[XFS] implement generic xfs_btree_rshift
[deliverable/linux.git] / fs / xfs / xfs_ialloc_btree.c
CommitLineData
1da177e4 1/*
7b718769
NS
2 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
1da177e4 4 *
7b718769
NS
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
1da177e4
LT
7 * published by the Free Software Foundation.
8 *
7b718769
NS
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
1da177e4 13 *
7b718769
NS
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
1da177e4 17 */
1da177e4 18#include "xfs.h"
a844f451 19#include "xfs_fs.h"
1da177e4 20#include "xfs_types.h"
a844f451 21#include "xfs_bit.h"
1da177e4 22#include "xfs_log.h"
a844f451 23#include "xfs_inum.h"
1da177e4
LT
24#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
a844f451 27#include "xfs_dir2.h"
1da177e4
LT
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
1da177e4 30#include "xfs_bmap_btree.h"
a844f451 31#include "xfs_alloc_btree.h"
1da177e4 32#include "xfs_ialloc_btree.h"
a844f451
NS
33#include "xfs_dir2_sf.h"
34#include "xfs_attr_sf.h"
35#include "xfs_dinode.h"
36#include "xfs_inode.h"
1da177e4
LT
37#include "xfs_btree.h"
38#include "xfs_ialloc.h"
39#include "xfs_alloc.h"
40#include "xfs_error.h"
41
1da177e4
LT
42STATIC void xfs_inobt_log_block(xfs_trans_t *, xfs_buf_t *, int);
43STATIC void xfs_inobt_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int);
44STATIC void xfs_inobt_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
45STATIC void xfs_inobt_log_recs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
46STATIC int xfs_inobt_lshift(xfs_btree_cur_t *, int, int *);
47STATIC int xfs_inobt_newroot(xfs_btree_cur_t *, int *);
1da177e4
LT
48STATIC int xfs_inobt_split(xfs_btree_cur_t *, int, xfs_agblock_t *,
49 xfs_inobt_key_t *, xfs_btree_cur_t **, int *);
1da177e4 50
1da177e4
LT
51/*
52 * Single level of the xfs_inobt_delete record deletion routine.
53 * Delete record pointed to by cur/level.
54 * Remove the record from its block then rebalance the tree.
55 * Return 0 for error, 1 for done, 2 to go on to the next level.
56 */
57STATIC int /* error */
58xfs_inobt_delrec(
59 xfs_btree_cur_t *cur, /* btree cursor */
60 int level, /* level removing record from */
61 int *stat) /* fail/done/go-on */
62{
63 xfs_buf_t *agbp; /* buffer for a.g. inode header */
64 xfs_mount_t *mp; /* mount structure */
65 xfs_agi_t *agi; /* allocation group inode header */
66 xfs_inobt_block_t *block; /* btree block record/key lives in */
67 xfs_agblock_t bno; /* btree block number */
68 xfs_buf_t *bp; /* buffer for block */
69 int error; /* error return value */
70 int i; /* loop index */
71 xfs_inobt_key_t key; /* kp points here if block is level 0 */
72 xfs_inobt_key_t *kp = NULL; /* pointer to btree keys */
73 xfs_agblock_t lbno; /* left block's block number */
74 xfs_buf_t *lbp; /* left block's buffer pointer */
75 xfs_inobt_block_t *left; /* left btree block */
76 xfs_inobt_key_t *lkp; /* left block key pointer */
77 xfs_inobt_ptr_t *lpp; /* left block address pointer */
78 int lrecs = 0; /* number of records in left block */
79 xfs_inobt_rec_t *lrp; /* left block record pointer */
80 xfs_inobt_ptr_t *pp = NULL; /* pointer to btree addresses */
81 int ptr; /* index in btree block for this rec */
82 xfs_agblock_t rbno; /* right block's block number */
83 xfs_buf_t *rbp; /* right block's buffer pointer */
84 xfs_inobt_block_t *right; /* right btree block */
85 xfs_inobt_key_t *rkp; /* right block key pointer */
86 xfs_inobt_rec_t *rp; /* pointer to btree records */
87 xfs_inobt_ptr_t *rpp; /* right block address pointer */
88 int rrecs = 0; /* number of records in right block */
89 int numrecs;
90 xfs_inobt_rec_t *rrp; /* right block record pointer */
91 xfs_btree_cur_t *tcur; /* temporary btree cursor */
92
93 mp = cur->bc_mp;
94
95 /*
96 * Get the index of the entry being deleted, check for nothing there.
97 */
98 ptr = cur->bc_ptrs[level];
99 if (ptr == 0) {
100 *stat = 0;
101 return 0;
102 }
103
104 /*
105 * Get the buffer & block containing the record or key/ptr.
106 */
107 bp = cur->bc_bufs[level];
108 block = XFS_BUF_TO_INOBT_BLOCK(bp);
109#ifdef DEBUG
110 if ((error = xfs_btree_check_sblock(cur, block, level, bp)))
111 return error;
112#endif
113 /*
114 * Fail if we're off the end of the block.
115 */
116
16259e7d 117 numrecs = be16_to_cpu(block->bb_numrecs);
1da177e4
LT
118 if (ptr > numrecs) {
119 *stat = 0;
120 return 0;
121 }
122 /*
123 * It's a nonleaf. Excise the key and ptr being deleted, by
124 * sliding the entries past them down one.
125 * Log the changed areas of the block.
126 */
127 if (level > 0) {
128 kp = XFS_INOBT_KEY_ADDR(block, 1, cur);
129 pp = XFS_INOBT_PTR_ADDR(block, 1, cur);
130#ifdef DEBUG
131 for (i = ptr; i < numrecs; i++) {
16259e7d 132 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(pp[i]), level)))
1da177e4
LT
133 return error;
134 }
135#endif
136 if (ptr < numrecs) {
137 memmove(&kp[ptr - 1], &kp[ptr],
138 (numrecs - ptr) * sizeof(*kp));
139 memmove(&pp[ptr - 1], &pp[ptr],
140 (numrecs - ptr) * sizeof(*kp));
141 xfs_inobt_log_keys(cur, bp, ptr, numrecs - 1);
142 xfs_inobt_log_ptrs(cur, bp, ptr, numrecs - 1);
143 }
144 }
145 /*
146 * It's a leaf. Excise the record being deleted, by sliding the
147 * entries past it down one. Log the changed areas of the block.
148 */
149 else {
150 rp = XFS_INOBT_REC_ADDR(block, 1, cur);
151 if (ptr < numrecs) {
152 memmove(&rp[ptr - 1], &rp[ptr],
153 (numrecs - ptr) * sizeof(*rp));
154 xfs_inobt_log_recs(cur, bp, ptr, numrecs - 1);
155 }
156 /*
157 * If it's the first record in the block, we'll need a key
158 * structure to pass up to the next level (updkey).
159 */
160 if (ptr == 1) {
161 key.ir_startino = rp->ir_startino;
162 kp = &key;
163 }
164 }
165 /*
166 * Decrement and log the number of entries in the block.
167 */
168 numrecs--;
16259e7d 169 block->bb_numrecs = cpu_to_be16(numrecs);
1da177e4
LT
170 xfs_inobt_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS);
171 /*
172 * Is this the root level? If so, we're almost done.
173 */
174 if (level == cur->bc_nlevels - 1) {
175 /*
176 * If this is the root level,
177 * and there's only one entry left,
178 * and it's NOT the leaf level,
179 * then we can get rid of this level.
180 */
181 if (numrecs == 1 && level > 0) {
169d6227 182 agbp = cur->bc_private.a.agbp;
1da177e4
LT
183 agi = XFS_BUF_TO_AGI(agbp);
184 /*
185 * pp is still set to the first pointer in the block.
186 * Make it the new root of the btree.
187 */
16259e7d 188 bno = be32_to_cpu(agi->agi_root);
1da177e4 189 agi->agi_root = *pp;
413d57c9 190 be32_add_cpu(&agi->agi_level, -1);
1da177e4
LT
191 /*
192 * Free the block.
193 */
194 if ((error = xfs_free_extent(cur->bc_tp,
169d6227 195 XFS_AGB_TO_FSB(mp, cur->bc_private.a.agno, bno), 1)))
1da177e4
LT
196 return error;
197 xfs_trans_binval(cur->bc_tp, bp);
198 xfs_ialloc_log_agi(cur->bc_tp, agbp,
199 XFS_AGI_ROOT | XFS_AGI_LEVEL);
200 /*
201 * Update the cursor so there's one fewer level.
202 */
203 cur->bc_bufs[level] = NULL;
204 cur->bc_nlevels--;
205 } else if (level > 0 &&
8df4da4a 206 (error = xfs_btree_decrement(cur, level, &i)))
1da177e4
LT
207 return error;
208 *stat = 1;
209 return 0;
210 }
211 /*
212 * If we deleted the leftmost entry in the block, update the
213 * key values above us in the tree.
214 */
38bb7423 215 if (ptr == 1 && (error = xfs_btree_updkey(cur, (union xfs_btree_key *)kp, level + 1)))
1da177e4
LT
216 return error;
217 /*
218 * If the number of records remaining in the block is at least
219 * the minimum, we're done.
220 */
221 if (numrecs >= XFS_INOBT_BLOCK_MINRECS(level, cur)) {
222 if (level > 0 &&
8df4da4a 223 (error = xfs_btree_decrement(cur, level, &i)))
1da177e4
LT
224 return error;
225 *stat = 1;
226 return 0;
227 }
228 /*
229 * Otherwise, we have to move some records around to keep the
230 * tree balanced. Look at the left and right sibling blocks to
231 * see if we can re-balance by moving only one record.
232 */
16259e7d
CH
233 rbno = be32_to_cpu(block->bb_rightsib);
234 lbno = be32_to_cpu(block->bb_leftsib);
1da177e4
LT
235 bno = NULLAGBLOCK;
236 ASSERT(rbno != NULLAGBLOCK || lbno != NULLAGBLOCK);
237 /*
238 * Duplicate the cursor so our btree manipulations here won't
239 * disrupt the next level up.
240 */
241 if ((error = xfs_btree_dup_cursor(cur, &tcur)))
242 return error;
243 /*
244 * If there's a right sibling, see if it's ok to shift an entry
245 * out of it.
246 */
247 if (rbno != NULLAGBLOCK) {
248 /*
249 * Move the temp cursor to the last entry in the next block.
250 * Actually any entry but the first would suffice.
251 */
252 i = xfs_btree_lastrec(tcur, level);
253 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
637aa50f 254 if ((error = xfs_btree_increment(tcur, level, &i)))
1da177e4
LT
255 goto error0;
256 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
257 i = xfs_btree_lastrec(tcur, level);
258 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
259 /*
260 * Grab a pointer to the block.
261 */
262 rbp = tcur->bc_bufs[level];
263 right = XFS_BUF_TO_INOBT_BLOCK(rbp);
264#ifdef DEBUG
265 if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
266 goto error0;
267#endif
268 /*
269 * Grab the current block number, for future use.
270 */
16259e7d 271 bno = be32_to_cpu(right->bb_leftsib);
1da177e4
LT
272 /*
273 * If right block is full enough so that removing one entry
274 * won't make it too empty, and left-shifting an entry out
275 * of right to us works, we're done.
276 */
16259e7d 277 if (be16_to_cpu(right->bb_numrecs) - 1 >=
1da177e4
LT
278 XFS_INOBT_BLOCK_MINRECS(level, cur)) {
279 if ((error = xfs_inobt_lshift(tcur, level, &i)))
280 goto error0;
281 if (i) {
16259e7d 282 ASSERT(be16_to_cpu(block->bb_numrecs) >=
1da177e4
LT
283 XFS_INOBT_BLOCK_MINRECS(level, cur));
284 xfs_btree_del_cursor(tcur,
285 XFS_BTREE_NOERROR);
286 if (level > 0 &&
8df4da4a 287 (error = xfs_btree_decrement(cur, level,
1da177e4
LT
288 &i)))
289 return error;
290 *stat = 1;
291 return 0;
292 }
293 }
294 /*
295 * Otherwise, grab the number of records in right for
296 * future reference, and fix up the temp cursor to point
297 * to our block again (last record).
298 */
16259e7d 299 rrecs = be16_to_cpu(right->bb_numrecs);
1da177e4
LT
300 if (lbno != NULLAGBLOCK) {
301 xfs_btree_firstrec(tcur, level);
8df4da4a 302 if ((error = xfs_btree_decrement(tcur, level, &i)))
1da177e4
LT
303 goto error0;
304 }
305 }
306 /*
307 * If there's a left sibling, see if it's ok to shift an entry
308 * out of it.
309 */
310 if (lbno != NULLAGBLOCK) {
311 /*
312 * Move the temp cursor to the first entry in the
313 * previous block.
314 */
315 xfs_btree_firstrec(tcur, level);
8df4da4a 316 if ((error = xfs_btree_decrement(tcur, level, &i)))
1da177e4
LT
317 goto error0;
318 xfs_btree_firstrec(tcur, level);
319 /*
320 * Grab a pointer to the block.
321 */
322 lbp = tcur->bc_bufs[level];
323 left = XFS_BUF_TO_INOBT_BLOCK(lbp);
324#ifdef DEBUG
325 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
326 goto error0;
327#endif
328 /*
329 * Grab the current block number, for future use.
330 */
16259e7d 331 bno = be32_to_cpu(left->bb_rightsib);
1da177e4
LT
332 /*
333 * If left block is full enough so that removing one entry
334 * won't make it too empty, and right-shifting an entry out
335 * of left to us works, we're done.
336 */
16259e7d 337 if (be16_to_cpu(left->bb_numrecs) - 1 >=
1da177e4 338 XFS_INOBT_BLOCK_MINRECS(level, cur)) {
9eaead51 339 if ((error = xfs_btree_rshift(tcur, level, &i)))
1da177e4
LT
340 goto error0;
341 if (i) {
16259e7d 342 ASSERT(be16_to_cpu(block->bb_numrecs) >=
1da177e4
LT
343 XFS_INOBT_BLOCK_MINRECS(level, cur));
344 xfs_btree_del_cursor(tcur,
345 XFS_BTREE_NOERROR);
346 if (level == 0)
347 cur->bc_ptrs[0]++;
348 *stat = 1;
349 return 0;
350 }
351 }
352 /*
353 * Otherwise, grab the number of records in right for
354 * future reference.
355 */
16259e7d 356 lrecs = be16_to_cpu(left->bb_numrecs);
1da177e4
LT
357 }
358 /*
359 * Delete the temp cursor, we're done with it.
360 */
361 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
362 /*
363 * If here, we need to do a join to keep the tree balanced.
364 */
365 ASSERT(bno != NULLAGBLOCK);
366 /*
367 * See if we can join with the left neighbor block.
368 */
369 if (lbno != NULLAGBLOCK &&
370 lrecs + numrecs <= XFS_INOBT_BLOCK_MAXRECS(level, cur)) {
371 /*
372 * Set "right" to be the starting block,
373 * "left" to be the left neighbor.
374 */
375 rbno = bno;
376 right = block;
16259e7d 377 rrecs = be16_to_cpu(right->bb_numrecs);
1da177e4
LT
378 rbp = bp;
379 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
169d6227 380 cur->bc_private.a.agno, lbno, 0, &lbp,
1da177e4
LT
381 XFS_INO_BTREE_REF)))
382 return error;
383 left = XFS_BUF_TO_INOBT_BLOCK(lbp);
16259e7d 384 lrecs = be16_to_cpu(left->bb_numrecs);
1da177e4
LT
385 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
386 return error;
387 }
388 /*
389 * If that won't work, see if we can join with the right neighbor block.
390 */
391 else if (rbno != NULLAGBLOCK &&
392 rrecs + numrecs <= XFS_INOBT_BLOCK_MAXRECS(level, cur)) {
393 /*
394 * Set "left" to be the starting block,
395 * "right" to be the right neighbor.
396 */
397 lbno = bno;
398 left = block;
16259e7d 399 lrecs = be16_to_cpu(left->bb_numrecs);
1da177e4
LT
400 lbp = bp;
401 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
169d6227 402 cur->bc_private.a.agno, rbno, 0, &rbp,
1da177e4
LT
403 XFS_INO_BTREE_REF)))
404 return error;
405 right = XFS_BUF_TO_INOBT_BLOCK(rbp);
16259e7d 406 rrecs = be16_to_cpu(right->bb_numrecs);
1da177e4
LT
407 if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
408 return error;
409 }
410 /*
411 * Otherwise, we can't fix the imbalance.
412 * Just return. This is probably a logic error, but it's not fatal.
413 */
414 else {
8df4da4a 415 if (level > 0 && (error = xfs_btree_decrement(cur, level, &i)))
1da177e4
LT
416 return error;
417 *stat = 1;
418 return 0;
419 }
420 /*
421 * We're now going to join "left" and "right" by moving all the stuff
422 * in "right" to "left" and deleting "right".
423 */
424 if (level > 0) {
425 /*
426 * It's a non-leaf. Move keys and pointers.
427 */
428 lkp = XFS_INOBT_KEY_ADDR(left, lrecs + 1, cur);
429 lpp = XFS_INOBT_PTR_ADDR(left, lrecs + 1, cur);
430 rkp = XFS_INOBT_KEY_ADDR(right, 1, cur);
431 rpp = XFS_INOBT_PTR_ADDR(right, 1, cur);
432#ifdef DEBUG
433 for (i = 0; i < rrecs; i++) {
16259e7d 434 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i]), level)))
1da177e4
LT
435 return error;
436 }
437#endif
438 memcpy(lkp, rkp, rrecs * sizeof(*lkp));
439 memcpy(lpp, rpp, rrecs * sizeof(*lpp));
440 xfs_inobt_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
441 xfs_inobt_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
442 } else {
443 /*
444 * It's a leaf. Move records.
445 */
446 lrp = XFS_INOBT_REC_ADDR(left, lrecs + 1, cur);
447 rrp = XFS_INOBT_REC_ADDR(right, 1, cur);
448 memcpy(lrp, rrp, rrecs * sizeof(*lrp));
449 xfs_inobt_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
450 }
451 /*
452 * If we joined with the left neighbor, set the buffer in the
453 * cursor to the left block, and fix up the index.
454 */
455 if (bp != lbp) {
456 xfs_btree_setbuf(cur, level, lbp);
457 cur->bc_ptrs[level] += lrecs;
458 }
459 /*
460 * If we joined with the right neighbor and there's a level above
461 * us, increment the cursor at that level.
462 */
463 else if (level + 1 < cur->bc_nlevels &&
637aa50f 464 (error = xfs_btree_increment(cur, level + 1, &i)))
1da177e4
LT
465 return error;
466 /*
467 * Fix up the number of records in the surviving block.
468 */
469 lrecs += rrecs;
16259e7d 470 left->bb_numrecs = cpu_to_be16(lrecs);
1da177e4
LT
471 /*
472 * Fix up the right block pointer in the surviving block, and log it.
473 */
474 left->bb_rightsib = right->bb_rightsib;
475 xfs_inobt_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
476 /*
477 * If there is a right sibling now, make it point to the
478 * remaining block.
479 */
16259e7d 480 if (be32_to_cpu(left->bb_rightsib) != NULLAGBLOCK) {
1da177e4
LT
481 xfs_inobt_block_t *rrblock;
482 xfs_buf_t *rrbp;
483
484 if ((error = xfs_btree_read_bufs(mp, cur->bc_tp,
169d6227 485 cur->bc_private.a.agno, be32_to_cpu(left->bb_rightsib), 0,
1da177e4
LT
486 &rrbp, XFS_INO_BTREE_REF)))
487 return error;
488 rrblock = XFS_BUF_TO_INOBT_BLOCK(rrbp);
489 if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp)))
490 return error;
16259e7d 491 rrblock->bb_leftsib = cpu_to_be32(lbno);
1da177e4
LT
492 xfs_inobt_log_block(cur->bc_tp, rrbp, XFS_BB_LEFTSIB);
493 }
494 /*
495 * Free the deleting block.
496 */
497 if ((error = xfs_free_extent(cur->bc_tp, XFS_AGB_TO_FSB(mp,
169d6227 498 cur->bc_private.a.agno, rbno), 1)))
1da177e4
LT
499 return error;
500 xfs_trans_binval(cur->bc_tp, rbp);
501 /*
502 * Readjust the ptr at this level if it's not a leaf, since it's
503 * still pointing at the deletion point, which makes the cursor
504 * inconsistent. If this makes the ptr 0, the caller fixes it up.
505 * We can't use decrement because it would change the next level up.
506 */
507 if (level > 0)
508 cur->bc_ptrs[level]--;
509 /*
510 * Return value means the next level up has something to do.
511 */
512 *stat = 2;
513 return 0;
514
515error0:
516 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
517 return error;
518}
519
520/*
521 * Insert one record/level. Return information to the caller
522 * allowing the next level up to proceed if necessary.
523 */
524STATIC int /* error */
525xfs_inobt_insrec(
526 xfs_btree_cur_t *cur, /* btree cursor */
527 int level, /* level to insert record at */
528 xfs_agblock_t *bnop, /* i/o: block number inserted */
529 xfs_inobt_rec_t *recp, /* i/o: record data inserted */
530 xfs_btree_cur_t **curp, /* output: new cursor replacing cur */
531 int *stat) /* success/failure */
532{
533 xfs_inobt_block_t *block; /* btree block record/key lives in */
534 xfs_buf_t *bp; /* buffer for block */
535 int error; /* error return value */
536 int i; /* loop index */
537 xfs_inobt_key_t key; /* key value being inserted */
538 xfs_inobt_key_t *kp=NULL; /* pointer to btree keys */
539 xfs_agblock_t nbno; /* block number of allocated block */
540 xfs_btree_cur_t *ncur; /* new cursor to be used at next lvl */
541 xfs_inobt_key_t nkey; /* new key value, from split */
542 xfs_inobt_rec_t nrec; /* new record value, for caller */
543 int numrecs;
544 int optr; /* old ptr value */
545 xfs_inobt_ptr_t *pp; /* pointer to btree addresses */
546 int ptr; /* index in btree block for this rec */
547 xfs_inobt_rec_t *rp=NULL; /* pointer to btree records */
548
5bde1ba9
CH
549 /*
550 * GCC doesn't understand the (arguably complex) control flow in
551 * this function and complains about uninitialized structure fields
552 * without this.
553 */
554 memset(&nrec, 0, sizeof(nrec));
555
1da177e4
LT
556 /*
557 * If we made it to the root level, allocate a new root block
558 * and we're done.
559 */
560 if (level >= cur->bc_nlevels) {
561 error = xfs_inobt_newroot(cur, &i);
562 *bnop = NULLAGBLOCK;
563 *stat = i;
564 return error;
565 }
566 /*
567 * Make a key out of the record data to be inserted, and save it.
568 */
61a25848 569 key.ir_startino = recp->ir_startino;
1da177e4
LT
570 optr = ptr = cur->bc_ptrs[level];
571 /*
572 * If we're off the left edge, return failure.
573 */
574 if (ptr == 0) {
575 *stat = 0;
576 return 0;
577 }
578 /*
579 * Get pointers to the btree buffer and block.
580 */
581 bp = cur->bc_bufs[level];
582 block = XFS_BUF_TO_INOBT_BLOCK(bp);
16259e7d 583 numrecs = be16_to_cpu(block->bb_numrecs);
1da177e4
LT
584#ifdef DEBUG
585 if ((error = xfs_btree_check_sblock(cur, block, level, bp)))
586 return error;
587 /*
588 * Check that the new entry is being inserted in the right place.
589 */
590 if (ptr <= numrecs) {
591 if (level == 0) {
592 rp = XFS_INOBT_REC_ADDR(block, ptr, cur);
593 xfs_btree_check_rec(cur->bc_btnum, recp, rp);
594 } else {
595 kp = XFS_INOBT_KEY_ADDR(block, ptr, cur);
596 xfs_btree_check_key(cur->bc_btnum, &key, kp);
597 }
598 }
599#endif
600 nbno = NULLAGBLOCK;
1121b219 601 ncur = NULL;
1da177e4
LT
602 /*
603 * If the block is full, we can't insert the new entry until we
604 * make the block un-full.
605 */
606 if (numrecs == XFS_INOBT_BLOCK_MAXRECS(level, cur)) {
607 /*
608 * First, try shifting an entry to the right neighbor.
609 */
9eaead51 610 if ((error = xfs_btree_rshift(cur, level, &i)))
1da177e4
LT
611 return error;
612 if (i) {
613 /* nothing */
614 }
615 /*
616 * Next, try shifting an entry to the left neighbor.
617 */
618 else {
619 if ((error = xfs_inobt_lshift(cur, level, &i)))
620 return error;
621 if (i) {
622 optr = ptr = cur->bc_ptrs[level];
623 } else {
624 /*
625 * Next, try splitting the current block
626 * in half. If this works we have to
627 * re-set our variables because
628 * we could be in a different block now.
629 */
630 if ((error = xfs_inobt_split(cur, level, &nbno,
631 &nkey, &ncur, &i)))
632 return error;
633 if (i) {
634 bp = cur->bc_bufs[level];
635 block = XFS_BUF_TO_INOBT_BLOCK(bp);
636#ifdef DEBUG
637 if ((error = xfs_btree_check_sblock(cur,
638 block, level, bp)))
639 return error;
640#endif
641 ptr = cur->bc_ptrs[level];
61a25848 642 nrec.ir_startino = nkey.ir_startino;
1da177e4
LT
643 } else {
644 /*
645 * Otherwise the insert fails.
646 */
647 *stat = 0;
648 return 0;
649 }
650 }
651 }
652 }
653 /*
654 * At this point we know there's room for our new entry in the block
655 * we're pointing at.
656 */
16259e7d 657 numrecs = be16_to_cpu(block->bb_numrecs);
1da177e4
LT
658 if (level > 0) {
659 /*
660 * It's a non-leaf entry. Make a hole for the new data
661 * in the key and ptr regions of the block.
662 */
663 kp = XFS_INOBT_KEY_ADDR(block, 1, cur);
664 pp = XFS_INOBT_PTR_ADDR(block, 1, cur);
665#ifdef DEBUG
666 for (i = numrecs; i >= ptr; i--) {
16259e7d 667 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(pp[i - 1]), level)))
1da177e4
LT
668 return error;
669 }
670#endif
671 memmove(&kp[ptr], &kp[ptr - 1],
672 (numrecs - ptr + 1) * sizeof(*kp));
673 memmove(&pp[ptr], &pp[ptr - 1],
674 (numrecs - ptr + 1) * sizeof(*pp));
675 /*
676 * Now stuff the new data in, bump numrecs and log the new data.
677 */
678#ifdef DEBUG
679 if ((error = xfs_btree_check_sptr(cur, *bnop, level)))
680 return error;
681#endif
c38e5e84 682 kp[ptr - 1] = key;
16259e7d 683 pp[ptr - 1] = cpu_to_be32(*bnop);
1da177e4 684 numrecs++;
16259e7d 685 block->bb_numrecs = cpu_to_be16(numrecs);
1da177e4
LT
686 xfs_inobt_log_keys(cur, bp, ptr, numrecs);
687 xfs_inobt_log_ptrs(cur, bp, ptr, numrecs);
688 } else {
689 /*
690 * It's a leaf entry. Make a hole for the new record.
691 */
692 rp = XFS_INOBT_REC_ADDR(block, 1, cur);
693 memmove(&rp[ptr], &rp[ptr - 1],
694 (numrecs - ptr + 1) * sizeof(*rp));
695 /*
696 * Now stuff the new record in, bump numrecs
697 * and log the new data.
698 */
c38e5e84 699 rp[ptr - 1] = *recp;
1da177e4 700 numrecs++;
16259e7d 701 block->bb_numrecs = cpu_to_be16(numrecs);
1da177e4
LT
702 xfs_inobt_log_recs(cur, bp, ptr, numrecs);
703 }
704 /*
705 * Log the new number of records in the btree header.
706 */
707 xfs_inobt_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS);
708#ifdef DEBUG
709 /*
710 * Check that the key/record is in the right place, now.
711 */
712 if (ptr < numrecs) {
713 if (level == 0)
714 xfs_btree_check_rec(cur->bc_btnum, rp + ptr - 1,
715 rp + ptr);
716 else
717 xfs_btree_check_key(cur->bc_btnum, kp + ptr - 1,
718 kp + ptr);
719 }
720#endif
721 /*
722 * If we inserted at the start of a block, update the parents' keys.
723 */
38bb7423 724 if (optr == 1 && (error = xfs_btree_updkey(cur, (union xfs_btree_key *)&key, level + 1)))
1da177e4
LT
725 return error;
726 /*
727 * Return the new block number, if any.
728 * If there is one, give back a record value and a cursor too.
729 */
730 *bnop = nbno;
731 if (nbno != NULLAGBLOCK) {
c38e5e84 732 *recp = nrec;
1da177e4
LT
733 *curp = ncur;
734 }
735 *stat = 1;
736 return 0;
737}
738
739/*
740 * Log header fields from a btree block.
741 */
742STATIC void
743xfs_inobt_log_block(
744 xfs_trans_t *tp, /* transaction pointer */
745 xfs_buf_t *bp, /* buffer containing btree block */
746 int fields) /* mask of fields: XFS_BB_... */
747{
748 int first; /* first byte offset logged */
749 int last; /* last byte offset logged */
750 static const short offsets[] = { /* table of offsets */
751 offsetof(xfs_inobt_block_t, bb_magic),
752 offsetof(xfs_inobt_block_t, bb_level),
753 offsetof(xfs_inobt_block_t, bb_numrecs),
754 offsetof(xfs_inobt_block_t, bb_leftsib),
755 offsetof(xfs_inobt_block_t, bb_rightsib),
756 sizeof(xfs_inobt_block_t)
757 };
758
759 xfs_btree_offsets(fields, offsets, XFS_BB_NUM_BITS, &first, &last);
760 xfs_trans_log_buf(tp, bp, first, last);
761}
762
763/*
764 * Log keys from a btree block (nonleaf).
765 */
766STATIC void
767xfs_inobt_log_keys(
768 xfs_btree_cur_t *cur, /* btree cursor */
769 xfs_buf_t *bp, /* buffer containing btree block */
770 int kfirst, /* index of first key to log */
771 int klast) /* index of last key to log */
772{
773 xfs_inobt_block_t *block; /* btree block to log from */
774 int first; /* first byte offset logged */
775 xfs_inobt_key_t *kp; /* key pointer in btree block */
776 int last; /* last byte offset logged */
777
778 block = XFS_BUF_TO_INOBT_BLOCK(bp);
779 kp = XFS_INOBT_KEY_ADDR(block, 1, cur);
780 first = (int)((xfs_caddr_t)&kp[kfirst - 1] - (xfs_caddr_t)block);
781 last = (int)(((xfs_caddr_t)&kp[klast] - 1) - (xfs_caddr_t)block);
782 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
783}
784
785/*
786 * Log block pointer fields from a btree block (nonleaf).
787 */
788STATIC void
789xfs_inobt_log_ptrs(
790 xfs_btree_cur_t *cur, /* btree cursor */
791 xfs_buf_t *bp, /* buffer containing btree block */
792 int pfirst, /* index of first pointer to log */
793 int plast) /* index of last pointer to log */
794{
795 xfs_inobt_block_t *block; /* btree block to log from */
796 int first; /* first byte offset logged */
797 int last; /* last byte offset logged */
798 xfs_inobt_ptr_t *pp; /* block-pointer pointer in btree blk */
799
800 block = XFS_BUF_TO_INOBT_BLOCK(bp);
801 pp = XFS_INOBT_PTR_ADDR(block, 1, cur);
802 first = (int)((xfs_caddr_t)&pp[pfirst - 1] - (xfs_caddr_t)block);
803 last = (int)(((xfs_caddr_t)&pp[plast] - 1) - (xfs_caddr_t)block);
804 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
805}
806
807/*
808 * Log records from a btree block (leaf).
809 */
810STATIC void
811xfs_inobt_log_recs(
812 xfs_btree_cur_t *cur, /* btree cursor */
813 xfs_buf_t *bp, /* buffer containing btree block */
814 int rfirst, /* index of first record to log */
815 int rlast) /* index of last record to log */
816{
817 xfs_inobt_block_t *block; /* btree block to log from */
818 int first; /* first byte offset logged */
819 int last; /* last byte offset logged */
820 xfs_inobt_rec_t *rp; /* record pointer for btree block */
821
822 block = XFS_BUF_TO_INOBT_BLOCK(bp);
823 rp = XFS_INOBT_REC_ADDR(block, 1, cur);
824 first = (int)((xfs_caddr_t)&rp[rfirst - 1] - (xfs_caddr_t)block);
825 last = (int)(((xfs_caddr_t)&rp[rlast] - 1) - (xfs_caddr_t)block);
826 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
827}
828
1da177e4
LT
829/*
830 * Move 1 record left from cur/level if possible.
831 * Update cur to reflect the new path.
832 */
833STATIC int /* error */
834xfs_inobt_lshift(
835 xfs_btree_cur_t *cur, /* btree cursor */
836 int level, /* level to shift record on */
837 int *stat) /* success/failure */
838{
839 int error; /* error return value */
840#ifdef DEBUG
841 int i; /* loop index */
842#endif
843 xfs_inobt_key_t key; /* key value for leaf level upward */
844 xfs_buf_t *lbp; /* buffer for left neighbor block */
845 xfs_inobt_block_t *left; /* left neighbor btree block */
846 xfs_inobt_key_t *lkp=NULL; /* key pointer for left block */
847 xfs_inobt_ptr_t *lpp; /* address pointer for left block */
848 xfs_inobt_rec_t *lrp=NULL; /* record pointer for left block */
849 int nrec; /* new number of left block entries */
850 xfs_buf_t *rbp; /* buffer for right (current) block */
851 xfs_inobt_block_t *right; /* right (current) btree block */
852 xfs_inobt_key_t *rkp=NULL; /* key pointer for right block */
853 xfs_inobt_ptr_t *rpp=NULL; /* address pointer for right block */
854 xfs_inobt_rec_t *rrp=NULL; /* record pointer for right block */
855
856 /*
857 * Set up variables for this block as "right".
858 */
859 rbp = cur->bc_bufs[level];
860 right = XFS_BUF_TO_INOBT_BLOCK(rbp);
861#ifdef DEBUG
862 if ((error = xfs_btree_check_sblock(cur, right, level, rbp)))
863 return error;
864#endif
865 /*
866 * If we've got no left sibling then we can't shift an entry left.
867 */
16259e7d 868 if (be32_to_cpu(right->bb_leftsib) == NULLAGBLOCK) {
1da177e4
LT
869 *stat = 0;
870 return 0;
871 }
872 /*
873 * If the cursor entry is the one that would be moved, don't
874 * do it... it's too complicated.
875 */
876 if (cur->bc_ptrs[level] <= 1) {
877 *stat = 0;
878 return 0;
879 }
880 /*
881 * Set up the left neighbor as "left".
882 */
883 if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
169d6227 884 cur->bc_private.a.agno, be32_to_cpu(right->bb_leftsib),
16259e7d 885 0, &lbp, XFS_INO_BTREE_REF)))
1da177e4
LT
886 return error;
887 left = XFS_BUF_TO_INOBT_BLOCK(lbp);
888 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
889 return error;
890 /*
891 * If it's full, it can't take another entry.
892 */
16259e7d 893 if (be16_to_cpu(left->bb_numrecs) == XFS_INOBT_BLOCK_MAXRECS(level, cur)) {
1da177e4
LT
894 *stat = 0;
895 return 0;
896 }
16259e7d 897 nrec = be16_to_cpu(left->bb_numrecs) + 1;
1da177e4
LT
898 /*
899 * If non-leaf, copy a key and a ptr to the left block.
900 */
901 if (level > 0) {
902 lkp = XFS_INOBT_KEY_ADDR(left, nrec, cur);
903 rkp = XFS_INOBT_KEY_ADDR(right, 1, cur);
904 *lkp = *rkp;
905 xfs_inobt_log_keys(cur, lbp, nrec, nrec);
906 lpp = XFS_INOBT_PTR_ADDR(left, nrec, cur);
907 rpp = XFS_INOBT_PTR_ADDR(right, 1, cur);
908#ifdef DEBUG
16259e7d 909 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(*rpp), level)))
1da177e4
LT
910 return error;
911#endif
c38e5e84 912 *lpp = *rpp;
1da177e4
LT
913 xfs_inobt_log_ptrs(cur, lbp, nrec, nrec);
914 }
915 /*
916 * If leaf, copy a record to the left block.
917 */
918 else {
919 lrp = XFS_INOBT_REC_ADDR(left, nrec, cur);
920 rrp = XFS_INOBT_REC_ADDR(right, 1, cur);
921 *lrp = *rrp;
922 xfs_inobt_log_recs(cur, lbp, nrec, nrec);
923 }
924 /*
925 * Bump and log left's numrecs, decrement and log right's numrecs.
926 */
413d57c9 927 be16_add_cpu(&left->bb_numrecs, 1);
1da177e4
LT
928 xfs_inobt_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS);
929#ifdef DEBUG
930 if (level > 0)
931 xfs_btree_check_key(cur->bc_btnum, lkp - 1, lkp);
932 else
933 xfs_btree_check_rec(cur->bc_btnum, lrp - 1, lrp);
934#endif
413d57c9 935 be16_add_cpu(&right->bb_numrecs, -1);
1da177e4
LT
936 xfs_inobt_log_block(cur->bc_tp, rbp, XFS_BB_NUMRECS);
937 /*
938 * Slide the contents of right down one entry.
939 */
940 if (level > 0) {
941#ifdef DEBUG
16259e7d
CH
942 for (i = 0; i < be16_to_cpu(right->bb_numrecs); i++) {
943 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(rpp[i + 1]),
1da177e4
LT
944 level)))
945 return error;
946 }
947#endif
16259e7d
CH
948 memmove(rkp, rkp + 1, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
949 memmove(rpp, rpp + 1, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
950 xfs_inobt_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
951 xfs_inobt_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
1da177e4 952 } else {
16259e7d
CH
953 memmove(rrp, rrp + 1, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
954 xfs_inobt_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
61a25848 955 key.ir_startino = rrp->ir_startino;
1da177e4
LT
956 rkp = &key;
957 }
958 /*
959 * Update the parent key values of right.
960 */
38bb7423 961 if ((error = xfs_btree_updkey(cur, (union xfs_btree_key *)rkp, level + 1)))
1da177e4
LT
962 return error;
963 /*
964 * Slide the cursor value left one.
965 */
966 cur->bc_ptrs[level]--;
967 *stat = 1;
968 return 0;
969}
970
971/*
972 * Allocate a new root block, fill it in.
973 */
974STATIC int /* error */
975xfs_inobt_newroot(
976 xfs_btree_cur_t *cur, /* btree cursor */
977 int *stat) /* success/failure */
978{
979 xfs_agi_t *agi; /* a.g. inode header */
980 xfs_alloc_arg_t args; /* allocation argument structure */
981 xfs_inobt_block_t *block; /* one half of the old root block */
982 xfs_buf_t *bp; /* buffer containing block */
983 int error; /* error return value */
984 xfs_inobt_key_t *kp; /* btree key pointer */
985 xfs_agblock_t lbno; /* left block number */
986 xfs_buf_t *lbp; /* left buffer pointer */
987 xfs_inobt_block_t *left; /* left btree block */
988 xfs_buf_t *nbp; /* new (root) buffer */
989 xfs_inobt_block_t *new; /* new (root) btree block */
990 int nptr; /* new value for key index, 1 or 2 */
991 xfs_inobt_ptr_t *pp; /* btree address pointer */
992 xfs_agblock_t rbno; /* right block number */
993 xfs_buf_t *rbp; /* right buffer pointer */
994 xfs_inobt_block_t *right; /* right btree block */
995 xfs_inobt_rec_t *rp; /* btree record pointer */
996
997 ASSERT(cur->bc_nlevels < XFS_IN_MAXLEVELS(cur->bc_mp));
998
999 /*
1000 * Get a block & a buffer.
1001 */
169d6227 1002 agi = XFS_BUF_TO_AGI(cur->bc_private.a.agbp);
1da177e4
LT
1003 args.tp = cur->bc_tp;
1004 args.mp = cur->bc_mp;
169d6227 1005 args.fsbno = XFS_AGB_TO_FSB(args.mp, cur->bc_private.a.agno,
16259e7d 1006 be32_to_cpu(agi->agi_root));
1da177e4
LT
1007 args.mod = args.minleft = args.alignment = args.total = args.wasdel =
1008 args.isfl = args.userdata = args.minalignslop = 0;
1009 args.minlen = args.maxlen = args.prod = 1;
1010 args.type = XFS_ALLOCTYPE_NEAR_BNO;
1011 if ((error = xfs_alloc_vextent(&args)))
1012 return error;
1013 /*
1014 * None available, we fail.
1015 */
1016 if (args.fsbno == NULLFSBLOCK) {
1017 *stat = 0;
1018 return 0;
1019 }
1020 ASSERT(args.len == 1);
1021 nbp = xfs_btree_get_bufs(args.mp, args.tp, args.agno, args.agbno, 0);
1022 new = XFS_BUF_TO_INOBT_BLOCK(nbp);
1023 /*
1024 * Set the root data in the a.g. inode structure.
1025 */
16259e7d 1026 agi->agi_root = cpu_to_be32(args.agbno);
413d57c9 1027 be32_add_cpu(&agi->agi_level, 1);
169d6227 1028 xfs_ialloc_log_agi(args.tp, cur->bc_private.a.agbp,
1da177e4
LT
1029 XFS_AGI_ROOT | XFS_AGI_LEVEL);
1030 /*
1031 * At the previous root level there are now two blocks: the old
1032 * root, and the new block generated when it was split.
1033 * We don't know which one the cursor is pointing at, so we
1034 * set up variables "left" and "right" for each case.
1035 */
1036 bp = cur->bc_bufs[cur->bc_nlevels - 1];
1037 block = XFS_BUF_TO_INOBT_BLOCK(bp);
1038#ifdef DEBUG
1039 if ((error = xfs_btree_check_sblock(cur, block, cur->bc_nlevels - 1, bp)))
1040 return error;
1041#endif
16259e7d 1042 if (be32_to_cpu(block->bb_rightsib) != NULLAGBLOCK) {
1da177e4
LT
1043 /*
1044 * Our block is left, pick up the right block.
1045 */
1046 lbp = bp;
1047 lbno = XFS_DADDR_TO_AGBNO(args.mp, XFS_BUF_ADDR(lbp));
1048 left = block;
16259e7d 1049 rbno = be32_to_cpu(left->bb_rightsib);
1da177e4
LT
1050 if ((error = xfs_btree_read_bufs(args.mp, args.tp, args.agno,
1051 rbno, 0, &rbp, XFS_INO_BTREE_REF)))
1052 return error;
1053 bp = rbp;
1054 right = XFS_BUF_TO_INOBT_BLOCK(rbp);
1055 if ((error = xfs_btree_check_sblock(cur, right,
1056 cur->bc_nlevels - 1, rbp)))
1057 return error;
1058 nptr = 1;
1059 } else {
1060 /*
1061 * Our block is right, pick up the left block.
1062 */
1063 rbp = bp;
1064 rbno = XFS_DADDR_TO_AGBNO(args.mp, XFS_BUF_ADDR(rbp));
1065 right = block;
16259e7d 1066 lbno = be32_to_cpu(right->bb_leftsib);
1da177e4
LT
1067 if ((error = xfs_btree_read_bufs(args.mp, args.tp, args.agno,
1068 lbno, 0, &lbp, XFS_INO_BTREE_REF)))
1069 return error;
1070 bp = lbp;
1071 left = XFS_BUF_TO_INOBT_BLOCK(lbp);
1072 if ((error = xfs_btree_check_sblock(cur, left,
1073 cur->bc_nlevels - 1, lbp)))
1074 return error;
1075 nptr = 2;
1076 }
1077 /*
1078 * Fill in the new block's btree header and log it.
1079 */
16259e7d
CH
1080 new->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
1081 new->bb_level = cpu_to_be16(cur->bc_nlevels);
1082 new->bb_numrecs = cpu_to_be16(2);
1083 new->bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1084 new->bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1da177e4
LT
1085 xfs_inobt_log_block(args.tp, nbp, XFS_BB_ALL_BITS);
1086 ASSERT(lbno != NULLAGBLOCK && rbno != NULLAGBLOCK);
1087 /*
1088 * Fill in the key data in the new root.
1089 */
1090 kp = XFS_INOBT_KEY_ADDR(new, 1, cur);
16259e7d 1091 if (be16_to_cpu(left->bb_level) > 0) {
c38e5e84
CH
1092 kp[0] = *XFS_INOBT_KEY_ADDR(left, 1, cur);
1093 kp[1] = *XFS_INOBT_KEY_ADDR(right, 1, cur);
1da177e4
LT
1094 } else {
1095 rp = XFS_INOBT_REC_ADDR(left, 1, cur);
61a25848 1096 kp[0].ir_startino = rp->ir_startino;
1da177e4 1097 rp = XFS_INOBT_REC_ADDR(right, 1, cur);
61a25848 1098 kp[1].ir_startino = rp->ir_startino;
1da177e4
LT
1099 }
1100 xfs_inobt_log_keys(cur, nbp, 1, 2);
1101 /*
1102 * Fill in the pointer data in the new root.
1103 */
1104 pp = XFS_INOBT_PTR_ADDR(new, 1, cur);
16259e7d
CH
1105 pp[0] = cpu_to_be32(lbno);
1106 pp[1] = cpu_to_be32(rbno);
1da177e4
LT
1107 xfs_inobt_log_ptrs(cur, nbp, 1, 2);
1108 /*
1109 * Fix up the cursor.
1110 */
1111 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
1112 cur->bc_ptrs[cur->bc_nlevels] = nptr;
1113 cur->bc_nlevels++;
1114 *stat = 1;
1115 return 0;
1116}
1117
1da177e4
LT
1118/*
1119 * Split cur/level block in half.
1120 * Return new block number and its first record (to be inserted into parent).
1121 */
1122STATIC int /* error */
1123xfs_inobt_split(
1124 xfs_btree_cur_t *cur, /* btree cursor */
1125 int level, /* level to split */
1126 xfs_agblock_t *bnop, /* output: block number allocated */
1127 xfs_inobt_key_t *keyp, /* output: first key of new block */
1128 xfs_btree_cur_t **curp, /* output: new cursor */
1129 int *stat) /* success/failure */
1130{
1131 xfs_alloc_arg_t args; /* allocation argument structure */
1132 int error; /* error return value */
1133 int i; /* loop index/record number */
1134 xfs_agblock_t lbno; /* left (current) block number */
1135 xfs_buf_t *lbp; /* buffer for left block */
1136 xfs_inobt_block_t *left; /* left (current) btree block */
1137 xfs_inobt_key_t *lkp; /* left btree key pointer */
1138 xfs_inobt_ptr_t *lpp; /* left btree address pointer */
1139 xfs_inobt_rec_t *lrp; /* left btree record pointer */
1140 xfs_buf_t *rbp; /* buffer for right block */
1141 xfs_inobt_block_t *right; /* right (new) btree block */
1142 xfs_inobt_key_t *rkp; /* right btree key pointer */
1143 xfs_inobt_ptr_t *rpp; /* right btree address pointer */
1144 xfs_inobt_rec_t *rrp; /* right btree record pointer */
1145
1146 /*
1147 * Set up left block (current one).
1148 */
1149 lbp = cur->bc_bufs[level];
1150 args.tp = cur->bc_tp;
1151 args.mp = cur->bc_mp;
1152 lbno = XFS_DADDR_TO_AGBNO(args.mp, XFS_BUF_ADDR(lbp));
1153 /*
1154 * Allocate the new block.
1155 * If we can't do it, we're toast. Give up.
1156 */
169d6227 1157 args.fsbno = XFS_AGB_TO_FSB(args.mp, cur->bc_private.a.agno, lbno);
1da177e4
LT
1158 args.mod = args.minleft = args.alignment = args.total = args.wasdel =
1159 args.isfl = args.userdata = args.minalignslop = 0;
1160 args.minlen = args.maxlen = args.prod = 1;
1161 args.type = XFS_ALLOCTYPE_NEAR_BNO;
1162 if ((error = xfs_alloc_vextent(&args)))
1163 return error;
1164 if (args.fsbno == NULLFSBLOCK) {
1165 *stat = 0;
1166 return 0;
1167 }
1168 ASSERT(args.len == 1);
1169 rbp = xfs_btree_get_bufs(args.mp, args.tp, args.agno, args.agbno, 0);
1170 /*
1171 * Set up the new block as "right".
1172 */
1173 right = XFS_BUF_TO_INOBT_BLOCK(rbp);
1174 /*
1175 * "Left" is the current (according to the cursor) block.
1176 */
1177 left = XFS_BUF_TO_INOBT_BLOCK(lbp);
1178#ifdef DEBUG
1179 if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
1180 return error;
1181#endif
1182 /*
1183 * Fill in the btree header for the new block.
1184 */
16259e7d
CH
1185 right->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
1186 right->bb_level = left->bb_level;
1187 right->bb_numrecs = cpu_to_be16(be16_to_cpu(left->bb_numrecs) / 2);
1da177e4
LT
1188 /*
1189 * Make sure that if there's an odd number of entries now, that
1190 * each new block will have the same number of entries.
1191 */
16259e7d
CH
1192 if ((be16_to_cpu(left->bb_numrecs) & 1) &&
1193 cur->bc_ptrs[level] <= be16_to_cpu(right->bb_numrecs) + 1)
413d57c9 1194 be16_add_cpu(&right->bb_numrecs, 1);
16259e7d 1195 i = be16_to_cpu(left->bb_numrecs) - be16_to_cpu(right->bb_numrecs) + 1;
1da177e4
LT
1196 /*
1197 * For non-leaf blocks, copy keys and addresses over to the new block.
1198 */
1199 if (level > 0) {
1200 lkp = XFS_INOBT_KEY_ADDR(left, i, cur);
1201 lpp = XFS_INOBT_PTR_ADDR(left, i, cur);
1202 rkp = XFS_INOBT_KEY_ADDR(right, 1, cur);
1203 rpp = XFS_INOBT_PTR_ADDR(right, 1, cur);
1204#ifdef DEBUG
16259e7d
CH
1205 for (i = 0; i < be16_to_cpu(right->bb_numrecs); i++) {
1206 if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(lpp[i]), level)))
1da177e4
LT
1207 return error;
1208 }
1209#endif
16259e7d
CH
1210 memcpy(rkp, lkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
1211 memcpy(rpp, lpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
1212 xfs_inobt_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
1213 xfs_inobt_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
1da177e4
LT
1214 *keyp = *rkp;
1215 }
1216 /*
1217 * For leaf blocks, copy records over to the new block.
1218 */
1219 else {
1220 lrp = XFS_INOBT_REC_ADDR(left, i, cur);
1221 rrp = XFS_INOBT_REC_ADDR(right, 1, cur);
16259e7d
CH
1222 memcpy(rrp, lrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
1223 xfs_inobt_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
61a25848 1224 keyp->ir_startino = rrp->ir_startino;
1da177e4
LT
1225 }
1226 /*
1227 * Find the left block number by looking in the buffer.
1228 * Adjust numrecs, sibling pointers.
1229 */
413d57c9 1230 be16_add_cpu(&left->bb_numrecs, -(be16_to_cpu(right->bb_numrecs)));
16259e7d
CH
1231 right->bb_rightsib = left->bb_rightsib;
1232 left->bb_rightsib = cpu_to_be32(args.agbno);
1233 right->bb_leftsib = cpu_to_be32(lbno);
1da177e4
LT
1234 xfs_inobt_log_block(args.tp, rbp, XFS_BB_ALL_BITS);
1235 xfs_inobt_log_block(args.tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
1236 /*
1237 * If there's a block to the new block's right, make that block
1238 * point back to right instead of to left.
1239 */
16259e7d 1240 if (be32_to_cpu(right->bb_rightsib) != NULLAGBLOCK) {
1da177e4
LT
1241 xfs_inobt_block_t *rrblock; /* rr btree block */
1242 xfs_buf_t *rrbp; /* buffer for rrblock */
1243
1244 if ((error = xfs_btree_read_bufs(args.mp, args.tp, args.agno,
16259e7d 1245 be32_to_cpu(right->bb_rightsib), 0, &rrbp,
1da177e4
LT
1246 XFS_INO_BTREE_REF)))
1247 return error;
1248 rrblock = XFS_BUF_TO_INOBT_BLOCK(rrbp);
1249 if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp)))
1250 return error;
16259e7d 1251 rrblock->bb_leftsib = cpu_to_be32(args.agbno);
1da177e4
LT
1252 xfs_inobt_log_block(args.tp, rrbp, XFS_BB_LEFTSIB);
1253 }
1254 /*
1255 * If the cursor is really in the right block, move it there.
1256 * If it's just pointing past the last entry in left, then we'll
1257 * insert there, so don't change anything in that case.
1258 */
16259e7d 1259 if (cur->bc_ptrs[level] > be16_to_cpu(left->bb_numrecs) + 1) {
1da177e4 1260 xfs_btree_setbuf(cur, level, rbp);
16259e7d 1261 cur->bc_ptrs[level] -= be16_to_cpu(left->bb_numrecs);
1da177e4
LT
1262 }
1263 /*
1264 * If there are more levels, we'll need another cursor which refers
1265 * the right block, no matter where this cursor was.
1266 */
1267 if (level + 1 < cur->bc_nlevels) {
1268 if ((error = xfs_btree_dup_cursor(cur, curp)))
1269 return error;
1270 (*curp)->bc_ptrs[level + 1]++;
1271 }
1272 *bnop = args.agbno;
1273 *stat = 1;
1274 return 0;
1275}
1276
1da177e4
LT
1277/*
1278 * Externally visible routines.
1279 */
1280
1da177e4
LT
1281/*
1282 * Delete the record pointed to by cur.
1283 * The cursor refers to the place where the record was (could be inserted)
1284 * when the operation returns.
1285 */
1286int /* error */
1287xfs_inobt_delete(
1288 xfs_btree_cur_t *cur, /* btree cursor */
1289 int *stat) /* success/failure */
1290{
1291 int error;
1292 int i; /* result code */
1293 int level; /* btree level */
1294
1295 /*
1296 * Go up the tree, starting at leaf level.
1297 * If 2 is returned then a join was done; go to the next level.
1298 * Otherwise we are done.
1299 */
1300 for (level = 0, i = 2; i == 2; level++) {
1301 if ((error = xfs_inobt_delrec(cur, level, &i)))
1302 return error;
1303 }
1304 if (i == 0) {
1305 for (level = 1; level < cur->bc_nlevels; level++) {
1306 if (cur->bc_ptrs[level] == 0) {
8df4da4a 1307 if ((error = xfs_btree_decrement(cur, level, &i)))
1da177e4
LT
1308 return error;
1309 break;
1310 }
1311 }
1312 }
1313 *stat = i;
1314 return 0;
1315}
1316
1317
1318/*
1319 * Get the data from the pointed-to record.
1320 */
1321int /* error */
1322xfs_inobt_get_rec(
1323 xfs_btree_cur_t *cur, /* btree cursor */
1324 xfs_agino_t *ino, /* output: starting inode of chunk */
1325 __int32_t *fcnt, /* output: number of free inodes */
1326 xfs_inofree_t *free, /* output: free inode mask */
1327 int *stat) /* output: success/failure */
1328{
1329 xfs_inobt_block_t *block; /* btree block */
1330 xfs_buf_t *bp; /* buffer containing btree block */
1331#ifdef DEBUG
1332 int error; /* error return value */
1333#endif
1334 int ptr; /* record number */
1335 xfs_inobt_rec_t *rec; /* record data */
1336
1337 bp = cur->bc_bufs[0];
1338 ptr = cur->bc_ptrs[0];
1339 block = XFS_BUF_TO_INOBT_BLOCK(bp);
1340#ifdef DEBUG
1341 if ((error = xfs_btree_check_sblock(cur, block, 0, bp)))
1342 return error;
1343#endif
1344 /*
1345 * Off the right end or left end, return failure.
1346 */
16259e7d 1347 if (ptr > be16_to_cpu(block->bb_numrecs) || ptr <= 0) {
1da177e4
LT
1348 *stat = 0;
1349 return 0;
1350 }
1351 /*
1352 * Point to the record and extract its data.
1353 */
1354 rec = XFS_INOBT_REC_ADDR(block, ptr, cur);
61a25848
CH
1355 *ino = be32_to_cpu(rec->ir_startino);
1356 *fcnt = be32_to_cpu(rec->ir_freecount);
1357 *free = be64_to_cpu(rec->ir_free);
1da177e4
LT
1358 *stat = 1;
1359 return 0;
1360}
1361
1da177e4
LT
1362/*
1363 * Insert the current record at the point referenced by cur.
1364 * The cursor may be inconsistent on return if splits have been done.
1365 */
1366int /* error */
1367xfs_inobt_insert(
1368 xfs_btree_cur_t *cur, /* btree cursor */
1369 int *stat) /* success/failure */
1370{
1371 int error; /* error return value */
1372 int i; /* result value, 0 for failure */
1373 int level; /* current level number in btree */
1374 xfs_agblock_t nbno; /* new block number (split result) */
1375 xfs_btree_cur_t *ncur; /* new cursor (split result) */
1376 xfs_inobt_rec_t nrec; /* record being inserted this level */
1377 xfs_btree_cur_t *pcur; /* previous level's cursor */
1378
1379 level = 0;
1380 nbno = NULLAGBLOCK;
61a25848
CH
1381 nrec.ir_startino = cpu_to_be32(cur->bc_rec.i.ir_startino);
1382 nrec.ir_freecount = cpu_to_be32(cur->bc_rec.i.ir_freecount);
1383 nrec.ir_free = cpu_to_be64(cur->bc_rec.i.ir_free);
1121b219 1384 ncur = NULL;
1da177e4
LT
1385 pcur = cur;
1386 /*
1387 * Loop going up the tree, starting at the leaf level.
1388 * Stop when we don't get a split block, that must mean that
1389 * the insert is finished with this level.
1390 */
1391 do {
1392 /*
1393 * Insert nrec/nbno into this level of the tree.
1394 * Note if we fail, nbno will be null.
1395 */
1396 if ((error = xfs_inobt_insrec(pcur, level++, &nbno, &nrec, &ncur,
1397 &i))) {
1398 if (pcur != cur)
1399 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
1400 return error;
1401 }
1402 /*
1403 * See if the cursor we just used is trash.
1404 * Can't trash the caller's cursor, but otherwise we should
1405 * if ncur is a new cursor or we're about to be done.
1406 */
1407 if (pcur != cur && (ncur || nbno == NULLAGBLOCK)) {
1408 cur->bc_nlevels = pcur->bc_nlevels;
1409 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
1410 }
1411 /*
1412 * If we got a new cursor, switch to it.
1413 */
1414 if (ncur) {
1415 pcur = ncur;
1121b219 1416 ncur = NULL;
1da177e4
LT
1417 }
1418 } while (nbno != NULLAGBLOCK);
1419 *stat = i;
1420 return 0;
1421}
1422
561f7d17
CH
1423STATIC struct xfs_btree_cur *
1424xfs_inobt_dup_cursor(
1425 struct xfs_btree_cur *cur)
1426{
1427 return xfs_inobt_init_cursor(cur->bc_mp, cur->bc_tp,
1428 cur->bc_private.a.agbp, cur->bc_private.a.agno);
1429}
1430
ce5e42db
CH
1431STATIC int
1432xfs_inobt_get_maxrecs(
1433 struct xfs_btree_cur *cur,
1434 int level)
1435{
1436 return cur->bc_mp->m_inobt_mxr[level != 0];
1437}
1438
fe033cc8
CH
1439STATIC void
1440xfs_inobt_init_key_from_rec(
1441 union xfs_btree_key *key,
1442 union xfs_btree_rec *rec)
1443{
1444 key->inobt.ir_startino = rec->inobt.ir_startino;
1445}
1446
1447/*
1448 * intial value of ptr for lookup
1449 */
1450STATIC void
1451xfs_inobt_init_ptr_from_cur(
1452 struct xfs_btree_cur *cur,
1453 union xfs_btree_ptr *ptr)
1454{
1455 struct xfs_agi *agi = XFS_BUF_TO_AGI(cur->bc_private.a.agbp);
1456
1457 ASSERT(cur->bc_private.a.agno == be32_to_cpu(agi->agi_seqno));
1458
1459 ptr->s = agi->agi_root;
1460}
1461
1462STATIC __int64_t
1463xfs_inobt_key_diff(
1464 struct xfs_btree_cur *cur,
1465 union xfs_btree_key *key)
1466{
1467 return (__int64_t)be32_to_cpu(key->inobt.ir_startino) -
1468 cur->bc_rec.i.ir_startino;
1469}
1470
8c4ed633
CH
1471#ifdef XFS_BTREE_TRACE
1472ktrace_t *xfs_inobt_trace_buf;
1473
1474STATIC void
1475xfs_inobt_trace_enter(
1476 struct xfs_btree_cur *cur,
1477 const char *func,
1478 char *s,
1479 int type,
1480 int line,
1481 __psunsigned_t a0,
1482 __psunsigned_t a1,
1483 __psunsigned_t a2,
1484 __psunsigned_t a3,
1485 __psunsigned_t a4,
1486 __psunsigned_t a5,
1487 __psunsigned_t a6,
1488 __psunsigned_t a7,
1489 __psunsigned_t a8,
1490 __psunsigned_t a9,
1491 __psunsigned_t a10)
1492{
1493 ktrace_enter(xfs_inobt_trace_buf, (void *)(__psint_t)type,
1494 (void *)func, (void *)s, NULL, (void *)cur,
1495 (void *)a0, (void *)a1, (void *)a2, (void *)a3,
1496 (void *)a4, (void *)a5, (void *)a6, (void *)a7,
1497 (void *)a8, (void *)a9, (void *)a10);
1498}
1499
1500STATIC void
1501xfs_inobt_trace_cursor(
1502 struct xfs_btree_cur *cur,
1503 __uint32_t *s0,
1504 __uint64_t *l0,
1505 __uint64_t *l1)
1506{
1507 *s0 = cur->bc_private.a.agno;
1508 *l0 = cur->bc_rec.i.ir_startino;
1509 *l1 = cur->bc_rec.i.ir_free;
1510}
1511
1512STATIC void
1513xfs_inobt_trace_key(
1514 struct xfs_btree_cur *cur,
1515 union xfs_btree_key *key,
1516 __uint64_t *l0,
1517 __uint64_t *l1)
1518{
1519 *l0 = be32_to_cpu(key->inobt.ir_startino);
1520 *l1 = 0;
1521}
1522
1523STATIC void
1524xfs_inobt_trace_record(
1525 struct xfs_btree_cur *cur,
1526 union xfs_btree_rec *rec,
1527 __uint64_t *l0,
1528 __uint64_t *l1,
1529 __uint64_t *l2)
1530{
1531 *l0 = be32_to_cpu(rec->inobt.ir_startino);
1532 *l1 = be32_to_cpu(rec->inobt.ir_freecount);
1533 *l2 = be64_to_cpu(rec->inobt.ir_free);
1534}
1535#endif /* XFS_BTREE_TRACE */
1536
561f7d17 1537static const struct xfs_btree_ops xfs_inobt_ops = {
65f1eaea
CH
1538 .rec_len = sizeof(xfs_inobt_rec_t),
1539 .key_len = sizeof(xfs_inobt_key_t),
1540
561f7d17 1541 .dup_cursor = xfs_inobt_dup_cursor,
ce5e42db 1542 .get_maxrecs = xfs_inobt_get_maxrecs,
fe033cc8
CH
1543 .init_key_from_rec = xfs_inobt_init_key_from_rec,
1544 .init_ptr_from_cur = xfs_inobt_init_ptr_from_cur,
1545 .key_diff = xfs_inobt_key_diff,
8c4ed633
CH
1546
1547#ifdef XFS_BTREE_TRACE
1548 .trace_enter = xfs_inobt_trace_enter,
1549 .trace_cursor = xfs_inobt_trace_cursor,
1550 .trace_key = xfs_inobt_trace_key,
1551 .trace_record = xfs_inobt_trace_record,
1552#endif
561f7d17
CH
1553};
1554
1555/*
1556 * Allocate a new inode btree cursor.
1557 */
1558struct xfs_btree_cur * /* new inode btree cursor */
1559xfs_inobt_init_cursor(
1560 struct xfs_mount *mp, /* file system mount point */
1561 struct xfs_trans *tp, /* transaction pointer */
1562 struct xfs_buf *agbp, /* buffer for agi structure */
1563 xfs_agnumber_t agno) /* allocation group number */
1564{
1565 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1566 struct xfs_btree_cur *cur;
1567
1568 cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_SLEEP);
1569
1570 cur->bc_tp = tp;
1571 cur->bc_mp = mp;
1572 cur->bc_nlevels = be32_to_cpu(agi->agi_level);
1573 cur->bc_btnum = XFS_BTNUM_INO;
1574 cur->bc_blocklog = mp->m_sb.sb_blocklog;
1575
1576 cur->bc_ops = &xfs_inobt_ops;
1577
1578 cur->bc_private.a.agbp = agbp;
1579 cur->bc_private.a.agno = agno;
1580
1581 return cur;
1582}
This page took 0.602948 seconds and 5 git commands to generate.