]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
75ca61c1 | 15 | #include <linux/mempool.h> |
5c676f6d | 16 | #include <linux/gfs2_ondisk.h> |
c969f58c SW |
17 | #include <linux/bio.h> |
18 | #include <linux/fs.h> | |
7f63257d | 19 | #include <linux/list_sort.h> |
b3b94faa DT |
20 | |
21 | #include "gfs2.h" | |
5c676f6d | 22 | #include "incore.h" |
2332c443 | 23 | #include "inode.h" |
b3b94faa DT |
24 | #include "glock.h" |
25 | #include "log.h" | |
26 | #include "lops.h" | |
27 | #include "meta_io.h" | |
28 | #include "recovery.h" | |
29 | #include "rgrp.h" | |
30 | #include "trans.h" | |
5c676f6d | 31 | #include "util.h" |
63997775 | 32 | #include "trace_gfs2.h" |
b3b94faa | 33 | |
9b9107a5 SW |
34 | /** |
35 | * gfs2_pin - Pin a buffer in memory | |
36 | * @sdp: The superblock | |
37 | * @bh: The buffer to be pinned | |
38 | * | |
39 | * The log lock must be held when calling this function | |
40 | */ | |
767f433f | 41 | void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh) |
9b9107a5 SW |
42 | { |
43 | struct gfs2_bufdata *bd; | |
44 | ||
29687a2a | 45 | BUG_ON(!current->journal_info); |
9b9107a5 SW |
46 | |
47 | clear_buffer_dirty(bh); | |
48 | if (test_set_buffer_pinned(bh)) | |
49 | gfs2_assert_withdraw(sdp, 0); | |
50 | if (!buffer_uptodate(bh)) | |
51 | gfs2_io_error_bh(sdp, bh); | |
52 | bd = bh->b_private; | |
53 | /* If this buffer is in the AIL and it has already been written | |
54 | * to in-place disk block, remove it from the AIL. | |
55 | */ | |
c618e87a | 56 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 BM |
57 | if (bd->bd_tr) |
58 | list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list); | |
c618e87a | 59 | spin_unlock(&sdp->sd_ail_lock); |
9b9107a5 | 60 | get_bh(bh); |
5e687eac | 61 | atomic_inc(&sdp->sd_log_pinned); |
63997775 | 62 | trace_gfs2_pin(bd, 1); |
9b9107a5 SW |
63 | } |
64 | ||
7c9ca621 BP |
65 | static bool buffer_is_rgrp(const struct gfs2_bufdata *bd) |
66 | { | |
67 | return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP; | |
68 | } | |
69 | ||
70 | static void maybe_release_space(struct gfs2_bufdata *bd) | |
71 | { | |
72 | struct gfs2_glock *gl = bd->bd_gl; | |
73 | struct gfs2_sbd *sdp = gl->gl_sbd; | |
74 | struct gfs2_rgrpd *rgd = gl->gl_object; | |
75 | unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number; | |
76 | struct gfs2_bitmap *bi = rgd->rd_bits + index; | |
77 | ||
78 | if (bi->bi_clone == 0) | |
79 | return; | |
80 | if (sdp->sd_args.ar_discard) | |
66fc061b | 81 | gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL); |
7c9ca621 BP |
82 | memcpy(bi->bi_clone + bi->bi_offset, |
83 | bd->bd_bh->b_data + bi->bi_offset, bi->bi_len); | |
84 | clear_bit(GBF_FULL, &bi->bi_flags); | |
85 | rgd->rd_free_clone = rgd->rd_free; | |
86 | } | |
87 | ||
9b9107a5 SW |
88 | /** |
89 | * gfs2_unpin - Unpin a buffer | |
90 | * @sdp: the filesystem the buffer belongs to | |
91 | * @bh: The buffer to unpin | |
92 | * @ai: | |
29687a2a | 93 | * @flags: The inode dirty flags |
9b9107a5 SW |
94 | * |
95 | */ | |
96 | ||
97 | static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh, | |
16ca9412 | 98 | struct gfs2_trans *tr) |
9b9107a5 SW |
99 | { |
100 | struct gfs2_bufdata *bd = bh->b_private; | |
101 | ||
29687a2a SW |
102 | BUG_ON(!buffer_uptodate(bh)); |
103 | BUG_ON(!buffer_pinned(bh)); | |
9b9107a5 SW |
104 | |
105 | lock_buffer(bh); | |
106 | mark_buffer_dirty(bh); | |
107 | clear_buffer_pinned(bh); | |
108 | ||
7c9ca621 BP |
109 | if (buffer_is_rgrp(bd)) |
110 | maybe_release_space(bd); | |
111 | ||
d6a079e8 | 112 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 | 113 | if (bd->bd_tr) { |
9b9107a5 SW |
114 | list_del(&bd->bd_ail_st_list); |
115 | brelse(bh); | |
116 | } else { | |
117 | struct gfs2_glock *gl = bd->bd_gl; | |
118 | list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list); | |
119 | atomic_inc(&gl->gl_ail_count); | |
120 | } | |
16ca9412 BM |
121 | bd->bd_tr = tr; |
122 | list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list); | |
d6a079e8 DC |
123 | spin_unlock(&sdp->sd_ail_lock); |
124 | ||
29687a2a | 125 | clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); |
63997775 | 126 | trace_gfs2_pin(bd, 0); |
9b9107a5 | 127 | unlock_buffer(bh); |
5e687eac | 128 | atomic_dec(&sdp->sd_log_pinned); |
9b9107a5 SW |
129 | } |
130 | ||
e8c92ed7 | 131 | static void gfs2_log_incr_head(struct gfs2_sbd *sdp) |
16615be1 | 132 | { |
e8c92ed7 SW |
133 | BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) && |
134 | (sdp->sd_log_flush_head != sdp->sd_log_head)); | |
135 | ||
136 | if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) { | |
137 | sdp->sd_log_flush_head = 0; | |
138 | sdp->sd_log_flush_wrapped = 1; | |
139 | } | |
16615be1 SW |
140 | } |
141 | ||
e8c92ed7 | 142 | static u64 gfs2_log_bmap(struct gfs2_sbd *sdp) |
16615be1 | 143 | { |
e8c92ed7 SW |
144 | unsigned int lbn = sdp->sd_log_flush_head; |
145 | struct gfs2_journal_extent *je; | |
146 | u64 block; | |
147 | ||
148 | list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) { | |
149 | if (lbn >= je->lblock && lbn < je->lblock + je->blocks) { | |
150 | block = je->dblock + lbn - je->lblock; | |
151 | gfs2_log_incr_head(sdp); | |
152 | return block; | |
153 | } | |
154 | } | |
155 | ||
156 | return -1; | |
16615be1 SW |
157 | } |
158 | ||
e8c92ed7 SW |
159 | /** |
160 | * gfs2_end_log_write_bh - end log write of pagecache data with buffers | |
161 | * @sdp: The superblock | |
162 | * @bvec: The bio_vec | |
163 | * @error: The i/o status | |
164 | * | |
165 | * This finds the relavent buffers and unlocks then and sets the | |
166 | * error flag according to the status of the i/o request. This is | |
167 | * used when the log is writing data which has an in-place version | |
168 | * that is pinned in the pagecache. | |
169 | */ | |
170 | ||
171 | static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec, | |
172 | int error) | |
16615be1 | 173 | { |
e8c92ed7 SW |
174 | struct buffer_head *bh, *next; |
175 | struct page *page = bvec->bv_page; | |
176 | unsigned size; | |
177 | ||
178 | bh = page_buffers(page); | |
179 | size = bvec->bv_len; | |
180 | while (bh_offset(bh) < bvec->bv_offset) | |
181 | bh = bh->b_this_page; | |
182 | do { | |
183 | if (error) | |
184 | set_buffer_write_io_error(bh); | |
185 | unlock_buffer(bh); | |
186 | next = bh->b_this_page; | |
187 | size -= bh->b_size; | |
188 | brelse(bh); | |
189 | bh = next; | |
190 | } while(bh && size); | |
16615be1 SW |
191 | } |
192 | ||
47ac5537 | 193 | /** |
e8c92ed7 SW |
194 | * gfs2_end_log_write - end of i/o to the log |
195 | * @bio: The bio | |
196 | * @error: Status of i/o request | |
197 | * | |
198 | * Each bio_vec contains either data from the pagecache or data | |
199 | * relating to the log itself. Here we iterate over the bio_vec | |
200 | * array, processing both kinds of data. | |
47ac5537 SW |
201 | * |
202 | */ | |
203 | ||
e8c92ed7 | 204 | static void gfs2_end_log_write(struct bio *bio, int error) |
47ac5537 | 205 | { |
e8c92ed7 SW |
206 | struct gfs2_sbd *sdp = bio->bi_private; |
207 | struct bio_vec *bvec; | |
208 | struct page *page; | |
209 | int i; | |
210 | ||
211 | if (error) { | |
212 | sdp->sd_log_error = error; | |
213 | fs_err(sdp, "Error %d writing to log\n", error); | |
214 | } | |
215 | ||
e97e548b | 216 | bio_for_each_segment_all(bvec, bio, i) { |
e8c92ed7 SW |
217 | page = bvec->bv_page; |
218 | if (page_has_buffers(page)) | |
219 | gfs2_end_log_write_bh(sdp, bvec, error); | |
220 | else | |
221 | mempool_free(page, gfs2_page_pool); | |
222 | } | |
47ac5537 | 223 | |
e8c92ed7 | 224 | bio_put(bio); |
47ac5537 SW |
225 | if (atomic_dec_and_test(&sdp->sd_log_in_flight)) |
226 | wake_up(&sdp->sd_log_flush_wait); | |
227 | } | |
228 | ||
229 | /** | |
e8c92ed7 SW |
230 | * gfs2_log_flush_bio - Submit any pending log bio |
231 | * @sdp: The superblock | |
232 | * @rw: The rw flags | |
47ac5537 | 233 | * |
e8c92ed7 SW |
234 | * Submit any pending part-built or full bio to the block device. If |
235 | * there is no pending bio, then this is a no-op. | |
47ac5537 SW |
236 | */ |
237 | ||
e8c92ed7 | 238 | void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int rw) |
47ac5537 | 239 | { |
e8c92ed7 SW |
240 | if (sdp->sd_log_bio) { |
241 | atomic_inc(&sdp->sd_log_in_flight); | |
242 | submit_bio(rw, sdp->sd_log_bio); | |
243 | sdp->sd_log_bio = NULL; | |
244 | } | |
245 | } | |
47ac5537 | 246 | |
e8c92ed7 SW |
247 | /** |
248 | * gfs2_log_alloc_bio - Allocate a new bio for log writing | |
249 | * @sdp: The superblock | |
250 | * @blkno: The next device block number we want to write to | |
251 | * | |
252 | * This should never be called when there is a cached bio in the | |
253 | * super block. When it returns, there will be a cached bio in the | |
254 | * super block which will have as many bio_vecs as the device is | |
255 | * happy to handle. | |
256 | * | |
257 | * Returns: Newly allocated bio | |
258 | */ | |
259 | ||
260 | static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno) | |
261 | { | |
262 | struct super_block *sb = sdp->sd_vfs; | |
263 | unsigned nrvecs = bio_get_nr_vecs(sb->s_bdev); | |
264 | struct bio *bio; | |
265 | ||
266 | BUG_ON(sdp->sd_log_bio); | |
267 | ||
268 | while (1) { | |
269 | bio = bio_alloc(GFP_NOIO, nrvecs); | |
270 | if (likely(bio)) | |
271 | break; | |
272 | nrvecs = max(nrvecs/2, 1U); | |
273 | } | |
47ac5537 | 274 | |
e8c92ed7 SW |
275 | bio->bi_sector = blkno * (sb->s_blocksize >> 9); |
276 | bio->bi_bdev = sb->s_bdev; | |
277 | bio->bi_end_io = gfs2_end_log_write; | |
278 | bio->bi_private = sdp; | |
279 | ||
280 | sdp->sd_log_bio = bio; | |
281 | ||
282 | return bio; | |
47ac5537 SW |
283 | } |
284 | ||
285 | /** | |
e8c92ed7 SW |
286 | * gfs2_log_get_bio - Get cached log bio, or allocate a new one |
287 | * @sdp: The superblock | |
288 | * @blkno: The device block number we want to write to | |
289 | * | |
290 | * If there is a cached bio, then if the next block number is sequential | |
291 | * with the previous one, return it, otherwise flush the bio to the | |
292 | * device. If there is not a cached bio, or we just flushed it, then | |
293 | * allocate a new one. | |
47ac5537 | 294 | * |
e8c92ed7 | 295 | * Returns: The bio to use for log writes |
47ac5537 SW |
296 | */ |
297 | ||
e8c92ed7 | 298 | static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno) |
47ac5537 | 299 | { |
e8c92ed7 SW |
300 | struct bio *bio = sdp->sd_log_bio; |
301 | u64 nblk; | |
302 | ||
303 | if (bio) { | |
f73a1c7d | 304 | nblk = bio_end_sector(bio); |
e8c92ed7 SW |
305 | nblk >>= sdp->sd_fsb2bb_shift; |
306 | if (blkno == nblk) | |
307 | return bio; | |
308 | gfs2_log_flush_bio(sdp, WRITE); | |
309 | } | |
310 | ||
311 | return gfs2_log_alloc_bio(sdp, blkno); | |
47ac5537 SW |
312 | } |
313 | ||
e8c92ed7 | 314 | |
47ac5537 | 315 | /** |
e8c92ed7 | 316 | * gfs2_log_write - write to log |
47ac5537 | 317 | * @sdp: the filesystem |
e8c92ed7 SW |
318 | * @page: the page to write |
319 | * @size: the size of the data to write | |
320 | * @offset: the offset within the page | |
47ac5537 | 321 | * |
e8c92ed7 SW |
322 | * Try and add the page segment to the current bio. If that fails, |
323 | * submit the current bio to the device and create a new one, and | |
324 | * then add the page segment to that. | |
47ac5537 SW |
325 | */ |
326 | ||
e8c92ed7 SW |
327 | static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page, |
328 | unsigned size, unsigned offset) | |
47ac5537 | 329 | { |
e8c92ed7 SW |
330 | u64 blkno = gfs2_log_bmap(sdp); |
331 | struct bio *bio; | |
332 | int ret; | |
333 | ||
334 | bio = gfs2_log_get_bio(sdp, blkno); | |
335 | ret = bio_add_page(bio, page, size, offset); | |
336 | if (ret == 0) { | |
337 | gfs2_log_flush_bio(sdp, WRITE); | |
338 | bio = gfs2_log_alloc_bio(sdp, blkno); | |
339 | ret = bio_add_page(bio, page, size, offset); | |
340 | WARN_ON(ret == 0); | |
341 | } | |
342 | } | |
47ac5537 | 343 | |
e8c92ed7 SW |
344 | /** |
345 | * gfs2_log_write_bh - write a buffer's content to the log | |
346 | * @sdp: The super block | |
347 | * @bh: The buffer pointing to the in-place location | |
348 | * | |
349 | * This writes the content of the buffer to the next available location | |
350 | * in the log. The buffer will be unlocked once the i/o to the log has | |
351 | * completed. | |
352 | */ | |
353 | ||
354 | static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh) | |
355 | { | |
356 | gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh)); | |
357 | } | |
47ac5537 | 358 | |
e8c92ed7 SW |
359 | /** |
360 | * gfs2_log_write_page - write one block stored in a page, into the log | |
361 | * @sdp: The superblock | |
362 | * @page: The struct page | |
363 | * | |
364 | * This writes the first block-sized part of the page into the log. Note | |
365 | * that the page must have been allocated from the gfs2_page_pool mempool | |
366 | * and that after this has been called, ownership has been transferred and | |
367 | * the page may be freed at any time. | |
368 | */ | |
47ac5537 | 369 | |
e8c92ed7 SW |
370 | void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page) |
371 | { | |
372 | struct super_block *sb = sdp->sd_vfs; | |
373 | gfs2_log_write(sdp, page, sb->s_blocksize, 0); | |
47ac5537 | 374 | } |
16615be1 | 375 | |
dad30e90 SW |
376 | static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type, |
377 | u32 ld_length, u32 ld_data1) | |
16615be1 | 378 | { |
144a4c2f | 379 | struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO); |
e8c92ed7 SW |
380 | struct gfs2_log_descriptor *ld = page_address(page); |
381 | clear_page(ld); | |
16615be1 SW |
382 | ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); |
383 | ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD); | |
384 | ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD); | |
385 | ld->ld_type = cpu_to_be32(ld_type); | |
dad30e90 SW |
386 | ld->ld_length = cpu_to_be32(ld_length); |
387 | ld->ld_data1 = cpu_to_be32(ld_data1); | |
16615be1 | 388 | ld->ld_data2 = 0; |
e8c92ed7 | 389 | return page; |
16615be1 SW |
390 | } |
391 | ||
dad30e90 SW |
392 | static void gfs2_check_magic(struct buffer_head *bh) |
393 | { | |
394 | void *kaddr; | |
395 | __be32 *ptr; | |
396 | ||
397 | clear_buffer_escaped(bh); | |
398 | kaddr = kmap_atomic(bh->b_page); | |
399 | ptr = kaddr + bh_offset(bh); | |
400 | if (*ptr == cpu_to_be32(GFS2_MAGIC)) | |
401 | set_buffer_escaped(bh); | |
402 | kunmap_atomic(kaddr); | |
403 | } | |
404 | ||
7f63257d BM |
405 | static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b) |
406 | { | |
407 | struct gfs2_bufdata *bda, *bdb; | |
408 | ||
409 | bda = list_entry(a, struct gfs2_bufdata, bd_list); | |
410 | bdb = list_entry(b, struct gfs2_bufdata, bd_list); | |
411 | ||
412 | if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr) | |
413 | return -1; | |
414 | if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr) | |
415 | return 1; | |
416 | return 0; | |
417 | } | |
418 | ||
dad30e90 SW |
419 | static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit, |
420 | unsigned int total, struct list_head *blist, | |
421 | bool is_databuf) | |
b3b94faa | 422 | { |
b3b94faa DT |
423 | struct gfs2_log_descriptor *ld; |
424 | struct gfs2_bufdata *bd1 = NULL, *bd2; | |
e8c92ed7 | 425 | struct page *page; |
b3b94faa DT |
426 | unsigned int num; |
427 | unsigned n; | |
428 | __be64 *ptr; | |
429 | ||
905d2aef | 430 | gfs2_log_lock(sdp); |
7f63257d | 431 | list_sort(NULL, blist, blocknr_cmp); |
c0752aa7 | 432 | bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list); |
b3b94faa DT |
433 | while(total) { |
434 | num = total; | |
435 | if (total > limit) | |
436 | num = limit; | |
905d2aef | 437 | gfs2_log_unlock(sdp); |
4a586812 BP |
438 | page = gfs2_get_log_desc(sdp, |
439 | is_databuf ? GFS2_LOG_DESC_JDATA : | |
440 | GFS2_LOG_DESC_METADATA, num + 1, num); | |
e8c92ed7 | 441 | ld = page_address(page); |
905d2aef | 442 | gfs2_log_lock(sdp); |
e8c92ed7 | 443 | ptr = (__be64 *)(ld + 1); |
b3b94faa DT |
444 | |
445 | n = 0; | |
c0752aa7 | 446 | list_for_each_entry_continue(bd1, blist, bd_list) { |
b3b94faa | 447 | *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr); |
dad30e90 SW |
448 | if (is_databuf) { |
449 | gfs2_check_magic(bd1->bd_bh); | |
450 | *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0); | |
451 | } | |
b3b94faa DT |
452 | if (++n >= num) |
453 | break; | |
454 | } | |
455 | ||
905d2aef | 456 | gfs2_log_unlock(sdp); |
e8c92ed7 | 457 | gfs2_log_write_page(sdp, page); |
905d2aef | 458 | gfs2_log_lock(sdp); |
b3b94faa DT |
459 | |
460 | n = 0; | |
c0752aa7 | 461 | list_for_each_entry_continue(bd2, blist, bd_list) { |
16615be1 | 462 | get_bh(bd2->bd_bh); |
905d2aef | 463 | gfs2_log_unlock(sdp); |
16615be1 | 464 | lock_buffer(bd2->bd_bh); |
dad30e90 SW |
465 | |
466 | if (buffer_escaped(bd2->bd_bh)) { | |
467 | void *kaddr; | |
468 | page = mempool_alloc(gfs2_page_pool, GFP_NOIO); | |
469 | ptr = page_address(page); | |
470 | kaddr = kmap_atomic(bd2->bd_bh->b_page); | |
471 | memcpy(ptr, kaddr + bh_offset(bd2->bd_bh), | |
472 | bd2->bd_bh->b_size); | |
473 | kunmap_atomic(kaddr); | |
474 | *(__be32 *)ptr = 0; | |
475 | clear_buffer_escaped(bd2->bd_bh); | |
476 | unlock_buffer(bd2->bd_bh); | |
477 | brelse(bd2->bd_bh); | |
478 | gfs2_log_write_page(sdp, page); | |
479 | } else { | |
480 | gfs2_log_write_bh(sdp, bd2->bd_bh); | |
481 | } | |
905d2aef | 482 | gfs2_log_lock(sdp); |
b3b94faa DT |
483 | if (++n >= num) |
484 | break; | |
485 | } | |
486 | ||
905d2aef | 487 | BUG_ON(total < num); |
b3b94faa DT |
488 | total -= num; |
489 | } | |
905d2aef | 490 | gfs2_log_unlock(sdp); |
b3b94faa DT |
491 | } |
492 | ||
dad30e90 SW |
493 | static void buf_lo_before_commit(struct gfs2_sbd *sdp) |
494 | { | |
495 | unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */ | |
496 | ||
497 | gfs2_before_commit(sdp, limit, sdp->sd_log_num_buf, | |
498 | &sdp->sd_log_le_buf, 0); | |
499 | } | |
500 | ||
16ca9412 | 501 | static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
b3b94faa DT |
502 | { |
503 | struct list_head *head = &sdp->sd_log_le_buf; | |
504 | struct gfs2_bufdata *bd; | |
505 | ||
16ca9412 BM |
506 | if (tr == NULL) { |
507 | gfs2_assert(sdp, list_empty(head)); | |
508 | return; | |
509 | } | |
510 | ||
b3b94faa | 511 | while (!list_empty(head)) { |
c0752aa7 BP |
512 | bd = list_entry(head->next, struct gfs2_bufdata, bd_list); |
513 | list_del_init(&bd->bd_list); | |
b3b94faa DT |
514 | sdp->sd_log_num_buf--; |
515 | ||
16ca9412 | 516 | gfs2_unpin(sdp, bd->bd_bh, tr); |
b3b94faa DT |
517 | } |
518 | gfs2_assert_warn(sdp, !sdp->sd_log_num_buf); | |
519 | } | |
520 | ||
521 | static void buf_lo_before_scan(struct gfs2_jdesc *jd, | |
55167622 | 522 | struct gfs2_log_header_host *head, int pass) |
b3b94faa | 523 | { |
feaa7bba | 524 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
525 | |
526 | if (pass != 0) | |
527 | return; | |
528 | ||
529 | sdp->sd_found_blocks = 0; | |
530 | sdp->sd_replayed_blocks = 0; | |
531 | } | |
532 | ||
533 | static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
534 | struct gfs2_log_descriptor *ld, __be64 *ptr, | |
535 | int pass) | |
536 | { | |
feaa7bba SW |
537 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
538 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
5c676f6d | 539 | struct gfs2_glock *gl = ip->i_gl; |
b3b94faa DT |
540 | unsigned int blks = be32_to_cpu(ld->ld_data1); |
541 | struct buffer_head *bh_log, *bh_ip; | |
cd915493 | 542 | u64 blkno; |
b3b94faa DT |
543 | int error = 0; |
544 | ||
545 | if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA) | |
546 | return 0; | |
547 | ||
548 | gfs2_replay_incr_blk(sdp, &start); | |
549 | ||
550 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
551 | blkno = be64_to_cpu(*ptr++); | |
552 | ||
553 | sdp->sd_found_blocks++; | |
554 | ||
555 | if (gfs2_revoke_check(sdp, blkno, start)) | |
556 | continue; | |
557 | ||
558 | error = gfs2_replay_read_block(jd, start, &bh_log); | |
82ffa516 SW |
559 | if (error) |
560 | return error; | |
b3b94faa DT |
561 | |
562 | bh_ip = gfs2_meta_new(gl, blkno); | |
563 | memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); | |
564 | ||
565 | if (gfs2_meta_check(sdp, bh_ip)) | |
566 | error = -EIO; | |
567 | else | |
568 | mark_buffer_dirty(bh_ip); | |
569 | ||
570 | brelse(bh_log); | |
571 | brelse(bh_ip); | |
572 | ||
573 | if (error) | |
574 | break; | |
575 | ||
576 | sdp->sd_replayed_blocks++; | |
577 | } | |
578 | ||
579 | return error; | |
580 | } | |
581 | ||
582 | static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
583 | { | |
feaa7bba SW |
584 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
585 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
b3b94faa DT |
586 | |
587 | if (error) { | |
7276b3b0 | 588 | gfs2_meta_sync(ip->i_gl); |
b3b94faa DT |
589 | return; |
590 | } | |
591 | if (pass != 1) | |
592 | return; | |
593 | ||
7276b3b0 | 594 | gfs2_meta_sync(ip->i_gl); |
b3b94faa DT |
595 | |
596 | fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n", | |
597 | jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); | |
598 | } | |
599 | ||
b3b94faa DT |
600 | static void revoke_lo_before_commit(struct gfs2_sbd *sdp) |
601 | { | |
b3b94faa | 602 | struct gfs2_meta_header *mh; |
b3b94faa DT |
603 | unsigned int offset; |
604 | struct list_head *head = &sdp->sd_log_le_revoke; | |
82e86087 | 605 | struct gfs2_bufdata *bd; |
e8c92ed7 | 606 | struct page *page; |
dad30e90 | 607 | unsigned int length; |
b3b94faa DT |
608 | |
609 | if (!sdp->sd_log_num_revoke) | |
610 | return; | |
611 | ||
dad30e90 SW |
612 | length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64)); |
613 | page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke); | |
b3b94faa DT |
614 | offset = sizeof(struct gfs2_log_descriptor); |
615 | ||
c0752aa7 | 616 | list_for_each_entry(bd, head, bd_list) { |
b3b94faa DT |
617 | sdp->sd_log_num_revoke--; |
618 | ||
cd915493 | 619 | if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) { |
b3b94faa | 620 | |
e8c92ed7 SW |
621 | gfs2_log_write_page(sdp, page); |
622 | page = mempool_alloc(gfs2_page_pool, GFP_NOIO); | |
623 | mh = page_address(page); | |
624 | clear_page(mh); | |
b3b94faa | 625 | mh->mh_magic = cpu_to_be32(GFS2_MAGIC); |
e3167ded SW |
626 | mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB); |
627 | mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB); | |
b3b94faa DT |
628 | offset = sizeof(struct gfs2_meta_header); |
629 | } | |
630 | ||
e8c92ed7 | 631 | *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno); |
cd915493 | 632 | offset += sizeof(u64); |
b3b94faa DT |
633 | } |
634 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
635 | ||
e8c92ed7 | 636 | gfs2_log_write_page(sdp, page); |
b3b94faa DT |
637 | } |
638 | ||
16ca9412 | 639 | static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
f42ab085 SW |
640 | { |
641 | struct list_head *head = &sdp->sd_log_le_revoke; | |
642 | struct gfs2_bufdata *bd; | |
643 | struct gfs2_glock *gl; | |
644 | ||
645 | while (!list_empty(head)) { | |
c0752aa7 BP |
646 | bd = list_entry(head->next, struct gfs2_bufdata, bd_list); |
647 | list_del_init(&bd->bd_list); | |
f42ab085 SW |
648 | gl = bd->bd_gl; |
649 | atomic_dec(&gl->gl_revokes); | |
650 | clear_bit(GLF_LFLUSH, &gl->gl_flags); | |
651 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
652 | } | |
653 | } | |
654 | ||
b3b94faa | 655 | static void revoke_lo_before_scan(struct gfs2_jdesc *jd, |
55167622 | 656 | struct gfs2_log_header_host *head, int pass) |
b3b94faa | 657 | { |
feaa7bba | 658 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
659 | |
660 | if (pass != 0) | |
661 | return; | |
662 | ||
663 | sdp->sd_found_revokes = 0; | |
664 | sdp->sd_replay_tail = head->lh_tail; | |
665 | } | |
666 | ||
667 | static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
668 | struct gfs2_log_descriptor *ld, __be64 *ptr, | |
669 | int pass) | |
670 | { | |
feaa7bba | 671 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
672 | unsigned int blks = be32_to_cpu(ld->ld_length); |
673 | unsigned int revokes = be32_to_cpu(ld->ld_data1); | |
674 | struct buffer_head *bh; | |
675 | unsigned int offset; | |
cd915493 | 676 | u64 blkno; |
b3b94faa DT |
677 | int first = 1; |
678 | int error; | |
679 | ||
680 | if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE) | |
681 | return 0; | |
682 | ||
683 | offset = sizeof(struct gfs2_log_descriptor); | |
684 | ||
685 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
686 | error = gfs2_replay_read_block(jd, start, &bh); | |
687 | if (error) | |
688 | return error; | |
689 | ||
690 | if (!first) | |
691 | gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB); | |
692 | ||
cd915493 | 693 | while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) { |
b3b94faa DT |
694 | blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset)); |
695 | ||
696 | error = gfs2_revoke_add(sdp, blkno, start); | |
3ad62e87 BP |
697 | if (error < 0) { |
698 | brelse(bh); | |
b3b94faa | 699 | return error; |
3ad62e87 | 700 | } |
b3b94faa DT |
701 | else if (error) |
702 | sdp->sd_found_revokes++; | |
703 | ||
704 | if (!--revokes) | |
705 | break; | |
cd915493 | 706 | offset += sizeof(u64); |
b3b94faa DT |
707 | } |
708 | ||
709 | brelse(bh); | |
710 | offset = sizeof(struct gfs2_meta_header); | |
711 | first = 0; | |
712 | } | |
713 | ||
714 | return 0; | |
715 | } | |
716 | ||
717 | static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
718 | { | |
feaa7bba | 719 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); |
b3b94faa DT |
720 | |
721 | if (error) { | |
722 | gfs2_revoke_clean(sdp); | |
723 | return; | |
724 | } | |
725 | if (pass != 1) | |
726 | return; | |
727 | ||
728 | fs_info(sdp, "jid=%u: Found %u revoke tags\n", | |
729 | jd->jd_jid, sdp->sd_found_revokes); | |
730 | ||
731 | gfs2_revoke_clean(sdp); | |
732 | } | |
733 | ||
16615be1 SW |
734 | /** |
735 | * databuf_lo_before_commit - Scan the data buffers, writing as we go | |
736 | * | |
737 | */ | |
738 | ||
739 | static void databuf_lo_before_commit(struct gfs2_sbd *sdp) | |
740 | { | |
dad30e90 | 741 | unsigned int limit = buf_limit(sdp) / 2; |
16615be1 | 742 | |
dad30e90 SW |
743 | gfs2_before_commit(sdp, limit, sdp->sd_log_num_databuf, |
744 | &sdp->sd_log_le_databuf, 1); | |
18ec7d5c SW |
745 | } |
746 | ||
747 | static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, | |
748 | struct gfs2_log_descriptor *ld, | |
749 | __be64 *ptr, int pass) | |
750 | { | |
feaa7bba SW |
751 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
752 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
5c676f6d | 753 | struct gfs2_glock *gl = ip->i_gl; |
18ec7d5c SW |
754 | unsigned int blks = be32_to_cpu(ld->ld_data1); |
755 | struct buffer_head *bh_log, *bh_ip; | |
cd915493 SW |
756 | u64 blkno; |
757 | u64 esc; | |
18ec7d5c SW |
758 | int error = 0; |
759 | ||
760 | if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA) | |
761 | return 0; | |
762 | ||
763 | gfs2_replay_incr_blk(sdp, &start); | |
764 | for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { | |
765 | blkno = be64_to_cpu(*ptr++); | |
766 | esc = be64_to_cpu(*ptr++); | |
767 | ||
768 | sdp->sd_found_blocks++; | |
769 | ||
770 | if (gfs2_revoke_check(sdp, blkno, start)) | |
771 | continue; | |
772 | ||
773 | error = gfs2_replay_read_block(jd, start, &bh_log); | |
774 | if (error) | |
775 | return error; | |
776 | ||
777 | bh_ip = gfs2_meta_new(gl, blkno); | |
778 | memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); | |
779 | ||
780 | /* Unescape */ | |
781 | if (esc) { | |
782 | __be32 *eptr = (__be32 *)bh_ip->b_data; | |
783 | *eptr = cpu_to_be32(GFS2_MAGIC); | |
784 | } | |
785 | mark_buffer_dirty(bh_ip); | |
786 | ||
787 | brelse(bh_log); | |
788 | brelse(bh_ip); | |
18ec7d5c SW |
789 | |
790 | sdp->sd_replayed_blocks++; | |
791 | } | |
792 | ||
793 | return error; | |
794 | } | |
795 | ||
796 | /* FIXME: sort out accounting for log blocks etc. */ | |
797 | ||
798 | static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) | |
799 | { | |
feaa7bba SW |
800 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
801 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
18ec7d5c SW |
802 | |
803 | if (error) { | |
7276b3b0 | 804 | gfs2_meta_sync(ip->i_gl); |
18ec7d5c SW |
805 | return; |
806 | } | |
807 | if (pass != 1) | |
808 | return; | |
809 | ||
810 | /* data sync? */ | |
7276b3b0 | 811 | gfs2_meta_sync(ip->i_gl); |
18ec7d5c SW |
812 | |
813 | fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n", | |
814 | jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); | |
815 | } | |
816 | ||
16ca9412 | 817 | static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
18ec7d5c SW |
818 | { |
819 | struct list_head *head = &sdp->sd_log_le_databuf; | |
820 | struct gfs2_bufdata *bd; | |
821 | ||
16ca9412 BM |
822 | if (tr == NULL) { |
823 | gfs2_assert(sdp, list_empty(head)); | |
824 | return; | |
825 | } | |
826 | ||
18ec7d5c | 827 | while (!list_empty(head)) { |
c0752aa7 BP |
828 | bd = list_entry(head->next, struct gfs2_bufdata, bd_list); |
829 | list_del_init(&bd->bd_list); | |
18ec7d5c | 830 | sdp->sd_log_num_databuf--; |
16ca9412 | 831 | gfs2_unpin(sdp, bd->bd_bh, tr); |
18ec7d5c | 832 | } |
b3b94faa DT |
833 | gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf); |
834 | } | |
835 | ||
18ec7d5c | 836 | |
b09e593d | 837 | const struct gfs2_log_operations gfs2_buf_lops = { |
b3b94faa DT |
838 | .lo_before_commit = buf_lo_before_commit, |
839 | .lo_after_commit = buf_lo_after_commit, | |
840 | .lo_before_scan = buf_lo_before_scan, | |
841 | .lo_scan_elements = buf_lo_scan_elements, | |
842 | .lo_after_scan = buf_lo_after_scan, | |
ea67eedb | 843 | .lo_name = "buf", |
b3b94faa DT |
844 | }; |
845 | ||
b09e593d | 846 | const struct gfs2_log_operations gfs2_revoke_lops = { |
b3b94faa | 847 | .lo_before_commit = revoke_lo_before_commit, |
f42ab085 | 848 | .lo_after_commit = revoke_lo_after_commit, |
b3b94faa DT |
849 | .lo_before_scan = revoke_lo_before_scan, |
850 | .lo_scan_elements = revoke_lo_scan_elements, | |
851 | .lo_after_scan = revoke_lo_after_scan, | |
ea67eedb | 852 | .lo_name = "revoke", |
b3b94faa DT |
853 | }; |
854 | ||
b09e593d | 855 | const struct gfs2_log_operations gfs2_databuf_lops = { |
b3b94faa | 856 | .lo_before_commit = databuf_lo_before_commit, |
18ec7d5c SW |
857 | .lo_after_commit = databuf_lo_after_commit, |
858 | .lo_scan_elements = databuf_lo_scan_elements, | |
859 | .lo_after_scan = databuf_lo_after_scan, | |
ea67eedb | 860 | .lo_name = "databuf", |
b3b94faa DT |
861 | }; |
862 | ||
b09e593d | 863 | const struct gfs2_log_operations *gfs2_log_ops[] = { |
16615be1 | 864 | &gfs2_databuf_lops, |
b3b94faa | 865 | &gfs2_buf_lops, |
16615be1 | 866 | &gfs2_revoke_lops, |
ea67eedb | 867 | NULL, |
b3b94faa DT |
868 | }; |
869 |