projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
nilfs2: add iterator for segment buffers
[linux-2.6-block.git] / fs / nilfs2 / segment.c
CommitLineData
9ff05123
RK		 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 "nilfs.h"
36#include "btnode.h"
37#include "page.h"
38#include "segment.h"
39#include "sufile.h"
40#include "cpfile.h"
41#include "ifile.h"
9ff05123
RK		 42#include "segbuf.h"
43
44
45/*
46 * Segment constructor
47 */
48#define SC_N_INODEVEC 16 /* Size of locally allocated inode vector */
49
50#define SC_MAX_SEGDELTA 64 /* Upper limit of the number of segments
51 appended in collection retry loop */
52
53/* Construction mode */
54enum {
55 SC_LSEG_SR = 1, /* Make a logical segment having a super root */
56 SC_LSEG_DSYNC, /* Flush data blocks of a given file and make
57 a logical segment without a super root */
58 SC_FLUSH_FILE, /* Flush data files, leads to segment writes without
59 creating a checkpoint */
60 SC_FLUSH_DAT, /* Flush DAT file. This also creates segments without
61 a checkpoint */
62};
63
64/* Stage numbers of dirty block collection */
65enum {
66 NILFS_ST_INIT = 0,
67 NILFS_ST_GC, /* Collecting dirty blocks for GC */
68 NILFS_ST_FILE,
9ff05123
RK		 69 NILFS_ST_IFILE,
70 NILFS_ST_CPFILE,
71 NILFS_ST_SUFILE,
72 NILFS_ST_DAT,
73 NILFS_ST_SR, /* Super root */
74 NILFS_ST_DSYNC, /* Data sync blocks */
75 NILFS_ST_DONE,
76};
77
78/* State flags of collection */
79#define NILFS_CF_NODE 0x0001 /* Collecting node blocks */
80#define NILFS_CF_IFILE_STARTED 0x0002 /* IFILE stage has started */
071cb4b8
RK		 81#define NILFS_CF_SUFREED 0x0004 /* segment usages has been freed */
82#define NILFS_CF_HISTORY_MASK (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
9ff05123
RK		 83
84/* Operations depending on the construction mode and file type */
85struct nilfs_sc_operations {
86 int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
87 struct inode *);
88 int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
89 struct inode *);
90 int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
91 struct inode *);
92 void (*write_data_binfo)(struct nilfs_sc_info *,
93 struct nilfs_segsum_pointer *,
94 union nilfs_binfo *);
95 void (*write_node_binfo)(struct nilfs_sc_info *,
96 struct nilfs_segsum_pointer *,
97 union nilfs_binfo *);
98};
99
100/*
101 * Other definitions
102 */
103static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
104static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
105static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
106static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
107 int);
108
109#define nilfs_cnt32_gt(a, b) \
110 (typecheck(__u32, a) && typecheck(__u32, b) && \
111 ((__s32)(b) - (__s32)(a) < 0))
112#define nilfs_cnt32_ge(a, b) \
113 (typecheck(__u32, a) && typecheck(__u32, b) && \
114 ((__s32)(a) - (__s32)(b) >= 0))
115#define nilfs_cnt32_lt(a, b) nilfs_cnt32_gt(b, a)
116#define nilfs_cnt32_le(a, b) nilfs_cnt32_ge(b, a)
117
118/*
119 * Transaction
120 */
121static struct kmem_cache *nilfs_transaction_cachep;
122
123/**
124 * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
125 *
126 * nilfs_init_transaction_cache() creates a slab cache for the struct
127 * nilfs_transaction_info.
128 *
129 * Return Value: On success, it returns 0. On error, one of the following
130 * negative error code is returned.
131 *
132 * %-ENOMEM - Insufficient memory available.
133 */
134int nilfs_init_transaction_cache(void)
135{
136 nilfs_transaction_cachep =
137 kmem_cache_create("nilfs2_transaction_cache",
138 sizeof(struct nilfs_transaction_info),
139 0, SLAB_RECLAIM_ACCOUNT, NULL);
140 return (nilfs_transaction_cachep == NULL) ? -ENOMEM : 0;
141}
142
143/**
144 * nilfs_detroy_transaction_cache - destroy the cache for transaction info
145 *
146 * nilfs_destroy_transaction_cache() frees the slab cache for the struct
147 * nilfs_transaction_info.
148 */
149void nilfs_destroy_transaction_cache(void)
150{
151 kmem_cache_destroy(nilfs_transaction_cachep);
152}
153
154static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
155{
156 struct nilfs_transaction_info *cur_ti = current->journal_info;
157 void *save = NULL;
158
159 if (cur_ti) {
160 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
161 return ++cur_ti->ti_count;
162 else {
163 /*
164 * If journal_info field is occupied by other FS,
47420c79
RK		 165 * it is saved and will be restored on
166 * nilfs_transaction_commit().
9ff05123
RK		 167 */
168 printk(KERN_WARNING
169 "NILFS warning: journal info from a different "
170 "FS\n");
171 save = current->journal_info;
172 }
173 }
174 if (!ti) {
175 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
176 if (!ti)
177 return -ENOMEM;
178 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
179 } else {
180 ti->ti_flags = 0;
181 }
182 ti->ti_count = 0;
183 ti->ti_save = save;
184 ti->ti_magic = NILFS_TI_MAGIC;
185 current->journal_info = ti;
186 return 0;
187}
188
189/**
190 * nilfs_transaction_begin - start indivisible file operations.
191 * @sb: super block
192 * @ti: nilfs_transaction_info
193 * @vacancy_check: flags for vacancy rate checks
194 *
195 * nilfs_transaction_begin() acquires a reader/writer semaphore, called
196 * the segment semaphore, to make a segment construction and write tasks
47420c79 197 * exclusive. The function is used with nilfs_transaction_commit() in pairs.
9ff05123
RK		 198 * The region enclosed by these two functions can be nested. To avoid a
199 * deadlock, the semaphore is only acquired or released in the outermost call.
200 *
201 * This function allocates a nilfs_transaction_info struct to keep context
202 * information on it. It is initialized and hooked onto the current task in
203 * the outermost call. If a pre-allocated struct is given to @ti, it is used
204 * instead; othewise a new struct is assigned from a slab.
205 *
206 * When @vacancy_check flag is set, this function will check the amount of
207 * free space, and will wait for the GC to reclaim disk space if low capacity.
208 *
209 * Return Value: On success, 0 is returned. On error, one of the following
210 * negative error code is returned.
211 *
212 * %-ENOMEM - Insufficient memory available.
213 *
9ff05123
RK		 214 * %-ENOSPC - No space left on device
215 */
216int nilfs_transaction_begin(struct super_block *sb,
217 struct nilfs_transaction_info *ti,
218 int vacancy_check)
219{
220 struct nilfs_sb_info *sbi;
221 struct the_nilfs *nilfs;
222 int ret = nilfs_prepare_segment_lock(ti);
223
224 if (unlikely(ret < 0))
225 return ret;
226 if (ret > 0)
227 return 0;
228
229 sbi = NILFS_SB(sb);
230 nilfs = sbi->s_nilfs;
231 down_read(&nilfs->ns_segctor_sem);
232 if (vacancy_check && nilfs_near_disk_full(nilfs)) {
233 up_read(&nilfs->ns_segctor_sem);
234 ret = -ENOSPC;
235 goto failed;
236 }
237 return 0;
238
239 failed:
240 ti = current->journal_info;
241 current->journal_info = ti->ti_save;
242 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
243 kmem_cache_free(nilfs_transaction_cachep, ti);
244 return ret;
245}
246
247/**
47420c79 248 * nilfs_transaction_commit - commit indivisible file operations.
9ff05123 249 * @sb: super block
9ff05123 250 *
47420c79
RK		 251 * nilfs_transaction_commit() releases the read semaphore which is
252 * acquired by nilfs_transaction_begin(). This is only performed
253 * in outermost call of this function. If a commit flag is set,
254 * nilfs_transaction_commit() sets a timer to start the segment
255 * constructor. If a sync flag is set, it starts construction
256 * directly.
9ff05123 257 */
47420c79 258int nilfs_transaction_commit(struct super_block *sb)
9ff05123
RK		 259{
260 struct nilfs_transaction_info *ti = current->journal_info;
261 struct nilfs_sb_info *sbi;
262 struct nilfs_sc_info *sci;
263 int err = 0;
264
265 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
47420c79 266 ti->ti_flags |= NILFS_TI_COMMIT;
9ff05123
RK		 267 if (ti->ti_count > 0) {
268 ti->ti_count--;
269 return 0;
270 }
271 sbi = NILFS_SB(sb);
272 sci = NILFS_SC(sbi);
273 if (sci != NULL) {
274 if (ti->ti_flags & NILFS_TI_COMMIT)
275 nilfs_segctor_start_timer(sci);
276 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
277 sci->sc_watermark)
278 nilfs_segctor_do_flush(sci, 0);
279 }
280 up_read(&sbi->s_nilfs->ns_segctor_sem);
281 current->journal_info = ti->ti_save;
282
283 if (ti->ti_flags & NILFS_TI_SYNC)
284 err = nilfs_construct_segment(sb);
285 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
286 kmem_cache_free(nilfs_transaction_cachep, ti);
287 return err;
288}
289
47420c79
RK		 290void nilfs_transaction_abort(struct super_block *sb)
291{
292 struct nilfs_transaction_info *ti = current->journal_info;
293
294 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
295 if (ti->ti_count > 0) {
296 ti->ti_count--;
297 return;
298 }
299 up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
300
301 current->journal_info = ti->ti_save;
302 if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
303 kmem_cache_free(nilfs_transaction_cachep, ti);
304}
305
9ff05123
RK		 306void nilfs_relax_pressure_in_lock(struct super_block *sb)
307{
308 struct nilfs_sb_info *sbi = NILFS_SB(sb);
309 struct nilfs_sc_info *sci = NILFS_SC(sbi);
310 struct the_nilfs *nilfs = sbi->s_nilfs;
311
312 if (!sci || !sci->sc_flush_request)
313 return;
314
315 set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
316 up_read(&nilfs->ns_segctor_sem);
317
318 down_write(&nilfs->ns_segctor_sem);
319 if (sci->sc_flush_request &&
320 test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
321 struct nilfs_transaction_info *ti = current->journal_info;
322
323 ti->ti_flags |= NILFS_TI_WRITER;
324 nilfs_segctor_do_immediate_flush(sci);
325 ti->ti_flags &= ~NILFS_TI_WRITER;
326 }
327 downgrade_write(&nilfs->ns_segctor_sem);
328}
329
330static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
331 struct nilfs_transaction_info *ti,
332 int gcflag)
333{
334 struct nilfs_transaction_info *cur_ti = current->journal_info;
335
1f5abe7e 336 WARN_ON(cur_ti);
9ff05123
RK		 337 ti->ti_flags = NILFS_TI_WRITER;
338 ti->ti_count = 0;
339 ti->ti_save = cur_ti;
340 ti->ti_magic = NILFS_TI_MAGIC;
341 INIT_LIST_HEAD(&ti->ti_garbage);
342 current->journal_info = ti;
343
344 for (;;) {
345 down_write(&sbi->s_nilfs->ns_segctor_sem);
346 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
347 break;
348
349 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
350
351 up_write(&sbi->s_nilfs->ns_segctor_sem);
352 yield();
353 }
354 if (gcflag)
355 ti->ti_flags |= NILFS_TI_GC;
356}
357
358static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
359{
360 struct nilfs_transaction_info *ti = current->journal_info;
361
362 BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
363 BUG_ON(ti->ti_count > 0);
364
365 up_write(&sbi->s_nilfs->ns_segctor_sem);
366 current->journal_info = ti->ti_save;
367 if (!list_empty(&ti->ti_garbage))
368 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
369}
370
371static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
372 struct nilfs_segsum_pointer *ssp,
373 unsigned bytes)
374{
375 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
376 unsigned blocksize = sci->sc_super->s_blocksize;
377 void *p;
378
379 if (unlikely(ssp->offset + bytes > blocksize)) {
380 ssp->offset = 0;
381 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
382 &segbuf->sb_segsum_buffers));
383 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
384 }
385 p = ssp->bh->b_data + ssp->offset;
386 ssp->offset += bytes;
387 return p;
388}
389
390/**
391 * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
392 * @sci: nilfs_sc_info
393 */
394static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
395{
396 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
397 struct buffer_head *sumbh;
398 unsigned sumbytes;
399 unsigned flags = 0;
400 int err;
401
402 if (nilfs_doing_gc())
403 flags = NILFS_SS_GC;
404 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
405 if (unlikely(err))
406 return err;
407
408 sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
409 sumbytes = segbuf->sb_sum.sumbytes;
410 sci->sc_finfo_ptr.bh = sumbh; sci->sc_finfo_ptr.offset = sumbytes;
411 sci->sc_binfo_ptr.bh = sumbh; sci->sc_binfo_ptr.offset = sumbytes;
412 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
413 return 0;
414}
415
416static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
417{
418 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
419 if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
420 return -E2BIG; /* The current segment is filled up
421 (internal code) */
422 sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
423 return nilfs_segctor_reset_segment_buffer(sci);
424}
425
426static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
427{
428 struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
429 int err;
430
431 if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
432 err = nilfs_segctor_feed_segment(sci);
433 if (err)
434 return err;
435 segbuf = sci->sc_curseg;
436 }
437 err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
438 if (likely(!err))
439 segbuf->sb_sum.flags |= NILFS_SS_SR;
440 return err;
441}
442
443/*
444 * Functions for making segment summary and payloads
445 */
446static int nilfs_segctor_segsum_block_required(
447 struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
448 unsigned binfo_size)
449{
450 unsigned blocksize = sci->sc_super->s_blocksize;
451 /* Size of finfo and binfo is enough small against blocksize */
452
453 return ssp->offset + binfo_size +
454 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
455 blocksize;
456}
457
458static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
459 struct inode *inode)
460{
461 sci->sc_curseg->sb_sum.nfinfo++;
462 sci->sc_binfo_ptr = sci->sc_finfo_ptr;
463 nilfs_segctor_map_segsum_entry(
464 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
c96fa464
RK		 465
466 if (inode->i_sb && !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
467 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
9ff05123
RK		 468 /* skip finfo */
469}
470
471static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
472 struct inode *inode)
473{
474 struct nilfs_finfo *finfo;
475 struct nilfs_inode_info *ii;
476 struct nilfs_segment_buffer *segbuf;
477
478 if (sci->sc_blk_cnt == 0)
479 return;
480
481 ii = NILFS_I(inode);
482 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
483 sizeof(*finfo));
484 finfo->fi_ino = cpu_to_le64(inode->i_ino);
485 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
486 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
487 finfo->fi_cno = cpu_to_le64(ii->i_cno);
488
489 segbuf = sci->sc_curseg;
490 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
491 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
492 sci->sc_finfo_ptr = sci->sc_binfo_ptr;
493 sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
494}
495
496static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
497 struct buffer_head *bh,
498 struct inode *inode,
499 unsigned binfo_size)
500{
501 struct nilfs_segment_buffer *segbuf;
502 int required, err = 0;
503
504 retry:
505 segbuf = sci->sc_curseg;
506 required = nilfs_segctor_segsum_block_required(
507 sci, &sci->sc_binfo_ptr, binfo_size);
508 if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
509 nilfs_segctor_end_finfo(sci, inode);
510 err = nilfs_segctor_feed_segment(sci);
511 if (err)
512 return err;
513 goto retry;
514 }
515 if (unlikely(required)) {
516 err = nilfs_segbuf_extend_segsum(segbuf);
517 if (unlikely(err))
518 goto failed;
519 }
520 if (sci->sc_blk_cnt == 0)
521 nilfs_segctor_begin_finfo(sci, inode);
522
523 nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
524 /* Substitution to vblocknr is delayed until update_blocknr() */
525 nilfs_segbuf_add_file_buffer(segbuf, bh);
526 sci->sc_blk_cnt++;
527 failed:
528 return err;
529}
530
531static int nilfs_handle_bmap_error(int err, const char *fname,
532 struct inode *inode, struct super_block *sb)
533{
534 if (err == -EINVAL) {
535 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
536 inode->i_ino);
537 err = -EIO;
538 }
539 return err;
540}
541
542/*
543 * Callback functions that enumerate, mark, and collect dirty blocks
544 */
545static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
546 struct buffer_head *bh, struct inode *inode)
547{
548 int err;
549
9ff05123
RK		 550 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
551 if (unlikely(err < 0))
552 return nilfs_handle_bmap_error(err, __func__, inode,
553 sci->sc_super);
554
555 err = nilfs_segctor_add_file_block(sci, bh, inode,
556 sizeof(struct nilfs_binfo_v));
557 if (!err)
558 sci->sc_datablk_cnt++;
559 return err;
560}
561
562static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
563 struct buffer_head *bh,
564 struct inode *inode)
565{
566 int err;
567
9ff05123
RK		 568 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
569 if (unlikely(err < 0))
570 return nilfs_handle_bmap_error(err, __func__, inode,
571 sci->sc_super);
572 return 0;
573}
574
575static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
576 struct buffer_head *bh,
577 struct inode *inode)
578{
1f5abe7e 579 WARN_ON(!buffer_dirty(bh));
9ff05123
RK		 580 return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
581}
582
583static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
584 struct nilfs_segsum_pointer *ssp,
585 union nilfs_binfo *binfo)
586{
587 struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
588 sci, ssp, sizeof(*binfo_v));
589 *binfo_v = binfo->bi_v;
590}
591
592static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
593 struct nilfs_segsum_pointer *ssp,
594 union nilfs_binfo *binfo)
595{
596 __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
597 sci, ssp, sizeof(*vblocknr));
598 *vblocknr = binfo->bi_v.bi_vblocknr;
599}
600
601struct nilfs_sc_operations nilfs_sc_file_ops = {
602 .collect_data = nilfs_collect_file_data,
603 .collect_node = nilfs_collect_file_node,
604 .collect_bmap = nilfs_collect_file_bmap,
605 .write_data_binfo = nilfs_write_file_data_binfo,
606 .write_node_binfo = nilfs_write_file_node_binfo,
607};
608
609static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
610 struct buffer_head *bh, struct inode *inode)
611{
612 int err;
613
614 err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
615 if (unlikely(err < 0))
616 return nilfs_handle_bmap_error(err, __func__, inode,
617 sci->sc_super);
618
619 err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
620 if (!err)
621 sci->sc_datablk_cnt++;
622 return err;
623}
624
625static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
626 struct buffer_head *bh, struct inode *inode)
627{
1f5abe7e 628 WARN_ON(!buffer_dirty(bh));
9ff05123
RK		 629 return nilfs_segctor_add_file_block(sci, bh, inode,
630 sizeof(struct nilfs_binfo_dat));
631}
632
633static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
634 struct nilfs_segsum_pointer *ssp,
635 union nilfs_binfo *binfo)
636{
637 __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
638 sizeof(*blkoff));
639 *blkoff = binfo->bi_dat.bi_blkoff;
640}
641
642static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
643 struct nilfs_segsum_pointer *ssp,
644 union nilfs_binfo *binfo)
645{
646 struct nilfs_binfo_dat *binfo_dat =
647 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
648 *binfo_dat = binfo->bi_dat;
649}
650
651struct nilfs_sc_operations nilfs_sc_dat_ops = {
652 .collect_data = nilfs_collect_dat_data,
653 .collect_node = nilfs_collect_file_node,
654 .collect_bmap = nilfs_collect_dat_bmap,
655 .write_data_binfo = nilfs_write_dat_data_binfo,
656 .write_node_binfo = nilfs_write_dat_node_binfo,
657};
658
659struct nilfs_sc_operations nilfs_sc_dsync_ops = {
660 .collect_data = nilfs_collect_file_data,
661 .collect_node = NULL,
662 .collect_bmap = NULL,
663 .write_data_binfo = nilfs_write_file_data_binfo,
664 .write_node_binfo = NULL,
665};
666
f30bf3e4
RK		 667static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
668 struct list_head *listp,
669 size_t nlimit,
670 loff_t start, loff_t end)
9ff05123 671{
9ff05123
RK		 672 struct address_space *mapping = inode->i_mapping;
673 struct pagevec pvec;
f30bf3e4
RK		 674 pgoff_t index = 0, last = ULONG_MAX;
675 size_t ndirties = 0;
676 int i;
9ff05123 677
f30bf3e4
RK		 678 if (unlikely(start != 0 || end != LLONG_MAX)) {
679 /*
680 * A valid range is given for sync-ing data pages. The
681 * range is rounded to per-page; extra dirty buffers
682 * may be included if blocksize < pagesize.
683 */
684 index = start >> PAGE_SHIFT;
685 last = end >> PAGE_SHIFT;
686 }
9ff05123
RK		 687 pagevec_init(&pvec, 0);
688 repeat:
f30bf3e4
RK		 689 if (unlikely(index > last) ||
690 !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
691 min_t(pgoff_t, last - index,
692 PAGEVEC_SIZE - 1) + 1))
693 return ndirties;
9ff05123
RK		 694
695 for (i = 0; i < pagevec_count(&pvec); i++) {
696 struct buffer_head *bh, *head;
697 struct page *page = pvec.pages[i];
698
f30bf3e4
RK		 699 if (unlikely(page->index > last))
700 break;
701
9ff05123
RK		 702 if (mapping->host) {
703 lock_page(page);
704 if (!page_has_buffers(page))
705 create_empty_buffers(page,
706 1 << inode->i_blkbits, 0);
707 unlock_page(page);
708 }
709
710 bh = head = page_buffers(page);
711 do {
f30bf3e4
RK		 712 if (!buffer_dirty(bh))
713 continue;
714 get_bh(bh);
715 list_add_tail(&bh->b_assoc_buffers, listp);
716 ndirties++;
717 if (unlikely(ndirties >= nlimit)) {
718 pagevec_release(&pvec);
719 cond_resched();
720 return ndirties;
9ff05123 721 }
f30bf3e4 722 } while (bh = bh->b_this_page, bh != head);
9ff05123
RK		 723 }
724 pagevec_release(&pvec);
725 cond_resched();
f30bf3e4 726 goto repeat;
9ff05123
RK		 727}
728
729static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
730 struct list_head *listp)
731{
732 struct nilfs_inode_info *ii = NILFS_I(inode);
733 struct address_space *mapping = &ii->i_btnode_cache;
734 struct pagevec pvec;
735 struct buffer_head *bh, *head;
736 unsigned int i;
737 pgoff_t index = 0;
738
739 pagevec_init(&pvec, 0);
740
741 while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
742 PAGEVEC_SIZE)) {
743 for (i = 0; i < pagevec_count(&pvec); i++) {
744 bh = head = page_buffers(pvec.pages[i]);
745 do {
746 if (buffer_dirty(bh)) {
747 get_bh(bh);
748 list_add_tail(&bh->b_assoc_buffers,
749 listp);
750 }
751 bh = bh->b_this_page;
752 } while (bh != head);
753 }
754 pagevec_release(&pvec);
755 cond_resched();
756 }
757}
758
759static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
760 struct list_head *head, int force)
761{
762 struct nilfs_inode_info *ii, *n;
763 struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
764 unsigned nv = 0;
765
766 while (!list_empty(head)) {
767 spin_lock(&sbi->s_inode_lock);
768 list_for_each_entry_safe(ii, n, head, i_dirty) {
769 list_del_init(&ii->i_dirty);
770 if (force) {
771 if (unlikely(ii->i_bh)) {
772 brelse(ii->i_bh);
773 ii->i_bh = NULL;
774 }
775 } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
776 set_bit(NILFS_I_QUEUED, &ii->i_state);
777 list_add_tail(&ii->i_dirty,
778 &sbi->s_dirty_files);
779 continue;
780 }
781 ivec[nv++] = ii;
782 if (nv == SC_N_INODEVEC)
783 break;
784 }
785 spin_unlock(&sbi->s_inode_lock);
786
787 for (pii = ivec; nv > 0; pii++, nv--)
788 iput(&(*pii)->vfs_inode);
789 }
790}
791
792static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
793{
794 struct the_nilfs *nilfs = sbi->s_nilfs;
795 int ret = 0;
796
797 if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
798 ret++;
799 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
800 ret++;
801 if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
802 ret++;
803 if (ret || nilfs_doing_gc())
804 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
805 ret++;
806 return ret;
807}
808
809static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
810{
811 return list_empty(&sci->sc_dirty_files) &&
812 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
071cb4b8 813 sci->sc_nfreesegs == 0 &&
9ff05123
RK		 814 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
815}
816
817static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
818{
819 struct nilfs_sb_info *sbi = sci->sc_sbi;
820 int ret = 0;
821
822 if (nilfs_test_metadata_dirty(sbi))
823 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
824
825 spin_lock(&sbi->s_inode_lock);
826 if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
827 ret++;
828
829 spin_unlock(&sbi->s_inode_lock);
830 return ret;
831}
832
833static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
834{
835 struct nilfs_sb_info *sbi = sci->sc_sbi;
836 struct the_nilfs *nilfs = sbi->s_nilfs;
837
838 nilfs_mdt_clear_dirty(sbi->s_ifile);
839 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
840 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
841 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
842}
843
844static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
845{
846 struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
847 struct buffer_head *bh_cp;
848 struct nilfs_checkpoint *raw_cp;
849 int err;
850
851 /* XXX: this interface will be changed */
852 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
853 &raw_cp, &bh_cp);
854 if (likely(!err)) {
855 /* The following code is duplicated with cpfile. But, it is
856 needed to collect the checkpoint even if it was not newly
857 created */
858 nilfs_mdt_mark_buffer_dirty(bh_cp);
859 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
860 nilfs_cpfile_put_checkpoint(
861 nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
1f5abe7e
RK		 862 } else
863 WARN_ON(err == -EINVAL || err == -ENOENT);
864
9ff05123
RK		 865 return err;
866}
867
868static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
869{
870 struct nilfs_sb_info *sbi = sci->sc_sbi;
871 struct the_nilfs *nilfs = sbi->s_nilfs;
872 struct buffer_head *bh_cp;
873 struct nilfs_checkpoint *raw_cp;
874 int err;
875
876 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
877 &raw_cp, &bh_cp);
878 if (unlikely(err)) {
1f5abe7e 879 WARN_ON(err == -EINVAL || err == -ENOENT);
9ff05123
RK		 880 goto failed_ibh;
881 }
882 raw_cp->cp_snapshot_list.ssl_next = 0;
883 raw_cp->cp_snapshot_list.ssl_prev = 0;
884 raw_cp->cp_inodes_count =
885 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
886 raw_cp->cp_blocks_count =
887 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
888 raw_cp->cp_nblk_inc =
889 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
890 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
891 raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
458c5b08 892
c96fa464
RK		 893 if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
894 nilfs_checkpoint_clear_minor(raw_cp);
895 else
896 nilfs_checkpoint_set_minor(raw_cp);
897
9ff05123
RK		 898 nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
899 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
900 return 0;
901
902 failed_ibh:
903 return err;
904}
905
906static void nilfs_fill_in_file_bmap(struct inode *ifile,
907 struct nilfs_inode_info *ii)
908
909{
910 struct buffer_head *ibh;
911 struct nilfs_inode *raw_inode;
912
913 if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
914 ibh = ii->i_bh;
915 BUG_ON(!ibh);
916 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
917 ibh);
918 nilfs_bmap_write(ii->i_bmap, raw_inode);
919 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
920 }
921}
922
923static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
924 struct inode *ifile)
925{
926 struct nilfs_inode_info *ii;
927
928 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
929 nilfs_fill_in_file_bmap(ifile, ii);
930 set_bit(NILFS_I_COLLECTED, &ii->i_state);
931 }
9ff05123
RK		 932}
933
934/*
935 * CRC calculation routines
936 */
937static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
938{
939 struct nilfs_super_root *raw_sr =
940 (struct nilfs_super_root *)bh_sr->b_data;
941 u32 crc;
942
9ff05123
RK		 943 crc = crc32_le(seed,
944 (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
945 NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
946 raw_sr->sr_sum = cpu_to_le32(crc);
947}
948
949static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
950 u32 seed)
951{
952 struct nilfs_segment_buffer *segbuf;
953
954 if (sci->sc_super_root)
955 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
956
957 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
958 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
959 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
960 }
961}
962
963static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
964 struct the_nilfs *nilfs)
965{
966 struct buffer_head *bh_sr = sci->sc_super_root;
967 struct nilfs_super_root *raw_sr =
968 (struct nilfs_super_root *)bh_sr->b_data;
969 unsigned isz = nilfs->ns_inode_size;
970
971 raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
972 raw_sr->sr_nongc_ctime
973 = cpu_to_le64(nilfs_doing_gc() ?
974 nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
975 raw_sr->sr_flags = 0;
976
3961f0e2
RK		 977 nilfs_write_inode_common(nilfs_dat_inode(nilfs), (void *)raw_sr +
978 NILFS_SR_DAT_OFFSET(isz), 1);
979 nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
980 NILFS_SR_CPFILE_OFFSET(isz), 1);
981 nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
982 NILFS_SR_SUFILE_OFFSET(isz), 1);
9ff05123
RK		 983}
984
985static void nilfs_redirty_inodes(struct list_head *head)
986{
987 struct nilfs_inode_info *ii;
988
989 list_for_each_entry(ii, head, i_dirty) {
990 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
991 clear_bit(NILFS_I_COLLECTED, &ii->i_state);
992 }
993}
994
995static void nilfs_drop_collected_inodes(struct list_head *head)
996{
997 struct nilfs_inode_info *ii;
998
999 list_for_each_entry(ii, head, i_dirty) {
1000 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1001 continue;
1002
1003 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
1004 set_bit(NILFS_I_UPDATED, &ii->i_state);
1005 }
1006}
1007
9ff05123
RK		 1008static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1009 struct inode *inode,
1010 struct list_head *listp,
1011 int (*collect)(struct nilfs_sc_info *,
1012 struct buffer_head *,
1013 struct inode *))
1014{
1015 struct buffer_head *bh, *n;
1016 int err = 0;
1017
1018 if (collect) {
1019 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1020 list_del_init(&bh->b_assoc_buffers);
1021 err = collect(sci, bh, inode);
1022 brelse(bh);
1023 if (unlikely(err))
1024 goto dispose_buffers;
1025 }
1026 return 0;
1027 }
1028
1029 dispose_buffers:
1030 while (!list_empty(listp)) {
1031 bh = list_entry(listp->next, struct buffer_head,
1032 b_assoc_buffers);
1033 list_del_init(&bh->b_assoc_buffers);
1034 brelse(bh);
1035 }
1036 return err;
1037}
1038
f30bf3e4
RK		 1039static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1040{
1041 /* Remaining number of blocks within segment buffer */
1042 return sci->sc_segbuf_nblocks -
1043 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1044}
1045
9ff05123
RK		 1046static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1047 struct inode *inode,
1048 struct nilfs_sc_operations *sc_ops)
1049{
1050 LIST_HEAD(data_buffers);
1051 LIST_HEAD(node_buffers);
f30bf3e4 1052 int err;
9ff05123
RK		 1053
1054 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
f30bf3e4
RK		 1055 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1056
1057 n = nilfs_lookup_dirty_data_buffers(
1058 inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1059 if (n > rest) {
1060 err = nilfs_segctor_apply_buffers(
9ff05123 1061 sci, inode, &data_buffers,
f30bf3e4
RK		 1062 sc_ops->collect_data);
1063 BUG_ON(!err); /* always receive -E2BIG or true error */
9ff05123
RK		 1064 goto break_or_fail;
1065 }
1066 }
1067 nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1068
1069 if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1070 err = nilfs_segctor_apply_buffers(
1071 sci, inode, &data_buffers, sc_ops->collect_data);
1072 if (unlikely(err)) {
1073 /* dispose node list */
1074 nilfs_segctor_apply_buffers(
1075 sci, inode, &node_buffers, NULL);
1076 goto break_or_fail;
1077 }
1078 sci->sc_stage.flags |= NILFS_CF_NODE;
1079 }
1080 /* Collect node */
1081 err = nilfs_segctor_apply_buffers(
1082 sci, inode, &node_buffers, sc_ops->collect_node);
1083 if (unlikely(err))
1084 goto break_or_fail;
1085
1086 nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1087 err = nilfs_segctor_apply_buffers(
1088 sci, inode, &node_buffers, sc_ops->collect_bmap);
1089 if (unlikely(err))
1090 goto break_or_fail;
1091
1092 nilfs_segctor_end_finfo(sci, inode);
1093 sci->sc_stage.flags &= ~NILFS_CF_NODE;
1094
1095 break_or_fail:
1096 return err;
1097}
1098
1099static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1100 struct inode *inode)
1101{
1102 LIST_HEAD(data_buffers);
f30bf3e4
RK		 1103 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1104 int err;
9ff05123 1105
f30bf3e4
RK		 1106 n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1107 sci->sc_dsync_start,
1108 sci->sc_dsync_end);
1109
1110 err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1111 nilfs_collect_file_data);
1112 if (!err) {
9ff05123 1113 nilfs_segctor_end_finfo(sci, inode);
f30bf3e4
RK		 1114 BUG_ON(n > rest);
1115 /* always receive -E2BIG or true error if n > rest */
1116 }
9ff05123
RK		 1117 return err;
1118}
1119
1120static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1121{
1122 struct nilfs_sb_info *sbi = sci->sc_sbi;
1123 struct the_nilfs *nilfs = sbi->s_nilfs;
1124 struct list_head *head;
1125 struct nilfs_inode_info *ii;
071cb4b8 1126 size_t ndone;
9ff05123
RK		 1127 int err = 0;
1128
1129 switch (sci->sc_stage.scnt) {
1130 case NILFS_ST_INIT:
1131 /* Pre-processes */
1132 sci->sc_stage.flags = 0;
1133
1134 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1135 sci->sc_nblk_inc = 0;
1136 sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1137 if (mode == SC_LSEG_DSYNC) {
1138 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1139 goto dsync_mode;
1140 }
1141 }
1142
1143 sci->sc_stage.dirty_file_ptr = NULL;
1144 sci->sc_stage.gc_inode_ptr = NULL;
1145 if (mode == SC_FLUSH_DAT) {
1146 sci->sc_stage.scnt = NILFS_ST_DAT;
1147 goto dat_stage;
1148 }
1149 sci->sc_stage.scnt++; /* Fall through */
1150 case NILFS_ST_GC:
1151 if (nilfs_doing_gc()) {
1152 head = &sci->sc_gc_inodes;
1153 ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1154 head, i_dirty);
1155 list_for_each_entry_continue(ii, head, i_dirty) {
1156 err = nilfs_segctor_scan_file(
1157 sci, &ii->vfs_inode,
1158 &nilfs_sc_file_ops);
1159 if (unlikely(err)) {
1160 sci->sc_stage.gc_inode_ptr = list_entry(
1161 ii->i_dirty.prev,
1162 struct nilfs_inode_info,
1163 i_dirty);
1164 goto break_or_fail;
1165 }
1166 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1167 }
1168 sci->sc_stage.gc_inode_ptr = NULL;
1169 }
1170 sci->sc_stage.scnt++; /* Fall through */
1171 case NILFS_ST_FILE:
1172 head = &sci->sc_dirty_files;
1173 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1174 i_dirty);
1175 list_for_each_entry_continue(ii, head, i_dirty) {
1176 clear_bit(NILFS_I_DIRTY, &ii->i_state);
1177
1178 err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1179 &nilfs_sc_file_ops);
1180 if (unlikely(err)) {
1181 sci->sc_stage.dirty_file_ptr =
1182 list_entry(ii->i_dirty.prev,
1183 struct nilfs_inode_info,
1184 i_dirty);
1185 goto break_or_fail;
1186 }
1187 /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1188 /* XXX: required ? */
1189 }
1190 sci->sc_stage.dirty_file_ptr = NULL;
1191 if (mode == SC_FLUSH_FILE) {
1192 sci->sc_stage.scnt = NILFS_ST_DONE;
1193 return 0;
1194 }
9ff05123
RK		 1195 sci->sc_stage.scnt++;
1196 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1197 /* Fall through */
1198 case NILFS_ST_IFILE:
1199 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1200 &nilfs_sc_file_ops);
1201 if (unlikely(err))
1202 break;
1203 sci->sc_stage.scnt++;
1204 /* Creating a checkpoint */
1205 err = nilfs_segctor_create_checkpoint(sci);
1206 if (unlikely(err))
1207 break;
1208 /* Fall through */
1209 case NILFS_ST_CPFILE:
1210 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1211 &nilfs_sc_file_ops);
1212 if (unlikely(err))
1213 break;
1214 sci->sc_stage.scnt++; /* Fall through */
1215 case NILFS_ST_SUFILE:
071cb4b8
RK		 1216 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1217 sci->sc_nfreesegs, &ndone);
1218 if (unlikely(err)) {
1219 nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1220 sci->sc_freesegs, ndone,
1221 NULL);
9ff05123 1222 break;
071cb4b8
RK		 1223 }
1224 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1225
9ff05123
RK		 1226 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1227 &nilfs_sc_file_ops);
1228 if (unlikely(err))
1229 break;
1230 sci->sc_stage.scnt++; /* Fall through */
1231 case NILFS_ST_DAT:
1232 dat_stage:
1233 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1234 &nilfs_sc_dat_ops);
1235 if (unlikely(err))
1236 break;
1237 if (mode == SC_FLUSH_DAT) {
1238 sci->sc_stage.scnt = NILFS_ST_DONE;
1239 return 0;
1240 }
1241 sci->sc_stage.scnt++; /* Fall through */
1242 case NILFS_ST_SR:
1243 if (mode == SC_LSEG_SR) {
1244 /* Appending a super root */
1245 err = nilfs_segctor_add_super_root(sci);
1246 if (unlikely(err))
1247 break;
1248 }
1249 /* End of a logical segment */
1250 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1251 sci->sc_stage.scnt = NILFS_ST_DONE;
1252 return 0;
1253 case NILFS_ST_DSYNC:
1254 dsync_mode:
1255 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
f30bf3e4 1256 ii = sci->sc_dsync_inode;
9ff05123
RK		 1257 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1258 break;
1259
1260 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1261 if (unlikely(err))
1262 break;
9ff05123
RK		 1263 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1264 sci->sc_stage.scnt = NILFS_ST_DONE;
1265 return 0;
1266 case NILFS_ST_DONE:
1267 return 0;
1268 default:
1269 BUG();
1270 }
1271
1272 break_or_fail:
1273 return err;
1274}
1275
9ff05123
RK		 1276static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1277 struct the_nilfs *nilfs)
1278{
e29df395 1279 struct nilfs_segment_buffer *segbuf;
9ff05123
RK		 1280 __u64 nextnum;
1281 int err;
1282
1283 if (list_empty(&sci->sc_segbufs)) {
1284 segbuf = nilfs_segbuf_new(sci->sc_super);
1285 if (unlikely(!segbuf))
1286 return -ENOMEM;
1287 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1288 } else
1289 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1290
cece5520
RK		 1291 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, nilfs->ns_pseg_offset,
1292 nilfs);
9ff05123
RK		 1293
1294 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
9ff05123 1295 nilfs_shift_to_next_segment(nilfs);
cece5520 1296 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
9ff05123
RK		 1297 }
1298 sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1299
61a189e9 1300 err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
9ff05123
RK		 1301 if (unlikely(err))
1302 return err;
1303
1304 if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1305 /* Start from the head of a new full segment */
cece5520 1306 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
9ff05123
RK		 1307 if (unlikely(err))
1308 return err;
1309 } else
1310 nextnum = nilfs->ns_nextnum;
1311
1312 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1313 nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1314
1315 /* truncating segment buffers */
e29df395 1316 nilfs_truncate_logs(&sci->sc_segbufs, segbuf);
cece5520 1317 return 0;
9ff05123
RK		 1318}
1319
1320static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1321 struct the_nilfs *nilfs, int nadd)
1322{
e29df395 1323 struct nilfs_segment_buffer *segbuf, *prev;
9ff05123
RK		 1324 struct inode *sufile = nilfs->ns_sufile;
1325 __u64 nextnextnum;
1326 LIST_HEAD(list);
1327 int err, ret, i;
1328
1329 prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1330 /*
1331 * Since the segment specified with nextnum might be allocated during
1332 * the previous construction, the buffer including its segusage may
1333 * not be dirty. The following call ensures that the buffer is dirty
1334 * and will pin the buffer on memory until the sufile is written.
1335 */
61a189e9 1336 err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
9ff05123
RK		 1337 if (unlikely(err))
1338 return err;
1339
1340 for (i = 0; i < nadd; i++) {
1341 /* extend segment info */
1342 err = -ENOMEM;
1343 segbuf = nilfs_segbuf_new(sci->sc_super);
1344 if (unlikely(!segbuf))
1345 goto failed;
1346
1347 /* map this buffer to region of segment on-disk */
cece5520 1348 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
9ff05123
RK		 1349 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1350
1351 /* allocate the next next full segment */
1352 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1353 if (unlikely(err))
1354 goto failed_segbuf;
1355
1356 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1357 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1358
1359 list_add_tail(&segbuf->sb_list, &list);
1360 prev = segbuf;
1361 }
0935db74 1362 list_splice_tail(&list, &sci->sc_segbufs);
9ff05123
RK		 1363 return 0;
1364
1365 failed_segbuf:
1366 nilfs_segbuf_free(segbuf);
1367 failed:
e29df395 1368 list_for_each_entry(segbuf, &list, sb_list) {
9ff05123 1369 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1f5abe7e 1370 WARN_ON(ret); /* never fails */
9ff05123 1371 }
e29df395 1372 nilfs_destroy_logs(&list);
9ff05123
RK		 1373 return err;
1374}
1375
1376static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1377 struct the_nilfs *nilfs)
1378{
1379 struct nilfs_segment_buffer *segbuf;
9284ad2a 1380 int ret;
9ff05123
RK		 1381
1382 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1383 if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1384 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1f5abe7e 1385 WARN_ON(ret); /* never fails */
9ff05123 1386 }
9284ad2a 1387 if (atomic_read(&segbuf->sb_err)) {
9ff05123
RK		 1388 /* Case 1: The first segment failed */
1389 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1390 /* Case 1a: Partial segment appended into an existing
1391 segment */
1392 nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1393 segbuf->sb_fseg_end);
1394 else /* Case 1b: New full segment */
1395 set_nilfs_discontinued(nilfs);
9ff05123
RK		 1396 }
1397
1398 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1399 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1f5abe7e 1400 WARN_ON(ret); /* never fails */
9284ad2a
RK		 1401 if (atomic_read(&segbuf->sb_err) &&
1402 segbuf->sb_segnum != nilfs->ns_nextnum)
1403 /* Case 2: extended segment (!= next) failed */
1404 nilfs_sufile_set_error(nilfs->ns_sufile,
1405 segbuf->sb_segnum);
9ff05123
RK		 1406 }
1407}
1408
9ff05123
RK		 1409static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1410 struct the_nilfs *nilfs, int err)
1411{
1412 if (unlikely(err)) {
1413 nilfs_segctor_free_incomplete_segments(sci, nilfs);
071cb4b8
RK		 1414 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1415 int ret;
1416
1417 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1418 sci->sc_freesegs,
1419 sci->sc_nfreesegs,
1420 NULL);
1421 WARN_ON(ret); /* do not happen */
1422 }
9ff05123 1423 }
e29df395
RK		 1424 nilfs_clear_logs(&sci->sc_segbufs);
1425 sci->sc_super_root = NULL;
9ff05123
RK		 1426}
1427
1428static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1429 struct inode *sufile)
1430{
1431 struct nilfs_segment_buffer *segbuf;
9ff05123
RK		 1432 unsigned long live_blocks;
1433 int ret;
1434
1435 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
9ff05123
RK		 1436 live_blocks = segbuf->sb_sum.nblocks +
1437 (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
071ec54d
RK		 1438 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1439 live_blocks,
1440 sci->sc_seg_ctime);
1441 WARN_ON(ret); /* always succeed because the segusage is dirty */
9ff05123
RK		 1442 }
1443}
1444
1445static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1446 struct inode *sufile)
1447{
1448 struct nilfs_segment_buffer *segbuf;
9ff05123
RK		 1449 int ret;
1450
1451 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
071ec54d
RK		 1452 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1453 segbuf->sb_pseg_start -
1454 segbuf->sb_fseg_start, 0);
1455 WARN_ON(ret); /* always succeed because the segusage is dirty */
9ff05123
RK		 1456
1457 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
071ec54d
RK		 1458 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1459 0, 0);
1f5abe7e 1460 WARN_ON(ret); /* always succeed */
9ff05123
RK		 1461 }
1462}
1463
1464static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1465 struct nilfs_segment_buffer *last,
1466 struct inode *sufile)
1467{
e29df395 1468 struct nilfs_segment_buffer *segbuf = last;
9ff05123
RK		 1469 int ret;
1470
e29df395 1471 list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
9ff05123
RK		 1472 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1473 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1f5abe7e 1474 WARN_ON(ret);
9ff05123 1475 }
e29df395 1476 nilfs_truncate_logs(&sci->sc_segbufs, last);
9ff05123
RK		 1477}
1478
1479
1480static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1481 struct the_nilfs *nilfs, int mode)
1482{
1483 struct nilfs_cstage prev_stage = sci->sc_stage;
1484 int err, nadd = 1;
1485
1486 /* Collection retry loop */
1487 for (;;) {
1488 sci->sc_super_root = NULL;
1489 sci->sc_nblk_this_inc = 0;
1490 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1491
1492 err = nilfs_segctor_reset_segment_buffer(sci);
1493 if (unlikely(err))
1494 goto failed;
1495
1496 err = nilfs_segctor_collect_blocks(sci, mode);
1497 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1498 if (!err)
1499 break;
1500
1501 if (unlikely(err != -E2BIG))
1502 goto failed;
1503
1504 /* The current segment is filled up */
1505 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1506 break;
1507
071cb4b8
RK		 1508 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1509 err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1510 sci->sc_freesegs,
1511 sci->sc_nfreesegs,
1512 NULL);
1513 WARN_ON(err); /* do not happen */
1514 }
e29df395 1515 nilfs_clear_logs(&sci->sc_segbufs);
9ff05123
RK		 1516
1517 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1518 if (unlikely(err))
1519 return err;
1520
1521 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1522 sci->sc_stage = prev_stage;
1523 }
1524 nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1525 return 0;
1526
1527 failed:
1528 return err;
1529}
1530
1531static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1532 struct buffer_head *new_bh)
1533{
1534 BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1535
1536 list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1537 /* The caller must release old_bh */
1538}
1539
1540static int
1541nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1542 struct nilfs_segment_buffer *segbuf,
1543 int mode)
1544{
1545 struct inode *inode = NULL;
1546 sector_t blocknr;
1547 unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1548 unsigned long nblocks = 0, ndatablk = 0;
1549 struct nilfs_sc_operations *sc_op = NULL;
1550 struct nilfs_segsum_pointer ssp;
1551 struct nilfs_finfo *finfo = NULL;
1552 union nilfs_binfo binfo;
1553 struct buffer_head *bh, *bh_org;
1554 ino_t ino = 0;
1555 int err = 0;
1556
1557 if (!nfinfo)
1558 goto out;
1559
1560 blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1561 ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1562 ssp.offset = sizeof(struct nilfs_segment_summary);
1563
1564 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1565 if (bh == sci->sc_super_root)
1566 break;
1567 if (!finfo) {
1568 finfo = nilfs_segctor_map_segsum_entry(
1569 sci, &ssp, sizeof(*finfo));
1570 ino = le64_to_cpu(finfo->fi_ino);
1571 nblocks = le32_to_cpu(finfo->fi_nblocks);
1572 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1573
1574 if (buffer_nilfs_node(bh))
1575 inode = NILFS_BTNC_I(bh->b_page->mapping);
1576 else
1577 inode = NILFS_AS_I(bh->b_page->mapping);
1578
1579 if (mode == SC_LSEG_DSYNC)
1580 sc_op = &nilfs_sc_dsync_ops;
1581 else if (ino == NILFS_DAT_INO)
1582 sc_op = &nilfs_sc_dat_ops;
1583 else /* file blocks */
1584 sc_op = &nilfs_sc_file_ops;
1585 }
1586 bh_org = bh;
1587 get_bh(bh_org);
1588 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1589 &binfo);
1590 if (bh != bh_org)
1591 nilfs_list_replace_buffer(bh_org, bh);
1592 brelse(bh_org);
1593 if (unlikely(err))
1594 goto failed_bmap;
1595
1596 if (ndatablk > 0)
1597 sc_op->write_data_binfo(sci, &ssp, &binfo);
1598 else
1599 sc_op->write_node_binfo(sci, &ssp, &binfo);
1600
1601 blocknr++;
1602 if (--nblocks == 0) {
1603 finfo = NULL;
1604 if (--nfinfo == 0)
1605 break;
1606 } else if (ndatablk > 0)
1607 ndatablk--;
1608 }
1609 out:
1610 return 0;
1611
1612 failed_bmap:
1613 err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1614 return err;
1615}
1616
1617static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1618{
1619 struct nilfs_segment_buffer *segbuf;
1620 int err;
1621
1622 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1623 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1624 if (unlikely(err))
1625 return err;
1626 nilfs_segbuf_fill_in_segsum(segbuf);
1627 }
1628 return 0;
1629}
1630
1631static int
1632nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1633{
1634 struct page *clone_page;
1635 struct buffer_head *bh, *head, *bh2;
1636 void *kaddr;
1637
1638 bh = head = page_buffers(page);
1639
1640 clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1641 if (unlikely(!clone_page))
1642 return -ENOMEM;
1643
1644 bh2 = page_buffers(clone_page);
1645 kaddr = kmap_atomic(page, KM_USER0);
1646 do {
1647 if (list_empty(&bh->b_assoc_buffers))
1648 continue;
1649 get_bh(bh2);
1650 page_cache_get(clone_page); /* for each bh */
1651 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1652 bh2->b_blocknr = bh->b_blocknr;
1653 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1654 list_add_tail(&bh->b_assoc_buffers, out);
1655 } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1656 kunmap_atomic(kaddr, KM_USER0);
1657
1658 if (!TestSetPageWriteback(clone_page))
1659 inc_zone_page_state(clone_page, NR_WRITEBACK);
1660 unlock_page(clone_page);
1661
1662 return 0;
1663}
1664
1665static int nilfs_test_page_to_be_frozen(struct page *page)
1666{
1667 struct address_space *mapping = page->mapping;
1668
1669 if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1670 return 0;
1671
1672 if (page_mapped(page)) {
1673 ClearPageChecked(page);
1674 return 1;
1675 }
1676 return PageChecked(page);
1677}
1678
1679static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1680{
1681 if (!page || PageWriteback(page))
1682 /* For split b-tree node pages, this function may be called
1683 twice. We ignore the 2nd or later calls by this check. */
1684 return 0;
1685
1686 lock_page(page);
1687 clear_page_dirty_for_io(page);
1688 set_page_writeback(page);
1689 unlock_page(page);
1690
1691 if (nilfs_test_page_to_be_frozen(page)) {
1692 int err = nilfs_copy_replace_page_buffers(page, out);
1693 if (unlikely(err))
1694 return err;
1695 }
1696 return 0;
1697}
1698
1699static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1700 struct page **failed_page)
1701{
1702 struct nilfs_segment_buffer *segbuf;
1703 struct page *bd_page = NULL, *fs_page = NULL;
1704 struct list_head *list = &sci->sc_copied_buffers;
1705 int err;
1706
1707 *failed_page = NULL;
1708 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1709 struct buffer_head *bh;
1710
1711 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1712 b_assoc_buffers) {
1713 if (bh->b_page != bd_page) {
1714 if (bd_page) {
1715 lock_page(bd_page);
1716 clear_page_dirty_for_io(bd_page);
1717 set_page_writeback(bd_page);
1718 unlock_page(bd_page);
1719 }
1720 bd_page = bh->b_page;
1721 }
1722 }
1723
1724 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1725 b_assoc_buffers) {
1726 if (bh == sci->sc_super_root) {
1727 if (bh->b_page != bd_page) {
1728 lock_page(bd_page);
1729 clear_page_dirty_for_io(bd_page);
1730 set_page_writeback(bd_page);
1731 unlock_page(bd_page);
1732 bd_page = bh->b_page;
1733 }
1734 break;
1735 }
1736 if (bh->b_page != fs_page) {
1737 err = nilfs_begin_page_io(fs_page, list);
1738 if (unlikely(err)) {
1739 *failed_page = fs_page;
1740 goto out;
1741 }
1742 fs_page = bh->b_page;
1743 }
1744 }
1745 }
1746 if (bd_page) {
1747 lock_page(bd_page);
1748 clear_page_dirty_for_io(bd_page);
1749 set_page_writeback(bd_page);
1750 unlock_page(bd_page);
1751 }
1752 err = nilfs_begin_page_io(fs_page, list);
1753 if (unlikely(err))
1754 *failed_page = fs_page;
1755 out:
1756 return err;
1757}
1758
1759static int nilfs_segctor_write(struct nilfs_sc_info *sci,
9c965bac 1760 struct the_nilfs *nilfs)
9ff05123
RK		 1761{
1762 struct nilfs_segment_buffer *segbuf;
9ff05123
RK		 1763 int err, res;
1764
9ff05123 1765 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
9c965bac 1766 err = nilfs_segbuf_write(segbuf, nilfs);
9ff05123 1767
9284ad2a 1768 res = nilfs_segbuf_wait(segbuf);
0cfae3d8
RK		 1769 err = err ? : res;
1770 if (err)
9ff05123
RK		 1771 return err;
1772 }
1773 return 0;
1774}
1775
9ff05123
RK		 1776static void __nilfs_end_page_io(struct page *page, int err)
1777{
9ff05123
RK		 1778 if (!err) {
1779 if (!nilfs_page_buffers_clean(page))
1780 __set_page_dirty_nobuffers(page);
1781 ClearPageError(page);
1782 } else {
1783 __set_page_dirty_nobuffers(page);
1784 SetPageError(page);
1785 }
1786
1787 if (buffer_nilfs_allocated(page_buffers(page))) {
1788 if (TestClearPageWriteback(page))
1789 dec_zone_page_state(page, NR_WRITEBACK);
1790 } else
1791 end_page_writeback(page);
1792}
1793
1794static void nilfs_end_page_io(struct page *page, int err)
1795{
1796 if (!page)
1797 return;
1798
a9777845 1799 if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
8227b297
RK		 1800 /*
1801 * For b-tree node pages, this function may be called twice
1802 * or more because they might be split in a segment.
1803 */
a9777845
RK		 1804 if (PageDirty(page)) {
1805 /*
1806 * For pages holding split b-tree node buffers, dirty
1807 * flag on the buffers may be cleared discretely.
1808 * In that case, the page is once redirtied for
1809 * remaining buffers, and it must be cancelled if
1810 * all the buffers get cleaned later.
1811 */
1812 lock_page(page);
1813 if (nilfs_page_buffers_clean(page))
1814 __nilfs_clear_page_dirty(page);
1815 unlock_page(page);
1816 }
9ff05123 1817 return;
a9777845 1818 }
9ff05123
RK		 1819
1820 __nilfs_end_page_io(page, err);
1821}
1822
1823static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1824{
1825 struct buffer_head *bh, *head;
1826 struct page *page;
1827
1828 while (!list_empty(list)) {
1829 bh = list_entry(list->next, struct buffer_head,
1830 b_assoc_buffers);
1831 page = bh->b_page;
1832 page_cache_get(page);
1833 head = bh = page_buffers(page);
1834 do {
1835 if (!list_empty(&bh->b_assoc_buffers)) {
1836 list_del_init(&bh->b_assoc_buffers);
1837 if (!err) {
1838 set_buffer_uptodate(bh);
1839 clear_buffer_dirty(bh);
1840 clear_buffer_nilfs_volatile(bh);
1841 }
1842 brelse(bh); /* for b_assoc_buffers */
1843 }
1844 } while ((bh = bh->b_this_page) != head);
1845
1846 __nilfs_end_page_io(page, err);
1847 page_cache_release(page);
1848 }
1849}
1850
1851static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1852 struct page *failed_page, int err)
1853{
1854 struct nilfs_segment_buffer *segbuf;
1855 struct page *bd_page = NULL, *fs_page = NULL;
1856
1857 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1858 struct buffer_head *bh;
1859
1860 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1861 b_assoc_buffers) {
1862 if (bh->b_page != bd_page) {
1863 if (bd_page)
1864 end_page_writeback(bd_page);
1865 bd_page = bh->b_page;
1866 }
1867 }
1868
1869 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1870 b_assoc_buffers) {
1871 if (bh == sci->sc_super_root) {
1872 if (bh->b_page != bd_page) {
1873 end_page_writeback(bd_page);
1874 bd_page = bh->b_page;
1875 }
1876 break;
1877 }
1878 if (bh->b_page != fs_page) {
1879 nilfs_end_page_io(fs_page, err);
8227b297 1880 if (fs_page && fs_page == failed_page)
9ff05123
RK		 1881 goto done;
1882 fs_page = bh->b_page;
1883 }
1884 }
1885 }
1886 if (bd_page)
1887 end_page_writeback(bd_page);
1888
1889 nilfs_end_page_io(fs_page, err);
1890 done:
1891 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1892}
1893
1894static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1895 struct nilfs_segment_buffer *segbuf)
1896{
1897 nilfs->ns_segnum = segbuf->sb_segnum;
1898 nilfs->ns_nextnum = segbuf->sb_nextnum;
1899 nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1900 + segbuf->sb_sum.nblocks;
1901 nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1902 nilfs->ns_ctime = segbuf->sb_sum.ctime;
1903}
1904
1905static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1906{
1907 struct nilfs_segment_buffer *segbuf;
1908 struct page *bd_page = NULL, *fs_page = NULL;
1909 struct nilfs_sb_info *sbi = sci->sc_sbi;
1910 struct the_nilfs *nilfs = sbi->s_nilfs;
1911 int update_sr = (sci->sc_super_root != NULL);
1912
1913 list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1914 struct buffer_head *bh;
1915
1916 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1917 b_assoc_buffers) {
1918 set_buffer_uptodate(bh);
1919 clear_buffer_dirty(bh);
1920 if (bh->b_page != bd_page) {
1921 if (bd_page)
1922 end_page_writeback(bd_page);
1923 bd_page = bh->b_page;
1924 }
1925 }
1926 /*
1927 * We assume that the buffers which belong to the same page
1928 * continue over the buffer list.
1929 * Under this assumption, the last BHs of pages is
1930 * identifiable by the discontinuity of bh->b_page
1931 * (page != fs_page).
1932 *
1933 * For B-tree node blocks, however, this assumption is not
1934 * guaranteed. The cleanup code of B-tree node pages needs
1935 * special care.
1936 */
1937 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1938 b_assoc_buffers) {
1939 set_buffer_uptodate(bh);
1940 clear_buffer_dirty(bh);
1941 clear_buffer_nilfs_volatile(bh);
1942 if (bh == sci->sc_super_root) {
1943 if (bh->b_page != bd_page) {
1944 end_page_writeback(bd_page);
1945 bd_page = bh->b_page;
1946 }
1947 break;
1948 }
1949 if (bh->b_page != fs_page) {
1950 nilfs_end_page_io(fs_page, 0);
1951 fs_page = bh->b_page;
1952 }
1953 }
1954
1955 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
1956 if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
1957 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1958 sci->sc_lseg_stime = jiffies;
1959 }
1960 if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
1961 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1962 }
1963 }
1964 /*
1965 * Since pages may continue over multiple segment buffers,
1966 * end of the last page must be checked outside of the loop.
1967 */
1968 if (bd_page)
1969 end_page_writeback(bd_page);
1970
1971 nilfs_end_page_io(fs_page, 0);
1972
1973 nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
1974
1975 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1976
1977 if (nilfs_doing_gc()) {
1978 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1979 if (update_sr)
1980 nilfs_commit_gcdat_inode(nilfs);
1088dcf4 1981 } else
9ff05123 1982 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
9ff05123
RK		 1983
1984 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1985
1986 segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1987 nilfs_set_next_segment(nilfs, segbuf);
1988
1989 if (update_sr) {
1990 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
e339ad31
RK		 1991 segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1992 sbi->s_super->s_dirt = 1;
9ff05123 1993
c96fa464 1994 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
9ff05123
RK		 1995 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1996 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1997 } else
1998 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1999}
2000
2001static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2002 struct nilfs_sb_info *sbi)
2003{
2004 struct nilfs_inode_info *ii, *n;
2005 __u64 cno = sbi->s_nilfs->ns_cno;
2006
2007 spin_lock(&sbi->s_inode_lock);
2008 retry:
2009 list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2010 if (!ii->i_bh) {
2011 struct buffer_head *ibh;
2012 int err;
2013
2014 spin_unlock(&sbi->s_inode_lock);
2015 err = nilfs_ifile_get_inode_block(
2016 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2017 if (unlikely(err)) {
2018 nilfs_warning(sbi->s_super, __func__,
2019 "failed to get inode block.\n");
2020 return err;
2021 }
2022 nilfs_mdt_mark_buffer_dirty(ibh);
2023 nilfs_mdt_mark_dirty(sbi->s_ifile);
2024 spin_lock(&sbi->s_inode_lock);
2025 if (likely(!ii->i_bh))
2026 ii->i_bh = ibh;
2027 else
2028 brelse(ibh);
2029 goto retry;
2030 }
2031 ii->i_cno = cno;
2032
2033 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2034 set_bit(NILFS_I_BUSY, &ii->i_state);
2035 list_del(&ii->i_dirty);
2036 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2037 }
2038 spin_unlock(&sbi->s_inode_lock);
2039
2040 NILFS_I(sbi->s_ifile)->i_cno = cno;
2041
2042 return 0;
2043}
2044
2045static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2046 struct nilfs_sb_info *sbi)
2047{
2048 struct nilfs_transaction_info *ti = current->journal_info;
2049 struct nilfs_inode_info *ii, *n;
2050 __u64 cno = sbi->s_nilfs->ns_cno;
2051
2052 spin_lock(&sbi->s_inode_lock);
2053 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2054 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2055 test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2056 /* The current checkpoint number (=nilfs->ns_cno) is
2057 changed between check-in and check-out only if the
2058 super root is written out. So, we can update i_cno
2059 for the inodes that remain in the dirty list. */
2060 ii->i_cno = cno;
2061 continue;
2062 }
2063 clear_bit(NILFS_I_BUSY, &ii->i_state);
2064 brelse(ii->i_bh);
2065 ii->i_bh = NULL;
2066 list_del(&ii->i_dirty);
2067 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2068 }
2069 spin_unlock(&sbi->s_inode_lock);
2070}
2071
9ff05123
RK		 2072/*
2073 * Main procedure of segment constructor
2074 */
2075static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2076{
2077 struct nilfs_sb_info *sbi = sci->sc_sbi;
2078 struct the_nilfs *nilfs = sbi->s_nilfs;
2079 struct page *failed_page;
2080 int err, has_sr = 0;
2081
2082 sci->sc_stage.scnt = NILFS_ST_INIT;
2083
2084 err = nilfs_segctor_check_in_files(sci, sbi);
2085 if (unlikely(err))
2086 goto out;
2087
2088 if (nilfs_test_metadata_dirty(sbi))
2089 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2090
2091 if (nilfs_segctor_clean(sci))
2092 goto out;
2093
2094 do {
2095 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2096
2097 err = nilfs_segctor_begin_construction(sci, nilfs);
2098 if (unlikely(err))
2099 goto out;
2100
2101 /* Update time stamp */
2102 sci->sc_seg_ctime = get_seconds();
2103
2104 err = nilfs_segctor_collect(sci, nilfs, mode);
2105 if (unlikely(err))
2106 goto failed;
2107
2108 has_sr = (sci->sc_super_root != NULL);
2109
2110 /* Avoid empty segment */
2111 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2112 NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
9ff05123
RK		 2113 nilfs_segctor_end_construction(sci, nilfs, 1);
2114 goto out;
2115 }
2116
2117 err = nilfs_segctor_assign(sci, mode);
2118 if (unlikely(err))
2119 goto failed;
2120
9ff05123
RK		 2121 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2122 nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2123
2124 if (has_sr) {
2125 err = nilfs_segctor_fill_in_checkpoint(sci);
2126 if (unlikely(err))
2127 goto failed_to_make_up;
2128
2129 nilfs_segctor_fill_in_super_root(sci, nilfs);
2130 }
2131 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2132
2133 /* Write partial segments */
2134 err = nilfs_segctor_prepare_write(sci, &failed_page);
2135 if (unlikely(err))
2136 goto failed_to_write;
2137
2138 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2139
9c965bac 2140 err = nilfs_segctor_write(sci, nilfs);
9ff05123
RK		 2141 if (unlikely(err))
2142 goto failed_to_write;
2143
2144 nilfs_segctor_complete_write(sci);
2145
2146 /* Commit segments */
071cb4b8 2147 if (has_sr)
9ff05123 2148 nilfs_segctor_clear_metadata_dirty(sci);
9ff05123
RK		 2149
2150 nilfs_segctor_end_construction(sci, nilfs, 0);
2151
2152 } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2153
9ff05123 2154 out:
e29df395 2155 nilfs_destroy_logs(&sci->sc_segbufs);
9ff05123
RK		 2156 nilfs_segctor_check_out_files(sci, sbi);
2157 return err;
2158
2159 failed_to_write:
2160 nilfs_segctor_abort_write(sci, failed_page, err);
2161 nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2162
2163 failed_to_make_up:
2164 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2165 nilfs_redirty_inodes(&sci->sc_dirty_files);
9ff05123
RK		 2166
2167 failed:
2168 if (nilfs_doing_gc())
2169 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2170 nilfs_segctor_end_construction(sci, nilfs, err);
2171 goto out;
2172}
2173
2174/**
2175 * nilfs_secgtor_start_timer - set timer of background write
2176 * @sci: nilfs_sc_info
2177 *
2178 * If the timer has already been set, it ignores the new request.
2179 * This function MUST be called within a section locking the segment
2180 * semaphore.
2181 */
2182static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2183{
2184 spin_lock(&sci->sc_state_lock);
2185 if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2186 sci->sc_timer->expires = jiffies + sci->sc_interval;
2187 add_timer(sci->sc_timer);
2188 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2189 }
2190 spin_unlock(&sci->sc_state_lock);
2191}
2192
2193static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2194{
2195 spin_lock(&sci->sc_state_lock);
2196 if (!(sci->sc_flush_request & (1 << bn))) {
2197 unsigned long prev_req = sci->sc_flush_request;
2198
2199 sci->sc_flush_request |= (1 << bn);
2200 if (!prev_req)
2201 wake_up(&sci->sc_wait_daemon);
2202 }
2203 spin_unlock(&sci->sc_state_lock);
2204}
2205
2206/**
2207 * nilfs_flush_segment - trigger a segment construction for resource control
2208 * @sb: super block
2209 * @ino: inode number of the file to be flushed out.
2210 */
2211void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2212{
2213 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2214 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2215
2216 if (!sci || nilfs_doing_construction())
2217 return;
2218 nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2219 /* assign bit 0 to data files */
2220}
2221
9ff05123
RK		 2222struct nilfs_segctor_wait_request {
2223 wait_queue_t wq;
2224 __u32 seq;
2225 int err;
2226 atomic_t done;
2227};
2228
2229static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2230{
2231 struct nilfs_segctor_wait_request wait_req;
2232 int err = 0;
2233
2234 spin_lock(&sci->sc_state_lock);
2235 init_wait(&wait_req.wq);
2236 wait_req.err = 0;
2237 atomic_set(&wait_req.done, 0);
2238 wait_req.seq = ++sci->sc_seq_request;
2239 spin_unlock(&sci->sc_state_lock);
2240
2241 init_waitqueue_entry(&wait_req.wq, current);
2242 add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2243 set_current_state(TASK_INTERRUPTIBLE);
2244 wake_up(&sci->sc_wait_daemon);
2245
2246 for (;;) {
2247 if (atomic_read(&wait_req.done)) {
2248 err = wait_req.err;
2249 break;
2250 }
2251 if (!signal_pending(current)) {
2252 schedule();
2253 continue;
2254 }
2255 err = -ERESTARTSYS;
2256 break;
2257 }
2258 finish_wait(&sci->sc_wait_request, &wait_req.wq);
2259 return err;
2260}
2261
2262static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2263{
2264 struct nilfs_segctor_wait_request *wrq, *n;
2265 unsigned long flags;
2266
2267 spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2268 list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2269 wq.task_list) {
2270 if (!atomic_read(&wrq->done) &&
2271 nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2272 wrq->err = err;
2273 atomic_set(&wrq->done, 1);
2274 }
2275 if (atomic_read(&wrq->done)) {
2276 wrq->wq.func(&wrq->wq,
2277 TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2278 0, NULL);
2279 }
2280 }
2281 spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2282}
2283
2284/**
2285 * nilfs_construct_segment - construct a logical segment
2286 * @sb: super block
2287 *
2288 * Return Value: On success, 0 is retured. On errors, one of the following
2289 * negative error code is returned.
2290 *
2291 * %-EROFS - Read only filesystem.
2292 *
2293 * %-EIO - I/O error
2294 *
2295 * %-ENOSPC - No space left on device (only in a panic state).
2296 *
2297 * %-ERESTARTSYS - Interrupted.
2298 *
2299 * %-ENOMEM - Insufficient memory available.
2300 */
2301int nilfs_construct_segment(struct super_block *sb)
2302{
2303 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2304 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2305 struct nilfs_transaction_info *ti;
2306 int err;
2307
2308 if (!sci)
2309 return -EROFS;
2310
2311 /* A call inside transactions causes a deadlock. */
2312 BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2313
2314 err = nilfs_segctor_sync(sci);
2315 return err;
2316}
2317
2318/**
2319 * nilfs_construct_dsync_segment - construct a data-only logical segment
2320 * @sb: super block
f30bf3e4
RK		 2321 * @inode: inode whose data blocks should be written out
2322 * @start: start byte offset
2323 * @end: end byte offset (inclusive)
9ff05123
RK		 2324 *
2325 * Return Value: On success, 0 is retured. On errors, one of the following
2326 * negative error code is returned.
2327 *
2328 * %-EROFS - Read only filesystem.
2329 *
2330 * %-EIO - I/O error
2331 *
2332 * %-ENOSPC - No space left on device (only in a panic state).
2333 *
2334 * %-ERESTARTSYS - Interrupted.
2335 *
2336 * %-ENOMEM - Insufficient memory available.
2337 */
f30bf3e4
RK		 2338int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2339 loff_t start, loff_t end)
9ff05123
RK		 2340{
2341 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2342 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2343 struct nilfs_inode_info *ii;
2344 struct nilfs_transaction_info ti;
2345 int err = 0;
2346
2347 if (!sci)
2348 return -EROFS;
2349
2350 nilfs_transaction_lock(sbi, &ti, 0);
2351
2352 ii = NILFS_I(inode);
2353 if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2354 nilfs_test_opt(sbi, STRICT_ORDER) ||
2355 test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2356 nilfs_discontinued(sbi->s_nilfs)) {
2357 nilfs_transaction_unlock(sbi);
2358 err = nilfs_segctor_sync(sci);
2359 return err;
2360 }
2361
2362 spin_lock(&sbi->s_inode_lock);
2363 if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2364 !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2365 spin_unlock(&sbi->s_inode_lock);
2366 nilfs_transaction_unlock(sbi);
2367 return 0;
2368 }
2369 spin_unlock(&sbi->s_inode_lock);
f30bf3e4
RK		 2370 sci->sc_dsync_inode = ii;
2371 sci->sc_dsync_start = start;
2372 sci->sc_dsync_end = end;
9ff05123
RK		 2373
2374 err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2375
2376 nilfs_transaction_unlock(sbi);
2377 return err;
2378}
2379
2380struct nilfs_segctor_req {
2381 int mode;
2382 __u32 seq_accepted;
2383 int sc_err; /* construction failure */
2384 int sb_err; /* super block writeback failure */
2385};
2386
2387#define FLUSH_FILE_BIT (0x1) /* data file only */
2388#define FLUSH_DAT_BIT (1 << NILFS_DAT_INO) /* DAT only */
2389
2390static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2391 struct nilfs_segctor_req *req)
2392{
9ff05123
RK		 2393 req->sc_err = req->sb_err = 0;
2394 spin_lock(&sci->sc_state_lock);
2395 req->seq_accepted = sci->sc_seq_request;
2396 spin_unlock(&sci->sc_state_lock);
2397
2398 if (sci->sc_timer)
2399 del_timer_sync(sci->sc_timer);
2400}
2401
2402static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2403 struct nilfs_segctor_req *req)
2404{
2405 /* Clear requests (even when the construction failed) */
2406 spin_lock(&sci->sc_state_lock);
2407
9ff05123 2408 if (req->mode == SC_LSEG_SR) {
aeda7f63 2409 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
9ff05123
RK		 2410 sci->sc_seq_done = req->seq_accepted;
2411 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2412 sci->sc_flush_request = 0;
aeda7f63
RK		 2413 } else {
2414 if (req->mode == SC_FLUSH_FILE)
2415 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2416 else if (req->mode == SC_FLUSH_DAT)
2417 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2418
2419 /* re-enable timer if checkpoint creation was not done */
2420 if (sci->sc_timer && (sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2421 time_before(jiffies, sci->sc_timer->expires))
2422 add_timer(sci->sc_timer);
2423 }
9ff05123
RK		 2424 spin_unlock(&sci->sc_state_lock);
2425}
2426
2427static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2428 struct nilfs_segctor_req *req)
2429{
2430 struct nilfs_sb_info *sbi = sci->sc_sbi;
2431 struct the_nilfs *nilfs = sbi->s_nilfs;
2432 int err = 0;
2433
2434 if (nilfs_discontinued(nilfs))
2435 req->mode = SC_LSEG_SR;
2436 if (!nilfs_segctor_confirm(sci)) {
2437 err = nilfs_segctor_do_construct(sci, req->mode);
2438 req->sc_err = err;
2439 }
2440 if (likely(!err)) {
2441 if (req->mode != SC_FLUSH_DAT)
2442 atomic_set(&nilfs->ns_ndirtyblks, 0);
2443 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2444 nilfs_discontinued(nilfs)) {
2445 down_write(&nilfs->ns_sem);
1dfa2710
JS		 2446 req->sb_err = nilfs_commit_super(sbi,
2447 nilfs_altsb_need_update(nilfs));
9ff05123
RK		 2448 up_write(&nilfs->ns_sem);
2449 }
2450 }
2451 return err;
2452}
2453
2454static void nilfs_construction_timeout(unsigned long data)
2455{
2456 struct task_struct *p = (struct task_struct *)data;
2457 wake_up_process(p);
2458}
2459
2460static void
2461nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2462{
2463 struct nilfs_inode_info *ii, *n;
2464
2465 list_for_each_entry_safe(ii, n, head, i_dirty) {
2466 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2467 continue;
2468 hlist_del_init(&ii->vfs_inode.i_hash);
2469 list_del_init(&ii->i_dirty);
2470 nilfs_clear_gcinode(&ii->vfs_inode);
2471 }
2472}
2473
4f6b8288
RK		 2474int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2475 void **kbufs)
9ff05123
RK		 2476{
2477 struct nilfs_sb_info *sbi = NILFS_SB(sb);
2478 struct nilfs_sc_info *sci = NILFS_SC(sbi);
2479 struct the_nilfs *nilfs = sbi->s_nilfs;
2480 struct nilfs_transaction_info ti;
2481 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2482 int err;
2483
2484 if (unlikely(!sci))
2485 return -EROFS;
2486
2487 nilfs_transaction_lock(sbi, &ti, 1);
2488
2489 err = nilfs_init_gcdat_inode(nilfs);
2490 if (unlikely(err))
2491 goto out_unlock;
071cb4b8 2492
4f6b8288 2493 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
9ff05123
RK		 2494 if (unlikely(err))
2495 goto out_unlock;
2496
071cb4b8
RK		 2497 sci->sc_freesegs = kbufs[4];
2498 sci->sc_nfreesegs = argv[4].v_nmembs;
0935db74 2499 list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
9ff05123
RK		 2500
2501 for (;;) {
2502 nilfs_segctor_accept(sci, &req);
2503 err = nilfs_segctor_construct(sci, &req);
2504 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2505 nilfs_segctor_notify(sci, &req);
2506
2507 if (likely(!err))
2508 break;
2509
2510 nilfs_warning(sb, __func__,
2511 "segment construction failed. (err=%d)", err);
2512 set_current_state(TASK_INTERRUPTIBLE);
2513 schedule_timeout(sci->sc_interval);
2514 }
2515
2516 out_unlock:
071cb4b8
RK		 2517 sci->sc_freesegs = NULL;
2518 sci->sc_nfreesegs = 0;
9ff05123
RK		 2519 nilfs_clear_gcdat_inode(nilfs);
2520 nilfs_transaction_unlock(sbi);
2521 return err;
2522}
2523
2524static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2525{
2526 struct nilfs_sb_info *sbi = sci->sc_sbi;
2527 struct nilfs_transaction_info ti;
2528 struct nilfs_segctor_req req = { .mode = mode };
2529
2530 nilfs_transaction_lock(sbi, &ti, 0);
2531
2532 nilfs_segctor_accept(sci, &req);
2533 nilfs_segctor_construct(sci, &req);
2534 nilfs_segctor_notify(sci, &req);
2535
2536 /*
2537 * Unclosed segment should be retried. We do this using sc_timer.
2538 * Timeout of sc_timer will invoke complete construction which leads
2539 * to close the current logical segment.
2540 */
2541 if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2542 nilfs_segctor_start_timer(sci);
2543
2544 nilfs_transaction_unlock(sbi);
2545}
2546
2547static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2548{
2549 int mode = 0;
2550 int err;
2551
2552 spin_lock(&sci->sc_state_lock);
2553 mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2554 SC_FLUSH_DAT : SC_FLUSH_FILE;
2555 spin_unlock(&sci->sc_state_lock);
2556
2557 if (mode) {
2558 err = nilfs_segctor_do_construct(sci, mode);
2559
2560 spin_lock(&sci->sc_state_lock);
2561 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2562 ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2563 spin_unlock(&sci->sc_state_lock);
2564 }
2565 clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2566}
2567
2568static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2569{
2570 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2571 time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2572 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2573 return SC_FLUSH_FILE;
2574 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2575 return SC_FLUSH_DAT;
2576 }
2577 return SC_LSEG_SR;
2578}
2579
2580/**
2581 * nilfs_segctor_thread - main loop of the segment constructor thread.
2582 * @arg: pointer to a struct nilfs_sc_info.
2583 *
2584 * nilfs_segctor_thread() initializes a timer and serves as a daemon
2585 * to execute segment constructions.
2586 */
2587static int nilfs_segctor_thread(void *arg)
2588{
2589 struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2590 struct timer_list timer;
2591 int timeout = 0;
2592
2593 init_timer(&timer);
2594 timer.data = (unsigned long)current;
2595 timer.function = nilfs_construction_timeout;
2596 sci->sc_timer = &timer;
2597
2598 /* start sync. */
2599 sci->sc_task = current;
2600 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2601 printk(KERN_INFO
2602 "segctord starting. Construction interval = %lu seconds, "
2603 "CP frequency < %lu seconds\n",
2604 sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2605
2606 spin_lock(&sci->sc_state_lock);
2607 loop:
2608 for (;;) {
2609 int mode;
2610
2611 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2612 goto end_thread;
2613
2614 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2615 mode = SC_LSEG_SR;
2616 else if (!sci->sc_flush_request)
2617 break;
2618 else
2619 mode = nilfs_segctor_flush_mode(sci);
2620
2621 spin_unlock(&sci->sc_state_lock);
2622 nilfs_segctor_thread_construct(sci, mode);
2623 spin_lock(&sci->sc_state_lock);
2624 timeout = 0;
2625 }
2626
2627
2628 if (freezing(current)) {
2629 spin_unlock(&sci->sc_state_lock);
2630 refrigerator();
2631 spin_lock(&sci->sc_state_lock);
2632 } else {
2633 DEFINE_WAIT(wait);
2634 int should_sleep = 1;
1dfa2710 2635 struct the_nilfs *nilfs;
9ff05123
RK		 2636
2637 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2638 TASK_INTERRUPTIBLE);
2639
2640 if (sci->sc_seq_request != sci->sc_seq_done)
2641 should_sleep = 0;
2642 else if (sci->sc_flush_request)
2643 should_sleep = 0;
2644 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2645 should_sleep = time_before(jiffies,
2646 sci->sc_timer->expires);
2647
2648 if (should_sleep) {
2649 spin_unlock(&sci->sc_state_lock);
2650 schedule();
2651 spin_lock(&sci->sc_state_lock);
2652 }
2653 finish_wait(&sci->sc_wait_daemon, &wait);
2654 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2655 time_after_eq(jiffies, sci->sc_timer->expires));
1dfa2710
JS		 2656 nilfs = sci->sc_sbi->s_nilfs;
2657 if (sci->sc_super->s_dirt && nilfs_sb_need_update(nilfs))
2658 set_nilfs_discontinued(nilfs);
9ff05123
RK		 2659 }
2660 goto loop;
2661
2662 end_thread:
2663 spin_unlock(&sci->sc_state_lock);
2664 del_timer_sync(sci->sc_timer);
2665 sci->sc_timer = NULL;
2666
2667 /* end sync. */
2668 sci->sc_task = NULL;
2669 wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2670 return 0;
2671}
2672
2673static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2674{
2675 struct task_struct *t;
2676
2677 t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2678 if (IS_ERR(t)) {
2679 int err = PTR_ERR(t);
2680
2681 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2682 err);
2683 return err;
2684 }
2685 wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2686 return 0;
2687}
2688
2689static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2690{
2691 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2692
2693 while (sci->sc_task) {
2694 wake_up(&sci->sc_wait_daemon);
2695 spin_unlock(&sci->sc_state_lock);
2696 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2697 spin_lock(&sci->sc_state_lock);
2698 }
2699}
2700
cece5520 2701static int nilfs_segctor_init(struct nilfs_sc_info *sci)
9ff05123 2702{
9ff05123 2703 sci->sc_seq_done = sci->sc_seq_request;
9ff05123 2704
cece5520 2705 return nilfs_segctor_start_thread(sci);
9ff05123
RK		 2706}
2707
2708/*
2709 * Setup & clean-up functions
2710 */
2711static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
2712{
2713 struct nilfs_sc_info *sci;
2714
2715 sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2716 if (!sci)
2717 return NULL;
2718
2719 sci->sc_sbi = sbi;
2720 sci->sc_super = sbi->s_super;
2721
2722 init_waitqueue_head(&sci->sc_wait_request);
2723 init_waitqueue_head(&sci->sc_wait_daemon);
2724 init_waitqueue_head(&sci->sc_wait_task);
2725 spin_lock_init(&sci->sc_state_lock);
2726 INIT_LIST_HEAD(&sci->sc_dirty_files);
2727 INIT_LIST_HEAD(&sci->sc_segbufs);
2728 INIT_LIST_HEAD(&sci->sc_gc_inodes);
9ff05123
RK		 2729 INIT_LIST_HEAD(&sci->sc_copied_buffers);
2730
2731 sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2732 sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2733 sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2734
2735 if (sbi->s_interval)
2736 sci->sc_interval = sbi->s_interval;
2737 if (sbi->s_watermark)
2738 sci->sc_watermark = sbi->s_watermark;
2739 return sci;
2740}
2741
2742static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2743{
2744 int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2745
2746 /* The segctord thread was stopped and its timer was removed.
2747 But some tasks remain. */
2748 do {
2749 struct nilfs_sb_info *sbi = sci->sc_sbi;
2750 struct nilfs_transaction_info ti;
2751 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2752
2753 nilfs_transaction_lock(sbi, &ti, 0);
2754 nilfs_segctor_accept(sci, &req);
2755 ret = nilfs_segctor_construct(sci, &req);
2756 nilfs_segctor_notify(sci, &req);
2757 nilfs_transaction_unlock(sbi);
2758
2759 } while (ret && retrycount-- > 0);
2760}
2761
2762/**
2763 * nilfs_segctor_destroy - destroy the segment constructor.
2764 * @sci: nilfs_sc_info
2765 *
2766 * nilfs_segctor_destroy() kills the segctord thread and frees
2767 * the nilfs_sc_info struct.
2768 * Caller must hold the segment semaphore.
2769 */
2770static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2771{
2772 struct nilfs_sb_info *sbi = sci->sc_sbi;
2773 int flag;
2774
2775 up_write(&sbi->s_nilfs->ns_segctor_sem);
2776
2777 spin_lock(&sci->sc_state_lock);
2778 nilfs_segctor_kill_thread(sci);
2779 flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2780 || sci->sc_seq_request != sci->sc_seq_done);
2781 spin_unlock(&sci->sc_state_lock);
2782
2783 if (flag || nilfs_segctor_confirm(sci))
2784 nilfs_segctor_write_out(sci);
2785
1f5abe7e 2786 WARN_ON(!list_empty(&sci->sc_copied_buffers));
9ff05123
RK		 2787
2788 if (!list_empty(&sci->sc_dirty_files)) {
2789 nilfs_warning(sbi->s_super, __func__,
2790 "dirty file(s) after the final construction\n");
2791 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
2792 }
9ff05123 2793
1f5abe7e 2794 WARN_ON(!list_empty(&sci->sc_segbufs));
9ff05123 2795
9ff05123
RK		 2796 down_write(&sbi->s_nilfs->ns_segctor_sem);
2797
2798 kfree(sci);
2799}
2800
2801/**
2802 * nilfs_attach_segment_constructor - attach a segment constructor
2803 * @sbi: nilfs_sb_info
9ff05123
RK		 2804 *
2805 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
2806 * initilizes it, and starts the segment constructor.
2807 *
2808 * Return Value: On success, 0 is returned. On error, one of the following
2809 * negative error code is returned.
2810 *
2811 * %-ENOMEM - Insufficient memory available.
2812 */
cece5520 2813int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi)
9ff05123
RK		 2814{
2815 struct the_nilfs *nilfs = sbi->s_nilfs;
2816 int err;
2817
2818 /* Each field of nilfs_segctor is cleared through the initialization
2819 of super-block info */
2820 sbi->s_sc_info = nilfs_segctor_new(sbi);
2821 if (!sbi->s_sc_info)
2822 return -ENOMEM;
2823
2824 nilfs_attach_writer(nilfs, sbi);
cece5520 2825 err = nilfs_segctor_init(NILFS_SC(sbi));
9ff05123
RK		 2826 if (err) {
2827 nilfs_detach_writer(nilfs, sbi);
2828 kfree(sbi->s_sc_info);
2829 sbi->s_sc_info = NULL;
2830 }
2831 return err;
2832}
2833
2834/**
2835 * nilfs_detach_segment_constructor - destroy the segment constructor
2836 * @sbi: nilfs_sb_info
2837 *
2838 * nilfs_detach_segment_constructor() kills the segment constructor daemon,
2839 * frees the struct nilfs_sc_info, and destroy the dirty file list.
2840 */
2841void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2842{
2843 struct the_nilfs *nilfs = sbi->s_nilfs;
2844 LIST_HEAD(garbage_list);
2845
2846 down_write(&nilfs->ns_segctor_sem);
2847 if (NILFS_SC(sbi)) {
2848 nilfs_segctor_destroy(NILFS_SC(sbi));
2849 sbi->s_sc_info = NULL;
2850 }
2851
2852 /* Force to free the list of dirty files */
2853 spin_lock(&sbi->s_inode_lock);
2854 if (!list_empty(&sbi->s_dirty_files)) {
2855 list_splice_init(&sbi->s_dirty_files, &garbage_list);
2856 nilfs_warning(sbi->s_super, __func__,
2857 "Non empty dirty list after the last "
2858 "segment construction\n");
2859 }
2860 spin_unlock(&sbi->s_inode_lock);
2861 up_write(&nilfs->ns_segctor_sem);
2862
2863 nilfs_dispose_list(sbi, &garbage_list, 1);
2864 nilfs_detach_writer(nilfs, sbi);
2865}
Linux 6.x block layer and io_uring tree(s)