diff options
Diffstat (limited to 'fs/xfs/xfs_alloc_btree.c')
-rw-r--r-- | fs/xfs/xfs_alloc_btree.c | 744 |
1 files changed, 59 insertions, 685 deletions
diff --git a/fs/xfs/xfs_alloc_btree.c b/fs/xfs/xfs_alloc_btree.c index f124ddd91c08..d256b51f913d 100644 --- a/fs/xfs/xfs_alloc_btree.c +++ b/fs/xfs/xfs_alloc_btree.c | |||
@@ -40,691 +40,6 @@ | |||
40 | #include "xfs_alloc.h" | 40 | #include "xfs_alloc.h" |
41 | #include "xfs_error.h" | 41 | #include "xfs_error.h" |
42 | 42 | ||
43 | /* | ||
44 | * Prototypes for internal functions. | ||
45 | */ | ||
46 | |||
47 | STATIC void xfs_alloc_log_block(xfs_trans_t *, xfs_buf_t *, int); | ||
48 | STATIC void xfs_alloc_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int); | ||
49 | STATIC void xfs_alloc_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int); | ||
50 | STATIC void xfs_alloc_log_recs(xfs_btree_cur_t *, xfs_buf_t *, int, int); | ||
51 | |||
52 | /* | ||
53 | * Internal functions. | ||
54 | */ | ||
55 | |||
56 | /* | ||
57 | * Single level of the xfs_alloc_delete record deletion routine. | ||
58 | * Delete record pointed to by cur/level. | ||
59 | * Remove the record from its block then rebalance the tree. | ||
60 | * Return 0 for error, 1 for done, 2 to go on to the next level. | ||
61 | */ | ||
62 | STATIC int /* error */ | ||
63 | xfs_alloc_delrec( | ||
64 | xfs_btree_cur_t *cur, /* btree cursor */ | ||
65 | int level, /* level removing record from */ | ||
66 | int *stat) /* fail/done/go-on */ | ||
67 | { | ||
68 | xfs_agf_t *agf; /* allocation group freelist header */ | ||
69 | xfs_alloc_block_t *block; /* btree block record/key lives in */ | ||
70 | xfs_agblock_t bno; /* btree block number */ | ||
71 | xfs_buf_t *bp; /* buffer for block */ | ||
72 | int error; /* error return value */ | ||
73 | int i; /* loop index */ | ||
74 | xfs_alloc_key_t key; /* kp points here if block is level 0 */ | ||
75 | xfs_agblock_t lbno; /* left block's block number */ | ||
76 | xfs_buf_t *lbp; /* left block's buffer pointer */ | ||
77 | xfs_alloc_block_t *left; /* left btree block */ | ||
78 | xfs_alloc_key_t *lkp=NULL; /* left block key pointer */ | ||
79 | xfs_alloc_ptr_t *lpp=NULL; /* left block address pointer */ | ||
80 | int lrecs=0; /* number of records in left block */ | ||
81 | xfs_alloc_rec_t *lrp; /* left block record pointer */ | ||
82 | xfs_mount_t *mp; /* mount structure */ | ||
83 | int ptr; /* index in btree block for this rec */ | ||
84 | xfs_agblock_t rbno; /* right block's block number */ | ||
85 | xfs_buf_t *rbp; /* right block's buffer pointer */ | ||
86 | xfs_alloc_block_t *right; /* right btree block */ | ||
87 | xfs_alloc_key_t *rkp; /* right block key pointer */ | ||
88 | xfs_alloc_ptr_t *rpp; /* right block address pointer */ | ||
89 | int rrecs=0; /* number of records in right block */ | ||
90 | int numrecs; | ||
91 | xfs_alloc_rec_t *rrp; /* right block record pointer */ | ||
92 | xfs_btree_cur_t *tcur; /* temporary btree cursor */ | ||
93 | |||
94 | /* | ||
95 | * Get the index of the entry being deleted, check for nothing there. | ||
96 | */ | ||
97 | ptr = cur->bc_ptrs[level]; | ||
98 | if (ptr == 0) { | ||
99 | *stat = 0; | ||
100 | return 0; | ||
101 | } | ||
102 | /* | ||
103 | * Get the buffer & block containing the record or key/ptr. | ||
104 | */ | ||
105 | bp = cur->bc_bufs[level]; | ||
106 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | ||
107 | #ifdef DEBUG | ||
108 | if ((error = xfs_btree_check_sblock(cur, block, level, bp))) | ||
109 | return error; | ||
110 | #endif | ||
111 | /* | ||
112 | * Fail if we're off the end of the block. | ||
113 | */ | ||
114 | numrecs = be16_to_cpu(block->bb_numrecs); | ||
115 | if (ptr > numrecs) { | ||
116 | *stat = 0; | ||
117 | return 0; | ||
118 | } | ||
119 | XFS_STATS_INC(xs_abt_delrec); | ||
120 | /* | ||
121 | * It's a nonleaf. Excise the key and ptr being deleted, by | ||
122 | * sliding the entries past them down one. | ||
123 | * Log the changed areas of the block. | ||
124 | */ | ||
125 | if (level > 0) { | ||
126 | lkp = XFS_ALLOC_KEY_ADDR(block, 1, cur); | ||
127 | lpp = XFS_ALLOC_PTR_ADDR(block, 1, cur); | ||
128 | #ifdef DEBUG | ||
129 | for (i = ptr; i < numrecs; i++) { | ||
130 | if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(lpp[i]), level))) | ||
131 | return error; | ||
132 | } | ||
133 | #endif | ||
134 | if (ptr < numrecs) { | ||
135 | memmove(&lkp[ptr - 1], &lkp[ptr], | ||
136 | (numrecs - ptr) * sizeof(*lkp)); | ||
137 | memmove(&lpp[ptr - 1], &lpp[ptr], | ||
138 | (numrecs - ptr) * sizeof(*lpp)); | ||
139 | xfs_alloc_log_ptrs(cur, bp, ptr, numrecs - 1); | ||
140 | xfs_alloc_log_keys(cur, bp, ptr, numrecs - 1); | ||
141 | } | ||
142 | } | ||
143 | /* | ||
144 | * It's a leaf. Excise the record being deleted, by sliding the | ||
145 | * entries past it down one. Log the changed areas of the block. | ||
146 | */ | ||
147 | else { | ||
148 | lrp = XFS_ALLOC_REC_ADDR(block, 1, cur); | ||
149 | if (ptr < numrecs) { | ||
150 | memmove(&lrp[ptr - 1], &lrp[ptr], | ||
151 | (numrecs - ptr) * sizeof(*lrp)); | ||
152 | xfs_alloc_log_recs(cur, bp, ptr, numrecs - 1); | ||
153 | } | ||
154 | /* | ||
155 | * If it's the first record in the block, we'll need a key | ||
156 | * structure to pass up to the next level (updkey). | ||
157 | */ | ||
158 | if (ptr == 1) { | ||
159 | key.ar_startblock = lrp->ar_startblock; | ||
160 | key.ar_blockcount = lrp->ar_blockcount; | ||
161 | lkp = &key; | ||
162 | } | ||
163 | } | ||
164 | /* | ||
165 | * Decrement and log the number of entries in the block. | ||
166 | */ | ||
167 | numrecs--; | ||
168 | block->bb_numrecs = cpu_to_be16(numrecs); | ||
169 | xfs_alloc_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS); | ||
170 | /* | ||
171 | * See if the longest free extent in the allocation group was | ||
172 | * changed by this operation. True if it's the by-size btree, and | ||
173 | * this is the leaf level, and there is no right sibling block, | ||
174 | * and this was the last record. | ||
175 | */ | ||
176 | agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | ||
177 | mp = cur->bc_mp; | ||
178 | |||
179 | if (level == 0 && | ||
180 | cur->bc_btnum == XFS_BTNUM_CNT && | ||
181 | be32_to_cpu(block->bb_rightsib) == NULLAGBLOCK && | ||
182 | ptr > numrecs) { | ||
183 | ASSERT(ptr == numrecs + 1); | ||
184 | /* | ||
185 | * There are still records in the block. Grab the size | ||
186 | * from the last one. | ||
187 | */ | ||
188 | if (numrecs) { | ||
189 | rrp = XFS_ALLOC_REC_ADDR(block, numrecs, cur); | ||
190 | agf->agf_longest = rrp->ar_blockcount; | ||
191 | } | ||
192 | /* | ||
193 | * No free extents left. | ||
194 | */ | ||
195 | else | ||
196 | agf->agf_longest = 0; | ||
197 | mp->m_perag[be32_to_cpu(agf->agf_seqno)].pagf_longest = | ||
198 | be32_to_cpu(agf->agf_longest); | ||
199 | xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, | ||
200 | XFS_AGF_LONGEST); | ||
201 | } | ||
202 | /* | ||
203 | * Is this the root level? If so, we're almost done. | ||
204 | */ | ||
205 | if (level == cur->bc_nlevels - 1) { | ||
206 | /* | ||
207 | * If this is the root level, | ||
208 | * and there's only one entry left, | ||
209 | * and it's NOT the leaf level, | ||
210 | * then we can get rid of this level. | ||
211 | */ | ||
212 | if (numrecs == 1 && level > 0) { | ||
213 | /* | ||
214 | * lpp is still set to the first pointer in the block. | ||
215 | * Make it the new root of the btree. | ||
216 | */ | ||
217 | bno = be32_to_cpu(agf->agf_roots[cur->bc_btnum]); | ||
218 | agf->agf_roots[cur->bc_btnum] = *lpp; | ||
219 | be32_add_cpu(&agf->agf_levels[cur->bc_btnum], -1); | ||
220 | mp->m_perag[be32_to_cpu(agf->agf_seqno)].pagf_levels[cur->bc_btnum]--; | ||
221 | /* | ||
222 | * Put this buffer/block on the ag's freelist. | ||
223 | */ | ||
224 | error = xfs_alloc_put_freelist(cur->bc_tp, | ||
225 | cur->bc_private.a.agbp, NULL, bno, 1); | ||
226 | if (error) | ||
227 | return error; | ||
228 | /* | ||
229 | * Since blocks move to the free list without the | ||
230 | * coordination used in xfs_bmap_finish, we can't allow | ||
231 | * block to be available for reallocation and | ||
232 | * non-transaction writing (user data) until we know | ||
233 | * that the transaction that moved it to the free list | ||
234 | * is permanently on disk. We track the blocks by | ||
235 | * declaring these blocks as "busy"; the busy list is | ||
236 | * maintained on a per-ag basis and each transaction | ||
237 | * records which entries should be removed when the | ||
238 | * iclog commits to disk. If a busy block is | ||
239 | * allocated, the iclog is pushed up to the LSN | ||
240 | * that freed the block. | ||
241 | */ | ||
242 | xfs_alloc_mark_busy(cur->bc_tp, | ||
243 | be32_to_cpu(agf->agf_seqno), bno, 1); | ||
244 | |||
245 | xfs_trans_agbtree_delta(cur->bc_tp, -1); | ||
246 | xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, | ||
247 | XFS_AGF_ROOTS | XFS_AGF_LEVELS); | ||
248 | /* | ||
249 | * Update the cursor so there's one fewer level. | ||
250 | */ | ||
251 | xfs_btree_setbuf(cur, level, NULL); | ||
252 | cur->bc_nlevels--; | ||
253 | } else if (level > 0 && | ||
254 | (error = xfs_btree_decrement(cur, level, &i))) | ||
255 | return error; | ||
256 | *stat = 1; | ||
257 | return 0; | ||
258 | } | ||
259 | /* | ||
260 | * If we deleted the leftmost entry in the block, update the | ||
261 | * key values above us in the tree. | ||
262 | */ | ||
263 | if (ptr == 1 && (error = xfs_btree_updkey(cur, (union xfs_btree_key *)lkp, level + 1))) | ||
264 | return error; | ||
265 | /* | ||
266 | * If the number of records remaining in the block is at least | ||
267 | * the minimum, we're done. | ||
268 | */ | ||
269 | if (numrecs >= XFS_ALLOC_BLOCK_MINRECS(level, cur)) { | ||
270 | if (level > 0 && (error = xfs_btree_decrement(cur, level, &i))) | ||
271 | return error; | ||
272 | *stat = 1; | ||
273 | return 0; | ||
274 | } | ||
275 | /* | ||
276 | * Otherwise, we have to move some records around to keep the | ||
277 | * tree balanced. Look at the left and right sibling blocks to | ||
278 | * see if we can re-balance by moving only one record. | ||
279 | */ | ||
280 | rbno = be32_to_cpu(block->bb_rightsib); | ||
281 | lbno = be32_to_cpu(block->bb_leftsib); | ||
282 | bno = NULLAGBLOCK; | ||
283 | ASSERT(rbno != NULLAGBLOCK || lbno != NULLAGBLOCK); | ||
284 | /* | ||
285 | * Duplicate the cursor so our btree manipulations here won't | ||
286 | * disrupt the next level up. | ||
287 | */ | ||
288 | if ((error = xfs_btree_dup_cursor(cur, &tcur))) | ||
289 | return error; | ||
290 | /* | ||
291 | * If there's a right sibling, see if it's ok to shift an entry | ||
292 | * out of it. | ||
293 | */ | ||
294 | if (rbno != NULLAGBLOCK) { | ||
295 | /* | ||
296 | * Move the temp cursor to the last entry in the next block. | ||
297 | * Actually any entry but the first would suffice. | ||
298 | */ | ||
299 | i = xfs_btree_lastrec(tcur, level); | ||
300 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | ||
301 | if ((error = xfs_btree_increment(tcur, level, &i))) | ||
302 | goto error0; | ||
303 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | ||
304 | i = xfs_btree_lastrec(tcur, level); | ||
305 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | ||
306 | /* | ||
307 | * Grab a pointer to the block. | ||
308 | */ | ||
309 | rbp = tcur->bc_bufs[level]; | ||
310 | right = XFS_BUF_TO_ALLOC_BLOCK(rbp); | ||
311 | #ifdef DEBUG | ||
312 | if ((error = xfs_btree_check_sblock(cur, right, level, rbp))) | ||
313 | goto error0; | ||
314 | #endif | ||
315 | /* | ||
316 | * Grab the current block number, for future use. | ||
317 | */ | ||
318 | bno = be32_to_cpu(right->bb_leftsib); | ||
319 | /* | ||
320 | * If right block is full enough so that removing one entry | ||
321 | * won't make it too empty, and left-shifting an entry out | ||
322 | * of right to us works, we're done. | ||
323 | */ | ||
324 | if (be16_to_cpu(right->bb_numrecs) - 1 >= | ||
325 | XFS_ALLOC_BLOCK_MINRECS(level, cur)) { | ||
326 | if ((error = xfs_btree_lshift(tcur, level, &i))) | ||
327 | goto error0; | ||
328 | if (i) { | ||
329 | ASSERT(be16_to_cpu(block->bb_numrecs) >= | ||
330 | XFS_ALLOC_BLOCK_MINRECS(level, cur)); | ||
331 | xfs_btree_del_cursor(tcur, | ||
332 | XFS_BTREE_NOERROR); | ||
333 | if (level > 0 && | ||
334 | (error = xfs_btree_decrement(cur, level, | ||
335 | &i))) | ||
336 | return error; | ||
337 | *stat = 1; | ||
338 | return 0; | ||
339 | } | ||
340 | } | ||
341 | /* | ||
342 | * Otherwise, grab the number of records in right for | ||
343 | * future reference, and fix up the temp cursor to point | ||
344 | * to our block again (last record). | ||
345 | */ | ||
346 | rrecs = be16_to_cpu(right->bb_numrecs); | ||
347 | if (lbno != NULLAGBLOCK) { | ||
348 | i = xfs_btree_firstrec(tcur, level); | ||
349 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | ||
350 | if ((error = xfs_btree_decrement(tcur, level, &i))) | ||
351 | goto error0; | ||
352 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | ||
353 | } | ||
354 | } | ||
355 | /* | ||
356 | * If there's a left sibling, see if it's ok to shift an entry | ||
357 | * out of it. | ||
358 | */ | ||
359 | if (lbno != NULLAGBLOCK) { | ||
360 | /* | ||
361 | * Move the temp cursor to the first entry in the | ||
362 | * previous block. | ||
363 | */ | ||
364 | i = xfs_btree_firstrec(tcur, level); | ||
365 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | ||
366 | if ((error = xfs_btree_decrement(tcur, level, &i))) | ||
367 | goto error0; | ||
368 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | ||
369 | xfs_btree_firstrec(tcur, level); | ||
370 | /* | ||
371 | * Grab a pointer to the block. | ||
372 | */ | ||
373 | lbp = tcur->bc_bufs[level]; | ||
374 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | ||
375 | #ifdef DEBUG | ||
376 | if ((error = xfs_btree_check_sblock(cur, left, level, lbp))) | ||
377 | goto error0; | ||
378 | #endif | ||
379 | /* | ||
380 | * Grab the current block number, for future use. | ||
381 | */ | ||
382 | bno = be32_to_cpu(left->bb_rightsib); | ||
383 | /* | ||
384 | * If left block is full enough so that removing one entry | ||
385 | * won't make it too empty, and right-shifting an entry out | ||
386 | * of left to us works, we're done. | ||
387 | */ | ||
388 | if (be16_to_cpu(left->bb_numrecs) - 1 >= | ||
389 | XFS_ALLOC_BLOCK_MINRECS(level, cur)) { | ||
390 | if ((error = xfs_btree_rshift(tcur, level, &i))) | ||
391 | goto error0; | ||
392 | if (i) { | ||
393 | ASSERT(be16_to_cpu(block->bb_numrecs) >= | ||
394 | XFS_ALLOC_BLOCK_MINRECS(level, cur)); | ||
395 | xfs_btree_del_cursor(tcur, | ||
396 | XFS_BTREE_NOERROR); | ||
397 | if (level == 0) | ||
398 | cur->bc_ptrs[0]++; | ||
399 | *stat = 1; | ||
400 | return 0; | ||
401 | } | ||
402 | } | ||
403 | /* | ||
404 | * Otherwise, grab the number of records in right for | ||
405 | * future reference. | ||
406 | */ | ||
407 | lrecs = be16_to_cpu(left->bb_numrecs); | ||
408 | } | ||
409 | /* | ||
410 | * Delete the temp cursor, we're done with it. | ||
411 | */ | ||
412 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | ||
413 | /* | ||
414 | * If here, we need to do a join to keep the tree balanced. | ||
415 | */ | ||
416 | ASSERT(bno != NULLAGBLOCK); | ||
417 | /* | ||
418 | * See if we can join with the left neighbor block. | ||
419 | */ | ||
420 | if (lbno != NULLAGBLOCK && | ||
421 | lrecs + numrecs <= XFS_ALLOC_BLOCK_MAXRECS(level, cur)) { | ||
422 | /* | ||
423 | * Set "right" to be the starting block, | ||
424 | * "left" to be the left neighbor. | ||
425 | */ | ||
426 | rbno = bno; | ||
427 | right = block; | ||
428 | rrecs = be16_to_cpu(right->bb_numrecs); | ||
429 | rbp = bp; | ||
430 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, | ||
431 | cur->bc_private.a.agno, lbno, 0, &lbp, | ||
432 | XFS_ALLOC_BTREE_REF))) | ||
433 | return error; | ||
434 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | ||
435 | lrecs = be16_to_cpu(left->bb_numrecs); | ||
436 | if ((error = xfs_btree_check_sblock(cur, left, level, lbp))) | ||
437 | return error; | ||
438 | } | ||
439 | /* | ||
440 | * If that won't work, see if we can join with the right neighbor block. | ||
441 | */ | ||
442 | else if (rbno != NULLAGBLOCK && | ||
443 | rrecs + numrecs <= XFS_ALLOC_BLOCK_MAXRECS(level, cur)) { | ||
444 | /* | ||
445 | * Set "left" to be the starting block, | ||
446 | * "right" to be the right neighbor. | ||
447 | */ | ||
448 | lbno = bno; | ||
449 | left = block; | ||
450 | lrecs = be16_to_cpu(left->bb_numrecs); | ||
451 | lbp = bp; | ||
452 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, | ||
453 | cur->bc_private.a.agno, rbno, 0, &rbp, | ||
454 | XFS_ALLOC_BTREE_REF))) | ||
455 | return error; | ||
456 | right = XFS_BUF_TO_ALLOC_BLOCK(rbp); | ||
457 | rrecs = be16_to_cpu(right->bb_numrecs); | ||
458 | if ((error = xfs_btree_check_sblock(cur, right, level, rbp))) | ||
459 | return error; | ||
460 | } | ||
461 | /* | ||
462 | * Otherwise, we can't fix the imbalance. | ||
463 | * Just return. This is probably a logic error, but it's not fatal. | ||
464 | */ | ||
465 | else { | ||
466 | if (level > 0 && (error = xfs_btree_decrement(cur, level, &i))) | ||
467 | return error; | ||
468 | *stat = 1; | ||
469 | return 0; | ||
470 | } | ||
471 | /* | ||
472 | * We're now going to join "left" and "right" by moving all the stuff | ||
473 | * in "right" to "left" and deleting "right". | ||
474 | */ | ||
475 | if (level > 0) { | ||
476 | /* | ||
477 | * It's a non-leaf. Move keys and pointers. | ||
478 | */ | ||
479 | lkp = XFS_ALLOC_KEY_ADDR(left, lrecs + 1, cur); | ||
480 | lpp = XFS_ALLOC_PTR_ADDR(left, lrecs + 1, cur); | ||
481 | rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur); | ||
482 | rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur); | ||
483 | #ifdef DEBUG | ||
484 | for (i = 0; i < rrecs; i++) { | ||
485 | if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i]), level))) | ||
486 | return error; | ||
487 | } | ||
488 | #endif | ||
489 | memcpy(lkp, rkp, rrecs * sizeof(*lkp)); | ||
490 | memcpy(lpp, rpp, rrecs * sizeof(*lpp)); | ||
491 | xfs_alloc_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs); | ||
492 | xfs_alloc_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs); | ||
493 | } else { | ||
494 | /* | ||
495 | * It's a leaf. Move records. | ||
496 | */ | ||
497 | lrp = XFS_ALLOC_REC_ADDR(left, lrecs + 1, cur); | ||
498 | rrp = XFS_ALLOC_REC_ADDR(right, 1, cur); | ||
499 | memcpy(lrp, rrp, rrecs * sizeof(*lrp)); | ||
500 | xfs_alloc_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs); | ||
501 | } | ||
502 | /* | ||
503 | * If we joined with the left neighbor, set the buffer in the | ||
504 | * cursor to the left block, and fix up the index. | ||
505 | */ | ||
506 | if (bp != lbp) { | ||
507 | xfs_btree_setbuf(cur, level, lbp); | ||
508 | cur->bc_ptrs[level] += lrecs; | ||
509 | } | ||
510 | /* | ||
511 | * If we joined with the right neighbor and there's a level above | ||
512 | * us, increment the cursor at that level. | ||
513 | */ | ||
514 | else if (level + 1 < cur->bc_nlevels && | ||
515 | (error = xfs_btree_increment(cur, level + 1, &i))) | ||
516 | return error; | ||
517 | /* | ||
518 | * Fix up the number of records in the surviving block. | ||
519 | */ | ||
520 | lrecs += rrecs; | ||
521 | left->bb_numrecs = cpu_to_be16(lrecs); | ||
522 | /* | ||
523 | * Fix up the right block pointer in the surviving block, and log it. | ||
524 | */ | ||
525 | left->bb_rightsib = right->bb_rightsib; | ||
526 | xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB); | ||
527 | /* | ||
528 | * If there is a right sibling now, make it point to the | ||
529 | * remaining block. | ||
530 | */ | ||
531 | if (be32_to_cpu(left->bb_rightsib) != NULLAGBLOCK) { | ||
532 | xfs_alloc_block_t *rrblock; | ||
533 | xfs_buf_t *rrbp; | ||
534 | |||
535 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, | ||
536 | cur->bc_private.a.agno, be32_to_cpu(left->bb_rightsib), 0, | ||
537 | &rrbp, XFS_ALLOC_BTREE_REF))) | ||
538 | return error; | ||
539 | rrblock = XFS_BUF_TO_ALLOC_BLOCK(rrbp); | ||
540 | if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp))) | ||
541 | return error; | ||
542 | rrblock->bb_leftsib = cpu_to_be32(lbno); | ||
543 | xfs_alloc_log_block(cur->bc_tp, rrbp, XFS_BB_LEFTSIB); | ||
544 | } | ||
545 | /* | ||
546 | * Free the deleting block by putting it on the freelist. | ||
547 | */ | ||
548 | error = xfs_alloc_put_freelist(cur->bc_tp, | ||
549 | cur->bc_private.a.agbp, NULL, rbno, 1); | ||
550 | if (error) | ||
551 | return error; | ||
552 | /* | ||
553 | * Since blocks move to the free list without the coordination | ||
554 | * used in xfs_bmap_finish, we can't allow block to be available | ||
555 | * for reallocation and non-transaction writing (user data) | ||
556 | * until we know that the transaction that moved it to the free | ||
557 | * list is permanently on disk. We track the blocks by declaring | ||
558 | * these blocks as "busy"; the busy list is maintained on a | ||
559 | * per-ag basis and each transaction records which entries | ||
560 | * should be removed when the iclog commits to disk. If a | ||
561 | * busy block is allocated, the iclog is pushed up to the | ||
562 | * LSN that freed the block. | ||
563 | */ | ||
564 | xfs_alloc_mark_busy(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1); | ||
565 | xfs_trans_agbtree_delta(cur->bc_tp, -1); | ||
566 | |||
567 | /* | ||
568 | * Adjust the current level's cursor so that we're left referring | ||
569 | * to the right node, after we're done. | ||
570 | * If this leaves the ptr value 0 our caller will fix it up. | ||
571 | */ | ||
572 | if (level > 0) | ||
573 | cur->bc_ptrs[level]--; | ||
574 | /* | ||
575 | * Return value means the next level up has something to do. | ||
576 | */ | ||
577 | *stat = 2; | ||
578 | return 0; | ||
579 | |||
580 | error0: | ||
581 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | ||
582 | return error; | ||
583 | } | ||
584 | |||
585 | /* | ||
586 | * Log header fields from a btree block. | ||
587 | */ | ||
588 | STATIC void | ||
589 | xfs_alloc_log_block( | ||
590 | xfs_trans_t *tp, /* transaction pointer */ | ||
591 | xfs_buf_t *bp, /* buffer containing btree block */ | ||
592 | int fields) /* mask of fields: XFS_BB_... */ | ||
593 | { | ||
594 | int first; /* first byte offset logged */ | ||
595 | int last; /* last byte offset logged */ | ||
596 | static const short offsets[] = { /* table of offsets */ | ||
597 | offsetof(xfs_alloc_block_t, bb_magic), | ||
598 | offsetof(xfs_alloc_block_t, bb_level), | ||
599 | offsetof(xfs_alloc_block_t, bb_numrecs), | ||
600 | offsetof(xfs_alloc_block_t, bb_leftsib), | ||
601 | offsetof(xfs_alloc_block_t, bb_rightsib), | ||
602 | sizeof(xfs_alloc_block_t) | ||
603 | }; | ||
604 | |||
605 | xfs_btree_offsets(fields, offsets, XFS_BB_NUM_BITS, &first, &last); | ||
606 | xfs_trans_log_buf(tp, bp, first, last); | ||
607 | } | ||
608 | |||
609 | /* | ||
610 | * Log keys from a btree block (nonleaf). | ||
611 | */ | ||
612 | STATIC void | ||
613 | xfs_alloc_log_keys( | ||
614 | xfs_btree_cur_t *cur, /* btree cursor */ | ||
615 | xfs_buf_t *bp, /* buffer containing btree block */ | ||
616 | int kfirst, /* index of first key to log */ | ||
617 | int klast) /* index of last key to log */ | ||
618 | { | ||
619 | xfs_alloc_block_t *block; /* btree block to log from */ | ||
620 | int first; /* first byte offset logged */ | ||
621 | xfs_alloc_key_t *kp; /* key pointer in btree block */ | ||
622 | int last; /* last byte offset logged */ | ||
623 | |||
624 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | ||
625 | kp = XFS_ALLOC_KEY_ADDR(block, 1, cur); | ||
626 | first = (int)((xfs_caddr_t)&kp[kfirst - 1] - (xfs_caddr_t)block); | ||
627 | last = (int)(((xfs_caddr_t)&kp[klast] - 1) - (xfs_caddr_t)block); | ||
628 | xfs_trans_log_buf(cur->bc_tp, bp, first, last); | ||
629 | } | ||
630 | |||
631 | /* | ||
632 | * Log block pointer fields from a btree block (nonleaf). | ||
633 | */ | ||
634 | STATIC void | ||
635 | xfs_alloc_log_ptrs( | ||
636 | xfs_btree_cur_t *cur, /* btree cursor */ | ||
637 | xfs_buf_t *bp, /* buffer containing btree block */ | ||
638 | int pfirst, /* index of first pointer to log */ | ||
639 | int plast) /* index of last pointer to log */ | ||
640 | { | ||
641 | xfs_alloc_block_t *block; /* btree block to log from */ | ||
642 | int first; /* first byte offset logged */ | ||
643 | int last; /* last byte offset logged */ | ||
644 | xfs_alloc_ptr_t *pp; /* block-pointer pointer in btree blk */ | ||
645 | |||
646 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | ||
647 | pp = XFS_ALLOC_PTR_ADDR(block, 1, cur); | ||
648 | first = (int)((xfs_caddr_t)&pp[pfirst - 1] - (xfs_caddr_t)block); | ||
649 | last = (int)(((xfs_caddr_t)&pp[plast] - 1) - (xfs_caddr_t)block); | ||
650 | xfs_trans_log_buf(cur->bc_tp, bp, first, last); | ||
651 | } | ||
652 | |||
653 | /* | ||
654 | * Log records from a btree block (leaf). | ||
655 | */ | ||
656 | STATIC void | ||
657 | xfs_alloc_log_recs( | ||
658 | xfs_btree_cur_t *cur, /* btree cursor */ | ||
659 | xfs_buf_t *bp, /* buffer containing btree block */ | ||
660 | int rfirst, /* index of first record to log */ | ||
661 | int rlast) /* index of last record to log */ | ||
662 | { | ||
663 | xfs_alloc_block_t *block; /* btree block to log from */ | ||
664 | int first; /* first byte offset logged */ | ||
665 | int last; /* last byte offset logged */ | ||
666 | xfs_alloc_rec_t *rp; /* record pointer for btree block */ | ||
667 | |||
668 | |||
669 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | ||
670 | rp = XFS_ALLOC_REC_ADDR(block, 1, cur); | ||
671 | #ifdef DEBUG | ||
672 | { | ||
673 | xfs_agf_t *agf; | ||
674 | xfs_alloc_rec_t *p; | ||
675 | |||
676 | agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | ||
677 | for (p = &rp[rfirst - 1]; p <= &rp[rlast - 1]; p++) | ||
678 | ASSERT(be32_to_cpu(p->ar_startblock) + | ||
679 | be32_to_cpu(p->ar_blockcount) <= | ||
680 | be32_to_cpu(agf->agf_length)); | ||
681 | } | ||
682 | #endif | ||
683 | first = (int)((xfs_caddr_t)&rp[rfirst - 1] - (xfs_caddr_t)block); | ||
684 | last = (int)(((xfs_caddr_t)&rp[rlast] - 1) - (xfs_caddr_t)block); | ||
685 | xfs_trans_log_buf(cur->bc_tp, bp, first, last); | ||
686 | } | ||
687 | |||
688 | |||
689 | /* | ||
690 | * Externally visible routines. | ||
691 | */ | ||
692 | |||
693 | /* | ||
694 | * Delete the record pointed to by cur. | ||
695 | * The cursor refers to the place where the record was (could be inserted) | ||
696 | * when the operation returns. | ||
697 | */ | ||
698 | int /* error */ | ||
699 | xfs_alloc_delete( | ||
700 | xfs_btree_cur_t *cur, /* btree cursor */ | ||
701 | int *stat) /* success/failure */ | ||
702 | { | ||
703 | int error; /* error return value */ | ||
704 | int i; /* result code */ | ||
705 | int level; /* btree level */ | ||
706 | |||
707 | /* | ||
708 | * Go up the tree, starting at leaf level. | ||
709 | * If 2 is returned then a join was done; go to the next level. | ||
710 | * Otherwise we are done. | ||
711 | */ | ||
712 | for (level = 0, i = 2; i == 2; level++) { | ||
713 | if ((error = xfs_alloc_delrec(cur, level, &i))) | ||
714 | return error; | ||
715 | } | ||
716 | if (i == 0) { | ||
717 | for (level = 1; level < cur->bc_nlevels; level++) { | ||
718 | if (cur->bc_ptrs[level] == 0) { | ||
719 | if ((error = xfs_btree_decrement(cur, level, &i))) | ||
720 | return error; | ||
721 | break; | ||
722 | } | ||
723 | } | ||
724 | } | ||
725 | *stat = i; | ||
726 | return 0; | ||
727 | } | ||
728 | 43 | ||
729 | /* | 44 | /* |
730 | * Get the data from the pointed-to record. | 45 | * Get the data from the pointed-to record. |
@@ -879,6 +194,7 @@ xfs_allocbt_update_lastrec( | |||
879 | struct xfs_agf *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | 194 | struct xfs_agf *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); |
880 | xfs_agnumber_t seqno = be32_to_cpu(agf->agf_seqno); | 195 | xfs_agnumber_t seqno = be32_to_cpu(agf->agf_seqno); |
881 | __be32 len; | 196 | __be32 len; |
197 | int numrecs; | ||
882 | 198 | ||
883 | ASSERT(cur->bc_btnum == XFS_BTNUM_CNT); | 199 | ASSERT(cur->bc_btnum == XFS_BTNUM_CNT); |
884 | 200 | ||
@@ -898,6 +214,22 @@ xfs_allocbt_update_lastrec( | |||
898 | return; | 214 | return; |
899 | len = rec->alloc.ar_blockcount; | 215 | len = rec->alloc.ar_blockcount; |
900 | break; | 216 | break; |
217 | case LASTREC_DELREC: | ||
218 | numrecs = xfs_btree_get_numrecs(block); | ||
219 | if (ptr <= numrecs) | ||
220 | return; | ||
221 | ASSERT(ptr == numrecs + 1); | ||
222 | |||
223 | if (numrecs) { | ||
224 | xfs_alloc_rec_t *rrp; | ||
225 | |||
226 | rrp = XFS_ALLOC_REC_ADDR(block, numrecs, cur); | ||
227 | len = rrp->ar_blockcount; | ||
228 | } else { | ||
229 | len = 0; | ||
230 | } | ||
231 | |||
232 | break; | ||
901 | default: | 233 | default: |
902 | ASSERT(0); | 234 | ASSERT(0); |
903 | return; | 235 | return; |
@@ -909,6 +241,14 @@ xfs_allocbt_update_lastrec( | |||
909 | } | 241 | } |
910 | 242 | ||
911 | STATIC int | 243 | STATIC int |
244 | xfs_allocbt_get_minrecs( | ||
245 | struct xfs_btree_cur *cur, | ||
246 | int level) | ||
247 | { | ||
248 | return cur->bc_mp->m_alloc_mnr[level != 0]; | ||
249 | } | ||
250 | |||
251 | STATIC int | ||
912 | xfs_allocbt_get_maxrecs( | 252 | xfs_allocbt_get_maxrecs( |
913 | struct xfs_btree_cur *cur, | 253 | struct xfs_btree_cur *cur, |
914 | int level) | 254 | int level) |
@@ -983,6 +323,38 @@ xfs_allocbt_key_diff( | |||
983 | return (__int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock; | 323 | return (__int64_t)be32_to_cpu(kp->ar_startblock) - rec->ar_startblock; |
984 | } | 324 | } |
985 | 325 | ||
326 | STATIC int | ||
327 | xfs_allocbt_kill_root( | ||
328 | struct xfs_btree_cur *cur, | ||
329 | struct xfs_buf *bp, | ||
330 | int level, | ||
331 | union xfs_btree_ptr *newroot) | ||
332 | { | ||
333 | int error; | ||
334 | |||
335 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | ||
336 | XFS_BTREE_STATS_INC(cur, killroot); | ||
337 | |||
338 | /* | ||
339 | * Update the root pointer, decreasing the level by 1 and then | ||
340 | * free the old root. | ||
341 | */ | ||
342 | xfs_allocbt_set_root(cur, newroot, -1); | ||
343 | error = xfs_allocbt_free_block(cur, bp); | ||
344 | if (error) { | ||
345 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | ||
346 | return error; | ||
347 | } | ||
348 | |||
349 | XFS_BTREE_STATS_INC(cur, free); | ||
350 | |||
351 | xfs_btree_setbuf(cur, level, NULL); | ||
352 | cur->bc_nlevels--; | ||
353 | |||
354 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | ||
355 | return 0; | ||
356 | } | ||
357 | |||
986 | #ifdef XFS_BTREE_TRACE | 358 | #ifdef XFS_BTREE_TRACE |
987 | ktrace_t *xfs_allocbt_trace_buf; | 359 | ktrace_t *xfs_allocbt_trace_buf; |
988 | 360 | ||
@@ -1055,9 +427,11 @@ static const struct xfs_btree_ops xfs_allocbt_ops = { | |||
1055 | 427 | ||
1056 | .dup_cursor = xfs_allocbt_dup_cursor, | 428 | .dup_cursor = xfs_allocbt_dup_cursor, |
1057 | .set_root = xfs_allocbt_set_root, | 429 | .set_root = xfs_allocbt_set_root, |
430 | .kill_root = xfs_allocbt_kill_root, | ||
1058 | .alloc_block = xfs_allocbt_alloc_block, | 431 | .alloc_block = xfs_allocbt_alloc_block, |
1059 | .free_block = xfs_allocbt_free_block, | 432 | .free_block = xfs_allocbt_free_block, |
1060 | .update_lastrec = xfs_allocbt_update_lastrec, | 433 | .update_lastrec = xfs_allocbt_update_lastrec, |
434 | .get_minrecs = xfs_allocbt_get_minrecs, | ||
1061 | .get_maxrecs = xfs_allocbt_get_maxrecs, | 435 | .get_maxrecs = xfs_allocbt_get_maxrecs, |
1062 | .init_key_from_rec = xfs_allocbt_init_key_from_rec, | 436 | .init_key_from_rec = xfs_allocbt_init_key_from_rec, |
1063 | .init_rec_from_key = xfs_allocbt_init_rec_from_key, | 437 | .init_rec_from_key = xfs_allocbt_init_rec_from_key, |