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 LT |
93 | |
94 | /*======================================================================== | |
95 | * Routines used for growing the Btree. | |
96 | *========================================================================*/ | |
97 | ||
98 | /* | |
99 | * Create the initial contents of an intermediate node. | |
100 | */ | |
101 | int | |
102 | xfs_da_node_create(xfs_da_args_t *args, xfs_dablk_t blkno, int level, | |
1d9025e5 | 103 | struct xfs_buf **bpp, int whichfork) |
1da177e4 LT |
104 | { |
105 | xfs_da_intnode_t *node; | |
1d9025e5 | 106 | struct xfs_buf *bp; |
1da177e4 LT |
107 | int error; |
108 | xfs_trans_t *tp; | |
109 | ||
5a5881cd DC |
110 | trace_xfs_da_node_create(args); |
111 | ||
1da177e4 LT |
112 | tp = args->trans; |
113 | error = xfs_da_get_buf(tp, args->dp, blkno, -1, &bp, whichfork); | |
114 | if (error) | |
115 | return(error); | |
116 | ASSERT(bp != NULL); | |
1d9025e5 | 117 | node = bp->b_addr; |
1da177e4 LT |
118 | node->hdr.info.forw = 0; |
119 | node->hdr.info.back = 0; | |
89da0544 | 120 | node->hdr.info.magic = cpu_to_be16(XFS_DA_NODE_MAGIC); |
1da177e4 LT |
121 | node->hdr.info.pad = 0; |
122 | node->hdr.count = 0; | |
fac80cce | 123 | node->hdr.level = cpu_to_be16(level); |
1da177e4 | 124 | |
1d9025e5 | 125 | xfs_trans_log_buf(tp, bp, |
1da177e4 LT |
126 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
127 | ||
128 | *bpp = bp; | |
129 | return(0); | |
130 | } | |
131 | ||
132 | /* | |
133 | * Split a leaf node, rebalance, then possibly split | |
134 | * intermediate nodes, rebalance, etc. | |
135 | */ | |
136 | int /* error */ | |
137 | xfs_da_split(xfs_da_state_t *state) | |
138 | { | |
139 | xfs_da_state_blk_t *oldblk, *newblk, *addblk; | |
140 | xfs_da_intnode_t *node; | |
1d9025e5 | 141 | struct xfs_buf *bp; |
1da177e4 LT |
142 | int max, action, error, i; |
143 | ||
5a5881cd DC |
144 | trace_xfs_da_split(state->args); |
145 | ||
1da177e4 LT |
146 | /* |
147 | * Walk back up the tree splitting/inserting/adjusting as necessary. | |
148 | * If we need to insert and there isn't room, split the node, then | |
149 | * decide which fragment to insert the new block from below into. | |
150 | * Note that we may split the root this way, but we need more fixup. | |
151 | */ | |
152 | max = state->path.active - 1; | |
153 | ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH)); | |
154 | ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC || | |
f6c2d1fa | 155 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); |
1da177e4 LT |
156 | |
157 | addblk = &state->path.blk[max]; /* initial dummy value */ | |
158 | for (i = max; (i >= 0) && addblk; state->path.active--, i--) { | |
159 | oldblk = &state->path.blk[i]; | |
160 | newblk = &state->altpath.blk[i]; | |
161 | ||
162 | /* | |
163 | * If a leaf node then | |
164 | * Allocate a new leaf node, then rebalance across them. | |
165 | * else if an intermediate node then | |
166 | * We split on the last layer, must we split the node? | |
167 | */ | |
168 | switch (oldblk->magic) { | |
169 | case XFS_ATTR_LEAF_MAGIC: | |
1da177e4 LT |
170 | error = xfs_attr_leaf_split(state, oldblk, newblk); |
171 | if ((error != 0) && (error != ENOSPC)) { | |
172 | return(error); /* GROT: attr is inconsistent */ | |
173 | } | |
174 | if (!error) { | |
175 | addblk = newblk; | |
176 | break; | |
177 | } | |
178 | /* | |
179 | * Entry wouldn't fit, split the leaf again. | |
180 | */ | |
181 | state->extravalid = 1; | |
182 | if (state->inleaf) { | |
183 | state->extraafter = 0; /* before newblk */ | |
5a5881cd | 184 | trace_xfs_attr_leaf_split_before(state->args); |
1da177e4 LT |
185 | error = xfs_attr_leaf_split(state, oldblk, |
186 | &state->extrablk); | |
187 | } else { | |
188 | state->extraafter = 1; /* after newblk */ | |
5a5881cd | 189 | trace_xfs_attr_leaf_split_after(state->args); |
1da177e4 LT |
190 | error = xfs_attr_leaf_split(state, newblk, |
191 | &state->extrablk); | |
192 | } | |
193 | if (error) | |
194 | return(error); /* GROT: attr inconsistent */ | |
195 | addblk = newblk; | |
196 | break; | |
1da177e4 | 197 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
198 | error = xfs_dir2_leafn_split(state, oldblk, newblk); |
199 | if (error) | |
200 | return error; | |
201 | addblk = newblk; | |
202 | break; | |
203 | case XFS_DA_NODE_MAGIC: | |
204 | error = xfs_da_node_split(state, oldblk, newblk, addblk, | |
205 | max - i, &action); | |
1da177e4 LT |
206 | addblk->bp = NULL; |
207 | if (error) | |
208 | return(error); /* GROT: dir is inconsistent */ | |
209 | /* | |
210 | * Record the newly split block for the next time thru? | |
211 | */ | |
212 | if (action) | |
213 | addblk = newblk; | |
214 | else | |
215 | addblk = NULL; | |
216 | break; | |
217 | } | |
218 | ||
219 | /* | |
220 | * Update the btree to show the new hashval for this child. | |
221 | */ | |
222 | xfs_da_fixhashpath(state, &state->path); | |
1da177e4 LT |
223 | } |
224 | if (!addblk) | |
225 | return(0); | |
226 | ||
227 | /* | |
228 | * Split the root node. | |
229 | */ | |
230 | ASSERT(state->path.active == 0); | |
231 | oldblk = &state->path.blk[0]; | |
232 | error = xfs_da_root_split(state, oldblk, addblk); | |
233 | if (error) { | |
1da177e4 LT |
234 | addblk->bp = NULL; |
235 | return(error); /* GROT: dir is inconsistent */ | |
236 | } | |
237 | ||
238 | /* | |
239 | * Update pointers to the node which used to be block 0 and | |
240 | * just got bumped because of the addition of a new root node. | |
241 | * There might be three blocks involved if a double split occurred, | |
242 | * and the original block 0 could be at any position in the list. | |
243 | */ | |
244 | ||
1d9025e5 | 245 | node = oldblk->bp->b_addr; |
1da177e4 | 246 | if (node->hdr.info.forw) { |
89da0544 | 247 | if (be32_to_cpu(node->hdr.info.forw) == addblk->blkno) { |
1da177e4 LT |
248 | bp = addblk->bp; |
249 | } else { | |
250 | ASSERT(state->extravalid); | |
251 | bp = state->extrablk.bp; | |
252 | } | |
1d9025e5 | 253 | node = bp->b_addr; |
89da0544 | 254 | node->hdr.info.back = cpu_to_be32(oldblk->blkno); |
1d9025e5 | 255 | xfs_trans_log_buf(state->args->trans, bp, |
1da177e4 LT |
256 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
257 | sizeof(node->hdr.info))); | |
258 | } | |
1d9025e5 | 259 | node = oldblk->bp->b_addr; |
89da0544 NS |
260 | if (node->hdr.info.back) { |
261 | if (be32_to_cpu(node->hdr.info.back) == addblk->blkno) { | |
1da177e4 LT |
262 | bp = addblk->bp; |
263 | } else { | |
264 | ASSERT(state->extravalid); | |
265 | bp = state->extrablk.bp; | |
266 | } | |
1d9025e5 | 267 | node = bp->b_addr; |
89da0544 | 268 | node->hdr.info.forw = cpu_to_be32(oldblk->blkno); |
1d9025e5 | 269 | xfs_trans_log_buf(state->args->trans, bp, |
1da177e4 LT |
270 | XFS_DA_LOGRANGE(node, &node->hdr.info, |
271 | sizeof(node->hdr.info))); | |
272 | } | |
1da177e4 LT |
273 | addblk->bp = NULL; |
274 | return(0); | |
275 | } | |
276 | ||
277 | /* | |
278 | * Split the root. We have to create a new root and point to the two | |
279 | * parts (the split old root) that we just created. Copy block zero to | |
280 | * the EOF, extending the inode in process. | |
281 | */ | |
282 | STATIC int /* error */ | |
283 | xfs_da_root_split(xfs_da_state_t *state, xfs_da_state_blk_t *blk1, | |
284 | xfs_da_state_blk_t *blk2) | |
285 | { | |
286 | xfs_da_intnode_t *node, *oldroot; | |
287 | xfs_da_args_t *args; | |
288 | xfs_dablk_t blkno; | |
1d9025e5 | 289 | struct xfs_buf *bp; |
1da177e4 LT |
290 | int error, size; |
291 | xfs_inode_t *dp; | |
292 | xfs_trans_t *tp; | |
293 | xfs_mount_t *mp; | |
294 | xfs_dir2_leaf_t *leaf; | |
295 | ||
5a5881cd DC |
296 | trace_xfs_da_root_split(state->args); |
297 | ||
1da177e4 LT |
298 | /* |
299 | * Copy the existing (incorrect) block from the root node position | |
300 | * to a free space somewhere. | |
301 | */ | |
302 | args = state->args; | |
303 | ASSERT(args != NULL); | |
304 | error = xfs_da_grow_inode(args, &blkno); | |
305 | if (error) | |
306 | return(error); | |
307 | dp = args->dp; | |
308 | tp = args->trans; | |
309 | mp = state->mp; | |
310 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork); | |
311 | if (error) | |
312 | return(error); | |
313 | ASSERT(bp != NULL); | |
1d9025e5 DC |
314 | node = bp->b_addr; |
315 | oldroot = blk1->bp->b_addr; | |
69ef921b | 316 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) { |
fac80cce | 317 | size = (int)((char *)&oldroot->btree[be16_to_cpu(oldroot->hdr.count)] - |
1da177e4 LT |
318 | (char *)oldroot); |
319 | } else { | |
69ef921b | 320 | ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)); |
1da177e4 | 321 | leaf = (xfs_dir2_leaf_t *)oldroot; |
a818e5de | 322 | size = (int)((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] - |
1da177e4 LT |
323 | (char *)leaf); |
324 | } | |
325 | memcpy(node, oldroot, size); | |
1d9025e5 | 326 | xfs_trans_log_buf(tp, bp, 0, size - 1); |
1da177e4 LT |
327 | blk1->bp = bp; |
328 | blk1->blkno = blkno; | |
329 | ||
330 | /* | |
331 | * Set up the new root node. | |
332 | */ | |
333 | error = xfs_da_node_create(args, | |
f6c2d1fa | 334 | (args->whichfork == XFS_DATA_FORK) ? mp->m_dirleafblk : 0, |
fac80cce | 335 | be16_to_cpu(node->hdr.level) + 1, &bp, args->whichfork); |
1da177e4 LT |
336 | if (error) |
337 | return(error); | |
1d9025e5 | 338 | node = bp->b_addr; |
403432dc NS |
339 | node->btree[0].hashval = cpu_to_be32(blk1->hashval); |
340 | node->btree[0].before = cpu_to_be32(blk1->blkno); | |
341 | node->btree[1].hashval = cpu_to_be32(blk2->hashval); | |
342 | node->btree[1].before = cpu_to_be32(blk2->blkno); | |
fac80cce | 343 | node->hdr.count = cpu_to_be16(2); |
1da177e4 LT |
344 | |
345 | #ifdef DEBUG | |
69ef921b | 346 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) { |
1da177e4 LT |
347 | ASSERT(blk1->blkno >= mp->m_dirleafblk && |
348 | blk1->blkno < mp->m_dirfreeblk); | |
349 | ASSERT(blk2->blkno >= mp->m_dirleafblk && | |
350 | blk2->blkno < mp->m_dirfreeblk); | |
351 | } | |
352 | #endif | |
353 | ||
354 | /* Header is already logged by xfs_da_node_create */ | |
1d9025e5 | 355 | xfs_trans_log_buf(tp, bp, |
1da177e4 LT |
356 | XFS_DA_LOGRANGE(node, node->btree, |
357 | sizeof(xfs_da_node_entry_t) * 2)); | |
1da177e4 LT |
358 | |
359 | return(0); | |
360 | } | |
361 | ||
362 | /* | |
363 | * Split the node, rebalance, then add the new entry. | |
364 | */ | |
365 | STATIC int /* error */ | |
366 | xfs_da_node_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk, | |
367 | xfs_da_state_blk_t *newblk, | |
368 | xfs_da_state_blk_t *addblk, | |
369 | int treelevel, int *result) | |
370 | { | |
371 | xfs_da_intnode_t *node; | |
372 | xfs_dablk_t blkno; | |
373 | int newcount, error; | |
374 | int useextra; | |
375 | ||
5a5881cd DC |
376 | trace_xfs_da_node_split(state->args); |
377 | ||
1d9025e5 | 378 | node = oldblk->bp->b_addr; |
69ef921b | 379 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 LT |
380 | |
381 | /* | |
f6c2d1fa | 382 | * With V2 dirs the extra block is data or freespace. |
1da177e4 | 383 | */ |
f6c2d1fa | 384 | useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK; |
1da177e4 LT |
385 | newcount = 1 + useextra; |
386 | /* | |
387 | * Do we have to split the node? | |
388 | */ | |
fac80cce | 389 | if ((be16_to_cpu(node->hdr.count) + newcount) > state->node_ents) { |
1da177e4 LT |
390 | /* |
391 | * Allocate a new node, add to the doubly linked chain of | |
392 | * nodes, then move some of our excess entries into it. | |
393 | */ | |
394 | error = xfs_da_grow_inode(state->args, &blkno); | |
395 | if (error) | |
396 | return(error); /* GROT: dir is inconsistent */ | |
397 | ||
398 | error = xfs_da_node_create(state->args, blkno, treelevel, | |
399 | &newblk->bp, state->args->whichfork); | |
400 | if (error) | |
401 | return(error); /* GROT: dir is inconsistent */ | |
402 | newblk->blkno = blkno; | |
403 | newblk->magic = XFS_DA_NODE_MAGIC; | |
404 | xfs_da_node_rebalance(state, oldblk, newblk); | |
405 | error = xfs_da_blk_link(state, oldblk, newblk); | |
406 | if (error) | |
407 | return(error); | |
408 | *result = 1; | |
409 | } else { | |
410 | *result = 0; | |
411 | } | |
412 | ||
413 | /* | |
414 | * Insert the new entry(s) into the correct block | |
415 | * (updating last hashval in the process). | |
416 | * | |
417 | * xfs_da_node_add() inserts BEFORE the given index, | |
418 | * and as a result of using node_lookup_int() we always | |
419 | * point to a valid entry (not after one), but a split | |
420 | * operation always results in a new block whose hashvals | |
421 | * FOLLOW the current block. | |
422 | * | |
423 | * If we had double-split op below us, then add the extra block too. | |
424 | */ | |
1d9025e5 | 425 | node = oldblk->bp->b_addr; |
fac80cce | 426 | if (oldblk->index <= be16_to_cpu(node->hdr.count)) { |
1da177e4 LT |
427 | oldblk->index++; |
428 | xfs_da_node_add(state, oldblk, addblk); | |
429 | if (useextra) { | |
430 | if (state->extraafter) | |
431 | oldblk->index++; | |
432 | xfs_da_node_add(state, oldblk, &state->extrablk); | |
433 | state->extravalid = 0; | |
434 | } | |
435 | } else { | |
436 | newblk->index++; | |
437 | xfs_da_node_add(state, newblk, addblk); | |
438 | if (useextra) { | |
439 | if (state->extraafter) | |
440 | newblk->index++; | |
441 | xfs_da_node_add(state, newblk, &state->extrablk); | |
442 | state->extravalid = 0; | |
443 | } | |
444 | } | |
445 | ||
446 | return(0); | |
447 | } | |
448 | ||
449 | /* | |
450 | * Balance the btree elements between two intermediate nodes, | |
451 | * usually one full and one empty. | |
452 | * | |
453 | * NOTE: if blk2 is empty, then it will get the upper half of blk1. | |
454 | */ | |
455 | STATIC void | |
456 | xfs_da_node_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1, | |
457 | xfs_da_state_blk_t *blk2) | |
458 | { | |
459 | xfs_da_intnode_t *node1, *node2, *tmpnode; | |
460 | xfs_da_node_entry_t *btree_s, *btree_d; | |
461 | int count, tmp; | |
462 | xfs_trans_t *tp; | |
463 | ||
5a5881cd DC |
464 | trace_xfs_da_node_rebalance(state->args); |
465 | ||
1d9025e5 DC |
466 | node1 = blk1->bp->b_addr; |
467 | node2 = blk2->bp->b_addr; | |
1da177e4 LT |
468 | /* |
469 | * Figure out how many entries need to move, and in which direction. | |
470 | * Swap the nodes around if that makes it simpler. | |
471 | */ | |
fac80cce | 472 | if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) && |
403432dc | 473 | ((be32_to_cpu(node2->btree[0].hashval) < be32_to_cpu(node1->btree[0].hashval)) || |
fac80cce NS |
474 | (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) < |
475 | be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) { | |
1da177e4 LT |
476 | tmpnode = node1; |
477 | node1 = node2; | |
478 | node2 = tmpnode; | |
479 | } | |
69ef921b CH |
480 | ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
481 | ASSERT(node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
fac80cce | 482 | count = (be16_to_cpu(node1->hdr.count) - be16_to_cpu(node2->hdr.count)) / 2; |
1da177e4 LT |
483 | if (count == 0) |
484 | return; | |
485 | tp = state->args->trans; | |
486 | /* | |
487 | * Two cases: high-to-low and low-to-high. | |
488 | */ | |
489 | if (count > 0) { | |
490 | /* | |
491 | * Move elements in node2 up to make a hole. | |
492 | */ | |
fac80cce | 493 | if ((tmp = be16_to_cpu(node2->hdr.count)) > 0) { |
1da177e4 LT |
494 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
495 | btree_s = &node2->btree[0]; | |
496 | btree_d = &node2->btree[count]; | |
497 | memmove(btree_d, btree_s, tmp); | |
498 | } | |
499 | ||
500 | /* | |
501 | * Move the req'd B-tree elements from high in node1 to | |
502 | * low in node2. | |
503 | */ | |
413d57c9 | 504 | be16_add_cpu(&node2->hdr.count, count); |
1da177e4 | 505 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
fac80cce | 506 | btree_s = &node1->btree[be16_to_cpu(node1->hdr.count) - count]; |
1da177e4 LT |
507 | btree_d = &node2->btree[0]; |
508 | memcpy(btree_d, btree_s, tmp); | |
413d57c9 | 509 | be16_add_cpu(&node1->hdr.count, -count); |
1da177e4 LT |
510 | } else { |
511 | /* | |
512 | * Move the req'd B-tree elements from low in node2 to | |
513 | * high in node1. | |
514 | */ | |
515 | count = -count; | |
516 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); | |
517 | btree_s = &node2->btree[0]; | |
fac80cce | 518 | btree_d = &node1->btree[be16_to_cpu(node1->hdr.count)]; |
1da177e4 | 519 | memcpy(btree_d, btree_s, tmp); |
413d57c9 | 520 | be16_add_cpu(&node1->hdr.count, count); |
1d9025e5 | 521 | xfs_trans_log_buf(tp, blk1->bp, |
1da177e4 LT |
522 | XFS_DA_LOGRANGE(node1, btree_d, tmp)); |
523 | ||
524 | /* | |
525 | * Move elements in node2 down to fill the hole. | |
526 | */ | |
fac80cce | 527 | tmp = be16_to_cpu(node2->hdr.count) - count; |
1da177e4 LT |
528 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
529 | btree_s = &node2->btree[count]; | |
530 | btree_d = &node2->btree[0]; | |
531 | memmove(btree_d, btree_s, tmp); | |
413d57c9 | 532 | be16_add_cpu(&node2->hdr.count, -count); |
1da177e4 LT |
533 | } |
534 | ||
535 | /* | |
536 | * Log header of node 1 and all current bits of node 2. | |
537 | */ | |
1d9025e5 | 538 | xfs_trans_log_buf(tp, blk1->bp, |
1da177e4 | 539 | XFS_DA_LOGRANGE(node1, &node1->hdr, sizeof(node1->hdr))); |
1d9025e5 | 540 | xfs_trans_log_buf(tp, blk2->bp, |
1da177e4 LT |
541 | XFS_DA_LOGRANGE(node2, &node2->hdr, |
542 | sizeof(node2->hdr) + | |
fac80cce | 543 | sizeof(node2->btree[0]) * be16_to_cpu(node2->hdr.count))); |
1da177e4 LT |
544 | |
545 | /* | |
546 | * Record the last hashval from each block for upward propagation. | |
547 | * (note: don't use the swapped node pointers) | |
548 | */ | |
1d9025e5 DC |
549 | node1 = blk1->bp->b_addr; |
550 | node2 = blk2->bp->b_addr; | |
fac80cce NS |
551 | blk1->hashval = be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval); |
552 | blk2->hashval = be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval); | |
1da177e4 LT |
553 | |
554 | /* | |
555 | * Adjust the expected index for insertion. | |
556 | */ | |
fac80cce NS |
557 | if (blk1->index >= be16_to_cpu(node1->hdr.count)) { |
558 | blk2->index = blk1->index - be16_to_cpu(node1->hdr.count); | |
559 | blk1->index = be16_to_cpu(node1->hdr.count) + 1; /* make it invalid */ | |
1da177e4 LT |
560 | } |
561 | } | |
562 | ||
563 | /* | |
564 | * Add a new entry to an intermediate node. | |
565 | */ | |
566 | STATIC void | |
567 | xfs_da_node_add(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk, | |
568 | xfs_da_state_blk_t *newblk) | |
569 | { | |
570 | xfs_da_intnode_t *node; | |
571 | xfs_da_node_entry_t *btree; | |
572 | int tmp; | |
1da177e4 | 573 | |
5a5881cd DC |
574 | trace_xfs_da_node_add(state->args); |
575 | ||
1d9025e5 | 576 | node = oldblk->bp->b_addr; |
69ef921b | 577 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
fac80cce | 578 | ASSERT((oldblk->index >= 0) && (oldblk->index <= be16_to_cpu(node->hdr.count))); |
1da177e4 | 579 | ASSERT(newblk->blkno != 0); |
f6c2d1fa | 580 | if (state->args->whichfork == XFS_DATA_FORK) |
73523a2e CH |
581 | ASSERT(newblk->blkno >= state->mp->m_dirleafblk && |
582 | newblk->blkno < state->mp->m_dirfreeblk); | |
1da177e4 LT |
583 | |
584 | /* | |
585 | * We may need to make some room before we insert the new node. | |
586 | */ | |
587 | tmp = 0; | |
588 | btree = &node->btree[ oldblk->index ]; | |
fac80cce NS |
589 | if (oldblk->index < be16_to_cpu(node->hdr.count)) { |
590 | tmp = (be16_to_cpu(node->hdr.count) - oldblk->index) * (uint)sizeof(*btree); | |
1da177e4 LT |
591 | memmove(btree + 1, btree, tmp); |
592 | } | |
403432dc NS |
593 | btree->hashval = cpu_to_be32(newblk->hashval); |
594 | btree->before = cpu_to_be32(newblk->blkno); | |
1d9025e5 | 595 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
1da177e4 | 596 | XFS_DA_LOGRANGE(node, btree, tmp + sizeof(*btree))); |
413d57c9 | 597 | be16_add_cpu(&node->hdr.count, 1); |
1d9025e5 | 598 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
1da177e4 LT |
599 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
600 | ||
601 | /* | |
602 | * Copy the last hash value from the oldblk to propagate upwards. | |
603 | */ | |
fac80cce | 604 | oldblk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1 ].hashval); |
1da177e4 LT |
605 | } |
606 | ||
607 | /*======================================================================== | |
608 | * Routines used for shrinking the Btree. | |
609 | *========================================================================*/ | |
610 | ||
611 | /* | |
612 | * Deallocate an empty leaf node, remove it from its parent, | |
613 | * possibly deallocating that block, etc... | |
614 | */ | |
615 | int | |
616 | xfs_da_join(xfs_da_state_t *state) | |
617 | { | |
618 | xfs_da_state_blk_t *drop_blk, *save_blk; | |
619 | int action, error; | |
620 | ||
5a5881cd DC |
621 | trace_xfs_da_join(state->args); |
622 | ||
1da177e4 LT |
623 | action = 0; |
624 | drop_blk = &state->path.blk[ state->path.active-1 ]; | |
625 | save_blk = &state->altpath.blk[ state->path.active-1 ]; | |
626 | ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC); | |
627 | ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC || | |
f6c2d1fa | 628 | drop_blk->magic == XFS_DIR2_LEAFN_MAGIC); |
1da177e4 LT |
629 | |
630 | /* | |
631 | * Walk back up the tree joining/deallocating as necessary. | |
632 | * When we stop dropping blocks, break out. | |
633 | */ | |
634 | for ( ; state->path.active >= 2; drop_blk--, save_blk--, | |
635 | state->path.active--) { | |
636 | /* | |
637 | * See if we can combine the block with a neighbor. | |
638 | * (action == 0) => no options, just leave | |
639 | * (action == 1) => coalesce, then unlink | |
640 | * (action == 2) => block empty, unlink it | |
641 | */ | |
642 | switch (drop_blk->magic) { | |
643 | case XFS_ATTR_LEAF_MAGIC: | |
1da177e4 | 644 | error = xfs_attr_leaf_toosmall(state, &action); |
1da177e4 LT |
645 | if (error) |
646 | return(error); | |
647 | if (action == 0) | |
648 | return(0); | |
1da177e4 | 649 | xfs_attr_leaf_unbalance(state, drop_blk, save_blk); |
1da177e4 | 650 | break; |
1da177e4 | 651 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
652 | error = xfs_dir2_leafn_toosmall(state, &action); |
653 | if (error) | |
654 | return error; | |
655 | if (action == 0) | |
656 | return 0; | |
657 | xfs_dir2_leafn_unbalance(state, drop_blk, save_blk); | |
658 | break; | |
659 | case XFS_DA_NODE_MAGIC: | |
660 | /* | |
661 | * Remove the offending node, fixup hashvals, | |
662 | * check for a toosmall neighbor. | |
663 | */ | |
664 | xfs_da_node_remove(state, drop_blk); | |
665 | xfs_da_fixhashpath(state, &state->path); | |
666 | error = xfs_da_node_toosmall(state, &action); | |
667 | if (error) | |
668 | return(error); | |
669 | if (action == 0) | |
670 | return 0; | |
671 | xfs_da_node_unbalance(state, drop_blk, save_blk); | |
672 | break; | |
673 | } | |
674 | xfs_da_fixhashpath(state, &state->altpath); | |
675 | error = xfs_da_blk_unlink(state, drop_blk, save_blk); | |
676 | xfs_da_state_kill_altpath(state); | |
677 | if (error) | |
678 | return(error); | |
679 | error = xfs_da_shrink_inode(state->args, drop_blk->blkno, | |
680 | drop_blk->bp); | |
681 | drop_blk->bp = NULL; | |
682 | if (error) | |
683 | return(error); | |
684 | } | |
685 | /* | |
686 | * We joined all the way to the top. If it turns out that | |
687 | * we only have one entry in the root, make the child block | |
688 | * the new root. | |
689 | */ | |
690 | xfs_da_node_remove(state, drop_blk); | |
691 | xfs_da_fixhashpath(state, &state->path); | |
692 | error = xfs_da_root_join(state, &state->path.blk[0]); | |
693 | return(error); | |
694 | } | |
695 | ||
1c4f3329 AE |
696 | #ifdef DEBUG |
697 | static void | |
698 | xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level) | |
699 | { | |
700 | __be16 magic = blkinfo->magic; | |
701 | ||
702 | if (level == 1) { | |
703 | ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
704 | magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)); | |
705 | } else | |
706 | ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
707 | ASSERT(!blkinfo->forw); | |
708 | ASSERT(!blkinfo->back); | |
709 | } | |
710 | #else /* !DEBUG */ | |
711 | #define xfs_da_blkinfo_onlychild_validate(blkinfo, level) | |
712 | #endif /* !DEBUG */ | |
713 | ||
1da177e4 LT |
714 | /* |
715 | * We have only one entry in the root. Copy the only remaining child of | |
716 | * the old root to block 0 as the new root node. | |
717 | */ | |
718 | STATIC int | |
719 | xfs_da_root_join(xfs_da_state_t *state, xfs_da_state_blk_t *root_blk) | |
720 | { | |
721 | xfs_da_intnode_t *oldroot; | |
1da177e4 LT |
722 | xfs_da_args_t *args; |
723 | xfs_dablk_t child; | |
1d9025e5 | 724 | struct xfs_buf *bp; |
1da177e4 LT |
725 | int error; |
726 | ||
5a5881cd DC |
727 | trace_xfs_da_root_join(state->args); |
728 | ||
1da177e4 LT |
729 | args = state->args; |
730 | ASSERT(args != NULL); | |
731 | ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC); | |
1d9025e5 | 732 | oldroot = root_blk->bp->b_addr; |
69ef921b | 733 | ASSERT(oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 LT |
734 | ASSERT(!oldroot->hdr.info.forw); |
735 | ASSERT(!oldroot->hdr.info.back); | |
736 | ||
737 | /* | |
738 | * If the root has more than one child, then don't do anything. | |
739 | */ | |
fac80cce | 740 | if (be16_to_cpu(oldroot->hdr.count) > 1) |
1da177e4 LT |
741 | return(0); |
742 | ||
743 | /* | |
744 | * Read in the (only) child block, then copy those bytes into | |
745 | * the root block's buffer and free the original child block. | |
746 | */ | |
403432dc | 747 | child = be32_to_cpu(oldroot->btree[0].before); |
1da177e4 LT |
748 | ASSERT(child != 0); |
749 | error = xfs_da_read_buf(args->trans, args->dp, child, -1, &bp, | |
4bb20a83 | 750 | args->whichfork, NULL); |
1da177e4 LT |
751 | if (error) |
752 | return(error); | |
753 | ASSERT(bp != NULL); | |
1d9025e5 | 754 | xfs_da_blkinfo_onlychild_validate(bp->b_addr, |
1c4f3329 AE |
755 | be16_to_cpu(oldroot->hdr.level)); |
756 | ||
1d9025e5 DC |
757 | memcpy(root_blk->bp->b_addr, bp->b_addr, state->blocksize); |
758 | xfs_trans_log_buf(args->trans, root_blk->bp, 0, state->blocksize - 1); | |
1da177e4 LT |
759 | error = xfs_da_shrink_inode(args, child, bp); |
760 | return(error); | |
761 | } | |
762 | ||
763 | /* | |
764 | * Check a node block and its neighbors to see if the block should be | |
765 | * collapsed into one or the other neighbor. Always keep the block | |
766 | * with the smaller block number. | |
767 | * If the current block is over 50% full, don't try to join it, return 0. | |
768 | * If the block is empty, fill in the state structure and return 2. | |
769 | * If it can be collapsed, fill in the state structure and return 1. | |
770 | * If nothing can be done, return 0. | |
771 | */ | |
772 | STATIC int | |
773 | xfs_da_node_toosmall(xfs_da_state_t *state, int *action) | |
774 | { | |
775 | xfs_da_intnode_t *node; | |
776 | xfs_da_state_blk_t *blk; | |
777 | xfs_da_blkinfo_t *info; | |
778 | int count, forward, error, retval, i; | |
779 | xfs_dablk_t blkno; | |
1d9025e5 | 780 | struct xfs_buf *bp; |
1da177e4 | 781 | |
ee73259b DC |
782 | trace_xfs_da_node_toosmall(state->args); |
783 | ||
1da177e4 LT |
784 | /* |
785 | * Check for the degenerate case of the block being over 50% full. | |
786 | * If so, it's not worth even looking to see if we might be able | |
787 | * to coalesce with a sibling. | |
788 | */ | |
789 | blk = &state->path.blk[ state->path.active-1 ]; | |
1d9025e5 | 790 | info = blk->bp->b_addr; |
69ef921b | 791 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 | 792 | node = (xfs_da_intnode_t *)info; |
fac80cce | 793 | count = be16_to_cpu(node->hdr.count); |
1da177e4 LT |
794 | if (count > (state->node_ents >> 1)) { |
795 | *action = 0; /* blk over 50%, don't try to join */ | |
796 | return(0); /* blk over 50%, don't try to join */ | |
797 | } | |
798 | ||
799 | /* | |
800 | * Check for the degenerate case of the block being empty. | |
801 | * If the block is empty, we'll simply delete it, no need to | |
c41564b5 | 802 | * coalesce it with a sibling block. We choose (arbitrarily) |
1da177e4 LT |
803 | * to merge with the forward block unless it is NULL. |
804 | */ | |
805 | if (count == 0) { | |
806 | /* | |
807 | * Make altpath point to the block we want to keep and | |
808 | * path point to the block we want to drop (this one). | |
809 | */ | |
89da0544 | 810 | forward = (info->forw != 0); |
1da177e4 LT |
811 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
812 | error = xfs_da_path_shift(state, &state->altpath, forward, | |
813 | 0, &retval); | |
814 | if (error) | |
815 | return(error); | |
816 | if (retval) { | |
817 | *action = 0; | |
818 | } else { | |
819 | *action = 2; | |
820 | } | |
821 | return(0); | |
822 | } | |
823 | ||
824 | /* | |
825 | * Examine each sibling block to see if we can coalesce with | |
826 | * at least 25% free space to spare. We need to figure out | |
827 | * whether to merge with the forward or the backward block. | |
828 | * We prefer coalescing with the lower numbered sibling so as | |
829 | * to shrink a directory over time. | |
830 | */ | |
831 | /* start with smaller blk num */ | |
89da0544 | 832 | forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back)); |
1da177e4 LT |
833 | for (i = 0; i < 2; forward = !forward, i++) { |
834 | if (forward) | |
89da0544 | 835 | blkno = be32_to_cpu(info->forw); |
1da177e4 | 836 | else |
89da0544 | 837 | blkno = be32_to_cpu(info->back); |
1da177e4 LT |
838 | if (blkno == 0) |
839 | continue; | |
840 | error = xfs_da_read_buf(state->args->trans, state->args->dp, | |
4bb20a83 DC |
841 | blkno, -1, &bp, state->args->whichfork, |
842 | NULL); | |
1da177e4 LT |
843 | if (error) |
844 | return(error); | |
845 | ASSERT(bp != NULL); | |
846 | ||
847 | node = (xfs_da_intnode_t *)info; | |
848 | count = state->node_ents; | |
849 | count -= state->node_ents >> 2; | |
fac80cce | 850 | count -= be16_to_cpu(node->hdr.count); |
1d9025e5 | 851 | node = bp->b_addr; |
69ef921b | 852 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
fac80cce | 853 | count -= be16_to_cpu(node->hdr.count); |
1d9025e5 | 854 | xfs_trans_brelse(state->args->trans, bp); |
1da177e4 LT |
855 | if (count >= 0) |
856 | break; /* fits with at least 25% to spare */ | |
857 | } | |
858 | if (i >= 2) { | |
859 | *action = 0; | |
860 | return(0); | |
861 | } | |
862 | ||
863 | /* | |
864 | * Make altpath point to the block we want to keep (the lower | |
865 | * numbered block) and path point to the block we want to drop. | |
866 | */ | |
867 | memcpy(&state->altpath, &state->path, sizeof(state->path)); | |
868 | if (blkno < blk->blkno) { | |
869 | error = xfs_da_path_shift(state, &state->altpath, forward, | |
870 | 0, &retval); | |
871 | if (error) { | |
872 | return(error); | |
873 | } | |
874 | if (retval) { | |
875 | *action = 0; | |
876 | return(0); | |
877 | } | |
878 | } else { | |
879 | error = xfs_da_path_shift(state, &state->path, forward, | |
880 | 0, &retval); | |
881 | if (error) { | |
882 | return(error); | |
883 | } | |
884 | if (retval) { | |
885 | *action = 0; | |
886 | return(0); | |
887 | } | |
888 | } | |
889 | *action = 1; | |
890 | return(0); | |
891 | } | |
892 | ||
893 | /* | |
894 | * Walk back up the tree adjusting hash values as necessary, | |
895 | * when we stop making changes, return. | |
896 | */ | |
897 | void | |
898 | xfs_da_fixhashpath(xfs_da_state_t *state, xfs_da_state_path_t *path) | |
899 | { | |
900 | xfs_da_state_blk_t *blk; | |
901 | xfs_da_intnode_t *node; | |
902 | xfs_da_node_entry_t *btree; | |
903 | xfs_dahash_t lasthash=0; | |
904 | int level, count; | |
905 | ||
ee73259b DC |
906 | trace_xfs_da_fixhashpath(state->args); |
907 | ||
1da177e4 LT |
908 | level = path->active-1; |
909 | blk = &path->blk[ level ]; | |
910 | switch (blk->magic) { | |
1da177e4 LT |
911 | case XFS_ATTR_LEAF_MAGIC: |
912 | lasthash = xfs_attr_leaf_lasthash(blk->bp, &count); | |
913 | if (count == 0) | |
914 | return; | |
915 | break; | |
1da177e4 | 916 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
917 | lasthash = xfs_dir2_leafn_lasthash(blk->bp, &count); |
918 | if (count == 0) | |
919 | return; | |
920 | break; | |
921 | case XFS_DA_NODE_MAGIC: | |
922 | lasthash = xfs_da_node_lasthash(blk->bp, &count); | |
923 | if (count == 0) | |
924 | return; | |
925 | break; | |
926 | } | |
927 | for (blk--, level--; level >= 0; blk--, level--) { | |
1d9025e5 | 928 | node = blk->bp->b_addr; |
69ef921b | 929 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 | 930 | btree = &node->btree[ blk->index ]; |
403432dc | 931 | if (be32_to_cpu(btree->hashval) == lasthash) |
1da177e4 LT |
932 | break; |
933 | blk->hashval = lasthash; | |
403432dc | 934 | btree->hashval = cpu_to_be32(lasthash); |
1d9025e5 | 935 | xfs_trans_log_buf(state->args->trans, blk->bp, |
1da177e4 LT |
936 | XFS_DA_LOGRANGE(node, btree, sizeof(*btree))); |
937 | ||
fac80cce | 938 | lasthash = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
1da177e4 LT |
939 | } |
940 | } | |
941 | ||
942 | /* | |
943 | * Remove an entry from an intermediate node. | |
944 | */ | |
945 | STATIC void | |
946 | xfs_da_node_remove(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk) | |
947 | { | |
948 | xfs_da_intnode_t *node; | |
949 | xfs_da_node_entry_t *btree; | |
950 | int tmp; | |
951 | ||
5a5881cd DC |
952 | trace_xfs_da_node_remove(state->args); |
953 | ||
1d9025e5 | 954 | node = drop_blk->bp->b_addr; |
fac80cce | 955 | ASSERT(drop_blk->index < be16_to_cpu(node->hdr.count)); |
1da177e4 LT |
956 | ASSERT(drop_blk->index >= 0); |
957 | ||
958 | /* | |
959 | * Copy over the offending entry, or just zero it out. | |
960 | */ | |
961 | btree = &node->btree[drop_blk->index]; | |
fac80cce NS |
962 | if (drop_blk->index < (be16_to_cpu(node->hdr.count)-1)) { |
963 | tmp = be16_to_cpu(node->hdr.count) - drop_blk->index - 1; | |
1da177e4 LT |
964 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
965 | memmove(btree, btree + 1, tmp); | |
1d9025e5 | 966 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
1da177e4 | 967 | XFS_DA_LOGRANGE(node, btree, tmp)); |
fac80cce | 968 | btree = &node->btree[be16_to_cpu(node->hdr.count)-1]; |
1da177e4 LT |
969 | } |
970 | memset((char *)btree, 0, sizeof(xfs_da_node_entry_t)); | |
1d9025e5 | 971 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
1da177e4 | 972 | XFS_DA_LOGRANGE(node, btree, sizeof(*btree))); |
413d57c9 | 973 | be16_add_cpu(&node->hdr.count, -1); |
1d9025e5 | 974 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
1da177e4 LT |
975 | XFS_DA_LOGRANGE(node, &node->hdr, sizeof(node->hdr))); |
976 | ||
977 | /* | |
978 | * Copy the last hash value from the block to propagate upwards. | |
979 | */ | |
980 | btree--; | |
403432dc | 981 | drop_blk->hashval = be32_to_cpu(btree->hashval); |
1da177e4 LT |
982 | } |
983 | ||
984 | /* | |
985 | * Unbalance the btree elements between two intermediate nodes, | |
986 | * move all Btree elements from one node into another. | |
987 | */ | |
988 | STATIC void | |
989 | xfs_da_node_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk, | |
990 | xfs_da_state_blk_t *save_blk) | |
991 | { | |
992 | xfs_da_intnode_t *drop_node, *save_node; | |
993 | xfs_da_node_entry_t *btree; | |
994 | int tmp; | |
995 | xfs_trans_t *tp; | |
996 | ||
5a5881cd DC |
997 | trace_xfs_da_node_unbalance(state->args); |
998 | ||
1d9025e5 DC |
999 | drop_node = drop_blk->bp->b_addr; |
1000 | save_node = save_blk->bp->b_addr; | |
69ef921b CH |
1001 | ASSERT(drop_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1002 | ASSERT(save_node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
1da177e4 LT |
1003 | tp = state->args->trans; |
1004 | ||
1005 | /* | |
1006 | * If the dying block has lower hashvals, then move all the | |
1007 | * elements in the remaining block up to make a hole. | |
1008 | */ | |
403432dc | 1009 | if ((be32_to_cpu(drop_node->btree[0].hashval) < be32_to_cpu(save_node->btree[ 0 ].hashval)) || |
fac80cce NS |
1010 | (be32_to_cpu(drop_node->btree[be16_to_cpu(drop_node->hdr.count)-1].hashval) < |
1011 | be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval))) | |
1da177e4 | 1012 | { |
fac80cce NS |
1013 | btree = &save_node->btree[be16_to_cpu(drop_node->hdr.count)]; |
1014 | tmp = be16_to_cpu(save_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t); | |
1da177e4 LT |
1015 | memmove(btree, &save_node->btree[0], tmp); |
1016 | btree = &save_node->btree[0]; | |
1d9025e5 | 1017 | xfs_trans_log_buf(tp, save_blk->bp, |
1da177e4 | 1018 | XFS_DA_LOGRANGE(save_node, btree, |
fac80cce | 1019 | (be16_to_cpu(save_node->hdr.count) + be16_to_cpu(drop_node->hdr.count)) * |
1da177e4 LT |
1020 | sizeof(xfs_da_node_entry_t))); |
1021 | } else { | |
fac80cce | 1022 | btree = &save_node->btree[be16_to_cpu(save_node->hdr.count)]; |
1d9025e5 | 1023 | xfs_trans_log_buf(tp, save_blk->bp, |
1da177e4 | 1024 | XFS_DA_LOGRANGE(save_node, btree, |
fac80cce | 1025 | be16_to_cpu(drop_node->hdr.count) * |
1da177e4 LT |
1026 | sizeof(xfs_da_node_entry_t))); |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * Move all the B-tree elements from drop_blk to save_blk. | |
1031 | */ | |
fac80cce | 1032 | tmp = be16_to_cpu(drop_node->hdr.count) * (uint)sizeof(xfs_da_node_entry_t); |
1da177e4 | 1033 | memcpy(btree, &drop_node->btree[0], tmp); |
413d57c9 | 1034 | be16_add_cpu(&save_node->hdr.count, be16_to_cpu(drop_node->hdr.count)); |
1da177e4 | 1035 | |
1d9025e5 | 1036 | xfs_trans_log_buf(tp, save_blk->bp, |
1da177e4 LT |
1037 | XFS_DA_LOGRANGE(save_node, &save_node->hdr, |
1038 | sizeof(save_node->hdr))); | |
1039 | ||
1040 | /* | |
1041 | * Save the last hashval in the remaining block for upward propagation. | |
1042 | */ | |
fac80cce | 1043 | save_blk->hashval = be32_to_cpu(save_node->btree[be16_to_cpu(save_node->hdr.count)-1].hashval); |
1da177e4 LT |
1044 | } |
1045 | ||
1046 | /*======================================================================== | |
1047 | * Routines used for finding things in the Btree. | |
1048 | *========================================================================*/ | |
1049 | ||
1050 | /* | |
1051 | * Walk down the Btree looking for a particular filename, filling | |
1052 | * in the state structure as we go. | |
1053 | * | |
1054 | * We will set the state structure to point to each of the elements | |
1055 | * in each of the nodes where either the hashval is or should be. | |
1056 | * | |
1057 | * We support duplicate hashval's so for each entry in the current | |
1058 | * node that could contain the desired hashval, descend. This is a | |
1059 | * pruned depth-first tree search. | |
1060 | */ | |
1061 | int /* error */ | |
1062 | xfs_da_node_lookup_int(xfs_da_state_t *state, int *result) | |
1063 | { | |
1064 | xfs_da_state_blk_t *blk; | |
1065 | xfs_da_blkinfo_t *curr; | |
1066 | xfs_da_intnode_t *node; | |
1067 | xfs_da_node_entry_t *btree; | |
1068 | xfs_dablk_t blkno; | |
1069 | int probe, span, max, error, retval; | |
d432c80e | 1070 | xfs_dahash_t hashval, btreehashval; |
1da177e4 LT |
1071 | xfs_da_args_t *args; |
1072 | ||
1073 | args = state->args; | |
1074 | ||
1075 | /* | |
1076 | * Descend thru the B-tree searching each level for the right | |
1077 | * node to use, until the right hashval is found. | |
1078 | */ | |
f6c2d1fa | 1079 | blkno = (args->whichfork == XFS_DATA_FORK)? state->mp->m_dirleafblk : 0; |
1da177e4 LT |
1080 | for (blk = &state->path.blk[0], state->path.active = 1; |
1081 | state->path.active <= XFS_DA_NODE_MAXDEPTH; | |
1082 | blk++, state->path.active++) { | |
1083 | /* | |
1084 | * Read the next node down in the tree. | |
1085 | */ | |
1086 | blk->blkno = blkno; | |
1087 | error = xfs_da_read_buf(args->trans, args->dp, blkno, | |
4bb20a83 | 1088 | -1, &blk->bp, args->whichfork, NULL); |
1da177e4 LT |
1089 | if (error) { |
1090 | blk->blkno = 0; | |
1091 | state->path.active--; | |
1092 | return(error); | |
1093 | } | |
1d9025e5 | 1094 | curr = blk->bp->b_addr; |
d432c80e NS |
1095 | blk->magic = be16_to_cpu(curr->magic); |
1096 | ASSERT(blk->magic == XFS_DA_NODE_MAGIC || | |
1097 | blk->magic == XFS_DIR2_LEAFN_MAGIC || | |
1098 | blk->magic == XFS_ATTR_LEAF_MAGIC); | |
1da177e4 LT |
1099 | |
1100 | /* | |
1101 | * Search an intermediate node for a match. | |
1102 | */ | |
89da0544 | 1103 | if (blk->magic == XFS_DA_NODE_MAGIC) { |
1d9025e5 | 1104 | node = blk->bp->b_addr; |
d432c80e | 1105 | max = be16_to_cpu(node->hdr.count); |
1c91ad3a | 1106 | blk->hashval = be32_to_cpu(node->btree[max-1].hashval); |
1da177e4 LT |
1107 | |
1108 | /* | |
1109 | * Binary search. (note: small blocks will skip loop) | |
1110 | */ | |
1da177e4 LT |
1111 | probe = span = max / 2; |
1112 | hashval = args->hashval; | |
1113 | for (btree = &node->btree[probe]; span > 4; | |
1114 | btree = &node->btree[probe]) { | |
1115 | span /= 2; | |
d432c80e NS |
1116 | btreehashval = be32_to_cpu(btree->hashval); |
1117 | if (btreehashval < hashval) | |
1da177e4 | 1118 | probe += span; |
d432c80e | 1119 | else if (btreehashval > hashval) |
1da177e4 LT |
1120 | probe -= span; |
1121 | else | |
1122 | break; | |
1123 | } | |
1124 | ASSERT((probe >= 0) && (probe < max)); | |
403432dc | 1125 | ASSERT((span <= 4) || (be32_to_cpu(btree->hashval) == hashval)); |
1da177e4 LT |
1126 | |
1127 | /* | |
1128 | * Since we may have duplicate hashval's, find the first | |
1129 | * matching hashval in the node. | |
1130 | */ | |
403432dc | 1131 | while ((probe > 0) && (be32_to_cpu(btree->hashval) >= hashval)) { |
1da177e4 LT |
1132 | btree--; |
1133 | probe--; | |
1134 | } | |
403432dc | 1135 | while ((probe < max) && (be32_to_cpu(btree->hashval) < hashval)) { |
1da177e4 LT |
1136 | btree++; |
1137 | probe++; | |
1138 | } | |
1139 | ||
1140 | /* | |
1141 | * Pick the right block to descend on. | |
1142 | */ | |
1143 | if (probe == max) { | |
1144 | blk->index = max-1; | |
403432dc | 1145 | blkno = be32_to_cpu(node->btree[max-1].before); |
1da177e4 LT |
1146 | } else { |
1147 | blk->index = probe; | |
403432dc | 1148 | blkno = be32_to_cpu(btree->before); |
1da177e4 | 1149 | } |
d432c80e | 1150 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
1da177e4 LT |
1151 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
1152 | break; | |
d432c80e | 1153 | } else if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
1da177e4 LT |
1154 | blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, NULL); |
1155 | break; | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | /* | |
1160 | * A leaf block that ends in the hashval that we are interested in | |
1161 | * (final hashval == search hashval) means that the next block may | |
1162 | * contain more entries with the same hashval, shift upward to the | |
1163 | * next leaf and keep searching. | |
1164 | */ | |
1165 | for (;;) { | |
f6c2d1fa | 1166 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
1da177e4 LT |
1167 | retval = xfs_dir2_leafn_lookup_int(blk->bp, args, |
1168 | &blk->index, state); | |
d432c80e | 1169 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
1da177e4 LT |
1170 | retval = xfs_attr_leaf_lookup_int(blk->bp, args); |
1171 | blk->index = args->index; | |
1172 | args->blkno = blk->blkno; | |
d432c80e NS |
1173 | } else { |
1174 | ASSERT(0); | |
1175 | return XFS_ERROR(EFSCORRUPTED); | |
1da177e4 | 1176 | } |
1da177e4 LT |
1177 | if (((retval == ENOENT) || (retval == ENOATTR)) && |
1178 | (blk->hashval == args->hashval)) { | |
1179 | error = xfs_da_path_shift(state, &state->path, 1, 1, | |
1180 | &retval); | |
1181 | if (error) | |
1182 | return(error); | |
1183 | if (retval == 0) { | |
1184 | continue; | |
d432c80e | 1185 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
1da177e4 LT |
1186 | /* path_shift() gives ENOENT */ |
1187 | retval = XFS_ERROR(ENOATTR); | |
1188 | } | |
1da177e4 LT |
1189 | } |
1190 | break; | |
1191 | } | |
1192 | *result = retval; | |
1193 | return(0); | |
1194 | } | |
1195 | ||
1196 | /*======================================================================== | |
1197 | * Utility routines. | |
1198 | *========================================================================*/ | |
1199 | ||
1200 | /* | |
1201 | * Link a new block into a doubly linked list of blocks (of whatever type). | |
1202 | */ | |
1203 | int /* error */ | |
1204 | xfs_da_blk_link(xfs_da_state_t *state, xfs_da_state_blk_t *old_blk, | |
1205 | xfs_da_state_blk_t *new_blk) | |
1206 | { | |
1207 | xfs_da_blkinfo_t *old_info, *new_info, *tmp_info; | |
1208 | xfs_da_args_t *args; | |
1209 | int before=0, error; | |
1d9025e5 | 1210 | struct xfs_buf *bp; |
1da177e4 LT |
1211 | |
1212 | /* | |
1213 | * Set up environment. | |
1214 | */ | |
1215 | args = state->args; | |
1216 | ASSERT(args != NULL); | |
1d9025e5 DC |
1217 | old_info = old_blk->bp->b_addr; |
1218 | new_info = new_blk->bp->b_addr; | |
1da177e4 | 1219 | ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC || |
f6c2d1fa | 1220 | old_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
1da177e4 | 1221 | old_blk->magic == XFS_ATTR_LEAF_MAGIC); |
89da0544 NS |
1222 | ASSERT(old_blk->magic == be16_to_cpu(old_info->magic)); |
1223 | ASSERT(new_blk->magic == be16_to_cpu(new_info->magic)); | |
1da177e4 LT |
1224 | ASSERT(old_blk->magic == new_blk->magic); |
1225 | ||
1226 | switch (old_blk->magic) { | |
1da177e4 LT |
1227 | case XFS_ATTR_LEAF_MAGIC: |
1228 | before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp); | |
1229 | break; | |
1da177e4 | 1230 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
1231 | before = xfs_dir2_leafn_order(old_blk->bp, new_blk->bp); |
1232 | break; | |
1233 | case XFS_DA_NODE_MAGIC: | |
1234 | before = xfs_da_node_order(old_blk->bp, new_blk->bp); | |
1235 | break; | |
1236 | } | |
1237 | ||
1238 | /* | |
1239 | * Link blocks in appropriate order. | |
1240 | */ | |
1241 | if (before) { | |
1242 | /* | |
1243 | * Link new block in before existing block. | |
1244 | */ | |
5a5881cd | 1245 | trace_xfs_da_link_before(args); |
89da0544 NS |
1246 | new_info->forw = cpu_to_be32(old_blk->blkno); |
1247 | new_info->back = old_info->back; | |
1248 | if (old_info->back) { | |
1da177e4 | 1249 | error = xfs_da_read_buf(args->trans, args->dp, |
89da0544 | 1250 | be32_to_cpu(old_info->back), |
4bb20a83 | 1251 | -1, &bp, args->whichfork, NULL); |
1da177e4 LT |
1252 | if (error) |
1253 | return(error); | |
1254 | ASSERT(bp != NULL); | |
1d9025e5 | 1255 | tmp_info = bp->b_addr; |
89da0544 NS |
1256 | ASSERT(be16_to_cpu(tmp_info->magic) == be16_to_cpu(old_info->magic)); |
1257 | ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno); | |
1258 | tmp_info->forw = cpu_to_be32(new_blk->blkno); | |
1d9025e5 | 1259 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
1da177e4 | 1260 | } |
89da0544 | 1261 | old_info->back = cpu_to_be32(new_blk->blkno); |
1da177e4 LT |
1262 | } else { |
1263 | /* | |
1264 | * Link new block in after existing block. | |
1265 | */ | |
5a5881cd | 1266 | trace_xfs_da_link_after(args); |
89da0544 NS |
1267 | new_info->forw = old_info->forw; |
1268 | new_info->back = cpu_to_be32(old_blk->blkno); | |
1269 | if (old_info->forw) { | |
1da177e4 | 1270 | error = xfs_da_read_buf(args->trans, args->dp, |
89da0544 | 1271 | be32_to_cpu(old_info->forw), |
4bb20a83 | 1272 | -1, &bp, args->whichfork, NULL); |
1da177e4 LT |
1273 | if (error) |
1274 | return(error); | |
1275 | ASSERT(bp != NULL); | |
1d9025e5 | 1276 | tmp_info = bp->b_addr; |
89da0544 NS |
1277 | ASSERT(tmp_info->magic == old_info->magic); |
1278 | ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno); | |
1279 | tmp_info->back = cpu_to_be32(new_blk->blkno); | |
1d9025e5 | 1280 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
1da177e4 | 1281 | } |
89da0544 | 1282 | old_info->forw = cpu_to_be32(new_blk->blkno); |
1da177e4 LT |
1283 | } |
1284 | ||
1d9025e5 DC |
1285 | xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1); |
1286 | xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1); | |
1da177e4 LT |
1287 | return(0); |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * Compare two intermediate nodes for "order". | |
1292 | */ | |
1293 | STATIC int | |
1d9025e5 DC |
1294 | xfs_da_node_order( |
1295 | struct xfs_buf *node1_bp, | |
1296 | struct xfs_buf *node2_bp) | |
1da177e4 LT |
1297 | { |
1298 | xfs_da_intnode_t *node1, *node2; | |
1299 | ||
1d9025e5 DC |
1300 | node1 = node1_bp->b_addr; |
1301 | node2 = node2_bp->b_addr; | |
69ef921b CH |
1302 | ASSERT(node1->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) && |
1303 | node2->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); | |
fac80cce | 1304 | if ((be16_to_cpu(node1->hdr.count) > 0) && (be16_to_cpu(node2->hdr.count) > 0) && |
403432dc NS |
1305 | ((be32_to_cpu(node2->btree[0].hashval) < |
1306 | be32_to_cpu(node1->btree[0].hashval)) || | |
fac80cce NS |
1307 | (be32_to_cpu(node2->btree[be16_to_cpu(node2->hdr.count)-1].hashval) < |
1308 | be32_to_cpu(node1->btree[be16_to_cpu(node1->hdr.count)-1].hashval)))) { | |
1da177e4 LT |
1309 | return(1); |
1310 | } | |
1311 | return(0); | |
1312 | } | |
1313 | ||
1314 | /* | |
1315 | * Pick up the last hashvalue from an intermediate node. | |
1316 | */ | |
1317 | STATIC uint | |
1d9025e5 DC |
1318 | xfs_da_node_lasthash( |
1319 | struct xfs_buf *bp, | |
1320 | int *count) | |
1da177e4 LT |
1321 | { |
1322 | xfs_da_intnode_t *node; | |
1323 | ||
1d9025e5 | 1324 | node = bp->b_addr; |
69ef921b | 1325 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 | 1326 | if (count) |
fac80cce | 1327 | *count = be16_to_cpu(node->hdr.count); |
1da177e4 LT |
1328 | if (!node->hdr.count) |
1329 | return(0); | |
fac80cce | 1330 | return be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
1da177e4 LT |
1331 | } |
1332 | ||
1333 | /* | |
1334 | * Unlink a block from a doubly linked list of blocks. | |
1335 | */ | |
ba0f32d4 | 1336 | STATIC int /* error */ |
1da177e4 LT |
1337 | xfs_da_blk_unlink(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk, |
1338 | xfs_da_state_blk_t *save_blk) | |
1339 | { | |
1340 | xfs_da_blkinfo_t *drop_info, *save_info, *tmp_info; | |
1341 | xfs_da_args_t *args; | |
1d9025e5 | 1342 | struct xfs_buf *bp; |
1da177e4 LT |
1343 | int error; |
1344 | ||
1345 | /* | |
1346 | * Set up environment. | |
1347 | */ | |
1348 | args = state->args; | |
1349 | ASSERT(args != NULL); | |
1d9025e5 DC |
1350 | save_info = save_blk->bp->b_addr; |
1351 | drop_info = drop_blk->bp->b_addr; | |
1da177e4 | 1352 | ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC || |
f6c2d1fa | 1353 | save_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
1da177e4 | 1354 | save_blk->magic == XFS_ATTR_LEAF_MAGIC); |
89da0544 NS |
1355 | ASSERT(save_blk->magic == be16_to_cpu(save_info->magic)); |
1356 | ASSERT(drop_blk->magic == be16_to_cpu(drop_info->magic)); | |
1da177e4 | 1357 | ASSERT(save_blk->magic == drop_blk->magic); |
89da0544 NS |
1358 | ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) || |
1359 | (be32_to_cpu(save_info->back) == drop_blk->blkno)); | |
1360 | ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) || | |
1361 | (be32_to_cpu(drop_info->back) == save_blk->blkno)); | |
1da177e4 LT |
1362 | |
1363 | /* | |
1364 | * Unlink the leaf block from the doubly linked chain of leaves. | |
1365 | */ | |
89da0544 | 1366 | if (be32_to_cpu(save_info->back) == drop_blk->blkno) { |
5a5881cd | 1367 | trace_xfs_da_unlink_back(args); |
89da0544 NS |
1368 | save_info->back = drop_info->back; |
1369 | if (drop_info->back) { | |
1da177e4 | 1370 | error = xfs_da_read_buf(args->trans, args->dp, |
89da0544 | 1371 | be32_to_cpu(drop_info->back), |
4bb20a83 | 1372 | -1, &bp, args->whichfork, NULL); |
1da177e4 LT |
1373 | if (error) |
1374 | return(error); | |
1375 | ASSERT(bp != NULL); | |
1d9025e5 | 1376 | tmp_info = bp->b_addr; |
89da0544 NS |
1377 | ASSERT(tmp_info->magic == save_info->magic); |
1378 | ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno); | |
1379 | tmp_info->forw = cpu_to_be32(save_blk->blkno); | |
1d9025e5 | 1380 | xfs_trans_log_buf(args->trans, bp, 0, |
1da177e4 | 1381 | sizeof(*tmp_info) - 1); |
1da177e4 LT |
1382 | } |
1383 | } else { | |
5a5881cd | 1384 | trace_xfs_da_unlink_forward(args); |
89da0544 NS |
1385 | save_info->forw = drop_info->forw; |
1386 | if (drop_info->forw) { | |
1da177e4 | 1387 | error = xfs_da_read_buf(args->trans, args->dp, |
89da0544 | 1388 | be32_to_cpu(drop_info->forw), |
4bb20a83 | 1389 | -1, &bp, args->whichfork, NULL); |
1da177e4 LT |
1390 | if (error) |
1391 | return(error); | |
1392 | ASSERT(bp != NULL); | |
1d9025e5 | 1393 | tmp_info = bp->b_addr; |
89da0544 NS |
1394 | ASSERT(tmp_info->magic == save_info->magic); |
1395 | ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno); | |
1396 | tmp_info->back = cpu_to_be32(save_blk->blkno); | |
1d9025e5 | 1397 | xfs_trans_log_buf(args->trans, bp, 0, |
1da177e4 | 1398 | sizeof(*tmp_info) - 1); |
1da177e4 LT |
1399 | } |
1400 | } | |
1401 | ||
1d9025e5 | 1402 | xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1); |
1da177e4 LT |
1403 | return(0); |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * Move a path "forward" or "!forward" one block at the current level. | |
1408 | * | |
1409 | * This routine will adjust a "path" to point to the next block | |
1410 | * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the | |
1411 | * Btree, including updating pointers to the intermediate nodes between | |
1412 | * the new bottom and the root. | |
1413 | */ | |
1414 | int /* error */ | |
1415 | xfs_da_path_shift(xfs_da_state_t *state, xfs_da_state_path_t *path, | |
1416 | int forward, int release, int *result) | |
1417 | { | |
1418 | xfs_da_state_blk_t *blk; | |
1419 | xfs_da_blkinfo_t *info; | |
1420 | xfs_da_intnode_t *node; | |
1421 | xfs_da_args_t *args; | |
1422 | xfs_dablk_t blkno=0; | |
1423 | int level, error; | |
1424 | ||
ee73259b DC |
1425 | trace_xfs_da_path_shift(state->args); |
1426 | ||
1da177e4 LT |
1427 | /* |
1428 | * Roll up the Btree looking for the first block where our | |
1429 | * current index is not at the edge of the block. Note that | |
1430 | * we skip the bottom layer because we want the sibling block. | |
1431 | */ | |
1432 | args = state->args; | |
1433 | ASSERT(args != NULL); | |
1434 | ASSERT(path != NULL); | |
1435 | ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH)); | |
1436 | level = (path->active-1) - 1; /* skip bottom layer in path */ | |
1437 | for (blk = &path->blk[level]; level >= 0; blk--, level--) { | |
1438 | ASSERT(blk->bp != NULL); | |
1d9025e5 | 1439 | node = blk->bp->b_addr; |
69ef921b | 1440 | ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
fac80cce | 1441 | if (forward && (blk->index < be16_to_cpu(node->hdr.count)-1)) { |
1da177e4 | 1442 | blk->index++; |
403432dc | 1443 | blkno = be32_to_cpu(node->btree[blk->index].before); |
1da177e4 LT |
1444 | break; |
1445 | } else if (!forward && (blk->index > 0)) { | |
1446 | blk->index--; | |
403432dc | 1447 | blkno = be32_to_cpu(node->btree[blk->index].before); |
1da177e4 LT |
1448 | break; |
1449 | } | |
1450 | } | |
1451 | if (level < 0) { | |
1452 | *result = XFS_ERROR(ENOENT); /* we're out of our tree */ | |
6a178100 | 1453 | ASSERT(args->op_flags & XFS_DA_OP_OKNOENT); |
1da177e4 LT |
1454 | return(0); |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * Roll down the edge of the subtree until we reach the | |
1459 | * same depth we were at originally. | |
1460 | */ | |
1461 | for (blk++, level++; level < path->active; blk++, level++) { | |
1462 | /* | |
1463 | * Release the old block. | |
1464 | * (if it's dirty, trans won't actually let go) | |
1465 | */ | |
1466 | if (release) | |
1d9025e5 | 1467 | xfs_trans_brelse(args->trans, blk->bp); |
1da177e4 LT |
1468 | |
1469 | /* | |
1470 | * Read the next child block. | |
1471 | */ | |
1472 | blk->blkno = blkno; | |
1473 | error = xfs_da_read_buf(args->trans, args->dp, blkno, -1, | |
4bb20a83 | 1474 | &blk->bp, args->whichfork, NULL); |
1da177e4 LT |
1475 | if (error) |
1476 | return(error); | |
1477 | ASSERT(blk->bp != NULL); | |
1d9025e5 | 1478 | info = blk->bp->b_addr; |
69ef921b CH |
1479 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
1480 | info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
1481 | info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)); | |
89da0544 NS |
1482 | blk->magic = be16_to_cpu(info->magic); |
1483 | if (blk->magic == XFS_DA_NODE_MAGIC) { | |
1da177e4 | 1484 | node = (xfs_da_intnode_t *)info; |
fac80cce | 1485 | blk->hashval = be32_to_cpu(node->btree[be16_to_cpu(node->hdr.count)-1].hashval); |
1da177e4 LT |
1486 | if (forward) |
1487 | blk->index = 0; | |
1488 | else | |
fac80cce | 1489 | blk->index = be16_to_cpu(node->hdr.count)-1; |
403432dc | 1490 | blkno = be32_to_cpu(node->btree[blk->index].before); |
1da177e4 LT |
1491 | } else { |
1492 | ASSERT(level == path->active-1); | |
1493 | blk->index = 0; | |
1494 | switch(blk->magic) { | |
1da177e4 LT |
1495 | case XFS_ATTR_LEAF_MAGIC: |
1496 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, | |
1497 | NULL); | |
1498 | break; | |
1da177e4 | 1499 | case XFS_DIR2_LEAFN_MAGIC: |
1da177e4 LT |
1500 | blk->hashval = xfs_dir2_leafn_lasthash(blk->bp, |
1501 | NULL); | |
1502 | break; | |
1503 | default: | |
1504 | ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC || | |
f6c2d1fa | 1505 | blk->magic == XFS_DIR2_LEAFN_MAGIC); |
1da177e4 LT |
1506 | break; |
1507 | } | |
1508 | } | |
1509 | } | |
1510 | *result = 0; | |
1511 | return(0); | |
1512 | } | |
1513 | ||
1514 | ||
1515 | /*======================================================================== | |
1516 | * Utility routines. | |
1517 | *========================================================================*/ | |
1518 | ||
1519 | /* | |
1520 | * Implement a simple hash on a character string. | |
1521 | * Rotate the hash value by 7 bits, then XOR each character in. | |
1522 | * This is implemented with some source-level loop unrolling. | |
1523 | */ | |
1524 | xfs_dahash_t | |
a5687787 | 1525 | xfs_da_hashname(const __uint8_t *name, int namelen) |
1da177e4 LT |
1526 | { |
1527 | xfs_dahash_t hash; | |
1528 | ||
1da177e4 LT |
1529 | /* |
1530 | * Do four characters at a time as long as we can. | |
1531 | */ | |
1532 | for (hash = 0; namelen >= 4; namelen -= 4, name += 4) | |
1533 | hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^ | |
1534 | (name[3] << 0) ^ rol32(hash, 7 * 4); | |
1535 | ||
1536 | /* | |
1537 | * Now do the rest of the characters. | |
1538 | */ | |
1539 | switch (namelen) { | |
1540 | case 3: | |
1541 | return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^ | |
1542 | rol32(hash, 7 * 3); | |
1543 | case 2: | |
1544 | return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2); | |
1545 | case 1: | |
1546 | return (name[0] << 0) ^ rol32(hash, 7 * 1); | |
2b3b6d07 | 1547 | default: /* case 0: */ |
1da177e4 LT |
1548 | return hash; |
1549 | } | |
1da177e4 LT |
1550 | } |
1551 | ||
5163f95a BN |
1552 | enum xfs_dacmp |
1553 | xfs_da_compname( | |
1554 | struct xfs_da_args *args, | |
2bc75421 DC |
1555 | const unsigned char *name, |
1556 | int len) | |
5163f95a BN |
1557 | { |
1558 | return (args->namelen == len && memcmp(args->name, name, len) == 0) ? | |
1559 | XFS_CMP_EXACT : XFS_CMP_DIFFERENT; | |
1560 | } | |
1561 | ||
1562 | static xfs_dahash_t | |
1563 | xfs_default_hashname( | |
1564 | struct xfs_name *name) | |
1565 | { | |
1566 | return xfs_da_hashname(name->name, name->len); | |
1567 | } | |
1568 | ||
1569 | const struct xfs_nameops xfs_default_nameops = { | |
1570 | .hashname = xfs_default_hashname, | |
1571 | .compname = xfs_da_compname | |
1572 | }; | |
1573 | ||
1da177e4 | 1574 | int |
77936d02 CH |
1575 | xfs_da_grow_inode_int( |
1576 | struct xfs_da_args *args, | |
1577 | xfs_fileoff_t *bno, | |
1578 | int count) | |
1da177e4 | 1579 | { |
77936d02 CH |
1580 | struct xfs_trans *tp = args->trans; |
1581 | struct xfs_inode *dp = args->dp; | |
1582 | int w = args->whichfork; | |
1583 | xfs_drfsbno_t nblks = dp->i_d.di_nblocks; | |
1584 | struct xfs_bmbt_irec map, *mapp; | |
1585 | int nmap, error, got, i, mapi; | |
1da177e4 | 1586 | |
1da177e4 LT |
1587 | /* |
1588 | * Find a spot in the file space to put the new block. | |
1589 | */ | |
77936d02 CH |
1590 | error = xfs_bmap_first_unused(tp, dp, count, bno, w); |
1591 | if (error) | |
1da177e4 | 1592 | return error; |
77936d02 | 1593 | |
1da177e4 LT |
1594 | /* |
1595 | * Try mapping it in one filesystem block. | |
1596 | */ | |
1597 | nmap = 1; | |
1598 | ASSERT(args->firstblock != NULL); | |
c0dc7828 DC |
1599 | error = xfs_bmapi_write(tp, dp, *bno, count, |
1600 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG, | |
1da177e4 | 1601 | args->firstblock, args->total, &map, &nmap, |
77936d02 CH |
1602 | args->flist); |
1603 | if (error) | |
1da177e4 | 1604 | return error; |
77936d02 | 1605 | |
1da177e4 LT |
1606 | ASSERT(nmap <= 1); |
1607 | if (nmap == 1) { | |
1608 | mapp = ↦ | |
1609 | mapi = 1; | |
77936d02 CH |
1610 | } else if (nmap == 0 && count > 1) { |
1611 | xfs_fileoff_t b; | |
1612 | int c; | |
1613 | ||
1614 | /* | |
1615 | * If we didn't get it and the block might work if fragmented, | |
1616 | * try without the CONTIG flag. Loop until we get it all. | |
1617 | */ | |
1da177e4 | 1618 | mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP); |
77936d02 | 1619 | for (b = *bno, mapi = 0; b < *bno + count; ) { |
1da177e4 | 1620 | nmap = MIN(XFS_BMAP_MAX_NMAP, count); |
77936d02 | 1621 | c = (int)(*bno + count - b); |
c0dc7828 DC |
1622 | error = xfs_bmapi_write(tp, dp, b, c, |
1623 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, | |
1da177e4 | 1624 | args->firstblock, args->total, |
77936d02 CH |
1625 | &mapp[mapi], &nmap, args->flist); |
1626 | if (error) | |
1627 | goto out_free_map; | |
1da177e4 LT |
1628 | if (nmap < 1) |
1629 | break; | |
1630 | mapi += nmap; | |
1631 | b = mapp[mapi - 1].br_startoff + | |
1632 | mapp[mapi - 1].br_blockcount; | |
1633 | } | |
1634 | } else { | |
1635 | mapi = 0; | |
1636 | mapp = NULL; | |
1637 | } | |
77936d02 | 1638 | |
1da177e4 LT |
1639 | /* |
1640 | * Count the blocks we got, make sure it matches the total. | |
1641 | */ | |
1642 | for (i = 0, got = 0; i < mapi; i++) | |
1643 | got += mapp[i].br_blockcount; | |
77936d02 | 1644 | if (got != count || mapp[0].br_startoff != *bno || |
1da177e4 | 1645 | mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount != |
77936d02 CH |
1646 | *bno + count) { |
1647 | error = XFS_ERROR(ENOSPC); | |
1648 | goto out_free_map; | |
1da177e4 | 1649 | } |
77936d02 | 1650 | |
a7444053 DC |
1651 | /* account for newly allocated blocks in reserved blocks total */ |
1652 | args->total -= dp->i_d.di_nblocks - nblks; | |
77936d02 CH |
1653 | |
1654 | out_free_map: | |
1655 | if (mapp != &map) | |
1656 | kmem_free(mapp); | |
1657 | return error; | |
1658 | } | |
1659 | ||
1660 | /* | |
1661 | * Add a block to the btree ahead of the file. | |
1662 | * Return the new block number to the caller. | |
1663 | */ | |
1664 | int | |
1665 | xfs_da_grow_inode( | |
1666 | struct xfs_da_args *args, | |
1667 | xfs_dablk_t *new_blkno) | |
1668 | { | |
1669 | xfs_fileoff_t bno; | |
1670 | int count; | |
1671 | int error; | |
1672 | ||
5a5881cd DC |
1673 | trace_xfs_da_grow_inode(args); |
1674 | ||
77936d02 CH |
1675 | if (args->whichfork == XFS_DATA_FORK) { |
1676 | bno = args->dp->i_mount->m_dirleafblk; | |
1677 | count = args->dp->i_mount->m_dirblkfsbs; | |
1678 | } else { | |
1679 | bno = 0; | |
1680 | count = 1; | |
1681 | } | |
1682 | ||
1683 | error = xfs_da_grow_inode_int(args, &bno, count); | |
1684 | if (!error) | |
1685 | *new_blkno = (xfs_dablk_t)bno; | |
1686 | return error; | |
1da177e4 LT |
1687 | } |
1688 | ||
1689 | /* | |
1690 | * Ick. We need to always be able to remove a btree block, even | |
1691 | * if there's no space reservation because the filesystem is full. | |
1692 | * This is called if xfs_bunmapi on a btree block fails due to ENOSPC. | |
1693 | * It swaps the target block with the last block in the file. The | |
1694 | * last block in the file can always be removed since it can't cause | |
1695 | * a bmap btree split to do that. | |
1696 | */ | |
1697 | STATIC int | |
1d9025e5 DC |
1698 | xfs_da_swap_lastblock( |
1699 | xfs_da_args_t *args, | |
1700 | xfs_dablk_t *dead_blknop, | |
1701 | struct xfs_buf **dead_bufp) | |
1da177e4 LT |
1702 | { |
1703 | xfs_dablk_t dead_blkno, last_blkno, sib_blkno, par_blkno; | |
1d9025e5 | 1704 | struct xfs_buf *dead_buf, *last_buf, *sib_buf, *par_buf; |
1da177e4 LT |
1705 | xfs_fileoff_t lastoff; |
1706 | xfs_inode_t *ip; | |
1707 | xfs_trans_t *tp; | |
1708 | xfs_mount_t *mp; | |
1709 | int error, w, entno, level, dead_level; | |
1710 | xfs_da_blkinfo_t *dead_info, *sib_info; | |
1711 | xfs_da_intnode_t *par_node, *dead_node; | |
1da177e4 LT |
1712 | xfs_dir2_leaf_t *dead_leaf2; |
1713 | xfs_dahash_t dead_hash; | |
1714 | ||
5a5881cd DC |
1715 | trace_xfs_da_swap_lastblock(args); |
1716 | ||
1da177e4 LT |
1717 | dead_buf = *dead_bufp; |
1718 | dead_blkno = *dead_blknop; | |
1719 | tp = args->trans; | |
1720 | ip = args->dp; | |
1721 | w = args->whichfork; | |
1722 | ASSERT(w == XFS_DATA_FORK); | |
1723 | mp = ip->i_mount; | |
f6c2d1fa NS |
1724 | lastoff = mp->m_dirfreeblk; |
1725 | error = xfs_bmap_last_before(tp, ip, &lastoff, w); | |
1da177e4 LT |
1726 | if (error) |
1727 | return error; | |
1728 | if (unlikely(lastoff == 0)) { | |
1729 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW, | |
1730 | mp); | |
1731 | return XFS_ERROR(EFSCORRUPTED); | |
1732 | } | |
1733 | /* | |
1734 | * Read the last block in the btree space. | |
1735 | */ | |
1736 | last_blkno = (xfs_dablk_t)lastoff - mp->m_dirblkfsbs; | |
4bb20a83 DC |
1737 | error = xfs_da_read_buf(tp, ip, last_blkno, -1, &last_buf, w, NULL); |
1738 | if (error) | |
1da177e4 LT |
1739 | return error; |
1740 | /* | |
1741 | * Copy the last block into the dead buffer and log it. | |
1742 | */ | |
1d9025e5 DC |
1743 | memcpy(dead_buf->b_addr, last_buf->b_addr, mp->m_dirblksize); |
1744 | xfs_trans_log_buf(tp, dead_buf, 0, mp->m_dirblksize - 1); | |
1745 | dead_info = dead_buf->b_addr; | |
1da177e4 LT |
1746 | /* |
1747 | * Get values from the moved block. | |
1748 | */ | |
69ef921b | 1749 | if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) { |
1da177e4 LT |
1750 | dead_leaf2 = (xfs_dir2_leaf_t *)dead_info; |
1751 | dead_level = 0; | |
3c1f9c15 | 1752 | dead_hash = be32_to_cpu(dead_leaf2->ents[be16_to_cpu(dead_leaf2->hdr.count) - 1].hashval); |
1da177e4 | 1753 | } else { |
69ef921b | 1754 | ASSERT(dead_info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)); |
1da177e4 | 1755 | dead_node = (xfs_da_intnode_t *)dead_info; |
fac80cce NS |
1756 | dead_level = be16_to_cpu(dead_node->hdr.level); |
1757 | dead_hash = be32_to_cpu(dead_node->btree[be16_to_cpu(dead_node->hdr.count) - 1].hashval); | |
1da177e4 LT |
1758 | } |
1759 | sib_buf = par_buf = NULL; | |
1760 | /* | |
1761 | * If the moved block has a left sibling, fix up the pointers. | |
1762 | */ | |
89da0544 | 1763 | if ((sib_blkno = be32_to_cpu(dead_info->back))) { |
4bb20a83 DC |
1764 | error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w, |
1765 | NULL); | |
1766 | if (error) | |
1da177e4 | 1767 | goto done; |
1d9025e5 | 1768 | sib_info = sib_buf->b_addr; |
1da177e4 | 1769 | if (unlikely( |
89da0544 NS |
1770 | be32_to_cpu(sib_info->forw) != last_blkno || |
1771 | sib_info->magic != dead_info->magic)) { | |
1da177e4 LT |
1772 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)", |
1773 | XFS_ERRLEVEL_LOW, mp); | |
1774 | error = XFS_ERROR(EFSCORRUPTED); | |
1775 | goto done; | |
1776 | } | |
89da0544 | 1777 | sib_info->forw = cpu_to_be32(dead_blkno); |
1d9025e5 | 1778 | xfs_trans_log_buf(tp, sib_buf, |
1da177e4 LT |
1779 | XFS_DA_LOGRANGE(sib_info, &sib_info->forw, |
1780 | sizeof(sib_info->forw))); | |
1da177e4 LT |
1781 | sib_buf = NULL; |
1782 | } | |
1783 | /* | |
1784 | * If the moved block has a right sibling, fix up the pointers. | |
1785 | */ | |
89da0544 | 1786 | if ((sib_blkno = be32_to_cpu(dead_info->forw))) { |
4bb20a83 DC |
1787 | error = xfs_da_read_buf(tp, ip, sib_blkno, -1, &sib_buf, w, |
1788 | NULL); | |
1789 | if (error) | |
1da177e4 | 1790 | goto done; |
1d9025e5 | 1791 | sib_info = sib_buf->b_addr; |
1da177e4 | 1792 | if (unlikely( |
89da0544 NS |
1793 | be32_to_cpu(sib_info->back) != last_blkno || |
1794 | sib_info->magic != dead_info->magic)) { | |
1da177e4 LT |
1795 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)", |
1796 | XFS_ERRLEVEL_LOW, mp); | |
1797 | error = XFS_ERROR(EFSCORRUPTED); | |
1798 | goto done; | |
1799 | } | |
89da0544 | 1800 | sib_info->back = cpu_to_be32(dead_blkno); |
1d9025e5 | 1801 | xfs_trans_log_buf(tp, sib_buf, |
1da177e4 LT |
1802 | XFS_DA_LOGRANGE(sib_info, &sib_info->back, |
1803 | sizeof(sib_info->back))); | |
1da177e4 LT |
1804 | sib_buf = NULL; |
1805 | } | |
f6c2d1fa | 1806 | par_blkno = mp->m_dirleafblk; |
1da177e4 LT |
1807 | level = -1; |
1808 | /* | |
1809 | * Walk down the tree looking for the parent of the moved block. | |
1810 | */ | |
1811 | for (;;) { | |
4bb20a83 DC |
1812 | error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w, |
1813 | NULL); | |
1814 | if (error) | |
1da177e4 | 1815 | goto done; |
1d9025e5 | 1816 | par_node = par_buf->b_addr; |
69ef921b CH |
1817 | if (unlikely(par_node->hdr.info.magic != |
1818 | cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
fac80cce | 1819 | (level >= 0 && level != be16_to_cpu(par_node->hdr.level) + 1))) { |
1da177e4 LT |
1820 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)", |
1821 | XFS_ERRLEVEL_LOW, mp); | |
1822 | error = XFS_ERROR(EFSCORRUPTED); | |
1823 | goto done; | |
1824 | } | |
fac80cce | 1825 | level = be16_to_cpu(par_node->hdr.level); |
1da177e4 | 1826 | for (entno = 0; |
fac80cce | 1827 | entno < be16_to_cpu(par_node->hdr.count) && |
403432dc | 1828 | be32_to_cpu(par_node->btree[entno].hashval) < dead_hash; |
1da177e4 LT |
1829 | entno++) |
1830 | continue; | |
fac80cce | 1831 | if (unlikely(entno == be16_to_cpu(par_node->hdr.count))) { |
1da177e4 LT |
1832 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)", |
1833 | XFS_ERRLEVEL_LOW, mp); | |
1834 | error = XFS_ERROR(EFSCORRUPTED); | |
1835 | goto done; | |
1836 | } | |
403432dc | 1837 | par_blkno = be32_to_cpu(par_node->btree[entno].before); |
1da177e4 LT |
1838 | if (level == dead_level + 1) |
1839 | break; | |
1d9025e5 | 1840 | xfs_trans_brelse(tp, par_buf); |
1da177e4 LT |
1841 | par_buf = NULL; |
1842 | } | |
1843 | /* | |
1844 | * We're in the right parent block. | |
1845 | * Look for the right entry. | |
1846 | */ | |
1847 | for (;;) { | |
1848 | for (; | |
fac80cce | 1849 | entno < be16_to_cpu(par_node->hdr.count) && |
403432dc | 1850 | be32_to_cpu(par_node->btree[entno].before) != last_blkno; |
1da177e4 LT |
1851 | entno++) |
1852 | continue; | |
fac80cce | 1853 | if (entno < be16_to_cpu(par_node->hdr.count)) |
1da177e4 | 1854 | break; |
89da0544 | 1855 | par_blkno = be32_to_cpu(par_node->hdr.info.forw); |
1d9025e5 | 1856 | xfs_trans_brelse(tp, par_buf); |
1da177e4 LT |
1857 | par_buf = NULL; |
1858 | if (unlikely(par_blkno == 0)) { | |
1859 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)", | |
1860 | XFS_ERRLEVEL_LOW, mp); | |
1861 | error = XFS_ERROR(EFSCORRUPTED); | |
1862 | goto done; | |
1863 | } | |
4bb20a83 DC |
1864 | error = xfs_da_read_buf(tp, ip, par_blkno, -1, &par_buf, w, |
1865 | NULL); | |
1866 | if (error) | |
1da177e4 | 1867 | goto done; |
1d9025e5 | 1868 | par_node = par_buf->b_addr; |
1da177e4 | 1869 | if (unlikely( |
fac80cce | 1870 | be16_to_cpu(par_node->hdr.level) != level || |
69ef921b | 1871 | par_node->hdr.info.magic != cpu_to_be16(XFS_DA_NODE_MAGIC))) { |
1da177e4 LT |
1872 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)", |
1873 | XFS_ERRLEVEL_LOW, mp); | |
1874 | error = XFS_ERROR(EFSCORRUPTED); | |
1875 | goto done; | |
1876 | } | |
1877 | entno = 0; | |
1878 | } | |
1879 | /* | |
1880 | * Update the parent entry pointing to the moved block. | |
1881 | */ | |
403432dc | 1882 | par_node->btree[entno].before = cpu_to_be32(dead_blkno); |
1d9025e5 | 1883 | xfs_trans_log_buf(tp, par_buf, |
1da177e4 LT |
1884 | XFS_DA_LOGRANGE(par_node, &par_node->btree[entno].before, |
1885 | sizeof(par_node->btree[entno].before))); | |
1da177e4 LT |
1886 | *dead_blknop = last_blkno; |
1887 | *dead_bufp = last_buf; | |
1888 | return 0; | |
1889 | done: | |
1890 | if (par_buf) | |
1d9025e5 | 1891 | xfs_trans_brelse(tp, par_buf); |
1da177e4 | 1892 | if (sib_buf) |
1d9025e5 DC |
1893 | xfs_trans_brelse(tp, sib_buf); |
1894 | xfs_trans_brelse(tp, last_buf); | |
1da177e4 LT |
1895 | return error; |
1896 | } | |
1897 | ||
1898 | /* | |
1899 | * Remove a btree block from a directory or attribute. | |
1900 | */ | |
1901 | int | |
1d9025e5 DC |
1902 | xfs_da_shrink_inode( |
1903 | xfs_da_args_t *args, | |
1904 | xfs_dablk_t dead_blkno, | |
1905 | struct xfs_buf *dead_buf) | |
1da177e4 LT |
1906 | { |
1907 | xfs_inode_t *dp; | |
1908 | int done, error, w, count; | |
1da177e4 LT |
1909 | xfs_trans_t *tp; |
1910 | xfs_mount_t *mp; | |
1911 | ||
5a5881cd DC |
1912 | trace_xfs_da_shrink_inode(args); |
1913 | ||
1da177e4 LT |
1914 | dp = args->dp; |
1915 | w = args->whichfork; | |
1916 | tp = args->trans; | |
1917 | mp = dp->i_mount; | |
f6c2d1fa | 1918 | if (w == XFS_DATA_FORK) |
1da177e4 LT |
1919 | count = mp->m_dirblkfsbs; |
1920 | else | |
1921 | count = 1; | |
1922 | for (;;) { | |
1923 | /* | |
1924 | * Remove extents. If we get ENOSPC for a dir we have to move | |
1925 | * the last block to the place we want to kill. | |
1926 | */ | |
1927 | if ((error = xfs_bunmapi(tp, dp, dead_blkno, count, | |
9d87c319 | 1928 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, |
b4e9181e | 1929 | 0, args->firstblock, args->flist, |
1da177e4 LT |
1930 | &done)) == ENOSPC) { |
1931 | if (w != XFS_DATA_FORK) | |
f6c2d1fa | 1932 | break; |
1da177e4 LT |
1933 | if ((error = xfs_da_swap_lastblock(args, &dead_blkno, |
1934 | &dead_buf))) | |
f6c2d1fa NS |
1935 | break; |
1936 | } else { | |
1da177e4 | 1937 | break; |
1da177e4 LT |
1938 | } |
1939 | } | |
1d9025e5 | 1940 | xfs_trans_binval(tp, dead_buf); |
1da177e4 LT |
1941 | return error; |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * See if the mapping(s) for this btree block are valid, i.e. | |
1946 | * don't contain holes, are logically contiguous, and cover the whole range. | |
1947 | */ | |
1948 | STATIC int | |
1949 | xfs_da_map_covers_blocks( | |
1950 | int nmap, | |
1951 | xfs_bmbt_irec_t *mapp, | |
1952 | xfs_dablk_t bno, | |
1953 | int count) | |
1954 | { | |
1955 | int i; | |
1956 | xfs_fileoff_t off; | |
1957 | ||
1958 | for (i = 0, off = bno; i < nmap; i++) { | |
1959 | if (mapp[i].br_startblock == HOLESTARTBLOCK || | |
1960 | mapp[i].br_startblock == DELAYSTARTBLOCK) { | |
1961 | return 0; | |
1962 | } | |
1963 | if (off != mapp[i].br_startoff) { | |
1964 | return 0; | |
1965 | } | |
1966 | off += mapp[i].br_blockcount; | |
1967 | } | |
1968 | return off == bno + count; | |
1969 | } | |
1970 | ||
1971 | /* | |
3605431f DC |
1972 | * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map. |
1973 | * | |
1974 | * For the single map case, it is assumed that the caller has provided a pointer | |
1975 | * to a valid xfs_buf_map. For the multiple map case, this function will | |
1976 | * allocate the xfs_buf_map to hold all the maps and replace the caller's single | |
1977 | * map pointer with the allocated map. | |
1da177e4 | 1978 | */ |
3605431f DC |
1979 | static int |
1980 | xfs_buf_map_from_irec( | |
1981 | struct xfs_mount *mp, | |
1982 | struct xfs_buf_map **mapp, | |
1983 | unsigned int *nmaps, | |
1984 | struct xfs_bmbt_irec *irecs, | |
1985 | unsigned int nirecs) | |
1da177e4 | 1986 | { |
3605431f DC |
1987 | struct xfs_buf_map *map; |
1988 | int i; | |
1989 | ||
1990 | ASSERT(*nmaps == 1); | |
1991 | ASSERT(nirecs >= 1); | |
1992 | ||
1993 | if (nirecs > 1) { | |
1994 | map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), KM_SLEEP); | |
1995 | if (!map) | |
1996 | return ENOMEM; | |
1997 | *mapp = map; | |
1998 | } | |
1999 | ||
2000 | *nmaps = nirecs; | |
2001 | map = *mapp; | |
2002 | for (i = 0; i < *nmaps; i++) { | |
2003 | ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK && | |
2004 | irecs[i].br_startblock != HOLESTARTBLOCK); | |
2005 | map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock); | |
2006 | map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount); | |
2007 | } | |
2008 | return 0; | |
2009 | } | |
2010 | ||
2011 | /* | |
2012 | * Map the block we are given ready for reading. There are three possible return | |
2013 | * values: | |
2014 | * -1 - will be returned if we land in a hole and mappedbno == -2 so the | |
2015 | * caller knows not to execute a subsequent read. | |
2016 | * 0 - if we mapped the block successfully | |
2017 | * >0 - positive error number if there was an error. | |
2018 | */ | |
2019 | static int | |
2020 | xfs_dabuf_map( | |
2021 | struct xfs_trans *trans, | |
2022 | struct xfs_inode *dp, | |
2023 | xfs_dablk_t bno, | |
2024 | xfs_daddr_t mappedbno, | |
2025 | int whichfork, | |
2026 | struct xfs_buf_map **map, | |
2027 | int *nmaps) | |
2028 | { | |
2029 | struct xfs_mount *mp = dp->i_mount; | |
2030 | int nfsb; | |
2031 | int error = 0; | |
2032 | struct xfs_bmbt_irec irec; | |
2033 | struct xfs_bmbt_irec *irecs = &irec; | |
2034 | int nirecs; | |
2035 | ||
2036 | ASSERT(map && *map); | |
2037 | ASSERT(*nmaps == 1); | |
1da177e4 | 2038 | |
f6c2d1fa | 2039 | nfsb = (whichfork == XFS_DATA_FORK) ? mp->m_dirblkfsbs : 1; |
3605431f | 2040 | |
1da177e4 LT |
2041 | /* |
2042 | * Caller doesn't have a mapping. -2 means don't complain | |
2043 | * if we land in a hole. | |
2044 | */ | |
2045 | if (mappedbno == -1 || mappedbno == -2) { | |
2046 | /* | |
2047 | * Optimize the one-block case. | |
2048 | */ | |
3605431f DC |
2049 | if (nfsb != 1) |
2050 | irecs = kmem_zalloc(sizeof(irec) * nfsb, KM_SLEEP); | |
5b777ad5 | 2051 | |
3605431f DC |
2052 | nirecs = nfsb; |
2053 | error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs, | |
2054 | &nirecs, xfs_bmapi_aflag(whichfork)); | |
5b777ad5 | 2055 | if (error) |
3605431f | 2056 | goto out; |
1da177e4 | 2057 | } else { |
3605431f DC |
2058 | irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno); |
2059 | irecs->br_startoff = (xfs_fileoff_t)bno; | |
2060 | irecs->br_blockcount = nfsb; | |
2061 | irecs->br_state = 0; | |
2062 | nirecs = 1; | |
1da177e4 | 2063 | } |
3605431f DC |
2064 | |
2065 | if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) { | |
2066 | error = mappedbno == -2 ? -1 : XFS_ERROR(EFSCORRUPTED); | |
1da177e4 LT |
2067 | if (unlikely(error == EFSCORRUPTED)) { |
2068 | if (xfs_error_level >= XFS_ERRLEVEL_LOW) { | |
3605431f | 2069 | int i; |
0b932ccc DC |
2070 | xfs_alert(mp, "%s: bno %lld dir: inode %lld", |
2071 | __func__, (long long)bno, | |
1da177e4 | 2072 | (long long)dp->i_ino); |
3605431f | 2073 | for (i = 0; i < *nmaps; i++) { |
0b932ccc DC |
2074 | xfs_alert(mp, |
2075 | "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d", | |
1da177e4 | 2076 | i, |
3605431f DC |
2077 | (long long)irecs[i].br_startoff, |
2078 | (long long)irecs[i].br_startblock, | |
2079 | (long long)irecs[i].br_blockcount, | |
2080 | irecs[i].br_state); | |
1da177e4 LT |
2081 | } |
2082 | } | |
2083 | XFS_ERROR_REPORT("xfs_da_do_buf(1)", | |
2084 | XFS_ERRLEVEL_LOW, mp); | |
2085 | } | |
3605431f | 2086 | goto out; |
1da177e4 | 2087 | } |
3605431f DC |
2088 | error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs); |
2089 | out: | |
2090 | if (irecs != &irec) | |
2091 | kmem_free(irecs); | |
2092 | return error; | |
2093 | } | |
2094 | ||
2095 | /* | |
2096 | * Get a buffer for the dir/attr block. | |
2097 | */ | |
2098 | int | |
2099 | xfs_da_get_buf( | |
2100 | struct xfs_trans *trans, | |
2101 | struct xfs_inode *dp, | |
2102 | xfs_dablk_t bno, | |
2103 | xfs_daddr_t mappedbno, | |
1d9025e5 | 2104 | struct xfs_buf **bpp, |
3605431f DC |
2105 | int whichfork) |
2106 | { | |
2107 | struct xfs_buf *bp; | |
2108 | struct xfs_buf_map map; | |
2109 | struct xfs_buf_map *mapp; | |
2110 | int nmap; | |
2111 | int error; | |
2112 | ||
2113 | *bpp = NULL; | |
2114 | mapp = ↦ | |
2115 | nmap = 1; | |
2116 | error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork, | |
2117 | &mapp, &nmap); | |
2118 | if (error) { | |
2119 | /* mapping a hole is not an error, but we don't continue */ | |
2120 | if (error == -1) | |
1da177e4 | 2121 | error = 0; |
3605431f | 2122 | goto out_free; |
1da177e4 | 2123 | } |
3605431f DC |
2124 | |
2125 | bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp, | |
2126 | mapp, nmap, 0); | |
2127 | error = bp ? bp->b_error : XFS_ERROR(EIO); | |
2128 | if (error) { | |
2129 | xfs_trans_brelse(trans, bp); | |
2130 | goto out_free; | |
2131 | } | |
2132 | ||
1d9025e5 | 2133 | *bpp = bp; |
3605431f DC |
2134 | |
2135 | out_free: | |
2136 | if (mapp != &map) | |
2137 | kmem_free(mapp); | |
2138 | ||
2139 | return error; | |
2140 | } | |
2141 | ||
2142 | /* | |
2143 | * Get a buffer for the dir/attr block, fill in the contents. | |
2144 | */ | |
2145 | int | |
2146 | xfs_da_read_buf( | |
2147 | struct xfs_trans *trans, | |
2148 | struct xfs_inode *dp, | |
2149 | xfs_dablk_t bno, | |
2150 | xfs_daddr_t mappedbno, | |
1d9025e5 | 2151 | struct xfs_buf **bpp, |
4bb20a83 DC |
2152 | int whichfork, |
2153 | xfs_buf_iodone_t verifier) | |
3605431f DC |
2154 | { |
2155 | struct xfs_buf *bp; | |
2156 | struct xfs_buf_map map; | |
2157 | struct xfs_buf_map *mapp; | |
2158 | int nmap; | |
2159 | int error; | |
2160 | ||
2161 | *bpp = NULL; | |
2162 | mapp = ↦ | |
2163 | nmap = 1; | |
2164 | error = xfs_dabuf_map(trans, dp, bno, mappedbno, whichfork, | |
2165 | &mapp, &nmap); | |
2166 | if (error) { | |
2167 | /* mapping a hole is not an error, but we don't continue */ | |
2168 | if (error == -1) | |
2169 | error = 0; | |
2170 | goto out_free; | |
2171 | } | |
2172 | ||
2173 | error = xfs_trans_read_buf_map(dp->i_mount, trans, | |
2174 | dp->i_mount->m_ddev_targp, | |
4bb20a83 | 2175 | mapp, nmap, 0, &bp, verifier); |
3605431f DC |
2176 | if (error) |
2177 | goto out_free; | |
2178 | ||
2179 | if (whichfork == XFS_ATTR_FORK) | |
2180 | xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF); | |
1da177e4 | 2181 | else |
3605431f DC |
2182 | xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF); |
2183 | ||
1da177e4 | 2184 | /* |
3605431f DC |
2185 | * This verification code will be moved to a CRC verification callback |
2186 | * function so just leave it here unchanged until then. | |
1da177e4 | 2187 | */ |
3605431f | 2188 | { |
1d9025e5 DC |
2189 | xfs_dir2_data_hdr_t *hdr = bp->b_addr; |
2190 | xfs_dir2_free_t *free = bp->b_addr; | |
2191 | xfs_da_blkinfo_t *info = bp->b_addr; | |
1da177e4 | 2192 | uint magic, magic1; |
3605431f | 2193 | struct xfs_mount *mp = dp->i_mount; |
1da177e4 | 2194 | |
89da0544 | 2195 | magic = be16_to_cpu(info->magic); |
c2066e26 | 2196 | magic1 = be32_to_cpu(hdr->magic); |
1da177e4 LT |
2197 | if (unlikely( |
2198 | XFS_TEST_ERROR((magic != XFS_DA_NODE_MAGIC) && | |
1da177e4 LT |
2199 | (magic != XFS_ATTR_LEAF_MAGIC) && |
2200 | (magic != XFS_DIR2_LEAF1_MAGIC) && | |
2201 | (magic != XFS_DIR2_LEAFN_MAGIC) && | |
2202 | (magic1 != XFS_DIR2_BLOCK_MAGIC) && | |
2203 | (magic1 != XFS_DIR2_DATA_MAGIC) && | |
69ef921b | 2204 | (free->hdr.magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC)), |
1da177e4 LT |
2205 | mp, XFS_ERRTAG_DA_READ_BUF, |
2206 | XFS_RANDOM_DA_READ_BUF))) { | |
3605431f | 2207 | trace_xfs_da_btree_corrupt(bp, _RET_IP_); |
1da177e4 LT |
2208 | XFS_CORRUPTION_ERROR("xfs_da_do_buf(2)", |
2209 | XFS_ERRLEVEL_LOW, mp, info); | |
2210 | error = XFS_ERROR(EFSCORRUPTED); | |
1d9025e5 | 2211 | xfs_trans_brelse(trans, bp); |
3605431f | 2212 | goto out_free; |
1da177e4 LT |
2213 | } |
2214 | } | |
1d9025e5 | 2215 | *bpp = bp; |
3605431f | 2216 | out_free: |
1da177e4 | 2217 | if (mapp != &map) |
f0e2d93c | 2218 | kmem_free(mapp); |
1da177e4 | 2219 | |
3605431f | 2220 | return error; |
1da177e4 LT |
2221 | } |
2222 | ||
2223 | /* | |
2224 | * Readahead the dir/attr block. | |
2225 | */ | |
2226 | xfs_daddr_t | |
2227 | xfs_da_reada_buf( | |
3605431f DC |
2228 | struct xfs_trans *trans, |
2229 | struct xfs_inode *dp, | |
2230 | xfs_dablk_t bno, | |
4bb20a83 DC |
2231 | int whichfork, |
2232 | xfs_buf_iodone_t verifier) | |
1da177e4 | 2233 | { |
3605431f DC |
2234 | xfs_daddr_t mappedbno = -1; |
2235 | struct xfs_buf_map map; | |
2236 | struct xfs_buf_map *mapp; | |
2237 | int nmap; | |
2238 | int error; | |
2239 | ||
2240 | mapp = ↦ | |
2241 | nmap = 1; | |
2242 | error = xfs_dabuf_map(trans, dp, bno, -1, whichfork, | |
2243 | &mapp, &nmap); | |
2244 | if (error) { | |
2245 | /* mapping a hole is not an error, but we don't continue */ | |
2246 | if (error == -1) | |
2247 | error = 0; | |
2248 | goto out_free; | |
2249 | } | |
1da177e4 | 2250 | |
3605431f | 2251 | mappedbno = mapp[0].bm_bn; |
c3f8fc73 | 2252 | xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, NULL); |
3605431f DC |
2253 | |
2254 | out_free: | |
2255 | if (mapp != &map) | |
2256 | kmem_free(mapp); | |
2257 | ||
2258 | if (error) | |
1da177e4 | 2259 | return -1; |
3605431f | 2260 | return mappedbno; |
1da177e4 LT |
2261 | } |
2262 | ||
1da177e4 | 2263 | kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */ |
1da177e4 LT |
2264 | |
2265 | /* | |
2266 | * Allocate a dir-state structure. | |
2267 | * We don't put them on the stack since they're large. | |
2268 | */ | |
2269 | xfs_da_state_t * | |
2270 | xfs_da_state_alloc(void) | |
2271 | { | |
f41d7fb9 | 2272 | return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS); |
1da177e4 LT |
2273 | } |
2274 | ||
2275 | /* | |
2276 | * Kill the altpath contents of a da-state structure. | |
2277 | */ | |
ba0f32d4 | 2278 | STATIC void |
1da177e4 LT |
2279 | xfs_da_state_kill_altpath(xfs_da_state_t *state) |
2280 | { | |
2281 | int i; | |
2282 | ||
1d9025e5 DC |
2283 | for (i = 0; i < state->altpath.active; i++) |
2284 | state->altpath.blk[i].bp = NULL; | |
1da177e4 LT |
2285 | state->altpath.active = 0; |
2286 | } | |
2287 | ||
2288 | /* | |
2289 | * Free a da-state structure. | |
2290 | */ | |
2291 | void | |
2292 | xfs_da_state_free(xfs_da_state_t *state) | |
2293 | { | |
1da177e4 | 2294 | xfs_da_state_kill_altpath(state); |
1da177e4 LT |
2295 | #ifdef DEBUG |
2296 | memset((char *)state, 0, sizeof(*state)); | |
2297 | #endif /* DEBUG */ | |
2298 | kmem_zone_free(xfs_da_state_zone, state); | |
2299 | } |