]>
Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * alloc.c | |
5 | * | |
6 | * Extent allocs and frees | |
7 | * | |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/fs.h> | |
27 | #include <linux/types.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
60b11392 | 30 | #include <linux/swap.h> |
a90714c1 | 31 | #include <linux/quotaops.h> |
e80de36d | 32 | #include <linux/blkdev.h> |
174cd4b1 | 33 | #include <linux/sched/signal.h> |
ccd979bd | 34 | |
ccd979bd MF |
35 | #include <cluster/masklog.h> |
36 | ||
37 | #include "ocfs2.h" | |
38 | ||
39 | #include "alloc.h" | |
60b11392 | 40 | #include "aops.h" |
d6b32bbb | 41 | #include "blockcheck.h" |
ccd979bd MF |
42 | #include "dlmglue.h" |
43 | #include "extent_map.h" | |
44 | #include "inode.h" | |
45 | #include "journal.h" | |
46 | #include "localalloc.h" | |
47 | #include "suballoc.h" | |
48 | #include "sysfile.h" | |
49 | #include "file.h" | |
50 | #include "super.h" | |
51 | #include "uptodate.h" | |
2a50a743 | 52 | #include "xattr.h" |
bcbbb24a | 53 | #include "refcounttree.h" |
a09d09b8 | 54 | #include "ocfs2_trace.h" |
ccd979bd MF |
55 | |
56 | #include "buffer_head_io.h" | |
57 | ||
853a3a14 TM |
58 | enum ocfs2_contig_type { |
59 | CONTIG_NONE = 0, | |
60 | CONTIG_LEFT, | |
61 | CONTIG_RIGHT, | |
62 | CONTIG_LEFTRIGHT, | |
63 | }; | |
e7d4cb6b | 64 | |
853a3a14 TM |
65 | static enum ocfs2_contig_type |
66 | ocfs2_extent_rec_contig(struct super_block *sb, | |
67 | struct ocfs2_extent_rec *ext, | |
68 | struct ocfs2_extent_rec *insert_rec); | |
1625f8ac JB |
69 | /* |
70 | * Operations for a specific extent tree type. | |
71 | * | |
72 | * To implement an on-disk btree (extent tree) type in ocfs2, add | |
73 | * an ocfs2_extent_tree_operations structure and the matching | |
8d6220d6 | 74 | * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it |
1625f8ac JB |
75 | * for the allocation portion of the extent tree. |
76 | */ | |
e7d4cb6b | 77 | struct ocfs2_extent_tree_operations { |
1625f8ac JB |
78 | /* |
79 | * last_eb_blk is the block number of the right most leaf extent | |
80 | * block. Most on-disk structures containing an extent tree store | |
81 | * this value for fast access. The ->eo_set_last_eb_blk() and | |
82 | * ->eo_get_last_eb_blk() operations access this value. They are | |
83 | * both required. | |
84 | */ | |
35dc0aa3 JB |
85 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
86 | u64 blkno); | |
87 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
1625f8ac JB |
88 | |
89 | /* | |
90 | * The on-disk structure usually keeps track of how many total | |
91 | * clusters are stored in this extent tree. This function updates | |
92 | * that value. new_clusters is the delta, and must be | |
93 | * added to the total. Required. | |
94 | */ | |
6136ca5f | 95 | void (*eo_update_clusters)(struct ocfs2_extent_tree *et, |
35dc0aa3 | 96 | u32 new_clusters); |
1625f8ac | 97 | |
92ba470c JB |
98 | /* |
99 | * If this extent tree is supported by an extent map, insert | |
100 | * a record into the map. | |
101 | */ | |
102 | void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et, | |
103 | struct ocfs2_extent_rec *rec); | |
104 | ||
4c911eef JB |
105 | /* |
106 | * If this extent tree is supported by an extent map, truncate the | |
107 | * map to clusters, | |
108 | */ | |
109 | void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et, | |
110 | u32 clusters); | |
111 | ||
1625f8ac JB |
112 | /* |
113 | * If ->eo_insert_check() exists, it is called before rec is | |
114 | * inserted into the extent tree. It is optional. | |
115 | */ | |
6136ca5f | 116 | int (*eo_insert_check)(struct ocfs2_extent_tree *et, |
1e61ee79 | 117 | struct ocfs2_extent_rec *rec); |
6136ca5f | 118 | int (*eo_sanity_check)(struct ocfs2_extent_tree *et); |
0ce1010f | 119 | |
1625f8ac JB |
120 | /* |
121 | * -------------------------------------------------------------- | |
122 | * The remaining are internal to ocfs2_extent_tree and don't have | |
123 | * accessor functions | |
124 | */ | |
125 | ||
126 | /* | |
127 | * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el. | |
128 | * It is required. | |
129 | */ | |
0ce1010f | 130 | void (*eo_fill_root_el)(struct ocfs2_extent_tree *et); |
1625f8ac JB |
131 | |
132 | /* | |
133 | * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if | |
134 | * it exists. If it does not, et->et_max_leaf_clusters is set | |
135 | * to 0 (unlimited). Optional. | |
136 | */ | |
6136ca5f | 137 | void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et); |
853a3a14 TM |
138 | |
139 | /* | |
140 | * ->eo_extent_contig test whether the 2 ocfs2_extent_rec | |
141 | * are contiguous or not. Optional. Don't need to set it if use | |
142 | * ocfs2_extent_rec as the tree leaf. | |
143 | */ | |
144 | enum ocfs2_contig_type | |
145 | (*eo_extent_contig)(struct ocfs2_extent_tree *et, | |
146 | struct ocfs2_extent_rec *ext, | |
147 | struct ocfs2_extent_rec *insert_rec); | |
e7d4cb6b TM |
148 | }; |
149 | ||
e7d4cb6b | 150 | |
f99b9b7c JB |
151 | /* |
152 | * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check | |
153 | * in the methods. | |
154 | */ | |
155 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et); | |
156 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
157 | u64 blkno); | |
6136ca5f | 158 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
f99b9b7c | 159 | u32 clusters); |
92ba470c JB |
160 | static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et, |
161 | struct ocfs2_extent_rec *rec); | |
4c911eef JB |
162 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
163 | u32 clusters); | |
6136ca5f | 164 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
f99b9b7c | 165 | struct ocfs2_extent_rec *rec); |
6136ca5f | 166 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et); |
f99b9b7c | 167 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et); |
9e62dc09 | 168 | static const struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { |
f99b9b7c JB |
169 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, |
170 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
171 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
92ba470c | 172 | .eo_extent_map_insert = ocfs2_dinode_extent_map_insert, |
4c911eef | 173 | .eo_extent_map_truncate = ocfs2_dinode_extent_map_truncate, |
f99b9b7c JB |
174 | .eo_insert_check = ocfs2_dinode_insert_check, |
175 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
176 | .eo_fill_root_el = ocfs2_dinode_fill_root_el, | |
177 | }; | |
0ce1010f | 178 | |
e7d4cb6b TM |
179 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, |
180 | u64 blkno) | |
181 | { | |
ea5efa15 | 182 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 183 | |
f99b9b7c | 184 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
185 | di->i_last_eb_blk = cpu_to_le64(blkno); |
186 | } | |
187 | ||
188 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
189 | { | |
ea5efa15 | 190 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 191 | |
f99b9b7c | 192 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
193 | return le64_to_cpu(di->i_last_eb_blk); |
194 | } | |
195 | ||
6136ca5f | 196 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
e7d4cb6b TM |
197 | u32 clusters) |
198 | { | |
6136ca5f | 199 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
ea5efa15 | 200 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b TM |
201 | |
202 | le32_add_cpu(&di->i_clusters, clusters); | |
6136ca5f JB |
203 | spin_lock(&oi->ip_lock); |
204 | oi->ip_clusters = le32_to_cpu(di->i_clusters); | |
205 | spin_unlock(&oi->ip_lock); | |
e7d4cb6b TM |
206 | } |
207 | ||
92ba470c JB |
208 | static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et, |
209 | struct ocfs2_extent_rec *rec) | |
210 | { | |
211 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
212 | ||
213 | ocfs2_extent_map_insert_rec(inode, rec); | |
214 | } | |
215 | ||
4c911eef JB |
216 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
217 | u32 clusters) | |
218 | { | |
219 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
220 | ||
221 | ocfs2_extent_map_trunc(inode, clusters); | |
222 | } | |
223 | ||
6136ca5f | 224 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
225 | struct ocfs2_extent_rec *rec) |
226 | { | |
6136ca5f JB |
227 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
228 | struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb); | |
1e61ee79 | 229 | |
6136ca5f | 230 | BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL); |
1e61ee79 | 231 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && |
6136ca5f | 232 | (oi->ip_clusters != le32_to_cpu(rec->e_cpos)), |
1e61ee79 JB |
233 | "Device %s, asking for sparse allocation: inode %llu, " |
234 | "cpos %u, clusters %u\n", | |
235 | osb->dev_str, | |
6136ca5f JB |
236 | (unsigned long long)oi->ip_blkno, |
237 | rec->e_cpos, oi->ip_clusters); | |
1e61ee79 JB |
238 | |
239 | return 0; | |
240 | } | |
241 | ||
6136ca5f | 242 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 243 | { |
10995aa2 | 244 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 245 | |
f99b9b7c | 246 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
10995aa2 | 247 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); |
e7d4cb6b | 248 | |
10995aa2 | 249 | return 0; |
e7d4cb6b TM |
250 | } |
251 | ||
f99b9b7c JB |
252 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et) |
253 | { | |
254 | struct ocfs2_dinode *di = et->et_object; | |
255 | ||
256 | et->et_root_el = &di->id2.i_list; | |
257 | } | |
258 | ||
e7d4cb6b | 259 | |
0ce1010f JB |
260 | static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et) |
261 | { | |
2a50a743 | 262 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
0ce1010f | 263 | |
2a50a743 | 264 | et->et_root_el = &vb->vb_xv->xr_list; |
0ce1010f JB |
265 | } |
266 | ||
f56654c4 TM |
267 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
268 | u64 blkno) | |
269 | { | |
2a50a743 | 270 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 271 | |
2a50a743 | 272 | vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno); |
f56654c4 TM |
273 | } |
274 | ||
275 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
276 | { | |
2a50a743 | 277 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 278 | |
2a50a743 | 279 | return le64_to_cpu(vb->vb_xv->xr_last_eb_blk); |
f56654c4 TM |
280 | } |
281 | ||
6136ca5f | 282 | static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et, |
f56654c4 TM |
283 | u32 clusters) |
284 | { | |
2a50a743 | 285 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 286 | |
2a50a743 | 287 | le32_add_cpu(&vb->vb_xv->xr_clusters, clusters); |
f56654c4 TM |
288 | } |
289 | ||
9e62dc09 | 290 | static const struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = { |
35dc0aa3 JB |
291 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
292 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
293 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
0ce1010f | 294 | .eo_fill_root_el = ocfs2_xattr_value_fill_root_el, |
f56654c4 TM |
295 | }; |
296 | ||
0ce1010f JB |
297 | static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et) |
298 | { | |
299 | struct ocfs2_xattr_block *xb = et->et_object; | |
300 | ||
301 | et->et_root_el = &xb->xb_attrs.xb_root.xt_list; | |
302 | } | |
303 | ||
6136ca5f | 304 | static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et) |
943cced3 | 305 | { |
6136ca5f | 306 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
943cced3 | 307 | et->et_max_leaf_clusters = |
6136ca5f | 308 | ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE); |
943cced3 JB |
309 | } |
310 | ||
ba492615 TM |
311 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
312 | u64 blkno) | |
313 | { | |
ea5efa15 | 314 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
315 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
316 | ||
317 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
318 | } | |
319 | ||
320 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
321 | { | |
ea5efa15 | 322 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
323 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
324 | ||
325 | return le64_to_cpu(xt->xt_last_eb_blk); | |
326 | } | |
327 | ||
6136ca5f | 328 | static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et, |
ba492615 TM |
329 | u32 clusters) |
330 | { | |
ea5efa15 | 331 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
332 | |
333 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
334 | } | |
335 | ||
9e62dc09 | 336 | static const struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { |
35dc0aa3 JB |
337 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
338 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
339 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
0ce1010f | 340 | .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el, |
943cced3 | 341 | .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters, |
ba492615 TM |
342 | }; |
343 | ||
9b7895ef MF |
344 | static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et, |
345 | u64 blkno) | |
346 | { | |
347 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
348 | ||
349 | dx_root->dr_last_eb_blk = cpu_to_le64(blkno); | |
350 | } | |
351 | ||
352 | static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
353 | { | |
354 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
355 | ||
356 | return le64_to_cpu(dx_root->dr_last_eb_blk); | |
357 | } | |
358 | ||
6136ca5f | 359 | static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et, |
9b7895ef MF |
360 | u32 clusters) |
361 | { | |
362 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
363 | ||
364 | le32_add_cpu(&dx_root->dr_clusters, clusters); | |
365 | } | |
366 | ||
6136ca5f | 367 | static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et) |
9b7895ef MF |
368 | { |
369 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
370 | ||
371 | BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root)); | |
372 | ||
373 | return 0; | |
374 | } | |
375 | ||
376 | static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et) | |
377 | { | |
378 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
379 | ||
380 | et->et_root_el = &dx_root->dr_list; | |
381 | } | |
382 | ||
9e62dc09 | 383 | static const struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = { |
9b7895ef MF |
384 | .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk, |
385 | .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk, | |
386 | .eo_update_clusters = ocfs2_dx_root_update_clusters, | |
387 | .eo_sanity_check = ocfs2_dx_root_sanity_check, | |
388 | .eo_fill_root_el = ocfs2_dx_root_fill_root_el, | |
389 | }; | |
390 | ||
fe924415 TM |
391 | static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et) |
392 | { | |
393 | struct ocfs2_refcount_block *rb = et->et_object; | |
394 | ||
395 | et->et_root_el = &rb->rf_list; | |
396 | } | |
397 | ||
398 | static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
399 | u64 blkno) | |
400 | { | |
401 | struct ocfs2_refcount_block *rb = et->et_object; | |
402 | ||
403 | rb->rf_last_eb_blk = cpu_to_le64(blkno); | |
404 | } | |
405 | ||
406 | static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
407 | { | |
408 | struct ocfs2_refcount_block *rb = et->et_object; | |
409 | ||
410 | return le64_to_cpu(rb->rf_last_eb_blk); | |
411 | } | |
412 | ||
413 | static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et, | |
414 | u32 clusters) | |
415 | { | |
416 | struct ocfs2_refcount_block *rb = et->et_object; | |
417 | ||
418 | le32_add_cpu(&rb->rf_clusters, clusters); | |
419 | } | |
420 | ||
421 | static enum ocfs2_contig_type | |
422 | ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et, | |
423 | struct ocfs2_extent_rec *ext, | |
424 | struct ocfs2_extent_rec *insert_rec) | |
425 | { | |
426 | return CONTIG_NONE; | |
427 | } | |
428 | ||
9e62dc09 | 429 | static const struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = { |
fe924415 TM |
430 | .eo_set_last_eb_blk = ocfs2_refcount_tree_set_last_eb_blk, |
431 | .eo_get_last_eb_blk = ocfs2_refcount_tree_get_last_eb_blk, | |
432 | .eo_update_clusters = ocfs2_refcount_tree_update_clusters, | |
433 | .eo_fill_root_el = ocfs2_refcount_tree_fill_root_el, | |
434 | .eo_extent_contig = ocfs2_refcount_tree_extent_contig, | |
435 | }; | |
436 | ||
8d6220d6 | 437 | static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 438 | struct ocfs2_caching_info *ci, |
8d6220d6 | 439 | struct buffer_head *bh, |
13723d00 | 440 | ocfs2_journal_access_func access, |
8d6220d6 | 441 | void *obj, |
9e62dc09 | 442 | const struct ocfs2_extent_tree_operations *ops) |
e7d4cb6b | 443 | { |
1a09f556 | 444 | et->et_ops = ops; |
ce1d9ea6 | 445 | et->et_root_bh = bh; |
5e404e9e | 446 | et->et_ci = ci; |
13723d00 | 447 | et->et_root_journal_access = access; |
ea5efa15 JB |
448 | if (!obj) |
449 | obj = (void *)bh->b_data; | |
450 | et->et_object = obj; | |
e7d4cb6b | 451 | |
0ce1010f | 452 | et->et_ops->eo_fill_root_el(et); |
943cced3 JB |
453 | if (!et->et_ops->eo_fill_max_leaf_clusters) |
454 | et->et_max_leaf_clusters = 0; | |
455 | else | |
6136ca5f | 456 | et->et_ops->eo_fill_max_leaf_clusters(et); |
e7d4cb6b TM |
457 | } |
458 | ||
8d6220d6 | 459 | void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 460 | struct ocfs2_caching_info *ci, |
8d6220d6 | 461 | struct buffer_head *bh) |
1a09f556 | 462 | { |
5e404e9e | 463 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di, |
13723d00 | 464 | NULL, &ocfs2_dinode_et_ops); |
1a09f556 JB |
465 | } |
466 | ||
8d6220d6 | 467 | void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 468 | struct ocfs2_caching_info *ci, |
8d6220d6 | 469 | struct buffer_head *bh) |
1a09f556 | 470 | { |
5e404e9e | 471 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb, |
13723d00 | 472 | NULL, &ocfs2_xattr_tree_et_ops); |
1a09f556 JB |
473 | } |
474 | ||
8d6220d6 | 475 | void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 476 | struct ocfs2_caching_info *ci, |
2a50a743 | 477 | struct ocfs2_xattr_value_buf *vb) |
e7d4cb6b | 478 | { |
5e404e9e | 479 | __ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb, |
8d6220d6 | 480 | &ocfs2_xattr_value_et_ops); |
e7d4cb6b TM |
481 | } |
482 | ||
9b7895ef | 483 | void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 484 | struct ocfs2_caching_info *ci, |
9b7895ef MF |
485 | struct buffer_head *bh) |
486 | { | |
5e404e9e | 487 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr, |
9b7895ef MF |
488 | NULL, &ocfs2_dx_root_et_ops); |
489 | } | |
490 | ||
fe924415 TM |
491 | void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et, |
492 | struct ocfs2_caching_info *ci, | |
493 | struct buffer_head *bh) | |
494 | { | |
495 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb, | |
496 | NULL, &ocfs2_refcount_tree_et_ops); | |
497 | } | |
498 | ||
35dc0aa3 JB |
499 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
500 | u64 new_last_eb_blk) | |
e7d4cb6b | 501 | { |
ce1d9ea6 | 502 | et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
503 | } |
504 | ||
35dc0aa3 | 505 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 506 | { |
ce1d9ea6 | 507 | return et->et_ops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
508 | } |
509 | ||
6136ca5f | 510 | static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et, |
35dc0aa3 JB |
511 | u32 clusters) |
512 | { | |
6136ca5f | 513 | et->et_ops->eo_update_clusters(et, clusters); |
35dc0aa3 JB |
514 | } |
515 | ||
92ba470c JB |
516 | static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et, |
517 | struct ocfs2_extent_rec *rec) | |
518 | { | |
519 | if (et->et_ops->eo_extent_map_insert) | |
520 | et->et_ops->eo_extent_map_insert(et, rec); | |
521 | } | |
522 | ||
4c911eef JB |
523 | static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et, |
524 | u32 clusters) | |
525 | { | |
526 | if (et->et_ops->eo_extent_map_truncate) | |
527 | et->et_ops->eo_extent_map_truncate(et, clusters); | |
528 | } | |
529 | ||
13723d00 | 530 | static inline int ocfs2_et_root_journal_access(handle_t *handle, |
13723d00 JB |
531 | struct ocfs2_extent_tree *et, |
532 | int type) | |
533 | { | |
d9a0a1f8 | 534 | return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh, |
13723d00 JB |
535 | type); |
536 | } | |
537 | ||
853a3a14 TM |
538 | static inline enum ocfs2_contig_type |
539 | ocfs2_et_extent_contig(struct ocfs2_extent_tree *et, | |
540 | struct ocfs2_extent_rec *rec, | |
541 | struct ocfs2_extent_rec *insert_rec) | |
542 | { | |
543 | if (et->et_ops->eo_extent_contig) | |
544 | return et->et_ops->eo_extent_contig(et, rec, insert_rec); | |
545 | ||
546 | return ocfs2_extent_rec_contig( | |
547 | ocfs2_metadata_cache_get_super(et->et_ci), | |
548 | rec, insert_rec); | |
549 | } | |
550 | ||
6136ca5f | 551 | static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
552 | struct ocfs2_extent_rec *rec) |
553 | { | |
554 | int ret = 0; | |
555 | ||
556 | if (et->et_ops->eo_insert_check) | |
6136ca5f | 557 | ret = et->et_ops->eo_insert_check(et, rec); |
1e61ee79 JB |
558 | return ret; |
559 | } | |
560 | ||
6136ca5f | 561 | static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 562 | { |
1e61ee79 JB |
563 | int ret = 0; |
564 | ||
565 | if (et->et_ops->eo_sanity_check) | |
6136ca5f | 566 | ret = et->et_ops->eo_sanity_check(et); |
1e61ee79 | 567 | return ret; |
e7d4cb6b TM |
568 | } |
569 | ||
59a5e416 MF |
570 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
571 | struct ocfs2_extent_block *eb); | |
d401dc12 JB |
572 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
573 | struct ocfs2_extent_tree *et, | |
6b791bcc TM |
574 | struct ocfs2_path *path, |
575 | struct ocfs2_extent_rec *insert_rec); | |
dcd0538f MF |
576 | /* |
577 | * Reset the actual path elements so that we can re-use the structure | |
578 | * to build another path. Generally, this involves freeing the buffer | |
579 | * heads. | |
580 | */ | |
e2e9f608 | 581 | void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) |
dcd0538f MF |
582 | { |
583 | int i, start = 0, depth = 0; | |
584 | struct ocfs2_path_item *node; | |
ccd979bd | 585 | |
dcd0538f MF |
586 | if (keep_root) |
587 | start = 1; | |
ccd979bd | 588 | |
dcd0538f MF |
589 | for(i = start; i < path_num_items(path); i++) { |
590 | node = &path->p_node[i]; | |
591 | ||
592 | brelse(node->bh); | |
593 | node->bh = NULL; | |
594 | node->el = NULL; | |
595 | } | |
596 | ||
597 | /* | |
598 | * Tree depth may change during truncate, or insert. If we're | |
599 | * keeping the root extent list, then make sure that our path | |
600 | * structure reflects the proper depth. | |
601 | */ | |
602 | if (keep_root) | |
603 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
13723d00 JB |
604 | else |
605 | path_root_access(path) = NULL; | |
dcd0538f MF |
606 | |
607 | path->p_tree_depth = depth; | |
608 | } | |
609 | ||
e2e9f608 | 610 | void ocfs2_free_path(struct ocfs2_path *path) |
dcd0538f MF |
611 | { |
612 | if (path) { | |
613 | ocfs2_reinit_path(path, 0); | |
614 | kfree(path); | |
615 | } | |
616 | } | |
617 | ||
328d5752 MF |
618 | /* |
619 | * All the elements of src into dest. After this call, src could be freed | |
620 | * without affecting dest. | |
621 | * | |
622 | * Both paths should have the same root. Any non-root elements of dest | |
623 | * will be freed. | |
624 | */ | |
625 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
626 | { | |
627 | int i; | |
628 | ||
629 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
630 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
13723d00 | 631 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
328d5752 MF |
632 | |
633 | ocfs2_reinit_path(dest, 1); | |
634 | ||
635 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
636 | dest->p_node[i].bh = src->p_node[i].bh; | |
637 | dest->p_node[i].el = src->p_node[i].el; | |
638 | ||
639 | if (dest->p_node[i].bh) | |
640 | get_bh(dest->p_node[i].bh); | |
641 | } | |
642 | } | |
643 | ||
dcd0538f MF |
644 | /* |
645 | * Make the *dest path the same as src and re-initialize src path to | |
646 | * have a root only. | |
647 | */ | |
648 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
649 | { | |
650 | int i; | |
651 | ||
652 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
13723d00 | 653 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
dcd0538f MF |
654 | |
655 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
656 | brelse(dest->p_node[i].bh); | |
657 | ||
658 | dest->p_node[i].bh = src->p_node[i].bh; | |
659 | dest->p_node[i].el = src->p_node[i].el; | |
660 | ||
661 | src->p_node[i].bh = NULL; | |
662 | src->p_node[i].el = NULL; | |
663 | } | |
664 | } | |
665 | ||
666 | /* | |
667 | * Insert an extent block at given index. | |
668 | * | |
669 | * This will not take an additional reference on eb_bh. | |
670 | */ | |
671 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
672 | struct buffer_head *eb_bh) | |
673 | { | |
674 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
675 | ||
676 | /* | |
677 | * Right now, no root bh is an extent block, so this helps | |
678 | * catch code errors with dinode trees. The assertion can be | |
679 | * safely removed if we ever need to insert extent block | |
680 | * structures at the root. | |
681 | */ | |
682 | BUG_ON(index == 0); | |
683 | ||
684 | path->p_node[index].bh = eb_bh; | |
685 | path->p_node[index].el = &eb->h_list; | |
686 | } | |
687 | ||
688 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
13723d00 JB |
689 | struct ocfs2_extent_list *root_el, |
690 | ocfs2_journal_access_func access) | |
dcd0538f MF |
691 | { |
692 | struct ocfs2_path *path; | |
ccd979bd | 693 | |
dcd0538f MF |
694 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
695 | ||
696 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
697 | if (path) { | |
698 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
699 | get_bh(root_bh); | |
700 | path_root_bh(path) = root_bh; | |
701 | path_root_el(path) = root_el; | |
13723d00 | 702 | path_root_access(path) = access; |
dcd0538f MF |
703 | } |
704 | ||
705 | return path; | |
706 | } | |
707 | ||
e2e9f608 | 708 | struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path) |
ffdd7a54 | 709 | { |
13723d00 JB |
710 | return ocfs2_new_path(path_root_bh(path), path_root_el(path), |
711 | path_root_access(path)); | |
ffdd7a54 JB |
712 | } |
713 | ||
e2e9f608 | 714 | struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et) |
ffdd7a54 | 715 | { |
13723d00 JB |
716 | return ocfs2_new_path(et->et_root_bh, et->et_root_el, |
717 | et->et_root_journal_access); | |
718 | } | |
719 | ||
720 | /* | |
721 | * Journal the buffer at depth idx. All idx>0 are extent_blocks, | |
722 | * otherwise it's the root_access function. | |
723 | * | |
724 | * I don't like the way this function's name looks next to | |
725 | * ocfs2_journal_access_path(), but I don't have a better one. | |
726 | */ | |
e2e9f608 TM |
727 | int ocfs2_path_bh_journal_access(handle_t *handle, |
728 | struct ocfs2_caching_info *ci, | |
729 | struct ocfs2_path *path, | |
730 | int idx) | |
13723d00 JB |
731 | { |
732 | ocfs2_journal_access_func access = path_root_access(path); | |
733 | ||
734 | if (!access) | |
735 | access = ocfs2_journal_access; | |
736 | ||
737 | if (idx) | |
738 | access = ocfs2_journal_access_eb; | |
739 | ||
0cf2f763 | 740 | return access(handle, ci, path->p_node[idx].bh, |
13723d00 | 741 | OCFS2_JOURNAL_ACCESS_WRITE); |
ffdd7a54 JB |
742 | } |
743 | ||
dcd0538f MF |
744 | /* |
745 | * Convenience function to journal all components in a path. | |
746 | */ | |
e2e9f608 TM |
747 | int ocfs2_journal_access_path(struct ocfs2_caching_info *ci, |
748 | handle_t *handle, | |
749 | struct ocfs2_path *path) | |
dcd0538f MF |
750 | { |
751 | int i, ret = 0; | |
752 | ||
753 | if (!path) | |
754 | goto out; | |
755 | ||
756 | for(i = 0; i < path_num_items(path); i++) { | |
0cf2f763 | 757 | ret = ocfs2_path_bh_journal_access(handle, ci, path, i); |
dcd0538f MF |
758 | if (ret < 0) { |
759 | mlog_errno(ret); | |
760 | goto out; | |
761 | } | |
762 | } | |
763 | ||
764 | out: | |
765 | return ret; | |
766 | } | |
767 | ||
328d5752 MF |
768 | /* |
769 | * Return the index of the extent record which contains cluster #v_cluster. | |
770 | * -1 is returned if it was not found. | |
771 | * | |
772 | * Should work fine on interior and exterior nodes. | |
773 | */ | |
774 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
775 | { | |
776 | int ret = -1; | |
777 | int i; | |
778 | struct ocfs2_extent_rec *rec; | |
779 | u32 rec_end, rec_start, clusters; | |
780 | ||
781 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
782 | rec = &el->l_recs[i]; | |
783 | ||
784 | rec_start = le32_to_cpu(rec->e_cpos); | |
785 | clusters = ocfs2_rec_clusters(el, rec); | |
786 | ||
787 | rec_end = rec_start + clusters; | |
788 | ||
789 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
790 | ret = i; | |
791 | break; | |
792 | } | |
793 | } | |
794 | ||
795 | return ret; | |
796 | } | |
797 | ||
e48edee2 MF |
798 | /* |
799 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
853a3a14 | 800 | * ocfs2_extent_rec_contig only work properly against leaf nodes! |
e48edee2 | 801 | */ |
dcd0538f MF |
802 | static int ocfs2_block_extent_contig(struct super_block *sb, |
803 | struct ocfs2_extent_rec *ext, | |
804 | u64 blkno) | |
ccd979bd | 805 | { |
e48edee2 MF |
806 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
807 | ||
808 | blk_end += ocfs2_clusters_to_blocks(sb, | |
809 | le16_to_cpu(ext->e_leaf_clusters)); | |
810 | ||
811 | return blkno == blk_end; | |
ccd979bd MF |
812 | } |
813 | ||
dcd0538f MF |
814 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
815 | struct ocfs2_extent_rec *right) | |
816 | { | |
e48edee2 MF |
817 | u32 left_range; |
818 | ||
819 | left_range = le32_to_cpu(left->e_cpos) + | |
820 | le16_to_cpu(left->e_leaf_clusters); | |
821 | ||
822 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
823 | } |
824 | ||
825 | static enum ocfs2_contig_type | |
853a3a14 TM |
826 | ocfs2_extent_rec_contig(struct super_block *sb, |
827 | struct ocfs2_extent_rec *ext, | |
828 | struct ocfs2_extent_rec *insert_rec) | |
dcd0538f MF |
829 | { |
830 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
831 | ||
328d5752 MF |
832 | /* |
833 | * Refuse to coalesce extent records with different flag | |
834 | * fields - we don't want to mix unwritten extents with user | |
835 | * data. | |
836 | */ | |
837 | if (ext->e_flags != insert_rec->e_flags) | |
838 | return CONTIG_NONE; | |
839 | ||
dcd0538f | 840 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
b4a17651 | 841 | ocfs2_block_extent_contig(sb, ext, blkno)) |
dcd0538f MF |
842 | return CONTIG_RIGHT; |
843 | ||
844 | blkno = le64_to_cpu(ext->e_blkno); | |
845 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
b4a17651 | 846 | ocfs2_block_extent_contig(sb, insert_rec, blkno)) |
dcd0538f MF |
847 | return CONTIG_LEFT; |
848 | ||
849 | return CONTIG_NONE; | |
850 | } | |
851 | ||
852 | /* | |
853 | * NOTE: We can have pretty much any combination of contiguousness and | |
854 | * appending. | |
855 | * | |
856 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
857 | * we'll have to update the path to that leaf. | |
858 | */ | |
859 | enum ocfs2_append_type { | |
860 | APPEND_NONE = 0, | |
861 | APPEND_TAIL, | |
862 | }; | |
863 | ||
328d5752 MF |
864 | enum ocfs2_split_type { |
865 | SPLIT_NONE = 0, | |
866 | SPLIT_LEFT, | |
867 | SPLIT_RIGHT, | |
868 | }; | |
869 | ||
dcd0538f | 870 | struct ocfs2_insert_type { |
328d5752 | 871 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
872 | enum ocfs2_append_type ins_appending; |
873 | enum ocfs2_contig_type ins_contig; | |
874 | int ins_contig_index; | |
dcd0538f MF |
875 | int ins_tree_depth; |
876 | }; | |
877 | ||
328d5752 MF |
878 | struct ocfs2_merge_ctxt { |
879 | enum ocfs2_contig_type c_contig_type; | |
880 | int c_has_empty_extent; | |
881 | int c_split_covers_rec; | |
328d5752 MF |
882 | }; |
883 | ||
5e96581a JB |
884 | static int ocfs2_validate_extent_block(struct super_block *sb, |
885 | struct buffer_head *bh) | |
886 | { | |
d6b32bbb | 887 | int rc; |
5e96581a JB |
888 | struct ocfs2_extent_block *eb = |
889 | (struct ocfs2_extent_block *)bh->b_data; | |
890 | ||
a09d09b8 | 891 | trace_ocfs2_validate_extent_block((unsigned long long)bh->b_blocknr); |
970e4936 | 892 | |
d6b32bbb JB |
893 | BUG_ON(!buffer_uptodate(bh)); |
894 | ||
895 | /* | |
896 | * If the ecc fails, we return the error but otherwise | |
897 | * leave the filesystem running. We know any error is | |
898 | * local to this block. | |
899 | */ | |
900 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check); | |
13723d00 JB |
901 | if (rc) { |
902 | mlog(ML_ERROR, "Checksum failed for extent block %llu\n", | |
903 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 904 | return rc; |
13723d00 | 905 | } |
d6b32bbb JB |
906 | |
907 | /* | |
908 | * Errors after here are fatal. | |
909 | */ | |
910 | ||
5e96581a | 911 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { |
17a5b9ab | 912 | rc = ocfs2_error(sb, |
7ecef14a JP |
913 | "Extent block #%llu has bad signature %.*s\n", |
914 | (unsigned long long)bh->b_blocknr, 7, | |
915 | eb->h_signature); | |
17a5b9ab | 916 | goto bail; |
5e96581a JB |
917 | } |
918 | ||
919 | if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) { | |
17a5b9ab | 920 | rc = ocfs2_error(sb, |
7ecef14a JP |
921 | "Extent block #%llu has an invalid h_blkno of %llu\n", |
922 | (unsigned long long)bh->b_blocknr, | |
923 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
17a5b9ab | 924 | goto bail; |
5e96581a JB |
925 | } |
926 | ||
927 | if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) { | |
17a5b9ab | 928 | rc = ocfs2_error(sb, |
7ecef14a JP |
929 | "Extent block #%llu has an invalid h_fs_generation of #%u\n", |
930 | (unsigned long long)bh->b_blocknr, | |
931 | le32_to_cpu(eb->h_fs_generation)); | |
17a5b9ab | 932 | goto bail; |
5e96581a | 933 | } |
17a5b9ab GR |
934 | bail: |
935 | return rc; | |
5e96581a JB |
936 | } |
937 | ||
3d03a305 | 938 | int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno, |
5e96581a JB |
939 | struct buffer_head **bh) |
940 | { | |
941 | int rc; | |
942 | struct buffer_head *tmp = *bh; | |
943 | ||
3d03a305 | 944 | rc = ocfs2_read_block(ci, eb_blkno, &tmp, |
970e4936 | 945 | ocfs2_validate_extent_block); |
5e96581a JB |
946 | |
947 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
970e4936 | 948 | if (!rc && !*bh) |
5e96581a JB |
949 | *bh = tmp; |
950 | ||
5e96581a JB |
951 | return rc; |
952 | } | |
953 | ||
954 | ||
ccd979bd MF |
955 | /* |
956 | * How many free extents have we got before we need more meta data? | |
957 | */ | |
958 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
f99b9b7c | 959 | struct ocfs2_extent_tree *et) |
ccd979bd MF |
960 | { |
961 | int retval; | |
e7d4cb6b | 962 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
963 | struct ocfs2_extent_block *eb; |
964 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 965 | u64 last_eb_blk = 0; |
ccd979bd | 966 | |
f99b9b7c JB |
967 | el = et->et_root_el; |
968 | last_eb_blk = ocfs2_et_get_last_eb_blk(et); | |
ccd979bd | 969 | |
e7d4cb6b | 970 | if (last_eb_blk) { |
3d03a305 JB |
971 | retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk, |
972 | &eb_bh); | |
ccd979bd MF |
973 | if (retval < 0) { |
974 | mlog_errno(retval); | |
975 | goto bail; | |
976 | } | |
977 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
978 | el = &eb->h_list; | |
e7d4cb6b | 979 | } |
ccd979bd MF |
980 | |
981 | BUG_ON(el->l_tree_depth != 0); | |
982 | ||
983 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
984 | bail: | |
a81cb88b | 985 | brelse(eb_bh); |
ccd979bd | 986 | |
a09d09b8 | 987 | trace_ocfs2_num_free_extents(retval); |
ccd979bd MF |
988 | return retval; |
989 | } | |
990 | ||
991 | /* expects array to already be allocated | |
992 | * | |
993 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
994 | * l_count for you | |
995 | */ | |
42a5a7a9 JB |
996 | static int ocfs2_create_new_meta_bhs(handle_t *handle, |
997 | struct ocfs2_extent_tree *et, | |
ccd979bd MF |
998 | int wanted, |
999 | struct ocfs2_alloc_context *meta_ac, | |
1000 | struct buffer_head *bhs[]) | |
1001 | { | |
1002 | int count, status, i; | |
1003 | u16 suballoc_bit_start; | |
1004 | u32 num_got; | |
2b6cb576 | 1005 | u64 suballoc_loc, first_blkno; |
42a5a7a9 JB |
1006 | struct ocfs2_super *osb = |
1007 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
ccd979bd MF |
1008 | struct ocfs2_extent_block *eb; |
1009 | ||
ccd979bd MF |
1010 | count = 0; |
1011 | while (count < wanted) { | |
1ed9b777 | 1012 | status = ocfs2_claim_metadata(handle, |
ccd979bd MF |
1013 | meta_ac, |
1014 | wanted - count, | |
2b6cb576 | 1015 | &suballoc_loc, |
ccd979bd MF |
1016 | &suballoc_bit_start, |
1017 | &num_got, | |
1018 | &first_blkno); | |
1019 | if (status < 0) { | |
1020 | mlog_errno(status); | |
1021 | goto bail; | |
1022 | } | |
1023 | ||
1024 | for(i = count; i < (num_got + count); i++) { | |
1025 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
1026 | if (bhs[i] == NULL) { | |
7391a294 | 1027 | status = -ENOMEM; |
ccd979bd MF |
1028 | mlog_errno(status); |
1029 | goto bail; | |
1030 | } | |
42a5a7a9 | 1031 | ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]); |
ccd979bd | 1032 | |
42a5a7a9 JB |
1033 | status = ocfs2_journal_access_eb(handle, et->et_ci, |
1034 | bhs[i], | |
13723d00 | 1035 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1036 | if (status < 0) { |
1037 | mlog_errno(status); | |
1038 | goto bail; | |
1039 | } | |
1040 | ||
1041 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
1042 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
1043 | /* Ok, setup the minimal stuff here. */ | |
1044 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
1045 | eb->h_blkno = cpu_to_le64(first_blkno); | |
1046 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
b89c5428 TY |
1047 | eb->h_suballoc_slot = |
1048 | cpu_to_le16(meta_ac->ac_alloc_slot); | |
2b6cb576 | 1049 | eb->h_suballoc_loc = cpu_to_le64(suballoc_loc); |
ccd979bd MF |
1050 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
1051 | eb->h_list.l_count = | |
1052 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
1053 | ||
1054 | suballoc_bit_start++; | |
1055 | first_blkno++; | |
1056 | ||
1057 | /* We'll also be dirtied by the caller, so | |
1058 | * this isn't absolutely necessary. */ | |
ec20cec7 | 1059 | ocfs2_journal_dirty(handle, bhs[i]); |
ccd979bd MF |
1060 | } |
1061 | ||
1062 | count += num_got; | |
1063 | } | |
1064 | ||
1065 | status = 0; | |
1066 | bail: | |
1067 | if (status < 0) { | |
1068 | for(i = 0; i < wanted; i++) { | |
a81cb88b | 1069 | brelse(bhs[i]); |
ccd979bd MF |
1070 | bhs[i] = NULL; |
1071 | } | |
c1e8d35e | 1072 | mlog_errno(status); |
ccd979bd | 1073 | } |
ccd979bd MF |
1074 | return status; |
1075 | } | |
1076 | ||
dcd0538f MF |
1077 | /* |
1078 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
1079 | * | |
1080 | * Returns the sum of the rightmost extent rec logical offset and | |
1081 | * cluster count. | |
1082 | * | |
1083 | * ocfs2_add_branch() uses this to determine what logical cluster | |
1084 | * value should be populated into the leftmost new branch records. | |
1085 | * | |
1086 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
1087 | * value for the new topmost tree record. | |
1088 | */ | |
1089 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
1090 | { | |
1091 | int i; | |
1092 | ||
1093 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1094 | ||
1095 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 1096 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
1097 | } |
1098 | ||
6b791bcc TM |
1099 | /* |
1100 | * Change range of the branches in the right most path according to the leaf | |
1101 | * extent block's rightmost record. | |
1102 | */ | |
1103 | static int ocfs2_adjust_rightmost_branch(handle_t *handle, | |
6b791bcc TM |
1104 | struct ocfs2_extent_tree *et) |
1105 | { | |
1106 | int status; | |
1107 | struct ocfs2_path *path = NULL; | |
1108 | struct ocfs2_extent_list *el; | |
1109 | struct ocfs2_extent_rec *rec; | |
1110 | ||
1111 | path = ocfs2_new_path_from_et(et); | |
1112 | if (!path) { | |
1113 | status = -ENOMEM; | |
1114 | return status; | |
1115 | } | |
1116 | ||
facdb77f | 1117 | status = ocfs2_find_path(et->et_ci, path, UINT_MAX); |
6b791bcc TM |
1118 | if (status < 0) { |
1119 | mlog_errno(status); | |
1120 | goto out; | |
1121 | } | |
1122 | ||
c901fb00 | 1123 | status = ocfs2_extend_trans(handle, path_num_items(path)); |
6b791bcc TM |
1124 | if (status < 0) { |
1125 | mlog_errno(status); | |
1126 | goto out; | |
1127 | } | |
1128 | ||
d401dc12 | 1129 | status = ocfs2_journal_access_path(et->et_ci, handle, path); |
6b791bcc TM |
1130 | if (status < 0) { |
1131 | mlog_errno(status); | |
1132 | goto out; | |
1133 | } | |
1134 | ||
1135 | el = path_leaf_el(path); | |
3a251f04 | 1136 | rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec) - 1]; |
6b791bcc | 1137 | |
d401dc12 | 1138 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
6b791bcc TM |
1139 | |
1140 | out: | |
1141 | ocfs2_free_path(path); | |
1142 | return status; | |
1143 | } | |
1144 | ||
ccd979bd MF |
1145 | /* |
1146 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
d401dc12 | 1147 | * to start at, if we don't want to start the branch at the root |
ccd979bd MF |
1148 | * structure. |
1149 | * | |
1150 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
1151 | * for the new last extent block. | |
1152 | * | |
1153 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 1154 | * contain a single record with cluster count == 0. |
ccd979bd | 1155 | */ |
d401dc12 | 1156 | static int ocfs2_add_branch(handle_t *handle, |
e7d4cb6b | 1157 | struct ocfs2_extent_tree *et, |
ccd979bd | 1158 | struct buffer_head *eb_bh, |
328d5752 | 1159 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
1160 | struct ocfs2_alloc_context *meta_ac) |
1161 | { | |
1162 | int status, new_blocks, i; | |
1163 | u64 next_blkno, new_last_eb_blk; | |
1164 | struct buffer_head *bh; | |
1165 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
1166 | struct ocfs2_extent_block *eb; |
1167 | struct ocfs2_extent_list *eb_el; | |
1168 | struct ocfs2_extent_list *el; | |
6b791bcc | 1169 | u32 new_cpos, root_end; |
ccd979bd | 1170 | |
328d5752 | 1171 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 1172 | |
ccd979bd MF |
1173 | if (eb_bh) { |
1174 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
1175 | el = &eb->h_list; | |
1176 | } else | |
ce1d9ea6 | 1177 | el = et->et_root_el; |
ccd979bd MF |
1178 | |
1179 | /* we never add a branch to a leaf. */ | |
1180 | BUG_ON(!el->l_tree_depth); | |
1181 | ||
1182 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
1183 | ||
6b791bcc TM |
1184 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
1185 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); | |
1186 | root_end = ocfs2_sum_rightmost_rec(et->et_root_el); | |
1187 | ||
1188 | /* | |
1189 | * If there is a gap before the root end and the real end | |
1190 | * of the righmost leaf block, we need to remove the gap | |
1191 | * between new_cpos and root_end first so that the tree | |
1192 | * is consistent after we add a new branch(it will start | |
1193 | * from new_cpos). | |
1194 | */ | |
1195 | if (root_end > new_cpos) { | |
a09d09b8 TM |
1196 | trace_ocfs2_adjust_rightmost_branch( |
1197 | (unsigned long long) | |
1198 | ocfs2_metadata_cache_owner(et->et_ci), | |
1199 | root_end, new_cpos); | |
1200 | ||
d401dc12 | 1201 | status = ocfs2_adjust_rightmost_branch(handle, et); |
6b791bcc TM |
1202 | if (status) { |
1203 | mlog_errno(status); | |
1204 | goto bail; | |
1205 | } | |
1206 | } | |
1207 | ||
ccd979bd MF |
1208 | /* allocate the number of new eb blocks we need */ |
1209 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
1210 | GFP_KERNEL); | |
1211 | if (!new_eb_bhs) { | |
1212 | status = -ENOMEM; | |
1213 | mlog_errno(status); | |
1214 | goto bail; | |
1215 | } | |
1216 | ||
42a5a7a9 | 1217 | status = ocfs2_create_new_meta_bhs(handle, et, new_blocks, |
ccd979bd MF |
1218 | meta_ac, new_eb_bhs); |
1219 | if (status < 0) { | |
1220 | mlog_errno(status); | |
1221 | goto bail; | |
1222 | } | |
1223 | ||
1224 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be | |
1225 | * linked with the rest of the tree. | |
1226 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
1227 | * | |
1228 | * when we leave the loop, new_last_eb_blk will point to the | |
1229 | * newest leaf, and next_blkno will point to the topmost extent | |
1230 | * block. */ | |
1231 | next_blkno = new_last_eb_blk = 0; | |
1232 | for(i = 0; i < new_blocks; i++) { | |
1233 | bh = new_eb_bhs[i]; | |
1234 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
5e96581a JB |
1235 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1236 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1237 | eb_el = &eb->h_list; |
1238 | ||
d401dc12 | 1239 | status = ocfs2_journal_access_eb(handle, et->et_ci, bh, |
13723d00 | 1240 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1241 | if (status < 0) { |
1242 | mlog_errno(status); | |
1243 | goto bail; | |
1244 | } | |
1245 | ||
1246 | eb->h_next_leaf_blk = 0; | |
1247 | eb_el->l_tree_depth = cpu_to_le16(i); | |
1248 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
1249 | /* |
1250 | * This actually counts as an empty extent as | |
1251 | * c_clusters == 0 | |
1252 | */ | |
1253 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 1254 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
1255 | /* |
1256 | * eb_el isn't always an interior node, but even leaf | |
1257 | * nodes want a zero'd flags and reserved field so | |
1258 | * this gets the whole 32 bits regardless of use. | |
1259 | */ | |
1260 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
1261 | if (!eb_el->l_tree_depth) |
1262 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
1263 | ||
ec20cec7 | 1264 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
1265 | next_blkno = le64_to_cpu(eb->h_blkno); |
1266 | } | |
1267 | ||
1268 | /* This is a bit hairy. We want to update up to three blocks | |
1269 | * here without leaving any of them in an inconsistent state | |
1270 | * in case of error. We don't have to worry about | |
1271 | * journal_dirty erroring as it won't unless we've aborted the | |
1272 | * handle (in which case we would never be here) so reserving | |
1273 | * the write with journal_access is all we need to do. */ | |
d401dc12 | 1274 | status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh, |
13723d00 | 1275 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1276 | if (status < 0) { |
1277 | mlog_errno(status); | |
1278 | goto bail; | |
1279 | } | |
d9a0a1f8 | 1280 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1281 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1282 | if (status < 0) { |
1283 | mlog_errno(status); | |
1284 | goto bail; | |
1285 | } | |
1286 | if (eb_bh) { | |
d401dc12 | 1287 | status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh, |
13723d00 | 1288 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1289 | if (status < 0) { |
1290 | mlog_errno(status); | |
1291 | goto bail; | |
1292 | } | |
1293 | } | |
1294 | ||
1295 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 1296 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
1297 | i = le16_to_cpu(el->l_next_free_rec); |
1298 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 1299 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 1300 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
1301 | le16_add_cpu(&el->l_next_free_rec, 1); |
1302 | ||
1303 | /* fe needs a new last extent block pointer, as does the | |
1304 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 1305 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 1306 | |
328d5752 | 1307 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
1308 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
1309 | ||
ec20cec7 JB |
1310 | ocfs2_journal_dirty(handle, *last_eb_bh); |
1311 | ocfs2_journal_dirty(handle, et->et_root_bh); | |
1312 | if (eb_bh) | |
1313 | ocfs2_journal_dirty(handle, eb_bh); | |
ccd979bd | 1314 | |
328d5752 MF |
1315 | /* |
1316 | * Some callers want to track the rightmost leaf so pass it | |
1317 | * back here. | |
1318 | */ | |
1319 | brelse(*last_eb_bh); | |
1320 | get_bh(new_eb_bhs[0]); | |
1321 | *last_eb_bh = new_eb_bhs[0]; | |
1322 | ||
ccd979bd MF |
1323 | status = 0; |
1324 | bail: | |
1325 | if (new_eb_bhs) { | |
1326 | for (i = 0; i < new_blocks; i++) | |
a81cb88b | 1327 | brelse(new_eb_bhs[i]); |
ccd979bd MF |
1328 | kfree(new_eb_bhs); |
1329 | } | |
1330 | ||
ccd979bd MF |
1331 | return status; |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * adds another level to the allocation tree. | |
1336 | * returns back the new extent block so you can add a branch to it | |
1337 | * after this call. | |
1338 | */ | |
d401dc12 | 1339 | static int ocfs2_shift_tree_depth(handle_t *handle, |
e7d4cb6b | 1340 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1341 | struct ocfs2_alloc_context *meta_ac, |
1342 | struct buffer_head **ret_new_eb_bh) | |
1343 | { | |
1344 | int status, i; | |
dcd0538f | 1345 | u32 new_clusters; |
ccd979bd | 1346 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 1347 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 1348 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
1349 | struct ocfs2_extent_list *eb_el; |
1350 | ||
42a5a7a9 | 1351 | status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac, |
ccd979bd MF |
1352 | &new_eb_bh); |
1353 | if (status < 0) { | |
1354 | mlog_errno(status); | |
1355 | goto bail; | |
1356 | } | |
1357 | ||
1358 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
5e96581a JB |
1359 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1360 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1361 | |
1362 | eb_el = &eb->h_list; | |
ce1d9ea6 | 1363 | root_el = et->et_root_el; |
ccd979bd | 1364 | |
d401dc12 | 1365 | status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh, |
13723d00 | 1366 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1367 | if (status < 0) { |
1368 | mlog_errno(status); | |
1369 | goto bail; | |
1370 | } | |
1371 | ||
e7d4cb6b TM |
1372 | /* copy the root extent list data into the new extent block */ |
1373 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1374 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1375 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1376 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd | 1377 | |
ec20cec7 | 1378 | ocfs2_journal_dirty(handle, new_eb_bh); |
ccd979bd | 1379 | |
d9a0a1f8 | 1380 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1381 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1382 | if (status < 0) { |
1383 | mlog_errno(status); | |
1384 | goto bail; | |
1385 | } | |
1386 | ||
dcd0538f MF |
1387 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1388 | ||
e7d4cb6b TM |
1389 | /* update root_bh now */ |
1390 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1391 | root_el->l_recs[0].e_cpos = 0; | |
1392 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1393 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1394 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1395 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1396 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1397 | |
1398 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1399 | * becomes the allocated extent block */ | |
e7d4cb6b | 1400 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1401 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1402 | |
ec20cec7 | 1403 | ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1404 | |
1405 | *ret_new_eb_bh = new_eb_bh; | |
1406 | new_eb_bh = NULL; | |
1407 | status = 0; | |
1408 | bail: | |
a81cb88b | 1409 | brelse(new_eb_bh); |
ccd979bd | 1410 | |
ccd979bd MF |
1411 | return status; |
1412 | } | |
1413 | ||
ccd979bd MF |
1414 | /* |
1415 | * Should only be called when there is no space left in any of the | |
1416 | * leaf nodes. What we want to do is find the lowest tree depth | |
1417 | * non-leaf extent block with room for new records. There are three | |
1418 | * valid results of this search: | |
1419 | * | |
1420 | * 1) a lowest extent block is found, then we pass it back in | |
1421 | * *lowest_eb_bh and return '0' | |
1422 | * | |
e7d4cb6b | 1423 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1424 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1425 | * | |
e7d4cb6b | 1426 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1427 | * which case we return > 0 |
1428 | * | |
1429 | * return status < 0 indicates an error. | |
1430 | */ | |
d401dc12 | 1431 | static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et, |
ccd979bd MF |
1432 | struct buffer_head **target_bh) |
1433 | { | |
1434 | int status = 0, i; | |
1435 | u64 blkno; | |
ccd979bd MF |
1436 | struct ocfs2_extent_block *eb; |
1437 | struct ocfs2_extent_list *el; | |
1438 | struct buffer_head *bh = NULL; | |
1439 | struct buffer_head *lowest_bh = NULL; | |
1440 | ||
ccd979bd MF |
1441 | *target_bh = NULL; |
1442 | ||
ce1d9ea6 | 1443 | el = et->et_root_el; |
ccd979bd MF |
1444 | |
1445 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1446 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3d03a305 | 1447 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
7ecef14a | 1448 | "Owner %llu has empty extent list (next_free_rec == 0)\n", |
3d03a305 | 1449 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); |
ccd979bd MF |
1450 | status = -EIO; |
1451 | goto bail; | |
1452 | } | |
1453 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1454 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1455 | if (!blkno) { | |
3d03a305 | 1456 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
7ecef14a | 1457 | "Owner %llu has extent list where extent # %d has no physical block start\n", |
3d03a305 | 1458 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i); |
ccd979bd MF |
1459 | status = -EIO; |
1460 | goto bail; | |
1461 | } | |
1462 | ||
a81cb88b MF |
1463 | brelse(bh); |
1464 | bh = NULL; | |
ccd979bd | 1465 | |
3d03a305 | 1466 | status = ocfs2_read_extent_block(et->et_ci, blkno, &bh); |
ccd979bd MF |
1467 | if (status < 0) { |
1468 | mlog_errno(status); | |
1469 | goto bail; | |
1470 | } | |
dcd0538f MF |
1471 | |
1472 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
dcd0538f MF |
1473 | el = &eb->h_list; |
1474 | ||
1475 | if (le16_to_cpu(el->l_next_free_rec) < | |
1476 | le16_to_cpu(el->l_count)) { | |
a81cb88b | 1477 | brelse(lowest_bh); |
dcd0538f MF |
1478 | lowest_bh = bh; |
1479 | get_bh(lowest_bh); | |
1480 | } | |
1481 | } | |
1482 | ||
1483 | /* If we didn't find one and the fe doesn't have any room, | |
1484 | * then return '1' */ | |
ce1d9ea6 | 1485 | el = et->et_root_el; |
e7d4cb6b | 1486 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) |
dcd0538f MF |
1487 | status = 1; |
1488 | ||
1489 | *target_bh = lowest_bh; | |
1490 | bail: | |
a81cb88b | 1491 | brelse(bh); |
dcd0538f | 1492 | |
dcd0538f MF |
1493 | return status; |
1494 | } | |
1495 | ||
c3afcbb3 MF |
1496 | /* |
1497 | * Grow a b-tree so that it has more records. | |
1498 | * | |
1499 | * We might shift the tree depth in which case existing paths should | |
1500 | * be considered invalid. | |
1501 | * | |
1502 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1503 | * |
1504 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 | 1505 | */ |
d401dc12 JB |
1506 | static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
1507 | int *final_depth, struct buffer_head **last_eb_bh, | |
c3afcbb3 MF |
1508 | struct ocfs2_alloc_context *meta_ac) |
1509 | { | |
1510 | int ret, shift; | |
ce1d9ea6 | 1511 | struct ocfs2_extent_list *el = et->et_root_el; |
e7d4cb6b | 1512 | int depth = le16_to_cpu(el->l_tree_depth); |
c3afcbb3 MF |
1513 | struct buffer_head *bh = NULL; |
1514 | ||
1515 | BUG_ON(meta_ac == NULL); | |
1516 | ||
d401dc12 | 1517 | shift = ocfs2_find_branch_target(et, &bh); |
c3afcbb3 MF |
1518 | if (shift < 0) { |
1519 | ret = shift; | |
1520 | mlog_errno(ret); | |
1521 | goto out; | |
1522 | } | |
1523 | ||
1524 | /* We traveled all the way to the bottom of the allocation tree | |
1525 | * and didn't find room for any more extents - we need to add | |
1526 | * another tree level */ | |
1527 | if (shift) { | |
1528 | BUG_ON(bh); | |
a09d09b8 TM |
1529 | trace_ocfs2_grow_tree( |
1530 | (unsigned long long) | |
1531 | ocfs2_metadata_cache_owner(et->et_ci), | |
1532 | depth); | |
c3afcbb3 MF |
1533 | |
1534 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1535 | * the new extent block (so we can pass that to | |
1536 | * ocfs2_add_branch). */ | |
d401dc12 | 1537 | ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh); |
c3afcbb3 MF |
1538 | if (ret < 0) { |
1539 | mlog_errno(ret); | |
1540 | goto out; | |
1541 | } | |
1542 | depth++; | |
328d5752 MF |
1543 | if (depth == 1) { |
1544 | /* | |
1545 | * Special case: we have room now if we shifted from | |
1546 | * tree_depth 0, so no more work needs to be done. | |
1547 | * | |
1548 | * We won't be calling add_branch, so pass | |
1549 | * back *last_eb_bh as the new leaf. At depth | |
1550 | * zero, it should always be null so there's | |
1551 | * no reason to brelse. | |
1552 | */ | |
1553 | BUG_ON(*last_eb_bh); | |
1554 | get_bh(bh); | |
1555 | *last_eb_bh = bh; | |
c3afcbb3 | 1556 | goto out; |
328d5752 | 1557 | } |
c3afcbb3 MF |
1558 | } |
1559 | ||
1560 | /* call ocfs2_add_branch to add the final part of the tree with | |
1561 | * the new data. */ | |
d401dc12 | 1562 | ret = ocfs2_add_branch(handle, et, bh, last_eb_bh, |
c3afcbb3 MF |
1563 | meta_ac); |
1564 | if (ret < 0) { | |
1565 | mlog_errno(ret); | |
1566 | goto out; | |
1567 | } | |
1568 | ||
1569 | out: | |
1570 | if (final_depth) | |
1571 | *final_depth = depth; | |
1572 | brelse(bh); | |
1573 | return ret; | |
1574 | } | |
1575 | ||
dcd0538f MF |
1576 | /* |
1577 | * This function will discard the rightmost extent record. | |
1578 | */ | |
1579 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1580 | { | |
1581 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1582 | int count = le16_to_cpu(el->l_count); | |
1583 | unsigned int num_bytes; | |
1584 | ||
1585 | BUG_ON(!next_free); | |
1586 | /* This will cause us to go off the end of our extent list. */ | |
1587 | BUG_ON(next_free >= count); | |
1588 | ||
1589 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1590 | ||
1591 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1592 | } | |
1593 | ||
1594 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1595 | struct ocfs2_extent_rec *insert_rec) | |
1596 | { | |
1597 | int i, insert_index, next_free, has_empty, num_bytes; | |
1598 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1599 | struct ocfs2_extent_rec *rec; | |
1600 | ||
1601 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1602 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1603 | ||
1604 | BUG_ON(!next_free); | |
1605 | ||
1606 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1607 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1608 | |
1609 | /* | |
1610 | * The easiest way to approach this is to just remove the | |
1611 | * empty extent and temporarily decrement next_free. | |
1612 | */ | |
1613 | if (has_empty) { | |
1614 | /* | |
1615 | * If next_free was 1 (only an empty extent), this | |
1616 | * loop won't execute, which is fine. We still want | |
1617 | * the decrement above to happen. | |
1618 | */ | |
1619 | for(i = 0; i < (next_free - 1); i++) | |
1620 | el->l_recs[i] = el->l_recs[i+1]; | |
1621 | ||
1622 | next_free--; | |
1623 | } | |
1624 | ||
1625 | /* | |
1626 | * Figure out what the new record index should be. | |
1627 | */ | |
1628 | for(i = 0; i < next_free; i++) { | |
1629 | rec = &el->l_recs[i]; | |
1630 | ||
1631 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1632 | break; | |
1633 | } | |
1634 | insert_index = i; | |
1635 | ||
a09d09b8 TM |
1636 | trace_ocfs2_rotate_leaf(insert_cpos, insert_index, |
1637 | has_empty, next_free, | |
1638 | le16_to_cpu(el->l_count)); | |
dcd0538f MF |
1639 | |
1640 | BUG_ON(insert_index < 0); | |
1641 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1642 | BUG_ON(insert_index > next_free); | |
1643 | ||
1644 | /* | |
1645 | * No need to memmove if we're just adding to the tail. | |
1646 | */ | |
1647 | if (insert_index != next_free) { | |
1648 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1649 | ||
1650 | num_bytes = next_free - insert_index; | |
1651 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1652 | memmove(&el->l_recs[insert_index + 1], | |
1653 | &el->l_recs[insert_index], | |
1654 | num_bytes); | |
1655 | } | |
1656 | ||
1657 | /* | |
1658 | * Either we had an empty extent, and need to re-increment or | |
1659 | * there was no empty extent on a non full rightmost leaf node, | |
1660 | * in which case we still need to increment. | |
1661 | */ | |
1662 | next_free++; | |
1663 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1664 | /* | |
1665 | * Make sure none of the math above just messed up our tree. | |
1666 | */ | |
1667 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1668 | ||
1669 | el->l_recs[insert_index] = *insert_rec; | |
1670 | ||
1671 | } | |
1672 | ||
328d5752 MF |
1673 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1674 | { | |
1675 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1676 | ||
1677 | BUG_ON(num_recs == 0); | |
1678 | ||
1679 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1680 | num_recs--; | |
1681 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1682 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1683 | memset(&el->l_recs[num_recs], 0, | |
1684 | sizeof(struct ocfs2_extent_rec)); | |
1685 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1686 | } | |
1687 | } | |
1688 | ||
dcd0538f MF |
1689 | /* |
1690 | * Create an empty extent record . | |
1691 | * | |
1692 | * l_next_free_rec may be updated. | |
1693 | * | |
1694 | * If an empty extent already exists do nothing. | |
1695 | */ | |
1696 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1697 | { | |
1698 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1699 | ||
e48edee2 MF |
1700 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1701 | ||
dcd0538f MF |
1702 | if (next_free == 0) |
1703 | goto set_and_inc; | |
1704 | ||
1705 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1706 | return; | |
1707 | ||
1708 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1709 | "Asked to create an empty extent in a full list:\n" | |
1710 | "count = %u, tree depth = %u", | |
1711 | le16_to_cpu(el->l_count), | |
1712 | le16_to_cpu(el->l_tree_depth)); | |
1713 | ||
1714 | ocfs2_shift_records_right(el); | |
1715 | ||
1716 | set_and_inc: | |
1717 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1718 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1719 | } | |
1720 | ||
1721 | /* | |
1722 | * For a rotation which involves two leaf nodes, the "root node" is | |
1723 | * the lowest level tree node which contains a path to both leafs. This | |
1724 | * resulting set of information can be used to form a complete "subtree" | |
1725 | * | |
1726 | * This function is passed two full paths from the dinode down to a | |
1727 | * pair of adjacent leaves. It's task is to figure out which path | |
1728 | * index contains the subtree root - this can be the root index itself | |
1729 | * in a worst-case rotation. | |
1730 | * | |
1731 | * The array index of the subtree root is passed back. | |
1732 | */ | |
38a04e43 TM |
1733 | int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et, |
1734 | struct ocfs2_path *left, | |
1735 | struct ocfs2_path *right) | |
dcd0538f MF |
1736 | { |
1737 | int i = 0; | |
1738 | ||
1739 | /* | |
1740 | * Check that the caller passed in two paths from the same tree. | |
1741 | */ | |
1742 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1743 | ||
1744 | do { | |
1745 | i++; | |
1746 | ||
1747 | /* | |
1748 | * The caller didn't pass two adjacent paths. | |
1749 | */ | |
1750 | mlog_bug_on_msg(i > left->p_tree_depth, | |
7dc02805 | 1751 | "Owner %llu, left depth %u, right depth %u\n" |
dcd0538f | 1752 | "left leaf blk %llu, right leaf blk %llu\n", |
7dc02805 JB |
1753 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
1754 | left->p_tree_depth, right->p_tree_depth, | |
dcd0538f MF |
1755 | (unsigned long long)path_leaf_bh(left)->b_blocknr, |
1756 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1757 | } while (left->p_node[i].bh->b_blocknr == | |
1758 | right->p_node[i].bh->b_blocknr); | |
1759 | ||
1760 | return i - 1; | |
1761 | } | |
1762 | ||
1763 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1764 | ||
1765 | /* | |
1766 | * Traverse a btree path in search of cpos, starting at root_el. | |
1767 | * | |
1768 | * This code can be called with a cpos larger than the tree, in which | |
1769 | * case it will return the rightmost path. | |
1770 | */ | |
facdb77f | 1771 | static int __ocfs2_find_path(struct ocfs2_caching_info *ci, |
dcd0538f MF |
1772 | struct ocfs2_extent_list *root_el, u32 cpos, |
1773 | path_insert_t *func, void *data) | |
1774 | { | |
1775 | int i, ret = 0; | |
1776 | u32 range; | |
1777 | u64 blkno; | |
1778 | struct buffer_head *bh = NULL; | |
1779 | struct ocfs2_extent_block *eb; | |
1780 | struct ocfs2_extent_list *el; | |
1781 | struct ocfs2_extent_rec *rec; | |
dcd0538f MF |
1782 | |
1783 | el = root_el; | |
1784 | while (el->l_tree_depth) { | |
1785 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
facdb77f | 1786 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
7ecef14a | 1787 | "Owner %llu has empty extent list at depth %u\n", |
facdb77f | 1788 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1789 | le16_to_cpu(el->l_tree_depth)); |
1790 | ret = -EROFS; | |
1791 | goto out; | |
1792 | ||
1793 | } | |
1794 | ||
1795 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1796 | rec = &el->l_recs[i]; | |
1797 | ||
1798 | /* | |
1799 | * In the case that cpos is off the allocation | |
1800 | * tree, this should just wind up returning the | |
1801 | * rightmost record. | |
1802 | */ | |
1803 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1804 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1805 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1806 | break; | |
1807 | } | |
1808 | ||
1809 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1810 | if (blkno == 0) { | |
facdb77f | 1811 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
7ecef14a | 1812 | "Owner %llu has bad blkno in extent list at depth %u (index %d)\n", |
facdb77f | 1813 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1814 | le16_to_cpu(el->l_tree_depth), i); |
1815 | ret = -EROFS; | |
1816 | goto out; | |
1817 | } | |
1818 | ||
1819 | brelse(bh); | |
1820 | bh = NULL; | |
facdb77f | 1821 | ret = ocfs2_read_extent_block(ci, blkno, &bh); |
dcd0538f MF |
1822 | if (ret) { |
1823 | mlog_errno(ret); | |
1824 | goto out; | |
1825 | } | |
1826 | ||
1827 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1828 | el = &eb->h_list; | |
dcd0538f MF |
1829 | |
1830 | if (le16_to_cpu(el->l_next_free_rec) > | |
1831 | le16_to_cpu(el->l_count)) { | |
facdb77f | 1832 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
7ecef14a | 1833 | "Owner %llu has bad count in extent list at block %llu (next free=%u, count=%u)\n", |
facdb77f | 1834 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1835 | (unsigned long long)bh->b_blocknr, |
1836 | le16_to_cpu(el->l_next_free_rec), | |
1837 | le16_to_cpu(el->l_count)); | |
1838 | ret = -EROFS; | |
1839 | goto out; | |
1840 | } | |
1841 | ||
1842 | if (func) | |
1843 | func(data, bh); | |
1844 | } | |
1845 | ||
1846 | out: | |
1847 | /* | |
1848 | * Catch any trailing bh that the loop didn't handle. | |
1849 | */ | |
1850 | brelse(bh); | |
1851 | ||
1852 | return ret; | |
1853 | } | |
1854 | ||
1855 | /* | |
1856 | * Given an initialized path (that is, it has a valid root extent | |
1857 | * list), this function will traverse the btree in search of the path | |
1858 | * which would contain cpos. | |
1859 | * | |
1860 | * The path traveled is recorded in the path structure. | |
1861 | * | |
1862 | * Note that this will not do any comparisons on leaf node extent | |
1863 | * records, so it will work fine in the case that we just added a tree | |
1864 | * branch. | |
1865 | */ | |
1866 | struct find_path_data { | |
1867 | int index; | |
1868 | struct ocfs2_path *path; | |
1869 | }; | |
1870 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1871 | { | |
1872 | struct find_path_data *fp = data; | |
1873 | ||
1874 | get_bh(bh); | |
1875 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1876 | fp->index++; | |
1877 | } | |
e2e9f608 TM |
1878 | int ocfs2_find_path(struct ocfs2_caching_info *ci, |
1879 | struct ocfs2_path *path, u32 cpos) | |
dcd0538f MF |
1880 | { |
1881 | struct find_path_data data; | |
1882 | ||
1883 | data.index = 1; | |
1884 | data.path = path; | |
facdb77f | 1885 | return __ocfs2_find_path(ci, path_root_el(path), cpos, |
dcd0538f MF |
1886 | find_path_ins, &data); |
1887 | } | |
1888 | ||
1889 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1890 | { | |
1891 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1892 | struct ocfs2_extent_list *el = &eb->h_list; | |
1893 | struct buffer_head **ret = data; | |
1894 | ||
1895 | /* We want to retain only the leaf block. */ | |
1896 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1897 | get_bh(bh); | |
1898 | *ret = bh; | |
1899 | } | |
1900 | } | |
1901 | /* | |
1902 | * Find the leaf block in the tree which would contain cpos. No | |
1903 | * checking of the actual leaf is done. | |
1904 | * | |
1905 | * Some paths want to call this instead of allocating a path structure | |
1906 | * and calling ocfs2_find_path(). | |
1907 | * | |
1908 | * This function doesn't handle non btree extent lists. | |
1909 | */ | |
facdb77f JB |
1910 | int ocfs2_find_leaf(struct ocfs2_caching_info *ci, |
1911 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1912 | struct buffer_head **leaf_bh) | |
dcd0538f MF |
1913 | { |
1914 | int ret; | |
1915 | struct buffer_head *bh = NULL; | |
1916 | ||
facdb77f | 1917 | ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh); |
dcd0538f MF |
1918 | if (ret) { |
1919 | mlog_errno(ret); | |
1920 | goto out; | |
1921 | } | |
1922 | ||
1923 | *leaf_bh = bh; | |
1924 | out: | |
1925 | return ret; | |
1926 | } | |
1927 | ||
1928 | /* | |
1929 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1930 | * | |
1931 | * Basically, we've moved stuff around at the bottom of the tree and | |
1932 | * we need to fix up the extent records above the changes to reflect | |
1933 | * the new changes. | |
1934 | * | |
1935 | * left_rec: the record on the left. | |
1936 | * left_child_el: is the child list pointed to by left_rec | |
1937 | * right_rec: the record to the right of left_rec | |
1938 | * right_child_el: is the child list pointed to by right_rec | |
1939 | * | |
1940 | * By definition, this only works on interior nodes. | |
1941 | */ | |
1942 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1943 | struct ocfs2_extent_list *left_child_el, | |
1944 | struct ocfs2_extent_rec *right_rec, | |
1945 | struct ocfs2_extent_list *right_child_el) | |
1946 | { | |
1947 | u32 left_clusters, right_end; | |
1948 | ||
1949 | /* | |
1950 | * Interior nodes never have holes. Their cpos is the cpos of | |
1951 | * the leftmost record in their child list. Their cluster | |
1952 | * count covers the full theoretical range of their child list | |
1953 | * - the range between their cpos and the cpos of the record | |
1954 | * immediately to their right. | |
1955 | */ | |
1956 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
82e12644 TM |
1957 | if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) { |
1958 | BUG_ON(right_child_el->l_tree_depth); | |
328d5752 MF |
1959 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); |
1960 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1961 | } | |
dcd0538f | 1962 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 1963 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
1964 | |
1965 | /* | |
1966 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 1967 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
1968 | * updating e_cpos to keep the same highest cluster count. |
1969 | */ | |
1970 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1971 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
1972 | |
1973 | right_rec->e_cpos = left_rec->e_cpos; | |
1974 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
1975 | ||
1976 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1977 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
1978 | } |
1979 | ||
1980 | /* | |
1981 | * Adjust the adjacent root node records involved in a | |
1982 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
1983 | * find it's index in the root list. | |
1984 | */ | |
1985 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
1986 | struct ocfs2_extent_list *left_el, | |
1987 | struct ocfs2_extent_list *right_el, | |
1988 | u64 left_el_blkno) | |
1989 | { | |
1990 | int i; | |
1991 | ||
1992 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
1993 | le16_to_cpu(left_el->l_tree_depth)); | |
1994 | ||
1995 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
1996 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
1997 | break; | |
1998 | } | |
1999 | ||
2000 | /* | |
2001 | * The path walking code should have never returned a root and | |
2002 | * two paths which are not adjacent. | |
2003 | */ | |
2004 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
2005 | ||
2006 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
2007 | &root_el->l_recs[i + 1], right_el); | |
2008 | } | |
2009 | ||
2010 | /* | |
2011 | * We've changed a leaf block (in right_path) and need to reflect that | |
2012 | * change back up the subtree. | |
2013 | * | |
2014 | * This happens in multiple places: | |
2015 | * - When we've moved an extent record from the left path leaf to the right | |
2016 | * path leaf to make room for an empty extent in the left path leaf. | |
2017 | * - When our insert into the right path leaf is at the leftmost edge | |
2018 | * and requires an update of the path immediately to it's left. This | |
2019 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
2020 | * - When we've adjusted the last extent record in the left path leaf and the |
2021 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f | 2022 | */ |
4619c73e | 2023 | static void ocfs2_complete_edge_insert(handle_t *handle, |
dcd0538f MF |
2024 | struct ocfs2_path *left_path, |
2025 | struct ocfs2_path *right_path, | |
2026 | int subtree_index) | |
2027 | { | |
ec20cec7 | 2028 | int i, idx; |
dcd0538f MF |
2029 | struct ocfs2_extent_list *el, *left_el, *right_el; |
2030 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
2031 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2032 | ||
2033 | /* | |
2034 | * Update the counts and position values within all the | |
2035 | * interior nodes to reflect the leaf rotation we just did. | |
2036 | * | |
2037 | * The root node is handled below the loop. | |
2038 | * | |
2039 | * We begin the loop with right_el and left_el pointing to the | |
2040 | * leaf lists and work our way up. | |
2041 | * | |
2042 | * NOTE: within this loop, left_el and right_el always refer | |
2043 | * to the *child* lists. | |
2044 | */ | |
2045 | left_el = path_leaf_el(left_path); | |
2046 | right_el = path_leaf_el(right_path); | |
2047 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
a09d09b8 | 2048 | trace_ocfs2_complete_edge_insert(i); |
dcd0538f MF |
2049 | |
2050 | /* | |
2051 | * One nice property of knowing that all of these | |
2052 | * nodes are below the root is that we only deal with | |
2053 | * the leftmost right node record and the rightmost | |
2054 | * left node record. | |
2055 | */ | |
2056 | el = left_path->p_node[i].el; | |
2057 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2058 | left_rec = &el->l_recs[idx]; | |
2059 | ||
2060 | el = right_path->p_node[i].el; | |
2061 | right_rec = &el->l_recs[0]; | |
2062 | ||
2063 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
2064 | right_el); | |
2065 | ||
ec20cec7 JB |
2066 | ocfs2_journal_dirty(handle, left_path->p_node[i].bh); |
2067 | ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
dcd0538f MF |
2068 | |
2069 | /* | |
2070 | * Setup our list pointers now so that the current | |
2071 | * parents become children in the next iteration. | |
2072 | */ | |
2073 | left_el = left_path->p_node[i].el; | |
2074 | right_el = right_path->p_node[i].el; | |
2075 | } | |
2076 | ||
2077 | /* | |
2078 | * At the root node, adjust the two adjacent records which | |
2079 | * begin our path to the leaves. | |
2080 | */ | |
2081 | ||
2082 | el = left_path->p_node[subtree_index].el; | |
2083 | left_el = left_path->p_node[subtree_index + 1].el; | |
2084 | right_el = right_path->p_node[subtree_index + 1].el; | |
2085 | ||
2086 | ocfs2_adjust_root_records(el, left_el, right_el, | |
2087 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
2088 | ||
2089 | root_bh = left_path->p_node[subtree_index].bh; | |
2090 | ||
ec20cec7 | 2091 | ocfs2_journal_dirty(handle, root_bh); |
dcd0538f MF |
2092 | } |
2093 | ||
5c601aba JB |
2094 | static int ocfs2_rotate_subtree_right(handle_t *handle, |
2095 | struct ocfs2_extent_tree *et, | |
dcd0538f MF |
2096 | struct ocfs2_path *left_path, |
2097 | struct ocfs2_path *right_path, | |
2098 | int subtree_index) | |
2099 | { | |
2100 | int ret, i; | |
2101 | struct buffer_head *right_leaf_bh; | |
2102 | struct buffer_head *left_leaf_bh = NULL; | |
2103 | struct buffer_head *root_bh; | |
2104 | struct ocfs2_extent_list *right_el, *left_el; | |
2105 | struct ocfs2_extent_rec move_rec; | |
2106 | ||
2107 | left_leaf_bh = path_leaf_bh(left_path); | |
2108 | left_el = path_leaf_el(left_path); | |
2109 | ||
2110 | if (left_el->l_next_free_rec != left_el->l_count) { | |
5c601aba | 2111 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
7ecef14a | 2112 | "Inode %llu has non-full interior leaf node %llu (next free = %u)\n", |
5c601aba | 2113 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2114 | (unsigned long long)left_leaf_bh->b_blocknr, |
2115 | le16_to_cpu(left_el->l_next_free_rec)); | |
2116 | return -EROFS; | |
2117 | } | |
2118 | ||
2119 | /* | |
2120 | * This extent block may already have an empty record, so we | |
2121 | * return early if so. | |
2122 | */ | |
2123 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
2124 | return 0; | |
2125 | ||
2126 | root_bh = left_path->p_node[subtree_index].bh; | |
2127 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2128 | ||
5c601aba | 2129 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2130 | subtree_index); |
dcd0538f MF |
2131 | if (ret) { |
2132 | mlog_errno(ret); | |
2133 | goto out; | |
2134 | } | |
2135 | ||
2136 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
5c601aba | 2137 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2138 | right_path, i); |
dcd0538f MF |
2139 | if (ret) { |
2140 | mlog_errno(ret); | |
2141 | goto out; | |
2142 | } | |
2143 | ||
5c601aba | 2144 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2145 | left_path, i); |
dcd0538f MF |
2146 | if (ret) { |
2147 | mlog_errno(ret); | |
2148 | goto out; | |
2149 | } | |
2150 | } | |
2151 | ||
2152 | right_leaf_bh = path_leaf_bh(right_path); | |
2153 | right_el = path_leaf_el(right_path); | |
2154 | ||
2155 | /* This is a code error, not a disk corruption. */ | |
2156 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
2157 | "because rightmost leaf block %llu is empty\n", | |
5c601aba | 2158 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2159 | (unsigned long long)right_leaf_bh->b_blocknr); |
2160 | ||
2161 | ocfs2_create_empty_extent(right_el); | |
2162 | ||
ec20cec7 | 2163 | ocfs2_journal_dirty(handle, right_leaf_bh); |
dcd0538f MF |
2164 | |
2165 | /* Do the copy now. */ | |
2166 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2167 | move_rec = left_el->l_recs[i]; | |
2168 | right_el->l_recs[0] = move_rec; | |
2169 | ||
2170 | /* | |
2171 | * Clear out the record we just copied and shift everything | |
2172 | * over, leaving an empty extent in the left leaf. | |
2173 | * | |
2174 | * We temporarily subtract from next_free_rec so that the | |
2175 | * shift will lose the tail record (which is now defunct). | |
2176 | */ | |
2177 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
2178 | ocfs2_shift_records_right(left_el); | |
2179 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2180 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
2181 | ||
ec20cec7 | 2182 | ocfs2_journal_dirty(handle, left_leaf_bh); |
dcd0538f | 2183 | |
4619c73e JB |
2184 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
2185 | subtree_index); | |
dcd0538f MF |
2186 | |
2187 | out: | |
2188 | return ret; | |
2189 | } | |
2190 | ||
2191 | /* | |
2192 | * Given a full path, determine what cpos value would return us a path | |
2193 | * containing the leaf immediately to the left of the current one. | |
2194 | * | |
2195 | * Will return zero if the path passed in is already the leftmost path. | |
2196 | */ | |
ee149a7c TY |
2197 | int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, |
2198 | struct ocfs2_path *path, u32 *cpos) | |
dcd0538f MF |
2199 | { |
2200 | int i, j, ret = 0; | |
2201 | u64 blkno; | |
2202 | struct ocfs2_extent_list *el; | |
2203 | ||
e48edee2 MF |
2204 | BUG_ON(path->p_tree_depth == 0); |
2205 | ||
dcd0538f MF |
2206 | *cpos = 0; |
2207 | ||
2208 | blkno = path_leaf_bh(path)->b_blocknr; | |
2209 | ||
2210 | /* Start at the tree node just above the leaf and work our way up. */ | |
2211 | i = path->p_tree_depth - 1; | |
2212 | while (i >= 0) { | |
2213 | el = path->p_node[i].el; | |
2214 | ||
2215 | /* | |
2216 | * Find the extent record just before the one in our | |
2217 | * path. | |
2218 | */ | |
2219 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2220 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2221 | if (j == 0) { | |
2222 | if (i == 0) { | |
2223 | /* | |
2224 | * We've determined that the | |
2225 | * path specified is already | |
2226 | * the leftmost one - return a | |
2227 | * cpos of zero. | |
2228 | */ | |
2229 | goto out; | |
2230 | } | |
2231 | /* | |
2232 | * The leftmost record points to our | |
2233 | * leaf - we need to travel up the | |
2234 | * tree one level. | |
2235 | */ | |
2236 | goto next_node; | |
2237 | } | |
2238 | ||
2239 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
2240 | *cpos = *cpos + ocfs2_rec_clusters(el, |
2241 | &el->l_recs[j - 1]); | |
2242 | *cpos = *cpos - 1; | |
dcd0538f MF |
2243 | goto out; |
2244 | } | |
2245 | } | |
2246 | ||
2247 | /* | |
2248 | * If we got here, we never found a valid node where | |
2249 | * the tree indicated one should be. | |
2250 | */ | |
7ecef14a | 2251 | ocfs2_error(sb, "Invalid extent tree at extent block %llu\n", |
dcd0538f MF |
2252 | (unsigned long long)blkno); |
2253 | ret = -EROFS; | |
2254 | goto out; | |
2255 | ||
2256 | next_node: | |
2257 | blkno = path->p_node[i].bh->b_blocknr; | |
2258 | i--; | |
2259 | } | |
2260 | ||
2261 | out: | |
2262 | return ret; | |
2263 | } | |
2264 | ||
328d5752 MF |
2265 | /* |
2266 | * Extend the transaction by enough credits to complete the rotation, | |
2267 | * and still leave at least the original number of credits allocated | |
2268 | * to this transaction. | |
2269 | */ | |
dcd0538f | 2270 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 2271 | int op_credits, |
dcd0538f MF |
2272 | struct ocfs2_path *path) |
2273 | { | |
c901fb00 | 2274 | int ret = 0; |
328d5752 | 2275 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f | 2276 | |
c901fb00 | 2277 | if (handle->h_buffer_credits < credits) |
c18b812d TM |
2278 | ret = ocfs2_extend_trans(handle, |
2279 | credits - handle->h_buffer_credits); | |
dcd0538f | 2280 | |
c901fb00 | 2281 | return ret; |
dcd0538f MF |
2282 | } |
2283 | ||
2284 | /* | |
2285 | * Trap the case where we're inserting into the theoretical range past | |
2286 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
2287 | * whose cpos is less than ours into the right leaf. | |
2288 | * | |
2289 | * It's only necessary to look at the rightmost record of the left | |
2290 | * leaf because the logic that calls us should ensure that the | |
2291 | * theoretical ranges in the path components above the leaves are | |
2292 | * correct. | |
2293 | */ | |
2294 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
2295 | u32 insert_cpos) | |
2296 | { | |
2297 | struct ocfs2_extent_list *left_el; | |
2298 | struct ocfs2_extent_rec *rec; | |
2299 | int next_free; | |
2300 | ||
2301 | left_el = path_leaf_el(left_path); | |
2302 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
2303 | rec = &left_el->l_recs[next_free - 1]; | |
2304 | ||
2305 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
2306 | return 1; | |
2307 | return 0; | |
2308 | } | |
2309 | ||
328d5752 MF |
2310 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
2311 | { | |
2312 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
2313 | unsigned int range; | |
2314 | struct ocfs2_extent_rec *rec; | |
2315 | ||
2316 | if (next_free == 0) | |
2317 | return 0; | |
2318 | ||
2319 | rec = &el->l_recs[0]; | |
2320 | if (ocfs2_is_empty_extent(rec)) { | |
2321 | /* Empty list. */ | |
2322 | if (next_free == 1) | |
2323 | return 0; | |
2324 | rec = &el->l_recs[1]; | |
2325 | } | |
2326 | ||
2327 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2328 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2329 | return 1; | |
2330 | return 0; | |
2331 | } | |
2332 | ||
dcd0538f MF |
2333 | /* |
2334 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2335 | * | |
2336 | * The path to the rightmost leaf should be passed in. | |
2337 | * | |
2338 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2339 | * | |
af901ca1 | 2340 | * Upon successful return from this function: |
dcd0538f MF |
2341 | * |
2342 | * - The 'right_path' array will contain a path to the leaf block | |
2343 | * whose range contains e_cpos. | |
2344 | * - That leaf block will have a single empty extent in list index 0. | |
2345 | * - In the case that the rotation requires a post-insert update, | |
2346 | * *ret_left_path will contain a valid path which can be passed to | |
2347 | * ocfs2_insert_path(). | |
2348 | */ | |
1bbf0b8d | 2349 | static int ocfs2_rotate_tree_right(handle_t *handle, |
5c601aba | 2350 | struct ocfs2_extent_tree *et, |
328d5752 | 2351 | enum ocfs2_split_type split, |
dcd0538f MF |
2352 | u32 insert_cpos, |
2353 | struct ocfs2_path *right_path, | |
2354 | struct ocfs2_path **ret_left_path) | |
2355 | { | |
328d5752 | 2356 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2357 | u32 cpos; |
2358 | struct ocfs2_path *left_path = NULL; | |
5c601aba | 2359 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
dcd0538f MF |
2360 | |
2361 | *ret_left_path = NULL; | |
2362 | ||
ffdd7a54 | 2363 | left_path = ocfs2_new_path_from_path(right_path); |
dcd0538f MF |
2364 | if (!left_path) { |
2365 | ret = -ENOMEM; | |
2366 | mlog_errno(ret); | |
2367 | goto out; | |
2368 | } | |
2369 | ||
5c601aba | 2370 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2371 | if (ret) { |
2372 | mlog_errno(ret); | |
2373 | goto out; | |
2374 | } | |
2375 | ||
a09d09b8 TM |
2376 | trace_ocfs2_rotate_tree_right( |
2377 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
2378 | insert_cpos, cpos); | |
dcd0538f MF |
2379 | |
2380 | /* | |
2381 | * What we want to do here is: | |
2382 | * | |
2383 | * 1) Start with the rightmost path. | |
2384 | * | |
2385 | * 2) Determine a path to the leaf block directly to the left | |
2386 | * of that leaf. | |
2387 | * | |
2388 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2389 | * which contains a path to both leaves. | |
2390 | * | |
2391 | * 4) Rotate the subtree. | |
2392 | * | |
2393 | * 5) Find the next subtree by considering the left path to be | |
2394 | * the new right path. | |
2395 | * | |
2396 | * The check at the top of this while loop also accepts | |
2397 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2398 | * value to get us the left path - insert_cpos might very well | |
2399 | * be filling that hole. | |
2400 | * | |
2401 | * Stop at a cpos of '0' because we either started at the | |
2402 | * leftmost branch (i.e., a tree with one branch and a | |
2403 | * rotation inside of it), or we've gone as far as we can in | |
2404 | * rotating subtrees. | |
2405 | */ | |
2406 | while (cpos && insert_cpos <= cpos) { | |
a09d09b8 TM |
2407 | trace_ocfs2_rotate_tree_right( |
2408 | (unsigned long long) | |
2409 | ocfs2_metadata_cache_owner(et->et_ci), | |
2410 | insert_cpos, cpos); | |
dcd0538f | 2411 | |
5c601aba | 2412 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
dcd0538f MF |
2413 | if (ret) { |
2414 | mlog_errno(ret); | |
2415 | goto out; | |
2416 | } | |
2417 | ||
2418 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2419 | path_leaf_bh(right_path), | |
5c601aba | 2420 | "Owner %llu: error during insert of %u " |
dcd0538f MF |
2421 | "(left path cpos %u) results in two identical " |
2422 | "paths ending at %llu\n", | |
5c601aba JB |
2423 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
2424 | insert_cpos, cpos, | |
dcd0538f MF |
2425 | (unsigned long long) |
2426 | path_leaf_bh(left_path)->b_blocknr); | |
2427 | ||
328d5752 MF |
2428 | if (split == SPLIT_NONE && |
2429 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2430 | insert_cpos)) { |
dcd0538f MF |
2431 | |
2432 | /* | |
2433 | * We've rotated the tree as much as we | |
2434 | * should. The rest is up to | |
2435 | * ocfs2_insert_path() to complete, after the | |
2436 | * record insertion. We indicate this | |
2437 | * situation by returning the left path. | |
2438 | * | |
2439 | * The reason we don't adjust the records here | |
2440 | * before the record insert is that an error | |
2441 | * later might break the rule where a parent | |
2442 | * record e_cpos will reflect the actual | |
2443 | * e_cpos of the 1st nonempty record of the | |
2444 | * child list. | |
2445 | */ | |
2446 | *ret_left_path = left_path; | |
2447 | goto out_ret_path; | |
2448 | } | |
2449 | ||
7dc02805 | 2450 | start = ocfs2_find_subtree_root(et, left_path, right_path); |
dcd0538f | 2451 | |
a09d09b8 TM |
2452 | trace_ocfs2_rotate_subtree(start, |
2453 | (unsigned long long) | |
2454 | right_path->p_node[start].bh->b_blocknr, | |
2455 | right_path->p_tree_depth); | |
dcd0538f MF |
2456 | |
2457 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2458 | orig_credits, right_path); |
dcd0538f MF |
2459 | if (ret) { |
2460 | mlog_errno(ret); | |
2461 | goto out; | |
2462 | } | |
2463 | ||
5c601aba | 2464 | ret = ocfs2_rotate_subtree_right(handle, et, left_path, |
dcd0538f MF |
2465 | right_path, start); |
2466 | if (ret) { | |
2467 | mlog_errno(ret); | |
2468 | goto out; | |
2469 | } | |
2470 | ||
328d5752 MF |
2471 | if (split != SPLIT_NONE && |
2472 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2473 | insert_cpos)) { | |
2474 | /* | |
2475 | * A rotate moves the rightmost left leaf | |
2476 | * record over to the leftmost right leaf | |
2477 | * slot. If we're doing an extent split | |
2478 | * instead of a real insert, then we have to | |
2479 | * check that the extent to be split wasn't | |
2480 | * just moved over. If it was, then we can | |
2481 | * exit here, passing left_path back - | |
2482 | * ocfs2_split_extent() is smart enough to | |
2483 | * search both leaves. | |
2484 | */ | |
2485 | *ret_left_path = left_path; | |
2486 | goto out_ret_path; | |
2487 | } | |
2488 | ||
dcd0538f MF |
2489 | /* |
2490 | * There is no need to re-read the next right path | |
2491 | * as we know that it'll be our current left | |
2492 | * path. Optimize by copying values instead. | |
2493 | */ | |
2494 | ocfs2_mv_path(right_path, left_path); | |
2495 | ||
5c601aba | 2496 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2497 | if (ret) { |
2498 | mlog_errno(ret); | |
2499 | goto out; | |
2500 | } | |
2501 | } | |
2502 | ||
2503 | out: | |
2504 | ocfs2_free_path(left_path); | |
2505 | ||
2506 | out_ret_path: | |
2507 | return ret; | |
2508 | } | |
2509 | ||
09106bae JB |
2510 | static int ocfs2_update_edge_lengths(handle_t *handle, |
2511 | struct ocfs2_extent_tree *et, | |
3c5e1068 | 2512 | int subtree_index, struct ocfs2_path *path) |
dcd0538f | 2513 | { |
3c5e1068 | 2514 | int i, idx, ret; |
dcd0538f | 2515 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2516 | struct ocfs2_extent_list *el; |
2517 | struct ocfs2_extent_block *eb; | |
2518 | u32 range; | |
dcd0538f | 2519 | |
09106bae | 2520 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
3c5e1068 TM |
2521 | if (ret) { |
2522 | mlog_errno(ret); | |
2523 | goto out; | |
2524 | } | |
2525 | ||
328d5752 MF |
2526 | /* Path should always be rightmost. */ |
2527 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2528 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2529 | |
328d5752 MF |
2530 | el = &eb->h_list; |
2531 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2532 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2533 | rec = &el->l_recs[idx]; | |
2534 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2535 | |
328d5752 MF |
2536 | for (i = 0; i < path->p_tree_depth; i++) { |
2537 | el = path->p_node[i].el; | |
2538 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2539 | rec = &el->l_recs[idx]; | |
dcd0538f | 2540 | |
328d5752 MF |
2541 | rec->e_int_clusters = cpu_to_le32(range); |
2542 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2543 | |
328d5752 | 2544 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2545 | } |
3c5e1068 TM |
2546 | out: |
2547 | return ret; | |
dcd0538f MF |
2548 | } |
2549 | ||
6641b0ce JB |
2550 | static void ocfs2_unlink_path(handle_t *handle, |
2551 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2552 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2553 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2554 | { |
328d5752 MF |
2555 | int ret, i; |
2556 | struct ocfs2_extent_block *eb; | |
2557 | struct ocfs2_extent_list *el; | |
2558 | struct buffer_head *bh; | |
2559 | ||
2560 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2561 | bh = path->p_node[i].bh; | |
2562 | ||
2563 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2564 | /* | |
2565 | * Not all nodes might have had their final count | |
2566 | * decremented by the caller - handle this here. | |
2567 | */ | |
2568 | el = &eb->h_list; | |
2569 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2570 | mlog(ML_ERROR, | |
2571 | "Inode %llu, attempted to remove extent block " | |
2572 | "%llu with %u records\n", | |
6641b0ce | 2573 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
2574 | (unsigned long long)le64_to_cpu(eb->h_blkno), |
2575 | le16_to_cpu(el->l_next_free_rec)); | |
2576 | ||
2577 | ocfs2_journal_dirty(handle, bh); | |
6641b0ce | 2578 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 MF |
2579 | continue; |
2580 | } | |
2581 | ||
2582 | el->l_next_free_rec = 0; | |
2583 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2584 | ||
2585 | ocfs2_journal_dirty(handle, bh); | |
2586 | ||
2587 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2588 | if (ret) | |
2589 | mlog_errno(ret); | |
2590 | ||
6641b0ce | 2591 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 | 2592 | } |
dcd0538f MF |
2593 | } |
2594 | ||
6641b0ce JB |
2595 | static void ocfs2_unlink_subtree(handle_t *handle, |
2596 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2597 | struct ocfs2_path *left_path, |
2598 | struct ocfs2_path *right_path, | |
2599 | int subtree_index, | |
2600 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2601 | { |
328d5752 MF |
2602 | int i; |
2603 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2604 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2605 | struct ocfs2_extent_list *el; |
328d5752 | 2606 | struct ocfs2_extent_block *eb; |
dcd0538f | 2607 | |
328d5752 | 2608 | el = path_leaf_el(left_path); |
dcd0538f | 2609 | |
328d5752 | 2610 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2611 | |
328d5752 MF |
2612 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2613 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2614 | break; | |
dcd0538f | 2615 | |
328d5752 | 2616 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2617 | |
328d5752 MF |
2618 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2619 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2620 | ||
2621 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2622 | eb->h_next_leaf_blk = 0; | |
2623 | ||
2624 | ocfs2_journal_dirty(handle, root_bh); | |
2625 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2626 | ||
6641b0ce | 2627 | ocfs2_unlink_path(handle, et, dealloc, right_path, |
328d5752 MF |
2628 | subtree_index + 1); |
2629 | } | |
2630 | ||
1e2dd63f JB |
2631 | static int ocfs2_rotate_subtree_left(handle_t *handle, |
2632 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2633 | struct ocfs2_path *left_path, |
2634 | struct ocfs2_path *right_path, | |
2635 | int subtree_index, | |
2636 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
1e2dd63f | 2637 | int *deleted) |
328d5752 MF |
2638 | { |
2639 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2640 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2641 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2642 | struct ocfs2_extent_block *eb; | |
2643 | ||
2644 | *deleted = 0; | |
2645 | ||
2646 | right_leaf_el = path_leaf_el(right_path); | |
2647 | left_leaf_el = path_leaf_el(left_path); | |
2648 | root_bh = left_path->p_node[subtree_index].bh; | |
2649 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2650 | ||
2651 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2652 | return 0; | |
dcd0538f | 2653 | |
328d5752 MF |
2654 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2655 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2656 | /* |
328d5752 MF |
2657 | * It's legal for us to proceed if the right leaf is |
2658 | * the rightmost one and it has an empty extent. There | |
2659 | * are two cases to handle - whether the leaf will be | |
2660 | * empty after removal or not. If the leaf isn't empty | |
2661 | * then just remove the empty extent up front. The | |
2662 | * next block will handle empty leaves by flagging | |
2663 | * them for unlink. | |
2664 | * | |
2665 | * Non rightmost leaves will throw -EAGAIN and the | |
2666 | * caller can manually move the subtree and retry. | |
dcd0538f | 2667 | */ |
dcd0538f | 2668 | |
328d5752 MF |
2669 | if (eb->h_next_leaf_blk != 0ULL) |
2670 | return -EAGAIN; | |
2671 | ||
2672 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
1e2dd63f | 2673 | ret = ocfs2_journal_access_eb(handle, et->et_ci, |
13723d00 JB |
2674 | path_leaf_bh(right_path), |
2675 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2676 | if (ret) { |
2677 | mlog_errno(ret); | |
2678 | goto out; | |
2679 | } | |
2680 | ||
328d5752 MF |
2681 | ocfs2_remove_empty_extent(right_leaf_el); |
2682 | } else | |
2683 | right_has_empty = 1; | |
dcd0538f MF |
2684 | } |
2685 | ||
328d5752 MF |
2686 | if (eb->h_next_leaf_blk == 0ULL && |
2687 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2688 | /* | |
2689 | * We have to update i_last_eb_blk during the meta | |
2690 | * data delete. | |
2691 | */ | |
d9a0a1f8 | 2692 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 2693 | OCFS2_JOURNAL_ACCESS_WRITE); |
328d5752 MF |
2694 | if (ret) { |
2695 | mlog_errno(ret); | |
2696 | goto out; | |
2697 | } | |
2698 | ||
2699 | del_right_subtree = 1; | |
2700 | } | |
2701 | ||
2702 | /* | |
2703 | * Getting here with an empty extent in the right path implies | |
2704 | * that it's the rightmost path and will be deleted. | |
2705 | */ | |
2706 | BUG_ON(right_has_empty && !del_right_subtree); | |
2707 | ||
1e2dd63f | 2708 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2709 | subtree_index); |
328d5752 MF |
2710 | if (ret) { |
2711 | mlog_errno(ret); | |
2712 | goto out; | |
2713 | } | |
2714 | ||
2715 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1e2dd63f | 2716 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2717 | right_path, i); |
328d5752 MF |
2718 | if (ret) { |
2719 | mlog_errno(ret); | |
2720 | goto out; | |
2721 | } | |
2722 | ||
1e2dd63f | 2723 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2724 | left_path, i); |
328d5752 MF |
2725 | if (ret) { |
2726 | mlog_errno(ret); | |
2727 | goto out; | |
2728 | } | |
2729 | } | |
2730 | ||
2731 | if (!right_has_empty) { | |
2732 | /* | |
2733 | * Only do this if we're moving a real | |
2734 | * record. Otherwise, the action is delayed until | |
2735 | * after removal of the right path in which case we | |
2736 | * can do a simple shift to remove the empty extent. | |
2737 | */ | |
2738 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2739 | memset(&right_leaf_el->l_recs[0], 0, | |
2740 | sizeof(struct ocfs2_extent_rec)); | |
2741 | } | |
2742 | if (eb->h_next_leaf_blk == 0ULL) { | |
2743 | /* | |
2744 | * Move recs over to get rid of empty extent, decrease | |
2745 | * next_free. This is allowed to remove the last | |
2746 | * extent in our leaf (setting l_next_free_rec to | |
2747 | * zero) - the delete code below won't care. | |
2748 | */ | |
2749 | ocfs2_remove_empty_extent(right_leaf_el); | |
2750 | } | |
2751 | ||
ec20cec7 JB |
2752 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); |
2753 | ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
328d5752 MF |
2754 | |
2755 | if (del_right_subtree) { | |
6641b0ce | 2756 | ocfs2_unlink_subtree(handle, et, left_path, right_path, |
328d5752 | 2757 | subtree_index, dealloc); |
09106bae | 2758 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
2759 | left_path); |
2760 | if (ret) { | |
2761 | mlog_errno(ret); | |
2762 | goto out; | |
2763 | } | |
328d5752 MF |
2764 | |
2765 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2766 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2767 | |
2768 | /* | |
2769 | * Removal of the extent in the left leaf was skipped | |
2770 | * above so we could delete the right path | |
2771 | * 1st. | |
2772 | */ | |
2773 | if (right_has_empty) | |
2774 | ocfs2_remove_empty_extent(left_leaf_el); | |
2775 | ||
ec20cec7 | 2776 | ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2777 | |
2778 | *deleted = 1; | |
2779 | } else | |
4619c73e | 2780 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
328d5752 MF |
2781 | subtree_index); |
2782 | ||
2783 | out: | |
2784 | return ret; | |
2785 | } | |
2786 | ||
2787 | /* | |
2788 | * Given a full path, determine what cpos value would return us a path | |
2789 | * containing the leaf immediately to the right of the current one. | |
2790 | * | |
2791 | * Will return zero if the path passed in is already the rightmost path. | |
2792 | * | |
2793 | * This looks similar, but is subtly different to | |
2794 | * ocfs2_find_cpos_for_left_leaf(). | |
2795 | */ | |
38a04e43 TM |
2796 | int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, |
2797 | struct ocfs2_path *path, u32 *cpos) | |
328d5752 MF |
2798 | { |
2799 | int i, j, ret = 0; | |
2800 | u64 blkno; | |
2801 | struct ocfs2_extent_list *el; | |
2802 | ||
2803 | *cpos = 0; | |
2804 | ||
2805 | if (path->p_tree_depth == 0) | |
2806 | return 0; | |
2807 | ||
2808 | blkno = path_leaf_bh(path)->b_blocknr; | |
2809 | ||
2810 | /* Start at the tree node just above the leaf and work our way up. */ | |
2811 | i = path->p_tree_depth - 1; | |
2812 | while (i >= 0) { | |
2813 | int next_free; | |
2814 | ||
2815 | el = path->p_node[i].el; | |
2816 | ||
2817 | /* | |
2818 | * Find the extent record just after the one in our | |
2819 | * path. | |
2820 | */ | |
2821 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2822 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2823 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2824 | if (j == (next_free - 1)) { | |
2825 | if (i == 0) { | |
2826 | /* | |
2827 | * We've determined that the | |
2828 | * path specified is already | |
2829 | * the rightmost one - return a | |
2830 | * cpos of zero. | |
2831 | */ | |
2832 | goto out; | |
2833 | } | |
2834 | /* | |
2835 | * The rightmost record points to our | |
2836 | * leaf - we need to travel up the | |
2837 | * tree one level. | |
2838 | */ | |
2839 | goto next_node; | |
2840 | } | |
2841 | ||
2842 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2843 | goto out; | |
2844 | } | |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * If we got here, we never found a valid node where | |
2849 | * the tree indicated one should be. | |
2850 | */ | |
7ecef14a | 2851 | ocfs2_error(sb, "Invalid extent tree at extent block %llu\n", |
328d5752 MF |
2852 | (unsigned long long)blkno); |
2853 | ret = -EROFS; | |
2854 | goto out; | |
2855 | ||
2856 | next_node: | |
2857 | blkno = path->p_node[i].bh->b_blocknr; | |
2858 | i--; | |
2859 | } | |
2860 | ||
2861 | out: | |
2862 | return ret; | |
2863 | } | |
2864 | ||
70f18c08 JB |
2865 | static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle, |
2866 | struct ocfs2_extent_tree *et, | |
13723d00 | 2867 | struct ocfs2_path *path) |
328d5752 MF |
2868 | { |
2869 | int ret; | |
13723d00 JB |
2870 | struct buffer_head *bh = path_leaf_bh(path); |
2871 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
328d5752 MF |
2872 | |
2873 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2874 | return 0; | |
2875 | ||
70f18c08 | 2876 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
13723d00 | 2877 | path_num_items(path) - 1); |
328d5752 MF |
2878 | if (ret) { |
2879 | mlog_errno(ret); | |
2880 | goto out; | |
2881 | } | |
2882 | ||
2883 | ocfs2_remove_empty_extent(el); | |
ec20cec7 | 2884 | ocfs2_journal_dirty(handle, bh); |
328d5752 MF |
2885 | |
2886 | out: | |
2887 | return ret; | |
2888 | } | |
2889 | ||
e46f74dc JB |
2890 | static int __ocfs2_rotate_tree_left(handle_t *handle, |
2891 | struct ocfs2_extent_tree *et, | |
2892 | int orig_credits, | |
328d5752 MF |
2893 | struct ocfs2_path *path, |
2894 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e46f74dc | 2895 | struct ocfs2_path **empty_extent_path) |
328d5752 MF |
2896 | { |
2897 | int ret, subtree_root, deleted; | |
2898 | u32 right_cpos; | |
2899 | struct ocfs2_path *left_path = NULL; | |
2900 | struct ocfs2_path *right_path = NULL; | |
e46f74dc | 2901 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
328d5752 | 2902 | |
099768b0 X |
2903 | if (!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))) |
2904 | return 0; | |
328d5752 MF |
2905 | |
2906 | *empty_extent_path = NULL; | |
2907 | ||
e46f74dc | 2908 | ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
328d5752 MF |
2909 | if (ret) { |
2910 | mlog_errno(ret); | |
2911 | goto out; | |
2912 | } | |
2913 | ||
ffdd7a54 | 2914 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2915 | if (!left_path) { |
2916 | ret = -ENOMEM; | |
2917 | mlog_errno(ret); | |
2918 | goto out; | |
2919 | } | |
2920 | ||
2921 | ocfs2_cp_path(left_path, path); | |
2922 | ||
ffdd7a54 | 2923 | right_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2924 | if (!right_path) { |
2925 | ret = -ENOMEM; | |
2926 | mlog_errno(ret); | |
2927 | goto out; | |
2928 | } | |
2929 | ||
2930 | while (right_cpos) { | |
facdb77f | 2931 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
328d5752 MF |
2932 | if (ret) { |
2933 | mlog_errno(ret); | |
2934 | goto out; | |
2935 | } | |
2936 | ||
7dc02805 | 2937 | subtree_root = ocfs2_find_subtree_root(et, left_path, |
328d5752 MF |
2938 | right_path); |
2939 | ||
a09d09b8 | 2940 | trace_ocfs2_rotate_subtree(subtree_root, |
328d5752 MF |
2941 | (unsigned long long) |
2942 | right_path->p_node[subtree_root].bh->b_blocknr, | |
2943 | right_path->p_tree_depth); | |
2944 | ||
17215989 | 2945 | ret = ocfs2_extend_rotate_transaction(handle, 0, |
328d5752 MF |
2946 | orig_credits, left_path); |
2947 | if (ret) { | |
2948 | mlog_errno(ret); | |
2949 | goto out; | |
2950 | } | |
2951 | ||
e8aed345 MF |
2952 | /* |
2953 | * Caller might still want to make changes to the | |
2954 | * tree root, so re-add it to the journal here. | |
2955 | */ | |
e46f74dc | 2956 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2957 | left_path, 0); |
e8aed345 MF |
2958 | if (ret) { |
2959 | mlog_errno(ret); | |
2960 | goto out; | |
2961 | } | |
2962 | ||
1e2dd63f | 2963 | ret = ocfs2_rotate_subtree_left(handle, et, left_path, |
328d5752 | 2964 | right_path, subtree_root, |
1e2dd63f | 2965 | dealloc, &deleted); |
328d5752 MF |
2966 | if (ret == -EAGAIN) { |
2967 | /* | |
2968 | * The rotation has to temporarily stop due to | |
2969 | * the right subtree having an empty | |
2970 | * extent. Pass it back to the caller for a | |
2971 | * fixup. | |
2972 | */ | |
2973 | *empty_extent_path = right_path; | |
2974 | right_path = NULL; | |
2975 | goto out; | |
2976 | } | |
2977 | if (ret) { | |
2978 | mlog_errno(ret); | |
2979 | goto out; | |
2980 | } | |
2981 | ||
2982 | /* | |
2983 | * The subtree rotate might have removed records on | |
2984 | * the rightmost edge. If so, then rotation is | |
2985 | * complete. | |
2986 | */ | |
2987 | if (deleted) | |
2988 | break; | |
2989 | ||
2990 | ocfs2_mv_path(left_path, right_path); | |
2991 | ||
e46f74dc | 2992 | ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, |
328d5752 MF |
2993 | &right_cpos); |
2994 | if (ret) { | |
2995 | mlog_errno(ret); | |
2996 | goto out; | |
2997 | } | |
2998 | } | |
2999 | ||
3000 | out: | |
3001 | ocfs2_free_path(right_path); | |
3002 | ocfs2_free_path(left_path); | |
3003 | ||
3004 | return ret; | |
3005 | } | |
3006 | ||
70f18c08 JB |
3007 | static int ocfs2_remove_rightmost_path(handle_t *handle, |
3008 | struct ocfs2_extent_tree *et, | |
e7d4cb6b | 3009 | struct ocfs2_path *path, |
70f18c08 | 3010 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3011 | { |
3012 | int ret, subtree_index; | |
3013 | u32 cpos; | |
3014 | struct ocfs2_path *left_path = NULL; | |
328d5752 MF |
3015 | struct ocfs2_extent_block *eb; |
3016 | struct ocfs2_extent_list *el; | |
3017 | ||
6136ca5f | 3018 | ret = ocfs2_et_sanity_check(et); |
e7d4cb6b TM |
3019 | if (ret) |
3020 | goto out; | |
328d5752 | 3021 | |
d9a0a1f8 | 3022 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
328d5752 MF |
3023 | if (ret) { |
3024 | mlog_errno(ret); | |
3025 | goto out; | |
3026 | } | |
3027 | ||
3d03a305 JB |
3028 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3029 | path, &cpos); | |
328d5752 MF |
3030 | if (ret) { |
3031 | mlog_errno(ret); | |
3032 | goto out; | |
3033 | } | |
3034 | ||
3035 | if (cpos) { | |
3036 | /* | |
3037 | * We have a path to the left of this one - it needs | |
3038 | * an update too. | |
3039 | */ | |
ffdd7a54 | 3040 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
3041 | if (!left_path) { |
3042 | ret = -ENOMEM; | |
3043 | mlog_errno(ret); | |
3044 | goto out; | |
3045 | } | |
3046 | ||
facdb77f | 3047 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
3048 | if (ret) { |
3049 | mlog_errno(ret); | |
3050 | goto out; | |
3051 | } | |
3052 | ||
d9a0a1f8 | 3053 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
328d5752 MF |
3054 | if (ret) { |
3055 | mlog_errno(ret); | |
3056 | goto out; | |
3057 | } | |
3058 | ||
7dc02805 | 3059 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
328d5752 | 3060 | |
6641b0ce | 3061 | ocfs2_unlink_subtree(handle, et, left_path, path, |
328d5752 | 3062 | subtree_index, dealloc); |
09106bae | 3063 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
3064 | left_path); |
3065 | if (ret) { | |
3066 | mlog_errno(ret); | |
3067 | goto out; | |
3068 | } | |
328d5752 MF |
3069 | |
3070 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 3071 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
3072 | } else { |
3073 | /* | |
3074 | * 'path' is also the leftmost path which | |
3075 | * means it must be the only one. This gets | |
3076 | * handled differently because we want to | |
70f18c08 | 3077 | * revert the root back to having extents |
328d5752 MF |
3078 | * in-line. |
3079 | */ | |
6641b0ce | 3080 | ocfs2_unlink_path(handle, et, dealloc, path, 1); |
328d5752 | 3081 | |
ce1d9ea6 | 3082 | el = et->et_root_el; |
328d5752 MF |
3083 | el->l_tree_depth = 0; |
3084 | el->l_next_free_rec = 0; | |
3085 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3086 | ||
35dc0aa3 | 3087 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
3088 | } |
3089 | ||
3090 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
3091 | ||
3092 | out: | |
3093 | ocfs2_free_path(left_path); | |
3094 | return ret; | |
3095 | } | |
3096 | ||
d0c97d52 X |
3097 | static int ocfs2_remove_rightmost_empty_extent(struct ocfs2_super *osb, |
3098 | struct ocfs2_extent_tree *et, | |
3099 | struct ocfs2_path *path, | |
3100 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
3101 | { | |
3102 | handle_t *handle; | |
3103 | int ret; | |
3104 | int credits = path->p_tree_depth * 2 + 1; | |
3105 | ||
3106 | handle = ocfs2_start_trans(osb, credits); | |
3107 | if (IS_ERR(handle)) { | |
3108 | ret = PTR_ERR(handle); | |
3109 | mlog_errno(ret); | |
3110 | return ret; | |
3111 | } | |
3112 | ||
3113 | ret = ocfs2_remove_rightmost_path(handle, et, path, dealloc); | |
3114 | if (ret) | |
3115 | mlog_errno(ret); | |
3116 | ||
3117 | ocfs2_commit_trans(osb, handle); | |
3118 | return ret; | |
3119 | } | |
3120 | ||
328d5752 MF |
3121 | /* |
3122 | * Left rotation of btree records. | |
3123 | * | |
3124 | * In many ways, this is (unsurprisingly) the opposite of right | |
3125 | * rotation. We start at some non-rightmost path containing an empty | |
3126 | * extent in the leaf block. The code works its way to the rightmost | |
3127 | * path by rotating records to the left in every subtree. | |
3128 | * | |
3129 | * This is used by any code which reduces the number of extent records | |
3130 | * in a leaf. After removal, an empty record should be placed in the | |
3131 | * leftmost list position. | |
3132 | * | |
3133 | * This won't handle a length update of the rightmost path records if | |
3134 | * the rightmost tree leaf record is removed so the caller is | |
3135 | * responsible for detecting and correcting that. | |
3136 | */ | |
70f18c08 JB |
3137 | static int ocfs2_rotate_tree_left(handle_t *handle, |
3138 | struct ocfs2_extent_tree *et, | |
328d5752 | 3139 | struct ocfs2_path *path, |
70f18c08 | 3140 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3141 | { |
3142 | int ret, orig_credits = handle->h_buffer_credits; | |
3143 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
3144 | struct ocfs2_extent_block *eb; | |
3145 | struct ocfs2_extent_list *el; | |
3146 | ||
3147 | el = path_leaf_el(path); | |
3148 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
3149 | return 0; | |
3150 | ||
3151 | if (path->p_tree_depth == 0) { | |
3152 | rightmost_no_delete: | |
3153 | /* | |
e7d4cb6b | 3154 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
3155 | * it up front. |
3156 | */ | |
70f18c08 | 3157 | ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path); |
328d5752 MF |
3158 | if (ret) |
3159 | mlog_errno(ret); | |
3160 | goto out; | |
3161 | } | |
3162 | ||
3163 | /* | |
3164 | * Handle rightmost branch now. There's several cases: | |
3165 | * 1) simple rotation leaving records in there. That's trivial. | |
3166 | * 2) rotation requiring a branch delete - there's no more | |
3167 | * records left. Two cases of this: | |
3168 | * a) There are branches to the left. | |
3169 | * b) This is also the leftmost (the only) branch. | |
3170 | * | |
3171 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
3172 | * 2a) we need the left branch so that we can update it with the unlink | |
70f18c08 | 3173 | * 2b) we need to bring the root back to inline extents. |
328d5752 MF |
3174 | */ |
3175 | ||
3176 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
3177 | el = &eb->h_list; | |
3178 | if (eb->h_next_leaf_blk == 0) { | |
3179 | /* | |
3180 | * This gets a bit tricky if we're going to delete the | |
3181 | * rightmost path. Get the other cases out of the way | |
3182 | * 1st. | |
3183 | */ | |
3184 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
3185 | goto rightmost_no_delete; | |
3186 | ||
3187 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3188 | ret = -EIO; | |
70f18c08 | 3189 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
7ecef14a | 3190 | "Owner %llu has empty extent block at %llu\n", |
70f18c08 | 3191 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
3192 | (unsigned long long)le64_to_cpu(eb->h_blkno)); |
3193 | goto out; | |
3194 | } | |
3195 | ||
3196 | /* | |
3197 | * XXX: The caller can not trust "path" any more after | |
3198 | * this as it will have been deleted. What do we do? | |
3199 | * | |
3200 | * In theory the rotate-for-merge code will never get | |
3201 | * here because it'll always ask for a rotate in a | |
3202 | * nonempty list. | |
3203 | */ | |
3204 | ||
70f18c08 JB |
3205 | ret = ocfs2_remove_rightmost_path(handle, et, path, |
3206 | dealloc); | |
328d5752 MF |
3207 | if (ret) |
3208 | mlog_errno(ret); | |
3209 | goto out; | |
3210 | } | |
3211 | ||
3212 | /* | |
3213 | * Now we can loop, remembering the path we get from -EAGAIN | |
3214 | * and restarting from there. | |
3215 | */ | |
3216 | try_rotate: | |
e46f74dc JB |
3217 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path, |
3218 | dealloc, &restart_path); | |
328d5752 MF |
3219 | if (ret && ret != -EAGAIN) { |
3220 | mlog_errno(ret); | |
3221 | goto out; | |
3222 | } | |
3223 | ||
3224 | while (ret == -EAGAIN) { | |
3225 | tmp_path = restart_path; | |
3226 | restart_path = NULL; | |
3227 | ||
e46f74dc | 3228 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, |
328d5752 | 3229 | tmp_path, dealloc, |
e46f74dc | 3230 | &restart_path); |
328d5752 MF |
3231 | if (ret && ret != -EAGAIN) { |
3232 | mlog_errno(ret); | |
3233 | goto out; | |
3234 | } | |
3235 | ||
3236 | ocfs2_free_path(tmp_path); | |
3237 | tmp_path = NULL; | |
3238 | ||
3239 | if (ret == 0) | |
3240 | goto try_rotate; | |
3241 | } | |
3242 | ||
3243 | out: | |
3244 | ocfs2_free_path(tmp_path); | |
3245 | ocfs2_free_path(restart_path); | |
3246 | return ret; | |
3247 | } | |
3248 | ||
3249 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
3250 | int index) | |
3251 | { | |
3252 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
3253 | unsigned int size; | |
3254 | ||
3255 | if (rec->e_leaf_clusters == 0) { | |
3256 | /* | |
3257 | * We consumed all of the merged-from record. An empty | |
3258 | * extent cannot exist anywhere but the 1st array | |
3259 | * position, so move things over if the merged-from | |
3260 | * record doesn't occupy that position. | |
3261 | * | |
3262 | * This creates a new empty extent so the caller | |
3263 | * should be smart enough to have removed any existing | |
3264 | * ones. | |
3265 | */ | |
3266 | if (index > 0) { | |
3267 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3268 | size = index * sizeof(struct ocfs2_extent_rec); | |
3269 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
3270 | } | |
3271 | ||
3272 | /* | |
3273 | * Always memset - the caller doesn't check whether it | |
3274 | * created an empty extent, so there could be junk in | |
3275 | * the other fields. | |
3276 | */ | |
3277 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3278 | } | |
3279 | } | |
3280 | ||
4fe82c31 | 3281 | static int ocfs2_get_right_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3282 | struct ocfs2_path *left_path, |
3283 | struct ocfs2_path **ret_right_path) | |
3284 | { | |
3285 | int ret; | |
3286 | u32 right_cpos; | |
3287 | struct ocfs2_path *right_path = NULL; | |
3288 | struct ocfs2_extent_list *left_el; | |
3289 | ||
3290 | *ret_right_path = NULL; | |
3291 | ||
3292 | /* This function shouldn't be called for non-trees. */ | |
3293 | BUG_ON(left_path->p_tree_depth == 0); | |
3294 | ||
3295 | left_el = path_leaf_el(left_path); | |
3296 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
3297 | ||
4fe82c31 JB |
3298 | ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3299 | left_path, &right_cpos); | |
677b9752 TM |
3300 | if (ret) { |
3301 | mlog_errno(ret); | |
3302 | goto out; | |
3303 | } | |
3304 | ||
3305 | /* This function shouldn't be called for the rightmost leaf. */ | |
3306 | BUG_ON(right_cpos == 0); | |
3307 | ||
ffdd7a54 | 3308 | right_path = ocfs2_new_path_from_path(left_path); |
677b9752 TM |
3309 | if (!right_path) { |
3310 | ret = -ENOMEM; | |
3311 | mlog_errno(ret); | |
3312 | goto out; | |
3313 | } | |
3314 | ||
4fe82c31 | 3315 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
677b9752 TM |
3316 | if (ret) { |
3317 | mlog_errno(ret); | |
3318 | goto out; | |
3319 | } | |
3320 | ||
3321 | *ret_right_path = right_path; | |
3322 | out: | |
3323 | if (ret) | |
3324 | ocfs2_free_path(right_path); | |
3325 | return ret; | |
3326 | } | |
3327 | ||
328d5752 MF |
3328 | /* |
3329 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3330 | * onto the beginning of the record "next" to it. |
3331 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
3332 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
3333 | * next extent block. | |
328d5752 | 3334 | */ |
4fe82c31 | 3335 | static int ocfs2_merge_rec_right(struct ocfs2_path *left_path, |
677b9752 | 3336 | handle_t *handle, |
7dc02805 | 3337 | struct ocfs2_extent_tree *et, |
677b9752 TM |
3338 | struct ocfs2_extent_rec *split_rec, |
3339 | int index) | |
328d5752 | 3340 | { |
677b9752 | 3341 | int ret, next_free, i; |
328d5752 MF |
3342 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3343 | struct ocfs2_extent_rec *left_rec; | |
3344 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3345 | struct ocfs2_extent_list *right_el; |
3346 | struct ocfs2_path *right_path = NULL; | |
3347 | int subtree_index = 0; | |
3348 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3349 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3350 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3351 | |
3352 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3353 | left_rec = &el->l_recs[index]; |
677b9752 | 3354 | |
9d8df6aa | 3355 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3356 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3357 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3358 | ret = ocfs2_get_right_path(et, left_path, &right_path); |
677b9752 TM |
3359 | if (ret) { |
3360 | mlog_errno(ret); | |
629a3b5f | 3361 | return ret; |
677b9752 TM |
3362 | } |
3363 | ||
3364 | right_el = path_leaf_el(right_path); | |
3365 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3366 | BUG_ON(next_free <= 0); | |
3367 | right_rec = &right_el->l_recs[0]; | |
3368 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3369 | BUG_ON(next_free <= 1); |
677b9752 TM |
3370 | right_rec = &right_el->l_recs[1]; |
3371 | } | |
3372 | ||
3373 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3374 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3375 | le32_to_cpu(right_rec->e_cpos)); | |
3376 | ||
7dc02805 JB |
3377 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3378 | right_path); | |
677b9752 TM |
3379 | |
3380 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3381 | handle->h_buffer_credits, | |
3382 | right_path); | |
3383 | if (ret) { | |
3384 | mlog_errno(ret); | |
3385 | goto out; | |
3386 | } | |
3387 | ||
3388 | root_bh = left_path->p_node[subtree_index].bh; | |
3389 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3390 | ||
7dc02805 | 3391 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3392 | subtree_index); |
677b9752 TM |
3393 | if (ret) { |
3394 | mlog_errno(ret); | |
3395 | goto out; | |
3396 | } | |
3397 | ||
3398 | for (i = subtree_index + 1; | |
3399 | i < path_num_items(right_path); i++) { | |
7dc02805 | 3400 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3401 | right_path, i); |
677b9752 TM |
3402 | if (ret) { |
3403 | mlog_errno(ret); | |
3404 | goto out; | |
3405 | } | |
3406 | ||
7dc02805 | 3407 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3408 | left_path, i); |
677b9752 TM |
3409 | if (ret) { |
3410 | mlog_errno(ret); | |
3411 | goto out; | |
3412 | } | |
3413 | } | |
3414 | ||
3415 | } else { | |
3416 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3417 | right_rec = &el->l_recs[index + 1]; | |
3418 | } | |
328d5752 | 3419 | |
7dc02805 | 3420 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path, |
13723d00 | 3421 | path_num_items(left_path) - 1); |
328d5752 MF |
3422 | if (ret) { |
3423 | mlog_errno(ret); | |
3424 | goto out; | |
3425 | } | |
3426 | ||
3427 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3428 | ||
3429 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3430 | le64_add_cpu(&right_rec->e_blkno, | |
7dc02805 JB |
3431 | -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3432 | split_clusters)); | |
328d5752 MF |
3433 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); |
3434 | ||
3435 | ocfs2_cleanup_merge(el, index); | |
3436 | ||
ec20cec7 | 3437 | ocfs2_journal_dirty(handle, bh); |
677b9752 | 3438 | if (right_path) { |
ec20cec7 | 3439 | ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); |
4619c73e JB |
3440 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
3441 | subtree_index); | |
677b9752 TM |
3442 | } |
3443 | out: | |
fd90d4df | 3444 | ocfs2_free_path(right_path); |
677b9752 TM |
3445 | return ret; |
3446 | } | |
3447 | ||
4fe82c31 | 3448 | static int ocfs2_get_left_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3449 | struct ocfs2_path *right_path, |
3450 | struct ocfs2_path **ret_left_path) | |
3451 | { | |
3452 | int ret; | |
3453 | u32 left_cpos; | |
3454 | struct ocfs2_path *left_path = NULL; | |
3455 | ||
3456 | *ret_left_path = NULL; | |
3457 | ||
3458 | /* This function shouldn't be called for non-trees. */ | |
3459 | BUG_ON(right_path->p_tree_depth == 0); | |
3460 | ||
4fe82c31 | 3461 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
677b9752 TM |
3462 | right_path, &left_cpos); |
3463 | if (ret) { | |
3464 | mlog_errno(ret); | |
3465 | goto out; | |
3466 | } | |
3467 | ||
3468 | /* This function shouldn't be called for the leftmost leaf. */ | |
3469 | BUG_ON(left_cpos == 0); | |
3470 | ||
ffdd7a54 | 3471 | left_path = ocfs2_new_path_from_path(right_path); |
677b9752 TM |
3472 | if (!left_path) { |
3473 | ret = -ENOMEM; | |
3474 | mlog_errno(ret); | |
3475 | goto out; | |
3476 | } | |
3477 | ||
4fe82c31 | 3478 | ret = ocfs2_find_path(et->et_ci, left_path, left_cpos); |
677b9752 TM |
3479 | if (ret) { |
3480 | mlog_errno(ret); | |
3481 | goto out; | |
3482 | } | |
3483 | ||
3484 | *ret_left_path = left_path; | |
328d5752 | 3485 | out: |
677b9752 TM |
3486 | if (ret) |
3487 | ocfs2_free_path(left_path); | |
328d5752 MF |
3488 | return ret; |
3489 | } | |
3490 | ||
3491 | /* | |
3492 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3493 | * onto the tail of the record "before" it. |
3494 | * For index > 0, the "before" means the extent rec at index - 1. | |
3495 | * | |
3496 | * For index == 0, the "before" means the last record of the previous | |
3497 | * extent block. And there is also a situation that we may need to | |
3498 | * remove the rightmost leaf extent block in the right_path and change | |
3499 | * the right path to indicate the new rightmost path. | |
328d5752 | 3500 | */ |
4fe82c31 | 3501 | static int ocfs2_merge_rec_left(struct ocfs2_path *right_path, |
328d5752 | 3502 | handle_t *handle, |
4fe82c31 | 3503 | struct ocfs2_extent_tree *et, |
328d5752 | 3504 | struct ocfs2_extent_rec *split_rec, |
677b9752 TM |
3505 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
3506 | int index) | |
328d5752 | 3507 | { |
677b9752 | 3508 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3509 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3510 | struct ocfs2_extent_rec *left_rec; | |
3511 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3512 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3513 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3514 | struct buffer_head *root_bh = NULL; | |
3515 | struct ocfs2_path *left_path = NULL; | |
3516 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3517 | |
677b9752 | 3518 | BUG_ON(index < 0); |
328d5752 | 3519 | |
328d5752 | 3520 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3521 | if (index == 0) { |
3522 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3523 | ret = ocfs2_get_left_path(et, right_path, &left_path); |
677b9752 TM |
3524 | if (ret) { |
3525 | mlog_errno(ret); | |
992ef6e7 | 3526 | return ret; |
677b9752 TM |
3527 | } |
3528 | ||
3529 | left_el = path_leaf_el(left_path); | |
3530 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3531 | le16_to_cpu(left_el->l_count)); | |
3532 | ||
3533 | left_rec = &left_el->l_recs[ | |
3534 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3535 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3536 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3537 | le32_to_cpu(split_rec->e_cpos)); | |
3538 | ||
7dc02805 JB |
3539 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3540 | right_path); | |
677b9752 TM |
3541 | |
3542 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3543 | handle->h_buffer_credits, | |
3544 | left_path); | |
3545 | if (ret) { | |
3546 | mlog_errno(ret); | |
3547 | goto out; | |
3548 | } | |
3549 | ||
3550 | root_bh = left_path->p_node[subtree_index].bh; | |
3551 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3552 | ||
4fe82c31 | 3553 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3554 | subtree_index); |
677b9752 TM |
3555 | if (ret) { |
3556 | mlog_errno(ret); | |
3557 | goto out; | |
3558 | } | |
3559 | ||
3560 | for (i = subtree_index + 1; | |
3561 | i < path_num_items(right_path); i++) { | |
4fe82c31 | 3562 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3563 | right_path, i); |
677b9752 TM |
3564 | if (ret) { |
3565 | mlog_errno(ret); | |
3566 | goto out; | |
3567 | } | |
3568 | ||
4fe82c31 | 3569 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3570 | left_path, i); |
677b9752 TM |
3571 | if (ret) { |
3572 | mlog_errno(ret); | |
3573 | goto out; | |
3574 | } | |
3575 | } | |
3576 | } else { | |
3577 | left_rec = &el->l_recs[index - 1]; | |
3578 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3579 | has_empty_extent = 1; | |
3580 | } | |
328d5752 | 3581 | |
4fe82c31 | 3582 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
9047beab | 3583 | path_num_items(right_path) - 1); |
328d5752 MF |
3584 | if (ret) { |
3585 | mlog_errno(ret); | |
3586 | goto out; | |
3587 | } | |
3588 | ||
3589 | if (has_empty_extent && index == 1) { | |
3590 | /* | |
3591 | * The easy case - we can just plop the record right in. | |
3592 | */ | |
3593 | *left_rec = *split_rec; | |
3594 | ||
3595 | has_empty_extent = 0; | |
677b9752 | 3596 | } else |
328d5752 | 3597 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3598 | |
3599 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3600 | le64_add_cpu(&right_rec->e_blkno, | |
4fe82c31 JB |
3601 | ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3602 | split_clusters)); | |
328d5752 MF |
3603 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); |
3604 | ||
3605 | ocfs2_cleanup_merge(el, index); | |
3606 | ||
ec20cec7 | 3607 | ocfs2_journal_dirty(handle, bh); |
677b9752 | 3608 | if (left_path) { |
ec20cec7 | 3609 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); |
677b9752 TM |
3610 | |
3611 | /* | |
3612 | * In the situation that the right_rec is empty and the extent | |
3613 | * block is empty also, ocfs2_complete_edge_insert can't handle | |
3614 | * it and we need to delete the right extent block. | |
3615 | */ | |
3616 | if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 && | |
3617 | le16_to_cpu(el->l_next_free_rec) == 1) { | |
17215989 X |
3618 | /* extend credit for ocfs2_remove_rightmost_path */ |
3619 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3620 | handle->h_buffer_credits, | |
3621 | right_path); | |
3622 | if (ret) { | |
3623 | mlog_errno(ret); | |
3624 | goto out; | |
3625 | } | |
677b9752 | 3626 | |
70f18c08 | 3627 | ret = ocfs2_remove_rightmost_path(handle, et, |
e7d4cb6b | 3628 | right_path, |
70f18c08 | 3629 | dealloc); |
677b9752 TM |
3630 | if (ret) { |
3631 | mlog_errno(ret); | |
3632 | goto out; | |
3633 | } | |
3634 | ||
3635 | /* Now the rightmost extent block has been deleted. | |
3636 | * So we use the new rightmost path. | |
3637 | */ | |
3638 | ocfs2_mv_path(right_path, left_path); | |
3639 | left_path = NULL; | |
3640 | } else | |
4619c73e | 3641 | ocfs2_complete_edge_insert(handle, left_path, |
677b9752 TM |
3642 | right_path, subtree_index); |
3643 | } | |
328d5752 | 3644 | out: |
fd90d4df | 3645 | ocfs2_free_path(left_path); |
328d5752 MF |
3646 | return ret; |
3647 | } | |
3648 | ||
c495dd24 JB |
3649 | static int ocfs2_try_to_merge_extent(handle_t *handle, |
3650 | struct ocfs2_extent_tree *et, | |
677b9752 | 3651 | struct ocfs2_path *path, |
328d5752 MF |
3652 | int split_index, |
3653 | struct ocfs2_extent_rec *split_rec, | |
3654 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
c495dd24 | 3655 | struct ocfs2_merge_ctxt *ctxt) |
328d5752 | 3656 | { |
518d7269 | 3657 | int ret = 0; |
677b9752 | 3658 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3659 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3660 | ||
3661 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3662 | ||
518d7269 | 3663 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
17215989 X |
3664 | /* extend credit for ocfs2_remove_rightmost_path */ |
3665 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3666 | handle->h_buffer_credits, | |
3667 | path); | |
3668 | if (ret) { | |
3669 | mlog_errno(ret); | |
3670 | goto out; | |
3671 | } | |
518d7269 TM |
3672 | /* |
3673 | * The merge code will need to create an empty | |
3674 | * extent to take the place of the newly | |
3675 | * emptied slot. Remove any pre-existing empty | |
3676 | * extents - having more than one in a leaf is | |
3677 | * illegal. | |
3678 | */ | |
70f18c08 | 3679 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
518d7269 TM |
3680 | if (ret) { |
3681 | mlog_errno(ret); | |
3682 | goto out; | |
328d5752 | 3683 | } |
518d7269 TM |
3684 | split_index--; |
3685 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3686 | } |
3687 | ||
3688 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3689 | /* | |
3690 | * Left-right contig implies this. | |
3691 | */ | |
3692 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3693 | |
3694 | /* | |
3695 | * Since the leftright insert always covers the entire | |
3696 | * extent, this call will delete the insert record | |
3697 | * entirely, resulting in an empty extent record added to | |
3698 | * the extent block. | |
3699 | * | |
3700 | * Since the adding of an empty extent shifts | |
3701 | * everything back to the right, there's no need to | |
3702 | * update split_index here. | |
677b9752 TM |
3703 | * |
3704 | * When the split_index is zero, we need to merge it to the | |
3705 | * prevoius extent block. It is more efficient and easier | |
3706 | * if we do merge_right first and merge_left later. | |
328d5752 | 3707 | */ |
4fe82c31 | 3708 | ret = ocfs2_merge_rec_right(path, handle, et, split_rec, |
677b9752 | 3709 | split_index); |
328d5752 MF |
3710 | if (ret) { |
3711 | mlog_errno(ret); | |
3712 | goto out; | |
3713 | } | |
3714 | ||
3715 | /* | |
3716 | * We can only get this from logic error above. | |
3717 | */ | |
3718 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3719 | ||
17215989 X |
3720 | /* extend credit for ocfs2_remove_rightmost_path */ |
3721 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3722 | handle->h_buffer_credits, | |
3723 | path); | |
3724 | if (ret) { | |
3725 | mlog_errno(ret); | |
3726 | goto out; | |
3727 | } | |
3728 | ||
677b9752 | 3729 | /* The merge left us with an empty extent, remove it. */ |
70f18c08 | 3730 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3731 | if (ret) { |
3732 | mlog_errno(ret); | |
3733 | goto out; | |
3734 | } | |
677b9752 | 3735 | |
328d5752 MF |
3736 | rec = &el->l_recs[split_index]; |
3737 | ||
3738 | /* | |
3739 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3740 | * we've merged it into the rec already. |
328d5752 | 3741 | */ |
4fe82c31 JB |
3742 | ret = ocfs2_merge_rec_left(path, handle, et, rec, |
3743 | dealloc, split_index); | |
677b9752 | 3744 | |
328d5752 MF |
3745 | if (ret) { |
3746 | mlog_errno(ret); | |
3747 | goto out; | |
3748 | } | |
3749 | ||
17215989 X |
3750 | /* extend credit for ocfs2_remove_rightmost_path */ |
3751 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3752 | handle->h_buffer_credits, | |
3753 | path); | |
3754 | if (ret) { | |
3755 | mlog_errno(ret); | |
3756 | goto out; | |
3757 | } | |
3758 | ||
70f18c08 | 3759 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3760 | /* |
3761 | * Error from this last rotate is not critical, so | |
3762 | * print but don't bubble it up. | |
3763 | */ | |
3764 | if (ret) | |
3765 | mlog_errno(ret); | |
3766 | ret = 0; | |
3767 | } else { | |
3768 | /* | |
3769 | * Merge a record to the left or right. | |
3770 | * | |
3771 | * 'contig_type' is relative to the existing record, | |
3772 | * so for example, if we're "right contig", it's to | |
3773 | * the record on the left (hence the left merge). | |
3774 | */ | |
3775 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
4fe82c31 JB |
3776 | ret = ocfs2_merge_rec_left(path, handle, et, |
3777 | split_rec, dealloc, | |
328d5752 MF |
3778 | split_index); |
3779 | if (ret) { | |
3780 | mlog_errno(ret); | |
3781 | goto out; | |
3782 | } | |
3783 | } else { | |
4fe82c31 | 3784 | ret = ocfs2_merge_rec_right(path, handle, |
7dc02805 | 3785 | et, split_rec, |
328d5752 MF |
3786 | split_index); |
3787 | if (ret) { | |
3788 | mlog_errno(ret); | |
3789 | goto out; | |
3790 | } | |
3791 | } | |
3792 | ||
3793 | if (ctxt->c_split_covers_rec) { | |
17215989 X |
3794 | /* extend credit for ocfs2_remove_rightmost_path */ |
3795 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3796 | handle->h_buffer_credits, | |
3797 | path); | |
3798 | if (ret) { | |
3799 | mlog_errno(ret); | |
3800 | ret = 0; | |
3801 | goto out; | |
3802 | } | |
3803 | ||
328d5752 MF |
3804 | /* |
3805 | * The merge may have left an empty extent in | |
3806 | * our leaf. Try to rotate it away. | |
3807 | */ | |
70f18c08 JB |
3808 | ret = ocfs2_rotate_tree_left(handle, et, path, |
3809 | dealloc); | |
328d5752 MF |
3810 | if (ret) |
3811 | mlog_errno(ret); | |
3812 | ret = 0; | |
3813 | } | |
3814 | } | |
3815 | ||
3816 | out: | |
3817 | return ret; | |
3818 | } | |
3819 | ||
3820 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3821 | enum ocfs2_split_type split, | |
3822 | struct ocfs2_extent_rec *rec, | |
3823 | struct ocfs2_extent_rec *split_rec) | |
3824 | { | |
3825 | u64 len_blocks; | |
3826 | ||
3827 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3828 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3829 | ||
3830 | if (split == SPLIT_LEFT) { | |
3831 | /* | |
3832 | * Region is on the left edge of the existing | |
3833 | * record. | |
3834 | */ | |
3835 | le32_add_cpu(&rec->e_cpos, | |
3836 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3837 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3838 | le16_add_cpu(&rec->e_leaf_clusters, | |
3839 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3840 | } else { | |
3841 | /* | |
3842 | * Region is on the right edge of the existing | |
3843 | * record. | |
3844 | */ | |
3845 | le16_add_cpu(&rec->e_leaf_clusters, | |
3846 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3847 | } | |
3848 | } | |
3849 | ||
3850 | /* | |
3851 | * Do the final bits of extent record insertion at the target leaf | |
3852 | * list. If this leaf is part of an allocation tree, it is assumed | |
3853 | * that the tree above has been prepared. | |
3854 | */ | |
d5628623 JB |
3855 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et, |
3856 | struct ocfs2_extent_rec *insert_rec, | |
328d5752 | 3857 | struct ocfs2_extent_list *el, |
d5628623 | 3858 | struct ocfs2_insert_type *insert) |
328d5752 MF |
3859 | { |
3860 | int i = insert->ins_contig_index; | |
3861 | unsigned int range; | |
3862 | struct ocfs2_extent_rec *rec; | |
3863 | ||
3864 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3865 | ||
3866 | if (insert->ins_split != SPLIT_NONE) { | |
3867 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3868 | BUG_ON(i == -1); | |
3869 | rec = &el->l_recs[i]; | |
d5628623 JB |
3870 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
3871 | insert->ins_split, rec, | |
328d5752 MF |
3872 | insert_rec); |
3873 | goto rotate; | |
3874 | } | |
3875 | ||
3876 | /* | |
3877 | * Contiguous insert - either left or right. | |
3878 | */ | |
3879 | if (insert->ins_contig != CONTIG_NONE) { | |
3880 | rec = &el->l_recs[i]; | |
3881 | if (insert->ins_contig == CONTIG_LEFT) { | |
3882 | rec->e_blkno = insert_rec->e_blkno; | |
3883 | rec->e_cpos = insert_rec->e_cpos; | |
3884 | } | |
3885 | le16_add_cpu(&rec->e_leaf_clusters, | |
3886 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3887 | return; | |
3888 | } | |
3889 | ||
3890 | /* | |
3891 | * Handle insert into an empty leaf. | |
3892 | */ | |
3893 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3894 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3895 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3896 | el->l_recs[0] = *insert_rec; | |
3897 | el->l_next_free_rec = cpu_to_le16(1); | |
3898 | return; | |
3899 | } | |
3900 | ||
3901 | /* | |
3902 | * Appending insert. | |
3903 | */ | |
3904 | if (insert->ins_appending == APPEND_TAIL) { | |
3905 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3906 | rec = &el->l_recs[i]; | |
3907 | range = le32_to_cpu(rec->e_cpos) | |
3908 | + le16_to_cpu(rec->e_leaf_clusters); | |
3909 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3910 | ||
3911 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3912 | le16_to_cpu(el->l_count), | |
d5628623 | 3913 | "owner %llu, depth %u, count %u, next free %u, " |
328d5752 MF |
3914 | "rec.cpos %u, rec.clusters %u, " |
3915 | "insert.cpos %u, insert.clusters %u\n", | |
d5628623 | 3916 | ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
3917 | le16_to_cpu(el->l_tree_depth), |
3918 | le16_to_cpu(el->l_count), | |
3919 | le16_to_cpu(el->l_next_free_rec), | |
3920 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3921 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3922 | le32_to_cpu(insert_rec->e_cpos), | |
3923 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3924 | i++; | |
3925 | el->l_recs[i] = *insert_rec; | |
3926 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3927 | return; | |
3928 | } | |
3929 | ||
3930 | rotate: | |
3931 | /* | |
3932 | * Ok, we have to rotate. | |
3933 | * | |
3934 | * At this point, it is safe to assume that inserting into an | |
3935 | * empty leaf and appending to a leaf have both been handled | |
3936 | * above. | |
3937 | * | |
3938 | * This leaf needs to have space, either by the empty 1st | |
3939 | * extent record, or by virtue of an l_next_rec < l_count. | |
3940 | */ | |
3941 | ocfs2_rotate_leaf(el, insert_rec); | |
3942 | } | |
3943 | ||
d401dc12 JB |
3944 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
3945 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3946 | struct ocfs2_path *path, |
3947 | struct ocfs2_extent_rec *insert_rec) | |
3948 | { | |
3949 | int ret, i, next_free; | |
3950 | struct buffer_head *bh; | |
3951 | struct ocfs2_extent_list *el; | |
3952 | struct ocfs2_extent_rec *rec; | |
3953 | ||
3954 | /* | |
3955 | * Update everything except the leaf block. | |
3956 | */ | |
3957 | for (i = 0; i < path->p_tree_depth; i++) { | |
3958 | bh = path->p_node[i].bh; | |
3959 | el = path->p_node[i].el; | |
3960 | ||
dcd0538f MF |
3961 | next_free = le16_to_cpu(el->l_next_free_rec); |
3962 | if (next_free == 0) { | |
d401dc12 | 3963 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
7ecef14a | 3964 | "Owner %llu has a bad extent list\n", |
d401dc12 | 3965 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); |
dcd0538f | 3966 | ret = -EIO; |
328d5752 MF |
3967 | return; |
3968 | } | |
3969 | ||
3970 | rec = &el->l_recs[next_free - 1]; | |
3971 | ||
3972 | rec->e_int_clusters = insert_rec->e_cpos; | |
3973 | le32_add_cpu(&rec->e_int_clusters, | |
3974 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3975 | le32_add_cpu(&rec->e_int_clusters, | |
3976 | -le32_to_cpu(rec->e_cpos)); | |
3977 | ||
ec20cec7 | 3978 | ocfs2_journal_dirty(handle, bh); |
328d5752 MF |
3979 | } |
3980 | } | |
3981 | ||
d401dc12 JB |
3982 | static int ocfs2_append_rec_to_path(handle_t *handle, |
3983 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3984 | struct ocfs2_extent_rec *insert_rec, |
3985 | struct ocfs2_path *right_path, | |
3986 | struct ocfs2_path **ret_left_path) | |
3987 | { | |
3988 | int ret, next_free; | |
3989 | struct ocfs2_extent_list *el; | |
3990 | struct ocfs2_path *left_path = NULL; | |
3991 | ||
3992 | *ret_left_path = NULL; | |
3993 | ||
3994 | /* | |
3995 | * This shouldn't happen for non-trees. The extent rec cluster | |
3996 | * count manipulation below only works for interior nodes. | |
3997 | */ | |
3998 | BUG_ON(right_path->p_tree_depth == 0); | |
3999 | ||
4000 | /* | |
4001 | * If our appending insert is at the leftmost edge of a leaf, | |
4002 | * then we might need to update the rightmost records of the | |
4003 | * neighboring path. | |
4004 | */ | |
4005 | el = path_leaf_el(right_path); | |
4006 | next_free = le16_to_cpu(el->l_next_free_rec); | |
4007 | if (next_free == 0 || | |
4008 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
4009 | u32 left_cpos; | |
4010 | ||
d401dc12 JB |
4011 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
4012 | right_path, &left_cpos); | |
328d5752 MF |
4013 | if (ret) { |
4014 | mlog_errno(ret); | |
dcd0538f MF |
4015 | goto out; |
4016 | } | |
4017 | ||
a09d09b8 TM |
4018 | trace_ocfs2_append_rec_to_path( |
4019 | (unsigned long long) | |
4020 | ocfs2_metadata_cache_owner(et->et_ci), | |
4021 | le32_to_cpu(insert_rec->e_cpos), | |
4022 | left_cpos); | |
e48edee2 | 4023 | |
328d5752 MF |
4024 | /* |
4025 | * No need to worry if the append is already in the | |
4026 | * leftmost leaf. | |
4027 | */ | |
4028 | if (left_cpos) { | |
ffdd7a54 | 4029 | left_path = ocfs2_new_path_from_path(right_path); |
328d5752 MF |
4030 | if (!left_path) { |
4031 | ret = -ENOMEM; | |
4032 | mlog_errno(ret); | |
4033 | goto out; | |
4034 | } | |
dcd0538f | 4035 | |
d401dc12 | 4036 | ret = ocfs2_find_path(et->et_ci, left_path, |
facdb77f | 4037 | left_cpos); |
328d5752 MF |
4038 | if (ret) { |
4039 | mlog_errno(ret); | |
4040 | goto out; | |
4041 | } | |
dcd0538f | 4042 | |
328d5752 MF |
4043 | /* |
4044 | * ocfs2_insert_path() will pass the left_path to the | |
4045 | * journal for us. | |
4046 | */ | |
4047 | } | |
4048 | } | |
dcd0538f | 4049 | |
d401dc12 | 4050 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
328d5752 MF |
4051 | if (ret) { |
4052 | mlog_errno(ret); | |
4053 | goto out; | |
dcd0538f MF |
4054 | } |
4055 | ||
d401dc12 | 4056 | ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec); |
328d5752 | 4057 | |
dcd0538f MF |
4058 | *ret_left_path = left_path; |
4059 | ret = 0; | |
4060 | out: | |
4061 | if (ret != 0) | |
4062 | ocfs2_free_path(left_path); | |
4063 | ||
4064 | return ret; | |
4065 | } | |
4066 | ||
c38e52bb | 4067 | static void ocfs2_split_record(struct ocfs2_extent_tree *et, |
328d5752 MF |
4068 | struct ocfs2_path *left_path, |
4069 | struct ocfs2_path *right_path, | |
4070 | struct ocfs2_extent_rec *split_rec, | |
4071 | enum ocfs2_split_type split) | |
4072 | { | |
4073 | int index; | |
4074 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
4075 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
4076 | struct ocfs2_extent_rec *rec, *tmprec; | |
4077 | ||
c19a28e1 | 4078 | right_el = path_leaf_el(right_path); |
328d5752 MF |
4079 | if (left_path) |
4080 | left_el = path_leaf_el(left_path); | |
4081 | ||
4082 | el = right_el; | |
4083 | insert_el = right_el; | |
4084 | index = ocfs2_search_extent_list(el, cpos); | |
4085 | if (index != -1) { | |
4086 | if (index == 0 && left_path) { | |
4087 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
4088 | ||
4089 | /* | |
4090 | * This typically means that the record | |
4091 | * started in the left path but moved to the | |
4092 | * right as a result of rotation. We either | |
4093 | * move the existing record to the left, or we | |
4094 | * do the later insert there. | |
4095 | * | |
4096 | * In this case, the left path should always | |
4097 | * exist as the rotate code will have passed | |
4098 | * it back for a post-insert update. | |
4099 | */ | |
4100 | ||
4101 | if (split == SPLIT_LEFT) { | |
4102 | /* | |
4103 | * It's a left split. Since we know | |
4104 | * that the rotate code gave us an | |
4105 | * empty extent in the left path, we | |
4106 | * can just do the insert there. | |
4107 | */ | |
4108 | insert_el = left_el; | |
4109 | } else { | |
4110 | /* | |
4111 | * Right split - we have to move the | |
4112 | * existing record over to the left | |
4113 | * leaf. The insert will be into the | |
4114 | * newly created empty extent in the | |
4115 | * right leaf. | |
4116 | */ | |
4117 | tmprec = &right_el->l_recs[index]; | |
4118 | ocfs2_rotate_leaf(left_el, tmprec); | |
4119 | el = left_el; | |
4120 | ||
4121 | memset(tmprec, 0, sizeof(*tmprec)); | |
4122 | index = ocfs2_search_extent_list(left_el, cpos); | |
4123 | BUG_ON(index == -1); | |
4124 | } | |
4125 | } | |
4126 | } else { | |
4127 | BUG_ON(!left_path); | |
4128 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
4129 | /* | |
4130 | * Left path is easy - we can just allow the insert to | |
4131 | * happen. | |
4132 | */ | |
4133 | el = left_el; | |
4134 | insert_el = left_el; | |
4135 | index = ocfs2_search_extent_list(el, cpos); | |
4136 | BUG_ON(index == -1); | |
4137 | } | |
4138 | ||
4139 | rec = &el->l_recs[index]; | |
c38e52bb JB |
4140 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
4141 | split, rec, split_rec); | |
328d5752 MF |
4142 | ocfs2_rotate_leaf(insert_el, split_rec); |
4143 | } | |
4144 | ||
dcd0538f | 4145 | /* |
e7d4cb6b TM |
4146 | * This function only does inserts on an allocation b-tree. For tree |
4147 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
4148 | * |
4149 | * right_path is the path we want to do the actual insert | |
4150 | * in. left_path should only be passed in if we need to update that | |
4151 | * portion of the tree after an edge insert. | |
4152 | */ | |
3505bec0 | 4153 | static int ocfs2_insert_path(handle_t *handle, |
7dc02805 | 4154 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4155 | struct ocfs2_path *left_path, |
4156 | struct ocfs2_path *right_path, | |
4157 | struct ocfs2_extent_rec *insert_rec, | |
4158 | struct ocfs2_insert_type *insert) | |
4159 | { | |
4160 | int ret, subtree_index; | |
4161 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 4162 | |
dcd0538f | 4163 | if (left_path) { |
dcd0538f MF |
4164 | /* |
4165 | * There's a chance that left_path got passed back to | |
4166 | * us without being accounted for in the | |
4167 | * journal. Extend our transaction here to be sure we | |
4168 | * can change those blocks. | |
4169 | */ | |
c901fb00 | 4170 | ret = ocfs2_extend_trans(handle, left_path->p_tree_depth); |
dcd0538f MF |
4171 | if (ret < 0) { |
4172 | mlog_errno(ret); | |
4173 | goto out; | |
4174 | } | |
4175 | ||
7dc02805 | 4176 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
dcd0538f MF |
4177 | if (ret < 0) { |
4178 | mlog_errno(ret); | |
4179 | goto out; | |
4180 | } | |
4181 | } | |
4182 | ||
e8aed345 MF |
4183 | /* |
4184 | * Pass both paths to the journal. The majority of inserts | |
4185 | * will be touching all components anyway. | |
4186 | */ | |
7dc02805 | 4187 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
e8aed345 MF |
4188 | if (ret < 0) { |
4189 | mlog_errno(ret); | |
4190 | goto out; | |
4191 | } | |
4192 | ||
328d5752 MF |
4193 | if (insert->ins_split != SPLIT_NONE) { |
4194 | /* | |
4195 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 4196 | * of splits, but it's easier to just let one separate |
328d5752 MF |
4197 | * function sort it all out. |
4198 | */ | |
c38e52bb | 4199 | ocfs2_split_record(et, left_path, right_path, |
328d5752 | 4200 | insert_rec, insert->ins_split); |
e8aed345 MF |
4201 | |
4202 | /* | |
4203 | * Split might have modified either leaf and we don't | |
4204 | * have a guarantee that the later edge insert will | |
4205 | * dirty this for us. | |
4206 | */ | |
4207 | if (left_path) | |
ec20cec7 JB |
4208 | ocfs2_journal_dirty(handle, |
4209 | path_leaf_bh(left_path)); | |
328d5752 | 4210 | } else |
d5628623 JB |
4211 | ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path), |
4212 | insert); | |
dcd0538f | 4213 | |
ec20cec7 | 4214 | ocfs2_journal_dirty(handle, leaf_bh); |
dcd0538f MF |
4215 | |
4216 | if (left_path) { | |
4217 | /* | |
4218 | * The rotate code has indicated that we need to fix | |
4219 | * up portions of the tree after the insert. | |
4220 | * | |
4221 | * XXX: Should we extend the transaction here? | |
4222 | */ | |
7dc02805 | 4223 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
dcd0538f | 4224 | right_path); |
4619c73e JB |
4225 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
4226 | subtree_index); | |
dcd0538f MF |
4227 | } |
4228 | ||
4229 | ret = 0; | |
4230 | out: | |
4231 | return ret; | |
4232 | } | |
4233 | ||
3505bec0 | 4234 | static int ocfs2_do_insert_extent(handle_t *handle, |
e7d4cb6b | 4235 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4236 | struct ocfs2_extent_rec *insert_rec, |
4237 | struct ocfs2_insert_type *type) | |
4238 | { | |
4239 | int ret, rotate = 0; | |
4240 | u32 cpos; | |
4241 | struct ocfs2_path *right_path = NULL; | |
4242 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
4243 | struct ocfs2_extent_list *el; |
4244 | ||
ce1d9ea6 | 4245 | el = et->et_root_el; |
dcd0538f | 4246 | |
d9a0a1f8 | 4247 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4248 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
4249 | if (ret) { |
4250 | mlog_errno(ret); | |
4251 | goto out; | |
4252 | } | |
4253 | ||
4254 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
d5628623 | 4255 | ocfs2_insert_at_leaf(et, insert_rec, el, type); |
dcd0538f MF |
4256 | goto out_update_clusters; |
4257 | } | |
4258 | ||
ffdd7a54 | 4259 | right_path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4260 | if (!right_path) { |
4261 | ret = -ENOMEM; | |
4262 | mlog_errno(ret); | |
4263 | goto out; | |
4264 | } | |
4265 | ||
4266 | /* | |
4267 | * Determine the path to start with. Rotations need the | |
4268 | * rightmost path, everything else can go directly to the | |
4269 | * target leaf. | |
4270 | */ | |
4271 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
4272 | if (type->ins_appending == APPEND_NONE && | |
4273 | type->ins_contig == CONTIG_NONE) { | |
4274 | rotate = 1; | |
4275 | cpos = UINT_MAX; | |
4276 | } | |
4277 | ||
facdb77f | 4278 | ret = ocfs2_find_path(et->et_ci, right_path, cpos); |
dcd0538f MF |
4279 | if (ret) { |
4280 | mlog_errno(ret); | |
4281 | goto out; | |
4282 | } | |
4283 | ||
4284 | /* | |
4285 | * Rotations and appends need special treatment - they modify | |
4286 | * parts of the tree's above them. | |
4287 | * | |
4288 | * Both might pass back a path immediate to the left of the | |
4289 | * one being inserted to. This will be cause | |
4290 | * ocfs2_insert_path() to modify the rightmost records of | |
4291 | * left_path to account for an edge insert. | |
4292 | * | |
4293 | * XXX: When modifying this code, keep in mind that an insert | |
4294 | * can wind up skipping both of these two special cases... | |
4295 | */ | |
4296 | if (rotate) { | |
1bbf0b8d | 4297 | ret = ocfs2_rotate_tree_right(handle, et, type->ins_split, |
dcd0538f MF |
4298 | le32_to_cpu(insert_rec->e_cpos), |
4299 | right_path, &left_path); | |
4300 | if (ret) { | |
4301 | mlog_errno(ret); | |
4302 | goto out; | |
4303 | } | |
e8aed345 MF |
4304 | |
4305 | /* | |
4306 | * ocfs2_rotate_tree_right() might have extended the | |
4307 | * transaction without re-journaling our tree root. | |
4308 | */ | |
d9a0a1f8 | 4309 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4310 | OCFS2_JOURNAL_ACCESS_WRITE); |
e8aed345 MF |
4311 | if (ret) { |
4312 | mlog_errno(ret); | |
4313 | goto out; | |
4314 | } | |
dcd0538f MF |
4315 | } else if (type->ins_appending == APPEND_TAIL |
4316 | && type->ins_contig != CONTIG_LEFT) { | |
d401dc12 | 4317 | ret = ocfs2_append_rec_to_path(handle, et, insert_rec, |
dcd0538f MF |
4318 | right_path, &left_path); |
4319 | if (ret) { | |
4320 | mlog_errno(ret); | |
4321 | goto out; | |
4322 | } | |
4323 | } | |
4324 | ||
3505bec0 | 4325 | ret = ocfs2_insert_path(handle, et, left_path, right_path, |
dcd0538f MF |
4326 | insert_rec, type); |
4327 | if (ret) { | |
4328 | mlog_errno(ret); | |
4329 | goto out; | |
4330 | } | |
4331 | ||
4332 | out_update_clusters: | |
328d5752 | 4333 | if (type->ins_split == SPLIT_NONE) |
6136ca5f | 4334 | ocfs2_et_update_clusters(et, |
35dc0aa3 | 4335 | le16_to_cpu(insert_rec->e_leaf_clusters)); |
dcd0538f | 4336 | |
ec20cec7 | 4337 | ocfs2_journal_dirty(handle, et->et_root_bh); |
dcd0538f MF |
4338 | |
4339 | out: | |
4340 | ocfs2_free_path(left_path); | |
4341 | ocfs2_free_path(right_path); | |
4342 | ||
4343 | return ret; | |
4344 | } | |
4345 | ||
9f99ad08 | 4346 | static int ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et, |
a2970291 | 4347 | struct ocfs2_path *path, |
328d5752 | 4348 | struct ocfs2_extent_list *el, int index, |
9f99ad08 X |
4349 | struct ocfs2_extent_rec *split_rec, |
4350 | struct ocfs2_merge_ctxt *ctxt) | |
328d5752 | 4351 | { |
9f99ad08 | 4352 | int status = 0; |
328d5752 | 4353 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4354 | u32 left_cpos, right_cpos; |
4355 | struct ocfs2_extent_rec *rec = NULL; | |
4356 | struct ocfs2_extent_list *new_el; | |
4357 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4358 | struct buffer_head *bh; | |
4359 | struct ocfs2_extent_block *eb; | |
a2970291 | 4360 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
ad5a4d70 TM |
4361 | |
4362 | if (index > 0) { | |
4363 | rec = &el->l_recs[index - 1]; | |
4364 | } else if (path->p_tree_depth > 0) { | |
a2970291 | 4365 | status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
ad5a4d70 | 4366 | if (status) |
06a269cc | 4367 | goto exit; |
ad5a4d70 TM |
4368 | |
4369 | if (left_cpos != 0) { | |
ffdd7a54 | 4370 | left_path = ocfs2_new_path_from_path(path); |
9f99ad08 X |
4371 | if (!left_path) { |
4372 | status = -ENOMEM; | |
4373 | mlog_errno(status); | |
06a269cc | 4374 | goto exit; |
9f99ad08 | 4375 | } |
ad5a4d70 | 4376 | |
a2970291 JB |
4377 | status = ocfs2_find_path(et->et_ci, left_path, |
4378 | left_cpos); | |
ad5a4d70 | 4379 | if (status) |
06a269cc | 4380 | goto free_left_path; |
ad5a4d70 TM |
4381 | |
4382 | new_el = path_leaf_el(left_path); | |
4383 | ||
4384 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4385 | le16_to_cpu(new_el->l_count)) { | |
4386 | bh = path_leaf_bh(left_path); | |
4387 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4388 | ocfs2_error(sb, |
7ecef14a | 4389 | "Extent block #%llu has an invalid l_next_free_rec of %d. It should have matched the l_count of %d\n", |
5e96581a JB |
4390 | (unsigned long long)le64_to_cpu(eb->h_blkno), |
4391 | le16_to_cpu(new_el->l_next_free_rec), | |
4392 | le16_to_cpu(new_el->l_count)); | |
4393 | status = -EINVAL; | |
06a269cc | 4394 | goto free_left_path; |
ad5a4d70 TM |
4395 | } |
4396 | rec = &new_el->l_recs[ | |
4397 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4398 | } | |
4399 | } | |
328d5752 MF |
4400 | |
4401 | /* | |
4402 | * We're careful to check for an empty extent record here - | |
4403 | * the merge code will know what to do if it sees one. | |
4404 | */ | |
ad5a4d70 | 4405 | if (rec) { |
328d5752 MF |
4406 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4407 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4408 | ret = CONTIG_RIGHT; | |
4409 | } else { | |
853a3a14 | 4410 | ret = ocfs2_et_extent_contig(et, rec, split_rec); |
328d5752 MF |
4411 | } |
4412 | } | |
4413 | ||
ad5a4d70 TM |
4414 | rec = NULL; |
4415 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4416 | rec = &el->l_recs[index + 1]; | |
4417 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4418 | path->p_tree_depth > 0) { | |
a2970291 | 4419 | status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
ad5a4d70 | 4420 | if (status) |
06a269cc | 4421 | goto free_left_path; |
ad5a4d70 TM |
4422 | |
4423 | if (right_cpos == 0) | |
06a269cc | 4424 | goto free_left_path; |
ad5a4d70 | 4425 | |
ffdd7a54 | 4426 | right_path = ocfs2_new_path_from_path(path); |
9f99ad08 X |
4427 | if (!right_path) { |
4428 | status = -ENOMEM; | |
4429 | mlog_errno(status); | |
06a269cc | 4430 | goto free_left_path; |
9f99ad08 | 4431 | } |
ad5a4d70 | 4432 | |
a2970291 | 4433 | status = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
ad5a4d70 | 4434 | if (status) |
06a269cc | 4435 | goto free_right_path; |
ad5a4d70 TM |
4436 | |
4437 | new_el = path_leaf_el(right_path); | |
4438 | rec = &new_el->l_recs[0]; | |
4439 | if (ocfs2_is_empty_extent(rec)) { | |
4440 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4441 | bh = path_leaf_bh(right_path); | |
4442 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4443 | ocfs2_error(sb, |
7ecef14a | 4444 | "Extent block #%llu has an invalid l_next_free_rec of %d\n", |
5e96581a JB |
4445 | (unsigned long long)le64_to_cpu(eb->h_blkno), |
4446 | le16_to_cpu(new_el->l_next_free_rec)); | |
4447 | status = -EINVAL; | |
06a269cc | 4448 | goto free_right_path; |
ad5a4d70 TM |
4449 | } |
4450 | rec = &new_el->l_recs[1]; | |
4451 | } | |
4452 | } | |
4453 | ||
4454 | if (rec) { | |
328d5752 MF |
4455 | enum ocfs2_contig_type contig_type; |
4456 | ||
853a3a14 | 4457 | contig_type = ocfs2_et_extent_contig(et, rec, split_rec); |
328d5752 MF |
4458 | |
4459 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4460 | ret = CONTIG_LEFTRIGHT; | |
4461 | else if (ret == CONTIG_NONE) | |
4462 | ret = contig_type; | |
4463 | } | |
4464 | ||
06a269cc | 4465 | free_right_path: |
fd90d4df | 4466 | ocfs2_free_path(right_path); |
06a269cc ME |
4467 | free_left_path: |
4468 | ocfs2_free_path(left_path); | |
4469 | exit: | |
9f99ad08 X |
4470 | if (status == 0) |
4471 | ctxt->c_contig_type = ret; | |
4472 | ||
4473 | return status; | |
328d5752 MF |
4474 | } |
4475 | ||
1ef61b33 | 4476 | static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et, |
dcd0538f MF |
4477 | struct ocfs2_insert_type *insert, |
4478 | struct ocfs2_extent_list *el, | |
1ef61b33 | 4479 | struct ocfs2_extent_rec *insert_rec) |
dcd0538f MF |
4480 | { |
4481 | int i; | |
4482 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4483 | ||
e48edee2 MF |
4484 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4485 | ||
dcd0538f | 4486 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
853a3a14 TM |
4487 | contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i], |
4488 | insert_rec); | |
dcd0538f MF |
4489 | if (contig_type != CONTIG_NONE) { |
4490 | insert->ins_contig_index = i; | |
4491 | break; | |
4492 | } | |
4493 | } | |
4494 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4495 | |
4496 | if (insert->ins_contig != CONTIG_NONE) { | |
4497 | struct ocfs2_extent_rec *rec = | |
4498 | &el->l_recs[insert->ins_contig_index]; | |
4499 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4500 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4501 | ||
4502 | /* | |
4503 | * Caller might want us to limit the size of extents, don't | |
4504 | * calculate contiguousness if we might exceed that limit. | |
4505 | */ | |
ce1d9ea6 JB |
4506 | if (et->et_max_leaf_clusters && |
4507 | (len > et->et_max_leaf_clusters)) | |
ca12b7c4 TM |
4508 | insert->ins_contig = CONTIG_NONE; |
4509 | } | |
dcd0538f MF |
4510 | } |
4511 | ||
4512 | /* | |
4513 | * This should only be called against the righmost leaf extent list. | |
4514 | * | |
4515 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4516 | * insert at the tail of the rightmost leaf. | |
4517 | * | |
e7d4cb6b TM |
4518 | * This should also work against the root extent list for tree's with 0 |
4519 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4520 | * then the logic here makes sense. |
4521 | */ | |
4522 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4523 | struct ocfs2_extent_list *el, | |
4524 | struct ocfs2_extent_rec *insert_rec) | |
4525 | { | |
4526 | int i; | |
4527 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4528 | struct ocfs2_extent_rec *rec; | |
4529 | ||
4530 | insert->ins_appending = APPEND_NONE; | |
4531 | ||
e48edee2 | 4532 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4533 | |
4534 | if (!el->l_next_free_rec) | |
4535 | goto set_tail_append; | |
4536 | ||
4537 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4538 | /* Were all records empty? */ | |
4539 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4540 | goto set_tail_append; | |
4541 | } | |
4542 | ||
4543 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4544 | rec = &el->l_recs[i]; | |
4545 | ||
e48edee2 MF |
4546 | if (cpos >= |
4547 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4548 | goto set_tail_append; |
4549 | ||
4550 | return; | |
4551 | ||
4552 | set_tail_append: | |
4553 | insert->ins_appending = APPEND_TAIL; | |
4554 | } | |
4555 | ||
4556 | /* | |
25985edc | 4557 | * Helper function called at the beginning of an insert. |
dcd0538f MF |
4558 | * |
4559 | * This computes a few things that are commonly used in the process of | |
4560 | * inserting into the btree: | |
4561 | * - Whether the new extent is contiguous with an existing one. | |
4562 | * - The current tree depth. | |
4563 | * - Whether the insert is an appending one. | |
4564 | * - The total # of free records in the tree. | |
4565 | * | |
4566 | * All of the information is stored on the ocfs2_insert_type | |
4567 | * structure. | |
4568 | */ | |
627961b7 | 4569 | static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et, |
dcd0538f MF |
4570 | struct buffer_head **last_eb_bh, |
4571 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4572 | int *free_records, |
dcd0538f MF |
4573 | struct ocfs2_insert_type *insert) |
4574 | { | |
4575 | int ret; | |
dcd0538f MF |
4576 | struct ocfs2_extent_block *eb; |
4577 | struct ocfs2_extent_list *el; | |
4578 | struct ocfs2_path *path = NULL; | |
4579 | struct buffer_head *bh = NULL; | |
4580 | ||
328d5752 MF |
4581 | insert->ins_split = SPLIT_NONE; |
4582 | ||
ce1d9ea6 | 4583 | el = et->et_root_el; |
dcd0538f MF |
4584 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4585 | ||
4586 | if (el->l_tree_depth) { | |
4587 | /* | |
4588 | * If we have tree depth, we read in the | |
4589 | * rightmost extent block ahead of time as | |
4590 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4591 | * may want it later. | |
4592 | */ | |
3d03a305 | 4593 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
4594 | ocfs2_et_get_last_eb_blk(et), |
4595 | &bh); | |
dcd0538f | 4596 | if (ret) { |
c1e8d35e | 4597 | mlog_errno(ret); |
dcd0538f MF |
4598 | goto out; |
4599 | } | |
ccd979bd | 4600 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4601 | el = &eb->h_list; |
dcd0538f | 4602 | } |
ccd979bd | 4603 | |
dcd0538f MF |
4604 | /* |
4605 | * Unless we have a contiguous insert, we'll need to know if | |
4606 | * there is room left in our allocation tree for another | |
4607 | * extent record. | |
4608 | * | |
4609 | * XXX: This test is simplistic, we can search for empty | |
4610 | * extent records too. | |
4611 | */ | |
c77534f6 | 4612 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4613 | le16_to_cpu(el->l_next_free_rec); |
4614 | ||
4615 | if (!insert->ins_tree_depth) { | |
1ef61b33 | 4616 | ocfs2_figure_contig_type(et, insert, el, insert_rec); |
dcd0538f MF |
4617 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4618 | return 0; | |
ccd979bd MF |
4619 | } |
4620 | ||
ffdd7a54 | 4621 | path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4622 | if (!path) { |
4623 | ret = -ENOMEM; | |
4624 | mlog_errno(ret); | |
4625 | goto out; | |
4626 | } | |
ccd979bd | 4627 | |
dcd0538f MF |
4628 | /* |
4629 | * In the case that we're inserting past what the tree | |
4630 | * currently accounts for, ocfs2_find_path() will return for | |
4631 | * us the rightmost tree path. This is accounted for below in | |
4632 | * the appending code. | |
4633 | */ | |
facdb77f | 4634 | ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos)); |
dcd0538f MF |
4635 | if (ret) { |
4636 | mlog_errno(ret); | |
4637 | goto out; | |
4638 | } | |
ccd979bd | 4639 | |
dcd0538f MF |
4640 | el = path_leaf_el(path); |
4641 | ||
4642 | /* | |
4643 | * Now that we have the path, there's two things we want to determine: | |
4644 | * 1) Contiguousness (also set contig_index if this is so) | |
4645 | * | |
4646 | * 2) Are we doing an append? We can trivially break this up | |
4647 | * into two types of appends: simple record append, or a | |
4648 | * rotate inside the tail leaf. | |
4649 | */ | |
1ef61b33 | 4650 | ocfs2_figure_contig_type(et, insert, el, insert_rec); |
dcd0538f MF |
4651 | |
4652 | /* | |
4653 | * The insert code isn't quite ready to deal with all cases of | |
4654 | * left contiguousness. Specifically, if it's an insert into | |
4655 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4656 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4657 | * left. For now, just catch that case and fool the layers |
4658 | * above us. This works just fine for tree_depth == 0, which | |
4659 | * is why we allow that above. | |
4660 | */ | |
4661 | if (insert->ins_contig == CONTIG_LEFT && | |
4662 | insert->ins_contig_index == 0) | |
4663 | insert->ins_contig = CONTIG_NONE; | |
4664 | ||
4665 | /* | |
4666 | * Ok, so we can simply compare against last_eb to figure out | |
4667 | * whether the path doesn't exist. This will only happen in | |
4668 | * the case that we're doing a tail append, so maybe we can | |
4669 | * take advantage of that information somehow. | |
4670 | */ | |
35dc0aa3 | 4671 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4672 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4673 | /* |
4674 | * Ok, ocfs2_find_path() returned us the rightmost | |
4675 | * tree path. This might be an appending insert. There are | |
4676 | * two cases: | |
4677 | * 1) We're doing a true append at the tail: | |
4678 | * -This might even be off the end of the leaf | |
4679 | * 2) We're "appending" by rotating in the tail | |
4680 | */ | |
4681 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4682 | } | |
4683 | ||
4684 | out: | |
4685 | ocfs2_free_path(path); | |
4686 | ||
4687 | if (ret == 0) | |
4688 | *last_eb_bh = bh; | |
4689 | else | |
4690 | brelse(bh); | |
4691 | return ret; | |
ccd979bd MF |
4692 | } |
4693 | ||
dcd0538f | 4694 | /* |
cc79d8c1 | 4695 | * Insert an extent into a btree. |
dcd0538f | 4696 | * |
cc79d8c1 | 4697 | * The caller needs to update the owning btree's cluster count. |
dcd0538f | 4698 | */ |
cc79d8c1 | 4699 | int ocfs2_insert_extent(handle_t *handle, |
f99b9b7c JB |
4700 | struct ocfs2_extent_tree *et, |
4701 | u32 cpos, | |
4702 | u64 start_blk, | |
4703 | u32 new_clusters, | |
4704 | u8 flags, | |
4705 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 4706 | { |
c3afcbb3 | 4707 | int status; |
c77534f6 | 4708 | int uninitialized_var(free_records); |
ccd979bd | 4709 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4710 | struct ocfs2_insert_type insert = {0, }; |
4711 | struct ocfs2_extent_rec rec; | |
4712 | ||
a09d09b8 TM |
4713 | trace_ocfs2_insert_extent_start( |
4714 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
4715 | cpos, new_clusters); | |
dcd0538f | 4716 | |
e48edee2 | 4717 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4718 | rec.e_cpos = cpu_to_le32(cpos); |
4719 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4720 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4721 | rec.e_flags = flags; |
6136ca5f | 4722 | status = ocfs2_et_insert_check(et, &rec); |
1e61ee79 JB |
4723 | if (status) { |
4724 | mlog_errno(status); | |
4725 | goto bail; | |
4726 | } | |
dcd0538f | 4727 | |
627961b7 | 4728 | status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec, |
c77534f6 | 4729 | &free_records, &insert); |
dcd0538f MF |
4730 | if (status < 0) { |
4731 | mlog_errno(status); | |
4732 | goto bail; | |
ccd979bd MF |
4733 | } |
4734 | ||
a09d09b8 TM |
4735 | trace_ocfs2_insert_extent(insert.ins_appending, insert.ins_contig, |
4736 | insert.ins_contig_index, free_records, | |
4737 | insert.ins_tree_depth); | |
ccd979bd | 4738 | |
c77534f6 | 4739 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
d401dc12 | 4740 | status = ocfs2_grow_tree(handle, et, |
328d5752 | 4741 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4742 | meta_ac); |
4743 | if (status) { | |
ccd979bd MF |
4744 | mlog_errno(status); |
4745 | goto bail; | |
4746 | } | |
ccd979bd MF |
4747 | } |
4748 | ||
dcd0538f | 4749 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
3505bec0 | 4750 | status = ocfs2_do_insert_extent(handle, et, &rec, &insert); |
ccd979bd MF |
4751 | if (status < 0) |
4752 | mlog_errno(status); | |
92ba470c JB |
4753 | else |
4754 | ocfs2_et_extent_map_insert(et, &rec); | |
ccd979bd MF |
4755 | |
4756 | bail: | |
a81cb88b | 4757 | brelse(last_eb_bh); |
ccd979bd | 4758 | |
f56654c4 TM |
4759 | return status; |
4760 | } | |
4761 | ||
0eb8d47e TM |
4762 | /* |
4763 | * Allcate and add clusters into the extent b-tree. | |
4764 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
f99b9b7c | 4765 | * The extent b-tree's root is specified by et, and |
0eb8d47e TM |
4766 | * it is not limited to the file storage. Any extent tree can use this |
4767 | * function if it implements the proper ocfs2_extent_tree. | |
4768 | */ | |
cbee7e1a JB |
4769 | int ocfs2_add_clusters_in_btree(handle_t *handle, |
4770 | struct ocfs2_extent_tree *et, | |
0eb8d47e TM |
4771 | u32 *logical_offset, |
4772 | u32 clusters_to_add, | |
4773 | int mark_unwritten, | |
0eb8d47e TM |
4774 | struct ocfs2_alloc_context *data_ac, |
4775 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 4776 | enum ocfs2_alloc_restarted *reason_ret) |
0eb8d47e | 4777 | { |
a09d09b8 | 4778 | int status = 0, err = 0; |
fb951eb5 | 4779 | int need_free = 0; |
0eb8d47e TM |
4780 | int free_extents; |
4781 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4782 | u32 bit_off, num_bits; | |
4783 | u64 block; | |
4784 | u8 flags = 0; | |
cbee7e1a JB |
4785 | struct ocfs2_super *osb = |
4786 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
0eb8d47e TM |
4787 | |
4788 | BUG_ON(!clusters_to_add); | |
4789 | ||
4790 | if (mark_unwritten) | |
4791 | flags = OCFS2_EXT_UNWRITTEN; | |
4792 | ||
3d03a305 | 4793 | free_extents = ocfs2_num_free_extents(osb, et); |
0eb8d47e TM |
4794 | if (free_extents < 0) { |
4795 | status = free_extents; | |
4796 | mlog_errno(status); | |
4797 | goto leave; | |
4798 | } | |
4799 | ||
4800 | /* there are two cases which could cause us to EAGAIN in the | |
4801 | * we-need-more-metadata case: | |
4802 | * 1) we haven't reserved *any* | |
4803 | * 2) we are so fragmented, we've needed to add metadata too | |
4804 | * many times. */ | |
4805 | if (!free_extents && !meta_ac) { | |
a09d09b8 | 4806 | err = -1; |
0eb8d47e TM |
4807 | status = -EAGAIN; |
4808 | reason = RESTART_META; | |
4809 | goto leave; | |
4810 | } else if ((!free_extents) | |
4811 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
f99b9b7c | 4812 | < ocfs2_extend_meta_needed(et->et_root_el))) { |
a09d09b8 | 4813 | err = -2; |
0eb8d47e TM |
4814 | status = -EAGAIN; |
4815 | reason = RESTART_META; | |
4816 | goto leave; | |
4817 | } | |
4818 | ||
1ed9b777 | 4819 | status = __ocfs2_claim_clusters(handle, data_ac, 1, |
0eb8d47e TM |
4820 | clusters_to_add, &bit_off, &num_bits); |
4821 | if (status < 0) { | |
4822 | if (status != -ENOSPC) | |
4823 | mlog_errno(status); | |
4824 | goto leave; | |
4825 | } | |
4826 | ||
4827 | BUG_ON(num_bits > clusters_to_add); | |
4828 | ||
13723d00 | 4829 | /* reserve our write early -- insert_extent may update the tree root */ |
d9a0a1f8 | 4830 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4831 | OCFS2_JOURNAL_ACCESS_WRITE); |
0eb8d47e TM |
4832 | if (status < 0) { |
4833 | mlog_errno(status); | |
fb951eb5 ZW |
4834 | need_free = 1; |
4835 | goto bail; | |
0eb8d47e TM |
4836 | } |
4837 | ||
4838 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
a09d09b8 TM |
4839 | trace_ocfs2_add_clusters_in_btree( |
4840 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
4841 | bit_off, num_bits); | |
cc79d8c1 | 4842 | status = ocfs2_insert_extent(handle, et, *logical_offset, block, |
f99b9b7c | 4843 | num_bits, flags, meta_ac); |
0eb8d47e TM |
4844 | if (status < 0) { |
4845 | mlog_errno(status); | |
fb951eb5 ZW |
4846 | need_free = 1; |
4847 | goto bail; | |
0eb8d47e TM |
4848 | } |
4849 | ||
ec20cec7 | 4850 | ocfs2_journal_dirty(handle, et->et_root_bh); |
0eb8d47e TM |
4851 | |
4852 | clusters_to_add -= num_bits; | |
4853 | *logical_offset += num_bits; | |
4854 | ||
4855 | if (clusters_to_add) { | |
a09d09b8 | 4856 | err = clusters_to_add; |
0eb8d47e TM |
4857 | status = -EAGAIN; |
4858 | reason = RESTART_TRANS; | |
4859 | } | |
4860 | ||
fb951eb5 ZW |
4861 | bail: |
4862 | if (need_free) { | |
4863 | if (data_ac->ac_which == OCFS2_AC_USE_LOCAL) | |
4864 | ocfs2_free_local_alloc_bits(osb, handle, data_ac, | |
4865 | bit_off, num_bits); | |
4866 | else | |
4867 | ocfs2_free_clusters(handle, | |
4868 | data_ac->ac_inode, | |
4869 | data_ac->ac_bh, | |
4870 | ocfs2_clusters_to_blocks(osb->sb, bit_off), | |
4871 | num_bits); | |
4872 | } | |
4873 | ||
0eb8d47e | 4874 | leave: |
0eb8d47e TM |
4875 | if (reason_ret) |
4876 | *reason_ret = reason; | |
a09d09b8 | 4877 | trace_ocfs2_add_clusters_in_btree_ret(status, reason, err); |
0eb8d47e TM |
4878 | return status; |
4879 | } | |
4880 | ||
328d5752 MF |
4881 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4882 | struct ocfs2_extent_rec *split_rec, | |
4883 | u32 cpos, | |
4884 | struct ocfs2_extent_rec *rec) | |
4885 | { | |
4886 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4887 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4888 | ||
4889 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4890 | ||
4891 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4892 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4893 | ||
4894 | split_rec->e_blkno = rec->e_blkno; | |
4895 | le64_add_cpu(&split_rec->e_blkno, | |
4896 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4897 | ||
4898 | split_rec->e_flags = rec->e_flags; | |
4899 | } | |
4900 | ||
d231129f | 4901 | static int ocfs2_split_and_insert(handle_t *handle, |
e7d4cb6b | 4902 | struct ocfs2_extent_tree *et, |
d231129f | 4903 | struct ocfs2_path *path, |
328d5752 MF |
4904 | struct buffer_head **last_eb_bh, |
4905 | int split_index, | |
4906 | struct ocfs2_extent_rec *orig_split_rec, | |
4907 | struct ocfs2_alloc_context *meta_ac) | |
4908 | { | |
4909 | int ret = 0, depth; | |
4910 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4911 | struct ocfs2_extent_rec tmprec; | |
4912 | struct ocfs2_extent_list *rightmost_el; | |
4913 | struct ocfs2_extent_rec rec; | |
4914 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4915 | struct ocfs2_insert_type insert; | |
4916 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4917 | |
4918 | leftright: | |
4919 | /* | |
4920 | * Store a copy of the record on the stack - it might move | |
4921 | * around as the tree is manipulated below. | |
4922 | */ | |
4923 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4924 | ||
ce1d9ea6 | 4925 | rightmost_el = et->et_root_el; |
328d5752 MF |
4926 | |
4927 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4928 | if (depth) { | |
4929 | BUG_ON(!(*last_eb_bh)); | |
4930 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4931 | rightmost_el = &eb->h_list; | |
4932 | } | |
4933 | ||
4934 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4935 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 4936 | ret = ocfs2_grow_tree(handle, et, |
e7d4cb6b | 4937 | &depth, last_eb_bh, meta_ac); |
328d5752 MF |
4938 | if (ret) { |
4939 | mlog_errno(ret); | |
4940 | goto out; | |
4941 | } | |
328d5752 MF |
4942 | } |
4943 | ||
4944 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4945 | insert.ins_appending = APPEND_NONE; | |
4946 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4947 | insert.ins_tree_depth = depth; |
4948 | ||
4949 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4950 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4951 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4952 | le16_to_cpu(rec.e_leaf_clusters); | |
4953 | ||
4954 | if (split_rec.e_cpos == rec.e_cpos) { | |
4955 | insert.ins_split = SPLIT_LEFT; | |
4956 | } else if (insert_range == rec_range) { | |
4957 | insert.ins_split = SPLIT_RIGHT; | |
4958 | } else { | |
4959 | /* | |
4960 | * Left/right split. We fake this as a right split | |
4961 | * first and then make a second pass as a left split. | |
4962 | */ | |
4963 | insert.ins_split = SPLIT_RIGHT; | |
4964 | ||
d231129f JB |
4965 | ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
4966 | &tmprec, insert_range, &rec); | |
328d5752 MF |
4967 | |
4968 | split_rec = tmprec; | |
4969 | ||
4970 | BUG_ON(do_leftright); | |
4971 | do_leftright = 1; | |
4972 | } | |
4973 | ||
3505bec0 | 4974 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
328d5752 MF |
4975 | if (ret) { |
4976 | mlog_errno(ret); | |
4977 | goto out; | |
4978 | } | |
4979 | ||
4980 | if (do_leftright == 1) { | |
4981 | u32 cpos; | |
4982 | struct ocfs2_extent_list *el; | |
4983 | ||
4984 | do_leftright++; | |
4985 | split_rec = *orig_split_rec; | |
4986 | ||
4987 | ocfs2_reinit_path(path, 1); | |
4988 | ||
4989 | cpos = le32_to_cpu(split_rec.e_cpos); | |
facdb77f | 4990 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
328d5752 MF |
4991 | if (ret) { |
4992 | mlog_errno(ret); | |
4993 | goto out; | |
4994 | } | |
4995 | ||
4996 | el = path_leaf_el(path); | |
4997 | split_index = ocfs2_search_extent_list(el, cpos); | |
981035b4 YX |
4998 | if (split_index == -1) { |
4999 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), | |
7ecef14a JP |
5000 | "Owner %llu has an extent at cpos %u which can no longer be found\n", |
5001 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5002 | cpos); | |
981035b4 YX |
5003 | ret = -EROFS; |
5004 | goto out; | |
5005 | } | |
328d5752 MF |
5006 | goto leftright; |
5007 | } | |
5008 | out: | |
5009 | ||
5010 | return ret; | |
5011 | } | |
5012 | ||
f3868d0f JB |
5013 | static int ocfs2_replace_extent_rec(handle_t *handle, |
5014 | struct ocfs2_extent_tree *et, | |
47be12e4 TM |
5015 | struct ocfs2_path *path, |
5016 | struct ocfs2_extent_list *el, | |
5017 | int split_index, | |
5018 | struct ocfs2_extent_rec *split_rec) | |
5019 | { | |
5020 | int ret; | |
5021 | ||
f3868d0f | 5022 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
47be12e4 TM |
5023 | path_num_items(path) - 1); |
5024 | if (ret) { | |
5025 | mlog_errno(ret); | |
5026 | goto out; | |
5027 | } | |
5028 | ||
5029 | el->l_recs[split_index] = *split_rec; | |
5030 | ||
5031 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5032 | out: | |
5033 | return ret; | |
5034 | } | |
5035 | ||
328d5752 | 5036 | /* |
555936bf TM |
5037 | * Split part or all of the extent record at split_index in the leaf |
5038 | * pointed to by path. Merge with the contiguous extent record if needed. | |
328d5752 MF |
5039 | * |
5040 | * Care is taken to handle contiguousness so as to not grow the tree. | |
5041 | * | |
5042 | * meta_ac is not strictly necessary - we only truly need it if growth | |
5043 | * of the tree is required. All other cases will degrade into a less | |
5044 | * optimal tree layout. | |
5045 | * | |
e7d4cb6b TM |
5046 | * last_eb_bh should be the rightmost leaf block for any extent |
5047 | * btree. Since a split may grow the tree or a merge might shrink it, | |
5048 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
5049 | * |
5050 | * This code is optimized for readability - several passes might be | |
5051 | * made over certain portions of the tree. All of those blocks will | |
5052 | * have been brought into cache (and pinned via the journal), so the | |
5053 | * extra overhead is not expressed in terms of disk reads. | |
5054 | */ | |
e2e9f608 TM |
5055 | int ocfs2_split_extent(handle_t *handle, |
5056 | struct ocfs2_extent_tree *et, | |
5057 | struct ocfs2_path *path, | |
5058 | int split_index, | |
5059 | struct ocfs2_extent_rec *split_rec, | |
5060 | struct ocfs2_alloc_context *meta_ac, | |
5061 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
328d5752 MF |
5062 | { |
5063 | int ret = 0; | |
5064 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 5065 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
5066 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
5067 | struct ocfs2_merge_ctxt ctxt; | |
5068 | struct ocfs2_extent_list *rightmost_el; | |
5069 | ||
328d5752 MF |
5070 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || |
5071 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
5072 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
5073 | ret = -EIO; | |
5074 | mlog_errno(ret); | |
5075 | goto out; | |
5076 | } | |
5077 | ||
9f99ad08 X |
5078 | ret = ocfs2_figure_merge_contig_type(et, path, el, |
5079 | split_index, | |
5080 | split_rec, | |
5081 | &ctxt); | |
5082 | if (ret) { | |
5083 | mlog_errno(ret); | |
5084 | goto out; | |
5085 | } | |
328d5752 MF |
5086 | |
5087 | /* | |
5088 | * The core merge / split code wants to know how much room is | |
a1cf076b | 5089 | * left in this allocation tree, so we pass the |
328d5752 MF |
5090 | * rightmost extent list. |
5091 | */ | |
5092 | if (path->p_tree_depth) { | |
5093 | struct ocfs2_extent_block *eb; | |
328d5752 | 5094 | |
3d03a305 | 5095 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5096 | ocfs2_et_get_last_eb_blk(et), |
5097 | &last_eb_bh); | |
328d5752 | 5098 | if (ret) { |
c1e8d35e | 5099 | mlog_errno(ret); |
328d5752 MF |
5100 | goto out; |
5101 | } | |
5102 | ||
5103 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
328d5752 MF |
5104 | rightmost_el = &eb->h_list; |
5105 | } else | |
5106 | rightmost_el = path_root_el(path); | |
5107 | ||
328d5752 MF |
5108 | if (rec->e_cpos == split_rec->e_cpos && |
5109 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
5110 | ctxt.c_split_covers_rec = 1; | |
5111 | else | |
5112 | ctxt.c_split_covers_rec = 0; | |
5113 | ||
5114 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
5115 | ||
a09d09b8 TM |
5116 | trace_ocfs2_split_extent(split_index, ctxt.c_contig_type, |
5117 | ctxt.c_has_empty_extent, | |
5118 | ctxt.c_split_covers_rec); | |
328d5752 MF |
5119 | |
5120 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
5121 | if (ctxt.c_split_covers_rec) | |
f3868d0f | 5122 | ret = ocfs2_replace_extent_rec(handle, et, path, el, |
47be12e4 | 5123 | split_index, split_rec); |
328d5752 | 5124 | else |
d231129f | 5125 | ret = ocfs2_split_and_insert(handle, et, path, |
328d5752 MF |
5126 | &last_eb_bh, split_index, |
5127 | split_rec, meta_ac); | |
5128 | if (ret) | |
5129 | mlog_errno(ret); | |
5130 | } else { | |
c495dd24 | 5131 | ret = ocfs2_try_to_merge_extent(handle, et, path, |
328d5752 | 5132 | split_index, split_rec, |
c495dd24 | 5133 | dealloc, &ctxt); |
328d5752 MF |
5134 | if (ret) |
5135 | mlog_errno(ret); | |
5136 | } | |
5137 | ||
328d5752 MF |
5138 | out: |
5139 | brelse(last_eb_bh); | |
5140 | return ret; | |
5141 | } | |
5142 | ||
5143 | /* | |
555936bf TM |
5144 | * Change the flags of the already-existing extent at cpos for len clusters. |
5145 | * | |
5146 | * new_flags: the flags we want to set. | |
5147 | * clear_flags: the flags we want to clear. | |
5148 | * phys: the new physical offset we want this new extent starts from. | |
328d5752 MF |
5149 | * |
5150 | * If the existing extent is larger than the request, initiate a | |
5151 | * split. An attempt will be made at merging with adjacent extents. | |
5152 | * | |
5153 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5154 | */ | |
1aa75fea TM |
5155 | int ocfs2_change_extent_flag(handle_t *handle, |
5156 | struct ocfs2_extent_tree *et, | |
5157 | u32 cpos, u32 len, u32 phys, | |
5158 | struct ocfs2_alloc_context *meta_ac, | |
5159 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
5160 | int new_flags, int clear_flags) | |
328d5752 MF |
5161 | { |
5162 | int ret, index; | |
555936bf TM |
5163 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
5164 | u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys); | |
328d5752 MF |
5165 | struct ocfs2_extent_rec split_rec; |
5166 | struct ocfs2_path *left_path = NULL; | |
5167 | struct ocfs2_extent_list *el; | |
555936bf | 5168 | struct ocfs2_extent_rec *rec; |
328d5752 | 5169 | |
ffdd7a54 | 5170 | left_path = ocfs2_new_path_from_et(et); |
328d5752 MF |
5171 | if (!left_path) { |
5172 | ret = -ENOMEM; | |
5173 | mlog_errno(ret); | |
5174 | goto out; | |
5175 | } | |
5176 | ||
facdb77f | 5177 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
5178 | if (ret) { |
5179 | mlog_errno(ret); | |
5180 | goto out; | |
5181 | } | |
5182 | el = path_leaf_el(left_path); | |
5183 | ||
5184 | index = ocfs2_search_extent_list(el, cpos); | |
981035b4 | 5185 | if (index == -1) { |
555936bf | 5186 | ocfs2_error(sb, |
7ecef14a JP |
5187 | "Owner %llu has an extent at cpos %u which can no longer be found\n", |
5188 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5189 | cpos); | |
328d5752 MF |
5190 | ret = -EROFS; |
5191 | goto out; | |
5192 | } | |
5193 | ||
555936bf TM |
5194 | ret = -EIO; |
5195 | rec = &el->l_recs[index]; | |
5196 | if (new_flags && (rec->e_flags & new_flags)) { | |
5197 | mlog(ML_ERROR, "Owner %llu tried to set %d flags on an " | |
86544fbd | 5198 | "extent that already had them\n", |
555936bf TM |
5199 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5200 | new_flags); | |
5201 | goto out; | |
5202 | } | |
5203 | ||
5204 | if (clear_flags && !(rec->e_flags & clear_flags)) { | |
5205 | mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an " | |
86544fbd | 5206 | "extent that didn't have them\n", |
555936bf TM |
5207 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5208 | clear_flags); | |
5209 | goto out; | |
5210 | } | |
5211 | ||
328d5752 MF |
5212 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); |
5213 | split_rec.e_cpos = cpu_to_le32(cpos); | |
5214 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
5215 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
555936bf TM |
5216 | split_rec.e_flags = rec->e_flags; |
5217 | if (new_flags) | |
5218 | split_rec.e_flags |= new_flags; | |
5219 | if (clear_flags) | |
5220 | split_rec.e_flags &= ~clear_flags; | |
5221 | ||
e2e9f608 TM |
5222 | ret = ocfs2_split_extent(handle, et, left_path, |
5223 | index, &split_rec, meta_ac, | |
5224 | dealloc); | |
328d5752 MF |
5225 | if (ret) |
5226 | mlog_errno(ret); | |
5227 | ||
5228 | out: | |
5229 | ocfs2_free_path(left_path); | |
5230 | return ret; | |
555936bf TM |
5231 | |
5232 | } | |
5233 | ||
5234 | /* | |
5235 | * Mark the already-existing extent at cpos as written for len clusters. | |
5236 | * This removes the unwritten extent flag. | |
5237 | * | |
5238 | * If the existing extent is larger than the request, initiate a | |
5239 | * split. An attempt will be made at merging with adjacent extents. | |
5240 | * | |
5241 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5242 | */ | |
5243 | int ocfs2_mark_extent_written(struct inode *inode, | |
5244 | struct ocfs2_extent_tree *et, | |
5245 | handle_t *handle, u32 cpos, u32 len, u32 phys, | |
5246 | struct ocfs2_alloc_context *meta_ac, | |
5247 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5248 | { | |
5249 | int ret; | |
5250 | ||
a09d09b8 TM |
5251 | trace_ocfs2_mark_extent_written( |
5252 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5253 | cpos, len, phys); | |
555936bf TM |
5254 | |
5255 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { | |
7ecef14a | 5256 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents that are being written to, but the feature bit is not set in the super block\n", |
555936bf TM |
5257 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
5258 | ret = -EROFS; | |
5259 | goto out; | |
5260 | } | |
5261 | ||
5262 | /* | |
5263 | * XXX: This should be fixed up so that we just re-insert the | |
5264 | * next extent records. | |
5265 | */ | |
5266 | ocfs2_et_extent_map_truncate(et, 0); | |
5267 | ||
5268 | ret = ocfs2_change_extent_flag(handle, et, cpos, | |
5269 | len, phys, meta_ac, dealloc, | |
5270 | 0, OCFS2_EXT_UNWRITTEN); | |
5271 | if (ret) | |
5272 | mlog_errno(ret); | |
5273 | ||
5274 | out: | |
5275 | return ret; | |
328d5752 MF |
5276 | } |
5277 | ||
dbdcf6a4 JB |
5278 | static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
5279 | struct ocfs2_path *path, | |
d0c7d708 MF |
5280 | int index, u32 new_range, |
5281 | struct ocfs2_alloc_context *meta_ac) | |
5282 | { | |
c901fb00 | 5283 | int ret, depth, credits; |
d0c7d708 MF |
5284 | struct buffer_head *last_eb_bh = NULL; |
5285 | struct ocfs2_extent_block *eb; | |
5286 | struct ocfs2_extent_list *rightmost_el, *el; | |
5287 | struct ocfs2_extent_rec split_rec; | |
5288 | struct ocfs2_extent_rec *rec; | |
5289 | struct ocfs2_insert_type insert; | |
5290 | ||
5291 | /* | |
5292 | * Setup the record to split before we grow the tree. | |
5293 | */ | |
5294 | el = path_leaf_el(path); | |
5295 | rec = &el->l_recs[index]; | |
dbdcf6a4 JB |
5296 | ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
5297 | &split_rec, new_range, rec); | |
d0c7d708 MF |
5298 | |
5299 | depth = path->p_tree_depth; | |
5300 | if (depth > 0) { | |
3d03a305 | 5301 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5302 | ocfs2_et_get_last_eb_blk(et), |
5303 | &last_eb_bh); | |
d0c7d708 MF |
5304 | if (ret < 0) { |
5305 | mlog_errno(ret); | |
5306 | goto out; | |
5307 | } | |
5308 | ||
5309 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
5310 | rightmost_el = &eb->h_list; | |
5311 | } else | |
5312 | rightmost_el = path_leaf_el(path); | |
5313 | ||
c901fb00 TM |
5314 | credits = path->p_tree_depth + |
5315 | ocfs2_extend_meta_needed(et->et_root_el); | |
d0c7d708 MF |
5316 | ret = ocfs2_extend_trans(handle, credits); |
5317 | if (ret) { | |
5318 | mlog_errno(ret); | |
5319 | goto out; | |
5320 | } | |
5321 | ||
5322 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
5323 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 5324 | ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
5325 | meta_ac); |
5326 | if (ret) { | |
5327 | mlog_errno(ret); | |
5328 | goto out; | |
5329 | } | |
d0c7d708 MF |
5330 | } |
5331 | ||
5332 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
5333 | insert.ins_appending = APPEND_NONE; | |
5334 | insert.ins_contig = CONTIG_NONE; | |
5335 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
5336 | insert.ins_tree_depth = depth; |
5337 | ||
3505bec0 | 5338 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
d0c7d708 MF |
5339 | if (ret) |
5340 | mlog_errno(ret); | |
5341 | ||
5342 | out: | |
5343 | brelse(last_eb_bh); | |
5344 | return ret; | |
5345 | } | |
5346 | ||
043beebb JB |
5347 | static int ocfs2_truncate_rec(handle_t *handle, |
5348 | struct ocfs2_extent_tree *et, | |
d0c7d708 MF |
5349 | struct ocfs2_path *path, int index, |
5350 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
043beebb | 5351 | u32 cpos, u32 len) |
d0c7d708 MF |
5352 | { |
5353 | int ret; | |
5354 | u32 left_cpos, rec_range, trunc_range; | |
c14688ea | 5355 | int is_rightmost_tree_rec = 0; |
043beebb | 5356 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
d0c7d708 MF |
5357 | struct ocfs2_path *left_path = NULL; |
5358 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
5359 | struct ocfs2_extent_rec *rec; | |
5360 | struct ocfs2_extent_block *eb; | |
5361 | ||
5362 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
17215989 X |
5363 | /* extend credit for ocfs2_remove_rightmost_path */ |
5364 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5365 | handle->h_buffer_credits, | |
5366 | path); | |
5367 | if (ret) { | |
5368 | mlog_errno(ret); | |
5369 | goto out; | |
5370 | } | |
5371 | ||
70f18c08 | 5372 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5373 | if (ret) { |
5374 | mlog_errno(ret); | |
5375 | goto out; | |
5376 | } | |
5377 | ||
5378 | index--; | |
5379 | } | |
5380 | ||
5381 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
5382 | path->p_tree_depth) { | |
5383 | /* | |
5384 | * Check whether this is the rightmost tree record. If | |
5385 | * we remove all of this record or part of its right | |
5386 | * edge then an update of the record lengths above it | |
5387 | * will be required. | |
5388 | */ | |
5389 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
5390 | if (eb->h_next_leaf_blk == 0) | |
5391 | is_rightmost_tree_rec = 1; | |
5392 | } | |
5393 | ||
5394 | rec = &el->l_recs[index]; | |
5395 | if (index == 0 && path->p_tree_depth && | |
5396 | le32_to_cpu(rec->e_cpos) == cpos) { | |
5397 | /* | |
5398 | * Changing the leftmost offset (via partial or whole | |
5399 | * record truncate) of an interior (or rightmost) path | |
5400 | * means we have to update the subtree that is formed | |
5401 | * by this leaf and the one to it's left. | |
5402 | * | |
5403 | * There are two cases we can skip: | |
043beebb | 5404 | * 1) Path is the leftmost one in our btree. |
d0c7d708 MF |
5405 | * 2) The leaf is rightmost and will be empty after |
5406 | * we remove the extent record - the rotate code | |
5407 | * knows how to update the newly formed edge. | |
5408 | */ | |
5409 | ||
043beebb | 5410 | ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
d0c7d708 MF |
5411 | if (ret) { |
5412 | mlog_errno(ret); | |
5413 | goto out; | |
5414 | } | |
5415 | ||
5416 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
ffdd7a54 | 5417 | left_path = ocfs2_new_path_from_path(path); |
d0c7d708 MF |
5418 | if (!left_path) { |
5419 | ret = -ENOMEM; | |
5420 | mlog_errno(ret); | |
5421 | goto out; | |
5422 | } | |
5423 | ||
facdb77f JB |
5424 | ret = ocfs2_find_path(et->et_ci, left_path, |
5425 | left_cpos); | |
d0c7d708 MF |
5426 | if (ret) { |
5427 | mlog_errno(ret); | |
5428 | goto out; | |
5429 | } | |
5430 | } | |
5431 | } | |
5432 | ||
5433 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5434 | handle->h_buffer_credits, | |
5435 | path); | |
5436 | if (ret) { | |
5437 | mlog_errno(ret); | |
5438 | goto out; | |
5439 | } | |
5440 | ||
d9a0a1f8 | 5441 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
d0c7d708 MF |
5442 | if (ret) { |
5443 | mlog_errno(ret); | |
5444 | goto out; | |
5445 | } | |
5446 | ||
d9a0a1f8 | 5447 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
d0c7d708 MF |
5448 | if (ret) { |
5449 | mlog_errno(ret); | |
5450 | goto out; | |
5451 | } | |
5452 | ||
5453 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5454 | trunc_range = cpos + len; | |
5455 | ||
5456 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5457 | int next_free; | |
5458 | ||
5459 | memset(rec, 0, sizeof(*rec)); | |
5460 | ocfs2_cleanup_merge(el, index); | |
d0c7d708 MF |
5461 | |
5462 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5463 | if (is_rightmost_tree_rec && next_free > 1) { | |
5464 | /* | |
5465 | * We skip the edge update if this path will | |
5466 | * be deleted by the rotate code. | |
5467 | */ | |
5468 | rec = &el->l_recs[next_free - 1]; | |
d401dc12 | 5469 | ocfs2_adjust_rightmost_records(handle, et, path, |
d0c7d708 MF |
5470 | rec); |
5471 | } | |
5472 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5473 | /* Remove leftmost portion of the record. */ | |
5474 | le32_add_cpu(&rec->e_cpos, len); | |
5475 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5476 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5477 | } else if (rec_range == trunc_range) { | |
5478 | /* Remove rightmost portion of the record */ | |
5479 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5480 | if (is_rightmost_tree_rec) | |
d401dc12 | 5481 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
d0c7d708 MF |
5482 | } else { |
5483 | /* Caller should have trapped this. */ | |
043beebb JB |
5484 | mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) " |
5485 | "(%u, %u)\n", | |
5486 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
d0c7d708 MF |
5487 | le32_to_cpu(rec->e_cpos), |
5488 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5489 | BUG(); | |
5490 | } | |
5491 | ||
5492 | if (left_path) { | |
5493 | int subtree_index; | |
5494 | ||
7dc02805 | 5495 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
4619c73e | 5496 | ocfs2_complete_edge_insert(handle, left_path, path, |
d0c7d708 MF |
5497 | subtree_index); |
5498 | } | |
5499 | ||
5500 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5501 | ||
70f18c08 | 5502 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5503 | if (ret) { |
5504 | mlog_errno(ret); | |
5505 | goto out; | |
5506 | } | |
5507 | ||
5508 | out: | |
5509 | ocfs2_free_path(left_path); | |
5510 | return ret; | |
5511 | } | |
5512 | ||
dbdcf6a4 | 5513 | int ocfs2_remove_extent(handle_t *handle, |
f99b9b7c | 5514 | struct ocfs2_extent_tree *et, |
dbdcf6a4 | 5515 | u32 cpos, u32 len, |
063c4561 | 5516 | struct ocfs2_alloc_context *meta_ac, |
f99b9b7c | 5517 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
d0c7d708 MF |
5518 | { |
5519 | int ret, index; | |
5520 | u32 rec_range, trunc_range; | |
5521 | struct ocfs2_extent_rec *rec; | |
5522 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 5523 | struct ocfs2_path *path = NULL; |
d0c7d708 | 5524 | |
4c911eef JB |
5525 | /* |
5526 | * XXX: Why are we truncating to 0 instead of wherever this | |
5527 | * affects us? | |
5528 | */ | |
5529 | ocfs2_et_extent_map_truncate(et, 0); | |
d0c7d708 | 5530 | |
ffdd7a54 | 5531 | path = ocfs2_new_path_from_et(et); |
d0c7d708 MF |
5532 | if (!path) { |
5533 | ret = -ENOMEM; | |
5534 | mlog_errno(ret); | |
5535 | goto out; | |
5536 | } | |
5537 | ||
facdb77f | 5538 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5539 | if (ret) { |
5540 | mlog_errno(ret); | |
5541 | goto out; | |
5542 | } | |
5543 | ||
5544 | el = path_leaf_el(path); | |
5545 | index = ocfs2_search_extent_list(el, cpos); | |
981035b4 | 5546 | if (index == -1) { |
dbdcf6a4 | 5547 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
7ecef14a | 5548 | "Owner %llu has an extent at cpos %u which can no longer be found\n", |
dbdcf6a4 JB |
5549 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5550 | cpos); | |
d0c7d708 MF |
5551 | ret = -EROFS; |
5552 | goto out; | |
5553 | } | |
5554 | ||
5555 | /* | |
5556 | * We have 3 cases of extent removal: | |
5557 | * 1) Range covers the entire extent rec | |
5558 | * 2) Range begins or ends on one edge of the extent rec | |
5559 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5560 | * | |
5561 | * For case 1 we remove the extent rec and left rotate to | |
5562 | * fill the hole. | |
5563 | * | |
5564 | * For case 2 we just shrink the existing extent rec, with a | |
5565 | * tree update if the shrinking edge is also the edge of an | |
5566 | * extent block. | |
5567 | * | |
5568 | * For case 3 we do a right split to turn the extent rec into | |
5569 | * something case 2 can handle. | |
5570 | */ | |
5571 | rec = &el->l_recs[index]; | |
5572 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5573 | trunc_range = cpos + len; | |
5574 | ||
5575 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5576 | ||
a09d09b8 TM |
5577 | trace_ocfs2_remove_extent( |
5578 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5579 | cpos, len, index, le32_to_cpu(rec->e_cpos), | |
5580 | ocfs2_rec_clusters(el, rec)); | |
d0c7d708 MF |
5581 | |
5582 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
043beebb JB |
5583 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5584 | cpos, len); | |
d0c7d708 MF |
5585 | if (ret) { |
5586 | mlog_errno(ret); | |
5587 | goto out; | |
5588 | } | |
5589 | } else { | |
dbdcf6a4 | 5590 | ret = ocfs2_split_tree(handle, et, path, index, |
d0c7d708 MF |
5591 | trunc_range, meta_ac); |
5592 | if (ret) { | |
5593 | mlog_errno(ret); | |
5594 | goto out; | |
5595 | } | |
5596 | ||
5597 | /* | |
5598 | * The split could have manipulated the tree enough to | |
5599 | * move the record location, so we have to look for it again. | |
5600 | */ | |
5601 | ocfs2_reinit_path(path, 1); | |
5602 | ||
facdb77f | 5603 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5604 | if (ret) { |
5605 | mlog_errno(ret); | |
5606 | goto out; | |
5607 | } | |
5608 | ||
5609 | el = path_leaf_el(path); | |
5610 | index = ocfs2_search_extent_list(el, cpos); | |
981035b4 | 5611 | if (index == -1) { |
dbdcf6a4 | 5612 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
7ecef14a | 5613 | "Owner %llu: split at cpos %u lost record\n", |
dbdcf6a4 | 5614 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
d0c7d708 MF |
5615 | cpos); |
5616 | ret = -EROFS; | |
5617 | goto out; | |
5618 | } | |
5619 | ||
5620 | /* | |
5621 | * Double check our values here. If anything is fishy, | |
5622 | * it's easier to catch it at the top level. | |
5623 | */ | |
5624 | rec = &el->l_recs[index]; | |
5625 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5626 | ocfs2_rec_clusters(el, rec); | |
5627 | if (rec_range != trunc_range) { | |
dbdcf6a4 | 5628 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
7ecef14a | 5629 | "Owner %llu: error after split at cpos %u trunc len %u, existing record is (%u,%u)\n", |
dbdcf6a4 | 5630 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
d0c7d708 MF |
5631 | cpos, len, le32_to_cpu(rec->e_cpos), |
5632 | ocfs2_rec_clusters(el, rec)); | |
5633 | ret = -EROFS; | |
5634 | goto out; | |
5635 | } | |
5636 | ||
043beebb JB |
5637 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5638 | cpos, len); | |
d0c7d708 MF |
5639 | if (ret) { |
5640 | mlog_errno(ret); | |
5641 | goto out; | |
5642 | } | |
5643 | } | |
5644 | ||
5645 | out: | |
5646 | ocfs2_free_path(path); | |
5647 | return ret; | |
5648 | } | |
5649 | ||
78f94673 TY |
5650 | /* |
5651 | * ocfs2_reserve_blocks_for_rec_trunc() would look basically the | |
5652 | * same as ocfs2_lock_alloctors(), except for it accepts a blocks | |
5653 | * number to reserve some extra blocks, and it only handles meta | |
5654 | * data allocations. | |
5655 | * | |
5656 | * Currently, only ocfs2_remove_btree_range() uses it for truncating | |
5657 | * and punching holes. | |
5658 | */ | |
5659 | static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode, | |
5660 | struct ocfs2_extent_tree *et, | |
5661 | u32 extents_to_split, | |
5662 | struct ocfs2_alloc_context **ac, | |
5663 | int extra_blocks) | |
5664 | { | |
5665 | int ret = 0, num_free_extents; | |
5666 | unsigned int max_recs_needed = 2 * extents_to_split; | |
5667 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5668 | ||
5669 | *ac = NULL; | |
5670 | ||
5671 | num_free_extents = ocfs2_num_free_extents(osb, et); | |
5672 | if (num_free_extents < 0) { | |
5673 | ret = num_free_extents; | |
5674 | mlog_errno(ret); | |
5675 | goto out; | |
5676 | } | |
5677 | ||
5678 | if (!num_free_extents || | |
5679 | (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) | |
5680 | extra_blocks += ocfs2_extend_meta_needed(et->et_root_el); | |
5681 | ||
5682 | if (extra_blocks) { | |
5683 | ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac); | |
5684 | if (ret < 0) { | |
5685 | if (ret != -ENOSPC) | |
5686 | mlog_errno(ret); | |
5687 | goto out; | |
5688 | } | |
5689 | } | |
5690 | ||
5691 | out: | |
5692 | if (ret) { | |
5693 | if (*ac) { | |
5694 | ocfs2_free_alloc_context(*ac); | |
5695 | *ac = NULL; | |
5696 | } | |
5697 | } | |
5698 | ||
5699 | return ret; | |
5700 | } | |
5701 | ||
fecc0112 MF |
5702 | int ocfs2_remove_btree_range(struct inode *inode, |
5703 | struct ocfs2_extent_tree *et, | |
78f94673 TY |
5704 | u32 cpos, u32 phys_cpos, u32 len, int flags, |
5705 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
f62f12b3 | 5706 | u64 refcount_loc, bool refcount_tree_locked) |
fecc0112 | 5707 | { |
78f94673 | 5708 | int ret, credits = 0, extra_blocks = 0; |
fecc0112 MF |
5709 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); |
5710 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5711 | struct inode *tl_inode = osb->osb_tl_inode; | |
5712 | handle_t *handle; | |
5713 | struct ocfs2_alloc_context *meta_ac = NULL; | |
78f94673 TY |
5714 | struct ocfs2_refcount_tree *ref_tree = NULL; |
5715 | ||
5716 | if ((flags & OCFS2_EXT_REFCOUNTED) && len) { | |
84e40080 | 5717 | BUG_ON(!ocfs2_is_refcount_inode(inode)); |
78f94673 | 5718 | |
f62f12b3 JB |
5719 | if (!refcount_tree_locked) { |
5720 | ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1, | |
5721 | &ref_tree, NULL); | |
5722 | if (ret) { | |
5723 | mlog_errno(ret); | |
5724 | goto bail; | |
5725 | } | |
78f94673 TY |
5726 | } |
5727 | ||
5728 | ret = ocfs2_prepare_refcount_change_for_del(inode, | |
5729 | refcount_loc, | |
5730 | phys_blkno, | |
5731 | len, | |
5732 | &credits, | |
5733 | &extra_blocks); | |
5734 | if (ret < 0) { | |
5735 | mlog_errno(ret); | |
33add0e3 | 5736 | goto bail; |
78f94673 TY |
5737 | } |
5738 | } | |
fecc0112 | 5739 | |
78f94673 TY |
5740 | ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac, |
5741 | extra_blocks); | |
fecc0112 MF |
5742 | if (ret) { |
5743 | mlog_errno(ret); | |
33add0e3 | 5744 | goto bail; |
fecc0112 MF |
5745 | } |
5746 | ||
5955102c | 5747 | inode_lock(tl_inode); |
fecc0112 MF |
5748 | |
5749 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5750 | ret = __ocfs2_flush_truncate_log(osb); | |
5751 | if (ret < 0) { | |
5752 | mlog_errno(ret); | |
5753 | goto out; | |
5754 | } | |
5755 | } | |
5756 | ||
78f94673 TY |
5757 | handle = ocfs2_start_trans(osb, |
5758 | ocfs2_remove_extent_credits(osb->sb) + credits); | |
fecc0112 MF |
5759 | if (IS_ERR(handle)) { |
5760 | ret = PTR_ERR(handle); | |
5761 | mlog_errno(ret); | |
5762 | goto out; | |
5763 | } | |
5764 | ||
d9a0a1f8 | 5765 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 5766 | OCFS2_JOURNAL_ACCESS_WRITE); |
fecc0112 MF |
5767 | if (ret) { |
5768 | mlog_errno(ret); | |
b8a0ae57 | 5769 | goto out_commit; |
fecc0112 MF |
5770 | } |
5771 | ||
5dd4056d | 5772 | dquot_free_space_nodirty(inode, |
fd4ef231 MF |
5773 | ocfs2_clusters_to_bytes(inode->i_sb, len)); |
5774 | ||
dbdcf6a4 | 5775 | ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc); |
fecc0112 MF |
5776 | if (ret) { |
5777 | mlog_errno(ret); | |
5778 | goto out_commit; | |
5779 | } | |
5780 | ||
6136ca5f | 5781 | ocfs2_et_update_clusters(et, -len); |
6fdb702d | 5782 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
fecc0112 | 5783 | |
ec20cec7 | 5784 | ocfs2_journal_dirty(handle, et->et_root_bh); |
fecc0112 | 5785 | |
78f94673 TY |
5786 | if (phys_blkno) { |
5787 | if (flags & OCFS2_EXT_REFCOUNTED) | |
5788 | ret = ocfs2_decrease_refcount(inode, handle, | |
5789 | ocfs2_blocks_to_clusters(osb->sb, | |
5790 | phys_blkno), | |
5791 | len, meta_ac, | |
5792 | dealloc, 1); | |
5793 | else | |
5794 | ret = ocfs2_truncate_log_append(osb, handle, | |
5795 | phys_blkno, len); | |
5796 | if (ret) | |
5797 | mlog_errno(ret); | |
5798 | ||
5799 | } | |
fecc0112 MF |
5800 | |
5801 | out_commit: | |
5802 | ocfs2_commit_trans(osb, handle); | |
5803 | out: | |
5955102c | 5804 | inode_unlock(tl_inode); |
33add0e3 | 5805 | bail: |
fecc0112 MF |
5806 | if (meta_ac) |
5807 | ocfs2_free_alloc_context(meta_ac); | |
5808 | ||
78f94673 TY |
5809 | if (ref_tree) |
5810 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
5811 | ||
fecc0112 MF |
5812 | return ret; |
5813 | } | |
5814 | ||
063c4561 | 5815 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5816 | { |
5817 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5818 | struct ocfs2_dinode *di; | |
5819 | struct ocfs2_truncate_log *tl; | |
5820 | ||
5821 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5822 | tl = &di->id2.i_dealloc; | |
5823 | ||
5824 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5825 | "slot %d, invalid truncate log parameters: used = " | |
5826 | "%u, count = %u\n", osb->slot_num, | |
5827 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5828 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5829 | } | |
5830 | ||
5831 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5832 | unsigned int new_start) | |
5833 | { | |
5834 | unsigned int tail_index; | |
5835 | unsigned int current_tail; | |
5836 | ||
5837 | /* No records, nothing to coalesce */ | |
5838 | if (!le16_to_cpu(tl->tl_used)) | |
5839 | return 0; | |
5840 | ||
5841 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5842 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5843 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5844 | ||
5845 | return current_tail == new_start; | |
5846 | } | |
5847 | ||
063c4561 MF |
5848 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5849 | handle_t *handle, | |
5850 | u64 start_blk, | |
5851 | unsigned int num_clusters) | |
ccd979bd MF |
5852 | { |
5853 | int status, index; | |
5854 | unsigned int start_cluster, tl_count; | |
5855 | struct inode *tl_inode = osb->osb_tl_inode; | |
5856 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5857 | struct ocfs2_dinode *di; | |
5858 | struct ocfs2_truncate_log *tl; | |
5859 | ||
5955102c | 5860 | BUG_ON(inode_trylock(tl_inode)); |
ccd979bd MF |
5861 | |
5862 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5863 | ||
5864 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5865 | |
10995aa2 JB |
5866 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5867 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5868 | * corruption is a code bug */ | |
5869 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5870 | ||
5871 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5872 | tl_count = le16_to_cpu(tl->tl_count); |
5873 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5874 | tl_count == 0, | |
b0697053 MF |
5875 | "Truncate record count on #%llu invalid " |
5876 | "wanted %u, actual %u\n", | |
5877 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5878 | ocfs2_truncate_recs_per_inode(osb->sb), |
5879 | le16_to_cpu(tl->tl_count)); | |
5880 | ||
5881 | /* Caller should have known to flush before calling us. */ | |
5882 | index = le16_to_cpu(tl->tl_used); | |
5883 | if (index >= tl_count) { | |
5884 | status = -ENOSPC; | |
5885 | mlog_errno(status); | |
5886 | goto bail; | |
5887 | } | |
5888 | ||
0cf2f763 | 5889 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5890 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5891 | if (status < 0) { |
5892 | mlog_errno(status); | |
5893 | goto bail; | |
5894 | } | |
5895 | ||
a09d09b8 TM |
5896 | trace_ocfs2_truncate_log_append( |
5897 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index, | |
5898 | start_cluster, num_clusters); | |
ccd979bd MF |
5899 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { |
5900 | /* | |
5901 | * Move index back to the record we are coalescing with. | |
5902 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5903 | */ | |
5904 | index--; | |
5905 | ||
5906 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
a09d09b8 TM |
5907 | trace_ocfs2_truncate_log_append( |
5908 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
5909 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5910 | num_clusters); | |
ccd979bd MF |
5911 | } else { |
5912 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5913 | tl->tl_used = cpu_to_le16(index + 1); | |
5914 | } | |
5915 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5916 | ||
ec20cec7 | 5917 | ocfs2_journal_dirty(handle, tl_bh); |
ccd979bd | 5918 | |
50308d81 | 5919 | osb->truncated_clusters += num_clusters; |
ccd979bd | 5920 | bail: |
ccd979bd MF |
5921 | return status; |
5922 | } | |
5923 | ||
5924 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
ccd979bd MF |
5925 | struct inode *data_alloc_inode, |
5926 | struct buffer_head *data_alloc_bh) | |
5927 | { | |
5928 | int status = 0; | |
5929 | int i; | |
5930 | unsigned int num_clusters; | |
5931 | u64 start_blk; | |
5932 | struct ocfs2_truncate_rec rec; | |
5933 | struct ocfs2_dinode *di; | |
5934 | struct ocfs2_truncate_log *tl; | |
5935 | struct inode *tl_inode = osb->osb_tl_inode; | |
5936 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
2b0ad008 | 5937 | handle_t *handle; |
ccd979bd | 5938 | |
ccd979bd MF |
5939 | di = (struct ocfs2_dinode *) tl_bh->b_data; |
5940 | tl = &di->id2.i_dealloc; | |
5941 | i = le16_to_cpu(tl->tl_used) - 1; | |
5942 | while (i >= 0) { | |
2b0ad008 JB |
5943 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); |
5944 | if (IS_ERR(handle)) { | |
5945 | status = PTR_ERR(handle); | |
5946 | mlog_errno(status); | |
5947 | goto bail; | |
5948 | } | |
5949 | ||
ccd979bd MF |
5950 | /* Caller has given us at least enough credits to |
5951 | * update the truncate log dinode */ | |
0cf2f763 | 5952 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5953 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5954 | if (status < 0) { |
5955 | mlog_errno(status); | |
5956 | goto bail; | |
5957 | } | |
5958 | ||
5959 | tl->tl_used = cpu_to_le16(i); | |
5960 | ||
ec20cec7 | 5961 | ocfs2_journal_dirty(handle, tl_bh); |
ccd979bd | 5962 | |
ccd979bd MF |
5963 | rec = tl->tl_recs[i]; |
5964 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5965 | le32_to_cpu(rec.t_start)); | |
5966 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5967 | ||
5968 | /* if start_blk is not set, we ignore the record as | |
5969 | * invalid. */ | |
5970 | if (start_blk) { | |
a09d09b8 TM |
5971 | trace_ocfs2_replay_truncate_records( |
5972 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
5973 | i, le32_to_cpu(rec.t_start), num_clusters); | |
ccd979bd MF |
5974 | |
5975 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5976 | data_alloc_bh, start_blk, | |
5977 | num_clusters); | |
5978 | if (status < 0) { | |
5979 | mlog_errno(status); | |
5980 | goto bail; | |
5981 | } | |
5982 | } | |
102c2595 | 5983 | |
2b0ad008 | 5984 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
5985 | i--; |
5986 | } | |
5987 | ||
50308d81 TM |
5988 | osb->truncated_clusters = 0; |
5989 | ||
ccd979bd | 5990 | bail: |
ccd979bd MF |
5991 | return status; |
5992 | } | |
5993 | ||
1b1dcc1b | 5994 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5995 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5996 | { |
5997 | int status; | |
5998 | unsigned int num_to_flush; | |
ccd979bd MF |
5999 | struct inode *tl_inode = osb->osb_tl_inode; |
6000 | struct inode *data_alloc_inode = NULL; | |
6001 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
6002 | struct buffer_head *data_alloc_bh = NULL; | |
6003 | struct ocfs2_dinode *di; | |
6004 | struct ocfs2_truncate_log *tl; | |
6005 | ||
5955102c | 6006 | BUG_ON(inode_trylock(tl_inode)); |
ccd979bd MF |
6007 | |
6008 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 6009 | |
10995aa2 JB |
6010 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
6011 | * by the underlying call to ocfs2_read_inode_block(), so any | |
6012 | * corruption is a code bug */ | |
6013 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
6014 | ||
6015 | tl = &di->id2.i_dealloc; | |
ccd979bd | 6016 | num_to_flush = le16_to_cpu(tl->tl_used); |
a09d09b8 TM |
6017 | trace_ocfs2_flush_truncate_log( |
6018 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
6019 | num_to_flush); | |
ccd979bd MF |
6020 | if (!num_to_flush) { |
6021 | status = 0; | |
e08dc8b9 | 6022 | goto out; |
ccd979bd MF |
6023 | } |
6024 | ||
6025 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
6026 | GLOBAL_BITMAP_SYSTEM_INODE, | |
6027 | OCFS2_INVALID_SLOT); | |
6028 | if (!data_alloc_inode) { | |
6029 | status = -EINVAL; | |
6030 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 6031 | goto out; |
ccd979bd MF |
6032 | } |
6033 | ||
5955102c | 6034 | inode_lock(data_alloc_inode); |
e08dc8b9 | 6035 | |
e63aecb6 | 6036 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
6037 | if (status < 0) { |
6038 | mlog_errno(status); | |
e08dc8b9 | 6039 | goto out_mutex; |
ccd979bd MF |
6040 | } |
6041 | ||
2b0ad008 | 6042 | status = ocfs2_replay_truncate_records(osb, data_alloc_inode, |
ccd979bd | 6043 | data_alloc_bh); |
e08dc8b9 | 6044 | if (status < 0) |
ccd979bd | 6045 | mlog_errno(status); |
ccd979bd | 6046 | |
e08dc8b9 | 6047 | brelse(data_alloc_bh); |
e63aecb6 | 6048 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 6049 | |
e08dc8b9 | 6050 | out_mutex: |
5955102c | 6051 | inode_unlock(data_alloc_inode); |
e08dc8b9 | 6052 | iput(data_alloc_inode); |
ccd979bd | 6053 | |
e08dc8b9 | 6054 | out: |
ccd979bd MF |
6055 | return status; |
6056 | } | |
6057 | ||
6058 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
6059 | { | |
6060 | int status; | |
6061 | struct inode *tl_inode = osb->osb_tl_inode; | |
6062 | ||
5955102c | 6063 | inode_lock(tl_inode); |
ccd979bd | 6064 | status = __ocfs2_flush_truncate_log(osb); |
5955102c | 6065 | inode_unlock(tl_inode); |
ccd979bd MF |
6066 | |
6067 | return status; | |
6068 | } | |
6069 | ||
c4028958 | 6070 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
6071 | { |
6072 | int status; | |
c4028958 DH |
6073 | struct ocfs2_super *osb = |
6074 | container_of(work, struct ocfs2_super, | |
6075 | osb_truncate_log_wq.work); | |
ccd979bd | 6076 | |
ccd979bd MF |
6077 | status = ocfs2_flush_truncate_log(osb); |
6078 | if (status < 0) | |
6079 | mlog_errno(status); | |
4d0ddb2c | 6080 | else |
b89c5428 | 6081 | ocfs2_init_steal_slots(osb); |
ccd979bd MF |
6082 | } |
6083 | ||
6084 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
6085 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
6086 | int cancel) | |
6087 | { | |
a9e9acae X |
6088 | if (osb->osb_tl_inode && |
6089 | atomic_read(&osb->osb_tl_disable) == 0) { | |
ccd979bd MF |
6090 | /* We want to push off log flushes while truncates are |
6091 | * still running. */ | |
6092 | if (cancel) | |
6093 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
6094 | ||
35ddf78e | 6095 | queue_delayed_work(osb->ocfs2_wq, &osb->osb_truncate_log_wq, |
ccd979bd MF |
6096 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); |
6097 | } | |
6098 | } | |
6099 | ||
2070ad1a ER |
6100 | /* |
6101 | * Try to flush truncate logs if we can free enough clusters from it. | |
6102 | * As for return value, "< 0" means error, "0" no space and "1" means | |
6103 | * we have freed enough spaces and let the caller try to allocate again. | |
6104 | */ | |
6105 | int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb, | |
6106 | unsigned int needed) | |
6107 | { | |
6108 | tid_t target; | |
6109 | int ret = 0; | |
6110 | unsigned int truncated_clusters; | |
6111 | ||
6112 | inode_lock(osb->osb_tl_inode); | |
6113 | truncated_clusters = osb->truncated_clusters; | |
6114 | inode_unlock(osb->osb_tl_inode); | |
6115 | ||
6116 | /* | |
6117 | * Check whether we can succeed in allocating if we free | |
6118 | * the truncate log. | |
6119 | */ | |
6120 | if (truncated_clusters < needed) | |
6121 | goto out; | |
6122 | ||
6123 | ret = ocfs2_flush_truncate_log(osb); | |
6124 | if (ret) { | |
6125 | mlog_errno(ret); | |
6126 | goto out; | |
6127 | } | |
6128 | ||
6129 | if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) { | |
6130 | jbd2_log_wait_commit(osb->journal->j_journal, target); | |
6131 | ret = 1; | |
6132 | } | |
6133 | out: | |
6134 | return ret; | |
6135 | } | |
6136 | ||
ccd979bd MF |
6137 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, |
6138 | int slot_num, | |
6139 | struct inode **tl_inode, | |
6140 | struct buffer_head **tl_bh) | |
6141 | { | |
6142 | int status; | |
6143 | struct inode *inode = NULL; | |
6144 | struct buffer_head *bh = NULL; | |
6145 | ||
6146 | inode = ocfs2_get_system_file_inode(osb, | |
6147 | TRUNCATE_LOG_SYSTEM_INODE, | |
6148 | slot_num); | |
6149 | if (!inode) { | |
6150 | status = -EINVAL; | |
6151 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
6152 | goto bail; | |
6153 | } | |
6154 | ||
b657c95c | 6155 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
6156 | if (status < 0) { |
6157 | iput(inode); | |
6158 | mlog_errno(status); | |
6159 | goto bail; | |
6160 | } | |
6161 | ||
6162 | *tl_inode = inode; | |
6163 | *tl_bh = bh; | |
6164 | bail: | |
ccd979bd MF |
6165 | return status; |
6166 | } | |
6167 | ||
6168 | /* called during the 1st stage of node recovery. we stamp a clean | |
6169 | * truncate log and pass back a copy for processing later. if the | |
6170 | * truncate log does not require processing, a *tl_copy is set to | |
6171 | * NULL. */ | |
6172 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
6173 | int slot_num, | |
6174 | struct ocfs2_dinode **tl_copy) | |
6175 | { | |
6176 | int status; | |
6177 | struct inode *tl_inode = NULL; | |
6178 | struct buffer_head *tl_bh = NULL; | |
6179 | struct ocfs2_dinode *di; | |
6180 | struct ocfs2_truncate_log *tl; | |
6181 | ||
6182 | *tl_copy = NULL; | |
6183 | ||
a09d09b8 | 6184 | trace_ocfs2_begin_truncate_log_recovery(slot_num); |
ccd979bd MF |
6185 | |
6186 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
6187 | if (status < 0) { | |
6188 | mlog_errno(status); | |
6189 | goto bail; | |
6190 | } | |
6191 | ||
6192 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 6193 | |
10995aa2 JB |
6194 | /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's |
6195 | * validated by the underlying call to ocfs2_read_inode_block(), | |
6196 | * so any corruption is a code bug */ | |
6197 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
6198 | ||
6199 | tl = &di->id2.i_dealloc; | |
ccd979bd | 6200 | if (le16_to_cpu(tl->tl_used)) { |
a09d09b8 | 6201 | trace_ocfs2_truncate_log_recovery_num(le16_to_cpu(tl->tl_used)); |
ccd979bd MF |
6202 | |
6203 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
6204 | if (!(*tl_copy)) { | |
6205 | status = -ENOMEM; | |
6206 | mlog_errno(status); | |
6207 | goto bail; | |
6208 | } | |
6209 | ||
6210 | /* Assuming the write-out below goes well, this copy | |
6211 | * will be passed back to recovery for processing. */ | |
6212 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
6213 | ||
6214 | /* All we need to do to clear the truncate log is set | |
6215 | * tl_used. */ | |
6216 | tl->tl_used = 0; | |
6217 | ||
13723d00 | 6218 | ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check); |
8cb471e8 | 6219 | status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode)); |
ccd979bd MF |
6220 | if (status < 0) { |
6221 | mlog_errno(status); | |
6222 | goto bail; | |
6223 | } | |
6224 | } | |
6225 | ||
6226 | bail: | |
72865d92 | 6227 | iput(tl_inode); |
a81cb88b | 6228 | brelse(tl_bh); |
ccd979bd | 6229 | |
46359295 | 6230 | if (status < 0) { |
ccd979bd MF |
6231 | kfree(*tl_copy); |
6232 | *tl_copy = NULL; | |
c1e8d35e | 6233 | mlog_errno(status); |
ccd979bd MF |
6234 | } |
6235 | ||
ccd979bd MF |
6236 | return status; |
6237 | } | |
6238 | ||
6239 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
6240 | struct ocfs2_dinode *tl_copy) | |
6241 | { | |
6242 | int status = 0; | |
6243 | int i; | |
6244 | unsigned int clusters, num_recs, start_cluster; | |
6245 | u64 start_blk; | |
1fabe148 | 6246 | handle_t *handle; |
ccd979bd MF |
6247 | struct inode *tl_inode = osb->osb_tl_inode; |
6248 | struct ocfs2_truncate_log *tl; | |
6249 | ||
ccd979bd MF |
6250 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { |
6251 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
6252 | return -EINVAL; | |
6253 | } | |
6254 | ||
6255 | tl = &tl_copy->id2.i_dealloc; | |
6256 | num_recs = le16_to_cpu(tl->tl_used); | |
a09d09b8 TM |
6257 | trace_ocfs2_complete_truncate_log_recovery( |
6258 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno), | |
6259 | num_recs); | |
ccd979bd | 6260 | |
5955102c | 6261 | inode_lock(tl_inode); |
ccd979bd MF |
6262 | for(i = 0; i < num_recs; i++) { |
6263 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6264 | status = __ocfs2_flush_truncate_log(osb); | |
6265 | if (status < 0) { | |
6266 | mlog_errno(status); | |
6267 | goto bail_up; | |
6268 | } | |
6269 | } | |
6270 | ||
65eff9cc | 6271 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
6272 | if (IS_ERR(handle)) { |
6273 | status = PTR_ERR(handle); | |
6274 | mlog_errno(status); | |
6275 | goto bail_up; | |
6276 | } | |
6277 | ||
6278 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
6279 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
6280 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
6281 | ||
6282 | status = ocfs2_truncate_log_append(osb, handle, | |
6283 | start_blk, clusters); | |
02dc1af4 | 6284 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
6285 | if (status < 0) { |
6286 | mlog_errno(status); | |
6287 | goto bail_up; | |
6288 | } | |
6289 | } | |
6290 | ||
6291 | bail_up: | |
5955102c | 6292 | inode_unlock(tl_inode); |
ccd979bd | 6293 | |
ccd979bd MF |
6294 | return status; |
6295 | } | |
6296 | ||
6297 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
6298 | { | |
6299 | int status; | |
6300 | struct inode *tl_inode = osb->osb_tl_inode; | |
6301 | ||
a9e9acae X |
6302 | atomic_set(&osb->osb_tl_disable, 1); |
6303 | ||
ccd979bd MF |
6304 | if (tl_inode) { |
6305 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
35ddf78e | 6306 | flush_workqueue(osb->ocfs2_wq); |
ccd979bd MF |
6307 | |
6308 | status = ocfs2_flush_truncate_log(osb); | |
6309 | if (status < 0) | |
6310 | mlog_errno(status); | |
6311 | ||
6312 | brelse(osb->osb_tl_bh); | |
6313 | iput(osb->osb_tl_inode); | |
6314 | } | |
ccd979bd MF |
6315 | } |
6316 | ||
6317 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
6318 | { | |
6319 | int status; | |
6320 | struct inode *tl_inode = NULL; | |
6321 | struct buffer_head *tl_bh = NULL; | |
6322 | ||
ccd979bd MF |
6323 | status = ocfs2_get_truncate_log_info(osb, |
6324 | osb->slot_num, | |
6325 | &tl_inode, | |
6326 | &tl_bh); | |
6327 | if (status < 0) | |
6328 | mlog_errno(status); | |
6329 | ||
6330 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
6331 | * osb->osb_tl_inode so we don't set any of the osb variables | |
6332 | * until we're sure all is well. */ | |
c4028958 DH |
6333 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
6334 | ocfs2_truncate_log_worker); | |
a9e9acae | 6335 | atomic_set(&osb->osb_tl_disable, 0); |
ccd979bd MF |
6336 | osb->osb_tl_bh = tl_bh; |
6337 | osb->osb_tl_inode = tl_inode; | |
6338 | ||
ccd979bd MF |
6339 | return status; |
6340 | } | |
6341 | ||
2b604351 MF |
6342 | /* |
6343 | * Delayed de-allocation of suballocator blocks. | |
6344 | * | |
6345 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
6346 | * | |
6347 | * The locking for this can get extremely complicated, especially when | |
6348 | * the suballocator inodes to delete from aren't known until deep | |
6349 | * within an unrelated codepath. | |
6350 | * | |
6351 | * ocfs2_extent_block structures are a good example of this - an inode | |
6352 | * btree could have been grown by any number of nodes each allocating | |
6353 | * out of their own suballoc inode. | |
6354 | * | |
6355 | * These structures allow the delay of block de-allocation until a | |
6356 | * later time, when locking of multiple cluster inodes won't cause | |
6357 | * deadlock. | |
6358 | */ | |
6359 | ||
6360 | /* | |
2891d290 TM |
6361 | * Describe a single bit freed from a suballocator. For the block |
6362 | * suballocators, it represents one block. For the global cluster | |
6363 | * allocator, it represents some clusters and free_bit indicates | |
6364 | * clusters number. | |
2b604351 MF |
6365 | */ |
6366 | struct ocfs2_cached_block_free { | |
6367 | struct ocfs2_cached_block_free *free_next; | |
74380c47 | 6368 | u64 free_bg; |
2b604351 MF |
6369 | u64 free_blk; |
6370 | unsigned int free_bit; | |
6371 | }; | |
6372 | ||
6373 | struct ocfs2_per_slot_free_list { | |
6374 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
6375 | int f_inode_type; | |
6376 | int f_slot; | |
6377 | struct ocfs2_cached_block_free *f_first; | |
6378 | }; | |
6379 | ||
2891d290 TM |
6380 | static int ocfs2_free_cached_blocks(struct ocfs2_super *osb, |
6381 | int sysfile_type, | |
6382 | int slot, | |
6383 | struct ocfs2_cached_block_free *head) | |
2b604351 MF |
6384 | { |
6385 | int ret; | |
6386 | u64 bg_blkno; | |
6387 | handle_t *handle; | |
6388 | struct inode *inode; | |
6389 | struct buffer_head *di_bh = NULL; | |
6390 | struct ocfs2_cached_block_free *tmp; | |
6391 | ||
6392 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
6393 | if (!inode) { | |
6394 | ret = -EINVAL; | |
6395 | mlog_errno(ret); | |
6396 | goto out; | |
6397 | } | |
6398 | ||
5955102c | 6399 | inode_lock(inode); |
2b604351 | 6400 | |
e63aecb6 | 6401 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
6402 | if (ret) { |
6403 | mlog_errno(ret); | |
6404 | goto out_mutex; | |
6405 | } | |
6406 | ||
2b604351 | 6407 | while (head) { |
74380c47 TM |
6408 | if (head->free_bg) |
6409 | bg_blkno = head->free_bg; | |
6410 | else | |
6411 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
6412 | head->free_bit); | |
d5bf1418 JB |
6413 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); |
6414 | if (IS_ERR(handle)) { | |
6415 | ret = PTR_ERR(handle); | |
6416 | mlog_errno(ret); | |
6417 | goto out_unlock; | |
6418 | } | |
6419 | ||
a09d09b8 TM |
6420 | trace_ocfs2_free_cached_blocks( |
6421 | (unsigned long long)head->free_blk, head->free_bit); | |
2b604351 MF |
6422 | |
6423 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
6424 | head->free_bit, bg_blkno, 1); | |
d5bf1418 | 6425 | if (ret) |
2b604351 | 6426 | mlog_errno(ret); |
2b604351 | 6427 | |
d5bf1418 | 6428 | ocfs2_commit_trans(osb, handle); |
2b604351 MF |
6429 | |
6430 | tmp = head; | |
6431 | head = head->free_next; | |
6432 | kfree(tmp); | |
6433 | } | |
6434 | ||
2b604351 | 6435 | out_unlock: |
e63aecb6 | 6436 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
6437 | brelse(di_bh); |
6438 | out_mutex: | |
5955102c | 6439 | inode_unlock(inode); |
2b604351 MF |
6440 | iput(inode); |
6441 | out: | |
6442 | while(head) { | |
6443 | /* Premature exit may have left some dangling items. */ | |
6444 | tmp = head; | |
6445 | head = head->free_next; | |
6446 | kfree(tmp); | |
6447 | } | |
6448 | ||
6449 | return ret; | |
6450 | } | |
6451 | ||
2891d290 TM |
6452 | int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6453 | u64 blkno, unsigned int bit) | |
6454 | { | |
6455 | int ret = 0; | |
6456 | struct ocfs2_cached_block_free *item; | |
6457 | ||
74380c47 | 6458 | item = kzalloc(sizeof(*item), GFP_NOFS); |
2891d290 TM |
6459 | if (item == NULL) { |
6460 | ret = -ENOMEM; | |
6461 | mlog_errno(ret); | |
6462 | return ret; | |
6463 | } | |
6464 | ||
a09d09b8 | 6465 | trace_ocfs2_cache_cluster_dealloc((unsigned long long)blkno, bit); |
2891d290 TM |
6466 | |
6467 | item->free_blk = blkno; | |
6468 | item->free_bit = bit; | |
6469 | item->free_next = ctxt->c_global_allocator; | |
6470 | ||
6471 | ctxt->c_global_allocator = item; | |
6472 | return ret; | |
6473 | } | |
6474 | ||
6475 | static int ocfs2_free_cached_clusters(struct ocfs2_super *osb, | |
6476 | struct ocfs2_cached_block_free *head) | |
6477 | { | |
6478 | struct ocfs2_cached_block_free *tmp; | |
6479 | struct inode *tl_inode = osb->osb_tl_inode; | |
6480 | handle_t *handle; | |
6481 | int ret = 0; | |
6482 | ||
5955102c | 6483 | inode_lock(tl_inode); |
2891d290 TM |
6484 | |
6485 | while (head) { | |
6486 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6487 | ret = __ocfs2_flush_truncate_log(osb); | |
6488 | if (ret < 0) { | |
6489 | mlog_errno(ret); | |
6490 | break; | |
6491 | } | |
6492 | } | |
6493 | ||
6494 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); | |
6495 | if (IS_ERR(handle)) { | |
6496 | ret = PTR_ERR(handle); | |
6497 | mlog_errno(ret); | |
6498 | break; | |
6499 | } | |
6500 | ||
6501 | ret = ocfs2_truncate_log_append(osb, handle, head->free_blk, | |
6502 | head->free_bit); | |
6503 | ||
6504 | ocfs2_commit_trans(osb, handle); | |
6505 | tmp = head; | |
6506 | head = head->free_next; | |
6507 | kfree(tmp); | |
6508 | ||
6509 | if (ret < 0) { | |
6510 | mlog_errno(ret); | |
6511 | break; | |
6512 | } | |
6513 | } | |
6514 | ||
5955102c | 6515 | inode_unlock(tl_inode); |
2891d290 TM |
6516 | |
6517 | while (head) { | |
6518 | /* Premature exit may have left some dangling items. */ | |
6519 | tmp = head; | |
6520 | head = head->free_next; | |
6521 | kfree(tmp); | |
6522 | } | |
6523 | ||
6524 | return ret; | |
6525 | } | |
6526 | ||
2b604351 MF |
6527 | int ocfs2_run_deallocs(struct ocfs2_super *osb, |
6528 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6529 | { | |
6530 | int ret = 0, ret2; | |
6531 | struct ocfs2_per_slot_free_list *fl; | |
6532 | ||
6533 | if (!ctxt) | |
6534 | return 0; | |
6535 | ||
6536 | while (ctxt->c_first_suballocator) { | |
6537 | fl = ctxt->c_first_suballocator; | |
6538 | ||
6539 | if (fl->f_first) { | |
a09d09b8 TM |
6540 | trace_ocfs2_run_deallocs(fl->f_inode_type, |
6541 | fl->f_slot); | |
2891d290 TM |
6542 | ret2 = ocfs2_free_cached_blocks(osb, |
6543 | fl->f_inode_type, | |
6544 | fl->f_slot, | |
6545 | fl->f_first); | |
2b604351 MF |
6546 | if (ret2) |
6547 | mlog_errno(ret2); | |
6548 | if (!ret) | |
6549 | ret = ret2; | |
6550 | } | |
6551 | ||
6552 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
6553 | kfree(fl); | |
6554 | } | |
6555 | ||
2891d290 TM |
6556 | if (ctxt->c_global_allocator) { |
6557 | ret2 = ocfs2_free_cached_clusters(osb, | |
6558 | ctxt->c_global_allocator); | |
6559 | if (ret2) | |
6560 | mlog_errno(ret2); | |
6561 | if (!ret) | |
6562 | ret = ret2; | |
6563 | ||
6564 | ctxt->c_global_allocator = NULL; | |
6565 | } | |
6566 | ||
2b604351 MF |
6567 | return ret; |
6568 | } | |
6569 | ||
6570 | static struct ocfs2_per_slot_free_list * | |
6571 | ocfs2_find_per_slot_free_list(int type, | |
6572 | int slot, | |
6573 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6574 | { | |
6575 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
6576 | ||
6577 | while (fl) { | |
6578 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
6579 | return fl; | |
6580 | ||
6581 | fl = fl->f_next_suballocator; | |
6582 | } | |
6583 | ||
6584 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
6585 | if (fl) { | |
6586 | fl->f_inode_type = type; | |
6587 | fl->f_slot = slot; | |
6588 | fl->f_first = NULL; | |
6589 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
6590 | ||
6591 | ctxt->c_first_suballocator = fl; | |
6592 | } | |
6593 | return fl; | |
6594 | } | |
6595 | ||
1823cb0b | 6596 | int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
74380c47 TM |
6597 | int type, int slot, u64 suballoc, |
6598 | u64 blkno, unsigned int bit) | |
2b604351 MF |
6599 | { |
6600 | int ret; | |
6601 | struct ocfs2_per_slot_free_list *fl; | |
6602 | struct ocfs2_cached_block_free *item; | |
6603 | ||
6604 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6605 | if (fl == NULL) { | |
6606 | ret = -ENOMEM; | |
6607 | mlog_errno(ret); | |
6608 | goto out; | |
6609 | } | |
6610 | ||
74380c47 | 6611 | item = kzalloc(sizeof(*item), GFP_NOFS); |
2b604351 MF |
6612 | if (item == NULL) { |
6613 | ret = -ENOMEM; | |
6614 | mlog_errno(ret); | |
6615 | goto out; | |
6616 | } | |
6617 | ||
a09d09b8 TM |
6618 | trace_ocfs2_cache_block_dealloc(type, slot, |
6619 | (unsigned long long)suballoc, | |
6620 | (unsigned long long)blkno, bit); | |
2b604351 | 6621 | |
74380c47 | 6622 | item->free_bg = suballoc; |
2b604351 MF |
6623 | item->free_blk = blkno; |
6624 | item->free_bit = bit; | |
6625 | item->free_next = fl->f_first; | |
6626 | ||
6627 | fl->f_first = item; | |
6628 | ||
6629 | ret = 0; | |
6630 | out: | |
6631 | return ret; | |
6632 | } | |
6633 | ||
59a5e416 MF |
6634 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6635 | struct ocfs2_extent_block *eb) | |
6636 | { | |
6637 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6638 | le16_to_cpu(eb->h_suballoc_slot), | |
74380c47 | 6639 | le64_to_cpu(eb->h_suballoc_loc), |
59a5e416 MF |
6640 | le64_to_cpu(eb->h_blkno), |
6641 | le16_to_cpu(eb->h_suballoc_bit)); | |
6642 | } | |
6643 | ||
2b4e30fb | 6644 | static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh) |
60b11392 MF |
6645 | { |
6646 | set_buffer_uptodate(bh); | |
6647 | mark_buffer_dirty(bh); | |
6648 | return 0; | |
6649 | } | |
6650 | ||
6f70fa51 TM |
6651 | void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
6652 | unsigned int from, unsigned int to, | |
6653 | struct page *page, int zero, u64 *phys) | |
1d410a6e MF |
6654 | { |
6655 | int ret, partial = 0; | |
6656 | ||
6657 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
6658 | if (ret) | |
6659 | mlog_errno(ret); | |
6660 | ||
6661 | if (zero) | |
eebd2aa3 | 6662 | zero_user_segment(page, from, to); |
1d410a6e MF |
6663 | |
6664 | /* | |
6665 | * Need to set the buffers we zero'd into uptodate | |
6666 | * here if they aren't - ocfs2_map_page_blocks() | |
6667 | * might've skipped some | |
6668 | */ | |
2b4e30fb JB |
6669 | ret = walk_page_buffers(handle, page_buffers(page), |
6670 | from, to, &partial, | |
6671 | ocfs2_zero_func); | |
6672 | if (ret < 0) | |
6673 | mlog_errno(ret); | |
6674 | else if (ocfs2_should_order_data(inode)) { | |
6675 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
1d410a6e MF |
6676 | if (ret < 0) |
6677 | mlog_errno(ret); | |
6678 | } | |
6679 | ||
6680 | if (!partial) | |
6681 | SetPageUptodate(page); | |
6682 | ||
6683 | flush_dcache_page(page); | |
6684 | } | |
6685 | ||
35edec1d MF |
6686 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
6687 | loff_t end, struct page **pages, | |
6688 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 6689 | { |
1d410a6e | 6690 | int i; |
60b11392 | 6691 | struct page *page; |
09cbfeaf | 6692 | unsigned int from, to = PAGE_SIZE; |
60b11392 MF |
6693 | struct super_block *sb = inode->i_sb; |
6694 | ||
6695 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
6696 | ||
6697 | if (numpages == 0) | |
6698 | goto out; | |
6699 | ||
09cbfeaf | 6700 | to = PAGE_SIZE; |
60b11392 MF |
6701 | for(i = 0; i < numpages; i++) { |
6702 | page = pages[i]; | |
6703 | ||
09cbfeaf KS |
6704 | from = start & (PAGE_SIZE - 1); |
6705 | if ((end >> PAGE_SHIFT) == page->index) | |
6706 | to = end & (PAGE_SIZE - 1); | |
35edec1d | 6707 | |
09cbfeaf KS |
6708 | BUG_ON(from > PAGE_SIZE); |
6709 | BUG_ON(to > PAGE_SIZE); | |
60b11392 | 6710 | |
1d410a6e MF |
6711 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
6712 | &phys); | |
60b11392 | 6713 | |
09cbfeaf | 6714 | start = (page->index + 1) << PAGE_SHIFT; |
60b11392 MF |
6715 | } |
6716 | out: | |
1d410a6e MF |
6717 | if (pages) |
6718 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6719 | } |
6720 | ||
6f70fa51 TM |
6721 | int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end, |
6722 | struct page **pages, int *num) | |
60b11392 | 6723 | { |
1d410a6e | 6724 | int numpages, ret = 0; |
60b11392 MF |
6725 | struct address_space *mapping = inode->i_mapping; |
6726 | unsigned long index; | |
35edec1d | 6727 | loff_t last_page_bytes; |
60b11392 | 6728 | |
35edec1d | 6729 | BUG_ON(start > end); |
60b11392 | 6730 | |
1d410a6e | 6731 | numpages = 0; |
35edec1d | 6732 | last_page_bytes = PAGE_ALIGN(end); |
09cbfeaf | 6733 | index = start >> PAGE_SHIFT; |
60b11392 | 6734 | do { |
9b4c0ff3 | 6735 | pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS); |
60b11392 MF |
6736 | if (!pages[numpages]) { |
6737 | ret = -ENOMEM; | |
6738 | mlog_errno(ret); | |
6739 | goto out; | |
6740 | } | |
6741 | ||
6742 | numpages++; | |
6743 | index++; | |
09cbfeaf | 6744 | } while (index < (last_page_bytes >> PAGE_SHIFT)); |
60b11392 MF |
6745 | |
6746 | out: | |
6747 | if (ret != 0) { | |
1d410a6e MF |
6748 | if (pages) |
6749 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6750 | numpages = 0; |
6751 | } | |
6752 | ||
6753 | *num = numpages; | |
6754 | ||
6755 | return ret; | |
6756 | } | |
6757 | ||
6f70fa51 TM |
6758 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
6759 | struct page **pages, int *num) | |
6760 | { | |
6761 | struct super_block *sb = inode->i_sb; | |
6762 | ||
6763 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != | |
6764 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
6765 | ||
6766 | return ocfs2_grab_pages(inode, start, end, pages, num); | |
6767 | } | |
6768 | ||
60b11392 MF |
6769 | /* |
6770 | * Zero the area past i_size but still within an allocated | |
6771 | * cluster. This avoids exposing nonzero data on subsequent file | |
6772 | * extends. | |
6773 | * | |
6774 | * We need to call this before i_size is updated on the inode because | |
6775 | * otherwise block_write_full_page() will skip writeout of pages past | |
6776 | * i_size. The new_i_size parameter is passed for this reason. | |
6777 | */ | |
35edec1d MF |
6778 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
6779 | u64 range_start, u64 range_end) | |
60b11392 | 6780 | { |
1d410a6e | 6781 | int ret = 0, numpages; |
60b11392 MF |
6782 | struct page **pages = NULL; |
6783 | u64 phys; | |
1d410a6e MF |
6784 | unsigned int ext_flags; |
6785 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
6786 | |
6787 | /* | |
6788 | * File systems which don't support sparse files zero on every | |
6789 | * extend. | |
6790 | */ | |
1d410a6e | 6791 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
6792 | return 0; |
6793 | ||
1d410a6e | 6794 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
6795 | sizeof(struct page *), GFP_NOFS); |
6796 | if (pages == NULL) { | |
6797 | ret = -ENOMEM; | |
6798 | mlog_errno(ret); | |
6799 | goto out; | |
6800 | } | |
6801 | ||
1d410a6e MF |
6802 | if (range_start == range_end) |
6803 | goto out; | |
6804 | ||
6805 | ret = ocfs2_extent_map_get_blocks(inode, | |
6806 | range_start >> sb->s_blocksize_bits, | |
6807 | &phys, NULL, &ext_flags); | |
60b11392 MF |
6808 | if (ret) { |
6809 | mlog_errno(ret); | |
6810 | goto out; | |
6811 | } | |
6812 | ||
1d410a6e MF |
6813 | /* |
6814 | * Tail is a hole, or is marked unwritten. In either case, we | |
6815 | * can count on read and write to return/push zero's. | |
6816 | */ | |
6817 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
6818 | goto out; |
6819 | ||
1d410a6e MF |
6820 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
6821 | &numpages); | |
6822 | if (ret) { | |
6823 | mlog_errno(ret); | |
6824 | goto out; | |
6825 | } | |
6826 | ||
35edec1d MF |
6827 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
6828 | numpages, phys, handle); | |
60b11392 MF |
6829 | |
6830 | /* | |
6831 | * Initiate writeout of the pages we zero'd here. We don't | |
6832 | * wait on them - the truncate_inode_pages() call later will | |
6833 | * do that for us. | |
6834 | */ | |
2cfd30ad CH |
6835 | ret = filemap_fdatawrite_range(inode->i_mapping, range_start, |
6836 | range_end - 1); | |
60b11392 MF |
6837 | if (ret) |
6838 | mlog_errno(ret); | |
6839 | ||
6840 | out: | |
d787ab09 | 6841 | kfree(pages); |
60b11392 MF |
6842 | |
6843 | return ret; | |
6844 | } | |
6845 | ||
fdd77704 TY |
6846 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
6847 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
6848 | { |
6849 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 6850 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 6851 | |
fdd77704 TY |
6852 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
6853 | memset(&di->id2, 0, blocksize - | |
6854 | offsetof(struct ocfs2_dinode, id2) - | |
6855 | xattrsize); | |
6856 | else | |
6857 | memset(&di->id2, 0, blocksize - | |
6858 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
6859 | } |
6860 | ||
5b6a3a2b MF |
6861 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
6862 | struct ocfs2_dinode *di) | |
6863 | { | |
fdd77704 | 6864 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
6865 | di->id2.i_list.l_tree_depth = 0; |
6866 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
6867 | di->id2.i_list.l_count = cpu_to_le16( |
6868 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
6869 | } |
6870 | ||
1afc32b9 MF |
6871 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
6872 | { | |
6873 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6874 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
6875 | ||
6876 | spin_lock(&oi->ip_lock); | |
6877 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
6878 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6879 | spin_unlock(&oi->ip_lock); | |
6880 | ||
6881 | /* | |
6882 | * We clear the entire i_data structure here so that all | |
6883 | * fields can be properly initialized. | |
6884 | */ | |
fdd77704 | 6885 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 6886 | |
fdd77704 TY |
6887 | idata->id_count = cpu_to_le16( |
6888 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
6889 | } |
6890 | ||
6891 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
6892 | struct buffer_head *di_bh) | |
6893 | { | |
6894 | int ret, i, has_data, num_pages = 0; | |
fb951eb5 ZW |
6895 | int need_free = 0; |
6896 | u32 bit_off, num; | |
1afc32b9 MF |
6897 | handle_t *handle; |
6898 | u64 uninitialized_var(block); | |
6899 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6900 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
6901 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
6902 | struct ocfs2_alloc_context *data_ac = NULL; |
6903 | struct page **pages = NULL; | |
6904 | loff_t end = osb->s_clustersize; | |
f99b9b7c | 6905 | struct ocfs2_extent_tree et; |
a90714c1 | 6906 | int did_quota = 0; |
1afc32b9 MF |
6907 | |
6908 | has_data = i_size_read(inode) ? 1 : 0; | |
6909 | ||
6910 | if (has_data) { | |
6911 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
6912 | sizeof(struct page *), GFP_NOFS); | |
6913 | if (pages == NULL) { | |
6914 | ret = -ENOMEM; | |
6915 | mlog_errno(ret); | |
3cc79b79 | 6916 | return ret; |
1afc32b9 MF |
6917 | } |
6918 | ||
6919 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
6920 | if (ret) { | |
6921 | mlog_errno(ret); | |
3cc79b79 | 6922 | goto free_pages; |
1afc32b9 MF |
6923 | } |
6924 | } | |
6925 | ||
a90714c1 JK |
6926 | handle = ocfs2_start_trans(osb, |
6927 | ocfs2_inline_to_extents_credits(osb->sb)); | |
1afc32b9 MF |
6928 | if (IS_ERR(handle)) { |
6929 | ret = PTR_ERR(handle); | |
6930 | mlog_errno(ret); | |
15eba0fe | 6931 | goto out; |
1afc32b9 MF |
6932 | } |
6933 | ||
0cf2f763 | 6934 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 6935 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
6936 | if (ret) { |
6937 | mlog_errno(ret); | |
6938 | goto out_commit; | |
6939 | } | |
6940 | ||
6941 | if (has_data) { | |
1afc32b9 MF |
6942 | unsigned int page_end; |
6943 | u64 phys; | |
6944 | ||
5dd4056d CH |
6945 | ret = dquot_alloc_space_nodirty(inode, |
6946 | ocfs2_clusters_to_bytes(osb->sb, 1)); | |
6947 | if (ret) | |
a90714c1 | 6948 | goto out_commit; |
a90714c1 JK |
6949 | did_quota = 1; |
6950 | ||
4fe370af MF |
6951 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; |
6952 | ||
1ed9b777 | 6953 | ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off, |
1afc32b9 MF |
6954 | &num); |
6955 | if (ret) { | |
6956 | mlog_errno(ret); | |
6957 | goto out_commit; | |
6958 | } | |
6959 | ||
6960 | /* | |
6961 | * Save two copies, one for insert, and one that can | |
6962 | * be changed by ocfs2_map_and_dirty_page() below. | |
6963 | */ | |
6964 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
6965 | ||
6966 | /* | |
6967 | * Non sparse file systems zero on extend, so no need | |
6968 | * to do that now. | |
6969 | */ | |
6970 | if (!ocfs2_sparse_alloc(osb) && | |
09cbfeaf KS |
6971 | PAGE_SIZE < osb->s_clustersize) |
6972 | end = PAGE_SIZE; | |
1afc32b9 MF |
6973 | |
6974 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
6975 | if (ret) { | |
6976 | mlog_errno(ret); | |
fb951eb5 | 6977 | need_free = 1; |
1afc32b9 MF |
6978 | goto out_commit; |
6979 | } | |
6980 | ||
6981 | /* | |
6982 | * This should populate the 1st page for us and mark | |
6983 | * it up to date. | |
6984 | */ | |
6985 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
6986 | if (ret) { | |
6987 | mlog_errno(ret); | |
fb951eb5 | 6988 | need_free = 1; |
15eba0fe | 6989 | goto out_unlock; |
1afc32b9 MF |
6990 | } |
6991 | ||
09cbfeaf KS |
6992 | page_end = PAGE_SIZE; |
6993 | if (PAGE_SIZE > osb->s_clustersize) | |
1afc32b9 MF |
6994 | page_end = osb->s_clustersize; |
6995 | ||
6996 | for (i = 0; i < num_pages; i++) | |
6997 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
6998 | pages[i], i > 0, &phys); | |
6999 | } | |
7000 | ||
7001 | spin_lock(&oi->ip_lock); | |
7002 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
7003 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7004 | spin_unlock(&oi->ip_lock); | |
7005 | ||
2931cdcb | 7006 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
5b6a3a2b | 7007 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
7008 | |
7009 | ocfs2_journal_dirty(handle, di_bh); | |
7010 | ||
7011 | if (has_data) { | |
7012 | /* | |
7013 | * An error at this point should be extremely rare. If | |
7014 | * this proves to be false, we could always re-build | |
7015 | * the in-inode data from our pages. | |
7016 | */ | |
5e404e9e | 7017 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
cc79d8c1 | 7018 | ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL); |
1afc32b9 MF |
7019 | if (ret) { |
7020 | mlog_errno(ret); | |
fb951eb5 | 7021 | need_free = 1; |
15eba0fe | 7022 | goto out_unlock; |
1afc32b9 MF |
7023 | } |
7024 | ||
7025 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7026 | } | |
7027 | ||
15eba0fe | 7028 | out_unlock: |
7029 | if (pages) | |
7030 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
7031 | ||
1afc32b9 | 7032 | out_commit: |
a90714c1 | 7033 | if (ret < 0 && did_quota) |
5dd4056d | 7034 | dquot_free_space_nodirty(inode, |
a90714c1 JK |
7035 | ocfs2_clusters_to_bytes(osb->sb, 1)); |
7036 | ||
fb951eb5 ZW |
7037 | if (need_free) { |
7038 | if (data_ac->ac_which == OCFS2_AC_USE_LOCAL) | |
7039 | ocfs2_free_local_alloc_bits(osb, handle, data_ac, | |
7040 | bit_off, num); | |
7041 | else | |
7042 | ocfs2_free_clusters(handle, | |
7043 | data_ac->ac_inode, | |
7044 | data_ac->ac_bh, | |
7045 | ocfs2_clusters_to_blocks(osb->sb, bit_off), | |
7046 | num); | |
7047 | } | |
7048 | ||
1afc32b9 MF |
7049 | ocfs2_commit_trans(osb, handle); |
7050 | ||
15eba0fe | 7051 | out: |
1afc32b9 MF |
7052 | if (data_ac) |
7053 | ocfs2_free_alloc_context(data_ac); | |
3cc79b79 | 7054 | free_pages: |
fd90d4df | 7055 | kfree(pages); |
1afc32b9 MF |
7056 | return ret; |
7057 | } | |
7058 | ||
ccd979bd MF |
7059 | /* |
7060 | * It is expected, that by the time you call this function, | |
7061 | * inode->i_size and fe->i_size have been adjusted. | |
7062 | * | |
7063 | * WARNING: This will kfree the truncate context | |
7064 | */ | |
7065 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
7066 | struct inode *inode, | |
78f94673 | 7067 | struct buffer_head *di_bh) |
ccd979bd | 7068 | { |
78f94673 TY |
7069 | int status = 0, i, flags = 0; |
7070 | u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff; | |
bcbbb24a | 7071 | u64 blkno = 0; |
ccd979bd | 7072 | struct ocfs2_extent_list *el; |
78f94673 | 7073 | struct ocfs2_extent_rec *rec; |
dcd0538f | 7074 | struct ocfs2_path *path = NULL; |
78f94673 TY |
7075 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
7076 | struct ocfs2_extent_list *root_el = &(di->id2.i_list); | |
7077 | u64 refcount_loc = le64_to_cpu(di->i_refcount_loc); | |
7078 | struct ocfs2_extent_tree et; | |
7079 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
f62f12b3 | 7080 | struct ocfs2_refcount_tree *ref_tree = NULL; |
ccd979bd | 7081 | |
78f94673 TY |
7082 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
7083 | ocfs2_init_dealloc_ctxt(&dealloc); | |
7084 | ||
dcd0538f | 7085 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
7086 | i_size_read(inode)); |
7087 | ||
78f94673 | 7088 | path = ocfs2_new_path(di_bh, &di->id2.i_list, |
13723d00 | 7089 | ocfs2_journal_access_di); |
dcd0538f MF |
7090 | if (!path) { |
7091 | status = -ENOMEM; | |
7092 | mlog_errno(status); | |
7093 | goto bail; | |
7094 | } | |
83418978 MF |
7095 | |
7096 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
7097 | ||
ccd979bd | 7098 | start: |
3a0782d0 MF |
7099 | /* |
7100 | * Check that we still have allocation to delete. | |
7101 | */ | |
7102 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
7103 | status = 0; | |
7104 | goto bail; | |
7105 | } | |
7106 | ||
dcd0538f MF |
7107 | /* |
7108 | * Truncate always works against the rightmost tree branch. | |
7109 | */ | |
facdb77f | 7110 | status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX); |
dcd0538f MF |
7111 | if (status) { |
7112 | mlog_errno(status); | |
7113 | goto bail; | |
ccd979bd MF |
7114 | } |
7115 | ||
a09d09b8 TM |
7116 | trace_ocfs2_commit_truncate( |
7117 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7118 | new_highest_cpos, | |
7119 | OCFS2_I(inode)->ip_clusters, | |
7120 | path->p_tree_depth); | |
dcd0538f MF |
7121 | |
7122 | /* | |
7123 | * By now, el will point to the extent list on the bottom most | |
7124 | * portion of this tree. Only the tail record is considered in | |
7125 | * each pass. | |
7126 | * | |
7127 | * We handle the following cases, in order: | |
7128 | * - empty extent: delete the remaining branch | |
7129 | * - remove the entire record | |
7130 | * - remove a partial record | |
7131 | * - no record needs to be removed (truncate has completed) | |
7132 | */ | |
7133 | el = path_leaf_el(path); | |
3a0782d0 MF |
7134 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
7135 | ocfs2_error(inode->i_sb, | |
7136 | "Inode %llu has empty extent block at %llu\n", | |
7137 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7138 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7139 | status = -EROFS; | |
7140 | goto bail; | |
7141 | } | |
7142 | ||
ccd979bd | 7143 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
78f94673 TY |
7144 | rec = &el->l_recs[i]; |
7145 | flags = rec->e_flags; | |
7146 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
7147 | ||
7148 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
7149 | /* | |
7150 | * Lower levels depend on this never happening, but it's best | |
7151 | * to check it up here before changing the tree. | |
7152 | */ | |
7153 | if (root_el->l_tree_depth && rec->e_int_clusters == 0) { | |
d0c97d52 | 7154 | mlog(ML_ERROR, "Inode %lu has an empty " |
78f94673 TY |
7155 | "extent record, depth %u\n", inode->i_ino, |
7156 | le16_to_cpu(root_el->l_tree_depth)); | |
d0c97d52 X |
7157 | status = ocfs2_remove_rightmost_empty_extent(osb, |
7158 | &et, path, &dealloc); | |
7159 | if (status) { | |
7160 | mlog_errno(status); | |
7161 | goto bail; | |
7162 | } | |
7163 | ||
7164 | ocfs2_reinit_path(path, 1); | |
7165 | goto start; | |
7166 | } else { | |
7167 | trunc_cpos = le32_to_cpu(rec->e_cpos); | |
7168 | trunc_len = 0; | |
7169 | blkno = 0; | |
78f94673 | 7170 | } |
78f94673 TY |
7171 | } else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) { |
7172 | /* | |
7173 | * Truncate entire record. | |
7174 | */ | |
7175 | trunc_cpos = le32_to_cpu(rec->e_cpos); | |
7176 | trunc_len = ocfs2_rec_clusters(el, rec); | |
7177 | blkno = le64_to_cpu(rec->e_blkno); | |
dcd0538f | 7178 | } else if (range > new_highest_cpos) { |
78f94673 TY |
7179 | /* |
7180 | * Partial truncate. it also should be | |
7181 | * the last truncate we're doing. | |
7182 | */ | |
7183 | trunc_cpos = new_highest_cpos; | |
7184 | trunc_len = range - new_highest_cpos; | |
7185 | coff = new_highest_cpos - le32_to_cpu(rec->e_cpos); | |
7186 | blkno = le64_to_cpu(rec->e_blkno) + | |
7187 | ocfs2_clusters_to_blocks(inode->i_sb, coff); | |
dcd0538f | 7188 | } else { |
78f94673 TY |
7189 | /* |
7190 | * Truncate completed, leave happily. | |
7191 | */ | |
dcd0538f MF |
7192 | status = 0; |
7193 | goto bail; | |
7194 | } | |
ccd979bd | 7195 | |
78f94673 | 7196 | phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno); |
bcbbb24a | 7197 | |
f62f12b3 JB |
7198 | if ((flags & OCFS2_EXT_REFCOUNTED) && trunc_len && !ref_tree) { |
7199 | status = ocfs2_lock_refcount_tree(osb, refcount_loc, 1, | |
7200 | &ref_tree, NULL); | |
7201 | if (status) { | |
7202 | mlog_errno(status); | |
7203 | goto bail; | |
7204 | } | |
7205 | } | |
7206 | ||
78f94673 TY |
7207 | status = ocfs2_remove_btree_range(inode, &et, trunc_cpos, |
7208 | phys_cpos, trunc_len, flags, &dealloc, | |
f62f12b3 | 7209 | refcount_loc, true); |
ccd979bd MF |
7210 | if (status < 0) { |
7211 | mlog_errno(status); | |
7212 | goto bail; | |
7213 | } | |
7214 | ||
dcd0538f MF |
7215 | ocfs2_reinit_path(path, 1); |
7216 | ||
7217 | /* | |
3a0782d0 MF |
7218 | * The check above will catch the case where we've truncated |
7219 | * away all allocation. | |
dcd0538f | 7220 | */ |
3a0782d0 MF |
7221 | goto start; |
7222 | ||
ccd979bd | 7223 | bail: |
f62f12b3 JB |
7224 | if (ref_tree) |
7225 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
ccd979bd MF |
7226 | |
7227 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
7228 | ||
78f94673 | 7229 | ocfs2_run_deallocs(osb, &dealloc); |
59a5e416 | 7230 | |
dcd0538f | 7231 | ocfs2_free_path(path); |
ccd979bd | 7232 | |
ccd979bd MF |
7233 | return status; |
7234 | } | |
7235 | ||
1afc32b9 MF |
7236 | /* |
7237 | * 'start' is inclusive, 'end' is not. | |
7238 | */ | |
7239 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7240 | unsigned int start, unsigned int end, int trunc) | |
7241 | { | |
7242 | int ret; | |
7243 | unsigned int numbytes; | |
7244 | handle_t *handle; | |
7245 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7246 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7247 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7248 | ||
7249 | if (end > i_size_read(inode)) | |
7250 | end = i_size_read(inode); | |
7251 | ||
d62e74be | 7252 | BUG_ON(start > end); |
1afc32b9 MF |
7253 | |
7254 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7255 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7256 | !ocfs2_supports_inline_data(osb)) { | |
7257 | ocfs2_error(inode->i_sb, | |
7ecef14a | 7258 | "Inline data flags for inode %llu don't agree! Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", |
1afc32b9 MF |
7259 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
7260 | le16_to_cpu(di->i_dyn_features), | |
7261 | OCFS2_I(inode)->ip_dyn_features, | |
7262 | osb->s_feature_incompat); | |
7263 | ret = -EROFS; | |
7264 | goto out; | |
7265 | } | |
7266 | ||
7267 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7268 | if (IS_ERR(handle)) { | |
7269 | ret = PTR_ERR(handle); | |
7270 | mlog_errno(ret); | |
7271 | goto out; | |
7272 | } | |
7273 | ||
0cf2f763 | 7274 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7275 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7276 | if (ret) { |
7277 | mlog_errno(ret); | |
7278 | goto out_commit; | |
7279 | } | |
7280 | ||
7281 | numbytes = end - start; | |
7282 | memset(idata->id_data + start, 0, numbytes); | |
7283 | ||
7284 | /* | |
7285 | * No need to worry about the data page here - it's been | |
7286 | * truncated already and inline data doesn't need it for | |
7287 | * pushing zero's to disk, so we'll let readpage pick it up | |
7288 | * later. | |
7289 | */ | |
7290 | if (trunc) { | |
7291 | i_size_write(inode, start); | |
7292 | di->i_size = cpu_to_le64(start); | |
7293 | } | |
7294 | ||
7295 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
078cd827 | 7296 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1afc32b9 MF |
7297 | |
7298 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7299 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7300 | ||
6fdb702d | 7301 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
1afc32b9 MF |
7302 | ocfs2_journal_dirty(handle, di_bh); |
7303 | ||
7304 | out_commit: | |
7305 | ocfs2_commit_trans(osb, handle); | |
7306 | ||
7307 | out: | |
7308 | return ret; | |
7309 | } | |
e80de36d TM |
7310 | |
7311 | static int ocfs2_trim_extent(struct super_block *sb, | |
7312 | struct ocfs2_group_desc *gd, | |
7313 | u32 start, u32 count) | |
7314 | { | |
7315 | u64 discard, bcount; | |
7316 | ||
7317 | bcount = ocfs2_clusters_to_blocks(sb, count); | |
7318 | discard = le64_to_cpu(gd->bg_blkno) + | |
7319 | ocfs2_clusters_to_blocks(sb, start); | |
7320 | ||
10fca35f TM |
7321 | trace_ocfs2_trim_extent(sb, (unsigned long long)discard, bcount); |
7322 | ||
e80de36d TM |
7323 | return sb_issue_discard(sb, discard, bcount, GFP_NOFS, 0); |
7324 | } | |
7325 | ||
7326 | static int ocfs2_trim_group(struct super_block *sb, | |
7327 | struct ocfs2_group_desc *gd, | |
7328 | u32 start, u32 max, u32 minbits) | |
7329 | { | |
7330 | int ret = 0, count = 0, next; | |
7331 | void *bitmap = gd->bg_bitmap; | |
7332 | ||
7333 | if (le16_to_cpu(gd->bg_free_bits_count) < minbits) | |
7334 | return 0; | |
7335 | ||
10fca35f TM |
7336 | trace_ocfs2_trim_group((unsigned long long)le64_to_cpu(gd->bg_blkno), |
7337 | start, max, minbits); | |
7338 | ||
e80de36d TM |
7339 | while (start < max) { |
7340 | start = ocfs2_find_next_zero_bit(bitmap, max, start); | |
7341 | if (start >= max) | |
7342 | break; | |
7343 | next = ocfs2_find_next_bit(bitmap, max, start); | |
7344 | ||
7345 | if ((next - start) >= minbits) { | |
7346 | ret = ocfs2_trim_extent(sb, gd, | |
7347 | start, next - start); | |
7348 | if (ret < 0) { | |
7349 | mlog_errno(ret); | |
7350 | break; | |
7351 | } | |
7352 | count += next - start; | |
7353 | } | |
7354 | start = next + 1; | |
7355 | ||
7356 | if (fatal_signal_pending(current)) { | |
7357 | count = -ERESTARTSYS; | |
7358 | break; | |
7359 | } | |
7360 | ||
7361 | if ((le16_to_cpu(gd->bg_free_bits_count) - count) < minbits) | |
7362 | break; | |
7363 | } | |
7364 | ||
7365 | if (ret < 0) | |
7366 | count = ret; | |
7367 | ||
7368 | return count; | |
7369 | } | |
7370 | ||
7371 | int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range) | |
7372 | { | |
7373 | struct ocfs2_super *osb = OCFS2_SB(sb); | |
7374 | u64 start, len, trimmed, first_group, last_group, group; | |
7375 | int ret, cnt; | |
7376 | u32 first_bit, last_bit, minlen; | |
7377 | struct buffer_head *main_bm_bh = NULL; | |
7378 | struct inode *main_bm_inode = NULL; | |
7379 | struct buffer_head *gd_bh = NULL; | |
7380 | struct ocfs2_dinode *main_bm; | |
7381 | struct ocfs2_group_desc *gd = NULL; | |
7382 | ||
7383 | start = range->start >> osb->s_clustersize_bits; | |
7384 | len = range->len >> osb->s_clustersize_bits; | |
7385 | minlen = range->minlen >> osb->s_clustersize_bits; | |
e80de36d | 7386 | |
aa89762c | 7387 | if (minlen >= osb->bitmap_cpg || range->len < sb->s_blocksize) |
e80de36d TM |
7388 | return -EINVAL; |
7389 | ||
7390 | main_bm_inode = ocfs2_get_system_file_inode(osb, | |
7391 | GLOBAL_BITMAP_SYSTEM_INODE, | |
7392 | OCFS2_INVALID_SLOT); | |
7393 | if (!main_bm_inode) { | |
7394 | ret = -EIO; | |
7395 | mlog_errno(ret); | |
7396 | goto out; | |
7397 | } | |
7398 | ||
5955102c | 7399 | inode_lock(main_bm_inode); |
e80de36d TM |
7400 | |
7401 | ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 0); | |
7402 | if (ret < 0) { | |
7403 | mlog_errno(ret); | |
7404 | goto out_mutex; | |
7405 | } | |
7406 | main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data; | |
7407 | ||
7408 | if (start >= le32_to_cpu(main_bm->i_clusters)) { | |
7409 | ret = -EINVAL; | |
7410 | goto out_unlock; | |
7411 | } | |
7412 | ||
aa89762c | 7413 | len = range->len >> osb->s_clustersize_bits; |
e80de36d TM |
7414 | if (start + len > le32_to_cpu(main_bm->i_clusters)) |
7415 | len = le32_to_cpu(main_bm->i_clusters) - start; | |
7416 | ||
10fca35f TM |
7417 | trace_ocfs2_trim_fs(start, len, minlen); |
7418 | ||
e80de36d TM |
7419 | /* Determine first and last group to examine based on start and len */ |
7420 | first_group = ocfs2_which_cluster_group(main_bm_inode, start); | |
7421 | if (first_group == osb->first_cluster_group_blkno) | |
7422 | first_bit = start; | |
7423 | else | |
7424 | first_bit = start - ocfs2_blocks_to_clusters(sb, first_group); | |
7425 | last_group = ocfs2_which_cluster_group(main_bm_inode, start + len - 1); | |
7426 | last_bit = osb->bitmap_cpg; | |
7427 | ||
aa89762c | 7428 | trimmed = 0; |
e80de36d TM |
7429 | for (group = first_group; group <= last_group;) { |
7430 | if (first_bit + len >= osb->bitmap_cpg) | |
7431 | last_bit = osb->bitmap_cpg; | |
7432 | else | |
7433 | last_bit = first_bit + len; | |
7434 | ||
7435 | ret = ocfs2_read_group_descriptor(main_bm_inode, | |
7436 | main_bm, group, | |
7437 | &gd_bh); | |
7438 | if (ret < 0) { | |
7439 | mlog_errno(ret); | |
7440 | break; | |
7441 | } | |
7442 | ||
7443 | gd = (struct ocfs2_group_desc *)gd_bh->b_data; | |
7444 | cnt = ocfs2_trim_group(sb, gd, first_bit, last_bit, minlen); | |
7445 | brelse(gd_bh); | |
7446 | gd_bh = NULL; | |
7447 | if (cnt < 0) { | |
7448 | ret = cnt; | |
7449 | mlog_errno(ret); | |
7450 | break; | |
7451 | } | |
7452 | ||
7453 | trimmed += cnt; | |
7454 | len -= osb->bitmap_cpg - first_bit; | |
7455 | first_bit = 0; | |
7456 | if (group == osb->first_cluster_group_blkno) | |
7457 | group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg); | |
7458 | else | |
7459 | group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg); | |
7460 | } | |
7461 | range->len = trimmed * sb->s_blocksize; | |
7462 | out_unlock: | |
7463 | ocfs2_inode_unlock(main_bm_inode, 0); | |
7464 | brelse(main_bm_bh); | |
7465 | out_mutex: | |
5955102c | 7466 | inode_unlock(main_bm_inode); |
e80de36d TM |
7467 | iput(main_bm_inode); |
7468 | out: | |
7469 | return ret; | |
7470 | } |