adbd9141aea165158b490cb3d3bbed19aa0a3bb7
[safe/jmp/linux-2.6] / fs / xfs / xfs_alloc_btree.c
1 /*
2  * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_btree_trace.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_alloc.h"
41 #include "xfs_error.h"
42 #include "xfs_trace.h"
43
44
45 STATIC struct xfs_btree_cur *
46 xfs_allocbt_dup_cursor(
47         struct xfs_btree_cur    *cur)
48 {
49         return xfs_allocbt_init_cursor(cur->bc_mp, cur->bc_tp,
50                         cur->bc_private.a.agbp, cur->bc_private.a.agno,
51                         cur->bc_btnum);
52 }
53
54 STATIC void
55 xfs_allocbt_set_root(
56         struct xfs_btree_cur    *cur,
57         union xfs_btree_ptr     *ptr,
58         int                     inc)
59 {
60         struct xfs_buf          *agbp = cur->bc_private.a.agbp;
61         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
62         xfs_agnumber_t          seqno = be32_to_cpu(agf->agf_seqno);
63         int                     btnum = cur->bc_btnum;
64
65         ASSERT(ptr->s != 0);
66
67         agf->agf_roots[btnum] = ptr->s;
68         be32_add_cpu(&agf->agf_levels[btnum], inc);
69         cur->bc_mp->m_perag[seqno].pagf_levels[btnum] += inc;
70
71         xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
72 }
73
74 STATIC int
75 xfs_allocbt_alloc_block(
76         struct xfs_btree_cur    *cur,
77         union xfs_btree_ptr     *start,
78         union xfs_btree_ptr     *new,
79         int                     length,
80         int                     *stat)
81 {
82         int                     error;
83         xfs_agblock_t           bno;
84
85         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
86
87         /* Allocate the new block from the freelist. If we can't, give up.  */
88         error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp,
89                                        &bno, 1);
90         if (error) {
91                 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
92                 return error;
93         }
94
95         if (bno == NULLAGBLOCK) {
96                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
97                 *stat = 0;
98                 return 0;
99         }
100
101         xfs_trans_agbtree_delta(cur->bc_tp, 1);
102         new->s = cpu_to_be32(bno);
103
104         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
105         *stat = 1;
106         return 0;
107 }
108
109 STATIC int
110 xfs_allocbt_free_block(
111         struct xfs_btree_cur    *cur,
112         struct xfs_buf          *bp)
113 {
114         struct xfs_buf          *agbp = cur->bc_private.a.agbp;
115         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
116         xfs_agblock_t           bno;
117         int                     error;
118
119         bno = xfs_daddr_to_agbno(cur->bc_mp, XFS_BUF_ADDR(bp));
120         error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
121         if (error)
122                 return error;
123
124         /*
125          * Since blocks move to the free list without the coordination used in
126          * xfs_bmap_finish, we can't allow block to be available for
127          * reallocation and non-transaction writing (user data) until we know
128          * that the transaction that moved it to the free list is permanently
129          * on disk. We track the blocks by declaring these blocks as "busy";
130          * the busy list is maintained on a per-ag basis and each transaction
131          * records which entries should be removed when the iclog commits to
132          * disk. If a busy block is allocated, the iclog is pushed up to the
133          * LSN that freed the block.
134          */
135         xfs_alloc_mark_busy(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1);
136         xfs_trans_agbtree_delta(cur->bc_tp, -1);
137         return 0;
138 }
139
140 /*
141  * Update the longest extent in the AGF
142  */
143 STATIC void
144 xfs_allocbt_update_lastrec(
145         struct xfs_btree_cur    *cur,
146         struct xfs_btree_block  *block,
147         union xfs_btree_rec     *rec,
148         int                     ptr,
149         int                     reason)
150 {
151         struct xfs_agf          *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
152         xfs_agnumber_t          seqno = be32_to_cpu(agf->agf_seqno);
153         __be32                  len;
154         int                     numrecs;
155
156         ASSERT(cur->bc_btnum == XFS_BTNUM_CNT);
157
158         switch (reason) {
159         case LASTREC_UPDATE:
160                 /*
161                  * If this is the last leaf block and it's the last record,
162                  * then update the size of the longest extent in the AG.
163                  */
164                 if (ptr != xfs_btree_get_numrecs(block))
165                         return;
166                 len = rec->alloc.ar_blockcount;
167                 break;
168         case LASTREC_INSREC:
169                 if (be32_to_cpu(rec->alloc.ar_blockcount) <=
170                     be32_to_cpu(agf->agf_longest))
171                         return;
172                 len = rec->alloc.ar_blockcount;
173                 break;
174         case LASTREC_DELREC:
175                 numrecs = xfs_btree_get_numrecs(block);
176                 if (ptr <= numrecs)
177                         return;
178                 ASSERT(ptr == numrecs + 1);
179
180                 if (numrecs) {
181                         xfs_alloc_rec_t *rrp;
182
183                         rrp = XFS_ALLOC_REC_ADDR(cur->bc_mp, block, numrecs);
184                         len = rrp->ar_blockcount;
185                 } else {
186                         len = 0;
187                 }
188
189                 break;
190         default:
191                 ASSERT(0);
192                 return;
193         }
194
195         agf->agf_longest = len;
196         cur->bc_mp->m_perag[seqno].pagf_longest = be32_to_cpu(len);
197         xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, XFS_AGF_LONGEST);
198 }
199
200 STATIC int
201 xfs_allocbt_get_minrecs(
202         struct xfs_btree_cur    *cur,
203         int                     level)
204 {
205         return cur->bc_mp->m_alloc_mnr[level != 0];
206 }
207
208 STATIC int
209 xfs_allocbt_get_maxrecs(
210         struct xfs_btree_cur    *cur,
211         int                     level)
212 {
213         return cur->bc_mp->m_alloc_mxr[level != 0];
214 }
215
216 STATIC void
217 xfs_allocbt_init_key_from_rec(
218         union xfs_btree_key     *key,
219         union xfs_btree_rec     *rec)
220 {
221         ASSERT(rec->alloc.ar_startblock != 0);
222
223         key->alloc.ar_startblock = rec->alloc.ar_startblock;
224         key->alloc.ar_blockcount = rec->alloc.ar_blockcount;
225 }
226
227 STATIC void
228 xfs_allocbt_init_rec_from_key(
229         union xfs_btree_key     *key,
230         union xfs_btree_rec     *rec)
231 {
232         ASSERT(key->alloc.ar_startblock != 0);
233
234         rec->alloc.ar_startblock = key->alloc.ar_startblock;
235         rec->alloc.ar_blockcount = key->alloc.ar_blockcount;
236 }
237
238 STATIC void
239 xfs_allocbt_init_rec_from_cur(
240         struct xfs_btree_cur    *cur,
241         union xfs_btree_rec     *rec)
242 {
243         ASSERT(cur->bc_rec.a.ar_startblock != 0);
244
245         rec->alloc.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
246         rec->alloc.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
247 }
248
249 STATIC void
250 xfs_allocbt_init_ptr_from_cur(
251         struct xfs_btree_cur    *cur,
252         union xfs_btree_ptr     *ptr)
253 {
254         struct xfs_agf          *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
255
256         ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
257         ASSERT(agf->agf_roots[cur->bc_btnum] != 0);
258
259         ptr->s = agf->agf_roots[cur->bc_btnum];
260 }
261
262 STATIC __int64_t
263 xfs_allocbt_key_diff(
264         struct xfs_btree_cur    *cur,
265         union xfs_btree_key     *key)
266 {
267         xfs_alloc_rec_incore_t  *rec = &cur->bc_rec.a;
268         xfs_alloc_key_t         *kp = &key->alloc;
269         __int64_t               diff;
270
271         if (cur->bc_btnum == XFS_BTNUM_BNO) {
272                 return (__int64_t)be32_to_cpu(kp->ar_startblock) -
273                                 rec->ar_startblock;
274         }
275
276         diff = (__int64_t)be32_to_cpu(kp->ar_blockcount) - rec->ar_blockcount;
277         if (diff)
278                 return diff;
279
280         return (__int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock;
281 }
282
283 STATIC int
284 xfs_allocbt_kill_root(
285         struct xfs_btree_cur    *cur,
286         struct xfs_buf          *bp,
287         int                     level,
288         union xfs_btree_ptr     *newroot)
289 {
290         int                     error;
291
292         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
293         XFS_BTREE_STATS_INC(cur, killroot);
294
295         /*
296          * Update the root pointer, decreasing the level by 1 and then
297          * free the old root.
298          */
299         xfs_allocbt_set_root(cur, newroot, -1);
300         error = xfs_allocbt_free_block(cur, bp);
301         if (error) {
302                 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
303                 return error;
304         }
305
306         XFS_BTREE_STATS_INC(cur, free);
307
308         xfs_btree_setbuf(cur, level, NULL);
309         cur->bc_nlevels--;
310
311         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
312         return 0;
313 }
314
315 #ifdef DEBUG
316 STATIC int
317 xfs_allocbt_keys_inorder(
318         struct xfs_btree_cur    *cur,
319         union xfs_btree_key     *k1,
320         union xfs_btree_key     *k2)
321 {
322         if (cur->bc_btnum == XFS_BTNUM_BNO) {
323                 return be32_to_cpu(k1->alloc.ar_startblock) <
324                        be32_to_cpu(k2->alloc.ar_startblock);
325         } else {
326                 return be32_to_cpu(k1->alloc.ar_blockcount) <
327                         be32_to_cpu(k2->alloc.ar_blockcount) ||
328                         (k1->alloc.ar_blockcount == k2->alloc.ar_blockcount &&
329                          be32_to_cpu(k1->alloc.ar_startblock) <
330                          be32_to_cpu(k2->alloc.ar_startblock));
331         }
332 }
333
334 STATIC int
335 xfs_allocbt_recs_inorder(
336         struct xfs_btree_cur    *cur,
337         union xfs_btree_rec     *r1,
338         union xfs_btree_rec     *r2)
339 {
340         if (cur->bc_btnum == XFS_BTNUM_BNO) {
341                 return be32_to_cpu(r1->alloc.ar_startblock) +
342                         be32_to_cpu(r1->alloc.ar_blockcount) <=
343                         be32_to_cpu(r2->alloc.ar_startblock);
344         } else {
345                 return be32_to_cpu(r1->alloc.ar_blockcount) <
346                         be32_to_cpu(r2->alloc.ar_blockcount) ||
347                         (r1->alloc.ar_blockcount == r2->alloc.ar_blockcount &&
348                          be32_to_cpu(r1->alloc.ar_startblock) <
349                          be32_to_cpu(r2->alloc.ar_startblock));
350         }
351 }
352 #endif  /* DEBUG */
353
354 #ifdef XFS_BTREE_TRACE
355 ktrace_t        *xfs_allocbt_trace_buf;
356
357 STATIC void
358 xfs_allocbt_trace_enter(
359         struct xfs_btree_cur    *cur,
360         const char              *func,
361         char                    *s,
362         int                     type,
363         int                     line,
364         __psunsigned_t          a0,
365         __psunsigned_t          a1,
366         __psunsigned_t          a2,
367         __psunsigned_t          a3,
368         __psunsigned_t          a4,
369         __psunsigned_t          a5,
370         __psunsigned_t          a6,
371         __psunsigned_t          a7,
372         __psunsigned_t          a8,
373         __psunsigned_t          a9,
374         __psunsigned_t          a10)
375 {
376         ktrace_enter(xfs_allocbt_trace_buf, (void *)(__psint_t)type,
377                 (void *)func, (void *)s, NULL, (void *)cur,
378                 (void *)a0, (void *)a1, (void *)a2, (void *)a3,
379                 (void *)a4, (void *)a5, (void *)a6, (void *)a7,
380                 (void *)a8, (void *)a9, (void *)a10);
381 }
382
383 STATIC void
384 xfs_allocbt_trace_cursor(
385         struct xfs_btree_cur    *cur,
386         __uint32_t              *s0,
387         __uint64_t              *l0,
388         __uint64_t              *l1)
389 {
390         *s0 = cur->bc_private.a.agno;
391         *l0 = cur->bc_rec.a.ar_startblock;
392         *l1 = cur->bc_rec.a.ar_blockcount;
393 }
394
395 STATIC void
396 xfs_allocbt_trace_key(
397         struct xfs_btree_cur    *cur,
398         union xfs_btree_key     *key,
399         __uint64_t              *l0,
400         __uint64_t              *l1)
401 {
402         *l0 = be32_to_cpu(key->alloc.ar_startblock);
403         *l1 = be32_to_cpu(key->alloc.ar_blockcount);
404 }
405
406 STATIC void
407 xfs_allocbt_trace_record(
408         struct xfs_btree_cur    *cur,
409         union xfs_btree_rec     *rec,
410         __uint64_t              *l0,
411         __uint64_t              *l1,
412         __uint64_t              *l2)
413 {
414         *l0 = be32_to_cpu(rec->alloc.ar_startblock);
415         *l1 = be32_to_cpu(rec->alloc.ar_blockcount);
416         *l2 = 0;
417 }
418 #endif /* XFS_BTREE_TRACE */
419
420 static const struct xfs_btree_ops xfs_allocbt_ops = {
421         .rec_len                = sizeof(xfs_alloc_rec_t),
422         .key_len                = sizeof(xfs_alloc_key_t),
423
424         .dup_cursor             = xfs_allocbt_dup_cursor,
425         .set_root               = xfs_allocbt_set_root,
426         .kill_root              = xfs_allocbt_kill_root,
427         .alloc_block            = xfs_allocbt_alloc_block,
428         .free_block             = xfs_allocbt_free_block,
429         .update_lastrec         = xfs_allocbt_update_lastrec,
430         .get_minrecs            = xfs_allocbt_get_minrecs,
431         .get_maxrecs            = xfs_allocbt_get_maxrecs,
432         .init_key_from_rec      = xfs_allocbt_init_key_from_rec,
433         .init_rec_from_key      = xfs_allocbt_init_rec_from_key,
434         .init_rec_from_cur      = xfs_allocbt_init_rec_from_cur,
435         .init_ptr_from_cur      = xfs_allocbt_init_ptr_from_cur,
436         .key_diff               = xfs_allocbt_key_diff,
437
438 #ifdef DEBUG
439         .keys_inorder           = xfs_allocbt_keys_inorder,
440         .recs_inorder           = xfs_allocbt_recs_inorder,
441 #endif
442
443 #ifdef XFS_BTREE_TRACE
444         .trace_enter            = xfs_allocbt_trace_enter,
445         .trace_cursor           = xfs_allocbt_trace_cursor,
446         .trace_key              = xfs_allocbt_trace_key,
447         .trace_record           = xfs_allocbt_trace_record,
448 #endif
449 };
450
451 /*
452  * Allocate a new allocation btree cursor.
453  */
454 struct xfs_btree_cur *                  /* new alloc btree cursor */
455 xfs_allocbt_init_cursor(
456         struct xfs_mount        *mp,            /* file system mount point */
457         struct xfs_trans        *tp,            /* transaction pointer */
458         struct xfs_buf          *agbp,          /* buffer for agf structure */
459         xfs_agnumber_t          agno,           /* allocation group number */
460         xfs_btnum_t             btnum)          /* btree identifier */
461 {
462         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
463         struct xfs_btree_cur    *cur;
464
465         ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
466
467         cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_SLEEP);
468
469         cur->bc_tp = tp;
470         cur->bc_mp = mp;
471         cur->bc_nlevels = be32_to_cpu(agf->agf_levels[btnum]);
472         cur->bc_btnum = btnum;
473         cur->bc_blocklog = mp->m_sb.sb_blocklog;
474
475         cur->bc_ops = &xfs_allocbt_ops;
476         if (btnum == XFS_BTNUM_CNT)
477                 cur->bc_flags = XFS_BTREE_LASTREC_UPDATE;
478
479         cur->bc_private.a.agbp = agbp;
480         cur->bc_private.a.agno = agno;
481
482         return cur;
483 }
484
485 /*
486  * Calculate number of records in an alloc btree block.
487  */
488 int
489 xfs_allocbt_maxrecs(
490         struct xfs_mount        *mp,
491         int                     blocklen,
492         int                     leaf)
493 {
494         blocklen -= XFS_ALLOC_BLOCK_LEN(mp);
495
496         if (leaf)
497                 return blocklen / sizeof(xfs_alloc_rec_t);
498         return blocklen / (sizeof(xfs_alloc_key_t) + sizeof(xfs_alloc_ptr_t));
499 }