blob: eef6763f3a6732084ede458d4a8b38b5e01e3fd4 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Nathan Scotta844f452005-11-02 14:38:42 +110027#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_dmapi.h"
29#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_dir2_sf.h"
34#include "xfs_attr_sf.h"
35#include "xfs_dinode.h"
36#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_btree.h"
38#include "xfs_ialloc.h"
39#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#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
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100048STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -070049xfs_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)
55ktrace_t *xfs_alloc_trace_buf;
56
57#define TRACE_ALLOC(s,a) \
58 xfs_alloc_trace_alloc(fname, s, a, __LINE__)
59#define TRACE_FREE(s,a,b,x,f) \
60 xfs_alloc_trace_free(fname, s, mp, a, b, x, f, __LINE__)
61#define TRACE_MODAGF(s,a,f) \
62 xfs_alloc_trace_modagf(fname, s, mp, a, f, __LINE__)
63#define TRACE_BUSY(fname,s,ag,agb,l,sl,tp) \
64 xfs_alloc_trace_busy(fname, s, mp, ag, agb, l, sl, tp, XFS_ALLOC_KTRACE_BUSY, __LINE__)
65#define TRACE_UNBUSY(fname,s,ag,sl,tp) \
66 xfs_alloc_trace_busy(fname, s, mp, ag, -1, -1, sl, tp, XFS_ALLOC_KTRACE_UNBUSY, __LINE__)
67#define TRACE_BUSYSEARCH(fname,s,ag,agb,l,sl,tp) \
68 xfs_alloc_trace_busy(fname, s, mp, ag, agb, l, sl, 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,sl,tp)
76#endif /* XFS_ALLOC_TRACE */
77
78/*
79 * Prototypes for per-ag allocation routines
80 */
81
82STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
83STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
84STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
85STATIC 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 */
96STATIC int /* success (>= minlen) */
97xfs_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 return len >= minlen;
120}
121
122/*
123 * Compute best start block and diff for "near" allocations.
124 * freelen >= wantlen already checked by caller.
125 */
126STATIC xfs_extlen_t /* difference value (absolute) */
127xfs_alloc_compute_diff(
128 xfs_agblock_t wantbno, /* target starting block */
129 xfs_extlen_t wantlen, /* target length */
130 xfs_extlen_t alignment, /* target alignment */
131 xfs_agblock_t freebno, /* freespace's starting block */
132 xfs_extlen_t freelen, /* freespace's length */
133 xfs_agblock_t *newbnop) /* result: best start block from free */
134{
135 xfs_agblock_t freeend; /* end of freespace extent */
136 xfs_agblock_t newbno1; /* return block number */
137 xfs_agblock_t newbno2; /* other new block number */
138 xfs_extlen_t newlen1=0; /* length with newbno1 */
139 xfs_extlen_t newlen2=0; /* length with newbno2 */
140 xfs_agblock_t wantend; /* end of target extent */
141
142 ASSERT(freelen >= wantlen);
143 freeend = freebno + freelen;
144 wantend = wantbno + wantlen;
145 if (freebno >= wantbno) {
146 if ((newbno1 = roundup(freebno, alignment)) >= freeend)
147 newbno1 = NULLAGBLOCK;
148 } else if (freeend >= wantend && alignment > 1) {
149 newbno1 = roundup(wantbno, alignment);
150 newbno2 = newbno1 - alignment;
151 if (newbno1 >= freeend)
152 newbno1 = NULLAGBLOCK;
153 else
154 newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
155 if (newbno2 < freebno)
156 newbno2 = NULLAGBLOCK;
157 else
158 newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
159 if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
160 if (newlen1 < newlen2 ||
161 (newlen1 == newlen2 &&
162 XFS_ABSDIFF(newbno1, wantbno) >
163 XFS_ABSDIFF(newbno2, wantbno)))
164 newbno1 = newbno2;
165 } else if (newbno2 != NULLAGBLOCK)
166 newbno1 = newbno2;
167 } else if (freeend >= wantend) {
168 newbno1 = wantbno;
169 } else if (alignment > 1) {
170 newbno1 = roundup(freeend - wantlen, alignment);
171 if (newbno1 > freeend - wantlen &&
172 newbno1 - alignment >= freebno)
173 newbno1 -= alignment;
174 else if (newbno1 >= freeend)
175 newbno1 = NULLAGBLOCK;
176 } else
177 newbno1 = freeend - wantlen;
178 *newbnop = newbno1;
179 return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
180}
181
182/*
183 * Fix up the length, based on mod and prod.
184 * len should be k * prod + mod for some k.
185 * If len is too small it is returned unchanged.
186 * If len hits maxlen it is left alone.
187 */
188STATIC void
189xfs_alloc_fix_len(
190 xfs_alloc_arg_t *args) /* allocation argument structure */
191{
192 xfs_extlen_t k;
193 xfs_extlen_t rlen;
194
195 ASSERT(args->mod < args->prod);
196 rlen = args->len;
197 ASSERT(rlen >= args->minlen);
198 ASSERT(rlen <= args->maxlen);
199 if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
200 (args->mod == 0 && rlen < args->prod))
201 return;
202 k = rlen % args->prod;
203 if (k == args->mod)
204 return;
205 if (k > args->mod) {
206 if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen)
207 return;
208 } else {
209 if ((int)(rlen = rlen - args->prod - (args->mod - k)) <
210 (int)args->minlen)
211 return;
212 }
213 ASSERT(rlen >= args->minlen);
214 ASSERT(rlen <= args->maxlen);
215 args->len = rlen;
216}
217
218/*
219 * Fix up length if there is too little space left in the a.g.
220 * Return 1 if ok, 0 if too little, should give up.
221 */
222STATIC int
223xfs_alloc_fix_minleft(
224 xfs_alloc_arg_t *args) /* allocation argument structure */
225{
226 xfs_agf_t *agf; /* a.g. freelist header */
227 int diff; /* free space difference */
228
229 if (args->minleft == 0)
230 return 1;
231 agf = XFS_BUF_TO_AGF(args->agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100232 diff = be32_to_cpu(agf->agf_freeblks)
233 + be32_to_cpu(agf->agf_flcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 - args->len - args->minleft;
235 if (diff >= 0)
236 return 1;
237 args->len += diff; /* shrink the allocated space */
238 if (args->len >= args->minlen)
239 return 1;
240 args->agbno = NULLAGBLOCK;
241 return 0;
242}
243
244/*
245 * Update the two btrees, logically removing from freespace the extent
246 * starting at rbno, rlen blocks. The extent is contained within the
247 * actual (current) free extent fbno for flen blocks.
248 * Flags are passed in indicating whether the cursors are set to the
249 * relevant records.
250 */
251STATIC int /* error code */
252xfs_alloc_fixup_trees(
253 xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */
254 xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */
255 xfs_agblock_t fbno, /* starting block of free extent */
256 xfs_extlen_t flen, /* length of free extent */
257 xfs_agblock_t rbno, /* starting block of returned extent */
258 xfs_extlen_t rlen, /* length of returned extent */
259 int flags) /* flags, XFSA_FIXUP_... */
260{
261 int error; /* error code */
262 int i; /* operation results */
263 xfs_agblock_t nfbno1; /* first new free startblock */
264 xfs_agblock_t nfbno2; /* second new free startblock */
265 xfs_extlen_t nflen1=0; /* first new free length */
266 xfs_extlen_t nflen2=0; /* second new free length */
267
268 /*
269 * Look up the record in the by-size tree if necessary.
270 */
271 if (flags & XFSA_FIXUP_CNT_OK) {
272#ifdef DEBUG
273 if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
274 return error;
275 XFS_WANT_CORRUPTED_RETURN(
276 i == 1 && nfbno1 == fbno && nflen1 == flen);
277#endif
278 } else {
279 if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
280 return error;
281 XFS_WANT_CORRUPTED_RETURN(i == 1);
282 }
283 /*
284 * Look up the record in the by-block tree if necessary.
285 */
286 if (flags & XFSA_FIXUP_BNO_OK) {
287#ifdef DEBUG
288 if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
289 return error;
290 XFS_WANT_CORRUPTED_RETURN(
291 i == 1 && nfbno1 == fbno && nflen1 == flen);
292#endif
293 } else {
294 if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
295 return error;
296 XFS_WANT_CORRUPTED_RETURN(i == 1);
297 }
298#ifdef DEBUG
299 {
300 xfs_alloc_block_t *bnoblock;
301 xfs_alloc_block_t *cntblock;
302
303 if (bno_cur->bc_nlevels == 1 &&
304 cnt_cur->bc_nlevels == 1) {
305 bnoblock = XFS_BUF_TO_ALLOC_BLOCK(bno_cur->bc_bufs[0]);
306 cntblock = XFS_BUF_TO_ALLOC_BLOCK(cnt_cur->bc_bufs[0]);
307 XFS_WANT_CORRUPTED_RETURN(
Christoph Hellwig16259e72005-11-02 15:11:25 +1100308 be16_to_cpu(bnoblock->bb_numrecs) ==
309 be16_to_cpu(cntblock->bb_numrecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 }
311 }
312#endif
313 /*
314 * Deal with all four cases: the allocated record is contained
315 * within the freespace record, so we can have new freespace
316 * at either (or both) end, or no freespace remaining.
317 */
318 if (rbno == fbno && rlen == flen)
319 nfbno1 = nfbno2 = NULLAGBLOCK;
320 else if (rbno == fbno) {
321 nfbno1 = rbno + rlen;
322 nflen1 = flen - rlen;
323 nfbno2 = NULLAGBLOCK;
324 } else if (rbno + rlen == fbno + flen) {
325 nfbno1 = fbno;
326 nflen1 = flen - rlen;
327 nfbno2 = NULLAGBLOCK;
328 } else {
329 nfbno1 = fbno;
330 nflen1 = rbno - fbno;
331 nfbno2 = rbno + rlen;
332 nflen2 = (fbno + flen) - nfbno2;
333 }
334 /*
335 * Delete the entry from the by-size btree.
336 */
337 if ((error = xfs_alloc_delete(cnt_cur, &i)))
338 return error;
339 XFS_WANT_CORRUPTED_RETURN(i == 1);
340 /*
341 * Add new by-size btree entry(s).
342 */
343 if (nfbno1 != NULLAGBLOCK) {
344 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
345 return error;
346 XFS_WANT_CORRUPTED_RETURN(i == 0);
347 if ((error = xfs_alloc_insert(cnt_cur, &i)))
348 return error;
349 XFS_WANT_CORRUPTED_RETURN(i == 1);
350 }
351 if (nfbno2 != NULLAGBLOCK) {
352 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
353 return error;
354 XFS_WANT_CORRUPTED_RETURN(i == 0);
355 if ((error = xfs_alloc_insert(cnt_cur, &i)))
356 return error;
357 XFS_WANT_CORRUPTED_RETURN(i == 1);
358 }
359 /*
360 * Fix up the by-block btree entry(s).
361 */
362 if (nfbno1 == NULLAGBLOCK) {
363 /*
364 * No remaining freespace, just delete the by-block tree entry.
365 */
366 if ((error = xfs_alloc_delete(bno_cur, &i)))
367 return error;
368 XFS_WANT_CORRUPTED_RETURN(i == 1);
369 } else {
370 /*
371 * Update the by-block entry to start later|be shorter.
372 */
373 if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
374 return error;
375 }
376 if (nfbno2 != NULLAGBLOCK) {
377 /*
378 * 2 resulting free entries, need to add one.
379 */
380 if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
381 return error;
382 XFS_WANT_CORRUPTED_RETURN(i == 0);
383 if ((error = xfs_alloc_insert(bno_cur, &i)))
384 return error;
385 XFS_WANT_CORRUPTED_RETURN(i == 1);
386 }
387 return 0;
388}
389
390/*
391 * Read in the allocation group free block array.
392 */
393STATIC int /* error */
394xfs_alloc_read_agfl(
395 xfs_mount_t *mp, /* mount point structure */
396 xfs_trans_t *tp, /* transaction pointer */
397 xfs_agnumber_t agno, /* allocation group number */
398 xfs_buf_t **bpp) /* buffer for the ag free block array */
399{
400 xfs_buf_t *bp; /* return value */
401 int error;
402
403 ASSERT(agno != NULLAGNUMBER);
404 error = xfs_trans_read_buf(
405 mp, tp, mp->m_ddev_targp,
406 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
407 XFS_FSS_TO_BB(mp, 1), 0, &bp);
408 if (error)
409 return error;
410 ASSERT(bp);
411 ASSERT(!XFS_BUF_GETERROR(bp));
412 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGFL, XFS_AGFL_REF);
413 *bpp = bp;
414 return 0;
415}
416
417#if defined(XFS_ALLOC_TRACE)
418/*
419 * Add an allocation trace entry for an alloc call.
420 */
421STATIC void
422xfs_alloc_trace_alloc(
423 char *name, /* function tag string */
424 char *str, /* additional string */
425 xfs_alloc_arg_t *args, /* allocation argument structure */
426 int line) /* source line number */
427{
428 ktrace_enter(xfs_alloc_trace_buf,
429 (void *)(__psint_t)(XFS_ALLOC_KTRACE_ALLOC | (line << 16)),
430 (void *)name,
431 (void *)str,
432 (void *)args->mp,
433 (void *)(__psunsigned_t)args->agno,
434 (void *)(__psunsigned_t)args->agbno,
435 (void *)(__psunsigned_t)args->minlen,
436 (void *)(__psunsigned_t)args->maxlen,
437 (void *)(__psunsigned_t)args->mod,
438 (void *)(__psunsigned_t)args->prod,
439 (void *)(__psunsigned_t)args->minleft,
440 (void *)(__psunsigned_t)args->total,
441 (void *)(__psunsigned_t)args->alignment,
442 (void *)(__psunsigned_t)args->len,
443 (void *)((((__psint_t)args->type) << 16) |
444 (__psint_t)args->otype),
445 (void *)(__psint_t)((args->wasdel << 3) |
446 (args->wasfromfl << 2) |
447 (args->isfl << 1) |
448 (args->userdata << 0)));
449}
450
451/*
452 * Add an allocation trace entry for a free call.
453 */
454STATIC void
455xfs_alloc_trace_free(
456 char *name, /* function tag string */
457 char *str, /* additional string */
458 xfs_mount_t *mp, /* file system mount point */
459 xfs_agnumber_t agno, /* allocation group number */
460 xfs_agblock_t agbno, /* a.g. relative block number */
461 xfs_extlen_t len, /* length of extent */
462 int isfl, /* set if is freelist allocation/free */
463 int line) /* source line number */
464{
465 ktrace_enter(xfs_alloc_trace_buf,
466 (void *)(__psint_t)(XFS_ALLOC_KTRACE_FREE | (line << 16)),
467 (void *)name,
468 (void *)str,
469 (void *)mp,
470 (void *)(__psunsigned_t)agno,
471 (void *)(__psunsigned_t)agbno,
472 (void *)(__psunsigned_t)len,
473 (void *)(__psint_t)isfl,
474 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL);
475}
476
477/*
478 * Add an allocation trace entry for modifying an agf.
479 */
480STATIC void
481xfs_alloc_trace_modagf(
482 char *name, /* function tag string */
483 char *str, /* additional string */
484 xfs_mount_t *mp, /* file system mount point */
485 xfs_agf_t *agf, /* new agf value */
486 int flags, /* logging flags for agf */
487 int line) /* source line number */
488{
489 ktrace_enter(xfs_alloc_trace_buf,
490 (void *)(__psint_t)(XFS_ALLOC_KTRACE_MODAGF | (line << 16)),
491 (void *)name,
492 (void *)str,
493 (void *)mp,
494 (void *)(__psint_t)flags,
Christoph Hellwig16259e72005-11-02 15:11:25 +1100495 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_seqno),
496 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_length),
497 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_BNO]),
498 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_CNT]),
499 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]),
500 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]),
501 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flfirst),
502 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_fllast),
503 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flcount),
504 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_freeblks),
505 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_longest));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
507
508STATIC void
509xfs_alloc_trace_busy(
510 char *name, /* function tag string */
511 char *str, /* additional string */
Nathan Scottc41564b2006-03-29 08:55:14 +1000512 xfs_mount_t *mp, /* file system mount point */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 xfs_agnumber_t agno, /* allocation group number */
514 xfs_agblock_t agbno, /* a.g. relative block number */
515 xfs_extlen_t len, /* length of extent */
516 int slot, /* perag Busy slot */
517 xfs_trans_t *tp,
518 int trtype, /* type: add, delete, search */
519 int line) /* source line number */
520{
521 ktrace_enter(xfs_alloc_trace_buf,
522 (void *)(__psint_t)(trtype | (line << 16)),
523 (void *)name,
524 (void *)str,
525 (void *)mp,
526 (void *)(__psunsigned_t)agno,
527 (void *)(__psunsigned_t)agbno,
528 (void *)(__psunsigned_t)len,
529 (void *)(__psint_t)slot,
530 (void *)tp,
531 NULL, NULL, NULL, NULL, NULL, NULL, NULL);
532}
533#endif /* XFS_ALLOC_TRACE */
534
535/*
536 * Allocation group level functions.
537 */
538
539/*
540 * Allocate a variable extent in the allocation group agno.
541 * Type and bno are used to determine where in the allocation group the
542 * extent will start.
543 * Extent's length (returned in *len) will be between minlen and maxlen,
544 * and of the form k * prod + mod unless there's nothing that large.
545 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
546 */
547STATIC int /* error */
548xfs_alloc_ag_vextent(
549 xfs_alloc_arg_t *args) /* argument structure for allocation */
550{
551 int error=0;
552#ifdef XFS_ALLOC_TRACE
553 static char fname[] = "xfs_alloc_ag_vextent";
554#endif
555
556 ASSERT(args->minlen > 0);
557 ASSERT(args->maxlen > 0);
558 ASSERT(args->minlen <= args->maxlen);
559 ASSERT(args->mod < args->prod);
560 ASSERT(args->alignment > 0);
561 /*
562 * Branch to correct routine based on the type.
563 */
564 args->wasfromfl = 0;
565 switch (args->type) {
566 case XFS_ALLOCTYPE_THIS_AG:
567 error = xfs_alloc_ag_vextent_size(args);
568 break;
569 case XFS_ALLOCTYPE_NEAR_BNO:
570 error = xfs_alloc_ag_vextent_near(args);
571 break;
572 case XFS_ALLOCTYPE_THIS_BNO:
573 error = xfs_alloc_ag_vextent_exact(args);
574 break;
575 default:
576 ASSERT(0);
577 /* NOTREACHED */
578 }
579 if (error)
580 return error;
581 /*
582 * If the allocation worked, need to change the agf structure
583 * (and log it), and the superblock.
584 */
585 if (args->agbno != NULLAGBLOCK) {
586 xfs_agf_t *agf; /* allocation group freelist header */
587#ifdef XFS_ALLOC_TRACE
588 xfs_mount_t *mp = args->mp;
589#endif
590 long slen = (long)args->len;
591
592 ASSERT(args->len >= args->minlen && args->len <= args->maxlen);
593 ASSERT(!(args->wasfromfl) || !args->isfl);
594 ASSERT(args->agbno % args->alignment == 0);
595 if (!(args->wasfromfl)) {
596
597 agf = XFS_BUF_TO_AGF(args->agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100598 be32_add(&agf->agf_freeblks, -(args->len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 xfs_trans_agblocks_delta(args->tp,
600 -((long)(args->len)));
601 args->pag->pagf_freeblks -= args->len;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100602 ASSERT(be32_to_cpu(agf->agf_freeblks) <=
603 be32_to_cpu(agf->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS);
605 xfs_alloc_log_agf(args->tp, args->agbp,
606 XFS_AGF_FREEBLKS);
607 /* search the busylist for these blocks */
608 xfs_alloc_search_busy(args->tp, args->agno,
609 args->agbno, args->len);
610 }
611 if (!args->isfl)
612 xfs_trans_mod_sb(args->tp,
613 args->wasdel ? XFS_TRANS_SB_RES_FDBLOCKS :
614 XFS_TRANS_SB_FDBLOCKS, -slen);
615 XFS_STATS_INC(xs_allocx);
616 XFS_STATS_ADD(xs_allocb, args->len);
617 }
618 return 0;
619}
620
621/*
622 * Allocate a variable extent at exactly agno/bno.
623 * Extent's length (returned in *len) will be between minlen and maxlen,
624 * and of the form k * prod + mod unless there's nothing that large.
625 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
626 */
627STATIC int /* error */
628xfs_alloc_ag_vextent_exact(
629 xfs_alloc_arg_t *args) /* allocation argument structure */
630{
631 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
632 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
633 xfs_agblock_t end; /* end of allocated extent */
634 int error;
635 xfs_agblock_t fbno; /* start block of found extent */
636 xfs_agblock_t fend; /* end block of found extent */
637 xfs_extlen_t flen; /* length of found extent */
638#ifdef XFS_ALLOC_TRACE
639 static char fname[] = "xfs_alloc_ag_vextent_exact";
640#endif
641 int i; /* success/failure of operation */
642 xfs_agblock_t maxend; /* end of maximal extent */
643 xfs_agblock_t minend; /* end of minimal extent */
644 xfs_extlen_t rlen; /* length of returned extent */
645
646 ASSERT(args->alignment == 1);
647 /*
648 * Allocate/initialize a cursor for the by-number freespace btree.
649 */
650 bno_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
651 args->agno, XFS_BTNUM_BNO, NULL, 0);
652 /*
653 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
654 * Look for the closest free block <= bno, it must contain bno
655 * if any free block does.
656 */
657 if ((error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i)))
658 goto error0;
659 if (!i) {
660 /*
661 * Didn't find it, return null.
662 */
663 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
664 args->agbno = NULLAGBLOCK;
665 return 0;
666 }
667 /*
668 * Grab the freespace record.
669 */
670 if ((error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i)))
671 goto error0;
672 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
673 ASSERT(fbno <= args->agbno);
674 minend = args->agbno + args->minlen;
675 maxend = args->agbno + args->maxlen;
676 fend = fbno + flen;
677 /*
678 * Give up if the freespace isn't long enough for the minimum request.
679 */
680 if (fend < minend) {
681 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
682 args->agbno = NULLAGBLOCK;
683 return 0;
684 }
685 /*
686 * End of extent will be smaller of the freespace end and the
687 * maximal requested end.
688 */
689 end = XFS_AGBLOCK_MIN(fend, maxend);
690 /*
691 * Fix the length according to mod and prod if given.
692 */
693 args->len = end - args->agbno;
694 xfs_alloc_fix_len(args);
695 if (!xfs_alloc_fix_minleft(args)) {
696 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
697 return 0;
698 }
699 rlen = args->len;
700 ASSERT(args->agbno + rlen <= fend);
701 end = args->agbno + rlen;
702 /*
703 * We are allocating agbno for rlen [agbno .. end]
704 * Allocate/initialize a cursor for the by-size btree.
705 */
706 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
707 args->agno, XFS_BTNUM_CNT, NULL, 0);
708 ASSERT(args->agbno + args->len <=
Christoph Hellwig16259e72005-11-02 15:11:25 +1100709 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
711 args->agbno, args->len, XFSA_FIXUP_BNO_OK))) {
712 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
713 goto error0;
714 }
715 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
716 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
717 TRACE_ALLOC("normal", args);
718 args->wasfromfl = 0;
719 return 0;
720
721error0:
722 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
723 TRACE_ALLOC("error", args);
724 return error;
725}
726
727/*
728 * Allocate a variable extent near bno in the allocation group agno.
729 * Extent's length (returned in len) will be between minlen and maxlen,
730 * and of the form k * prod + mod unless there's nothing that large.
731 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
732 */
733STATIC int /* error */
734xfs_alloc_ag_vextent_near(
735 xfs_alloc_arg_t *args) /* allocation argument structure */
736{
737 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
738 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
739 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
740#ifdef XFS_ALLOC_TRACE
741 static char fname[] = "xfs_alloc_ag_vextent_near";
742#endif
743 xfs_agblock_t gtbno; /* start bno of right side entry */
744 xfs_agblock_t gtbnoa; /* aligned ... */
745 xfs_extlen_t gtdiff; /* difference to right side entry */
746 xfs_extlen_t gtlen; /* length of right side entry */
747 xfs_extlen_t gtlena; /* aligned ... */
748 xfs_agblock_t gtnew; /* useful start bno of right side */
749 int error; /* error code */
750 int i; /* result code, temporary */
751 int j; /* result code, temporary */
752 xfs_agblock_t ltbno; /* start bno of left side entry */
753 xfs_agblock_t ltbnoa; /* aligned ... */
754 xfs_extlen_t ltdiff; /* difference to left side entry */
755 /*REFERENCED*/
756 xfs_agblock_t ltend; /* end bno of left side entry */
757 xfs_extlen_t ltlen; /* length of left side entry */
758 xfs_extlen_t ltlena; /* aligned ... */
759 xfs_agblock_t ltnew; /* useful start bno of left side */
760 xfs_extlen_t rlen; /* length of returned extent */
761#if defined(DEBUG) && defined(__KERNEL__)
762 /*
763 * Randomly don't execute the first algorithm.
764 */
765 int dofirst; /* set to do first algorithm */
766
767 dofirst = random() & 1;
768#endif
769 /*
770 * Get a cursor for the by-size btree.
771 */
772 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
773 args->agno, XFS_BTNUM_CNT, NULL, 0);
774 ltlen = 0;
775 bno_cur_lt = bno_cur_gt = NULL;
776 /*
777 * See if there are any free extents as big as maxlen.
778 */
779 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
780 goto error0;
781 /*
782 * If none, then pick up the last entry in the tree unless the
783 * tree is empty.
784 */
785 if (!i) {
786 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &ltbno,
787 &ltlen, &i)))
788 goto error0;
789 if (i == 0 || ltlen == 0) {
790 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
791 return 0;
792 }
793 ASSERT(i == 1);
794 }
795 args->wasfromfl = 0;
796 /*
797 * First algorithm.
798 * If the requested extent is large wrt the freespaces available
799 * in this a.g., then the cursor will be pointing to a btree entry
800 * near the right edge of the tree. If it's in the last btree leaf
801 * block, then we just examine all the entries in that block
802 * that are big enough, and pick the best one.
803 * This is written as a while loop so we can break out of it,
804 * but we never loop back to the top.
805 */
806 while (xfs_btree_islastblock(cnt_cur, 0)) {
807 xfs_extlen_t bdiff;
808 int besti=0;
809 xfs_extlen_t blen=0;
810 xfs_agblock_t bnew=0;
811
812#if defined(DEBUG) && defined(__KERNEL__)
813 if (!dofirst)
814 break;
815#endif
816 /*
817 * Start from the entry that lookup found, sequence through
818 * all larger free blocks. If we're actually pointing at a
819 * record smaller than maxlen, go to the start of this block,
820 * and skip all those smaller than minlen.
821 */
822 if (ltlen || args->alignment > 1) {
823 cnt_cur->bc_ptrs[0] = 1;
824 do {
825 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno,
826 &ltlen, &i)))
827 goto error0;
828 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
829 if (ltlen >= args->minlen)
830 break;
831 if ((error = xfs_alloc_increment(cnt_cur, 0, &i)))
832 goto error0;
833 } while (i);
834 ASSERT(ltlen >= args->minlen);
835 if (!i)
836 break;
837 }
838 i = cnt_cur->bc_ptrs[0];
839 for (j = 1, blen = 0, bdiff = 0;
840 !error && j && (blen < args->maxlen || bdiff > 0);
841 error = xfs_alloc_increment(cnt_cur, 0, &j)) {
842 /*
843 * For each entry, decide if it's better than
844 * the previous best entry.
845 */
846 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno, &ltlen, &i)))
847 goto error0;
848 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
849 if (!xfs_alloc_compute_aligned(ltbno, ltlen,
850 args->alignment, args->minlen,
851 &ltbnoa, &ltlena))
852 continue;
853 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
854 xfs_alloc_fix_len(args);
855 ASSERT(args->len >= args->minlen);
856 if (args->len < blen)
857 continue;
858 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
859 args->alignment, ltbno, ltlen, &ltnew);
860 if (ltnew != NULLAGBLOCK &&
861 (args->len > blen || ltdiff < bdiff)) {
862 bdiff = ltdiff;
863 bnew = ltnew;
864 blen = args->len;
865 besti = cnt_cur->bc_ptrs[0];
866 }
867 }
868 /*
869 * It didn't work. We COULD be in a case where
870 * there's a good record somewhere, so try again.
871 */
872 if (blen == 0)
873 break;
874 /*
875 * Point at the best entry, and retrieve it again.
876 */
877 cnt_cur->bc_ptrs[0] = besti;
878 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno, &ltlen, &i)))
879 goto error0;
880 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
881 ltend = ltbno + ltlen;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100882 ASSERT(ltend <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 args->len = blen;
884 if (!xfs_alloc_fix_minleft(args)) {
885 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
886 TRACE_ALLOC("nominleft", args);
887 return 0;
888 }
889 blen = args->len;
890 /*
891 * We are allocating starting at bnew for blen blocks.
892 */
893 args->agbno = bnew;
894 ASSERT(bnew >= ltbno);
895 ASSERT(bnew + blen <= ltend);
896 /*
897 * Set up a cursor for the by-bno tree.
898 */
899 bno_cur_lt = xfs_btree_init_cursor(args->mp, args->tp,
900 args->agbp, args->agno, XFS_BTNUM_BNO, NULL, 0);
901 /*
902 * Fix up the btree entries.
903 */
904 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
905 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
906 goto error0;
907 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
908 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
909 TRACE_ALLOC("first", args);
910 return 0;
911 }
912 /*
913 * Second algorithm.
914 * Search in the by-bno tree to the left and to the right
915 * simultaneously, until in each case we find a space big enough,
916 * or run into the edge of the tree. When we run into the edge,
917 * we deallocate that cursor.
918 * If both searches succeed, we compare the two spaces and pick
919 * the better one.
920 * With alignment, it's possible for both to fail; the upper
921 * level algorithm that picks allocation groups for allocations
922 * is not supposed to do this.
923 */
924 /*
925 * Allocate and initialize the cursor for the leftward search.
926 */
927 bno_cur_lt = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
928 args->agno, XFS_BTNUM_BNO, NULL, 0);
929 /*
930 * Lookup <= bno to find the leftward search's starting point.
931 */
932 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
933 goto error0;
934 if (!i) {
935 /*
936 * Didn't find anything; use this cursor for the rightward
937 * search.
938 */
939 bno_cur_gt = bno_cur_lt;
940 bno_cur_lt = NULL;
941 }
942 /*
943 * Found something. Duplicate the cursor for the rightward search.
944 */
945 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
946 goto error0;
947 /*
948 * Increment the cursor, so we will point at the entry just right
949 * of the leftward entry if any, or to the leftmost entry.
950 */
951 if ((error = xfs_alloc_increment(bno_cur_gt, 0, &i)))
952 goto error0;
953 if (!i) {
954 /*
955 * It failed, there are no rightward entries.
956 */
957 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
958 bno_cur_gt = NULL;
959 }
960 /*
961 * Loop going left with the leftward cursor, right with the
962 * rightward cursor, until either both directions give up or
963 * we find an entry at least as big as minlen.
964 */
965 do {
966 if (bno_cur_lt) {
967 if ((error = xfs_alloc_get_rec(bno_cur_lt, &ltbno, &ltlen, &i)))
968 goto error0;
969 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
970 if (xfs_alloc_compute_aligned(ltbno, ltlen,
971 args->alignment, args->minlen,
972 &ltbnoa, &ltlena))
973 break;
974 if ((error = xfs_alloc_decrement(bno_cur_lt, 0, &i)))
975 goto error0;
976 if (!i) {
977 xfs_btree_del_cursor(bno_cur_lt,
978 XFS_BTREE_NOERROR);
979 bno_cur_lt = NULL;
980 }
981 }
982 if (bno_cur_gt) {
983 if ((error = xfs_alloc_get_rec(bno_cur_gt, &gtbno, &gtlen, &i)))
984 goto error0;
985 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
986 if (xfs_alloc_compute_aligned(gtbno, gtlen,
987 args->alignment, args->minlen,
988 &gtbnoa, &gtlena))
989 break;
990 if ((error = xfs_alloc_increment(bno_cur_gt, 0, &i)))
991 goto error0;
992 if (!i) {
993 xfs_btree_del_cursor(bno_cur_gt,
994 XFS_BTREE_NOERROR);
995 bno_cur_gt = NULL;
996 }
997 }
998 } while (bno_cur_lt || bno_cur_gt);
999 /*
1000 * Got both cursors still active, need to find better entry.
1001 */
1002 if (bno_cur_lt && bno_cur_gt) {
1003 /*
1004 * Left side is long enough, look for a right side entry.
1005 */
1006 if (ltlena >= args->minlen) {
1007 /*
1008 * Fix up the length.
1009 */
1010 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1011 xfs_alloc_fix_len(args);
1012 rlen = args->len;
1013 ltdiff = xfs_alloc_compute_diff(args->agbno, rlen,
1014 args->alignment, ltbno, ltlen, &ltnew);
1015 /*
1016 * Not perfect.
1017 */
1018 if (ltdiff) {
1019 /*
1020 * Look until we find a better one, run out of
1021 * space, or run off the end.
1022 */
1023 while (bno_cur_lt && bno_cur_gt) {
1024 if ((error = xfs_alloc_get_rec(
1025 bno_cur_gt, &gtbno,
1026 &gtlen, &i)))
1027 goto error0;
1028 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1029 xfs_alloc_compute_aligned(gtbno, gtlen,
1030 args->alignment, args->minlen,
1031 &gtbnoa, &gtlena);
1032 /*
1033 * The left one is clearly better.
1034 */
1035 if (gtbnoa >= args->agbno + ltdiff) {
1036 xfs_btree_del_cursor(
1037 bno_cur_gt,
1038 XFS_BTREE_NOERROR);
1039 bno_cur_gt = NULL;
1040 break;
1041 }
1042 /*
1043 * If we reach a big enough entry,
1044 * compare the two and pick the best.
1045 */
1046 if (gtlena >= args->minlen) {
1047 args->len =
1048 XFS_EXTLEN_MIN(gtlena,
1049 args->maxlen);
1050 xfs_alloc_fix_len(args);
1051 rlen = args->len;
1052 gtdiff = xfs_alloc_compute_diff(
1053 args->agbno, rlen,
1054 args->alignment,
1055 gtbno, gtlen, &gtnew);
1056 /*
1057 * Right side is better.
1058 */
1059 if (gtdiff < ltdiff) {
1060 xfs_btree_del_cursor(
1061 bno_cur_lt,
1062 XFS_BTREE_NOERROR);
1063 bno_cur_lt = NULL;
1064 }
1065 /*
1066 * Left side is better.
1067 */
1068 else {
1069 xfs_btree_del_cursor(
1070 bno_cur_gt,
1071 XFS_BTREE_NOERROR);
1072 bno_cur_gt = NULL;
1073 }
1074 break;
1075 }
1076 /*
1077 * Fell off the right end.
1078 */
1079 if ((error = xfs_alloc_increment(
1080 bno_cur_gt, 0, &i)))
1081 goto error0;
1082 if (!i) {
1083 xfs_btree_del_cursor(
1084 bno_cur_gt,
1085 XFS_BTREE_NOERROR);
1086 bno_cur_gt = NULL;
1087 break;
1088 }
1089 }
1090 }
1091 /*
1092 * The left side is perfect, trash the right side.
1093 */
1094 else {
1095 xfs_btree_del_cursor(bno_cur_gt,
1096 XFS_BTREE_NOERROR);
1097 bno_cur_gt = NULL;
1098 }
1099 }
1100 /*
1101 * It's the right side that was found first, look left.
1102 */
1103 else {
1104 /*
1105 * Fix up the length.
1106 */
1107 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1108 xfs_alloc_fix_len(args);
1109 rlen = args->len;
1110 gtdiff = xfs_alloc_compute_diff(args->agbno, rlen,
1111 args->alignment, gtbno, gtlen, &gtnew);
1112 /*
1113 * Right side entry isn't perfect.
1114 */
1115 if (gtdiff) {
1116 /*
1117 * Look until we find a better one, run out of
1118 * space, or run off the end.
1119 */
1120 while (bno_cur_lt && bno_cur_gt) {
1121 if ((error = xfs_alloc_get_rec(
1122 bno_cur_lt, &ltbno,
1123 &ltlen, &i)))
1124 goto error0;
1125 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1126 xfs_alloc_compute_aligned(ltbno, ltlen,
1127 args->alignment, args->minlen,
1128 &ltbnoa, &ltlena);
1129 /*
1130 * The right one is clearly better.
1131 */
1132 if (ltbnoa <= args->agbno - gtdiff) {
1133 xfs_btree_del_cursor(
1134 bno_cur_lt,
1135 XFS_BTREE_NOERROR);
1136 bno_cur_lt = NULL;
1137 break;
1138 }
1139 /*
1140 * If we reach a big enough entry,
1141 * compare the two and pick the best.
1142 */
1143 if (ltlena >= args->minlen) {
1144 args->len = XFS_EXTLEN_MIN(
1145 ltlena, args->maxlen);
1146 xfs_alloc_fix_len(args);
1147 rlen = args->len;
1148 ltdiff = xfs_alloc_compute_diff(
1149 args->agbno, rlen,
1150 args->alignment,
1151 ltbno, ltlen, &ltnew);
1152 /*
1153 * Left side is better.
1154 */
1155 if (ltdiff < gtdiff) {
1156 xfs_btree_del_cursor(
1157 bno_cur_gt,
1158 XFS_BTREE_NOERROR);
1159 bno_cur_gt = NULL;
1160 }
1161 /*
1162 * Right side is better.
1163 */
1164 else {
1165 xfs_btree_del_cursor(
1166 bno_cur_lt,
1167 XFS_BTREE_NOERROR);
1168 bno_cur_lt = NULL;
1169 }
1170 break;
1171 }
1172 /*
1173 * Fell off the left end.
1174 */
1175 if ((error = xfs_alloc_decrement(
1176 bno_cur_lt, 0, &i)))
1177 goto error0;
1178 if (!i) {
1179 xfs_btree_del_cursor(bno_cur_lt,
1180 XFS_BTREE_NOERROR);
1181 bno_cur_lt = NULL;
1182 break;
1183 }
1184 }
1185 }
1186 /*
1187 * The right side is perfect, trash the left side.
1188 */
1189 else {
1190 xfs_btree_del_cursor(bno_cur_lt,
1191 XFS_BTREE_NOERROR);
1192 bno_cur_lt = NULL;
1193 }
1194 }
1195 }
1196 /*
1197 * If we couldn't get anything, give up.
1198 */
1199 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1200 TRACE_ALLOC("neither", args);
1201 args->agbno = NULLAGBLOCK;
1202 return 0;
1203 }
1204 /*
1205 * At this point we have selected a freespace entry, either to the
1206 * left or to the right. If it's on the right, copy all the
1207 * useful variables to the "left" set so we only have one
1208 * copy of this code.
1209 */
1210 if (bno_cur_gt) {
1211 bno_cur_lt = bno_cur_gt;
1212 bno_cur_gt = NULL;
1213 ltbno = gtbno;
1214 ltbnoa = gtbnoa;
1215 ltlen = gtlen;
1216 ltlena = gtlena;
1217 j = 1;
1218 } else
1219 j = 0;
1220 /*
1221 * Fix up the length and compute the useful address.
1222 */
1223 ltend = ltbno + ltlen;
1224 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1225 xfs_alloc_fix_len(args);
1226 if (!xfs_alloc_fix_minleft(args)) {
1227 TRACE_ALLOC("nominleft", args);
1228 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1229 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1230 return 0;
1231 }
1232 rlen = args->len;
1233 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, ltbno,
1234 ltlen, &ltnew);
1235 ASSERT(ltnew >= ltbno);
1236 ASSERT(ltnew + rlen <= ltend);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001237 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 args->agbno = ltnew;
1239 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1240 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1241 goto error0;
1242 TRACE_ALLOC(j ? "gt" : "lt", args);
1243 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1244 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1245 return 0;
1246
1247 error0:
1248 TRACE_ALLOC("error", args);
1249 if (cnt_cur != NULL)
1250 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1251 if (bno_cur_lt != NULL)
1252 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1253 if (bno_cur_gt != NULL)
1254 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1255 return error;
1256}
1257
1258/*
1259 * Allocate a variable extent anywhere in the allocation group agno.
1260 * Extent's length (returned in len) will be between minlen and maxlen,
1261 * and of the form k * prod + mod unless there's nothing that large.
1262 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1263 */
1264STATIC int /* error */
1265xfs_alloc_ag_vextent_size(
1266 xfs_alloc_arg_t *args) /* allocation argument structure */
1267{
1268 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1269 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1270 int error; /* error result */
1271 xfs_agblock_t fbno; /* start of found freespace */
1272 xfs_extlen_t flen; /* length of found freespace */
1273#ifdef XFS_ALLOC_TRACE
1274 static char fname[] = "xfs_alloc_ag_vextent_size";
1275#endif
1276 int i; /* temp status variable */
1277 xfs_agblock_t rbno; /* returned block number */
1278 xfs_extlen_t rlen; /* length of returned extent */
1279
1280 /*
1281 * Allocate and initialize a cursor for the by-size btree.
1282 */
1283 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
1284 args->agno, XFS_BTNUM_CNT, NULL, 0);
1285 bno_cur = NULL;
1286 /*
1287 * Look for an entry >= maxlen+alignment-1 blocks.
1288 */
1289 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1290 args->maxlen + args->alignment - 1, &i)))
1291 goto error0;
1292 /*
1293 * If none, then pick up the last entry in the tree unless the
1294 * tree is empty.
1295 */
1296 if (!i) {
1297 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &fbno,
1298 &flen, &i)))
1299 goto error0;
1300 if (i == 0 || flen == 0) {
1301 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1302 TRACE_ALLOC("noentry", args);
1303 return 0;
1304 }
1305 ASSERT(i == 1);
1306 }
1307 /*
1308 * There's a freespace as big as maxlen+alignment-1, get it.
1309 */
1310 else {
1311 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i)))
1312 goto error0;
1313 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1314 }
1315 /*
1316 * In the first case above, we got the last entry in the
1317 * by-size btree. Now we check to see if the space hits maxlen
1318 * once aligned; if not, we search left for something better.
1319 * This can't happen in the second case above.
1320 */
1321 xfs_alloc_compute_aligned(fbno, flen, args->alignment, args->minlen,
1322 &rbno, &rlen);
1323 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1324 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1325 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1326 if (rlen < args->maxlen) {
1327 xfs_agblock_t bestfbno;
1328 xfs_extlen_t bestflen;
1329 xfs_agblock_t bestrbno;
1330 xfs_extlen_t bestrlen;
1331
1332 bestrlen = rlen;
1333 bestrbno = rbno;
1334 bestflen = flen;
1335 bestfbno = fbno;
1336 for (;;) {
1337 if ((error = xfs_alloc_decrement(cnt_cur, 0, &i)))
1338 goto error0;
1339 if (i == 0)
1340 break;
1341 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1342 &i)))
1343 goto error0;
1344 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1345 if (flen < bestrlen)
1346 break;
1347 xfs_alloc_compute_aligned(fbno, flen, args->alignment,
1348 args->minlen, &rbno, &rlen);
1349 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1350 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1351 (rlen <= flen && rbno + rlen <= fbno + flen),
1352 error0);
1353 if (rlen > bestrlen) {
1354 bestrlen = rlen;
1355 bestrbno = rbno;
1356 bestflen = flen;
1357 bestfbno = fbno;
1358 if (rlen == args->maxlen)
1359 break;
1360 }
1361 }
1362 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1363 &i)))
1364 goto error0;
1365 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1366 rlen = bestrlen;
1367 rbno = bestrbno;
1368 flen = bestflen;
1369 fbno = bestfbno;
1370 }
1371 args->wasfromfl = 0;
1372 /*
1373 * Fix up the length.
1374 */
1375 args->len = rlen;
1376 xfs_alloc_fix_len(args);
1377 if (rlen < args->minlen || !xfs_alloc_fix_minleft(args)) {
1378 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1379 TRACE_ALLOC("nominleft", args);
1380 args->agbno = NULLAGBLOCK;
1381 return 0;
1382 }
1383 rlen = args->len;
1384 XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1385 /*
1386 * Allocate and initialize a cursor for the by-block tree.
1387 */
1388 bno_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
1389 args->agno, XFS_BTNUM_BNO, NULL, 0);
1390 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1391 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1392 goto error0;
1393 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1394 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1395 cnt_cur = bno_cur = NULL;
1396 args->len = rlen;
1397 args->agbno = rbno;
1398 XFS_WANT_CORRUPTED_GOTO(
1399 args->agbno + args->len <=
Christoph Hellwig16259e72005-11-02 15:11:25 +11001400 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 error0);
1402 TRACE_ALLOC("normal", args);
1403 return 0;
1404
1405error0:
1406 TRACE_ALLOC("error", args);
1407 if (cnt_cur)
1408 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1409 if (bno_cur)
1410 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1411 return error;
1412}
1413
1414/*
1415 * Deal with the case where only small freespaces remain.
1416 * Either return the contents of the last freespace record,
1417 * or allocate space from the freelist if there is nothing in the tree.
1418 */
1419STATIC int /* error */
1420xfs_alloc_ag_vextent_small(
1421 xfs_alloc_arg_t *args, /* allocation argument structure */
1422 xfs_btree_cur_t *ccur, /* by-size cursor */
1423 xfs_agblock_t *fbnop, /* result block number */
1424 xfs_extlen_t *flenp, /* result length */
1425 int *stat) /* status: 0-freelist, 1-normal/none */
1426{
1427 int error;
1428 xfs_agblock_t fbno;
1429 xfs_extlen_t flen;
1430#ifdef XFS_ALLOC_TRACE
1431 static char fname[] = "xfs_alloc_ag_vextent_small";
1432#endif
1433 int i;
1434
1435 if ((error = xfs_alloc_decrement(ccur, 0, &i)))
1436 goto error0;
1437 if (i) {
1438 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1439 goto error0;
1440 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1441 }
1442 /*
1443 * Nothing in the btree, try the freelist. Make sure
1444 * to respect minleft even when pulling from the
1445 * freelist.
1446 */
1447 else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
Christoph Hellwig16259e72005-11-02 15:11:25 +11001448 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1449 > args->minleft)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 if ((error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno)))
1451 goto error0;
1452 if (fbno != NULLAGBLOCK) {
1453 if (args->userdata) {
1454 xfs_buf_t *bp;
1455
1456 bp = xfs_btree_get_bufs(args->mp, args->tp,
1457 args->agno, fbno, 0);
1458 xfs_trans_binval(args->tp, bp);
1459 }
1460 args->len = 1;
1461 args->agbno = fbno;
1462 XFS_WANT_CORRUPTED_GOTO(
1463 args->agbno + args->len <=
Christoph Hellwig16259e72005-11-02 15:11:25 +11001464 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 error0);
1466 args->wasfromfl = 1;
1467 TRACE_ALLOC("freelist", args);
1468 *stat = 0;
1469 return 0;
1470 }
1471 /*
1472 * Nothing in the freelist.
1473 */
1474 else
1475 flen = 0;
1476 }
1477 /*
1478 * Can't allocate from the freelist for some reason.
1479 */
1480 else
1481 flen = 0;
1482 /*
1483 * Can't do the allocation, give up.
1484 */
1485 if (flen < args->minlen) {
1486 args->agbno = NULLAGBLOCK;
1487 TRACE_ALLOC("notenough", args);
1488 flen = 0;
1489 }
1490 *fbnop = fbno;
1491 *flenp = flen;
1492 *stat = 1;
1493 TRACE_ALLOC("normal", args);
1494 return 0;
1495
1496error0:
1497 TRACE_ALLOC("error", args);
1498 return error;
1499}
1500
1501/*
1502 * Free the extent starting at agno/bno for length.
1503 */
1504STATIC int /* error */
1505xfs_free_ag_extent(
1506 xfs_trans_t *tp, /* transaction pointer */
1507 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1508 xfs_agnumber_t agno, /* allocation group number */
1509 xfs_agblock_t bno, /* starting block number */
1510 xfs_extlen_t len, /* length of extent */
1511 int isfl) /* set if is freelist blocks - no sb acctg */
1512{
1513 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1514 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1515 int error; /* error return value */
1516#ifdef XFS_ALLOC_TRACE
1517 static char fname[] = "xfs_free_ag_extent";
1518#endif
1519 xfs_agblock_t gtbno; /* start of right neighbor block */
1520 xfs_extlen_t gtlen; /* length of right neighbor block */
1521 int haveleft; /* have a left neighbor block */
1522 int haveright; /* have a right neighbor block */
1523 int i; /* temp, result code */
1524 xfs_agblock_t ltbno; /* start of left neighbor block */
1525 xfs_extlen_t ltlen; /* length of left neighbor block */
1526 xfs_mount_t *mp; /* mount point struct for filesystem */
1527 xfs_agblock_t nbno; /* new starting block of freespace */
1528 xfs_extlen_t nlen; /* new length of freespace */
1529
1530 mp = tp->t_mountp;
1531 /*
1532 * Allocate and initialize a cursor for the by-block btree.
1533 */
1534 bno_cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO, NULL,
1535 0);
1536 cnt_cur = NULL;
1537 /*
1538 * Look for a neighboring block on the left (lower block numbers)
1539 * that is contiguous with this space.
1540 */
1541 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1542 goto error0;
1543 if (haveleft) {
1544 /*
1545 * There is a block to our left.
1546 */
1547 if ((error = xfs_alloc_get_rec(bno_cur, &ltbno, &ltlen, &i)))
1548 goto error0;
1549 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1550 /*
1551 * It's not contiguous, though.
1552 */
1553 if (ltbno + ltlen < bno)
1554 haveleft = 0;
1555 else {
1556 /*
1557 * If this failure happens the request to free this
1558 * space was invalid, it's (partly) already free.
1559 * Very bad.
1560 */
1561 XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1562 }
1563 }
1564 /*
1565 * Look for a neighboring block on the right (higher block numbers)
1566 * that is contiguous with this space.
1567 */
1568 if ((error = xfs_alloc_increment(bno_cur, 0, &haveright)))
1569 goto error0;
1570 if (haveright) {
1571 /*
1572 * There is a block to our right.
1573 */
1574 if ((error = xfs_alloc_get_rec(bno_cur, &gtbno, &gtlen, &i)))
1575 goto error0;
1576 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1577 /*
1578 * It's not contiguous, though.
1579 */
1580 if (bno + len < gtbno)
1581 haveright = 0;
1582 else {
1583 /*
1584 * If this failure happens the request to free this
1585 * space was invalid, it's (partly) already free.
1586 * Very bad.
1587 */
1588 XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1589 }
1590 }
1591 /*
1592 * Now allocate and initialize a cursor for the by-size tree.
1593 */
1594 cnt_cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT, NULL,
1595 0);
1596 /*
1597 * Have both left and right contiguous neighbors.
1598 * Merge all three into a single free block.
1599 */
1600 if (haveleft && haveright) {
1601 /*
1602 * Delete the old by-size entry on the left.
1603 */
1604 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1605 goto error0;
1606 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1607 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1608 goto error0;
1609 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1610 /*
1611 * Delete the old by-size entry on the right.
1612 */
1613 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1614 goto error0;
1615 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1616 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1617 goto error0;
1618 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1619 /*
1620 * Delete the old by-block entry for the right block.
1621 */
1622 if ((error = xfs_alloc_delete(bno_cur, &i)))
1623 goto error0;
1624 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1625 /*
1626 * Move the by-block cursor back to the left neighbor.
1627 */
1628 if ((error = xfs_alloc_decrement(bno_cur, 0, &i)))
1629 goto error0;
1630 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1631#ifdef DEBUG
1632 /*
1633 * Check that this is the right record: delete didn't
1634 * mangle the cursor.
1635 */
1636 {
1637 xfs_agblock_t xxbno;
1638 xfs_extlen_t xxlen;
1639
1640 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1641 &i)))
1642 goto error0;
1643 XFS_WANT_CORRUPTED_GOTO(
1644 i == 1 && xxbno == ltbno && xxlen == ltlen,
1645 error0);
1646 }
1647#endif
1648 /*
1649 * Update remaining by-block entry to the new, joined block.
1650 */
1651 nbno = ltbno;
1652 nlen = len + ltlen + gtlen;
1653 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1654 goto error0;
1655 }
1656 /*
1657 * Have only a left contiguous neighbor.
1658 * Merge it together with the new freespace.
1659 */
1660 else if (haveleft) {
1661 /*
1662 * Delete the old by-size entry on the left.
1663 */
1664 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1665 goto error0;
1666 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1667 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1668 goto error0;
1669 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1670 /*
1671 * Back up the by-block cursor to the left neighbor, and
1672 * update its length.
1673 */
1674 if ((error = xfs_alloc_decrement(bno_cur, 0, &i)))
1675 goto error0;
1676 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1677 nbno = ltbno;
1678 nlen = len + ltlen;
1679 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1680 goto error0;
1681 }
1682 /*
1683 * Have only a right contiguous neighbor.
1684 * Merge it together with the new freespace.
1685 */
1686 else if (haveright) {
1687 /*
1688 * Delete the old by-size entry on the right.
1689 */
1690 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1691 goto error0;
1692 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1693 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1694 goto error0;
1695 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1696 /*
1697 * Update the starting block and length of the right
1698 * neighbor in the by-block tree.
1699 */
1700 nbno = bno;
1701 nlen = len + gtlen;
1702 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1703 goto error0;
1704 }
1705 /*
1706 * No contiguous neighbors.
1707 * Insert the new freespace into the by-block tree.
1708 */
1709 else {
1710 nbno = bno;
1711 nlen = len;
1712 if ((error = xfs_alloc_insert(bno_cur, &i)))
1713 goto error0;
1714 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1715 }
1716 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1717 bno_cur = NULL;
1718 /*
1719 * In all cases we need to insert the new freespace in the by-size tree.
1720 */
1721 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1722 goto error0;
1723 XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1724 if ((error = xfs_alloc_insert(cnt_cur, &i)))
1725 goto error0;
1726 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1727 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1728 cnt_cur = NULL;
1729 /*
1730 * Update the freespace totals in the ag and superblock.
1731 */
1732 {
1733 xfs_agf_t *agf;
1734 xfs_perag_t *pag; /* per allocation group data */
1735
1736 agf = XFS_BUF_TO_AGF(agbp);
1737 pag = &mp->m_perag[agno];
Christoph Hellwig16259e72005-11-02 15:11:25 +11001738 be32_add(&agf->agf_freeblks, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 xfs_trans_agblocks_delta(tp, len);
1740 pag->pagf_freeblks += len;
1741 XFS_WANT_CORRUPTED_GOTO(
Christoph Hellwig16259e72005-11-02 15:11:25 +11001742 be32_to_cpu(agf->agf_freeblks) <=
1743 be32_to_cpu(agf->agf_length),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 error0);
1745 TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS);
1746 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
1747 if (!isfl)
1748 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1749 XFS_STATS_INC(xs_freex);
1750 XFS_STATS_ADD(xs_freeb, len);
1751 }
1752 TRACE_FREE(haveleft ?
1753 (haveright ? "both" : "left") :
1754 (haveright ? "right" : "none"),
1755 agno, bno, len, isfl);
1756
1757 /*
1758 * Since blocks move to the free list without the coordination
1759 * used in xfs_bmap_finish, we can't allow block to be available
1760 * for reallocation and non-transaction writing (user data)
1761 * until we know that the transaction that moved it to the free
1762 * list is permanently on disk. We track the blocks by declaring
1763 * these blocks as "busy"; the busy list is maintained on a per-ag
1764 * basis and each transaction records which entries should be removed
1765 * when the iclog commits to disk. If a busy block is allocated,
1766 * the iclog is pushed up to the LSN that freed the block.
1767 */
1768 xfs_alloc_mark_busy(tp, agno, bno, len);
1769 return 0;
1770
1771 error0:
1772 TRACE_FREE("error", agno, bno, len, isfl);
1773 if (bno_cur)
1774 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1775 if (cnt_cur)
1776 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1777 return error;
1778}
1779
1780/*
1781 * Visible (exported) allocation/free functions.
1782 * Some of these are used just by xfs_alloc_btree.c and this file.
1783 */
1784
1785/*
1786 * Compute and fill in value of m_ag_maxlevels.
1787 */
1788void
1789xfs_alloc_compute_maxlevels(
1790 xfs_mount_t *mp) /* file system mount structure */
1791{
1792 int level;
1793 uint maxblocks;
1794 uint maxleafents;
1795 int minleafrecs;
1796 int minnoderecs;
1797
1798 maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1799 minleafrecs = mp->m_alloc_mnr[0];
1800 minnoderecs = mp->m_alloc_mnr[1];
1801 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1802 for (level = 1; maxblocks > 1; level++)
1803 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1804 mp->m_ag_maxlevels = level;
1805}
1806
1807/*
1808 * Decide whether to use this allocation group for this allocation.
1809 * If so, fix up the btree freelist's size.
1810 */
1811STATIC int /* error */
1812xfs_alloc_fix_freelist(
1813 xfs_alloc_arg_t *args, /* allocation argument structure */
1814 int flags) /* XFS_ALLOC_FLAG_... */
1815{
1816 xfs_buf_t *agbp; /* agf buffer pointer */
1817 xfs_agf_t *agf; /* a.g. freespace structure pointer */
1818 xfs_buf_t *agflbp;/* agfl buffer pointer */
1819 xfs_agblock_t bno; /* freelist block */
1820 xfs_extlen_t delta; /* new blocks needed in freelist */
1821 int error; /* error result code */
1822 xfs_extlen_t longest;/* longest extent in allocation group */
1823 xfs_mount_t *mp; /* file system mount point structure */
1824 xfs_extlen_t need; /* total blocks needed in freelist */
1825 xfs_perag_t *pag; /* per-ag information structure */
1826 xfs_alloc_arg_t targs; /* local allocation arguments */
1827 xfs_trans_t *tp; /* transaction pointer */
1828
1829 mp = args->mp;
1830
1831 pag = args->pag;
1832 tp = args->tp;
1833 if (!pag->pagf_init) {
1834 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1835 &agbp)))
1836 return error;
1837 if (!pag->pagf_init) {
1838 args->agbp = NULL;
1839 return 0;
1840 }
1841 } else
1842 agbp = NULL;
1843
Nathan Scottc41564b2006-03-29 08:55:14 +10001844 /* If this is a metadata preferred pag and we are user data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 * then try somewhere else if we are not being asked to
1846 * try harder at this point
1847 */
1848 if (pag->pagf_metadata && args->userdata && flags) {
1849 args->agbp = NULL;
1850 return 0;
1851 }
1852
1853 need = XFS_MIN_FREELIST_PAG(pag, mp);
1854 delta = need > pag->pagf_flcount ? need - pag->pagf_flcount : 0;
1855 /*
1856 * If it looks like there isn't a long enough extent, or enough
1857 * total blocks, reject it.
1858 */
1859 longest = (pag->pagf_longest > delta) ?
1860 (pag->pagf_longest - delta) :
1861 (pag->pagf_flcount > 0 || pag->pagf_longest > 0);
1862 if (args->minlen + args->alignment + args->minalignslop - 1 > longest ||
Yingping Lud210a282006-06-09 14:55:18 +10001863 (!(flags & XFS_ALLOC_FLAG_FREEING) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 (int)(pag->pagf_freeblks + pag->pagf_flcount -
1865 need - args->total) <
1866 (int)args->minleft)) {
1867 if (agbp)
1868 xfs_trans_brelse(tp, agbp);
1869 args->agbp = NULL;
1870 return 0;
1871 }
1872 /*
1873 * Get the a.g. freespace buffer.
1874 * Can fail if we're not blocking on locks, and it's held.
1875 */
1876 if (agbp == NULL) {
1877 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1878 &agbp)))
1879 return error;
1880 if (agbp == NULL) {
1881 args->agbp = NULL;
1882 return 0;
1883 }
1884 }
1885 /*
1886 * Figure out how many blocks we should have in the freelist.
1887 */
1888 agf = XFS_BUF_TO_AGF(agbp);
1889 need = XFS_MIN_FREELIST(agf, mp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001890 delta = need > be32_to_cpu(agf->agf_flcount) ?
1891 (need - be32_to_cpu(agf->agf_flcount)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 /*
1893 * If there isn't enough total or single-extent, reject it.
1894 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001895 longest = be32_to_cpu(agf->agf_longest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 longest = (longest > delta) ? (longest - delta) :
Christoph Hellwig16259e72005-11-02 15:11:25 +11001897 (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 if (args->minlen + args->alignment + args->minalignslop - 1 > longest ||
Yingping Lud210a282006-06-09 14:55:18 +10001899 (!(flags & XFS_ALLOC_FLAG_FREEING) &&
Christoph Hellwig16259e72005-11-02 15:11:25 +11001900 (int)(be32_to_cpu(agf->agf_freeblks) +
1901 be32_to_cpu(agf->agf_flcount) - need - args->total) <
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 (int)args->minleft)) {
1903 xfs_trans_brelse(tp, agbp);
1904 args->agbp = NULL;
1905 return 0;
1906 }
1907 /*
1908 * Make the freelist shorter if it's too long.
1909 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001910 while (be32_to_cpu(agf->agf_flcount) > need) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 xfs_buf_t *bp;
1912
1913 if ((error = xfs_alloc_get_freelist(tp, agbp, &bno)))
1914 return error;
1915 if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1916 return error;
1917 bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1918 xfs_trans_binval(tp, bp);
1919 }
1920 /*
1921 * Initialize the args structure.
1922 */
1923 targs.tp = tp;
1924 targs.mp = mp;
1925 targs.agbp = agbp;
1926 targs.agno = args->agno;
1927 targs.mod = targs.minleft = targs.wasdel = targs.userdata =
1928 targs.minalignslop = 0;
1929 targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1930 targs.type = XFS_ALLOCTYPE_THIS_AG;
1931 targs.pag = pag;
1932 if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1933 return error;
1934 /*
1935 * Make the freelist longer if it's too short.
1936 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001937 while (be32_to_cpu(agf->agf_flcount) < need) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 targs.agbno = 0;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001939 targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 /*
1941 * Allocate as many blocks as possible at once.
1942 */
Nathan Scotte63a3692006-05-08 19:51:58 +10001943 if ((error = xfs_alloc_ag_vextent(&targs))) {
1944 xfs_trans_brelse(tp, agflbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 return error;
Nathan Scotte63a3692006-05-08 19:51:58 +10001946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 /*
1948 * Stop if we run out. Won't happen if callers are obeying
1949 * the restrictions correctly. Can happen for free calls
1950 * on a completely full ag.
1951 */
Yingping Lud210a282006-06-09 14:55:18 +10001952 if (targs.agbno == NULLAGBLOCK) {
1953 if (!(flags & XFS_ALLOC_FLAG_FREEING)) {
1954 xfs_trans_brelse(tp, agflbp);
1955 args->agbp = NULL;
1956 return 0;
1957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 break;
Yingping Lud210a282006-06-09 14:55:18 +10001959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 /*
1961 * Put each allocated block on the list.
1962 */
1963 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
1964 if ((error = xfs_alloc_put_freelist(tp, agbp, agflbp,
1965 bno)))
1966 return error;
1967 }
1968 }
Nathan Scotte63a3692006-05-08 19:51:58 +10001969 xfs_trans_brelse(tp, agflbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 args->agbp = agbp;
1971 return 0;
1972}
1973
1974/*
1975 * Get a block from the freelist.
1976 * Returns with the buffer for the block gotten.
1977 */
1978int /* error */
1979xfs_alloc_get_freelist(
1980 xfs_trans_t *tp, /* transaction pointer */
1981 xfs_buf_t *agbp, /* buffer containing the agf structure */
1982 xfs_agblock_t *bnop) /* block address retrieved from freelist */
1983{
1984 xfs_agf_t *agf; /* a.g. freespace structure */
1985 xfs_agfl_t *agfl; /* a.g. freelist structure */
1986 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
1987 xfs_agblock_t bno; /* block number returned */
1988 int error;
1989#ifdef XFS_ALLOC_TRACE
1990 static char fname[] = "xfs_alloc_get_freelist";
1991#endif
1992 xfs_mount_t *mp; /* mount structure */
1993 xfs_perag_t *pag; /* per allocation group data */
1994
1995 agf = XFS_BUF_TO_AGF(agbp);
1996 /*
1997 * Freelist is empty, give up.
1998 */
1999 if (!agf->agf_flcount) {
2000 *bnop = NULLAGBLOCK;
2001 return 0;
2002 }
2003 /*
2004 * Read the array of free blocks.
2005 */
2006 mp = tp->t_mountp;
2007 if ((error = xfs_alloc_read_agfl(mp, tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +11002008 be32_to_cpu(agf->agf_seqno), &agflbp)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 return error;
2010 agfl = XFS_BUF_TO_AGFL(agflbp);
2011 /*
2012 * Get the block number and update the data structures.
2013 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002014 bno = INT_GET(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)], ARCH_CONVERT);
2015 be32_add(&agf->agf_flfirst, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 xfs_trans_brelse(tp, agflbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002017 if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 agf->agf_flfirst = 0;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002019 pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)];
2020 be32_add(&agf->agf_flcount, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 xfs_trans_agflist_delta(tp, -1);
2022 pag->pagf_flcount--;
2023 TRACE_MODAGF(NULL, agf, XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT);
2024 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT);
2025 *bnop = bno;
2026
2027 /*
2028 * As blocks are freed, they are added to the per-ag busy list
2029 * and remain there until the freeing transaction is committed to
2030 * disk. Now that we have allocated blocks, this list must be
2031 * searched to see if a block is being reused. If one is, then
2032 * the freeing transaction must be pushed to disk NOW by forcing
2033 * to disk all iclogs up that transaction's LSN.
2034 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002035 xfs_alloc_search_busy(tp, be32_to_cpu(agf->agf_seqno), bno, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 return 0;
2037}
2038
2039/*
2040 * Log the given fields from the agf structure.
2041 */
2042void
2043xfs_alloc_log_agf(
2044 xfs_trans_t *tp, /* transaction pointer */
2045 xfs_buf_t *bp, /* buffer for a.g. freelist header */
2046 int fields) /* mask of fields to be logged (XFS_AGF_...) */
2047{
2048 int first; /* first byte offset */
2049 int last; /* last byte offset */
2050 static const short offsets[] = {
2051 offsetof(xfs_agf_t, agf_magicnum),
2052 offsetof(xfs_agf_t, agf_versionnum),
2053 offsetof(xfs_agf_t, agf_seqno),
2054 offsetof(xfs_agf_t, agf_length),
2055 offsetof(xfs_agf_t, agf_roots[0]),
2056 offsetof(xfs_agf_t, agf_levels[0]),
2057 offsetof(xfs_agf_t, agf_flfirst),
2058 offsetof(xfs_agf_t, agf_fllast),
2059 offsetof(xfs_agf_t, agf_flcount),
2060 offsetof(xfs_agf_t, agf_freeblks),
2061 offsetof(xfs_agf_t, agf_longest),
2062 sizeof(xfs_agf_t)
2063 };
2064
2065 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2066 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2067}
2068
2069/*
2070 * Interface for inode allocation to force the pag data to be initialized.
2071 */
2072int /* error */
2073xfs_alloc_pagf_init(
2074 xfs_mount_t *mp, /* file system mount structure */
2075 xfs_trans_t *tp, /* transaction pointer */
2076 xfs_agnumber_t agno, /* allocation group number */
2077 int flags) /* XFS_ALLOC_FLAGS_... */
2078{
2079 xfs_buf_t *bp;
2080 int error;
2081
2082 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2083 return error;
2084 if (bp)
2085 xfs_trans_brelse(tp, bp);
2086 return 0;
2087}
2088
2089/*
2090 * Put the block on the freelist for the allocation group.
2091 */
2092int /* error */
2093xfs_alloc_put_freelist(
2094 xfs_trans_t *tp, /* transaction pointer */
2095 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2096 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2097 xfs_agblock_t bno) /* block being freed */
2098{
2099 xfs_agf_t *agf; /* a.g. freespace structure */
2100 xfs_agfl_t *agfl; /* a.g. free block array */
2101 xfs_agblock_t *blockp;/* pointer to array entry */
2102 int error;
2103#ifdef XFS_ALLOC_TRACE
2104 static char fname[] = "xfs_alloc_put_freelist";
2105#endif
2106 xfs_mount_t *mp; /* mount structure */
2107 xfs_perag_t *pag; /* per allocation group data */
2108
2109 agf = XFS_BUF_TO_AGF(agbp);
2110 mp = tp->t_mountp;
2111
2112 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +11002113 be32_to_cpu(agf->agf_seqno), &agflbp)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 return error;
2115 agfl = XFS_BUF_TO_AGFL(agflbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002116 be32_add(&agf->agf_fllast, 1);
2117 if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 agf->agf_fllast = 0;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002119 pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)];
2120 be32_add(&agf->agf_flcount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 xfs_trans_agflist_delta(tp, 1);
2122 pag->pagf_flcount++;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002123 ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2124 blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 INT_SET(*blockp, ARCH_CONVERT, bno);
2126 TRACE_MODAGF(NULL, agf, XFS_AGF_FLLAST | XFS_AGF_FLCOUNT);
2127 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLLAST | XFS_AGF_FLCOUNT);
2128 xfs_trans_log_buf(tp, agflbp,
2129 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl),
2130 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl +
2131 sizeof(xfs_agblock_t) - 1));
2132 return 0;
2133}
2134
2135/*
2136 * Read in the allocation group header (free/alloc section).
2137 */
2138int /* error */
2139xfs_alloc_read_agf(
2140 xfs_mount_t *mp, /* mount point structure */
2141 xfs_trans_t *tp, /* transaction pointer */
2142 xfs_agnumber_t agno, /* allocation group number */
2143 int flags, /* XFS_ALLOC_FLAG_... */
2144 xfs_buf_t **bpp) /* buffer for the ag freelist header */
2145{
2146 xfs_agf_t *agf; /* ag freelist header */
2147 int agf_ok; /* set if agf is consistent */
2148 xfs_buf_t *bp; /* return value */
2149 xfs_perag_t *pag; /* per allocation group data */
2150 int error;
2151
2152 ASSERT(agno != NULLAGNUMBER);
2153 error = xfs_trans_read_buf(
2154 mp, tp, mp->m_ddev_targp,
2155 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2156 XFS_FSS_TO_BB(mp, 1),
2157 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XFS_BUF_TRYLOCK : 0U,
2158 &bp);
2159 if (error)
2160 return error;
2161 ASSERT(!bp || !XFS_BUF_GETERROR(bp));
2162 if (!bp) {
2163 *bpp = NULL;
2164 return 0;
2165 }
2166 /*
2167 * Validate the magic number of the agf block.
2168 */
2169 agf = XFS_BUF_TO_AGF(bp);
2170 agf_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11002171 be32_to_cpu(agf->agf_magicnum) == XFS_AGF_MAGIC &&
2172 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2173 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2174 be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2175 be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2176 be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF,
2178 XFS_RANDOM_ALLOC_READ_AGF))) {
2179 XFS_CORRUPTION_ERROR("xfs_alloc_read_agf",
2180 XFS_ERRLEVEL_LOW, mp, agf);
2181 xfs_trans_brelse(tp, bp);
2182 return XFS_ERROR(EFSCORRUPTED);
2183 }
2184 pag = &mp->m_perag[agno];
2185 if (!pag->pagf_init) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11002186 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2187 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2188 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 pag->pagf_levels[XFS_BTNUM_BNOi] =
Christoph Hellwig16259e72005-11-02 15:11:25 +11002190 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 pag->pagf_levels[XFS_BTNUM_CNTi] =
Christoph Hellwig16259e72005-11-02 15:11:25 +11002192 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 spinlock_init(&pag->pagb_lock, "xfspagb");
2194 pag->pagb_list = kmem_zalloc(XFS_PAGB_NUM_SLOTS *
2195 sizeof(xfs_perag_busy_t), KM_SLEEP);
2196 pag->pagf_init = 1;
2197 }
2198#ifdef DEBUG
2199 else if (!XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11002200 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2201 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2202 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
Christoph Hellwig16259e72005-11-02 15:11:25 +11002204 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
Christoph Hellwig16259e72005-11-02 15:11:25 +11002206 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 }
2208#endif
2209 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGF, XFS_AGF_REF);
2210 *bpp = bp;
2211 return 0;
2212}
2213
2214/*
2215 * Allocate an extent (variable-size).
2216 * Depending on the allocation type, we either look in a single allocation
2217 * group or loop over the allocation groups to find the result.
2218 */
2219int /* error */
2220xfs_alloc_vextent(
2221 xfs_alloc_arg_t *args) /* allocation argument structure */
2222{
2223 xfs_agblock_t agsize; /* allocation group size */
2224 int error;
2225 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2226#ifdef XFS_ALLOC_TRACE
2227 static char fname[] = "xfs_alloc_vextent";
2228#endif
2229 xfs_extlen_t minleft;/* minimum left value, temp copy */
2230 xfs_mount_t *mp; /* mount structure pointer */
2231 xfs_agnumber_t sagno; /* starting allocation group number */
2232 xfs_alloctype_t type; /* input allocation type */
2233 int bump_rotor = 0;
2234 int no_min = 0;
2235 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2236
2237 mp = args->mp;
2238 type = args->otype = args->type;
2239 args->agbno = NULLAGBLOCK;
2240 /*
2241 * Just fix this up, for the case where the last a.g. is shorter
2242 * (or there's only one a.g.) and the caller couldn't easily figure
2243 * that out (xfs_bmap_alloc).
2244 */
2245 agsize = mp->m_sb.sb_agblocks;
2246 if (args->maxlen > agsize)
2247 args->maxlen = agsize;
2248 if (args->alignment == 0)
2249 args->alignment = 1;
2250 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2251 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2252 ASSERT(args->minlen <= args->maxlen);
2253 ASSERT(args->minlen <= agsize);
2254 ASSERT(args->mod < args->prod);
2255 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2256 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2257 args->minlen > args->maxlen || args->minlen > agsize ||
2258 args->mod >= args->prod) {
2259 args->fsbno = NULLFSBLOCK;
2260 TRACE_ALLOC("badargs", args);
2261 return 0;
2262 }
2263 minleft = args->minleft;
2264
2265 switch (type) {
2266 case XFS_ALLOCTYPE_THIS_AG:
2267 case XFS_ALLOCTYPE_NEAR_BNO:
2268 case XFS_ALLOCTYPE_THIS_BNO:
2269 /*
2270 * These three force us into a single a.g.
2271 */
2272 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2273 down_read(&mp->m_peraglock);
2274 args->pag = &mp->m_perag[args->agno];
2275 args->minleft = 0;
2276 error = xfs_alloc_fix_freelist(args, 0);
2277 args->minleft = minleft;
2278 if (error) {
2279 TRACE_ALLOC("nofix", args);
2280 goto error0;
2281 }
2282 if (!args->agbp) {
2283 up_read(&mp->m_peraglock);
2284 TRACE_ALLOC("noagbp", args);
2285 break;
2286 }
2287 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2288 if ((error = xfs_alloc_ag_vextent(args)))
2289 goto error0;
2290 up_read(&mp->m_peraglock);
2291 break;
2292 case XFS_ALLOCTYPE_START_BNO:
2293 /*
2294 * Try near allocation first, then anywhere-in-ag after
2295 * the first a.g. fails.
2296 */
2297 if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2298 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2299 args->fsbno = XFS_AGB_TO_FSB(mp,
2300 ((mp->m_agfrotor / rotorstep) %
2301 mp->m_sb.sb_agcount), 0);
2302 bump_rotor = 1;
2303 }
2304 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2305 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2306 /* FALLTHROUGH */
2307 case XFS_ALLOCTYPE_ANY_AG:
2308 case XFS_ALLOCTYPE_START_AG:
2309 case XFS_ALLOCTYPE_FIRST_AG:
2310 /*
2311 * Rotate through the allocation groups looking for a winner.
2312 */
2313 if (type == XFS_ALLOCTYPE_ANY_AG) {
2314 /*
2315 * Start with the last place we left off.
2316 */
2317 args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2318 mp->m_sb.sb_agcount;
2319 args->type = XFS_ALLOCTYPE_THIS_AG;
2320 flags = XFS_ALLOC_FLAG_TRYLOCK;
2321 } else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2322 /*
2323 * Start with allocation group given by bno.
2324 */
2325 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2326 args->type = XFS_ALLOCTYPE_THIS_AG;
2327 sagno = 0;
2328 flags = 0;
2329 } else {
2330 if (type == XFS_ALLOCTYPE_START_AG)
2331 args->type = XFS_ALLOCTYPE_THIS_AG;
2332 /*
2333 * Start with the given allocation group.
2334 */
2335 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2336 flags = XFS_ALLOC_FLAG_TRYLOCK;
2337 }
2338 /*
2339 * Loop over allocation groups twice; first time with
2340 * trylock set, second time without.
2341 */
2342 down_read(&mp->m_peraglock);
2343 for (;;) {
2344 args->pag = &mp->m_perag[args->agno];
2345 if (no_min) args->minleft = 0;
2346 error = xfs_alloc_fix_freelist(args, flags);
2347 args->minleft = minleft;
2348 if (error) {
2349 TRACE_ALLOC("nofix", args);
2350 goto error0;
2351 }
2352 /*
2353 * If we get a buffer back then the allocation will fly.
2354 */
2355 if (args->agbp) {
2356 if ((error = xfs_alloc_ag_vextent(args)))
2357 goto error0;
2358 break;
2359 }
2360 TRACE_ALLOC("loopfailed", args);
2361 /*
2362 * Didn't work, figure out the next iteration.
2363 */
2364 if (args->agno == sagno &&
2365 type == XFS_ALLOCTYPE_START_BNO)
2366 args->type = XFS_ALLOCTYPE_THIS_AG;
Yingping Lud210a282006-06-09 14:55:18 +10002367 /*
2368 * For the first allocation, we can try any AG to get
2369 * space. However, if we already have allocated a
2370 * block, we don't want to try AGs whose number is below
2371 * sagno. Otherwise, we may end up with out-of-order
2372 * locking of AGF, which might cause deadlock.
2373 */
2374 if (++(args->agno) == mp->m_sb.sb_agcount) {
2375 if (args->firstblock != NULLFSBLOCK)
2376 args->agno = sagno;
2377 else
2378 args->agno = 0;
2379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 /*
2381 * Reached the starting a.g., must either be done
2382 * or switch to non-trylock mode.
2383 */
2384 if (args->agno == sagno) {
2385 if (no_min == 1) {
2386 args->agbno = NULLAGBLOCK;
2387 TRACE_ALLOC("allfailed", args);
2388 break;
2389 }
2390 if (flags == 0) {
2391 no_min = 1;
2392 } else {
2393 flags = 0;
2394 if (type == XFS_ALLOCTYPE_START_BNO) {
2395 args->agbno = XFS_FSB_TO_AGBNO(mp,
2396 args->fsbno);
2397 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2398 }
2399 }
2400 }
2401 }
2402 up_read(&mp->m_peraglock);
2403 if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2404 if (args->agno == sagno)
2405 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2406 (mp->m_sb.sb_agcount * rotorstep);
2407 else
2408 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2409 (mp->m_sb.sb_agcount * rotorstep);
2410 }
2411 break;
2412 default:
2413 ASSERT(0);
2414 /* NOTREACHED */
2415 }
2416 if (args->agbno == NULLAGBLOCK)
2417 args->fsbno = NULLFSBLOCK;
2418 else {
2419 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2420#ifdef DEBUG
2421 ASSERT(args->len >= args->minlen);
2422 ASSERT(args->len <= args->maxlen);
2423 ASSERT(args->agbno % args->alignment == 0);
2424 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2425 args->len);
2426#endif
2427 }
2428 return 0;
2429error0:
2430 up_read(&mp->m_peraglock);
2431 return error;
2432}
2433
2434/*
2435 * Free an extent.
2436 * Just break up the extent address and hand off to xfs_free_ag_extent
2437 * after fixing up the freelist.
2438 */
2439int /* error */
2440xfs_free_extent(
2441 xfs_trans_t *tp, /* transaction pointer */
2442 xfs_fsblock_t bno, /* starting block number of extent */
2443 xfs_extlen_t len) /* length of extent */
2444{
2445#ifdef DEBUG
2446 xfs_agf_t *agf; /* a.g. freespace header */
2447#endif
2448 xfs_alloc_arg_t args; /* allocation argument structure */
2449 int error;
2450
2451 ASSERT(len != 0);
2452 args.tp = tp;
2453 args.mp = tp->t_mountp;
2454 args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2455 ASSERT(args.agno < args.mp->m_sb.sb_agcount);
2456 args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2457 args.alignment = 1;
2458 args.minlen = args.minleft = args.minalignslop = 0;
2459 down_read(&args.mp->m_peraglock);
2460 args.pag = &args.mp->m_perag[args.agno];
Yingping Lud210a282006-06-09 14:55:18 +10002461 if ((error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 goto error0;
2463#ifdef DEBUG
2464 ASSERT(args.agbp != NULL);
2465 agf = XFS_BUF_TO_AGF(args.agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002466 ASSERT(args.agbno + len <= be32_to_cpu(agf->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467#endif
2468 error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno,
2469 len, 0);
2470error0:
2471 up_read(&args.mp->m_peraglock);
2472 return error;
2473}
2474
2475
2476/*
2477 * AG Busy list management
2478 * The busy list contains block ranges that have been freed but whose
Nathan Scottc41564b2006-03-29 08:55:14 +10002479 * transactions have not yet hit disk. If any block listed in a busy
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 * list is reused, the transaction that freed it must be forced to disk
2481 * before continuing to use the block.
2482 *
2483 * xfs_alloc_mark_busy - add to the per-ag busy list
2484 * xfs_alloc_clear_busy - remove an item from the per-ag busy list
2485 */
2486void
2487xfs_alloc_mark_busy(xfs_trans_t *tp,
2488 xfs_agnumber_t agno,
2489 xfs_agblock_t bno,
2490 xfs_extlen_t len)
2491{
2492 xfs_mount_t *mp;
2493 xfs_perag_busy_t *bsy;
2494 int n;
2495 SPLDECL(s);
2496
2497 mp = tp->t_mountp;
2498 s = mutex_spinlock(&mp->m_perag[agno].pagb_lock);
2499
2500 /* search pagb_list for an open slot */
2501 for (bsy = mp->m_perag[agno].pagb_list, n = 0;
2502 n < XFS_PAGB_NUM_SLOTS;
2503 bsy++, n++) {
2504 if (bsy->busy_tp == NULL) {
2505 break;
2506 }
2507 }
2508
2509 if (n < XFS_PAGB_NUM_SLOTS) {
2510 bsy = &mp->m_perag[agno].pagb_list[n];
2511 mp->m_perag[agno].pagb_count++;
2512 TRACE_BUSY("xfs_alloc_mark_busy", "got", agno, bno, len, n, tp);
2513 bsy->busy_start = bno;
2514 bsy->busy_length = len;
2515 bsy->busy_tp = tp;
2516 xfs_trans_add_busy(tp, agno, n);
2517 } else {
2518 TRACE_BUSY("xfs_alloc_mark_busy", "FULL", agno, bno, len, -1, tp);
2519 /*
2520 * The busy list is full! Since it is now not possible to
2521 * track the free block, make this a synchronous transaction
2522 * to insure that the block is not reused before this
2523 * transaction commits.
2524 */
2525 xfs_trans_set_sync(tp);
2526 }
2527
2528 mutex_spinunlock(&mp->m_perag[agno].pagb_lock, s);
2529}
2530
2531void
2532xfs_alloc_clear_busy(xfs_trans_t *tp,
2533 xfs_agnumber_t agno,
2534 int idx)
2535{
2536 xfs_mount_t *mp;
2537 xfs_perag_busy_t *list;
2538 SPLDECL(s);
2539
2540 mp = tp->t_mountp;
2541
2542 s = mutex_spinlock(&mp->m_perag[agno].pagb_lock);
2543 list = mp->m_perag[agno].pagb_list;
2544
2545 ASSERT(idx < XFS_PAGB_NUM_SLOTS);
2546 if (list[idx].busy_tp == tp) {
2547 TRACE_UNBUSY("xfs_alloc_clear_busy", "found", agno, idx, tp);
2548 list[idx].busy_tp = NULL;
2549 mp->m_perag[agno].pagb_count--;
2550 } else {
2551 TRACE_UNBUSY("xfs_alloc_clear_busy", "missing", agno, idx, tp);
2552 }
2553
2554 mutex_spinunlock(&mp->m_perag[agno].pagb_lock, s);
2555}
2556
2557
2558/*
2559 * returns non-zero if any of (agno,bno):len is in a busy list
2560 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002561STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562xfs_alloc_search_busy(xfs_trans_t *tp,
2563 xfs_agnumber_t agno,
2564 xfs_agblock_t bno,
2565 xfs_extlen_t len)
2566{
2567 xfs_mount_t *mp;
2568 xfs_perag_busy_t *bsy;
2569 int n;
2570 xfs_agblock_t uend, bend;
2571 xfs_lsn_t lsn;
2572 int cnt;
2573 SPLDECL(s);
2574
2575 mp = tp->t_mountp;
2576
2577 s = mutex_spinlock(&mp->m_perag[agno].pagb_lock);
2578 cnt = mp->m_perag[agno].pagb_count;
2579
2580 uend = bno + len - 1;
2581
2582 /* search pagb_list for this slot, skipping open slots */
2583 for (bsy = mp->m_perag[agno].pagb_list, n = 0;
2584 cnt; bsy++, n++) {
2585
2586 /*
2587 * (start1,length1) within (start2, length2)
2588 */
2589 if (bsy->busy_tp != NULL) {
2590 bend = bsy->busy_start + bsy->busy_length - 1;
2591 if ((bno > bend) ||
2592 (uend < bsy->busy_start)) {
2593 cnt--;
2594 } else {
2595 TRACE_BUSYSEARCH("xfs_alloc_search_busy",
2596 "found1", agno, bno, len, n,
2597 tp);
2598 break;
2599 }
2600 }
2601 }
2602
2603 /*
2604 * If a block was found, force the log through the LSN of the
2605 * transaction that freed the block
2606 */
2607 if (cnt) {
2608 TRACE_BUSYSEARCH("xfs_alloc_search_busy", "found", agno, bno, len, n, tp);
2609 lsn = bsy->busy_tp->t_commit_lsn;
2610 mutex_spinunlock(&mp->m_perag[agno].pagb_lock, s);
2611 xfs_log_force(mp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC);
2612 } else {
2613 TRACE_BUSYSEARCH("xfs_alloc_search_busy", "not-found", agno, bno, len, n, tp);
2614 n = -1;
2615 mutex_spinunlock(&mp->m_perag[agno].pagb_lock, s);
2616 }
2617
2618 return n;
2619}