blob: f4328e1e2a744c6a590024707b2588f7307b2767 [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"
27#include "xfs_dir.h"
Nathan Scotta844f452005-11-02 14:38:42 +110028#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_dmapi.h"
30#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_dir_sf.h"
35#include "xfs_dir2_sf.h"
36#include "xfs_attr_sf.h"
37#include "xfs_dinode.h"
38#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_btree.h"
40#include "xfs_ialloc.h"
41#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_error.h"
43
44
45#define XFS_ABSDIFF(a,b) (((a) <= (b)) ? ((b) - (a)) : ((a) - (b)))
46
47#define XFSA_FIXUP_BNO_OK 1
48#define XFSA_FIXUP_CNT_OK 2
49
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100050STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -070051xfs_alloc_search_busy(xfs_trans_t *tp,
52 xfs_agnumber_t agno,
53 xfs_agblock_t bno,
54 xfs_extlen_t len);
55
56#if defined(XFS_ALLOC_TRACE)
57ktrace_t *xfs_alloc_trace_buf;
58
59#define TRACE_ALLOC(s,a) \
60 xfs_alloc_trace_alloc(fname, s, a, __LINE__)
61#define TRACE_FREE(s,a,b,x,f) \
62 xfs_alloc_trace_free(fname, s, mp, a, b, x, f, __LINE__)
63#define TRACE_MODAGF(s,a,f) \
64 xfs_alloc_trace_modagf(fname, s, mp, a, f, __LINE__)
65#define TRACE_BUSY(fname,s,ag,agb,l,sl,tp) \
66 xfs_alloc_trace_busy(fname, s, mp, ag, agb, l, sl, tp, XFS_ALLOC_KTRACE_BUSY, __LINE__)
67#define TRACE_UNBUSY(fname,s,ag,sl,tp) \
68 xfs_alloc_trace_busy(fname, s, mp, ag, -1, -1, sl, tp, XFS_ALLOC_KTRACE_UNBUSY, __LINE__)
69#define TRACE_BUSYSEARCH(fname,s,ag,agb,l,sl,tp) \
70 xfs_alloc_trace_busy(fname, s, mp, ag, agb, l, sl, tp, XFS_ALLOC_KTRACE_BUSYSEARCH, __LINE__)
71#else
72#define TRACE_ALLOC(s,a)
73#define TRACE_FREE(s,a,b,x,f)
74#define TRACE_MODAGF(s,a,f)
75#define TRACE_BUSY(s,a,ag,agb,l,sl,tp)
76#define TRACE_UNBUSY(fname,s,ag,sl,tp)
77#define TRACE_BUSYSEARCH(fname,s,ag,agb,l,sl,tp)
78#endif /* XFS_ALLOC_TRACE */
79
80/*
81 * Prototypes for per-ag allocation routines
82 */
83
84STATIC int xfs_alloc_ag_vextent_exact(xfs_alloc_arg_t *);
85STATIC int xfs_alloc_ag_vextent_near(xfs_alloc_arg_t *);
86STATIC int xfs_alloc_ag_vextent_size(xfs_alloc_arg_t *);
87STATIC int xfs_alloc_ag_vextent_small(xfs_alloc_arg_t *,
88 xfs_btree_cur_t *, xfs_agblock_t *, xfs_extlen_t *, int *);
89
90/*
91 * Internal functions.
92 */
93
94/*
95 * Compute aligned version of the found extent.
96 * Takes alignment and min length into account.
97 */
98STATIC int /* success (>= minlen) */
99xfs_alloc_compute_aligned(
100 xfs_agblock_t foundbno, /* starting block in found extent */
101 xfs_extlen_t foundlen, /* length in found extent */
102 xfs_extlen_t alignment, /* alignment for allocation */
103 xfs_extlen_t minlen, /* minimum length for allocation */
104 xfs_agblock_t *resbno, /* result block number */
105 xfs_extlen_t *reslen) /* result length */
106{
107 xfs_agblock_t bno;
108 xfs_extlen_t diff;
109 xfs_extlen_t len;
110
111 if (alignment > 1 && foundlen >= minlen) {
112 bno = roundup(foundbno, alignment);
113 diff = bno - foundbno;
114 len = diff >= foundlen ? 0 : foundlen - diff;
115 } else {
116 bno = foundbno;
117 len = foundlen;
118 }
119 *resbno = bno;
120 *reslen = len;
121 return len >= minlen;
122}
123
124/*
125 * Compute best start block and diff for "near" allocations.
126 * freelen >= wantlen already checked by caller.
127 */
128STATIC xfs_extlen_t /* difference value (absolute) */
129xfs_alloc_compute_diff(
130 xfs_agblock_t wantbno, /* target starting block */
131 xfs_extlen_t wantlen, /* target length */
132 xfs_extlen_t alignment, /* target alignment */
133 xfs_agblock_t freebno, /* freespace's starting block */
134 xfs_extlen_t freelen, /* freespace's length */
135 xfs_agblock_t *newbnop) /* result: best start block from free */
136{
137 xfs_agblock_t freeend; /* end of freespace extent */
138 xfs_agblock_t newbno1; /* return block number */
139 xfs_agblock_t newbno2; /* other new block number */
140 xfs_extlen_t newlen1=0; /* length with newbno1 */
141 xfs_extlen_t newlen2=0; /* length with newbno2 */
142 xfs_agblock_t wantend; /* end of target extent */
143
144 ASSERT(freelen >= wantlen);
145 freeend = freebno + freelen;
146 wantend = wantbno + wantlen;
147 if (freebno >= wantbno) {
148 if ((newbno1 = roundup(freebno, alignment)) >= freeend)
149 newbno1 = NULLAGBLOCK;
150 } else if (freeend >= wantend && alignment > 1) {
151 newbno1 = roundup(wantbno, alignment);
152 newbno2 = newbno1 - alignment;
153 if (newbno1 >= freeend)
154 newbno1 = NULLAGBLOCK;
155 else
156 newlen1 = XFS_EXTLEN_MIN(wantlen, freeend - newbno1);
157 if (newbno2 < freebno)
158 newbno2 = NULLAGBLOCK;
159 else
160 newlen2 = XFS_EXTLEN_MIN(wantlen, freeend - newbno2);
161 if (newbno1 != NULLAGBLOCK && newbno2 != NULLAGBLOCK) {
162 if (newlen1 < newlen2 ||
163 (newlen1 == newlen2 &&
164 XFS_ABSDIFF(newbno1, wantbno) >
165 XFS_ABSDIFF(newbno2, wantbno)))
166 newbno1 = newbno2;
167 } else if (newbno2 != NULLAGBLOCK)
168 newbno1 = newbno2;
169 } else if (freeend >= wantend) {
170 newbno1 = wantbno;
171 } else if (alignment > 1) {
172 newbno1 = roundup(freeend - wantlen, alignment);
173 if (newbno1 > freeend - wantlen &&
174 newbno1 - alignment >= freebno)
175 newbno1 -= alignment;
176 else if (newbno1 >= freeend)
177 newbno1 = NULLAGBLOCK;
178 } else
179 newbno1 = freeend - wantlen;
180 *newbnop = newbno1;
181 return newbno1 == NULLAGBLOCK ? 0 : XFS_ABSDIFF(newbno1, wantbno);
182}
183
184/*
185 * Fix up the length, based on mod and prod.
186 * len should be k * prod + mod for some k.
187 * If len is too small it is returned unchanged.
188 * If len hits maxlen it is left alone.
189 */
190STATIC void
191xfs_alloc_fix_len(
192 xfs_alloc_arg_t *args) /* allocation argument structure */
193{
194 xfs_extlen_t k;
195 xfs_extlen_t rlen;
196
197 ASSERT(args->mod < args->prod);
198 rlen = args->len;
199 ASSERT(rlen >= args->minlen);
200 ASSERT(rlen <= args->maxlen);
201 if (args->prod <= 1 || rlen < args->mod || rlen == args->maxlen ||
202 (args->mod == 0 && rlen < args->prod))
203 return;
204 k = rlen % args->prod;
205 if (k == args->mod)
206 return;
207 if (k > args->mod) {
208 if ((int)(rlen = rlen - k - args->mod) < (int)args->minlen)
209 return;
210 } else {
211 if ((int)(rlen = rlen - args->prod - (args->mod - k)) <
212 (int)args->minlen)
213 return;
214 }
215 ASSERT(rlen >= args->minlen);
216 ASSERT(rlen <= args->maxlen);
217 args->len = rlen;
218}
219
220/*
221 * Fix up length if there is too little space left in the a.g.
222 * Return 1 if ok, 0 if too little, should give up.
223 */
224STATIC int
225xfs_alloc_fix_minleft(
226 xfs_alloc_arg_t *args) /* allocation argument structure */
227{
228 xfs_agf_t *agf; /* a.g. freelist header */
229 int diff; /* free space difference */
230
231 if (args->minleft == 0)
232 return 1;
233 agf = XFS_BUF_TO_AGF(args->agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100234 diff = be32_to_cpu(agf->agf_freeblks)
235 + be32_to_cpu(agf->agf_flcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 - args->len - args->minleft;
237 if (diff >= 0)
238 return 1;
239 args->len += diff; /* shrink the allocated space */
240 if (args->len >= args->minlen)
241 return 1;
242 args->agbno = NULLAGBLOCK;
243 return 0;
244}
245
246/*
247 * Update the two btrees, logically removing from freespace the extent
248 * starting at rbno, rlen blocks. The extent is contained within the
249 * actual (current) free extent fbno for flen blocks.
250 * Flags are passed in indicating whether the cursors are set to the
251 * relevant records.
252 */
253STATIC int /* error code */
254xfs_alloc_fixup_trees(
255 xfs_btree_cur_t *cnt_cur, /* cursor for by-size btree */
256 xfs_btree_cur_t *bno_cur, /* cursor for by-block btree */
257 xfs_agblock_t fbno, /* starting block of free extent */
258 xfs_extlen_t flen, /* length of free extent */
259 xfs_agblock_t rbno, /* starting block of returned extent */
260 xfs_extlen_t rlen, /* length of returned extent */
261 int flags) /* flags, XFSA_FIXUP_... */
262{
263 int error; /* error code */
264 int i; /* operation results */
265 xfs_agblock_t nfbno1; /* first new free startblock */
266 xfs_agblock_t nfbno2; /* second new free startblock */
267 xfs_extlen_t nflen1=0; /* first new free length */
268 xfs_extlen_t nflen2=0; /* second new free length */
269
270 /*
271 * Look up the record in the by-size tree if necessary.
272 */
273 if (flags & XFSA_FIXUP_CNT_OK) {
274#ifdef DEBUG
275 if ((error = xfs_alloc_get_rec(cnt_cur, &nfbno1, &nflen1, &i)))
276 return error;
277 XFS_WANT_CORRUPTED_RETURN(
278 i == 1 && nfbno1 == fbno && nflen1 == flen);
279#endif
280 } else {
281 if ((error = xfs_alloc_lookup_eq(cnt_cur, fbno, flen, &i)))
282 return error;
283 XFS_WANT_CORRUPTED_RETURN(i == 1);
284 }
285 /*
286 * Look up the record in the by-block tree if necessary.
287 */
288 if (flags & XFSA_FIXUP_BNO_OK) {
289#ifdef DEBUG
290 if ((error = xfs_alloc_get_rec(bno_cur, &nfbno1, &nflen1, &i)))
291 return error;
292 XFS_WANT_CORRUPTED_RETURN(
293 i == 1 && nfbno1 == fbno && nflen1 == flen);
294#endif
295 } else {
296 if ((error = xfs_alloc_lookup_eq(bno_cur, fbno, flen, &i)))
297 return error;
298 XFS_WANT_CORRUPTED_RETURN(i == 1);
299 }
300#ifdef DEBUG
301 {
302 xfs_alloc_block_t *bnoblock;
303 xfs_alloc_block_t *cntblock;
304
305 if (bno_cur->bc_nlevels == 1 &&
306 cnt_cur->bc_nlevels == 1) {
307 bnoblock = XFS_BUF_TO_ALLOC_BLOCK(bno_cur->bc_bufs[0]);
308 cntblock = XFS_BUF_TO_ALLOC_BLOCK(cnt_cur->bc_bufs[0]);
309 XFS_WANT_CORRUPTED_RETURN(
Christoph Hellwig16259e72005-11-02 15:11:25 +1100310 be16_to_cpu(bnoblock->bb_numrecs) ==
311 be16_to_cpu(cntblock->bb_numrecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 }
313 }
314#endif
315 /*
316 * Deal with all four cases: the allocated record is contained
317 * within the freespace record, so we can have new freespace
318 * at either (or both) end, or no freespace remaining.
319 */
320 if (rbno == fbno && rlen == flen)
321 nfbno1 = nfbno2 = NULLAGBLOCK;
322 else if (rbno == fbno) {
323 nfbno1 = rbno + rlen;
324 nflen1 = flen - rlen;
325 nfbno2 = NULLAGBLOCK;
326 } else if (rbno + rlen == fbno + flen) {
327 nfbno1 = fbno;
328 nflen1 = flen - rlen;
329 nfbno2 = NULLAGBLOCK;
330 } else {
331 nfbno1 = fbno;
332 nflen1 = rbno - fbno;
333 nfbno2 = rbno + rlen;
334 nflen2 = (fbno + flen) - nfbno2;
335 }
336 /*
337 * Delete the entry from the by-size btree.
338 */
339 if ((error = xfs_alloc_delete(cnt_cur, &i)))
340 return error;
341 XFS_WANT_CORRUPTED_RETURN(i == 1);
342 /*
343 * Add new by-size btree entry(s).
344 */
345 if (nfbno1 != NULLAGBLOCK) {
346 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno1, nflen1, &i)))
347 return error;
348 XFS_WANT_CORRUPTED_RETURN(i == 0);
349 if ((error = xfs_alloc_insert(cnt_cur, &i)))
350 return error;
351 XFS_WANT_CORRUPTED_RETURN(i == 1);
352 }
353 if (nfbno2 != NULLAGBLOCK) {
354 if ((error = xfs_alloc_lookup_eq(cnt_cur, nfbno2, nflen2, &i)))
355 return error;
356 XFS_WANT_CORRUPTED_RETURN(i == 0);
357 if ((error = xfs_alloc_insert(cnt_cur, &i)))
358 return error;
359 XFS_WANT_CORRUPTED_RETURN(i == 1);
360 }
361 /*
362 * Fix up the by-block btree entry(s).
363 */
364 if (nfbno1 == NULLAGBLOCK) {
365 /*
366 * No remaining freespace, just delete the by-block tree entry.
367 */
368 if ((error = xfs_alloc_delete(bno_cur, &i)))
369 return error;
370 XFS_WANT_CORRUPTED_RETURN(i == 1);
371 } else {
372 /*
373 * Update the by-block entry to start later|be shorter.
374 */
375 if ((error = xfs_alloc_update(bno_cur, nfbno1, nflen1)))
376 return error;
377 }
378 if (nfbno2 != NULLAGBLOCK) {
379 /*
380 * 2 resulting free entries, need to add one.
381 */
382 if ((error = xfs_alloc_lookup_eq(bno_cur, nfbno2, nflen2, &i)))
383 return error;
384 XFS_WANT_CORRUPTED_RETURN(i == 0);
385 if ((error = xfs_alloc_insert(bno_cur, &i)))
386 return error;
387 XFS_WANT_CORRUPTED_RETURN(i == 1);
388 }
389 return 0;
390}
391
392/*
393 * Read in the allocation group free block array.
394 */
395STATIC int /* error */
396xfs_alloc_read_agfl(
397 xfs_mount_t *mp, /* mount point structure */
398 xfs_trans_t *tp, /* transaction pointer */
399 xfs_agnumber_t agno, /* allocation group number */
400 xfs_buf_t **bpp) /* buffer for the ag free block array */
401{
402 xfs_buf_t *bp; /* return value */
403 int error;
404
405 ASSERT(agno != NULLAGNUMBER);
406 error = xfs_trans_read_buf(
407 mp, tp, mp->m_ddev_targp,
408 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
409 XFS_FSS_TO_BB(mp, 1), 0, &bp);
410 if (error)
411 return error;
412 ASSERT(bp);
413 ASSERT(!XFS_BUF_GETERROR(bp));
414 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGFL, XFS_AGFL_REF);
415 *bpp = bp;
416 return 0;
417}
418
419#if defined(XFS_ALLOC_TRACE)
420/*
421 * Add an allocation trace entry for an alloc call.
422 */
423STATIC void
424xfs_alloc_trace_alloc(
425 char *name, /* function tag string */
426 char *str, /* additional string */
427 xfs_alloc_arg_t *args, /* allocation argument structure */
428 int line) /* source line number */
429{
430 ktrace_enter(xfs_alloc_trace_buf,
431 (void *)(__psint_t)(XFS_ALLOC_KTRACE_ALLOC | (line << 16)),
432 (void *)name,
433 (void *)str,
434 (void *)args->mp,
435 (void *)(__psunsigned_t)args->agno,
436 (void *)(__psunsigned_t)args->agbno,
437 (void *)(__psunsigned_t)args->minlen,
438 (void *)(__psunsigned_t)args->maxlen,
439 (void *)(__psunsigned_t)args->mod,
440 (void *)(__psunsigned_t)args->prod,
441 (void *)(__psunsigned_t)args->minleft,
442 (void *)(__psunsigned_t)args->total,
443 (void *)(__psunsigned_t)args->alignment,
444 (void *)(__psunsigned_t)args->len,
445 (void *)((((__psint_t)args->type) << 16) |
446 (__psint_t)args->otype),
447 (void *)(__psint_t)((args->wasdel << 3) |
448 (args->wasfromfl << 2) |
449 (args->isfl << 1) |
450 (args->userdata << 0)));
451}
452
453/*
454 * Add an allocation trace entry for a free call.
455 */
456STATIC void
457xfs_alloc_trace_free(
458 char *name, /* function tag string */
459 char *str, /* additional string */
460 xfs_mount_t *mp, /* file system mount point */
461 xfs_agnumber_t agno, /* allocation group number */
462 xfs_agblock_t agbno, /* a.g. relative block number */
463 xfs_extlen_t len, /* length of extent */
464 int isfl, /* set if is freelist allocation/free */
465 int line) /* source line number */
466{
467 ktrace_enter(xfs_alloc_trace_buf,
468 (void *)(__psint_t)(XFS_ALLOC_KTRACE_FREE | (line << 16)),
469 (void *)name,
470 (void *)str,
471 (void *)mp,
472 (void *)(__psunsigned_t)agno,
473 (void *)(__psunsigned_t)agbno,
474 (void *)(__psunsigned_t)len,
475 (void *)(__psint_t)isfl,
476 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL);
477}
478
479/*
480 * Add an allocation trace entry for modifying an agf.
481 */
482STATIC void
483xfs_alloc_trace_modagf(
484 char *name, /* function tag string */
485 char *str, /* additional string */
486 xfs_mount_t *mp, /* file system mount point */
487 xfs_agf_t *agf, /* new agf value */
488 int flags, /* logging flags for agf */
489 int line) /* source line number */
490{
491 ktrace_enter(xfs_alloc_trace_buf,
492 (void *)(__psint_t)(XFS_ALLOC_KTRACE_MODAGF | (line << 16)),
493 (void *)name,
494 (void *)str,
495 (void *)mp,
496 (void *)(__psint_t)flags,
Christoph Hellwig16259e72005-11-02 15:11:25 +1100497 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_seqno),
498 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_length),
499 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_BNO]),
500 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_roots[XFS_BTNUM_CNT]),
501 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNO]),
502 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNT]),
503 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flfirst),
504 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_fllast),
505 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_flcount),
506 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_freeblks),
507 (void *)(__psunsigned_t)be32_to_cpu(agf->agf_longest));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508}
509
510STATIC void
511xfs_alloc_trace_busy(
512 char *name, /* function tag string */
513 char *str, /* additional string */
514 xfs_mount_t *mp, /* file system mount poing */
515 xfs_agnumber_t agno, /* allocation group number */
516 xfs_agblock_t agbno, /* a.g. relative block number */
517 xfs_extlen_t len, /* length of extent */
518 int slot, /* perag Busy slot */
519 xfs_trans_t *tp,
520 int trtype, /* type: add, delete, search */
521 int line) /* source line number */
522{
523 ktrace_enter(xfs_alloc_trace_buf,
524 (void *)(__psint_t)(trtype | (line << 16)),
525 (void *)name,
526 (void *)str,
527 (void *)mp,
528 (void *)(__psunsigned_t)agno,
529 (void *)(__psunsigned_t)agbno,
530 (void *)(__psunsigned_t)len,
531 (void *)(__psint_t)slot,
532 (void *)tp,
533 NULL, NULL, NULL, NULL, NULL, NULL, NULL);
534}
535#endif /* XFS_ALLOC_TRACE */
536
537/*
538 * Allocation group level functions.
539 */
540
541/*
542 * Allocate a variable extent in the allocation group agno.
543 * Type and bno are used to determine where in the allocation group the
544 * extent will start.
545 * Extent's length (returned in *len) will be between minlen and maxlen,
546 * and of the form k * prod + mod unless there's nothing that large.
547 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
548 */
549STATIC int /* error */
550xfs_alloc_ag_vextent(
551 xfs_alloc_arg_t *args) /* argument structure for allocation */
552{
553 int error=0;
554#ifdef XFS_ALLOC_TRACE
555 static char fname[] = "xfs_alloc_ag_vextent";
556#endif
557
558 ASSERT(args->minlen > 0);
559 ASSERT(args->maxlen > 0);
560 ASSERT(args->minlen <= args->maxlen);
561 ASSERT(args->mod < args->prod);
562 ASSERT(args->alignment > 0);
563 /*
564 * Branch to correct routine based on the type.
565 */
566 args->wasfromfl = 0;
567 switch (args->type) {
568 case XFS_ALLOCTYPE_THIS_AG:
569 error = xfs_alloc_ag_vextent_size(args);
570 break;
571 case XFS_ALLOCTYPE_NEAR_BNO:
572 error = xfs_alloc_ag_vextent_near(args);
573 break;
574 case XFS_ALLOCTYPE_THIS_BNO:
575 error = xfs_alloc_ag_vextent_exact(args);
576 break;
577 default:
578 ASSERT(0);
579 /* NOTREACHED */
580 }
581 if (error)
582 return error;
583 /*
584 * If the allocation worked, need to change the agf structure
585 * (and log it), and the superblock.
586 */
587 if (args->agbno != NULLAGBLOCK) {
588 xfs_agf_t *agf; /* allocation group freelist header */
589#ifdef XFS_ALLOC_TRACE
590 xfs_mount_t *mp = args->mp;
591#endif
592 long slen = (long)args->len;
593
594 ASSERT(args->len >= args->minlen && args->len <= args->maxlen);
595 ASSERT(!(args->wasfromfl) || !args->isfl);
596 ASSERT(args->agbno % args->alignment == 0);
597 if (!(args->wasfromfl)) {
598
599 agf = XFS_BUF_TO_AGF(args->agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100600 be32_add(&agf->agf_freeblks, -(args->len));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 xfs_trans_agblocks_delta(args->tp,
602 -((long)(args->len)));
603 args->pag->pagf_freeblks -= args->len;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100604 ASSERT(be32_to_cpu(agf->agf_freeblks) <=
605 be32_to_cpu(agf->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS);
607 xfs_alloc_log_agf(args->tp, args->agbp,
608 XFS_AGF_FREEBLKS);
609 /* search the busylist for these blocks */
610 xfs_alloc_search_busy(args->tp, args->agno,
611 args->agbno, args->len);
612 }
613 if (!args->isfl)
614 xfs_trans_mod_sb(args->tp,
615 args->wasdel ? XFS_TRANS_SB_RES_FDBLOCKS :
616 XFS_TRANS_SB_FDBLOCKS, -slen);
617 XFS_STATS_INC(xs_allocx);
618 XFS_STATS_ADD(xs_allocb, args->len);
619 }
620 return 0;
621}
622
623/*
624 * Allocate a variable extent at exactly agno/bno.
625 * Extent's length (returned in *len) will be between minlen and maxlen,
626 * and of the form k * prod + mod unless there's nothing that large.
627 * Return the starting a.g. block (bno), or NULLAGBLOCK if we can't do it.
628 */
629STATIC int /* error */
630xfs_alloc_ag_vextent_exact(
631 xfs_alloc_arg_t *args) /* allocation argument structure */
632{
633 xfs_btree_cur_t *bno_cur;/* by block-number btree cursor */
634 xfs_btree_cur_t *cnt_cur;/* by count btree cursor */
635 xfs_agblock_t end; /* end of allocated extent */
636 int error;
637 xfs_agblock_t fbno; /* start block of found extent */
638 xfs_agblock_t fend; /* end block of found extent */
639 xfs_extlen_t flen; /* length of found extent */
640#ifdef XFS_ALLOC_TRACE
641 static char fname[] = "xfs_alloc_ag_vextent_exact";
642#endif
643 int i; /* success/failure of operation */
644 xfs_agblock_t maxend; /* end of maximal extent */
645 xfs_agblock_t minend; /* end of minimal extent */
646 xfs_extlen_t rlen; /* length of returned extent */
647
648 ASSERT(args->alignment == 1);
649 /*
650 * Allocate/initialize a cursor for the by-number freespace btree.
651 */
652 bno_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
653 args->agno, XFS_BTNUM_BNO, NULL, 0);
654 /*
655 * Lookup bno and minlen in the btree (minlen is irrelevant, really).
656 * Look for the closest free block <= bno, it must contain bno
657 * if any free block does.
658 */
659 if ((error = xfs_alloc_lookup_le(bno_cur, args->agbno, args->minlen, &i)))
660 goto error0;
661 if (!i) {
662 /*
663 * Didn't find it, return null.
664 */
665 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
666 args->agbno = NULLAGBLOCK;
667 return 0;
668 }
669 /*
670 * Grab the freespace record.
671 */
672 if ((error = xfs_alloc_get_rec(bno_cur, &fbno, &flen, &i)))
673 goto error0;
674 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
675 ASSERT(fbno <= args->agbno);
676 minend = args->agbno + args->minlen;
677 maxend = args->agbno + args->maxlen;
678 fend = fbno + flen;
679 /*
680 * Give up if the freespace isn't long enough for the minimum request.
681 */
682 if (fend < minend) {
683 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
684 args->agbno = NULLAGBLOCK;
685 return 0;
686 }
687 /*
688 * End of extent will be smaller of the freespace end and the
689 * maximal requested end.
690 */
691 end = XFS_AGBLOCK_MIN(fend, maxend);
692 /*
693 * Fix the length according to mod and prod if given.
694 */
695 args->len = end - args->agbno;
696 xfs_alloc_fix_len(args);
697 if (!xfs_alloc_fix_minleft(args)) {
698 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
699 return 0;
700 }
701 rlen = args->len;
702 ASSERT(args->agbno + rlen <= fend);
703 end = args->agbno + rlen;
704 /*
705 * We are allocating agbno for rlen [agbno .. end]
706 * Allocate/initialize a cursor for the by-size btree.
707 */
708 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
709 args->agno, XFS_BTNUM_CNT, NULL, 0);
710 ASSERT(args->agbno + args->len <=
Christoph Hellwig16259e72005-11-02 15:11:25 +1100711 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
713 args->agbno, args->len, XFSA_FIXUP_BNO_OK))) {
714 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
715 goto error0;
716 }
717 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
718 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
719 TRACE_ALLOC("normal", args);
720 args->wasfromfl = 0;
721 return 0;
722
723error0:
724 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
725 TRACE_ALLOC("error", args);
726 return error;
727}
728
729/*
730 * Allocate a variable extent near bno in the allocation group agno.
731 * Extent's length (returned in len) will be between minlen and maxlen,
732 * and of the form k * prod + mod unless there's nothing that large.
733 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
734 */
735STATIC int /* error */
736xfs_alloc_ag_vextent_near(
737 xfs_alloc_arg_t *args) /* allocation argument structure */
738{
739 xfs_btree_cur_t *bno_cur_gt; /* cursor for bno btree, right side */
740 xfs_btree_cur_t *bno_cur_lt; /* cursor for bno btree, left side */
741 xfs_btree_cur_t *cnt_cur; /* cursor for count btree */
742#ifdef XFS_ALLOC_TRACE
743 static char fname[] = "xfs_alloc_ag_vextent_near";
744#endif
745 xfs_agblock_t gtbno; /* start bno of right side entry */
746 xfs_agblock_t gtbnoa; /* aligned ... */
747 xfs_extlen_t gtdiff; /* difference to right side entry */
748 xfs_extlen_t gtlen; /* length of right side entry */
749 xfs_extlen_t gtlena; /* aligned ... */
750 xfs_agblock_t gtnew; /* useful start bno of right side */
751 int error; /* error code */
752 int i; /* result code, temporary */
753 int j; /* result code, temporary */
754 xfs_agblock_t ltbno; /* start bno of left side entry */
755 xfs_agblock_t ltbnoa; /* aligned ... */
756 xfs_extlen_t ltdiff; /* difference to left side entry */
757 /*REFERENCED*/
758 xfs_agblock_t ltend; /* end bno of left side entry */
759 xfs_extlen_t ltlen; /* length of left side entry */
760 xfs_extlen_t ltlena; /* aligned ... */
761 xfs_agblock_t ltnew; /* useful start bno of left side */
762 xfs_extlen_t rlen; /* length of returned extent */
763#if defined(DEBUG) && defined(__KERNEL__)
764 /*
765 * Randomly don't execute the first algorithm.
766 */
767 int dofirst; /* set to do first algorithm */
768
769 dofirst = random() & 1;
770#endif
771 /*
772 * Get a cursor for the by-size btree.
773 */
774 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
775 args->agno, XFS_BTNUM_CNT, NULL, 0);
776 ltlen = 0;
777 bno_cur_lt = bno_cur_gt = NULL;
778 /*
779 * See if there are any free extents as big as maxlen.
780 */
781 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0, args->maxlen, &i)))
782 goto error0;
783 /*
784 * If none, then pick up the last entry in the tree unless the
785 * tree is empty.
786 */
787 if (!i) {
788 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &ltbno,
789 &ltlen, &i)))
790 goto error0;
791 if (i == 0 || ltlen == 0) {
792 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
793 return 0;
794 }
795 ASSERT(i == 1);
796 }
797 args->wasfromfl = 0;
798 /*
799 * First algorithm.
800 * If the requested extent is large wrt the freespaces available
801 * in this a.g., then the cursor will be pointing to a btree entry
802 * near the right edge of the tree. If it's in the last btree leaf
803 * block, then we just examine all the entries in that block
804 * that are big enough, and pick the best one.
805 * This is written as a while loop so we can break out of it,
806 * but we never loop back to the top.
807 */
808 while (xfs_btree_islastblock(cnt_cur, 0)) {
809 xfs_extlen_t bdiff;
810 int besti=0;
811 xfs_extlen_t blen=0;
812 xfs_agblock_t bnew=0;
813
814#if defined(DEBUG) && defined(__KERNEL__)
815 if (!dofirst)
816 break;
817#endif
818 /*
819 * Start from the entry that lookup found, sequence through
820 * all larger free blocks. If we're actually pointing at a
821 * record smaller than maxlen, go to the start of this block,
822 * and skip all those smaller than minlen.
823 */
824 if (ltlen || args->alignment > 1) {
825 cnt_cur->bc_ptrs[0] = 1;
826 do {
827 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno,
828 &ltlen, &i)))
829 goto error0;
830 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
831 if (ltlen >= args->minlen)
832 break;
833 if ((error = xfs_alloc_increment(cnt_cur, 0, &i)))
834 goto error0;
835 } while (i);
836 ASSERT(ltlen >= args->minlen);
837 if (!i)
838 break;
839 }
840 i = cnt_cur->bc_ptrs[0];
841 for (j = 1, blen = 0, bdiff = 0;
842 !error && j && (blen < args->maxlen || bdiff > 0);
843 error = xfs_alloc_increment(cnt_cur, 0, &j)) {
844 /*
845 * For each entry, decide if it's better than
846 * the previous best entry.
847 */
848 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno, &ltlen, &i)))
849 goto error0;
850 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
851 if (!xfs_alloc_compute_aligned(ltbno, ltlen,
852 args->alignment, args->minlen,
853 &ltbnoa, &ltlena))
854 continue;
855 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
856 xfs_alloc_fix_len(args);
857 ASSERT(args->len >= args->minlen);
858 if (args->len < blen)
859 continue;
860 ltdiff = xfs_alloc_compute_diff(args->agbno, args->len,
861 args->alignment, ltbno, ltlen, &ltnew);
862 if (ltnew != NULLAGBLOCK &&
863 (args->len > blen || ltdiff < bdiff)) {
864 bdiff = ltdiff;
865 bnew = ltnew;
866 blen = args->len;
867 besti = cnt_cur->bc_ptrs[0];
868 }
869 }
870 /*
871 * It didn't work. We COULD be in a case where
872 * there's a good record somewhere, so try again.
873 */
874 if (blen == 0)
875 break;
876 /*
877 * Point at the best entry, and retrieve it again.
878 */
879 cnt_cur->bc_ptrs[0] = besti;
880 if ((error = xfs_alloc_get_rec(cnt_cur, &ltbno, &ltlen, &i)))
881 goto error0;
882 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
883 ltend = ltbno + ltlen;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100884 ASSERT(ltend <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 args->len = blen;
886 if (!xfs_alloc_fix_minleft(args)) {
887 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
888 TRACE_ALLOC("nominleft", args);
889 return 0;
890 }
891 blen = args->len;
892 /*
893 * We are allocating starting at bnew for blen blocks.
894 */
895 args->agbno = bnew;
896 ASSERT(bnew >= ltbno);
897 ASSERT(bnew + blen <= ltend);
898 /*
899 * Set up a cursor for the by-bno tree.
900 */
901 bno_cur_lt = xfs_btree_init_cursor(args->mp, args->tp,
902 args->agbp, args->agno, XFS_BTNUM_BNO, NULL, 0);
903 /*
904 * Fix up the btree entries.
905 */
906 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno,
907 ltlen, bnew, blen, XFSA_FIXUP_CNT_OK)))
908 goto error0;
909 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
910 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
911 TRACE_ALLOC("first", args);
912 return 0;
913 }
914 /*
915 * Second algorithm.
916 * Search in the by-bno tree to the left and to the right
917 * simultaneously, until in each case we find a space big enough,
918 * or run into the edge of the tree. When we run into the edge,
919 * we deallocate that cursor.
920 * If both searches succeed, we compare the two spaces and pick
921 * the better one.
922 * With alignment, it's possible for both to fail; the upper
923 * level algorithm that picks allocation groups for allocations
924 * is not supposed to do this.
925 */
926 /*
927 * Allocate and initialize the cursor for the leftward search.
928 */
929 bno_cur_lt = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
930 args->agno, XFS_BTNUM_BNO, NULL, 0);
931 /*
932 * Lookup <= bno to find the leftward search's starting point.
933 */
934 if ((error = xfs_alloc_lookup_le(bno_cur_lt, args->agbno, args->maxlen, &i)))
935 goto error0;
936 if (!i) {
937 /*
938 * Didn't find anything; use this cursor for the rightward
939 * search.
940 */
941 bno_cur_gt = bno_cur_lt;
942 bno_cur_lt = NULL;
943 }
944 /*
945 * Found something. Duplicate the cursor for the rightward search.
946 */
947 else if ((error = xfs_btree_dup_cursor(bno_cur_lt, &bno_cur_gt)))
948 goto error0;
949 /*
950 * Increment the cursor, so we will point at the entry just right
951 * of the leftward entry if any, or to the leftmost entry.
952 */
953 if ((error = xfs_alloc_increment(bno_cur_gt, 0, &i)))
954 goto error0;
955 if (!i) {
956 /*
957 * It failed, there are no rightward entries.
958 */
959 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_NOERROR);
960 bno_cur_gt = NULL;
961 }
962 /*
963 * Loop going left with the leftward cursor, right with the
964 * rightward cursor, until either both directions give up or
965 * we find an entry at least as big as minlen.
966 */
967 do {
968 if (bno_cur_lt) {
969 if ((error = xfs_alloc_get_rec(bno_cur_lt, &ltbno, &ltlen, &i)))
970 goto error0;
971 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
972 if (xfs_alloc_compute_aligned(ltbno, ltlen,
973 args->alignment, args->minlen,
974 &ltbnoa, &ltlena))
975 break;
976 if ((error = xfs_alloc_decrement(bno_cur_lt, 0, &i)))
977 goto error0;
978 if (!i) {
979 xfs_btree_del_cursor(bno_cur_lt,
980 XFS_BTREE_NOERROR);
981 bno_cur_lt = NULL;
982 }
983 }
984 if (bno_cur_gt) {
985 if ((error = xfs_alloc_get_rec(bno_cur_gt, &gtbno, &gtlen, &i)))
986 goto error0;
987 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
988 if (xfs_alloc_compute_aligned(gtbno, gtlen,
989 args->alignment, args->minlen,
990 &gtbnoa, &gtlena))
991 break;
992 if ((error = xfs_alloc_increment(bno_cur_gt, 0, &i)))
993 goto error0;
994 if (!i) {
995 xfs_btree_del_cursor(bno_cur_gt,
996 XFS_BTREE_NOERROR);
997 bno_cur_gt = NULL;
998 }
999 }
1000 } while (bno_cur_lt || bno_cur_gt);
1001 /*
1002 * Got both cursors still active, need to find better entry.
1003 */
1004 if (bno_cur_lt && bno_cur_gt) {
1005 /*
1006 * Left side is long enough, look for a right side entry.
1007 */
1008 if (ltlena >= args->minlen) {
1009 /*
1010 * Fix up the length.
1011 */
1012 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1013 xfs_alloc_fix_len(args);
1014 rlen = args->len;
1015 ltdiff = xfs_alloc_compute_diff(args->agbno, rlen,
1016 args->alignment, ltbno, ltlen, &ltnew);
1017 /*
1018 * Not perfect.
1019 */
1020 if (ltdiff) {
1021 /*
1022 * Look until we find a better one, run out of
1023 * space, or run off the end.
1024 */
1025 while (bno_cur_lt && bno_cur_gt) {
1026 if ((error = xfs_alloc_get_rec(
1027 bno_cur_gt, &gtbno,
1028 &gtlen, &i)))
1029 goto error0;
1030 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1031 xfs_alloc_compute_aligned(gtbno, gtlen,
1032 args->alignment, args->minlen,
1033 &gtbnoa, &gtlena);
1034 /*
1035 * The left one is clearly better.
1036 */
1037 if (gtbnoa >= args->agbno + ltdiff) {
1038 xfs_btree_del_cursor(
1039 bno_cur_gt,
1040 XFS_BTREE_NOERROR);
1041 bno_cur_gt = NULL;
1042 break;
1043 }
1044 /*
1045 * If we reach a big enough entry,
1046 * compare the two and pick the best.
1047 */
1048 if (gtlena >= args->minlen) {
1049 args->len =
1050 XFS_EXTLEN_MIN(gtlena,
1051 args->maxlen);
1052 xfs_alloc_fix_len(args);
1053 rlen = args->len;
1054 gtdiff = xfs_alloc_compute_diff(
1055 args->agbno, rlen,
1056 args->alignment,
1057 gtbno, gtlen, &gtnew);
1058 /*
1059 * Right side is better.
1060 */
1061 if (gtdiff < ltdiff) {
1062 xfs_btree_del_cursor(
1063 bno_cur_lt,
1064 XFS_BTREE_NOERROR);
1065 bno_cur_lt = NULL;
1066 }
1067 /*
1068 * Left side is better.
1069 */
1070 else {
1071 xfs_btree_del_cursor(
1072 bno_cur_gt,
1073 XFS_BTREE_NOERROR);
1074 bno_cur_gt = NULL;
1075 }
1076 break;
1077 }
1078 /*
1079 * Fell off the right end.
1080 */
1081 if ((error = xfs_alloc_increment(
1082 bno_cur_gt, 0, &i)))
1083 goto error0;
1084 if (!i) {
1085 xfs_btree_del_cursor(
1086 bno_cur_gt,
1087 XFS_BTREE_NOERROR);
1088 bno_cur_gt = NULL;
1089 break;
1090 }
1091 }
1092 }
1093 /*
1094 * The left side is perfect, trash the right side.
1095 */
1096 else {
1097 xfs_btree_del_cursor(bno_cur_gt,
1098 XFS_BTREE_NOERROR);
1099 bno_cur_gt = NULL;
1100 }
1101 }
1102 /*
1103 * It's the right side that was found first, look left.
1104 */
1105 else {
1106 /*
1107 * Fix up the length.
1108 */
1109 args->len = XFS_EXTLEN_MIN(gtlena, args->maxlen);
1110 xfs_alloc_fix_len(args);
1111 rlen = args->len;
1112 gtdiff = xfs_alloc_compute_diff(args->agbno, rlen,
1113 args->alignment, gtbno, gtlen, &gtnew);
1114 /*
1115 * Right side entry isn't perfect.
1116 */
1117 if (gtdiff) {
1118 /*
1119 * Look until we find a better one, run out of
1120 * space, or run off the end.
1121 */
1122 while (bno_cur_lt && bno_cur_gt) {
1123 if ((error = xfs_alloc_get_rec(
1124 bno_cur_lt, &ltbno,
1125 &ltlen, &i)))
1126 goto error0;
1127 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1128 xfs_alloc_compute_aligned(ltbno, ltlen,
1129 args->alignment, args->minlen,
1130 &ltbnoa, &ltlena);
1131 /*
1132 * The right one is clearly better.
1133 */
1134 if (ltbnoa <= args->agbno - gtdiff) {
1135 xfs_btree_del_cursor(
1136 bno_cur_lt,
1137 XFS_BTREE_NOERROR);
1138 bno_cur_lt = NULL;
1139 break;
1140 }
1141 /*
1142 * If we reach a big enough entry,
1143 * compare the two and pick the best.
1144 */
1145 if (ltlena >= args->minlen) {
1146 args->len = XFS_EXTLEN_MIN(
1147 ltlena, args->maxlen);
1148 xfs_alloc_fix_len(args);
1149 rlen = args->len;
1150 ltdiff = xfs_alloc_compute_diff(
1151 args->agbno, rlen,
1152 args->alignment,
1153 ltbno, ltlen, &ltnew);
1154 /*
1155 * Left side is better.
1156 */
1157 if (ltdiff < gtdiff) {
1158 xfs_btree_del_cursor(
1159 bno_cur_gt,
1160 XFS_BTREE_NOERROR);
1161 bno_cur_gt = NULL;
1162 }
1163 /*
1164 * Right side is better.
1165 */
1166 else {
1167 xfs_btree_del_cursor(
1168 bno_cur_lt,
1169 XFS_BTREE_NOERROR);
1170 bno_cur_lt = NULL;
1171 }
1172 break;
1173 }
1174 /*
1175 * Fell off the left end.
1176 */
1177 if ((error = xfs_alloc_decrement(
1178 bno_cur_lt, 0, &i)))
1179 goto error0;
1180 if (!i) {
1181 xfs_btree_del_cursor(bno_cur_lt,
1182 XFS_BTREE_NOERROR);
1183 bno_cur_lt = NULL;
1184 break;
1185 }
1186 }
1187 }
1188 /*
1189 * The right side is perfect, trash the left side.
1190 */
1191 else {
1192 xfs_btree_del_cursor(bno_cur_lt,
1193 XFS_BTREE_NOERROR);
1194 bno_cur_lt = NULL;
1195 }
1196 }
1197 }
1198 /*
1199 * If we couldn't get anything, give up.
1200 */
1201 if (bno_cur_lt == NULL && bno_cur_gt == NULL) {
1202 TRACE_ALLOC("neither", args);
1203 args->agbno = NULLAGBLOCK;
1204 return 0;
1205 }
1206 /*
1207 * At this point we have selected a freespace entry, either to the
1208 * left or to the right. If it's on the right, copy all the
1209 * useful variables to the "left" set so we only have one
1210 * copy of this code.
1211 */
1212 if (bno_cur_gt) {
1213 bno_cur_lt = bno_cur_gt;
1214 bno_cur_gt = NULL;
1215 ltbno = gtbno;
1216 ltbnoa = gtbnoa;
1217 ltlen = gtlen;
1218 ltlena = gtlena;
1219 j = 1;
1220 } else
1221 j = 0;
1222 /*
1223 * Fix up the length and compute the useful address.
1224 */
1225 ltend = ltbno + ltlen;
1226 args->len = XFS_EXTLEN_MIN(ltlena, args->maxlen);
1227 xfs_alloc_fix_len(args);
1228 if (!xfs_alloc_fix_minleft(args)) {
1229 TRACE_ALLOC("nominleft", args);
1230 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1231 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1232 return 0;
1233 }
1234 rlen = args->len;
1235 (void)xfs_alloc_compute_diff(args->agbno, rlen, args->alignment, ltbno,
1236 ltlen, &ltnew);
1237 ASSERT(ltnew >= ltbno);
1238 ASSERT(ltnew + rlen <= ltend);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001239 ASSERT(ltnew + rlen <= be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 args->agbno = ltnew;
1241 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur_lt, ltbno, ltlen,
1242 ltnew, rlen, XFSA_FIXUP_BNO_OK)))
1243 goto error0;
1244 TRACE_ALLOC(j ? "gt" : "lt", args);
1245 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1246 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_NOERROR);
1247 return 0;
1248
1249 error0:
1250 TRACE_ALLOC("error", args);
1251 if (cnt_cur != NULL)
1252 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1253 if (bno_cur_lt != NULL)
1254 xfs_btree_del_cursor(bno_cur_lt, XFS_BTREE_ERROR);
1255 if (bno_cur_gt != NULL)
1256 xfs_btree_del_cursor(bno_cur_gt, XFS_BTREE_ERROR);
1257 return error;
1258}
1259
1260/*
1261 * Allocate a variable extent anywhere in the allocation group agno.
1262 * Extent's length (returned in len) will be between minlen and maxlen,
1263 * and of the form k * prod + mod unless there's nothing that large.
1264 * Return the starting a.g. block, or NULLAGBLOCK if we can't do it.
1265 */
1266STATIC int /* error */
1267xfs_alloc_ag_vextent_size(
1268 xfs_alloc_arg_t *args) /* allocation argument structure */
1269{
1270 xfs_btree_cur_t *bno_cur; /* cursor for bno btree */
1271 xfs_btree_cur_t *cnt_cur; /* cursor for cnt btree */
1272 int error; /* error result */
1273 xfs_agblock_t fbno; /* start of found freespace */
1274 xfs_extlen_t flen; /* length of found freespace */
1275#ifdef XFS_ALLOC_TRACE
1276 static char fname[] = "xfs_alloc_ag_vextent_size";
1277#endif
1278 int i; /* temp status variable */
1279 xfs_agblock_t rbno; /* returned block number */
1280 xfs_extlen_t rlen; /* length of returned extent */
1281
1282 /*
1283 * Allocate and initialize a cursor for the by-size btree.
1284 */
1285 cnt_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
1286 args->agno, XFS_BTNUM_CNT, NULL, 0);
1287 bno_cur = NULL;
1288 /*
1289 * Look for an entry >= maxlen+alignment-1 blocks.
1290 */
1291 if ((error = xfs_alloc_lookup_ge(cnt_cur, 0,
1292 args->maxlen + args->alignment - 1, &i)))
1293 goto error0;
1294 /*
1295 * If none, then pick up the last entry in the tree unless the
1296 * tree is empty.
1297 */
1298 if (!i) {
1299 if ((error = xfs_alloc_ag_vextent_small(args, cnt_cur, &fbno,
1300 &flen, &i)))
1301 goto error0;
1302 if (i == 0 || flen == 0) {
1303 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1304 TRACE_ALLOC("noentry", args);
1305 return 0;
1306 }
1307 ASSERT(i == 1);
1308 }
1309 /*
1310 * There's a freespace as big as maxlen+alignment-1, get it.
1311 */
1312 else {
1313 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen, &i)))
1314 goto error0;
1315 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1316 }
1317 /*
1318 * In the first case above, we got the last entry in the
1319 * by-size btree. Now we check to see if the space hits maxlen
1320 * once aligned; if not, we search left for something better.
1321 * This can't happen in the second case above.
1322 */
1323 xfs_alloc_compute_aligned(fbno, flen, args->alignment, args->minlen,
1324 &rbno, &rlen);
1325 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1326 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1327 (rlen <= flen && rbno + rlen <= fbno + flen), error0);
1328 if (rlen < args->maxlen) {
1329 xfs_agblock_t bestfbno;
1330 xfs_extlen_t bestflen;
1331 xfs_agblock_t bestrbno;
1332 xfs_extlen_t bestrlen;
1333
1334 bestrlen = rlen;
1335 bestrbno = rbno;
1336 bestflen = flen;
1337 bestfbno = fbno;
1338 for (;;) {
1339 if ((error = xfs_alloc_decrement(cnt_cur, 0, &i)))
1340 goto error0;
1341 if (i == 0)
1342 break;
1343 if ((error = xfs_alloc_get_rec(cnt_cur, &fbno, &flen,
1344 &i)))
1345 goto error0;
1346 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1347 if (flen < bestrlen)
1348 break;
1349 xfs_alloc_compute_aligned(fbno, flen, args->alignment,
1350 args->minlen, &rbno, &rlen);
1351 rlen = XFS_EXTLEN_MIN(args->maxlen, rlen);
1352 XFS_WANT_CORRUPTED_GOTO(rlen == 0 ||
1353 (rlen <= flen && rbno + rlen <= fbno + flen),
1354 error0);
1355 if (rlen > bestrlen) {
1356 bestrlen = rlen;
1357 bestrbno = rbno;
1358 bestflen = flen;
1359 bestfbno = fbno;
1360 if (rlen == args->maxlen)
1361 break;
1362 }
1363 }
1364 if ((error = xfs_alloc_lookup_eq(cnt_cur, bestfbno, bestflen,
1365 &i)))
1366 goto error0;
1367 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1368 rlen = bestrlen;
1369 rbno = bestrbno;
1370 flen = bestflen;
1371 fbno = bestfbno;
1372 }
1373 args->wasfromfl = 0;
1374 /*
1375 * Fix up the length.
1376 */
1377 args->len = rlen;
1378 xfs_alloc_fix_len(args);
1379 if (rlen < args->minlen || !xfs_alloc_fix_minleft(args)) {
1380 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1381 TRACE_ALLOC("nominleft", args);
1382 args->agbno = NULLAGBLOCK;
1383 return 0;
1384 }
1385 rlen = args->len;
1386 XFS_WANT_CORRUPTED_GOTO(rlen <= flen, error0);
1387 /*
1388 * Allocate and initialize a cursor for the by-block tree.
1389 */
1390 bno_cur = xfs_btree_init_cursor(args->mp, args->tp, args->agbp,
1391 args->agno, XFS_BTNUM_BNO, NULL, 0);
1392 if ((error = xfs_alloc_fixup_trees(cnt_cur, bno_cur, fbno, flen,
1393 rbno, rlen, XFSA_FIXUP_CNT_OK)))
1394 goto error0;
1395 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1396 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1397 cnt_cur = bno_cur = NULL;
1398 args->len = rlen;
1399 args->agbno = rbno;
1400 XFS_WANT_CORRUPTED_GOTO(
1401 args->agbno + args->len <=
Christoph Hellwig16259e72005-11-02 15:11:25 +11001402 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 error0);
1404 TRACE_ALLOC("normal", args);
1405 return 0;
1406
1407error0:
1408 TRACE_ALLOC("error", args);
1409 if (cnt_cur)
1410 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1411 if (bno_cur)
1412 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1413 return error;
1414}
1415
1416/*
1417 * Deal with the case where only small freespaces remain.
1418 * Either return the contents of the last freespace record,
1419 * or allocate space from the freelist if there is nothing in the tree.
1420 */
1421STATIC int /* error */
1422xfs_alloc_ag_vextent_small(
1423 xfs_alloc_arg_t *args, /* allocation argument structure */
1424 xfs_btree_cur_t *ccur, /* by-size cursor */
1425 xfs_agblock_t *fbnop, /* result block number */
1426 xfs_extlen_t *flenp, /* result length */
1427 int *stat) /* status: 0-freelist, 1-normal/none */
1428{
1429 int error;
1430 xfs_agblock_t fbno;
1431 xfs_extlen_t flen;
1432#ifdef XFS_ALLOC_TRACE
1433 static char fname[] = "xfs_alloc_ag_vextent_small";
1434#endif
1435 int i;
1436
1437 if ((error = xfs_alloc_decrement(ccur, 0, &i)))
1438 goto error0;
1439 if (i) {
1440 if ((error = xfs_alloc_get_rec(ccur, &fbno, &flen, &i)))
1441 goto error0;
1442 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1443 }
1444 /*
1445 * Nothing in the btree, try the freelist. Make sure
1446 * to respect minleft even when pulling from the
1447 * freelist.
1448 */
1449 else if (args->minlen == 1 && args->alignment == 1 && !args->isfl &&
Christoph Hellwig16259e72005-11-02 15:11:25 +11001450 (be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_flcount)
1451 > args->minleft)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 if ((error = xfs_alloc_get_freelist(args->tp, args->agbp, &fbno)))
1453 goto error0;
1454 if (fbno != NULLAGBLOCK) {
1455 if (args->userdata) {
1456 xfs_buf_t *bp;
1457
1458 bp = xfs_btree_get_bufs(args->mp, args->tp,
1459 args->agno, fbno, 0);
1460 xfs_trans_binval(args->tp, bp);
1461 }
1462 args->len = 1;
1463 args->agbno = fbno;
1464 XFS_WANT_CORRUPTED_GOTO(
1465 args->agbno + args->len <=
Christoph Hellwig16259e72005-11-02 15:11:25 +11001466 be32_to_cpu(XFS_BUF_TO_AGF(args->agbp)->agf_length),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 error0);
1468 args->wasfromfl = 1;
1469 TRACE_ALLOC("freelist", args);
1470 *stat = 0;
1471 return 0;
1472 }
1473 /*
1474 * Nothing in the freelist.
1475 */
1476 else
1477 flen = 0;
1478 }
1479 /*
1480 * Can't allocate from the freelist for some reason.
1481 */
1482 else
1483 flen = 0;
1484 /*
1485 * Can't do the allocation, give up.
1486 */
1487 if (flen < args->minlen) {
1488 args->agbno = NULLAGBLOCK;
1489 TRACE_ALLOC("notenough", args);
1490 flen = 0;
1491 }
1492 *fbnop = fbno;
1493 *flenp = flen;
1494 *stat = 1;
1495 TRACE_ALLOC("normal", args);
1496 return 0;
1497
1498error0:
1499 TRACE_ALLOC("error", args);
1500 return error;
1501}
1502
1503/*
1504 * Free the extent starting at agno/bno for length.
1505 */
1506STATIC int /* error */
1507xfs_free_ag_extent(
1508 xfs_trans_t *tp, /* transaction pointer */
1509 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
1510 xfs_agnumber_t agno, /* allocation group number */
1511 xfs_agblock_t bno, /* starting block number */
1512 xfs_extlen_t len, /* length of extent */
1513 int isfl) /* set if is freelist blocks - no sb acctg */
1514{
1515 xfs_btree_cur_t *bno_cur; /* cursor for by-block btree */
1516 xfs_btree_cur_t *cnt_cur; /* cursor for by-size btree */
1517 int error; /* error return value */
1518#ifdef XFS_ALLOC_TRACE
1519 static char fname[] = "xfs_free_ag_extent";
1520#endif
1521 xfs_agblock_t gtbno; /* start of right neighbor block */
1522 xfs_extlen_t gtlen; /* length of right neighbor block */
1523 int haveleft; /* have a left neighbor block */
1524 int haveright; /* have a right neighbor block */
1525 int i; /* temp, result code */
1526 xfs_agblock_t ltbno; /* start of left neighbor block */
1527 xfs_extlen_t ltlen; /* length of left neighbor block */
1528 xfs_mount_t *mp; /* mount point struct for filesystem */
1529 xfs_agblock_t nbno; /* new starting block of freespace */
1530 xfs_extlen_t nlen; /* new length of freespace */
1531
1532 mp = tp->t_mountp;
1533 /*
1534 * Allocate and initialize a cursor for the by-block btree.
1535 */
1536 bno_cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_BNO, NULL,
1537 0);
1538 cnt_cur = NULL;
1539 /*
1540 * Look for a neighboring block on the left (lower block numbers)
1541 * that is contiguous with this space.
1542 */
1543 if ((error = xfs_alloc_lookup_le(bno_cur, bno, len, &haveleft)))
1544 goto error0;
1545 if (haveleft) {
1546 /*
1547 * There is a block to our left.
1548 */
1549 if ((error = xfs_alloc_get_rec(bno_cur, &ltbno, &ltlen, &i)))
1550 goto error0;
1551 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1552 /*
1553 * It's not contiguous, though.
1554 */
1555 if (ltbno + ltlen < bno)
1556 haveleft = 0;
1557 else {
1558 /*
1559 * If this failure happens the request to free this
1560 * space was invalid, it's (partly) already free.
1561 * Very bad.
1562 */
1563 XFS_WANT_CORRUPTED_GOTO(ltbno + ltlen <= bno, error0);
1564 }
1565 }
1566 /*
1567 * Look for a neighboring block on the right (higher block numbers)
1568 * that is contiguous with this space.
1569 */
1570 if ((error = xfs_alloc_increment(bno_cur, 0, &haveright)))
1571 goto error0;
1572 if (haveright) {
1573 /*
1574 * There is a block to our right.
1575 */
1576 if ((error = xfs_alloc_get_rec(bno_cur, &gtbno, &gtlen, &i)))
1577 goto error0;
1578 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1579 /*
1580 * It's not contiguous, though.
1581 */
1582 if (bno + len < gtbno)
1583 haveright = 0;
1584 else {
1585 /*
1586 * If this failure happens the request to free this
1587 * space was invalid, it's (partly) already free.
1588 * Very bad.
1589 */
1590 XFS_WANT_CORRUPTED_GOTO(gtbno >= bno + len, error0);
1591 }
1592 }
1593 /*
1594 * Now allocate and initialize a cursor for the by-size tree.
1595 */
1596 cnt_cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_CNT, NULL,
1597 0);
1598 /*
1599 * Have both left and right contiguous neighbors.
1600 * Merge all three into a single free block.
1601 */
1602 if (haveleft && haveright) {
1603 /*
1604 * Delete the old by-size entry on the left.
1605 */
1606 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1607 goto error0;
1608 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1609 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1610 goto error0;
1611 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1612 /*
1613 * Delete the old by-size entry on the right.
1614 */
1615 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1616 goto error0;
1617 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1618 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1619 goto error0;
1620 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1621 /*
1622 * Delete the old by-block entry for the right block.
1623 */
1624 if ((error = xfs_alloc_delete(bno_cur, &i)))
1625 goto error0;
1626 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1627 /*
1628 * Move the by-block cursor back to the left neighbor.
1629 */
1630 if ((error = xfs_alloc_decrement(bno_cur, 0, &i)))
1631 goto error0;
1632 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1633#ifdef DEBUG
1634 /*
1635 * Check that this is the right record: delete didn't
1636 * mangle the cursor.
1637 */
1638 {
1639 xfs_agblock_t xxbno;
1640 xfs_extlen_t xxlen;
1641
1642 if ((error = xfs_alloc_get_rec(bno_cur, &xxbno, &xxlen,
1643 &i)))
1644 goto error0;
1645 XFS_WANT_CORRUPTED_GOTO(
1646 i == 1 && xxbno == ltbno && xxlen == ltlen,
1647 error0);
1648 }
1649#endif
1650 /*
1651 * Update remaining by-block entry to the new, joined block.
1652 */
1653 nbno = ltbno;
1654 nlen = len + ltlen + gtlen;
1655 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1656 goto error0;
1657 }
1658 /*
1659 * Have only a left contiguous neighbor.
1660 * Merge it together with the new freespace.
1661 */
1662 else if (haveleft) {
1663 /*
1664 * Delete the old by-size entry on the left.
1665 */
1666 if ((error = xfs_alloc_lookup_eq(cnt_cur, ltbno, ltlen, &i)))
1667 goto error0;
1668 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1669 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1670 goto error0;
1671 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1672 /*
1673 * Back up the by-block cursor to the left neighbor, and
1674 * update its length.
1675 */
1676 if ((error = xfs_alloc_decrement(bno_cur, 0, &i)))
1677 goto error0;
1678 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1679 nbno = ltbno;
1680 nlen = len + ltlen;
1681 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1682 goto error0;
1683 }
1684 /*
1685 * Have only a right contiguous neighbor.
1686 * Merge it together with the new freespace.
1687 */
1688 else if (haveright) {
1689 /*
1690 * Delete the old by-size entry on the right.
1691 */
1692 if ((error = xfs_alloc_lookup_eq(cnt_cur, gtbno, gtlen, &i)))
1693 goto error0;
1694 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1695 if ((error = xfs_alloc_delete(cnt_cur, &i)))
1696 goto error0;
1697 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1698 /*
1699 * Update the starting block and length of the right
1700 * neighbor in the by-block tree.
1701 */
1702 nbno = bno;
1703 nlen = len + gtlen;
1704 if ((error = xfs_alloc_update(bno_cur, nbno, nlen)))
1705 goto error0;
1706 }
1707 /*
1708 * No contiguous neighbors.
1709 * Insert the new freespace into the by-block tree.
1710 */
1711 else {
1712 nbno = bno;
1713 nlen = len;
1714 if ((error = xfs_alloc_insert(bno_cur, &i)))
1715 goto error0;
1716 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1717 }
1718 xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
1719 bno_cur = NULL;
1720 /*
1721 * In all cases we need to insert the new freespace in the by-size tree.
1722 */
1723 if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
1724 goto error0;
1725 XFS_WANT_CORRUPTED_GOTO(i == 0, error0);
1726 if ((error = xfs_alloc_insert(cnt_cur, &i)))
1727 goto error0;
1728 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1729 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
1730 cnt_cur = NULL;
1731 /*
1732 * Update the freespace totals in the ag and superblock.
1733 */
1734 {
1735 xfs_agf_t *agf;
1736 xfs_perag_t *pag; /* per allocation group data */
1737
1738 agf = XFS_BUF_TO_AGF(agbp);
1739 pag = &mp->m_perag[agno];
Christoph Hellwig16259e72005-11-02 15:11:25 +11001740 be32_add(&agf->agf_freeblks, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 xfs_trans_agblocks_delta(tp, len);
1742 pag->pagf_freeblks += len;
1743 XFS_WANT_CORRUPTED_GOTO(
Christoph Hellwig16259e72005-11-02 15:11:25 +11001744 be32_to_cpu(agf->agf_freeblks) <=
1745 be32_to_cpu(agf->agf_length),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 error0);
1747 TRACE_MODAGF(NULL, agf, XFS_AGF_FREEBLKS);
1748 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FREEBLKS);
1749 if (!isfl)
1750 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (long)len);
1751 XFS_STATS_INC(xs_freex);
1752 XFS_STATS_ADD(xs_freeb, len);
1753 }
1754 TRACE_FREE(haveleft ?
1755 (haveright ? "both" : "left") :
1756 (haveright ? "right" : "none"),
1757 agno, bno, len, isfl);
1758
1759 /*
1760 * Since blocks move to the free list without the coordination
1761 * used in xfs_bmap_finish, we can't allow block to be available
1762 * for reallocation and non-transaction writing (user data)
1763 * until we know that the transaction that moved it to the free
1764 * list is permanently on disk. We track the blocks by declaring
1765 * these blocks as "busy"; the busy list is maintained on a per-ag
1766 * basis and each transaction records which entries should be removed
1767 * when the iclog commits to disk. If a busy block is allocated,
1768 * the iclog is pushed up to the LSN that freed the block.
1769 */
1770 xfs_alloc_mark_busy(tp, agno, bno, len);
1771 return 0;
1772
1773 error0:
1774 TRACE_FREE("error", agno, bno, len, isfl);
1775 if (bno_cur)
1776 xfs_btree_del_cursor(bno_cur, XFS_BTREE_ERROR);
1777 if (cnt_cur)
1778 xfs_btree_del_cursor(cnt_cur, XFS_BTREE_ERROR);
1779 return error;
1780}
1781
1782/*
1783 * Visible (exported) allocation/free functions.
1784 * Some of these are used just by xfs_alloc_btree.c and this file.
1785 */
1786
1787/*
1788 * Compute and fill in value of m_ag_maxlevels.
1789 */
1790void
1791xfs_alloc_compute_maxlevels(
1792 xfs_mount_t *mp) /* file system mount structure */
1793{
1794 int level;
1795 uint maxblocks;
1796 uint maxleafents;
1797 int minleafrecs;
1798 int minnoderecs;
1799
1800 maxleafents = (mp->m_sb.sb_agblocks + 1) / 2;
1801 minleafrecs = mp->m_alloc_mnr[0];
1802 minnoderecs = mp->m_alloc_mnr[1];
1803 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1804 for (level = 1; maxblocks > 1; level++)
1805 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1806 mp->m_ag_maxlevels = level;
1807}
1808
1809/*
1810 * Decide whether to use this allocation group for this allocation.
1811 * If so, fix up the btree freelist's size.
1812 */
1813STATIC int /* error */
1814xfs_alloc_fix_freelist(
1815 xfs_alloc_arg_t *args, /* allocation argument structure */
1816 int flags) /* XFS_ALLOC_FLAG_... */
1817{
1818 xfs_buf_t *agbp; /* agf buffer pointer */
1819 xfs_agf_t *agf; /* a.g. freespace structure pointer */
1820 xfs_buf_t *agflbp;/* agfl buffer pointer */
1821 xfs_agblock_t bno; /* freelist block */
1822 xfs_extlen_t delta; /* new blocks needed in freelist */
1823 int error; /* error result code */
1824 xfs_extlen_t longest;/* longest extent in allocation group */
1825 xfs_mount_t *mp; /* file system mount point structure */
1826 xfs_extlen_t need; /* total blocks needed in freelist */
1827 xfs_perag_t *pag; /* per-ag information structure */
1828 xfs_alloc_arg_t targs; /* local allocation arguments */
1829 xfs_trans_t *tp; /* transaction pointer */
1830
1831 mp = args->mp;
1832
1833 pag = args->pag;
1834 tp = args->tp;
1835 if (!pag->pagf_init) {
1836 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1837 &agbp)))
1838 return error;
1839 if (!pag->pagf_init) {
1840 args->agbp = NULL;
1841 return 0;
1842 }
1843 } else
1844 agbp = NULL;
1845
1846 /* If this is a metadata prefered pag and we are user data
1847 * then try somewhere else if we are not being asked to
1848 * try harder at this point
1849 */
1850 if (pag->pagf_metadata && args->userdata && flags) {
1851 args->agbp = NULL;
1852 return 0;
1853 }
1854
1855 need = XFS_MIN_FREELIST_PAG(pag, mp);
1856 delta = need > pag->pagf_flcount ? need - pag->pagf_flcount : 0;
1857 /*
1858 * If it looks like there isn't a long enough extent, or enough
1859 * total blocks, reject it.
1860 */
1861 longest = (pag->pagf_longest > delta) ?
1862 (pag->pagf_longest - delta) :
1863 (pag->pagf_flcount > 0 || pag->pagf_longest > 0);
1864 if (args->minlen + args->alignment + args->minalignslop - 1 > longest ||
1865 (args->minleft &&
1866 (int)(pag->pagf_freeblks + pag->pagf_flcount -
1867 need - args->total) <
1868 (int)args->minleft)) {
1869 if (agbp)
1870 xfs_trans_brelse(tp, agbp);
1871 args->agbp = NULL;
1872 return 0;
1873 }
1874 /*
1875 * Get the a.g. freespace buffer.
1876 * Can fail if we're not blocking on locks, and it's held.
1877 */
1878 if (agbp == NULL) {
1879 if ((error = xfs_alloc_read_agf(mp, tp, args->agno, flags,
1880 &agbp)))
1881 return error;
1882 if (agbp == NULL) {
1883 args->agbp = NULL;
1884 return 0;
1885 }
1886 }
1887 /*
1888 * Figure out how many blocks we should have in the freelist.
1889 */
1890 agf = XFS_BUF_TO_AGF(agbp);
1891 need = XFS_MIN_FREELIST(agf, mp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001892 delta = need > be32_to_cpu(agf->agf_flcount) ?
1893 (need - be32_to_cpu(agf->agf_flcount)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 /*
1895 * If there isn't enough total or single-extent, reject it.
1896 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001897 longest = be32_to_cpu(agf->agf_longest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 longest = (longest > delta) ? (longest - delta) :
Christoph Hellwig16259e72005-11-02 15:11:25 +11001899 (be32_to_cpu(agf->agf_flcount) > 0 || longest > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 if (args->minlen + args->alignment + args->minalignslop - 1 > longest ||
1901 (args->minleft &&
Christoph Hellwig16259e72005-11-02 15:11:25 +11001902 (int)(be32_to_cpu(agf->agf_freeblks) +
1903 be32_to_cpu(agf->agf_flcount) - need - args->total) <
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 (int)args->minleft)) {
1905 xfs_trans_brelse(tp, agbp);
1906 args->agbp = NULL;
1907 return 0;
1908 }
1909 /*
1910 * Make the freelist shorter if it's too long.
1911 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001912 while (be32_to_cpu(agf->agf_flcount) > need) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 xfs_buf_t *bp;
1914
1915 if ((error = xfs_alloc_get_freelist(tp, agbp, &bno)))
1916 return error;
1917 if ((error = xfs_free_ag_extent(tp, agbp, args->agno, bno, 1, 1)))
1918 return error;
1919 bp = xfs_btree_get_bufs(mp, tp, args->agno, bno, 0);
1920 xfs_trans_binval(tp, bp);
1921 }
1922 /*
1923 * Initialize the args structure.
1924 */
1925 targs.tp = tp;
1926 targs.mp = mp;
1927 targs.agbp = agbp;
1928 targs.agno = args->agno;
1929 targs.mod = targs.minleft = targs.wasdel = targs.userdata =
1930 targs.minalignslop = 0;
1931 targs.alignment = targs.minlen = targs.prod = targs.isfl = 1;
1932 targs.type = XFS_ALLOCTYPE_THIS_AG;
1933 targs.pag = pag;
1934 if ((error = xfs_alloc_read_agfl(mp, tp, targs.agno, &agflbp)))
1935 return error;
1936 /*
1937 * Make the freelist longer if it's too short.
1938 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001939 while (be32_to_cpu(agf->agf_flcount) < need) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 targs.agbno = 0;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001941 targs.maxlen = need - be32_to_cpu(agf->agf_flcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 /*
1943 * Allocate as many blocks as possible at once.
1944 */
1945 if ((error = xfs_alloc_ag_vextent(&targs)))
1946 return error;
1947 /*
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 */
1952 if (targs.agbno == NULLAGBLOCK)
1953 break;
1954 /*
1955 * Put each allocated block on the list.
1956 */
1957 for (bno = targs.agbno; bno < targs.agbno + targs.len; bno++) {
1958 if ((error = xfs_alloc_put_freelist(tp, agbp, agflbp,
1959 bno)))
1960 return error;
1961 }
1962 }
1963 args->agbp = agbp;
1964 return 0;
1965}
1966
1967/*
1968 * Get a block from the freelist.
1969 * Returns with the buffer for the block gotten.
1970 */
1971int /* error */
1972xfs_alloc_get_freelist(
1973 xfs_trans_t *tp, /* transaction pointer */
1974 xfs_buf_t *agbp, /* buffer containing the agf structure */
1975 xfs_agblock_t *bnop) /* block address retrieved from freelist */
1976{
1977 xfs_agf_t *agf; /* a.g. freespace structure */
1978 xfs_agfl_t *agfl; /* a.g. freelist structure */
1979 xfs_buf_t *agflbp;/* buffer for a.g. freelist structure */
1980 xfs_agblock_t bno; /* block number returned */
1981 int error;
1982#ifdef XFS_ALLOC_TRACE
1983 static char fname[] = "xfs_alloc_get_freelist";
1984#endif
1985 xfs_mount_t *mp; /* mount structure */
1986 xfs_perag_t *pag; /* per allocation group data */
1987
1988 agf = XFS_BUF_TO_AGF(agbp);
1989 /*
1990 * Freelist is empty, give up.
1991 */
1992 if (!agf->agf_flcount) {
1993 *bnop = NULLAGBLOCK;
1994 return 0;
1995 }
1996 /*
1997 * Read the array of free blocks.
1998 */
1999 mp = tp->t_mountp;
2000 if ((error = xfs_alloc_read_agfl(mp, tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +11002001 be32_to_cpu(agf->agf_seqno), &agflbp)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 return error;
2003 agfl = XFS_BUF_TO_AGFL(agflbp);
2004 /*
2005 * Get the block number and update the data structures.
2006 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002007 bno = INT_GET(agfl->agfl_bno[be32_to_cpu(agf->agf_flfirst)], ARCH_CONVERT);
2008 be32_add(&agf->agf_flfirst, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 xfs_trans_brelse(tp, agflbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002010 if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 agf->agf_flfirst = 0;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002012 pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)];
2013 be32_add(&agf->agf_flcount, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 xfs_trans_agflist_delta(tp, -1);
2015 pag->pagf_flcount--;
2016 TRACE_MODAGF(NULL, agf, XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT);
2017 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT);
2018 *bnop = bno;
2019
2020 /*
2021 * As blocks are freed, they are added to the per-ag busy list
2022 * and remain there until the freeing transaction is committed to
2023 * disk. Now that we have allocated blocks, this list must be
2024 * searched to see if a block is being reused. If one is, then
2025 * the freeing transaction must be pushed to disk NOW by forcing
2026 * to disk all iclogs up that transaction's LSN.
2027 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002028 xfs_alloc_search_busy(tp, be32_to_cpu(agf->agf_seqno), bno, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 return 0;
2030}
2031
2032/*
2033 * Log the given fields from the agf structure.
2034 */
2035void
2036xfs_alloc_log_agf(
2037 xfs_trans_t *tp, /* transaction pointer */
2038 xfs_buf_t *bp, /* buffer for a.g. freelist header */
2039 int fields) /* mask of fields to be logged (XFS_AGF_...) */
2040{
2041 int first; /* first byte offset */
2042 int last; /* last byte offset */
2043 static const short offsets[] = {
2044 offsetof(xfs_agf_t, agf_magicnum),
2045 offsetof(xfs_agf_t, agf_versionnum),
2046 offsetof(xfs_agf_t, agf_seqno),
2047 offsetof(xfs_agf_t, agf_length),
2048 offsetof(xfs_agf_t, agf_roots[0]),
2049 offsetof(xfs_agf_t, agf_levels[0]),
2050 offsetof(xfs_agf_t, agf_flfirst),
2051 offsetof(xfs_agf_t, agf_fllast),
2052 offsetof(xfs_agf_t, agf_flcount),
2053 offsetof(xfs_agf_t, agf_freeblks),
2054 offsetof(xfs_agf_t, agf_longest),
2055 sizeof(xfs_agf_t)
2056 };
2057
2058 xfs_btree_offsets(fields, offsets, XFS_AGF_NUM_BITS, &first, &last);
2059 xfs_trans_log_buf(tp, bp, (uint)first, (uint)last);
2060}
2061
2062/*
2063 * Interface for inode allocation to force the pag data to be initialized.
2064 */
2065int /* error */
2066xfs_alloc_pagf_init(
2067 xfs_mount_t *mp, /* file system mount structure */
2068 xfs_trans_t *tp, /* transaction pointer */
2069 xfs_agnumber_t agno, /* allocation group number */
2070 int flags) /* XFS_ALLOC_FLAGS_... */
2071{
2072 xfs_buf_t *bp;
2073 int error;
2074
2075 if ((error = xfs_alloc_read_agf(mp, tp, agno, flags, &bp)))
2076 return error;
2077 if (bp)
2078 xfs_trans_brelse(tp, bp);
2079 return 0;
2080}
2081
2082/*
2083 * Put the block on the freelist for the allocation group.
2084 */
2085int /* error */
2086xfs_alloc_put_freelist(
2087 xfs_trans_t *tp, /* transaction pointer */
2088 xfs_buf_t *agbp, /* buffer for a.g. freelist header */
2089 xfs_buf_t *agflbp,/* buffer for a.g. free block array */
2090 xfs_agblock_t bno) /* block being freed */
2091{
2092 xfs_agf_t *agf; /* a.g. freespace structure */
2093 xfs_agfl_t *agfl; /* a.g. free block array */
2094 xfs_agblock_t *blockp;/* pointer to array entry */
2095 int error;
2096#ifdef XFS_ALLOC_TRACE
2097 static char fname[] = "xfs_alloc_put_freelist";
2098#endif
2099 xfs_mount_t *mp; /* mount structure */
2100 xfs_perag_t *pag; /* per allocation group data */
2101
2102 agf = XFS_BUF_TO_AGF(agbp);
2103 mp = tp->t_mountp;
2104
2105 if (!agflbp && (error = xfs_alloc_read_agfl(mp, tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +11002106 be32_to_cpu(agf->agf_seqno), &agflbp)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 return error;
2108 agfl = XFS_BUF_TO_AGFL(agflbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002109 be32_add(&agf->agf_fllast, 1);
2110 if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 agf->agf_fllast = 0;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002112 pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)];
2113 be32_add(&agf->agf_flcount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 xfs_trans_agflist_delta(tp, 1);
2115 pag->pagf_flcount++;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002116 ASSERT(be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp));
2117 blockp = &agfl->agfl_bno[be32_to_cpu(agf->agf_fllast)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 INT_SET(*blockp, ARCH_CONVERT, bno);
2119 TRACE_MODAGF(NULL, agf, XFS_AGF_FLLAST | XFS_AGF_FLCOUNT);
2120 xfs_alloc_log_agf(tp, agbp, XFS_AGF_FLLAST | XFS_AGF_FLCOUNT);
2121 xfs_trans_log_buf(tp, agflbp,
2122 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl),
2123 (int)((xfs_caddr_t)blockp - (xfs_caddr_t)agfl +
2124 sizeof(xfs_agblock_t) - 1));
2125 return 0;
2126}
2127
2128/*
2129 * Read in the allocation group header (free/alloc section).
2130 */
2131int /* error */
2132xfs_alloc_read_agf(
2133 xfs_mount_t *mp, /* mount point structure */
2134 xfs_trans_t *tp, /* transaction pointer */
2135 xfs_agnumber_t agno, /* allocation group number */
2136 int flags, /* XFS_ALLOC_FLAG_... */
2137 xfs_buf_t **bpp) /* buffer for the ag freelist header */
2138{
2139 xfs_agf_t *agf; /* ag freelist header */
2140 int agf_ok; /* set if agf is consistent */
2141 xfs_buf_t *bp; /* return value */
2142 xfs_perag_t *pag; /* per allocation group data */
2143 int error;
2144
2145 ASSERT(agno != NULLAGNUMBER);
2146 error = xfs_trans_read_buf(
2147 mp, tp, mp->m_ddev_targp,
2148 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
2149 XFS_FSS_TO_BB(mp, 1),
2150 (flags & XFS_ALLOC_FLAG_TRYLOCK) ? XFS_BUF_TRYLOCK : 0U,
2151 &bp);
2152 if (error)
2153 return error;
2154 ASSERT(!bp || !XFS_BUF_GETERROR(bp));
2155 if (!bp) {
2156 *bpp = NULL;
2157 return 0;
2158 }
2159 /*
2160 * Validate the magic number of the agf block.
2161 */
2162 agf = XFS_BUF_TO_AGF(bp);
2163 agf_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11002164 be32_to_cpu(agf->agf_magicnum) == XFS_AGF_MAGIC &&
2165 XFS_AGF_GOOD_VERSION(be32_to_cpu(agf->agf_versionnum)) &&
2166 be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length) &&
2167 be32_to_cpu(agf->agf_flfirst) < XFS_AGFL_SIZE(mp) &&
2168 be32_to_cpu(agf->agf_fllast) < XFS_AGFL_SIZE(mp) &&
2169 be32_to_cpu(agf->agf_flcount) <= XFS_AGFL_SIZE(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 if (unlikely(XFS_TEST_ERROR(!agf_ok, mp, XFS_ERRTAG_ALLOC_READ_AGF,
2171 XFS_RANDOM_ALLOC_READ_AGF))) {
2172 XFS_CORRUPTION_ERROR("xfs_alloc_read_agf",
2173 XFS_ERRLEVEL_LOW, mp, agf);
2174 xfs_trans_brelse(tp, bp);
2175 return XFS_ERROR(EFSCORRUPTED);
2176 }
2177 pag = &mp->m_perag[agno];
2178 if (!pag->pagf_init) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11002179 pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks);
2180 pag->pagf_flcount = be32_to_cpu(agf->agf_flcount);
2181 pag->pagf_longest = be32_to_cpu(agf->agf_longest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 pag->pagf_levels[XFS_BTNUM_BNOi] =
Christoph Hellwig16259e72005-11-02 15:11:25 +11002183 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 pag->pagf_levels[XFS_BTNUM_CNTi] =
Christoph Hellwig16259e72005-11-02 15:11:25 +11002185 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 spinlock_init(&pag->pagb_lock, "xfspagb");
2187 pag->pagb_list = kmem_zalloc(XFS_PAGB_NUM_SLOTS *
2188 sizeof(xfs_perag_busy_t), KM_SLEEP);
2189 pag->pagf_init = 1;
2190 }
2191#ifdef DEBUG
2192 else if (!XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11002193 ASSERT(pag->pagf_freeblks == be32_to_cpu(agf->agf_freeblks));
2194 ASSERT(pag->pagf_flcount == be32_to_cpu(agf->agf_flcount));
2195 ASSERT(pag->pagf_longest == be32_to_cpu(agf->agf_longest));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 ASSERT(pag->pagf_levels[XFS_BTNUM_BNOi] ==
Christoph Hellwig16259e72005-11-02 15:11:25 +11002197 be32_to_cpu(agf->agf_levels[XFS_BTNUM_BNOi]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 ASSERT(pag->pagf_levels[XFS_BTNUM_CNTi] ==
Christoph Hellwig16259e72005-11-02 15:11:25 +11002199 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 }
2201#endif
2202 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGF, XFS_AGF_REF);
2203 *bpp = bp;
2204 return 0;
2205}
2206
2207/*
2208 * Allocate an extent (variable-size).
2209 * Depending on the allocation type, we either look in a single allocation
2210 * group or loop over the allocation groups to find the result.
2211 */
2212int /* error */
2213xfs_alloc_vextent(
2214 xfs_alloc_arg_t *args) /* allocation argument structure */
2215{
2216 xfs_agblock_t agsize; /* allocation group size */
2217 int error;
2218 int flags; /* XFS_ALLOC_FLAG_... locking flags */
2219#ifdef XFS_ALLOC_TRACE
2220 static char fname[] = "xfs_alloc_vextent";
2221#endif
2222 xfs_extlen_t minleft;/* minimum left value, temp copy */
2223 xfs_mount_t *mp; /* mount structure pointer */
2224 xfs_agnumber_t sagno; /* starting allocation group number */
2225 xfs_alloctype_t type; /* input allocation type */
2226 int bump_rotor = 0;
2227 int no_min = 0;
2228 xfs_agnumber_t rotorstep = xfs_rotorstep; /* inode32 agf stepper */
2229
2230 mp = args->mp;
2231 type = args->otype = args->type;
2232 args->agbno = NULLAGBLOCK;
2233 /*
2234 * Just fix this up, for the case where the last a.g. is shorter
2235 * (or there's only one a.g.) and the caller couldn't easily figure
2236 * that out (xfs_bmap_alloc).
2237 */
2238 agsize = mp->m_sb.sb_agblocks;
2239 if (args->maxlen > agsize)
2240 args->maxlen = agsize;
2241 if (args->alignment == 0)
2242 args->alignment = 1;
2243 ASSERT(XFS_FSB_TO_AGNO(mp, args->fsbno) < mp->m_sb.sb_agcount);
2244 ASSERT(XFS_FSB_TO_AGBNO(mp, args->fsbno) < agsize);
2245 ASSERT(args->minlen <= args->maxlen);
2246 ASSERT(args->minlen <= agsize);
2247 ASSERT(args->mod < args->prod);
2248 if (XFS_FSB_TO_AGNO(mp, args->fsbno) >= mp->m_sb.sb_agcount ||
2249 XFS_FSB_TO_AGBNO(mp, args->fsbno) >= agsize ||
2250 args->minlen > args->maxlen || args->minlen > agsize ||
2251 args->mod >= args->prod) {
2252 args->fsbno = NULLFSBLOCK;
2253 TRACE_ALLOC("badargs", args);
2254 return 0;
2255 }
2256 minleft = args->minleft;
2257
2258 switch (type) {
2259 case XFS_ALLOCTYPE_THIS_AG:
2260 case XFS_ALLOCTYPE_NEAR_BNO:
2261 case XFS_ALLOCTYPE_THIS_BNO:
2262 /*
2263 * These three force us into a single a.g.
2264 */
2265 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2266 down_read(&mp->m_peraglock);
2267 args->pag = &mp->m_perag[args->agno];
2268 args->minleft = 0;
2269 error = xfs_alloc_fix_freelist(args, 0);
2270 args->minleft = minleft;
2271 if (error) {
2272 TRACE_ALLOC("nofix", args);
2273 goto error0;
2274 }
2275 if (!args->agbp) {
2276 up_read(&mp->m_peraglock);
2277 TRACE_ALLOC("noagbp", args);
2278 break;
2279 }
2280 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2281 if ((error = xfs_alloc_ag_vextent(args)))
2282 goto error0;
2283 up_read(&mp->m_peraglock);
2284 break;
2285 case XFS_ALLOCTYPE_START_BNO:
2286 /*
2287 * Try near allocation first, then anywhere-in-ag after
2288 * the first a.g. fails.
2289 */
2290 if ((args->userdata == XFS_ALLOC_INITIAL_USER_DATA) &&
2291 (mp->m_flags & XFS_MOUNT_32BITINODES)) {
2292 args->fsbno = XFS_AGB_TO_FSB(mp,
2293 ((mp->m_agfrotor / rotorstep) %
2294 mp->m_sb.sb_agcount), 0);
2295 bump_rotor = 1;
2296 }
2297 args->agbno = XFS_FSB_TO_AGBNO(mp, args->fsbno);
2298 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2299 /* FALLTHROUGH */
2300 case XFS_ALLOCTYPE_ANY_AG:
2301 case XFS_ALLOCTYPE_START_AG:
2302 case XFS_ALLOCTYPE_FIRST_AG:
2303 /*
2304 * Rotate through the allocation groups looking for a winner.
2305 */
2306 if (type == XFS_ALLOCTYPE_ANY_AG) {
2307 /*
2308 * Start with the last place we left off.
2309 */
2310 args->agno = sagno = (mp->m_agfrotor / rotorstep) %
2311 mp->m_sb.sb_agcount;
2312 args->type = XFS_ALLOCTYPE_THIS_AG;
2313 flags = XFS_ALLOC_FLAG_TRYLOCK;
2314 } else if (type == XFS_ALLOCTYPE_FIRST_AG) {
2315 /*
2316 * Start with allocation group given by bno.
2317 */
2318 args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2319 args->type = XFS_ALLOCTYPE_THIS_AG;
2320 sagno = 0;
2321 flags = 0;
2322 } else {
2323 if (type == XFS_ALLOCTYPE_START_AG)
2324 args->type = XFS_ALLOCTYPE_THIS_AG;
2325 /*
2326 * Start with the given allocation group.
2327 */
2328 args->agno = sagno = XFS_FSB_TO_AGNO(mp, args->fsbno);
2329 flags = XFS_ALLOC_FLAG_TRYLOCK;
2330 }
2331 /*
2332 * Loop over allocation groups twice; first time with
2333 * trylock set, second time without.
2334 */
2335 down_read(&mp->m_peraglock);
2336 for (;;) {
2337 args->pag = &mp->m_perag[args->agno];
2338 if (no_min) args->minleft = 0;
2339 error = xfs_alloc_fix_freelist(args, flags);
2340 args->minleft = minleft;
2341 if (error) {
2342 TRACE_ALLOC("nofix", args);
2343 goto error0;
2344 }
2345 /*
2346 * If we get a buffer back then the allocation will fly.
2347 */
2348 if (args->agbp) {
2349 if ((error = xfs_alloc_ag_vextent(args)))
2350 goto error0;
2351 break;
2352 }
2353 TRACE_ALLOC("loopfailed", args);
2354 /*
2355 * Didn't work, figure out the next iteration.
2356 */
2357 if (args->agno == sagno &&
2358 type == XFS_ALLOCTYPE_START_BNO)
2359 args->type = XFS_ALLOCTYPE_THIS_AG;
2360 if (++(args->agno) == mp->m_sb.sb_agcount)
2361 args->agno = 0;
2362 /*
2363 * Reached the starting a.g., must either be done
2364 * or switch to non-trylock mode.
2365 */
2366 if (args->agno == sagno) {
2367 if (no_min == 1) {
2368 args->agbno = NULLAGBLOCK;
2369 TRACE_ALLOC("allfailed", args);
2370 break;
2371 }
2372 if (flags == 0) {
2373 no_min = 1;
2374 } else {
2375 flags = 0;
2376 if (type == XFS_ALLOCTYPE_START_BNO) {
2377 args->agbno = XFS_FSB_TO_AGBNO(mp,
2378 args->fsbno);
2379 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2380 }
2381 }
2382 }
2383 }
2384 up_read(&mp->m_peraglock);
2385 if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) {
2386 if (args->agno == sagno)
2387 mp->m_agfrotor = (mp->m_agfrotor + 1) %
2388 (mp->m_sb.sb_agcount * rotorstep);
2389 else
2390 mp->m_agfrotor = (args->agno * rotorstep + 1) %
2391 (mp->m_sb.sb_agcount * rotorstep);
2392 }
2393 break;
2394 default:
2395 ASSERT(0);
2396 /* NOTREACHED */
2397 }
2398 if (args->agbno == NULLAGBLOCK)
2399 args->fsbno = NULLFSBLOCK;
2400 else {
2401 args->fsbno = XFS_AGB_TO_FSB(mp, args->agno, args->agbno);
2402#ifdef DEBUG
2403 ASSERT(args->len >= args->minlen);
2404 ASSERT(args->len <= args->maxlen);
2405 ASSERT(args->agbno % args->alignment == 0);
2406 XFS_AG_CHECK_DADDR(mp, XFS_FSB_TO_DADDR(mp, args->fsbno),
2407 args->len);
2408#endif
2409 }
2410 return 0;
2411error0:
2412 up_read(&mp->m_peraglock);
2413 return error;
2414}
2415
2416/*
2417 * Free an extent.
2418 * Just break up the extent address and hand off to xfs_free_ag_extent
2419 * after fixing up the freelist.
2420 */
2421int /* error */
2422xfs_free_extent(
2423 xfs_trans_t *tp, /* transaction pointer */
2424 xfs_fsblock_t bno, /* starting block number of extent */
2425 xfs_extlen_t len) /* length of extent */
2426{
2427#ifdef DEBUG
2428 xfs_agf_t *agf; /* a.g. freespace header */
2429#endif
2430 xfs_alloc_arg_t args; /* allocation argument structure */
2431 int error;
2432
2433 ASSERT(len != 0);
2434 args.tp = tp;
2435 args.mp = tp->t_mountp;
2436 args.agno = XFS_FSB_TO_AGNO(args.mp, bno);
2437 ASSERT(args.agno < args.mp->m_sb.sb_agcount);
2438 args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno);
2439 args.alignment = 1;
2440 args.minlen = args.minleft = args.minalignslop = 0;
2441 down_read(&args.mp->m_peraglock);
2442 args.pag = &args.mp->m_perag[args.agno];
2443 if ((error = xfs_alloc_fix_freelist(&args, 0)))
2444 goto error0;
2445#ifdef DEBUG
2446 ASSERT(args.agbp != NULL);
2447 agf = XFS_BUF_TO_AGF(args.agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002448 ASSERT(args.agbno + len <= be32_to_cpu(agf->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449#endif
2450 error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno,
2451 len, 0);
2452error0:
2453 up_read(&args.mp->m_peraglock);
2454 return error;
2455}
2456
2457
2458/*
2459 * AG Busy list management
2460 * The busy list contains block ranges that have been freed but whose
2461 * transacations have not yet hit disk. If any block listed in a busy
2462 * list is reused, the transaction that freed it must be forced to disk
2463 * before continuing to use the block.
2464 *
2465 * xfs_alloc_mark_busy - add to the per-ag busy list
2466 * xfs_alloc_clear_busy - remove an item from the per-ag busy list
2467 */
2468void
2469xfs_alloc_mark_busy(xfs_trans_t *tp,
2470 xfs_agnumber_t agno,
2471 xfs_agblock_t bno,
2472 xfs_extlen_t len)
2473{
2474 xfs_mount_t *mp;
2475 xfs_perag_busy_t *bsy;
2476 int n;
2477 SPLDECL(s);
2478
2479 mp = tp->t_mountp;
2480 s = mutex_spinlock(&mp->m_perag[agno].pagb_lock);
2481
2482 /* search pagb_list for an open slot */
2483 for (bsy = mp->m_perag[agno].pagb_list, n = 0;
2484 n < XFS_PAGB_NUM_SLOTS;
2485 bsy++, n++) {
2486 if (bsy->busy_tp == NULL) {
2487 break;
2488 }
2489 }
2490
2491 if (n < XFS_PAGB_NUM_SLOTS) {
2492 bsy = &mp->m_perag[agno].pagb_list[n];
2493 mp->m_perag[agno].pagb_count++;
2494 TRACE_BUSY("xfs_alloc_mark_busy", "got", agno, bno, len, n, tp);
2495 bsy->busy_start = bno;
2496 bsy->busy_length = len;
2497 bsy->busy_tp = tp;
2498 xfs_trans_add_busy(tp, agno, n);
2499 } else {
2500 TRACE_BUSY("xfs_alloc_mark_busy", "FULL", agno, bno, len, -1, tp);
2501 /*
2502 * The busy list is full! Since it is now not possible to
2503 * track the free block, make this a synchronous transaction
2504 * to insure that the block is not reused before this
2505 * transaction commits.
2506 */
2507 xfs_trans_set_sync(tp);
2508 }
2509
2510 mutex_spinunlock(&mp->m_perag[agno].pagb_lock, s);
2511}
2512
2513void
2514xfs_alloc_clear_busy(xfs_trans_t *tp,
2515 xfs_agnumber_t agno,
2516 int idx)
2517{
2518 xfs_mount_t *mp;
2519 xfs_perag_busy_t *list;
2520 SPLDECL(s);
2521
2522 mp = tp->t_mountp;
2523
2524 s = mutex_spinlock(&mp->m_perag[agno].pagb_lock);
2525 list = mp->m_perag[agno].pagb_list;
2526
2527 ASSERT(idx < XFS_PAGB_NUM_SLOTS);
2528 if (list[idx].busy_tp == tp) {
2529 TRACE_UNBUSY("xfs_alloc_clear_busy", "found", agno, idx, tp);
2530 list[idx].busy_tp = NULL;
2531 mp->m_perag[agno].pagb_count--;
2532 } else {
2533 TRACE_UNBUSY("xfs_alloc_clear_busy", "missing", agno, idx, tp);
2534 }
2535
2536 mutex_spinunlock(&mp->m_perag[agno].pagb_lock, s);
2537}
2538
2539
2540/*
2541 * returns non-zero if any of (agno,bno):len is in a busy list
2542 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002543STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544xfs_alloc_search_busy(xfs_trans_t *tp,
2545 xfs_agnumber_t agno,
2546 xfs_agblock_t bno,
2547 xfs_extlen_t len)
2548{
2549 xfs_mount_t *mp;
2550 xfs_perag_busy_t *bsy;
2551 int n;
2552 xfs_agblock_t uend, bend;
2553 xfs_lsn_t lsn;
2554 int cnt;
2555 SPLDECL(s);
2556
2557 mp = tp->t_mountp;
2558
2559 s = mutex_spinlock(&mp->m_perag[agno].pagb_lock);
2560 cnt = mp->m_perag[agno].pagb_count;
2561
2562 uend = bno + len - 1;
2563
2564 /* search pagb_list for this slot, skipping open slots */
2565 for (bsy = mp->m_perag[agno].pagb_list, n = 0;
2566 cnt; bsy++, n++) {
2567
2568 /*
2569 * (start1,length1) within (start2, length2)
2570 */
2571 if (bsy->busy_tp != NULL) {
2572 bend = bsy->busy_start + bsy->busy_length - 1;
2573 if ((bno > bend) ||
2574 (uend < bsy->busy_start)) {
2575 cnt--;
2576 } else {
2577 TRACE_BUSYSEARCH("xfs_alloc_search_busy",
2578 "found1", agno, bno, len, n,
2579 tp);
2580 break;
2581 }
2582 }
2583 }
2584
2585 /*
2586 * If a block was found, force the log through the LSN of the
2587 * transaction that freed the block
2588 */
2589 if (cnt) {
2590 TRACE_BUSYSEARCH("xfs_alloc_search_busy", "found", agno, bno, len, n, tp);
2591 lsn = bsy->busy_tp->t_commit_lsn;
2592 mutex_spinunlock(&mp->m_perag[agno].pagb_lock, s);
2593 xfs_log_force(mp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC);
2594 } else {
2595 TRACE_BUSYSEARCH("xfs_alloc_search_busy", "not-found", agno, bno, len, n, tp);
2596 n = -1;
2597 mutex_spinunlock(&mp->m_perag[agno].pagb_lock, s);
2598 }
2599
2600 return n;
2601}