 LT |
2113 | if (error) |
2114 | return error; | |
4805621a | 2115 | if (!*bpp) |
1da177e4 | 2116 | return 0; |
4805621a | 2117 | |
5a52c2a5 | 2118 | ASSERT(!(*bpp)->b_error); |
4805621a FCH |
2119 | agf = XFS_BUF_TO_AGF(*bpp); |
2120 | ||
1da177e4 LT |
2121 | /* |
2122 | * Validate the magic number of the agf block. | |
2123 | */ | |
1da177e4 | 2124 | agf_ok = |
69ef921b | 2125 | agf->agf_magicnum == cpu_to_be32(XFS_AGF_MAGIC) && |
16259e7d CH |
2126 | XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) && |
2127 | be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) && | |
2128 | be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) && | |
2129 | be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) && | |
4805621a FCH |
2130 | be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp) && |
2131 | be32_to_cpu(agf->agf_seqno) == agno; | |
89b28393 BN |
2132 | if (xfs_sb_version_haslazysbcount(&mp->m_sb)) |
2133 | agf_ok = agf_ok && be32_to_cpu(agf->agf_btreeblks) <= | |
2134 | be32_to_cpu(agf->agf_length); | |
1da177e4 LT |
2135 | if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF, |
2136 | XFS_RANDOM_ALLOC_READ_AGF))) { | |
2137 | XFS_CORRUPTION_ERROR("xfs_alloc_read_agf", | |
2138 | XFS_ERRLEVEL_LOW, mp, agf); | |
4805621a | 2139 | xfs_trans_brelse(tp, *bpp); |
1da177e4 LT |
2140 | return XFS_ERROR(EFSCORRUPTED); |
2141 | } | |
38f23232 | 2142 | xfs_buf_set_ref(*bpp, XFS_AGF_REF); |
4805621a FCH |
2143 | return 0; |
2144 | } | |
2145 | ||
2146 | /* | |
2147 | * Read in the allocation group header (free/alloc section). | |
2148 | */ | |
2149 | int /* error */ | |
2150 | xfs_alloc_read_agf( | |
2151 | struct xfs_mount *mp, /* mount point structure */ | |
2152 | struct xfs_trans *tp, /* transaction pointer */ | |
2153 | xfs_agnumber_t agno, /* allocation group number */ | |
2154 | int flags, /* XFS_ALLOC_FLAG_... */ | |
2155 | struct xfs_buf **bpp) /* buffer for the ag freelist header */ | |
2156 | { | |
2157 | struct xfs_agf *agf; /* ag freelist header */ | |
2158 | struct xfs_perag *pag; /* per allocation group data */ | |
2159 | int error; | |
2160 | ||
2161 | ASSERT(agno != NULLAGNUMBER); | |
2162 | ||
2163 | error = xfs_read_agf(mp, tp, agno, | |
0cadda1c | 2164 | (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0, |
4805621a FCH |
2165 | bpp); |
2166 | if (error) | |
2167 | return error; | |
2168 | if (!*bpp) | |
2169 | return 0; | |
5a52c2a5 | 2170 | ASSERT(!(*bpp)->b_error); |
4805621a FCH |
2171 | |
2172 | agf = XFS_BUF_TO_AGF(*bpp); | |
a862e0fd | 2173 | pag = xfs_perag_get(mp, agno); |
1da177e4 | 2174 | if (!pag->pagf_init) { |
16259e7d | 2175 | pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); |
92821e2b | 2176 | pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); |
16259e7d CH |
2177 | pag->pagf_flcount = be32_to_cpu(agf->agf_flcount); |
2178 | pag->pagf_longest = be32_to_cpu(agf->agf_longest); | |
1da177e4 | 2179 | pag->pagf_levels[XFS_BTNUM_BNOi] = |
16259e7d | 2180 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]); |
1da177e4 | 2181 | pag->pagf_levels[XFS_BTNUM_CNTi] = |
16259e7d | 2182 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); |
007c61c6 | 2183 | spin_lock_init(&pag->pagb_lock); |
e57336ff | 2184 | pag->pagb_count = 0; |
ed3b4d6c | 2185 | pag->pagb_tree = RB_ROOT; |
1da177e4 LT |
2186 | pag->pagf_init = 1; |
2187 | } | |
2188 | #ifdef DEBUG | |
2189 | else if (!XFS_FORCED_SHUTDOWN(mp)) { | |
16259e7d | 2190 | ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks)); |
89b28393 | 2191 | ASSERT(pag->pagf_btreeblks == be32_to_cpu(agf->agf_btreeblks)); |
16259e7d CH |
2192 | ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount)); |
2193 | ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest)); | |
1da177e4 | 2194 | ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] == |
16259e7d | 2195 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi])); |
1da177e4 | 2196 | ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] == |
16259e7d | 2197 | be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); |
1da177e4 LT |
2198 | } |
2199 | #endif | |
a862e0fd | 2200 | xfs_perag_put(pag); |
1da177e4 LT |
2201 | return 0; |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * Allocate an extent (variable-size). | |
2206 | * Depending on the allocation type, we either look in a single allocation | |
2207 | * group or loop over the allocation groups to find the result. | |
2208 | */ | |
2209 | int /* error */ | |
c999a223 | 2210 | __xfs_alloc_vextent( |
1da177e4 LT |
2211 | xfs_alloc_arg_t *args) /* allocation argument structure */ |
2212 | { | |
2213 | xfs_agblock_t agsize; /* allocation group size */ | |
2214 | int error; | |
2215 | int flags; /* XFS_ALLOC_FLAG_... locking flags */ | |
1da177e4 LT |
2216 | xfs_extlen_t minleft;/* minimum left value, temp copy */ |
2217 | xfs_mount_t *mp; /* mount structure pointer */ | |
2218 | xfs_agnumber_t sagno; /* starting allocation group number */ | |
2219 | xfs_alloctype_t type; /* input allocation type */ | |
2220 | int bump_rotor = 0; | |
2221 | int no_min = 0; | |
2222 | xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */ | |
2223 | ||
2224 | mp = args->mp; | |
2225 | type = args->otype = args->type; | |
2226 | args->agbno = NULLAGBLOCK; | |
2227 | /* | |
2228 | * Just fix this up, for the case where the last a.g. is shorter | |
2229 | * (or there's only one a.g.) and the caller couldn't easily figure | |
2230 | * that out (xfs_bmap_alloc). | |
2231 | */ | |
2232 | agsize = mp->m_sb.sb_agblocks; | |
2233 | if (args->maxlen > agsize) | |
2234 | args->maxlen = agsize; | |
2235 | if (args->alignment == 0) | |
2236 | args->alignment = 1; | |
2237 | ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount); | |
2238 | ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize); | |
2239 | ASSERT(args->minlen <= args->maxlen); | |
2240 | ASSERT(args->minlen <= agsize); | |
2241 | ASSERT(args->mod < args->prod); | |
2242 | if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount || | |
2243 | XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize || | |
2244 | args->minlen > args->maxlen || args->minlen > agsize || | |
2245 | args->mod >= args->prod) { | |
2246 | args->fsbno = NULLFSBLOCK; | |
0b1b213f | 2247 | trace_xfs_alloc_vextent_badargs(args); |
1da177e4 LT |
2248 | return 0; |
2249 | } | |
2250 | minleft = args->minleft; | |
2251 | ||
2252 | switch (type) { | |
2253 | case XFS_ALLOCTYPE_THIS_AG: | |
2254 | case XFS_ALLOCTYPE_NEAR_BNO: | |
2255 | case XFS_ALLOCTYPE_THIS_BNO: | |
2256 | /* | |
2257 | * These three force us into a single a.g. | |
2258 | */ | |
2259 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
a862e0fd | 2260 | args->pag = xfs_perag_get(mp, args->agno); |
1da177e4 LT |
2261 | args->minleft = 0; |
2262 | error = xfs_alloc_fix_freelist(args, 0); | |
2263 | args->minleft = minleft; | |
2264 | if (error) { | |
0b1b213f | 2265 | trace_xfs_alloc_vextent_nofix(args); |
1da177e4 LT |
2266 | goto error0; |
2267 | } | |
2268 | if (!args->agbp) { | |
0b1b213f | 2269 | trace_xfs_alloc_vextent_noagbp(args); |
1da177e4 LT |
2270 | break; |
2271 | } | |
2272 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2273 | if ((error = xfs_alloc_ag_vextent(args))) | |
2274 | goto error0; | |
1da177e4 LT |
2275 | break; |
2276 | case XFS_ALLOCTYPE_START_BNO: | |
2277 | /* | |
2278 | * Try near allocation first, then anywhere-in-ag after | |
2279 | * the first a.g. fails. | |
2280 | */ | |
2281 | if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) && | |
2282 | (mp->m_flags & XFS_MOUNT_32BITINODES)) { | |
2283 | args->fsbno = XFS_AGB_TO_FSB(mp, | |
2284 | ((mp->m_agfrotor / rotorstep) % | |
2285 | mp->m_sb.sb_agcount), 0); | |
2286 | bump_rotor = 1; | |
2287 | } | |
2288 | args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno); | |
2289 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2290 | /* FALLTHROUGH */ | |
2291 | case XFS_ALLOCTYPE_ANY_AG: | |
2292 | case XFS_ALLOCTYPE_START_AG: | |
2293 | case XFS_ALLOCTYPE_FIRST_AG: | |
2294 | /* | |
2295 | * Rotate through the allocation groups looking for a winner. | |
2296 | */ | |
2297 | if (type == XFS_ALLOCTYPE_ANY_AG) { | |
2298 | /* | |
2299 | * Start with the last place we left off. | |
2300 | */ | |
2301 | args->agno = sagno = (mp->m_agfrotor / rotorstep) % | |
2302 | mp->m_sb.sb_agcount; | |
2303 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2304 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2305 | } else if (type == XFS_ALLOCTYPE_FIRST_AG) { | |
2306 | /* | |
2307 | * Start with allocation group given by bno. | |
2308 | */ | |
2309 | args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2310 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2311 | sagno = 0; | |
2312 | flags = 0; | |
2313 | } else { | |
2314 | if (type == XFS_ALLOCTYPE_START_AG) | |
2315 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
2316 | /* | |
2317 | * Start with the given allocation group. | |
2318 | */ | |
2319 | args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno); | |
2320 | flags = XFS_ALLOC_FLAG_TRYLOCK; | |
2321 | } | |
2322 | /* | |
2323 | * Loop over allocation groups twice; first time with | |
2324 | * trylock set, second time without. | |
2325 | */ | |
1da177e4 | 2326 | for (;;) { |
a862e0fd | 2327 | args->pag = xfs_perag_get(mp, args->agno); |
1da177e4 LT |
2328 | if (no_min) args->minleft = 0; |
2329 | error = xfs_alloc_fix_freelist(args, flags); | |
2330 | args->minleft = minleft; | |
2331 | if (error) { | |
0b1b213f | 2332 | trace_xfs_alloc_vextent_nofix(args); |
1da177e4 LT |
2333 | goto error0; |
2334 | } | |
2335 | /* | |
2336 | * If we get a buffer back then the allocation will fly. | |
2337 | */ | |
2338 | if (args->agbp) { | |
2339 | if ((error = xfs_alloc_ag_vextent(args))) | |
2340 | goto error0; | |
2341 | break; | |
2342 | } | |
0b1b213f CH |
2343 | |
2344 | trace_xfs_alloc_vextent_loopfailed(args); | |
2345 | ||
1da177e4 LT |
2346 | /* |
2347 | * Didn't work, figure out the next iteration. | |
2348 | */ | |
2349 | if (args->agno == sagno && | |
2350 | type == XFS_ALLOCTYPE_START_BNO) | |
2351 | args->type = XFS_ALLOCTYPE_THIS_AG; | |
d210a28c YL |
2352 | /* |
2353 | * For the first allocation, we can try any AG to get | |
2354 | * space. However, if we already have allocated a | |
2355 | * block, we don't want to try AGs whose number is below | |
2356 | * sagno. Otherwise, we may end up with out-of-order | |
2357 | * locking of AGF, which might cause deadlock. | |
2358 | */ | |
2359 | if (++(args->agno) == mp->m_sb.sb_agcount) { | |
2360 | if (args->firstblock != NULLFSBLOCK) | |
2361 | args->agno = sagno; | |
2362 | else | |
2363 | args->agno = 0; | |
2364 | } | |
1da177e4 LT |
2365 | /* |
2366 | * Reached the starting a.g., must either be done | |
2367 | * or switch to non-trylock mode. | |
2368 | */ | |
2369 | if (args->agno == sagno) { | |
2370 | if (no_min == 1) { | |
2371 | args->agbno = NULLAGBLOCK; | |
0b1b213f | 2372 | trace_xfs_alloc_vextent_allfailed(args); |
1da177e4 LT |
2373 | break; |
2374 | } | |
2375 | if (flags == 0) { | |
2376 | no_min = 1; | |
2377 | } else { | |
2378 | flags = 0; | |
2379 | if (type == XFS_ALLOCTYPE_START_BNO) { | |
2380 | args->agbno = XFS_FSB_TO_AGBNO(mp, | |
2381 | args->fsbno); | |
2382 | args->type = XFS_ALLOCTYPE_NEAR_BNO; | |
2383 | } | |
2384 | } | |
2385 | } | |
a862e0fd | 2386 | xfs_perag_put(args->pag); |
1da177e4 | 2387 | } |
1da177e4 LT |
2388 | if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) { |
2389 | if (args->agno == sagno) | |
2390 | mp->m_agfrotor = (mp->m_agfrotor + 1) % | |
2391 | (mp->m_sb.sb_agcount * rotorstep); | |
2392 | else | |
2393 | mp->m_agfrotor = (args->agno * rotorstep + 1) % | |
2394 | (mp->m_sb.sb_agcount * rotorstep); | |
2395 | } | |
2396 | break; | |
2397 | default: | |
2398 | ASSERT(0); | |
2399 | /* NOTREACHED */ | |
2400 | } | |
2401 | if (args->agbno == NULLAGBLOCK) | |
2402 | args->fsbno = NULLFSBLOCK; | |
2403 | else { | |
2404 | args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno); | |
2405 | #ifdef DEBUG | |
2406 | ASSERT(args->len >= args->minlen); | |
2407 | ASSERT(args->len <= args->maxlen); | |
2408 | ASSERT(args->agbno % args->alignment == 0); | |
2409 | XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno), | |
2410 | args->len); | |
2411 | #endif | |
2412 | } | |
a862e0fd | 2413 | xfs_perag_put(args->pag); |
1da177e4 LT |
2414 | return 0; |
2415 | error0: | |
a862e0fd | 2416 | xfs_perag_put(args->pag); |
1da177e4 LT |
2417 | return error; |
2418 | } | |
2419 | ||
c999a223 DC |
2420 | static void |
2421 | xfs_alloc_vextent_worker( | |
2422 | struct work_struct *work) | |
2423 | { | |
2424 | struct xfs_alloc_arg *args = container_of(work, | |
2425 | struct xfs_alloc_arg, work); | |
2426 | unsigned long pflags; | |
2427 | ||
2428 | /* we are in a transaction context here */ | |
2429 | current_set_flags_nested(&pflags, PF_FSTRANS); | |
2430 | ||
2431 | args->result = __xfs_alloc_vextent(args); | |
2432 | complete(args->done); | |
2433 | ||
2434 | current_restore_flags_nested(&pflags, PF_FSTRANS); | |
2435 | } | |
2436 | ||
aa292847 DC |
2437 | /* |
2438 | * Data allocation requests often come in with little stack to work on. Push | |
2439 | * them off to a worker thread so there is lots of stack to use. Metadata | |
2440 | * requests, OTOH, are generally from low stack usage paths, so avoid the | |
2441 | * context switch overhead here. | |
2442 | */ | |
2443 | int | |
c999a223 | 2444 | xfs_alloc_vextent( |
aa292847 | 2445 | struct xfs_alloc_arg *args) |
c999a223 DC |
2446 | { |
2447 | DECLARE_COMPLETION_ONSTACK(done); | |
2448 | ||
aa292847 DC |
2449 | if (!args->userdata) |
2450 | return __xfs_alloc_vextent(args); | |
2451 | ||
2452 | ||
c999a223 | 2453 | args->done = &done; |
3b876c8f | 2454 | INIT_WORK_ONSTACK(&args->work, xfs_alloc_vextent_worker); |
c999a223 DC |
2455 | queue_work(xfs_alloc_wq, &args->work); |
2456 | wait_for_completion(&done); | |
2457 | return args->result; | |
2458 | } | |
2459 | ||
1da177e4 LT |
2460 | /* |
2461 | * Free an extent. | |
2462 | * Just break up the extent address and hand off to xfs_free_ag_extent | |
2463 | * after fixing up the freelist. | |
2464 | */ | |
2465 | int /* error */ | |
2466 | xfs_free_extent( | |
2467 | xfs_trans_t *tp, /* transaction pointer */ | |
2468 | xfs_fsblock_t bno, /* starting block number of extent */ | |
2469 | xfs_extlen_t len) /* length of extent */ | |
2470 | { | |
0e1edbd9 | 2471 | xfs_alloc_arg_t args; |
1da177e4 LT |
2472 | int error; |
2473 | ||
2474 | ASSERT(len != 0); | |
0e1edbd9 | 2475 | memset(&args, 0, sizeof(xfs_alloc_arg_t)); |
1da177e4 LT |
2476 | args.tp = tp; |
2477 | args.mp = tp->t_mountp; | |
be65b18a DC |
2478 | |
2479 | /* | |
2480 | * validate that the block number is legal - the enables us to detect | |
2481 | * and handle a silent filesystem corruption rather than crashing. | |
2482 | */ | |
1da177e4 | 2483 | args.agno = XFS_FSB_TO_AGNO(args.mp, bno); |
be65b18a DC |
2484 | if (args.agno >= args.mp->m_sb.sb_agcount) |
2485 | return EFSCORRUPTED; | |
2486 | ||
1da177e4 | 2487 | args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno); |
be65b18a DC |
2488 | if (args.agbno >= args.mp->m_sb.sb_agblocks) |
2489 | return EFSCORRUPTED; | |
2490 | ||
a862e0fd | 2491 | args.pag = xfs_perag_get(args.mp, args.agno); |
be65b18a DC |
2492 | ASSERT(args.pag); |
2493 | ||
2494 | error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING); | |
2495 | if (error) | |
1da177e4 | 2496 | goto error0; |
be65b18a DC |
2497 | |
2498 | /* validate the extent size is legal now we have the agf locked */ | |
2499 | if (args.agbno + len > | |
2500 | be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length)) { | |
2501 | error = EFSCORRUPTED; | |
2502 | goto error0; | |
2503 | } | |
2504 | ||
0e1edbd9 | 2505 | error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0); |
a870acd9 | 2506 | if (!error) |
4ecbfe63 | 2507 | xfs_extent_busy_insert(tp, args.agno, args.agbno, len, 0); |
1da177e4 | 2508 | error0: |
a862e0fd | 2509 | xfs_perag_put(args.pag); |
1da177e4 LT |
2510 | return error; |
2511 | } |