Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 | 2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
f5ea1100 | 3 | * Copyright (c) 2013 Red Hat, Inc. |
7b718769 | 4 | * All Rights Reserved. |
1da177e4 | 5 | * |
7b718769 NS |
6 | * This program is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
8 | * published by the Free Software Foundation. |
9 | * | |
7b718769 NS |
10 | * This program is distributed in the hope that it would be useful, |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
1da177e4 | 14 | * |
7b718769 NS |
15 | * You should have received a copy of the GNU General Public License |
16 | * along with this program; if not, write the Free Software Foundation, | |
17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 18 | */ |
1da177e4 | 19 | #include "xfs.h" |
a844f451 | 20 | #include "xfs_fs.h" |
70a9883c | 21 | #include "xfs_shared.h" |
239880ef DC |
22 | #include "xfs_format.h" |
23 | #include "xfs_log_format.h" | |
24 | #include "xfs_trans_resv.h" | |
a844f451 | 25 | #include "xfs_bit.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
57062787 | 27 | #include "xfs_da_format.h" |
a844f451 | 28 | #include "xfs_da_btree.h" |
2b9ab5ab | 29 | #include "xfs_dir2.h" |
57926640 | 30 | #include "xfs_dir2_priv.h" |
1da177e4 | 31 | #include "xfs_inode.h" |
239880ef | 32 | #include "xfs_trans.h" |
a844f451 NS |
33 | #include "xfs_inode_item.h" |
34 | #include "xfs_alloc.h" | |
1da177e4 | 35 | #include "xfs_bmap.h" |
1da177e4 LT |
36 | #include "xfs_attr.h" |
37 | #include "xfs_attr_leaf.h" | |
1da177e4 | 38 | #include "xfs_error.h" |
0b1b213f | 39 | #include "xfs_trace.h" |
f5ea1100 DC |
40 | #include "xfs_cksum.h" |
41 | #include "xfs_buf_item.h" | |
a45086e2 | 42 | #include "xfs_log.h" |
1da177e4 LT |
43 | |
44 | /* | |
45 | * xfs_da_btree.c | |
46 | * | |
47 | * Routines to implement directories as Btrees of hashed names. | |
48 | */ | |
49 | ||
50 | /*======================================================================== | |
51 | * Function prototypes for the kernel. | |
52 | *========================================================================*/ | |
53 | ||
54 | /* | |
55 | * Routines used for growing the Btree. | |
56 | */ | |
f5ea1100 | 57 | STATIC int xfs_da3_root_split(xfs_da_state_t *state, |
1da177e4 LT |
58 | xfs_da_state_blk_t *existing_root, |
59 | xfs_da_state_blk_t *new_child); | |
f5ea1100 | 60 | STATIC int xfs_da3_node_split(xfs_da_state_t *state, |
1da177e4 LT |
61 | xfs_da_state_blk_t *existing_blk, |
62 | xfs_da_state_blk_t *split_blk, | |
63 | xfs_da_state_blk_t *blk_to_add, | |
64 | int treelevel, | |
65 | int *result); | |
f5ea1100 | 66 | STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state, |
1da177e4 LT |
67 | xfs_da_state_blk_t *node_blk_1, |
68 | xfs_da_state_blk_t *node_blk_2); | |
f5ea1100 | 69 | STATIC void xfs_da3_node_add(xfs_da_state_t *state, |
1da177e4 LT |
70 | xfs_da_state_blk_t *old_node_blk, |
71 | xfs_da_state_blk_t *new_node_blk); | |
72 | ||
73 | /* | |
74 | * Routines used for shrinking the Btree. | |
75 | */ | |
f5ea1100 | 76 | STATIC int xfs_da3_root_join(xfs_da_state_t *state, |
1da177e4 | 77 | xfs_da_state_blk_t *root_blk); |
f5ea1100 DC |
78 | STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval); |
79 | STATIC void xfs_da3_node_remove(xfs_da_state_t *state, | |
1da177e4 | 80 | xfs_da_state_blk_t *drop_blk); |
f5ea1100 | 81 | STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state, |
1da177e4 LT |
82 | xfs_da_state_blk_t *src_node_blk, |
83 | xfs_da_state_blk_t *dst_node_blk); | |
84 | ||
85 | /* | |
86 | * Utility routines. | |
87 | */ | |
f5ea1100 | 88 | STATIC int xfs_da3_blk_unlink(xfs_da_state_t *state, |
ba0f32d4 CH |
89 | xfs_da_state_blk_t *drop_blk, |
90 | xfs_da_state_blk_t *save_blk); | |
1da177e4 | 91 | |
f5ea1100 DC |
92 | |
93 | kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */ | |
94 | ||
95 | /* | |
96 | * Allocate a dir-state structure. | |
97 | * We don't put them on the stack since they're large. | |
98 | */ | |
99 | xfs_da_state_t * | |
100 | xfs_da_state_alloc(void) | |
101 | { | |
102 | return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS); | |
103 | } | |
104 | ||
105 | /* | |
106 | * Kill the altpath contents of a da-state structure. | |
107 | */ | |
108 | STATIC void | |
109 | xfs_da_state_kill_altpath(xfs_da_state_t *state) | |
110 | { | |
111 | int i; | |
112 | ||
113 | for (i = 0; i < state->altpath.active; i++) | |
114 | state->altpath.blk[i].bp = NULL; | |
115 | state->altpath.active = 0; | |
116 | } | |
117 | ||
118 | /* | |
119 | * Free a da-state structure. | |
120 | */ | |
121 | void | |
122 | xfs_da_state_free(xfs_da_state_t *state) | |
123 | { | |
124 | xfs_da_state_kill_altpath(state); | |
125 | #ifdef DEBUG | |
126 | memset((char *)state, 0, sizeof(*state)); | |
127 | #endif /* DEBUG */ | |
128 | kmem_zone_free(xfs_da_state_zone, state); | |
129 | } | |
130 | ||
a6a781a5 | 131 | static xfs_failaddr_t |
f5ea1100 | 132 | xfs_da3_node_verify( |
d9392a4b DC |
133 | struct xfs_buf *bp) |
134 | { | |
135 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
f5ea1100 DC |
136 | struct xfs_da_intnode *hdr = bp->b_addr; |
137 | struct xfs_da3_icnode_hdr ichdr; | |
01ba43b8 DC |
138 | const struct xfs_dir_ops *ops; |
139 | ||
140 | ops = xfs_dir_get_ops(mp, NULL); | |
f5ea1100 | 141 | |
01ba43b8 | 142 | ops->node_hdr_from_disk(&ichdr, hdr); |
f5ea1100 DC |
143 | |
144 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
145 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; | |
146 | ||
147 | if (ichdr.magic != XFS_DA3_NODE_MAGIC) | |
a6a781a5 | 148 | return __this_address; |
f5ea1100 | 149 | |
ce748eaa | 150 | if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid)) |
a6a781a5 | 151 | return __this_address; |
f5ea1100 | 152 | if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn) |
a6a781a5 | 153 | return __this_address; |
a45086e2 | 154 | if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn))) |
a6a781a5 | 155 | return __this_address; |
f5ea1100 DC |
156 | } else { |
157 | if (ichdr.magic != XFS_DA_NODE_MAGIC) | |
a6a781a5 | 158 | return __this_address; |
d9392a4b | 159 | } |
f5ea1100 | 160 | if (ichdr.level == 0) |
a6a781a5 | 161 | return __this_address; |
f5ea1100 | 162 | if (ichdr.level > XFS_DA_NODE_MAXDEPTH) |
a6a781a5 | 163 | return __this_address; |
f5ea1100 | 164 | if (ichdr.count == 0) |
a6a781a5 | 165 | return __this_address; |
f5ea1100 DC |
166 | |
167 | /* | |
168 | * we don't know if the node is for and attribute or directory tree, | |
169 | * so only fail if the count is outside both bounds | |
170 | */ | |
7ab610f9 DC |
171 | if (ichdr.count > mp->m_dir_geo->node_ents && |
172 | ichdr.count > mp->m_attr_geo->node_ents) | |
a6a781a5 | 173 | return __this_address; |
f5ea1100 DC |
174 | |
175 | /* XXX: hash order check? */ | |
d9392a4b | 176 | |
a6a781a5 | 177 | return NULL; |
d9392a4b DC |
178 | } |
179 | ||
180 | static void | |
f5ea1100 | 181 | xfs_da3_node_write_verify( |
612cfbfe DC |
182 | struct xfs_buf *bp) |
183 | { | |
f5ea1100 DC |
184 | struct xfs_mount *mp = bp->b_target->bt_mount; |
185 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
186 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; | |
187 | ||
a6a781a5 | 188 | if (xfs_da3_node_verify(bp)) { |
31ca03c9 | 189 | xfs_verifier_error(bp, -EFSCORRUPTED); |
f5ea1100 DC |
190 | return; |
191 | } | |
192 | ||
193 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
194 | return; | |
195 | ||
196 | if (bip) | |
197 | hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
198 | ||
f1dbcd7e | 199 | xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF); |
612cfbfe DC |
200 | } |
201 | ||
1813dd64 DC |
202 | /* |
203 | * leaf/node format detection on trees is sketchy, so a node read can be done on | |
204 | * leaf level blocks when detection identifies the tree as a node format tree | |
205 | * incorrectly. In this case, we need to swap the verifier to match the correct | |
206 | * format of the block being read. | |
207 | */ | |
612cfbfe | 208 | static void |
f5ea1100 | 209 | xfs_da3_node_read_verify( |
d9392a4b DC |
210 | struct xfs_buf *bp) |
211 | { | |
d9392a4b DC |
212 | struct xfs_da_blkinfo *info = bp->b_addr; |
213 | ||
214 | switch (be16_to_cpu(info->magic)) { | |
f5ea1100 | 215 | case XFS_DA3_NODE_MAGIC: |
ce5028cf | 216 | if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) { |
31ca03c9 | 217 | xfs_verifier_error(bp, -EFSBADCRC); |
f5ea1100 | 218 | break; |
ce5028cf | 219 | } |
f5ea1100 | 220 | /* fall through */ |
d9392a4b | 221 | case XFS_DA_NODE_MAGIC: |
a6a781a5 | 222 | if (xfs_da3_node_verify(bp)) |
31ca03c9 | 223 | xfs_verifier_error(bp, -EFSCORRUPTED); |
f5ea1100 | 224 | return; |
d9392a4b | 225 | case XFS_ATTR_LEAF_MAGIC: |
7ced60ca | 226 | case XFS_ATTR3_LEAF_MAGIC: |
517c2220 | 227 | bp->b_ops = &xfs_attr3_leaf_buf_ops; |
1813dd64 | 228 | bp->b_ops->verify_read(bp); |
d9392a4b DC |
229 | return; |
230 | case XFS_DIR2_LEAFN_MAGIC: | |
24df33b4 DC |
231 | case XFS_DIR3_LEAFN_MAGIC: |
232 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
1813dd64 | 233 | bp->b_ops->verify_read(bp); |
d9392a4b DC |
234 | return; |
235 | default: | |
31ca03c9 | 236 | xfs_verifier_error(bp, -EFSCORRUPTED); |
d9392a4b DC |
237 | break; |
238 | } | |
d9392a4b DC |
239 | } |
240 | ||
f5ea1100 | 241 | const struct xfs_buf_ops xfs_da3_node_buf_ops = { |
233135b7 | 242 | .name = "xfs_da3_node", |
f5ea1100 DC |
243 | .verify_read = xfs_da3_node_read_verify, |
244 | .verify_write = xfs_da3_node_write_verify, | |
1813dd64 DC |
245 | }; |
246 | ||
d9392a4b | 247 | int |
f5ea1100 | 248 | xfs_da3_node_read( |
d9392a4b DC |
249 | struct xfs_trans *tp, |
250 | struct xfs_inode *dp, | |
251 | xfs_dablk_t bno, | |
252 | xfs_daddr_t mappedbno, | |
253 | struct xfs_buf **bpp, | |
254 | int which_fork) | |
255 | { | |
d75afeb3 DC |
256 | int err; |
257 | ||
258 | err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp, | |
f5ea1100 | 259 | which_fork, &xfs_da3_node_buf_ops); |
cd87d867 | 260 | if (!err && tp && *bpp) { |
d75afeb3 DC |
261 | struct xfs_da_blkinfo *info = (*bpp)->b_addr; |
262 | int type; | |
263 | ||
264 | switch (be16_to_cpu(info->magic)) { | |
d75afeb3 | 265 | case XFS_DA_NODE_MAGIC: |
cab09a81 | 266 | case XFS_DA3_NODE_MAGIC: |
61fe135c | 267 | type = XFS_BLFT_DA_NODE_BUF; |
d75afeb3 DC |
268 | break; |
269 | case XFS_ATTR_LEAF_MAGIC: | |
270 | case XFS_ATTR3_LEAF_MAGIC: | |
61fe135c | 271 | type = XFS_BLFT_ATTR_LEAF_BUF; |
d75afeb3 DC |
272 | break; |
273 | case XFS_DIR2_LEAFN_MAGIC: | |
274 | case XFS_DIR3_LEAFN_MAGIC: | |
61fe135c | 275 | type = XFS_BLFT_DIR_LEAFN_BUF; |
d75afeb3 DC |
276 | break; |
277 | default: | |
278 | type = 0; | |
279 | ASSERT(0); | |
280 | break; | |
281 | } | |
282 | xfs_trans_buf_set_type(tp, *bpp, type); | |
283 | } | |
284 | return err; | |
d9392a4b DC |
285 | } |
286 | ||
1da177e4 LT |
287 | /*======================================================================== |
288 | * Routines used for growing the Btree. | |
289 | *========================================================================*/ | |
290 | ||
291 | /* | |
292 | * Create the initial contents of an intermediate node. | |
293 | */ | |
294 | int | |
f5ea1100 DC |
295 | xfs_da3_node_create( |
296 | struct xfs_da_args *args, | |
297 | xfs_dablk_t blkno, | |
298 | int level, | |
299 | struct xfs_buf **bpp, | |
300 | int whichfork) | |
1da177e4 | 301 | { |
f5ea1100 DC |
302 | struct xfs_da_intnode *node; |
303 | struct xfs_trans *tp = args->trans; | |
304 | struct xfs_mount *mp = tp->t_mountp; | |
305 | struct xfs_da3_icnode_hdr ichdr = {0}; | |
306 | struct xfs_buf *bp; | |
307 | int error; | |
01ba43b8 | 308 | struct xfs_inode *dp = args->dp; |
1da177e4 | 309 | |
5a5881cd | 310 | trace_xfs_da_node_create(args); |
f5ea1100 | 311 | ASSERT(level <= XFS_DA_NODE_MAXDEPTH); |
5a5881cd | 312 | |
01ba43b8 | 313 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork); |
1da177e4 | 314 | if (error) |
d99831ff | 315 | return error; |
d75afeb3 | 316 | bp->b_ops = &xfs_da3_node_buf_ops; |
61fe135c | 317 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); |
1d9025e5 | 318 | node = bp->b_addr; |
1da177e4 | 319 | |
f5ea1100 DC |
320 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
321 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; | |
322 | ||
a45086e2 | 323 | memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr)); |
f5ea1100 DC |
324 | ichdr.magic = XFS_DA3_NODE_MAGIC; |
325 | hdr3->info.blkno = cpu_to_be64(bp->b_bn); | |
326 | hdr3->info.owner = cpu_to_be64(args->dp->i_ino); | |
ce748eaa | 327 | uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid); |
f5ea1100 DC |
328 | } else { |
329 | ichdr.magic = XFS_DA_NODE_MAGIC; | |
330 | } | |
331 | ichdr.level = level; | |
332 | ||
01ba43b8 | 333 | dp->d_ops->node_hdr_to_disk(node, &ichdr); |
1d9025e5 | 334 | xfs_trans_log_buf(tp, bp, |
1c9a5b2e | 335 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
1da177e4 LT |
336 | |
337 | *bpp = bp; | |
d99831ff | 338 | return 0; |
1da177e4 LT |
339 | } |
340 | ||
341 | /* | |
342 | * Split a leaf node, rebalance, then possibly split | |
343 | * intermediate nodes, rebalance, etc. | |
344 | */ | |
345 | int /* error */ | |
f5ea1100 DC |
346 | xfs_da3_split( |
347 | struct xfs_da_state *state) | |
1da177e4 | 348 | { |
f5ea1100 DC |
349 | struct xfs_da_state_blk *oldblk; |
350 | struct xfs_da_state_blk *newblk; | |
351 | struct xfs_da_state_blk *addblk; | |
352 | struct xfs_da_intnode *node; | |
f5ea1100 | 353 | int max; |
836a94ad | 354 | int action = 0; |
f5ea1100 DC |
355 | int error; |
356 | int i; | |
1da177e4 | 357 | |
5a5881cd DC |
358 | trace_xfs_da_split(state->args); |
359 | ||
1da177e4 LT |
360 | /* |
361 | * Walk back up the tree splitting/inserting/adjusting as necessary. | |
362 | * If we need to insert and there isn't room, split the node, then | |
363 | * decide which fragment to insert the new block from below into. | |
364 | * Note that we may split the root this way, but we need more fixup. | |
365 | */ | |
366 | max = state->path.active - 1; | |
367 | ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH)); | |
368 | ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC || | |
f6c2d1fa | 369 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); |
1da177e4 LT |
370 | |
371 | addblk = &state->path.blk[max]; /* initial dummy value */ | |
372 | for (i = max; (i >= 0) && addblk; state->path.active--, i--) { | |
373 | oldblk = &state->path.blk[i]; | |
374 | newblk = &state->altpath.blk[i]; | |
375 | ||
376 | /* | |
377 | * If a leaf node then | |
378 | * Allocate a new leaf node, then rebalance across them. | |
379 | * else if an intermediate node then | |
380 | * We split on the last layer, must we split the node? | |
381 | */ | |
382 | switch (oldblk->magic) { | |
383 | case XFS_ATTR_LEAF_MAGIC: | |
517c2220 | 384 | error = xfs_attr3_leaf_split(state, oldblk, newblk); |
2451337d | 385 | if ((error != 0) && (error != -ENOSPC)) { |
d99831ff | 386 | return error; /* GROT: attr is inconsistent */ |
1da177e4 LT |
387 | } |
388 | if (!error) { | |
389 | addblk = newblk; | |
390 | break; | |
391 | } | |
392 | /* | |
160ae76f DC |
393 | * Entry wouldn't fit, split the leaf again. The new |
394 | * extrablk will be consumed by xfs_da3_node_split if | |
395 | * the node is split. | |
1da177e4 LT |
396 | */ |
397 | state->extravalid = 1; | |
398 | if (state->inleaf) { | |
399 | state->extraafter = 0; /* before newblk */ | |
5a5881cd | 400 | trace_xfs_attr_leaf_split_before(state->args); |
517c2220 | 401 | error = xfs_attr3_leaf_split(state, oldblk, |
1da177e4 LT |
402 | &state->extrablk); |
403 | } else { | |
404 | state->extraafter = 1; /* after newblk */ | |
5a5881cd | 405 | trace_xfs_attr_leaf_split_after(state->args); |
517c2220 | 406 | error = xfs_attr3_leaf_split(state, newblk, |
1da177e4 LT |
407 | &state->extrablk); |
408 | } | |
409 | if (error) | |
d99831ff | 410 | return error; /* GROT: attr inconsistent */ |
1da177e4 LT |
411 | addblk = newblk; |
412 | break; | |
1da177e4 | 413 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
414 | error = xfs_dir2_leafn_split(state, oldblk, newblk); |
415 | if (error) | |
416 | return error; | |
417 | addblk = newblk; | |
418 | break; | |
419 | case XFS_DA_NODE_MAGIC: | |
f5ea1100 | 420 | error = xfs_da3_node_split(state, oldblk, newblk, addblk, |
1da177e4 | 421 | max - i, &action); |
1da177e4 LT |
422 | addblk->bp = NULL; |
423 | if (error) | |
d99831ff | 424 | return error; /* GROT: dir is inconsistent */ |
1da177e4 LT |
425 | /* |
426 | * Record the newly split block for the next time thru? | |
427 | */ | |
428 | if (action) | |
429 | addblk = newblk; | |
430 | else | |
431 | addblk = NULL; | |
432 | break; | |
433 | } | |
434 | ||
435 | /* | |
436 | * Update the btree to show the new hashval for this child. | |
437 | */ | |
f5ea1100 | 438 | xfs_da3_fixhashpath(state, &state->path); |
1da177e4 LT |
439 | } |
440 | if (!addblk) | |
d99831ff | 441 | return 0; |
1da177e4 | 442 | |
160ae76f DC |
443 | /* |
444 | * xfs_da3_node_split() should have consumed any extra blocks we added | |
445 | * during a double leaf split in the attr fork. This is guaranteed as | |
446 | * we can't be here if the attr fork only has a single leaf block. | |
447 | */ | |
448 | ASSERT(state->extravalid == 0 || | |
449 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); | |
450 | ||
1da177e4 LT |
451 | /* |
452 | * Split the root node. | |
453 | */ | |
454 | ASSERT(state->path.active == 0); | |
455 | oldblk = &state->path.blk[0]; | |
f5ea1100 | 456 | error = xfs_da3_root_split(state, oldblk, addblk); |
1da177e4 | 457 | if (error) { |
1da177e4 | 458 | addblk->bp = NULL; |
d99831ff | 459 | return error; /* GROT: dir is inconsistent */ |
1da177e4 LT |
460 | } |
461 | ||
462 | /* | |
160ae76f DC |
463 | * Update pointers to the node which used to be block 0 and just got |
464 | * bumped because of the addition of a new root node. Note that the | |
465 | * original block 0 could be at any position in the list of blocks in | |
466 | * the tree. | |
f5ea1100 | 467 | * |
160ae76f DC |
468 | * Note: the magic numbers and sibling pointers are in the same physical |
469 | * place for both v2 and v3 headers (by design). Hence it doesn't matter | |
470 | * which version of the xfs_da_intnode structure we use here as the | |
471 | * result will be the same using either structure. | |
1da177e4 | 472 | */ |
1d9025e5 | 473 | node = oldblk->bp->b_addr; |
1da177e4 | 474 | if (node->hdr.info.forw) { |
160ae76f DC |
475 | ASSERT(be32_to_cpu(node->hdr.info.forw) == addblk->blkno); |
476 | node = addblk->bp->b_addr; | |
89da0544 | 477 | node->hdr.info.back = cpu_to_be32(oldblk->blkno); |
160ae76f DC |
478 | xfs_trans_log_buf(state->args->trans, addblk->bp, |
479 | XFS_DA_LOGRANGE(node, &node->hdr.info, | |
480 | sizeof(node->hdr.info))); | |
1da177e4 | 481 | } |
1d9025e5 | 482 | node = oldblk->bp->b_addr; |
89da0544 | 483 | if (node->hdr.info.back) { |
160ae76f DC |
484 | ASSERT(be32_to_cpu(node->hdr.info.back) == addblk->blkno); |
485 | node = addblk->bp->b_addr; | |
89da0544 | 486 | node->hdr.info.forw = cpu_to_be32(oldblk->blkno); |
160ae76f DC |
487 | xfs_trans_log_buf(state->args->trans, addblk->bp, |
488 | XFS_DA_LOGRANGE(node, &node->hdr.info, | |
489 | sizeof(node->hdr.info))); | |
1da177e4 | 490 | } |
1da177e4 | 491 | addblk->bp = NULL; |
d99831ff | 492 | return 0; |
1da177e4 LT |
493 | } |
494 | ||
495 | /* | |
496 | * Split the root. We have to create a new root and point to the two | |
497 | * parts (the split old root) that we just created. Copy block zero to | |
498 | * the EOF, extending the inode in process. | |
499 | */ | |
500 | STATIC int /* error */ | |
f5ea1100 DC |
501 | xfs_da3_root_split( |
502 | struct xfs_da_state *state, | |
503 | struct xfs_da_state_blk *blk1, | |
504 | struct xfs_da_state_blk *blk2) | |
1da177e4 | 505 | { |
f5ea1100 DC |
506 | struct xfs_da_intnode *node; |
507 | struct xfs_da_intnode *oldroot; | |
508 | struct xfs_da_node_entry *btree; | |
509 | struct xfs_da3_icnode_hdr nodehdr; | |
510 | struct xfs_da_args *args; | |
511 | struct xfs_buf *bp; | |
512 | struct xfs_inode *dp; | |
513 | struct xfs_trans *tp; | |
f5ea1100 DC |
514 | struct xfs_dir2_leaf *leaf; |
515 | xfs_dablk_t blkno; | |
516 | int level; | |
517 | int error; | |
518 | int size; | |
1da177e4 | 519 | |
5a5881cd DC |
520 | trace_xfs_da_root_split(state->args); |
521 | ||
1da177e4 LT |
522 | /* |
523 | * Copy the existing (incorrect) block from the root node position | |
524 | * to a free space somewhere. | |
525 | */ | |
526 | args = state->args; | |
1da177e4 LT |
527 | error = xfs_da_grow_inode(args, &blkno); |
528 | if (error) | |
f5ea1100 DC |
529 | return error; |
530 | ||
1da177e4 LT |
531 | dp = args->dp; |
532 | tp = args->trans; | |
1da177e4 LT |
533 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork); |
534 | if (error) | |
f5ea1100 | 535 | return error; |
1d9025e5 DC |
536 | node = bp->b_addr; |
537 | oldroot = blk1->bp->b_addr; | |
f5ea1100 DC |
538 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
539 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) { | |
29916df0 | 540 | struct xfs_da3_icnode_hdr icnodehdr; |
f5ea1100 | 541 | |
29916df0 | 542 | dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot); |
4bceb18f | 543 | btree = dp->d_ops->node_tree_p(oldroot); |
29916df0 FF |
544 | size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot); |
545 | level = icnodehdr.level; | |
d75afeb3 DC |
546 | |
547 | /* | |
548 | * we are about to copy oldroot to bp, so set up the type | |
549 | * of bp while we know exactly what it will be. | |
550 | */ | |
61fe135c | 551 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); |
1da177e4 | 552 | } else { |
24df33b4 DC |
553 | struct xfs_dir3_icleaf_hdr leafhdr; |
554 | struct xfs_dir2_leaf_entry *ents; | |
555 | ||
1da177e4 | 556 | leaf = (xfs_dir2_leaf_t *)oldroot; |
01ba43b8 | 557 | dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf); |
4141956a | 558 | ents = dp->d_ops->leaf_ents_p(leaf); |
24df33b4 DC |
559 | |
560 | ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC || | |
561 | leafhdr.magic == XFS_DIR3_LEAFN_MAGIC); | |
562 | size = (int)((char *)&ents[leafhdr.count] - (char *)leaf); | |
f5ea1100 | 563 | level = 0; |
d75afeb3 DC |
564 | |
565 | /* | |
566 | * we are about to copy oldroot to bp, so set up the type | |
567 | * of bp while we know exactly what it will be. | |
568 | */ | |
61fe135c | 569 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF); |
1da177e4 | 570 | } |
f5ea1100 DC |
571 | |
572 | /* | |
573 | * we can copy most of the information in the node from one block to | |
574 | * another, but for CRC enabled headers we have to make sure that the | |
575 | * block specific identifiers are kept intact. We update the buffer | |
576 | * directly for this. | |
577 | */ | |
1da177e4 | 578 | memcpy(node, oldroot, size); |
f5ea1100 DC |
579 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) || |
580 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
581 | struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node; | |
582 | ||
583 | node3->hdr.info.blkno = cpu_to_be64(bp->b_bn); | |
584 | } | |
1d9025e5 | 585 | xfs_trans_log_buf(tp, bp, 0, size - 1); |
b0f539de | 586 | |
1813dd64 | 587 | bp->b_ops = blk1->bp->b_ops; |
0a4edc8f | 588 | xfs_trans_buf_copy_type(bp, blk1->bp); |
1da177e4 LT |
589 | blk1->bp = bp; |
590 | blk1->blkno = blkno; | |
591 | ||
592 | /* | |
593 | * Set up the new root node. | |
594 | */ | |
f5ea1100 | 595 | error = xfs_da3_node_create(args, |
7dda6e86 | 596 | (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0, |
f5ea1100 | 597 | level + 1, &bp, args->whichfork); |
1da177e4 | 598 | if (error) |
f5ea1100 DC |
599 | return error; |
600 | ||
1d9025e5 | 601 | node = bp->b_addr; |
01ba43b8 | 602 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
4bceb18f | 603 | btree = dp->d_ops->node_tree_p(node); |
f5ea1100 DC |
604 | btree[0].hashval = cpu_to_be32(blk1->hashval); |
605 | btree[0].before = cpu_to_be32(blk1->blkno); | |
606 | btree[1].hashval = cpu_to_be32(blk2->hashval); | |
607 | btree[1].before = cpu_to_be32(blk2->blkno); | |
608 | nodehdr.count = 2; | |
01ba43b8 | 609 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
1da177e4 LT |
610 | |
611 | #ifdef DEBUG | |
24df33b4 DC |
612 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
613 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
7dda6e86 DC |
614 | ASSERT(blk1->blkno >= args->geo->leafblk && |
615 | blk1->blkno < args->geo->freeblk); | |
616 | ASSERT(blk2->blkno >= args->geo->leafblk && | |
617 | blk2->blkno < args->geo->freeblk); | |
1da177e4 LT |
618 | } |
619 | #endif | |
620 | ||
621 | /* Header is already logged by xfs_da_node_create */ | |
1d9025e5 | 622 | xfs_trans_log_buf(tp, bp, |
f5ea1100 | 623 | XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2)); |
1da177e4 | 624 | |
f5ea1100 | 625 | return 0; |
1da177e4 LT |
626 | } |
627 | ||
628 | /* | |
629 | * Split the node, rebalance, then add the new entry. | |
630 | */ | |
631 | STATIC int /* error */ | |
f5ea1100 DC |
632 | xfs_da3_node_split( |
633 | struct xfs_da_state *state, | |
634 | struct xfs_da_state_blk *oldblk, | |
635 | struct xfs_da_state_blk *newblk, | |
636 | struct xfs_da_state_blk *addblk, | |
637 | int treelevel, | |
638 | int *result) | |
1da177e4 | 639 | { |
f5ea1100 DC |
640 | struct xfs_da_intnode *node; |
641 | struct xfs_da3_icnode_hdr nodehdr; | |
642 | xfs_dablk_t blkno; | |
643 | int newcount; | |
644 | int error; | |
645 | int useextra; | |
01ba43b8 | 646 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 647 | |
5a5881cd DC |
648 | trace_xfs_da_node_split(state->args); |
649 | ||
1d9025e5 | 650 | node = oldblk->bp->b_addr; |
01ba43b8 | 651 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
1da177e4 LT |
652 | |
653 | /* | |
f6c2d1fa | 654 | * With V2 dirs the extra block is data or freespace. |
1da177e4 | 655 | */ |
f6c2d1fa | 656 | useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK; |
1da177e4 LT |
657 | newcount = 1 + useextra; |
658 | /* | |
659 | * Do we have to split the node? | |
660 | */ | |
b2a21e7a | 661 | if (nodehdr.count + newcount > state->args->geo->node_ents) { |
1da177e4 LT |
662 | /* |
663 | * Allocate a new node, add to the doubly linked chain of | |
664 | * nodes, then move some of our excess entries into it. | |
665 | */ | |
666 | error = xfs_da_grow_inode(state->args, &blkno); | |
667 | if (error) | |
d99831ff | 668 | return error; /* GROT: dir is inconsistent */ |
1da177e4 | 669 | |
f5ea1100 | 670 | error = xfs_da3_node_create(state->args, blkno, treelevel, |
1da177e4 LT |
671 | &newblk->bp, state->args->whichfork); |
672 | if (error) | |
d99831ff | 673 | return error; /* GROT: dir is inconsistent */ |
1da177e4 LT |
674 | newblk->blkno = blkno; |
675 | newblk->magic = XFS_DA_NODE_MAGIC; | |
f5ea1100 DC |
676 | xfs_da3_node_rebalance(state, oldblk, newblk); |
677 | error = xfs_da3_blk_link(state, oldblk, newblk); | |
1da177e4 | 678 | if (error) |
d99831ff | 679 | return error; |
1da177e4 LT |
680 | *result = 1; |
681 | } else { | |
682 | *result = 0; | |
683 | } | |
684 | ||
685 | /* | |
686 | * Insert the new entry(s) into the correct block | |
687 | * (updating last hashval in the process). | |
688 | * | |
f5ea1100 | 689 | * xfs_da3_node_add() inserts BEFORE the given index, |
1da177e4 LT |
690 | * and as a result of using node_lookup_int() we always |
691 | * point to a valid entry (not after one), but a split | |
692 | * operation always results in a new block whose hashvals | |
693 | * FOLLOW the current block. | |
694 | * | |
695 | * If we had double-split op below us, then add the extra block too. | |
696 | */ | |
1d9025e5 | 697 | node = oldblk->bp->b_addr; |
01ba43b8 | 698 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
f5ea1100 | 699 | if (oldblk->index <= nodehdr.count) { |
1da177e4 | 700 | oldblk->index++; |
f5ea1100 | 701 | xfs_da3_node_add(state, oldblk, addblk); |
1da177e4 LT |
702 | if (useextra) { |
703 | if (state->extraafter) | |
704 | oldblk->index++; | |
f5ea1100 | 705 | xfs_da3_node_add(state, oldblk, &state->extrablk); |
1da177e4 LT |
706 | state->extravalid = 0; |
707 | } | |
708 | } else { | |
709 | newblk->index++; | |
f5ea1100 | 710 | xfs_da3_node_add(state, newblk, addblk); |
1da177e4 LT |
711 | if (useextra) { |
712 | if (state->extraafter) | |
713 | newblk->index++; | |
f5ea1100 | 714 | xfs_da3_node_add(state, newblk, &state->extrablk); |
1da177e4 LT |
715 | state->extravalid = 0; |
716 | } | |
717 | } | |
718 | ||
d99831ff | 719 | return 0; |
1da177e4 LT |
720 | } |
721 | ||
722 | /* | |
723 | * Balance the btree elements between two intermediate nodes, | |
724 | * usually one full and one empty. | |
725 | * | |
726 | * NOTE: if blk2 is empty, then it will get the upper half of blk1. | |
727 | */ | |
728 | STATIC void | |
f5ea1100 DC |
729 | xfs_da3_node_rebalance( |
730 | struct xfs_da_state *state, | |
731 | struct xfs_da_state_blk *blk1, | |
732 | struct xfs_da_state_blk *blk2) | |
1da177e4 | 733 | { |
f5ea1100 DC |
734 | struct xfs_da_intnode *node1; |
735 | struct xfs_da_intnode *node2; | |
736 | struct xfs_da_intnode *tmpnode; | |
737 | struct xfs_da_node_entry *btree1; | |
738 | struct xfs_da_node_entry *btree2; | |
739 | struct xfs_da_node_entry *btree_s; | |
740 | struct xfs_da_node_entry *btree_d; | |
741 | struct xfs_da3_icnode_hdr nodehdr1; | |
742 | struct xfs_da3_icnode_hdr nodehdr2; | |
743 | struct xfs_trans *tp; | |
744 | int count; | |
745 | int tmp; | |
746 | int swap = 0; | |
4bceb18f | 747 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 748 | |
5a5881cd DC |
749 | trace_xfs_da_node_rebalance(state->args); |
750 | ||
1d9025e5 DC |
751 | node1 = blk1->bp->b_addr; |
752 | node2 = blk2->bp->b_addr; | |
01ba43b8 DC |
753 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
754 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); | |
4bceb18f DC |
755 | btree1 = dp->d_ops->node_tree_p(node1); |
756 | btree2 = dp->d_ops->node_tree_p(node2); | |
f5ea1100 | 757 | |
1da177e4 LT |
758 | /* |
759 | * Figure out how many entries need to move, and in which direction. | |
760 | * Swap the nodes around if that makes it simpler. | |
761 | */ | |
f5ea1100 DC |
762 | if (nodehdr1.count > 0 && nodehdr2.count > 0 && |
763 | ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) || | |
764 | (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) < | |
765 | be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) { | |
1da177e4 LT |
766 | tmpnode = node1; |
767 | node1 = node2; | |
768 | node2 = tmpnode; | |
01ba43b8 DC |
769 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
770 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); | |
4bceb18f DC |
771 | btree1 = dp->d_ops->node_tree_p(node1); |
772 | btree2 = dp->d_ops->node_tree_p(node2); | |
f5ea1100 | 773 | swap = 1; |
1da177e4 | 774 | } |
f5ea1100 DC |
775 | |
776 | count = (nodehdr1.count - nodehdr2.count) / 2; | |
1da177e4 LT |
777 | if (count == 0) |
778 | return; | |
779 | tp = state->args->trans; | |
780 | /* | |
781 | * Two cases: high-to-low and low-to-high. | |
782 | */ | |
783 | if (count > 0) { | |
784 | /* | |
785 | * Move elements in node2 up to make a hole. | |
786 | */ | |
f5ea1100 DC |
787 | tmp = nodehdr2.count; |
788 | if (tmp > 0) { | |
1da177e4 | 789 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
f5ea1100 DC |
790 | btree_s = &btree2[0]; |
791 | btree_d = &btree2[count]; | |
1da177e4 LT |
792 | memmove(btree_d, btree_s, tmp); |
793 | } | |
794 | ||
795 | /* | |
796 | * Move the req'd B-tree elements from high in node1 to | |
797 | * low in node2. | |
798 | */ | |
f5ea1100 | 799 | nodehdr2.count += count; |
1da177e4 | 800 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
f5ea1100 DC |
801 | btree_s = &btree1[nodehdr1.count - count]; |
802 | btree_d = &btree2[0]; | |
1da177e4 | 803 | memcpy(btree_d, btree_s, tmp); |
f5ea1100 | 804 | nodehdr1.count -= count; |
1da177e4 LT |
805 | } else { |
806 | /* | |
807 | * Move the req'd B-tree elements from low in node2 to | |
808 | * high in node1. | |
809 | */ | |
810 | count = -count; | |
811 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); | |
f5ea1100 DC |
812 | btree_s = &btree2[0]; |
813 | btree_d = &btree1[nodehdr1.count]; | |
1da177e4 | 814 | memcpy(btree_d, btree_s, tmp); |
f5ea1100 DC |
815 | nodehdr1.count += count; |
816 | ||
1d9025e5 | 817 | xfs_trans_log_buf(tp, blk1->bp, |
1da177e4 LT |
818 | XFS_DA_LOGRANGE(node1, btree_d, tmp)); |
819 | ||
820 | /* | |
821 | * Move elements in node2 down to fill the hole. | |
822 | */ | |
f5ea1100 | 823 | tmp = nodehdr2.count - count; |
1da177e4 | 824 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
f5ea1100 DC |
825 | btree_s = &btree2[count]; |
826 | btree_d = &btree2[0]; | |
1da177e4 | 827 | memmove(btree_d, btree_s, tmp); |
f5ea1100 | 828 | nodehdr2.count -= count; |
1da177e4 LT |
829 | } |
830 | ||
831 | /* | |
832 | * Log header of node 1 and all current bits of node 2. | |
833 | */ | |
01ba43b8 | 834 | dp->d_ops->node_hdr_to_disk(node1, &nodehdr1); |
1d9025e5 | 835 | xfs_trans_log_buf(tp, blk1->bp, |
1c9a5b2e | 836 | XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size)); |
f5ea1100 | 837 | |
01ba43b8 | 838 | dp->d_ops->node_hdr_to_disk(node2, &nodehdr2); |
1d9025e5 | 839 | xfs_trans_log_buf(tp, blk2->bp, |
1da177e4 | 840 | XFS_DA_LOGRANGE(node2, &node2->hdr, |
1c9a5b2e | 841 | dp->d_ops->node_hdr_size + |
f5ea1100 | 842 | (sizeof(btree2[0]) * nodehdr2.count))); |
1da177e4 LT |
843 | |
844 | /* | |
845 | * Record the last hashval from each block for upward propagation. | |
846 | * (note: don't use the swapped node pointers) | |
847 | */ | |
f5ea1100 DC |
848 | if (swap) { |
849 | node1 = blk1->bp->b_addr; | |
850 | node2 = blk2->bp->b_addr; | |
01ba43b8 DC |
851 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
852 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); | |
4bceb18f DC |
853 | btree1 = dp->d_ops->node_tree_p(node1); |
854 | btree2 = dp->d_ops->node_tree_p(node2); | |
f5ea1100 DC |
855 | } |
856 | blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval); | |
857 | blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval); | |
1da177e4 LT |
858 | |
859 | /* | |
860 | * Adjust the expected index for insertion. | |
861 | */ | |
f5ea1100 DC |
862 | if (blk1->index >= nodehdr1.count) { |
863 | blk2->index = blk1->index - nodehdr1.count; | |
864 | blk1->index = nodehdr1.count + 1; /* make it invalid */ | |
1da177e4 LT |
865 | } |
866 | } | |
867 | ||
868 | /* | |
869 | * Add a new entry to an intermediate node. | |
870 | */ | |
871 | STATIC void | |
f5ea1100 DC |
872 | xfs_da3_node_add( |
873 | struct xfs_da_state *state, | |
874 | struct xfs_da_state_blk *oldblk, | |
875 | struct xfs_da_state_blk *newblk) | |
1da177e4 | 876 | { |
f5ea1100 DC |
877 | struct xfs_da_intnode *node; |
878 | struct xfs_da3_icnode_hdr nodehdr; | |
879 | struct xfs_da_node_entry *btree; | |
880 | int tmp; | |
4bceb18f | 881 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 882 | |
5a5881cd DC |
883 | trace_xfs_da_node_add(state->args); |
884 | ||
1d9025e5 | 885 | node = oldblk->bp->b_addr; |
01ba43b8 | 886 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
4bceb18f | 887 | btree = dp->d_ops->node_tree_p(node); |
f5ea1100 DC |
888 | |
889 | ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count); | |
1da177e4 | 890 | ASSERT(newblk->blkno != 0); |
f6c2d1fa | 891 | if (state->args->whichfork == XFS_DATA_FORK) |
7dda6e86 DC |
892 | ASSERT(newblk->blkno >= state->args->geo->leafblk && |
893 | newblk->blkno < state->args->geo->freeblk); | |
1da177e4 LT |
894 | |
895 | /* | |
896 | * We may need to make some room before we insert the new node. | |
897 | */ | |
898 | tmp = 0; | |
f5ea1100 DC |
899 | if (oldblk->index < nodehdr.count) { |
900 | tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree); | |
901 | memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp); | |
1da177e4 | 902 | } |
f5ea1100 DC |
903 | btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval); |
904 | btree[oldblk->index].before = cpu_to_be32(newblk->blkno); | |
1d9025e5 | 905 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
f5ea1100 DC |
906 | XFS_DA_LOGRANGE(node, &btree[oldblk->index], |
907 | tmp + sizeof(*btree))); | |
908 | ||
909 | nodehdr.count += 1; | |
01ba43b8 | 910 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
1d9025e5 | 911 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
1c9a5b2e | 912 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
1da177e4 LT |
913 | |
914 | /* | |
915 | * Copy the last hash value from the oldblk to propagate upwards. | |
916 | */ | |
f5ea1100 | 917 | oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
1da177e4 LT |
918 | } |
919 | ||
920 | /*======================================================================== | |
921 | * Routines used for shrinking the Btree. | |
922 | *========================================================================*/ | |
923 | ||
924 | /* | |
925 | * Deallocate an empty leaf node, remove it from its parent, | |
926 | * possibly deallocating that block, etc... | |
927 | */ | |
928 | int | |
f5ea1100 DC |
929 | xfs_da3_join( |
930 | struct xfs_da_state *state) | |
1da177e4 | 931 | { |
f5ea1100 DC |
932 | struct xfs_da_state_blk *drop_blk; |
933 | struct xfs_da_state_blk *save_blk; | |
934 | int action = 0; | |
935 | int error; | |
1da177e4 | 936 | |
5a5881cd DC |
937 | trace_xfs_da_join(state->args); |
938 | ||
1da177e4 LT |
939 | drop_blk = &state->path.blk[ state->path.active-1 ]; |
940 | save_blk = &state->altpath.blk[ state->path.active-1 ]; | |
941 | ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC); | |
942 | ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC || | |
f6c2d1fa | 943 | drop_blk->magic == XFS_DIR2_LEAFN_MAGIC); |
1da177e4 LT |
944 | |
945 | /* | |
946 | * Walk back up the tree joining/deallocating as necessary. | |
947 | * When we stop dropping blocks, break out. | |
948 | */ | |
949 | for ( ; state->path.active >= 2; drop_blk--, save_blk--, | |
950 | state->path.active--) { | |
951 | /* | |
952 | * See if we can combine the block with a neighbor. | |
953 | * (action == 0) => no options, just leave | |
954 | * (action == 1) => coalesce, then unlink | |
955 | * (action == 2) => block empty, unlink it | |
956 | */ | |
957 | switch (drop_blk->magic) { | |
958 | case XFS_ATTR_LEAF_MAGIC: | |
517c2220 | 959 | error = xfs_attr3_leaf_toosmall(state, &action); |
1da177e4 | 960 | if (error) |
d99831ff | 961 | return error; |
1da177e4 | 962 | if (action == 0) |
d99831ff | 963 | return 0; |
517c2220 | 964 | xfs_attr3_leaf_unbalance(state, drop_blk, save_blk); |
1da177e4 | 965 | break; |
1da177e4 | 966 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
967 | error = xfs_dir2_leafn_toosmall(state, &action); |
968 | if (error) | |
969 | return error; | |
970 | if (action == 0) | |
971 | return 0; | |
972 | xfs_dir2_leafn_unbalance(state, drop_blk, save_blk); | |
973 | break; | |
974 | case XFS_DA_NODE_MAGIC: | |
975 | /* | |
976 | * Remove the offending node, fixup hashvals, | |
977 | * check for a toosmall neighbor. | |
978 | */ | |
f5ea1100 DC |
979 | xfs_da3_node_remove(state, drop_blk); |
980 | xfs_da3_fixhashpath(state, &state->path); | |
981 | error = xfs_da3_node_toosmall(state, &action); | |
1da177e4 | 982 | if (error) |
d99831ff | 983 | return error; |
1da177e4 LT |
984 | if (action == 0) |
985 | return 0; | |
f5ea1100 | 986 | xfs_da3_node_unbalance(state, drop_blk, save_blk); |
1da177e4 LT |
987 | break; |
988 | } | |
f5ea1100 DC |
989 | xfs_da3_fixhashpath(state, &state->altpath); |
990 | error = xfs_da3_blk_unlink(state, drop_blk, save_blk); | |
1da177e4 LT |
991 | xfs_da_state_kill_altpath(state); |
992 | if (error) | |
d99831ff | 993 | return error; |
1da177e4 LT |
994 | error = xfs_da_shrink_inode(state->args, drop_blk->blkno, |
995 | drop_blk->bp); | |
996 | drop_blk->bp = NULL; | |
997 | if (error) | |
d99831ff | 998 | return error; |
1da177e4 LT |
999 | } |
1000 | /* | |
1001 | * We joined all the way to the top. If it turns out that | |
1002 | * we only have one entry in the root, make the child block | |
1003 | * the new root. | |
1004 | */ | |
f5ea1100 DC |
1005 | xfs_da3_node_remove(state, drop_blk); |
1006 | xfs_da3_fixhashpath(state, &state->path); | |
1007 | error = xfs_da3_root_join(state, &state->path.blk[0]); | |
d99831ff | 1008 | return error; |
1da177e4 LT |
1009 | } |
1010 | ||
1c4f3329 AE |
1011 | #ifdef DEBUG |
1012 | static void | |
1013 | xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level) | |
1014 | { | |
1015 | __be16 magic = blkinfo->magic; | |
1016 | ||
1017 | if (level == 1) { | |
1018 | ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
24df33b4 | 1019 | magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
517c2220 DC |
1020 | magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || |
1021 | magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); | |
f5ea1100 DC |
1022 | } else { |
1023 | ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
1024 | magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)); | |
1025 | } | |
1c4f3329 AE |
1026 | ASSERT(!blkinfo->forw); |
1027 | ASSERT(!blkinfo->back); | |
1028 | } | |
1029 | #else /* !DEBUG */ | |
1030 | #define xfs_da_blkinfo_onlychild_validate(blkinfo, level) | |
1031 | #endif /* !DEBUG */ | |
1032 | ||
1da177e4 LT |
1033 | /* |
1034 | * We have only one entry in the root. Copy the only remaining child of | |
1035 | * the old root to block 0 as the new root node. | |
1036 | */ | |
1037 | STATIC int | |
f5ea1100 DC |
1038 | xfs_da3_root_join( |
1039 | struct xfs_da_state *state, | |
1040 | struct xfs_da_state_blk *root_blk) | |
1da177e4 | 1041 | { |
f5ea1100 DC |
1042 | struct xfs_da_intnode *oldroot; |
1043 | struct xfs_da_args *args; | |
1044 | xfs_dablk_t child; | |
1045 | struct xfs_buf *bp; | |
1046 | struct xfs_da3_icnode_hdr oldroothdr; | |
1047 | struct xfs_da_node_entry *btree; | |
1048 | int error; | |
01ba43b8 | 1049 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 1050 | |
5a5881cd DC |
1051 | trace_xfs_da_root_join(state->args); |
1052 | ||
1da177e4 | 1053 | ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC); |
f5ea1100 DC |
1054 | |
1055 | args = state->args; | |
1d9025e5 | 1056 | oldroot = root_blk->bp->b_addr; |
01ba43b8 | 1057 | dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot); |
f5ea1100 DC |
1058 | ASSERT(oldroothdr.forw == 0); |
1059 | ASSERT(oldroothdr.back == 0); | |
1da177e4 LT |
1060 | |
1061 | /* | |
1062 | * If the root has more than one child, then don't do anything. | |
1063 | */ | |
f5ea1100 DC |
1064 | if (oldroothdr.count > 1) |
1065 | return 0; | |
1da177e4 LT |
1066 | |
1067 | /* | |
1068 | * Read in the (only) child block, then copy those bytes into | |
1069 | * the root block's buffer and free the original child block. | |
1070 | */ | |
01ba43b8 | 1071 | btree = dp->d_ops->node_tree_p(oldroot); |
f5ea1100 | 1072 | child = be32_to_cpu(btree[0].before); |
1da177e4 | 1073 | ASSERT(child != 0); |
01ba43b8 | 1074 | error = xfs_da3_node_read(args->trans, dp, child, -1, &bp, |
d9392a4b | 1075 | args->whichfork); |
1da177e4 | 1076 | if (error) |
f5ea1100 DC |
1077 | return error; |
1078 | xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level); | |
1c4f3329 | 1079 | |
612cfbfe DC |
1080 | /* |
1081 | * This could be copying a leaf back into the root block in the case of | |
1082 | * there only being a single leaf block left in the tree. Hence we have | |
1813dd64 | 1083 | * to update the b_ops pointer as well to match the buffer type change |
f5ea1100 DC |
1084 | * that could occur. For dir3 blocks we also need to update the block |
1085 | * number in the buffer header. | |
612cfbfe | 1086 | */ |
b2a21e7a | 1087 | memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize); |
1813dd64 | 1088 | root_blk->bp->b_ops = bp->b_ops; |
d75afeb3 | 1089 | xfs_trans_buf_copy_type(root_blk->bp, bp); |
f5ea1100 DC |
1090 | if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) { |
1091 | struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr; | |
1092 | da3->blkno = cpu_to_be64(root_blk->bp->b_bn); | |
1093 | } | |
b2a21e7a DC |
1094 | xfs_trans_log_buf(args->trans, root_blk->bp, 0, |
1095 | args->geo->blksize - 1); | |
1da177e4 | 1096 | error = xfs_da_shrink_inode(args, child, bp); |
d99831ff | 1097 | return error; |
1da177e4 LT |
1098 | } |
1099 | ||
1100 | /* | |
1101 | * Check a node block and its neighbors to see if the block should be | |
1102 | * collapsed into one or the other neighbor. Always keep the block | |
1103 | * with the smaller block number. | |
1104 | * If the current block is over 50% full, don't try to join it, return 0. | |
1105 | * If the block is empty, fill in the state structure and return 2. | |
1106 | * If it can be collapsed, fill in the state structure and return 1. | |
1107 | * If nothing can be done, return 0. | |
1108 | */ | |
1109 | STATIC int | |
f5ea1100 DC |
1110 | xfs_da3_node_toosmall( |
1111 | struct xfs_da_state *state, | |
1112 | int *action) | |
1da177e4 | 1113 | { |
f5ea1100 DC |
1114 | struct xfs_da_intnode *node; |
1115 | struct xfs_da_state_blk *blk; | |
1116 | struct xfs_da_blkinfo *info; | |
1117 | xfs_dablk_t blkno; | |
1118 | struct xfs_buf *bp; | |
1119 | struct xfs_da3_icnode_hdr nodehdr; | |
1120 | int count; | |
1121 | int forward; | |
1122 | int error; | |
1123 | int retval; | |
1124 | int i; | |
01ba43b8 | 1125 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 1126 | |
ee73259b DC |
1127 | trace_xfs_da_node_toosmall(state->args); |
1128 | ||
1da177e4 LT |
1129 | /* |
1130 | * Check for the degenerate case of the block being over 50% full. | |
1131 | * If so, it's not worth even looking to see if we might be able | |
1132 | * to coalesce with a sibling. | |
1133 | */ | |
1134 | blk = &state->path.blk[ state->path.active-1 ]; | |
1d9025e5 | 1135 | info = blk->bp->b_addr; |
1da177e4 | 1136 | node = (xfs_da_intnode_t *)info; |
01ba43b8 | 1137 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
b2a21e7a | 1138 | if (nodehdr.count > (state->args->geo->node_ents >> 1)) { |
1da177e4 | 1139 | *action = 0; /* blk over 50%, don't try to join */ |
d99831ff | 1140 | return 0; /* blk over 50%, don't try to join */ |
1da177e4 LT |
1141 | } |
1142 | ||
1143 | /* | |
1144 | * Check for the degenerate case of the block being empty. | |
1145 | * If the block is empty, we'll simply delete it, no need to | |
c41564b5 | 1146 | * coalesce it with a sibling block. We choose (arbitrarily) |
1da177e4 LT |
1147 | * to merge with the forward block unless it is NULL. |
1148 | */ | |
f5ea1100 | 1149 | if (nodehdr.count == 0) { |
1da177e4 LT |
1150 | /* |
1151 | * Make altpath point to the block we want to keep and | |
1152 | * path point to the block we want to drop (this one). | |
1153 | */ | |
89da0544 | 1154 | forward = (info->forw != 0); |
1da177e4 | 1155 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
f5ea1100 | 1156 | error = xfs_da3_path_shift(state, &state->altpath, forward, |
1da177e4 LT |
1157 | 0, &retval); |
1158 | if (error) | |
d99831ff | 1159 | return error; |
1da177e4 LT |
1160 | if (retval) { |
1161 | *action = 0; | |
1162 | } else { | |
1163 | *action = 2; | |
1164 | } | |
d99831ff | 1165 | return 0; |
1da177e4 LT |
1166 | } |
1167 | ||
1168 | /* | |
1169 | * Examine each sibling block to see if we can coalesce with | |
1170 | * at least 25% free space to spare. We need to figure out | |
1171 | * whether to merge with the forward or the backward block. | |
1172 | * We prefer coalescing with the lower numbered sibling so as | |
1173 | * to shrink a directory over time. | |
1174 | */ | |
b2a21e7a DC |
1175 | count = state->args->geo->node_ents; |
1176 | count -= state->args->geo->node_ents >> 2; | |
f5ea1100 DC |
1177 | count -= nodehdr.count; |
1178 | ||
1da177e4 | 1179 | /* start with smaller blk num */ |
f5ea1100 | 1180 | forward = nodehdr.forw < nodehdr.back; |
1da177e4 | 1181 | for (i = 0; i < 2; forward = !forward, i++) { |
997def25 | 1182 | struct xfs_da3_icnode_hdr thdr; |
1da177e4 | 1183 | if (forward) |
f5ea1100 | 1184 | blkno = nodehdr.forw; |
1da177e4 | 1185 | else |
f5ea1100 | 1186 | blkno = nodehdr.back; |
1da177e4 LT |
1187 | if (blkno == 0) |
1188 | continue; | |
01ba43b8 | 1189 | error = xfs_da3_node_read(state->args->trans, dp, |
d9392a4b | 1190 | blkno, -1, &bp, state->args->whichfork); |
1da177e4 | 1191 | if (error) |
d99831ff | 1192 | return error; |
1da177e4 | 1193 | |
1d9025e5 | 1194 | node = bp->b_addr; |
01ba43b8 | 1195 | dp->d_ops->node_hdr_from_disk(&thdr, node); |
1d9025e5 | 1196 | xfs_trans_brelse(state->args->trans, bp); |
f5ea1100 | 1197 | |
997def25 | 1198 | if (count - thdr.count >= 0) |
1da177e4 LT |
1199 | break; /* fits with at least 25% to spare */ |
1200 | } | |
1201 | if (i >= 2) { | |
1202 | *action = 0; | |
f5ea1100 | 1203 | return 0; |
1da177e4 LT |
1204 | } |
1205 | ||
1206 | /* | |
1207 | * Make altpath point to the block we want to keep (the lower | |
1208 | * numbered block) and path point to the block we want to drop. | |
1209 | */ | |
1210 | memcpy(&state->altpath, &state->path, sizeof(state->path)); | |
1211 | if (blkno < blk->blkno) { | |
f5ea1100 | 1212 | error = xfs_da3_path_shift(state, &state->altpath, forward, |
1da177e4 | 1213 | 0, &retval); |
1da177e4 | 1214 | } else { |
f5ea1100 | 1215 | error = xfs_da3_path_shift(state, &state->path, forward, |
1da177e4 | 1216 | 0, &retval); |
f5ea1100 DC |
1217 | } |
1218 | if (error) | |
1219 | return error; | |
1220 | if (retval) { | |
1221 | *action = 0; | |
1222 | return 0; | |
1da177e4 LT |
1223 | } |
1224 | *action = 1; | |
f5ea1100 DC |
1225 | return 0; |
1226 | } | |
1227 | ||
1228 | /* | |
1229 | * Pick up the last hashvalue from an intermediate node. | |
1230 | */ | |
1231 | STATIC uint | |
1232 | xfs_da3_node_lasthash( | |
4bceb18f | 1233 | struct xfs_inode *dp, |
f5ea1100 DC |
1234 | struct xfs_buf *bp, |
1235 | int *count) | |
1236 | { | |
1237 | struct xfs_da_intnode *node; | |
1238 | struct xfs_da_node_entry *btree; | |
1239 | struct xfs_da3_icnode_hdr nodehdr; | |
1240 | ||
1241 | node = bp->b_addr; | |
01ba43b8 | 1242 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
f5ea1100 DC |
1243 | if (count) |
1244 | *count = nodehdr.count; | |
1245 | if (!nodehdr.count) | |
1246 | return 0; | |
4bceb18f | 1247 | btree = dp->d_ops->node_tree_p(node); |
f5ea1100 | 1248 | return be32_to_cpu(btree[nodehdr.count - 1].hashval); |
1da177e4 LT |
1249 | } |
1250 | ||
1251 | /* | |
1252 | * Walk back up the tree adjusting hash values as necessary, | |
1253 | * when we stop making changes, return. | |
1254 | */ | |
1255 | void | |
f5ea1100 DC |
1256 | xfs_da3_fixhashpath( |
1257 | struct xfs_da_state *state, | |
1258 | struct xfs_da_state_path *path) | |
1da177e4 | 1259 | { |
f5ea1100 DC |
1260 | struct xfs_da_state_blk *blk; |
1261 | struct xfs_da_intnode *node; | |
1262 | struct xfs_da_node_entry *btree; | |
1263 | xfs_dahash_t lasthash=0; | |
1264 | int level; | |
1265 | int count; | |
4bceb18f | 1266 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 1267 | |
ee73259b DC |
1268 | trace_xfs_da_fixhashpath(state->args); |
1269 | ||
1da177e4 LT |
1270 | level = path->active-1; |
1271 | blk = &path->blk[ level ]; | |
1272 | switch (blk->magic) { | |
1da177e4 LT |
1273 | case XFS_ATTR_LEAF_MAGIC: |
1274 | lasthash = xfs_attr_leaf_lasthash(blk->bp, &count); | |
1275 | if (count == 0) | |
1276 | return; | |
1277 | break; | |
1da177e4 | 1278 | case XFS_DIR2_LEAFN_MAGIC: |
8e8877e6 | 1279 | lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count); |
1da177e4 LT |
1280 | if (count == 0) |
1281 | return; | |
1282 | break; | |
1283 | case XFS_DA_NODE_MAGIC: | |
4bceb18f | 1284 | lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count); |
1da177e4 LT |
1285 | if (count == 0) |
1286 | return; | |
1287 | break; | |
1288 | } | |
1289 | for (blk--, level--; level >= 0; blk--, level--) { | |
f5ea1100 DC |
1290 | struct xfs_da3_icnode_hdr nodehdr; |
1291 | ||
1d9025e5 | 1292 | node = blk->bp->b_addr; |
01ba43b8 | 1293 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
4bceb18f | 1294 | btree = dp->d_ops->node_tree_p(node); |
c88547a8 | 1295 | if (be32_to_cpu(btree[blk->index].hashval) == lasthash) |
1da177e4 LT |
1296 | break; |
1297 | blk->hashval = lasthash; | |
f5ea1100 | 1298 | btree[blk->index].hashval = cpu_to_be32(lasthash); |
1d9025e5 | 1299 | xfs_trans_log_buf(state->args->trans, blk->bp, |
f5ea1100 DC |
1300 | XFS_DA_LOGRANGE(node, &btree[blk->index], |
1301 | sizeof(*btree))); | |
1da177e4 | 1302 | |
f5ea1100 | 1303 | lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
1da177e4 LT |
1304 | } |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * Remove an entry from an intermediate node. | |
1309 | */ | |
1310 | STATIC void | |
f5ea1100 DC |
1311 | xfs_da3_node_remove( |
1312 | struct xfs_da_state *state, | |
1313 | struct xfs_da_state_blk *drop_blk) | |
1da177e4 | 1314 | { |
f5ea1100 DC |
1315 | struct xfs_da_intnode *node; |
1316 | struct xfs_da3_icnode_hdr nodehdr; | |
1317 | struct xfs_da_node_entry *btree; | |
1318 | int index; | |
1319 | int tmp; | |
4bceb18f | 1320 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 1321 | |
5a5881cd DC |
1322 | trace_xfs_da_node_remove(state->args); |
1323 | ||
1d9025e5 | 1324 | node = drop_blk->bp->b_addr; |
01ba43b8 | 1325 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
f5ea1100 | 1326 | ASSERT(drop_blk->index < nodehdr.count); |
1da177e4 LT |
1327 | ASSERT(drop_blk->index >= 0); |
1328 | ||
1329 | /* | |
1330 | * Copy over the offending entry, or just zero it out. | |
1331 | */ | |
f5ea1100 | 1332 | index = drop_blk->index; |
4bceb18f | 1333 | btree = dp->d_ops->node_tree_p(node); |
f5ea1100 DC |
1334 | if (index < nodehdr.count - 1) { |
1335 | tmp = nodehdr.count - index - 1; | |
1da177e4 | 1336 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
f5ea1100 | 1337 | memmove(&btree[index], &btree[index + 1], tmp); |
1d9025e5 | 1338 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
f5ea1100 DC |
1339 | XFS_DA_LOGRANGE(node, &btree[index], tmp)); |
1340 | index = nodehdr.count - 1; | |
1da177e4 | 1341 | } |
f5ea1100 | 1342 | memset(&btree[index], 0, sizeof(xfs_da_node_entry_t)); |
1d9025e5 | 1343 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
f5ea1100 DC |
1344 | XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index]))); |
1345 | nodehdr.count -= 1; | |
01ba43b8 | 1346 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
1d9025e5 | 1347 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
1c9a5b2e | 1348 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
1da177e4 LT |
1349 | |
1350 | /* | |
1351 | * Copy the last hash value from the block to propagate upwards. | |
1352 | */ | |
f5ea1100 | 1353 | drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval); |
1da177e4 LT |
1354 | } |
1355 | ||
1356 | /* | |
f5ea1100 | 1357 | * Unbalance the elements between two intermediate nodes, |
1da177e4 LT |
1358 | * move all Btree elements from one node into another. |
1359 | */ | |
1360 | STATIC void | |
f5ea1100 DC |
1361 | xfs_da3_node_unbalance( |
1362 | struct xfs_da_state *state, | |
1363 | struct xfs_da_state_blk *drop_blk, | |
1364 | struct xfs_da_state_blk *save_blk) | |
1da177e4 | 1365 | { |
f5ea1100 DC |
1366 | struct xfs_da_intnode *drop_node; |
1367 | struct xfs_da_intnode *save_node; | |
1368 | struct xfs_da_node_entry *drop_btree; | |
1369 | struct xfs_da_node_entry *save_btree; | |
1370 | struct xfs_da3_icnode_hdr drop_hdr; | |
1371 | struct xfs_da3_icnode_hdr save_hdr; | |
1372 | struct xfs_trans *tp; | |
1373 | int sindex; | |
1374 | int tmp; | |
4bceb18f | 1375 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 1376 | |
5a5881cd DC |
1377 | trace_xfs_da_node_unbalance(state->args); |
1378 | ||
1d9025e5 DC |
1379 | drop_node = drop_blk->bp->b_addr; |
1380 | save_node = save_blk->bp->b_addr; | |
01ba43b8 DC |
1381 | dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node); |
1382 | dp->d_ops->node_hdr_from_disk(&save_hdr, save_node); | |
4bceb18f DC |
1383 | drop_btree = dp->d_ops->node_tree_p(drop_node); |
1384 | save_btree = dp->d_ops->node_tree_p(save_node); | |
1da177e4 LT |
1385 | tp = state->args->trans; |
1386 | ||
1387 | /* | |
1388 | * If the dying block has lower hashvals, then move all the | |
1389 | * elements in the remaining block up to make a hole. | |
1390 | */ | |
f5ea1100 DC |
1391 | if ((be32_to_cpu(drop_btree[0].hashval) < |
1392 | be32_to_cpu(save_btree[0].hashval)) || | |
1393 | (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) < | |
1394 | be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) { | |
1395 | /* XXX: check this - is memmove dst correct? */ | |
1396 | tmp = save_hdr.count * sizeof(xfs_da_node_entry_t); | |
1397 | memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp); | |
1398 | ||
1399 | sindex = 0; | |
1d9025e5 | 1400 | xfs_trans_log_buf(tp, save_blk->bp, |
f5ea1100 DC |
1401 | XFS_DA_LOGRANGE(save_node, &save_btree[0], |
1402 | (save_hdr.count + drop_hdr.count) * | |
1403 | sizeof(xfs_da_node_entry_t))); | |
1da177e4 | 1404 | } else { |
f5ea1100 | 1405 | sindex = save_hdr.count; |
1d9025e5 | 1406 | xfs_trans_log_buf(tp, save_blk->bp, |
f5ea1100 DC |
1407 | XFS_DA_LOGRANGE(save_node, &save_btree[sindex], |
1408 | drop_hdr.count * sizeof(xfs_da_node_entry_t))); | |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | /* | |
1412 | * Move all the B-tree elements from drop_blk to save_blk. | |
1413 | */ | |
f5ea1100 DC |
1414 | tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t); |
1415 | memcpy(&save_btree[sindex], &drop_btree[0], tmp); | |
1416 | save_hdr.count += drop_hdr.count; | |
1da177e4 | 1417 | |
01ba43b8 | 1418 | dp->d_ops->node_hdr_to_disk(save_node, &save_hdr); |
1d9025e5 | 1419 | xfs_trans_log_buf(tp, save_blk->bp, |
1da177e4 | 1420 | XFS_DA_LOGRANGE(save_node, &save_node->hdr, |
1c9a5b2e | 1421 | dp->d_ops->node_hdr_size)); |
1da177e4 LT |
1422 | |
1423 | /* | |
1424 | * Save the last hashval in the remaining block for upward propagation. | |
1425 | */ | |
f5ea1100 | 1426 | save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval); |
1da177e4 LT |
1427 | } |
1428 | ||
1429 | /*======================================================================== | |
1430 | * Routines used for finding things in the Btree. | |
1431 | *========================================================================*/ | |
1432 | ||
1433 | /* | |
1434 | * Walk down the Btree looking for a particular filename, filling | |
1435 | * in the state structure as we go. | |
1436 | * | |
1437 | * We will set the state structure to point to each of the elements | |
1438 | * in each of the nodes where either the hashval is or should be. | |
1439 | * | |
1440 | * We support duplicate hashval's so for each entry in the current | |
1441 | * node that could contain the desired hashval, descend. This is a | |
1442 | * pruned depth-first tree search. | |
1443 | */ | |
1444 | int /* error */ | |
f5ea1100 DC |
1445 | xfs_da3_node_lookup_int( |
1446 | struct xfs_da_state *state, | |
1447 | int *result) | |
1da177e4 | 1448 | { |
f5ea1100 DC |
1449 | struct xfs_da_state_blk *blk; |
1450 | struct xfs_da_blkinfo *curr; | |
1451 | struct xfs_da_intnode *node; | |
1452 | struct xfs_da_node_entry *btree; | |
1453 | struct xfs_da3_icnode_hdr nodehdr; | |
1454 | struct xfs_da_args *args; | |
1455 | xfs_dablk_t blkno; | |
1456 | xfs_dahash_t hashval; | |
1457 | xfs_dahash_t btreehashval; | |
1458 | int probe; | |
1459 | int span; | |
1460 | int max; | |
1461 | int error; | |
1462 | int retval; | |
8210f4dd | 1463 | unsigned int expected_level = 0; |
4bceb18f | 1464 | struct xfs_inode *dp = state->args->dp; |
1da177e4 LT |
1465 | |
1466 | args = state->args; | |
1467 | ||
1468 | /* | |
1469 | * Descend thru the B-tree searching each level for the right | |
1470 | * node to use, until the right hashval is found. | |
1471 | */ | |
8210f4dd | 1472 | blkno = args->geo->leafblk; |
1da177e4 LT |
1473 | for (blk = &state->path.blk[0], state->path.active = 1; |
1474 | state->path.active <= XFS_DA_NODE_MAXDEPTH; | |
1475 | blk++, state->path.active++) { | |
1476 | /* | |
1477 | * Read the next node down in the tree. | |
1478 | */ | |
1479 | blk->blkno = blkno; | |
f5ea1100 | 1480 | error = xfs_da3_node_read(args->trans, args->dp, blkno, |
d9392a4b | 1481 | -1, &blk->bp, args->whichfork); |
1da177e4 LT |
1482 | if (error) { |
1483 | blk->blkno = 0; | |
1484 | state->path.active--; | |
d99831ff | 1485 | return error; |
1da177e4 | 1486 | } |
1d9025e5 | 1487 | curr = blk->bp->b_addr; |
d432c80e | 1488 | blk->magic = be16_to_cpu(curr->magic); |
f5ea1100 | 1489 | |
517c2220 DC |
1490 | if (blk->magic == XFS_ATTR_LEAF_MAGIC || |
1491 | blk->magic == XFS_ATTR3_LEAF_MAGIC) { | |
1492 | blk->magic = XFS_ATTR_LEAF_MAGIC; | |
f5ea1100 DC |
1493 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
1494 | break; | |
1495 | } | |
1496 | ||
1497 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC || | |
1498 | blk->magic == XFS_DIR3_LEAFN_MAGIC) { | |
1499 | blk->magic = XFS_DIR2_LEAFN_MAGIC; | |
8e8877e6 DW |
1500 | blk->hashval = xfs_dir2_leaf_lasthash(args->dp, |
1501 | blk->bp, NULL); | |
f5ea1100 DC |
1502 | break; |
1503 | } | |
1504 | ||
1505 | blk->magic = XFS_DA_NODE_MAGIC; | |
1506 | ||
1da177e4 LT |
1507 | |
1508 | /* | |
1509 | * Search an intermediate node for a match. | |
1510 | */ | |
f5ea1100 | 1511 | node = blk->bp->b_addr; |
01ba43b8 | 1512 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
4bceb18f | 1513 | btree = dp->d_ops->node_tree_p(node); |
1da177e4 | 1514 | |
8210f4dd DW |
1515 | /* Tree taller than we can handle; bail out! */ |
1516 | if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) | |
1517 | return -EFSCORRUPTED; | |
1518 | ||
1519 | /* Check the level from the root. */ | |
1520 | if (blkno == args->geo->leafblk) | |
1521 | expected_level = nodehdr.level - 1; | |
1522 | else if (expected_level != nodehdr.level) | |
1523 | return -EFSCORRUPTED; | |
1524 | else | |
1525 | expected_level--; | |
1526 | ||
f5ea1100 DC |
1527 | max = nodehdr.count; |
1528 | blk->hashval = be32_to_cpu(btree[max - 1].hashval); | |
1da177e4 | 1529 | |
f5ea1100 DC |
1530 | /* |
1531 | * Binary search. (note: small blocks will skip loop) | |
1532 | */ | |
1533 | probe = span = max / 2; | |
1534 | hashval = args->hashval; | |
1535 | while (span > 4) { | |
1536 | span /= 2; | |
1537 | btreehashval = be32_to_cpu(btree[probe].hashval); | |
1538 | if (btreehashval < hashval) | |
1539 | probe += span; | |
1540 | else if (btreehashval > hashval) | |
1541 | probe -= span; | |
1542 | else | |
1543 | break; | |
1544 | } | |
1545 | ASSERT((probe >= 0) && (probe < max)); | |
1546 | ASSERT((span <= 4) || | |
1547 | (be32_to_cpu(btree[probe].hashval) == hashval)); | |
1da177e4 | 1548 | |
f5ea1100 DC |
1549 | /* |
1550 | * Since we may have duplicate hashval's, find the first | |
1551 | * matching hashval in the node. | |
1552 | */ | |
1553 | while (probe > 0 && | |
1554 | be32_to_cpu(btree[probe].hashval) >= hashval) { | |
1555 | probe--; | |
1556 | } | |
1557 | while (probe < max && | |
1558 | be32_to_cpu(btree[probe].hashval) < hashval) { | |
1559 | probe++; | |
1560 | } | |
1561 | ||
1562 | /* | |
1563 | * Pick the right block to descend on. | |
1564 | */ | |
1565 | if (probe == max) { | |
1566 | blk->index = max - 1; | |
1567 | blkno = be32_to_cpu(btree[max - 1].before); | |
1568 | } else { | |
1569 | blk->index = probe; | |
1570 | blkno = be32_to_cpu(btree[probe].before); | |
1da177e4 | 1571 | } |
8210f4dd DW |
1572 | |
1573 | /* We can't point back to the root. */ | |
1574 | if (blkno == args->geo->leafblk) | |
1575 | return -EFSCORRUPTED; | |
1da177e4 LT |
1576 | } |
1577 | ||
8210f4dd DW |
1578 | if (expected_level != 0) |
1579 | return -EFSCORRUPTED; | |
1580 | ||
1da177e4 LT |
1581 | /* |
1582 | * A leaf block that ends in the hashval that we are interested in | |
1583 | * (final hashval == search hashval) means that the next block may | |
1584 | * contain more entries with the same hashval, shift upward to the | |
1585 | * next leaf and keep searching. | |
1586 | */ | |
1587 | for (;;) { | |
f6c2d1fa | 1588 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
1da177e4 LT |
1589 | retval = xfs_dir2_leafn_lookup_int(blk->bp, args, |
1590 | &blk->index, state); | |
d432c80e | 1591 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
517c2220 | 1592 | retval = xfs_attr3_leaf_lookup_int(blk->bp, args); |
1da177e4 LT |
1593 | blk->index = args->index; |
1594 | args->blkno = blk->blkno; | |
d432c80e NS |
1595 | } else { |
1596 | ASSERT(0); | |
2451337d | 1597 | return -EFSCORRUPTED; |
1da177e4 | 1598 | } |
2451337d | 1599 | if (((retval == -ENOENT) || (retval == -ENOATTR)) && |
1da177e4 | 1600 | (blk->hashval == args->hashval)) { |
f5ea1100 | 1601 | error = xfs_da3_path_shift(state, &state->path, 1, 1, |
1da177e4 LT |
1602 | &retval); |
1603 | if (error) | |
d99831ff | 1604 | return error; |
1da177e4 LT |
1605 | if (retval == 0) { |
1606 | continue; | |
d432c80e | 1607 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
1da177e4 | 1608 | /* path_shift() gives ENOENT */ |
2451337d | 1609 | retval = -ENOATTR; |
1da177e4 | 1610 | } |
1da177e4 LT |
1611 | } |
1612 | break; | |
1613 | } | |
1614 | *result = retval; | |
d99831ff | 1615 | return 0; |
1da177e4 LT |
1616 | } |
1617 | ||
1618 | /*======================================================================== | |
1619 | * Utility routines. | |
1620 | *========================================================================*/ | |
1621 | ||
f5ea1100 DC |
1622 | /* |
1623 | * Compare two intermediate nodes for "order". | |
1624 | */ | |
1625 | STATIC int | |
1626 | xfs_da3_node_order( | |
4bceb18f | 1627 | struct xfs_inode *dp, |
f5ea1100 DC |
1628 | struct xfs_buf *node1_bp, |
1629 | struct xfs_buf *node2_bp) | |
1630 | { | |
1631 | struct xfs_da_intnode *node1; | |
1632 | struct xfs_da_intnode *node2; | |
1633 | struct xfs_da_node_entry *btree1; | |
1634 | struct xfs_da_node_entry *btree2; | |
1635 | struct xfs_da3_icnode_hdr node1hdr; | |
1636 | struct xfs_da3_icnode_hdr node2hdr; | |
1637 | ||
1638 | node1 = node1_bp->b_addr; | |
1639 | node2 = node2_bp->b_addr; | |
01ba43b8 DC |
1640 | dp->d_ops->node_hdr_from_disk(&node1hdr, node1); |
1641 | dp->d_ops->node_hdr_from_disk(&node2hdr, node2); | |
4bceb18f DC |
1642 | btree1 = dp->d_ops->node_tree_p(node1); |
1643 | btree2 = dp->d_ops->node_tree_p(node2); | |
f5ea1100 DC |
1644 | |
1645 | if (node1hdr.count > 0 && node2hdr.count > 0 && | |
1646 | ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) || | |
1647 | (be32_to_cpu(btree2[node2hdr.count - 1].hashval) < | |
1648 | be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) { | |
1649 | return 1; | |
1650 | } | |
1651 | return 0; | |
1652 | } | |
1653 | ||
1da177e4 LT |
1654 | /* |
1655 | * Link a new block into a doubly linked list of blocks (of whatever type). | |
1656 | */ | |
1657 | int /* error */ | |
f5ea1100 DC |
1658 | xfs_da3_blk_link( |
1659 | struct xfs_da_state *state, | |
1660 | struct xfs_da_state_blk *old_blk, | |
1661 | struct xfs_da_state_blk *new_blk) | |
1da177e4 | 1662 | { |
f5ea1100 DC |
1663 | struct xfs_da_blkinfo *old_info; |
1664 | struct xfs_da_blkinfo *new_info; | |
1665 | struct xfs_da_blkinfo *tmp_info; | |
1666 | struct xfs_da_args *args; | |
1667 | struct xfs_buf *bp; | |
1668 | int before = 0; | |
1669 | int error; | |
4bceb18f | 1670 | struct xfs_inode *dp = state->args->dp; |
1da177e4 LT |
1671 | |
1672 | /* | |
1673 | * Set up environment. | |
1674 | */ | |
1675 | args = state->args; | |
1676 | ASSERT(args != NULL); | |
1d9025e5 DC |
1677 | old_info = old_blk->bp->b_addr; |
1678 | new_info = new_blk->bp->b_addr; | |
1da177e4 | 1679 | ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC || |
f6c2d1fa | 1680 | old_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
1da177e4 | 1681 | old_blk->magic == XFS_ATTR_LEAF_MAGIC); |
1da177e4 LT |
1682 | |
1683 | switch (old_blk->magic) { | |
1da177e4 LT |
1684 | case XFS_ATTR_LEAF_MAGIC: |
1685 | before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp); | |
1686 | break; | |
1da177e4 | 1687 | case XFS_DIR2_LEAFN_MAGIC: |
4bceb18f | 1688 | before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp); |
1da177e4 LT |
1689 | break; |
1690 | case XFS_DA_NODE_MAGIC: | |
4bceb18f | 1691 | before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp); |
1da177e4 LT |
1692 | break; |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * Link blocks in appropriate order. | |
1697 | */ | |
1698 | if (before) { | |
1699 | /* | |
1700 | * Link new block in before existing block. | |
1701 | */ | |
5a5881cd | 1702 | trace_xfs_da_link_before(args); |
89da0544 NS |
1703 | new_info->forw = cpu_to_be32(old_blk->blkno); |
1704 | new_info->back = old_info->back; | |
1705 | if (old_info->back) { | |
4bceb18f | 1706 | error = xfs_da3_node_read(args->trans, dp, |
89da0544 | 1707 | be32_to_cpu(old_info->back), |
d9392a4b | 1708 | -1, &bp, args->whichfork); |
1da177e4 | 1709 | if (error) |
d99831ff | 1710 | return error; |
1da177e4 | 1711 | ASSERT(bp != NULL); |
1d9025e5 | 1712 | tmp_info = bp->b_addr; |
f5ea1100 | 1713 | ASSERT(tmp_info->magic == old_info->magic); |
89da0544 NS |
1714 | ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno); |
1715 | tmp_info->forw = cpu_to_be32(new_blk->blkno); | |
1d9025e5 | 1716 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
1da177e4 | 1717 | } |
89da0544 | 1718 | old_info->back = cpu_to_be32(new_blk->blkno); |
1da177e4 LT |
1719 | } else { |
1720 | /* | |
1721 | * Link new block in after existing block. | |
1722 | */ | |
5a5881cd | 1723 | trace_xfs_da_link_after(args); |
89da0544 NS |
1724 | new_info->forw = old_info->forw; |
1725 | new_info->back = cpu_to_be32(old_blk->blkno); | |
1726 | if (old_info->forw) { | |
4bceb18f | 1727 | error = xfs_da3_node_read(args->trans, dp, |
89da0544 | 1728 | be32_to_cpu(old_info->forw), |
d9392a4b | 1729 | -1, &bp, args->whichfork); |
1da177e4 | 1730 | if (error) |
d99831ff | 1731 | return error; |
1da177e4 | 1732 | ASSERT(bp != NULL); |
1d9025e5 | 1733 | tmp_info = bp->b_addr; |
89da0544 NS |
1734 | ASSERT(tmp_info->magic == old_info->magic); |
1735 | ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno); | |
1736 | tmp_info->back = cpu_to_be32(new_blk->blkno); | |
1d9025e5 | 1737 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
1da177e4 | 1738 | } |
89da0544 | 1739 | old_info->forw = cpu_to_be32(new_blk->blkno); |
1da177e4 LT |
1740 | } |
1741 | ||
1d9025e5 DC |
1742 | xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1); |
1743 | xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1); | |
d99831ff | 1744 | return 0; |
1da177e4 LT |
1745 | } |
1746 | ||
1da177e4 LT |
1747 | /* |
1748 | * Unlink a block from a doubly linked list of blocks. | |
1749 | */ | |
ba0f32d4 | 1750 | STATIC int /* error */ |
f5ea1100 DC |
1751 | xfs_da3_blk_unlink( |
1752 | struct xfs_da_state *state, | |
1753 | struct xfs_da_state_blk *drop_blk, | |
1754 | struct xfs_da_state_blk *save_blk) | |
1da177e4 | 1755 | { |
f5ea1100 DC |
1756 | struct xfs_da_blkinfo *drop_info; |
1757 | struct xfs_da_blkinfo *save_info; | |
1758 | struct xfs_da_blkinfo *tmp_info; | |
1759 | struct xfs_da_args *args; | |
1760 | struct xfs_buf *bp; | |
1761 | int error; | |
1da177e4 LT |
1762 | |
1763 | /* | |
1764 | * Set up environment. | |
1765 | */ | |
1766 | args = state->args; | |
1767 | ASSERT(args != NULL); | |
1d9025e5 DC |
1768 | save_info = save_blk->bp->b_addr; |
1769 | drop_info = drop_blk->bp->b_addr; | |
1da177e4 | 1770 | ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC || |
f6c2d1fa | 1771 | save_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
1da177e4 | 1772 | save_blk->magic == XFS_ATTR_LEAF_MAGIC); |
1da177e4 | 1773 | ASSERT(save_blk->magic == drop_blk->magic); |
89da0544 NS |
1774 | ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) || |
1775 | (be32_to_cpu(save_info->back) == drop_blk->blkno)); | |
1776 | ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) || | |
1777 | (be32_to_cpu(drop_info->back) == save_blk->blkno)); | |
1da177e4 LT |
1778 | |
1779 | /* | |
1780 | * Unlink the leaf block from the doubly linked chain of leaves. | |
1781 | */ | |
89da0544 | 1782 | if (be32_to_cpu(save_info->back) == drop_blk->blkno) { |
5a5881cd | 1783 | trace_xfs_da_unlink_back(args); |
89da0544 NS |
1784 | save_info->back = drop_info->back; |
1785 | if (drop_info->back) { | |
f5ea1100 | 1786 | error = xfs_da3_node_read(args->trans, args->dp, |
89da0544 | 1787 | be32_to_cpu(drop_info->back), |
d9392a4b | 1788 | -1, &bp, args->whichfork); |
1da177e4 | 1789 | if (error) |
d99831ff | 1790 | return error; |
1da177e4 | 1791 | ASSERT(bp != NULL); |
1d9025e5 | 1792 | tmp_info = bp->b_addr; |
89da0544 NS |
1793 | ASSERT(tmp_info->magic == save_info->magic); |
1794 | ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno); | |
1795 | tmp_info->forw = cpu_to_be32(save_blk->blkno); | |
1d9025e5 | 1796 | xfs_trans_log_buf(args->trans, bp, 0, |
1da177e4 | 1797 | sizeof(*tmp_info) - 1); |
1da177e4 LT |
1798 | } |
1799 | } else { | |
5a5881cd | 1800 | trace_xfs_da_unlink_forward(args); |
89da0544 NS |
1801 | save_info->forw = drop_info->forw; |
1802 | if (drop_info->forw) { | |
f5ea1100 | 1803 | error = xfs_da3_node_read(args->trans, args->dp, |
89da0544 | 1804 | be32_to_cpu(drop_info->forw), |
d9392a4b | 1805 | -1, &bp, args->whichfork); |
1da177e4 | 1806 | if (error) |
d99831ff | 1807 | return error; |
1da177e4 | 1808 | ASSERT(bp != NULL); |
1d9025e5 | 1809 | tmp_info = bp->b_addr; |
89da0544 NS |
1810 | ASSERT(tmp_info->magic == save_info->magic); |
1811 | ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno); | |
1812 | tmp_info->back = cpu_to_be32(save_blk->blkno); | |
1d9025e5 | 1813 | xfs_trans_log_buf(args->trans, bp, 0, |
1da177e4 | 1814 | sizeof(*tmp_info) - 1); |
1da177e4 LT |
1815 | } |
1816 | } | |
1817 | ||
1d9025e5 | 1818 | xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1); |
d99831ff | 1819 | return 0; |
1da177e4 LT |
1820 | } |
1821 | ||
1822 | /* | |
1823 | * Move a path "forward" or "!forward" one block at the current level. | |
1824 | * | |
1825 | * This routine will adjust a "path" to point to the next block | |
1826 | * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the | |
1827 | * Btree, including updating pointers to the intermediate nodes between | |
1828 | * the new bottom and the root. | |
1829 | */ | |
1830 | int /* error */ | |
f5ea1100 DC |
1831 | xfs_da3_path_shift( |
1832 | struct xfs_da_state *state, | |
1833 | struct xfs_da_state_path *path, | |
1834 | int forward, | |
1835 | int release, | |
1836 | int *result) | |
1da177e4 | 1837 | { |
f5ea1100 DC |
1838 | struct xfs_da_state_blk *blk; |
1839 | struct xfs_da_blkinfo *info; | |
1840 | struct xfs_da_intnode *node; | |
1841 | struct xfs_da_args *args; | |
1842 | struct xfs_da_node_entry *btree; | |
1843 | struct xfs_da3_icnode_hdr nodehdr; | |
7df1c170 | 1844 | struct xfs_buf *bp; |
f5ea1100 DC |
1845 | xfs_dablk_t blkno = 0; |
1846 | int level; | |
1847 | int error; | |
4bceb18f | 1848 | struct xfs_inode *dp = state->args->dp; |
1da177e4 | 1849 | |
ee73259b DC |
1850 | trace_xfs_da_path_shift(state->args); |
1851 | ||
1da177e4 LT |
1852 | /* |
1853 | * Roll up the Btree looking for the first block where our | |
1854 | * current index is not at the edge of the block. Note that | |
1855 | * we skip the bottom layer because we want the sibling block. | |
1856 | */ | |
1857 | args = state->args; | |
1858 | ASSERT(args != NULL); | |
1859 | ASSERT(path != NULL); | |
1860 | ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH)); | |
1861 | level = (path->active-1) - 1; /* skip bottom layer in path */ | |
1862 | for (blk = &path->blk[level]; level >= 0; blk--, level--) { | |
1d9025e5 | 1863 | node = blk->bp->b_addr; |
01ba43b8 | 1864 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
4bceb18f | 1865 | btree = dp->d_ops->node_tree_p(node); |
f5ea1100 DC |
1866 | |
1867 | if (forward && (blk->index < nodehdr.count - 1)) { | |
1da177e4 | 1868 | blk->index++; |
f5ea1100 | 1869 | blkno = be32_to_cpu(btree[blk->index].before); |
1da177e4 LT |
1870 | break; |
1871 | } else if (!forward && (blk->index > 0)) { | |
1872 | blk->index--; | |
f5ea1100 | 1873 | blkno = be32_to_cpu(btree[blk->index].before); |
1da177e4 LT |
1874 | break; |
1875 | } | |
1876 | } | |
1877 | if (level < 0) { | |
2451337d | 1878 | *result = -ENOENT; /* we're out of our tree */ |
6a178100 | 1879 | ASSERT(args->op_flags & XFS_DA_OP_OKNOENT); |
d99831ff | 1880 | return 0; |
1da177e4 LT |
1881 | } |
1882 | ||
1883 | /* | |
1884 | * Roll down the edge of the subtree until we reach the | |
1885 | * same depth we were at originally. | |
1886 | */ | |
1887 | for (blk++, level++; level < path->active; blk++, level++) { | |
1888 | /* | |
7df1c170 | 1889 | * Read the next child block into a local buffer. |
1da177e4 | 1890 | */ |
7df1c170 BF |
1891 | error = xfs_da3_node_read(args->trans, dp, blkno, -1, &bp, |
1892 | args->whichfork); | |
1893 | if (error) | |
1894 | return error; | |
1da177e4 LT |
1895 | |
1896 | /* | |
7df1c170 BF |
1897 | * Release the old block (if it's dirty, the trans doesn't |
1898 | * actually let go) and swap the local buffer into the path | |
1899 | * structure. This ensures failure of the above read doesn't set | |
1900 | * a NULL buffer in an active slot in the path. | |
1da177e4 | 1901 | */ |
7df1c170 BF |
1902 | if (release) |
1903 | xfs_trans_brelse(args->trans, blk->bp); | |
1da177e4 | 1904 | blk->blkno = blkno; |
7df1c170 BF |
1905 | blk->bp = bp; |
1906 | ||
1d9025e5 | 1907 | info = blk->bp->b_addr; |
69ef921b | 1908 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
f5ea1100 | 1909 | info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) || |
69ef921b | 1910 | info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
24df33b4 | 1911 | info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
517c2220 DC |
1912 | info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || |
1913 | info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); | |
f5ea1100 DC |
1914 | |
1915 | ||
1916 | /* | |
1917 | * Note: we flatten the magic number to a single type so we | |
1918 | * don't have to compare against crc/non-crc types elsewhere. | |
1919 | */ | |
1920 | switch (be16_to_cpu(info->magic)) { | |
1921 | case XFS_DA_NODE_MAGIC: | |
1922 | case XFS_DA3_NODE_MAGIC: | |
1923 | blk->magic = XFS_DA_NODE_MAGIC; | |
1da177e4 | 1924 | node = (xfs_da_intnode_t *)info; |
01ba43b8 | 1925 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
4bceb18f | 1926 | btree = dp->d_ops->node_tree_p(node); |
f5ea1100 | 1927 | blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
1da177e4 LT |
1928 | if (forward) |
1929 | blk->index = 0; | |
1930 | else | |
f5ea1100 DC |
1931 | blk->index = nodehdr.count - 1; |
1932 | blkno = be32_to_cpu(btree[blk->index].before); | |
1933 | break; | |
1934 | case XFS_ATTR_LEAF_MAGIC: | |
517c2220 | 1935 | case XFS_ATTR3_LEAF_MAGIC: |
f5ea1100 | 1936 | blk->magic = XFS_ATTR_LEAF_MAGIC; |
1da177e4 LT |
1937 | ASSERT(level == path->active-1); |
1938 | blk->index = 0; | |
4141956a | 1939 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
f5ea1100 DC |
1940 | break; |
1941 | case XFS_DIR2_LEAFN_MAGIC: | |
1942 | case XFS_DIR3_LEAFN_MAGIC: | |
1943 | blk->magic = XFS_DIR2_LEAFN_MAGIC; | |
1944 | ASSERT(level == path->active-1); | |
1945 | blk->index = 0; | |
8e8877e6 DW |
1946 | blk->hashval = xfs_dir2_leaf_lasthash(args->dp, |
1947 | blk->bp, NULL); | |
f5ea1100 DC |
1948 | break; |
1949 | default: | |
1950 | ASSERT(0); | |
1951 | break; | |
1da177e4 LT |
1952 | } |
1953 | } | |
1954 | *result = 0; | |
f5ea1100 | 1955 | return 0; |
1da177e4 LT |
1956 | } |
1957 | ||
1958 | ||
1959 | /*======================================================================== | |
1960 | * Utility routines. | |
1961 | *========================================================================*/ | |
1962 | ||
1963 | /* | |
1964 | * Implement a simple hash on a character string. | |
1965 | * Rotate the hash value by 7 bits, then XOR each character in. | |
1966 | * This is implemented with some source-level loop unrolling. | |
1967 | */ | |
1968 | xfs_dahash_t | |
c8ce540d | 1969 | xfs_da_hashname(const uint8_t *name, int namelen) |
1da177e4 LT |
1970 | { |
1971 | xfs_dahash_t hash; | |
1972 | ||
1da177e4 LT |
1973 | /* |
1974 | * Do four characters at a time as long as we can. | |
1975 | */ | |
1976 | for (hash = 0; namelen >= 4; namelen -= 4, name += 4) | |
1977 | hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^ | |
1978 | (name[3] << 0) ^ rol32(hash, 7 * 4); | |
1979 | ||
1980 | /* | |
1981 | * Now do the rest of the characters. | |
1982 | */ | |
1983 | switch (namelen) { | |
1984 | case 3: | |
1985 | return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^ | |
1986 | rol32(hash, 7 * 3); | |
1987 | case 2: | |
1988 | return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2); | |
1989 | case 1: | |
1990 | return (name[0] << 0) ^ rol32(hash, 7 * 1); | |
2b3b6d07 | 1991 | default: /* case 0: */ |
1da177e4 LT |
1992 | return hash; |
1993 | } | |
1da177e4 LT |
1994 | } |
1995 | ||
5163f95a BN |
1996 | enum xfs_dacmp |
1997 | xfs_da_compname( | |
1998 | struct xfs_da_args *args, | |
2bc75421 DC |
1999 | const unsigned char *name, |
2000 | int len) | |
5163f95a BN |
2001 | { |
2002 | return (args->namelen == len && memcmp(args->name, name, len) == 0) ? | |
2003 | XFS_CMP_EXACT : XFS_CMP_DIFFERENT; | |
2004 | } | |
2005 | ||
2006 | static xfs_dahash_t | |
2007 | xfs_default_hashname( | |
2008 | struct xfs_name *name) | |
2009 | { | |
2010 | return xfs_da_hashname(name->name, name->len); | |
2011 | } | |
2012 | ||
2013 | const struct xfs_nameops xfs_default_nameops = { | |
2014 | .hashname = xfs_default_hashname, | |
2015 | .compname = xfs_da_compname | |
2016 | }; | |
2017 | ||
1da177e4 | 2018 | int |
77936d02 CH |
2019 | xfs_da_grow_inode_int( |
2020 | struct xfs_da_args *args, | |
2021 | xfs_fileoff_t *bno, | |
2022 | int count) | |
1da177e4 | 2023 | { |
77936d02 CH |
2024 | struct xfs_trans *tp = args->trans; |
2025 | struct xfs_inode *dp = args->dp; | |
2026 | int w = args->whichfork; | |
d5cf09ba | 2027 | xfs_rfsblock_t nblks = dp->i_d.di_nblocks; |
77936d02 CH |
2028 | struct xfs_bmbt_irec map, *mapp; |
2029 | int nmap, error, got, i, mapi; | |
1da177e4 | 2030 | |
1da177e4 LT |
2031 | /* |
2032 | * Find a spot in the file space to put the new block. | |
2033 | */ | |
77936d02 CH |
2034 | error = xfs_bmap_first_unused(tp, dp, count, bno, w); |
2035 | if (error) | |
1da177e4 | 2036 | return error; |
77936d02 | 2037 | |
1da177e4 LT |
2038 | /* |
2039 | * Try mapping it in one filesystem block. | |
2040 | */ | |
2041 | nmap = 1; | |
2042 | ASSERT(args->firstblock != NULL); | |
c0dc7828 DC |
2043 | error = xfs_bmapi_write(tp, dp, *bno, count, |
2044 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG, | |
1da177e4 | 2045 | args->firstblock, args->total, &map, &nmap, |
2c3234d1 | 2046 | args->dfops); |
77936d02 | 2047 | if (error) |
1da177e4 | 2048 | return error; |
77936d02 | 2049 | |
1da177e4 LT |
2050 | ASSERT(nmap <= 1); |
2051 | if (nmap == 1) { | |
2052 | mapp = ↦ | |
2053 | mapi = 1; | |
77936d02 CH |
2054 | } else if (nmap == 0 && count > 1) { |
2055 | xfs_fileoff_t b; | |
2056 | int c; | |
2057 | ||
2058 | /* | |
2059 | * If we didn't get it and the block might work if fragmented, | |
2060 | * try without the CONTIG flag. Loop until we get it all. | |
2061 | */ | |
1da177e4 | 2062 | mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP); |
77936d02 | 2063 | for (b = *bno, mapi = 0; b < *bno + count; ) { |
1da177e4 | 2064 | nmap = MIN(XFS_BMAP_MAX_NMAP, count); |
77936d02 | 2065 | c = (int)(*bno + count - b); |
c0dc7828 DC |
2066 | error = xfs_bmapi_write(tp, dp, b, c, |
2067 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, | |
1da177e4 | 2068 | args->firstblock, args->total, |
2c3234d1 | 2069 | &mapp[mapi], &nmap, args->dfops); |
77936d02 CH |
2070 | if (error) |
2071 | goto out_free_map; | |
1da177e4 LT |
2072 | if (nmap < 1) |
2073 | break; | |
2074 | mapi += nmap; | |
2075 | b = mapp[mapi - 1].br_startoff + | |
2076 | mapp[mapi - 1].br_blockcount; | |
2077 | } | |
2078 | } else { | |
2079 | mapi = 0; | |
2080 | mapp = NULL; | |
2081 | } | |
77936d02 | 2082 | |
1da177e4 LT |
2083 | /* |
2084 | * Count the blocks we got, make sure it matches the total. | |
2085 | */ | |
2086 | for (i = 0, got = 0; i < mapi; i++) | |
2087 | got += mapp[i].br_blockcount; | |
77936d02 | 2088 | if (got != count || mapp[0].br_startoff != *bno || |
1da177e4 | 2089 | mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount != |
77936d02 | 2090 | *bno + count) { |
2451337d | 2091 | error = -ENOSPC; |
77936d02 | 2092 | goto out_free_map; |
1da177e4 | 2093 | } |
77936d02 | 2094 | |
a7444053 DC |
2095 | /* account for newly allocated blocks in reserved blocks total */ |
2096 | args->total -= dp->i_d.di_nblocks - nblks; | |
77936d02 CH |
2097 | |
2098 | out_free_map: | |
2099 | if (mapp != &map) | |
2100 | kmem_free(mapp); | |
2101 | return error; | |
2102 | } | |
2103 | ||
2104 | /* | |
2105 | * Add a block to the btree ahead of the file. | |
2106 | * Return the new block number to the caller. | |
2107 | */ | |
2108 | int | |
2109 | xfs_da_grow_inode( | |
2110 | struct xfs_da_args *args, | |
2111 | xfs_dablk_t *new_blkno) | |
2112 | { | |
2113 | xfs_fileoff_t bno; | |
77936d02 CH |
2114 | int error; |
2115 | ||
5a5881cd DC |
2116 | trace_xfs_da_grow_inode(args); |
2117 | ||
d6cf1305 DC |
2118 | bno = args->geo->leafblk; |
2119 | error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount); | |
77936d02 CH |
2120 | if (!error) |
2121 | *new_blkno = (xfs_dablk_t)bno; | |
2122 | return error; | |
1da177e4 LT |
2123 | } |
2124 | ||
2125 | /* | |
2126 | * Ick. We need to always be able to remove a btree block, even | |
2127 | * if there's no space reservation because the filesystem is full. | |
2128 | * This is called if xfs_bunmapi on a btree block fails due to ENOSPC. | |
2129 | * It swaps the target block with the last block in the file. The | |
2130 | * last block in the file can always be removed since it can't cause | |
2131 | * a bmap btree split to do that. | |
2132 | */ | |
2133 | STATIC int | |
f5ea1100 DC |
2134 | xfs_da3_swap_lastblock( |
2135 | struct xfs_da_args *args, | |
2136 | xfs_dablk_t *dead_blknop, | |
2137 | struct xfs_buf **dead_bufp) | |
1da177e4 | 2138 | { |
f5ea1100 DC |
2139 | struct xfs_da_blkinfo *dead_info; |
2140 | struct xfs_da_blkinfo *sib_info; | |
2141 | struct xfs_da_intnode *par_node; | |
2142 | struct xfs_da_intnode *dead_node; | |
2143 | struct xfs_dir2_leaf *dead_leaf2; | |
2144 | struct xfs_da_node_entry *btree; | |
2145 | struct xfs_da3_icnode_hdr par_hdr; | |
4bceb18f | 2146 | struct xfs_inode *dp; |
f5ea1100 DC |
2147 | struct xfs_trans *tp; |
2148 | struct xfs_mount *mp; | |
2149 | struct xfs_buf *dead_buf; | |
2150 | struct xfs_buf *last_buf; | |
2151 | struct xfs_buf *sib_buf; | |
2152 | struct xfs_buf *par_buf; | |
2153 | xfs_dahash_t dead_hash; | |
2154 | xfs_fileoff_t lastoff; | |
2155 | xfs_dablk_t dead_blkno; | |
2156 | xfs_dablk_t last_blkno; | |
2157 | xfs_dablk_t sib_blkno; | |
2158 | xfs_dablk_t par_blkno; | |
2159 | int error; | |
2160 | int w; | |
2161 | int entno; | |
2162 | int level; | |
2163 | int dead_level; | |
1da177e4 | 2164 | |
5a5881cd DC |
2165 | trace_xfs_da_swap_lastblock(args); |
2166 | ||
1da177e4 LT |
2167 | dead_buf = *dead_bufp; |
2168 | dead_blkno = *dead_blknop; | |
2169 | tp = args->trans; | |
4bceb18f | 2170 | dp = args->dp; |
1da177e4 LT |
2171 | w = args->whichfork; |
2172 | ASSERT(w == XFS_DATA_FORK); | |
4bceb18f | 2173 | mp = dp->i_mount; |
7dda6e86 | 2174 | lastoff = args->geo->freeblk; |
4bceb18f | 2175 | error = xfs_bmap_last_before(tp, dp, &lastoff, w); |
1da177e4 LT |
2176 | if (error) |
2177 | return error; | |
2178 | if (unlikely(lastoff == 0)) { | |
2179 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW, | |
2180 | mp); | |
2451337d | 2181 | return -EFSCORRUPTED; |
1da177e4 LT |
2182 | } |
2183 | /* | |
2184 | * Read the last block in the btree space. | |
2185 | */ | |
d6cf1305 | 2186 | last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount; |
4bceb18f | 2187 | error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w); |
4bb20a83 | 2188 | if (error) |
1da177e4 LT |
2189 | return error; |
2190 | /* | |
2191 | * Copy the last block into the dead buffer and log it. | |
2192 | */ | |
8f66193c DC |
2193 | memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize); |
2194 | xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1); | |
1d9025e5 | 2195 | dead_info = dead_buf->b_addr; |
1da177e4 LT |
2196 | /* |
2197 | * Get values from the moved block. | |
2198 | */ | |
24df33b4 DC |
2199 | if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
2200 | dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
2201 | struct xfs_dir3_icleaf_hdr leafhdr; | |
2202 | struct xfs_dir2_leaf_entry *ents; | |
2203 | ||
1da177e4 | 2204 | dead_leaf2 = (xfs_dir2_leaf_t *)dead_info; |
01ba43b8 | 2205 | dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2); |
4bceb18f | 2206 | ents = dp->d_ops->leaf_ents_p(dead_leaf2); |
1da177e4 | 2207 | dead_level = 0; |
24df33b4 | 2208 | dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval); |
1da177e4 | 2209 | } else { |
f5ea1100 DC |
2210 | struct xfs_da3_icnode_hdr deadhdr; |
2211 | ||
1da177e4 | 2212 | dead_node = (xfs_da_intnode_t *)dead_info; |
01ba43b8 | 2213 | dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node); |
4bceb18f | 2214 | btree = dp->d_ops->node_tree_p(dead_node); |
f5ea1100 DC |
2215 | dead_level = deadhdr.level; |
2216 | dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval); | |
1da177e4 LT |
2217 | } |
2218 | sib_buf = par_buf = NULL; | |
2219 | /* | |
2220 | * If the moved block has a left sibling, fix up the pointers. | |
2221 | */ | |
89da0544 | 2222 | if ((sib_blkno = be32_to_cpu(dead_info->back))) { |
4bceb18f | 2223 | error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w); |
4bb20a83 | 2224 | if (error) |
1da177e4 | 2225 | goto done; |
1d9025e5 | 2226 | sib_info = sib_buf->b_addr; |
1da177e4 | 2227 | if (unlikely( |
89da0544 NS |
2228 | be32_to_cpu(sib_info->forw) != last_blkno || |
2229 | sib_info->magic != dead_info->magic)) { | |
1da177e4 LT |
2230 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)", |
2231 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 2232 | error = -EFSCORRUPTED; |
1da177e4 LT |
2233 | goto done; |
2234 | } | |
89da0544 | 2235 | sib_info->forw = cpu_to_be32(dead_blkno); |
1d9025e5 | 2236 | xfs_trans_log_buf(tp, sib_buf, |
1da177e4 LT |
2237 | XFS_DA_LOGRANGE(sib_info, &sib_info->forw, |
2238 | sizeof(sib_info->forw))); | |
1da177e4 LT |
2239 | sib_buf = NULL; |
2240 | } | |
2241 | /* | |
2242 | * If the moved block has a right sibling, fix up the pointers. | |
2243 | */ | |
89da0544 | 2244 | if ((sib_blkno = be32_to_cpu(dead_info->forw))) { |
4bceb18f | 2245 | error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w); |
4bb20a83 | 2246 | if (error) |
1da177e4 | 2247 | goto done; |
1d9025e5 | 2248 | sib_info = sib_buf->b_addr; |
1da177e4 | 2249 | if (unlikely( |
89da0544 NS |
2250 | be32_to_cpu(sib_info->back) != last_blkno || |
2251 | sib_info->magic != dead_info->magic)) { | |
1da177e4 LT |
2252 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)", |
2253 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 2254 | error = -EFSCORRUPTED; |
1da177e4 LT |
2255 | goto done; |
2256 | } | |
89da0544 | 2257 | sib_info->back = cpu_to_be32(dead_blkno); |
1d9025e5 | 2258 | xfs_trans_log_buf(tp, sib_buf, |
1da177e4 LT |
2259 | XFS_DA_LOGRANGE(sib_info, &sib_info->back, |
2260 | sizeof(sib_info->back))); | |
1da177e4 LT |
2261 | sib_buf = NULL; |
2262 | } | |
7dda6e86 | 2263 | par_blkno = args->geo->leafblk; |
1da177e4 LT |
2264 | level = -1; |
2265 | /* | |
2266 | * Walk down the tree looking for the parent of the moved block. | |
2267 | */ | |
2268 | for (;;) { | |
4bceb18f | 2269 | error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w); |
4bb20a83 | 2270 | if (error) |
1da177e4 | 2271 | goto done; |
1d9025e5 | 2272 | par_node = par_buf->b_addr; |
01ba43b8 | 2273 | dp->d_ops->node_hdr_from_disk(&par_hdr, par_node); |
f5ea1100 | 2274 | if (level >= 0 && level != par_hdr.level + 1) { |
1da177e4 LT |
2275 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)", |
2276 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 2277 | error = -EFSCORRUPTED; |
1da177e4 LT |
2278 | goto done; |
2279 | } | |
f5ea1100 | 2280 | level = par_hdr.level; |
4bceb18f | 2281 | btree = dp->d_ops->node_tree_p(par_node); |
1da177e4 | 2282 | for (entno = 0; |
f5ea1100 DC |
2283 | entno < par_hdr.count && |
2284 | be32_to_cpu(btree[entno].hashval) < dead_hash; | |
1da177e4 LT |
2285 | entno++) |
2286 | continue; | |
f5ea1100 | 2287 | if (entno == par_hdr.count) { |
1da177e4 LT |
2288 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)", |
2289 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 2290 | error = -EFSCORRUPTED; |
1da177e4 LT |
2291 | goto done; |
2292 | } | |
f5ea1100 | 2293 | par_blkno = be32_to_cpu(btree[entno].before); |
1da177e4 LT |
2294 | if (level == dead_level + 1) |
2295 | break; | |
1d9025e5 | 2296 | xfs_trans_brelse(tp, par_buf); |
1da177e4 LT |
2297 | par_buf = NULL; |
2298 | } | |
2299 | /* | |
2300 | * We're in the right parent block. | |
2301 | * Look for the right entry. | |
2302 | */ | |
2303 | for (;;) { | |
2304 | for (; | |
f5ea1100 DC |
2305 | entno < par_hdr.count && |
2306 | be32_to_cpu(btree[entno].before) != last_blkno; | |
1da177e4 LT |
2307 | entno++) |
2308 | continue; | |
f5ea1100 | 2309 | if (entno < par_hdr.count) |
1da177e4 | 2310 | break; |
f5ea1100 | 2311 | par_blkno = par_hdr.forw; |
1d9025e5 | 2312 | xfs_trans_brelse(tp, par_buf); |
1da177e4 LT |
2313 | par_buf = NULL; |
2314 | if (unlikely(par_blkno == 0)) { | |
2315 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)", | |
2316 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 2317 | error = -EFSCORRUPTED; |
1da177e4 LT |
2318 | goto done; |
2319 | } | |
4bceb18f | 2320 | error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w); |
4bb20a83 | 2321 | if (error) |
1da177e4 | 2322 | goto done; |
1d9025e5 | 2323 | par_node = par_buf->b_addr; |
01ba43b8 | 2324 | dp->d_ops->node_hdr_from_disk(&par_hdr, par_node); |
f5ea1100 | 2325 | if (par_hdr.level != level) { |
1da177e4 LT |
2326 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)", |
2327 | XFS_ERRLEVEL_LOW, mp); | |
2451337d | 2328 | error = -EFSCORRUPTED; |
1da177e4 LT |
2329 | goto done; |
2330 | } | |
4bceb18f | 2331 | btree = dp->d_ops->node_tree_p(par_node); |
1da177e4 LT |
2332 | entno = 0; |
2333 | } | |
2334 | /* | |
2335 | * Update the parent entry pointing to the moved block. | |
2336 | */ | |
f5ea1100 | 2337 | btree[entno].before = cpu_to_be32(dead_blkno); |
1d9025e5 | 2338 | xfs_trans_log_buf(tp, par_buf, |
f5ea1100 DC |
2339 | XFS_DA_LOGRANGE(par_node, &btree[entno].before, |
2340 | sizeof(btree[entno].before))); | |
1da177e4 LT |
2341 | *dead_blknop = last_blkno; |
2342 | *dead_bufp = last_buf; | |
2343 | return 0; | |
2344 | done: | |
2345 | if (par_buf) | |
1d9025e5 | 2346 | xfs_trans_brelse(tp, par_buf); |
1da177e4 | 2347 | if (sib_buf) |
1d9025e5 DC |
2348 | xfs_trans_brelse(tp, sib_buf); |
2349 | xfs_trans_brelse(tp, last_buf); | |
1da177e4 LT |
2350 | return error; |
2351 | } | |
2352 | ||
2353 | /* | |
2354 | * Remove a btree block from a directory or attribute. | |
2355 | */ | |
2356 | int | |
1d9025e5 DC |
2357 | xfs_da_shrink_inode( |
2358 | xfs_da_args_t *args, | |
2359 | xfs_dablk_t dead_blkno, | |
2360 | struct xfs_buf *dead_buf) | |
1da177e4 LT |
2361 | { |
2362 | xfs_inode_t *dp; | |
2363 | int done, error, w, count; | |
1da177e4 | 2364 | xfs_trans_t *tp; |
1da177e4 | 2365 | |
5a5881cd DC |
2366 | trace_xfs_da_shrink_inode(args); |
2367 | ||
1da177e4 LT |
2368 | dp = args->dp; |
2369 | w = args->whichfork; | |
2370 | tp = args->trans; | |
d6cf1305 | 2371 | count = args->geo->fsbcount; |
1da177e4 LT |
2372 | for (;;) { |
2373 | /* | |
2374 | * Remove extents. If we get ENOSPC for a dir we have to move | |
2375 | * the last block to the place we want to kill. | |
2376 | */ | |
f5ea1100 | 2377 | error = xfs_bunmapi(tp, dp, dead_blkno, count, |
ab7bb610 | 2378 | xfs_bmapi_aflag(w), 0, args->firstblock, |
2c3234d1 | 2379 | args->dfops, &done); |
2451337d | 2380 | if (error == -ENOSPC) { |
1da177e4 | 2381 | if (w != XFS_DATA_FORK) |
f6c2d1fa | 2382 | break; |
f5ea1100 DC |
2383 | error = xfs_da3_swap_lastblock(args, &dead_blkno, |
2384 | &dead_buf); | |
2385 | if (error) | |
f6c2d1fa NS |
2386 | break; |
2387 | } else { | |
1da177e4 | 2388 | break; |
1da177e4 LT |
2389 | } |
2390 | } | |
1d9025e5 | 2391 | xfs_trans_binval(tp, dead_buf); |
1da177e4 LT |
2392 | return error; |
2393 | } | |
2394 | ||
2395 | /* | |
2396 | * See if the mapping(s) for this btree block are valid, i.e. | |
2397 | * don't contain holes, are logically contiguous, and cover the whole range. | |
2398 | */ | |
2399 | STATIC int | |
2400 | xfs_da_map_covers_blocks( | |
2401 | int nmap, | |
2402 | xfs_bmbt_irec_t *mapp, | |
2403 | xfs_dablk_t bno, | |
2404 | int count) | |
2405 | { | |
2406 | int i; | |
2407 | xfs_fileoff_t off; | |
2408 | ||
2409 | for (i = 0, off = bno; i < nmap; i++) { | |
2410 | if (mapp[i].br_startblock == HOLESTARTBLOCK || | |
2411 | mapp[i].br_startblock == DELAYSTARTBLOCK) { | |
2412 | return 0; | |
2413 | } | |
2414 | if (off != mapp[i].br_startoff) { | |
2415 | return 0; | |
2416 | } | |
2417 | off += mapp[i].br_blockcount; | |
2418 | } | |
2419 | return off == bno + count; | |
2420 | } | |
2421 | ||
2422 | /* | |
3605431f DC |
2423 | * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map. |
2424 | * | |
2425 | * For the single map case, it is assumed that the caller has provided a pointer | |
2426 | * to a valid xfs_buf_map. For the multiple map case, this function will | |
2427 | * allocate the xfs_buf_map to hold all the maps and replace the caller's single | |
2428 | * map pointer with the allocated map. | |
1da177e4 | 2429 | */ |
3605431f DC |
2430 | static int |
2431 | xfs_buf_map_from_irec( | |
2432 | struct xfs_mount *mp, | |
2433 | struct xfs_buf_map **mapp, | |
836a94ad | 2434 | int *nmaps, |
3605431f | 2435 | struct xfs_bmbt_irec *irecs, |
836a94ad | 2436 | int nirecs) |
1da177e4 | 2437 | { |
3605431f DC |
2438 | struct xfs_buf_map *map; |
2439 | int i; | |
2440 | ||
2441 | ASSERT(*nmaps == 1); | |
2442 | ASSERT(nirecs >= 1); | |
2443 | ||
2444 | if (nirecs > 1) { | |
b17cb364 DC |
2445 | map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), |
2446 | KM_SLEEP | KM_NOFS); | |
3605431f | 2447 | if (!map) |
2451337d | 2448 | return -ENOMEM; |
3605431f DC |
2449 | *mapp = map; |
2450 | } | |
2451 | ||
2452 | *nmaps = nirecs; | |
2453 | map = *mapp; | |
2454 | for (i = 0; i < *nmaps; i++) { | |
2455 | ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK && | |
2456 | irecs[i].br_startblock != HOLESTARTBLOCK); | |
2457 | map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock); | |
2458 | map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount); | |
2459 | } | |
2460 | return 0; | |
2461 | } | |
2462 | ||
2463 | /* | |
2464 | * Map the block we are given ready for reading. There are three possible return | |
2465 | * values: | |
2466 | * -1 - will be returned if we land in a hole and mappedbno == -2 so the | |
2467 | * caller knows not to execute a subsequent read. | |
2468 | * 0 - if we mapped the block successfully | |
2469 | * >0 - positive error number if there was an error. | |
2470 | */ | |
2471 | static int | |
2472 | xfs_dabuf_map( | |
3605431f DC |
2473 | struct xfs_inode *dp, |
2474 | xfs_dablk_t bno, | |
2475 | xfs_daddr_t mappedbno, | |
2476 | int whichfork, | |
2477 | struct xfs_buf_map **map, | |
2478 | int *nmaps) | |
2479 | { | |
2480 | struct xfs_mount *mp = dp->i_mount; | |
2481 | int nfsb; | |
2482 | int error = 0; | |
2483 | struct xfs_bmbt_irec irec; | |
2484 | struct xfs_bmbt_irec *irecs = &irec; | |
2485 | int nirecs; | |
2486 | ||
2487 | ASSERT(map && *map); | |
2488 | ASSERT(*nmaps == 1); | |
1da177e4 | 2489 | |
d6cf1305 DC |
2490 | if (whichfork == XFS_DATA_FORK) |
2491 | nfsb = mp->m_dir_geo->fsbcount; | |
2492 | else | |
2493 | nfsb = mp->m_attr_geo->fsbcount; | |
3605431f | 2494 | |
1da177e4 LT |
2495 | /* |
2496 | * Caller doesn't have a mapping. -2 means don't complain | |
2497 | * if we land in a hole. | |
2498 | */ | |
2499 | if (mappedbno == -1 || mappedbno == -2) { | |
2500 | /* | |
2501 | * Optimize the one-block case. | |
2502 | */ | |
3605431f | 2503 | if (nfsb != 1) |
b17cb364 DC |
2504 | irecs = kmem_zalloc(sizeof(irec) * nfsb, |
2505 | KM_SLEEP | KM_NOFS); | |
5b777ad5 | 2506 | |
3605431f DC |
2507 | nirecs = nfsb; |
2508 | error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs, | |
2509 | &nirecs, xfs_bmapi_aflag(whichfork)); | |
5b777ad5 | 2510 | if (error) |
3605431f | 2511 | goto out; |
1da177e4 | 2512 | } else { |
3605431f DC |
2513 | irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno); |
2514 | irecs->br_startoff = (xfs_fileoff_t)bno; | |
2515 | irecs->br_blockcount = nfsb; | |
2516 | irecs->br_state = 0; | |
2517 | nirecs = 1; | |
1da177e4 | 2518 | } |
3605431f DC |
2519 | |
2520 | if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) { | |
2451337d DC |
2521 | error = mappedbno == -2 ? -1 : -EFSCORRUPTED; |
2522 | if (unlikely(error == -EFSCORRUPTED)) { | |
1da177e4 | 2523 | if (xfs_error_level >= XFS_ERRLEVEL_LOW) { |
3605431f | 2524 | int i; |
0b932ccc DC |
2525 | xfs_alert(mp, "%s: bno %lld dir: inode %lld", |
2526 | __func__, (long long)bno, | |
1da177e4 | 2527 | (long long)dp->i_ino); |
3605431f | 2528 | for (i = 0; i < *nmaps; i++) { |
0b932ccc DC |
2529 | xfs_alert(mp, |
2530 | "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d", | |
1da177e4 | 2531 | i, |
3605431f DC |
2532 | (long long)irecs[i].br_startoff, |
2533 | (long long)irecs[i].br_startblock, | |
2534 | (long long)irecs[i].br_blockcount, | |
2535 | irecs[i].br_state); | |
1da177e4 LT |
2536 | } |
2537 | } | |
2538 | XFS_ERROR_REPORT("xfs_da_do_buf(1)", | |
2539 | XFS_ERRLEVEL_LOW, mp); | |
2540 | } | |
3605431f | 2541 | goto out; |
1da177e4 | 2542 | } |
3605431f DC |
2543 | error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs); |
2544 | out: | |
2545 | if (irecs != &irec) | |
2546 | kmem_free(irecs); | |
2547 | return error; | |
2548 | } | |
2549 | ||
2550 | /* | |
2551 | * Get a buffer for the dir/attr block. | |
2552 | */ | |
2553 | int | |
2554 | xfs_da_get_buf( | |
2555 | struct xfs_trans *trans, | |
2556 | struct xfs_inode *dp, | |
2557 | xfs_dablk_t bno, | |
2558 | xfs_daddr_t mappedbno, | |
1d9025e5 | 2559 | struct xfs_buf **bpp, |
3605431f DC |
2560 | int whichfork) |
2561 | { | |
2562 | struct xfs_buf *bp; | |
2563 | struct xfs_buf_map map; | |
2564 | struct xfs_buf_map *mapp; | |
2565 | int nmap; | |
2566 | int error; | |
2567 | ||
2568 | *bpp = NULL; | |
2569 | mapp = ↦ | |
2570 | nmap = 1; | |
9df2dd0b | 2571 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
3605431f DC |
2572 | &mapp, &nmap); |
2573 | if (error) { | |
2574 | /* mapping a hole is not an error, but we don't continue */ | |
2575 | if (error == -1) | |
1da177e4 | 2576 | error = 0; |
3605431f | 2577 | goto out_free; |
1da177e4 | 2578 | } |
3605431f DC |
2579 | |
2580 | bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp, | |
2581 | mapp, nmap, 0); | |
2451337d | 2582 | error = bp ? bp->b_error : -EIO; |
3605431f | 2583 | if (error) { |
6ee49a20 ES |
2584 | if (bp) |
2585 | xfs_trans_brelse(trans, bp); | |
3605431f DC |
2586 | goto out_free; |
2587 | } | |
2588 | ||
1d9025e5 | 2589 | *bpp = bp; |
3605431f DC |
2590 | |
2591 | out_free: | |
2592 | if (mapp != &map) | |
2593 | kmem_free(mapp); | |
2594 | ||
2595 | return error; | |
2596 | } | |
2597 | ||
2598 | /* | |
2599 | * Get a buffer for the dir/attr block, fill in the contents. | |
2600 | */ | |
2601 | int | |
2602 | xfs_da_read_buf( | |
2603 | struct xfs_trans *trans, | |
2604 | struct xfs_inode *dp, | |
2605 | xfs_dablk_t bno, | |
2606 | xfs_daddr_t mappedbno, | |
1d9025e5 | 2607 | struct xfs_buf **bpp, |
4bb20a83 | 2608 | int whichfork, |
1813dd64 | 2609 | const struct xfs_buf_ops *ops) |
3605431f DC |
2610 | { |
2611 | struct xfs_buf *bp; | |
2612 | struct xfs_buf_map map; | |
2613 | struct xfs_buf_map *mapp; | |
2614 | int nmap; | |
2615 | int error; | |
2616 | ||
2617 | *bpp = NULL; | |
2618 | mapp = ↦ | |
2619 | nmap = 1; | |
9df2dd0b | 2620 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
3605431f DC |
2621 | &mapp, &nmap); |
2622 | if (error) { | |
2623 | /* mapping a hole is not an error, but we don't continue */ | |
2624 | if (error == -1) | |
2625 | error = 0; | |
2626 | goto out_free; | |
2627 | } | |
2628 | ||
2629 | error = xfs_trans_read_buf_map(dp->i_mount, trans, | |
2630 | dp->i_mount->m_ddev_targp, | |
1813dd64 | 2631 | mapp, nmap, 0, &bp, ops); |
3605431f DC |
2632 | if (error) |
2633 | goto out_free; | |
2634 | ||
2635 | if (whichfork == XFS_ATTR_FORK) | |
2636 | xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF); | |
1da177e4 | 2637 | else |
3605431f | 2638 | xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF); |
1d9025e5 | 2639 | *bpp = bp; |
3605431f | 2640 | out_free: |
1da177e4 | 2641 | if (mapp != &map) |
f0e2d93c | 2642 | kmem_free(mapp); |
1da177e4 | 2643 | |
3605431f | 2644 | return error; |
1da177e4 LT |
2645 | } |
2646 | ||
2647 | /* | |
2648 | * Readahead the dir/attr block. | |
2649 | */ | |
7a652bbe | 2650 | int |
1da177e4 | 2651 | xfs_da_reada_buf( |
3605431f DC |
2652 | struct xfs_inode *dp, |
2653 | xfs_dablk_t bno, | |
da6958c8 | 2654 | xfs_daddr_t mappedbno, |
4bb20a83 | 2655 | int whichfork, |
1813dd64 | 2656 | const struct xfs_buf_ops *ops) |
1da177e4 | 2657 | { |
3605431f DC |
2658 | struct xfs_buf_map map; |
2659 | struct xfs_buf_map *mapp; | |
2660 | int nmap; | |
2661 | int error; | |
2662 | ||
2663 | mapp = ↦ | |
2664 | nmap = 1; | |
9df2dd0b | 2665 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
3605431f DC |
2666 | &mapp, &nmap); |
2667 | if (error) { | |
2668 | /* mapping a hole is not an error, but we don't continue */ | |
2669 | if (error == -1) | |
2670 | error = 0; | |
2671 | goto out_free; | |
2672 | } | |
1da177e4 | 2673 | |
3605431f | 2674 | mappedbno = mapp[0].bm_bn; |
1813dd64 | 2675 | xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops); |
3605431f DC |
2676 | |
2677 | out_free: | |
2678 | if (mapp != &map) | |
2679 | kmem_free(mapp); | |
2680 | ||
7a652bbe | 2681 | return error; |
1da177e4 | 2682 | } |