Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 LT |
22 | #include "xfs_log.h" |
23 | #include "xfs_trans.h" | |
24 | #include "xfs_sb.h" | |
25 | #include "xfs_ag.h" | |
1da177e4 | 26 | #include "xfs_mount.h" |
a844f451 | 27 | #include "xfs_da_btree.h" |
1da177e4 | 28 | #include "xfs_bmap_btree.h" |
57926640 CH |
29 | #include "xfs_dir2.h" |
30 | #include "xfs_dir2_format.h" | |
31 | #include "xfs_dir2_priv.h" | |
1da177e4 | 32 | #include "xfs_dinode.h" |
1da177e4 | 33 | #include "xfs_inode.h" |
a844f451 NS |
34 | #include "xfs_inode_item.h" |
35 | #include "xfs_alloc.h" | |
1da177e4 | 36 | #include "xfs_bmap.h" |
1da177e4 LT |
37 | #include "xfs_attr.h" |
38 | #include "xfs_attr_leaf.h" | |
1da177e4 | 39 | #include "xfs_error.h" |
0b1b213f | 40 | #include "xfs_trace.h" |
1da177e4 LT |
41 | |
42 | /* | |
43 | * xfs_da_btree.c | |
44 | * | |
45 | * Routines to implement directories as Btrees of hashed names. | |
46 | */ | |
47 | ||
48 | /*======================================================================== | |
49 | * Function prototypes for the kernel. | |
50 | *========================================================================*/ | |
51 | ||
52 | /* | |
53 | * Routines used for growing the Btree. | |
54 | */ | |
55 | STATIC int xfs_da_root_split(xfs_da_state_t *state, | |
56 | xfs_da_state_blk_t *existing_root, | |
57 | xfs_da_state_blk_t *new_child); | |
58 | STATIC int xfs_da_node_split(xfs_da_state_t *state, | |
59 | xfs_da_state_blk_t *existing_blk, | |
60 | xfs_da_state_blk_t *split_blk, | |
61 | xfs_da_state_blk_t *blk_to_add, | |
62 | int treelevel, | |
63 | int *result); | |
64 | STATIC void xfs_da_node_rebalance(xfs_da_state_t *state, | |
65 | xfs_da_state_blk_t *node_blk_1, | |
66 | xfs_da_state_blk_t *node_blk_2); | |
67 | STATIC void xfs_da_node_add(xfs_da_state_t *state, | |
68 | xfs_da_state_blk_t *old_node_blk, | |
69 | xfs_da_state_blk_t *new_node_blk); | |
70 | ||
71 | /* | |
72 | * Routines used for shrinking the Btree. | |
73 | */ | |
74 | STATIC int xfs_da_root_join(xfs_da_state_t *state, | |
75 | xfs_da_state_blk_t *root_blk); | |
76 | STATIC int xfs_da_node_toosmall(xfs_da_state_t *state, int *retval); | |
77 | STATIC void xfs_da_node_remove(xfs_da_state_t *state, | |
78 | xfs_da_state_blk_t *drop_blk); | |
79 | STATIC void xfs_da_node_unbalance(xfs_da_state_t *state, | |
80 | xfs_da_state_blk_t *src_node_blk, | |
81 | xfs_da_state_blk_t *dst_node_blk); | |
82 | ||
83 | /* | |
84 | * Utility routines. | |
85 | */ | |
1d9025e5 DC |
86 | STATIC uint xfs_da_node_lasthash(struct xfs_buf *bp, int *count); |
87 | STATIC int xfs_da_node_order(struct xfs_buf *node1_bp, | |
88 | struct xfs_buf *node2_bp); | |
ba0f32d4 CH |
89 | STATIC int xfs_da_blk_unlink(xfs_da_state_t *state, |
90 | xfs_da_state_blk_t *drop_blk, | |
91 | xfs_da_state_blk_t *save_blk); | |
92 | STATIC void xfs_da_state_kill_altpath(xfs_da_state_t *state); | |
1da177e4 | 93 | |
d9392a4b | 94 | static void |
612cfbfe | 95 | xfs_da_node_verify( |
d9392a4b DC |
96 | struct xfs_buf *bp) |
97 | { | |
98 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
99 | struct xfs_da_node_hdr *hdr = bp->b_addr; | |
100 | int block_ok = 0; | |
101 | ||
102 | block_ok = hdr->info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC); | |
103 | block_ok = block_ok && | |
104 | be16_to_cpu(hdr->level) > 0 && | |
105 | be16_to_cpu(hdr->count) > 0 ; | |
106 | if (!block_ok) { | |
107 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr); | |
108 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
109 | } | |
110 | ||
d9392a4b DC |
111 | } |
112 | ||
113 | static void | |
612cfbfe DC |
114 | xfs_da_node_write_verify( |
115 | struct xfs_buf *bp) | |
116 | { | |
117 | xfs_da_node_verify(bp); | |
118 | } | |
119 | ||
1813dd64 DC |
120 | /* |
121 | * leaf/node format detection on trees is sketchy, so a node read can be done on | |
122 | * leaf level blocks when detection identifies the tree as a node format tree | |
123 | * incorrectly. In this case, we need to swap the verifier to match the correct | |
124 | * format of the block being read. | |
125 | */ | |
612cfbfe DC |
126 | static void |
127 | xfs_da_node_read_verify( | |
d9392a4b DC |
128 | struct xfs_buf *bp) |
129 | { | |
130 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
131 | struct xfs_da_blkinfo *info = bp->b_addr; | |
132 | ||
133 | switch (be16_to_cpu(info->magic)) { | |
134 | case XFS_DA_NODE_MAGIC: | |
612cfbfe DC |
135 | xfs_da_node_verify(bp); |
136 | break; | |
d9392a4b | 137 | case XFS_ATTR_LEAF_MAGIC: |
1813dd64 DC |
138 | bp->b_ops = &xfs_attr_leaf_buf_ops; |
139 | bp->b_ops->verify_read(bp); | |
d9392a4b DC |
140 | return; |
141 | case XFS_DIR2_LEAFN_MAGIC: | |
24df33b4 DC |
142 | case XFS_DIR3_LEAFN_MAGIC: |
143 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
1813dd64 | 144 | bp->b_ops->verify_read(bp); |
d9392a4b DC |
145 | return; |
146 | default: | |
612cfbfe DC |
147 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, |
148 | mp, info); | |
149 | xfs_buf_ioerror(bp, EFSCORRUPTED); | |
d9392a4b DC |
150 | break; |
151 | } | |
d9392a4b DC |
152 | } |
153 | ||
1813dd64 DC |
154 | const struct xfs_buf_ops xfs_da_node_buf_ops = { |
155 | .verify_read = xfs_da_node_read_verify, | |
156 | .verify_write = xfs_da_node_write_verify, | |
157 | }; | |
158 | ||
159 | ||
d9392a4b DC |
160 | int |
161 | xfs_da_node_read( | |
162 | struct xfs_trans *tp, | |
163 | struct xfs_inode *dp, | |
164 | xfs_dablk_t bno, | |
165 | xfs_daddr_t mappedbno, | |
166 | struct xfs_buf **bpp, | |
167 | int which_fork) | |
168 | { | |
169 | return xfs_da_read_buf(tp, dp, bno, mappedbno, bpp, | |
1813dd64 | 170 | which_fork, &xfs_da_node_buf_ops); |
d9392a4b DC |
171 | } |
172 | ||
1da177e4 LT |
173 | /*======================================================================== |
174 | * Routines used for growing the Btree. | |
175 | *========================================================================*/ | |
176 | ||
177 | /* | |
178 | * Create the initial contents of an intermediate node. | |
179 | */ | |
180 | int | |
181 | xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level, | |
1d9025e5 | 182 | struct xfs_buf **bpp, int whichfork) |
1da177e4 LT |
183 | { |
184 | xfs_da_intnode_t *node; | |
1d9025e5 | 185 | struct xfs_buf *bp; |
1da177e4 LT |
186 | int error; |
187 | xfs_trans_t *tp; | |
188 | ||
5a5881cd DC |
189 | trace_xfs_da_node_create(args); |
190 | ||
1da177e4 LT |
191 | tp = args->trans; |
192 | error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork); | |
193 | if (error) | |
194 | return(error); | |
195 | ASSERT(bp != NULL); | |
1d9025e5 | 196 | node = bp->b_addr; |
1da177e4 LT |
197 | node->hdr.info.forw = 0; |
198 | node->hdr.info.back = 0; | |
89da0544 | 199 | node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC); |
1da177e4 LT |
200 | node->hdr.info.pad = 0; |
201 | node->hdr.count = 0; | |
fac80cce | 202 | node->hdr.level = cpu_to_be16(level); |
1da177e4 | 203 | |
1d9025e5 | 204 | xfs_trans_log_buf(tp, bp, |
1da177e4 LT |
205 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
206 | ||
1813dd64 | 207 | bp->b_ops = &xfs_da_node_buf_ops; |
1da177e4 LT |
208 | *bpp = bp; |
209 | return(0); | |
210 | } | |
211 | ||
212 | /* | |
213 | * Split a leaf node, rebalance, then possibly split | |
214 | * intermediate nodes, rebalance, etc. | |
215 | */ | |
216 | int /* error */ | |
217 | xfs_da_split(xfs_da_state_t *state) | |
218 | { | |
219 | xfs_da_state_blk_t *oldblk, *newblk, *addblk; | |
220 | xfs_da_intnode_t *node; | |
1d9025e5 | 221 | struct xfs_buf *bp; |
1da177e4 LT |
222 | int max, action, error, i; |
223 | ||
5a5881cd DC |
224 | trace_xfs_da_split(state->args); |
225 | ||
1da177e4 LT |
226 | /* |
227 | * Walk back up the tree splitting/inserting/adjusting as necessary. | |
228 | * If we need to insert and there isn't room, split the node, then | |
229 | * decide which fragment to insert the new block from below into. | |
230 | * Note that we may split the root this way, but we need more fixup. | |
231 | */ | |
232 | max = state->path.active - 1; | |
233 | ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH)); | |
234 | ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC || | |
f6c2d1fa | 235 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); |
1da177e4 LT |
236 | |
237 | addblk = &state->path.blk[max]; /* initial dummy value */ | |
238 | for (i = max; (i >= 0) && addblk; state->path.active--, i--) { | |
239 | oldblk = &state->path.blk[i]; | |
240 | newblk = &state->altpath.blk[i]; | |
241 | ||
242 | /* | |
243 | * If a leaf node then | |
244 | * Allocate a new leaf node, then rebalance across them. | |
245 | * else if an intermediate node then | |
246 | * We split on the last layer, must we split the node? | |
247 | */ | |
248 | switch (oldblk->magic) { | |
249 | case XFS_ATTR_LEAF_MAGIC: | |
1da177e4 LT |
250 | error = xfs_attr_leaf_split(state, oldblk, newblk); |
251 | if ((error != 0) && (error != ENOSPC)) { | |
252 | return(error); /* GROT: attr is inconsistent */ | |
253 | } | |
254 | if (!error) { | |
255 | addblk = newblk; | |
256 | break; | |
257 | } | |
258 | /* | |
259 | * Entry wouldn't fit, split the leaf again. | |
260 | */ | |
261 | state->extravalid = 1; | |
262 | if (state->inleaf) { | |
263 | state->extraafter = 0; /* before newblk */ | |
5a5881cd | 264 | trace_xfs_attr_leaf_split_before(state->args); |
1da177e4 LT |
265 | error = xfs_attr_leaf_split(state, oldblk, |
266 | &state->extrablk); | |
267 | } else { | |
268 | state->extraafter = 1; /* after newblk */ | |
5a5881cd | 269 | trace_xfs_attr_leaf_split_after(state->args); |
1da177e4 LT |
270 | error = xfs_attr_leaf_split(state, newblk, |
271 | &state->extrablk); | |
272 | } | |
273 | if (error) | |
274 | return(error); /* GROT: attr inconsistent */ | |
275 | addblk = newblk; | |
276 | break; | |
1da177e4 | 277 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
278 | error = xfs_dir2_leafn_split(state, oldblk, newblk); |
279 | if (error) | |
280 | return error; | |
281 | addblk = newblk; | |
282 | break; | |
283 | case XFS_DA_NODE_MAGIC: | |
284 | error = xfs_da_node_split(state, oldblk, newblk, addblk, | |
285 | max - i, &action); | |
1da177e4 LT |
286 | addblk->bp = NULL; |
287 | if (error) | |
288 | return(error); /* GROT: dir is inconsistent */ | |
289 | /* | |
290 | * Record the newly split block for the next time thru? | |
291 | */ | |
292 | if (action) | |
293 | addblk = newblk; | |
294 | else | |
295 | addblk = NULL; | |
296 | break; | |
297 | } | |
298 | ||
299 | /* | |
300 | * Update the btree to show the new hashval for this child. | |
301 | */ | |
302 | xfs_da_fixhashpath(state, &state->path); | |
1da177e4 LT |
303 | } |
304 | if (!addblk) | |
305 | return(0); | |
306 | ||
307 | /* | |
308 | * Split the root node. | |
309 | */ | |
310 | ASSERT(state->path.active == 0); | |
311 | oldblk = &state->path.blk[0]; | |
312 | error = xfs_da_root_split(state, oldblk, addblk); | |
313 | if (error) { | |
1da177e4 LT |
314 | addblk->bp = NULL; |
315 | return(error); /* GROT: dir is inconsistent */ | |
316 | } | |
317 | ||
318 | /* | |
319 | * Update pointers to the node which used to be block 0 and | |
320 | * just got bumped because of the addition of a new root node. | |
321 | * There might be three blocks involved if a double split occurred, | |
322 | * and the original block 0 could be at any position in the list. | |
323 | */ | |
324 | ||
1d9025e5 | 325 | node = oldblk->bp->b_addr; |
1da177e4 | 326 | if (node->hdr.info.forw) { |
89da0544 | 327 | if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) { |
1da177e4 LT |
328 | bp = addblk->bp; |
329 | } else { | |
330 | ASSERT(state->extravalid); | |
331 | bp = state->extrablk.bp; | |
332 | } | |
1d9025e5 | 333 | node = bp->b_addr; |
89da0544 | 334 | node->hdr.info.back = cpu_to_be32(oldblk->blkno); |
1d9025e5 | 335 | xfs_trans_log_buf(state->args->trans, bp, |
1da177e4 LT |
336 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
337 | sizeof(node->hdr.info))); | |
338 | } | |
1d9025e5 | 339 | node = oldblk->bp->b_addr; |
89da0544 NS |
340 | if (node->hdr.info.back) { |
341 | if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) { | |
1da177e4 LT |
342 | bp = addblk->bp; |
343 | } else { | |
344 | ASSERT(state->extravalid); | |
345 | bp = state->extrablk.bp; | |
346 | } | |
1d9025e5 | 347 | node = bp->b_addr; |
89da0544 | 348 | node->hdr.info.forw = cpu_to_be32(oldblk->blkno); |
1d9025e5 | 349 | xfs_trans_log_buf(state->args->trans, bp, |
1da177e4 LT |
350 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
351 | sizeof(node->hdr.info))); | |
352 | } | |
1da177e4 LT |
353 | addblk->bp = NULL; |
354 | return(0); | |
355 | } | |
356 | ||
357 | /* | |
358 | * Split the root. We have to create a new root and point to the two | |
359 | * parts (the split old root) that we just created. Copy block zero to | |
360 | * the EOF, extending the inode in process. | |
361 | */ | |
362 | STATIC int /* error */ | |
363 | xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1, | |
364 | xfs_da_state_blk_t *blk2) | |
365 | { | |
366 | xfs_da_intnode_t *node, *oldroot; | |
367 | xfs_da_args_t *args; | |
368 | xfs_dablk_t blkno; | |
1d9025e5 | 369 | struct xfs_buf *bp; |
1da177e4 LT |
370 | int error, size; |
371 | xfs_inode_t *dp; | |
372 | xfs_trans_t *tp; | |
373 | xfs_mount_t *mp; | |
374 | xfs_dir2_leaf_t *leaf; | |
375 | ||
5a5881cd DC |
376 | trace_xfs_da_root_split(state->args); |
377 | ||
1da177e4 LT |
378 | /* |
379 | * Copy the existing (incorrect) block from the root node position | |
380 | * to a free space somewhere. | |
381 | */ | |
382 | args = state->args; | |
383 | ASSERT(args != NULL); | |
384 | error = xfs_da_grow_inode(args, &blkno); | |
385 | if (error) | |
386 | return(error); | |
387 | dp = args->dp; | |
388 | tp = args->trans; | |
389 | mp = state->mp; | |
390 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork); | |
391 | if (error) | |
392 | return(error); | |
393 | ASSERT(bp != NULL); | |
1d9025e5 DC |
394 | node = bp->b_addr; |
395 | oldroot = blk1->bp->b_addr; | |
69ef921b | 396 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) { |
fac80cce | 397 | size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] - |
1da177e4 LT |
398 | (char *)oldroot); |
399 | } else { | |
24df33b4 DC |
400 | struct xfs_dir3_icleaf_hdr leafhdr; |
401 | struct xfs_dir2_leaf_entry *ents; | |
402 | ||
1da177e4 | 403 | leaf = (xfs_dir2_leaf_t *)oldroot; |
24df33b4 DC |
404 | xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf); |
405 | ents = xfs_dir3_leaf_ents_p(leaf); | |
406 | ||
407 | ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC || | |
408 | leafhdr.magic == XFS_DIR3_LEAFN_MAGIC); | |
409 | size = (int)((char *)&ents[leafhdr.count] - (char *)leaf); | |
1da177e4 | 410 | } |
24df33b4 | 411 | /* XXX: can't just copy CRC headers from one block to another */ |
1da177e4 | 412 | memcpy(node, oldroot, size); |
1d9025e5 | 413 | xfs_trans_log_buf(tp, bp, 0, size - 1); |
b0f539de | 414 | |
1813dd64 | 415 | bp->b_ops = blk1->bp->b_ops; |
1da177e4 LT |
416 | blk1->bp = bp; |
417 | blk1->blkno = blkno; | |
418 | ||
419 | /* | |
420 | * Set up the new root node. | |
421 | */ | |
422 | error = xfs_da_node_create(args, | |
f6c2d1fa | 423 | (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0, |
fac80cce | 424 | be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork); |
1da177e4 LT |
425 | if (error) |
426 | return(error); | |
1d9025e5 | 427 | node = bp->b_addr; |
403432dc NS |
428 | node->btree[0].hashval = cpu_to_be32(blk1->hashval); |
429 | node->btree[0].before = cpu_to_be32(blk1->blkno); | |
430 | node->btree[1].hashval = cpu_to_be32(blk2->hashval); | |
431 | node->btree[1].before = cpu_to_be32(blk2->blkno); | |
fac80cce | 432 | node->hdr.count = cpu_to_be16(2); |
1da177e4 LT |
433 | |
434 | #ifdef DEBUG | |
24df33b4 DC |
435 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
436 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
1da177e4 LT |
437 | ASSERT(blk1->blkno >= mp->m_dirleafblk && |
438 | blk1->blkno < mp->m_dirfreeblk); | |
439 | ASSERT(blk2->blkno >= mp->m_dirleafblk && | |
440 | blk2->blkno < mp->m_dirfreeblk); | |
441 | } | |
442 | #endif | |
443 | ||
444 | /* Header is already logged by xfs_da_node_create */ | |
1d9025e5 | 445 | xfs_trans_log_buf(tp, bp, |
1da177e4 LT |
446 | XFS_DA_LOGRANGE(node, node->btree, |
447 | sizeof(xfs_da_node_entry_t) * 2)); | |
1da177e4 LT |
448 | |
449 | return(0); | |
450 | } | |
451 | ||
452 | /* | |
453 | * Split the node, rebalance, then add the new entry. | |
454 | */ | |
455 | STATIC int /* error */ | |
456 | xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk, | |
457 | xfs_da_state_blk_t *newblk, | |
458 | xfs_da_state_blk_t *addblk, | |
459 | int treelevel, int *result) | |
460 | { | |
461 | xfs_da_intnode_t *node; | |
462 | xfs_dablk_t blkno; | |
463 | int newcount, error; | |
464 | int useextra; | |
465 | ||
5a5881cd DC |
466 | trace_xfs_da_node_split(state->args); |
467 | ||
1d9025e5 | 468 | node = oldblk->bp->b_addr; |
69ef921b | 469 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 LT |
470 | |
471 | /* | |
f6c2d1fa | 472 | * With V2 dirs the extra block is data or freespace. |
1da177e4 | 473 | */ |
f6c2d1fa | 474 | useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK; |
1da177e4 LT |
475 | newcount = 1 + useextra; |
476 | /* | |
477 | * Do we have to split the node? | |
478 | */ | |
fac80cce | 479 | if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) { |
1da177e4 LT |
480 | /* |
481 | * Allocate a new node, add to the doubly linked chain of | |
482 | * nodes, then move some of our excess entries into it. | |
483 | */ | |
484 | error = xfs_da_grow_inode(state->args, &blkno); | |
485 | if (error) | |
486 | return(error); /* GROT: dir is inconsistent */ | |
487 | ||
488 | error = xfs_da_node_create(state->args, blkno, treelevel, | |
489 | &newblk->bp, state->args->whichfork); | |
490 | if (error) | |
491 | return(error); /* GROT: dir is inconsistent */ | |
492 | newblk->blkno = blkno; | |
493 | newblk->magic = XFS_DA_NODE_MAGIC; | |
494 | xfs_da_node_rebalance(state, oldblk, newblk); | |
495 | error = xfs_da_blk_link(state, oldblk, newblk); | |
496 | if (error) | |
497 | return(error); | |
498 | *result = 1; | |
499 | } else { | |
500 | *result = 0; | |
501 | } | |
502 | ||
503 | /* | |
504 | * Insert the new entry(s) into the correct block | |
505 | * (updating last hashval in the process). | |
506 | * | |
507 | * xfs_da_node_add() inserts BEFORE the given index, | |
508 | * and as a result of using node_lookup_int() we always | |
509 | * point to a valid entry (not after one), but a split | |
510 | * operation always results in a new block whose hashvals | |
511 | * FOLLOW the current block. | |
512 | * | |
513 | * If we had double-split op below us, then add the extra block too. | |
514 | */ | |
1d9025e5 | 515 | node = oldblk->bp->b_addr; |
fac80cce | 516 | if (oldblk->index <= be16_to_cpu(node->hdr.count)) { |
1da177e4 LT |
517 | oldblk->index++; |
518 | xfs_da_node_add(state, oldblk, addblk); | |
519 | if (useextra) { | |
520 | if (state->extraafter) | |
521 | oldblk->index++; | |
522 | xfs_da_node_add(state, oldblk, &state->extrablk); | |
523 | state->extravalid = 0; | |
524 | } | |
525 | } else { | |
526 | newblk->index++; | |
527 | xfs_da_node_add(state, newblk, addblk); | |
528 | if (useextra) { | |
529 | if (state->extraafter) | |
530 | newblk->index++; | |
531 | xfs_da_node_add(state, newblk, &state->extrablk); | |
532 | state->extravalid = 0; | |
533 | } | |
534 | } | |
535 | ||
536 | return(0); | |
537 | } | |
538 | ||
539 | /* | |
540 | * Balance the btree elements between two intermediate nodes, | |
541 | * usually one full and one empty. | |
542 | * | |
543 | * NOTE: if blk2 is empty, then it will get the upper half of blk1. | |
544 | */ | |
545 | STATIC void | |
546 | xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1, | |
547 | xfs_da_state_blk_t *blk2) | |
548 | { | |
549 | xfs_da_intnode_t *node1, *node2, *tmpnode; | |
550 | xfs_da_node_entry_t *btree_s, *btree_d; | |
551 | int count, tmp; | |
552 | xfs_trans_t *tp; | |
553 | ||
5a5881cd DC |
554 | trace_xfs_da_node_rebalance(state->args); |
555 | ||
1d9025e5 DC |
556 | node1 = blk1->bp->b_addr; |
557 | node2 = blk2->bp->b_addr; | |
1da177e4 LT |
558 | /* |
559 | * Figure out how many entries need to move, and in which direction. | |
560 | * Swap the nodes around if that makes it simpler. | |
561 | */ | |
fac80cce | 562 | if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) && |
403432dc | 563 | ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) || |
fac80cce NS |
564 | (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) < |
565 | be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) { | |
1da177e4 LT |
566 | tmpnode = node1; |
567 | node1 = node2; | |
568 | node2 = tmpnode; | |
569 | } | |
69ef921b CH |
570 | ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
571 | ASSERT(node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
fac80cce | 572 | count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2; |
1da177e4 LT |
573 | if (count == 0) |
574 | return; | |
575 | tp = state->args->trans; | |
576 | /* | |
577 | * Two cases: high-to-low and low-to-high. | |
578 | */ | |
579 | if (count > 0) { | |
580 | /* | |
581 | * Move elements in node2 up to make a hole. | |
582 | */ | |
fac80cce | 583 | if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) { |
1da177e4 LT |
584 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
585 | btree_s = &node2->btree[0]; | |
586 | btree_d = &node2->btree[count]; | |
587 | memmove(btree_d, btree_s, tmp); | |
588 | } | |
589 | ||
590 | /* | |
591 | * Move the req'd B-tree elements from high in node1 to | |
592 | * low in node2. | |
593 | */ | |
413d57c9 | 594 | be16_add_cpu(&node2->hdr.count, count); |
1da177e4 | 595 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
fac80cce | 596 | btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count]; |
1da177e4 LT |
597 | btree_d = &node2->btree[0]; |
598 | memcpy(btree_d, btree_s, tmp); | |
413d57c9 | 599 | be16_add_cpu(&node1->hdr.count, -count); |
1da177e4 LT |
600 | } else { |
601 | /* | |
602 | * Move the req'd B-tree elements from low in node2 to | |
603 | * high in node1. | |
604 | */ | |
605 | count = -count; | |
606 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); | |
607 | btree_s = &node2->btree[0]; | |
fac80cce | 608 | btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)]; |
1da177e4 | 609 | memcpy(btree_d, btree_s, tmp); |
413d57c9 | 610 | be16_add_cpu(&node1->hdr.count, count); |
1d9025e5 | 611 | xfs_trans_log_buf(tp, blk1->bp, |
1da177e4 LT |
612 | XFS_DA_LOGRANGE(node1, btree_d, tmp)); |
613 | ||
614 | /* | |
615 | * Move elements in node2 down to fill the hole. | |
616 | */ | |
fac80cce | 617 | tmp = be16_to_cpu(node2->hdr.count) - count; |
1da177e4 LT |
618 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
619 | btree_s = &node2->btree[count]; | |
620 | btree_d = &node2->btree[0]; | |
621 | memmove(btree_d, btree_s, tmp); | |
413d57c9 | 622 | be16_add_cpu(&node2->hdr.count, -count); |
1da177e4 LT |
623 | } |
624 | ||
625 | /* | |
626 | * Log header of node 1 and all current bits of node 2. | |
627 | */ | |
1d9025e5 | 628 | xfs_trans_log_buf(tp, blk1->bp, |
1da177e4 | 629 | XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr))); |
1d9025e5 | 630 | xfs_trans_log_buf(tp, blk2->bp, |
1da177e4 LT |
631 | XFS_DA_LOGRANGE(node2, &node2->hdr, |
632 | sizeof(node2->hdr) + | |
fac80cce | 633 | sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count))); |
1da177e4 LT |
634 | |
635 | /* | |
636 | * Record the last hashval from each block for upward propagation. | |
637 | * (note: don't use the swapped node pointers) | |
638 | */ | |
1d9025e5 DC |
639 | node1 = blk1->bp->b_addr; |
640 | node2 = blk2->bp->b_addr; | |
fac80cce NS |
641 | blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval); |
642 | blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval); | |
1da177e4 LT |
643 | |
644 | /* | |
645 | * Adjust the expected index for insertion. | |
646 | */ | |
fac80cce NS |
647 | if (blk1->index >= be16_to_cpu(node1->hdr.count)) { |
648 | blk2->index = blk1->index - be16_to_cpu(node1->hdr.count); | |
649 | blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */ | |
1da177e4 LT |
650 | } |
651 | } | |
652 | ||
653 | /* | |
654 | * Add a new entry to an intermediate node. | |
655 | */ | |
656 | STATIC void | |
657 | xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk, | |
658 | xfs_da_state_blk_t *newblk) | |
659 | { | |
660 | xfs_da_intnode_t *node; | |
661 | xfs_da_node_entry_t *btree; | |
662 | int tmp; | |
1da177e4 | 663 | |
5a5881cd DC |
664 | trace_xfs_da_node_add(state->args); |
665 | ||
1d9025e5 | 666 | node = oldblk->bp->b_addr; |
69ef921b | 667 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
fac80cce | 668 | ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count))); |
1da177e4 | 669 | ASSERT(newblk->blkno != 0); |
f6c2d1fa | 670 | if (state->args->whichfork == XFS_DATA_FORK) |
73523a2e CH |
671 | ASSERT(newblk->blkno >= state->mp->m_dirleafblk && |
672 | newblk->blkno < state->mp->m_dirfreeblk); | |
1da177e4 LT |
673 | |
674 | /* | |
675 | * We may need to make some room before we insert the new node. | |
676 | */ | |
677 | tmp = 0; | |
678 | btree = &node->btree[ oldblk->index ]; | |
fac80cce NS |
679 | if (oldblk->index < be16_to_cpu(node->hdr.count)) { |
680 | tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree); | |
1da177e4 LT |
681 | memmove(btree + 1, btree, tmp); |
682 | } | |
403432dc NS |
683 | btree->hashval = cpu_to_be32(newblk->hashval); |
684 | btree->before = cpu_to_be32(newblk->blkno); | |
1d9025e5 | 685 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
1da177e4 | 686 | XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree))); |
413d57c9 | 687 | be16_add_cpu(&node->hdr.count, 1); |
1d9025e5 | 688 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
1da177e4 LT |
689 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
690 | ||
691 | /* | |
692 | * Copy the last hash value from the oldblk to propagate upwards. | |
693 | */ | |
fac80cce | 694 | oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval); |
1da177e4 LT |
695 | } |
696 | ||
697 | /*======================================================================== | |
698 | * Routines used for shrinking the Btree. | |
699 | *========================================================================*/ | |
700 | ||
701 | /* | |
702 | * Deallocate an empty leaf node, remove it from its parent, | |
703 | * possibly deallocating that block, etc... | |
704 | */ | |
705 | int | |
706 | xfs_da_join(xfs_da_state_t *state) | |
707 | { | |
708 | xfs_da_state_blk_t *drop_blk, *save_blk; | |
709 | int action, error; | |
710 | ||
5a5881cd DC |
711 | trace_xfs_da_join(state->args); |
712 | ||
1da177e4 LT |
713 | action = 0; |
714 | drop_blk = &state->path.blk[ state->path.active-1 ]; | |
715 | save_blk = &state->altpath.blk[ state->path.active-1 ]; | |
716 | ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC); | |
717 | ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC || | |
f6c2d1fa | 718 | drop_blk->magic == XFS_DIR2_LEAFN_MAGIC); |
1da177e4 LT |
719 | |
720 | /* | |
721 | * Walk back up the tree joining/deallocating as necessary. | |
722 | * When we stop dropping blocks, break out. | |
723 | */ | |
724 | for ( ; state->path.active >= 2; drop_blk--, save_blk--, | |
725 | state->path.active--) { | |
726 | /* | |
727 | * See if we can combine the block with a neighbor. | |
728 | * (action == 0) => no options, just leave | |
729 | * (action == 1) => coalesce, then unlink | |
730 | * (action == 2) => block empty, unlink it | |
731 | */ | |
732 | switch (drop_blk->magic) { | |
733 | case XFS_ATTR_LEAF_MAGIC: | |
1da177e4 | 734 | error = xfs_attr_leaf_toosmall(state, &action); |
1da177e4 LT |
735 | if (error) |
736 | return(error); | |
737 | if (action == 0) | |
738 | return(0); | |
1da177e4 | 739 | xfs_attr_leaf_unbalance(state, drop_blk, save_blk); |
1da177e4 | 740 | break; |
1da177e4 | 741 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
742 | error = xfs_dir2_leafn_toosmall(state, &action); |
743 | if (error) | |
744 | return error; | |
745 | if (action == 0) | |
746 | return 0; | |
747 | xfs_dir2_leafn_unbalance(state, drop_blk, save_blk); | |
748 | break; | |
749 | case XFS_DA_NODE_MAGIC: | |
750 | /* | |
751 | * Remove the offending node, fixup hashvals, | |
752 | * check for a toosmall neighbor. | |
753 | */ | |
754 | xfs_da_node_remove(state, drop_blk); | |
755 | xfs_da_fixhashpath(state, &state->path); | |
756 | error = xfs_da_node_toosmall(state, &action); | |
757 | if (error) | |
758 | return(error); | |
759 | if (action == 0) | |
760 | return 0; | |
761 | xfs_da_node_unbalance(state, drop_blk, save_blk); | |
762 | break; | |
763 | } | |
764 | xfs_da_fixhashpath(state, &state->altpath); | |
765 | error = xfs_da_blk_unlink(state, drop_blk, save_blk); | |
766 | xfs_da_state_kill_altpath(state); | |
767 | if (error) | |
768 | return(error); | |
769 | error = xfs_da_shrink_inode(state->args, drop_blk->blkno, | |
770 | drop_blk->bp); | |
771 | drop_blk->bp = NULL; | |
772 | if (error) | |
773 | return(error); | |
774 | } | |
775 | /* | |
776 | * We joined all the way to the top. If it turns out that | |
777 | * we only have one entry in the root, make the child block | |
778 | * the new root. | |
779 | */ | |
780 | xfs_da_node_remove(state, drop_blk); | |
781 | xfs_da_fixhashpath(state, &state->path); | |
782 | error = xfs_da_root_join(state, &state->path.blk[0]); | |
783 | return(error); | |
784 | } | |
785 | ||
1c4f3329 AE |
786 | #ifdef DEBUG |
787 | static void | |
788 | xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level) | |
789 | { | |
790 | __be16 magic = blkinfo->magic; | |
791 | ||
792 | if (level == 1) { | |
793 | ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
24df33b4 | 794 | magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
1c4f3329 AE |
795 | magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)); |
796 | } else | |
797 | ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
798 | ASSERT(!blkinfo->forw); | |
799 | ASSERT(!blkinfo->back); | |
800 | } | |
801 | #else /* !DEBUG */ | |
802 | #define xfs_da_blkinfo_onlychild_validate(blkinfo, level) | |
803 | #endif /* !DEBUG */ | |
804 | ||
1da177e4 LT |
805 | /* |
806 | * We have only one entry in the root. Copy the only remaining child of | |
807 | * the old root to block 0 as the new root node. | |
808 | */ | |
809 | STATIC int | |
810 | xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk) | |
811 | { | |
812 | xfs_da_intnode_t *oldroot; | |
1da177e4 LT |
813 | xfs_da_args_t *args; |
814 | xfs_dablk_t child; | |
1d9025e5 | 815 | struct xfs_buf *bp; |
1da177e4 LT |
816 | int error; |
817 | ||
5a5881cd DC |
818 | trace_xfs_da_root_join(state->args); |
819 | ||
1da177e4 LT |
820 | args = state->args; |
821 | ASSERT(args != NULL); | |
822 | ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC); | |
1d9025e5 | 823 | oldroot = root_blk->bp->b_addr; |
69ef921b | 824 | ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 LT |
825 | ASSERT(!oldroot->hdr.info.forw); |
826 | ASSERT(!oldroot->hdr.info.back); | |
827 | ||
828 | /* | |
829 | * If the root has more than one child, then don't do anything. | |
830 | */ | |
fac80cce | 831 | if (be16_to_cpu(oldroot->hdr.count) > 1) |
1da177e4 LT |
832 | return(0); |
833 | ||
834 | /* | |
835 | * Read in the (only) child block, then copy those bytes into | |
836 | * the root block's buffer and free the original child block. | |
837 | */ | |
403432dc | 838 | child = be32_to_cpu(oldroot->btree[0].before); |
1da177e4 | 839 | ASSERT(child != 0); |
d9392a4b DC |
840 | error = xfs_da_node_read(args->trans, args->dp, child, -1, &bp, |
841 | args->whichfork); | |
1da177e4 LT |
842 | if (error) |
843 | return(error); | |
844 | ASSERT(bp != NULL); | |
1d9025e5 | 845 | xfs_da_blkinfo_onlychild_validate(bp->b_addr, |
1c4f3329 AE |
846 | be16_to_cpu(oldroot->hdr.level)); |
847 | ||
612cfbfe DC |
848 | /* |
849 | * This could be copying a leaf back into the root block in the case of | |
850 | * there only being a single leaf block left in the tree. Hence we have | |
1813dd64 | 851 | * to update the b_ops pointer as well to match the buffer type change |
612cfbfe DC |
852 | * that could occur. |
853 | */ | |
1d9025e5 | 854 | memcpy(root_blk->bp->b_addr, bp->b_addr, state->blocksize); |
1813dd64 | 855 | root_blk->bp->b_ops = bp->b_ops; |
1d9025e5 | 856 | xfs_trans_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1); |
1da177e4 LT |
857 | error = xfs_da_shrink_inode(args, child, bp); |
858 | return(error); | |
859 | } | |
860 | ||
861 | /* | |
862 | * Check a node block and its neighbors to see if the block should be | |
863 | * collapsed into one or the other neighbor. Always keep the block | |
864 | * with the smaller block number. | |
865 | * If the current block is over 50% full, don't try to join it, return 0. | |
866 | * If the block is empty, fill in the state structure and return 2. | |
867 | * If it can be collapsed, fill in the state structure and return 1. | |
868 | * If nothing can be done, return 0. | |
869 | */ | |
870 | STATIC int | |
871 | xfs_da_node_toosmall(xfs_da_state_t *state, int *action) | |
872 | { | |
873 | xfs_da_intnode_t *node; | |
874 | xfs_da_state_blk_t *blk; | |
875 | xfs_da_blkinfo_t *info; | |
876 | int count, forward, error, retval, i; | |
877 | xfs_dablk_t blkno; | |
1d9025e5 | 878 | struct xfs_buf *bp; |
1da177e4 | 879 | |
ee73259b DC |
880 | trace_xfs_da_node_toosmall(state->args); |
881 | ||
1da177e4 LT |
882 | /* |
883 | * Check for the degenerate case of the block being over 50% full. | |
884 | * If so, it's not worth even looking to see if we might be able | |
885 | * to coalesce with a sibling. | |
886 | */ | |
887 | blk = &state->path.blk[ state->path.active-1 ]; | |
1d9025e5 | 888 | info = blk->bp->b_addr; |
69ef921b | 889 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 | 890 | node = (xfs_da_intnode_t *)info; |
fac80cce | 891 | count = be16_to_cpu(node->hdr.count); |
1da177e4 LT |
892 | if (count > (state->node_ents >> 1)) { |
893 | *action = 0; /* blk over 50%, don't try to join */ | |
894 | return(0); /* blk over 50%, don't try to join */ | |
895 | } | |
896 | ||
897 | /* | |
898 | * Check for the degenerate case of the block being empty. | |
899 | * If the block is empty, we'll simply delete it, no need to | |
c41564b5 | 900 | * coalesce it with a sibling block. We choose (arbitrarily) |
1da177e4 LT |
901 | * to merge with the forward block unless it is NULL. |
902 | */ | |
903 | if (count == 0) { | |
904 | /* | |
905 | * Make altpath point to the block we want to keep and | |
906 | * path point to the block we want to drop (this one). | |
907 | */ | |
89da0544 | 908 | forward = (info->forw != 0); |
1da177e4 LT |
909 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
910 | error = xfs_da_path_shift(state, &state->altpath, forward, | |
911 | 0, &retval); | |
912 | if (error) | |
913 | return(error); | |
914 | if (retval) { | |
915 | *action = 0; | |
916 | } else { | |
917 | *action = 2; | |
918 | } | |
919 | return(0); | |
920 | } | |
921 | ||
922 | /* | |
923 | * Examine each sibling block to see if we can coalesce with | |
924 | * at least 25% free space to spare. We need to figure out | |
925 | * whether to merge with the forward or the backward block. | |
926 | * We prefer coalescing with the lower numbered sibling so as | |
927 | * to shrink a directory over time. | |
928 | */ | |
929 | /* start with smaller blk num */ | |
89da0544 | 930 | forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back)); |
1da177e4 LT |
931 | for (i = 0; i < 2; forward = !forward, i++) { |
932 | if (forward) | |
89da0544 | 933 | blkno = be32_to_cpu(info->forw); |
1da177e4 | 934 | else |
89da0544 | 935 | blkno = be32_to_cpu(info->back); |
1da177e4 LT |
936 | if (blkno == 0) |
937 | continue; | |
d9392a4b DC |
938 | error = xfs_da_node_read(state->args->trans, state->args->dp, |
939 | blkno, -1, &bp, state->args->whichfork); | |
1da177e4 LT |
940 | if (error) |
941 | return(error); | |
942 | ASSERT(bp != NULL); | |
943 | ||
944 | node = (xfs_da_intnode_t *)info; | |
945 | count = state->node_ents; | |
946 | count -= state->node_ents >> 2; | |
fac80cce | 947 | count -= be16_to_cpu(node->hdr.count); |
1d9025e5 | 948 | node = bp->b_addr; |
69ef921b | 949 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
fac80cce | 950 | count -= be16_to_cpu(node->hdr.count); |
1d9025e5 | 951 | xfs_trans_brelse(state->args->trans, bp); |
1da177e4 LT |
952 | if (count >= 0) |
953 | break; /* fits with at least 25% to spare */ | |
954 | } | |
955 | if (i >= 2) { | |
956 | *action = 0; | |
957 | return(0); | |
958 | } | |
959 | ||
960 | /* | |
961 | * Make altpath point to the block we want to keep (the lower | |
962 | * numbered block) and path point to the block we want to drop. | |
963 | */ | |
964 | memcpy(&state->altpath, &state->path, sizeof(state->path)); | |
965 | if (blkno < blk->blkno) { | |
966 | error = xfs_da_path_shift(state, &state->altpath, forward, | |
967 | 0, &retval); | |
968 | if (error) { | |
969 | return(error); | |
970 | } | |
971 | if (retval) { | |
972 | *action = 0; | |
973 | return(0); | |
974 | } | |
975 | } else { | |
976 | error = xfs_da_path_shift(state, &state->path, forward, | |
977 | 0, &retval); | |
978 | if (error) { | |
979 | return(error); | |
980 | } | |
981 | if (retval) { | |
982 | *action = 0; | |
983 | return(0); | |
984 | } | |
985 | } | |
986 | *action = 1; | |
987 | return(0); | |
988 | } | |
989 | ||
990 | /* | |
991 | * Walk back up the tree adjusting hash values as necessary, | |
992 | * when we stop making changes, return. | |
993 | */ | |
994 | void | |
995 | xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path) | |
996 | { | |
997 | xfs_da_state_blk_t *blk; | |
998 | xfs_da_intnode_t *node; | |
999 | xfs_da_node_entry_t *btree; | |
1000 | xfs_dahash_t lasthash=0; | |
1001 | int level, count; | |
1002 | ||
ee73259b DC |
1003 | trace_xfs_da_fixhashpath(state->args); |
1004 | ||
1da177e4 LT |
1005 | level = path->active-1; |
1006 | blk = &path->blk[ level ]; | |
1007 | switch (blk->magic) { | |
1da177e4 LT |
1008 | case XFS_ATTR_LEAF_MAGIC: |
1009 | lasthash = xfs_attr_leaf_lasthash(blk->bp, &count); | |
1010 | if (count == 0) | |
1011 | return; | |
1012 | break; | |
1da177e4 | 1013 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
1014 | lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count); |
1015 | if (count == 0) | |
1016 | return; | |
1017 | break; | |
1018 | case XFS_DA_NODE_MAGIC: | |
1019 | lasthash = xfs_da_node_lasthash(blk->bp, &count); | |
1020 | if (count == 0) | |
1021 | return; | |
1022 | break; | |
1023 | } | |
1024 | for (blk--, level--; level >= 0; blk--, level--) { | |
1d9025e5 | 1025 | node = blk->bp->b_addr; |
69ef921b | 1026 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 | 1027 | btree = &node->btree[ blk->index ]; |
403432dc | 1028 | if (be32_to_cpu(btree->hashval) == lasthash) |
1da177e4 LT |
1029 | break; |
1030 | blk->hashval = lasthash; | |
403432dc | 1031 | btree->hashval = cpu_to_be32(lasthash); |
1d9025e5 | 1032 | xfs_trans_log_buf(state->args->trans, blk->bp, |
1da177e4 LT |
1033 | XFS_DA_LOGRANGE(node, btree, sizeof(*btree))); |
1034 | ||
fac80cce | 1035 | lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
1da177e4 LT |
1036 | } |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * Remove an entry from an intermediate node. | |
1041 | */ | |
1042 | STATIC void | |
1043 | xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk) | |
1044 | { | |
1045 | xfs_da_intnode_t *node; | |
1046 | xfs_da_node_entry_t *btree; | |
1047 | int tmp; | |
1048 | ||
5a5881cd DC |
1049 | trace_xfs_da_node_remove(state->args); |
1050 | ||
1d9025e5 | 1051 | node = drop_blk->bp->b_addr; |
fac80cce | 1052 | ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count)); |
1da177e4 LT |
1053 | ASSERT(drop_blk->index >= 0); |
1054 | ||
1055 | /* | |
1056 | * Copy over the offending entry, or just zero it out. | |
1057 | */ | |
1058 | btree = &node->btree[drop_blk->index]; | |
fac80cce NS |
1059 | if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) { |
1060 | tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1; | |
1da177e4 LT |
1061 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
1062 | memmove(btree, btree + 1, tmp); | |
1d9025e5 | 1063 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
1da177e4 | 1064 | XFS_DA_LOGRANGE(node, btree, tmp)); |
fac80cce | 1065 | btree = &node->btree[be16_to_cpu(node->hdr.count)-1]; |
1da177e4 LT |
1066 | } |
1067 | memset((char *)btree, 0, sizeof(xfs_da_node_entry_t)); | |
1d9025e5 | 1068 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
1da177e4 | 1069 | XFS_DA_LOGRANGE(node, btree, sizeof(*btree))); |
413d57c9 | 1070 | be16_add_cpu(&node->hdr.count, -1); |
1d9025e5 | 1071 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
1da177e4 LT |
1072 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
1073 | ||
1074 | /* | |
1075 | * Copy the last hash value from the block to propagate upwards. | |
1076 | */ | |
1077 | btree--; | |
403432dc | 1078 | drop_blk->hashval = be32_to_cpu(btree->hashval); |
1da177e4 LT |
1079 | } |
1080 | ||
1081 | /* | |
1082 | * Unbalance the btree elements between two intermediate nodes, | |
1083 | * move all Btree elements from one node into another. | |
1084 | */ | |
1085 | STATIC void | |
1086 | xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk, | |
1087 | xfs_da_state_blk_t *save_blk) | |
1088 | { | |
1089 | xfs_da_intnode_t *drop_node, *save_node; | |
1090 | xfs_da_node_entry_t *btree; | |
1091 | int tmp; | |
1092 | xfs_trans_t *tp; | |
1093 | ||
5a5881cd DC |
1094 | trace_xfs_da_node_unbalance(state->args); |
1095 | ||
1d9025e5 DC |
1096 | drop_node = drop_blk->bp->b_addr; |
1097 | save_node = save_blk->bp->b_addr; | |
69ef921b CH |
1098 | ASSERT(drop_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1099 | ASSERT(save_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
1da177e4 LT |
1100 | tp = state->args->trans; |
1101 | ||
1102 | /* | |
1103 | * If the dying block has lower hashvals, then move all the | |
1104 | * elements in the remaining block up to make a hole. | |
1105 | */ | |
403432dc | 1106 | if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) || |
fac80cce NS |
1107 | (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) < |
1108 | be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval))) | |
1da177e4 | 1109 | { |
fac80cce NS |
1110 | btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)]; |
1111 | tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t); | |
1da177e4 LT |
1112 | memmove(btree, &save_node->btree[0], tmp); |
1113 | btree = &save_node->btree[0]; | |
1d9025e5 | 1114 | xfs_trans_log_buf(tp, save_blk->bp, |
1da177e4 | 1115 | XFS_DA_LOGRANGE(save_node, btree, |
fac80cce | 1116 | (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) * |
1da177e4 LT |
1117 | sizeof(xfs_da_node_entry_t))); |
1118 | } else { | |
fac80cce | 1119 | btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)]; |
1d9025e5 | 1120 | xfs_trans_log_buf(tp, save_blk->bp, |
1da177e4 | 1121 | XFS_DA_LOGRANGE(save_node, btree, |
fac80cce | 1122 | be16_to_cpu(drop_node->hdr.count) * |
1da177e4 LT |
1123 | sizeof(xfs_da_node_entry_t))); |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * Move all the B-tree elements from drop_blk to save_blk. | |
1128 | */ | |
fac80cce | 1129 | tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t); |
1da177e4 | 1130 | memcpy(btree, &drop_node->btree[0], tmp); |
413d57c9 | 1131 | be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count)); |
1da177e4 | 1132 | |
1d9025e5 | 1133 | xfs_trans_log_buf(tp, save_blk->bp, |
1da177e4 LT |
1134 | XFS_DA_LOGRANGE(save_node, &save_node->hdr, |
1135 | sizeof(save_node->hdr))); | |
1136 | ||
1137 | /* | |
1138 | * Save the last hashval in the remaining block for upward propagation. | |
1139 | */ | |
fac80cce | 1140 | save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval); |
1da177e4 LT |
1141 | } |
1142 | ||
1143 | /*======================================================================== | |
1144 | * Routines used for finding things in the Btree. | |
1145 | *========================================================================*/ | |
1146 | ||
1147 | /* | |
1148 | * Walk down the Btree looking for a particular filename, filling | |
1149 | * in the state structure as we go. | |
1150 | * | |
1151 | * We will set the state structure to point to each of the elements | |
1152 | * in each of the nodes where either the hashval is or should be. | |
1153 | * | |
1154 | * We support duplicate hashval's so for each entry in the current | |
1155 | * node that could contain the desired hashval, descend. This is a | |
1156 | * pruned depth-first tree search. | |
1157 | */ | |
1158 | int /* error */ | |
1159 | xfs_da_node_lookup_int(xfs_da_state_t *state, int *result) | |
1160 | { | |
1161 | xfs_da_state_blk_t *blk; | |
1162 | xfs_da_blkinfo_t *curr; | |
1163 | xfs_da_intnode_t *node; | |
1164 | xfs_da_node_entry_t *btree; | |
1165 | xfs_dablk_t blkno; | |
1166 | int probe, span, max, error, retval; | |
d432c80e | 1167 | xfs_dahash_t hashval, btreehashval; |
1da177e4 LT |
1168 | xfs_da_args_t *args; |
1169 | ||
1170 | args = state->args; | |
1171 | ||
1172 | /* | |
1173 | * Descend thru the B-tree searching each level for the right | |
1174 | * node to use, until the right hashval is found. | |
1175 | */ | |
f6c2d1fa | 1176 | blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0; |
1da177e4 LT |
1177 | for (blk = &state->path.blk[0], state->path.active = 1; |
1178 | state->path.active <= XFS_DA_NODE_MAXDEPTH; | |
1179 | blk++, state->path.active++) { | |
1180 | /* | |
1181 | * Read the next node down in the tree. | |
1182 | */ | |
1183 | blk->blkno = blkno; | |
d9392a4b DC |
1184 | error = xfs_da_node_read(args->trans, args->dp, blkno, |
1185 | -1, &blk->bp, args->whichfork); | |
1da177e4 LT |
1186 | if (error) { |
1187 | blk->blkno = 0; | |
1188 | state->path.active--; | |
1189 | return(error); | |
1190 | } | |
1d9025e5 | 1191 | curr = blk->bp->b_addr; |
d432c80e NS |
1192 | blk->magic = be16_to_cpu(curr->magic); |
1193 | ASSERT(blk->magic == XFS_DA_NODE_MAGIC || | |
1194 | blk->magic == XFS_DIR2_LEAFN_MAGIC || | |
1195 | blk->magic == XFS_ATTR_LEAF_MAGIC); | |
1da177e4 LT |
1196 | |
1197 | /* | |
1198 | * Search an intermediate node for a match. | |
1199 | */ | |
89da0544 | 1200 | if (blk->magic == XFS_DA_NODE_MAGIC) { |
1d9025e5 | 1201 | node = blk->bp->b_addr; |
d432c80e | 1202 | max = be16_to_cpu(node->hdr.count); |
1c91ad3a | 1203 | blk->hashval = be32_to_cpu(node->btree[max-1].hashval); |
1da177e4 LT |
1204 | |
1205 | /* | |
1206 | * Binary search. (note: small blocks will skip loop) | |
1207 | */ | |
1da177e4 LT |
1208 | probe = span = max / 2; |
1209 | hashval = args->hashval; | |
1210 | for (btree = &node->btree[probe]; span > 4; | |
1211 | btree = &node->btree[probe]) { | |
1212 | span /= 2; | |
d432c80e NS |
1213 | btreehashval = be32_to_cpu(btree->hashval); |
1214 | if (btreehashval < hashval) | |
1da177e4 | 1215 | probe += span; |
d432c80e | 1216 | else if (btreehashval > hashval) |
1da177e4 LT |
1217 | probe -= span; |
1218 | else | |
1219 | break; | |
1220 | } | |
1221 | ASSERT((probe >= 0) && (probe < max)); | |
403432dc | 1222 | ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval)); |
1da177e4 LT |
1223 | |
1224 | /* | |
1225 | * Since we may have duplicate hashval's, find the first | |
1226 | * matching hashval in the node. | |
1227 | */ | |
403432dc | 1228 | while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) { |
1da177e4 LT |
1229 | btree--; |
1230 | probe--; | |
1231 | } | |
403432dc | 1232 | while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) { |
1da177e4 LT |
1233 | btree++; |
1234 | probe++; | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * Pick the right block to descend on. | |
1239 | */ | |
1240 | if (probe == max) { | |
1241 | blk->index = max-1; | |
403432dc | 1242 | blkno = be32_to_cpu(node->btree[max-1].before); |
1da177e4 LT |
1243 | } else { |
1244 | blk->index = probe; | |
403432dc | 1245 | blkno = be32_to_cpu(btree->before); |
1da177e4 | 1246 | } |
d432c80e | 1247 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
1da177e4 LT |
1248 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
1249 | break; | |
d432c80e | 1250 | } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
1da177e4 LT |
1251 | blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL); |
1252 | break; | |
1253 | } | |
1254 | } | |
1255 | ||
1256 | /* | |
1257 | * A leaf block that ends in the hashval that we are interested in | |
1258 | * (final hashval == search hashval) means that the next block may | |
1259 | * contain more entries with the same hashval, shift upward to the | |
1260 | * next leaf and keep searching. | |
1261 | */ | |
1262 | for (;;) { | |
f6c2d1fa | 1263 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
1da177e4 LT |
1264 | retval = xfs_dir2_leafn_lookup_int(blk->bp, args, |
1265 | &blk->index, state); | |
d432c80e | 1266 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
1da177e4 LT |
1267 | retval = xfs_attr_leaf_lookup_int(blk->bp, args); |
1268 | blk->index = args->index; | |
1269 | args->blkno = blk->blkno; | |
d432c80e NS |
1270 | } else { |
1271 | ASSERT(0); | |
1272 | return XFS_ERROR(EFSCORRUPTED); | |
1da177e4 | 1273 | } |
1da177e4 LT |
1274 | if (((retval == ENOENT) || (retval == ENOATTR)) && |
1275 | (blk->hashval == args->hashval)) { | |
1276 | error = xfs_da_path_shift(state, &state->path, 1, 1, | |
1277 | &retval); | |
1278 | if (error) | |
1279 | return(error); | |
1280 | if (retval == 0) { | |
1281 | continue; | |
d432c80e | 1282 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
1da177e4 LT |
1283 | /* path_shift() gives ENOENT */ |
1284 | retval = XFS_ERROR(ENOATTR); | |
1285 | } | |
1da177e4 LT |
1286 | } |
1287 | break; | |
1288 | } | |
1289 | *result = retval; | |
1290 | return(0); | |
1291 | } | |
1292 | ||
1293 | /*======================================================================== | |
1294 | * Utility routines. | |
1295 | *========================================================================*/ | |
1296 | ||
1297 | /* | |
1298 | * Link a new block into a doubly linked list of blocks (of whatever type). | |
1299 | */ | |
1300 | int /* error */ | |
1301 | xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk, | |
1302 | xfs_da_state_blk_t *new_blk) | |
1303 | { | |
1304 | xfs_da_blkinfo_t *old_info, *new_info, *tmp_info; | |
1305 | xfs_da_args_t *args; | |
1306 | int before=0, error; | |
1d9025e5 | 1307 | struct xfs_buf *bp; |
1da177e4 LT |
1308 | |
1309 | /* | |
1310 | * Set up environment. | |
1311 | */ | |
1312 | args = state->args; | |
1313 | ASSERT(args != NULL); | |
1d9025e5 DC |
1314 | old_info = old_blk->bp->b_addr; |
1315 | new_info = new_blk->bp->b_addr; | |
1da177e4 | 1316 | ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC || |
f6c2d1fa | 1317 | old_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
1da177e4 | 1318 | old_blk->magic == XFS_ATTR_LEAF_MAGIC); |
89da0544 NS |
1319 | ASSERT(old_blk->magic == be16_to_cpu(old_info->magic)); |
1320 | ASSERT(new_blk->magic == be16_to_cpu(new_info->magic)); | |
1da177e4 LT |
1321 | ASSERT(old_blk->magic == new_blk->magic); |
1322 | ||
1323 | switch (old_blk->magic) { | |
1da177e4 LT |
1324 | case XFS_ATTR_LEAF_MAGIC: |
1325 | before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp); | |
1326 | break; | |
1da177e4 | 1327 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
1328 | before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp); |
1329 | break; | |
1330 | case XFS_DA_NODE_MAGIC: | |
1331 | before = xfs_da_node_order(old_blk->bp, new_blk->bp); | |
1332 | break; | |
1333 | } | |
1334 | ||
1335 | /* | |
1336 | * Link blocks in appropriate order. | |
1337 | */ | |
1338 | if (before) { | |
1339 | /* | |
1340 | * Link new block in before existing block. | |
1341 | */ | |
5a5881cd | 1342 | trace_xfs_da_link_before(args); |
89da0544 NS |
1343 | new_info->forw = cpu_to_be32(old_blk->blkno); |
1344 | new_info->back = old_info->back; | |
1345 | if (old_info->back) { | |
d9392a4b | 1346 | error = xfs_da_node_read(args->trans, args->dp, |
89da0544 | 1347 | be32_to_cpu(old_info->back), |
d9392a4b | 1348 | -1, &bp, args->whichfork); |
1da177e4 LT |
1349 | if (error) |
1350 | return(error); | |
1351 | ASSERT(bp != NULL); | |
1d9025e5 | 1352 | tmp_info = bp->b_addr; |
89da0544 NS |
1353 | ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic)); |
1354 | ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno); | |
1355 | tmp_info->forw = cpu_to_be32(new_blk->blkno); | |
1d9025e5 | 1356 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
1da177e4 | 1357 | } |
89da0544 | 1358 | old_info->back = cpu_to_be32(new_blk->blkno); |
1da177e4 LT |
1359 | } else { |
1360 | /* | |
1361 | * Link new block in after existing block. | |
1362 | */ | |
5a5881cd | 1363 | trace_xfs_da_link_after(args); |
89da0544 NS |
1364 | new_info->forw = old_info->forw; |
1365 | new_info->back = cpu_to_be32(old_blk->blkno); | |
1366 | if (old_info->forw) { | |
d9392a4b | 1367 | error = xfs_da_node_read(args->trans, args->dp, |
89da0544 | 1368 | be32_to_cpu(old_info->forw), |
d9392a4b | 1369 | -1, &bp, args->whichfork); |
1da177e4 LT |
1370 | if (error) |
1371 | return(error); | |
1372 | ASSERT(bp != NULL); | |
1d9025e5 | 1373 | tmp_info = bp->b_addr; |
89da0544 NS |
1374 | ASSERT(tmp_info->magic == old_info->magic); |
1375 | ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno); | |
1376 | tmp_info->back = cpu_to_be32(new_blk->blkno); | |
1d9025e5 | 1377 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
1da177e4 | 1378 | } |
89da0544 | 1379 | old_info->forw = cpu_to_be32(new_blk->blkno); |
1da177e4 LT |
1380 | } |
1381 | ||
1d9025e5 DC |
1382 | xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1); |
1383 | xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1); | |
1da177e4 LT |
1384 | return(0); |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * Compare two intermediate nodes for "order". | |
1389 | */ | |
1390 | STATIC int | |
1d9025e5 DC |
1391 | xfs_da_node_order( |
1392 | struct xfs_buf *node1_bp, | |
1393 | struct xfs_buf *node2_bp) | |
1da177e4 LT |
1394 | { |
1395 | xfs_da_intnode_t *node1, *node2; | |
1396 | ||
1d9025e5 DC |
1397 | node1 = node1_bp->b_addr; |
1398 | node2 = node2_bp->b_addr; | |
69ef921b CH |
1399 | ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) && |
1400 | node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
fac80cce | 1401 | if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) && |
403432dc NS |
1402 | ((be32_to_cpu(node2->btree[0].hashval) < |
1403 | be32_to_cpu(node1->btree[0].hashval)) || | |
fac80cce NS |
1404 | (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) < |
1405 | be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) { | |
1da177e4 LT |
1406 | return(1); |
1407 | } | |
1408 | return(0); | |
1409 | } | |
1410 | ||
1411 | /* | |
1412 | * Pick up the last hashvalue from an intermediate node. | |
1413 | */ | |
1414 | STATIC uint | |
1d9025e5 DC |
1415 | xfs_da_node_lasthash( |
1416 | struct xfs_buf *bp, | |
1417 | int *count) | |
1da177e4 LT |
1418 | { |
1419 | xfs_da_intnode_t *node; | |
1420 | ||
1d9025e5 | 1421 | node = bp->b_addr; |
69ef921b | 1422 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 | 1423 | if (count) |
fac80cce | 1424 | *count = be16_to_cpu(node->hdr.count); |
1da177e4 LT |
1425 | if (!node->hdr.count) |
1426 | return(0); | |
fac80cce | 1427 | return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
1da177e4 LT |
1428 | } |
1429 | ||
1430 | /* | |
1431 | * Unlink a block from a doubly linked list of blocks. | |
1432 | */ | |
ba0f32d4 | 1433 | STATIC int /* error */ |
1da177e4 LT |
1434 | xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk, |
1435 | xfs_da_state_blk_t *save_blk) | |
1436 | { | |
1437 | xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info; | |
1438 | xfs_da_args_t *args; | |
1d9025e5 | 1439 | struct xfs_buf *bp; |
1da177e4 LT |
1440 | int error; |
1441 | ||
1442 | /* | |
1443 | * Set up environment. | |
1444 | */ | |
1445 | args = state->args; | |
1446 | ASSERT(args != NULL); | |
1d9025e5 DC |
1447 | save_info = save_blk->bp->b_addr; |
1448 | drop_info = drop_blk->bp->b_addr; | |
1da177e4 | 1449 | ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC || |
f6c2d1fa | 1450 | save_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
1da177e4 | 1451 | save_blk->magic == XFS_ATTR_LEAF_MAGIC); |
89da0544 NS |
1452 | ASSERT(save_blk->magic == be16_to_cpu(save_info->magic)); |
1453 | ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic)); | |
1da177e4 | 1454 | ASSERT(save_blk->magic == drop_blk->magic); |
89da0544 NS |
1455 | ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) || |
1456 | (be32_to_cpu(save_info->back) == drop_blk->blkno)); | |
1457 | ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) || | |
1458 | (be32_to_cpu(drop_info->back) == save_blk->blkno)); | |
1da177e4 LT |
1459 | |
1460 | /* | |
1461 | * Unlink the leaf block from the doubly linked chain of leaves. | |
1462 | */ | |
89da0544 | 1463 | if (be32_to_cpu(save_info->back) == drop_blk->blkno) { |
5a5881cd | 1464 | trace_xfs_da_unlink_back(args); |
89da0544 NS |
1465 | save_info->back = drop_info->back; |
1466 | if (drop_info->back) { | |
d9392a4b | 1467 | error = xfs_da_node_read(args->trans, args->dp, |
89da0544 | 1468 | be32_to_cpu(drop_info->back), |
d9392a4b | 1469 | -1, &bp, args->whichfork); |
1da177e4 LT |
1470 | if (error) |
1471 | return(error); | |
1472 | ASSERT(bp != NULL); | |
1d9025e5 | 1473 | tmp_info = bp->b_addr; |
89da0544 NS |
1474 | ASSERT(tmp_info->magic == save_info->magic); |
1475 | ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno); | |
1476 | tmp_info->forw = cpu_to_be32(save_blk->blkno); | |
1d9025e5 | 1477 | xfs_trans_log_buf(args->trans, bp, 0, |
1da177e4 | 1478 | sizeof(*tmp_info) - 1); |
1da177e4 LT |
1479 | } |
1480 | } else { | |
5a5881cd | 1481 | trace_xfs_da_unlink_forward(args); |
89da0544 NS |
1482 | save_info->forw = drop_info->forw; |
1483 | if (drop_info->forw) { | |
d9392a4b | 1484 | error = xfs_da_node_read(args->trans, args->dp, |
89da0544 | 1485 | be32_to_cpu(drop_info->forw), |
d9392a4b | 1486 | -1, &bp, args->whichfork); |
1da177e4 LT |
1487 | if (error) |
1488 | return(error); | |
1489 | ASSERT(bp != NULL); | |
1d9025e5 | 1490 | tmp_info = bp->b_addr; |
89da0544 NS |
1491 | ASSERT(tmp_info->magic == save_info->magic); |
1492 | ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno); | |
1493 | tmp_info->back = cpu_to_be32(save_blk->blkno); | |
1d9025e5 | 1494 | xfs_trans_log_buf(args->trans, bp, 0, |
1da177e4 | 1495 | sizeof(*tmp_info) - 1); |
1da177e4 LT |
1496 | } |
1497 | } | |
1498 | ||
1d9025e5 | 1499 | xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1); |
1da177e4 LT |
1500 | return(0); |
1501 | } | |
1502 | ||
1503 | /* | |
1504 | * Move a path "forward" or "!forward" one block at the current level. | |
1505 | * | |
1506 | * This routine will adjust a "path" to point to the next block | |
1507 | * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the | |
1508 | * Btree, including updating pointers to the intermediate nodes between | |
1509 | * the new bottom and the root. | |
1510 | */ | |
1511 | int /* error */ | |
1512 | xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path, | |
1513 | int forward, int release, int *result) | |
1514 | { | |
1515 | xfs_da_state_blk_t *blk; | |
1516 | xfs_da_blkinfo_t *info; | |
1517 | xfs_da_intnode_t *node; | |
1518 | xfs_da_args_t *args; | |
1519 | xfs_dablk_t blkno=0; | |
1520 | int level, error; | |
1521 | ||
ee73259b DC |
1522 | trace_xfs_da_path_shift(state->args); |
1523 | ||
1da177e4 LT |
1524 | /* |
1525 | * Roll up the Btree looking for the first block where our | |
1526 | * current index is not at the edge of the block. Note that | |
1527 | * we skip the bottom layer because we want the sibling block. | |
1528 | */ | |
1529 | args = state->args; | |
1530 | ASSERT(args != NULL); | |
1531 | ASSERT(path != NULL); | |
1532 | ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH)); | |
1533 | level = (path->active-1) - 1; /* skip bottom layer in path */ | |
1534 | for (blk = &path->blk[level]; level >= 0; blk--, level--) { | |
1535 | ASSERT(blk->bp != NULL); | |
1d9025e5 | 1536 | node = blk->bp->b_addr; |
69ef921b | 1537 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
fac80cce | 1538 | if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) { |
1da177e4 | 1539 | blk->index++; |
403432dc | 1540 | blkno = be32_to_cpu(node->btree[blk->index].before); |
1da177e4 LT |
1541 | break; |
1542 | } else if (!forward && (blk->index > 0)) { | |
1543 | blk->index--; | |
403432dc | 1544 | blkno = be32_to_cpu(node->btree[blk->index].before); |
1da177e4 LT |
1545 | break; |
1546 | } | |
1547 | } | |
1548 | if (level < 0) { | |
1549 | *result = XFS_ERROR(ENOENT); /* we're out of our tree */ | |
6a178100 | 1550 | ASSERT(args->op_flags & XFS_DA_OP_OKNOENT); |
1da177e4 LT |
1551 | return(0); |
1552 | } | |
1553 | ||
1554 | /* | |
1555 | * Roll down the edge of the subtree until we reach the | |
1556 | * same depth we were at originally. | |
1557 | */ | |
1558 | for (blk++, level++; level < path->active; blk++, level++) { | |
1559 | /* | |
1560 | * Release the old block. | |
1561 | * (if it's dirty, trans won't actually let go) | |
1562 | */ | |
1563 | if (release) | |
1d9025e5 | 1564 | xfs_trans_brelse(args->trans, blk->bp); |
1da177e4 LT |
1565 | |
1566 | /* | |
1567 | * Read the next child block. | |
1568 | */ | |
1569 | blk->blkno = blkno; | |
d9392a4b DC |
1570 | error = xfs_da_node_read(args->trans, args->dp, blkno, -1, |
1571 | &blk->bp, args->whichfork); | |
1da177e4 LT |
1572 | if (error) |
1573 | return(error); | |
1574 | ASSERT(blk->bp != NULL); | |
1d9025e5 | 1575 | info = blk->bp->b_addr; |
69ef921b CH |
1576 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
1577 | info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
24df33b4 | 1578 | info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
69ef921b | 1579 | info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)); |
89da0544 NS |
1580 | blk->magic = be16_to_cpu(info->magic); |
1581 | if (blk->magic == XFS_DA_NODE_MAGIC) { | |
1da177e4 | 1582 | node = (xfs_da_intnode_t *)info; |
fac80cce | 1583 | blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
1da177e4 LT |
1584 | if (forward) |
1585 | blk->index = 0; | |
1586 | else | |
fac80cce | 1587 | blk->index = be16_to_cpu(node->hdr.count)-1; |
403432dc | 1588 | blkno = be32_to_cpu(node->btree[blk->index].before); |
1da177e4 LT |
1589 | } else { |
1590 | ASSERT(level == path->active-1); | |
1591 | blk->index = 0; | |
1592 | switch(blk->magic) { | |
1da177e4 LT |
1593 | case XFS_ATTR_LEAF_MAGIC: |
1594 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, | |
1595 | NULL); | |
1596 | break; | |
1da177e4 | 1597 | case XFS_DIR2_LEAFN_MAGIC: |
24df33b4 DC |
1598 | case XFS_DIR3_LEAFN_MAGIC: |
1599 | blk->magic = XFS_DIR2_LEAFN_MAGIC; | |
1da177e4 LT |
1600 | blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, |
1601 | NULL); | |
1602 | break; | |
1603 | default: | |
24df33b4 | 1604 | ASSERT(0); |
1da177e4 LT |
1605 | break; |
1606 | } | |
1607 | } | |
1608 | } | |
1609 | *result = 0; | |
1610 | return(0); | |
1611 | } | |
1612 | ||
1613 | ||
1614 | /*======================================================================== | |
1615 | * Utility routines. | |
1616 | *========================================================================*/ | |
1617 | ||
1618 | /* | |
1619 | * Implement a simple hash on a character string. | |
1620 | * Rotate the hash value by 7 bits, then XOR each character in. | |
1621 | * This is implemented with some source-level loop unrolling. | |
1622 | */ | |
1623 | xfs_dahash_t | |
a5687787 | 1624 | xfs_da_hashname(const __uint8_t *name, int namelen) |
1da177e4 LT |
1625 | { |
1626 | xfs_dahash_t hash; | |
1627 | ||
1da177e4 LT |
1628 | /* |
1629 | * Do four characters at a time as long as we can. | |
1630 | */ | |
1631 | for (hash = 0; namelen >= 4; namelen -= 4, name += 4) | |
1632 | hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^ | |
1633 | (name[3] << 0) ^ rol32(hash, 7 * 4); | |
1634 | ||
1635 | /* | |
1636 | * Now do the rest of the characters. | |
1637 | */ | |
1638 | switch (namelen) { | |
1639 | case 3: | |
1640 | return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^ | |
1641 | rol32(hash, 7 * 3); | |
1642 | case 2: | |
1643 | return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2); | |
1644 | case 1: | |
1645 | return (name[0] << 0) ^ rol32(hash, 7 * 1); | |
2b3b6d07 | 1646 | default: /* case 0: */ |
1da177e4 LT |
1647 | return hash; |
1648 | } | |
1da177e4 LT |
1649 | } |
1650 | ||
5163f95a BN |
1651 | enum xfs_dacmp |
1652 | xfs_da_compname( | |
1653 | struct xfs_da_args *args, | |
2bc75421 DC |
1654 | const unsigned char *name, |
1655 | int len) | |
5163f95a BN |
1656 | { |
1657 | return (args->namelen == len && memcmp(args->name, name, len) == 0) ? | |
1658 | XFS_CMP_EXACT : XFS_CMP_DIFFERENT; | |
1659 | } | |
1660 | ||
1661 | static xfs_dahash_t | |
1662 | xfs_default_hashname( | |
1663 | struct xfs_name *name) | |
1664 | { | |
1665 | return xfs_da_hashname(name->name, name->len); | |
1666 | } | |
1667 | ||
1668 | const struct xfs_nameops xfs_default_nameops = { | |
1669 | .hashname = xfs_default_hashname, | |
1670 | .compname = xfs_da_compname | |
1671 | }; | |
1672 | ||
1da177e4 | 1673 | int |
77936d02 CH |
1674 | xfs_da_grow_inode_int( |
1675 | struct xfs_da_args *args, | |
1676 | xfs_fileoff_t *bno, | |
1677 | int count) | |
1da177e4 | 1678 | { |
77936d02 CH |
1679 | struct xfs_trans *tp = args->trans; |
1680 | struct xfs_inode *dp = args->dp; | |
1681 | int w = args->whichfork; | |
1682 | xfs_drfsbno_t nblks = dp->i_d.di_nblocks; | |
1683 | struct xfs_bmbt_irec map, *mapp; | |
1684 | int nmap, error, got, i, mapi; | |
1da177e4 | 1685 | |
1da177e4 LT |
1686 | /* |
1687 | * Find a spot in the file space to put the new block. | |
1688 | */ | |
77936d02 CH |
1689 | error = xfs_bmap_first_unused(tp, dp, count, bno, w); |
1690 | if (error) | |
1da177e4 | 1691 | return error; |
77936d02 | 1692 | |
1da177e4 LT |
1693 | /* |
1694 | * Try mapping it in one filesystem block. | |
1695 | */ | |
1696 | nmap = 1; | |
1697 | ASSERT(args->firstblock != NULL); | |
c0dc7828 DC |
1698 | error = xfs_bmapi_write(tp, dp, *bno, count, |
1699 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG, | |
1da177e4 | 1700 | args->firstblock, args->total, &map, &nmap, |
77936d02 CH |
1701 | args->flist); |
1702 | if (error) | |
1da177e4 | 1703 | return error; |
77936d02 | 1704 | |
1da177e4 LT |
1705 | ASSERT(nmap <= 1); |
1706 | if (nmap == 1) { | |
1707 | mapp = ↦ | |
1708 | mapi = 1; | |
77936d02 CH |
1709 | } else if (nmap == 0 && count > 1) { |
1710 | xfs_fileoff_t b; | |
1711 | int c; | |
1712 | ||
1713 | /* | |
1714 | * If we didn't get it and the block might work if fragmented, | |
1715 | * try without the CONTIG flag. Loop until we get it all. | |
1716 | */ | |
1da177e4 | 1717 | mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP); |
77936d02 | 1718 | for (b = *bno, mapi = 0; b < *bno + count; ) { |
1da177e4 | 1719 | nmap = MIN(XFS_BMAP_MAX_NMAP, count); |
77936d02 | 1720 | c = (int)(*bno + count - b); |
c0dc7828 DC |
1721 | error = xfs_bmapi_write(tp, dp, b, c, |
1722 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, | |
1da177e4 | 1723 | args->firstblock, args->total, |
77936d02 CH |
1724 | &mapp[mapi], &nmap, args->flist); |
1725 | if (error) | |
1726 | goto out_free_map; | |
1da177e4 LT |
1727 | if (nmap < 1) |
1728 | break; | |
1729 | mapi += nmap; | |
1730 | b = mapp[mapi - 1].br_startoff + | |
1731 | mapp[mapi - 1].br_blockcount; | |
1732 | } | |
1733 | } else { | |
1734 | mapi = 0; | |
1735 | mapp = NULL; | |
1736 | } | |
77936d02 | 1737 | |
1da177e4 LT |
1738 | /* |
1739 | * Count the blocks we got, make sure it matches the total. | |
1740 | */ | |
1741 | for (i = 0, got = 0; i < mapi; i++) | |
1742 | got += mapp[i].br_blockcount; | |
77936d02 | 1743 | if (got != count || mapp[0].br_startoff != *bno || |
1da177e4 | 1744 | mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount != |
77936d02 CH |
1745 | *bno + count) { |
1746 | error = XFS_ERROR(ENOSPC); | |
1747 | goto out_free_map; | |
1da177e4 | 1748 | } |
77936d02 | 1749 | |
a7444053 DC |
1750 | /* account for newly allocated blocks in reserved blocks total */ |
1751 | args->total -= dp->i_d.di_nblocks - nblks; | |
77936d02 CH |
1752 | |
1753 | out_free_map: | |
1754 | if (mapp != &map) | |
1755 | kmem_free(mapp); | |
1756 | return error; | |
1757 | } | |
1758 | ||
1759 | /* | |
1760 | * Add a block to the btree ahead of the file. | |
1761 | * Return the new block number to the caller. | |
1762 | */ | |
1763 | int | |
1764 | xfs_da_grow_inode( | |
1765 | struct xfs_da_args *args, | |
1766 | xfs_dablk_t *new_blkno) | |
1767 | { | |
1768 | xfs_fileoff_t bno; | |
1769 | int count; | |
1770 | int error; | |
1771 | ||
5a5881cd DC |
1772 | trace_xfs_da_grow_inode(args); |
1773 | ||
77936d02 CH |
1774 | if (args->whichfork == XFS_DATA_FORK) { |
1775 | bno = args->dp->i_mount->m_dirleafblk; | |
1776 | count = args->dp->i_mount->m_dirblkfsbs; | |
1777 | } else { | |
1778 | bno = 0; | |
1779 | count = 1; | |
1780 | } | |
1781 | ||
1782 | error = xfs_da_grow_inode_int(args, &bno, count); | |
1783 | if (!error) | |
1784 | *new_blkno = (xfs_dablk_t)bno; | |
1785 | return error; | |
1da177e4 LT |
1786 | } |
1787 | ||
1788 | /* | |
1789 | * Ick. We need to always be able to remove a btree block, even | |
1790 | * if there's no space reservation because the filesystem is full. | |
1791 | * This is called if xfs_bunmapi on a btree block fails due to ENOSPC. | |
1792 | * It swaps the target block with the last block in the file. The | |
1793 | * last block in the file can always be removed since it can't cause | |
1794 | * a bmap btree split to do that. | |
1795 | */ | |
1796 | STATIC int | |
1d9025e5 DC |
1797 | xfs_da_swap_lastblock( |
1798 | xfs_da_args_t *args, | |
1799 | xfs_dablk_t *dead_blknop, | |
1800 | struct xfs_buf **dead_bufp) | |
1da177e4 LT |
1801 | { |
1802 | xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno; | |
1d9025e5 | 1803 | struct xfs_buf *dead_buf, *last_buf, *sib_buf, *par_buf; |
1da177e4 LT |
1804 | xfs_fileoff_t lastoff; |
1805 | xfs_inode_t *ip; | |
1806 | xfs_trans_t *tp; | |
1807 | xfs_mount_t *mp; | |
1808 | int error, w, entno, level, dead_level; | |
1809 | xfs_da_blkinfo_t *dead_info, *sib_info; | |
1810 | xfs_da_intnode_t *par_node, *dead_node; | |
1da177e4 LT |
1811 | xfs_dir2_leaf_t *dead_leaf2; |
1812 | xfs_dahash_t dead_hash; | |
1813 | ||
5a5881cd DC |
1814 | trace_xfs_da_swap_lastblock(args); |
1815 | ||
1da177e4 LT |
1816 | dead_buf = *dead_bufp; |
1817 | dead_blkno = *dead_blknop; | |
1818 | tp = args->trans; | |
1819 | ip = args->dp; | |
1820 | w = args->whichfork; | |
1821 | ASSERT(w == XFS_DATA_FORK); | |
1822 | mp = ip->i_mount; | |
f6c2d1fa NS |
1823 | lastoff = mp->m_dirfreeblk; |
1824 | error = xfs_bmap_last_before(tp, ip, &lastoff, w); | |
1da177e4 LT |
1825 | if (error) |
1826 | return error; | |
1827 | if (unlikely(lastoff == 0)) { | |
1828 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW, | |
1829 | mp); | |
1830 | return XFS_ERROR(EFSCORRUPTED); | |
1831 | } | |
1832 | /* | |
1833 | * Read the last block in the btree space. | |
1834 | */ | |
1835 | last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs; | |
d9392a4b | 1836 | error = xfs_da_node_read(tp, ip, last_blkno, -1, &last_buf, w); |
4bb20a83 | 1837 | if (error) |
1da177e4 LT |
1838 | return error; |
1839 | /* | |
1840 | * Copy the last block into the dead buffer and log it. | |
1841 | */ | |
1d9025e5 DC |
1842 | memcpy(dead_buf->b_addr, last_buf->b_addr, mp->m_dirblksize); |
1843 | xfs_trans_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1); | |
1844 | dead_info = dead_buf->b_addr; | |
1da177e4 LT |
1845 | /* |
1846 | * Get values from the moved block. | |
1847 | */ | |
24df33b4 DC |
1848 | if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
1849 | dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
1850 | struct xfs_dir3_icleaf_hdr leafhdr; | |
1851 | struct xfs_dir2_leaf_entry *ents; | |
1852 | ||
1da177e4 | 1853 | dead_leaf2 = (xfs_dir2_leaf_t *)dead_info; |
24df33b4 DC |
1854 | xfs_dir3_leaf_hdr_from_disk(&leafhdr, dead_leaf2); |
1855 | ents = xfs_dir3_leaf_ents_p(dead_leaf2); | |
1da177e4 | 1856 | dead_level = 0; |
24df33b4 | 1857 | dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval); |
1da177e4 | 1858 | } else { |
69ef921b | 1859 | ASSERT(dead_info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 | 1860 | dead_node = (xfs_da_intnode_t *)dead_info; |
fac80cce NS |
1861 | dead_level = be16_to_cpu(dead_node->hdr.level); |
1862 | dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval); | |
1da177e4 LT |
1863 | } |
1864 | sib_buf = par_buf = NULL; | |
1865 | /* | |
1866 | * If the moved block has a left sibling, fix up the pointers. | |
1867 | */ | |
89da0544 | 1868 | if ((sib_blkno = be32_to_cpu(dead_info->back))) { |
d9392a4b | 1869 | error = xfs_da_node_read(tp, ip, sib_blkno, -1, &sib_buf, w); |
4bb20a83 | 1870 | if (error) |
1da177e4 | 1871 | goto done; |
1d9025e5 | 1872 | sib_info = sib_buf->b_addr; |
1da177e4 | 1873 | if (unlikely( |
89da0544 NS |
1874 | be32_to_cpu(sib_info->forw) != last_blkno || |
1875 | sib_info->magic != dead_info->magic)) { | |
1da177e4 LT |
1876 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)", |
1877 | XFS_ERRLEVEL_LOW, mp); | |
1878 | error = XFS_ERROR(EFSCORRUPTED); | |
1879 | goto done; | |
1880 | } | |
89da0544 | 1881 | sib_info->forw = cpu_to_be32(dead_blkno); |
1d9025e5 | 1882 | xfs_trans_log_buf(tp, sib_buf, |
1da177e4 LT |
1883 | XFS_DA_LOGRANGE(sib_info, &sib_info->forw, |
1884 | sizeof(sib_info->forw))); | |
1da177e4 LT |
1885 | sib_buf = NULL; |
1886 | } | |
1887 | /* | |
1888 | * If the moved block has a right sibling, fix up the pointers. | |
1889 | */ | |
89da0544 | 1890 | if ((sib_blkno = be32_to_cpu(dead_info->forw))) { |
d9392a4b | 1891 | error = xfs_da_node_read(tp, ip, sib_blkno, -1, &sib_buf, w); |
4bb20a83 | 1892 | if (error) |
1da177e4 | 1893 | goto done; |
1d9025e5 | 1894 | sib_info = sib_buf->b_addr; |
1da177e4 | 1895 | if (unlikely( |
89da0544 NS |
1896 | be32_to_cpu(sib_info->back) != last_blkno || |
1897 | sib_info->magic != dead_info->magic)) { | |
1da177e4 LT |
1898 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)", |
1899 | XFS_ERRLEVEL_LOW, mp); | |
1900 | error = XFS_ERROR(EFSCORRUPTED); | |
1901 | goto done; | |
1902 | } | |
89da0544 | 1903 | sib_info->back = cpu_to_be32(dead_blkno); |
1d9025e5 | 1904 | xfs_trans_log_buf(tp, sib_buf, |
1da177e4 LT |
1905 | XFS_DA_LOGRANGE(sib_info, &sib_info->back, |
1906 | sizeof(sib_info->back))); | |
1da177e4 LT |
1907 | sib_buf = NULL; |
1908 | } | |
f6c2d1fa | 1909 | par_blkno = mp->m_dirleafblk; |
1da177e4 LT |
1910 | level = -1; |
1911 | /* | |
1912 | * Walk down the tree looking for the parent of the moved block. | |
1913 | */ | |
1914 | for (;;) { | |
d9392a4b | 1915 | error = xfs_da_node_read(tp, ip, par_blkno, -1, &par_buf, w); |
4bb20a83 | 1916 | if (error) |
1da177e4 | 1917 | goto done; |
1d9025e5 | 1918 | par_node = par_buf->b_addr; |
69ef921b CH |
1919 | if (unlikely(par_node->hdr.info.magic != |
1920 | cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
fac80cce | 1921 | (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) { |
1da177e4 LT |
1922 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)", |
1923 | XFS_ERRLEVEL_LOW, mp); | |
1924 | error = XFS_ERROR(EFSCORRUPTED); | |
1925 | goto done; | |
1926 | } | |
fac80cce | 1927 | level = be16_to_cpu(par_node->hdr.level); |
1da177e4 | 1928 | for (entno = 0; |
fac80cce | 1929 | entno < be16_to_cpu(par_node->hdr.count) && |
403432dc | 1930 | be32_to_cpu(par_node->btree[entno].hashval) < dead_hash; |
1da177e4 LT |
1931 | entno++) |
1932 | continue; | |
fac80cce | 1933 | if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) { |
1da177e4 LT |
1934 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)", |
1935 | XFS_ERRLEVEL_LOW, mp); | |
1936 | error = XFS_ERROR(EFSCORRUPTED); | |
1937 | goto done; | |
1938 | } | |
403432dc | 1939 | par_blkno = be32_to_cpu(par_node->btree[entno].before); |
1da177e4 LT |
1940 | if (level == dead_level + 1) |
1941 | break; | |
1d9025e5 | 1942 | xfs_trans_brelse(tp, par_buf); |
1da177e4 LT |
1943 | par_buf = NULL; |
1944 | } | |
1945 | /* | |
1946 | * We're in the right parent block. | |
1947 | * Look for the right entry. | |
1948 | */ | |
1949 | for (;;) { | |
1950 | for (; | |
fac80cce | 1951 | entno < be16_to_cpu(par_node->hdr.count) && |
403432dc | 1952 | be32_to_cpu(par_node->btree[entno].before) != last_blkno; |
1da177e4 LT |
1953 | entno++) |
1954 | continue; | |
fac80cce | 1955 | if (entno < be16_to_cpu(par_node->hdr.count)) |
1da177e4 | 1956 | break; |
89da0544 | 1957 | par_blkno = be32_to_cpu(par_node->hdr.info.forw); |
1d9025e5 | 1958 | xfs_trans_brelse(tp, par_buf); |
1da177e4 LT |
1959 | par_buf = NULL; |
1960 | if (unlikely(par_blkno == 0)) { | |
1961 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)", | |
1962 | XFS_ERRLEVEL_LOW, mp); | |
1963 | error = XFS_ERROR(EFSCORRUPTED); | |
1964 | goto done; | |
1965 | } | |
d9392a4b | 1966 | error = xfs_da_node_read(tp, ip, par_blkno, -1, &par_buf, w); |
4bb20a83 | 1967 | if (error) |
1da177e4 | 1968 | goto done; |
1d9025e5 | 1969 | par_node = par_buf->b_addr; |
1da177e4 | 1970 | if (unlikely( |
fac80cce | 1971 | be16_to_cpu(par_node->hdr.level) != level || |
69ef921b | 1972 | par_node->hdr.info.magic != cpu_to_be16(XFS_DA_NODE_MAGIC))) { |
1da177e4 LT |
1973 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)", |
1974 | XFS_ERRLEVEL_LOW, mp); | |
1975 | error = XFS_ERROR(EFSCORRUPTED); | |
1976 | goto done; | |
1977 | } | |
1978 | entno = 0; | |
1979 | } | |
1980 | /* | |
1981 | * Update the parent entry pointing to the moved block. | |
1982 | */ | |
403432dc | 1983 | par_node->btree[entno].before = cpu_to_be32(dead_blkno); |
1d9025e5 | 1984 | xfs_trans_log_buf(tp, par_buf, |
1da177e4 LT |
1985 | XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before, |
1986 | sizeof(par_node->btree[entno].before))); | |
1da177e4 LT |
1987 | *dead_blknop = last_blkno; |
1988 | *dead_bufp = last_buf; | |
1989 | return 0; | |
1990 | done: | |
1991 | if (par_buf) | |
1d9025e5 | 1992 | xfs_trans_brelse(tp, par_buf); |
1da177e4 | 1993 | if (sib_buf) |
1d9025e5 DC |
1994 | xfs_trans_brelse(tp, sib_buf); |
1995 | xfs_trans_brelse(tp, last_buf); | |
1da177e4 LT |
1996 | return error; |
1997 | } | |
1998 | ||
1999 | /* | |
2000 | * Remove a btree block from a directory or attribute. | |
2001 | */ | |
2002 | int | |
1d9025e5 DC |
2003 | xfs_da_shrink_inode( |
2004 | xfs_da_args_t *args, | |
2005 | xfs_dablk_t dead_blkno, | |
2006 | struct xfs_buf *dead_buf) | |
1da177e4 LT |
2007 | { |
2008 | xfs_inode_t *dp; | |
2009 | int done, error, w, count; | |
1da177e4 LT |
2010 | xfs_trans_t *tp; |
2011 | xfs_mount_t *mp; | |
2012 | ||
5a5881cd DC |
2013 | trace_xfs_da_shrink_inode(args); |
2014 | ||
1da177e4 LT |
2015 | dp = args->dp; |
2016 | w = args->whichfork; | |
2017 | tp = args->trans; | |
2018 | mp = dp->i_mount; | |
f6c2d1fa | 2019 | if (w == XFS_DATA_FORK) |
1da177e4 LT |
2020 | count = mp->m_dirblkfsbs; |
2021 | else | |
2022 | count = 1; | |
2023 | for (;;) { | |
2024 | /* | |
2025 | * Remove extents. If we get ENOSPC for a dir we have to move | |
2026 | * the last block to the place we want to kill. | |
2027 | */ | |
2028 | if ((error = xfs_bunmapi(tp, dp, dead_blkno, count, | |
9d87c319 | 2029 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, |
b4e9181e | 2030 | 0, args->firstblock, args->flist, |
1da177e4 LT |
2031 | &done)) == ENOSPC) { |
2032 | if (w != XFS_DATA_FORK) | |
f6c2d1fa | 2033 | break; |
1da177e4 LT |
2034 | if ((error = xfs_da_swap_lastblock(args, &dead_blkno, |
2035 | &dead_buf))) | |
f6c2d1fa NS |
2036 | break; |
2037 | } else { | |
1da177e4 | 2038 | break; |
1da177e4 LT |
2039 | } |
2040 | } | |
1d9025e5 | 2041 | xfs_trans_binval(tp, dead_buf); |
1da177e4 LT |
2042 | return error; |
2043 | } | |
2044 | ||
2045 | /* | |
2046 | * See if the mapping(s) for this btree block are valid, i.e. | |
2047 | * don't contain holes, are logically contiguous, and cover the whole range. | |
2048 | */ | |
2049 | STATIC int | |
2050 | xfs_da_map_covers_blocks( | |
2051 | int nmap, | |
2052 | xfs_bmbt_irec_t *mapp, | |
2053 | xfs_dablk_t bno, | |
2054 | int count) | |
2055 | { | |
2056 | int i; | |
2057 | xfs_fileoff_t off; | |
2058 | ||
2059 | for (i = 0, off = bno; i < nmap; i++) { | |
2060 | if (mapp[i].br_startblock == HOLESTARTBLOCK || | |
2061 | mapp[i].br_startblock == DELAYSTARTBLOCK) { | |
2062 | return 0; | |
2063 | } | |
2064 | if (off != mapp[i].br_startoff) { | |
2065 | return 0; | |
2066 | } | |
2067 | off += mapp[i].br_blockcount; | |
2068 | } | |
2069 | return off == bno + count; | |
2070 | } | |
2071 | ||
2072 | /* | |
3605431f DC |
2073 | * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map. |
2074 | * | |
2075 | * For the single map case, it is assumed that the caller has provided a pointer | |
2076 | * to a valid xfs_buf_map. For the multiple map case, this function will | |
2077 | * allocate the xfs_buf_map to hold all the maps and replace the caller's single | |
2078 | * map pointer with the allocated map. | |
1da177e4 | 2079 | */ |
3605431f DC |
2080 | static int |
2081 | xfs_buf_map_from_irec( | |
2082 | struct xfs_mount *mp, | |
2083 | struct xfs_buf_map **mapp, | |
2084 | unsigned int *nmaps, | |
2085 | struct xfs_bmbt_irec *irecs, | |
2086 | unsigned int nirecs) | |
1da177e4 | 2087 | { |
3605431f DC |
2088 | struct xfs_buf_map *map; |
2089 | int i; | |
2090 | ||
2091 | ASSERT(*nmaps == 1); | |
2092 | ASSERT(nirecs >= 1); | |
2093 | ||
2094 | if (nirecs > 1) { | |
2095 | map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), KM_SLEEP); | |
2096 | if (!map) | |
2097 | return ENOMEM; | |
2098 | *mapp = map; | |
2099 | } | |
2100 | ||
2101 | *nmaps = nirecs; | |
2102 | map = *mapp; | |
2103 | for (i = 0; i < *nmaps; i++) { | |
2104 | ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK && | |
2105 | irecs[i].br_startblock != HOLESTARTBLOCK); | |
2106 | map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock); | |
2107 | map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount); | |
2108 | } | |
2109 | return 0; | |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * Map the block we are given ready for reading. There are three possible return | |
2114 | * values: | |
2115 | * -1 - will be returned if we land in a hole and mappedbno == -2 so the | |
2116 | * caller knows not to execute a subsequent read. | |
2117 | * 0 - if we mapped the block successfully | |
2118 | * >0 - positive error number if there was an error. | |
2119 | */ | |
2120 | static int | |
2121 | xfs_dabuf_map( | |
2122 | struct xfs_trans *trans, | |
2123 | struct xfs_inode *dp, | |
2124 | xfs_dablk_t bno, | |
2125 | xfs_daddr_t mappedbno, | |
2126 | int whichfork, | |
2127 | struct xfs_buf_map **map, | |
2128 | int *nmaps) | |
2129 | { | |
2130 | struct xfs_mount *mp = dp->i_mount; | |
2131 | int nfsb; | |
2132 | int error = 0; | |
2133 | struct xfs_bmbt_irec irec; | |
2134 | struct xfs_bmbt_irec *irecs = &irec; | |
2135 | int nirecs; | |
2136 | ||
2137 | ASSERT(map && *map); | |
2138 | ASSERT(*nmaps == 1); | |
1da177e4 | 2139 | |
f6c2d1fa | 2140 | nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1; |
3605431f | 2141 | |
1da177e4 LT |
2142 | /* |
2143 | * Caller doesn't have a mapping. -2 means don't complain | |
2144 | * if we land in a hole. | |
2145 | */ | |
2146 | if (mappedbno == -1 || mappedbno == -2) { | |
2147 | /* | |
2148 | * Optimize the one-block case. | |
2149 | */ | |
3605431f DC |
2150 | if (nfsb != 1) |
2151 | irecs = kmem_zalloc(sizeof(irec) * nfsb, KM_SLEEP); | |
5b777ad5 | 2152 | |
3605431f DC |
2153 | nirecs = nfsb; |
2154 | error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs, | |
2155 | &nirecs, xfs_bmapi_aflag(whichfork)); | |
5b777ad5 | 2156 | if (error) |
3605431f | 2157 | goto out; |
1da177e4 | 2158 | } else { |
3605431f DC |
2159 | irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno); |
2160 | irecs->br_startoff = (xfs_fileoff_t)bno; | |
2161 | irecs->br_blockcount = nfsb; | |
2162 | irecs->br_state = 0; | |
2163 | nirecs = 1; | |
1da177e4 | 2164 | } |
3605431f DC |
2165 | |
2166 | if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) { | |
2167 | error = mappedbno == -2 ? -1 : XFS_ERROR(EFSCORRUPTED); | |
1da177e4 LT |
2168 | if (unlikely(error == EFSCORRUPTED)) { |
2169 | if (xfs_error_level >= XFS_ERRLEVEL_LOW) { | |
3605431f | 2170 | int i; |
0b932ccc DC |
2171 | xfs_alert(mp, "%s: bno %lld dir: inode %lld", |
2172 | __func__, (long long)bno, | |
1da177e4 | 2173 | (long long)dp->i_ino); |
3605431f | 2174 | for (i = 0; i < *nmaps; i++) { |
0b932ccc DC |
2175 | xfs_alert(mp, |
2176 | "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d", | |
1da177e4 | 2177 | i, |
3605431f DC |
2178 | (long long)irecs[i].br_startoff, |
2179 | (long long)irecs[i].br_startblock, | |
2180 | (long long)irecs[i].br_blockcount, | |
2181 | irecs[i].br_state); | |
1da177e4 LT |
2182 | } |
2183 | } | |
2184 | XFS_ERROR_REPORT("xfs_da_do_buf(1)", | |
2185 | XFS_ERRLEVEL_LOW, mp); | |
2186 | } | |
3605431f | 2187 | goto out; |
1da177e4 | 2188 | } |
3605431f DC |
2189 | error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs); |
2190 | out: | |
2191 | if (irecs != &irec) | |
2192 | kmem_free(irecs); | |
2193 | return error; | |
2194 | } | |
2195 | ||
2196 | /* | |
2197 | * Get a buffer for the dir/attr block. | |
2198 | */ | |
2199 | int | |
2200 | xfs_da_get_buf( | |
2201 | struct xfs_trans *trans, | |
2202 | struct xfs_inode *dp, | |
2203 | xfs_dablk_t bno, | |
2204 | xfs_daddr_t mappedbno, | |
1d9025e5 | 2205 | struct xfs_buf **bpp, |
3605431f DC |
2206 | int whichfork) |
2207 | { | |
2208 | struct xfs_buf *bp; | |
2209 | struct xfs_buf_map map; | |
2210 | struct xfs_buf_map *mapp; | |
2211 | int nmap; | |
2212 | int error; | |
2213 | ||
2214 | *bpp = NULL; | |
2215 | mapp = ↦ | |
2216 | nmap = 1; | |
2217 | error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork, | |
2218 | &mapp, &nmap); | |
2219 | if (error) { | |
2220 | /* mapping a hole is not an error, but we don't continue */ | |
2221 | if (error == -1) | |
1da177e4 | 2222 | error = 0; |
3605431f | 2223 | goto out_free; |
1da177e4 | 2224 | } |
3605431f DC |
2225 | |
2226 | bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp, | |
2227 | mapp, nmap, 0); | |
2228 | error = bp ? bp->b_error : XFS_ERROR(EIO); | |
2229 | if (error) { | |
2230 | xfs_trans_brelse(trans, bp); | |
2231 | goto out_free; | |
2232 | } | |
2233 | ||
1d9025e5 | 2234 | *bpp = bp; |
3605431f DC |
2235 | |
2236 | out_free: | |
2237 | if (mapp != &map) | |
2238 | kmem_free(mapp); | |
2239 | ||
2240 | return error; | |
2241 | } | |
2242 | ||
2243 | /* | |
2244 | * Get a buffer for the dir/attr block, fill in the contents. | |
2245 | */ | |
2246 | int | |
2247 | xfs_da_read_buf( | |
2248 | struct xfs_trans *trans, | |
2249 | struct xfs_inode *dp, | |
2250 | xfs_dablk_t bno, | |
2251 | xfs_daddr_t mappedbno, | |
1d9025e5 | 2252 | struct xfs_buf **bpp, |
4bb20a83 | 2253 | int whichfork, |
1813dd64 | 2254 | const struct xfs_buf_ops *ops) |
3605431f DC |
2255 | { |
2256 | struct xfs_buf *bp; | |
2257 | struct xfs_buf_map map; | |
2258 | struct xfs_buf_map *mapp; | |
2259 | int nmap; | |
2260 | int error; | |
2261 | ||
2262 | *bpp = NULL; | |
2263 | mapp = ↦ | |
2264 | nmap = 1; | |
2265 | error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork, | |
2266 | &mapp, &nmap); | |
2267 | if (error) { | |
2268 | /* mapping a hole is not an error, but we don't continue */ | |
2269 | if (error == -1) | |
2270 | error = 0; | |
2271 | goto out_free; | |
2272 | } | |
2273 | ||
2274 | error = xfs_trans_read_buf_map(dp->i_mount, trans, | |
2275 | dp->i_mount->m_ddev_targp, | |
1813dd64 | 2276 | mapp, nmap, 0, &bp, ops); |
3605431f DC |
2277 | if (error) |
2278 | goto out_free; | |
2279 | ||
2280 | if (whichfork == XFS_ATTR_FORK) | |
2281 | xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF); | |
1da177e4 | 2282 | else |
3605431f DC |
2283 | xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF); |
2284 | ||
1da177e4 | 2285 | /* |
3605431f DC |
2286 | * This verification code will be moved to a CRC verification callback |
2287 | * function so just leave it here unchanged until then. | |
1da177e4 | 2288 | */ |
3605431f | 2289 | { |
1d9025e5 DC |
2290 | xfs_dir2_data_hdr_t *hdr = bp->b_addr; |
2291 | xfs_dir2_free_t *free = bp->b_addr; | |
2292 | xfs_da_blkinfo_t *info = bp->b_addr; | |
1da177e4 | 2293 | uint magic, magic1; |
3605431f | 2294 | struct xfs_mount *mp = dp->i_mount; |
1da177e4 | 2295 | |
89da0544 | 2296 | magic = be16_to_cpu(info->magic); |
c2066e26 | 2297 | magic1 = be32_to_cpu(hdr->magic); |
1da177e4 LT |
2298 | if (unlikely( |
2299 | XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) && | |
1da177e4 LT |
2300 | (magic != XFS_ATTR_LEAF_MAGIC) && |
2301 | (magic != XFS_DIR2_LEAF1_MAGIC) && | |
24df33b4 | 2302 | (magic != XFS_DIR3_LEAF1_MAGIC) && |
1da177e4 | 2303 | (magic != XFS_DIR2_LEAFN_MAGIC) && |
24df33b4 | 2304 | (magic != XFS_DIR3_LEAFN_MAGIC) && |
1da177e4 | 2305 | (magic1 != XFS_DIR2_BLOCK_MAGIC) && |
24df33b4 | 2306 | (magic1 != XFS_DIR3_BLOCK_MAGIC) && |
1da177e4 | 2307 | (magic1 != XFS_DIR2_DATA_MAGIC) && |
24df33b4 DC |
2308 | (magic1 != XFS_DIR3_DATA_MAGIC) && |
2309 | (free->hdr.magic != | |
2310 | cpu_to_be32(XFS_DIR2_FREE_MAGIC)) && | |
2311 | (free->hdr.magic != | |
2312 | cpu_to_be32(XFS_DIR3_FREE_MAGIC)), | |
1da177e4 LT |
2313 | mp, XFS_ERRTAG_DA_READ_BUF, |
2314 | XFS_RANDOM_DA_READ_BUF))) { | |
3605431f | 2315 | trace_xfs_da_btree_corrupt(bp, _RET_IP_); |
1da177e4 LT |
2316 | XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)", |
2317 | XFS_ERRLEVEL_LOW, mp, info); | |
2318 | error = XFS_ERROR(EFSCORRUPTED); | |
1d9025e5 | 2319 | xfs_trans_brelse(trans, bp); |
3605431f | 2320 | goto out_free; |
1da177e4 LT |
2321 | } |
2322 | } | |
1d9025e5 | 2323 | *bpp = bp; |
3605431f | 2324 | out_free: |
1da177e4 | 2325 | if (mapp != &map) |
f0e2d93c | 2326 | kmem_free(mapp); |
1da177e4 | 2327 | |
3605431f | 2328 | return error; |
1da177e4 LT |
2329 | } |
2330 | ||
2331 | /* | |
2332 | * Readahead the dir/attr block. | |
2333 | */ | |
2334 | xfs_daddr_t | |
2335 | xfs_da_reada_buf( | |
3605431f DC |
2336 | struct xfs_trans *trans, |
2337 | struct xfs_inode *dp, | |
2338 | xfs_dablk_t bno, | |
da6958c8 | 2339 | xfs_daddr_t mappedbno, |
4bb20a83 | 2340 | int whichfork, |
1813dd64 | 2341 | const struct xfs_buf_ops *ops) |
1da177e4 | 2342 | { |
3605431f DC |
2343 | struct xfs_buf_map map; |
2344 | struct xfs_buf_map *mapp; | |
2345 | int nmap; | |
2346 | int error; | |
2347 | ||
2348 | mapp = ↦ | |
2349 | nmap = 1; | |
da6958c8 | 2350 | error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork, |
3605431f DC |
2351 | &mapp, &nmap); |
2352 | if (error) { | |
2353 | /* mapping a hole is not an error, but we don't continue */ | |
2354 | if (error == -1) | |
2355 | error = 0; | |
2356 | goto out_free; | |
2357 | } | |
1da177e4 | 2358 | |
3605431f | 2359 | mappedbno = mapp[0].bm_bn; |
1813dd64 | 2360 | xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops); |
3605431f DC |
2361 | |
2362 | out_free: | |
2363 | if (mapp != &map) | |
2364 | kmem_free(mapp); | |
2365 | ||
2366 | if (error) | |
1da177e4 | 2367 | return -1; |
3605431f | 2368 | return mappedbno; |
1da177e4 LT |
2369 | } |
2370 | ||
1da177e4 | 2371 | kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */ |
1da177e4 LT |
2372 | |
2373 | /* | |
2374 | * Allocate a dir-state structure. | |
2375 | * We don't put them on the stack since they're large. | |
2376 | */ | |
2377 | xfs_da_state_t * | |
2378 | xfs_da_state_alloc(void) | |
2379 | { | |
f41d7fb9 | 2380 | return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS); |
1da177e4 LT |
2381 | } |
2382 | ||
2383 | /* | |
2384 | * Kill the altpath contents of a da-state structure. | |
2385 | */ | |
ba0f32d4 | 2386 | STATIC void |
1da177e4 LT |
2387 | xfs_da_state_kill_altpath(xfs_da_state_t *state) |
2388 | { | |
2389 | int i; | |
2390 | ||
1d9025e5 DC |
2391 | for (i = 0; i < state->altpath.active; i++) |
2392 | state->altpath.blk[i].bp = NULL; | |
1da177e4 LT |
2393 | state->altpath.active = 0; |
2394 | } | |
2395 | ||
2396 | /* | |
2397 | * Free a da-state structure. | |
2398 | */ | |
2399 | void | |
2400 | xfs_da_state_free(xfs_da_state_t *state) | |
2401 | { | |
1da177e4 | 2402 | xfs_da_state_kill_altpath(state); |
1da177e4 LT |
2403 | #ifdef DEBUG |
2404 | memset((char *)state, 0, sizeof(*state)); | |
2405 | #endif /* DEBUG */ | |
2406 | kmem_zone_free(xfs_da_state_zone, state); | |
2407 | } |