]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
da23017d NS |
2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
5000d01d | 4 | * |
da23017d 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 | |
2bd0ea18 | 7 | * published by the Free Software Foundation. |
5000d01d | 8 | * |
da23017d 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. | |
5000d01d | 13 | * |
da23017d 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 | |
2bd0ea18 | 17 | */ |
9c799827 | 18 | #include "libxfs_priv.h" |
b626fb59 DC |
19 | #include "xfs_fs.h" |
20 | #include "xfs_shared.h" | |
21 | #include "xfs_format.h" | |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
24 | #include "xfs_bit.h" | |
25 | #include "xfs_mount.h" | |
f944d3d0 | 26 | #include "xfs_defer.h" |
b626fb59 DC |
27 | #include "xfs_inode.h" |
28 | #include "xfs_trans.h" | |
29 | #include "xfs_btree.h" | |
56d3fc2b | 30 | #include "xfs_errortag.h" |
b626fb59 DC |
31 | #include "xfs_trace.h" |
32 | #include "xfs_cksum.h" | |
33 | #include "xfs_alloc.h" | |
2bd0ea18 NS |
34 | |
35 | /* | |
36 | * Cursor allocation zone. | |
37 | */ | |
5e656dbb | 38 | kmem_zone_t *xfs_btree_cur_zone; |
2bd0ea18 NS |
39 | |
40 | /* | |
41 | * Btree magic numbers. | |
42 | */ | |
4a492e72 | 43 | static const uint32_t xfs_magics[2][XFS_BTNUM_MAX] = { |
e37838e5 | 44 | { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, 0, XFS_BMAP_MAGIC, XFS_IBT_MAGIC, |
88ce0792 | 45 | XFS_FIBT_MAGIC, 0 }, |
e37838e5 | 46 | { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC, XFS_RMAP_CRC_MAGIC, |
88ce0792 DW |
47 | XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC, |
48 | XFS_REFC_CRC_MAGIC } | |
2bd0ea18 | 49 | }; |
4c6b3277 | 50 | |
4a492e72 | 51 | uint32_t |
4c6b3277 ES |
52 | xfs_btree_magic( |
53 | int crc, | |
54 | xfs_btnum_t btnum) | |
55 | { | |
4a492e72 | 56 | uint32_t magic = xfs_magics[crc][btnum]; |
4c6b3277 ES |
57 | |
58 | /* Ensure we asked for crc for crc-only magics. */ | |
59 | ASSERT(magic != 0); | |
60 | return magic; | |
61 | } | |
2bd0ea18 | 62 | |
b47beff8 DW |
63 | /* |
64 | * Check a long btree block header. Return the address of the failing check, | |
65 | * or NULL if everything is ok. | |
66 | */ | |
67 | xfs_failaddr_t | |
68 | __xfs_btree_check_lblock( | |
69 | struct xfs_btree_cur *cur, | |
70 | struct xfs_btree_block *block, | |
71 | int level, | |
72 | struct xfs_buf *bp) | |
2bd0ea18 | 73 | { |
b47beff8 | 74 | struct xfs_mount *mp = cur->bc_mp; |
4c6b3277 | 75 | xfs_btnum_t btnum = cur->bc_btnum; |
b47beff8 | 76 | int crc = xfs_sb_version_hascrc(&mp->m_sb); |
5dfa5cd2 | 77 | |
4c6b3277 | 78 | if (crc) { |
b47beff8 DW |
79 | if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid)) |
80 | return __this_address; | |
81 | if (block->bb_u.l.bb_blkno != | |
82 | cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL)) | |
83 | return __this_address; | |
84 | if (block->bb_u.l.bb_pad != cpu_to_be32(0)) | |
85 | return __this_address; | |
5dfa5cd2 DC |
86 | } |
87 | ||
b47beff8 DW |
88 | if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum)) |
89 | return __this_address; | |
90 | if (be16_to_cpu(block->bb_level) != level) | |
91 | return __this_address; | |
92 | if (be16_to_cpu(block->bb_numrecs) > | |
93 | cur->bc_ops->get_maxrecs(cur, level)) | |
94 | return __this_address; | |
95 | if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) && | |
96 | !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_leftsib), | |
97 | level + 1)) | |
98 | return __this_address; | |
99 | if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) && | |
100 | !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_rightsib), | |
101 | level + 1)) | |
102 | return __this_address; | |
103 | ||
104 | return NULL; | |
105 | } | |
106 | ||
107 | /* Check a long btree block header. */ | |
246486c1 | 108 | static int |
b47beff8 DW |
109 | xfs_btree_check_lblock( |
110 | struct xfs_btree_cur *cur, | |
111 | struct xfs_btree_block *block, | |
112 | int level, | |
113 | struct xfs_buf *bp) | |
114 | { | |
115 | struct xfs_mount *mp = cur->bc_mp; | |
116 | xfs_failaddr_t fa; | |
117 | ||
118 | fa = __xfs_btree_check_lblock(cur, block, level, bp); | |
119 | if (unlikely(XFS_TEST_ERROR(fa != NULL, mp, | |
e2a190dd | 120 | XFS_ERRTAG_BTREE_CHECK_LBLOCK))) { |
2bd0ea18 | 121 | if (bp) |
56b2de80 | 122 | trace_xfs_btree_corrupt(bp, _RET_IP_); |
5dfa5cd2 | 123 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); |
12b53197 | 124 | return -EFSCORRUPTED; |
2bd0ea18 NS |
125 | } |
126 | return 0; | |
127 | } | |
128 | ||
b47beff8 DW |
129 | /* |
130 | * Check a short btree block header. Return the address of the failing check, | |
131 | * or NULL if everything is ok. | |
132 | */ | |
133 | xfs_failaddr_t | |
134 | __xfs_btree_check_sblock( | |
135 | struct xfs_btree_cur *cur, | |
136 | struct xfs_btree_block *block, | |
137 | int level, | |
138 | struct xfs_buf *bp) | |
2bd0ea18 | 139 | { |
b47beff8 | 140 | struct xfs_mount *mp = cur->bc_mp; |
4c6b3277 | 141 | xfs_btnum_t btnum = cur->bc_btnum; |
b47beff8 | 142 | int crc = xfs_sb_version_hascrc(&mp->m_sb); |
5dfa5cd2 | 143 | |
4c6b3277 | 144 | if (crc) { |
b47beff8 DW |
145 | if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid)) |
146 | return __this_address; | |
147 | if (block->bb_u.s.bb_blkno != | |
148 | cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL)) | |
149 | return __this_address; | |
5dfa5cd2 DC |
150 | } |
151 | ||
b47beff8 DW |
152 | if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum)) |
153 | return __this_address; | |
154 | if (be16_to_cpu(block->bb_level) != level) | |
155 | return __this_address; | |
156 | if (be16_to_cpu(block->bb_numrecs) > | |
157 | cur->bc_ops->get_maxrecs(cur, level)) | |
158 | return __this_address; | |
159 | if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) && | |
160 | !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_leftsib), | |
161 | level + 1)) | |
162 | return __this_address; | |
163 | if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) && | |
164 | !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_rightsib), | |
165 | level + 1)) | |
166 | return __this_address; | |
167 | ||
168 | return NULL; | |
169 | } | |
170 | ||
171 | /* Check a short btree block header. */ | |
172 | STATIC int | |
173 | xfs_btree_check_sblock( | |
174 | struct xfs_btree_cur *cur, | |
175 | struct xfs_btree_block *block, | |
176 | int level, | |
177 | struct xfs_buf *bp) | |
178 | { | |
179 | struct xfs_mount *mp = cur->bc_mp; | |
180 | xfs_failaddr_t fa; | |
181 | ||
182 | fa = __xfs_btree_check_sblock(cur, block, level, bp); | |
183 | if (unlikely(XFS_TEST_ERROR(fa != NULL, mp, | |
e2a190dd | 184 | XFS_ERRTAG_BTREE_CHECK_SBLOCK))) { |
2bd0ea18 | 185 | if (bp) |
56b2de80 | 186 | trace_xfs_btree_corrupt(bp, _RET_IP_); |
5dfa5cd2 | 187 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); |
12b53197 | 188 | return -EFSCORRUPTED; |
2bd0ea18 NS |
189 | } |
190 | return 0; | |
191 | } | |
192 | ||
193 | /* | |
b194c7d8 BN |
194 | * Debug routine: check that block header is ok. |
195 | */ | |
196 | int | |
197 | xfs_btree_check_block( | |
198 | struct xfs_btree_cur *cur, /* btree cursor */ | |
199 | struct xfs_btree_block *block, /* generic btree block pointer */ | |
200 | int level, /* level of the btree block */ | |
201 | struct xfs_buf *bp) /* buffer containing block, if any */ | |
202 | { | |
b3563c19 BN |
203 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) |
204 | return xfs_btree_check_lblock(cur, block, level, bp); | |
205 | else | |
206 | return xfs_btree_check_sblock(cur, block, level, bp); | |
b194c7d8 BN |
207 | } |
208 | ||
d5c546df DW |
209 | /* Check that this long pointer is valid and points within the fs. */ |
210 | bool | |
b194c7d8 | 211 | xfs_btree_check_lptr( |
d5c546df DW |
212 | struct xfs_btree_cur *cur, |
213 | xfs_fsblock_t fsbno, | |
214 | int level) | |
b194c7d8 | 215 | { |
d5c546df DW |
216 | if (level <= 0) |
217 | return false; | |
218 | return xfs_verify_fsbno(cur->bc_mp, fsbno); | |
b194c7d8 BN |
219 | } |
220 | ||
d5c546df DW |
221 | /* Check that this short pointer is valid and points within the AG. */ |
222 | bool | |
2bd0ea18 | 223 | xfs_btree_check_sptr( |
d5c546df DW |
224 | struct xfs_btree_cur *cur, |
225 | xfs_agblock_t agbno, | |
226 | int level) | |
2bd0ea18 | 227 | { |
d5c546df DW |
228 | if (level <= 0) |
229 | return false; | |
230 | return xfs_verify_agbno(cur->bc_mp, cur->bc_private.a.agno, agbno); | |
2bd0ea18 NS |
231 | } |
232 | ||
d5c546df | 233 | #ifdef DEBUG |
b194c7d8 | 234 | /* |
d5c546df DW |
235 | * Check that a given (indexed) btree pointer at a certain level of a |
236 | * btree is valid and doesn't point past where it should. | |
b194c7d8 | 237 | */ |
246486c1 | 238 | static int |
b194c7d8 | 239 | xfs_btree_check_ptr( |
d5c546df DW |
240 | struct xfs_btree_cur *cur, |
241 | union xfs_btree_ptr *ptr, | |
242 | int index, | |
243 | int level) | |
b194c7d8 BN |
244 | { |
245 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) { | |
d5c546df DW |
246 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, |
247 | xfs_btree_check_lptr(cur, | |
248 | be64_to_cpu((&ptr->l)[index]), level)); | |
b194c7d8 | 249 | } else { |
d5c546df DW |
250 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, |
251 | xfs_btree_check_sptr(cur, | |
252 | be32_to_cpu((&ptr->s)[index]), level)); | |
b194c7d8 | 253 | } |
d5c546df DW |
254 | |
255 | return 0; | |
b194c7d8 | 256 | } |
870d4cbc | 257 | #endif |
b194c7d8 | 258 | |
5dfa5cd2 DC |
259 | /* |
260 | * Calculate CRC on the whole btree block and stuff it into the | |
261 | * long-form btree header. | |
262 | * | |
263 | * Prior to calculting the CRC, pull the LSN out of the buffer log item and put | |
eab16f4c | 264 | * it into the buffer so recovery knows what the last modification was that made |
5dfa5cd2 DC |
265 | * it to disk. |
266 | */ | |
267 | void | |
268 | xfs_btree_lblock_calc_crc( | |
269 | struct xfs_buf *bp) | |
270 | { | |
271 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); | |
37d086ca | 272 | struct xfs_buf_log_item *bip = bp->b_log_item; |
5dfa5cd2 DC |
273 | |
274 | if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb)) | |
275 | return; | |
276 | if (bip) | |
277 | block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
43b5aeed | 278 | xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF); |
5dfa5cd2 DC |
279 | } |
280 | ||
281 | bool | |
282 | xfs_btree_lblock_verify_crc( | |
283 | struct xfs_buf *bp) | |
284 | { | |
a65d8d29 BF |
285 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); |
286 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
287 | ||
4ca4eb89 | 288 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
a65d8d29 BF |
289 | if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn))) |
290 | return false; | |
d21ca64d | 291 | return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF); |
a65d8d29 | 292 | } |
d21ca64d | 293 | |
5dfa5cd2 DC |
294 | return true; |
295 | } | |
296 | ||
297 | /* | |
298 | * Calculate CRC on the whole btree block and stuff it into the | |
299 | * short-form btree header. | |
300 | * | |
301 | * Prior to calculting the CRC, pull the LSN out of the buffer log item and put | |
eab16f4c | 302 | * it into the buffer so recovery knows what the last modification was that made |
5dfa5cd2 DC |
303 | * it to disk. |
304 | */ | |
305 | void | |
306 | xfs_btree_sblock_calc_crc( | |
307 | struct xfs_buf *bp) | |
308 | { | |
309 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); | |
37d086ca | 310 | struct xfs_buf_log_item *bip = bp->b_log_item; |
5dfa5cd2 DC |
311 | |
312 | if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb)) | |
313 | return; | |
314 | if (bip) | |
315 | block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
43b5aeed | 316 | xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF); |
5dfa5cd2 DC |
317 | } |
318 | ||
319 | bool | |
320 | xfs_btree_sblock_verify_crc( | |
321 | struct xfs_buf *bp) | |
322 | { | |
a65d8d29 BF |
323 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); |
324 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
325 | ||
4ca4eb89 | 326 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
a65d8d29 | 327 | if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn))) |
bc01119d | 328 | return __this_address; |
d21ca64d | 329 | return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF); |
a65d8d29 | 330 | } |
d21ca64d | 331 | |
5dfa5cd2 DC |
332 | return true; |
333 | } | |
334 | ||
c261f8c0 CH |
335 | static int |
336 | xfs_btree_free_block( | |
337 | struct xfs_btree_cur *cur, | |
338 | struct xfs_buf *bp) | |
339 | { | |
340 | int error; | |
341 | ||
342 | error = cur->bc_ops->free_block(cur, bp); | |
08caf14f CH |
343 | if (!error) { |
344 | xfs_trans_binval(cur->bc_tp, bp); | |
c261f8c0 | 345 | XFS_BTREE_STATS_INC(cur, free); |
08caf14f | 346 | } |
c261f8c0 CH |
347 | return error; |
348 | } | |
349 | ||
2bd0ea18 NS |
350 | /* |
351 | * Delete the btree cursor. | |
352 | */ | |
353 | void | |
354 | xfs_btree_del_cursor( | |
dfc130f3 | 355 | xfs_btree_cur_t *cur, /* btree cursor */ |
2bd0ea18 NS |
356 | int error) /* del because of error */ |
357 | { | |
358 | int i; /* btree level */ | |
359 | ||
360 | /* | |
361 | * Clear the buffer pointers, and release the buffers. | |
362 | * If we're doing this in the face of an error, we | |
363 | * need to make sure to inspect all of the entries | |
364 | * in the bc_bufs array for buffers to be unlocked. | |
365 | * This is because some of the btree code works from | |
366 | * level n down to 0, and if we get an error along | |
367 | * the way we won't have initialized all the entries | |
368 | * down to 0. | |
369 | */ | |
370 | for (i = 0; i < cur->bc_nlevels; i++) { | |
371 | if (cur->bc_bufs[i]) | |
56b2de80 | 372 | xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]); |
2bd0ea18 NS |
373 | else if (!error) |
374 | break; | |
375 | } | |
376 | /* | |
5000d01d | 377 | * Can't free a bmap cursor without having dealt with the |
2bd0ea18 NS |
378 | * allocated indirect blocks' accounting. |
379 | */ | |
380 | ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP || | |
381 | cur->bc_private.b.allocated == 0); | |
382 | /* | |
383 | * Free the cursor. | |
384 | */ | |
385 | kmem_zone_free(xfs_btree_cur_zone, cur); | |
386 | } | |
387 | ||
388 | /* | |
389 | * Duplicate the btree cursor. | |
390 | * Allocate a new one, copy the record, re-get the buffers. | |
391 | */ | |
392 | int /* error */ | |
393 | xfs_btree_dup_cursor( | |
dfc130f3 RC |
394 | xfs_btree_cur_t *cur, /* input cursor */ |
395 | xfs_btree_cur_t **ncur) /* output cursor */ | |
2bd0ea18 NS |
396 | { |
397 | xfs_buf_t *bp; /* btree block's buffer pointer */ | |
5000d01d | 398 | int error; /* error return value */ |
2bd0ea18 NS |
399 | int i; /* level number of btree block */ |
400 | xfs_mount_t *mp; /* mount structure for filesystem */ | |
dfc130f3 | 401 | xfs_btree_cur_t *new; /* new cursor value */ |
2bd0ea18 NS |
402 | xfs_trans_t *tp; /* transaction pointer, can be NULL */ |
403 | ||
404 | tp = cur->bc_tp; | |
405 | mp = cur->bc_mp; | |
b194c7d8 | 406 | |
2bd0ea18 NS |
407 | /* |
408 | * Allocate a new cursor like the old one. | |
409 | */ | |
b194c7d8 BN |
410 | new = cur->bc_ops->dup_cursor(cur); |
411 | ||
2bd0ea18 NS |
412 | /* |
413 | * Copy the record currently in the cursor. | |
414 | */ | |
415 | new->bc_rec = cur->bc_rec; | |
b194c7d8 | 416 | |
2bd0ea18 NS |
417 | /* |
418 | * For each level current, re-get the buffer and copy the ptr value. | |
419 | */ | |
420 | for (i = 0; i < new->bc_nlevels; i++) { | |
421 | new->bc_ptrs[i] = cur->bc_ptrs[i]; | |
422 | new->bc_ra[i] = cur->bc_ra[i]; | |
a2ceac1f DC |
423 | bp = cur->bc_bufs[i]; |
424 | if (bp) { | |
425 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, | |
426 | XFS_BUF_ADDR(bp), mp->m_bsize, | |
427 | 0, &bp, | |
428 | cur->bc_ops->buf_ops); | |
429 | if (error) { | |
2bd0ea18 NS |
430 | xfs_btree_del_cursor(new, error); |
431 | *ncur = NULL; | |
432 | return error; | |
433 | } | |
5dfa5cd2 DC |
434 | } |
435 | new->bc_bufs[i] = bp; | |
2bd0ea18 | 436 | } |
2bd0ea18 NS |
437 | *ncur = new; |
438 | return 0; | |
439 | } | |
440 | ||
b194c7d8 BN |
441 | /* |
442 | * XFS btree block layout and addressing: | |
443 | * | |
444 | * There are two types of blocks in the btree: leaf and non-leaf blocks. | |
445 | * | |
446 | * The leaf record start with a header then followed by records containing | |
447 | * the values. A non-leaf block also starts with the same header, and | |
448 | * then first contains lookup keys followed by an equal number of pointers | |
449 | * to the btree blocks at the previous level. | |
450 | * | |
451 | * +--------+-------+-------+-------+-------+-------+-------+ | |
452 | * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N | | |
453 | * +--------+-------+-------+-------+-------+-------+-------+ | |
454 | * | |
455 | * +--------+-------+-------+-------+-------+-------+-------+ | |
456 | * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N | | |
457 | * +--------+-------+-------+-------+-------+-------+-------+ | |
458 | * | |
459 | * The header is called struct xfs_btree_block for reasons better left unknown | |
460 | * and comes in different versions for short (32bit) and long (64bit) block | |
461 | * pointers. The record and key structures are defined by the btree instances | |
462 | * and opaque to the btree core. The block pointers are simple disk endian | |
463 | * integers, available in a short (32bit) and long (64bit) variant. | |
464 | * | |
465 | * The helpers below calculate the offset of a given record, key or pointer | |
466 | * into a btree block (xfs_btree_*_offset) or return a pointer to the given | |
467 | * record, key or pointer (xfs_btree_*_addr). Note that all addressing | |
468 | * inside the btree block is done using indices starting at one, not zero! | |
13e831e0 DW |
469 | * |
470 | * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing | |
471 | * overlapping intervals. In such a tree, records are still sorted lowest to | |
472 | * highest and indexed by the smallest key value that refers to the record. | |
473 | * However, nodes are different: each pointer has two associated keys -- one | |
474 | * indexing the lowest key available in the block(s) below (the same behavior | |
475 | * as the key in a regular btree) and another indexing the highest key | |
476 | * available in the block(s) below. Because records are /not/ sorted by the | |
477 | * highest key, all leaf block updates require us to compute the highest key | |
478 | * that matches any record in the leaf and to recursively update the high keys | |
479 | * in the nodes going further up in the tree, if necessary. Nodes look like | |
480 | * this: | |
481 | * | |
482 | * +--------+-----+-----+-----+-----+-----+-------+-------+-----+ | |
483 | * Non-Leaf: | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... | | |
484 | * +--------+-----+-----+-----+-----+-----+-------+-------+-----+ | |
485 | * | |
486 | * To perform an interval query on an overlapped tree, perform the usual | |
487 | * depth-first search and use the low and high keys to decide if we can skip | |
488 | * that particular node. If a leaf node is reached, return the records that | |
489 | * intersect the interval. Note that an interval query may return numerous | |
490 | * entries. For a non-overlapped tree, simply search for the record associated | |
491 | * with the lowest key and iterate forward until a non-matching record is | |
492 | * found. Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by | |
493 | * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in | |
494 | * more detail. | |
495 | * | |
496 | * Why do we care about overlapping intervals? Let's say you have a bunch of | |
497 | * reverse mapping records on a reflink filesystem: | |
498 | * | |
499 | * 1: +- file A startblock B offset C length D -----------+ | |
500 | * 2: +- file E startblock F offset G length H --------------+ | |
501 | * 3: +- file I startblock F offset J length K --+ | |
502 | * 4: +- file L... --+ | |
503 | * | |
504 | * Now say we want to map block (B+D) into file A at offset (C+D). Ideally, | |
505 | * we'd simply increment the length of record 1. But how do we find the record | |
506 | * that ends at (B+D-1) (i.e. record 1)? A LE lookup of (B+D-1) would return | |
507 | * record 3 because the keys are ordered first by startblock. An interval | |
508 | * query would return records 1 and 2 because they both overlap (B+D-1), and | |
509 | * from that we can pick out record 1 as the appropriate left neighbor. | |
510 | * | |
511 | * In the non-overlapped case you can do a LE lookup and decrement the cursor | |
512 | * because a record's interval must end before the next record. | |
b194c7d8 BN |
513 | */ |
514 | ||
515 | /* | |
516 | * Return size of the btree block header for this btree instance. | |
517 | */ | |
518 | static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur) | |
519 | { | |
e0607266 DC |
520 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) { |
521 | if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) | |
522 | return XFS_BTREE_LBLOCK_CRC_LEN; | |
523 | return XFS_BTREE_LBLOCK_LEN; | |
524 | } | |
5dfa5cd2 | 525 | if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) |
e0607266 DC |
526 | return XFS_BTREE_SBLOCK_CRC_LEN; |
527 | return XFS_BTREE_SBLOCK_LEN; | |
b194c7d8 BN |
528 | } |
529 | ||
530 | /* | |
531 | * Return size of btree block pointers for this btree instance. | |
532 | */ | |
533 | static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur) | |
534 | { | |
535 | return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ? | |
536 | sizeof(__be64) : sizeof(__be32); | |
537 | } | |
538 | ||
539 | /* | |
540 | * Calculate offset of the n-th record in a btree block. | |
541 | */ | |
542 | STATIC size_t | |
543 | xfs_btree_rec_offset( | |
544 | struct xfs_btree_cur *cur, | |
545 | int n) | |
546 | { | |
547 | return xfs_btree_block_len(cur) + | |
548 | (n - 1) * cur->bc_ops->rec_len; | |
549 | } | |
550 | ||
551 | /* | |
552 | * Calculate offset of the n-th key in a btree block. | |
553 | */ | |
554 | STATIC size_t | |
555 | xfs_btree_key_offset( | |
556 | struct xfs_btree_cur *cur, | |
557 | int n) | |
558 | { | |
559 | return xfs_btree_block_len(cur) + | |
560 | (n - 1) * cur->bc_ops->key_len; | |
561 | } | |
562 | ||
13e831e0 DW |
563 | /* |
564 | * Calculate offset of the n-th high key in a btree block. | |
565 | */ | |
566 | STATIC size_t | |
567 | xfs_btree_high_key_offset( | |
568 | struct xfs_btree_cur *cur, | |
569 | int n) | |
570 | { | |
571 | return xfs_btree_block_len(cur) + | |
572 | (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2); | |
573 | } | |
574 | ||
b194c7d8 BN |
575 | /* |
576 | * Calculate offset of the n-th block pointer in a btree block. | |
577 | */ | |
578 | STATIC size_t | |
579 | xfs_btree_ptr_offset( | |
580 | struct xfs_btree_cur *cur, | |
581 | int n, | |
582 | int level) | |
583 | { | |
584 | return xfs_btree_block_len(cur) + | |
585 | cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len + | |
586 | (n - 1) * xfs_btree_ptr_len(cur); | |
587 | } | |
588 | ||
589 | /* | |
590 | * Return a pointer to the n-th record in the btree block. | |
591 | */ | |
50bb67d6 | 592 | union xfs_btree_rec * |
b194c7d8 BN |
593 | xfs_btree_rec_addr( |
594 | struct xfs_btree_cur *cur, | |
595 | int n, | |
596 | struct xfs_btree_block *block) | |
597 | { | |
598 | return (union xfs_btree_rec *) | |
599 | ((char *)block + xfs_btree_rec_offset(cur, n)); | |
600 | } | |
601 | ||
602 | /* | |
603 | * Return a pointer to the n-th key in the btree block. | |
604 | */ | |
50bb67d6 | 605 | union xfs_btree_key * |
b194c7d8 BN |
606 | xfs_btree_key_addr( |
607 | struct xfs_btree_cur *cur, | |
608 | int n, | |
609 | struct xfs_btree_block *block) | |
610 | { | |
611 | return (union xfs_btree_key *) | |
612 | ((char *)block + xfs_btree_key_offset(cur, n)); | |
613 | } | |
614 | ||
13e831e0 DW |
615 | /* |
616 | * Return a pointer to the n-th high key in the btree block. | |
617 | */ | |
50bb67d6 | 618 | union xfs_btree_key * |
13e831e0 DW |
619 | xfs_btree_high_key_addr( |
620 | struct xfs_btree_cur *cur, | |
621 | int n, | |
622 | struct xfs_btree_block *block) | |
623 | { | |
624 | return (union xfs_btree_key *) | |
625 | ((char *)block + xfs_btree_high_key_offset(cur, n)); | |
626 | } | |
627 | ||
b194c7d8 BN |
628 | /* |
629 | * Return a pointer to the n-th block pointer in the btree block. | |
630 | */ | |
50bb67d6 | 631 | union xfs_btree_ptr * |
b194c7d8 BN |
632 | xfs_btree_ptr_addr( |
633 | struct xfs_btree_cur *cur, | |
634 | int n, | |
635 | struct xfs_btree_block *block) | |
636 | { | |
637 | int level = xfs_btree_get_level(block); | |
638 | ||
639 | ASSERT(block->bb_level != 0); | |
640 | ||
641 | return (union xfs_btree_ptr *) | |
642 | ((char *)block + xfs_btree_ptr_offset(cur, n, level)); | |
643 | } | |
644 | ||
645 | /* | |
10851b18 | 646 | * Get the root block which is stored in the inode. |
b194c7d8 BN |
647 | * |
648 | * For now this btree implementation assumes the btree root is always | |
649 | * stored in the if_broot field of an inode fork. | |
650 | */ | |
651 | STATIC struct xfs_btree_block * | |
652 | xfs_btree_get_iroot( | |
dcaff8ac | 653 | struct xfs_btree_cur *cur) |
b194c7d8 | 654 | { |
dcaff8ac | 655 | struct xfs_ifork *ifp; |
b194c7d8 | 656 | |
dcaff8ac KN |
657 | ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork); |
658 | return (struct xfs_btree_block *)ifp->if_broot; | |
b194c7d8 BN |
659 | } |
660 | ||
5000d01d | 661 | /* |
2bd0ea18 | 662 | * Retrieve the block pointer from the cursor at the given level. |
b194c7d8 | 663 | * This may be an inode btree root or from a buffer. |
2bd0ea18 | 664 | */ |
50bb67d6 | 665 | struct xfs_btree_block * /* generic btree block pointer */ |
2bd0ea18 | 666 | xfs_btree_get_block( |
b194c7d8 | 667 | struct xfs_btree_cur *cur, /* btree cursor */ |
2bd0ea18 | 668 | int level, /* level in btree */ |
b194c7d8 BN |
669 | struct xfs_buf **bpp) /* buffer containing the block */ |
670 | { | |
671 | if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) && | |
672 | (level == cur->bc_nlevels - 1)) { | |
673 | *bpp = NULL; | |
674 | return xfs_btree_get_iroot(cur); | |
2bd0ea18 | 675 | } |
b194c7d8 BN |
676 | |
677 | *bpp = cur->bc_bufs[level]; | |
678 | return XFS_BUF_TO_BLOCK(*bpp); | |
2bd0ea18 NS |
679 | } |
680 | ||
681 | /* | |
682 | * Get a buffer for the block, return it with no data read. | |
683 | * Long-form addressing. | |
684 | */ | |
685 | xfs_buf_t * /* buffer for fsbno */ | |
686 | xfs_btree_get_bufl( | |
687 | xfs_mount_t *mp, /* file system mount point */ | |
688 | xfs_trans_t *tp, /* transaction pointer */ | |
689 | xfs_fsblock_t fsbno, /* file system block number */ | |
690 | uint lock) /* lock flags for get_buf */ | |
691 | { | |
2bd0ea18 NS |
692 | xfs_daddr_t d; /* real disk block address */ |
693 | ||
694 | ASSERT(fsbno != NULLFSBLOCK); | |
695 | d = XFS_FSB_TO_DADDR(mp, fsbno); | |
ff105f75 | 696 | return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock); |
2bd0ea18 NS |
697 | } |
698 | ||
699 | /* | |
700 | * Get a buffer for the block, return it with no data read. | |
701 | * Short-form addressing. | |
702 | */ | |
703 | xfs_buf_t * /* buffer for agno/agbno */ | |
704 | xfs_btree_get_bufs( | |
705 | xfs_mount_t *mp, /* file system mount point */ | |
706 | xfs_trans_t *tp, /* transaction pointer */ | |
707 | xfs_agnumber_t agno, /* allocation group number */ | |
708 | xfs_agblock_t agbno, /* allocation group block number */ | |
709 | uint lock) /* lock flags for get_buf */ | |
710 | { | |
2bd0ea18 NS |
711 | xfs_daddr_t d; /* real disk block address */ |
712 | ||
713 | ASSERT(agno != NULLAGNUMBER); | |
714 | ASSERT(agbno != NULLAGBLOCK); | |
715 | d = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
ff105f75 | 716 | return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock); |
2bd0ea18 NS |
717 | } |
718 | ||
2bd0ea18 NS |
719 | /* |
720 | * Check for the cursor referring to the last block at the given level. | |
721 | */ | |
722 | int /* 1=is last block, 0=not last block */ | |
723 | xfs_btree_islastblock( | |
724 | xfs_btree_cur_t *cur, /* btree cursor */ | |
725 | int level) /* level to check */ | |
726 | { | |
b3563c19 | 727 | struct xfs_btree_block *block; /* generic btree block pointer */ |
2bd0ea18 NS |
728 | xfs_buf_t *bp; /* buffer containing block */ |
729 | ||
730 | block = xfs_btree_get_block(cur, level, &bp); | |
731 | xfs_btree_check_block(cur, block, level, bp); | |
b194c7d8 | 732 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) |
5a35bf2c | 733 | return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK); |
2bd0ea18 | 734 | else |
a2ceac1f | 735 | return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK); |
2bd0ea18 NS |
736 | } |
737 | ||
5e656dbb BN |
738 | /* |
739 | * Change the cursor to point to the first record at the given level. | |
740 | * Other levels are unaffected. | |
741 | */ | |
b194c7d8 | 742 | STATIC int /* success=1, failure=0 */ |
5e656dbb BN |
743 | xfs_btree_firstrec( |
744 | xfs_btree_cur_t *cur, /* btree cursor */ | |
745 | int level) /* level to change */ | |
746 | { | |
b3563c19 | 747 | struct xfs_btree_block *block; /* generic btree block pointer */ |
5e656dbb BN |
748 | xfs_buf_t *bp; /* buffer containing block */ |
749 | ||
750 | /* | |
751 | * Get the block pointer for this level. | |
752 | */ | |
753 | block = xfs_btree_get_block(cur, level, &bp); | |
c3d6d07a DW |
754 | if (xfs_btree_check_block(cur, block, level, bp)) |
755 | return 0; | |
5e656dbb BN |
756 | /* |
757 | * It's empty, there is no such record. | |
758 | */ | |
b194c7d8 | 759 | if (!block->bb_numrecs) |
5e656dbb BN |
760 | return 0; |
761 | /* | |
762 | * Set the ptr value to 1, that's the first record/key. | |
763 | */ | |
764 | cur->bc_ptrs[level] = 1; | |
765 | return 1; | |
766 | } | |
767 | ||
2bd0ea18 NS |
768 | /* |
769 | * Change the cursor to point to the last record in the current block | |
dfc130f3 | 770 | * at the given level. Other levels are unaffected. |
2bd0ea18 | 771 | */ |
b194c7d8 | 772 | STATIC int /* success=1, failure=0 */ |
2bd0ea18 NS |
773 | xfs_btree_lastrec( |
774 | xfs_btree_cur_t *cur, /* btree cursor */ | |
775 | int level) /* level to change */ | |
776 | { | |
b3563c19 | 777 | struct xfs_btree_block *block; /* generic btree block pointer */ |
2bd0ea18 NS |
778 | xfs_buf_t *bp; /* buffer containing block */ |
779 | ||
780 | /* | |
781 | * Get the block pointer for this level. | |
782 | */ | |
783 | block = xfs_btree_get_block(cur, level, &bp); | |
c3d6d07a DW |
784 | if (xfs_btree_check_block(cur, block, level, bp)) |
785 | return 0; | |
2bd0ea18 NS |
786 | /* |
787 | * It's empty, there is no such record. | |
788 | */ | |
b194c7d8 | 789 | if (!block->bb_numrecs) |
2bd0ea18 NS |
790 | return 0; |
791 | /* | |
792 | * Set the ptr value to numrecs, that's the last record/key. | |
793 | */ | |
b194c7d8 | 794 | cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs); |
2bd0ea18 NS |
795 | return 1; |
796 | } | |
797 | ||
798 | /* | |
799 | * Compute first and last byte offsets for the fields given. | |
800 | * Interprets the offsets table, which contains struct field offsets. | |
801 | */ | |
802 | void | |
803 | xfs_btree_offsets( | |
4a492e72 | 804 | int64_t fields, /* bitmask of fields */ |
2bd0ea18 NS |
805 | const short *offsets, /* table of field offsets */ |
806 | int nbits, /* number of bits to inspect */ | |
807 | int *first, /* output: first byte offset */ | |
808 | int *last) /* output: last byte offset */ | |
809 | { | |
810 | int i; /* current bit number */ | |
4a492e72 | 811 | int64_t imask; /* mask for current bit number */ |
2bd0ea18 NS |
812 | |
813 | ASSERT(fields != 0); | |
814 | /* | |
815 | * Find the lowest bit, so the first byte offset. | |
816 | */ | |
817 | for (i = 0, imask = 1LL; ; i++, imask <<= 1) { | |
818 | if (imask & fields) { | |
819 | *first = offsets[i]; | |
820 | break; | |
821 | } | |
822 | } | |
823 | /* | |
824 | * Find the highest bit, so the last byte offset. | |
825 | */ | |
826 | for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) { | |
827 | if (imask & fields) { | |
828 | *last = offsets[i + 1] - 1; | |
829 | break; | |
830 | } | |
831 | } | |
832 | } | |
833 | ||
834 | /* | |
835 | * Get a buffer for the block, return it read in. | |
836 | * Long-form addressing. | |
837 | */ | |
a2ceac1f | 838 | int |
2bd0ea18 | 839 | xfs_btree_read_bufl( |
a2ceac1f DC |
840 | struct xfs_mount *mp, /* file system mount point */ |
841 | struct xfs_trans *tp, /* transaction pointer */ | |
842 | xfs_fsblock_t fsbno, /* file system block number */ | |
843 | uint lock, /* lock flags for read_buf */ | |
844 | struct xfs_buf **bpp, /* buffer for fsbno */ | |
845 | int refval, /* ref count value for buffer */ | |
846 | const struct xfs_buf_ops *ops) | |
2bd0ea18 | 847 | { |
a2ceac1f | 848 | struct xfs_buf *bp; /* return value */ |
2bd0ea18 | 849 | xfs_daddr_t d; /* real disk block address */ |
a2ceac1f | 850 | int error; |
2bd0ea18 | 851 | |
ecdc52ff | 852 | if (!xfs_verify_fsbno(mp, fsbno)) |
e2376544 | 853 | return -EFSCORRUPTED; |
2bd0ea18 | 854 | d = XFS_FSB_TO_DADDR(mp, fsbno); |
a2ceac1f DC |
855 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d, |
856 | mp->m_bsize, lock, &bp, ops); | |
857 | if (error) | |
2bd0ea18 | 858 | return error; |
a2ceac1f DC |
859 | if (bp) |
860 | xfs_buf_set_ref(bp, refval); | |
2bd0ea18 NS |
861 | *bpp = bp; |
862 | return 0; | |
863 | } | |
864 | ||
10851b18 DC |
865 | /* |
866 | * Read-ahead the block, don't wait for it, don't return a buffer. | |
867 | * Long-form addressing. | |
868 | */ | |
869 | /* ARGSUSED */ | |
870 | void | |
871 | xfs_btree_reada_bufl( | |
872 | struct xfs_mount *mp, /* file system mount point */ | |
873 | xfs_fsblock_t fsbno, /* file system block number */ | |
874 | xfs_extlen_t count, /* count of filesystem blocks */ | |
875 | const struct xfs_buf_ops *ops) | |
876 | { | |
877 | xfs_daddr_t d; | |
878 | ||
879 | ASSERT(fsbno != NULLFSBLOCK); | |
880 | d = XFS_FSB_TO_DADDR(mp, fsbno); | |
881 | xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops); | |
882 | } | |
883 | ||
884 | /* | |
885 | * Read-ahead the block, don't wait for it, don't return a buffer. | |
886 | * Short-form addressing. | |
887 | */ | |
888 | /* ARGSUSED */ | |
889 | void | |
890 | xfs_btree_reada_bufs( | |
891 | struct xfs_mount *mp, /* file system mount point */ | |
892 | xfs_agnumber_t agno, /* allocation group number */ | |
893 | xfs_agblock_t agbno, /* allocation group block number */ | |
894 | xfs_extlen_t count, /* count of filesystem blocks */ | |
895 | const struct xfs_buf_ops *ops) | |
896 | { | |
897 | xfs_daddr_t d; | |
898 | ||
899 | ASSERT(agno != NULLAGNUMBER); | |
900 | ASSERT(agbno != NULLAGBLOCK); | |
901 | d = XFS_AGB_TO_DADDR(mp, agno, agbno); | |
902 | xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops); | |
903 | } | |
904 | ||
b194c7d8 BN |
905 | STATIC int |
906 | xfs_btree_readahead_lblock( | |
907 | struct xfs_btree_cur *cur, | |
908 | int lr, | |
909 | struct xfs_btree_block *block) | |
2bd0ea18 | 910 | { |
2bd0ea18 | 911 | int rval = 0; |
5a35bf2c DC |
912 | xfs_fsblock_t left = be64_to_cpu(block->bb_u.l.bb_leftsib); |
913 | xfs_fsblock_t right = be64_to_cpu(block->bb_u.l.bb_rightsib); | |
2bd0ea18 | 914 | |
5a35bf2c | 915 | if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) { |
a2ceac1f DC |
916 | xfs_btree_reada_bufl(cur->bc_mp, left, 1, |
917 | cur->bc_ops->buf_ops); | |
b194c7d8 BN |
918 | rval++; |
919 | } | |
920 | ||
5a35bf2c | 921 | if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) { |
a2ceac1f DC |
922 | xfs_btree_reada_bufl(cur->bc_mp, right, 1, |
923 | cur->bc_ops->buf_ops); | |
b194c7d8 | 924 | rval++; |
2bd0ea18 | 925 | } |
b194c7d8 | 926 | |
2bd0ea18 NS |
927 | return rval; |
928 | } | |
929 | ||
b194c7d8 BN |
930 | STATIC int |
931 | xfs_btree_readahead_sblock( | |
932 | struct xfs_btree_cur *cur, | |
933 | int lr, | |
934 | struct xfs_btree_block *block) | |
2bd0ea18 | 935 | { |
b194c7d8 BN |
936 | int rval = 0; |
937 | xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib); | |
938 | xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib); | |
2bd0ea18 | 939 | |
b194c7d8 BN |
940 | |
941 | if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) { | |
942 | xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno, | |
a2ceac1f | 943 | left, 1, cur->bc_ops->buf_ops); |
b194c7d8 BN |
944 | rval++; |
945 | } | |
946 | ||
947 | if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) { | |
948 | xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno, | |
a2ceac1f | 949 | right, 1, cur->bc_ops->buf_ops); |
b194c7d8 BN |
950 | rval++; |
951 | } | |
952 | ||
953 | return rval; | |
954 | } | |
955 | ||
956 | /* | |
957 | * Read-ahead btree blocks, at the given level. | |
958 | * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA. | |
959 | */ | |
960 | STATIC int | |
961 | xfs_btree_readahead( | |
962 | struct xfs_btree_cur *cur, /* btree cursor */ | |
963 | int lev, /* level in btree */ | |
964 | int lr) /* left/right bits */ | |
965 | { | |
966 | struct xfs_btree_block *block; | |
967 | ||
968 | /* | |
969 | * No readahead needed if we are at the root level and the | |
970 | * btree root is stored in the inode. | |
971 | */ | |
972 | if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) && | |
973 | (lev == cur->bc_nlevels - 1)) | |
974 | return 0; | |
975 | ||
976 | if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev]) | |
977 | return 0; | |
978 | ||
979 | cur->bc_ra[lev] |= lr; | |
980 | block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]); | |
981 | ||
982 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) | |
983 | return xfs_btree_readahead_lblock(cur, lr, block); | |
984 | return xfs_btree_readahead_sblock(cur, lr, block); | |
985 | } | |
986 | ||
9c6ebc42 DC |
987 | STATIC xfs_daddr_t |
988 | xfs_btree_ptr_to_daddr( | |
989 | struct xfs_btree_cur *cur, | |
990 | union xfs_btree_ptr *ptr) | |
991 | { | |
992 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) { | |
5a35bf2c | 993 | ASSERT(ptr->l != cpu_to_be64(NULLFSBLOCK)); |
9c6ebc42 DC |
994 | |
995 | return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l)); | |
996 | } else { | |
997 | ASSERT(cur->bc_private.a.agno != NULLAGNUMBER); | |
998 | ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK)); | |
999 | ||
1000 | return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno, | |
1001 | be32_to_cpu(ptr->s)); | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | /* | |
1006 | * Readahead @count btree blocks at the given @ptr location. | |
1007 | * | |
1008 | * We don't need to care about long or short form btrees here as we have a | |
1009 | * method of converting the ptr directly to a daddr available to us. | |
1010 | */ | |
1011 | STATIC void | |
1012 | xfs_btree_readahead_ptr( | |
1013 | struct xfs_btree_cur *cur, | |
1014 | union xfs_btree_ptr *ptr, | |
1015 | xfs_extlen_t count) | |
1016 | { | |
1017 | xfs_buf_readahead(cur->bc_mp->m_ddev_targp, | |
1018 | xfs_btree_ptr_to_daddr(cur, ptr), | |
1019 | cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops); | |
1020 | } | |
1021 | ||
b194c7d8 BN |
1022 | /* |
1023 | * Set the buffer for level "lev" in the cursor to bp, releasing | |
1024 | * any previous buffer. | |
1025 | */ | |
56b2de80 | 1026 | STATIC void |
b194c7d8 BN |
1027 | xfs_btree_setbuf( |
1028 | xfs_btree_cur_t *cur, /* btree cursor */ | |
1029 | int lev, /* level in btree */ | |
1030 | xfs_buf_t *bp) /* new buffer to set */ | |
1031 | { | |
b3563c19 | 1032 | struct xfs_btree_block *b; /* btree block */ |
b194c7d8 | 1033 | |
56b2de80 DC |
1034 | if (cur->bc_bufs[lev]) |
1035 | xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]); | |
2bd0ea18 NS |
1036 | cur->bc_bufs[lev] = bp; |
1037 | cur->bc_ra[lev] = 0; | |
56b2de80 | 1038 | |
2bd0ea18 | 1039 | b = XFS_BUF_TO_BLOCK(bp); |
b194c7d8 | 1040 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) { |
5a35bf2c | 1041 | if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK)) |
2bd0ea18 | 1042 | cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA; |
5a35bf2c | 1043 | if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK)) |
2bd0ea18 NS |
1044 | cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA; |
1045 | } else { | |
a2ceac1f | 1046 | if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK)) |
2bd0ea18 | 1047 | cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA; |
a2ceac1f | 1048 | if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK)) |
2bd0ea18 NS |
1049 | cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA; |
1050 | } | |
1051 | } | |
b194c7d8 | 1052 | |
15c4e6e4 | 1053 | bool |
b194c7d8 BN |
1054 | xfs_btree_ptr_is_null( |
1055 | struct xfs_btree_cur *cur, | |
1056 | union xfs_btree_ptr *ptr) | |
1057 | { | |
1058 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) | |
5a35bf2c | 1059 | return ptr->l == cpu_to_be64(NULLFSBLOCK); |
b194c7d8 | 1060 | else |
a2ceac1f | 1061 | return ptr->s == cpu_to_be32(NULLAGBLOCK); |
b194c7d8 BN |
1062 | } |
1063 | ||
1064 | STATIC void | |
1065 | xfs_btree_set_ptr_null( | |
1066 | struct xfs_btree_cur *cur, | |
1067 | union xfs_btree_ptr *ptr) | |
1068 | { | |
1069 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) | |
5a35bf2c | 1070 | ptr->l = cpu_to_be64(NULLFSBLOCK); |
b194c7d8 BN |
1071 | else |
1072 | ptr->s = cpu_to_be32(NULLAGBLOCK); | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * Get/set/init sibling pointers | |
1077 | */ | |
15c4e6e4 | 1078 | void |
b194c7d8 BN |
1079 | xfs_btree_get_sibling( |
1080 | struct xfs_btree_cur *cur, | |
1081 | struct xfs_btree_block *block, | |
1082 | union xfs_btree_ptr *ptr, | |
1083 | int lr) | |
1084 | { | |
1085 | ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB); | |
1086 | ||
1087 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) { | |
1088 | if (lr == XFS_BB_RIGHTSIB) | |
1089 | ptr->l = block->bb_u.l.bb_rightsib; | |
1090 | else | |
1091 | ptr->l = block->bb_u.l.bb_leftsib; | |
1092 | } else { | |
1093 | if (lr == XFS_BB_RIGHTSIB) | |
1094 | ptr->s = block->bb_u.s.bb_rightsib; | |
1095 | else | |
1096 | ptr->s = block->bb_u.s.bb_leftsib; | |
1097 | } | |
1098 | } | |
1099 | ||
1100 | STATIC void | |
1101 | xfs_btree_set_sibling( | |
1102 | struct xfs_btree_cur *cur, | |
1103 | struct xfs_btree_block *block, | |
1104 | union xfs_btree_ptr *ptr, | |
1105 | int lr) | |
1106 | { | |
1107 | ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB); | |
1108 | ||
1109 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) { | |
1110 | if (lr == XFS_BB_RIGHTSIB) | |
1111 | block->bb_u.l.bb_rightsib = ptr->l; | |
1112 | else | |
1113 | block->bb_u.l.bb_leftsib = ptr->l; | |
1114 | } else { | |
1115 | if (lr == XFS_BB_RIGHTSIB) | |
1116 | block->bb_u.s.bb_rightsib = ptr->s; | |
1117 | else | |
1118 | block->bb_u.s.bb_leftsib = ptr->s; | |
1119 | } | |
1120 | } | |
1121 | ||
5dfa5cd2 DC |
1122 | void |
1123 | xfs_btree_init_block_int( | |
1124 | struct xfs_mount *mp, | |
1125 | struct xfs_btree_block *buf, | |
1126 | xfs_daddr_t blkno, | |
e394a4b1 | 1127 | xfs_btnum_t btnum, |
5dfa5cd2 DC |
1128 | __u16 level, |
1129 | __u16 numrecs, | |
1130 | __u64 owner, | |
1131 | unsigned int flags) | |
1132 | { | |
f4241a08 | 1133 | int crc = xfs_sb_version_hascrc(&mp->m_sb); |
e394a4b1 | 1134 | __u32 magic = xfs_btree_magic(crc, btnum); |
f4241a08 | 1135 | |
5dfa5cd2 DC |
1136 | buf->bb_magic = cpu_to_be32(magic); |
1137 | buf->bb_level = cpu_to_be16(level); | |
1138 | buf->bb_numrecs = cpu_to_be16(numrecs); | |
1139 | ||
1140 | if (flags & XFS_BTREE_LONG_PTRS) { | |
5a35bf2c DC |
1141 | buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK); |
1142 | buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK); | |
f4241a08 | 1143 | if (crc) { |
5dfa5cd2 DC |
1144 | buf->bb_u.l.bb_blkno = cpu_to_be64(blkno); |
1145 | buf->bb_u.l.bb_owner = cpu_to_be64(owner); | |
9c4e12fb | 1146 | uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid); |
5dfa5cd2 | 1147 | buf->bb_u.l.bb_pad = 0; |
6f9ea829 | 1148 | buf->bb_u.l.bb_lsn = 0; |
5dfa5cd2 DC |
1149 | } |
1150 | } else { | |
1151 | /* owner is a 32 bit value on short blocks */ | |
1152 | __u32 __owner = (__u32)owner; | |
1153 | ||
1154 | buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK); | |
1155 | buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK); | |
f4241a08 | 1156 | if (crc) { |
5dfa5cd2 DC |
1157 | buf->bb_u.s.bb_blkno = cpu_to_be64(blkno); |
1158 | buf->bb_u.s.bb_owner = cpu_to_be32(__owner); | |
9c4e12fb | 1159 | uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid); |
6f9ea829 | 1160 | buf->bb_u.s.bb_lsn = 0; |
5dfa5cd2 DC |
1161 | } |
1162 | } | |
1163 | } | |
1164 | ||
a2ceac1f | 1165 | void |
b194c7d8 | 1166 | xfs_btree_init_block( |
a2ceac1f DC |
1167 | struct xfs_mount *mp, |
1168 | struct xfs_buf *bp, | |
e394a4b1 | 1169 | xfs_btnum_t btnum, |
a2ceac1f DC |
1170 | __u16 level, |
1171 | __u16 numrecs, | |
5dfa5cd2 | 1172 | __u64 owner, |
a2ceac1f | 1173 | unsigned int flags) |
b194c7d8 | 1174 | { |
5dfa5cd2 | 1175 | xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn, |
e394a4b1 | 1176 | btnum, level, numrecs, owner, flags); |
b194c7d8 BN |
1177 | } |
1178 | ||
a2ceac1f DC |
1179 | STATIC void |
1180 | xfs_btree_init_block_cur( | |
1181 | struct xfs_btree_cur *cur, | |
5dfa5cd2 | 1182 | struct xfs_buf *bp, |
a2ceac1f | 1183 | int level, |
5dfa5cd2 | 1184 | int numrecs) |
a2ceac1f | 1185 | { |
4c6b3277 | 1186 | __u64 owner; |
5dfa5cd2 DC |
1187 | |
1188 | /* | |
1189 | * we can pull the owner from the cursor right now as the different | |
1190 | * owners align directly with the pointer size of the btree. This may | |
1191 | * change in future, but is safe for current users of the generic btree | |
1192 | * code. | |
1193 | */ | |
1194 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) | |
1195 | owner = cur->bc_private.b.ip->i_ino; | |
1196 | else | |
1197 | owner = cur->bc_private.a.agno; | |
1198 | ||
1199 | xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn, | |
e394a4b1 | 1200 | cur->bc_btnum, level, numrecs, |
5dfa5cd2 | 1201 | owner, cur->bc_flags); |
a2ceac1f DC |
1202 | } |
1203 | ||
b194c7d8 BN |
1204 | /* |
1205 | * Return true if ptr is the last record in the btree and | |
5dfa5cd2 | 1206 | * we need to track updates to this record. The decision |
b194c7d8 BN |
1207 | * will be further refined in the update_lastrec method. |
1208 | */ | |
1209 | STATIC int | |
1210 | xfs_btree_is_lastrec( | |
1211 | struct xfs_btree_cur *cur, | |
1212 | struct xfs_btree_block *block, | |
1213 | int level) | |
1214 | { | |
1215 | union xfs_btree_ptr ptr; | |
1216 | ||
1217 | if (level > 0) | |
1218 | return 0; | |
1219 | if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE)) | |
1220 | return 0; | |
1221 | ||
1222 | xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB); | |
1223 | if (!xfs_btree_ptr_is_null(cur, &ptr)) | |
1224 | return 0; | |
1225 | return 1; | |
1226 | } | |
1227 | ||
1228 | STATIC void | |
1229 | xfs_btree_buf_to_ptr( | |
1230 | struct xfs_btree_cur *cur, | |
1231 | struct xfs_buf *bp, | |
1232 | union xfs_btree_ptr *ptr) | |
1233 | { | |
1234 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) | |
1235 | ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp, | |
1236 | XFS_BUF_ADDR(bp))); | |
1237 | else { | |
56b2de80 | 1238 | ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp, |
b194c7d8 BN |
1239 | XFS_BUF_ADDR(bp))); |
1240 | } | |
1241 | } | |
1242 | ||
b194c7d8 BN |
1243 | STATIC void |
1244 | xfs_btree_set_refs( | |
1245 | struct xfs_btree_cur *cur, | |
1246 | struct xfs_buf *bp) | |
1247 | { | |
1248 | switch (cur->bc_btnum) { | |
1249 | case XFS_BTNUM_BNO: | |
1250 | case XFS_BTNUM_CNT: | |
a2ceac1f | 1251 | xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF); |
b194c7d8 BN |
1252 | break; |
1253 | case XFS_BTNUM_INO: | |
c0a4c227 | 1254 | case XFS_BTNUM_FINO: |
a2ceac1f | 1255 | xfs_buf_set_ref(bp, XFS_INO_BTREE_REF); |
b194c7d8 BN |
1256 | break; |
1257 | case XFS_BTNUM_BMAP: | |
a2ceac1f | 1258 | xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF); |
b194c7d8 | 1259 | break; |
b3a96b46 DW |
1260 | case XFS_BTNUM_RMAP: |
1261 | xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF); | |
1262 | break; | |
d8079fe0 DW |
1263 | case XFS_BTNUM_REFC: |
1264 | xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF); | |
1265 | break; | |
b194c7d8 BN |
1266 | default: |
1267 | ASSERT(0); | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | STATIC int | |
1272 | xfs_btree_get_buf_block( | |
1273 | struct xfs_btree_cur *cur, | |
1274 | union xfs_btree_ptr *ptr, | |
1275 | int flags, | |
1276 | struct xfs_btree_block **block, | |
1277 | struct xfs_buf **bpp) | |
1278 | { | |
1279 | struct xfs_mount *mp = cur->bc_mp; | |
1280 | xfs_daddr_t d; | |
1281 | ||
1282 | /* need to sort out how callers deal with failures first */ | |
56b2de80 | 1283 | ASSERT(!(flags & XBF_TRYLOCK)); |
b194c7d8 BN |
1284 | |
1285 | d = xfs_btree_ptr_to_daddr(cur, ptr); | |
1286 | *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d, | |
1287 | mp->m_bsize, flags); | |
1288 | ||
a2ceac1f | 1289 | if (!*bpp) |
12b53197 | 1290 | return -ENOMEM; |
b194c7d8 | 1291 | |
a2ceac1f | 1292 | (*bpp)->b_ops = cur->bc_ops->buf_ops; |
b194c7d8 BN |
1293 | *block = XFS_BUF_TO_BLOCK(*bpp); |
1294 | return 0; | |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * Read in the buffer at the given ptr and return the buffer and | |
1299 | * the block pointer within the buffer. | |
1300 | */ | |
1301 | STATIC int | |
1302 | xfs_btree_read_buf_block( | |
1303 | struct xfs_btree_cur *cur, | |
1304 | union xfs_btree_ptr *ptr, | |
b194c7d8 BN |
1305 | int flags, |
1306 | struct xfs_btree_block **block, | |
1307 | struct xfs_buf **bpp) | |
1308 | { | |
1309 | struct xfs_mount *mp = cur->bc_mp; | |
1310 | xfs_daddr_t d; | |
1311 | int error; | |
1312 | ||
1313 | /* need to sort out how callers deal with failures first */ | |
56b2de80 | 1314 | ASSERT(!(flags & XBF_TRYLOCK)); |
b194c7d8 BN |
1315 | |
1316 | d = xfs_btree_ptr_to_daddr(cur, ptr); | |
1317 | error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d, | |
a2ceac1f DC |
1318 | mp->m_bsize, flags, bpp, |
1319 | cur->bc_ops->buf_ops); | |
b194c7d8 BN |
1320 | if (error) |
1321 | return error; | |
1322 | ||
b194c7d8 BN |
1323 | xfs_btree_set_refs(cur, *bpp); |
1324 | *block = XFS_BUF_TO_BLOCK(*bpp); | |
a2ceac1f | 1325 | return 0; |
b194c7d8 BN |
1326 | } |
1327 | ||
1328 | /* | |
1329 | * Copy keys from one btree block to another. | |
1330 | */ | |
1331 | STATIC void | |
1332 | xfs_btree_copy_keys( | |
1333 | struct xfs_btree_cur *cur, | |
1334 | union xfs_btree_key *dst_key, | |
1335 | union xfs_btree_key *src_key, | |
1336 | int numkeys) | |
1337 | { | |
1338 | ASSERT(numkeys >= 0); | |
1339 | memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len); | |
1340 | } | |
1341 | ||
1342 | /* | |
1343 | * Copy records from one btree block to another. | |
1344 | */ | |
1345 | STATIC void | |
1346 | xfs_btree_copy_recs( | |
1347 | struct xfs_btree_cur *cur, | |
1348 | union xfs_btree_rec *dst_rec, | |
1349 | union xfs_btree_rec *src_rec, | |
1350 | int numrecs) | |
1351 | { | |
1352 | ASSERT(numrecs >= 0); | |
1353 | memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len); | |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * Copy block pointers from one btree block to another. | |
1358 | */ | |
1359 | STATIC void | |
1360 | xfs_btree_copy_ptrs( | |
1361 | struct xfs_btree_cur *cur, | |
1362 | union xfs_btree_ptr *dst_ptr, | |
1363 | union xfs_btree_ptr *src_ptr, | |
1364 | int numptrs) | |
1365 | { | |
1366 | ASSERT(numptrs >= 0); | |
1367 | memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur)); | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Shift keys one index left/right inside a single btree block. | |
1372 | */ | |
1373 | STATIC void | |
1374 | xfs_btree_shift_keys( | |
1375 | struct xfs_btree_cur *cur, | |
1376 | union xfs_btree_key *key, | |
1377 | int dir, | |
1378 | int numkeys) | |
1379 | { | |
1380 | char *dst_key; | |
1381 | ||
1382 | ASSERT(numkeys >= 0); | |
1383 | ASSERT(dir == 1 || dir == -1); | |
1384 | ||
1385 | dst_key = (char *)key + (dir * cur->bc_ops->key_len); | |
1386 | memmove(dst_key, key, numkeys * cur->bc_ops->key_len); | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * Shift records one index left/right inside a single btree block. | |
1391 | */ | |
1392 | STATIC void | |
1393 | xfs_btree_shift_recs( | |
1394 | struct xfs_btree_cur *cur, | |
1395 | union xfs_btree_rec *rec, | |
1396 | int dir, | |
1397 | int numrecs) | |
1398 | { | |
1399 | char *dst_rec; | |
1400 | ||
1401 | ASSERT(numrecs >= 0); | |
1402 | ASSERT(dir == 1 || dir == -1); | |
1403 | ||
1404 | dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len); | |
1405 | memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len); | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * Shift block pointers one index left/right inside a single btree block. | |
1410 | */ | |
1411 | STATIC void | |
1412 | xfs_btree_shift_ptrs( | |
1413 | struct xfs_btree_cur *cur, | |
1414 | union xfs_btree_ptr *ptr, | |
1415 | int dir, | |
1416 | int numptrs) | |
1417 | { | |
1418 | char *dst_ptr; | |
1419 | ||
1420 | ASSERT(numptrs >= 0); | |
1421 | ASSERT(dir == 1 || dir == -1); | |
1422 | ||
1423 | dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur)); | |
1424 | memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur)); | |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * Log key values from the btree block. | |
1429 | */ | |
1430 | STATIC void | |
1431 | xfs_btree_log_keys( | |
1432 | struct xfs_btree_cur *cur, | |
1433 | struct xfs_buf *bp, | |
1434 | int first, | |
1435 | int last) | |
1436 | { | |
1437 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
1438 | XFS_BTREE_TRACE_ARGBII(cur, bp, first, last); | |
1439 | ||
1440 | if (bp) { | |
bdc16ee5 | 1441 | xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF); |
b194c7d8 BN |
1442 | xfs_trans_log_buf(cur->bc_tp, bp, |
1443 | xfs_btree_key_offset(cur, first), | |
1444 | xfs_btree_key_offset(cur, last + 1) - 1); | |
1445 | } else { | |
1446 | xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip, | |
1447 | xfs_ilog_fbroot(cur->bc_private.b.whichfork)); | |
1448 | } | |
1449 | ||
1450 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1451 | } | |
1452 | ||
1453 | /* | |
1454 | * Log record values from the btree block. | |
1455 | */ | |
1456 | void | |
1457 | xfs_btree_log_recs( | |
1458 | struct xfs_btree_cur *cur, | |
1459 | struct xfs_buf *bp, | |
1460 | int first, | |
1461 | int last) | |
1462 | { | |
1463 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
1464 | XFS_BTREE_TRACE_ARGBII(cur, bp, first, last); | |
1465 | ||
bdc16ee5 | 1466 | xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF); |
b194c7d8 BN |
1467 | xfs_trans_log_buf(cur->bc_tp, bp, |
1468 | xfs_btree_rec_offset(cur, first), | |
1469 | xfs_btree_rec_offset(cur, last + 1) - 1); | |
1470 | ||
1471 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1472 | } | |
1473 | ||
1474 | /* | |
1475 | * Log block pointer fields from a btree block (nonleaf). | |
1476 | */ | |
1477 | STATIC void | |
1478 | xfs_btree_log_ptrs( | |
1479 | struct xfs_btree_cur *cur, /* btree cursor */ | |
1480 | struct xfs_buf *bp, /* buffer containing btree block */ | |
1481 | int first, /* index of first pointer to log */ | |
1482 | int last) /* index of last pointer to log */ | |
1483 | { | |
1484 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
1485 | XFS_BTREE_TRACE_ARGBII(cur, bp, first, last); | |
1486 | ||
1487 | if (bp) { | |
1488 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); | |
1489 | int level = xfs_btree_get_level(block); | |
1490 | ||
bdc16ee5 | 1491 | xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF); |
b194c7d8 BN |
1492 | xfs_trans_log_buf(cur->bc_tp, bp, |
1493 | xfs_btree_ptr_offset(cur, first, level), | |
1494 | xfs_btree_ptr_offset(cur, last + 1, level) - 1); | |
1495 | } else { | |
1496 | xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip, | |
1497 | xfs_ilog_fbroot(cur->bc_private.b.whichfork)); | |
1498 | } | |
1499 | ||
1500 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1501 | } | |
1502 | ||
1503 | /* | |
1504 | * Log fields from a btree block header. | |
1505 | */ | |
1506 | void | |
1507 | xfs_btree_log_block( | |
1508 | struct xfs_btree_cur *cur, /* btree cursor */ | |
1509 | struct xfs_buf *bp, /* buffer containing btree block */ | |
1510 | int fields) /* mask of fields: XFS_BB_... */ | |
1511 | { | |
1512 | int first; /* first byte offset logged */ | |
1513 | int last; /* last byte offset logged */ | |
1514 | static const short soffsets[] = { /* table of offsets (short) */ | |
b3563c19 BN |
1515 | offsetof(struct xfs_btree_block, bb_magic), |
1516 | offsetof(struct xfs_btree_block, bb_level), | |
1517 | offsetof(struct xfs_btree_block, bb_numrecs), | |
1518 | offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib), | |
1519 | offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib), | |
5dfa5cd2 DC |
1520 | offsetof(struct xfs_btree_block, bb_u.s.bb_blkno), |
1521 | offsetof(struct xfs_btree_block, bb_u.s.bb_lsn), | |
1522 | offsetof(struct xfs_btree_block, bb_u.s.bb_uuid), | |
1523 | offsetof(struct xfs_btree_block, bb_u.s.bb_owner), | |
1524 | offsetof(struct xfs_btree_block, bb_u.s.bb_crc), | |
e0607266 | 1525 | XFS_BTREE_SBLOCK_CRC_LEN |
b194c7d8 BN |
1526 | }; |
1527 | static const short loffsets[] = { /* table of offsets (long) */ | |
b3563c19 BN |
1528 | offsetof(struct xfs_btree_block, bb_magic), |
1529 | offsetof(struct xfs_btree_block, bb_level), | |
1530 | offsetof(struct xfs_btree_block, bb_numrecs), | |
1531 | offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib), | |
1532 | offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib), | |
5dfa5cd2 DC |
1533 | offsetof(struct xfs_btree_block, bb_u.l.bb_blkno), |
1534 | offsetof(struct xfs_btree_block, bb_u.l.bb_lsn), | |
1535 | offsetof(struct xfs_btree_block, bb_u.l.bb_uuid), | |
1536 | offsetof(struct xfs_btree_block, bb_u.l.bb_owner), | |
1537 | offsetof(struct xfs_btree_block, bb_u.l.bb_crc), | |
1538 | offsetof(struct xfs_btree_block, bb_u.l.bb_pad), | |
e0607266 | 1539 | XFS_BTREE_LBLOCK_CRC_LEN |
b194c7d8 BN |
1540 | }; |
1541 | ||
1542 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
1543 | XFS_BTREE_TRACE_ARGBI(cur, bp, fields); | |
1544 | ||
1545 | if (bp) { | |
5dfa5cd2 DC |
1546 | int nbits; |
1547 | ||
1548 | if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) { | |
1549 | /* | |
1550 | * We don't log the CRC when updating a btree | |
1551 | * block but instead recreate it during log | |
1552 | * recovery. As the log buffers have checksums | |
10851b18 | 1553 | * of their own this is safe and avoids logging a crc |
5dfa5cd2 DC |
1554 | * update in a lot of places. |
1555 | */ | |
1556 | if (fields == XFS_BB_ALL_BITS) | |
1557 | fields = XFS_BB_ALL_BITS_CRC; | |
1558 | nbits = XFS_BB_NUM_BITS_CRC; | |
1559 | } else { | |
1560 | nbits = XFS_BB_NUM_BITS; | |
1561 | } | |
b194c7d8 BN |
1562 | xfs_btree_offsets(fields, |
1563 | (cur->bc_flags & XFS_BTREE_LONG_PTRS) ? | |
1564 | loffsets : soffsets, | |
5dfa5cd2 | 1565 | nbits, &first, &last); |
bdc16ee5 | 1566 | xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF); |
b194c7d8 BN |
1567 | xfs_trans_log_buf(cur->bc_tp, bp, first, last); |
1568 | } else { | |
1569 | xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip, | |
1570 | xfs_ilog_fbroot(cur->bc_private.b.whichfork)); | |
1571 | } | |
1572 | ||
1573 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * Increment cursor by one record at the level. | |
1578 | * For nonzero levels the leaf-ward information is untouched. | |
1579 | */ | |
1580 | int /* error */ | |
1581 | xfs_btree_increment( | |
1582 | struct xfs_btree_cur *cur, | |
1583 | int level, | |
1584 | int *stat) /* success/failure */ | |
1585 | { | |
1586 | struct xfs_btree_block *block; | |
1587 | union xfs_btree_ptr ptr; | |
1588 | struct xfs_buf *bp; | |
1589 | int error; /* error return value */ | |
1590 | int lev; | |
1591 | ||
1592 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
1593 | XFS_BTREE_TRACE_ARGI(cur, level); | |
1594 | ||
1595 | ASSERT(level < cur->bc_nlevels); | |
1596 | ||
1597 | /* Read-ahead to the right at this level. */ | |
1598 | xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA); | |
1599 | ||
1600 | /* Get a pointer to the btree block. */ | |
1601 | block = xfs_btree_get_block(cur, level, &bp); | |
1602 | ||
1603 | #ifdef DEBUG | |
1604 | error = xfs_btree_check_block(cur, block, level, bp); | |
1605 | if (error) | |
1606 | goto error0; | |
1607 | #endif | |
1608 | ||
1609 | /* We're done if we remain in the block after the increment. */ | |
1610 | if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block)) | |
1611 | goto out1; | |
1612 | ||
1613 | /* Fail if we just went off the right edge of the tree. */ | |
1614 | xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB); | |
1615 | if (xfs_btree_ptr_is_null(cur, &ptr)) | |
1616 | goto out0; | |
1617 | ||
1618 | XFS_BTREE_STATS_INC(cur, increment); | |
1619 | ||
1620 | /* | |
1621 | * March up the tree incrementing pointers. | |
1622 | * Stop when we don't go off the right edge of a block. | |
1623 | */ | |
1624 | for (lev = level + 1; lev < cur->bc_nlevels; lev++) { | |
1625 | block = xfs_btree_get_block(cur, lev, &bp); | |
1626 | ||
1627 | #ifdef DEBUG | |
1628 | error = xfs_btree_check_block(cur, block, lev, bp); | |
1629 | if (error) | |
1630 | goto error0; | |
1631 | #endif | |
1632 | ||
1633 | if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block)) | |
1634 | break; | |
1635 | ||
1636 | /* Read-ahead the right block for the next loop. */ | |
1637 | xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA); | |
1638 | } | |
1639 | ||
1640 | /* | |
1641 | * If we went off the root then we are either seriously | |
1642 | * confused or have the tree root in an inode. | |
1643 | */ | |
1644 | if (lev == cur->bc_nlevels) { | |
1645 | if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) | |
1646 | goto out0; | |
1647 | ASSERT(0); | |
12b53197 | 1648 | error = -EFSCORRUPTED; |
b194c7d8 BN |
1649 | goto error0; |
1650 | } | |
1651 | ASSERT(lev < cur->bc_nlevels); | |
1652 | ||
1653 | /* | |
1654 | * Now walk back down the tree, fixing up the cursor's buffer | |
1655 | * pointers and key numbers. | |
1656 | */ | |
1657 | for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) { | |
1658 | union xfs_btree_ptr *ptrp; | |
1659 | ||
1660 | ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block); | |
ff105f75 DC |
1661 | --lev; |
1662 | error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp); | |
b194c7d8 BN |
1663 | if (error) |
1664 | goto error0; | |
1665 | ||
1666 | xfs_btree_setbuf(cur, lev, bp); | |
1667 | cur->bc_ptrs[lev] = 1; | |
1668 | } | |
1669 | out1: | |
1670 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1671 | *stat = 1; | |
1672 | return 0; | |
1673 | ||
1674 | out0: | |
1675 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1676 | *stat = 0; | |
1677 | return 0; | |
1678 | ||
1679 | error0: | |
1680 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
1681 | return error; | |
1682 | } | |
1683 | ||
1684 | /* | |
1685 | * Decrement cursor by one record at the level. | |
1686 | * For nonzero levels the leaf-ward information is untouched. | |
1687 | */ | |
1688 | int /* error */ | |
1689 | xfs_btree_decrement( | |
1690 | struct xfs_btree_cur *cur, | |
1691 | int level, | |
1692 | int *stat) /* success/failure */ | |
1693 | { | |
1694 | struct xfs_btree_block *block; | |
1695 | xfs_buf_t *bp; | |
1696 | int error; /* error return value */ | |
1697 | int lev; | |
1698 | union xfs_btree_ptr ptr; | |
1699 | ||
1700 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
1701 | XFS_BTREE_TRACE_ARGI(cur, level); | |
1702 | ||
1703 | ASSERT(level < cur->bc_nlevels); | |
1704 | ||
1705 | /* Read-ahead to the left at this level. */ | |
1706 | xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA); | |
1707 | ||
1708 | /* We're done if we remain in the block after the decrement. */ | |
1709 | if (--cur->bc_ptrs[level] > 0) | |
1710 | goto out1; | |
1711 | ||
1712 | /* Get a pointer to the btree block. */ | |
1713 | block = xfs_btree_get_block(cur, level, &bp); | |
1714 | ||
1715 | #ifdef DEBUG | |
1716 | error = xfs_btree_check_block(cur, block, level, bp); | |
1717 | if (error) | |
1718 | goto error0; | |
1719 | #endif | |
1720 | ||
1721 | /* Fail if we just went off the left edge of the tree. */ | |
1722 | xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB); | |
1723 | if (xfs_btree_ptr_is_null(cur, &ptr)) | |
1724 | goto out0; | |
1725 | ||
1726 | XFS_BTREE_STATS_INC(cur, decrement); | |
1727 | ||
1728 | /* | |
1729 | * March up the tree decrementing pointers. | |
1730 | * Stop when we don't go off the left edge of a block. | |
1731 | */ | |
1732 | for (lev = level + 1; lev < cur->bc_nlevels; lev++) { | |
1733 | if (--cur->bc_ptrs[lev] > 0) | |
1734 | break; | |
1735 | /* Read-ahead the left block for the next loop. */ | |
1736 | xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA); | |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * If we went off the root then we are seriously confused. | |
1741 | * or the root of the tree is in an inode. | |
1742 | */ | |
1743 | if (lev == cur->bc_nlevels) { | |
1744 | if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) | |
1745 | goto out0; | |
1746 | ASSERT(0); | |
12b53197 | 1747 | error = -EFSCORRUPTED; |
b194c7d8 BN |
1748 | goto error0; |
1749 | } | |
1750 | ASSERT(lev < cur->bc_nlevels); | |
1751 | ||
1752 | /* | |
1753 | * Now walk back down the tree, fixing up the cursor's buffer | |
1754 | * pointers and key numbers. | |
1755 | */ | |
1756 | for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) { | |
1757 | union xfs_btree_ptr *ptrp; | |
1758 | ||
1759 | ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block); | |
ff105f75 DC |
1760 | --lev; |
1761 | error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp); | |
b194c7d8 BN |
1762 | if (error) |
1763 | goto error0; | |
1764 | xfs_btree_setbuf(cur, lev, bp); | |
1765 | cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block); | |
1766 | } | |
1767 | out1: | |
1768 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1769 | *stat = 1; | |
1770 | return 0; | |
1771 | ||
1772 | out0: | |
1773 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1774 | *stat = 0; | |
1775 | return 0; | |
1776 | ||
1777 | error0: | |
1778 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
1779 | return error; | |
1780 | } | |
1781 | ||
50bb67d6 | 1782 | int |
b194c7d8 BN |
1783 | xfs_btree_lookup_get_block( |
1784 | struct xfs_btree_cur *cur, /* btree cursor */ | |
1785 | int level, /* level in the btree */ | |
1786 | union xfs_btree_ptr *pp, /* ptr to btree block */ | |
1787 | struct xfs_btree_block **blkp) /* return btree block */ | |
1788 | { | |
1789 | struct xfs_buf *bp; /* buffer pointer for btree block */ | |
1790 | int error = 0; | |
1791 | ||
1792 | /* special case the root block if in an inode */ | |
1793 | if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) && | |
1794 | (level == cur->bc_nlevels - 1)) { | |
1795 | *blkp = xfs_btree_get_iroot(cur); | |
1796 | return 0; | |
1797 | } | |
1798 | ||
1799 | /* | |
1800 | * If the old buffer at this level for the disk address we are | |
1801 | * looking for re-use it. | |
1802 | * | |
1803 | * Otherwise throw it away and get a new one. | |
1804 | */ | |
1805 | bp = cur->bc_bufs[level]; | |
1806 | if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) { | |
1807 | *blkp = XFS_BUF_TO_BLOCK(bp); | |
1808 | return 0; | |
1809 | } | |
1810 | ||
ff105f75 | 1811 | error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp); |
b194c7d8 BN |
1812 | if (error) |
1813 | return error; | |
1814 | ||
312eea24 DW |
1815 | /* Check the inode owner since the verifiers don't. */ |
1816 | if (xfs_sb_version_hascrc(&cur->bc_mp->m_sb) && | |
c7ba1731 | 1817 | !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_INVALID_OWNER) && |
312eea24 DW |
1818 | (cur->bc_flags & XFS_BTREE_LONG_PTRS) && |
1819 | be64_to_cpu((*blkp)->bb_u.l.bb_owner) != | |
1820 | cur->bc_private.b.ip->i_ino) | |
1821 | goto out_bad; | |
1822 | ||
1823 | /* Did we get the level we were looking for? */ | |
1824 | if (be16_to_cpu((*blkp)->bb_level) != level) | |
1825 | goto out_bad; | |
1826 | ||
1827 | /* Check that internal nodes have at least one record. */ | |
1828 | if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0) | |
1829 | goto out_bad; | |
1830 | ||
b194c7d8 BN |
1831 | xfs_btree_setbuf(cur, level, bp); |
1832 | return 0; | |
312eea24 DW |
1833 | |
1834 | out_bad: | |
1835 | *blkp = NULL; | |
1836 | xfs_trans_brelse(cur->bc_tp, bp); | |
1837 | return -EFSCORRUPTED; | |
b194c7d8 BN |
1838 | } |
1839 | ||
1840 | /* | |
1841 | * Get current search key. For level 0 we don't actually have a key | |
1842 | * structure so we make one up from the record. For all other levels | |
1843 | * we just return the right key. | |
1844 | */ | |
1845 | STATIC union xfs_btree_key * | |
1846 | xfs_lookup_get_search_key( | |
1847 | struct xfs_btree_cur *cur, | |
1848 | int level, | |
1849 | int keyno, | |
1850 | struct xfs_btree_block *block, | |
1851 | union xfs_btree_key *kp) | |
1852 | { | |
1853 | if (level == 0) { | |
1854 | cur->bc_ops->init_key_from_rec(kp, | |
1855 | xfs_btree_rec_addr(cur, keyno, block)); | |
1856 | return kp; | |
1857 | } | |
1858 | ||
1859 | return xfs_btree_key_addr(cur, keyno, block); | |
1860 | } | |
1861 | ||
1862 | /* | |
1863 | * Lookup the record. The cursor is made to point to it, based on dir. | |
10851b18 | 1864 | * stat is set to 0 if can't find any such record, 1 for success. |
b194c7d8 BN |
1865 | */ |
1866 | int /* error */ | |
1867 | xfs_btree_lookup( | |
1868 | struct xfs_btree_cur *cur, /* btree cursor */ | |
1869 | xfs_lookup_t dir, /* <=, ==, or >= */ | |
1870 | int *stat) /* success/failure */ | |
1871 | { | |
1872 | struct xfs_btree_block *block; /* current btree block */ | |
4a492e72 | 1873 | int64_t diff; /* difference for the current key */ |
b194c7d8 BN |
1874 | int error; /* error return value */ |
1875 | int keyno; /* current key number */ | |
1876 | int level; /* level in the btree */ | |
1877 | union xfs_btree_ptr *pp; /* ptr to btree block */ | |
1878 | union xfs_btree_ptr ptr; /* ptr to btree block */ | |
1879 | ||
1880 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
1881 | XFS_BTREE_TRACE_ARGI(cur, dir); | |
1882 | ||
1883 | XFS_BTREE_STATS_INC(cur, lookup); | |
1884 | ||
574b4153 DW |
1885 | /* No such thing as a zero-level tree. */ |
1886 | if (cur->bc_nlevels == 0) | |
1887 | return -EFSCORRUPTED; | |
1888 | ||
b194c7d8 BN |
1889 | block = NULL; |
1890 | keyno = 0; | |
1891 | ||
1892 | /* initialise start pointer from cursor */ | |
1893 | cur->bc_ops->init_ptr_from_cur(cur, &ptr); | |
1894 | pp = &ptr; | |
1895 | ||
1896 | /* | |
1897 | * Iterate over each level in the btree, starting at the root. | |
1898 | * For each level above the leaves, find the key we need, based | |
1899 | * on the lookup record, then follow the corresponding block | |
1900 | * pointer down to the next level. | |
1901 | */ | |
1902 | for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) { | |
1903 | /* Get the block we need to do the lookup on. */ | |
1904 | error = xfs_btree_lookup_get_block(cur, level, pp, &block); | |
1905 | if (error) | |
1906 | goto error0; | |
1907 | ||
1908 | if (diff == 0) { | |
1909 | /* | |
1910 | * If we already had a key match at a higher level, we | |
1911 | * know we need to use the first entry in this block. | |
1912 | */ | |
1913 | keyno = 1; | |
1914 | } else { | |
1915 | /* Otherwise search this block. Do a binary search. */ | |
1916 | ||
1917 | int high; /* high entry number */ | |
1918 | int low; /* low entry number */ | |
1919 | ||
1920 | /* Set low and high entry numbers, 1-based. */ | |
1921 | low = 1; | |
1922 | high = xfs_btree_get_numrecs(block); | |
1923 | if (!high) { | |
1924 | /* Block is empty, must be an empty leaf. */ | |
1925 | ASSERT(level == 0 && cur->bc_nlevels == 1); | |
1926 | ||
1927 | cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE; | |
1928 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
1929 | *stat = 0; | |
1930 | return 0; | |
1931 | } | |
1932 | ||
1933 | /* Binary search the block. */ | |
1934 | while (low <= high) { | |
1935 | union xfs_btree_key key; | |
1936 | union xfs_btree_key *kp; | |
1937 | ||
1938 | XFS_BTREE_STATS_INC(cur, compare); | |
1939 | ||
1940 | /* keyno is average of low and high. */ | |
1941 | keyno = (low + high) >> 1; | |
1942 | ||
1943 | /* Get current search key */ | |
1944 | kp = xfs_lookup_get_search_key(cur, level, | |
1945 | keyno, block, &key); | |
1946 | ||
1947 | /* | |
1948 | * Compute difference to get next direction: | |
1949 | * - less than, move right | |
1950 | * - greater than, move left | |
1951 | * - equal, we're done | |
1952 | */ | |
1953 | diff = cur->bc_ops->key_diff(cur, kp); | |
1954 | if (diff < 0) | |
1955 | low = keyno + 1; | |
1956 | else if (diff > 0) | |
1957 | high = keyno - 1; | |
1958 | else | |
1959 | break; | |
1960 | } | |
1961 | } | |
1962 | ||
1963 | /* | |
1964 | * If there are more levels, set up for the next level | |
1965 | * by getting the block number and filling in the cursor. | |
1966 | */ | |
1967 | if (level > 0) { | |
1968 | /* | |
1969 | * If we moved left, need the previous key number, | |
1970 | * unless there isn't one. | |
1971 | */ | |
1972 | if (diff > 0 && --keyno < 1) | |
1973 | keyno = 1; | |
1974 | pp = xfs_btree_ptr_addr(cur, keyno, block); | |
1975 | ||
1976 | #ifdef DEBUG | |
1977 | error = xfs_btree_check_ptr(cur, pp, 0, level); | |
1978 | if (error) | |
1979 | goto error0; | |
1980 | #endif | |
1981 | cur->bc_ptrs[level] = keyno; | |
1982 | } | |
1983 | } | |
1984 | ||
1985 | /* Done with the search. See if we need to adjust the results. */ | |
1986 | if (dir != XFS_LOOKUP_LE && diff < 0) { | |
1987 | keyno++; | |
1988 | /* | |
1989 | * If ge search and we went off the end of the block, but it's | |
1990 | * not the last block, we're in the wrong block. | |
1991 | */ | |
1992 | xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB); | |
1993 | if (dir == XFS_LOOKUP_GE && | |
1994 | keyno > xfs_btree_get_numrecs(block) && | |
1995 | !xfs_btree_ptr_is_null(cur, &ptr)) { | |
1996 | int i; | |
1997 | ||
1998 | cur->bc_ptrs[0] = keyno; | |
1999 | error = xfs_btree_increment(cur, 0, &i); | |
2000 | if (error) | |
2001 | goto error0; | |
19ebedcf | 2002 | XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1); |
b194c7d8 BN |
2003 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); |
2004 | *stat = 1; | |
2005 | return 0; | |
2006 | } | |
2007 | } else if (dir == XFS_LOOKUP_LE && diff > 0) | |
2008 | keyno--; | |
2009 | cur->bc_ptrs[0] = keyno; | |
2010 | ||
2011 | /* Return if we succeeded or not. */ | |
2012 | if (keyno == 0 || keyno > xfs_btree_get_numrecs(block)) | |
2013 | *stat = 0; | |
2014 | else if (dir != XFS_LOOKUP_EQ || diff == 0) | |
2015 | *stat = 1; | |
2016 | else | |
2017 | *stat = 0; | |
2018 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2019 | return 0; | |
2020 | ||
2021 | error0: | |
2022 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
2023 | return error; | |
2024 | } | |
2025 | ||
13e831e0 | 2026 | /* Find the high key storage area from a regular key. */ |
0ad12e7f | 2027 | union xfs_btree_key * |
13e831e0 DW |
2028 | xfs_btree_high_key_from_key( |
2029 | struct xfs_btree_cur *cur, | |
2030 | union xfs_btree_key *key) | |
2031 | { | |
2032 | ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING); | |
2033 | return (union xfs_btree_key *)((char *)key + | |
2034 | (cur->bc_ops->key_len / 2)); | |
2035 | } | |
2036 | ||
64dbe047 DW |
2037 | /* Determine the low (and high if overlapped) keys of a leaf block */ |
2038 | STATIC void | |
2039 | xfs_btree_get_leaf_keys( | |
13e831e0 DW |
2040 | struct xfs_btree_cur *cur, |
2041 | struct xfs_btree_block *block, | |
2042 | union xfs_btree_key *key) | |
2043 | { | |
13e831e0 DW |
2044 | union xfs_btree_key max_hkey; |
2045 | union xfs_btree_key hkey; | |
64dbe047 | 2046 | union xfs_btree_rec *rec; |
13e831e0 | 2047 | union xfs_btree_key *high; |
64dbe047 | 2048 | int n; |
13e831e0 | 2049 | |
13e831e0 DW |
2050 | rec = xfs_btree_rec_addr(cur, 1, block); |
2051 | cur->bc_ops->init_key_from_rec(key, rec); | |
2052 | ||
64dbe047 DW |
2053 | if (cur->bc_flags & XFS_BTREE_OVERLAPPING) { |
2054 | ||
2055 | cur->bc_ops->init_high_key_from_rec(&max_hkey, rec); | |
2056 | for (n = 2; n <= xfs_btree_get_numrecs(block); n++) { | |
2057 | rec = xfs_btree_rec_addr(cur, n, block); | |
2058 | cur->bc_ops->init_high_key_from_rec(&hkey, rec); | |
2059 | if (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey) | |
2060 | > 0) | |
2061 | max_hkey = hkey; | |
2062 | } | |
13e831e0 | 2063 | |
64dbe047 DW |
2064 | high = xfs_btree_high_key_from_key(cur, key); |
2065 | memcpy(high, &max_hkey, cur->bc_ops->key_len / 2); | |
2066 | } | |
13e831e0 DW |
2067 | } |
2068 | ||
64dbe047 DW |
2069 | /* Determine the low (and high if overlapped) keys of a node block */ |
2070 | STATIC void | |
2071 | xfs_btree_get_node_keys( | |
13e831e0 DW |
2072 | struct xfs_btree_cur *cur, |
2073 | struct xfs_btree_block *block, | |
2074 | union xfs_btree_key *key) | |
2075 | { | |
13e831e0 DW |
2076 | union xfs_btree_key *hkey; |
2077 | union xfs_btree_key *max_hkey; | |
2078 | union xfs_btree_key *high; | |
64dbe047 | 2079 | int n; |
13e831e0 | 2080 | |
64dbe047 DW |
2081 | if (cur->bc_flags & XFS_BTREE_OVERLAPPING) { |
2082 | memcpy(key, xfs_btree_key_addr(cur, 1, block), | |
2083 | cur->bc_ops->key_len / 2); | |
2084 | ||
2085 | max_hkey = xfs_btree_high_key_addr(cur, 1, block); | |
2086 | for (n = 2; n <= xfs_btree_get_numrecs(block); n++) { | |
2087 | hkey = xfs_btree_high_key_addr(cur, n, block); | |
2088 | if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0) | |
2089 | max_hkey = hkey; | |
2090 | } | |
13e831e0 | 2091 | |
64dbe047 DW |
2092 | high = xfs_btree_high_key_from_key(cur, key); |
2093 | memcpy(high, max_hkey, cur->bc_ops->key_len / 2); | |
2094 | } else { | |
2095 | memcpy(key, xfs_btree_key_addr(cur, 1, block), | |
2096 | cur->bc_ops->key_len); | |
13e831e0 | 2097 | } |
13e831e0 DW |
2098 | } |
2099 | ||
a3c9cb10 | 2100 | /* Derive the keys for any btree block. */ |
0ad12e7f | 2101 | void |
a3c9cb10 DW |
2102 | xfs_btree_get_keys( |
2103 | struct xfs_btree_cur *cur, | |
2104 | struct xfs_btree_block *block, | |
2105 | union xfs_btree_key *key) | |
2106 | { | |
2107 | if (be16_to_cpu(block->bb_level) == 0) | |
64dbe047 | 2108 | xfs_btree_get_leaf_keys(cur, block, key); |
a3c9cb10 | 2109 | else |
64dbe047 | 2110 | xfs_btree_get_node_keys(cur, block, key); |
a3c9cb10 DW |
2111 | } |
2112 | ||
b194c7d8 | 2113 | /* |
a3c9cb10 DW |
2114 | * Decide if we need to update the parent keys of a btree block. For |
2115 | * a standard btree this is only necessary if we're updating the first | |
13e831e0 DW |
2116 | * record/key. For an overlapping btree, we must always update the |
2117 | * keys because the highest key can be in any of the records or keys | |
2118 | * in the block. | |
b194c7d8 | 2119 | */ |
a3c9cb10 DW |
2120 | static inline bool |
2121 | xfs_btree_needs_key_update( | |
2122 | struct xfs_btree_cur *cur, | |
2123 | int ptr) | |
2124 | { | |
13e831e0 DW |
2125 | return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1; |
2126 | } | |
2127 | ||
2128 | /* | |
2129 | * Update the low and high parent keys of the given level, progressing | |
2130 | * towards the root. If force_all is false, stop if the keys for a given | |
2131 | * level do not need updating. | |
2132 | */ | |
2133 | STATIC int | |
2134 | __xfs_btree_updkeys( | |
2135 | struct xfs_btree_cur *cur, | |
2136 | int level, | |
2137 | struct xfs_btree_block *block, | |
2138 | struct xfs_buf *bp0, | |
2139 | bool force_all) | |
2140 | { | |
45413937 | 2141 | union xfs_btree_key key; /* keys from current level */ |
13e831e0 DW |
2142 | union xfs_btree_key *lkey; /* keys from the next level up */ |
2143 | union xfs_btree_key *hkey; | |
2144 | union xfs_btree_key *nlkey; /* keys from the next level up */ | |
2145 | union xfs_btree_key *nhkey; | |
2146 | struct xfs_buf *bp; | |
2147 | int ptr; | |
2148 | ||
2149 | ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING); | |
2150 | ||
2151 | /* Exit if there aren't any parent levels to update. */ | |
2152 | if (level + 1 >= cur->bc_nlevels) | |
2153 | return 0; | |
2154 | ||
2155 | trace_xfs_btree_updkeys(cur, level, bp0); | |
2156 | ||
45413937 | 2157 | lkey = &key; |
13e831e0 DW |
2158 | hkey = xfs_btree_high_key_from_key(cur, lkey); |
2159 | xfs_btree_get_keys(cur, block, lkey); | |
2160 | for (level++; level < cur->bc_nlevels; level++) { | |
2161 | #ifdef DEBUG | |
2162 | int error; | |
2163 | #endif | |
2164 | block = xfs_btree_get_block(cur, level, &bp); | |
2165 | trace_xfs_btree_updkeys(cur, level, bp); | |
2166 | #ifdef DEBUG | |
2167 | error = xfs_btree_check_block(cur, block, level, bp); | |
2168 | if (error) { | |
2169 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
2170 | return error; | |
2171 | } | |
2172 | #endif | |
2173 | ptr = cur->bc_ptrs[level]; | |
2174 | nlkey = xfs_btree_key_addr(cur, ptr, block); | |
2175 | nhkey = xfs_btree_high_key_addr(cur, ptr, block); | |
2176 | if (!force_all && | |
2177 | !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 || | |
2178 | cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0)) | |
2179 | break; | |
2180 | xfs_btree_copy_keys(cur, nlkey, lkey, 1); | |
2181 | xfs_btree_log_keys(cur, bp, ptr, ptr); | |
2182 | if (level + 1 >= cur->bc_nlevels) | |
2183 | break; | |
64dbe047 | 2184 | xfs_btree_get_node_keys(cur, block, lkey); |
13e831e0 DW |
2185 | } |
2186 | ||
2187 | return 0; | |
2188 | } | |
2189 | ||
13e831e0 DW |
2190 | /* Update all the keys from some level in cursor back to the root. */ |
2191 | STATIC int | |
2192 | xfs_btree_updkeys_force( | |
2193 | struct xfs_btree_cur *cur, | |
2194 | int level) | |
2195 | { | |
2196 | struct xfs_buf *bp; | |
2197 | struct xfs_btree_block *block; | |
2198 | ||
2199 | block = xfs_btree_get_block(cur, level, &bp); | |
2200 | return __xfs_btree_updkeys(cur, level, block, bp, true); | |
a3c9cb10 DW |
2201 | } |
2202 | ||
2203 | /* | |
2204 | * Update the parent keys of the given level, progressing towards the root. | |
2205 | */ | |
64dbe047 | 2206 | STATIC int |
a3c9cb10 | 2207 | xfs_btree_update_keys( |
b194c7d8 | 2208 | struct xfs_btree_cur *cur, |
b194c7d8 BN |
2209 | int level) |
2210 | { | |
2211 | struct xfs_btree_block *block; | |
2212 | struct xfs_buf *bp; | |
2213 | union xfs_btree_key *kp; | |
a3c9cb10 | 2214 | union xfs_btree_key key; |
b194c7d8 BN |
2215 | int ptr; |
2216 | ||
64dbe047 DW |
2217 | ASSERT(level >= 0); |
2218 | ||
2219 | block = xfs_btree_get_block(cur, level, &bp); | |
2220 | if (cur->bc_flags & XFS_BTREE_OVERLAPPING) | |
2221 | return __xfs_btree_updkeys(cur, level, block, bp, false); | |
13e831e0 | 2222 | |
b194c7d8 BN |
2223 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); |
2224 | XFS_BTREE_TRACE_ARGIK(cur, level, keyp); | |
2225 | ||
b194c7d8 BN |
2226 | /* |
2227 | * Go up the tree from this level toward the root. | |
2228 | * At each level, update the key value to the value input. | |
2229 | * Stop when we reach a level where the cursor isn't pointing | |
2230 | * at the first entry in the block. | |
2231 | */ | |
a3c9cb10 DW |
2232 | xfs_btree_get_keys(cur, block, &key); |
2233 | for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) { | |
b194c7d8 BN |
2234 | #ifdef DEBUG |
2235 | int error; | |
2236 | #endif | |
2237 | block = xfs_btree_get_block(cur, level, &bp); | |
2238 | #ifdef DEBUG | |
2239 | error = xfs_btree_check_block(cur, block, level, bp); | |
2240 | if (error) { | |
2241 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
2242 | return error; | |
2243 | } | |
2244 | #endif | |
2245 | ptr = cur->bc_ptrs[level]; | |
2246 | kp = xfs_btree_key_addr(cur, ptr, block); | |
a3c9cb10 | 2247 | xfs_btree_copy_keys(cur, kp, &key, 1); |
b194c7d8 BN |
2248 | xfs_btree_log_keys(cur, bp, ptr, ptr); |
2249 | } | |
2250 | ||
2251 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2252 | return 0; | |
2253 | } | |
2254 | ||
2255 | /* | |
2256 | * Update the record referred to by cur to the value in the | |
2257 | * given record. This either works (return 0) or gets an | |
2258 | * EFSCORRUPTED error. | |
2259 | */ | |
2260 | int | |
2261 | xfs_btree_update( | |
2262 | struct xfs_btree_cur *cur, | |
2263 | union xfs_btree_rec *rec) | |
2264 | { | |
2265 | struct xfs_btree_block *block; | |
2266 | struct xfs_buf *bp; | |
2267 | int error; | |
2268 | int ptr; | |
2269 | union xfs_btree_rec *rp; | |
2270 | ||
2271 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
2272 | XFS_BTREE_TRACE_ARGR(cur, rec); | |
2273 | ||
2274 | /* Pick up the current block. */ | |
2275 | block = xfs_btree_get_block(cur, 0, &bp); | |
2276 | ||
2277 | #ifdef DEBUG | |
2278 | error = xfs_btree_check_block(cur, block, 0, bp); | |
2279 | if (error) | |
2280 | goto error0; | |
2281 | #endif | |
2282 | /* Get the address of the rec to be updated. */ | |
2283 | ptr = cur->bc_ptrs[0]; | |
2284 | rp = xfs_btree_rec_addr(cur, ptr, block); | |
2285 | ||
2286 | /* Fill in the new contents and log them. */ | |
2287 | xfs_btree_copy_recs(cur, rp, rec, 1); | |
2288 | xfs_btree_log_recs(cur, bp, ptr, ptr); | |
2289 | ||
2290 | /* | |
2291 | * If we are tracking the last record in the tree and | |
2292 | * we are at the far right edge of the tree, update it. | |
2293 | */ | |
2294 | if (xfs_btree_is_lastrec(cur, block, 0)) { | |
2295 | cur->bc_ops->update_lastrec(cur, block, rec, | |
2296 | ptr, LASTREC_UPDATE); | |
2297 | } | |
2298 | ||
13e831e0 | 2299 | /* Pass new key value up to our parent. */ |
a3c9cb10 | 2300 | if (xfs_btree_needs_key_update(cur, ptr)) { |
64dbe047 | 2301 | error = xfs_btree_update_keys(cur, 0); |
b194c7d8 BN |
2302 | if (error) |
2303 | goto error0; | |
2304 | } | |
2305 | ||
2306 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2307 | return 0; | |
2308 | ||
2309 | error0: | |
2310 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
2311 | return error; | |
2312 | } | |
2313 | ||
2314 | /* | |
2315 | * Move 1 record left from cur/level if possible. | |
2316 | * Update cur to reflect the new path. | |
2317 | */ | |
2318 | STATIC int /* error */ | |
2319 | xfs_btree_lshift( | |
2320 | struct xfs_btree_cur *cur, | |
2321 | int level, | |
2322 | int *stat) /* success/failure */ | |
2323 | { | |
b194c7d8 BN |
2324 | struct xfs_buf *lbp; /* left buffer pointer */ |
2325 | struct xfs_btree_block *left; /* left btree block */ | |
2326 | int lrecs; /* left record count */ | |
2327 | struct xfs_buf *rbp; /* right buffer pointer */ | |
2328 | struct xfs_btree_block *right; /* right btree block */ | |
13e831e0 | 2329 | struct xfs_btree_cur *tcur; /* temporary btree cursor */ |
b194c7d8 BN |
2330 | int rrecs; /* right record count */ |
2331 | union xfs_btree_ptr lptr; /* left btree pointer */ | |
2332 | union xfs_btree_key *rkp = NULL; /* right btree key */ | |
2333 | union xfs_btree_ptr *rpp = NULL; /* right address pointer */ | |
2334 | union xfs_btree_rec *rrp = NULL; /* right record pointer */ | |
2335 | int error; /* error return value */ | |
13e831e0 | 2336 | int i; |
b194c7d8 BN |
2337 | |
2338 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
2339 | XFS_BTREE_TRACE_ARGI(cur, level); | |
2340 | ||
2341 | if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) && | |
2342 | level == cur->bc_nlevels - 1) | |
2343 | goto out0; | |
2344 | ||
2345 | /* Set up variables for this block as "right". */ | |
2346 | right = xfs_btree_get_block(cur, level, &rbp); | |
2347 | ||
2348 | #ifdef DEBUG | |
2349 | error = xfs_btree_check_block(cur, right, level, rbp); | |
2350 | if (error) | |
2351 | goto error0; | |
2352 | #endif | |
2353 | ||
2354 | /* If we've got no left sibling then we can't shift an entry left. */ | |
2355 | xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB); | |
2356 | if (xfs_btree_ptr_is_null(cur, &lptr)) | |
2357 | goto out0; | |
2358 | ||
2359 | /* | |
2360 | * If the cursor entry is the one that would be moved, don't | |
2361 | * do it... it's too complicated. | |
2362 | */ | |
2363 | if (cur->bc_ptrs[level] <= 1) | |
2364 | goto out0; | |
2365 | ||
2366 | /* Set up the left neighbor as "left". */ | |
ff105f75 | 2367 | error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp); |
b194c7d8 BN |
2368 | if (error) |
2369 | goto error0; | |
2370 | ||
2371 | /* If it's full, it can't take another entry. */ | |
2372 | lrecs = xfs_btree_get_numrecs(left); | |
2373 | if (lrecs == cur->bc_ops->get_maxrecs(cur, level)) | |
2374 | goto out0; | |
2375 | ||
2376 | rrecs = xfs_btree_get_numrecs(right); | |
2377 | ||
2378 | /* | |
2379 | * We add one entry to the left side and remove one for the right side. | |
56b2de80 | 2380 | * Account for it here, the changes will be updated on disk and logged |
b194c7d8 BN |
2381 | * later. |
2382 | */ | |
2383 | lrecs++; | |
2384 | rrecs--; | |
2385 | ||
2386 | XFS_BTREE_STATS_INC(cur, lshift); | |
2387 | XFS_BTREE_STATS_ADD(cur, moves, 1); | |
2388 | ||
2389 | /* | |
2390 | * If non-leaf, copy a key and a ptr to the left block. | |
2391 | * Log the changes to the left block. | |
2392 | */ | |
2393 | if (level > 0) { | |
2394 | /* It's a non-leaf. Move keys and pointers. */ | |
2395 | union xfs_btree_key *lkp; /* left btree key */ | |
2396 | union xfs_btree_ptr *lpp; /* left address pointer */ | |
2397 | ||
2398 | lkp = xfs_btree_key_addr(cur, lrecs, left); | |
2399 | rkp = xfs_btree_key_addr(cur, 1, right); | |
2400 | ||
2401 | lpp = xfs_btree_ptr_addr(cur, lrecs, left); | |
2402 | rpp = xfs_btree_ptr_addr(cur, 1, right); | |
2403 | #ifdef DEBUG | |
2404 | error = xfs_btree_check_ptr(cur, rpp, 0, level); | |
2405 | if (error) | |
2406 | goto error0; | |
2407 | #endif | |
2408 | xfs_btree_copy_keys(cur, lkp, rkp, 1); | |
2409 | xfs_btree_copy_ptrs(cur, lpp, rpp, 1); | |
2410 | ||
2411 | xfs_btree_log_keys(cur, lbp, lrecs, lrecs); | |
2412 | xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs); | |
2413 | ||
2414 | ASSERT(cur->bc_ops->keys_inorder(cur, | |
2415 | xfs_btree_key_addr(cur, lrecs - 1, left), lkp)); | |
2416 | } else { | |
2417 | /* It's a leaf. Move records. */ | |
2418 | union xfs_btree_rec *lrp; /* left record pointer */ | |
2419 | ||
2420 | lrp = xfs_btree_rec_addr(cur, lrecs, left); | |
2421 | rrp = xfs_btree_rec_addr(cur, 1, right); | |
2422 | ||
2423 | xfs_btree_copy_recs(cur, lrp, rrp, 1); | |
2424 | xfs_btree_log_recs(cur, lbp, lrecs, lrecs); | |
2425 | ||
2426 | ASSERT(cur->bc_ops->recs_inorder(cur, | |
2427 | xfs_btree_rec_addr(cur, lrecs - 1, left), lrp)); | |
2428 | } | |
2429 | ||
2430 | xfs_btree_set_numrecs(left, lrecs); | |
2431 | xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS); | |
2432 | ||
2433 | xfs_btree_set_numrecs(right, rrecs); | |
2434 | xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS); | |
2435 | ||
2436 | /* | |
2437 | * Slide the contents of right down one entry. | |
2438 | */ | |
2439 | XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1); | |
2440 | if (level > 0) { | |
2441 | /* It's a nonleaf. operate on keys and ptrs */ | |
2442 | #ifdef DEBUG | |
2443 | int i; /* loop index */ | |
2444 | ||
2445 | for (i = 0; i < rrecs; i++) { | |
2446 | error = xfs_btree_check_ptr(cur, rpp, i + 1, level); | |
2447 | if (error) | |
2448 | goto error0; | |
2449 | } | |
2450 | #endif | |
2451 | xfs_btree_shift_keys(cur, | |
2452 | xfs_btree_key_addr(cur, 2, right), | |
2453 | -1, rrecs); | |
2454 | xfs_btree_shift_ptrs(cur, | |
2455 | xfs_btree_ptr_addr(cur, 2, right), | |
2456 | -1, rrecs); | |
2457 | ||
2458 | xfs_btree_log_keys(cur, rbp, 1, rrecs); | |
2459 | xfs_btree_log_ptrs(cur, rbp, 1, rrecs); | |
2460 | } else { | |
2461 | /* It's a leaf. operate on records */ | |
2462 | xfs_btree_shift_recs(cur, | |
2463 | xfs_btree_rec_addr(cur, 2, right), | |
2464 | -1, rrecs); | |
2465 | xfs_btree_log_recs(cur, rbp, 1, rrecs); | |
b194c7d8 BN |
2466 | } |
2467 | ||
13e831e0 DW |
2468 | /* |
2469 | * Using a temporary cursor, update the parent key values of the | |
2470 | * block on the left. | |
2471 | */ | |
e6358021 DW |
2472 | if (cur->bc_flags & XFS_BTREE_OVERLAPPING) { |
2473 | error = xfs_btree_dup_cursor(cur, &tcur); | |
2474 | if (error) | |
2475 | goto error0; | |
2476 | i = xfs_btree_firstrec(tcur, level); | |
2477 | XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0); | |
13e831e0 | 2478 | |
e6358021 DW |
2479 | error = xfs_btree_decrement(tcur, level, &i); |
2480 | if (error) | |
2481 | goto error1; | |
13e831e0 | 2482 | |
e6358021 | 2483 | /* Update the parent high keys of the left block, if needed. */ |
64dbe047 | 2484 | error = xfs_btree_update_keys(tcur, level); |
13e831e0 DW |
2485 | if (error) |
2486 | goto error1; | |
e6358021 DW |
2487 | |
2488 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
13e831e0 DW |
2489 | } |
2490 | ||
e6358021 DW |
2491 | /* Update the parent keys of the right block. */ |
2492 | error = xfs_btree_update_keys(cur, level); | |
2493 | if (error) | |
2494 | goto error0; | |
b194c7d8 BN |
2495 | |
2496 | /* Slide the cursor value left one. */ | |
2497 | cur->bc_ptrs[level]--; | |
2498 | ||
2499 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2500 | *stat = 1; | |
2501 | return 0; | |
2502 | ||
2503 | out0: | |
2504 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2505 | *stat = 0; | |
2506 | return 0; | |
2507 | ||
2508 | error0: | |
2509 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
2510 | return error; | |
13e831e0 DW |
2511 | |
2512 | error1: | |
2513 | XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR); | |
2514 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
2515 | return error; | |
b194c7d8 BN |
2516 | } |
2517 | ||
2518 | /* | |
2519 | * Move 1 record right from cur/level if possible. | |
2520 | * Update cur to reflect the new path. | |
2521 | */ | |
2522 | STATIC int /* error */ | |
2523 | xfs_btree_rshift( | |
2524 | struct xfs_btree_cur *cur, | |
2525 | int level, | |
2526 | int *stat) /* success/failure */ | |
2527 | { | |
b194c7d8 BN |
2528 | struct xfs_buf *lbp; /* left buffer pointer */ |
2529 | struct xfs_btree_block *left; /* left btree block */ | |
2530 | struct xfs_buf *rbp; /* right buffer pointer */ | |
2531 | struct xfs_btree_block *right; /* right btree block */ | |
2532 | struct xfs_btree_cur *tcur; /* temporary btree cursor */ | |
2533 | union xfs_btree_ptr rptr; /* right block pointer */ | |
2534 | union xfs_btree_key *rkp; /* right btree key */ | |
2535 | int rrecs; /* right record count */ | |
2536 | int lrecs; /* left record count */ | |
2537 | int error; /* error return value */ | |
2538 | int i; /* loop counter */ | |
2539 | ||
2540 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
2541 | XFS_BTREE_TRACE_ARGI(cur, level); | |
2542 | ||
2543 | if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) && | |
2544 | (level == cur->bc_nlevels - 1)) | |
2545 | goto out0; | |
2546 | ||
2547 | /* Set up variables for this block as "left". */ | |
2548 | left = xfs_btree_get_block(cur, level, &lbp); | |
2549 | ||
2550 | #ifdef DEBUG | |
2551 | error = xfs_btree_check_block(cur, left, level, lbp); | |
2552 | if (error) | |
2553 | goto error0; | |
2554 | #endif | |
2555 | ||
2556 | /* If we've got no right sibling then we can't shift an entry right. */ | |
2557 | xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB); | |
2558 | if (xfs_btree_ptr_is_null(cur, &rptr)) | |
2559 | goto out0; | |
2560 | ||
2561 | /* | |
2562 | * If the cursor entry is the one that would be moved, don't | |
2563 | * do it... it's too complicated. | |
2564 | */ | |
2565 | lrecs = xfs_btree_get_numrecs(left); | |
2566 | if (cur->bc_ptrs[level] >= lrecs) | |
2567 | goto out0; | |
2568 | ||
2569 | /* Set up the right neighbor as "right". */ | |
ff105f75 | 2570 | error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp); |
b194c7d8 BN |
2571 | if (error) |
2572 | goto error0; | |
2573 | ||
2574 | /* If it's full, it can't take another entry. */ | |
2575 | rrecs = xfs_btree_get_numrecs(right); | |
2576 | if (rrecs == cur->bc_ops->get_maxrecs(cur, level)) | |
2577 | goto out0; | |
2578 | ||
2579 | XFS_BTREE_STATS_INC(cur, rshift); | |
2580 | XFS_BTREE_STATS_ADD(cur, moves, rrecs); | |
2581 | ||
2582 | /* | |
2583 | * Make a hole at the start of the right neighbor block, then | |
2584 | * copy the last left block entry to the hole. | |
2585 | */ | |
2586 | if (level > 0) { | |
2587 | /* It's a nonleaf. make a hole in the keys and ptrs */ | |
2588 | union xfs_btree_key *lkp; | |
2589 | union xfs_btree_ptr *lpp; | |
2590 | union xfs_btree_ptr *rpp; | |
2591 | ||
2592 | lkp = xfs_btree_key_addr(cur, lrecs, left); | |
2593 | lpp = xfs_btree_ptr_addr(cur, lrecs, left); | |
2594 | rkp = xfs_btree_key_addr(cur, 1, right); | |
2595 | rpp = xfs_btree_ptr_addr(cur, 1, right); | |
2596 | ||
2597 | #ifdef DEBUG | |
2598 | for (i = rrecs - 1; i >= 0; i--) { | |
2599 | error = xfs_btree_check_ptr(cur, rpp, i, level); | |
2600 | if (error) | |
2601 | goto error0; | |
2602 | } | |
2603 | #endif | |
2604 | ||
2605 | xfs_btree_shift_keys(cur, rkp, 1, rrecs); | |
2606 | xfs_btree_shift_ptrs(cur, rpp, 1, rrecs); | |
2607 | ||
2608 | #ifdef DEBUG | |
2609 | error = xfs_btree_check_ptr(cur, lpp, 0, level); | |
2610 | if (error) | |
2611 | goto error0; | |
2612 | #endif | |
2613 | ||
2614 | /* Now put the new data in, and log it. */ | |
2615 | xfs_btree_copy_keys(cur, rkp, lkp, 1); | |
2616 | xfs_btree_copy_ptrs(cur, rpp, lpp, 1); | |
2617 | ||
2618 | xfs_btree_log_keys(cur, rbp, 1, rrecs + 1); | |
2619 | xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1); | |
2620 | ||
2621 | ASSERT(cur->bc_ops->keys_inorder(cur, rkp, | |
2622 | xfs_btree_key_addr(cur, 2, right))); | |
2623 | } else { | |
2624 | /* It's a leaf. make a hole in the records */ | |
2625 | union xfs_btree_rec *lrp; | |
2626 | union xfs_btree_rec *rrp; | |
2627 | ||
2628 | lrp = xfs_btree_rec_addr(cur, lrecs, left); | |
2629 | rrp = xfs_btree_rec_addr(cur, 1, right); | |
2630 | ||
2631 | xfs_btree_shift_recs(cur, rrp, 1, rrecs); | |
2632 | ||
2633 | /* Now put the new data in, and log it. */ | |
2634 | xfs_btree_copy_recs(cur, rrp, lrp, 1); | |
2635 | xfs_btree_log_recs(cur, rbp, 1, rrecs + 1); | |
b194c7d8 BN |
2636 | } |
2637 | ||
2638 | /* | |
2639 | * Decrement and log left's numrecs, bump and log right's numrecs. | |
2640 | */ | |
2641 | xfs_btree_set_numrecs(left, --lrecs); | |
2642 | xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS); | |
2643 | ||
2644 | xfs_btree_set_numrecs(right, ++rrecs); | |
2645 | xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS); | |
2646 | ||
2647 | /* | |
2648 | * Using a temporary cursor, update the parent key values of the | |
2649 | * block on the right. | |
2650 | */ | |
2651 | error = xfs_btree_dup_cursor(cur, &tcur); | |
2652 | if (error) | |
2653 | goto error0; | |
2654 | i = xfs_btree_lastrec(tcur, level); | |
e6358021 | 2655 | XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0); |
b194c7d8 BN |
2656 | |
2657 | error = xfs_btree_increment(tcur, level, &i); | |
2658 | if (error) | |
2659 | goto error1; | |
2660 | ||
13e831e0 DW |
2661 | /* Update the parent high keys of the left block, if needed. */ |
2662 | if (cur->bc_flags & XFS_BTREE_OVERLAPPING) { | |
64dbe047 | 2663 | error = xfs_btree_update_keys(cur, level); |
13e831e0 DW |
2664 | if (error) |
2665 | goto error1; | |
2666 | } | |
2667 | ||
a3c9cb10 | 2668 | /* Update the parent keys of the right block. */ |
64dbe047 | 2669 | error = xfs_btree_update_keys(tcur, level); |
b194c7d8 BN |
2670 | if (error) |
2671 | goto error1; | |
2672 | ||
2673 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
2674 | ||
2675 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2676 | *stat = 1; | |
2677 | return 0; | |
2678 | ||
2679 | out0: | |
2680 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2681 | *stat = 0; | |
2682 | return 0; | |
2683 | ||
2684 | error0: | |
2685 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
2686 | return error; | |
2687 | ||
2688 | error1: | |
2689 | XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR); | |
2690 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
2691 | return error; | |
2692 | } | |
2693 | ||
2694 | /* | |
2695 | * Split cur/level block in half. | |
2696 | * Return new block number and the key to its first | |
2697 | * record (to be inserted into parent). | |
2698 | */ | |
2699 | STATIC int /* error */ | |
ff105f75 | 2700 | __xfs_btree_split( |
b194c7d8 BN |
2701 | struct xfs_btree_cur *cur, |
2702 | int level, | |
2703 | union xfs_btree_ptr *ptrp, | |
2704 | union xfs_btree_key *key, | |
2705 | struct xfs_btree_cur **curp, | |
2706 | int *stat) /* success/failure */ | |
2707 | { | |
2708 | union xfs_btree_ptr lptr; /* left sibling block ptr */ | |
2709 | struct xfs_buf *lbp; /* left buffer pointer */ | |
2710 | struct xfs_btree_block *left; /* left btree block */ | |
2711 | union xfs_btree_ptr rptr; /* right sibling block ptr */ | |
2712 | struct xfs_buf *rbp; /* right buffer pointer */ | |
2713 | struct xfs_btree_block *right; /* right btree block */ | |
2714 | union xfs_btree_ptr rrptr; /* right-right sibling ptr */ | |
2715 | struct xfs_buf *rrbp; /* right-right buffer pointer */ | |
2716 | struct xfs_btree_block *rrblock; /* right-right btree block */ | |
2717 | int lrecs; | |
2718 | int rrecs; | |
2719 | int src_index; | |
2720 | int error; /* error return value */ | |
2721 | #ifdef DEBUG | |
2722 | int i; | |
2723 | #endif | |
2724 | ||
2725 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
2726 | XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key); | |
2727 | ||
2728 | XFS_BTREE_STATS_INC(cur, split); | |
2729 | ||
2730 | /* Set up left block (current one). */ | |
2731 | left = xfs_btree_get_block(cur, level, &lbp); | |
2732 | ||
2733 | #ifdef DEBUG | |
2734 | error = xfs_btree_check_block(cur, left, level, lbp); | |
2735 | if (error) | |
2736 | goto error0; | |
2737 | #endif | |
2738 | ||
2739 | xfs_btree_buf_to_ptr(cur, lbp, &lptr); | |
2740 | ||
2741 | /* Allocate the new block. If we can't do it, we're toast. Give up. */ | |
ff105f75 | 2742 | error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat); |
b194c7d8 BN |
2743 | if (error) |
2744 | goto error0; | |
2745 | if (*stat == 0) | |
2746 | goto out0; | |
2747 | XFS_BTREE_STATS_INC(cur, alloc); | |
2748 | ||
2749 | /* Set up the new block as "right". */ | |
2750 | error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp); | |
2751 | if (error) | |
2752 | goto error0; | |
2753 | ||
2754 | /* Fill in the btree header for the new right block. */ | |
5dfa5cd2 | 2755 | xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0); |
b194c7d8 BN |
2756 | |
2757 | /* | |
2758 | * Split the entries between the old and the new block evenly. | |
2759 | * Make sure that if there's an odd number of entries now, that | |
2760 | * each new block will have the same number of entries. | |
2761 | */ | |
2762 | lrecs = xfs_btree_get_numrecs(left); | |
2763 | rrecs = lrecs / 2; | |
2764 | if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1) | |
2765 | rrecs++; | |
2766 | src_index = (lrecs - rrecs + 1); | |
2767 | ||
2768 | XFS_BTREE_STATS_ADD(cur, moves, rrecs); | |
2769 | ||
a3c9cb10 DW |
2770 | /* Adjust numrecs for the later get_*_keys() calls. */ |
2771 | lrecs -= rrecs; | |
2772 | xfs_btree_set_numrecs(left, lrecs); | |
2773 | xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs); | |
2774 | ||
b194c7d8 BN |
2775 | /* |
2776 | * Copy btree block entries from the left block over to the | |
2777 | * new block, the right. Update the right block and log the | |
2778 | * changes. | |
2779 | */ | |
2780 | if (level > 0) { | |
2781 | /* It's a non-leaf. Move keys and pointers. */ | |
2782 | union xfs_btree_key *lkp; /* left btree key */ | |
2783 | union xfs_btree_ptr *lpp; /* left address pointer */ | |
2784 | union xfs_btree_key *rkp; /* right btree key */ | |
2785 | union xfs_btree_ptr *rpp; /* right address pointer */ | |
2786 | ||
2787 | lkp = xfs_btree_key_addr(cur, src_index, left); | |
2788 | lpp = xfs_btree_ptr_addr(cur, src_index, left); | |
2789 | rkp = xfs_btree_key_addr(cur, 1, right); | |
2790 | rpp = xfs_btree_ptr_addr(cur, 1, right); | |
2791 | ||
2792 | #ifdef DEBUG | |
2793 | for (i = src_index; i < rrecs; i++) { | |
2794 | error = xfs_btree_check_ptr(cur, lpp, i, level); | |
2795 | if (error) | |
2796 | goto error0; | |
2797 | } | |
2798 | #endif | |
2799 | ||
a3c9cb10 | 2800 | /* Copy the keys & pointers to the new block. */ |
b194c7d8 BN |
2801 | xfs_btree_copy_keys(cur, rkp, lkp, rrecs); |
2802 | xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs); | |
2803 | ||
2804 | xfs_btree_log_keys(cur, rbp, 1, rrecs); | |
2805 | xfs_btree_log_ptrs(cur, rbp, 1, rrecs); | |
2806 | ||
a3c9cb10 | 2807 | /* Stash the keys of the new block for later insertion. */ |
64dbe047 | 2808 | xfs_btree_get_node_keys(cur, right, key); |
b194c7d8 BN |
2809 | } else { |
2810 | /* It's a leaf. Move records. */ | |
2811 | union xfs_btree_rec *lrp; /* left record pointer */ | |
2812 | union xfs_btree_rec *rrp; /* right record pointer */ | |
2813 | ||
2814 | lrp = xfs_btree_rec_addr(cur, src_index, left); | |
2815 | rrp = xfs_btree_rec_addr(cur, 1, right); | |
2816 | ||
a3c9cb10 | 2817 | /* Copy records to the new block. */ |
b194c7d8 BN |
2818 | xfs_btree_copy_recs(cur, rrp, lrp, rrecs); |
2819 | xfs_btree_log_recs(cur, rbp, 1, rrecs); | |
2820 | ||
a3c9cb10 | 2821 | /* Stash the keys of the new block for later insertion. */ |
64dbe047 | 2822 | xfs_btree_get_leaf_keys(cur, right, key); |
b194c7d8 BN |
2823 | } |
2824 | ||
b194c7d8 BN |
2825 | /* |
2826 | * Find the left block number by looking in the buffer. | |
a3c9cb10 | 2827 | * Adjust sibling pointers. |
b194c7d8 BN |
2828 | */ |
2829 | xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB); | |
2830 | xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB); | |
2831 | xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB); | |
2832 | xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB); | |
2833 | ||
b194c7d8 BN |
2834 | xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS); |
2835 | xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB); | |
2836 | ||
2837 | /* | |
2838 | * If there's a block to the new block's right, make that block | |
2839 | * point back to right instead of to left. | |
2840 | */ | |
2841 | if (!xfs_btree_ptr_is_null(cur, &rrptr)) { | |
ff105f75 | 2842 | error = xfs_btree_read_buf_block(cur, &rrptr, |
b194c7d8 BN |
2843 | 0, &rrblock, &rrbp); |
2844 | if (error) | |
2845 | goto error0; | |
2846 | xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB); | |
2847 | xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB); | |
2848 | } | |
13e831e0 DW |
2849 | |
2850 | /* Update the parent high keys of the left block, if needed. */ | |
2851 | if (cur->bc_flags & XFS_BTREE_OVERLAPPING) { | |
64dbe047 | 2852 | error = xfs_btree_update_keys(cur, level); |
13e831e0 DW |
2853 | if (error) |
2854 | goto error0; | |
2855 | } | |
2856 | ||
b194c7d8 BN |
2857 | /* |
2858 | * If the cursor is really in the right block, move it there. | |
2859 | * If it's just pointing past the last entry in left, then we'll | |
2860 | * insert there, so don't change anything in that case. | |
2861 | */ | |
2862 | if (cur->bc_ptrs[level] > lrecs + 1) { | |
2863 | xfs_btree_setbuf(cur, level, rbp); | |
2864 | cur->bc_ptrs[level] -= lrecs; | |
2865 | } | |
2866 | /* | |
2867 | * If there are more levels, we'll need another cursor which refers | |
2868 | * the right block, no matter where this cursor was. | |
2869 | */ | |
2870 | if (level + 1 < cur->bc_nlevels) { | |
2871 | error = xfs_btree_dup_cursor(cur, curp); | |
2872 | if (error) | |
2873 | goto error0; | |
2874 | (*curp)->bc_ptrs[level + 1]++; | |
2875 | } | |
2876 | *ptrp = rptr; | |
2877 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2878 | *stat = 1; | |
2879 | return 0; | |
2880 | out0: | |
2881 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
2882 | *stat = 0; | |
2883 | return 0; | |
2884 | ||
2885 | error0: | |
2886 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
2887 | return error; | |
2888 | } | |
2889 | ||
19ebedcf | 2890 | #ifdef __KERNEL__ |
ff105f75 DC |
2891 | struct xfs_btree_split_args { |
2892 | struct xfs_btree_cur *cur; | |
2893 | int level; | |
2894 | union xfs_btree_ptr *ptrp; | |
2895 | union xfs_btree_key *key; | |
2896 | struct xfs_btree_cur **curp; | |
2897 | int *stat; /* success/failure */ | |
2898 | int result; | |
2899 | bool kswapd; /* allocation in kswapd context */ | |
2900 | struct completion *done; | |
2901 | struct work_struct work; | |
2902 | }; | |
2903 | ||
2904 | /* | |
2905 | * Stack switching interfaces for allocation | |
2906 | */ | |
2907 | static void | |
2908 | xfs_btree_split_worker( | |
2909 | struct work_struct *work) | |
2910 | { | |
2911 | struct xfs_btree_split_args *args = container_of(work, | |
2912 | struct xfs_btree_split_args, work); | |
2913 | unsigned long pflags; | |
b455713b | 2914 | unsigned long new_pflags = PF_MEMALLOC_NOFS; |
ff105f75 DC |
2915 | |
2916 | /* | |
2917 | * we are in a transaction context here, but may also be doing work | |
2918 | * in kswapd context, and hence we may need to inherit that state | |
2919 | * temporarily to ensure that we don't block waiting for memory reclaim | |
2920 | * in any way. | |
2921 | */ | |
2922 | if (args->kswapd) | |
2923 | new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; | |
2924 | ||
2925 | current_set_flags_nested(&pflags, new_pflags); | |
2926 | ||
2927 | args->result = __xfs_btree_split(args->cur, args->level, args->ptrp, | |
2928 | args->key, args->curp, args->stat); | |
2929 | complete(args->done); | |
2930 | ||
2931 | current_restore_flags_nested(&pflags, new_pflags); | |
2932 | } | |
ff105f75 DC |
2933 | |
2934 | /* | |
2935 | * BMBT split requests often come in with little stack to work on. Push | |
2936 | * them off to a worker thread so there is lots of stack to use. For the other | |
2937 | * btree types, just call directly to avoid the context switch overhead here. | |
2938 | */ | |
2939 | STATIC int /* error */ | |
2940 | xfs_btree_split( | |
2941 | struct xfs_btree_cur *cur, | |
2942 | int level, | |
2943 | union xfs_btree_ptr *ptrp, | |
2944 | union xfs_btree_key *key, | |
2945 | struct xfs_btree_cur **curp, | |
2946 | int *stat) /* success/failure */ | |
2947 | { | |
ff105f75 DC |
2948 | struct xfs_btree_split_args args; |
2949 | DECLARE_COMPLETION_ONSTACK(done); | |
2950 | ||
2951 | if (cur->bc_btnum != XFS_BTNUM_BMAP) | |
ff105f75 DC |
2952 | return __xfs_btree_split(cur, level, ptrp, key, curp, stat); |
2953 | ||
ff105f75 DC |
2954 | args.cur = cur; |
2955 | args.level = level; | |
2956 | args.ptrp = ptrp; | |
2957 | args.key = key; | |
2958 | args.curp = curp; | |
2959 | args.stat = stat; | |
2960 | args.done = &done; | |
2961 | args.kswapd = current_is_kswapd(); | |
2962 | INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker); | |
2963 | queue_work(xfs_alloc_wq, &args.work); | |
2964 | wait_for_completion(&done); | |
2965 | destroy_work_on_stack(&args.work); | |
2966 | return args.result; | |
ff105f75 | 2967 | } |
19ebedcf DC |
2968 | #else /* !KERNEL */ |
2969 | #define xfs_btree_split __xfs_btree_split | |
2970 | #endif | |
ff105f75 DC |
2971 | |
2972 | ||
b194c7d8 BN |
2973 | /* |
2974 | * Copy the old inode root contents into a real block and make the | |
2975 | * broot point to it. | |
2976 | */ | |
2977 | int /* error */ | |
2978 | xfs_btree_new_iroot( | |
2979 | struct xfs_btree_cur *cur, /* btree cursor */ | |
2980 | int *logflags, /* logging flags for inode */ | |
2981 | int *stat) /* return status - 0 fail */ | |
2982 | { | |
2983 | struct xfs_buf *cbp; /* buffer for cblock */ | |
2984 | struct xfs_btree_block *block; /* btree block */ | |
2985 | struct xfs_btree_block *cblock; /* child btree block */ | |
2986 | union xfs_btree_key *ckp; /* child key pointer */ | |
2987 | union xfs_btree_ptr *cpp; /* child ptr pointer */ | |
2988 | union xfs_btree_key *kp; /* pointer to btree key */ | |
2989 | union xfs_btree_ptr *pp; /* pointer to block addr */ | |
2990 | union xfs_btree_ptr nptr; /* new block addr */ | |
2991 | int level; /* btree level */ | |
2992 | int error; /* error return code */ | |
2993 | #ifdef DEBUG | |
2994 | int i; /* loop counter */ | |
2995 | #endif | |
2996 | ||
2997 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
2998 | XFS_BTREE_STATS_INC(cur, newroot); | |
2999 | ||
3000 | ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE); | |
3001 | ||
3002 | level = cur->bc_nlevels - 1; | |
3003 | ||
3004 | block = xfs_btree_get_iroot(cur); | |
3005 | pp = xfs_btree_ptr_addr(cur, 1, block); | |
3006 | ||
3007 | /* Allocate the new block. If we can't do it, we're toast. Give up. */ | |
ff105f75 | 3008 | error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat); |
b194c7d8 BN |
3009 | if (error) |
3010 | goto error0; | |
3011 | if (*stat == 0) { | |
3012 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3013 | return 0; | |
3014 | } | |
3015 | XFS_BTREE_STATS_INC(cur, alloc); | |
3016 | ||
3017 | /* Copy the root into a real block. */ | |
3018 | error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp); | |
3019 | if (error) | |
3020 | goto error0; | |
3021 | ||
77ec5ff4 DC |
3022 | /* |
3023 | * we can't just memcpy() the root in for CRC enabled btree blocks. | |
3024 | * In that case have to also ensure the blkno remains correct | |
3025 | */ | |
b194c7d8 | 3026 | memcpy(cblock, block, xfs_btree_block_len(cur)); |
77ec5ff4 DC |
3027 | if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) { |
3028 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) | |
3029 | cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn); | |
3030 | else | |
3031 | cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn); | |
3032 | } | |
b194c7d8 BN |
3033 | |
3034 | be16_add_cpu(&block->bb_level, 1); | |
3035 | xfs_btree_set_numrecs(block, 1); | |
3036 | cur->bc_nlevels++; | |
3037 | cur->bc_ptrs[level + 1] = 1; | |
3038 | ||
3039 | kp = xfs_btree_key_addr(cur, 1, block); | |
3040 | ckp = xfs_btree_key_addr(cur, 1, cblock); | |
3041 | xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock)); | |
3042 | ||
3043 | cpp = xfs_btree_ptr_addr(cur, 1, cblock); | |
3044 | #ifdef DEBUG | |
3045 | for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) { | |
3046 | error = xfs_btree_check_ptr(cur, pp, i, level); | |
3047 | if (error) | |
3048 | goto error0; | |
3049 | } | |
3050 | #endif | |
3051 | xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock)); | |
3052 | ||
3053 | #ifdef DEBUG | |
3054 | error = xfs_btree_check_ptr(cur, &nptr, 0, level); | |
3055 | if (error) | |
3056 | goto error0; | |
3057 | #endif | |
3058 | xfs_btree_copy_ptrs(cur, pp, &nptr, 1); | |
3059 | ||
3060 | xfs_iroot_realloc(cur->bc_private.b.ip, | |
3061 | 1 - xfs_btree_get_numrecs(cblock), | |
3062 | cur->bc_private.b.whichfork); | |
3063 | ||
3064 | xfs_btree_setbuf(cur, level, cbp); | |
3065 | ||
3066 | /* | |
3067 | * Do all this logging at the end so that | |
3068 | * the root is at the right level. | |
3069 | */ | |
3070 | xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS); | |
3071 | xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs)); | |
3072 | xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs)); | |
3073 | ||
3074 | *logflags |= | |
56b2de80 | 3075 | XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork); |
b194c7d8 BN |
3076 | *stat = 1; |
3077 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3078 | return 0; | |
3079 | error0: | |
3080 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
3081 | return error; | |
3082 | } | |
3083 | ||
3084 | /* | |
3085 | * Allocate a new root block, fill it in. | |
3086 | */ | |
3087 | STATIC int /* error */ | |
3088 | xfs_btree_new_root( | |
3089 | struct xfs_btree_cur *cur, /* btree cursor */ | |
3090 | int *stat) /* success/failure */ | |
3091 | { | |
3092 | struct xfs_btree_block *block; /* one half of the old root block */ | |
3093 | struct xfs_buf *bp; /* buffer containing block */ | |
3094 | int error; /* error return value */ | |
3095 | struct xfs_buf *lbp; /* left buffer pointer */ | |
3096 | struct xfs_btree_block *left; /* left btree block */ | |
3097 | struct xfs_buf *nbp; /* new (root) buffer */ | |
3098 | struct xfs_btree_block *new; /* new (root) btree block */ | |
3099 | int nptr; /* new value for key index, 1 or 2 */ | |
3100 | struct xfs_buf *rbp; /* right buffer pointer */ | |
3101 | struct xfs_btree_block *right; /* right btree block */ | |
3102 | union xfs_btree_ptr rptr; | |
3103 | union xfs_btree_ptr lptr; | |
3104 | ||
3105 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
3106 | XFS_BTREE_STATS_INC(cur, newroot); | |
3107 | ||
3108 | /* initialise our start point from the cursor */ | |
3109 | cur->bc_ops->init_ptr_from_cur(cur, &rptr); | |
3110 | ||
3111 | /* Allocate the new block. If we can't do it, we're toast. Give up. */ | |
ff105f75 | 3112 | error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat); |
b194c7d8 BN |
3113 | if (error) |
3114 | goto error0; | |
3115 | if (*stat == 0) | |
3116 | goto out0; | |
3117 | XFS_BTREE_STATS_INC(cur, alloc); | |
3118 | ||
3119 | /* Set up the new block. */ | |
3120 | error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp); | |
3121 | if (error) | |
3122 | goto error0; | |
3123 | ||
3124 | /* Set the root in the holding structure increasing the level by 1. */ | |
3125 | cur->bc_ops->set_root(cur, &lptr, 1); | |
3126 | ||
3127 | /* | |
3128 | * At the previous root level there are now two blocks: the old root, | |
3129 | * and the new block generated when it was split. We don't know which | |
3130 | * one the cursor is pointing at, so we set up variables "left" and | |
3131 | * "right" for each case. | |
3132 | */ | |
3133 | block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp); | |
3134 | ||
3135 | #ifdef DEBUG | |
3136 | error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp); | |
3137 | if (error) | |
3138 | goto error0; | |
3139 | #endif | |
3140 | ||
3141 | xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB); | |
3142 | if (!xfs_btree_ptr_is_null(cur, &rptr)) { | |
3143 | /* Our block is left, pick up the right block. */ | |
3144 | lbp = bp; | |
3145 | xfs_btree_buf_to_ptr(cur, lbp, &lptr); | |
3146 | left = block; | |
ff105f75 | 3147 | error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp); |
b194c7d8 BN |
3148 | if (error) |
3149 | goto error0; | |
3150 | bp = rbp; | |
3151 | nptr = 1; | |
3152 | } else { | |
3153 | /* Our block is right, pick up the left block. */ | |
3154 | rbp = bp; | |
3155 | xfs_btree_buf_to_ptr(cur, rbp, &rptr); | |
3156 | right = block; | |
3157 | xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB); | |
ff105f75 | 3158 | error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp); |
b194c7d8 BN |
3159 | if (error) |
3160 | goto error0; | |
3161 | bp = lbp; | |
3162 | nptr = 2; | |
3163 | } | |
a3c9cb10 | 3164 | |
b194c7d8 | 3165 | /* Fill in the new block's btree header and log it. */ |
5dfa5cd2 | 3166 | xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2); |
b194c7d8 BN |
3167 | xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS); |
3168 | ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) && | |
3169 | !xfs_btree_ptr_is_null(cur, &rptr)); | |
3170 | ||
3171 | /* Fill in the key data in the new root. */ | |
3172 | if (xfs_btree_get_level(left) > 0) { | |
a3c9cb10 DW |
3173 | /* |
3174 | * Get the keys for the left block's keys and put them directly | |
3175 | * in the parent block. Do the same for the right block. | |
3176 | */ | |
64dbe047 | 3177 | xfs_btree_get_node_keys(cur, left, |
a3c9cb10 | 3178 | xfs_btree_key_addr(cur, 1, new)); |
64dbe047 | 3179 | xfs_btree_get_node_keys(cur, right, |
a3c9cb10 | 3180 | xfs_btree_key_addr(cur, 2, new)); |
b194c7d8 | 3181 | } else { |
a3c9cb10 DW |
3182 | /* |
3183 | * Get the keys for the left block's records and put them | |
3184 | * directly in the parent block. Do the same for the right | |
3185 | * block. | |
3186 | */ | |
64dbe047 | 3187 | xfs_btree_get_leaf_keys(cur, left, |
a3c9cb10 | 3188 | xfs_btree_key_addr(cur, 1, new)); |
64dbe047 | 3189 | xfs_btree_get_leaf_keys(cur, right, |
a3c9cb10 | 3190 | xfs_btree_key_addr(cur, 2, new)); |
b194c7d8 BN |
3191 | } |
3192 | xfs_btree_log_keys(cur, nbp, 1, 2); | |
3193 | ||
3194 | /* Fill in the pointer data in the new root. */ | |
3195 | xfs_btree_copy_ptrs(cur, | |
3196 | xfs_btree_ptr_addr(cur, 1, new), &lptr, 1); | |
3197 | xfs_btree_copy_ptrs(cur, | |
3198 | xfs_btree_ptr_addr(cur, 2, new), &rptr, 1); | |
3199 | xfs_btree_log_ptrs(cur, nbp, 1, 2); | |
3200 | ||
3201 | /* Fix up the cursor. */ | |
3202 | xfs_btree_setbuf(cur, cur->bc_nlevels, nbp); | |
3203 | cur->bc_ptrs[cur->bc_nlevels] = nptr; | |
3204 | cur->bc_nlevels++; | |
3205 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3206 | *stat = 1; | |
3207 | return 0; | |
3208 | error0: | |
3209 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
3210 | return error; | |
3211 | out0: | |
3212 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3213 | *stat = 0; | |
3214 | return 0; | |
3215 | } | |
3216 | ||
3217 | STATIC int | |
3218 | xfs_btree_make_block_unfull( | |
3219 | struct xfs_btree_cur *cur, /* btree cursor */ | |
3220 | int level, /* btree level */ | |
3221 | int numrecs,/* # of recs in block */ | |
3222 | int *oindex,/* old tree index */ | |
3223 | int *index, /* new tree index */ | |
3224 | union xfs_btree_ptr *nptr, /* new btree ptr */ | |
3225 | struct xfs_btree_cur **ncur, /* new btree cursor */ | |
a3c9cb10 | 3226 | union xfs_btree_key *key, /* key of new block */ |
b194c7d8 BN |
3227 | int *stat) |
3228 | { | |
b194c7d8 BN |
3229 | int error = 0; |
3230 | ||
3231 | if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) && | |
3232 | level == cur->bc_nlevels - 1) { | |
3233 | struct xfs_inode *ip = cur->bc_private.b.ip; | |
3234 | ||
3235 | if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) { | |
3236 | /* A root block that can be made bigger. */ | |
b194c7d8 | 3237 | xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork); |
cff8bf94 | 3238 | *stat = 1; |
b194c7d8 BN |
3239 | } else { |
3240 | /* A root block that needs replacing */ | |
3241 | int logflags = 0; | |
3242 | ||
3243 | error = xfs_btree_new_iroot(cur, &logflags, stat); | |
3244 | if (error || *stat == 0) | |
3245 | return error; | |
3246 | ||
3247 | xfs_trans_log_inode(cur->bc_tp, ip, logflags); | |
3248 | } | |
3249 | ||
3250 | return 0; | |
3251 | } | |
3252 | ||
3253 | /* First, try shifting an entry to the right neighbor. */ | |
3254 | error = xfs_btree_rshift(cur, level, stat); | |
3255 | if (error || *stat) | |
3256 | return error; | |
3257 | ||
3258 | /* Next, try shifting an entry to the left neighbor. */ | |
3259 | error = xfs_btree_lshift(cur, level, stat); | |
3260 | if (error) | |
3261 | return error; | |
3262 | ||
3263 | if (*stat) { | |
3264 | *oindex = *index = cur->bc_ptrs[level]; | |
3265 | return 0; | |
3266 | } | |
3267 | ||
3268 | /* | |
3269 | * Next, try splitting the current block in half. | |
3270 | * | |
3271 | * If this works we have to re-set our variables because we | |
3272 | * could be in a different block now. | |
3273 | */ | |
d3cd7a27 | 3274 | error = xfs_btree_split(cur, level, nptr, key, ncur, stat); |
b194c7d8 BN |
3275 | if (error || *stat == 0) |
3276 | return error; | |
3277 | ||
3278 | ||
3279 | *index = cur->bc_ptrs[level]; | |
b194c7d8 BN |
3280 | return 0; |
3281 | } | |
3282 | ||
3283 | /* | |
3284 | * Insert one record/level. Return information to the caller | |
3285 | * allowing the next level up to proceed if necessary. | |
3286 | */ | |
3287 | STATIC int | |
3288 | xfs_btree_insrec( | |
3289 | struct xfs_btree_cur *cur, /* btree cursor */ | |
3290 | int level, /* level to insert record at */ | |
3291 | union xfs_btree_ptr *ptrp, /* i/o: block number inserted */ | |
d3cd7a27 DW |
3292 | union xfs_btree_rec *rec, /* record to insert */ |
3293 | union xfs_btree_key *key, /* i/o: block key for ptrp */ | |
b194c7d8 BN |
3294 | struct xfs_btree_cur **curp, /* output: new cursor replacing cur */ |
3295 | int *stat) /* success/failure */ | |
3296 | { | |
3297 | struct xfs_btree_block *block; /* btree block */ | |
3298 | struct xfs_buf *bp; /* buffer for block */ | |
b194c7d8 BN |
3299 | union xfs_btree_ptr nptr; /* new block ptr */ |
3300 | struct xfs_btree_cur *ncur; /* new btree cursor */ | |
45413937 | 3301 | union xfs_btree_key nkey; /* new block key */ |
13e831e0 | 3302 | union xfs_btree_key *lkey; |
b194c7d8 BN |
3303 | int optr; /* old key/record index */ |
3304 | int ptr; /* key/record index */ | |
3305 | int numrecs;/* number of records */ | |
3306 | int error; /* error return value */ | |
3307 | #ifdef DEBUG | |
3308 | int i; | |
3309 | #endif | |
13e831e0 | 3310 | xfs_daddr_t old_bn; |
b194c7d8 BN |
3311 | |
3312 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
d3cd7a27 | 3313 | XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, &rec); |
b194c7d8 BN |
3314 | |
3315 | ncur = NULL; | |
45413937 | 3316 | lkey = &nkey; |
b194c7d8 BN |
3317 | |
3318 | /* | |
3319 | * If we have an external root pointer, and we've made it to the | |
3320 | * root level, allocate a new root block and we're done. | |
3321 | */ | |
3322 | if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) && | |
3323 | (level >= cur->bc_nlevels)) { | |
3324 | error = xfs_btree_new_root(cur, stat); | |
3325 | xfs_btree_set_ptr_null(cur, ptrp); | |
3326 | ||
3327 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3328 | return error; | |
3329 | } | |
3330 | ||
3331 | /* If we're off the left edge, return failure. */ | |
3332 | ptr = cur->bc_ptrs[level]; | |
3333 | if (ptr == 0) { | |
3334 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3335 | *stat = 0; | |
3336 | return 0; | |
3337 | } | |
3338 | ||
b194c7d8 BN |
3339 | optr = ptr; |
3340 | ||
3341 | XFS_BTREE_STATS_INC(cur, insrec); | |
3342 | ||
3343 | /* Get pointers to the btree buffer and block. */ | |
3344 | block = xfs_btree_get_block(cur, level, &bp); | |
13e831e0 | 3345 | old_bn = bp ? bp->b_bn : XFS_BUF_DADDR_NULL; |
b194c7d8 BN |
3346 | numrecs = xfs_btree_get_numrecs(block); |
3347 | ||
3348 | #ifdef DEBUG | |
3349 | error = xfs_btree_check_block(cur, block, level, bp); | |
3350 | if (error) | |
3351 | goto error0; | |
3352 | ||
3353 | /* Check that the new entry is being inserted in the right place. */ | |
3354 | if (ptr <= numrecs) { | |
3355 | if (level == 0) { | |
d3cd7a27 | 3356 | ASSERT(cur->bc_ops->recs_inorder(cur, rec, |
b194c7d8 BN |
3357 | xfs_btree_rec_addr(cur, ptr, block))); |
3358 | } else { | |
d3cd7a27 | 3359 | ASSERT(cur->bc_ops->keys_inorder(cur, key, |
b194c7d8 BN |
3360 | xfs_btree_key_addr(cur, ptr, block))); |
3361 | } | |
3362 | } | |
3363 | #endif | |
3364 | ||
3365 | /* | |
3366 | * If the block is full, we can't insert the new entry until we | |
3367 | * make the block un-full. | |
3368 | */ | |
3369 | xfs_btree_set_ptr_null(cur, &nptr); | |
3370 | if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) { | |
3371 | error = xfs_btree_make_block_unfull(cur, level, numrecs, | |
13e831e0 | 3372 | &optr, &ptr, &nptr, &ncur, lkey, stat); |
b194c7d8 BN |
3373 | if (error || *stat == 0) |
3374 | goto error0; | |
3375 | } | |
3376 | ||
3377 | /* | |
3378 | * The current block may have changed if the block was | |
3379 | * previously full and we have just made space in it. | |
3380 | */ | |
3381 | block = xfs_btree_get_block(cur, level, &bp); | |
3382 | numrecs = xfs_btree_get_numrecs(block); | |
3383 | ||
3384 | #ifdef DEBUG | |
3385 | error = xfs_btree_check_block(cur, block, level, bp); | |
3386 | if (error) | |
3387 | return error; | |
3388 | #endif | |
3389 | ||
3390 | /* | |
3391 | * At this point we know there's room for our new entry in the block | |
3392 | * we're pointing at. | |
3393 | */ | |
3394 | XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1); | |
3395 | ||
3396 | if (level > 0) { | |
3397 | /* It's a nonleaf. make a hole in the keys and ptrs */ | |
3398 | union xfs_btree_key *kp; | |
3399 | union xfs_btree_ptr *pp; | |
3400 | ||
3401 | kp = xfs_btree_key_addr(cur, ptr, block); | |
3402 | pp = xfs_btree_ptr_addr(cur, ptr, block); | |
3403 | ||
3404 | #ifdef DEBUG | |
3405 | for (i = numrecs - ptr; i >= 0; i--) { | |
3406 | error = xfs_btree_check_ptr(cur, pp, i, level); | |
3407 | if (error) | |
3408 | return error; | |
3409 | } | |
3410 | #endif | |
3411 | ||
3412 | xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1); | |
3413 | xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1); | |
3414 | ||
3415 | #ifdef DEBUG | |
3416 | error = xfs_btree_check_ptr(cur, ptrp, 0, level); | |
3417 | if (error) | |
3418 | goto error0; | |
3419 | #endif | |
3420 | ||
3421 | /* Now put the new data in, bump numrecs and log it. */ | |
d3cd7a27 | 3422 | xfs_btree_copy_keys(cur, kp, key, 1); |
b194c7d8 BN |
3423 | xfs_btree_copy_ptrs(cur, pp, ptrp, 1); |
3424 | numrecs++; | |
3425 | xfs_btree_set_numrecs(block, numrecs); | |
3426 | xfs_btree_log_ptrs(cur, bp, ptr, numrecs); | |
3427 | xfs_btree_log_keys(cur, bp, ptr, numrecs); | |
3428 | #ifdef DEBUG | |
3429 | if (ptr < numrecs) { | |
3430 | ASSERT(cur->bc_ops->keys_inorder(cur, kp, | |
3431 | xfs_btree_key_addr(cur, ptr + 1, block))); | |
3432 | } | |
3433 | #endif | |
3434 | } else { | |
3435 | /* It's a leaf. make a hole in the records */ | |
3436 | union xfs_btree_rec *rp; | |
3437 | ||
3438 | rp = xfs_btree_rec_addr(cur, ptr, block); | |
3439 | ||
3440 | xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1); | |
3441 | ||
3442 | /* Now put the new data in, bump numrecs and log it. */ | |
d3cd7a27 | 3443 | xfs_btree_copy_recs(cur, rp, rec, 1); |
b194c7d8 BN |
3444 | xfs_btree_set_numrecs(block, ++numrecs); |
3445 | xfs_btree_log_recs(cur, bp, ptr, numrecs); | |
3446 | #ifdef DEBUG | |
3447 | if (ptr < numrecs) { | |
3448 | ASSERT(cur->bc_ops->recs_inorder(cur, rp, | |
3449 | xfs_btree_rec_addr(cur, ptr + 1, block))); | |
3450 | } | |
3451 | #endif | |
3452 | } | |
3453 | ||
3454 | /* Log the new number of records in the btree header. */ | |
3455 | xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS); | |
3456 | ||
13e831e0 DW |
3457 | /* |
3458 | * If we just inserted into a new tree block, we have to | |
3459 | * recalculate nkey here because nkey is out of date. | |
3460 | * | |
3461 | * Otherwise we're just updating an existing block (having shoved | |
3462 | * some records into the new tree block), so use the regular key | |
3463 | * update mechanism. | |
3464 | */ | |
3465 | if (bp && bp->b_bn != old_bn) { | |
3466 | xfs_btree_get_keys(cur, block, lkey); | |
3467 | } else if (xfs_btree_needs_key_update(cur, optr)) { | |
64dbe047 | 3468 | error = xfs_btree_update_keys(cur, level); |
b194c7d8 BN |
3469 | if (error) |
3470 | goto error0; | |
3471 | } | |
3472 | ||
3473 | /* | |
3474 | * If we are tracking the last record in the tree and | |
3475 | * we are at the far right edge of the tree, update it. | |
3476 | */ | |
3477 | if (xfs_btree_is_lastrec(cur, block, level)) { | |
d3cd7a27 | 3478 | cur->bc_ops->update_lastrec(cur, block, rec, |
b194c7d8 BN |
3479 | ptr, LASTREC_INSREC); |
3480 | } | |
3481 | ||
3482 | /* | |
3483 | * Return the new block number, if any. | |
3484 | * If there is one, give back a record value and a cursor too. | |
3485 | */ | |
3486 | *ptrp = nptr; | |
3487 | if (!xfs_btree_ptr_is_null(cur, &nptr)) { | |
13e831e0 | 3488 | xfs_btree_copy_keys(cur, key, lkey, 1); |
b194c7d8 BN |
3489 | *curp = ncur; |
3490 | } | |
3491 | ||
3492 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3493 | *stat = 1; | |
3494 | return 0; | |
3495 | ||
3496 | error0: | |
3497 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
3498 | return error; | |
3499 | } | |
3500 | ||
3501 | /* | |
3502 | * Insert the record at the point referenced by cur. | |
3503 | * | |
3504 | * A multi-level split of the tree on insert will invalidate the original | |
3505 | * cursor. All callers of this function should assume that the cursor is | |
3506 | * no longer valid and revalidate it. | |
3507 | */ | |
3508 | int | |
3509 | xfs_btree_insert( | |
3510 | struct xfs_btree_cur *cur, | |
3511 | int *stat) | |
3512 | { | |
3513 | int error; /* error return value */ | |
3514 | int i; /* result value, 0 for failure */ | |
3515 | int level; /* current level number in btree */ | |
3516 | union xfs_btree_ptr nptr; /* new block number (split result) */ | |
3517 | struct xfs_btree_cur *ncur; /* new cursor (split result) */ | |
3518 | struct xfs_btree_cur *pcur; /* previous level's cursor */ | |
45413937 | 3519 | union xfs_btree_key bkey; /* key of block to insert */ |
13e831e0 | 3520 | union xfs_btree_key *key; |
b194c7d8 BN |
3521 | union xfs_btree_rec rec; /* record to insert */ |
3522 | ||
3523 | level = 0; | |
3524 | ncur = NULL; | |
3525 | pcur = cur; | |
45413937 | 3526 | key = &bkey; |
b194c7d8 BN |
3527 | |
3528 | xfs_btree_set_ptr_null(cur, &nptr); | |
d3cd7a27 DW |
3529 | |
3530 | /* Make a key out of the record data to be inserted, and save it. */ | |
b194c7d8 | 3531 | cur->bc_ops->init_rec_from_cur(cur, &rec); |
13e831e0 | 3532 | cur->bc_ops->init_key_from_rec(key, &rec); |
b194c7d8 BN |
3533 | |
3534 | /* | |
3535 | * Loop going up the tree, starting at the leaf level. | |
3536 | * Stop when we don't get a split block, that must mean that | |
3537 | * the insert is finished with this level. | |
3538 | */ | |
3539 | do { | |
3540 | /* | |
3541 | * Insert nrec/nptr into this level of the tree. | |
3542 | * Note if we fail, nptr will be null. | |
3543 | */ | |
13e831e0 | 3544 | error = xfs_btree_insrec(pcur, level, &nptr, &rec, key, |
d3cd7a27 | 3545 | &ncur, &i); |
b194c7d8 BN |
3546 | if (error) { |
3547 | if (pcur != cur) | |
3548 | xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR); | |
3549 | goto error0; | |
3550 | } | |
3551 | ||
19ebedcf | 3552 | XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0); |
b194c7d8 BN |
3553 | level++; |
3554 | ||
3555 | /* | |
3556 | * See if the cursor we just used is trash. | |
3557 | * Can't trash the caller's cursor, but otherwise we should | |
3558 | * if ncur is a new cursor or we're about to be done. | |
3559 | */ | |
3560 | if (pcur != cur && | |
3561 | (ncur || xfs_btree_ptr_is_null(cur, &nptr))) { | |
3562 | /* Save the state from the cursor before we trash it */ | |
3563 | if (cur->bc_ops->update_cursor) | |
3564 | cur->bc_ops->update_cursor(pcur, cur); | |
3565 | cur->bc_nlevels = pcur->bc_nlevels; | |
3566 | xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR); | |
3567 | } | |
3568 | /* If we got a new cursor, switch to it. */ | |
3569 | if (ncur) { | |
3570 | pcur = ncur; | |
3571 | ncur = NULL; | |
3572 | } | |
3573 | } while (!xfs_btree_ptr_is_null(cur, &nptr)); | |
3574 | ||
3575 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3576 | *stat = i; | |
3577 | return 0; | |
3578 | error0: | |
3579 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
3580 | return error; | |
3581 | } | |
3582 | ||
3583 | /* | |
3584 | * Try to merge a non-leaf block back into the inode root. | |
3585 | * | |
3586 | * Note: the killroot names comes from the fact that we're effectively | |
3587 | * killing the old root block. But because we can't just delete the | |
3588 | * inode we have to copy the single block it was pointing to into the | |
3589 | * inode. | |
3590 | */ | |
56b2de80 | 3591 | STATIC int |
b194c7d8 BN |
3592 | xfs_btree_kill_iroot( |
3593 | struct xfs_btree_cur *cur) | |
3594 | { | |
3595 | int whichfork = cur->bc_private.b.whichfork; | |
3596 | struct xfs_inode *ip = cur->bc_private.b.ip; | |
3597 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); | |
3598 | struct xfs_btree_block *block; | |
3599 | struct xfs_btree_block *cblock; | |
3600 | union xfs_btree_key *kp; | |
3601 | union xfs_btree_key *ckp; | |
3602 | union xfs_btree_ptr *pp; | |
3603 | union xfs_btree_ptr *cpp; | |
3604 | struct xfs_buf *cbp; | |
3605 | int level; | |
3606 | int index; | |
3607 | int numrecs; | |
410c3de5 | 3608 | int error; |
b194c7d8 BN |
3609 | #ifdef DEBUG |
3610 | union xfs_btree_ptr ptr; | |
3611 | int i; | |
3612 | #endif | |
3613 | ||
3614 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
3615 | ||
3616 | ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE); | |
3617 | ASSERT(cur->bc_nlevels > 1); | |
3618 | ||
3619 | /* | |
3620 | * Don't deal with the root block needs to be a leaf case. | |
3621 | * We're just going to turn the thing back into extents anyway. | |
3622 | */ | |
3623 | level = cur->bc_nlevels - 1; | |
3624 | if (level == 1) | |
3625 | goto out0; | |
3626 | ||
3627 | /* | |
3628 | * Give up if the root has multiple children. | |
3629 | */ | |
3630 | block = xfs_btree_get_iroot(cur); | |
3631 | if (xfs_btree_get_numrecs(block) != 1) | |
3632 | goto out0; | |
3633 | ||
3634 | cblock = xfs_btree_get_block(cur, level - 1, &cbp); | |
3635 | numrecs = xfs_btree_get_numrecs(cblock); | |
3636 | ||
3637 | /* | |
3638 | * Only do this if the next level will fit. | |
3639 | * Then the data must be copied up to the inode, | |
3640 | * instead of freeing the root you free the next level. | |
3641 | */ | |
3642 | if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level)) | |
3643 | goto out0; | |
3644 | ||
3645 | XFS_BTREE_STATS_INC(cur, killroot); | |
3646 | ||
3647 | #ifdef DEBUG | |
3648 | xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB); | |
3649 | ASSERT(xfs_btree_ptr_is_null(cur, &ptr)); | |
3650 | xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB); | |
3651 | ASSERT(xfs_btree_ptr_is_null(cur, &ptr)); | |
3652 | #endif | |
3653 | ||
3654 | index = numrecs - cur->bc_ops->get_maxrecs(cur, level); | |
3655 | if (index) { | |
3656 | xfs_iroot_realloc(cur->bc_private.b.ip, index, | |
3657 | cur->bc_private.b.whichfork); | |
b3563c19 | 3658 | block = ifp->if_broot; |
b194c7d8 BN |
3659 | } |
3660 | ||
3661 | be16_add_cpu(&block->bb_numrecs, index); | |
3662 | ASSERT(block->bb_numrecs == cblock->bb_numrecs); | |
3663 | ||
3664 | kp = xfs_btree_key_addr(cur, 1, block); | |
3665 | ckp = xfs_btree_key_addr(cur, 1, cblock); | |
3666 | xfs_btree_copy_keys(cur, kp, ckp, numrecs); | |
3667 | ||
3668 | pp = xfs_btree_ptr_addr(cur, 1, block); | |
3669 | cpp = xfs_btree_ptr_addr(cur, 1, cblock); | |
3670 | #ifdef DEBUG | |
3671 | for (i = 0; i < numrecs; i++) { | |
b194c7d8 BN |
3672 | error = xfs_btree_check_ptr(cur, cpp, i, level - 1); |
3673 | if (error) { | |
3674 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
3675 | return error; | |
3676 | } | |
3677 | } | |
3678 | #endif | |
3679 | xfs_btree_copy_ptrs(cur, pp, cpp, numrecs); | |
3680 | ||
c261f8c0 | 3681 | error = xfs_btree_free_block(cur, cbp); |
410c3de5 CH |
3682 | if (error) { |
3683 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
3684 | return error; | |
3685 | } | |
b194c7d8 BN |
3686 | |
3687 | cur->bc_bufs[level - 1] = NULL; | |
3688 | be16_add_cpu(&block->bb_level, -1); | |
3689 | xfs_trans_log_inode(cur->bc_tp, ip, | |
56b2de80 | 3690 | XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork)); |
b194c7d8 BN |
3691 | cur->bc_nlevels--; |
3692 | out0: | |
3693 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3694 | return 0; | |
3695 | } | |
3696 | ||
56b2de80 DC |
3697 | /* |
3698 | * Kill the current root node, and replace it with it's only child node. | |
3699 | */ | |
3700 | STATIC int | |
3701 | xfs_btree_kill_root( | |
3702 | struct xfs_btree_cur *cur, | |
3703 | struct xfs_buf *bp, | |
3704 | int level, | |
3705 | union xfs_btree_ptr *newroot) | |
3706 | { | |
3707 | int error; | |
3708 | ||
3709 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
3710 | XFS_BTREE_STATS_INC(cur, killroot); | |
3711 | ||
3712 | /* | |
3713 | * Update the root pointer, decreasing the level by 1 and then | |
3714 | * free the old root. | |
3715 | */ | |
3716 | cur->bc_ops->set_root(cur, newroot, -1); | |
3717 | ||
c261f8c0 | 3718 | error = xfs_btree_free_block(cur, bp); |
56b2de80 DC |
3719 | if (error) { |
3720 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
3721 | return error; | |
3722 | } | |
3723 | ||
56b2de80 DC |
3724 | cur->bc_bufs[level] = NULL; |
3725 | cur->bc_ra[level] = 0; | |
3726 | cur->bc_nlevels--; | |
3727 | ||
3728 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3729 | return 0; | |
3730 | } | |
3731 | ||
b194c7d8 BN |
3732 | STATIC int |
3733 | xfs_btree_dec_cursor( | |
3734 | struct xfs_btree_cur *cur, | |
3735 | int level, | |
3736 | int *stat) | |
3737 | { | |
3738 | int error; | |
3739 | int i; | |
3740 | ||
3741 | if (level > 0) { | |
3742 | error = xfs_btree_decrement(cur, level, &i); | |
3743 | if (error) | |
3744 | return error; | |
3745 | } | |
3746 | ||
3747 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3748 | *stat = 1; | |
3749 | return 0; | |
3750 | } | |
3751 | ||
3752 | /* | |
3753 | * Single level of the btree record deletion routine. | |
3754 | * Delete record pointed to by cur/level. | |
3755 | * Remove the record from its block then rebalance the tree. | |
3756 | * Return 0 for error, 1 for done, 2 to go on to the next level. | |
3757 | */ | |
3758 | STATIC int /* error */ | |
3759 | xfs_btree_delrec( | |
3760 | struct xfs_btree_cur *cur, /* btree cursor */ | |
3761 | int level, /* level removing record from */ | |
3762 | int *stat) /* fail/done/go-on */ | |
3763 | { | |
3764 | struct xfs_btree_block *block; /* btree block */ | |
3765 | union xfs_btree_ptr cptr; /* current block ptr */ | |
3766 | struct xfs_buf *bp; /* buffer for block */ | |
3767 | int error; /* error return value */ | |
3768 | int i; /* loop counter */ | |
b194c7d8 BN |
3769 | union xfs_btree_ptr lptr; /* left sibling block ptr */ |
3770 | struct xfs_buf *lbp; /* left buffer pointer */ | |
3771 | struct xfs_btree_block *left; /* left btree block */ | |
3772 | int lrecs = 0; /* left record count */ | |
3773 | int ptr; /* key/record index */ | |
3774 | union xfs_btree_ptr rptr; /* right sibling block ptr */ | |
3775 | struct xfs_buf *rbp; /* right buffer pointer */ | |
3776 | struct xfs_btree_block *right; /* right btree block */ | |
3777 | struct xfs_btree_block *rrblock; /* right-right btree block */ | |
3778 | struct xfs_buf *rrbp; /* right-right buffer pointer */ | |
3779 | int rrecs = 0; /* right record count */ | |
3780 | struct xfs_btree_cur *tcur; /* temporary btree cursor */ | |
3781 | int numrecs; /* temporary numrec count */ | |
3782 | ||
3783 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
3784 | XFS_BTREE_TRACE_ARGI(cur, level); | |
3785 | ||
3786 | tcur = NULL; | |
3787 | ||
3788 | /* Get the index of the entry being deleted, check for nothing there. */ | |
3789 | ptr = cur->bc_ptrs[level]; | |
3790 | if (ptr == 0) { | |
3791 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3792 | *stat = 0; | |
3793 | return 0; | |
3794 | } | |
3795 | ||
3796 | /* Get the buffer & block containing the record or key/ptr. */ | |
3797 | block = xfs_btree_get_block(cur, level, &bp); | |
3798 | numrecs = xfs_btree_get_numrecs(block); | |
3799 | ||
3800 | #ifdef DEBUG | |
3801 | error = xfs_btree_check_block(cur, block, level, bp); | |
3802 | if (error) | |
3803 | goto error0; | |
3804 | #endif | |
3805 | ||
3806 | /* Fail if we're off the end of the block. */ | |
3807 | if (ptr > numrecs) { | |
3808 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
3809 | *stat = 0; | |
3810 | return 0; | |
3811 | } | |
3812 | ||
3813 | XFS_BTREE_STATS_INC(cur, delrec); | |
3814 | XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr); | |
3815 | ||
3816 | /* Excise the entries being deleted. */ | |
3817 | if (level > 0) { | |
3818 | /* It's a nonleaf. operate on keys and ptrs */ | |
3819 | union xfs_btree_key *lkp; | |
3820 | union xfs_btree_ptr *lpp; | |
3821 | ||
3822 | lkp = xfs_btree_key_addr(cur, ptr + 1, block); | |
3823 | lpp = xfs_btree_ptr_addr(cur, ptr + 1, block); | |
3824 | ||
3825 | #ifdef DEBUG | |
3826 | for (i = 0; i < numrecs - ptr; i++) { | |
3827 | error = xfs_btree_check_ptr(cur, lpp, i, level); | |
3828 | if (error) | |
3829 | goto error0; | |
3830 | } | |
3831 | #endif | |
3832 | ||
3833 | if (ptr < numrecs) { | |
3834 | xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr); | |
3835 | xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr); | |
3836 | xfs_btree_log_keys(cur, bp, ptr, numrecs - 1); | |
3837 | xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1); | |
3838 | } | |
b194c7d8 BN |
3839 | } else { |
3840 | /* It's a leaf. operate on records */ | |
3841 | if (ptr < numrecs) { | |
3842 | xfs_btree_shift_recs(cur, | |
3843 | xfs_btree_rec_addr(cur, ptr + 1, block), | |
3844 | -1, numrecs - ptr); | |
3845 | xfs_btree_log_recs(cur, bp, ptr, numrecs - 1); | |
3846 | } | |
b194c7d8 BN |
3847 | } |
3848 | ||
3849 | /* | |
3850 | * Decrement and log the number of entries in the block. | |
3851 | */ | |
3852 | xfs_btree_set_numrecs(block, --numrecs); | |
3853 | xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS); | |
3854 | ||
3855 | /* | |
3856 | * If we are tracking the last record in the tree and | |
3857 | * we are at the far right edge of the tree, update it. | |
3858 | */ | |
3859 | if (xfs_btree_is_lastrec(cur, block, level)) { | |
3860 | cur->bc_ops->update_lastrec(cur, block, NULL, | |
3861 | ptr, LASTREC_DELREC); | |
3862 | } | |
3863 | ||
3864 | /* | |
3865 | * We're at the root level. First, shrink the root block in-memory. | |
3866 | * Try to get rid of the next level down. If we can't then there's | |
3867 | * nothing left to do. | |
3868 | */ | |
3869 | if (level == cur->bc_nlevels - 1) { | |
3870 | if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) { | |
3871 | xfs_iroot_realloc(cur->bc_private.b.ip, -1, | |
3872 | cur->bc_private.b.whichfork); | |
3873 | ||
3874 | error = xfs_btree_kill_iroot(cur); | |
3875 | if (error) | |
3876 | goto error0; | |
3877 | ||
3878 | error = xfs_btree_dec_cursor(cur, level, stat); | |
3879 | if (error) | |
3880 | goto error0; | |
3881 | *stat = 1; | |
3882 | return 0; | |
3883 | } | |
3884 | ||
3885 | /* | |
3886 | * If this is the root level, and there's only one entry left, | |
3887 | * and it's NOT the leaf level, then we can get rid of this | |
3888 | * level. | |
3889 | */ | |
3890 | if (numrecs == 1 && level > 0) { | |
3891 | union xfs_btree_ptr *pp; | |
3892 | /* | |
3893 | * pp is still set to the first pointer in the block. | |
3894 | * Make it the new root of the btree. | |
3895 | */ | |
3896 | pp = xfs_btree_ptr_addr(cur, 1, block); | |
56b2de80 | 3897 | error = xfs_btree_kill_root(cur, bp, level, pp); |
b194c7d8 BN |
3898 | if (error) |
3899 | goto error0; | |
3900 | } else if (level > 0) { | |
3901 | error = xfs_btree_dec_cursor(cur, level, stat); | |
3902 | if (error) | |
3903 | goto error0; | |
3904 | } | |
3905 | *stat = 1; | |
3906 | return 0; | |
3907 | } | |
3908 | ||
3909 | /* | |
3910 | * If we deleted the leftmost entry in the block, update the | |
3911 | * key values above us in the tree. | |
3912 | */ | |
a3c9cb10 | 3913 | if (xfs_btree_needs_key_update(cur, ptr)) { |
64dbe047 | 3914 | error = xfs_btree_update_keys(cur, level); |
b194c7d8 BN |
3915 | if (error) |
3916 | goto error0; | |
3917 | } | |
3918 | ||
3919 | /* | |
3920 | * If the number of records remaining in the block is at least | |
3921 | * the minimum, we're done. | |
3922 | */ | |
3923 | if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) { | |
3924 | error = xfs_btree_dec_cursor(cur, level, stat); | |
3925 | if (error) | |
3926 | goto error0; | |
3927 | return 0; | |
3928 | } | |
3929 | ||
3930 | /* | |
3931 | * Otherwise, we have to move some records around to keep the | |
3932 | * tree balanced. Look at the left and right sibling blocks to | |
3933 | * see if we can re-balance by moving only one record. | |
3934 | */ | |
3935 | xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB); | |
3936 | xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB); | |
3937 | ||
3938 | if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) { | |
3939 | /* | |
3940 | * One child of root, need to get a chance to copy its contents | |
3941 | * into the root and delete it. Can't go up to next level, | |
3942 | * there's nothing to delete there. | |
3943 | */ | |
3944 | if (xfs_btree_ptr_is_null(cur, &rptr) && | |
3945 | xfs_btree_ptr_is_null(cur, &lptr) && | |
3946 | level == cur->bc_nlevels - 2) { | |
3947 | error = xfs_btree_kill_iroot(cur); | |
3948 | if (!error) | |
3949 | error = xfs_btree_dec_cursor(cur, level, stat); | |
3950 | if (error) | |
3951 | goto error0; | |
3952 | return 0; | |
3953 | } | |
3954 | } | |
3955 | ||
3956 | ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) || | |
3957 | !xfs_btree_ptr_is_null(cur, &lptr)); | |
3958 | ||
3959 | /* | |
3960 | * Duplicate the cursor so our btree manipulations here won't | |
3961 | * disrupt the next level up. | |
3962 | */ | |
3963 | error = xfs_btree_dup_cursor(cur, &tcur); | |
3964 | if (error) | |
3965 | goto error0; | |
3966 | ||
3967 | /* | |
3968 | * If there's a right sibling, see if it's ok to shift an entry | |
3969 | * out of it. | |
3970 | */ | |
3971 | if (!xfs_btree_ptr_is_null(cur, &rptr)) { | |
3972 | /* | |
3973 | * Move the temp cursor to the last entry in the next block. | |
3974 | * Actually any entry but the first would suffice. | |
3975 | */ | |
3976 | i = xfs_btree_lastrec(tcur, level); | |
19ebedcf | 3977 | XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0); |
b194c7d8 BN |
3978 | |
3979 | error = xfs_btree_increment(tcur, level, &i); | |
3980 | if (error) | |
3981 | goto error0; | |
19ebedcf | 3982 | XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0); |
b194c7d8 BN |
3983 | |
3984 | i = xfs_btree_lastrec(tcur, level); | |
19ebedcf | 3985 | XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0); |
b194c7d8 BN |
3986 | |
3987 | /* Grab a pointer to the block. */ | |
3988 | right = xfs_btree_get_block(tcur, level, &rbp); | |
3989 | #ifdef DEBUG | |
3990 | error = xfs_btree_check_block(tcur, right, level, rbp); | |
3991 | if (error) | |
3992 | goto error0; | |
3993 | #endif | |
3994 | /* Grab the current block number, for future use. */ | |
3995 | xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB); | |
3996 | ||
3997 | /* | |
3998 | * If right block is full enough so that removing one entry | |
3999 | * won't make it too empty, and left-shifting an entry out | |
4000 | * of right to us works, we're done. | |
4001 | */ | |
4002 | if (xfs_btree_get_numrecs(right) - 1 >= | |
4003 | cur->bc_ops->get_minrecs(tcur, level)) { | |
4004 | error = xfs_btree_lshift(tcur, level, &i); | |
4005 | if (error) | |
4006 | goto error0; | |
4007 | if (i) { | |
4008 | ASSERT(xfs_btree_get_numrecs(block) >= | |
4009 | cur->bc_ops->get_minrecs(tcur, level)); | |
4010 | ||
4011 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
4012 | tcur = NULL; | |
4013 | ||
4014 | error = xfs_btree_dec_cursor(cur, level, stat); | |
4015 | if (error) | |
4016 | goto error0; | |
4017 | return 0; | |
4018 | } | |
4019 | } | |
4020 | ||
4021 | /* | |
4022 | * Otherwise, grab the number of records in right for | |
4023 | * future reference, and fix up the temp cursor to point | |
4024 | * to our block again (last record). | |
4025 | */ | |
4026 | rrecs = xfs_btree_get_numrecs(right); | |
4027 | if (!xfs_btree_ptr_is_null(cur, &lptr)) { | |
4028 | i = xfs_btree_firstrec(tcur, level); | |
19ebedcf | 4029 | XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0); |
b194c7d8 BN |
4030 | |
4031 | error = xfs_btree_decrement(tcur, level, &i); | |
4032 | if (error) | |
4033 | goto error0; | |
19ebedcf | 4034 | XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0); |
b194c7d8 BN |
4035 | } |
4036 | } | |
4037 | ||
4038 | /* | |
4039 | * If there's a left sibling, see if it's ok to shift an entry | |
4040 | * out of it. | |
4041 | */ | |
4042 | if (!xfs_btree_ptr_is_null(cur, &lptr)) { | |
4043 | /* | |
4044 | * Move the temp cursor to the first entry in the | |
4045 | * previous block. | |
4046 | */ | |
4047 | i = xfs_btree_firstrec(tcur, level); | |
19ebedcf | 4048 | XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0); |
b194c7d8 BN |
4049 | |
4050 | error = xfs_btree_decrement(tcur, level, &i); | |
4051 | if (error) | |
4052 | goto error0; | |
4053 | i = xfs_btree_firstrec(tcur, level); | |
19ebedcf | 4054 | XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0); |
b194c7d8 BN |
4055 | |
4056 | /* Grab a pointer to the block. */ | |
4057 | left = xfs_btree_get_block(tcur, level, &lbp); | |
4058 | #ifdef DEBUG | |
4059 | error = xfs_btree_check_block(cur, left, level, lbp); | |
4060 | if (error) | |
4061 | goto error0; | |
4062 | #endif | |
4063 | /* Grab the current block number, for future use. */ | |
4064 | xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB); | |
4065 | ||
4066 | /* | |
4067 | * If left block is full enough so that removing one entry | |
4068 | * won't make it too empty, and right-shifting an entry out | |
4069 | * of left to us works, we're done. | |
4070 | */ | |
4071 | if (xfs_btree_get_numrecs(left) - 1 >= | |
4072 | cur->bc_ops->get_minrecs(tcur, level)) { | |
4073 | error = xfs_btree_rshift(tcur, level, &i); | |
4074 | if (error) | |
4075 | goto error0; | |
4076 | if (i) { | |
4077 | ASSERT(xfs_btree_get_numrecs(block) >= | |
4078 | cur->bc_ops->get_minrecs(tcur, level)); | |
4079 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
4080 | tcur = NULL; | |
4081 | if (level == 0) | |
4082 | cur->bc_ptrs[0]++; | |
4083 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
4084 | *stat = 1; | |
4085 | return 0; | |
4086 | } | |
4087 | } | |
4088 | ||
4089 | /* | |
4090 | * Otherwise, grab the number of records in right for | |
4091 | * future reference. | |
4092 | */ | |
4093 | lrecs = xfs_btree_get_numrecs(left); | |
4094 | } | |
4095 | ||
4096 | /* Delete the temp cursor, we're done with it. */ | |
4097 | xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR); | |
4098 | tcur = NULL; | |
4099 | ||
4100 | /* If here, we need to do a join to keep the tree balanced. */ | |
4101 | ASSERT(!xfs_btree_ptr_is_null(cur, &cptr)); | |
4102 | ||
4103 | if (!xfs_btree_ptr_is_null(cur, &lptr) && | |
4104 | lrecs + xfs_btree_get_numrecs(block) <= | |
4105 | cur->bc_ops->get_maxrecs(cur, level)) { | |
4106 | /* | |
4107 | * Set "right" to be the starting block, | |
4108 | * "left" to be the left neighbor. | |
4109 | */ | |
4110 | rptr = cptr; | |
4111 | right = block; | |
4112 | rbp = bp; | |
ff105f75 | 4113 | error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp); |
b194c7d8 BN |
4114 | if (error) |
4115 | goto error0; | |
4116 | ||
4117 | /* | |
4118 | * If that won't work, see if we can join with the right neighbor block. | |
4119 | */ | |
4120 | } else if (!xfs_btree_ptr_is_null(cur, &rptr) && | |
4121 | rrecs + xfs_btree_get_numrecs(block) <= | |
4122 | cur->bc_ops->get_maxrecs(cur, level)) { | |
4123 | /* | |
4124 | * Set "left" to be the starting block, | |
4125 | * "right" to be the right neighbor. | |
4126 | */ | |
4127 | lptr = cptr; | |
4128 | left = block; | |
4129 | lbp = bp; | |
ff105f75 | 4130 | error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp); |
b194c7d8 BN |
4131 | if (error) |
4132 | goto error0; | |
4133 | ||
4134 | /* | |
4135 | * Otherwise, we can't fix the imbalance. | |
4136 | * Just return. This is probably a logic error, but it's not fatal. | |
4137 | */ | |
4138 | } else { | |
4139 | error = xfs_btree_dec_cursor(cur, level, stat); | |
4140 | if (error) | |
4141 | goto error0; | |
4142 | return 0; | |
4143 | } | |
4144 | ||
4145 | rrecs = xfs_btree_get_numrecs(right); | |
4146 | lrecs = xfs_btree_get_numrecs(left); | |
4147 | ||
4148 | /* | |
4149 | * We're now going to join "left" and "right" by moving all the stuff | |
4150 | * in "right" to "left" and deleting "right". | |
4151 | */ | |
4152 | XFS_BTREE_STATS_ADD(cur, moves, rrecs); | |
4153 | if (level > 0) { | |
4154 | /* It's a non-leaf. Move keys and pointers. */ | |
4155 | union xfs_btree_key *lkp; /* left btree key */ | |
4156 | union xfs_btree_ptr *lpp; /* left address pointer */ | |
4157 | union xfs_btree_key *rkp; /* right btree key */ | |
4158 | union xfs_btree_ptr *rpp; /* right address pointer */ | |
4159 | ||
4160 | lkp = xfs_btree_key_addr(cur, lrecs + 1, left); | |
4161 | lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left); | |
4162 | rkp = xfs_btree_key_addr(cur, 1, right); | |
4163 | rpp = xfs_btree_ptr_addr(cur, 1, right); | |
4164 | #ifdef DEBUG | |
4165 | for (i = 1; i < rrecs; i++) { | |
4166 | error = xfs_btree_check_ptr(cur, rpp, i, level); | |
4167 | if (error) | |
4168 | goto error0; | |
4169 | } | |
4170 | #endif | |
4171 | xfs_btree_copy_keys(cur, lkp, rkp, rrecs); | |
4172 | xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs); | |
4173 | ||
4174 | xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs); | |
4175 | xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs); | |
4176 | } else { | |
4177 | /* It's a leaf. Move records. */ | |
4178 | union xfs_btree_rec *lrp; /* left record pointer */ | |
4179 | union xfs_btree_rec *rrp; /* right record pointer */ | |
4180 | ||
4181 | lrp = xfs_btree_rec_addr(cur, lrecs + 1, left); | |
4182 | rrp = xfs_btree_rec_addr(cur, 1, right); | |
4183 | ||
4184 | xfs_btree_copy_recs(cur, lrp, rrp, rrecs); | |
4185 | xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs); | |
4186 | } | |
4187 | ||
4188 | XFS_BTREE_STATS_INC(cur, join); | |
4189 | ||
4190 | /* | |
56b2de80 | 4191 | * Fix up the number of records and right block pointer in the |
b194c7d8 BN |
4192 | * surviving block, and log it. |
4193 | */ | |
4194 | xfs_btree_set_numrecs(left, lrecs + rrecs); | |
4195 | xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB), | |
4196 | xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB); | |
4197 | xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB); | |
4198 | ||
4199 | /* If there is a right sibling, point it to the remaining block. */ | |
4200 | xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB); | |
4201 | if (!xfs_btree_ptr_is_null(cur, &cptr)) { | |
ff105f75 | 4202 | error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp); |
b194c7d8 BN |
4203 | if (error) |
4204 | goto error0; | |
4205 | xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB); | |
4206 | xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB); | |
4207 | } | |
4208 | ||
4209 | /* Free the deleted block. */ | |
c261f8c0 | 4210 | error = xfs_btree_free_block(cur, rbp); |
b194c7d8 BN |
4211 | if (error) |
4212 | goto error0; | |
b194c7d8 BN |
4213 | |
4214 | /* | |
4215 | * If we joined with the left neighbor, set the buffer in the | |
4216 | * cursor to the left block, and fix up the index. | |
4217 | */ | |
4218 | if (bp != lbp) { | |
4219 | cur->bc_bufs[level] = lbp; | |
4220 | cur->bc_ptrs[level] += lrecs; | |
4221 | cur->bc_ra[level] = 0; | |
4222 | } | |
4223 | /* | |
4224 | * If we joined with the right neighbor and there's a level above | |
4225 | * us, increment the cursor at that level. | |
4226 | */ | |
4227 | else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) || | |
4228 | (level + 1 < cur->bc_nlevels)) { | |
4229 | error = xfs_btree_increment(cur, level + 1, &i); | |
4230 | if (error) | |
4231 | goto error0; | |
4232 | } | |
4233 | ||
4234 | /* | |
4235 | * Readjust the ptr at this level if it's not a leaf, since it's | |
4236 | * still pointing at the deletion point, which makes the cursor | |
4237 | * inconsistent. If this makes the ptr 0, the caller fixes it up. | |
4238 | * We can't use decrement because it would change the next level up. | |
4239 | */ | |
4240 | if (level > 0) | |
4241 | cur->bc_ptrs[level]--; | |
4242 | ||
13e831e0 DW |
4243 | /* |
4244 | * We combined blocks, so we have to update the parent keys if the | |
4245 | * btree supports overlapped intervals. However, bc_ptrs[level + 1] | |
4246 | * points to the old block so that the caller knows which record to | |
4247 | * delete. Therefore, the caller must be savvy enough to call updkeys | |
4248 | * for us if we return stat == 2. The other exit points from this | |
4249 | * function don't require deletions further up the tree, so they can | |
4250 | * call updkeys directly. | |
4251 | */ | |
4252 | ||
b194c7d8 BN |
4253 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); |
4254 | /* Return value means the next level up has something to do. */ | |
4255 | *stat = 2; | |
4256 | return 0; | |
4257 | ||
4258 | error0: | |
4259 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
4260 | if (tcur) | |
4261 | xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR); | |
4262 | return error; | |
4263 | } | |
4264 | ||
4265 | /* | |
4266 | * Delete the record pointed to by cur. | |
4267 | * The cursor refers to the place where the record was (could be inserted) | |
4268 | * when the operation returns. | |
4269 | */ | |
4270 | int /* error */ | |
4271 | xfs_btree_delete( | |
4272 | struct xfs_btree_cur *cur, | |
4273 | int *stat) /* success/failure */ | |
4274 | { | |
4275 | int error; /* error return value */ | |
4276 | int level; | |
4277 | int i; | |
13e831e0 | 4278 | bool joined = false; |
b194c7d8 BN |
4279 | |
4280 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY); | |
4281 | ||
4282 | /* | |
4283 | * Go up the tree, starting at leaf level. | |
4284 | * | |
4285 | * If 2 is returned then a join was done; go to the next level. | |
4286 | * Otherwise we are done. | |
4287 | */ | |
4288 | for (level = 0, i = 2; i == 2; level++) { | |
4289 | error = xfs_btree_delrec(cur, level, &i); | |
4290 | if (error) | |
4291 | goto error0; | |
13e831e0 DW |
4292 | if (i == 2) |
4293 | joined = true; | |
4294 | } | |
4295 | ||
4296 | /* | |
4297 | * If we combined blocks as part of deleting the record, delrec won't | |
4298 | * have updated the parent high keys so we have to do that here. | |
4299 | */ | |
4300 | if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) { | |
4301 | error = xfs_btree_updkeys_force(cur, 0); | |
4302 | if (error) | |
4303 | goto error0; | |
b194c7d8 BN |
4304 | } |
4305 | ||
4306 | if (i == 0) { | |
4307 | for (level = 1; level < cur->bc_nlevels; level++) { | |
4308 | if (cur->bc_ptrs[level] == 0) { | |
4309 | error = xfs_btree_decrement(cur, level, &i); | |
4310 | if (error) | |
4311 | goto error0; | |
4312 | break; | |
4313 | } | |
4314 | } | |
4315 | } | |
4316 | ||
4317 | XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT); | |
4318 | *stat = i; | |
4319 | return 0; | |
4320 | error0: | |
4321 | XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR); | |
4322 | return error; | |
4323 | } | |
4324 | ||
4325 | /* | |
4326 | * Get the data from the pointed-to record. | |
4327 | */ | |
4328 | int /* error */ | |
4329 | xfs_btree_get_rec( | |
4330 | struct xfs_btree_cur *cur, /* btree cursor */ | |
4331 | union xfs_btree_rec **recp, /* output: btree record */ | |
4332 | int *stat) /* output: success/failure */ | |
4333 | { | |
4334 | struct xfs_btree_block *block; /* btree block */ | |
4335 | struct xfs_buf *bp; /* buffer pointer */ | |
4336 | int ptr; /* record number */ | |
4337 | #ifdef DEBUG | |
4338 | int error; /* error return value */ | |
4339 | #endif | |
4340 | ||
4341 | ptr = cur->bc_ptrs[0]; | |
4342 | block = xfs_btree_get_block(cur, 0, &bp); | |
4343 | ||
4344 | #ifdef DEBUG | |
4345 | error = xfs_btree_check_block(cur, block, 0, bp); | |
4346 | if (error) | |
4347 | return error; | |
4348 | #endif | |
4349 | ||
4350 | /* | |
4351 | * Off the right end or left end, return failure. | |
4352 | */ | |
4353 | if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) { | |
4354 | *stat = 0; | |
4355 | return 0; | |
4356 | } | |
4357 | ||
4358 | /* | |
4359 | * Point to the record and extract its data. | |
4360 | */ | |
4361 | *recp = xfs_btree_rec_addr(cur, ptr, block); | |
4362 | *stat = 1; | |
4363 | return 0; | |
4364 | } | |
9c6ebc42 | 4365 | |
f31736bf DW |
4366 | /* Visit a block in a btree. */ |
4367 | STATIC int | |
4368 | xfs_btree_visit_block( | |
4369 | struct xfs_btree_cur *cur, | |
4370 | int level, | |
4371 | xfs_btree_visit_blocks_fn fn, | |
4372 | void *data) | |
4373 | { | |
4374 | struct xfs_btree_block *block; | |
4375 | struct xfs_buf *bp; | |
4376 | union xfs_btree_ptr rptr; | |
4377 | int error; | |
4378 | ||
4379 | /* do right sibling readahead */ | |
4380 | xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA); | |
4381 | block = xfs_btree_get_block(cur, level, &bp); | |
4382 | ||
4383 | /* process the block */ | |
4384 | error = fn(cur, level, data); | |
4385 | if (error) | |
4386 | return error; | |
4387 | ||
4388 | /* now read rh sibling block for next iteration */ | |
4389 | xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB); | |
4390 | if (xfs_btree_ptr_is_null(cur, &rptr)) | |
4391 | return -ENOENT; | |
4392 | ||
4393 | return xfs_btree_lookup_get_block(cur, level, &rptr, &block); | |
4394 | } | |
4395 | ||
4396 | ||
4397 | /* Visit every block in a btree. */ | |
4398 | int | |
4399 | xfs_btree_visit_blocks( | |
4400 | struct xfs_btree_cur *cur, | |
4401 | xfs_btree_visit_blocks_fn fn, | |
4402 | void *data) | |
4403 | { | |
4404 | union xfs_btree_ptr lptr; | |
4405 | int level; | |
4406 | struct xfs_btree_block *block = NULL; | |
4407 | int error = 0; | |
4408 | ||
4409 | cur->bc_ops->init_ptr_from_cur(cur, &lptr); | |
4410 | ||
4411 | /* for each level */ | |
4412 | for (level = cur->bc_nlevels - 1; level >= 0; level--) { | |
4413 | /* grab the left hand block */ | |
4414 | error = xfs_btree_lookup_get_block(cur, level, &lptr, &block); | |
4415 | if (error) | |
4416 | return error; | |
4417 | ||
4418 | /* readahead the left most block for the next level down */ | |
4419 | if (level > 0) { | |
4420 | union xfs_btree_ptr *ptr; | |
4421 | ||
4422 | ptr = xfs_btree_ptr_addr(cur, 1, block); | |
4423 | xfs_btree_readahead_ptr(cur, ptr, 1); | |
4424 | ||
4425 | /* save for the next iteration of the loop */ | |
f9a6d642 | 4426 | xfs_btree_copy_ptrs(cur, &lptr, ptr, 1); |
f31736bf DW |
4427 | } |
4428 | ||
4429 | /* for each buffer in the level */ | |
4430 | do { | |
4431 | error = xfs_btree_visit_block(cur, level, fn, data); | |
4432 | } while (!error); | |
4433 | ||
4434 | if (error != -ENOENT) | |
4435 | return error; | |
4436 | } | |
4437 | ||
4438 | return 0; | |
4439 | } | |
4440 | ||
9c6ebc42 DC |
4441 | /* |
4442 | * Change the owner of a btree. | |
4443 | * | |
4444 | * The mechanism we use here is ordered buffer logging. Because we don't know | |
4445 | * how many buffers were are going to need to modify, we don't really want to | |
4446 | * have to make transaction reservations for the worst case of every buffer in a | |
4447 | * full size btree as that may be more space that we can fit in the log.... | |
4448 | * | |
4449 | * We do the btree walk in the most optimal manner possible - we have sibling | |
4450 | * pointers so we can just walk all the blocks on each level from left to right | |
4451 | * in a single pass, and then move to the next level and do the same. We can | |
4452 | * also do readahead on the sibling pointers to get IO moving more quickly, | |
4453 | * though for slow disks this is unlikely to make much difference to performance | |
4454 | * as the amount of CPU work we have to do before moving to the next block is | |
4455 | * relatively small. | |
4456 | * | |
4457 | * For each btree block that we load, modify the owner appropriately, set the | |
4458 | * buffer as an ordered buffer and log it appropriately. We need to ensure that | |
4459 | * we mark the region we change dirty so that if the buffer is relogged in | |
4460 | * a subsequent transaction the changes we make here as an ordered buffer are | |
4461 | * correctly relogged in that transaction. If we are in recovery context, then | |
4462 | * just queue the modified buffer as delayed write buffer so the transaction | |
4463 | * recovery completion writes the changes to disk. | |
4464 | */ | |
f31736bf | 4465 | struct xfs_btree_block_change_owner_info { |
4a492e72 | 4466 | uint64_t new_owner; |
f31736bf DW |
4467 | struct list_head *buffer_list; |
4468 | }; | |
4469 | ||
9c6ebc42 DC |
4470 | static int |
4471 | xfs_btree_block_change_owner( | |
4472 | struct xfs_btree_cur *cur, | |
4473 | int level, | |
f31736bf | 4474 | void *data) |
9c6ebc42 | 4475 | { |
f31736bf | 4476 | struct xfs_btree_block_change_owner_info *bbcoi = data; |
9c6ebc42 DC |
4477 | struct xfs_btree_block *block; |
4478 | struct xfs_buf *bp; | |
9c6ebc42 DC |
4479 | |
4480 | /* modify the owner */ | |
4481 | block = xfs_btree_get_block(cur, level, &bp); | |
38fa71a7 BF |
4482 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) { |
4483 | if (block->bb_u.l.bb_owner == cpu_to_be64(bbcoi->new_owner)) | |
4484 | return 0; | |
f31736bf | 4485 | block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner); |
38fa71a7 BF |
4486 | } else { |
4487 | if (block->bb_u.s.bb_owner == cpu_to_be32(bbcoi->new_owner)) | |
4488 | return 0; | |
f31736bf | 4489 | block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner); |
38fa71a7 | 4490 | } |
9c6ebc42 DC |
4491 | |
4492 | /* | |
4493 | * If the block is a root block hosted in an inode, we might not have a | |
4494 | * buffer pointer here and we shouldn't attempt to log the change as the | |
4495 | * information is already held in the inode and discarded when the root | |
4496 | * block is formatted into the on-disk inode fork. We still change it, | |
4497 | * though, so everything is consistent in memory. | |
4498 | */ | |
38fa71a7 | 4499 | if (!bp) { |
9c6ebc42 DC |
4500 | ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE); |
4501 | ASSERT(level == cur->bc_nlevels - 1); | |
38fa71a7 BF |
4502 | return 0; |
4503 | } | |
4504 | ||
4505 | if (cur->bc_tp) { | |
4506 | if (!xfs_trans_ordered_buf(cur->bc_tp, bp)) { | |
4507 | xfs_btree_log_block(cur, bp, XFS_BB_OWNER); | |
4508 | return -EAGAIN; | |
4509 | } | |
4510 | } else { | |
4511 | xfs_buf_delwri_queue(bp, bbcoi->buffer_list); | |
9c6ebc42 DC |
4512 | } |
4513 | ||
f31736bf | 4514 | return 0; |
9c6ebc42 DC |
4515 | } |
4516 | ||
4517 | int | |
4518 | xfs_btree_change_owner( | |
4519 | struct xfs_btree_cur *cur, | |
4a492e72 | 4520 | uint64_t new_owner, |
9c6ebc42 DC |
4521 | struct list_head *buffer_list) |
4522 | { | |
f31736bf | 4523 | struct xfs_btree_block_change_owner_info bbcoi; |
9c6ebc42 | 4524 | |
f31736bf DW |
4525 | bbcoi.new_owner = new_owner; |
4526 | bbcoi.buffer_list = buffer_list; | |
9c6ebc42 | 4527 | |
f31736bf DW |
4528 | return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner, |
4529 | &bbcoi); | |
9c6ebc42 | 4530 | } |
dbca0167 | 4531 | |
fb75464f | 4532 | /* Verify the v5 fields of a long-format btree block. */ |
bc01119d | 4533 | xfs_failaddr_t |
fb75464f DW |
4534 | xfs_btree_lblock_v5hdr_verify( |
4535 | struct xfs_buf *bp, | |
4536 | uint64_t owner) | |
4537 | { | |
4538 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
4539 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); | |
4540 | ||
4541 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
bc01119d | 4542 | return __this_address; |
fb75464f | 4543 | if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid)) |
bc01119d | 4544 | return __this_address; |
fb75464f | 4545 | if (block->bb_u.l.bb_blkno != cpu_to_be64(bp->b_bn)) |
bc01119d | 4546 | return __this_address; |
fb75464f DW |
4547 | if (owner != XFS_RMAP_OWN_UNKNOWN && |
4548 | be64_to_cpu(block->bb_u.l.bb_owner) != owner) | |
bc01119d DW |
4549 | return __this_address; |
4550 | return NULL; | |
fb75464f DW |
4551 | } |
4552 | ||
4553 | /* Verify a long-format btree block. */ | |
bc01119d | 4554 | xfs_failaddr_t |
fb75464f DW |
4555 | xfs_btree_lblock_verify( |
4556 | struct xfs_buf *bp, | |
4557 | unsigned int max_recs) | |
4558 | { | |
4559 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
4560 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); | |
4561 | ||
4562 | /* numrecs verification */ | |
4563 | if (be16_to_cpu(block->bb_numrecs) > max_recs) | |
bc01119d | 4564 | return __this_address; |
fb75464f DW |
4565 | |
4566 | /* sibling pointer verification */ | |
4567 | if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) && | |
4568 | !xfs_verify_fsbno(mp, be64_to_cpu(block->bb_u.l.bb_leftsib))) | |
bc01119d | 4569 | return __this_address; |
fb75464f DW |
4570 | if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) && |
4571 | !xfs_verify_fsbno(mp, be64_to_cpu(block->bb_u.l.bb_rightsib))) | |
bc01119d | 4572 | return __this_address; |
fb75464f | 4573 | |
bc01119d | 4574 | return NULL; |
fb75464f DW |
4575 | } |
4576 | ||
dbca0167 DW |
4577 | /** |
4578 | * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format | |
4579 | * btree block | |
4580 | * | |
4581 | * @bp: buffer containing the btree block | |
4582 | * @max_recs: pointer to the m_*_mxr max records field in the xfs mount | |
4583 | * @pag_max_level: pointer to the per-ag max level field | |
4584 | */ | |
bc01119d | 4585 | xfs_failaddr_t |
dbca0167 DW |
4586 | xfs_btree_sblock_v5hdr_verify( |
4587 | struct xfs_buf *bp) | |
4588 | { | |
4589 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
4590 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); | |
4591 | struct xfs_perag *pag = bp->b_pag; | |
4592 | ||
4593 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
bc01119d | 4594 | return __this_address; |
dbca0167 | 4595 | if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid)) |
bc01119d | 4596 | return __this_address; |
dbca0167 | 4597 | if (block->bb_u.s.bb_blkno != cpu_to_be64(bp->b_bn)) |
bc01119d | 4598 | return __this_address; |
dbca0167 | 4599 | if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno) |
bc01119d DW |
4600 | return __this_address; |
4601 | return NULL; | |
dbca0167 DW |
4602 | } |
4603 | ||
4604 | /** | |
4605 | * xfs_btree_sblock_verify() -- verify a short-format btree block | |
4606 | * | |
4607 | * @bp: buffer containing the btree block | |
4608 | * @max_recs: maximum records allowed in this btree node | |
4609 | */ | |
bc01119d | 4610 | xfs_failaddr_t |
dbca0167 DW |
4611 | xfs_btree_sblock_verify( |
4612 | struct xfs_buf *bp, | |
4613 | unsigned int max_recs) | |
4614 | { | |
4615 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
4616 | struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp); | |
bb2a125a | 4617 | xfs_agblock_t agno; |
dbca0167 DW |
4618 | |
4619 | /* numrecs verification */ | |
4620 | if (be16_to_cpu(block->bb_numrecs) > max_recs) | |
bc01119d | 4621 | return __this_address; |
dbca0167 DW |
4622 | |
4623 | /* sibling pointer verification */ | |
bb2a125a DW |
4624 | agno = xfs_daddr_to_agno(mp, XFS_BUF_ADDR(bp)); |
4625 | if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) && | |
4626 | !xfs_verify_agbno(mp, agno, be32_to_cpu(block->bb_u.s.bb_leftsib))) | |
bc01119d | 4627 | return __this_address; |
bb2a125a DW |
4628 | if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) && |
4629 | !xfs_verify_agbno(mp, agno, be32_to_cpu(block->bb_u.s.bb_rightsib))) | |
bc01119d | 4630 | return __this_address; |
dbca0167 | 4631 | |
bc01119d | 4632 | return NULL; |
dbca0167 | 4633 | } |
730e2a19 DW |
4634 | |
4635 | /* | |
4636 | * Calculate the number of btree levels needed to store a given number of | |
4637 | * records in a short-format btree. | |
4638 | */ | |
4639 | uint | |
4640 | xfs_btree_compute_maxlevels( | |
4641 | struct xfs_mount *mp, | |
4642 | uint *limits, | |
4643 | unsigned long len) | |
4644 | { | |
4645 | uint level; | |
4646 | unsigned long maxblocks; | |
4647 | ||
4648 | maxblocks = (len + limits[0] - 1) / limits[0]; | |
4649 | for (level = 1; maxblocks > 1; level++) | |
4650 | maxblocks = (maxblocks + limits[1] - 1) / limits[1]; | |
4651 | return level; | |
4652 | } | |
5d3b7fe1 DW |
4653 | |
4654 | /* | |
4655 | * Query a regular btree for all records overlapping a given interval. | |
4656 | * Start with a LE lookup of the key of low_rec and return all records | |
4657 | * until we find a record with a key greater than the key of high_rec. | |
4658 | */ | |
4659 | STATIC int | |
4660 | xfs_btree_simple_query_range( | |
4661 | struct xfs_btree_cur *cur, | |
4662 | union xfs_btree_key *low_key, | |
4663 | union xfs_btree_key *high_key, | |
4664 | xfs_btree_query_range_fn fn, | |
4665 | void *priv) | |
4666 | { | |
4667 | union xfs_btree_rec *recp; | |
4668 | union xfs_btree_key rec_key; | |
4a492e72 | 4669 | int64_t diff; |
5d3b7fe1 DW |
4670 | int stat; |
4671 | bool firstrec = true; | |
4672 | int error; | |
4673 | ||
4674 | ASSERT(cur->bc_ops->init_high_key_from_rec); | |
4675 | ASSERT(cur->bc_ops->diff_two_keys); | |
4676 | ||
4677 | /* | |
4678 | * Find the leftmost record. The btree cursor must be set | |
4679 | * to the low record used to generate low_key. | |
4680 | */ | |
4681 | stat = 0; | |
4682 | error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat); | |
4683 | if (error) | |
4684 | goto out; | |
4685 | ||
a3654981 DW |
4686 | /* Nothing? See if there's anything to the right. */ |
4687 | if (!stat) { | |
4688 | error = xfs_btree_increment(cur, 0, &stat); | |
4689 | if (error) | |
4690 | goto out; | |
4691 | } | |
4692 | ||
5d3b7fe1 DW |
4693 | while (stat) { |
4694 | /* Find the record. */ | |
4695 | error = xfs_btree_get_rec(cur, &recp, &stat); | |
4696 | if (error || !stat) | |
4697 | break; | |
5d3b7fe1 DW |
4698 | |
4699 | /* Skip if high_key(rec) < low_key. */ | |
4700 | if (firstrec) { | |
94a22dec | 4701 | cur->bc_ops->init_high_key_from_rec(&rec_key, recp); |
5d3b7fe1 DW |
4702 | firstrec = false; |
4703 | diff = cur->bc_ops->diff_two_keys(cur, low_key, | |
4704 | &rec_key); | |
4705 | if (diff > 0) | |
4706 | goto advloop; | |
4707 | } | |
4708 | ||
4709 | /* Stop if high_key < low_key(rec). */ | |
94a22dec | 4710 | cur->bc_ops->init_key_from_rec(&rec_key, recp); |
5d3b7fe1 DW |
4711 | diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key); |
4712 | if (diff > 0) | |
4713 | break; | |
4714 | ||
4715 | /* Callback */ | |
4716 | error = fn(cur, recp, priv); | |
4717 | if (error < 0 || error == XFS_BTREE_QUERY_RANGE_ABORT) | |
4718 | break; | |
4719 | ||
4720 | advloop: | |
4721 | /* Move on to the next record. */ | |
4722 | error = xfs_btree_increment(cur, 0, &stat); | |
4723 | if (error) | |
4724 | break; | |
4725 | } | |
4726 | ||
4727 | out: | |
4728 | return error; | |
4729 | } | |
4730 | ||
4731 | /* | |
4732 | * Query an overlapped interval btree for all records overlapping a given | |
4733 | * interval. This function roughly follows the algorithm given in | |
4734 | * "Interval Trees" of _Introduction to Algorithms_, which is section | |
4735 | * 14.3 in the 2nd and 3rd editions. | |
4736 | * | |
4737 | * First, generate keys for the low and high records passed in. | |
4738 | * | |
4739 | * For any leaf node, generate the high and low keys for the record. | |
4740 | * If the record keys overlap with the query low/high keys, pass the | |
4741 | * record to the function iterator. | |
4742 | * | |
4743 | * For any internal node, compare the low and high keys of each | |
4744 | * pointer against the query low/high keys. If there's an overlap, | |
4745 | * follow the pointer. | |
4746 | * | |
4747 | * As an optimization, we stop scanning a block when we find a low key | |
4748 | * that is greater than the query's high key. | |
4749 | */ | |
4750 | STATIC int | |
4751 | xfs_btree_overlapped_query_range( | |
4752 | struct xfs_btree_cur *cur, | |
4753 | union xfs_btree_key *low_key, | |
4754 | union xfs_btree_key *high_key, | |
4755 | xfs_btree_query_range_fn fn, | |
4756 | void *priv) | |
4757 | { | |
4758 | union xfs_btree_ptr ptr; | |
4759 | union xfs_btree_ptr *pp; | |
4760 | union xfs_btree_key rec_key; | |
4761 | union xfs_btree_key rec_hkey; | |
4762 | union xfs_btree_key *lkp; | |
4763 | union xfs_btree_key *hkp; | |
4764 | union xfs_btree_rec *recp; | |
4765 | struct xfs_btree_block *block; | |
4a492e72 DW |
4766 | int64_t ldiff; |
4767 | int64_t hdiff; | |
5d3b7fe1 DW |
4768 | int level; |
4769 | struct xfs_buf *bp; | |
4770 | int i; | |
4771 | int error; | |
4772 | ||
4773 | /* Load the root of the btree. */ | |
4774 | level = cur->bc_nlevels - 1; | |
4775 | cur->bc_ops->init_ptr_from_cur(cur, &ptr); | |
4776 | error = xfs_btree_lookup_get_block(cur, level, &ptr, &block); | |
4777 | if (error) | |
4778 | return error; | |
4779 | xfs_btree_get_block(cur, level, &bp); | |
4780 | trace_xfs_btree_overlapped_query_range(cur, level, bp); | |
4781 | #ifdef DEBUG | |
4782 | error = xfs_btree_check_block(cur, block, level, bp); | |
4783 | if (error) | |
4784 | goto out; | |
4785 | #endif | |
4786 | cur->bc_ptrs[level] = 1; | |
4787 | ||
4788 | while (level < cur->bc_nlevels) { | |
4789 | block = xfs_btree_get_block(cur, level, &bp); | |
4790 | ||
4791 | /* End of node, pop back towards the root. */ | |
4792 | if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) { | |
4793 | pop_up: | |
4794 | if (level < cur->bc_nlevels - 1) | |
4795 | cur->bc_ptrs[level + 1]++; | |
4796 | level++; | |
4797 | continue; | |
4798 | } | |
4799 | ||
4800 | if (level == 0) { | |
4801 | /* Handle a leaf node. */ | |
4802 | recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block); | |
4803 | ||
4804 | cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp); | |
4805 | ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey, | |
4806 | low_key); | |
4807 | ||
4808 | cur->bc_ops->init_key_from_rec(&rec_key, recp); | |
4809 | hdiff = cur->bc_ops->diff_two_keys(cur, high_key, | |
4810 | &rec_key); | |
4811 | ||
4812 | /* | |
4813 | * If (record's high key >= query's low key) and | |
4814 | * (query's high key >= record's low key), then | |
4815 | * this record overlaps the query range; callback. | |
4816 | */ | |
4817 | if (ldiff >= 0 && hdiff >= 0) { | |
4818 | error = fn(cur, recp, priv); | |
4819 | if (error < 0 || | |
4820 | error == XFS_BTREE_QUERY_RANGE_ABORT) | |
4821 | break; | |
4822 | } else if (hdiff < 0) { | |
4823 | /* Record is larger than high key; pop. */ | |
4824 | goto pop_up; | |
4825 | } | |
4826 | cur->bc_ptrs[level]++; | |
4827 | continue; | |
4828 | } | |
4829 | ||
4830 | /* Handle an internal node. */ | |
4831 | lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block); | |
4832 | hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block); | |
4833 | pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block); | |
4834 | ||
4835 | ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key); | |
4836 | hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp); | |
4837 | ||
4838 | /* | |
4839 | * If (pointer's high key >= query's low key) and | |
4840 | * (query's high key >= pointer's low key), then | |
4841 | * this record overlaps the query range; follow pointer. | |
4842 | */ | |
4843 | if (ldiff >= 0 && hdiff >= 0) { | |
4844 | level--; | |
4845 | error = xfs_btree_lookup_get_block(cur, level, pp, | |
4846 | &block); | |
4847 | if (error) | |
4848 | goto out; | |
4849 | xfs_btree_get_block(cur, level, &bp); | |
4850 | trace_xfs_btree_overlapped_query_range(cur, level, bp); | |
4851 | #ifdef DEBUG | |
4852 | error = xfs_btree_check_block(cur, block, level, bp); | |
4853 | if (error) | |
4854 | goto out; | |
4855 | #endif | |
4856 | cur->bc_ptrs[level] = 1; | |
4857 | continue; | |
4858 | } else if (hdiff < 0) { | |
4859 | /* The low key is larger than the upper range; pop. */ | |
4860 | goto pop_up; | |
4861 | } | |
4862 | cur->bc_ptrs[level]++; | |
4863 | } | |
4864 | ||
4865 | out: | |
4866 | /* | |
4867 | * If we don't end this function with the cursor pointing at a record | |
4868 | * block, a subsequent non-error cursor deletion will not release | |
4869 | * node-level buffers, causing a buffer leak. This is quite possible | |
4870 | * with a zero-results range query, so release the buffers if we | |
4871 | * failed to return any results. | |
4872 | */ | |
4873 | if (cur->bc_bufs[0] == NULL) { | |
4874 | for (i = 0; i < cur->bc_nlevels; i++) { | |
4875 | if (cur->bc_bufs[i]) { | |
4876 | xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]); | |
4877 | cur->bc_bufs[i] = NULL; | |
4878 | cur->bc_ptrs[i] = 0; | |
4879 | cur->bc_ra[i] = 0; | |
4880 | } | |
4881 | } | |
4882 | } | |
4883 | ||
4884 | return error; | |
4885 | } | |
4886 | ||
4887 | /* | |
4888 | * Query a btree for all records overlapping a given interval of keys. The | |
4889 | * supplied function will be called with each record found; return one of the | |
4890 | * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error | |
4891 | * code. This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a | |
4892 | * negative error code. | |
4893 | */ | |
4894 | int | |
4895 | xfs_btree_query_range( | |
4896 | struct xfs_btree_cur *cur, | |
4897 | union xfs_btree_irec *low_rec, | |
4898 | union xfs_btree_irec *high_rec, | |
4899 | xfs_btree_query_range_fn fn, | |
4900 | void *priv) | |
4901 | { | |
4902 | union xfs_btree_rec rec; | |
4903 | union xfs_btree_key low_key; | |
4904 | union xfs_btree_key high_key; | |
4905 | ||
4906 | /* Find the keys of both ends of the interval. */ | |
4907 | cur->bc_rec = *high_rec; | |
4908 | cur->bc_ops->init_rec_from_cur(cur, &rec); | |
4909 | cur->bc_ops->init_key_from_rec(&high_key, &rec); | |
4910 | ||
4911 | cur->bc_rec = *low_rec; | |
4912 | cur->bc_ops->init_rec_from_cur(cur, &rec); | |
4913 | cur->bc_ops->init_key_from_rec(&low_key, &rec); | |
4914 | ||
4915 | /* Enforce low key < high key. */ | |
4916 | if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0) | |
4917 | return -EINVAL; | |
4918 | ||
4919 | if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING)) | |
4920 | return xfs_btree_simple_query_range(cur, &low_key, | |
4921 | &high_key, fn, priv); | |
4922 | return xfs_btree_overlapped_query_range(cur, &low_key, &high_key, | |
4923 | fn, priv); | |
4924 | } | |
fc8c849d | 4925 | |
7e05e856 DW |
4926 | /* Query a btree for all records. */ |
4927 | int | |
4928 | xfs_btree_query_all( | |
4929 | struct xfs_btree_cur *cur, | |
4930 | xfs_btree_query_range_fn fn, | |
4931 | void *priv) | |
4932 | { | |
f296858d DW |
4933 | union xfs_btree_key low_key; |
4934 | union xfs_btree_key high_key; | |
4935 | ||
4936 | memset(&cur->bc_rec, 0, sizeof(cur->bc_rec)); | |
4937 | memset(&low_key, 0, sizeof(low_key)); | |
4938 | memset(&high_key, 0xFF, sizeof(high_key)); | |
7e05e856 | 4939 | |
f296858d | 4940 | return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv); |
7e05e856 DW |
4941 | } |
4942 | ||
fc8c849d DW |
4943 | /* |
4944 | * Calculate the number of blocks needed to store a given number of records | |
4945 | * in a short-format (per-AG metadata) btree. | |
4946 | */ | |
4947 | xfs_extlen_t | |
4948 | xfs_btree_calc_size( | |
4949 | struct xfs_mount *mp, | |
4950 | uint *limits, | |
4951 | unsigned long long len) | |
4952 | { | |
4953 | int level; | |
4954 | int maxrecs; | |
4955 | xfs_extlen_t rval; | |
4956 | ||
4957 | maxrecs = limits[0]; | |
4958 | for (level = 0, rval = 0; len > 1; level++) { | |
4959 | len += maxrecs - 1; | |
4960 | do_div(len, maxrecs); | |
4961 | maxrecs = limits[1]; | |
4962 | rval += len; | |
4963 | } | |
4964 | return rval; | |
4965 | } | |
2cccc8ce | 4966 | |
1e85c20d | 4967 | static int |
2cccc8ce DW |
4968 | xfs_btree_count_blocks_helper( |
4969 | struct xfs_btree_cur *cur, | |
4970 | int level, | |
4971 | void *data) | |
4972 | { | |
4973 | xfs_extlen_t *blocks = data; | |
4974 | (*blocks)++; | |
4975 | ||
4976 | return 0; | |
4977 | } | |
4978 | ||
4979 | /* Count the blocks in a btree and return the result in *blocks. */ | |
4980 | int | |
4981 | xfs_btree_count_blocks( | |
4982 | struct xfs_btree_cur *cur, | |
4983 | xfs_extlen_t *blocks) | |
4984 | { | |
4985 | *blocks = 0; | |
4986 | return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper, | |
4987 | blocks); | |
4988 | } | |
15c4e6e4 DW |
4989 | |
4990 | /* Compare two btree pointers. */ | |
4991 | int64_t | |
4992 | xfs_btree_diff_two_ptrs( | |
4993 | struct xfs_btree_cur *cur, | |
4994 | const union xfs_btree_ptr *a, | |
4995 | const union xfs_btree_ptr *b) | |
4996 | { | |
4997 | if (cur->bc_flags & XFS_BTREE_LONG_PTRS) | |
4998 | return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l); | |
4999 | return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s); | |
5000 | } | |
1fe41a73 DW |
5001 | |
5002 | /* If there's an extent, we're done. */ | |
5003 | STATIC int | |
5004 | xfs_btree_has_record_helper( | |
5005 | struct xfs_btree_cur *cur, | |
5006 | union xfs_btree_rec *rec, | |
5007 | void *priv) | |
5008 | { | |
5009 | return XFS_BTREE_QUERY_RANGE_ABORT; | |
5010 | } | |
5011 | ||
5012 | /* Is there a record covering a given range of keys? */ | |
5013 | int | |
5014 | xfs_btree_has_record( | |
5015 | struct xfs_btree_cur *cur, | |
5016 | union xfs_btree_irec *low, | |
5017 | union xfs_btree_irec *high, | |
5018 | bool *exists) | |
5019 | { | |
5020 | int error; | |
5021 | ||
5022 | error = xfs_btree_query_range(cur, low, high, | |
5023 | &xfs_btree_has_record_helper, NULL); | |
5024 | if (error == XFS_BTREE_QUERY_RANGE_ABORT) { | |
5025 | *exists = true; | |
5026 | return 0; | |
5027 | } | |
5028 | *exists = false; | |
5029 | return error; | |
5030 | } |