?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux...
[linux-2.6/kmemtrace.git] / fs / xfs / xfs_alloc.c
blob1956f83489f1b8378f4459844e422547d421e0ec
1 /*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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
7 * published by the Free Software Foundation.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_error.h"
41
42
43 #define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
44
45 #define XFSA_FIXUP_BNO_OK 1
46 #define XFSA_FIXUP_CNT_OK 2
47
48 STATIC void
49 xfs_alloc_search_busy(xfs_trans_t *tp,
50 xfs_agnumber_t agno,
51 xfs_agblock_t bno,
52 xfs_extlen_t len);
53
54 #if defined(XFS_ALLOC_TRACE)
55 ktrace_t *xfs_alloc_trace_buf;
56
57 #define TRACE_ALLOC(s,a) \
58 xfs_alloc_trace_alloc(__func__, s, a, __LINE__)
59 #define TRACE_FREE(s,a,b,x,f) \
60 xfs_alloc_trace_free(__func__, s, mp, a, b, x, f, __LINE__)
61 #define TRACE_MODAGF(s,a,f) \
62 xfs_alloc_trace_modagf(__func__, s, mp, a, f, __LINE__)
63 #define TRACE_BUSY(__func__,s,ag,agb,l,sl,tp) \
64 xfs_alloc_trace_busy(__func__, s, mp, ag, agb, l, sl, tp, XFS_ALLOC_KTRACE_BUSY, __LINE__)
65 #define TRACE_UNBUSY(__func__,s,ag,sl,tp) \
66 xfs_alloc_trace_busy(__func__, s, mp, ag, -1, -1, sl, tp, XFS_ALLOC_KTRACE_UNBUSY, __LINE__)
67 #define TRACE_BUSYSEARCH(__func__,s,ag,agb,l,tp) \
68 xfs_alloc_trace_busy(__func__, s, mp, ag, agb, l, 0, tp, XFS_ALLOC_KTRACE_BUSYSEARCH, __LINE__)
69 #else
70 #define TRACE_ALLOC(s,a)
71 #define TRACE_FREE(s,a,b,x,f)
72 #define TRACE_MODAGF(s,a,f)
73 #define TRACE_BUSY(s,a,ag,agb,l,sl,tp)
74 #define TRACE_UNBUSY(fname,s,ag,sl,tp)
75 #define TRACE_BUSYSEARCH(fname,s,ag,agb,l,tp)
76 #endif /* XFS_ALLOC_TRACE */
77
78 /*
79 * Prototypes for per-ag allocation routines
80 */
81
82 STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
83 STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
84 STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
85 STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *,
86 xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *);
87
88 /*
89 * Internal functions.
90 */
91
92 /*
93 * Compute aligned version of the found extent.
94 * Takes alignment and min length into account.
95 */
96 STATIC void
97 xfs_alloc_compute_aligned(
98 xfs_agblock_t foundbno, /* starting block in found extent */
99 xfs_extlen_t foundlen, /* length in found extent */
100 xfs_extlen_t alignment, /* alignment for allocation */
101 xfs_extlen_t minlen, /* minimum length for allocation */
102 xfs_agblock_t *resbno, /* result block number */
103 xfs_extlen_t *reslen) /* result length */
104 {
105 xfs_agblock_t bno;
106 xfs_extlen_t diff;
107 xfs_extlen_t len;
108
109 if (alignment > 1 && foundlen >= minlen) {
110 bno = roundup(foundbno, alignment);
111 diff = bno - foundbno;
112 len = diff >= foundlen ? 0 : foundlen - diff;
113 } else {
114 bno = foundbno;
115 len = foundlen;
116 }
117 *resbno = bno;
118 *reslen = len;
119 }
120
121 /*
122 * Compute best start block and diff for "near" allocations.
123 * freelen >= wantlen already checked by caller.
124 */
125 STATIC xfs_extlen_t /* difference value (absolute) */
126 xfs_alloc_compute_diff(
127 xfs_agblock_t wantbno, /* target starting block */
128 xfs_extlen_t wantlen, /* target length */
129 xfs_extlen_t alignment, /* target alignment */
130 xfs_agblock_t freebno, /* freespace's starting block */
131 xfs_extlen_t freelen, /* freespace's length */
132 xfs_agblock_t *newbnop) /* result: best start block from free */
133 {
134 xfs_agblock_t freeend; /* end of freespace extent */
135 xfs_agblock_t newbno1; /* return block number */
136 xfs_agblock_t newbno2; /* other new block number */
137 xfs_extlen_t newlen1=0; /* length with newbno1 */
138 xfs_extlen_t newlen2=0; /* length with newbno2 */
139 xfs_agblock_t wantend; /* end of target extent */
140
141 ASSERT(freelen >= wantlen);
142 freeend = freebno + freelen;
143 wantend = wantbno + wantlen;
144 if (freebno >= wantbno) {
145 if ((newbno1 = roundup(freebno, alignment)) >= freeend)
146 newbno1 = NULLAGBLOCK;
147 } else if (freeend >= wantend && alignment > 1) {
148 newbno1 = roundup(wantbno, alignment);
149 newbno2 = newbno1 - alignment;
150 if (newbno1 >= freeend)
151 newbno1 = NULLAGBLOCK;
152 else
153 newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
154 if (newbno2 < freebno)
155 newbno2 = NULLAGBLOCK;
156 else
157 newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
158 if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
159 if (newlen1 < newlen2 ||
160 (newlen1 == newlen2 &&
161 XFS_ABSDIFF(newbno1, wantbno) >
162 XFS_ABSDIFF(newbno2, wantbno)))
163 newbno1 = newbno2;
164 } else if (newbno2 != NULLAGBLOCK)
165 newbno1 = newbno2;
166 } else if (freeend >= wantend) {
167 newbno1 = wantbno;
168 } else if (alignment > 1) {
169 newbno1 = roundup(freeend - wantlen, alignment);
170 if (newbno1 > freeend - wantlen &&
171 newbno1 - alignment >= freebno)
172 newbno1 -= alignment;
173 else if (newbno1 >= freeend)
174 newbno1 = NULLAGBLOCK;
175 } else
176 newbno1 = freeend - wantlen;
177 *newbnop = newbno1;
178 return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
179 }
180
181 /*
182 * Fix up the length, based on mod and prod.
183 * len should be k * prod + mod for some k.
184 * If len is too small it is returned unchanged.
185 * If len hits maxlen it is left alone.
186 */
187 STATIC void
188 xfs_alloc_fix_len(
189 xfs_alloc_arg_t *args) /* allocation argument structure */
190 {
191 xfs_extlen_t k;
192 xfs_extlen_t rlen;
193
194 ASSERT(args->mod < args->prod);
195 rlen = args->len;
196 ASSERT(rlen >= args->minlen);
197 ASSERT(rlen <= args->maxlen);
198 if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
199 (args->mod == 0 && rlen < args->prod))
200 return;
201 k = rlen % args->prod;
202 if (k == args->mod)
203 return;
204 if (k > args->mod) {
205 if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen)
206 return;
207 } else {
208 if ((int)(rlen = rlen - args->prod - (args->mod - k)) <
209 (int)args->minlen)
210 return;
211 }
212 ASSERT(rlen >= args->minlen);
213 ASSERT(rlen <= args->maxlen);
214 args->len = rlen;
215 }
216
217 /*
218 * Fix up length if there is too little space left in the a.g.
219 * Return 1 if ok, 0 if too little, should give up.
220 */
221 STATIC int
222 xfs_alloc_fix_minleft(
223 xfs_alloc_arg_t *args) /* allocation argument structure */
224 {
225 xfs_agf_t *agf; /* a.g. freelist header */
226 int diff; /* free space difference */
227
228 if (args->minleft == 0)
229 return 1;
230 agf = XFS_BUF_TO_AGF(args->agbp);
231 diff = be32_to_cpu(agf->agf_freeblks)
232 + be32_to_cpu(agf->agf_flcount)
233 - args->len - args->minleft;
234 if (diff >= 0)
235 return 1;
236 args->len += diff; /* shrink the allocated space */
237 if (args->len >= args->minlen)
238 return 1;
239 args->agbno = NULLAGBLOCK;
240 return 0;
241 }
242
243 /*
244 * Update the two btrees, logically removing from freespace the extent
245 * starting at rbno, rlen blocks. The extent is contained within the
246 * actual (current) free extent fbno for flen blocks.
247 * Flags are passed in indicating whether the cursors are set to the
248 * relevant records.
249 */
250 STATIC int /* error code */
251 xfs_alloc_fixup_trees(
252 xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */
253 xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */
254 xfs_agblock_t fbno, /* starting block of free extent */
255 xfs_extlen_t flen, /* length of free extent */
256 xfs_agblock_t rbno, /* starting block of returned extent */
257 xfs_extlen_t rlen, /* length of returned extent */
258 int flags) /* flags, XFSA_FIXUP_... */
259 {
260 int error; /* error code */
261 int i; /* operation results */
262 xfs_agblock_t nfbno1; /* first new free startblock */
263 xfs_agblock_t nfbno2; /* second new free startblock */
264 xfs_extlen_t nflen1=0; /* first new free length */
265 xfs_extlen_t nflen2=0; /* second new free length */
266
267 /*
268 * Look up the record in the by-size tree if necessary.
269 */
270 if (flags & XFSA_FIXUP_CNT_OK) {
271 #ifdef DEBUG
272 if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
273 return error;
274 XFS_WANT_CORRUPTED_RETURN(
275 i == 1 && nfbno1 == fbno && nflen1 == flen);
276 #endif
277 } else {
278 if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
279 return error;
280 XFS_WANT_CORRUPTED_RETURN(i == 1);
281 }
282 /*
283 * Look up the record in the by-block tree if necessary.
284 */
285 if (flags & XFSA_FIXUP_BNO_OK) {
286 #ifdef DEBUG
287 if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
288 return error;
289 XFS_WANT_CORRUPTED_RETURN(
290 i == 1 && nfbno1 == fbno && nflen1 == flen);
291 #endif
292 } else {
293 if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
294 return error;
295 XFS_WANT_CORRUPTED_RETURN(i == 1);
296 }
297 #ifdef DEBUG
298 {
299 xfs_alloc_block_t *bnoblock;
300 xfs_alloc_block_t *cntblock;
301
302 if (bno_cur->bc_nlevels == 1 &&
303 cnt_cur->bc_nlevels == 1) {
304 bnoblock = XFS_BUF_TO_ALLOC_BLOCK(bno_cur->bc_bufs[0]);
305 cntblock = XFS_BUF_TO_ALLOC_BLOCK(cnt_cur->bc_bufs[0]);
306 XFS_WANT_CORRUPTED_RETURN(
307 be16_to_cpu(bnoblock->bb_numrecs) ==
308 be16_to_cpu(cntblock->bb_numrecs));
309 }
310 }
311 #endif
312 /*
313 * Deal with all four cases: the allocated record is contained
314 * within the freespace record, so we can have new freespace
315 * at either (or both) end, or no freespace remaining.
316 */
317 if (rbno == fbno && rlen == flen)
318 nfbno1 = nfbno2 = NULLAGBLOCK;
319 else if (rbno == fbno) {
320 nfbno1 = rbno + rlen;
321 nflen1 = flen - rlen;
322 nfbno2 = NULLAGBLOCK;
323 } else if (rbno + rlen == fbno + flen) {
324 nfbno1 = fbno;
325 nflen1 = flen - rlen;
326 nfbno2 = NULLAGBLOCK;
327 } else {
328 nfbno1 = fbno;
329 nflen1 = rbno - fbno;
330 nfbno2 = rbno + rlen;
331 nflen2 = (fbno + flen) - nfbno2;
332 }
333 /*
334 * Delete the entry from the by-size btree.
335 */
336 if ((error = xfs_alloc_delete(cnt_cur, &i)))
337 return error;
338 XFS_WANT_CORRUPTED_RETURN(i == 1);
339 /*
340 * Add new by-size btree entry(s).
341 */
342 if (nfbno1 != NULLAGBLOCK) {
343 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
344 return error;
345 XFS_WANT_CORRUPTED_RETURN(i == 0);
346 if ((error = xfs_alloc_insert(cnt_cur, &i)))
347 return error;
348 XFS_WANT_CORRUPTED_RETURN(i == 1);
349 }
350 if (nfbno2 != NULLAGBLOCK) {
351 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
352 return error;
353 XFS_WANT_CORRUPTED_RETURN(i == 0);
354 if ((error = xfs_alloc_insert(cnt_cur, &i)))
355 return error;
356 XFS_WANT_CORRUPTED_RETURN(i == 1);
357 }
358 /*
359 * Fix up the by-block btree entry(s).
360 */
361 if (nfbno1 == NULLAGBLOCK) {
362 /*
363 * No remaining freespace, just delete the by-block tree entry.
364 */
365 if ((error = xfs_alloc_delete(bno_cur, &i)))
366 return error;
367 XFS_WANT_CORRUPTED_RETURN(i == 1);
368 } else {
369 /*
370 * Update the by-block entry to start later|be shorter.
371 */
372 if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
373 return error;
374 }
375 if (nfbno2 != NULLAGBLOCK) {
376 /*
377 * 2 resulting free entries, need to add one.
378 */
379 if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
380 return error;
381 XFS_WANT_CORRUPTED_RETURN(i == 0);
382 if ((error = xfs_alloc_insert(bno_cur, &i)))
383 return error;
384 XFS_WANT_CORRUPTED_RETURN(i == 1);
385 }
386 return 0;
387 }
388
389 /*
390 * Read in the allocation group free block array.
391 */
392 STATIC int /* error */
393 xfs_alloc_read_agfl(
394 xfs_mount_t *mp, /* mount point structure */
395 xfs_trans_t *tp, /* transaction pointer */
396 xfs_agnumber_t agno, /* allocation group number */
397 xfs_buf_t **bpp) /* buffer for the ag free block array */
398 {
399 xfs_buf_t *bp; /* return value */
400 int error;
401
402 ASSERT(agno != NULLAGNUMBER);
403 error = xfs_trans_read_buf(
404 mp, tp, mp->m_ddev_targp,
405 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
406 XFS_FSS_TO_BB(mp, 1), 0, &bp);
407 if (error)
408 return error;
409 ASSERT(bp);
410 ASSERT(!XFS_BUF_GETERROR(bp));
411 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGFL, XFS_AGFL_REF);
412 *bpp = bp;
413 return 0;
414 }
415
416 #if defined(XFS_ALLOC_TRACE)
417 /*
418 * Add an allocation trace entry for an alloc call.
419 */
420 STATIC void
421 xfs_alloc_trace_alloc(
422 const char *name, /* function tag string */
423 char *str, /* additional string */
424 xfs_alloc_arg_t *args, /* allocation argument structure */
425 int line) /* source line number */
426 {
427 ktrace_enter(xfs_alloc_trace_buf,
428 (void *)(__psint_t)(XFS_ALLOC_KTRACE_ALLOC | (line << 16)),
429 (void *)name,
430 (void *)str,
431 (void *)args->mp,
432 (void *)(__psunsigned_t)args->agno,
433 (void *)(__psunsigned_t)args->agbno,
434 (void *)(__psunsigned_t)args->minlen,
435 (void *)(__psunsigned_t)args->maxlen,
436 (void *)(__psunsigned_t)args->mod,
437 (void *)(__psunsigned_t)args->prod,
438 (void *)(__psunsigned_t)args->minleft,
439 (void *)(__psunsigned_t)args->total,
440 (void *)(__psunsigned_t)args->alignment,
441 (void *)(__psunsigned_t)args->len,
442 (void *)((((__psint_t)args->type) << 16) |
443 (__psint_t)args->otype),
444 (void *)(__psint_t)((args->wasdel << 3) |
445 (args->wasfromfl << 2) |
446 (args->isfl << 1) |
447 (args->userdata << 0)));
448 }
449
450 /*
451 * Add an allocation trace entry for a free call.
452 */
453 STATIC void
454 xfs_alloc_trace_free(
455 const char *name, /* function tag string */
456 char *str, /* additional string */
457 xfs_mount_t *mp, /* file system mount point */
458 xfs_agnumber_t agno, /* allocation group number */
459 xfs_agblock_t agbno, /* a.g. relative block number */
460 xfs_extlen_t len, /* length of extent */
461 int isfl, /* set if is freelist allocation/free */
462 int line) /* source line number */
463 {
464 ktrace_enter(xfs_alloc_trace_buf,
465 (void *)(__psint_t)(XFS_ALLOC_KTRACE_FREE | (line << 16)),
466 (void *)name,
467 (void *)str,
468 (void *)mp,
469 (void *)(__psunsigned_t)agno,
470 (void *)(__psunsigned_t)agbno,
471 (void *)(__psunsigned_t)len,
472 (void *)(__psint_t)isfl,
473 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL);
474 }
475
476 /*
477 * Add an allocation trace entry for modifying an agf.
478 */
479 STATIC void
480 xfs_alloc_trace_modagf(
481 const char *name, /* function tag string */
482 char *str, /* additional string */
483 xfs_mount_t *mp, /* file system mount point */
484 xfs_agf_t *agf, /* new agf value */
485 int flags, /* logging flags for agf */
486 int line) /* source line number */
487 {
488 ktrace_enter(xfs_alloc_trace_buf,
489 (void *)(__psint_t)(XFS_ALLOC_KTRACE_MODAGF | (line << 16)),
490 (void *)name,
491 (void *)str,
492 (void *)mp,
493 (void *)(__psint_t)flags,
494 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_seqno),
495 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_length),
496 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_BNO]),
497 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_CNT]),
498 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]),
499 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]),
500 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flfirst),
501 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_fllast),
502 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flcount),
503 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_freeblks),
504 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_longest));
505 }
506
507 STATIC void
508 xfs_alloc_trace_busy(
509 const char *name, /* function tag string */
510 char *str, /* additional string */
511 xfs_mount_t *mp, /* file system mount point */
512 xfs_agnumber_t agno, /* allocation group number */
513 xfs_agblock_t agbno, /* a.g. relative block number */
514 xfs_extlen_t len, /* length of extent */
515 int slot, /* perag Busy slot */
516 xfs_trans_t *tp,
517 int trtype, /* type: add, delete, search */
518 int line) /* source line number */
519 {
520 ktrace_enter(xfs_alloc_trace_buf,
521 (void *)(__psint_t)(trtype | (line << 16)),
522 (void *)name,
523 (void *)str,
524 (void *)mp,
525 (void *)(__psunsigned_t)agno,
526 (void *)(__psunsigned_t)agbno,
527 (void *)(__psunsigned_t)len,
528 (void *)(__psint_t)slot,
529 (void *)tp,
530 NULL, NULL, NULL, NULL, NULL, NULL, NULL);
531 }
532 #endif /* XFS_ALLOC_TRACE */
533
534 /*
535 * Allocation group level functions.
536 */
537
538 /*
539 * Allocate a variable extent in the allocation group agno.
540 * Type and bno are used to determine where in the allocation group the
541 * extent will start.
542 * Extent's length (returned in *len) will be between minlen and maxlen,
543 * and of the form k * prod + mod unless there's nothing that large.
544 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
545 */
546 STATIC int /* error */
547 xfs_alloc_ag_vextent(
548 xfs_alloc_arg_t *args) /* argument structure for allocation */
549 {
550 int error=0;
551
552 ASSERT(args->minlen > 0);
553 ASSERT(args->maxlen > 0);
554 ASSERT(args->minlen <= args->maxlen);
555 ASSERT(args->mod < args->prod);
556 ASSERT(args->alignment > 0);
557 /*
558 * Branch to correct routine based on the type.
559 */
560 args->wasfromfl = 0;
561 switch (args->type) {
562 case XFS_ALLOCTYPE_THIS_AG:
563 error = xfs_alloc_ag_vextent_size(args);
564 break;
565 case XFS_ALLOCTYPE_NEAR_BNO:
566 error = xfs_alloc_ag_vextent_near(args);
567 break;
568 case XFS_ALLOCTYPE_THIS_BNO:
569 error = xfs_alloc_ag_vextent_exact(args);
570 break;
571 default:
572 ASSERT(0);
573 /* NOTREACHED */
574 }
575 if (error)
576 return error;
577 /*
578 * If the allocation worked, need to change the agf structure
579 * (and log it), and the superblock.
580 */
581 if (args->agbno != NULLAGBLOCK) {
582 xfs_agf_t *agf; /* allocation group freelist header */
583 #ifdef XFS_ALLOC_TRACE
584 xfs_mount_t *mp = args->mp;
585 #endif
586 long slen = (long)args->len;
587
588 ASSERT(args->len >= args->minlen && args->len <= args->maxlen);
589 ASSERT(!(args->wasfromfl) || !args->isfl);
590 ASSERT(args->agbno % args->alignment == 0);
591 if (!(args->wasfromfl)) {
592
593 agf = XFS_BUF_TO_AGF(args->agbp);
594 be32_add_cpu(&agf->agf_freeblks, -(args->len));
595 xfs_trans_agblocks_delta(args->tp,
596 -((long)(args->len)));
597 args->pag->pagf_freeblks -= args->len;
598 ASSERT(be32_to_cpu(agf->agf_freeblks) <=
599 be32_to_cpu(agf->agf_length));
600 TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS);
601 xfs_alloc_log_agf(args->tp, args->agbp,
602 XFS_AGF_FREEBLKS);
603 /* search the busylist for these blocks */
604 xfs_alloc_search_busy(args->tp, args->agno,
605 args->agbno, args->len);
606 }
607 if (!args->isfl)
608 xfs_trans_mod_sb(args->tp,
609 args->wasdel ? XFS_TRANS_SB_RES_FDBLOCKS :
610 XFS_TRANS_SB_FDBLOCKS, -slen);
611 XFS_STATS_INC(xs_allocx);
612 XFS_STATS_ADD(xs_allocb, args->len);
613 }
614 return 0;
615 }
616
617 /*
618 * Allocate a variable extent at exactly agno/bno.
619 * Extent's length (returned in *len) will be between minlen and maxlen,
620 * and of the form k * prod + mod unless there's nothing that large.
621 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
622 */
623 STATIC int /* error */
624 xfs_alloc_ag_vextent_exact(
625 xfs_alloc_arg_t *args) /* allocation argument structure */
626 {
627 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
628 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
629 xfs_agblock_t end; /* end of allocated extent */
630 int error;
631 xfs_agblock_t fbno; /* start block of found extent */
632 xfs_agblock_t fend; /* end block of found extent */
633 xfs_extlen_t flen; /* length of found extent */
634 int i; /* success/failure of operation */
635 xfs_agblock_t maxend; /* end of maximal extent */
636 xfs_agblock_t minend; /* end of minimal extent */
637 xfs_extlen_t rlen; /* length of returned extent */
638
639 ASSERT(args->alignment == 1);
640 /*
641 * Allocate/initialize a cursor for the by-number freespace btree.
642 */
643 bno_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
644 args->agno, XFS_BTNUM_BNO, NULL, 0);
645 /*
646 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
647 * Look for the closest free block <= bno, it must contain bno
648 * if any free block does.
649 */
650 if ((error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i)))
651 goto error0;
652 if (!i) {
653 /*
654 * Didn't find it, return null.
655 */
656 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
657 args->agbno = NULLAGBLOCK;
658 return 0;
659 }
660 /*
661 * Grab the freespace record.
662 */
663 if ((error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i)))
664 goto error0;
665 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
666 ASSERT(fbno <= args->agbno);
667 minend = args->agbno + args->minlen;
668 maxend = args->agbno + args->maxlen;
669 fend = fbno + flen;
670 /*
671 * Give up if the freespace isn't long enough for the minimum request.
672 */
673 if (fend < minend) {
674 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
675 args->agbno = NULLAGBLOCK;
676 return 0;
677 }
678 /*
679 * End of extent will be smaller of the freespace end and the
680 * maximal requested end.
681 */
682 end = XFS_AGBLOCK_MIN(fend, maxend);
683 /*
684 * Fix the length according to mod and prod if given.
685 */
686 args->len = end - args->agbno;
687 xfs_alloc_fix_len(args);
688 if (!xfs_alloc_fix_minleft(args)) {
689 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
690 return 0;
691 }
692 rlen = args->len;
693 ASSERT(args->agbno + rlen <= fend);
694 end = args->agbno + rlen;
695 /*
696 * We are allocating agbno for rlen [agbno .. end]
697 * Allocate/initialize a cursor for the by-size btree.
698 */
699 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
700 args->agno, XFS_BTNUM_CNT, NULL, 0);
701 ASSERT(args->agbno + args->len <=
702 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
703 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
704 args->agbno, args->len, XFSA_FIXUP_BNO_OK))) {
705 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
706 goto error0;
707 }
708 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
709 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
710 TRACE_ALLOC("normal", args);
711 args->wasfromfl = 0;
712 return 0;
713
714 error0:
715 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
716 TRACE_ALLOC("error", args);
717 return error;
718 }
719
720 /*
721 * Allocate a variable extent near bno in the allocation group agno.
722 * Extent's length (returned in len) will be between minlen and maxlen,
723 * and of the form k * prod + mod unless there's nothing that large.
724 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
725 */
726 STATIC int /* error */
727 xfs_alloc_ag_vextent_near(
728 xfs_alloc_arg_t *args) /* allocation argument structure */
729 {
730 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
731 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
732 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
733 xfs_agblock_t gtbno; /* start bno of right side entry */
734 xfs_agblock_t gtbnoa; /* aligned ... */
735 xfs_extlen_t gtdiff; /* difference to right side entry */
736 xfs_extlen_t gtlen; /* length of right side entry */
737 xfs_extlen_t gtlena; /* aligned ... */
738 xfs_agblock_t gtnew; /* useful start bno of right side */
739 int error; /* error code */
740 int i; /* result code, temporary */
741 int j; /* result code, temporary */
742 xfs_agblock_t ltbno; /* start bno of left side entry */
743 xfs_agblock_t ltbnoa; /* aligned ... */
744 xfs_extlen_t ltdiff; /* difference to left side entry */
745 /*REFERENCED*/
746 xfs_agblock_t ltend; /* end bno of left side entry */
747 xfs_extlen_t ltlen; /* length of left side entry */
748 xfs_extlen_t ltlena; /* aligned ... */
749 xfs_agblock_t ltnew; /* useful start bno of left side */
750 xfs_extlen_t rlen; /* length of returned extent */
751 #if defined(DEBUG) && defined(__KERNEL__)
752 /*
753 * Randomly don't execute the first algorithm.
754 */
755 int dofirst; /* set to do first algorithm */
756
757 dofirst = random32() & 1;
758 #endif
759 /*
760 * Get a cursor for the by-size btree.
761 */
762 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
763 args->agno, XFS_BTNUM_CNT, NULL, 0);
764 ltlen = 0;
765 bno_cur_lt = bno_cur_gt = NULL;
766 /*
767 * See if there are any free extents as big as maxlen.
768 */
769 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
770 goto error0;
771 /*
772 * If none, then pick up the last entry in the tree unless the
773 * tree is empty.
774 */
775 if (!i) {
776 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &ltbno,
777 &ltlen, &i)))
778 goto error0;
779 if (i == 0 || ltlen == 0) {
780 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
781 return 0;
782 }
783 ASSERT(i == 1);
784 }
785 args->wasfromfl = 0;
786 /*
787 * First algorithm.
788 * If the requested extent is large wrt the freespaces available
789 * in this a.g., then the cursor will be pointing to a btree entry
790 * near the right edge of the tree. If it's in the last btree leaf
791 * block, then we just examine all the entries in that block
792 * that are big enough, and pick the best one.
793 * This is written as a while loop so we can break out of it,
794 * but we never loop back to the top.
795 */
796 while (xfs_btree_islastblock(cnt_cur, 0)) {
797 xfs_extlen_t bdiff;
798 int besti=0;
799 xfs_extlen_t blen=0;
800 xfs_agblock_t bnew=0;
801
802 #if defined(DEBUG) && defined(__KERNEL__)
803 if (!dofirst)
804 break;
805 #endif
806 /*
807 * Start from the entry that lookup found, sequence through
808 * all larger free blocks. If we're actually pointing at a
809 * record smaller than maxlen, go to the start of this block,
810 * and skip all those smaller than minlen.
811 */
812 if (ltlen || args->alignment > 1) {
813 cnt_cur->bc_ptrs[0] = 1;
814 do {
815 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno,
816 &ltlen, &i)))
817 goto error0;
818 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
819 if (ltlen >= args->minlen)
820 break;
821 if ((error = xfs_alloc_increment(cnt_cur, 0, &i)))
822 goto error0;
823 } while (i);
824 ASSERT(ltlen >= args->minlen);
825 if (!i)
826 break;
827 }
828 i = cnt_cur->bc_ptrs[0];
829 for (j = 1, blen = 0, bdiff = 0;
830 !error && j && (blen < args->maxlen || bdiff > 0);
831 error = xfs_alloc_increment(cnt_cur, 0, &j)) {
832 /*
833 * For each entry, decide if it's better than
834 * the previous best entry.
835 */
836 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno, &ltlen, &i)))
837 goto error0;
838 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
839 xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment,
840 args->minlen, &ltbnoa, &ltlena);
841 if (ltlena < args->minlen)
842 continue;
843 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
844 xfs_alloc_fix_len(args);
845 ASSERT(args->len >= args->minlen);
846 if (args->len < blen)
847 continue;
848 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
849 args->alignment, ltbno, ltlen, &ltnew);
850 if (ltnew != NULLAGBLOCK &&
851 (args->len > blen || ltdiff < bdiff)) {
852 bdiff = ltdiff;
853 bnew = ltnew;
854 blen = args->len;
855 besti = cnt_cur->bc_ptrs[0];
856 }
857 }
858 /*
859 * It didn't work. We COULD be in a case where
860 * there's a good record somewhere, so try again.
861 */
862 if (blen == 0)
863 break;
864 /*
865 * Point at the best entry, and retrieve it again.
866 */
867 cnt_cur->bc_ptrs[0] = besti;
868 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno, &ltlen, &i)))
869 goto error0;
870 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
871 ltend = ltbno + ltlen;
872 ASSERT(ltend <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
873 args->len = blen;
874 if (!xfs_alloc_fix_minleft(args)) {
875 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
876 TRACE_ALLOC("nominleft", args);
877 return 0;
878 }
879 blen = args->len;
880 /*
881 * We are allocating starting at bnew for blen blocks.
882 */
883 args->agbno = bnew;
884 ASSERT(bnew >= ltbno);
885 ASSERT(bnew + blen <= ltend);
886 /*
887 * Set up a cursor for the by-bno tree.
888 */
889 bno_cur_lt = xfs_btree_init_cursor(args->mp, args->tp,
890 args->agbp, args->agno, XFS_BTNUM_BNO, NULL, 0);
891 /*
892 * Fix up the btree entries.
893 */
894 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
895 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
896 goto error0;
897 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
898 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
899 TRACE_ALLOC("first", args);
900 return 0;
901 }
902 /*
903 * Second algorithm.
904 * Search in the by-bno tree to the left and to the right
905 * simultaneously, until in each case we find a space big enough,
906 * or run into the edge of the tree. When we run into the edge,
907 * we deallocate that cursor.
908 * If both searches succeed, we compare the two spaces and pick
909 * the better one.
910 * With alignment, it's possible for both to fail; the upper
911 * level algorithm that picks allocation groups for allocations
912 * is not supposed to do this.
913 */
914 /*
915 * Allocate and initialize the cursor for the leftward search.
916 */
917 bno_cur_lt = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
918 args->agno, XFS_BTNUM_BNO, NULL, 0);
919 /*
920 * Lookup <= bno to find the leftward search's starting point.
921 */
922 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
923 goto error0;
924 if (!i) {
925 /*
926 * Didn't find anything; use this cursor for the rightward
927 * search.
928 */
929 bno_cur_gt = bno_cur_lt;
930 bno_cur_lt = NULL;
931 }
932 /*
933 * Found something. Duplicate the cursor for the rightward search.
934 */
935 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
936 goto error0;
937 /*
938 * Increment the cursor, so we will point at the entry just right
939 * of the leftward entry if any, or to the leftmost entry.
940 */
941 if ((error = xfs_alloc_increment(bno_cur_gt, 0, &i)))
942 goto error0;
943 if (!i) {
944 /*
945 * It failed, there are no rightward entries.
946 */
947 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
948 bno_cur_gt = NULL;
949 }
950 /*
951 * Loop going left with the leftward cursor, right with the
952 * rightward cursor, until either both directions give up or
953 * we find an entry at least as big as minlen.
954 */
955 do {
956 if (bno_cur_lt) {
957 if ((error = xfs_alloc_get_rec(bno_cur_lt, &ltbno, &ltlen, &i)))
958 goto error0;
959 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
960 xfs_alloc_compute_aligned(ltbno, ltlen, args->alignment,
961 args->minlen, &ltbnoa, &ltlena);
962 if (ltlena >= args->minlen)
963 break;
964 if ((error = xfs_alloc_decrement(bno_cur_lt, 0, &i)))
965 goto error0;
966 if (!i) {
967 xfs_btree_del_cursor(bno_cur_lt,
968 XFS_BTREE_NOERROR);
969 bno_cur_lt = NULL;
970 }
971 }
972 if (bno_cur_gt) {
973 if ((error = xfs_alloc_get_rec(bno_cur_gt, &gtbno, &gtlen, &i)))
974 goto error0;
975 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
976 xfs_alloc_compute_aligned(gtbno, gtlen, args->alignment,
977 args->minlen, &gtbnoa, &gtlena);
978 if (gtlena >= args->minlen)
979 break;
980 if ((error = xfs_alloc_increment(bno_cur_gt, 0, &i)))
981 goto error0;
982 if (!i) {
983 xfs_btree_del_cursor(bno_cur_gt,
984 XFS_BTREE_NOERROR);
985 bno_cur_gt = NULL;
986 }
987 }
988 } while (bno_cur_lt || bno_cur_gt);
989 /*
990 * Got both cursors still active, need to find better entry.
991 */
992 if (bno_cur_lt && bno_cur_gt) {
993 /*
994 * Left side is long enough, look for a right side entry.
995 */
996 if (ltlena >= args->minlen) {
997 /*
998 * Fix up the length.
999 */
1000 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1001 xfs_alloc_fix_len(args);
1002 rlen = args->len;
1003 ltdiff = xfs_alloc_compute_diff(args->agbno, rlen,
1004 args->alignment, ltbno, ltlen, &ltnew);
1005 /*
1006 * Not perfect.
1007 */
1008 if (ltdiff) {
1009 /*
1010 * Look until we find a better one, run out of
1011 * space, or run off the end.
1012 */
1013 while (bno_cur_lt && bno_cur_gt) {
1014 if ((error = xfs_alloc_get_rec(
1015 bno_cur_gt, &gtbno,
1016 &gtlen, &i)))
1017 goto error0;
1018 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1019 xfs_alloc_compute_aligned(gtbno, gtlen,
1020 args->alignment, args->minlen,
1021 &gtbnoa, &gtlena);
1022 /*
1023 * The left one is clearly better.
1024 */
1025 if (gtbnoa >= args->agbno + ltdiff) {
1026 xfs_btree_del_cursor(
1027 bno_cur_gt,
1028 XFS_BTREE_NOERROR);
1029 bno_cur_gt = NULL;
1030 break;
1031 }
1032 /*
1033 * If we reach a big enough entry,
1034 * compare the two and pick the best.
1035 */
1036 if (gtlena >= args->minlen) {
1037 args->len =
1038 XFS_EXTLEN_MIN(gtlena,
1039 args->maxlen);
1040 xfs_alloc_fix_len(args);
1041 rlen = args->len;
1042 gtdiff = xfs_alloc_compute_diff(
1043 args->agbno, rlen,
1044 args->alignment,
1045 gtbno, gtlen, &gtnew);
1046 /*
1047 * Right side is better.
1048 */
1049 if (gtdiff < ltdiff) {
1050 xfs_btree_del_cursor(
1051 bno_cur_lt,
1052 XFS_BTREE_NOERROR);
1053 bno_cur_lt = NULL;
1054 }
1055 /*
1056 * Left side is better.
1057 */
1058 else {
1059 xfs_btree_del_cursor(
1060 bno_cur_gt,
1061 XFS_BTREE_NOERROR);
1062 bno_cur_gt = NULL;
1063 }
1064 break;
1065 }
1066 /*
1067 * Fell off the right end.
1068 */
1069 if ((error = xfs_alloc_increment(
1070 bno_cur_gt, 0, &i)))
1071 goto error0;
1072 if (!i) {
1073 xfs_btree_del_cursor(
1074 bno_cur_gt,
1075 XFS_BTREE_NOERROR);
1076 bno_cur_gt = NULL;
1077 break;
1078 }
1079 }
1080 }
1081 /*
1082 * The left side is perfect, trash the right side.
1083 */
1084 else {
1085 xfs_btree_del_cursor(bno_cur_gt,
1086 XFS_BTREE_NOERROR);
1087 bno_cur_gt = NULL;
1088 }
1089 }
1090 /*
1091 * It's the right side that was found first, look left.
1092 */
1093 else {
1094 /*
1095 * Fix up the length.
1096 */
1097 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1098 xfs_alloc_fix_len(args);
1099 rlen = args->len;
1100 gtdiff = xfs_alloc_compute_diff(args->agbno, rlen,
1101 args->alignment, gtbno, gtlen, &gtnew);
1102 /*
1103 * Right side entry isn't perfect.
1104 */
1105 if (gtdiff) {
1106 /*
1107 * Look until we find a better one, run out of
1108 * space, or run off the end.
1109 */
1110 while (bno_cur_lt && bno_cur_gt) {
1111 if ((error = xfs_alloc_get_rec(
1112 bno_cur_lt, &ltbno,
1113 &ltlen, &i)))
1114 goto error0;
1115 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1116 xfs_alloc_compute_aligned(ltbno, ltlen,
1117 args->alignment, args->minlen,
1118 &ltbnoa, &ltlena);
1119 /*
1120 * The right one is clearly better.
1121 */
1122 if (ltbnoa <= args->agbno - gtdiff) {
1123 xfs_btree_del_cursor(
1124 bno_cur_lt,
1125 XFS_BTREE_NOERROR);
1126 bno_cur_lt = NULL;
1127 break;
1128 }
1129 /*
1130 * If we reach a big enough entry,
1131 * compare the two and pick the best.
1132 */
1133 if (ltlena >= args->minlen) {
1134 args->len = XFS_EXTLEN_MIN(
1135 ltlena, args->maxlen);
1136 xfs_alloc_fix_len(args);
1137 rlen = args->len;
1138 ltdiff = xfs_alloc_compute_diff(
1139 args->agbno, rlen,
1140 args->alignment,
1141 ltbno, ltlen, &ltnew);
1142 /*
1143 * Left side is better.
1144 */
1145 if (ltdiff < gtdiff) {
1146 xfs_btree_del_cursor(
1147 bno_cur_gt,
1148 XFS_BTREE_NOERROR);
1149 bno_cur_gt = NULL;
1150 }
1151 /*
1152 * Right side is better.
1153 */
1154 else {
1155 xfs_btree_del_cursor(
1156 bno_cur_lt,
1157 XFS_BTREE_NOERROR);
1158 bno_cur_lt = NULL;
1159 }
1160 break;
1161 }
1162 /*
1163 * Fell off the left end.
1164 */
1165 if ((error = xfs_alloc_decrement(
1166 bno_cur_lt, 0, &i)))
1167 goto error0;
1168 if (!i) {
1169 xfs_btree_del_cursor(bno_cur_lt,
1170 XFS_BTREE_NOERROR);
1171 bno_cur_lt = NULL;
1172 break;
1173 }
1174 }
1175 }
1176 /*
1177 * The right side is perfect, trash the left side.
1178 */
1179 else {
1180 xfs_btree_del_cursor(bno_cur_lt,
1181 XFS_BTREE_NOERROR);
1182 bno_cur_lt = NULL;
1183 }
1184 }
1185 }
1186 /*
1187 * If we couldn't get anything, give up.
1188 */
1189 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1190 TRACE_ALLOC("neither", args);
1191 args->agbno = NULLAGBLOCK;
1192 return 0;
1193 }
1194 /*
1195 * At this point we have selected a freespace entry, either to the
1196 * left or to the right. If it's on the right, copy all the
1197 * useful variables to the "left" set so we only have one
1198 * copy of this code.
1199 */
1200 if (bno_cur_gt) {
1201 bno_cur_lt = bno_cur_gt;
1202 bno_cur_gt = NULL;
1203 ltbno = gtbno;
1204 ltbnoa = gtbnoa;
1205 ltlen = gtlen;
1206 ltlena = gtlena;
1207 j = 1;
1208 } else
1209 j = 0;
1210 /*
1211 * Fix up the length and compute the useful address.
1212 */
1213 ltend = ltbno + ltlen;
1214 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1215 xfs_alloc_fix_len(args);
1216 if (!xfs_alloc_fix_minleft(args)) {
1217 TRACE_ALLOC("nominleft", args);
1218 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1219 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1220 return 0;
1221 }
1222 rlen = args->len;
1223 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, ltbno,
1224 ltlen, &ltnew);
1225 ASSERT(ltnew >= ltbno);
1226 ASSERT(ltnew + rlen <= ltend);
1227 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
1228 args->agbno = ltnew;
1229 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1230 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1231 goto error0;
1232 TRACE_ALLOC(j ? "gt" : "lt", args);
1233 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1234 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1235 return 0;
1236
1237 error0:
1238 TRACE_ALLOC("error", args);
1239 if (cnt_cur != NULL)
1240 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1241 if (bno_cur_lt != NULL)
1242 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1243 if (bno_cur_gt != NULL)
1244 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1245 return error;
1246 }
1247
1248 /*
1249 * Allocate a variable extent anywhere in the allocation group agno.
1250 * Extent's length (returned in len) will be between minlen and maxlen,
1251 * and of the form k * prod + mod unless there's nothing that large.
1252 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1253 */
1254 STATIC int /* error */
1255 xfs_alloc_ag_vextent_size(
1256 xfs_alloc_arg_t *args) /* allocation argument structure */
1257 {
1258 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1259 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1260 int error; /* error result */
1261 xfs_agblock_t fbno; /* start of found freespace */
1262 xfs_extlen_t flen; /* length of found freespace */
1263 int i; /* temp status variable */
1264 xfs_agblock_t rbno; /* returned block number */
1265 xfs_extlen_t rlen; /* length of returned extent */
1266
1267 /*
1268 * Allocate and initialize a cursor for the by-size btree.
1269 */
1270 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
1271 args->agno, XFS_BTNUM_CNT, NULL, 0);
1272 bno_cur = NULL;
1273 /*
1274 * Look for an entry >= maxlen+alignment-1 blocks.
1275 */
1276 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1277 args->maxlen + args->alignment - 1, &i)))
1278 goto error0;
1279 /*
1280 * If none, then pick up the last entry in the tree unless the
1281 * tree is empty.
1282 */
1283 if (!i) {
1284 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &fbno,
1285 &flen, &i)))
1286 goto error0;
1287 if (i == 0 || flen == 0) {
1288 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1289 TRACE_ALLOC("noentry", args);
1290 return 0;
1291 }
1292 ASSERT(i == 1);
1293 }
1294 /*
1295 * There's a freespace as big as maxlen+alignment-1, get it.
1296 */
1297 else {
1298 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i)))
1299 goto error0;
1300 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1301 }
1302 /*
1303 * In the first case above, we got the last entry in the
1304 * by-size btree. Now we check to see if the space hits maxlen
1305 * once aligned; if not, we search left for something better.
1306 * This can't happen in the second case above.
1307 */
1308 xfs_alloc_compute_aligned(fbno, flen, args->alignment, args->minlen,
1309 &rbno, &rlen);
1310 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1311 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1312 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1313 if (rlen < args->maxlen) {
1314 xfs_agblock_t bestfbno;
1315 xfs_extlen_t bestflen;
1316 xfs_agblock_t bestrbno;
1317 xfs_extlen_t bestrlen;
1318
1319 bestrlen = rlen;
1320 bestrbno = rbno;
1321 bestflen = flen;
1322 bestfbno = fbno;
1323 for (;;) {
1324 if ((error = xfs_alloc_decrement(cnt_cur, 0, &i)))
1325 goto error0;
1326 if (i == 0)
1327 break;
1328 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1329 &i)))
1330 goto error0;
1331 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1332 if (flen < bestrlen)
1333 break;
1334 xfs_alloc_compute_aligned(fbno, flen, args->alignment,
1335 args->minlen, &rbno, &rlen);
1336 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1337 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1338 (rlen <= flen && rbno + rlen <= fbno + flen),
1339 error0);
1340 if (rlen > bestrlen) {
1341 bestrlen = rlen;
1342 bestrbno = rbno;
1343 bestflen = flen;
1344 bestfbno = fbno;
1345 if (rlen == args->maxlen)
1346 break;
1347 }
1348 }
1349 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1350 &i)))
1351 goto error0;
1352 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1353 rlen = bestrlen;
1354 rbno = bestrbno;
1355 flen = bestflen;
1356 fbno = bestfbno;
1357 }
1358 args->wasfromfl = 0;
1359 /*
1360 * Fix up the length.
1361 */
1362 args->len = rlen;
1363 xfs_alloc_fix_len(args);
1364 if (rlen < args->minlen || !xfs_alloc_fix_minleft(args)) {
1365 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1366 TRACE_ALLOC("nominleft", args);
1367 args->agbno = NULLAGBLOCK;
1368 return 0;
1369 }
1370 rlen = args->len;
1371 XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1372 /*
1373 * Allocate and initialize a cursor for the by-block tree.
1374 */
1375 bno_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
1376 args->agno, XFS_BTNUM_BNO, NULL, 0);
1377 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1378 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1379 goto error0;
1380 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1381 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1382 cnt_cur = bno_cur = NULL;
1383 args->len = rlen;
1384 args->agbno = rbno;
1385 XFS_WANT_CORRUPTED_GOTO(
1386 args->agbno + args->len <=
1387 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1388 error0);
1389 TRACE_ALLOC("normal", args);
1390 return 0;
1391
1392 error0:
1393 TRACE_ALLOC("error", args);
1394 if (cnt_cur)
1395 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1396 if (bno_cur)
1397 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1398 return error;
1399 }
1400
1401 /*
1402 * Deal with the case where only small freespaces remain.
1403 * Either return the contents of the last freespace record,
1404 * or allocate space from the freelist if there is nothing in the tree.
1405 */
1406 STATIC int /* error */
1407 xfs_alloc_ag_vextent_small(
1408 xfs_alloc_arg_t *args, /* allocation argument structure */
1409 xfs_btree_cur_t *ccur, /* by-size cursor */
1410 xfs_agblock_t *fbnop, /* result block number */
1411 xfs_extlen_t *flenp, /* result length */
1412 int *stat) /* status: 0-freelist, 1-normal/none */
1413 {
1414 int error;
1415 xfs_agblock_t fbno;
1416 xfs_extlen_t flen;
1417 int i;
1418
1419 if ((error = xfs_alloc_decrement(ccur, 0, &i)))
1420 goto error0;
1421 if (i) {
1422 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1423 goto error0;
1424 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1425 }
1426 /*
1427 * Nothing in the btree, try the freelist. Make sure
1428 * to respect minleft even when pulling from the
1429 * freelist.
1430 */
1431 else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
1432 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1433 > args->minleft)) {
1434 error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno, 0);
1435 if (error)
1436 goto error0;
1437 if (fbno != NULLAGBLOCK) {
1438 if (args->userdata) {
1439 xfs_buf_t *bp;
1440
1441 bp = xfs_btree_get_bufs(args->mp, args->tp,
1442 args->agno, fbno, 0);
1443 xfs_trans_binval(args->tp, bp);
1444 }
1445 args->len = 1;
1446 args->agbno = fbno;
1447 XFS_WANT_CORRUPTED_GOTO(
1448 args->agbno + args->len <=
1449 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
1450 error0);
1451 args->wasfromfl = 1;
1452 TRACE_ALLOC("freelist", args);
1453 *stat = 0;
1454 return 0;
1455 }
1456 /*
1457 * Nothing in the freelist.
1458 */
1459 else
1460 flen = 0;
1461 }
1462 /*
1463 * Can't allocate from the freelist for some reason.
1464 */
1465 else {
1466 fbno = NULLAGBLOCK;
1467 flen = 0;
1468 }
1469 /*
1470 * Can't do the allocation, give up.
1471 */
1472 if (flen < args->minlen) {
1473 args->agbno = NULLAGBLOCK;
1474 TRACE_ALLOC("notenough", args);
1475 flen = 0;
1476 }
1477 *fbnop = fbno;
1478 *flenp = flen;
1479 *stat = 1;
1480 TRACE_ALLOC("normal", args);
1481 return 0;
1482
1483 error0:
1484 TRACE_ALLOC("error", args);
1485 return error;
1486 }
1487
1488 /*
1489 * Free the extent starting at agno/bno for length.
1490 */
1491 STATIC int /* error */
1492 xfs_free_ag_extent(
1493 xfs_trans_t *tp, /* transaction pointer */
1494 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1495 xfs_agnumber_t agno, /* allocation group number */
1496 xfs_agblock_t bno, /* starting block number */
1497 xfs_extlen_t len, /* length of extent */
1498 int isfl) /* set if is freelist blocks - no sb acctg */
1499 {
1500 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1501 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1502 int error; /* error return value */
1503 xfs_agblock_t gtbno; /* start of right neighbor block */
1504 xfs_extlen_t gtlen; /* length of right neighbor block */
1505 int haveleft; /* have a left neighbor block */
1506 int haveright; /* have a right neighbor block */
1507 int i; /* temp, result code */
1508 xfs_agblock_t ltbno; /* start of left neighbor block */
1509 xfs_extlen_t ltlen; /* length of left neighbor block */
1510 xfs_mount_t *mp; /* mount point struct for filesystem */
1511 xfs_agblock_t nbno; /* new starting block of freespace */
1512 xfs_extlen_t nlen; /* new length of freespace */
1513
1514 mp = tp->t_mountp;
1515 /*
1516 * Allocate and initialize a cursor for the by-block btree.
1517 */
1518 bno_cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO, NULL,
1519 0);
1520 cnt_cur = NULL;
1521 /*
1522 * Look for a neighboring block on the left (lower block numbers)
1523 * that is contiguous with this space.
1524 */
1525 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1526 goto error0;
1527 if (haveleft) {
1528 /*
1529 * There is a block to our left.
1530 */
1531 if ((error = xfs_alloc_get_rec(bno_cur, &ltbno, &ltlen, &i)))
1532 goto error0;
1533 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1534 /*
1535 * It's not contiguous, though.
1536 */
1537 if (ltbno + ltlen < bno)
1538 haveleft = 0;
1539 else {
1540 /*
1541 * If this failure happens the request to free this
1542 * space was invalid, it's (partly) already free.
1543 * Very bad.
1544 */
1545 XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1546 }
1547 }
1548 /*
1549 * Look for a neighboring block on the right (higher block numbers)
1550 * that is contiguous with this space.
1551 */
1552 if ((error = xfs_alloc_increment(bno_cur, 0, &haveright)))
1553 goto error0;
1554 if (haveright) {
1555 /*
1556 * There is a block to our right.
1557 */
1558 if ((error = xfs_alloc_get_rec(bno_cur, &gtbno, &gtlen, &i)))
1559 goto error0;
1560 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1561 /*
1562 * It's not contiguous, though.
1563 */
1564 if (bno + len < gtbno)
1565 haveright = 0;
1566 else {
1567 /*
1568 * If this failure happens the request to free this
1569 * space was invalid, it's (partly) already free.
1570 * Very bad.
1571 */
1572 XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1573 }
1574 }
1575 /*
1576 * Now allocate and initialize a cursor for the by-size tree.
1577 */
1578 cnt_cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT, NULL,
1579 0);
1580 /*
1581 * Have both left and right contiguous neighbors.
1582 * Merge all three into a single free block.
1583 */
1584 if (haveleft && haveright) {
1585 /*
1586 * Delete the old by-size entry on the left.
1587 */
1588 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1589 goto error0;
1590 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1591 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1592 goto error0;
1593 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1594 /*
1595 * Delete the old by-size entry on the right.
1596 */
1597 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1598 goto error0;
1599 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1600 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1601 goto error0;
1602 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1603 /*
1604 * Delete the old by-block entry for the right block.
1605 */
1606 if ((error = xfs_alloc_delete(bno_cur, &i)))
1607 goto error0;
1608 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1609 /*
1610 * Move the by-block cursor back to the left neighbor.
1611 */
1612 if ((error = xfs_alloc_decrement(bno_cur, 0, &i)))
1613 goto error0;
1614 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1615 #ifdef DEBUG
1616 /*
1617 * Check that this is the right record: delete didn't
1618 * mangle the cursor.
1619 */
1620 {
1621 xfs_agblock_t xxbno;
1622 xfs_extlen_t xxlen;
1623
1624 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1625 &i)))
1626 goto error0;
1627 XFS_WANT_CORRUPTED_GOTO(
1628 i == 1 && xxbno == ltbno && xxlen == ltlen,
1629 error0);
1630 }
1631 #endif
1632 /*
1633 * Update remaining by-block entry to the new, joined block.
1634 */
1635 nbno = ltbno;
1636 nlen = len + ltlen + gtlen;
1637 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1638 goto error0;
1639 }
1640 /*
1641 * Have only a left contiguous neighbor.
1642 * Merge it together with the new freespace.
1643 */
1644 else if (haveleft) {
1645 /*
1646 * Delete the old by-size entry on the left.
1647 */
1648 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1649 goto error0;
1650 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1651 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1652 goto error0;
1653 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1654 /*
1655 * Back up the by-block cursor to the left neighbor, and
1656 * update its length.
1657 */
1658 if ((error = xfs_alloc_decrement(bno_cur, 0, &i)))
1659 goto error0;
1660 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1661 nbno = ltbno;
1662 nlen = len + ltlen;
1663 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1664 goto error0;
1665 }
1666 /*
1667 * Have only a right contiguous neighbor.
1668 * Merge it together with the new freespace.
1669 */
1670 else if (haveright) {
1671 /*
1672 * Delete the old by-size entry on the right.
1673 */
1674 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1675 goto error0;
1676 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1677 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1678 goto error0;
1679 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1680 /*
1681 * Update the starting block and length of the right
1682 * neighbor in the by-block tree.
1683 */
1684 nbno = bno;
1685 nlen = len + gtlen;
1686 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1687 goto error0;
1688 }
1689 /*
1690 * No contiguous neighbors.
1691 * Insert the new freespace into the by-block tree.
1692 */
1693 else {
1694 nbno = bno;
1695 nlen = len;
1696 if ((error = xfs_alloc_insert(bno_cur, &i)))
1697 goto error0;
1698 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1699 }
1700 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1701 bno_cur = NULL;
1702 /*
1703 * In all cases we need to insert the new freespace in the by-size tree.
1704 */
1705 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1706 goto error0;
1707 XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1708 if ((error = xfs_alloc_insert(cnt_cur, &i)))
1709 goto error0;
1710 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1711 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1712 cnt_cur = NULL;
1713 /*
1714 * Update the freespace totals in the ag and superblock.
1715 */
1716 {
1717 xfs_agf_t *agf;
1718 xfs_perag_t *pag; /* per allocation group data */
1719
1720 agf = XFS_BUF_TO_AGF(agbp);
1721 pag = &mp->m_perag[agno];
1722 be32_add_cpu(&agf->agf_freeblks, len);
1723 xfs_trans_agblocks_delta(tp, len);
1724 pag->pagf_freeblks += len;
1725 XFS_WANT_CORRUPTED_GOTO(
1726 be32_to_cpu(agf->agf_freeblks) <=
1727 be32_to_cpu(agf->agf_length),
1728 error0);
1729 TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS);
1730 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
1731 if (!isfl)
1732 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1733 XFS_STATS_INC(xs_freex);
1734 XFS_STATS_ADD(xs_freeb, len);
1735 }
1736 TRACE_FREE(haveleft ?
1737 (haveright ? "both" : "left") :
1738 (haveright ? "right" : "none"),
1739 agno, bno, len, isfl);
1740
1741 /*
1742 * Since blocks move to the free list without the coordination
1743 * used in xfs_bmap_finish, we can't allow block to be available
1744 * for reallocation and non-transaction writing (user data)
1745 * until we know that the transaction that moved it to the free
1746 * list is permanently on disk. We track the blocks by declaring
1747 * these blocks as "busy"; the busy list is maintained on a per-ag
1748 * basis and each transaction records which entries should be removed
1749 * when the iclog commits to disk. If a busy block is allocated,
1750 * the iclog is pushed up to the LSN that freed the block.
1751 */
1752 xfs_alloc_mark_busy(tp, agno, bno, len);
1753 return 0;
1754
1755 error0:
1756 TRACE_FREE("error", agno, bno, len, isfl);
1757 if (bno_cur)
1758 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1759 if (cnt_cur)
1760 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1761 return error;
1762 }
1763
1764 /*
1765 * Visible (exported) allocation/free functions.
1766 * Some of these are used just by xfs_alloc_btree.c and this file.
1767 */
1768
1769 /*
1770 * Compute and fill in value of m_ag_maxlevels.
1771 */
1772 void
1773 xfs_alloc_compute_maxlevels(
1774 xfs_mount_t *mp) /* file system mount structure */
1775 {
1776 int level;
1777 uint maxblocks;
1778 uint maxleafents;
1779 int minleafrecs;
1780 int minnoderecs;
1781
1782 maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1783 minleafrecs = mp->m_alloc_mnr[0];
1784 minnoderecs = mp->m_alloc_mnr[1];
1785 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1786 for (level = 1; maxblocks > 1; level++)
1787 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1788 mp->m_ag_maxlevels = level;
1789 }
1790
1791 /*
1792 * Decide whether to use this allocation group for this allocation.
1793 * If so, fix up the btree freelist's size.
1794 */
1795 STATIC int /* error */
1796 xfs_alloc_fix_freelist(
1797 xfs_alloc_arg_t *args, /* allocation argument structure */
1798 int flags) /* XFS_ALLOC_FLAG_... */
1799 {
1800 xfs_buf_t *agbp; /* agf buffer pointer */
1801 xfs_agf_t *agf; /* a.g. freespace structure pointer */
1802 xfs_buf_t *agflbp;/* agfl buffer pointer */
1803 xfs_agblock_t bno; /* freelist block */
1804 xfs_extlen_t delta; /* new blocks needed in freelist */
1805 int error; /* error result code */
1806 xfs_extlen_t longest;/* longest extent in allocation group */
1807 xfs_mount_t *mp; /* file system mount point structure */
1808 xfs_extlen_t need; /* total blocks needed in freelist */
1809 xfs_perag_t *pag; /* per-ag information structure */
1810 xfs_alloc_arg_t targs; /* local allocation arguments */
1811 xfs_trans_t *tp; /* transaction pointer */
1812
1813 mp = args->mp;
1814
1815 pag = args->pag;
1816 tp = args->tp;
1817 if (!pag->pagf_init) {
1818 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1819 &agbp)))
1820 return error;
1821 if (!pag->pagf_init) {
1822 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1823 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1824 args->agbp = NULL;
1825 return 0;
1826 }
1827 } else
1828 agbp = NULL;
1829
1830 /*
1831 * If this is a metadata preferred pag and we are user data
1832 * then try somewhere else if we are not being asked to
1833 * try harder at this point
1834 */
1835 if (pag->pagf_metadata && args->userdata &&
1836 (flags & XFS_ALLOC_FLAG_TRYLOCK)) {
1837 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1838 args->agbp = NULL;
1839 return 0;
1840 }
1841
1842 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1843 need = XFS_MIN_FREELIST_PAG(pag, mp);
1844 delta = need > pag->pagf_flcount ? need - pag->pagf_flcount : 0;
1845 /*
1846 * If it looks like there isn't a long enough extent, or enough
1847 * total blocks, reject it.
1848 */
1849 longest = (pag->pagf_longest > delta) ?
1850 (pag->pagf_longest - delta) :
1851 (pag->pagf_flcount > 0 || pag->pagf_longest > 0);
1852 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1853 longest ||
1854 ((int)(pag->pagf_freeblks + pag->pagf_flcount -
1855 need - args->total) < (int)args->minleft)) {
1856 if (agbp)
1857 xfs_trans_brelse(tp, agbp);
1858 args->agbp = NULL;
1859 return 0;
1860 }
1861 }
1862
1863 /*
1864 * Get the a.g. freespace buffer.
1865 * Can fail if we're not blocking on locks, and it's held.
1866 */
1867 if (agbp == NULL) {
1868 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1869 &agbp)))
1870 return error;
1871 if (agbp == NULL) {
1872 ASSERT(flags & XFS_ALLOC_FLAG_TRYLOCK);
1873 ASSERT(!(flags & XFS_ALLOC_FLAG_FREEING));
1874 args->agbp = NULL;
1875 return 0;
1876 }
1877 }
1878 /*
1879 * Figure out how many blocks we should have in the freelist.
1880 */
1881 agf = XFS_BUF_TO_AGF(agbp);
1882 need = XFS_MIN_FREELIST(agf, mp);
1883 /*
1884 * If there isn't enough total or single-extent, reject it.
1885 */
1886 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1887 delta = need > be32_to_cpu(agf->agf_flcount) ?
1888 (need - be32_to_cpu(agf->agf_flcount)) : 0;
1889 longest = be32_to_cpu(agf->agf_longest);
1890 longest = (longest > delta) ? (longest - delta) :
1891 (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
1892 if ((args->minlen + args->alignment + args->minalignslop - 1) >
1893 longest ||
1894 ((int)(be32_to_cpu(agf->agf_freeblks) +
1895 be32_to_cpu(agf->agf_flcount) - need - args->total) <
1896 (int)args->minleft)) {
1897 xfs_trans_brelse(tp, agbp);
1898 args->agbp = NULL;
1899 return 0;
1900 }
1901 }
1902 /*
1903 * Make the freelist shorter if it's too long.
1904 */
1905 while (be32_to_cpu(agf->agf_flcount) > need) {
1906 xfs_buf_t *bp;
1907
1908 error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
1909 if (error)
1910 return error;
1911 if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1912 return error;
1913 bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1914 xfs_trans_binval(tp, bp);
1915 }
1916 /*
1917 * Initialize the args structure.
1918 */
1919 targs.tp = tp;
1920 targs.mp = mp;
1921 targs.agbp = agbp;
1922 targs.agno = args->agno;
1923 targs.mod = targs.minleft = targs.wasdel = targs.userdata =
1924 targs.minalignslop = 0;
1925 targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1926 targs.type = XFS_ALLOCTYPE_THIS_AG;
1927 targs.pag = pag;
1928 if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1929 return error;
1930 /*
1931 * Make the freelist longer if it's too short.
1932 */
1933 while (be32_to_cpu(agf->agf_flcount) < need) {
1934 targs.agbno = 0;
1935 targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
1936 /*
1937 * Allocate as many blocks as possible at once.
1938 */
1939 if ((error = xfs_alloc_ag_vextent(&targs))) {
1940 xfs_trans_brelse(tp, agflbp);
1941 return error;
1942 }
1943 /*
1944 * Stop if we run out. Won't happen if callers are obeying
1945 * the restrictions correctly. Can happen for free calls
1946 * on a completely full ag.
1947 */
1948 if (targs.agbno == NULLAGBLOCK) {
1949 if (flags & XFS_ALLOC_FLAG_FREEING)
1950 break;
1951 xfs_trans_brelse(tp, agflbp);
1952 args->agbp = NULL;
1953 return 0;
1954 }
1955 /*
1956 * Put each allocated block on the list.
1957 */
1958 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
1959 error = xfs_alloc_put_freelist(tp, agbp,
1960 agflbp, bno, 0);
1961 if (error)
1962 return error;
1963 }
1964 }
1965 xfs_trans_brelse(tp, agflbp);
1966 args->agbp = agbp;
1967 return 0;
1968 }
1969
1970 /*
1971 * Get a block from the freelist.
1972 * Returns with the buffer for the block gotten.
1973 */
1974 int /* error */
1975 xfs_alloc_get_freelist(
1976 xfs_trans_t *tp, /* transaction pointer */
1977 xfs_buf_t *agbp, /* buffer containing the agf structure */
1978 xfs_agblock_t *bnop, /* block address retrieved from freelist */
1979 int btreeblk) /* destination is a AGF btree */
1980 {
1981 xfs_agf_t *agf; /* a.g. freespace structure */
1982 xfs_agfl_t *agfl; /* a.g. freelist structure */
1983 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
1984 xfs_agblock_t bno; /* block number returned */
1985 int error;
1986 int logflags;
1987 xfs_mount_t *mp; /* mount structure */
1988 xfs_perag_t *pag; /* per allocation group data */
1989
1990 agf = XFS_BUF_TO_AGF(agbp);
1991 /*
1992 * Freelist is empty, give up.
1993 */
1994 if (!agf->agf_flcount) {
1995 *bnop = NULLAGBLOCK;
1996 return 0;
1997 }
1998 /*
1999 * Read the array of free blocks.
2000 */
2001 mp = tp->t_mountp;
2002 if ((error = xfs_alloc_read_agfl(mp, tp,
2003 be32_to_cpu(agf->agf_seqno), &agflbp)))
2004 return error;
2005 agfl = XFS_BUF_TO_AGFL(agflbp);
2006 /*
2007 * Get the block number and update the data structures.
2008 */
2009 bno = be32_to_cpu(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
2010 be32_add_cpu(&agf->agf_flfirst, 1);
2011 xfs_trans_brelse(tp, agflbp);
2012 if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
2013 agf->agf_flfirst = 0;
2014 pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)];
2015 be32_add_cpu(&agf->agf_flcount, -1);
2016 xfs_trans_agflist_delta(tp, -1);
2017 pag->pagf_flcount--;
2018
2019 logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT;
2020 if (btreeblk) {
2021 be32_add_cpu(&agf->agf_btreeblks, 1);
2022 pag->pagf_btreeblks++;
2023 logflags |= XFS_AGF_BTREEBLKS;
2024 }
2025
2026 TRACE_MODAGF(NULL, agf, logflags);
2027 xfs_alloc_log_agf(tp, agbp, logflags);
2028 *bnop = bno;
2029
2030 /*
2031 * As blocks are freed, they are added to the per-ag busy list
2032 * and remain there until the freeing transaction is committed to
2033 * disk. Now that we have allocated blocks, this list must be
2034 * searched to see if a block is being reused. If one is, then
2035 * the freeing transaction must be pushed to disk NOW by forcing
2036 * to disk all iclogs up that transaction's LSN.
2037 */
2038 xfs_alloc_search_busy(tp, be32_to_cpu(agf->agf_seqno), bno, 1);
2039 return 0;
2040 }
2041
2042 /*
2043 * Log the given fields from the agf structure.
2044 */
2045 void
2046 xfs_alloc_log_agf(
2047 xfs_trans_t *tp, /* transaction pointer */
2048 xfs_buf_t *bp, /* buffer for a.g. freelist header */
2049 int fields) /* mask of fields to be logged (XFS_AGF_...) */
2050 {
2051 int first; /* first byte offset */
2052 int last; /* last byte offset */
2053 static const short offsets[] = {
2054 offsetof(xfs_agf_t, agf_magicnum),
2055 offsetof(xfs_agf_t, agf_versionnum),
2056 offsetof(xfs_agf_t, agf_seqno),
2057 offsetof(xfs_agf_t, agf_length),
2058 offsetof(xfs_agf_t, agf_roots[0]),
2059 offsetof(xfs_agf_t, agf_levels[0]),
2060 offsetof(xfs_agf_t, agf_flfirst),
2061 offsetof(xfs_agf_t, agf_fllast),
2062 offsetof(xfs_agf_t, agf_flcount),
2063 offsetof(xfs_agf_t, agf_freeblks),
2064 offsetof(xfs_agf_t, agf_longest),
2065 offsetof(xfs_agf_t, agf_btreeblks),
2066 sizeof(xfs_agf_t)
2067 };
2068
2069 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2070 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2071 }
2072
2073 /*
2074 * Interface for inode allocation to force the pag data to be initialized.
2075 */
2076 int /* error */
2077 xfs_alloc_pagf_init(
2078 xfs_mount_t *mp, /* file system mount structure */
2079 xfs_trans_t *tp, /* transaction pointer */
2080 xfs_agnumber_t agno, /* allocation group number */
2081 int flags) /* XFS_ALLOC_FLAGS_... */
2082 {
2083 xfs_buf_t *bp;
2084 int error;
2085
2086 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2087 return error;
2088 if (bp)
2089 xfs_trans_brelse(tp, bp);
2090 return 0;
2091 }
2092
2093 /*
2094 * Put the block on the freelist for the allocation group.
2095 */
2096 int /* error */
2097 xfs_alloc_put_freelist(
2098 xfs_trans_t *tp, /* transaction pointer */
2099 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2100 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2101 xfs_agblock_t bno, /* block being freed */
2102 int btreeblk) /* block came from a AGF btree */
2103 {
2104 xfs_agf_t *agf; /* a.g. freespace structure */
2105 xfs_agfl_t *agfl; /* a.g. free block array */
2106 __be32 *blockp;/* pointer to array entry */
2107 int error;
2108 int logflags;
2109 xfs_mount_t *mp; /* mount structure */
2110 xfs_perag_t *pag; /* per allocation group data */
2111
2112 agf = XFS_BUF_TO_AGF(agbp);
2113 mp = tp->t_mountp;
2114
2115 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
2116 be32_to_cpu(agf->agf_seqno), &agflbp)))
2117 return error;
2118 agfl = XFS_BUF_TO_AGFL(agflbp);
2119 be32_add_cpu(&agf->agf_fllast, 1);
2120 if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
2121 agf->agf_fllast = 0;
2122 pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)];
2123 be32_add_cpu(&agf->agf_flcount, 1);
2124 xfs_trans_agflist_delta(tp, 1);
2125 pag->pagf_flcount++;
2126
2127 logflags = XFS_AGF_FLLAST | XFS_AGF_FLCOUNT;
2128 if (btreeblk) {
2129 be32_add_cpu(&agf->agf_btreeblks, -1);
2130 pag->pagf_btreeblks--;
2131 logflags |= XFS_AGF_BTREEBLKS;
2132 }
2133
2134 TRACE_MODAGF(NULL, agf, logflags);
2135 xfs_alloc_log_agf(tp, agbp, logflags);
2136
2137 ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2138 blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)];
2139 *blockp = cpu_to_be32(bno);
2140 TRACE_MODAGF(NULL, agf, logflags);
2141 xfs_alloc_log_agf(tp, agbp, logflags);
2142 xfs_trans_log_buf(tp, agflbp,
2143 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl),
2144 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl +
2145 sizeof(xfs_agblock_t) - 1));
2146 return 0;
2147 }
2148
2149 /*
2150 * Read in the allocation group header (free/alloc section).
2151 */
2152 int /* error */
2153 xfs_alloc_read_agf(
2154 xfs_mount_t *mp, /* mount point structure */
2155 xfs_trans_t *tp, /* transaction pointer */
2156 xfs_agnumber_t agno, /* allocation group number */
2157 int flags, /* XFS_ALLOC_FLAG_... */
2158 xfs_buf_t **bpp) /* buffer for the ag freelist header */
2159 {
2160 xfs_agf_t *agf; /* ag freelist header */
2161 int agf_ok; /* set if agf is consistent */
2162 xfs_buf_t *bp; /* return value */
2163 xfs_perag_t *pag; /* per allocation group data */
2164 int error;
2165
2166 ASSERT(agno != NULLAGNUMBER);
2167 error = xfs_trans_read_buf(
2168 mp, tp, mp->m_ddev_targp,
2169 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2170 XFS_FSS_TO_BB(mp, 1),
2171 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XFS_BUF_TRYLOCK : 0U,
2172 &bp);
2173 if (error)
2174 return error;
2175 ASSERT(!bp || !XFS_BUF_GETERROR(bp));
2176 if (!bp) {
2177 *bpp = NULL;
2178 return 0;
2179 }
2180 /*
2181 * Validate the magic number of the agf block.
2182 */
2183 agf = XFS_BUF_TO_AGF(bp);
2184 agf_ok =
2185 be32_to_cpu(agf->agf_magicnum) == XFS_AGF_MAGIC &&
2186 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2187 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2188 be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2189 be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2190 be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp);
2191 if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF,
2192 XFS_RANDOM_ALLOC_READ_AGF))) {
2193 XFS_CORRUPTION_ERROR("xfs_alloc_read_agf",
2194 XFS_ERRLEVEL_LOW, mp, agf);
2195 xfs_trans_brelse(tp, bp);
2196 return XFS_ERROR(EFSCORRUPTED);
2197 }
2198 pag = &mp->m_perag[agno];
2199 if (!pag->pagf_init) {
2200 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2201 pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks);
2202 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2203 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
2204 pag->pagf_levels[XFS_BTNUM_BNOi] =
2205 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
2206 pag->pagf_levels[XFS_BTNUM_CNTi] =
2207 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2208 spin_lock_init(&pag->pagb_lock);
2209 pag->pagb_list = kmem_zalloc(XFS_PAGB_NUM_SLOTS *
2210 sizeof(xfs_perag_busy_t), KM_SLEEP);
2211 pag->pagf_init = 1;
2212 }
2213 #ifdef DEBUG
2214 else if (!XFS_FORCED_SHUTDOWN(mp)) {
2215 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2216 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2217 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
2218 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
2219 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
2220 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
2221 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
2222 }
2223 #endif
2224 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGF, XFS_AGF_REF);
2225 *bpp = bp;
2226 return 0;
2227 }
2228
2229 /*
2230 * Allocate an extent (variable-size).
2231 * Depending on the allocation type, we either look in a single allocation
2232 * group or loop over the allocation groups to find the result.
2233 */
2234 int /* error */
2235 xfs_alloc_vextent(
2236 xfs_alloc_arg_t *args) /* allocation argument structure */
2237 {
2238 xfs_agblock_t agsize; /* allocation group size */
2239 int error;
2240 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2241 xfs_extlen_t minleft;/* minimum left value, temp copy */
2242 xfs_mount_t *mp; /* mount structure pointer */
2243 xfs_agnumber_t sagno; /* starting allocation group number */
2244 xfs_alloctype_t type; /* input allocation type */
2245 int bump_rotor = 0;
2246 int no_min = 0;
2247 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2248
2249 mp = args->mp;
2250 type = args->otype = args->type;
2251 args->agbno = NULLAGBLOCK;
2252 /*
2253 * Just fix this up, for the case where the last a.g. is shorter
2254 * (or there's only one a.g.) and the caller couldn't easily figure
2255 * that out (xfs_bmap_alloc).
2256 */
2257 agsize = mp->m_sb.sb_agblocks;
2258 if (args->maxlen > agsize)
2259 args->maxlen = agsize;
2260 if (args->alignment == 0)
2261 args->alignment = 1;
2262 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2263 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2264 ASSERT(args->minlen <= args->maxlen);
2265 ASSERT(args->minlen <= agsize);
2266 ASSERT(args->mod < args->prod);
2267 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2268 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2269 args->minlen > args->maxlen || args->minlen > agsize ||
2270 args->mod >= args->prod) {
2271 args->fsbno = NULLFSBLOCK;
2272 TRACE_ALLOC("badargs", args);
2273 return 0;
2274 }
2275 minleft = args->minleft;
2276
2277 switch (type) {
2278 case XFS_ALLOCTYPE_THIS_AG:
2279 case XFS_ALLOCTYPE_NEAR_BNO:
2280 case XFS_ALLOCTYPE_THIS_BNO:
2281 /*
2282 * These three force us into a single a.g.
2283 */
2284 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2285 down_read(&mp->m_peraglock);
2286 args->pag = &mp->m_perag[args->agno];
2287 args->minleft = 0;
2288 error = xfs_alloc_fix_freelist(args, 0);
2289 args->minleft = minleft;
2290 if (error) {
2291 TRACE_ALLOC("nofix", args);
2292 goto error0;
2293 }
2294 if (!args->agbp) {
2295 up_read(&mp->m_peraglock);
2296 TRACE_ALLOC("noagbp", args);
2297 break;
2298 }
2299 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2300 if ((error = xfs_alloc_ag_vextent(args)))
2301 goto error0;
2302 up_read(&mp->m_peraglock);
2303 break;
2304 case XFS_ALLOCTYPE_START_BNO:
2305 /*
2306 * Try near allocation first, then anywhere-in-ag after
2307 * the first a.g. fails.
2308 */
2309 if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2310 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2311 args->fsbno = XFS_AGB_TO_FSB(mp,
2312 ((mp->m_agfrotor / rotorstep) %
2313 mp->m_sb.sb_agcount), 0);
2314 bump_rotor = 1;
2315 }
2316 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2317 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2318 /* FALLTHROUGH */
2319 case XFS_ALLOCTYPE_ANY_AG:
2320 case XFS_ALLOCTYPE_START_AG:
2321 case XFS_ALLOCTYPE_FIRST_AG:
2322 /*
2323 * Rotate through the allocation groups looking for a winner.
2324 */
2325 if (type == XFS_ALLOCTYPE_ANY_AG) {
2326 /*
2327 * Start with the last place we left off.
2328 */
2329 args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2330 mp->m_sb.sb_agcount;
2331 args->type = XFS_ALLOCTYPE_THIS_AG;
2332 flags = XFS_ALLOC_FLAG_TRYLOCK;
2333 } else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2334 /*
2335 * Start with allocation group given by bno.
2336 */
2337 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2338 args->type = XFS_ALLOCTYPE_THIS_AG;
2339 sagno = 0;
2340 flags = 0;
2341 } else {
2342 if (type == XFS_ALLOCTYPE_START_AG)
2343 args->type = XFS_ALLOCTYPE_THIS_AG;
2344 /*
2345 * Start with the given allocation group.
2346 */
2347 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2348 flags = XFS_ALLOC_FLAG_TRYLOCK;
2349 }
2350 /*
2351 * Loop over allocation groups twice; first time with
2352 * trylock set, second time without.
2353 */
2354 down_read(&mp->m_peraglock);
2355 for (;;) {
2356 args->pag = &mp->m_perag[args->agno];
2357 if (no_min) args->minleft = 0;
2358 error = xfs_alloc_fix_freelist(args, flags);
2359 args->minleft = minleft;
2360 if (error) {
2361 TRACE_ALLOC("nofix", args);
2362 goto error0;
2363 }
2364 /*
2365 * If we get a buffer back then the allocation will fly.
2366 */
2367 if (args->agbp) {
2368 if ((error = xfs_alloc_ag_vextent(args)))
2369 goto error0;
2370 break;
2371 }
2372 TRACE_ALLOC("loopfailed", args);
2373 /*
2374 * Didn't work, figure out the next iteration.
2375 */
2376 if (args->agno == sagno &&
2377 type == XFS_ALLOCTYPE_START_BNO)
2378 args->type = XFS_ALLOCTYPE_THIS_AG;
2379 /*
2380 * For the first allocation, we can try any AG to get
2381 * space. However, if we already have allocated a
2382 * block, we don't want to try AGs whose number is below
2383 * sagno. Otherwise, we may end up with out-of-order
2384 * locking of AGF, which might cause deadlock.
2385 */
2386 if (++(args->agno) == mp->m_sb.sb_agcount) {
2387 if (args->firstblock != NULLFSBLOCK)
2388 args->agno = sagno;
2389 else
2390 args->agno = 0;
2391 }
2392 /*
2393 * Reached the starting a.g., must either be done
2394 * or switch to non-trylock mode.
2395 */
2396 if (args->agno == sagno) {
2397 if (no_min == 1) {
2398 args->agbno = NULLAGBLOCK;
2399 TRACE_ALLOC("allfailed", args);
2400 break;
2401 }
2402 if (flags == 0) {
2403 no_min = 1;
2404 } else {
2405 flags = 0;
2406 if (type == XFS_ALLOCTYPE_START_BNO) {
2407 args->agbno = XFS_FSB_TO_AGBNO(mp,
2408 args->fsbno);
2409 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2410 }
2411 }
2412 }
2413 }
2414 up_read(&mp->m_peraglock);
2415 if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2416 if (args->agno == sagno)
2417 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2418 (mp->m_sb.sb_agcount * rotorstep);
2419 else
2420 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2421 (mp->m_sb.sb_agcount * rotorstep);
2422 }
2423 break;
2424 default:
2425 ASSERT(0);
2426 /* NOTREACHED */
2427 }
2428 if (args->agbno == NULLAGBLOCK)
2429 args->fsbno = NULLFSBLOCK;
2430 else {
2431 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2432 #ifdef DEBUG
2433 ASSERT(args->len >= args->minlen);
2434 ASSERT(args->len <= args->maxlen);
2435 ASSERT(args->agbno % args->alignment == 0);
2436 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2437 args->len);
2438 #endif
2439 }
2440 return 0;
2441 error0:
2442 up_read(&mp->m_peraglock);
2443 return error;
2444 }
2445
2446 /*
2447 * Free an extent.
2448 * Just break up the extent address and hand off to xfs_free_ag_extent
2449 * after fixing up the freelist.
2450 */
2451 int /* error */
2452 xfs_free_extent(
2453 xfs_trans_t *tp, /* transaction pointer */
2454 xfs_fsblock_t bno, /* starting block number of extent */
2455 xfs_extlen_t len) /* length of extent */
2456 {
2457 xfs_alloc_arg_t args;
2458 int error;
2459
2460 ASSERT(len != 0);
2461 memset(&args, 0, sizeof(xfs_alloc_arg_t));
2462 args.tp = tp;
2463 args.mp = tp->t_mountp;
2464 args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2465 ASSERT(args.agno < args.mp->m_sb.sb_agcount);
2466 args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2467 down_read(&args.mp->m_peraglock);
2468 args.pag = &args.mp->m_perag[args.agno];
2469 if ((error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING)))
2470 goto error0;
2471 #ifdef DEBUG
2472 ASSERT(args.agbp != NULL);
2473 ASSERT((args.agbno + len) <=
2474 be32_to_cpu(XFS_BUF_TO_AGF(args.agbp)->agf_length));
2475 #endif
2476 error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0);
2477 error0:
2478 up_read(&args.mp->m_peraglock);
2479 return error;
2480 }
2481
2482
2483 /*
2484 * AG Busy list management
2485 * The busy list contains block ranges that have been freed but whose
2486 * transactions have not yet hit disk. If any block listed in a busy
2487 * list is reused, the transaction that freed it must be forced to disk
2488 * before continuing to use the block.
2489 *
2490 * xfs_alloc_mark_busy - add to the per-ag busy list
2491 * xfs_alloc_clear_busy - remove an item from the per-ag busy list
2492 */
2493 void
2494 xfs_alloc_mark_busy(xfs_trans_t *tp,
2495 xfs_agnumber_t agno,
2496 xfs_agblock_t bno,
2497 xfs_extlen_t len)
2498 {
2499 xfs_mount_t *mp;
2500 xfs_perag_busy_t *bsy;
2501 int n;
2502
2503 mp = tp->t_mountp;
2504 spin_lock(&mp->m_perag[agno].pagb_lock);
2505
2506 /* search pagb_list for an open slot */
2507 for (bsy = mp->m_perag[agno].pagb_list, n = 0;
2508 n < XFS_PAGB_NUM_SLOTS;
2509 bsy++, n++) {
2510 if (bsy->busy_tp == NULL) {
2511 break;
2512 }
2513 }
2514
2515 if (n < XFS_PAGB_NUM_SLOTS) {
2516 bsy = &mp->m_perag[agno].pagb_list[n];
2517 mp->m_perag[agno].pagb_count++;
2518 TRACE_BUSY("xfs_alloc_mark_busy", "got", agno, bno, len, n, tp);
2519 bsy->busy_start = bno;
2520 bsy->busy_length = len;
2521 bsy->busy_tp = tp;
2522 xfs_trans_add_busy(tp, agno, n);
2523 } else {
2524 TRACE_BUSY("xfs_alloc_mark_busy", "FULL", agno, bno, len, -1, tp);
2525 /*
2526 * The busy list is full! Since it is now not possible to
2527 * track the free block, make this a synchronous transaction
2528 * to insure that the block is not reused before this
2529 * transaction commits.
2530 */
2531 xfs_trans_set_sync(tp);
2532 }
2533
2534 spin_unlock(&mp->m_perag[agno].pagb_lock);
2535 }
2536
2537 void
2538 xfs_alloc_clear_busy(xfs_trans_t *tp,
2539 xfs_agnumber_t agno,
2540 int idx)
2541 {
2542 xfs_mount_t *mp;
2543 xfs_perag_busy_t *list;
2544
2545 mp = tp->t_mountp;
2546
2547 spin_lock(&mp->m_perag[agno].pagb_lock);
2548 list = mp->m_perag[agno].pagb_list;
2549
2550 ASSERT(idx < XFS_PAGB_NUM_SLOTS);
2551 if (list[idx].busy_tp == tp) {
2552 TRACE_UNBUSY("xfs_alloc_clear_busy", "found", agno, idx, tp);
2553 list[idx].busy_tp = NULL;
2554 mp->m_perag[agno].pagb_count--;
2555 } else {
2556 TRACE_UNBUSY("xfs_alloc_clear_busy", "missing", agno, idx, tp);
2557 }
2558
2559 spin_unlock(&mp->m_perag[agno].pagb_lock);
2560 }
2561
2562
2563 /*
2564 * If we find the extent in the busy list, force the log out to get the
2565 * extent out of the busy list so the caller can use it straight away.
2566 */
2567 STATIC void
2568 xfs_alloc_search_busy(xfs_trans_t *tp,
2569 xfs_agnumber_t agno,
2570 xfs_agblock_t bno,
2571 xfs_extlen_t len)
2572 {
2573 xfs_mount_t *mp;
2574 xfs_perag_busy_t *bsy;
2575 xfs_agblock_t uend, bend;
2576 xfs_lsn_t lsn;
2577 int cnt;
2578
2579 mp = tp->t_mountp;
2580
2581 spin_lock(&mp->m_perag[agno].pagb_lock);
2582 cnt = mp->m_perag[agno].pagb_count;
2583
2584 uend = bno + len - 1;
2585
2586 /* search pagb_list for this slot, skipping open slots */
2587 for (bsy = mp->m_perag[agno].pagb_list; cnt; bsy++) {
2588
2589 /*
2590 * (start1,length1) within (start2, length2)
2591 */
2592 if (bsy->busy_tp != NULL) {
2593 bend = bsy->busy_start + bsy->busy_length - 1;
2594 if ((bno > bend) || (uend < bsy->busy_start)) {
2595 cnt--;
2596 } else {
2597 TRACE_BUSYSEARCH("xfs_alloc_search_busy",
2598 "found1", agno, bno, len, tp);
2599 break;
2600 }
2601 }
2602 }
2603
2604 /*
2605 * If a block was found, force the log through the LSN of the
2606 * transaction that freed the block
2607 */
2608 if (cnt) {
2609 TRACE_BUSYSEARCH("xfs_alloc_search_busy", "found", agno, bno, len, tp);
2610 lsn = bsy->busy_tp->t_commit_lsn;
2611 spin_unlock(&mp->m_perag[agno].pagb_lock);
2612 xfs_log_force(mp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC);
2613 } else {
2614 TRACE_BUSYSEARCH("xfs_alloc_search_busy", "not-found", agno, bno, len, tp);
2615 spin_unlock(&mp->m_perag[agno].pagb_lock);
2616 }
2617 }
kmemtrace - Kernel memory tracer