Commit | Line | Data |
---|---|---|
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" | |
27 | #include "xfs_dir.h" | |
a844f451 | 28 | #include "xfs_dir2.h" |
1da177e4 LT |
29 | #include "xfs_dmapi.h" |
30 | #include "xfs_mount.h" | |
a844f451 | 31 | #include "xfs_bmap_btree.h" |
1da177e4 LT |
32 | #include "xfs_alloc_btree.h" |
33 | #include "xfs_ialloc_btree.h" | |
a844f451 NS |
34 | #include "xfs_dir_sf.h" |
35 | #include "xfs_dir2_sf.h" | |
36 | #include "xfs_attr_sf.h" | |
37 | #include "xfs_dinode.h" | |
38 | #include "xfs_inode.h" | |
1da177e4 LT |
39 | #include "xfs_btree.h" |
40 | #include "xfs_ialloc.h" | |
41 | #include "xfs_alloc.h" | |
42 | #include "xfs_error.h" | |
43 | ||
44 | /* | |
45 | * Prototypes for internal functions. | |
46 | */ | |
47 | ||
48 | STATIC void xfs_alloc_log_block(xfs_trans_t *, xfs_buf_t *, int); | |
49 | STATIC void xfs_alloc_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int); | |
50 | STATIC void xfs_alloc_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int); | |
51 | STATIC void xfs_alloc_log_recs(xfs_btree_cur_t *, xfs_buf_t *, int, int); | |
52 | STATIC int xfs_alloc_lshift(xfs_btree_cur_t *, int, int *); | |
53 | STATIC int xfs_alloc_newroot(xfs_btree_cur_t *, int *); | |
54 | STATIC int xfs_alloc_rshift(xfs_btree_cur_t *, int, int *); | |
55 | STATIC int xfs_alloc_split(xfs_btree_cur_t *, int, xfs_agblock_t *, | |
56 | xfs_alloc_key_t *, xfs_btree_cur_t **, int *); | |
57 | STATIC int xfs_alloc_updkey(xfs_btree_cur_t *, xfs_alloc_key_t *, int); | |
58 | ||
59 | /* | |
60 | * Internal functions. | |
61 | */ | |
62 | ||
63 | /* | |
64 | * Single level of the xfs_alloc_delete record deletion routine. | |
65 | * Delete record pointed to by cur/level. | |
66 | * Remove the record from its block then rebalance the tree. | |
67 | * Return 0 for error, 1 for done, 2 to go on to the next level. | |
68 | */ | |
69 | STATIC int /* error */ | |
70 | xfs_alloc_delrec( | |
71 | xfs_btree_cur_t *cur, /* btree cursor */ | |
72 | int level, /* level removing record from */ | |
73 | int *stat) /* fail/done/go-on */ | |
74 | { | |
75 | xfs_agf_t *agf; /* allocation group freelist header */ | |
76 | xfs_alloc_block_t *block; /* btree block record/key lives in */ | |
77 | xfs_agblock_t bno; /* btree block number */ | |
78 | xfs_buf_t *bp; /* buffer for block */ | |
79 | int error; /* error return value */ | |
80 | int i; /* loop index */ | |
81 | xfs_alloc_key_t key; /* kp points here if block is level 0 */ | |
82 | xfs_agblock_t lbno; /* left block's block number */ | |
83 | xfs_buf_t *lbp; /* left block's buffer pointer */ | |
84 | xfs_alloc_block_t *left; /* left btree block */ | |
85 | xfs_alloc_key_t *lkp=NULL; /* left block key pointer */ | |
86 | xfs_alloc_ptr_t *lpp=NULL; /* left block address pointer */ | |
87 | int lrecs=0; /* number of records in left block */ | |
88 | xfs_alloc_rec_t *lrp; /* left block record pointer */ | |
89 | xfs_mount_t *mp; /* mount structure */ | |
90 | int ptr; /* index in btree block for this rec */ | |
91 | xfs_agblock_t rbno; /* right block's block number */ | |
92 | xfs_buf_t *rbp; /* right block's buffer pointer */ | |
93 | xfs_alloc_block_t *right; /* right btree block */ | |
94 | xfs_alloc_key_t *rkp; /* right block key pointer */ | |
95 | xfs_alloc_ptr_t *rpp; /* right block address pointer */ | |
96 | int rrecs=0; /* number of records in right block */ | |
97 | xfs_alloc_rec_t *rrp; /* right block record pointer */ | |
98 | xfs_btree_cur_t *tcur; /* temporary btree cursor */ | |
99 | ||
100 | /* | |
101 | * Get the index of the entry being deleted, check for nothing there. | |
102 | */ | |
103 | ptr = cur->bc_ptrs[level]; | |
104 | if (ptr == 0) { | |
105 | *stat = 0; | |
106 | return 0; | |
107 | } | |
108 | /* | |
109 | * Get the buffer & block containing the record or key/ptr. | |
110 | */ | |
111 | bp = cur->bc_bufs[level]; | |
112 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
113 | #ifdef DEBUG | |
114 | if ((error = xfs_btree_check_sblock(cur, block, level, bp))) | |
115 | return error; | |
116 | #endif | |
117 | /* | |
118 | * Fail if we're off the end of the block. | |
119 | */ | |
120 | if (ptr > INT_GET(block->bb_numrecs, ARCH_CONVERT)) { | |
121 | *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 | |
134 | for (i = ptr; i < INT_GET(block->bb_numrecs, ARCH_CONVERT); i++) { | |
135 | if ((error = xfs_btree_check_sptr(cur, INT_GET(lpp[i], ARCH_CONVERT), level))) | |
136 | return error; | |
137 | } | |
138 | #endif | |
139 | if (ptr < INT_GET(block->bb_numrecs, ARCH_CONVERT)) { | |
140 | memmove(&lkp[ptr - 1], &lkp[ptr], | |
141 | (INT_GET(block->bb_numrecs, ARCH_CONVERT) - ptr) * sizeof(*lkp)); /* INT_: mem copy */ | |
142 | memmove(&lpp[ptr - 1], &lpp[ptr], | |
143 | (INT_GET(block->bb_numrecs, ARCH_CONVERT) - ptr) * sizeof(*lpp)); /* INT_: mem copy */ | |
144 | xfs_alloc_log_ptrs(cur, bp, ptr, INT_GET(block->bb_numrecs, ARCH_CONVERT) - 1); | |
145 | xfs_alloc_log_keys(cur, bp, ptr, INT_GET(block->bb_numrecs, ARCH_CONVERT) - 1); | |
146 | } | |
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); | |
154 | if (ptr < INT_GET(block->bb_numrecs, ARCH_CONVERT)) { | |
155 | memmove(&lrp[ptr - 1], &lrp[ptr], | |
156 | (INT_GET(block->bb_numrecs, ARCH_CONVERT) - ptr) * sizeof(*lrp)); | |
157 | xfs_alloc_log_recs(cur, bp, ptr, INT_GET(block->bb_numrecs, ARCH_CONVERT) - 1); | |
158 | } | |
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) { | |
164 | key.ar_startblock = lrp->ar_startblock; /* INT_: direct copy */ | |
165 | key.ar_blockcount = lrp->ar_blockcount; /* INT_: direct copy */ | |
166 | lkp = &key; | |
167 | } | |
168 | } | |
169 | /* | |
170 | * Decrement and log the number of entries in the block. | |
171 | */ | |
172 | INT_MOD(block->bb_numrecs, ARCH_CONVERT, -1); | |
173 | xfs_alloc_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS); | |
174 | /* | |
175 | * See if the longest free extent in the allocation group was | |
176 | * changed by this operation. True if it's the by-size btree, and | |
177 | * this is the leaf level, and there is no right sibling block, | |
178 | * and this was the last record. | |
179 | */ | |
180 | agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | |
181 | mp = cur->bc_mp; | |
182 | ||
183 | if (level == 0 && | |
184 | cur->bc_btnum == XFS_BTNUM_CNT && | |
185 | INT_GET(block->bb_rightsib, ARCH_CONVERT) == NULLAGBLOCK && | |
186 | ptr > INT_GET(block->bb_numrecs, ARCH_CONVERT)) { | |
187 | ASSERT(ptr == INT_GET(block->bb_numrecs, ARCH_CONVERT) + 1); | |
188 | /* | |
189 | * There are still records in the block. Grab the size | |
190 | * from the last one. | |
191 | */ | |
192 | if (INT_GET(block->bb_numrecs, ARCH_CONVERT)) { | |
193 | rrp = XFS_ALLOC_REC_ADDR(block, INT_GET(block->bb_numrecs, ARCH_CONVERT), cur); | |
194 | INT_COPY(agf->agf_longest, rrp->ar_blockcount, ARCH_CONVERT); | |
195 | } | |
196 | /* | |
197 | * No free extents left. | |
198 | */ | |
199 | else | |
200 | agf->agf_longest = 0; | |
201 | mp->m_perag[INT_GET(agf->agf_seqno, ARCH_CONVERT)].pagf_longest = | |
202 | INT_GET(agf->agf_longest, ARCH_CONVERT); | |
203 | xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, | |
204 | XFS_AGF_LONGEST); | |
205 | } | |
206 | /* | |
207 | * Is this the root level? If so, we're almost done. | |
208 | */ | |
209 | if (level == cur->bc_nlevels - 1) { | |
210 | /* | |
211 | * If this is the root level, | |
212 | * and there's only one entry left, | |
213 | * and it's NOT the leaf level, | |
214 | * then we can get rid of this level. | |
215 | */ | |
216 | if (INT_GET(block->bb_numrecs, ARCH_CONVERT) == 1 && level > 0) { | |
217 | /* | |
218 | * lpp is still set to the first pointer in the block. | |
219 | * Make it the new root of the btree. | |
220 | */ | |
221 | bno = INT_GET(agf->agf_roots[cur->bc_btnum], ARCH_CONVERT); | |
222 | INT_COPY(agf->agf_roots[cur->bc_btnum], *lpp, ARCH_CONVERT); | |
223 | INT_MOD(agf->agf_levels[cur->bc_btnum], ARCH_CONVERT, -1); | |
224 | mp->m_perag[INT_GET(agf->agf_seqno, ARCH_CONVERT)].pagf_levels[cur->bc_btnum]--; | |
225 | /* | |
226 | * Put this buffer/block on the ag's freelist. | |
227 | */ | |
228 | if ((error = xfs_alloc_put_freelist(cur->bc_tp, | |
229 | cur->bc_private.a.agbp, NULL, bno))) | |
230 | return error; | |
231 | /* | |
232 | * Since blocks move to the free list without the | |
233 | * coordination used in xfs_bmap_finish, we can't allow | |
234 | * block to be available for reallocation and | |
235 | * non-transaction writing (user data) until we know | |
236 | * that the transaction that moved it to the free list | |
237 | * is permanently on disk. We track the blocks by | |
238 | * declaring these blocks as "busy"; the busy list is | |
239 | * maintained on a per-ag basis and each transaction | |
240 | * records which entries should be removed when the | |
241 | * iclog commits to disk. If a busy block is | |
242 | * allocated, the iclog is pushed up to the LSN | |
243 | * that freed the block. | |
244 | */ | |
245 | xfs_alloc_mark_busy(cur->bc_tp, | |
246 | INT_GET(agf->agf_seqno, ARCH_CONVERT), bno, 1); | |
247 | ||
248 | xfs_trans_agbtree_delta(cur->bc_tp, -1); | |
249 | xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, | |
250 | XFS_AGF_ROOTS | XFS_AGF_LEVELS); | |
251 | /* | |
252 | * Update the cursor so there's one fewer level. | |
253 | */ | |
254 | xfs_btree_setbuf(cur, level, NULL); | |
255 | cur->bc_nlevels--; | |
256 | } else if (level > 0 && | |
257 | (error = xfs_alloc_decrement(cur, level, &i))) | |
258 | return error; | |
259 | *stat = 1; | |
260 | return 0; | |
261 | } | |
262 | /* | |
263 | * If we deleted the leftmost entry in the block, update the | |
264 | * key values above us in the tree. | |
265 | */ | |
266 | if (ptr == 1 && (error = xfs_alloc_updkey(cur, lkp, level + 1))) | |
267 | return error; | |
268 | /* | |
269 | * If the number of records remaining in the block is at least | |
270 | * the minimum, we're done. | |
271 | */ | |
272 | if (INT_GET(block->bb_numrecs, ARCH_CONVERT) >= XFS_ALLOC_BLOCK_MINRECS(level, cur)) { | |
273 | if (level > 0 && (error = xfs_alloc_decrement(cur, level, &i))) | |
274 | return error; | |
275 | *stat = 1; | |
276 | return 0; | |
277 | } | |
278 | /* | |
279 | * Otherwise, we have to move some records around to keep the | |
280 | * tree balanced. Look at the left and right sibling blocks to | |
281 | * see if we can re-balance by moving only one record. | |
282 | */ | |
283 | rbno = INT_GET(block->bb_rightsib, ARCH_CONVERT); | |
284 | lbno = INT_GET(block->bb_leftsib, ARCH_CONVERT); | |
285 | bno = NULLAGBLOCK; | |
286 | ASSERT(rbno != NULLAGBLOCK || lbno != NULLAGBLOCK); | |
287 | /* | |
288 | * Duplicate the cursor so our btree manipulations here won't | |
289 | * disrupt the next level up. | |
290 | */ | |
291 | if ((error = xfs_btree_dup_cursor(cur, &tcur))) | |
292 | return error; | |
293 | /* | |
294 | * If there's a right sibling, see if it's ok to shift an entry | |
295 | * out of it. | |
296 | */ | |
297 | if (rbno != NULLAGBLOCK) { | |
298 | /* | |
299 | * Move the temp cursor to the last entry in the next block. | |
300 | * Actually any entry but the first would suffice. | |
301 | */ | |
302 | i = xfs_btree_lastrec(tcur, level); | |
303 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
304 | if ((error = xfs_alloc_increment(tcur, level, &i))) | |
305 | goto error0; | |
306 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
307 | i = xfs_btree_lastrec(tcur, level); | |
308 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
309 | /* | |
310 | * Grab a pointer to the block. | |
311 | */ | |
312 | rbp = tcur->bc_bufs[level]; | |
313 | right = XFS_BUF_TO_ALLOC_BLOCK(rbp); | |
314 | #ifdef DEBUG | |
315 | if ((error = xfs_btree_check_sblock(cur, right, level, rbp))) | |
316 | goto error0; | |
317 | #endif | |
318 | /* | |
319 | * Grab the current block number, for future use. | |
320 | */ | |
321 | bno = INT_GET(right->bb_leftsib, ARCH_CONVERT); | |
322 | /* | |
323 | * If right block is full enough so that removing one entry | |
324 | * won't make it too empty, and left-shifting an entry out | |
325 | * of right to us works, we're done. | |
326 | */ | |
327 | if (INT_GET(right->bb_numrecs, ARCH_CONVERT) - 1 >= | |
328 | XFS_ALLOC_BLOCK_MINRECS(level, cur)) { | |
329 | if ((error = xfs_alloc_lshift(tcur, level, &i))) | |
330 | goto error0; | |
331 | if (i) { | |
332 | ASSERT(INT_GET(block->bb_numrecs, ARCH_CONVERT) >= | |
333 | XFS_ALLOC_BLOCK_MINRECS(level, cur)); | |
334 | xfs_btree_del_cursor(tcur, | |
335 | XFS_BTREE_NOERROR); | |
336 | if (level > 0 && | |
337 | (error = xfs_alloc_decrement(cur, level, | |
338 | &i))) | |
339 | return error; | |
340 | *stat = 1; | |
341 | return 0; | |
342 | } | |
343 | } | |
344 | /* | |
345 | * Otherwise, grab the number of records in right for | |
346 | * future reference, and fix up the temp cursor to point | |
347 | * to our block again (last record). | |
348 | */ | |
349 | rrecs = INT_GET(right->bb_numrecs, ARCH_CONVERT); | |
350 | if (lbno != NULLAGBLOCK) { | |
351 | i = xfs_btree_firstrec(tcur, level); | |
352 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
353 | if ((error = xfs_alloc_decrement(tcur, level, &i))) | |
354 | goto error0; | |
355 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
356 | } | |
357 | } | |
358 | /* | |
359 | * If there's a left sibling, see if it's ok to shift an entry | |
360 | * out of it. | |
361 | */ | |
362 | if (lbno != NULLAGBLOCK) { | |
363 | /* | |
364 | * Move the temp cursor to the first entry in the | |
365 | * previous block. | |
366 | */ | |
367 | i = xfs_btree_firstrec(tcur, level); | |
368 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
369 | if ((error = xfs_alloc_decrement(tcur, level, &i))) | |
370 | goto error0; | |
371 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
372 | xfs_btree_firstrec(tcur, level); | |
373 | /* | |
374 | * Grab a pointer to the block. | |
375 | */ | |
376 | lbp = tcur->bc_bufs[level]; | |
377 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | |
378 | #ifdef DEBUG | |
379 | if ((error = xfs_btree_check_sblock(cur, left, level, lbp))) | |
380 | goto error0; | |
381 | #endif | |
382 | /* | |
383 | * Grab the current block number, for future use. | |
384 | */ | |
385 | bno = INT_GET(left->bb_rightsib, ARCH_CONVERT); | |
386 | /* | |
387 | * If left block is full enough so that removing one entry | |
388 | * won't make it too empty, and right-shifting an entry out | |
389 | * of left to us works, we're done. | |
390 | */ | |
391 | if (INT_GET(left->bb_numrecs, ARCH_CONVERT) - 1 >= | |
392 | XFS_ALLOC_BLOCK_MINRECS(level, cur)) { | |
393 | if ((error = xfs_alloc_rshift(tcur, level, &i))) | |
394 | goto error0; | |
395 | if (i) { | |
396 | ASSERT(INT_GET(block->bb_numrecs, ARCH_CONVERT) >= | |
397 | XFS_ALLOC_BLOCK_MINRECS(level, cur)); | |
398 | xfs_btree_del_cursor(tcur, | |
399 | XFS_BTREE_NOERROR); | |
400 | if (level == 0) | |
401 | cur->bc_ptrs[0]++; | |
402 | *stat = 1; | |
403 | return 0; | |
404 | } | |
405 | } | |
406 | /* | |
407 | * Otherwise, grab the number of records in right for | |
408 | * future reference. | |
409 | */ | |
410 | lrecs = INT_GET(left->bb_numrecs, ARCH_CONVERT); | |
411 | } | |
412 | /* | |
413 | * Delete the temp cursor, we're done with it. | |
414 | */ | |
415 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
416 | /* | |
417 | * If here, we need to do a join to keep the tree balanced. | |
418 | */ | |
419 | ASSERT(bno != NULLAGBLOCK); | |
420 | /* | |
421 | * See if we can join with the left neighbor block. | |
422 | */ | |
423 | if (lbno != NULLAGBLOCK && | |
424 | lrecs + INT_GET(block->bb_numrecs, ARCH_CONVERT) <= XFS_ALLOC_BLOCK_MAXRECS(level, cur)) { | |
425 | /* | |
426 | * Set "right" to be the starting block, | |
427 | * "left" to be the left neighbor. | |
428 | */ | |
429 | rbno = bno; | |
430 | right = block; | |
431 | rbp = bp; | |
432 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, | |
433 | cur->bc_private.a.agno, lbno, 0, &lbp, | |
434 | XFS_ALLOC_BTREE_REF))) | |
435 | return error; | |
436 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | |
437 | if ((error = xfs_btree_check_sblock(cur, left, level, lbp))) | |
438 | return error; | |
439 | } | |
440 | /* | |
441 | * If that won't work, see if we can join with the right neighbor block. | |
442 | */ | |
443 | else if (rbno != NULLAGBLOCK && | |
444 | rrecs + INT_GET(block->bb_numrecs, ARCH_CONVERT) <= | |
445 | XFS_ALLOC_BLOCK_MAXRECS(level, cur)) { | |
446 | /* | |
447 | * Set "left" to be the starting block, | |
448 | * "right" to be the right neighbor. | |
449 | */ | |
450 | lbno = bno; | |
451 | left = block; | |
452 | lbp = bp; | |
453 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, | |
454 | cur->bc_private.a.agno, rbno, 0, &rbp, | |
455 | XFS_ALLOC_BTREE_REF))) | |
456 | return error; | |
457 | right = XFS_BUF_TO_ALLOC_BLOCK(rbp); | |
458 | if ((error = xfs_btree_check_sblock(cur, right, level, rbp))) | |
459 | return error; | |
460 | } | |
461 | /* | |
462 | * Otherwise, we can't fix the imbalance. | |
463 | * Just return. This is probably a logic error, but it's not fatal. | |
464 | */ | |
465 | else { | |
466 | if (level > 0 && (error = xfs_alloc_decrement(cur, level, &i))) | |
467 | return error; | |
468 | *stat = 1; | |
469 | return 0; | |
470 | } | |
471 | /* | |
472 | * We're now going to join "left" and "right" by moving all the stuff | |
473 | * in "right" to "left" and deleting "right". | |
474 | */ | |
475 | if (level > 0) { | |
476 | /* | |
477 | * It's a non-leaf. Move keys and pointers. | |
478 | */ | |
479 | lkp = XFS_ALLOC_KEY_ADDR(left, INT_GET(left->bb_numrecs, ARCH_CONVERT) + 1, cur); | |
480 | lpp = XFS_ALLOC_PTR_ADDR(left, INT_GET(left->bb_numrecs, ARCH_CONVERT) + 1, cur); | |
481 | rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur); | |
482 | rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur); | |
483 | #ifdef DEBUG | |
484 | for (i = 0; i < INT_GET(right->bb_numrecs, ARCH_CONVERT); i++) { | |
485 | if ((error = xfs_btree_check_sptr(cur, INT_GET(rpp[i], ARCH_CONVERT), level))) | |
486 | return error; | |
487 | } | |
488 | #endif | |
489 | memcpy(lkp, rkp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*lkp)); /* INT_: structure copy */ | |
490 | memcpy(lpp, rpp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*lpp)); /* INT_: structure copy */ | |
491 | xfs_alloc_log_keys(cur, lbp, INT_GET(left->bb_numrecs, ARCH_CONVERT) + 1, | |
492 | INT_GET(left->bb_numrecs, ARCH_CONVERT) + INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
493 | xfs_alloc_log_ptrs(cur, lbp, INT_GET(left->bb_numrecs, ARCH_CONVERT) + 1, | |
494 | INT_GET(left->bb_numrecs, ARCH_CONVERT) + INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
495 | } else { | |
496 | /* | |
497 | * It's a leaf. Move records. | |
498 | */ | |
499 | lrp = XFS_ALLOC_REC_ADDR(left, INT_GET(left->bb_numrecs, ARCH_CONVERT) + 1, cur); | |
500 | rrp = XFS_ALLOC_REC_ADDR(right, 1, cur); | |
501 | memcpy(lrp, rrp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*lrp)); | |
502 | xfs_alloc_log_recs(cur, lbp, INT_GET(left->bb_numrecs, ARCH_CONVERT) + 1, | |
503 | INT_GET(left->bb_numrecs, ARCH_CONVERT) + INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
504 | } | |
505 | /* | |
506 | * If we joined with the left neighbor, set the buffer in the | |
507 | * cursor to the left block, and fix up the index. | |
508 | */ | |
509 | if (bp != lbp) { | |
510 | xfs_btree_setbuf(cur, level, lbp); | |
511 | cur->bc_ptrs[level] += INT_GET(left->bb_numrecs, ARCH_CONVERT); | |
512 | } | |
513 | /* | |
514 | * If we joined with the right neighbor and there's a level above | |
515 | * us, increment the cursor at that level. | |
516 | */ | |
517 | else if (level + 1 < cur->bc_nlevels && | |
518 | (error = xfs_alloc_increment(cur, level + 1, &i))) | |
519 | return error; | |
520 | /* | |
521 | * Fix up the number of records in the surviving block. | |
522 | */ | |
523 | INT_MOD(left->bb_numrecs, ARCH_CONVERT, INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
524 | /* | |
525 | * Fix up the right block pointer in the surviving block, and log it. | |
526 | */ | |
527 | left->bb_rightsib = right->bb_rightsib; /* INT_: direct copy */ | |
528 | xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB); | |
529 | /* | |
530 | * If there is a right sibling now, make it point to the | |
531 | * remaining block. | |
532 | */ | |
533 | if (INT_GET(left->bb_rightsib, ARCH_CONVERT) != NULLAGBLOCK) { | |
534 | xfs_alloc_block_t *rrblock; | |
535 | xfs_buf_t *rrbp; | |
536 | ||
537 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, | |
538 | cur->bc_private.a.agno, INT_GET(left->bb_rightsib, ARCH_CONVERT), 0, | |
539 | &rrbp, XFS_ALLOC_BTREE_REF))) | |
540 | return error; | |
541 | rrblock = XFS_BUF_TO_ALLOC_BLOCK(rrbp); | |
542 | if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp))) | |
543 | return error; | |
544 | INT_SET(rrblock->bb_leftsib, ARCH_CONVERT, lbno); | |
545 | xfs_alloc_log_block(cur->bc_tp, rrbp, XFS_BB_LEFTSIB); | |
546 | } | |
547 | /* | |
548 | * Free the deleting block by putting it on the freelist. | |
549 | */ | |
550 | if ((error = xfs_alloc_put_freelist(cur->bc_tp, cur->bc_private.a.agbp, | |
551 | NULL, rbno))) | |
552 | return error; | |
553 | /* | |
554 | * Since blocks move to the free list without the coordination | |
555 | * used in xfs_bmap_finish, we can't allow block to be available | |
556 | * for reallocation and non-transaction writing (user data) | |
557 | * until we know that the transaction that moved it to the free | |
558 | * list is permanently on disk. We track the blocks by declaring | |
559 | * these blocks as "busy"; the busy list is maintained on a | |
560 | * per-ag basis and each transaction records which entries | |
561 | * should be removed when the iclog commits to disk. If a | |
562 | * busy block is allocated, the iclog is pushed up to the | |
563 | * LSN that freed the block. | |
564 | */ | |
565 | xfs_alloc_mark_busy(cur->bc_tp, | |
566 | INT_GET(agf->agf_seqno, ARCH_CONVERT), bno, 1); | |
567 | ||
568 | xfs_trans_agbtree_delta(cur->bc_tp, -1); | |
569 | /* | |
570 | * Adjust the current level's cursor so that we're left referring | |
571 | * to the right node, after we're done. | |
572 | * If this leaves the ptr value 0 our caller will fix it up. | |
573 | */ | |
574 | if (level > 0) | |
575 | cur->bc_ptrs[level]--; | |
576 | /* | |
577 | * Return value means the next level up has something to do. | |
578 | */ | |
579 | *stat = 2; | |
580 | return 0; | |
581 | ||
582 | error0: | |
583 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
584 | return error; | |
585 | } | |
586 | ||
587 | /* | |
588 | * Insert one record/level. Return information to the caller | |
589 | * allowing the next level up to proceed if necessary. | |
590 | */ | |
591 | STATIC int /* error */ | |
592 | xfs_alloc_insrec( | |
593 | xfs_btree_cur_t *cur, /* btree cursor */ | |
594 | int level, /* level to insert record at */ | |
595 | xfs_agblock_t *bnop, /* i/o: block number inserted */ | |
596 | xfs_alloc_rec_t *recp, /* i/o: record data inserted */ | |
597 | xfs_btree_cur_t **curp, /* output: new cursor replacing cur */ | |
598 | int *stat) /* output: success/failure */ | |
599 | { | |
600 | xfs_agf_t *agf; /* allocation group freelist header */ | |
601 | xfs_alloc_block_t *block; /* btree block record/key lives in */ | |
602 | xfs_buf_t *bp; /* buffer for block */ | |
603 | int error; /* error return value */ | |
604 | int i; /* loop index */ | |
605 | xfs_alloc_key_t key; /* key value being inserted */ | |
606 | xfs_alloc_key_t *kp; /* pointer to btree keys */ | |
607 | xfs_agblock_t nbno; /* block number of allocated block */ | |
608 | xfs_btree_cur_t *ncur; /* new cursor to be used at next lvl */ | |
609 | xfs_alloc_key_t nkey; /* new key value, from split */ | |
610 | xfs_alloc_rec_t nrec; /* new record value, for caller */ | |
611 | int optr; /* old ptr value */ | |
612 | xfs_alloc_ptr_t *pp; /* pointer to btree addresses */ | |
613 | int ptr; /* index in btree block for this rec */ | |
614 | xfs_alloc_rec_t *rp; /* pointer to btree records */ | |
615 | ||
616 | ASSERT(INT_GET(recp->ar_blockcount, ARCH_CONVERT) > 0); | |
5bde1ba9 CH |
617 | |
618 | /* | |
619 | * GCC doesn't understand the (arguably complex) control flow in | |
620 | * this function and complains about uninitialized structure fields | |
621 | * without this. | |
622 | */ | |
623 | memset(&nrec, 0, sizeof(nrec)); | |
624 | ||
1da177e4 LT |
625 | /* |
626 | * If we made it to the root level, allocate a new root block | |
627 | * and we're done. | |
628 | */ | |
629 | if (level >= cur->bc_nlevels) { | |
630 | XFS_STATS_INC(xs_abt_insrec); | |
631 | if ((error = xfs_alloc_newroot(cur, &i))) | |
632 | return error; | |
633 | *bnop = NULLAGBLOCK; | |
634 | *stat = i; | |
635 | return 0; | |
636 | } | |
637 | /* | |
638 | * Make a key out of the record data to be inserted, and save it. | |
639 | */ | |
640 | key.ar_startblock = recp->ar_startblock; /* INT_: direct copy */ | |
641 | key.ar_blockcount = recp->ar_blockcount; /* INT_: direct copy */ | |
642 | optr = ptr = cur->bc_ptrs[level]; | |
643 | /* | |
644 | * If we're off the left edge, return failure. | |
645 | */ | |
646 | if (ptr == 0) { | |
647 | *stat = 0; | |
648 | return 0; | |
649 | } | |
650 | XFS_STATS_INC(xs_abt_insrec); | |
651 | /* | |
652 | * Get pointers to the btree buffer and block. | |
653 | */ | |
654 | bp = cur->bc_bufs[level]; | |
655 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
656 | #ifdef DEBUG | |
657 | if ((error = xfs_btree_check_sblock(cur, block, level, bp))) | |
658 | return error; | |
659 | /* | |
660 | * Check that the new entry is being inserted in the right place. | |
661 | */ | |
662 | if (ptr <= INT_GET(block->bb_numrecs, ARCH_CONVERT)) { | |
663 | if (level == 0) { | |
664 | rp = XFS_ALLOC_REC_ADDR(block, ptr, cur); | |
665 | xfs_btree_check_rec(cur->bc_btnum, recp, rp); | |
666 | } else { | |
667 | kp = XFS_ALLOC_KEY_ADDR(block, ptr, cur); | |
668 | xfs_btree_check_key(cur->bc_btnum, &key, kp); | |
669 | } | |
670 | } | |
671 | #endif | |
672 | nbno = NULLAGBLOCK; | |
673 | ncur = (xfs_btree_cur_t *)0; | |
674 | /* | |
675 | * If the block is full, we can't insert the new entry until we | |
676 | * make the block un-full. | |
677 | */ | |
678 | if (INT_GET(block->bb_numrecs, ARCH_CONVERT) == XFS_ALLOC_BLOCK_MAXRECS(level, cur)) { | |
679 | /* | |
680 | * First, try shifting an entry to the right neighbor. | |
681 | */ | |
682 | if ((error = xfs_alloc_rshift(cur, level, &i))) | |
683 | return error; | |
684 | if (i) { | |
685 | /* nothing */ | |
686 | } | |
687 | /* | |
688 | * Next, try shifting an entry to the left neighbor. | |
689 | */ | |
690 | else { | |
691 | if ((error = xfs_alloc_lshift(cur, level, &i))) | |
692 | return error; | |
693 | if (i) | |
694 | optr = ptr = cur->bc_ptrs[level]; | |
695 | else { | |
696 | /* | |
697 | * Next, try splitting the current block in | |
698 | * half. If this works we have to re-set our | |
699 | * variables because we could be in a | |
700 | * different block now. | |
701 | */ | |
702 | if ((error = xfs_alloc_split(cur, level, &nbno, | |
703 | &nkey, &ncur, &i))) | |
704 | return error; | |
705 | if (i) { | |
706 | bp = cur->bc_bufs[level]; | |
707 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
708 | #ifdef DEBUG | |
709 | if ((error = | |
710 | xfs_btree_check_sblock(cur, | |
711 | block, level, bp))) | |
712 | return error; | |
713 | #endif | |
714 | ptr = cur->bc_ptrs[level]; | |
715 | nrec.ar_startblock = nkey.ar_startblock; /* INT_: direct copy */ | |
716 | nrec.ar_blockcount = nkey.ar_blockcount; /* INT_: direct copy */ | |
717 | } | |
718 | /* | |
719 | * Otherwise the insert fails. | |
720 | */ | |
721 | else { | |
722 | *stat = 0; | |
723 | return 0; | |
724 | } | |
725 | } | |
726 | } | |
727 | } | |
728 | /* | |
729 | * At this point we know there's room for our new entry in the block | |
730 | * we're pointing at. | |
731 | */ | |
732 | if (level > 0) { | |
733 | /* | |
734 | * It's a non-leaf entry. Make a hole for the new data | |
735 | * in the key and ptr regions of the block. | |
736 | */ | |
737 | kp = XFS_ALLOC_KEY_ADDR(block, 1, cur); | |
738 | pp = XFS_ALLOC_PTR_ADDR(block, 1, cur); | |
739 | #ifdef DEBUG | |
740 | for (i = INT_GET(block->bb_numrecs, ARCH_CONVERT); i >= ptr; i--) { | |
741 | if ((error = xfs_btree_check_sptr(cur, INT_GET(pp[i - 1], ARCH_CONVERT), level))) | |
742 | return error; | |
743 | } | |
744 | #endif | |
745 | memmove(&kp[ptr], &kp[ptr - 1], | |
746 | (INT_GET(block->bb_numrecs, ARCH_CONVERT) - ptr + 1) * sizeof(*kp)); /* INT_: copy */ | |
747 | memmove(&pp[ptr], &pp[ptr - 1], | |
748 | (INT_GET(block->bb_numrecs, ARCH_CONVERT) - ptr + 1) * sizeof(*pp)); /* INT_: copy */ | |
749 | #ifdef DEBUG | |
750 | if ((error = xfs_btree_check_sptr(cur, *bnop, level))) | |
751 | return error; | |
752 | #endif | |
753 | /* | |
754 | * Now stuff the new data in, bump numrecs and log the new data. | |
755 | */ | |
756 | kp[ptr - 1] = key; | |
757 | INT_SET(pp[ptr - 1], ARCH_CONVERT, *bnop); | |
758 | INT_MOD(block->bb_numrecs, ARCH_CONVERT, +1); | |
759 | xfs_alloc_log_keys(cur, bp, ptr, INT_GET(block->bb_numrecs, ARCH_CONVERT)); | |
760 | xfs_alloc_log_ptrs(cur, bp, ptr, INT_GET(block->bb_numrecs, ARCH_CONVERT)); | |
761 | #ifdef DEBUG | |
762 | if (ptr < INT_GET(block->bb_numrecs, ARCH_CONVERT)) | |
763 | xfs_btree_check_key(cur->bc_btnum, kp + ptr - 1, | |
764 | kp + ptr); | |
765 | #endif | |
766 | } else { | |
767 | /* | |
768 | * It's a leaf entry. Make a hole for the new record. | |
769 | */ | |
770 | rp = XFS_ALLOC_REC_ADDR(block, 1, cur); | |
771 | memmove(&rp[ptr], &rp[ptr - 1], | |
772 | (INT_GET(block->bb_numrecs, ARCH_CONVERT) - ptr + 1) * sizeof(*rp)); | |
773 | /* | |
774 | * Now stuff the new record in, bump numrecs | |
775 | * and log the new data. | |
776 | */ | |
777 | rp[ptr - 1] = *recp; /* INT_: struct copy */ | |
778 | INT_MOD(block->bb_numrecs, ARCH_CONVERT, +1); | |
779 | xfs_alloc_log_recs(cur, bp, ptr, INT_GET(block->bb_numrecs, ARCH_CONVERT)); | |
780 | #ifdef DEBUG | |
781 | if (ptr < INT_GET(block->bb_numrecs, ARCH_CONVERT)) | |
782 | xfs_btree_check_rec(cur->bc_btnum, rp + ptr - 1, | |
783 | rp + ptr); | |
784 | #endif | |
785 | } | |
786 | /* | |
787 | * Log the new number of records in the btree header. | |
788 | */ | |
789 | xfs_alloc_log_block(cur->bc_tp, bp, XFS_BB_NUMRECS); | |
790 | /* | |
791 | * If we inserted at the start of a block, update the parents' keys. | |
792 | */ | |
793 | if (optr == 1 && (error = xfs_alloc_updkey(cur, &key, level + 1))) | |
794 | return error; | |
795 | /* | |
796 | * Look to see if the longest extent in the allocation group | |
797 | * needs to be updated. | |
798 | */ | |
799 | ||
800 | agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | |
801 | if (level == 0 && | |
802 | cur->bc_btnum == XFS_BTNUM_CNT && | |
803 | INT_GET(block->bb_rightsib, ARCH_CONVERT) == NULLAGBLOCK && | |
804 | INT_GET(recp->ar_blockcount, ARCH_CONVERT) > INT_GET(agf->agf_longest, ARCH_CONVERT)) { | |
805 | /* | |
806 | * If this is a leaf in the by-size btree and there | |
807 | * is no right sibling block and this block is bigger | |
808 | * than the previous longest block, update it. | |
809 | */ | |
810 | INT_COPY(agf->agf_longest, recp->ar_blockcount, ARCH_CONVERT); | |
811 | cur->bc_mp->m_perag[INT_GET(agf->agf_seqno, ARCH_CONVERT)].pagf_longest | |
812 | = INT_GET(recp->ar_blockcount, ARCH_CONVERT); | |
813 | xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, | |
814 | XFS_AGF_LONGEST); | |
815 | } | |
816 | /* | |
817 | * Return the new block number, if any. | |
818 | * If there is one, give back a record value and a cursor too. | |
819 | */ | |
820 | *bnop = nbno; | |
821 | if (nbno != NULLAGBLOCK) { | |
822 | *recp = nrec; /* INT_: struct copy */ | |
823 | *curp = ncur; /* INT_: struct copy */ | |
824 | } | |
825 | *stat = 1; | |
826 | return 0; | |
827 | } | |
828 | ||
829 | /* | |
830 | * Log header fields from a btree block. | |
831 | */ | |
832 | STATIC void | |
833 | xfs_alloc_log_block( | |
834 | xfs_trans_t *tp, /* transaction pointer */ | |
835 | xfs_buf_t *bp, /* buffer containing btree block */ | |
836 | int fields) /* mask of fields: XFS_BB_... */ | |
837 | { | |
838 | int first; /* first byte offset logged */ | |
839 | int last; /* last byte offset logged */ | |
840 | static const short offsets[] = { /* table of offsets */ | |
841 | offsetof(xfs_alloc_block_t, bb_magic), | |
842 | offsetof(xfs_alloc_block_t, bb_level), | |
843 | offsetof(xfs_alloc_block_t, bb_numrecs), | |
844 | offsetof(xfs_alloc_block_t, bb_leftsib), | |
845 | offsetof(xfs_alloc_block_t, bb_rightsib), | |
846 | sizeof(xfs_alloc_block_t) | |
847 | }; | |
848 | ||
849 | xfs_btree_offsets(fields, offsets, XFS_BB_NUM_BITS, &first, &last); | |
850 | xfs_trans_log_buf(tp, bp, first, last); | |
851 | } | |
852 | ||
853 | /* | |
854 | * Log keys from a btree block (nonleaf). | |
855 | */ | |
856 | STATIC void | |
857 | xfs_alloc_log_keys( | |
858 | xfs_btree_cur_t *cur, /* btree cursor */ | |
859 | xfs_buf_t *bp, /* buffer containing btree block */ | |
860 | int kfirst, /* index of first key to log */ | |
861 | int klast) /* index of last key to log */ | |
862 | { | |
863 | xfs_alloc_block_t *block; /* btree block to log from */ | |
864 | int first; /* first byte offset logged */ | |
865 | xfs_alloc_key_t *kp; /* key pointer in btree block */ | |
866 | int last; /* last byte offset logged */ | |
867 | ||
868 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
869 | kp = XFS_ALLOC_KEY_ADDR(block, 1, cur); | |
870 | first = (int)((xfs_caddr_t)&kp[kfirst - 1] - (xfs_caddr_t)block); | |
871 | last = (int)(((xfs_caddr_t)&kp[klast] - 1) - (xfs_caddr_t)block); | |
872 | xfs_trans_log_buf(cur->bc_tp, bp, first, last); | |
873 | } | |
874 | ||
875 | /* | |
876 | * Log block pointer fields from a btree block (nonleaf). | |
877 | */ | |
878 | STATIC void | |
879 | xfs_alloc_log_ptrs( | |
880 | xfs_btree_cur_t *cur, /* btree cursor */ | |
881 | xfs_buf_t *bp, /* buffer containing btree block */ | |
882 | int pfirst, /* index of first pointer to log */ | |
883 | int plast) /* index of last pointer to log */ | |
884 | { | |
885 | xfs_alloc_block_t *block; /* btree block to log from */ | |
886 | int first; /* first byte offset logged */ | |
887 | int last; /* last byte offset logged */ | |
888 | xfs_alloc_ptr_t *pp; /* block-pointer pointer in btree blk */ | |
889 | ||
890 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
891 | pp = XFS_ALLOC_PTR_ADDR(block, 1, cur); | |
892 | first = (int)((xfs_caddr_t)&pp[pfirst - 1] - (xfs_caddr_t)block); | |
893 | last = (int)(((xfs_caddr_t)&pp[plast] - 1) - (xfs_caddr_t)block); | |
894 | xfs_trans_log_buf(cur->bc_tp, bp, first, last); | |
895 | } | |
896 | ||
897 | /* | |
898 | * Log records from a btree block (leaf). | |
899 | */ | |
900 | STATIC void | |
901 | xfs_alloc_log_recs( | |
902 | xfs_btree_cur_t *cur, /* btree cursor */ | |
903 | xfs_buf_t *bp, /* buffer containing btree block */ | |
904 | int rfirst, /* index of first record to log */ | |
905 | int rlast) /* index of last record to log */ | |
906 | { | |
907 | xfs_alloc_block_t *block; /* btree block to log from */ | |
908 | int first; /* first byte offset logged */ | |
909 | int last; /* last byte offset logged */ | |
910 | xfs_alloc_rec_t *rp; /* record pointer for btree block */ | |
911 | ||
912 | ||
913 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
914 | rp = XFS_ALLOC_REC_ADDR(block, 1, cur); | |
915 | #ifdef DEBUG | |
916 | { | |
917 | xfs_agf_t *agf; | |
918 | xfs_alloc_rec_t *p; | |
919 | ||
920 | agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | |
921 | for (p = &rp[rfirst - 1]; p <= &rp[rlast - 1]; p++) | |
922 | ASSERT(INT_GET(p->ar_startblock, ARCH_CONVERT) + INT_GET(p->ar_blockcount, ARCH_CONVERT) <= | |
923 | INT_GET(agf->agf_length, ARCH_CONVERT)); | |
924 | } | |
925 | #endif | |
926 | first = (int)((xfs_caddr_t)&rp[rfirst - 1] - (xfs_caddr_t)block); | |
927 | last = (int)(((xfs_caddr_t)&rp[rlast] - 1) - (xfs_caddr_t)block); | |
928 | xfs_trans_log_buf(cur->bc_tp, bp, first, last); | |
929 | } | |
930 | ||
931 | /* | |
932 | * Lookup the record. The cursor is made to point to it, based on dir. | |
933 | * Return 0 if can't find any such record, 1 for success. | |
934 | */ | |
935 | STATIC int /* error */ | |
936 | xfs_alloc_lookup( | |
937 | xfs_btree_cur_t *cur, /* btree cursor */ | |
938 | xfs_lookup_t dir, /* <=, ==, or >= */ | |
939 | int *stat) /* success/failure */ | |
940 | { | |
941 | xfs_agblock_t agbno; /* a.g. relative btree block number */ | |
942 | xfs_agnumber_t agno; /* allocation group number */ | |
943 | xfs_alloc_block_t *block=NULL; /* current btree block */ | |
944 | int diff; /* difference for the current key */ | |
945 | int error; /* error return value */ | |
946 | int keyno=0; /* current key number */ | |
947 | int level; /* level in the btree */ | |
948 | xfs_mount_t *mp; /* file system mount point */ | |
949 | ||
950 | XFS_STATS_INC(xs_abt_lookup); | |
951 | /* | |
952 | * Get the allocation group header, and the root block number. | |
953 | */ | |
954 | mp = cur->bc_mp; | |
955 | ||
956 | { | |
957 | xfs_agf_t *agf; /* a.g. freespace header */ | |
958 | ||
959 | agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | |
960 | agno = INT_GET(agf->agf_seqno, ARCH_CONVERT); | |
961 | agbno = INT_GET(agf->agf_roots[cur->bc_btnum], ARCH_CONVERT); | |
962 | } | |
963 | /* | |
964 | * Iterate over each level in the btree, starting at the root. | |
965 | * For each level above the leaves, find the key we need, based | |
966 | * on the lookup record, then follow the corresponding block | |
967 | * pointer down to the next level. | |
968 | */ | |
969 | for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) { | |
970 | xfs_buf_t *bp; /* buffer pointer for btree block */ | |
971 | xfs_daddr_t d; /* disk address of btree block */ | |
972 | ||
973 | /* | |
974 | * Get the disk address we're looking for. | |
975 | */ | |
976 | d = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
977 | /* | |
978 | * If the old buffer at this level is for a different block, | |
979 | * throw it away, otherwise just use it. | |
980 | */ | |
981 | bp = cur->bc_bufs[level]; | |
982 | if (bp && XFS_BUF_ADDR(bp) != d) | |
983 | bp = (xfs_buf_t *)0; | |
984 | if (!bp) { | |
985 | /* | |
986 | * Need to get a new buffer. Read it, then | |
987 | * set it in the cursor, releasing the old one. | |
988 | */ | |
989 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, agno, | |
990 | agbno, 0, &bp, XFS_ALLOC_BTREE_REF))) | |
991 | return error; | |
992 | xfs_btree_setbuf(cur, level, bp); | |
993 | /* | |
994 | * Point to the btree block, now that we have the buffer | |
995 | */ | |
996 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
997 | if ((error = xfs_btree_check_sblock(cur, block, level, | |
998 | bp))) | |
999 | return error; | |
1000 | } else | |
1001 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
1002 | /* | |
1003 | * If we already had a key match at a higher level, we know | |
1004 | * we need to use the first entry in this block. | |
1005 | */ | |
1006 | if (diff == 0) | |
1007 | keyno = 1; | |
1008 | /* | |
1009 | * Otherwise we need to search this block. Do a binary search. | |
1010 | */ | |
1011 | else { | |
1012 | int high; /* high entry number */ | |
1013 | xfs_alloc_key_t *kkbase=NULL;/* base of keys in block */ | |
1014 | xfs_alloc_rec_t *krbase=NULL;/* base of records in block */ | |
1015 | int low; /* low entry number */ | |
1016 | ||
1017 | /* | |
1018 | * Get a pointer to keys or records. | |
1019 | */ | |
1020 | if (level > 0) | |
1021 | kkbase = XFS_ALLOC_KEY_ADDR(block, 1, cur); | |
1022 | else | |
1023 | krbase = XFS_ALLOC_REC_ADDR(block, 1, cur); | |
1024 | /* | |
1025 | * Set low and high entry numbers, 1-based. | |
1026 | */ | |
1027 | low = 1; | |
1028 | if (!(high = INT_GET(block->bb_numrecs, ARCH_CONVERT))) { | |
1029 | /* | |
1030 | * If the block is empty, the tree must | |
1031 | * be an empty leaf. | |
1032 | */ | |
1033 | ASSERT(level == 0 && cur->bc_nlevels == 1); | |
1034 | cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE; | |
1035 | *stat = 0; | |
1036 | return 0; | |
1037 | } | |
1038 | /* | |
1039 | * Binary search the block. | |
1040 | */ | |
1041 | while (low <= high) { | |
1042 | xfs_extlen_t blockcount; /* key value */ | |
1043 | xfs_agblock_t startblock; /* key value */ | |
1044 | ||
1045 | XFS_STATS_INC(xs_abt_compare); | |
1046 | /* | |
1047 | * keyno is average of low and high. | |
1048 | */ | |
1049 | keyno = (low + high) >> 1; | |
1050 | /* | |
1051 | * Get startblock & blockcount. | |
1052 | */ | |
1053 | if (level > 0) { | |
1054 | xfs_alloc_key_t *kkp; | |
1055 | ||
1056 | kkp = kkbase + keyno - 1; | |
1057 | startblock = INT_GET(kkp->ar_startblock, ARCH_CONVERT); | |
1058 | blockcount = INT_GET(kkp->ar_blockcount, ARCH_CONVERT); | |
1059 | } else { | |
1060 | xfs_alloc_rec_t *krp; | |
1061 | ||
1062 | krp = krbase + keyno - 1; | |
1063 | startblock = INT_GET(krp->ar_startblock, ARCH_CONVERT); | |
1064 | blockcount = INT_GET(krp->ar_blockcount, ARCH_CONVERT); | |
1065 | } | |
1066 | /* | |
1067 | * Compute difference to get next direction. | |
1068 | */ | |
1069 | if (cur->bc_btnum == XFS_BTNUM_BNO) | |
1070 | diff = (int)startblock - | |
1071 | (int)cur->bc_rec.a.ar_startblock; | |
1072 | else if (!(diff = (int)blockcount - | |
1073 | (int)cur->bc_rec.a.ar_blockcount)) | |
1074 | diff = (int)startblock - | |
1075 | (int)cur->bc_rec.a.ar_startblock; | |
1076 | /* | |
1077 | * Less than, move right. | |
1078 | */ | |
1079 | if (diff < 0) | |
1080 | low = keyno + 1; | |
1081 | /* | |
1082 | * Greater than, move left. | |
1083 | */ | |
1084 | else if (diff > 0) | |
1085 | high = keyno - 1; | |
1086 | /* | |
1087 | * Equal, we're done. | |
1088 | */ | |
1089 | else | |
1090 | break; | |
1091 | } | |
1092 | } | |
1093 | /* | |
1094 | * If there are more levels, set up for the next level | |
1095 | * by getting the block number and filling in the cursor. | |
1096 | */ | |
1097 | if (level > 0) { | |
1098 | /* | |
1099 | * If we moved left, need the previous key number, | |
1100 | * unless there isn't one. | |
1101 | */ | |
1102 | if (diff > 0 && --keyno < 1) | |
1103 | keyno = 1; | |
1104 | agbno = INT_GET(*XFS_ALLOC_PTR_ADDR(block, keyno, cur), ARCH_CONVERT); | |
1105 | #ifdef DEBUG | |
1106 | if ((error = xfs_btree_check_sptr(cur, agbno, level))) | |
1107 | return error; | |
1108 | #endif | |
1109 | cur->bc_ptrs[level] = keyno; | |
1110 | } | |
1111 | } | |
1112 | /* | |
1113 | * Done with the search. | |
1114 | * See if we need to adjust the results. | |
1115 | */ | |
1116 | if (dir != XFS_LOOKUP_LE && diff < 0) { | |
1117 | keyno++; | |
1118 | /* | |
1119 | * If ge search and we went off the end of the block, but it's | |
1120 | * not the last block, we're in the wrong block. | |
1121 | */ | |
1122 | if (dir == XFS_LOOKUP_GE && | |
1123 | keyno > INT_GET(block->bb_numrecs, ARCH_CONVERT) && | |
1124 | INT_GET(block->bb_rightsib, ARCH_CONVERT) != NULLAGBLOCK) { | |
1125 | int i; | |
1126 | ||
1127 | cur->bc_ptrs[0] = keyno; | |
1128 | if ((error = xfs_alloc_increment(cur, 0, &i))) | |
1129 | return error; | |
1130 | XFS_WANT_CORRUPTED_RETURN(i == 1); | |
1131 | *stat = 1; | |
1132 | return 0; | |
1133 | } | |
1134 | } | |
1135 | else if (dir == XFS_LOOKUP_LE && diff > 0) | |
1136 | keyno--; | |
1137 | cur->bc_ptrs[0] = keyno; | |
1138 | /* | |
1139 | * Return if we succeeded or not. | |
1140 | */ | |
1141 | if (keyno == 0 || keyno > INT_GET(block->bb_numrecs, ARCH_CONVERT)) | |
1142 | *stat = 0; | |
1143 | else | |
1144 | *stat = ((dir != XFS_LOOKUP_EQ) || (diff == 0)); | |
1145 | return 0; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * Move 1 record left from cur/level if possible. | |
1150 | * Update cur to reflect the new path. | |
1151 | */ | |
1152 | STATIC int /* error */ | |
1153 | xfs_alloc_lshift( | |
1154 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1155 | int level, /* level to shift record on */ | |
1156 | int *stat) /* success/failure */ | |
1157 | { | |
1158 | int error; /* error return value */ | |
1159 | #ifdef DEBUG | |
1160 | int i; /* loop index */ | |
1161 | #endif | |
1162 | xfs_alloc_key_t key; /* key value for leaf level upward */ | |
1163 | xfs_buf_t *lbp; /* buffer for left neighbor block */ | |
1164 | xfs_alloc_block_t *left; /* left neighbor btree block */ | |
1165 | int nrec; /* new number of left block entries */ | |
1166 | xfs_buf_t *rbp; /* buffer for right (current) block */ | |
1167 | xfs_alloc_block_t *right; /* right (current) btree block */ | |
1168 | xfs_alloc_key_t *rkp=NULL; /* key pointer for right block */ | |
1169 | xfs_alloc_ptr_t *rpp=NULL; /* address pointer for right block */ | |
1170 | xfs_alloc_rec_t *rrp=NULL; /* record pointer for right block */ | |
1171 | ||
1172 | /* | |
1173 | * Set up variables for this block as "right". | |
1174 | */ | |
1175 | rbp = cur->bc_bufs[level]; | |
1176 | right = XFS_BUF_TO_ALLOC_BLOCK(rbp); | |
1177 | #ifdef DEBUG | |
1178 | if ((error = xfs_btree_check_sblock(cur, right, level, rbp))) | |
1179 | return error; | |
1180 | #endif | |
1181 | /* | |
1182 | * If we've got no left sibling then we can't shift an entry left. | |
1183 | */ | |
1184 | if (INT_GET(right->bb_leftsib, ARCH_CONVERT) == NULLAGBLOCK) { | |
1185 | *stat = 0; | |
1186 | return 0; | |
1187 | } | |
1188 | /* | |
1189 | * If the cursor entry is the one that would be moved, don't | |
1190 | * do it... it's too complicated. | |
1191 | */ | |
1192 | if (cur->bc_ptrs[level] <= 1) { | |
1193 | *stat = 0; | |
1194 | return 0; | |
1195 | } | |
1196 | /* | |
1197 | * Set up the left neighbor as "left". | |
1198 | */ | |
1199 | if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp, | |
1200 | cur->bc_private.a.agno, INT_GET(right->bb_leftsib, ARCH_CONVERT), 0, &lbp, | |
1201 | XFS_ALLOC_BTREE_REF))) | |
1202 | return error; | |
1203 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | |
1204 | if ((error = xfs_btree_check_sblock(cur, left, level, lbp))) | |
1205 | return error; | |
1206 | /* | |
1207 | * If it's full, it can't take another entry. | |
1208 | */ | |
1209 | if (INT_GET(left->bb_numrecs, ARCH_CONVERT) == XFS_ALLOC_BLOCK_MAXRECS(level, cur)) { | |
1210 | *stat = 0; | |
1211 | return 0; | |
1212 | } | |
1213 | nrec = INT_GET(left->bb_numrecs, ARCH_CONVERT) + 1; | |
1214 | /* | |
1215 | * If non-leaf, copy a key and a ptr to the left block. | |
1216 | */ | |
1217 | if (level > 0) { | |
1218 | xfs_alloc_key_t *lkp; /* key pointer for left block */ | |
1219 | xfs_alloc_ptr_t *lpp; /* address pointer for left block */ | |
1220 | ||
1221 | lkp = XFS_ALLOC_KEY_ADDR(left, nrec, cur); | |
1222 | rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur); | |
1223 | *lkp = *rkp; | |
1224 | xfs_alloc_log_keys(cur, lbp, nrec, nrec); | |
1225 | lpp = XFS_ALLOC_PTR_ADDR(left, nrec, cur); | |
1226 | rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur); | |
1227 | #ifdef DEBUG | |
1228 | if ((error = xfs_btree_check_sptr(cur, INT_GET(*rpp, ARCH_CONVERT), level))) | |
1229 | return error; | |
1230 | #endif | |
1231 | *lpp = *rpp; /* INT_: copy */ | |
1232 | xfs_alloc_log_ptrs(cur, lbp, nrec, nrec); | |
1233 | xfs_btree_check_key(cur->bc_btnum, lkp - 1, lkp); | |
1234 | } | |
1235 | /* | |
1236 | * If leaf, copy a record to the left block. | |
1237 | */ | |
1238 | else { | |
1239 | xfs_alloc_rec_t *lrp; /* record pointer for left block */ | |
1240 | ||
1241 | lrp = XFS_ALLOC_REC_ADDR(left, nrec, cur); | |
1242 | rrp = XFS_ALLOC_REC_ADDR(right, 1, cur); | |
1243 | *lrp = *rrp; | |
1244 | xfs_alloc_log_recs(cur, lbp, nrec, nrec); | |
1245 | xfs_btree_check_rec(cur->bc_btnum, lrp - 1, lrp); | |
1246 | } | |
1247 | /* | |
1248 | * Bump and log left's numrecs, decrement and log right's numrecs. | |
1249 | */ | |
1250 | INT_MOD(left->bb_numrecs, ARCH_CONVERT, +1); | |
1251 | xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS); | |
1252 | INT_MOD(right->bb_numrecs, ARCH_CONVERT, -1); | |
1253 | xfs_alloc_log_block(cur->bc_tp, rbp, XFS_BB_NUMRECS); | |
1254 | /* | |
1255 | * Slide the contents of right down one entry. | |
1256 | */ | |
1257 | if (level > 0) { | |
1258 | #ifdef DEBUG | |
1259 | for (i = 0; i < INT_GET(right->bb_numrecs, ARCH_CONVERT); i++) { | |
1260 | if ((error = xfs_btree_check_sptr(cur, INT_GET(rpp[i + 1], ARCH_CONVERT), | |
1261 | level))) | |
1262 | return error; | |
1263 | } | |
1264 | #endif | |
1265 | memmove(rkp, rkp + 1, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rkp)); | |
1266 | memmove(rpp, rpp + 1, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rpp)); | |
1267 | xfs_alloc_log_keys(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
1268 | xfs_alloc_log_ptrs(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
1269 | } else { | |
1270 | memmove(rrp, rrp + 1, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rrp)); | |
1271 | xfs_alloc_log_recs(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
1272 | key.ar_startblock = rrp->ar_startblock; /* INT_: direct copy */ | |
1273 | key.ar_blockcount = rrp->ar_blockcount; /* INT_: direct copy */ | |
1274 | rkp = &key; | |
1275 | } | |
1276 | /* | |
1277 | * Update the parent key values of right. | |
1278 | */ | |
1279 | if ((error = xfs_alloc_updkey(cur, rkp, level + 1))) | |
1280 | return error; | |
1281 | /* | |
1282 | * Slide the cursor value left one. | |
1283 | */ | |
1284 | cur->bc_ptrs[level]--; | |
1285 | *stat = 1; | |
1286 | return 0; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * Allocate a new root block, fill it in. | |
1291 | */ | |
1292 | STATIC int /* error */ | |
1293 | xfs_alloc_newroot( | |
1294 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1295 | int *stat) /* success/failure */ | |
1296 | { | |
1297 | int error; /* error return value */ | |
1298 | xfs_agblock_t lbno; /* left block number */ | |
1299 | xfs_buf_t *lbp; /* left btree buffer */ | |
1300 | xfs_alloc_block_t *left; /* left btree block */ | |
1301 | xfs_mount_t *mp; /* mount structure */ | |
1302 | xfs_agblock_t nbno; /* new block number */ | |
1303 | xfs_buf_t *nbp; /* new (root) buffer */ | |
1304 | xfs_alloc_block_t *new; /* new (root) btree block */ | |
1305 | int nptr; /* new value for key index, 1 or 2 */ | |
1306 | xfs_agblock_t rbno; /* right block number */ | |
1307 | xfs_buf_t *rbp; /* right btree buffer */ | |
1308 | xfs_alloc_block_t *right; /* right btree block */ | |
1309 | ||
1310 | mp = cur->bc_mp; | |
1311 | ||
1312 | ASSERT(cur->bc_nlevels < XFS_AG_MAXLEVELS(mp)); | |
1313 | /* | |
1314 | * Get a buffer from the freelist blocks, for the new root. | |
1315 | */ | |
1316 | if ((error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp, | |
1317 | &nbno))) | |
1318 | return error; | |
1319 | /* | |
1320 | * None available, we fail. | |
1321 | */ | |
1322 | if (nbno == NULLAGBLOCK) { | |
1323 | *stat = 0; | |
1324 | return 0; | |
1325 | } | |
1326 | xfs_trans_agbtree_delta(cur->bc_tp, 1); | |
1327 | nbp = xfs_btree_get_bufs(mp, cur->bc_tp, cur->bc_private.a.agno, nbno, | |
1328 | 0); | |
1329 | new = XFS_BUF_TO_ALLOC_BLOCK(nbp); | |
1330 | /* | |
1331 | * Set the root data in the a.g. freespace structure. | |
1332 | */ | |
1333 | { | |
1334 | xfs_agf_t *agf; /* a.g. freespace header */ | |
1335 | xfs_agnumber_t seqno; | |
1336 | ||
1337 | agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | |
1338 | INT_SET(agf->agf_roots[cur->bc_btnum], ARCH_CONVERT, nbno); | |
1339 | INT_MOD(agf->agf_levels[cur->bc_btnum], ARCH_CONVERT, 1); | |
1340 | seqno = INT_GET(agf->agf_seqno, ARCH_CONVERT); | |
1341 | mp->m_perag[seqno].pagf_levels[cur->bc_btnum]++; | |
1342 | xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, | |
1343 | XFS_AGF_ROOTS | XFS_AGF_LEVELS); | |
1344 | } | |
1345 | /* | |
1346 | * At the previous root level there are now two blocks: the old | |
1347 | * root, and the new block generated when it was split. | |
1348 | * We don't know which one the cursor is pointing at, so we | |
1349 | * set up variables "left" and "right" for each case. | |
1350 | */ | |
1351 | lbp = cur->bc_bufs[cur->bc_nlevels - 1]; | |
1352 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | |
1353 | #ifdef DEBUG | |
1354 | if ((error = xfs_btree_check_sblock(cur, left, cur->bc_nlevels - 1, lbp))) | |
1355 | return error; | |
1356 | #endif | |
1357 | if (INT_GET(left->bb_rightsib, ARCH_CONVERT) != NULLAGBLOCK) { | |
1358 | /* | |
1359 | * Our block is left, pick up the right block. | |
1360 | */ | |
1361 | lbno = XFS_DADDR_TO_AGBNO(mp, XFS_BUF_ADDR(lbp)); | |
1362 | rbno = INT_GET(left->bb_rightsib, ARCH_CONVERT); | |
1363 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, | |
1364 | cur->bc_private.a.agno, rbno, 0, &rbp, | |
1365 | XFS_ALLOC_BTREE_REF))) | |
1366 | return error; | |
1367 | right = XFS_BUF_TO_ALLOC_BLOCK(rbp); | |
1368 | if ((error = xfs_btree_check_sblock(cur, right, | |
1369 | cur->bc_nlevels - 1, rbp))) | |
1370 | return error; | |
1371 | nptr = 1; | |
1372 | } else { | |
1373 | /* | |
1374 | * Our block is right, pick up the left block. | |
1375 | */ | |
1376 | rbp = lbp; | |
1377 | right = left; | |
1378 | rbno = XFS_DADDR_TO_AGBNO(mp, XFS_BUF_ADDR(rbp)); | |
1379 | lbno = INT_GET(right->bb_leftsib, ARCH_CONVERT); | |
1380 | if ((error = xfs_btree_read_bufs(mp, cur->bc_tp, | |
1381 | cur->bc_private.a.agno, lbno, 0, &lbp, | |
1382 | XFS_ALLOC_BTREE_REF))) | |
1383 | return error; | |
1384 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | |
1385 | if ((error = xfs_btree_check_sblock(cur, left, | |
1386 | cur->bc_nlevels - 1, lbp))) | |
1387 | return error; | |
1388 | nptr = 2; | |
1389 | } | |
1390 | /* | |
1391 | * Fill in the new block's btree header and log it. | |
1392 | */ | |
1393 | INT_SET(new->bb_magic, ARCH_CONVERT, xfs_magics[cur->bc_btnum]); | |
1394 | INT_SET(new->bb_level, ARCH_CONVERT, (__uint16_t)cur->bc_nlevels); | |
1395 | INT_SET(new->bb_numrecs, ARCH_CONVERT, 2); | |
1396 | INT_SET(new->bb_leftsib, ARCH_CONVERT, NULLAGBLOCK); | |
1397 | INT_SET(new->bb_rightsib, ARCH_CONVERT, NULLAGBLOCK); | |
1398 | xfs_alloc_log_block(cur->bc_tp, nbp, XFS_BB_ALL_BITS); | |
1399 | ASSERT(lbno != NULLAGBLOCK && rbno != NULLAGBLOCK); | |
1400 | /* | |
1401 | * Fill in the key data in the new root. | |
1402 | */ | |
1403 | { | |
1404 | xfs_alloc_key_t *kp; /* btree key pointer */ | |
1405 | ||
1406 | kp = XFS_ALLOC_KEY_ADDR(new, 1, cur); | |
1407 | if (INT_GET(left->bb_level, ARCH_CONVERT) > 0) { | |
1408 | kp[0] = *XFS_ALLOC_KEY_ADDR(left, 1, cur); /* INT_: structure copy */ | |
1409 | kp[1] = *XFS_ALLOC_KEY_ADDR(right, 1, cur);/* INT_: structure copy */ | |
1410 | } else { | |
1411 | xfs_alloc_rec_t *rp; /* btree record pointer */ | |
1412 | ||
1413 | rp = XFS_ALLOC_REC_ADDR(left, 1, cur); | |
1414 | kp[0].ar_startblock = rp->ar_startblock; /* INT_: direct copy */ | |
1415 | kp[0].ar_blockcount = rp->ar_blockcount; /* INT_: direct copy */ | |
1416 | rp = XFS_ALLOC_REC_ADDR(right, 1, cur); | |
1417 | kp[1].ar_startblock = rp->ar_startblock; /* INT_: direct copy */ | |
1418 | kp[1].ar_blockcount = rp->ar_blockcount; /* INT_: direct copy */ | |
1419 | } | |
1420 | } | |
1421 | xfs_alloc_log_keys(cur, nbp, 1, 2); | |
1422 | /* | |
1423 | * Fill in the pointer data in the new root. | |
1424 | */ | |
1425 | { | |
1426 | xfs_alloc_ptr_t *pp; /* btree address pointer */ | |
1427 | ||
1428 | pp = XFS_ALLOC_PTR_ADDR(new, 1, cur); | |
1429 | INT_SET(pp[0], ARCH_CONVERT, lbno); | |
1430 | INT_SET(pp[1], ARCH_CONVERT, rbno); | |
1431 | } | |
1432 | xfs_alloc_log_ptrs(cur, nbp, 1, 2); | |
1433 | /* | |
1434 | * Fix up the cursor. | |
1435 | */ | |
1436 | xfs_btree_setbuf(cur, cur->bc_nlevels, nbp); | |
1437 | cur->bc_ptrs[cur->bc_nlevels] = nptr; | |
1438 | cur->bc_nlevels++; | |
1439 | *stat = 1; | |
1440 | return 0; | |
1441 | } | |
1442 | ||
1443 | /* | |
1444 | * Move 1 record right from cur/level if possible. | |
1445 | * Update cur to reflect the new path. | |
1446 | */ | |
1447 | STATIC int /* error */ | |
1448 | xfs_alloc_rshift( | |
1449 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1450 | int level, /* level to shift record on */ | |
1451 | int *stat) /* success/failure */ | |
1452 | { | |
1453 | int error; /* error return value */ | |
1454 | int i; /* loop index */ | |
1455 | xfs_alloc_key_t key; /* key value for leaf level upward */ | |
1456 | xfs_buf_t *lbp; /* buffer for left (current) block */ | |
1457 | xfs_alloc_block_t *left; /* left (current) btree block */ | |
1458 | xfs_buf_t *rbp; /* buffer for right neighbor block */ | |
1459 | xfs_alloc_block_t *right; /* right neighbor btree block */ | |
1460 | xfs_alloc_key_t *rkp; /* key pointer for right block */ | |
1461 | xfs_btree_cur_t *tcur; /* temporary cursor */ | |
1462 | ||
1463 | /* | |
1464 | * Set up variables for this block as "left". | |
1465 | */ | |
1466 | lbp = cur->bc_bufs[level]; | |
1467 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | |
1468 | #ifdef DEBUG | |
1469 | if ((error = xfs_btree_check_sblock(cur, left, level, lbp))) | |
1470 | return error; | |
1471 | #endif | |
1472 | /* | |
1473 | * If we've got no right sibling then we can't shift an entry right. | |
1474 | */ | |
1475 | if (INT_GET(left->bb_rightsib, ARCH_CONVERT) == NULLAGBLOCK) { | |
1476 | *stat = 0; | |
1477 | return 0; | |
1478 | } | |
1479 | /* | |
1480 | * If the cursor entry is the one that would be moved, don't | |
1481 | * do it... it's too complicated. | |
1482 | */ | |
1483 | if (cur->bc_ptrs[level] >= INT_GET(left->bb_numrecs, ARCH_CONVERT)) { | |
1484 | *stat = 0; | |
1485 | return 0; | |
1486 | } | |
1487 | /* | |
1488 | * Set up the right neighbor as "right". | |
1489 | */ | |
1490 | if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp, | |
1491 | cur->bc_private.a.agno, INT_GET(left->bb_rightsib, ARCH_CONVERT), 0, &rbp, | |
1492 | XFS_ALLOC_BTREE_REF))) | |
1493 | return error; | |
1494 | right = XFS_BUF_TO_ALLOC_BLOCK(rbp); | |
1495 | if ((error = xfs_btree_check_sblock(cur, right, level, rbp))) | |
1496 | return error; | |
1497 | /* | |
1498 | * If it's full, it can't take another entry. | |
1499 | */ | |
1500 | if (INT_GET(right->bb_numrecs, ARCH_CONVERT) == XFS_ALLOC_BLOCK_MAXRECS(level, cur)) { | |
1501 | *stat = 0; | |
1502 | return 0; | |
1503 | } | |
1504 | /* | |
1505 | * Make a hole at the start of the right neighbor block, then | |
1506 | * copy the last left block entry to the hole. | |
1507 | */ | |
1508 | if (level > 0) { | |
1509 | xfs_alloc_key_t *lkp; /* key pointer for left block */ | |
1510 | xfs_alloc_ptr_t *lpp; /* address pointer for left block */ | |
1511 | xfs_alloc_ptr_t *rpp; /* address pointer for right block */ | |
1512 | ||
1513 | lkp = XFS_ALLOC_KEY_ADDR(left, INT_GET(left->bb_numrecs, ARCH_CONVERT), cur); | |
1514 | lpp = XFS_ALLOC_PTR_ADDR(left, INT_GET(left->bb_numrecs, ARCH_CONVERT), cur); | |
1515 | rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur); | |
1516 | rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur); | |
1517 | #ifdef DEBUG | |
1518 | for (i = INT_GET(right->bb_numrecs, ARCH_CONVERT) - 1; i >= 0; i--) { | |
1519 | if ((error = xfs_btree_check_sptr(cur, INT_GET(rpp[i], ARCH_CONVERT), level))) | |
1520 | return error; | |
1521 | } | |
1522 | #endif | |
1523 | memmove(rkp + 1, rkp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rkp)); | |
1524 | memmove(rpp + 1, rpp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rpp)); | |
1525 | #ifdef DEBUG | |
1526 | if ((error = xfs_btree_check_sptr(cur, INT_GET(*lpp, ARCH_CONVERT), level))) | |
1527 | return error; | |
1528 | #endif | |
1529 | *rkp = *lkp; /* INT_: copy */ | |
1530 | *rpp = *lpp; /* INT_: copy */ | |
1531 | xfs_alloc_log_keys(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT) + 1); | |
1532 | xfs_alloc_log_ptrs(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT) + 1); | |
1533 | xfs_btree_check_key(cur->bc_btnum, rkp, rkp + 1); | |
1534 | } else { | |
1535 | xfs_alloc_rec_t *lrp; /* record pointer for left block */ | |
1536 | xfs_alloc_rec_t *rrp; /* record pointer for right block */ | |
1537 | ||
1538 | lrp = XFS_ALLOC_REC_ADDR(left, INT_GET(left->bb_numrecs, ARCH_CONVERT), cur); | |
1539 | rrp = XFS_ALLOC_REC_ADDR(right, 1, cur); | |
1540 | memmove(rrp + 1, rrp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rrp)); | |
1541 | *rrp = *lrp; | |
1542 | xfs_alloc_log_recs(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT) + 1); | |
1543 | key.ar_startblock = rrp->ar_startblock; /* INT_: direct copy */ | |
1544 | key.ar_blockcount = rrp->ar_blockcount; /* INT_: direct copy */ | |
1545 | rkp = &key; | |
1546 | xfs_btree_check_rec(cur->bc_btnum, rrp, rrp + 1); | |
1547 | } | |
1548 | /* | |
1549 | * Decrement and log left's numrecs, bump and log right's numrecs. | |
1550 | */ | |
1551 | INT_MOD(left->bb_numrecs, ARCH_CONVERT, -1); | |
1552 | xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS); | |
1553 | INT_MOD(right->bb_numrecs, ARCH_CONVERT, +1); | |
1554 | xfs_alloc_log_block(cur->bc_tp, rbp, XFS_BB_NUMRECS); | |
1555 | /* | |
1556 | * Using a temporary cursor, update the parent key values of the | |
1557 | * block on the right. | |
1558 | */ | |
1559 | if ((error = xfs_btree_dup_cursor(cur, &tcur))) | |
1560 | return error; | |
1561 | i = xfs_btree_lastrec(tcur, level); | |
1562 | XFS_WANT_CORRUPTED_GOTO(i == 1, error0); | |
1563 | if ((error = xfs_alloc_increment(tcur, level, &i)) || | |
1564 | (error = xfs_alloc_updkey(tcur, rkp, level + 1))) | |
1565 | goto error0; | |
1566 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
1567 | *stat = 1; | |
1568 | return 0; | |
1569 | error0: | |
1570 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
1571 | return error; | |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * Split cur/level block in half. | |
1576 | * Return new block number and its first record (to be inserted into parent). | |
1577 | */ | |
1578 | STATIC int /* error */ | |
1579 | xfs_alloc_split( | |
1580 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1581 | int level, /* level to split */ | |
1582 | xfs_agblock_t *bnop, /* output: block number allocated */ | |
1583 | xfs_alloc_key_t *keyp, /* output: first key of new block */ | |
1584 | xfs_btree_cur_t **curp, /* output: new cursor */ | |
1585 | int *stat) /* success/failure */ | |
1586 | { | |
1587 | int error; /* error return value */ | |
1588 | int i; /* loop index/record number */ | |
1589 | xfs_agblock_t lbno; /* left (current) block number */ | |
1590 | xfs_buf_t *lbp; /* buffer for left block */ | |
1591 | xfs_alloc_block_t *left; /* left (current) btree block */ | |
1592 | xfs_agblock_t rbno; /* right (new) block number */ | |
1593 | xfs_buf_t *rbp; /* buffer for right block */ | |
1594 | xfs_alloc_block_t *right; /* right (new) btree block */ | |
1595 | ||
1596 | /* | |
1597 | * Allocate the new block from the freelist. | |
1598 | * If we can't do it, we're toast. Give up. | |
1599 | */ | |
1600 | if ((error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp, | |
1601 | &rbno))) | |
1602 | return error; | |
1603 | if (rbno == NULLAGBLOCK) { | |
1604 | *stat = 0; | |
1605 | return 0; | |
1606 | } | |
1607 | xfs_trans_agbtree_delta(cur->bc_tp, 1); | |
1608 | rbp = xfs_btree_get_bufs(cur->bc_mp, cur->bc_tp, cur->bc_private.a.agno, | |
1609 | rbno, 0); | |
1610 | /* | |
1611 | * Set up the new block as "right". | |
1612 | */ | |
1613 | right = XFS_BUF_TO_ALLOC_BLOCK(rbp); | |
1614 | /* | |
1615 | * "Left" is the current (according to the cursor) block. | |
1616 | */ | |
1617 | lbp = cur->bc_bufs[level]; | |
1618 | left = XFS_BUF_TO_ALLOC_BLOCK(lbp); | |
1619 | #ifdef DEBUG | |
1620 | if ((error = xfs_btree_check_sblock(cur, left, level, lbp))) | |
1621 | return error; | |
1622 | #endif | |
1623 | /* | |
1624 | * Fill in the btree header for the new block. | |
1625 | */ | |
1626 | INT_SET(right->bb_magic, ARCH_CONVERT, xfs_magics[cur->bc_btnum]); | |
1627 | right->bb_level = left->bb_level; /* INT_: direct copy */ | |
1628 | INT_SET(right->bb_numrecs, ARCH_CONVERT, (__uint16_t)(INT_GET(left->bb_numrecs, ARCH_CONVERT) / 2)); | |
1629 | /* | |
1630 | * Make sure that if there's an odd number of entries now, that | |
1631 | * each new block will have the same number of entries. | |
1632 | */ | |
1633 | if ((INT_GET(left->bb_numrecs, ARCH_CONVERT) & 1) && | |
1634 | cur->bc_ptrs[level] <= INT_GET(right->bb_numrecs, ARCH_CONVERT) + 1) | |
1635 | INT_MOD(right->bb_numrecs, ARCH_CONVERT, +1); | |
1636 | i = INT_GET(left->bb_numrecs, ARCH_CONVERT) - INT_GET(right->bb_numrecs, ARCH_CONVERT) + 1; | |
1637 | /* | |
1638 | * For non-leaf blocks, copy keys and addresses over to the new block. | |
1639 | */ | |
1640 | if (level > 0) { | |
1641 | xfs_alloc_key_t *lkp; /* left btree key pointer */ | |
1642 | xfs_alloc_ptr_t *lpp; /* left btree address pointer */ | |
1643 | xfs_alloc_key_t *rkp; /* right btree key pointer */ | |
1644 | xfs_alloc_ptr_t *rpp; /* right btree address pointer */ | |
1645 | ||
1646 | lkp = XFS_ALLOC_KEY_ADDR(left, i, cur); | |
1647 | lpp = XFS_ALLOC_PTR_ADDR(left, i, cur); | |
1648 | rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur); | |
1649 | rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur); | |
1650 | #ifdef DEBUG | |
1651 | for (i = 0; i < INT_GET(right->bb_numrecs, ARCH_CONVERT); i++) { | |
1652 | if ((error = xfs_btree_check_sptr(cur, INT_GET(lpp[i], ARCH_CONVERT), level))) | |
1653 | return error; | |
1654 | } | |
1655 | #endif | |
1656 | memcpy(rkp, lkp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rkp)); /* INT_: copy */ | |
1657 | memcpy(rpp, lpp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rpp)); /* INT_: copy */ | |
1658 | xfs_alloc_log_keys(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
1659 | xfs_alloc_log_ptrs(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
1660 | *keyp = *rkp; | |
1661 | } | |
1662 | /* | |
1663 | * For leaf blocks, copy records over to the new block. | |
1664 | */ | |
1665 | else { | |
1666 | xfs_alloc_rec_t *lrp; /* left btree record pointer */ | |
1667 | xfs_alloc_rec_t *rrp; /* right btree record pointer */ | |
1668 | ||
1669 | lrp = XFS_ALLOC_REC_ADDR(left, i, cur); | |
1670 | rrp = XFS_ALLOC_REC_ADDR(right, 1, cur); | |
1671 | memcpy(rrp, lrp, INT_GET(right->bb_numrecs, ARCH_CONVERT) * sizeof(*rrp)); | |
1672 | xfs_alloc_log_recs(cur, rbp, 1, INT_GET(right->bb_numrecs, ARCH_CONVERT)); | |
1673 | keyp->ar_startblock = rrp->ar_startblock; /* INT_: direct copy */ | |
1674 | keyp->ar_blockcount = rrp->ar_blockcount; /* INT_: direct copy */ | |
1675 | } | |
1676 | /* | |
1677 | * Find the left block number by looking in the buffer. | |
1678 | * Adjust numrecs, sibling pointers. | |
1679 | */ | |
1680 | lbno = XFS_DADDR_TO_AGBNO(cur->bc_mp, XFS_BUF_ADDR(lbp)); | |
1681 | INT_MOD(left->bb_numrecs, ARCH_CONVERT, -(INT_GET(right->bb_numrecs, ARCH_CONVERT))); | |
1682 | right->bb_rightsib = left->bb_rightsib; /* INT_: direct copy */ | |
1683 | INT_SET(left->bb_rightsib, ARCH_CONVERT, rbno); | |
1684 | INT_SET(right->bb_leftsib, ARCH_CONVERT, lbno); | |
1685 | xfs_alloc_log_block(cur->bc_tp, rbp, XFS_BB_ALL_BITS); | |
1686 | xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB); | |
1687 | /* | |
1688 | * If there's a block to the new block's right, make that block | |
1689 | * point back to right instead of to left. | |
1690 | */ | |
1691 | if (INT_GET(right->bb_rightsib, ARCH_CONVERT) != NULLAGBLOCK) { | |
1692 | xfs_alloc_block_t *rrblock; /* rr btree block */ | |
1693 | xfs_buf_t *rrbp; /* buffer for rrblock */ | |
1694 | ||
1695 | if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp, | |
1696 | cur->bc_private.a.agno, INT_GET(right->bb_rightsib, ARCH_CONVERT), 0, | |
1697 | &rrbp, XFS_ALLOC_BTREE_REF))) | |
1698 | return error; | |
1699 | rrblock = XFS_BUF_TO_ALLOC_BLOCK(rrbp); | |
1700 | if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp))) | |
1701 | return error; | |
1702 | INT_SET(rrblock->bb_leftsib, ARCH_CONVERT, rbno); | |
1703 | xfs_alloc_log_block(cur->bc_tp, rrbp, XFS_BB_LEFTSIB); | |
1704 | } | |
1705 | /* | |
1706 | * If the cursor is really in the right block, move it there. | |
1707 | * If it's just pointing past the last entry in left, then we'll | |
1708 | * insert there, so don't change anything in that case. | |
1709 | */ | |
1710 | if (cur->bc_ptrs[level] > INT_GET(left->bb_numrecs, ARCH_CONVERT) + 1) { | |
1711 | xfs_btree_setbuf(cur, level, rbp); | |
1712 | cur->bc_ptrs[level] -= INT_GET(left->bb_numrecs, ARCH_CONVERT); | |
1713 | } | |
1714 | /* | |
1715 | * If there are more levels, we'll need another cursor which refers to | |
1716 | * the right block, no matter where this cursor was. | |
1717 | */ | |
1718 | if (level + 1 < cur->bc_nlevels) { | |
1719 | if ((error = xfs_btree_dup_cursor(cur, curp))) | |
1720 | return error; | |
1721 | (*curp)->bc_ptrs[level + 1]++; | |
1722 | } | |
1723 | *bnop = rbno; | |
1724 | *stat = 1; | |
1725 | return 0; | |
1726 | } | |
1727 | ||
1728 | /* | |
1729 | * Update keys at all levels from here to the root along the cursor's path. | |
1730 | */ | |
1731 | STATIC int /* error */ | |
1732 | xfs_alloc_updkey( | |
1733 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1734 | xfs_alloc_key_t *keyp, /* new key value to update to */ | |
1735 | int level) /* starting level for update */ | |
1736 | { | |
1737 | int ptr; /* index of key in block */ | |
1738 | ||
1739 | /* | |
1740 | * Go up the tree from this level toward the root. | |
1741 | * At each level, update the key value to the value input. | |
1742 | * Stop when we reach a level where the cursor isn't pointing | |
1743 | * at the first entry in the block. | |
1744 | */ | |
1745 | for (ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) { | |
1746 | xfs_alloc_block_t *block; /* btree block */ | |
1747 | xfs_buf_t *bp; /* buffer for block */ | |
1748 | #ifdef DEBUG | |
1749 | int error; /* error return value */ | |
1750 | #endif | |
1751 | xfs_alloc_key_t *kp; /* ptr to btree block keys */ | |
1752 | ||
1753 | bp = cur->bc_bufs[level]; | |
1754 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
1755 | #ifdef DEBUG | |
1756 | if ((error = xfs_btree_check_sblock(cur, block, level, bp))) | |
1757 | return error; | |
1758 | #endif | |
1759 | ptr = cur->bc_ptrs[level]; | |
1760 | kp = XFS_ALLOC_KEY_ADDR(block, ptr, cur); | |
1761 | *kp = *keyp; | |
1762 | xfs_alloc_log_keys(cur, bp, ptr, ptr); | |
1763 | } | |
1764 | return 0; | |
1765 | } | |
1766 | ||
1767 | /* | |
1768 | * Externally visible routines. | |
1769 | */ | |
1770 | ||
1771 | /* | |
1772 | * Decrement cursor by one record at the level. | |
1773 | * For nonzero levels the leaf-ward information is untouched. | |
1774 | */ | |
1775 | int /* error */ | |
1776 | xfs_alloc_decrement( | |
1777 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1778 | int level, /* level in btree, 0 is leaf */ | |
1779 | int *stat) /* success/failure */ | |
1780 | { | |
1781 | xfs_alloc_block_t *block; /* btree block */ | |
1782 | int error; /* error return value */ | |
1783 | int lev; /* btree level */ | |
1784 | ||
1785 | ASSERT(level < cur->bc_nlevels); | |
1786 | /* | |
1787 | * Read-ahead to the left at this level. | |
1788 | */ | |
1789 | xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA); | |
1790 | /* | |
1791 | * Decrement the ptr at this level. If we're still in the block | |
1792 | * then we're done. | |
1793 | */ | |
1794 | if (--cur->bc_ptrs[level] > 0) { | |
1795 | *stat = 1; | |
1796 | return 0; | |
1797 | } | |
1798 | /* | |
1799 | * Get a pointer to the btree block. | |
1800 | */ | |
1801 | block = XFS_BUF_TO_ALLOC_BLOCK(cur->bc_bufs[level]); | |
1802 | #ifdef DEBUG | |
1803 | if ((error = xfs_btree_check_sblock(cur, block, level, | |
1804 | cur->bc_bufs[level]))) | |
1805 | return error; | |
1806 | #endif | |
1807 | /* | |
1808 | * If we just went off the left edge of the tree, return failure. | |
1809 | */ | |
1810 | if (INT_GET(block->bb_leftsib, ARCH_CONVERT) == NULLAGBLOCK) { | |
1811 | *stat = 0; | |
1812 | return 0; | |
1813 | } | |
1814 | /* | |
1815 | * March up the tree decrementing pointers. | |
1816 | * Stop when we don't go off the left edge of a block. | |
1817 | */ | |
1818 | for (lev = level + 1; lev < cur->bc_nlevels; lev++) { | |
1819 | if (--cur->bc_ptrs[lev] > 0) | |
1820 | break; | |
1821 | /* | |
1822 | * Read-ahead the left block, we're going to read it | |
1823 | * in the next loop. | |
1824 | */ | |
1825 | xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA); | |
1826 | } | |
1827 | /* | |
1828 | * If we went off the root then we are seriously confused. | |
1829 | */ | |
1830 | ASSERT(lev < cur->bc_nlevels); | |
1831 | /* | |
1832 | * Now walk back down the tree, fixing up the cursor's buffer | |
1833 | * pointers and key numbers. | |
1834 | */ | |
1835 | for (block = XFS_BUF_TO_ALLOC_BLOCK(cur->bc_bufs[lev]); lev > level; ) { | |
1836 | xfs_agblock_t agbno; /* block number of btree block */ | |
1837 | xfs_buf_t *bp; /* buffer pointer for block */ | |
1838 | ||
1839 | agbno = INT_GET(*XFS_ALLOC_PTR_ADDR(block, cur->bc_ptrs[lev], cur), ARCH_CONVERT); | |
1840 | if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp, | |
1841 | cur->bc_private.a.agno, agbno, 0, &bp, | |
1842 | XFS_ALLOC_BTREE_REF))) | |
1843 | return error; | |
1844 | lev--; | |
1845 | xfs_btree_setbuf(cur, lev, bp); | |
1846 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
1847 | if ((error = xfs_btree_check_sblock(cur, block, lev, bp))) | |
1848 | return error; | |
1849 | cur->bc_ptrs[lev] = INT_GET(block->bb_numrecs, ARCH_CONVERT); | |
1850 | } | |
1851 | *stat = 1; | |
1852 | return 0; | |
1853 | } | |
1854 | ||
1855 | /* | |
1856 | * Delete the record pointed to by cur. | |
1857 | * The cursor refers to the place where the record was (could be inserted) | |
1858 | * when the operation returns. | |
1859 | */ | |
1860 | int /* error */ | |
1861 | xfs_alloc_delete( | |
1862 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1863 | int *stat) /* success/failure */ | |
1864 | { | |
1865 | int error; /* error return value */ | |
1866 | int i; /* result code */ | |
1867 | int level; /* btree level */ | |
1868 | ||
1869 | /* | |
1870 | * Go up the tree, starting at leaf level. | |
1871 | * If 2 is returned then a join was done; go to the next level. | |
1872 | * Otherwise we are done. | |
1873 | */ | |
1874 | for (level = 0, i = 2; i == 2; level++) { | |
1875 | if ((error = xfs_alloc_delrec(cur, level, &i))) | |
1876 | return error; | |
1877 | } | |
1878 | if (i == 0) { | |
1879 | for (level = 1; level < cur->bc_nlevels; level++) { | |
1880 | if (cur->bc_ptrs[level] == 0) { | |
1881 | if ((error = xfs_alloc_decrement(cur, level, &i))) | |
1882 | return error; | |
1883 | break; | |
1884 | } | |
1885 | } | |
1886 | } | |
1887 | *stat = i; | |
1888 | return 0; | |
1889 | } | |
1890 | ||
1891 | /* | |
1892 | * Get the data from the pointed-to record. | |
1893 | */ | |
1894 | int /* error */ | |
1895 | xfs_alloc_get_rec( | |
1896 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1897 | xfs_agblock_t *bno, /* output: starting block of extent */ | |
1898 | xfs_extlen_t *len, /* output: length of extent */ | |
1899 | int *stat) /* output: success/failure */ | |
1900 | { | |
1901 | xfs_alloc_block_t *block; /* btree block */ | |
1902 | #ifdef DEBUG | |
1903 | int error; /* error return value */ | |
1904 | #endif | |
1905 | int ptr; /* record number */ | |
1906 | ||
1907 | ptr = cur->bc_ptrs[0]; | |
1908 | block = XFS_BUF_TO_ALLOC_BLOCK(cur->bc_bufs[0]); | |
1909 | #ifdef DEBUG | |
1910 | if ((error = xfs_btree_check_sblock(cur, block, 0, cur->bc_bufs[0]))) | |
1911 | return error; | |
1912 | #endif | |
1913 | /* | |
1914 | * Off the right end or left end, return failure. | |
1915 | */ | |
1916 | if (ptr > INT_GET(block->bb_numrecs, ARCH_CONVERT) || ptr <= 0) { | |
1917 | *stat = 0; | |
1918 | return 0; | |
1919 | } | |
1920 | /* | |
1921 | * Point to the record and extract its data. | |
1922 | */ | |
1923 | { | |
1924 | xfs_alloc_rec_t *rec; /* record data */ | |
1925 | ||
1926 | rec = XFS_ALLOC_REC_ADDR(block, ptr, cur); | |
1927 | *bno = INT_GET(rec->ar_startblock, ARCH_CONVERT); | |
1928 | *len = INT_GET(rec->ar_blockcount, ARCH_CONVERT); | |
1929 | } | |
1930 | *stat = 1; | |
1931 | return 0; | |
1932 | } | |
1933 | ||
1934 | /* | |
1935 | * Increment cursor by one record at the level. | |
1936 | * For nonzero levels the leaf-ward information is untouched. | |
1937 | */ | |
1938 | int /* error */ | |
1939 | xfs_alloc_increment( | |
1940 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1941 | int level, /* level in btree, 0 is leaf */ | |
1942 | int *stat) /* success/failure */ | |
1943 | { | |
1944 | xfs_alloc_block_t *block; /* btree block */ | |
1945 | xfs_buf_t *bp; /* tree block buffer */ | |
1946 | int error; /* error return value */ | |
1947 | int lev; /* btree level */ | |
1948 | ||
1949 | ASSERT(level < cur->bc_nlevels); | |
1950 | /* | |
1951 | * Read-ahead to the right at this level. | |
1952 | */ | |
1953 | xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA); | |
1954 | /* | |
1955 | * Get a pointer to the btree block. | |
1956 | */ | |
1957 | bp = cur->bc_bufs[level]; | |
1958 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
1959 | #ifdef DEBUG | |
1960 | if ((error = xfs_btree_check_sblock(cur, block, level, bp))) | |
1961 | return error; | |
1962 | #endif | |
1963 | /* | |
1964 | * Increment the ptr at this level. If we're still in the block | |
1965 | * then we're done. | |
1966 | */ | |
1967 | if (++cur->bc_ptrs[level] <= INT_GET(block->bb_numrecs, ARCH_CONVERT)) { | |
1968 | *stat = 1; | |
1969 | return 0; | |
1970 | } | |
1971 | /* | |
1972 | * If we just went off the right edge of the tree, return failure. | |
1973 | */ | |
1974 | if (INT_GET(block->bb_rightsib, ARCH_CONVERT) == NULLAGBLOCK) { | |
1975 | *stat = 0; | |
1976 | return 0; | |
1977 | } | |
1978 | /* | |
1979 | * March up the tree incrementing pointers. | |
1980 | * Stop when we don't go off the right edge of a block. | |
1981 | */ | |
1982 | for (lev = level + 1; lev < cur->bc_nlevels; lev++) { | |
1983 | bp = cur->bc_bufs[lev]; | |
1984 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
1985 | #ifdef DEBUG | |
1986 | if ((error = xfs_btree_check_sblock(cur, block, lev, bp))) | |
1987 | return error; | |
1988 | #endif | |
1989 | if (++cur->bc_ptrs[lev] <= INT_GET(block->bb_numrecs, ARCH_CONVERT)) | |
1990 | break; | |
1991 | /* | |
1992 | * Read-ahead the right block, we're going to read it | |
1993 | * in the next loop. | |
1994 | */ | |
1995 | xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA); | |
1996 | } | |
1997 | /* | |
1998 | * If we went off the root then we are seriously confused. | |
1999 | */ | |
2000 | ASSERT(lev < cur->bc_nlevels); | |
2001 | /* | |
2002 | * Now walk back down the tree, fixing up the cursor's buffer | |
2003 | * pointers and key numbers. | |
2004 | */ | |
2005 | for (bp = cur->bc_bufs[lev], block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
2006 | lev > level; ) { | |
2007 | xfs_agblock_t agbno; /* block number of btree block */ | |
2008 | ||
2009 | agbno = INT_GET(*XFS_ALLOC_PTR_ADDR(block, cur->bc_ptrs[lev], cur), ARCH_CONVERT); | |
2010 | if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp, | |
2011 | cur->bc_private.a.agno, agbno, 0, &bp, | |
2012 | XFS_ALLOC_BTREE_REF))) | |
2013 | return error; | |
2014 | lev--; | |
2015 | xfs_btree_setbuf(cur, lev, bp); | |
2016 | block = XFS_BUF_TO_ALLOC_BLOCK(bp); | |
2017 | if ((error = xfs_btree_check_sblock(cur, block, lev, bp))) | |
2018 | return error; | |
2019 | cur->bc_ptrs[lev] = 1; | |
2020 | } | |
2021 | *stat = 1; | |
2022 | return 0; | |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * Insert the current record at the point referenced by cur. | |
2027 | * The cursor may be inconsistent on return if splits have been done. | |
2028 | */ | |
2029 | int /* error */ | |
2030 | xfs_alloc_insert( | |
2031 | xfs_btree_cur_t *cur, /* btree cursor */ | |
2032 | int *stat) /* success/failure */ | |
2033 | { | |
2034 | int error; /* error return value */ | |
2035 | int i; /* result value, 0 for failure */ | |
2036 | int level; /* current level number in btree */ | |
2037 | xfs_agblock_t nbno; /* new block number (split result) */ | |
2038 | xfs_btree_cur_t *ncur; /* new cursor (split result) */ | |
2039 | xfs_alloc_rec_t nrec; /* record being inserted this level */ | |
2040 | xfs_btree_cur_t *pcur; /* previous level's cursor */ | |
2041 | ||
2042 | level = 0; | |
2043 | nbno = NULLAGBLOCK; | |
2044 | INT_SET(nrec.ar_startblock, ARCH_CONVERT, cur->bc_rec.a.ar_startblock); | |
2045 | INT_SET(nrec.ar_blockcount, ARCH_CONVERT, cur->bc_rec.a.ar_blockcount); | |
2046 | ncur = (xfs_btree_cur_t *)0; | |
2047 | pcur = cur; | |
2048 | /* | |
2049 | * Loop going up the tree, starting at the leaf level. | |
2050 | * Stop when we don't get a split block, that must mean that | |
2051 | * the insert is finished with this level. | |
2052 | */ | |
2053 | do { | |
2054 | /* | |
2055 | * Insert nrec/nbno into this level of the tree. | |
2056 | * Note if we fail, nbno will be null. | |
2057 | */ | |
2058 | if ((error = xfs_alloc_insrec(pcur, level++, &nbno, &nrec, &ncur, | |
2059 | &i))) { | |
2060 | if (pcur != cur) | |
2061 | xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR); | |
2062 | return error; | |
2063 | } | |
2064 | /* | |
2065 | * See if the cursor we just used is trash. | |
2066 | * Can't trash the caller's cursor, but otherwise we should | |
2067 | * if ncur is a new cursor or we're about to be done. | |
2068 | */ | |
2069 | if (pcur != cur && (ncur || nbno == NULLAGBLOCK)) { | |
2070 | cur->bc_nlevels = pcur->bc_nlevels; | |
2071 | xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR); | |
2072 | } | |
2073 | /* | |
2074 | * If we got a new cursor, switch to it. | |
2075 | */ | |
2076 | if (ncur) { | |
2077 | pcur = ncur; | |
2078 | ncur = (xfs_btree_cur_t *)0; | |
2079 | } | |
2080 | } while (nbno != NULLAGBLOCK); | |
2081 | *stat = i; | |
2082 | return 0; | |
2083 | } | |
2084 | ||
2085 | /* | |
2086 | * Lookup the record equal to [bno, len] in the btree given by cur. | |
2087 | */ | |
2088 | int /* error */ | |
2089 | xfs_alloc_lookup_eq( | |
2090 | xfs_btree_cur_t *cur, /* btree cursor */ | |
2091 | xfs_agblock_t bno, /* starting block of extent */ | |
2092 | xfs_extlen_t len, /* length of extent */ | |
2093 | int *stat) /* success/failure */ | |
2094 | { | |
2095 | cur->bc_rec.a.ar_startblock = bno; | |
2096 | cur->bc_rec.a.ar_blockcount = len; | |
2097 | return xfs_alloc_lookup(cur, XFS_LOOKUP_EQ, stat); | |
2098 | } | |
2099 | ||
2100 | /* | |
2101 | * Lookup the first record greater than or equal to [bno, len] | |
2102 | * in the btree given by cur. | |
2103 | */ | |
2104 | int /* error */ | |
2105 | xfs_alloc_lookup_ge( | |
2106 | xfs_btree_cur_t *cur, /* btree cursor */ | |
2107 | xfs_agblock_t bno, /* starting block of extent */ | |
2108 | xfs_extlen_t len, /* length of extent */ | |
2109 | int *stat) /* success/failure */ | |
2110 | { | |
2111 | cur->bc_rec.a.ar_startblock = bno; | |
2112 | cur->bc_rec.a.ar_blockcount = len; | |
2113 | return xfs_alloc_lookup(cur, XFS_LOOKUP_GE, stat); | |
2114 | } | |
2115 | ||
2116 | /* | |
2117 | * Lookup the first record less than or equal to [bno, len] | |
2118 | * in the btree given by cur. | |
2119 | */ | |
2120 | int /* error */ | |
2121 | xfs_alloc_lookup_le( | |
2122 | xfs_btree_cur_t *cur, /* btree cursor */ | |
2123 | xfs_agblock_t bno, /* starting block of extent */ | |
2124 | xfs_extlen_t len, /* length of extent */ | |
2125 | int *stat) /* success/failure */ | |
2126 | { | |
2127 | cur->bc_rec.a.ar_startblock = bno; | |
2128 | cur->bc_rec.a.ar_blockcount = len; | |
2129 | return xfs_alloc_lookup(cur, XFS_LOOKUP_LE, stat); | |
2130 | } | |
2131 | ||
2132 | /* | |
2133 | * Update the record referred to by cur, to the value given by [bno, len]. | |
2134 | * This either works (return 0) or gets an EFSCORRUPTED error. | |
2135 | */ | |
2136 | int /* error */ | |
2137 | xfs_alloc_update( | |
2138 | xfs_btree_cur_t *cur, /* btree cursor */ | |
2139 | xfs_agblock_t bno, /* starting block of extent */ | |
2140 | xfs_extlen_t len) /* length of extent */ | |
2141 | { | |
2142 | xfs_alloc_block_t *block; /* btree block to update */ | |
2143 | int error; /* error return value */ | |
2144 | int ptr; /* current record number (updating) */ | |
2145 | ||
2146 | ASSERT(len > 0); | |
2147 | /* | |
2148 | * Pick up the a.g. freelist struct and the current block. | |
2149 | */ | |
2150 | block = XFS_BUF_TO_ALLOC_BLOCK(cur->bc_bufs[0]); | |
2151 | #ifdef DEBUG | |
2152 | if ((error = xfs_btree_check_sblock(cur, block, 0, cur->bc_bufs[0]))) | |
2153 | return error; | |
2154 | #endif | |
2155 | /* | |
2156 | * Get the address of the rec to be updated. | |
2157 | */ | |
2158 | ptr = cur->bc_ptrs[0]; | |
2159 | { | |
2160 | xfs_alloc_rec_t *rp; /* pointer to updated record */ | |
2161 | ||
2162 | rp = XFS_ALLOC_REC_ADDR(block, ptr, cur); | |
2163 | /* | |
2164 | * Fill in the new contents and log them. | |
2165 | */ | |
2166 | INT_SET(rp->ar_startblock, ARCH_CONVERT, bno); | |
2167 | INT_SET(rp->ar_blockcount, ARCH_CONVERT, len); | |
2168 | xfs_alloc_log_recs(cur, cur->bc_bufs[0], ptr, ptr); | |
2169 | } | |
2170 | /* | |
2171 | * If it's the by-size btree and it's the last leaf block and | |
2172 | * it's the last record... then update the size of the longest | |
2173 | * extent in the a.g., which we cache in the a.g. freelist header. | |
2174 | */ | |
2175 | if (cur->bc_btnum == XFS_BTNUM_CNT && | |
2176 | INT_GET(block->bb_rightsib, ARCH_CONVERT) == NULLAGBLOCK && | |
2177 | ptr == INT_GET(block->bb_numrecs, ARCH_CONVERT)) { | |
2178 | xfs_agf_t *agf; /* a.g. freespace header */ | |
2179 | xfs_agnumber_t seqno; | |
2180 | ||
2181 | agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); | |
2182 | seqno = INT_GET(agf->agf_seqno, ARCH_CONVERT); | |
2183 | cur->bc_mp->m_perag[seqno].pagf_longest = len; | |
2184 | INT_SET(agf->agf_longest, ARCH_CONVERT, len); | |
2185 | xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, | |
2186 | XFS_AGF_LONGEST); | |
2187 | } | |
2188 | /* | |
2189 | * Updating first record in leaf. Pass new key value up to our parent. | |
2190 | */ | |
2191 | if (ptr == 1) { | |
2192 | xfs_alloc_key_t key; /* key containing [bno, len] */ | |
2193 | ||
2194 | INT_SET(key.ar_startblock, ARCH_CONVERT, bno); | |
2195 | INT_SET(key.ar_blockcount, ARCH_CONVERT, len); | |
2196 | if ((error = xfs_alloc_updkey(cur, &key, 1))) | |
2197 | return error; | |
2198 | } | |
2199 | return 0; | |
2200 | } |