]>
Commit | Line | Data |
---|---|---|
37b3b4d6 | 1 | // SPDX-License-Identifier: GPL-2.0 |
2bd0ea18 | 2 | /* |
da23017d | 3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
88b32f06 | 4 | * Copyright (c) 2013 Red Hat, Inc. |
da23017d | 5 | * All Rights Reserved. |
2bd0ea18 | 6 | */ |
9c799827 | 7 | #include "libxfs_priv.h" |
b626fb59 DC |
8 | #include "xfs_fs.h" |
9 | #include "xfs_shared.h" | |
10 | #include "xfs_format.h" | |
11 | #include "xfs_log_format.h" | |
12 | #include "xfs_trans_resv.h" | |
13 | #include "xfs_bit.h" | |
14 | #include "xfs_mount.h" | |
e169cc9b | 15 | #include "xfs_inode.h" |
b626fb59 DC |
16 | #include "xfs_dir2.h" |
17 | #include "xfs_dir2_priv.h" | |
b626fb59 | 18 | #include "xfs_trans.h" |
b626fb59 DC |
19 | #include "xfs_bmap.h" |
20 | #include "xfs_attr_leaf.h" | |
21 | #include "xfs_trace.h" | |
2bd0ea18 NS |
22 | |
23 | /* | |
24 | * xfs_da_btree.c | |
25 | * | |
26 | * Routines to implement directories as Btrees of hashed names. | |
27 | */ | |
28 | ||
5e656dbb BN |
29 | /*======================================================================== |
30 | * Function prototypes for the kernel. | |
31 | *========================================================================*/ | |
32 | ||
33 | /* | |
34 | * Routines used for growing the Btree. | |
35 | */ | |
88b32f06 | 36 | STATIC int xfs_da3_root_split(xfs_da_state_t *state, |
5e656dbb BN |
37 | xfs_da_state_blk_t *existing_root, |
38 | xfs_da_state_blk_t *new_child); | |
88b32f06 | 39 | STATIC int xfs_da3_node_split(xfs_da_state_t *state, |
5e656dbb BN |
40 | xfs_da_state_blk_t *existing_blk, |
41 | xfs_da_state_blk_t *split_blk, | |
42 | xfs_da_state_blk_t *blk_to_add, | |
43 | int treelevel, | |
44 | int *result); | |
88b32f06 | 45 | STATIC void xfs_da3_node_rebalance(xfs_da_state_t *state, |
5e656dbb BN |
46 | xfs_da_state_blk_t *node_blk_1, |
47 | xfs_da_state_blk_t *node_blk_2); | |
88b32f06 | 48 | STATIC void xfs_da3_node_add(xfs_da_state_t *state, |
5e656dbb BN |
49 | xfs_da_state_blk_t *old_node_blk, |
50 | xfs_da_state_blk_t *new_node_blk); | |
51 | ||
52 | /* | |
53 | * Routines used for shrinking the Btree. | |
54 | */ | |
88b32f06 | 55 | STATIC int xfs_da3_root_join(xfs_da_state_t *state, |
5e656dbb | 56 | xfs_da_state_blk_t *root_blk); |
88b32f06 DC |
57 | STATIC int xfs_da3_node_toosmall(xfs_da_state_t *state, int *retval); |
58 | STATIC void xfs_da3_node_remove(xfs_da_state_t *state, | |
5e656dbb | 59 | xfs_da_state_blk_t *drop_blk); |
88b32f06 | 60 | STATIC void xfs_da3_node_unbalance(xfs_da_state_t *state, |
5e656dbb BN |
61 | xfs_da_state_blk_t *src_node_blk, |
62 | xfs_da_state_blk_t *dst_node_blk); | |
63 | ||
64 | /* | |
65 | * Utility routines. | |
66 | */ | |
88b32f06 | 67 | STATIC int xfs_da3_blk_unlink(xfs_da_state_t *state, |
5e656dbb BN |
68 | xfs_da_state_blk_t *drop_blk, |
69 | xfs_da_state_blk_t *save_blk); | |
5e656dbb | 70 | |
88b32f06 | 71 | |
2e1394fc | 72 | struct kmem_cache *xfs_da_state_cache; /* anchor for dir/attr state */ |
88b32f06 DC |
73 | |
74 | /* | |
75 | * Allocate a dir-state structure. | |
76 | * We don't put them on the stack since they're large. | |
77 | */ | |
2db3075b CM |
78 | struct xfs_da_state * |
79 | xfs_da_state_alloc( | |
80 | struct xfs_da_args *args) | |
88b32f06 | 81 | { |
2db3075b CM |
82 | struct xfs_da_state *state; |
83 | ||
2e1394fc | 84 | state = kmem_cache_zalloc(xfs_da_state_cache, GFP_NOFS | __GFP_NOFAIL); |
2db3075b CM |
85 | state->args = args; |
86 | state->mp = args->dp->i_mount; | |
87 | return state; | |
88b32f06 DC |
88 | } |
89 | ||
90 | /* | |
91 | * Kill the altpath contents of a da-state structure. | |
92 | */ | |
93 | STATIC void | |
94 | xfs_da_state_kill_altpath(xfs_da_state_t *state) | |
95 | { | |
96 | int i; | |
97 | ||
98 | for (i = 0; i < state->altpath.active; i++) | |
99 | state->altpath.blk[i].bp = NULL; | |
100 | state->altpath.active = 0; | |
101 | } | |
102 | ||
103 | /* | |
104 | * Free a da-state structure. | |
105 | */ | |
106 | void | |
107 | xfs_da_state_free(xfs_da_state_t *state) | |
108 | { | |
109 | xfs_da_state_kill_altpath(state); | |
110 | #ifdef DEBUG | |
111 | memset((char *)state, 0, sizeof(*state)); | |
112 | #endif /* DEBUG */ | |
2e1394fc | 113 | kmem_cache_free(xfs_da_state_cache, state); |
88b32f06 DC |
114 | } |
115 | ||
ecc6ab2b DW |
116 | void |
117 | xfs_da_state_reset( | |
118 | struct xfs_da_state *state, | |
119 | struct xfs_da_args *args) | |
120 | { | |
121 | xfs_da_state_kill_altpath(state); | |
122 | memset(state, 0, sizeof(struct xfs_da_state)); | |
123 | state->args = args; | |
124 | state->mp = state->args->dp->i_mount; | |
125 | } | |
126 | ||
59ab3748 CH |
127 | static inline int xfs_dabuf_nfsb(struct xfs_mount *mp, int whichfork) |
128 | { | |
129 | if (whichfork == XFS_DATA_FORK) | |
130 | return mp->m_dir_geo->fsbcount; | |
131 | return mp->m_attr_geo->fsbcount; | |
132 | } | |
133 | ||
08c16786 CH |
134 | void |
135 | xfs_da3_node_hdr_from_disk( | |
136 | struct xfs_mount *mp, | |
137 | struct xfs_da3_icnode_hdr *to, | |
138 | struct xfs_da_intnode *from) | |
139 | { | |
94541a16 | 140 | if (xfs_has_crc(mp)) { |
08c16786 CH |
141 | struct xfs_da3_intnode *from3 = (struct xfs_da3_intnode *)from; |
142 | ||
143 | to->forw = be32_to_cpu(from3->hdr.info.hdr.forw); | |
144 | to->back = be32_to_cpu(from3->hdr.info.hdr.back); | |
145 | to->magic = be16_to_cpu(from3->hdr.info.hdr.magic); | |
146 | to->count = be16_to_cpu(from3->hdr.__count); | |
147 | to->level = be16_to_cpu(from3->hdr.__level); | |
8faa51a8 | 148 | to->btree = from3->__btree; |
08c16786 CH |
149 | ASSERT(to->magic == XFS_DA3_NODE_MAGIC); |
150 | } else { | |
151 | to->forw = be32_to_cpu(from->hdr.info.forw); | |
152 | to->back = be32_to_cpu(from->hdr.info.back); | |
153 | to->magic = be16_to_cpu(from->hdr.info.magic); | |
154 | to->count = be16_to_cpu(from->hdr.__count); | |
155 | to->level = be16_to_cpu(from->hdr.__level); | |
8faa51a8 | 156 | to->btree = from->__btree; |
08c16786 CH |
157 | ASSERT(to->magic == XFS_DA_NODE_MAGIC); |
158 | } | |
159 | } | |
160 | ||
b81278fa CH |
161 | void |
162 | xfs_da3_node_hdr_to_disk( | |
163 | struct xfs_mount *mp, | |
164 | struct xfs_da_intnode *to, | |
165 | struct xfs_da3_icnode_hdr *from) | |
166 | { | |
94541a16 | 167 | if (xfs_has_crc(mp)) { |
b81278fa CH |
168 | struct xfs_da3_intnode *to3 = (struct xfs_da3_intnode *)to; |
169 | ||
170 | ASSERT(from->magic == XFS_DA3_NODE_MAGIC); | |
171 | to3->hdr.info.hdr.forw = cpu_to_be32(from->forw); | |
172 | to3->hdr.info.hdr.back = cpu_to_be32(from->back); | |
173 | to3->hdr.info.hdr.magic = cpu_to_be16(from->magic); | |
174 | to3->hdr.__count = cpu_to_be16(from->count); | |
175 | to3->hdr.__level = cpu_to_be16(from->level); | |
176 | } else { | |
177 | ASSERT(from->magic == XFS_DA_NODE_MAGIC); | |
178 | to->hdr.info.forw = cpu_to_be32(from->forw); | |
179 | to->hdr.info.back = cpu_to_be32(from->back); | |
180 | to->hdr.info.magic = cpu_to_be16(from->magic); | |
181 | to->hdr.__count = cpu_to_be16(from->count); | |
182 | to->hdr.__level = cpu_to_be16(from->level); | |
183 | } | |
184 | } | |
185 | ||
6b27f70a BF |
186 | /* |
187 | * Verify an xfs_da3_blkinfo structure. Note that the da3 fields are only | |
188 | * accessible on v5 filesystems. This header format is common across da node, | |
189 | * attr leaf and dir leaf blocks. | |
190 | */ | |
191 | xfs_failaddr_t | |
192 | xfs_da3_blkinfo_verify( | |
193 | struct xfs_buf *bp, | |
194 | struct xfs_da3_blkinfo *hdr3) | |
195 | { | |
7861ef77 | 196 | struct xfs_mount *mp = bp->b_mount; |
6b27f70a BF |
197 | struct xfs_da_blkinfo *hdr = &hdr3->hdr; |
198 | ||
9e26de8f | 199 | if (!xfs_verify_magic16(bp, hdr->magic)) |
6b27f70a BF |
200 | return __this_address; |
201 | ||
94541a16 | 202 | if (xfs_has_crc(mp)) { |
6b27f70a BF |
203 | if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid)) |
204 | return __this_address; | |
f1208396 | 205 | if (be64_to_cpu(hdr3->blkno) != xfs_buf_daddr(bp)) |
6b27f70a BF |
206 | return __this_address; |
207 | if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn))) | |
208 | return __this_address; | |
209 | } | |
210 | ||
9e26de8f | 211 | return NULL; |
6b27f70a BF |
212 | } |
213 | ||
bc01119d | 214 | static xfs_failaddr_t |
88b32f06 | 215 | xfs_da3_node_verify( |
a2ceac1f DC |
216 | struct xfs_buf *bp) |
217 | { | |
7861ef77 | 218 | struct xfs_mount *mp = bp->b_mount; |
88b32f06 DC |
219 | struct xfs_da_intnode *hdr = bp->b_addr; |
220 | struct xfs_da3_icnode_hdr ichdr; | |
6b27f70a | 221 | xfs_failaddr_t fa; |
ff105f75 | 222 | |
08c16786 | 223 | xfs_da3_node_hdr_from_disk(mp, &ichdr, hdr); |
88b32f06 | 224 | |
6b27f70a BF |
225 | fa = xfs_da3_blkinfo_verify(bp, bp->b_addr); |
226 | if (fa) | |
227 | return fa; | |
68dbe77f | 228 | |
88b32f06 | 229 | if (ichdr.level == 0) |
bc01119d | 230 | return __this_address; |
88b32f06 | 231 | if (ichdr.level > XFS_DA_NODE_MAXDEPTH) |
bc01119d | 232 | return __this_address; |
88b32f06 | 233 | if (ichdr.count == 0) |
bc01119d | 234 | return __this_address; |
88b32f06 DC |
235 | |
236 | /* | |
237 | * we don't know if the node is for and attribute or directory tree, | |
238 | * so only fail if the count is outside both bounds | |
239 | */ | |
ff105f75 DC |
240 | if (ichdr.count > mp->m_dir_geo->node_ents && |
241 | ichdr.count > mp->m_attr_geo->node_ents) | |
bc01119d | 242 | return __this_address; |
88b32f06 DC |
243 | |
244 | /* XXX: hash order check? */ | |
a2ceac1f | 245 | |
bc01119d | 246 | return NULL; |
a2ceac1f DC |
247 | } |
248 | ||
249 | static void | |
88b32f06 | 250 | xfs_da3_node_write_verify( |
a2ceac1f DC |
251 | struct xfs_buf *bp) |
252 | { | |
7861ef77 | 253 | struct xfs_mount *mp = bp->b_mount; |
37d086ca | 254 | struct xfs_buf_log_item *bip = bp->b_log_item; |
88b32f06 | 255 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; |
1e697959 | 256 | xfs_failaddr_t fa; |
88b32f06 | 257 | |
1e697959 DW |
258 | fa = xfs_da3_node_verify(bp); |
259 | if (fa) { | |
260 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
88b32f06 DC |
261 | return; |
262 | } | |
263 | ||
b16a427a | 264 | if (!xfs_has_crc(mp)) |
88b32f06 DC |
265 | return; |
266 | ||
267 | if (bip) | |
268 | hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
269 | ||
43b5aeed | 270 | xfs_buf_update_cksum(bp, XFS_DA3_NODE_CRC_OFF); |
a2ceac1f DC |
271 | } |
272 | ||
273 | /* | |
274 | * leaf/node format detection on trees is sketchy, so a node read can be done on | |
275 | * leaf level blocks when detection identifies the tree as a node format tree | |
276 | * incorrectly. In this case, we need to swap the verifier to match the correct | |
277 | * format of the block being read. | |
278 | */ | |
279 | static void | |
88b32f06 | 280 | xfs_da3_node_read_verify( |
a2ceac1f DC |
281 | struct xfs_buf *bp) |
282 | { | |
a2ceac1f | 283 | struct xfs_da_blkinfo *info = bp->b_addr; |
1e697959 | 284 | xfs_failaddr_t fa; |
a2ceac1f DC |
285 | |
286 | switch (be16_to_cpu(info->magic)) { | |
88b32f06 | 287 | case XFS_DA3_NODE_MAGIC: |
45922933 | 288 | if (!xfs_buf_verify_cksum(bp, XFS_DA3_NODE_CRC_OFF)) { |
1e697959 DW |
289 | xfs_verifier_error(bp, -EFSBADCRC, |
290 | __this_address); | |
88b32f06 | 291 | break; |
45922933 | 292 | } |
df9c7d8d | 293 | fallthrough; |
a2ceac1f | 294 | case XFS_DA_NODE_MAGIC: |
1e697959 DW |
295 | fa = xfs_da3_node_verify(bp); |
296 | if (fa) | |
297 | xfs_verifier_error(bp, -EFSCORRUPTED, fa); | |
88b32f06 | 298 | return; |
a2ceac1f | 299 | case XFS_ATTR_LEAF_MAGIC: |
a24374f4 DC |
300 | case XFS_ATTR3_LEAF_MAGIC: |
301 | bp->b_ops = &xfs_attr3_leaf_buf_ops; | |
a2ceac1f DC |
302 | bp->b_ops->verify_read(bp); |
303 | return; | |
304 | case XFS_DIR2_LEAFN_MAGIC: | |
65b80c98 DC |
305 | case XFS_DIR3_LEAFN_MAGIC: |
306 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
a2ceac1f DC |
307 | bp->b_ops->verify_read(bp); |
308 | return; | |
309 | default: | |
1e697959 | 310 | xfs_verifier_error(bp, -EFSCORRUPTED, __this_address); |
a2ceac1f DC |
311 | break; |
312 | } | |
313 | } | |
314 | ||
95d9582b DW |
315 | /* Verify the structure of a da3 block. */ |
316 | static xfs_failaddr_t | |
317 | xfs_da3_node_verify_struct( | |
318 | struct xfs_buf *bp) | |
319 | { | |
320 | struct xfs_da_blkinfo *info = bp->b_addr; | |
321 | ||
322 | switch (be16_to_cpu(info->magic)) { | |
323 | case XFS_DA3_NODE_MAGIC: | |
324 | case XFS_DA_NODE_MAGIC: | |
325 | return xfs_da3_node_verify(bp); | |
326 | case XFS_ATTR_LEAF_MAGIC: | |
327 | case XFS_ATTR3_LEAF_MAGIC: | |
328 | bp->b_ops = &xfs_attr3_leaf_buf_ops; | |
329 | return bp->b_ops->verify_struct(bp); | |
330 | case XFS_DIR2_LEAFN_MAGIC: | |
331 | case XFS_DIR3_LEAFN_MAGIC: | |
332 | bp->b_ops = &xfs_dir3_leafn_buf_ops; | |
333 | return bp->b_ops->verify_struct(bp); | |
334 | default: | |
335 | return __this_address; | |
336 | } | |
337 | } | |
338 | ||
88b32f06 | 339 | const struct xfs_buf_ops xfs_da3_node_buf_ops = { |
a3fac935 | 340 | .name = "xfs_da3_node", |
9e26de8f DW |
341 | .magic16 = { cpu_to_be16(XFS_DA_NODE_MAGIC), |
342 | cpu_to_be16(XFS_DA3_NODE_MAGIC) }, | |
88b32f06 DC |
343 | .verify_read = xfs_da3_node_read_verify, |
344 | .verify_write = xfs_da3_node_write_verify, | |
95d9582b | 345 | .verify_struct = xfs_da3_node_verify_struct, |
a2ceac1f DC |
346 | }; |
347 | ||
02cc4995 CH |
348 | static int |
349 | xfs_da3_node_set_type( | |
350 | struct xfs_trans *tp, | |
351 | struct xfs_buf *bp) | |
352 | { | |
353 | struct xfs_da_blkinfo *info = bp->b_addr; | |
354 | ||
355 | switch (be16_to_cpu(info->magic)) { | |
356 | case XFS_DA_NODE_MAGIC: | |
357 | case XFS_DA3_NODE_MAGIC: | |
358 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); | |
359 | return 0; | |
360 | case XFS_ATTR_LEAF_MAGIC: | |
361 | case XFS_ATTR3_LEAF_MAGIC: | |
362 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_ATTR_LEAF_BUF); | |
363 | return 0; | |
364 | case XFS_DIR2_LEAFN_MAGIC: | |
365 | case XFS_DIR3_LEAFN_MAGIC: | |
366 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF); | |
367 | return 0; | |
368 | default: | |
369 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, tp->t_mountp, | |
370 | info, sizeof(*info)); | |
371 | xfs_trans_brelse(tp, bp); | |
372 | return -EFSCORRUPTED; | |
373 | } | |
374 | } | |
375 | ||
a2ceac1f | 376 | int |
88b32f06 | 377 | xfs_da3_node_read( |
a2ceac1f DC |
378 | struct xfs_trans *tp, |
379 | struct xfs_inode *dp, | |
380 | xfs_dablk_t bno, | |
a2ceac1f | 381 | struct xfs_buf **bpp, |
02cc4995 | 382 | int whichfork) |
a2ceac1f | 383 | { |
02cc4995 | 384 | int error; |
8b4dc4a9 | 385 | |
5f356ae6 | 386 | error = xfs_da_read_buf(tp, dp, bno, 0, bpp, whichfork, |
02cc4995 CH |
387 | &xfs_da3_node_buf_ops); |
388 | if (error || !*bpp || !tp) | |
389 | return error; | |
390 | return xfs_da3_node_set_type(tp, *bpp); | |
391 | } | |
8b4dc4a9 | 392 | |
02cc4995 CH |
393 | int |
394 | xfs_da3_node_read_mapped( | |
395 | struct xfs_trans *tp, | |
396 | struct xfs_inode *dp, | |
397 | xfs_daddr_t mappedbno, | |
398 | struct xfs_buf **bpp, | |
399 | int whichfork) | |
400 | { | |
5f356ae6 | 401 | struct xfs_mount *mp = dp->i_mount; |
02cc4995 CH |
402 | int error; |
403 | ||
5f356ae6 CH |
404 | error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, mappedbno, |
405 | XFS_FSB_TO_BB(mp, xfs_dabuf_nfsb(mp, whichfork)), 0, | |
406 | bpp, &xfs_da3_node_buf_ops); | |
407 | if (error || !*bpp) | |
02cc4995 | 408 | return error; |
5f356ae6 CH |
409 | |
410 | if (whichfork == XFS_ATTR_FORK) | |
411 | xfs_buf_set_ref(*bpp, XFS_ATTR_BTREE_REF); | |
412 | else | |
413 | xfs_buf_set_ref(*bpp, XFS_DIR_BTREE_REF); | |
414 | ||
415 | if (!tp) | |
416 | return 0; | |
02cc4995 | 417 | return xfs_da3_node_set_type(tp, *bpp); |
a2ceac1f DC |
418 | } |
419 | ||
2bd0ea18 NS |
420 | /*======================================================================== |
421 | * Routines used for growing the Btree. | |
422 | *========================================================================*/ | |
423 | ||
424 | /* | |
425 | * Create the initial contents of an intermediate node. | |
426 | */ | |
427 | int | |
88b32f06 DC |
428 | xfs_da3_node_create( |
429 | struct xfs_da_args *args, | |
430 | xfs_dablk_t blkno, | |
431 | int level, | |
432 | struct xfs_buf **bpp, | |
433 | int whichfork) | |
2bd0ea18 | 434 | { |
88b32f06 DC |
435 | struct xfs_da_intnode *node; |
436 | struct xfs_trans *tp = args->trans; | |
437 | struct xfs_mount *mp = tp->t_mountp; | |
438 | struct xfs_da3_icnode_hdr ichdr = {0}; | |
439 | struct xfs_buf *bp; | |
440 | int error; | |
ff105f75 | 441 | struct xfs_inode *dp = args->dp; |
2bd0ea18 | 442 | |
a2ceac1f | 443 | trace_xfs_da_node_create(args); |
88b32f06 | 444 | ASSERT(level <= XFS_DA_NODE_MAXDEPTH); |
a2ceac1f | 445 | |
c1d19744 | 446 | error = xfs_da_get_buf(tp, dp, blkno, &bp, whichfork); |
2bd0ea18 | 447 | if (error) |
af43ca9f | 448 | return error; |
8b4dc4a9 | 449 | bp->b_ops = &xfs_da3_node_buf_ops; |
bdc16ee5 | 450 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); |
a2ceac1f | 451 | node = bp->b_addr; |
2bd0ea18 | 452 | |
b16a427a | 453 | if (xfs_has_crc(mp)) { |
88b32f06 DC |
454 | struct xfs_da3_node_hdr *hdr3 = bp->b_addr; |
455 | ||
a65d8d29 | 456 | memset(hdr3, 0, sizeof(struct xfs_da3_node_hdr)); |
88b32f06 | 457 | ichdr.magic = XFS_DA3_NODE_MAGIC; |
f1208396 | 458 | hdr3->info.blkno = cpu_to_be64(xfs_buf_daddr(bp)); |
88b32f06 | 459 | hdr3->info.owner = cpu_to_be64(args->dp->i_ino); |
9c4e12fb | 460 | uuid_copy(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid); |
88b32f06 DC |
461 | } else { |
462 | ichdr.magic = XFS_DA_NODE_MAGIC; | |
463 | } | |
464 | ichdr.level = level; | |
465 | ||
b81278fa | 466 | xfs_da3_node_hdr_to_disk(dp->i_mount, node, &ichdr); |
a2ceac1f | 467 | xfs_trans_log_buf(tp, bp, |
52be9b6a | 468 | XFS_DA_LOGRANGE(node, &node->hdr, args->geo->node_hdr_size)); |
2bd0ea18 NS |
469 | |
470 | *bpp = bp; | |
af43ca9f | 471 | return 0; |
2bd0ea18 NS |
472 | } |
473 | ||
474 | /* | |
475 | * Split a leaf node, rebalance, then possibly split | |
476 | * intermediate nodes, rebalance, etc. | |
477 | */ | |
478 | int /* error */ | |
88b32f06 DC |
479 | xfs_da3_split( |
480 | struct xfs_da_state *state) | |
2bd0ea18 | 481 | { |
88b32f06 DC |
482 | struct xfs_da_state_blk *oldblk; |
483 | struct xfs_da_state_blk *newblk; | |
484 | struct xfs_da_state_blk *addblk; | |
485 | struct xfs_da_intnode *node; | |
88b32f06 | 486 | int max; |
6bddecbc | 487 | int action = 0; |
88b32f06 DC |
488 | int error; |
489 | int i; | |
2bd0ea18 | 490 | |
a2ceac1f DC |
491 | trace_xfs_da_split(state->args); |
492 | ||
7f92d9ee AH |
493 | if (XFS_TEST_ERROR(false, state->mp, XFS_ERRTAG_DA_LEAF_SPLIT)) |
494 | return -EIO; | |
495 | ||
2bd0ea18 NS |
496 | /* |
497 | * Walk back up the tree splitting/inserting/adjusting as necessary. | |
498 | * If we need to insert and there isn't room, split the node, then | |
499 | * decide which fragment to insert the new block from below into. | |
500 | * Note that we may split the root this way, but we need more fixup. | |
501 | */ | |
502 | max = state->path.active - 1; | |
503 | ASSERT((max >= 0) && (max < XFS_DA_NODE_MAXDEPTH)); | |
504 | ASSERT(state->path.blk[max].magic == XFS_ATTR_LEAF_MAGIC || | |
5e656dbb | 505 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); |
2bd0ea18 NS |
506 | |
507 | addblk = &state->path.blk[max]; /* initial dummy value */ | |
508 | for (i = max; (i >= 0) && addblk; state->path.active--, i--) { | |
509 | oldblk = &state->path.blk[i]; | |
510 | newblk = &state->altpath.blk[i]; | |
511 | ||
512 | /* | |
513 | * If a leaf node then | |
514 | * Allocate a new leaf node, then rebalance across them. | |
515 | * else if an intermediate node then | |
516 | * We split on the last layer, must we split the node? | |
517 | */ | |
518 | switch (oldblk->magic) { | |
519 | case XFS_ATTR_LEAF_MAGIC: | |
a24374f4 | 520 | error = xfs_attr3_leaf_split(state, oldblk, newblk); |
12b53197 | 521 | if ((error != 0) && (error != -ENOSPC)) { |
af43ca9f | 522 | return error; /* GROT: attr is inconsistent */ |
2bd0ea18 NS |
523 | } |
524 | if (!error) { | |
525 | addblk = newblk; | |
526 | break; | |
527 | } | |
528 | /* | |
4280e59d DC |
529 | * Entry wouldn't fit, split the leaf again. The new |
530 | * extrablk will be consumed by xfs_da3_node_split if | |
531 | * the node is split. | |
2bd0ea18 NS |
532 | */ |
533 | state->extravalid = 1; | |
534 | if (state->inleaf) { | |
535 | state->extraafter = 0; /* before newblk */ | |
a2ceac1f | 536 | trace_xfs_attr_leaf_split_before(state->args); |
a24374f4 | 537 | error = xfs_attr3_leaf_split(state, oldblk, |
2bd0ea18 NS |
538 | &state->extrablk); |
539 | } else { | |
540 | state->extraafter = 1; /* after newblk */ | |
a2ceac1f | 541 | trace_xfs_attr_leaf_split_after(state->args); |
a24374f4 | 542 | error = xfs_attr3_leaf_split(state, newblk, |
2bd0ea18 NS |
543 | &state->extrablk); |
544 | } | |
545 | if (error) | |
af43ca9f | 546 | return error; /* GROT: attr inconsistent */ |
2bd0ea18 NS |
547 | addblk = newblk; |
548 | break; | |
2bd0ea18 | 549 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
550 | error = xfs_dir2_leafn_split(state, oldblk, newblk); |
551 | if (error) | |
552 | return error; | |
553 | addblk = newblk; | |
554 | break; | |
555 | case XFS_DA_NODE_MAGIC: | |
88b32f06 | 556 | error = xfs_da3_node_split(state, oldblk, newblk, addblk, |
2bd0ea18 | 557 | max - i, &action); |
2bd0ea18 NS |
558 | addblk->bp = NULL; |
559 | if (error) | |
af43ca9f | 560 | return error; /* GROT: dir is inconsistent */ |
2bd0ea18 NS |
561 | /* |
562 | * Record the newly split block for the next time thru? | |
563 | */ | |
564 | if (action) | |
565 | addblk = newblk; | |
566 | else | |
567 | addblk = NULL; | |
568 | break; | |
569 | } | |
570 | ||
571 | /* | |
572 | * Update the btree to show the new hashval for this child. | |
573 | */ | |
88b32f06 | 574 | xfs_da3_fixhashpath(state, &state->path); |
2bd0ea18 NS |
575 | } |
576 | if (!addblk) | |
af43ca9f | 577 | return 0; |
2bd0ea18 | 578 | |
4280e59d DC |
579 | /* |
580 | * xfs_da3_node_split() should have consumed any extra blocks we added | |
581 | * during a double leaf split in the attr fork. This is guaranteed as | |
582 | * we can't be here if the attr fork only has a single leaf block. | |
583 | */ | |
584 | ASSERT(state->extravalid == 0 || | |
585 | state->path.blk[max].magic == XFS_DIR2_LEAFN_MAGIC); | |
586 | ||
2bd0ea18 NS |
587 | /* |
588 | * Split the root node. | |
589 | */ | |
590 | ASSERT(state->path.active == 0); | |
591 | oldblk = &state->path.blk[0]; | |
88b32f06 | 592 | error = xfs_da3_root_split(state, oldblk, addblk); |
ffab1122 DW |
593 | if (error) |
594 | goto out; | |
2bd0ea18 NS |
595 | |
596 | /* | |
4280e59d DC |
597 | * Update pointers to the node which used to be block 0 and just got |
598 | * bumped because of the addition of a new root node. Note that the | |
599 | * original block 0 could be at any position in the list of blocks in | |
600 | * the tree. | |
88b32f06 | 601 | * |
4280e59d DC |
602 | * Note: the magic numbers and sibling pointers are in the same physical |
603 | * place for both v2 and v3 headers (by design). Hence it doesn't matter | |
604 | * which version of the xfs_da_intnode structure we use here as the | |
605 | * result will be the same using either structure. | |
2bd0ea18 | 606 | */ |
a2ceac1f | 607 | node = oldblk->bp->b_addr; |
46eca962 | 608 | if (node->hdr.info.forw) { |
ffab1122 | 609 | if (be32_to_cpu(node->hdr.info.forw) != addblk->blkno) { |
1be76d11 | 610 | xfs_buf_mark_corrupt(oldblk->bp); |
ffab1122 DW |
611 | error = -EFSCORRUPTED; |
612 | goto out; | |
613 | } | |
4280e59d | 614 | node = addblk->bp->b_addr; |
5e656dbb | 615 | node->hdr.info.back = cpu_to_be32(oldblk->blkno); |
4280e59d DC |
616 | xfs_trans_log_buf(state->args->trans, addblk->bp, |
617 | XFS_DA_LOGRANGE(node, &node->hdr.info, | |
618 | sizeof(node->hdr.info))); | |
2bd0ea18 | 619 | } |
a2ceac1f | 620 | node = oldblk->bp->b_addr; |
5e656dbb | 621 | if (node->hdr.info.back) { |
ffab1122 | 622 | if (be32_to_cpu(node->hdr.info.back) != addblk->blkno) { |
1be76d11 | 623 | xfs_buf_mark_corrupt(oldblk->bp); |
ffab1122 DW |
624 | error = -EFSCORRUPTED; |
625 | goto out; | |
626 | } | |
4280e59d | 627 | node = addblk->bp->b_addr; |
5e656dbb | 628 | node->hdr.info.forw = cpu_to_be32(oldblk->blkno); |
4280e59d DC |
629 | xfs_trans_log_buf(state->args->trans, addblk->bp, |
630 | XFS_DA_LOGRANGE(node, &node->hdr.info, | |
631 | sizeof(node->hdr.info))); | |
2bd0ea18 | 632 | } |
ffab1122 | 633 | out: |
2bd0ea18 | 634 | addblk->bp = NULL; |
ffab1122 | 635 | return error; |
2bd0ea18 NS |
636 | } |
637 | ||
638 | /* | |
639 | * Split the root. We have to create a new root and point to the two | |
640 | * parts (the split old root) that we just created. Copy block zero to | |
641 | * the EOF, extending the inode in process. | |
642 | */ | |
643 | STATIC int /* error */ | |
88b32f06 DC |
644 | xfs_da3_root_split( |
645 | struct xfs_da_state *state, | |
646 | struct xfs_da_state_blk *blk1, | |
647 | struct xfs_da_state_blk *blk2) | |
2bd0ea18 | 648 | { |
88b32f06 DC |
649 | struct xfs_da_intnode *node; |
650 | struct xfs_da_intnode *oldroot; | |
651 | struct xfs_da_node_entry *btree; | |
652 | struct xfs_da3_icnode_hdr nodehdr; | |
653 | struct xfs_da_args *args; | |
654 | struct xfs_buf *bp; | |
655 | struct xfs_inode *dp; | |
656 | struct xfs_trans *tp; | |
88b32f06 DC |
657 | struct xfs_dir2_leaf *leaf; |
658 | xfs_dablk_t blkno; | |
659 | int level; | |
660 | int error; | |
661 | int size; | |
2bd0ea18 | 662 | |
a2ceac1f DC |
663 | trace_xfs_da_root_split(state->args); |
664 | ||
2bd0ea18 NS |
665 | /* |
666 | * Copy the existing (incorrect) block from the root node position | |
667 | * to a free space somewhere. | |
668 | */ | |
669 | args = state->args; | |
2bd0ea18 NS |
670 | error = xfs_da_grow_inode(args, &blkno); |
671 | if (error) | |
88b32f06 DC |
672 | return error; |
673 | ||
2bd0ea18 NS |
674 | dp = args->dp; |
675 | tp = args->trans; | |
c1d19744 | 676 | error = xfs_da_get_buf(tp, dp, blkno, &bp, args->whichfork); |
2bd0ea18 | 677 | if (error) |
88b32f06 | 678 | return error; |
a2ceac1f DC |
679 | node = bp->b_addr; |
680 | oldroot = blk1->bp->b_addr; | |
88b32f06 DC |
681 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || |
682 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)) { | |
19ebedcf | 683 | struct xfs_da3_icnode_hdr icnodehdr; |
88b32f06 | 684 | |
08c16786 | 685 | xfs_da3_node_hdr_from_disk(dp->i_mount, &icnodehdr, oldroot); |
8faa51a8 | 686 | btree = icnodehdr.btree; |
19ebedcf DC |
687 | size = (int)((char *)&btree[icnodehdr.count] - (char *)oldroot); |
688 | level = icnodehdr.level; | |
8b4dc4a9 DC |
689 | |
690 | /* | |
691 | * we are about to copy oldroot to bp, so set up the type | |
692 | * of bp while we know exactly what it will be. | |
693 | */ | |
bdc16ee5 | 694 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DA_NODE_BUF); |
2bd0ea18 | 695 | } else { |
65b80c98 | 696 | struct xfs_dir3_icleaf_hdr leafhdr; |
65b80c98 | 697 | |
2bd0ea18 | 698 | leaf = (xfs_dir2_leaf_t *)oldroot; |
9db68faf | 699 | xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf); |
65b80c98 DC |
700 | |
701 | ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC || | |
702 | leafhdr.magic == XFS_DIR3_LEAFN_MAGIC); | |
a2279497 CH |
703 | size = (int)((char *)&leafhdr.ents[leafhdr.count] - |
704 | (char *)leaf); | |
88b32f06 | 705 | level = 0; |
8b4dc4a9 DC |
706 | |
707 | /* | |
708 | * we are about to copy oldroot to bp, so set up the type | |
709 | * of bp while we know exactly what it will be. | |
710 | */ | |
bdc16ee5 | 711 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF); |
2bd0ea18 | 712 | } |
88b32f06 DC |
713 | |
714 | /* | |
715 | * we can copy most of the information in the node from one block to | |
716 | * another, but for CRC enabled headers we have to make sure that the | |
717 | * block specific identifiers are kept intact. We update the buffer | |
718 | * directly for this. | |
719 | */ | |
32181a02 | 720 | memcpy(node, oldroot, size); |
88b32f06 DC |
721 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) || |
722 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
723 | struct xfs_da3_intnode *node3 = (struct xfs_da3_intnode *)node; | |
724 | ||
f1208396 | 725 | node3->hdr.info.blkno = cpu_to_be64(xfs_buf_daddr(bp)); |
88b32f06 | 726 | } |
a2ceac1f DC |
727 | xfs_trans_log_buf(tp, bp, 0, size - 1); |
728 | ||
729 | bp->b_ops = blk1->bp->b_ops; | |
e26915ee | 730 | xfs_trans_buf_copy_type(bp, blk1->bp); |
2bd0ea18 NS |
731 | blk1->bp = bp; |
732 | blk1->blkno = blkno; | |
733 | ||
734 | /* | |
735 | * Set up the new root node. | |
736 | */ | |
88b32f06 | 737 | error = xfs_da3_node_create(args, |
ff105f75 | 738 | (args->whichfork == XFS_DATA_FORK) ? args->geo->leafblk : 0, |
88b32f06 | 739 | level + 1, &bp, args->whichfork); |
2bd0ea18 | 740 | if (error) |
88b32f06 DC |
741 | return error; |
742 | ||
a2ceac1f | 743 | node = bp->b_addr; |
08c16786 | 744 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node); |
8faa51a8 | 745 | btree = nodehdr.btree; |
88b32f06 DC |
746 | btree[0].hashval = cpu_to_be32(blk1->hashval); |
747 | btree[0].before = cpu_to_be32(blk1->blkno); | |
748 | btree[1].hashval = cpu_to_be32(blk2->hashval); | |
749 | btree[1].before = cpu_to_be32(blk2->blkno); | |
750 | nodehdr.count = 2; | |
b81278fa | 751 | xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr); |
6bef826c NS |
752 | |
753 | #ifdef DEBUG | |
65b80c98 DC |
754 | if (oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
755 | oldroot->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
ff105f75 DC |
756 | ASSERT(blk1->blkno >= args->geo->leafblk && |
757 | blk1->blkno < args->geo->freeblk); | |
758 | ASSERT(blk2->blkno >= args->geo->leafblk && | |
759 | blk2->blkno < args->geo->freeblk); | |
2bd0ea18 | 760 | } |
6bef826c NS |
761 | #endif |
762 | ||
2bd0ea18 | 763 | /* Header is already logged by xfs_da_node_create */ |
a2ceac1f | 764 | xfs_trans_log_buf(tp, bp, |
88b32f06 | 765 | XFS_DA_LOGRANGE(node, btree, sizeof(xfs_da_node_entry_t) * 2)); |
2bd0ea18 | 766 | |
88b32f06 | 767 | return 0; |
2bd0ea18 NS |
768 | } |
769 | ||
770 | /* | |
771 | * Split the node, rebalance, then add the new entry. | |
772 | */ | |
773 | STATIC int /* error */ | |
88b32f06 DC |
774 | xfs_da3_node_split( |
775 | struct xfs_da_state *state, | |
776 | struct xfs_da_state_blk *oldblk, | |
777 | struct xfs_da_state_blk *newblk, | |
778 | struct xfs_da_state_blk *addblk, | |
779 | int treelevel, | |
780 | int *result) | |
2bd0ea18 | 781 | { |
88b32f06 DC |
782 | struct xfs_da_intnode *node; |
783 | struct xfs_da3_icnode_hdr nodehdr; | |
784 | xfs_dablk_t blkno; | |
785 | int newcount; | |
786 | int error; | |
787 | int useextra; | |
ff105f75 | 788 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 789 | |
a2ceac1f DC |
790 | trace_xfs_da_node_split(state->args); |
791 | ||
792 | node = oldblk->bp->b_addr; | |
08c16786 | 793 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node); |
2bd0ea18 NS |
794 | |
795 | /* | |
5e656dbb | 796 | * With V2 dirs the extra block is data or freespace. |
2bd0ea18 | 797 | */ |
5e656dbb | 798 | useextra = state->extravalid && state->args->whichfork == XFS_ATTR_FORK; |
2bd0ea18 NS |
799 | newcount = 1 + useextra; |
800 | /* | |
801 | * Do we have to split the node? | |
802 | */ | |
ff105f75 | 803 | if (nodehdr.count + newcount > state->args->geo->node_ents) { |
2bd0ea18 NS |
804 | /* |
805 | * Allocate a new node, add to the doubly linked chain of | |
806 | * nodes, then move some of our excess entries into it. | |
807 | */ | |
808 | error = xfs_da_grow_inode(state->args, &blkno); | |
809 | if (error) | |
af43ca9f | 810 | return error; /* GROT: dir is inconsistent */ |
5000d01d | 811 | |
88b32f06 | 812 | error = xfs_da3_node_create(state->args, blkno, treelevel, |
2bd0ea18 NS |
813 | &newblk->bp, state->args->whichfork); |
814 | if (error) | |
af43ca9f | 815 | return error; /* GROT: dir is inconsistent */ |
2bd0ea18 NS |
816 | newblk->blkno = blkno; |
817 | newblk->magic = XFS_DA_NODE_MAGIC; | |
88b32f06 DC |
818 | xfs_da3_node_rebalance(state, oldblk, newblk); |
819 | error = xfs_da3_blk_link(state, oldblk, newblk); | |
2bd0ea18 | 820 | if (error) |
af43ca9f | 821 | return error; |
2bd0ea18 NS |
822 | *result = 1; |
823 | } else { | |
824 | *result = 0; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Insert the new entry(s) into the correct block | |
829 | * (updating last hashval in the process). | |
830 | * | |
88b32f06 | 831 | * xfs_da3_node_add() inserts BEFORE the given index, |
2bd0ea18 NS |
832 | * and as a result of using node_lookup_int() we always |
833 | * point to a valid entry (not after one), but a split | |
834 | * operation always results in a new block whose hashvals | |
835 | * FOLLOW the current block. | |
836 | * | |
837 | * If we had double-split op below us, then add the extra block too. | |
838 | */ | |
a2ceac1f | 839 | node = oldblk->bp->b_addr; |
08c16786 | 840 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node); |
88b32f06 | 841 | if (oldblk->index <= nodehdr.count) { |
2bd0ea18 | 842 | oldblk->index++; |
88b32f06 | 843 | xfs_da3_node_add(state, oldblk, addblk); |
2bd0ea18 NS |
844 | if (useextra) { |
845 | if (state->extraafter) | |
846 | oldblk->index++; | |
88b32f06 | 847 | xfs_da3_node_add(state, oldblk, &state->extrablk); |
2bd0ea18 NS |
848 | state->extravalid = 0; |
849 | } | |
850 | } else { | |
851 | newblk->index++; | |
88b32f06 | 852 | xfs_da3_node_add(state, newblk, addblk); |
2bd0ea18 NS |
853 | if (useextra) { |
854 | if (state->extraafter) | |
855 | newblk->index++; | |
88b32f06 | 856 | xfs_da3_node_add(state, newblk, &state->extrablk); |
2bd0ea18 NS |
857 | state->extravalid = 0; |
858 | } | |
859 | } | |
860 | ||
af43ca9f | 861 | return 0; |
2bd0ea18 NS |
862 | } |
863 | ||
864 | /* | |
865 | * Balance the btree elements between two intermediate nodes, | |
866 | * usually one full and one empty. | |
867 | * | |
868 | * NOTE: if blk2 is empty, then it will get the upper half of blk1. | |
869 | */ | |
870 | STATIC void | |
88b32f06 DC |
871 | xfs_da3_node_rebalance( |
872 | struct xfs_da_state *state, | |
873 | struct xfs_da_state_blk *blk1, | |
874 | struct xfs_da_state_blk *blk2) | |
2bd0ea18 | 875 | { |
88b32f06 DC |
876 | struct xfs_da_intnode *node1; |
877 | struct xfs_da_intnode *node2; | |
88b32f06 DC |
878 | struct xfs_da_node_entry *btree1; |
879 | struct xfs_da_node_entry *btree2; | |
880 | struct xfs_da_node_entry *btree_s; | |
881 | struct xfs_da_node_entry *btree_d; | |
882 | struct xfs_da3_icnode_hdr nodehdr1; | |
883 | struct xfs_da3_icnode_hdr nodehdr2; | |
884 | struct xfs_trans *tp; | |
885 | int count; | |
886 | int tmp; | |
887 | int swap = 0; | |
ff105f75 | 888 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 889 | |
a2ceac1f DC |
890 | trace_xfs_da_node_rebalance(state->args); |
891 | ||
892 | node1 = blk1->bp->b_addr; | |
893 | node2 = blk2->bp->b_addr; | |
08c16786 CH |
894 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1); |
895 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2); | |
8faa51a8 CH |
896 | btree1 = nodehdr1.btree; |
897 | btree2 = nodehdr2.btree; | |
88b32f06 | 898 | |
2bd0ea18 NS |
899 | /* |
900 | * Figure out how many entries need to move, and in which direction. | |
901 | * Swap the nodes around if that makes it simpler. | |
902 | */ | |
88b32f06 DC |
903 | if (nodehdr1.count > 0 && nodehdr2.count > 0 && |
904 | ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) || | |
905 | (be32_to_cpu(btree2[nodehdr2.count - 1].hashval) < | |
906 | be32_to_cpu(btree1[nodehdr1.count - 1].hashval)))) { | |
016bb3d1 | 907 | swap(node1, node2); |
08c16786 CH |
908 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1); |
909 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2); | |
8faa51a8 CH |
910 | btree1 = nodehdr1.btree; |
911 | btree2 = nodehdr2.btree; | |
88b32f06 | 912 | swap = 1; |
2bd0ea18 | 913 | } |
88b32f06 DC |
914 | |
915 | count = (nodehdr1.count - nodehdr2.count) / 2; | |
2bd0ea18 NS |
916 | if (count == 0) |
917 | return; | |
918 | tp = state->args->trans; | |
919 | /* | |
920 | * Two cases: high-to-low and low-to-high. | |
921 | */ | |
922 | if (count > 0) { | |
923 | /* | |
924 | * Move elements in node2 up to make a hole. | |
925 | */ | |
88b32f06 DC |
926 | tmp = nodehdr2.count; |
927 | if (tmp > 0) { | |
2bd0ea18 | 928 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
88b32f06 DC |
929 | btree_s = &btree2[0]; |
930 | btree_d = &btree2[count]; | |
32181a02 | 931 | memmove(btree_d, btree_s, tmp); |
2bd0ea18 NS |
932 | } |
933 | ||
934 | /* | |
935 | * Move the req'd B-tree elements from high in node1 to | |
936 | * low in node2. | |
937 | */ | |
88b32f06 | 938 | nodehdr2.count += count; |
2bd0ea18 | 939 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); |
7b02556d | 940 | btree_s = &btree1[nodehdr1.count - count]; |
88b32f06 | 941 | btree_d = &btree2[0]; |
32181a02 | 942 | memcpy(btree_d, btree_s, tmp); |
88b32f06 | 943 | nodehdr1.count -= count; |
2bd0ea18 NS |
944 | } else { |
945 | /* | |
946 | * Move the req'd B-tree elements from low in node2 to | |
947 | * high in node1. | |
948 | */ | |
949 | count = -count; | |
950 | tmp = count * (uint)sizeof(xfs_da_node_entry_t); | |
88b32f06 DC |
951 | btree_s = &btree2[0]; |
952 | btree_d = &btree1[nodehdr1.count]; | |
32181a02 | 953 | memcpy(btree_d, btree_s, tmp); |
88b32f06 DC |
954 | nodehdr1.count += count; |
955 | ||
a2ceac1f | 956 | xfs_trans_log_buf(tp, blk1->bp, |
2bd0ea18 NS |
957 | XFS_DA_LOGRANGE(node1, btree_d, tmp)); |
958 | ||
959 | /* | |
960 | * Move elements in node2 down to fill the hole. | |
961 | */ | |
88b32f06 | 962 | tmp = nodehdr2.count - count; |
2bd0ea18 | 963 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
88b32f06 DC |
964 | btree_s = &btree2[count]; |
965 | btree_d = &btree2[0]; | |
32181a02 | 966 | memmove(btree_d, btree_s, tmp); |
88b32f06 | 967 | nodehdr2.count -= count; |
2bd0ea18 NS |
968 | } |
969 | ||
970 | /* | |
971 | * Log header of node 1 and all current bits of node 2. | |
972 | */ | |
b81278fa | 973 | xfs_da3_node_hdr_to_disk(dp->i_mount, node1, &nodehdr1); |
a2ceac1f | 974 | xfs_trans_log_buf(tp, blk1->bp, |
52be9b6a CH |
975 | XFS_DA_LOGRANGE(node1, &node1->hdr, |
976 | state->args->geo->node_hdr_size)); | |
88b32f06 | 977 | |
b81278fa | 978 | xfs_da3_node_hdr_to_disk(dp->i_mount, node2, &nodehdr2); |
a2ceac1f | 979 | xfs_trans_log_buf(tp, blk2->bp, |
2bd0ea18 | 980 | XFS_DA_LOGRANGE(node2, &node2->hdr, |
52be9b6a | 981 | state->args->geo->node_hdr_size + |
88b32f06 | 982 | (sizeof(btree2[0]) * nodehdr2.count))); |
2bd0ea18 NS |
983 | |
984 | /* | |
985 | * Record the last hashval from each block for upward propagation. | |
986 | * (note: don't use the swapped node pointers) | |
987 | */ | |
88b32f06 DC |
988 | if (swap) { |
989 | node1 = blk1->bp->b_addr; | |
990 | node2 = blk2->bp->b_addr; | |
08c16786 CH |
991 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr1, node1); |
992 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr2, node2); | |
8faa51a8 CH |
993 | btree1 = nodehdr1.btree; |
994 | btree2 = nodehdr2.btree; | |
88b32f06 DC |
995 | } |
996 | blk1->hashval = be32_to_cpu(btree1[nodehdr1.count - 1].hashval); | |
997 | blk2->hashval = be32_to_cpu(btree2[nodehdr2.count - 1].hashval); | |
2bd0ea18 NS |
998 | |
999 | /* | |
1000 | * Adjust the expected index for insertion. | |
1001 | */ | |
88b32f06 DC |
1002 | if (blk1->index >= nodehdr1.count) { |
1003 | blk2->index = blk1->index - nodehdr1.count; | |
1004 | blk1->index = nodehdr1.count + 1; /* make it invalid */ | |
2bd0ea18 NS |
1005 | } |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * Add a new entry to an intermediate node. | |
1010 | */ | |
1011 | STATIC void | |
88b32f06 DC |
1012 | xfs_da3_node_add( |
1013 | struct xfs_da_state *state, | |
1014 | struct xfs_da_state_blk *oldblk, | |
1015 | struct xfs_da_state_blk *newblk) | |
2bd0ea18 | 1016 | { |
88b32f06 DC |
1017 | struct xfs_da_intnode *node; |
1018 | struct xfs_da3_icnode_hdr nodehdr; | |
1019 | struct xfs_da_node_entry *btree; | |
1020 | int tmp; | |
ff105f75 | 1021 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1022 | |
a2ceac1f DC |
1023 | trace_xfs_da_node_add(state->args); |
1024 | ||
1025 | node = oldblk->bp->b_addr; | |
08c16786 | 1026 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node); |
8faa51a8 | 1027 | btree = nodehdr.btree; |
88b32f06 DC |
1028 | |
1029 | ASSERT(oldblk->index >= 0 && oldblk->index <= nodehdr.count); | |
2bd0ea18 | 1030 | ASSERT(newblk->blkno != 0); |
5e656dbb | 1031 | if (state->args->whichfork == XFS_DATA_FORK) |
ff105f75 DC |
1032 | ASSERT(newblk->blkno >= state->args->geo->leafblk && |
1033 | newblk->blkno < state->args->geo->freeblk); | |
2bd0ea18 NS |
1034 | |
1035 | /* | |
1036 | * We may need to make some room before we insert the new node. | |
1037 | */ | |
1038 | tmp = 0; | |
88b32f06 DC |
1039 | if (oldblk->index < nodehdr.count) { |
1040 | tmp = (nodehdr.count - oldblk->index) * (uint)sizeof(*btree); | |
1041 | memmove(&btree[oldblk->index + 1], &btree[oldblk->index], tmp); | |
2bd0ea18 | 1042 | } |
88b32f06 DC |
1043 | btree[oldblk->index].hashval = cpu_to_be32(newblk->hashval); |
1044 | btree[oldblk->index].before = cpu_to_be32(newblk->blkno); | |
a2ceac1f | 1045 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
88b32f06 DC |
1046 | XFS_DA_LOGRANGE(node, &btree[oldblk->index], |
1047 | tmp + sizeof(*btree))); | |
1048 | ||
1049 | nodehdr.count += 1; | |
b81278fa | 1050 | xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr); |
a2ceac1f | 1051 | xfs_trans_log_buf(state->args->trans, oldblk->bp, |
52be9b6a CH |
1052 | XFS_DA_LOGRANGE(node, &node->hdr, |
1053 | state->args->geo->node_hdr_size)); | |
2bd0ea18 NS |
1054 | |
1055 | /* | |
1056 | * Copy the last hash value from the oldblk to propagate upwards. | |
1057 | */ | |
88b32f06 | 1058 | oldblk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
2bd0ea18 NS |
1059 | } |
1060 | ||
1061 | /*======================================================================== | |
1062 | * Routines used for shrinking the Btree. | |
1063 | *========================================================================*/ | |
1064 | ||
1065 | /* | |
1066 | * Deallocate an empty leaf node, remove it from its parent, | |
1067 | * possibly deallocating that block, etc... | |
1068 | */ | |
1069 | int | |
88b32f06 DC |
1070 | xfs_da3_join( |
1071 | struct xfs_da_state *state) | |
2bd0ea18 | 1072 | { |
88b32f06 DC |
1073 | struct xfs_da_state_blk *drop_blk; |
1074 | struct xfs_da_state_blk *save_blk; | |
1075 | int action = 0; | |
1076 | int error; | |
2bd0ea18 | 1077 | |
a2ceac1f DC |
1078 | trace_xfs_da_join(state->args); |
1079 | ||
2bd0ea18 NS |
1080 | drop_blk = &state->path.blk[ state->path.active-1 ]; |
1081 | save_blk = &state->altpath.blk[ state->path.active-1 ]; | |
1082 | ASSERT(state->path.blk[0].magic == XFS_DA_NODE_MAGIC); | |
1083 | ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC || | |
5e656dbb | 1084 | drop_blk->magic == XFS_DIR2_LEAFN_MAGIC); |
2bd0ea18 NS |
1085 | |
1086 | /* | |
1087 | * Walk back up the tree joining/deallocating as necessary. | |
1088 | * When we stop dropping blocks, break out. | |
1089 | */ | |
1090 | for ( ; state->path.active >= 2; drop_blk--, save_blk--, | |
1091 | state->path.active--) { | |
1092 | /* | |
1093 | * See if we can combine the block with a neighbor. | |
1094 | * (action == 0) => no options, just leave | |
1095 | * (action == 1) => coalesce, then unlink | |
1096 | * (action == 2) => block empty, unlink it | |
1097 | */ | |
1098 | switch (drop_blk->magic) { | |
1099 | case XFS_ATTR_LEAF_MAGIC: | |
a24374f4 | 1100 | error = xfs_attr3_leaf_toosmall(state, &action); |
2bd0ea18 | 1101 | if (error) |
af43ca9f | 1102 | return error; |
2bd0ea18 | 1103 | if (action == 0) |
af43ca9f | 1104 | return 0; |
a24374f4 | 1105 | xfs_attr3_leaf_unbalance(state, drop_blk, save_blk); |
2bd0ea18 | 1106 | break; |
2bd0ea18 | 1107 | case XFS_DIR2_LEAFN_MAGIC: |
2bd0ea18 NS |
1108 | error = xfs_dir2_leafn_toosmall(state, &action); |
1109 | if (error) | |
1110 | return error; | |
1111 | if (action == 0) | |
1112 | return 0; | |
1113 | xfs_dir2_leafn_unbalance(state, drop_blk, save_blk); | |
1114 | break; | |
1115 | case XFS_DA_NODE_MAGIC: | |
1116 | /* | |
1117 | * Remove the offending node, fixup hashvals, | |
1118 | * check for a toosmall neighbor. | |
1119 | */ | |
88b32f06 DC |
1120 | xfs_da3_node_remove(state, drop_blk); |
1121 | xfs_da3_fixhashpath(state, &state->path); | |
1122 | error = xfs_da3_node_toosmall(state, &action); | |
2bd0ea18 | 1123 | if (error) |
af43ca9f | 1124 | return error; |
2bd0ea18 NS |
1125 | if (action == 0) |
1126 | return 0; | |
88b32f06 | 1127 | xfs_da3_node_unbalance(state, drop_blk, save_blk); |
2bd0ea18 NS |
1128 | break; |
1129 | } | |
88b32f06 DC |
1130 | xfs_da3_fixhashpath(state, &state->altpath); |
1131 | error = xfs_da3_blk_unlink(state, drop_blk, save_blk); | |
2bd0ea18 NS |
1132 | xfs_da_state_kill_altpath(state); |
1133 | if (error) | |
af43ca9f | 1134 | return error; |
2bd0ea18 NS |
1135 | error = xfs_da_shrink_inode(state->args, drop_blk->blkno, |
1136 | drop_blk->bp); | |
1137 | drop_blk->bp = NULL; | |
1138 | if (error) | |
af43ca9f | 1139 | return error; |
2bd0ea18 NS |
1140 | } |
1141 | /* | |
1142 | * We joined all the way to the top. If it turns out that | |
1143 | * we only have one entry in the root, make the child block | |
1144 | * the new root. | |
1145 | */ | |
88b32f06 DC |
1146 | xfs_da3_node_remove(state, drop_blk); |
1147 | xfs_da3_fixhashpath(state, &state->path); | |
1148 | error = xfs_da3_root_join(state, &state->path.blk[0]); | |
af43ca9f | 1149 | return error; |
2bd0ea18 NS |
1150 | } |
1151 | ||
a2ceac1f DC |
1152 | #ifdef DEBUG |
1153 | static void | |
1154 | xfs_da_blkinfo_onlychild_validate(struct xfs_da_blkinfo *blkinfo, __u16 level) | |
1155 | { | |
1156 | __be16 magic = blkinfo->magic; | |
1157 | ||
1158 | if (level == 1) { | |
1159 | ASSERT(magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || | |
65b80c98 | 1160 | magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
a24374f4 DC |
1161 | magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || |
1162 | magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); | |
88b32f06 DC |
1163 | } else { |
1164 | ASSERT(magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
1165 | magic == cpu_to_be16(XFS_DA3_NODE_MAGIC)); | |
1166 | } | |
a2ceac1f DC |
1167 | ASSERT(!blkinfo->forw); |
1168 | ASSERT(!blkinfo->back); | |
1169 | } | |
1170 | #else /* !DEBUG */ | |
1171 | #define xfs_da_blkinfo_onlychild_validate(blkinfo, level) | |
1172 | #endif /* !DEBUG */ | |
1173 | ||
2bd0ea18 | 1174 | /* |
dfc130f3 | 1175 | * We have only one entry in the root. Copy the only remaining child of |
2bd0ea18 NS |
1176 | * the old root to block 0 as the new root node. |
1177 | */ | |
1178 | STATIC int | |
88b32f06 DC |
1179 | xfs_da3_root_join( |
1180 | struct xfs_da_state *state, | |
1181 | struct xfs_da_state_blk *root_blk) | |
2bd0ea18 | 1182 | { |
88b32f06 DC |
1183 | struct xfs_da_intnode *oldroot; |
1184 | struct xfs_da_args *args; | |
1185 | xfs_dablk_t child; | |
1186 | struct xfs_buf *bp; | |
1187 | struct xfs_da3_icnode_hdr oldroothdr; | |
88b32f06 | 1188 | int error; |
ff105f75 | 1189 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1190 | |
a2ceac1f DC |
1191 | trace_xfs_da_root_join(state->args); |
1192 | ||
2bd0ea18 | 1193 | ASSERT(root_blk->magic == XFS_DA_NODE_MAGIC); |
88b32f06 DC |
1194 | |
1195 | args = state->args; | |
a2ceac1f | 1196 | oldroot = root_blk->bp->b_addr; |
08c16786 | 1197 | xfs_da3_node_hdr_from_disk(dp->i_mount, &oldroothdr, oldroot); |
88b32f06 DC |
1198 | ASSERT(oldroothdr.forw == 0); |
1199 | ASSERT(oldroothdr.back == 0); | |
2bd0ea18 NS |
1200 | |
1201 | /* | |
1202 | * If the root has more than one child, then don't do anything. | |
1203 | */ | |
88b32f06 DC |
1204 | if (oldroothdr.count > 1) |
1205 | return 0; | |
2bd0ea18 NS |
1206 | |
1207 | /* | |
1208 | * Read in the (only) child block, then copy those bytes into | |
1209 | * the root block's buffer and free the original child block. | |
1210 | */ | |
8faa51a8 | 1211 | child = be32_to_cpu(oldroothdr.btree[0].before); |
2bd0ea18 | 1212 | ASSERT(child != 0); |
02cc4995 | 1213 | error = xfs_da3_node_read(args->trans, dp, child, &bp, args->whichfork); |
2bd0ea18 | 1214 | if (error) |
88b32f06 DC |
1215 | return error; |
1216 | xfs_da_blkinfo_onlychild_validate(bp->b_addr, oldroothdr.level); | |
a2ceac1f DC |
1217 | |
1218 | /* | |
1219 | * This could be copying a leaf back into the root block in the case of | |
1220 | * there only being a single leaf block left in the tree. Hence we have | |
1221 | * to update the b_ops pointer as well to match the buffer type change | |
88b32f06 DC |
1222 | * that could occur. For dir3 blocks we also need to update the block |
1223 | * number in the buffer header. | |
a2ceac1f | 1224 | */ |
ff105f75 | 1225 | memcpy(root_blk->bp->b_addr, bp->b_addr, args->geo->blksize); |
a2ceac1f | 1226 | root_blk->bp->b_ops = bp->b_ops; |
8b4dc4a9 | 1227 | xfs_trans_buf_copy_type(root_blk->bp, bp); |
88b32f06 DC |
1228 | if (oldroothdr.magic == XFS_DA3_NODE_MAGIC) { |
1229 | struct xfs_da3_blkinfo *da3 = root_blk->bp->b_addr; | |
f1208396 | 1230 | da3->blkno = cpu_to_be64(xfs_buf_daddr(root_blk->bp)); |
88b32f06 | 1231 | } |
ff105f75 DC |
1232 | xfs_trans_log_buf(args->trans, root_blk->bp, 0, |
1233 | args->geo->blksize - 1); | |
2bd0ea18 | 1234 | error = xfs_da_shrink_inode(args, child, bp); |
af43ca9f | 1235 | return error; |
2bd0ea18 NS |
1236 | } |
1237 | ||
1238 | /* | |
1239 | * Check a node block and its neighbors to see if the block should be | |
1240 | * collapsed into one or the other neighbor. Always keep the block | |
1241 | * with the smaller block number. | |
1242 | * If the current block is over 50% full, don't try to join it, return 0. | |
1243 | * If the block is empty, fill in the state structure and return 2. | |
1244 | * If it can be collapsed, fill in the state structure and return 1. | |
1245 | * If nothing can be done, return 0. | |
1246 | */ | |
1247 | STATIC int | |
88b32f06 DC |
1248 | xfs_da3_node_toosmall( |
1249 | struct xfs_da_state *state, | |
1250 | int *action) | |
2bd0ea18 | 1251 | { |
88b32f06 DC |
1252 | struct xfs_da_intnode *node; |
1253 | struct xfs_da_state_blk *blk; | |
1254 | struct xfs_da_blkinfo *info; | |
1255 | xfs_dablk_t blkno; | |
1256 | struct xfs_buf *bp; | |
1257 | struct xfs_da3_icnode_hdr nodehdr; | |
1258 | int count; | |
1259 | int forward; | |
1260 | int error; | |
1261 | int retval; | |
1262 | int i; | |
ff105f75 | 1263 | struct xfs_inode *dp = state->args->dp; |
a2ceac1f DC |
1264 | |
1265 | trace_xfs_da_node_toosmall(state->args); | |
2bd0ea18 NS |
1266 | |
1267 | /* | |
1268 | * Check for the degenerate case of the block being over 50% full. | |
1269 | * If so, it's not worth even looking to see if we might be able | |
1270 | * to coalesce with a sibling. | |
1271 | */ | |
1272 | blk = &state->path.blk[ state->path.active-1 ]; | |
a2ceac1f | 1273 | info = blk->bp->b_addr; |
2bd0ea18 | 1274 | node = (xfs_da_intnode_t *)info; |
08c16786 | 1275 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node); |
ff105f75 | 1276 | if (nodehdr.count > (state->args->geo->node_ents >> 1)) { |
4ca431fc | 1277 | *action = 0; /* blk over 50%, don't try to join */ |
af43ca9f | 1278 | return 0; /* blk over 50%, don't try to join */ |
2bd0ea18 NS |
1279 | } |
1280 | ||
1281 | /* | |
1282 | * Check for the degenerate case of the block being empty. | |
1283 | * If the block is empty, we'll simply delete it, no need to | |
5e656dbb | 1284 | * coalesce it with a sibling block. We choose (arbitrarily) |
2bd0ea18 NS |
1285 | * to merge with the forward block unless it is NULL. |
1286 | */ | |
88b32f06 | 1287 | if (nodehdr.count == 0) { |
2bd0ea18 NS |
1288 | /* |
1289 | * Make altpath point to the block we want to keep and | |
1290 | * path point to the block we want to drop (this one). | |
1291 | */ | |
5e656dbb | 1292 | forward = (info->forw != 0); |
32181a02 | 1293 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
88b32f06 | 1294 | error = xfs_da3_path_shift(state, &state->altpath, forward, |
2bd0ea18 NS |
1295 | 0, &retval); |
1296 | if (error) | |
af43ca9f | 1297 | return error; |
2bd0ea18 NS |
1298 | if (retval) { |
1299 | *action = 0; | |
1300 | } else { | |
1301 | *action = 2; | |
1302 | } | |
af43ca9f | 1303 | return 0; |
2bd0ea18 NS |
1304 | } |
1305 | ||
1306 | /* | |
1307 | * Examine each sibling block to see if we can coalesce with | |
1308 | * at least 25% free space to spare. We need to figure out | |
1309 | * whether to merge with the forward or the backward block. | |
1310 | * We prefer coalescing with the lower numbered sibling so as | |
1311 | * to shrink a directory over time. | |
1312 | */ | |
ff105f75 DC |
1313 | count = state->args->geo->node_ents; |
1314 | count -= state->args->geo->node_ents >> 2; | |
88b32f06 DC |
1315 | count -= nodehdr.count; |
1316 | ||
2bd0ea18 | 1317 | /* start with smaller blk num */ |
88b32f06 | 1318 | forward = nodehdr.forw < nodehdr.back; |
2bd0ea18 | 1319 | for (i = 0; i < 2; forward = !forward, i++) { |
c9522f4d | 1320 | struct xfs_da3_icnode_hdr thdr; |
2bd0ea18 | 1321 | if (forward) |
88b32f06 | 1322 | blkno = nodehdr.forw; |
2bd0ea18 | 1323 | else |
88b32f06 | 1324 | blkno = nodehdr.back; |
2bd0ea18 NS |
1325 | if (blkno == 0) |
1326 | continue; | |
02cc4995 CH |
1327 | error = xfs_da3_node_read(state->args->trans, dp, blkno, &bp, |
1328 | state->args->whichfork); | |
2bd0ea18 | 1329 | if (error) |
af43ca9f | 1330 | return error; |
2bd0ea18 | 1331 | |
a2ceac1f | 1332 | node = bp->b_addr; |
08c16786 | 1333 | xfs_da3_node_hdr_from_disk(dp->i_mount, &thdr, node); |
a2ceac1f | 1334 | xfs_trans_brelse(state->args->trans, bp); |
88b32f06 | 1335 | |
c9522f4d | 1336 | if (count - thdr.count >= 0) |
2bd0ea18 NS |
1337 | break; /* fits with at least 25% to spare */ |
1338 | } | |
1339 | if (i >= 2) { | |
1340 | *action = 0; | |
88b32f06 | 1341 | return 0; |
2bd0ea18 NS |
1342 | } |
1343 | ||
1344 | /* | |
1345 | * Make altpath point to the block we want to keep (the lower | |
1346 | * numbered block) and path point to the block we want to drop. | |
1347 | */ | |
32181a02 | 1348 | memcpy(&state->altpath, &state->path, sizeof(state->path)); |
2bd0ea18 | 1349 | if (blkno < blk->blkno) { |
88b32f06 | 1350 | error = xfs_da3_path_shift(state, &state->altpath, forward, |
2bd0ea18 | 1351 | 0, &retval); |
2bd0ea18 | 1352 | } else { |
88b32f06 | 1353 | error = xfs_da3_path_shift(state, &state->path, forward, |
2bd0ea18 | 1354 | 0, &retval); |
88b32f06 DC |
1355 | } |
1356 | if (error) | |
1357 | return error; | |
1358 | if (retval) { | |
1359 | *action = 0; | |
1360 | return 0; | |
2bd0ea18 NS |
1361 | } |
1362 | *action = 1; | |
88b32f06 DC |
1363 | return 0; |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Pick up the last hashvalue from an intermediate node. | |
1368 | */ | |
1369 | STATIC uint | |
1370 | xfs_da3_node_lasthash( | |
ff105f75 | 1371 | struct xfs_inode *dp, |
88b32f06 DC |
1372 | struct xfs_buf *bp, |
1373 | int *count) | |
1374 | { | |
88b32f06 DC |
1375 | struct xfs_da3_icnode_hdr nodehdr; |
1376 | ||
8faa51a8 | 1377 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, bp->b_addr); |
88b32f06 DC |
1378 | if (count) |
1379 | *count = nodehdr.count; | |
1380 | if (!nodehdr.count) | |
1381 | return 0; | |
8faa51a8 | 1382 | return be32_to_cpu(nodehdr.btree[nodehdr.count - 1].hashval); |
2bd0ea18 NS |
1383 | } |
1384 | ||
2bd0ea18 NS |
1385 | /* |
1386 | * Walk back up the tree adjusting hash values as necessary, | |
1387 | * when we stop making changes, return. | |
1388 | */ | |
1389 | void | |
88b32f06 DC |
1390 | xfs_da3_fixhashpath( |
1391 | struct xfs_da_state *state, | |
1392 | struct xfs_da_state_path *path) | |
2bd0ea18 | 1393 | { |
88b32f06 DC |
1394 | struct xfs_da_state_blk *blk; |
1395 | struct xfs_da_intnode *node; | |
1396 | struct xfs_da_node_entry *btree; | |
1397 | xfs_dahash_t lasthash=0; | |
1398 | int level; | |
1399 | int count; | |
ff105f75 | 1400 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1401 | |
a2ceac1f DC |
1402 | trace_xfs_da_fixhashpath(state->args); |
1403 | ||
2bd0ea18 NS |
1404 | level = path->active-1; |
1405 | blk = &path->blk[ level ]; | |
1406 | switch (blk->magic) { | |
2bd0ea18 NS |
1407 | case XFS_ATTR_LEAF_MAGIC: |
1408 | lasthash = xfs_attr_leaf_lasthash(blk->bp, &count); | |
1409 | if (count == 0) | |
1410 | return; | |
1411 | break; | |
2bd0ea18 | 1412 | case XFS_DIR2_LEAFN_MAGIC: |
94ffb80a | 1413 | lasthash = xfs_dir2_leaf_lasthash(dp, blk->bp, &count); |
2bd0ea18 NS |
1414 | if (count == 0) |
1415 | return; | |
1416 | break; | |
1417 | case XFS_DA_NODE_MAGIC: | |
ff105f75 | 1418 | lasthash = xfs_da3_node_lasthash(dp, blk->bp, &count); |
2bd0ea18 NS |
1419 | if (count == 0) |
1420 | return; | |
1421 | break; | |
1422 | } | |
1423 | for (blk--, level--; level >= 0; blk--, level--) { | |
88b32f06 DC |
1424 | struct xfs_da3_icnode_hdr nodehdr; |
1425 | ||
a2ceac1f | 1426 | node = blk->bp->b_addr; |
08c16786 | 1427 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node); |
8faa51a8 | 1428 | btree = nodehdr.btree; |
0a3717c9 | 1429 | if (be32_to_cpu(btree[blk->index].hashval) == lasthash) |
2bd0ea18 NS |
1430 | break; |
1431 | blk->hashval = lasthash; | |
88b32f06 | 1432 | btree[blk->index].hashval = cpu_to_be32(lasthash); |
a2ceac1f | 1433 | xfs_trans_log_buf(state->args->trans, blk->bp, |
88b32f06 DC |
1434 | XFS_DA_LOGRANGE(node, &btree[blk->index], |
1435 | sizeof(*btree))); | |
2bd0ea18 | 1436 | |
88b32f06 | 1437 | lasthash = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
2bd0ea18 NS |
1438 | } |
1439 | } | |
1440 | ||
2bd0ea18 NS |
1441 | /* |
1442 | * Remove an entry from an intermediate node. | |
1443 | */ | |
1444 | STATIC void | |
88b32f06 DC |
1445 | xfs_da3_node_remove( |
1446 | struct xfs_da_state *state, | |
1447 | struct xfs_da_state_blk *drop_blk) | |
2bd0ea18 | 1448 | { |
88b32f06 DC |
1449 | struct xfs_da_intnode *node; |
1450 | struct xfs_da3_icnode_hdr nodehdr; | |
1451 | struct xfs_da_node_entry *btree; | |
1452 | int index; | |
1453 | int tmp; | |
ff105f75 | 1454 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1455 | |
a2ceac1f DC |
1456 | trace_xfs_da_node_remove(state->args); |
1457 | ||
1458 | node = drop_blk->bp->b_addr; | |
08c16786 | 1459 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node); |
88b32f06 | 1460 | ASSERT(drop_blk->index < nodehdr.count); |
2bd0ea18 NS |
1461 | ASSERT(drop_blk->index >= 0); |
1462 | ||
1463 | /* | |
1464 | * Copy over the offending entry, or just zero it out. | |
1465 | */ | |
88b32f06 | 1466 | index = drop_blk->index; |
8faa51a8 | 1467 | btree = nodehdr.btree; |
88b32f06 DC |
1468 | if (index < nodehdr.count - 1) { |
1469 | tmp = nodehdr.count - index - 1; | |
2bd0ea18 | 1470 | tmp *= (uint)sizeof(xfs_da_node_entry_t); |
88b32f06 | 1471 | memmove(&btree[index], &btree[index + 1], tmp); |
a2ceac1f | 1472 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
88b32f06 DC |
1473 | XFS_DA_LOGRANGE(node, &btree[index], tmp)); |
1474 | index = nodehdr.count - 1; | |
2bd0ea18 | 1475 | } |
88b32f06 | 1476 | memset(&btree[index], 0, sizeof(xfs_da_node_entry_t)); |
a2ceac1f | 1477 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
88b32f06 DC |
1478 | XFS_DA_LOGRANGE(node, &btree[index], sizeof(btree[index]))); |
1479 | nodehdr.count -= 1; | |
b81278fa | 1480 | xfs_da3_node_hdr_to_disk(dp->i_mount, node, &nodehdr); |
a2ceac1f | 1481 | xfs_trans_log_buf(state->args->trans, drop_blk->bp, |
52be9b6a | 1482 | XFS_DA_LOGRANGE(node, &node->hdr, state->args->geo->node_hdr_size)); |
2bd0ea18 NS |
1483 | |
1484 | /* | |
1485 | * Copy the last hash value from the block to propagate upwards. | |
1486 | */ | |
88b32f06 | 1487 | drop_blk->hashval = be32_to_cpu(btree[index - 1].hashval); |
2bd0ea18 NS |
1488 | } |
1489 | ||
1490 | /* | |
88b32f06 | 1491 | * Unbalance the elements between two intermediate nodes, |
2bd0ea18 NS |
1492 | * move all Btree elements from one node into another. |
1493 | */ | |
1494 | STATIC void | |
88b32f06 DC |
1495 | xfs_da3_node_unbalance( |
1496 | struct xfs_da_state *state, | |
1497 | struct xfs_da_state_blk *drop_blk, | |
1498 | struct xfs_da_state_blk *save_blk) | |
2bd0ea18 | 1499 | { |
88b32f06 DC |
1500 | struct xfs_da_intnode *drop_node; |
1501 | struct xfs_da_intnode *save_node; | |
7b02556d DC |
1502 | struct xfs_da_node_entry *drop_btree; |
1503 | struct xfs_da_node_entry *save_btree; | |
1504 | struct xfs_da3_icnode_hdr drop_hdr; | |
1505 | struct xfs_da3_icnode_hdr save_hdr; | |
88b32f06 DC |
1506 | struct xfs_trans *tp; |
1507 | int sindex; | |
1508 | int tmp; | |
ff105f75 | 1509 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1510 | |
a2ceac1f DC |
1511 | trace_xfs_da_node_unbalance(state->args); |
1512 | ||
1513 | drop_node = drop_blk->bp->b_addr; | |
1514 | save_node = save_blk->bp->b_addr; | |
08c16786 CH |
1515 | xfs_da3_node_hdr_from_disk(dp->i_mount, &drop_hdr, drop_node); |
1516 | xfs_da3_node_hdr_from_disk(dp->i_mount, &save_hdr, save_node); | |
8faa51a8 CH |
1517 | drop_btree = drop_hdr.btree; |
1518 | save_btree = save_hdr.btree; | |
2bd0ea18 NS |
1519 | tp = state->args->trans; |
1520 | ||
1521 | /* | |
1522 | * If the dying block has lower hashvals, then move all the | |
1523 | * elements in the remaining block up to make a hole. | |
1524 | */ | |
7b02556d DC |
1525 | if ((be32_to_cpu(drop_btree[0].hashval) < |
1526 | be32_to_cpu(save_btree[0].hashval)) || | |
1527 | (be32_to_cpu(drop_btree[drop_hdr.count - 1].hashval) < | |
1528 | be32_to_cpu(save_btree[save_hdr.count - 1].hashval))) { | |
88b32f06 | 1529 | /* XXX: check this - is memmove dst correct? */ |
7b02556d DC |
1530 | tmp = save_hdr.count * sizeof(xfs_da_node_entry_t); |
1531 | memmove(&save_btree[drop_hdr.count], &save_btree[0], tmp); | |
88b32f06 DC |
1532 | |
1533 | sindex = 0; | |
a2ceac1f | 1534 | xfs_trans_log_buf(tp, save_blk->bp, |
7b02556d DC |
1535 | XFS_DA_LOGRANGE(save_node, &save_btree[0], |
1536 | (save_hdr.count + drop_hdr.count) * | |
88b32f06 | 1537 | sizeof(xfs_da_node_entry_t))); |
2bd0ea18 | 1538 | } else { |
7b02556d | 1539 | sindex = save_hdr.count; |
a2ceac1f | 1540 | xfs_trans_log_buf(tp, save_blk->bp, |
7b02556d DC |
1541 | XFS_DA_LOGRANGE(save_node, &save_btree[sindex], |
1542 | drop_hdr.count * sizeof(xfs_da_node_entry_t))); | |
2bd0ea18 NS |
1543 | } |
1544 | ||
1545 | /* | |
1546 | * Move all the B-tree elements from drop_blk to save_blk. | |
1547 | */ | |
7b02556d DC |
1548 | tmp = drop_hdr.count * (uint)sizeof(xfs_da_node_entry_t); |
1549 | memcpy(&save_btree[sindex], &drop_btree[0], tmp); | |
1550 | save_hdr.count += drop_hdr.count; | |
2bd0ea18 | 1551 | |
b81278fa | 1552 | xfs_da3_node_hdr_to_disk(dp->i_mount, save_node, &save_hdr); |
a2ceac1f | 1553 | xfs_trans_log_buf(tp, save_blk->bp, |
2bd0ea18 | 1554 | XFS_DA_LOGRANGE(save_node, &save_node->hdr, |
52be9b6a | 1555 | state->args->geo->node_hdr_size)); |
2bd0ea18 NS |
1556 | |
1557 | /* | |
1558 | * Save the last hashval in the remaining block for upward propagation. | |
1559 | */ | |
7b02556d | 1560 | save_blk->hashval = be32_to_cpu(save_btree[save_hdr.count - 1].hashval); |
2bd0ea18 NS |
1561 | } |
1562 | ||
2bd0ea18 NS |
1563 | /*======================================================================== |
1564 | * Routines used for finding things in the Btree. | |
1565 | *========================================================================*/ | |
1566 | ||
1567 | /* | |
1568 | * Walk down the Btree looking for a particular filename, filling | |
1569 | * in the state structure as we go. | |
1570 | * | |
1571 | * We will set the state structure to point to each of the elements | |
1572 | * in each of the nodes where either the hashval is or should be. | |
1573 | * | |
1574 | * We support duplicate hashval's so for each entry in the current | |
1575 | * node that could contain the desired hashval, descend. This is a | |
1576 | * pruned depth-first tree search. | |
1577 | */ | |
1578 | int /* error */ | |
88b32f06 DC |
1579 | xfs_da3_node_lookup_int( |
1580 | struct xfs_da_state *state, | |
1581 | int *result) | |
2bd0ea18 | 1582 | { |
88b32f06 DC |
1583 | struct xfs_da_state_blk *blk; |
1584 | struct xfs_da_blkinfo *curr; | |
1585 | struct xfs_da_intnode *node; | |
1586 | struct xfs_da_node_entry *btree; | |
1587 | struct xfs_da3_icnode_hdr nodehdr; | |
1588 | struct xfs_da_args *args; | |
1589 | xfs_dablk_t blkno; | |
1590 | xfs_dahash_t hashval; | |
1591 | xfs_dahash_t btreehashval; | |
1592 | int probe; | |
1593 | int span; | |
1594 | int max; | |
1595 | int error; | |
1596 | int retval; | |
9473ee32 | 1597 | unsigned int expected_level = 0; |
ea09a747 | 1598 | uint16_t magic; |
ff105f75 | 1599 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 NS |
1600 | |
1601 | args = state->args; | |
32a82561 | 1602 | |
2bd0ea18 NS |
1603 | /* |
1604 | * Descend thru the B-tree searching each level for the right | |
1605 | * node to use, until the right hashval is found. | |
1606 | */ | |
9473ee32 | 1607 | blkno = args->geo->leafblk; |
2bd0ea18 NS |
1608 | for (blk = &state->path.blk[0], state->path.active = 1; |
1609 | state->path.active <= XFS_DA_NODE_MAXDEPTH; | |
1610 | blk++, state->path.active++) { | |
1611 | /* | |
1612 | * Read the next node down in the tree. | |
1613 | */ | |
1614 | blk->blkno = blkno; | |
88b32f06 | 1615 | error = xfs_da3_node_read(args->trans, args->dp, blkno, |
02cc4995 | 1616 | &blk->bp, args->whichfork); |
2bd0ea18 NS |
1617 | if (error) { |
1618 | blk->blkno = 0; | |
1619 | state->path.active--; | |
af43ca9f | 1620 | return error; |
2bd0ea18 | 1621 | } |
a2ceac1f | 1622 | curr = blk->bp->b_addr; |
ea09a747 | 1623 | magic = be16_to_cpu(curr->magic); |
88b32f06 | 1624 | |
ea09a747 DW |
1625 | if (magic == XFS_ATTR_LEAF_MAGIC || |
1626 | magic == XFS_ATTR3_LEAF_MAGIC) { | |
a24374f4 | 1627 | blk->magic = XFS_ATTR_LEAF_MAGIC; |
88b32f06 DC |
1628 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
1629 | break; | |
1630 | } | |
1631 | ||
ea09a747 DW |
1632 | if (magic == XFS_DIR2_LEAFN_MAGIC || |
1633 | magic == XFS_DIR3_LEAFN_MAGIC) { | |
88b32f06 | 1634 | blk->magic = XFS_DIR2_LEAFN_MAGIC; |
94ffb80a DW |
1635 | blk->hashval = xfs_dir2_leaf_lasthash(args->dp, |
1636 | blk->bp, NULL); | |
88b32f06 DC |
1637 | break; |
1638 | } | |
1639 | ||
a0264b73 | 1640 | if (magic != XFS_DA_NODE_MAGIC && magic != XFS_DA3_NODE_MAGIC) { |
1be76d11 | 1641 | xfs_buf_mark_corrupt(blk->bp); |
b403e029 | 1642 | return -EFSCORRUPTED; |
a0264b73 | 1643 | } |
88b32f06 | 1644 | |
b403e029 | 1645 | blk->magic = XFS_DA_NODE_MAGIC; |
2bd0ea18 NS |
1646 | |
1647 | /* | |
1648 | * Search an intermediate node for a match. | |
1649 | */ | |
88b32f06 | 1650 | node = blk->bp->b_addr; |
08c16786 | 1651 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, node); |
8faa51a8 | 1652 | btree = nodehdr.btree; |
2bd0ea18 | 1653 | |
9473ee32 | 1654 | /* Tree taller than we can handle; bail out! */ |
a0264b73 | 1655 | if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) { |
1be76d11 | 1656 | xfs_buf_mark_corrupt(blk->bp); |
9473ee32 | 1657 | return -EFSCORRUPTED; |
a0264b73 | 1658 | } |
9473ee32 DW |
1659 | |
1660 | /* Check the level from the root. */ | |
1661 | if (blkno == args->geo->leafblk) | |
1662 | expected_level = nodehdr.level - 1; | |
a0264b73 | 1663 | else if (expected_level != nodehdr.level) { |
1be76d11 | 1664 | xfs_buf_mark_corrupt(blk->bp); |
9473ee32 | 1665 | return -EFSCORRUPTED; |
a0264b73 | 1666 | } else |
9473ee32 DW |
1667 | expected_level--; |
1668 | ||
88b32f06 DC |
1669 | max = nodehdr.count; |
1670 | blk->hashval = be32_to_cpu(btree[max - 1].hashval); | |
2bd0ea18 | 1671 | |
88b32f06 DC |
1672 | /* |
1673 | * Binary search. (note: small blocks will skip loop) | |
1674 | */ | |
1675 | probe = span = max / 2; | |
1676 | hashval = args->hashval; | |
1677 | while (span > 4) { | |
1678 | span /= 2; | |
1679 | btreehashval = be32_to_cpu(btree[probe].hashval); | |
1680 | if (btreehashval < hashval) | |
1681 | probe += span; | |
1682 | else if (btreehashval > hashval) | |
1683 | probe -= span; | |
1684 | else | |
1685 | break; | |
1686 | } | |
1687 | ASSERT((probe >= 0) && (probe < max)); | |
1688 | ASSERT((span <= 4) || | |
1689 | (be32_to_cpu(btree[probe].hashval) == hashval)); | |
2bd0ea18 | 1690 | |
88b32f06 DC |
1691 | /* |
1692 | * Since we may have duplicate hashval's, find the first | |
1693 | * matching hashval in the node. | |
1694 | */ | |
1695 | while (probe > 0 && | |
1696 | be32_to_cpu(btree[probe].hashval) >= hashval) { | |
1697 | probe--; | |
1698 | } | |
1699 | while (probe < max && | |
1700 | be32_to_cpu(btree[probe].hashval) < hashval) { | |
1701 | probe++; | |
1702 | } | |
1703 | ||
1704 | /* | |
1705 | * Pick the right block to descend on. | |
1706 | */ | |
1707 | if (probe == max) { | |
1708 | blk->index = max - 1; | |
1709 | blkno = be32_to_cpu(btree[max - 1].before); | |
1710 | } else { | |
1711 | blk->index = probe; | |
1712 | blkno = be32_to_cpu(btree[probe].before); | |
2bd0ea18 | 1713 | } |
9473ee32 DW |
1714 | |
1715 | /* We can't point back to the root. */ | |
bc73da84 | 1716 | if (XFS_IS_CORRUPT(dp->i_mount, blkno == args->geo->leafblk)) |
9473ee32 | 1717 | return -EFSCORRUPTED; |
2bd0ea18 NS |
1718 | } |
1719 | ||
bc73da84 | 1720 | if (XFS_IS_CORRUPT(dp->i_mount, expected_level != 0)) |
9473ee32 DW |
1721 | return -EFSCORRUPTED; |
1722 | ||
2bd0ea18 NS |
1723 | /* |
1724 | * A leaf block that ends in the hashval that we are interested in | |
1725 | * (final hashval == search hashval) means that the next block may | |
1726 | * contain more entries with the same hashval, shift upward to the | |
1727 | * next leaf and keep searching. | |
1728 | */ | |
1729 | for (;;) { | |
5e656dbb | 1730 | if (blk->magic == XFS_DIR2_LEAFN_MAGIC) { |
32a82561 | 1731 | retval = xfs_dir2_leafn_lookup_int(blk->bp, args, |
2bd0ea18 | 1732 | &blk->index, state); |
5e656dbb | 1733 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
a24374f4 | 1734 | retval = xfs_attr3_leaf_lookup_int(blk->bp, args); |
32a82561 NS |
1735 | blk->index = args->index; |
1736 | args->blkno = blk->blkno; | |
5e656dbb BN |
1737 | } else { |
1738 | ASSERT(0); | |
12b53197 | 1739 | return -EFSCORRUPTED; |
2bd0ea18 | 1740 | } |
12b53197 | 1741 | if (((retval == -ENOENT) || (retval == -ENOATTR)) && |
32a82561 | 1742 | (blk->hashval == args->hashval)) { |
88b32f06 | 1743 | error = xfs_da3_path_shift(state, &state->path, 1, 1, |
2bd0ea18 NS |
1744 | &retval); |
1745 | if (error) | |
af43ca9f | 1746 | return error; |
2bd0ea18 NS |
1747 | if (retval == 0) { |
1748 | continue; | |
5e656dbb | 1749 | } else if (blk->magic == XFS_ATTR_LEAF_MAGIC) { |
2bd0ea18 | 1750 | /* path_shift() gives ENOENT */ |
12b53197 | 1751 | retval = -ENOATTR; |
2bd0ea18 | 1752 | } |
2bd0ea18 NS |
1753 | } |
1754 | break; | |
1755 | } | |
1756 | *result = retval; | |
af43ca9f | 1757 | return 0; |
2bd0ea18 NS |
1758 | } |
1759 | ||
2bd0ea18 NS |
1760 | /*======================================================================== |
1761 | * Utility routines. | |
1762 | *========================================================================*/ | |
1763 | ||
88b32f06 DC |
1764 | /* |
1765 | * Compare two intermediate nodes for "order". | |
1766 | */ | |
1767 | STATIC int | |
1768 | xfs_da3_node_order( | |
ff105f75 | 1769 | struct xfs_inode *dp, |
88b32f06 DC |
1770 | struct xfs_buf *node1_bp, |
1771 | struct xfs_buf *node2_bp) | |
1772 | { | |
1773 | struct xfs_da_intnode *node1; | |
1774 | struct xfs_da_intnode *node2; | |
1775 | struct xfs_da_node_entry *btree1; | |
1776 | struct xfs_da_node_entry *btree2; | |
1777 | struct xfs_da3_icnode_hdr node1hdr; | |
1778 | struct xfs_da3_icnode_hdr node2hdr; | |
1779 | ||
1780 | node1 = node1_bp->b_addr; | |
1781 | node2 = node2_bp->b_addr; | |
08c16786 CH |
1782 | xfs_da3_node_hdr_from_disk(dp->i_mount, &node1hdr, node1); |
1783 | xfs_da3_node_hdr_from_disk(dp->i_mount, &node2hdr, node2); | |
8faa51a8 CH |
1784 | btree1 = node1hdr.btree; |
1785 | btree2 = node2hdr.btree; | |
88b32f06 DC |
1786 | |
1787 | if (node1hdr.count > 0 && node2hdr.count > 0 && | |
1788 | ((be32_to_cpu(btree2[0].hashval) < be32_to_cpu(btree1[0].hashval)) || | |
1789 | (be32_to_cpu(btree2[node2hdr.count - 1].hashval) < | |
1790 | be32_to_cpu(btree1[node1hdr.count - 1].hashval)))) { | |
1791 | return 1; | |
1792 | } | |
1793 | return 0; | |
1794 | } | |
1795 | ||
2bd0ea18 NS |
1796 | /* |
1797 | * Link a new block into a doubly linked list of blocks (of whatever type). | |
1798 | */ | |
1799 | int /* error */ | |
88b32f06 DC |
1800 | xfs_da3_blk_link( |
1801 | struct xfs_da_state *state, | |
1802 | struct xfs_da_state_blk *old_blk, | |
1803 | struct xfs_da_state_blk *new_blk) | |
2bd0ea18 | 1804 | { |
88b32f06 DC |
1805 | struct xfs_da_blkinfo *old_info; |
1806 | struct xfs_da_blkinfo *new_info; | |
1807 | struct xfs_da_blkinfo *tmp_info; | |
1808 | struct xfs_da_args *args; | |
1809 | struct xfs_buf *bp; | |
1810 | int before = 0; | |
1811 | int error; | |
ff105f75 | 1812 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 NS |
1813 | |
1814 | /* | |
1815 | * Set up environment. | |
1816 | */ | |
1817 | args = state->args; | |
1818 | ASSERT(args != NULL); | |
a2ceac1f DC |
1819 | old_info = old_blk->bp->b_addr; |
1820 | new_info = new_blk->bp->b_addr; | |
2bd0ea18 | 1821 | ASSERT(old_blk->magic == XFS_DA_NODE_MAGIC || |
5e656dbb | 1822 | old_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
2bd0ea18 | 1823 | old_blk->magic == XFS_ATTR_LEAF_MAGIC); |
2bd0ea18 NS |
1824 | |
1825 | switch (old_blk->magic) { | |
2bd0ea18 NS |
1826 | case XFS_ATTR_LEAF_MAGIC: |
1827 | before = xfs_attr_leaf_order(old_blk->bp, new_blk->bp); | |
1828 | break; | |
2bd0ea18 | 1829 | case XFS_DIR2_LEAFN_MAGIC: |
ff105f75 | 1830 | before = xfs_dir2_leafn_order(dp, old_blk->bp, new_blk->bp); |
2bd0ea18 NS |
1831 | break; |
1832 | case XFS_DA_NODE_MAGIC: | |
ff105f75 | 1833 | before = xfs_da3_node_order(dp, old_blk->bp, new_blk->bp); |
2bd0ea18 NS |
1834 | break; |
1835 | } | |
1836 | ||
1837 | /* | |
1838 | * Link blocks in appropriate order. | |
1839 | */ | |
1840 | if (before) { | |
1841 | /* | |
1842 | * Link new block in before existing block. | |
1843 | */ | |
a2ceac1f | 1844 | trace_xfs_da_link_before(args); |
5e656dbb BN |
1845 | new_info->forw = cpu_to_be32(old_blk->blkno); |
1846 | new_info->back = old_info->back; | |
1847 | if (old_info->back) { | |
ff105f75 | 1848 | error = xfs_da3_node_read(args->trans, dp, |
5e656dbb | 1849 | be32_to_cpu(old_info->back), |
02cc4995 | 1850 | &bp, args->whichfork); |
2bd0ea18 | 1851 | if (error) |
af43ca9f | 1852 | return error; |
2bd0ea18 | 1853 | ASSERT(bp != NULL); |
a2ceac1f | 1854 | tmp_info = bp->b_addr; |
88b32f06 | 1855 | ASSERT(tmp_info->magic == old_info->magic); |
5e656dbb BN |
1856 | ASSERT(be32_to_cpu(tmp_info->forw) == old_blk->blkno); |
1857 | tmp_info->forw = cpu_to_be32(new_blk->blkno); | |
a2ceac1f | 1858 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
2bd0ea18 | 1859 | } |
5e656dbb | 1860 | old_info->back = cpu_to_be32(new_blk->blkno); |
2bd0ea18 NS |
1861 | } else { |
1862 | /* | |
1863 | * Link new block in after existing block. | |
1864 | */ | |
a2ceac1f | 1865 | trace_xfs_da_link_after(args); |
5e656dbb BN |
1866 | new_info->forw = old_info->forw; |
1867 | new_info->back = cpu_to_be32(old_blk->blkno); | |
1868 | if (old_info->forw) { | |
ff105f75 | 1869 | error = xfs_da3_node_read(args->trans, dp, |
5e656dbb | 1870 | be32_to_cpu(old_info->forw), |
02cc4995 | 1871 | &bp, args->whichfork); |
2bd0ea18 | 1872 | if (error) |
af43ca9f | 1873 | return error; |
2bd0ea18 | 1874 | ASSERT(bp != NULL); |
a2ceac1f | 1875 | tmp_info = bp->b_addr; |
5e656dbb BN |
1876 | ASSERT(tmp_info->magic == old_info->magic); |
1877 | ASSERT(be32_to_cpu(tmp_info->back) == old_blk->blkno); | |
1878 | tmp_info->back = cpu_to_be32(new_blk->blkno); | |
a2ceac1f | 1879 | xfs_trans_log_buf(args->trans, bp, 0, sizeof(*tmp_info)-1); |
2bd0ea18 | 1880 | } |
5e656dbb | 1881 | old_info->forw = cpu_to_be32(new_blk->blkno); |
2bd0ea18 NS |
1882 | } |
1883 | ||
a2ceac1f DC |
1884 | xfs_trans_log_buf(args->trans, old_blk->bp, 0, sizeof(*tmp_info) - 1); |
1885 | xfs_trans_log_buf(args->trans, new_blk->bp, 0, sizeof(*tmp_info) - 1); | |
af43ca9f | 1886 | return 0; |
2bd0ea18 NS |
1887 | } |
1888 | ||
2bd0ea18 NS |
1889 | /* |
1890 | * Unlink a block from a doubly linked list of blocks. | |
1891 | */ | |
5e656dbb | 1892 | STATIC int /* error */ |
88b32f06 DC |
1893 | xfs_da3_blk_unlink( |
1894 | struct xfs_da_state *state, | |
1895 | struct xfs_da_state_blk *drop_blk, | |
1896 | struct xfs_da_state_blk *save_blk) | |
2bd0ea18 | 1897 | { |
88b32f06 DC |
1898 | struct xfs_da_blkinfo *drop_info; |
1899 | struct xfs_da_blkinfo *save_info; | |
1900 | struct xfs_da_blkinfo *tmp_info; | |
1901 | struct xfs_da_args *args; | |
1902 | struct xfs_buf *bp; | |
1903 | int error; | |
2bd0ea18 NS |
1904 | |
1905 | /* | |
1906 | * Set up environment. | |
1907 | */ | |
1908 | args = state->args; | |
1909 | ASSERT(args != NULL); | |
a2ceac1f DC |
1910 | save_info = save_blk->bp->b_addr; |
1911 | drop_info = drop_blk->bp->b_addr; | |
2bd0ea18 | 1912 | ASSERT(save_blk->magic == XFS_DA_NODE_MAGIC || |
5e656dbb | 1913 | save_blk->magic == XFS_DIR2_LEAFN_MAGIC || |
2bd0ea18 | 1914 | save_blk->magic == XFS_ATTR_LEAF_MAGIC); |
2bd0ea18 | 1915 | ASSERT(save_blk->magic == drop_blk->magic); |
5e656dbb BN |
1916 | ASSERT((be32_to_cpu(save_info->forw) == drop_blk->blkno) || |
1917 | (be32_to_cpu(save_info->back) == drop_blk->blkno)); | |
1918 | ASSERT((be32_to_cpu(drop_info->forw) == save_blk->blkno) || | |
1919 | (be32_to_cpu(drop_info->back) == save_blk->blkno)); | |
2bd0ea18 NS |
1920 | |
1921 | /* | |
1922 | * Unlink the leaf block from the doubly linked chain of leaves. | |
1923 | */ | |
5e656dbb | 1924 | if (be32_to_cpu(save_info->back) == drop_blk->blkno) { |
a2ceac1f | 1925 | trace_xfs_da_unlink_back(args); |
5e656dbb BN |
1926 | save_info->back = drop_info->back; |
1927 | if (drop_info->back) { | |
88b32f06 | 1928 | error = xfs_da3_node_read(args->trans, args->dp, |
5e656dbb | 1929 | be32_to_cpu(drop_info->back), |
02cc4995 | 1930 | &bp, args->whichfork); |
2bd0ea18 | 1931 | if (error) |
af43ca9f | 1932 | return error; |
2bd0ea18 | 1933 | ASSERT(bp != NULL); |
a2ceac1f | 1934 | tmp_info = bp->b_addr; |
5e656dbb BN |
1935 | ASSERT(tmp_info->magic == save_info->magic); |
1936 | ASSERT(be32_to_cpu(tmp_info->forw) == drop_blk->blkno); | |
1937 | tmp_info->forw = cpu_to_be32(save_blk->blkno); | |
a2ceac1f | 1938 | xfs_trans_log_buf(args->trans, bp, 0, |
2bd0ea18 | 1939 | sizeof(*tmp_info) - 1); |
2bd0ea18 NS |
1940 | } |
1941 | } else { | |
a2ceac1f | 1942 | trace_xfs_da_unlink_forward(args); |
5e656dbb BN |
1943 | save_info->forw = drop_info->forw; |
1944 | if (drop_info->forw) { | |
88b32f06 | 1945 | error = xfs_da3_node_read(args->trans, args->dp, |
5e656dbb | 1946 | be32_to_cpu(drop_info->forw), |
02cc4995 | 1947 | &bp, args->whichfork); |
2bd0ea18 | 1948 | if (error) |
af43ca9f | 1949 | return error; |
2bd0ea18 | 1950 | ASSERT(bp != NULL); |
a2ceac1f | 1951 | tmp_info = bp->b_addr; |
5e656dbb BN |
1952 | ASSERT(tmp_info->magic == save_info->magic); |
1953 | ASSERT(be32_to_cpu(tmp_info->back) == drop_blk->blkno); | |
1954 | tmp_info->back = cpu_to_be32(save_blk->blkno); | |
a2ceac1f | 1955 | xfs_trans_log_buf(args->trans, bp, 0, |
2bd0ea18 | 1956 | sizeof(*tmp_info) - 1); |
2bd0ea18 NS |
1957 | } |
1958 | } | |
1959 | ||
a2ceac1f | 1960 | xfs_trans_log_buf(args->trans, save_blk->bp, 0, sizeof(*save_info) - 1); |
af43ca9f | 1961 | return 0; |
2bd0ea18 NS |
1962 | } |
1963 | ||
1964 | /* | |
1965 | * Move a path "forward" or "!forward" one block at the current level. | |
1966 | * | |
1967 | * This routine will adjust a "path" to point to the next block | |
1968 | * "forward" (higher hashvalues) or "!forward" (lower hashvals) in the | |
1969 | * Btree, including updating pointers to the intermediate nodes between | |
1970 | * the new bottom and the root. | |
1971 | */ | |
1972 | int /* error */ | |
88b32f06 DC |
1973 | xfs_da3_path_shift( |
1974 | struct xfs_da_state *state, | |
1975 | struct xfs_da_state_path *path, | |
1976 | int forward, | |
1977 | int release, | |
1978 | int *result) | |
2bd0ea18 | 1979 | { |
88b32f06 DC |
1980 | struct xfs_da_state_blk *blk; |
1981 | struct xfs_da_blkinfo *info; | |
88b32f06 DC |
1982 | struct xfs_da_args *args; |
1983 | struct xfs_da_node_entry *btree; | |
1984 | struct xfs_da3_icnode_hdr nodehdr; | |
c063b960 | 1985 | struct xfs_buf *bp; |
88b32f06 DC |
1986 | xfs_dablk_t blkno = 0; |
1987 | int level; | |
1988 | int error; | |
ff105f75 | 1989 | struct xfs_inode *dp = state->args->dp; |
2bd0ea18 | 1990 | |
a2ceac1f DC |
1991 | trace_xfs_da_path_shift(state->args); |
1992 | ||
2bd0ea18 NS |
1993 | /* |
1994 | * Roll up the Btree looking for the first block where our | |
1995 | * current index is not at the edge of the block. Note that | |
1996 | * we skip the bottom layer because we want the sibling block. | |
1997 | */ | |
1998 | args = state->args; | |
1999 | ASSERT(args != NULL); | |
2000 | ASSERT(path != NULL); | |
2001 | ASSERT((path->active > 0) && (path->active < XFS_DA_NODE_MAXDEPTH)); | |
2002 | level = (path->active-1) - 1; /* skip bottom layer in path */ | |
dfab3d49 QC |
2003 | for (; level >= 0; level--) { |
2004 | blk = &path->blk[level]; | |
8faa51a8 CH |
2005 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, |
2006 | blk->bp->b_addr); | |
88b32f06 DC |
2007 | |
2008 | if (forward && (blk->index < nodehdr.count - 1)) { | |
2bd0ea18 | 2009 | blk->index++; |
8faa51a8 | 2010 | blkno = be32_to_cpu(nodehdr.btree[blk->index].before); |
2bd0ea18 NS |
2011 | break; |
2012 | } else if (!forward && (blk->index > 0)) { | |
2013 | blk->index--; | |
8faa51a8 | 2014 | blkno = be32_to_cpu(nodehdr.btree[blk->index].before); |
2bd0ea18 NS |
2015 | break; |
2016 | } | |
2017 | } | |
2018 | if (level < 0) { | |
12b53197 | 2019 | *result = -ENOENT; /* we're out of our tree */ |
5e656dbb | 2020 | ASSERT(args->op_flags & XFS_DA_OP_OKNOENT); |
af43ca9f | 2021 | return 0; |
2bd0ea18 NS |
2022 | } |
2023 | ||
2024 | /* | |
2025 | * Roll down the edge of the subtree until we reach the | |
2026 | * same depth we were at originally. | |
2027 | */ | |
2028 | for (blk++, level++; level < path->active; blk++, level++) { | |
2029 | /* | |
c063b960 | 2030 | * Read the next child block into a local buffer. |
2bd0ea18 | 2031 | */ |
02cc4995 | 2032 | error = xfs_da3_node_read(args->trans, dp, blkno, &bp, |
c063b960 BF |
2033 | args->whichfork); |
2034 | if (error) | |
2035 | return error; | |
2bd0ea18 NS |
2036 | |
2037 | /* | |
c063b960 BF |
2038 | * Release the old block (if it's dirty, the trans doesn't |
2039 | * actually let go) and swap the local buffer into the path | |
2040 | * structure. This ensures failure of the above read doesn't set | |
2041 | * a NULL buffer in an active slot in the path. | |
2bd0ea18 | 2042 | */ |
c063b960 BF |
2043 | if (release) |
2044 | xfs_trans_brelse(args->trans, blk->bp); | |
2bd0ea18 | 2045 | blk->blkno = blkno; |
c063b960 BF |
2046 | blk->bp = bp; |
2047 | ||
a2ceac1f DC |
2048 | info = blk->bp->b_addr; |
2049 | ASSERT(info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC) || | |
88b32f06 | 2050 | info->magic == cpu_to_be16(XFS_DA3_NODE_MAGIC) || |
a2ceac1f | 2051 | info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
65b80c98 | 2052 | info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) || |
a24374f4 DC |
2053 | info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) || |
2054 | info->magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)); | |
88b32f06 DC |
2055 | |
2056 | ||
2057 | /* | |
2058 | * Note: we flatten the magic number to a single type so we | |
2059 | * don't have to compare against crc/non-crc types elsewhere. | |
2060 | */ | |
2061 | switch (be16_to_cpu(info->magic)) { | |
2062 | case XFS_DA_NODE_MAGIC: | |
2063 | case XFS_DA3_NODE_MAGIC: | |
2064 | blk->magic = XFS_DA_NODE_MAGIC; | |
8faa51a8 CH |
2065 | xfs_da3_node_hdr_from_disk(dp->i_mount, &nodehdr, |
2066 | bp->b_addr); | |
2067 | btree = nodehdr.btree; | |
88b32f06 | 2068 | blk->hashval = be32_to_cpu(btree[nodehdr.count - 1].hashval); |
2bd0ea18 NS |
2069 | if (forward) |
2070 | blk->index = 0; | |
2071 | else | |
88b32f06 DC |
2072 | blk->index = nodehdr.count - 1; |
2073 | blkno = be32_to_cpu(btree[blk->index].before); | |
2074 | break; | |
2075 | case XFS_ATTR_LEAF_MAGIC: | |
a24374f4 | 2076 | case XFS_ATTR3_LEAF_MAGIC: |
88b32f06 | 2077 | blk->magic = XFS_ATTR_LEAF_MAGIC; |
2bd0ea18 NS |
2078 | ASSERT(level == path->active-1); |
2079 | blk->index = 0; | |
ff105f75 | 2080 | blk->hashval = xfs_attr_leaf_lasthash(blk->bp, NULL); |
88b32f06 DC |
2081 | break; |
2082 | case XFS_DIR2_LEAFN_MAGIC: | |
2083 | case XFS_DIR3_LEAFN_MAGIC: | |
2084 | blk->magic = XFS_DIR2_LEAFN_MAGIC; | |
2085 | ASSERT(level == path->active-1); | |
2086 | blk->index = 0; | |
94ffb80a DW |
2087 | blk->hashval = xfs_dir2_leaf_lasthash(args->dp, |
2088 | blk->bp, NULL); | |
88b32f06 DC |
2089 | break; |
2090 | default: | |
2091 | ASSERT(0); | |
2092 | break; | |
2bd0ea18 NS |
2093 | } |
2094 | } | |
2095 | *result = 0; | |
88b32f06 | 2096 | return 0; |
2bd0ea18 NS |
2097 | } |
2098 | ||
2099 | ||
2100 | /*======================================================================== | |
2101 | * Utility routines. | |
2102 | *========================================================================*/ | |
2103 | ||
2104 | /* | |
2105 | * Implement a simple hash on a character string. | |
2106 | * Rotate the hash value by 7 bits, then XOR each character in. | |
2107 | * This is implemented with some source-level loop unrolling. | |
2108 | */ | |
2109 | xfs_dahash_t | |
4a492e72 | 2110 | xfs_da_hashname(const uint8_t *name, int namelen) |
2bd0ea18 NS |
2111 | { |
2112 | xfs_dahash_t hash; | |
2113 | ||
2bd0ea18 NS |
2114 | /* |
2115 | * Do four characters at a time as long as we can. | |
2116 | */ | |
f9c48a87 | 2117 | for (hash = 0; namelen >= 4; namelen -= 4, name += 4) |
2bd0ea18 | 2118 | hash = (name[0] << 21) ^ (name[1] << 14) ^ (name[2] << 7) ^ |
f9c48a87 NS |
2119 | (name[3] << 0) ^ rol32(hash, 7 * 4); |
2120 | ||
2bd0ea18 NS |
2121 | /* |
2122 | * Now do the rest of the characters. | |
2123 | */ | |
2124 | switch (namelen) { | |
2125 | case 3: | |
2126 | return (name[0] << 14) ^ (name[1] << 7) ^ (name[2] << 0) ^ | |
f9c48a87 | 2127 | rol32(hash, 7 * 3); |
2bd0ea18 | 2128 | case 2: |
f9c48a87 | 2129 | return (name[0] << 7) ^ (name[1] << 0) ^ rol32(hash, 7 * 2); |
2bd0ea18 | 2130 | case 1: |
f9c48a87 NS |
2131 | return (name[0] << 0) ^ rol32(hash, 7 * 1); |
2132 | default: /* case 0: */ | |
2bd0ea18 NS |
2133 | return hash; |
2134 | } | |
2bd0ea18 NS |
2135 | } |
2136 | ||
51ca7008 | 2137 | enum xfs_dacmp |
5e656dbb BN |
2138 | xfs_da_compname( |
2139 | struct xfs_da_args *args, | |
56b2de80 DC |
2140 | const unsigned char *name, |
2141 | int len) | |
51ca7008 | 2142 | { |
5e656dbb | 2143 | return (args->namelen == len && memcmp(args->name, name, len) == 0) ? |
51ca7008 BN |
2144 | XFS_CMP_EXACT : XFS_CMP_DIFFERENT; |
2145 | } | |
2146 | ||
2bd0ea18 | 2147 | int |
a2ceac1f DC |
2148 | xfs_da_grow_inode_int( |
2149 | struct xfs_da_args *args, | |
2150 | xfs_fileoff_t *bno, | |
2151 | int count) | |
2bd0ea18 | 2152 | { |
a2ceac1f DC |
2153 | struct xfs_trans *tp = args->trans; |
2154 | struct xfs_inode *dp = args->dp; | |
2155 | int w = args->whichfork; | |
aa00f286 | 2156 | xfs_rfsblock_t nblks = dp->i_nblocks; |
a2ceac1f DC |
2157 | struct xfs_bmbt_irec map, *mapp; |
2158 | int nmap, error, got, i, mapi; | |
56b2de80 | 2159 | |
2bd0ea18 NS |
2160 | /* |
2161 | * Find a spot in the file space to put the new block. | |
2162 | */ | |
a2ceac1f DC |
2163 | error = xfs_bmap_first_unused(tp, dp, count, bno, w); |
2164 | if (error) | |
2bd0ea18 | 2165 | return error; |
a2ceac1f | 2166 | |
2bd0ea18 NS |
2167 | /* |
2168 | * Try mapping it in one filesystem block. | |
2169 | */ | |
2170 | nmap = 1; | |
a2ceac1f DC |
2171 | error = xfs_bmapi_write(tp, dp, *bno, count, |
2172 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG, | |
b39f9fcb | 2173 | args->total, &map, &nmap); |
a2ceac1f | 2174 | if (error) |
2bd0ea18 | 2175 | return error; |
a2ceac1f | 2176 | |
2bd0ea18 NS |
2177 | ASSERT(nmap <= 1); |
2178 | if (nmap == 1) { | |
2179 | mapp = ↦ | |
2180 | mapi = 1; | |
a2ceac1f DC |
2181 | } else if (nmap == 0 && count > 1) { |
2182 | xfs_fileoff_t b; | |
2183 | int c; | |
2184 | ||
2185 | /* | |
2186 | * If we didn't get it and the block might work if fragmented, | |
2187 | * try without the CONTIG flag. Loop until we get it all. | |
2188 | */ | |
6cd1e6db | 2189 | mapp = kmem_alloc(sizeof(*mapp) * count, 0); |
a2ceac1f | 2190 | for (b = *bno, mapi = 0; b < *bno + count; ) { |
a2ceac1f | 2191 | c = (int)(*bno + count - b); |
04d4c27a | 2192 | nmap = min(XFS_BMAP_MAX_NMAP, c); |
a2ceac1f DC |
2193 | error = xfs_bmapi_write(tp, dp, b, c, |
2194 | xfs_bmapi_aflag(w)|XFS_BMAPI_METADATA, | |
b39f9fcb | 2195 | args->total, &mapp[mapi], &nmap); |
a2ceac1f DC |
2196 | if (error) |
2197 | goto out_free_map; | |
2bd0ea18 NS |
2198 | if (nmap < 1) |
2199 | break; | |
2200 | mapi += nmap; | |
2201 | b = mapp[mapi - 1].br_startoff + | |
2202 | mapp[mapi - 1].br_blockcount; | |
2203 | } | |
2204 | } else { | |
2bd0ea18 NS |
2205 | mapi = 0; |
2206 | mapp = NULL; | |
2207 | } | |
a2ceac1f | 2208 | |
2bd0ea18 NS |
2209 | /* |
2210 | * Count the blocks we got, make sure it matches the total. | |
2211 | */ | |
2212 | for (i = 0, got = 0; i < mapi; i++) | |
2213 | got += mapp[i].br_blockcount; | |
a2ceac1f | 2214 | if (got != count || mapp[0].br_startoff != *bno || |
2bd0ea18 | 2215 | mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount != |
a2ceac1f | 2216 | *bno + count) { |
12b53197 | 2217 | error = -ENOSPC; |
a2ceac1f | 2218 | goto out_free_map; |
2bd0ea18 | 2219 | } |
a2ceac1f | 2220 | |
56b2de80 | 2221 | /* account for newly allocated blocks in reserved blocks total */ |
aa00f286 | 2222 | args->total -= dp->i_nblocks - nblks; |
a2ceac1f DC |
2223 | |
2224 | out_free_map: | |
2225 | if (mapp != &map) | |
2226 | kmem_free(mapp); | |
2227 | return error; | |
2228 | } | |
2229 | ||
2230 | /* | |
2231 | * Add a block to the btree ahead of the file. | |
2232 | * Return the new block number to the caller. | |
2233 | */ | |
2234 | int | |
2235 | xfs_da_grow_inode( | |
2236 | struct xfs_da_args *args, | |
2237 | xfs_dablk_t *new_blkno) | |
2238 | { | |
2239 | xfs_fileoff_t bno; | |
a2ceac1f DC |
2240 | int error; |
2241 | ||
2242 | trace_xfs_da_grow_inode(args); | |
2243 | ||
ff105f75 DC |
2244 | bno = args->geo->leafblk; |
2245 | error = xfs_da_grow_inode_int(args, &bno, args->geo->fsbcount); | |
a2ceac1f DC |
2246 | if (!error) |
2247 | *new_blkno = (xfs_dablk_t)bno; | |
2248 | return error; | |
2bd0ea18 NS |
2249 | } |
2250 | ||
2bd0ea18 | 2251 | /* |
dfc130f3 | 2252 | * Ick. We need to always be able to remove a btree block, even |
2bd0ea18 NS |
2253 | * if there's no space reservation because the filesystem is full. |
2254 | * This is called if xfs_bunmapi on a btree block fails due to ENOSPC. | |
2255 | * It swaps the target block with the last block in the file. The | |
2256 | * last block in the file can always be removed since it can't cause | |
2257 | * a bmap btree split to do that. | |
2258 | */ | |
2259 | STATIC int | |
88b32f06 DC |
2260 | xfs_da3_swap_lastblock( |
2261 | struct xfs_da_args *args, | |
2262 | xfs_dablk_t *dead_blknop, | |
2263 | struct xfs_buf **dead_bufp) | |
2bd0ea18 | 2264 | { |
88b32f06 DC |
2265 | struct xfs_da_blkinfo *dead_info; |
2266 | struct xfs_da_blkinfo *sib_info; | |
2267 | struct xfs_da_intnode *par_node; | |
2268 | struct xfs_da_intnode *dead_node; | |
2269 | struct xfs_dir2_leaf *dead_leaf2; | |
2270 | struct xfs_da_node_entry *btree; | |
2271 | struct xfs_da3_icnode_hdr par_hdr; | |
ff105f75 | 2272 | struct xfs_inode *dp; |
88b32f06 DC |
2273 | struct xfs_trans *tp; |
2274 | struct xfs_mount *mp; | |
2275 | struct xfs_buf *dead_buf; | |
2276 | struct xfs_buf *last_buf; | |
2277 | struct xfs_buf *sib_buf; | |
2278 | struct xfs_buf *par_buf; | |
2279 | xfs_dahash_t dead_hash; | |
2280 | xfs_fileoff_t lastoff; | |
2281 | xfs_dablk_t dead_blkno; | |
2282 | xfs_dablk_t last_blkno; | |
2283 | xfs_dablk_t sib_blkno; | |
2284 | xfs_dablk_t par_blkno; | |
2285 | int error; | |
2286 | int w; | |
2287 | int entno; | |
2288 | int level; | |
2289 | int dead_level; | |
2bd0ea18 | 2290 | |
a2ceac1f DC |
2291 | trace_xfs_da_swap_lastblock(args); |
2292 | ||
2bd0ea18 NS |
2293 | dead_buf = *dead_bufp; |
2294 | dead_blkno = *dead_blknop; | |
2295 | tp = args->trans; | |
ff105f75 | 2296 | dp = args->dp; |
2bd0ea18 NS |
2297 | w = args->whichfork; |
2298 | ASSERT(w == XFS_DATA_FORK); | |
ff105f75 DC |
2299 | mp = dp->i_mount; |
2300 | lastoff = args->geo->freeblk; | |
2301 | error = xfs_bmap_last_before(tp, dp, &lastoff, w); | |
2bd0ea18 NS |
2302 | if (error) |
2303 | return error; | |
bc73da84 | 2304 | if (XFS_IS_CORRUPT(mp, lastoff == 0)) |
12b53197 | 2305 | return -EFSCORRUPTED; |
2bd0ea18 NS |
2306 | /* |
2307 | * Read the last block in the btree space. | |
2308 | */ | |
ff105f75 | 2309 | last_blkno = (xfs_dablk_t)lastoff - args->geo->fsbcount; |
02cc4995 | 2310 | error = xfs_da3_node_read(tp, dp, last_blkno, &last_buf, w); |
a2ceac1f | 2311 | if (error) |
2bd0ea18 NS |
2312 | return error; |
2313 | /* | |
2314 | * Copy the last block into the dead buffer and log it. | |
2315 | */ | |
ff105f75 DC |
2316 | memcpy(dead_buf->b_addr, last_buf->b_addr, args->geo->blksize); |
2317 | xfs_trans_log_buf(tp, dead_buf, 0, args->geo->blksize - 1); | |
a2ceac1f | 2318 | dead_info = dead_buf->b_addr; |
2bd0ea18 NS |
2319 | /* |
2320 | * Get values from the moved block. | |
2321 | */ | |
65b80c98 DC |
2322 | if (dead_info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) || |
2323 | dead_info->magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC)) { | |
2324 | struct xfs_dir3_icleaf_hdr leafhdr; | |
2325 | struct xfs_dir2_leaf_entry *ents; | |
2326 | ||
2bd0ea18 | 2327 | dead_leaf2 = (xfs_dir2_leaf_t *)dead_info; |
9db68faf CH |
2328 | xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, |
2329 | dead_leaf2); | |
a2279497 | 2330 | ents = leafhdr.ents; |
2bd0ea18 | 2331 | dead_level = 0; |
65b80c98 | 2332 | dead_hash = be32_to_cpu(ents[leafhdr.count - 1].hashval); |
2bd0ea18 | 2333 | } else { |
88b32f06 DC |
2334 | struct xfs_da3_icnode_hdr deadhdr; |
2335 | ||
2bd0ea18 | 2336 | dead_node = (xfs_da_intnode_t *)dead_info; |
08c16786 | 2337 | xfs_da3_node_hdr_from_disk(dp->i_mount, &deadhdr, dead_node); |
8faa51a8 | 2338 | btree = deadhdr.btree; |
88b32f06 DC |
2339 | dead_level = deadhdr.level; |
2340 | dead_hash = be32_to_cpu(btree[deadhdr.count - 1].hashval); | |
2bd0ea18 NS |
2341 | } |
2342 | sib_buf = par_buf = NULL; | |
2343 | /* | |
2344 | * If the moved block has a left sibling, fix up the pointers. | |
2345 | */ | |
5e656dbb | 2346 | if ((sib_blkno = be32_to_cpu(dead_info->back))) { |
02cc4995 | 2347 | error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w); |
a2ceac1f | 2348 | if (error) |
2bd0ea18 | 2349 | goto done; |
a2ceac1f | 2350 | sib_info = sib_buf->b_addr; |
bc73da84 DW |
2351 | if (XFS_IS_CORRUPT(mp, |
2352 | be32_to_cpu(sib_info->forw) != last_blkno || | |
2353 | sib_info->magic != dead_info->magic)) { | |
12b53197 | 2354 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2355 | goto done; |
2356 | } | |
5e656dbb | 2357 | sib_info->forw = cpu_to_be32(dead_blkno); |
a2ceac1f | 2358 | xfs_trans_log_buf(tp, sib_buf, |
2bd0ea18 NS |
2359 | XFS_DA_LOGRANGE(sib_info, &sib_info->forw, |
2360 | sizeof(sib_info->forw))); | |
2bd0ea18 NS |
2361 | sib_buf = NULL; |
2362 | } | |
2363 | /* | |
2364 | * If the moved block has a right sibling, fix up the pointers. | |
2365 | */ | |
5e656dbb | 2366 | if ((sib_blkno = be32_to_cpu(dead_info->forw))) { |
02cc4995 | 2367 | error = xfs_da3_node_read(tp, dp, sib_blkno, &sib_buf, w); |
a2ceac1f | 2368 | if (error) |
2bd0ea18 | 2369 | goto done; |
a2ceac1f | 2370 | sib_info = sib_buf->b_addr; |
bc73da84 DW |
2371 | if (XFS_IS_CORRUPT(mp, |
2372 | be32_to_cpu(sib_info->back) != last_blkno || | |
2373 | sib_info->magic != dead_info->magic)) { | |
12b53197 | 2374 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2375 | goto done; |
2376 | } | |
5e656dbb | 2377 | sib_info->back = cpu_to_be32(dead_blkno); |
a2ceac1f | 2378 | xfs_trans_log_buf(tp, sib_buf, |
2bd0ea18 NS |
2379 | XFS_DA_LOGRANGE(sib_info, &sib_info->back, |
2380 | sizeof(sib_info->back))); | |
2bd0ea18 NS |
2381 | sib_buf = NULL; |
2382 | } | |
ff105f75 | 2383 | par_blkno = args->geo->leafblk; |
2bd0ea18 NS |
2384 | level = -1; |
2385 | /* | |
2386 | * Walk down the tree looking for the parent of the moved block. | |
2387 | */ | |
2388 | for (;;) { | |
02cc4995 | 2389 | error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w); |
a2ceac1f | 2390 | if (error) |
2bd0ea18 | 2391 | goto done; |
a2ceac1f | 2392 | par_node = par_buf->b_addr; |
08c16786 | 2393 | xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node); |
bc73da84 DW |
2394 | if (XFS_IS_CORRUPT(mp, |
2395 | level >= 0 && level != par_hdr.level + 1)) { | |
12b53197 | 2396 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2397 | goto done; |
2398 | } | |
88b32f06 | 2399 | level = par_hdr.level; |
8faa51a8 | 2400 | btree = par_hdr.btree; |
2bd0ea18 | 2401 | for (entno = 0; |
88b32f06 DC |
2402 | entno < par_hdr.count && |
2403 | be32_to_cpu(btree[entno].hashval) < dead_hash; | |
2bd0ea18 NS |
2404 | entno++) |
2405 | continue; | |
bc73da84 | 2406 | if (XFS_IS_CORRUPT(mp, entno == par_hdr.count)) { |
12b53197 | 2407 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2408 | goto done; |
2409 | } | |
88b32f06 | 2410 | par_blkno = be32_to_cpu(btree[entno].before); |
2bd0ea18 NS |
2411 | if (level == dead_level + 1) |
2412 | break; | |
a2ceac1f | 2413 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 NS |
2414 | par_buf = NULL; |
2415 | } | |
2416 | /* | |
2417 | * We're in the right parent block. | |
2418 | * Look for the right entry. | |
2419 | */ | |
2420 | for (;;) { | |
2421 | for (; | |
88b32f06 DC |
2422 | entno < par_hdr.count && |
2423 | be32_to_cpu(btree[entno].before) != last_blkno; | |
2bd0ea18 NS |
2424 | entno++) |
2425 | continue; | |
88b32f06 | 2426 | if (entno < par_hdr.count) |
2bd0ea18 | 2427 | break; |
88b32f06 | 2428 | par_blkno = par_hdr.forw; |
a2ceac1f | 2429 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 | 2430 | par_buf = NULL; |
bc73da84 | 2431 | if (XFS_IS_CORRUPT(mp, par_blkno == 0)) { |
12b53197 | 2432 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2433 | goto done; |
2434 | } | |
02cc4995 | 2435 | error = xfs_da3_node_read(tp, dp, par_blkno, &par_buf, w); |
a2ceac1f | 2436 | if (error) |
2bd0ea18 | 2437 | goto done; |
a2ceac1f | 2438 | par_node = par_buf->b_addr; |
08c16786 | 2439 | xfs_da3_node_hdr_from_disk(dp->i_mount, &par_hdr, par_node); |
bc73da84 | 2440 | if (XFS_IS_CORRUPT(mp, par_hdr.level != level)) { |
12b53197 | 2441 | error = -EFSCORRUPTED; |
2bd0ea18 NS |
2442 | goto done; |
2443 | } | |
8faa51a8 | 2444 | btree = par_hdr.btree; |
2bd0ea18 NS |
2445 | entno = 0; |
2446 | } | |
2447 | /* | |
2448 | * Update the parent entry pointing to the moved block. | |
2449 | */ | |
88b32f06 | 2450 | btree[entno].before = cpu_to_be32(dead_blkno); |
a2ceac1f | 2451 | xfs_trans_log_buf(tp, par_buf, |
88b32f06 DC |
2452 | XFS_DA_LOGRANGE(par_node, &btree[entno].before, |
2453 | sizeof(btree[entno].before))); | |
2bd0ea18 NS |
2454 | *dead_blknop = last_blkno; |
2455 | *dead_bufp = last_buf; | |
2456 | return 0; | |
2457 | done: | |
2458 | if (par_buf) | |
a2ceac1f | 2459 | xfs_trans_brelse(tp, par_buf); |
2bd0ea18 | 2460 | if (sib_buf) |
a2ceac1f DC |
2461 | xfs_trans_brelse(tp, sib_buf); |
2462 | xfs_trans_brelse(tp, last_buf); | |
2bd0ea18 NS |
2463 | return error; |
2464 | } | |
2465 | ||
2466 | /* | |
2467 | * Remove a btree block from a directory or attribute. | |
2468 | */ | |
2469 | int | |
a2ceac1f | 2470 | xfs_da_shrink_inode( |
a50d2ab0 BF |
2471 | struct xfs_da_args *args, |
2472 | xfs_dablk_t dead_blkno, | |
2473 | struct xfs_buf *dead_buf) | |
2bd0ea18 | 2474 | { |
a50d2ab0 BF |
2475 | struct xfs_inode *dp; |
2476 | int done, error, w, count; | |
2477 | struct xfs_trans *tp; | |
2bd0ea18 | 2478 | |
a2ceac1f DC |
2479 | trace_xfs_da_shrink_inode(args); |
2480 | ||
2bd0ea18 NS |
2481 | dp = args->dp; |
2482 | w = args->whichfork; | |
2483 | tp = args->trans; | |
ff105f75 | 2484 | count = args->geo->fsbcount; |
2bd0ea18 NS |
2485 | for (;;) { |
2486 | /* | |
2487 | * Remove extents. If we get ENOSPC for a dir we have to move | |
2488 | * the last block to the place we want to kill. | |
2489 | */ | |
88b32f06 | 2490 | error = xfs_bunmapi(tp, dp, dead_blkno, count, |
d3c5f3dd | 2491 | xfs_bmapi_aflag(w), 0, &done); |
12b53197 | 2492 | if (error == -ENOSPC) { |
2bd0ea18 | 2493 | if (w != XFS_DATA_FORK) |
5e656dbb | 2494 | break; |
88b32f06 DC |
2495 | error = xfs_da3_swap_lastblock(args, &dead_blkno, |
2496 | &dead_buf); | |
2497 | if (error) | |
5e656dbb BN |
2498 | break; |
2499 | } else { | |
2bd0ea18 | 2500 | break; |
2bd0ea18 NS |
2501 | } |
2502 | } | |
a2ceac1f | 2503 | xfs_trans_binval(tp, dead_buf); |
2bd0ea18 NS |
2504 | return error; |
2505 | } | |
2506 | ||
a2ceac1f DC |
2507 | static int |
2508 | xfs_dabuf_map( | |
a2ceac1f DC |
2509 | struct xfs_inode *dp, |
2510 | xfs_dablk_t bno, | |
48d1399b | 2511 | unsigned int flags, |
a2ceac1f | 2512 | int whichfork, |
571973e8 | 2513 | struct xfs_buf_map **mapp, |
a2ceac1f DC |
2514 | int *nmaps) |
2515 | { | |
2516 | struct xfs_mount *mp = dp->i_mount; | |
59ab3748 | 2517 | int nfsb = xfs_dabuf_nfsb(mp, whichfork); |
571973e8 CH |
2518 | struct xfs_bmbt_irec irec, *irecs = &irec; |
2519 | struct xfs_buf_map *map = *mapp; | |
2520 | xfs_fileoff_t off = bno; | |
2521 | int error = 0, nirecs, i; | |
2bd0ea18 | 2522 | |
571973e8 | 2523 | if (nfsb > 1) |
59ab3748 | 2524 | irecs = kmem_zalloc(sizeof(irec) * nfsb, KM_NOFS); |
571973e8 | 2525 | |
59ab3748 | 2526 | nirecs = nfsb; |
571973e8 CH |
2527 | error = xfs_bmapi_read(dp, bno, nfsb, irecs, &nirecs, |
2528 | xfs_bmapi_aflag(whichfork)); | |
59ab3748 | 2529 | if (error) |
571973e8 | 2530 | goto out_free_irecs; |
a2ceac1f | 2531 | |
571973e8 CH |
2532 | /* |
2533 | * Use the caller provided map for the single map case, else allocate a | |
2534 | * larger one that needs to be free by the caller. | |
2535 | */ | |
2536 | if (nirecs > 1) { | |
2537 | map = kmem_zalloc(nirecs * sizeof(struct xfs_buf_map), KM_NOFS); | |
a2a0a99a DW |
2538 | if (!map) { |
2539 | error = -ENOMEM; | |
571973e8 | 2540 | goto out_free_irecs; |
a2a0a99a | 2541 | } |
571973e8 CH |
2542 | *mapp = map; |
2543 | } | |
3af5535c | 2544 | |
571973e8 CH |
2545 | for (i = 0; i < nirecs; i++) { |
2546 | if (irecs[i].br_startblock == HOLESTARTBLOCK || | |
2547 | irecs[i].br_startblock == DELAYSTARTBLOCK) | |
2548 | goto invalid_mapping; | |
2549 | if (off != irecs[i].br_startoff) | |
2550 | goto invalid_mapping; | |
2551 | ||
2552 | map[i].bm_bn = XFS_FSB_TO_DADDR(mp, irecs[i].br_startblock); | |
2553 | map[i].bm_len = XFS_FSB_TO_BB(mp, irecs[i].br_blockcount); | |
2554 | off += irecs[i].br_blockcount; | |
2555 | } | |
2556 | ||
2557 | if (off != bno + nfsb) | |
2558 | goto invalid_mapping; | |
2559 | ||
2560 | *nmaps = nirecs; | |
2561 | out_free_irecs: | |
2562 | if (irecs != &irec) | |
2563 | kmem_free(irecs); | |
2564 | return error; | |
2565 | ||
2566 | invalid_mapping: | |
2567 | /* Caller ok with no mapping. */ | |
48d1399b | 2568 | if (XFS_IS_CORRUPT(mp, !(flags & XFS_DABUF_MAP_HOLE_OK))) { |
571973e8 | 2569 | error = -EFSCORRUPTED; |
3af5535c | 2570 | if (xfs_error_level >= XFS_ERRLEVEL_LOW) { |
571973e8 CH |
2571 | xfs_alert(mp, "%s: bno %u inode %llu", |
2572 | __func__, bno, dp->i_ino); | |
3af5535c | 2573 | |
571973e8 | 2574 | for (i = 0; i < nirecs; i++) { |
3af5535c | 2575 | xfs_alert(mp, |
a2ceac1f | 2576 | "[%02d] br_startoff %lld br_startblock %lld br_blockcount %lld br_state %d", |
571973e8 CH |
2577 | i, irecs[i].br_startoff, |
2578 | irecs[i].br_startblock, | |
2579 | irecs[i].br_blockcount, | |
3af5535c | 2580 | irecs[i].br_state); |
4ca431fc | 2581 | } |
4ca431fc | 2582 | } |
571973e8 CH |
2583 | } else { |
2584 | *nmaps = 0; | |
2bd0ea18 | 2585 | } |
571973e8 | 2586 | goto out_free_irecs; |
a2ceac1f DC |
2587 | } |
2588 | ||
2589 | /* | |
2590 | * Get a buffer for the dir/attr block. | |
2591 | */ | |
2592 | int | |
2593 | xfs_da_get_buf( | |
59ab3748 | 2594 | struct xfs_trans *tp, |
a2ceac1f DC |
2595 | struct xfs_inode *dp, |
2596 | xfs_dablk_t bno, | |
a2ceac1f DC |
2597 | struct xfs_buf **bpp, |
2598 | int whichfork) | |
2599 | { | |
59ab3748 | 2600 | struct xfs_mount *mp = dp->i_mount; |
a2ceac1f | 2601 | struct xfs_buf *bp; |
59ab3748 CH |
2602 | struct xfs_buf_map map, *mapp = ↦ |
2603 | int nmap = 1; | |
a2ceac1f DC |
2604 | int error; |
2605 | ||
2606 | *bpp = NULL; | |
c1d19744 | 2607 | error = xfs_dabuf_map(dp, bno, 0, whichfork, &mapp, &nmap); |
48d1399b | 2608 | if (error || nmap == 0) |
a2ceac1f | 2609 | goto out_free; |
a2ceac1f | 2610 | |
51409fcc DW |
2611 | error = xfs_trans_get_buf_map(tp, mp->m_ddev_targp, mapp, nmap, 0, &bp); |
2612 | if (error) | |
a2ceac1f | 2613 | goto out_free; |
a2ceac1f DC |
2614 | |
2615 | *bpp = bp; | |
2616 | ||
2617 | out_free: | |
2618 | if (mapp != &map) | |
2619 | kmem_free(mapp); | |
2620 | ||
2621 | return error; | |
2622 | } | |
2623 | ||
2624 | /* | |
2625 | * Get a buffer for the dir/attr block, fill in the contents. | |
2626 | */ | |
2627 | int | |
2628 | xfs_da_read_buf( | |
59ab3748 | 2629 | struct xfs_trans *tp, |
a2ceac1f DC |
2630 | struct xfs_inode *dp, |
2631 | xfs_dablk_t bno, | |
5f356ae6 | 2632 | unsigned int flags, |
a2ceac1f DC |
2633 | struct xfs_buf **bpp, |
2634 | int whichfork, | |
2635 | const struct xfs_buf_ops *ops) | |
2636 | { | |
59ab3748 | 2637 | struct xfs_mount *mp = dp->i_mount; |
a2ceac1f | 2638 | struct xfs_buf *bp; |
59ab3748 CH |
2639 | struct xfs_buf_map map, *mapp = ↦ |
2640 | int nmap = 1; | |
a2ceac1f DC |
2641 | int error; |
2642 | ||
2643 | *bpp = NULL; | |
5f356ae6 | 2644 | error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap); |
48d1399b | 2645 | if (error || !nmap) |
a2ceac1f | 2646 | goto out_free; |
a2ceac1f | 2647 | |
59ab3748 CH |
2648 | error = xfs_trans_read_buf_map(mp, tp, mp->m_ddev_targp, mapp, nmap, 0, |
2649 | &bp, ops); | |
a2ceac1f DC |
2650 | if (error) |
2651 | goto out_free; | |
2652 | ||
2653 | if (whichfork == XFS_ATTR_FORK) | |
2654 | xfs_buf_set_ref(bp, XFS_ATTR_BTREE_REF); | |
2bd0ea18 | 2655 | else |
a2ceac1f | 2656 | xfs_buf_set_ref(bp, XFS_DIR_BTREE_REF); |
a2ceac1f DC |
2657 | *bpp = bp; |
2658 | out_free: | |
2bd0ea18 | 2659 | if (mapp != &map) |
5e656dbb | 2660 | kmem_free(mapp); |
2bd0ea18 | 2661 | |
a2ceac1f | 2662 | return error; |
2bd0ea18 NS |
2663 | } |
2664 | ||
2665 | /* | |
5e656dbb | 2666 | * Readahead the dir/attr block. |
2bd0ea18 | 2667 | */ |
3a65062f | 2668 | int |
5e656dbb | 2669 | xfs_da_reada_buf( |
a2ceac1f DC |
2670 | struct xfs_inode *dp, |
2671 | xfs_dablk_t bno, | |
c6c4bbf3 | 2672 | unsigned int flags, |
a2ceac1f DC |
2673 | int whichfork, |
2674 | const struct xfs_buf_ops *ops) | |
2bd0ea18 | 2675 | { |
a2ceac1f DC |
2676 | struct xfs_buf_map map; |
2677 | struct xfs_buf_map *mapp; | |
2678 | int nmap; | |
2679 | int error; | |
2bd0ea18 | 2680 | |
a2ceac1f DC |
2681 | mapp = ↦ |
2682 | nmap = 1; | |
c6c4bbf3 | 2683 | error = xfs_dabuf_map(dp, bno, flags, whichfork, &mapp, &nmap); |
48d1399b | 2684 | if (error || !nmap) |
a2ceac1f | 2685 | goto out_free; |
a2ceac1f | 2686 | |
a2ceac1f DC |
2687 | xfs_buf_readahead_map(dp->i_mount->m_ddev_targp, mapp, nmap, ops); |
2688 | ||
2689 | out_free: | |
2690 | if (mapp != &map) | |
2691 | kmem_free(mapp); | |
2692 | ||
3a65062f | 2693 | return error; |
2bd0ea18 | 2694 | } |