]>
Commit | Line | Data |
---|---|---|
2bd0ea18 | 1 | /* |
da23017d | 2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
88b32f06 | 3 | * Copyright (c) 2013 Red Hat, Inc. |
da23017d | 4 | * All Rights Reserved. |
5000d01d | 5 | * |
da23017d NS |
6 | * This program is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU General Public License as | |
2bd0ea18 | 8 | * published by the Free Software Foundation. |
5000d01d | 9 | * |
da23017d NS |
10 | * This program is distributed in the hope that it would be useful, |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
5000d01d | 14 | * |
da23017d NS |
15 | * You should have received a copy of the GNU General Public License |
16 | * along with this program; if not, write the Free Software Foundation, | |
17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
2bd0ea18 | 18 | */ |
9c799827 | 19 | #include "libxfs_priv.h" |
b626fb59 DC |
20 | #include "xfs_fs.h" |
21 | #include "xfs_shared.h" | |
22 | #include "xfs_format.h" | |
23 | #include "xfs_log_format.h" | |
24 | #include "xfs_trans_resv.h" | |
25 | #include "xfs_bit.h" | |
26 | #include "xfs_mount.h" | |
27 | #include "xfs_da_format.h" | |
28 | #include "xfs_da_btree.h" | |
29 | #include "xfs_dir2.h" | |
30 | #include "xfs_dir2_priv.h" | |
31 | #include "xfs_inode.h" | |
32 | #include "xfs_trans.h" | |
33 | #include "xfs_alloc.h" | |
34 | #include "xfs_bmap.h" | |
35 | #include "xfs_attr_leaf.h" | |
36 | #include "xfs_trace.h" | |
37 | #include "xfs_cksum.h" | |
2bd0ea18 NS |
38 | |
39 | /* | |
40 | * xfs_da_btree.c | |
41 | * | |
42 | * Routines to implement directories as Btrees of hashed names. | |
43 | */ | |
44 | ||
5e656dbb BN |
45 | /*======================================================================== |
46 | * Function prototypes for the kernel. | |
47 | *========================================================================*/ | |
48 | ||
49 | /* | |
50 | * Routines used for growing the Btree. | |
51 | */ | |
88b32f06 | 52 | STATIC int xfs_da3_root_split(xfs_da_state_t *state, |
5e656dbb BN |
53 | xfs_da_state_blk_t *existing_root, |
54 | xfs_da_state_blk_t *new_child); | |
88b32f06 | 55 | STATIC int xfs_da3_node_split(xfs_da_state_t *state, |
5e656dbb BN |
56 | xfs_da_state_blk_t *existing_blk, |
57 | xfs_da_state_blk_t *split_blk, | |
58 | xfs_da_state_blk_t *blk_to_add, | |
59 | int treelevel, | |
60 | int *result); | |
88b32f06 | 61 | STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state, |
5e656dbb BN |
62 | xfs_da_state_blk_t *node_blk_1, |
63 | xfs_da_state_blk_t *node_blk_2); | |
88b32f06 | 64 | STATIC void xfs_da3_node_add(xfs_da_state_t *state, |
5e656dbb BN |
65 | xfs_da_state_blk_t *old_node_blk, |
66 | xfs_da_state_blk_t *new_node_blk); | |
67 | ||
68 | /* | |
69 | * Routines used for shrinking the Btree. | |
70 | */ | |
88b32f06 | 71 | STATIC int xfs_da3_root_join(xfs_da_state_t *state, |
5e656dbb | 72 | xfs_da_state_blk_t *root_blk); |
88b32f06 DC |
73 | STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval); |
74 | STATIC void xfs_da3_node_remove(xfs_da_state_t *state, | |
5e656dbb | 75 | xfs_da_state_blk_t *drop_blk); |
88b32f06 | 76 | STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state, |
5e656dbb BN |
77 | xfs_da_state_blk_t *src_node_blk, |
78 | xfs_da_state_blk_t *dst_node_blk); | |
79 | ||
80 | /* | |
81 | * Utility routines. | |
82 | */ | |
88b32f06 | 83 | STATIC int xfs_da3_blk_unlink(xfs_da_state_t *state, |
5e656dbb BN |
84 | xfs_da_state_blk_t *drop_blk, |
85 | xfs_da_state_blk_t *save_blk); | |
5e656dbb | 86 | |
88b32f06 DC |
87 | |
88 | kmem_zone_t *xfs_da_state_zone; /* anchor for state struct zone */ | |
89 | ||
90 | /* | |
91 | * Allocate a dir-state structure. | |
92 | * We don't put them on the stack since they're large. | |
93 | */ | |
94 | xfs_da_state_t * | |
95 | xfs_da_state_alloc(void) | |
96 | { | |
97 | return kmem_zone_zalloc(xfs_da_state_zone, KM_NOFS); | |
98 | } | |
99 | ||
100 | /* | |
101 | * Kill the altpath contents of a da-state structure. | |
102 | */ | |
103 | STATIC void | |
104 | xfs_da_state_kill_altpath(xfs_da_state_t *state) | |
105 | { | |
106 | int i; | |
107 | ||
108 | for (i = 0; i < state->altpath.active; i++) | |
109 | state->altpath.blk[i].bp = NULL; | |
110 | state->altpath.active = 0; | |
111 | } | |
112 | ||
113 | /* | |
114 | * Free a da-state structure. | |
115 | */ | |
116 | void | |
117 | xfs_da_state_free(xfs_da_state_t *state) | |
118 | { | |
119 | xfs_da_state_kill_altpath(state); | |
120 | #ifdef DEBUG | |
121 | memset((char *)state, 0, sizeof(*state)); | |
122 | #endif /* DEBUG */ | |
123 | kmem_zone_free(xfs_da_state_zone, state); | |
124 | } | |
125 | ||
88b32f06 DC |
126 | static bool |
127 | xfs_da3_node_verify( | |
a2ceac1f DC |
128 | struct xfs_buf *bp) |
129 | { | |
130 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
88b32f06 DC |
131 | struct xfs_da_intnode *hdr = bp->b_addr; |
132 | struct xfs_da3_icnode_hdr ichdr; | |
ff105f75 DC |
133 | const struct xfs_dir_ops *ops; |
134 | ||
135 | ops = xfs_dir_get_ops(mp, NULL); | |
88b32f06 | 136 | |
ff105f75 | 137 | ops->node_hdr_from_disk(&ichdr, hdr); |
88b32f06 DC |
138 | |
139 | if (xfs_sb_version_hascrc(&mp->m_sb)) { | |
140 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; | |
141 | ||
142 | if (ichdr.magic != XFS_DA3_NODE_MAGIC) | |
143 | return false; | |
144 | ||
9c4e12fb | 145 | if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid)) |
88b32f06 DC |
146 | return false; |
147 | if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn) | |
148 | return false; | |
a65d8d29 BF |
149 | if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn))) |
150 | return false; | |
88b32f06 DC |
151 | } else { |
152 | if (ichdr.magic != XFS_DA_NODE_MAGIC) | |
153 | return false; | |
a2ceac1f | 154 | } |
88b32f06 DC |
155 | if (ichdr.level == 0) |
156 | return false; | |
157 | if (ichdr.level > XFS_DA_NODE_MAXDEPTH) | |
158 | return false; | |
159 | if (ichdr.count == 0) | |
160 | return false; | |
161 | ||
162 | /* | |
163 | * we don't know if the node is for and attribute or directory tree, | |
164 | * so only fail if the count is outside both bounds | |
165 | */ | |
ff105f75 DC |
166 | if (ichdr.count > mp->m_dir_geo->node_ents && |
167 | ichdr.count > mp->m_attr_geo->node_ents) | |
88b32f06 DC |
168 | return false; |
169 | ||
170 | /* XXX: hash order check? */ | |
a2ceac1f | 171 | |
88b32f06 | 172 | return true; |
a2ceac1f DC |
173 | } |
174 | ||
175 | static void | |
88b32f06 | 176 | xfs_da3_node_write_verify( |
a2ceac1f DC |
177 | struct xfs_buf *bp) |
178 | { | |
88b32f06 DC |
179 | struct xfs_mount *mp = bp->b_target->bt_mount; |
180 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
181 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; | |
182 | ||
183 | if (!xfs_da3_node_verify(bp)) { | |
7e6c95f1 | 184 | xfs_verifier_error(bp, -EFSCORRUPTED); |
88b32f06 DC |
185 | return; |
186 | } | |
187 | ||
188 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
189 | return; | |
190 | ||
191 | if (bip) | |
192 | hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
193 | ||
43b5aeed | 194 | xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF); |
a2ceac1f DC |
195 | } |
196 | ||
197 | /* | |
198 | * leaf/node format detection on trees is sketchy, so a node read can be done on | |
199 | * leaf level blocks when detection identifies the tree as a node format tree | |
200 | * incorrectly. In this case, we need to swap the verifier to match the correct | |
201 | * format of the block being read. | |
202 | */ | |
203 | static void | |
88b32f06 | 204 | xfs_da3_node_read_verify( |
a2ceac1f DC |
205 | struct xfs_buf *bp) |
206 | { | |
a2ceac1f DC |
207 | struct xfs_da_blkinfo *info = bp->b_addr; |
208 | ||
209 | switch (be16_to_cpu(info->magic)) { | |
88b32f06 | 210 | case XFS_DA3_NODE_MAGIC: |
45922933 | 211 | if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) { |
7e6c95f1 | 212 | xfs_verifier_error(bp, -EFSBADCRC); |
88b32f06 | 213 | break; |
45922933 | 214 | } |
88b32f06 | 215 | /* fall through */ |
a2ceac1f | 216 | case XFS_DA_NODE_MAGIC: |
7e6c95f1 DW |
217 | if (!xfs_da3_node_verify(bp)) |
218 | xfs_verifier_error(bp, -EFSCORRUPTED); | |
88b32f06 | 219 | return; |
a2ceac1f | 220 | case XFS_ATTR_LEAF_MAGIC: |
a24374f4 DC |
221 | case XFS_ATTR3_LEAF_MAGIC: |
222 | bp->b_ops = &xfs_attr3_leaf_buf_ops; | |
a2ceac1f DC |
223 | bp->b_ops->verify_read(bp); |
224 | return; | |
225 | case XFS_DIR2_LEAFN_MAGIC: | |
65b80c98 DC |
226 | case XFS_DIR3_LEAFN_MAGIC: |
227 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
a2ceac1f DC |
228 | bp->b_ops->verify_read(bp); |
229 | return; | |
230 | default: | |
7e6c95f1 | 231 | xfs_verifier_error(bp, -EFSCORRUPTED); |
a2ceac1f DC |
232 | break; |
233 | } | |
234 | } | |
235 | ||
88b32f06 | 236 | const struct xfs_buf_ops xfs_da3_node_buf_ops = { |
a3fac935 | 237 | .name = "xfs_da3_node", |
88b32f06 DC |
238 | .verify_read = xfs_da3_node_read_verify, |
239 | .verify_write = xfs_da3_node_write_verify, | |
a2ceac1f DC |
240 | }; |
241 | ||
a2ceac1f | 242 | int |
88b32f06 | 243 | xfs_da3_node_read( |
a2ceac1f DC |
244 | struct xfs_trans *tp, |
245 | struct xfs_inode *dp, | |
246 | xfs_dablk_t bno, | |
247 | xfs_daddr_t mappedbno, | |
248 | struct xfs_buf **bpp, | |
249 | int which_fork) | |
250 | { | |
8b4dc4a9 DC |
251 | int err; |
252 | ||
253 | err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp, | |
88b32f06 | 254 | which_fork, &xfs_da3_node_buf_ops); |
a6c6b1bd | 255 | if (!err && tp && *bpp) { |
8b4dc4a9 DC |
256 | struct xfs_da_blkinfo *info = (*bpp)->b_addr; |
257 | int type; | |
258 | ||
259 | switch (be16_to_cpu(info->magic)) { | |
8b4dc4a9 | 260 | case XFS_DA_NODE_MAGIC: |
f383ee8c | 261 | case XFS_DA3_NODE_MAGIC: |
bdc16ee5 | 262 | type = XFS_BLFT_DA_NODE_BUF; |
8b4dc4a9 DC |
263 | break; |
264 | case XFS_ATTR_LEAF_MAGIC: | |
265 | case XFS_ATTR3_LEAF_MAGIC: | |
bdc16ee5 | 266 | type = XFS_BLFT_ATTR_LEAF_BUF; |
8b4dc4a9 DC |
267 | break; |
268 | case XFS_DIR2_LEAFN_MAGIC: | |
269 | case XFS_DIR3_LEAFN_MAGIC: | |
bdc16ee5 | 270 | type = XFS_BLFT_DIR_LEAFN_BUF; |
8b4dc4a9 DC |
271 | break; |
272 | default: | |
273 | type = 0; | |
274 | ASSERT(0); | |
275 | break; | |
276 | } | |
277 | xfs_trans_buf_set_type(tp, *bpp, type); | |
278 | } | |
279 | return err; | |
a2ceac1f DC |
280 | } |
281 | ||
2bd0ea18 NS |
282 | /*======================================================================== |
283 | * Routines used for growing the Btree. | |
284 | *========================================================================*/ | |
285 | ||
286 | /* | |
287 | * Create the initial contents of an intermediate node. | |
288 | */ | |
289 | int | |
88b32f06 DC |
290 | xfs_da3_node_create( |
291 | struct xfs_da_args *args, | |
292 | xfs_dablk_t blkno, | |
293 | int level, | |
294 | struct xfs_buf **bpp, | |
295 | int whichfork) | |
2bd0ea18 | 296 | { |
88b32f06 DC |
297 | struct xfs_da_intnode *node; |
298 | struct xfs_trans *tp = args->trans; | |
299 | struct xfs_mount *mp = tp->t_mountp; | |
300 | struct xfs_da3_icnode_hdr ichdr = {0}; | |
301 | struct xfs_buf *bp; | |
302 | int error; | |
ff105f75 | 303 | struct xfs_inode *dp = args->dp; |
2bd0ea18 | 304 | |
a2ceac1f | 305 | trace_xfs_da_node_create(args); |
88b32f06 | 306 | ASSERT(level <= XFS_DA_NODE_MAXDEPTH); |
a2ceac1f | 307 | |
ff105f75 | 308 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, whichfork); |
2bd0ea18 | 309 | if (error) |
af43ca9f | 310 | return error; |
8b4dc4a9 | 311 | bp->b_ops = &xfs_da3_node_buf_ops; |
bdc16ee5 | 312 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); |
a2ceac1f | 313 | node = bp->b_addr; |
2bd0ea18 | 314 | |
88b32f06 DC |
315 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
316 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; | |
317 | ||
a65d8d29 | 318 | memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr)); |
88b32f06 DC |
319 | ichdr.magic = XFS_DA3_NODE_MAGIC; |
320 | hdr3->info.blkno = cpu_to_be64(bp->b_bn); | |
321 | hdr3->info.owner = cpu_to_be64(args->dp->i_ino); | |
9c4e12fb | 322 | uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid); |
88b32f06 DC |
323 | } else { |
324 | ichdr.magic = XFS_DA_NODE_MAGIC; | |
325 | } | |
326 | ichdr.level = level; | |
327 | ||
ff105f75 | 328 | dp->d_ops->node_hdr_to_disk(node, &ichdr); |
a2ceac1f | 329 | xfs_trans_log_buf(tp, bp, |
ff105f75 | 330 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
2bd0ea18 NS |
331 | |
332 | *bpp = bp; | |
af43ca9f | 333 | return 0; |
2bd0ea18 NS |
334 | } |
335 | ||
336 | /* | |
337 | * Split a leaf node, rebalance, then possibly split | |
338 | * intermediate nodes, rebalance, etc. | |
339 | */ | |
340 | int /* error */ | |
88b32f06 DC |
341 | xfs_da3_split( |
342 | struct xfs_da_state *state) | |
2bd0ea18 | 343 | { |
88b32f06 DC |
344 | struct xfs_da_state_blk *oldblk; |
345 | struct xfs_da_state_blk *newblk; | |
346 | struct xfs_da_state_blk *addblk; | |
347 | struct xfs_da_intnode *node; | |
88b32f06 | 348 | int max; |
6bddecbc | 349 | int action = 0; |
88b32f06 DC |
350 | int error; |
351 | int i; | |
2bd0ea18 | 352 | |
a2ceac1f DC |
353 | trace_xfs_da_split(state->args); |
354 | ||
2bd0ea18 NS |
355 | /* |
356 | * Walk back up the tree splitting/inserting/adjusting as necessary. | |
357 | * If we need to insert and there isn't room, split the node, then | |
358 | * decide which fragment to insert the new block from below into. | |
359 | * Note that we may split the root this way, but we need more fixup. | |
360 | */ | |
361 | max = state->path.active - 1; | |
362 | ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH)); | |
363 | ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC || | |
5e656dbb | 364 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); |
2bd0ea18 NS |
365 | |
366 | addblk = &state->path.blk[max]; /* initial dummy value */ | |
367 | for (i = max; (i >= 0) && addblk; state->path.active--, i--) { | |
368 | oldblk = &state->path.blk[i]; | |
369 | newblk = &state->altpath.blk[i]; | |
370 | ||
371 | /* | |
372 | * If a leaf node then | |
373 | * Allocate a new leaf node, then rebalance across them. | |
374 | * else if an intermediate node then | |
375 | * We split on the last layer, must we split the node? | |
376 | */ | |
377 | switch (oldblk->magic) { | |
378 | case XFS_ATTR_LEAF_MAGIC: | |
a24374f4 | 379 | error = xfs_attr3_leaf_split(state, oldblk, newblk); |
12b53197 | 380 | if ((error != 0) && (error != -ENOSPC)) { |
af43ca9f | 381 | return error; /* GROT: attr is inconsistent */ |
2bd0ea18 NS |
382 | } |
383 | if (!error) { | |
384 | addblk = newblk; | |
385 | break; | |
386 | } | |
387 | /* | |
4280e59d DC |
388 | * Entry wouldn't fit, split the leaf again. The new |
389 | * extrablk will be consumed by xfs_da3_node_split if | |
390 | * the node is split. | |
2bd0ea18 NS |
391 | */ |
392 | state->extravalid = 1; | |
393 | if (state->inleaf) { | |
394 | state->extraafter = 0; /* before newblk */ | |
a2ceac1f | 395 | trace_xfs_attr_leaf_split_before(state->args); |
a24374f4 | 396 | error = xfs_attr3_leaf_split(state, oldblk, |
2bd0ea18 NS |
397 | &state->extrablk); |
398 | } else { | |
399 | state->extraafter = 1; /* after newblk */ | |
a2ceac1f | 400 | trace_xfs_attr_leaf_split_after(state->args); |
a24374f4 | 401 | error = xfs_attr3_leaf_split(state, newblk, |
2bd0ea18 NS |
402 | &state->extrablk); |
403 | } | |
404 | if (error) | |
af43ca9f | 405 | return error; /* GROT: attr inconsistent */ |
2bd0ea18 NS |
406 | addblk = newblk; |
407 | break; | |
2bd0ea18 | 408 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
409 | error = xfs_dir2_leafn_split(state, oldblk, newblk); |
410 | if (error) | |
411 | return error; | |
412 | addblk = newblk; | |
413 | break; | |
414 | case XFS_DA_NODE_MAGIC: | |
88b32f06 | 415 | error = xfs_da3_node_split(state, oldblk, newblk, addblk, |
2bd0ea18 | 416 | max - i, &action); |
2bd0ea18 NS |
417 | addblk->bp = NULL; |
418 | if (error) | |
af43ca9f | 419 | return error; /* GROT: dir is inconsistent */ |
2bd0ea18 NS |
420 | /* |
421 | * Record the newly split block for the next time thru? | |
422 | */ | |
423 | if (action) | |
424 | addblk = newblk; | |
425 | else | |
426 | addblk = NULL; | |
427 | break; | |
428 | } | |
429 | ||
430 | /* | |
431 | * Update the btree to show the new hashval for this child. | |
432 | */ | |
88b32f06 | 433 | xfs_da3_fixhashpath(state, &state->path); |
2bd0ea18 NS |
434 | } |
435 | if (!addblk) | |
af43ca9f | 436 | return 0; |
2bd0ea18 | 437 | |
4280e59d DC |
438 | /* |
439 | * xfs_da3_node_split() should have consumed any extra blocks we added | |
440 | * during a double leaf split in the attr fork. This is guaranteed as | |
441 | * we can't be here if the attr fork only has a single leaf block. | |
442 | */ | |
443 | ASSERT(state->extravalid == 0 || | |
444 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); | |
445 | ||
2bd0ea18 NS |
446 | /* |
447 | * Split the root node. | |
448 | */ | |
449 | ASSERT(state->path.active == 0); | |
450 | oldblk = &state->path.blk[0]; | |
88b32f06 | 451 | error = xfs_da3_root_split(state, oldblk, addblk); |
2bd0ea18 | 452 | if (error) { |
2bd0ea18 | 453 | addblk->bp = NULL; |
af43ca9f | 454 | return error; /* GROT: dir is inconsistent */ |
2bd0ea18 NS |
455 | } |
456 | ||
457 | /* | |
4280e59d DC |
458 | * Update pointers to the node which used to be block 0 and just got |
459 | * bumped because of the addition of a new root node. Note that the | |
460 | * original block 0 could be at any position in the list of blocks in | |
461 | * the tree. | |
88b32f06 | 462 | * |
4280e59d DC |
463 | * Note: the magic numbers and sibling pointers are in the same physical |
464 | * place for both v2 and v3 headers (by design). Hence it doesn't matter | |
465 | * which version of the xfs_da_intnode structure we use here as the | |
466 | * result will be the same using either structure. | |
2bd0ea18 | 467 | */ |
a2ceac1f | 468 | node = oldblk->bp->b_addr; |
46eca962 | 469 | if (node->hdr.info.forw) { |
4280e59d DC |
470 | ASSERT(be32_to_cpu(node->hdr.info.forw) == addblk->blkno); |
471 | node = addblk->bp->b_addr; | |
5e656dbb | 472 | node->hdr.info.back = cpu_to_be32(oldblk->blkno); |
4280e59d DC |
473 | xfs_trans_log_buf(state->args->trans, addblk->bp, |
474 | XFS_DA_LOGRANGE(node, &node->hdr.info, | |
475 | sizeof(node->hdr.info))); | |
2bd0ea18 | 476 | } |
a2ceac1f | 477 | node = oldblk->bp->b_addr; |
5e656dbb | 478 | if (node->hdr.info.back) { |
4280e59d DC |
479 | ASSERT(be32_to_cpu(node->hdr.info.back) == addblk->blkno); |
480 | node = addblk->bp->b_addr; | |
5e656dbb | 481 | node->hdr.info.forw = cpu_to_be32(oldblk->blkno); |
4280e59d DC |
482 | xfs_trans_log_buf(state->args->trans, addblk->bp, |
483 | XFS_DA_LOGRANGE(node, &node->hdr.info, | |
484 | sizeof(node->hdr.info))); | |
2bd0ea18 | 485 | } |
2bd0ea18 | 486 | addblk->bp = NULL; |
af43ca9f | 487 | return 0; |
2bd0ea18 NS |
488 | } |
489 | ||
490 | /* | |
491 | * Split the root. We have to create a new root and point to the two | |
492 | * parts (the split old root) that we just created. Copy block zero to | |
493 | * the EOF, extending the inode in process. | |
494 | */ | |
495 | STATIC int /* error */ | |
88b32f06 DC |
496 | xfs_da3_root_split( |
497 | struct xfs_da_state *state, | |
498 | struct xfs_da_state_blk *blk1, | |
499 | struct xfs_da_state_blk *blk2) | |
2bd0ea18 | 500 | { |
88b32f06 DC |
501 | struct xfs_da_intnode *node; |
502 | struct xfs_da_intnode *oldroot; | |
503 | struct xfs_da_node_entry *btree; | |
504 | struct xfs_da3_icnode_hdr nodehdr; | |
505 | struct xfs_da_args *args; | |
506 | struct xfs_buf *bp; | |
507 | struct xfs_inode *dp; | |
508 | struct xfs_trans *tp; | |
88b32f06 DC |
509 | struct xfs_dir2_leaf *leaf; |
510 | xfs_dablk_t blkno; | |
511 | int level; | |
512 | int error; | |
513 | int size; | |
2bd0ea18 | 514 | |
a2ceac1f DC |
515 | trace_xfs_da_root_split(state->args); |
516 | ||
2bd0ea18 NS |
517 | /* |
518 | * Copy the existing (incorrect) block from the root node position | |
519 | * to a free space somewhere. | |
520 | */ | |
521 | args = state->args; | |
2bd0ea18 NS |
522 | error = xfs_da_grow_inode(args, &blkno); |
523 | if (error) | |
88b32f06 DC |
524 | return error; |
525 | ||
2bd0ea18 NS |
526 | dp = args->dp; |
527 | tp = args->trans; | |
2bd0ea18 NS |
528 | error = xfs_da_get_buf(tp, dp, blkno, -1, &bp, args->whichfork); |
529 | if (error) | |
88b32f06 | 530 | return error; |
a2ceac1f DC |
531 | node = bp->b_addr; |
532 | oldroot = blk1->bp->b_addr; | |
88b32f06 DC |
533 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
534 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) { | |
19ebedcf | 535 | struct xfs_da3_icnode_hdr icnodehdr; |
88b32f06 | 536 | |
19ebedcf | 537 | dp->d_ops->node_hdr_from_disk(&icnodehdr, oldroot); |
ff105f75 | 538 | btree = dp->d_ops->node_tree_p(oldroot); |
19ebedcf DC |
539 | size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot); |
540 | level = icnodehdr.level; | |
8b4dc4a9 DC |
541 | |
542 | /* | |
543 | * we are about to copy oldroot to bp, so set up the type | |
544 | * of bp while we know exactly what it will be. | |
545 | */ | |
bdc16ee5 | 546 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); |
2bd0ea18 | 547 | } else { |
65b80c98 DC |
548 | struct xfs_dir3_icleaf_hdr leafhdr; |
549 | struct xfs_dir2_leaf_entry *ents; | |
550 | ||
2bd0ea18 | 551 | leaf = (xfs_dir2_leaf_t *)oldroot; |
ff105f75 DC |
552 | dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf); |
553 | ents = dp->d_ops->leaf_ents_p(leaf); | |
65b80c98 DC |
554 | |
555 | ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC || | |
556 | leafhdr.magic == XFS_DIR3_LEAFN_MAGIC); | |
557 | size = (int)((char *)&ents[leafhdr.count] - (char *)leaf); | |
88b32f06 | 558 | level = 0; |
8b4dc4a9 DC |
559 | |
560 | /* | |
561 | * we are about to copy oldroot to bp, so set up the type | |
562 | * of bp while we know exactly what it will be. | |
563 | */ | |
bdc16ee5 | 564 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF); |
2bd0ea18 | 565 | } |
88b32f06 DC |
566 | |
567 | /* | |
568 | * we can copy most of the information in the node from one block to | |
569 | * another, but for CRC enabled headers we have to make sure that the | |
570 | * block specific identifiers are kept intact. We update the buffer | |
571 | * directly for this. | |
572 | */ | |
32181a02 | 573 | memcpy(node, oldroot, size); |
88b32f06 DC |
574 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) || |
575 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
576 | struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node; | |
577 | ||
578 | node3->hdr.info.blkno = cpu_to_be64(bp->b_bn); | |
579 | } | |
a2ceac1f DC |
580 | xfs_trans_log_buf(tp, bp, 0, size - 1); |
581 | ||
582 | bp->b_ops = blk1->bp->b_ops; | |
e26915ee | 583 | xfs_trans_buf_copy_type(bp, blk1->bp); |
2bd0ea18 NS |
584 | blk1->bp = bp; |
585 | blk1->blkno = blkno; | |
586 | ||
587 | /* | |
588 | * Set up the new root node. | |
589 | */ | |
88b32f06 | 590 | error = xfs_da3_node_create(args, |
ff105f75 | 591 | (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0, |
88b32f06 | 592 | level + 1, &bp, args->whichfork); |
2bd0ea18 | 593 | if (error) |
88b32f06 DC |
594 | return error; |
595 | ||
a2ceac1f | 596 | node = bp->b_addr; |
ff105f75 DC |
597 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
598 | btree = dp->d_ops->node_tree_p(node); | |
88b32f06 DC |
599 | btree[0].hashval = cpu_to_be32(blk1->hashval); |
600 | btree[0].before = cpu_to_be32(blk1->blkno); | |
601 | btree[1].hashval = cpu_to_be32(blk2->hashval); | |
602 | btree[1].before = cpu_to_be32(blk2->blkno); | |
603 | nodehdr.count = 2; | |
ff105f75 | 604 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
6bef826c NS |
605 | |
606 | #ifdef DEBUG | |
65b80c98 DC |
607 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
608 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
ff105f75 DC |
609 | ASSERT(blk1->blkno >= args->geo->leafblk && |
610 | blk1->blkno < args->geo->freeblk); | |
611 | ASSERT(blk2->blkno >= args->geo->leafblk && | |
612 | blk2->blkno < args->geo->freeblk); | |
2bd0ea18 | 613 | } |
6bef826c NS |
614 | #endif |
615 | ||
2bd0ea18 | 616 | /* Header is already logged by xfs_da_node_create */ |
a2ceac1f | 617 | xfs_trans_log_buf(tp, bp, |
88b32f06 | 618 | XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2)); |
2bd0ea18 | 619 | |
88b32f06 | 620 | return 0; |
2bd0ea18 NS |
621 | } |
622 | ||
623 | /* | |
624 | * Split the node, rebalance, then add the new entry. | |
625 | */ | |
626 | STATIC int /* error */ | |
88b32f06 DC |
627 | xfs_da3_node_split( |
628 | struct xfs_da_state *state, | |
629 | struct xfs_da_state_blk *oldblk, | |
630 | struct xfs_da_state_blk *newblk, | |
631 | struct xfs_da_state_blk *addblk, | |
632 | int treelevel, | |
633 | int *result) | |
2bd0ea18 | 634 | { |
88b32f06 DC |
635 | struct xfs_da_intnode *node; |
636 | struct xfs_da3_icnode_hdr nodehdr; | |
637 | xfs_dablk_t blkno; | |
638 | int newcount; | |
639 | int error; | |
640 | int useextra; | |
ff105f75 | 641 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 642 | |
a2ceac1f DC |
643 | trace_xfs_da_node_split(state->args); |
644 | ||
645 | node = oldblk->bp->b_addr; | |
ff105f75 | 646 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
2bd0ea18 NS |
647 | |
648 | /* | |
5e656dbb | 649 | * With V2 dirs the extra block is data or freespace. |
2bd0ea18 | 650 | */ |
5e656dbb | 651 | useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK; |
2bd0ea18 NS |
652 | newcount = 1 + useextra; |
653 | /* | |
654 | * Do we have to split the node? | |
655 | */ | |
ff105f75 | 656 | if (nodehdr.count + newcount > state->args->geo->node_ents) { |
2bd0ea18 NS |
657 | /* |
658 | * Allocate a new node, add to the doubly linked chain of | |
659 | * nodes, then move some of our excess entries into it. | |
660 | */ | |
661 | error = xfs_da_grow_inode(state->args, &blkno); | |
662 | if (error) | |
af43ca9f | 663 | return error; /* GROT: dir is inconsistent */ |
5000d01d | 664 | |
88b32f06 | 665 | error = xfs_da3_node_create(state->args, blkno, treelevel, |
2bd0ea18 NS |
666 | &newblk->bp, state->args->whichfork); |
667 | if (error) | |
af43ca9f | 668 | return error; /* GROT: dir is inconsistent */ |
2bd0ea18 NS |
669 | newblk->blkno = blkno; |
670 | newblk->magic = XFS_DA_NODE_MAGIC; | |
88b32f06 DC |
671 | xfs_da3_node_rebalance(state, oldblk, newblk); |
672 | error = xfs_da3_blk_link(state, oldblk, newblk); | |
2bd0ea18 | 673 | if (error) |
af43ca9f | 674 | return error; |
2bd0ea18 NS |
675 | *result = 1; |
676 | } else { | |
677 | *result = 0; | |
678 | } | |
679 | ||
680 | /* | |
681 | * Insert the new entry(s) into the correct block | |
682 | * (updating last hashval in the process). | |
683 | * | |
88b32f06 | 684 | * xfs_da3_node_add() inserts BEFORE the given index, |
2bd0ea18 NS |
685 | * and as a result of using node_lookup_int() we always |
686 | * point to a valid entry (not after one), but a split | |
687 | * operation always results in a new block whose hashvals | |
688 | * FOLLOW the current block. | |
689 | * | |
690 | * If we had double-split op below us, then add the extra block too. | |
691 | */ | |
a2ceac1f | 692 | node = oldblk->bp->b_addr; |
ff105f75 | 693 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
88b32f06 | 694 | if (oldblk->index <= nodehdr.count) { |
2bd0ea18 | 695 | oldblk->index++; |
88b32f06 | 696 | xfs_da3_node_add(state, oldblk, addblk); |
2bd0ea18 NS |
697 | if (useextra) { |
698 | if (state->extraafter) | |
699 | oldblk->index++; | |
88b32f06 | 700 | xfs_da3_node_add(state, oldblk, &state->extrablk); |
2bd0ea18 NS |
701 | state->extravalid = 0; |
702 | } | |
703 | } else { | |
704 | newblk->index++; | |
88b32f06 | 705 | xfs_da3_node_add(state, newblk, addblk); |
2bd0ea18 NS |
706 | if (useextra) { |
707 | if (state->extraafter) | |
708 | newblk->index++; | |
88b32f06 | 709 | xfs_da3_node_add(state, newblk, &state->extrablk); |
2bd0ea18 NS |
710 | state->extravalid = 0; |
711 | } | |
712 | } | |
713 | ||
af43ca9f | 714 | return 0; |
2bd0ea18 NS |
715 | } |
716 | ||
717 | /* | |
718 | * Balance the btree elements between two intermediate nodes, | |
719 | * usually one full and one empty. | |
720 | * | |
721 | * NOTE: if blk2 is empty, then it will get the upper half of blk1. | |
722 | */ | |
723 | STATIC void | |
88b32f06 DC |
724 | xfs_da3_node_rebalance( |
725 | struct xfs_da_state *state, | |
726 | struct xfs_da_state_blk *blk1, | |
727 | struct xfs_da_state_blk *blk2) | |
2bd0ea18 | 728 | { |
88b32f06 DC |
729 | struct xfs_da_intnode *node1; |
730 | struct xfs_da_intnode *node2; | |
731 | struct xfs_da_intnode *tmpnode; | |
732 | struct xfs_da_node_entry *btree1; | |
733 | struct xfs_da_node_entry *btree2; | |
734 | struct xfs_da_node_entry *btree_s; | |
735 | struct xfs_da_node_entry *btree_d; | |
736 | struct xfs_da3_icnode_hdr nodehdr1; | |
737 | struct xfs_da3_icnode_hdr nodehdr2; | |
738 | struct xfs_trans *tp; | |
739 | int count; | |
740 | int tmp; | |
741 | int swap = 0; | |
ff105f75 | 742 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 743 | |
a2ceac1f DC |
744 | trace_xfs_da_node_rebalance(state->args); |
745 | ||
746 | node1 = blk1->bp->b_addr; | |
747 | node2 = blk2->bp->b_addr; | |
ff105f75 DC |
748 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
749 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); | |
750 | btree1 = dp->d_ops->node_tree_p(node1); | |
751 | btree2 = dp->d_ops->node_tree_p(node2); | |
88b32f06 | 752 | |
2bd0ea18 NS |
753 | /* |
754 | * Figure out how many entries need to move, and in which direction. | |
755 | * Swap the nodes around if that makes it simpler. | |
756 | */ | |
88b32f06 DC |
757 | if (nodehdr1.count > 0 && nodehdr2.count > 0 && |
758 | ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) || | |
759 | (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) < | |
760 | be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) { | |
2bd0ea18 NS |
761 | tmpnode = node1; |
762 | node1 = node2; | |
763 | node2 = tmpnode; | |
ff105f75 DC |
764 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
765 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); | |
766 | btree1 = dp->d_ops->node_tree_p(node1); | |
767 | btree2 = dp->d_ops->node_tree_p(node2); | |
88b32f06 | 768 | swap = 1; |
2bd0ea18 | 769 | } |
88b32f06 DC |
770 | |
771 | count = (nodehdr1.count - nodehdr2.count) / 2; | |
2bd0ea18 NS |
772 | if (count == 0) |
773 | return; | |
774 | tp = state->args->trans; | |
775 | /* | |
776 | * Two cases: high-to-low and low-to-high. | |
777 | */ | |
778 | if (count > 0) { | |
779 | /* | |
780 | * Move elements in node2 up to make a hole. | |
781 | */ | |
88b32f06 DC |
782 | tmp = nodehdr2.count; |
783 | if (tmp > 0) { | |
2bd0ea18 | 784 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
88b32f06 DC |
785 | btree_s = &btree2[0]; |
786 | btree_d = &btree2[count]; | |
32181a02 | 787 | memmove(btree_d, btree_s, tmp); |
2bd0ea18 NS |
788 | } |
789 | ||
790 | /* | |
791 | * Move the req'd B-tree elements from high in node1 to | |
792 | * low in node2. | |
793 | */ | |
88b32f06 | 794 | nodehdr2.count += count; |
2bd0ea18 | 795 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
7b02556d | 796 | btree_s = &btree1[nodehdr1.count - count]; |
88b32f06 | 797 | btree_d = &btree2[0]; |
32181a02 | 798 | memcpy(btree_d, btree_s, tmp); |
88b32f06 | 799 | nodehdr1.count -= count; |
2bd0ea18 NS |
800 | } else { |
801 | /* | |
802 | * Move the req'd B-tree elements from low in node2 to | |
803 | * high in node1. | |
804 | */ | |
805 | count = -count; | |
806 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); | |
88b32f06 DC |
807 | btree_s = &btree2[0]; |
808 | btree_d = &btree1[nodehdr1.count]; | |
32181a02 | 809 | memcpy(btree_d, btree_s, tmp); |
88b32f06 DC |
810 | nodehdr1.count += count; |
811 | ||
a2ceac1f | 812 | xfs_trans_log_buf(tp, blk1->bp, |
2bd0ea18 NS |
813 | XFS_DA_LOGRANGE(node1, btree_d, tmp)); |
814 | ||
815 | /* | |
816 | * Move elements in node2 down to fill the hole. | |
817 | */ | |
88b32f06 | 818 | tmp = nodehdr2.count - count; |
2bd0ea18 | 819 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
88b32f06 DC |
820 | btree_s = &btree2[count]; |
821 | btree_d = &btree2[0]; | |
32181a02 | 822 | memmove(btree_d, btree_s, tmp); |
88b32f06 | 823 | nodehdr2.count -= count; |
2bd0ea18 NS |
824 | } |
825 | ||
826 | /* | |
827 | * Log header of node 1 and all current bits of node 2. | |
828 | */ | |
ff105f75 | 829 | dp->d_ops->node_hdr_to_disk(node1, &nodehdr1); |
a2ceac1f | 830 | xfs_trans_log_buf(tp, blk1->bp, |
ff105f75 | 831 | XFS_DA_LOGRANGE(node1, &node1->hdr, dp->d_ops->node_hdr_size)); |
88b32f06 | 832 | |
ff105f75 | 833 | dp->d_ops->node_hdr_to_disk(node2, &nodehdr2); |
a2ceac1f | 834 | xfs_trans_log_buf(tp, blk2->bp, |
2bd0ea18 | 835 | XFS_DA_LOGRANGE(node2, &node2->hdr, |
ff105f75 | 836 | dp->d_ops->node_hdr_size + |
88b32f06 | 837 | (sizeof(btree2[0]) * nodehdr2.count))); |
2bd0ea18 NS |
838 | |
839 | /* | |
840 | * Record the last hashval from each block for upward propagation. | |
841 | * (note: don't use the swapped node pointers) | |
842 | */ | |
88b32f06 DC |
843 | if (swap) { |
844 | node1 = blk1->bp->b_addr; | |
845 | node2 = blk2->bp->b_addr; | |
ff105f75 DC |
846 | dp->d_ops->node_hdr_from_disk(&nodehdr1, node1); |
847 | dp->d_ops->node_hdr_from_disk(&nodehdr2, node2); | |
848 | btree1 = dp->d_ops->node_tree_p(node1); | |
849 | btree2 = dp->d_ops->node_tree_p(node2); | |
88b32f06 DC |
850 | } |
851 | blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval); | |
852 | blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval); | |
2bd0ea18 NS |
853 | |
854 | /* | |
855 | * Adjust the expected index for insertion. | |
856 | */ | |
88b32f06 DC |
857 | if (blk1->index >= nodehdr1.count) { |
858 | blk2->index = blk1->index - nodehdr1.count; | |
859 | blk1->index = nodehdr1.count + 1; /* make it invalid */ | |
2bd0ea18 NS |
860 | } |
861 | } | |
862 | ||
863 | /* | |
864 | * Add a new entry to an intermediate node. | |
865 | */ | |
866 | STATIC void | |
88b32f06 DC |
867 | xfs_da3_node_add( |
868 | struct xfs_da_state *state, | |
869 | struct xfs_da_state_blk *oldblk, | |
870 | struct xfs_da_state_blk *newblk) | |
2bd0ea18 | 871 | { |
88b32f06 DC |
872 | struct xfs_da_intnode *node; |
873 | struct xfs_da3_icnode_hdr nodehdr; | |
874 | struct xfs_da_node_entry *btree; | |
875 | int tmp; | |
ff105f75 | 876 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 877 | |
a2ceac1f DC |
878 | trace_xfs_da_node_add(state->args); |
879 | ||
880 | node = oldblk->bp->b_addr; | |
ff105f75 DC |
881 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
882 | btree = dp->d_ops->node_tree_p(node); | |
88b32f06 DC |
883 | |
884 | ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count); | |
2bd0ea18 | 885 | ASSERT(newblk->blkno != 0); |
5e656dbb | 886 | if (state->args->whichfork == XFS_DATA_FORK) |
ff105f75 DC |
887 | ASSERT(newblk->blkno >= state->args->geo->leafblk && |
888 | newblk->blkno < state->args->geo->freeblk); | |
2bd0ea18 NS |
889 | |
890 | /* | |
891 | * We may need to make some room before we insert the new node. | |
892 | */ | |
893 | tmp = 0; | |
88b32f06 DC |
894 | if (oldblk->index < nodehdr.count) { |
895 | tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree); | |
896 | memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp); | |
2bd0ea18 | 897 | } |
88b32f06 DC |
898 | btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval); |
899 | btree[oldblk->index].before = cpu_to_be32(newblk->blkno); | |
a2ceac1f | 900 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
88b32f06 DC |
901 | XFS_DA_LOGRANGE(node, &btree[oldblk->index], |
902 | tmp + sizeof(*btree))); | |
903 | ||
904 | nodehdr.count += 1; | |
ff105f75 | 905 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
a2ceac1f | 906 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
ff105f75 | 907 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
2bd0ea18 NS |
908 | |
909 | /* | |
910 | * Copy the last hash value from the oldblk to propagate upwards. | |
911 | */ | |
88b32f06 | 912 | oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
2bd0ea18 NS |
913 | } |
914 | ||
915 | /*======================================================================== | |
916 | * Routines used for shrinking the Btree. | |
917 | *========================================================================*/ | |
918 | ||
919 | /* | |
920 | * Deallocate an empty leaf node, remove it from its parent, | |
921 | * possibly deallocating that block, etc... | |
922 | */ | |
923 | int | |
88b32f06 DC |
924 | xfs_da3_join( |
925 | struct xfs_da_state *state) | |
2bd0ea18 | 926 | { |
88b32f06 DC |
927 | struct xfs_da_state_blk *drop_blk; |
928 | struct xfs_da_state_blk *save_blk; | |
929 | int action = 0; | |
930 | int error; | |
2bd0ea18 | 931 | |
a2ceac1f DC |
932 | trace_xfs_da_join(state->args); |
933 | ||
2bd0ea18 NS |
934 | drop_blk = &state->path.blk[ state->path.active-1 ]; |
935 | save_blk = &state->altpath.blk[ state->path.active-1 ]; | |
936 | ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC); | |
937 | ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC || | |
5e656dbb | 938 | drop_blk->magic == XFS_DIR2_LEAFN_MAGIC); |
2bd0ea18 NS |
939 | |
940 | /* | |
941 | * Walk back up the tree joining/deallocating as necessary. | |
942 | * When we stop dropping blocks, break out. | |
943 | */ | |
944 | for ( ; state->path.active >= 2; drop_blk--, save_blk--, | |
945 | state->path.active--) { | |
946 | /* | |
947 | * See if we can combine the block with a neighbor. | |
948 | * (action == 0) => no options, just leave | |
949 | * (action == 1) => coalesce, then unlink | |
950 | * (action == 2) => block empty, unlink it | |
951 | */ | |
952 | switch (drop_blk->magic) { | |
953 | case XFS_ATTR_LEAF_MAGIC: | |
a24374f4 | 954 | error = xfs_attr3_leaf_toosmall(state, &action); |
2bd0ea18 | 955 | if (error) |
af43ca9f | 956 | return error; |
2bd0ea18 | 957 | if (action == 0) |
af43ca9f | 958 | return 0; |
a24374f4 | 959 | xfs_attr3_leaf_unbalance(state, drop_blk, save_blk); |
2bd0ea18 | 960 | break; |
2bd0ea18 | 961 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
962 | error = xfs_dir2_leafn_toosmall(state, &action); |
963 | if (error) | |
964 | return error; | |
965 | if (action == 0) | |
966 | return 0; | |
967 | xfs_dir2_leafn_unbalance(state, drop_blk, save_blk); | |
968 | break; | |
969 | case XFS_DA_NODE_MAGIC: | |
970 | /* | |
971 | * Remove the offending node, fixup hashvals, | |
972 | * check for a toosmall neighbor. | |
973 | */ | |
88b32f06 DC |
974 | xfs_da3_node_remove(state, drop_blk); |
975 | xfs_da3_fixhashpath(state, &state->path); | |
976 | error = xfs_da3_node_toosmall(state, &action); | |
2bd0ea18 | 977 | if (error) |
af43ca9f | 978 | return error; |
2bd0ea18 NS |
979 | if (action == 0) |
980 | return 0; | |
88b32f06 | 981 | xfs_da3_node_unbalance(state, drop_blk, save_blk); |
2bd0ea18 NS |
982 | break; |
983 | } | |
88b32f06 DC |
984 | xfs_da3_fixhashpath(state, &state->altpath); |
985 | error = xfs_da3_blk_unlink(state, drop_blk, save_blk); | |
2bd0ea18 NS |
986 | xfs_da_state_kill_altpath(state); |
987 | if (error) | |
af43ca9f | 988 | return error; |
2bd0ea18 NS |
989 | error = xfs_da_shrink_inode(state->args, drop_blk->blkno, |
990 | drop_blk->bp); | |
991 | drop_blk->bp = NULL; | |
992 | if (error) | |
af43ca9f | 993 | return error; |
2bd0ea18 NS |
994 | } |
995 | /* | |
996 | * We joined all the way to the top. If it turns out that | |
997 | * we only have one entry in the root, make the child block | |
998 | * the new root. | |
999 | */ | |
88b32f06 DC |
1000 | xfs_da3_node_remove(state, drop_blk); |
1001 | xfs_da3_fixhashpath(state, &state->path); | |
1002 | error = xfs_da3_root_join(state, &state->path.blk[0]); | |
af43ca9f | 1003 | return error; |
2bd0ea18 NS |
1004 | } |
1005 | ||
a2ceac1f DC |
1006 | #ifdef DEBUG |
1007 | static void | |
1008 | xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level) | |
1009 | { | |
1010 | __be16 magic = blkinfo->magic; | |
1011 | ||
1012 | if (level == 1) { | |
1013 | ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
65b80c98 | 1014 | magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
a24374f4 DC |
1015 | magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || |
1016 | magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); | |
88b32f06 DC |
1017 | } else { |
1018 | ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
1019 | magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)); | |
1020 | } | |
a2ceac1f DC |
1021 | ASSERT(!blkinfo->forw); |
1022 | ASSERT(!blkinfo->back); | |
1023 | } | |
1024 | #else /* !DEBUG */ | |
1025 | #define xfs_da_blkinfo_onlychild_validate(blkinfo, level) | |
1026 | #endif /* !DEBUG */ | |
1027 | ||
2bd0ea18 | 1028 | /* |
dfc130f3 | 1029 | * We have only one entry in the root. Copy the only remaining child of |
2bd0ea18 NS |
1030 | * the old root to block 0 as the new root node. |
1031 | */ | |
1032 | STATIC int | |
88b32f06 DC |
1033 | xfs_da3_root_join( |
1034 | struct xfs_da_state *state, | |
1035 | struct xfs_da_state_blk *root_blk) | |
2bd0ea18 | 1036 | { |
88b32f06 DC |
1037 | struct xfs_da_intnode *oldroot; |
1038 | struct xfs_da_args *args; | |
1039 | xfs_dablk_t child; | |
1040 | struct xfs_buf *bp; | |
1041 | struct xfs_da3_icnode_hdr oldroothdr; | |
1042 | struct xfs_da_node_entry *btree; | |
1043 | int error; | |
ff105f75 | 1044 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1045 | |
a2ceac1f DC |
1046 | trace_xfs_da_root_join(state->args); |
1047 | ||
2bd0ea18 | 1048 | ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC); |
88b32f06 DC |
1049 | |
1050 | args = state->args; | |
a2ceac1f | 1051 | oldroot = root_blk->bp->b_addr; |
ff105f75 | 1052 | dp->d_ops->node_hdr_from_disk(&oldroothdr, oldroot); |
88b32f06 DC |
1053 | ASSERT(oldroothdr.forw == 0); |
1054 | ASSERT(oldroothdr.back == 0); | |
2bd0ea18 NS |
1055 | |
1056 | /* | |
1057 | * If the root has more than one child, then don't do anything. | |
1058 | */ | |
88b32f06 DC |
1059 | if (oldroothdr.count > 1) |
1060 | return 0; | |
2bd0ea18 NS |
1061 | |
1062 | /* | |
1063 | * Read in the (only) child block, then copy those bytes into | |
1064 | * the root block's buffer and free the original child block. | |
1065 | */ | |
ff105f75 | 1066 | btree = dp->d_ops->node_tree_p(oldroot); |
88b32f06 | 1067 | child = be32_to_cpu(btree[0].before); |
2bd0ea18 | 1068 | ASSERT(child != 0); |
ff105f75 | 1069 | error = xfs_da3_node_read(args->trans, dp, child, -1, &bp, |
2bd0ea18 NS |
1070 | args->whichfork); |
1071 | if (error) | |
88b32f06 DC |
1072 | return error; |
1073 | xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level); | |
a2ceac1f DC |
1074 | |
1075 | /* | |
1076 | * This could be copying a leaf back into the root block in the case of | |
1077 | * there only being a single leaf block left in the tree. Hence we have | |
1078 | * to update the b_ops pointer as well to match the buffer type change | |
88b32f06 DC |
1079 | * that could occur. For dir3 blocks we also need to update the block |
1080 | * number in the buffer header. | |
a2ceac1f | 1081 | */ |
ff105f75 | 1082 | memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize); |
a2ceac1f | 1083 | root_blk->bp->b_ops = bp->b_ops; |
8b4dc4a9 | 1084 | xfs_trans_buf_copy_type(root_blk->bp, bp); |
88b32f06 DC |
1085 | if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) { |
1086 | struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr; | |
1087 | da3->blkno = cpu_to_be64(root_blk->bp->b_bn); | |
1088 | } | |
ff105f75 DC |
1089 | xfs_trans_log_buf(args->trans, root_blk->bp, 0, |
1090 | args->geo->blksize - 1); | |
2bd0ea18 | 1091 | error = xfs_da_shrink_inode(args, child, bp); |
af43ca9f | 1092 | return error; |
2bd0ea18 NS |
1093 | } |
1094 | ||
1095 | /* | |
1096 | * Check a node block and its neighbors to see if the block should be | |
1097 | * collapsed into one or the other neighbor. Always keep the block | |
1098 | * with the smaller block number. | |
1099 | * If the current block is over 50% full, don't try to join it, return 0. | |
1100 | * If the block is empty, fill in the state structure and return 2. | |
1101 | * If it can be collapsed, fill in the state structure and return 1. | |
1102 | * If nothing can be done, return 0. | |
1103 | */ | |
1104 | STATIC int | |
88b32f06 DC |
1105 | xfs_da3_node_toosmall( |
1106 | struct xfs_da_state *state, | |
1107 | int *action) | |
2bd0ea18 | 1108 | { |
88b32f06 DC |
1109 | struct xfs_da_intnode *node; |
1110 | struct xfs_da_state_blk *blk; | |
1111 | struct xfs_da_blkinfo *info; | |
1112 | xfs_dablk_t blkno; | |
1113 | struct xfs_buf *bp; | |
1114 | struct xfs_da3_icnode_hdr nodehdr; | |
1115 | int count; | |
1116 | int forward; | |
1117 | int error; | |
1118 | int retval; | |
1119 | int i; | |
ff105f75 | 1120 | struct xfs_inode *dp = state->args->dp; |
a2ceac1f DC |
1121 | |
1122 | trace_xfs_da_node_toosmall(state->args); | |
2bd0ea18 NS |
1123 | |
1124 | /* | |
1125 | * Check for the degenerate case of the block being over 50% full. | |
1126 | * If so, it's not worth even looking to see if we might be able | |
1127 | * to coalesce with a sibling. | |
1128 | */ | |
1129 | blk = &state->path.blk[ state->path.active-1 ]; | |
a2ceac1f | 1130 | info = blk->bp->b_addr; |
2bd0ea18 | 1131 | node = (xfs_da_intnode_t *)info; |
ff105f75 DC |
1132 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
1133 | if (nodehdr.count > (state->args->geo->node_ents >> 1)) { | |
4ca431fc | 1134 | *action = 0; /* blk over 50%, don't try to join */ |
af43ca9f | 1135 | return 0; /* blk over 50%, don't try to join */ |
2bd0ea18 NS |
1136 | } |
1137 | ||
1138 | /* | |
1139 | * Check for the degenerate case of the block being empty. | |
1140 | * If the block is empty, we'll simply delete it, no need to | |
5e656dbb | 1141 | * coalesce it with a sibling block. We choose (arbitrarily) |
2bd0ea18 NS |
1142 | * to merge with the forward block unless it is NULL. |
1143 | */ | |
88b32f06 | 1144 | if (nodehdr.count == 0) { |
2bd0ea18 NS |
1145 | /* |
1146 | * Make altpath point to the block we want to keep and | |
1147 | * path point to the block we want to drop (this one). | |
1148 | */ | |
5e656dbb | 1149 | forward = (info->forw != 0); |
32181a02 | 1150 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
88b32f06 | 1151 | error = xfs_da3_path_shift(state, &state->altpath, forward, |
2bd0ea18 NS |
1152 | 0, &retval); |
1153 | if (error) | |
af43ca9f | 1154 | return error; |
2bd0ea18 NS |
1155 | if (retval) { |
1156 | *action = 0; | |
1157 | } else { | |
1158 | *action = 2; | |
1159 | } | |
af43ca9f | 1160 | return 0; |
2bd0ea18 NS |
1161 | } |
1162 | ||
1163 | /* | |
1164 | * Examine each sibling block to see if we can coalesce with | |
1165 | * at least 25% free space to spare. We need to figure out | |
1166 | * whether to merge with the forward or the backward block. | |
1167 | * We prefer coalescing with the lower numbered sibling so as | |
1168 | * to shrink a directory over time. | |
1169 | */ | |
ff105f75 DC |
1170 | count = state->args->geo->node_ents; |
1171 | count -= state->args->geo->node_ents >> 2; | |
88b32f06 DC |
1172 | count -= nodehdr.count; |
1173 | ||
2bd0ea18 | 1174 | /* start with smaller blk num */ |
88b32f06 | 1175 | forward = nodehdr.forw < nodehdr.back; |
2bd0ea18 | 1176 | for (i = 0; i < 2; forward = !forward, i++) { |
c9522f4d | 1177 | struct xfs_da3_icnode_hdr thdr; |
2bd0ea18 | 1178 | if (forward) |
88b32f06 | 1179 | blkno = nodehdr.forw; |
2bd0ea18 | 1180 | else |
88b32f06 | 1181 | blkno = nodehdr.back; |
2bd0ea18 NS |
1182 | if (blkno == 0) |
1183 | continue; | |
ff105f75 | 1184 | error = xfs_da3_node_read(state->args->trans, dp, |
2bd0ea18 NS |
1185 | blkno, -1, &bp, state->args->whichfork); |
1186 | if (error) | |
af43ca9f | 1187 | return error; |
2bd0ea18 | 1188 | |
a2ceac1f | 1189 | node = bp->b_addr; |
ff105f75 | 1190 | dp->d_ops->node_hdr_from_disk(&thdr, node); |
a2ceac1f | 1191 | xfs_trans_brelse(state->args->trans, bp); |
88b32f06 | 1192 | |
c9522f4d | 1193 | if (count - thdr.count >= 0) |
2bd0ea18 NS |
1194 | break; /* fits with at least 25% to spare */ |
1195 | } | |
1196 | if (i >= 2) { | |
1197 | *action = 0; | |
88b32f06 | 1198 | return 0; |
2bd0ea18 NS |
1199 | } |
1200 | ||
1201 | /* | |
1202 | * Make altpath point to the block we want to keep (the lower | |
1203 | * numbered block) and path point to the block we want to drop. | |
1204 | */ | |
32181a02 | 1205 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
2bd0ea18 | 1206 | if (blkno < blk->blkno) { |
88b32f06 | 1207 | error = xfs_da3_path_shift(state, &state->altpath, forward, |
2bd0ea18 | 1208 | 0, &retval); |
2bd0ea18 | 1209 | } else { |
88b32f06 | 1210 | error = xfs_da3_path_shift(state, &state->path, forward, |
2bd0ea18 | 1211 | 0, &retval); |
88b32f06 DC |
1212 | } |
1213 | if (error) | |
1214 | return error; | |
1215 | if (retval) { | |
1216 | *action = 0; | |
1217 | return 0; | |
2bd0ea18 NS |
1218 | } |
1219 | *action = 1; | |
88b32f06 DC |
1220 | return 0; |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * Pick up the last hashvalue from an intermediate node. | |
1225 | */ | |
1226 | STATIC uint | |
1227 | xfs_da3_node_lasthash( | |
ff105f75 | 1228 | struct xfs_inode *dp, |
88b32f06 DC |
1229 | struct xfs_buf *bp, |
1230 | int *count) | |
1231 | { | |
1232 | struct xfs_da_intnode *node; | |
1233 | struct xfs_da_node_entry *btree; | |
1234 | struct xfs_da3_icnode_hdr nodehdr; | |
1235 | ||
1236 | node = bp->b_addr; | |
ff105f75 | 1237 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
88b32f06 DC |
1238 | if (count) |
1239 | *count = nodehdr.count; | |
1240 | if (!nodehdr.count) | |
1241 | return 0; | |
ff105f75 | 1242 | btree = dp->d_ops->node_tree_p(node); |
88b32f06 | 1243 | return be32_to_cpu(btree[nodehdr.count - 1].hashval); |
2bd0ea18 NS |
1244 | } |
1245 | ||
2bd0ea18 NS |
1246 | /* |
1247 | * Walk back up the tree adjusting hash values as necessary, | |
1248 | * when we stop making changes, return. | |
1249 | */ | |
1250 | void | |
88b32f06 DC |
1251 | xfs_da3_fixhashpath( |
1252 | struct xfs_da_state *state, | |
1253 | struct xfs_da_state_path *path) | |
2bd0ea18 | 1254 | { |
88b32f06 DC |
1255 | struct xfs_da_state_blk *blk; |
1256 | struct xfs_da_intnode *node; | |
1257 | struct xfs_da_node_entry *btree; | |
1258 | xfs_dahash_t lasthash=0; | |
1259 | int level; | |
1260 | int count; | |
ff105f75 | 1261 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1262 | |
a2ceac1f DC |
1263 | trace_xfs_da_fixhashpath(state->args); |
1264 | ||
2bd0ea18 NS |
1265 | level = path->active-1; |
1266 | blk = &path->blk[ level ]; | |
1267 | switch (blk->magic) { | |
2bd0ea18 NS |
1268 | case XFS_ATTR_LEAF_MAGIC: |
1269 | lasthash = xfs_attr_leaf_lasthash(blk->bp, &count); | |
1270 | if (count == 0) | |
1271 | return; | |
1272 | break; | |
2bd0ea18 | 1273 | case XFS_DIR2_LEAFN_MAGIC: |
94ffb80a | 1274 | lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count); |
2bd0ea18 NS |
1275 | if (count == 0) |
1276 | return; | |
1277 | break; | |
1278 | case XFS_DA_NODE_MAGIC: | |
ff105f75 | 1279 | lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count); |
2bd0ea18 NS |
1280 | if (count == 0) |
1281 | return; | |
1282 | break; | |
1283 | } | |
1284 | for (blk--, level--; level >= 0; blk--, level--) { | |
88b32f06 DC |
1285 | struct xfs_da3_icnode_hdr nodehdr; |
1286 | ||
a2ceac1f | 1287 | node = blk->bp->b_addr; |
ff105f75 DC |
1288 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
1289 | btree = dp->d_ops->node_tree_p(node); | |
0a3717c9 | 1290 | if (be32_to_cpu(btree[blk->index].hashval) == lasthash) |
2bd0ea18 NS |
1291 | break; |
1292 | blk->hashval = lasthash; | |
88b32f06 | 1293 | btree[blk->index].hashval = cpu_to_be32(lasthash); |
a2ceac1f | 1294 | xfs_trans_log_buf(state->args->trans, blk->bp, |
88b32f06 DC |
1295 | XFS_DA_LOGRANGE(node, &btree[blk->index], |
1296 | sizeof(*btree))); | |
2bd0ea18 | 1297 | |
88b32f06 | 1298 | lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
2bd0ea18 NS |
1299 | } |
1300 | } | |
1301 | ||
2bd0ea18 NS |
1302 | /* |
1303 | * Remove an entry from an intermediate node. | |
1304 | */ | |
1305 | STATIC void | |
88b32f06 DC |
1306 | xfs_da3_node_remove( |
1307 | struct xfs_da_state *state, | |
1308 | struct xfs_da_state_blk *drop_blk) | |
2bd0ea18 | 1309 | { |
88b32f06 DC |
1310 | struct xfs_da_intnode *node; |
1311 | struct xfs_da3_icnode_hdr nodehdr; | |
1312 | struct xfs_da_node_entry *btree; | |
1313 | int index; | |
1314 | int tmp; | |
ff105f75 | 1315 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1316 | |
a2ceac1f DC |
1317 | trace_xfs_da_node_remove(state->args); |
1318 | ||
1319 | node = drop_blk->bp->b_addr; | |
ff105f75 | 1320 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
88b32f06 | 1321 | ASSERT(drop_blk->index < nodehdr.count); |
2bd0ea18 NS |
1322 | ASSERT(drop_blk->index >= 0); |
1323 | ||
1324 | /* | |
1325 | * Copy over the offending entry, or just zero it out. | |
1326 | */ | |
88b32f06 | 1327 | index = drop_blk->index; |
ff105f75 | 1328 | btree = dp->d_ops->node_tree_p(node); |
88b32f06 DC |
1329 | if (index < nodehdr.count - 1) { |
1330 | tmp = nodehdr.count - index - 1; | |
2bd0ea18 | 1331 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
88b32f06 | 1332 | memmove(&btree[index], &btree[index + 1], tmp); |
a2ceac1f | 1333 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
88b32f06 DC |
1334 | XFS_DA_LOGRANGE(node, &btree[index], tmp)); |
1335 | index = nodehdr.count - 1; | |
2bd0ea18 | 1336 | } |
88b32f06 | 1337 | memset(&btree[index], 0, sizeof(xfs_da_node_entry_t)); |
a2ceac1f | 1338 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
88b32f06 DC |
1339 | XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index]))); |
1340 | nodehdr.count -= 1; | |
ff105f75 | 1341 | dp->d_ops->node_hdr_to_disk(node, &nodehdr); |
a2ceac1f | 1342 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
ff105f75 | 1343 | XFS_DA_LOGRANGE(node, &node->hdr, dp->d_ops->node_hdr_size)); |
2bd0ea18 NS |
1344 | |
1345 | /* | |
1346 | * Copy the last hash value from the block to propagate upwards. | |
1347 | */ | |
88b32f06 | 1348 | drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval); |
2bd0ea18 NS |
1349 | } |
1350 | ||
1351 | /* | |
88b32f06 | 1352 | * Unbalance the elements between two intermediate nodes, |
2bd0ea18 NS |
1353 | * move all Btree elements from one node into another. |
1354 | */ | |
1355 | STATIC void | |
88b32f06 DC |
1356 | xfs_da3_node_unbalance( |
1357 | struct xfs_da_state *state, | |
1358 | struct xfs_da_state_blk *drop_blk, | |
1359 | struct xfs_da_state_blk *save_blk) | |
2bd0ea18 | 1360 | { |
88b32f06 DC |
1361 | struct xfs_da_intnode *drop_node; |
1362 | struct xfs_da_intnode *save_node; | |
7b02556d DC |
1363 | struct xfs_da_node_entry *drop_btree; |
1364 | struct xfs_da_node_entry *save_btree; | |
1365 | struct xfs_da3_icnode_hdr drop_hdr; | |
1366 | struct xfs_da3_icnode_hdr save_hdr; | |
88b32f06 DC |
1367 | struct xfs_trans *tp; |
1368 | int sindex; | |
1369 | int tmp; | |
ff105f75 | 1370 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1371 | |
a2ceac1f DC |
1372 | trace_xfs_da_node_unbalance(state->args); |
1373 | ||
1374 | drop_node = drop_blk->bp->b_addr; | |
1375 | save_node = save_blk->bp->b_addr; | |
ff105f75 DC |
1376 | dp->d_ops->node_hdr_from_disk(&drop_hdr, drop_node); |
1377 | dp->d_ops->node_hdr_from_disk(&save_hdr, save_node); | |
1378 | drop_btree = dp->d_ops->node_tree_p(drop_node); | |
1379 | save_btree = dp->d_ops->node_tree_p(save_node); | |
2bd0ea18 NS |
1380 | tp = state->args->trans; |
1381 | ||
1382 | /* | |
1383 | * If the dying block has lower hashvals, then move all the | |
1384 | * elements in the remaining block up to make a hole. | |
1385 | */ | |
7b02556d DC |
1386 | if ((be32_to_cpu(drop_btree[0].hashval) < |
1387 | be32_to_cpu(save_btree[0].hashval)) || | |
1388 | (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) < | |
1389 | be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) { | |
88b32f06 | 1390 | /* XXX: check this - is memmove dst correct? */ |
7b02556d DC |
1391 | tmp = save_hdr.count * sizeof(xfs_da_node_entry_t); |
1392 | memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp); | |
88b32f06 DC |
1393 | |
1394 | sindex = 0; | |
a2ceac1f | 1395 | xfs_trans_log_buf(tp, save_blk->bp, |
7b02556d DC |
1396 | XFS_DA_LOGRANGE(save_node, &save_btree[0], |
1397 | (save_hdr.count + drop_hdr.count) * | |
88b32f06 | 1398 | sizeof(xfs_da_node_entry_t))); |
2bd0ea18 | 1399 | } else { |
7b02556d | 1400 | sindex = save_hdr.count; |
a2ceac1f | 1401 | xfs_trans_log_buf(tp, save_blk->bp, |
7b02556d DC |
1402 | XFS_DA_LOGRANGE(save_node, &save_btree[sindex], |
1403 | drop_hdr.count * sizeof(xfs_da_node_entry_t))); | |
2bd0ea18 NS |
1404 | } |
1405 | ||
1406 | /* | |
1407 | * Move all the B-tree elements from drop_blk to save_blk. | |
1408 | */ | |
7b02556d DC |
1409 | tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t); |
1410 | memcpy(&save_btree[sindex], &drop_btree[0], tmp); | |
1411 | save_hdr.count += drop_hdr.count; | |
2bd0ea18 | 1412 | |
ff105f75 | 1413 | dp->d_ops->node_hdr_to_disk(save_node, &save_hdr); |
a2ceac1f | 1414 | xfs_trans_log_buf(tp, save_blk->bp, |
2bd0ea18 | 1415 | XFS_DA_LOGRANGE(save_node, &save_node->hdr, |
ff105f75 | 1416 | dp->d_ops->node_hdr_size)); |
2bd0ea18 NS |
1417 | |
1418 | /* | |
1419 | * Save the last hashval in the remaining block for upward propagation. | |
1420 | */ | |
7b02556d | 1421 | save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval); |
2bd0ea18 NS |
1422 | } |
1423 | ||
2bd0ea18 NS |
1424 | /*======================================================================== |
1425 | * Routines used for finding things in the Btree. | |
1426 | *========================================================================*/ | |
1427 | ||
1428 | /* | |
1429 | * Walk down the Btree looking for a particular filename, filling | |
1430 | * in the state structure as we go. | |
1431 | * | |
1432 | * We will set the state structure to point to each of the elements | |
1433 | * in each of the nodes where either the hashval is or should be. | |
1434 | * | |
1435 | * We support duplicate hashval's so for each entry in the current | |
1436 | * node that could contain the desired hashval, descend. This is a | |
1437 | * pruned depth-first tree search. | |
1438 | */ | |
1439 | int /* error */ | |
88b32f06 DC |
1440 | xfs_da3_node_lookup_int( |
1441 | struct xfs_da_state *state, | |
1442 | int *result) | |
2bd0ea18 | 1443 | { |
88b32f06 DC |
1444 | struct xfs_da_state_blk *blk; |
1445 | struct xfs_da_blkinfo *curr; | |
1446 | struct xfs_da_intnode *node; | |
1447 | struct xfs_da_node_entry *btree; | |
1448 | struct xfs_da3_icnode_hdr nodehdr; | |
1449 | struct xfs_da_args *args; | |
1450 | xfs_dablk_t blkno; | |
1451 | xfs_dahash_t hashval; | |
1452 | xfs_dahash_t btreehashval; | |
1453 | int probe; | |
1454 | int span; | |
1455 | int max; | |
1456 | int error; | |
1457 | int retval; | |
9473ee32 | 1458 | unsigned int expected_level = 0; |
ff105f75 | 1459 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 NS |
1460 | |
1461 | args = state->args; | |
32a82561 | 1462 | |
2bd0ea18 NS |
1463 | /* |
1464 | * Descend thru the B-tree searching each level for the right | |
1465 | * node to use, until the right hashval is found. | |
1466 | */ | |
9473ee32 | 1467 | blkno = args->geo->leafblk; |
2bd0ea18 NS |
1468 | for (blk = &state->path.blk[0], state->path.active = 1; |
1469 | state->path.active <= XFS_DA_NODE_MAXDEPTH; | |
1470 | blk++, state->path.active++) { | |
1471 | /* | |
1472 | * Read the next node down in the tree. | |
1473 | */ | |
1474 | blk->blkno = blkno; | |
88b32f06 | 1475 | error = xfs_da3_node_read(args->trans, args->dp, blkno, |
32a82561 | 1476 | -1, &blk->bp, args->whichfork); |
2bd0ea18 NS |
1477 | if (error) { |
1478 | blk->blkno = 0; | |
1479 | state->path.active--; | |
af43ca9f | 1480 | return error; |
2bd0ea18 | 1481 | } |
a2ceac1f | 1482 | curr = blk->bp->b_addr; |
5e656dbb | 1483 | blk->magic = be16_to_cpu(curr->magic); |
88b32f06 | 1484 | |
a24374f4 DC |
1485 | if (blk->magic == XFS_ATTR_LEAF_MAGIC || |
1486 | blk->magic == XFS_ATTR3_LEAF_MAGIC) { | |
1487 | blk->magic = XFS_ATTR_LEAF_MAGIC; | |
88b32f06 DC |
1488 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
1489 | break; | |
1490 | } | |
1491 | ||
1492 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC || | |
1493 | blk->magic == XFS_DIR3_LEAFN_MAGIC) { | |
1494 | blk->magic = XFS_DIR2_LEAFN_MAGIC; | |
94ffb80a DW |
1495 | blk->hashval = xfs_dir2_leaf_lasthash(args->dp, |
1496 | blk->bp, NULL); | |
88b32f06 DC |
1497 | break; |
1498 | } | |
1499 | ||
1500 | blk->magic = XFS_DA_NODE_MAGIC; | |
1501 | ||
2bd0ea18 NS |
1502 | |
1503 | /* | |
1504 | * Search an intermediate node for a match. | |
1505 | */ | |
88b32f06 | 1506 | node = blk->bp->b_addr; |
ff105f75 DC |
1507 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
1508 | btree = dp->d_ops->node_tree_p(node); | |
2bd0ea18 | 1509 | |
9473ee32 DW |
1510 | /* Tree taller than we can handle; bail out! */ |
1511 | if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) | |
1512 | return -EFSCORRUPTED; | |
1513 | ||
1514 | /* Check the level from the root. */ | |
1515 | if (blkno == args->geo->leafblk) | |
1516 | expected_level = nodehdr.level - 1; | |
1517 | else if (expected_level != nodehdr.level) | |
1518 | return -EFSCORRUPTED; | |
1519 | else | |
1520 | expected_level--; | |
1521 | ||
88b32f06 DC |
1522 | max = nodehdr.count; |
1523 | blk->hashval = be32_to_cpu(btree[max - 1].hashval); | |
2bd0ea18 | 1524 | |
88b32f06 DC |
1525 | /* |
1526 | * Binary search. (note: small blocks will skip loop) | |
1527 | */ | |
1528 | probe = span = max / 2; | |
1529 | hashval = args->hashval; | |
1530 | while (span > 4) { | |
1531 | span /= 2; | |
1532 | btreehashval = be32_to_cpu(btree[probe].hashval); | |
1533 | if (btreehashval < hashval) | |
1534 | probe += span; | |
1535 | else if (btreehashval > hashval) | |
1536 | probe -= span; | |
1537 | else | |
1538 | break; | |
1539 | } | |
1540 | ASSERT((probe >= 0) && (probe < max)); | |
1541 | ASSERT((span <= 4) || | |
1542 | (be32_to_cpu(btree[probe].hashval) == hashval)); | |
2bd0ea18 | 1543 | |
88b32f06 DC |
1544 | /* |
1545 | * Since we may have duplicate hashval's, find the first | |
1546 | * matching hashval in the node. | |
1547 | */ | |
1548 | while (probe > 0 && | |
1549 | be32_to_cpu(btree[probe].hashval) >= hashval) { | |
1550 | probe--; | |
1551 | } | |
1552 | while (probe < max && | |
1553 | be32_to_cpu(btree[probe].hashval) < hashval) { | |
1554 | probe++; | |
1555 | } | |
1556 | ||
1557 | /* | |
1558 | * Pick the right block to descend on. | |
1559 | */ | |
1560 | if (probe == max) { | |
1561 | blk->index = max - 1; | |
1562 | blkno = be32_to_cpu(btree[max - 1].before); | |
1563 | } else { | |
1564 | blk->index = probe; | |
1565 | blkno = be32_to_cpu(btree[probe].before); | |
2bd0ea18 | 1566 | } |
9473ee32 DW |
1567 | |
1568 | /* We can't point back to the root. */ | |
1569 | if (blkno == args->geo->leafblk) | |
1570 | return -EFSCORRUPTED; | |
2bd0ea18 NS |
1571 | } |
1572 | ||
9473ee32 DW |
1573 | if (expected_level != 0) |
1574 | return -EFSCORRUPTED; | |
1575 | ||
2bd0ea18 NS |
1576 | /* |
1577 | * A leaf block that ends in the hashval that we are interested in | |
1578 | * (final hashval == search hashval) means that the next block may | |
1579 | * contain more entries with the same hashval, shift upward to the | |
1580 | * next leaf and keep searching. | |
1581 | */ | |
1582 | for (;;) { | |
5e656dbb | 1583 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
32a82561 | 1584 | retval = xfs_dir2_leafn_lookup_int(blk->bp, args, |
2bd0ea18 | 1585 | &blk->index, state); |
5e656dbb | 1586 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
a24374f4 | 1587 | retval = xfs_attr3_leaf_lookup_int(blk->bp, args); |
32a82561 NS |
1588 | blk->index = args->index; |
1589 | args->blkno = blk->blkno; | |
5e656dbb BN |
1590 | } else { |
1591 | ASSERT(0); | |
12b53197 | 1592 | return -EFSCORRUPTED; |
2bd0ea18 | 1593 | } |
12b53197 | 1594 | if (((retval == -ENOENT) || (retval == -ENOATTR)) && |
32a82561 | 1595 | (blk->hashval == args->hashval)) { |
88b32f06 | 1596 | error = xfs_da3_path_shift(state, &state->path, 1, 1, |
2bd0ea18 NS |
1597 | &retval); |
1598 | if (error) | |
af43ca9f | 1599 | return error; |
2bd0ea18 NS |
1600 | if (retval == 0) { |
1601 | continue; | |
5e656dbb | 1602 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
2bd0ea18 | 1603 | /* path_shift() gives ENOENT */ |
12b53197 | 1604 | retval = -ENOATTR; |
2bd0ea18 | 1605 | } |
2bd0ea18 NS |
1606 | } |
1607 | break; | |
1608 | } | |
1609 | *result = retval; | |
af43ca9f | 1610 | return 0; |
2bd0ea18 NS |
1611 | } |
1612 | ||
2bd0ea18 NS |
1613 | /*======================================================================== |
1614 | * Utility routines. | |
1615 | *========================================================================*/ | |
1616 | ||
88b32f06 DC |
1617 | /* |
1618 | * Compare two intermediate nodes for "order". | |
1619 | */ | |
1620 | STATIC int | |
1621 | xfs_da3_node_order( | |
ff105f75 | 1622 | struct xfs_inode *dp, |
88b32f06 DC |
1623 | struct xfs_buf *node1_bp, |
1624 | struct xfs_buf *node2_bp) | |
1625 | { | |
1626 | struct xfs_da_intnode *node1; | |
1627 | struct xfs_da_intnode *node2; | |
1628 | struct xfs_da_node_entry *btree1; | |
1629 | struct xfs_da_node_entry *btree2; | |
1630 | struct xfs_da3_icnode_hdr node1hdr; | |
1631 | struct xfs_da3_icnode_hdr node2hdr; | |
1632 | ||
1633 | node1 = node1_bp->b_addr; | |
1634 | node2 = node2_bp->b_addr; | |
ff105f75 DC |
1635 | dp->d_ops->node_hdr_from_disk(&node1hdr, node1); |
1636 | dp->d_ops->node_hdr_from_disk(&node2hdr, node2); | |
1637 | btree1 = dp->d_ops->node_tree_p(node1); | |
1638 | btree2 = dp->d_ops->node_tree_p(node2); | |
88b32f06 DC |
1639 | |
1640 | if (node1hdr.count > 0 && node2hdr.count > 0 && | |
1641 | ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) || | |
1642 | (be32_to_cpu(btree2[node2hdr.count - 1].hashval) < | |
1643 | be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) { | |
1644 | return 1; | |
1645 | } | |
1646 | return 0; | |
1647 | } | |
1648 | ||
2bd0ea18 NS |
1649 | /* |
1650 | * Link a new block into a doubly linked list of blocks (of whatever type). | |
1651 | */ | |
1652 | int /* error */ | |
88b32f06 DC |
1653 | xfs_da3_blk_link( |
1654 | struct xfs_da_state *state, | |
1655 | struct xfs_da_state_blk *old_blk, | |
1656 | struct xfs_da_state_blk *new_blk) | |
2bd0ea18 | 1657 | { |
88b32f06 DC |
1658 | struct xfs_da_blkinfo *old_info; |
1659 | struct xfs_da_blkinfo *new_info; | |
1660 | struct xfs_da_blkinfo *tmp_info; | |
1661 | struct xfs_da_args *args; | |
1662 | struct xfs_buf *bp; | |
1663 | int before = 0; | |
1664 | int error; | |
ff105f75 | 1665 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 NS |
1666 | |
1667 | /* | |
1668 | * Set up environment. | |
1669 | */ | |
1670 | args = state->args; | |
1671 | ASSERT(args != NULL); | |
a2ceac1f DC |
1672 | old_info = old_blk->bp->b_addr; |
1673 | new_info = new_blk->bp->b_addr; | |
2bd0ea18 | 1674 | ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC || |
5e656dbb | 1675 | old_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
2bd0ea18 | 1676 | old_blk->magic == XFS_ATTR_LEAF_MAGIC); |
2bd0ea18 NS |
1677 | |
1678 | switch (old_blk->magic) { | |
2bd0ea18 NS |
1679 | case XFS_ATTR_LEAF_MAGIC: |
1680 | before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp); | |
1681 | break; | |
2bd0ea18 | 1682 | case XFS_DIR2_LEAFN_MAGIC: |
ff105f75 | 1683 | before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp); |
2bd0ea18 NS |
1684 | break; |
1685 | case XFS_DA_NODE_MAGIC: | |
ff105f75 | 1686 | before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp); |
2bd0ea18 NS |
1687 | break; |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * Link blocks in appropriate order. | |
1692 | */ | |
1693 | if (before) { | |
1694 | /* | |
1695 | * Link new block in before existing block. | |
1696 | */ | |
a2ceac1f | 1697 | trace_xfs_da_link_before(args); |
5e656dbb BN |
1698 | new_info->forw = cpu_to_be32(old_blk->blkno); |
1699 | new_info->back = old_info->back; | |
1700 | if (old_info->back) { | |
ff105f75 | 1701 | error = xfs_da3_node_read(args->trans, dp, |
5e656dbb BN |
1702 | be32_to_cpu(old_info->back), |
1703 | -1, &bp, args->whichfork); | |
2bd0ea18 | 1704 | if (error) |
af43ca9f | 1705 | return error; |
2bd0ea18 | 1706 | ASSERT(bp != NULL); |
a2ceac1f | 1707 | tmp_info = bp->b_addr; |
88b32f06 | 1708 | ASSERT(tmp_info->magic == old_info->magic); |
5e656dbb BN |
1709 | ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno); |
1710 | tmp_info->forw = cpu_to_be32(new_blk->blkno); | |
a2ceac1f | 1711 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
2bd0ea18 | 1712 | } |
5e656dbb | 1713 | old_info->back = cpu_to_be32(new_blk->blkno); |
2bd0ea18 NS |
1714 | } else { |
1715 | /* | |
1716 | * Link new block in after existing block. | |
1717 | */ | |
a2ceac1f | 1718 | trace_xfs_da_link_after(args); |
5e656dbb BN |
1719 | new_info->forw = old_info->forw; |
1720 | new_info->back = cpu_to_be32(old_blk->blkno); | |
1721 | if (old_info->forw) { | |
ff105f75 | 1722 | error = xfs_da3_node_read(args->trans, dp, |
5e656dbb BN |
1723 | be32_to_cpu(old_info->forw), |
1724 | -1, &bp, args->whichfork); | |
2bd0ea18 | 1725 | if (error) |
af43ca9f | 1726 | return error; |
2bd0ea18 | 1727 | ASSERT(bp != NULL); |
a2ceac1f | 1728 | tmp_info = bp->b_addr; |
5e656dbb BN |
1729 | ASSERT(tmp_info->magic == old_info->magic); |
1730 | ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno); | |
1731 | tmp_info->back = cpu_to_be32(new_blk->blkno); | |
a2ceac1f | 1732 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
2bd0ea18 | 1733 | } |
5e656dbb | 1734 | old_info->forw = cpu_to_be32(new_blk->blkno); |
2bd0ea18 NS |
1735 | } |
1736 | ||
a2ceac1f DC |
1737 | xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1); |
1738 | xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1); | |
af43ca9f | 1739 | return 0; |
2bd0ea18 NS |
1740 | } |
1741 | ||
2bd0ea18 NS |
1742 | /* |
1743 | * Unlink a block from a doubly linked list of blocks. | |
1744 | */ | |
5e656dbb | 1745 | STATIC int /* error */ |
88b32f06 DC |
1746 | xfs_da3_blk_unlink( |
1747 | struct xfs_da_state *state, | |
1748 | struct xfs_da_state_blk *drop_blk, | |
1749 | struct xfs_da_state_blk *save_blk) | |
2bd0ea18 | 1750 | { |
88b32f06 DC |
1751 | struct xfs_da_blkinfo *drop_info; |
1752 | struct xfs_da_blkinfo *save_info; | |
1753 | struct xfs_da_blkinfo *tmp_info; | |
1754 | struct xfs_da_args *args; | |
1755 | struct xfs_buf *bp; | |
1756 | int error; | |
2bd0ea18 NS |
1757 | |
1758 | /* | |
1759 | * Set up environment. | |
1760 | */ | |
1761 | args = state->args; | |
1762 | ASSERT(args != NULL); | |
a2ceac1f DC |
1763 | save_info = save_blk->bp->b_addr; |
1764 | drop_info = drop_blk->bp->b_addr; | |
2bd0ea18 | 1765 | ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC || |
5e656dbb | 1766 | save_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
2bd0ea18 | 1767 | save_blk->magic == XFS_ATTR_LEAF_MAGIC); |
2bd0ea18 | 1768 | ASSERT(save_blk->magic == drop_blk->magic); |
5e656dbb BN |
1769 | ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) || |
1770 | (be32_to_cpu(save_info->back) == drop_blk->blkno)); | |
1771 | ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) || | |
1772 | (be32_to_cpu(drop_info->back) == save_blk->blkno)); | |
2bd0ea18 NS |
1773 | |
1774 | /* | |
1775 | * Unlink the leaf block from the doubly linked chain of leaves. | |
1776 | */ | |
5e656dbb | 1777 | if (be32_to_cpu(save_info->back) == drop_blk->blkno) { |
a2ceac1f | 1778 | trace_xfs_da_unlink_back(args); |
5e656dbb BN |
1779 | save_info->back = drop_info->back; |
1780 | if (drop_info->back) { | |
88b32f06 | 1781 | error = xfs_da3_node_read(args->trans, args->dp, |
5e656dbb BN |
1782 | be32_to_cpu(drop_info->back), |
1783 | -1, &bp, args->whichfork); | |
2bd0ea18 | 1784 | if (error) |
af43ca9f | 1785 | return error; |
2bd0ea18 | 1786 | ASSERT(bp != NULL); |
a2ceac1f | 1787 | tmp_info = bp->b_addr; |
5e656dbb BN |
1788 | ASSERT(tmp_info->magic == save_info->magic); |
1789 | ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno); | |
1790 | tmp_info->forw = cpu_to_be32(save_blk->blkno); | |
a2ceac1f | 1791 | xfs_trans_log_buf(args->trans, bp, 0, |
2bd0ea18 | 1792 | sizeof(*tmp_info) - 1); |
2bd0ea18 NS |
1793 | } |
1794 | } else { | |
a2ceac1f | 1795 | trace_xfs_da_unlink_forward(args); |
5e656dbb BN |
1796 | save_info->forw = drop_info->forw; |
1797 | if (drop_info->forw) { | |
88b32f06 | 1798 | error = xfs_da3_node_read(args->trans, args->dp, |
5e656dbb BN |
1799 | be32_to_cpu(drop_info->forw), |
1800 | -1, &bp, args->whichfork); | |
2bd0ea18 | 1801 | if (error) |
af43ca9f | 1802 | return error; |
2bd0ea18 | 1803 | ASSERT(bp != NULL); |
a2ceac1f | 1804 | tmp_info = bp->b_addr; |
5e656dbb BN |
1805 | ASSERT(tmp_info->magic == save_info->magic); |
1806 | ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno); | |
1807 | tmp_info->back = cpu_to_be32(save_blk->blkno); | |
a2ceac1f | 1808 | xfs_trans_log_buf(args->trans, bp, 0, |
2bd0ea18 | 1809 | sizeof(*tmp_info) - 1); |
2bd0ea18 NS |
1810 | } |
1811 | } | |
1812 | ||
a2ceac1f | 1813 | xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1); |
af43ca9f | 1814 | return 0; |
2bd0ea18 NS |
1815 | } |
1816 | ||
1817 | /* | |
1818 | * Move a path "forward" or "!forward" one block at the current level. | |
1819 | * | |
1820 | * This routine will adjust a "path" to point to the next block | |
1821 | * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the | |
1822 | * Btree, including updating pointers to the intermediate nodes between | |
1823 | * the new bottom and the root. | |
1824 | */ | |
1825 | int /* error */ | |
88b32f06 DC |
1826 | xfs_da3_path_shift( |
1827 | struct xfs_da_state *state, | |
1828 | struct xfs_da_state_path *path, | |
1829 | int forward, | |
1830 | int release, | |
1831 | int *result) | |
2bd0ea18 | 1832 | { |
88b32f06 DC |
1833 | struct xfs_da_state_blk *blk; |
1834 | struct xfs_da_blkinfo *info; | |
1835 | struct xfs_da_intnode *node; | |
1836 | struct xfs_da_args *args; | |
1837 | struct xfs_da_node_entry *btree; | |
1838 | struct xfs_da3_icnode_hdr nodehdr; | |
c063b960 | 1839 | struct xfs_buf *bp; |
88b32f06 DC |
1840 | xfs_dablk_t blkno = 0; |
1841 | int level; | |
1842 | int error; | |
ff105f75 | 1843 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1844 | |
a2ceac1f DC |
1845 | trace_xfs_da_path_shift(state->args); |
1846 | ||
2bd0ea18 NS |
1847 | /* |
1848 | * Roll up the Btree looking for the first block where our | |
1849 | * current index is not at the edge of the block. Note that | |
1850 | * we skip the bottom layer because we want the sibling block. | |
1851 | */ | |
1852 | args = state->args; | |
1853 | ASSERT(args != NULL); | |
1854 | ASSERT(path != NULL); | |
1855 | ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH)); | |
1856 | level = (path->active-1) - 1; /* skip bottom layer in path */ | |
1857 | for (blk = &path->blk[level]; level >= 0; blk--, level--) { | |
a2ceac1f | 1858 | node = blk->bp->b_addr; |
ff105f75 DC |
1859 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
1860 | btree = dp->d_ops->node_tree_p(node); | |
88b32f06 DC |
1861 | |
1862 | if (forward && (blk->index < nodehdr.count - 1)) { | |
2bd0ea18 | 1863 | blk->index++; |
88b32f06 | 1864 | blkno = be32_to_cpu(btree[blk->index].before); |
2bd0ea18 NS |
1865 | break; |
1866 | } else if (!forward && (blk->index > 0)) { | |
1867 | blk->index--; | |
88b32f06 | 1868 | blkno = be32_to_cpu(btree[blk->index].before); |
2bd0ea18 NS |
1869 | break; |
1870 | } | |
1871 | } | |
1872 | if (level < 0) { | |
12b53197 | 1873 | *result = -ENOENT; /* we're out of our tree */ |
5e656dbb | 1874 | ASSERT(args->op_flags & XFS_DA_OP_OKNOENT); |
af43ca9f | 1875 | return 0; |
2bd0ea18 NS |
1876 | } |
1877 | ||
1878 | /* | |
1879 | * Roll down the edge of the subtree until we reach the | |
1880 | * same depth we were at originally. | |
1881 | */ | |
1882 | for (blk++, level++; level < path->active; blk++, level++) { | |
1883 | /* | |
c063b960 | 1884 | * Read the next child block into a local buffer. |
2bd0ea18 | 1885 | */ |
c063b960 BF |
1886 | error = xfs_da3_node_read(args->trans, dp, blkno, -1, &bp, |
1887 | args->whichfork); | |
1888 | if (error) | |
1889 | return error; | |
2bd0ea18 NS |
1890 | |
1891 | /* | |
c063b960 BF |
1892 | * Release the old block (if it's dirty, the trans doesn't |
1893 | * actually let go) and swap the local buffer into the path | |
1894 | * structure. This ensures failure of the above read doesn't set | |
1895 | * a NULL buffer in an active slot in the path. | |
2bd0ea18 | 1896 | */ |
c063b960 BF |
1897 | if (release) |
1898 | xfs_trans_brelse(args->trans, blk->bp); | |
2bd0ea18 | 1899 | blk->blkno = blkno; |
c063b960 BF |
1900 | blk->bp = bp; |
1901 | ||
a2ceac1f DC |
1902 | info = blk->bp->b_addr; |
1903 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
88b32f06 | 1904 | info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) || |
a2ceac1f | 1905 | info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
65b80c98 | 1906 | info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
a24374f4 DC |
1907 | info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || |
1908 | info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); | |
88b32f06 DC |
1909 | |
1910 | ||
1911 | /* | |
1912 | * Note: we flatten the magic number to a single type so we | |
1913 | * don't have to compare against crc/non-crc types elsewhere. | |
1914 | */ | |
1915 | switch (be16_to_cpu(info->magic)) { | |
1916 | case XFS_DA_NODE_MAGIC: | |
1917 | case XFS_DA3_NODE_MAGIC: | |
1918 | blk->magic = XFS_DA_NODE_MAGIC; | |
2bd0ea18 | 1919 | node = (xfs_da_intnode_t *)info; |
ff105f75 DC |
1920 | dp->d_ops->node_hdr_from_disk(&nodehdr, node); |
1921 | btree = dp->d_ops->node_tree_p(node); | |
88b32f06 | 1922 | blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
2bd0ea18 NS |
1923 | if (forward) |
1924 | blk->index = 0; | |
1925 | else | |
88b32f06 DC |
1926 | blk->index = nodehdr.count - 1; |
1927 | blkno = be32_to_cpu(btree[blk->index].before); | |
1928 | break; | |
1929 | case XFS_ATTR_LEAF_MAGIC: | |
a24374f4 | 1930 | case XFS_ATTR3_LEAF_MAGIC: |
88b32f06 | 1931 | blk->magic = XFS_ATTR_LEAF_MAGIC; |
2bd0ea18 NS |
1932 | ASSERT(level == path->active-1); |
1933 | blk->index = 0; | |
ff105f75 | 1934 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
88b32f06 DC |
1935 | break; |
1936 | case XFS_DIR2_LEAFN_MAGIC: | |
1937 | case XFS_DIR3_LEAFN_MAGIC: | |
1938 | blk->magic = XFS_DIR2_LEAFN_MAGIC; | |
1939 | ASSERT(level == path->active-1); | |
1940 | blk->index = 0; | |
94ffb80a DW |
1941 | blk->hashval = xfs_dir2_leaf_lasthash(args->dp, |
1942 | blk->bp, NULL); | |
88b32f06 DC |
1943 | break; |
1944 | default: | |
1945 | ASSERT(0); | |
1946 | break; | |
2bd0ea18 NS |
1947 | } |
1948 | } | |
1949 | *result = 0; | |
88b32f06 | 1950 | return 0; |
2bd0ea18 NS |
1951 | } |
1952 | ||
1953 | ||
1954 | /*======================================================================== | |
1955 | * Utility routines. | |
1956 | *========================================================================*/ | |
1957 | ||
1958 | /* | |
1959 | * Implement a simple hash on a character string. | |
1960 | * Rotate the hash value by 7 bits, then XOR each character in. | |
1961 | * This is implemented with some source-level loop unrolling. | |
1962 | */ | |
1963 | xfs_dahash_t | |
4a492e72 | 1964 | xfs_da_hashname(const uint8_t *name, int namelen) |
2bd0ea18 NS |
1965 | { |
1966 | xfs_dahash_t hash; | |
1967 | ||
2bd0ea18 NS |
1968 | /* |
1969 | * Do four characters at a time as long as we can. | |
1970 | */ | |
f9c48a87 | 1971 | for (hash = 0; namelen >= 4; namelen -= 4, name += 4) |
2bd0ea18 | 1972 | hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^ |
f9c48a87 NS |
1973 | (name[3] << 0) ^ rol32(hash, 7 * 4); |
1974 | ||
2bd0ea18 NS |
1975 | /* |
1976 | * Now do the rest of the characters. | |
1977 | */ | |
1978 | switch (namelen) { | |
1979 | case 3: | |
1980 | return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^ | |
f9c48a87 | 1981 | rol32(hash, 7 * 3); |
2bd0ea18 | 1982 | case 2: |
f9c48a87 | 1983 | return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2); |
2bd0ea18 | 1984 | case 1: |
f9c48a87 NS |
1985 | return (name[0] << 0) ^ rol32(hash, 7 * 1); |
1986 | default: /* case 0: */ | |
2bd0ea18 NS |
1987 | return hash; |
1988 | } | |
2bd0ea18 NS |
1989 | } |
1990 | ||
51ca7008 | 1991 | enum xfs_dacmp |
5e656dbb BN |
1992 | xfs_da_compname( |
1993 | struct xfs_da_args *args, | |
56b2de80 DC |
1994 | const unsigned char *name, |
1995 | int len) | |
51ca7008 | 1996 | { |
5e656dbb | 1997 | return (args->namelen == len && memcmp(args->name, name, len) == 0) ? |
51ca7008 BN |
1998 | XFS_CMP_EXACT : XFS_CMP_DIFFERENT; |
1999 | } | |
2000 | ||
5e656dbb BN |
2001 | static xfs_dahash_t |
2002 | xfs_default_hashname( | |
2003 | struct xfs_name *name) | |
2004 | { | |
2005 | return xfs_da_hashname(name->name, name->len); | |
2006 | } | |
2007 | ||
51ca7008 | 2008 | const struct xfs_nameops xfs_default_nameops = { |
5e656dbb | 2009 | .hashname = xfs_default_hashname, |
51ca7008 BN |
2010 | .compname = xfs_da_compname |
2011 | }; | |
2012 | ||
2bd0ea18 | 2013 | int |
a2ceac1f DC |
2014 | xfs_da_grow_inode_int( |
2015 | struct xfs_da_args *args, | |
2016 | xfs_fileoff_t *bno, | |
2017 | int count) | |
2bd0ea18 | 2018 | { |
a2ceac1f DC |
2019 | struct xfs_trans *tp = args->trans; |
2020 | struct xfs_inode *dp = args->dp; | |
2021 | int w = args->whichfork; | |
5a35bf2c | 2022 | xfs_rfsblock_t nblks = dp->i_d.di_nblocks; |
a2ceac1f DC |
2023 | struct xfs_bmbt_irec map, *mapp; |
2024 | int nmap, error, got, i, mapi; | |
56b2de80 | 2025 | |
2bd0ea18 NS |
2026 | /* |
2027 | * Find a spot in the file space to put the new block. | |
2028 | */ | |
a2ceac1f DC |
2029 | error = xfs_bmap_first_unused(tp, dp, count, bno, w); |
2030 | if (error) | |
2bd0ea18 | 2031 | return error; |
a2ceac1f | 2032 | |
2bd0ea18 NS |
2033 | /* |
2034 | * Try mapping it in one filesystem block. | |
2035 | */ | |
2036 | nmap = 1; | |
2037 | ASSERT(args->firstblock != NULL); | |
a2ceac1f DC |
2038 | error = xfs_bmapi_write(tp, dp, *bno, count, |
2039 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG, | |
2bd0ea18 | 2040 | args->firstblock, args->total, &map, &nmap, |
f33cea1a | 2041 | args->dfops); |
a2ceac1f | 2042 | if (error) |
2bd0ea18 | 2043 | return error; |
a2ceac1f | 2044 | |
2bd0ea18 NS |
2045 | ASSERT(nmap <= 1); |
2046 | if (nmap == 1) { | |
2047 | mapp = ↦ | |
2048 | mapi = 1; | |
a2ceac1f DC |
2049 | } else if (nmap == 0 && count > 1) { |
2050 | xfs_fileoff_t b; | |
2051 | int c; | |
2052 | ||
2053 | /* | |
2054 | * If we didn't get it and the block might work if fragmented, | |
2055 | * try without the CONTIG flag. Loop until we get it all. | |
2056 | */ | |
2bd0ea18 | 2057 | mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP); |
a2ceac1f | 2058 | for (b = *bno, mapi = 0; b < *bno + count; ) { |
2bd0ea18 | 2059 | nmap = MIN(XFS_BMAP_MAX_NMAP, count); |
a2ceac1f DC |
2060 | c = (int)(*bno + count - b); |
2061 | error = xfs_bmapi_write(tp, dp, b, c, | |
2062 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, | |
2bd0ea18 | 2063 | args->firstblock, args->total, |
f33cea1a | 2064 | &mapp[mapi], &nmap, args->dfops); |
a2ceac1f DC |
2065 | if (error) |
2066 | goto out_free_map; | |
2bd0ea18 NS |
2067 | if (nmap < 1) |
2068 | break; | |
2069 | mapi += nmap; | |
2070 | b = mapp[mapi - 1].br_startoff + | |
2071 | mapp[mapi - 1].br_blockcount; | |
2072 | } | |
2073 | } else { | |
2bd0ea18 NS |
2074 | mapi = 0; |
2075 | mapp = NULL; | |
2076 | } | |
a2ceac1f | 2077 | |
2bd0ea18 NS |
2078 | /* |
2079 | * Count the blocks we got, make sure it matches the total. | |
2080 | */ | |
2081 | for (i = 0, got = 0; i < mapi; i++) | |
2082 | got += mapp[i].br_blockcount; | |
a2ceac1f | 2083 | if (got != count || mapp[0].br_startoff != *bno || |
2bd0ea18 | 2084 | mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount != |
a2ceac1f | 2085 | *bno + count) { |
12b53197 | 2086 | error = -ENOSPC; |
a2ceac1f | 2087 | goto out_free_map; |
2bd0ea18 | 2088 | } |
a2ceac1f | 2089 | |
56b2de80 DC |
2090 | /* account for newly allocated blocks in reserved blocks total */ |
2091 | args->total -= dp->i_d.di_nblocks - nblks; | |
a2ceac1f DC |
2092 | |
2093 | out_free_map: | |
2094 | if (mapp != &map) | |
2095 | kmem_free(mapp); | |
2096 | return error; | |
2097 | } | |
2098 | ||
2099 | /* | |
2100 | * Add a block to the btree ahead of the file. | |
2101 | * Return the new block number to the caller. | |
2102 | */ | |
2103 | int | |
2104 | xfs_da_grow_inode( | |
2105 | struct xfs_da_args *args, | |
2106 | xfs_dablk_t *new_blkno) | |
2107 | { | |
2108 | xfs_fileoff_t bno; | |
a2ceac1f DC |
2109 | int error; |
2110 | ||
2111 | trace_xfs_da_grow_inode(args); | |
2112 | ||
ff105f75 DC |
2113 | bno = args->geo->leafblk; |
2114 | error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount); | |
a2ceac1f DC |
2115 | if (!error) |
2116 | *new_blkno = (xfs_dablk_t)bno; | |
2117 | return error; | |
2bd0ea18 NS |
2118 | } |
2119 | ||
2bd0ea18 | 2120 | /* |
dfc130f3 | 2121 | * Ick. We need to always be able to remove a btree block, even |
2bd0ea18 NS |
2122 | * if there's no space reservation because the filesystem is full. |
2123 | * This is called if xfs_bunmapi on a btree block fails due to ENOSPC. | |
2124 | * It swaps the target block with the last block in the file. The | |
2125 | * last block in the file can always be removed since it can't cause | |
2126 | * a bmap btree split to do that. | |
2127 | */ | |
2128 | STATIC int | |
88b32f06 DC |
2129 | xfs_da3_swap_lastblock( |
2130 | struct xfs_da_args *args, | |
2131 | xfs_dablk_t *dead_blknop, | |
2132 | struct xfs_buf **dead_bufp) | |
2bd0ea18 | 2133 | { |
88b32f06 DC |
2134 | struct xfs_da_blkinfo *dead_info; |
2135 | struct xfs_da_blkinfo *sib_info; | |
2136 | struct xfs_da_intnode *par_node; | |
2137 | struct xfs_da_intnode *dead_node; | |
2138 | struct xfs_dir2_leaf *dead_leaf2; | |
2139 | struct xfs_da_node_entry *btree; | |
2140 | struct xfs_da3_icnode_hdr par_hdr; | |
ff105f75 | 2141 | struct xfs_inode *dp; |
88b32f06 DC |
2142 | struct xfs_trans *tp; |
2143 | struct xfs_mount *mp; | |
2144 | struct xfs_buf *dead_buf; | |
2145 | struct xfs_buf *last_buf; | |
2146 | struct xfs_buf *sib_buf; | |
2147 | struct xfs_buf *par_buf; | |
2148 | xfs_dahash_t dead_hash; | |
2149 | xfs_fileoff_t lastoff; | |
2150 | xfs_dablk_t dead_blkno; | |
2151 | xfs_dablk_t last_blkno; | |
2152 | xfs_dablk_t sib_blkno; | |
2153 | xfs_dablk_t par_blkno; | |
2154 | int error; | |
2155 | int w; | |
2156 | int entno; | |
2157 | int level; | |
2158 | int dead_level; | |
2bd0ea18 | 2159 | |
a2ceac1f DC |
2160 | trace_xfs_da_swap_lastblock(args); |
2161 | ||
2bd0ea18 NS |
2162 | dead_buf = *dead_bufp; |
2163 | dead_blkno = *dead_blknop; | |
2164 | tp = args->trans; | |
ff105f75 | 2165 | dp = args->dp; |
2bd0ea18 NS |
2166 | w = args->whichfork; |
2167 | ASSERT(w == XFS_DATA_FORK); | |
ff105f75 DC |
2168 | mp = dp->i_mount; |
2169 | lastoff = args->geo->freeblk; | |
2170 | error = xfs_bmap_last_before(tp, dp, &lastoff, w); | |
2bd0ea18 NS |
2171 | if (error) |
2172 | return error; | |
4ca431fc NS |
2173 | if (unlikely(lastoff == 0)) { |
2174 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(1)", XFS_ERRLEVEL_LOW, | |
2175 | mp); | |
12b53197 | 2176 | return -EFSCORRUPTED; |
4ca431fc | 2177 | } |
2bd0ea18 NS |
2178 | /* |
2179 | * Read the last block in the btree space. | |
2180 | */ | |
ff105f75 DC |
2181 | last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount; |
2182 | error = xfs_da3_node_read(tp, dp, last_blkno, -1, &last_buf, w); | |
a2ceac1f | 2183 | if (error) |
2bd0ea18 NS |
2184 | return error; |
2185 | /* | |
2186 | * Copy the last block into the dead buffer and log it. | |
2187 | */ | |
ff105f75 DC |
2188 | memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize); |
2189 | xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1); | |
a2ceac1f | 2190 | dead_info = dead_buf->b_addr; |
2bd0ea18 NS |
2191 | /* |
2192 | * Get values from the moved block. | |
2193 | */ | |
65b80c98 DC |
2194 | if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
2195 | dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
2196 | struct xfs_dir3_icleaf_hdr leafhdr; | |
2197 | struct xfs_dir2_leaf_entry *ents; | |
2198 | ||
2bd0ea18 | 2199 | dead_leaf2 = (xfs_dir2_leaf_t *)dead_info; |
ff105f75 DC |
2200 | dp->d_ops->leaf_hdr_from_disk(&leafhdr, dead_leaf2); |
2201 | ents = dp->d_ops->leaf_ents_p(dead_leaf2); | |
2bd0ea18 | 2202 | dead_level = 0; |
65b80c98 | 2203 | dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval); |
2bd0ea18 | 2204 | } else { |
88b32f06 DC |
2205 | struct xfs_da3_icnode_hdr deadhdr; |
2206 | ||
2bd0ea18 | 2207 | dead_node = (xfs_da_intnode_t *)dead_info; |
ff105f75 DC |
2208 | dp->d_ops->node_hdr_from_disk(&deadhdr, dead_node); |
2209 | btree = dp->d_ops->node_tree_p(dead_node); | |
88b32f06 DC |
2210 | dead_level = deadhdr.level; |
2211 | dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval); | |
2bd0ea18 NS |
2212 | } |
2213 | sib_buf = par_buf = NULL; | |
2214 | /* | |
2215 | * If the moved block has a left sibling, fix up the pointers. | |
2216 | */ | |
5e656dbb | 2217 | if ((sib_blkno = be32_to_cpu(dead_info->back))) { |
ff105f75 | 2218 | error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w); |
a2ceac1f | 2219 | if (error) |
2bd0ea18 | 2220 | goto done; |
a2ceac1f | 2221 | sib_info = sib_buf->b_addr; |
4ca431fc | 2222 | if (unlikely( |
5e656dbb BN |
2223 | be32_to_cpu(sib_info->forw) != last_blkno || |
2224 | sib_info->magic != dead_info->magic)) { | |
4ca431fc NS |
2225 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(2)", |
2226 | XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 2227 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2228 | goto done; |
2229 | } | |
5e656dbb | 2230 | sib_info->forw = cpu_to_be32(dead_blkno); |
a2ceac1f | 2231 | xfs_trans_log_buf(tp, sib_buf, |
2bd0ea18 NS |
2232 | XFS_DA_LOGRANGE(sib_info, &sib_info->forw, |
2233 | sizeof(sib_info->forw))); | |
2bd0ea18 NS |
2234 | sib_buf = NULL; |
2235 | } | |
2236 | /* | |
2237 | * If the moved block has a right sibling, fix up the pointers. | |
2238 | */ | |
5e656dbb | 2239 | if ((sib_blkno = be32_to_cpu(dead_info->forw))) { |
ff105f75 | 2240 | error = xfs_da3_node_read(tp, dp, sib_blkno, -1, &sib_buf, w); |
a2ceac1f | 2241 | if (error) |
2bd0ea18 | 2242 | goto done; |
a2ceac1f | 2243 | sib_info = sib_buf->b_addr; |
4ca431fc | 2244 | if (unlikely( |
5e656dbb BN |
2245 | be32_to_cpu(sib_info->back) != last_blkno || |
2246 | sib_info->magic != dead_info->magic)) { | |
4ca431fc NS |
2247 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(3)", |
2248 | XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 2249 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2250 | goto done; |
2251 | } | |
5e656dbb | 2252 | sib_info->back = cpu_to_be32(dead_blkno); |
a2ceac1f | 2253 | xfs_trans_log_buf(tp, sib_buf, |
2bd0ea18 NS |
2254 | XFS_DA_LOGRANGE(sib_info, &sib_info->back, |
2255 | sizeof(sib_info->back))); | |
2bd0ea18 NS |
2256 | sib_buf = NULL; |
2257 | } | |
ff105f75 | 2258 | par_blkno = args->geo->leafblk; |
2bd0ea18 NS |
2259 | level = -1; |
2260 | /* | |
2261 | * Walk down the tree looking for the parent of the moved block. | |
2262 | */ | |
2263 | for (;;) { | |
ff105f75 | 2264 | error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w); |
a2ceac1f | 2265 | if (error) |
2bd0ea18 | 2266 | goto done; |
a2ceac1f | 2267 | par_node = par_buf->b_addr; |
ff105f75 | 2268 | dp->d_ops->node_hdr_from_disk(&par_hdr, par_node); |
88b32f06 | 2269 | if (level >= 0 && level != par_hdr.level + 1) { |
4ca431fc NS |
2270 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(4)", |
2271 | XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 2272 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2273 | goto done; |
2274 | } | |
88b32f06 | 2275 | level = par_hdr.level; |
ff105f75 | 2276 | btree = dp->d_ops->node_tree_p(par_node); |
2bd0ea18 | 2277 | for (entno = 0; |
88b32f06 DC |
2278 | entno < par_hdr.count && |
2279 | be32_to_cpu(btree[entno].hashval) < dead_hash; | |
2bd0ea18 NS |
2280 | entno++) |
2281 | continue; | |
88b32f06 | 2282 | if (entno == par_hdr.count) { |
4ca431fc NS |
2283 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(5)", |
2284 | XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 2285 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2286 | goto done; |
2287 | } | |
88b32f06 | 2288 | par_blkno = be32_to_cpu(btree[entno].before); |
2bd0ea18 NS |
2289 | if (level == dead_level + 1) |
2290 | break; | |
a2ceac1f | 2291 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 NS |
2292 | par_buf = NULL; |
2293 | } | |
2294 | /* | |
2295 | * We're in the right parent block. | |
2296 | * Look for the right entry. | |
2297 | */ | |
2298 | for (;;) { | |
2299 | for (; | |
88b32f06 DC |
2300 | entno < par_hdr.count && |
2301 | be32_to_cpu(btree[entno].before) != last_blkno; | |
2bd0ea18 NS |
2302 | entno++) |
2303 | continue; | |
88b32f06 | 2304 | if (entno < par_hdr.count) |
2bd0ea18 | 2305 | break; |
88b32f06 | 2306 | par_blkno = par_hdr.forw; |
a2ceac1f | 2307 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 | 2308 | par_buf = NULL; |
4ca431fc NS |
2309 | if (unlikely(par_blkno == 0)) { |
2310 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(6)", | |
2311 | XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 2312 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2313 | goto done; |
2314 | } | |
ff105f75 | 2315 | error = xfs_da3_node_read(tp, dp, par_blkno, -1, &par_buf, w); |
a2ceac1f | 2316 | if (error) |
2bd0ea18 | 2317 | goto done; |
a2ceac1f | 2318 | par_node = par_buf->b_addr; |
ff105f75 | 2319 | dp->d_ops->node_hdr_from_disk(&par_hdr, par_node); |
88b32f06 | 2320 | if (par_hdr.level != level) { |
4ca431fc NS |
2321 | XFS_ERROR_REPORT("xfs_da_swap_lastblock(7)", |
2322 | XFS_ERRLEVEL_LOW, mp); | |
12b53197 | 2323 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2324 | goto done; |
2325 | } | |
ff105f75 | 2326 | btree = dp->d_ops->node_tree_p(par_node); |
2bd0ea18 NS |
2327 | entno = 0; |
2328 | } | |
2329 | /* | |
2330 | * Update the parent entry pointing to the moved block. | |
2331 | */ | |
88b32f06 | 2332 | btree[entno].before = cpu_to_be32(dead_blkno); |
a2ceac1f | 2333 | xfs_trans_log_buf(tp, par_buf, |
88b32f06 DC |
2334 | XFS_DA_LOGRANGE(par_node, &btree[entno].before, |
2335 | sizeof(btree[entno].before))); | |
2bd0ea18 NS |
2336 | *dead_blknop = last_blkno; |
2337 | *dead_bufp = last_buf; | |
2338 | return 0; | |
2339 | done: | |
2340 | if (par_buf) | |
a2ceac1f | 2341 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 | 2342 | if (sib_buf) |
a2ceac1f DC |
2343 | xfs_trans_brelse(tp, sib_buf); |
2344 | xfs_trans_brelse(tp, last_buf); | |
2bd0ea18 NS |
2345 | return error; |
2346 | } | |
2347 | ||
2348 | /* | |
2349 | * Remove a btree block from a directory or attribute. | |
2350 | */ | |
2351 | int | |
a2ceac1f DC |
2352 | xfs_da_shrink_inode( |
2353 | xfs_da_args_t *args, | |
2354 | xfs_dablk_t dead_blkno, | |
2355 | struct xfs_buf *dead_buf) | |
2bd0ea18 NS |
2356 | { |
2357 | xfs_inode_t *dp; | |
2358 | int done, error, w, count; | |
2bd0ea18 | 2359 | xfs_trans_t *tp; |
2bd0ea18 | 2360 | |
a2ceac1f DC |
2361 | trace_xfs_da_shrink_inode(args); |
2362 | ||
2bd0ea18 NS |
2363 | dp = args->dp; |
2364 | w = args->whichfork; | |
2365 | tp = args->trans; | |
ff105f75 | 2366 | count = args->geo->fsbcount; |
2bd0ea18 NS |
2367 | for (;;) { |
2368 | /* | |
2369 | * Remove extents. If we get ENOSPC for a dir we have to move | |
2370 | * the last block to the place we want to kill. | |
2371 | */ | |
88b32f06 | 2372 | error = xfs_bunmapi(tp, dp, dead_blkno, count, |
db9cba4e | 2373 | xfs_bmapi_aflag(w), 0, args->firstblock, |
f33cea1a | 2374 | args->dfops, &done); |
12b53197 | 2375 | if (error == -ENOSPC) { |
2bd0ea18 | 2376 | if (w != XFS_DATA_FORK) |
5e656dbb | 2377 | break; |
88b32f06 DC |
2378 | error = xfs_da3_swap_lastblock(args, &dead_blkno, |
2379 | &dead_buf); | |
2380 | if (error) | |
5e656dbb BN |
2381 | break; |
2382 | } else { | |
2bd0ea18 | 2383 | break; |
2bd0ea18 NS |
2384 | } |
2385 | } | |
a2ceac1f | 2386 | xfs_trans_binval(tp, dead_buf); |
2bd0ea18 NS |
2387 | return error; |
2388 | } | |
2389 | ||
2390 | /* | |
2391 | * See if the mapping(s) for this btree block are valid, i.e. | |
2392 | * don't contain holes, are logically contiguous, and cover the whole range. | |
2393 | */ | |
2394 | STATIC int | |
2395 | xfs_da_map_covers_blocks( | |
2396 | int nmap, | |
dfc130f3 | 2397 | xfs_bmbt_irec_t *mapp, |
2bd0ea18 NS |
2398 | xfs_dablk_t bno, |
2399 | int count) | |
2400 | { | |
2401 | int i; | |
2402 | xfs_fileoff_t off; | |
2403 | ||
2404 | for (i = 0, off = bno; i < nmap; i++) { | |
2405 | if (mapp[i].br_startblock == HOLESTARTBLOCK || | |
2406 | mapp[i].br_startblock == DELAYSTARTBLOCK) { | |
2bd0ea18 NS |
2407 | return 0; |
2408 | } | |
2409 | if (off != mapp[i].br_startoff) { | |
2bd0ea18 NS |
2410 | return 0; |
2411 | } | |
2412 | off += mapp[i].br_blockcount; | |
2413 | } | |
2414 | return off == bno + count; | |
2415 | } | |
2416 | ||
2417 | /* | |
a2ceac1f DC |
2418 | * Convert a struct xfs_bmbt_irec to a struct xfs_buf_map. |
2419 | * | |
2420 | * For the single map case, it is assumed that the caller has provided a pointer | |
2421 | * to a valid xfs_buf_map. For the multiple map case, this function will | |
2422 | * allocate the xfs_buf_map to hold all the maps and replace the caller's single | |
2423 | * map pointer with the allocated map. | |
2bd0ea18 | 2424 | */ |
a2ceac1f DC |
2425 | static int |
2426 | xfs_buf_map_from_irec( | |
2427 | struct xfs_mount *mp, | |
2428 | struct xfs_buf_map **mapp, | |
6bddecbc | 2429 | int *nmaps, |
a2ceac1f | 2430 | struct xfs_bmbt_irec *irecs, |
6bddecbc | 2431 | int nirecs) |
2bd0ea18 | 2432 | { |
a2ceac1f DC |
2433 | struct xfs_buf_map *map; |
2434 | int i; | |
2435 | ||
2436 | ASSERT(*nmaps == 1); | |
2437 | ASSERT(nirecs >= 1); | |
2438 | ||
2439 | if (nirecs > 1) { | |
7b02556d DC |
2440 | map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), |
2441 | KM_SLEEP | KM_NOFS); | |
a2ceac1f | 2442 | if (!map) |
12b53197 | 2443 | return -ENOMEM; |
a2ceac1f DC |
2444 | *mapp = map; |
2445 | } | |
2446 | ||
2447 | *nmaps = nirecs; | |
2448 | map = *mapp; | |
2449 | for (i = 0; i < *nmaps; i++) { | |
2450 | ASSERT(irecs[i].br_startblock != DELAYSTARTBLOCK && | |
2451 | irecs[i].br_startblock != HOLESTARTBLOCK); | |
2452 | map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock); | |
2453 | map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount); | |
2454 | } | |
2455 | return 0; | |
2456 | } | |
2457 | ||
2458 | /* | |
2459 | * Map the block we are given ready for reading. There are three possible return | |
2460 | * values: | |
2461 | * -1 - will be returned if we land in a hole and mappedbno == -2 so the | |
2462 | * caller knows not to execute a subsequent read. | |
2463 | * 0 - if we mapped the block successfully | |
2464 | * >0 - positive error number if there was an error. | |
2465 | */ | |
2466 | static int | |
2467 | xfs_dabuf_map( | |
a2ceac1f DC |
2468 | struct xfs_inode *dp, |
2469 | xfs_dablk_t bno, | |
2470 | xfs_daddr_t mappedbno, | |
2471 | int whichfork, | |
2472 | struct xfs_buf_map **map, | |
2473 | int *nmaps) | |
2474 | { | |
2475 | struct xfs_mount *mp = dp->i_mount; | |
2476 | int nfsb; | |
2477 | int error = 0; | |
2478 | struct xfs_bmbt_irec irec; | |
2479 | struct xfs_bmbt_irec *irecs = &irec; | |
2480 | int nirecs; | |
2481 | ||
2482 | ASSERT(map && *map); | |
2483 | ASSERT(*nmaps == 1); | |
2bd0ea18 | 2484 | |
ff105f75 DC |
2485 | if (whichfork == XFS_DATA_FORK) |
2486 | nfsb = mp->m_dir_geo->fsbcount; | |
2487 | else | |
2488 | nfsb = mp->m_attr_geo->fsbcount; | |
a2ceac1f | 2489 | |
2bd0ea18 NS |
2490 | /* |
2491 | * Caller doesn't have a mapping. -2 means don't complain | |
2492 | * if we land in a hole. | |
2493 | */ | |
2494 | if (mappedbno == -1 || mappedbno == -2) { | |
2495 | /* | |
2496 | * Optimize the one-block case. | |
2497 | */ | |
a2ceac1f | 2498 | if (nfsb != 1) |
7b02556d DC |
2499 | irecs = kmem_zalloc(sizeof(irec) * nfsb, |
2500 | KM_SLEEP | KM_NOFS); | |
2bd0ea18 | 2501 | |
a2ceac1f DC |
2502 | nirecs = nfsb; |
2503 | error = xfs_bmapi_read(dp, (xfs_fileoff_t)bno, nfsb, irecs, | |
2504 | &nirecs, xfs_bmapi_aflag(whichfork)); | |
2505 | if (error) | |
2506 | goto out; | |
2bd0ea18 | 2507 | } else { |
a2ceac1f DC |
2508 | irecs->br_startblock = XFS_DADDR_TO_FSB(mp, mappedbno); |
2509 | irecs->br_startoff = (xfs_fileoff_t)bno; | |
2510 | irecs->br_blockcount = nfsb; | |
2511 | irecs->br_state = 0; | |
2512 | nirecs = 1; | |
2bd0ea18 | 2513 | } |
a2ceac1f DC |
2514 | |
2515 | if (!xfs_da_map_covers_blocks(nirecs, irecs, bno, nfsb)) { | |
12b53197 DC |
2516 | error = mappedbno == -2 ? -1 : -EFSCORRUPTED; |
2517 | if (unlikely(error == -EFSCORRUPTED)) { | |
4ca431fc | 2518 | if (xfs_error_level >= XFS_ERRLEVEL_LOW) { |
a2ceac1f DC |
2519 | int i; |
2520 | xfs_alert(mp, "%s: bno %lld dir: inode %lld", | |
2521 | __func__, (long long)bno, | |
4ca431fc | 2522 | (long long)dp->i_ino); |
a2ceac1f DC |
2523 | for (i = 0; i < *nmaps; i++) { |
2524 | xfs_alert(mp, | |
2525 | "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d", | |
4ca431fc | 2526 | i, |
a2ceac1f DC |
2527 | (long long)irecs[i].br_startoff, |
2528 | (long long)irecs[i].br_startblock, | |
2529 | (long long)irecs[i].br_blockcount, | |
2530 | irecs[i].br_state); | |
4ca431fc NS |
2531 | } |
2532 | } | |
2533 | XFS_ERROR_REPORT("xfs_da_do_buf(1)", | |
2534 | XFS_ERRLEVEL_LOW, mp); | |
2535 | } | |
a2ceac1f | 2536 | goto out; |
2bd0ea18 | 2537 | } |
a2ceac1f DC |
2538 | error = xfs_buf_map_from_irec(mp, map, nmaps, irecs, nirecs); |
2539 | out: | |
2540 | if (irecs != &irec) | |
2541 | kmem_free(irecs); | |
2542 | return error; | |
2543 | } | |
2544 | ||
2545 | /* | |
2546 | * Get a buffer for the dir/attr block. | |
2547 | */ | |
2548 | int | |
2549 | xfs_da_get_buf( | |
2550 | struct xfs_trans *trans, | |
2551 | struct xfs_inode *dp, | |
2552 | xfs_dablk_t bno, | |
2553 | xfs_daddr_t mappedbno, | |
2554 | struct xfs_buf **bpp, | |
2555 | int whichfork) | |
2556 | { | |
2557 | struct xfs_buf *bp; | |
2558 | struct xfs_buf_map map; | |
2559 | struct xfs_buf_map *mapp; | |
2560 | int nmap; | |
2561 | int error; | |
2562 | ||
2563 | *bpp = NULL; | |
2564 | mapp = ↦ | |
2565 | nmap = 1; | |
ff105f75 | 2566 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
a2ceac1f DC |
2567 | &mapp, &nmap); |
2568 | if (error) { | |
2569 | /* mapping a hole is not an error, but we don't continue */ | |
2570 | if (error == -1) | |
2bd0ea18 | 2571 | error = 0; |
a2ceac1f | 2572 | goto out_free; |
2bd0ea18 | 2573 | } |
a2ceac1f DC |
2574 | |
2575 | bp = xfs_trans_get_buf_map(trans, dp->i_mount->m_ddev_targp, | |
2576 | mapp, nmap, 0); | |
12b53197 | 2577 | error = bp ? bp->b_error : -EIO; |
a2ceac1f | 2578 | if (error) { |
5a35bf2c DC |
2579 | if (bp) |
2580 | xfs_trans_brelse(trans, bp); | |
a2ceac1f DC |
2581 | goto out_free; |
2582 | } | |
2583 | ||
2584 | *bpp = bp; | |
2585 | ||
2586 | out_free: | |
2587 | if (mapp != &map) | |
2588 | kmem_free(mapp); | |
2589 | ||
2590 | return error; | |
2591 | } | |
2592 | ||
2593 | /* | |
2594 | * Get a buffer for the dir/attr block, fill in the contents. | |
2595 | */ | |
2596 | int | |
2597 | xfs_da_read_buf( | |
2598 | struct xfs_trans *trans, | |
2599 | struct xfs_inode *dp, | |
2600 | xfs_dablk_t bno, | |
2601 | xfs_daddr_t mappedbno, | |
2602 | struct xfs_buf **bpp, | |
2603 | int whichfork, | |
2604 | const struct xfs_buf_ops *ops) | |
2605 | { | |
2606 | struct xfs_buf *bp; | |
2607 | struct xfs_buf_map map; | |
2608 | struct xfs_buf_map *mapp; | |
2609 | int nmap; | |
2610 | int error; | |
2611 | ||
2612 | *bpp = NULL; | |
2613 | mapp = ↦ | |
2614 | nmap = 1; | |
ff105f75 | 2615 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
a2ceac1f DC |
2616 | &mapp, &nmap); |
2617 | if (error) { | |
2618 | /* mapping a hole is not an error, but we don't continue */ | |
2619 | if (error == -1) | |
2620 | error = 0; | |
2621 | goto out_free; | |
2622 | } | |
2623 | ||
2624 | error = xfs_trans_read_buf_map(dp->i_mount, trans, | |
2625 | dp->i_mount->m_ddev_targp, | |
2626 | mapp, nmap, 0, &bp, ops); | |
2627 | if (error) | |
2628 | goto out_free; | |
2629 | ||
2630 | if (whichfork == XFS_ATTR_FORK) | |
2631 | xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF); | |
2bd0ea18 | 2632 | else |
a2ceac1f | 2633 | xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF); |
a2ceac1f DC |
2634 | *bpp = bp; |
2635 | out_free: | |
2bd0ea18 | 2636 | if (mapp != &map) |
5e656dbb | 2637 | kmem_free(mapp); |
2bd0ea18 | 2638 | |
a2ceac1f | 2639 | return error; |
2bd0ea18 NS |
2640 | } |
2641 | ||
2642 | /* | |
5e656dbb | 2643 | * Readahead the dir/attr block. |
2bd0ea18 | 2644 | */ |
3a65062f | 2645 | int |
5e656dbb | 2646 | xfs_da_reada_buf( |
a2ceac1f DC |
2647 | struct xfs_inode *dp, |
2648 | xfs_dablk_t bno, | |
2649 | xfs_daddr_t mappedbno, | |
2650 | int whichfork, | |
2651 | const struct xfs_buf_ops *ops) | |
2bd0ea18 | 2652 | { |
a2ceac1f DC |
2653 | struct xfs_buf_map map; |
2654 | struct xfs_buf_map *mapp; | |
2655 | int nmap; | |
2656 | int error; | |
2bd0ea18 | 2657 | |
a2ceac1f DC |
2658 | mapp = ↦ |
2659 | nmap = 1; | |
ff105f75 | 2660 | error = xfs_dabuf_map(dp, bno, mappedbno, whichfork, |
a2ceac1f DC |
2661 | &mapp, &nmap); |
2662 | if (error) { | |
2663 | /* mapping a hole is not an error, but we don't continue */ | |
2664 | if (error == -1) | |
2665 | error = 0; | |
2666 | goto out_free; | |
2667 | } | |
2668 | ||
2669 | mappedbno = mapp[0].bm_bn; | |
2670 | xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops); | |
2671 | ||
2672 | out_free: | |
2673 | if (mapp != &map) | |
2674 | kmem_free(mapp); | |
2675 | ||
3a65062f | 2676 | return error; |
2bd0ea18 | 2677 | } |