?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
RT-AC56 3.0.0.4.374.37 core
[tomato.git] / release / src-rt-6.x.4708 / linux / linux-2.6.36 / fs / nilfs2 / segment.c
blobf6a7744c27c2e6ad26c0daa425e8899028910488
1 /*
2 * segment.c - NILFS segment constructor.
3 *
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
21 *
22 */
23
24 #include <linux/pagemap.h>
25 #include <linux/buffer_head.h>
26 #include <linux/writeback.h>
27 #include <linux/bio.h>
28 #include <linux/completion.h>
29 #include <linux/blkdev.h>
30 #include <linux/backing-dev.h>
31 #include <linux/freezer.h>
32 #include <linux/kthread.h>
33 #include <linux/crc32.h>
34 #include <linux/pagevec.h>
35 #include <linux/slab.h>
36 #include "nilfs.h"
37 #include "btnode.h"
38 #include "page.h"
39 #include "segment.h"
40 #include "sufile.h"
41 #include "cpfile.h"
42 #include "ifile.h"
43 #include "segbuf.h"
44
45
46 /*
47 * Segment constructor
48 */
49 #define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */
50
51 #define SC_MAX_SEGDELTA 64 /* Upper limit of the number of segments
52 appended in collection retry loop */
53
54 /* Construction mode */
55 enum {
56 SC_LSEG_SR = 1, /* Make a logical segment having a super root */
57 SC_LSEG_DSYNC, /* Flush data blocks of a given file and make
58 a logical segment without a super root */
59 SC_FLUSH_FILE, /* Flush data files, leads to segment writes without
60 creating a checkpoint */
61 SC_FLUSH_DAT, /* Flush DAT file. This also creates segments without
62 a checkpoint */
63 };
64
65 /* Stage numbers of dirty block collection */
66 enum {
67 NILFS_ST_INIT = 0,
68 NILFS_ST_GC, /* Collecting dirty blocks for GC */
69 NILFS_ST_FILE,
70 NILFS_ST_IFILE,
71 NILFS_ST_CPFILE,
72 NILFS_ST_SUFILE,
73 NILFS_ST_DAT,
74 NILFS_ST_SR, /* Super root */
75 NILFS_ST_DSYNC, /* Data sync blocks */
76 NILFS_ST_DONE,
77 };
78
79 /* State flags of collection */
80 #define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
81 #define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
82 #define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
83 #define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
84
85 /* Operations depending on the construction mode and file type */
86 struct nilfs_sc_operations {
87 int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
88 struct inode *);
89 int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
90 struct inode *);
91 int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
92 struct inode *);
93 void (*write_data_binfo)(struct nilfs_sc_info *,
94 struct nilfs_segsum_pointer *,
95 union nilfs_binfo *);
96 void (*write_node_binfo)(struct nilfs_sc_info *,
97 struct nilfs_segsum_pointer *,
98 union nilfs_binfo *);
99 };
100
101 /*
102 * Other definitions
103 */
104 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
105 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
106 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
107 static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
108 int);
109
110 #define nilfs_cnt32_gt(a, b) \
111 (typecheck(__u32, a) && typecheck(__u32, b) && \
112 ((__s32)(b) - (__s32)(a) < 0))
113 #define nilfs_cnt32_ge(a, b) \
114 (typecheck(__u32, a) && typecheck(__u32, b) && \
115 ((__s32)(a) - (__s32)(b) >= 0))
116 #define nilfs_cnt32_lt(a, b) nilfs_cnt32_gt(b, a)
117 #define nilfs_cnt32_le(a, b) nilfs_cnt32_ge(b, a)
118
119 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
120 {
121 struct nilfs_transaction_info *cur_ti = current->journal_info;
122 void *save = NULL;
123
124 if (cur_ti) {
125 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
126 return ++cur_ti->ti_count;
127 else {
128 /*
129 * If journal_info field is occupied by other FS,
130 * it is saved and will be restored on
131 * nilfs_transaction_commit().
132 */
133 printk(KERN_WARNING
134 "NILFS warning: journal info from a different "
135 "FS\n");
136 save = current->journal_info;
137 }
138 }
139 if (!ti) {
140 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
141 if (!ti)
142 return -ENOMEM;
143 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
144 } else {
145 ti->ti_flags = 0;
146 }
147 ti->ti_count = 0;
148 ti->ti_save = save;
149 ti->ti_magic = NILFS_TI_MAGIC;
150 current->journal_info = ti;
151 return 0;
152 }
153
154 /**
155 * nilfs_transaction_begin - start indivisible file operations.
156 * @sb: super block
157 * @ti: nilfs_transaction_info
158 * @vacancy_check: flags for vacancy rate checks
159 *
160 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
161 * the segment semaphore, to make a segment construction and write tasks
162 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
163 * The region enclosed by these two functions can be nested. To avoid a
164 * deadlock, the semaphore is only acquired or released in the outermost call.
165 *
166 * This function allocates a nilfs_transaction_info struct to keep context
167 * information on it. It is initialized and hooked onto the current task in
168 * the outermost call. If a pre-allocated struct is given to @ti, it is used
169 * instead; otherwise a new struct is assigned from a slab.
170 *
171 * When @vacancy_check flag is set, this function will check the amount of
172 * free space, and will wait for the GC to reclaim disk space if low capacity.
173 *
174 * Return Value: On success, 0 is returned. On error, one of the following
175 * negative error code is returned.
176 *
177 * %-ENOMEM - Insufficient memory available.
178 *
179 * %-ENOSPC - No space left on device
180 */
181 int nilfs_transaction_begin(struct super_block *sb,
182 struct nilfs_transaction_info *ti,
183 int vacancy_check)
184 {
185 struct nilfs_sb_info *sbi;
186 struct the_nilfs *nilfs;
187 int ret = nilfs_prepare_segment_lock(ti);
188
189 if (unlikely(ret < 0))
190 return ret;
191 if (ret > 0)
192 return 0;
193
194 sbi = NILFS_SB(sb);
195 nilfs = sbi->s_nilfs;
196 down_read(&nilfs->ns_segctor_sem);
197 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
198 up_read(&nilfs->ns_segctor_sem);
199 ret = -ENOSPC;
200 goto failed;
201 }
202 return 0;
203
204 failed:
205 ti = current->journal_info;
206 current->journal_info = ti->ti_save;
207 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
208 kmem_cache_free(nilfs_transaction_cachep, ti);
209 return ret;
210 }
211
212 /**
213 * nilfs_transaction_commit - commit indivisible file operations.
214 * @sb: super block
215 *
216 * nilfs_transaction_commit() releases the read semaphore which is
217 * acquired by nilfs_transaction_begin(). This is only performed
218 * in outermost call of this function. If a commit flag is set,
219 * nilfs_transaction_commit() sets a timer to start the segment
220 * constructor. If a sync flag is set, it starts construction
221 * directly.
222 */
223 int nilfs_transaction_commit(struct super_block *sb)
224 {
225 struct nilfs_transaction_info *ti = current->journal_info;
226 struct nilfs_sb_info *sbi;
227 struct nilfs_sc_info *sci;
228 int err = 0;
229
230 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
231 ti->ti_flags |= NILFS_TI_COMMIT;
232 if (ti->ti_count > 0) {
233 ti->ti_count--;
234 return 0;
235 }
236 sbi = NILFS_SB(sb);
237 sci = NILFS_SC(sbi);
238 if (sci != NULL) {
239 if (ti->ti_flags & NILFS_TI_COMMIT)
240 nilfs_segctor_start_timer(sci);
241 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
242 sci->sc_watermark)
243 nilfs_segctor_do_flush(sci, 0);
244 }
245 up_read(&sbi->s_nilfs->ns_segctor_sem);
246 current->journal_info = ti->ti_save;
247
248 if (ti->ti_flags & NILFS_TI_SYNC)
249 err = nilfs_construct_segment(sb);
250 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
251 kmem_cache_free(nilfs_transaction_cachep, ti);
252 return err;
253 }
254
255 void nilfs_transaction_abort(struct super_block *sb)
256 {
257 struct nilfs_transaction_info *ti = current->journal_info;
258
259 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
260 if (ti->ti_count > 0) {
261 ti->ti_count--;
262 return;
263 }
264 up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
265
266 current->journal_info = ti->ti_save;
267 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
268 kmem_cache_free(nilfs_transaction_cachep, ti);
269 }
270
271 void nilfs_relax_pressure_in_lock(struct super_block *sb)
272 {
273 struct nilfs_sb_info *sbi = NILFS_SB(sb);
274 struct nilfs_sc_info *sci = NILFS_SC(sbi);
275 struct the_nilfs *nilfs = sbi->s_nilfs;
276
277 if (!sci || !sci->sc_flush_request)
278 return;
279
280 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
281 up_read(&nilfs->ns_segctor_sem);
282
283 down_write(&nilfs->ns_segctor_sem);
284 if (sci->sc_flush_request &&
285 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
286 struct nilfs_transaction_info *ti = current->journal_info;
287
288 ti->ti_flags |= NILFS_TI_WRITER;
289 nilfs_segctor_do_immediate_flush(sci);
290 ti->ti_flags &= ~NILFS_TI_WRITER;
291 }
292 downgrade_write(&nilfs->ns_segctor_sem);
293 }
294
295 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
296 struct nilfs_transaction_info *ti,
297 int gcflag)
298 {
299 struct nilfs_transaction_info *cur_ti = current->journal_info;
300
301 WARN_ON(cur_ti);
302 ti->ti_flags = NILFS_TI_WRITER;
303 ti->ti_count = 0;
304 ti->ti_save = cur_ti;
305 ti->ti_magic = NILFS_TI_MAGIC;
306 INIT_LIST_HEAD(&ti->ti_garbage);
307 current->journal_info = ti;
308
309 for (;;) {
310 down_write(&sbi->s_nilfs->ns_segctor_sem);
311 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
312 break;
313
314 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
315
316 up_write(&sbi->s_nilfs->ns_segctor_sem);
317 yield();
318 }
319 if (gcflag)
320 ti->ti_flags |= NILFS_TI_GC;
321 }
322
323 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
324 {
325 struct nilfs_transaction_info *ti = current->journal_info;
326
327 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
328 BUG_ON(ti->ti_count > 0);
329
330 up_write(&sbi->s_nilfs->ns_segctor_sem);
331 current->journal_info = ti->ti_save;
332 if (!list_empty(&ti->ti_garbage))
333 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
334 }
335
336 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
337 struct nilfs_segsum_pointer *ssp,
338 unsigned bytes)
339 {
340 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
341 unsigned blocksize = sci->sc_super->s_blocksize;
342 void *p;
343
344 if (unlikely(ssp->offset + bytes > blocksize)) {
345 ssp->offset = 0;
346 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
347 &segbuf->sb_segsum_buffers));
348 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
349 }
350 p = ssp->bh->b_data + ssp->offset;
351 ssp->offset += bytes;
352 return p;
353 }
354
355 /**
356 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
357 * @sci: nilfs_sc_info
358 */
359 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
360 {
361 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
362 struct buffer_head *sumbh;
363 unsigned sumbytes;
364 unsigned flags = 0;
365 int err;
366
367 if (nilfs_doing_gc())
368 flags = NILFS_SS_GC;
369 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime,
370 sci->sc_sbi->s_nilfs->ns_cno);
371 if (unlikely(err))
372 return err;
373
374 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
375 sumbytes = segbuf->sb_sum.sumbytes;
376 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
377 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
378 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
379 return 0;
380 }
381
382 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
383 {
384 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
385 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
386 return -E2BIG; /* The current segment is filled up
387 (internal code) */
388 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
389 return nilfs_segctor_reset_segment_buffer(sci);
390 }
391
392 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
393 {
394 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
395 int err;
396
397 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
398 err = nilfs_segctor_feed_segment(sci);
399 if (err)
400 return err;
401 segbuf = sci->sc_curseg;
402 }
403 err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
404 if (likely(!err))
405 segbuf->sb_sum.flags |= NILFS_SS_SR;
406 return err;
407 }
408
409 /*
410 * Functions for making segment summary and payloads
411 */
412 static int nilfs_segctor_segsum_block_required(
413 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
414 unsigned binfo_size)
415 {
416 unsigned blocksize = sci->sc_super->s_blocksize;
417 /* Size of finfo and binfo is enough small against blocksize */
418
419 return ssp->offset + binfo_size +
420 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
421 blocksize;
422 }
423
424 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
425 struct inode *inode)
426 {
427 sci->sc_curseg->sb_sum.nfinfo++;
428 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
429 nilfs_segctor_map_segsum_entry(
430 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
431
432 if (inode->i_sb && !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
433 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
434 /* skip finfo */
435 }
436
437 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
438 struct inode *inode)
439 {
440 struct nilfs_finfo *finfo;
441 struct nilfs_inode_info *ii;
442 struct nilfs_segment_buffer *segbuf;
443
444 if (sci->sc_blk_cnt == 0)
445 return;
446
447 ii = NILFS_I(inode);
448 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
449 sizeof(*finfo));
450 finfo->fi_ino = cpu_to_le64(inode->i_ino);
451 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
452 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
453 finfo->fi_cno = cpu_to_le64(ii->i_cno);
454
455 segbuf = sci->sc_curseg;
456 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
457 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
458 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
459 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
460 }
461
462 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
463 struct buffer_head *bh,
464 struct inode *inode,
465 unsigned binfo_size)
466 {
467 struct nilfs_segment_buffer *segbuf;
468 int required, err = 0;
469
470 retry:
471 segbuf = sci->sc_curseg;
472 required = nilfs_segctor_segsum_block_required(
473 sci, &sci->sc_binfo_ptr, binfo_size);
474 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
475 nilfs_segctor_end_finfo(sci, inode);
476 err = nilfs_segctor_feed_segment(sci);
477 if (err)
478 return err;
479 goto retry;
480 }
481 if (unlikely(required)) {
482 err = nilfs_segbuf_extend_segsum(segbuf);
483 if (unlikely(err))
484 goto failed;
485 }
486 if (sci->sc_blk_cnt == 0)
487 nilfs_segctor_begin_finfo(sci, inode);
488
489 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
490 /* Substitution to vblocknr is delayed until update_blocknr() */
491 nilfs_segbuf_add_file_buffer(segbuf, bh);
492 sci->sc_blk_cnt++;
493 failed:
494 return err;
495 }
496
497 static int nilfs_handle_bmap_error(int err, const char *fname,
498 struct inode *inode, struct super_block *sb)
499 {
500 if (err == -EINVAL) {
501 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
502 inode->i_ino);
503 err = -EIO;
504 }
505 return err;
506 }
507
508 /*
509 * Callback functions that enumerate, mark, and collect dirty blocks
510 */
511 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
512 struct buffer_head *bh, struct inode *inode)
513 {
514 int err;
515
516 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
517 if (unlikely(err < 0))
518 return nilfs_handle_bmap_error(err, __func__, inode,
519 sci->sc_super);
520
521 err = nilfs_segctor_add_file_block(sci, bh, inode,
522 sizeof(struct nilfs_binfo_v));
523 if (!err)
524 sci->sc_datablk_cnt++;
525 return err;
526 }
527
528 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
529 struct buffer_head *bh,
530 struct inode *inode)
531 {
532 int err;
533
534 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
535 if (unlikely(err < 0))
536 return nilfs_handle_bmap_error(err, __func__, inode,
537 sci->sc_super);
538 return 0;
539 }
540
541 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
542 struct buffer_head *bh,
543 struct inode *inode)
544 {
545 WARN_ON(!buffer_dirty(bh));
546 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
547 }
548
549 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
550 struct nilfs_segsum_pointer *ssp,
551 union nilfs_binfo *binfo)
552 {
553 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
554 sci, ssp, sizeof(*binfo_v));
555 *binfo_v = binfo->bi_v;
556 }
557
558 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
559 struct nilfs_segsum_pointer *ssp,
560 union nilfs_binfo *binfo)
561 {
562 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
563 sci, ssp, sizeof(*vblocknr));
564 *vblocknr = binfo->bi_v.bi_vblocknr;
565 }
566
567 static struct nilfs_sc_operations nilfs_sc_file_ops = {
568 .collect_data = nilfs_collect_file_data,
569 .collect_node = nilfs_collect_file_node,
570 .collect_bmap = nilfs_collect_file_bmap,
571 .write_data_binfo = nilfs_write_file_data_binfo,
572 .write_node_binfo = nilfs_write_file_node_binfo,
573 };
574
575 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
576 struct buffer_head *bh, struct inode *inode)
577 {
578 int err;
579
580 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
581 if (unlikely(err < 0))
582 return nilfs_handle_bmap_error(err, __func__, inode,
583 sci->sc_super);
584
585 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
586 if (!err)
587 sci->sc_datablk_cnt++;
588 return err;
589 }
590
591 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
592 struct buffer_head *bh, struct inode *inode)
593 {
594 WARN_ON(!buffer_dirty(bh));
595 return nilfs_segctor_add_file_block(sci, bh, inode,
596 sizeof(struct nilfs_binfo_dat));
597 }
598
599 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
600 struct nilfs_segsum_pointer *ssp,
601 union nilfs_binfo *binfo)
602 {
603 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
604 sizeof(*blkoff));
605 *blkoff = binfo->bi_dat.bi_blkoff;
606 }
607
608 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
609 struct nilfs_segsum_pointer *ssp,
610 union nilfs_binfo *binfo)
611 {
612 struct nilfs_binfo_dat *binfo_dat =
613 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
614 *binfo_dat = binfo->bi_dat;
615 }
616
617 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
618 .collect_data = nilfs_collect_dat_data,
619 .collect_node = nilfs_collect_file_node,
620 .collect_bmap = nilfs_collect_dat_bmap,
621 .write_data_binfo = nilfs_write_dat_data_binfo,
622 .write_node_binfo = nilfs_write_dat_node_binfo,
623 };
624
625 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
626 .collect_data = nilfs_collect_file_data,
627 .collect_node = NULL,
628 .collect_bmap = NULL,
629 .write_data_binfo = nilfs_write_file_data_binfo,
630 .write_node_binfo = NULL,
631 };
632
633 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
634 struct list_head *listp,
635 size_t nlimit,
636 loff_t start, loff_t end)
637 {
638 struct address_space *mapping = inode->i_mapping;
639 struct pagevec pvec;
640 pgoff_t index = 0, last = ULONG_MAX;
641 size_t ndirties = 0;
642 int i;
643
644 if (unlikely(start != 0 || end != LLONG_MAX)) {
645 /*
646 * A valid range is given for sync-ing data pages. The
647 * range is rounded to per-page; extra dirty buffers
648 * may be included if blocksize < pagesize.
649 */
650 index = start >> PAGE_SHIFT;
651 last = end >> PAGE_SHIFT;
652 }
653 pagevec_init(&pvec, 0);
654 repeat:
655 if (unlikely(index > last) ||
656 !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
657 min_t(pgoff_t, last - index,
658 PAGEVEC_SIZE - 1) + 1))
659 return ndirties;
660
661 for (i = 0; i < pagevec_count(&pvec); i++) {
662 struct buffer_head *bh, *head;
663 struct page *page = pvec.pages[i];
664
665 if (unlikely(page->index > last))
666 break;
667
668 if (mapping->host) {
669 lock_page(page);
670 if (!page_has_buffers(page))
671 create_empty_buffers(page,
672 1 << inode->i_blkbits, 0);
673 unlock_page(page);
674 }
675
676 bh = head = page_buffers(page);
677 do {
678 if (!buffer_dirty(bh))
679 continue;
680 get_bh(bh);
681 list_add_tail(&bh->b_assoc_buffers, listp);
682 ndirties++;
683 if (unlikely(ndirties >= nlimit)) {
684 pagevec_release(&pvec);
685 cond_resched();
686 return ndirties;
687 }
688 } while (bh = bh->b_this_page, bh != head);
689 }
690 pagevec_release(&pvec);
691 cond_resched();
692 goto repeat;
693 }
694
695 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
696 struct list_head *listp)
697 {
698 struct nilfs_inode_info *ii = NILFS_I(inode);
699 struct address_space *mapping = &ii->i_btnode_cache;
700 struct pagevec pvec;
701 struct buffer_head *bh, *head;
702 unsigned int i;
703 pgoff_t index = 0;
704
705 pagevec_init(&pvec, 0);
706
707 while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
708 PAGEVEC_SIZE)) {
709 for (i = 0; i < pagevec_count(&pvec); i++) {
710 bh = head = page_buffers(pvec.pages[i]);
711 do {
712 if (buffer_dirty(bh)) {
713 get_bh(bh);
714 list_add_tail(&bh->b_assoc_buffers,
715 listp);
716 }
717 bh = bh->b_this_page;
718 } while (bh != head);
719 }
720 pagevec_release(&pvec);
721 cond_resched();
722 }
723 }
724
725 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
726 struct list_head *head, int force)
727 {
728 struct nilfs_inode_info *ii, *n;
729 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
730 unsigned nv = 0;
731
732 while (!list_empty(head)) {
733 spin_lock(&sbi->s_inode_lock);
734 list_for_each_entry_safe(ii, n, head, i_dirty) {
735 list_del_init(&ii->i_dirty);
736 if (force) {
737 if (unlikely(ii->i_bh)) {
738 brelse(ii->i_bh);
739 ii->i_bh = NULL;
740 }
741 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
742 set_bit(NILFS_I_QUEUED, &ii->i_state);
743 list_add_tail(&ii->i_dirty,
744 &sbi->s_dirty_files);
745 continue;
746 }
747 ivec[nv++] = ii;
748 if (nv == SC_N_INODEVEC)
749 break;
750 }
751 spin_unlock(&sbi->s_inode_lock);
752
753 for (pii = ivec; nv > 0; pii++, nv--)
754 iput(&(*pii)->vfs_inode);
755 }
756 }
757
758 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
759 {
760 struct the_nilfs *nilfs = sbi->s_nilfs;
761 int ret = 0;
762
763 if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
764 ret++;
765 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
766 ret++;
767 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
768 ret++;
769 if (ret || nilfs_doing_gc())
770 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
771 ret++;
772 return ret;
773 }
774
775 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
776 {
777 return list_empty(&sci->sc_dirty_files) &&
778 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
779 sci->sc_nfreesegs == 0 &&
780 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
781 }
782
783 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
784 {
785 struct nilfs_sb_info *sbi = sci->sc_sbi;
786 int ret = 0;
787
788 if (nilfs_test_metadata_dirty(sbi))
789 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
790
791 spin_lock(&sbi->s_inode_lock);
792 if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
793 ret++;
794
795 spin_unlock(&sbi->s_inode_lock);
796 return ret;
797 }
798
799 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
800 {
801 struct nilfs_sb_info *sbi = sci->sc_sbi;
802 struct the_nilfs *nilfs = sbi->s_nilfs;
803
804 nilfs_mdt_clear_dirty(sbi->s_ifile);
805 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
806 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
807 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
808 }
809
810 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
811 {
812 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
813 struct buffer_head *bh_cp;
814 struct nilfs_checkpoint *raw_cp;
815 int err;
816
817 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
818 &raw_cp, &bh_cp);
819 if (likely(!err)) {
820 /* The following code is duplicated with cpfile. But, it is
821 needed to collect the checkpoint even if it was not newly
822 created */
823 nilfs_mdt_mark_buffer_dirty(bh_cp);
824 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
825 nilfs_cpfile_put_checkpoint(
826 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
827 } else
828 WARN_ON(err == -EINVAL || err == -ENOENT);
829
830 return err;
831 }
832
833 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
834 {
835 struct nilfs_sb_info *sbi = sci->sc_sbi;
836 struct the_nilfs *nilfs = sbi->s_nilfs;
837 struct buffer_head *bh_cp;
838 struct nilfs_checkpoint *raw_cp;
839 int err;
840
841 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
842 &raw_cp, &bh_cp);
843 if (unlikely(err)) {
844 WARN_ON(err == -EINVAL || err == -ENOENT);
845 goto failed_ibh;
846 }
847 raw_cp->cp_snapshot_list.ssl_next = 0;
848 raw_cp->cp_snapshot_list.ssl_prev = 0;
849 raw_cp->cp_inodes_count =
850 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
851 raw_cp->cp_blocks_count =
852 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
853 raw_cp->cp_nblk_inc =
854 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
855 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
856 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
857
858 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
859 nilfs_checkpoint_clear_minor(raw_cp);
860 else
861 nilfs_checkpoint_set_minor(raw_cp);
862
863 nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
864 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
865 return 0;
866
867 failed_ibh:
868 return err;
869 }
870
871 static void nilfs_fill_in_file_bmap(struct inode *ifile,
872 struct nilfs_inode_info *ii)
873
874 {
875 struct buffer_head *ibh;
876 struct nilfs_inode *raw_inode;
877
878 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
879 ibh = ii->i_bh;
880 BUG_ON(!ibh);
881 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
882 ibh);
883 nilfs_bmap_write(ii->i_bmap, raw_inode);
884 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
885 }
886 }
887
888 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
889 struct inode *ifile)
890 {
891 struct nilfs_inode_info *ii;
892
893 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
894 nilfs_fill_in_file_bmap(ifile, ii);
895 set_bit(NILFS_I_COLLECTED, &ii->i_state);
896 }
897 }
898
899 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
900 struct the_nilfs *nilfs)
901 {
902 struct buffer_head *bh_sr;
903 struct nilfs_super_root *raw_sr;
904 unsigned isz = nilfs->ns_inode_size;
905
906 bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
907 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
908
909 raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
910 raw_sr->sr_nongc_ctime
911 = cpu_to_le64(nilfs_doing_gc() ?
912 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
913 raw_sr->sr_flags = 0;
914
915 nilfs_write_inode_common(nilfs_dat_inode(nilfs), (void *)raw_sr +
916 NILFS_SR_DAT_OFFSET(isz), 1);
917 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
918 NILFS_SR_CPFILE_OFFSET(isz), 1);
919 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
920 NILFS_SR_SUFILE_OFFSET(isz), 1);
921 }
922
923 static void nilfs_redirty_inodes(struct list_head *head)
924 {
925 struct nilfs_inode_info *ii;
926
927 list_for_each_entry(ii, head, i_dirty) {
928 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
929 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
930 }
931 }
932
933 static void nilfs_drop_collected_inodes(struct list_head *head)
934 {
935 struct nilfs_inode_info *ii;
936
937 list_for_each_entry(ii, head, i_dirty) {
938 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
939 continue;
940
941 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
942 set_bit(NILFS_I_UPDATED, &ii->i_state);
943 }
944 }
945
946 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
947 struct inode *inode,
948 struct list_head *listp,
949 int (*collect)(struct nilfs_sc_info *,
950 struct buffer_head *,
951 struct inode *))
952 {
953 struct buffer_head *bh, *n;
954 int err = 0;
955
956 if (collect) {
957 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
958 list_del_init(&bh->b_assoc_buffers);
959 err = collect(sci, bh, inode);
960 brelse(bh);
961 if (unlikely(err))
962 goto dispose_buffers;
963 }
964 return 0;
965 }
966
967 dispose_buffers:
968 while (!list_empty(listp)) {
969 bh = list_entry(listp->next, struct buffer_head,
970 b_assoc_buffers);
971 list_del_init(&bh->b_assoc_buffers);
972 brelse(bh);
973 }
974 return err;
975 }
976
977 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
978 {
979 /* Remaining number of blocks within segment buffer */
980 return sci->sc_segbuf_nblocks -
981 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
982 }
983
984 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
985 struct inode *inode,
986 struct nilfs_sc_operations *sc_ops)
987 {
988 LIST_HEAD(data_buffers);
989 LIST_HEAD(node_buffers);
990 int err;
991
992 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
993 size_t n, rest = nilfs_segctor_buffer_rest(sci);
994
995 n = nilfs_lookup_dirty_data_buffers(
996 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
997 if (n > rest) {
998 err = nilfs_segctor_apply_buffers(
999 sci, inode, &data_buffers,
1000 sc_ops->collect_data);
1001 BUG_ON(!err); /* always receive -E2BIG or true error */
1002 goto break_or_fail;
1003 }
1004 }
1005 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1006
1007 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1008 err = nilfs_segctor_apply_buffers(
1009 sci, inode, &data_buffers, sc_ops->collect_data);
1010 if (unlikely(err)) {
1011 /* dispose node list */
1012 nilfs_segctor_apply_buffers(
1013 sci, inode, &node_buffers, NULL);
1014 goto break_or_fail;
1015 }
1016 sci->sc_stage.flags |= NILFS_CF_NODE;
1017 }
1018 /* Collect node */
1019 err = nilfs_segctor_apply_buffers(
1020 sci, inode, &node_buffers, sc_ops->collect_node);
1021 if (unlikely(err))
1022 goto break_or_fail;
1023
1024 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1025 err = nilfs_segctor_apply_buffers(
1026 sci, inode, &node_buffers, sc_ops->collect_bmap);
1027 if (unlikely(err))
1028 goto break_or_fail;
1029
1030 nilfs_segctor_end_finfo(sci, inode);
1031 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1032
1033 break_or_fail:
1034 return err;
1035 }
1036
1037 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1038 struct inode *inode)
1039 {
1040 LIST_HEAD(data_buffers);
1041 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1042 int err;
1043
1044 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1045 sci->sc_dsync_start,
1046 sci->sc_dsync_end);
1047
1048 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1049 nilfs_collect_file_data);
1050 if (!err) {
1051 nilfs_segctor_end_finfo(sci, inode);
1052 BUG_ON(n > rest);
1053 /* always receive -E2BIG or true error if n > rest */
1054 }
1055 return err;
1056 }
1057
1058 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1059 {
1060 struct nilfs_sb_info *sbi = sci->sc_sbi;
1061 struct the_nilfs *nilfs = sbi->s_nilfs;
1062 struct list_head *head;
1063 struct nilfs_inode_info *ii;
1064 size_t ndone;
1065 int err = 0;
1066
1067 switch (sci->sc_stage.scnt) {
1068 case NILFS_ST_INIT:
1069 /* Pre-processes */
1070 sci->sc_stage.flags = 0;
1071
1072 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1073 sci->sc_nblk_inc = 0;
1074 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1075 if (mode == SC_LSEG_DSYNC) {
1076 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1077 goto dsync_mode;
1078 }
1079 }
1080
1081 sci->sc_stage.dirty_file_ptr = NULL;
1082 sci->sc_stage.gc_inode_ptr = NULL;
1083 if (mode == SC_FLUSH_DAT) {
1084 sci->sc_stage.scnt = NILFS_ST_DAT;
1085 goto dat_stage;
1086 }
1087 sci->sc_stage.scnt++; /* Fall through */
1088 case NILFS_ST_GC:
1089 if (nilfs_doing_gc()) {
1090 head = &sci->sc_gc_inodes;
1091 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1092 head, i_dirty);
1093 list_for_each_entry_continue(ii, head, i_dirty) {
1094 err = nilfs_segctor_scan_file(
1095 sci, &ii->vfs_inode,
1096 &nilfs_sc_file_ops);
1097 if (unlikely(err)) {
1098 sci->sc_stage.gc_inode_ptr = list_entry(
1099 ii->i_dirty.prev,
1100 struct nilfs_inode_info,
1101 i_dirty);
1102 goto break_or_fail;
1103 }
1104 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1105 }
1106 sci->sc_stage.gc_inode_ptr = NULL;
1107 }
1108 sci->sc_stage.scnt++; /* Fall through */
1109 case NILFS_ST_FILE:
1110 head = &sci->sc_dirty_files;
1111 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1112 i_dirty);
1113 list_for_each_entry_continue(ii, head, i_dirty) {
1114 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1115
1116 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1117 &nilfs_sc_file_ops);
1118 if (unlikely(err)) {
1119 sci->sc_stage.dirty_file_ptr =
1120 list_entry(ii->i_dirty.prev,
1121 struct nilfs_inode_info,
1122 i_dirty);
1123 goto break_or_fail;
1124 }
1125 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1126 }
1127 sci->sc_stage.dirty_file_ptr = NULL;
1128 if (mode == SC_FLUSH_FILE) {
1129 sci->sc_stage.scnt = NILFS_ST_DONE;
1130 return 0;
1131 }
1132 sci->sc_stage.scnt++;
1133 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1134 /* Fall through */
1135 case NILFS_ST_IFILE:
1136 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1137 &nilfs_sc_file_ops);
1138 if (unlikely(err))
1139 break;
1140 sci->sc_stage.scnt++;
1141 /* Creating a checkpoint */
1142 err = nilfs_segctor_create_checkpoint(sci);
1143 if (unlikely(err))
1144 break;
1145 /* Fall through */
1146 case NILFS_ST_CPFILE:
1147 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1148 &nilfs_sc_file_ops);
1149 if (unlikely(err))
1150 break;
1151 sci->sc_stage.scnt++; /* Fall through */
1152 case NILFS_ST_SUFILE:
1153 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1154 sci->sc_nfreesegs, &ndone);
1155 if (unlikely(err)) {
1156 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1157 sci->sc_freesegs, ndone,
1158 NULL);
1159 break;
1160 }
1161 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1162
1163 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1164 &nilfs_sc_file_ops);
1165 if (unlikely(err))
1166 break;
1167 sci->sc_stage.scnt++; /* Fall through */
1168 case NILFS_ST_DAT:
1169 dat_stage:
1170 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1171 &nilfs_sc_dat_ops);
1172 if (unlikely(err))
1173 break;
1174 if (mode == SC_FLUSH_DAT) {
1175 sci->sc_stage.scnt = NILFS_ST_DONE;
1176 return 0;
1177 }
1178 sci->sc_stage.scnt++; /* Fall through */
1179 case NILFS_ST_SR:
1180 if (mode == SC_LSEG_SR) {
1181 /* Appending a super root */
1182 err = nilfs_segctor_add_super_root(sci);
1183 if (unlikely(err))
1184 break;
1185 }
1186 /* End of a logical segment */
1187 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1188 sci->sc_stage.scnt = NILFS_ST_DONE;
1189 return 0;
1190 case NILFS_ST_DSYNC:
1191 dsync_mode:
1192 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1193 ii = sci->sc_dsync_inode;
1194 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1195 break;
1196
1197 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1198 if (unlikely(err))
1199 break;
1200 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1201 sci->sc_stage.scnt = NILFS_ST_DONE;
1202 return 0;
1203 case NILFS_ST_DONE:
1204 return 0;
1205 default:
1206 BUG();
1207 }
1208
1209 break_or_fail:
1210 return err;
1211 }
1212
1213 /**
1214 * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1215 * @sci: nilfs_sc_info
1216 * @nilfs: nilfs object
1217 */
1218 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1219 struct the_nilfs *nilfs)
1220 {
1221 struct nilfs_segment_buffer *segbuf, *prev;
1222 __u64 nextnum;
1223 int err, alloc = 0;
1224
1225 segbuf = nilfs_segbuf_new(sci->sc_super);
1226 if (unlikely(!segbuf))
1227 return -ENOMEM;
1228
1229 if (list_empty(&sci->sc_write_logs)) {
1230 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1231 nilfs->ns_pseg_offset, nilfs);
1232 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1233 nilfs_shift_to_next_segment(nilfs);
1234 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1235 }
1236
1237 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1238 nextnum = nilfs->ns_nextnum;
1239
1240 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1241 /* Start from the head of a new full segment */
1242 alloc++;
1243 } else {
1244 /* Continue logs */
1245 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1246 nilfs_segbuf_map_cont(segbuf, prev);
1247 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1248 nextnum = prev->sb_nextnum;
1249
1250 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1251 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1252 segbuf->sb_sum.seg_seq++;
1253 alloc++;
1254 }
1255 }
1256
1257 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1258 if (err)
1259 goto failed;
1260
1261 if (alloc) {
1262 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1263 if (err)
1264 goto failed;
1265 }
1266 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1267
1268 BUG_ON(!list_empty(&sci->sc_segbufs));
1269 list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1270 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1271 return 0;
1272
1273 failed:
1274 nilfs_segbuf_free(segbuf);
1275 return err;
1276 }
1277
1278 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1279 struct the_nilfs *nilfs, int nadd)
1280 {
1281 struct nilfs_segment_buffer *segbuf, *prev;
1282 struct inode *sufile = nilfs->ns_sufile;
1283 __u64 nextnextnum;
1284 LIST_HEAD(list);
1285 int err, ret, i;
1286
1287 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1288 /*
1289 * Since the segment specified with nextnum might be allocated during
1290 * the previous construction, the buffer including its segusage may
1291 * not be dirty. The following call ensures that the buffer is dirty
1292 * and will pin the buffer on memory until the sufile is written.
1293 */
1294 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1295 if (unlikely(err))
1296 return err;
1297
1298 for (i = 0; i < nadd; i++) {
1299 /* extend segment info */
1300 err = -ENOMEM;
1301 segbuf = nilfs_segbuf_new(sci->sc_super);
1302 if (unlikely(!segbuf))
1303 goto failed;
1304
1305 /* map this buffer to region of segment on-disk */
1306 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1307 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1308
1309 /* allocate the next next full segment */
1310 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1311 if (unlikely(err))
1312 goto failed_segbuf;
1313
1314 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1315 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1316
1317 list_add_tail(&segbuf->sb_list, &list);
1318 prev = segbuf;
1319 }
1320 list_splice_tail(&list, &sci->sc_segbufs);
1321 return 0;
1322
1323 failed_segbuf:
1324 nilfs_segbuf_free(segbuf);
1325 failed:
1326 list_for_each_entry(segbuf, &list, sb_list) {
1327 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1328 WARN_ON(ret); /* never fails */
1329 }
1330 nilfs_destroy_logs(&list);
1331 return err;
1332 }
1333
1334 static void nilfs_free_incomplete_logs(struct list_head *logs,
1335 struct the_nilfs *nilfs)
1336 {
1337 struct nilfs_segment_buffer *segbuf, *prev;
1338 struct inode *sufile = nilfs->ns_sufile;
1339 int ret;
1340
1341 segbuf = NILFS_FIRST_SEGBUF(logs);
1342 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1343 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1344 WARN_ON(ret); /* never fails */
1345 }
1346 if (atomic_read(&segbuf->sb_err)) {
1347 /* Case 1: The first segment failed */
1348 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1349 /* Case 1a: Partial segment appended into an existing
1350 segment */
1351 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1352 segbuf->sb_fseg_end);
1353 else /* Case 1b: New full segment */
1354 set_nilfs_discontinued(nilfs);
1355 }
1356
1357 prev = segbuf;
1358 list_for_each_entry_continue(segbuf, logs, sb_list) {
1359 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1360 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1361 WARN_ON(ret); /* never fails */
1362 }
1363 if (atomic_read(&segbuf->sb_err) &&
1364 segbuf->sb_segnum != nilfs->ns_nextnum)
1365 /* Case 2: extended segment (!= next) failed */
1366 nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1367 prev = segbuf;
1368 }
1369 }
1370
1371 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1372 struct inode *sufile)
1373 {
1374 struct nilfs_segment_buffer *segbuf;
1375 unsigned long live_blocks;
1376 int ret;
1377
1378 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1379 live_blocks = segbuf->sb_sum.nblocks +
1380 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1381 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1382 live_blocks,
1383 sci->sc_seg_ctime);
1384 WARN_ON(ret); /* always succeed because the segusage is dirty */
1385 }
1386 }
1387
1388 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1389 {
1390 struct nilfs_segment_buffer *segbuf;
1391 int ret;
1392
1393 segbuf = NILFS_FIRST_SEGBUF(logs);
1394 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1395 segbuf->sb_pseg_start -
1396 segbuf->sb_fseg_start, 0);
1397 WARN_ON(ret); /* always succeed because the segusage is dirty */
1398
1399 list_for_each_entry_continue(segbuf, logs, sb_list) {
1400 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1401 0, 0);
1402 WARN_ON(ret); /* always succeed */
1403 }
1404 }
1405
1406 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1407 struct nilfs_segment_buffer *last,
1408 struct inode *sufile)
1409 {
1410 struct nilfs_segment_buffer *segbuf = last;
1411 int ret;
1412
1413 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1414 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1415 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1416 WARN_ON(ret);
1417 }
1418 nilfs_truncate_logs(&sci->sc_segbufs, last);
1419 }
1420
1421
1422 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1423 struct the_nilfs *nilfs, int mode)
1424 {
1425 struct nilfs_cstage prev_stage = sci->sc_stage;
1426 int err, nadd = 1;
1427
1428 /* Collection retry loop */
1429 for (;;) {
1430 sci->sc_nblk_this_inc = 0;
1431 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1432
1433 err = nilfs_segctor_reset_segment_buffer(sci);
1434 if (unlikely(err))
1435 goto failed;
1436
1437 err = nilfs_segctor_collect_blocks(sci, mode);
1438 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1439 if (!err)
1440 break;
1441
1442 if (unlikely(err != -E2BIG))
1443 goto failed;
1444
1445 /* The current segment is filled up */
1446 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1447 break;
1448
1449 nilfs_clear_logs(&sci->sc_segbufs);
1450
1451 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1452 if (unlikely(err))
1453 return err;
1454
1455 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1456 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1457 sci->sc_freesegs,
1458 sci->sc_nfreesegs,
1459 NULL);
1460 WARN_ON(err); /* do not happen */
1461 }
1462 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1463 sci->sc_stage = prev_stage;
1464 }
1465 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1466 return 0;
1467
1468 failed:
1469 return err;
1470 }
1471
1472 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1473 struct buffer_head *new_bh)
1474 {
1475 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1476
1477 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1478 /* The caller must release old_bh */
1479 }
1480
1481 static int
1482 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1483 struct nilfs_segment_buffer *segbuf,
1484 int mode)
1485 {
1486 struct inode *inode = NULL;
1487 sector_t blocknr;
1488 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1489 unsigned long nblocks = 0, ndatablk = 0;
1490 struct nilfs_sc_operations *sc_op = NULL;
1491 struct nilfs_segsum_pointer ssp;
1492 struct nilfs_finfo *finfo = NULL;
1493 union nilfs_binfo binfo;
1494 struct buffer_head *bh, *bh_org;
1495 ino_t ino = 0;
1496 int err = 0;
1497
1498 if (!nfinfo)
1499 goto out;
1500
1501 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1502 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1503 ssp.offset = sizeof(struct nilfs_segment_summary);
1504
1505 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1506 if (bh == segbuf->sb_super_root)
1507 break;
1508 if (!finfo) {
1509 finfo = nilfs_segctor_map_segsum_entry(
1510 sci, &ssp, sizeof(*finfo));
1511 ino = le64_to_cpu(finfo->fi_ino);
1512 nblocks = le32_to_cpu(finfo->fi_nblocks);
1513 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1514
1515 if (buffer_nilfs_node(bh))
1516 inode = NILFS_BTNC_I(bh->b_page->mapping);
1517 else
1518 inode = NILFS_AS_I(bh->b_page->mapping);
1519
1520 if (mode == SC_LSEG_DSYNC)
1521 sc_op = &nilfs_sc_dsync_ops;
1522 else if (ino == NILFS_DAT_INO)
1523 sc_op = &nilfs_sc_dat_ops;
1524 else /* file blocks */
1525 sc_op = &nilfs_sc_file_ops;
1526 }
1527 bh_org = bh;
1528 get_bh(bh_org);
1529 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1530 &binfo);
1531 if (bh != bh_org)
1532 nilfs_list_replace_buffer(bh_org, bh);
1533 brelse(bh_org);
1534 if (unlikely(err))
1535 goto failed_bmap;
1536
1537 if (ndatablk > 0)
1538 sc_op->write_data_binfo(sci, &ssp, &binfo);
1539 else
1540 sc_op->write_node_binfo(sci, &ssp, &binfo);
1541
1542 blocknr++;
1543 if (--nblocks == 0) {
1544 finfo = NULL;
1545 if (--nfinfo == 0)
1546 break;
1547 } else if (ndatablk > 0)
1548 ndatablk--;
1549 }
1550 out:
1551 return 0;
1552
1553 failed_bmap:
1554 err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1555 return err;
1556 }
1557
1558 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1559 {
1560 struct nilfs_segment_buffer *segbuf;
1561 int err;
1562
1563 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1564 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1565 if (unlikely(err))
1566 return err;
1567 nilfs_segbuf_fill_in_segsum(segbuf);
1568 }
1569 return 0;
1570 }
1571
1572 static int
1573 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1574 {
1575 struct page *clone_page;
1576 struct buffer_head *bh, *head, *bh2;
1577 void *kaddr;
1578
1579 bh = head = page_buffers(page);
1580
1581 clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1582 if (unlikely(!clone_page))
1583 return -ENOMEM;
1584
1585 bh2 = page_buffers(clone_page);
1586 kaddr = kmap_atomic(page, KM_USER0);
1587 do {
1588 if (list_empty(&bh->b_assoc_buffers))
1589 continue;
1590 get_bh(bh2);
1591 page_cache_get(clone_page); /* for each bh */
1592 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1593 bh2->b_blocknr = bh->b_blocknr;
1594 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1595 list_add_tail(&bh->b_assoc_buffers, out);
1596 } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1597 kunmap_atomic(kaddr, KM_USER0);
1598
1599 if (!TestSetPageWriteback(clone_page))
1600 inc_zone_page_state(clone_page, NR_WRITEBACK);
1601 unlock_page(clone_page);
1602
1603 return 0;
1604 }
1605
1606 static int nilfs_test_page_to_be_frozen(struct page *page)
1607 {
1608 struct address_space *mapping = page->mapping;
1609
1610 if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1611 return 0;
1612
1613 if (page_mapped(page)) {
1614 ClearPageChecked(page);
1615 return 1;
1616 }
1617 return PageChecked(page);
1618 }
1619
1620 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1621 {
1622 if (!page || PageWriteback(page))
1623 /* For split b-tree node pages, this function may be called
1624 twice. We ignore the 2nd or later calls by this check. */
1625 return 0;
1626
1627 lock_page(page);
1628 clear_page_dirty_for_io(page);
1629 set_page_writeback(page);
1630 unlock_page(page);
1631
1632 if (nilfs_test_page_to_be_frozen(page)) {
1633 int err = nilfs_copy_replace_page_buffers(page, out);
1634 if (unlikely(err))
1635 return err;
1636 }
1637 return 0;
1638 }
1639
1640 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1641 struct page **failed_page)
1642 {
1643 struct nilfs_segment_buffer *segbuf;
1644 struct page *bd_page = NULL, *fs_page = NULL;
1645 struct list_head *list = &sci->sc_copied_buffers;
1646 int err;
1647
1648 *failed_page = NULL;
1649 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1650 struct buffer_head *bh;
1651
1652 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1653 b_assoc_buffers) {
1654 if (bh->b_page != bd_page) {
1655 if (bd_page) {
1656 lock_page(bd_page);
1657 clear_page_dirty_for_io(bd_page);
1658 set_page_writeback(bd_page);
1659 unlock_page(bd_page);
1660 }
1661 bd_page = bh->b_page;
1662 }
1663 }
1664
1665 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1666 b_assoc_buffers) {
1667 if (bh == segbuf->sb_super_root) {
1668 if (bh->b_page != bd_page) {
1669 lock_page(bd_page);
1670 clear_page_dirty_for_io(bd_page);
1671 set_page_writeback(bd_page);
1672 unlock_page(bd_page);
1673 bd_page = bh->b_page;
1674 }
1675 break;
1676 }
1677 if (bh->b_page != fs_page) {
1678 err = nilfs_begin_page_io(fs_page, list);
1679 if (unlikely(err)) {
1680 *failed_page = fs_page;
1681 goto out;
1682 }
1683 fs_page = bh->b_page;
1684 }
1685 }
1686 }
1687 if (bd_page) {
1688 lock_page(bd_page);
1689 clear_page_dirty_for_io(bd_page);
1690 set_page_writeback(bd_page);
1691 unlock_page(bd_page);
1692 }
1693 err = nilfs_begin_page_io(fs_page, list);
1694 if (unlikely(err))
1695 *failed_page = fs_page;
1696 out:
1697 return err;
1698 }
1699
1700 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1701 struct the_nilfs *nilfs)
1702 {
1703 int ret;
1704
1705 ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1706 list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1707 return ret;
1708 }
1709
1710 static void __nilfs_end_page_io(struct page *page, int err)
1711 {
1712 if (!err) {
1713 if (!nilfs_page_buffers_clean(page))
1714 __set_page_dirty_nobuffers(page);
1715 ClearPageError(page);
1716 } else {
1717 __set_page_dirty_nobuffers(page);
1718 SetPageError(page);
1719 }
1720
1721 if (buffer_nilfs_allocated(page_buffers(page))) {
1722 if (TestClearPageWriteback(page))
1723 dec_zone_page_state(page, NR_WRITEBACK);
1724 } else
1725 end_page_writeback(page);
1726 }
1727
1728 static void nilfs_end_page_io(struct page *page, int err)
1729 {
1730 if (!page)
1731 return;
1732
1733 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1734 /*
1735 * For b-tree node pages, this function may be called twice
1736 * or more because they might be split in a segment.
1737 */
1738 if (PageDirty(page)) {
1739 /*
1740 * For pages holding split b-tree node buffers, dirty
1741 * flag on the buffers may be cleared discretely.
1742 * In that case, the page is once redirtied for
1743 * remaining buffers, and it must be cancelled if
1744 * all the buffers get cleaned later.
1745 */
1746 lock_page(page);
1747 if (nilfs_page_buffers_clean(page))
1748 __nilfs_clear_page_dirty(page);
1749 unlock_page(page);
1750 }
1751 return;
1752 }
1753
1754 __nilfs_end_page_io(page, err);
1755 }
1756
1757 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1758 {
1759 struct buffer_head *bh, *head;
1760 struct page *page;
1761
1762 while (!list_empty(list)) {
1763 bh = list_entry(list->next, struct buffer_head,
1764 b_assoc_buffers);
1765 page = bh->b_page;
1766 page_cache_get(page);
1767 head = bh = page_buffers(page);
1768 do {
1769 if (!list_empty(&bh->b_assoc_buffers)) {
1770 list_del_init(&bh->b_assoc_buffers);
1771 if (!err) {
1772 set_buffer_uptodate(bh);
1773 clear_buffer_dirty(bh);
1774 clear_buffer_nilfs_volatile(bh);
1775 }
1776 brelse(bh); /* for b_assoc_buffers */
1777 }
1778 } while ((bh = bh->b_this_page) != head);
1779
1780 __nilfs_end_page_io(page, err);
1781 page_cache_release(page);
1782 }
1783 }
1784
1785 static void nilfs_abort_logs(struct list_head *logs, struct page *failed_page,
1786 int err)
1787 {
1788 struct nilfs_segment_buffer *segbuf;
1789 struct page *bd_page = NULL, *fs_page = NULL;
1790 struct buffer_head *bh;
1791
1792 if (list_empty(logs))
1793 return;
1794
1795 list_for_each_entry(segbuf, logs, sb_list) {
1796 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1797 b_assoc_buffers) {
1798 if (bh->b_page != bd_page) {
1799 if (bd_page)
1800 end_page_writeback(bd_page);
1801 bd_page = bh->b_page;
1802 }
1803 }
1804
1805 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1806 b_assoc_buffers) {
1807 if (bh == segbuf->sb_super_root) {
1808 if (bh->b_page != bd_page) {
1809 end_page_writeback(bd_page);
1810 bd_page = bh->b_page;
1811 }
1812 break;
1813 }
1814 if (bh->b_page != fs_page) {
1815 nilfs_end_page_io(fs_page, err);
1816 if (fs_page && fs_page == failed_page)
1817 return;
1818 fs_page = bh->b_page;
1819 }
1820 }
1821 }
1822 if (bd_page)
1823 end_page_writeback(bd_page);
1824
1825 nilfs_end_page_io(fs_page, err);
1826 }
1827
1828 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1829 struct the_nilfs *nilfs, int err)
1830 {
1831 LIST_HEAD(logs);
1832 int ret;
1833
1834 list_splice_tail_init(&sci->sc_write_logs, &logs);
1835 ret = nilfs_wait_on_logs(&logs);
1836 nilfs_abort_logs(&logs, NULL, ret ? : err);
1837
1838 list_splice_tail_init(&sci->sc_segbufs, &logs);
1839 nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1840 nilfs_free_incomplete_logs(&logs, nilfs);
1841 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1842
1843 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1844 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1845 sci->sc_freesegs,
1846 sci->sc_nfreesegs,
1847 NULL);
1848 WARN_ON(ret); /* do not happen */
1849 }
1850
1851 nilfs_destroy_logs(&logs);
1852 }
1853
1854 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1855 struct nilfs_segment_buffer *segbuf)
1856 {
1857 nilfs->ns_segnum = segbuf->sb_segnum;
1858 nilfs->ns_nextnum = segbuf->sb_nextnum;
1859 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1860 + segbuf->sb_sum.nblocks;
1861 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1862 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1863 }
1864
1865 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1866 {
1867 struct nilfs_segment_buffer *segbuf;
1868 struct page *bd_page = NULL, *fs_page = NULL;
1869 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
1870 int update_sr = false;
1871
1872 list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1873 struct buffer_head *bh;
1874
1875 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1876 b_assoc_buffers) {
1877 set_buffer_uptodate(bh);
1878 clear_buffer_dirty(bh);
1879 if (bh->b_page != bd_page) {
1880 if (bd_page)
1881 end_page_writeback(bd_page);
1882 bd_page = bh->b_page;
1883 }
1884 }
1885 /*
1886 * We assume that the buffers which belong to the same page
1887 * continue over the buffer list.
1888 * Under this assumption, the last BHs of pages is
1889 * identifiable by the discontinuity of bh->b_page
1890 * (page != fs_page).
1891 *
1892 * For B-tree node blocks, however, this assumption is not
1893 * guaranteed. The cleanup code of B-tree node pages needs
1894 * special care.
1895 */
1896 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1897 b_assoc_buffers) {
1898 set_buffer_uptodate(bh);
1899 clear_buffer_dirty(bh);
1900 clear_buffer_nilfs_volatile(bh);
1901 if (bh == segbuf->sb_super_root) {
1902 if (bh->b_page != bd_page) {
1903 end_page_writeback(bd_page);
1904 bd_page = bh->b_page;
1905 }
1906 update_sr = true;
1907 break;
1908 }
1909 if (bh->b_page != fs_page) {
1910 nilfs_end_page_io(fs_page, 0);
1911 fs_page = bh->b_page;
1912 }
1913 }
1914
1915 if (!nilfs_segbuf_simplex(segbuf)) {
1916 if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1917 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1918 sci->sc_lseg_stime = jiffies;
1919 }
1920 if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1921 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1922 }
1923 }
1924 /*
1925 * Since pages may continue over multiple segment buffers,
1926 * end of the last page must be checked outside of the loop.
1927 */
1928 if (bd_page)
1929 end_page_writeback(bd_page);
1930
1931 nilfs_end_page_io(fs_page, 0);
1932
1933 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
1934
1935 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1936
1937 if (nilfs_doing_gc()) {
1938 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1939 if (update_sr)
1940 nilfs_commit_gcdat_inode(nilfs);
1941 } else
1942 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1943
1944 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1945
1946 segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1947 nilfs_set_next_segment(nilfs, segbuf);
1948
1949 if (update_sr) {
1950 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1951 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1952
1953 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1954 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1955 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1956 nilfs_segctor_clear_metadata_dirty(sci);
1957 } else
1958 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1959 }
1960
1961 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1962 {
1963 int ret;
1964
1965 ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1966 if (!ret) {
1967 nilfs_segctor_complete_write(sci);
1968 nilfs_destroy_logs(&sci->sc_write_logs);
1969 }
1970 return ret;
1971 }
1972
1973 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
1974 struct nilfs_sb_info *sbi)
1975 {
1976 struct nilfs_inode_info *ii, *n;
1977 __u64 cno = sbi->s_nilfs->ns_cno;
1978
1979 spin_lock(&sbi->s_inode_lock);
1980 retry:
1981 list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
1982 if (!ii->i_bh) {
1983 struct buffer_head *ibh;
1984 int err;
1985
1986 spin_unlock(&sbi->s_inode_lock);
1987 err = nilfs_ifile_get_inode_block(
1988 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
1989 if (unlikely(err)) {
1990 nilfs_warning(sbi->s_super, __func__,
1991 "failed to get inode block.\n");
1992 return err;
1993 }
1994 nilfs_mdt_mark_buffer_dirty(ibh);
1995 nilfs_mdt_mark_dirty(sbi->s_ifile);
1996 spin_lock(&sbi->s_inode_lock);
1997 if (likely(!ii->i_bh))
1998 ii->i_bh = ibh;
1999 else
2000 brelse(ibh);
2001 goto retry;
2002 }
2003 ii->i_cno = cno;
2004
2005 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2006 set_bit(NILFS_I_BUSY, &ii->i_state);
2007 list_del(&ii->i_dirty);
2008 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2009 }
2010 spin_unlock(&sbi->s_inode_lock);
2011
2012 NILFS_I(sbi->s_ifile)->i_cno = cno;
2013
2014 return 0;
2015 }
2016
2017 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2018 struct nilfs_sb_info *sbi)
2019 {
2020 struct nilfs_transaction_info *ti = current->journal_info;
2021 struct nilfs_inode_info *ii, *n;
2022 __u64 cno = sbi->s_nilfs->ns_cno;
2023
2024 spin_lock(&sbi->s_inode_lock);
2025 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2026 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2027 test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2028 /* The current checkpoint number (=nilfs->ns_cno) is
2029 changed between check-in and check-out only if the
2030 super root is written out. So, we can update i_cno
2031 for the inodes that remain in the dirty list. */
2032 ii->i_cno = cno;
2033 continue;
2034 }
2035 clear_bit(NILFS_I_BUSY, &ii->i_state);
2036 brelse(ii->i_bh);
2037 ii->i_bh = NULL;
2038 list_del(&ii->i_dirty);
2039 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2040 }
2041 spin_unlock(&sbi->s_inode_lock);
2042 }
2043
2044 /*
2045 * Main procedure of segment constructor
2046 */
2047 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2048 {
2049 struct nilfs_sb_info *sbi = sci->sc_sbi;
2050 struct the_nilfs *nilfs = sbi->s_nilfs;
2051 struct page *failed_page;
2052 int err;
2053
2054 sci->sc_stage.scnt = NILFS_ST_INIT;
2055
2056 err = nilfs_segctor_check_in_files(sci, sbi);
2057 if (unlikely(err))
2058 goto out;
2059
2060 if (nilfs_test_metadata_dirty(sbi))
2061 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2062
2063 if (nilfs_segctor_clean(sci))
2064 goto out;
2065
2066 do {
2067 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2068
2069 err = nilfs_segctor_begin_construction(sci, nilfs);
2070 if (unlikely(err))
2071 goto out;
2072
2073 /* Update time stamp */
2074 sci->sc_seg_ctime = get_seconds();
2075
2076 err = nilfs_segctor_collect(sci, nilfs, mode);
2077 if (unlikely(err))
2078 goto failed;
2079
2080 /* Avoid empty segment */
2081 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2082 nilfs_segbuf_empty(sci->sc_curseg)) {
2083 nilfs_segctor_abort_construction(sci, nilfs, 1);
2084 goto out;
2085 }
2086
2087 err = nilfs_segctor_assign(sci, mode);
2088 if (unlikely(err))
2089 goto failed;
2090
2091 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2092 nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2093
2094 if (mode == SC_LSEG_SR &&
2095 sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
2096 err = nilfs_segctor_fill_in_checkpoint(sci);
2097 if (unlikely(err))
2098 goto failed_to_write;
2099
2100 nilfs_segctor_fill_in_super_root(sci, nilfs);
2101 }
2102 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2103
2104 /* Write partial segments */
2105 err = nilfs_segctor_prepare_write(sci, &failed_page);
2106 if (err) {
2107 nilfs_abort_logs(&sci->sc_segbufs, failed_page, err);
2108 goto failed_to_write;
2109 }
2110
2111 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2112 nilfs->ns_crc_seed);
2113
2114 err = nilfs_segctor_write(sci, nilfs);
2115 if (unlikely(err))
2116 goto failed_to_write;
2117
2118 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2119 nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2120 /*
2121 * At this point, we avoid double buffering
2122 * for blocksize < pagesize because page dirty
2123 * flag is turned off during write and dirty
2124 * buffers are not properly collected for
2125 * pages crossing over segments.
2126 */
2127 err = nilfs_segctor_wait(sci);
2128 if (err)
2129 goto failed_to_write;
2130 }
2131 } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2132
2133 out:
2134 nilfs_segctor_check_out_files(sci, sbi);
2135 return err;
2136
2137 failed_to_write:
2138 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2139 nilfs_redirty_inodes(&sci->sc_dirty_files);
2140
2141 failed:
2142 if (nilfs_doing_gc())
2143 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2144 nilfs_segctor_abort_construction(sci, nilfs, err);
2145 goto out;
2146 }
2147
2148 /**
2149 * nilfs_segctor_start_timer - set timer of background write
2150 * @sci: nilfs_sc_info
2151 *
2152 * If the timer has already been set, it ignores the new request.
2153 * This function MUST be called within a section locking the segment
2154 * semaphore.
2155 */
2156 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2157 {
2158 spin_lock(&sci->sc_state_lock);
2159 if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2160 sci->sc_timer.expires = jiffies + sci->sc_interval;
2161 add_timer(&sci->sc_timer);
2162 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2163 }
2164 spin_unlock(&sci->sc_state_lock);
2165 }
2166
2167 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2168 {
2169 spin_lock(&sci->sc_state_lock);
2170 if (!(sci->sc_flush_request & (1 << bn))) {
2171 unsigned long prev_req = sci->sc_flush_request;
2172
2173 sci->sc_flush_request |= (1 << bn);
2174 if (!prev_req)
2175 wake_up(&sci->sc_wait_daemon);
2176 }
2177 spin_unlock(&sci->sc_state_lock);
2178 }
2179
2180 /**
2181 * nilfs_flush_segment - trigger a segment construction for resource control
2182 * @sb: super block
2183 * @ino: inode number of the file to be flushed out.
2184 */
2185 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2186 {
2187 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2188 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2189
2190 if (!sci || nilfs_doing_construction())
2191 return;
2192 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2193 /* assign bit 0 to data files */
2194 }
2195
2196 struct nilfs_segctor_wait_request {
2197 wait_queue_t wq;
2198 __u32 seq;
2199 int err;
2200 atomic_t done;
2201 };
2202
2203 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2204 {
2205 struct nilfs_segctor_wait_request wait_req;
2206 int err = 0;
2207
2208 spin_lock(&sci->sc_state_lock);
2209 init_wait(&wait_req.wq);
2210 wait_req.err = 0;
2211 atomic_set(&wait_req.done, 0);
2212 wait_req.seq = ++sci->sc_seq_request;
2213 spin_unlock(&sci->sc_state_lock);
2214
2215 init_waitqueue_entry(&wait_req.wq, current);
2216 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2217 set_current_state(TASK_INTERRUPTIBLE);
2218 wake_up(&sci->sc_wait_daemon);
2219
2220 for (;;) {
2221 if (atomic_read(&wait_req.done)) {
2222 err = wait_req.err;
2223 break;
2224 }
2225 if (!signal_pending(current)) {
2226 schedule();
2227 continue;
2228 }
2229 err = -ERESTARTSYS;
2230 break;
2231 }
2232 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2233 return err;
2234 }
2235
2236 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2237 {
2238 struct nilfs_segctor_wait_request *wrq, *n;
2239 unsigned long flags;
2240
2241 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2242 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2243 wq.task_list) {
2244 if (!atomic_read(&wrq->done) &&
2245 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2246 wrq->err = err;
2247 atomic_set(&wrq->done, 1);
2248 }
2249 if (atomic_read(&wrq->done)) {
2250 wrq->wq.func(&wrq->wq,
2251 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2252 0, NULL);
2253 }
2254 }
2255 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2256 }
2257
2258 /**
2259 * nilfs_construct_segment - construct a logical segment
2260 * @sb: super block
2261 *
2262 * Return Value: On success, 0 is retured. On errors, one of the following
2263 * negative error code is returned.
2264 *
2265 * %-EROFS - Read only filesystem.
2266 *
2267 * %-EIO - I/O error
2268 *
2269 * %-ENOSPC - No space left on device (only in a panic state).
2270 *
2271 * %-ERESTARTSYS - Interrupted.
2272 *
2273 * %-ENOMEM - Insufficient memory available.
2274 */
2275 int nilfs_construct_segment(struct super_block *sb)
2276 {
2277 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2278 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2279 struct nilfs_transaction_info *ti;
2280 int err;
2281
2282 if (!sci)
2283 return -EROFS;
2284
2285 /* A call inside transactions causes a deadlock. */
2286 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2287
2288 err = nilfs_segctor_sync(sci);
2289 return err;
2290 }
2291
2292 /**
2293 * nilfs_construct_dsync_segment - construct a data-only logical segment
2294 * @sb: super block
2295 * @inode: inode whose data blocks should be written out
2296 * @start: start byte offset
2297 * @end: end byte offset (inclusive)
2298 *
2299 * Return Value: On success, 0 is retured. On errors, one of the following
2300 * negative error code is returned.
2301 *
2302 * %-EROFS - Read only filesystem.
2303 *
2304 * %-EIO - I/O error
2305 *
2306 * %-ENOSPC - No space left on device (only in a panic state).
2307 *
2308 * %-ERESTARTSYS - Interrupted.
2309 *
2310 * %-ENOMEM - Insufficient memory available.
2311 */
2312 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2313 loff_t start, loff_t end)
2314 {
2315 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2316 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2317 struct nilfs_inode_info *ii;
2318 struct nilfs_transaction_info ti;
2319 int err = 0;
2320
2321 if (!sci)
2322 return -EROFS;
2323
2324 nilfs_transaction_lock(sbi, &ti, 0);
2325
2326 ii = NILFS_I(inode);
2327 if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2328 nilfs_test_opt(sbi, STRICT_ORDER) ||
2329 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2330 nilfs_discontinued(sbi->s_nilfs)) {
2331 nilfs_transaction_unlock(sbi);
2332 err = nilfs_segctor_sync(sci);
2333 return err;
2334 }
2335
2336 spin_lock(&sbi->s_inode_lock);
2337 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2338 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2339 spin_unlock(&sbi->s_inode_lock);
2340 nilfs_transaction_unlock(sbi);
2341 return 0;
2342 }
2343 spin_unlock(&sbi->s_inode_lock);
2344 sci->sc_dsync_inode = ii;
2345 sci->sc_dsync_start = start;
2346 sci->sc_dsync_end = end;
2347
2348 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2349
2350 nilfs_transaction_unlock(sbi);
2351 return err;
2352 }
2353
2354 #define FLUSH_FILE_BIT (0x1) /* data file only */
2355 #define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2356
2357 /**
2358 * nilfs_segctor_accept - record accepted sequence count of log-write requests
2359 * @sci: segment constructor object
2360 */
2361 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2362 {
2363 spin_lock(&sci->sc_state_lock);
2364 sci->sc_seq_accepted = sci->sc_seq_request;
2365 spin_unlock(&sci->sc_state_lock);
2366 del_timer_sync(&sci->sc_timer);
2367 }
2368
2369 /**
2370 * nilfs_segctor_notify - notify the result of request to caller threads
2371 * @sci: segment constructor object
2372 * @mode: mode of log forming
2373 * @err: error code to be notified
2374 */
2375 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2376 {
2377 /* Clear requests (even when the construction failed) */
2378 spin_lock(&sci->sc_state_lock);
2379
2380 if (mode == SC_LSEG_SR) {
2381 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2382 sci->sc_seq_done = sci->sc_seq_accepted;
2383 nilfs_segctor_wakeup(sci, err);
2384 sci->sc_flush_request = 0;
2385 } else {
2386 if (mode == SC_FLUSH_FILE)
2387 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2388 else if (mode == SC_FLUSH_DAT)
2389 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2390
2391 /* re-enable timer if checkpoint creation was not done */
2392 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2393 time_before(jiffies, sci->sc_timer.expires))
2394 add_timer(&sci->sc_timer);
2395 }
2396 spin_unlock(&sci->sc_state_lock);
2397 }
2398
2399 /**
2400 * nilfs_segctor_construct - form logs and write them to disk
2401 * @sci: segment constructor object
2402 * @mode: mode of log forming
2403 */
2404 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2405 {
2406 struct nilfs_sb_info *sbi = sci->sc_sbi;
2407 struct the_nilfs *nilfs = sbi->s_nilfs;
2408 struct nilfs_super_block **sbp;
2409 int err = 0;
2410
2411 nilfs_segctor_accept(sci);
2412
2413 if (nilfs_discontinued(nilfs))
2414 mode = SC_LSEG_SR;
2415 if (!nilfs_segctor_confirm(sci))
2416 err = nilfs_segctor_do_construct(sci, mode);
2417
2418 if (likely(!err)) {
2419 if (mode != SC_FLUSH_DAT)
2420 atomic_set(&nilfs->ns_ndirtyblks, 0);
2421 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2422 nilfs_discontinued(nilfs)) {
2423 down_write(&nilfs->ns_sem);
2424 err = -EIO;
2425 sbp = nilfs_prepare_super(sbi,
2426 nilfs_sb_will_flip(nilfs));
2427 if (likely(sbp)) {
2428 nilfs_set_log_cursor(sbp[0], nilfs);
2429 err = nilfs_commit_super(sbi, NILFS_SB_COMMIT);
2430 }
2431 up_write(&nilfs->ns_sem);
2432 }
2433 }
2434
2435 nilfs_segctor_notify(sci, mode, err);
2436 return err;
2437 }
2438
2439 static void nilfs_construction_timeout(unsigned long data)
2440 {
2441 struct task_struct *p = (struct task_struct *)data;
2442 wake_up_process(p);
2443 }
2444
2445 static void
2446 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2447 {
2448 struct nilfs_inode_info *ii, *n;
2449
2450 list_for_each_entry_safe(ii, n, head, i_dirty) {
2451 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2452 continue;
2453 hlist_del_init(&ii->vfs_inode.i_hash);
2454 list_del_init(&ii->i_dirty);
2455 nilfs_clear_gcinode(&ii->vfs_inode);
2456 }
2457 }
2458
2459 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2460 void **kbufs)
2461 {
2462 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2463 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2464 struct the_nilfs *nilfs = sbi->s_nilfs;
2465 struct nilfs_transaction_info ti;
2466 int err;
2467
2468 if (unlikely(!sci))
2469 return -EROFS;
2470
2471 nilfs_transaction_lock(sbi, &ti, 1);
2472
2473 err = nilfs_init_gcdat_inode(nilfs);
2474 if (unlikely(err))
2475 goto out_unlock;
2476
2477 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2478 if (unlikely(err))
2479 goto out_unlock;
2480
2481 sci->sc_freesegs = kbufs[4];
2482 sci->sc_nfreesegs = argv[4].v_nmembs;
2483 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2484
2485 for (;;) {
2486 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2487 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2488
2489 if (likely(!err))
2490 break;
2491
2492 nilfs_warning(sb, __func__,
2493 "segment construction failed. (err=%d)", err);
2494 set_current_state(TASK_INTERRUPTIBLE);
2495 schedule_timeout(sci->sc_interval);
2496 }
2497 if (nilfs_test_opt(sbi, DISCARD)) {
2498 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2499 sci->sc_nfreesegs);
2500 if (ret) {
2501 printk(KERN_WARNING
2502 "NILFS warning: error %d on discard request, "
2503 "turning discards off for the device\n", ret);
2504 nilfs_clear_opt(sbi, DISCARD);
2505 }
2506 }
2507
2508 out_unlock:
2509 sci->sc_freesegs = NULL;
2510 sci->sc_nfreesegs = 0;
2511 nilfs_clear_gcdat_inode(nilfs);
2512 nilfs_transaction_unlock(sbi);
2513 return err;
2514 }
2515
2516 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2517 {
2518 struct nilfs_sb_info *sbi = sci->sc_sbi;
2519 struct nilfs_transaction_info ti;
2520
2521 nilfs_transaction_lock(sbi, &ti, 0);
2522 nilfs_segctor_construct(sci, mode);
2523
2524 /*
2525 * Unclosed segment should be retried. We do this using sc_timer.
2526 * Timeout of sc_timer will invoke complete construction which leads
2527 * to close the current logical segment.
2528 */
2529 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2530 nilfs_segctor_start_timer(sci);
2531
2532 nilfs_transaction_unlock(sbi);
2533 }
2534
2535 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2536 {
2537 int mode = 0;
2538 int err;
2539
2540 spin_lock(&sci->sc_state_lock);
2541 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2542 SC_FLUSH_DAT : SC_FLUSH_FILE;
2543 spin_unlock(&sci->sc_state_lock);
2544
2545 if (mode) {
2546 err = nilfs_segctor_do_construct(sci, mode);
2547
2548 spin_lock(&sci->sc_state_lock);
2549 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2550 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2551 spin_unlock(&sci->sc_state_lock);
2552 }
2553 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2554 }
2555
2556 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2557 {
2558 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2559 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2560 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2561 return SC_FLUSH_FILE;
2562 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2563 return SC_FLUSH_DAT;
2564 }
2565 return SC_LSEG_SR;
2566 }
2567
2568 /**
2569 * nilfs_segctor_thread - main loop of the segment constructor thread.
2570 * @arg: pointer to a struct nilfs_sc_info.
2571 *
2572 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2573 * to execute segment constructions.
2574 */
2575 static int nilfs_segctor_thread(void *arg)
2576 {
2577 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2578 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2579 int timeout = 0;
2580
2581 sci->sc_timer.data = (unsigned long)current;
2582 sci->sc_timer.function = nilfs_construction_timeout;
2583
2584 /* start sync. */
2585 sci->sc_task = current;
2586 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2587 printk(KERN_INFO
2588 "segctord starting. Construction interval = %lu seconds, "
2589 "CP frequency < %lu seconds\n",
2590 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2591
2592 spin_lock(&sci->sc_state_lock);
2593 loop:
2594 for (;;) {
2595 int mode;
2596
2597 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2598 goto end_thread;
2599
2600 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2601 mode = SC_LSEG_SR;
2602 else if (!sci->sc_flush_request)
2603 break;
2604 else
2605 mode = nilfs_segctor_flush_mode(sci);
2606
2607 spin_unlock(&sci->sc_state_lock);
2608 nilfs_segctor_thread_construct(sci, mode);
2609 spin_lock(&sci->sc_state_lock);
2610 timeout = 0;
2611 }
2612
2613
2614 if (freezing(current)) {
2615 spin_unlock(&sci->sc_state_lock);
2616 refrigerator();
2617 spin_lock(&sci->sc_state_lock);
2618 } else {
2619 DEFINE_WAIT(wait);
2620 int should_sleep = 1;
2621
2622 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2623 TASK_INTERRUPTIBLE);
2624
2625 if (sci->sc_seq_request != sci->sc_seq_done)
2626 should_sleep = 0;
2627 else if (sci->sc_flush_request)
2628 should_sleep = 0;
2629 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2630 should_sleep = time_before(jiffies,
2631 sci->sc_timer.expires);
2632
2633 if (should_sleep) {
2634 spin_unlock(&sci->sc_state_lock);
2635 schedule();
2636 spin_lock(&sci->sc_state_lock);
2637 }
2638 finish_wait(&sci->sc_wait_daemon, &wait);
2639 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2640 time_after_eq(jiffies, sci->sc_timer.expires));
2641
2642 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2643 set_nilfs_discontinued(nilfs);
2644 }
2645 goto loop;
2646
2647 end_thread:
2648 spin_unlock(&sci->sc_state_lock);
2649
2650 /* end sync. */
2651 sci->sc_task = NULL;
2652 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2653 return 0;
2654 }
2655
2656 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2657 {
2658 struct task_struct *t;
2659
2660 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2661 if (IS_ERR(t)) {
2662 int err = PTR_ERR(t);
2663
2664 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2665 err);
2666 return err;
2667 }
2668 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2669 return 0;
2670 }
2671
2672 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2673 {
2674 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2675
2676 while (sci->sc_task) {
2677 wake_up(&sci->sc_wait_daemon);
2678 spin_unlock(&sci->sc_state_lock);
2679 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2680 spin_lock(&sci->sc_state_lock);
2681 }
2682 }
2683
2684 /*
2685 * Setup & clean-up functions
2686 */
2687 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
2688 {
2689 struct nilfs_sc_info *sci;
2690
2691 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2692 if (!sci)
2693 return NULL;
2694
2695 sci->sc_sbi = sbi;
2696 sci->sc_super = sbi->s_super;
2697
2698 init_waitqueue_head(&sci->sc_wait_request);
2699 init_waitqueue_head(&sci->sc_wait_daemon);
2700 init_waitqueue_head(&sci->sc_wait_task);
2701 spin_lock_init(&sci->sc_state_lock);
2702 INIT_LIST_HEAD(&sci->sc_dirty_files);
2703 INIT_LIST_HEAD(&sci->sc_segbufs);
2704 INIT_LIST_HEAD(&sci->sc_write_logs);
2705 INIT_LIST_HEAD(&sci->sc_gc_inodes);
2706 INIT_LIST_HEAD(&sci->sc_copied_buffers);
2707 init_timer(&sci->sc_timer);
2708
2709 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2710 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2711 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2712
2713 if (sbi->s_interval)
2714 sci->sc_interval = sbi->s_interval;
2715 if (sbi->s_watermark)
2716 sci->sc_watermark = sbi->s_watermark;
2717 return sci;
2718 }
2719
2720 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2721 {
2722 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2723
2724 /* The segctord thread was stopped and its timer was removed.
2725 But some tasks remain. */
2726 do {
2727 struct nilfs_sb_info *sbi = sci->sc_sbi;
2728 struct nilfs_transaction_info ti;
2729
2730 nilfs_transaction_lock(sbi, &ti, 0);
2731 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2732 nilfs_transaction_unlock(sbi);
2733
2734 } while (ret && retrycount-- > 0);
2735 }
2736
2737 /**
2738 * nilfs_segctor_destroy - destroy the segment constructor.
2739 * @sci: nilfs_sc_info
2740 *
2741 * nilfs_segctor_destroy() kills the segctord thread and frees
2742 * the nilfs_sc_info struct.
2743 * Caller must hold the segment semaphore.
2744 */
2745 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2746 {
2747 struct nilfs_sb_info *sbi = sci->sc_sbi;
2748 int flag;
2749
2750 up_write(&sbi->s_nilfs->ns_segctor_sem);
2751
2752 spin_lock(&sci->sc_state_lock);
2753 nilfs_segctor_kill_thread(sci);
2754 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2755 || sci->sc_seq_request != sci->sc_seq_done);
2756 spin_unlock(&sci->sc_state_lock);
2757
2758 if (flag || !nilfs_segctor_confirm(sci))
2759 nilfs_segctor_write_out(sci);
2760
2761 WARN_ON(!list_empty(&sci->sc_copied_buffers));
2762
2763 if (!list_empty(&sci->sc_dirty_files)) {
2764 nilfs_warning(sbi->s_super, __func__,
2765 "dirty file(s) after the final construction\n");
2766 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
2767 }
2768
2769 WARN_ON(!list_empty(&sci->sc_segbufs));
2770 WARN_ON(!list_empty(&sci->sc_write_logs));
2771
2772 down_write(&sbi->s_nilfs->ns_segctor_sem);
2773
2774 del_timer_sync(&sci->sc_timer);
2775 kfree(sci);
2776 }
2777
2778 /**
2779 * nilfs_attach_segment_constructor - attach a segment constructor
2780 * @sbi: nilfs_sb_info
2781 *
2782 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
2783 * initializes it, and starts the segment constructor.
2784 *
2785 * Return Value: On success, 0 is returned. On error, one of the following
2786 * negative error code is returned.
2787 *
2788 * %-ENOMEM - Insufficient memory available.
2789 */
2790 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi)
2791 {
2792 struct the_nilfs *nilfs = sbi->s_nilfs;
2793 int err;
2794
2795 if (NILFS_SC(sbi)) {
2796 /*
2797 * This happens if the filesystem was remounted
2798 * read/write after nilfs_error degenerated it into a
2799 * read-only mount.
2800 */
2801 nilfs_detach_segment_constructor(sbi);
2802 }
2803
2804 sbi->s_sc_info = nilfs_segctor_new(sbi);
2805 if (!sbi->s_sc_info)
2806 return -ENOMEM;
2807
2808 nilfs_attach_writer(nilfs, sbi);
2809 err = nilfs_segctor_start_thread(NILFS_SC(sbi));
2810 if (err) {
2811 nilfs_detach_writer(nilfs, sbi);
2812 kfree(sbi->s_sc_info);
2813 sbi->s_sc_info = NULL;
2814 }
2815 return err;
2816 }
2817
2818 /**
2819 * nilfs_detach_segment_constructor - destroy the segment constructor
2820 * @sbi: nilfs_sb_info
2821 *
2822 * nilfs_detach_segment_constructor() kills the segment constructor daemon,
2823 * frees the struct nilfs_sc_info, and destroy the dirty file list.
2824 */
2825 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2826 {
2827 struct the_nilfs *nilfs = sbi->s_nilfs;
2828 LIST_HEAD(garbage_list);
2829
2830 down_write(&nilfs->ns_segctor_sem);
2831 if (NILFS_SC(sbi)) {
2832 nilfs_segctor_destroy(NILFS_SC(sbi));
2833 sbi->s_sc_info = NULL;
2834 }
2835
2836 /* Force to free the list of dirty files */
2837 spin_lock(&sbi->s_inode_lock);
2838 if (!list_empty(&sbi->s_dirty_files)) {
2839 list_splice_init(&sbi->s_dirty_files, &garbage_list);
2840 nilfs_warning(sbi->s_super, __func__,
2841 "Non empty dirty list after the last "
2842 "segment construction\n");
2843 }
2844 spin_unlock(&sbi->s_inode_lock);
2845 up_write(&nilfs->ns_segctor_sem);
2846
2847 nilfs_dispose_list(sbi, &garbage_list, 1);
2848 nilfs_detach_writer(nilfs, sbi);
2849 }
Tomato firmware and modifications.