]>
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> |
ccd979bd MF |
32 | |
33 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | |
34 | #include <cluster/masklog.h> | |
35 | ||
36 | #include "ocfs2.h" | |
37 | ||
38 | #include "alloc.h" | |
60b11392 | 39 | #include "aops.h" |
d6b32bbb | 40 | #include "blockcheck.h" |
ccd979bd MF |
41 | #include "dlmglue.h" |
42 | #include "extent_map.h" | |
43 | #include "inode.h" | |
44 | #include "journal.h" | |
45 | #include "localalloc.h" | |
46 | #include "suballoc.h" | |
47 | #include "sysfile.h" | |
48 | #include "file.h" | |
49 | #include "super.h" | |
50 | #include "uptodate.h" | |
2a50a743 | 51 | #include "xattr.h" |
ccd979bd MF |
52 | |
53 | #include "buffer_head_io.h" | |
54 | ||
e7d4cb6b | 55 | |
1625f8ac JB |
56 | /* |
57 | * Operations for a specific extent tree type. | |
58 | * | |
59 | * To implement an on-disk btree (extent tree) type in ocfs2, add | |
60 | * an ocfs2_extent_tree_operations structure and the matching | |
8d6220d6 | 61 | * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it |
1625f8ac JB |
62 | * for the allocation portion of the extent tree. |
63 | */ | |
e7d4cb6b | 64 | struct ocfs2_extent_tree_operations { |
1625f8ac JB |
65 | /* |
66 | * last_eb_blk is the block number of the right most leaf extent | |
67 | * block. Most on-disk structures containing an extent tree store | |
68 | * this value for fast access. The ->eo_set_last_eb_blk() and | |
69 | * ->eo_get_last_eb_blk() operations access this value. They are | |
70 | * both required. | |
71 | */ | |
35dc0aa3 JB |
72 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
73 | u64 blkno); | |
74 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
1625f8ac JB |
75 | |
76 | /* | |
77 | * The on-disk structure usually keeps track of how many total | |
78 | * clusters are stored in this extent tree. This function updates | |
79 | * that value. new_clusters is the delta, and must be | |
80 | * added to the total. Required. | |
81 | */ | |
6136ca5f | 82 | void (*eo_update_clusters)(struct ocfs2_extent_tree *et, |
35dc0aa3 | 83 | u32 new_clusters); |
1625f8ac | 84 | |
4c911eef JB |
85 | /* |
86 | * If this extent tree is supported by an extent map, truncate the | |
87 | * map to clusters, | |
88 | */ | |
89 | void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et, | |
90 | u32 clusters); | |
91 | ||
1625f8ac JB |
92 | /* |
93 | * If ->eo_insert_check() exists, it is called before rec is | |
94 | * inserted into the extent tree. It is optional. | |
95 | */ | |
6136ca5f | 96 | int (*eo_insert_check)(struct ocfs2_extent_tree *et, |
1e61ee79 | 97 | struct ocfs2_extent_rec *rec); |
6136ca5f | 98 | int (*eo_sanity_check)(struct ocfs2_extent_tree *et); |
0ce1010f | 99 | |
1625f8ac JB |
100 | /* |
101 | * -------------------------------------------------------------- | |
102 | * The remaining are internal to ocfs2_extent_tree and don't have | |
103 | * accessor functions | |
104 | */ | |
105 | ||
106 | /* | |
107 | * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el. | |
108 | * It is required. | |
109 | */ | |
0ce1010f | 110 | void (*eo_fill_root_el)(struct ocfs2_extent_tree *et); |
1625f8ac JB |
111 | |
112 | /* | |
113 | * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if | |
114 | * it exists. If it does not, et->et_max_leaf_clusters is set | |
115 | * to 0 (unlimited). Optional. | |
116 | */ | |
6136ca5f | 117 | void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et); |
e7d4cb6b TM |
118 | }; |
119 | ||
e7d4cb6b | 120 | |
f99b9b7c JB |
121 | /* |
122 | * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check | |
123 | * in the methods. | |
124 | */ | |
125 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et); | |
126 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
127 | u64 blkno); | |
6136ca5f | 128 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
f99b9b7c | 129 | u32 clusters); |
4c911eef JB |
130 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
131 | u32 clusters); | |
6136ca5f | 132 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
f99b9b7c | 133 | struct ocfs2_extent_rec *rec); |
6136ca5f | 134 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et); |
f99b9b7c JB |
135 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et); |
136 | static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { | |
137 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, | |
138 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
139 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
4c911eef | 140 | .eo_extent_map_truncate = ocfs2_dinode_extent_map_truncate, |
f99b9b7c JB |
141 | .eo_insert_check = ocfs2_dinode_insert_check, |
142 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
143 | .eo_fill_root_el = ocfs2_dinode_fill_root_el, | |
144 | }; | |
0ce1010f | 145 | |
e7d4cb6b TM |
146 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, |
147 | u64 blkno) | |
148 | { | |
ea5efa15 | 149 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 150 | |
f99b9b7c | 151 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
152 | di->i_last_eb_blk = cpu_to_le64(blkno); |
153 | } | |
154 | ||
155 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
156 | { | |
ea5efa15 | 157 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 158 | |
f99b9b7c | 159 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
160 | return le64_to_cpu(di->i_last_eb_blk); |
161 | } | |
162 | ||
6136ca5f | 163 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
e7d4cb6b TM |
164 | u32 clusters) |
165 | { | |
6136ca5f | 166 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
ea5efa15 | 167 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b TM |
168 | |
169 | le32_add_cpu(&di->i_clusters, clusters); | |
6136ca5f JB |
170 | spin_lock(&oi->ip_lock); |
171 | oi->ip_clusters = le32_to_cpu(di->i_clusters); | |
172 | spin_unlock(&oi->ip_lock); | |
e7d4cb6b TM |
173 | } |
174 | ||
4c911eef JB |
175 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
176 | u32 clusters) | |
177 | { | |
178 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
179 | ||
180 | ocfs2_extent_map_trunc(inode, clusters); | |
181 | } | |
182 | ||
6136ca5f | 183 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
184 | struct ocfs2_extent_rec *rec) |
185 | { | |
6136ca5f JB |
186 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
187 | struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb); | |
1e61ee79 | 188 | |
6136ca5f | 189 | BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL); |
1e61ee79 | 190 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && |
6136ca5f | 191 | (oi->ip_clusters != le32_to_cpu(rec->e_cpos)), |
1e61ee79 JB |
192 | "Device %s, asking for sparse allocation: inode %llu, " |
193 | "cpos %u, clusters %u\n", | |
194 | osb->dev_str, | |
6136ca5f JB |
195 | (unsigned long long)oi->ip_blkno, |
196 | rec->e_cpos, oi->ip_clusters); | |
1e61ee79 JB |
197 | |
198 | return 0; | |
199 | } | |
200 | ||
6136ca5f | 201 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 202 | { |
10995aa2 | 203 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 204 | |
f99b9b7c | 205 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
10995aa2 | 206 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); |
e7d4cb6b | 207 | |
10995aa2 | 208 | return 0; |
e7d4cb6b TM |
209 | } |
210 | ||
f99b9b7c JB |
211 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et) |
212 | { | |
213 | struct ocfs2_dinode *di = et->et_object; | |
214 | ||
215 | et->et_root_el = &di->id2.i_list; | |
216 | } | |
217 | ||
e7d4cb6b | 218 | |
0ce1010f JB |
219 | static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et) |
220 | { | |
2a50a743 | 221 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
0ce1010f | 222 | |
2a50a743 | 223 | et->et_root_el = &vb->vb_xv->xr_list; |
0ce1010f JB |
224 | } |
225 | ||
f56654c4 TM |
226 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
227 | u64 blkno) | |
228 | { | |
2a50a743 | 229 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 230 | |
2a50a743 | 231 | vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno); |
f56654c4 TM |
232 | } |
233 | ||
234 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
235 | { | |
2a50a743 | 236 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 237 | |
2a50a743 | 238 | return le64_to_cpu(vb->vb_xv->xr_last_eb_blk); |
f56654c4 TM |
239 | } |
240 | ||
6136ca5f | 241 | static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et, |
f56654c4 TM |
242 | u32 clusters) |
243 | { | |
2a50a743 | 244 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 245 | |
2a50a743 | 246 | le32_add_cpu(&vb->vb_xv->xr_clusters, clusters); |
f56654c4 TM |
247 | } |
248 | ||
1a09f556 | 249 | static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = { |
35dc0aa3 JB |
250 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
251 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
252 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
0ce1010f | 253 | .eo_fill_root_el = ocfs2_xattr_value_fill_root_el, |
f56654c4 TM |
254 | }; |
255 | ||
0ce1010f JB |
256 | static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et) |
257 | { | |
258 | struct ocfs2_xattr_block *xb = et->et_object; | |
259 | ||
260 | et->et_root_el = &xb->xb_attrs.xb_root.xt_list; | |
261 | } | |
262 | ||
6136ca5f | 263 | static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et) |
943cced3 | 264 | { |
6136ca5f | 265 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
943cced3 | 266 | et->et_max_leaf_clusters = |
6136ca5f | 267 | ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE); |
943cced3 JB |
268 | } |
269 | ||
ba492615 TM |
270 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
271 | u64 blkno) | |
272 | { | |
ea5efa15 | 273 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
274 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
275 | ||
276 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
277 | } | |
278 | ||
279 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
280 | { | |
ea5efa15 | 281 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
282 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
283 | ||
284 | return le64_to_cpu(xt->xt_last_eb_blk); | |
285 | } | |
286 | ||
6136ca5f | 287 | static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et, |
ba492615 TM |
288 | u32 clusters) |
289 | { | |
ea5efa15 | 290 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
291 | |
292 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
293 | } | |
294 | ||
ba492615 | 295 | static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { |
35dc0aa3 JB |
296 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
297 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
298 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
0ce1010f | 299 | .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el, |
943cced3 | 300 | .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters, |
ba492615 TM |
301 | }; |
302 | ||
9b7895ef MF |
303 | static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et, |
304 | u64 blkno) | |
305 | { | |
306 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
307 | ||
308 | dx_root->dr_last_eb_blk = cpu_to_le64(blkno); | |
309 | } | |
310 | ||
311 | static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
312 | { | |
313 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
314 | ||
315 | return le64_to_cpu(dx_root->dr_last_eb_blk); | |
316 | } | |
317 | ||
6136ca5f | 318 | static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et, |
9b7895ef MF |
319 | u32 clusters) |
320 | { | |
321 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
322 | ||
323 | le32_add_cpu(&dx_root->dr_clusters, clusters); | |
324 | } | |
325 | ||
6136ca5f | 326 | static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et) |
9b7895ef MF |
327 | { |
328 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
329 | ||
330 | BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root)); | |
331 | ||
332 | return 0; | |
333 | } | |
334 | ||
335 | static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et) | |
336 | { | |
337 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
338 | ||
339 | et->et_root_el = &dx_root->dr_list; | |
340 | } | |
341 | ||
342 | static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = { | |
343 | .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk, | |
344 | .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk, | |
345 | .eo_update_clusters = ocfs2_dx_root_update_clusters, | |
346 | .eo_sanity_check = ocfs2_dx_root_sanity_check, | |
347 | .eo_fill_root_el = ocfs2_dx_root_fill_root_el, | |
348 | }; | |
349 | ||
8d6220d6 JB |
350 | static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et, |
351 | struct inode *inode, | |
352 | struct buffer_head *bh, | |
13723d00 | 353 | ocfs2_journal_access_func access, |
8d6220d6 JB |
354 | void *obj, |
355 | struct ocfs2_extent_tree_operations *ops) | |
e7d4cb6b | 356 | { |
1a09f556 | 357 | et->et_ops = ops; |
ce1d9ea6 | 358 | et->et_root_bh = bh; |
d9a0a1f8 | 359 | et->et_ci = INODE_CACHE(inode); |
13723d00 | 360 | et->et_root_journal_access = access; |
ea5efa15 JB |
361 | if (!obj) |
362 | obj = (void *)bh->b_data; | |
363 | et->et_object = obj; | |
e7d4cb6b | 364 | |
0ce1010f | 365 | et->et_ops->eo_fill_root_el(et); |
943cced3 JB |
366 | if (!et->et_ops->eo_fill_max_leaf_clusters) |
367 | et->et_max_leaf_clusters = 0; | |
368 | else | |
6136ca5f | 369 | et->et_ops->eo_fill_max_leaf_clusters(et); |
e7d4cb6b TM |
370 | } |
371 | ||
8d6220d6 JB |
372 | void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et, |
373 | struct inode *inode, | |
374 | struct buffer_head *bh) | |
1a09f556 | 375 | { |
13723d00 JB |
376 | __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_di, |
377 | NULL, &ocfs2_dinode_et_ops); | |
1a09f556 JB |
378 | } |
379 | ||
8d6220d6 | 380 | void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et, |
f99b9b7c | 381 | struct inode *inode, |
8d6220d6 | 382 | struct buffer_head *bh) |
1a09f556 | 383 | { |
13723d00 JB |
384 | __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_xb, |
385 | NULL, &ocfs2_xattr_tree_et_ops); | |
1a09f556 JB |
386 | } |
387 | ||
8d6220d6 JB |
388 | void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et, |
389 | struct inode *inode, | |
2a50a743 | 390 | struct ocfs2_xattr_value_buf *vb) |
e7d4cb6b | 391 | { |
2a50a743 | 392 | __ocfs2_init_extent_tree(et, inode, vb->vb_bh, vb->vb_access, vb, |
8d6220d6 | 393 | &ocfs2_xattr_value_et_ops); |
e7d4cb6b TM |
394 | } |
395 | ||
9b7895ef MF |
396 | void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et, |
397 | struct inode *inode, | |
398 | struct buffer_head *bh) | |
399 | { | |
400 | __ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_dr, | |
401 | NULL, &ocfs2_dx_root_et_ops); | |
402 | } | |
403 | ||
35dc0aa3 JB |
404 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
405 | u64 new_last_eb_blk) | |
e7d4cb6b | 406 | { |
ce1d9ea6 | 407 | et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
408 | } |
409 | ||
35dc0aa3 | 410 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 411 | { |
ce1d9ea6 | 412 | return et->et_ops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
413 | } |
414 | ||
6136ca5f | 415 | static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et, |
35dc0aa3 JB |
416 | u32 clusters) |
417 | { | |
6136ca5f | 418 | et->et_ops->eo_update_clusters(et, clusters); |
35dc0aa3 JB |
419 | } |
420 | ||
4c911eef JB |
421 | static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et, |
422 | u32 clusters) | |
423 | { | |
424 | if (et->et_ops->eo_extent_map_truncate) | |
425 | et->et_ops->eo_extent_map_truncate(et, clusters); | |
426 | } | |
427 | ||
13723d00 | 428 | static inline int ocfs2_et_root_journal_access(handle_t *handle, |
13723d00 JB |
429 | struct ocfs2_extent_tree *et, |
430 | int type) | |
431 | { | |
d9a0a1f8 | 432 | return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh, |
13723d00 JB |
433 | type); |
434 | } | |
435 | ||
6136ca5f | 436 | static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
437 | struct ocfs2_extent_rec *rec) |
438 | { | |
439 | int ret = 0; | |
440 | ||
441 | if (et->et_ops->eo_insert_check) | |
6136ca5f | 442 | ret = et->et_ops->eo_insert_check(et, rec); |
1e61ee79 JB |
443 | return ret; |
444 | } | |
445 | ||
6136ca5f | 446 | static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 447 | { |
1e61ee79 JB |
448 | int ret = 0; |
449 | ||
450 | if (et->et_ops->eo_sanity_check) | |
6136ca5f | 451 | ret = et->et_ops->eo_sanity_check(et); |
1e61ee79 | 452 | return ret; |
e7d4cb6b TM |
453 | } |
454 | ||
dcd0538f | 455 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); |
59a5e416 MF |
456 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
457 | struct ocfs2_extent_block *eb); | |
ccd979bd | 458 | |
dcd0538f MF |
459 | /* |
460 | * Structures which describe a path through a btree, and functions to | |
461 | * manipulate them. | |
462 | * | |
463 | * The idea here is to be as generic as possible with the tree | |
464 | * manipulation code. | |
465 | */ | |
466 | struct ocfs2_path_item { | |
467 | struct buffer_head *bh; | |
468 | struct ocfs2_extent_list *el; | |
469 | }; | |
ccd979bd | 470 | |
dcd0538f | 471 | #define OCFS2_MAX_PATH_DEPTH 5 |
ccd979bd | 472 | |
dcd0538f | 473 | struct ocfs2_path { |
13723d00 JB |
474 | int p_tree_depth; |
475 | ocfs2_journal_access_func p_root_access; | |
476 | struct ocfs2_path_item p_node[OCFS2_MAX_PATH_DEPTH]; | |
dcd0538f | 477 | }; |
ccd979bd | 478 | |
dcd0538f MF |
479 | #define path_root_bh(_path) ((_path)->p_node[0].bh) |
480 | #define path_root_el(_path) ((_path)->p_node[0].el) | |
13723d00 | 481 | #define path_root_access(_path)((_path)->p_root_access) |
dcd0538f MF |
482 | #define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh) |
483 | #define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el) | |
484 | #define path_num_items(_path) ((_path)->p_tree_depth + 1) | |
ccd979bd | 485 | |
facdb77f JB |
486 | static int ocfs2_find_path(struct ocfs2_caching_info *ci, |
487 | struct ocfs2_path *path, u32 cpos); | |
d401dc12 JB |
488 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
489 | struct ocfs2_extent_tree *et, | |
6b791bcc TM |
490 | struct ocfs2_path *path, |
491 | struct ocfs2_extent_rec *insert_rec); | |
dcd0538f MF |
492 | /* |
493 | * Reset the actual path elements so that we can re-use the structure | |
494 | * to build another path. Generally, this involves freeing the buffer | |
495 | * heads. | |
496 | */ | |
497 | static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) | |
498 | { | |
499 | int i, start = 0, depth = 0; | |
500 | struct ocfs2_path_item *node; | |
ccd979bd | 501 | |
dcd0538f MF |
502 | if (keep_root) |
503 | start = 1; | |
ccd979bd | 504 | |
dcd0538f MF |
505 | for(i = start; i < path_num_items(path); i++) { |
506 | node = &path->p_node[i]; | |
507 | ||
508 | brelse(node->bh); | |
509 | node->bh = NULL; | |
510 | node->el = NULL; | |
511 | } | |
512 | ||
513 | /* | |
514 | * Tree depth may change during truncate, or insert. If we're | |
515 | * keeping the root extent list, then make sure that our path | |
516 | * structure reflects the proper depth. | |
517 | */ | |
518 | if (keep_root) | |
519 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
13723d00 JB |
520 | else |
521 | path_root_access(path) = NULL; | |
dcd0538f MF |
522 | |
523 | path->p_tree_depth = depth; | |
524 | } | |
525 | ||
526 | static void ocfs2_free_path(struct ocfs2_path *path) | |
527 | { | |
528 | if (path) { | |
529 | ocfs2_reinit_path(path, 0); | |
530 | kfree(path); | |
531 | } | |
532 | } | |
533 | ||
328d5752 MF |
534 | /* |
535 | * All the elements of src into dest. After this call, src could be freed | |
536 | * without affecting dest. | |
537 | * | |
538 | * Both paths should have the same root. Any non-root elements of dest | |
539 | * will be freed. | |
540 | */ | |
541 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
542 | { | |
543 | int i; | |
544 | ||
545 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
546 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
13723d00 | 547 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
328d5752 MF |
548 | |
549 | ocfs2_reinit_path(dest, 1); | |
550 | ||
551 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
552 | dest->p_node[i].bh = src->p_node[i].bh; | |
553 | dest->p_node[i].el = src->p_node[i].el; | |
554 | ||
555 | if (dest->p_node[i].bh) | |
556 | get_bh(dest->p_node[i].bh); | |
557 | } | |
558 | } | |
559 | ||
dcd0538f MF |
560 | /* |
561 | * Make the *dest path the same as src and re-initialize src path to | |
562 | * have a root only. | |
563 | */ | |
564 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
565 | { | |
566 | int i; | |
567 | ||
568 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
13723d00 | 569 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
dcd0538f MF |
570 | |
571 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
572 | brelse(dest->p_node[i].bh); | |
573 | ||
574 | dest->p_node[i].bh = src->p_node[i].bh; | |
575 | dest->p_node[i].el = src->p_node[i].el; | |
576 | ||
577 | src->p_node[i].bh = NULL; | |
578 | src->p_node[i].el = NULL; | |
579 | } | |
580 | } | |
581 | ||
582 | /* | |
583 | * Insert an extent block at given index. | |
584 | * | |
585 | * This will not take an additional reference on eb_bh. | |
586 | */ | |
587 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
588 | struct buffer_head *eb_bh) | |
589 | { | |
590 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
591 | ||
592 | /* | |
593 | * Right now, no root bh is an extent block, so this helps | |
594 | * catch code errors with dinode trees. The assertion can be | |
595 | * safely removed if we ever need to insert extent block | |
596 | * structures at the root. | |
597 | */ | |
598 | BUG_ON(index == 0); | |
599 | ||
600 | path->p_node[index].bh = eb_bh; | |
601 | path->p_node[index].el = &eb->h_list; | |
602 | } | |
603 | ||
604 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
13723d00 JB |
605 | struct ocfs2_extent_list *root_el, |
606 | ocfs2_journal_access_func access) | |
dcd0538f MF |
607 | { |
608 | struct ocfs2_path *path; | |
ccd979bd | 609 | |
dcd0538f MF |
610 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
611 | ||
612 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
613 | if (path) { | |
614 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
615 | get_bh(root_bh); | |
616 | path_root_bh(path) = root_bh; | |
617 | path_root_el(path) = root_el; | |
13723d00 | 618 | path_root_access(path) = access; |
dcd0538f MF |
619 | } |
620 | ||
621 | return path; | |
622 | } | |
623 | ||
ffdd7a54 JB |
624 | static struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path) |
625 | { | |
13723d00 JB |
626 | return ocfs2_new_path(path_root_bh(path), path_root_el(path), |
627 | path_root_access(path)); | |
ffdd7a54 JB |
628 | } |
629 | ||
630 | static struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et) | |
631 | { | |
13723d00 JB |
632 | return ocfs2_new_path(et->et_root_bh, et->et_root_el, |
633 | et->et_root_journal_access); | |
634 | } | |
635 | ||
636 | /* | |
637 | * Journal the buffer at depth idx. All idx>0 are extent_blocks, | |
638 | * otherwise it's the root_access function. | |
639 | * | |
640 | * I don't like the way this function's name looks next to | |
641 | * ocfs2_journal_access_path(), but I don't have a better one. | |
642 | */ | |
643 | static int ocfs2_path_bh_journal_access(handle_t *handle, | |
0cf2f763 | 644 | struct ocfs2_caching_info *ci, |
13723d00 JB |
645 | struct ocfs2_path *path, |
646 | int idx) | |
647 | { | |
648 | ocfs2_journal_access_func access = path_root_access(path); | |
649 | ||
650 | if (!access) | |
651 | access = ocfs2_journal_access; | |
652 | ||
653 | if (idx) | |
654 | access = ocfs2_journal_access_eb; | |
655 | ||
0cf2f763 | 656 | return access(handle, ci, path->p_node[idx].bh, |
13723d00 | 657 | OCFS2_JOURNAL_ACCESS_WRITE); |
ffdd7a54 JB |
658 | } |
659 | ||
dcd0538f MF |
660 | /* |
661 | * Convenience function to journal all components in a path. | |
662 | */ | |
0cf2f763 JB |
663 | static int ocfs2_journal_access_path(struct ocfs2_caching_info *ci, |
664 | handle_t *handle, | |
dcd0538f MF |
665 | struct ocfs2_path *path) |
666 | { | |
667 | int i, ret = 0; | |
668 | ||
669 | if (!path) | |
670 | goto out; | |
671 | ||
672 | for(i = 0; i < path_num_items(path); i++) { | |
0cf2f763 | 673 | ret = ocfs2_path_bh_journal_access(handle, ci, path, i); |
dcd0538f MF |
674 | if (ret < 0) { |
675 | mlog_errno(ret); | |
676 | goto out; | |
677 | } | |
678 | } | |
679 | ||
680 | out: | |
681 | return ret; | |
682 | } | |
683 | ||
328d5752 MF |
684 | /* |
685 | * Return the index of the extent record which contains cluster #v_cluster. | |
686 | * -1 is returned if it was not found. | |
687 | * | |
688 | * Should work fine on interior and exterior nodes. | |
689 | */ | |
690 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
691 | { | |
692 | int ret = -1; | |
693 | int i; | |
694 | struct ocfs2_extent_rec *rec; | |
695 | u32 rec_end, rec_start, clusters; | |
696 | ||
697 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
698 | rec = &el->l_recs[i]; | |
699 | ||
700 | rec_start = le32_to_cpu(rec->e_cpos); | |
701 | clusters = ocfs2_rec_clusters(el, rec); | |
702 | ||
703 | rec_end = rec_start + clusters; | |
704 | ||
705 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
706 | ret = i; | |
707 | break; | |
708 | } | |
709 | } | |
710 | ||
711 | return ret; | |
712 | } | |
713 | ||
dcd0538f MF |
714 | enum ocfs2_contig_type { |
715 | CONTIG_NONE = 0, | |
716 | CONTIG_LEFT, | |
328d5752 MF |
717 | CONTIG_RIGHT, |
718 | CONTIG_LEFTRIGHT, | |
dcd0538f MF |
719 | }; |
720 | ||
e48edee2 MF |
721 | |
722 | /* | |
723 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
724 | * ocfs2_extent_contig only work properly against leaf nodes! | |
725 | */ | |
dcd0538f MF |
726 | static int ocfs2_block_extent_contig(struct super_block *sb, |
727 | struct ocfs2_extent_rec *ext, | |
728 | u64 blkno) | |
ccd979bd | 729 | { |
e48edee2 MF |
730 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
731 | ||
732 | blk_end += ocfs2_clusters_to_blocks(sb, | |
733 | le16_to_cpu(ext->e_leaf_clusters)); | |
734 | ||
735 | return blkno == blk_end; | |
ccd979bd MF |
736 | } |
737 | ||
dcd0538f MF |
738 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
739 | struct ocfs2_extent_rec *right) | |
740 | { | |
e48edee2 MF |
741 | u32 left_range; |
742 | ||
743 | left_range = le32_to_cpu(left->e_cpos) + | |
744 | le16_to_cpu(left->e_leaf_clusters); | |
745 | ||
746 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
747 | } |
748 | ||
749 | static enum ocfs2_contig_type | |
b4a17651 | 750 | ocfs2_extent_contig(struct super_block *sb, |
dcd0538f MF |
751 | struct ocfs2_extent_rec *ext, |
752 | struct ocfs2_extent_rec *insert_rec) | |
753 | { | |
754 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
755 | ||
328d5752 MF |
756 | /* |
757 | * Refuse to coalesce extent records with different flag | |
758 | * fields - we don't want to mix unwritten extents with user | |
759 | * data. | |
760 | */ | |
761 | if (ext->e_flags != insert_rec->e_flags) | |
762 | return CONTIG_NONE; | |
763 | ||
dcd0538f | 764 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
b4a17651 | 765 | ocfs2_block_extent_contig(sb, ext, blkno)) |
dcd0538f MF |
766 | return CONTIG_RIGHT; |
767 | ||
768 | blkno = le64_to_cpu(ext->e_blkno); | |
769 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
b4a17651 | 770 | ocfs2_block_extent_contig(sb, insert_rec, blkno)) |
dcd0538f MF |
771 | return CONTIG_LEFT; |
772 | ||
773 | return CONTIG_NONE; | |
774 | } | |
775 | ||
776 | /* | |
777 | * NOTE: We can have pretty much any combination of contiguousness and | |
778 | * appending. | |
779 | * | |
780 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
781 | * we'll have to update the path to that leaf. | |
782 | */ | |
783 | enum ocfs2_append_type { | |
784 | APPEND_NONE = 0, | |
785 | APPEND_TAIL, | |
786 | }; | |
787 | ||
328d5752 MF |
788 | enum ocfs2_split_type { |
789 | SPLIT_NONE = 0, | |
790 | SPLIT_LEFT, | |
791 | SPLIT_RIGHT, | |
792 | }; | |
793 | ||
dcd0538f | 794 | struct ocfs2_insert_type { |
328d5752 | 795 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
796 | enum ocfs2_append_type ins_appending; |
797 | enum ocfs2_contig_type ins_contig; | |
798 | int ins_contig_index; | |
dcd0538f MF |
799 | int ins_tree_depth; |
800 | }; | |
801 | ||
328d5752 MF |
802 | struct ocfs2_merge_ctxt { |
803 | enum ocfs2_contig_type c_contig_type; | |
804 | int c_has_empty_extent; | |
805 | int c_split_covers_rec; | |
328d5752 MF |
806 | }; |
807 | ||
5e96581a JB |
808 | static int ocfs2_validate_extent_block(struct super_block *sb, |
809 | struct buffer_head *bh) | |
810 | { | |
d6b32bbb | 811 | int rc; |
5e96581a JB |
812 | struct ocfs2_extent_block *eb = |
813 | (struct ocfs2_extent_block *)bh->b_data; | |
814 | ||
970e4936 JB |
815 | mlog(0, "Validating extent block %llu\n", |
816 | (unsigned long long)bh->b_blocknr); | |
817 | ||
d6b32bbb JB |
818 | BUG_ON(!buffer_uptodate(bh)); |
819 | ||
820 | /* | |
821 | * If the ecc fails, we return the error but otherwise | |
822 | * leave the filesystem running. We know any error is | |
823 | * local to this block. | |
824 | */ | |
825 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check); | |
13723d00 JB |
826 | if (rc) { |
827 | mlog(ML_ERROR, "Checksum failed for extent block %llu\n", | |
828 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 829 | return rc; |
13723d00 | 830 | } |
d6b32bbb JB |
831 | |
832 | /* | |
833 | * Errors after here are fatal. | |
834 | */ | |
835 | ||
5e96581a JB |
836 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { |
837 | ocfs2_error(sb, | |
838 | "Extent block #%llu has bad signature %.*s", | |
839 | (unsigned long long)bh->b_blocknr, 7, | |
840 | eb->h_signature); | |
841 | return -EINVAL; | |
842 | } | |
843 | ||
844 | if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) { | |
845 | ocfs2_error(sb, | |
846 | "Extent block #%llu has an invalid h_blkno " | |
847 | "of %llu", | |
848 | (unsigned long long)bh->b_blocknr, | |
849 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
850 | return -EINVAL; | |
851 | } | |
852 | ||
853 | if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) { | |
854 | ocfs2_error(sb, | |
855 | "Extent block #%llu has an invalid " | |
856 | "h_fs_generation of #%u", | |
857 | (unsigned long long)bh->b_blocknr, | |
858 | le32_to_cpu(eb->h_fs_generation)); | |
859 | return -EINVAL; | |
860 | } | |
861 | ||
862 | return 0; | |
863 | } | |
864 | ||
3d03a305 | 865 | int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno, |
5e96581a JB |
866 | struct buffer_head **bh) |
867 | { | |
868 | int rc; | |
869 | struct buffer_head *tmp = *bh; | |
870 | ||
3d03a305 | 871 | rc = ocfs2_read_block(ci, eb_blkno, &tmp, |
970e4936 | 872 | ocfs2_validate_extent_block); |
5e96581a JB |
873 | |
874 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
970e4936 | 875 | if (!rc && !*bh) |
5e96581a JB |
876 | *bh = tmp; |
877 | ||
5e96581a JB |
878 | return rc; |
879 | } | |
880 | ||
881 | ||
ccd979bd MF |
882 | /* |
883 | * How many free extents have we got before we need more meta data? | |
884 | */ | |
885 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
f99b9b7c | 886 | struct ocfs2_extent_tree *et) |
ccd979bd MF |
887 | { |
888 | int retval; | |
e7d4cb6b | 889 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
890 | struct ocfs2_extent_block *eb; |
891 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 892 | u64 last_eb_blk = 0; |
ccd979bd MF |
893 | |
894 | mlog_entry_void(); | |
895 | ||
f99b9b7c JB |
896 | el = et->et_root_el; |
897 | last_eb_blk = ocfs2_et_get_last_eb_blk(et); | |
ccd979bd | 898 | |
e7d4cb6b | 899 | if (last_eb_blk) { |
3d03a305 JB |
900 | retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk, |
901 | &eb_bh); | |
ccd979bd MF |
902 | if (retval < 0) { |
903 | mlog_errno(retval); | |
904 | goto bail; | |
905 | } | |
906 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
907 | el = &eb->h_list; | |
e7d4cb6b | 908 | } |
ccd979bd MF |
909 | |
910 | BUG_ON(el->l_tree_depth != 0); | |
911 | ||
912 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
913 | bail: | |
a81cb88b | 914 | brelse(eb_bh); |
ccd979bd MF |
915 | |
916 | mlog_exit(retval); | |
917 | return retval; | |
918 | } | |
919 | ||
920 | /* expects array to already be allocated | |
921 | * | |
922 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
923 | * l_count for you | |
924 | */ | |
42a5a7a9 JB |
925 | static int ocfs2_create_new_meta_bhs(handle_t *handle, |
926 | struct ocfs2_extent_tree *et, | |
ccd979bd MF |
927 | int wanted, |
928 | struct ocfs2_alloc_context *meta_ac, | |
929 | struct buffer_head *bhs[]) | |
930 | { | |
931 | int count, status, i; | |
932 | u16 suballoc_bit_start; | |
933 | u32 num_got; | |
934 | u64 first_blkno; | |
42a5a7a9 JB |
935 | struct ocfs2_super *osb = |
936 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
ccd979bd MF |
937 | struct ocfs2_extent_block *eb; |
938 | ||
939 | mlog_entry_void(); | |
940 | ||
941 | count = 0; | |
942 | while (count < wanted) { | |
943 | status = ocfs2_claim_metadata(osb, | |
944 | handle, | |
945 | meta_ac, | |
946 | wanted - count, | |
947 | &suballoc_bit_start, | |
948 | &num_got, | |
949 | &first_blkno); | |
950 | if (status < 0) { | |
951 | mlog_errno(status); | |
952 | goto bail; | |
953 | } | |
954 | ||
955 | for(i = count; i < (num_got + count); i++) { | |
956 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
957 | if (bhs[i] == NULL) { | |
958 | status = -EIO; | |
959 | mlog_errno(status); | |
960 | goto bail; | |
961 | } | |
42a5a7a9 | 962 | ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]); |
ccd979bd | 963 | |
42a5a7a9 JB |
964 | status = ocfs2_journal_access_eb(handle, et->et_ci, |
965 | bhs[i], | |
13723d00 | 966 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
967 | if (status < 0) { |
968 | mlog_errno(status); | |
969 | goto bail; | |
970 | } | |
971 | ||
972 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
973 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
974 | /* Ok, setup the minimal stuff here. */ | |
975 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
976 | eb->h_blkno = cpu_to_le64(first_blkno); | |
977 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
ccd979bd | 978 | eb->h_suballoc_slot = cpu_to_le16(osb->slot_num); |
ccd979bd MF |
979 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
980 | eb->h_list.l_count = | |
981 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
982 | ||
983 | suballoc_bit_start++; | |
984 | first_blkno++; | |
985 | ||
986 | /* We'll also be dirtied by the caller, so | |
987 | * this isn't absolutely necessary. */ | |
988 | status = ocfs2_journal_dirty(handle, bhs[i]); | |
989 | if (status < 0) { | |
990 | mlog_errno(status); | |
991 | goto bail; | |
992 | } | |
993 | } | |
994 | ||
995 | count += num_got; | |
996 | } | |
997 | ||
998 | status = 0; | |
999 | bail: | |
1000 | if (status < 0) { | |
1001 | for(i = 0; i < wanted; i++) { | |
a81cb88b | 1002 | brelse(bhs[i]); |
ccd979bd MF |
1003 | bhs[i] = NULL; |
1004 | } | |
1005 | } | |
1006 | mlog_exit(status); | |
1007 | return status; | |
1008 | } | |
1009 | ||
dcd0538f MF |
1010 | /* |
1011 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
1012 | * | |
1013 | * Returns the sum of the rightmost extent rec logical offset and | |
1014 | * cluster count. | |
1015 | * | |
1016 | * ocfs2_add_branch() uses this to determine what logical cluster | |
1017 | * value should be populated into the leftmost new branch records. | |
1018 | * | |
1019 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
1020 | * value for the new topmost tree record. | |
1021 | */ | |
1022 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
1023 | { | |
1024 | int i; | |
1025 | ||
1026 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1027 | ||
1028 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 1029 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
1030 | } |
1031 | ||
6b791bcc TM |
1032 | /* |
1033 | * Change range of the branches in the right most path according to the leaf | |
1034 | * extent block's rightmost record. | |
1035 | */ | |
1036 | static int ocfs2_adjust_rightmost_branch(handle_t *handle, | |
6b791bcc TM |
1037 | struct ocfs2_extent_tree *et) |
1038 | { | |
1039 | int status; | |
1040 | struct ocfs2_path *path = NULL; | |
1041 | struct ocfs2_extent_list *el; | |
1042 | struct ocfs2_extent_rec *rec; | |
1043 | ||
1044 | path = ocfs2_new_path_from_et(et); | |
1045 | if (!path) { | |
1046 | status = -ENOMEM; | |
1047 | return status; | |
1048 | } | |
1049 | ||
facdb77f | 1050 | status = ocfs2_find_path(et->et_ci, path, UINT_MAX); |
6b791bcc TM |
1051 | if (status < 0) { |
1052 | mlog_errno(status); | |
1053 | goto out; | |
1054 | } | |
1055 | ||
1056 | status = ocfs2_extend_trans(handle, path_num_items(path) + | |
1057 | handle->h_buffer_credits); | |
1058 | if (status < 0) { | |
1059 | mlog_errno(status); | |
1060 | goto out; | |
1061 | } | |
1062 | ||
d401dc12 | 1063 | status = ocfs2_journal_access_path(et->et_ci, handle, path); |
6b791bcc TM |
1064 | if (status < 0) { |
1065 | mlog_errno(status); | |
1066 | goto out; | |
1067 | } | |
1068 | ||
1069 | el = path_leaf_el(path); | |
1070 | rec = &el->l_recs[le32_to_cpu(el->l_next_free_rec) - 1]; | |
1071 | ||
d401dc12 | 1072 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
6b791bcc TM |
1073 | |
1074 | out: | |
1075 | ocfs2_free_path(path); | |
1076 | return status; | |
1077 | } | |
1078 | ||
ccd979bd MF |
1079 | /* |
1080 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
d401dc12 | 1081 | * to start at, if we don't want to start the branch at the root |
ccd979bd MF |
1082 | * structure. |
1083 | * | |
1084 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
1085 | * for the new last extent block. | |
1086 | * | |
1087 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 1088 | * contain a single record with cluster count == 0. |
ccd979bd | 1089 | */ |
d401dc12 | 1090 | static int ocfs2_add_branch(handle_t *handle, |
e7d4cb6b | 1091 | struct ocfs2_extent_tree *et, |
ccd979bd | 1092 | struct buffer_head *eb_bh, |
328d5752 | 1093 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
1094 | struct ocfs2_alloc_context *meta_ac) |
1095 | { | |
1096 | int status, new_blocks, i; | |
1097 | u64 next_blkno, new_last_eb_blk; | |
1098 | struct buffer_head *bh; | |
1099 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
1100 | struct ocfs2_extent_block *eb; |
1101 | struct ocfs2_extent_list *eb_el; | |
1102 | struct ocfs2_extent_list *el; | |
6b791bcc | 1103 | u32 new_cpos, root_end; |
ccd979bd MF |
1104 | |
1105 | mlog_entry_void(); | |
1106 | ||
328d5752 | 1107 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 1108 | |
ccd979bd MF |
1109 | if (eb_bh) { |
1110 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
1111 | el = &eb->h_list; | |
1112 | } else | |
ce1d9ea6 | 1113 | el = et->et_root_el; |
ccd979bd MF |
1114 | |
1115 | /* we never add a branch to a leaf. */ | |
1116 | BUG_ON(!el->l_tree_depth); | |
1117 | ||
1118 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
1119 | ||
6b791bcc TM |
1120 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
1121 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); | |
1122 | root_end = ocfs2_sum_rightmost_rec(et->et_root_el); | |
1123 | ||
1124 | /* | |
1125 | * If there is a gap before the root end and the real end | |
1126 | * of the righmost leaf block, we need to remove the gap | |
1127 | * between new_cpos and root_end first so that the tree | |
1128 | * is consistent after we add a new branch(it will start | |
1129 | * from new_cpos). | |
1130 | */ | |
1131 | if (root_end > new_cpos) { | |
1132 | mlog(0, "adjust the cluster end from %u to %u\n", | |
1133 | root_end, new_cpos); | |
d401dc12 | 1134 | status = ocfs2_adjust_rightmost_branch(handle, et); |
6b791bcc TM |
1135 | if (status) { |
1136 | mlog_errno(status); | |
1137 | goto bail; | |
1138 | } | |
1139 | } | |
1140 | ||
ccd979bd MF |
1141 | /* allocate the number of new eb blocks we need */ |
1142 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
1143 | GFP_KERNEL); | |
1144 | if (!new_eb_bhs) { | |
1145 | status = -ENOMEM; | |
1146 | mlog_errno(status); | |
1147 | goto bail; | |
1148 | } | |
1149 | ||
42a5a7a9 | 1150 | status = ocfs2_create_new_meta_bhs(handle, et, new_blocks, |
ccd979bd MF |
1151 | meta_ac, new_eb_bhs); |
1152 | if (status < 0) { | |
1153 | mlog_errno(status); | |
1154 | goto bail; | |
1155 | } | |
1156 | ||
1157 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be | |
1158 | * linked with the rest of the tree. | |
1159 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
1160 | * | |
1161 | * when we leave the loop, new_last_eb_blk will point to the | |
1162 | * newest leaf, and next_blkno will point to the topmost extent | |
1163 | * block. */ | |
1164 | next_blkno = new_last_eb_blk = 0; | |
1165 | for(i = 0; i < new_blocks; i++) { | |
1166 | bh = new_eb_bhs[i]; | |
1167 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
5e96581a JB |
1168 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1169 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1170 | eb_el = &eb->h_list; |
1171 | ||
d401dc12 | 1172 | status = ocfs2_journal_access_eb(handle, et->et_ci, bh, |
13723d00 | 1173 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1174 | if (status < 0) { |
1175 | mlog_errno(status); | |
1176 | goto bail; | |
1177 | } | |
1178 | ||
1179 | eb->h_next_leaf_blk = 0; | |
1180 | eb_el->l_tree_depth = cpu_to_le16(i); | |
1181 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
1182 | /* |
1183 | * This actually counts as an empty extent as | |
1184 | * c_clusters == 0 | |
1185 | */ | |
1186 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 1187 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
1188 | /* |
1189 | * eb_el isn't always an interior node, but even leaf | |
1190 | * nodes want a zero'd flags and reserved field so | |
1191 | * this gets the whole 32 bits regardless of use. | |
1192 | */ | |
1193 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
1194 | if (!eb_el->l_tree_depth) |
1195 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
1196 | ||
1197 | status = ocfs2_journal_dirty(handle, bh); | |
1198 | if (status < 0) { | |
1199 | mlog_errno(status); | |
1200 | goto bail; | |
1201 | } | |
1202 | ||
1203 | next_blkno = le64_to_cpu(eb->h_blkno); | |
1204 | } | |
1205 | ||
1206 | /* This is a bit hairy. We want to update up to three blocks | |
1207 | * here without leaving any of them in an inconsistent state | |
1208 | * in case of error. We don't have to worry about | |
1209 | * journal_dirty erroring as it won't unless we've aborted the | |
1210 | * handle (in which case we would never be here) so reserving | |
1211 | * the write with journal_access is all we need to do. */ | |
d401dc12 | 1212 | status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh, |
13723d00 | 1213 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1214 | if (status < 0) { |
1215 | mlog_errno(status); | |
1216 | goto bail; | |
1217 | } | |
d9a0a1f8 | 1218 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1219 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1220 | if (status < 0) { |
1221 | mlog_errno(status); | |
1222 | goto bail; | |
1223 | } | |
1224 | if (eb_bh) { | |
d401dc12 | 1225 | status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh, |
13723d00 | 1226 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1227 | if (status < 0) { |
1228 | mlog_errno(status); | |
1229 | goto bail; | |
1230 | } | |
1231 | } | |
1232 | ||
1233 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 1234 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
1235 | i = le16_to_cpu(el->l_next_free_rec); |
1236 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 1237 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 1238 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
1239 | le16_add_cpu(&el->l_next_free_rec, 1); |
1240 | ||
1241 | /* fe needs a new last extent block pointer, as does the | |
1242 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 1243 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 1244 | |
328d5752 | 1245 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
1246 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
1247 | ||
328d5752 | 1248 | status = ocfs2_journal_dirty(handle, *last_eb_bh); |
ccd979bd MF |
1249 | if (status < 0) |
1250 | mlog_errno(status); | |
ce1d9ea6 | 1251 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1252 | if (status < 0) |
1253 | mlog_errno(status); | |
1254 | if (eb_bh) { | |
1255 | status = ocfs2_journal_dirty(handle, eb_bh); | |
1256 | if (status < 0) | |
1257 | mlog_errno(status); | |
1258 | } | |
1259 | ||
328d5752 MF |
1260 | /* |
1261 | * Some callers want to track the rightmost leaf so pass it | |
1262 | * back here. | |
1263 | */ | |
1264 | brelse(*last_eb_bh); | |
1265 | get_bh(new_eb_bhs[0]); | |
1266 | *last_eb_bh = new_eb_bhs[0]; | |
1267 | ||
ccd979bd MF |
1268 | status = 0; |
1269 | bail: | |
1270 | if (new_eb_bhs) { | |
1271 | for (i = 0; i < new_blocks; i++) | |
a81cb88b | 1272 | brelse(new_eb_bhs[i]); |
ccd979bd MF |
1273 | kfree(new_eb_bhs); |
1274 | } | |
1275 | ||
1276 | mlog_exit(status); | |
1277 | return status; | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * adds another level to the allocation tree. | |
1282 | * returns back the new extent block so you can add a branch to it | |
1283 | * after this call. | |
1284 | */ | |
d401dc12 | 1285 | static int ocfs2_shift_tree_depth(handle_t *handle, |
e7d4cb6b | 1286 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1287 | struct ocfs2_alloc_context *meta_ac, |
1288 | struct buffer_head **ret_new_eb_bh) | |
1289 | { | |
1290 | int status, i; | |
dcd0538f | 1291 | u32 new_clusters; |
ccd979bd | 1292 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 1293 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 1294 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
1295 | struct ocfs2_extent_list *eb_el; |
1296 | ||
1297 | mlog_entry_void(); | |
1298 | ||
42a5a7a9 | 1299 | status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac, |
ccd979bd MF |
1300 | &new_eb_bh); |
1301 | if (status < 0) { | |
1302 | mlog_errno(status); | |
1303 | goto bail; | |
1304 | } | |
1305 | ||
1306 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
5e96581a JB |
1307 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1308 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1309 | |
1310 | eb_el = &eb->h_list; | |
ce1d9ea6 | 1311 | root_el = et->et_root_el; |
ccd979bd | 1312 | |
d401dc12 | 1313 | status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh, |
13723d00 | 1314 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1315 | if (status < 0) { |
1316 | mlog_errno(status); | |
1317 | goto bail; | |
1318 | } | |
1319 | ||
e7d4cb6b TM |
1320 | /* copy the root extent list data into the new extent block */ |
1321 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1322 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1323 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1324 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd MF |
1325 | |
1326 | status = ocfs2_journal_dirty(handle, new_eb_bh); | |
1327 | if (status < 0) { | |
1328 | mlog_errno(status); | |
1329 | goto bail; | |
1330 | } | |
1331 | ||
d9a0a1f8 | 1332 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1333 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1334 | if (status < 0) { |
1335 | mlog_errno(status); | |
1336 | goto bail; | |
1337 | } | |
1338 | ||
dcd0538f MF |
1339 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1340 | ||
e7d4cb6b TM |
1341 | /* update root_bh now */ |
1342 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1343 | root_el->l_recs[0].e_cpos = 0; | |
1344 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1345 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1346 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1347 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1348 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1349 | |
1350 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1351 | * becomes the allocated extent block */ | |
e7d4cb6b | 1352 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1353 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1354 | |
ce1d9ea6 | 1355 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1356 | if (status < 0) { |
1357 | mlog_errno(status); | |
1358 | goto bail; | |
1359 | } | |
1360 | ||
1361 | *ret_new_eb_bh = new_eb_bh; | |
1362 | new_eb_bh = NULL; | |
1363 | status = 0; | |
1364 | bail: | |
a81cb88b | 1365 | brelse(new_eb_bh); |
ccd979bd MF |
1366 | |
1367 | mlog_exit(status); | |
1368 | return status; | |
1369 | } | |
1370 | ||
ccd979bd MF |
1371 | /* |
1372 | * Should only be called when there is no space left in any of the | |
1373 | * leaf nodes. What we want to do is find the lowest tree depth | |
1374 | * non-leaf extent block with room for new records. There are three | |
1375 | * valid results of this search: | |
1376 | * | |
1377 | * 1) a lowest extent block is found, then we pass it back in | |
1378 | * *lowest_eb_bh and return '0' | |
1379 | * | |
e7d4cb6b | 1380 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1381 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1382 | * | |
e7d4cb6b | 1383 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1384 | * which case we return > 0 |
1385 | * | |
1386 | * return status < 0 indicates an error. | |
1387 | */ | |
d401dc12 | 1388 | static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et, |
ccd979bd MF |
1389 | struct buffer_head **target_bh) |
1390 | { | |
1391 | int status = 0, i; | |
1392 | u64 blkno; | |
ccd979bd MF |
1393 | struct ocfs2_extent_block *eb; |
1394 | struct ocfs2_extent_list *el; | |
1395 | struct buffer_head *bh = NULL; | |
1396 | struct buffer_head *lowest_bh = NULL; | |
1397 | ||
1398 | mlog_entry_void(); | |
1399 | ||
1400 | *target_bh = NULL; | |
1401 | ||
ce1d9ea6 | 1402 | el = et->et_root_el; |
ccd979bd MF |
1403 | |
1404 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1405 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3d03a305 JB |
1406 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1407 | "Owner %llu has empty " | |
ccd979bd | 1408 | "extent list (next_free_rec == 0)", |
3d03a305 | 1409 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); |
ccd979bd MF |
1410 | status = -EIO; |
1411 | goto bail; | |
1412 | } | |
1413 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1414 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1415 | if (!blkno) { | |
3d03a305 JB |
1416 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1417 | "Owner %llu has extent " | |
ccd979bd MF |
1418 | "list where extent # %d has no physical " |
1419 | "block start", | |
3d03a305 | 1420 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i); |
ccd979bd MF |
1421 | status = -EIO; |
1422 | goto bail; | |
1423 | } | |
1424 | ||
a81cb88b MF |
1425 | brelse(bh); |
1426 | bh = NULL; | |
ccd979bd | 1427 | |
3d03a305 | 1428 | status = ocfs2_read_extent_block(et->et_ci, blkno, &bh); |
ccd979bd MF |
1429 | if (status < 0) { |
1430 | mlog_errno(status); | |
1431 | goto bail; | |
1432 | } | |
dcd0538f MF |
1433 | |
1434 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
dcd0538f MF |
1435 | el = &eb->h_list; |
1436 | ||
1437 | if (le16_to_cpu(el->l_next_free_rec) < | |
1438 | le16_to_cpu(el->l_count)) { | |
a81cb88b | 1439 | brelse(lowest_bh); |
dcd0538f MF |
1440 | lowest_bh = bh; |
1441 | get_bh(lowest_bh); | |
1442 | } | |
1443 | } | |
1444 | ||
1445 | /* If we didn't find one and the fe doesn't have any room, | |
1446 | * then return '1' */ | |
ce1d9ea6 | 1447 | el = et->et_root_el; |
e7d4cb6b | 1448 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) |
dcd0538f MF |
1449 | status = 1; |
1450 | ||
1451 | *target_bh = lowest_bh; | |
1452 | bail: | |
a81cb88b | 1453 | brelse(bh); |
dcd0538f MF |
1454 | |
1455 | mlog_exit(status); | |
1456 | return status; | |
1457 | } | |
1458 | ||
c3afcbb3 MF |
1459 | /* |
1460 | * Grow a b-tree so that it has more records. | |
1461 | * | |
1462 | * We might shift the tree depth in which case existing paths should | |
1463 | * be considered invalid. | |
1464 | * | |
1465 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1466 | * |
1467 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 | 1468 | */ |
d401dc12 JB |
1469 | static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
1470 | int *final_depth, struct buffer_head **last_eb_bh, | |
c3afcbb3 MF |
1471 | struct ocfs2_alloc_context *meta_ac) |
1472 | { | |
1473 | int ret, shift; | |
ce1d9ea6 | 1474 | struct ocfs2_extent_list *el = et->et_root_el; |
e7d4cb6b | 1475 | int depth = le16_to_cpu(el->l_tree_depth); |
c3afcbb3 MF |
1476 | struct buffer_head *bh = NULL; |
1477 | ||
1478 | BUG_ON(meta_ac == NULL); | |
1479 | ||
d401dc12 | 1480 | shift = ocfs2_find_branch_target(et, &bh); |
c3afcbb3 MF |
1481 | if (shift < 0) { |
1482 | ret = shift; | |
1483 | mlog_errno(ret); | |
1484 | goto out; | |
1485 | } | |
1486 | ||
1487 | /* We traveled all the way to the bottom of the allocation tree | |
1488 | * and didn't find room for any more extents - we need to add | |
1489 | * another tree level */ | |
1490 | if (shift) { | |
1491 | BUG_ON(bh); | |
1492 | mlog(0, "need to shift tree depth (current = %d)\n", depth); | |
1493 | ||
1494 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1495 | * the new extent block (so we can pass that to | |
1496 | * ocfs2_add_branch). */ | |
d401dc12 | 1497 | ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh); |
c3afcbb3 MF |
1498 | if (ret < 0) { |
1499 | mlog_errno(ret); | |
1500 | goto out; | |
1501 | } | |
1502 | depth++; | |
328d5752 MF |
1503 | if (depth == 1) { |
1504 | /* | |
1505 | * Special case: we have room now if we shifted from | |
1506 | * tree_depth 0, so no more work needs to be done. | |
1507 | * | |
1508 | * We won't be calling add_branch, so pass | |
1509 | * back *last_eb_bh as the new leaf. At depth | |
1510 | * zero, it should always be null so there's | |
1511 | * no reason to brelse. | |
1512 | */ | |
1513 | BUG_ON(*last_eb_bh); | |
1514 | get_bh(bh); | |
1515 | *last_eb_bh = bh; | |
c3afcbb3 | 1516 | goto out; |
328d5752 | 1517 | } |
c3afcbb3 MF |
1518 | } |
1519 | ||
1520 | /* call ocfs2_add_branch to add the final part of the tree with | |
1521 | * the new data. */ | |
1522 | mlog(0, "add branch. bh = %p\n", bh); | |
d401dc12 | 1523 | ret = ocfs2_add_branch(handle, et, bh, last_eb_bh, |
c3afcbb3 MF |
1524 | meta_ac); |
1525 | if (ret < 0) { | |
1526 | mlog_errno(ret); | |
1527 | goto out; | |
1528 | } | |
1529 | ||
1530 | out: | |
1531 | if (final_depth) | |
1532 | *final_depth = depth; | |
1533 | brelse(bh); | |
1534 | return ret; | |
1535 | } | |
1536 | ||
dcd0538f MF |
1537 | /* |
1538 | * This function will discard the rightmost extent record. | |
1539 | */ | |
1540 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1541 | { | |
1542 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1543 | int count = le16_to_cpu(el->l_count); | |
1544 | unsigned int num_bytes; | |
1545 | ||
1546 | BUG_ON(!next_free); | |
1547 | /* This will cause us to go off the end of our extent list. */ | |
1548 | BUG_ON(next_free >= count); | |
1549 | ||
1550 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1551 | ||
1552 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1553 | } | |
1554 | ||
1555 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1556 | struct ocfs2_extent_rec *insert_rec) | |
1557 | { | |
1558 | int i, insert_index, next_free, has_empty, num_bytes; | |
1559 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1560 | struct ocfs2_extent_rec *rec; | |
1561 | ||
1562 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1563 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1564 | ||
1565 | BUG_ON(!next_free); | |
1566 | ||
1567 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1568 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1569 | |
1570 | /* | |
1571 | * The easiest way to approach this is to just remove the | |
1572 | * empty extent and temporarily decrement next_free. | |
1573 | */ | |
1574 | if (has_empty) { | |
1575 | /* | |
1576 | * If next_free was 1 (only an empty extent), this | |
1577 | * loop won't execute, which is fine. We still want | |
1578 | * the decrement above to happen. | |
1579 | */ | |
1580 | for(i = 0; i < (next_free - 1); i++) | |
1581 | el->l_recs[i] = el->l_recs[i+1]; | |
1582 | ||
1583 | next_free--; | |
1584 | } | |
1585 | ||
1586 | /* | |
1587 | * Figure out what the new record index should be. | |
1588 | */ | |
1589 | for(i = 0; i < next_free; i++) { | |
1590 | rec = &el->l_recs[i]; | |
1591 | ||
1592 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1593 | break; | |
1594 | } | |
1595 | insert_index = i; | |
1596 | ||
1597 | mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n", | |
1598 | insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count)); | |
1599 | ||
1600 | BUG_ON(insert_index < 0); | |
1601 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1602 | BUG_ON(insert_index > next_free); | |
1603 | ||
1604 | /* | |
1605 | * No need to memmove if we're just adding to the tail. | |
1606 | */ | |
1607 | if (insert_index != next_free) { | |
1608 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1609 | ||
1610 | num_bytes = next_free - insert_index; | |
1611 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1612 | memmove(&el->l_recs[insert_index + 1], | |
1613 | &el->l_recs[insert_index], | |
1614 | num_bytes); | |
1615 | } | |
1616 | ||
1617 | /* | |
1618 | * Either we had an empty extent, and need to re-increment or | |
1619 | * there was no empty extent on a non full rightmost leaf node, | |
1620 | * in which case we still need to increment. | |
1621 | */ | |
1622 | next_free++; | |
1623 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1624 | /* | |
1625 | * Make sure none of the math above just messed up our tree. | |
1626 | */ | |
1627 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1628 | ||
1629 | el->l_recs[insert_index] = *insert_rec; | |
1630 | ||
1631 | } | |
1632 | ||
328d5752 MF |
1633 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1634 | { | |
1635 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1636 | ||
1637 | BUG_ON(num_recs == 0); | |
1638 | ||
1639 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1640 | num_recs--; | |
1641 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1642 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1643 | memset(&el->l_recs[num_recs], 0, | |
1644 | sizeof(struct ocfs2_extent_rec)); | |
1645 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1646 | } | |
1647 | } | |
1648 | ||
dcd0538f MF |
1649 | /* |
1650 | * Create an empty extent record . | |
1651 | * | |
1652 | * l_next_free_rec may be updated. | |
1653 | * | |
1654 | * If an empty extent already exists do nothing. | |
1655 | */ | |
1656 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1657 | { | |
1658 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1659 | ||
e48edee2 MF |
1660 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1661 | ||
dcd0538f MF |
1662 | if (next_free == 0) |
1663 | goto set_and_inc; | |
1664 | ||
1665 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1666 | return; | |
1667 | ||
1668 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1669 | "Asked to create an empty extent in a full list:\n" | |
1670 | "count = %u, tree depth = %u", | |
1671 | le16_to_cpu(el->l_count), | |
1672 | le16_to_cpu(el->l_tree_depth)); | |
1673 | ||
1674 | ocfs2_shift_records_right(el); | |
1675 | ||
1676 | set_and_inc: | |
1677 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1678 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1679 | } | |
1680 | ||
1681 | /* | |
1682 | * For a rotation which involves two leaf nodes, the "root node" is | |
1683 | * the lowest level tree node which contains a path to both leafs. This | |
1684 | * resulting set of information can be used to form a complete "subtree" | |
1685 | * | |
1686 | * This function is passed two full paths from the dinode down to a | |
1687 | * pair of adjacent leaves. It's task is to figure out which path | |
1688 | * index contains the subtree root - this can be the root index itself | |
1689 | * in a worst-case rotation. | |
1690 | * | |
1691 | * The array index of the subtree root is passed back. | |
1692 | */ | |
7dc02805 | 1693 | static int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et, |
dcd0538f MF |
1694 | struct ocfs2_path *left, |
1695 | struct ocfs2_path *right) | |
1696 | { | |
1697 | int i = 0; | |
1698 | ||
1699 | /* | |
1700 | * Check that the caller passed in two paths from the same tree. | |
1701 | */ | |
1702 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1703 | ||
1704 | do { | |
1705 | i++; | |
1706 | ||
1707 | /* | |
1708 | * The caller didn't pass two adjacent paths. | |
1709 | */ | |
1710 | mlog_bug_on_msg(i > left->p_tree_depth, | |
7dc02805 | 1711 | "Owner %llu, left depth %u, right depth %u\n" |
dcd0538f | 1712 | "left leaf blk %llu, right leaf blk %llu\n", |
7dc02805 JB |
1713 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
1714 | left->p_tree_depth, right->p_tree_depth, | |
dcd0538f MF |
1715 | (unsigned long long)path_leaf_bh(left)->b_blocknr, |
1716 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1717 | } while (left->p_node[i].bh->b_blocknr == | |
1718 | right->p_node[i].bh->b_blocknr); | |
1719 | ||
1720 | return i - 1; | |
1721 | } | |
1722 | ||
1723 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1724 | ||
1725 | /* | |
1726 | * Traverse a btree path in search of cpos, starting at root_el. | |
1727 | * | |
1728 | * This code can be called with a cpos larger than the tree, in which | |
1729 | * case it will return the rightmost path. | |
1730 | */ | |
facdb77f | 1731 | static int __ocfs2_find_path(struct ocfs2_caching_info *ci, |
dcd0538f MF |
1732 | struct ocfs2_extent_list *root_el, u32 cpos, |
1733 | path_insert_t *func, void *data) | |
1734 | { | |
1735 | int i, ret = 0; | |
1736 | u32 range; | |
1737 | u64 blkno; | |
1738 | struct buffer_head *bh = NULL; | |
1739 | struct ocfs2_extent_block *eb; | |
1740 | struct ocfs2_extent_list *el; | |
1741 | struct ocfs2_extent_rec *rec; | |
dcd0538f MF |
1742 | |
1743 | el = root_el; | |
1744 | while (el->l_tree_depth) { | |
1745 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
facdb77f JB |
1746 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1747 | "Owner %llu has empty extent list at " | |
dcd0538f | 1748 | "depth %u\n", |
facdb77f | 1749 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1750 | le16_to_cpu(el->l_tree_depth)); |
1751 | ret = -EROFS; | |
1752 | goto out; | |
1753 | ||
1754 | } | |
1755 | ||
1756 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1757 | rec = &el->l_recs[i]; | |
1758 | ||
1759 | /* | |
1760 | * In the case that cpos is off the allocation | |
1761 | * tree, this should just wind up returning the | |
1762 | * rightmost record. | |
1763 | */ | |
1764 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1765 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1766 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1767 | break; | |
1768 | } | |
1769 | ||
1770 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1771 | if (blkno == 0) { | |
facdb77f JB |
1772 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1773 | "Owner %llu has bad blkno in extent list " | |
dcd0538f | 1774 | "at depth %u (index %d)\n", |
facdb77f | 1775 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1776 | le16_to_cpu(el->l_tree_depth), i); |
1777 | ret = -EROFS; | |
1778 | goto out; | |
1779 | } | |
1780 | ||
1781 | brelse(bh); | |
1782 | bh = NULL; | |
facdb77f | 1783 | ret = ocfs2_read_extent_block(ci, blkno, &bh); |
dcd0538f MF |
1784 | if (ret) { |
1785 | mlog_errno(ret); | |
1786 | goto out; | |
1787 | } | |
1788 | ||
1789 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1790 | el = &eb->h_list; | |
dcd0538f MF |
1791 | |
1792 | if (le16_to_cpu(el->l_next_free_rec) > | |
1793 | le16_to_cpu(el->l_count)) { | |
facdb77f JB |
1794 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1795 | "Owner %llu has bad count in extent list " | |
dcd0538f | 1796 | "at block %llu (next free=%u, count=%u)\n", |
facdb77f | 1797 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1798 | (unsigned long long)bh->b_blocknr, |
1799 | le16_to_cpu(el->l_next_free_rec), | |
1800 | le16_to_cpu(el->l_count)); | |
1801 | ret = -EROFS; | |
1802 | goto out; | |
1803 | } | |
1804 | ||
1805 | if (func) | |
1806 | func(data, bh); | |
1807 | } | |
1808 | ||
1809 | out: | |
1810 | /* | |
1811 | * Catch any trailing bh that the loop didn't handle. | |
1812 | */ | |
1813 | brelse(bh); | |
1814 | ||
1815 | return ret; | |
1816 | } | |
1817 | ||
1818 | /* | |
1819 | * Given an initialized path (that is, it has a valid root extent | |
1820 | * list), this function will traverse the btree in search of the path | |
1821 | * which would contain cpos. | |
1822 | * | |
1823 | * The path traveled is recorded in the path structure. | |
1824 | * | |
1825 | * Note that this will not do any comparisons on leaf node extent | |
1826 | * records, so it will work fine in the case that we just added a tree | |
1827 | * branch. | |
1828 | */ | |
1829 | struct find_path_data { | |
1830 | int index; | |
1831 | struct ocfs2_path *path; | |
1832 | }; | |
1833 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1834 | { | |
1835 | struct find_path_data *fp = data; | |
1836 | ||
1837 | get_bh(bh); | |
1838 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1839 | fp->index++; | |
1840 | } | |
facdb77f JB |
1841 | static int ocfs2_find_path(struct ocfs2_caching_info *ci, |
1842 | struct ocfs2_path *path, u32 cpos) | |
dcd0538f MF |
1843 | { |
1844 | struct find_path_data data; | |
1845 | ||
1846 | data.index = 1; | |
1847 | data.path = path; | |
facdb77f | 1848 | return __ocfs2_find_path(ci, path_root_el(path), cpos, |
dcd0538f MF |
1849 | find_path_ins, &data); |
1850 | } | |
1851 | ||
1852 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1853 | { | |
1854 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1855 | struct ocfs2_extent_list *el = &eb->h_list; | |
1856 | struct buffer_head **ret = data; | |
1857 | ||
1858 | /* We want to retain only the leaf block. */ | |
1859 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1860 | get_bh(bh); | |
1861 | *ret = bh; | |
1862 | } | |
1863 | } | |
1864 | /* | |
1865 | * Find the leaf block in the tree which would contain cpos. No | |
1866 | * checking of the actual leaf is done. | |
1867 | * | |
1868 | * Some paths want to call this instead of allocating a path structure | |
1869 | * and calling ocfs2_find_path(). | |
1870 | * | |
1871 | * This function doesn't handle non btree extent lists. | |
1872 | */ | |
facdb77f JB |
1873 | int ocfs2_find_leaf(struct ocfs2_caching_info *ci, |
1874 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1875 | struct buffer_head **leaf_bh) | |
dcd0538f MF |
1876 | { |
1877 | int ret; | |
1878 | struct buffer_head *bh = NULL; | |
1879 | ||
facdb77f | 1880 | ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh); |
dcd0538f MF |
1881 | if (ret) { |
1882 | mlog_errno(ret); | |
1883 | goto out; | |
1884 | } | |
1885 | ||
1886 | *leaf_bh = bh; | |
1887 | out: | |
1888 | return ret; | |
1889 | } | |
1890 | ||
1891 | /* | |
1892 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1893 | * | |
1894 | * Basically, we've moved stuff around at the bottom of the tree and | |
1895 | * we need to fix up the extent records above the changes to reflect | |
1896 | * the new changes. | |
1897 | * | |
1898 | * left_rec: the record on the left. | |
1899 | * left_child_el: is the child list pointed to by left_rec | |
1900 | * right_rec: the record to the right of left_rec | |
1901 | * right_child_el: is the child list pointed to by right_rec | |
1902 | * | |
1903 | * By definition, this only works on interior nodes. | |
1904 | */ | |
1905 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1906 | struct ocfs2_extent_list *left_child_el, | |
1907 | struct ocfs2_extent_rec *right_rec, | |
1908 | struct ocfs2_extent_list *right_child_el) | |
1909 | { | |
1910 | u32 left_clusters, right_end; | |
1911 | ||
1912 | /* | |
1913 | * Interior nodes never have holes. Their cpos is the cpos of | |
1914 | * the leftmost record in their child list. Their cluster | |
1915 | * count covers the full theoretical range of their child list | |
1916 | * - the range between their cpos and the cpos of the record | |
1917 | * immediately to their right. | |
1918 | */ | |
1919 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
82e12644 TM |
1920 | if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) { |
1921 | BUG_ON(right_child_el->l_tree_depth); | |
328d5752 MF |
1922 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); |
1923 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1924 | } | |
dcd0538f | 1925 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 1926 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
1927 | |
1928 | /* | |
1929 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 1930 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
1931 | * updating e_cpos to keep the same highest cluster count. |
1932 | */ | |
1933 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1934 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
1935 | |
1936 | right_rec->e_cpos = left_rec->e_cpos; | |
1937 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
1938 | ||
1939 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1940 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
1941 | } |
1942 | ||
1943 | /* | |
1944 | * Adjust the adjacent root node records involved in a | |
1945 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
1946 | * find it's index in the root list. | |
1947 | */ | |
1948 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
1949 | struct ocfs2_extent_list *left_el, | |
1950 | struct ocfs2_extent_list *right_el, | |
1951 | u64 left_el_blkno) | |
1952 | { | |
1953 | int i; | |
1954 | ||
1955 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
1956 | le16_to_cpu(left_el->l_tree_depth)); | |
1957 | ||
1958 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
1959 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
1960 | break; | |
1961 | } | |
1962 | ||
1963 | /* | |
1964 | * The path walking code should have never returned a root and | |
1965 | * two paths which are not adjacent. | |
1966 | */ | |
1967 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
1968 | ||
1969 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
1970 | &root_el->l_recs[i + 1], right_el); | |
1971 | } | |
1972 | ||
1973 | /* | |
1974 | * We've changed a leaf block (in right_path) and need to reflect that | |
1975 | * change back up the subtree. | |
1976 | * | |
1977 | * This happens in multiple places: | |
1978 | * - When we've moved an extent record from the left path leaf to the right | |
1979 | * path leaf to make room for an empty extent in the left path leaf. | |
1980 | * - When our insert into the right path leaf is at the leftmost edge | |
1981 | * and requires an update of the path immediately to it's left. This | |
1982 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
1983 | * - When we've adjusted the last extent record in the left path leaf and the |
1984 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f | 1985 | */ |
4619c73e | 1986 | static void ocfs2_complete_edge_insert(handle_t *handle, |
dcd0538f MF |
1987 | struct ocfs2_path *left_path, |
1988 | struct ocfs2_path *right_path, | |
1989 | int subtree_index) | |
1990 | { | |
1991 | int ret, i, idx; | |
1992 | struct ocfs2_extent_list *el, *left_el, *right_el; | |
1993 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
1994 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
1995 | ||
1996 | /* | |
1997 | * Update the counts and position values within all the | |
1998 | * interior nodes to reflect the leaf rotation we just did. | |
1999 | * | |
2000 | * The root node is handled below the loop. | |
2001 | * | |
2002 | * We begin the loop with right_el and left_el pointing to the | |
2003 | * leaf lists and work our way up. | |
2004 | * | |
2005 | * NOTE: within this loop, left_el and right_el always refer | |
2006 | * to the *child* lists. | |
2007 | */ | |
2008 | left_el = path_leaf_el(left_path); | |
2009 | right_el = path_leaf_el(right_path); | |
2010 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
2011 | mlog(0, "Adjust records at index %u\n", i); | |
2012 | ||
2013 | /* | |
2014 | * One nice property of knowing that all of these | |
2015 | * nodes are below the root is that we only deal with | |
2016 | * the leftmost right node record and the rightmost | |
2017 | * left node record. | |
2018 | */ | |
2019 | el = left_path->p_node[i].el; | |
2020 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2021 | left_rec = &el->l_recs[idx]; | |
2022 | ||
2023 | el = right_path->p_node[i].el; | |
2024 | right_rec = &el->l_recs[0]; | |
2025 | ||
2026 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
2027 | right_el); | |
2028 | ||
2029 | ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh); | |
2030 | if (ret) | |
2031 | mlog_errno(ret); | |
2032 | ||
2033 | ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
2034 | if (ret) | |
2035 | mlog_errno(ret); | |
2036 | ||
2037 | /* | |
2038 | * Setup our list pointers now so that the current | |
2039 | * parents become children in the next iteration. | |
2040 | */ | |
2041 | left_el = left_path->p_node[i].el; | |
2042 | right_el = right_path->p_node[i].el; | |
2043 | } | |
2044 | ||
2045 | /* | |
2046 | * At the root node, adjust the two adjacent records which | |
2047 | * begin our path to the leaves. | |
2048 | */ | |
2049 | ||
2050 | el = left_path->p_node[subtree_index].el; | |
2051 | left_el = left_path->p_node[subtree_index + 1].el; | |
2052 | right_el = right_path->p_node[subtree_index + 1].el; | |
2053 | ||
2054 | ocfs2_adjust_root_records(el, left_el, right_el, | |
2055 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
2056 | ||
2057 | root_bh = left_path->p_node[subtree_index].bh; | |
2058 | ||
2059 | ret = ocfs2_journal_dirty(handle, root_bh); | |
2060 | if (ret) | |
2061 | mlog_errno(ret); | |
2062 | } | |
2063 | ||
5c601aba JB |
2064 | static int ocfs2_rotate_subtree_right(handle_t *handle, |
2065 | struct ocfs2_extent_tree *et, | |
dcd0538f MF |
2066 | struct ocfs2_path *left_path, |
2067 | struct ocfs2_path *right_path, | |
2068 | int subtree_index) | |
2069 | { | |
2070 | int ret, i; | |
2071 | struct buffer_head *right_leaf_bh; | |
2072 | struct buffer_head *left_leaf_bh = NULL; | |
2073 | struct buffer_head *root_bh; | |
2074 | struct ocfs2_extent_list *right_el, *left_el; | |
2075 | struct ocfs2_extent_rec move_rec; | |
2076 | ||
2077 | left_leaf_bh = path_leaf_bh(left_path); | |
2078 | left_el = path_leaf_el(left_path); | |
2079 | ||
2080 | if (left_el->l_next_free_rec != left_el->l_count) { | |
5c601aba | 2081 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
dcd0538f MF |
2082 | "Inode %llu has non-full interior leaf node %llu" |
2083 | "(next free = %u)", | |
5c601aba | 2084 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2085 | (unsigned long long)left_leaf_bh->b_blocknr, |
2086 | le16_to_cpu(left_el->l_next_free_rec)); | |
2087 | return -EROFS; | |
2088 | } | |
2089 | ||
2090 | /* | |
2091 | * This extent block may already have an empty record, so we | |
2092 | * return early if so. | |
2093 | */ | |
2094 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
2095 | return 0; | |
2096 | ||
2097 | root_bh = left_path->p_node[subtree_index].bh; | |
2098 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2099 | ||
5c601aba | 2100 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2101 | subtree_index); |
dcd0538f MF |
2102 | if (ret) { |
2103 | mlog_errno(ret); | |
2104 | goto out; | |
2105 | } | |
2106 | ||
2107 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
5c601aba | 2108 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2109 | right_path, i); |
dcd0538f MF |
2110 | if (ret) { |
2111 | mlog_errno(ret); | |
2112 | goto out; | |
2113 | } | |
2114 | ||
5c601aba | 2115 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2116 | left_path, i); |
dcd0538f MF |
2117 | if (ret) { |
2118 | mlog_errno(ret); | |
2119 | goto out; | |
2120 | } | |
2121 | } | |
2122 | ||
2123 | right_leaf_bh = path_leaf_bh(right_path); | |
2124 | right_el = path_leaf_el(right_path); | |
2125 | ||
2126 | /* This is a code error, not a disk corruption. */ | |
2127 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
2128 | "because rightmost leaf block %llu is empty\n", | |
5c601aba | 2129 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2130 | (unsigned long long)right_leaf_bh->b_blocknr); |
2131 | ||
2132 | ocfs2_create_empty_extent(right_el); | |
2133 | ||
2134 | ret = ocfs2_journal_dirty(handle, right_leaf_bh); | |
2135 | if (ret) { | |
2136 | mlog_errno(ret); | |
2137 | goto out; | |
2138 | } | |
2139 | ||
2140 | /* Do the copy now. */ | |
2141 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2142 | move_rec = left_el->l_recs[i]; | |
2143 | right_el->l_recs[0] = move_rec; | |
2144 | ||
2145 | /* | |
2146 | * Clear out the record we just copied and shift everything | |
2147 | * over, leaving an empty extent in the left leaf. | |
2148 | * | |
2149 | * We temporarily subtract from next_free_rec so that the | |
2150 | * shift will lose the tail record (which is now defunct). | |
2151 | */ | |
2152 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
2153 | ocfs2_shift_records_right(left_el); | |
2154 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2155 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
2156 | ||
2157 | ret = ocfs2_journal_dirty(handle, left_leaf_bh); | |
2158 | if (ret) { | |
2159 | mlog_errno(ret); | |
2160 | goto out; | |
2161 | } | |
2162 | ||
4619c73e JB |
2163 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
2164 | subtree_index); | |
dcd0538f MF |
2165 | |
2166 | out: | |
2167 | return ret; | |
2168 | } | |
2169 | ||
2170 | /* | |
2171 | * Given a full path, determine what cpos value would return us a path | |
2172 | * containing the leaf immediately to the left of the current one. | |
2173 | * | |
2174 | * Will return zero if the path passed in is already the leftmost path. | |
2175 | */ | |
2176 | static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, | |
2177 | struct ocfs2_path *path, u32 *cpos) | |
2178 | { | |
2179 | int i, j, ret = 0; | |
2180 | u64 blkno; | |
2181 | struct ocfs2_extent_list *el; | |
2182 | ||
e48edee2 MF |
2183 | BUG_ON(path->p_tree_depth == 0); |
2184 | ||
dcd0538f MF |
2185 | *cpos = 0; |
2186 | ||
2187 | blkno = path_leaf_bh(path)->b_blocknr; | |
2188 | ||
2189 | /* Start at the tree node just above the leaf and work our way up. */ | |
2190 | i = path->p_tree_depth - 1; | |
2191 | while (i >= 0) { | |
2192 | el = path->p_node[i].el; | |
2193 | ||
2194 | /* | |
2195 | * Find the extent record just before the one in our | |
2196 | * path. | |
2197 | */ | |
2198 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2199 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2200 | if (j == 0) { | |
2201 | if (i == 0) { | |
2202 | /* | |
2203 | * We've determined that the | |
2204 | * path specified is already | |
2205 | * the leftmost one - return a | |
2206 | * cpos of zero. | |
2207 | */ | |
2208 | goto out; | |
2209 | } | |
2210 | /* | |
2211 | * The leftmost record points to our | |
2212 | * leaf - we need to travel up the | |
2213 | * tree one level. | |
2214 | */ | |
2215 | goto next_node; | |
2216 | } | |
2217 | ||
2218 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
2219 | *cpos = *cpos + ocfs2_rec_clusters(el, |
2220 | &el->l_recs[j - 1]); | |
2221 | *cpos = *cpos - 1; | |
dcd0538f MF |
2222 | goto out; |
2223 | } | |
2224 | } | |
2225 | ||
2226 | /* | |
2227 | * If we got here, we never found a valid node where | |
2228 | * the tree indicated one should be. | |
2229 | */ | |
2230 | ocfs2_error(sb, | |
2231 | "Invalid extent tree at extent block %llu\n", | |
2232 | (unsigned long long)blkno); | |
2233 | ret = -EROFS; | |
2234 | goto out; | |
2235 | ||
2236 | next_node: | |
2237 | blkno = path->p_node[i].bh->b_blocknr; | |
2238 | i--; | |
2239 | } | |
2240 | ||
2241 | out: | |
2242 | return ret; | |
2243 | } | |
2244 | ||
328d5752 MF |
2245 | /* |
2246 | * Extend the transaction by enough credits to complete the rotation, | |
2247 | * and still leave at least the original number of credits allocated | |
2248 | * to this transaction. | |
2249 | */ | |
dcd0538f | 2250 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 2251 | int op_credits, |
dcd0538f MF |
2252 | struct ocfs2_path *path) |
2253 | { | |
328d5752 | 2254 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f MF |
2255 | |
2256 | if (handle->h_buffer_credits < credits) | |
2257 | return ocfs2_extend_trans(handle, credits); | |
2258 | ||
2259 | return 0; | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * Trap the case where we're inserting into the theoretical range past | |
2264 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
2265 | * whose cpos is less than ours into the right leaf. | |
2266 | * | |
2267 | * It's only necessary to look at the rightmost record of the left | |
2268 | * leaf because the logic that calls us should ensure that the | |
2269 | * theoretical ranges in the path components above the leaves are | |
2270 | * correct. | |
2271 | */ | |
2272 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
2273 | u32 insert_cpos) | |
2274 | { | |
2275 | struct ocfs2_extent_list *left_el; | |
2276 | struct ocfs2_extent_rec *rec; | |
2277 | int next_free; | |
2278 | ||
2279 | left_el = path_leaf_el(left_path); | |
2280 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
2281 | rec = &left_el->l_recs[next_free - 1]; | |
2282 | ||
2283 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
2284 | return 1; | |
2285 | return 0; | |
2286 | } | |
2287 | ||
328d5752 MF |
2288 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
2289 | { | |
2290 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
2291 | unsigned int range; | |
2292 | struct ocfs2_extent_rec *rec; | |
2293 | ||
2294 | if (next_free == 0) | |
2295 | return 0; | |
2296 | ||
2297 | rec = &el->l_recs[0]; | |
2298 | if (ocfs2_is_empty_extent(rec)) { | |
2299 | /* Empty list. */ | |
2300 | if (next_free == 1) | |
2301 | return 0; | |
2302 | rec = &el->l_recs[1]; | |
2303 | } | |
2304 | ||
2305 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2306 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2307 | return 1; | |
2308 | return 0; | |
2309 | } | |
2310 | ||
dcd0538f MF |
2311 | /* |
2312 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2313 | * | |
2314 | * The path to the rightmost leaf should be passed in. | |
2315 | * | |
2316 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2317 | * | |
2318 | * Upon succesful return from this function: | |
2319 | * | |
2320 | * - The 'right_path' array will contain a path to the leaf block | |
2321 | * whose range contains e_cpos. | |
2322 | * - That leaf block will have a single empty extent in list index 0. | |
2323 | * - In the case that the rotation requires a post-insert update, | |
2324 | * *ret_left_path will contain a valid path which can be passed to | |
2325 | * ocfs2_insert_path(). | |
2326 | */ | |
1bbf0b8d | 2327 | static int ocfs2_rotate_tree_right(handle_t *handle, |
5c601aba | 2328 | struct ocfs2_extent_tree *et, |
328d5752 | 2329 | enum ocfs2_split_type split, |
dcd0538f MF |
2330 | u32 insert_cpos, |
2331 | struct ocfs2_path *right_path, | |
2332 | struct ocfs2_path **ret_left_path) | |
2333 | { | |
328d5752 | 2334 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2335 | u32 cpos; |
2336 | struct ocfs2_path *left_path = NULL; | |
5c601aba | 2337 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
dcd0538f MF |
2338 | |
2339 | *ret_left_path = NULL; | |
2340 | ||
ffdd7a54 | 2341 | left_path = ocfs2_new_path_from_path(right_path); |
dcd0538f MF |
2342 | if (!left_path) { |
2343 | ret = -ENOMEM; | |
2344 | mlog_errno(ret); | |
2345 | goto out; | |
2346 | } | |
2347 | ||
5c601aba | 2348 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2349 | if (ret) { |
2350 | mlog_errno(ret); | |
2351 | goto out; | |
2352 | } | |
2353 | ||
2354 | mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos); | |
2355 | ||
2356 | /* | |
2357 | * What we want to do here is: | |
2358 | * | |
2359 | * 1) Start with the rightmost path. | |
2360 | * | |
2361 | * 2) Determine a path to the leaf block directly to the left | |
2362 | * of that leaf. | |
2363 | * | |
2364 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2365 | * which contains a path to both leaves. | |
2366 | * | |
2367 | * 4) Rotate the subtree. | |
2368 | * | |
2369 | * 5) Find the next subtree by considering the left path to be | |
2370 | * the new right path. | |
2371 | * | |
2372 | * The check at the top of this while loop also accepts | |
2373 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2374 | * value to get us the left path - insert_cpos might very well | |
2375 | * be filling that hole. | |
2376 | * | |
2377 | * Stop at a cpos of '0' because we either started at the | |
2378 | * leftmost branch (i.e., a tree with one branch and a | |
2379 | * rotation inside of it), or we've gone as far as we can in | |
2380 | * rotating subtrees. | |
2381 | */ | |
2382 | while (cpos && insert_cpos <= cpos) { | |
2383 | mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n", | |
2384 | insert_cpos, cpos); | |
2385 | ||
5c601aba | 2386 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
dcd0538f MF |
2387 | if (ret) { |
2388 | mlog_errno(ret); | |
2389 | goto out; | |
2390 | } | |
2391 | ||
2392 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2393 | path_leaf_bh(right_path), | |
5c601aba | 2394 | "Owner %llu: error during insert of %u " |
dcd0538f MF |
2395 | "(left path cpos %u) results in two identical " |
2396 | "paths ending at %llu\n", | |
5c601aba JB |
2397 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
2398 | insert_cpos, cpos, | |
dcd0538f MF |
2399 | (unsigned long long) |
2400 | path_leaf_bh(left_path)->b_blocknr); | |
2401 | ||
328d5752 MF |
2402 | if (split == SPLIT_NONE && |
2403 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2404 | insert_cpos)) { |
dcd0538f MF |
2405 | |
2406 | /* | |
2407 | * We've rotated the tree as much as we | |
2408 | * should. The rest is up to | |
2409 | * ocfs2_insert_path() to complete, after the | |
2410 | * record insertion. We indicate this | |
2411 | * situation by returning the left path. | |
2412 | * | |
2413 | * The reason we don't adjust the records here | |
2414 | * before the record insert is that an error | |
2415 | * later might break the rule where a parent | |
2416 | * record e_cpos will reflect the actual | |
2417 | * e_cpos of the 1st nonempty record of the | |
2418 | * child list. | |
2419 | */ | |
2420 | *ret_left_path = left_path; | |
2421 | goto out_ret_path; | |
2422 | } | |
2423 | ||
7dc02805 | 2424 | start = ocfs2_find_subtree_root(et, left_path, right_path); |
dcd0538f MF |
2425 | |
2426 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2427 | start, | |
2428 | (unsigned long long) right_path->p_node[start].bh->b_blocknr, | |
2429 | right_path->p_tree_depth); | |
2430 | ||
2431 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2432 | orig_credits, right_path); |
dcd0538f MF |
2433 | if (ret) { |
2434 | mlog_errno(ret); | |
2435 | goto out; | |
2436 | } | |
2437 | ||
5c601aba | 2438 | ret = ocfs2_rotate_subtree_right(handle, et, left_path, |
dcd0538f MF |
2439 | right_path, start); |
2440 | if (ret) { | |
2441 | mlog_errno(ret); | |
2442 | goto out; | |
2443 | } | |
2444 | ||
328d5752 MF |
2445 | if (split != SPLIT_NONE && |
2446 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2447 | insert_cpos)) { | |
2448 | /* | |
2449 | * A rotate moves the rightmost left leaf | |
2450 | * record over to the leftmost right leaf | |
2451 | * slot. If we're doing an extent split | |
2452 | * instead of a real insert, then we have to | |
2453 | * check that the extent to be split wasn't | |
2454 | * just moved over. If it was, then we can | |
2455 | * exit here, passing left_path back - | |
2456 | * ocfs2_split_extent() is smart enough to | |
2457 | * search both leaves. | |
2458 | */ | |
2459 | *ret_left_path = left_path; | |
2460 | goto out_ret_path; | |
2461 | } | |
2462 | ||
dcd0538f MF |
2463 | /* |
2464 | * There is no need to re-read the next right path | |
2465 | * as we know that it'll be our current left | |
2466 | * path. Optimize by copying values instead. | |
2467 | */ | |
2468 | ocfs2_mv_path(right_path, left_path); | |
2469 | ||
5c601aba | 2470 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2471 | if (ret) { |
2472 | mlog_errno(ret); | |
2473 | goto out; | |
2474 | } | |
2475 | } | |
2476 | ||
2477 | out: | |
2478 | ocfs2_free_path(left_path); | |
2479 | ||
2480 | out_ret_path: | |
2481 | return ret; | |
2482 | } | |
2483 | ||
09106bae JB |
2484 | static int ocfs2_update_edge_lengths(handle_t *handle, |
2485 | struct ocfs2_extent_tree *et, | |
3c5e1068 | 2486 | int subtree_index, struct ocfs2_path *path) |
dcd0538f | 2487 | { |
3c5e1068 | 2488 | int i, idx, ret; |
dcd0538f | 2489 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2490 | struct ocfs2_extent_list *el; |
2491 | struct ocfs2_extent_block *eb; | |
2492 | u32 range; | |
dcd0538f | 2493 | |
3c5e1068 TM |
2494 | /* |
2495 | * In normal tree rotation process, we will never touch the | |
2496 | * tree branch above subtree_index and ocfs2_extend_rotate_transaction | |
2497 | * doesn't reserve the credits for them either. | |
2498 | * | |
2499 | * But we do have a special case here which will update the rightmost | |
2500 | * records for all the bh in the path. | |
2501 | * So we have to allocate extra credits and access them. | |
2502 | */ | |
2503 | ret = ocfs2_extend_trans(handle, | |
2504 | handle->h_buffer_credits + subtree_index); | |
2505 | if (ret) { | |
2506 | mlog_errno(ret); | |
2507 | goto out; | |
2508 | } | |
2509 | ||
09106bae | 2510 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
3c5e1068 TM |
2511 | if (ret) { |
2512 | mlog_errno(ret); | |
2513 | goto out; | |
2514 | } | |
2515 | ||
328d5752 MF |
2516 | /* Path should always be rightmost. */ |
2517 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2518 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2519 | |
328d5752 MF |
2520 | el = &eb->h_list; |
2521 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2522 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2523 | rec = &el->l_recs[idx]; | |
2524 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2525 | |
328d5752 MF |
2526 | for (i = 0; i < path->p_tree_depth; i++) { |
2527 | el = path->p_node[i].el; | |
2528 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2529 | rec = &el->l_recs[idx]; | |
dcd0538f | 2530 | |
328d5752 MF |
2531 | rec->e_int_clusters = cpu_to_le32(range); |
2532 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2533 | |
328d5752 | 2534 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2535 | } |
3c5e1068 TM |
2536 | out: |
2537 | return ret; | |
dcd0538f MF |
2538 | } |
2539 | ||
6641b0ce JB |
2540 | static void ocfs2_unlink_path(handle_t *handle, |
2541 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2542 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2543 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2544 | { |
328d5752 MF |
2545 | int ret, i; |
2546 | struct ocfs2_extent_block *eb; | |
2547 | struct ocfs2_extent_list *el; | |
2548 | struct buffer_head *bh; | |
2549 | ||
2550 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2551 | bh = path->p_node[i].bh; | |
2552 | ||
2553 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2554 | /* | |
2555 | * Not all nodes might have had their final count | |
2556 | * decremented by the caller - handle this here. | |
2557 | */ | |
2558 | el = &eb->h_list; | |
2559 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2560 | mlog(ML_ERROR, | |
2561 | "Inode %llu, attempted to remove extent block " | |
2562 | "%llu with %u records\n", | |
6641b0ce | 2563 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
2564 | (unsigned long long)le64_to_cpu(eb->h_blkno), |
2565 | le16_to_cpu(el->l_next_free_rec)); | |
2566 | ||
2567 | ocfs2_journal_dirty(handle, bh); | |
6641b0ce | 2568 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 MF |
2569 | continue; |
2570 | } | |
2571 | ||
2572 | el->l_next_free_rec = 0; | |
2573 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2574 | ||
2575 | ocfs2_journal_dirty(handle, bh); | |
2576 | ||
2577 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2578 | if (ret) | |
2579 | mlog_errno(ret); | |
2580 | ||
6641b0ce | 2581 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 | 2582 | } |
dcd0538f MF |
2583 | } |
2584 | ||
6641b0ce JB |
2585 | static void ocfs2_unlink_subtree(handle_t *handle, |
2586 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2587 | struct ocfs2_path *left_path, |
2588 | struct ocfs2_path *right_path, | |
2589 | int subtree_index, | |
2590 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2591 | { |
328d5752 MF |
2592 | int i; |
2593 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2594 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2595 | struct ocfs2_extent_list *el; |
328d5752 | 2596 | struct ocfs2_extent_block *eb; |
dcd0538f | 2597 | |
328d5752 | 2598 | el = path_leaf_el(left_path); |
dcd0538f | 2599 | |
328d5752 | 2600 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2601 | |
328d5752 MF |
2602 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2603 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2604 | break; | |
dcd0538f | 2605 | |
328d5752 | 2606 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2607 | |
328d5752 MF |
2608 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2609 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2610 | ||
2611 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2612 | eb->h_next_leaf_blk = 0; | |
2613 | ||
2614 | ocfs2_journal_dirty(handle, root_bh); | |
2615 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2616 | ||
6641b0ce | 2617 | ocfs2_unlink_path(handle, et, dealloc, right_path, |
328d5752 MF |
2618 | subtree_index + 1); |
2619 | } | |
2620 | ||
1e2dd63f JB |
2621 | static int ocfs2_rotate_subtree_left(handle_t *handle, |
2622 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2623 | struct ocfs2_path *left_path, |
2624 | struct ocfs2_path *right_path, | |
2625 | int subtree_index, | |
2626 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
1e2dd63f | 2627 | int *deleted) |
328d5752 MF |
2628 | { |
2629 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2630 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2631 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2632 | struct ocfs2_extent_block *eb; | |
2633 | ||
2634 | *deleted = 0; | |
2635 | ||
2636 | right_leaf_el = path_leaf_el(right_path); | |
2637 | left_leaf_el = path_leaf_el(left_path); | |
2638 | root_bh = left_path->p_node[subtree_index].bh; | |
2639 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2640 | ||
2641 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2642 | return 0; | |
dcd0538f | 2643 | |
328d5752 MF |
2644 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2645 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2646 | /* |
328d5752 MF |
2647 | * It's legal for us to proceed if the right leaf is |
2648 | * the rightmost one and it has an empty extent. There | |
2649 | * are two cases to handle - whether the leaf will be | |
2650 | * empty after removal or not. If the leaf isn't empty | |
2651 | * then just remove the empty extent up front. The | |
2652 | * next block will handle empty leaves by flagging | |
2653 | * them for unlink. | |
2654 | * | |
2655 | * Non rightmost leaves will throw -EAGAIN and the | |
2656 | * caller can manually move the subtree and retry. | |
dcd0538f | 2657 | */ |
dcd0538f | 2658 | |
328d5752 MF |
2659 | if (eb->h_next_leaf_blk != 0ULL) |
2660 | return -EAGAIN; | |
2661 | ||
2662 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
1e2dd63f | 2663 | ret = ocfs2_journal_access_eb(handle, et->et_ci, |
13723d00 JB |
2664 | path_leaf_bh(right_path), |
2665 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2666 | if (ret) { |
2667 | mlog_errno(ret); | |
2668 | goto out; | |
2669 | } | |
2670 | ||
328d5752 MF |
2671 | ocfs2_remove_empty_extent(right_leaf_el); |
2672 | } else | |
2673 | right_has_empty = 1; | |
dcd0538f MF |
2674 | } |
2675 | ||
328d5752 MF |
2676 | if (eb->h_next_leaf_blk == 0ULL && |
2677 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2678 | /* | |
2679 | * We have to update i_last_eb_blk during the meta | |
2680 | * data delete. | |
2681 | */ | |
d9a0a1f8 | 2682 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 2683 | OCFS2_JOURNAL_ACCESS_WRITE); |
328d5752 MF |
2684 | if (ret) { |
2685 | mlog_errno(ret); | |
2686 | goto out; | |
2687 | } | |
2688 | ||
2689 | del_right_subtree = 1; | |
2690 | } | |
2691 | ||
2692 | /* | |
2693 | * Getting here with an empty extent in the right path implies | |
2694 | * that it's the rightmost path and will be deleted. | |
2695 | */ | |
2696 | BUG_ON(right_has_empty && !del_right_subtree); | |
2697 | ||
1e2dd63f | 2698 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2699 | subtree_index); |
328d5752 MF |
2700 | if (ret) { |
2701 | mlog_errno(ret); | |
2702 | goto out; | |
2703 | } | |
2704 | ||
2705 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1e2dd63f | 2706 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2707 | right_path, i); |
328d5752 MF |
2708 | if (ret) { |
2709 | mlog_errno(ret); | |
2710 | goto out; | |
2711 | } | |
2712 | ||
1e2dd63f | 2713 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2714 | left_path, i); |
328d5752 MF |
2715 | if (ret) { |
2716 | mlog_errno(ret); | |
2717 | goto out; | |
2718 | } | |
2719 | } | |
2720 | ||
2721 | if (!right_has_empty) { | |
2722 | /* | |
2723 | * Only do this if we're moving a real | |
2724 | * record. Otherwise, the action is delayed until | |
2725 | * after removal of the right path in which case we | |
2726 | * can do a simple shift to remove the empty extent. | |
2727 | */ | |
2728 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2729 | memset(&right_leaf_el->l_recs[0], 0, | |
2730 | sizeof(struct ocfs2_extent_rec)); | |
2731 | } | |
2732 | if (eb->h_next_leaf_blk == 0ULL) { | |
2733 | /* | |
2734 | * Move recs over to get rid of empty extent, decrease | |
2735 | * next_free. This is allowed to remove the last | |
2736 | * extent in our leaf (setting l_next_free_rec to | |
2737 | * zero) - the delete code below won't care. | |
2738 | */ | |
2739 | ocfs2_remove_empty_extent(right_leaf_el); | |
2740 | } | |
2741 | ||
2742 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2743 | if (ret) | |
2744 | mlog_errno(ret); | |
2745 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
2746 | if (ret) | |
2747 | mlog_errno(ret); | |
2748 | ||
2749 | if (del_right_subtree) { | |
6641b0ce | 2750 | ocfs2_unlink_subtree(handle, et, left_path, right_path, |
328d5752 | 2751 | subtree_index, dealloc); |
09106bae | 2752 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
2753 | left_path); |
2754 | if (ret) { | |
2755 | mlog_errno(ret); | |
2756 | goto out; | |
2757 | } | |
328d5752 MF |
2758 | |
2759 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2760 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2761 | |
2762 | /* | |
2763 | * Removal of the extent in the left leaf was skipped | |
2764 | * above so we could delete the right path | |
2765 | * 1st. | |
2766 | */ | |
2767 | if (right_has_empty) | |
2768 | ocfs2_remove_empty_extent(left_leaf_el); | |
2769 | ||
e7d4cb6b | 2770 | ret = ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2771 | if (ret) |
2772 | mlog_errno(ret); | |
2773 | ||
2774 | *deleted = 1; | |
2775 | } else | |
4619c73e | 2776 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
328d5752 MF |
2777 | subtree_index); |
2778 | ||
2779 | out: | |
2780 | return ret; | |
2781 | } | |
2782 | ||
2783 | /* | |
2784 | * Given a full path, determine what cpos value would return us a path | |
2785 | * containing the leaf immediately to the right of the current one. | |
2786 | * | |
2787 | * Will return zero if the path passed in is already the rightmost path. | |
2788 | * | |
2789 | * This looks similar, but is subtly different to | |
2790 | * ocfs2_find_cpos_for_left_leaf(). | |
2791 | */ | |
2792 | static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, | |
2793 | struct ocfs2_path *path, u32 *cpos) | |
2794 | { | |
2795 | int i, j, ret = 0; | |
2796 | u64 blkno; | |
2797 | struct ocfs2_extent_list *el; | |
2798 | ||
2799 | *cpos = 0; | |
2800 | ||
2801 | if (path->p_tree_depth == 0) | |
2802 | return 0; | |
2803 | ||
2804 | blkno = path_leaf_bh(path)->b_blocknr; | |
2805 | ||
2806 | /* Start at the tree node just above the leaf and work our way up. */ | |
2807 | i = path->p_tree_depth - 1; | |
2808 | while (i >= 0) { | |
2809 | int next_free; | |
2810 | ||
2811 | el = path->p_node[i].el; | |
2812 | ||
2813 | /* | |
2814 | * Find the extent record just after the one in our | |
2815 | * path. | |
2816 | */ | |
2817 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2818 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2819 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2820 | if (j == (next_free - 1)) { | |
2821 | if (i == 0) { | |
2822 | /* | |
2823 | * We've determined that the | |
2824 | * path specified is already | |
2825 | * the rightmost one - return a | |
2826 | * cpos of zero. | |
2827 | */ | |
2828 | goto out; | |
2829 | } | |
2830 | /* | |
2831 | * The rightmost record points to our | |
2832 | * leaf - we need to travel up the | |
2833 | * tree one level. | |
2834 | */ | |
2835 | goto next_node; | |
2836 | } | |
2837 | ||
2838 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2839 | goto out; | |
2840 | } | |
2841 | } | |
2842 | ||
2843 | /* | |
2844 | * If we got here, we never found a valid node where | |
2845 | * the tree indicated one should be. | |
2846 | */ | |
2847 | ocfs2_error(sb, | |
2848 | "Invalid extent tree at extent block %llu\n", | |
2849 | (unsigned long long)blkno); | |
2850 | ret = -EROFS; | |
2851 | goto out; | |
2852 | ||
2853 | next_node: | |
2854 | blkno = path->p_node[i].bh->b_blocknr; | |
2855 | i--; | |
2856 | } | |
2857 | ||
2858 | out: | |
2859 | return ret; | |
2860 | } | |
2861 | ||
70f18c08 JB |
2862 | static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle, |
2863 | struct ocfs2_extent_tree *et, | |
13723d00 | 2864 | struct ocfs2_path *path) |
328d5752 MF |
2865 | { |
2866 | int ret; | |
13723d00 JB |
2867 | struct buffer_head *bh = path_leaf_bh(path); |
2868 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
328d5752 MF |
2869 | |
2870 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2871 | return 0; | |
2872 | ||
70f18c08 | 2873 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
13723d00 | 2874 | path_num_items(path) - 1); |
328d5752 MF |
2875 | if (ret) { |
2876 | mlog_errno(ret); | |
2877 | goto out; | |
2878 | } | |
2879 | ||
2880 | ocfs2_remove_empty_extent(el); | |
2881 | ||
2882 | ret = ocfs2_journal_dirty(handle, bh); | |
2883 | if (ret) | |
2884 | mlog_errno(ret); | |
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 MF |
2902 | |
2903 | BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))); | |
2904 | ||
2905 | *empty_extent_path = NULL; | |
2906 | ||
e46f74dc | 2907 | ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
328d5752 MF |
2908 | if (ret) { |
2909 | mlog_errno(ret); | |
2910 | goto out; | |
2911 | } | |
2912 | ||
ffdd7a54 | 2913 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2914 | if (!left_path) { |
2915 | ret = -ENOMEM; | |
2916 | mlog_errno(ret); | |
2917 | goto out; | |
2918 | } | |
2919 | ||
2920 | ocfs2_cp_path(left_path, path); | |
2921 | ||
ffdd7a54 | 2922 | right_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2923 | if (!right_path) { |
2924 | ret = -ENOMEM; | |
2925 | mlog_errno(ret); | |
2926 | goto out; | |
2927 | } | |
2928 | ||
2929 | while (right_cpos) { | |
facdb77f | 2930 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
328d5752 MF |
2931 | if (ret) { |
2932 | mlog_errno(ret); | |
2933 | goto out; | |
2934 | } | |
2935 | ||
7dc02805 | 2936 | subtree_root = ocfs2_find_subtree_root(et, left_path, |
328d5752 MF |
2937 | right_path); |
2938 | ||
2939 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2940 | subtree_root, | |
2941 | (unsigned long long) | |
2942 | right_path->p_node[subtree_root].bh->b_blocknr, | |
2943 | right_path->p_tree_depth); | |
2944 | ||
2945 | ret = ocfs2_extend_rotate_transaction(handle, subtree_root, | |
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 | ||
328d5752 | 3018 | |
6136ca5f | 3019 | ret = ocfs2_et_sanity_check(et); |
e7d4cb6b TM |
3020 | if (ret) |
3021 | goto out; | |
328d5752 MF |
3022 | /* |
3023 | * There's two ways we handle this depending on | |
3024 | * whether path is the only existing one. | |
3025 | */ | |
3026 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3027 | handle->h_buffer_credits, | |
3028 | path); | |
3029 | if (ret) { | |
3030 | mlog_errno(ret); | |
3031 | goto out; | |
3032 | } | |
3033 | ||
d9a0a1f8 | 3034 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
328d5752 MF |
3035 | if (ret) { |
3036 | mlog_errno(ret); | |
3037 | goto out; | |
3038 | } | |
3039 | ||
3d03a305 JB |
3040 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3041 | path, &cpos); | |
328d5752 MF |
3042 | if (ret) { |
3043 | mlog_errno(ret); | |
3044 | goto out; | |
3045 | } | |
3046 | ||
3047 | if (cpos) { | |
3048 | /* | |
3049 | * We have a path to the left of this one - it needs | |
3050 | * an update too. | |
3051 | */ | |
ffdd7a54 | 3052 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
3053 | if (!left_path) { |
3054 | ret = -ENOMEM; | |
3055 | mlog_errno(ret); | |
3056 | goto out; | |
3057 | } | |
3058 | ||
facdb77f | 3059 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
3060 | if (ret) { |
3061 | mlog_errno(ret); | |
3062 | goto out; | |
3063 | } | |
3064 | ||
d9a0a1f8 | 3065 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
328d5752 MF |
3066 | if (ret) { |
3067 | mlog_errno(ret); | |
3068 | goto out; | |
3069 | } | |
3070 | ||
7dc02805 | 3071 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
328d5752 | 3072 | |
6641b0ce | 3073 | ocfs2_unlink_subtree(handle, et, left_path, path, |
328d5752 | 3074 | subtree_index, dealloc); |
09106bae | 3075 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
3076 | left_path); |
3077 | if (ret) { | |
3078 | mlog_errno(ret); | |
3079 | goto out; | |
3080 | } | |
328d5752 MF |
3081 | |
3082 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 3083 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
3084 | } else { |
3085 | /* | |
3086 | * 'path' is also the leftmost path which | |
3087 | * means it must be the only one. This gets | |
3088 | * handled differently because we want to | |
70f18c08 | 3089 | * revert the root back to having extents |
328d5752 MF |
3090 | * in-line. |
3091 | */ | |
6641b0ce | 3092 | ocfs2_unlink_path(handle, et, dealloc, path, 1); |
328d5752 | 3093 | |
ce1d9ea6 | 3094 | el = et->et_root_el; |
328d5752 MF |
3095 | el->l_tree_depth = 0; |
3096 | el->l_next_free_rec = 0; | |
3097 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3098 | ||
35dc0aa3 | 3099 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
3100 | } |
3101 | ||
3102 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
3103 | ||
3104 | out: | |
3105 | ocfs2_free_path(left_path); | |
3106 | return ret; | |
3107 | } | |
3108 | ||
3109 | /* | |
3110 | * Left rotation of btree records. | |
3111 | * | |
3112 | * In many ways, this is (unsurprisingly) the opposite of right | |
3113 | * rotation. We start at some non-rightmost path containing an empty | |
3114 | * extent in the leaf block. The code works its way to the rightmost | |
3115 | * path by rotating records to the left in every subtree. | |
3116 | * | |
3117 | * This is used by any code which reduces the number of extent records | |
3118 | * in a leaf. After removal, an empty record should be placed in the | |
3119 | * leftmost list position. | |
3120 | * | |
3121 | * This won't handle a length update of the rightmost path records if | |
3122 | * the rightmost tree leaf record is removed so the caller is | |
3123 | * responsible for detecting and correcting that. | |
3124 | */ | |
70f18c08 JB |
3125 | static int ocfs2_rotate_tree_left(handle_t *handle, |
3126 | struct ocfs2_extent_tree *et, | |
328d5752 | 3127 | struct ocfs2_path *path, |
70f18c08 | 3128 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3129 | { |
3130 | int ret, orig_credits = handle->h_buffer_credits; | |
3131 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
3132 | struct ocfs2_extent_block *eb; | |
3133 | struct ocfs2_extent_list *el; | |
3134 | ||
3135 | el = path_leaf_el(path); | |
3136 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
3137 | return 0; | |
3138 | ||
3139 | if (path->p_tree_depth == 0) { | |
3140 | rightmost_no_delete: | |
3141 | /* | |
e7d4cb6b | 3142 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
3143 | * it up front. |
3144 | */ | |
70f18c08 | 3145 | ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path); |
328d5752 MF |
3146 | if (ret) |
3147 | mlog_errno(ret); | |
3148 | goto out; | |
3149 | } | |
3150 | ||
3151 | /* | |
3152 | * Handle rightmost branch now. There's several cases: | |
3153 | * 1) simple rotation leaving records in there. That's trivial. | |
3154 | * 2) rotation requiring a branch delete - there's no more | |
3155 | * records left. Two cases of this: | |
3156 | * a) There are branches to the left. | |
3157 | * b) This is also the leftmost (the only) branch. | |
3158 | * | |
3159 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
3160 | * 2a) we need the left branch so that we can update it with the unlink | |
70f18c08 | 3161 | * 2b) we need to bring the root back to inline extents. |
328d5752 MF |
3162 | */ |
3163 | ||
3164 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
3165 | el = &eb->h_list; | |
3166 | if (eb->h_next_leaf_blk == 0) { | |
3167 | /* | |
3168 | * This gets a bit tricky if we're going to delete the | |
3169 | * rightmost path. Get the other cases out of the way | |
3170 | * 1st. | |
3171 | */ | |
3172 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
3173 | goto rightmost_no_delete; | |
3174 | ||
3175 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3176 | ret = -EIO; | |
70f18c08 JB |
3177 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3178 | "Owner %llu has empty extent block at %llu", | |
3179 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
328d5752 MF |
3180 | (unsigned long long)le64_to_cpu(eb->h_blkno)); |
3181 | goto out; | |
3182 | } | |
3183 | ||
3184 | /* | |
3185 | * XXX: The caller can not trust "path" any more after | |
3186 | * this as it will have been deleted. What do we do? | |
3187 | * | |
3188 | * In theory the rotate-for-merge code will never get | |
3189 | * here because it'll always ask for a rotate in a | |
3190 | * nonempty list. | |
3191 | */ | |
3192 | ||
70f18c08 JB |
3193 | ret = ocfs2_remove_rightmost_path(handle, et, path, |
3194 | dealloc); | |
328d5752 MF |
3195 | if (ret) |
3196 | mlog_errno(ret); | |
3197 | goto out; | |
3198 | } | |
3199 | ||
3200 | /* | |
3201 | * Now we can loop, remembering the path we get from -EAGAIN | |
3202 | * and restarting from there. | |
3203 | */ | |
3204 | try_rotate: | |
e46f74dc JB |
3205 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path, |
3206 | dealloc, &restart_path); | |
328d5752 MF |
3207 | if (ret && ret != -EAGAIN) { |
3208 | mlog_errno(ret); | |
3209 | goto out; | |
3210 | } | |
3211 | ||
3212 | while (ret == -EAGAIN) { | |
3213 | tmp_path = restart_path; | |
3214 | restart_path = NULL; | |
3215 | ||
e46f74dc | 3216 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, |
328d5752 | 3217 | tmp_path, dealloc, |
e46f74dc | 3218 | &restart_path); |
328d5752 MF |
3219 | if (ret && ret != -EAGAIN) { |
3220 | mlog_errno(ret); | |
3221 | goto out; | |
3222 | } | |
3223 | ||
3224 | ocfs2_free_path(tmp_path); | |
3225 | tmp_path = NULL; | |
3226 | ||
3227 | if (ret == 0) | |
3228 | goto try_rotate; | |
3229 | } | |
3230 | ||
3231 | out: | |
3232 | ocfs2_free_path(tmp_path); | |
3233 | ocfs2_free_path(restart_path); | |
3234 | return ret; | |
3235 | } | |
3236 | ||
3237 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
3238 | int index) | |
3239 | { | |
3240 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
3241 | unsigned int size; | |
3242 | ||
3243 | if (rec->e_leaf_clusters == 0) { | |
3244 | /* | |
3245 | * We consumed all of the merged-from record. An empty | |
3246 | * extent cannot exist anywhere but the 1st array | |
3247 | * position, so move things over if the merged-from | |
3248 | * record doesn't occupy that position. | |
3249 | * | |
3250 | * This creates a new empty extent so the caller | |
3251 | * should be smart enough to have removed any existing | |
3252 | * ones. | |
3253 | */ | |
3254 | if (index > 0) { | |
3255 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3256 | size = index * sizeof(struct ocfs2_extent_rec); | |
3257 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
3258 | } | |
3259 | ||
3260 | /* | |
3261 | * Always memset - the caller doesn't check whether it | |
3262 | * created an empty extent, so there could be junk in | |
3263 | * the other fields. | |
3264 | */ | |
3265 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3266 | } | |
3267 | } | |
3268 | ||
4fe82c31 | 3269 | static int ocfs2_get_right_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3270 | struct ocfs2_path *left_path, |
3271 | struct ocfs2_path **ret_right_path) | |
3272 | { | |
3273 | int ret; | |
3274 | u32 right_cpos; | |
3275 | struct ocfs2_path *right_path = NULL; | |
3276 | struct ocfs2_extent_list *left_el; | |
3277 | ||
3278 | *ret_right_path = NULL; | |
3279 | ||
3280 | /* This function shouldn't be called for non-trees. */ | |
3281 | BUG_ON(left_path->p_tree_depth == 0); | |
3282 | ||
3283 | left_el = path_leaf_el(left_path); | |
3284 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
3285 | ||
4fe82c31 JB |
3286 | ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3287 | left_path, &right_cpos); | |
677b9752 TM |
3288 | if (ret) { |
3289 | mlog_errno(ret); | |
3290 | goto out; | |
3291 | } | |
3292 | ||
3293 | /* This function shouldn't be called for the rightmost leaf. */ | |
3294 | BUG_ON(right_cpos == 0); | |
3295 | ||
ffdd7a54 | 3296 | right_path = ocfs2_new_path_from_path(left_path); |
677b9752 TM |
3297 | if (!right_path) { |
3298 | ret = -ENOMEM; | |
3299 | mlog_errno(ret); | |
3300 | goto out; | |
3301 | } | |
3302 | ||
4fe82c31 | 3303 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
677b9752 TM |
3304 | if (ret) { |
3305 | mlog_errno(ret); | |
3306 | goto out; | |
3307 | } | |
3308 | ||
3309 | *ret_right_path = right_path; | |
3310 | out: | |
3311 | if (ret) | |
3312 | ocfs2_free_path(right_path); | |
3313 | return ret; | |
3314 | } | |
3315 | ||
328d5752 MF |
3316 | /* |
3317 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3318 | * onto the beginning of the record "next" to it. |
3319 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
3320 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
3321 | * next extent block. | |
328d5752 | 3322 | */ |
4fe82c31 | 3323 | static int ocfs2_merge_rec_right(struct ocfs2_path *left_path, |
677b9752 | 3324 | handle_t *handle, |
7dc02805 | 3325 | struct ocfs2_extent_tree *et, |
677b9752 TM |
3326 | struct ocfs2_extent_rec *split_rec, |
3327 | int index) | |
328d5752 | 3328 | { |
677b9752 | 3329 | int ret, next_free, i; |
328d5752 MF |
3330 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3331 | struct ocfs2_extent_rec *left_rec; | |
3332 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3333 | struct ocfs2_extent_list *right_el; |
3334 | struct ocfs2_path *right_path = NULL; | |
3335 | int subtree_index = 0; | |
3336 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3337 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3338 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3339 | |
3340 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3341 | left_rec = &el->l_recs[index]; |
677b9752 | 3342 | |
9d8df6aa | 3343 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3344 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3345 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3346 | ret = ocfs2_get_right_path(et, left_path, &right_path); |
677b9752 TM |
3347 | if (ret) { |
3348 | mlog_errno(ret); | |
3349 | goto out; | |
3350 | } | |
3351 | ||
3352 | right_el = path_leaf_el(right_path); | |
3353 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3354 | BUG_ON(next_free <= 0); | |
3355 | right_rec = &right_el->l_recs[0]; | |
3356 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3357 | BUG_ON(next_free <= 1); |
677b9752 TM |
3358 | right_rec = &right_el->l_recs[1]; |
3359 | } | |
3360 | ||
3361 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3362 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3363 | le32_to_cpu(right_rec->e_cpos)); | |
3364 | ||
7dc02805 JB |
3365 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3366 | right_path); | |
677b9752 TM |
3367 | |
3368 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3369 | handle->h_buffer_credits, | |
3370 | right_path); | |
3371 | if (ret) { | |
3372 | mlog_errno(ret); | |
3373 | goto out; | |
3374 | } | |
3375 | ||
3376 | root_bh = left_path->p_node[subtree_index].bh; | |
3377 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3378 | ||
7dc02805 | 3379 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3380 | subtree_index); |
677b9752 TM |
3381 | if (ret) { |
3382 | mlog_errno(ret); | |
3383 | goto out; | |
3384 | } | |
3385 | ||
3386 | for (i = subtree_index + 1; | |
3387 | i < path_num_items(right_path); i++) { | |
7dc02805 | 3388 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3389 | right_path, i); |
677b9752 TM |
3390 | if (ret) { |
3391 | mlog_errno(ret); | |
3392 | goto out; | |
3393 | } | |
3394 | ||
7dc02805 | 3395 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3396 | left_path, i); |
677b9752 TM |
3397 | if (ret) { |
3398 | mlog_errno(ret); | |
3399 | goto out; | |
3400 | } | |
3401 | } | |
3402 | ||
3403 | } else { | |
3404 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3405 | right_rec = &el->l_recs[index + 1]; | |
3406 | } | |
328d5752 | 3407 | |
7dc02805 | 3408 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path, |
13723d00 | 3409 | path_num_items(left_path) - 1); |
328d5752 MF |
3410 | if (ret) { |
3411 | mlog_errno(ret); | |
3412 | goto out; | |
3413 | } | |
3414 | ||
3415 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3416 | ||
3417 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3418 | le64_add_cpu(&right_rec->e_blkno, | |
7dc02805 JB |
3419 | -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3420 | split_clusters)); | |
328d5752 MF |
3421 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); |
3422 | ||
3423 | ocfs2_cleanup_merge(el, index); | |
3424 | ||
3425 | ret = ocfs2_journal_dirty(handle, bh); | |
3426 | if (ret) | |
3427 | mlog_errno(ret); | |
3428 | ||
677b9752 TM |
3429 | if (right_path) { |
3430 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
3431 | if (ret) | |
3432 | mlog_errno(ret); | |
3433 | ||
4619c73e JB |
3434 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
3435 | subtree_index); | |
677b9752 TM |
3436 | } |
3437 | out: | |
3438 | if (right_path) | |
3439 | ocfs2_free_path(right_path); | |
3440 | return ret; | |
3441 | } | |
3442 | ||
4fe82c31 | 3443 | static int ocfs2_get_left_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3444 | struct ocfs2_path *right_path, |
3445 | struct ocfs2_path **ret_left_path) | |
3446 | { | |
3447 | int ret; | |
3448 | u32 left_cpos; | |
3449 | struct ocfs2_path *left_path = NULL; | |
3450 | ||
3451 | *ret_left_path = NULL; | |
3452 | ||
3453 | /* This function shouldn't be called for non-trees. */ | |
3454 | BUG_ON(right_path->p_tree_depth == 0); | |
3455 | ||
4fe82c31 | 3456 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
677b9752 TM |
3457 | right_path, &left_cpos); |
3458 | if (ret) { | |
3459 | mlog_errno(ret); | |
3460 | goto out; | |
3461 | } | |
3462 | ||
3463 | /* This function shouldn't be called for the leftmost leaf. */ | |
3464 | BUG_ON(left_cpos == 0); | |
3465 | ||
ffdd7a54 | 3466 | left_path = ocfs2_new_path_from_path(right_path); |
677b9752 TM |
3467 | if (!left_path) { |
3468 | ret = -ENOMEM; | |
3469 | mlog_errno(ret); | |
3470 | goto out; | |
3471 | } | |
3472 | ||
4fe82c31 | 3473 | ret = ocfs2_find_path(et->et_ci, left_path, left_cpos); |
677b9752 TM |
3474 | if (ret) { |
3475 | mlog_errno(ret); | |
3476 | goto out; | |
3477 | } | |
3478 | ||
3479 | *ret_left_path = left_path; | |
328d5752 | 3480 | out: |
677b9752 TM |
3481 | if (ret) |
3482 | ocfs2_free_path(left_path); | |
328d5752 MF |
3483 | return ret; |
3484 | } | |
3485 | ||
3486 | /* | |
3487 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3488 | * onto the tail of the record "before" it. |
3489 | * For index > 0, the "before" means the extent rec at index - 1. | |
3490 | * | |
3491 | * For index == 0, the "before" means the last record of the previous | |
3492 | * extent block. And there is also a situation that we may need to | |
3493 | * remove the rightmost leaf extent block in the right_path and change | |
3494 | * the right path to indicate the new rightmost path. | |
328d5752 | 3495 | */ |
4fe82c31 | 3496 | static int ocfs2_merge_rec_left(struct ocfs2_path *right_path, |
328d5752 | 3497 | handle_t *handle, |
4fe82c31 | 3498 | struct ocfs2_extent_tree *et, |
328d5752 | 3499 | struct ocfs2_extent_rec *split_rec, |
677b9752 TM |
3500 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
3501 | int index) | |
328d5752 | 3502 | { |
677b9752 | 3503 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3504 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3505 | struct ocfs2_extent_rec *left_rec; | |
3506 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3507 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3508 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3509 | struct buffer_head *root_bh = NULL; | |
3510 | struct ocfs2_path *left_path = NULL; | |
3511 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3512 | |
677b9752 | 3513 | BUG_ON(index < 0); |
328d5752 | 3514 | |
328d5752 | 3515 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3516 | if (index == 0) { |
3517 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3518 | ret = ocfs2_get_left_path(et, right_path, &left_path); |
677b9752 TM |
3519 | if (ret) { |
3520 | mlog_errno(ret); | |
3521 | goto out; | |
3522 | } | |
3523 | ||
3524 | left_el = path_leaf_el(left_path); | |
3525 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3526 | le16_to_cpu(left_el->l_count)); | |
3527 | ||
3528 | left_rec = &left_el->l_recs[ | |
3529 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3530 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3531 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3532 | le32_to_cpu(split_rec->e_cpos)); | |
3533 | ||
7dc02805 JB |
3534 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3535 | right_path); | |
677b9752 TM |
3536 | |
3537 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3538 | handle->h_buffer_credits, | |
3539 | left_path); | |
3540 | if (ret) { | |
3541 | mlog_errno(ret); | |
3542 | goto out; | |
3543 | } | |
3544 | ||
3545 | root_bh = left_path->p_node[subtree_index].bh; | |
3546 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3547 | ||
4fe82c31 | 3548 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3549 | subtree_index); |
677b9752 TM |
3550 | if (ret) { |
3551 | mlog_errno(ret); | |
3552 | goto out; | |
3553 | } | |
3554 | ||
3555 | for (i = subtree_index + 1; | |
3556 | i < path_num_items(right_path); i++) { | |
4fe82c31 | 3557 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3558 | right_path, i); |
677b9752 TM |
3559 | if (ret) { |
3560 | mlog_errno(ret); | |
3561 | goto out; | |
3562 | } | |
3563 | ||
4fe82c31 | 3564 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3565 | left_path, i); |
677b9752 TM |
3566 | if (ret) { |
3567 | mlog_errno(ret); | |
3568 | goto out; | |
3569 | } | |
3570 | } | |
3571 | } else { | |
3572 | left_rec = &el->l_recs[index - 1]; | |
3573 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3574 | has_empty_extent = 1; | |
3575 | } | |
328d5752 | 3576 | |
4fe82c31 | 3577 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
9047beab | 3578 | path_num_items(right_path) - 1); |
328d5752 MF |
3579 | if (ret) { |
3580 | mlog_errno(ret); | |
3581 | goto out; | |
3582 | } | |
3583 | ||
3584 | if (has_empty_extent && index == 1) { | |
3585 | /* | |
3586 | * The easy case - we can just plop the record right in. | |
3587 | */ | |
3588 | *left_rec = *split_rec; | |
3589 | ||
3590 | has_empty_extent = 0; | |
677b9752 | 3591 | } else |
328d5752 | 3592 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3593 | |
3594 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3595 | le64_add_cpu(&right_rec->e_blkno, | |
4fe82c31 JB |
3596 | ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3597 | split_clusters)); | |
328d5752 MF |
3598 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); |
3599 | ||
3600 | ocfs2_cleanup_merge(el, index); | |
3601 | ||
3602 | ret = ocfs2_journal_dirty(handle, bh); | |
3603 | if (ret) | |
3604 | mlog_errno(ret); | |
3605 | ||
677b9752 TM |
3606 | if (left_path) { |
3607 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
3608 | if (ret) | |
3609 | mlog_errno(ret); | |
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) { | |
3618 | ||
70f18c08 | 3619 | ret = ocfs2_remove_rightmost_path(handle, et, |
e7d4cb6b | 3620 | right_path, |
70f18c08 | 3621 | dealloc); |
677b9752 TM |
3622 | if (ret) { |
3623 | mlog_errno(ret); | |
3624 | goto out; | |
3625 | } | |
3626 | ||
3627 | /* Now the rightmost extent block has been deleted. | |
3628 | * So we use the new rightmost path. | |
3629 | */ | |
3630 | ocfs2_mv_path(right_path, left_path); | |
3631 | left_path = NULL; | |
3632 | } else | |
4619c73e | 3633 | ocfs2_complete_edge_insert(handle, left_path, |
677b9752 TM |
3634 | right_path, subtree_index); |
3635 | } | |
328d5752 | 3636 | out: |
677b9752 TM |
3637 | if (left_path) |
3638 | ocfs2_free_path(left_path); | |
328d5752 MF |
3639 | return ret; |
3640 | } | |
3641 | ||
c495dd24 JB |
3642 | static int ocfs2_try_to_merge_extent(handle_t *handle, |
3643 | struct ocfs2_extent_tree *et, | |
677b9752 | 3644 | struct ocfs2_path *path, |
328d5752 MF |
3645 | int split_index, |
3646 | struct ocfs2_extent_rec *split_rec, | |
3647 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
c495dd24 | 3648 | struct ocfs2_merge_ctxt *ctxt) |
328d5752 | 3649 | { |
518d7269 | 3650 | int ret = 0; |
677b9752 | 3651 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3652 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3653 | ||
3654 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3655 | ||
518d7269 TM |
3656 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
3657 | /* | |
3658 | * The merge code will need to create an empty | |
3659 | * extent to take the place of the newly | |
3660 | * emptied slot. Remove any pre-existing empty | |
3661 | * extents - having more than one in a leaf is | |
3662 | * illegal. | |
3663 | */ | |
70f18c08 | 3664 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
518d7269 TM |
3665 | if (ret) { |
3666 | mlog_errno(ret); | |
3667 | goto out; | |
328d5752 | 3668 | } |
518d7269 TM |
3669 | split_index--; |
3670 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3671 | } |
3672 | ||
3673 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3674 | /* | |
3675 | * Left-right contig implies this. | |
3676 | */ | |
3677 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3678 | |
3679 | /* | |
3680 | * Since the leftright insert always covers the entire | |
3681 | * extent, this call will delete the insert record | |
3682 | * entirely, resulting in an empty extent record added to | |
3683 | * the extent block. | |
3684 | * | |
3685 | * Since the adding of an empty extent shifts | |
3686 | * everything back to the right, there's no need to | |
3687 | * update split_index here. | |
677b9752 TM |
3688 | * |
3689 | * When the split_index is zero, we need to merge it to the | |
3690 | * prevoius extent block. It is more efficient and easier | |
3691 | * if we do merge_right first and merge_left later. | |
328d5752 | 3692 | */ |
4fe82c31 | 3693 | ret = ocfs2_merge_rec_right(path, handle, et, split_rec, |
677b9752 | 3694 | split_index); |
328d5752 MF |
3695 | if (ret) { |
3696 | mlog_errno(ret); | |
3697 | goto out; | |
3698 | } | |
3699 | ||
3700 | /* | |
3701 | * We can only get this from logic error above. | |
3702 | */ | |
3703 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3704 | ||
677b9752 | 3705 | /* The merge left us with an empty extent, remove it. */ |
70f18c08 | 3706 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3707 | if (ret) { |
3708 | mlog_errno(ret); | |
3709 | goto out; | |
3710 | } | |
677b9752 | 3711 | |
328d5752 MF |
3712 | rec = &el->l_recs[split_index]; |
3713 | ||
3714 | /* | |
3715 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3716 | * we've merged it into the rec already. |
328d5752 | 3717 | */ |
4fe82c31 JB |
3718 | ret = ocfs2_merge_rec_left(path, handle, et, rec, |
3719 | dealloc, split_index); | |
677b9752 | 3720 | |
328d5752 MF |
3721 | if (ret) { |
3722 | mlog_errno(ret); | |
3723 | goto out; | |
3724 | } | |
3725 | ||
70f18c08 | 3726 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3727 | /* |
3728 | * Error from this last rotate is not critical, so | |
3729 | * print but don't bubble it up. | |
3730 | */ | |
3731 | if (ret) | |
3732 | mlog_errno(ret); | |
3733 | ret = 0; | |
3734 | } else { | |
3735 | /* | |
3736 | * Merge a record to the left or right. | |
3737 | * | |
3738 | * 'contig_type' is relative to the existing record, | |
3739 | * so for example, if we're "right contig", it's to | |
3740 | * the record on the left (hence the left merge). | |
3741 | */ | |
3742 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
4fe82c31 JB |
3743 | ret = ocfs2_merge_rec_left(path, handle, et, |
3744 | split_rec, dealloc, | |
328d5752 MF |
3745 | split_index); |
3746 | if (ret) { | |
3747 | mlog_errno(ret); | |
3748 | goto out; | |
3749 | } | |
3750 | } else { | |
4fe82c31 | 3751 | ret = ocfs2_merge_rec_right(path, handle, |
7dc02805 | 3752 | et, split_rec, |
328d5752 MF |
3753 | split_index); |
3754 | if (ret) { | |
3755 | mlog_errno(ret); | |
3756 | goto out; | |
3757 | } | |
3758 | } | |
3759 | ||
3760 | if (ctxt->c_split_covers_rec) { | |
3761 | /* | |
3762 | * The merge may have left an empty extent in | |
3763 | * our leaf. Try to rotate it away. | |
3764 | */ | |
70f18c08 JB |
3765 | ret = ocfs2_rotate_tree_left(handle, et, path, |
3766 | dealloc); | |
328d5752 MF |
3767 | if (ret) |
3768 | mlog_errno(ret); | |
3769 | ret = 0; | |
3770 | } | |
3771 | } | |
3772 | ||
3773 | out: | |
3774 | return ret; | |
3775 | } | |
3776 | ||
3777 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3778 | enum ocfs2_split_type split, | |
3779 | struct ocfs2_extent_rec *rec, | |
3780 | struct ocfs2_extent_rec *split_rec) | |
3781 | { | |
3782 | u64 len_blocks; | |
3783 | ||
3784 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3785 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3786 | ||
3787 | if (split == SPLIT_LEFT) { | |
3788 | /* | |
3789 | * Region is on the left edge of the existing | |
3790 | * record. | |
3791 | */ | |
3792 | le32_add_cpu(&rec->e_cpos, | |
3793 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3794 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3795 | le16_add_cpu(&rec->e_leaf_clusters, | |
3796 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3797 | } else { | |
3798 | /* | |
3799 | * Region is on the right edge of the existing | |
3800 | * record. | |
3801 | */ | |
3802 | le16_add_cpu(&rec->e_leaf_clusters, | |
3803 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3804 | } | |
3805 | } | |
3806 | ||
3807 | /* | |
3808 | * Do the final bits of extent record insertion at the target leaf | |
3809 | * list. If this leaf is part of an allocation tree, it is assumed | |
3810 | * that the tree above has been prepared. | |
3811 | */ | |
d5628623 JB |
3812 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et, |
3813 | struct ocfs2_extent_rec *insert_rec, | |
328d5752 | 3814 | struct ocfs2_extent_list *el, |
d5628623 | 3815 | struct ocfs2_insert_type *insert) |
328d5752 MF |
3816 | { |
3817 | int i = insert->ins_contig_index; | |
3818 | unsigned int range; | |
3819 | struct ocfs2_extent_rec *rec; | |
3820 | ||
3821 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3822 | ||
3823 | if (insert->ins_split != SPLIT_NONE) { | |
3824 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3825 | BUG_ON(i == -1); | |
3826 | rec = &el->l_recs[i]; | |
d5628623 JB |
3827 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
3828 | insert->ins_split, rec, | |
328d5752 MF |
3829 | insert_rec); |
3830 | goto rotate; | |
3831 | } | |
3832 | ||
3833 | /* | |
3834 | * Contiguous insert - either left or right. | |
3835 | */ | |
3836 | if (insert->ins_contig != CONTIG_NONE) { | |
3837 | rec = &el->l_recs[i]; | |
3838 | if (insert->ins_contig == CONTIG_LEFT) { | |
3839 | rec->e_blkno = insert_rec->e_blkno; | |
3840 | rec->e_cpos = insert_rec->e_cpos; | |
3841 | } | |
3842 | le16_add_cpu(&rec->e_leaf_clusters, | |
3843 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3844 | return; | |
3845 | } | |
3846 | ||
3847 | /* | |
3848 | * Handle insert into an empty leaf. | |
3849 | */ | |
3850 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3851 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3852 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3853 | el->l_recs[0] = *insert_rec; | |
3854 | el->l_next_free_rec = cpu_to_le16(1); | |
3855 | return; | |
3856 | } | |
3857 | ||
3858 | /* | |
3859 | * Appending insert. | |
3860 | */ | |
3861 | if (insert->ins_appending == APPEND_TAIL) { | |
3862 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3863 | rec = &el->l_recs[i]; | |
3864 | range = le32_to_cpu(rec->e_cpos) | |
3865 | + le16_to_cpu(rec->e_leaf_clusters); | |
3866 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3867 | ||
3868 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3869 | le16_to_cpu(el->l_count), | |
d5628623 | 3870 | "owner %llu, depth %u, count %u, next free %u, " |
328d5752 MF |
3871 | "rec.cpos %u, rec.clusters %u, " |
3872 | "insert.cpos %u, insert.clusters %u\n", | |
d5628623 | 3873 | ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
3874 | le16_to_cpu(el->l_tree_depth), |
3875 | le16_to_cpu(el->l_count), | |
3876 | le16_to_cpu(el->l_next_free_rec), | |
3877 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3878 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3879 | le32_to_cpu(insert_rec->e_cpos), | |
3880 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3881 | i++; | |
3882 | el->l_recs[i] = *insert_rec; | |
3883 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3884 | return; | |
3885 | } | |
3886 | ||
3887 | rotate: | |
3888 | /* | |
3889 | * Ok, we have to rotate. | |
3890 | * | |
3891 | * At this point, it is safe to assume that inserting into an | |
3892 | * empty leaf and appending to a leaf have both been handled | |
3893 | * above. | |
3894 | * | |
3895 | * This leaf needs to have space, either by the empty 1st | |
3896 | * extent record, or by virtue of an l_next_rec < l_count. | |
3897 | */ | |
3898 | ocfs2_rotate_leaf(el, insert_rec); | |
3899 | } | |
3900 | ||
d401dc12 JB |
3901 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
3902 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3903 | struct ocfs2_path *path, |
3904 | struct ocfs2_extent_rec *insert_rec) | |
3905 | { | |
3906 | int ret, i, next_free; | |
3907 | struct buffer_head *bh; | |
3908 | struct ocfs2_extent_list *el; | |
3909 | struct ocfs2_extent_rec *rec; | |
3910 | ||
3911 | /* | |
3912 | * Update everything except the leaf block. | |
3913 | */ | |
3914 | for (i = 0; i < path->p_tree_depth; i++) { | |
3915 | bh = path->p_node[i].bh; | |
3916 | el = path->p_node[i].el; | |
3917 | ||
dcd0538f MF |
3918 | next_free = le16_to_cpu(el->l_next_free_rec); |
3919 | if (next_free == 0) { | |
d401dc12 JB |
3920 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3921 | "Owner %llu has a bad extent list", | |
3922 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
dcd0538f | 3923 | ret = -EIO; |
328d5752 MF |
3924 | return; |
3925 | } | |
3926 | ||
3927 | rec = &el->l_recs[next_free - 1]; | |
3928 | ||
3929 | rec->e_int_clusters = insert_rec->e_cpos; | |
3930 | le32_add_cpu(&rec->e_int_clusters, | |
3931 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3932 | le32_add_cpu(&rec->e_int_clusters, | |
3933 | -le32_to_cpu(rec->e_cpos)); | |
3934 | ||
3935 | ret = ocfs2_journal_dirty(handle, bh); | |
3936 | if (ret) | |
3937 | mlog_errno(ret); | |
3938 | ||
3939 | } | |
3940 | } | |
3941 | ||
d401dc12 JB |
3942 | static int ocfs2_append_rec_to_path(handle_t *handle, |
3943 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3944 | struct ocfs2_extent_rec *insert_rec, |
3945 | struct ocfs2_path *right_path, | |
3946 | struct ocfs2_path **ret_left_path) | |
3947 | { | |
3948 | int ret, next_free; | |
3949 | struct ocfs2_extent_list *el; | |
3950 | struct ocfs2_path *left_path = NULL; | |
3951 | ||
3952 | *ret_left_path = NULL; | |
3953 | ||
3954 | /* | |
3955 | * This shouldn't happen for non-trees. The extent rec cluster | |
3956 | * count manipulation below only works for interior nodes. | |
3957 | */ | |
3958 | BUG_ON(right_path->p_tree_depth == 0); | |
3959 | ||
3960 | /* | |
3961 | * If our appending insert is at the leftmost edge of a leaf, | |
3962 | * then we might need to update the rightmost records of the | |
3963 | * neighboring path. | |
3964 | */ | |
3965 | el = path_leaf_el(right_path); | |
3966 | next_free = le16_to_cpu(el->l_next_free_rec); | |
3967 | if (next_free == 0 || | |
3968 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3969 | u32 left_cpos; | |
3970 | ||
d401dc12 JB |
3971 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3972 | right_path, &left_cpos); | |
328d5752 MF |
3973 | if (ret) { |
3974 | mlog_errno(ret); | |
dcd0538f MF |
3975 | goto out; |
3976 | } | |
3977 | ||
328d5752 MF |
3978 | mlog(0, "Append may need a left path update. cpos: %u, " |
3979 | "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos), | |
3980 | left_cpos); | |
e48edee2 | 3981 | |
328d5752 MF |
3982 | /* |
3983 | * No need to worry if the append is already in the | |
3984 | * leftmost leaf. | |
3985 | */ | |
3986 | if (left_cpos) { | |
ffdd7a54 | 3987 | left_path = ocfs2_new_path_from_path(right_path); |
328d5752 MF |
3988 | if (!left_path) { |
3989 | ret = -ENOMEM; | |
3990 | mlog_errno(ret); | |
3991 | goto out; | |
3992 | } | |
dcd0538f | 3993 | |
d401dc12 | 3994 | ret = ocfs2_find_path(et->et_ci, left_path, |
facdb77f | 3995 | left_cpos); |
328d5752 MF |
3996 | if (ret) { |
3997 | mlog_errno(ret); | |
3998 | goto out; | |
3999 | } | |
dcd0538f | 4000 | |
328d5752 MF |
4001 | /* |
4002 | * ocfs2_insert_path() will pass the left_path to the | |
4003 | * journal for us. | |
4004 | */ | |
4005 | } | |
4006 | } | |
dcd0538f | 4007 | |
d401dc12 | 4008 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
328d5752 MF |
4009 | if (ret) { |
4010 | mlog_errno(ret); | |
4011 | goto out; | |
dcd0538f MF |
4012 | } |
4013 | ||
d401dc12 | 4014 | ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec); |
328d5752 | 4015 | |
dcd0538f MF |
4016 | *ret_left_path = left_path; |
4017 | ret = 0; | |
4018 | out: | |
4019 | if (ret != 0) | |
4020 | ocfs2_free_path(left_path); | |
4021 | ||
4022 | return ret; | |
4023 | } | |
4024 | ||
c38e52bb | 4025 | static void ocfs2_split_record(struct ocfs2_extent_tree *et, |
328d5752 MF |
4026 | struct ocfs2_path *left_path, |
4027 | struct ocfs2_path *right_path, | |
4028 | struct ocfs2_extent_rec *split_rec, | |
4029 | enum ocfs2_split_type split) | |
4030 | { | |
4031 | int index; | |
4032 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
4033 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
4034 | struct ocfs2_extent_rec *rec, *tmprec; | |
4035 | ||
c19a28e1 | 4036 | right_el = path_leaf_el(right_path); |
328d5752 MF |
4037 | if (left_path) |
4038 | left_el = path_leaf_el(left_path); | |
4039 | ||
4040 | el = right_el; | |
4041 | insert_el = right_el; | |
4042 | index = ocfs2_search_extent_list(el, cpos); | |
4043 | if (index != -1) { | |
4044 | if (index == 0 && left_path) { | |
4045 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
4046 | ||
4047 | /* | |
4048 | * This typically means that the record | |
4049 | * started in the left path but moved to the | |
4050 | * right as a result of rotation. We either | |
4051 | * move the existing record to the left, or we | |
4052 | * do the later insert there. | |
4053 | * | |
4054 | * In this case, the left path should always | |
4055 | * exist as the rotate code will have passed | |
4056 | * it back for a post-insert update. | |
4057 | */ | |
4058 | ||
4059 | if (split == SPLIT_LEFT) { | |
4060 | /* | |
4061 | * It's a left split. Since we know | |
4062 | * that the rotate code gave us an | |
4063 | * empty extent in the left path, we | |
4064 | * can just do the insert there. | |
4065 | */ | |
4066 | insert_el = left_el; | |
4067 | } else { | |
4068 | /* | |
4069 | * Right split - we have to move the | |
4070 | * existing record over to the left | |
4071 | * leaf. The insert will be into the | |
4072 | * newly created empty extent in the | |
4073 | * right leaf. | |
4074 | */ | |
4075 | tmprec = &right_el->l_recs[index]; | |
4076 | ocfs2_rotate_leaf(left_el, tmprec); | |
4077 | el = left_el; | |
4078 | ||
4079 | memset(tmprec, 0, sizeof(*tmprec)); | |
4080 | index = ocfs2_search_extent_list(left_el, cpos); | |
4081 | BUG_ON(index == -1); | |
4082 | } | |
4083 | } | |
4084 | } else { | |
4085 | BUG_ON(!left_path); | |
4086 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
4087 | /* | |
4088 | * Left path is easy - we can just allow the insert to | |
4089 | * happen. | |
4090 | */ | |
4091 | el = left_el; | |
4092 | insert_el = left_el; | |
4093 | index = ocfs2_search_extent_list(el, cpos); | |
4094 | BUG_ON(index == -1); | |
4095 | } | |
4096 | ||
4097 | rec = &el->l_recs[index]; | |
c38e52bb JB |
4098 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
4099 | split, rec, split_rec); | |
328d5752 MF |
4100 | ocfs2_rotate_leaf(insert_el, split_rec); |
4101 | } | |
4102 | ||
dcd0538f | 4103 | /* |
e7d4cb6b TM |
4104 | * This function only does inserts on an allocation b-tree. For tree |
4105 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
4106 | * |
4107 | * right_path is the path we want to do the actual insert | |
4108 | * in. left_path should only be passed in if we need to update that | |
4109 | * portion of the tree after an edge insert. | |
4110 | */ | |
3505bec0 | 4111 | static int ocfs2_insert_path(handle_t *handle, |
7dc02805 | 4112 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4113 | struct ocfs2_path *left_path, |
4114 | struct ocfs2_path *right_path, | |
4115 | struct ocfs2_extent_rec *insert_rec, | |
4116 | struct ocfs2_insert_type *insert) | |
4117 | { | |
4118 | int ret, subtree_index; | |
4119 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 4120 | |
dcd0538f MF |
4121 | if (left_path) { |
4122 | int credits = handle->h_buffer_credits; | |
4123 | ||
4124 | /* | |
4125 | * There's a chance that left_path got passed back to | |
4126 | * us without being accounted for in the | |
4127 | * journal. Extend our transaction here to be sure we | |
4128 | * can change those blocks. | |
4129 | */ | |
4130 | credits += left_path->p_tree_depth; | |
4131 | ||
4132 | ret = ocfs2_extend_trans(handle, credits); | |
4133 | if (ret < 0) { | |
4134 | mlog_errno(ret); | |
4135 | goto out; | |
4136 | } | |
4137 | ||
7dc02805 | 4138 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
dcd0538f MF |
4139 | if (ret < 0) { |
4140 | mlog_errno(ret); | |
4141 | goto out; | |
4142 | } | |
4143 | } | |
4144 | ||
e8aed345 MF |
4145 | /* |
4146 | * Pass both paths to the journal. The majority of inserts | |
4147 | * will be touching all components anyway. | |
4148 | */ | |
7dc02805 | 4149 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
e8aed345 MF |
4150 | if (ret < 0) { |
4151 | mlog_errno(ret); | |
4152 | goto out; | |
4153 | } | |
4154 | ||
328d5752 MF |
4155 | if (insert->ins_split != SPLIT_NONE) { |
4156 | /* | |
4157 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 4158 | * of splits, but it's easier to just let one separate |
328d5752 MF |
4159 | * function sort it all out. |
4160 | */ | |
c38e52bb | 4161 | ocfs2_split_record(et, left_path, right_path, |
328d5752 | 4162 | insert_rec, insert->ins_split); |
e8aed345 MF |
4163 | |
4164 | /* | |
4165 | * Split might have modified either leaf and we don't | |
4166 | * have a guarantee that the later edge insert will | |
4167 | * dirty this for us. | |
4168 | */ | |
4169 | if (left_path) | |
4170 | ret = ocfs2_journal_dirty(handle, | |
4171 | path_leaf_bh(left_path)); | |
4172 | if (ret) | |
4173 | mlog_errno(ret); | |
328d5752 | 4174 | } else |
d5628623 JB |
4175 | ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path), |
4176 | insert); | |
dcd0538f | 4177 | |
dcd0538f MF |
4178 | ret = ocfs2_journal_dirty(handle, leaf_bh); |
4179 | if (ret) | |
4180 | mlog_errno(ret); | |
4181 | ||
4182 | if (left_path) { | |
4183 | /* | |
4184 | * The rotate code has indicated that we need to fix | |
4185 | * up portions of the tree after the insert. | |
4186 | * | |
4187 | * XXX: Should we extend the transaction here? | |
4188 | */ | |
7dc02805 | 4189 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
dcd0538f | 4190 | right_path); |
4619c73e JB |
4191 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
4192 | subtree_index); | |
dcd0538f MF |
4193 | } |
4194 | ||
4195 | ret = 0; | |
4196 | out: | |
4197 | return ret; | |
4198 | } | |
4199 | ||
3505bec0 | 4200 | static int ocfs2_do_insert_extent(handle_t *handle, |
e7d4cb6b | 4201 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4202 | struct ocfs2_extent_rec *insert_rec, |
4203 | struct ocfs2_insert_type *type) | |
4204 | { | |
4205 | int ret, rotate = 0; | |
4206 | u32 cpos; | |
4207 | struct ocfs2_path *right_path = NULL; | |
4208 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
4209 | struct ocfs2_extent_list *el; |
4210 | ||
ce1d9ea6 | 4211 | el = et->et_root_el; |
dcd0538f | 4212 | |
d9a0a1f8 | 4213 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4214 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
4215 | if (ret) { |
4216 | mlog_errno(ret); | |
4217 | goto out; | |
4218 | } | |
4219 | ||
4220 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
d5628623 | 4221 | ocfs2_insert_at_leaf(et, insert_rec, el, type); |
dcd0538f MF |
4222 | goto out_update_clusters; |
4223 | } | |
4224 | ||
ffdd7a54 | 4225 | right_path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4226 | if (!right_path) { |
4227 | ret = -ENOMEM; | |
4228 | mlog_errno(ret); | |
4229 | goto out; | |
4230 | } | |
4231 | ||
4232 | /* | |
4233 | * Determine the path to start with. Rotations need the | |
4234 | * rightmost path, everything else can go directly to the | |
4235 | * target leaf. | |
4236 | */ | |
4237 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
4238 | if (type->ins_appending == APPEND_NONE && | |
4239 | type->ins_contig == CONTIG_NONE) { | |
4240 | rotate = 1; | |
4241 | cpos = UINT_MAX; | |
4242 | } | |
4243 | ||
facdb77f | 4244 | ret = ocfs2_find_path(et->et_ci, right_path, cpos); |
dcd0538f MF |
4245 | if (ret) { |
4246 | mlog_errno(ret); | |
4247 | goto out; | |
4248 | } | |
4249 | ||
4250 | /* | |
4251 | * Rotations and appends need special treatment - they modify | |
4252 | * parts of the tree's above them. | |
4253 | * | |
4254 | * Both might pass back a path immediate to the left of the | |
4255 | * one being inserted to. This will be cause | |
4256 | * ocfs2_insert_path() to modify the rightmost records of | |
4257 | * left_path to account for an edge insert. | |
4258 | * | |
4259 | * XXX: When modifying this code, keep in mind that an insert | |
4260 | * can wind up skipping both of these two special cases... | |
4261 | */ | |
4262 | if (rotate) { | |
1bbf0b8d | 4263 | ret = ocfs2_rotate_tree_right(handle, et, type->ins_split, |
dcd0538f MF |
4264 | le32_to_cpu(insert_rec->e_cpos), |
4265 | right_path, &left_path); | |
4266 | if (ret) { | |
4267 | mlog_errno(ret); | |
4268 | goto out; | |
4269 | } | |
e8aed345 MF |
4270 | |
4271 | /* | |
4272 | * ocfs2_rotate_tree_right() might have extended the | |
4273 | * transaction without re-journaling our tree root. | |
4274 | */ | |
d9a0a1f8 | 4275 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4276 | OCFS2_JOURNAL_ACCESS_WRITE); |
e8aed345 MF |
4277 | if (ret) { |
4278 | mlog_errno(ret); | |
4279 | goto out; | |
4280 | } | |
dcd0538f MF |
4281 | } else if (type->ins_appending == APPEND_TAIL |
4282 | && type->ins_contig != CONTIG_LEFT) { | |
d401dc12 | 4283 | ret = ocfs2_append_rec_to_path(handle, et, insert_rec, |
dcd0538f MF |
4284 | right_path, &left_path); |
4285 | if (ret) { | |
4286 | mlog_errno(ret); | |
4287 | goto out; | |
4288 | } | |
4289 | } | |
4290 | ||
3505bec0 | 4291 | ret = ocfs2_insert_path(handle, et, left_path, right_path, |
dcd0538f MF |
4292 | insert_rec, type); |
4293 | if (ret) { | |
4294 | mlog_errno(ret); | |
4295 | goto out; | |
4296 | } | |
4297 | ||
4298 | out_update_clusters: | |
328d5752 | 4299 | if (type->ins_split == SPLIT_NONE) |
6136ca5f | 4300 | ocfs2_et_update_clusters(et, |
35dc0aa3 | 4301 | le16_to_cpu(insert_rec->e_leaf_clusters)); |
dcd0538f | 4302 | |
ce1d9ea6 | 4303 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); |
dcd0538f MF |
4304 | if (ret) |
4305 | mlog_errno(ret); | |
4306 | ||
4307 | out: | |
4308 | ocfs2_free_path(left_path); | |
4309 | ocfs2_free_path(right_path); | |
4310 | ||
4311 | return ret; | |
4312 | } | |
4313 | ||
328d5752 | 4314 | static enum ocfs2_contig_type |
a2970291 JB |
4315 | ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et, |
4316 | struct ocfs2_path *path, | |
328d5752 MF |
4317 | struct ocfs2_extent_list *el, int index, |
4318 | struct ocfs2_extent_rec *split_rec) | |
4319 | { | |
ad5a4d70 | 4320 | int status; |
328d5752 | 4321 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4322 | u32 left_cpos, right_cpos; |
4323 | struct ocfs2_extent_rec *rec = NULL; | |
4324 | struct ocfs2_extent_list *new_el; | |
4325 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4326 | struct buffer_head *bh; | |
4327 | struct ocfs2_extent_block *eb; | |
a2970291 | 4328 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
ad5a4d70 TM |
4329 | |
4330 | if (index > 0) { | |
4331 | rec = &el->l_recs[index - 1]; | |
4332 | } else if (path->p_tree_depth > 0) { | |
a2970291 | 4333 | status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
ad5a4d70 TM |
4334 | if (status) |
4335 | goto out; | |
4336 | ||
4337 | if (left_cpos != 0) { | |
ffdd7a54 | 4338 | left_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4339 | if (!left_path) |
4340 | goto out; | |
4341 | ||
a2970291 JB |
4342 | status = ocfs2_find_path(et->et_ci, left_path, |
4343 | left_cpos); | |
ad5a4d70 TM |
4344 | if (status) |
4345 | goto out; | |
4346 | ||
4347 | new_el = path_leaf_el(left_path); | |
4348 | ||
4349 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4350 | le16_to_cpu(new_el->l_count)) { | |
4351 | bh = path_leaf_bh(left_path); | |
4352 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4353 | ocfs2_error(sb, |
5e96581a JB |
4354 | "Extent block #%llu has an " |
4355 | "invalid l_next_free_rec of " | |
4356 | "%d. It should have " | |
4357 | "matched the l_count of %d", | |
4358 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4359 | le16_to_cpu(new_el->l_next_free_rec), | |
4360 | le16_to_cpu(new_el->l_count)); | |
4361 | status = -EINVAL; | |
ad5a4d70 TM |
4362 | goto out; |
4363 | } | |
4364 | rec = &new_el->l_recs[ | |
4365 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4366 | } | |
4367 | } | |
328d5752 MF |
4368 | |
4369 | /* | |
4370 | * We're careful to check for an empty extent record here - | |
4371 | * the merge code will know what to do if it sees one. | |
4372 | */ | |
ad5a4d70 | 4373 | if (rec) { |
328d5752 MF |
4374 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4375 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4376 | ret = CONTIG_RIGHT; | |
4377 | } else { | |
a2970291 | 4378 | ret = ocfs2_extent_contig(sb, rec, split_rec); |
328d5752 MF |
4379 | } |
4380 | } | |
4381 | ||
ad5a4d70 TM |
4382 | rec = NULL; |
4383 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4384 | rec = &el->l_recs[index + 1]; | |
4385 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4386 | path->p_tree_depth > 0) { | |
a2970291 | 4387 | status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
ad5a4d70 TM |
4388 | if (status) |
4389 | goto out; | |
4390 | ||
4391 | if (right_cpos == 0) | |
4392 | goto out; | |
4393 | ||
ffdd7a54 | 4394 | right_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4395 | if (!right_path) |
4396 | goto out; | |
4397 | ||
a2970291 | 4398 | status = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
ad5a4d70 TM |
4399 | if (status) |
4400 | goto out; | |
4401 | ||
4402 | new_el = path_leaf_el(right_path); | |
4403 | rec = &new_el->l_recs[0]; | |
4404 | if (ocfs2_is_empty_extent(rec)) { | |
4405 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4406 | bh = path_leaf_bh(right_path); | |
4407 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4408 | ocfs2_error(sb, |
5e96581a JB |
4409 | "Extent block #%llu has an " |
4410 | "invalid l_next_free_rec of %d", | |
4411 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4412 | le16_to_cpu(new_el->l_next_free_rec)); | |
4413 | status = -EINVAL; | |
ad5a4d70 TM |
4414 | goto out; |
4415 | } | |
4416 | rec = &new_el->l_recs[1]; | |
4417 | } | |
4418 | } | |
4419 | ||
4420 | if (rec) { | |
328d5752 MF |
4421 | enum ocfs2_contig_type contig_type; |
4422 | ||
a2970291 | 4423 | contig_type = ocfs2_extent_contig(sb, rec, split_rec); |
328d5752 MF |
4424 | |
4425 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4426 | ret = CONTIG_LEFTRIGHT; | |
4427 | else if (ret == CONTIG_NONE) | |
4428 | ret = contig_type; | |
4429 | } | |
4430 | ||
ad5a4d70 TM |
4431 | out: |
4432 | if (left_path) | |
4433 | ocfs2_free_path(left_path); | |
4434 | if (right_path) | |
4435 | ocfs2_free_path(right_path); | |
4436 | ||
328d5752 MF |
4437 | return ret; |
4438 | } | |
4439 | ||
dcd0538f MF |
4440 | static void ocfs2_figure_contig_type(struct inode *inode, |
4441 | struct ocfs2_insert_type *insert, | |
4442 | struct ocfs2_extent_list *el, | |
ca12b7c4 TM |
4443 | struct ocfs2_extent_rec *insert_rec, |
4444 | struct ocfs2_extent_tree *et) | |
dcd0538f MF |
4445 | { |
4446 | int i; | |
4447 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4448 | ||
e48edee2 MF |
4449 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4450 | ||
dcd0538f | 4451 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
b4a17651 | 4452 | contig_type = ocfs2_extent_contig(inode->i_sb, &el->l_recs[i], |
dcd0538f MF |
4453 | insert_rec); |
4454 | if (contig_type != CONTIG_NONE) { | |
4455 | insert->ins_contig_index = i; | |
4456 | break; | |
4457 | } | |
4458 | } | |
4459 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4460 | |
4461 | if (insert->ins_contig != CONTIG_NONE) { | |
4462 | struct ocfs2_extent_rec *rec = | |
4463 | &el->l_recs[insert->ins_contig_index]; | |
4464 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4465 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4466 | ||
4467 | /* | |
4468 | * Caller might want us to limit the size of extents, don't | |
4469 | * calculate contiguousness if we might exceed that limit. | |
4470 | */ | |
ce1d9ea6 JB |
4471 | if (et->et_max_leaf_clusters && |
4472 | (len > et->et_max_leaf_clusters)) | |
ca12b7c4 TM |
4473 | insert->ins_contig = CONTIG_NONE; |
4474 | } | |
dcd0538f MF |
4475 | } |
4476 | ||
4477 | /* | |
4478 | * This should only be called against the righmost leaf extent list. | |
4479 | * | |
4480 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4481 | * insert at the tail of the rightmost leaf. | |
4482 | * | |
e7d4cb6b TM |
4483 | * This should also work against the root extent list for tree's with 0 |
4484 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4485 | * then the logic here makes sense. |
4486 | */ | |
4487 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4488 | struct ocfs2_extent_list *el, | |
4489 | struct ocfs2_extent_rec *insert_rec) | |
4490 | { | |
4491 | int i; | |
4492 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4493 | struct ocfs2_extent_rec *rec; | |
4494 | ||
4495 | insert->ins_appending = APPEND_NONE; | |
4496 | ||
e48edee2 | 4497 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4498 | |
4499 | if (!el->l_next_free_rec) | |
4500 | goto set_tail_append; | |
4501 | ||
4502 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4503 | /* Were all records empty? */ | |
4504 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4505 | goto set_tail_append; | |
4506 | } | |
4507 | ||
4508 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4509 | rec = &el->l_recs[i]; | |
4510 | ||
e48edee2 MF |
4511 | if (cpos >= |
4512 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4513 | goto set_tail_append; |
4514 | ||
4515 | return; | |
4516 | ||
4517 | set_tail_append: | |
4518 | insert->ins_appending = APPEND_TAIL; | |
4519 | } | |
4520 | ||
4521 | /* | |
4522 | * Helper function called at the begining of an insert. | |
4523 | * | |
4524 | * This computes a few things that are commonly used in the process of | |
4525 | * inserting into the btree: | |
4526 | * - Whether the new extent is contiguous with an existing one. | |
4527 | * - The current tree depth. | |
4528 | * - Whether the insert is an appending one. | |
4529 | * - The total # of free records in the tree. | |
4530 | * | |
4531 | * All of the information is stored on the ocfs2_insert_type | |
4532 | * structure. | |
4533 | */ | |
4534 | static int ocfs2_figure_insert_type(struct inode *inode, | |
e7d4cb6b | 4535 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4536 | struct buffer_head **last_eb_bh, |
4537 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4538 | int *free_records, |
dcd0538f MF |
4539 | struct ocfs2_insert_type *insert) |
4540 | { | |
4541 | int ret; | |
dcd0538f MF |
4542 | struct ocfs2_extent_block *eb; |
4543 | struct ocfs2_extent_list *el; | |
4544 | struct ocfs2_path *path = NULL; | |
4545 | struct buffer_head *bh = NULL; | |
4546 | ||
328d5752 MF |
4547 | insert->ins_split = SPLIT_NONE; |
4548 | ||
ce1d9ea6 | 4549 | el = et->et_root_el; |
dcd0538f MF |
4550 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4551 | ||
4552 | if (el->l_tree_depth) { | |
4553 | /* | |
4554 | * If we have tree depth, we read in the | |
4555 | * rightmost extent block ahead of time as | |
4556 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4557 | * may want it later. | |
4558 | */ | |
3d03a305 | 4559 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
4560 | ocfs2_et_get_last_eb_blk(et), |
4561 | &bh); | |
dcd0538f MF |
4562 | if (ret) { |
4563 | mlog_exit(ret); | |
4564 | goto out; | |
4565 | } | |
ccd979bd | 4566 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4567 | el = &eb->h_list; |
dcd0538f | 4568 | } |
ccd979bd | 4569 | |
dcd0538f MF |
4570 | /* |
4571 | * Unless we have a contiguous insert, we'll need to know if | |
4572 | * there is room left in our allocation tree for another | |
4573 | * extent record. | |
4574 | * | |
4575 | * XXX: This test is simplistic, we can search for empty | |
4576 | * extent records too. | |
4577 | */ | |
c77534f6 | 4578 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4579 | le16_to_cpu(el->l_next_free_rec); |
4580 | ||
4581 | if (!insert->ins_tree_depth) { | |
ca12b7c4 | 4582 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4583 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4584 | return 0; | |
ccd979bd MF |
4585 | } |
4586 | ||
ffdd7a54 | 4587 | path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4588 | if (!path) { |
4589 | ret = -ENOMEM; | |
4590 | mlog_errno(ret); | |
4591 | goto out; | |
4592 | } | |
ccd979bd | 4593 | |
dcd0538f MF |
4594 | /* |
4595 | * In the case that we're inserting past what the tree | |
4596 | * currently accounts for, ocfs2_find_path() will return for | |
4597 | * us the rightmost tree path. This is accounted for below in | |
4598 | * the appending code. | |
4599 | */ | |
facdb77f | 4600 | ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos)); |
dcd0538f MF |
4601 | if (ret) { |
4602 | mlog_errno(ret); | |
4603 | goto out; | |
4604 | } | |
ccd979bd | 4605 | |
dcd0538f MF |
4606 | el = path_leaf_el(path); |
4607 | ||
4608 | /* | |
4609 | * Now that we have the path, there's two things we want to determine: | |
4610 | * 1) Contiguousness (also set contig_index if this is so) | |
4611 | * | |
4612 | * 2) Are we doing an append? We can trivially break this up | |
4613 | * into two types of appends: simple record append, or a | |
4614 | * rotate inside the tail leaf. | |
4615 | */ | |
ca12b7c4 | 4616 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4617 | |
4618 | /* | |
4619 | * The insert code isn't quite ready to deal with all cases of | |
4620 | * left contiguousness. Specifically, if it's an insert into | |
4621 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4622 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4623 | * left. For now, just catch that case and fool the layers |
4624 | * above us. This works just fine for tree_depth == 0, which | |
4625 | * is why we allow that above. | |
4626 | */ | |
4627 | if (insert->ins_contig == CONTIG_LEFT && | |
4628 | insert->ins_contig_index == 0) | |
4629 | insert->ins_contig = CONTIG_NONE; | |
4630 | ||
4631 | /* | |
4632 | * Ok, so we can simply compare against last_eb to figure out | |
4633 | * whether the path doesn't exist. This will only happen in | |
4634 | * the case that we're doing a tail append, so maybe we can | |
4635 | * take advantage of that information somehow. | |
4636 | */ | |
35dc0aa3 | 4637 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4638 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4639 | /* |
4640 | * Ok, ocfs2_find_path() returned us the rightmost | |
4641 | * tree path. This might be an appending insert. There are | |
4642 | * two cases: | |
4643 | * 1) We're doing a true append at the tail: | |
4644 | * -This might even be off the end of the leaf | |
4645 | * 2) We're "appending" by rotating in the tail | |
4646 | */ | |
4647 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4648 | } | |
4649 | ||
4650 | out: | |
4651 | ocfs2_free_path(path); | |
4652 | ||
4653 | if (ret == 0) | |
4654 | *last_eb_bh = bh; | |
4655 | else | |
4656 | brelse(bh); | |
4657 | return ret; | |
ccd979bd MF |
4658 | } |
4659 | ||
dcd0538f MF |
4660 | /* |
4661 | * Insert an extent into an inode btree. | |
4662 | * | |
4663 | * The caller needs to update fe->i_clusters | |
4664 | */ | |
f99b9b7c JB |
4665 | int ocfs2_insert_extent(struct ocfs2_super *osb, |
4666 | handle_t *handle, | |
4667 | struct inode *inode, | |
4668 | struct ocfs2_extent_tree *et, | |
4669 | u32 cpos, | |
4670 | u64 start_blk, | |
4671 | u32 new_clusters, | |
4672 | u8 flags, | |
4673 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 4674 | { |
c3afcbb3 | 4675 | int status; |
c77534f6 | 4676 | int uninitialized_var(free_records); |
ccd979bd | 4677 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4678 | struct ocfs2_insert_type insert = {0, }; |
4679 | struct ocfs2_extent_rec rec; | |
4680 | ||
4681 | mlog(0, "add %u clusters at position %u to inode %llu\n", | |
4682 | new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
4683 | ||
e48edee2 | 4684 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4685 | rec.e_cpos = cpu_to_le32(cpos); |
4686 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4687 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4688 | rec.e_flags = flags; |
6136ca5f | 4689 | status = ocfs2_et_insert_check(et, &rec); |
1e61ee79 JB |
4690 | if (status) { |
4691 | mlog_errno(status); | |
4692 | goto bail; | |
4693 | } | |
dcd0538f | 4694 | |
e7d4cb6b | 4695 | status = ocfs2_figure_insert_type(inode, et, &last_eb_bh, &rec, |
c77534f6 | 4696 | &free_records, &insert); |
dcd0538f MF |
4697 | if (status < 0) { |
4698 | mlog_errno(status); | |
4699 | goto bail; | |
ccd979bd MF |
4700 | } |
4701 | ||
dcd0538f MF |
4702 | mlog(0, "Insert.appending: %u, Insert.Contig: %u, " |
4703 | "Insert.contig_index: %d, Insert.free_records: %d, " | |
4704 | "Insert.tree_depth: %d\n", | |
4705 | insert.ins_appending, insert.ins_contig, insert.ins_contig_index, | |
c77534f6 | 4706 | free_records, insert.ins_tree_depth); |
ccd979bd | 4707 | |
c77534f6 | 4708 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
d401dc12 | 4709 | status = ocfs2_grow_tree(handle, et, |
328d5752 | 4710 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4711 | meta_ac); |
4712 | if (status) { | |
ccd979bd MF |
4713 | mlog_errno(status); |
4714 | goto bail; | |
4715 | } | |
ccd979bd MF |
4716 | } |
4717 | ||
dcd0538f | 4718 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
3505bec0 | 4719 | status = ocfs2_do_insert_extent(handle, et, &rec, &insert); |
ccd979bd MF |
4720 | if (status < 0) |
4721 | mlog_errno(status); | |
f99b9b7c | 4722 | else if (et->et_ops == &ocfs2_dinode_et_ops) |
83418978 | 4723 | ocfs2_extent_map_insert_rec(inode, &rec); |
ccd979bd MF |
4724 | |
4725 | bail: | |
a81cb88b | 4726 | brelse(last_eb_bh); |
ccd979bd | 4727 | |
f56654c4 TM |
4728 | mlog_exit(status); |
4729 | return status; | |
4730 | } | |
4731 | ||
0eb8d47e TM |
4732 | /* |
4733 | * Allcate and add clusters into the extent b-tree. | |
4734 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
f99b9b7c | 4735 | * The extent b-tree's root is specified by et, and |
0eb8d47e TM |
4736 | * it is not limited to the file storage. Any extent tree can use this |
4737 | * function if it implements the proper ocfs2_extent_tree. | |
4738 | */ | |
4739 | int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb, | |
4740 | struct inode *inode, | |
4741 | u32 *logical_offset, | |
4742 | u32 clusters_to_add, | |
4743 | int mark_unwritten, | |
f99b9b7c | 4744 | struct ocfs2_extent_tree *et, |
0eb8d47e TM |
4745 | handle_t *handle, |
4746 | struct ocfs2_alloc_context *data_ac, | |
4747 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 4748 | enum ocfs2_alloc_restarted *reason_ret) |
0eb8d47e TM |
4749 | { |
4750 | int status = 0; | |
4751 | int free_extents; | |
4752 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4753 | u32 bit_off, num_bits; | |
4754 | u64 block; | |
4755 | u8 flags = 0; | |
4756 | ||
4757 | BUG_ON(!clusters_to_add); | |
4758 | ||
4759 | if (mark_unwritten) | |
4760 | flags = OCFS2_EXT_UNWRITTEN; | |
4761 | ||
3d03a305 | 4762 | free_extents = ocfs2_num_free_extents(osb, et); |
0eb8d47e TM |
4763 | if (free_extents < 0) { |
4764 | status = free_extents; | |
4765 | mlog_errno(status); | |
4766 | goto leave; | |
4767 | } | |
4768 | ||
4769 | /* there are two cases which could cause us to EAGAIN in the | |
4770 | * we-need-more-metadata case: | |
4771 | * 1) we haven't reserved *any* | |
4772 | * 2) we are so fragmented, we've needed to add metadata too | |
4773 | * many times. */ | |
4774 | if (!free_extents && !meta_ac) { | |
4775 | mlog(0, "we haven't reserved any metadata!\n"); | |
4776 | status = -EAGAIN; | |
4777 | reason = RESTART_META; | |
4778 | goto leave; | |
4779 | } else if ((!free_extents) | |
4780 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
f99b9b7c | 4781 | < ocfs2_extend_meta_needed(et->et_root_el))) { |
0eb8d47e TM |
4782 | mlog(0, "filesystem is really fragmented...\n"); |
4783 | status = -EAGAIN; | |
4784 | reason = RESTART_META; | |
4785 | goto leave; | |
4786 | } | |
4787 | ||
4788 | status = __ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
4789 | clusters_to_add, &bit_off, &num_bits); | |
4790 | if (status < 0) { | |
4791 | if (status != -ENOSPC) | |
4792 | mlog_errno(status); | |
4793 | goto leave; | |
4794 | } | |
4795 | ||
4796 | BUG_ON(num_bits > clusters_to_add); | |
4797 | ||
13723d00 | 4798 | /* reserve our write early -- insert_extent may update the tree root */ |
d9a0a1f8 | 4799 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4800 | OCFS2_JOURNAL_ACCESS_WRITE); |
0eb8d47e TM |
4801 | if (status < 0) { |
4802 | mlog_errno(status); | |
4803 | goto leave; | |
4804 | } | |
4805 | ||
4806 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
4807 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", | |
4808 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
f99b9b7c JB |
4809 | status = ocfs2_insert_extent(osb, handle, inode, et, |
4810 | *logical_offset, block, | |
4811 | num_bits, flags, meta_ac); | |
0eb8d47e TM |
4812 | if (status < 0) { |
4813 | mlog_errno(status); | |
4814 | goto leave; | |
4815 | } | |
4816 | ||
f99b9b7c | 4817 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
0eb8d47e TM |
4818 | if (status < 0) { |
4819 | mlog_errno(status); | |
4820 | goto leave; | |
4821 | } | |
4822 | ||
4823 | clusters_to_add -= num_bits; | |
4824 | *logical_offset += num_bits; | |
4825 | ||
4826 | if (clusters_to_add) { | |
4827 | mlog(0, "need to alloc once more, wanted = %u\n", | |
4828 | clusters_to_add); | |
4829 | status = -EAGAIN; | |
4830 | reason = RESTART_TRANS; | |
4831 | } | |
4832 | ||
4833 | leave: | |
4834 | mlog_exit(status); | |
4835 | if (reason_ret) | |
4836 | *reason_ret = reason; | |
4837 | return status; | |
4838 | } | |
4839 | ||
328d5752 MF |
4840 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4841 | struct ocfs2_extent_rec *split_rec, | |
4842 | u32 cpos, | |
4843 | struct ocfs2_extent_rec *rec) | |
4844 | { | |
4845 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4846 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4847 | ||
4848 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4849 | ||
4850 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4851 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4852 | ||
4853 | split_rec->e_blkno = rec->e_blkno; | |
4854 | le64_add_cpu(&split_rec->e_blkno, | |
4855 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4856 | ||
4857 | split_rec->e_flags = rec->e_flags; | |
4858 | } | |
4859 | ||
4860 | static int ocfs2_split_and_insert(struct inode *inode, | |
4861 | handle_t *handle, | |
4862 | struct ocfs2_path *path, | |
e7d4cb6b | 4863 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4864 | struct buffer_head **last_eb_bh, |
4865 | int split_index, | |
4866 | struct ocfs2_extent_rec *orig_split_rec, | |
4867 | struct ocfs2_alloc_context *meta_ac) | |
4868 | { | |
4869 | int ret = 0, depth; | |
4870 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4871 | struct ocfs2_extent_rec tmprec; | |
4872 | struct ocfs2_extent_list *rightmost_el; | |
4873 | struct ocfs2_extent_rec rec; | |
4874 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4875 | struct ocfs2_insert_type insert; | |
4876 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4877 | |
4878 | leftright: | |
4879 | /* | |
4880 | * Store a copy of the record on the stack - it might move | |
4881 | * around as the tree is manipulated below. | |
4882 | */ | |
4883 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4884 | ||
ce1d9ea6 | 4885 | rightmost_el = et->et_root_el; |
328d5752 MF |
4886 | |
4887 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4888 | if (depth) { | |
4889 | BUG_ON(!(*last_eb_bh)); | |
4890 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4891 | rightmost_el = &eb->h_list; | |
4892 | } | |
4893 | ||
4894 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4895 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 4896 | ret = ocfs2_grow_tree(handle, et, |
e7d4cb6b | 4897 | &depth, last_eb_bh, meta_ac); |
328d5752 MF |
4898 | if (ret) { |
4899 | mlog_errno(ret); | |
4900 | goto out; | |
4901 | } | |
328d5752 MF |
4902 | } |
4903 | ||
4904 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4905 | insert.ins_appending = APPEND_NONE; | |
4906 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4907 | insert.ins_tree_depth = depth; |
4908 | ||
4909 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4910 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4911 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4912 | le16_to_cpu(rec.e_leaf_clusters); | |
4913 | ||
4914 | if (split_rec.e_cpos == rec.e_cpos) { | |
4915 | insert.ins_split = SPLIT_LEFT; | |
4916 | } else if (insert_range == rec_range) { | |
4917 | insert.ins_split = SPLIT_RIGHT; | |
4918 | } else { | |
4919 | /* | |
4920 | * Left/right split. We fake this as a right split | |
4921 | * first and then make a second pass as a left split. | |
4922 | */ | |
4923 | insert.ins_split = SPLIT_RIGHT; | |
4924 | ||
4925 | ocfs2_make_right_split_rec(inode->i_sb, &tmprec, insert_range, | |
4926 | &rec); | |
4927 | ||
4928 | split_rec = tmprec; | |
4929 | ||
4930 | BUG_ON(do_leftright); | |
4931 | do_leftright = 1; | |
4932 | } | |
4933 | ||
3505bec0 | 4934 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
328d5752 MF |
4935 | if (ret) { |
4936 | mlog_errno(ret); | |
4937 | goto out; | |
4938 | } | |
4939 | ||
4940 | if (do_leftright == 1) { | |
4941 | u32 cpos; | |
4942 | struct ocfs2_extent_list *el; | |
4943 | ||
4944 | do_leftright++; | |
4945 | split_rec = *orig_split_rec; | |
4946 | ||
4947 | ocfs2_reinit_path(path, 1); | |
4948 | ||
4949 | cpos = le32_to_cpu(split_rec.e_cpos); | |
facdb77f | 4950 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
328d5752 MF |
4951 | if (ret) { |
4952 | mlog_errno(ret); | |
4953 | goto out; | |
4954 | } | |
4955 | ||
4956 | el = path_leaf_el(path); | |
4957 | split_index = ocfs2_search_extent_list(el, cpos); | |
4958 | goto leftright; | |
4959 | } | |
4960 | out: | |
4961 | ||
4962 | return ret; | |
4963 | } | |
4964 | ||
47be12e4 TM |
4965 | static int ocfs2_replace_extent_rec(struct inode *inode, |
4966 | handle_t *handle, | |
4967 | struct ocfs2_path *path, | |
4968 | struct ocfs2_extent_list *el, | |
4969 | int split_index, | |
4970 | struct ocfs2_extent_rec *split_rec) | |
4971 | { | |
4972 | int ret; | |
4973 | ||
0cf2f763 | 4974 | ret = ocfs2_path_bh_journal_access(handle, INODE_CACHE(inode), path, |
47be12e4 TM |
4975 | path_num_items(path) - 1); |
4976 | if (ret) { | |
4977 | mlog_errno(ret); | |
4978 | goto out; | |
4979 | } | |
4980 | ||
4981 | el->l_recs[split_index] = *split_rec; | |
4982 | ||
4983 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
4984 | out: | |
4985 | return ret; | |
4986 | } | |
4987 | ||
328d5752 MF |
4988 | /* |
4989 | * Mark part or all of the extent record at split_index in the leaf | |
4990 | * pointed to by path as written. This removes the unwritten | |
4991 | * extent flag. | |
4992 | * | |
4993 | * Care is taken to handle contiguousness so as to not grow the tree. | |
4994 | * | |
4995 | * meta_ac is not strictly necessary - we only truly need it if growth | |
4996 | * of the tree is required. All other cases will degrade into a less | |
4997 | * optimal tree layout. | |
4998 | * | |
e7d4cb6b TM |
4999 | * last_eb_bh should be the rightmost leaf block for any extent |
5000 | * btree. Since a split may grow the tree or a merge might shrink it, | |
5001 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
5002 | * |
5003 | * This code is optimized for readability - several passes might be | |
5004 | * made over certain portions of the tree. All of those blocks will | |
5005 | * have been brought into cache (and pinned via the journal), so the | |
5006 | * extra overhead is not expressed in terms of disk reads. | |
5007 | */ | |
5008 | static int __ocfs2_mark_extent_written(struct inode *inode, | |
e7d4cb6b | 5009 | struct ocfs2_extent_tree *et, |
328d5752 MF |
5010 | handle_t *handle, |
5011 | struct ocfs2_path *path, | |
5012 | int split_index, | |
5013 | struct ocfs2_extent_rec *split_rec, | |
5014 | struct ocfs2_alloc_context *meta_ac, | |
5015 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5016 | { | |
5017 | int ret = 0; | |
5018 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 5019 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
5020 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
5021 | struct ocfs2_merge_ctxt ctxt; | |
5022 | struct ocfs2_extent_list *rightmost_el; | |
5023 | ||
3cf0c507 | 5024 | if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) { |
328d5752 MF |
5025 | ret = -EIO; |
5026 | mlog_errno(ret); | |
5027 | goto out; | |
5028 | } | |
5029 | ||
5030 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || | |
5031 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
5032 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
5033 | ret = -EIO; | |
5034 | mlog_errno(ret); | |
5035 | goto out; | |
5036 | } | |
5037 | ||
a2970291 | 5038 | ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el, |
328d5752 MF |
5039 | split_index, |
5040 | split_rec); | |
5041 | ||
5042 | /* | |
5043 | * The core merge / split code wants to know how much room is | |
5044 | * left in this inodes allocation tree, so we pass the | |
5045 | * rightmost extent list. | |
5046 | */ | |
5047 | if (path->p_tree_depth) { | |
5048 | struct ocfs2_extent_block *eb; | |
328d5752 | 5049 | |
3d03a305 | 5050 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5051 | ocfs2_et_get_last_eb_blk(et), |
5052 | &last_eb_bh); | |
328d5752 MF |
5053 | if (ret) { |
5054 | mlog_exit(ret); | |
5055 | goto out; | |
5056 | } | |
5057 | ||
5058 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
328d5752 MF |
5059 | rightmost_el = &eb->h_list; |
5060 | } else | |
5061 | rightmost_el = path_root_el(path); | |
5062 | ||
328d5752 MF |
5063 | if (rec->e_cpos == split_rec->e_cpos && |
5064 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
5065 | ctxt.c_split_covers_rec = 1; | |
5066 | else | |
5067 | ctxt.c_split_covers_rec = 0; | |
5068 | ||
5069 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
5070 | ||
015452b1 MF |
5071 | mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n", |
5072 | split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent, | |
5073 | ctxt.c_split_covers_rec); | |
328d5752 MF |
5074 | |
5075 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
5076 | if (ctxt.c_split_covers_rec) | |
47be12e4 TM |
5077 | ret = ocfs2_replace_extent_rec(inode, handle, |
5078 | path, el, | |
5079 | split_index, split_rec); | |
328d5752 | 5080 | else |
e7d4cb6b | 5081 | ret = ocfs2_split_and_insert(inode, handle, path, et, |
328d5752 MF |
5082 | &last_eb_bh, split_index, |
5083 | split_rec, meta_ac); | |
5084 | if (ret) | |
5085 | mlog_errno(ret); | |
5086 | } else { | |
c495dd24 | 5087 | ret = ocfs2_try_to_merge_extent(handle, et, path, |
328d5752 | 5088 | split_index, split_rec, |
c495dd24 | 5089 | dealloc, &ctxt); |
328d5752 MF |
5090 | if (ret) |
5091 | mlog_errno(ret); | |
5092 | } | |
5093 | ||
328d5752 MF |
5094 | out: |
5095 | brelse(last_eb_bh); | |
5096 | return ret; | |
5097 | } | |
5098 | ||
5099 | /* | |
5100 | * Mark the already-existing extent at cpos as written for len clusters. | |
5101 | * | |
5102 | * If the existing extent is larger than the request, initiate a | |
5103 | * split. An attempt will be made at merging with adjacent extents. | |
5104 | * | |
5105 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5106 | */ | |
f99b9b7c JB |
5107 | int ocfs2_mark_extent_written(struct inode *inode, |
5108 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
5109 | handle_t *handle, u32 cpos, u32 len, u32 phys, |
5110 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 5111 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
5112 | { |
5113 | int ret, index; | |
5114 | u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys); | |
5115 | struct ocfs2_extent_rec split_rec; | |
5116 | struct ocfs2_path *left_path = NULL; | |
5117 | struct ocfs2_extent_list *el; | |
5118 | ||
5119 | mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n", | |
5120 | inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno); | |
5121 | ||
5122 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { | |
5123 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents " | |
5124 | "that are being written to, but the feature bit " | |
5125 | "is not set in the super block.", | |
5126 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
5127 | ret = -EROFS; | |
5128 | goto out; | |
5129 | } | |
5130 | ||
5131 | /* | |
5132 | * XXX: This should be fixed up so that we just re-insert the | |
5133 | * next extent records. | |
5134 | */ | |
4c911eef | 5135 | ocfs2_et_extent_map_truncate(et, 0); |
328d5752 | 5136 | |
ffdd7a54 | 5137 | left_path = ocfs2_new_path_from_et(et); |
328d5752 MF |
5138 | if (!left_path) { |
5139 | ret = -ENOMEM; | |
5140 | mlog_errno(ret); | |
5141 | goto out; | |
5142 | } | |
5143 | ||
facdb77f | 5144 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
5145 | if (ret) { |
5146 | mlog_errno(ret); | |
5147 | goto out; | |
5148 | } | |
5149 | el = path_leaf_el(left_path); | |
5150 | ||
5151 | index = ocfs2_search_extent_list(el, cpos); | |
5152 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5153 | ocfs2_error(inode->i_sb, | |
5154 | "Inode %llu has an extent at cpos %u which can no " | |
5155 | "longer be found.\n", | |
5156 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
5157 | ret = -EROFS; | |
5158 | goto out; | |
5159 | } | |
5160 | ||
5161 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
5162 | split_rec.e_cpos = cpu_to_le32(cpos); | |
5163 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
5164 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
5165 | split_rec.e_flags = path_leaf_el(left_path)->l_recs[index].e_flags; | |
5166 | split_rec.e_flags &= ~OCFS2_EXT_UNWRITTEN; | |
5167 | ||
f99b9b7c | 5168 | ret = __ocfs2_mark_extent_written(inode, et, handle, left_path, |
e7d4cb6b TM |
5169 | index, &split_rec, meta_ac, |
5170 | dealloc); | |
328d5752 MF |
5171 | if (ret) |
5172 | mlog_errno(ret); | |
5173 | ||
5174 | out: | |
5175 | ocfs2_free_path(left_path); | |
5176 | return ret; | |
5177 | } | |
5178 | ||
e7d4cb6b | 5179 | static int ocfs2_split_tree(struct inode *inode, struct ocfs2_extent_tree *et, |
d0c7d708 MF |
5180 | handle_t *handle, struct ocfs2_path *path, |
5181 | int index, u32 new_range, | |
5182 | struct ocfs2_alloc_context *meta_ac) | |
5183 | { | |
5184 | int ret, depth, credits = handle->h_buffer_credits; | |
d0c7d708 MF |
5185 | struct buffer_head *last_eb_bh = NULL; |
5186 | struct ocfs2_extent_block *eb; | |
5187 | struct ocfs2_extent_list *rightmost_el, *el; | |
5188 | struct ocfs2_extent_rec split_rec; | |
5189 | struct ocfs2_extent_rec *rec; | |
5190 | struct ocfs2_insert_type insert; | |
5191 | ||
5192 | /* | |
5193 | * Setup the record to split before we grow the tree. | |
5194 | */ | |
5195 | el = path_leaf_el(path); | |
5196 | rec = &el->l_recs[index]; | |
5197 | ocfs2_make_right_split_rec(inode->i_sb, &split_rec, new_range, rec); | |
5198 | ||
5199 | depth = path->p_tree_depth; | |
5200 | if (depth > 0) { | |
3d03a305 | 5201 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5202 | ocfs2_et_get_last_eb_blk(et), |
5203 | &last_eb_bh); | |
d0c7d708 MF |
5204 | if (ret < 0) { |
5205 | mlog_errno(ret); | |
5206 | goto out; | |
5207 | } | |
5208 | ||
5209 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
5210 | rightmost_el = &eb->h_list; | |
5211 | } else | |
5212 | rightmost_el = path_leaf_el(path); | |
5213 | ||
811f933d | 5214 | credits += path->p_tree_depth + |
ce1d9ea6 | 5215 | ocfs2_extend_meta_needed(et->et_root_el); |
d0c7d708 MF |
5216 | ret = ocfs2_extend_trans(handle, credits); |
5217 | if (ret) { | |
5218 | mlog_errno(ret); | |
5219 | goto out; | |
5220 | } | |
5221 | ||
5222 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
5223 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 5224 | ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
5225 | meta_ac); |
5226 | if (ret) { | |
5227 | mlog_errno(ret); | |
5228 | goto out; | |
5229 | } | |
d0c7d708 MF |
5230 | } |
5231 | ||
5232 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
5233 | insert.ins_appending = APPEND_NONE; | |
5234 | insert.ins_contig = CONTIG_NONE; | |
5235 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
5236 | insert.ins_tree_depth = depth; |
5237 | ||
3505bec0 | 5238 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
d0c7d708 MF |
5239 | if (ret) |
5240 | mlog_errno(ret); | |
5241 | ||
5242 | out: | |
5243 | brelse(last_eb_bh); | |
5244 | return ret; | |
5245 | } | |
5246 | ||
043beebb JB |
5247 | static int ocfs2_truncate_rec(handle_t *handle, |
5248 | struct ocfs2_extent_tree *et, | |
d0c7d708 MF |
5249 | struct ocfs2_path *path, int index, |
5250 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
043beebb | 5251 | u32 cpos, u32 len) |
d0c7d708 MF |
5252 | { |
5253 | int ret; | |
5254 | u32 left_cpos, rec_range, trunc_range; | |
5255 | int wants_rotate = 0, is_rightmost_tree_rec = 0; | |
043beebb | 5256 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
d0c7d708 MF |
5257 | struct ocfs2_path *left_path = NULL; |
5258 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
5259 | struct ocfs2_extent_rec *rec; | |
5260 | struct ocfs2_extent_block *eb; | |
5261 | ||
5262 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
70f18c08 | 5263 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5264 | if (ret) { |
5265 | mlog_errno(ret); | |
5266 | goto out; | |
5267 | } | |
5268 | ||
5269 | index--; | |
5270 | } | |
5271 | ||
5272 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
5273 | path->p_tree_depth) { | |
5274 | /* | |
5275 | * Check whether this is the rightmost tree record. If | |
5276 | * we remove all of this record or part of its right | |
5277 | * edge then an update of the record lengths above it | |
5278 | * will be required. | |
5279 | */ | |
5280 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
5281 | if (eb->h_next_leaf_blk == 0) | |
5282 | is_rightmost_tree_rec = 1; | |
5283 | } | |
5284 | ||
5285 | rec = &el->l_recs[index]; | |
5286 | if (index == 0 && path->p_tree_depth && | |
5287 | le32_to_cpu(rec->e_cpos) == cpos) { | |
5288 | /* | |
5289 | * Changing the leftmost offset (via partial or whole | |
5290 | * record truncate) of an interior (or rightmost) path | |
5291 | * means we have to update the subtree that is formed | |
5292 | * by this leaf and the one to it's left. | |
5293 | * | |
5294 | * There are two cases we can skip: | |
043beebb | 5295 | * 1) Path is the leftmost one in our btree. |
d0c7d708 MF |
5296 | * 2) The leaf is rightmost and will be empty after |
5297 | * we remove the extent record - the rotate code | |
5298 | * knows how to update the newly formed edge. | |
5299 | */ | |
5300 | ||
043beebb | 5301 | ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
d0c7d708 MF |
5302 | if (ret) { |
5303 | mlog_errno(ret); | |
5304 | goto out; | |
5305 | } | |
5306 | ||
5307 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
ffdd7a54 | 5308 | left_path = ocfs2_new_path_from_path(path); |
d0c7d708 MF |
5309 | if (!left_path) { |
5310 | ret = -ENOMEM; | |
5311 | mlog_errno(ret); | |
5312 | goto out; | |
5313 | } | |
5314 | ||
facdb77f JB |
5315 | ret = ocfs2_find_path(et->et_ci, left_path, |
5316 | left_cpos); | |
d0c7d708 MF |
5317 | if (ret) { |
5318 | mlog_errno(ret); | |
5319 | goto out; | |
5320 | } | |
5321 | } | |
5322 | } | |
5323 | ||
5324 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5325 | handle->h_buffer_credits, | |
5326 | path); | |
5327 | if (ret) { | |
5328 | mlog_errno(ret); | |
5329 | goto out; | |
5330 | } | |
5331 | ||
d9a0a1f8 | 5332 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
d0c7d708 MF |
5333 | if (ret) { |
5334 | mlog_errno(ret); | |
5335 | goto out; | |
5336 | } | |
5337 | ||
d9a0a1f8 | 5338 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
d0c7d708 MF |
5339 | if (ret) { |
5340 | mlog_errno(ret); | |
5341 | goto out; | |
5342 | } | |
5343 | ||
5344 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5345 | trunc_range = cpos + len; | |
5346 | ||
5347 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5348 | int next_free; | |
5349 | ||
5350 | memset(rec, 0, sizeof(*rec)); | |
5351 | ocfs2_cleanup_merge(el, index); | |
5352 | wants_rotate = 1; | |
5353 | ||
5354 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5355 | if (is_rightmost_tree_rec && next_free > 1) { | |
5356 | /* | |
5357 | * We skip the edge update if this path will | |
5358 | * be deleted by the rotate code. | |
5359 | */ | |
5360 | rec = &el->l_recs[next_free - 1]; | |
d401dc12 | 5361 | ocfs2_adjust_rightmost_records(handle, et, path, |
d0c7d708 MF |
5362 | rec); |
5363 | } | |
5364 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5365 | /* Remove leftmost portion of the record. */ | |
5366 | le32_add_cpu(&rec->e_cpos, len); | |
5367 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5368 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5369 | } else if (rec_range == trunc_range) { | |
5370 | /* Remove rightmost portion of the record */ | |
5371 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5372 | if (is_rightmost_tree_rec) | |
d401dc12 | 5373 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
d0c7d708 MF |
5374 | } else { |
5375 | /* Caller should have trapped this. */ | |
043beebb JB |
5376 | mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) " |
5377 | "(%u, %u)\n", | |
5378 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
d0c7d708 MF |
5379 | le32_to_cpu(rec->e_cpos), |
5380 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5381 | BUG(); | |
5382 | } | |
5383 | ||
5384 | if (left_path) { | |
5385 | int subtree_index; | |
5386 | ||
7dc02805 | 5387 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
4619c73e | 5388 | ocfs2_complete_edge_insert(handle, left_path, path, |
d0c7d708 MF |
5389 | subtree_index); |
5390 | } | |
5391 | ||
5392 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5393 | ||
70f18c08 | 5394 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5395 | if (ret) { |
5396 | mlog_errno(ret); | |
5397 | goto out; | |
5398 | } | |
5399 | ||
5400 | out: | |
5401 | ocfs2_free_path(left_path); | |
5402 | return ret; | |
5403 | } | |
5404 | ||
f99b9b7c JB |
5405 | int ocfs2_remove_extent(struct inode *inode, |
5406 | struct ocfs2_extent_tree *et, | |
063c4561 MF |
5407 | u32 cpos, u32 len, handle_t *handle, |
5408 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 5409 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
d0c7d708 MF |
5410 | { |
5411 | int ret, index; | |
5412 | u32 rec_range, trunc_range; | |
5413 | struct ocfs2_extent_rec *rec; | |
5414 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 5415 | struct ocfs2_path *path = NULL; |
d0c7d708 | 5416 | |
4c911eef JB |
5417 | /* |
5418 | * XXX: Why are we truncating to 0 instead of wherever this | |
5419 | * affects us? | |
5420 | */ | |
5421 | ocfs2_et_extent_map_truncate(et, 0); | |
d0c7d708 | 5422 | |
ffdd7a54 | 5423 | path = ocfs2_new_path_from_et(et); |
d0c7d708 MF |
5424 | if (!path) { |
5425 | ret = -ENOMEM; | |
5426 | mlog_errno(ret); | |
5427 | goto out; | |
5428 | } | |
5429 | ||
facdb77f | 5430 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5431 | if (ret) { |
5432 | mlog_errno(ret); | |
5433 | goto out; | |
5434 | } | |
5435 | ||
5436 | el = path_leaf_el(path); | |
5437 | index = ocfs2_search_extent_list(el, cpos); | |
5438 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5439 | ocfs2_error(inode->i_sb, | |
5440 | "Inode %llu has an extent at cpos %u which can no " | |
5441 | "longer be found.\n", | |
5442 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
5443 | ret = -EROFS; | |
5444 | goto out; | |
5445 | } | |
5446 | ||
5447 | /* | |
5448 | * We have 3 cases of extent removal: | |
5449 | * 1) Range covers the entire extent rec | |
5450 | * 2) Range begins or ends on one edge of the extent rec | |
5451 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5452 | * | |
5453 | * For case 1 we remove the extent rec and left rotate to | |
5454 | * fill the hole. | |
5455 | * | |
5456 | * For case 2 we just shrink the existing extent rec, with a | |
5457 | * tree update if the shrinking edge is also the edge of an | |
5458 | * extent block. | |
5459 | * | |
5460 | * For case 3 we do a right split to turn the extent rec into | |
5461 | * something case 2 can handle. | |
5462 | */ | |
5463 | rec = &el->l_recs[index]; | |
5464 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5465 | trunc_range = cpos + len; | |
5466 | ||
5467 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5468 | ||
5469 | mlog(0, "Inode %llu, remove (cpos %u, len %u). Existing index %d " | |
5470 | "(cpos %u, len %u)\n", | |
5471 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, len, index, | |
5472 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec)); | |
5473 | ||
5474 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
043beebb JB |
5475 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5476 | cpos, len); | |
d0c7d708 MF |
5477 | if (ret) { |
5478 | mlog_errno(ret); | |
5479 | goto out; | |
5480 | } | |
5481 | } else { | |
f99b9b7c | 5482 | ret = ocfs2_split_tree(inode, et, handle, path, index, |
d0c7d708 MF |
5483 | trunc_range, meta_ac); |
5484 | if (ret) { | |
5485 | mlog_errno(ret); | |
5486 | goto out; | |
5487 | } | |
5488 | ||
5489 | /* | |
5490 | * The split could have manipulated the tree enough to | |
5491 | * move the record location, so we have to look for it again. | |
5492 | */ | |
5493 | ocfs2_reinit_path(path, 1); | |
5494 | ||
facdb77f | 5495 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5496 | if (ret) { |
5497 | mlog_errno(ret); | |
5498 | goto out; | |
5499 | } | |
5500 | ||
5501 | el = path_leaf_el(path); | |
5502 | index = ocfs2_search_extent_list(el, cpos); | |
5503 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5504 | ocfs2_error(inode->i_sb, | |
5505 | "Inode %llu: split at cpos %u lost record.", | |
5506 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5507 | cpos); | |
5508 | ret = -EROFS; | |
5509 | goto out; | |
5510 | } | |
5511 | ||
5512 | /* | |
5513 | * Double check our values here. If anything is fishy, | |
5514 | * it's easier to catch it at the top level. | |
5515 | */ | |
5516 | rec = &el->l_recs[index]; | |
5517 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5518 | ocfs2_rec_clusters(el, rec); | |
5519 | if (rec_range != trunc_range) { | |
5520 | ocfs2_error(inode->i_sb, | |
5521 | "Inode %llu: error after split at cpos %u" | |
5522 | "trunc len %u, existing record is (%u,%u)", | |
5523 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5524 | cpos, len, le32_to_cpu(rec->e_cpos), | |
5525 | ocfs2_rec_clusters(el, rec)); | |
5526 | ret = -EROFS; | |
5527 | goto out; | |
5528 | } | |
5529 | ||
043beebb JB |
5530 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5531 | cpos, len); | |
d0c7d708 MF |
5532 | if (ret) { |
5533 | mlog_errno(ret); | |
5534 | goto out; | |
5535 | } | |
5536 | } | |
5537 | ||
5538 | out: | |
5539 | ocfs2_free_path(path); | |
5540 | return ret; | |
5541 | } | |
5542 | ||
fecc0112 MF |
5543 | int ocfs2_remove_btree_range(struct inode *inode, |
5544 | struct ocfs2_extent_tree *et, | |
5545 | u32 cpos, u32 phys_cpos, u32 len, | |
5546 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5547 | { | |
5548 | int ret; | |
5549 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
5550 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5551 | struct inode *tl_inode = osb->osb_tl_inode; | |
5552 | handle_t *handle; | |
5553 | struct ocfs2_alloc_context *meta_ac = NULL; | |
5554 | ||
5555 | ret = ocfs2_lock_allocators(inode, et, 0, 1, NULL, &meta_ac); | |
5556 | if (ret) { | |
5557 | mlog_errno(ret); | |
5558 | return ret; | |
5559 | } | |
5560 | ||
5561 | mutex_lock(&tl_inode->i_mutex); | |
5562 | ||
5563 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5564 | ret = __ocfs2_flush_truncate_log(osb); | |
5565 | if (ret < 0) { | |
5566 | mlog_errno(ret); | |
5567 | goto out; | |
5568 | } | |
5569 | } | |
5570 | ||
a90714c1 | 5571 | handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb)); |
fecc0112 MF |
5572 | if (IS_ERR(handle)) { |
5573 | ret = PTR_ERR(handle); | |
5574 | mlog_errno(ret); | |
5575 | goto out; | |
5576 | } | |
5577 | ||
d9a0a1f8 | 5578 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 5579 | OCFS2_JOURNAL_ACCESS_WRITE); |
fecc0112 MF |
5580 | if (ret) { |
5581 | mlog_errno(ret); | |
5582 | goto out; | |
5583 | } | |
5584 | ||
fd4ef231 MF |
5585 | vfs_dq_free_space_nodirty(inode, |
5586 | ocfs2_clusters_to_bytes(inode->i_sb, len)); | |
5587 | ||
fecc0112 MF |
5588 | ret = ocfs2_remove_extent(inode, et, cpos, len, handle, meta_ac, |
5589 | dealloc); | |
5590 | if (ret) { | |
5591 | mlog_errno(ret); | |
5592 | goto out_commit; | |
5593 | } | |
5594 | ||
6136ca5f | 5595 | ocfs2_et_update_clusters(et, -len); |
fecc0112 MF |
5596 | |
5597 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); | |
5598 | if (ret) { | |
5599 | mlog_errno(ret); | |
5600 | goto out_commit; | |
5601 | } | |
5602 | ||
5603 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); | |
5604 | if (ret) | |
5605 | mlog_errno(ret); | |
5606 | ||
5607 | out_commit: | |
5608 | ocfs2_commit_trans(osb, handle); | |
5609 | out: | |
5610 | mutex_unlock(&tl_inode->i_mutex); | |
5611 | ||
5612 | if (meta_ac) | |
5613 | ocfs2_free_alloc_context(meta_ac); | |
5614 | ||
5615 | return ret; | |
5616 | } | |
5617 | ||
063c4561 | 5618 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5619 | { |
5620 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5621 | struct ocfs2_dinode *di; | |
5622 | struct ocfs2_truncate_log *tl; | |
5623 | ||
5624 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5625 | tl = &di->id2.i_dealloc; | |
5626 | ||
5627 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5628 | "slot %d, invalid truncate log parameters: used = " | |
5629 | "%u, count = %u\n", osb->slot_num, | |
5630 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5631 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5632 | } | |
5633 | ||
5634 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5635 | unsigned int new_start) | |
5636 | { | |
5637 | unsigned int tail_index; | |
5638 | unsigned int current_tail; | |
5639 | ||
5640 | /* No records, nothing to coalesce */ | |
5641 | if (!le16_to_cpu(tl->tl_used)) | |
5642 | return 0; | |
5643 | ||
5644 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5645 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5646 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5647 | ||
5648 | return current_tail == new_start; | |
5649 | } | |
5650 | ||
063c4561 MF |
5651 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5652 | handle_t *handle, | |
5653 | u64 start_blk, | |
5654 | unsigned int num_clusters) | |
ccd979bd MF |
5655 | { |
5656 | int status, index; | |
5657 | unsigned int start_cluster, tl_count; | |
5658 | struct inode *tl_inode = osb->osb_tl_inode; | |
5659 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5660 | struct ocfs2_dinode *di; | |
5661 | struct ocfs2_truncate_log *tl; | |
5662 | ||
b0697053 MF |
5663 | mlog_entry("start_blk = %llu, num_clusters = %u\n", |
5664 | (unsigned long long)start_blk, num_clusters); | |
ccd979bd | 5665 | |
1b1dcc1b | 5666 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5667 | |
5668 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5669 | ||
5670 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5671 | |
10995aa2 JB |
5672 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5673 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5674 | * corruption is a code bug */ | |
5675 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5676 | ||
5677 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5678 | tl_count = le16_to_cpu(tl->tl_count); |
5679 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5680 | tl_count == 0, | |
b0697053 MF |
5681 | "Truncate record count on #%llu invalid " |
5682 | "wanted %u, actual %u\n", | |
5683 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5684 | ocfs2_truncate_recs_per_inode(osb->sb), |
5685 | le16_to_cpu(tl->tl_count)); | |
5686 | ||
5687 | /* Caller should have known to flush before calling us. */ | |
5688 | index = le16_to_cpu(tl->tl_used); | |
5689 | if (index >= tl_count) { | |
5690 | status = -ENOSPC; | |
5691 | mlog_errno(status); | |
5692 | goto bail; | |
5693 | } | |
5694 | ||
0cf2f763 | 5695 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5696 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5697 | if (status < 0) { |
5698 | mlog_errno(status); | |
5699 | goto bail; | |
5700 | } | |
5701 | ||
5702 | mlog(0, "Log truncate of %u clusters starting at cluster %u to " | |
b0697053 MF |
5703 | "%llu (index = %d)\n", num_clusters, start_cluster, |
5704 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index); | |
ccd979bd MF |
5705 | |
5706 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { | |
5707 | /* | |
5708 | * Move index back to the record we are coalescing with. | |
5709 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5710 | */ | |
5711 | index--; | |
5712 | ||
5713 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
5714 | mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n", | |
5715 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5716 | num_clusters); | |
5717 | } else { | |
5718 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5719 | tl->tl_used = cpu_to_le16(index + 1); | |
5720 | } | |
5721 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5722 | ||
5723 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5724 | if (status < 0) { | |
5725 | mlog_errno(status); | |
5726 | goto bail; | |
5727 | } | |
5728 | ||
5729 | bail: | |
5730 | mlog_exit(status); | |
5731 | return status; | |
5732 | } | |
5733 | ||
5734 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
1fabe148 | 5735 | handle_t *handle, |
ccd979bd MF |
5736 | struct inode *data_alloc_inode, |
5737 | struct buffer_head *data_alloc_bh) | |
5738 | { | |
5739 | int status = 0; | |
5740 | int i; | |
5741 | unsigned int num_clusters; | |
5742 | u64 start_blk; | |
5743 | struct ocfs2_truncate_rec rec; | |
5744 | struct ocfs2_dinode *di; | |
5745 | struct ocfs2_truncate_log *tl; | |
5746 | struct inode *tl_inode = osb->osb_tl_inode; | |
5747 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5748 | ||
5749 | mlog_entry_void(); | |
5750 | ||
5751 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5752 | tl = &di->id2.i_dealloc; | |
5753 | i = le16_to_cpu(tl->tl_used) - 1; | |
5754 | while (i >= 0) { | |
5755 | /* Caller has given us at least enough credits to | |
5756 | * update the truncate log dinode */ | |
0cf2f763 | 5757 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5758 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5759 | if (status < 0) { |
5760 | mlog_errno(status); | |
5761 | goto bail; | |
5762 | } | |
5763 | ||
5764 | tl->tl_used = cpu_to_le16(i); | |
5765 | ||
5766 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5767 | if (status < 0) { | |
5768 | mlog_errno(status); | |
5769 | goto bail; | |
5770 | } | |
5771 | ||
5772 | /* TODO: Perhaps we can calculate the bulk of the | |
5773 | * credits up front rather than extending like | |
5774 | * this. */ | |
5775 | status = ocfs2_extend_trans(handle, | |
5776 | OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); | |
5777 | if (status < 0) { | |
5778 | mlog_errno(status); | |
5779 | goto bail; | |
5780 | } | |
5781 | ||
5782 | rec = tl->tl_recs[i]; | |
5783 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5784 | le32_to_cpu(rec.t_start)); | |
5785 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5786 | ||
5787 | /* if start_blk is not set, we ignore the record as | |
5788 | * invalid. */ | |
5789 | if (start_blk) { | |
5790 | mlog(0, "free record %d, start = %u, clusters = %u\n", | |
5791 | i, le32_to_cpu(rec.t_start), num_clusters); | |
5792 | ||
5793 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5794 | data_alloc_bh, start_blk, | |
5795 | num_clusters); | |
5796 | if (status < 0) { | |
5797 | mlog_errno(status); | |
5798 | goto bail; | |
5799 | } | |
5800 | } | |
5801 | i--; | |
5802 | } | |
5803 | ||
5804 | bail: | |
5805 | mlog_exit(status); | |
5806 | return status; | |
5807 | } | |
5808 | ||
1b1dcc1b | 5809 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5810 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5811 | { |
5812 | int status; | |
5813 | unsigned int num_to_flush; | |
1fabe148 | 5814 | handle_t *handle; |
ccd979bd MF |
5815 | struct inode *tl_inode = osb->osb_tl_inode; |
5816 | struct inode *data_alloc_inode = NULL; | |
5817 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5818 | struct buffer_head *data_alloc_bh = NULL; | |
5819 | struct ocfs2_dinode *di; | |
5820 | struct ocfs2_truncate_log *tl; | |
5821 | ||
5822 | mlog_entry_void(); | |
5823 | ||
1b1dcc1b | 5824 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5825 | |
5826 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5827 | |
10995aa2 JB |
5828 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5829 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5830 | * corruption is a code bug */ | |
5831 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5832 | ||
5833 | tl = &di->id2.i_dealloc; | |
ccd979bd | 5834 | num_to_flush = le16_to_cpu(tl->tl_used); |
b0697053 MF |
5835 | mlog(0, "Flush %u records from truncate log #%llu\n", |
5836 | num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno); | |
ccd979bd MF |
5837 | if (!num_to_flush) { |
5838 | status = 0; | |
e08dc8b9 | 5839 | goto out; |
ccd979bd MF |
5840 | } |
5841 | ||
5842 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
5843 | GLOBAL_BITMAP_SYSTEM_INODE, | |
5844 | OCFS2_INVALID_SLOT); | |
5845 | if (!data_alloc_inode) { | |
5846 | status = -EINVAL; | |
5847 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 5848 | goto out; |
ccd979bd MF |
5849 | } |
5850 | ||
e08dc8b9 MF |
5851 | mutex_lock(&data_alloc_inode->i_mutex); |
5852 | ||
e63aecb6 | 5853 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
5854 | if (status < 0) { |
5855 | mlog_errno(status); | |
e08dc8b9 | 5856 | goto out_mutex; |
ccd979bd MF |
5857 | } |
5858 | ||
65eff9cc | 5859 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5860 | if (IS_ERR(handle)) { |
5861 | status = PTR_ERR(handle); | |
ccd979bd | 5862 | mlog_errno(status); |
e08dc8b9 | 5863 | goto out_unlock; |
ccd979bd MF |
5864 | } |
5865 | ||
5866 | status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode, | |
5867 | data_alloc_bh); | |
e08dc8b9 | 5868 | if (status < 0) |
ccd979bd | 5869 | mlog_errno(status); |
ccd979bd | 5870 | |
02dc1af4 | 5871 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 5872 | |
e08dc8b9 MF |
5873 | out_unlock: |
5874 | brelse(data_alloc_bh); | |
e63aecb6 | 5875 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 5876 | |
e08dc8b9 MF |
5877 | out_mutex: |
5878 | mutex_unlock(&data_alloc_inode->i_mutex); | |
5879 | iput(data_alloc_inode); | |
ccd979bd | 5880 | |
e08dc8b9 | 5881 | out: |
ccd979bd MF |
5882 | mlog_exit(status); |
5883 | return status; | |
5884 | } | |
5885 | ||
5886 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
5887 | { | |
5888 | int status; | |
5889 | struct inode *tl_inode = osb->osb_tl_inode; | |
5890 | ||
1b1dcc1b | 5891 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd | 5892 | status = __ocfs2_flush_truncate_log(osb); |
1b1dcc1b | 5893 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5894 | |
5895 | return status; | |
5896 | } | |
5897 | ||
c4028958 | 5898 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
5899 | { |
5900 | int status; | |
c4028958 DH |
5901 | struct ocfs2_super *osb = |
5902 | container_of(work, struct ocfs2_super, | |
5903 | osb_truncate_log_wq.work); | |
ccd979bd MF |
5904 | |
5905 | mlog_entry_void(); | |
5906 | ||
5907 | status = ocfs2_flush_truncate_log(osb); | |
5908 | if (status < 0) | |
5909 | mlog_errno(status); | |
4d0ddb2c TM |
5910 | else |
5911 | ocfs2_init_inode_steal_slot(osb); | |
ccd979bd MF |
5912 | |
5913 | mlog_exit(status); | |
5914 | } | |
5915 | ||
5916 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
5917 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
5918 | int cancel) | |
5919 | { | |
5920 | if (osb->osb_tl_inode) { | |
5921 | /* We want to push off log flushes while truncates are | |
5922 | * still running. */ | |
5923 | if (cancel) | |
5924 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5925 | ||
5926 | queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq, | |
5927 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); | |
5928 | } | |
5929 | } | |
5930 | ||
5931 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, | |
5932 | int slot_num, | |
5933 | struct inode **tl_inode, | |
5934 | struct buffer_head **tl_bh) | |
5935 | { | |
5936 | int status; | |
5937 | struct inode *inode = NULL; | |
5938 | struct buffer_head *bh = NULL; | |
5939 | ||
5940 | inode = ocfs2_get_system_file_inode(osb, | |
5941 | TRUNCATE_LOG_SYSTEM_INODE, | |
5942 | slot_num); | |
5943 | if (!inode) { | |
5944 | status = -EINVAL; | |
5945 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
5946 | goto bail; | |
5947 | } | |
5948 | ||
b657c95c | 5949 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
5950 | if (status < 0) { |
5951 | iput(inode); | |
5952 | mlog_errno(status); | |
5953 | goto bail; | |
5954 | } | |
5955 | ||
5956 | *tl_inode = inode; | |
5957 | *tl_bh = bh; | |
5958 | bail: | |
5959 | mlog_exit(status); | |
5960 | return status; | |
5961 | } | |
5962 | ||
5963 | /* called during the 1st stage of node recovery. we stamp a clean | |
5964 | * truncate log and pass back a copy for processing later. if the | |
5965 | * truncate log does not require processing, a *tl_copy is set to | |
5966 | * NULL. */ | |
5967 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
5968 | int slot_num, | |
5969 | struct ocfs2_dinode **tl_copy) | |
5970 | { | |
5971 | int status; | |
5972 | struct inode *tl_inode = NULL; | |
5973 | struct buffer_head *tl_bh = NULL; | |
5974 | struct ocfs2_dinode *di; | |
5975 | struct ocfs2_truncate_log *tl; | |
5976 | ||
5977 | *tl_copy = NULL; | |
5978 | ||
5979 | mlog(0, "recover truncate log from slot %d\n", slot_num); | |
5980 | ||
5981 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
5982 | if (status < 0) { | |
5983 | mlog_errno(status); | |
5984 | goto bail; | |
5985 | } | |
5986 | ||
5987 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5988 | |
10995aa2 JB |
5989 | /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's |
5990 | * validated by the underlying call to ocfs2_read_inode_block(), | |
5991 | * so any corruption is a code bug */ | |
5992 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5993 | ||
5994 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5995 | if (le16_to_cpu(tl->tl_used)) { |
5996 | mlog(0, "We'll have %u logs to recover\n", | |
5997 | le16_to_cpu(tl->tl_used)); | |
5998 | ||
5999 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
6000 | if (!(*tl_copy)) { | |
6001 | status = -ENOMEM; | |
6002 | mlog_errno(status); | |
6003 | goto bail; | |
6004 | } | |
6005 | ||
6006 | /* Assuming the write-out below goes well, this copy | |
6007 | * will be passed back to recovery for processing. */ | |
6008 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
6009 | ||
6010 | /* All we need to do to clear the truncate log is set | |
6011 | * tl_used. */ | |
6012 | tl->tl_used = 0; | |
6013 | ||
13723d00 | 6014 | ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check); |
8cb471e8 | 6015 | status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode)); |
ccd979bd MF |
6016 | if (status < 0) { |
6017 | mlog_errno(status); | |
6018 | goto bail; | |
6019 | } | |
6020 | } | |
6021 | ||
6022 | bail: | |
6023 | if (tl_inode) | |
6024 | iput(tl_inode); | |
a81cb88b | 6025 | brelse(tl_bh); |
ccd979bd MF |
6026 | |
6027 | if (status < 0 && (*tl_copy)) { | |
6028 | kfree(*tl_copy); | |
6029 | *tl_copy = NULL; | |
6030 | } | |
6031 | ||
6032 | mlog_exit(status); | |
6033 | return status; | |
6034 | } | |
6035 | ||
6036 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
6037 | struct ocfs2_dinode *tl_copy) | |
6038 | { | |
6039 | int status = 0; | |
6040 | int i; | |
6041 | unsigned int clusters, num_recs, start_cluster; | |
6042 | u64 start_blk; | |
1fabe148 | 6043 | handle_t *handle; |
ccd979bd MF |
6044 | struct inode *tl_inode = osb->osb_tl_inode; |
6045 | struct ocfs2_truncate_log *tl; | |
6046 | ||
6047 | mlog_entry_void(); | |
6048 | ||
6049 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { | |
6050 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
6051 | return -EINVAL; | |
6052 | } | |
6053 | ||
6054 | tl = &tl_copy->id2.i_dealloc; | |
6055 | num_recs = le16_to_cpu(tl->tl_used); | |
b0697053 | 6056 | mlog(0, "cleanup %u records from %llu\n", num_recs, |
1ca1a111 | 6057 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno)); |
ccd979bd | 6058 | |
1b1dcc1b | 6059 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
6060 | for(i = 0; i < num_recs; i++) { |
6061 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6062 | status = __ocfs2_flush_truncate_log(osb); | |
6063 | if (status < 0) { | |
6064 | mlog_errno(status); | |
6065 | goto bail_up; | |
6066 | } | |
6067 | } | |
6068 | ||
65eff9cc | 6069 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
6070 | if (IS_ERR(handle)) { |
6071 | status = PTR_ERR(handle); | |
6072 | mlog_errno(status); | |
6073 | goto bail_up; | |
6074 | } | |
6075 | ||
6076 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
6077 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
6078 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
6079 | ||
6080 | status = ocfs2_truncate_log_append(osb, handle, | |
6081 | start_blk, clusters); | |
02dc1af4 | 6082 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
6083 | if (status < 0) { |
6084 | mlog_errno(status); | |
6085 | goto bail_up; | |
6086 | } | |
6087 | } | |
6088 | ||
6089 | bail_up: | |
1b1dcc1b | 6090 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6091 | |
6092 | mlog_exit(status); | |
6093 | return status; | |
6094 | } | |
6095 | ||
6096 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
6097 | { | |
6098 | int status; | |
6099 | struct inode *tl_inode = osb->osb_tl_inode; | |
6100 | ||
6101 | mlog_entry_void(); | |
6102 | ||
6103 | if (tl_inode) { | |
6104 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
6105 | flush_workqueue(ocfs2_wq); | |
6106 | ||
6107 | status = ocfs2_flush_truncate_log(osb); | |
6108 | if (status < 0) | |
6109 | mlog_errno(status); | |
6110 | ||
6111 | brelse(osb->osb_tl_bh); | |
6112 | iput(osb->osb_tl_inode); | |
6113 | } | |
6114 | ||
6115 | mlog_exit_void(); | |
6116 | } | |
6117 | ||
6118 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
6119 | { | |
6120 | int status; | |
6121 | struct inode *tl_inode = NULL; | |
6122 | struct buffer_head *tl_bh = NULL; | |
6123 | ||
6124 | mlog_entry_void(); | |
6125 | ||
6126 | status = ocfs2_get_truncate_log_info(osb, | |
6127 | osb->slot_num, | |
6128 | &tl_inode, | |
6129 | &tl_bh); | |
6130 | if (status < 0) | |
6131 | mlog_errno(status); | |
6132 | ||
6133 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
6134 | * osb->osb_tl_inode so we don't set any of the osb variables | |
6135 | * until we're sure all is well. */ | |
c4028958 DH |
6136 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
6137 | ocfs2_truncate_log_worker); | |
ccd979bd MF |
6138 | osb->osb_tl_bh = tl_bh; |
6139 | osb->osb_tl_inode = tl_inode; | |
6140 | ||
6141 | mlog_exit(status); | |
6142 | return status; | |
6143 | } | |
6144 | ||
2b604351 MF |
6145 | /* |
6146 | * Delayed de-allocation of suballocator blocks. | |
6147 | * | |
6148 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
6149 | * | |
6150 | * The locking for this can get extremely complicated, especially when | |
6151 | * the suballocator inodes to delete from aren't known until deep | |
6152 | * within an unrelated codepath. | |
6153 | * | |
6154 | * ocfs2_extent_block structures are a good example of this - an inode | |
6155 | * btree could have been grown by any number of nodes each allocating | |
6156 | * out of their own suballoc inode. | |
6157 | * | |
6158 | * These structures allow the delay of block de-allocation until a | |
6159 | * later time, when locking of multiple cluster inodes won't cause | |
6160 | * deadlock. | |
6161 | */ | |
6162 | ||
6163 | /* | |
2891d290 TM |
6164 | * Describe a single bit freed from a suballocator. For the block |
6165 | * suballocators, it represents one block. For the global cluster | |
6166 | * allocator, it represents some clusters and free_bit indicates | |
6167 | * clusters number. | |
2b604351 MF |
6168 | */ |
6169 | struct ocfs2_cached_block_free { | |
6170 | struct ocfs2_cached_block_free *free_next; | |
6171 | u64 free_blk; | |
6172 | unsigned int free_bit; | |
6173 | }; | |
6174 | ||
6175 | struct ocfs2_per_slot_free_list { | |
6176 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
6177 | int f_inode_type; | |
6178 | int f_slot; | |
6179 | struct ocfs2_cached_block_free *f_first; | |
6180 | }; | |
6181 | ||
2891d290 TM |
6182 | static int ocfs2_free_cached_blocks(struct ocfs2_super *osb, |
6183 | int sysfile_type, | |
6184 | int slot, | |
6185 | struct ocfs2_cached_block_free *head) | |
2b604351 MF |
6186 | { |
6187 | int ret; | |
6188 | u64 bg_blkno; | |
6189 | handle_t *handle; | |
6190 | struct inode *inode; | |
6191 | struct buffer_head *di_bh = NULL; | |
6192 | struct ocfs2_cached_block_free *tmp; | |
6193 | ||
6194 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
6195 | if (!inode) { | |
6196 | ret = -EINVAL; | |
6197 | mlog_errno(ret); | |
6198 | goto out; | |
6199 | } | |
6200 | ||
6201 | mutex_lock(&inode->i_mutex); | |
6202 | ||
e63aecb6 | 6203 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
6204 | if (ret) { |
6205 | mlog_errno(ret); | |
6206 | goto out_mutex; | |
6207 | } | |
6208 | ||
6209 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); | |
6210 | if (IS_ERR(handle)) { | |
6211 | ret = PTR_ERR(handle); | |
6212 | mlog_errno(ret); | |
6213 | goto out_unlock; | |
6214 | } | |
6215 | ||
6216 | while (head) { | |
6217 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
6218 | head->free_bit); | |
6219 | mlog(0, "Free bit: (bit %u, blkno %llu)\n", | |
6220 | head->free_bit, (unsigned long long)head->free_blk); | |
6221 | ||
6222 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
6223 | head->free_bit, bg_blkno, 1); | |
6224 | if (ret) { | |
6225 | mlog_errno(ret); | |
6226 | goto out_journal; | |
6227 | } | |
6228 | ||
6229 | ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE); | |
6230 | if (ret) { | |
6231 | mlog_errno(ret); | |
6232 | goto out_journal; | |
6233 | } | |
6234 | ||
6235 | tmp = head; | |
6236 | head = head->free_next; | |
6237 | kfree(tmp); | |
6238 | } | |
6239 | ||
6240 | out_journal: | |
6241 | ocfs2_commit_trans(osb, handle); | |
6242 | ||
6243 | out_unlock: | |
e63aecb6 | 6244 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
6245 | brelse(di_bh); |
6246 | out_mutex: | |
6247 | mutex_unlock(&inode->i_mutex); | |
6248 | iput(inode); | |
6249 | out: | |
6250 | while(head) { | |
6251 | /* Premature exit may have left some dangling items. */ | |
6252 | tmp = head; | |
6253 | head = head->free_next; | |
6254 | kfree(tmp); | |
6255 | } | |
6256 | ||
6257 | return ret; | |
6258 | } | |
6259 | ||
2891d290 TM |
6260 | int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6261 | u64 blkno, unsigned int bit) | |
6262 | { | |
6263 | int ret = 0; | |
6264 | struct ocfs2_cached_block_free *item; | |
6265 | ||
6266 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6267 | if (item == NULL) { | |
6268 | ret = -ENOMEM; | |
6269 | mlog_errno(ret); | |
6270 | return ret; | |
6271 | } | |
6272 | ||
6273 | mlog(0, "Insert clusters: (bit %u, blk %llu)\n", | |
6274 | bit, (unsigned long long)blkno); | |
6275 | ||
6276 | item->free_blk = blkno; | |
6277 | item->free_bit = bit; | |
6278 | item->free_next = ctxt->c_global_allocator; | |
6279 | ||
6280 | ctxt->c_global_allocator = item; | |
6281 | return ret; | |
6282 | } | |
6283 | ||
6284 | static int ocfs2_free_cached_clusters(struct ocfs2_super *osb, | |
6285 | struct ocfs2_cached_block_free *head) | |
6286 | { | |
6287 | struct ocfs2_cached_block_free *tmp; | |
6288 | struct inode *tl_inode = osb->osb_tl_inode; | |
6289 | handle_t *handle; | |
6290 | int ret = 0; | |
6291 | ||
6292 | mutex_lock(&tl_inode->i_mutex); | |
6293 | ||
6294 | while (head) { | |
6295 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6296 | ret = __ocfs2_flush_truncate_log(osb); | |
6297 | if (ret < 0) { | |
6298 | mlog_errno(ret); | |
6299 | break; | |
6300 | } | |
6301 | } | |
6302 | ||
6303 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); | |
6304 | if (IS_ERR(handle)) { | |
6305 | ret = PTR_ERR(handle); | |
6306 | mlog_errno(ret); | |
6307 | break; | |
6308 | } | |
6309 | ||
6310 | ret = ocfs2_truncate_log_append(osb, handle, head->free_blk, | |
6311 | head->free_bit); | |
6312 | ||
6313 | ocfs2_commit_trans(osb, handle); | |
6314 | tmp = head; | |
6315 | head = head->free_next; | |
6316 | kfree(tmp); | |
6317 | ||
6318 | if (ret < 0) { | |
6319 | mlog_errno(ret); | |
6320 | break; | |
6321 | } | |
6322 | } | |
6323 | ||
6324 | mutex_unlock(&tl_inode->i_mutex); | |
6325 | ||
6326 | while (head) { | |
6327 | /* Premature exit may have left some dangling items. */ | |
6328 | tmp = head; | |
6329 | head = head->free_next; | |
6330 | kfree(tmp); | |
6331 | } | |
6332 | ||
6333 | return ret; | |
6334 | } | |
6335 | ||
2b604351 MF |
6336 | int ocfs2_run_deallocs(struct ocfs2_super *osb, |
6337 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6338 | { | |
6339 | int ret = 0, ret2; | |
6340 | struct ocfs2_per_slot_free_list *fl; | |
6341 | ||
6342 | if (!ctxt) | |
6343 | return 0; | |
6344 | ||
6345 | while (ctxt->c_first_suballocator) { | |
6346 | fl = ctxt->c_first_suballocator; | |
6347 | ||
6348 | if (fl->f_first) { | |
6349 | mlog(0, "Free items: (type %u, slot %d)\n", | |
6350 | fl->f_inode_type, fl->f_slot); | |
2891d290 TM |
6351 | ret2 = ocfs2_free_cached_blocks(osb, |
6352 | fl->f_inode_type, | |
6353 | fl->f_slot, | |
6354 | fl->f_first); | |
2b604351 MF |
6355 | if (ret2) |
6356 | mlog_errno(ret2); | |
6357 | if (!ret) | |
6358 | ret = ret2; | |
6359 | } | |
6360 | ||
6361 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
6362 | kfree(fl); | |
6363 | } | |
6364 | ||
2891d290 TM |
6365 | if (ctxt->c_global_allocator) { |
6366 | ret2 = ocfs2_free_cached_clusters(osb, | |
6367 | ctxt->c_global_allocator); | |
6368 | if (ret2) | |
6369 | mlog_errno(ret2); | |
6370 | if (!ret) | |
6371 | ret = ret2; | |
6372 | ||
6373 | ctxt->c_global_allocator = NULL; | |
6374 | } | |
6375 | ||
2b604351 MF |
6376 | return ret; |
6377 | } | |
6378 | ||
6379 | static struct ocfs2_per_slot_free_list * | |
6380 | ocfs2_find_per_slot_free_list(int type, | |
6381 | int slot, | |
6382 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6383 | { | |
6384 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
6385 | ||
6386 | while (fl) { | |
6387 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
6388 | return fl; | |
6389 | ||
6390 | fl = fl->f_next_suballocator; | |
6391 | } | |
6392 | ||
6393 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
6394 | if (fl) { | |
6395 | fl->f_inode_type = type; | |
6396 | fl->f_slot = slot; | |
6397 | fl->f_first = NULL; | |
6398 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
6399 | ||
6400 | ctxt->c_first_suballocator = fl; | |
6401 | } | |
6402 | return fl; | |
6403 | } | |
6404 | ||
6405 | static int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, | |
6406 | int type, int slot, u64 blkno, | |
6407 | unsigned int bit) | |
6408 | { | |
6409 | int ret; | |
6410 | struct ocfs2_per_slot_free_list *fl; | |
6411 | struct ocfs2_cached_block_free *item; | |
6412 | ||
6413 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6414 | if (fl == NULL) { | |
6415 | ret = -ENOMEM; | |
6416 | mlog_errno(ret); | |
6417 | goto out; | |
6418 | } | |
6419 | ||
6420 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6421 | if (item == NULL) { | |
6422 | ret = -ENOMEM; | |
6423 | mlog_errno(ret); | |
6424 | goto out; | |
6425 | } | |
6426 | ||
6427 | mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n", | |
6428 | type, slot, bit, (unsigned long long)blkno); | |
6429 | ||
6430 | item->free_blk = blkno; | |
6431 | item->free_bit = bit; | |
6432 | item->free_next = fl->f_first; | |
6433 | ||
6434 | fl->f_first = item; | |
6435 | ||
6436 | ret = 0; | |
6437 | out: | |
6438 | return ret; | |
6439 | } | |
6440 | ||
59a5e416 MF |
6441 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6442 | struct ocfs2_extent_block *eb) | |
6443 | { | |
6444 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6445 | le16_to_cpu(eb->h_suballoc_slot), | |
6446 | le64_to_cpu(eb->h_blkno), | |
6447 | le16_to_cpu(eb->h_suballoc_bit)); | |
6448 | } | |
6449 | ||
ccd979bd MF |
6450 | /* This function will figure out whether the currently last extent |
6451 | * block will be deleted, and if it will, what the new last extent | |
6452 | * block will be so we can update his h_next_leaf_blk field, as well | |
6453 | * as the dinodes i_last_eb_blk */ | |
dcd0538f | 6454 | static int ocfs2_find_new_last_ext_blk(struct inode *inode, |
3a0782d0 | 6455 | unsigned int clusters_to_del, |
dcd0538f | 6456 | struct ocfs2_path *path, |
ccd979bd MF |
6457 | struct buffer_head **new_last_eb) |
6458 | { | |
3a0782d0 | 6459 | int next_free, ret = 0; |
dcd0538f | 6460 | u32 cpos; |
3a0782d0 | 6461 | struct ocfs2_extent_rec *rec; |
ccd979bd MF |
6462 | struct ocfs2_extent_block *eb; |
6463 | struct ocfs2_extent_list *el; | |
6464 | struct buffer_head *bh = NULL; | |
6465 | ||
6466 | *new_last_eb = NULL; | |
6467 | ||
ccd979bd | 6468 | /* we have no tree, so of course, no last_eb. */ |
dcd0538f MF |
6469 | if (!path->p_tree_depth) |
6470 | goto out; | |
ccd979bd MF |
6471 | |
6472 | /* trunc to zero special case - this makes tree_depth = 0 | |
6473 | * regardless of what it is. */ | |
3a0782d0 | 6474 | if (OCFS2_I(inode)->ip_clusters == clusters_to_del) |
dcd0538f | 6475 | goto out; |
ccd979bd | 6476 | |
dcd0538f | 6477 | el = path_leaf_el(path); |
ccd979bd MF |
6478 | BUG_ON(!el->l_next_free_rec); |
6479 | ||
3a0782d0 MF |
6480 | /* |
6481 | * Make sure that this extent list will actually be empty | |
6482 | * after we clear away the data. We can shortcut out if | |
6483 | * there's more than one non-empty extent in the | |
6484 | * list. Otherwise, a check of the remaining extent is | |
6485 | * necessary. | |
6486 | */ | |
6487 | next_free = le16_to_cpu(el->l_next_free_rec); | |
6488 | rec = NULL; | |
dcd0538f | 6489 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { |
3a0782d0 | 6490 | if (next_free > 2) |
dcd0538f | 6491 | goto out; |
3a0782d0 MF |
6492 | |
6493 | /* We may have a valid extent in index 1, check it. */ | |
6494 | if (next_free == 2) | |
6495 | rec = &el->l_recs[1]; | |
6496 | ||
6497 | /* | |
6498 | * Fall through - no more nonempty extents, so we want | |
6499 | * to delete this leaf. | |
6500 | */ | |
6501 | } else { | |
6502 | if (next_free > 1) | |
6503 | goto out; | |
6504 | ||
6505 | rec = &el->l_recs[0]; | |
6506 | } | |
6507 | ||
6508 | if (rec) { | |
6509 | /* | |
6510 | * Check it we'll only be trimming off the end of this | |
6511 | * cluster. | |
6512 | */ | |
e48edee2 | 6513 | if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del) |
3a0782d0 MF |
6514 | goto out; |
6515 | } | |
ccd979bd | 6516 | |
dcd0538f MF |
6517 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); |
6518 | if (ret) { | |
6519 | mlog_errno(ret); | |
6520 | goto out; | |
6521 | } | |
ccd979bd | 6522 | |
facdb77f | 6523 | ret = ocfs2_find_leaf(INODE_CACHE(inode), path_root_el(path), cpos, &bh); |
dcd0538f MF |
6524 | if (ret) { |
6525 | mlog_errno(ret); | |
6526 | goto out; | |
6527 | } | |
ccd979bd | 6528 | |
dcd0538f MF |
6529 | eb = (struct ocfs2_extent_block *) bh->b_data; |
6530 | el = &eb->h_list; | |
5e96581a JB |
6531 | |
6532 | /* ocfs2_find_leaf() gets the eb from ocfs2_read_extent_block(). | |
6533 | * Any corruption is a code bug. */ | |
6534 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
6535 | |
6536 | *new_last_eb = bh; | |
6537 | get_bh(*new_last_eb); | |
dcd0538f MF |
6538 | mlog(0, "returning block %llu, (cpos: %u)\n", |
6539 | (unsigned long long)le64_to_cpu(eb->h_blkno), cpos); | |
6540 | out: | |
6541 | brelse(bh); | |
ccd979bd | 6542 | |
dcd0538f | 6543 | return ret; |
ccd979bd MF |
6544 | } |
6545 | ||
3a0782d0 MF |
6546 | /* |
6547 | * Trim some clusters off the rightmost edge of a tree. Only called | |
6548 | * during truncate. | |
6549 | * | |
6550 | * The caller needs to: | |
6551 | * - start journaling of each path component. | |
6552 | * - compute and fully set up any new last ext block | |
6553 | */ | |
6554 | static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path, | |
6555 | handle_t *handle, struct ocfs2_truncate_context *tc, | |
6556 | u32 clusters_to_del, u64 *delete_start) | |
6557 | { | |
6558 | int ret, i, index = path->p_tree_depth; | |
6559 | u32 new_edge = 0; | |
6560 | u64 deleted_eb = 0; | |
6561 | struct buffer_head *bh; | |
6562 | struct ocfs2_extent_list *el; | |
6563 | struct ocfs2_extent_rec *rec; | |
6564 | ||
6565 | *delete_start = 0; | |
6566 | ||
6567 | while (index >= 0) { | |
6568 | bh = path->p_node[index].bh; | |
6569 | el = path->p_node[index].el; | |
6570 | ||
6571 | mlog(0, "traveling tree (index = %d, block = %llu)\n", | |
6572 | index, (unsigned long long)bh->b_blocknr); | |
6573 | ||
6574 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
6575 | ||
6576 | if (index != | |
6577 | (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) { | |
6578 | ocfs2_error(inode->i_sb, | |
6579 | "Inode %lu has invalid ext. block %llu", | |
6580 | inode->i_ino, | |
6581 | (unsigned long long)bh->b_blocknr); | |
6582 | ret = -EROFS; | |
6583 | goto out; | |
6584 | } | |
6585 | ||
6586 | find_tail_record: | |
6587 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
6588 | rec = &el->l_recs[i]; | |
6589 | ||
6590 | mlog(0, "Extent list before: record %d: (%u, %u, %llu), " | |
6591 | "next = %u\n", i, le32_to_cpu(rec->e_cpos), | |
e48edee2 | 6592 | ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6593 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6594 | le16_to_cpu(el->l_next_free_rec)); | |
6595 | ||
e48edee2 | 6596 | BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del); |
3a0782d0 MF |
6597 | |
6598 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
6599 | /* | |
6600 | * If the leaf block contains a single empty | |
6601 | * extent and no records, we can just remove | |
6602 | * the block. | |
6603 | */ | |
6604 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
6605 | memset(rec, 0, | |
6606 | sizeof(struct ocfs2_extent_rec)); | |
6607 | el->l_next_free_rec = cpu_to_le16(0); | |
6608 | ||
6609 | goto delete; | |
6610 | } | |
6611 | ||
6612 | /* | |
6613 | * Remove any empty extents by shifting things | |
6614 | * left. That should make life much easier on | |
6615 | * the code below. This condition is rare | |
6616 | * enough that we shouldn't see a performance | |
6617 | * hit. | |
6618 | */ | |
6619 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
6620 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6621 | ||
6622 | for(i = 0; | |
6623 | i < le16_to_cpu(el->l_next_free_rec); i++) | |
6624 | el->l_recs[i] = el->l_recs[i + 1]; | |
6625 | ||
6626 | memset(&el->l_recs[i], 0, | |
6627 | sizeof(struct ocfs2_extent_rec)); | |
6628 | ||
6629 | /* | |
6630 | * We've modified our extent list. The | |
6631 | * simplest way to handle this change | |
6632 | * is to being the search from the | |
6633 | * start again. | |
6634 | */ | |
6635 | goto find_tail_record; | |
6636 | } | |
6637 | ||
e48edee2 | 6638 | le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del); |
3a0782d0 MF |
6639 | |
6640 | /* | |
6641 | * We'll use "new_edge" on our way back up the | |
6642 | * tree to know what our rightmost cpos is. | |
6643 | */ | |
e48edee2 | 6644 | new_edge = le16_to_cpu(rec->e_leaf_clusters); |
3a0782d0 MF |
6645 | new_edge += le32_to_cpu(rec->e_cpos); |
6646 | ||
6647 | /* | |
6648 | * The caller will use this to delete data blocks. | |
6649 | */ | |
6650 | *delete_start = le64_to_cpu(rec->e_blkno) | |
6651 | + ocfs2_clusters_to_blocks(inode->i_sb, | |
e48edee2 | 6652 | le16_to_cpu(rec->e_leaf_clusters)); |
3a0782d0 MF |
6653 | |
6654 | /* | |
6655 | * If it's now empty, remove this record. | |
6656 | */ | |
e48edee2 | 6657 | if (le16_to_cpu(rec->e_leaf_clusters) == 0) { |
3a0782d0 MF |
6658 | memset(rec, 0, |
6659 | sizeof(struct ocfs2_extent_rec)); | |
6660 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6661 | } | |
6662 | } else { | |
6663 | if (le64_to_cpu(rec->e_blkno) == deleted_eb) { | |
6664 | memset(rec, 0, | |
6665 | sizeof(struct ocfs2_extent_rec)); | |
6666 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6667 | ||
6668 | goto delete; | |
6669 | } | |
6670 | ||
6671 | /* Can this actually happen? */ | |
6672 | if (le16_to_cpu(el->l_next_free_rec) == 0) | |
6673 | goto delete; | |
6674 | ||
6675 | /* | |
6676 | * We never actually deleted any clusters | |
6677 | * because our leaf was empty. There's no | |
6678 | * reason to adjust the rightmost edge then. | |
6679 | */ | |
6680 | if (new_edge == 0) | |
6681 | goto delete; | |
6682 | ||
e48edee2 MF |
6683 | rec->e_int_clusters = cpu_to_le32(new_edge); |
6684 | le32_add_cpu(&rec->e_int_clusters, | |
3a0782d0 MF |
6685 | -le32_to_cpu(rec->e_cpos)); |
6686 | ||
6687 | /* | |
6688 | * A deleted child record should have been | |
6689 | * caught above. | |
6690 | */ | |
e48edee2 | 6691 | BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0); |
3a0782d0 MF |
6692 | } |
6693 | ||
6694 | delete: | |
6695 | ret = ocfs2_journal_dirty(handle, bh); | |
6696 | if (ret) { | |
6697 | mlog_errno(ret); | |
6698 | goto out; | |
6699 | } | |
6700 | ||
6701 | mlog(0, "extent list container %llu, after: record %d: " | |
6702 | "(%u, %u, %llu), next = %u.\n", | |
6703 | (unsigned long long)bh->b_blocknr, i, | |
e48edee2 | 6704 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6705 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6706 | le16_to_cpu(el->l_next_free_rec)); | |
6707 | ||
6708 | /* | |
6709 | * We must be careful to only attempt delete of an | |
6710 | * extent block (and not the root inode block). | |
6711 | */ | |
6712 | if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) { | |
6713 | struct ocfs2_extent_block *eb = | |
6714 | (struct ocfs2_extent_block *)bh->b_data; | |
6715 | ||
6716 | /* | |
6717 | * Save this for use when processing the | |
6718 | * parent block. | |
6719 | */ | |
6720 | deleted_eb = le64_to_cpu(eb->h_blkno); | |
6721 | ||
6722 | mlog(0, "deleting this extent block.\n"); | |
6723 | ||
8cb471e8 | 6724 | ocfs2_remove_from_cache(INODE_CACHE(inode), bh); |
3a0782d0 | 6725 | |
e48edee2 | 6726 | BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0])); |
3a0782d0 MF |
6727 | BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos)); |
6728 | BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno)); | |
6729 | ||
59a5e416 MF |
6730 | ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb); |
6731 | /* An error here is not fatal. */ | |
6732 | if (ret < 0) | |
6733 | mlog_errno(ret); | |
3a0782d0 MF |
6734 | } else { |
6735 | deleted_eb = 0; | |
6736 | } | |
6737 | ||
6738 | index--; | |
6739 | } | |
6740 | ||
6741 | ret = 0; | |
6742 | out: | |
6743 | return ret; | |
6744 | } | |
6745 | ||
ccd979bd MF |
6746 | static int ocfs2_do_truncate(struct ocfs2_super *osb, |
6747 | unsigned int clusters_to_del, | |
6748 | struct inode *inode, | |
6749 | struct buffer_head *fe_bh, | |
1fabe148 | 6750 | handle_t *handle, |
dcd0538f MF |
6751 | struct ocfs2_truncate_context *tc, |
6752 | struct ocfs2_path *path) | |
ccd979bd | 6753 | { |
3a0782d0 | 6754 | int status; |
ccd979bd | 6755 | struct ocfs2_dinode *fe; |
ccd979bd MF |
6756 | struct ocfs2_extent_block *last_eb = NULL; |
6757 | struct ocfs2_extent_list *el; | |
ccd979bd | 6758 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
6759 | u64 delete_blk = 0; |
6760 | ||
6761 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
6762 | ||
3a0782d0 | 6763 | status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del, |
dcd0538f | 6764 | path, &last_eb_bh); |
ccd979bd MF |
6765 | if (status < 0) { |
6766 | mlog_errno(status); | |
6767 | goto bail; | |
6768 | } | |
dcd0538f MF |
6769 | |
6770 | /* | |
6771 | * Each component will be touched, so we might as well journal | |
6772 | * here to avoid having to handle errors later. | |
6773 | */ | |
0cf2f763 | 6774 | status = ocfs2_journal_access_path(INODE_CACHE(inode), handle, path); |
3a0782d0 MF |
6775 | if (status < 0) { |
6776 | mlog_errno(status); | |
6777 | goto bail; | |
dcd0538f MF |
6778 | } |
6779 | ||
6780 | if (last_eb_bh) { | |
0cf2f763 | 6781 | status = ocfs2_journal_access_eb(handle, INODE_CACHE(inode), last_eb_bh, |
13723d00 | 6782 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
6783 | if (status < 0) { |
6784 | mlog_errno(status); | |
6785 | goto bail; | |
6786 | } | |
6787 | ||
ccd979bd | 6788 | last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; |
dcd0538f | 6789 | } |
ccd979bd | 6790 | |
dcd0538f MF |
6791 | el = &(fe->id2.i_list); |
6792 | ||
6793 | /* | |
6794 | * Lower levels depend on this never happening, but it's best | |
6795 | * to check it up here before changing the tree. | |
6796 | */ | |
e48edee2 | 6797 | if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) { |
dcd0538f MF |
6798 | ocfs2_error(inode->i_sb, |
6799 | "Inode %lu has an empty extent record, depth %u\n", | |
6800 | inode->i_ino, le16_to_cpu(el->l_tree_depth)); | |
3a0782d0 | 6801 | status = -EROFS; |
ccd979bd MF |
6802 | goto bail; |
6803 | } | |
ccd979bd | 6804 | |
a90714c1 JK |
6805 | vfs_dq_free_space_nodirty(inode, |
6806 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_del)); | |
ccd979bd MF |
6807 | spin_lock(&OCFS2_I(inode)->ip_lock); |
6808 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) - | |
6809 | clusters_to_del; | |
6810 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
6811 | le32_add_cpu(&fe->i_clusters, -clusters_to_del); | |
e535e2ef | 6812 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 6813 | |
3a0782d0 MF |
6814 | status = ocfs2_trim_tree(inode, path, handle, tc, |
6815 | clusters_to_del, &delete_blk); | |
6816 | if (status) { | |
6817 | mlog_errno(status); | |
6818 | goto bail; | |
ccd979bd MF |
6819 | } |
6820 | ||
dcd0538f | 6821 | if (le32_to_cpu(fe->i_clusters) == 0) { |
ccd979bd MF |
6822 | /* trunc to zero is a special case. */ |
6823 | el->l_tree_depth = 0; | |
6824 | fe->i_last_eb_blk = 0; | |
6825 | } else if (last_eb) | |
6826 | fe->i_last_eb_blk = last_eb->h_blkno; | |
6827 | ||
6828 | status = ocfs2_journal_dirty(handle, fe_bh); | |
6829 | if (status < 0) { | |
6830 | mlog_errno(status); | |
6831 | goto bail; | |
6832 | } | |
6833 | ||
6834 | if (last_eb) { | |
6835 | /* If there will be a new last extent block, then by | |
6836 | * definition, there cannot be any leaves to the right of | |
6837 | * him. */ | |
ccd979bd MF |
6838 | last_eb->h_next_leaf_blk = 0; |
6839 | status = ocfs2_journal_dirty(handle, last_eb_bh); | |
6840 | if (status < 0) { | |
6841 | mlog_errno(status); | |
6842 | goto bail; | |
6843 | } | |
6844 | } | |
6845 | ||
3a0782d0 MF |
6846 | if (delete_blk) { |
6847 | status = ocfs2_truncate_log_append(osb, handle, delete_blk, | |
6848 | clusters_to_del); | |
ccd979bd MF |
6849 | if (status < 0) { |
6850 | mlog_errno(status); | |
6851 | goto bail; | |
6852 | } | |
ccd979bd MF |
6853 | } |
6854 | status = 0; | |
6855 | bail: | |
60e2ec48 | 6856 | brelse(last_eb_bh); |
ccd979bd MF |
6857 | mlog_exit(status); |
6858 | return status; | |
6859 | } | |
6860 | ||
2b4e30fb | 6861 | static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh) |
60b11392 MF |
6862 | { |
6863 | set_buffer_uptodate(bh); | |
6864 | mark_buffer_dirty(bh); | |
6865 | return 0; | |
6866 | } | |
6867 | ||
1d410a6e MF |
6868 | static void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
6869 | unsigned int from, unsigned int to, | |
6870 | struct page *page, int zero, u64 *phys) | |
6871 | { | |
6872 | int ret, partial = 0; | |
6873 | ||
6874 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
6875 | if (ret) | |
6876 | mlog_errno(ret); | |
6877 | ||
6878 | if (zero) | |
eebd2aa3 | 6879 | zero_user_segment(page, from, to); |
1d410a6e MF |
6880 | |
6881 | /* | |
6882 | * Need to set the buffers we zero'd into uptodate | |
6883 | * here if they aren't - ocfs2_map_page_blocks() | |
6884 | * might've skipped some | |
6885 | */ | |
2b4e30fb JB |
6886 | ret = walk_page_buffers(handle, page_buffers(page), |
6887 | from, to, &partial, | |
6888 | ocfs2_zero_func); | |
6889 | if (ret < 0) | |
6890 | mlog_errno(ret); | |
6891 | else if (ocfs2_should_order_data(inode)) { | |
6892 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
1d410a6e MF |
6893 | if (ret < 0) |
6894 | mlog_errno(ret); | |
6895 | } | |
6896 | ||
6897 | if (!partial) | |
6898 | SetPageUptodate(page); | |
6899 | ||
6900 | flush_dcache_page(page); | |
6901 | } | |
6902 | ||
35edec1d MF |
6903 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
6904 | loff_t end, struct page **pages, | |
6905 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 6906 | { |
1d410a6e | 6907 | int i; |
60b11392 MF |
6908 | struct page *page; |
6909 | unsigned int from, to = PAGE_CACHE_SIZE; | |
6910 | struct super_block *sb = inode->i_sb; | |
6911 | ||
6912 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
6913 | ||
6914 | if (numpages == 0) | |
6915 | goto out; | |
6916 | ||
35edec1d | 6917 | to = PAGE_CACHE_SIZE; |
60b11392 MF |
6918 | for(i = 0; i < numpages; i++) { |
6919 | page = pages[i]; | |
6920 | ||
35edec1d MF |
6921 | from = start & (PAGE_CACHE_SIZE - 1); |
6922 | if ((end >> PAGE_CACHE_SHIFT) == page->index) | |
6923 | to = end & (PAGE_CACHE_SIZE - 1); | |
6924 | ||
60b11392 MF |
6925 | BUG_ON(from > PAGE_CACHE_SIZE); |
6926 | BUG_ON(to > PAGE_CACHE_SIZE); | |
6927 | ||
1d410a6e MF |
6928 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
6929 | &phys); | |
60b11392 | 6930 | |
35edec1d | 6931 | start = (page->index + 1) << PAGE_CACHE_SHIFT; |
60b11392 MF |
6932 | } |
6933 | out: | |
1d410a6e MF |
6934 | if (pages) |
6935 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6936 | } |
6937 | ||
35edec1d | 6938 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
1d410a6e | 6939 | struct page **pages, int *num) |
60b11392 | 6940 | { |
1d410a6e | 6941 | int numpages, ret = 0; |
60b11392 MF |
6942 | struct super_block *sb = inode->i_sb; |
6943 | struct address_space *mapping = inode->i_mapping; | |
6944 | unsigned long index; | |
35edec1d | 6945 | loff_t last_page_bytes; |
60b11392 | 6946 | |
35edec1d | 6947 | BUG_ON(start > end); |
60b11392 | 6948 | |
35edec1d MF |
6949 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != |
6950 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
6951 | ||
1d410a6e | 6952 | numpages = 0; |
35edec1d MF |
6953 | last_page_bytes = PAGE_ALIGN(end); |
6954 | index = start >> PAGE_CACHE_SHIFT; | |
60b11392 MF |
6955 | do { |
6956 | pages[numpages] = grab_cache_page(mapping, index); | |
6957 | if (!pages[numpages]) { | |
6958 | ret = -ENOMEM; | |
6959 | mlog_errno(ret); | |
6960 | goto out; | |
6961 | } | |
6962 | ||
6963 | numpages++; | |
6964 | index++; | |
35edec1d | 6965 | } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT)); |
60b11392 MF |
6966 | |
6967 | out: | |
6968 | if (ret != 0) { | |
1d410a6e MF |
6969 | if (pages) |
6970 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6971 | numpages = 0; |
6972 | } | |
6973 | ||
6974 | *num = numpages; | |
6975 | ||
6976 | return ret; | |
6977 | } | |
6978 | ||
6979 | /* | |
6980 | * Zero the area past i_size but still within an allocated | |
6981 | * cluster. This avoids exposing nonzero data on subsequent file | |
6982 | * extends. | |
6983 | * | |
6984 | * We need to call this before i_size is updated on the inode because | |
6985 | * otherwise block_write_full_page() will skip writeout of pages past | |
6986 | * i_size. The new_i_size parameter is passed for this reason. | |
6987 | */ | |
35edec1d MF |
6988 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
6989 | u64 range_start, u64 range_end) | |
60b11392 | 6990 | { |
1d410a6e | 6991 | int ret = 0, numpages; |
60b11392 MF |
6992 | struct page **pages = NULL; |
6993 | u64 phys; | |
1d410a6e MF |
6994 | unsigned int ext_flags; |
6995 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
6996 | |
6997 | /* | |
6998 | * File systems which don't support sparse files zero on every | |
6999 | * extend. | |
7000 | */ | |
1d410a6e | 7001 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
7002 | return 0; |
7003 | ||
1d410a6e | 7004 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
7005 | sizeof(struct page *), GFP_NOFS); |
7006 | if (pages == NULL) { | |
7007 | ret = -ENOMEM; | |
7008 | mlog_errno(ret); | |
7009 | goto out; | |
7010 | } | |
7011 | ||
1d410a6e MF |
7012 | if (range_start == range_end) |
7013 | goto out; | |
7014 | ||
7015 | ret = ocfs2_extent_map_get_blocks(inode, | |
7016 | range_start >> sb->s_blocksize_bits, | |
7017 | &phys, NULL, &ext_flags); | |
60b11392 MF |
7018 | if (ret) { |
7019 | mlog_errno(ret); | |
7020 | goto out; | |
7021 | } | |
7022 | ||
1d410a6e MF |
7023 | /* |
7024 | * Tail is a hole, or is marked unwritten. In either case, we | |
7025 | * can count on read and write to return/push zero's. | |
7026 | */ | |
7027 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
7028 | goto out; |
7029 | ||
1d410a6e MF |
7030 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
7031 | &numpages); | |
7032 | if (ret) { | |
7033 | mlog_errno(ret); | |
7034 | goto out; | |
7035 | } | |
7036 | ||
35edec1d MF |
7037 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
7038 | numpages, phys, handle); | |
60b11392 MF |
7039 | |
7040 | /* | |
7041 | * Initiate writeout of the pages we zero'd here. We don't | |
7042 | * wait on them - the truncate_inode_pages() call later will | |
7043 | * do that for us. | |
7044 | */ | |
35edec1d MF |
7045 | ret = do_sync_mapping_range(inode->i_mapping, range_start, |
7046 | range_end - 1, SYNC_FILE_RANGE_WRITE); | |
60b11392 MF |
7047 | if (ret) |
7048 | mlog_errno(ret); | |
7049 | ||
7050 | out: | |
7051 | if (pages) | |
7052 | kfree(pages); | |
7053 | ||
7054 | return ret; | |
7055 | } | |
7056 | ||
fdd77704 TY |
7057 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
7058 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
7059 | { |
7060 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 7061 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 7062 | |
fdd77704 TY |
7063 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
7064 | memset(&di->id2, 0, blocksize - | |
7065 | offsetof(struct ocfs2_dinode, id2) - | |
7066 | xattrsize); | |
7067 | else | |
7068 | memset(&di->id2, 0, blocksize - | |
7069 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
7070 | } |
7071 | ||
5b6a3a2b MF |
7072 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
7073 | struct ocfs2_dinode *di) | |
7074 | { | |
fdd77704 | 7075 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
7076 | di->id2.i_list.l_tree_depth = 0; |
7077 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
7078 | di->id2.i_list.l_count = cpu_to_le16( |
7079 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
7080 | } |
7081 | ||
1afc32b9 MF |
7082 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
7083 | { | |
7084 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
7085 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7086 | ||
7087 | spin_lock(&oi->ip_lock); | |
7088 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
7089 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7090 | spin_unlock(&oi->ip_lock); | |
7091 | ||
7092 | /* | |
7093 | * We clear the entire i_data structure here so that all | |
7094 | * fields can be properly initialized. | |
7095 | */ | |
fdd77704 | 7096 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 7097 | |
fdd77704 TY |
7098 | idata->id_count = cpu_to_le16( |
7099 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
7100 | } |
7101 | ||
7102 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
7103 | struct buffer_head *di_bh) | |
7104 | { | |
7105 | int ret, i, has_data, num_pages = 0; | |
7106 | handle_t *handle; | |
7107 | u64 uninitialized_var(block); | |
7108 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
7109 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7110 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
7111 | struct ocfs2_alloc_context *data_ac = NULL; |
7112 | struct page **pages = NULL; | |
7113 | loff_t end = osb->s_clustersize; | |
f99b9b7c | 7114 | struct ocfs2_extent_tree et; |
a90714c1 | 7115 | int did_quota = 0; |
1afc32b9 MF |
7116 | |
7117 | has_data = i_size_read(inode) ? 1 : 0; | |
7118 | ||
7119 | if (has_data) { | |
7120 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
7121 | sizeof(struct page *), GFP_NOFS); | |
7122 | if (pages == NULL) { | |
7123 | ret = -ENOMEM; | |
7124 | mlog_errno(ret); | |
7125 | goto out; | |
7126 | } | |
7127 | ||
7128 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
7129 | if (ret) { | |
7130 | mlog_errno(ret); | |
7131 | goto out; | |
7132 | } | |
7133 | } | |
7134 | ||
a90714c1 JK |
7135 | handle = ocfs2_start_trans(osb, |
7136 | ocfs2_inline_to_extents_credits(osb->sb)); | |
1afc32b9 MF |
7137 | if (IS_ERR(handle)) { |
7138 | ret = PTR_ERR(handle); | |
7139 | mlog_errno(ret); | |
7140 | goto out_unlock; | |
7141 | } | |
7142 | ||
0cf2f763 | 7143 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7144 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7145 | if (ret) { |
7146 | mlog_errno(ret); | |
7147 | goto out_commit; | |
7148 | } | |
7149 | ||
7150 | if (has_data) { | |
7151 | u32 bit_off, num; | |
7152 | unsigned int page_end; | |
7153 | u64 phys; | |
7154 | ||
a90714c1 JK |
7155 | if (vfs_dq_alloc_space_nodirty(inode, |
7156 | ocfs2_clusters_to_bytes(osb->sb, 1))) { | |
7157 | ret = -EDQUOT; | |
7158 | goto out_commit; | |
7159 | } | |
7160 | did_quota = 1; | |
7161 | ||
1afc32b9 MF |
7162 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, |
7163 | &num); | |
7164 | if (ret) { | |
7165 | mlog_errno(ret); | |
7166 | goto out_commit; | |
7167 | } | |
7168 | ||
7169 | /* | |
7170 | * Save two copies, one for insert, and one that can | |
7171 | * be changed by ocfs2_map_and_dirty_page() below. | |
7172 | */ | |
7173 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
7174 | ||
7175 | /* | |
7176 | * Non sparse file systems zero on extend, so no need | |
7177 | * to do that now. | |
7178 | */ | |
7179 | if (!ocfs2_sparse_alloc(osb) && | |
7180 | PAGE_CACHE_SIZE < osb->s_clustersize) | |
7181 | end = PAGE_CACHE_SIZE; | |
7182 | ||
7183 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
7184 | if (ret) { | |
7185 | mlog_errno(ret); | |
7186 | goto out_commit; | |
7187 | } | |
7188 | ||
7189 | /* | |
7190 | * This should populate the 1st page for us and mark | |
7191 | * it up to date. | |
7192 | */ | |
7193 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
7194 | if (ret) { | |
7195 | mlog_errno(ret); | |
7196 | goto out_commit; | |
7197 | } | |
7198 | ||
7199 | page_end = PAGE_CACHE_SIZE; | |
7200 | if (PAGE_CACHE_SIZE > osb->s_clustersize) | |
7201 | page_end = osb->s_clustersize; | |
7202 | ||
7203 | for (i = 0; i < num_pages; i++) | |
7204 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
7205 | pages[i], i > 0, &phys); | |
7206 | } | |
7207 | ||
7208 | spin_lock(&oi->ip_lock); | |
7209 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
7210 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7211 | spin_unlock(&oi->ip_lock); | |
7212 | ||
5b6a3a2b | 7213 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
7214 | |
7215 | ocfs2_journal_dirty(handle, di_bh); | |
7216 | ||
7217 | if (has_data) { | |
7218 | /* | |
7219 | * An error at this point should be extremely rare. If | |
7220 | * this proves to be false, we could always re-build | |
7221 | * the in-inode data from our pages. | |
7222 | */ | |
8d6220d6 | 7223 | ocfs2_init_dinode_extent_tree(&et, inode, di_bh); |
f99b9b7c JB |
7224 | ret = ocfs2_insert_extent(osb, handle, inode, &et, |
7225 | 0, block, 1, 0, NULL); | |
1afc32b9 MF |
7226 | if (ret) { |
7227 | mlog_errno(ret); | |
7228 | goto out_commit; | |
7229 | } | |
7230 | ||
7231 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7232 | } | |
7233 | ||
7234 | out_commit: | |
a90714c1 JK |
7235 | if (ret < 0 && did_quota) |
7236 | vfs_dq_free_space_nodirty(inode, | |
7237 | ocfs2_clusters_to_bytes(osb->sb, 1)); | |
7238 | ||
1afc32b9 MF |
7239 | ocfs2_commit_trans(osb, handle); |
7240 | ||
7241 | out_unlock: | |
7242 | if (data_ac) | |
7243 | ocfs2_free_alloc_context(data_ac); | |
7244 | ||
7245 | out: | |
7246 | if (pages) { | |
7247 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
7248 | kfree(pages); | |
7249 | } | |
7250 | ||
7251 | return ret; | |
7252 | } | |
7253 | ||
ccd979bd MF |
7254 | /* |
7255 | * It is expected, that by the time you call this function, | |
7256 | * inode->i_size and fe->i_size have been adjusted. | |
7257 | * | |
7258 | * WARNING: This will kfree the truncate context | |
7259 | */ | |
7260 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
7261 | struct inode *inode, | |
7262 | struct buffer_head *fe_bh, | |
7263 | struct ocfs2_truncate_context *tc) | |
7264 | { | |
7265 | int status, i, credits, tl_sem = 0; | |
dcd0538f | 7266 | u32 clusters_to_del, new_highest_cpos, range; |
ccd979bd | 7267 | struct ocfs2_extent_list *el; |
1fabe148 | 7268 | handle_t *handle = NULL; |
ccd979bd | 7269 | struct inode *tl_inode = osb->osb_tl_inode; |
dcd0538f | 7270 | struct ocfs2_path *path = NULL; |
e7d4cb6b | 7271 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data; |
ccd979bd MF |
7272 | |
7273 | mlog_entry_void(); | |
7274 | ||
dcd0538f | 7275 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
7276 | i_size_read(inode)); |
7277 | ||
13723d00 JB |
7278 | path = ocfs2_new_path(fe_bh, &di->id2.i_list, |
7279 | ocfs2_journal_access_di); | |
dcd0538f MF |
7280 | if (!path) { |
7281 | status = -ENOMEM; | |
7282 | mlog_errno(status); | |
7283 | goto bail; | |
7284 | } | |
83418978 MF |
7285 | |
7286 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
7287 | ||
ccd979bd | 7288 | start: |
3a0782d0 MF |
7289 | /* |
7290 | * Check that we still have allocation to delete. | |
7291 | */ | |
7292 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
7293 | status = 0; | |
7294 | goto bail; | |
7295 | } | |
7296 | ||
dcd0538f MF |
7297 | /* |
7298 | * Truncate always works against the rightmost tree branch. | |
7299 | */ | |
facdb77f | 7300 | status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX); |
dcd0538f MF |
7301 | if (status) { |
7302 | mlog_errno(status); | |
7303 | goto bail; | |
ccd979bd MF |
7304 | } |
7305 | ||
dcd0538f MF |
7306 | mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n", |
7307 | OCFS2_I(inode)->ip_clusters, path->p_tree_depth); | |
7308 | ||
7309 | /* | |
7310 | * By now, el will point to the extent list on the bottom most | |
7311 | * portion of this tree. Only the tail record is considered in | |
7312 | * each pass. | |
7313 | * | |
7314 | * We handle the following cases, in order: | |
7315 | * - empty extent: delete the remaining branch | |
7316 | * - remove the entire record | |
7317 | * - remove a partial record | |
7318 | * - no record needs to be removed (truncate has completed) | |
7319 | */ | |
7320 | el = path_leaf_el(path); | |
3a0782d0 MF |
7321 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
7322 | ocfs2_error(inode->i_sb, | |
7323 | "Inode %llu has empty extent block at %llu\n", | |
7324 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7325 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7326 | status = -EROFS; | |
7327 | goto bail; | |
7328 | } | |
7329 | ||
ccd979bd | 7330 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
dcd0538f | 7331 | range = le32_to_cpu(el->l_recs[i].e_cpos) + |
e48edee2 | 7332 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
7333 | if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) { |
7334 | clusters_to_del = 0; | |
7335 | } else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) { | |
e48edee2 | 7336 | clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f | 7337 | } else if (range > new_highest_cpos) { |
e48edee2 | 7338 | clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) + |
ccd979bd | 7339 | le32_to_cpu(el->l_recs[i].e_cpos)) - |
dcd0538f MF |
7340 | new_highest_cpos; |
7341 | } else { | |
7342 | status = 0; | |
7343 | goto bail; | |
7344 | } | |
ccd979bd | 7345 | |
dcd0538f MF |
7346 | mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n", |
7347 | clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7348 | ||
1b1dcc1b | 7349 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
7350 | tl_sem = 1; |
7351 | /* ocfs2_truncate_log_needs_flush guarantees us at least one | |
7352 | * record is free for use. If there isn't any, we flush to get | |
7353 | * an empty truncate log. */ | |
7354 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
7355 | status = __ocfs2_flush_truncate_log(osb); | |
7356 | if (status < 0) { | |
7357 | mlog_errno(status); | |
7358 | goto bail; | |
7359 | } | |
7360 | } | |
7361 | ||
7362 | credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del, | |
dcd0538f MF |
7363 | (struct ocfs2_dinode *)fe_bh->b_data, |
7364 | el); | |
65eff9cc | 7365 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
7366 | if (IS_ERR(handle)) { |
7367 | status = PTR_ERR(handle); | |
7368 | handle = NULL; | |
7369 | mlog_errno(status); | |
7370 | goto bail; | |
7371 | } | |
7372 | ||
dcd0538f MF |
7373 | status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle, |
7374 | tc, path); | |
ccd979bd MF |
7375 | if (status < 0) { |
7376 | mlog_errno(status); | |
7377 | goto bail; | |
7378 | } | |
7379 | ||
1b1dcc1b | 7380 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7381 | tl_sem = 0; |
7382 | ||
02dc1af4 | 7383 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
7384 | handle = NULL; |
7385 | ||
dcd0538f MF |
7386 | ocfs2_reinit_path(path, 1); |
7387 | ||
7388 | /* | |
3a0782d0 MF |
7389 | * The check above will catch the case where we've truncated |
7390 | * away all allocation. | |
dcd0538f | 7391 | */ |
3a0782d0 MF |
7392 | goto start; |
7393 | ||
ccd979bd | 7394 | bail: |
ccd979bd MF |
7395 | |
7396 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
7397 | ||
7398 | if (tl_sem) | |
1b1dcc1b | 7399 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7400 | |
7401 | if (handle) | |
02dc1af4 | 7402 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 7403 | |
59a5e416 MF |
7404 | ocfs2_run_deallocs(osb, &tc->tc_dealloc); |
7405 | ||
dcd0538f | 7406 | ocfs2_free_path(path); |
ccd979bd MF |
7407 | |
7408 | /* This will drop the ext_alloc cluster lock for us */ | |
7409 | ocfs2_free_truncate_context(tc); | |
7410 | ||
7411 | mlog_exit(status); | |
7412 | return status; | |
7413 | } | |
7414 | ||
ccd979bd | 7415 | /* |
59a5e416 | 7416 | * Expects the inode to already be locked. |
ccd979bd MF |
7417 | */ |
7418 | int ocfs2_prepare_truncate(struct ocfs2_super *osb, | |
7419 | struct inode *inode, | |
7420 | struct buffer_head *fe_bh, | |
7421 | struct ocfs2_truncate_context **tc) | |
7422 | { | |
59a5e416 | 7423 | int status; |
ccd979bd MF |
7424 | unsigned int new_i_clusters; |
7425 | struct ocfs2_dinode *fe; | |
7426 | struct ocfs2_extent_block *eb; | |
ccd979bd | 7427 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
7428 | |
7429 | mlog_entry_void(); | |
7430 | ||
7431 | *tc = NULL; | |
7432 | ||
7433 | new_i_clusters = ocfs2_clusters_for_bytes(osb->sb, | |
7434 | i_size_read(inode)); | |
7435 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
7436 | ||
7437 | mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size =" | |
1ca1a111 MF |
7438 | "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters, |
7439 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 7440 | |
cd861280 | 7441 | *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL); |
ccd979bd MF |
7442 | if (!(*tc)) { |
7443 | status = -ENOMEM; | |
7444 | mlog_errno(status); | |
7445 | goto bail; | |
7446 | } | |
59a5e416 | 7447 | ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc); |
ccd979bd | 7448 | |
ccd979bd | 7449 | if (fe->id2.i_list.l_tree_depth) { |
3d03a305 | 7450 | status = ocfs2_read_extent_block(INODE_CACHE(inode), |
5e96581a JB |
7451 | le64_to_cpu(fe->i_last_eb_blk), |
7452 | &last_eb_bh); | |
ccd979bd MF |
7453 | if (status < 0) { |
7454 | mlog_errno(status); | |
7455 | goto bail; | |
7456 | } | |
7457 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
ccd979bd MF |
7458 | } |
7459 | ||
7460 | (*tc)->tc_last_eb_bh = last_eb_bh; | |
7461 | ||
ccd979bd MF |
7462 | status = 0; |
7463 | bail: | |
7464 | if (status < 0) { | |
7465 | if (*tc) | |
7466 | ocfs2_free_truncate_context(*tc); | |
7467 | *tc = NULL; | |
7468 | } | |
7469 | mlog_exit_void(); | |
7470 | return status; | |
7471 | } | |
7472 | ||
1afc32b9 MF |
7473 | /* |
7474 | * 'start' is inclusive, 'end' is not. | |
7475 | */ | |
7476 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7477 | unsigned int start, unsigned int end, int trunc) | |
7478 | { | |
7479 | int ret; | |
7480 | unsigned int numbytes; | |
7481 | handle_t *handle; | |
7482 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7483 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7484 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7485 | ||
7486 | if (end > i_size_read(inode)) | |
7487 | end = i_size_read(inode); | |
7488 | ||
7489 | BUG_ON(start >= end); | |
7490 | ||
7491 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7492 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7493 | !ocfs2_supports_inline_data(osb)) { | |
7494 | ocfs2_error(inode->i_sb, | |
7495 | "Inline data flags for inode %llu don't agree! " | |
7496 | "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", | |
7497 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7498 | le16_to_cpu(di->i_dyn_features), | |
7499 | OCFS2_I(inode)->ip_dyn_features, | |
7500 | osb->s_feature_incompat); | |
7501 | ret = -EROFS; | |
7502 | goto out; | |
7503 | } | |
7504 | ||
7505 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7506 | if (IS_ERR(handle)) { | |
7507 | ret = PTR_ERR(handle); | |
7508 | mlog_errno(ret); | |
7509 | goto out; | |
7510 | } | |
7511 | ||
0cf2f763 | 7512 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7513 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7514 | if (ret) { |
7515 | mlog_errno(ret); | |
7516 | goto out_commit; | |
7517 | } | |
7518 | ||
7519 | numbytes = end - start; | |
7520 | memset(idata->id_data + start, 0, numbytes); | |
7521 | ||
7522 | /* | |
7523 | * No need to worry about the data page here - it's been | |
7524 | * truncated already and inline data doesn't need it for | |
7525 | * pushing zero's to disk, so we'll let readpage pick it up | |
7526 | * later. | |
7527 | */ | |
7528 | if (trunc) { | |
7529 | i_size_write(inode, start); | |
7530 | di->i_size = cpu_to_le64(start); | |
7531 | } | |
7532 | ||
7533 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7534 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
7535 | ||
7536 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7537 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7538 | ||
7539 | ocfs2_journal_dirty(handle, di_bh); | |
7540 | ||
7541 | out_commit: | |
7542 | ocfs2_commit_trans(osb, handle); | |
7543 | ||
7544 | out: | |
7545 | return ret; | |
7546 | } | |
7547 | ||
ccd979bd MF |
7548 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc) |
7549 | { | |
59a5e416 MF |
7550 | /* |
7551 | * The caller is responsible for completing deallocation | |
7552 | * before freeing the context. | |
7553 | */ | |
7554 | if (tc->tc_dealloc.c_first_suballocator != NULL) | |
7555 | mlog(ML_NOTICE, | |
7556 | "Truncate completion has non-empty dealloc context\n"); | |
ccd979bd | 7557 | |
a81cb88b | 7558 | brelse(tc->tc_last_eb_bh); |
ccd979bd MF |
7559 | |
7560 | kfree(tc); | |
7561 | } |