Commit | Line | Data |
---|---|---|
328970de | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
ccd979bd MF |
2 | /* -*- mode: c; c-basic-offset: 8; -*- |
3 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
4 | * | |
5 | * localalloc.c | |
6 | * | |
7 | * Node local data allocation | |
8 | * | |
9 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
ccd979bd MF |
10 | */ |
11 | ||
12 | #include <linux/fs.h> | |
13 | #include <linux/types.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/highmem.h> | |
16 | #include <linux/bitops.h> | |
17 | ||
ccd979bd MF |
18 | #include <cluster/masklog.h> |
19 | ||
20 | #include "ocfs2.h" | |
21 | ||
22 | #include "alloc.h" | |
13723d00 | 23 | #include "blockcheck.h" |
ccd979bd MF |
24 | #include "dlmglue.h" |
25 | #include "inode.h" | |
26 | #include "journal.h" | |
27 | #include "localalloc.h" | |
28 | #include "suballoc.h" | |
29 | #include "super.h" | |
30 | #include "sysfile.h" | |
a04733d8 | 31 | #include "ocfs2_trace.h" |
ccd979bd MF |
32 | |
33 | #include "buffer_head_io.h" | |
34 | ||
35 | #define OCFS2_LOCAL_ALLOC(dinode) (&((dinode)->id2.i_lab)) | |
36 | ||
ccd979bd MF |
37 | static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc); |
38 | ||
39 | static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb, | |
40 | struct ocfs2_dinode *alloc, | |
d02f00cc MF |
41 | u32 *numbits, |
42 | struct ocfs2_alloc_reservation *resv); | |
ccd979bd MF |
43 | |
44 | static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc); | |
45 | ||
46 | static int ocfs2_sync_local_to_main(struct ocfs2_super *osb, | |
1fabe148 | 47 | handle_t *handle, |
ccd979bd MF |
48 | struct ocfs2_dinode *alloc, |
49 | struct inode *main_bm_inode, | |
50 | struct buffer_head *main_bm_bh); | |
51 | ||
52 | static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb, | |
ccd979bd MF |
53 | struct ocfs2_alloc_context **ac, |
54 | struct inode **bitmap_inode, | |
55 | struct buffer_head **bitmap_bh); | |
56 | ||
57 | static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb, | |
1fabe148 | 58 | handle_t *handle, |
ccd979bd MF |
59 | struct ocfs2_alloc_context *ac); |
60 | ||
61 | static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb, | |
62 | struct inode *local_alloc_inode); | |
63 | ||
6b82021b MF |
64 | /* |
65 | * ocfs2_la_default_mb() - determine a default size, in megabytes of | |
66 | * the local alloc. | |
67 | * | |
68 | * Generally, we'd like to pick as large a local alloc as | |
69 | * possible. Performance on large workloads tends to scale | |
70 | * proportionally to la size. In addition to that, the reservations | |
71 | * code functions more efficiently as it can reserve more windows for | |
72 | * write. | |
73 | * | |
74 | * Some things work against us when trying to choose a large local alloc: | |
75 | * | |
76 | * - We need to ensure our sizing is picked to leave enough space in | |
77 | * group descriptors for other allocations (such as block groups, | |
78 | * etc). Picking default sizes which are a multiple of 4 could help | |
79 | * - block groups are allocated in 2mb and 4mb chunks. | |
80 | * | |
81 | * - Likewise, we don't want to starve other nodes of bits on small | |
82 | * file systems. This can easily be taken care of by limiting our | |
83 | * default to a reasonable size (256M) on larger cluster sizes. | |
84 | * | |
85 | * - Some file systems can't support very large sizes - 4k and 8k in | |
86 | * particular are limited to less than 128 and 256 megabytes respectively. | |
87 | * | |
88 | * The following reference table shows group descriptor and local | |
89 | * alloc maximums at various cluster sizes (4k blocksize) | |
90 | * | |
91 | * csize: 4K group: 126M la: 121M | |
92 | * csize: 8K group: 252M la: 243M | |
93 | * csize: 16K group: 504M la: 486M | |
94 | * csize: 32K group: 1008M la: 972M | |
95 | * csize: 64K group: 2016M la: 1944M | |
96 | * csize: 128K group: 4032M la: 3888M | |
97 | * csize: 256K group: 8064M la: 7776M | |
98 | * csize: 512K group: 16128M la: 15552M | |
99 | * csize: 1024K group: 32256M la: 31104M | |
100 | */ | |
101 | #define OCFS2_LA_MAX_DEFAULT_MB 256 | |
102 | #define OCFS2_LA_OLD_DEFAULT 8 | |
103 | unsigned int ocfs2_la_default_mb(struct ocfs2_super *osb) | |
104 | { | |
105 | unsigned int la_mb; | |
106 | unsigned int gd_mb; | |
1739da40 | 107 | unsigned int la_max_mb; |
6b82021b MF |
108 | unsigned int megs_per_slot; |
109 | struct super_block *sb = osb->sb; | |
110 | ||
111 | gd_mb = ocfs2_clusters_to_megabytes(osb->sb, | |
8571882c | 112 | 8 * ocfs2_group_bitmap_size(sb, 0, osb->s_feature_incompat)); |
6b82021b MF |
113 | |
114 | /* | |
115 | * This takes care of files systems with very small group | |
116 | * descriptors - 512 byte blocksize at cluster sizes lower | |
117 | * than 16K and also 1k blocksize with 4k cluster size. | |
118 | */ | |
119 | if ((sb->s_blocksize == 512 && osb->s_clustersize <= 8192) | |
120 | || (sb->s_blocksize == 1024 && osb->s_clustersize == 4096)) | |
121 | return OCFS2_LA_OLD_DEFAULT; | |
122 | ||
123 | /* | |
124 | * Leave enough room for some block groups and make the final | |
125 | * value we work from a multiple of 4. | |
126 | */ | |
127 | gd_mb -= 16; | |
128 | gd_mb &= 0xFFFFFFFB; | |
129 | ||
130 | la_mb = gd_mb; | |
131 | ||
132 | /* | |
133 | * Keep window sizes down to a reasonable default | |
134 | */ | |
135 | if (la_mb > OCFS2_LA_MAX_DEFAULT_MB) { | |
136 | /* | |
137 | * Some clustersize / blocksize combinations will have | |
138 | * given us a larger than OCFS2_LA_MAX_DEFAULT_MB | |
139 | * default size, but get poor distribution when | |
140 | * limited to exactly 256 megabytes. | |
141 | * | |
142 | * As an example, 16K clustersize at 4K blocksize | |
143 | * gives us a cluster group size of 504M. Paring the | |
144 | * local alloc size down to 256 however, would give us | |
145 | * only one window and around 200MB left in the | |
146 | * cluster group. Instead, find the first size below | |
147 | * 256 which would give us an even distribution. | |
148 | * | |
149 | * Larger cluster group sizes actually work out pretty | |
150 | * well when pared to 256, so we don't have to do this | |
151 | * for any group that fits more than two | |
152 | * OCFS2_LA_MAX_DEFAULT_MB windows. | |
153 | */ | |
154 | if (gd_mb > (2 * OCFS2_LA_MAX_DEFAULT_MB)) | |
155 | la_mb = 256; | |
156 | else { | |
157 | unsigned int gd_mult = gd_mb; | |
158 | ||
159 | while (gd_mult > 256) | |
160 | gd_mult = gd_mult >> 1; | |
161 | ||
162 | la_mb = gd_mult; | |
163 | } | |
164 | } | |
165 | ||
166 | megs_per_slot = osb->osb_clusters_at_boot / osb->max_slots; | |
167 | megs_per_slot = ocfs2_clusters_to_megabytes(osb->sb, megs_per_slot); | |
168 | /* Too many nodes, too few disk clusters. */ | |
169 | if (megs_per_slot < la_mb) | |
170 | la_mb = megs_per_slot; | |
171 | ||
1739da40 TM |
172 | /* We can't store more bits than we can in a block. */ |
173 | la_max_mb = ocfs2_clusters_to_megabytes(osb->sb, | |
174 | ocfs2_local_alloc_size(sb) * 8); | |
175 | if (la_mb > la_max_mb) | |
176 | la_mb = la_max_mb; | |
177 | ||
6b82021b MF |
178 | return la_mb; |
179 | } | |
180 | ||
73c8a800 MF |
181 | void ocfs2_la_set_sizes(struct ocfs2_super *osb, int requested_mb) |
182 | { | |
183 | struct super_block *sb = osb->sb; | |
6b82021b | 184 | unsigned int la_default_mb = ocfs2_la_default_mb(osb); |
73c8a800 MF |
185 | unsigned int la_max_mb; |
186 | ||
187 | la_max_mb = ocfs2_clusters_to_megabytes(sb, | |
188 | ocfs2_local_alloc_size(sb) * 8); | |
189 | ||
a04733d8 | 190 | trace_ocfs2_la_set_sizes(requested_mb, la_max_mb, la_default_mb); |
73c8a800 MF |
191 | |
192 | if (requested_mb == -1) { | |
193 | /* No user request - use defaults */ | |
194 | osb->local_alloc_default_bits = | |
195 | ocfs2_megabytes_to_clusters(sb, la_default_mb); | |
196 | } else if (requested_mb > la_max_mb) { | |
197 | /* Request is too big, we give the maximum available */ | |
198 | osb->local_alloc_default_bits = | |
199 | ocfs2_megabytes_to_clusters(sb, la_max_mb); | |
200 | } else { | |
201 | osb->local_alloc_default_bits = | |
202 | ocfs2_megabytes_to_clusters(sb, requested_mb); | |
203 | } | |
204 | ||
205 | osb->local_alloc_bits = osb->local_alloc_default_bits; | |
206 | } | |
207 | ||
9c7af40b MF |
208 | static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb) |
209 | { | |
210 | return (osb->local_alloc_state == OCFS2_LA_THROTTLED || | |
211 | osb->local_alloc_state == OCFS2_LA_ENABLED); | |
212 | } | |
213 | ||
214 | void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb, | |
215 | unsigned int num_clusters) | |
216 | { | |
217 | spin_lock(&osb->osb_lock); | |
218 | if (osb->local_alloc_state == OCFS2_LA_DISABLED || | |
219 | osb->local_alloc_state == OCFS2_LA_THROTTLED) | |
220 | if (num_clusters >= osb->local_alloc_default_bits) { | |
221 | cancel_delayed_work(&osb->la_enable_wq); | |
222 | osb->local_alloc_state = OCFS2_LA_ENABLED; | |
223 | } | |
224 | spin_unlock(&osb->osb_lock); | |
225 | } | |
226 | ||
227 | void ocfs2_la_enable_worker(struct work_struct *work) | |
228 | { | |
229 | struct ocfs2_super *osb = | |
230 | container_of(work, struct ocfs2_super, | |
231 | la_enable_wq.work); | |
232 | spin_lock(&osb->osb_lock); | |
233 | osb->local_alloc_state = OCFS2_LA_ENABLED; | |
234 | spin_unlock(&osb->osb_lock); | |
235 | } | |
236 | ||
ccd979bd MF |
237 | /* |
238 | * Tell us whether a given allocation should use the local alloc | |
239 | * file. Otherwise, it has to go to the main bitmap. | |
9c7af40b MF |
240 | * |
241 | * This function does semi-dirty reads of local alloc size and state! | |
242 | * This is ok however, as the values are re-checked once under mutex. | |
ccd979bd MF |
243 | */ |
244 | int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits) | |
245 | { | |
2fbe8d1e | 246 | int ret = 0; |
9c7af40b MF |
247 | int la_bits; |
248 | ||
249 | spin_lock(&osb->osb_lock); | |
250 | la_bits = osb->local_alloc_bits; | |
ccd979bd | 251 | |
9c7af40b | 252 | if (!ocfs2_la_state_enabled(osb)) |
2fbe8d1e | 253 | goto bail; |
ccd979bd MF |
254 | |
255 | /* la_bits should be at least twice the size (in clusters) of | |
256 | * a new block group. We want to be sure block group | |
257 | * allocations go through the local alloc, so allow an | |
258 | * allocation to take up to half the bitmap. */ | |
259 | if (bits > (la_bits / 2)) | |
2fbe8d1e | 260 | goto bail; |
ccd979bd | 261 | |
2fbe8d1e SM |
262 | ret = 1; |
263 | bail: | |
a04733d8 TM |
264 | trace_ocfs2_alloc_should_use_local( |
265 | (unsigned long long)bits, osb->local_alloc_state, la_bits, ret); | |
9c7af40b | 266 | spin_unlock(&osb->osb_lock); |
2fbe8d1e | 267 | return ret; |
ccd979bd MF |
268 | } |
269 | ||
270 | int ocfs2_load_local_alloc(struct ocfs2_super *osb) | |
271 | { | |
272 | int status = 0; | |
273 | struct ocfs2_dinode *alloc = NULL; | |
274 | struct buffer_head *alloc_bh = NULL; | |
275 | u32 num_used; | |
276 | struct inode *inode = NULL; | |
277 | struct ocfs2_local_alloc *la; | |
278 | ||
ebcee4b5 | 279 | if (osb->local_alloc_bits == 0) |
2fbe8d1e SM |
280 | goto bail; |
281 | ||
ebcee4b5 | 282 | if (osb->local_alloc_bits >= osb->bitmap_cpg) { |
2fbe8d1e SM |
283 | mlog(ML_NOTICE, "Requested local alloc window %d is larger " |
284 | "than max possible %u. Using defaults.\n", | |
ebcee4b5 MF |
285 | osb->local_alloc_bits, (osb->bitmap_cpg - 1)); |
286 | osb->local_alloc_bits = | |
287 | ocfs2_megabytes_to_clusters(osb->sb, | |
6b82021b | 288 | ocfs2_la_default_mb(osb)); |
2fbe8d1e SM |
289 | } |
290 | ||
ccd979bd MF |
291 | /* read the alloc off disk */ |
292 | inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE, | |
293 | osb->slot_num); | |
294 | if (!inode) { | |
295 | status = -EINVAL; | |
296 | mlog_errno(status); | |
297 | goto bail; | |
298 | } | |
299 | ||
b657c95c JB |
300 | status = ocfs2_read_inode_block_full(inode, &alloc_bh, |
301 | OCFS2_BH_IGNORE_CACHE); | |
ccd979bd MF |
302 | if (status < 0) { |
303 | mlog_errno(status); | |
304 | goto bail; | |
305 | } | |
306 | ||
307 | alloc = (struct ocfs2_dinode *) alloc_bh->b_data; | |
308 | la = OCFS2_LOCAL_ALLOC(alloc); | |
309 | ||
310 | if (!(le32_to_cpu(alloc->i_flags) & | |
311 | (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) { | |
b0697053 MF |
312 | mlog(ML_ERROR, "Invalid local alloc inode, %llu\n", |
313 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
314 | status = -EINVAL; |
315 | goto bail; | |
316 | } | |
317 | ||
318 | if ((la->la_size == 0) || | |
319 | (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) { | |
320 | mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n", | |
321 | le16_to_cpu(la->la_size)); | |
322 | status = -EINVAL; | |
323 | goto bail; | |
324 | } | |
325 | ||
326 | /* do a little verification. */ | |
327 | num_used = ocfs2_local_alloc_count_bits(alloc); | |
328 | ||
329 | /* hopefully the local alloc has always been recovered before | |
330 | * we load it. */ | |
331 | if (num_used | |
332 | || alloc->id1.bitmap1.i_used | |
333 | || alloc->id1.bitmap1.i_total | |
532e1e54 JB |
334 | || la->la_bm_off) { |
335 | mlog(ML_ERROR, "inconsistent detected, clean journal with" | |
336 | " unrecovered local alloc, please run fsck.ocfs2!\n" | |
ccd979bd MF |
337 | "found = %u, set = %u, taken = %u, off = %u\n", |
338 | num_used, le32_to_cpu(alloc->id1.bitmap1.i_used), | |
339 | le32_to_cpu(alloc->id1.bitmap1.i_total), | |
340 | OCFS2_LOCAL_ALLOC(alloc)->la_bm_off); | |
341 | ||
532e1e54 JB |
342 | status = -EINVAL; |
343 | goto bail; | |
344 | } | |
345 | ||
ccd979bd MF |
346 | osb->local_alloc_bh = alloc_bh; |
347 | osb->local_alloc_state = OCFS2_LA_ENABLED; | |
348 | ||
349 | bail: | |
350 | if (status < 0) | |
a81cb88b | 351 | brelse(alloc_bh); |
72865d92 | 352 | iput(inode); |
ccd979bd | 353 | |
a04733d8 | 354 | trace_ocfs2_load_local_alloc(osb->local_alloc_bits); |
2fbe8d1e | 355 | |
c1e8d35e TM |
356 | if (status) |
357 | mlog_errno(status); | |
ccd979bd MF |
358 | return status; |
359 | } | |
360 | ||
361 | /* | |
362 | * return any unused bits to the bitmap and write out a clean | |
363 | * local_alloc. | |
364 | * | |
365 | * local_alloc_bh is optional. If not passed, we will simply use the | |
366 | * one off osb. If you do pass it however, be warned that it *will* be | |
367 | * returned brelse'd and NULL'd out.*/ | |
368 | void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb) | |
369 | { | |
370 | int status; | |
1fabe148 | 371 | handle_t *handle; |
ccd979bd MF |
372 | struct inode *local_alloc_inode = NULL; |
373 | struct buffer_head *bh = NULL; | |
374 | struct buffer_head *main_bm_bh = NULL; | |
375 | struct inode *main_bm_inode = NULL; | |
376 | struct ocfs2_dinode *alloc_copy = NULL; | |
377 | struct ocfs2_dinode *alloc = NULL; | |
378 | ||
9c7af40b | 379 | cancel_delayed_work(&osb->la_enable_wq); |
35ddf78e | 380 | flush_workqueue(osb->ocfs2_wq); |
9c7af40b | 381 | |
ccd979bd | 382 | if (osb->local_alloc_state == OCFS2_LA_UNUSED) |
8898a5a5 | 383 | goto out; |
ccd979bd MF |
384 | |
385 | local_alloc_inode = | |
386 | ocfs2_get_system_file_inode(osb, | |
387 | LOCAL_ALLOC_SYSTEM_INODE, | |
388 | osb->slot_num); | |
389 | if (!local_alloc_inode) { | |
390 | status = -ENOENT; | |
391 | mlog_errno(status); | |
8898a5a5 | 392 | goto out; |
ccd979bd MF |
393 | } |
394 | ||
395 | osb->local_alloc_state = OCFS2_LA_DISABLED; | |
396 | ||
d02f00cc MF |
397 | ocfs2_resmap_uninit(&osb->osb_la_resmap); |
398 | ||
ccd979bd MF |
399 | main_bm_inode = ocfs2_get_system_file_inode(osb, |
400 | GLOBAL_BITMAP_SYSTEM_INODE, | |
401 | OCFS2_INVALID_SLOT); | |
402 | if (!main_bm_inode) { | |
403 | status = -EINVAL; | |
404 | mlog_errno(status); | |
8898a5a5 | 405 | goto out; |
ccd979bd MF |
406 | } |
407 | ||
5955102c | 408 | inode_lock(main_bm_inode); |
8898a5a5 | 409 | |
e63aecb6 | 410 | status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); |
ccd979bd MF |
411 | if (status < 0) { |
412 | mlog_errno(status); | |
8898a5a5 | 413 | goto out_mutex; |
ccd979bd MF |
414 | } |
415 | ||
416 | /* WINDOW_MOVE_CREDITS is a bit heavy... */ | |
65eff9cc | 417 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
418 | if (IS_ERR(handle)) { |
419 | mlog_errno(PTR_ERR(handle)); | |
420 | handle = NULL; | |
8898a5a5 | 421 | goto out_unlock; |
ccd979bd MF |
422 | } |
423 | ||
424 | bh = osb->local_alloc_bh; | |
425 | alloc = (struct ocfs2_dinode *) bh->b_data; | |
426 | ||
d8b2fa65 | 427 | alloc_copy = kmemdup(alloc, bh->b_size, GFP_NOFS); |
ccd979bd MF |
428 | if (!alloc_copy) { |
429 | status = -ENOMEM; | |
8898a5a5 | 430 | goto out_commit; |
ccd979bd | 431 | } |
ccd979bd | 432 | |
0cf2f763 JB |
433 | status = ocfs2_journal_access_di(handle, INODE_CACHE(local_alloc_inode), |
434 | bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
435 | if (status < 0) { |
436 | mlog_errno(status); | |
8898a5a5 | 437 | goto out_commit; |
ccd979bd MF |
438 | } |
439 | ||
440 | ocfs2_clear_local_alloc(alloc); | |
ec20cec7 | 441 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
442 | |
443 | brelse(bh); | |
444 | osb->local_alloc_bh = NULL; | |
445 | osb->local_alloc_state = OCFS2_LA_UNUSED; | |
446 | ||
447 | status = ocfs2_sync_local_to_main(osb, handle, alloc_copy, | |
448 | main_bm_inode, main_bm_bh); | |
449 | if (status < 0) | |
450 | mlog_errno(status); | |
451 | ||
8898a5a5 | 452 | out_commit: |
02dc1af4 | 453 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 454 | |
8898a5a5 | 455 | out_unlock: |
a81cb88b | 456 | brelse(main_bm_bh); |
ccd979bd | 457 | |
e63aecb6 | 458 | ocfs2_inode_unlock(main_bm_inode, 1); |
ccd979bd | 459 | |
8898a5a5 | 460 | out_mutex: |
5955102c | 461 | inode_unlock(main_bm_inode); |
8898a5a5 MF |
462 | iput(main_bm_inode); |
463 | ||
464 | out: | |
72865d92 | 465 | iput(local_alloc_inode); |
ccd979bd | 466 | |
d787ab09 | 467 | kfree(alloc_copy); |
ccd979bd MF |
468 | } |
469 | ||
470 | /* | |
471 | * We want to free the bitmap bits outside of any recovery context as | |
472 | * we'll need a cluster lock to do so, but we must clear the local | |
473 | * alloc before giving up the recovered nodes journal. To solve this, | |
474 | * we kmalloc a copy of the local alloc before it's change for the | |
475 | * caller to process with ocfs2_complete_local_alloc_recovery | |
476 | */ | |
477 | int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb, | |
478 | int slot_num, | |
479 | struct ocfs2_dinode **alloc_copy) | |
480 | { | |
481 | int status = 0; | |
482 | struct buffer_head *alloc_bh = NULL; | |
483 | struct inode *inode = NULL; | |
484 | struct ocfs2_dinode *alloc; | |
485 | ||
a04733d8 | 486 | trace_ocfs2_begin_local_alloc_recovery(slot_num); |
ccd979bd MF |
487 | |
488 | *alloc_copy = NULL; | |
489 | ||
490 | inode = ocfs2_get_system_file_inode(osb, | |
491 | LOCAL_ALLOC_SYSTEM_INODE, | |
492 | slot_num); | |
493 | if (!inode) { | |
494 | status = -EINVAL; | |
495 | mlog_errno(status); | |
496 | goto bail; | |
497 | } | |
498 | ||
5955102c | 499 | inode_lock(inode); |
ccd979bd | 500 | |
b657c95c JB |
501 | status = ocfs2_read_inode_block_full(inode, &alloc_bh, |
502 | OCFS2_BH_IGNORE_CACHE); | |
ccd979bd MF |
503 | if (status < 0) { |
504 | mlog_errno(status); | |
505 | goto bail; | |
506 | } | |
507 | ||
508 | *alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL); | |
509 | if (!(*alloc_copy)) { | |
510 | status = -ENOMEM; | |
511 | goto bail; | |
512 | } | |
513 | memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size); | |
514 | ||
515 | alloc = (struct ocfs2_dinode *) alloc_bh->b_data; | |
516 | ocfs2_clear_local_alloc(alloc); | |
517 | ||
13723d00 | 518 | ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check); |
8cb471e8 | 519 | status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode)); |
ccd979bd MF |
520 | if (status < 0) |
521 | mlog_errno(status); | |
522 | ||
523 | bail: | |
d787ab09 | 524 | if (status < 0) { |
ccd979bd MF |
525 | kfree(*alloc_copy); |
526 | *alloc_copy = NULL; | |
527 | } | |
528 | ||
a81cb88b | 529 | brelse(alloc_bh); |
ccd979bd MF |
530 | |
531 | if (inode) { | |
5955102c | 532 | inode_unlock(inode); |
ccd979bd MF |
533 | iput(inode); |
534 | } | |
535 | ||
c1e8d35e TM |
536 | if (status) |
537 | mlog_errno(status); | |
ccd979bd MF |
538 | return status; |
539 | } | |
540 | ||
541 | /* | |
542 | * Step 2: By now, we've completed the journal recovery, we've stamped | |
543 | * a clean local alloc on disk and dropped the node out of the | |
544 | * recovery map. Dlm locks will no longer stall, so lets clear out the | |
545 | * main bitmap. | |
546 | */ | |
547 | int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb, | |
548 | struct ocfs2_dinode *alloc) | |
549 | { | |
550 | int status; | |
1fabe148 | 551 | handle_t *handle; |
ccd979bd | 552 | struct buffer_head *main_bm_bh = NULL; |
8898a5a5 | 553 | struct inode *main_bm_inode; |
ccd979bd | 554 | |
ccd979bd MF |
555 | main_bm_inode = ocfs2_get_system_file_inode(osb, |
556 | GLOBAL_BITMAP_SYSTEM_INODE, | |
557 | OCFS2_INVALID_SLOT); | |
558 | if (!main_bm_inode) { | |
559 | status = -EINVAL; | |
560 | mlog_errno(status); | |
8898a5a5 | 561 | goto out; |
ccd979bd MF |
562 | } |
563 | ||
5955102c | 564 | inode_lock(main_bm_inode); |
8898a5a5 | 565 | |
e63aecb6 | 566 | status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1); |
ccd979bd MF |
567 | if (status < 0) { |
568 | mlog_errno(status); | |
8898a5a5 | 569 | goto out_mutex; |
ccd979bd MF |
570 | } |
571 | ||
65eff9cc | 572 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
573 | if (IS_ERR(handle)) { |
574 | status = PTR_ERR(handle); | |
575 | handle = NULL; | |
576 | mlog_errno(status); | |
8898a5a5 | 577 | goto out_unlock; |
ccd979bd MF |
578 | } |
579 | ||
580 | /* we want the bitmap change to be recorded on disk asap */ | |
1fabe148 | 581 | handle->h_sync = 1; |
ccd979bd MF |
582 | |
583 | status = ocfs2_sync_local_to_main(osb, handle, alloc, | |
584 | main_bm_inode, main_bm_bh); | |
585 | if (status < 0) | |
586 | mlog_errno(status); | |
587 | ||
02dc1af4 | 588 | ocfs2_commit_trans(osb, handle); |
8898a5a5 MF |
589 | |
590 | out_unlock: | |
e63aecb6 | 591 | ocfs2_inode_unlock(main_bm_inode, 1); |
8898a5a5 MF |
592 | |
593 | out_mutex: | |
5955102c | 594 | inode_unlock(main_bm_inode); |
ccd979bd | 595 | |
a81cb88b | 596 | brelse(main_bm_bh); |
ccd979bd | 597 | |
8898a5a5 | 598 | iput(main_bm_inode); |
ccd979bd | 599 | |
8898a5a5 | 600 | out: |
4d0ddb2c | 601 | if (!status) |
b89c5428 | 602 | ocfs2_init_steal_slots(osb); |
c1e8d35e TM |
603 | if (status) |
604 | mlog_errno(status); | |
ccd979bd MF |
605 | return status; |
606 | } | |
607 | ||
608 | /* | |
9c7af40b | 609 | * make sure we've got at least bits_wanted contiguous bits in the |
1b1dcc1b | 610 | * local alloc. You lose them when you drop i_mutex. |
ccd979bd MF |
611 | * |
612 | * We will add ourselves to the transaction passed in, but may start | |
613 | * our own in order to shift windows. | |
614 | */ | |
615 | int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb, | |
ccd979bd MF |
616 | u32 bits_wanted, |
617 | struct ocfs2_alloc_context *ac) | |
618 | { | |
619 | int status; | |
620 | struct ocfs2_dinode *alloc; | |
621 | struct inode *local_alloc_inode; | |
622 | unsigned int free_bits; | |
623 | ||
ccd979bd | 624 | BUG_ON(!ac); |
ccd979bd MF |
625 | |
626 | local_alloc_inode = | |
627 | ocfs2_get_system_file_inode(osb, | |
628 | LOCAL_ALLOC_SYSTEM_INODE, | |
629 | osb->slot_num); | |
630 | if (!local_alloc_inode) { | |
631 | status = -ENOENT; | |
632 | mlog_errno(status); | |
633 | goto bail; | |
634 | } | |
da5cbf2f | 635 | |
5955102c | 636 | inode_lock(local_alloc_inode); |
da5cbf2f | 637 | |
9c7af40b MF |
638 | /* |
639 | * We must double check state and allocator bits because | |
640 | * another process may have changed them while holding i_mutex. | |
641 | */ | |
642 | spin_lock(&osb->osb_lock); | |
643 | if (!ocfs2_la_state_enabled(osb) || | |
644 | (bits_wanted > osb->local_alloc_bits)) { | |
645 | spin_unlock(&osb->osb_lock); | |
ccd979bd MF |
646 | status = -ENOSPC; |
647 | goto bail; | |
648 | } | |
9c7af40b | 649 | spin_unlock(&osb->osb_lock); |
ccd979bd MF |
650 | |
651 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
652 | ||
e407e397 | 653 | #ifdef CONFIG_OCFS2_DEBUG_FS |
ccd979bd MF |
654 | if (le32_to_cpu(alloc->id1.bitmap1.i_used) != |
655 | ocfs2_local_alloc_count_bits(alloc)) { | |
93f5920d JP |
656 | status = ocfs2_error(osb->sb, "local alloc inode %llu says it has %u used bits, but a count shows %u\n", |
657 | (unsigned long long)le64_to_cpu(alloc->i_blkno), | |
658 | le32_to_cpu(alloc->id1.bitmap1.i_used), | |
659 | ocfs2_local_alloc_count_bits(alloc)); | |
ccd979bd MF |
660 | goto bail; |
661 | } | |
5a58c3ef | 662 | #endif |
ccd979bd MF |
663 | |
664 | free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) - | |
665 | le32_to_cpu(alloc->id1.bitmap1.i_used); | |
666 | if (bits_wanted > free_bits) { | |
667 | /* uhoh, window change time. */ | |
668 | status = | |
669 | ocfs2_local_alloc_slide_window(osb, local_alloc_inode); | |
670 | if (status < 0) { | |
671 | if (status != -ENOSPC) | |
672 | mlog_errno(status); | |
673 | goto bail; | |
674 | } | |
9c7af40b MF |
675 | |
676 | /* | |
677 | * Under certain conditions, the window slide code | |
678 | * might have reduced the number of bits available or | |
679 | * disabled the the local alloc entirely. Re-check | |
680 | * here and return -ENOSPC if necessary. | |
681 | */ | |
682 | status = -ENOSPC; | |
683 | if (!ocfs2_la_state_enabled(osb)) | |
684 | goto bail; | |
685 | ||
686 | free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) - | |
687 | le32_to_cpu(alloc->id1.bitmap1.i_used); | |
688 | if (bits_wanted > free_bits) | |
689 | goto bail; | |
ccd979bd MF |
690 | } |
691 | ||
8fccfc82 | 692 | ac->ac_inode = local_alloc_inode; |
a4a48911 TM |
693 | /* We should never use localalloc from another slot */ |
694 | ac->ac_alloc_slot = osb->slot_num; | |
8fccfc82 | 695 | ac->ac_which = OCFS2_AC_USE_LOCAL; |
ccd979bd MF |
696 | get_bh(osb->local_alloc_bh); |
697 | ac->ac_bh = osb->local_alloc_bh; | |
ccd979bd MF |
698 | status = 0; |
699 | bail: | |
bda0233b | 700 | if (status < 0 && local_alloc_inode) { |
5955102c | 701 | inode_unlock(local_alloc_inode); |
8fccfc82 | 702 | iput(local_alloc_inode); |
bda0233b | 703 | } |
ccd979bd | 704 | |
a04733d8 TM |
705 | trace_ocfs2_reserve_local_alloc_bits( |
706 | (unsigned long long)ac->ac_max_block, | |
707 | bits_wanted, osb->slot_num, status); | |
2fbe8d1e | 708 | |
c1e8d35e TM |
709 | if (status) |
710 | mlog_errno(status); | |
ccd979bd MF |
711 | return status; |
712 | } | |
713 | ||
714 | int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb, | |
1fabe148 | 715 | handle_t *handle, |
ccd979bd | 716 | struct ocfs2_alloc_context *ac, |
415cb800 | 717 | u32 bits_wanted, |
ccd979bd MF |
718 | u32 *bit_off, |
719 | u32 *num_bits) | |
720 | { | |
721 | int status, start; | |
722 | struct inode *local_alloc_inode; | |
ccd979bd MF |
723 | void *bitmap; |
724 | struct ocfs2_dinode *alloc; | |
725 | struct ocfs2_local_alloc *la; | |
726 | ||
ccd979bd MF |
727 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL); |
728 | ||
ccd979bd MF |
729 | local_alloc_inode = ac->ac_inode; |
730 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
731 | la = OCFS2_LOCAL_ALLOC(alloc); | |
732 | ||
d02f00cc MF |
733 | start = ocfs2_local_alloc_find_clear_bits(osb, alloc, &bits_wanted, |
734 | ac->ac_resv); | |
ccd979bd MF |
735 | if (start == -1) { |
736 | /* TODO: Shouldn't we just BUG here? */ | |
737 | status = -ENOSPC; | |
738 | mlog_errno(status); | |
739 | goto bail; | |
740 | } | |
741 | ||
742 | bitmap = la->la_bitmap; | |
743 | *bit_off = le32_to_cpu(la->la_bm_off) + start; | |
ccd979bd MF |
744 | *num_bits = bits_wanted; |
745 | ||
0cf2f763 JB |
746 | status = ocfs2_journal_access_di(handle, |
747 | INODE_CACHE(local_alloc_inode), | |
13723d00 JB |
748 | osb->local_alloc_bh, |
749 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
750 | if (status < 0) { |
751 | mlog_errno(status); | |
752 | goto bail; | |
753 | } | |
754 | ||
d02f00cc MF |
755 | ocfs2_resmap_claimed_bits(&osb->osb_la_resmap, ac->ac_resv, start, |
756 | bits_wanted); | |
757 | ||
ccd979bd MF |
758 | while(bits_wanted--) |
759 | ocfs2_set_bit(start++, bitmap); | |
760 | ||
0dd3256e | 761 | le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits); |
ec20cec7 | 762 | ocfs2_journal_dirty(handle, osb->local_alloc_bh); |
ccd979bd | 763 | |
ccd979bd | 764 | bail: |
c1e8d35e TM |
765 | if (status) |
766 | mlog_errno(status); | |
ccd979bd MF |
767 | return status; |
768 | } | |
769 | ||
fb951eb5 ZW |
770 | int ocfs2_free_local_alloc_bits(struct ocfs2_super *osb, |
771 | handle_t *handle, | |
772 | struct ocfs2_alloc_context *ac, | |
773 | u32 bit_off, | |
774 | u32 num_bits) | |
775 | { | |
776 | int status, start; | |
777 | u32 clear_bits; | |
778 | struct inode *local_alloc_inode; | |
779 | void *bitmap; | |
780 | struct ocfs2_dinode *alloc; | |
781 | struct ocfs2_local_alloc *la; | |
782 | ||
783 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL); | |
784 | ||
785 | local_alloc_inode = ac->ac_inode; | |
786 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
787 | la = OCFS2_LOCAL_ALLOC(alloc); | |
788 | ||
789 | bitmap = la->la_bitmap; | |
790 | start = bit_off - le32_to_cpu(la->la_bm_off); | |
791 | clear_bits = num_bits; | |
792 | ||
793 | status = ocfs2_journal_access_di(handle, | |
794 | INODE_CACHE(local_alloc_inode), | |
795 | osb->local_alloc_bh, | |
796 | OCFS2_JOURNAL_ACCESS_WRITE); | |
797 | if (status < 0) { | |
798 | mlog_errno(status); | |
799 | goto bail; | |
800 | } | |
801 | ||
802 | while (clear_bits--) | |
803 | ocfs2_clear_bit(start++, bitmap); | |
804 | ||
805 | le32_add_cpu(&alloc->id1.bitmap1.i_used, -num_bits); | |
806 | ocfs2_journal_dirty(handle, osb->local_alloc_bh); | |
807 | ||
808 | bail: | |
809 | return status; | |
810 | } | |
811 | ||
ccd979bd MF |
812 | static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc) |
813 | { | |
a75613ec | 814 | u32 count; |
ccd979bd MF |
815 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); |
816 | ||
a75613ec | 817 | count = memweight(la->la_bitmap, le16_to_cpu(la->la_size)); |
ccd979bd | 818 | |
a04733d8 | 819 | trace_ocfs2_local_alloc_count_bits(count); |
ccd979bd MF |
820 | return count; |
821 | } | |
822 | ||
823 | static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb, | |
d02f00cc MF |
824 | struct ocfs2_dinode *alloc, |
825 | u32 *numbits, | |
826 | struct ocfs2_alloc_reservation *resv) | |
ccd979bd | 827 | { |
dec5b0d4 | 828 | int numfound = 0, bitoff, left, startoff; |
d02f00cc MF |
829 | int local_resv = 0; |
830 | struct ocfs2_alloc_reservation r; | |
ccd979bd | 831 | void *bitmap = NULL; |
d02f00cc | 832 | struct ocfs2_reservation_map *resmap = &osb->osb_la_resmap; |
ccd979bd | 833 | |
ccd979bd | 834 | if (!alloc->id1.bitmap1.i_total) { |
ccd979bd MF |
835 | bitoff = -1; |
836 | goto bail; | |
837 | } | |
838 | ||
d02f00cc MF |
839 | if (!resv) { |
840 | local_resv = 1; | |
841 | ocfs2_resv_init_once(&r); | |
842 | ocfs2_resv_set_type(&r, OCFS2_RESV_FLAG_TMP); | |
843 | resv = &r; | |
844 | } | |
845 | ||
846 | numfound = *numbits; | |
847 | if (ocfs2_resmap_resv_bits(resmap, resv, &bitoff, &numfound) == 0) { | |
848 | if (numfound < *numbits) | |
849 | *numbits = numfound; | |
850 | goto bail; | |
851 | } | |
852 | ||
853 | /* | |
854 | * Code error. While reservations are enabled, local | |
855 | * allocation should _always_ go through them. | |
856 | */ | |
857 | BUG_ON(osb->osb_resv_level != 0); | |
858 | ||
859 | /* | |
860 | * Reservations are disabled. Handle this the old way. | |
861 | */ | |
862 | ||
ccd979bd MF |
863 | bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap; |
864 | ||
865 | numfound = bitoff = startoff = 0; | |
ccd979bd MF |
866 | left = le32_to_cpu(alloc->id1.bitmap1.i_total); |
867 | while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) { | |
868 | if (bitoff == left) { | |
869 | /* mlog(0, "bitoff (%d) == left", bitoff); */ | |
870 | break; | |
871 | } | |
872 | /* mlog(0, "Found a zero: bitoff = %d, startoff = %d, " | |
873 | "numfound = %d\n", bitoff, startoff, numfound);*/ | |
874 | ||
875 | /* Ok, we found a zero bit... is it contig. or do we | |
876 | * start over?*/ | |
877 | if (bitoff == startoff) { | |
878 | /* we found a zero */ | |
879 | numfound++; | |
880 | startoff++; | |
881 | } else { | |
882 | /* got a zero after some ones */ | |
883 | numfound = 1; | |
884 | startoff = bitoff+1; | |
885 | } | |
886 | /* we got everything we needed */ | |
d02f00cc | 887 | if (numfound == *numbits) { |
ccd979bd MF |
888 | /* mlog(0, "Found it all!\n"); */ |
889 | break; | |
890 | } | |
891 | } | |
892 | ||
a04733d8 | 893 | trace_ocfs2_local_alloc_find_clear_bits_search_bitmap(bitoff, numfound); |
ccd979bd | 894 | |
3e4218df | 895 | if (numfound == *numbits) |
ccd979bd | 896 | bitoff = startoff - numfound; |
3e4218df | 897 | else |
ccd979bd MF |
898 | bitoff = -1; |
899 | ||
900 | bail: | |
d02f00cc MF |
901 | if (local_resv) |
902 | ocfs2_resv_discard(resmap, resv); | |
903 | ||
a04733d8 TM |
904 | trace_ocfs2_local_alloc_find_clear_bits(*numbits, |
905 | le32_to_cpu(alloc->id1.bitmap1.i_total), | |
906 | bitoff, numfound); | |
907 | ||
ccd979bd MF |
908 | return bitoff; |
909 | } | |
910 | ||
911 | static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc) | |
912 | { | |
913 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | |
914 | int i; | |
ccd979bd MF |
915 | |
916 | alloc->id1.bitmap1.i_total = 0; | |
917 | alloc->id1.bitmap1.i_used = 0; | |
918 | la->la_bm_off = 0; | |
919 | for(i = 0; i < le16_to_cpu(la->la_size); i++) | |
920 | la->la_bitmap[i] = 0; | |
ccd979bd MF |
921 | } |
922 | ||
923 | #if 0 | |
924 | /* turn this on and uncomment below to aid debugging window shifts. */ | |
925 | static void ocfs2_verify_zero_bits(unsigned long *bitmap, | |
926 | unsigned int start, | |
927 | unsigned int count) | |
928 | { | |
929 | unsigned int tmp = count; | |
930 | while(tmp--) { | |
931 | if (ocfs2_test_bit(start + tmp, bitmap)) { | |
932 | printk("ocfs2_verify_zero_bits: start = %u, count = " | |
933 | "%u\n", start, count); | |
934 | printk("ocfs2_verify_zero_bits: bit %u is set!", | |
935 | start + tmp); | |
936 | BUG(); | |
937 | } | |
938 | } | |
939 | } | |
940 | #endif | |
941 | ||
942 | /* | |
943 | * sync the local alloc to main bitmap. | |
944 | * | |
945 | * assumes you've already locked the main bitmap -- the bitmap inode | |
946 | * passed is used for caching. | |
947 | */ | |
948 | static int ocfs2_sync_local_to_main(struct ocfs2_super *osb, | |
1fabe148 | 949 | handle_t *handle, |
ccd979bd MF |
950 | struct ocfs2_dinode *alloc, |
951 | struct inode *main_bm_inode, | |
952 | struct buffer_head *main_bm_bh) | |
953 | { | |
954 | int status = 0; | |
955 | int bit_off, left, count, start; | |
956 | u64 la_start_blk; | |
957 | u64 blkno; | |
958 | void *bitmap; | |
959 | struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc); | |
960 | ||
a04733d8 | 961 | trace_ocfs2_sync_local_to_main( |
ef6b689b TM |
962 | le32_to_cpu(alloc->id1.bitmap1.i_total), |
963 | le32_to_cpu(alloc->id1.bitmap1.i_used)); | |
ccd979bd MF |
964 | |
965 | if (!alloc->id1.bitmap1.i_total) { | |
ccd979bd MF |
966 | goto bail; |
967 | } | |
968 | ||
969 | if (le32_to_cpu(alloc->id1.bitmap1.i_used) == | |
970 | le32_to_cpu(alloc->id1.bitmap1.i_total)) { | |
ccd979bd MF |
971 | goto bail; |
972 | } | |
973 | ||
974 | la_start_blk = ocfs2_clusters_to_blocks(osb->sb, | |
975 | le32_to_cpu(la->la_bm_off)); | |
976 | bitmap = la->la_bitmap; | |
977 | start = count = bit_off = 0; | |
978 | left = le32_to_cpu(alloc->id1.bitmap1.i_total); | |
979 | ||
980 | while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start)) | |
981 | != -1) { | |
982 | if ((bit_off < left) && (bit_off == start)) { | |
983 | count++; | |
984 | start++; | |
985 | continue; | |
986 | } | |
987 | if (count) { | |
988 | blkno = la_start_blk + | |
989 | ocfs2_clusters_to_blocks(osb->sb, | |
990 | start - count); | |
991 | ||
a04733d8 | 992 | trace_ocfs2_sync_local_to_main_free( |
b0697053 MF |
993 | count, start - count, |
994 | (unsigned long long)la_start_blk, | |
995 | (unsigned long long)blkno); | |
ccd979bd | 996 | |
b4414eea MF |
997 | status = ocfs2_release_clusters(handle, |
998 | main_bm_inode, | |
999 | main_bm_bh, blkno, | |
1000 | count); | |
ccd979bd MF |
1001 | if (status < 0) { |
1002 | mlog_errno(status); | |
1003 | goto bail; | |
1004 | } | |
1005 | } | |
1006 | if (bit_off >= left) | |
1007 | break; | |
1008 | count = 1; | |
1009 | start = bit_off + 1; | |
1010 | } | |
1011 | ||
1012 | bail: | |
c1e8d35e TM |
1013 | if (status) |
1014 | mlog_errno(status); | |
ccd979bd MF |
1015 | return status; |
1016 | } | |
1017 | ||
9c7af40b MF |
1018 | enum ocfs2_la_event { |
1019 | OCFS2_LA_EVENT_SLIDE, /* Normal window slide. */ | |
1020 | OCFS2_LA_EVENT_FRAGMENTED, /* The global bitmap has | |
1021 | * enough bits theoretically | |
1022 | * free, but a contiguous | |
1023 | * allocation could not be | |
1024 | * found. */ | |
1025 | OCFS2_LA_EVENT_ENOSPC, /* Global bitmap doesn't have | |
1026 | * enough bits free to satisfy | |
1027 | * our request. */ | |
1028 | }; | |
1029 | #define OCFS2_LA_ENABLE_INTERVAL (30 * HZ) | |
1030 | /* | |
1031 | * Given an event, calculate the size of our next local alloc window. | |
1032 | * | |
1033 | * This should always be called under i_mutex of the local alloc inode | |
1034 | * so that local alloc disabling doesn't race with processes trying to | |
1035 | * use the allocator. | |
1036 | * | |
1037 | * Returns the state which the local alloc was left in. This value can | |
1038 | * be ignored by some paths. | |
1039 | */ | |
1040 | static int ocfs2_recalc_la_window(struct ocfs2_super *osb, | |
1041 | enum ocfs2_la_event event) | |
1042 | { | |
1043 | unsigned int bits; | |
1044 | int state; | |
1045 | ||
1046 | spin_lock(&osb->osb_lock); | |
1047 | if (osb->local_alloc_state == OCFS2_LA_DISABLED) { | |
1048 | WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED); | |
1049 | goto out_unlock; | |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * ENOSPC and fragmentation are treated similarly for now. | |
1054 | */ | |
1055 | if (event == OCFS2_LA_EVENT_ENOSPC || | |
1056 | event == OCFS2_LA_EVENT_FRAGMENTED) { | |
1057 | /* | |
1058 | * We ran out of contiguous space in the primary | |
1059 | * bitmap. Drastically reduce the number of bits used | |
1060 | * by local alloc until we have to disable it. | |
1061 | */ | |
1062 | bits = osb->local_alloc_bits >> 1; | |
1063 | if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) { | |
1064 | /* | |
1065 | * By setting state to THROTTLED, we'll keep | |
1066 | * the number of local alloc bits used down | |
1067 | * until an event occurs which would give us | |
1068 | * reason to assume the bitmap situation might | |
1069 | * have changed. | |
1070 | */ | |
1071 | osb->local_alloc_state = OCFS2_LA_THROTTLED; | |
1072 | osb->local_alloc_bits = bits; | |
1073 | } else { | |
1074 | osb->local_alloc_state = OCFS2_LA_DISABLED; | |
1075 | } | |
35ddf78e | 1076 | queue_delayed_work(osb->ocfs2_wq, &osb->la_enable_wq, |
9c7af40b MF |
1077 | OCFS2_LA_ENABLE_INTERVAL); |
1078 | goto out_unlock; | |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * Don't increase the size of the local alloc window until we | |
1083 | * know we might be able to fulfill the request. Otherwise, we | |
1084 | * risk bouncing around the global bitmap during periods of | |
1085 | * low space. | |
1086 | */ | |
1087 | if (osb->local_alloc_state != OCFS2_LA_THROTTLED) | |
1088 | osb->local_alloc_bits = osb->local_alloc_default_bits; | |
1089 | ||
1090 | out_unlock: | |
1091 | state = osb->local_alloc_state; | |
1092 | spin_unlock(&osb->osb_lock); | |
1093 | ||
1094 | return state; | |
1095 | } | |
1096 | ||
ccd979bd | 1097 | static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb, |
ccd979bd MF |
1098 | struct ocfs2_alloc_context **ac, |
1099 | struct inode **bitmap_inode, | |
1100 | struct buffer_head **bitmap_bh) | |
1101 | { | |
1102 | int status; | |
1103 | ||
cd861280 | 1104 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
1105 | if (!(*ac)) { |
1106 | status = -ENOMEM; | |
1107 | mlog_errno(status); | |
1108 | goto bail; | |
1109 | } | |
1110 | ||
9c7af40b | 1111 | retry_enospc: |
7e9b7937 | 1112 | (*ac)->ac_bits_wanted = osb->local_alloc_bits; |
ccd979bd | 1113 | status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac); |
9c7af40b MF |
1114 | if (status == -ENOSPC) { |
1115 | if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) == | |
1116 | OCFS2_LA_DISABLED) | |
1117 | goto bail; | |
1118 | ||
1119 | ocfs2_free_ac_resource(*ac); | |
1120 | memset(*ac, 0, sizeof(struct ocfs2_alloc_context)); | |
1121 | goto retry_enospc; | |
1122 | } | |
ccd979bd | 1123 | if (status < 0) { |
9c7af40b | 1124 | mlog_errno(status); |
ccd979bd MF |
1125 | goto bail; |
1126 | } | |
1127 | ||
1128 | *bitmap_inode = (*ac)->ac_inode; | |
1129 | igrab(*bitmap_inode); | |
1130 | *bitmap_bh = (*ac)->ac_bh; | |
1131 | get_bh(*bitmap_bh); | |
1132 | status = 0; | |
1133 | bail: | |
1134 | if ((status < 0) && *ac) { | |
1135 | ocfs2_free_alloc_context(*ac); | |
1136 | *ac = NULL; | |
1137 | } | |
1138 | ||
c1e8d35e TM |
1139 | if (status) |
1140 | mlog_errno(status); | |
ccd979bd MF |
1141 | return status; |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * pass it the bitmap lock in lock_bh if you have it. | |
1146 | */ | |
1147 | static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb, | |
1fabe148 | 1148 | handle_t *handle, |
ccd979bd MF |
1149 | struct ocfs2_alloc_context *ac) |
1150 | { | |
1151 | int status = 0; | |
1152 | u32 cluster_off, cluster_count; | |
1153 | struct ocfs2_dinode *alloc = NULL; | |
1154 | struct ocfs2_local_alloc *la; | |
1155 | ||
ccd979bd MF |
1156 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; |
1157 | la = OCFS2_LOCAL_ALLOC(alloc); | |
1158 | ||
a04733d8 TM |
1159 | trace_ocfs2_local_alloc_new_window( |
1160 | le32_to_cpu(alloc->id1.bitmap1.i_total), | |
1161 | osb->local_alloc_bits); | |
883d4cae MF |
1162 | |
1163 | /* Instruct the allocation code to try the most recently used | |
1164 | * cluster group. We'll re-record the group used this pass | |
1165 | * below. */ | |
1166 | ac->ac_last_group = osb->la_last_gd; | |
1167 | ||
ccd979bd MF |
1168 | /* we used the generic suballoc reserve function, but we set |
1169 | * everything up nicely, so there's no reason why we can't use | |
1170 | * the more specific cluster api to claim bits. */ | |
1ed9b777 | 1171 | status = ocfs2_claim_clusters(handle, ac, osb->local_alloc_bits, |
ccd979bd | 1172 | &cluster_off, &cluster_count); |
9c7af40b MF |
1173 | if (status == -ENOSPC) { |
1174 | retry_enospc: | |
1175 | /* | |
1176 | * Note: We could also try syncing the journal here to | |
1177 | * allow use of any free bits which the current | |
1178 | * transaction can't give us access to. --Mark | |
1179 | */ | |
1180 | if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) == | |
1181 | OCFS2_LA_DISABLED) | |
1182 | goto bail; | |
1183 | ||
7e9b7937 | 1184 | ac->ac_bits_wanted = osb->local_alloc_bits; |
1ed9b777 | 1185 | status = ocfs2_claim_clusters(handle, ac, |
9c7af40b MF |
1186 | osb->local_alloc_bits, |
1187 | &cluster_off, | |
1188 | &cluster_count); | |
1189 | if (status == -ENOSPC) | |
1190 | goto retry_enospc; | |
1191 | /* | |
1192 | * We only shrunk the *minimum* number of in our | |
1193 | * request - it's entirely possible that the allocator | |
1194 | * might give us more than we asked for. | |
1195 | */ | |
1196 | if (status == 0) { | |
1197 | spin_lock(&osb->osb_lock); | |
1198 | osb->local_alloc_bits = cluster_count; | |
1199 | spin_unlock(&osb->osb_lock); | |
1200 | } | |
1201 | } | |
ccd979bd MF |
1202 | if (status < 0) { |
1203 | if (status != -ENOSPC) | |
1204 | mlog_errno(status); | |
1205 | goto bail; | |
1206 | } | |
1207 | ||
883d4cae MF |
1208 | osb->la_last_gd = ac->ac_last_group; |
1209 | ||
ccd979bd MF |
1210 | la->la_bm_off = cpu_to_le32(cluster_off); |
1211 | alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count); | |
1212 | /* just in case... In the future when we find space ourselves, | |
1213 | * we don't have to get all contiguous -- but we'll have to | |
1214 | * set all previously used bits in bitmap and update | |
1215 | * la_bits_set before setting the bits in the main bitmap. */ | |
1216 | alloc->id1.bitmap1.i_used = 0; | |
1217 | memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0, | |
1218 | le16_to_cpu(la->la_size)); | |
1219 | ||
d02f00cc MF |
1220 | ocfs2_resmap_restart(&osb->osb_la_resmap, cluster_count, |
1221 | OCFS2_LOCAL_ALLOC(alloc)->la_bitmap); | |
1222 | ||
a04733d8 TM |
1223 | trace_ocfs2_local_alloc_new_window_result( |
1224 | OCFS2_LOCAL_ALLOC(alloc)->la_bm_off, | |
1225 | le32_to_cpu(alloc->id1.bitmap1.i_total)); | |
ccd979bd MF |
1226 | |
1227 | bail: | |
c1e8d35e TM |
1228 | if (status) |
1229 | mlog_errno(status); | |
ccd979bd MF |
1230 | return status; |
1231 | } | |
1232 | ||
1233 | /* Note that we do *NOT* lock the local alloc inode here as | |
1234 | * it's been locked already for us. */ | |
1235 | static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb, | |
1236 | struct inode *local_alloc_inode) | |
1237 | { | |
1238 | int status = 0; | |
1239 | struct buffer_head *main_bm_bh = NULL; | |
1240 | struct inode *main_bm_inode = NULL; | |
1fabe148 | 1241 | handle_t *handle = NULL; |
ccd979bd MF |
1242 | struct ocfs2_dinode *alloc; |
1243 | struct ocfs2_dinode *alloc_copy = NULL; | |
1244 | struct ocfs2_alloc_context *ac = NULL; | |
1245 | ||
9c7af40b MF |
1246 | ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE); |
1247 | ||
ccd979bd MF |
1248 | /* This will lock the main bitmap for us. */ |
1249 | status = ocfs2_local_alloc_reserve_for_window(osb, | |
ccd979bd MF |
1250 | &ac, |
1251 | &main_bm_inode, | |
1252 | &main_bm_bh); | |
1253 | if (status < 0) { | |
1254 | if (status != -ENOSPC) | |
1255 | mlog_errno(status); | |
1256 | goto bail; | |
1257 | } | |
1258 | ||
65eff9cc | 1259 | handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS); |
ccd979bd MF |
1260 | if (IS_ERR(handle)) { |
1261 | status = PTR_ERR(handle); | |
1262 | handle = NULL; | |
1263 | mlog_errno(status); | |
1264 | goto bail; | |
1265 | } | |
1266 | ||
1267 | alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data; | |
1268 | ||
1269 | /* We want to clear the local alloc before doing anything | |
1270 | * else, so that if we error later during this operation, | |
1271 | * local alloc shutdown won't try to double free main bitmap | |
1272 | * bits. Make a copy so the sync function knows which bits to | |
1273 | * free. */ | |
d8b2fa65 | 1274 | alloc_copy = kmemdup(alloc, osb->local_alloc_bh->b_size, GFP_NOFS); |
ccd979bd MF |
1275 | if (!alloc_copy) { |
1276 | status = -ENOMEM; | |
1277 | mlog_errno(status); | |
1278 | goto bail; | |
1279 | } | |
ccd979bd | 1280 | |
0cf2f763 JB |
1281 | status = ocfs2_journal_access_di(handle, |
1282 | INODE_CACHE(local_alloc_inode), | |
13723d00 JB |
1283 | osb->local_alloc_bh, |
1284 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
1285 | if (status < 0) { |
1286 | mlog_errno(status); | |
1287 | goto bail; | |
1288 | } | |
1289 | ||
1290 | ocfs2_clear_local_alloc(alloc); | |
ec20cec7 | 1291 | ocfs2_journal_dirty(handle, osb->local_alloc_bh); |
ccd979bd MF |
1292 | |
1293 | status = ocfs2_sync_local_to_main(osb, handle, alloc_copy, | |
1294 | main_bm_inode, main_bm_bh); | |
1295 | if (status < 0) { | |
1296 | mlog_errno(status); | |
1297 | goto bail; | |
1298 | } | |
1299 | ||
1300 | status = ocfs2_local_alloc_new_window(osb, handle, ac); | |
1301 | if (status < 0) { | |
1302 | if (status != -ENOSPC) | |
1303 | mlog_errno(status); | |
1304 | goto bail; | |
1305 | } | |
1306 | ||
1307 | atomic_inc(&osb->alloc_stats.moves); | |
1308 | ||
ccd979bd MF |
1309 | bail: |
1310 | if (handle) | |
02dc1af4 | 1311 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1312 | |
a81cb88b | 1313 | brelse(main_bm_bh); |
ccd979bd | 1314 | |
72865d92 | 1315 | iput(main_bm_inode); |
d787ab09 | 1316 | kfree(alloc_copy); |
ccd979bd MF |
1317 | |
1318 | if (ac) | |
1319 | ocfs2_free_alloc_context(ac); | |
1320 | ||
c1e8d35e TM |
1321 | if (status) |
1322 | mlog_errno(status); | |
ccd979bd MF |
1323 | return status; |
1324 | } | |
1325 |