blob: 3ce2645508ae70ea826cc3b77cb2aeeaa1830785 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
Nathan Scotta844f452005-11-02 14:38:42 +110027#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_dmapi.h"
29#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_alloc_btree.h"
32#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_dir2_sf.h"
34#include "xfs_attr_sf.h"
35#include "xfs_dinode.h"
36#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_btree.h"
38#include "xfs_ialloc.h"
39#include "xfs_alloc.h"
40#include "xfs_error.h"
41
42/*
43 * Prototypes for internal functions.
44 */
45
46STATIC void xfs_alloc_log_block(xfs_trans_t *, xfs_buf_t *, int);
47STATIC void xfs_alloc_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int);
48STATIC void xfs_alloc_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
49STATIC void xfs_alloc_log_recs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
50STATIC int xfs_alloc_lshift(xfs_btree_cur_t *, int, int *);
51STATIC int xfs_alloc_newroot(xfs_btree_cur_t *, int *);
52STATIC int xfs_alloc_rshift(xfs_btree_cur_t *, int, int *);
53STATIC int xfs_alloc_split(xfs_btree_cur_t *, int, xfs_agblock_t *,
54 xfs_alloc_key_t *, xfs_btree_cur_t **, int *);
55STATIC int xfs_alloc_updkey(xfs_btree_cur_t *, xfs_alloc_key_t *, int);
56
57/*
58 * Internal functions.
59 */
60
61/*
62 * Single level of the xfs_alloc_delete record deletion routine.
63 * Delete record pointed to by cur/level.
64 * Remove the record from its block then rebalance the tree.
65 * Return 0 for error, 1 for done, 2 to go on to the next level.
66 */
67STATIC int /* error */
68xfs_alloc_delrec(
69 xfs_btree_cur_t *cur, /* btree cursor */
70 int level, /* level removing record from */
71 int *stat) /* fail/done/go-on */
72{
73 xfs_agf_t *agf; /* allocation group freelist header */
74 xfs_alloc_block_t *block; /* btree block record/key lives in */
75 xfs_agblock_t bno; /* btree block number */
76 xfs_buf_t *bp; /* buffer for block */
77 int error; /* error return value */
78 int i; /* loop index */
79 xfs_alloc_key_t key; /* kp points here if block is level 0 */
80 xfs_agblock_t lbno; /* left block's block number */
81 xfs_buf_t *lbp; /* left block's buffer pointer */
82 xfs_alloc_block_t *left; /* left btree block */
83 xfs_alloc_key_t *lkp=NULL; /* left block key pointer */
84 xfs_alloc_ptr_t *lpp=NULL; /* left block address pointer */
85 int lrecs=0; /* number of records in left block */
86 xfs_alloc_rec_t *lrp; /* left block record pointer */
87 xfs_mount_t *mp; /* mount structure */
88 int ptr; /* index in btree block for this rec */
89 xfs_agblock_t rbno; /* right block's block number */
90 xfs_buf_t *rbp; /* right block's buffer pointer */
91 xfs_alloc_block_t *right; /* right btree block */
92 xfs_alloc_key_t *rkp; /* right block key pointer */
93 xfs_alloc_ptr_t *rpp; /* right block address pointer */
94 int rrecs=0; /* number of records in right block */
Eric Sandeen91d87232006-09-28 11:05:40 +100095 int numrecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 xfs_alloc_rec_t *rrp; /* right block record pointer */
97 xfs_btree_cur_t *tcur; /* temporary btree cursor */
98
99 /*
100 * Get the index of the entry being deleted, check for nothing there.
101 */
102 ptr = cur->bc_ptrs[level];
103 if (ptr == 0) {
104 *stat = 0;
105 return 0;
106 }
107 /*
108 * Get the buffer & block containing the record or key/ptr.
109 */
110 bp = cur->bc_bufs[level];
111 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
112#ifdef DEBUG
113 if ((error = xfs_btree_check_sblock(cur, block, level, bp)))
114 return error;
115#endif
116 /*
117 * Fail if we're off the end of the block.
118 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000119 numrecs = be16_to_cpu(block->bb_numrecs);
120 if (ptr > numrecs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 *stat = 0;
122 return 0;
123 }
124 XFS_STATS_INC(xs_abt_delrec);
125 /*
126 * It's a nonleaf. Excise the key and ptr being deleted, by
127 * sliding the entries past them down one.
128 * Log the changed areas of the block.
129 */
130 if (level > 0) {
131 lkp = XFS_ALLOC_KEY_ADDR(block, 1, cur);
132 lpp = XFS_ALLOC_PTR_ADDR(block, 1, cur);
133#ifdef DEBUG
Eric Sandeen91d87232006-09-28 11:05:40 +1000134 for (i = ptr; i < numrecs; i++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +1100135 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(lpp[i]), level)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 return error;
137 }
138#endif
Eric Sandeen91d87232006-09-28 11:05:40 +1000139 if (ptr < numrecs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 memmove(&lkp[ptr - 1], &lkp[ptr],
Eric Sandeen91d87232006-09-28 11:05:40 +1000141 (numrecs - ptr) * sizeof(*lkp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 memmove(&lpp[ptr - 1], &lpp[ptr],
Eric Sandeen91d87232006-09-28 11:05:40 +1000143 (numrecs - ptr) * sizeof(*lpp));
144 xfs_alloc_log_ptrs(cur, bp, ptr, numrecs - 1);
145 xfs_alloc_log_keys(cur, bp, ptr, numrecs - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 }
147 }
148 /*
149 * It's a leaf. Excise the record being deleted, by sliding the
150 * entries past it down one. Log the changed areas of the block.
151 */
152 else {
153 lrp = XFS_ALLOC_REC_ADDR(block, 1, cur);
Eric Sandeen91d87232006-09-28 11:05:40 +1000154 if (ptr < numrecs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 memmove(&lrp[ptr - 1], &lrp[ptr],
Eric Sandeen91d87232006-09-28 11:05:40 +1000156 (numrecs - ptr) * sizeof(*lrp));
157 xfs_alloc_log_recs(cur, bp, ptr, numrecs - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 }
159 /*
160 * If it's the first record in the block, we'll need a key
161 * structure to pass up to the next level (updkey).
162 */
163 if (ptr == 1) {
Christoph Hellwig16259e72005-11-02 15:11:25 +1100164 key.ar_startblock = lrp->ar_startblock;
165 key.ar_blockcount = lrp->ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 lkp = &key;
167 }
168 }
169 /*
170 * Decrement and log the number of entries in the block.
171 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000172 numrecs--;
173 block->bb_numrecs = cpu_to_be16(numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 xfs_alloc_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS);
175 /*
176 * See if the longest free extent in the allocation group was
177 * changed by this operation. True if it's the by-size btree, and
178 * this is the leaf level, and there is no right sibling block,
179 * and this was the last record.
180 */
181 agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
182 mp = cur->bc_mp;
183
184 if (level == 0 &&
185 cur->bc_btnum == XFS_BTNUM_CNT &&
Christoph Hellwig16259e72005-11-02 15:11:25 +1100186 be32_to_cpu(block->bb_rightsib) == NULLAGBLOCK &&
Eric Sandeen91d87232006-09-28 11:05:40 +1000187 ptr > numrecs) {
188 ASSERT(ptr == numrecs + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 /*
190 * There are still records in the block. Grab the size
191 * from the last one.
192 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000193 if (numrecs) {
194 rrp = XFS_ALLOC_REC_ADDR(block, numrecs, cur);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100195 agf->agf_longest = rrp->ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 }
197 /*
198 * No free extents left.
199 */
200 else
201 agf->agf_longest = 0;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100202 mp->m_perag[be32_to_cpu(agf->agf_seqno)].pagf_longest =
203 be32_to_cpu(agf->agf_longest);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp,
205 XFS_AGF_LONGEST);
206 }
207 /*
208 * Is this the root level? If so, we're almost done.
209 */
210 if (level == cur->bc_nlevels - 1) {
211 /*
212 * If this is the root level,
213 * and there's only one entry left,
214 * and it's NOT the leaf level,
215 * then we can get rid of this level.
216 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000217 if (numrecs == 1 && level > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 /*
219 * lpp is still set to the first pointer in the block.
220 * Make it the new root of the btree.
221 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100222 bno = be32_to_cpu(agf->agf_roots[cur->bc_btnum]);
223 agf->agf_roots[cur->bc_btnum] = *lpp;
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800224 be32_add_cpu(&agf->agf_levels[cur->bc_btnum], -1);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100225 mp->m_perag[be32_to_cpu(agf->agf_seqno)].pagf_levels[cur->bc_btnum]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 /*
227 * Put this buffer/block on the ag's freelist.
228 */
David Chinner92821e22007-05-24 15:26:31 +1000229 error = xfs_alloc_put_freelist(cur->bc_tp,
230 cur->bc_private.a.agbp, NULL, bno, 1);
231 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 return error;
233 /*
234 * Since blocks move to the free list without the
235 * coordination used in xfs_bmap_finish, we can't allow
236 * block to be available for reallocation and
237 * non-transaction writing (user data) until we know
238 * that the transaction that moved it to the free list
239 * is permanently on disk. We track the blocks by
240 * declaring these blocks as "busy"; the busy list is
241 * maintained on a per-ag basis and each transaction
242 * records which entries should be removed when the
243 * iclog commits to disk. If a busy block is
244 * allocated, the iclog is pushed up to the LSN
245 * that freed the block.
246 */
247 xfs_alloc_mark_busy(cur->bc_tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +1100248 be32_to_cpu(agf->agf_seqno), bno, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250 xfs_trans_agbtree_delta(cur->bc_tp, -1);
251 xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp,
252 XFS_AGF_ROOTS | XFS_AGF_LEVELS);
253 /*
254 * Update the cursor so there's one fewer level.
255 */
256 xfs_btree_setbuf(cur, level, NULL);
257 cur->bc_nlevels--;
258 } else if (level > 0 &&
259 (error = xfs_alloc_decrement(cur, level, &i)))
260 return error;
261 *stat = 1;
262 return 0;
263 }
264 /*
265 * If we deleted the leftmost entry in the block, update the
266 * key values above us in the tree.
267 */
268 if (ptr == 1 && (error = xfs_alloc_updkey(cur, lkp, level + 1)))
269 return error;
270 /*
271 * If the number of records remaining in the block is at least
272 * the minimum, we're done.
273 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000274 if (numrecs >= XFS_ALLOC_BLOCK_MINRECS(level, cur)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (level > 0 && (error = xfs_alloc_decrement(cur, level, &i)))
276 return error;
277 *stat = 1;
278 return 0;
279 }
280 /*
281 * Otherwise, we have to move some records around to keep the
282 * tree balanced. Look at the left and right sibling blocks to
283 * see if we can re-balance by moving only one record.
284 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100285 rbno = be32_to_cpu(block->bb_rightsib);
286 lbno = be32_to_cpu(block->bb_leftsib);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 bno = NULLAGBLOCK;
288 ASSERT(rbno != NULLAGBLOCK || lbno != NULLAGBLOCK);
289 /*
290 * Duplicate the cursor so our btree manipulations here won't
291 * disrupt the next level up.
292 */
293 if ((error = xfs_btree_dup_cursor(cur, &tcur)))
294 return error;
295 /*
296 * If there's a right sibling, see if it's ok to shift an entry
297 * out of it.
298 */
299 if (rbno != NULLAGBLOCK) {
300 /*
301 * Move the temp cursor to the last entry in the next block.
302 * Actually any entry but the first would suffice.
303 */
304 i = xfs_btree_lastrec(tcur, level);
305 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
306 if ((error = xfs_alloc_increment(tcur, level, &i)))
307 goto error0;
308 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
309 i = xfs_btree_lastrec(tcur, level);
310 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
311 /*
312 * Grab a pointer to the block.
313 */
314 rbp = tcur->bc_bufs[level];
315 right = XFS_BUF_TO_ALLOC_BLOCK(rbp);
316#ifdef DEBUG
317 if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
318 goto error0;
319#endif
320 /*
321 * Grab the current block number, for future use.
322 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100323 bno = be32_to_cpu(right->bb_leftsib);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 /*
325 * If right block is full enough so that removing one entry
326 * won't make it too empty, and left-shifting an entry out
327 * of right to us works, we're done.
328 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100329 if (be16_to_cpu(right->bb_numrecs) - 1 >=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 XFS_ALLOC_BLOCK_MINRECS(level, cur)) {
331 if ((error = xfs_alloc_lshift(tcur, level, &i)))
332 goto error0;
333 if (i) {
Christoph Hellwig16259e72005-11-02 15:11:25 +1100334 ASSERT(be16_to_cpu(block->bb_numrecs) >=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 XFS_ALLOC_BLOCK_MINRECS(level, cur));
336 xfs_btree_del_cursor(tcur,
337 XFS_BTREE_NOERROR);
338 if (level > 0 &&
339 (error = xfs_alloc_decrement(cur, level,
340 &i)))
341 return error;
342 *stat = 1;
343 return 0;
344 }
345 }
346 /*
347 * Otherwise, grab the number of records in right for
348 * future reference, and fix up the temp cursor to point
349 * to our block again (last record).
350 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100351 rrecs = be16_to_cpu(right->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 if (lbno != NULLAGBLOCK) {
353 i = xfs_btree_firstrec(tcur, level);
354 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
355 if ((error = xfs_alloc_decrement(tcur, level, &i)))
356 goto error0;
357 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
358 }
359 }
360 /*
361 * If there's a left sibling, see if it's ok to shift an entry
362 * out of it.
363 */
364 if (lbno != NULLAGBLOCK) {
365 /*
366 * Move the temp cursor to the first entry in the
367 * previous block.
368 */
369 i = xfs_btree_firstrec(tcur, level);
370 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
371 if ((error = xfs_alloc_decrement(tcur, level, &i)))
372 goto error0;
373 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
374 xfs_btree_firstrec(tcur, level);
375 /*
376 * Grab a pointer to the block.
377 */
378 lbp = tcur->bc_bufs[level];
379 left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
380#ifdef DEBUG
381 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
382 goto error0;
383#endif
384 /*
385 * Grab the current block number, for future use.
386 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100387 bno = be32_to_cpu(left->bb_rightsib);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 /*
389 * If left block is full enough so that removing one entry
390 * won't make it too empty, and right-shifting an entry out
391 * of left to us works, we're done.
392 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100393 if (be16_to_cpu(left->bb_numrecs) - 1 >=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 XFS_ALLOC_BLOCK_MINRECS(level, cur)) {
395 if ((error = xfs_alloc_rshift(tcur, level, &i)))
396 goto error0;
397 if (i) {
Christoph Hellwig16259e72005-11-02 15:11:25 +1100398 ASSERT(be16_to_cpu(block->bb_numrecs) >=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 XFS_ALLOC_BLOCK_MINRECS(level, cur));
400 xfs_btree_del_cursor(tcur,
401 XFS_BTREE_NOERROR);
402 if (level == 0)
403 cur->bc_ptrs[0]++;
404 *stat = 1;
405 return 0;
406 }
407 }
408 /*
409 * Otherwise, grab the number of records in right for
410 * future reference.
411 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100412 lrecs = be16_to_cpu(left->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
414 /*
415 * Delete the temp cursor, we're done with it.
416 */
417 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
418 /*
419 * If here, we need to do a join to keep the tree balanced.
420 */
421 ASSERT(bno != NULLAGBLOCK);
422 /*
423 * See if we can join with the left neighbor block.
424 */
425 if (lbno != NULLAGBLOCK &&
Eric Sandeen91d87232006-09-28 11:05:40 +1000426 lrecs + numrecs <= XFS_ALLOC_BLOCK_MAXRECS(level, cur)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 /*
428 * Set "right" to be the starting block,
429 * "left" to be the left neighbor.
430 */
431 rbno = bno;
432 right = block;
Eric Sandeen91d87232006-09-28 11:05:40 +1000433 rrecs = be16_to_cpu(right->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 rbp = bp;
435 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
436 cur->bc_private.a.agno, lbno, 0, &lbp,
437 XFS_ALLOC_BTREE_REF)))
438 return error;
439 left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
Eric Sandeen91d87232006-09-28 11:05:40 +1000440 lrecs = be16_to_cpu(left->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
442 return error;
443 }
444 /*
445 * If that won't work, see if we can join with the right neighbor block.
446 */
447 else if (rbno != NULLAGBLOCK &&
Eric Sandeen91d87232006-09-28 11:05:40 +1000448 rrecs + numrecs <= XFS_ALLOC_BLOCK_MAXRECS(level, cur)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 /*
450 * Set "left" to be the starting block,
451 * "right" to be the right neighbor.
452 */
453 lbno = bno;
454 left = block;
Eric Sandeen91d87232006-09-28 11:05:40 +1000455 lrecs = be16_to_cpu(left->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 lbp = bp;
457 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
458 cur->bc_private.a.agno, rbno, 0, &rbp,
459 XFS_ALLOC_BTREE_REF)))
460 return error;
461 right = XFS_BUF_TO_ALLOC_BLOCK(rbp);
Eric Sandeen91d87232006-09-28 11:05:40 +1000462 rrecs = be16_to_cpu(right->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
464 return error;
465 }
466 /*
467 * Otherwise, we can't fix the imbalance.
468 * Just return. This is probably a logic error, but it's not fatal.
469 */
470 else {
471 if (level > 0 && (error = xfs_alloc_decrement(cur, level, &i)))
472 return error;
473 *stat = 1;
474 return 0;
475 }
476 /*
477 * We're now going to join "left" and "right" by moving all the stuff
478 * in "right" to "left" and deleting "right".
479 */
480 if (level > 0) {
481 /*
482 * It's a non-leaf. Move keys and pointers.
483 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000484 lkp = XFS_ALLOC_KEY_ADDR(left, lrecs + 1, cur);
485 lpp = XFS_ALLOC_PTR_ADDR(left, lrecs + 1, cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur);
487 rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur);
488#ifdef DEBUG
Eric Sandeen91d87232006-09-28 11:05:40 +1000489 for (i = 0; i < rrecs; i++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +1100490 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i]), level)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 return error;
492 }
493#endif
Eric Sandeen91d87232006-09-28 11:05:40 +1000494 memcpy(lkp, rkp, rrecs * sizeof(*lkp));
495 memcpy(lpp, rpp, rrecs * sizeof(*lpp));
496 xfs_alloc_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
497 xfs_alloc_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 } else {
499 /*
500 * It's a leaf. Move records.
501 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000502 lrp = XFS_ALLOC_REC_ADDR(left, lrecs + 1, cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 rrp = XFS_ALLOC_REC_ADDR(right, 1, cur);
Eric Sandeen91d87232006-09-28 11:05:40 +1000504 memcpy(lrp, rrp, rrecs * sizeof(*lrp));
505 xfs_alloc_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 }
507 /*
508 * If we joined with the left neighbor, set the buffer in the
509 * cursor to the left block, and fix up the index.
510 */
511 if (bp != lbp) {
512 xfs_btree_setbuf(cur, level, lbp);
Eric Sandeen91d87232006-09-28 11:05:40 +1000513 cur->bc_ptrs[level] += lrecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 }
515 /*
516 * If we joined with the right neighbor and there's a level above
517 * us, increment the cursor at that level.
518 */
519 else if (level + 1 < cur->bc_nlevels &&
520 (error = xfs_alloc_increment(cur, level + 1, &i)))
521 return error;
522 /*
523 * Fix up the number of records in the surviving block.
524 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000525 lrecs += rrecs;
526 left->bb_numrecs = cpu_to_be16(lrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 /*
528 * Fix up the right block pointer in the surviving block, and log it.
529 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100530 left->bb_rightsib = right->bb_rightsib;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
532 /*
533 * If there is a right sibling now, make it point to the
534 * remaining block.
535 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100536 if (be32_to_cpu(left->bb_rightsib) != NULLAGBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 xfs_alloc_block_t *rrblock;
538 xfs_buf_t *rrbp;
539
540 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +1100541 cur->bc_private.a.agno, be32_to_cpu(left->bb_rightsib), 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 &rrbp, XFS_ALLOC_BTREE_REF)))
543 return error;
544 rrblock = XFS_BUF_TO_ALLOC_BLOCK(rrbp);
545 if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp)))
546 return error;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100547 rrblock->bb_leftsib = cpu_to_be32(lbno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 xfs_alloc_log_block(cur->bc_tp, rrbp, XFS_BB_LEFTSIB);
549 }
550 /*
551 * Free the deleting block by putting it on the freelist.
552 */
David Chinner92821e22007-05-24 15:26:31 +1000553 error = xfs_alloc_put_freelist(cur->bc_tp,
554 cur->bc_private.a.agbp, NULL, rbno, 1);
555 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return error;
557 /*
558 * Since blocks move to the free list without the coordination
559 * used in xfs_bmap_finish, we can't allow block to be available
560 * for reallocation and non-transaction writing (user data)
561 * until we know that the transaction that moved it to the free
562 * list is permanently on disk. We track the blocks by declaring
563 * these blocks as "busy"; the busy list is maintained on a
564 * per-ag basis and each transaction records which entries
565 * should be removed when the iclog commits to disk. If a
566 * busy block is allocated, the iclog is pushed up to the
567 * LSN that freed the block.
568 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100569 xfs_alloc_mark_busy(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 xfs_trans_agbtree_delta(cur->bc_tp, -1);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 /*
573 * Adjust the current level's cursor so that we're left referring
574 * to the right node, after we're done.
575 * If this leaves the ptr value 0 our caller will fix it up.
576 */
577 if (level > 0)
578 cur->bc_ptrs[level]--;
579 /*
580 * Return value means the next level up has something to do.
581 */
582 *stat = 2;
583 return 0;
584
585error0:
586 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
587 return error;
588}
589
590/*
591 * Insert one record/level. Return information to the caller
592 * allowing the next level up to proceed if necessary.
593 */
594STATIC int /* error */
595xfs_alloc_insrec(
596 xfs_btree_cur_t *cur, /* btree cursor */
597 int level, /* level to insert record at */
598 xfs_agblock_t *bnop, /* i/o: block number inserted */
599 xfs_alloc_rec_t *recp, /* i/o: record data inserted */
600 xfs_btree_cur_t **curp, /* output: new cursor replacing cur */
601 int *stat) /* output: success/failure */
602{
603 xfs_agf_t *agf; /* allocation group freelist header */
604 xfs_alloc_block_t *block; /* btree block record/key lives in */
605 xfs_buf_t *bp; /* buffer for block */
606 int error; /* error return value */
607 int i; /* loop index */
608 xfs_alloc_key_t key; /* key value being inserted */
609 xfs_alloc_key_t *kp; /* pointer to btree keys */
610 xfs_agblock_t nbno; /* block number of allocated block */
611 xfs_btree_cur_t *ncur; /* new cursor to be used at next lvl */
612 xfs_alloc_key_t nkey; /* new key value, from split */
613 xfs_alloc_rec_t nrec; /* new record value, for caller */
Eric Sandeen91d87232006-09-28 11:05:40 +1000614 int numrecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 int optr; /* old ptr value */
616 xfs_alloc_ptr_t *pp; /* pointer to btree addresses */
617 int ptr; /* index in btree block for this rec */
618 xfs_alloc_rec_t *rp; /* pointer to btree records */
619
Christoph Hellwig16259e72005-11-02 15:11:25 +1100620 ASSERT(be32_to_cpu(recp->ar_blockcount) > 0);
Christoph Hellwig5bde1ba92005-11-02 15:06:18 +1100621
622 /*
623 * GCC doesn't understand the (arguably complex) control flow in
624 * this function and complains about uninitialized structure fields
625 * without this.
626 */
627 memset(&nrec, 0, sizeof(nrec));
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 /*
630 * If we made it to the root level, allocate a new root block
631 * and we're done.
632 */
633 if (level >= cur->bc_nlevels) {
634 XFS_STATS_INC(xs_abt_insrec);
635 if ((error = xfs_alloc_newroot(cur, &i)))
636 return error;
637 *bnop = NULLAGBLOCK;
638 *stat = i;
639 return 0;
640 }
641 /*
642 * Make a key out of the record data to be inserted, and save it.
643 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100644 key.ar_startblock = recp->ar_startblock;
645 key.ar_blockcount = recp->ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 optr = ptr = cur->bc_ptrs[level];
647 /*
648 * If we're off the left edge, return failure.
649 */
650 if (ptr == 0) {
651 *stat = 0;
652 return 0;
653 }
654 XFS_STATS_INC(xs_abt_insrec);
655 /*
656 * Get pointers to the btree buffer and block.
657 */
658 bp = cur->bc_bufs[level];
659 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
Eric Sandeen91d87232006-09-28 11:05:40 +1000660 numrecs = be16_to_cpu(block->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661#ifdef DEBUG
662 if ((error = xfs_btree_check_sblock(cur, block, level, bp)))
663 return error;
664 /*
665 * Check that the new entry is being inserted in the right place.
666 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000667 if (ptr <= numrecs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 if (level == 0) {
669 rp = XFS_ALLOC_REC_ADDR(block, ptr, cur);
670 xfs_btree_check_rec(cur->bc_btnum, recp, rp);
671 } else {
672 kp = XFS_ALLOC_KEY_ADDR(block, ptr, cur);
673 xfs_btree_check_key(cur->bc_btnum, &key, kp);
674 }
675 }
676#endif
677 nbno = NULLAGBLOCK;
Nathan Scott1121b212006-09-28 10:58:40 +1000678 ncur = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 /*
680 * If the block is full, we can't insert the new entry until we
681 * make the block un-full.
682 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000683 if (numrecs == XFS_ALLOC_BLOCK_MAXRECS(level, cur)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 /*
685 * First, try shifting an entry to the right neighbor.
686 */
687 if ((error = xfs_alloc_rshift(cur, level, &i)))
688 return error;
689 if (i) {
690 /* nothing */
691 }
692 /*
693 * Next, try shifting an entry to the left neighbor.
694 */
695 else {
696 if ((error = xfs_alloc_lshift(cur, level, &i)))
697 return error;
698 if (i)
699 optr = ptr = cur->bc_ptrs[level];
700 else {
701 /*
702 * Next, try splitting the current block in
703 * half. If this works we have to re-set our
704 * variables because we could be in a
705 * different block now.
706 */
707 if ((error = xfs_alloc_split(cur, level, &nbno,
708 &nkey, &ncur, &i)))
709 return error;
710 if (i) {
711 bp = cur->bc_bufs[level];
712 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
713#ifdef DEBUG
714 if ((error =
715 xfs_btree_check_sblock(cur,
716 block, level, bp)))
717 return error;
718#endif
719 ptr = cur->bc_ptrs[level];
Christoph Hellwig16259e72005-11-02 15:11:25 +1100720 nrec.ar_startblock = nkey.ar_startblock;
721 nrec.ar_blockcount = nkey.ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
723 /*
724 * Otherwise the insert fails.
725 */
726 else {
727 *stat = 0;
728 return 0;
729 }
730 }
731 }
732 }
733 /*
734 * At this point we know there's room for our new entry in the block
735 * we're pointing at.
736 */
Eric Sandeen91d87232006-09-28 11:05:40 +1000737 numrecs = be16_to_cpu(block->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 if (level > 0) {
739 /*
740 * It's a non-leaf entry. Make a hole for the new data
741 * in the key and ptr regions of the block.
742 */
743 kp = XFS_ALLOC_KEY_ADDR(block, 1, cur);
744 pp = XFS_ALLOC_PTR_ADDR(block, 1, cur);
745#ifdef DEBUG
Eric Sandeen91d87232006-09-28 11:05:40 +1000746 for (i = numrecs; i >= ptr; i--) {
Christoph Hellwig16259e72005-11-02 15:11:25 +1100747 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(pp[i - 1]), level)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 return error;
749 }
750#endif
751 memmove(&kp[ptr], &kp[ptr - 1],
Eric Sandeen91d87232006-09-28 11:05:40 +1000752 (numrecs - ptr + 1) * sizeof(*kp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 memmove(&pp[ptr], &pp[ptr - 1],
Eric Sandeen91d87232006-09-28 11:05:40 +1000754 (numrecs - ptr + 1) * sizeof(*pp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755#ifdef DEBUG
756 if ((error = xfs_btree_check_sptr(cur, *bnop, level)))
757 return error;
758#endif
759 /*
760 * Now stuff the new data in, bump numrecs and log the new data.
761 */
762 kp[ptr - 1] = key;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100763 pp[ptr - 1] = cpu_to_be32(*bnop);
Eric Sandeen91d87232006-09-28 11:05:40 +1000764 numrecs++;
765 block->bb_numrecs = cpu_to_be16(numrecs);
766 xfs_alloc_log_keys(cur, bp, ptr, numrecs);
767 xfs_alloc_log_ptrs(cur, bp, ptr, numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768#ifdef DEBUG
Eric Sandeen91d87232006-09-28 11:05:40 +1000769 if (ptr < numrecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 xfs_btree_check_key(cur->bc_btnum, kp + ptr - 1,
771 kp + ptr);
772#endif
773 } else {
774 /*
775 * It's a leaf entry. Make a hole for the new record.
776 */
777 rp = XFS_ALLOC_REC_ADDR(block, 1, cur);
778 memmove(&rp[ptr], &rp[ptr - 1],
Eric Sandeen91d87232006-09-28 11:05:40 +1000779 (numrecs - ptr + 1) * sizeof(*rp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 /*
781 * Now stuff the new record in, bump numrecs
782 * and log the new data.
783 */
Christoph Hellwigc38e5e82006-09-28 10:57:17 +1000784 rp[ptr - 1] = *recp;
Eric Sandeen91d87232006-09-28 11:05:40 +1000785 numrecs++;
786 block->bb_numrecs = cpu_to_be16(numrecs);
787 xfs_alloc_log_recs(cur, bp, ptr, numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788#ifdef DEBUG
Eric Sandeen91d87232006-09-28 11:05:40 +1000789 if (ptr < numrecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 xfs_btree_check_rec(cur->bc_btnum, rp + ptr - 1,
791 rp + ptr);
792#endif
793 }
794 /*
795 * Log the new number of records in the btree header.
796 */
797 xfs_alloc_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS);
798 /*
799 * If we inserted at the start of a block, update the parents' keys.
800 */
801 if (optr == 1 && (error = xfs_alloc_updkey(cur, &key, level + 1)))
802 return error;
803 /*
804 * Look to see if the longest extent in the allocation group
805 * needs to be updated.
806 */
807
808 agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
809 if (level == 0 &&
810 cur->bc_btnum == XFS_BTNUM_CNT &&
Christoph Hellwig16259e72005-11-02 15:11:25 +1100811 be32_to_cpu(block->bb_rightsib) == NULLAGBLOCK &&
812 be32_to_cpu(recp->ar_blockcount) > be32_to_cpu(agf->agf_longest)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 /*
814 * If this is a leaf in the by-size btree and there
815 * is no right sibling block and this block is bigger
816 * than the previous longest block, update it.
817 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100818 agf->agf_longest = recp->ar_blockcount;
819 cur->bc_mp->m_perag[be32_to_cpu(agf->agf_seqno)].pagf_longest
820 = be32_to_cpu(recp->ar_blockcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp,
822 XFS_AGF_LONGEST);
823 }
824 /*
825 * Return the new block number, if any.
826 * If there is one, give back a record value and a cursor too.
827 */
828 *bnop = nbno;
829 if (nbno != NULLAGBLOCK) {
Christoph Hellwigc38e5e82006-09-28 10:57:17 +1000830 *recp = nrec;
831 *curp = ncur;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 }
833 *stat = 1;
834 return 0;
835}
836
837/*
838 * Log header fields from a btree block.
839 */
840STATIC void
841xfs_alloc_log_block(
842 xfs_trans_t *tp, /* transaction pointer */
843 xfs_buf_t *bp, /* buffer containing btree block */
844 int fields) /* mask of fields: XFS_BB_... */
845{
846 int first; /* first byte offset logged */
847 int last; /* last byte offset logged */
848 static const short offsets[] = { /* table of offsets */
849 offsetof(xfs_alloc_block_t, bb_magic),
850 offsetof(xfs_alloc_block_t, bb_level),
851 offsetof(xfs_alloc_block_t, bb_numrecs),
852 offsetof(xfs_alloc_block_t, bb_leftsib),
853 offsetof(xfs_alloc_block_t, bb_rightsib),
854 sizeof(xfs_alloc_block_t)
855 };
856
857 xfs_btree_offsets(fields, offsets, XFS_BB_NUM_BITS, &first, &last);
858 xfs_trans_log_buf(tp, bp, first, last);
859}
860
861/*
862 * Log keys from a btree block (nonleaf).
863 */
864STATIC void
865xfs_alloc_log_keys(
866 xfs_btree_cur_t *cur, /* btree cursor */
867 xfs_buf_t *bp, /* buffer containing btree block */
868 int kfirst, /* index of first key to log */
869 int klast) /* index of last key to log */
870{
871 xfs_alloc_block_t *block; /* btree block to log from */
872 int first; /* first byte offset logged */
873 xfs_alloc_key_t *kp; /* key pointer in btree block */
874 int last; /* last byte offset logged */
875
876 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
877 kp = XFS_ALLOC_KEY_ADDR(block, 1, cur);
878 first = (int)((xfs_caddr_t)&kp[kfirst - 1] - (xfs_caddr_t)block);
879 last = (int)(((xfs_caddr_t)&kp[klast] - 1) - (xfs_caddr_t)block);
880 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
881}
882
883/*
884 * Log block pointer fields from a btree block (nonleaf).
885 */
886STATIC void
887xfs_alloc_log_ptrs(
888 xfs_btree_cur_t *cur, /* btree cursor */
889 xfs_buf_t *bp, /* buffer containing btree block */
890 int pfirst, /* index of first pointer to log */
891 int plast) /* index of last pointer to log */
892{
893 xfs_alloc_block_t *block; /* btree block to log from */
894 int first; /* first byte offset logged */
895 int last; /* last byte offset logged */
896 xfs_alloc_ptr_t *pp; /* block-pointer pointer in btree blk */
897
898 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
899 pp = XFS_ALLOC_PTR_ADDR(block, 1, cur);
900 first = (int)((xfs_caddr_t)&pp[pfirst - 1] - (xfs_caddr_t)block);
901 last = (int)(((xfs_caddr_t)&pp[plast] - 1) - (xfs_caddr_t)block);
902 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
903}
904
905/*
906 * Log records from a btree block (leaf).
907 */
908STATIC void
909xfs_alloc_log_recs(
910 xfs_btree_cur_t *cur, /* btree cursor */
911 xfs_buf_t *bp, /* buffer containing btree block */
912 int rfirst, /* index of first record to log */
913 int rlast) /* index of last record to log */
914{
915 xfs_alloc_block_t *block; /* btree block to log from */
916 int first; /* first byte offset logged */
917 int last; /* last byte offset logged */
918 xfs_alloc_rec_t *rp; /* record pointer for btree block */
919
920
921 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
922 rp = XFS_ALLOC_REC_ADDR(block, 1, cur);
923#ifdef DEBUG
924 {
925 xfs_agf_t *agf;
926 xfs_alloc_rec_t *p;
927
928 agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
929 for (p = &rp[rfirst - 1]; p <= &rp[rlast - 1]; p++)
Christoph Hellwig16259e72005-11-02 15:11:25 +1100930 ASSERT(be32_to_cpu(p->ar_startblock) +
931 be32_to_cpu(p->ar_blockcount) <=
932 be32_to_cpu(agf->agf_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 }
934#endif
935 first = (int)((xfs_caddr_t)&rp[rfirst - 1] - (xfs_caddr_t)block);
936 last = (int)(((xfs_caddr_t)&rp[rlast] - 1) - (xfs_caddr_t)block);
937 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
938}
939
940/*
941 * Lookup the record. The cursor is made to point to it, based on dir.
942 * Return 0 if can't find any such record, 1 for success.
943 */
944STATIC int /* error */
945xfs_alloc_lookup(
946 xfs_btree_cur_t *cur, /* btree cursor */
947 xfs_lookup_t dir, /* <=, ==, or >= */
948 int *stat) /* success/failure */
949{
950 xfs_agblock_t agbno; /* a.g. relative btree block number */
951 xfs_agnumber_t agno; /* allocation group number */
952 xfs_alloc_block_t *block=NULL; /* current btree block */
953 int diff; /* difference for the current key */
954 int error; /* error return value */
955 int keyno=0; /* current key number */
956 int level; /* level in the btree */
957 xfs_mount_t *mp; /* file system mount point */
958
959 XFS_STATS_INC(xs_abt_lookup);
960 /*
961 * Get the allocation group header, and the root block number.
962 */
963 mp = cur->bc_mp;
964
965 {
966 xfs_agf_t *agf; /* a.g. freespace header */
967
968 agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100969 agno = be32_to_cpu(agf->agf_seqno);
970 agbno = be32_to_cpu(agf->agf_roots[cur->bc_btnum]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 }
972 /*
973 * Iterate over each level in the btree, starting at the root.
974 * For each level above the leaves, find the key we need, based
975 * on the lookup record, then follow the corresponding block
976 * pointer down to the next level.
977 */
978 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
979 xfs_buf_t *bp; /* buffer pointer for btree block */
980 xfs_daddr_t d; /* disk address of btree block */
981
982 /*
983 * Get the disk address we're looking for.
984 */
985 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
986 /*
987 * If the old buffer at this level is for a different block,
988 * throw it away, otherwise just use it.
989 */
990 bp = cur->bc_bufs[level];
991 if (bp && XFS_BUF_ADDR(bp) != d)
Nathan Scott1121b212006-09-28 10:58:40 +1000992 bp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 if (!bp) {
994 /*
995 * Need to get a new buffer. Read it, then
996 * set it in the cursor, releasing the old one.
997 */
998 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, agno,
999 agbno, 0, &bp, XFS_ALLOC_BTREE_REF)))
1000 return error;
1001 xfs_btree_setbuf(cur, level, bp);
1002 /*
1003 * Point to the btree block, now that we have the buffer
1004 */
1005 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
1006 if ((error = xfs_btree_check_sblock(cur, block, level,
1007 bp)))
1008 return error;
1009 } else
1010 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
1011 /*
1012 * If we already had a key match at a higher level, we know
1013 * we need to use the first entry in this block.
1014 */
1015 if (diff == 0)
1016 keyno = 1;
1017 /*
1018 * Otherwise we need to search this block. Do a binary search.
1019 */
1020 else {
1021 int high; /* high entry number */
1022 xfs_alloc_key_t *kkbase=NULL;/* base of keys in block */
1023 xfs_alloc_rec_t *krbase=NULL;/* base of records in block */
1024 int low; /* low entry number */
1025
1026 /*
1027 * Get a pointer to keys or records.
1028 */
1029 if (level > 0)
1030 kkbase = XFS_ALLOC_KEY_ADDR(block, 1, cur);
1031 else
1032 krbase = XFS_ALLOC_REC_ADDR(block, 1, cur);
1033 /*
1034 * Set low and high entry numbers, 1-based.
1035 */
1036 low = 1;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001037 if (!(high = be16_to_cpu(block->bb_numrecs))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 /*
1039 * If the block is empty, the tree must
1040 * be an empty leaf.
1041 */
1042 ASSERT(level == 0 && cur->bc_nlevels == 1);
1043 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1044 *stat = 0;
1045 return 0;
1046 }
1047 /*
1048 * Binary search the block.
1049 */
1050 while (low <= high) {
1051 xfs_extlen_t blockcount; /* key value */
1052 xfs_agblock_t startblock; /* key value */
1053
1054 XFS_STATS_INC(xs_abt_compare);
1055 /*
1056 * keyno is average of low and high.
1057 */
1058 keyno = (low + high) >> 1;
1059 /*
1060 * Get startblock & blockcount.
1061 */
1062 if (level > 0) {
1063 xfs_alloc_key_t *kkp;
1064
1065 kkp = kkbase + keyno - 1;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001066 startblock = be32_to_cpu(kkp->ar_startblock);
1067 blockcount = be32_to_cpu(kkp->ar_blockcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 } else {
1069 xfs_alloc_rec_t *krp;
1070
1071 krp = krbase + keyno - 1;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001072 startblock = be32_to_cpu(krp->ar_startblock);
1073 blockcount = be32_to_cpu(krp->ar_blockcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 }
1075 /*
1076 * Compute difference to get next direction.
1077 */
1078 if (cur->bc_btnum == XFS_BTNUM_BNO)
1079 diff = (int)startblock -
1080 (int)cur->bc_rec.a.ar_startblock;
1081 else if (!(diff = (int)blockcount -
1082 (int)cur->bc_rec.a.ar_blockcount))
1083 diff = (int)startblock -
1084 (int)cur->bc_rec.a.ar_startblock;
1085 /*
1086 * Less than, move right.
1087 */
1088 if (diff < 0)
1089 low = keyno + 1;
1090 /*
1091 * Greater than, move left.
1092 */
1093 else if (diff > 0)
1094 high = keyno - 1;
1095 /*
1096 * Equal, we're done.
1097 */
1098 else
1099 break;
1100 }
1101 }
1102 /*
1103 * If there are more levels, set up for the next level
1104 * by getting the block number and filling in the cursor.
1105 */
1106 if (level > 0) {
1107 /*
1108 * If we moved left, need the previous key number,
1109 * unless there isn't one.
1110 */
1111 if (diff > 0 && --keyno < 1)
1112 keyno = 1;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001113 agbno = be32_to_cpu(*XFS_ALLOC_PTR_ADDR(block, keyno, cur));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114#ifdef DEBUG
1115 if ((error = xfs_btree_check_sptr(cur, agbno, level)))
1116 return error;
1117#endif
1118 cur->bc_ptrs[level] = keyno;
1119 }
1120 }
1121 /*
1122 * Done with the search.
1123 * See if we need to adjust the results.
1124 */
1125 if (dir != XFS_LOOKUP_LE && diff < 0) {
1126 keyno++;
1127 /*
1128 * If ge search and we went off the end of the block, but it's
1129 * not the last block, we're in the wrong block.
1130 */
1131 if (dir == XFS_LOOKUP_GE &&
Christoph Hellwig16259e72005-11-02 15:11:25 +11001132 keyno > be16_to_cpu(block->bb_numrecs) &&
1133 be32_to_cpu(block->bb_rightsib) != NULLAGBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 int i;
1135
1136 cur->bc_ptrs[0] = keyno;
1137 if ((error = xfs_alloc_increment(cur, 0, &i)))
1138 return error;
1139 XFS_WANT_CORRUPTED_RETURN(i == 1);
1140 *stat = 1;
1141 return 0;
1142 }
1143 }
1144 else if (dir == XFS_LOOKUP_LE && diff > 0)
1145 keyno--;
1146 cur->bc_ptrs[0] = keyno;
1147 /*
1148 * Return if we succeeded or not.
1149 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001150 if (keyno == 0 || keyno > be16_to_cpu(block->bb_numrecs))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 *stat = 0;
1152 else
1153 *stat = ((dir != XFS_LOOKUP_EQ) || (diff == 0));
1154 return 0;
1155}
1156
1157/*
1158 * Move 1 record left from cur/level if possible.
1159 * Update cur to reflect the new path.
1160 */
1161STATIC int /* error */
1162xfs_alloc_lshift(
1163 xfs_btree_cur_t *cur, /* btree cursor */
1164 int level, /* level to shift record on */
1165 int *stat) /* success/failure */
1166{
1167 int error; /* error return value */
1168#ifdef DEBUG
1169 int i; /* loop index */
1170#endif
1171 xfs_alloc_key_t key; /* key value for leaf level upward */
1172 xfs_buf_t *lbp; /* buffer for left neighbor block */
1173 xfs_alloc_block_t *left; /* left neighbor btree block */
1174 int nrec; /* new number of left block entries */
1175 xfs_buf_t *rbp; /* buffer for right (current) block */
1176 xfs_alloc_block_t *right; /* right (current) btree block */
1177 xfs_alloc_key_t *rkp=NULL; /* key pointer for right block */
1178 xfs_alloc_ptr_t *rpp=NULL; /* address pointer for right block */
1179 xfs_alloc_rec_t *rrp=NULL; /* record pointer for right block */
1180
1181 /*
1182 * Set up variables for this block as "right".
1183 */
1184 rbp = cur->bc_bufs[level];
1185 right = XFS_BUF_TO_ALLOC_BLOCK(rbp);
1186#ifdef DEBUG
1187 if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
1188 return error;
1189#endif
1190 /*
1191 * If we've got no left sibling then we can't shift an entry left.
1192 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001193 if (be32_to_cpu(right->bb_leftsib) == NULLAGBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 *stat = 0;
1195 return 0;
1196 }
1197 /*
1198 * If the cursor entry is the one that would be moved, don't
1199 * do it... it's too complicated.
1200 */
1201 if (cur->bc_ptrs[level] <= 1) {
1202 *stat = 0;
1203 return 0;
1204 }
1205 /*
1206 * Set up the left neighbor as "left".
1207 */
1208 if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +11001209 cur->bc_private.a.agno, be32_to_cpu(right->bb_leftsib),
1210 0, &lbp, XFS_ALLOC_BTREE_REF)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 return error;
1212 left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
1213 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
1214 return error;
1215 /*
1216 * If it's full, it can't take another entry.
1217 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001218 if (be16_to_cpu(left->bb_numrecs) == XFS_ALLOC_BLOCK_MAXRECS(level, cur)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 *stat = 0;
1220 return 0;
1221 }
Christoph Hellwig16259e72005-11-02 15:11:25 +11001222 nrec = be16_to_cpu(left->bb_numrecs) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 /*
1224 * If non-leaf, copy a key and a ptr to the left block.
1225 */
1226 if (level > 0) {
1227 xfs_alloc_key_t *lkp; /* key pointer for left block */
1228 xfs_alloc_ptr_t *lpp; /* address pointer for left block */
1229
1230 lkp = XFS_ALLOC_KEY_ADDR(left, nrec, cur);
1231 rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur);
1232 *lkp = *rkp;
1233 xfs_alloc_log_keys(cur, lbp, nrec, nrec);
1234 lpp = XFS_ALLOC_PTR_ADDR(left, nrec, cur);
1235 rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur);
1236#ifdef DEBUG
Christoph Hellwig16259e72005-11-02 15:11:25 +11001237 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(*rpp), level)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 return error;
1239#endif
Christoph Hellwigc38e5e82006-09-28 10:57:17 +10001240 *lpp = *rpp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 xfs_alloc_log_ptrs(cur, lbp, nrec, nrec);
1242 xfs_btree_check_key(cur->bc_btnum, lkp - 1, lkp);
1243 }
1244 /*
1245 * If leaf, copy a record to the left block.
1246 */
1247 else {
1248 xfs_alloc_rec_t *lrp; /* record pointer for left block */
1249
1250 lrp = XFS_ALLOC_REC_ADDR(left, nrec, cur);
1251 rrp = XFS_ALLOC_REC_ADDR(right, 1, cur);
1252 *lrp = *rrp;
1253 xfs_alloc_log_recs(cur, lbp, nrec, nrec);
1254 xfs_btree_check_rec(cur->bc_btnum, lrp - 1, lrp);
1255 }
1256 /*
1257 * Bump and log left's numrecs, decrement and log right's numrecs.
1258 */
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001259 be16_add_cpu(&left->bb_numrecs, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS);
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001261 be16_add_cpu(&right->bb_numrecs, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 xfs_alloc_log_block(cur->bc_tp, rbp, XFS_BB_NUMRECS);
1263 /*
1264 * Slide the contents of right down one entry.
1265 */
1266 if (level > 0) {
1267#ifdef DEBUG
Christoph Hellwig16259e72005-11-02 15:11:25 +11001268 for (i = 0; i < be16_to_cpu(right->bb_numrecs); i++) {
1269 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i + 1]),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 level)))
1271 return error;
1272 }
1273#endif
Christoph Hellwig16259e72005-11-02 15:11:25 +11001274 memmove(rkp, rkp + 1, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
1275 memmove(rpp, rpp + 1, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
1276 xfs_alloc_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
1277 xfs_alloc_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 } else {
Christoph Hellwig16259e72005-11-02 15:11:25 +11001279 memmove(rrp, rrp + 1, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
1280 xfs_alloc_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
1281 key.ar_startblock = rrp->ar_startblock;
1282 key.ar_blockcount = rrp->ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 rkp = &key;
1284 }
1285 /*
1286 * Update the parent key values of right.
1287 */
1288 if ((error = xfs_alloc_updkey(cur, rkp, level + 1)))
1289 return error;
1290 /*
1291 * Slide the cursor value left one.
1292 */
1293 cur->bc_ptrs[level]--;
1294 *stat = 1;
1295 return 0;
1296}
1297
1298/*
1299 * Allocate a new root block, fill it in.
1300 */
1301STATIC int /* error */
1302xfs_alloc_newroot(
1303 xfs_btree_cur_t *cur, /* btree cursor */
1304 int *stat) /* success/failure */
1305{
1306 int error; /* error return value */
1307 xfs_agblock_t lbno; /* left block number */
1308 xfs_buf_t *lbp; /* left btree buffer */
1309 xfs_alloc_block_t *left; /* left btree block */
1310 xfs_mount_t *mp; /* mount structure */
1311 xfs_agblock_t nbno; /* new block number */
1312 xfs_buf_t *nbp; /* new (root) buffer */
1313 xfs_alloc_block_t *new; /* new (root) btree block */
1314 int nptr; /* new value for key index, 1 or 2 */
1315 xfs_agblock_t rbno; /* right block number */
1316 xfs_buf_t *rbp; /* right btree buffer */
1317 xfs_alloc_block_t *right; /* right btree block */
1318
1319 mp = cur->bc_mp;
1320
1321 ASSERT(cur->bc_nlevels < XFS_AG_MAXLEVELS(mp));
1322 /*
1323 * Get a buffer from the freelist blocks, for the new root.
1324 */
David Chinner92821e22007-05-24 15:26:31 +10001325 error = xfs_alloc_get_freelist(cur->bc_tp,
1326 cur->bc_private.a.agbp, &nbno, 1);
1327 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 return error;
1329 /*
1330 * None available, we fail.
1331 */
1332 if (nbno == NULLAGBLOCK) {
1333 *stat = 0;
1334 return 0;
1335 }
1336 xfs_trans_agbtree_delta(cur->bc_tp, 1);
1337 nbp = xfs_btree_get_bufs(mp, cur->bc_tp, cur->bc_private.a.agno, nbno,
1338 0);
1339 new = XFS_BUF_TO_ALLOC_BLOCK(nbp);
1340 /*
1341 * Set the root data in the a.g. freespace structure.
1342 */
1343 {
1344 xfs_agf_t *agf; /* a.g. freespace header */
1345 xfs_agnumber_t seqno;
1346
1347 agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001348 agf->agf_roots[cur->bc_btnum] = cpu_to_be32(nbno);
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001349 be32_add_cpu(&agf->agf_levels[cur->bc_btnum], 1);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001350 seqno = be32_to_cpu(agf->agf_seqno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 mp->m_perag[seqno].pagf_levels[cur->bc_btnum]++;
1352 xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp,
1353 XFS_AGF_ROOTS | XFS_AGF_LEVELS);
1354 }
1355 /*
1356 * At the previous root level there are now two blocks: the old
1357 * root, and the new block generated when it was split.
1358 * We don't know which one the cursor is pointing at, so we
1359 * set up variables "left" and "right" for each case.
1360 */
1361 lbp = cur->bc_bufs[cur->bc_nlevels - 1];
1362 left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
1363#ifdef DEBUG
1364 if ((error = xfs_btree_check_sblock(cur, left, cur->bc_nlevels - 1, lbp)))
1365 return error;
1366#endif
Christoph Hellwig16259e72005-11-02 15:11:25 +11001367 if (be32_to_cpu(left->bb_rightsib) != NULLAGBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 /*
1369 * Our block is left, pick up the right block.
1370 */
1371 lbno = XFS_DADDR_TO_AGBNO(mp, XFS_BUF_ADDR(lbp));
Christoph Hellwig16259e72005-11-02 15:11:25 +11001372 rbno = be32_to_cpu(left->bb_rightsib);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
1374 cur->bc_private.a.agno, rbno, 0, &rbp,
1375 XFS_ALLOC_BTREE_REF)))
1376 return error;
1377 right = XFS_BUF_TO_ALLOC_BLOCK(rbp);
1378 if ((error = xfs_btree_check_sblock(cur, right,
1379 cur->bc_nlevels - 1, rbp)))
1380 return error;
1381 nptr = 1;
1382 } else {
1383 /*
1384 * Our block is right, pick up the left block.
1385 */
1386 rbp = lbp;
1387 right = left;
1388 rbno = XFS_DADDR_TO_AGBNO(mp, XFS_BUF_ADDR(rbp));
Christoph Hellwig16259e72005-11-02 15:11:25 +11001389 lbno = be32_to_cpu(right->bb_leftsib);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
1391 cur->bc_private.a.agno, lbno, 0, &lbp,
1392 XFS_ALLOC_BTREE_REF)))
1393 return error;
1394 left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
1395 if ((error = xfs_btree_check_sblock(cur, left,
1396 cur->bc_nlevels - 1, lbp)))
1397 return error;
1398 nptr = 2;
1399 }
1400 /*
1401 * Fill in the new block's btree header and log it.
1402 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001403 new->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
1404 new->bb_level = cpu_to_be16(cur->bc_nlevels);
1405 new->bb_numrecs = cpu_to_be16(2);
1406 new->bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1407 new->bb_rightsib = cpu_to_be32(NULLAGBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 xfs_alloc_log_block(cur->bc_tp, nbp, XFS_BB_ALL_BITS);
1409 ASSERT(lbno != NULLAGBLOCK && rbno != NULLAGBLOCK);
1410 /*
1411 * Fill in the key data in the new root.
1412 */
1413 {
1414 xfs_alloc_key_t *kp; /* btree key pointer */
1415
1416 kp = XFS_ALLOC_KEY_ADDR(new, 1, cur);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001417 if (be16_to_cpu(left->bb_level) > 0) {
Christoph Hellwigc38e5e82006-09-28 10:57:17 +10001418 kp[0] = *XFS_ALLOC_KEY_ADDR(left, 1, cur);
1419 kp[1] = *XFS_ALLOC_KEY_ADDR(right, 1, cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 } else {
1421 xfs_alloc_rec_t *rp; /* btree record pointer */
1422
1423 rp = XFS_ALLOC_REC_ADDR(left, 1, cur);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001424 kp[0].ar_startblock = rp->ar_startblock;
1425 kp[0].ar_blockcount = rp->ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 rp = XFS_ALLOC_REC_ADDR(right, 1, cur);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001427 kp[1].ar_startblock = rp->ar_startblock;
1428 kp[1].ar_blockcount = rp->ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 }
1430 }
1431 xfs_alloc_log_keys(cur, nbp, 1, 2);
1432 /*
1433 * Fill in the pointer data in the new root.
1434 */
1435 {
1436 xfs_alloc_ptr_t *pp; /* btree address pointer */
1437
1438 pp = XFS_ALLOC_PTR_ADDR(new, 1, cur);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001439 pp[0] = cpu_to_be32(lbno);
1440 pp[1] = cpu_to_be32(rbno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 }
1442 xfs_alloc_log_ptrs(cur, nbp, 1, 2);
1443 /*
1444 * Fix up the cursor.
1445 */
1446 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
1447 cur->bc_ptrs[cur->bc_nlevels] = nptr;
1448 cur->bc_nlevels++;
1449 *stat = 1;
1450 return 0;
1451}
1452
1453/*
1454 * Move 1 record right from cur/level if possible.
1455 * Update cur to reflect the new path.
1456 */
1457STATIC int /* error */
1458xfs_alloc_rshift(
1459 xfs_btree_cur_t *cur, /* btree cursor */
1460 int level, /* level to shift record on */
1461 int *stat) /* success/failure */
1462{
1463 int error; /* error return value */
1464 int i; /* loop index */
1465 xfs_alloc_key_t key; /* key value for leaf level upward */
1466 xfs_buf_t *lbp; /* buffer for left (current) block */
1467 xfs_alloc_block_t *left; /* left (current) btree block */
1468 xfs_buf_t *rbp; /* buffer for right neighbor block */
1469 xfs_alloc_block_t *right; /* right neighbor btree block */
1470 xfs_alloc_key_t *rkp; /* key pointer for right block */
1471 xfs_btree_cur_t *tcur; /* temporary cursor */
1472
1473 /*
1474 * Set up variables for this block as "left".
1475 */
1476 lbp = cur->bc_bufs[level];
1477 left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
1478#ifdef DEBUG
1479 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
1480 return error;
1481#endif
1482 /*
1483 * If we've got no right sibling then we can't shift an entry right.
1484 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001485 if (be32_to_cpu(left->bb_rightsib) == NULLAGBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 *stat = 0;
1487 return 0;
1488 }
1489 /*
1490 * If the cursor entry is the one that would be moved, don't
1491 * do it... it's too complicated.
1492 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001493 if (cur->bc_ptrs[level] >= be16_to_cpu(left->bb_numrecs)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 *stat = 0;
1495 return 0;
1496 }
1497 /*
1498 * Set up the right neighbor as "right".
1499 */
1500 if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +11001501 cur->bc_private.a.agno, be32_to_cpu(left->bb_rightsib),
1502 0, &rbp, XFS_ALLOC_BTREE_REF)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 return error;
1504 right = XFS_BUF_TO_ALLOC_BLOCK(rbp);
1505 if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
1506 return error;
1507 /*
1508 * If it's full, it can't take another entry.
1509 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001510 if (be16_to_cpu(right->bb_numrecs) == XFS_ALLOC_BLOCK_MAXRECS(level, cur)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 *stat = 0;
1512 return 0;
1513 }
1514 /*
1515 * Make a hole at the start of the right neighbor block, then
1516 * copy the last left block entry to the hole.
1517 */
1518 if (level > 0) {
1519 xfs_alloc_key_t *lkp; /* key pointer for left block */
1520 xfs_alloc_ptr_t *lpp; /* address pointer for left block */
1521 xfs_alloc_ptr_t *rpp; /* address pointer for right block */
1522
Christoph Hellwig16259e72005-11-02 15:11:25 +11001523 lkp = XFS_ALLOC_KEY_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
1524 lpp = XFS_ALLOC_PTR_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur);
1526 rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur);
1527#ifdef DEBUG
Christoph Hellwig16259e72005-11-02 15:11:25 +11001528 for (i = be16_to_cpu(right->bb_numrecs) - 1; i >= 0; i--) {
1529 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i]), level)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 return error;
1531 }
1532#endif
Christoph Hellwig16259e72005-11-02 15:11:25 +11001533 memmove(rkp + 1, rkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
1534 memmove(rpp + 1, rpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535#ifdef DEBUG
Christoph Hellwig16259e72005-11-02 15:11:25 +11001536 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(*lpp), level)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 return error;
1538#endif
Christoph Hellwigc38e5e82006-09-28 10:57:17 +10001539 *rkp = *lkp;
1540 *rpp = *lpp;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001541 xfs_alloc_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
1542 xfs_alloc_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 xfs_btree_check_key(cur->bc_btnum, rkp, rkp + 1);
1544 } else {
1545 xfs_alloc_rec_t *lrp; /* record pointer for left block */
1546 xfs_alloc_rec_t *rrp; /* record pointer for right block */
1547
Christoph Hellwig16259e72005-11-02 15:11:25 +11001548 lrp = XFS_ALLOC_REC_ADDR(left, be16_to_cpu(left->bb_numrecs), cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 rrp = XFS_ALLOC_REC_ADDR(right, 1, cur);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001550 memmove(rrp + 1, rrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 *rrp = *lrp;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001552 xfs_alloc_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs) + 1);
1553 key.ar_startblock = rrp->ar_startblock;
1554 key.ar_blockcount = rrp->ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 rkp = &key;
1556 xfs_btree_check_rec(cur->bc_btnum, rrp, rrp + 1);
1557 }
1558 /*
1559 * Decrement and log left's numrecs, bump and log right's numrecs.
1560 */
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001561 be16_add_cpu(&left->bb_numrecs, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS);
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001563 be16_add_cpu(&right->bb_numrecs, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 xfs_alloc_log_block(cur->bc_tp, rbp, XFS_BB_NUMRECS);
1565 /*
1566 * Using a temporary cursor, update the parent key values of the
1567 * block on the right.
1568 */
1569 if ((error = xfs_btree_dup_cursor(cur, &tcur)))
1570 return error;
1571 i = xfs_btree_lastrec(tcur, level);
1572 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1573 if ((error = xfs_alloc_increment(tcur, level, &i)) ||
1574 (error = xfs_alloc_updkey(tcur, rkp, level + 1)))
1575 goto error0;
1576 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1577 *stat = 1;
1578 return 0;
1579error0:
1580 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1581 return error;
1582}
1583
1584/*
1585 * Split cur/level block in half.
1586 * Return new block number and its first record (to be inserted into parent).
1587 */
1588STATIC int /* error */
1589xfs_alloc_split(
1590 xfs_btree_cur_t *cur, /* btree cursor */
1591 int level, /* level to split */
1592 xfs_agblock_t *bnop, /* output: block number allocated */
1593 xfs_alloc_key_t *keyp, /* output: first key of new block */
1594 xfs_btree_cur_t **curp, /* output: new cursor */
1595 int *stat) /* success/failure */
1596{
1597 int error; /* error return value */
1598 int i; /* loop index/record number */
1599 xfs_agblock_t lbno; /* left (current) block number */
1600 xfs_buf_t *lbp; /* buffer for left block */
1601 xfs_alloc_block_t *left; /* left (current) btree block */
1602 xfs_agblock_t rbno; /* right (new) block number */
1603 xfs_buf_t *rbp; /* buffer for right block */
1604 xfs_alloc_block_t *right; /* right (new) btree block */
1605
1606 /*
1607 * Allocate the new block from the freelist.
1608 * If we can't do it, we're toast. Give up.
1609 */
David Chinner92821e22007-05-24 15:26:31 +10001610 error = xfs_alloc_get_freelist(cur->bc_tp,
1611 cur->bc_private.a.agbp, &rbno, 1);
1612 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 return error;
1614 if (rbno == NULLAGBLOCK) {
1615 *stat = 0;
1616 return 0;
1617 }
1618 xfs_trans_agbtree_delta(cur->bc_tp, 1);
1619 rbp = xfs_btree_get_bufs(cur->bc_mp, cur->bc_tp, cur->bc_private.a.agno,
1620 rbno, 0);
1621 /*
1622 * Set up the new block as "right".
1623 */
1624 right = XFS_BUF_TO_ALLOC_BLOCK(rbp);
1625 /*
1626 * "Left" is the current (according to the cursor) block.
1627 */
1628 lbp = cur->bc_bufs[level];
1629 left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
1630#ifdef DEBUG
1631 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
1632 return error;
1633#endif
1634 /*
1635 * Fill in the btree header for the new block.
1636 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001637 right->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
1638 right->bb_level = left->bb_level;
1639 right->bb_numrecs = cpu_to_be16(be16_to_cpu(left->bb_numrecs) / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 /*
1641 * Make sure that if there's an odd number of entries now, that
1642 * each new block will have the same number of entries.
1643 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001644 if ((be16_to_cpu(left->bb_numrecs) & 1) &&
1645 cur->bc_ptrs[level] <= be16_to_cpu(right->bb_numrecs) + 1)
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001646 be16_add_cpu(&right->bb_numrecs, 1);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001647 i = be16_to_cpu(left->bb_numrecs) - be16_to_cpu(right->bb_numrecs) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 /*
1649 * For non-leaf blocks, copy keys and addresses over to the new block.
1650 */
1651 if (level > 0) {
1652 xfs_alloc_key_t *lkp; /* left btree key pointer */
1653 xfs_alloc_ptr_t *lpp; /* left btree address pointer */
1654 xfs_alloc_key_t *rkp; /* right btree key pointer */
1655 xfs_alloc_ptr_t *rpp; /* right btree address pointer */
1656
1657 lkp = XFS_ALLOC_KEY_ADDR(left, i, cur);
1658 lpp = XFS_ALLOC_PTR_ADDR(left, i, cur);
1659 rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur);
1660 rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur);
1661#ifdef DEBUG
Christoph Hellwig16259e72005-11-02 15:11:25 +11001662 for (i = 0; i < be16_to_cpu(right->bb_numrecs); i++) {
1663 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(lpp[i]), level)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 return error;
1665 }
1666#endif
Christoph Hellwig16259e72005-11-02 15:11:25 +11001667 memcpy(rkp, lkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
1668 memcpy(rpp, lpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
1669 xfs_alloc_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
1670 xfs_alloc_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 *keyp = *rkp;
1672 }
1673 /*
1674 * For leaf blocks, copy records over to the new block.
1675 */
1676 else {
1677 xfs_alloc_rec_t *lrp; /* left btree record pointer */
1678 xfs_alloc_rec_t *rrp; /* right btree record pointer */
1679
1680 lrp = XFS_ALLOC_REC_ADDR(left, i, cur);
1681 rrp = XFS_ALLOC_REC_ADDR(right, 1, cur);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001682 memcpy(rrp, lrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
1683 xfs_alloc_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
1684 keyp->ar_startblock = rrp->ar_startblock;
1685 keyp->ar_blockcount = rrp->ar_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 }
1687 /*
1688 * Find the left block number by looking in the buffer.
1689 * Adjust numrecs, sibling pointers.
1690 */
1691 lbno = XFS_DADDR_TO_AGBNO(cur->bc_mp, XFS_BUF_ADDR(lbp));
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001692 be16_add_cpu(&left->bb_numrecs, -(be16_to_cpu(right->bb_numrecs)));
Christoph Hellwig16259e72005-11-02 15:11:25 +11001693 right->bb_rightsib = left->bb_rightsib;
1694 left->bb_rightsib = cpu_to_be32(rbno);
1695 right->bb_leftsib = cpu_to_be32(lbno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 xfs_alloc_log_block(cur->bc_tp, rbp, XFS_BB_ALL_BITS);
1697 xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
1698 /*
1699 * If there's a block to the new block's right, make that block
1700 * point back to right instead of to left.
1701 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001702 if (be32_to_cpu(right->bb_rightsib) != NULLAGBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 xfs_alloc_block_t *rrblock; /* rr btree block */
1704 xfs_buf_t *rrbp; /* buffer for rrblock */
1705
1706 if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
Christoph Hellwig16259e72005-11-02 15:11:25 +11001707 cur->bc_private.a.agno, be32_to_cpu(right->bb_rightsib), 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 &rrbp, XFS_ALLOC_BTREE_REF)))
1709 return error;
1710 rrblock = XFS_BUF_TO_ALLOC_BLOCK(rrbp);
1711 if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp)))
1712 return error;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001713 rrblock->bb_leftsib = cpu_to_be32(rbno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 xfs_alloc_log_block(cur->bc_tp, rrbp, XFS_BB_LEFTSIB);
1715 }
1716 /*
1717 * If the cursor is really in the right block, move it there.
1718 * If it's just pointing past the last entry in left, then we'll
1719 * insert there, so don't change anything in that case.
1720 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001721 if (cur->bc_ptrs[level] > be16_to_cpu(left->bb_numrecs) + 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 xfs_btree_setbuf(cur, level, rbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001723 cur->bc_ptrs[level] -= be16_to_cpu(left->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 }
1725 /*
1726 * If there are more levels, we'll need another cursor which refers to
1727 * the right block, no matter where this cursor was.
1728 */
1729 if (level + 1 < cur->bc_nlevels) {
1730 if ((error = xfs_btree_dup_cursor(cur, curp)))
1731 return error;
1732 (*curp)->bc_ptrs[level + 1]++;
1733 }
1734 *bnop = rbno;
1735 *stat = 1;
1736 return 0;
1737}
1738
1739/*
1740 * Update keys at all levels from here to the root along the cursor's path.
1741 */
1742STATIC int /* error */
1743xfs_alloc_updkey(
1744 xfs_btree_cur_t *cur, /* btree cursor */
1745 xfs_alloc_key_t *keyp, /* new key value to update to */
1746 int level) /* starting level for update */
1747{
1748 int ptr; /* index of key in block */
1749
1750 /*
1751 * Go up the tree from this level toward the root.
1752 * At each level, update the key value to the value input.
1753 * Stop when we reach a level where the cursor isn't pointing
1754 * at the first entry in the block.
1755 */
1756 for (ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
1757 xfs_alloc_block_t *block; /* btree block */
1758 xfs_buf_t *bp; /* buffer for block */
1759#ifdef DEBUG
1760 int error; /* error return value */
1761#endif
1762 xfs_alloc_key_t *kp; /* ptr to btree block keys */
1763
1764 bp = cur->bc_bufs[level];
1765 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
1766#ifdef DEBUG
1767 if ((error = xfs_btree_check_sblock(cur, block, level, bp)))
1768 return error;
1769#endif
1770 ptr = cur->bc_ptrs[level];
1771 kp = XFS_ALLOC_KEY_ADDR(block, ptr, cur);
1772 *kp = *keyp;
1773 xfs_alloc_log_keys(cur, bp, ptr, ptr);
1774 }
1775 return 0;
1776}
1777
1778/*
1779 * Externally visible routines.
1780 */
1781
1782/*
1783 * Decrement cursor by one record at the level.
1784 * For nonzero levels the leaf-ward information is untouched.
1785 */
1786int /* error */
1787xfs_alloc_decrement(
1788 xfs_btree_cur_t *cur, /* btree cursor */
1789 int level, /* level in btree, 0 is leaf */
1790 int *stat) /* success/failure */
1791{
1792 xfs_alloc_block_t *block; /* btree block */
1793 int error; /* error return value */
1794 int lev; /* btree level */
1795
1796 ASSERT(level < cur->bc_nlevels);
1797 /*
1798 * Read-ahead to the left at this level.
1799 */
1800 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1801 /*
1802 * Decrement the ptr at this level. If we're still in the block
1803 * then we're done.
1804 */
1805 if (--cur->bc_ptrs[level] > 0) {
1806 *stat = 1;
1807 return 0;
1808 }
1809 /*
1810 * Get a pointer to the btree block.
1811 */
1812 block = XFS_BUF_TO_ALLOC_BLOCK(cur->bc_bufs[level]);
1813#ifdef DEBUG
1814 if ((error = xfs_btree_check_sblock(cur, block, level,
1815 cur->bc_bufs[level])))
1816 return error;
1817#endif
1818 /*
1819 * If we just went off the left edge of the tree, return failure.
1820 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001821 if (be32_to_cpu(block->bb_leftsib) == NULLAGBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 *stat = 0;
1823 return 0;
1824 }
1825 /*
1826 * March up the tree decrementing pointers.
1827 * Stop when we don't go off the left edge of a block.
1828 */
1829 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1830 if (--cur->bc_ptrs[lev] > 0)
1831 break;
1832 /*
1833 * Read-ahead the left block, we're going to read it
1834 * in the next loop.
1835 */
1836 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1837 }
1838 /*
1839 * If we went off the root then we are seriously confused.
1840 */
1841 ASSERT(lev < cur->bc_nlevels);
1842 /*
1843 * Now walk back down the tree, fixing up the cursor's buffer
1844 * pointers and key numbers.
1845 */
1846 for (block = XFS_BUF_TO_ALLOC_BLOCK(cur->bc_bufs[lev]); lev > level; ) {
1847 xfs_agblock_t agbno; /* block number of btree block */
1848 xfs_buf_t *bp; /* buffer pointer for block */
1849
Christoph Hellwig16259e72005-11-02 15:11:25 +11001850 agbno = be32_to_cpu(*XFS_ALLOC_PTR_ADDR(block, cur->bc_ptrs[lev], cur));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
1852 cur->bc_private.a.agno, agbno, 0, &bp,
1853 XFS_ALLOC_BTREE_REF)))
1854 return error;
1855 lev--;
1856 xfs_btree_setbuf(cur, lev, bp);
1857 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
1858 if ((error = xfs_btree_check_sblock(cur, block, lev, bp)))
1859 return error;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001860 cur->bc_ptrs[lev] = be16_to_cpu(block->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 }
1862 *stat = 1;
1863 return 0;
1864}
1865
1866/*
1867 * Delete the record pointed to by cur.
1868 * The cursor refers to the place where the record was (could be inserted)
1869 * when the operation returns.
1870 */
1871int /* error */
1872xfs_alloc_delete(
1873 xfs_btree_cur_t *cur, /* btree cursor */
1874 int *stat) /* success/failure */
1875{
1876 int error; /* error return value */
1877 int i; /* result code */
1878 int level; /* btree level */
1879
1880 /*
1881 * Go up the tree, starting at leaf level.
1882 * If 2 is returned then a join was done; go to the next level.
1883 * Otherwise we are done.
1884 */
1885 for (level = 0, i = 2; i == 2; level++) {
1886 if ((error = xfs_alloc_delrec(cur, level, &i)))
1887 return error;
1888 }
1889 if (i == 0) {
1890 for (level = 1; level < cur->bc_nlevels; level++) {
1891 if (cur->bc_ptrs[level] == 0) {
1892 if ((error = xfs_alloc_decrement(cur, level, &i)))
1893 return error;
1894 break;
1895 }
1896 }
1897 }
1898 *stat = i;
1899 return 0;
1900}
1901
1902/*
1903 * Get the data from the pointed-to record.
1904 */
1905int /* error */
1906xfs_alloc_get_rec(
1907 xfs_btree_cur_t *cur, /* btree cursor */
1908 xfs_agblock_t *bno, /* output: starting block of extent */
1909 xfs_extlen_t *len, /* output: length of extent */
1910 int *stat) /* output: success/failure */
1911{
1912 xfs_alloc_block_t *block; /* btree block */
1913#ifdef DEBUG
1914 int error; /* error return value */
1915#endif
1916 int ptr; /* record number */
1917
1918 ptr = cur->bc_ptrs[0];
1919 block = XFS_BUF_TO_ALLOC_BLOCK(cur->bc_bufs[0]);
1920#ifdef DEBUG
1921 if ((error = xfs_btree_check_sblock(cur, block, 0, cur->bc_bufs[0])))
1922 return error;
1923#endif
1924 /*
1925 * Off the right end or left end, return failure.
1926 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001927 if (ptr > be16_to_cpu(block->bb_numrecs) || ptr <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 *stat = 0;
1929 return 0;
1930 }
1931 /*
1932 * Point to the record and extract its data.
1933 */
1934 {
1935 xfs_alloc_rec_t *rec; /* record data */
1936
1937 rec = XFS_ALLOC_REC_ADDR(block, ptr, cur);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001938 *bno = be32_to_cpu(rec->ar_startblock);
1939 *len = be32_to_cpu(rec->ar_blockcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 }
1941 *stat = 1;
1942 return 0;
1943}
1944
1945/*
1946 * Increment cursor by one record at the level.
1947 * For nonzero levels the leaf-ward information is untouched.
1948 */
1949int /* error */
1950xfs_alloc_increment(
1951 xfs_btree_cur_t *cur, /* btree cursor */
1952 int level, /* level in btree, 0 is leaf */
1953 int *stat) /* success/failure */
1954{
1955 xfs_alloc_block_t *block; /* btree block */
1956 xfs_buf_t *bp; /* tree block buffer */
1957 int error; /* error return value */
1958 int lev; /* btree level */
1959
1960 ASSERT(level < cur->bc_nlevels);
1961 /*
1962 * Read-ahead to the right at this level.
1963 */
1964 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1965 /*
1966 * Get a pointer to the btree block.
1967 */
1968 bp = cur->bc_bufs[level];
1969 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
1970#ifdef DEBUG
1971 if ((error = xfs_btree_check_sblock(cur, block, level, bp)))
1972 return error;
1973#endif
1974 /*
1975 * Increment the ptr at this level. If we're still in the block
1976 * then we're done.
1977 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001978 if (++cur->bc_ptrs[level] <= be16_to_cpu(block->bb_numrecs)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 *stat = 1;
1980 return 0;
1981 }
1982 /*
1983 * If we just went off the right edge of the tree, return failure.
1984 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001985 if (be32_to_cpu(block->bb_rightsib) == NULLAGBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 *stat = 0;
1987 return 0;
1988 }
1989 /*
1990 * March up the tree incrementing pointers.
1991 * Stop when we don't go off the right edge of a block.
1992 */
1993 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1994 bp = cur->bc_bufs[lev];
1995 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
1996#ifdef DEBUG
1997 if ((error = xfs_btree_check_sblock(cur, block, lev, bp)))
1998 return error;
1999#endif
Christoph Hellwig16259e72005-11-02 15:11:25 +11002000 if (++cur->bc_ptrs[lev] <= be16_to_cpu(block->bb_numrecs))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 break;
2002 /*
2003 * Read-ahead the right block, we're going to read it
2004 * in the next loop.
2005 */
2006 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
2007 }
2008 /*
2009 * If we went off the root then we are seriously confused.
2010 */
2011 ASSERT(lev < cur->bc_nlevels);
2012 /*
2013 * Now walk back down the tree, fixing up the cursor's buffer
2014 * pointers and key numbers.
2015 */
2016 for (bp = cur->bc_bufs[lev], block = XFS_BUF_TO_ALLOC_BLOCK(bp);
2017 lev > level; ) {
2018 xfs_agblock_t agbno; /* block number of btree block */
2019
Christoph Hellwig16259e72005-11-02 15:11:25 +11002020 agbno = be32_to_cpu(*XFS_ALLOC_PTR_ADDR(block, cur->bc_ptrs[lev], cur));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
2022 cur->bc_private.a.agno, agbno, 0, &bp,
2023 XFS_ALLOC_BTREE_REF)))
2024 return error;
2025 lev--;
2026 xfs_btree_setbuf(cur, lev, bp);
2027 block = XFS_BUF_TO_ALLOC_BLOCK(bp);
2028 if ((error = xfs_btree_check_sblock(cur, block, lev, bp)))
2029 return error;
2030 cur->bc_ptrs[lev] = 1;
2031 }
2032 *stat = 1;
2033 return 0;
2034}
2035
2036/*
2037 * Insert the current record at the point referenced by cur.
2038 * The cursor may be inconsistent on return if splits have been done.
2039 */
2040int /* error */
2041xfs_alloc_insert(
2042 xfs_btree_cur_t *cur, /* btree cursor */
2043 int *stat) /* success/failure */
2044{
2045 int error; /* error return value */
2046 int i; /* result value, 0 for failure */
2047 int level; /* current level number in btree */
2048 xfs_agblock_t nbno; /* new block number (split result) */
2049 xfs_btree_cur_t *ncur; /* new cursor (split result) */
2050 xfs_alloc_rec_t nrec; /* record being inserted this level */
2051 xfs_btree_cur_t *pcur; /* previous level's cursor */
2052
2053 level = 0;
2054 nbno = NULLAGBLOCK;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002055 nrec.ar_startblock = cpu_to_be32(cur->bc_rec.a.ar_startblock);
2056 nrec.ar_blockcount = cpu_to_be32(cur->bc_rec.a.ar_blockcount);
Nathan Scott1121b212006-09-28 10:58:40 +10002057 ncur = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 pcur = cur;
2059 /*
2060 * Loop going up the tree, starting at the leaf level.
2061 * Stop when we don't get a split block, that must mean that
2062 * the insert is finished with this level.
2063 */
2064 do {
2065 /*
2066 * Insert nrec/nbno into this level of the tree.
2067 * Note if we fail, nbno will be null.
2068 */
2069 if ((error = xfs_alloc_insrec(pcur, level++, &nbno, &nrec, &ncur,
2070 &i))) {
2071 if (pcur != cur)
2072 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
2073 return error;
2074 }
2075 /*
2076 * See if the cursor we just used is trash.
2077 * Can't trash the caller's cursor, but otherwise we should
2078 * if ncur is a new cursor or we're about to be done.
2079 */
2080 if (pcur != cur && (ncur || nbno == NULLAGBLOCK)) {
2081 cur->bc_nlevels = pcur->bc_nlevels;
2082 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
2083 }
2084 /*
2085 * If we got a new cursor, switch to it.
2086 */
2087 if (ncur) {
2088 pcur = ncur;
Nathan Scott1121b212006-09-28 10:58:40 +10002089 ncur = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 }
2091 } while (nbno != NULLAGBLOCK);
2092 *stat = i;
2093 return 0;
2094}
2095
2096/*
2097 * Lookup the record equal to [bno, len] in the btree given by cur.
2098 */
2099int /* error */
2100xfs_alloc_lookup_eq(
2101 xfs_btree_cur_t *cur, /* btree cursor */
2102 xfs_agblock_t bno, /* starting block of extent */
2103 xfs_extlen_t len, /* length of extent */
2104 int *stat) /* success/failure */
2105{
2106 cur->bc_rec.a.ar_startblock = bno;
2107 cur->bc_rec.a.ar_blockcount = len;
2108 return xfs_alloc_lookup(cur, XFS_LOOKUP_EQ, stat);
2109}
2110
2111/*
2112 * Lookup the first record greater than or equal to [bno, len]
2113 * in the btree given by cur.
2114 */
2115int /* error */
2116xfs_alloc_lookup_ge(
2117 xfs_btree_cur_t *cur, /* btree cursor */
2118 xfs_agblock_t bno, /* starting block of extent */
2119 xfs_extlen_t len, /* length of extent */
2120 int *stat) /* success/failure */
2121{
2122 cur->bc_rec.a.ar_startblock = bno;
2123 cur->bc_rec.a.ar_blockcount = len;
2124 return xfs_alloc_lookup(cur, XFS_LOOKUP_GE, stat);
2125}
2126
2127/*
2128 * Lookup the first record less than or equal to [bno, len]
2129 * in the btree given by cur.
2130 */
2131int /* error */
2132xfs_alloc_lookup_le(
2133 xfs_btree_cur_t *cur, /* btree cursor */
2134 xfs_agblock_t bno, /* starting block of extent */
2135 xfs_extlen_t len, /* length of extent */
2136 int *stat) /* success/failure */
2137{
2138 cur->bc_rec.a.ar_startblock = bno;
2139 cur->bc_rec.a.ar_blockcount = len;
2140 return xfs_alloc_lookup(cur, XFS_LOOKUP_LE, stat);
2141}
2142
2143/*
2144 * Update the record referred to by cur, to the value given by [bno, len].
2145 * This either works (return 0) or gets an EFSCORRUPTED error.
2146 */
2147int /* error */
2148xfs_alloc_update(
2149 xfs_btree_cur_t *cur, /* btree cursor */
2150 xfs_agblock_t bno, /* starting block of extent */
2151 xfs_extlen_t len) /* length of extent */
2152{
2153 xfs_alloc_block_t *block; /* btree block to update */
2154 int error; /* error return value */
2155 int ptr; /* current record number (updating) */
2156
2157 ASSERT(len > 0);
2158 /*
2159 * Pick up the a.g. freelist struct and the current block.
2160 */
2161 block = XFS_BUF_TO_ALLOC_BLOCK(cur->bc_bufs[0]);
2162#ifdef DEBUG
2163 if ((error = xfs_btree_check_sblock(cur, block, 0, cur->bc_bufs[0])))
2164 return error;
2165#endif
2166 /*
2167 * Get the address of the rec to be updated.
2168 */
2169 ptr = cur->bc_ptrs[0];
2170 {
2171 xfs_alloc_rec_t *rp; /* pointer to updated record */
2172
2173 rp = XFS_ALLOC_REC_ADDR(block, ptr, cur);
2174 /*
2175 * Fill in the new contents and log them.
2176 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002177 rp->ar_startblock = cpu_to_be32(bno);
2178 rp->ar_blockcount = cpu_to_be32(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 xfs_alloc_log_recs(cur, cur->bc_bufs[0], ptr, ptr);
2180 }
2181 /*
2182 * If it's the by-size btree and it's the last leaf block and
2183 * it's the last record... then update the size of the longest
2184 * extent in the a.g., which we cache in the a.g. freelist header.
2185 */
2186 if (cur->bc_btnum == XFS_BTNUM_CNT &&
Christoph Hellwig16259e72005-11-02 15:11:25 +11002187 be32_to_cpu(block->bb_rightsib) == NULLAGBLOCK &&
2188 ptr == be16_to_cpu(block->bb_numrecs)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 xfs_agf_t *agf; /* a.g. freespace header */
2190 xfs_agnumber_t seqno;
2191
2192 agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002193 seqno = be32_to_cpu(agf->agf_seqno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 cur->bc_mp->m_perag[seqno].pagf_longest = len;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002195 agf->agf_longest = cpu_to_be32(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp,
2197 XFS_AGF_LONGEST);
2198 }
2199 /*
2200 * Updating first record in leaf. Pass new key value up to our parent.
2201 */
2202 if (ptr == 1) {
2203 xfs_alloc_key_t key; /* key containing [bno, len] */
2204
Christoph Hellwig16259e72005-11-02 15:11:25 +11002205 key.ar_startblock = cpu_to_be32(bno);
2206 key.ar_blockcount = cpu_to_be32(len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 if ((error = xfs_alloc_updkey(cur, &key, 1)))
2208 return error;
2209 }
2210 return 0;
2211}